diff options
| -rw-r--r-- | ChangeLog | 6 | ||||
| -rw-r--r-- | generic/tclPathObj.c | 304 |
2 files changed, 226 insertions, 84 deletions
@@ -1,3 +1,9 @@ +2003-12-14 Vince Darley <vincentdarley@users.sourceforge.net> + + * generic/tclPathObj.c: complete rewrite of generic file + normalization code to cope with links followed by '..'. + [Bug 849514], and parts of [859251] + 2003-12-12 David Gravereaux <davygrvy@pobox.com> * win/tclWinChan.c: Win32's SetFilePointer() takes LONGs not diff --git a/generic/tclPathObj.c b/generic/tclPathObj.c index 40570aa..1402aad 100644 --- a/generic/tclPathObj.c +++ b/generic/tclPathObj.c @@ -10,7 +10,7 @@ * See the file "license.terms" for information on usage and redistribution * of this file, and for a DISCLAIMER OF ALL WARRANTIES. * - * RCS: @(#) $Id: tclPathObj.c,v 1.14 2003/11/24 10:13:36 vincentdarley Exp $ + * RCS: @(#) $Id: tclPathObj.c,v 1.15 2003/12/14 10:51:12 vincentdarley Exp $ */ #include "tclInt.h" @@ -30,7 +30,9 @@ static void FreeFsPathInternalRep _ANSI_ARGS_((Tcl_Obj *listPtr)); static void UpdateStringOfFsPath _ANSI_ARGS_((Tcl_Obj *objPtr)); static int SetFsPathFromAny _ANSI_ARGS_((Tcl_Interp *interp, Tcl_Obj *objPtr)); -static int FindSplitPos _ANSI_ARGS_((char *path, char *separator)); +static int FindSplitPos _ANSI_ARGS_((CONST char *path, + CONST char separator)); +static int IsSeparatorOrNull _ANSI_ARGS_((CONST char ch)); /* @@ -115,6 +117,9 @@ typedef struct FsPath { * * The behaviour of this function if passed a non-absolute path * is NOT defined. + * + * pathPtr may have a refCount of zero, or may be a shared + * object. * * Results: * The result is returned in a Tcl_Obj with a refCount of 1, @@ -125,94 +130,195 @@ typedef struct FsPath { * None (beyond the memory allocation for the result). * * Special note: - * This code is based on code from Matt Newman and Jean-Claude - * Wippler, with additions from Vince Darley and is copyright - * those respective authors. + * This code was originally based on code from Matt Newman and + * Jean-Claude Wippler, but has since been totally rewritten by + * Vince Darley to deal with symbolic links. * *--------------------------------------------------------------------------- */ Tcl_Obj* TclFSNormalizeAbsolutePath(interp, pathPtr, clientDataPtr) - Tcl_Interp* interp; /* Interpreter to use */ - Tcl_Obj *pathPtr; /* Absolute path to normalize */ - ClientData *clientDataPtr; + Tcl_Interp* interp; /* Interpreter to use */ + Tcl_Obj *pathPtr; /* Absolute path to normalize */ + ClientData *clientDataPtr; /* If non-NULL, then may be set to the + * fs-specific clientData for this path. + * This will happen when that extra + * information can be calculated efficiently + * as a side-effect of normalization. */ { - int splen = 0, nplen, eltLen, i; - char *eltName; - Tcl_Obj *retVal; - Tcl_Obj *split; - Tcl_Obj *elt; + ClientData clientData = NULL; + CONST char *dirSep, *oldDirSep; + int first = 1; /* Set to zero once we've passed the first + * directory separator - we can't use '..' to + * remove the volume in a path. */ + Tcl_Obj *retVal = NULL; + dirSep = Tcl_GetString(pathPtr); + + if (tclPlatform == TCL_PLATFORM_WINDOWS) { + if (dirSep[0] != 0 && dirSep[1] == ':' && + (dirSep[2] == '/' || dirSep[2] == '\\')) { + /* Do nothing */ + } else if ((dirSep[0] == '/' || dirSep[0] == '\\') + && (dirSep[1] == '/' || dirSep[1] == '\\')) { + /* + * UNC style path, where we must skip over the + * first separator, since the first two segments + * are actually inseparable. + */ + dirSep += 2; + dirSep += FindSplitPos(dirSep, '/'); + if (*dirSep != 0) { + dirSep++; + } + } + } - /* Split has refCount zero */ - split = Tcl_FSSplitPath(pathPtr, &splen); - /* - * Modify the list of entries in place, by removing '.', and - * removing '..' and the entry before -- unless that entry before - * is the top-level entry, i.e. the name of a volume. + * Scan forward from one directory separator to the next, + * checking for '..' and '.' sequences which must be handled + * specially. In particular handling of '..' can be complicated + * if the directory before is a link, since we will have to + * expand the link to be able to back up one level. */ - nplen = 0; - for (i = 0; i < splen; i++) { - Tcl_ListObjIndex(NULL, split, nplen, &elt); - eltName = Tcl_GetStringFromObj(elt, &eltLen); - - if ((eltLen == 1) && (eltName[0] == '.')) { - Tcl_ListObjReplace(NULL, split, nplen, 1, 0, NULL); - } else if ((eltLen == 2) - && (eltName[0] == '.') && (eltName[1] == '.')) { - if (nplen > 1) { - nplen--; - Tcl_ListObjReplace(NULL, split, nplen, 2, 0, NULL); - } else { - Tcl_ListObjReplace(NULL, split, nplen, 1, 0, NULL); + while (*dirSep != 0) { + oldDirSep = dirSep; + dirSep += 1+FindSplitPos(dirSep+1, '/'); + if (dirSep[0] == 0 || dirSep[1] == 0) { + if (retVal != NULL) { + Tcl_AppendToObj(retVal, oldDirSep, dirSep - oldDirSep); + } + break; + } + if (dirSep[1] == '.') { + if (retVal != NULL) { + Tcl_AppendToObj(retVal, oldDirSep, dirSep - oldDirSep); + } + again: + if (IsSeparatorOrNull(dirSep[2])) { + /* Need to skip '.' in the path */ + if (retVal == NULL) { + CONST char *path = Tcl_GetString(pathPtr); + retVal = Tcl_NewStringObj(path, dirSep - path); + Tcl_IncrRefCount(retVal); + } + dirSep += 2; + if (dirSep[0] != 0 && dirSep[1] == '.') { + goto again; + } + continue; + } + if (dirSep[2] == '.' && IsSeparatorOrNull(dirSep[3])) { + Tcl_Obj *link; + int curLen; + char *linkStr; + /* Have '..' so need to skip previous directory */ + if (retVal == NULL) { + CONST char *path = Tcl_GetString(pathPtr); + retVal = Tcl_NewStringObj(path, dirSep - path); + Tcl_IncrRefCount(retVal); + } + if (!first) { + link = Tcl_FSLink(retVal, NULL, 0); + if (link != NULL) { + /* Got a link */ + Tcl_DecrRefCount(retVal); + retVal = link; + linkStr = Tcl_GetStringFromObj(retVal, &curLen); + /* Convert to forward-slashes on windows */ + if (tclPlatform == TCL_PLATFORM_WINDOWS) { + int i; + for (i = 0; i < curLen; i++) { + if (linkStr[i] == '\\') { + linkStr[i] = '/'; + } + } + } + } else { + linkStr = Tcl_GetStringFromObj(retVal, &curLen); + } + /* Either way, we now remove the last path element */ + while (--curLen > 0) { + if (IsSeparatorOrNull(linkStr[curLen])) { + Tcl_SetObjLength(retVal, curLen); + break; + } + } + } + dirSep += 3; + if (dirSep[0] != 0 && dirSep[1] == '.') { + goto again; + } + continue; } - } else { - nplen++; + } + first = 0; + if (retVal != NULL) { + Tcl_AppendToObj(retVal, oldDirSep, dirSep - oldDirSep); } } - if (nplen > 0) { - ClientData clientData = NULL; - - retVal = Tcl_FSJoinPath(split, nplen); - /* - * Now we have an absolute path, with no '..', '.' sequences, - * but it still may not be in 'unique' form, depending on the - * platform. For instance, Unix is case-sensitive, so the - * path is ok. Windows is case-insensitive, and also has the - * weird 'longname/shortname' thing (e.g. C:/Program Files/ and - * C:/Progra~1/ are equivalent). MacOS is case-insensitive. - * - * Virtual file systems which may be registered may have - * other criteria for normalizing a path. - */ + + /* + * If we didn't make any changes, just use the input path + */ + if (retVal == NULL) { + retVal = pathPtr; Tcl_IncrRefCount(retVal); - TclFSNormalizeToUniquePath(interp, retVal, 0, &clientData); - /* - * Since we know it is a normalized path, we can - * actually convert this object into an FsPath for - * greater efficiency - */ - TclFSMakePathFromNormalized(interp, retVal, clientData); - if (clientDataPtr != NULL) { - *clientDataPtr = clientData; + + if (Tcl_IsShared(retVal)) { + /* + * Unfortunately, the platform-specific normalization code + * which will be called below has no way of dealing with the + * case where an object is shared. It is expecting to + * modify an object in place. So, we must duplicate this + * here to ensure an object with a single ref-count. + * + * If that changes in the future (e.g. the normalize proc is + * given one object and is able to return a different one), + * then we could remove this code. + */ + Tcl_DecrRefCount(retVal); + retVal = Tcl_DuplicateObj(pathPtr); + Tcl_IncrRefCount(retVal); } - } else { - /* Init to an empty string */ - retVal = Tcl_NewStringObj("",0); - Tcl_IncrRefCount(retVal); } + /* - * We increment and then decrement the refCount of split to free - * it. We do this right at the end, in case there are - * optimisations in Tcl_FSJoinPath(split, nplen) above which would - * let it make use of split more effectively if it has a refCount - * of zero. Also we can't just decrement the ref count, in case - * 'split' was actually returned by the join call above, in a - * single-element optimisation when nplen == 1. + * Ensure a windows drive like C:/ has a trailing separator */ - Tcl_IncrRefCount(split); - Tcl_DecrRefCount(split); + if (tclPlatform == TCL_PLATFORM_WINDOWS) { + int len; + CONST char *path = Tcl_GetStringFromObj(retVal, &len); + if (len == 2 && path[0] != 0 && path[1] == ':') { + if (Tcl_IsShared(retVal)) { + Tcl_DecrRefCount(retVal); + retVal = Tcl_DuplicateObj(retVal); + Tcl_IncrRefCount(retVal); + } + Tcl_AppendToObj(retVal, "/", 1); + } + } + /* + * Now we have an absolute path, with no '..', '.' sequences, + * but it still may not be in 'unique' form, depending on the + * platform. For instance, Unix is case-sensitive, so the + * path is ok. Windows is case-insensitive, and also has the + * weird 'longname/shortname' thing (e.g. C:/Program Files/ and + * C:/Progra~1/ are equivalent). MacOS is case-insensitive. + * + * Virtual file systems which may be registered may have + * other criteria for normalizing a path. + */ + TclFSNormalizeToUniquePath(interp, retVal, 0, &clientData); + /* + * Since we know it is a normalized path, we can + * actually convert this object into an FsPath for + * greater efficiency + */ + TclFSMakePathFromNormalized(interp, retVal, clientData); + if (clientDataPtr != NULL) { + *clientDataPtr = clientData; + } /* This has a refCount of 1 for the caller */ return retVal; } @@ -575,20 +681,45 @@ Tcl_FSConvertToPathType(interp, objPtr) } /* + * Helper function for normalization. + */ +static int +IsSeparatorOrNull(ch) + CONST char ch; +{ + if (ch == 0) { + return 1; + } + switch (tclPlatform) { + case TCL_PLATFORM_UNIX: { + return (ch == '/' ? 1 : 0); + } + case TCL_PLATFORM_MAC: { + return (ch == ':' ? 1 : 0); + } + case TCL_PLATFORM_WINDOWS: { + return ((ch == '/' || ch == '\\') ? 1 : 0); + } + } + return 0; +} + +/* * Helper function for SetFsPathFromAny. Returns position of first - * directory delimiter in the path. + * directory delimiter in the path. If no separator is found, then + * returns the position of the end of the string. */ static int FindSplitPos(path, separator) - char *path; - char *separator; + CONST char *path; + CONST char separator; { int count = 0; switch (tclPlatform) { case TCL_PLATFORM_UNIX: case TCL_PLATFORM_MAC: while (path[count] != 0) { - if (path[count] == *separator) { + if (path[count] == separator) { return count; } count++; @@ -597,7 +728,7 @@ FindSplitPos(path, separator) case TCL_PLATFORM_WINDOWS: while (path[count] != 0) { - if (path[count] == *separator || path[count] == '\\') { + if (path[count] == separator || path[count] == '\\') { return count; } count++; @@ -1259,12 +1390,14 @@ Tcl_FSGetNormalizedPath(interp, pathObjPtr) * Path of form C:foo/bar, but this only makes * sense if the cwd is also on drive C. */ - CONST char *drive = Tcl_GetString(useThisCwd); - char drive_c = path[0]; - if (drive_c >= 'a') { - drive_c -= ('a' - 'A'); + int cwdLen; + CONST char *drive = Tcl_GetStringFromObj(useThisCwd, + &cwdLen); + char drive_cur = path[0]; + if (drive_cur >= 'a') { + drive_cur -= ('a' - 'A'); } - if (drive[0] == drive_c) { + if (drive[0] == drive_cur) { absolutePath = Tcl_DuplicateObj(useThisCwd); /* We have a refCount on the cwd */ } else { @@ -1280,7 +1413,10 @@ Tcl_FSGetNormalizedPath(interp, pathObjPtr) absolutePath = Tcl_NewStringObj(path, 2); } Tcl_IncrRefCount(absolutePath); - Tcl_AppendToObj(absolutePath, "/", 1); + if (drive[cwdLen-1] != '/') { + /* Only add a trailing '/' if needed */ + Tcl_AppendToObj(absolutePath, "/", 1); + } Tcl_AppendToObj(absolutePath, path+2, -1); } #endif /* __WIN32__ */ @@ -1623,7 +1759,7 @@ SetFsPathFromAny(interp, objPtr) if (strchr(name, ':') != NULL) separator = ':'; } - split = FindSplitPos(name, &separator); + split = FindSplitPos(name, separator); if (split != len) { /* We have multiple pieces '~user/foo/bar...' */ name[split] = '\0'; |