summaryrefslogtreecommitdiffstats
path: root/libxml2/parser.c
diff options
context:
space:
mode:
authorWilliam Joye <wjoye@cfa.harvard.edu>2016-10-19 18:22:05 (GMT)
committerWilliam Joye <wjoye@cfa.harvard.edu>2016-10-19 18:22:05 (GMT)
commit34be72f9ed749a5c013d3f7f47d810e8caf652cb (patch)
tree2f75b6050c088da53d774b3da3fb71e7fa49ec78 /libxml2/parser.c
parent604eac5688c5d437ed6d585ee63e53fba6f558c0 (diff)
parent0f851b6f17365c29b56d6a17564d58a33120508a (diff)
downloadblt-34be72f9ed749a5c013d3f7f47d810e8caf652cb.zip
blt-34be72f9ed749a5c013d3f7f47d810e8caf652cb.tar.gz
blt-34be72f9ed749a5c013d3f7f47d810e8caf652cb.tar.bz2
Merge commit '0f851b6f17365c29b56d6a17564d58a33120508a' as 'libxml2'
Diffstat (limited to 'libxml2/parser.c')
-rw-r--r--libxml2/parser.c15802
1 files changed, 15802 insertions, 0 deletions
diff --git a/libxml2/parser.c b/libxml2/parser.c
new file mode 100644
index 0000000..f6d652e
--- /dev/null
+++ b/libxml2/parser.c
@@ -0,0 +1,15802 @@
+/*
+ * parser.c : an XML 1.0 parser, namespaces and validity support are mostly
+ * implemented on top of the SAX interfaces
+ *
+ * References:
+ * The XML specification:
+ * http://www.w3.org/TR/REC-xml
+ * Original 1.0 version:
+ * http://www.w3.org/TR/1998/REC-xml-19980210
+ * XML second edition working draft
+ * http://www.w3.org/TR/2000/WD-xml-2e-20000814
+ *
+ * Okay this is a big file, the parser core is around 7000 lines, then it
+ * is followed by the progressive parser top routines, then the various
+ * high level APIs to call the parser and a few miscellaneous functions.
+ * A number of helper functions and deprecated ones have been moved to
+ * parserInternals.c to reduce this file size.
+ * As much as possible the functions are associated with their relative
+ * production in the XML specification. A few productions defining the
+ * different ranges of character are actually implanted either in
+ * parserInternals.h or parserInternals.c
+ * The DOM tree build is realized from the default SAX callbacks in
+ * the module SAX.c.
+ * The routines doing the validation checks are in valid.c and called either
+ * from the SAX callbacks or as standalone functions using a preparsed
+ * document.
+ *
+ * See Copyright for the status of this software.
+ *
+ * daniel@veillard.com
+ */
+
+#define IN_LIBXML
+#include "libxml.h"
+
+#if defined(WIN32) && !defined (__CYGWIN__)
+#define XML_DIR_SEP '\\'
+#else
+#define XML_DIR_SEP '/'
+#endif
+
+#include <stdlib.h>
+#include <limits.h>
+#include <string.h>
+#include <stdarg.h>
+#include <libxml/xmlmemory.h>
+#include <libxml/threads.h>
+#include <libxml/globals.h>
+#include <libxml/tree.h>
+#include <libxml/parser.h>
+#include <libxml/parserInternals.h>
+#include <libxml/valid.h>
+#include <libxml/entities.h>
+#include <libxml/xmlerror.h>
+#include <libxml/encoding.h>
+#include <libxml/xmlIO.h>
+#include <libxml/uri.h>
+#ifdef LIBXML_CATALOG_ENABLED
+#include <libxml/catalog.h>
+#endif
+#ifdef LIBXML_SCHEMAS_ENABLED
+#include <libxml/xmlschemastypes.h>
+#include <libxml/relaxng.h>
+#endif
+#ifdef HAVE_CTYPE_H
+#include <ctype.h>
+#endif
+#ifdef HAVE_STDLIB_H
+#include <stdlib.h>
+#endif
+#ifdef HAVE_SYS_STAT_H
+#include <sys/stat.h>
+#endif
+#ifdef HAVE_FCNTL_H
+#include <fcntl.h>
+#endif
+#ifdef HAVE_UNISTD_H
+#include <unistd.h>
+#endif
+#ifdef HAVE_ZLIB_H
+#include <zlib.h>
+#endif
+#ifdef HAVE_LZMA_H
+#include <lzma.h>
+#endif
+
+#include "buf.h"
+#include "enc.h"
+
+static void
+xmlFatalErr(xmlParserCtxtPtr ctxt, xmlParserErrors error, const char *info);
+
+static xmlParserCtxtPtr
+xmlCreateEntityParserCtxtInternal(const xmlChar *URL, const xmlChar *ID,
+ const xmlChar *base, xmlParserCtxtPtr pctx);
+
+static void xmlHaltParser(xmlParserCtxtPtr ctxt);
+
+/************************************************************************
+ * *
+ * Arbitrary limits set in the parser. See XML_PARSE_HUGE *
+ * *
+ ************************************************************************/
+
+#define XML_PARSER_BIG_ENTITY 1000
+#define XML_PARSER_LOT_ENTITY 5000
+
+/*
+ * XML_PARSER_NON_LINEAR is the threshold where the ratio of parsed entity
+ * replacement over the size in byte of the input indicates that you have
+ * and eponential behaviour. A value of 10 correspond to at least 3 entity
+ * replacement per byte of input.
+ */
+#define XML_PARSER_NON_LINEAR 10
+
+/*
+ * xmlParserEntityCheck
+ *
+ * Function to check non-linear entity expansion behaviour
+ * This is here to detect and stop exponential linear entity expansion
+ * This is not a limitation of the parser but a safety
+ * boundary feature. It can be disabled with the XML_PARSE_HUGE
+ * parser option.
+ */
+static int
+xmlParserEntityCheck(xmlParserCtxtPtr ctxt, size_t size,
+ xmlEntityPtr ent, size_t replacement)
+{
+ size_t consumed = 0;
+
+ if ((ctxt == NULL) || (ctxt->options & XML_PARSE_HUGE))
+ return (0);
+ if (ctxt->lastError.code == XML_ERR_ENTITY_LOOP)
+ return (1);
+
+ /*
+ * This may look absurd but is needed to detect
+ * entities problems
+ */
+ if ((ent != NULL) && (ent->etype != XML_INTERNAL_PREDEFINED_ENTITY) &&
+ (ent->content != NULL) && (ent->checked == 0)) {
+ unsigned long oldnbent = ctxt->nbentities;
+ xmlChar *rep;
+
+ ent->checked = 1;
+
+ ++ctxt->depth;
+ rep = xmlStringDecodeEntities(ctxt, ent->content,
+ XML_SUBSTITUTE_REF, 0, 0, 0);
+ --ctxt->depth;
+
+ ent->checked = (ctxt->nbentities - oldnbent + 1) * 2;
+ if (rep != NULL) {
+ if (xmlStrchr(rep, '<'))
+ ent->checked |= 1;
+ xmlFree(rep);
+ rep = NULL;
+ }
+ }
+ if (replacement != 0) {
+ if (replacement < XML_MAX_TEXT_LENGTH)
+ return(0);
+
+ /*
+ * If the volume of entity copy reaches 10 times the
+ * amount of parsed data and over the large text threshold
+ * then that's very likely to be an abuse.
+ */
+ if (ctxt->input != NULL) {
+ consumed = ctxt->input->consumed +
+ (ctxt->input->cur - ctxt->input->base);
+ }
+ consumed += ctxt->sizeentities;
+
+ if (replacement < XML_PARSER_NON_LINEAR * consumed)
+ return(0);
+ } else if (size != 0) {
+ /*
+ * Do the check based on the replacement size of the entity
+ */
+ if (size < XML_PARSER_BIG_ENTITY)
+ return(0);
+
+ /*
+ * A limit on the amount of text data reasonably used
+ */
+ if (ctxt->input != NULL) {
+ consumed = ctxt->input->consumed +
+ (ctxt->input->cur - ctxt->input->base);
+ }
+ consumed += ctxt->sizeentities;
+
+ if ((size < XML_PARSER_NON_LINEAR * consumed) &&
+ (ctxt->nbentities * 3 < XML_PARSER_NON_LINEAR * consumed))
+ return (0);
+ } else if (ent != NULL) {
+ /*
+ * use the number of parsed entities in the replacement
+ */
+ size = ent->checked / 2;
+
+ /*
+ * The amount of data parsed counting entities size only once
+ */
+ if (ctxt->input != NULL) {
+ consumed = ctxt->input->consumed +
+ (ctxt->input->cur - ctxt->input->base);
+ }
+ consumed += ctxt->sizeentities;
+
+ /*
+ * Check the density of entities for the amount of data
+ * knowing an entity reference will take at least 3 bytes
+ */
+ if (size * 3 < consumed * XML_PARSER_NON_LINEAR)
+ return (0);
+ } else {
+ /*
+ * strange we got no data for checking
+ */
+ if (((ctxt->lastError.code != XML_ERR_UNDECLARED_ENTITY) &&
+ (ctxt->lastError.code != XML_WAR_UNDECLARED_ENTITY)) ||
+ (ctxt->nbentities <= 10000))
+ return (0);
+ }
+ xmlFatalErr(ctxt, XML_ERR_ENTITY_LOOP, NULL);
+ return (1);
+}
+
+/**
+ * xmlParserMaxDepth:
+ *
+ * arbitrary depth limit for the XML documents that we allow to
+ * process. This is not a limitation of the parser but a safety
+ * boundary feature. It can be disabled with the XML_PARSE_HUGE
+ * parser option.
+ */
+unsigned int xmlParserMaxDepth = 256;
+
+
+
+#define SAX2 1
+#define XML_PARSER_BIG_BUFFER_SIZE 300
+#define XML_PARSER_BUFFER_SIZE 100
+#define SAX_COMPAT_MODE BAD_CAST "SAX compatibility mode document"
+
+/**
+ * XML_PARSER_CHUNK_SIZE
+ *
+ * When calling GROW that's the minimal amount of data
+ * the parser expected to have received. It is not a hard
+ * limit but an optimization when reading strings like Names
+ * It is not strictly needed as long as inputs available characters
+ * are followed by 0, which should be provided by the I/O level
+ */
+#define XML_PARSER_CHUNK_SIZE 100
+
+/*
+ * List of XML prefixed PI allowed by W3C specs
+ */
+
+static const char *xmlW3CPIs[] = {
+ "xml-stylesheet",
+ "xml-model",
+ NULL
+};
+
+
+/* DEPR void xmlParserHandleReference(xmlParserCtxtPtr ctxt); */
+static xmlEntityPtr xmlParseStringPEReference(xmlParserCtxtPtr ctxt,
+ const xmlChar **str);
+
+static xmlParserErrors
+xmlParseExternalEntityPrivate(xmlDocPtr doc, xmlParserCtxtPtr oldctxt,
+ xmlSAXHandlerPtr sax,
+ void *user_data, int depth, const xmlChar *URL,
+ const xmlChar *ID, xmlNodePtr *list);
+
+static int
+xmlCtxtUseOptionsInternal(xmlParserCtxtPtr ctxt, int options,
+ const char *encoding);
+#ifdef LIBXML_LEGACY_ENABLED
+static void
+xmlAddEntityReference(xmlEntityPtr ent, xmlNodePtr firstNode,
+ xmlNodePtr lastNode);
+#endif /* LIBXML_LEGACY_ENABLED */
+
+static xmlParserErrors
+xmlParseBalancedChunkMemoryInternal(xmlParserCtxtPtr oldctxt,
+ const xmlChar *string, void *user_data, xmlNodePtr *lst);
+
+static int
+xmlLoadEntityContent(xmlParserCtxtPtr ctxt, xmlEntityPtr entity);
+
+/************************************************************************
+ * *
+ * Some factorized error routines *
+ * *
+ ************************************************************************/
+
+/**
+ * xmlErrAttributeDup:
+ * @ctxt: an XML parser context
+ * @prefix: the attribute prefix
+ * @localname: the attribute localname
+ *
+ * Handle a redefinition of attribute error
+ */
+static void
+xmlErrAttributeDup(xmlParserCtxtPtr ctxt, const xmlChar * prefix,
+ const xmlChar * localname)
+{
+ if ((ctxt != NULL) && (ctxt->disableSAX != 0) &&
+ (ctxt->instate == XML_PARSER_EOF))
+ return;
+ if (ctxt != NULL)
+ ctxt->errNo = XML_ERR_ATTRIBUTE_REDEFINED;
+
+ if (prefix == NULL)
+ __xmlRaiseError(NULL, NULL, NULL, ctxt, NULL, XML_FROM_PARSER,
+ XML_ERR_ATTRIBUTE_REDEFINED, XML_ERR_FATAL, NULL, 0,
+ (const char *) localname, NULL, NULL, 0, 0,
+ "Attribute %s redefined\n", localname);
+ else
+ __xmlRaiseError(NULL, NULL, NULL, ctxt, NULL, XML_FROM_PARSER,
+ XML_ERR_ATTRIBUTE_REDEFINED, XML_ERR_FATAL, NULL, 0,
+ (const char *) prefix, (const char *) localname,
+ NULL, 0, 0, "Attribute %s:%s redefined\n", prefix,
+ localname);
+ if (ctxt != NULL) {
+ ctxt->wellFormed = 0;
+ if (ctxt->recovery == 0)
+ ctxt->disableSAX = 1;
+ }
+}
+
+/**
+ * xmlFatalErr:
+ * @ctxt: an XML parser context
+ * @error: the error number
+ * @extra: extra information string
+ *
+ * Handle a fatal parser error, i.e. violating Well-Formedness constraints
+ */
+static void
+xmlFatalErr(xmlParserCtxtPtr ctxt, xmlParserErrors error, const char *info)
+{
+ const char *errmsg;
+ char errstr[129] = "";
+
+ if ((ctxt != NULL) && (ctxt->disableSAX != 0) &&
+ (ctxt->instate == XML_PARSER_EOF))
+ return;
+ switch (error) {
+ case XML_ERR_INVALID_HEX_CHARREF:
+ errmsg = "CharRef: invalid hexadecimal value";
+ break;
+ case XML_ERR_INVALID_DEC_CHARREF:
+ errmsg = "CharRef: invalid decimal value";
+ break;
+ case XML_ERR_INVALID_CHARREF:
+ errmsg = "CharRef: invalid value";
+ break;
+ case XML_ERR_INTERNAL_ERROR:
+ errmsg = "internal error";
+ break;
+ case XML_ERR_PEREF_AT_EOF:
+ errmsg = "PEReference at end of document";
+ break;
+ case XML_ERR_PEREF_IN_PROLOG:
+ errmsg = "PEReference in prolog";
+ break;
+ case XML_ERR_PEREF_IN_EPILOG:
+ errmsg = "PEReference in epilog";
+ break;
+ case XML_ERR_PEREF_NO_NAME:
+ errmsg = "PEReference: no name";
+ break;
+ case XML_ERR_PEREF_SEMICOL_MISSING:
+ errmsg = "PEReference: expecting ';'";
+ break;
+ case XML_ERR_ENTITY_LOOP:
+ errmsg = "Detected an entity reference loop";
+ break;
+ case XML_ERR_ENTITY_NOT_STARTED:
+ errmsg = "EntityValue: \" or ' expected";
+ break;
+ case XML_ERR_ENTITY_PE_INTERNAL:
+ errmsg = "PEReferences forbidden in internal subset";
+ break;
+ case XML_ERR_ENTITY_NOT_FINISHED:
+ errmsg = "EntityValue: \" or ' expected";
+ break;
+ case XML_ERR_ATTRIBUTE_NOT_STARTED:
+ errmsg = "AttValue: \" or ' expected";
+ break;
+ case XML_ERR_LT_IN_ATTRIBUTE:
+ errmsg = "Unescaped '<' not allowed in attributes values";
+ break;
+ case XML_ERR_LITERAL_NOT_STARTED:
+ errmsg = "SystemLiteral \" or ' expected";
+ break;
+ case XML_ERR_LITERAL_NOT_FINISHED:
+ errmsg = "Unfinished System or Public ID \" or ' expected";
+ break;
+ case XML_ERR_MISPLACED_CDATA_END:
+ errmsg = "Sequence ']]>' not allowed in content";
+ break;
+ case XML_ERR_URI_REQUIRED:
+ errmsg = "SYSTEM or PUBLIC, the URI is missing";
+ break;
+ case XML_ERR_PUBID_REQUIRED:
+ errmsg = "PUBLIC, the Public Identifier is missing";
+ break;
+ case XML_ERR_HYPHEN_IN_COMMENT:
+ errmsg = "Comment must not contain '--' (double-hyphen)";
+ break;
+ case XML_ERR_PI_NOT_STARTED:
+ errmsg = "xmlParsePI : no target name";
+ break;
+ case XML_ERR_RESERVED_XML_NAME:
+ errmsg = "Invalid PI name";
+ break;
+ case XML_ERR_NOTATION_NOT_STARTED:
+ errmsg = "NOTATION: Name expected here";
+ break;
+ case XML_ERR_NOTATION_NOT_FINISHED:
+ errmsg = "'>' required to close NOTATION declaration";
+ break;
+ case XML_ERR_VALUE_REQUIRED:
+ errmsg = "Entity value required";
+ break;
+ case XML_ERR_URI_FRAGMENT:
+ errmsg = "Fragment not allowed";
+ break;
+ case XML_ERR_ATTLIST_NOT_STARTED:
+ errmsg = "'(' required to start ATTLIST enumeration";
+ break;
+ case XML_ERR_NMTOKEN_REQUIRED:
+ errmsg = "NmToken expected in ATTLIST enumeration";
+ break;
+ case XML_ERR_ATTLIST_NOT_FINISHED:
+ errmsg = "')' required to finish ATTLIST enumeration";
+ break;
+ case XML_ERR_MIXED_NOT_STARTED:
+ errmsg = "MixedContentDecl : '|' or ')*' expected";
+ break;
+ case XML_ERR_PCDATA_REQUIRED:
+ errmsg = "MixedContentDecl : '#PCDATA' expected";
+ break;
+ case XML_ERR_ELEMCONTENT_NOT_STARTED:
+ errmsg = "ContentDecl : Name or '(' expected";
+ break;
+ case XML_ERR_ELEMCONTENT_NOT_FINISHED:
+ errmsg = "ContentDecl : ',' '|' or ')' expected";
+ break;
+ case XML_ERR_PEREF_IN_INT_SUBSET:
+ errmsg =
+ "PEReference: forbidden within markup decl in internal subset";
+ break;
+ case XML_ERR_GT_REQUIRED:
+ errmsg = "expected '>'";
+ break;
+ case XML_ERR_CONDSEC_INVALID:
+ errmsg = "XML conditional section '[' expected";
+ break;
+ case XML_ERR_EXT_SUBSET_NOT_FINISHED:
+ errmsg = "Content error in the external subset";
+ break;
+ case XML_ERR_CONDSEC_INVALID_KEYWORD:
+ errmsg =
+ "conditional section INCLUDE or IGNORE keyword expected";
+ break;
+ case XML_ERR_CONDSEC_NOT_FINISHED:
+ errmsg = "XML conditional section not closed";
+ break;
+ case XML_ERR_XMLDECL_NOT_STARTED:
+ errmsg = "Text declaration '<?xml' required";
+ break;
+ case XML_ERR_XMLDECL_NOT_FINISHED:
+ errmsg = "parsing XML declaration: '?>' expected";
+ break;
+ case XML_ERR_EXT_ENTITY_STANDALONE:
+ errmsg = "external parsed entities cannot be standalone";
+ break;
+ case XML_ERR_ENTITYREF_SEMICOL_MISSING:
+ errmsg = "EntityRef: expecting ';'";
+ break;
+ case XML_ERR_DOCTYPE_NOT_FINISHED:
+ errmsg = "DOCTYPE improperly terminated";
+ break;
+ case XML_ERR_LTSLASH_REQUIRED:
+ errmsg = "EndTag: '</' not found";
+ break;
+ case XML_ERR_EQUAL_REQUIRED:
+ errmsg = "expected '='";
+ break;
+ case XML_ERR_STRING_NOT_CLOSED:
+ errmsg = "String not closed expecting \" or '";
+ break;
+ case XML_ERR_STRING_NOT_STARTED:
+ errmsg = "String not started expecting ' or \"";
+ break;
+ case XML_ERR_ENCODING_NAME:
+ errmsg = "Invalid XML encoding name";
+ break;
+ case XML_ERR_STANDALONE_VALUE:
+ errmsg = "standalone accepts only 'yes' or 'no'";
+ break;
+ case XML_ERR_DOCUMENT_EMPTY:
+ errmsg = "Document is empty";
+ break;
+ case XML_ERR_DOCUMENT_END:
+ errmsg = "Extra content at the end of the document";
+ break;
+ case XML_ERR_NOT_WELL_BALANCED:
+ errmsg = "chunk is not well balanced";
+ break;
+ case XML_ERR_EXTRA_CONTENT:
+ errmsg = "extra content at the end of well balanced chunk";
+ break;
+ case XML_ERR_VERSION_MISSING:
+ errmsg = "Malformed declaration expecting version";
+ break;
+ case XML_ERR_NAME_TOO_LONG:
+ errmsg = "Name too long use XML_PARSE_HUGE option";
+ break;
+#if 0
+ case:
+ errmsg = "";
+ break;
+#endif
+ default:
+ errmsg = "Unregistered error message";
+ }
+ if (info == NULL)
+ snprintf(errstr, 128, "%s\n", errmsg);
+ else
+ snprintf(errstr, 128, "%s: %%s\n", errmsg);
+ if (ctxt != NULL)
+ ctxt->errNo = error;
+ __xmlRaiseError(NULL, NULL, NULL, ctxt, NULL, XML_FROM_PARSER, error,
+ XML_ERR_FATAL, NULL, 0, info, NULL, NULL, 0, 0, &errstr[0],
+ info);
+ if (ctxt != NULL) {
+ ctxt->wellFormed = 0;
+ if (ctxt->recovery == 0)
+ ctxt->disableSAX = 1;
+ }
+}
+
+/**
+ * xmlFatalErrMsg:
+ * @ctxt: an XML parser context
+ * @error: the error number
+ * @msg: the error message
+ *
+ * Handle a fatal parser error, i.e. violating Well-Formedness constraints
+ */
+static void
+xmlFatalErrMsg(xmlParserCtxtPtr ctxt, xmlParserErrors error,
+ const char *msg)
+{
+ if ((ctxt != NULL) && (ctxt->disableSAX != 0) &&
+ (ctxt->instate == XML_PARSER_EOF))
+ return;
+ if (ctxt != NULL)
+ ctxt->errNo = error;
+ __xmlRaiseError(NULL, NULL, NULL, ctxt, NULL, XML_FROM_PARSER, error,
+ XML_ERR_FATAL, NULL, 0, NULL, NULL, NULL, 0, 0, "%s", msg);
+ if (ctxt != NULL) {
+ ctxt->wellFormed = 0;
+ if (ctxt->recovery == 0)
+ ctxt->disableSAX = 1;
+ }
+}
+
+/**
+ * xmlWarningMsg:
+ * @ctxt: an XML parser context
+ * @error: the error number
+ * @msg: the error message
+ * @str1: extra data
+ * @str2: extra data
+ *
+ * Handle a warning.
+ */
+static void
+xmlWarningMsg(xmlParserCtxtPtr ctxt, xmlParserErrors error,
+ const char *msg, const xmlChar *str1, const xmlChar *str2)
+{
+ xmlStructuredErrorFunc schannel = NULL;
+
+ if ((ctxt != NULL) && (ctxt->disableSAX != 0) &&
+ (ctxt->instate == XML_PARSER_EOF))
+ return;
+ if ((ctxt != NULL) && (ctxt->sax != NULL) &&
+ (ctxt->sax->initialized == XML_SAX2_MAGIC))
+ schannel = ctxt->sax->serror;
+ if (ctxt != NULL) {
+ __xmlRaiseError(schannel,
+ (ctxt->sax) ? ctxt->sax->warning : NULL,
+ ctxt->userData,
+ ctxt, NULL, XML_FROM_PARSER, error,
+ XML_ERR_WARNING, NULL, 0,
+ (const char *) str1, (const char *) str2, NULL, 0, 0,
+ msg, (const char *) str1, (const char *) str2);
+ } else {
+ __xmlRaiseError(schannel, NULL, NULL,
+ ctxt, NULL, XML_FROM_PARSER, error,
+ XML_ERR_WARNING, NULL, 0,
+ (const char *) str1, (const char *) str2, NULL, 0, 0,
+ msg, (const char *) str1, (const char *) str2);
+ }
+}
+
+/**
+ * xmlValidityError:
+ * @ctxt: an XML parser context
+ * @error: the error number
+ * @msg: the error message
+ * @str1: extra data
+ *
+ * Handle a validity error.
+ */
+static void
+xmlValidityError(xmlParserCtxtPtr ctxt, xmlParserErrors error,
+ const char *msg, const xmlChar *str1, const xmlChar *str2)
+{
+ xmlStructuredErrorFunc schannel = NULL;
+
+ if ((ctxt != NULL) && (ctxt->disableSAX != 0) &&
+ (ctxt->instate == XML_PARSER_EOF))
+ return;
+ if (ctxt != NULL) {
+ ctxt->errNo = error;
+ if ((ctxt->sax != NULL) && (ctxt->sax->initialized == XML_SAX2_MAGIC))
+ schannel = ctxt->sax->serror;
+ }
+ if (ctxt != NULL) {
+ __xmlRaiseError(schannel,
+ ctxt->vctxt.error, ctxt->vctxt.userData,
+ ctxt, NULL, XML_FROM_DTD, error,
+ XML_ERR_ERROR, NULL, 0, (const char *) str1,
+ (const char *) str2, NULL, 0, 0,
+ msg, (const char *) str1, (const char *) str2);
+ ctxt->valid = 0;
+ } else {
+ __xmlRaiseError(schannel, NULL, NULL,
+ ctxt, NULL, XML_FROM_DTD, error,
+ XML_ERR_ERROR, NULL, 0, (const char *) str1,
+ (const char *) str2, NULL, 0, 0,
+ msg, (const char *) str1, (const char *) str2);
+ }
+}
+
+/**
+ * xmlFatalErrMsgInt:
+ * @ctxt: an XML parser context
+ * @error: the error number
+ * @msg: the error message
+ * @val: an integer value
+ *
+ * Handle a fatal parser error, i.e. violating Well-Formedness constraints
+ */
+static void
+xmlFatalErrMsgInt(xmlParserCtxtPtr ctxt, xmlParserErrors error,
+ const char *msg, int val)
+{
+ if ((ctxt != NULL) && (ctxt->disableSAX != 0) &&
+ (ctxt->instate == XML_PARSER_EOF))
+ return;
+ if (ctxt != NULL)
+ ctxt->errNo = error;
+ __xmlRaiseError(NULL, NULL, NULL,
+ ctxt, NULL, XML_FROM_PARSER, error, XML_ERR_FATAL,
+ NULL, 0, NULL, NULL, NULL, val, 0, msg, val);
+ if (ctxt != NULL) {
+ ctxt->wellFormed = 0;
+ if (ctxt->recovery == 0)
+ ctxt->disableSAX = 1;
+ }
+}
+
+/**
+ * xmlFatalErrMsgStrIntStr:
+ * @ctxt: an XML parser context
+ * @error: the error number
+ * @msg: the error message
+ * @str1: an string info
+ * @val: an integer value
+ * @str2: an string info
+ *
+ * Handle a fatal parser error, i.e. violating Well-Formedness constraints
+ */
+static void
+xmlFatalErrMsgStrIntStr(xmlParserCtxtPtr ctxt, xmlParserErrors error,
+ const char *msg, const xmlChar *str1, int val,
+ const xmlChar *str2)
+{
+ if ((ctxt != NULL) && (ctxt->disableSAX != 0) &&
+ (ctxt->instate == XML_PARSER_EOF))
+ return;
+ if (ctxt != NULL)
+ ctxt->errNo = error;
+ __xmlRaiseError(NULL, NULL, NULL,
+ ctxt, NULL, XML_FROM_PARSER, error, XML_ERR_FATAL,
+ NULL, 0, (const char *) str1, (const char *) str2,
+ NULL, val, 0, msg, str1, val, str2);
+ if (ctxt != NULL) {
+ ctxt->wellFormed = 0;
+ if (ctxt->recovery == 0)
+ ctxt->disableSAX = 1;
+ }
+}
+
+/**
+ * xmlFatalErrMsgStr:
+ * @ctxt: an XML parser context
+ * @error: the error number
+ * @msg: the error message
+ * @val: a string value
+ *
+ * Handle a fatal parser error, i.e. violating Well-Formedness constraints
+ */
+static void
+xmlFatalErrMsgStr(xmlParserCtxtPtr ctxt, xmlParserErrors error,
+ const char *msg, const xmlChar * val)
+{
+ if ((ctxt != NULL) && (ctxt->disableSAX != 0) &&
+ (ctxt->instate == XML_PARSER_EOF))
+ return;
+ if (ctxt != NULL)
+ ctxt->errNo = error;
+ __xmlRaiseError(NULL, NULL, NULL, ctxt, NULL,
+ XML_FROM_PARSER, error, XML_ERR_FATAL,
+ NULL, 0, (const char *) val, NULL, NULL, 0, 0, msg,
+ val);
+ if (ctxt != NULL) {
+ ctxt->wellFormed = 0;
+ if (ctxt->recovery == 0)
+ ctxt->disableSAX = 1;
+ }
+}
+
+/**
+ * xmlErrMsgStr:
+ * @ctxt: an XML parser context
+ * @error: the error number
+ * @msg: the error message
+ * @val: a string value
+ *
+ * Handle a non fatal parser error
+ */
+static void
+xmlErrMsgStr(xmlParserCtxtPtr ctxt, xmlParserErrors error,
+ const char *msg, const xmlChar * val)
+{
+ if ((ctxt != NULL) && (ctxt->disableSAX != 0) &&
+ (ctxt->instate == XML_PARSER_EOF))
+ return;
+ if (ctxt != NULL)
+ ctxt->errNo = error;
+ __xmlRaiseError(NULL, NULL, NULL, ctxt, NULL,
+ XML_FROM_PARSER, error, XML_ERR_ERROR,
+ NULL, 0, (const char *) val, NULL, NULL, 0, 0, msg,
+ val);
+}
+
+/**
+ * xmlNsErr:
+ * @ctxt: an XML parser context
+ * @error: the error number
+ * @msg: the message
+ * @info1: extra information string
+ * @info2: extra information string
+ *
+ * Handle a fatal parser error, i.e. violating Well-Formedness constraints
+ */
+static void
+xmlNsErr(xmlParserCtxtPtr ctxt, xmlParserErrors error,
+ const char *msg,
+ const xmlChar * info1, const xmlChar * info2,
+ const xmlChar * info3)
+{
+ if ((ctxt != NULL) && (ctxt->disableSAX != 0) &&
+ (ctxt->instate == XML_PARSER_EOF))
+ return;
+ if (ctxt != NULL)
+ ctxt->errNo = error;
+ __xmlRaiseError(NULL, NULL, NULL, ctxt, NULL, XML_FROM_NAMESPACE, error,
+ XML_ERR_ERROR, NULL, 0, (const char *) info1,
+ (const char *) info2, (const char *) info3, 0, 0, msg,
+ info1, info2, info3);
+ if (ctxt != NULL)
+ ctxt->nsWellFormed = 0;
+}
+
+/**
+ * xmlNsWarn
+ * @ctxt: an XML parser context
+ * @error: the error number
+ * @msg: the message
+ * @info1: extra information string
+ * @info2: extra information string
+ *
+ * Handle a namespace warning error
+ */
+static void
+xmlNsWarn(xmlParserCtxtPtr ctxt, xmlParserErrors error,
+ const char *msg,
+ const xmlChar * info1, const xmlChar * info2,
+ const xmlChar * info3)
+{
+ if ((ctxt != NULL) && (ctxt->disableSAX != 0) &&
+ (ctxt->instate == XML_PARSER_EOF))
+ return;
+ __xmlRaiseError(NULL, NULL, NULL, ctxt, NULL, XML_FROM_NAMESPACE, error,
+ XML_ERR_WARNING, NULL, 0, (const char *) info1,
+ (const char *) info2, (const char *) info3, 0, 0, msg,
+ info1, info2, info3);
+}
+
+/************************************************************************
+ * *
+ * Library wide options *
+ * *
+ ************************************************************************/
+
+/**
+ * xmlHasFeature:
+ * @feature: the feature to be examined
+ *
+ * Examines if the library has been compiled with a given feature.
+ *
+ * Returns a non-zero value if the feature exist, otherwise zero.
+ * Returns zero (0) if the feature does not exist or an unknown
+ * unknown feature is requested, non-zero otherwise.
+ */
+int
+xmlHasFeature(xmlFeature feature)
+{
+ switch (feature) {
+ case XML_WITH_THREAD:
+#ifdef LIBXML_THREAD_ENABLED
+ return(1);
+#else
+ return(0);
+#endif
+ case XML_WITH_TREE:
+#ifdef LIBXML_TREE_ENABLED
+ return(1);
+#else
+ return(0);
+#endif
+ case XML_WITH_OUTPUT:
+#ifdef LIBXML_OUTPUT_ENABLED
+ return(1);
+#else
+ return(0);
+#endif
+ case XML_WITH_PUSH:
+#ifdef LIBXML_PUSH_ENABLED
+ return(1);
+#else
+ return(0);
+#endif
+ case XML_WITH_READER:
+#ifdef LIBXML_READER_ENABLED
+ return(1);
+#else
+ return(0);
+#endif
+ case XML_WITH_PATTERN:
+#ifdef LIBXML_PATTERN_ENABLED
+ return(1);
+#else
+ return(0);
+#endif
+ case XML_WITH_WRITER:
+#ifdef LIBXML_WRITER_ENABLED
+ return(1);
+#else
+ return(0);
+#endif
+ case XML_WITH_SAX1:
+#ifdef LIBXML_SAX1_ENABLED
+ return(1);
+#else
+ return(0);
+#endif
+ case XML_WITH_FTP:
+#ifdef LIBXML_FTP_ENABLED
+ return(1);
+#else
+ return(0);
+#endif
+ case XML_WITH_HTTP:
+#ifdef LIBXML_HTTP_ENABLED
+ return(1);
+#else
+ return(0);
+#endif
+ case XML_WITH_VALID:
+#ifdef LIBXML_VALID_ENABLED
+ return(1);
+#else
+ return(0);
+#endif
+ case XML_WITH_HTML:
+#ifdef LIBXML_HTML_ENABLED
+ return(1);
+#else
+ return(0);
+#endif
+ case XML_WITH_LEGACY:
+#ifdef LIBXML_LEGACY_ENABLED
+ return(1);
+#else
+ return(0);
+#endif
+ case XML_WITH_C14N:
+#ifdef LIBXML_C14N_ENABLED
+ return(1);
+#else
+ return(0);
+#endif
+ case XML_WITH_CATALOG:
+#ifdef LIBXML_CATALOG_ENABLED
+ return(1);
+#else
+ return(0);
+#endif
+ case XML_WITH_XPATH:
+#ifdef LIBXML_XPATH_ENABLED
+ return(1);
+#else
+ return(0);
+#endif
+ case XML_WITH_XPTR:
+#ifdef LIBXML_XPTR_ENABLED
+ return(1);
+#else
+ return(0);
+#endif
+ case XML_WITH_XINCLUDE:
+#ifdef LIBXML_XINCLUDE_ENABLED
+ return(1);
+#else
+ return(0);
+#endif
+ case XML_WITH_ICONV:
+#ifdef LIBXML_ICONV_ENABLED
+ return(1);
+#else
+ return(0);
+#endif
+ case XML_WITH_ISO8859X:
+#ifdef LIBXML_ISO8859X_ENABLED
+ return(1);
+#else
+ return(0);
+#endif
+ case XML_WITH_UNICODE:
+#ifdef LIBXML_UNICODE_ENABLED
+ return(1);
+#else
+ return(0);
+#endif
+ case XML_WITH_REGEXP:
+#ifdef LIBXML_REGEXP_ENABLED
+ return(1);
+#else
+ return(0);
+#endif
+ case XML_WITH_AUTOMATA:
+#ifdef LIBXML_AUTOMATA_ENABLED
+ return(1);
+#else
+ return(0);
+#endif
+ case XML_WITH_EXPR:
+#ifdef LIBXML_EXPR_ENABLED
+ return(1);
+#else
+ return(0);
+#endif
+ case XML_WITH_SCHEMAS:
+#ifdef LIBXML_SCHEMAS_ENABLED
+ return(1);
+#else
+ return(0);
+#endif
+ case XML_WITH_SCHEMATRON:
+#ifdef LIBXML_SCHEMATRON_ENABLED
+ return(1);
+#else
+ return(0);
+#endif
+ case XML_WITH_MODULES:
+#ifdef LIBXML_MODULES_ENABLED
+ return(1);
+#else
+ return(0);
+#endif
+ case XML_WITH_DEBUG:
+#ifdef LIBXML_DEBUG_ENABLED
+ return(1);
+#else
+ return(0);
+#endif
+ case XML_WITH_DEBUG_MEM:
+#ifdef DEBUG_MEMORY_LOCATION
+ return(1);
+#else
+ return(0);
+#endif
+ case XML_WITH_DEBUG_RUN:
+#ifdef LIBXML_DEBUG_RUNTIME
+ return(1);
+#else
+ return(0);
+#endif
+ case XML_WITH_ZLIB:
+#ifdef LIBXML_ZLIB_ENABLED
+ return(1);
+#else
+ return(0);
+#endif
+ case XML_WITH_LZMA:
+#ifdef LIBXML_LZMA_ENABLED
+ return(1);
+#else
+ return(0);
+#endif
+ case XML_WITH_ICU:
+#ifdef LIBXML_ICU_ENABLED
+ return(1);
+#else
+ return(0);
+#endif
+ default:
+ break;
+ }
+ return(0);
+}
+
+/************************************************************************
+ * *
+ * SAX2 defaulted attributes handling *
+ * *
+ ************************************************************************/
+
+/**
+ * xmlDetectSAX2:
+ * @ctxt: an XML parser context
+ *
+ * Do the SAX2 detection and specific intialization
+ */
+static void
+xmlDetectSAX2(xmlParserCtxtPtr ctxt) {
+ if (ctxt == NULL) return;
+#ifdef LIBXML_SAX1_ENABLED
+ if ((ctxt->sax) && (ctxt->sax->initialized == XML_SAX2_MAGIC) &&
+ ((ctxt->sax->startElementNs != NULL) ||
+ (ctxt->sax->endElementNs != NULL))) ctxt->sax2 = 1;
+#else
+ ctxt->sax2 = 1;
+#endif /* LIBXML_SAX1_ENABLED */
+
+ ctxt->str_xml = xmlDictLookup(ctxt->dict, BAD_CAST "xml", 3);
+ ctxt->str_xmlns = xmlDictLookup(ctxt->dict, BAD_CAST "xmlns", 5);
+ ctxt->str_xml_ns = xmlDictLookup(ctxt->dict, XML_XML_NAMESPACE, 36);
+ if ((ctxt->str_xml==NULL) || (ctxt->str_xmlns==NULL) ||
+ (ctxt->str_xml_ns == NULL)) {
+ xmlErrMemory(ctxt, NULL);
+ }
+}
+
+typedef struct _xmlDefAttrs xmlDefAttrs;
+typedef xmlDefAttrs *xmlDefAttrsPtr;
+struct _xmlDefAttrs {
+ int nbAttrs; /* number of defaulted attributes on that element */
+ int maxAttrs; /* the size of the array */
+ const xmlChar *values[5]; /* array of localname/prefix/values/external */
+};
+
+/**
+ * xmlAttrNormalizeSpace:
+ * @src: the source string
+ * @dst: the target string
+ *
+ * Normalize the space in non CDATA attribute values:
+ * If the attribute type is not CDATA, then the XML processor MUST further
+ * process the normalized attribute value by discarding any leading and
+ * trailing space (#x20) characters, and by replacing sequences of space
+ * (#x20) characters by a single space (#x20) character.
+ * Note that the size of dst need to be at least src, and if one doesn't need
+ * to preserve dst (and it doesn't come from a dictionary or read-only) then
+ * passing src as dst is just fine.
+ *
+ * Returns a pointer to the normalized value (dst) or NULL if no conversion
+ * is needed.
+ */
+static xmlChar *
+xmlAttrNormalizeSpace(const xmlChar *src, xmlChar *dst)
+{
+ if ((src == NULL) || (dst == NULL))
+ return(NULL);
+
+ while (*src == 0x20) src++;
+ while (*src != 0) {
+ if (*src == 0x20) {
+ while (*src == 0x20) src++;
+ if (*src != 0)
+ *dst++ = 0x20;
+ } else {
+ *dst++ = *src++;
+ }
+ }
+ *dst = 0;
+ if (dst == src)
+ return(NULL);
+ return(dst);
+}
+
+/**
+ * xmlAttrNormalizeSpace2:
+ * @src: the source string
+ *
+ * Normalize the space in non CDATA attribute values, a slightly more complex
+ * front end to avoid allocation problems when running on attribute values
+ * coming from the input.
+ *
+ * Returns a pointer to the normalized value (dst) or NULL if no conversion
+ * is needed.
+ */
+static const xmlChar *
+xmlAttrNormalizeSpace2(xmlParserCtxtPtr ctxt, xmlChar *src, int *len)
+{
+ int i;
+ int remove_head = 0;
+ int need_realloc = 0;
+ const xmlChar *cur;
+
+ if ((ctxt == NULL) || (src == NULL) || (len == NULL))
+ return(NULL);
+ i = *len;
+ if (i <= 0)
+ return(NULL);
+
+ cur = src;
+ while (*cur == 0x20) {
+ cur++;
+ remove_head++;
+ }
+ while (*cur != 0) {
+ if (*cur == 0x20) {
+ cur++;
+ if ((*cur == 0x20) || (*cur == 0)) {
+ need_realloc = 1;
+ break;
+ }
+ } else
+ cur++;
+ }
+ if (need_realloc) {
+ xmlChar *ret;
+
+ ret = xmlStrndup(src + remove_head, i - remove_head + 1);
+ if (ret == NULL) {
+ xmlErrMemory(ctxt, NULL);
+ return(NULL);
+ }
+ xmlAttrNormalizeSpace(ret, ret);
+ *len = (int) strlen((const char *)ret);
+ return(ret);
+ } else if (remove_head) {
+ *len -= remove_head;
+ memmove(src, src + remove_head, 1 + *len);
+ return(src);
+ }
+ return(NULL);
+}
+
+/**
+ * xmlAddDefAttrs:
+ * @ctxt: an XML parser context
+ * @fullname: the element fullname
+ * @fullattr: the attribute fullname
+ * @value: the attribute value
+ *
+ * Add a defaulted attribute for an element
+ */
+static void
+xmlAddDefAttrs(xmlParserCtxtPtr ctxt,
+ const xmlChar *fullname,
+ const xmlChar *fullattr,
+ const xmlChar *value) {
+ xmlDefAttrsPtr defaults;
+ int len;
+ const xmlChar *name;
+ const xmlChar *prefix;
+
+ /*
+ * Allows to detect attribute redefinitions
+ */
+ if (ctxt->attsSpecial != NULL) {
+ if (xmlHashLookup2(ctxt->attsSpecial, fullname, fullattr) != NULL)
+ return;
+ }
+
+ if (ctxt->attsDefault == NULL) {
+ ctxt->attsDefault = xmlHashCreateDict(10, ctxt->dict);
+ if (ctxt->attsDefault == NULL)
+ goto mem_error;
+ }
+
+ /*
+ * split the element name into prefix:localname , the string found
+ * are within the DTD and then not associated to namespace names.
+ */
+ name = xmlSplitQName3(fullname, &len);
+ if (name == NULL) {
+ name = xmlDictLookup(ctxt->dict, fullname, -1);
+ prefix = NULL;
+ } else {
+ name = xmlDictLookup(ctxt->dict, name, -1);
+ prefix = xmlDictLookup(ctxt->dict, fullname, len);
+ }
+
+ /*
+ * make sure there is some storage
+ */
+ defaults = xmlHashLookup2(ctxt->attsDefault, name, prefix);
+ if (defaults == NULL) {
+ defaults = (xmlDefAttrsPtr) xmlMalloc(sizeof(xmlDefAttrs) +
+ (4 * 5) * sizeof(const xmlChar *));
+ if (defaults == NULL)
+ goto mem_error;
+ defaults->nbAttrs = 0;
+ defaults->maxAttrs = 4;
+ if (xmlHashUpdateEntry2(ctxt->attsDefault, name, prefix,
+ defaults, NULL) < 0) {
+ xmlFree(defaults);
+ goto mem_error;
+ }
+ } else if (defaults->nbAttrs >= defaults->maxAttrs) {
+ xmlDefAttrsPtr temp;
+
+ temp = (xmlDefAttrsPtr) xmlRealloc(defaults, sizeof(xmlDefAttrs) +
+ (2 * defaults->maxAttrs * 5) * sizeof(const xmlChar *));
+ if (temp == NULL)
+ goto mem_error;
+ defaults = temp;
+ defaults->maxAttrs *= 2;
+ if (xmlHashUpdateEntry2(ctxt->attsDefault, name, prefix,
+ defaults, NULL) < 0) {
+ xmlFree(defaults);
+ goto mem_error;
+ }
+ }
+
+ /*
+ * Split the element name into prefix:localname , the string found
+ * are within the DTD and hen not associated to namespace names.
+ */
+ name = xmlSplitQName3(fullattr, &len);
+ if (name == NULL) {
+ name = xmlDictLookup(ctxt->dict, fullattr, -1);
+ prefix = NULL;
+ } else {
+ name = xmlDictLookup(ctxt->dict, name, -1);
+ prefix = xmlDictLookup(ctxt->dict, fullattr, len);
+ }
+
+ defaults->values[5 * defaults->nbAttrs] = name;
+ defaults->values[5 * defaults->nbAttrs + 1] = prefix;
+ /* intern the string and precompute the end */
+ len = xmlStrlen(value);
+ value = xmlDictLookup(ctxt->dict, value, len);
+ defaults->values[5 * defaults->nbAttrs + 2] = value;
+ defaults->values[5 * defaults->nbAttrs + 3] = value + len;
+ if (ctxt->external)
+ defaults->values[5 * defaults->nbAttrs + 4] = BAD_CAST "external";
+ else
+ defaults->values[5 * defaults->nbAttrs + 4] = NULL;
+ defaults->nbAttrs++;
+
+ return;
+
+mem_error:
+ xmlErrMemory(ctxt, NULL);
+ return;
+}
+
+/**
+ * xmlAddSpecialAttr:
+ * @ctxt: an XML parser context
+ * @fullname: the element fullname
+ * @fullattr: the attribute fullname
+ * @type: the attribute type
+ *
+ * Register this attribute type
+ */
+static void
+xmlAddSpecialAttr(xmlParserCtxtPtr ctxt,
+ const xmlChar *fullname,
+ const xmlChar *fullattr,
+ int type)
+{
+ if (ctxt->attsSpecial == NULL) {
+ ctxt->attsSpecial = xmlHashCreateDict(10, ctxt->dict);
+ if (ctxt->attsSpecial == NULL)
+ goto mem_error;
+ }
+
+ if (xmlHashLookup2(ctxt->attsSpecial, fullname, fullattr) != NULL)
+ return;
+
+ xmlHashAddEntry2(ctxt->attsSpecial, fullname, fullattr,
+ (void *) (long) type);
+ return;
+
+mem_error:
+ xmlErrMemory(ctxt, NULL);
+ return;
+}
+
+/**
+ * xmlCleanSpecialAttrCallback:
+ *
+ * Removes CDATA attributes from the special attribute table
+ */
+static void
+xmlCleanSpecialAttrCallback(void *payload, void *data,
+ const xmlChar *fullname, const xmlChar *fullattr,
+ const xmlChar *unused ATTRIBUTE_UNUSED) {
+ xmlParserCtxtPtr ctxt = (xmlParserCtxtPtr) data;
+
+ if (((long) payload) == XML_ATTRIBUTE_CDATA) {
+ xmlHashRemoveEntry2(ctxt->attsSpecial, fullname, fullattr, NULL);
+ }
+}
+
+/**
+ * xmlCleanSpecialAttr:
+ * @ctxt: an XML parser context
+ *
+ * Trim the list of attributes defined to remove all those of type
+ * CDATA as they are not special. This call should be done when finishing
+ * to parse the DTD and before starting to parse the document root.
+ */
+static void
+xmlCleanSpecialAttr(xmlParserCtxtPtr ctxt)
+{
+ if (ctxt->attsSpecial == NULL)
+ return;
+
+ xmlHashScanFull(ctxt->attsSpecial, xmlCleanSpecialAttrCallback, ctxt);
+
+ if (xmlHashSize(ctxt->attsSpecial) == 0) {
+ xmlHashFree(ctxt->attsSpecial, NULL);
+ ctxt->attsSpecial = NULL;
+ }
+ return;
+}
+
+/**
+ * xmlCheckLanguageID:
+ * @lang: pointer to the string value
+ *
+ * Checks that the value conforms to the LanguageID production:
+ *
+ * NOTE: this is somewhat deprecated, those productions were removed from
+ * the XML Second edition.
+ *
+ * [33] LanguageID ::= Langcode ('-' Subcode)*
+ * [34] Langcode ::= ISO639Code | IanaCode | UserCode
+ * [35] ISO639Code ::= ([a-z] | [A-Z]) ([a-z] | [A-Z])
+ * [36] IanaCode ::= ('i' | 'I') '-' ([a-z] | [A-Z])+
+ * [37] UserCode ::= ('x' | 'X') '-' ([a-z] | [A-Z])+
+ * [38] Subcode ::= ([a-z] | [A-Z])+
+ *
+ * The current REC reference the sucessors of RFC 1766, currently 5646
+ *
+ * http://www.rfc-editor.org/rfc/rfc5646.txt
+ * langtag = language
+ * ["-" script]
+ * ["-" region]
+ * *("-" variant)
+ * *("-" extension)
+ * ["-" privateuse]
+ * language = 2*3ALPHA ; shortest ISO 639 code
+ * ["-" extlang] ; sometimes followed by
+ * ; extended language subtags
+ * / 4ALPHA ; or reserved for future use
+ * / 5*8ALPHA ; or registered language subtag
+ *
+ * extlang = 3ALPHA ; selected ISO 639 codes
+ * *2("-" 3ALPHA) ; permanently reserved
+ *
+ * script = 4ALPHA ; ISO 15924 code
+ *
+ * region = 2ALPHA ; ISO 3166-1 code
+ * / 3DIGIT ; UN M.49 code
+ *
+ * variant = 5*8alphanum ; registered variants
+ * / (DIGIT 3alphanum)
+ *
+ * extension = singleton 1*("-" (2*8alphanum))
+ *
+ * ; Single alphanumerics
+ * ; "x" reserved for private use
+ * singleton = DIGIT ; 0 - 9
+ * / %x41-57 ; A - W
+ * / %x59-5A ; Y - Z
+ * / %x61-77 ; a - w
+ * / %x79-7A ; y - z
+ *
+ * it sounds right to still allow Irregular i-xxx IANA and user codes too
+ * The parser below doesn't try to cope with extension or privateuse
+ * that could be added but that's not interoperable anyway
+ *
+ * Returns 1 if correct 0 otherwise
+ **/
+int
+xmlCheckLanguageID(const xmlChar * lang)
+{
+ const xmlChar *cur = lang, *nxt;
+
+ if (cur == NULL)
+ return (0);
+ if (((cur[0] == 'i') && (cur[1] == '-')) ||
+ ((cur[0] == 'I') && (cur[1] == '-')) ||
+ ((cur[0] == 'x') && (cur[1] == '-')) ||
+ ((cur[0] == 'X') && (cur[1] == '-'))) {
+ /*
+ * Still allow IANA code and user code which were coming
+ * from the previous version of the XML-1.0 specification
+ * it's deprecated but we should not fail
+ */
+ cur += 2;
+ while (((cur[0] >= 'A') && (cur[0] <= 'Z')) ||
+ ((cur[0] >= 'a') && (cur[0] <= 'z')))
+ cur++;
+ return(cur[0] == 0);
+ }
+ nxt = cur;
+ while (((nxt[0] >= 'A') && (nxt[0] <= 'Z')) ||
+ ((nxt[0] >= 'a') && (nxt[0] <= 'z')))
+ nxt++;
+ if (nxt - cur >= 4) {
+ /*
+ * Reserved
+ */
+ if ((nxt - cur > 8) || (nxt[0] != 0))
+ return(0);
+ return(1);
+ }
+ if (nxt - cur < 2)
+ return(0);
+ /* we got an ISO 639 code */
+ if (nxt[0] == 0)
+ return(1);
+ if (nxt[0] != '-')
+ return(0);
+
+ nxt++;
+ cur = nxt;
+ /* now we can have extlang or script or region or variant */
+ if ((nxt[0] >= '0') && (nxt[0] <= '9'))
+ goto region_m49;
+
+ while (((nxt[0] >= 'A') && (nxt[0] <= 'Z')) ||
+ ((nxt[0] >= 'a') && (nxt[0] <= 'z')))
+ nxt++;
+ if (nxt - cur == 4)
+ goto script;
+ if (nxt - cur == 2)
+ goto region;
+ if ((nxt - cur >= 5) && (nxt - cur <= 8))
+ goto variant;
+ if (nxt - cur != 3)
+ return(0);
+ /* we parsed an extlang */
+ if (nxt[0] == 0)
+ return(1);
+ if (nxt[0] != '-')
+ return(0);
+
+ nxt++;
+ cur = nxt;
+ /* now we can have script or region or variant */
+ if ((nxt[0] >= '0') && (nxt[0] <= '9'))
+ goto region_m49;
+
+ while (((nxt[0] >= 'A') && (nxt[0] <= 'Z')) ||
+ ((nxt[0] >= 'a') && (nxt[0] <= 'z')))
+ nxt++;
+ if (nxt - cur == 2)
+ goto region;
+ if ((nxt - cur >= 5) && (nxt - cur <= 8))
+ goto variant;
+ if (nxt - cur != 4)
+ return(0);
+ /* we parsed a script */
+script:
+ if (nxt[0] == 0)
+ return(1);
+ if (nxt[0] != '-')
+ return(0);
+
+ nxt++;
+ cur = nxt;
+ /* now we can have region or variant */
+ if ((nxt[0] >= '0') && (nxt[0] <= '9'))
+ goto region_m49;
+
+ while (((nxt[0] >= 'A') && (nxt[0] <= 'Z')) ||
+ ((nxt[0] >= 'a') && (nxt[0] <= 'z')))
+ nxt++;
+
+ if ((nxt - cur >= 5) && (nxt - cur <= 8))
+ goto variant;
+ if (nxt - cur != 2)
+ return(0);
+ /* we parsed a region */
+region:
+ if (nxt[0] == 0)
+ return(1);
+ if (nxt[0] != '-')
+ return(0);
+
+ nxt++;
+ cur = nxt;
+ /* now we can just have a variant */
+ while (((nxt[0] >= 'A') && (nxt[0] <= 'Z')) ||
+ ((nxt[0] >= 'a') && (nxt[0] <= 'z')))
+ nxt++;
+
+ if ((nxt - cur < 5) || (nxt - cur > 8))
+ return(0);
+
+ /* we parsed a variant */
+variant:
+ if (nxt[0] == 0)
+ return(1);
+ if (nxt[0] != '-')
+ return(0);
+ /* extensions and private use subtags not checked */
+ return (1);
+
+region_m49:
+ if (((nxt[1] >= '0') && (nxt[1] <= '9')) &&
+ ((nxt[2] >= '0') && (nxt[2] <= '9'))) {
+ nxt += 3;
+ goto region;
+ }
+ return(0);
+}
+
+/************************************************************************
+ * *
+ * Parser stacks related functions and macros *
+ * *
+ ************************************************************************/
+
+static xmlEntityPtr xmlParseStringEntityRef(xmlParserCtxtPtr ctxt,
+ const xmlChar ** str);
+
+#ifdef SAX2
+/**
+ * nsPush:
+ * @ctxt: an XML parser context
+ * @prefix: the namespace prefix or NULL
+ * @URL: the namespace name
+ *
+ * Pushes a new parser namespace on top of the ns stack
+ *
+ * Returns -1 in case of error, -2 if the namespace should be discarded
+ * and the index in the stack otherwise.
+ */
+static int
+nsPush(xmlParserCtxtPtr ctxt, const xmlChar *prefix, const xmlChar *URL)
+{
+ if (ctxt->options & XML_PARSE_NSCLEAN) {
+ int i;
+ for (i = ctxt->nsNr - 2;i >= 0;i -= 2) {
+ if (ctxt->nsTab[i] == prefix) {
+ /* in scope */
+ if (ctxt->nsTab[i + 1] == URL)
+ return(-2);
+ /* out of scope keep it */
+ break;
+ }
+ }
+ }
+ if ((ctxt->nsMax == 0) || (ctxt->nsTab == NULL)) {
+ ctxt->nsMax = 10;
+ ctxt->nsNr = 0;
+ ctxt->nsTab = (const xmlChar **)
+ xmlMalloc(ctxt->nsMax * sizeof(xmlChar *));
+ if (ctxt->nsTab == NULL) {
+ xmlErrMemory(ctxt, NULL);
+ ctxt->nsMax = 0;
+ return (-1);
+ }
+ } else if (ctxt->nsNr >= ctxt->nsMax) {
+ const xmlChar ** tmp;
+ ctxt->nsMax *= 2;
+ tmp = (const xmlChar **) xmlRealloc((char *) ctxt->nsTab,
+ ctxt->nsMax * sizeof(ctxt->nsTab[0]));
+ if (tmp == NULL) {
+ xmlErrMemory(ctxt, NULL);
+ ctxt->nsMax /= 2;
+ return (-1);
+ }
+ ctxt->nsTab = tmp;
+ }
+ ctxt->nsTab[ctxt->nsNr++] = prefix;
+ ctxt->nsTab[ctxt->nsNr++] = URL;
+ return (ctxt->nsNr);
+}
+/**
+ * nsPop:
+ * @ctxt: an XML parser context
+ * @nr: the number to pop
+ *
+ * Pops the top @nr parser prefix/namespace from the ns stack
+ *
+ * Returns the number of namespaces removed
+ */
+static int
+nsPop(xmlParserCtxtPtr ctxt, int nr)
+{
+ int i;
+
+ if (ctxt->nsTab == NULL) return(0);
+ if (ctxt->nsNr < nr) {
+ xmlGenericError(xmlGenericErrorContext, "Pbm popping %d NS\n", nr);
+ nr = ctxt->nsNr;
+ }
+ if (ctxt->nsNr <= 0)
+ return (0);
+
+ for (i = 0;i < nr;i++) {
+ ctxt->nsNr--;
+ ctxt->nsTab[ctxt->nsNr] = NULL;
+ }
+ return(nr);
+}
+#endif
+
+static int
+xmlCtxtGrowAttrs(xmlParserCtxtPtr ctxt, int nr) {
+ const xmlChar **atts;
+ int *attallocs;
+ int maxatts;
+
+ if (ctxt->atts == NULL) {
+ maxatts = 55; /* allow for 10 attrs by default */
+ atts = (const xmlChar **)
+ xmlMalloc(maxatts * sizeof(xmlChar *));
+ if (atts == NULL) goto mem_error;
+ ctxt->atts = atts;
+ attallocs = (int *) xmlMalloc((maxatts / 5) * sizeof(int));
+ if (attallocs == NULL) goto mem_error;
+ ctxt->attallocs = attallocs;
+ ctxt->maxatts = maxatts;
+ } else if (nr + 5 > ctxt->maxatts) {
+ maxatts = (nr + 5) * 2;
+ atts = (const xmlChar **) xmlRealloc((void *) ctxt->atts,
+ maxatts * sizeof(const xmlChar *));
+ if (atts == NULL) goto mem_error;
+ ctxt->atts = atts;
+ attallocs = (int *) xmlRealloc((void *) ctxt->attallocs,
+ (maxatts / 5) * sizeof(int));
+ if (attallocs == NULL) goto mem_error;
+ ctxt->attallocs = attallocs;
+ ctxt->maxatts = maxatts;
+ }
+ return(ctxt->maxatts);
+mem_error:
+ xmlErrMemory(ctxt, NULL);
+ return(-1);
+}
+
+/**
+ * inputPush:
+ * @ctxt: an XML parser context
+ * @value: the parser input
+ *
+ * Pushes a new parser input on top of the input stack
+ *
+ * Returns -1 in case of error, the index in the stack otherwise
+ */
+int
+inputPush(xmlParserCtxtPtr ctxt, xmlParserInputPtr value)
+{
+ if ((ctxt == NULL) || (value == NULL))
+ return(-1);
+ if (ctxt->inputNr >= ctxt->inputMax) {
+ ctxt->inputMax *= 2;
+ ctxt->inputTab =
+ (xmlParserInputPtr *) xmlRealloc(ctxt->inputTab,
+ ctxt->inputMax *
+ sizeof(ctxt->inputTab[0]));
+ if (ctxt->inputTab == NULL) {
+ xmlErrMemory(ctxt, NULL);
+ xmlFreeInputStream(value);
+ ctxt->inputMax /= 2;
+ value = NULL;
+ return (-1);
+ }
+ }
+ ctxt->inputTab[ctxt->inputNr] = value;
+ ctxt->input = value;
+ return (ctxt->inputNr++);
+}
+/**
+ * inputPop:
+ * @ctxt: an XML parser context
+ *
+ * Pops the top parser input from the input stack
+ *
+ * Returns the input just removed
+ */
+xmlParserInputPtr
+inputPop(xmlParserCtxtPtr ctxt)
+{
+ xmlParserInputPtr ret;
+
+ if (ctxt == NULL)
+ return(NULL);
+ if (ctxt->inputNr <= 0)
+ return (NULL);
+ ctxt->inputNr--;
+ if (ctxt->inputNr > 0)
+ ctxt->input = ctxt->inputTab[ctxt->inputNr - 1];
+ else
+ ctxt->input = NULL;
+ ret = ctxt->inputTab[ctxt->inputNr];
+ ctxt->inputTab[ctxt->inputNr] = NULL;
+ return (ret);
+}
+/**
+ * nodePush:
+ * @ctxt: an XML parser context
+ * @value: the element node
+ *
+ * Pushes a new element node on top of the node stack
+ *
+ * Returns -1 in case of error, the index in the stack otherwise
+ */
+int
+nodePush(xmlParserCtxtPtr ctxt, xmlNodePtr value)
+{
+ if (ctxt == NULL) return(0);
+ if (ctxt->nodeNr >= ctxt->nodeMax) {
+ xmlNodePtr *tmp;
+
+ tmp = (xmlNodePtr *) xmlRealloc(ctxt->nodeTab,
+ ctxt->nodeMax * 2 *
+ sizeof(ctxt->nodeTab[0]));
+ if (tmp == NULL) {
+ xmlErrMemory(ctxt, NULL);
+ return (-1);
+ }
+ ctxt->nodeTab = tmp;
+ ctxt->nodeMax *= 2;
+ }
+ if ((((unsigned int) ctxt->nodeNr) > xmlParserMaxDepth) &&
+ ((ctxt->options & XML_PARSE_HUGE) == 0)) {
+ xmlFatalErrMsgInt(ctxt, XML_ERR_INTERNAL_ERROR,
+ "Excessive depth in document: %d use XML_PARSE_HUGE option\n",
+ xmlParserMaxDepth);
+ xmlHaltParser(ctxt);
+ return(-1);
+ }
+ ctxt->nodeTab[ctxt->nodeNr] = value;
+ ctxt->node = value;
+ return (ctxt->nodeNr++);
+}
+
+/**
+ * nodePop:
+ * @ctxt: an XML parser context
+ *
+ * Pops the top element node from the node stack
+ *
+ * Returns the node just removed
+ */
+xmlNodePtr
+nodePop(xmlParserCtxtPtr ctxt)
+{
+ xmlNodePtr ret;
+
+ if (ctxt == NULL) return(NULL);
+ if (ctxt->nodeNr <= 0)
+ return (NULL);
+ ctxt->nodeNr--;
+ if (ctxt->nodeNr > 0)
+ ctxt->node = ctxt->nodeTab[ctxt->nodeNr - 1];
+ else
+ ctxt->node = NULL;
+ ret = ctxt->nodeTab[ctxt->nodeNr];
+ ctxt->nodeTab[ctxt->nodeNr] = NULL;
+ return (ret);
+}
+
+#ifdef LIBXML_PUSH_ENABLED
+/**
+ * nameNsPush:
+ * @ctxt: an XML parser context
+ * @value: the element name
+ * @prefix: the element prefix
+ * @URI: the element namespace name
+ *
+ * Pushes a new element name/prefix/URL on top of the name stack
+ *
+ * Returns -1 in case of error, the index in the stack otherwise
+ */
+static int
+nameNsPush(xmlParserCtxtPtr ctxt, const xmlChar * value,
+ const xmlChar *prefix, const xmlChar *URI, int nsNr)
+{
+ if (ctxt->nameNr >= ctxt->nameMax) {
+ const xmlChar * *tmp;
+ void **tmp2;
+ ctxt->nameMax *= 2;
+ tmp = (const xmlChar * *) xmlRealloc((xmlChar * *)ctxt->nameTab,
+ ctxt->nameMax *
+ sizeof(ctxt->nameTab[0]));
+ if (tmp == NULL) {
+ ctxt->nameMax /= 2;
+ goto mem_error;
+ }
+ ctxt->nameTab = tmp;
+ tmp2 = (void **) xmlRealloc((void * *)ctxt->pushTab,
+ ctxt->nameMax * 3 *
+ sizeof(ctxt->pushTab[0]));
+ if (tmp2 == NULL) {
+ ctxt->nameMax /= 2;
+ goto mem_error;
+ }
+ ctxt->pushTab = tmp2;
+ }
+ ctxt->nameTab[ctxt->nameNr] = value;
+ ctxt->name = value;
+ ctxt->pushTab[ctxt->nameNr * 3] = (void *) prefix;
+ ctxt->pushTab[ctxt->nameNr * 3 + 1] = (void *) URI;
+ ctxt->pushTab[ctxt->nameNr * 3 + 2] = (void *) (long) nsNr;
+ return (ctxt->nameNr++);
+mem_error:
+ xmlErrMemory(ctxt, NULL);
+ return (-1);
+}
+/**
+ * nameNsPop:
+ * @ctxt: an XML parser context
+ *
+ * Pops the top element/prefix/URI name from the name stack
+ *
+ * Returns the name just removed
+ */
+static const xmlChar *
+nameNsPop(xmlParserCtxtPtr ctxt)
+{
+ const xmlChar *ret;
+
+ if (ctxt->nameNr <= 0)
+ return (NULL);
+ ctxt->nameNr--;
+ if (ctxt->nameNr > 0)
+ ctxt->name = ctxt->nameTab[ctxt->nameNr - 1];
+ else
+ ctxt->name = NULL;
+ ret = ctxt->nameTab[ctxt->nameNr];
+ ctxt->nameTab[ctxt->nameNr] = NULL;
+ return (ret);
+}
+#endif /* LIBXML_PUSH_ENABLED */
+
+/**
+ * namePush:
+ * @ctxt: an XML parser context
+ * @value: the element name
+ *
+ * Pushes a new element name on top of the name stack
+ *
+ * Returns -1 in case of error, the index in the stack otherwise
+ */
+int
+namePush(xmlParserCtxtPtr ctxt, const xmlChar * value)
+{
+ if (ctxt == NULL) return (-1);
+
+ if (ctxt->nameNr >= ctxt->nameMax) {
+ const xmlChar * *tmp;
+ tmp = (const xmlChar * *) xmlRealloc((xmlChar * *)ctxt->nameTab,
+ ctxt->nameMax * 2 *
+ sizeof(ctxt->nameTab[0]));
+ if (tmp == NULL) {
+ goto mem_error;
+ }
+ ctxt->nameTab = tmp;
+ ctxt->nameMax *= 2;
+ }
+ ctxt->nameTab[ctxt->nameNr] = value;
+ ctxt->name = value;
+ return (ctxt->nameNr++);
+mem_error:
+ xmlErrMemory(ctxt, NULL);
+ return (-1);
+}
+/**
+ * namePop:
+ * @ctxt: an XML parser context
+ *
+ * Pops the top element name from the name stack
+ *
+ * Returns the name just removed
+ */
+const xmlChar *
+namePop(xmlParserCtxtPtr ctxt)
+{
+ const xmlChar *ret;
+
+ if ((ctxt == NULL) || (ctxt->nameNr <= 0))
+ return (NULL);
+ ctxt->nameNr--;
+ if (ctxt->nameNr > 0)
+ ctxt->name = ctxt->nameTab[ctxt->nameNr - 1];
+ else
+ ctxt->name = NULL;
+ ret = ctxt->nameTab[ctxt->nameNr];
+ ctxt->nameTab[ctxt->nameNr] = NULL;
+ return (ret);
+}
+
+static int spacePush(xmlParserCtxtPtr ctxt, int val) {
+ if (ctxt->spaceNr >= ctxt->spaceMax) {
+ int *tmp;
+
+ ctxt->spaceMax *= 2;
+ tmp = (int *) xmlRealloc(ctxt->spaceTab,
+ ctxt->spaceMax * sizeof(ctxt->spaceTab[0]));
+ if (tmp == NULL) {
+ xmlErrMemory(ctxt, NULL);
+ ctxt->spaceMax /=2;
+ return(-1);
+ }
+ ctxt->spaceTab = tmp;
+ }
+ ctxt->spaceTab[ctxt->spaceNr] = val;
+ ctxt->space = &ctxt->spaceTab[ctxt->spaceNr];
+ return(ctxt->spaceNr++);
+}
+
+static int spacePop(xmlParserCtxtPtr ctxt) {
+ int ret;
+ if (ctxt->spaceNr <= 0) return(0);
+ ctxt->spaceNr--;
+ if (ctxt->spaceNr > 0)
+ ctxt->space = &ctxt->spaceTab[ctxt->spaceNr - 1];
+ else
+ ctxt->space = &ctxt->spaceTab[0];
+ ret = ctxt->spaceTab[ctxt->spaceNr];
+ ctxt->spaceTab[ctxt->spaceNr] = -1;
+ return(ret);
+}
+
+/*
+ * Macros for accessing the content. Those should be used only by the parser,
+ * and not exported.
+ *
+ * Dirty macros, i.e. one often need to make assumption on the context to
+ * use them
+ *
+ * CUR_PTR return the current pointer to the xmlChar to be parsed.
+ * To be used with extreme caution since operations consuming
+ * characters may move the input buffer to a different location !
+ * CUR returns the current xmlChar value, i.e. a 8 bit value if compiled
+ * This should be used internally by the parser
+ * only to compare to ASCII values otherwise it would break when
+ * running with UTF-8 encoding.
+ * RAW same as CUR but in the input buffer, bypass any token
+ * extraction that may have been done
+ * NXT(n) returns the n'th next xmlChar. Same as CUR is should be used only
+ * to compare on ASCII based substring.
+ * SKIP(n) Skip n xmlChar, and must also be used only to skip ASCII defined
+ * strings without newlines within the parser.
+ * NEXT1(l) Skip 1 xmlChar, and must also be used only to skip 1 non-newline ASCII
+ * defined char within the parser.
+ * Clean macros, not dependent of an ASCII context, expect UTF-8 encoding
+ *
+ * NEXT Skip to the next character, this does the proper decoding
+ * in UTF-8 mode. It also pop-up unfinished entities on the fly.
+ * NEXTL(l) Skip the current unicode character of l xmlChars long.
+ * CUR_CHAR(l) returns the current unicode character (int), set l
+ * to the number of xmlChars used for the encoding [0-5].
+ * CUR_SCHAR same but operate on a string instead of the context
+ * COPY_BUF copy the current unicode char to the target buffer, increment
+ * the index
+ * GROW, SHRINK handling of input buffers
+ */
+
+#define RAW (*ctxt->input->cur)
+#define CUR (*ctxt->input->cur)
+#define NXT(val) ctxt->input->cur[(val)]
+#define CUR_PTR ctxt->input->cur
+#define BASE_PTR ctxt->input->base
+
+#define CMP4( s, c1, c2, c3, c4 ) \
+ ( ((unsigned char *) s)[ 0 ] == c1 && ((unsigned char *) s)[ 1 ] == c2 && \
+ ((unsigned char *) s)[ 2 ] == c3 && ((unsigned char *) s)[ 3 ] == c4 )
+#define CMP5( s, c1, c2, c3, c4, c5 ) \
+ ( CMP4( s, c1, c2, c3, c4 ) && ((unsigned char *) s)[ 4 ] == c5 )
+#define CMP6( s, c1, c2, c3, c4, c5, c6 ) \
+ ( CMP5( s, c1, c2, c3, c4, c5 ) && ((unsigned char *) s)[ 5 ] == c6 )
+#define CMP7( s, c1, c2, c3, c4, c5, c6, c7 ) \
+ ( CMP6( s, c1, c2, c3, c4, c5, c6 ) && ((unsigned char *) s)[ 6 ] == c7 )
+#define CMP8( s, c1, c2, c3, c4, c5, c6, c7, c8 ) \
+ ( CMP7( s, c1, c2, c3, c4, c5, c6, c7 ) && ((unsigned char *) s)[ 7 ] == c8 )
+#define CMP9( s, c1, c2, c3, c4, c5, c6, c7, c8, c9 ) \
+ ( CMP8( s, c1, c2, c3, c4, c5, c6, c7, c8 ) && \
+ ((unsigned char *) s)[ 8 ] == c9 )
+#define CMP10( s, c1, c2, c3, c4, c5, c6, c7, c8, c9, c10 ) \
+ ( CMP9( s, c1, c2, c3, c4, c5, c6, c7, c8, c9 ) && \
+ ((unsigned char *) s)[ 9 ] == c10 )
+
+#define SKIP(val) do { \
+ ctxt->nbChars += (val),ctxt->input->cur += (val),ctxt->input->col+=(val); \
+ if (*ctxt->input->cur == '%') xmlParserHandlePEReference(ctxt); \
+ if ((*ctxt->input->cur == 0) && \
+ (xmlParserInputGrow(ctxt->input, INPUT_CHUNK) <= 0)) \
+ xmlPopInput(ctxt); \
+ } while (0)
+
+#define SKIPL(val) do { \
+ int skipl; \
+ for(skipl=0; skipl<val; skipl++) { \
+ if (*(ctxt->input->cur) == '\n') { \
+ ctxt->input->line++; ctxt->input->col = 1; \
+ } else ctxt->input->col++; \
+ ctxt->nbChars++; \
+ ctxt->input->cur++; \
+ } \
+ if (*ctxt->input->cur == '%') xmlParserHandlePEReference(ctxt); \
+ if ((*ctxt->input->cur == 0) && \
+ (xmlParserInputGrow(ctxt->input, INPUT_CHUNK) <= 0)) \
+ xmlPopInput(ctxt); \
+ } while (0)
+
+#define SHRINK if ((ctxt->progressive == 0) && \
+ (ctxt->input->cur - ctxt->input->base > 2 * INPUT_CHUNK) && \
+ (ctxt->input->end - ctxt->input->cur < 2 * INPUT_CHUNK)) \
+ xmlSHRINK (ctxt);
+
+static void xmlSHRINK (xmlParserCtxtPtr ctxt) {
+ xmlParserInputShrink(ctxt->input);
+ if ((*ctxt->input->cur == 0) &&
+ (xmlParserInputGrow(ctxt->input, INPUT_CHUNK) <= 0))
+ xmlPopInput(ctxt);
+ }
+
+#define GROW if ((ctxt->progressive == 0) && \
+ (ctxt->input->end - ctxt->input->cur < INPUT_CHUNK)) \
+ xmlGROW (ctxt);
+
+static void xmlGROW (xmlParserCtxtPtr ctxt) {
+ unsigned long curEnd = ctxt->input->end - ctxt->input->cur;
+ unsigned long curBase = ctxt->input->cur - ctxt->input->base;
+
+ if (((curEnd > (unsigned long) XML_MAX_LOOKUP_LIMIT) ||
+ (curBase > (unsigned long) XML_MAX_LOOKUP_LIMIT)) &&
+ ((ctxt->input->buf) && (ctxt->input->buf->readcallback != (xmlInputReadCallback) xmlNop)) &&
+ ((ctxt->options & XML_PARSE_HUGE) == 0)) {
+ xmlFatalErr(ctxt, XML_ERR_INTERNAL_ERROR, "Huge input lookup");
+ xmlHaltParser(ctxt);
+ return;
+ }
+ xmlParserInputGrow(ctxt->input, INPUT_CHUNK);
+ if ((ctxt->input->cur > ctxt->input->end) ||
+ (ctxt->input->cur < ctxt->input->base)) {
+ xmlHaltParser(ctxt);
+ xmlFatalErr(ctxt, XML_ERR_INTERNAL_ERROR, "cur index out of bound");
+ return;
+ }
+ if ((ctxt->input->cur != NULL) && (*ctxt->input->cur == 0) &&
+ (xmlParserInputGrow(ctxt->input, INPUT_CHUNK) <= 0))
+ xmlPopInput(ctxt);
+}
+
+#define SKIP_BLANKS xmlSkipBlankChars(ctxt)
+
+#define NEXT xmlNextChar(ctxt)
+
+#define NEXT1 { \
+ ctxt->input->col++; \
+ ctxt->input->cur++; \
+ ctxt->nbChars++; \
+ if (*ctxt->input->cur == 0) \
+ xmlParserInputGrow(ctxt->input, INPUT_CHUNK); \
+ }
+
+#define NEXTL(l) do { \
+ if (*(ctxt->input->cur) == '\n') { \
+ ctxt->input->line++; ctxt->input->col = 1; \
+ } else ctxt->input->col++; \
+ ctxt->input->cur += l; \
+ if (*ctxt->input->cur == '%') xmlParserHandlePEReference(ctxt); \
+ } while (0)
+
+#define CUR_CHAR(l) xmlCurrentChar(ctxt, &l)
+#define CUR_SCHAR(s, l) xmlStringCurrentChar(ctxt, s, &l)
+
+#define COPY_BUF(l,b,i,v) \
+ if (l == 1) b[i++] = (xmlChar) v; \
+ else i += xmlCopyCharMultiByte(&b[i],v)
+
+/**
+ * xmlSkipBlankChars:
+ * @ctxt: the XML parser context
+ *
+ * skip all blanks character found at that point in the input streams.
+ * It pops up finished entities in the process if allowable at that point.
+ *
+ * Returns the number of space chars skipped
+ */
+
+int
+xmlSkipBlankChars(xmlParserCtxtPtr ctxt) {
+ int res = 0;
+
+ /*
+ * It's Okay to use CUR/NEXT here since all the blanks are on
+ * the ASCII range.
+ */
+ if ((ctxt->inputNr == 1) && (ctxt->instate != XML_PARSER_DTD)) {
+ const xmlChar *cur;
+ /*
+ * if we are in the document content, go really fast
+ */
+ cur = ctxt->input->cur;
+ while (IS_BLANK_CH(*cur)) {
+ if (*cur == '\n') {
+ ctxt->input->line++; ctxt->input->col = 1;
+ } else {
+ ctxt->input->col++;
+ }
+ cur++;
+ res++;
+ if (*cur == 0) {
+ ctxt->input->cur = cur;
+ xmlParserInputGrow(ctxt->input, INPUT_CHUNK);
+ cur = ctxt->input->cur;
+ }
+ }
+ ctxt->input->cur = cur;
+ } else {
+ int cur;
+ do {
+ cur = CUR;
+ while ((IS_BLANK_CH(cur) && /* CHECKED tstblanks.xml */
+ (ctxt->instate != XML_PARSER_EOF))) {
+ NEXT;
+ cur = CUR;
+ res++;
+ }
+ while ((cur == 0) && (ctxt->inputNr > 1) &&
+ (ctxt->instate != XML_PARSER_COMMENT)) {
+ xmlPopInput(ctxt);
+ cur = CUR;
+ }
+ /*
+ * Need to handle support of entities branching here
+ */
+ if (*ctxt->input->cur == '%') xmlParserHandlePEReference(ctxt);
+ } while ((IS_BLANK(cur)) && /* CHECKED tstblanks.xml */
+ (ctxt->instate != XML_PARSER_EOF));
+ }
+ return(res);
+}
+
+/************************************************************************
+ * *
+ * Commodity functions to handle entities *
+ * *
+ ************************************************************************/
+
+/**
+ * xmlPopInput:
+ * @ctxt: an XML parser context
+ *
+ * xmlPopInput: the current input pointed by ctxt->input came to an end
+ * pop it and return the next char.
+ *
+ * Returns the current xmlChar in the parser context
+ */
+xmlChar
+xmlPopInput(xmlParserCtxtPtr ctxt) {
+ if ((ctxt == NULL) || (ctxt->inputNr <= 1)) return(0);
+ if (xmlParserDebugEntities)
+ xmlGenericError(xmlGenericErrorContext,
+ "Popping input %d\n", ctxt->inputNr);
+ xmlFreeInputStream(inputPop(ctxt));
+ if ((*ctxt->input->cur == 0) &&
+ (xmlParserInputGrow(ctxt->input, INPUT_CHUNK) <= 0))
+ return(xmlPopInput(ctxt));
+ return(CUR);
+}
+
+/**
+ * xmlPushInput:
+ * @ctxt: an XML parser context
+ * @input: an XML parser input fragment (entity, XML fragment ...).
+ *
+ * xmlPushInput: switch to a new input stream which is stacked on top
+ * of the previous one(s).
+ * Returns -1 in case of error or the index in the input stack
+ */
+int
+xmlPushInput(xmlParserCtxtPtr ctxt, xmlParserInputPtr input) {
+ int ret;
+ if (input == NULL) return(-1);
+
+ if (xmlParserDebugEntities) {
+ if ((ctxt->input != NULL) && (ctxt->input->filename))
+ xmlGenericError(xmlGenericErrorContext,
+ "%s(%d): ", ctxt->input->filename,
+ ctxt->input->line);
+ xmlGenericError(xmlGenericErrorContext,
+ "Pushing input %d : %.30s\n", ctxt->inputNr+1, input->cur);
+ }
+ ret = inputPush(ctxt, input);
+ if (ctxt->instate == XML_PARSER_EOF)
+ return(-1);
+ GROW;
+ return(ret);
+}
+
+/**
+ * xmlParseCharRef:
+ * @ctxt: an XML parser context
+ *
+ * parse Reference declarations
+ *
+ * [66] CharRef ::= '&#' [0-9]+ ';' |
+ * '&#x' [0-9a-fA-F]+ ';'
+ *
+ * [ WFC: Legal Character ]
+ * Characters referred to using character references must match the
+ * production for Char.
+ *
+ * Returns the value parsed (as an int), 0 in case of error
+ */
+int
+xmlParseCharRef(xmlParserCtxtPtr ctxt) {
+ unsigned int val = 0;
+ int count = 0;
+ unsigned int outofrange = 0;
+
+ /*
+ * Using RAW/CUR/NEXT is okay since we are working on ASCII range here
+ */
+ if ((RAW == '&') && (NXT(1) == '#') &&
+ (NXT(2) == 'x')) {
+ SKIP(3);
+ GROW;
+ while (RAW != ';') { /* loop blocked by count */
+ if (count++ > 20) {
+ count = 0;
+ GROW;
+ if (ctxt->instate == XML_PARSER_EOF)
+ return(0);
+ }
+ if ((RAW >= '0') && (RAW <= '9'))
+ val = val * 16 + (CUR - '0');
+ else if ((RAW >= 'a') && (RAW <= 'f') && (count < 20))
+ val = val * 16 + (CUR - 'a') + 10;
+ else if ((RAW >= 'A') && (RAW <= 'F') && (count < 20))
+ val = val * 16 + (CUR - 'A') + 10;
+ else {
+ xmlFatalErr(ctxt, XML_ERR_INVALID_HEX_CHARREF, NULL);
+ val = 0;
+ break;
+ }
+ if (val > 0x10FFFF)
+ outofrange = val;
+
+ NEXT;
+ count++;
+ }
+ if (RAW == ';') {
+ /* on purpose to avoid reentrancy problems with NEXT and SKIP */
+ ctxt->input->col++;
+ ctxt->nbChars ++;
+ ctxt->input->cur++;
+ }
+ } else if ((RAW == '&') && (NXT(1) == '#')) {
+ SKIP(2);
+ GROW;
+ while (RAW != ';') { /* loop blocked by count */
+ if (count++ > 20) {
+ count = 0;
+ GROW;
+ if (ctxt->instate == XML_PARSER_EOF)
+ return(0);
+ }
+ if ((RAW >= '0') && (RAW <= '9'))
+ val = val * 10 + (CUR - '0');
+ else {
+ xmlFatalErr(ctxt, XML_ERR_INVALID_DEC_CHARREF, NULL);
+ val = 0;
+ break;
+ }
+ if (val > 0x10FFFF)
+ outofrange = val;
+
+ NEXT;
+ count++;
+ }
+ if (RAW == ';') {
+ /* on purpose to avoid reentrancy problems with NEXT and SKIP */
+ ctxt->input->col++;
+ ctxt->nbChars ++;
+ ctxt->input->cur++;
+ }
+ } else {
+ xmlFatalErr(ctxt, XML_ERR_INVALID_CHARREF, NULL);
+ }
+
+ /*
+ * [ WFC: Legal Character ]
+ * Characters referred to using character references must match the
+ * production for Char.
+ */
+ if ((IS_CHAR(val) && (outofrange == 0))) {
+ return(val);
+ } else {
+ xmlFatalErrMsgInt(ctxt, XML_ERR_INVALID_CHAR,
+ "xmlParseCharRef: invalid xmlChar value %d\n",
+ val);
+ }
+ return(0);
+}
+
+/**
+ * xmlParseStringCharRef:
+ * @ctxt: an XML parser context
+ * @str: a pointer to an index in the string
+ *
+ * parse Reference declarations, variant parsing from a string rather
+ * than an an input flow.
+ *
+ * [66] CharRef ::= '&#' [0-9]+ ';' |
+ * '&#x' [0-9a-fA-F]+ ';'
+ *
+ * [ WFC: Legal Character ]
+ * Characters referred to using character references must match the
+ * production for Char.
+ *
+ * Returns the value parsed (as an int), 0 in case of error, str will be
+ * updated to the current value of the index
+ */
+static int
+xmlParseStringCharRef(xmlParserCtxtPtr ctxt, const xmlChar **str) {
+ const xmlChar *ptr;
+ xmlChar cur;
+ unsigned int val = 0;
+ unsigned int outofrange = 0;
+
+ if ((str == NULL) || (*str == NULL)) return(0);
+ ptr = *str;
+ cur = *ptr;
+ if ((cur == '&') && (ptr[1] == '#') && (ptr[2] == 'x')) {
+ ptr += 3;
+ cur = *ptr;
+ while (cur != ';') { /* Non input consuming loop */
+ if ((cur >= '0') && (cur <= '9'))
+ val = val * 16 + (cur - '0');
+ else if ((cur >= 'a') && (cur <= 'f'))
+ val = val * 16 + (cur - 'a') + 10;
+ else if ((cur >= 'A') && (cur <= 'F'))
+ val = val * 16 + (cur - 'A') + 10;
+ else {
+ xmlFatalErr(ctxt, XML_ERR_INVALID_HEX_CHARREF, NULL);
+ val = 0;
+ break;
+ }
+ if (val > 0x10FFFF)
+ outofrange = val;
+
+ ptr++;
+ cur = *ptr;
+ }
+ if (cur == ';')
+ ptr++;
+ } else if ((cur == '&') && (ptr[1] == '#')){
+ ptr += 2;
+ cur = *ptr;
+ while (cur != ';') { /* Non input consuming loops */
+ if ((cur >= '0') && (cur <= '9'))
+ val = val * 10 + (cur - '0');
+ else {
+ xmlFatalErr(ctxt, XML_ERR_INVALID_DEC_CHARREF, NULL);
+ val = 0;
+ break;
+ }
+ if (val > 0x10FFFF)
+ outofrange = val;
+
+ ptr++;
+ cur = *ptr;
+ }
+ if (cur == ';')
+ ptr++;
+ } else {
+ xmlFatalErr(ctxt, XML_ERR_INVALID_CHARREF, NULL);
+ return(0);
+ }
+ *str = ptr;
+
+ /*
+ * [ WFC: Legal Character ]
+ * Characters referred to using character references must match the
+ * production for Char.
+ */
+ if ((IS_CHAR(val) && (outofrange == 0))) {
+ return(val);
+ } else {
+ xmlFatalErrMsgInt(ctxt, XML_ERR_INVALID_CHAR,
+ "xmlParseStringCharRef: invalid xmlChar value %d\n",
+ val);
+ }
+ return(0);
+}
+
+/**
+ * xmlNewBlanksWrapperInputStream:
+ * @ctxt: an XML parser context
+ * @entity: an Entity pointer
+ *
+ * Create a new input stream for wrapping
+ * blanks around a PEReference
+ *
+ * Returns the new input stream or NULL
+ */
+
+static void deallocblankswrapper (xmlChar *str) {xmlFree(str);}
+
+static xmlParserInputPtr
+xmlNewBlanksWrapperInputStream(xmlParserCtxtPtr ctxt, xmlEntityPtr entity) {
+ xmlParserInputPtr input;
+ xmlChar *buffer;
+ size_t length;
+ if (entity == NULL) {
+ xmlFatalErr(ctxt, XML_ERR_INTERNAL_ERROR,
+ "xmlNewBlanksWrapperInputStream entity\n");
+ return(NULL);
+ }
+ if (xmlParserDebugEntities)
+ xmlGenericError(xmlGenericErrorContext,
+ "new blanks wrapper for entity: %s\n", entity->name);
+ input = xmlNewInputStream(ctxt);
+ if (input == NULL) {
+ return(NULL);
+ }
+ length = xmlStrlen(entity->name) + 5;
+ buffer = xmlMallocAtomic(length);
+ if (buffer == NULL) {
+ xmlErrMemory(ctxt, NULL);
+ xmlFree(input);
+ return(NULL);
+ }
+ buffer [0] = ' ';
+ buffer [1] = '%';
+ buffer [length-3] = ';';
+ buffer [length-2] = ' ';
+ buffer [length-1] = 0;
+ memcpy(buffer + 2, entity->name, length - 5);
+ input->free = deallocblankswrapper;
+ input->base = buffer;
+ input->cur = buffer;
+ input->length = length;
+ input->end = &buffer[length];
+ return(input);
+}
+
+/**
+ * xmlParserHandlePEReference:
+ * @ctxt: the parser context
+ *
+ * [69] PEReference ::= '%' Name ';'
+ *
+ * [ WFC: No Recursion ]
+ * A parsed entity must not contain a recursive
+ * reference to itself, either directly or indirectly.
+ *
+ * [ WFC: Entity Declared ]
+ * In a document without any DTD, a document with only an internal DTD
+ * subset which contains no parameter entity references, or a document
+ * with "standalone='yes'", ... ... The declaration of a parameter
+ * entity must precede any reference to it...
+ *
+ * [ VC: Entity Declared ]
+ * In a document with an external subset or external parameter entities
+ * with "standalone='no'", ... ... The declaration of a parameter entity
+ * must precede any reference to it...
+ *
+ * [ WFC: In DTD ]
+ * Parameter-entity references may only appear in the DTD.
+ * NOTE: misleading but this is handled.
+ *
+ * A PEReference may have been detected in the current input stream
+ * the handling is done accordingly to
+ * http://www.w3.org/TR/REC-xml#entproc
+ * i.e.
+ * - Included in literal in entity values
+ * - Included as Parameter Entity reference within DTDs
+ */
+void
+xmlParserHandlePEReference(xmlParserCtxtPtr ctxt) {
+ const xmlChar *name;
+ xmlEntityPtr entity = NULL;
+ xmlParserInputPtr input;
+
+ if (RAW != '%') return;
+ switch(ctxt->instate) {
+ case XML_PARSER_CDATA_SECTION:
+ return;
+ case XML_PARSER_COMMENT:
+ return;
+ case XML_PARSER_START_TAG:
+ return;
+ case XML_PARSER_END_TAG:
+ return;
+ case XML_PARSER_EOF:
+ xmlFatalErr(ctxt, XML_ERR_PEREF_AT_EOF, NULL);
+ return;
+ case XML_PARSER_PROLOG:
+ case XML_PARSER_START:
+ case XML_PARSER_MISC:
+ xmlFatalErr(ctxt, XML_ERR_PEREF_IN_PROLOG, NULL);
+ return;
+ case XML_PARSER_ENTITY_DECL:
+ case XML_PARSER_CONTENT:
+ case XML_PARSER_ATTRIBUTE_VALUE:
+ case XML_PARSER_PI:
+ case XML_PARSER_SYSTEM_LITERAL:
+ case XML_PARSER_PUBLIC_LITERAL:
+ /* we just ignore it there */
+ return;
+ case XML_PARSER_EPILOG:
+ xmlFatalErr(ctxt, XML_ERR_PEREF_IN_EPILOG, NULL);
+ return;
+ case XML_PARSER_ENTITY_VALUE:
+ /*
+ * NOTE: in the case of entity values, we don't do the
+ * substitution here since we need the literal
+ * entity value to be able to save the internal
+ * subset of the document.
+ * This will be handled by xmlStringDecodeEntities
+ */
+ return;
+ case XML_PARSER_DTD:
+ /*
+ * [WFC: Well-Formedness Constraint: PEs in Internal Subset]
+ * In the internal DTD subset, parameter-entity references
+ * can occur only where markup declarations can occur, not
+ * within markup declarations.
+ * In that case this is handled in xmlParseMarkupDecl
+ */
+ if ((ctxt->external == 0) && (ctxt->inputNr == 1))
+ return;
+ if (IS_BLANK_CH(NXT(1)) || NXT(1) == 0)
+ return;
+ break;
+ case XML_PARSER_IGNORE:
+ return;
+ }
+
+ NEXT;
+ name = xmlParseName(ctxt);
+ if (xmlParserDebugEntities)
+ xmlGenericError(xmlGenericErrorContext,
+ "PEReference: %s\n", name);
+ if (name == NULL) {
+ xmlFatalErr(ctxt, XML_ERR_PEREF_NO_NAME, NULL);
+ } else {
+ if (RAW == ';') {
+ NEXT;
+ if ((ctxt->sax != NULL) && (ctxt->sax->getParameterEntity != NULL))
+ entity = ctxt->sax->getParameterEntity(ctxt->userData, name);
+ if (ctxt->instate == XML_PARSER_EOF)
+ return;
+ if (entity == NULL) {
+
+ /*
+ * [ WFC: Entity Declared ]
+ * In a document without any DTD, a document with only an
+ * internal DTD subset which contains no parameter entity
+ * references, or a document with "standalone='yes'", ...
+ * ... The declaration of a parameter entity must precede
+ * any reference to it...
+ */
+ if ((ctxt->standalone == 1) ||
+ ((ctxt->hasExternalSubset == 0) &&
+ (ctxt->hasPErefs == 0))) {
+ xmlFatalErrMsgStr(ctxt, XML_ERR_UNDECLARED_ENTITY,
+ "PEReference: %%%s; not found\n", name);
+ } else {
+ /*
+ * [ VC: Entity Declared ]
+ * In a document with an external subset or external
+ * parameter entities with "standalone='no'", ...
+ * ... The declaration of a parameter entity must precede
+ * any reference to it...
+ */
+ if ((ctxt->validate) && (ctxt->vctxt.error != NULL)) {
+ xmlValidityError(ctxt, XML_WAR_UNDECLARED_ENTITY,
+ "PEReference: %%%s; not found\n",
+ name, NULL);
+ } else
+ xmlWarningMsg(ctxt, XML_WAR_UNDECLARED_ENTITY,
+ "PEReference: %%%s; not found\n",
+ name, NULL);
+ ctxt->valid = 0;
+ }
+ xmlParserEntityCheck(ctxt, 0, NULL, 0);
+ } else if (ctxt->input->free != deallocblankswrapper) {
+ input = xmlNewBlanksWrapperInputStream(ctxt, entity);
+ if (xmlPushInput(ctxt, input) < 0)
+ return;
+ } else {
+ if ((entity->etype == XML_INTERNAL_PARAMETER_ENTITY) ||
+ (entity->etype == XML_EXTERNAL_PARAMETER_ENTITY)) {
+ xmlChar start[4];
+ xmlCharEncoding enc;
+
+ /*
+ * Note: external parameter entities will not be loaded, it
+ * is not required for a non-validating parser, unless the
+ * option of validating, or substituting entities were
+ * given. Doing so is far more secure as the parser will
+ * only process data coming from the document entity by
+ * default.
+ */
+ if ((entity->etype == XML_EXTERNAL_PARAMETER_ENTITY) &&
+ ((ctxt->options & XML_PARSE_NOENT) == 0) &&
+ ((ctxt->options & XML_PARSE_DTDVALID) == 0) &&
+ ((ctxt->options & XML_PARSE_DTDLOAD) == 0) &&
+ ((ctxt->options & XML_PARSE_DTDATTR) == 0) &&
+ (ctxt->replaceEntities == 0) &&
+ (ctxt->validate == 0))
+ return;
+
+ /*
+ * handle the extra spaces added before and after
+ * c.f. http://www.w3.org/TR/REC-xml#as-PE
+ * this is done independently.
+ */
+ input = xmlNewEntityInputStream(ctxt, entity);
+ if (xmlPushInput(ctxt, input) < 0)
+ return;
+
+ /*
+ * Get the 4 first bytes and decode the charset
+ * if enc != XML_CHAR_ENCODING_NONE
+ * plug some encoding conversion routines.
+ * Note that, since we may have some non-UTF8
+ * encoding (like UTF16, bug 135229), the 'length'
+ * is not known, but we can calculate based upon
+ * the amount of data in the buffer.
+ */
+ GROW
+ if (ctxt->instate == XML_PARSER_EOF)
+ return;
+ if ((ctxt->input->end - ctxt->input->cur)>=4) {
+ start[0] = RAW;
+ start[1] = NXT(1);
+ start[2] = NXT(2);
+ start[3] = NXT(3);
+ enc = xmlDetectCharEncoding(start, 4);
+ if (enc != XML_CHAR_ENCODING_NONE) {
+ xmlSwitchEncoding(ctxt, enc);
+ }
+ }
+
+ if ((entity->etype == XML_EXTERNAL_PARAMETER_ENTITY) &&
+ (CMP5(CUR_PTR, '<', '?', 'x', 'm', 'l' )) &&
+ (IS_BLANK_CH(NXT(5)))) {
+ xmlParseTextDecl(ctxt);
+ }
+ } else {
+ xmlFatalErrMsgStr(ctxt, XML_ERR_ENTITY_IS_PARAMETER,
+ "PEReference: %s is not a parameter entity\n",
+ name);
+ }
+ }
+ } else {
+ xmlFatalErr(ctxt, XML_ERR_PEREF_SEMICOL_MISSING, NULL);
+ }
+ }
+}
+
+/*
+ * Macro used to grow the current buffer.
+ * buffer##_size is expected to be a size_t
+ * mem_error: is expected to handle memory allocation failures
+ */
+#define growBuffer(buffer, n) { \
+ xmlChar *tmp; \
+ size_t new_size = buffer##_size * 2 + n; \
+ if (new_size < buffer##_size) goto mem_error; \
+ tmp = (xmlChar *) xmlRealloc(buffer, new_size); \
+ if (tmp == NULL) goto mem_error; \
+ buffer = tmp; \
+ buffer##_size = new_size; \
+}
+
+/**
+ * xmlStringLenDecodeEntities:
+ * @ctxt: the parser context
+ * @str: the input string
+ * @len: the string length
+ * @what: combination of XML_SUBSTITUTE_REF and XML_SUBSTITUTE_PEREF
+ * @end: an end marker xmlChar, 0 if none
+ * @end2: an end marker xmlChar, 0 if none
+ * @end3: an end marker xmlChar, 0 if none
+ *
+ * Takes a entity string content and process to do the adequate substitutions.
+ *
+ * [67] Reference ::= EntityRef | CharRef
+ *
+ * [69] PEReference ::= '%' Name ';'
+ *
+ * Returns A newly allocated string with the substitution done. The caller
+ * must deallocate it !
+ */
+xmlChar *
+xmlStringLenDecodeEntities(xmlParserCtxtPtr ctxt, const xmlChar *str, int len,
+ int what, xmlChar end, xmlChar end2, xmlChar end3) {
+ xmlChar *buffer = NULL;
+ size_t buffer_size = 0;
+ size_t nbchars = 0;
+
+ xmlChar *current = NULL;
+ xmlChar *rep = NULL;
+ const xmlChar *last;
+ xmlEntityPtr ent;
+ int c,l;
+
+ if ((ctxt == NULL) || (str == NULL) || (len < 0))
+ return(NULL);
+ last = str + len;
+
+ if (((ctxt->depth > 40) &&
+ ((ctxt->options & XML_PARSE_HUGE) == 0)) ||
+ (ctxt->depth > 1024)) {
+ xmlFatalErr(ctxt, XML_ERR_ENTITY_LOOP, NULL);
+ return(NULL);
+ }
+
+ /*
+ * allocate a translation buffer.
+ */
+ buffer_size = XML_PARSER_BIG_BUFFER_SIZE;
+ buffer = (xmlChar *) xmlMallocAtomic(buffer_size);
+ if (buffer == NULL) goto mem_error;
+
+ /*
+ * OK loop until we reach one of the ending char or a size limit.
+ * we are operating on already parsed values.
+ */
+ if (str < last)
+ c = CUR_SCHAR(str, l);
+ else
+ c = 0;
+ while ((c != 0) && (c != end) && /* non input consuming loop */
+ (c != end2) && (c != end3)) {
+
+ if (c == 0) break;
+ if ((c == '&') && (str[1] == '#')) {
+ int val = xmlParseStringCharRef(ctxt, &str);
+ if (val != 0) {
+ COPY_BUF(0,buffer,nbchars,val);
+ }
+ if (nbchars + XML_PARSER_BUFFER_SIZE > buffer_size) {
+ growBuffer(buffer, XML_PARSER_BUFFER_SIZE);
+ }
+ } else if ((c == '&') && (what & XML_SUBSTITUTE_REF)) {
+ if (xmlParserDebugEntities)
+ xmlGenericError(xmlGenericErrorContext,
+ "String decoding Entity Reference: %.30s\n",
+ str);
+ ent = xmlParseStringEntityRef(ctxt, &str);
+ if ((ctxt->lastError.code == XML_ERR_ENTITY_LOOP) ||
+ (ctxt->lastError.code == XML_ERR_INTERNAL_ERROR))
+ goto int_error;
+ xmlParserEntityCheck(ctxt, 0, ent, 0);
+ if (ent != NULL)
+ ctxt->nbentities += ent->checked / 2;
+ if ((ent != NULL) &&
+ (ent->etype == XML_INTERNAL_PREDEFINED_ENTITY)) {
+ if (ent->content != NULL) {
+ COPY_BUF(0,buffer,nbchars,ent->content[0]);
+ if (nbchars + XML_PARSER_BUFFER_SIZE > buffer_size) {
+ growBuffer(buffer, XML_PARSER_BUFFER_SIZE);
+ }
+ } else {
+ xmlFatalErrMsg(ctxt, XML_ERR_INTERNAL_ERROR,
+ "predefined entity has no content\n");
+ }
+ } else if ((ent != NULL) && (ent->content != NULL)) {
+ ctxt->depth++;
+ rep = xmlStringDecodeEntities(ctxt, ent->content, what,
+ 0, 0, 0);
+ ctxt->depth--;
+
+ if ((ctxt->lastError.code == XML_ERR_ENTITY_LOOP) ||
+ (ctxt->lastError.code == XML_ERR_INTERNAL_ERROR))
+ goto int_error;
+
+ if (rep != NULL) {
+ current = rep;
+ while (*current != 0) { /* non input consuming loop */
+ buffer[nbchars++] = *current++;
+ if (nbchars + XML_PARSER_BUFFER_SIZE > buffer_size) {
+ if (xmlParserEntityCheck(ctxt, nbchars, ent, 0))
+ goto int_error;
+ growBuffer(buffer, XML_PARSER_BUFFER_SIZE);
+ }
+ }
+ xmlFree(rep);
+ rep = NULL;
+ }
+ } else if (ent != NULL) {
+ int i = xmlStrlen(ent->name);
+ const xmlChar *cur = ent->name;
+
+ buffer[nbchars++] = '&';
+ if (nbchars + i + XML_PARSER_BUFFER_SIZE > buffer_size) {
+ growBuffer(buffer, i + XML_PARSER_BUFFER_SIZE);
+ }
+ for (;i > 0;i--)
+ buffer[nbchars++] = *cur++;
+ buffer[nbchars++] = ';';
+ }
+ } else if (c == '%' && (what & XML_SUBSTITUTE_PEREF)) {
+ if (xmlParserDebugEntities)
+ xmlGenericError(xmlGenericErrorContext,
+ "String decoding PE Reference: %.30s\n", str);
+ ent = xmlParseStringPEReference(ctxt, &str);
+ if (ctxt->lastError.code == XML_ERR_ENTITY_LOOP)
+ goto int_error;
+ xmlParserEntityCheck(ctxt, 0, ent, 0);
+ if (ent != NULL)
+ ctxt->nbentities += ent->checked / 2;
+ if (ent != NULL) {
+ if (ent->content == NULL) {
+ /*
+ * Note: external parsed entities will not be loaded,
+ * it is not required for a non-validating parser to
+ * complete external PEreferences coming from the
+ * internal subset
+ */
+ if (((ctxt->options & XML_PARSE_NOENT) != 0) ||
+ ((ctxt->options & XML_PARSE_DTDVALID) != 0) ||
+ (ctxt->validate != 0)) {
+ xmlLoadEntityContent(ctxt, ent);
+ } else {
+ xmlWarningMsg(ctxt, XML_ERR_ENTITY_PROCESSING,
+ "not validating will not read content for PE entity %s\n",
+ ent->name, NULL);
+ }
+ }
+ ctxt->depth++;
+ rep = xmlStringDecodeEntities(ctxt, ent->content, what,
+ 0, 0, 0);
+ ctxt->depth--;
+ if (rep != NULL) {
+ current = rep;
+ while (*current != 0) { /* non input consuming loop */
+ buffer[nbchars++] = *current++;
+ if (nbchars + XML_PARSER_BUFFER_SIZE > buffer_size) {
+ if (xmlParserEntityCheck(ctxt, nbchars, ent, 0))
+ goto int_error;
+ growBuffer(buffer, XML_PARSER_BUFFER_SIZE);
+ }
+ }
+ xmlFree(rep);
+ rep = NULL;
+ }
+ }
+ } else {
+ COPY_BUF(l,buffer,nbchars,c);
+ str += l;
+ if (nbchars + XML_PARSER_BUFFER_SIZE > buffer_size) {
+ growBuffer(buffer, XML_PARSER_BUFFER_SIZE);
+ }
+ }
+ if (str < last)
+ c = CUR_SCHAR(str, l);
+ else
+ c = 0;
+ }
+ buffer[nbchars] = 0;
+ return(buffer);
+
+mem_error:
+ xmlErrMemory(ctxt, NULL);
+int_error:
+ if (rep != NULL)
+ xmlFree(rep);
+ if (buffer != NULL)
+ xmlFree(buffer);
+ return(NULL);
+}
+
+/**
+ * xmlStringDecodeEntities:
+ * @ctxt: the parser context
+ * @str: the input string
+ * @what: combination of XML_SUBSTITUTE_REF and XML_SUBSTITUTE_PEREF
+ * @end: an end marker xmlChar, 0 if none
+ * @end2: an end marker xmlChar, 0 if none
+ * @end3: an end marker xmlChar, 0 if none
+ *
+ * Takes a entity string content and process to do the adequate substitutions.
+ *
+ * [67] Reference ::= EntityRef | CharRef
+ *
+ * [69] PEReference ::= '%' Name ';'
+ *
+ * Returns A newly allocated string with the substitution done. The caller
+ * must deallocate it !
+ */
+xmlChar *
+xmlStringDecodeEntities(xmlParserCtxtPtr ctxt, const xmlChar *str, int what,
+ xmlChar end, xmlChar end2, xmlChar end3) {
+ if ((ctxt == NULL) || (str == NULL)) return(NULL);
+ return(xmlStringLenDecodeEntities(ctxt, str, xmlStrlen(str), what,
+ end, end2, end3));
+}
+
+/************************************************************************
+ * *
+ * Commodity functions, cleanup needed ? *
+ * *
+ ************************************************************************/
+
+/**
+ * areBlanks:
+ * @ctxt: an XML parser context
+ * @str: a xmlChar *
+ * @len: the size of @str
+ * @blank_chars: we know the chars are blanks
+ *
+ * Is this a sequence of blank chars that one can ignore ?
+ *
+ * Returns 1 if ignorable 0 otherwise.
+ */
+
+static int areBlanks(xmlParserCtxtPtr ctxt, const xmlChar *str, int len,
+ int blank_chars) {
+ int i, ret;
+ xmlNodePtr lastChild;
+
+ /*
+ * Don't spend time trying to differentiate them, the same callback is
+ * used !
+ */
+ if (ctxt->sax->ignorableWhitespace == ctxt->sax->characters)
+ return(0);
+
+ /*
+ * Check for xml:space value.
+ */
+ if ((ctxt->space == NULL) || (*(ctxt->space) == 1) ||
+ (*(ctxt->space) == -2))
+ return(0);
+
+ /*
+ * Check that the string is made of blanks
+ */
+ if (blank_chars == 0) {
+ for (i = 0;i < len;i++)
+ if (!(IS_BLANK_CH(str[i]))) return(0);
+ }
+
+ /*
+ * Look if the element is mixed content in the DTD if available
+ */
+ if (ctxt->node == NULL) return(0);
+ if (ctxt->myDoc != NULL) {
+ ret = xmlIsMixedElement(ctxt->myDoc, ctxt->node->name);
+ if (ret == 0) return(1);
+ if (ret == 1) return(0);
+ }
+
+ /*
+ * Otherwise, heuristic :-\
+ */
+ if ((RAW != '<') && (RAW != 0xD)) return(0);
+ if ((ctxt->node->children == NULL) &&
+ (RAW == '<') && (NXT(1) == '/')) return(0);
+
+ lastChild = xmlGetLastChild(ctxt->node);
+ if (lastChild == NULL) {
+ if ((ctxt->node->type != XML_ELEMENT_NODE) &&
+ (ctxt->node->content != NULL)) return(0);
+ } else if (xmlNodeIsText(lastChild))
+ return(0);
+ else if ((ctxt->node->children != NULL) &&
+ (xmlNodeIsText(ctxt->node->children)))
+ return(0);
+ return(1);
+}
+
+/************************************************************************
+ * *
+ * Extra stuff for namespace support *
+ * Relates to http://www.w3.org/TR/WD-xml-names *
+ * *
+ ************************************************************************/
+
+/**
+ * xmlSplitQName:
+ * @ctxt: an XML parser context
+ * @name: an XML parser context
+ * @prefix: a xmlChar **
+ *
+ * parse an UTF8 encoded XML qualified name string
+ *
+ * [NS 5] QName ::= (Prefix ':')? LocalPart
+ *
+ * [NS 6] Prefix ::= NCName
+ *
+ * [NS 7] LocalPart ::= NCName
+ *
+ * Returns the local part, and prefix is updated
+ * to get the Prefix if any.
+ */
+
+xmlChar *
+xmlSplitQName(xmlParserCtxtPtr ctxt, const xmlChar *name, xmlChar **prefix) {
+ xmlChar buf[XML_MAX_NAMELEN + 5];
+ xmlChar *buffer = NULL;
+ int len = 0;
+ int max = XML_MAX_NAMELEN;
+ xmlChar *ret = NULL;
+ const xmlChar *cur = name;
+ int c;
+
+ if (prefix == NULL) return(NULL);
+ *prefix = NULL;
+
+ if (cur == NULL) return(NULL);
+
+#ifndef XML_XML_NAMESPACE
+ /* xml: prefix is not really a namespace */
+ if ((cur[0] == 'x') && (cur[1] == 'm') &&
+ (cur[2] == 'l') && (cur[3] == ':'))
+ return(xmlStrdup(name));
+#endif
+
+ /* nasty but well=formed */
+ if (cur[0] == ':')
+ return(xmlStrdup(name));
+
+ c = *cur++;
+ while ((c != 0) && (c != ':') && (len < max)) { /* tested bigname.xml */
+ buf[len++] = c;
+ c = *cur++;
+ }
+ if (len >= max) {
+ /*
+ * Okay someone managed to make a huge name, so he's ready to pay
+ * for the processing speed.
+ */
+ max = len * 2;
+
+ buffer = (xmlChar *) xmlMallocAtomic(max * sizeof(xmlChar));
+ if (buffer == NULL) {
+ xmlErrMemory(ctxt, NULL);
+ return(NULL);
+ }
+ memcpy(buffer, buf, len);
+ while ((c != 0) && (c != ':')) { /* tested bigname.xml */
+ if (len + 10 > max) {
+ xmlChar *tmp;
+
+ max *= 2;
+ tmp = (xmlChar *) xmlRealloc(buffer,
+ max * sizeof(xmlChar));
+ if (tmp == NULL) {
+ xmlFree(buffer);
+ xmlErrMemory(ctxt, NULL);
+ return(NULL);
+ }
+ buffer = tmp;
+ }
+ buffer[len++] = c;
+ c = *cur++;
+ }
+ buffer[len] = 0;
+ }
+
+ if ((c == ':') && (*cur == 0)) {
+ if (buffer != NULL)
+ xmlFree(buffer);
+ *prefix = NULL;
+ return(xmlStrdup(name));
+ }
+
+ if (buffer == NULL)
+ ret = xmlStrndup(buf, len);
+ else {
+ ret = buffer;
+ buffer = NULL;
+ max = XML_MAX_NAMELEN;
+ }
+
+
+ if (c == ':') {
+ c = *cur;
+ *prefix = ret;
+ if (c == 0) {
+ return(xmlStrndup(BAD_CAST "", 0));
+ }
+ len = 0;
+
+ /*
+ * Check that the first character is proper to start
+ * a new name
+ */
+ if (!(((c >= 0x61) && (c <= 0x7A)) ||
+ ((c >= 0x41) && (c <= 0x5A)) ||
+ (c == '_') || (c == ':'))) {
+ int l;
+ int first = CUR_SCHAR(cur, l);
+
+ if (!IS_LETTER(first) && (first != '_')) {
+ xmlFatalErrMsgStr(ctxt, XML_NS_ERR_QNAME,
+ "Name %s is not XML Namespace compliant\n",
+ name);
+ }
+ }
+ cur++;
+
+ while ((c != 0) && (len < max)) { /* tested bigname2.xml */
+ buf[len++] = c;
+ c = *cur++;
+ }
+ if (len >= max) {
+ /*
+ * Okay someone managed to make a huge name, so he's ready to pay
+ * for the processing speed.
+ */
+ max = len * 2;
+
+ buffer = (xmlChar *) xmlMallocAtomic(max * sizeof(xmlChar));
+ if (buffer == NULL) {
+ xmlErrMemory(ctxt, NULL);
+ return(NULL);
+ }
+ memcpy(buffer, buf, len);
+ while (c != 0) { /* tested bigname2.xml */
+ if (len + 10 > max) {
+ xmlChar *tmp;
+
+ max *= 2;
+ tmp = (xmlChar *) xmlRealloc(buffer,
+ max * sizeof(xmlChar));
+ if (tmp == NULL) {
+ xmlErrMemory(ctxt, NULL);
+ xmlFree(buffer);
+ return(NULL);
+ }
+ buffer = tmp;
+ }
+ buffer[len++] = c;
+ c = *cur++;
+ }
+ buffer[len] = 0;
+ }
+
+ if (buffer == NULL)
+ ret = xmlStrndup(buf, len);
+ else {
+ ret = buffer;
+ }
+ }
+
+ return(ret);
+}
+
+/************************************************************************
+ * *
+ * The parser itself *
+ * Relates to http://www.w3.org/TR/REC-xml *
+ * *
+ ************************************************************************/
+
+/************************************************************************
+ * *
+ * Routines to parse Name, NCName and NmToken *
+ * *
+ ************************************************************************/
+#ifdef DEBUG
+static unsigned long nbParseName = 0;
+static unsigned long nbParseNmToken = 0;
+static unsigned long nbParseNCName = 0;
+static unsigned long nbParseNCNameComplex = 0;
+static unsigned long nbParseNameComplex = 0;
+static unsigned long nbParseStringName = 0;
+#endif
+
+/*
+ * The two following functions are related to the change of accepted
+ * characters for Name and NmToken in the Revision 5 of XML-1.0
+ * They correspond to the modified production [4] and the new production [4a]
+ * changes in that revision. Also note that the macros used for the
+ * productions Letter, Digit, CombiningChar and Extender are not needed
+ * anymore.
+ * We still keep compatibility to pre-revision5 parsing semantic if the
+ * new XML_PARSE_OLD10 option is given to the parser.
+ */
+static int
+xmlIsNameStartChar(xmlParserCtxtPtr ctxt, int c) {
+ if ((ctxt->options & XML_PARSE_OLD10) == 0) {
+ /*
+ * Use the new checks of production [4] [4a] amd [5] of the
+ * Update 5 of XML-1.0
+ */
+ if ((c != ' ') && (c != '>') && (c != '/') && /* accelerators */
+ (((c >= 'a') && (c <= 'z')) ||
+ ((c >= 'A') && (c <= 'Z')) ||
+ (c == '_') || (c == ':') ||
+ ((c >= 0xC0) && (c <= 0xD6)) ||
+ ((c >= 0xD8) && (c <= 0xF6)) ||
+ ((c >= 0xF8) && (c <= 0x2FF)) ||
+ ((c >= 0x370) && (c <= 0x37D)) ||
+ ((c >= 0x37F) && (c <= 0x1FFF)) ||
+ ((c >= 0x200C) && (c <= 0x200D)) ||
+ ((c >= 0x2070) && (c <= 0x218F)) ||
+ ((c >= 0x2C00) && (c <= 0x2FEF)) ||
+ ((c >= 0x3001) && (c <= 0xD7FF)) ||
+ ((c >= 0xF900) && (c <= 0xFDCF)) ||
+ ((c >= 0xFDF0) && (c <= 0xFFFD)) ||
+ ((c >= 0x10000) && (c <= 0xEFFFF))))
+ return(1);
+ } else {
+ if (IS_LETTER(c) || (c == '_') || (c == ':'))
+ return(1);
+ }
+ return(0);
+}
+
+static int
+xmlIsNameChar(xmlParserCtxtPtr ctxt, int c) {
+ if ((ctxt->options & XML_PARSE_OLD10) == 0) {
+ /*
+ * Use the new checks of production [4] [4a] amd [5] of the
+ * Update 5 of XML-1.0
+ */
+ if ((c != ' ') && (c != '>') && (c != '/') && /* accelerators */
+ (((c >= 'a') && (c <= 'z')) ||
+ ((c >= 'A') && (c <= 'Z')) ||
+ ((c >= '0') && (c <= '9')) || /* !start */
+ (c == '_') || (c == ':') ||
+ (c == '-') || (c == '.') || (c == 0xB7) || /* !start */
+ ((c >= 0xC0) && (c <= 0xD6)) ||
+ ((c >= 0xD8) && (c <= 0xF6)) ||
+ ((c >= 0xF8) && (c <= 0x2FF)) ||
+ ((c >= 0x300) && (c <= 0x36F)) || /* !start */
+ ((c >= 0x370) && (c <= 0x37D)) ||
+ ((c >= 0x37F) && (c <= 0x1FFF)) ||
+ ((c >= 0x200C) && (c <= 0x200D)) ||
+ ((c >= 0x203F) && (c <= 0x2040)) || /* !start */
+ ((c >= 0x2070) && (c <= 0x218F)) ||
+ ((c >= 0x2C00) && (c <= 0x2FEF)) ||
+ ((c >= 0x3001) && (c <= 0xD7FF)) ||
+ ((c >= 0xF900) && (c <= 0xFDCF)) ||
+ ((c >= 0xFDF0) && (c <= 0xFFFD)) ||
+ ((c >= 0x10000) && (c <= 0xEFFFF))))
+ return(1);
+ } else {
+ if ((IS_LETTER(c)) || (IS_DIGIT(c)) ||
+ (c == '.') || (c == '-') ||
+ (c == '_') || (c == ':') ||
+ (IS_COMBINING(c)) ||
+ (IS_EXTENDER(c)))
+ return(1);
+ }
+ return(0);
+}
+
+static xmlChar * xmlParseAttValueInternal(xmlParserCtxtPtr ctxt,
+ int *len, int *alloc, int normalize);
+
+static const xmlChar *
+xmlParseNameComplex(xmlParserCtxtPtr ctxt) {
+ int len = 0, l;
+ int c;
+ int count = 0;
+
+#ifdef DEBUG
+ nbParseNameComplex++;
+#endif
+
+ /*
+ * Handler for more complex cases
+ */
+ GROW;
+ if (ctxt->instate == XML_PARSER_EOF)
+ return(NULL);
+ c = CUR_CHAR(l);
+ if ((ctxt->options & XML_PARSE_OLD10) == 0) {
+ /*
+ * Use the new checks of production [4] [4a] amd [5] of the
+ * Update 5 of XML-1.0
+ */
+ if ((c == ' ') || (c == '>') || (c == '/') || /* accelerators */
+ (!(((c >= 'a') && (c <= 'z')) ||
+ ((c >= 'A') && (c <= 'Z')) ||
+ (c == '_') || (c == ':') ||
+ ((c >= 0xC0) && (c <= 0xD6)) ||
+ ((c >= 0xD8) && (c <= 0xF6)) ||
+ ((c >= 0xF8) && (c <= 0x2FF)) ||
+ ((c >= 0x370) && (c <= 0x37D)) ||
+ ((c >= 0x37F) && (c <= 0x1FFF)) ||
+ ((c >= 0x200C) && (c <= 0x200D)) ||
+ ((c >= 0x2070) && (c <= 0x218F)) ||
+ ((c >= 0x2C00) && (c <= 0x2FEF)) ||
+ ((c >= 0x3001) && (c <= 0xD7FF)) ||
+ ((c >= 0xF900) && (c <= 0xFDCF)) ||
+ ((c >= 0xFDF0) && (c <= 0xFFFD)) ||
+ ((c >= 0x10000) && (c <= 0xEFFFF))))) {
+ return(NULL);
+ }
+ len += l;
+ NEXTL(l);
+ c = CUR_CHAR(l);
+ while ((c != ' ') && (c != '>') && (c != '/') && /* accelerators */
+ (((c >= 'a') && (c <= 'z')) ||
+ ((c >= 'A') && (c <= 'Z')) ||
+ ((c >= '0') && (c <= '9')) || /* !start */
+ (c == '_') || (c == ':') ||
+ (c == '-') || (c == '.') || (c == 0xB7) || /* !start */
+ ((c >= 0xC0) && (c <= 0xD6)) ||
+ ((c >= 0xD8) && (c <= 0xF6)) ||
+ ((c >= 0xF8) && (c <= 0x2FF)) ||
+ ((c >= 0x300) && (c <= 0x36F)) || /* !start */
+ ((c >= 0x370) && (c <= 0x37D)) ||
+ ((c >= 0x37F) && (c <= 0x1FFF)) ||
+ ((c >= 0x200C) && (c <= 0x200D)) ||
+ ((c >= 0x203F) && (c <= 0x2040)) || /* !start */
+ ((c >= 0x2070) && (c <= 0x218F)) ||
+ ((c >= 0x2C00) && (c <= 0x2FEF)) ||
+ ((c >= 0x3001) && (c <= 0xD7FF)) ||
+ ((c >= 0xF900) && (c <= 0xFDCF)) ||
+ ((c >= 0xFDF0) && (c <= 0xFFFD)) ||
+ ((c >= 0x10000) && (c <= 0xEFFFF))
+ )) {
+ if (count++ > XML_PARSER_CHUNK_SIZE) {
+ count = 0;
+ GROW;
+ if (ctxt->instate == XML_PARSER_EOF)
+ return(NULL);
+ }
+ len += l;
+ NEXTL(l);
+ c = CUR_CHAR(l);
+ }
+ } else {
+ if ((c == ' ') || (c == '>') || (c == '/') || /* accelerators */
+ (!IS_LETTER(c) && (c != '_') &&
+ (c != ':'))) {
+ return(NULL);
+ }
+ len += l;
+ NEXTL(l);
+ c = CUR_CHAR(l);
+
+ while ((c != ' ') && (c != '>') && (c != '/') && /* test bigname.xml */
+ ((IS_LETTER(c)) || (IS_DIGIT(c)) ||
+ (c == '.') || (c == '-') ||
+ (c == '_') || (c == ':') ||
+ (IS_COMBINING(c)) ||
+ (IS_EXTENDER(c)))) {
+ if (count++ > XML_PARSER_CHUNK_SIZE) {
+ count = 0;
+ GROW;
+ if (ctxt->instate == XML_PARSER_EOF)
+ return(NULL);
+ }
+ len += l;
+ NEXTL(l);
+ c = CUR_CHAR(l);
+ if (c == 0) {
+ count = 0;
+ GROW;
+ if (ctxt->instate == XML_PARSER_EOF)
+ return(NULL);
+ c = CUR_CHAR(l);
+ }
+ }
+ }
+ if ((len > XML_MAX_NAME_LENGTH) &&
+ ((ctxt->options & XML_PARSE_HUGE) == 0)) {
+ xmlFatalErr(ctxt, XML_ERR_NAME_TOO_LONG, "Name");
+ return(NULL);
+ }
+ if ((*ctxt->input->cur == '\n') && (ctxt->input->cur[-1] == '\r'))
+ return(xmlDictLookup(ctxt->dict, ctxt->input->cur - (len + 1), len));
+ return(xmlDictLookup(ctxt->dict, ctxt->input->cur - len, len));
+}
+
+/**
+ * xmlParseName:
+ * @ctxt: an XML parser context
+ *
+ * parse an XML name.
+ *
+ * [4] NameChar ::= Letter | Digit | '.' | '-' | '_' | ':' |
+ * CombiningChar | Extender
+ *
+ * [5] Name ::= (Letter | '_' | ':') (NameChar)*
+ *
+ * [6] Names ::= Name (#x20 Name)*
+ *
+ * Returns the Name parsed or NULL
+ */
+
+const xmlChar *
+xmlParseName(xmlParserCtxtPtr ctxt) {
+ const xmlChar *in;
+ const xmlChar *ret;
+ int count = 0;
+
+ GROW;
+
+#ifdef DEBUG
+ nbParseName++;
+#endif
+
+ /*
+ * Accelerator for simple ASCII names
+ */
+ in = ctxt->input->cur;
+ if (((*in >= 0x61) && (*in <= 0x7A)) ||
+ ((*in >= 0x41) && (*in <= 0x5A)) ||
+ (*in == '_') || (*in == ':')) {
+ in++;
+ while (((*in >= 0x61) && (*in <= 0x7A)) ||
+ ((*in >= 0x41) && (*in <= 0x5A)) ||
+ ((*in >= 0x30) && (*in <= 0x39)) ||
+ (*in == '_') || (*in == '-') ||
+ (*in == ':') || (*in == '.'))
+ in++;
+ if ((*in > 0) && (*in < 0x80)) {
+ count = in - ctxt->input->cur;
+ if ((count > XML_MAX_NAME_LENGTH) &&
+ ((ctxt->options & XML_PARSE_HUGE) == 0)) {
+ xmlFatalErr(ctxt, XML_ERR_NAME_TOO_LONG, "Name");
+ return(NULL);
+ }
+ ret = xmlDictLookup(ctxt->dict, ctxt->input->cur, count);
+ ctxt->input->cur = in;
+ ctxt->nbChars += count;
+ ctxt->input->col += count;
+ if (ret == NULL)
+ xmlErrMemory(ctxt, NULL);
+ return(ret);
+ }
+ }
+ /* accelerator for special cases */
+ return(xmlParseNameComplex(ctxt));
+}
+
+static const xmlChar *
+xmlParseNCNameComplex(xmlParserCtxtPtr ctxt) {
+ int len = 0, l;
+ int c;
+ int count = 0;
+ size_t startPosition = 0;
+
+#ifdef DEBUG
+ nbParseNCNameComplex++;
+#endif
+
+ /*
+ * Handler for more complex cases
+ */
+ GROW;
+ startPosition = CUR_PTR - BASE_PTR;
+ c = CUR_CHAR(l);
+ if ((c == ' ') || (c == '>') || (c == '/') || /* accelerators */
+ (!xmlIsNameStartChar(ctxt, c) || (c == ':'))) {
+ return(NULL);
+ }
+
+ while ((c != ' ') && (c != '>') && (c != '/') && /* test bigname.xml */
+ (xmlIsNameChar(ctxt, c) && (c != ':'))) {
+ if (count++ > XML_PARSER_CHUNK_SIZE) {
+ if ((len > XML_MAX_NAME_LENGTH) &&
+ ((ctxt->options & XML_PARSE_HUGE) == 0)) {
+ xmlFatalErr(ctxt, XML_ERR_NAME_TOO_LONG, "NCName");
+ return(NULL);
+ }
+ count = 0;
+ GROW;
+ if (ctxt->instate == XML_PARSER_EOF)
+ return(NULL);
+ }
+ len += l;
+ NEXTL(l);
+ c = CUR_CHAR(l);
+ if (c == 0) {
+ count = 0;
+ /*
+ * when shrinking to extend the buffer we really need to preserve
+ * the part of the name we already parsed. Hence rolling back
+ * by current lenght.
+ */
+ ctxt->input->cur -= l;
+ GROW;
+ ctxt->input->cur += l;
+ if (ctxt->instate == XML_PARSER_EOF)
+ return(NULL);
+ c = CUR_CHAR(l);
+ }
+ }
+ if ((len > XML_MAX_NAME_LENGTH) &&
+ ((ctxt->options & XML_PARSE_HUGE) == 0)) {
+ xmlFatalErr(ctxt, XML_ERR_NAME_TOO_LONG, "NCName");
+ return(NULL);
+ }
+ return(xmlDictLookup(ctxt->dict, (BASE_PTR + startPosition), len));
+}
+
+/**
+ * xmlParseNCName:
+ * @ctxt: an XML parser context
+ * @len: length of the string parsed
+ *
+ * parse an XML name.
+ *
+ * [4NS] NCNameChar ::= Letter | Digit | '.' | '-' | '_' |
+ * CombiningChar | Extender
+ *
+ * [5NS] NCName ::= (Letter | '_') (NCNameChar)*
+ *
+ * Returns the Name parsed or NULL
+ */
+
+static const xmlChar *
+xmlParseNCName(xmlParserCtxtPtr ctxt) {
+ const xmlChar *in, *e;
+ const xmlChar *ret;
+ int count = 0;
+
+#ifdef DEBUG
+ nbParseNCName++;
+#endif
+
+ /*
+ * Accelerator for simple ASCII names
+ */
+ in = ctxt->input->cur;
+ e = ctxt->input->end;
+ if ((((*in >= 0x61) && (*in <= 0x7A)) ||
+ ((*in >= 0x41) && (*in <= 0x5A)) ||
+ (*in == '_')) && (in < e)) {
+ in++;
+ while ((((*in >= 0x61) && (*in <= 0x7A)) ||
+ ((*in >= 0x41) && (*in <= 0x5A)) ||
+ ((*in >= 0x30) && (*in <= 0x39)) ||
+ (*in == '_') || (*in == '-') ||
+ (*in == '.')) && (in < e))
+ in++;
+ if (in >= e)
+ goto complex;
+ if ((*in > 0) && (*in < 0x80)) {
+ count = in - ctxt->input->cur;
+ if ((count > XML_MAX_NAME_LENGTH) &&
+ ((ctxt->options & XML_PARSE_HUGE) == 0)) {
+ xmlFatalErr(ctxt, XML_ERR_NAME_TOO_LONG, "NCName");
+ return(NULL);
+ }
+ ret = xmlDictLookup(ctxt->dict, ctxt->input->cur, count);
+ ctxt->input->cur = in;
+ ctxt->nbChars += count;
+ ctxt->input->col += count;
+ if (ret == NULL) {
+ xmlErrMemory(ctxt, NULL);
+ }
+ return(ret);
+ }
+ }
+complex:
+ return(xmlParseNCNameComplex(ctxt));
+}
+
+/**
+ * xmlParseNameAndCompare:
+ * @ctxt: an XML parser context
+ *
+ * parse an XML name and compares for match
+ * (specialized for endtag parsing)
+ *
+ * Returns NULL for an illegal name, (xmlChar*) 1 for success
+ * and the name for mismatch
+ */
+
+static const xmlChar *
+xmlParseNameAndCompare(xmlParserCtxtPtr ctxt, xmlChar const *other) {
+ register const xmlChar *cmp = other;
+ register const xmlChar *in;
+ const xmlChar *ret;
+
+ GROW;
+ if (ctxt->instate == XML_PARSER_EOF)
+ return(NULL);
+
+ in = ctxt->input->cur;
+ while (*in != 0 && *in == *cmp) {
+ ++in;
+ ++cmp;
+ ctxt->input->col++;
+ }
+ if (*cmp == 0 && (*in == '>' || IS_BLANK_CH (*in))) {
+ /* success */
+ ctxt->input->cur = in;
+ return (const xmlChar*) 1;
+ }
+ /* failure (or end of input buffer), check with full function */
+ ret = xmlParseName (ctxt);
+ /* strings coming from the dictionary direct compare possible */
+ if (ret == other) {
+ return (const xmlChar*) 1;
+ }
+ return ret;
+}
+
+/**
+ * xmlParseStringName:
+ * @ctxt: an XML parser context
+ * @str: a pointer to the string pointer (IN/OUT)
+ *
+ * parse an XML name.
+ *
+ * [4] NameChar ::= Letter | Digit | '.' | '-' | '_' | ':' |
+ * CombiningChar | Extender
+ *
+ * [5] Name ::= (Letter | '_' | ':') (NameChar)*
+ *
+ * [6] Names ::= Name (#x20 Name)*
+ *
+ * Returns the Name parsed or NULL. The @str pointer
+ * is updated to the current location in the string.
+ */
+
+static xmlChar *
+xmlParseStringName(xmlParserCtxtPtr ctxt, const xmlChar** str) {
+ xmlChar buf[XML_MAX_NAMELEN + 5];
+ const xmlChar *cur = *str;
+ int len = 0, l;
+ int c;
+
+#ifdef DEBUG
+ nbParseStringName++;
+#endif
+
+ c = CUR_SCHAR(cur, l);
+ if (!xmlIsNameStartChar(ctxt, c)) {
+ return(NULL);
+ }
+
+ COPY_BUF(l,buf,len,c);
+ cur += l;
+ c = CUR_SCHAR(cur, l);
+ while (xmlIsNameChar(ctxt, c)) {
+ COPY_BUF(l,buf,len,c);
+ cur += l;
+ c = CUR_SCHAR(cur, l);
+ if (len >= XML_MAX_NAMELEN) { /* test bigentname.xml */
+ /*
+ * Okay someone managed to make a huge name, so he's ready to pay
+ * for the processing speed.
+ */
+ xmlChar *buffer;
+ int max = len * 2;
+
+ buffer = (xmlChar *) xmlMallocAtomic(max * sizeof(xmlChar));
+ if (buffer == NULL) {
+ xmlErrMemory(ctxt, NULL);
+ return(NULL);
+ }
+ memcpy(buffer, buf, len);
+ while (xmlIsNameChar(ctxt, c)) {
+ if (len + 10 > max) {
+ xmlChar *tmp;
+
+ if ((len > XML_MAX_NAME_LENGTH) &&
+ ((ctxt->options & XML_PARSE_HUGE) == 0)) {
+ xmlFatalErr(ctxt, XML_ERR_NAME_TOO_LONG, "NCName");
+ xmlFree(buffer);
+ return(NULL);
+ }
+ max *= 2;
+ tmp = (xmlChar *) xmlRealloc(buffer,
+ max * sizeof(xmlChar));
+ if (tmp == NULL) {
+ xmlErrMemory(ctxt, NULL);
+ xmlFree(buffer);
+ return(NULL);
+ }
+ buffer = tmp;
+ }
+ COPY_BUF(l,buffer,len,c);
+ cur += l;
+ c = CUR_SCHAR(cur, l);
+ }
+ buffer[len] = 0;
+ *str = cur;
+ return(buffer);
+ }
+ }
+ if ((len > XML_MAX_NAME_LENGTH) &&
+ ((ctxt->options & XML_PARSE_HUGE) == 0)) {
+ xmlFatalErr(ctxt, XML_ERR_NAME_TOO_LONG, "NCName");
+ return(NULL);
+ }
+ *str = cur;
+ return(xmlStrndup(buf, len));
+}
+
+/**
+ * xmlParseNmtoken:
+ * @ctxt: an XML parser context
+ *
+ * parse an XML Nmtoken.
+ *
+ * [7] Nmtoken ::= (NameChar)+
+ *
+ * [8] Nmtokens ::= Nmtoken (#x20 Nmtoken)*
+ *
+ * Returns the Nmtoken parsed or NULL
+ */
+
+xmlChar *
+xmlParseNmtoken(xmlParserCtxtPtr ctxt) {
+ xmlChar buf[XML_MAX_NAMELEN + 5];
+ int len = 0, l;
+ int c;
+ int count = 0;
+
+#ifdef DEBUG
+ nbParseNmToken++;
+#endif
+
+ GROW;
+ if (ctxt->instate == XML_PARSER_EOF)
+ return(NULL);
+ c = CUR_CHAR(l);
+
+ while (xmlIsNameChar(ctxt, c)) {
+ if (count++ > XML_PARSER_CHUNK_SIZE) {
+ count = 0;
+ GROW;
+ }
+ COPY_BUF(l,buf,len,c);
+ NEXTL(l);
+ c = CUR_CHAR(l);
+ if (c == 0) {
+ count = 0;
+ GROW;
+ if (ctxt->instate == XML_PARSER_EOF)
+ return(NULL);
+ c = CUR_CHAR(l);
+ }
+ if (len >= XML_MAX_NAMELEN) {
+ /*
+ * Okay someone managed to make a huge token, so he's ready to pay
+ * for the processing speed.
+ */
+ xmlChar *buffer;
+ int max = len * 2;
+
+ buffer = (xmlChar *) xmlMallocAtomic(max * sizeof(xmlChar));
+ if (buffer == NULL) {
+ xmlErrMemory(ctxt, NULL);
+ return(NULL);
+ }
+ memcpy(buffer, buf, len);
+ while (xmlIsNameChar(ctxt, c)) {
+ if (count++ > XML_PARSER_CHUNK_SIZE) {
+ count = 0;
+ GROW;
+ if (ctxt->instate == XML_PARSER_EOF) {
+ xmlFree(buffer);
+ return(NULL);
+ }
+ }
+ if (len + 10 > max) {
+ xmlChar *tmp;
+
+ if ((max > XML_MAX_NAME_LENGTH) &&
+ ((ctxt->options & XML_PARSE_HUGE) == 0)) {
+ xmlFatalErr(ctxt, XML_ERR_NAME_TOO_LONG, "NmToken");
+ xmlFree(buffer);
+ return(NULL);
+ }
+ max *= 2;
+ tmp = (xmlChar *) xmlRealloc(buffer,
+ max * sizeof(xmlChar));
+ if (tmp == NULL) {
+ xmlErrMemory(ctxt, NULL);
+ xmlFree(buffer);
+ return(NULL);
+ }
+ buffer = tmp;
+ }
+ COPY_BUF(l,buffer,len,c);
+ NEXTL(l);
+ c = CUR_CHAR(l);
+ }
+ buffer[len] = 0;
+ return(buffer);
+ }
+ }
+ if (len == 0)
+ return(NULL);
+ if ((len > XML_MAX_NAME_LENGTH) &&
+ ((ctxt->options & XML_PARSE_HUGE) == 0)) {
+ xmlFatalErr(ctxt, XML_ERR_NAME_TOO_LONG, "NmToken");
+ return(NULL);
+ }
+ return(xmlStrndup(buf, len));
+}
+
+/**
+ * xmlParseEntityValue:
+ * @ctxt: an XML parser context
+ * @orig: if non-NULL store a copy of the original entity value
+ *
+ * parse a value for ENTITY declarations
+ *
+ * [9] EntityValue ::= '"' ([^%&"] | PEReference | Reference)* '"' |
+ * "'" ([^%&'] | PEReference | Reference)* "'"
+ *
+ * Returns the EntityValue parsed with reference substituted or NULL
+ */
+
+xmlChar *
+xmlParseEntityValue(xmlParserCtxtPtr ctxt, xmlChar **orig) {
+ xmlChar *buf = NULL;
+ int len = 0;
+ int size = XML_PARSER_BUFFER_SIZE;
+ int c, l;
+ xmlChar stop;
+ xmlChar *ret = NULL;
+ const xmlChar *cur = NULL;
+ xmlParserInputPtr input;
+
+ if (RAW == '"') stop = '"';
+ else if (RAW == '\'') stop = '\'';
+ else {
+ xmlFatalErr(ctxt, XML_ERR_ENTITY_NOT_STARTED, NULL);
+ return(NULL);
+ }
+ buf = (xmlChar *) xmlMallocAtomic(size * sizeof(xmlChar));
+ if (buf == NULL) {
+ xmlErrMemory(ctxt, NULL);
+ return(NULL);
+ }
+
+ /*
+ * The content of the entity definition is copied in a buffer.
+ */
+
+ ctxt->instate = XML_PARSER_ENTITY_VALUE;
+ input = ctxt->input;
+ GROW;
+ if (ctxt->instate == XML_PARSER_EOF) {
+ xmlFree(buf);
+ return(NULL);
+ }
+ NEXT;
+ c = CUR_CHAR(l);
+ /*
+ * NOTE: 4.4.5 Included in Literal
+ * When a parameter entity reference appears in a literal entity
+ * value, ... a single or double quote character in the replacement
+ * text is always treated as a normal data character and will not
+ * terminate the literal.
+ * In practice it means we stop the loop only when back at parsing
+ * the initial entity and the quote is found
+ */
+ while (((IS_CHAR(c)) && ((c != stop) || /* checked */
+ (ctxt->input != input))) && (ctxt->instate != XML_PARSER_EOF)) {
+ if (len + 5 >= size) {
+ xmlChar *tmp;
+
+ size *= 2;
+ tmp = (xmlChar *) xmlRealloc(buf, size * sizeof(xmlChar));
+ if (tmp == NULL) {
+ xmlErrMemory(ctxt, NULL);
+ xmlFree(buf);
+ return(NULL);
+ }
+ buf = tmp;
+ }
+ COPY_BUF(l,buf,len,c);
+ NEXTL(l);
+ /*
+ * Pop-up of finished entities.
+ */
+ while ((RAW == 0) && (ctxt->inputNr > 1)) /* non input consuming */
+ xmlPopInput(ctxt);
+
+ GROW;
+ c = CUR_CHAR(l);
+ if (c == 0) {
+ GROW;
+ c = CUR_CHAR(l);
+ }
+ }
+ buf[len] = 0;
+ if (ctxt->instate == XML_PARSER_EOF) {
+ xmlFree(buf);
+ return(NULL);
+ }
+
+ /*
+ * Raise problem w.r.t. '&' and '%' being used in non-entities
+ * reference constructs. Note Charref will be handled in
+ * xmlStringDecodeEntities()
+ */
+ cur = buf;
+ while (*cur != 0) { /* non input consuming */
+ if ((*cur == '%') || ((*cur == '&') && (cur[1] != '#'))) {
+ xmlChar *name;
+ xmlChar tmp = *cur;
+
+ cur++;
+ name = xmlParseStringName(ctxt, &cur);
+ if ((name == NULL) || (*cur != ';')) {
+ xmlFatalErrMsgInt(ctxt, XML_ERR_ENTITY_CHAR_ERROR,
+ "EntityValue: '%c' forbidden except for entities references\n",
+ tmp);
+ }
+ if ((tmp == '%') && (ctxt->inSubset == 1) &&
+ (ctxt->inputNr == 1)) {
+ xmlFatalErr(ctxt, XML_ERR_ENTITY_PE_INTERNAL, NULL);
+ }
+ if (name != NULL)
+ xmlFree(name);
+ if (*cur == 0)
+ break;
+ }
+ cur++;
+ }
+
+ /*
+ * Then PEReference entities are substituted.
+ */
+ if (c != stop) {
+ xmlFatalErr(ctxt, XML_ERR_ENTITY_NOT_FINISHED, NULL);
+ xmlFree(buf);
+ } else {
+ NEXT;
+ /*
+ * NOTE: 4.4.7 Bypassed
+ * When a general entity reference appears in the EntityValue in
+ * an entity declaration, it is bypassed and left as is.
+ * so XML_SUBSTITUTE_REF is not set here.
+ */
+ ++ctxt->depth;
+ ret = xmlStringDecodeEntities(ctxt, buf, XML_SUBSTITUTE_PEREF,
+ 0, 0, 0);
+ --ctxt->depth;
+ if (orig != NULL)
+ *orig = buf;
+ else
+ xmlFree(buf);
+ }
+
+ return(ret);
+}
+
+/**
+ * xmlParseAttValueComplex:
+ * @ctxt: an XML parser context
+ * @len: the resulting attribute len
+ * @normalize: wether to apply the inner normalization
+ *
+ * parse a value for an attribute, this is the fallback function
+ * of xmlParseAttValue() when the attribute parsing requires handling
+ * of non-ASCII characters, or normalization compaction.
+ *
+ * Returns the AttValue parsed or NULL. The value has to be freed by the caller.
+ */
+static xmlChar *
+xmlParseAttValueComplex(xmlParserCtxtPtr ctxt, int *attlen, int normalize) {
+ xmlChar limit = 0;
+ xmlChar *buf = NULL;
+ xmlChar *rep = NULL;
+ size_t len = 0;
+ size_t buf_size = 0;
+ int c, l, in_space = 0;
+ xmlChar *current = NULL;
+ xmlEntityPtr ent;
+
+ if (NXT(0) == '"') {
+ ctxt->instate = XML_PARSER_ATTRIBUTE_VALUE;
+ limit = '"';
+ NEXT;
+ } else if (NXT(0) == '\'') {
+ limit = '\'';
+ ctxt->instate = XML_PARSER_ATTRIBUTE_VALUE;
+ NEXT;
+ } else {
+ xmlFatalErr(ctxt, XML_ERR_ATTRIBUTE_NOT_STARTED, NULL);
+ return(NULL);
+ }
+
+ /*
+ * allocate a translation buffer.
+ */
+ buf_size = XML_PARSER_BUFFER_SIZE;
+ buf = (xmlChar *) xmlMallocAtomic(buf_size);
+ if (buf == NULL) goto mem_error;
+
+ /*
+ * OK loop until we reach one of the ending char or a size limit.
+ */
+ c = CUR_CHAR(l);
+ while (((NXT(0) != limit) && /* checked */
+ (IS_CHAR(c)) && (c != '<')) &&
+ (ctxt->instate != XML_PARSER_EOF)) {
+ /*
+ * Impose a reasonable limit on attribute size, unless XML_PARSE_HUGE
+ * special option is given
+ */
+ if ((len > XML_MAX_TEXT_LENGTH) &&
+ ((ctxt->options & XML_PARSE_HUGE) == 0)) {
+ xmlFatalErrMsg(ctxt, XML_ERR_ATTRIBUTE_NOT_FINISHED,
+ "AttValue length too long\n");
+ goto mem_error;
+ }
+ if (c == 0) break;
+ if (c == '&') {
+ in_space = 0;
+ if (NXT(1) == '#') {
+ int val = xmlParseCharRef(ctxt);
+
+ if (val == '&') {
+ if (ctxt->replaceEntities) {
+ if (len + 10 > buf_size) {
+ growBuffer(buf, 10);
+ }
+ buf[len++] = '&';
+ } else {
+ /*
+ * The reparsing will be done in xmlStringGetNodeList()
+ * called by the attribute() function in SAX.c
+ */
+ if (len + 10 > buf_size) {
+ growBuffer(buf, 10);
+ }
+ buf[len++] = '&';
+ buf[len++] = '#';
+ buf[len++] = '3';
+ buf[len++] = '8';
+ buf[len++] = ';';
+ }
+ } else if (val != 0) {
+ if (len + 10 > buf_size) {
+ growBuffer(buf, 10);
+ }
+ len += xmlCopyChar(0, &buf[len], val);
+ }
+ } else {
+ ent = xmlParseEntityRef(ctxt);
+ ctxt->nbentities++;
+ if (ent != NULL)
+ ctxt->nbentities += ent->owner;
+ if ((ent != NULL) &&
+ (ent->etype == XML_INTERNAL_PREDEFINED_ENTITY)) {
+ if (len + 10 > buf_size) {
+ growBuffer(buf, 10);
+ }
+ if ((ctxt->replaceEntities == 0) &&
+ (ent->content[0] == '&')) {
+ buf[len++] = '&';
+ buf[len++] = '#';
+ buf[len++] = '3';
+ buf[len++] = '8';
+ buf[len++] = ';';
+ } else {
+ buf[len++] = ent->content[0];
+ }
+ } else if ((ent != NULL) &&
+ (ctxt->replaceEntities != 0)) {
+ if (ent->etype != XML_INTERNAL_PREDEFINED_ENTITY) {
+ ++ctxt->depth;
+ rep = xmlStringDecodeEntities(ctxt, ent->content,
+ XML_SUBSTITUTE_REF,
+ 0, 0, 0);
+ --ctxt->depth;
+ if (rep != NULL) {
+ current = rep;
+ while (*current != 0) { /* non input consuming */
+ if ((*current == 0xD) || (*current == 0xA) ||
+ (*current == 0x9)) {
+ buf[len++] = 0x20;
+ current++;
+ } else
+ buf[len++] = *current++;
+ if (len + 10 > buf_size) {
+ growBuffer(buf, 10);
+ }
+ }
+ xmlFree(rep);
+ rep = NULL;
+ }
+ } else {
+ if (len + 10 > buf_size) {
+ growBuffer(buf, 10);
+ }
+ if (ent->content != NULL)
+ buf[len++] = ent->content[0];
+ }
+ } else if (ent != NULL) {
+ int i = xmlStrlen(ent->name);
+ const xmlChar *cur = ent->name;
+
+ /*
+ * This may look absurd but is needed to detect
+ * entities problems
+ */
+ if ((ent->etype != XML_INTERNAL_PREDEFINED_ENTITY) &&
+ (ent->content != NULL) && (ent->checked == 0)) {
+ unsigned long oldnbent = ctxt->nbentities;
+
+ ++ctxt->depth;
+ rep = xmlStringDecodeEntities(ctxt, ent->content,
+ XML_SUBSTITUTE_REF, 0, 0, 0);
+ --ctxt->depth;
+
+ ent->checked = (ctxt->nbentities - oldnbent + 1) * 2;
+ if (rep != NULL) {
+ if (xmlStrchr(rep, '<'))
+ ent->checked |= 1;
+ xmlFree(rep);
+ rep = NULL;
+ }
+ }
+
+ /*
+ * Just output the reference
+ */
+ buf[len++] = '&';
+ while (len + i + 10 > buf_size) {
+ growBuffer(buf, i + 10);
+ }
+ for (;i > 0;i--)
+ buf[len++] = *cur++;
+ buf[len++] = ';';
+ }
+ }
+ } else {
+ if ((c == 0x20) || (c == 0xD) || (c == 0xA) || (c == 0x9)) {
+ if ((len != 0) || (!normalize)) {
+ if ((!normalize) || (!in_space)) {
+ COPY_BUF(l,buf,len,0x20);
+ while (len + 10 > buf_size) {
+ growBuffer(buf, 10);
+ }
+ }
+ in_space = 1;
+ }
+ } else {
+ in_space = 0;
+ COPY_BUF(l,buf,len,c);
+ if (len + 10 > buf_size) {
+ growBuffer(buf, 10);
+ }
+ }
+ NEXTL(l);
+ }
+ GROW;
+ c = CUR_CHAR(l);
+ }
+ if (ctxt->instate == XML_PARSER_EOF)
+ goto error;
+
+ if ((in_space) && (normalize)) {
+ while ((len > 0) && (buf[len - 1] == 0x20)) len--;
+ }
+ buf[len] = 0;
+ if (RAW == '<') {
+ xmlFatalErr(ctxt, XML_ERR_LT_IN_ATTRIBUTE, NULL);
+ } else if (RAW != limit) {
+ if ((c != 0) && (!IS_CHAR(c))) {
+ xmlFatalErrMsg(ctxt, XML_ERR_INVALID_CHAR,
+ "invalid character in attribute value\n");
+ } else {
+ xmlFatalErrMsg(ctxt, XML_ERR_ATTRIBUTE_NOT_FINISHED,
+ "AttValue: ' expected\n");
+ }
+ } else
+ NEXT;
+
+ /*
+ * There we potentially risk an overflow, don't allow attribute value of
+ * length more than INT_MAX it is a very reasonnable assumption !
+ */
+ if (len >= INT_MAX) {
+ xmlFatalErrMsg(ctxt, XML_ERR_ATTRIBUTE_NOT_FINISHED,
+ "AttValue length too long\n");
+ goto mem_error;
+ }
+
+ if (attlen != NULL) *attlen = (int) len;
+ return(buf);
+
+mem_error:
+ xmlErrMemory(ctxt, NULL);
+error:
+ if (buf != NULL)
+ xmlFree(buf);
+ if (rep != NULL)
+ xmlFree(rep);
+ return(NULL);
+}
+
+/**
+ * xmlParseAttValue:
+ * @ctxt: an XML parser context
+ *
+ * parse a value for an attribute
+ * Note: the parser won't do substitution of entities here, this
+ * will be handled later in xmlStringGetNodeList
+ *
+ * [10] AttValue ::= '"' ([^<&"] | Reference)* '"' |
+ * "'" ([^<&'] | Reference)* "'"
+ *
+ * 3.3.3 Attribute-Value Normalization:
+ * Before the value of an attribute is passed to the application or
+ * checked for validity, the XML processor must normalize it as follows:
+ * - a character reference is processed by appending the referenced
+ * character to the attribute value
+ * - an entity reference is processed by recursively processing the
+ * replacement text of the entity
+ * - a whitespace character (#x20, #xD, #xA, #x9) is processed by
+ * appending #x20 to the normalized value, except that only a single
+ * #x20 is appended for a "#xD#xA" sequence that is part of an external
+ * parsed entity or the literal entity value of an internal parsed entity
+ * - other characters are processed by appending them to the normalized value
+ * If the declared value is not CDATA, then the XML processor must further
+ * process the normalized attribute value by discarding any leading and
+ * trailing space (#x20) characters, and by replacing sequences of space
+ * (#x20) characters by a single space (#x20) character.
+ * All attributes for which no declaration has been read should be treated
+ * by a non-validating parser as if declared CDATA.
+ *
+ * Returns the AttValue parsed or NULL. The value has to be freed by the caller.
+ */
+
+
+xmlChar *
+xmlParseAttValue(xmlParserCtxtPtr ctxt) {
+ if ((ctxt == NULL) || (ctxt->input == NULL)) return(NULL);
+ return(xmlParseAttValueInternal(ctxt, NULL, NULL, 0));
+}
+
+/**
+ * xmlParseSystemLiteral:
+ * @ctxt: an XML parser context
+ *
+ * parse an XML Literal
+ *
+ * [11] SystemLiteral ::= ('"' [^"]* '"') | ("'" [^']* "'")
+ *
+ * Returns the SystemLiteral parsed or NULL
+ */
+
+xmlChar *
+xmlParseSystemLiteral(xmlParserCtxtPtr ctxt) {
+ xmlChar *buf = NULL;
+ int len = 0;
+ int size = XML_PARSER_BUFFER_SIZE;
+ int cur, l;
+ xmlChar stop;
+ int state = ctxt->instate;
+ int count = 0;
+
+ SHRINK;
+ if (RAW == '"') {
+ NEXT;
+ stop = '"';
+ } else if (RAW == '\'') {
+ NEXT;
+ stop = '\'';
+ } else {
+ xmlFatalErr(ctxt, XML_ERR_LITERAL_NOT_STARTED, NULL);
+ return(NULL);
+ }
+
+ buf = (xmlChar *) xmlMallocAtomic(size * sizeof(xmlChar));
+ if (buf == NULL) {
+ xmlErrMemory(ctxt, NULL);
+ return(NULL);
+ }
+ ctxt->instate = XML_PARSER_SYSTEM_LITERAL;
+ cur = CUR_CHAR(l);
+ while ((IS_CHAR(cur)) && (cur != stop)) { /* checked */
+ if (len + 5 >= size) {
+ xmlChar *tmp;
+
+ if ((size > XML_MAX_NAME_LENGTH) &&
+ ((ctxt->options & XML_PARSE_HUGE) == 0)) {
+ xmlFatalErr(ctxt, XML_ERR_NAME_TOO_LONG, "SystemLiteral");
+ xmlFree(buf);
+ ctxt->instate = (xmlParserInputState) state;
+ return(NULL);
+ }
+ size *= 2;
+ tmp = (xmlChar *) xmlRealloc(buf, size * sizeof(xmlChar));
+ if (tmp == NULL) {
+ xmlFree(buf);
+ xmlErrMemory(ctxt, NULL);
+ ctxt->instate = (xmlParserInputState) state;
+ return(NULL);
+ }
+ buf = tmp;
+ }
+ count++;
+ if (count > 50) {
+ GROW;
+ count = 0;
+ if (ctxt->instate == XML_PARSER_EOF) {
+ xmlFree(buf);
+ return(NULL);
+ }
+ }
+ COPY_BUF(l,buf,len,cur);
+ NEXTL(l);
+ cur = CUR_CHAR(l);
+ if (cur == 0) {
+ GROW;
+ SHRINK;
+ cur = CUR_CHAR(l);
+ }
+ }
+ buf[len] = 0;
+ ctxt->instate = (xmlParserInputState) state;
+ if (!IS_CHAR(cur)) {
+ xmlFatalErr(ctxt, XML_ERR_LITERAL_NOT_FINISHED, NULL);
+ } else {
+ NEXT;
+ }
+ return(buf);
+}
+
+/**
+ * xmlParsePubidLiteral:
+ * @ctxt: an XML parser context
+ *
+ * parse an XML public literal
+ *
+ * [12] PubidLiteral ::= '"' PubidChar* '"' | "'" (PubidChar - "'")* "'"
+ *
+ * Returns the PubidLiteral parsed or NULL.
+ */
+
+xmlChar *
+xmlParsePubidLiteral(xmlParserCtxtPtr ctxt) {
+ xmlChar *buf = NULL;
+ int len = 0;
+ int size = XML_PARSER_BUFFER_SIZE;
+ xmlChar cur;
+ xmlChar stop;
+ int count = 0;
+ xmlParserInputState oldstate = ctxt->instate;
+
+ SHRINK;
+ if (RAW == '"') {
+ NEXT;
+ stop = '"';
+ } else if (RAW == '\'') {
+ NEXT;
+ stop = '\'';
+ } else {
+ xmlFatalErr(ctxt, XML_ERR_LITERAL_NOT_STARTED, NULL);
+ return(NULL);
+ }
+ buf = (xmlChar *) xmlMallocAtomic(size * sizeof(xmlChar));
+ if (buf == NULL) {
+ xmlErrMemory(ctxt, NULL);
+ return(NULL);
+ }
+ ctxt->instate = XML_PARSER_PUBLIC_LITERAL;
+ cur = CUR;
+ while ((IS_PUBIDCHAR_CH(cur)) && (cur != stop)) { /* checked */
+ if (len + 1 >= size) {
+ xmlChar *tmp;
+
+ if ((size > XML_MAX_NAME_LENGTH) &&
+ ((ctxt->options & XML_PARSE_HUGE) == 0)) {
+ xmlFatalErr(ctxt, XML_ERR_NAME_TOO_LONG, "Public ID");
+ xmlFree(buf);
+ return(NULL);
+ }
+ size *= 2;
+ tmp = (xmlChar *) xmlRealloc(buf, size * sizeof(xmlChar));
+ if (tmp == NULL) {
+ xmlErrMemory(ctxt, NULL);
+ xmlFree(buf);
+ return(NULL);
+ }
+ buf = tmp;
+ }
+ buf[len++] = cur;
+ count++;
+ if (count > 50) {
+ GROW;
+ count = 0;
+ if (ctxt->instate == XML_PARSER_EOF) {
+ xmlFree(buf);
+ return(NULL);
+ }
+ }
+ NEXT;
+ cur = CUR;
+ if (cur == 0) {
+ GROW;
+ SHRINK;
+ cur = CUR;
+ }
+ }
+ buf[len] = 0;
+ if (cur != stop) {
+ xmlFatalErr(ctxt, XML_ERR_LITERAL_NOT_FINISHED, NULL);
+ } else {
+ NEXT;
+ }
+ ctxt->instate = oldstate;
+ return(buf);
+}
+
+static void xmlParseCharDataComplex(xmlParserCtxtPtr ctxt, int cdata);
+
+/*
+ * used for the test in the inner loop of the char data testing
+ */
+static const unsigned char test_char_data[256] = {
+ 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+ 0x00, 0x09, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, /* 0x9, CR/LF separated */
+ 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+ 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+ 0x20, 0x21, 0x22, 0x23, 0x24, 0x25, 0x00, 0x27, /* & */
+ 0x28, 0x29, 0x2A, 0x2B, 0x2C, 0x2D, 0x2E, 0x2F,
+ 0x30, 0x31, 0x32, 0x33, 0x34, 0x35, 0x36, 0x37,
+ 0x38, 0x39, 0x3A, 0x3B, 0x00, 0x3D, 0x3E, 0x3F, /* < */
+ 0x40, 0x41, 0x42, 0x43, 0x44, 0x45, 0x46, 0x47,
+ 0x48, 0x49, 0x4A, 0x4B, 0x4C, 0x4D, 0x4E, 0x4F,
+ 0x50, 0x51, 0x52, 0x53, 0x54, 0x55, 0x56, 0x57,
+ 0x58, 0x59, 0x5A, 0x5B, 0x5C, 0x00, 0x5E, 0x5F, /* ] */
+ 0x60, 0x61, 0x62, 0x63, 0x64, 0x65, 0x66, 0x67,
+ 0x68, 0x69, 0x6A, 0x6B, 0x6C, 0x6D, 0x6E, 0x6F,
+ 0x70, 0x71, 0x72, 0x73, 0x74, 0x75, 0x76, 0x77,
+ 0x78, 0x79, 0x7A, 0x7B, 0x7C, 0x7D, 0x7E, 0x7F,
+ 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, /* non-ascii */
+ 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+ 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+ 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+ 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+ 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+ 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+ 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+ 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+ 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+ 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+ 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+ 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+ 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+ 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+ 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00
+};
+
+/**
+ * xmlParseCharData:
+ * @ctxt: an XML parser context
+ * @cdata: int indicating whether we are within a CDATA section
+ *
+ * parse a CharData section.
+ * if we are within a CDATA section ']]>' marks an end of section.
+ *
+ * The right angle bracket (>) may be represented using the string "&gt;",
+ * and must, for compatibility, be escaped using "&gt;" or a character
+ * reference when it appears in the string "]]>" in content, when that
+ * string is not marking the end of a CDATA section.
+ *
+ * [14] CharData ::= [^<&]* - ([^<&]* ']]>' [^<&]*)
+ */
+
+void
+xmlParseCharData(xmlParserCtxtPtr ctxt, int cdata) {
+ const xmlChar *in;
+ int nbchar = 0;
+ int line = ctxt->input->line;
+ int col = ctxt->input->col;
+ int ccol;
+
+ SHRINK;
+ GROW;
+ /*
+ * Accelerated common case where input don't need to be
+ * modified before passing it to the handler.
+ */
+ if (!cdata) {
+ in = ctxt->input->cur;
+ do {
+get_more_space:
+ while (*in == 0x20) { in++; ctxt->input->col++; }
+ if (*in == 0xA) {
+ do {
+ ctxt->input->line++; ctxt->input->col = 1;
+ in++;
+ } while (*in == 0xA);
+ goto get_more_space;
+ }
+ if (*in == '<') {
+ nbchar = in - ctxt->input->cur;
+ if (nbchar > 0) {
+ const xmlChar *tmp = ctxt->input->cur;
+ ctxt->input->cur = in;
+
+ if ((ctxt->sax != NULL) &&
+ (ctxt->sax->ignorableWhitespace !=
+ ctxt->sax->characters)) {
+ if (areBlanks(ctxt, tmp, nbchar, 1)) {
+ if (ctxt->sax->ignorableWhitespace != NULL)
+ ctxt->sax->ignorableWhitespace(ctxt->userData,
+ tmp, nbchar);
+ } else {
+ if (ctxt->sax->characters != NULL)
+ ctxt->sax->characters(ctxt->userData,
+ tmp, nbchar);
+ if (*ctxt->space == -1)
+ *ctxt->space = -2;
+ }
+ } else if ((ctxt->sax != NULL) &&
+ (ctxt->sax->characters != NULL)) {
+ ctxt->sax->characters(ctxt->userData,
+ tmp, nbchar);
+ }
+ }
+ return;
+ }
+
+get_more:
+ ccol = ctxt->input->col;
+ while (test_char_data[*in]) {
+ in++;
+ ccol++;
+ }
+ ctxt->input->col = ccol;
+ if (*in == 0xA) {
+ do {
+ ctxt->input->line++; ctxt->input->col = 1;
+ in++;
+ } while (*in == 0xA);
+ goto get_more;
+ }
+ if (*in == ']') {
+ if ((in[1] == ']') && (in[2] == '>')) {
+ xmlFatalErr(ctxt, XML_ERR_MISPLACED_CDATA_END, NULL);
+ ctxt->input->cur = in;
+ return;
+ }
+ in++;
+ ctxt->input->col++;
+ goto get_more;
+ }
+ nbchar = in - ctxt->input->cur;
+ if (nbchar > 0) {
+ if ((ctxt->sax != NULL) &&
+ (ctxt->sax->ignorableWhitespace !=
+ ctxt->sax->characters) &&
+ (IS_BLANK_CH(*ctxt->input->cur))) {
+ const xmlChar *tmp = ctxt->input->cur;
+ ctxt->input->cur = in;
+
+ if (areBlanks(ctxt, tmp, nbchar, 0)) {
+ if (ctxt->sax->ignorableWhitespace != NULL)
+ ctxt->sax->ignorableWhitespace(ctxt->userData,
+ tmp, nbchar);
+ } else {
+ if (ctxt->sax->characters != NULL)
+ ctxt->sax->characters(ctxt->userData,
+ tmp, nbchar);
+ if (*ctxt->space == -1)
+ *ctxt->space = -2;
+ }
+ line = ctxt->input->line;
+ col = ctxt->input->col;
+ } else if (ctxt->sax != NULL) {
+ if (ctxt->sax->characters != NULL)
+ ctxt->sax->characters(ctxt->userData,
+ ctxt->input->cur, nbchar);
+ line = ctxt->input->line;
+ col = ctxt->input->col;
+ }
+ /* something really bad happened in the SAX callback */
+ if (ctxt->instate != XML_PARSER_CONTENT)
+ return;
+ }
+ ctxt->input->cur = in;
+ if (*in == 0xD) {
+ in++;
+ if (*in == 0xA) {
+ ctxt->input->cur = in;
+ in++;
+ ctxt->input->line++; ctxt->input->col = 1;
+ continue; /* while */
+ }
+ in--;
+ }
+ if (*in == '<') {
+ return;
+ }
+ if (*in == '&') {
+ return;
+ }
+ SHRINK;
+ GROW;
+ if (ctxt->instate == XML_PARSER_EOF)
+ return;
+ in = ctxt->input->cur;
+ } while (((*in >= 0x20) && (*in <= 0x7F)) || (*in == 0x09));
+ nbchar = 0;
+ }
+ ctxt->input->line = line;
+ ctxt->input->col = col;
+ xmlParseCharDataComplex(ctxt, cdata);
+}
+
+/**
+ * xmlParseCharDataComplex:
+ * @ctxt: an XML parser context
+ * @cdata: int indicating whether we are within a CDATA section
+ *
+ * parse a CharData section.this is the fallback function
+ * of xmlParseCharData() when the parsing requires handling
+ * of non-ASCII characters.
+ */
+static void
+xmlParseCharDataComplex(xmlParserCtxtPtr ctxt, int cdata) {
+ xmlChar buf[XML_PARSER_BIG_BUFFER_SIZE + 5];
+ int nbchar = 0;
+ int cur, l;
+ int count = 0;
+
+ SHRINK;
+ GROW;
+ cur = CUR_CHAR(l);
+ while ((cur != '<') && /* checked */
+ (cur != '&') &&
+ (IS_CHAR(cur))) /* test also done in xmlCurrentChar() */ {
+ if ((cur == ']') && (NXT(1) == ']') &&
+ (NXT(2) == '>')) {
+ if (cdata) break;
+ else {
+ xmlFatalErr(ctxt, XML_ERR_MISPLACED_CDATA_END, NULL);
+ }
+ }
+ COPY_BUF(l,buf,nbchar,cur);
+ if (nbchar >= XML_PARSER_BIG_BUFFER_SIZE) {
+ buf[nbchar] = 0;
+
+ /*
+ * OK the segment is to be consumed as chars.
+ */
+ if ((ctxt->sax != NULL) && (!ctxt->disableSAX)) {
+ if (areBlanks(ctxt, buf, nbchar, 0)) {
+ if (ctxt->sax->ignorableWhitespace != NULL)
+ ctxt->sax->ignorableWhitespace(ctxt->userData,
+ buf, nbchar);
+ } else {
+ if (ctxt->sax->characters != NULL)
+ ctxt->sax->characters(ctxt->userData, buf, nbchar);
+ if ((ctxt->sax->characters !=
+ ctxt->sax->ignorableWhitespace) &&
+ (*ctxt->space == -1))
+ *ctxt->space = -2;
+ }
+ }
+ nbchar = 0;
+ /* something really bad happened in the SAX callback */
+ if (ctxt->instate != XML_PARSER_CONTENT)
+ return;
+ }
+ count++;
+ if (count > 50) {
+ GROW;
+ count = 0;
+ if (ctxt->instate == XML_PARSER_EOF)
+ return;
+ }
+ NEXTL(l);
+ cur = CUR_CHAR(l);
+ }
+ if (nbchar != 0) {
+ buf[nbchar] = 0;
+ /*
+ * OK the segment is to be consumed as chars.
+ */
+ if ((ctxt->sax != NULL) && (!ctxt->disableSAX)) {
+ if (areBlanks(ctxt, buf, nbchar, 0)) {
+ if (ctxt->sax->ignorableWhitespace != NULL)
+ ctxt->sax->ignorableWhitespace(ctxt->userData, buf, nbchar);
+ } else {
+ if (ctxt->sax->characters != NULL)
+ ctxt->sax->characters(ctxt->userData, buf, nbchar);
+ if ((ctxt->sax->characters != ctxt->sax->ignorableWhitespace) &&
+ (*ctxt->space == -1))
+ *ctxt->space = -2;
+ }
+ }
+ }
+ if ((cur != 0) && (!IS_CHAR(cur))) {
+ /* Generate the error and skip the offending character */
+ xmlFatalErrMsgInt(ctxt, XML_ERR_INVALID_CHAR,
+ "PCDATA invalid Char value %d\n",
+ cur);
+ NEXTL(l);
+ }
+}
+
+/**
+ * xmlParseExternalID:
+ * @ctxt: an XML parser context
+ * @publicID: a xmlChar** receiving PubidLiteral
+ * @strict: indicate whether we should restrict parsing to only
+ * production [75], see NOTE below
+ *
+ * Parse an External ID or a Public ID
+ *
+ * NOTE: Productions [75] and [83] interact badly since [75] can generate
+ * 'PUBLIC' S PubidLiteral S SystemLiteral
+ *
+ * [75] ExternalID ::= 'SYSTEM' S SystemLiteral
+ * | 'PUBLIC' S PubidLiteral S SystemLiteral
+ *
+ * [83] PublicID ::= 'PUBLIC' S PubidLiteral
+ *
+ * Returns the function returns SystemLiteral and in the second
+ * case publicID receives PubidLiteral, is strict is off
+ * it is possible to return NULL and have publicID set.
+ */
+
+xmlChar *
+xmlParseExternalID(xmlParserCtxtPtr ctxt, xmlChar **publicID, int strict) {
+ xmlChar *URI = NULL;
+
+ SHRINK;
+
+ *publicID = NULL;
+ if (CMP6(CUR_PTR, 'S', 'Y', 'S', 'T', 'E', 'M')) {
+ SKIP(6);
+ if (!IS_BLANK_CH(CUR)) {
+ xmlFatalErrMsg(ctxt, XML_ERR_SPACE_REQUIRED,
+ "Space required after 'SYSTEM'\n");
+ }
+ SKIP_BLANKS;
+ URI = xmlParseSystemLiteral(ctxt);
+ if (URI == NULL) {
+ xmlFatalErr(ctxt, XML_ERR_URI_REQUIRED, NULL);
+ }
+ } else if (CMP6(CUR_PTR, 'P', 'U', 'B', 'L', 'I', 'C')) {
+ SKIP(6);
+ if (!IS_BLANK_CH(CUR)) {
+ xmlFatalErrMsg(ctxt, XML_ERR_SPACE_REQUIRED,
+ "Space required after 'PUBLIC'\n");
+ }
+ SKIP_BLANKS;
+ *publicID = xmlParsePubidLiteral(ctxt);
+ if (*publicID == NULL) {
+ xmlFatalErr(ctxt, XML_ERR_PUBID_REQUIRED, NULL);
+ }
+ if (strict) {
+ /*
+ * We don't handle [83] so "S SystemLiteral" is required.
+ */
+ if (!IS_BLANK_CH(CUR)) {
+ xmlFatalErrMsg(ctxt, XML_ERR_SPACE_REQUIRED,
+ "Space required after the Public Identifier\n");
+ }
+ } else {
+ /*
+ * We handle [83] so we return immediately, if
+ * "S SystemLiteral" is not detected. From a purely parsing
+ * point of view that's a nice mess.
+ */
+ const xmlChar *ptr;
+ GROW;
+
+ ptr = CUR_PTR;
+ if (!IS_BLANK_CH(*ptr)) return(NULL);
+
+ while (IS_BLANK_CH(*ptr)) ptr++; /* TODO: dangerous, fix ! */
+ if ((*ptr != '\'') && (*ptr != '"')) return(NULL);
+ }
+ SKIP_BLANKS;
+ URI = xmlParseSystemLiteral(ctxt);
+ if (URI == NULL) {
+ xmlFatalErr(ctxt, XML_ERR_URI_REQUIRED, NULL);
+ }
+ }
+ return(URI);
+}
+
+/**
+ * xmlParseCommentComplex:
+ * @ctxt: an XML parser context
+ * @buf: the already parsed part of the buffer
+ * @len: number of bytes filles in the buffer
+ * @size: allocated size of the buffer
+ *
+ * Skip an XML (SGML) comment <!-- .... -->
+ * The spec says that "For compatibility, the string "--" (double-hyphen)
+ * must not occur within comments. "
+ * This is the slow routine in case the accelerator for ascii didn't work
+ *
+ * [15] Comment ::= '<!--' ((Char - '-') | ('-' (Char - '-')))* '-->'
+ */
+static void
+xmlParseCommentComplex(xmlParserCtxtPtr ctxt, xmlChar *buf,
+ size_t len, size_t size) {
+ int q, ql;
+ int r, rl;
+ int cur, l;
+ size_t count = 0;
+ int inputid;
+
+ inputid = ctxt->input->id;
+
+ if (buf == NULL) {
+ len = 0;
+ size = XML_PARSER_BUFFER_SIZE;
+ buf = (xmlChar *) xmlMallocAtomic(size * sizeof(xmlChar));
+ if (buf == NULL) {
+ xmlErrMemory(ctxt, NULL);
+ return;
+ }
+ }
+ GROW; /* Assure there's enough input data */
+ q = CUR_CHAR(ql);
+ if (q == 0)
+ goto not_terminated;
+ if (!IS_CHAR(q)) {
+ xmlFatalErrMsgInt(ctxt, XML_ERR_INVALID_CHAR,
+ "xmlParseComment: invalid xmlChar value %d\n",
+ q);
+ xmlFree (buf);
+ return;
+ }
+ NEXTL(ql);
+ r = CUR_CHAR(rl);
+ if (r == 0)
+ goto not_terminated;
+ if (!IS_CHAR(r)) {
+ xmlFatalErrMsgInt(ctxt, XML_ERR_INVALID_CHAR,
+ "xmlParseComment: invalid xmlChar value %d\n",
+ q);
+ xmlFree (buf);
+ return;
+ }
+ NEXTL(rl);
+ cur = CUR_CHAR(l);
+ if (cur == 0)
+ goto not_terminated;
+ while (IS_CHAR(cur) && /* checked */
+ ((cur != '>') ||
+ (r != '-') || (q != '-'))) {
+ if ((r == '-') && (q == '-')) {
+ xmlFatalErr(ctxt, XML_ERR_HYPHEN_IN_COMMENT, NULL);
+ }
+ if ((len > XML_MAX_TEXT_LENGTH) &&
+ ((ctxt->options & XML_PARSE_HUGE) == 0)) {
+ xmlFatalErrMsgStr(ctxt, XML_ERR_COMMENT_NOT_FINISHED,
+ "Comment too big found", NULL);
+ xmlFree (buf);
+ return;
+ }
+ if (len + 5 >= size) {
+ xmlChar *new_buf;
+ size_t new_size;
+
+ new_size = size * 2;
+ new_buf = (xmlChar *) xmlRealloc(buf, new_size);
+ if (new_buf == NULL) {
+ xmlFree (buf);
+ xmlErrMemory(ctxt, NULL);
+ return;
+ }
+ buf = new_buf;
+ size = new_size;
+ }
+ COPY_BUF(ql,buf,len,q);
+ q = r;
+ ql = rl;
+ r = cur;
+ rl = l;
+
+ count++;
+ if (count > 50) {
+ GROW;
+ count = 0;
+ if (ctxt->instate == XML_PARSER_EOF) {
+ xmlFree(buf);
+ return;
+ }
+ }
+ NEXTL(l);
+ cur = CUR_CHAR(l);
+ if (cur == 0) {
+ SHRINK;
+ GROW;
+ cur = CUR_CHAR(l);
+ }
+ }
+ buf[len] = 0;
+ if (cur == 0) {
+ xmlFatalErrMsgStr(ctxt, XML_ERR_COMMENT_NOT_FINISHED,
+ "Comment not terminated \n<!--%.50s\n", buf);
+ } else if (!IS_CHAR(cur)) {
+ xmlFatalErrMsgInt(ctxt, XML_ERR_INVALID_CHAR,
+ "xmlParseComment: invalid xmlChar value %d\n",
+ cur);
+ } else {
+ if (inputid != ctxt->input->id) {
+ xmlFatalErrMsg(ctxt, XML_ERR_ENTITY_BOUNDARY,
+ "Comment doesn't start and stop in the same entity\n");
+ }
+ NEXT;
+ if ((ctxt->sax != NULL) && (ctxt->sax->comment != NULL) &&
+ (!ctxt->disableSAX))
+ ctxt->sax->comment(ctxt->userData, buf);
+ }
+ xmlFree(buf);
+ return;
+not_terminated:
+ xmlFatalErrMsgStr(ctxt, XML_ERR_COMMENT_NOT_FINISHED,
+ "Comment not terminated\n", NULL);
+ xmlFree(buf);
+ return;
+}
+
+/**
+ * xmlParseComment:
+ * @ctxt: an XML parser context
+ *
+ * Skip an XML (SGML) comment <!-- .... -->
+ * The spec says that "For compatibility, the string "--" (double-hyphen)
+ * must not occur within comments. "
+ *
+ * [15] Comment ::= '<!--' ((Char - '-') | ('-' (Char - '-')))* '-->'
+ */
+void
+xmlParseComment(xmlParserCtxtPtr ctxt) {
+ xmlChar *buf = NULL;
+ size_t size = XML_PARSER_BUFFER_SIZE;
+ size_t len = 0;
+ xmlParserInputState state;
+ const xmlChar *in;
+ size_t nbchar = 0;
+ int ccol;
+ int inputid;
+
+ /*
+ * Check that there is a comment right here.
+ */
+ if ((RAW != '<') || (NXT(1) != '!') ||
+ (NXT(2) != '-') || (NXT(3) != '-')) return;
+ state = ctxt->instate;
+ ctxt->instate = XML_PARSER_COMMENT;
+ inputid = ctxt->input->id;
+ SKIP(4);
+ SHRINK;
+ GROW;
+
+ /*
+ * Accelerated common case where input don't need to be
+ * modified before passing it to the handler.
+ */
+ in = ctxt->input->cur;
+ do {
+ if (*in == 0xA) {
+ do {
+ ctxt->input->line++; ctxt->input->col = 1;
+ in++;
+ } while (*in == 0xA);
+ }
+get_more:
+ ccol = ctxt->input->col;
+ while (((*in > '-') && (*in <= 0x7F)) ||
+ ((*in >= 0x20) && (*in < '-')) ||
+ (*in == 0x09)) {
+ in++;
+ ccol++;
+ }
+ ctxt->input->col = ccol;
+ if (*in == 0xA) {
+ do {
+ ctxt->input->line++; ctxt->input->col = 1;
+ in++;
+ } while (*in == 0xA);
+ goto get_more;
+ }
+ nbchar = in - ctxt->input->cur;
+ /*
+ * save current set of data
+ */
+ if (nbchar > 0) {
+ if ((ctxt->sax != NULL) &&
+ (ctxt->sax->comment != NULL)) {
+ if (buf == NULL) {
+ if ((*in == '-') && (in[1] == '-'))
+ size = nbchar + 1;
+ else
+ size = XML_PARSER_BUFFER_SIZE + nbchar;
+ buf = (xmlChar *) xmlMallocAtomic(size * sizeof(xmlChar));
+ if (buf == NULL) {
+ xmlErrMemory(ctxt, NULL);
+ ctxt->instate = state;
+ return;
+ }
+ len = 0;
+ } else if (len + nbchar + 1 >= size) {
+ xmlChar *new_buf;
+ size += len + nbchar + XML_PARSER_BUFFER_SIZE;
+ new_buf = (xmlChar *) xmlRealloc(buf,
+ size * sizeof(xmlChar));
+ if (new_buf == NULL) {
+ xmlFree (buf);
+ xmlErrMemory(ctxt, NULL);
+ ctxt->instate = state;
+ return;
+ }
+ buf = new_buf;
+ }
+ memcpy(&buf[len], ctxt->input->cur, nbchar);
+ len += nbchar;
+ buf[len] = 0;
+ }
+ }
+ if ((len > XML_MAX_TEXT_LENGTH) &&
+ ((ctxt->options & XML_PARSE_HUGE) == 0)) {
+ xmlFatalErrMsgStr(ctxt, XML_ERR_COMMENT_NOT_FINISHED,
+ "Comment too big found", NULL);
+ xmlFree (buf);
+ return;
+ }
+ ctxt->input->cur = in;
+ if (*in == 0xA) {
+ in++;
+ ctxt->input->line++; ctxt->input->col = 1;
+ }
+ if (*in == 0xD) {
+ in++;
+ if (*in == 0xA) {
+ ctxt->input->cur = in;
+ in++;
+ ctxt->input->line++; ctxt->input->col = 1;
+ continue; /* while */
+ }
+ in--;
+ }
+ SHRINK;
+ GROW;
+ if (ctxt->instate == XML_PARSER_EOF) {
+ xmlFree(buf);
+ return;
+ }
+ in = ctxt->input->cur;
+ if (*in == '-') {
+ if (in[1] == '-') {
+ if (in[2] == '>') {
+ if (ctxt->input->id != inputid) {
+ xmlFatalErrMsg(ctxt, XML_ERR_ENTITY_BOUNDARY,
+ "comment doesn't start and stop in the same entity\n");
+ }
+ SKIP(3);
+ if ((ctxt->sax != NULL) && (ctxt->sax->comment != NULL) &&
+ (!ctxt->disableSAX)) {
+ if (buf != NULL)
+ ctxt->sax->comment(ctxt->userData, buf);
+ else
+ ctxt->sax->comment(ctxt->userData, BAD_CAST "");
+ }
+ if (buf != NULL)
+ xmlFree(buf);
+ if (ctxt->instate != XML_PARSER_EOF)
+ ctxt->instate = state;
+ return;
+ }
+ if (buf != NULL) {
+ xmlFatalErrMsgStr(ctxt, XML_ERR_HYPHEN_IN_COMMENT,
+ "Double hyphen within comment: "
+ "<!--%.50s\n",
+ buf);
+ } else
+ xmlFatalErrMsgStr(ctxt, XML_ERR_HYPHEN_IN_COMMENT,
+ "Double hyphen within comment\n", NULL);
+ in++;
+ ctxt->input->col++;
+ }
+ in++;
+ ctxt->input->col++;
+ goto get_more;
+ }
+ } while (((*in >= 0x20) && (*in <= 0x7F)) || (*in == 0x09));
+ xmlParseCommentComplex(ctxt, buf, len, size);
+ ctxt->instate = state;
+ return;
+}
+
+
+/**
+ * xmlParsePITarget:
+ * @ctxt: an XML parser context
+ *
+ * parse the name of a PI
+ *
+ * [17] PITarget ::= Name - (('X' | 'x') ('M' | 'm') ('L' | 'l'))
+ *
+ * Returns the PITarget name or NULL
+ */
+
+const xmlChar *
+xmlParsePITarget(xmlParserCtxtPtr ctxt) {
+ const xmlChar *name;
+
+ name = xmlParseName(ctxt);
+ if ((name != NULL) &&
+ ((name[0] == 'x') || (name[0] == 'X')) &&
+ ((name[1] == 'm') || (name[1] == 'M')) &&
+ ((name[2] == 'l') || (name[2] == 'L'))) {
+ int i;
+ if ((name[0] == 'x') && (name[1] == 'm') &&
+ (name[2] == 'l') && (name[3] == 0)) {
+ xmlFatalErrMsg(ctxt, XML_ERR_RESERVED_XML_NAME,
+ "XML declaration allowed only at the start of the document\n");
+ return(name);
+ } else if (name[3] == 0) {
+ xmlFatalErr(ctxt, XML_ERR_RESERVED_XML_NAME, NULL);
+ return(name);
+ }
+ for (i = 0;;i++) {
+ if (xmlW3CPIs[i] == NULL) break;
+ if (xmlStrEqual(name, (const xmlChar *)xmlW3CPIs[i]))
+ return(name);
+ }
+ xmlWarningMsg(ctxt, XML_ERR_RESERVED_XML_NAME,
+ "xmlParsePITarget: invalid name prefix 'xml'\n",
+ NULL, NULL);
+ }
+ if ((name != NULL) && (xmlStrchr(name, ':') != NULL)) {
+ xmlNsErr(ctxt, XML_NS_ERR_COLON,
+ "colons are forbidden from PI names '%s'\n", name, NULL, NULL);
+ }
+ return(name);
+}
+
+#ifdef LIBXML_CATALOG_ENABLED
+/**
+ * xmlParseCatalogPI:
+ * @ctxt: an XML parser context
+ * @catalog: the PI value string
+ *
+ * parse an XML Catalog Processing Instruction.
+ *
+ * <?oasis-xml-catalog catalog="http://example.com/catalog.xml"?>
+ *
+ * Occurs only if allowed by the user and if happening in the Misc
+ * part of the document before any doctype informations
+ * This will add the given catalog to the parsing context in order
+ * to be used if there is a resolution need further down in the document
+ */
+
+static void
+xmlParseCatalogPI(xmlParserCtxtPtr ctxt, const xmlChar *catalog) {
+ xmlChar *URL = NULL;
+ const xmlChar *tmp, *base;
+ xmlChar marker;
+
+ tmp = catalog;
+ while (IS_BLANK_CH(*tmp)) tmp++;
+ if (xmlStrncmp(tmp, BAD_CAST"catalog", 7))
+ goto error;
+ tmp += 7;
+ while (IS_BLANK_CH(*tmp)) tmp++;
+ if (*tmp != '=') {
+ return;
+ }
+ tmp++;
+ while (IS_BLANK_CH(*tmp)) tmp++;
+ marker = *tmp;
+ if ((marker != '\'') && (marker != '"'))
+ goto error;
+ tmp++;
+ base = tmp;
+ while ((*tmp != 0) && (*tmp != marker)) tmp++;
+ if (*tmp == 0)
+ goto error;
+ URL = xmlStrndup(base, tmp - base);
+ tmp++;
+ while (IS_BLANK_CH(*tmp)) tmp++;
+ if (*tmp != 0)
+ goto error;
+
+ if (URL != NULL) {
+ ctxt->catalogs = xmlCatalogAddLocal(ctxt->catalogs, URL);
+ xmlFree(URL);
+ }
+ return;
+
+error:
+ xmlWarningMsg(ctxt, XML_WAR_CATALOG_PI,
+ "Catalog PI syntax error: %s\n",
+ catalog, NULL);
+ if (URL != NULL)
+ xmlFree(URL);
+}
+#endif
+
+/**
+ * xmlParsePI:
+ * @ctxt: an XML parser context
+ *
+ * parse an XML Processing Instruction.
+ *
+ * [16] PI ::= '<?' PITarget (S (Char* - (Char* '?>' Char*)))? '?>'
+ *
+ * The processing is transfered to SAX once parsed.
+ */
+
+void
+xmlParsePI(xmlParserCtxtPtr ctxt) {
+ xmlChar *buf = NULL;
+ size_t len = 0;
+ size_t size = XML_PARSER_BUFFER_SIZE;
+ int cur, l;
+ const xmlChar *target;
+ xmlParserInputState state;
+ int count = 0;
+
+ if ((RAW == '<') && (NXT(1) == '?')) {
+ xmlParserInputPtr input = ctxt->input;
+ state = ctxt->instate;
+ ctxt->instate = XML_PARSER_PI;
+ /*
+ * this is a Processing Instruction.
+ */
+ SKIP(2);
+ SHRINK;
+
+ /*
+ * Parse the target name and check for special support like
+ * namespace.
+ */
+ target = xmlParsePITarget(ctxt);
+ if (target != NULL) {
+ if ((RAW == '?') && (NXT(1) == '>')) {
+ if (input != ctxt->input) {
+ xmlFatalErrMsg(ctxt, XML_ERR_ENTITY_BOUNDARY,
+ "PI declaration doesn't start and stop in the same entity\n");
+ }
+ SKIP(2);
+
+ /*
+ * SAX: PI detected.
+ */
+ if ((ctxt->sax) && (!ctxt->disableSAX) &&
+ (ctxt->sax->processingInstruction != NULL))
+ ctxt->sax->processingInstruction(ctxt->userData,
+ target, NULL);
+ if (ctxt->instate != XML_PARSER_EOF)
+ ctxt->instate = state;
+ return;
+ }
+ buf = (xmlChar *) xmlMallocAtomic(size * sizeof(xmlChar));
+ if (buf == NULL) {
+ xmlErrMemory(ctxt, NULL);
+ ctxt->instate = state;
+ return;
+ }
+ cur = CUR;
+ if (!IS_BLANK(cur)) {
+ xmlFatalErrMsgStr(ctxt, XML_ERR_SPACE_REQUIRED,
+ "ParsePI: PI %s space expected\n", target);
+ }
+ SKIP_BLANKS;
+ cur = CUR_CHAR(l);
+ while (IS_CHAR(cur) && /* checked */
+ ((cur != '?') || (NXT(1) != '>'))) {
+ if (len + 5 >= size) {
+ xmlChar *tmp;
+ size_t new_size = size * 2;
+ tmp = (xmlChar *) xmlRealloc(buf, new_size);
+ if (tmp == NULL) {
+ xmlErrMemory(ctxt, NULL);
+ xmlFree(buf);
+ ctxt->instate = state;
+ return;
+ }
+ buf = tmp;
+ size = new_size;
+ }
+ count++;
+ if (count > 50) {
+ GROW;
+ if (ctxt->instate == XML_PARSER_EOF) {
+ xmlFree(buf);
+ return;
+ }
+ count = 0;
+ if ((len > XML_MAX_TEXT_LENGTH) &&
+ ((ctxt->options & XML_PARSE_HUGE) == 0)) {
+ xmlFatalErrMsgStr(ctxt, XML_ERR_PI_NOT_FINISHED,
+ "PI %s too big found", target);
+ xmlFree(buf);
+ ctxt->instate = state;
+ return;
+ }
+ }
+ COPY_BUF(l,buf,len,cur);
+ NEXTL(l);
+ cur = CUR_CHAR(l);
+ if (cur == 0) {
+ SHRINK;
+ GROW;
+ cur = CUR_CHAR(l);
+ }
+ }
+ if ((len > XML_MAX_TEXT_LENGTH) &&
+ ((ctxt->options & XML_PARSE_HUGE) == 0)) {
+ xmlFatalErrMsgStr(ctxt, XML_ERR_PI_NOT_FINISHED,
+ "PI %s too big found", target);
+ xmlFree(buf);
+ ctxt->instate = state;
+ return;
+ }
+ buf[len] = 0;
+ if (cur != '?') {
+ xmlFatalErrMsgStr(ctxt, XML_ERR_PI_NOT_FINISHED,
+ "ParsePI: PI %s never end ...\n", target);
+ } else {
+ if (input != ctxt->input) {
+ xmlFatalErrMsg(ctxt, XML_ERR_SPACE_REQUIRED,
+ "PI declaration doesn't start and stop in the same entity\n");
+ }
+ SKIP(2);
+
+#ifdef LIBXML_CATALOG_ENABLED
+ if (((state == XML_PARSER_MISC) ||
+ (state == XML_PARSER_START)) &&
+ (xmlStrEqual(target, XML_CATALOG_PI))) {
+ xmlCatalogAllow allow = xmlCatalogGetDefaults();
+ if ((allow == XML_CATA_ALLOW_DOCUMENT) ||
+ (allow == XML_CATA_ALLOW_ALL))
+ xmlParseCatalogPI(ctxt, buf);
+ }
+#endif
+
+
+ /*
+ * SAX: PI detected.
+ */
+ if ((ctxt->sax) && (!ctxt->disableSAX) &&
+ (ctxt->sax->processingInstruction != NULL))
+ ctxt->sax->processingInstruction(ctxt->userData,
+ target, buf);
+ }
+ xmlFree(buf);
+ } else {
+ xmlFatalErr(ctxt, XML_ERR_PI_NOT_STARTED, NULL);
+ }
+ if (ctxt->instate != XML_PARSER_EOF)
+ ctxt->instate = state;
+ }
+}
+
+/**
+ * xmlParseNotationDecl:
+ * @ctxt: an XML parser context
+ *
+ * parse a notation declaration
+ *
+ * [82] NotationDecl ::= '<!NOTATION' S Name S (ExternalID | PublicID) S? '>'
+ *
+ * Hence there is actually 3 choices:
+ * 'PUBLIC' S PubidLiteral
+ * 'PUBLIC' S PubidLiteral S SystemLiteral
+ * and 'SYSTEM' S SystemLiteral
+ *
+ * See the NOTE on xmlParseExternalID().
+ */
+
+void
+xmlParseNotationDecl(xmlParserCtxtPtr ctxt) {
+ const xmlChar *name;
+ xmlChar *Pubid;
+ xmlChar *Systemid;
+
+ if (CMP10(CUR_PTR, '<', '!', 'N', 'O', 'T', 'A', 'T', 'I', 'O', 'N')) {
+ xmlParserInputPtr input = ctxt->input;
+ SHRINK;
+ SKIP(10);
+ if (!IS_BLANK_CH(CUR)) {
+ xmlFatalErrMsg(ctxt, XML_ERR_SPACE_REQUIRED,
+ "Space required after '<!NOTATION'\n");
+ return;
+ }
+ SKIP_BLANKS;
+
+ name = xmlParseName(ctxt);
+ if (name == NULL) {
+ xmlFatalErr(ctxt, XML_ERR_NOTATION_NOT_STARTED, NULL);
+ return;
+ }
+ if (!IS_BLANK_CH(CUR)) {
+ xmlFatalErrMsg(ctxt, XML_ERR_SPACE_REQUIRED,
+ "Space required after the NOTATION name'\n");
+ return;
+ }
+ if (xmlStrchr(name, ':') != NULL) {
+ xmlNsErr(ctxt, XML_NS_ERR_COLON,
+ "colons are forbidden from notation names '%s'\n",
+ name, NULL, NULL);
+ }
+ SKIP_BLANKS;
+
+ /*
+ * Parse the IDs.
+ */
+ Systemid = xmlParseExternalID(ctxt, &Pubid, 0);
+ SKIP_BLANKS;
+
+ if (RAW == '>') {
+ if (input != ctxt->input) {
+ xmlFatalErrMsg(ctxt, XML_ERR_SPACE_REQUIRED,
+ "Notation declaration doesn't start and stop in the same entity\n");
+ }
+ NEXT;
+ if ((ctxt->sax != NULL) && (!ctxt->disableSAX) &&
+ (ctxt->sax->notationDecl != NULL))
+ ctxt->sax->notationDecl(ctxt->userData, name, Pubid, Systemid);
+ } else {
+ xmlFatalErr(ctxt, XML_ERR_NOTATION_NOT_FINISHED, NULL);
+ }
+ if (Systemid != NULL) xmlFree(Systemid);
+ if (Pubid != NULL) xmlFree(Pubid);
+ }
+}
+
+/**
+ * xmlParseEntityDecl:
+ * @ctxt: an XML parser context
+ *
+ * parse <!ENTITY declarations
+ *
+ * [70] EntityDecl ::= GEDecl | PEDecl
+ *
+ * [71] GEDecl ::= '<!ENTITY' S Name S EntityDef S? '>'
+ *
+ * [72] PEDecl ::= '<!ENTITY' S '%' S Name S PEDef S? '>'
+ *
+ * [73] EntityDef ::= EntityValue | (ExternalID NDataDecl?)
+ *
+ * [74] PEDef ::= EntityValue | ExternalID
+ *
+ * [76] NDataDecl ::= S 'NDATA' S Name
+ *
+ * [ VC: Notation Declared ]
+ * The Name must match the declared name of a notation.
+ */
+
+void
+xmlParseEntityDecl(xmlParserCtxtPtr ctxt) {
+ const xmlChar *name = NULL;
+ xmlChar *value = NULL;
+ xmlChar *URI = NULL, *literal = NULL;
+ const xmlChar *ndata = NULL;
+ int isParameter = 0;
+ xmlChar *orig = NULL;
+ int skipped;
+
+ /* GROW; done in the caller */
+ if (CMP8(CUR_PTR, '<', '!', 'E', 'N', 'T', 'I', 'T', 'Y')) {
+ xmlParserInputPtr input = ctxt->input;
+ SHRINK;
+ SKIP(8);
+ skipped = SKIP_BLANKS;
+ if (skipped == 0) {
+ xmlFatalErrMsg(ctxt, XML_ERR_SPACE_REQUIRED,
+ "Space required after '<!ENTITY'\n");
+ }
+
+ if (RAW == '%') {
+ NEXT;
+ skipped = SKIP_BLANKS;
+ if (skipped == 0) {
+ xmlFatalErrMsg(ctxt, XML_ERR_SPACE_REQUIRED,
+ "Space required after '%'\n");
+ }
+ isParameter = 1;
+ }
+
+ name = xmlParseName(ctxt);
+ if (name == NULL) {
+ xmlFatalErrMsg(ctxt, XML_ERR_NAME_REQUIRED,
+ "xmlParseEntityDecl: no name\n");
+ return;
+ }
+ if (xmlStrchr(name, ':') != NULL) {
+ xmlNsErr(ctxt, XML_NS_ERR_COLON,
+ "colons are forbidden from entities names '%s'\n",
+ name, NULL, NULL);
+ }
+ skipped = SKIP_BLANKS;
+ if (skipped == 0) {
+ xmlFatalErrMsg(ctxt, XML_ERR_SPACE_REQUIRED,
+ "Space required after the entity name\n");
+ }
+
+ ctxt->instate = XML_PARSER_ENTITY_DECL;
+ /*
+ * handle the various case of definitions...
+ */
+ if (isParameter) {
+ if ((RAW == '"') || (RAW == '\'')) {
+ value = xmlParseEntityValue(ctxt, &orig);
+ if (value) {
+ if ((ctxt->sax != NULL) &&
+ (!ctxt->disableSAX) && (ctxt->sax->entityDecl != NULL))
+ ctxt->sax->entityDecl(ctxt->userData, name,
+ XML_INTERNAL_PARAMETER_ENTITY,
+ NULL, NULL, value);
+ }
+ } else {
+ URI = xmlParseExternalID(ctxt, &literal, 1);
+ if ((URI == NULL) && (literal == NULL)) {
+ xmlFatalErr(ctxt, XML_ERR_VALUE_REQUIRED, NULL);
+ }
+ if (URI) {
+ xmlURIPtr uri;
+
+ uri = xmlParseURI((const char *) URI);
+ if (uri == NULL) {
+ xmlErrMsgStr(ctxt, XML_ERR_INVALID_URI,
+ "Invalid URI: %s\n", URI);
+ /*
+ * This really ought to be a well formedness error
+ * but the XML Core WG decided otherwise c.f. issue
+ * E26 of the XML erratas.
+ */
+ } else {
+ if (uri->fragment != NULL) {
+ /*
+ * Okay this is foolish to block those but not
+ * invalid URIs.
+ */
+ xmlFatalErr(ctxt, XML_ERR_URI_FRAGMENT, NULL);
+ } else {
+ if ((ctxt->sax != NULL) &&
+ (!ctxt->disableSAX) &&
+ (ctxt->sax->entityDecl != NULL))
+ ctxt->sax->entityDecl(ctxt->userData, name,
+ XML_EXTERNAL_PARAMETER_ENTITY,
+ literal, URI, NULL);
+ }
+ xmlFreeURI(uri);
+ }
+ }
+ }
+ } else {
+ if ((RAW == '"') || (RAW == '\'')) {
+ value = xmlParseEntityValue(ctxt, &orig);
+ if ((ctxt->sax != NULL) &&
+ (!ctxt->disableSAX) && (ctxt->sax->entityDecl != NULL))
+ ctxt->sax->entityDecl(ctxt->userData, name,
+ XML_INTERNAL_GENERAL_ENTITY,
+ NULL, NULL, value);
+ /*
+ * For expat compatibility in SAX mode.
+ */
+ if ((ctxt->myDoc == NULL) ||
+ (xmlStrEqual(ctxt->myDoc->version, SAX_COMPAT_MODE))) {
+ if (ctxt->myDoc == NULL) {
+ ctxt->myDoc = xmlNewDoc(SAX_COMPAT_MODE);
+ if (ctxt->myDoc == NULL) {
+ xmlErrMemory(ctxt, "New Doc failed");
+ return;
+ }
+ ctxt->myDoc->properties = XML_DOC_INTERNAL;
+ }
+ if (ctxt->myDoc->intSubset == NULL)
+ ctxt->myDoc->intSubset = xmlNewDtd(ctxt->myDoc,
+ BAD_CAST "fake", NULL, NULL);
+
+ xmlSAX2EntityDecl(ctxt, name, XML_INTERNAL_GENERAL_ENTITY,
+ NULL, NULL, value);
+ }
+ } else {
+ URI = xmlParseExternalID(ctxt, &literal, 1);
+ if ((URI == NULL) && (literal == NULL)) {
+ xmlFatalErr(ctxt, XML_ERR_VALUE_REQUIRED, NULL);
+ }
+ if (URI) {
+ xmlURIPtr uri;
+
+ uri = xmlParseURI((const char *)URI);
+ if (uri == NULL) {
+ xmlErrMsgStr(ctxt, XML_ERR_INVALID_URI,
+ "Invalid URI: %s\n", URI);
+ /*
+ * This really ought to be a well formedness error
+ * but the XML Core WG decided otherwise c.f. issue
+ * E26 of the XML erratas.
+ */
+ } else {
+ if (uri->fragment != NULL) {
+ /*
+ * Okay this is foolish to block those but not
+ * invalid URIs.
+ */
+ xmlFatalErr(ctxt, XML_ERR_URI_FRAGMENT, NULL);
+ }
+ xmlFreeURI(uri);
+ }
+ }
+ if ((RAW != '>') && (!IS_BLANK_CH(CUR))) {
+ xmlFatalErrMsg(ctxt, XML_ERR_SPACE_REQUIRED,
+ "Space required before 'NDATA'\n");
+ }
+ SKIP_BLANKS;
+ if (CMP5(CUR_PTR, 'N', 'D', 'A', 'T', 'A')) {
+ SKIP(5);
+ if (!IS_BLANK_CH(CUR)) {
+ xmlFatalErrMsg(ctxt, XML_ERR_SPACE_REQUIRED,
+ "Space required after 'NDATA'\n");
+ }
+ SKIP_BLANKS;
+ ndata = xmlParseName(ctxt);
+ if ((ctxt->sax != NULL) && (!ctxt->disableSAX) &&
+ (ctxt->sax->unparsedEntityDecl != NULL))
+ ctxt->sax->unparsedEntityDecl(ctxt->userData, name,
+ literal, URI, ndata);
+ } else {
+ if ((ctxt->sax != NULL) &&
+ (!ctxt->disableSAX) && (ctxt->sax->entityDecl != NULL))
+ ctxt->sax->entityDecl(ctxt->userData, name,
+ XML_EXTERNAL_GENERAL_PARSED_ENTITY,
+ literal, URI, NULL);
+ /*
+ * For expat compatibility in SAX mode.
+ * assuming the entity repalcement was asked for
+ */
+ if ((ctxt->replaceEntities != 0) &&
+ ((ctxt->myDoc == NULL) ||
+ (xmlStrEqual(ctxt->myDoc->version, SAX_COMPAT_MODE)))) {
+ if (ctxt->myDoc == NULL) {
+ ctxt->myDoc = xmlNewDoc(SAX_COMPAT_MODE);
+ if (ctxt->myDoc == NULL) {
+ xmlErrMemory(ctxt, "New Doc failed");
+ return;
+ }
+ ctxt->myDoc->properties = XML_DOC_INTERNAL;
+ }
+
+ if (ctxt->myDoc->intSubset == NULL)
+ ctxt->myDoc->intSubset = xmlNewDtd(ctxt->myDoc,
+ BAD_CAST "fake", NULL, NULL);
+ xmlSAX2EntityDecl(ctxt, name,
+ XML_EXTERNAL_GENERAL_PARSED_ENTITY,
+ literal, URI, NULL);
+ }
+ }
+ }
+ }
+ if (ctxt->instate == XML_PARSER_EOF)
+ return;
+ SKIP_BLANKS;
+ if (RAW != '>') {
+ xmlFatalErrMsgStr(ctxt, XML_ERR_ENTITY_NOT_FINISHED,
+ "xmlParseEntityDecl: entity %s not terminated\n", name);
+ xmlHaltParser(ctxt);
+ } else {
+ if (input != ctxt->input) {
+ xmlFatalErrMsg(ctxt, XML_ERR_ENTITY_BOUNDARY,
+ "Entity declaration doesn't start and stop in the same entity\n");
+ }
+ NEXT;
+ }
+ if (orig != NULL) {
+ /*
+ * Ugly mechanism to save the raw entity value.
+ */
+ xmlEntityPtr cur = NULL;
+
+ if (isParameter) {
+ if ((ctxt->sax != NULL) &&
+ (ctxt->sax->getParameterEntity != NULL))
+ cur = ctxt->sax->getParameterEntity(ctxt->userData, name);
+ } else {
+ if ((ctxt->sax != NULL) &&
+ (ctxt->sax->getEntity != NULL))
+ cur = ctxt->sax->getEntity(ctxt->userData, name);
+ if ((cur == NULL) && (ctxt->userData==ctxt)) {
+ cur = xmlSAX2GetEntity(ctxt, name);
+ }
+ }
+ if (cur != NULL) {
+ if (cur->orig != NULL)
+ xmlFree(orig);
+ else
+ cur->orig = orig;
+ } else
+ xmlFree(orig);
+ }
+ if (value != NULL) xmlFree(value);
+ if (URI != NULL) xmlFree(URI);
+ if (literal != NULL) xmlFree(literal);
+ }
+}
+
+/**
+ * xmlParseDefaultDecl:
+ * @ctxt: an XML parser context
+ * @value: Receive a possible fixed default value for the attribute
+ *
+ * Parse an attribute default declaration
+ *
+ * [60] DefaultDecl ::= '#REQUIRED' | '#IMPLIED' | (('#FIXED' S)? AttValue)
+ *
+ * [ VC: Required Attribute ]
+ * if the default declaration is the keyword #REQUIRED, then the
+ * attribute must be specified for all elements of the type in the
+ * attribute-list declaration.
+ *
+ * [ VC: Attribute Default Legal ]
+ * The declared default value must meet the lexical constraints of
+ * the declared attribute type c.f. xmlValidateAttributeDecl()
+ *
+ * [ VC: Fixed Attribute Default ]
+ * if an attribute has a default value declared with the #FIXED
+ * keyword, instances of that attribute must match the default value.
+ *
+ * [ WFC: No < in Attribute Values ]
+ * handled in xmlParseAttValue()
+ *
+ * returns: XML_ATTRIBUTE_NONE, XML_ATTRIBUTE_REQUIRED, XML_ATTRIBUTE_IMPLIED
+ * or XML_ATTRIBUTE_FIXED.
+ */
+
+int
+xmlParseDefaultDecl(xmlParserCtxtPtr ctxt, xmlChar **value) {
+ int val;
+ xmlChar *ret;
+
+ *value = NULL;
+ if (CMP9(CUR_PTR, '#', 'R', 'E', 'Q', 'U', 'I', 'R', 'E', 'D')) {
+ SKIP(9);
+ return(XML_ATTRIBUTE_REQUIRED);
+ }
+ if (CMP8(CUR_PTR, '#', 'I', 'M', 'P', 'L', 'I', 'E', 'D')) {
+ SKIP(8);
+ return(XML_ATTRIBUTE_IMPLIED);
+ }
+ val = XML_ATTRIBUTE_NONE;
+ if (CMP6(CUR_PTR, '#', 'F', 'I', 'X', 'E', 'D')) {
+ SKIP(6);
+ val = XML_ATTRIBUTE_FIXED;
+ if (!IS_BLANK_CH(CUR)) {
+ xmlFatalErrMsg(ctxt, XML_ERR_SPACE_REQUIRED,
+ "Space required after '#FIXED'\n");
+ }
+ SKIP_BLANKS;
+ }
+ ret = xmlParseAttValue(ctxt);
+ ctxt->instate = XML_PARSER_DTD;
+ if (ret == NULL) {
+ xmlFatalErrMsg(ctxt, (xmlParserErrors)ctxt->errNo,
+ "Attribute default value declaration error\n");
+ } else
+ *value = ret;
+ return(val);
+}
+
+/**
+ * xmlParseNotationType:
+ * @ctxt: an XML parser context
+ *
+ * parse an Notation attribute type.
+ *
+ * Note: the leading 'NOTATION' S part has already being parsed...
+ *
+ * [58] NotationType ::= 'NOTATION' S '(' S? Name (S? '|' S? Name)* S? ')'
+ *
+ * [ VC: Notation Attributes ]
+ * Values of this type must match one of the notation names included
+ * in the declaration; all notation names in the declaration must be declared.
+ *
+ * Returns: the notation attribute tree built while parsing
+ */
+
+xmlEnumerationPtr
+xmlParseNotationType(xmlParserCtxtPtr ctxt) {
+ const xmlChar *name;
+ xmlEnumerationPtr ret = NULL, last = NULL, cur, tmp;
+
+ if (RAW != '(') {
+ xmlFatalErr(ctxt, XML_ERR_NOTATION_NOT_STARTED, NULL);
+ return(NULL);
+ }
+ SHRINK;
+ do {
+ NEXT;
+ SKIP_BLANKS;
+ name = xmlParseName(ctxt);
+ if (name == NULL) {
+ xmlFatalErrMsg(ctxt, XML_ERR_NAME_REQUIRED,
+ "Name expected in NOTATION declaration\n");
+ xmlFreeEnumeration(ret);
+ return(NULL);
+ }
+ tmp = ret;
+ while (tmp != NULL) {
+ if (xmlStrEqual(name, tmp->name)) {
+ xmlValidityError(ctxt, XML_DTD_DUP_TOKEN,
+ "standalone: attribute notation value token %s duplicated\n",
+ name, NULL);
+ if (!xmlDictOwns(ctxt->dict, name))
+ xmlFree((xmlChar *) name);
+ break;
+ }
+ tmp = tmp->next;
+ }
+ if (tmp == NULL) {
+ cur = xmlCreateEnumeration(name);
+ if (cur == NULL) {
+ xmlFreeEnumeration(ret);
+ return(NULL);
+ }
+ if (last == NULL) ret = last = cur;
+ else {
+ last->next = cur;
+ last = cur;
+ }
+ }
+ SKIP_BLANKS;
+ } while (RAW == '|');
+ if (RAW != ')') {
+ xmlFatalErr(ctxt, XML_ERR_NOTATION_NOT_FINISHED, NULL);
+ xmlFreeEnumeration(ret);
+ return(NULL);
+ }
+ NEXT;
+ return(ret);
+}
+
+/**
+ * xmlParseEnumerationType:
+ * @ctxt: an XML parser context
+ *
+ * parse an Enumeration attribute type.
+ *
+ * [59] Enumeration ::= '(' S? Nmtoken (S? '|' S? Nmtoken)* S? ')'
+ *
+ * [ VC: Enumeration ]
+ * Values of this type must match one of the Nmtoken tokens in
+ * the declaration
+ *
+ * Returns: the enumeration attribute tree built while parsing
+ */
+
+xmlEnumerationPtr
+xmlParseEnumerationType(xmlParserCtxtPtr ctxt) {
+ xmlChar *name;
+ xmlEnumerationPtr ret = NULL, last = NULL, cur, tmp;
+
+ if (RAW != '(') {
+ xmlFatalErr(ctxt, XML_ERR_ATTLIST_NOT_STARTED, NULL);
+ return(NULL);
+ }
+ SHRINK;
+ do {
+ NEXT;
+ SKIP_BLANKS;
+ name = xmlParseNmtoken(ctxt);
+ if (name == NULL) {
+ xmlFatalErr(ctxt, XML_ERR_NMTOKEN_REQUIRED, NULL);
+ return(ret);
+ }
+ tmp = ret;
+ while (tmp != NULL) {
+ if (xmlStrEqual(name, tmp->name)) {
+ xmlValidityError(ctxt, XML_DTD_DUP_TOKEN,
+ "standalone: attribute enumeration value token %s duplicated\n",
+ name, NULL);
+ if (!xmlDictOwns(ctxt->dict, name))
+ xmlFree(name);
+ break;
+ }
+ tmp = tmp->next;
+ }
+ if (tmp == NULL) {
+ cur = xmlCreateEnumeration(name);
+ if (!xmlDictOwns(ctxt->dict, name))
+ xmlFree(name);
+ if (cur == NULL) {
+ xmlFreeEnumeration(ret);
+ return(NULL);
+ }
+ if (last == NULL) ret = last = cur;
+ else {
+ last->next = cur;
+ last = cur;
+ }
+ }
+ SKIP_BLANKS;
+ } while (RAW == '|');
+ if (RAW != ')') {
+ xmlFatalErr(ctxt, XML_ERR_ATTLIST_NOT_FINISHED, NULL);
+ return(ret);
+ }
+ NEXT;
+ return(ret);
+}
+
+/**
+ * xmlParseEnumeratedType:
+ * @ctxt: an XML parser context
+ * @tree: the enumeration tree built while parsing
+ *
+ * parse an Enumerated attribute type.
+ *
+ * [57] EnumeratedType ::= NotationType | Enumeration
+ *
+ * [58] NotationType ::= 'NOTATION' S '(' S? Name (S? '|' S? Name)* S? ')'
+ *
+ *
+ * Returns: XML_ATTRIBUTE_ENUMERATION or XML_ATTRIBUTE_NOTATION
+ */
+
+int
+xmlParseEnumeratedType(xmlParserCtxtPtr ctxt, xmlEnumerationPtr *tree) {
+ if (CMP8(CUR_PTR, 'N', 'O', 'T', 'A', 'T', 'I', 'O', 'N')) {
+ SKIP(8);
+ if (!IS_BLANK_CH(CUR)) {
+ xmlFatalErrMsg(ctxt, XML_ERR_SPACE_REQUIRED,
+ "Space required after 'NOTATION'\n");
+ return(0);
+ }
+ SKIP_BLANKS;
+ *tree = xmlParseNotationType(ctxt);
+ if (*tree == NULL) return(0);
+ return(XML_ATTRIBUTE_NOTATION);
+ }
+ *tree = xmlParseEnumerationType(ctxt);
+ if (*tree == NULL) return(0);
+ return(XML_ATTRIBUTE_ENUMERATION);
+}
+
+/**
+ * xmlParseAttributeType:
+ * @ctxt: an XML parser context
+ * @tree: the enumeration tree built while parsing
+ *
+ * parse the Attribute list def for an element
+ *
+ * [54] AttType ::= StringType | TokenizedType | EnumeratedType
+ *
+ * [55] StringType ::= 'CDATA'
+ *
+ * [56] TokenizedType ::= 'ID' | 'IDREF' | 'IDREFS' | 'ENTITY' |
+ * 'ENTITIES' | 'NMTOKEN' | 'NMTOKENS'
+ *
+ * Validity constraints for attribute values syntax are checked in
+ * xmlValidateAttributeValue()
+ *
+ * [ VC: ID ]
+ * Values of type ID must match the Name production. A name must not
+ * appear more than once in an XML document as a value of this type;
+ * i.e., ID values must uniquely identify the elements which bear them.
+ *
+ * [ VC: One ID per Element Type ]
+ * No element type may have more than one ID attribute specified.
+ *
+ * [ VC: ID Attribute Default ]
+ * An ID attribute must have a declared default of #IMPLIED or #REQUIRED.
+ *
+ * [ VC: IDREF ]
+ * Values of type IDREF must match the Name production, and values
+ * of type IDREFS must match Names; each IDREF Name must match the value
+ * of an ID attribute on some element in the XML document; i.e. IDREF
+ * values must match the value of some ID attribute.
+ *
+ * [ VC: Entity Name ]
+ * Values of type ENTITY must match the Name production, values
+ * of type ENTITIES must match Names; each Entity Name must match the
+ * name of an unparsed entity declared in the DTD.
+ *
+ * [ VC: Name Token ]
+ * Values of type NMTOKEN must match the Nmtoken production; values
+ * of type NMTOKENS must match Nmtokens.
+ *
+ * Returns the attribute type
+ */
+int
+xmlParseAttributeType(xmlParserCtxtPtr ctxt, xmlEnumerationPtr *tree) {
+ SHRINK;
+ if (CMP5(CUR_PTR, 'C', 'D', 'A', 'T', 'A')) {
+ SKIP(5);
+ return(XML_ATTRIBUTE_CDATA);
+ } else if (CMP6(CUR_PTR, 'I', 'D', 'R', 'E', 'F', 'S')) {
+ SKIP(6);
+ return(XML_ATTRIBUTE_IDREFS);
+ } else if (CMP5(CUR_PTR, 'I', 'D', 'R', 'E', 'F')) {
+ SKIP(5);
+ return(XML_ATTRIBUTE_IDREF);
+ } else if ((RAW == 'I') && (NXT(1) == 'D')) {
+ SKIP(2);
+ return(XML_ATTRIBUTE_ID);
+ } else if (CMP6(CUR_PTR, 'E', 'N', 'T', 'I', 'T', 'Y')) {
+ SKIP(6);
+ return(XML_ATTRIBUTE_ENTITY);
+ } else if (CMP8(CUR_PTR, 'E', 'N', 'T', 'I', 'T', 'I', 'E', 'S')) {
+ SKIP(8);
+ return(XML_ATTRIBUTE_ENTITIES);
+ } else if (CMP8(CUR_PTR, 'N', 'M', 'T', 'O', 'K', 'E', 'N', 'S')) {
+ SKIP(8);
+ return(XML_ATTRIBUTE_NMTOKENS);
+ } else if (CMP7(CUR_PTR, 'N', 'M', 'T', 'O', 'K', 'E', 'N')) {
+ SKIP(7);
+ return(XML_ATTRIBUTE_NMTOKEN);
+ }
+ return(xmlParseEnumeratedType(ctxt, tree));
+}
+
+/**
+ * xmlParseAttributeListDecl:
+ * @ctxt: an XML parser context
+ *
+ * : parse the Attribute list def for an element
+ *
+ * [52] AttlistDecl ::= '<!ATTLIST' S Name AttDef* S? '>'
+ *
+ * [53] AttDef ::= S Name S AttType S DefaultDecl
+ *
+ */
+void
+xmlParseAttributeListDecl(xmlParserCtxtPtr ctxt) {
+ const xmlChar *elemName;
+ const xmlChar *attrName;
+ xmlEnumerationPtr tree;
+
+ if (CMP9(CUR_PTR, '<', '!', 'A', 'T', 'T', 'L', 'I', 'S', 'T')) {
+ xmlParserInputPtr input = ctxt->input;
+
+ SKIP(9);
+ if (!IS_BLANK_CH(CUR)) {
+ xmlFatalErrMsg(ctxt, XML_ERR_SPACE_REQUIRED,
+ "Space required after '<!ATTLIST'\n");
+ }
+ SKIP_BLANKS;
+ elemName = xmlParseName(ctxt);
+ if (elemName == NULL) {
+ xmlFatalErrMsg(ctxt, XML_ERR_NAME_REQUIRED,
+ "ATTLIST: no name for Element\n");
+ return;
+ }
+ SKIP_BLANKS;
+ GROW;
+ while ((RAW != '>') && (ctxt->instate != XML_PARSER_EOF)) {
+ const xmlChar *check = CUR_PTR;
+ int type;
+ int def;
+ xmlChar *defaultValue = NULL;
+
+ GROW;
+ tree = NULL;
+ attrName = xmlParseName(ctxt);
+ if (attrName == NULL) {
+ xmlFatalErrMsg(ctxt, XML_ERR_NAME_REQUIRED,
+ "ATTLIST: no name for Attribute\n");
+ break;
+ }
+ GROW;
+ if (!IS_BLANK_CH(CUR)) {
+ xmlFatalErrMsg(ctxt, XML_ERR_SPACE_REQUIRED,
+ "Space required after the attribute name\n");
+ break;
+ }
+ SKIP_BLANKS;
+
+ type = xmlParseAttributeType(ctxt, &tree);
+ if (type <= 0) {
+ break;
+ }
+
+ GROW;
+ if (!IS_BLANK_CH(CUR)) {
+ xmlFatalErrMsg(ctxt, XML_ERR_SPACE_REQUIRED,
+ "Space required after the attribute type\n");
+ if (tree != NULL)
+ xmlFreeEnumeration(tree);
+ break;
+ }
+ SKIP_BLANKS;
+
+ def = xmlParseDefaultDecl(ctxt, &defaultValue);
+ if (def <= 0) {
+ if (defaultValue != NULL)
+ xmlFree(defaultValue);
+ if (tree != NULL)
+ xmlFreeEnumeration(tree);
+ break;
+ }
+ if ((type != XML_ATTRIBUTE_CDATA) && (defaultValue != NULL))
+ xmlAttrNormalizeSpace(defaultValue, defaultValue);
+
+ GROW;
+ if (RAW != '>') {
+ if (!IS_BLANK_CH(CUR)) {
+ xmlFatalErrMsg(ctxt, XML_ERR_SPACE_REQUIRED,
+ "Space required after the attribute default value\n");
+ if (defaultValue != NULL)
+ xmlFree(defaultValue);
+ if (tree != NULL)
+ xmlFreeEnumeration(tree);
+ break;
+ }
+ SKIP_BLANKS;
+ }
+ if (check == CUR_PTR) {
+ xmlFatalErr(ctxt, XML_ERR_INTERNAL_ERROR,
+ "in xmlParseAttributeListDecl\n");
+ if (defaultValue != NULL)
+ xmlFree(defaultValue);
+ if (tree != NULL)
+ xmlFreeEnumeration(tree);
+ break;
+ }
+ if ((ctxt->sax != NULL) && (!ctxt->disableSAX) &&
+ (ctxt->sax->attributeDecl != NULL))
+ ctxt->sax->attributeDecl(ctxt->userData, elemName, attrName,
+ type, def, defaultValue, tree);
+ else if (tree != NULL)
+ xmlFreeEnumeration(tree);
+
+ if ((ctxt->sax2) && (defaultValue != NULL) &&
+ (def != XML_ATTRIBUTE_IMPLIED) &&
+ (def != XML_ATTRIBUTE_REQUIRED)) {
+ xmlAddDefAttrs(ctxt, elemName, attrName, defaultValue);
+ }
+ if (ctxt->sax2) {
+ xmlAddSpecialAttr(ctxt, elemName, attrName, type);
+ }
+ if (defaultValue != NULL)
+ xmlFree(defaultValue);
+ GROW;
+ }
+ if (RAW == '>') {
+ if (input != ctxt->input) {
+ xmlValidityError(ctxt, XML_ERR_ENTITY_BOUNDARY,
+ "Attribute list declaration doesn't start and stop in the same entity\n",
+ NULL, NULL);
+ }
+ NEXT;
+ }
+ }
+}
+
+/**
+ * xmlParseElementMixedContentDecl:
+ * @ctxt: an XML parser context
+ * @inputchk: the input used for the current entity, needed for boundary checks
+ *
+ * parse the declaration for a Mixed Element content
+ * The leading '(' and spaces have been skipped in xmlParseElementContentDecl
+ *
+ * [51] Mixed ::= '(' S? '#PCDATA' (S? '|' S? Name)* S? ')*' |
+ * '(' S? '#PCDATA' S? ')'
+ *
+ * [ VC: Proper Group/PE Nesting ] applies to [51] too (see [49])
+ *
+ * [ VC: No Duplicate Types ]
+ * The same name must not appear more than once in a single
+ * mixed-content declaration.
+ *
+ * returns: the list of the xmlElementContentPtr describing the element choices
+ */
+xmlElementContentPtr
+xmlParseElementMixedContentDecl(xmlParserCtxtPtr ctxt, int inputchk) {
+ xmlElementContentPtr ret = NULL, cur = NULL, n;
+ const xmlChar *elem = NULL;
+
+ GROW;
+ if (CMP7(CUR_PTR, '#', 'P', 'C', 'D', 'A', 'T', 'A')) {
+ SKIP(7);
+ SKIP_BLANKS;
+ SHRINK;
+ if (RAW == ')') {
+ if ((ctxt->validate) && (ctxt->input->id != inputchk)) {
+ xmlValidityError(ctxt, XML_ERR_ENTITY_BOUNDARY,
+"Element content declaration doesn't start and stop in the same entity\n",
+ NULL, NULL);
+ }
+ NEXT;
+ ret = xmlNewDocElementContent(ctxt->myDoc, NULL, XML_ELEMENT_CONTENT_PCDATA);
+ if (ret == NULL)
+ return(NULL);
+ if (RAW == '*') {
+ ret->ocur = XML_ELEMENT_CONTENT_MULT;
+ NEXT;
+ }
+ return(ret);
+ }
+ if ((RAW == '(') || (RAW == '|')) {
+ ret = cur = xmlNewDocElementContent(ctxt->myDoc, NULL, XML_ELEMENT_CONTENT_PCDATA);
+ if (ret == NULL) return(NULL);
+ }
+ while ((RAW == '|') && (ctxt->instate != XML_PARSER_EOF)) {
+ NEXT;
+ if (elem == NULL) {
+ ret = xmlNewDocElementContent(ctxt->myDoc, NULL, XML_ELEMENT_CONTENT_OR);
+ if (ret == NULL) return(NULL);
+ ret->c1 = cur;
+ if (cur != NULL)
+ cur->parent = ret;
+ cur = ret;
+ } else {
+ n = xmlNewDocElementContent(ctxt->myDoc, NULL, XML_ELEMENT_CONTENT_OR);
+ if (n == NULL) return(NULL);
+ n->c1 = xmlNewDocElementContent(ctxt->myDoc, elem, XML_ELEMENT_CONTENT_ELEMENT);
+ if (n->c1 != NULL)
+ n->c1->parent = n;
+ cur->c2 = n;
+ if (n != NULL)
+ n->parent = cur;
+ cur = n;
+ }
+ SKIP_BLANKS;
+ elem = xmlParseName(ctxt);
+ if (elem == NULL) {
+ xmlFatalErrMsg(ctxt, XML_ERR_NAME_REQUIRED,
+ "xmlParseElementMixedContentDecl : Name expected\n");
+ xmlFreeDocElementContent(ctxt->myDoc, cur);
+ return(NULL);
+ }
+ SKIP_BLANKS;
+ GROW;
+ }
+ if ((RAW == ')') && (NXT(1) == '*')) {
+ if (elem != NULL) {
+ cur->c2 = xmlNewDocElementContent(ctxt->myDoc, elem,
+ XML_ELEMENT_CONTENT_ELEMENT);
+ if (cur->c2 != NULL)
+ cur->c2->parent = cur;
+ }
+ if (ret != NULL)
+ ret->ocur = XML_ELEMENT_CONTENT_MULT;
+ if ((ctxt->validate) && (ctxt->input->id != inputchk)) {
+ xmlValidityError(ctxt, XML_ERR_ENTITY_BOUNDARY,
+"Element content declaration doesn't start and stop in the same entity\n",
+ NULL, NULL);
+ }
+ SKIP(2);
+ } else {
+ xmlFreeDocElementContent(ctxt->myDoc, ret);
+ xmlFatalErr(ctxt, XML_ERR_MIXED_NOT_STARTED, NULL);
+ return(NULL);
+ }
+
+ } else {
+ xmlFatalErr(ctxt, XML_ERR_PCDATA_REQUIRED, NULL);
+ }
+ return(ret);
+}
+
+/**
+ * xmlParseElementChildrenContentDeclPriv:
+ * @ctxt: an XML parser context
+ * @inputchk: the input used for the current entity, needed for boundary checks
+ * @depth: the level of recursion
+ *
+ * parse the declaration for a Mixed Element content
+ * The leading '(' and spaces have been skipped in xmlParseElementContentDecl
+ *
+ *
+ * [47] children ::= (choice | seq) ('?' | '*' | '+')?
+ *
+ * [48] cp ::= (Name | choice | seq) ('?' | '*' | '+')?
+ *
+ * [49] choice ::= '(' S? cp ( S? '|' S? cp )* S? ')'
+ *
+ * [50] seq ::= '(' S? cp ( S? ',' S? cp )* S? ')'
+ *
+ * [ VC: Proper Group/PE Nesting ] applies to [49] and [50]
+ * TODO Parameter-entity replacement text must be properly nested
+ * with parenthesized groups. That is to say, if either of the
+ * opening or closing parentheses in a choice, seq, or Mixed
+ * construct is contained in the replacement text for a parameter
+ * entity, both must be contained in the same replacement text. For
+ * interoperability, if a parameter-entity reference appears in a
+ * choice, seq, or Mixed construct, its replacement text should not
+ * be empty, and neither the first nor last non-blank character of
+ * the replacement text should be a connector (| or ,).
+ *
+ * Returns the tree of xmlElementContentPtr describing the element
+ * hierarchy.
+ */
+static xmlElementContentPtr
+xmlParseElementChildrenContentDeclPriv(xmlParserCtxtPtr ctxt, int inputchk,
+ int depth) {
+ xmlElementContentPtr ret = NULL, cur = NULL, last = NULL, op = NULL;
+ const xmlChar *elem;
+ xmlChar type = 0;
+
+ if (((depth > 128) && ((ctxt->options & XML_PARSE_HUGE) == 0)) ||
+ (depth > 2048)) {
+ xmlFatalErrMsgInt(ctxt, XML_ERR_ELEMCONTENT_NOT_FINISHED,
+"xmlParseElementChildrenContentDecl : depth %d too deep, use XML_PARSE_HUGE\n",
+ depth);
+ return(NULL);
+ }
+ SKIP_BLANKS;
+ GROW;
+ if (RAW == '(') {
+ int inputid = ctxt->input->id;
+
+ /* Recurse on first child */
+ NEXT;
+ SKIP_BLANKS;
+ cur = ret = xmlParseElementChildrenContentDeclPriv(ctxt, inputid,
+ depth + 1);
+ SKIP_BLANKS;
+ GROW;
+ } else {
+ elem = xmlParseName(ctxt);
+ if (elem == NULL) {
+ xmlFatalErr(ctxt, XML_ERR_ELEMCONTENT_NOT_STARTED, NULL);
+ return(NULL);
+ }
+ cur = ret = xmlNewDocElementContent(ctxt->myDoc, elem, XML_ELEMENT_CONTENT_ELEMENT);
+ if (cur == NULL) {
+ xmlErrMemory(ctxt, NULL);
+ return(NULL);
+ }
+ GROW;
+ if (RAW == '?') {
+ cur->ocur = XML_ELEMENT_CONTENT_OPT;
+ NEXT;
+ } else if (RAW == '*') {
+ cur->ocur = XML_ELEMENT_CONTENT_MULT;
+ NEXT;
+ } else if (RAW == '+') {
+ cur->ocur = XML_ELEMENT_CONTENT_PLUS;
+ NEXT;
+ } else {
+ cur->ocur = XML_ELEMENT_CONTENT_ONCE;
+ }
+ GROW;
+ }
+ SKIP_BLANKS;
+ SHRINK;
+ while ((RAW != ')') && (ctxt->instate != XML_PARSER_EOF)) {
+ /*
+ * Each loop we parse one separator and one element.
+ */
+ if (RAW == ',') {
+ if (type == 0) type = CUR;
+
+ /*
+ * Detect "Name | Name , Name" error
+ */
+ else if (type != CUR) {
+ xmlFatalErrMsgInt(ctxt, XML_ERR_SEPARATOR_REQUIRED,
+ "xmlParseElementChildrenContentDecl : '%c' expected\n",
+ type);
+ if ((last != NULL) && (last != ret))
+ xmlFreeDocElementContent(ctxt->myDoc, last);
+ if (ret != NULL)
+ xmlFreeDocElementContent(ctxt->myDoc, ret);
+ return(NULL);
+ }
+ NEXT;
+
+ op = xmlNewDocElementContent(ctxt->myDoc, NULL, XML_ELEMENT_CONTENT_SEQ);
+ if (op == NULL) {
+ if ((last != NULL) && (last != ret))
+ xmlFreeDocElementContent(ctxt->myDoc, last);
+ xmlFreeDocElementContent(ctxt->myDoc, ret);
+ return(NULL);
+ }
+ if (last == NULL) {
+ op->c1 = ret;
+ if (ret != NULL)
+ ret->parent = op;
+ ret = cur = op;
+ } else {
+ cur->c2 = op;
+ if (op != NULL)
+ op->parent = cur;
+ op->c1 = last;
+ if (last != NULL)
+ last->parent = op;
+ cur =op;
+ last = NULL;
+ }
+ } else if (RAW == '|') {
+ if (type == 0) type = CUR;
+
+ /*
+ * Detect "Name , Name | Name" error
+ */
+ else if (type != CUR) {
+ xmlFatalErrMsgInt(ctxt, XML_ERR_SEPARATOR_REQUIRED,
+ "xmlParseElementChildrenContentDecl : '%c' expected\n",
+ type);
+ if ((last != NULL) && (last != ret))
+ xmlFreeDocElementContent(ctxt->myDoc, last);
+ if (ret != NULL)
+ xmlFreeDocElementContent(ctxt->myDoc, ret);
+ return(NULL);
+ }
+ NEXT;
+
+ op = xmlNewDocElementContent(ctxt->myDoc, NULL, XML_ELEMENT_CONTENT_OR);
+ if (op == NULL) {
+ if ((last != NULL) && (last != ret))
+ xmlFreeDocElementContent(ctxt->myDoc, last);
+ if (ret != NULL)
+ xmlFreeDocElementContent(ctxt->myDoc, ret);
+ return(NULL);
+ }
+ if (last == NULL) {
+ op->c1 = ret;
+ if (ret != NULL)
+ ret->parent = op;
+ ret = cur = op;
+ } else {
+ cur->c2 = op;
+ if (op != NULL)
+ op->parent = cur;
+ op->c1 = last;
+ if (last != NULL)
+ last->parent = op;
+ cur =op;
+ last = NULL;
+ }
+ } else {
+ xmlFatalErr(ctxt, XML_ERR_ELEMCONTENT_NOT_FINISHED, NULL);
+ if ((last != NULL) && (last != ret))
+ xmlFreeDocElementContent(ctxt->myDoc, last);
+ if (ret != NULL)
+ xmlFreeDocElementContent(ctxt->myDoc, ret);
+ return(NULL);
+ }
+ GROW;
+ SKIP_BLANKS;
+ GROW;
+ if (RAW == '(') {
+ int inputid = ctxt->input->id;
+ /* Recurse on second child */
+ NEXT;
+ SKIP_BLANKS;
+ last = xmlParseElementChildrenContentDeclPriv(ctxt, inputid,
+ depth + 1);
+ SKIP_BLANKS;
+ } else {
+ elem = xmlParseName(ctxt);
+ if (elem == NULL) {
+ xmlFatalErr(ctxt, XML_ERR_ELEMCONTENT_NOT_STARTED, NULL);
+ if (ret != NULL)
+ xmlFreeDocElementContent(ctxt->myDoc, ret);
+ return(NULL);
+ }
+ last = xmlNewDocElementContent(ctxt->myDoc, elem, XML_ELEMENT_CONTENT_ELEMENT);
+ if (last == NULL) {
+ if (ret != NULL)
+ xmlFreeDocElementContent(ctxt->myDoc, ret);
+ return(NULL);
+ }
+ if (RAW == '?') {
+ last->ocur = XML_ELEMENT_CONTENT_OPT;
+ NEXT;
+ } else if (RAW == '*') {
+ last->ocur = XML_ELEMENT_CONTENT_MULT;
+ NEXT;
+ } else if (RAW == '+') {
+ last->ocur = XML_ELEMENT_CONTENT_PLUS;
+ NEXT;
+ } else {
+ last->ocur = XML_ELEMENT_CONTENT_ONCE;
+ }
+ }
+ SKIP_BLANKS;
+ GROW;
+ }
+ if ((cur != NULL) && (last != NULL)) {
+ cur->c2 = last;
+ if (last != NULL)
+ last->parent = cur;
+ }
+ if ((ctxt->validate) && (ctxt->input->id != inputchk)) {
+ xmlValidityError(ctxt, XML_ERR_ENTITY_BOUNDARY,
+"Element content declaration doesn't start and stop in the same entity\n",
+ NULL, NULL);
+ }
+ NEXT;
+ if (RAW == '?') {
+ if (ret != NULL) {
+ if ((ret->ocur == XML_ELEMENT_CONTENT_PLUS) ||
+ (ret->ocur == XML_ELEMENT_CONTENT_MULT))
+ ret->ocur = XML_ELEMENT_CONTENT_MULT;
+ else
+ ret->ocur = XML_ELEMENT_CONTENT_OPT;
+ }
+ NEXT;
+ } else if (RAW == '*') {
+ if (ret != NULL) {
+ ret->ocur = XML_ELEMENT_CONTENT_MULT;
+ cur = ret;
+ /*
+ * Some normalization:
+ * (a | b* | c?)* == (a | b | c)*
+ */
+ while ((cur != NULL) && (cur->type == XML_ELEMENT_CONTENT_OR)) {
+ if ((cur->c1 != NULL) &&
+ ((cur->c1->ocur == XML_ELEMENT_CONTENT_OPT) ||
+ (cur->c1->ocur == XML_ELEMENT_CONTENT_MULT)))
+ cur->c1->ocur = XML_ELEMENT_CONTENT_ONCE;
+ if ((cur->c2 != NULL) &&
+ ((cur->c2->ocur == XML_ELEMENT_CONTENT_OPT) ||
+ (cur->c2->ocur == XML_ELEMENT_CONTENT_MULT)))
+ cur->c2->ocur = XML_ELEMENT_CONTENT_ONCE;
+ cur = cur->c2;
+ }
+ }
+ NEXT;
+ } else if (RAW == '+') {
+ if (ret != NULL) {
+ int found = 0;
+
+ if ((ret->ocur == XML_ELEMENT_CONTENT_OPT) ||
+ (ret->ocur == XML_ELEMENT_CONTENT_MULT))
+ ret->ocur = XML_ELEMENT_CONTENT_MULT;
+ else
+ ret->ocur = XML_ELEMENT_CONTENT_PLUS;
+ /*
+ * Some normalization:
+ * (a | b*)+ == (a | b)*
+ * (a | b?)+ == (a | b)*
+ */
+ while ((cur != NULL) && (cur->type == XML_ELEMENT_CONTENT_OR)) {
+ if ((cur->c1 != NULL) &&
+ ((cur->c1->ocur == XML_ELEMENT_CONTENT_OPT) ||
+ (cur->c1->ocur == XML_ELEMENT_CONTENT_MULT))) {
+ cur->c1->ocur = XML_ELEMENT_CONTENT_ONCE;
+ found = 1;
+ }
+ if ((cur->c2 != NULL) &&
+ ((cur->c2->ocur == XML_ELEMENT_CONTENT_OPT) ||
+ (cur->c2->ocur == XML_ELEMENT_CONTENT_MULT))) {
+ cur->c2->ocur = XML_ELEMENT_CONTENT_ONCE;
+ found = 1;
+ }
+ cur = cur->c2;
+ }
+ if (found)
+ ret->ocur = XML_ELEMENT_CONTENT_MULT;
+ }
+ NEXT;
+ }
+ return(ret);
+}
+
+/**
+ * xmlParseElementChildrenContentDecl:
+ * @ctxt: an XML parser context
+ * @inputchk: the input used for the current entity, needed for boundary checks
+ *
+ * parse the declaration for a Mixed Element content
+ * The leading '(' and spaces have been skipped in xmlParseElementContentDecl
+ *
+ * [47] children ::= (choice | seq) ('?' | '*' | '+')?
+ *
+ * [48] cp ::= (Name | choice | seq) ('?' | '*' | '+')?
+ *
+ * [49] choice ::= '(' S? cp ( S? '|' S? cp )* S? ')'
+ *
+ * [50] seq ::= '(' S? cp ( S? ',' S? cp )* S? ')'
+ *
+ * [ VC: Proper Group/PE Nesting ] applies to [49] and [50]
+ * TODO Parameter-entity replacement text must be properly nested
+ * with parenthesized groups. That is to say, if either of the
+ * opening or closing parentheses in a choice, seq, or Mixed
+ * construct is contained in the replacement text for a parameter
+ * entity, both must be contained in the same replacement text. For
+ * interoperability, if a parameter-entity reference appears in a
+ * choice, seq, or Mixed construct, its replacement text should not
+ * be empty, and neither the first nor last non-blank character of
+ * the replacement text should be a connector (| or ,).
+ *
+ * Returns the tree of xmlElementContentPtr describing the element
+ * hierarchy.
+ */
+xmlElementContentPtr
+xmlParseElementChildrenContentDecl(xmlParserCtxtPtr ctxt, int inputchk) {
+ /* stub left for API/ABI compat */
+ return(xmlParseElementChildrenContentDeclPriv(ctxt, inputchk, 1));
+}
+
+/**
+ * xmlParseElementContentDecl:
+ * @ctxt: an XML parser context
+ * @name: the name of the element being defined.
+ * @result: the Element Content pointer will be stored here if any
+ *
+ * parse the declaration for an Element content either Mixed or Children,
+ * the cases EMPTY and ANY are handled directly in xmlParseElementDecl
+ *
+ * [46] contentspec ::= 'EMPTY' | 'ANY' | Mixed | children
+ *
+ * returns: the type of element content XML_ELEMENT_TYPE_xxx
+ */
+
+int
+xmlParseElementContentDecl(xmlParserCtxtPtr ctxt, const xmlChar *name,
+ xmlElementContentPtr *result) {
+
+ xmlElementContentPtr tree = NULL;
+ int inputid = ctxt->input->id;
+ int res;
+
+ *result = NULL;
+
+ if (RAW != '(') {
+ xmlFatalErrMsgStr(ctxt, XML_ERR_ELEMCONTENT_NOT_STARTED,
+ "xmlParseElementContentDecl : %s '(' expected\n", name);
+ return(-1);
+ }
+ NEXT;
+ GROW;
+ if (ctxt->instate == XML_PARSER_EOF)
+ return(-1);
+ SKIP_BLANKS;
+ if (CMP7(CUR_PTR, '#', 'P', 'C', 'D', 'A', 'T', 'A')) {
+ tree = xmlParseElementMixedContentDecl(ctxt, inputid);
+ res = XML_ELEMENT_TYPE_MIXED;
+ } else {
+ tree = xmlParseElementChildrenContentDeclPriv(ctxt, inputid, 1);
+ res = XML_ELEMENT_TYPE_ELEMENT;
+ }
+ SKIP_BLANKS;
+ *result = tree;
+ return(res);
+}
+
+/**
+ * xmlParseElementDecl:
+ * @ctxt: an XML parser context
+ *
+ * parse an Element declaration.
+ *
+ * [45] elementdecl ::= '<!ELEMENT' S Name S contentspec S? '>'
+ *
+ * [ VC: Unique Element Type Declaration ]
+ * No element type may be declared more than once
+ *
+ * Returns the type of the element, or -1 in case of error
+ */
+int
+xmlParseElementDecl(xmlParserCtxtPtr ctxt) {
+ const xmlChar *name;
+ int ret = -1;
+ xmlElementContentPtr content = NULL;
+
+ /* GROW; done in the caller */
+ if (CMP9(CUR_PTR, '<', '!', 'E', 'L', 'E', 'M', 'E', 'N', 'T')) {
+ xmlParserInputPtr input = ctxt->input;
+
+ SKIP(9);
+ if (!IS_BLANK_CH(CUR)) {
+ xmlFatalErrMsg(ctxt, XML_ERR_SPACE_REQUIRED,
+ "Space required after 'ELEMENT'\n");
+ }
+ SKIP_BLANKS;
+ name = xmlParseName(ctxt);
+ if (name == NULL) {
+ xmlFatalErrMsg(ctxt, XML_ERR_NAME_REQUIRED,
+ "xmlParseElementDecl: no name for Element\n");
+ return(-1);
+ }
+ while ((RAW == 0) && (ctxt->inputNr > 1))
+ xmlPopInput(ctxt);
+ if (!IS_BLANK_CH(CUR)) {
+ xmlFatalErrMsg(ctxt, XML_ERR_SPACE_REQUIRED,
+ "Space required after the element name\n");
+ }
+ SKIP_BLANKS;
+ if (CMP5(CUR_PTR, 'E', 'M', 'P', 'T', 'Y')) {
+ SKIP(5);
+ /*
+ * Element must always be empty.
+ */
+ ret = XML_ELEMENT_TYPE_EMPTY;
+ } else if ((RAW == 'A') && (NXT(1) == 'N') &&
+ (NXT(2) == 'Y')) {
+ SKIP(3);
+ /*
+ * Element is a generic container.
+ */
+ ret = XML_ELEMENT_TYPE_ANY;
+ } else if (RAW == '(') {
+ ret = xmlParseElementContentDecl(ctxt, name, &content);
+ } else {
+ /*
+ * [ WFC: PEs in Internal Subset ] error handling.
+ */
+ if ((RAW == '%') && (ctxt->external == 0) &&
+ (ctxt->inputNr == 1)) {
+ xmlFatalErrMsg(ctxt, XML_ERR_PEREF_IN_INT_SUBSET,
+ "PEReference: forbidden within markup decl in internal subset\n");
+ } else {
+ xmlFatalErrMsg(ctxt, XML_ERR_ELEMCONTENT_NOT_STARTED,
+ "xmlParseElementDecl: 'EMPTY', 'ANY' or '(' expected\n");
+ }
+ return(-1);
+ }
+
+ SKIP_BLANKS;
+ /*
+ * Pop-up of finished entities.
+ */
+ while ((RAW == 0) && (ctxt->inputNr > 1))
+ xmlPopInput(ctxt);
+ SKIP_BLANKS;
+
+ if (RAW != '>') {
+ xmlFatalErr(ctxt, XML_ERR_GT_REQUIRED, NULL);
+ if (content != NULL) {
+ xmlFreeDocElementContent(ctxt->myDoc, content);
+ }
+ } else {
+ if (input != ctxt->input) {
+ xmlFatalErrMsg(ctxt, XML_ERR_ENTITY_BOUNDARY,
+ "Element declaration doesn't start and stop in the same entity\n");
+ }
+
+ NEXT;
+ if ((ctxt->sax != NULL) && (!ctxt->disableSAX) &&
+ (ctxt->sax->elementDecl != NULL)) {
+ if (content != NULL)
+ content->parent = NULL;
+ ctxt->sax->elementDecl(ctxt->userData, name, ret,
+ content);
+ if ((content != NULL) && (content->parent == NULL)) {
+ /*
+ * this is a trick: if xmlAddElementDecl is called,
+ * instead of copying the full tree it is plugged directly
+ * if called from the parser. Avoid duplicating the
+ * interfaces or change the API/ABI
+ */
+ xmlFreeDocElementContent(ctxt->myDoc, content);
+ }
+ } else if (content != NULL) {
+ xmlFreeDocElementContent(ctxt->myDoc, content);
+ }
+ }
+ }
+ return(ret);
+}
+
+/**
+ * xmlParseConditionalSections
+ * @ctxt: an XML parser context
+ *
+ * [61] conditionalSect ::= includeSect | ignoreSect
+ * [62] includeSect ::= '<![' S? 'INCLUDE' S? '[' extSubsetDecl ']]>'
+ * [63] ignoreSect ::= '<![' S? 'IGNORE' S? '[' ignoreSectContents* ']]>'
+ * [64] ignoreSectContents ::= Ignore ('<![' ignoreSectContents ']]>' Ignore)*
+ * [65] Ignore ::= Char* - (Char* ('<![' | ']]>') Char*)
+ */
+
+static void
+xmlParseConditionalSections(xmlParserCtxtPtr ctxt) {
+ int id = ctxt->input->id;
+
+ SKIP(3);
+ SKIP_BLANKS;
+ if (CMP7(CUR_PTR, 'I', 'N', 'C', 'L', 'U', 'D', 'E')) {
+ SKIP(7);
+ SKIP_BLANKS;
+ if (RAW != '[') {
+ xmlFatalErr(ctxt, XML_ERR_CONDSEC_INVALID, NULL);
+ xmlHaltParser(ctxt);
+ return;
+ } else {
+ if (ctxt->input->id != id) {
+ xmlValidityError(ctxt, XML_ERR_ENTITY_BOUNDARY,
+ "All markup of the conditional section is not in the same entity\n",
+ NULL, NULL);
+ }
+ NEXT;
+ }
+ if (xmlParserDebugEntities) {
+ if ((ctxt->input != NULL) && (ctxt->input->filename))
+ xmlGenericError(xmlGenericErrorContext,
+ "%s(%d): ", ctxt->input->filename,
+ ctxt->input->line);
+ xmlGenericError(xmlGenericErrorContext,
+ "Entering INCLUDE Conditional Section\n");
+ }
+
+ while (((RAW != 0) && ((RAW != ']') || (NXT(1) != ']') ||
+ (NXT(2) != '>'))) && (ctxt->instate != XML_PARSER_EOF)) {
+ const xmlChar *check = CUR_PTR;
+ unsigned int cons = ctxt->input->consumed;
+
+ if ((RAW == '<') && (NXT(1) == '!') && (NXT(2) == '[')) {
+ xmlParseConditionalSections(ctxt);
+ } else if (IS_BLANK_CH(CUR)) {
+ NEXT;
+ } else if (RAW == '%') {
+ xmlParsePEReference(ctxt);
+ } else
+ xmlParseMarkupDecl(ctxt);
+
+ /*
+ * Pop-up of finished entities.
+ */
+ while ((RAW == 0) && (ctxt->inputNr > 1))
+ xmlPopInput(ctxt);
+
+ if ((CUR_PTR == check) && (cons == ctxt->input->consumed)) {
+ xmlFatalErr(ctxt, XML_ERR_EXT_SUBSET_NOT_FINISHED, NULL);
+ break;
+ }
+ }
+ if (xmlParserDebugEntities) {
+ if ((ctxt->input != NULL) && (ctxt->input->filename))
+ xmlGenericError(xmlGenericErrorContext,
+ "%s(%d): ", ctxt->input->filename,
+ ctxt->input->line);
+ xmlGenericError(xmlGenericErrorContext,
+ "Leaving INCLUDE Conditional Section\n");
+ }
+
+ } else if (CMP6(CUR_PTR, 'I', 'G', 'N', 'O', 'R', 'E')) {
+ int state;
+ xmlParserInputState instate;
+ int depth = 0;
+
+ SKIP(6);
+ SKIP_BLANKS;
+ if (RAW != '[') {
+ xmlFatalErr(ctxt, XML_ERR_CONDSEC_INVALID, NULL);
+ xmlHaltParser(ctxt);
+ return;
+ } else {
+ if (ctxt->input->id != id) {
+ xmlValidityError(ctxt, XML_ERR_ENTITY_BOUNDARY,
+ "All markup of the conditional section is not in the same entity\n",
+ NULL, NULL);
+ }
+ NEXT;
+ }
+ if (xmlParserDebugEntities) {
+ if ((ctxt->input != NULL) && (ctxt->input->filename))
+ xmlGenericError(xmlGenericErrorContext,
+ "%s(%d): ", ctxt->input->filename,
+ ctxt->input->line);
+ xmlGenericError(xmlGenericErrorContext,
+ "Entering IGNORE Conditional Section\n");
+ }
+
+ /*
+ * Parse up to the end of the conditional section
+ * But disable SAX event generating DTD building in the meantime
+ */
+ state = ctxt->disableSAX;
+ instate = ctxt->instate;
+ if (ctxt->recovery == 0) ctxt->disableSAX = 1;
+ ctxt->instate = XML_PARSER_IGNORE;
+
+ while (((depth >= 0) && (RAW != 0)) &&
+ (ctxt->instate != XML_PARSER_EOF)) {
+ if ((RAW == '<') && (NXT(1) == '!') && (NXT(2) == '[')) {
+ depth++;
+ SKIP(3);
+ continue;
+ }
+ if ((RAW == ']') && (NXT(1) == ']') && (NXT(2) == '>')) {
+ if (--depth >= 0) SKIP(3);
+ continue;
+ }
+ NEXT;
+ continue;
+ }
+
+ ctxt->disableSAX = state;
+ ctxt->instate = instate;
+
+ if (xmlParserDebugEntities) {
+ if ((ctxt->input != NULL) && (ctxt->input->filename))
+ xmlGenericError(xmlGenericErrorContext,
+ "%s(%d): ", ctxt->input->filename,
+ ctxt->input->line);
+ xmlGenericError(xmlGenericErrorContext,
+ "Leaving IGNORE Conditional Section\n");
+ }
+
+ } else {
+ xmlFatalErr(ctxt, XML_ERR_CONDSEC_INVALID_KEYWORD, NULL);
+ xmlHaltParser(ctxt);
+ return;
+ }
+
+ if (RAW == 0)
+ SHRINK;
+
+ if (RAW == 0) {
+ xmlFatalErr(ctxt, XML_ERR_CONDSEC_NOT_FINISHED, NULL);
+ } else {
+ if (ctxt->input->id != id) {
+ xmlValidityError(ctxt, XML_ERR_ENTITY_BOUNDARY,
+ "All markup of the conditional section is not in the same entity\n",
+ NULL, NULL);
+ }
+ if ((ctxt-> instate != XML_PARSER_EOF) &&
+ ((ctxt->input->cur + 3) <= ctxt->input->end))
+ SKIP(3);
+ }
+}
+
+/**
+ * xmlParseMarkupDecl:
+ * @ctxt: an XML parser context
+ *
+ * parse Markup declarations
+ *
+ * [29] markupdecl ::= elementdecl | AttlistDecl | EntityDecl |
+ * NotationDecl | PI | Comment
+ *
+ * [ VC: Proper Declaration/PE Nesting ]
+ * Parameter-entity replacement text must be properly nested with
+ * markup declarations. That is to say, if either the first character
+ * or the last character of a markup declaration (markupdecl above) is
+ * contained in the replacement text for a parameter-entity reference,
+ * both must be contained in the same replacement text.
+ *
+ * [ WFC: PEs in Internal Subset ]
+ * In the internal DTD subset, parameter-entity references can occur
+ * only where markup declarations can occur, not within markup declarations.
+ * (This does not apply to references that occur in external parameter
+ * entities or to the external subset.)
+ */
+void
+xmlParseMarkupDecl(xmlParserCtxtPtr ctxt) {
+ GROW;
+ if (CUR == '<') {
+ if (NXT(1) == '!') {
+ switch (NXT(2)) {
+ case 'E':
+ if (NXT(3) == 'L')
+ xmlParseElementDecl(ctxt);
+ else if (NXT(3) == 'N')
+ xmlParseEntityDecl(ctxt);
+ break;
+ case 'A':
+ xmlParseAttributeListDecl(ctxt);
+ break;
+ case 'N':
+ xmlParseNotationDecl(ctxt);
+ break;
+ case '-':
+ xmlParseComment(ctxt);
+ break;
+ default:
+ /* there is an error but it will be detected later */
+ break;
+ }
+ } else if (NXT(1) == '?') {
+ xmlParsePI(ctxt);
+ }
+ }
+
+ /*
+ * detect requirement to exit there and act accordingly
+ * and avoid having instate overriden later on
+ */
+ if (ctxt->instate == XML_PARSER_EOF)
+ return;
+
+ /*
+ * This is only for internal subset. On external entities,
+ * the replacement is done before parsing stage
+ */
+ if ((ctxt->external == 0) && (ctxt->inputNr == 1))
+ xmlParsePEReference(ctxt);
+
+ /*
+ * Conditional sections are allowed from entities included
+ * by PE References in the internal subset.
+ */
+ if ((ctxt->external == 0) && (ctxt->inputNr > 1)) {
+ if ((RAW == '<') && (NXT(1) == '!') && (NXT(2) == '[')) {
+ xmlParseConditionalSections(ctxt);
+ }
+ }
+
+ ctxt->instate = XML_PARSER_DTD;
+}
+
+/**
+ * xmlParseTextDecl:
+ * @ctxt: an XML parser context
+ *
+ * parse an XML declaration header for external entities
+ *
+ * [77] TextDecl ::= '<?xml' VersionInfo? EncodingDecl S? '?>'
+ */
+
+void
+xmlParseTextDecl(xmlParserCtxtPtr ctxt) {
+ xmlChar *version;
+ const xmlChar *encoding;
+
+ /*
+ * We know that '<?xml' is here.
+ */
+ if ((CMP5(CUR_PTR, '<', '?', 'x', 'm', 'l')) && (IS_BLANK_CH(NXT(5)))) {
+ SKIP(5);
+ } else {
+ xmlFatalErr(ctxt, XML_ERR_XMLDECL_NOT_STARTED, NULL);
+ return;
+ }
+
+ if (!IS_BLANK_CH(CUR)) {
+ xmlFatalErrMsg(ctxt, XML_ERR_SPACE_REQUIRED,
+ "Space needed after '<?xml'\n");
+ }
+ SKIP_BLANKS;
+
+ /*
+ * We may have the VersionInfo here.
+ */
+ version = xmlParseVersionInfo(ctxt);
+ if (version == NULL)
+ version = xmlCharStrdup(XML_DEFAULT_VERSION);
+ else {
+ if (!IS_BLANK_CH(CUR)) {
+ xmlFatalErrMsg(ctxt, XML_ERR_SPACE_REQUIRED,
+ "Space needed here\n");
+ }
+ }
+ ctxt->input->version = version;
+
+ /*
+ * We must have the encoding declaration
+ */
+ encoding = xmlParseEncodingDecl(ctxt);
+ if (ctxt->errNo == XML_ERR_UNSUPPORTED_ENCODING) {
+ /*
+ * The XML REC instructs us to stop parsing right here
+ */
+ return;
+ }
+ if ((encoding == NULL) && (ctxt->errNo == XML_ERR_OK)) {
+ xmlFatalErrMsg(ctxt, XML_ERR_MISSING_ENCODING,
+ "Missing encoding in text declaration\n");
+ }
+
+ SKIP_BLANKS;
+ if ((RAW == '?') && (NXT(1) == '>')) {
+ SKIP(2);
+ } else if (RAW == '>') {
+ /* Deprecated old WD ... */
+ xmlFatalErr(ctxt, XML_ERR_XMLDECL_NOT_FINISHED, NULL);
+ NEXT;
+ } else {
+ xmlFatalErr(ctxt, XML_ERR_XMLDECL_NOT_FINISHED, NULL);
+ MOVETO_ENDTAG(CUR_PTR);
+ NEXT;
+ }
+}
+
+/**
+ * xmlParseExternalSubset:
+ * @ctxt: an XML parser context
+ * @ExternalID: the external identifier
+ * @SystemID: the system identifier (or URL)
+ *
+ * parse Markup declarations from an external subset
+ *
+ * [30] extSubset ::= textDecl? extSubsetDecl
+ *
+ * [31] extSubsetDecl ::= (markupdecl | conditionalSect | PEReference | S) *
+ */
+void
+xmlParseExternalSubset(xmlParserCtxtPtr ctxt, const xmlChar *ExternalID,
+ const xmlChar *SystemID) {
+ xmlDetectSAX2(ctxt);
+ GROW;
+
+ if ((ctxt->encoding == NULL) &&
+ (ctxt->input->end - ctxt->input->cur >= 4)) {
+ xmlChar start[4];
+ xmlCharEncoding enc;
+
+ start[0] = RAW;
+ start[1] = NXT(1);
+ start[2] = NXT(2);
+ start[3] = NXT(3);
+ enc = xmlDetectCharEncoding(start, 4);
+ if (enc != XML_CHAR_ENCODING_NONE)
+ xmlSwitchEncoding(ctxt, enc);
+ }
+
+ if (CMP5(CUR_PTR, '<', '?', 'x', 'm', 'l')) {
+ xmlParseTextDecl(ctxt);
+ if (ctxt->errNo == XML_ERR_UNSUPPORTED_ENCODING) {
+ /*
+ * The XML REC instructs us to stop parsing right here
+ */
+ xmlHaltParser(ctxt);
+ return;
+ }
+ }
+ if (ctxt->myDoc == NULL) {
+ ctxt->myDoc = xmlNewDoc(BAD_CAST "1.0");
+ if (ctxt->myDoc == NULL) {
+ xmlErrMemory(ctxt, "New Doc failed");
+ return;
+ }
+ ctxt->myDoc->properties = XML_DOC_INTERNAL;
+ }
+ if ((ctxt->myDoc != NULL) && (ctxt->myDoc->intSubset == NULL))
+ xmlCreateIntSubset(ctxt->myDoc, NULL, ExternalID, SystemID);
+
+ ctxt->instate = XML_PARSER_DTD;
+ ctxt->external = 1;
+ while (((RAW == '<') && (NXT(1) == '?')) ||
+ ((RAW == '<') && (NXT(1) == '!')) ||
+ (RAW == '%') || IS_BLANK_CH(CUR)) {
+ const xmlChar *check = CUR_PTR;
+ unsigned int cons = ctxt->input->consumed;
+
+ GROW;
+ if ((RAW == '<') && (NXT(1) == '!') && (NXT(2) == '[')) {
+ xmlParseConditionalSections(ctxt);
+ } else if (IS_BLANK_CH(CUR)) {
+ NEXT;
+ } else if (RAW == '%') {
+ xmlParsePEReference(ctxt);
+ } else
+ xmlParseMarkupDecl(ctxt);
+
+ /*
+ * Pop-up of finished entities.
+ */
+ while ((RAW == 0) && (ctxt->inputNr > 1))
+ xmlPopInput(ctxt);
+
+ if ((CUR_PTR == check) && (cons == ctxt->input->consumed)) {
+ xmlFatalErr(ctxt, XML_ERR_EXT_SUBSET_NOT_FINISHED, NULL);
+ break;
+ }
+ }
+
+ if (RAW != 0) {
+ xmlFatalErr(ctxt, XML_ERR_EXT_SUBSET_NOT_FINISHED, NULL);
+ }
+
+}
+
+/**
+ * xmlParseReference:
+ * @ctxt: an XML parser context
+ *
+ * parse and handle entity references in content, depending on the SAX
+ * interface, this may end-up in a call to character() if this is a
+ * CharRef, a predefined entity, if there is no reference() callback.
+ * or if the parser was asked to switch to that mode.
+ *
+ * [67] Reference ::= EntityRef | CharRef
+ */
+void
+xmlParseReference(xmlParserCtxtPtr ctxt) {
+ xmlEntityPtr ent;
+ xmlChar *val;
+ int was_checked;
+ xmlNodePtr list = NULL;
+ xmlParserErrors ret = XML_ERR_OK;
+
+
+ if (RAW != '&')
+ return;
+
+ /*
+ * Simple case of a CharRef
+ */
+ if (NXT(1) == '#') {
+ int i = 0;
+ xmlChar out[10];
+ int hex = NXT(2);
+ int value = xmlParseCharRef(ctxt);
+
+ if (value == 0)
+ return;
+ if (ctxt->charset != XML_CHAR_ENCODING_UTF8) {
+ /*
+ * So we are using non-UTF-8 buffers
+ * Check that the char fit on 8bits, if not
+ * generate a CharRef.
+ */
+ if (value <= 0xFF) {
+ out[0] = value;
+ out[1] = 0;
+ if ((ctxt->sax != NULL) && (ctxt->sax->characters != NULL) &&
+ (!ctxt->disableSAX))
+ ctxt->sax->characters(ctxt->userData, out, 1);
+ } else {
+ if ((hex == 'x') || (hex == 'X'))
+ snprintf((char *)out, sizeof(out), "#x%X", value);
+ else
+ snprintf((char *)out, sizeof(out), "#%d", value);
+ if ((ctxt->sax != NULL) && (ctxt->sax->reference != NULL) &&
+ (!ctxt->disableSAX))
+ ctxt->sax->reference(ctxt->userData, out);
+ }
+ } else {
+ /*
+ * Just encode the value in UTF-8
+ */
+ COPY_BUF(0 ,out, i, value);
+ out[i] = 0;
+ if ((ctxt->sax != NULL) && (ctxt->sax->characters != NULL) &&
+ (!ctxt->disableSAX))
+ ctxt->sax->characters(ctxt->userData, out, i);
+ }
+ return;
+ }
+
+ /*
+ * We are seeing an entity reference
+ */
+ ent = xmlParseEntityRef(ctxt);
+ if (ent == NULL) return;
+ if (!ctxt->wellFormed)
+ return;
+ was_checked = ent->checked;
+
+ /* special case of predefined entities */
+ if ((ent->name == NULL) ||
+ (ent->etype == XML_INTERNAL_PREDEFINED_ENTITY)) {
+ val = ent->content;
+ if (val == NULL) return;
+ /*
+ * inline the entity.
+ */
+ if ((ctxt->sax != NULL) && (ctxt->sax->characters != NULL) &&
+ (!ctxt->disableSAX))
+ ctxt->sax->characters(ctxt->userData, val, xmlStrlen(val));
+ return;
+ }
+
+ /*
+ * The first reference to the entity trigger a parsing phase
+ * where the ent->children is filled with the result from
+ * the parsing.
+ * Note: external parsed entities will not be loaded, it is not
+ * required for a non-validating parser, unless the parsing option
+ * of validating, or substituting entities were given. Doing so is
+ * far more secure as the parser will only process data coming from
+ * the document entity by default.
+ */
+ if (((ent->checked == 0) ||
+ ((ent->children == NULL) && (ctxt->options & XML_PARSE_NOENT))) &&
+ ((ent->etype != XML_EXTERNAL_GENERAL_PARSED_ENTITY) ||
+ (ctxt->options & (XML_PARSE_NOENT | XML_PARSE_DTDVALID)))) {
+ unsigned long oldnbent = ctxt->nbentities;
+
+ /*
+ * This is a bit hackish but this seems the best
+ * way to make sure both SAX and DOM entity support
+ * behaves okay.
+ */
+ void *user_data;
+ if (ctxt->userData == ctxt)
+ user_data = NULL;
+ else
+ user_data = ctxt->userData;
+
+ /*
+ * Check that this entity is well formed
+ * 4.3.2: An internal general parsed entity is well-formed
+ * if its replacement text matches the production labeled
+ * content.
+ */
+ if (ent->etype == XML_INTERNAL_GENERAL_ENTITY) {
+ ctxt->depth++;
+ ret = xmlParseBalancedChunkMemoryInternal(ctxt, ent->content,
+ user_data, &list);
+ ctxt->depth--;
+
+ } else if (ent->etype == XML_EXTERNAL_GENERAL_PARSED_ENTITY) {
+ ctxt->depth++;
+ ret = xmlParseExternalEntityPrivate(ctxt->myDoc, ctxt, ctxt->sax,
+ user_data, ctxt->depth, ent->URI,
+ ent->ExternalID, &list);
+ ctxt->depth--;
+ } else {
+ ret = XML_ERR_ENTITY_PE_INTERNAL;
+ xmlErrMsgStr(ctxt, XML_ERR_INTERNAL_ERROR,
+ "invalid entity type found\n", NULL);
+ }
+
+ /*
+ * Store the number of entities needing parsing for this entity
+ * content and do checkings
+ */
+ ent->checked = (ctxt->nbentities - oldnbent + 1) * 2;
+ if ((ent->content != NULL) && (xmlStrchr(ent->content, '<')))
+ ent->checked |= 1;
+ if (ret == XML_ERR_ENTITY_LOOP) {
+ xmlFatalErr(ctxt, XML_ERR_ENTITY_LOOP, NULL);
+ xmlFreeNodeList(list);
+ return;
+ }
+ if (xmlParserEntityCheck(ctxt, 0, ent, 0)) {
+ xmlFreeNodeList(list);
+ return;
+ }
+
+ if ((ret == XML_ERR_OK) && (list != NULL)) {
+ if (((ent->etype == XML_INTERNAL_GENERAL_ENTITY) ||
+ (ent->etype == XML_EXTERNAL_GENERAL_PARSED_ENTITY))&&
+ (ent->children == NULL)) {
+ ent->children = list;
+ if (ctxt->replaceEntities) {
+ /*
+ * Prune it directly in the generated document
+ * except for single text nodes.
+ */
+ if (((list->type == XML_TEXT_NODE) &&
+ (list->next == NULL)) ||
+ (ctxt->parseMode == XML_PARSE_READER)) {
+ list->parent = (xmlNodePtr) ent;
+ list = NULL;
+ ent->owner = 1;
+ } else {
+ ent->owner = 0;
+ while (list != NULL) {
+ list->parent = (xmlNodePtr) ctxt->node;
+ list->doc = ctxt->myDoc;
+ if (list->next == NULL)
+ ent->last = list;
+ list = list->next;
+ }
+ list = ent->children;
+#ifdef LIBXML_LEGACY_ENABLED
+ if (ent->etype == XML_EXTERNAL_GENERAL_PARSED_ENTITY)
+ xmlAddEntityReference(ent, list, NULL);
+#endif /* LIBXML_LEGACY_ENABLED */
+ }
+ } else {
+ ent->owner = 1;
+ while (list != NULL) {
+ list->parent = (xmlNodePtr) ent;
+ xmlSetTreeDoc(list, ent->doc);
+ if (list->next == NULL)
+ ent->last = list;
+ list = list->next;
+ }
+ }
+ } else {
+ xmlFreeNodeList(list);
+ list = NULL;
+ }
+ } else if ((ret != XML_ERR_OK) &&
+ (ret != XML_WAR_UNDECLARED_ENTITY)) {
+ xmlFatalErrMsgStr(ctxt, XML_ERR_UNDECLARED_ENTITY,
+ "Entity '%s' failed to parse\n", ent->name);
+ xmlParserEntityCheck(ctxt, 0, ent, 0);
+ } else if (list != NULL) {
+ xmlFreeNodeList(list);
+ list = NULL;
+ }
+ if (ent->checked == 0)
+ ent->checked = 2;
+ } else if (ent->checked != 1) {
+ ctxt->nbentities += ent->checked / 2;
+ }
+
+ /*
+ * Now that the entity content has been gathered
+ * provide it to the application, this can take different forms based
+ * on the parsing modes.
+ */
+ if (ent->children == NULL) {
+ /*
+ * Probably running in SAX mode and the callbacks don't
+ * build the entity content. So unless we already went
+ * though parsing for first checking go though the entity
+ * content to generate callbacks associated to the entity
+ */
+ if (was_checked != 0) {
+ void *user_data;
+ /*
+ * This is a bit hackish but this seems the best
+ * way to make sure both SAX and DOM entity support
+ * behaves okay.
+ */
+ if (ctxt->userData == ctxt)
+ user_data = NULL;
+ else
+ user_data = ctxt->userData;
+
+ if (ent->etype == XML_INTERNAL_GENERAL_ENTITY) {
+ ctxt->depth++;
+ ret = xmlParseBalancedChunkMemoryInternal(ctxt,
+ ent->content, user_data, NULL);
+ ctxt->depth--;
+ } else if (ent->etype ==
+ XML_EXTERNAL_GENERAL_PARSED_ENTITY) {
+ ctxt->depth++;
+ ret = xmlParseExternalEntityPrivate(ctxt->myDoc, ctxt,
+ ctxt->sax, user_data, ctxt->depth,
+ ent->URI, ent->ExternalID, NULL);
+ ctxt->depth--;
+ } else {
+ ret = XML_ERR_ENTITY_PE_INTERNAL;
+ xmlErrMsgStr(ctxt, XML_ERR_INTERNAL_ERROR,
+ "invalid entity type found\n", NULL);
+ }
+ if (ret == XML_ERR_ENTITY_LOOP) {
+ xmlFatalErr(ctxt, XML_ERR_ENTITY_LOOP, NULL);
+ return;
+ }
+ }
+ if ((ctxt->sax != NULL) && (ctxt->sax->reference != NULL) &&
+ (ctxt->replaceEntities == 0) && (!ctxt->disableSAX)) {
+ /*
+ * Entity reference callback comes second, it's somewhat
+ * superfluous but a compatibility to historical behaviour
+ */
+ ctxt->sax->reference(ctxt->userData, ent->name);
+ }
+ return;
+ }
+
+ /*
+ * If we didn't get any children for the entity being built
+ */
+ if ((ctxt->sax != NULL) && (ctxt->sax->reference != NULL) &&
+ (ctxt->replaceEntities == 0) && (!ctxt->disableSAX)) {
+ /*
+ * Create a node.
+ */
+ ctxt->sax->reference(ctxt->userData, ent->name);
+ return;
+ }
+
+ if ((ctxt->replaceEntities) || (ent->children == NULL)) {
+ /*
+ * There is a problem on the handling of _private for entities
+ * (bug 155816): Should we copy the content of the field from
+ * the entity (possibly overwriting some value set by the user
+ * when a copy is created), should we leave it alone, or should
+ * we try to take care of different situations? The problem
+ * is exacerbated by the usage of this field by the xmlReader.
+ * To fix this bug, we look at _private on the created node
+ * and, if it's NULL, we copy in whatever was in the entity.
+ * If it's not NULL we leave it alone. This is somewhat of a
+ * hack - maybe we should have further tests to determine
+ * what to do.
+ */
+ if ((ctxt->node != NULL) && (ent->children != NULL)) {
+ /*
+ * Seems we are generating the DOM content, do
+ * a simple tree copy for all references except the first
+ * In the first occurrence list contains the replacement.
+ */
+ if (((list == NULL) && (ent->owner == 0)) ||
+ (ctxt->parseMode == XML_PARSE_READER)) {
+ xmlNodePtr nw = NULL, cur, firstChild = NULL;
+
+ /*
+ * We are copying here, make sure there is no abuse
+ */
+ ctxt->sizeentcopy += ent->length + 5;
+ if (xmlParserEntityCheck(ctxt, 0, ent, ctxt->sizeentcopy))
+ return;
+
+ /*
+ * when operating on a reader, the entities definitions
+ * are always owning the entities subtree.
+ if (ctxt->parseMode == XML_PARSE_READER)
+ ent->owner = 1;
+ */
+
+ cur = ent->children;
+ while (cur != NULL) {
+ nw = xmlDocCopyNode(cur, ctxt->myDoc, 1);
+ if (nw != NULL) {
+ if (nw->_private == NULL)
+ nw->_private = cur->_private;
+ if (firstChild == NULL){
+ firstChild = nw;
+ }
+ nw = xmlAddChild(ctxt->node, nw);
+ }
+ if (cur == ent->last) {
+ /*
+ * needed to detect some strange empty
+ * node cases in the reader tests
+ */
+ if ((ctxt->parseMode == XML_PARSE_READER) &&
+ (nw != NULL) &&
+ (nw->type == XML_ELEMENT_NODE) &&
+ (nw->children == NULL))
+ nw->extra = 1;
+
+ break;
+ }
+ cur = cur->next;
+ }
+#ifdef LIBXML_LEGACY_ENABLED
+ if (ent->etype == XML_EXTERNAL_GENERAL_PARSED_ENTITY)
+ xmlAddEntityReference(ent, firstChild, nw);
+#endif /* LIBXML_LEGACY_ENABLED */
+ } else if ((list == NULL) || (ctxt->inputNr > 0)) {
+ xmlNodePtr nw = NULL, cur, next, last,
+ firstChild = NULL;
+
+ /*
+ * We are copying here, make sure there is no abuse
+ */
+ ctxt->sizeentcopy += ent->length + 5;
+ if (xmlParserEntityCheck(ctxt, 0, ent, ctxt->sizeentcopy))
+ return;
+
+ /*
+ * Copy the entity child list and make it the new
+ * entity child list. The goal is to make sure any
+ * ID or REF referenced will be the one from the
+ * document content and not the entity copy.
+ */
+ cur = ent->children;
+ ent->children = NULL;
+ last = ent->last;
+ ent->last = NULL;
+ while (cur != NULL) {
+ next = cur->next;
+ cur->next = NULL;
+ cur->parent = NULL;
+ nw = xmlDocCopyNode(cur, ctxt->myDoc, 1);
+ if (nw != NULL) {
+ if (nw->_private == NULL)
+ nw->_private = cur->_private;
+ if (firstChild == NULL){
+ firstChild = cur;
+ }
+ xmlAddChild((xmlNodePtr) ent, nw);
+ xmlAddChild(ctxt->node, cur);
+ }
+ if (cur == last)
+ break;
+ cur = next;
+ }
+ if (ent->owner == 0)
+ ent->owner = 1;
+#ifdef LIBXML_LEGACY_ENABLED
+ if (ent->etype == XML_EXTERNAL_GENERAL_PARSED_ENTITY)
+ xmlAddEntityReference(ent, firstChild, nw);
+#endif /* LIBXML_LEGACY_ENABLED */
+ } else {
+ const xmlChar *nbktext;
+
+ /*
+ * the name change is to avoid coalescing of the
+ * node with a possible previous text one which
+ * would make ent->children a dangling pointer
+ */
+ nbktext = xmlDictLookup(ctxt->dict, BAD_CAST "nbktext",
+ -1);
+ if (ent->children->type == XML_TEXT_NODE)
+ ent->children->name = nbktext;
+ if ((ent->last != ent->children) &&
+ (ent->last->type == XML_TEXT_NODE))
+ ent->last->name = nbktext;
+ xmlAddChildList(ctxt->node, ent->children);
+ }
+
+ /*
+ * This is to avoid a nasty side effect, see
+ * characters() in SAX.c
+ */
+ ctxt->nodemem = 0;
+ ctxt->nodelen = 0;
+ return;
+ }
+ }
+}
+
+/**
+ * xmlParseEntityRef:
+ * @ctxt: an XML parser context
+ *
+ * parse ENTITY references declarations
+ *
+ * [68] EntityRef ::= '&' Name ';'
+ *
+ * [ WFC: Entity Declared ]
+ * In a document without any DTD, a document with only an internal DTD
+ * subset which contains no parameter entity references, or a document
+ * with "standalone='yes'", the Name given in the entity reference
+ * must match that in an entity declaration, except that well-formed
+ * documents need not declare any of the following entities: amp, lt,
+ * gt, apos, quot. The declaration of a parameter entity must precede
+ * any reference to it. Similarly, the declaration of a general entity
+ * must precede any reference to it which appears in a default value in an
+ * attribute-list declaration. Note that if entities are declared in the
+ * external subset or in external parameter entities, a non-validating
+ * processor is not obligated to read and process their declarations;
+ * for such documents, the rule that an entity must be declared is a
+ * well-formedness constraint only if standalone='yes'.
+ *
+ * [ WFC: Parsed Entity ]
+ * An entity reference must not contain the name of an unparsed entity
+ *
+ * Returns the xmlEntityPtr if found, or NULL otherwise.
+ */
+xmlEntityPtr
+xmlParseEntityRef(xmlParserCtxtPtr ctxt) {
+ const xmlChar *name;
+ xmlEntityPtr ent = NULL;
+
+ GROW;
+ if (ctxt->instate == XML_PARSER_EOF)
+ return(NULL);
+
+ if (RAW != '&')
+ return(NULL);
+ NEXT;
+ name = xmlParseName(ctxt);
+ if (name == NULL) {
+ xmlFatalErrMsg(ctxt, XML_ERR_NAME_REQUIRED,
+ "xmlParseEntityRef: no name\n");
+ return(NULL);
+ }
+ if (RAW != ';') {
+ xmlFatalErr(ctxt, XML_ERR_ENTITYREF_SEMICOL_MISSING, NULL);
+ return(NULL);
+ }
+ NEXT;
+
+ /*
+ * Predefined entities override any extra definition
+ */
+ if ((ctxt->options & XML_PARSE_OLDSAX) == 0) {
+ ent = xmlGetPredefinedEntity(name);
+ if (ent != NULL)
+ return(ent);
+ }
+
+ /*
+ * Increase the number of entity references parsed
+ */
+ ctxt->nbentities++;
+
+ /*
+ * Ask first SAX for entity resolution, otherwise try the
+ * entities which may have stored in the parser context.
+ */
+ if (ctxt->sax != NULL) {
+ if (ctxt->sax->getEntity != NULL)
+ ent = ctxt->sax->getEntity(ctxt->userData, name);
+ if ((ctxt->wellFormed == 1 ) && (ent == NULL) &&
+ (ctxt->options & XML_PARSE_OLDSAX))
+ ent = xmlGetPredefinedEntity(name);
+ if ((ctxt->wellFormed == 1 ) && (ent == NULL) &&
+ (ctxt->userData==ctxt)) {
+ ent = xmlSAX2GetEntity(ctxt, name);
+ }
+ }
+ if (ctxt->instate == XML_PARSER_EOF)
+ return(NULL);
+ /*
+ * [ WFC: Entity Declared ]
+ * In a document without any DTD, a document with only an
+ * internal DTD subset which contains no parameter entity
+ * references, or a document with "standalone='yes'", the
+ * Name given in the entity reference must match that in an
+ * entity declaration, except that well-formed documents
+ * need not declare any of the following entities: amp, lt,
+ * gt, apos, quot.
+ * The declaration of a parameter entity must precede any
+ * reference to it.
+ * Similarly, the declaration of a general entity must
+ * precede any reference to it which appears in a default
+ * value in an attribute-list declaration. Note that if
+ * entities are declared in the external subset or in
+ * external parameter entities, a non-validating processor
+ * is not obligated to read and process their declarations;
+ * for such documents, the rule that an entity must be
+ * declared is a well-formedness constraint only if
+ * standalone='yes'.
+ */
+ if (ent == NULL) {
+ if ((ctxt->standalone == 1) ||
+ ((ctxt->hasExternalSubset == 0) &&
+ (ctxt->hasPErefs == 0))) {
+ xmlFatalErrMsgStr(ctxt, XML_ERR_UNDECLARED_ENTITY,
+ "Entity '%s' not defined\n", name);
+ } else {
+ xmlErrMsgStr(ctxt, XML_WAR_UNDECLARED_ENTITY,
+ "Entity '%s' not defined\n", name);
+ if ((ctxt->inSubset == 0) &&
+ (ctxt->sax != NULL) &&
+ (ctxt->sax->reference != NULL)) {
+ ctxt->sax->reference(ctxt->userData, name);
+ }
+ }
+ xmlParserEntityCheck(ctxt, 0, ent, 0);
+ ctxt->valid = 0;
+ }
+
+ /*
+ * [ WFC: Parsed Entity ]
+ * An entity reference must not contain the name of an
+ * unparsed entity
+ */
+ else if (ent->etype == XML_EXTERNAL_GENERAL_UNPARSED_ENTITY) {
+ xmlFatalErrMsgStr(ctxt, XML_ERR_UNPARSED_ENTITY,
+ "Entity reference to unparsed entity %s\n", name);
+ }
+
+ /*
+ * [ WFC: No External Entity References ]
+ * Attribute values cannot contain direct or indirect
+ * entity references to external entities.
+ */
+ else if ((ctxt->instate == XML_PARSER_ATTRIBUTE_VALUE) &&
+ (ent->etype == XML_EXTERNAL_GENERAL_PARSED_ENTITY)) {
+ xmlFatalErrMsgStr(ctxt, XML_ERR_ENTITY_IS_EXTERNAL,
+ "Attribute references external entity '%s'\n", name);
+ }
+ /*
+ * [ WFC: No < in Attribute Values ]
+ * The replacement text of any entity referred to directly or
+ * indirectly in an attribute value (other than "&lt;") must
+ * not contain a <.
+ */
+ else if ((ctxt->instate == XML_PARSER_ATTRIBUTE_VALUE) &&
+ (ent != NULL) &&
+ (ent->etype != XML_INTERNAL_PREDEFINED_ENTITY)) {
+ if (((ent->checked & 1) || (ent->checked == 0)) &&
+ (ent->content != NULL) && (xmlStrchr(ent->content, '<'))) {
+ xmlFatalErrMsgStr(ctxt, XML_ERR_LT_IN_ATTRIBUTE,
+ "'<' in entity '%s' is not allowed in attributes values\n", name);
+ }
+ }
+
+ /*
+ * Internal check, no parameter entities here ...
+ */
+ else {
+ switch (ent->etype) {
+ case XML_INTERNAL_PARAMETER_ENTITY:
+ case XML_EXTERNAL_PARAMETER_ENTITY:
+ xmlFatalErrMsgStr(ctxt, XML_ERR_ENTITY_IS_PARAMETER,
+ "Attempt to reference the parameter entity '%s'\n",
+ name);
+ break;
+ default:
+ break;
+ }
+ }
+
+ /*
+ * [ WFC: No Recursion ]
+ * A parsed entity must not contain a recursive reference
+ * to itself, either directly or indirectly.
+ * Done somewhere else
+ */
+ return(ent);
+}
+
+/**
+ * xmlParseStringEntityRef:
+ * @ctxt: an XML parser context
+ * @str: a pointer to an index in the string
+ *
+ * parse ENTITY references declarations, but this version parses it from
+ * a string value.
+ *
+ * [68] EntityRef ::= '&' Name ';'
+ *
+ * [ WFC: Entity Declared ]
+ * In a document without any DTD, a document with only an internal DTD
+ * subset which contains no parameter entity references, or a document
+ * with "standalone='yes'", the Name given in the entity reference
+ * must match that in an entity declaration, except that well-formed
+ * documents need not declare any of the following entities: amp, lt,
+ * gt, apos, quot. The declaration of a parameter entity must precede
+ * any reference to it. Similarly, the declaration of a general entity
+ * must precede any reference to it which appears in a default value in an
+ * attribute-list declaration. Note that if entities are declared in the
+ * external subset or in external parameter entities, a non-validating
+ * processor is not obligated to read and process their declarations;
+ * for such documents, the rule that an entity must be declared is a
+ * well-formedness constraint only if standalone='yes'.
+ *
+ * [ WFC: Parsed Entity ]
+ * An entity reference must not contain the name of an unparsed entity
+ *
+ * Returns the xmlEntityPtr if found, or NULL otherwise. The str pointer
+ * is updated to the current location in the string.
+ */
+static xmlEntityPtr
+xmlParseStringEntityRef(xmlParserCtxtPtr ctxt, const xmlChar ** str) {
+ xmlChar *name;
+ const xmlChar *ptr;
+ xmlChar cur;
+ xmlEntityPtr ent = NULL;
+
+ if ((str == NULL) || (*str == NULL))
+ return(NULL);
+ ptr = *str;
+ cur = *ptr;
+ if (cur != '&')
+ return(NULL);
+
+ ptr++;
+ name = xmlParseStringName(ctxt, &ptr);
+ if (name == NULL) {
+ xmlFatalErrMsg(ctxt, XML_ERR_NAME_REQUIRED,
+ "xmlParseStringEntityRef: no name\n");
+ *str = ptr;
+ return(NULL);
+ }
+ if (*ptr != ';') {
+ xmlFatalErr(ctxt, XML_ERR_ENTITYREF_SEMICOL_MISSING, NULL);
+ xmlFree(name);
+ *str = ptr;
+ return(NULL);
+ }
+ ptr++;
+
+
+ /*
+ * Predefined entities override any extra definition
+ */
+ if ((ctxt->options & XML_PARSE_OLDSAX) == 0) {
+ ent = xmlGetPredefinedEntity(name);
+ if (ent != NULL) {
+ xmlFree(name);
+ *str = ptr;
+ return(ent);
+ }
+ }
+
+ /*
+ * Increate the number of entity references parsed
+ */
+ ctxt->nbentities++;
+
+ /*
+ * Ask first SAX for entity resolution, otherwise try the
+ * entities which may have stored in the parser context.
+ */
+ if (ctxt->sax != NULL) {
+ if (ctxt->sax->getEntity != NULL)
+ ent = ctxt->sax->getEntity(ctxt->userData, name);
+ if ((ent == NULL) && (ctxt->options & XML_PARSE_OLDSAX))
+ ent = xmlGetPredefinedEntity(name);
+ if ((ent == NULL) && (ctxt->userData==ctxt)) {
+ ent = xmlSAX2GetEntity(ctxt, name);
+ }
+ }
+ if (ctxt->instate == XML_PARSER_EOF) {
+ xmlFree(name);
+ return(NULL);
+ }
+
+ /*
+ * [ WFC: Entity Declared ]
+ * In a document without any DTD, a document with only an
+ * internal DTD subset which contains no parameter entity
+ * references, or a document with "standalone='yes'", the
+ * Name given in the entity reference must match that in an
+ * entity declaration, except that well-formed documents
+ * need not declare any of the following entities: amp, lt,
+ * gt, apos, quot.
+ * The declaration of a parameter entity must precede any
+ * reference to it.
+ * Similarly, the declaration of a general entity must
+ * precede any reference to it which appears in a default
+ * value in an attribute-list declaration. Note that if
+ * entities are declared in the external subset or in
+ * external parameter entities, a non-validating processor
+ * is not obligated to read and process their declarations;
+ * for such documents, the rule that an entity must be
+ * declared is a well-formedness constraint only if
+ * standalone='yes'.
+ */
+ if (ent == NULL) {
+ if ((ctxt->standalone == 1) ||
+ ((ctxt->hasExternalSubset == 0) &&
+ (ctxt->hasPErefs == 0))) {
+ xmlFatalErrMsgStr(ctxt, XML_ERR_UNDECLARED_ENTITY,
+ "Entity '%s' not defined\n", name);
+ } else {
+ xmlErrMsgStr(ctxt, XML_WAR_UNDECLARED_ENTITY,
+ "Entity '%s' not defined\n",
+ name);
+ }
+ xmlParserEntityCheck(ctxt, 0, ent, 0);
+ /* TODO ? check regressions ctxt->valid = 0; */
+ }
+
+ /*
+ * [ WFC: Parsed Entity ]
+ * An entity reference must not contain the name of an
+ * unparsed entity
+ */
+ else if (ent->etype == XML_EXTERNAL_GENERAL_UNPARSED_ENTITY) {
+ xmlFatalErrMsgStr(ctxt, XML_ERR_UNPARSED_ENTITY,
+ "Entity reference to unparsed entity %s\n", name);
+ }
+
+ /*
+ * [ WFC: No External Entity References ]
+ * Attribute values cannot contain direct or indirect
+ * entity references to external entities.
+ */
+ else if ((ctxt->instate == XML_PARSER_ATTRIBUTE_VALUE) &&
+ (ent->etype == XML_EXTERNAL_GENERAL_PARSED_ENTITY)) {
+ xmlFatalErrMsgStr(ctxt, XML_ERR_ENTITY_IS_EXTERNAL,
+ "Attribute references external entity '%s'\n", name);
+ }
+ /*
+ * [ WFC: No < in Attribute Values ]
+ * The replacement text of any entity referred to directly or
+ * indirectly in an attribute value (other than "&lt;") must
+ * not contain a <.
+ */
+ else if ((ctxt->instate == XML_PARSER_ATTRIBUTE_VALUE) &&
+ (ent != NULL) && (ent->content != NULL) &&
+ (ent->etype != XML_INTERNAL_PREDEFINED_ENTITY) &&
+ (xmlStrchr(ent->content, '<'))) {
+ xmlFatalErrMsgStr(ctxt, XML_ERR_LT_IN_ATTRIBUTE,
+ "'<' in entity '%s' is not allowed in attributes values\n",
+ name);
+ }
+
+ /*
+ * Internal check, no parameter entities here ...
+ */
+ else {
+ switch (ent->etype) {
+ case XML_INTERNAL_PARAMETER_ENTITY:
+ case XML_EXTERNAL_PARAMETER_ENTITY:
+ xmlFatalErrMsgStr(ctxt, XML_ERR_ENTITY_IS_PARAMETER,
+ "Attempt to reference the parameter entity '%s'\n",
+ name);
+ break;
+ default:
+ break;
+ }
+ }
+
+ /*
+ * [ WFC: No Recursion ]
+ * A parsed entity must not contain a recursive reference
+ * to itself, either directly or indirectly.
+ * Done somewhere else
+ */
+
+ xmlFree(name);
+ *str = ptr;
+ return(ent);
+}
+
+/**
+ * xmlParsePEReference:
+ * @ctxt: an XML parser context
+ *
+ * parse PEReference declarations
+ * The entity content is handled directly by pushing it's content as
+ * a new input stream.
+ *
+ * [69] PEReference ::= '%' Name ';'
+ *
+ * [ WFC: No Recursion ]
+ * A parsed entity must not contain a recursive
+ * reference to itself, either directly or indirectly.
+ *
+ * [ WFC: Entity Declared ]
+ * In a document without any DTD, a document with only an internal DTD
+ * subset which contains no parameter entity references, or a document
+ * with "standalone='yes'", ... ... The declaration of a parameter
+ * entity must precede any reference to it...
+ *
+ * [ VC: Entity Declared ]
+ * In a document with an external subset or external parameter entities
+ * with "standalone='no'", ... ... The declaration of a parameter entity
+ * must precede any reference to it...
+ *
+ * [ WFC: In DTD ]
+ * Parameter-entity references may only appear in the DTD.
+ * NOTE: misleading but this is handled.
+ */
+void
+xmlParsePEReference(xmlParserCtxtPtr ctxt)
+{
+ const xmlChar *name;
+ xmlEntityPtr entity = NULL;
+ xmlParserInputPtr input;
+
+ if (RAW != '%')
+ return;
+ NEXT;
+ name = xmlParseName(ctxt);
+ if (name == NULL) {
+ xmlFatalErrMsg(ctxt, XML_ERR_NAME_REQUIRED,
+ "xmlParsePEReference: no name\n");
+ return;
+ }
+ if (RAW != ';') {
+ xmlFatalErr(ctxt, XML_ERR_ENTITYREF_SEMICOL_MISSING, NULL);
+ return;
+ }
+
+ NEXT;
+
+ /*
+ * Increate the number of entity references parsed
+ */
+ ctxt->nbentities++;
+
+ /*
+ * Request the entity from SAX
+ */
+ if ((ctxt->sax != NULL) &&
+ (ctxt->sax->getParameterEntity != NULL))
+ entity = ctxt->sax->getParameterEntity(ctxt->userData, name);
+ if (ctxt->instate == XML_PARSER_EOF)
+ return;
+ if (entity == NULL) {
+ /*
+ * [ WFC: Entity Declared ]
+ * In a document without any DTD, a document with only an
+ * internal DTD subset which contains no parameter entity
+ * references, or a document with "standalone='yes'", ...
+ * ... The declaration of a parameter entity must precede
+ * any reference to it...
+ */
+ if ((ctxt->standalone == 1) ||
+ ((ctxt->hasExternalSubset == 0) &&
+ (ctxt->hasPErefs == 0))) {
+ xmlFatalErrMsgStr(ctxt, XML_ERR_UNDECLARED_ENTITY,
+ "PEReference: %%%s; not found\n",
+ name);
+ } else {
+ /*
+ * [ VC: Entity Declared ]
+ * In a document with an external subset or external
+ * parameter entities with "standalone='no'", ...
+ * ... The declaration of a parameter entity must
+ * precede any reference to it...
+ */
+ xmlWarningMsg(ctxt, XML_WAR_UNDECLARED_ENTITY,
+ "PEReference: %%%s; not found\n",
+ name, NULL);
+ ctxt->valid = 0;
+ }
+ xmlParserEntityCheck(ctxt, 0, NULL, 0);
+ } else {
+ /*
+ * Internal checking in case the entity quest barfed
+ */
+ if ((entity->etype != XML_INTERNAL_PARAMETER_ENTITY) &&
+ (entity->etype != XML_EXTERNAL_PARAMETER_ENTITY)) {
+ xmlWarningMsg(ctxt, XML_WAR_UNDECLARED_ENTITY,
+ "Internal: %%%s; is not a parameter entity\n",
+ name, NULL);
+ } else if (ctxt->input->free != deallocblankswrapper) {
+ input = xmlNewBlanksWrapperInputStream(ctxt, entity);
+ if (xmlPushInput(ctxt, input) < 0)
+ return;
+ } else {
+ /*
+ * TODO !!!
+ * handle the extra spaces added before and after
+ * c.f. http://www.w3.org/TR/REC-xml#as-PE
+ */
+ input = xmlNewEntityInputStream(ctxt, entity);
+ if (xmlPushInput(ctxt, input) < 0)
+ return;
+ if ((entity->etype == XML_EXTERNAL_PARAMETER_ENTITY) &&
+ (CMP5(CUR_PTR, '<', '?', 'x', 'm', 'l')) &&
+ (IS_BLANK_CH(NXT(5)))) {
+ xmlParseTextDecl(ctxt);
+ if (ctxt->errNo ==
+ XML_ERR_UNSUPPORTED_ENCODING) {
+ /*
+ * The XML REC instructs us to stop parsing
+ * right here
+ */
+ xmlHaltParser(ctxt);
+ return;
+ }
+ }
+ }
+ }
+ ctxt->hasPErefs = 1;
+}
+
+/**
+ * xmlLoadEntityContent:
+ * @ctxt: an XML parser context
+ * @entity: an unloaded system entity
+ *
+ * Load the original content of the given system entity from the
+ * ExternalID/SystemID given. This is to be used for Included in Literal
+ * http://www.w3.org/TR/REC-xml/#inliteral processing of entities references
+ *
+ * Returns 0 in case of success and -1 in case of failure
+ */
+static int
+xmlLoadEntityContent(xmlParserCtxtPtr ctxt, xmlEntityPtr entity) {
+ xmlParserInputPtr input;
+ xmlBufferPtr buf;
+ int l, c;
+ int count = 0;
+
+ if ((ctxt == NULL) || (entity == NULL) ||
+ ((entity->etype != XML_EXTERNAL_PARAMETER_ENTITY) &&
+ (entity->etype != XML_EXTERNAL_GENERAL_PARSED_ENTITY)) ||
+ (entity->content != NULL)) {
+ xmlFatalErr(ctxt, XML_ERR_INTERNAL_ERROR,
+ "xmlLoadEntityContent parameter error");
+ return(-1);
+ }
+
+ if (xmlParserDebugEntities)
+ xmlGenericError(xmlGenericErrorContext,
+ "Reading %s entity content input\n", entity->name);
+
+ buf = xmlBufferCreate();
+ if (buf == NULL) {
+ xmlFatalErr(ctxt, XML_ERR_INTERNAL_ERROR,
+ "xmlLoadEntityContent parameter error");
+ return(-1);
+ }
+
+ input = xmlNewEntityInputStream(ctxt, entity);
+ if (input == NULL) {
+ xmlFatalErr(ctxt, XML_ERR_INTERNAL_ERROR,
+ "xmlLoadEntityContent input error");
+ xmlBufferFree(buf);
+ return(-1);
+ }
+
+ /*
+ * Push the entity as the current input, read char by char
+ * saving to the buffer until the end of the entity or an error
+ */
+ if (xmlPushInput(ctxt, input) < 0) {
+ xmlBufferFree(buf);
+ return(-1);
+ }
+
+ GROW;
+ c = CUR_CHAR(l);
+ while ((ctxt->input == input) && (ctxt->input->cur < ctxt->input->end) &&
+ (IS_CHAR(c))) {
+ xmlBufferAdd(buf, ctxt->input->cur, l);
+ if (count++ > XML_PARSER_CHUNK_SIZE) {
+ count = 0;
+ GROW;
+ if (ctxt->instate == XML_PARSER_EOF) {
+ xmlBufferFree(buf);
+ return(-1);
+ }
+ }
+ NEXTL(l);
+ c = CUR_CHAR(l);
+ if (c == 0) {
+ count = 0;
+ GROW;
+ if (ctxt->instate == XML_PARSER_EOF) {
+ xmlBufferFree(buf);
+ return(-1);
+ }
+ c = CUR_CHAR(l);
+ }
+ }
+
+ if ((ctxt->input == input) && (ctxt->input->cur >= ctxt->input->end)) {
+ xmlPopInput(ctxt);
+ } else if (!IS_CHAR(c)) {
+ xmlFatalErrMsgInt(ctxt, XML_ERR_INVALID_CHAR,
+ "xmlLoadEntityContent: invalid char value %d\n",
+ c);
+ xmlBufferFree(buf);
+ return(-1);
+ }
+ entity->content = buf->content;
+ buf->content = NULL;
+ xmlBufferFree(buf);
+
+ return(0);
+}
+
+/**
+ * xmlParseStringPEReference:
+ * @ctxt: an XML parser context
+ * @str: a pointer to an index in the string
+ *
+ * parse PEReference declarations
+ *
+ * [69] PEReference ::= '%' Name ';'
+ *
+ * [ WFC: No Recursion ]
+ * A parsed entity must not contain a recursive
+ * reference to itself, either directly or indirectly.
+ *
+ * [ WFC: Entity Declared ]
+ * In a document without any DTD, a document with only an internal DTD
+ * subset which contains no parameter entity references, or a document
+ * with "standalone='yes'", ... ... The declaration of a parameter
+ * entity must precede any reference to it...
+ *
+ * [ VC: Entity Declared ]
+ * In a document with an external subset or external parameter entities
+ * with "standalone='no'", ... ... The declaration of a parameter entity
+ * must precede any reference to it...
+ *
+ * [ WFC: In DTD ]
+ * Parameter-entity references may only appear in the DTD.
+ * NOTE: misleading but this is handled.
+ *
+ * Returns the string of the entity content.
+ * str is updated to the current value of the index
+ */
+static xmlEntityPtr
+xmlParseStringPEReference(xmlParserCtxtPtr ctxt, const xmlChar **str) {
+ const xmlChar *ptr;
+ xmlChar cur;
+ xmlChar *name;
+ xmlEntityPtr entity = NULL;
+
+ if ((str == NULL) || (*str == NULL)) return(NULL);
+ ptr = *str;
+ cur = *ptr;
+ if (cur != '%')
+ return(NULL);
+ ptr++;
+ name = xmlParseStringName(ctxt, &ptr);
+ if (name == NULL) {
+ xmlFatalErrMsg(ctxt, XML_ERR_NAME_REQUIRED,
+ "xmlParseStringPEReference: no name\n");
+ *str = ptr;
+ return(NULL);
+ }
+ cur = *ptr;
+ if (cur != ';') {
+ xmlFatalErr(ctxt, XML_ERR_ENTITYREF_SEMICOL_MISSING, NULL);
+ xmlFree(name);
+ *str = ptr;
+ return(NULL);
+ }
+ ptr++;
+
+ /*
+ * Increate the number of entity references parsed
+ */
+ ctxt->nbentities++;
+
+ /*
+ * Request the entity from SAX
+ */
+ if ((ctxt->sax != NULL) &&
+ (ctxt->sax->getParameterEntity != NULL))
+ entity = ctxt->sax->getParameterEntity(ctxt->userData, name);
+ if (ctxt->instate == XML_PARSER_EOF) {
+ xmlFree(name);
+ return(NULL);
+ }
+ if (entity == NULL) {
+ /*
+ * [ WFC: Entity Declared ]
+ * In a document without any DTD, a document with only an
+ * internal DTD subset which contains no parameter entity
+ * references, or a document with "standalone='yes'", ...
+ * ... The declaration of a parameter entity must precede
+ * any reference to it...
+ */
+ if ((ctxt->standalone == 1) ||
+ ((ctxt->hasExternalSubset == 0) && (ctxt->hasPErefs == 0))) {
+ xmlFatalErrMsgStr(ctxt, XML_ERR_UNDECLARED_ENTITY,
+ "PEReference: %%%s; not found\n", name);
+ } else {
+ /*
+ * [ VC: Entity Declared ]
+ * In a document with an external subset or external
+ * parameter entities with "standalone='no'", ...
+ * ... The declaration of a parameter entity must
+ * precede any reference to it...
+ */
+ xmlWarningMsg(ctxt, XML_WAR_UNDECLARED_ENTITY,
+ "PEReference: %%%s; not found\n",
+ name, NULL);
+ ctxt->valid = 0;
+ }
+ xmlParserEntityCheck(ctxt, 0, NULL, 0);
+ } else {
+ /*
+ * Internal checking in case the entity quest barfed
+ */
+ if ((entity->etype != XML_INTERNAL_PARAMETER_ENTITY) &&
+ (entity->etype != XML_EXTERNAL_PARAMETER_ENTITY)) {
+ xmlWarningMsg(ctxt, XML_WAR_UNDECLARED_ENTITY,
+ "%%%s; is not a parameter entity\n",
+ name, NULL);
+ }
+ }
+ ctxt->hasPErefs = 1;
+ xmlFree(name);
+ *str = ptr;
+ return(entity);
+}
+
+/**
+ * xmlParseDocTypeDecl:
+ * @ctxt: an XML parser context
+ *
+ * parse a DOCTYPE declaration
+ *
+ * [28] doctypedecl ::= '<!DOCTYPE' S Name (S ExternalID)? S?
+ * ('[' (markupdecl | PEReference | S)* ']' S?)? '>'
+ *
+ * [ VC: Root Element Type ]
+ * The Name in the document type declaration must match the element
+ * type of the root element.
+ */
+
+void
+xmlParseDocTypeDecl(xmlParserCtxtPtr ctxt) {
+ const xmlChar *name = NULL;
+ xmlChar *ExternalID = NULL;
+ xmlChar *URI = NULL;
+
+ /*
+ * We know that '<!DOCTYPE' has been detected.
+ */
+ SKIP(9);
+
+ SKIP_BLANKS;
+
+ /*
+ * Parse the DOCTYPE name.
+ */
+ name = xmlParseName(ctxt);
+ if (name == NULL) {
+ xmlFatalErrMsg(ctxt, XML_ERR_NAME_REQUIRED,
+ "xmlParseDocTypeDecl : no DOCTYPE name !\n");
+ }
+ ctxt->intSubName = name;
+
+ SKIP_BLANKS;
+
+ /*
+ * Check for SystemID and ExternalID
+ */
+ URI = xmlParseExternalID(ctxt, &ExternalID, 1);
+
+ if ((URI != NULL) || (ExternalID != NULL)) {
+ ctxt->hasExternalSubset = 1;
+ }
+ ctxt->extSubURI = URI;
+ ctxt->extSubSystem = ExternalID;
+
+ SKIP_BLANKS;
+
+ /*
+ * Create and update the internal subset.
+ */
+ if ((ctxt->sax != NULL) && (ctxt->sax->internalSubset != NULL) &&
+ (!ctxt->disableSAX))
+ ctxt->sax->internalSubset(ctxt->userData, name, ExternalID, URI);
+ if (ctxt->instate == XML_PARSER_EOF)
+ return;
+
+ /*
+ * Is there any internal subset declarations ?
+ * they are handled separately in xmlParseInternalSubset()
+ */
+ if (RAW == '[')
+ return;
+
+ /*
+ * We should be at the end of the DOCTYPE declaration.
+ */
+ if (RAW != '>') {
+ xmlFatalErr(ctxt, XML_ERR_DOCTYPE_NOT_FINISHED, NULL);
+ }
+ NEXT;
+}
+
+/**
+ * xmlParseInternalSubset:
+ * @ctxt: an XML parser context
+ *
+ * parse the internal subset declaration
+ *
+ * [28 end] ('[' (markupdecl | PEReference | S)* ']' S?)? '>'
+ */
+
+static void
+xmlParseInternalSubset(xmlParserCtxtPtr ctxt) {
+ /*
+ * Is there any DTD definition ?
+ */
+ if (RAW == '[') {
+ ctxt->instate = XML_PARSER_DTD;
+ NEXT;
+ /*
+ * Parse the succession of Markup declarations and
+ * PEReferences.
+ * Subsequence (markupdecl | PEReference | S)*
+ */
+ while ((RAW != ']') && (ctxt->instate != XML_PARSER_EOF)) {
+ const xmlChar *check = CUR_PTR;
+ unsigned int cons = ctxt->input->consumed;
+
+ SKIP_BLANKS;
+ xmlParseMarkupDecl(ctxt);
+ xmlParsePEReference(ctxt);
+
+ /*
+ * Pop-up of finished entities.
+ */
+ while ((RAW == 0) && (ctxt->inputNr > 1))
+ xmlPopInput(ctxt);
+
+ if ((CUR_PTR == check) && (cons == ctxt->input->consumed)) {
+ xmlFatalErr(ctxt, XML_ERR_INTERNAL_ERROR,
+ "xmlParseInternalSubset: error detected in Markup declaration\n");
+ break;
+ }
+ }
+ if (RAW == ']') {
+ NEXT;
+ SKIP_BLANKS;
+ }
+ }
+
+ /*
+ * We should be at the end of the DOCTYPE declaration.
+ */
+ if (RAW != '>') {
+ xmlFatalErr(ctxt, XML_ERR_DOCTYPE_NOT_FINISHED, NULL);
+ return;
+ }
+ NEXT;
+}
+
+#ifdef LIBXML_SAX1_ENABLED
+/**
+ * xmlParseAttribute:
+ * @ctxt: an XML parser context
+ * @value: a xmlChar ** used to store the value of the attribute
+ *
+ * parse an attribute
+ *
+ * [41] Attribute ::= Name Eq AttValue
+ *
+ * [ WFC: No External Entity References ]
+ * Attribute values cannot contain direct or indirect entity references
+ * to external entities.
+ *
+ * [ WFC: No < in Attribute Values ]
+ * The replacement text of any entity referred to directly or indirectly in
+ * an attribute value (other than "&lt;") must not contain a <.
+ *
+ * [ VC: Attribute Value Type ]
+ * The attribute must have been declared; the value must be of the type
+ * declared for it.
+ *
+ * [25] Eq ::= S? '=' S?
+ *
+ * With namespace:
+ *
+ * [NS 11] Attribute ::= QName Eq AttValue
+ *
+ * Also the case QName == xmlns:??? is handled independently as a namespace
+ * definition.
+ *
+ * Returns the attribute name, and the value in *value.
+ */
+
+const xmlChar *
+xmlParseAttribute(xmlParserCtxtPtr ctxt, xmlChar **value) {
+ const xmlChar *name;
+ xmlChar *val;
+
+ *value = NULL;
+ GROW;
+ name = xmlParseName(ctxt);
+ if (name == NULL) {
+ xmlFatalErrMsg(ctxt, XML_ERR_NAME_REQUIRED,
+ "error parsing attribute name\n");
+ return(NULL);
+ }
+
+ /*
+ * read the value
+ */
+ SKIP_BLANKS;
+ if (RAW == '=') {
+ NEXT;
+ SKIP_BLANKS;
+ val = xmlParseAttValue(ctxt);
+ ctxt->instate = XML_PARSER_CONTENT;
+ } else {
+ xmlFatalErrMsgStr(ctxt, XML_ERR_ATTRIBUTE_WITHOUT_VALUE,
+ "Specification mandate value for attribute %s\n", name);
+ return(NULL);
+ }
+
+ /*
+ * Check that xml:lang conforms to the specification
+ * No more registered as an error, just generate a warning now
+ * since this was deprecated in XML second edition
+ */
+ if ((ctxt->pedantic) && (xmlStrEqual(name, BAD_CAST "xml:lang"))) {
+ if (!xmlCheckLanguageID(val)) {
+ xmlWarningMsg(ctxt, XML_WAR_LANG_VALUE,
+ "Malformed value for xml:lang : %s\n",
+ val, NULL);
+ }
+ }
+
+ /*
+ * Check that xml:space conforms to the specification
+ */
+ if (xmlStrEqual(name, BAD_CAST "xml:space")) {
+ if (xmlStrEqual(val, BAD_CAST "default"))
+ *(ctxt->space) = 0;
+ else if (xmlStrEqual(val, BAD_CAST "preserve"))
+ *(ctxt->space) = 1;
+ else {
+ xmlWarningMsg(ctxt, XML_WAR_SPACE_VALUE,
+"Invalid value \"%s\" for xml:space : \"default\" or \"preserve\" expected\n",
+ val, NULL);
+ }
+ }
+
+ *value = val;
+ return(name);
+}
+
+/**
+ * xmlParseStartTag:
+ * @ctxt: an XML parser context
+ *
+ * parse a start of tag either for rule element or
+ * EmptyElement. In both case we don't parse the tag closing chars.
+ *
+ * [40] STag ::= '<' Name (S Attribute)* S? '>'
+ *
+ * [ WFC: Unique Att Spec ]
+ * No attribute name may appear more than once in the same start-tag or
+ * empty-element tag.
+ *
+ * [44] EmptyElemTag ::= '<' Name (S Attribute)* S? '/>'
+ *
+ * [ WFC: Unique Att Spec ]
+ * No attribute name may appear more than once in the same start-tag or
+ * empty-element tag.
+ *
+ * With namespace:
+ *
+ * [NS 8] STag ::= '<' QName (S Attribute)* S? '>'
+ *
+ * [NS 10] EmptyElement ::= '<' QName (S Attribute)* S? '/>'
+ *
+ * Returns the element name parsed
+ */
+
+const xmlChar *
+xmlParseStartTag(xmlParserCtxtPtr ctxt) {
+ const xmlChar *name;
+ const xmlChar *attname;
+ xmlChar *attvalue;
+ const xmlChar **atts = ctxt->atts;
+ int nbatts = 0;
+ int maxatts = ctxt->maxatts;
+ int i;
+
+ if (RAW != '<') return(NULL);
+ NEXT1;
+
+ name = xmlParseName(ctxt);
+ if (name == NULL) {
+ xmlFatalErrMsg(ctxt, XML_ERR_NAME_REQUIRED,
+ "xmlParseStartTag: invalid element name\n");
+ return(NULL);
+ }
+
+ /*
+ * Now parse the attributes, it ends up with the ending
+ *
+ * (S Attribute)* S?
+ */
+ SKIP_BLANKS;
+ GROW;
+
+ while (((RAW != '>') &&
+ ((RAW != '/') || (NXT(1) != '>')) &&
+ (IS_BYTE_CHAR(RAW))) && (ctxt->instate != XML_PARSER_EOF)) {
+ const xmlChar *q = CUR_PTR;
+ unsigned int cons = ctxt->input->consumed;
+
+ attname = xmlParseAttribute(ctxt, &attvalue);
+ if ((attname != NULL) && (attvalue != NULL)) {
+ /*
+ * [ WFC: Unique Att Spec ]
+ * No attribute name may appear more than once in the same
+ * start-tag or empty-element tag.
+ */
+ for (i = 0; i < nbatts;i += 2) {
+ if (xmlStrEqual(atts[i], attname)) {
+ xmlErrAttributeDup(ctxt, NULL, attname);
+ xmlFree(attvalue);
+ goto failed;
+ }
+ }
+ /*
+ * Add the pair to atts
+ */
+ if (atts == NULL) {
+ maxatts = 22; /* allow for 10 attrs by default */
+ atts = (const xmlChar **)
+ xmlMalloc(maxatts * sizeof(xmlChar *));
+ if (atts == NULL) {
+ xmlErrMemory(ctxt, NULL);
+ if (attvalue != NULL)
+ xmlFree(attvalue);
+ goto failed;
+ }
+ ctxt->atts = atts;
+ ctxt->maxatts = maxatts;
+ } else if (nbatts + 4 > maxatts) {
+ const xmlChar **n;
+
+ maxatts *= 2;
+ n = (const xmlChar **) xmlRealloc((void *) atts,
+ maxatts * sizeof(const xmlChar *));
+ if (n == NULL) {
+ xmlErrMemory(ctxt, NULL);
+ if (attvalue != NULL)
+ xmlFree(attvalue);
+ goto failed;
+ }
+ atts = n;
+ ctxt->atts = atts;
+ ctxt->maxatts = maxatts;
+ }
+ atts[nbatts++] = attname;
+ atts[nbatts++] = attvalue;
+ atts[nbatts] = NULL;
+ atts[nbatts + 1] = NULL;
+ } else {
+ if (attvalue != NULL)
+ xmlFree(attvalue);
+ }
+
+failed:
+
+ GROW
+ if ((RAW == '>') || (((RAW == '/') && (NXT(1) == '>'))))
+ break;
+ if (!IS_BLANK_CH(RAW)) {
+ xmlFatalErrMsg(ctxt, XML_ERR_SPACE_REQUIRED,
+ "attributes construct error\n");
+ }
+ SKIP_BLANKS;
+ if ((cons == ctxt->input->consumed) && (q == CUR_PTR) &&
+ (attname == NULL) && (attvalue == NULL)) {
+ xmlFatalErrMsg(ctxt, XML_ERR_INTERNAL_ERROR,
+ "xmlParseStartTag: problem parsing attributes\n");
+ break;
+ }
+ SHRINK;
+ GROW;
+ }
+
+ /*
+ * SAX: Start of Element !
+ */
+ if ((ctxt->sax != NULL) && (ctxt->sax->startElement != NULL) &&
+ (!ctxt->disableSAX)) {
+ if (nbatts > 0)
+ ctxt->sax->startElement(ctxt->userData, name, atts);
+ else
+ ctxt->sax->startElement(ctxt->userData, name, NULL);
+ }
+
+ if (atts != NULL) {
+ /* Free only the content strings */
+ for (i = 1;i < nbatts;i+=2)
+ if (atts[i] != NULL)
+ xmlFree((xmlChar *) atts[i]);
+ }
+ return(name);
+}
+
+/**
+ * xmlParseEndTag1:
+ * @ctxt: an XML parser context
+ * @line: line of the start tag
+ * @nsNr: number of namespaces on the start tag
+ *
+ * parse an end of tag
+ *
+ * [42] ETag ::= '</' Name S? '>'
+ *
+ * With namespace
+ *
+ * [NS 9] ETag ::= '</' QName S? '>'
+ */
+
+static void
+xmlParseEndTag1(xmlParserCtxtPtr ctxt, int line) {
+ const xmlChar *name;
+
+ GROW;
+ if ((RAW != '<') || (NXT(1) != '/')) {
+ xmlFatalErrMsg(ctxt, XML_ERR_LTSLASH_REQUIRED,
+ "xmlParseEndTag: '</' not found\n");
+ return;
+ }
+ SKIP(2);
+
+ name = xmlParseNameAndCompare(ctxt,ctxt->name);
+
+ /*
+ * We should definitely be at the ending "S? '>'" part
+ */
+ GROW;
+ SKIP_BLANKS;
+ if ((!IS_BYTE_CHAR(RAW)) || (RAW != '>')) {
+ xmlFatalErr(ctxt, XML_ERR_GT_REQUIRED, NULL);
+ } else
+ NEXT1;
+
+ /*
+ * [ WFC: Element Type Match ]
+ * The Name in an element's end-tag must match the element type in the
+ * start-tag.
+ *
+ */
+ if (name != (xmlChar*)1) {
+ if (name == NULL) name = BAD_CAST "unparseable";
+ xmlFatalErrMsgStrIntStr(ctxt, XML_ERR_TAG_NAME_MISMATCH,
+ "Opening and ending tag mismatch: %s line %d and %s\n",
+ ctxt->name, line, name);
+ }
+
+ /*
+ * SAX: End of Tag
+ */
+ if ((ctxt->sax != NULL) && (ctxt->sax->endElement != NULL) &&
+ (!ctxt->disableSAX))
+ ctxt->sax->endElement(ctxt->userData, ctxt->name);
+
+ namePop(ctxt);
+ spacePop(ctxt);
+ return;
+}
+
+/**
+ * xmlParseEndTag:
+ * @ctxt: an XML parser context
+ *
+ * parse an end of tag
+ *
+ * [42] ETag ::= '</' Name S? '>'
+ *
+ * With namespace
+ *
+ * [NS 9] ETag ::= '</' QName S? '>'
+ */
+
+void
+xmlParseEndTag(xmlParserCtxtPtr ctxt) {
+ xmlParseEndTag1(ctxt, 0);
+}
+#endif /* LIBXML_SAX1_ENABLED */
+
+/************************************************************************
+ * *
+ * SAX 2 specific operations *
+ * *
+ ************************************************************************/
+
+/*
+ * xmlGetNamespace:
+ * @ctxt: an XML parser context
+ * @prefix: the prefix to lookup
+ *
+ * Lookup the namespace name for the @prefix (which ca be NULL)
+ * The prefix must come from the @ctxt->dict dictionary
+ *
+ * Returns the namespace name or NULL if not bound
+ */
+static const xmlChar *
+xmlGetNamespace(xmlParserCtxtPtr ctxt, const xmlChar *prefix) {
+ int i;
+
+ if (prefix == ctxt->str_xml) return(ctxt->str_xml_ns);
+ for (i = ctxt->nsNr - 2;i >= 0;i-=2)
+ if (ctxt->nsTab[i] == prefix) {
+ if ((prefix == NULL) && (*ctxt->nsTab[i + 1] == 0))
+ return(NULL);
+ return(ctxt->nsTab[i + 1]);
+ }
+ return(NULL);
+}
+
+/**
+ * xmlParseQName:
+ * @ctxt: an XML parser context
+ * @prefix: pointer to store the prefix part
+ *
+ * parse an XML Namespace QName
+ *
+ * [6] QName ::= (Prefix ':')? LocalPart
+ * [7] Prefix ::= NCName
+ * [8] LocalPart ::= NCName
+ *
+ * Returns the Name parsed or NULL
+ */
+
+static const xmlChar *
+xmlParseQName(xmlParserCtxtPtr ctxt, const xmlChar **prefix) {
+ const xmlChar *l, *p;
+
+ GROW;
+
+ l = xmlParseNCName(ctxt);
+ if (l == NULL) {
+ if (CUR == ':') {
+ l = xmlParseName(ctxt);
+ if (l != NULL) {
+ xmlNsErr(ctxt, XML_NS_ERR_QNAME,
+ "Failed to parse QName '%s'\n", l, NULL, NULL);
+ *prefix = NULL;
+ return(l);
+ }
+ }
+ return(NULL);
+ }
+ if (CUR == ':') {
+ NEXT;
+ p = l;
+ l = xmlParseNCName(ctxt);
+ if (l == NULL) {
+ xmlChar *tmp;
+
+ xmlNsErr(ctxt, XML_NS_ERR_QNAME,
+ "Failed to parse QName '%s:'\n", p, NULL, NULL);
+ l = xmlParseNmtoken(ctxt);
+ if (l == NULL)
+ tmp = xmlBuildQName(BAD_CAST "", p, NULL, 0);
+ else {
+ tmp = xmlBuildQName(l, p, NULL, 0);
+ xmlFree((char *)l);
+ }
+ p = xmlDictLookup(ctxt->dict, tmp, -1);
+ if (tmp != NULL) xmlFree(tmp);
+ *prefix = NULL;
+ return(p);
+ }
+ if (CUR == ':') {
+ xmlChar *tmp;
+
+ xmlNsErr(ctxt, XML_NS_ERR_QNAME,
+ "Failed to parse QName '%s:%s:'\n", p, l, NULL);
+ NEXT;
+ tmp = (xmlChar *) xmlParseName(ctxt);
+ if (tmp != NULL) {
+ tmp = xmlBuildQName(tmp, l, NULL, 0);
+ l = xmlDictLookup(ctxt->dict, tmp, -1);
+ if (tmp != NULL) xmlFree(tmp);
+ *prefix = p;
+ return(l);
+ }
+ tmp = xmlBuildQName(BAD_CAST "", l, NULL, 0);
+ l = xmlDictLookup(ctxt->dict, tmp, -1);
+ if (tmp != NULL) xmlFree(tmp);
+ *prefix = p;
+ return(l);
+ }
+ *prefix = p;
+ } else
+ *prefix = NULL;
+ return(l);
+}
+
+/**
+ * xmlParseQNameAndCompare:
+ * @ctxt: an XML parser context
+ * @name: the localname
+ * @prefix: the prefix, if any.
+ *
+ * parse an XML name and compares for match
+ * (specialized for endtag parsing)
+ *
+ * Returns NULL for an illegal name, (xmlChar*) 1 for success
+ * and the name for mismatch
+ */
+
+static const xmlChar *
+xmlParseQNameAndCompare(xmlParserCtxtPtr ctxt, xmlChar const *name,
+ xmlChar const *prefix) {
+ const xmlChar *cmp;
+ const xmlChar *in;
+ const xmlChar *ret;
+ const xmlChar *prefix2;
+
+ if (prefix == NULL) return(xmlParseNameAndCompare(ctxt, name));
+
+ GROW;
+ in = ctxt->input->cur;
+
+ cmp = prefix;
+ while (*in != 0 && *in == *cmp) {
+ ++in;
+ ++cmp;
+ }
+ if ((*cmp == 0) && (*in == ':')) {
+ in++;
+ cmp = name;
+ while (*in != 0 && *in == *cmp) {
+ ++in;
+ ++cmp;
+ }
+ if (*cmp == 0 && (*in == '>' || IS_BLANK_CH (*in))) {
+ /* success */
+ ctxt->input->cur = in;
+ return((const xmlChar*) 1);
+ }
+ }
+ /*
+ * all strings coms from the dictionary, equality can be done directly
+ */
+ ret = xmlParseQName (ctxt, &prefix2);
+ if ((ret == name) && (prefix == prefix2))
+ return((const xmlChar*) 1);
+ return ret;
+}
+
+/**
+ * xmlParseAttValueInternal:
+ * @ctxt: an XML parser context
+ * @len: attribute len result
+ * @alloc: whether the attribute was reallocated as a new string
+ * @normalize: if 1 then further non-CDATA normalization must be done
+ *
+ * parse a value for an attribute.
+ * NOTE: if no normalization is needed, the routine will return pointers
+ * directly from the data buffer.
+ *
+ * 3.3.3 Attribute-Value Normalization:
+ * Before the value of an attribute is passed to the application or
+ * checked for validity, the XML processor must normalize it as follows:
+ * - a character reference is processed by appending the referenced
+ * character to the attribute value
+ * - an entity reference is processed by recursively processing the
+ * replacement text of the entity
+ * - a whitespace character (#x20, #xD, #xA, #x9) is processed by
+ * appending #x20 to the normalized value, except that only a single
+ * #x20 is appended for a "#xD#xA" sequence that is part of an external
+ * parsed entity or the literal entity value of an internal parsed entity
+ * - other characters are processed by appending them to the normalized value
+ * If the declared value is not CDATA, then the XML processor must further
+ * process the normalized attribute value by discarding any leading and
+ * trailing space (#x20) characters, and by replacing sequences of space
+ * (#x20) characters by a single space (#x20) character.
+ * All attributes for which no declaration has been read should be treated
+ * by a non-validating parser as if declared CDATA.
+ *
+ * Returns the AttValue parsed or NULL. The value has to be freed by the
+ * caller if it was copied, this can be detected by val[*len] == 0.
+ */
+
+static xmlChar *
+xmlParseAttValueInternal(xmlParserCtxtPtr ctxt, int *len, int *alloc,
+ int normalize)
+{
+ xmlChar limit = 0;
+ const xmlChar *in = NULL, *start, *end, *last;
+ xmlChar *ret = NULL;
+ int line, col;
+
+ GROW;
+ in = (xmlChar *) CUR_PTR;
+ line = ctxt->input->line;
+ col = ctxt->input->col;
+ if (*in != '"' && *in != '\'') {
+ xmlFatalErr(ctxt, XML_ERR_ATTRIBUTE_NOT_STARTED, NULL);
+ return (NULL);
+ }
+ ctxt->instate = XML_PARSER_ATTRIBUTE_VALUE;
+
+ /*
+ * try to handle in this routine the most common case where no
+ * allocation of a new string is required and where content is
+ * pure ASCII.
+ */
+ limit = *in++;
+ col++;
+ end = ctxt->input->end;
+ start = in;
+ if (in >= end) {
+ const xmlChar *oldbase = ctxt->input->base;
+ GROW;
+ if (oldbase != ctxt->input->base) {
+ long delta = ctxt->input->base - oldbase;
+ start = start + delta;
+ in = in + delta;
+ }
+ end = ctxt->input->end;
+ }
+ if (normalize) {
+ /*
+ * Skip any leading spaces
+ */
+ while ((in < end) && (*in != limit) &&
+ ((*in == 0x20) || (*in == 0x9) ||
+ (*in == 0xA) || (*in == 0xD))) {
+ if (*in == 0xA) {
+ line++; col = 1;
+ } else {
+ col++;
+ }
+ in++;
+ start = in;
+ if (in >= end) {
+ const xmlChar *oldbase = ctxt->input->base;
+ GROW;
+ if (ctxt->instate == XML_PARSER_EOF)
+ return(NULL);
+ if (oldbase != ctxt->input->base) {
+ long delta = ctxt->input->base - oldbase;
+ start = start + delta;
+ in = in + delta;
+ }
+ end = ctxt->input->end;
+ if (((in - start) > XML_MAX_TEXT_LENGTH) &&
+ ((ctxt->options & XML_PARSE_HUGE) == 0)) {
+ xmlFatalErrMsg(ctxt, XML_ERR_ATTRIBUTE_NOT_FINISHED,
+ "AttValue length too long\n");
+ return(NULL);
+ }
+ }
+ }
+ while ((in < end) && (*in != limit) && (*in >= 0x20) &&
+ (*in <= 0x7f) && (*in != '&') && (*in != '<')) {
+ col++;
+ if ((*in++ == 0x20) && (*in == 0x20)) break;
+ if (in >= end) {
+ const xmlChar *oldbase = ctxt->input->base;
+ GROW;
+ if (ctxt->instate == XML_PARSER_EOF)
+ return(NULL);
+ if (oldbase != ctxt->input->base) {
+ long delta = ctxt->input->base - oldbase;
+ start = start + delta;
+ in = in + delta;
+ }
+ end = ctxt->input->end;
+ if (((in - start) > XML_MAX_TEXT_LENGTH) &&
+ ((ctxt->options & XML_PARSE_HUGE) == 0)) {
+ xmlFatalErrMsg(ctxt, XML_ERR_ATTRIBUTE_NOT_FINISHED,
+ "AttValue length too long\n");
+ return(NULL);
+ }
+ }
+ }
+ last = in;
+ /*
+ * skip the trailing blanks
+ */
+ while ((last[-1] == 0x20) && (last > start)) last--;
+ while ((in < end) && (*in != limit) &&
+ ((*in == 0x20) || (*in == 0x9) ||
+ (*in == 0xA) || (*in == 0xD))) {
+ if (*in == 0xA) {
+ line++, col = 1;
+ } else {
+ col++;
+ }
+ in++;
+ if (in >= end) {
+ const xmlChar *oldbase = ctxt->input->base;
+ GROW;
+ if (ctxt->instate == XML_PARSER_EOF)
+ return(NULL);
+ if (oldbase != ctxt->input->base) {
+ long delta = ctxt->input->base - oldbase;
+ start = start + delta;
+ in = in + delta;
+ last = last + delta;
+ }
+ end = ctxt->input->end;
+ if (((in - start) > XML_MAX_TEXT_LENGTH) &&
+ ((ctxt->options & XML_PARSE_HUGE) == 0)) {
+ xmlFatalErrMsg(ctxt, XML_ERR_ATTRIBUTE_NOT_FINISHED,
+ "AttValue length too long\n");
+ return(NULL);
+ }
+ }
+ }
+ if (((in - start) > XML_MAX_TEXT_LENGTH) &&
+ ((ctxt->options & XML_PARSE_HUGE) == 0)) {
+ xmlFatalErrMsg(ctxt, XML_ERR_ATTRIBUTE_NOT_FINISHED,
+ "AttValue length too long\n");
+ return(NULL);
+ }
+ if (*in != limit) goto need_complex;
+ } else {
+ while ((in < end) && (*in != limit) && (*in >= 0x20) &&
+ (*in <= 0x7f) && (*in != '&') && (*in != '<')) {
+ in++;
+ col++;
+ if (in >= end) {
+ const xmlChar *oldbase = ctxt->input->base;
+ GROW;
+ if (ctxt->instate == XML_PARSER_EOF)
+ return(NULL);
+ if (oldbase != ctxt->input->base) {
+ long delta = ctxt->input->base - oldbase;
+ start = start + delta;
+ in = in + delta;
+ }
+ end = ctxt->input->end;
+ if (((in - start) > XML_MAX_TEXT_LENGTH) &&
+ ((ctxt->options & XML_PARSE_HUGE) == 0)) {
+ xmlFatalErrMsg(ctxt, XML_ERR_ATTRIBUTE_NOT_FINISHED,
+ "AttValue length too long\n");
+ return(NULL);
+ }
+ }
+ }
+ last = in;
+ if (((in - start) > XML_MAX_TEXT_LENGTH) &&
+ ((ctxt->options & XML_PARSE_HUGE) == 0)) {
+ xmlFatalErrMsg(ctxt, XML_ERR_ATTRIBUTE_NOT_FINISHED,
+ "AttValue length too long\n");
+ return(NULL);
+ }
+ if (*in != limit) goto need_complex;
+ }
+ in++;
+ col++;
+ if (len != NULL) {
+ *len = last - start;
+ ret = (xmlChar *) start;
+ } else {
+ if (alloc) *alloc = 1;
+ ret = xmlStrndup(start, last - start);
+ }
+ CUR_PTR = in;
+ ctxt->input->line = line;
+ ctxt->input->col = col;
+ if (alloc) *alloc = 0;
+ return ret;
+need_complex:
+ if (alloc) *alloc = 1;
+ return xmlParseAttValueComplex(ctxt, len, normalize);
+}
+
+/**
+ * xmlParseAttribute2:
+ * @ctxt: an XML parser context
+ * @pref: the element prefix
+ * @elem: the element name
+ * @prefix: a xmlChar ** used to store the value of the attribute prefix
+ * @value: a xmlChar ** used to store the value of the attribute
+ * @len: an int * to save the length of the attribute
+ * @alloc: an int * to indicate if the attribute was allocated
+ *
+ * parse an attribute in the new SAX2 framework.
+ *
+ * Returns the attribute name, and the value in *value, .
+ */
+
+static const xmlChar *
+xmlParseAttribute2(xmlParserCtxtPtr ctxt,
+ const xmlChar * pref, const xmlChar * elem,
+ const xmlChar ** prefix, xmlChar ** value,
+ int *len, int *alloc)
+{
+ const xmlChar *name;
+ xmlChar *val, *internal_val = NULL;
+ int normalize = 0;
+
+ *value = NULL;
+ GROW;
+ name = xmlParseQName(ctxt, prefix);
+ if (name == NULL) {
+ xmlFatalErrMsg(ctxt, XML_ERR_NAME_REQUIRED,
+ "error parsing attribute name\n");
+ return (NULL);
+ }
+
+ /*
+ * get the type if needed
+ */
+ if (ctxt->attsSpecial != NULL) {
+ int type;
+
+ type = (int) (long) xmlHashQLookup2(ctxt->attsSpecial,
+ pref, elem, *prefix, name);
+ if (type != 0)
+ normalize = 1;
+ }
+
+ /*
+ * read the value
+ */
+ SKIP_BLANKS;
+ if (RAW == '=') {
+ NEXT;
+ SKIP_BLANKS;
+ val = xmlParseAttValueInternal(ctxt, len, alloc, normalize);
+ if (normalize) {
+ /*
+ * Sometimes a second normalisation pass for spaces is needed
+ * but that only happens if charrefs or entities refernces
+ * have been used in the attribute value, i.e. the attribute
+ * value have been extracted in an allocated string already.
+ */
+ if (*alloc) {
+ const xmlChar *val2;
+
+ val2 = xmlAttrNormalizeSpace2(ctxt, val, len);
+ if ((val2 != NULL) && (val2 != val)) {
+ xmlFree(val);
+ val = (xmlChar *) val2;
+ }
+ }
+ }
+ ctxt->instate = XML_PARSER_CONTENT;
+ } else {
+ xmlFatalErrMsgStr(ctxt, XML_ERR_ATTRIBUTE_WITHOUT_VALUE,
+ "Specification mandate value for attribute %s\n",
+ name);
+ return (NULL);
+ }
+
+ if (*prefix == ctxt->str_xml) {
+ /*
+ * Check that xml:lang conforms to the specification
+ * No more registered as an error, just generate a warning now
+ * since this was deprecated in XML second edition
+ */
+ if ((ctxt->pedantic) && (xmlStrEqual(name, BAD_CAST "lang"))) {
+ internal_val = xmlStrndup(val, *len);
+ if (!xmlCheckLanguageID(internal_val)) {
+ xmlWarningMsg(ctxt, XML_WAR_LANG_VALUE,
+ "Malformed value for xml:lang : %s\n",
+ internal_val, NULL);
+ }
+ }
+
+ /*
+ * Check that xml:space conforms to the specification
+ */
+ if (xmlStrEqual(name, BAD_CAST "space")) {
+ internal_val = xmlStrndup(val, *len);
+ if (xmlStrEqual(internal_val, BAD_CAST "default"))
+ *(ctxt->space) = 0;
+ else if (xmlStrEqual(internal_val, BAD_CAST "preserve"))
+ *(ctxt->space) = 1;
+ else {
+ xmlWarningMsg(ctxt, XML_WAR_SPACE_VALUE,
+ "Invalid value \"%s\" for xml:space : \"default\" or \"preserve\" expected\n",
+ internal_val, NULL);
+ }
+ }
+ if (internal_val) {
+ xmlFree(internal_val);
+ }
+ }
+
+ *value = val;
+ return (name);
+}
+/**
+ * xmlParseStartTag2:
+ * @ctxt: an XML parser context
+ *
+ * parse a start of tag either for rule element or
+ * EmptyElement. In both case we don't parse the tag closing chars.
+ * This routine is called when running SAX2 parsing
+ *
+ * [40] STag ::= '<' Name (S Attribute)* S? '>'
+ *
+ * [ WFC: Unique Att Spec ]
+ * No attribute name may appear more than once in the same start-tag or
+ * empty-element tag.
+ *
+ * [44] EmptyElemTag ::= '<' Name (S Attribute)* S? '/>'
+ *
+ * [ WFC: Unique Att Spec ]
+ * No attribute name may appear more than once in the same start-tag or
+ * empty-element tag.
+ *
+ * With namespace:
+ *
+ * [NS 8] STag ::= '<' QName (S Attribute)* S? '>'
+ *
+ * [NS 10] EmptyElement ::= '<' QName (S Attribute)* S? '/>'
+ *
+ * Returns the element name parsed
+ */
+
+static const xmlChar *
+xmlParseStartTag2(xmlParserCtxtPtr ctxt, const xmlChar **pref,
+ const xmlChar **URI, int *tlen) {
+ const xmlChar *localname;
+ const xmlChar *prefix;
+ const xmlChar *attname;
+ const xmlChar *aprefix;
+ const xmlChar *nsname;
+ xmlChar *attvalue;
+ const xmlChar **atts = ctxt->atts;
+ int maxatts = ctxt->maxatts;
+ int nratts, nbatts, nbdef;
+ int i, j, nbNs, attval, oldline, oldcol, inputNr;
+ const xmlChar *base;
+ unsigned long cur;
+ int nsNr = ctxt->nsNr;
+
+ if (RAW != '<') return(NULL);
+ NEXT1;
+
+ /*
+ * NOTE: it is crucial with the SAX2 API to never call SHRINK beyond that
+ * point since the attribute values may be stored as pointers to
+ * the buffer and calling SHRINK would destroy them !
+ * The Shrinking is only possible once the full set of attribute
+ * callbacks have been done.
+ */
+reparse:
+ SHRINK;
+ base = ctxt->input->base;
+ cur = ctxt->input->cur - ctxt->input->base;
+ inputNr = ctxt->inputNr;
+ oldline = ctxt->input->line;
+ oldcol = ctxt->input->col;
+ nbatts = 0;
+ nratts = 0;
+ nbdef = 0;
+ nbNs = 0;
+ attval = 0;
+ /* Forget any namespaces added during an earlier parse of this element. */
+ ctxt->nsNr = nsNr;
+
+ localname = xmlParseQName(ctxt, &prefix);
+ if (localname == NULL) {
+ xmlFatalErrMsg(ctxt, XML_ERR_NAME_REQUIRED,
+ "StartTag: invalid element name\n");
+ return(NULL);
+ }
+ *tlen = ctxt->input->cur - ctxt->input->base - cur;
+
+ /*
+ * Now parse the attributes, it ends up with the ending
+ *
+ * (S Attribute)* S?
+ */
+ SKIP_BLANKS;
+ GROW;
+ if ((ctxt->input->base != base) || (inputNr != ctxt->inputNr))
+ goto base_changed;
+
+ while (((RAW != '>') &&
+ ((RAW != '/') || (NXT(1) != '>')) &&
+ (IS_BYTE_CHAR(RAW))) && (ctxt->instate != XML_PARSER_EOF)) {
+ const xmlChar *q = CUR_PTR;
+ unsigned int cons = ctxt->input->consumed;
+ int len = -1, alloc = 0;
+
+ attname = xmlParseAttribute2(ctxt, prefix, localname,
+ &aprefix, &attvalue, &len, &alloc);
+ if ((ctxt->input->base != base) || (inputNr != ctxt->inputNr)) {
+ if ((attvalue != NULL) && (alloc != 0))
+ xmlFree(attvalue);
+ attvalue = NULL;
+ goto base_changed;
+ }
+ if ((attname != NULL) && (attvalue != NULL)) {
+ if (len < 0) len = xmlStrlen(attvalue);
+ if ((attname == ctxt->str_xmlns) && (aprefix == NULL)) {
+ const xmlChar *URL = xmlDictLookup(ctxt->dict, attvalue, len);
+ xmlURIPtr uri;
+
+ if (URL == NULL) {
+ xmlErrMemory(ctxt, "dictionary allocation failure");
+ if ((attvalue != NULL) && (alloc != 0))
+ xmlFree(attvalue);
+ return(NULL);
+ }
+ if (*URL != 0) {
+ uri = xmlParseURI((const char *) URL);
+ if (uri == NULL) {
+ xmlNsErr(ctxt, XML_WAR_NS_URI,
+ "xmlns: '%s' is not a valid URI\n",
+ URL, NULL, NULL);
+ } else {
+ if (uri->scheme == NULL) {
+ xmlNsWarn(ctxt, XML_WAR_NS_URI_RELATIVE,
+ "xmlns: URI %s is not absolute\n",
+ URL, NULL, NULL);
+ }
+ xmlFreeURI(uri);
+ }
+ if (URL == ctxt->str_xml_ns) {
+ if (attname != ctxt->str_xml) {
+ xmlNsErr(ctxt, XML_NS_ERR_XML_NAMESPACE,
+ "xml namespace URI cannot be the default namespace\n",
+ NULL, NULL, NULL);
+ }
+ goto skip_default_ns;
+ }
+ if ((len == 29) &&
+ (xmlStrEqual(URL,
+ BAD_CAST "http://www.w3.org/2000/xmlns/"))) {
+ xmlNsErr(ctxt, XML_NS_ERR_XML_NAMESPACE,
+ "reuse of the xmlns namespace name is forbidden\n",
+ NULL, NULL, NULL);
+ goto skip_default_ns;
+ }
+ }
+ /*
+ * check that it's not a defined namespace
+ */
+ for (j = 1;j <= nbNs;j++)
+ if (ctxt->nsTab[ctxt->nsNr - 2 * j] == NULL)
+ break;
+ if (j <= nbNs)
+ xmlErrAttributeDup(ctxt, NULL, attname);
+ else
+ if (nsPush(ctxt, NULL, URL) > 0) nbNs++;
+skip_default_ns:
+ if (alloc != 0) xmlFree(attvalue);
+ if ((RAW == '>') || (((RAW == '/') && (NXT(1) == '>'))))
+ break;
+ if (!IS_BLANK_CH(RAW)) {
+ xmlFatalErrMsg(ctxt, XML_ERR_SPACE_REQUIRED,
+ "attributes construct error\n");
+ break;
+ }
+ SKIP_BLANKS;
+ continue;
+ }
+ if (aprefix == ctxt->str_xmlns) {
+ const xmlChar *URL = xmlDictLookup(ctxt->dict, attvalue, len);
+ xmlURIPtr uri;
+
+ if (attname == ctxt->str_xml) {
+ if (URL != ctxt->str_xml_ns) {
+ xmlNsErr(ctxt, XML_NS_ERR_XML_NAMESPACE,
+ "xml namespace prefix mapped to wrong URI\n",
+ NULL, NULL, NULL);
+ }
+ /*
+ * Do not keep a namespace definition node
+ */
+ goto skip_ns;
+ }
+ if (URL == ctxt->str_xml_ns) {
+ if (attname != ctxt->str_xml) {
+ xmlNsErr(ctxt, XML_NS_ERR_XML_NAMESPACE,
+ "xml namespace URI mapped to wrong prefix\n",
+ NULL, NULL, NULL);
+ }
+ goto skip_ns;
+ }
+ if (attname == ctxt->str_xmlns) {
+ xmlNsErr(ctxt, XML_NS_ERR_XML_NAMESPACE,
+ "redefinition of the xmlns prefix is forbidden\n",
+ NULL, NULL, NULL);
+ goto skip_ns;
+ }
+ if ((len == 29) &&
+ (xmlStrEqual(URL,
+ BAD_CAST "http://www.w3.org/2000/xmlns/"))) {
+ xmlNsErr(ctxt, XML_NS_ERR_XML_NAMESPACE,
+ "reuse of the xmlns namespace name is forbidden\n",
+ NULL, NULL, NULL);
+ goto skip_ns;
+ }
+ if ((URL == NULL) || (URL[0] == 0)) {
+ xmlNsErr(ctxt, XML_NS_ERR_XML_NAMESPACE,
+ "xmlns:%s: Empty XML namespace is not allowed\n",
+ attname, NULL, NULL);
+ goto skip_ns;
+ } else {
+ uri = xmlParseURI((const char *) URL);
+ if (uri == NULL) {
+ xmlNsErr(ctxt, XML_WAR_NS_URI,
+ "xmlns:%s: '%s' is not a valid URI\n",
+ attname, URL, NULL);
+ } else {
+ if ((ctxt->pedantic) && (uri->scheme == NULL)) {
+ xmlNsWarn(ctxt, XML_WAR_NS_URI_RELATIVE,
+ "xmlns:%s: URI %s is not absolute\n",
+ attname, URL, NULL);
+ }
+ xmlFreeURI(uri);
+ }
+ }
+
+ /*
+ * check that it's not a defined namespace
+ */
+ for (j = 1;j <= nbNs;j++)
+ if (ctxt->nsTab[ctxt->nsNr - 2 * j] == attname)
+ break;
+ if (j <= nbNs)
+ xmlErrAttributeDup(ctxt, aprefix, attname);
+ else
+ if (nsPush(ctxt, attname, URL) > 0) nbNs++;
+skip_ns:
+ if (alloc != 0) xmlFree(attvalue);
+ if ((RAW == '>') || (((RAW == '/') && (NXT(1) == '>'))))
+ break;
+ if (!IS_BLANK_CH(RAW)) {
+ xmlFatalErrMsg(ctxt, XML_ERR_SPACE_REQUIRED,
+ "attributes construct error\n");
+ break;
+ }
+ SKIP_BLANKS;
+ if ((ctxt->input->base != base) || (inputNr != ctxt->inputNr))
+ goto base_changed;
+ continue;
+ }
+
+ /*
+ * Add the pair to atts
+ */
+ if ((atts == NULL) || (nbatts + 5 > maxatts)) {
+ if (xmlCtxtGrowAttrs(ctxt, nbatts + 5) < 0) {
+ if (attvalue[len] == 0)
+ xmlFree(attvalue);
+ goto failed;
+ }
+ maxatts = ctxt->maxatts;
+ atts = ctxt->atts;
+ }
+ ctxt->attallocs[nratts++] = alloc;
+ atts[nbatts++] = attname;
+ atts[nbatts++] = aprefix;
+ atts[nbatts++] = NULL; /* the URI will be fetched later */
+ atts[nbatts++] = attvalue;
+ attvalue += len;
+ atts[nbatts++] = attvalue;
+ /*
+ * tag if some deallocation is needed
+ */
+ if (alloc != 0) attval = 1;
+ } else {
+ if ((attvalue != NULL) && (attvalue[len] == 0))
+ xmlFree(attvalue);
+ }
+
+failed:
+
+ GROW
+ if (ctxt->instate == XML_PARSER_EOF)
+ break;
+ if ((ctxt->input->base != base) || (inputNr != ctxt->inputNr))
+ goto base_changed;
+ if ((RAW == '>') || (((RAW == '/') && (NXT(1) == '>'))))
+ break;
+ if (!IS_BLANK_CH(RAW)) {
+ xmlFatalErrMsg(ctxt, XML_ERR_SPACE_REQUIRED,
+ "attributes construct error\n");
+ break;
+ }
+ SKIP_BLANKS;
+ if ((cons == ctxt->input->consumed) && (q == CUR_PTR) &&
+ (attname == NULL) && (attvalue == NULL)) {
+ xmlFatalErr(ctxt, XML_ERR_INTERNAL_ERROR,
+ "xmlParseStartTag: problem parsing attributes\n");
+ break;
+ }
+ GROW;
+ if ((ctxt->input->base != base) || (inputNr != ctxt->inputNr))
+ goto base_changed;
+ }
+
+ /*
+ * The attributes defaulting
+ */
+ if (ctxt->attsDefault != NULL) {
+ xmlDefAttrsPtr defaults;
+
+ defaults = xmlHashLookup2(ctxt->attsDefault, localname, prefix);
+ if (defaults != NULL) {
+ for (i = 0;i < defaults->nbAttrs;i++) {
+ attname = defaults->values[5 * i];
+ aprefix = defaults->values[5 * i + 1];
+
+ /*
+ * special work for namespaces defaulted defs
+ */
+ if ((attname == ctxt->str_xmlns) && (aprefix == NULL)) {
+ /*
+ * check that it's not a defined namespace
+ */
+ for (j = 1;j <= nbNs;j++)
+ if (ctxt->nsTab[ctxt->nsNr - 2 * j] == NULL)
+ break;
+ if (j <= nbNs) continue;
+
+ nsname = xmlGetNamespace(ctxt, NULL);
+ if (nsname != defaults->values[5 * i + 2]) {
+ if (nsPush(ctxt, NULL,
+ defaults->values[5 * i + 2]) > 0)
+ nbNs++;
+ }
+ } else if (aprefix == ctxt->str_xmlns) {
+ /*
+ * check that it's not a defined namespace
+ */
+ for (j = 1;j <= nbNs;j++)
+ if (ctxt->nsTab[ctxt->nsNr - 2 * j] == attname)
+ break;
+ if (j <= nbNs) continue;
+
+ nsname = xmlGetNamespace(ctxt, attname);
+ if (nsname != defaults->values[2]) {
+ if (nsPush(ctxt, attname,
+ defaults->values[5 * i + 2]) > 0)
+ nbNs++;
+ }
+ } else {
+ /*
+ * check that it's not a defined attribute
+ */
+ for (j = 0;j < nbatts;j+=5) {
+ if ((attname == atts[j]) && (aprefix == atts[j+1]))
+ break;
+ }
+ if (j < nbatts) continue;
+
+ if ((atts == NULL) || (nbatts + 5 > maxatts)) {
+ if (xmlCtxtGrowAttrs(ctxt, nbatts + 5) < 0) {
+ return(NULL);
+ }
+ maxatts = ctxt->maxatts;
+ atts = ctxt->atts;
+ }
+ atts[nbatts++] = attname;
+ atts[nbatts++] = aprefix;
+ if (aprefix == NULL)
+ atts[nbatts++] = NULL;
+ else
+ atts[nbatts++] = xmlGetNamespace(ctxt, aprefix);
+ atts[nbatts++] = defaults->values[5 * i + 2];
+ atts[nbatts++] = defaults->values[5 * i + 3];
+ if ((ctxt->standalone == 1) &&
+ (defaults->values[5 * i + 4] != NULL)) {
+ xmlValidityError(ctxt, XML_DTD_STANDALONE_DEFAULTED,
+ "standalone: attribute %s on %s defaulted from external subset\n",
+ attname, localname);
+ }
+ nbdef++;
+ }
+ }
+ }
+ }
+
+ /*
+ * The attributes checkings
+ */
+ for (i = 0; i < nbatts;i += 5) {
+ /*
+ * The default namespace does not apply to attribute names.
+ */
+ if (atts[i + 1] != NULL) {
+ nsname = xmlGetNamespace(ctxt, atts[i + 1]);
+ if (nsname == NULL) {
+ xmlNsErr(ctxt, XML_NS_ERR_UNDEFINED_NAMESPACE,
+ "Namespace prefix %s for %s on %s is not defined\n",
+ atts[i + 1], atts[i], localname);
+ }
+ atts[i + 2] = nsname;
+ } else
+ nsname = NULL;
+ /*
+ * [ WFC: Unique Att Spec ]
+ * No attribute name may appear more than once in the same
+ * start-tag or empty-element tag.
+ * As extended by the Namespace in XML REC.
+ */
+ for (j = 0; j < i;j += 5) {
+ if (atts[i] == atts[j]) {
+ if (atts[i+1] == atts[j+1]) {
+ xmlErrAttributeDup(ctxt, atts[i+1], atts[i]);
+ break;
+ }
+ if ((nsname != NULL) && (atts[j + 2] == nsname)) {
+ xmlNsErr(ctxt, XML_NS_ERR_ATTRIBUTE_REDEFINED,
+ "Namespaced Attribute %s in '%s' redefined\n",
+ atts[i], nsname, NULL);
+ break;
+ }
+ }
+ }
+ }
+
+ nsname = xmlGetNamespace(ctxt, prefix);
+ if ((prefix != NULL) && (nsname == NULL)) {
+ xmlNsErr(ctxt, XML_NS_ERR_UNDEFINED_NAMESPACE,
+ "Namespace prefix %s on %s is not defined\n",
+ prefix, localname, NULL);
+ }
+ *pref = prefix;
+ *URI = nsname;
+
+ /*
+ * SAX: Start of Element !
+ */
+ if ((ctxt->sax != NULL) && (ctxt->sax->startElementNs != NULL) &&
+ (!ctxt->disableSAX)) {
+ if (nbNs > 0)
+ ctxt->sax->startElementNs(ctxt->userData, localname, prefix,
+ nsname, nbNs, &ctxt->nsTab[ctxt->nsNr - 2 * nbNs],
+ nbatts / 5, nbdef, atts);
+ else
+ ctxt->sax->startElementNs(ctxt->userData, localname, prefix,
+ nsname, 0, NULL, nbatts / 5, nbdef, atts);
+ }
+
+ /*
+ * Free up attribute allocated strings if needed
+ */
+ if (attval != 0) {
+ for (i = 3,j = 0; j < nratts;i += 5,j++)
+ if ((ctxt->attallocs[j] != 0) && (atts[i] != NULL))
+ xmlFree((xmlChar *) atts[i]);
+ }
+
+ return(localname);
+
+base_changed:
+ /*
+ * the attribute strings are valid iif the base didn't changed
+ */
+ if (attval != 0) {
+ for (i = 3,j = 0; j < nratts;i += 5,j++)
+ if ((ctxt->attallocs[j] != 0) && (atts[i] != NULL))
+ xmlFree((xmlChar *) atts[i]);
+ }
+
+ /*
+ * We can't switch from one entity to another in the middle
+ * of a start tag
+ */
+ if (inputNr != ctxt->inputNr) {
+ xmlFatalErrMsg(ctxt, XML_ERR_ENTITY_BOUNDARY,
+ "Start tag doesn't start and stop in the same entity\n");
+ return(NULL);
+ }
+
+ ctxt->input->cur = ctxt->input->base + cur;
+ ctxt->input->line = oldline;
+ ctxt->input->col = oldcol;
+ if (ctxt->wellFormed == 1) {
+ goto reparse;
+ }
+ return(NULL);
+}
+
+/**
+ * xmlParseEndTag2:
+ * @ctxt: an XML parser context
+ * @line: line of the start tag
+ * @nsNr: number of namespaces on the start tag
+ *
+ * parse an end of tag
+ *
+ * [42] ETag ::= '</' Name S? '>'
+ *
+ * With namespace
+ *
+ * [NS 9] ETag ::= '</' QName S? '>'
+ */
+
+static void
+xmlParseEndTag2(xmlParserCtxtPtr ctxt, const xmlChar *prefix,
+ const xmlChar *URI, int line, int nsNr, int tlen) {
+ const xmlChar *name;
+ size_t curLength;
+
+ GROW;
+ if ((RAW != '<') || (NXT(1) != '/')) {
+ xmlFatalErr(ctxt, XML_ERR_LTSLASH_REQUIRED, NULL);
+ return;
+ }
+ SKIP(2);
+
+ curLength = ctxt->input->end - ctxt->input->cur;
+ if ((tlen > 0) && (curLength >= (size_t)tlen) &&
+ (xmlStrncmp(ctxt->input->cur, ctxt->name, tlen) == 0)) {
+ if ((curLength >= (size_t)(tlen + 1)) &&
+ (ctxt->input->cur[tlen] == '>')) {
+ ctxt->input->cur += tlen + 1;
+ ctxt->input->col += tlen + 1;
+ goto done;
+ }
+ ctxt->input->cur += tlen;
+ ctxt->input->col += tlen;
+ name = (xmlChar*)1;
+ } else {
+ if (prefix == NULL)
+ name = xmlParseNameAndCompare(ctxt, ctxt->name);
+ else
+ name = xmlParseQNameAndCompare(ctxt, ctxt->name, prefix);
+ }
+
+ /*
+ * We should definitely be at the ending "S? '>'" part
+ */
+ GROW;
+ if (ctxt->instate == XML_PARSER_EOF)
+ return;
+ SKIP_BLANKS;
+ if ((!IS_BYTE_CHAR(RAW)) || (RAW != '>')) {
+ xmlFatalErr(ctxt, XML_ERR_GT_REQUIRED, NULL);
+ } else
+ NEXT1;
+
+ /*
+ * [ WFC: Element Type Match ]
+ * The Name in an element's end-tag must match the element type in the
+ * start-tag.
+ *
+ */
+ if (name != (xmlChar*)1) {
+ if (name == NULL) name = BAD_CAST "unparseable";
+ if ((line == 0) && (ctxt->node != NULL))
+ line = ctxt->node->line;
+ xmlFatalErrMsgStrIntStr(ctxt, XML_ERR_TAG_NAME_MISMATCH,
+ "Opening and ending tag mismatch: %s line %d and %s\n",
+ ctxt->name, line, name);
+ }
+
+ /*
+ * SAX: End of Tag
+ */
+done:
+ if ((ctxt->sax != NULL) && (ctxt->sax->endElementNs != NULL) &&
+ (!ctxt->disableSAX))
+ ctxt->sax->endElementNs(ctxt->userData, ctxt->name, prefix, URI);
+
+ spacePop(ctxt);
+ if (nsNr != 0)
+ nsPop(ctxt, nsNr);
+ return;
+}
+
+/**
+ * xmlParseCDSect:
+ * @ctxt: an XML parser context
+ *
+ * Parse escaped pure raw content.
+ *
+ * [18] CDSect ::= CDStart CData CDEnd
+ *
+ * [19] CDStart ::= '<![CDATA['
+ *
+ * [20] Data ::= (Char* - (Char* ']]>' Char*))
+ *
+ * [21] CDEnd ::= ']]>'
+ */
+void
+xmlParseCDSect(xmlParserCtxtPtr ctxt) {
+ xmlChar *buf = NULL;
+ int len = 0;
+ int size = XML_PARSER_BUFFER_SIZE;
+ int r, rl;
+ int s, sl;
+ int cur, l;
+ int count = 0;
+
+ /* Check 2.6.0 was NXT(0) not RAW */
+ if (CMP9(CUR_PTR, '<', '!', '[', 'C', 'D', 'A', 'T', 'A', '[')) {
+ SKIP(9);
+ } else
+ return;
+
+ ctxt->instate = XML_PARSER_CDATA_SECTION;
+ r = CUR_CHAR(rl);
+ if (!IS_CHAR(r)) {
+ xmlFatalErr(ctxt, XML_ERR_CDATA_NOT_FINISHED, NULL);
+ ctxt->instate = XML_PARSER_CONTENT;
+ return;
+ }
+ NEXTL(rl);
+ s = CUR_CHAR(sl);
+ if (!IS_CHAR(s)) {
+ xmlFatalErr(ctxt, XML_ERR_CDATA_NOT_FINISHED, NULL);
+ ctxt->instate = XML_PARSER_CONTENT;
+ return;
+ }
+ NEXTL(sl);
+ cur = CUR_CHAR(l);
+ buf = (xmlChar *) xmlMallocAtomic(size * sizeof(xmlChar));
+ if (buf == NULL) {
+ xmlErrMemory(ctxt, NULL);
+ return;
+ }
+ while (IS_CHAR(cur) &&
+ ((r != ']') || (s != ']') || (cur != '>'))) {
+ if (len + 5 >= size) {
+ xmlChar *tmp;
+
+ if ((size > XML_MAX_TEXT_LENGTH) &&
+ ((ctxt->options & XML_PARSE_HUGE) == 0)) {
+ xmlFatalErrMsgStr(ctxt, XML_ERR_CDATA_NOT_FINISHED,
+ "CData section too big found", NULL);
+ xmlFree (buf);
+ return;
+ }
+ tmp = (xmlChar *) xmlRealloc(buf, size * 2 * sizeof(xmlChar));
+ if (tmp == NULL) {
+ xmlFree(buf);
+ xmlErrMemory(ctxt, NULL);
+ return;
+ }
+ buf = tmp;
+ size *= 2;
+ }
+ COPY_BUF(rl,buf,len,r);
+ r = s;
+ rl = sl;
+ s = cur;
+ sl = l;
+ count++;
+ if (count > 50) {
+ GROW;
+ if (ctxt->instate == XML_PARSER_EOF) {
+ xmlFree(buf);
+ return;
+ }
+ count = 0;
+ }
+ NEXTL(l);
+ cur = CUR_CHAR(l);
+ }
+ buf[len] = 0;
+ ctxt->instate = XML_PARSER_CONTENT;
+ if (cur != '>') {
+ xmlFatalErrMsgStr(ctxt, XML_ERR_CDATA_NOT_FINISHED,
+ "CData section not finished\n%.50s\n", buf);
+ xmlFree(buf);
+ return;
+ }
+ NEXTL(l);
+
+ /*
+ * OK the buffer is to be consumed as cdata.
+ */
+ if ((ctxt->sax != NULL) && (!ctxt->disableSAX)) {
+ if (ctxt->sax->cdataBlock != NULL)
+ ctxt->sax->cdataBlock(ctxt->userData, buf, len);
+ else if (ctxt->sax->characters != NULL)
+ ctxt->sax->characters(ctxt->userData, buf, len);
+ }
+ xmlFree(buf);
+}
+
+/**
+ * xmlParseContent:
+ * @ctxt: an XML parser context
+ *
+ * Parse a content:
+ *
+ * [43] content ::= (element | CharData | Reference | CDSect | PI | Comment)*
+ */
+
+void
+xmlParseContent(xmlParserCtxtPtr ctxt) {
+ GROW;
+ while ((RAW != 0) &&
+ ((RAW != '<') || (NXT(1) != '/')) &&
+ (ctxt->instate != XML_PARSER_EOF)) {
+ const xmlChar *test = CUR_PTR;
+ unsigned int cons = ctxt->input->consumed;
+ const xmlChar *cur = ctxt->input->cur;
+
+ /*
+ * First case : a Processing Instruction.
+ */
+ if ((*cur == '<') && (cur[1] == '?')) {
+ xmlParsePI(ctxt);
+ }
+
+ /*
+ * Second case : a CDSection
+ */
+ /* 2.6.0 test was *cur not RAW */
+ else if (CMP9(CUR_PTR, '<', '!', '[', 'C', 'D', 'A', 'T', 'A', '[')) {
+ xmlParseCDSect(ctxt);
+ }
+
+ /*
+ * Third case : a comment
+ */
+ else if ((*cur == '<') && (NXT(1) == '!') &&
+ (NXT(2) == '-') && (NXT(3) == '-')) {
+ xmlParseComment(ctxt);
+ ctxt->instate = XML_PARSER_CONTENT;
+ }
+
+ /*
+ * Fourth case : a sub-element.
+ */
+ else if (*cur == '<') {
+ xmlParseElement(ctxt);
+ }
+
+ /*
+ * Fifth case : a reference. If if has not been resolved,
+ * parsing returns it's Name, create the node
+ */
+
+ else if (*cur == '&') {
+ xmlParseReference(ctxt);
+ }
+
+ /*
+ * Last case, text. Note that References are handled directly.
+ */
+ else {
+ xmlParseCharData(ctxt, 0);
+ }
+
+ GROW;
+ /*
+ * Pop-up of finished entities.
+ */
+ while ((RAW == 0) && (ctxt->inputNr > 1))
+ xmlPopInput(ctxt);
+ SHRINK;
+
+ if ((cons == ctxt->input->consumed) && (test == CUR_PTR)) {
+ xmlFatalErr(ctxt, XML_ERR_INTERNAL_ERROR,
+ "detected an error in element content\n");
+ xmlHaltParser(ctxt);
+ break;
+ }
+ }
+}
+
+/**
+ * xmlParseElement:
+ * @ctxt: an XML parser context
+ *
+ * parse an XML element, this is highly recursive
+ *
+ * [39] element ::= EmptyElemTag | STag content ETag
+ *
+ * [ WFC: Element Type Match ]
+ * The Name in an element's end-tag must match the element type in the
+ * start-tag.
+ *
+ */
+
+void
+xmlParseElement(xmlParserCtxtPtr ctxt) {
+ const xmlChar *name;
+ const xmlChar *prefix = NULL;
+ const xmlChar *URI = NULL;
+ xmlParserNodeInfo node_info;
+ int line, tlen = 0;
+ xmlNodePtr ret;
+ int nsNr = ctxt->nsNr;
+
+ if (((unsigned int) ctxt->nameNr > xmlParserMaxDepth) &&
+ ((ctxt->options & XML_PARSE_HUGE) == 0)) {
+ xmlFatalErrMsgInt(ctxt, XML_ERR_INTERNAL_ERROR,
+ "Excessive depth in document: %d use XML_PARSE_HUGE option\n",
+ xmlParserMaxDepth);
+ xmlHaltParser(ctxt);
+ return;
+ }
+
+ /* Capture start position */
+ if (ctxt->record_info) {
+ node_info.begin_pos = ctxt->input->consumed +
+ (CUR_PTR - ctxt->input->base);
+ node_info.begin_line = ctxt->input->line;
+ }
+
+ if (ctxt->spaceNr == 0)
+ spacePush(ctxt, -1);
+ else if (*ctxt->space == -2)
+ spacePush(ctxt, -1);
+ else
+ spacePush(ctxt, *ctxt->space);
+
+ line = ctxt->input->line;
+#ifdef LIBXML_SAX1_ENABLED
+ if (ctxt->sax2)
+#endif /* LIBXML_SAX1_ENABLED */
+ name = xmlParseStartTag2(ctxt, &prefix, &URI, &tlen);
+#ifdef LIBXML_SAX1_ENABLED
+ else
+ name = xmlParseStartTag(ctxt);
+#endif /* LIBXML_SAX1_ENABLED */
+ if (ctxt->instate == XML_PARSER_EOF)
+ return;
+ if (name == NULL) {
+ spacePop(ctxt);
+ return;
+ }
+ namePush(ctxt, name);
+ ret = ctxt->node;
+
+#ifdef LIBXML_VALID_ENABLED
+ /*
+ * [ VC: Root Element Type ]
+ * The Name in the document type declaration must match the element
+ * type of the root element.
+ */
+ if (ctxt->validate && ctxt->wellFormed && ctxt->myDoc &&
+ ctxt->node && (ctxt->node == ctxt->myDoc->children))
+ ctxt->valid &= xmlValidateRoot(&ctxt->vctxt, ctxt->myDoc);
+#endif /* LIBXML_VALID_ENABLED */
+
+ /*
+ * Check for an Empty Element.
+ */
+ if ((RAW == '/') && (NXT(1) == '>')) {
+ SKIP(2);
+ if (ctxt->sax2) {
+ if ((ctxt->sax != NULL) && (ctxt->sax->endElementNs != NULL) &&
+ (!ctxt->disableSAX))
+ ctxt->sax->endElementNs(ctxt->userData, name, prefix, URI);
+#ifdef LIBXML_SAX1_ENABLED
+ } else {
+ if ((ctxt->sax != NULL) && (ctxt->sax->endElement != NULL) &&
+ (!ctxt->disableSAX))
+ ctxt->sax->endElement(ctxt->userData, name);
+#endif /* LIBXML_SAX1_ENABLED */
+ }
+ namePop(ctxt);
+ spacePop(ctxt);
+ if (nsNr != ctxt->nsNr)
+ nsPop(ctxt, ctxt->nsNr - nsNr);
+ if ( ret != NULL && ctxt->record_info ) {
+ node_info.end_pos = ctxt->input->consumed +
+ (CUR_PTR - ctxt->input->base);
+ node_info.end_line = ctxt->input->line;
+ node_info.node = ret;
+ xmlParserAddNodeInfo(ctxt, &node_info);
+ }
+ return;
+ }
+ if (RAW == '>') {
+ NEXT1;
+ } else {
+ xmlFatalErrMsgStrIntStr(ctxt, XML_ERR_GT_REQUIRED,
+ "Couldn't find end of Start Tag %s line %d\n",
+ name, line, NULL);
+
+ /*
+ * end of parsing of this node.
+ */
+ nodePop(ctxt);
+ namePop(ctxt);
+ spacePop(ctxt);
+ if (nsNr != ctxt->nsNr)
+ nsPop(ctxt, ctxt->nsNr - nsNr);
+
+ /*
+ * Capture end position and add node
+ */
+ if ( ret != NULL && ctxt->record_info ) {
+ node_info.end_pos = ctxt->input->consumed +
+ (CUR_PTR - ctxt->input->base);
+ node_info.end_line = ctxt->input->line;
+ node_info.node = ret;
+ xmlParserAddNodeInfo(ctxt, &node_info);
+ }
+ return;
+ }
+
+ /*
+ * Parse the content of the element:
+ */
+ xmlParseContent(ctxt);
+ if (ctxt->instate == XML_PARSER_EOF)
+ return;
+ if (!IS_BYTE_CHAR(RAW)) {
+ xmlFatalErrMsgStrIntStr(ctxt, XML_ERR_TAG_NOT_FINISHED,
+ "Premature end of data in tag %s line %d\n",
+ name, line, NULL);
+
+ /*
+ * end of parsing of this node.
+ */
+ nodePop(ctxt);
+ namePop(ctxt);
+ spacePop(ctxt);
+ if (nsNr != ctxt->nsNr)
+ nsPop(ctxt, ctxt->nsNr - nsNr);
+ return;
+ }
+
+ /*
+ * parse the end of tag: '</' should be here.
+ */
+ if (ctxt->sax2) {
+ xmlParseEndTag2(ctxt, prefix, URI, line, ctxt->nsNr - nsNr, tlen);
+ namePop(ctxt);
+ }
+#ifdef LIBXML_SAX1_ENABLED
+ else
+ xmlParseEndTag1(ctxt, line);
+#endif /* LIBXML_SAX1_ENABLED */
+
+ /*
+ * Capture end position and add node
+ */
+ if ( ret != NULL && ctxt->record_info ) {
+ node_info.end_pos = ctxt->input->consumed +
+ (CUR_PTR - ctxt->input->base);
+ node_info.end_line = ctxt->input->line;
+ node_info.node = ret;
+ xmlParserAddNodeInfo(ctxt, &node_info);
+ }
+}
+
+/**
+ * xmlParseVersionNum:
+ * @ctxt: an XML parser context
+ *
+ * parse the XML version value.
+ *
+ * [26] VersionNum ::= '1.' [0-9]+
+ *
+ * In practice allow [0-9].[0-9]+ at that level
+ *
+ * Returns the string giving the XML version number, or NULL
+ */
+xmlChar *
+xmlParseVersionNum(xmlParserCtxtPtr ctxt) {
+ xmlChar *buf = NULL;
+ int len = 0;
+ int size = 10;
+ xmlChar cur;
+
+ buf = (xmlChar *) xmlMallocAtomic(size * sizeof(xmlChar));
+ if (buf == NULL) {
+ xmlErrMemory(ctxt, NULL);
+ return(NULL);
+ }
+ cur = CUR;
+ if (!((cur >= '0') && (cur <= '9'))) {
+ xmlFree(buf);
+ return(NULL);
+ }
+ buf[len++] = cur;
+ NEXT;
+ cur=CUR;
+ if (cur != '.') {
+ xmlFree(buf);
+ return(NULL);
+ }
+ buf[len++] = cur;
+ NEXT;
+ cur=CUR;
+ while ((cur >= '0') && (cur <= '9')) {
+ if (len + 1 >= size) {
+ xmlChar *tmp;
+
+ size *= 2;
+ tmp = (xmlChar *) xmlRealloc(buf, size * sizeof(xmlChar));
+ if (tmp == NULL) {
+ xmlFree(buf);
+ xmlErrMemory(ctxt, NULL);
+ return(NULL);
+ }
+ buf = tmp;
+ }
+ buf[len++] = cur;
+ NEXT;
+ cur=CUR;
+ }
+ buf[len] = 0;
+ return(buf);
+}
+
+/**
+ * xmlParseVersionInfo:
+ * @ctxt: an XML parser context
+ *
+ * parse the XML version.
+ *
+ * [24] VersionInfo ::= S 'version' Eq (' VersionNum ' | " VersionNum ")
+ *
+ * [25] Eq ::= S? '=' S?
+ *
+ * Returns the version string, e.g. "1.0"
+ */
+
+xmlChar *
+xmlParseVersionInfo(xmlParserCtxtPtr ctxt) {
+ xmlChar *version = NULL;
+
+ if (CMP7(CUR_PTR, 'v', 'e', 'r', 's', 'i', 'o', 'n')) {
+ SKIP(7);
+ SKIP_BLANKS;
+ if (RAW != '=') {
+ xmlFatalErr(ctxt, XML_ERR_EQUAL_REQUIRED, NULL);
+ return(NULL);
+ }
+ NEXT;
+ SKIP_BLANKS;
+ if (RAW == '"') {
+ NEXT;
+ version = xmlParseVersionNum(ctxt);
+ if (RAW != '"') {
+ xmlFatalErr(ctxt, XML_ERR_STRING_NOT_CLOSED, NULL);
+ } else
+ NEXT;
+ } else if (RAW == '\''){
+ NEXT;
+ version = xmlParseVersionNum(ctxt);
+ if (RAW != '\'') {
+ xmlFatalErr(ctxt, XML_ERR_STRING_NOT_CLOSED, NULL);
+ } else
+ NEXT;
+ } else {
+ xmlFatalErr(ctxt, XML_ERR_STRING_NOT_STARTED, NULL);
+ }
+ }
+ return(version);
+}
+
+/**
+ * xmlParseEncName:
+ * @ctxt: an XML parser context
+ *
+ * parse the XML encoding name
+ *
+ * [81] EncName ::= [A-Za-z] ([A-Za-z0-9._] | '-')*
+ *
+ * Returns the encoding name value or NULL
+ */
+xmlChar *
+xmlParseEncName(xmlParserCtxtPtr ctxt) {
+ xmlChar *buf = NULL;
+ int len = 0;
+ int size = 10;
+ xmlChar cur;
+
+ cur = CUR;
+ if (((cur >= 'a') && (cur <= 'z')) ||
+ ((cur >= 'A') && (cur <= 'Z'))) {
+ buf = (xmlChar *) xmlMallocAtomic(size * sizeof(xmlChar));
+ if (buf == NULL) {
+ xmlErrMemory(ctxt, NULL);
+ return(NULL);
+ }
+
+ buf[len++] = cur;
+ NEXT;
+ cur = CUR;
+ while (((cur >= 'a') && (cur <= 'z')) ||
+ ((cur >= 'A') && (cur <= 'Z')) ||
+ ((cur >= '0') && (cur <= '9')) ||
+ (cur == '.') || (cur == '_') ||
+ (cur == '-')) {
+ if (len + 1 >= size) {
+ xmlChar *tmp;
+
+ size *= 2;
+ tmp = (xmlChar *) xmlRealloc(buf, size * sizeof(xmlChar));
+ if (tmp == NULL) {
+ xmlErrMemory(ctxt, NULL);
+ xmlFree(buf);
+ return(NULL);
+ }
+ buf = tmp;
+ }
+ buf[len++] = cur;
+ NEXT;
+ cur = CUR;
+ if (cur == 0) {
+ SHRINK;
+ GROW;
+ cur = CUR;
+ }
+ }
+ buf[len] = 0;
+ } else {
+ xmlFatalErr(ctxt, XML_ERR_ENCODING_NAME, NULL);
+ }
+ return(buf);
+}
+
+/**
+ * xmlParseEncodingDecl:
+ * @ctxt: an XML parser context
+ *
+ * parse the XML encoding declaration
+ *
+ * [80] EncodingDecl ::= S 'encoding' Eq ('"' EncName '"' | "'" EncName "'")
+ *
+ * this setups the conversion filters.
+ *
+ * Returns the encoding value or NULL
+ */
+
+const xmlChar *
+xmlParseEncodingDecl(xmlParserCtxtPtr ctxt) {
+ xmlChar *encoding = NULL;
+
+ SKIP_BLANKS;
+ if (CMP8(CUR_PTR, 'e', 'n', 'c', 'o', 'd', 'i', 'n', 'g')) {
+ SKIP(8);
+ SKIP_BLANKS;
+ if (RAW != '=') {
+ xmlFatalErr(ctxt, XML_ERR_EQUAL_REQUIRED, NULL);
+ return(NULL);
+ }
+ NEXT;
+ SKIP_BLANKS;
+ if (RAW == '"') {
+ NEXT;
+ encoding = xmlParseEncName(ctxt);
+ if (RAW != '"') {
+ xmlFatalErr(ctxt, XML_ERR_STRING_NOT_CLOSED, NULL);
+ xmlFree((xmlChar *) encoding);
+ return(NULL);
+ } else
+ NEXT;
+ } else if (RAW == '\''){
+ NEXT;
+ encoding = xmlParseEncName(ctxt);
+ if (RAW != '\'') {
+ xmlFatalErr(ctxt, XML_ERR_STRING_NOT_CLOSED, NULL);
+ xmlFree((xmlChar *) encoding);
+ return(NULL);
+ } else
+ NEXT;
+ } else {
+ xmlFatalErr(ctxt, XML_ERR_STRING_NOT_STARTED, NULL);
+ }
+
+ /*
+ * Non standard parsing, allowing the user to ignore encoding
+ */
+ if (ctxt->options & XML_PARSE_IGNORE_ENC) {
+ xmlFree((xmlChar *) encoding);
+ return(NULL);
+ }
+
+ /*
+ * UTF-16 encoding stwich has already taken place at this stage,
+ * more over the little-endian/big-endian selection is already done
+ */
+ if ((encoding != NULL) &&
+ ((!xmlStrcasecmp(encoding, BAD_CAST "UTF-16")) ||
+ (!xmlStrcasecmp(encoding, BAD_CAST "UTF16")))) {
+ /*
+ * If no encoding was passed to the parser, that we are
+ * using UTF-16 and no decoder is present i.e. the
+ * document is apparently UTF-8 compatible, then raise an
+ * encoding mismatch fatal error
+ */
+ if ((ctxt->encoding == NULL) &&
+ (ctxt->input->buf != NULL) &&
+ (ctxt->input->buf->encoder == NULL)) {
+ xmlFatalErrMsg(ctxt, XML_ERR_INVALID_ENCODING,
+ "Document labelled UTF-16 but has UTF-8 content\n");
+ }
+ if (ctxt->encoding != NULL)
+ xmlFree((xmlChar *) ctxt->encoding);
+ ctxt->encoding = encoding;
+ }
+ /*
+ * UTF-8 encoding is handled natively
+ */
+ else if ((encoding != NULL) &&
+ ((!xmlStrcasecmp(encoding, BAD_CAST "UTF-8")) ||
+ (!xmlStrcasecmp(encoding, BAD_CAST "UTF8")))) {
+ if (ctxt->encoding != NULL)
+ xmlFree((xmlChar *) ctxt->encoding);
+ ctxt->encoding = encoding;
+ }
+ else if (encoding != NULL) {
+ xmlCharEncodingHandlerPtr handler;
+
+ if (ctxt->input->encoding != NULL)
+ xmlFree((xmlChar *) ctxt->input->encoding);
+ ctxt->input->encoding = encoding;
+
+ handler = xmlFindCharEncodingHandler((const char *) encoding);
+ if (handler != NULL) {
+ if (xmlSwitchToEncoding(ctxt, handler) < 0) {
+ /* failed to convert */
+ ctxt->errNo = XML_ERR_UNSUPPORTED_ENCODING;
+ return(NULL);
+ }
+ } else {
+ xmlFatalErrMsgStr(ctxt, XML_ERR_UNSUPPORTED_ENCODING,
+ "Unsupported encoding %s\n", encoding);
+ return(NULL);
+ }
+ }
+ }
+ return(encoding);
+}
+
+/**
+ * xmlParseSDDecl:
+ * @ctxt: an XML parser context
+ *
+ * parse the XML standalone declaration
+ *
+ * [32] SDDecl ::= S 'standalone' Eq
+ * (("'" ('yes' | 'no') "'") | ('"' ('yes' | 'no')'"'))
+ *
+ * [ VC: Standalone Document Declaration ]
+ * TODO The standalone document declaration must have the value "no"
+ * if any external markup declarations contain declarations of:
+ * - attributes with default values, if elements to which these
+ * attributes apply appear in the document without specifications
+ * of values for these attributes, or
+ * - entities (other than amp, lt, gt, apos, quot), if references
+ * to those entities appear in the document, or
+ * - attributes with values subject to normalization, where the
+ * attribute appears in the document with a value which will change
+ * as a result of normalization, or
+ * - element types with element content, if white space occurs directly
+ * within any instance of those types.
+ *
+ * Returns:
+ * 1 if standalone="yes"
+ * 0 if standalone="no"
+ * -2 if standalone attribute is missing or invalid
+ * (A standalone value of -2 means that the XML declaration was found,
+ * but no value was specified for the standalone attribute).
+ */
+
+int
+xmlParseSDDecl(xmlParserCtxtPtr ctxt) {
+ int standalone = -2;
+
+ SKIP_BLANKS;
+ if (CMP10(CUR_PTR, 's', 't', 'a', 'n', 'd', 'a', 'l', 'o', 'n', 'e')) {
+ SKIP(10);
+ SKIP_BLANKS;
+ if (RAW != '=') {
+ xmlFatalErr(ctxt, XML_ERR_EQUAL_REQUIRED, NULL);
+ return(standalone);
+ }
+ NEXT;
+ SKIP_BLANKS;
+ if (RAW == '\''){
+ NEXT;
+ if ((RAW == 'n') && (NXT(1) == 'o')) {
+ standalone = 0;
+ SKIP(2);
+ } else if ((RAW == 'y') && (NXT(1) == 'e') &&
+ (NXT(2) == 's')) {
+ standalone = 1;
+ SKIP(3);
+ } else {
+ xmlFatalErr(ctxt, XML_ERR_STANDALONE_VALUE, NULL);
+ }
+ if (RAW != '\'') {
+ xmlFatalErr(ctxt, XML_ERR_STRING_NOT_CLOSED, NULL);
+ } else
+ NEXT;
+ } else if (RAW == '"'){
+ NEXT;
+ if ((RAW == 'n') && (NXT(1) == 'o')) {
+ standalone = 0;
+ SKIP(2);
+ } else if ((RAW == 'y') && (NXT(1) == 'e') &&
+ (NXT(2) == 's')) {
+ standalone = 1;
+ SKIP(3);
+ } else {
+ xmlFatalErr(ctxt, XML_ERR_STANDALONE_VALUE, NULL);
+ }
+ if (RAW != '"') {
+ xmlFatalErr(ctxt, XML_ERR_STRING_NOT_CLOSED, NULL);
+ } else
+ NEXT;
+ } else {
+ xmlFatalErr(ctxt, XML_ERR_STRING_NOT_STARTED, NULL);
+ }
+ }
+ return(standalone);
+}
+
+/**
+ * xmlParseXMLDecl:
+ * @ctxt: an XML parser context
+ *
+ * parse an XML declaration header
+ *
+ * [23] XMLDecl ::= '<?xml' VersionInfo EncodingDecl? SDDecl? S? '?>'
+ */
+
+void
+xmlParseXMLDecl(xmlParserCtxtPtr ctxt) {
+ xmlChar *version;
+
+ /*
+ * This value for standalone indicates that the document has an
+ * XML declaration but it does not have a standalone attribute.
+ * It will be overwritten later if a standalone attribute is found.
+ */
+ ctxt->input->standalone = -2;
+
+ /*
+ * We know that '<?xml' is here.
+ */
+ SKIP(5);
+
+ if (!IS_BLANK_CH(RAW)) {
+ xmlFatalErrMsg(ctxt, XML_ERR_SPACE_REQUIRED,
+ "Blank needed after '<?xml'\n");
+ }
+ SKIP_BLANKS;
+
+ /*
+ * We must have the VersionInfo here.
+ */
+ version = xmlParseVersionInfo(ctxt);
+ if (version == NULL) {
+ xmlFatalErr(ctxt, XML_ERR_VERSION_MISSING, NULL);
+ } else {
+ if (!xmlStrEqual(version, (const xmlChar *) XML_DEFAULT_VERSION)) {
+ /*
+ * Changed here for XML-1.0 5th edition
+ */
+ if (ctxt->options & XML_PARSE_OLD10) {
+ xmlFatalErrMsgStr(ctxt, XML_ERR_UNKNOWN_VERSION,
+ "Unsupported version '%s'\n",
+ version);
+ } else {
+ if ((version[0] == '1') && ((version[1] == '.'))) {
+ xmlWarningMsg(ctxt, XML_WAR_UNKNOWN_VERSION,
+ "Unsupported version '%s'\n",
+ version, NULL);
+ } else {
+ xmlFatalErrMsgStr(ctxt, XML_ERR_UNKNOWN_VERSION,
+ "Unsupported version '%s'\n",
+ version);
+ }
+ }
+ }
+ if (ctxt->version != NULL)
+ xmlFree((void *) ctxt->version);
+ ctxt->version = version;
+ }
+
+ /*
+ * We may have the encoding declaration
+ */
+ if (!IS_BLANK_CH(RAW)) {
+ if ((RAW == '?') && (NXT(1) == '>')) {
+ SKIP(2);
+ return;
+ }
+ xmlFatalErrMsg(ctxt, XML_ERR_SPACE_REQUIRED, "Blank needed here\n");
+ }
+ xmlParseEncodingDecl(ctxt);
+ if ((ctxt->errNo == XML_ERR_UNSUPPORTED_ENCODING) ||
+ (ctxt->instate == XML_PARSER_EOF)) {
+ /*
+ * The XML REC instructs us to stop parsing right here
+ */
+ return;
+ }
+
+ /*
+ * We may have the standalone status.
+ */
+ if ((ctxt->input->encoding != NULL) && (!IS_BLANK_CH(RAW))) {
+ if ((RAW == '?') && (NXT(1) == '>')) {
+ SKIP(2);
+ return;
+ }
+ xmlFatalErrMsg(ctxt, XML_ERR_SPACE_REQUIRED, "Blank needed here\n");
+ }
+
+ /*
+ * We can grow the input buffer freely at that point
+ */
+ GROW;
+
+ SKIP_BLANKS;
+ ctxt->input->standalone = xmlParseSDDecl(ctxt);
+
+ SKIP_BLANKS;
+ if ((RAW == '?') && (NXT(1) == '>')) {
+ SKIP(2);
+ } else if (RAW == '>') {
+ /* Deprecated old WD ... */
+ xmlFatalErr(ctxt, XML_ERR_XMLDECL_NOT_FINISHED, NULL);
+ NEXT;
+ } else {
+ xmlFatalErr(ctxt, XML_ERR_XMLDECL_NOT_FINISHED, NULL);
+ MOVETO_ENDTAG(CUR_PTR);
+ NEXT;
+ }
+}
+
+/**
+ * xmlParseMisc:
+ * @ctxt: an XML parser context
+ *
+ * parse an XML Misc* optional field.
+ *
+ * [27] Misc ::= Comment | PI | S
+ */
+
+void
+xmlParseMisc(xmlParserCtxtPtr ctxt) {
+ while ((ctxt->instate != XML_PARSER_EOF) &&
+ (((RAW == '<') && (NXT(1) == '?')) ||
+ (CMP4(CUR_PTR, '<', '!', '-', '-')) ||
+ IS_BLANK_CH(CUR))) {
+ if ((RAW == '<') && (NXT(1) == '?')) {
+ xmlParsePI(ctxt);
+ } else if (IS_BLANK_CH(CUR)) {
+ NEXT;
+ } else
+ xmlParseComment(ctxt);
+ }
+}
+
+/**
+ * xmlParseDocument:
+ * @ctxt: an XML parser context
+ *
+ * parse an XML document (and build a tree if using the standard SAX
+ * interface).
+ *
+ * [1] document ::= prolog element Misc*
+ *
+ * [22] prolog ::= XMLDecl? Misc* (doctypedecl Misc*)?
+ *
+ * Returns 0, -1 in case of error. the parser context is augmented
+ * as a result of the parsing.
+ */
+
+int
+xmlParseDocument(xmlParserCtxtPtr ctxt) {
+ xmlChar start[4];
+ xmlCharEncoding enc;
+
+ xmlInitParser();
+
+ if ((ctxt == NULL) || (ctxt->input == NULL))
+ return(-1);
+
+ GROW;
+
+ /*
+ * SAX: detecting the level.
+ */
+ xmlDetectSAX2(ctxt);
+
+ /*
+ * SAX: beginning of the document processing.
+ */
+ if ((ctxt->sax) && (ctxt->sax->setDocumentLocator))
+ ctxt->sax->setDocumentLocator(ctxt->userData, &xmlDefaultSAXLocator);
+ if (ctxt->instate == XML_PARSER_EOF)
+ return(-1);
+
+ if ((ctxt->encoding == NULL) &&
+ ((ctxt->input->end - ctxt->input->cur) >= 4)) {
+ /*
+ * Get the 4 first bytes and decode the charset
+ * if enc != XML_CHAR_ENCODING_NONE
+ * plug some encoding conversion routines.
+ */
+ start[0] = RAW;
+ start[1] = NXT(1);
+ start[2] = NXT(2);
+ start[3] = NXT(3);
+ enc = xmlDetectCharEncoding(&start[0], 4);
+ if (enc != XML_CHAR_ENCODING_NONE) {
+ xmlSwitchEncoding(ctxt, enc);
+ }
+ }
+
+
+ if (CUR == 0) {
+ xmlFatalErr(ctxt, XML_ERR_DOCUMENT_EMPTY, NULL);
+ return(-1);
+ }
+
+ /*
+ * Check for the XMLDecl in the Prolog.
+ * do not GROW here to avoid the detected encoder to decode more
+ * than just the first line, unless the amount of data is really
+ * too small to hold "<?xml version="1.0" encoding="foo"
+ */
+ if ((ctxt->input->end - ctxt->input->cur) < 35) {
+ GROW;
+ }
+ if ((CMP5(CUR_PTR, '<', '?', 'x', 'm', 'l')) && (IS_BLANK_CH(NXT(5)))) {
+
+ /*
+ * Note that we will switch encoding on the fly.
+ */
+ xmlParseXMLDecl(ctxt);
+ if ((ctxt->errNo == XML_ERR_UNSUPPORTED_ENCODING) ||
+ (ctxt->instate == XML_PARSER_EOF)) {
+ /*
+ * The XML REC instructs us to stop parsing right here
+ */
+ return(-1);
+ }
+ ctxt->standalone = ctxt->input->standalone;
+ SKIP_BLANKS;
+ } else {
+ ctxt->version = xmlCharStrdup(XML_DEFAULT_VERSION);
+ }
+ if ((ctxt->sax) && (ctxt->sax->startDocument) && (!ctxt->disableSAX))
+ ctxt->sax->startDocument(ctxt->userData);
+ if (ctxt->instate == XML_PARSER_EOF)
+ return(-1);
+ if ((ctxt->myDoc != NULL) && (ctxt->input != NULL) &&
+ (ctxt->input->buf != NULL) && (ctxt->input->buf->compressed >= 0)) {
+ ctxt->myDoc->compression = ctxt->input->buf->compressed;
+ }
+
+ /*
+ * The Misc part of the Prolog
+ */
+ GROW;
+ xmlParseMisc(ctxt);
+
+ /*
+ * Then possibly doc type declaration(s) and more Misc
+ * (doctypedecl Misc*)?
+ */
+ GROW;
+ if (CMP9(CUR_PTR, '<', '!', 'D', 'O', 'C', 'T', 'Y', 'P', 'E')) {
+
+ ctxt->inSubset = 1;
+ xmlParseDocTypeDecl(ctxt);
+ if (RAW == '[') {
+ ctxt->instate = XML_PARSER_DTD;
+ xmlParseInternalSubset(ctxt);
+ if (ctxt->instate == XML_PARSER_EOF)
+ return(-1);
+ }
+
+ /*
+ * Create and update the external subset.
+ */
+ ctxt->inSubset = 2;
+ if ((ctxt->sax != NULL) && (ctxt->sax->externalSubset != NULL) &&
+ (!ctxt->disableSAX))
+ ctxt->sax->externalSubset(ctxt->userData, ctxt->intSubName,
+ ctxt->extSubSystem, ctxt->extSubURI);
+ if (ctxt->instate == XML_PARSER_EOF)
+ return(-1);
+ ctxt->inSubset = 0;
+
+ xmlCleanSpecialAttr(ctxt);
+
+ ctxt->instate = XML_PARSER_PROLOG;
+ xmlParseMisc(ctxt);
+ }
+
+ /*
+ * Time to start parsing the tree itself
+ */
+ GROW;
+ if (RAW != '<') {
+ xmlFatalErrMsg(ctxt, XML_ERR_DOCUMENT_EMPTY,
+ "Start tag expected, '<' not found\n");
+ } else {
+ ctxt->instate = XML_PARSER_CONTENT;
+ xmlParseElement(ctxt);
+ ctxt->instate = XML_PARSER_EPILOG;
+
+
+ /*
+ * The Misc part at the end
+ */
+ xmlParseMisc(ctxt);
+
+ if (RAW != 0) {
+ xmlFatalErr(ctxt, XML_ERR_DOCUMENT_END, NULL);
+ }
+ ctxt->instate = XML_PARSER_EOF;
+ }
+
+ /*
+ * SAX: end of the document processing.
+ */
+ if ((ctxt->sax) && (ctxt->sax->endDocument != NULL))
+ ctxt->sax->endDocument(ctxt->userData);
+
+ /*
+ * Remove locally kept entity definitions if the tree was not built
+ */
+ if ((ctxt->myDoc != NULL) &&
+ (xmlStrEqual(ctxt->myDoc->version, SAX_COMPAT_MODE))) {
+ xmlFreeDoc(ctxt->myDoc);
+ ctxt->myDoc = NULL;
+ }
+
+ if ((ctxt->wellFormed) && (ctxt->myDoc != NULL)) {
+ ctxt->myDoc->properties |= XML_DOC_WELLFORMED;
+ if (ctxt->valid)
+ ctxt->myDoc->properties |= XML_DOC_DTDVALID;
+ if (ctxt->nsWellFormed)
+ ctxt->myDoc->properties |= XML_DOC_NSVALID;
+ if (ctxt->options & XML_PARSE_OLD10)
+ ctxt->myDoc->properties |= XML_DOC_OLD10;
+ }
+ if (! ctxt->wellFormed) {
+ ctxt->valid = 0;
+ return(-1);
+ }
+ return(0);
+}
+
+/**
+ * xmlParseExtParsedEnt:
+ * @ctxt: an XML parser context
+ *
+ * parse a general parsed entity
+ * An external general parsed entity is well-formed if it matches the
+ * production labeled extParsedEnt.
+ *
+ * [78] extParsedEnt ::= TextDecl? content
+ *
+ * Returns 0, -1 in case of error. the parser context is augmented
+ * as a result of the parsing.
+ */
+
+int
+xmlParseExtParsedEnt(xmlParserCtxtPtr ctxt) {
+ xmlChar start[4];
+ xmlCharEncoding enc;
+
+ if ((ctxt == NULL) || (ctxt->input == NULL))
+ return(-1);
+
+ xmlDefaultSAXHandlerInit();
+
+ xmlDetectSAX2(ctxt);
+
+ GROW;
+
+ /*
+ * SAX: beginning of the document processing.
+ */
+ if ((ctxt->sax) && (ctxt->sax->setDocumentLocator))
+ ctxt->sax->setDocumentLocator(ctxt->userData, &xmlDefaultSAXLocator);
+
+ /*
+ * Get the 4 first bytes and decode the charset
+ * if enc != XML_CHAR_ENCODING_NONE
+ * plug some encoding conversion routines.
+ */
+ if ((ctxt->input->end - ctxt->input->cur) >= 4) {
+ start[0] = RAW;
+ start[1] = NXT(1);
+ start[2] = NXT(2);
+ start[3] = NXT(3);
+ enc = xmlDetectCharEncoding(start, 4);
+ if (enc != XML_CHAR_ENCODING_NONE) {
+ xmlSwitchEncoding(ctxt, enc);
+ }
+ }
+
+
+ if (CUR == 0) {
+ xmlFatalErr(ctxt, XML_ERR_DOCUMENT_EMPTY, NULL);
+ }
+
+ /*
+ * Check for the XMLDecl in the Prolog.
+ */
+ GROW;
+ if ((CMP5(CUR_PTR, '<', '?', 'x', 'm', 'l')) && (IS_BLANK_CH(NXT(5)))) {
+
+ /*
+ * Note that we will switch encoding on the fly.
+ */
+ xmlParseXMLDecl(ctxt);
+ if (ctxt->errNo == XML_ERR_UNSUPPORTED_ENCODING) {
+ /*
+ * The XML REC instructs us to stop parsing right here
+ */
+ return(-1);
+ }
+ SKIP_BLANKS;
+ } else {
+ ctxt->version = xmlCharStrdup(XML_DEFAULT_VERSION);
+ }
+ if ((ctxt->sax) && (ctxt->sax->startDocument) && (!ctxt->disableSAX))
+ ctxt->sax->startDocument(ctxt->userData);
+ if (ctxt->instate == XML_PARSER_EOF)
+ return(-1);
+
+ /*
+ * Doing validity checking on chunk doesn't make sense
+ */
+ ctxt->instate = XML_PARSER_CONTENT;
+ ctxt->validate = 0;
+ ctxt->loadsubset = 0;
+ ctxt->depth = 0;
+
+ xmlParseContent(ctxt);
+ if (ctxt->instate == XML_PARSER_EOF)
+ return(-1);
+
+ if ((RAW == '<') && (NXT(1) == '/')) {
+ xmlFatalErr(ctxt, XML_ERR_NOT_WELL_BALANCED, NULL);
+ } else if (RAW != 0) {
+ xmlFatalErr(ctxt, XML_ERR_EXTRA_CONTENT, NULL);
+ }
+
+ /*
+ * SAX: end of the document processing.
+ */
+ if ((ctxt->sax) && (ctxt->sax->endDocument != NULL))
+ ctxt->sax->endDocument(ctxt->userData);
+
+ if (! ctxt->wellFormed) return(-1);
+ return(0);
+}
+
+#ifdef LIBXML_PUSH_ENABLED
+/************************************************************************
+ * *
+ * Progressive parsing interfaces *
+ * *
+ ************************************************************************/
+
+/**
+ * xmlParseLookupSequence:
+ * @ctxt: an XML parser context
+ * @first: the first char to lookup
+ * @next: the next char to lookup or zero
+ * @third: the next char to lookup or zero
+ *
+ * Try to find if a sequence (first, next, third) or just (first next) or
+ * (first) is available in the input stream.
+ * This function has a side effect of (possibly) incrementing ctxt->checkIndex
+ * to avoid rescanning sequences of bytes, it DOES change the state of the
+ * parser, do not use liberally.
+ *
+ * Returns the index to the current parsing point if the full sequence
+ * is available, -1 otherwise.
+ */
+static int
+xmlParseLookupSequence(xmlParserCtxtPtr ctxt, xmlChar first,
+ xmlChar next, xmlChar third) {
+ int base, len;
+ xmlParserInputPtr in;
+ const xmlChar *buf;
+
+ in = ctxt->input;
+ if (in == NULL) return(-1);
+ base = in->cur - in->base;
+ if (base < 0) return(-1);
+ if (ctxt->checkIndex > base)
+ base = ctxt->checkIndex;
+ if (in->buf == NULL) {
+ buf = in->base;
+ len = in->length;
+ } else {
+ buf = xmlBufContent(in->buf->buffer);
+ len = xmlBufUse(in->buf->buffer);
+ }
+ /* take into account the sequence length */
+ if (third) len -= 2;
+ else if (next) len --;
+ for (;base < len;base++) {
+ if (buf[base] == first) {
+ if (third != 0) {
+ if ((buf[base + 1] != next) ||
+ (buf[base + 2] != third)) continue;
+ } else if (next != 0) {
+ if (buf[base + 1] != next) continue;
+ }
+ ctxt->checkIndex = 0;
+#ifdef DEBUG_PUSH
+ if (next == 0)
+ xmlGenericError(xmlGenericErrorContext,
+ "PP: lookup '%c' found at %d\n",
+ first, base);
+ else if (third == 0)
+ xmlGenericError(xmlGenericErrorContext,
+ "PP: lookup '%c%c' found at %d\n",
+ first, next, base);
+ else
+ xmlGenericError(xmlGenericErrorContext,
+ "PP: lookup '%c%c%c' found at %d\n",
+ first, next, third, base);
+#endif
+ return(base - (in->cur - in->base));
+ }
+ }
+ ctxt->checkIndex = base;
+#ifdef DEBUG_PUSH
+ if (next == 0)
+ xmlGenericError(xmlGenericErrorContext,
+ "PP: lookup '%c' failed\n", first);
+ else if (third == 0)
+ xmlGenericError(xmlGenericErrorContext,
+ "PP: lookup '%c%c' failed\n", first, next);
+ else
+ xmlGenericError(xmlGenericErrorContext,
+ "PP: lookup '%c%c%c' failed\n", first, next, third);
+#endif
+ return(-1);
+}
+
+/**
+ * xmlParseGetLasts:
+ * @ctxt: an XML parser context
+ * @lastlt: pointer to store the last '<' from the input
+ * @lastgt: pointer to store the last '>' from the input
+ *
+ * Lookup the last < and > in the current chunk
+ */
+static void
+xmlParseGetLasts(xmlParserCtxtPtr ctxt, const xmlChar **lastlt,
+ const xmlChar **lastgt) {
+ const xmlChar *tmp;
+
+ if ((ctxt == NULL) || (lastlt == NULL) || (lastgt == NULL)) {
+ xmlGenericError(xmlGenericErrorContext,
+ "Internal error: xmlParseGetLasts\n");
+ return;
+ }
+ if ((ctxt->progressive != 0) && (ctxt->inputNr == 1)) {
+ tmp = ctxt->input->end;
+ tmp--;
+ while ((tmp >= ctxt->input->base) && (*tmp != '<')) tmp--;
+ if (tmp < ctxt->input->base) {
+ *lastlt = NULL;
+ *lastgt = NULL;
+ } else {
+ *lastlt = tmp;
+ tmp++;
+ while ((tmp < ctxt->input->end) && (*tmp != '>')) {
+ if (*tmp == '\'') {
+ tmp++;
+ while ((tmp < ctxt->input->end) && (*tmp != '\'')) tmp++;
+ if (tmp < ctxt->input->end) tmp++;
+ } else if (*tmp == '"') {
+ tmp++;
+ while ((tmp < ctxt->input->end) && (*tmp != '"')) tmp++;
+ if (tmp < ctxt->input->end) tmp++;
+ } else
+ tmp++;
+ }
+ if (tmp < ctxt->input->end)
+ *lastgt = tmp;
+ else {
+ tmp = *lastlt;
+ tmp--;
+ while ((tmp >= ctxt->input->base) && (*tmp != '>')) tmp--;
+ if (tmp >= ctxt->input->base)
+ *lastgt = tmp;
+ else
+ *lastgt = NULL;
+ }
+ }
+ } else {
+ *lastlt = NULL;
+ *lastgt = NULL;
+ }
+}
+/**
+ * xmlCheckCdataPush:
+ * @cur: pointer to the block of characters
+ * @len: length of the block in bytes
+ * @complete: 1 if complete CDATA block is passed in, 0 if partial block
+ *
+ * Check that the block of characters is okay as SCdata content [20]
+ *
+ * Returns the number of bytes to pass if okay, a negative index where an
+ * UTF-8 error occured otherwise
+ */
+static int
+xmlCheckCdataPush(const xmlChar *utf, int len, int complete) {
+ int ix;
+ unsigned char c;
+ int codepoint;
+
+ if ((utf == NULL) || (len <= 0))
+ return(0);
+
+ for (ix = 0; ix < len;) { /* string is 0-terminated */
+ c = utf[ix];
+ if ((c & 0x80) == 0x00) { /* 1-byte code, starts with 10 */
+ if (c >= 0x20)
+ ix++;
+ else if ((c == 0xA) || (c == 0xD) || (c == 0x9))
+ ix++;
+ else
+ return(-ix);
+ } else if ((c & 0xe0) == 0xc0) {/* 2-byte code, starts with 110 */
+ if (ix + 2 > len) return(complete ? -ix : ix);
+ if ((utf[ix+1] & 0xc0 ) != 0x80)
+ return(-ix);
+ codepoint = (utf[ix] & 0x1f) << 6;
+ codepoint |= utf[ix+1] & 0x3f;
+ if (!xmlIsCharQ(codepoint))
+ return(-ix);
+ ix += 2;
+ } else if ((c & 0xf0) == 0xe0) {/* 3-byte code, starts with 1110 */
+ if (ix + 3 > len) return(complete ? -ix : ix);
+ if (((utf[ix+1] & 0xc0) != 0x80) ||
+ ((utf[ix+2] & 0xc0) != 0x80))
+ return(-ix);
+ codepoint = (utf[ix] & 0xf) << 12;
+ codepoint |= (utf[ix+1] & 0x3f) << 6;
+ codepoint |= utf[ix+2] & 0x3f;
+ if (!xmlIsCharQ(codepoint))
+ return(-ix);
+ ix += 3;
+ } else if ((c & 0xf8) == 0xf0) {/* 4-byte code, starts with 11110 */
+ if (ix + 4 > len) return(complete ? -ix : ix);
+ if (((utf[ix+1] & 0xc0) != 0x80) ||
+ ((utf[ix+2] & 0xc0) != 0x80) ||
+ ((utf[ix+3] & 0xc0) != 0x80))
+ return(-ix);
+ codepoint = (utf[ix] & 0x7) << 18;
+ codepoint |= (utf[ix+1] & 0x3f) << 12;
+ codepoint |= (utf[ix+2] & 0x3f) << 6;
+ codepoint |= utf[ix+3] & 0x3f;
+ if (!xmlIsCharQ(codepoint))
+ return(-ix);
+ ix += 4;
+ } else /* unknown encoding */
+ return(-ix);
+ }
+ return(ix);
+}
+
+/**
+ * xmlParseTryOrFinish:
+ * @ctxt: an XML parser context
+ * @terminate: last chunk indicator
+ *
+ * Try to progress on parsing
+ *
+ * Returns zero if no parsing was possible
+ */
+static int
+xmlParseTryOrFinish(xmlParserCtxtPtr ctxt, int terminate) {
+ int ret = 0;
+ int avail, tlen;
+ xmlChar cur, next;
+ const xmlChar *lastlt, *lastgt;
+
+ if (ctxt->input == NULL)
+ return(0);
+
+#ifdef DEBUG_PUSH
+ switch (ctxt->instate) {
+ case XML_PARSER_EOF:
+ xmlGenericError(xmlGenericErrorContext,
+ "PP: try EOF\n"); break;
+ case XML_PARSER_START:
+ xmlGenericError(xmlGenericErrorContext,
+ "PP: try START\n"); break;
+ case XML_PARSER_MISC:
+ xmlGenericError(xmlGenericErrorContext,
+ "PP: try MISC\n");break;
+ case XML_PARSER_COMMENT:
+ xmlGenericError(xmlGenericErrorContext,
+ "PP: try COMMENT\n");break;
+ case XML_PARSER_PROLOG:
+ xmlGenericError(xmlGenericErrorContext,
+ "PP: try PROLOG\n");break;
+ case XML_PARSER_START_TAG:
+ xmlGenericError(xmlGenericErrorContext,
+ "PP: try START_TAG\n");break;
+ case XML_PARSER_CONTENT:
+ xmlGenericError(xmlGenericErrorContext,
+ "PP: try CONTENT\n");break;
+ case XML_PARSER_CDATA_SECTION:
+ xmlGenericError(xmlGenericErrorContext,
+ "PP: try CDATA_SECTION\n");break;
+ case XML_PARSER_END_TAG:
+ xmlGenericError(xmlGenericErrorContext,
+ "PP: try END_TAG\n");break;
+ case XML_PARSER_ENTITY_DECL:
+ xmlGenericError(xmlGenericErrorContext,
+ "PP: try ENTITY_DECL\n");break;
+ case XML_PARSER_ENTITY_VALUE:
+ xmlGenericError(xmlGenericErrorContext,
+ "PP: try ENTITY_VALUE\n");break;
+ case XML_PARSER_ATTRIBUTE_VALUE:
+ xmlGenericError(xmlGenericErrorContext,
+ "PP: try ATTRIBUTE_VALUE\n");break;
+ case XML_PARSER_DTD:
+ xmlGenericError(xmlGenericErrorContext,
+ "PP: try DTD\n");break;
+ case XML_PARSER_EPILOG:
+ xmlGenericError(xmlGenericErrorContext,
+ "PP: try EPILOG\n");break;
+ case XML_PARSER_PI:
+ xmlGenericError(xmlGenericErrorContext,
+ "PP: try PI\n");break;
+ case XML_PARSER_IGNORE:
+ xmlGenericError(xmlGenericErrorContext,
+ "PP: try IGNORE\n");break;
+ }
+#endif
+
+ if ((ctxt->input != NULL) &&
+ (ctxt->input->cur - ctxt->input->base > 4096)) {
+ xmlSHRINK(ctxt);
+ ctxt->checkIndex = 0;
+ }
+ xmlParseGetLasts(ctxt, &lastlt, &lastgt);
+
+ while (ctxt->instate != XML_PARSER_EOF) {
+ if ((ctxt->errNo != XML_ERR_OK) && (ctxt->disableSAX == 1))
+ return(0);
+
+
+ /*
+ * Pop-up of finished entities.
+ */
+ while ((RAW == 0) && (ctxt->inputNr > 1))
+ xmlPopInput(ctxt);
+
+ if (ctxt->input == NULL) break;
+ if (ctxt->input->buf == NULL)
+ avail = ctxt->input->length -
+ (ctxt->input->cur - ctxt->input->base);
+ else {
+ /*
+ * If we are operating on converted input, try to flush
+ * remainng chars to avoid them stalling in the non-converted
+ * buffer. But do not do this in document start where
+ * encoding="..." may not have been read and we work on a
+ * guessed encoding.
+ */
+ if ((ctxt->instate != XML_PARSER_START) &&
+ (ctxt->input->buf->raw != NULL) &&
+ (xmlBufIsEmpty(ctxt->input->buf->raw) == 0)) {
+ size_t base = xmlBufGetInputBase(ctxt->input->buf->buffer,
+ ctxt->input);
+ size_t current = ctxt->input->cur - ctxt->input->base;
+
+ xmlParserInputBufferPush(ctxt->input->buf, 0, "");
+ xmlBufSetInputBaseCur(ctxt->input->buf->buffer, ctxt->input,
+ base, current);
+ }
+ avail = xmlBufUse(ctxt->input->buf->buffer) -
+ (ctxt->input->cur - ctxt->input->base);
+ }
+ if (avail < 1)
+ goto done;
+ switch (ctxt->instate) {
+ case XML_PARSER_EOF:
+ /*
+ * Document parsing is done !
+ */
+ goto done;
+ case XML_PARSER_START:
+ if (ctxt->charset == XML_CHAR_ENCODING_NONE) {
+ xmlChar start[4];
+ xmlCharEncoding enc;
+
+ /*
+ * Very first chars read from the document flow.
+ */
+ if (avail < 4)
+ goto done;
+
+ /*
+ * Get the 4 first bytes and decode the charset
+ * if enc != XML_CHAR_ENCODING_NONE
+ * plug some encoding conversion routines,
+ * else xmlSwitchEncoding will set to (default)
+ * UTF8.
+ */
+ start[0] = RAW;
+ start[1] = NXT(1);
+ start[2] = NXT(2);
+ start[3] = NXT(3);
+ enc = xmlDetectCharEncoding(start, 4);
+ xmlSwitchEncoding(ctxt, enc);
+ break;
+ }
+
+ if (avail < 2)
+ goto done;
+ cur = ctxt->input->cur[0];
+ next = ctxt->input->cur[1];
+ if (cur == 0) {
+ if ((ctxt->sax) && (ctxt->sax->setDocumentLocator))
+ ctxt->sax->setDocumentLocator(ctxt->userData,
+ &xmlDefaultSAXLocator);
+ xmlFatalErr(ctxt, XML_ERR_DOCUMENT_EMPTY, NULL);
+ xmlHaltParser(ctxt);
+#ifdef DEBUG_PUSH
+ xmlGenericError(xmlGenericErrorContext,
+ "PP: entering EOF\n");
+#endif
+ if ((ctxt->sax) && (ctxt->sax->endDocument != NULL))
+ ctxt->sax->endDocument(ctxt->userData);
+ goto done;
+ }
+ if ((cur == '<') && (next == '?')) {
+ /* PI or XML decl */
+ if (avail < 5) return(ret);
+ if ((!terminate) &&
+ (xmlParseLookupSequence(ctxt, '?', '>', 0) < 0))
+ return(ret);
+ if ((ctxt->sax) && (ctxt->sax->setDocumentLocator))
+ ctxt->sax->setDocumentLocator(ctxt->userData,
+ &xmlDefaultSAXLocator);
+ if ((ctxt->input->cur[2] == 'x') &&
+ (ctxt->input->cur[3] == 'm') &&
+ (ctxt->input->cur[4] == 'l') &&
+ (IS_BLANK_CH(ctxt->input->cur[5]))) {
+ ret += 5;
+#ifdef DEBUG_PUSH
+ xmlGenericError(xmlGenericErrorContext,
+ "PP: Parsing XML Decl\n");
+#endif
+ xmlParseXMLDecl(ctxt);
+ if (ctxt->errNo == XML_ERR_UNSUPPORTED_ENCODING) {
+ /*
+ * The XML REC instructs us to stop parsing right
+ * here
+ */
+ xmlHaltParser(ctxt);
+ return(0);
+ }
+ ctxt->standalone = ctxt->input->standalone;
+ if ((ctxt->encoding == NULL) &&
+ (ctxt->input->encoding != NULL))
+ ctxt->encoding = xmlStrdup(ctxt->input->encoding);
+ if ((ctxt->sax) && (ctxt->sax->startDocument) &&
+ (!ctxt->disableSAX))
+ ctxt->sax->startDocument(ctxt->userData);
+ ctxt->instate = XML_PARSER_MISC;
+#ifdef DEBUG_PUSH
+ xmlGenericError(xmlGenericErrorContext,
+ "PP: entering MISC\n");
+#endif
+ } else {
+ ctxt->version = xmlCharStrdup(XML_DEFAULT_VERSION);
+ if ((ctxt->sax) && (ctxt->sax->startDocument) &&
+ (!ctxt->disableSAX))
+ ctxt->sax->startDocument(ctxt->userData);
+ ctxt->instate = XML_PARSER_MISC;
+#ifdef DEBUG_PUSH
+ xmlGenericError(xmlGenericErrorContext,
+ "PP: entering MISC\n");
+#endif
+ }
+ } else {
+ if ((ctxt->sax) && (ctxt->sax->setDocumentLocator))
+ ctxt->sax->setDocumentLocator(ctxt->userData,
+ &xmlDefaultSAXLocator);
+ ctxt->version = xmlCharStrdup(XML_DEFAULT_VERSION);
+ if (ctxt->version == NULL) {
+ xmlErrMemory(ctxt, NULL);
+ break;
+ }
+ if ((ctxt->sax) && (ctxt->sax->startDocument) &&
+ (!ctxt->disableSAX))
+ ctxt->sax->startDocument(ctxt->userData);
+ ctxt->instate = XML_PARSER_MISC;
+#ifdef DEBUG_PUSH
+ xmlGenericError(xmlGenericErrorContext,
+ "PP: entering MISC\n");
+#endif
+ }
+ break;
+ case XML_PARSER_START_TAG: {
+ const xmlChar *name;
+ const xmlChar *prefix = NULL;
+ const xmlChar *URI = NULL;
+ int nsNr = ctxt->nsNr;
+
+ if ((avail < 2) && (ctxt->inputNr == 1))
+ goto done;
+ cur = ctxt->input->cur[0];
+ if (cur != '<') {
+ xmlFatalErr(ctxt, XML_ERR_DOCUMENT_EMPTY, NULL);
+ xmlHaltParser(ctxt);
+ if ((ctxt->sax) && (ctxt->sax->endDocument != NULL))
+ ctxt->sax->endDocument(ctxt->userData);
+ goto done;
+ }
+ if (!terminate) {
+ if (ctxt->progressive) {
+ /* > can be found unescaped in attribute values */
+ if ((lastgt == NULL) || (ctxt->input->cur >= lastgt))
+ goto done;
+ } else if (xmlParseLookupSequence(ctxt, '>', 0, 0) < 0) {
+ goto done;
+ }
+ }
+ if (ctxt->spaceNr == 0)
+ spacePush(ctxt, -1);
+ else if (*ctxt->space == -2)
+ spacePush(ctxt, -1);
+ else
+ spacePush(ctxt, *ctxt->space);
+#ifdef LIBXML_SAX1_ENABLED
+ if (ctxt->sax2)
+#endif /* LIBXML_SAX1_ENABLED */
+ name = xmlParseStartTag2(ctxt, &prefix, &URI, &tlen);
+#ifdef LIBXML_SAX1_ENABLED
+ else
+ name = xmlParseStartTag(ctxt);
+#endif /* LIBXML_SAX1_ENABLED */
+ if (ctxt->instate == XML_PARSER_EOF)
+ goto done;
+ if (name == NULL) {
+ spacePop(ctxt);
+ xmlHaltParser(ctxt);
+ if ((ctxt->sax) && (ctxt->sax->endDocument != NULL))
+ ctxt->sax->endDocument(ctxt->userData);
+ goto done;
+ }
+#ifdef LIBXML_VALID_ENABLED
+ /*
+ * [ VC: Root Element Type ]
+ * The Name in the document type declaration must match
+ * the element type of the root element.
+ */
+ if (ctxt->validate && ctxt->wellFormed && ctxt->myDoc &&
+ ctxt->node && (ctxt->node == ctxt->myDoc->children))
+ ctxt->valid &= xmlValidateRoot(&ctxt->vctxt, ctxt->myDoc);
+#endif /* LIBXML_VALID_ENABLED */
+
+ /*
+ * Check for an Empty Element.
+ */
+ if ((RAW == '/') && (NXT(1) == '>')) {
+ SKIP(2);
+
+ if (ctxt->sax2) {
+ if ((ctxt->sax != NULL) &&
+ (ctxt->sax->endElementNs != NULL) &&
+ (!ctxt->disableSAX))
+ ctxt->sax->endElementNs(ctxt->userData, name,
+ prefix, URI);
+ if (ctxt->nsNr - nsNr > 0)
+ nsPop(ctxt, ctxt->nsNr - nsNr);
+#ifdef LIBXML_SAX1_ENABLED
+ } else {
+ if ((ctxt->sax != NULL) &&
+ (ctxt->sax->endElement != NULL) &&
+ (!ctxt->disableSAX))
+ ctxt->sax->endElement(ctxt->userData, name);
+#endif /* LIBXML_SAX1_ENABLED */
+ }
+ if (ctxt->instate == XML_PARSER_EOF)
+ goto done;
+ spacePop(ctxt);
+ if (ctxt->nameNr == 0) {
+ ctxt->instate = XML_PARSER_EPILOG;
+ } else {
+ ctxt->instate = XML_PARSER_CONTENT;
+ }
+ ctxt->progressive = 1;
+ break;
+ }
+ if (RAW == '>') {
+ NEXT;
+ } else {
+ xmlFatalErrMsgStr(ctxt, XML_ERR_GT_REQUIRED,
+ "Couldn't find end of Start Tag %s\n",
+ name);
+ nodePop(ctxt);
+ spacePop(ctxt);
+ }
+ if (ctxt->sax2)
+ nameNsPush(ctxt, name, prefix, URI, ctxt->nsNr - nsNr);
+#ifdef LIBXML_SAX1_ENABLED
+ else
+ namePush(ctxt, name);
+#endif /* LIBXML_SAX1_ENABLED */
+
+ ctxt->instate = XML_PARSER_CONTENT;
+ ctxt->progressive = 1;
+ break;
+ }
+ case XML_PARSER_CONTENT: {
+ const xmlChar *test;
+ unsigned int cons;
+ if ((avail < 2) && (ctxt->inputNr == 1))
+ goto done;
+ cur = ctxt->input->cur[0];
+ next = ctxt->input->cur[1];
+
+ test = CUR_PTR;
+ cons = ctxt->input->consumed;
+ if ((cur == '<') && (next == '/')) {
+ ctxt->instate = XML_PARSER_END_TAG;
+ break;
+ } else if ((cur == '<') && (next == '?')) {
+ if ((!terminate) &&
+ (xmlParseLookupSequence(ctxt, '?', '>', 0) < 0)) {
+ ctxt->progressive = XML_PARSER_PI;
+ goto done;
+ }
+ xmlParsePI(ctxt);
+ ctxt->instate = XML_PARSER_CONTENT;
+ ctxt->progressive = 1;
+ } else if ((cur == '<') && (next != '!')) {
+ ctxt->instate = XML_PARSER_START_TAG;
+ break;
+ } else if ((cur == '<') && (next == '!') &&
+ (ctxt->input->cur[2] == '-') &&
+ (ctxt->input->cur[3] == '-')) {
+ int term;
+
+ if (avail < 4)
+ goto done;
+ ctxt->input->cur += 4;
+ term = xmlParseLookupSequence(ctxt, '-', '-', '>');
+ ctxt->input->cur -= 4;
+ if ((!terminate) && (term < 0)) {
+ ctxt->progressive = XML_PARSER_COMMENT;
+ goto done;
+ }
+ xmlParseComment(ctxt);
+ ctxt->instate = XML_PARSER_CONTENT;
+ ctxt->progressive = 1;
+ } else if ((cur == '<') && (ctxt->input->cur[1] == '!') &&
+ (ctxt->input->cur[2] == '[') &&
+ (ctxt->input->cur[3] == 'C') &&
+ (ctxt->input->cur[4] == 'D') &&
+ (ctxt->input->cur[5] == 'A') &&
+ (ctxt->input->cur[6] == 'T') &&
+ (ctxt->input->cur[7] == 'A') &&
+ (ctxt->input->cur[8] == '[')) {
+ SKIP(9);
+ ctxt->instate = XML_PARSER_CDATA_SECTION;
+ break;
+ } else if ((cur == '<') && (next == '!') &&
+ (avail < 9)) {
+ goto done;
+ } else if (cur == '&') {
+ if ((!terminate) &&
+ (xmlParseLookupSequence(ctxt, ';', 0, 0) < 0))
+ goto done;
+ xmlParseReference(ctxt);
+ } else {
+ /* TODO Avoid the extra copy, handle directly !!! */
+ /*
+ * Goal of the following test is:
+ * - minimize calls to the SAX 'character' callback
+ * when they are mergeable
+ * - handle an problem for isBlank when we only parse
+ * a sequence of blank chars and the next one is
+ * not available to check against '<' presence.
+ * - tries to homogenize the differences in SAX
+ * callbacks between the push and pull versions
+ * of the parser.
+ */
+ if ((ctxt->inputNr == 1) &&
+ (avail < XML_PARSER_BIG_BUFFER_SIZE)) {
+ if (!terminate) {
+ if (ctxt->progressive) {
+ if ((lastlt == NULL) ||
+ (ctxt->input->cur > lastlt))
+ goto done;
+ } else if (xmlParseLookupSequence(ctxt,
+ '<', 0, 0) < 0) {
+ goto done;
+ }
+ }
+ }
+ ctxt->checkIndex = 0;
+ xmlParseCharData(ctxt, 0);
+ }
+ /*
+ * Pop-up of finished entities.
+ */
+ while ((RAW == 0) && (ctxt->inputNr > 1))
+ xmlPopInput(ctxt);
+ if ((cons == ctxt->input->consumed) && (test == CUR_PTR)) {
+ xmlFatalErr(ctxt, XML_ERR_INTERNAL_ERROR,
+ "detected an error in element content\n");
+ xmlHaltParser(ctxt);
+ break;
+ }
+ break;
+ }
+ case XML_PARSER_END_TAG:
+ if (avail < 2)
+ goto done;
+ if (!terminate) {
+ if (ctxt->progressive) {
+ /* > can be found unescaped in attribute values */
+ if ((lastgt == NULL) || (ctxt->input->cur >= lastgt))
+ goto done;
+ } else if (xmlParseLookupSequence(ctxt, '>', 0, 0) < 0) {
+ goto done;
+ }
+ }
+ if (ctxt->sax2) {
+ xmlParseEndTag2(ctxt,
+ (void *) ctxt->pushTab[ctxt->nameNr * 3 - 3],
+ (void *) ctxt->pushTab[ctxt->nameNr * 3 - 2], 0,
+ (int) (long) ctxt->pushTab[ctxt->nameNr * 3 - 1], 0);
+ nameNsPop(ctxt);
+ }
+#ifdef LIBXML_SAX1_ENABLED
+ else
+ xmlParseEndTag1(ctxt, 0);
+#endif /* LIBXML_SAX1_ENABLED */
+ if (ctxt->instate == XML_PARSER_EOF) {
+ /* Nothing */
+ } else if (ctxt->nameNr == 0) {
+ ctxt->instate = XML_PARSER_EPILOG;
+ } else {
+ ctxt->instate = XML_PARSER_CONTENT;
+ }
+ break;
+ case XML_PARSER_CDATA_SECTION: {
+ /*
+ * The Push mode need to have the SAX callback for
+ * cdataBlock merge back contiguous callbacks.
+ */
+ int base;
+
+ base = xmlParseLookupSequence(ctxt, ']', ']', '>');
+ if (base < 0) {
+ if (avail >= XML_PARSER_BIG_BUFFER_SIZE + 2) {
+ int tmp;
+
+ tmp = xmlCheckCdataPush(ctxt->input->cur,
+ XML_PARSER_BIG_BUFFER_SIZE, 0);
+ if (tmp < 0) {
+ tmp = -tmp;
+ ctxt->input->cur += tmp;
+ goto encoding_error;
+ }
+ if ((ctxt->sax != NULL) && (!ctxt->disableSAX)) {
+ if (ctxt->sax->cdataBlock != NULL)
+ ctxt->sax->cdataBlock(ctxt->userData,
+ ctxt->input->cur, tmp);
+ else if (ctxt->sax->characters != NULL)
+ ctxt->sax->characters(ctxt->userData,
+ ctxt->input->cur, tmp);
+ }
+ if (ctxt->instate == XML_PARSER_EOF)
+ goto done;
+ SKIPL(tmp);
+ ctxt->checkIndex = 0;
+ }
+ goto done;
+ } else {
+ int tmp;
+
+ tmp = xmlCheckCdataPush(ctxt->input->cur, base, 1);
+ if ((tmp < 0) || (tmp != base)) {
+ tmp = -tmp;
+ ctxt->input->cur += tmp;
+ goto encoding_error;
+ }
+ if ((ctxt->sax != NULL) && (base == 0) &&
+ (ctxt->sax->cdataBlock != NULL) &&
+ (!ctxt->disableSAX)) {
+ /*
+ * Special case to provide identical behaviour
+ * between pull and push parsers on enpty CDATA
+ * sections
+ */
+ if ((ctxt->input->cur - ctxt->input->base >= 9) &&
+ (!strncmp((const char *)&ctxt->input->cur[-9],
+ "<![CDATA[", 9)))
+ ctxt->sax->cdataBlock(ctxt->userData,
+ BAD_CAST "", 0);
+ } else if ((ctxt->sax != NULL) && (base > 0) &&
+ (!ctxt->disableSAX)) {
+ if (ctxt->sax->cdataBlock != NULL)
+ ctxt->sax->cdataBlock(ctxt->userData,
+ ctxt->input->cur, base);
+ else if (ctxt->sax->characters != NULL)
+ ctxt->sax->characters(ctxt->userData,
+ ctxt->input->cur, base);
+ }
+ if (ctxt->instate == XML_PARSER_EOF)
+ goto done;
+ SKIPL(base + 3);
+ ctxt->checkIndex = 0;
+ ctxt->instate = XML_PARSER_CONTENT;
+#ifdef DEBUG_PUSH
+ xmlGenericError(xmlGenericErrorContext,
+ "PP: entering CONTENT\n");
+#endif
+ }
+ break;
+ }
+ case XML_PARSER_MISC:
+ SKIP_BLANKS;
+ if (ctxt->input->buf == NULL)
+ avail = ctxt->input->length -
+ (ctxt->input->cur - ctxt->input->base);
+ else
+ avail = xmlBufUse(ctxt->input->buf->buffer) -
+ (ctxt->input->cur - ctxt->input->base);
+ if (avail < 2)
+ goto done;
+ cur = ctxt->input->cur[0];
+ next = ctxt->input->cur[1];
+ if ((cur == '<') && (next == '?')) {
+ if ((!terminate) &&
+ (xmlParseLookupSequence(ctxt, '?', '>', 0) < 0)) {
+ ctxt->progressive = XML_PARSER_PI;
+ goto done;
+ }
+#ifdef DEBUG_PUSH
+ xmlGenericError(xmlGenericErrorContext,
+ "PP: Parsing PI\n");
+#endif
+ xmlParsePI(ctxt);
+ if (ctxt->instate == XML_PARSER_EOF)
+ goto done;
+ ctxt->instate = XML_PARSER_MISC;
+ ctxt->progressive = 1;
+ ctxt->checkIndex = 0;
+ } else if ((cur == '<') && (next == '!') &&
+ (ctxt->input->cur[2] == '-') &&
+ (ctxt->input->cur[3] == '-')) {
+ if ((!terminate) &&
+ (xmlParseLookupSequence(ctxt, '-', '-', '>') < 0)) {
+ ctxt->progressive = XML_PARSER_COMMENT;
+ goto done;
+ }
+#ifdef DEBUG_PUSH
+ xmlGenericError(xmlGenericErrorContext,
+ "PP: Parsing Comment\n");
+#endif
+ xmlParseComment(ctxt);
+ if (ctxt->instate == XML_PARSER_EOF)
+ goto done;
+ ctxt->instate = XML_PARSER_MISC;
+ ctxt->progressive = 1;
+ ctxt->checkIndex = 0;
+ } else if ((cur == '<') && (next == '!') &&
+ (ctxt->input->cur[2] == 'D') &&
+ (ctxt->input->cur[3] == 'O') &&
+ (ctxt->input->cur[4] == 'C') &&
+ (ctxt->input->cur[5] == 'T') &&
+ (ctxt->input->cur[6] == 'Y') &&
+ (ctxt->input->cur[7] == 'P') &&
+ (ctxt->input->cur[8] == 'E')) {
+ if ((!terminate) &&
+ (xmlParseLookupSequence(ctxt, '>', 0, 0) < 0)) {
+ ctxt->progressive = XML_PARSER_DTD;
+ goto done;
+ }
+#ifdef DEBUG_PUSH
+ xmlGenericError(xmlGenericErrorContext,
+ "PP: Parsing internal subset\n");
+#endif
+ ctxt->inSubset = 1;
+ ctxt->progressive = 0;
+ ctxt->checkIndex = 0;
+ xmlParseDocTypeDecl(ctxt);
+ if (ctxt->instate == XML_PARSER_EOF)
+ goto done;
+ if (RAW == '[') {
+ ctxt->instate = XML_PARSER_DTD;
+#ifdef DEBUG_PUSH
+ xmlGenericError(xmlGenericErrorContext,
+ "PP: entering DTD\n");
+#endif
+ } else {
+ /*
+ * Create and update the external subset.
+ */
+ ctxt->inSubset = 2;
+ if ((ctxt->sax != NULL) && (!ctxt->disableSAX) &&
+ (ctxt->sax->externalSubset != NULL))
+ ctxt->sax->externalSubset(ctxt->userData,
+ ctxt->intSubName, ctxt->extSubSystem,
+ ctxt->extSubURI);
+ ctxt->inSubset = 0;
+ xmlCleanSpecialAttr(ctxt);
+ ctxt->instate = XML_PARSER_PROLOG;
+#ifdef DEBUG_PUSH
+ xmlGenericError(xmlGenericErrorContext,
+ "PP: entering PROLOG\n");
+#endif
+ }
+ } else if ((cur == '<') && (next == '!') &&
+ (avail < 9)) {
+ goto done;
+ } else {
+ ctxt->instate = XML_PARSER_START_TAG;
+ ctxt->progressive = XML_PARSER_START_TAG;
+ xmlParseGetLasts(ctxt, &lastlt, &lastgt);
+#ifdef DEBUG_PUSH
+ xmlGenericError(xmlGenericErrorContext,
+ "PP: entering START_TAG\n");
+#endif
+ }
+ break;
+ case XML_PARSER_PROLOG:
+ SKIP_BLANKS;
+ if (ctxt->input->buf == NULL)
+ avail = ctxt->input->length - (ctxt->input->cur - ctxt->input->base);
+ else
+ avail = xmlBufUse(ctxt->input->buf->buffer) -
+ (ctxt->input->cur - ctxt->input->base);
+ if (avail < 2)
+ goto done;
+ cur = ctxt->input->cur[0];
+ next = ctxt->input->cur[1];
+ if ((cur == '<') && (next == '?')) {
+ if ((!terminate) &&
+ (xmlParseLookupSequence(ctxt, '?', '>', 0) < 0)) {
+ ctxt->progressive = XML_PARSER_PI;
+ goto done;
+ }
+#ifdef DEBUG_PUSH
+ xmlGenericError(xmlGenericErrorContext,
+ "PP: Parsing PI\n");
+#endif
+ xmlParsePI(ctxt);
+ if (ctxt->instate == XML_PARSER_EOF)
+ goto done;
+ ctxt->instate = XML_PARSER_PROLOG;
+ ctxt->progressive = 1;
+ } else if ((cur == '<') && (next == '!') &&
+ (ctxt->input->cur[2] == '-') && (ctxt->input->cur[3] == '-')) {
+ if ((!terminate) &&
+ (xmlParseLookupSequence(ctxt, '-', '-', '>') < 0)) {
+ ctxt->progressive = XML_PARSER_COMMENT;
+ goto done;
+ }
+#ifdef DEBUG_PUSH
+ xmlGenericError(xmlGenericErrorContext,
+ "PP: Parsing Comment\n");
+#endif
+ xmlParseComment(ctxt);
+ if (ctxt->instate == XML_PARSER_EOF)
+ goto done;
+ ctxt->instate = XML_PARSER_PROLOG;
+ ctxt->progressive = 1;
+ } else if ((cur == '<') && (next == '!') &&
+ (avail < 4)) {
+ goto done;
+ } else {
+ ctxt->instate = XML_PARSER_START_TAG;
+ if (ctxt->progressive == 0)
+ ctxt->progressive = XML_PARSER_START_TAG;
+ xmlParseGetLasts(ctxt, &lastlt, &lastgt);
+#ifdef DEBUG_PUSH
+ xmlGenericError(xmlGenericErrorContext,
+ "PP: entering START_TAG\n");
+#endif
+ }
+ break;
+ case XML_PARSER_EPILOG:
+ SKIP_BLANKS;
+ if (ctxt->input->buf == NULL)
+ avail = ctxt->input->length - (ctxt->input->cur - ctxt->input->base);
+ else
+ avail = xmlBufUse(ctxt->input->buf->buffer) -
+ (ctxt->input->cur - ctxt->input->base);
+ if (avail < 2)
+ goto done;
+ cur = ctxt->input->cur[0];
+ next = ctxt->input->cur[1];
+ if ((cur == '<') && (next == '?')) {
+ if ((!terminate) &&
+ (xmlParseLookupSequence(ctxt, '?', '>', 0) < 0)) {
+ ctxt->progressive = XML_PARSER_PI;
+ goto done;
+ }
+#ifdef DEBUG_PUSH
+ xmlGenericError(xmlGenericErrorContext,
+ "PP: Parsing PI\n");
+#endif
+ xmlParsePI(ctxt);
+ if (ctxt->instate == XML_PARSER_EOF)
+ goto done;
+ ctxt->instate = XML_PARSER_EPILOG;
+ ctxt->progressive = 1;
+ } else if ((cur == '<') && (next == '!') &&
+ (ctxt->input->cur[2] == '-') && (ctxt->input->cur[3] == '-')) {
+ if ((!terminate) &&
+ (xmlParseLookupSequence(ctxt, '-', '-', '>') < 0)) {
+ ctxt->progressive = XML_PARSER_COMMENT;
+ goto done;
+ }
+#ifdef DEBUG_PUSH
+ xmlGenericError(xmlGenericErrorContext,
+ "PP: Parsing Comment\n");
+#endif
+ xmlParseComment(ctxt);
+ if (ctxt->instate == XML_PARSER_EOF)
+ goto done;
+ ctxt->instate = XML_PARSER_EPILOG;
+ ctxt->progressive = 1;
+ } else if ((cur == '<') && (next == '!') &&
+ (avail < 4)) {
+ goto done;
+ } else {
+ xmlFatalErr(ctxt, XML_ERR_DOCUMENT_END, NULL);
+ xmlHaltParser(ctxt);
+#ifdef DEBUG_PUSH
+ xmlGenericError(xmlGenericErrorContext,
+ "PP: entering EOF\n");
+#endif
+ if ((ctxt->sax) && (ctxt->sax->endDocument != NULL))
+ ctxt->sax->endDocument(ctxt->userData);
+ goto done;
+ }
+ break;
+ case XML_PARSER_DTD: {
+ /*
+ * Sorry but progressive parsing of the internal subset
+ * is not expected to be supported. We first check that
+ * the full content of the internal subset is available and
+ * the parsing is launched only at that point.
+ * Internal subset ends up with "']' S? '>'" in an unescaped
+ * section and not in a ']]>' sequence which are conditional
+ * sections (whoever argued to keep that crap in XML deserve
+ * a place in hell !).
+ */
+ int base, i;
+ xmlChar *buf;
+ xmlChar quote = 0;
+ size_t use;
+
+ base = ctxt->input->cur - ctxt->input->base;
+ if (base < 0) return(0);
+ if (ctxt->checkIndex > base)
+ base = ctxt->checkIndex;
+ buf = xmlBufContent(ctxt->input->buf->buffer);
+ use = xmlBufUse(ctxt->input->buf->buffer);
+ for (;(unsigned int) base < use; base++) {
+ if (quote != 0) {
+ if (buf[base] == quote)
+ quote = 0;
+ continue;
+ }
+ if ((quote == 0) && (buf[base] == '<')) {
+ int found = 0;
+ /* special handling of comments */
+ if (((unsigned int) base + 4 < use) &&
+ (buf[base + 1] == '!') &&
+ (buf[base + 2] == '-') &&
+ (buf[base + 3] == '-')) {
+ for (;(unsigned int) base + 3 < use; base++) {
+ if ((buf[base] == '-') &&
+ (buf[base + 1] == '-') &&
+ (buf[base + 2] == '>')) {
+ found = 1;
+ base += 2;
+ break;
+ }
+ }
+ if (!found) {
+#if 0
+ fprintf(stderr, "unfinished comment\n");
+#endif
+ break; /* for */
+ }
+ continue;
+ }
+ }
+ if (buf[base] == '"') {
+ quote = '"';
+ continue;
+ }
+ if (buf[base] == '\'') {
+ quote = '\'';
+ continue;
+ }
+ if (buf[base] == ']') {
+#if 0
+ fprintf(stderr, "%c%c%c%c: ", buf[base],
+ buf[base + 1], buf[base + 2], buf[base + 3]);
+#endif
+ if ((unsigned int) base +1 >= use)
+ break;
+ if (buf[base + 1] == ']') {
+ /* conditional crap, skip both ']' ! */
+ base++;
+ continue;
+ }
+ for (i = 1; (unsigned int) base + i < use; i++) {
+ if (buf[base + i] == '>') {
+#if 0
+ fprintf(stderr, "found\n");
+#endif
+ goto found_end_int_subset;
+ }
+ if (!IS_BLANK_CH(buf[base + i])) {
+#if 0
+ fprintf(stderr, "not found\n");
+#endif
+ goto not_end_of_int_subset;
+ }
+ }
+#if 0
+ fprintf(stderr, "end of stream\n");
+#endif
+ break;
+
+ }
+not_end_of_int_subset:
+ continue; /* for */
+ }
+ /*
+ * We didn't found the end of the Internal subset
+ */
+ if (quote == 0)
+ ctxt->checkIndex = base;
+ else
+ ctxt->checkIndex = 0;
+#ifdef DEBUG_PUSH
+ if (next == 0)
+ xmlGenericError(xmlGenericErrorContext,
+ "PP: lookup of int subset end filed\n");
+#endif
+ goto done;
+
+found_end_int_subset:
+ ctxt->checkIndex = 0;
+ xmlParseInternalSubset(ctxt);
+ if (ctxt->instate == XML_PARSER_EOF)
+ goto done;
+ ctxt->inSubset = 2;
+ if ((ctxt->sax != NULL) && (!ctxt->disableSAX) &&
+ (ctxt->sax->externalSubset != NULL))
+ ctxt->sax->externalSubset(ctxt->userData, ctxt->intSubName,
+ ctxt->extSubSystem, ctxt->extSubURI);
+ ctxt->inSubset = 0;
+ xmlCleanSpecialAttr(ctxt);
+ if (ctxt->instate == XML_PARSER_EOF)
+ goto done;
+ ctxt->instate = XML_PARSER_PROLOG;
+ ctxt->checkIndex = 0;
+#ifdef DEBUG_PUSH
+ xmlGenericError(xmlGenericErrorContext,
+ "PP: entering PROLOG\n");
+#endif
+ break;
+ }
+ case XML_PARSER_COMMENT:
+ xmlGenericError(xmlGenericErrorContext,
+ "PP: internal error, state == COMMENT\n");
+ ctxt->instate = XML_PARSER_CONTENT;
+#ifdef DEBUG_PUSH
+ xmlGenericError(xmlGenericErrorContext,
+ "PP: entering CONTENT\n");
+#endif
+ break;
+ case XML_PARSER_IGNORE:
+ xmlGenericError(xmlGenericErrorContext,
+ "PP: internal error, state == IGNORE");
+ ctxt->instate = XML_PARSER_DTD;
+#ifdef DEBUG_PUSH
+ xmlGenericError(xmlGenericErrorContext,
+ "PP: entering DTD\n");
+#endif
+ break;
+ case XML_PARSER_PI:
+ xmlGenericError(xmlGenericErrorContext,
+ "PP: internal error, state == PI\n");
+ ctxt->instate = XML_PARSER_CONTENT;
+#ifdef DEBUG_PUSH
+ xmlGenericError(xmlGenericErrorContext,
+ "PP: entering CONTENT\n");
+#endif
+ break;
+ case XML_PARSER_ENTITY_DECL:
+ xmlGenericError(xmlGenericErrorContext,
+ "PP: internal error, state == ENTITY_DECL\n");
+ ctxt->instate = XML_PARSER_DTD;
+#ifdef DEBUG_PUSH
+ xmlGenericError(xmlGenericErrorContext,
+ "PP: entering DTD\n");
+#endif
+ break;
+ case XML_PARSER_ENTITY_VALUE:
+ xmlGenericError(xmlGenericErrorContext,
+ "PP: internal error, state == ENTITY_VALUE\n");
+ ctxt->instate = XML_PARSER_CONTENT;
+#ifdef DEBUG_PUSH
+ xmlGenericError(xmlGenericErrorContext,
+ "PP: entering DTD\n");
+#endif
+ break;
+ case XML_PARSER_ATTRIBUTE_VALUE:
+ xmlGenericError(xmlGenericErrorContext,
+ "PP: internal error, state == ATTRIBUTE_VALUE\n");
+ ctxt->instate = XML_PARSER_START_TAG;
+#ifdef DEBUG_PUSH
+ xmlGenericError(xmlGenericErrorContext,
+ "PP: entering START_TAG\n");
+#endif
+ break;
+ case XML_PARSER_SYSTEM_LITERAL:
+ xmlGenericError(xmlGenericErrorContext,
+ "PP: internal error, state == SYSTEM_LITERAL\n");
+ ctxt->instate = XML_PARSER_START_TAG;
+#ifdef DEBUG_PUSH
+ xmlGenericError(xmlGenericErrorContext,
+ "PP: entering START_TAG\n");
+#endif
+ break;
+ case XML_PARSER_PUBLIC_LITERAL:
+ xmlGenericError(xmlGenericErrorContext,
+ "PP: internal error, state == PUBLIC_LITERAL\n");
+ ctxt->instate = XML_PARSER_START_TAG;
+#ifdef DEBUG_PUSH
+ xmlGenericError(xmlGenericErrorContext,
+ "PP: entering START_TAG\n");
+#endif
+ break;
+ }
+ }
+done:
+#ifdef DEBUG_PUSH
+ xmlGenericError(xmlGenericErrorContext, "PP: done %d\n", ret);
+#endif
+ return(ret);
+encoding_error:
+ {
+ char buffer[150];
+
+ snprintf(buffer, 149, "Bytes: 0x%02X 0x%02X 0x%02X 0x%02X\n",
+ ctxt->input->cur[0], ctxt->input->cur[1],
+ ctxt->input->cur[2], ctxt->input->cur[3]);
+ __xmlErrEncoding(ctxt, XML_ERR_INVALID_CHAR,
+ "Input is not proper UTF-8, indicate encoding !\n%s",
+ BAD_CAST buffer, NULL);
+ }
+ return(0);
+}
+
+/**
+ * xmlParseCheckTransition:
+ * @ctxt: an XML parser context
+ * @chunk: a char array
+ * @size: the size in byte of the chunk
+ *
+ * Check depending on the current parser state if the chunk given must be
+ * processed immediately or one need more data to advance on parsing.
+ *
+ * Returns -1 in case of error, 0 if the push is not needed and 1 if needed
+ */
+static int
+xmlParseCheckTransition(xmlParserCtxtPtr ctxt, const char *chunk, int size) {
+ if ((ctxt == NULL) || (chunk == NULL) || (size < 0))
+ return(-1);
+ if (ctxt->instate == XML_PARSER_START_TAG) {
+ if (memchr(chunk, '>', size) != NULL)
+ return(1);
+ return(0);
+ }
+ if (ctxt->progressive == XML_PARSER_COMMENT) {
+ if (memchr(chunk, '>', size) != NULL)
+ return(1);
+ return(0);
+ }
+ if (ctxt->instate == XML_PARSER_CDATA_SECTION) {
+ if (memchr(chunk, '>', size) != NULL)
+ return(1);
+ return(0);
+ }
+ if (ctxt->progressive == XML_PARSER_PI) {
+ if (memchr(chunk, '>', size) != NULL)
+ return(1);
+ return(0);
+ }
+ if (ctxt->instate == XML_PARSER_END_TAG) {
+ if (memchr(chunk, '>', size) != NULL)
+ return(1);
+ return(0);
+ }
+ if ((ctxt->progressive == XML_PARSER_DTD) ||
+ (ctxt->instate == XML_PARSER_DTD)) {
+ if (memchr(chunk, '>', size) != NULL)
+ return(1);
+ return(0);
+ }
+ return(1);
+}
+
+/**
+ * xmlParseChunk:
+ * @ctxt: an XML parser context
+ * @chunk: an char array
+ * @size: the size in byte of the chunk
+ * @terminate: last chunk indicator
+ *
+ * Parse a Chunk of memory
+ *
+ * Returns zero if no error, the xmlParserErrors otherwise.
+ */
+int
+xmlParseChunk(xmlParserCtxtPtr ctxt, const char *chunk, int size,
+ int terminate) {
+ int end_in_lf = 0;
+ int remain = 0;
+ size_t old_avail = 0;
+ size_t avail = 0;
+
+ if (ctxt == NULL)
+ return(XML_ERR_INTERNAL_ERROR);
+ if ((ctxt->errNo != XML_ERR_OK) && (ctxt->disableSAX == 1))
+ return(ctxt->errNo);
+ if (ctxt->instate == XML_PARSER_EOF)
+ return(-1);
+ if (ctxt->instate == XML_PARSER_START)
+ xmlDetectSAX2(ctxt);
+ if ((size > 0) && (chunk != NULL) && (!terminate) &&
+ (chunk[size - 1] == '\r')) {
+ end_in_lf = 1;
+ size--;
+ }
+
+xmldecl_done:
+
+ if ((size > 0) && (chunk != NULL) && (ctxt->input != NULL) &&
+ (ctxt->input->buf != NULL) && (ctxt->instate != XML_PARSER_EOF)) {
+ size_t base = xmlBufGetInputBase(ctxt->input->buf->buffer, ctxt->input);
+ size_t cur = ctxt->input->cur - ctxt->input->base;
+ int res;
+
+ old_avail = xmlBufUse(ctxt->input->buf->buffer);
+ /*
+ * Specific handling if we autodetected an encoding, we should not
+ * push more than the first line ... which depend on the encoding
+ * And only push the rest once the final encoding was detected
+ */
+ if ((ctxt->instate == XML_PARSER_START) && (ctxt->input != NULL) &&
+ (ctxt->input->buf != NULL) && (ctxt->input->buf->encoder != NULL)) {
+ unsigned int len = 45;
+
+ if ((xmlStrcasestr(BAD_CAST ctxt->input->buf->encoder->name,
+ BAD_CAST "UTF-16")) ||
+ (xmlStrcasestr(BAD_CAST ctxt->input->buf->encoder->name,
+ BAD_CAST "UTF16")))
+ len = 90;
+ else if ((xmlStrcasestr(BAD_CAST ctxt->input->buf->encoder->name,
+ BAD_CAST "UCS-4")) ||
+ (xmlStrcasestr(BAD_CAST ctxt->input->buf->encoder->name,
+ BAD_CAST "UCS4")))
+ len = 180;
+
+ if (ctxt->input->buf->rawconsumed < len)
+ len -= ctxt->input->buf->rawconsumed;
+
+ /*
+ * Change size for reading the initial declaration only
+ * if size is greater than len. Otherwise, memmove in xmlBufferAdd
+ * will blindly copy extra bytes from memory.
+ */
+ if ((unsigned int) size > len) {
+ remain = size - len;
+ size = len;
+ } else {
+ remain = 0;
+ }
+ }
+ res = xmlParserInputBufferPush(ctxt->input->buf, size, chunk);
+ if (res < 0) {
+ ctxt->errNo = XML_PARSER_EOF;
+ xmlHaltParser(ctxt);
+ return (XML_PARSER_EOF);
+ }
+ xmlBufSetInputBaseCur(ctxt->input->buf->buffer, ctxt->input, base, cur);
+#ifdef DEBUG_PUSH
+ xmlGenericError(xmlGenericErrorContext, "PP: pushed %d\n", size);
+#endif
+
+ } else if (ctxt->instate != XML_PARSER_EOF) {
+ if ((ctxt->input != NULL) && ctxt->input->buf != NULL) {
+ xmlParserInputBufferPtr in = ctxt->input->buf;
+ if ((in->encoder != NULL) && (in->buffer != NULL) &&
+ (in->raw != NULL)) {
+ int nbchars;
+ size_t base = xmlBufGetInputBase(in->buffer, ctxt->input);
+ size_t current = ctxt->input->cur - ctxt->input->base;
+
+ nbchars = xmlCharEncInput(in, terminate);
+ if (nbchars < 0) {
+ /* TODO 2.6.0 */
+ xmlGenericError(xmlGenericErrorContext,
+ "xmlParseChunk: encoder error\n");
+ return(XML_ERR_INVALID_ENCODING);
+ }
+ xmlBufSetInputBaseCur(in->buffer, ctxt->input, base, current);
+ }
+ }
+ }
+ if (remain != 0) {
+ xmlParseTryOrFinish(ctxt, 0);
+ } else {
+ if ((ctxt->input != NULL) && (ctxt->input->buf != NULL))
+ avail = xmlBufUse(ctxt->input->buf->buffer);
+ /*
+ * Depending on the current state it may not be such
+ * a good idea to try parsing if there is nothing in the chunk
+ * which would be worth doing a parser state transition and we
+ * need to wait for more data
+ */
+ if ((terminate) || (avail > XML_MAX_TEXT_LENGTH) ||
+ (old_avail == 0) || (avail == 0) ||
+ (xmlParseCheckTransition(ctxt,
+ (const char *)&ctxt->input->base[old_avail],
+ avail - old_avail)))
+ xmlParseTryOrFinish(ctxt, terminate);
+ }
+ if (ctxt->instate == XML_PARSER_EOF)
+ return(ctxt->errNo);
+
+ if ((ctxt->input != NULL) &&
+ (((ctxt->input->end - ctxt->input->cur) > XML_MAX_LOOKUP_LIMIT) ||
+ ((ctxt->input->cur - ctxt->input->base) > XML_MAX_LOOKUP_LIMIT)) &&
+ ((ctxt->options & XML_PARSE_HUGE) == 0)) {
+ xmlFatalErr(ctxt, XML_ERR_INTERNAL_ERROR, "Huge input lookup");
+ xmlHaltParser(ctxt);
+ }
+ if ((ctxt->errNo != XML_ERR_OK) && (ctxt->disableSAX == 1))
+ return(ctxt->errNo);
+
+ if (remain != 0) {
+ chunk += size;
+ size = remain;
+ remain = 0;
+ goto xmldecl_done;
+ }
+ if ((end_in_lf == 1) && (ctxt->input != NULL) &&
+ (ctxt->input->buf != NULL)) {
+ size_t base = xmlBufGetInputBase(ctxt->input->buf->buffer,
+ ctxt->input);
+ size_t current = ctxt->input->cur - ctxt->input->base;
+
+ xmlParserInputBufferPush(ctxt->input->buf, 1, "\r");
+
+ xmlBufSetInputBaseCur(ctxt->input->buf->buffer, ctxt->input,
+ base, current);
+ }
+ if (terminate) {
+ /*
+ * Check for termination
+ */
+ int cur_avail = 0;
+
+ if (ctxt->input != NULL) {
+ if (ctxt->input->buf == NULL)
+ cur_avail = ctxt->input->length -
+ (ctxt->input->cur - ctxt->input->base);
+ else
+ cur_avail = xmlBufUse(ctxt->input->buf->buffer) -
+ (ctxt->input->cur - ctxt->input->base);
+ }
+
+ if ((ctxt->instate != XML_PARSER_EOF) &&
+ (ctxt->instate != XML_PARSER_EPILOG)) {
+ xmlFatalErr(ctxt, XML_ERR_DOCUMENT_END, NULL);
+ }
+ if ((ctxt->instate == XML_PARSER_EPILOG) && (cur_avail > 0)) {
+ xmlFatalErr(ctxt, XML_ERR_DOCUMENT_END, NULL);
+ }
+ if (ctxt->instate != XML_PARSER_EOF) {
+ if ((ctxt->sax) && (ctxt->sax->endDocument != NULL))
+ ctxt->sax->endDocument(ctxt->userData);
+ }
+ ctxt->instate = XML_PARSER_EOF;
+ }
+ if (ctxt->wellFormed == 0)
+ return((xmlParserErrors) ctxt->errNo);
+ else
+ return(0);
+}
+
+/************************************************************************
+ * *
+ * I/O front end functions to the parser *
+ * *
+ ************************************************************************/
+
+/**
+ * xmlCreatePushParserCtxt:
+ * @sax: a SAX handler
+ * @user_data: The user data returned on SAX callbacks
+ * @chunk: a pointer to an array of chars
+ * @size: number of chars in the array
+ * @filename: an optional file name or URI
+ *
+ * Create a parser context for using the XML parser in push mode.
+ * If @buffer and @size are non-NULL, the data is used to detect
+ * the encoding. The remaining characters will be parsed so they
+ * don't need to be fed in again through xmlParseChunk.
+ * To allow content encoding detection, @size should be >= 4
+ * The value of @filename is used for fetching external entities
+ * and error/warning reports.
+ *
+ * Returns the new parser context or NULL
+ */
+
+xmlParserCtxtPtr
+xmlCreatePushParserCtxt(xmlSAXHandlerPtr sax, void *user_data,
+ const char *chunk, int size, const char *filename) {
+ xmlParserCtxtPtr ctxt;
+ xmlParserInputPtr inputStream;
+ xmlParserInputBufferPtr buf;
+ xmlCharEncoding enc = XML_CHAR_ENCODING_NONE;
+
+ /*
+ * plug some encoding conversion routines
+ */
+ if ((chunk != NULL) && (size >= 4))
+ enc = xmlDetectCharEncoding((const xmlChar *) chunk, size);
+
+ buf = xmlAllocParserInputBuffer(enc);
+ if (buf == NULL) return(NULL);
+
+ ctxt = xmlNewParserCtxt();
+ if (ctxt == NULL) {
+ xmlErrMemory(NULL, "creating parser: out of memory\n");
+ xmlFreeParserInputBuffer(buf);
+ return(NULL);
+ }
+ ctxt->dictNames = 1;
+ ctxt->pushTab = (void **) xmlMalloc(ctxt->nameMax * 3 * sizeof(xmlChar *));
+ if (ctxt->pushTab == NULL) {
+ xmlErrMemory(ctxt, NULL);
+ xmlFreeParserInputBuffer(buf);
+ xmlFreeParserCtxt(ctxt);
+ return(NULL);
+ }
+ if (sax != NULL) {
+#ifdef LIBXML_SAX1_ENABLED
+ if (ctxt->sax != (xmlSAXHandlerPtr) &xmlDefaultSAXHandler)
+#endif /* LIBXML_SAX1_ENABLED */
+ xmlFree(ctxt->sax);
+ ctxt->sax = (xmlSAXHandlerPtr) xmlMalloc(sizeof(xmlSAXHandler));
+ if (ctxt->sax == NULL) {
+ xmlErrMemory(ctxt, NULL);
+ xmlFreeParserInputBuffer(buf);
+ xmlFreeParserCtxt(ctxt);
+ return(NULL);
+ }
+ memset(ctxt->sax, 0, sizeof(xmlSAXHandler));
+ if (sax->initialized == XML_SAX2_MAGIC)
+ memcpy(ctxt->sax, sax, sizeof(xmlSAXHandler));
+ else
+ memcpy(ctxt->sax, sax, sizeof(xmlSAXHandlerV1));
+ if (user_data != NULL)
+ ctxt->userData = user_data;
+ }
+ if (filename == NULL) {
+ ctxt->directory = NULL;
+ } else {
+ ctxt->directory = xmlParserGetDirectory(filename);
+ }
+
+ inputStream = xmlNewInputStream(ctxt);
+ if (inputStream == NULL) {
+ xmlFreeParserCtxt(ctxt);
+ xmlFreeParserInputBuffer(buf);
+ return(NULL);
+ }
+
+ if (filename == NULL)
+ inputStream->filename = NULL;
+ else {
+ inputStream->filename = (char *)
+ xmlCanonicPath((const xmlChar *) filename);
+ if (inputStream->filename == NULL) {
+ xmlFreeParserCtxt(ctxt);
+ xmlFreeParserInputBuffer(buf);
+ return(NULL);
+ }
+ }
+ inputStream->buf = buf;
+ xmlBufResetInput(inputStream->buf->buffer, inputStream);
+ inputPush(ctxt, inputStream);
+
+ /*
+ * If the caller didn't provide an initial 'chunk' for determining
+ * the encoding, we set the context to XML_CHAR_ENCODING_NONE so
+ * that it can be automatically determined later
+ */
+ if ((size == 0) || (chunk == NULL)) {
+ ctxt->charset = XML_CHAR_ENCODING_NONE;
+ } else if ((ctxt->input != NULL) && (ctxt->input->buf != NULL)) {
+ size_t base = xmlBufGetInputBase(ctxt->input->buf->buffer, ctxt->input);
+ size_t cur = ctxt->input->cur - ctxt->input->base;
+
+ xmlParserInputBufferPush(ctxt->input->buf, size, chunk);
+
+ xmlBufSetInputBaseCur(ctxt->input->buf->buffer, ctxt->input, base, cur);
+#ifdef DEBUG_PUSH
+ xmlGenericError(xmlGenericErrorContext, "PP: pushed %d\n", size);
+#endif
+ }
+
+ if (enc != XML_CHAR_ENCODING_NONE) {
+ xmlSwitchEncoding(ctxt, enc);
+ }
+
+ return(ctxt);
+}
+#endif /* LIBXML_PUSH_ENABLED */
+
+/**
+ * xmlHaltParser:
+ * @ctxt: an XML parser context
+ *
+ * Blocks further parser processing don't override error
+ * for internal use
+ */
+static void
+xmlHaltParser(xmlParserCtxtPtr ctxt) {
+ if (ctxt == NULL)
+ return;
+ ctxt->instate = XML_PARSER_EOF;
+ ctxt->disableSAX = 1;
+ if (ctxt->input != NULL) {
+ /*
+ * in case there was a specific allocation deallocate before
+ * overriding base
+ */
+ if (ctxt->input->free != NULL) {
+ ctxt->input->free((xmlChar *) ctxt->input->base);
+ ctxt->input->free = NULL;
+ }
+ ctxt->input->cur = BAD_CAST"";
+ ctxt->input->base = ctxt->input->cur;
+ }
+}
+
+/**
+ * xmlStopParser:
+ * @ctxt: an XML parser context
+ *
+ * Blocks further parser processing
+ */
+void
+xmlStopParser(xmlParserCtxtPtr ctxt) {
+ if (ctxt == NULL)
+ return;
+ xmlHaltParser(ctxt);
+ ctxt->errNo = XML_ERR_USER_STOP;
+}
+
+/**
+ * xmlCreateIOParserCtxt:
+ * @sax: a SAX handler
+ * @user_data: The user data returned on SAX callbacks
+ * @ioread: an I/O read function
+ * @ioclose: an I/O close function
+ * @ioctx: an I/O handler
+ * @enc: the charset encoding if known
+ *
+ * Create a parser context for using the XML parser with an existing
+ * I/O stream
+ *
+ * Returns the new parser context or NULL
+ */
+xmlParserCtxtPtr
+xmlCreateIOParserCtxt(xmlSAXHandlerPtr sax, void *user_data,
+ xmlInputReadCallback ioread, xmlInputCloseCallback ioclose,
+ void *ioctx, xmlCharEncoding enc) {
+ xmlParserCtxtPtr ctxt;
+ xmlParserInputPtr inputStream;
+ xmlParserInputBufferPtr buf;
+
+ if (ioread == NULL) return(NULL);
+
+ buf = xmlParserInputBufferCreateIO(ioread, ioclose, ioctx, enc);
+ if (buf == NULL) {
+ if (ioclose != NULL)
+ ioclose(ioctx);
+ return (NULL);
+ }
+
+ ctxt = xmlNewParserCtxt();
+ if (ctxt == NULL) {
+ xmlFreeParserInputBuffer(buf);
+ return(NULL);
+ }
+ if (sax != NULL) {
+#ifdef LIBXML_SAX1_ENABLED
+ if (ctxt->sax != (xmlSAXHandlerPtr) &xmlDefaultSAXHandler)
+#endif /* LIBXML_SAX1_ENABLED */
+ xmlFree(ctxt->sax);
+ ctxt->sax = (xmlSAXHandlerPtr) xmlMalloc(sizeof(xmlSAXHandler));
+ if (ctxt->sax == NULL) {
+ xmlErrMemory(ctxt, NULL);
+ xmlFreeParserCtxt(ctxt);
+ return(NULL);
+ }
+ memset(ctxt->sax, 0, sizeof(xmlSAXHandler));
+ if (sax->initialized == XML_SAX2_MAGIC)
+ memcpy(ctxt->sax, sax, sizeof(xmlSAXHandler));
+ else
+ memcpy(ctxt->sax, sax, sizeof(xmlSAXHandlerV1));
+ if (user_data != NULL)
+ ctxt->userData = user_data;
+ }
+
+ inputStream = xmlNewIOInputStream(ctxt, buf, enc);
+ if (inputStream == NULL) {
+ xmlFreeParserCtxt(ctxt);
+ return(NULL);
+ }
+ inputPush(ctxt, inputStream);
+
+ return(ctxt);
+}
+
+#ifdef LIBXML_VALID_ENABLED
+/************************************************************************
+ * *
+ * Front ends when parsing a DTD *
+ * *
+ ************************************************************************/
+
+/**
+ * xmlIOParseDTD:
+ * @sax: the SAX handler block or NULL
+ * @input: an Input Buffer
+ * @enc: the charset encoding if known
+ *
+ * Load and parse a DTD
+ *
+ * Returns the resulting xmlDtdPtr or NULL in case of error.
+ * @input will be freed by the function in any case.
+ */
+
+xmlDtdPtr
+xmlIOParseDTD(xmlSAXHandlerPtr sax, xmlParserInputBufferPtr input,
+ xmlCharEncoding enc) {
+ xmlDtdPtr ret = NULL;
+ xmlParserCtxtPtr ctxt;
+ xmlParserInputPtr pinput = NULL;
+ xmlChar start[4];
+
+ if (input == NULL)
+ return(NULL);
+
+ ctxt = xmlNewParserCtxt();
+ if (ctxt == NULL) {
+ xmlFreeParserInputBuffer(input);
+ return(NULL);
+ }
+
+ /* We are loading a DTD */
+ ctxt->options |= XML_PARSE_DTDLOAD;
+
+ /*
+ * Set-up the SAX context
+ */
+ if (sax != NULL) {
+ if (ctxt->sax != NULL)
+ xmlFree(ctxt->sax);
+ ctxt->sax = sax;
+ ctxt->userData = ctxt;
+ }
+ xmlDetectSAX2(ctxt);
+
+ /*
+ * generate a parser input from the I/O handler
+ */
+
+ pinput = xmlNewIOInputStream(ctxt, input, XML_CHAR_ENCODING_NONE);
+ if (pinput == NULL) {
+ if (sax != NULL) ctxt->sax = NULL;
+ xmlFreeParserInputBuffer(input);
+ xmlFreeParserCtxt(ctxt);
+ return(NULL);
+ }
+
+ /*
+ * plug some encoding conversion routines here.
+ */
+ if (xmlPushInput(ctxt, pinput) < 0) {
+ if (sax != NULL) ctxt->sax = NULL;
+ xmlFreeParserCtxt(ctxt);
+ return(NULL);
+ }
+ if (enc != XML_CHAR_ENCODING_NONE) {
+ xmlSwitchEncoding(ctxt, enc);
+ }
+
+ pinput->filename = NULL;
+ pinput->line = 1;
+ pinput->col = 1;
+ pinput->base = ctxt->input->cur;
+ pinput->cur = ctxt->input->cur;
+ pinput->free = NULL;
+
+ /*
+ * let's parse that entity knowing it's an external subset.
+ */
+ ctxt->inSubset = 2;
+ ctxt->myDoc = xmlNewDoc(BAD_CAST "1.0");
+ if (ctxt->myDoc == NULL) {
+ xmlErrMemory(ctxt, "New Doc failed");
+ return(NULL);
+ }
+ ctxt->myDoc->properties = XML_DOC_INTERNAL;
+ ctxt->myDoc->extSubset = xmlNewDtd(ctxt->myDoc, BAD_CAST "none",
+ BAD_CAST "none", BAD_CAST "none");
+
+ if ((enc == XML_CHAR_ENCODING_NONE) &&
+ ((ctxt->input->end - ctxt->input->cur) >= 4)) {
+ /*
+ * Get the 4 first bytes and decode the charset
+ * if enc != XML_CHAR_ENCODING_NONE
+ * plug some encoding conversion routines.
+ */
+ start[0] = RAW;
+ start[1] = NXT(1);
+ start[2] = NXT(2);
+ start[3] = NXT(3);
+ enc = xmlDetectCharEncoding(start, 4);
+ if (enc != XML_CHAR_ENCODING_NONE) {
+ xmlSwitchEncoding(ctxt, enc);
+ }
+ }
+
+ xmlParseExternalSubset(ctxt, BAD_CAST "none", BAD_CAST "none");
+
+ if (ctxt->myDoc != NULL) {
+ if (ctxt->wellFormed) {
+ ret = ctxt->myDoc->extSubset;
+ ctxt->myDoc->extSubset = NULL;
+ if (ret != NULL) {
+ xmlNodePtr tmp;
+
+ ret->doc = NULL;
+ tmp = ret->children;
+ while (tmp != NULL) {
+ tmp->doc = NULL;
+ tmp = tmp->next;
+ }
+ }
+ } else {
+ ret = NULL;
+ }
+ xmlFreeDoc(ctxt->myDoc);
+ ctxt->myDoc = NULL;
+ }
+ if (sax != NULL) ctxt->sax = NULL;
+ xmlFreeParserCtxt(ctxt);
+
+ return(ret);
+}
+
+/**
+ * xmlSAXParseDTD:
+ * @sax: the SAX handler block
+ * @ExternalID: a NAME* containing the External ID of the DTD
+ * @SystemID: a NAME* containing the URL to the DTD
+ *
+ * Load and parse an external subset.
+ *
+ * Returns the resulting xmlDtdPtr or NULL in case of error.
+ */
+
+xmlDtdPtr
+xmlSAXParseDTD(xmlSAXHandlerPtr sax, const xmlChar *ExternalID,
+ const xmlChar *SystemID) {
+ xmlDtdPtr ret = NULL;
+ xmlParserCtxtPtr ctxt;
+ xmlParserInputPtr input = NULL;
+ xmlCharEncoding enc;
+ xmlChar* systemIdCanonic;
+
+ if ((ExternalID == NULL) && (SystemID == NULL)) return(NULL);
+
+ ctxt = xmlNewParserCtxt();
+ if (ctxt == NULL) {
+ return(NULL);
+ }
+
+ /* We are loading a DTD */
+ ctxt->options |= XML_PARSE_DTDLOAD;
+
+ /*
+ * Set-up the SAX context
+ */
+ if (sax != NULL) {
+ if (ctxt->sax != NULL)
+ xmlFree(ctxt->sax);
+ ctxt->sax = sax;
+ ctxt->userData = ctxt;
+ }
+
+ /*
+ * Canonicalise the system ID
+ */
+ systemIdCanonic = xmlCanonicPath(SystemID);
+ if ((SystemID != NULL) && (systemIdCanonic == NULL)) {
+ xmlFreeParserCtxt(ctxt);
+ return(NULL);
+ }
+
+ /*
+ * Ask the Entity resolver to load the damn thing
+ */
+
+ if ((ctxt->sax != NULL) && (ctxt->sax->resolveEntity != NULL))
+ input = ctxt->sax->resolveEntity(ctxt->userData, ExternalID,
+ systemIdCanonic);
+ if (input == NULL) {
+ if (sax != NULL) ctxt->sax = NULL;
+ xmlFreeParserCtxt(ctxt);
+ if (systemIdCanonic != NULL)
+ xmlFree(systemIdCanonic);
+ return(NULL);
+ }
+
+ /*
+ * plug some encoding conversion routines here.
+ */
+ if (xmlPushInput(ctxt, input) < 0) {
+ if (sax != NULL) ctxt->sax = NULL;
+ xmlFreeParserCtxt(ctxt);
+ if (systemIdCanonic != NULL)
+ xmlFree(systemIdCanonic);
+ return(NULL);
+ }
+ if ((ctxt->input->end - ctxt->input->cur) >= 4) {
+ enc = xmlDetectCharEncoding(ctxt->input->cur, 4);
+ xmlSwitchEncoding(ctxt, enc);
+ }
+
+ if (input->filename == NULL)
+ input->filename = (char *) systemIdCanonic;
+ else
+ xmlFree(systemIdCanonic);
+ input->line = 1;
+ input->col = 1;
+ input->base = ctxt->input->cur;
+ input->cur = ctxt->input->cur;
+ input->free = NULL;
+
+ /*
+ * let's parse that entity knowing it's an external subset.
+ */
+ ctxt->inSubset = 2;
+ ctxt->myDoc = xmlNewDoc(BAD_CAST "1.0");
+ if (ctxt->myDoc == NULL) {
+ xmlErrMemory(ctxt, "New Doc failed");
+ if (sax != NULL) ctxt->sax = NULL;
+ xmlFreeParserCtxt(ctxt);
+ return(NULL);
+ }
+ ctxt->myDoc->properties = XML_DOC_INTERNAL;
+ ctxt->myDoc->extSubset = xmlNewDtd(ctxt->myDoc, BAD_CAST "none",
+ ExternalID, SystemID);
+ xmlParseExternalSubset(ctxt, ExternalID, SystemID);
+
+ if (ctxt->myDoc != NULL) {
+ if (ctxt->wellFormed) {
+ ret = ctxt->myDoc->extSubset;
+ ctxt->myDoc->extSubset = NULL;
+ if (ret != NULL) {
+ xmlNodePtr tmp;
+
+ ret->doc = NULL;
+ tmp = ret->children;
+ while (tmp != NULL) {
+ tmp->doc = NULL;
+ tmp = tmp->next;
+ }
+ }
+ } else {
+ ret = NULL;
+ }
+ xmlFreeDoc(ctxt->myDoc);
+ ctxt->myDoc = NULL;
+ }
+ if (sax != NULL) ctxt->sax = NULL;
+ xmlFreeParserCtxt(ctxt);
+
+ return(ret);
+}
+
+
+/**
+ * xmlParseDTD:
+ * @ExternalID: a NAME* containing the External ID of the DTD
+ * @SystemID: a NAME* containing the URL to the DTD
+ *
+ * Load and parse an external subset.
+ *
+ * Returns the resulting xmlDtdPtr or NULL in case of error.
+ */
+
+xmlDtdPtr
+xmlParseDTD(const xmlChar *ExternalID, const xmlChar *SystemID) {
+ return(xmlSAXParseDTD(NULL, ExternalID, SystemID));
+}
+#endif /* LIBXML_VALID_ENABLED */
+
+/************************************************************************
+ * *
+ * Front ends when parsing an Entity *
+ * *
+ ************************************************************************/
+
+/**
+ * xmlParseCtxtExternalEntity:
+ * @ctx: the existing parsing context
+ * @URL: the URL for the entity to load
+ * @ID: the System ID for the entity to load
+ * @lst: the return value for the set of parsed nodes
+ *
+ * Parse an external general entity within an existing parsing context
+ * An external general parsed entity is well-formed if it matches the
+ * production labeled extParsedEnt.
+ *
+ * [78] extParsedEnt ::= TextDecl? content
+ *
+ * Returns 0 if the entity is well formed, -1 in case of args problem and
+ * the parser error code otherwise
+ */
+
+int
+xmlParseCtxtExternalEntity(xmlParserCtxtPtr ctx, const xmlChar *URL,
+ const xmlChar *ID, xmlNodePtr *lst) {
+ xmlParserCtxtPtr ctxt;
+ xmlDocPtr newDoc;
+ xmlNodePtr newRoot;
+ xmlSAXHandlerPtr oldsax = NULL;
+ int ret = 0;
+ xmlChar start[4];
+ xmlCharEncoding enc;
+
+ if (ctx == NULL) return(-1);
+
+ if (((ctx->depth > 40) && ((ctx->options & XML_PARSE_HUGE) == 0)) ||
+ (ctx->depth > 1024)) {
+ return(XML_ERR_ENTITY_LOOP);
+ }
+
+ if (lst != NULL)
+ *lst = NULL;
+ if ((URL == NULL) && (ID == NULL))
+ return(-1);
+ if (ctx->myDoc == NULL) /* @@ relax but check for dereferences */
+ return(-1);
+
+ ctxt = xmlCreateEntityParserCtxtInternal(URL, ID, NULL, ctx);
+ if (ctxt == NULL) {
+ return(-1);
+ }
+
+ oldsax = ctxt->sax;
+ ctxt->sax = ctx->sax;
+ xmlDetectSAX2(ctxt);
+ newDoc = xmlNewDoc(BAD_CAST "1.0");
+ if (newDoc == NULL) {
+ xmlFreeParserCtxt(ctxt);
+ return(-1);
+ }
+ newDoc->properties = XML_DOC_INTERNAL;
+ if (ctx->myDoc->dict) {
+ newDoc->dict = ctx->myDoc->dict;
+ xmlDictReference(newDoc->dict);
+ }
+ if (ctx->myDoc != NULL) {
+ newDoc->intSubset = ctx->myDoc->intSubset;
+ newDoc->extSubset = ctx->myDoc->extSubset;
+ }
+ if (ctx->myDoc->URL != NULL) {
+ newDoc->URL = xmlStrdup(ctx->myDoc->URL);
+ }
+ newRoot = xmlNewDocNode(newDoc, NULL, BAD_CAST "pseudoroot", NULL);
+ if (newRoot == NULL) {
+ ctxt->sax = oldsax;
+ xmlFreeParserCtxt(ctxt);
+ newDoc->intSubset = NULL;
+ newDoc->extSubset = NULL;
+ xmlFreeDoc(newDoc);
+ return(-1);
+ }
+ xmlAddChild((xmlNodePtr) newDoc, newRoot);
+ nodePush(ctxt, newDoc->children);
+ if (ctx->myDoc == NULL) {
+ ctxt->myDoc = newDoc;
+ } else {
+ ctxt->myDoc = ctx->myDoc;
+ newDoc->children->doc = ctx->myDoc;
+ }
+
+ /*
+ * Get the 4 first bytes and decode the charset
+ * if enc != XML_CHAR_ENCODING_NONE
+ * plug some encoding conversion routines.
+ */
+ GROW
+ if ((ctxt->input->end - ctxt->input->cur) >= 4) {
+ start[0] = RAW;
+ start[1] = NXT(1);
+ start[2] = NXT(2);
+ start[3] = NXT(3);
+ enc = xmlDetectCharEncoding(start, 4);
+ if (enc != XML_CHAR_ENCODING_NONE) {
+ xmlSwitchEncoding(ctxt, enc);
+ }
+ }
+
+ /*
+ * Parse a possible text declaration first
+ */
+ if ((CMP5(CUR_PTR, '<', '?', 'x', 'm', 'l')) && (IS_BLANK_CH(NXT(5)))) {
+ xmlParseTextDecl(ctxt);
+ /*
+ * An XML-1.0 document can't reference an entity not XML-1.0
+ */
+ if ((xmlStrEqual(ctx->version, BAD_CAST "1.0")) &&
+ (!xmlStrEqual(ctxt->input->version, BAD_CAST "1.0"))) {
+ xmlFatalErrMsg(ctxt, XML_ERR_VERSION_MISMATCH,
+ "Version mismatch between document and entity\n");
+ }
+ }
+
+ /*
+ * If the user provided its own SAX callbacks then reuse the
+ * useData callback field, otherwise the expected setup in a
+ * DOM builder is to have userData == ctxt
+ */
+ if (ctx->userData == ctx)
+ ctxt->userData = ctxt;
+ else
+ ctxt->userData = ctx->userData;
+
+ /*
+ * Doing validity checking on chunk doesn't make sense
+ */
+ ctxt->instate = XML_PARSER_CONTENT;
+ ctxt->validate = ctx->validate;
+ ctxt->valid = ctx->valid;
+ ctxt->loadsubset = ctx->loadsubset;
+ ctxt->depth = ctx->depth + 1;
+ ctxt->replaceEntities = ctx->replaceEntities;
+ if (ctxt->validate) {
+ ctxt->vctxt.error = ctx->vctxt.error;
+ ctxt->vctxt.warning = ctx->vctxt.warning;
+ } else {
+ ctxt->vctxt.error = NULL;
+ ctxt->vctxt.warning = NULL;
+ }
+ ctxt->vctxt.nodeTab = NULL;
+ ctxt->vctxt.nodeNr = 0;
+ ctxt->vctxt.nodeMax = 0;
+ ctxt->vctxt.node = NULL;
+ if (ctxt->dict != NULL) xmlDictFree(ctxt->dict);
+ ctxt->dict = ctx->dict;
+ ctxt->str_xml = xmlDictLookup(ctxt->dict, BAD_CAST "xml", 3);
+ ctxt->str_xmlns = xmlDictLookup(ctxt->dict, BAD_CAST "xmlns", 5);
+ ctxt->str_xml_ns = xmlDictLookup(ctxt->dict, XML_XML_NAMESPACE, 36);
+ ctxt->dictNames = ctx->dictNames;
+ ctxt->attsDefault = ctx->attsDefault;
+ ctxt->attsSpecial = ctx->attsSpecial;
+ ctxt->linenumbers = ctx->linenumbers;
+
+ xmlParseContent(ctxt);
+
+ ctx->validate = ctxt->validate;
+ ctx->valid = ctxt->valid;
+ if ((RAW == '<') && (NXT(1) == '/')) {
+ xmlFatalErr(ctxt, XML_ERR_NOT_WELL_BALANCED, NULL);
+ } else if (RAW != 0) {
+ xmlFatalErr(ctxt, XML_ERR_EXTRA_CONTENT, NULL);
+ }
+ if (ctxt->node != newDoc->children) {
+ xmlFatalErr(ctxt, XML_ERR_NOT_WELL_BALANCED, NULL);
+ }
+
+ if (!ctxt->wellFormed) {
+ if (ctxt->errNo == 0)
+ ret = 1;
+ else
+ ret = ctxt->errNo;
+ } else {
+ if (lst != NULL) {
+ xmlNodePtr cur;
+
+ /*
+ * Return the newly created nodeset after unlinking it from
+ * they pseudo parent.
+ */
+ cur = newDoc->children->children;
+ *lst = cur;
+ while (cur != NULL) {
+ cur->parent = NULL;
+ cur = cur->next;
+ }
+ newDoc->children->children = NULL;
+ }
+ ret = 0;
+ }
+ ctxt->sax = oldsax;
+ ctxt->dict = NULL;
+ ctxt->attsDefault = NULL;
+ ctxt->attsSpecial = NULL;
+ xmlFreeParserCtxt(ctxt);
+ newDoc->intSubset = NULL;
+ newDoc->extSubset = NULL;
+ xmlFreeDoc(newDoc);
+
+ return(ret);
+}
+
+/**
+ * xmlParseExternalEntityPrivate:
+ * @doc: the document the chunk pertains to
+ * @oldctxt: the previous parser context if available
+ * @sax: the SAX handler bloc (possibly NULL)
+ * @user_data: The user data returned on SAX callbacks (possibly NULL)
+ * @depth: Used for loop detection, use 0
+ * @URL: the URL for the entity to load
+ * @ID: the System ID for the entity to load
+ * @list: the return value for the set of parsed nodes
+ *
+ * Private version of xmlParseExternalEntity()
+ *
+ * Returns 0 if the entity is well formed, -1 in case of args problem and
+ * the parser error code otherwise
+ */
+
+static xmlParserErrors
+xmlParseExternalEntityPrivate(xmlDocPtr doc, xmlParserCtxtPtr oldctxt,
+ xmlSAXHandlerPtr sax,
+ void *user_data, int depth, const xmlChar *URL,
+ const xmlChar *ID, xmlNodePtr *list) {
+ xmlParserCtxtPtr ctxt;
+ xmlDocPtr newDoc;
+ xmlNodePtr newRoot;
+ xmlSAXHandlerPtr oldsax = NULL;
+ xmlParserErrors ret = XML_ERR_OK;
+ xmlChar start[4];
+ xmlCharEncoding enc;
+
+ if (((depth > 40) &&
+ ((oldctxt == NULL) || (oldctxt->options & XML_PARSE_HUGE) == 0)) ||
+ (depth > 1024)) {
+ return(XML_ERR_ENTITY_LOOP);
+ }
+
+ if (list != NULL)
+ *list = NULL;
+ if ((URL == NULL) && (ID == NULL))
+ return(XML_ERR_INTERNAL_ERROR);
+ if (doc == NULL)
+ return(XML_ERR_INTERNAL_ERROR);
+
+
+ ctxt = xmlCreateEntityParserCtxtInternal(URL, ID, NULL, oldctxt);
+ if (ctxt == NULL) return(XML_WAR_UNDECLARED_ENTITY);
+ ctxt->userData = ctxt;
+ if (oldctxt != NULL) {
+ ctxt->_private = oldctxt->_private;
+ ctxt->loadsubset = oldctxt->loadsubset;
+ ctxt->validate = oldctxt->validate;
+ ctxt->external = oldctxt->external;
+ ctxt->record_info = oldctxt->record_info;
+ ctxt->node_seq.maximum = oldctxt->node_seq.maximum;
+ ctxt->node_seq.length = oldctxt->node_seq.length;
+ ctxt->node_seq.buffer = oldctxt->node_seq.buffer;
+ } else {
+ /*
+ * Doing validity checking on chunk without context
+ * doesn't make sense
+ */
+ ctxt->_private = NULL;
+ ctxt->validate = 0;
+ ctxt->external = 2;
+ ctxt->loadsubset = 0;
+ }
+ if (sax != NULL) {
+ oldsax = ctxt->sax;
+ ctxt->sax = sax;
+ if (user_data != NULL)
+ ctxt->userData = user_data;
+ }
+ xmlDetectSAX2(ctxt);
+ newDoc = xmlNewDoc(BAD_CAST "1.0");
+ if (newDoc == NULL) {
+ ctxt->node_seq.maximum = 0;
+ ctxt->node_seq.length = 0;
+ ctxt->node_seq.buffer = NULL;
+ xmlFreeParserCtxt(ctxt);
+ return(XML_ERR_INTERNAL_ERROR);
+ }
+ newDoc->properties = XML_DOC_INTERNAL;
+ newDoc->intSubset = doc->intSubset;
+ newDoc->extSubset = doc->extSubset;
+ newDoc->dict = doc->dict;
+ xmlDictReference(newDoc->dict);
+
+ if (doc->URL != NULL) {
+ newDoc->URL = xmlStrdup(doc->URL);
+ }
+ newRoot = xmlNewDocNode(newDoc, NULL, BAD_CAST "pseudoroot", NULL);
+ if (newRoot == NULL) {
+ if (sax != NULL)
+ ctxt->sax = oldsax;
+ ctxt->node_seq.maximum = 0;
+ ctxt->node_seq.length = 0;
+ ctxt->node_seq.buffer = NULL;
+ xmlFreeParserCtxt(ctxt);
+ newDoc->intSubset = NULL;
+ newDoc->extSubset = NULL;
+ xmlFreeDoc(newDoc);
+ return(XML_ERR_INTERNAL_ERROR);
+ }
+ xmlAddChild((xmlNodePtr) newDoc, newRoot);
+ nodePush(ctxt, newDoc->children);
+ ctxt->myDoc = doc;
+ newRoot->doc = doc;
+
+ /*
+ * Get the 4 first bytes and decode the charset
+ * if enc != XML_CHAR_ENCODING_NONE
+ * plug some encoding conversion routines.
+ */
+ GROW;
+ if ((ctxt->input->end - ctxt->input->cur) >= 4) {
+ start[0] = RAW;
+ start[1] = NXT(1);
+ start[2] = NXT(2);
+ start[3] = NXT(3);
+ enc = xmlDetectCharEncoding(start, 4);
+ if (enc != XML_CHAR_ENCODING_NONE) {
+ xmlSwitchEncoding(ctxt, enc);
+ }
+ }
+
+ /*
+ * Parse a possible text declaration first
+ */
+ if ((CMP5(CUR_PTR, '<', '?', 'x', 'm', 'l')) && (IS_BLANK_CH(NXT(5)))) {
+ xmlParseTextDecl(ctxt);
+ }
+
+ ctxt->instate = XML_PARSER_CONTENT;
+ ctxt->depth = depth;
+
+ xmlParseContent(ctxt);
+
+ if ((RAW == '<') && (NXT(1) == '/')) {
+ xmlFatalErr(ctxt, XML_ERR_NOT_WELL_BALANCED, NULL);
+ } else if (RAW != 0) {
+ xmlFatalErr(ctxt, XML_ERR_EXTRA_CONTENT, NULL);
+ }
+ if (ctxt->node != newDoc->children) {
+ xmlFatalErr(ctxt, XML_ERR_NOT_WELL_BALANCED, NULL);
+ }
+
+ if (!ctxt->wellFormed) {
+ if (ctxt->errNo == 0)
+ ret = XML_ERR_INTERNAL_ERROR;
+ else
+ ret = (xmlParserErrors)ctxt->errNo;
+ } else {
+ if (list != NULL) {
+ xmlNodePtr cur;
+
+ /*
+ * Return the newly created nodeset after unlinking it from
+ * they pseudo parent.
+ */
+ cur = newDoc->children->children;
+ *list = cur;
+ while (cur != NULL) {
+ cur->parent = NULL;
+ cur = cur->next;
+ }
+ newDoc->children->children = NULL;
+ }
+ ret = XML_ERR_OK;
+ }
+
+ /*
+ * Record in the parent context the number of entities replacement
+ * done when parsing that reference.
+ */
+ if (oldctxt != NULL)
+ oldctxt->nbentities += ctxt->nbentities;
+
+ /*
+ * Also record the size of the entity parsed
+ */
+ if (ctxt->input != NULL && oldctxt != NULL) {
+ oldctxt->sizeentities += ctxt->input->consumed;
+ oldctxt->sizeentities += (ctxt->input->cur - ctxt->input->base);
+ }
+ /*
+ * And record the last error if any
+ */
+ if (ctxt->lastError.code != XML_ERR_OK)
+ xmlCopyError(&ctxt->lastError, &oldctxt->lastError);
+
+ if (sax != NULL)
+ ctxt->sax = oldsax;
+ if (oldctxt != NULL) {
+ oldctxt->node_seq.maximum = ctxt->node_seq.maximum;
+ oldctxt->node_seq.length = ctxt->node_seq.length;
+ oldctxt->node_seq.buffer = ctxt->node_seq.buffer;
+ }
+ ctxt->node_seq.maximum = 0;
+ ctxt->node_seq.length = 0;
+ ctxt->node_seq.buffer = NULL;
+ xmlFreeParserCtxt(ctxt);
+ newDoc->intSubset = NULL;
+ newDoc->extSubset = NULL;
+ xmlFreeDoc(newDoc);
+
+ return(ret);
+}
+
+#ifdef LIBXML_SAX1_ENABLED
+/**
+ * xmlParseExternalEntity:
+ * @doc: the document the chunk pertains to
+ * @sax: the SAX handler bloc (possibly NULL)
+ * @user_data: The user data returned on SAX callbacks (possibly NULL)
+ * @depth: Used for loop detection, use 0
+ * @URL: the URL for the entity to load
+ * @ID: the System ID for the entity to load
+ * @lst: the return value for the set of parsed nodes
+ *
+ * Parse an external general entity
+ * An external general parsed entity is well-formed if it matches the
+ * production labeled extParsedEnt.
+ *
+ * [78] extParsedEnt ::= TextDecl? content
+ *
+ * Returns 0 if the entity is well formed, -1 in case of args problem and
+ * the parser error code otherwise
+ */
+
+int
+xmlParseExternalEntity(xmlDocPtr doc, xmlSAXHandlerPtr sax, void *user_data,
+ int depth, const xmlChar *URL, const xmlChar *ID, xmlNodePtr *lst) {
+ return(xmlParseExternalEntityPrivate(doc, NULL, sax, user_data, depth, URL,
+ ID, lst));
+}
+
+/**
+ * xmlParseBalancedChunkMemory:
+ * @doc: the document the chunk pertains to
+ * @sax: the SAX handler bloc (possibly NULL)
+ * @user_data: The user data returned on SAX callbacks (possibly NULL)
+ * @depth: Used for loop detection, use 0
+ * @string: the input string in UTF8 or ISO-Latin (zero terminated)
+ * @lst: the return value for the set of parsed nodes
+ *
+ * Parse a well-balanced chunk of an XML document
+ * called by the parser
+ * The allowed sequence for the Well Balanced Chunk is the one defined by
+ * the content production in the XML grammar:
+ *
+ * [43] content ::= (element | CharData | Reference | CDSect | PI | Comment)*
+ *
+ * Returns 0 if the chunk is well balanced, -1 in case of args problem and
+ * the parser error code otherwise
+ */
+
+int
+xmlParseBalancedChunkMemory(xmlDocPtr doc, xmlSAXHandlerPtr sax,
+ void *user_data, int depth, const xmlChar *string, xmlNodePtr *lst) {
+ return xmlParseBalancedChunkMemoryRecover( doc, sax, user_data,
+ depth, string, lst, 0 );
+}
+#endif /* LIBXML_SAX1_ENABLED */
+
+/**
+ * xmlParseBalancedChunkMemoryInternal:
+ * @oldctxt: the existing parsing context
+ * @string: the input string in UTF8 or ISO-Latin (zero terminated)
+ * @user_data: the user data field for the parser context
+ * @lst: the return value for the set of parsed nodes
+ *
+ *
+ * Parse a well-balanced chunk of an XML document
+ * called by the parser
+ * The allowed sequence for the Well Balanced Chunk is the one defined by
+ * the content production in the XML grammar:
+ *
+ * [43] content ::= (element | CharData | Reference | CDSect | PI | Comment)*
+ *
+ * Returns XML_ERR_OK if the chunk is well balanced, and the parser
+ * error code otherwise
+ *
+ * In case recover is set to 1, the nodelist will not be empty even if
+ * the parsed chunk is not well balanced.
+ */
+static xmlParserErrors
+xmlParseBalancedChunkMemoryInternal(xmlParserCtxtPtr oldctxt,
+ const xmlChar *string, void *user_data, xmlNodePtr *lst) {
+ xmlParserCtxtPtr ctxt;
+ xmlDocPtr newDoc = NULL;
+ xmlNodePtr newRoot;
+ xmlSAXHandlerPtr oldsax = NULL;
+ xmlNodePtr content = NULL;
+ xmlNodePtr last = NULL;
+ int size;
+ xmlParserErrors ret = XML_ERR_OK;
+#ifdef SAX2
+ int i;
+#endif
+
+ if (((oldctxt->depth > 40) && ((oldctxt->options & XML_PARSE_HUGE) == 0)) ||
+ (oldctxt->depth > 1024)) {
+ return(XML_ERR_ENTITY_LOOP);
+ }
+
+
+ if (lst != NULL)
+ *lst = NULL;
+ if (string == NULL)
+ return(XML_ERR_INTERNAL_ERROR);
+
+ size = xmlStrlen(string);
+
+ ctxt = xmlCreateMemoryParserCtxt((char *) string, size);
+ if (ctxt == NULL) return(XML_WAR_UNDECLARED_ENTITY);
+ if (user_data != NULL)
+ ctxt->userData = user_data;
+ else
+ ctxt->userData = ctxt;
+ if (ctxt->dict != NULL) xmlDictFree(ctxt->dict);
+ ctxt->dict = oldctxt->dict;
+ ctxt->str_xml = xmlDictLookup(ctxt->dict, BAD_CAST "xml", 3);
+ ctxt->str_xmlns = xmlDictLookup(ctxt->dict, BAD_CAST "xmlns", 5);
+ ctxt->str_xml_ns = xmlDictLookup(ctxt->dict, XML_XML_NAMESPACE, 36);
+
+#ifdef SAX2
+ /* propagate namespaces down the entity */
+ for (i = 0;i < oldctxt->nsNr;i += 2) {
+ nsPush(ctxt, oldctxt->nsTab[i], oldctxt->nsTab[i+1]);
+ }
+#endif
+
+ oldsax = ctxt->sax;
+ ctxt->sax = oldctxt->sax;
+ xmlDetectSAX2(ctxt);
+ ctxt->replaceEntities = oldctxt->replaceEntities;
+ ctxt->options = oldctxt->options;
+
+ ctxt->_private = oldctxt->_private;
+ if (oldctxt->myDoc == NULL) {
+ newDoc = xmlNewDoc(BAD_CAST "1.0");
+ if (newDoc == NULL) {
+ ctxt->sax = oldsax;
+ ctxt->dict = NULL;
+ xmlFreeParserCtxt(ctxt);
+ return(XML_ERR_INTERNAL_ERROR);
+ }
+ newDoc->properties = XML_DOC_INTERNAL;
+ newDoc->dict = ctxt->dict;
+ xmlDictReference(newDoc->dict);
+ ctxt->myDoc = newDoc;
+ } else {
+ ctxt->myDoc = oldctxt->myDoc;
+ content = ctxt->myDoc->children;
+ last = ctxt->myDoc->last;
+ }
+ newRoot = xmlNewDocNode(ctxt->myDoc, NULL, BAD_CAST "pseudoroot", NULL);
+ if (newRoot == NULL) {
+ ctxt->sax = oldsax;
+ ctxt->dict = NULL;
+ xmlFreeParserCtxt(ctxt);
+ if (newDoc != NULL) {
+ xmlFreeDoc(newDoc);
+ }
+ return(XML_ERR_INTERNAL_ERROR);
+ }
+ ctxt->myDoc->children = NULL;
+ ctxt->myDoc->last = NULL;
+ xmlAddChild((xmlNodePtr) ctxt->myDoc, newRoot);
+ nodePush(ctxt, ctxt->myDoc->children);
+ ctxt->instate = XML_PARSER_CONTENT;
+ ctxt->depth = oldctxt->depth + 1;
+
+ ctxt->validate = 0;
+ ctxt->loadsubset = oldctxt->loadsubset;
+ if ((oldctxt->validate) || (oldctxt->replaceEntities != 0)) {
+ /*
+ * ID/IDREF registration will be done in xmlValidateElement below
+ */
+ ctxt->loadsubset |= XML_SKIP_IDS;
+ }
+ ctxt->dictNames = oldctxt->dictNames;
+ ctxt->attsDefault = oldctxt->attsDefault;
+ ctxt->attsSpecial = oldctxt->attsSpecial;
+
+ xmlParseContent(ctxt);
+ if ((RAW == '<') && (NXT(1) == '/')) {
+ xmlFatalErr(ctxt, XML_ERR_NOT_WELL_BALANCED, NULL);
+ } else if (RAW != 0) {
+ xmlFatalErr(ctxt, XML_ERR_EXTRA_CONTENT, NULL);
+ }
+ if (ctxt->node != ctxt->myDoc->children) {
+ xmlFatalErr(ctxt, XML_ERR_NOT_WELL_BALANCED, NULL);
+ }
+
+ if (!ctxt->wellFormed) {
+ if (ctxt->errNo == 0)
+ ret = XML_ERR_INTERNAL_ERROR;
+ else
+ ret = (xmlParserErrors)ctxt->errNo;
+ } else {
+ ret = XML_ERR_OK;
+ }
+
+ if ((lst != NULL) && (ret == XML_ERR_OK)) {
+ xmlNodePtr cur;
+
+ /*
+ * Return the newly created nodeset after unlinking it from
+ * they pseudo parent.
+ */
+ cur = ctxt->myDoc->children->children;
+ *lst = cur;
+ while (cur != NULL) {
+#ifdef LIBXML_VALID_ENABLED
+ if ((oldctxt->validate) && (oldctxt->wellFormed) &&
+ (oldctxt->myDoc) && (oldctxt->myDoc->intSubset) &&
+ (cur->type == XML_ELEMENT_NODE)) {
+ oldctxt->valid &= xmlValidateElement(&oldctxt->vctxt,
+ oldctxt->myDoc, cur);
+ }
+#endif /* LIBXML_VALID_ENABLED */
+ cur->parent = NULL;
+ cur = cur->next;
+ }
+ ctxt->myDoc->children->children = NULL;
+ }
+ if (ctxt->myDoc != NULL) {
+ xmlFreeNode(ctxt->myDoc->children);
+ ctxt->myDoc->children = content;
+ ctxt->myDoc->last = last;
+ }
+
+ /*
+ * Record in the parent context the number of entities replacement
+ * done when parsing that reference.
+ */
+ if (oldctxt != NULL)
+ oldctxt->nbentities += ctxt->nbentities;
+
+ /*
+ * Also record the last error if any
+ */
+ if (ctxt->lastError.code != XML_ERR_OK)
+ xmlCopyError(&ctxt->lastError, &oldctxt->lastError);
+
+ ctxt->sax = oldsax;
+ ctxt->dict = NULL;
+ ctxt->attsDefault = NULL;
+ ctxt->attsSpecial = NULL;
+ xmlFreeParserCtxt(ctxt);
+ if (newDoc != NULL) {
+ xmlFreeDoc(newDoc);
+ }
+
+ return(ret);
+}
+
+/**
+ * xmlParseInNodeContext:
+ * @node: the context node
+ * @data: the input string
+ * @datalen: the input string length in bytes
+ * @options: a combination of xmlParserOption
+ * @lst: the return value for the set of parsed nodes
+ *
+ * Parse a well-balanced chunk of an XML document
+ * within the context (DTD, namespaces, etc ...) of the given node.
+ *
+ * The allowed sequence for the data is a Well Balanced Chunk defined by
+ * the content production in the XML grammar:
+ *
+ * [43] content ::= (element | CharData | Reference | CDSect | PI | Comment)*
+ *
+ * Returns XML_ERR_OK if the chunk is well balanced, and the parser
+ * error code otherwise
+ */
+xmlParserErrors
+xmlParseInNodeContext(xmlNodePtr node, const char *data, int datalen,
+ int options, xmlNodePtr *lst) {
+#ifdef SAX2
+ xmlParserCtxtPtr ctxt;
+ xmlDocPtr doc = NULL;
+ xmlNodePtr fake, cur;
+ int nsnr = 0;
+
+ xmlParserErrors ret = XML_ERR_OK;
+
+ /*
+ * check all input parameters, grab the document
+ */
+ if ((lst == NULL) || (node == NULL) || (data == NULL) || (datalen < 0))
+ return(XML_ERR_INTERNAL_ERROR);
+ switch (node->type) {
+ case XML_ELEMENT_NODE:
+ case XML_ATTRIBUTE_NODE:
+ case XML_TEXT_NODE:
+ case XML_CDATA_SECTION_NODE:
+ case XML_ENTITY_REF_NODE:
+ case XML_PI_NODE:
+ case XML_COMMENT_NODE:
+ case XML_DOCUMENT_NODE:
+ case XML_HTML_DOCUMENT_NODE:
+ break;
+ default:
+ return(XML_ERR_INTERNAL_ERROR);
+
+ }
+ while ((node != NULL) && (node->type != XML_ELEMENT_NODE) &&
+ (node->type != XML_DOCUMENT_NODE) &&
+ (node->type != XML_HTML_DOCUMENT_NODE))
+ node = node->parent;
+ if (node == NULL)
+ return(XML_ERR_INTERNAL_ERROR);
+ if (node->type == XML_ELEMENT_NODE)
+ doc = node->doc;
+ else
+ doc = (xmlDocPtr) node;
+ if (doc == NULL)
+ return(XML_ERR_INTERNAL_ERROR);
+
+ /*
+ * allocate a context and set-up everything not related to the
+ * node position in the tree
+ */
+ if (doc->type == XML_DOCUMENT_NODE)
+ ctxt = xmlCreateMemoryParserCtxt((char *) data, datalen);
+#ifdef LIBXML_HTML_ENABLED
+ else if (doc->type == XML_HTML_DOCUMENT_NODE) {
+ ctxt = htmlCreateMemoryParserCtxt((char *) data, datalen);
+ /*
+ * When parsing in context, it makes no sense to add implied
+ * elements like html/body/etc...
+ */
+ options |= HTML_PARSE_NOIMPLIED;
+ }
+#endif
+ else
+ return(XML_ERR_INTERNAL_ERROR);
+
+ if (ctxt == NULL)
+ return(XML_ERR_NO_MEMORY);
+
+ /*
+ * Use input doc's dict if present, else assure XML_PARSE_NODICT is set.
+ * We need a dictionary for xmlDetectSAX2, so if there's no doc dict
+ * we must wait until the last moment to free the original one.
+ */
+ if (doc->dict != NULL) {
+ if (ctxt->dict != NULL)
+ xmlDictFree(ctxt->dict);
+ ctxt->dict = doc->dict;
+ } else
+ options |= XML_PARSE_NODICT;
+
+ if (doc->encoding != NULL) {
+ xmlCharEncodingHandlerPtr hdlr;
+
+ if (ctxt->encoding != NULL)
+ xmlFree((xmlChar *) ctxt->encoding);
+ ctxt->encoding = xmlStrdup((const xmlChar *) doc->encoding);
+
+ hdlr = xmlFindCharEncodingHandler((const char *) doc->encoding);
+ if (hdlr != NULL) {
+ xmlSwitchToEncoding(ctxt, hdlr);
+ } else {
+ return(XML_ERR_UNSUPPORTED_ENCODING);
+ }
+ }
+
+ xmlCtxtUseOptionsInternal(ctxt, options, NULL);
+ xmlDetectSAX2(ctxt);
+ ctxt->myDoc = doc;
+ /* parsing in context, i.e. as within existing content */
+ ctxt->instate = XML_PARSER_CONTENT;
+
+ fake = xmlNewComment(NULL);
+ if (fake == NULL) {
+ xmlFreeParserCtxt(ctxt);
+ return(XML_ERR_NO_MEMORY);
+ }
+ xmlAddChild(node, fake);
+
+ if (node->type == XML_ELEMENT_NODE) {
+ nodePush(ctxt, node);
+ /*
+ * initialize the SAX2 namespaces stack
+ */
+ cur = node;
+ while ((cur != NULL) && (cur->type == XML_ELEMENT_NODE)) {
+ xmlNsPtr ns = cur->nsDef;
+ const xmlChar *iprefix, *ihref;
+
+ while (ns != NULL) {
+ if (ctxt->dict) {
+ iprefix = xmlDictLookup(ctxt->dict, ns->prefix, -1);
+ ihref = xmlDictLookup(ctxt->dict, ns->href, -1);
+ } else {
+ iprefix = ns->prefix;
+ ihref = ns->href;
+ }
+
+ if (xmlGetNamespace(ctxt, iprefix) == NULL) {
+ nsPush(ctxt, iprefix, ihref);
+ nsnr++;
+ }
+ ns = ns->next;
+ }
+ cur = cur->parent;
+ }
+ }
+
+ if ((ctxt->validate) || (ctxt->replaceEntities != 0)) {
+ /*
+ * ID/IDREF registration will be done in xmlValidateElement below
+ */
+ ctxt->loadsubset |= XML_SKIP_IDS;
+ }
+
+#ifdef LIBXML_HTML_ENABLED
+ if (doc->type == XML_HTML_DOCUMENT_NODE)
+ __htmlParseContent(ctxt);
+ else
+#endif
+ xmlParseContent(ctxt);
+
+ nsPop(ctxt, nsnr);
+ if ((RAW == '<') && (NXT(1) == '/')) {
+ xmlFatalErr(ctxt, XML_ERR_NOT_WELL_BALANCED, NULL);
+ } else if (RAW != 0) {
+ xmlFatalErr(ctxt, XML_ERR_EXTRA_CONTENT, NULL);
+ }
+ if ((ctxt->node != NULL) && (ctxt->node != node)) {
+ xmlFatalErr(ctxt, XML_ERR_NOT_WELL_BALANCED, NULL);
+ ctxt->wellFormed = 0;
+ }
+
+ if (!ctxt->wellFormed) {
+ if (ctxt->errNo == 0)
+ ret = XML_ERR_INTERNAL_ERROR;
+ else
+ ret = (xmlParserErrors)ctxt->errNo;
+ } else {
+ ret = XML_ERR_OK;
+ }
+
+ /*
+ * Return the newly created nodeset after unlinking it from
+ * the pseudo sibling.
+ */
+
+ cur = fake->next;
+ fake->next = NULL;
+ node->last = fake;
+
+ if (cur != NULL) {
+ cur->prev = NULL;
+ }
+
+ *lst = cur;
+
+ while (cur != NULL) {
+ cur->parent = NULL;
+ cur = cur->next;
+ }
+
+ xmlUnlinkNode(fake);
+ xmlFreeNode(fake);
+
+
+ if (ret != XML_ERR_OK) {
+ xmlFreeNodeList(*lst);
+ *lst = NULL;
+ }
+
+ if (doc->dict != NULL)
+ ctxt->dict = NULL;
+ xmlFreeParserCtxt(ctxt);
+
+ return(ret);
+#else /* !SAX2 */
+ return(XML_ERR_INTERNAL_ERROR);
+#endif
+}
+
+#ifdef LIBXML_SAX1_ENABLED
+/**
+ * xmlParseBalancedChunkMemoryRecover:
+ * @doc: the document the chunk pertains to
+ * @sax: the SAX handler bloc (possibly NULL)
+ * @user_data: The user data returned on SAX callbacks (possibly NULL)
+ * @depth: Used for loop detection, use 0
+ * @string: the input string in UTF8 or ISO-Latin (zero terminated)
+ * @lst: the return value for the set of parsed nodes
+ * @recover: return nodes even if the data is broken (use 0)
+ *
+ *
+ * Parse a well-balanced chunk of an XML document
+ * called by the parser
+ * The allowed sequence for the Well Balanced Chunk is the one defined by
+ * the content production in the XML grammar:
+ *
+ * [43] content ::= (element | CharData | Reference | CDSect | PI | Comment)*
+ *
+ * Returns 0 if the chunk is well balanced, -1 in case of args problem and
+ * the parser error code otherwise
+ *
+ * In case recover is set to 1, the nodelist will not be empty even if
+ * the parsed chunk is not well balanced, assuming the parsing succeeded to
+ * some extent.
+ */
+int
+xmlParseBalancedChunkMemoryRecover(xmlDocPtr doc, xmlSAXHandlerPtr sax,
+ void *user_data, int depth, const xmlChar *string, xmlNodePtr *lst,
+ int recover) {
+ xmlParserCtxtPtr ctxt;
+ xmlDocPtr newDoc;
+ xmlSAXHandlerPtr oldsax = NULL;
+ xmlNodePtr content, newRoot;
+ int size;
+ int ret = 0;
+
+ if (depth > 40) {
+ return(XML_ERR_ENTITY_LOOP);
+ }
+
+
+ if (lst != NULL)
+ *lst = NULL;
+ if (string == NULL)
+ return(-1);
+
+ size = xmlStrlen(string);
+
+ ctxt = xmlCreateMemoryParserCtxt((char *) string, size);
+ if (ctxt == NULL) return(-1);
+ ctxt->userData = ctxt;
+ if (sax != NULL) {
+ oldsax = ctxt->sax;
+ ctxt->sax = sax;
+ if (user_data != NULL)
+ ctxt->userData = user_data;
+ }
+ newDoc = xmlNewDoc(BAD_CAST "1.0");
+ if (newDoc == NULL) {
+ xmlFreeParserCtxt(ctxt);
+ return(-1);
+ }
+ newDoc->properties = XML_DOC_INTERNAL;
+ if ((doc != NULL) && (doc->dict != NULL)) {
+ xmlDictFree(ctxt->dict);
+ ctxt->dict = doc->dict;
+ xmlDictReference(ctxt->dict);
+ ctxt->str_xml = xmlDictLookup(ctxt->dict, BAD_CAST "xml", 3);
+ ctxt->str_xmlns = xmlDictLookup(ctxt->dict, BAD_CAST "xmlns", 5);
+ ctxt->str_xml_ns = xmlDictLookup(ctxt->dict, XML_XML_NAMESPACE, 36);
+ ctxt->dictNames = 1;
+ } else {
+ xmlCtxtUseOptionsInternal(ctxt, XML_PARSE_NODICT, NULL);
+ }
+ if (doc != NULL) {
+ newDoc->intSubset = doc->intSubset;
+ newDoc->extSubset = doc->extSubset;
+ }
+ newRoot = xmlNewDocNode(newDoc, NULL, BAD_CAST "pseudoroot", NULL);
+ if (newRoot == NULL) {
+ if (sax != NULL)
+ ctxt->sax = oldsax;
+ xmlFreeParserCtxt(ctxt);
+ newDoc->intSubset = NULL;
+ newDoc->extSubset = NULL;
+ xmlFreeDoc(newDoc);
+ return(-1);
+ }
+ xmlAddChild((xmlNodePtr) newDoc, newRoot);
+ nodePush(ctxt, newRoot);
+ if (doc == NULL) {
+ ctxt->myDoc = newDoc;
+ } else {
+ ctxt->myDoc = newDoc;
+ newDoc->children->doc = doc;
+ /* Ensure that doc has XML spec namespace */
+ xmlSearchNsByHref(doc, (xmlNodePtr)doc, XML_XML_NAMESPACE);
+ newDoc->oldNs = doc->oldNs;
+ }
+ ctxt->instate = XML_PARSER_CONTENT;
+ ctxt->depth = depth;
+
+ /*
+ * Doing validity checking on chunk doesn't make sense
+ */
+ ctxt->validate = 0;
+ ctxt->loadsubset = 0;
+ xmlDetectSAX2(ctxt);
+
+ if ( doc != NULL ){
+ content = doc->children;
+ doc->children = NULL;
+ xmlParseContent(ctxt);
+ doc->children = content;
+ }
+ else {
+ xmlParseContent(ctxt);
+ }
+ if ((RAW == '<') && (NXT(1) == '/')) {
+ xmlFatalErr(ctxt, XML_ERR_NOT_WELL_BALANCED, NULL);
+ } else if (RAW != 0) {
+ xmlFatalErr(ctxt, XML_ERR_EXTRA_CONTENT, NULL);
+ }
+ if (ctxt->node != newDoc->children) {
+ xmlFatalErr(ctxt, XML_ERR_NOT_WELL_BALANCED, NULL);
+ }
+
+ if (!ctxt->wellFormed) {
+ if (ctxt->errNo == 0)
+ ret = 1;
+ else
+ ret = ctxt->errNo;
+ } else {
+ ret = 0;
+ }
+
+ if ((lst != NULL) && ((ret == 0) || (recover == 1))) {
+ xmlNodePtr cur;
+
+ /*
+ * Return the newly created nodeset after unlinking it from
+ * they pseudo parent.
+ */
+ cur = newDoc->children->children;
+ *lst = cur;
+ while (cur != NULL) {
+ xmlSetTreeDoc(cur, doc);
+ cur->parent = NULL;
+ cur = cur->next;
+ }
+ newDoc->children->children = NULL;
+ }
+
+ if (sax != NULL)
+ ctxt->sax = oldsax;
+ xmlFreeParserCtxt(ctxt);
+ newDoc->intSubset = NULL;
+ newDoc->extSubset = NULL;
+ newDoc->oldNs = NULL;
+ xmlFreeDoc(newDoc);
+
+ return(ret);
+}
+
+/**
+ * xmlSAXParseEntity:
+ * @sax: the SAX handler block
+ * @filename: the filename
+ *
+ * parse an XML external entity out of context and build a tree.
+ * It use the given SAX function block to handle the parsing callback.
+ * If sax is NULL, fallback to the default DOM tree building routines.
+ *
+ * [78] extParsedEnt ::= TextDecl? content
+ *
+ * This correspond to a "Well Balanced" chunk
+ *
+ * Returns the resulting document tree
+ */
+
+xmlDocPtr
+xmlSAXParseEntity(xmlSAXHandlerPtr sax, const char *filename) {
+ xmlDocPtr ret;
+ xmlParserCtxtPtr ctxt;
+
+ ctxt = xmlCreateFileParserCtxt(filename);
+ if (ctxt == NULL) {
+ return(NULL);
+ }
+ if (sax != NULL) {
+ if (ctxt->sax != NULL)
+ xmlFree(ctxt->sax);
+ ctxt->sax = sax;
+ ctxt->userData = NULL;
+ }
+
+ xmlParseExtParsedEnt(ctxt);
+
+ if (ctxt->wellFormed)
+ ret = ctxt->myDoc;
+ else {
+ ret = NULL;
+ xmlFreeDoc(ctxt->myDoc);
+ ctxt->myDoc = NULL;
+ }
+ if (sax != NULL)
+ ctxt->sax = NULL;
+ xmlFreeParserCtxt(ctxt);
+
+ return(ret);
+}
+
+/**
+ * xmlParseEntity:
+ * @filename: the filename
+ *
+ * parse an XML external entity out of context and build a tree.
+ *
+ * [78] extParsedEnt ::= TextDecl? content
+ *
+ * This correspond to a "Well Balanced" chunk
+ *
+ * Returns the resulting document tree
+ */
+
+xmlDocPtr
+xmlParseEntity(const char *filename) {
+ return(xmlSAXParseEntity(NULL, filename));
+}
+#endif /* LIBXML_SAX1_ENABLED */
+
+/**
+ * xmlCreateEntityParserCtxtInternal:
+ * @URL: the entity URL
+ * @ID: the entity PUBLIC ID
+ * @base: a possible base for the target URI
+ * @pctx: parser context used to set options on new context
+ *
+ * Create a parser context for an external entity
+ * Automatic support for ZLIB/Compress compressed document is provided
+ * by default if found at compile-time.
+ *
+ * Returns the new parser context or NULL
+ */
+static xmlParserCtxtPtr
+xmlCreateEntityParserCtxtInternal(const xmlChar *URL, const xmlChar *ID,
+ const xmlChar *base, xmlParserCtxtPtr pctx) {
+ xmlParserCtxtPtr ctxt;
+ xmlParserInputPtr inputStream;
+ char *directory = NULL;
+ xmlChar *uri;
+
+ ctxt = xmlNewParserCtxt();
+ if (ctxt == NULL) {
+ return(NULL);
+ }
+
+ if (pctx != NULL) {
+ ctxt->options = pctx->options;
+ ctxt->_private = pctx->_private;
+ }
+
+ uri = xmlBuildURI(URL, base);
+
+ if (uri == NULL) {
+ inputStream = xmlLoadExternalEntity((char *)URL, (char *)ID, ctxt);
+ if (inputStream == NULL) {
+ xmlFreeParserCtxt(ctxt);
+ return(NULL);
+ }
+
+ inputPush(ctxt, inputStream);
+
+ if ((ctxt->directory == NULL) && (directory == NULL))
+ directory = xmlParserGetDirectory((char *)URL);
+ if ((ctxt->directory == NULL) && (directory != NULL))
+ ctxt->directory = directory;
+ } else {
+ inputStream = xmlLoadExternalEntity((char *)uri, (char *)ID, ctxt);
+ if (inputStream == NULL) {
+ xmlFree(uri);
+ xmlFreeParserCtxt(ctxt);
+ return(NULL);
+ }
+
+ inputPush(ctxt, inputStream);
+
+ if ((ctxt->directory == NULL) && (directory == NULL))
+ directory = xmlParserGetDirectory((char *)uri);
+ if ((ctxt->directory == NULL) && (directory != NULL))
+ ctxt->directory = directory;
+ xmlFree(uri);
+ }
+ return(ctxt);
+}
+
+/**
+ * xmlCreateEntityParserCtxt:
+ * @URL: the entity URL
+ * @ID: the entity PUBLIC ID
+ * @base: a possible base for the target URI
+ *
+ * Create a parser context for an external entity
+ * Automatic support for ZLIB/Compress compressed document is provided
+ * by default if found at compile-time.
+ *
+ * Returns the new parser context or NULL
+ */
+xmlParserCtxtPtr
+xmlCreateEntityParserCtxt(const xmlChar *URL, const xmlChar *ID,
+ const xmlChar *base) {
+ return xmlCreateEntityParserCtxtInternal(URL, ID, base, NULL);
+
+}
+
+/************************************************************************
+ * *
+ * Front ends when parsing from a file *
+ * *
+ ************************************************************************/
+
+/**
+ * xmlCreateURLParserCtxt:
+ * @filename: the filename or URL
+ * @options: a combination of xmlParserOption
+ *
+ * Create a parser context for a file or URL content.
+ * Automatic support for ZLIB/Compress compressed document is provided
+ * by default if found at compile-time and for file accesses
+ *
+ * Returns the new parser context or NULL
+ */
+xmlParserCtxtPtr
+xmlCreateURLParserCtxt(const char *filename, int options)
+{
+ xmlParserCtxtPtr ctxt;
+ xmlParserInputPtr inputStream;
+ char *directory = NULL;
+
+ ctxt = xmlNewParserCtxt();
+ if (ctxt == NULL) {
+ xmlErrMemory(NULL, "cannot allocate parser context");
+ return(NULL);
+ }
+
+ if (options)
+ xmlCtxtUseOptionsInternal(ctxt, options, NULL);
+ ctxt->linenumbers = 1;
+
+ inputStream = xmlLoadExternalEntity(filename, NULL, ctxt);
+ if (inputStream == NULL) {
+ xmlFreeParserCtxt(ctxt);
+ return(NULL);
+ }
+
+ inputPush(ctxt, inputStream);
+ if ((ctxt->directory == NULL) && (directory == NULL))
+ directory = xmlParserGetDirectory(filename);
+ if ((ctxt->directory == NULL) && (directory != NULL))
+ ctxt->directory = directory;
+
+ return(ctxt);
+}
+
+/**
+ * xmlCreateFileParserCtxt:
+ * @filename: the filename
+ *
+ * Create a parser context for a file content.
+ * Automatic support for ZLIB/Compress compressed document is provided
+ * by default if found at compile-time.
+ *
+ * Returns the new parser context or NULL
+ */
+xmlParserCtxtPtr
+xmlCreateFileParserCtxt(const char *filename)
+{
+ return(xmlCreateURLParserCtxt(filename, 0));
+}
+
+#ifdef LIBXML_SAX1_ENABLED
+/**
+ * xmlSAXParseFileWithData:
+ * @sax: the SAX handler block
+ * @filename: the filename
+ * @recovery: work in recovery mode, i.e. tries to read no Well Formed
+ * documents
+ * @data: the userdata
+ *
+ * parse an XML file and build a tree. Automatic support for ZLIB/Compress
+ * compressed document is provided by default if found at compile-time.
+ * It use the given SAX function block to handle the parsing callback.
+ * If sax is NULL, fallback to the default DOM tree building routines.
+ *
+ * User data (void *) is stored within the parser context in the
+ * context's _private member, so it is available nearly everywhere in libxml
+ *
+ * Returns the resulting document tree
+ */
+
+xmlDocPtr
+xmlSAXParseFileWithData(xmlSAXHandlerPtr sax, const char *filename,
+ int recovery, void *data) {
+ xmlDocPtr ret;
+ xmlParserCtxtPtr ctxt;
+
+ xmlInitParser();
+
+ ctxt = xmlCreateFileParserCtxt(filename);
+ if (ctxt == NULL) {
+ return(NULL);
+ }
+ if (sax != NULL) {
+ if (ctxt->sax != NULL)
+ xmlFree(ctxt->sax);
+ ctxt->sax = sax;
+ }
+ xmlDetectSAX2(ctxt);
+ if (data!=NULL) {
+ ctxt->_private = data;
+ }
+
+ if (ctxt->directory == NULL)
+ ctxt->directory = xmlParserGetDirectory(filename);
+
+ ctxt->recovery = recovery;
+
+ xmlParseDocument(ctxt);
+
+ if ((ctxt->wellFormed) || recovery) {
+ ret = ctxt->myDoc;
+ if (ret != NULL) {
+ if (ctxt->input->buf->compressed > 0)
+ ret->compression = 9;
+ else
+ ret->compression = ctxt->input->buf->compressed;
+ }
+ }
+ else {
+ ret = NULL;
+ xmlFreeDoc(ctxt->myDoc);
+ ctxt->myDoc = NULL;
+ }
+ if (sax != NULL)
+ ctxt->sax = NULL;
+ xmlFreeParserCtxt(ctxt);
+
+ return(ret);
+}
+
+/**
+ * xmlSAXParseFile:
+ * @sax: the SAX handler block
+ * @filename: the filename
+ * @recovery: work in recovery mode, i.e. tries to read no Well Formed
+ * documents
+ *
+ * parse an XML file and build a tree. Automatic support for ZLIB/Compress
+ * compressed document is provided by default if found at compile-time.
+ * It use the given SAX function block to handle the parsing callback.
+ * If sax is NULL, fallback to the default DOM tree building routines.
+ *
+ * Returns the resulting document tree
+ */
+
+xmlDocPtr
+xmlSAXParseFile(xmlSAXHandlerPtr sax, const char *filename,
+ int recovery) {
+ return(xmlSAXParseFileWithData(sax,filename,recovery,NULL));
+}
+
+/**
+ * xmlRecoverDoc:
+ * @cur: a pointer to an array of xmlChar
+ *
+ * parse an XML in-memory document and build a tree.
+ * In the case the document is not Well Formed, a attempt to build a
+ * tree is tried anyway
+ *
+ * Returns the resulting document tree or NULL in case of failure
+ */
+
+xmlDocPtr
+xmlRecoverDoc(const xmlChar *cur) {
+ return(xmlSAXParseDoc(NULL, cur, 1));
+}
+
+/**
+ * xmlParseFile:
+ * @filename: the filename
+ *
+ * parse an XML file and build a tree. Automatic support for ZLIB/Compress
+ * compressed document is provided by default if found at compile-time.
+ *
+ * Returns the resulting document tree if the file was wellformed,
+ * NULL otherwise.
+ */
+
+xmlDocPtr
+xmlParseFile(const char *filename) {
+ return(xmlSAXParseFile(NULL, filename, 0));
+}
+
+/**
+ * xmlRecoverFile:
+ * @filename: the filename
+ *
+ * parse an XML file and build a tree. Automatic support for ZLIB/Compress
+ * compressed document is provided by default if found at compile-time.
+ * In the case the document is not Well Formed, it attempts to build
+ * a tree anyway
+ *
+ * Returns the resulting document tree or NULL in case of failure
+ */
+
+xmlDocPtr
+xmlRecoverFile(const char *filename) {
+ return(xmlSAXParseFile(NULL, filename, 1));
+}
+
+
+/**
+ * xmlSetupParserForBuffer:
+ * @ctxt: an XML parser context
+ * @buffer: a xmlChar * buffer
+ * @filename: a file name
+ *
+ * Setup the parser context to parse a new buffer; Clears any prior
+ * contents from the parser context. The buffer parameter must not be
+ * NULL, but the filename parameter can be
+ */
+void
+xmlSetupParserForBuffer(xmlParserCtxtPtr ctxt, const xmlChar* buffer,
+ const char* filename)
+{
+ xmlParserInputPtr input;
+
+ if ((ctxt == NULL) || (buffer == NULL))
+ return;
+
+ input = xmlNewInputStream(ctxt);
+ if (input == NULL) {
+ xmlErrMemory(NULL, "parsing new buffer: out of memory\n");
+ xmlClearParserCtxt(ctxt);
+ return;
+ }
+
+ xmlClearParserCtxt(ctxt);
+ if (filename != NULL)
+ input->filename = (char *) xmlCanonicPath((const xmlChar *)filename);
+ input->base = buffer;
+ input->cur = buffer;
+ input->end = &buffer[xmlStrlen(buffer)];
+ inputPush(ctxt, input);
+}
+
+/**
+ * xmlSAXUserParseFile:
+ * @sax: a SAX handler
+ * @user_data: The user data returned on SAX callbacks
+ * @filename: a file name
+ *
+ * parse an XML file and call the given SAX handler routines.
+ * Automatic support for ZLIB/Compress compressed document is provided
+ *
+ * Returns 0 in case of success or a error number otherwise
+ */
+int
+xmlSAXUserParseFile(xmlSAXHandlerPtr sax, void *user_data,
+ const char *filename) {
+ int ret = 0;
+ xmlParserCtxtPtr ctxt;
+
+ ctxt = xmlCreateFileParserCtxt(filename);
+ if (ctxt == NULL) return -1;
+ if (ctxt->sax != (xmlSAXHandlerPtr) &xmlDefaultSAXHandler)
+ xmlFree(ctxt->sax);
+ ctxt->sax = sax;
+ xmlDetectSAX2(ctxt);
+
+ if (user_data != NULL)
+ ctxt->userData = user_data;
+
+ xmlParseDocument(ctxt);
+
+ if (ctxt->wellFormed)
+ ret = 0;
+ else {
+ if (ctxt->errNo != 0)
+ ret = ctxt->errNo;
+ else
+ ret = -1;
+ }
+ if (sax != NULL)
+ ctxt->sax = NULL;
+ if (ctxt->myDoc != NULL) {
+ xmlFreeDoc(ctxt->myDoc);
+ ctxt->myDoc = NULL;
+ }
+ xmlFreeParserCtxt(ctxt);
+
+ return ret;
+}
+#endif /* LIBXML_SAX1_ENABLED */
+
+/************************************************************************
+ * *
+ * Front ends when parsing from memory *
+ * *
+ ************************************************************************/
+
+/**
+ * xmlCreateMemoryParserCtxt:
+ * @buffer: a pointer to a char array
+ * @size: the size of the array
+ *
+ * Create a parser context for an XML in-memory document.
+ *
+ * Returns the new parser context or NULL
+ */
+xmlParserCtxtPtr
+xmlCreateMemoryParserCtxt(const char *buffer, int size) {
+ xmlParserCtxtPtr ctxt;
+ xmlParserInputPtr input;
+ xmlParserInputBufferPtr buf;
+
+ if (buffer == NULL)
+ return(NULL);
+ if (size <= 0)
+ return(NULL);
+
+ ctxt = xmlNewParserCtxt();
+ if (ctxt == NULL)
+ return(NULL);
+
+ /* TODO: xmlParserInputBufferCreateStatic, requires some serious changes */
+ buf = xmlParserInputBufferCreateMem(buffer, size, XML_CHAR_ENCODING_NONE);
+ if (buf == NULL) {
+ xmlFreeParserCtxt(ctxt);
+ return(NULL);
+ }
+
+ input = xmlNewInputStream(ctxt);
+ if (input == NULL) {
+ xmlFreeParserInputBuffer(buf);
+ xmlFreeParserCtxt(ctxt);
+ return(NULL);
+ }
+
+ input->filename = NULL;
+ input->buf = buf;
+ xmlBufResetInput(input->buf->buffer, input);
+
+ inputPush(ctxt, input);
+ return(ctxt);
+}
+
+#ifdef LIBXML_SAX1_ENABLED
+/**
+ * xmlSAXParseMemoryWithData:
+ * @sax: the SAX handler block
+ * @buffer: an pointer to a char array
+ * @size: the size of the array
+ * @recovery: work in recovery mode, i.e. tries to read no Well Formed
+ * documents
+ * @data: the userdata
+ *
+ * parse an XML in-memory block and use the given SAX function block
+ * to handle the parsing callback. If sax is NULL, fallback to the default
+ * DOM tree building routines.
+ *
+ * User data (void *) is stored within the parser context in the
+ * context's _private member, so it is available nearly everywhere in libxml
+ *
+ * Returns the resulting document tree
+ */
+
+xmlDocPtr
+xmlSAXParseMemoryWithData(xmlSAXHandlerPtr sax, const char *buffer,
+ int size, int recovery, void *data) {
+ xmlDocPtr ret;
+ xmlParserCtxtPtr ctxt;
+
+ xmlInitParser();
+
+ ctxt = xmlCreateMemoryParserCtxt(buffer, size);
+ if (ctxt == NULL) return(NULL);
+ if (sax != NULL) {
+ if (ctxt->sax != NULL)
+ xmlFree(ctxt->sax);
+ ctxt->sax = sax;
+ }
+ xmlDetectSAX2(ctxt);
+ if (data!=NULL) {
+ ctxt->_private=data;
+ }
+
+ ctxt->recovery = recovery;
+
+ xmlParseDocument(ctxt);
+
+ if ((ctxt->wellFormed) || recovery) ret = ctxt->myDoc;
+ else {
+ ret = NULL;
+ xmlFreeDoc(ctxt->myDoc);
+ ctxt->myDoc = NULL;
+ }
+ if (sax != NULL)
+ ctxt->sax = NULL;
+ xmlFreeParserCtxt(ctxt);
+
+ return(ret);
+}
+
+/**
+ * xmlSAXParseMemory:
+ * @sax: the SAX handler block
+ * @buffer: an pointer to a char array
+ * @size: the size of the array
+ * @recovery: work in recovery mode, i.e. tries to read not Well Formed
+ * documents
+ *
+ * parse an XML in-memory block and use the given SAX function block
+ * to handle the parsing callback. If sax is NULL, fallback to the default
+ * DOM tree building routines.
+ *
+ * Returns the resulting document tree
+ */
+xmlDocPtr
+xmlSAXParseMemory(xmlSAXHandlerPtr sax, const char *buffer,
+ int size, int recovery) {
+ return xmlSAXParseMemoryWithData(sax, buffer, size, recovery, NULL);
+}
+
+/**
+ * xmlParseMemory:
+ * @buffer: an pointer to a char array
+ * @size: the size of the array
+ *
+ * parse an XML in-memory block and build a tree.
+ *
+ * Returns the resulting document tree
+ */
+
+xmlDocPtr xmlParseMemory(const char *buffer, int size) {
+ return(xmlSAXParseMemory(NULL, buffer, size, 0));
+}
+
+/**
+ * xmlRecoverMemory:
+ * @buffer: an pointer to a char array
+ * @size: the size of the array
+ *
+ * parse an XML in-memory block and build a tree.
+ * In the case the document is not Well Formed, an attempt to
+ * build a tree is tried anyway
+ *
+ * Returns the resulting document tree or NULL in case of error
+ */
+
+xmlDocPtr xmlRecoverMemory(const char *buffer, int size) {
+ return(xmlSAXParseMemory(NULL, buffer, size, 1));
+}
+
+/**
+ * xmlSAXUserParseMemory:
+ * @sax: a SAX handler
+ * @user_data: The user data returned on SAX callbacks
+ * @buffer: an in-memory XML document input
+ * @size: the length of the XML document in bytes
+ *
+ * A better SAX parsing routine.
+ * parse an XML in-memory buffer and call the given SAX handler routines.
+ *
+ * Returns 0 in case of success or a error number otherwise
+ */
+int xmlSAXUserParseMemory(xmlSAXHandlerPtr sax, void *user_data,
+ const char *buffer, int size) {
+ int ret = 0;
+ xmlParserCtxtPtr ctxt;
+
+ xmlInitParser();
+
+ ctxt = xmlCreateMemoryParserCtxt(buffer, size);
+ if (ctxt == NULL) return -1;
+ if (ctxt->sax != (xmlSAXHandlerPtr) &xmlDefaultSAXHandler)
+ xmlFree(ctxt->sax);
+ ctxt->sax = sax;
+ xmlDetectSAX2(ctxt);
+
+ if (user_data != NULL)
+ ctxt->userData = user_data;
+
+ xmlParseDocument(ctxt);
+
+ if (ctxt->wellFormed)
+ ret = 0;
+ else {
+ if (ctxt->errNo != 0)
+ ret = ctxt->errNo;
+ else
+ ret = -1;
+ }
+ if (sax != NULL)
+ ctxt->sax = NULL;
+ if (ctxt->myDoc != NULL) {
+ xmlFreeDoc(ctxt->myDoc);
+ ctxt->myDoc = NULL;
+ }
+ xmlFreeParserCtxt(ctxt);
+
+ return ret;
+}
+#endif /* LIBXML_SAX1_ENABLED */
+
+/**
+ * xmlCreateDocParserCtxt:
+ * @cur: a pointer to an array of xmlChar
+ *
+ * Creates a parser context for an XML in-memory document.
+ *
+ * Returns the new parser context or NULL
+ */
+xmlParserCtxtPtr
+xmlCreateDocParserCtxt(const xmlChar *cur) {
+ int len;
+
+ if (cur == NULL)
+ return(NULL);
+ len = xmlStrlen(cur);
+ return(xmlCreateMemoryParserCtxt((const char *)cur, len));
+}
+
+#ifdef LIBXML_SAX1_ENABLED
+/**
+ * xmlSAXParseDoc:
+ * @sax: the SAX handler block
+ * @cur: a pointer to an array of xmlChar
+ * @recovery: work in recovery mode, i.e. tries to read no Well Formed
+ * documents
+ *
+ * parse an XML in-memory document and build a tree.
+ * It use the given SAX function block to handle the parsing callback.
+ * If sax is NULL, fallback to the default DOM tree building routines.
+ *
+ * Returns the resulting document tree
+ */
+
+xmlDocPtr
+xmlSAXParseDoc(xmlSAXHandlerPtr sax, const xmlChar *cur, int recovery) {
+ xmlDocPtr ret;
+ xmlParserCtxtPtr ctxt;
+ xmlSAXHandlerPtr oldsax = NULL;
+
+ if (cur == NULL) return(NULL);
+
+
+ ctxt = xmlCreateDocParserCtxt(cur);
+ if (ctxt == NULL) return(NULL);
+ if (sax != NULL) {
+ oldsax = ctxt->sax;
+ ctxt->sax = sax;
+ ctxt->userData = NULL;
+ }
+ xmlDetectSAX2(ctxt);
+
+ xmlParseDocument(ctxt);
+ if ((ctxt->wellFormed) || recovery) ret = ctxt->myDoc;
+ else {
+ ret = NULL;
+ xmlFreeDoc(ctxt->myDoc);
+ ctxt->myDoc = NULL;
+ }
+ if (sax != NULL)
+ ctxt->sax = oldsax;
+ xmlFreeParserCtxt(ctxt);
+
+ return(ret);
+}
+
+/**
+ * xmlParseDoc:
+ * @cur: a pointer to an array of xmlChar
+ *
+ * parse an XML in-memory document and build a tree.
+ *
+ * Returns the resulting document tree
+ */
+
+xmlDocPtr
+xmlParseDoc(const xmlChar *cur) {
+ return(xmlSAXParseDoc(NULL, cur, 0));
+}
+#endif /* LIBXML_SAX1_ENABLED */
+
+#ifdef LIBXML_LEGACY_ENABLED
+/************************************************************************
+ * *
+ * Specific function to keep track of entities references *
+ * and used by the XSLT debugger *
+ * *
+ ************************************************************************/
+
+static xmlEntityReferenceFunc xmlEntityRefFunc = NULL;
+
+/**
+ * xmlAddEntityReference:
+ * @ent : A valid entity
+ * @firstNode : A valid first node for children of entity
+ * @lastNode : A valid last node of children entity
+ *
+ * Notify of a reference to an entity of type XML_EXTERNAL_GENERAL_PARSED_ENTITY
+ */
+static void
+xmlAddEntityReference(xmlEntityPtr ent, xmlNodePtr firstNode,
+ xmlNodePtr lastNode)
+{
+ if (xmlEntityRefFunc != NULL) {
+ (*xmlEntityRefFunc) (ent, firstNode, lastNode);
+ }
+}
+
+
+/**
+ * xmlSetEntityReferenceFunc:
+ * @func: A valid function
+ *
+ * Set the function to call call back when a xml reference has been made
+ */
+void
+xmlSetEntityReferenceFunc(xmlEntityReferenceFunc func)
+{
+ xmlEntityRefFunc = func;
+}
+#endif /* LIBXML_LEGACY_ENABLED */
+
+/************************************************************************
+ * *
+ * Miscellaneous *
+ * *
+ ************************************************************************/
+
+#ifdef LIBXML_XPATH_ENABLED
+#include <libxml/xpath.h>
+#endif
+
+extern void XMLCDECL xmlGenericErrorDefaultFunc(void *ctx, const char *msg, ...);
+static int xmlParserInitialized = 0;
+
+/**
+ * xmlInitParser:
+ *
+ * Initialization function for the XML parser.
+ * This is not reentrant. Call once before processing in case of
+ * use in multithreaded programs.
+ */
+
+void
+xmlInitParser(void) {
+ if (xmlParserInitialized != 0)
+ return;
+
+#ifdef LIBXML_THREAD_ENABLED
+ __xmlGlobalInitMutexLock();
+ if (xmlParserInitialized == 0) {
+#endif
+ xmlInitThreads();
+ xmlInitGlobals();
+ if ((xmlGenericError == xmlGenericErrorDefaultFunc) ||
+ (xmlGenericError == NULL))
+ initGenericErrorDefaultFunc(NULL);
+ xmlInitMemory();
+ xmlInitializeDict();
+ xmlInitCharEncodingHandlers();
+ xmlDefaultSAXHandlerInit();
+ xmlRegisterDefaultInputCallbacks();
+#ifdef LIBXML_OUTPUT_ENABLED
+ xmlRegisterDefaultOutputCallbacks();
+#endif /* LIBXML_OUTPUT_ENABLED */
+#ifdef LIBXML_HTML_ENABLED
+ htmlInitAutoClose();
+ htmlDefaultSAXHandlerInit();
+#endif
+#ifdef LIBXML_XPATH_ENABLED
+ xmlXPathInit();
+#endif
+ xmlParserInitialized = 1;
+#ifdef LIBXML_THREAD_ENABLED
+ }
+ __xmlGlobalInitMutexUnlock();
+#endif
+}
+
+/**
+ * xmlCleanupParser:
+ *
+ * This function name is somewhat misleading. It does not clean up
+ * parser state, it cleans up memory allocated by the library itself.
+ * It is a cleanup function for the XML library. It tries to reclaim all
+ * related global memory allocated for the library processing.
+ * It doesn't deallocate any document related memory. One should
+ * call xmlCleanupParser() only when the process has finished using
+ * the library and all XML/HTML documents built with it.
+ * See also xmlInitParser() which has the opposite function of preparing
+ * the library for operations.
+ *
+ * WARNING: if your application is multithreaded or has plugin support
+ * calling this may crash the application if another thread or
+ * a plugin is still using libxml2. It's sometimes very hard to
+ * guess if libxml2 is in use in the application, some libraries
+ * or plugins may use it without notice. In case of doubt abstain
+ * from calling this function or do it just before calling exit()
+ * to avoid leak reports from valgrind !
+ */
+
+void
+xmlCleanupParser(void) {
+ if (!xmlParserInitialized)
+ return;
+
+ xmlCleanupCharEncodingHandlers();
+#ifdef LIBXML_CATALOG_ENABLED
+ xmlCatalogCleanup();
+#endif
+ xmlDictCleanup();
+ xmlCleanupInputCallbacks();
+#ifdef LIBXML_OUTPUT_ENABLED
+ xmlCleanupOutputCallbacks();
+#endif
+#ifdef LIBXML_SCHEMAS_ENABLED
+ xmlSchemaCleanupTypes();
+ xmlRelaxNGCleanupTypes();
+#endif
+ xmlResetLastError();
+ xmlCleanupGlobals();
+ xmlCleanupThreads(); /* must be last if called not from the main thread */
+ xmlCleanupMemory();
+ xmlParserInitialized = 0;
+}
+
+/************************************************************************
+ * *
+ * New set (2.6.0) of simpler and more flexible APIs *
+ * *
+ ************************************************************************/
+
+/**
+ * DICT_FREE:
+ * @str: a string
+ *
+ * Free a string if it is not owned by the "dict" dictionary in the
+ * current scope
+ */
+#define DICT_FREE(str) \
+ if ((str) && ((!dict) || \
+ (xmlDictOwns(dict, (const xmlChar *)(str)) == 0))) \
+ xmlFree((char *)(str));
+
+/**
+ * xmlCtxtReset:
+ * @ctxt: an XML parser context
+ *
+ * Reset a parser context
+ */
+void
+xmlCtxtReset(xmlParserCtxtPtr ctxt)
+{
+ xmlParserInputPtr input;
+ xmlDictPtr dict;
+
+ if (ctxt == NULL)
+ return;
+
+ dict = ctxt->dict;
+
+ while ((input = inputPop(ctxt)) != NULL) { /* Non consuming */
+ xmlFreeInputStream(input);
+ }
+ ctxt->inputNr = 0;
+ ctxt->input = NULL;
+
+ ctxt->spaceNr = 0;
+ if (ctxt->spaceTab != NULL) {
+ ctxt->spaceTab[0] = -1;
+ ctxt->space = &ctxt->spaceTab[0];
+ } else {
+ ctxt->space = NULL;
+ }
+
+
+ ctxt->nodeNr = 0;
+ ctxt->node = NULL;
+
+ ctxt->nameNr = 0;
+ ctxt->name = NULL;
+
+ DICT_FREE(ctxt->version);
+ ctxt->version = NULL;
+ DICT_FREE(ctxt->encoding);
+ ctxt->encoding = NULL;
+ DICT_FREE(ctxt->directory);
+ ctxt->directory = NULL;
+ DICT_FREE(ctxt->extSubURI);
+ ctxt->extSubURI = NULL;
+ DICT_FREE(ctxt->extSubSystem);
+ ctxt->extSubSystem = NULL;
+ if (ctxt->myDoc != NULL)
+ xmlFreeDoc(ctxt->myDoc);
+ ctxt->myDoc = NULL;
+
+ ctxt->standalone = -1;
+ ctxt->hasExternalSubset = 0;
+ ctxt->hasPErefs = 0;
+ ctxt->html = 0;
+ ctxt->external = 0;
+ ctxt->instate = XML_PARSER_START;
+ ctxt->token = 0;
+
+ ctxt->wellFormed = 1;
+ ctxt->nsWellFormed = 1;
+ ctxt->disableSAX = 0;
+ ctxt->valid = 1;
+#if 0
+ ctxt->vctxt.userData = ctxt;
+ ctxt->vctxt.error = xmlParserValidityError;
+ ctxt->vctxt.warning = xmlParserValidityWarning;
+#endif
+ ctxt->record_info = 0;
+ ctxt->nbChars = 0;
+ ctxt->checkIndex = 0;
+ ctxt->inSubset = 0;
+ ctxt->errNo = XML_ERR_OK;
+ ctxt->depth = 0;
+ ctxt->charset = XML_CHAR_ENCODING_UTF8;
+ ctxt->catalogs = NULL;
+ ctxt->nbentities = 0;
+ ctxt->sizeentities = 0;
+ ctxt->sizeentcopy = 0;
+ xmlInitNodeInfoSeq(&ctxt->node_seq);
+
+ if (ctxt->attsDefault != NULL) {
+ xmlHashFree(ctxt->attsDefault, (xmlHashDeallocator) xmlFree);
+ ctxt->attsDefault = NULL;
+ }
+ if (ctxt->attsSpecial != NULL) {
+ xmlHashFree(ctxt->attsSpecial, NULL);
+ ctxt->attsSpecial = NULL;
+ }
+
+#ifdef LIBXML_CATALOG_ENABLED
+ if (ctxt->catalogs != NULL)
+ xmlCatalogFreeLocal(ctxt->catalogs);
+#endif
+ if (ctxt->lastError.code != XML_ERR_OK)
+ xmlResetError(&ctxt->lastError);
+}
+
+/**
+ * xmlCtxtResetPush:
+ * @ctxt: an XML parser context
+ * @chunk: a pointer to an array of chars
+ * @size: number of chars in the array
+ * @filename: an optional file name or URI
+ * @encoding: the document encoding, or NULL
+ *
+ * Reset a push parser context
+ *
+ * Returns 0 in case of success and 1 in case of error
+ */
+int
+xmlCtxtResetPush(xmlParserCtxtPtr ctxt, const char *chunk,
+ int size, const char *filename, const char *encoding)
+{
+ xmlParserInputPtr inputStream;
+ xmlParserInputBufferPtr buf;
+ xmlCharEncoding enc = XML_CHAR_ENCODING_NONE;
+
+ if (ctxt == NULL)
+ return(1);
+
+ if ((encoding == NULL) && (chunk != NULL) && (size >= 4))
+ enc = xmlDetectCharEncoding((const xmlChar *) chunk, size);
+
+ buf = xmlAllocParserInputBuffer(enc);
+ if (buf == NULL)
+ return(1);
+
+ if (ctxt == NULL) {
+ xmlFreeParserInputBuffer(buf);
+ return(1);
+ }
+
+ xmlCtxtReset(ctxt);
+
+ if (ctxt->pushTab == NULL) {
+ ctxt->pushTab = (void **) xmlMalloc(ctxt->nameMax * 3 *
+ sizeof(xmlChar *));
+ if (ctxt->pushTab == NULL) {
+ xmlErrMemory(ctxt, NULL);
+ xmlFreeParserInputBuffer(buf);
+ return(1);
+ }
+ }
+
+ if (filename == NULL) {
+ ctxt->directory = NULL;
+ } else {
+ ctxt->directory = xmlParserGetDirectory(filename);
+ }
+
+ inputStream = xmlNewInputStream(ctxt);
+ if (inputStream == NULL) {
+ xmlFreeParserInputBuffer(buf);
+ return(1);
+ }
+
+ if (filename == NULL)
+ inputStream->filename = NULL;
+ else
+ inputStream->filename = (char *)
+ xmlCanonicPath((const xmlChar *) filename);
+ inputStream->buf = buf;
+ xmlBufResetInput(buf->buffer, inputStream);
+
+ inputPush(ctxt, inputStream);
+
+ if ((size > 0) && (chunk != NULL) && (ctxt->input != NULL) &&
+ (ctxt->input->buf != NULL)) {
+ size_t base = xmlBufGetInputBase(ctxt->input->buf->buffer, ctxt->input);
+ size_t cur = ctxt->input->cur - ctxt->input->base;
+
+ xmlParserInputBufferPush(ctxt->input->buf, size, chunk);
+
+ xmlBufSetInputBaseCur(ctxt->input->buf->buffer, ctxt->input, base, cur);
+#ifdef DEBUG_PUSH
+ xmlGenericError(xmlGenericErrorContext, "PP: pushed %d\n", size);
+#endif
+ }
+
+ if (encoding != NULL) {
+ xmlCharEncodingHandlerPtr hdlr;
+
+ if (ctxt->encoding != NULL)
+ xmlFree((xmlChar *) ctxt->encoding);
+ ctxt->encoding = xmlStrdup((const xmlChar *) encoding);
+
+ hdlr = xmlFindCharEncodingHandler(encoding);
+ if (hdlr != NULL) {
+ xmlSwitchToEncoding(ctxt, hdlr);
+ } else {
+ xmlFatalErrMsgStr(ctxt, XML_ERR_UNSUPPORTED_ENCODING,
+ "Unsupported encoding %s\n", BAD_CAST encoding);
+ }
+ } else if (enc != XML_CHAR_ENCODING_NONE) {
+ xmlSwitchEncoding(ctxt, enc);
+ }
+
+ return(0);
+}
+
+
+/**
+ * xmlCtxtUseOptionsInternal:
+ * @ctxt: an XML parser context
+ * @options: a combination of xmlParserOption
+ * @encoding: the user provided encoding to use
+ *
+ * Applies the options to the parser context
+ *
+ * Returns 0 in case of success, the set of unknown or unimplemented options
+ * in case of error.
+ */
+static int
+xmlCtxtUseOptionsInternal(xmlParserCtxtPtr ctxt, int options, const char *encoding)
+{
+ if (ctxt == NULL)
+ return(-1);
+ if (encoding != NULL) {
+ if (ctxt->encoding != NULL)
+ xmlFree((xmlChar *) ctxt->encoding);
+ ctxt->encoding = xmlStrdup((const xmlChar *) encoding);
+ }
+ if (options & XML_PARSE_RECOVER) {
+ ctxt->recovery = 1;
+ options -= XML_PARSE_RECOVER;
+ ctxt->options |= XML_PARSE_RECOVER;
+ } else
+ ctxt->recovery = 0;
+ if (options & XML_PARSE_DTDLOAD) {
+ ctxt->loadsubset = XML_DETECT_IDS;
+ options -= XML_PARSE_DTDLOAD;
+ ctxt->options |= XML_PARSE_DTDLOAD;
+ } else
+ ctxt->loadsubset = 0;
+ if (options & XML_PARSE_DTDATTR) {
+ ctxt->loadsubset |= XML_COMPLETE_ATTRS;
+ options -= XML_PARSE_DTDATTR;
+ ctxt->options |= XML_PARSE_DTDATTR;
+ }
+ if (options & XML_PARSE_NOENT) {
+ ctxt->replaceEntities = 1;
+ /* ctxt->loadsubset |= XML_DETECT_IDS; */
+ options -= XML_PARSE_NOENT;
+ ctxt->options |= XML_PARSE_NOENT;
+ } else
+ ctxt->replaceEntities = 0;
+ if (options & XML_PARSE_PEDANTIC) {
+ ctxt->pedantic = 1;
+ options -= XML_PARSE_PEDANTIC;
+ ctxt->options |= XML_PARSE_PEDANTIC;
+ } else
+ ctxt->pedantic = 0;
+ if (options & XML_PARSE_NOBLANKS) {
+ ctxt->keepBlanks = 0;
+ ctxt->sax->ignorableWhitespace = xmlSAX2IgnorableWhitespace;
+ options -= XML_PARSE_NOBLANKS;
+ ctxt->options |= XML_PARSE_NOBLANKS;
+ } else
+ ctxt->keepBlanks = 1;
+ if (options & XML_PARSE_DTDVALID) {
+ ctxt->validate = 1;
+ if (options & XML_PARSE_NOWARNING)
+ ctxt->vctxt.warning = NULL;
+ if (options & XML_PARSE_NOERROR)
+ ctxt->vctxt.error = NULL;
+ options -= XML_PARSE_DTDVALID;
+ ctxt->options |= XML_PARSE_DTDVALID;
+ } else
+ ctxt->validate = 0;
+ if (options & XML_PARSE_NOWARNING) {
+ ctxt->sax->warning = NULL;
+ options -= XML_PARSE_NOWARNING;
+ }
+ if (options & XML_PARSE_NOERROR) {
+ ctxt->sax->error = NULL;
+ ctxt->sax->fatalError = NULL;
+ options -= XML_PARSE_NOERROR;
+ }
+#ifdef LIBXML_SAX1_ENABLED
+ if (options & XML_PARSE_SAX1) {
+ ctxt->sax->startElement = xmlSAX2StartElement;
+ ctxt->sax->endElement = xmlSAX2EndElement;
+ ctxt->sax->startElementNs = NULL;
+ ctxt->sax->endElementNs = NULL;
+ ctxt->sax->initialized = 1;
+ options -= XML_PARSE_SAX1;
+ ctxt->options |= XML_PARSE_SAX1;
+ }
+#endif /* LIBXML_SAX1_ENABLED */
+ if (options & XML_PARSE_NODICT) {
+ ctxt->dictNames = 0;
+ options -= XML_PARSE_NODICT;
+ ctxt->options |= XML_PARSE_NODICT;
+ } else {
+ ctxt->dictNames = 1;
+ }
+ if (options & XML_PARSE_NOCDATA) {
+ ctxt->sax->cdataBlock = NULL;
+ options -= XML_PARSE_NOCDATA;
+ ctxt->options |= XML_PARSE_NOCDATA;
+ }
+ if (options & XML_PARSE_NSCLEAN) {
+ ctxt->options |= XML_PARSE_NSCLEAN;
+ options -= XML_PARSE_NSCLEAN;
+ }
+ if (options & XML_PARSE_NONET) {
+ ctxt->options |= XML_PARSE_NONET;
+ options -= XML_PARSE_NONET;
+ }
+ if (options & XML_PARSE_COMPACT) {
+ ctxt->options |= XML_PARSE_COMPACT;
+ options -= XML_PARSE_COMPACT;
+ }
+ if (options & XML_PARSE_OLD10) {
+ ctxt->options |= XML_PARSE_OLD10;
+ options -= XML_PARSE_OLD10;
+ }
+ if (options & XML_PARSE_NOBASEFIX) {
+ ctxt->options |= XML_PARSE_NOBASEFIX;
+ options -= XML_PARSE_NOBASEFIX;
+ }
+ if (options & XML_PARSE_HUGE) {
+ ctxt->options |= XML_PARSE_HUGE;
+ options -= XML_PARSE_HUGE;
+ if (ctxt->dict != NULL)
+ xmlDictSetLimit(ctxt->dict, 0);
+ }
+ if (options & XML_PARSE_OLDSAX) {
+ ctxt->options |= XML_PARSE_OLDSAX;
+ options -= XML_PARSE_OLDSAX;
+ }
+ if (options & XML_PARSE_IGNORE_ENC) {
+ ctxt->options |= XML_PARSE_IGNORE_ENC;
+ options -= XML_PARSE_IGNORE_ENC;
+ }
+ if (options & XML_PARSE_BIG_LINES) {
+ ctxt->options |= XML_PARSE_BIG_LINES;
+ options -= XML_PARSE_BIG_LINES;
+ }
+ ctxt->linenumbers = 1;
+ return (options);
+}
+
+/**
+ * xmlCtxtUseOptions:
+ * @ctxt: an XML parser context
+ * @options: a combination of xmlParserOption
+ *
+ * Applies the options to the parser context
+ *
+ * Returns 0 in case of success, the set of unknown or unimplemented options
+ * in case of error.
+ */
+int
+xmlCtxtUseOptions(xmlParserCtxtPtr ctxt, int options)
+{
+ return(xmlCtxtUseOptionsInternal(ctxt, options, NULL));
+}
+
+/**
+ * xmlDoRead:
+ * @ctxt: an XML parser context
+ * @URL: the base URL to use for the document
+ * @encoding: the document encoding, or NULL
+ * @options: a combination of xmlParserOption
+ * @reuse: keep the context for reuse
+ *
+ * Common front-end for the xmlRead functions
+ *
+ * Returns the resulting document tree or NULL
+ */
+static xmlDocPtr
+xmlDoRead(xmlParserCtxtPtr ctxt, const char *URL, const char *encoding,
+ int options, int reuse)
+{
+ xmlDocPtr ret;
+
+ xmlCtxtUseOptionsInternal(ctxt, options, encoding);
+ if (encoding != NULL) {
+ xmlCharEncodingHandlerPtr hdlr;
+
+ hdlr = xmlFindCharEncodingHandler(encoding);
+ if (hdlr != NULL)
+ xmlSwitchToEncoding(ctxt, hdlr);
+ }
+ if ((URL != NULL) && (ctxt->input != NULL) &&
+ (ctxt->input->filename == NULL))
+ ctxt->input->filename = (char *) xmlStrdup((const xmlChar *) URL);
+ xmlParseDocument(ctxt);
+ if ((ctxt->wellFormed) || ctxt->recovery)
+ ret = ctxt->myDoc;
+ else {
+ ret = NULL;
+ if (ctxt->myDoc != NULL) {
+ xmlFreeDoc(ctxt->myDoc);
+ }
+ }
+ ctxt->myDoc = NULL;
+ if (!reuse) {
+ xmlFreeParserCtxt(ctxt);
+ }
+
+ return (ret);
+}
+
+/**
+ * xmlReadDoc:
+ * @cur: a pointer to a zero terminated string
+ * @URL: the base URL to use for the document
+ * @encoding: the document encoding, or NULL
+ * @options: a combination of xmlParserOption
+ *
+ * parse an XML in-memory document and build a tree.
+ *
+ * Returns the resulting document tree
+ */
+xmlDocPtr
+xmlReadDoc(const xmlChar * cur, const char *URL, const char *encoding, int options)
+{
+ xmlParserCtxtPtr ctxt;
+
+ if (cur == NULL)
+ return (NULL);
+ xmlInitParser();
+
+ ctxt = xmlCreateDocParserCtxt(cur);
+ if (ctxt == NULL)
+ return (NULL);
+ return (xmlDoRead(ctxt, URL, encoding, options, 0));
+}
+
+/**
+ * xmlReadFile:
+ * @filename: a file or URL
+ * @encoding: the document encoding, or NULL
+ * @options: a combination of xmlParserOption
+ *
+ * parse an XML file from the filesystem or the network.
+ *
+ * Returns the resulting document tree
+ */
+xmlDocPtr
+xmlReadFile(const char *filename, const char *encoding, int options)
+{
+ xmlParserCtxtPtr ctxt;
+
+ xmlInitParser();
+ ctxt = xmlCreateURLParserCtxt(filename, options);
+ if (ctxt == NULL)
+ return (NULL);
+ return (xmlDoRead(ctxt, NULL, encoding, options, 0));
+}
+
+/**
+ * xmlReadMemory:
+ * @buffer: a pointer to a char array
+ * @size: the size of the array
+ * @URL: the base URL to use for the document
+ * @encoding: the document encoding, or NULL
+ * @options: a combination of xmlParserOption
+ *
+ * parse an XML in-memory document and build a tree.
+ *
+ * Returns the resulting document tree
+ */
+xmlDocPtr
+xmlReadMemory(const char *buffer, int size, const char *URL, const char *encoding, int options)
+{
+ xmlParserCtxtPtr ctxt;
+
+ xmlInitParser();
+ ctxt = xmlCreateMemoryParserCtxt(buffer, size);
+ if (ctxt == NULL)
+ return (NULL);
+ return (xmlDoRead(ctxt, URL, encoding, options, 0));
+}
+
+/**
+ * xmlReadFd:
+ * @fd: an open file descriptor
+ * @URL: the base URL to use for the document
+ * @encoding: the document encoding, or NULL
+ * @options: a combination of xmlParserOption
+ *
+ * parse an XML from a file descriptor and build a tree.
+ * NOTE that the file descriptor will not be closed when the
+ * reader is closed or reset.
+ *
+ * Returns the resulting document tree
+ */
+xmlDocPtr
+xmlReadFd(int fd, const char *URL, const char *encoding, int options)
+{
+ xmlParserCtxtPtr ctxt;
+ xmlParserInputBufferPtr input;
+ xmlParserInputPtr stream;
+
+ if (fd < 0)
+ return (NULL);
+ xmlInitParser();
+
+ input = xmlParserInputBufferCreateFd(fd, XML_CHAR_ENCODING_NONE);
+ if (input == NULL)
+ return (NULL);
+ input->closecallback = NULL;
+ ctxt = xmlNewParserCtxt();
+ if (ctxt == NULL) {
+ xmlFreeParserInputBuffer(input);
+ return (NULL);
+ }
+ stream = xmlNewIOInputStream(ctxt, input, XML_CHAR_ENCODING_NONE);
+ if (stream == NULL) {
+ xmlFreeParserInputBuffer(input);
+ xmlFreeParserCtxt(ctxt);
+ return (NULL);
+ }
+ inputPush(ctxt, stream);
+ return (xmlDoRead(ctxt, URL, encoding, options, 0));
+}
+
+/**
+ * xmlReadIO:
+ * @ioread: an I/O read function
+ * @ioclose: an I/O close function
+ * @ioctx: an I/O handler
+ * @URL: the base URL to use for the document
+ * @encoding: the document encoding, or NULL
+ * @options: a combination of xmlParserOption
+ *
+ * parse an XML document from I/O functions and source and build a tree.
+ *
+ * Returns the resulting document tree
+ */
+xmlDocPtr
+xmlReadIO(xmlInputReadCallback ioread, xmlInputCloseCallback ioclose,
+ void *ioctx, const char *URL, const char *encoding, int options)
+{
+ xmlParserCtxtPtr ctxt;
+ xmlParserInputBufferPtr input;
+ xmlParserInputPtr stream;
+
+ if (ioread == NULL)
+ return (NULL);
+ xmlInitParser();
+
+ input = xmlParserInputBufferCreateIO(ioread, ioclose, ioctx,
+ XML_CHAR_ENCODING_NONE);
+ if (input == NULL) {
+ if (ioclose != NULL)
+ ioclose(ioctx);
+ return (NULL);
+ }
+ ctxt = xmlNewParserCtxt();
+ if (ctxt == NULL) {
+ xmlFreeParserInputBuffer(input);
+ return (NULL);
+ }
+ stream = xmlNewIOInputStream(ctxt, input, XML_CHAR_ENCODING_NONE);
+ if (stream == NULL) {
+ xmlFreeParserInputBuffer(input);
+ xmlFreeParserCtxt(ctxt);
+ return (NULL);
+ }
+ inputPush(ctxt, stream);
+ return (xmlDoRead(ctxt, URL, encoding, options, 0));
+}
+
+/**
+ * xmlCtxtReadDoc:
+ * @ctxt: an XML parser context
+ * @cur: a pointer to a zero terminated string
+ * @URL: the base URL to use for the document
+ * @encoding: the document encoding, or NULL
+ * @options: a combination of xmlParserOption
+ *
+ * parse an XML in-memory document and build a tree.
+ * This reuses the existing @ctxt parser context
+ *
+ * Returns the resulting document tree
+ */
+xmlDocPtr
+xmlCtxtReadDoc(xmlParserCtxtPtr ctxt, const xmlChar * cur,
+ const char *URL, const char *encoding, int options)
+{
+ xmlParserInputPtr stream;
+
+ if (cur == NULL)
+ return (NULL);
+ if (ctxt == NULL)
+ return (NULL);
+ xmlInitParser();
+
+ xmlCtxtReset(ctxt);
+
+ stream = xmlNewStringInputStream(ctxt, cur);
+ if (stream == NULL) {
+ return (NULL);
+ }
+ inputPush(ctxt, stream);
+ return (xmlDoRead(ctxt, URL, encoding, options, 1));
+}
+
+/**
+ * xmlCtxtReadFile:
+ * @ctxt: an XML parser context
+ * @filename: a file or URL
+ * @encoding: the document encoding, or NULL
+ * @options: a combination of xmlParserOption
+ *
+ * parse an XML file from the filesystem or the network.
+ * This reuses the existing @ctxt parser context
+ *
+ * Returns the resulting document tree
+ */
+xmlDocPtr
+xmlCtxtReadFile(xmlParserCtxtPtr ctxt, const char *filename,
+ const char *encoding, int options)
+{
+ xmlParserInputPtr stream;
+
+ if (filename == NULL)
+ return (NULL);
+ if (ctxt == NULL)
+ return (NULL);
+ xmlInitParser();
+
+ xmlCtxtReset(ctxt);
+
+ stream = xmlLoadExternalEntity(filename, NULL, ctxt);
+ if (stream == NULL) {
+ return (NULL);
+ }
+ inputPush(ctxt, stream);
+ return (xmlDoRead(ctxt, NULL, encoding, options, 1));
+}
+
+/**
+ * xmlCtxtReadMemory:
+ * @ctxt: an XML parser context
+ * @buffer: a pointer to a char array
+ * @size: the size of the array
+ * @URL: the base URL to use for the document
+ * @encoding: the document encoding, or NULL
+ * @options: a combination of xmlParserOption
+ *
+ * parse an XML in-memory document and build a tree.
+ * This reuses the existing @ctxt parser context
+ *
+ * Returns the resulting document tree
+ */
+xmlDocPtr
+xmlCtxtReadMemory(xmlParserCtxtPtr ctxt, const char *buffer, int size,
+ const char *URL, const char *encoding, int options)
+{
+ xmlParserInputBufferPtr input;
+ xmlParserInputPtr stream;
+
+ if (ctxt == NULL)
+ return (NULL);
+ if (buffer == NULL)
+ return (NULL);
+ xmlInitParser();
+
+ xmlCtxtReset(ctxt);
+
+ input = xmlParserInputBufferCreateMem(buffer, size, XML_CHAR_ENCODING_NONE);
+ if (input == NULL) {
+ return(NULL);
+ }
+
+ stream = xmlNewIOInputStream(ctxt, input, XML_CHAR_ENCODING_NONE);
+ if (stream == NULL) {
+ xmlFreeParserInputBuffer(input);
+ return(NULL);
+ }
+
+ inputPush(ctxt, stream);
+ return (xmlDoRead(ctxt, URL, encoding, options, 1));
+}
+
+/**
+ * xmlCtxtReadFd:
+ * @ctxt: an XML parser context
+ * @fd: an open file descriptor
+ * @URL: the base URL to use for the document
+ * @encoding: the document encoding, or NULL
+ * @options: a combination of xmlParserOption
+ *
+ * parse an XML from a file descriptor and build a tree.
+ * This reuses the existing @ctxt parser context
+ * NOTE that the file descriptor will not be closed when the
+ * reader is closed or reset.
+ *
+ * Returns the resulting document tree
+ */
+xmlDocPtr
+xmlCtxtReadFd(xmlParserCtxtPtr ctxt, int fd,
+ const char *URL, const char *encoding, int options)
+{
+ xmlParserInputBufferPtr input;
+ xmlParserInputPtr stream;
+
+ if (fd < 0)
+ return (NULL);
+ if (ctxt == NULL)
+ return (NULL);
+ xmlInitParser();
+
+ xmlCtxtReset(ctxt);
+
+
+ input = xmlParserInputBufferCreateFd(fd, XML_CHAR_ENCODING_NONE);
+ if (input == NULL)
+ return (NULL);
+ input->closecallback = NULL;
+ stream = xmlNewIOInputStream(ctxt, input, XML_CHAR_ENCODING_NONE);
+ if (stream == NULL) {
+ xmlFreeParserInputBuffer(input);
+ return (NULL);
+ }
+ inputPush(ctxt, stream);
+ return (xmlDoRead(ctxt, URL, encoding, options, 1));
+}
+
+/**
+ * xmlCtxtReadIO:
+ * @ctxt: an XML parser context
+ * @ioread: an I/O read function
+ * @ioclose: an I/O close function
+ * @ioctx: an I/O handler
+ * @URL: the base URL to use for the document
+ * @encoding: the document encoding, or NULL
+ * @options: a combination of xmlParserOption
+ *
+ * parse an XML document from I/O functions and source and build a tree.
+ * This reuses the existing @ctxt parser context
+ *
+ * Returns the resulting document tree
+ */
+xmlDocPtr
+xmlCtxtReadIO(xmlParserCtxtPtr ctxt, xmlInputReadCallback ioread,
+ xmlInputCloseCallback ioclose, void *ioctx,
+ const char *URL,
+ const char *encoding, int options)
+{
+ xmlParserInputBufferPtr input;
+ xmlParserInputPtr stream;
+
+ if (ioread == NULL)
+ return (NULL);
+ if (ctxt == NULL)
+ return (NULL);
+ xmlInitParser();
+
+ xmlCtxtReset(ctxt);
+
+ input = xmlParserInputBufferCreateIO(ioread, ioclose, ioctx,
+ XML_CHAR_ENCODING_NONE);
+ if (input == NULL) {
+ if (ioclose != NULL)
+ ioclose(ioctx);
+ return (NULL);
+ }
+ stream = xmlNewIOInputStream(ctxt, input, XML_CHAR_ENCODING_NONE);
+ if (stream == NULL) {
+ xmlFreeParserInputBuffer(input);
+ return (NULL);
+ }
+ inputPush(ctxt, stream);
+ return (xmlDoRead(ctxt, URL, encoding, options, 1));
+}
+
+#define bottom_parser
+#include "elfgcchack.h"