/* * tclScan.c -- * * This file contains the implementation of the "scan" command. * * Copyright (c) 1998 by Scriptics Corporation. * * See the file "license.terms" for information on usage and redistribution of * this file, and for a DISCLAIMER OF ALL WARRANTIES. */ #include "tclInt.h" /* * Flag values used by Tcl_ScanObjCmd. */ #define SCAN_NOSKIP 0x1 /* Don't skip blanks. */ #define SCAN_SUPPRESS 0x2 /* Suppress assignment. */ #define SCAN_UNSIGNED 0x4 /* Read an unsigned value. */ #define SCAN_WIDTH 0x8 /* A width value was supplied. */ #define SCAN_LONGER 0x400 /* Asked for a wide value. */ #define SCAN_BIG 0x800 /* Asked for a bignum value. */ /* * The following structure contains the information associated with a * character set. */ typedef struct CharSet { int exclude; /* 1 if this is an exclusion set. */ int nchars; Tcl_UniChar *chars; int nranges; struct Range { Tcl_UniChar start; Tcl_UniChar end; } *ranges; } CharSet; /* * Declarations for functions used only in this file. */ static const char * BuildCharSet(CharSet *cset, const char *format); static int CharInSet(CharSet *cset, int ch); static void ReleaseCharSet(CharSet *cset); static int ValidateFormat(Tcl_Interp *interp, const char *format, int numVars, int *totalVars); /* *---------------------------------------------------------------------- * * BuildCharSet -- * * This function examines a character set format specification and builds * a CharSet containing the individual characters and character ranges * specified. * * Results: * Returns the next format position. * * Side effects: * Initializes the charset. * *---------------------------------------------------------------------- */ static const char * BuildCharSet( CharSet *cset, const char *format) /* Points to first char of set. */ { Tcl_UniChar ch = 0, start; int offset, nranges; const char *end; memset(cset, 0, sizeof(CharSet)); offset = TclUtfToUniChar(format, &ch); if (ch == '^') { cset->exclude = 1; format += offset; offset = TclUtfToUniChar(format, &ch); } end = format + offset; /* * Find the close bracket so we can overallocate the set. */ if (ch == ']') { end += TclUtfToUniChar(end, &ch); } nranges = 0; while (ch != ']') { if (ch == '-') { nranges++; } end += TclUtfToUniChar(end, &ch); } cset->chars = ckalloc(sizeof(Tcl_UniChar) * (end - format - 1)); if (nranges > 0) { cset->ranges = ckalloc(sizeof(struct Range) * nranges); } else { cset->ranges = NULL; } /* * Now build the character set. */ cset->nchars = cset->nranges = 0; format += TclUtfToUniChar(format, &ch); start = ch; if (ch == ']' || ch == '-') { cset->chars[cset->nchars++] = ch; format += TclUtfToUniChar(format, &ch); } while (ch != ']') { if (*format == '-') { /* * This may be the first character of a range, so don't add it * yet. */ start = ch; } else if (ch == '-') { /* * Check to see if this is the last character in the set, in which * case it is not a range and we should add the previous character * as well as the dash. */ if (*format == ']' || !cset->ranges) { cset->chars[cset->nchars++] = start; cset->chars[cset->nchars++] = ch; } else { format += TclUtfToUniChar(format, &ch); /* * Check to see if the range is in reverse order. */ if (start < ch) { cset->ranges[cset->nranges].start = start; cset->ranges[cset->nranges].end = ch; } else { cset->ranges[cset->nranges].start = ch; cset->ranges[cset->nranges].end = start; } cset->nranges++; } } else { cset->chars[cset->nchars++] = ch; } format += TclUtfToUniChar(format, &ch); } return format; } /* *---------------------------------------------------------------------- * * CharInSet -- * * Check to see if a character matches the given set. * * Results: * Returns non-zero if the character matches the given set. * * Side effects: * None. * *---------------------------------------------------------------------- */ static int CharInSet( CharSet *cset, int c) /* Character to test, passed as int because of * non-ANSI prototypes. */ { Tcl_UniChar ch = (Tcl_UniChar) c; int i, match = 0; for (i = 0; i < cset->nchars; i++) { if (cset->chars[i] == ch) { match = 1; break; } } if (!match) { for (i = 0; i < cset->nranges; i++) { if ((cset->ranges[i].start <= ch) && (ch <= cset->ranges[i].end)) { match = 1; break; } } } return (cset->exclude ? !match : match); } /* *---------------------------------------------------------------------- * * ReleaseCharSet -- * * Free the storage associated with a character set. * * Results: * None. * * Side effects: * None. * *---------------------------------------------------------------------- */ static void ReleaseCharSet( CharSet *cset) { ckfree(cset->chars); if (cset->ranges) { ckfree(cset->ranges); } } /* *---------------------------------------------------------------------- * * ValidateFormat -- * * Parse the format string and verify that it is properly formed and that * there are exactly enough variables on the command line. * * Results: * A standard Tcl result. * * Side effects: * May place an error in the interpreter result. * *---------------------------------------------------------------------- */ static int ValidateFormat( Tcl_Interp *interp, /* Current interpreter. */ const char *format, /* The format string. */ int numVars, /* The number of variables passed to the scan * command. */ int *totalSubs) /* The number of variables that will be * required. */ { int gotXpg, gotSequential, value, i, flags; char *end; Tcl_UniChar ch = 0; int objIndex, xpgSize, nspace = numVars; int *nassign = TclStackAlloc(interp, nspace * sizeof(int)); char buf[TCL_UTF_MAX+1] = ""; Tcl_Obj *errorMsg; /* Place to build an error messages. Note that * these are messy operations because we do * not want to use the formatting engine; * we're inside there! */ /* * Initialize an array that records the number of times a variable is * assigned to by the format string. We use this to detect if a variable * is multiply assigned or left unassigned. */ for (i = 0; i < nspace; i++) { nassign[i] = 0; } xpgSize = objIndex = gotXpg = gotSequential = 0; while (*format != '\0') { format += TclUtfToUniChar(format, &ch); flags = 0; if (ch != '%') { continue; } format += TclUtfToUniChar(format, &ch); if (ch == '%') { continue; } if (ch == '*') { flags |= SCAN_SUPPRESS; format += TclUtfToUniChar(format, &ch); goto xpgCheckDone; } if ((ch < 0x80) && isdigit(UCHAR(ch))) { /* INTL: "C" locale. */ /* * Check for an XPG3-style %n$ specification. Note: there must * not be a mixture of XPG3 specs and non-XPG3 specs in the same * format string. */ value = strtoul(format-1, &end, 10); /* INTL: "C" locale. */ if (*end != '$') { goto notXpg; } format = end+1; format += TclUtfToUniChar(format, &ch); gotXpg = 1; if (gotSequential) { goto mixedXPG; } objIndex = value - 1; if ((objIndex < 0) || (numVars && (objIndex >= numVars))) { goto badIndex; } else if (numVars == 0) { /* * In the case where no vars are specified, the user can * specify %9999$ legally, so we have to consider special * rules for growing the assign array. 'value' is guaranteed * to be > 0. */ xpgSize = (xpgSize > value) ? xpgSize : value; } goto xpgCheckDone; } notXpg: gotSequential = 1; if (gotXpg) { mixedXPG: Tcl_SetObjResult(interp, Tcl_NewStringObj( "cannot mix \"%\" and \"%n$\" conversion specifiers", -1)); Tcl_SetErrorCode(interp, "TCL", "FORMAT", "MIXEDSPECTYPES", NULL); goto error; } xpgCheckDone: /* * Parse any width specifier. */ if ((ch < 0x80) && isdigit(UCHAR(ch))) { /* INTL: "C" locale. */ value = strtoul(format-1, (char **) &format, 10); /* INTL: "C" locale. */ flags |= SCAN_WIDTH; format += TclUtfToUniChar(format, &ch); } /* * Handle any size specifier. */ switch (ch) { case 'l': if (*format == 'l') { flags |= SCAN_BIG; format += 1; format += TclUtfToUniChar(format, &ch); break; } /* FALLTHRU */ case 'L': flags |= SCAN_LONGER; /* FALLTHRU */ case 'h': format += TclUtfToUniChar(format, &ch); } if (!(flags & SCAN_SUPPRESS) && numVars && (objIndex >= numVars)) { goto badIndex; } /* * Handle the various field types. */ switch (ch) { case 'c': if (flags & SCAN_WIDTH) { Tcl_SetObjResult(interp, Tcl_NewStringObj( "field width may not be specified in %c conversion", -1)); Tcl_SetErrorCode(interp, "TCL", "FORMAT", "BADWIDTH", NULL); goto error; } /* FALLTHRU */ case 'n': case 's': if (flags & (SCAN_LONGER|SCAN_BIG)) { invalidFieldSize: buf[Tcl_UniCharToUtf(ch, buf)] = '\0'; errorMsg = Tcl_NewStringObj( "field size modifier may not be specified in %", -1); Tcl_AppendToObj(errorMsg, buf, -1); Tcl_AppendToObj(errorMsg, " conversion", -1); Tcl_SetObjResult(interp, errorMsg); Tcl_SetErrorCode(interp, "TCL", "FORMAT", "BADSIZE", NULL); goto error; } /* * Fall through! */ case 'd': case 'e': case 'E': case 'f': case 'g': case 'G': case 'i': case 'o': case 'x': case 'X': case 'b': break; case 'u': if (flags & SCAN_BIG) { Tcl_SetObjResult(interp, Tcl_NewStringObj( "unsigned bignum scans are invalid", -1)); Tcl_SetErrorCode(interp, "TCL", "FORMAT", "BADUNSIGNED",NULL); goto error; } break; /* * Bracket terms need special checking */ case '[': if (flags & (SCAN_LONGER|SCAN_BIG)) { goto invalidFieldSize; } if (*format == '\0') { goto badSet; } format += TclUtfToUniChar(format, &ch); if (ch == '^') { if (*format == '\0') { goto badSet; } format += TclUtfToUniChar(format, &ch); } if (ch == ']') { if (*format == '\0') { goto badSet; } format += TclUtfToUniChar(format, &ch); } while (ch != ']') { if (*format == '\0') { goto badSet; } format += TclUtfToUniChar(format, &ch); } break; badSet: Tcl_SetObjResult(interp, Tcl_NewStringObj( "unmatched [ in format string", -1)); Tcl_SetErrorCode(interp, "TCL", "FORMAT", "BRACKET", NULL); goto error; default: buf[Tcl_UniCharToUtf(ch, buf)] = '\0'; errorMsg = Tcl_NewStringObj( "bad scan conversion character \"", -1); Tcl_AppendToObj(errorMsg, buf, -1); Tcl_AppendToObj(errorMsg, "\"", -1); Tcl_SetObjResult(interp, errorMsg); Tcl_SetErrorCode(interp, "TCL", "FORMAT", "BADTYPE", NULL); goto error; } if (!(flags & SCAN_SUPPRESS)) { if (objIndex >= nspace) { /* * Expand the nassign buffer. If we are using XPG specifiers, * make sure that we grow to a large enough size. xpgSize is * guaranteed to be at least one larger than objIndex. */ value = nspace; if (xpgSize) { nspace = xpgSize; } else { nspace += 16; /* formerly STATIC_LIST_SIZE */ } nassign = TclStackRealloc(interp, nassign, nspace * sizeof(int)); for (i = value; i < nspace; i++) { nassign[i] = 0; } } nassign[objIndex]++; objIndex++; } } /* * Verify that all of the variable were assigned exactly once. */ if (numVars == 0) { if (xpgSize) { numVars = xpgSize; } else { numVars = objIndex; } } if (totalSubs) { *totalSubs = numVars; } for (i = 0; i < numVars; i++) { if (nassign[i] > 1) { Tcl_SetObjResult(interp, Tcl_NewStringObj( "variable is assigned by multiple \"%n$\" conversion specifiers", -1)); Tcl_SetErrorCode(interp, "TCL", "FORMAT", "POLYASSIGNED", NULL); goto error; } else if (!xpgSize && (nassign[i] == 0)) { /* * If the space is empty, and xpgSize is 0 (means XPG wasn't used, * and/or numVars != 0), then too many vars were given */ Tcl_SetObjResult(interp, Tcl_NewStringObj( "variable is not assigned by any conversion specifiers", -1)); Tcl_SetErrorCode(interp, "TCL", "FORMAT", "UNASSIGNED", NULL); goto error; } } TclStackFree(interp, nassign); return TCL_OK; badIndex: if (gotXpg) { Tcl_SetObjResult(interp, Tcl_NewStringObj( "\"%n$\" argument index out of range", -1)); Tcl_SetErrorCode(interp, "TCL", "FORMAT", "INDEXRANGE", NULL); } else { Tcl_SetObjResult(interp, Tcl_NewStringObj( "different numbers of variable names and field specifiers", -1)); Tcl_SetErrorCode(interp, "TCL", "FORMAT", "FIELDVARMISMATCH", NULL); } error: TclStackFree(interp, nassign); return TCL_ERROR; } /* *---------------------------------------------------------------------- * * Tcl_ScanObjCmd -- * * This function is invoked to process the "scan" Tcl command. See the * user documentation for details on what it does. * * Results: * A standard Tcl result. * * Side effects: * See the user documentation. * *---------------------------------------------------------------------- */ /* ARGSUSED */ int Tcl_ScanObjCmd( ClientData dummy, /* Not used. */ Tcl_Interp *interp, /* Current interpreter. */ int objc, /* Number of arguments. */ Tcl_Obj *const objv[]) /* Argument objects. */ { const char *format; int numVars, nconversions, totalVars = -1; int objIndex, offset, i, result, code; long value; const char *string, *end, *baseString; char op = 0; int width, underflow = 0; Tcl_WideInt wideValue; Tcl_UniChar ch = 0, sch = 0; Tcl_Obj **objs = NULL, *objPtr = NULL; int flags; if (objc < 3) { Tcl_WrongNumArgs(interp, 1, objv, "string format ?varName ...?"); return TCL_ERROR; } format = Tcl_GetString(objv[2]); numVars = objc-3; /* * Check for errors in the format string. */ if (ValidateFormat(interp, format, numVars, &totalVars) == TCL_ERROR) { return TCL_ERROR; } /* * Allocate space for the result objects. */ if (totalVars > 0) { objs = ckalloc(sizeof(Tcl_Obj *) * totalVars); for (i = 0; i < totalVars; i++) { objs[i] = NULL; } } string = Tcl_GetString(objv[1]); baseString = string; /* * Iterate over the format string filling in the result objects until we * reach the end of input, the end of the format string, or there is a * mismatch. */ objIndex = 0; nconversions = 0; while (*format != '\0') { int parseFlag = TCL_PARSE_NO_WHITESPACE; format += TclUtfToUniChar(format, &ch); flags = 0; /* * If we see whitespace in the format, skip whitespace in the string. */ if (Tcl_UniCharIsSpace(ch)) { offset = TclUtfToUniChar(string, &sch); while (Tcl_UniCharIsSpace(sch)) { if (*string == '\0') { goto done; } string += offset; offset = TclUtfToUniChar(string, &sch); } continue; } if (ch != '%') { literal: if (*string == '\0') { underflow = 1; goto done; } string += TclUtfToUniChar(string, &sch); if (ch != sch) { goto done; } continue; } format += TclUtfToUniChar(format, &ch); if (ch == '%') { goto literal; } /* * Check for assignment suppression ('*') or an XPG3-style assignment * ('%n$'). */ if (ch == '*') { flags |= SCAN_SUPPRESS; format += TclUtfToUniChar(format, &ch); } else if ((ch < 0x80) && isdigit(UCHAR(ch))) { /* INTL: "C" locale. */ char *formatEnd; value = strtoul(format-1, &formatEnd, 10);/* INTL: "C" locale. */ if (*formatEnd == '$') { format = formatEnd+1; format += TclUtfToUniChar(format, &ch); objIndex = (int) value - 1; } } /* * Parse any width specifier. */ if ((ch < 0x80) && isdigit(UCHAR(ch))) { /* INTL: "C" locale. */ width = (int) strtoul(format-1, (char **) &format, 10);/* INTL: "C" locale. */ format += TclUtfToUniChar(format, &ch); } else { width = 0; } /* * Handle any size specifier. */ switch (ch) { case 'l': if (*format == 'l') { flags |= SCAN_BIG; format += 1; format += TclUtfToUniChar(format, &ch); break; } /* FALLTHRU */ case 'L': flags |= SCAN_LONGER; /* FALLTHRU */ case 'h': format += TclUtfToUniChar(format, &ch); } /* * Handle the various field types. */ switch (ch) { case 'n': if (!(flags & SCAN_SUPPRESS)) { objPtr = Tcl_NewIntObj(string - baseString); Tcl_IncrRefCount(objPtr); CLANG_ASSERT(objs); objs[objIndex++] = objPtr; } nconversions++; continue; case 'd': op = 'i'; parseFlag |= TCL_PARSE_DECIMAL_ONLY; break; case 'i': op = 'i'; parseFlag |= TCL_PARSE_SCAN_PREFIXES; break; case 'o': op = 'i'; parseFlag |= TCL_PARSE_OCTAL_ONLY | TCL_PARSE_SCAN_PREFIXES; break; case 'x': case 'X': op = 'i'; parseFlag |= TCL_PARSE_HEXADECIMAL_ONLY; break; case 'b': op = 'i'; parseFlag |= TCL_PARSE_BINARY_ONLY; break; case 'u': op = 'i'; parseFlag |= TCL_PARSE_DECIMAL_ONLY; flags |= SCAN_UNSIGNED; break; case 'f': case 'e': case 'E': case 'g': case 'G': op = 'f'; break; case 's': op = 's'; break; case 'c': op = 'c'; flags |= SCAN_NOSKIP; break; case '[': op = '['; flags |= SCAN_NOSKIP; break; } /* * At this point, we will need additional characters from the string * to proceed. */ if (*string == '\0') { underflow = 1; goto done; } /* * Skip any leading whitespace at the beginning of a field unless the * format suppresses this behavior. */ if (!(flags & SCAN_NOSKIP)) { while (*string != '\0') { offset = TclUtfToUniChar(string, &sch); if (!Tcl_UniCharIsSpace(sch)) { break; } string += offset; } if (*string == '\0') { underflow = 1; goto done; } } /* * Perform the requested scanning operation. */ switch (op) { case 's': /* * Scan a string up to width characters or whitespace. */ if (width == 0) { width = ~0; } end = string; while (*end != '\0') { offset = TclUtfToUniChar(end, &sch); if (Tcl_UniCharIsSpace(sch)) { break; } end += offset; if (--width == 0) { break; } } if (!(flags & SCAN_SUPPRESS)) { objPtr = Tcl_NewStringObj(string, end-string); Tcl_IncrRefCount(objPtr); CLANG_ASSERT(objs); objs[objIndex++] = objPtr; } string = end; break; case '[': { CharSet cset; if (width == 0) { width = ~0; } end = string; format = BuildCharSet(&cset, format); while (*end != '\0') { offset = TclUtfToUniChar(end, &sch); if (!CharInSet(&cset, (int)sch)) { break; } end += offset; if (--width == 0) { break; } } ReleaseCharSet(&cset); if (string == end) { /* * Nothing matched the range, stop processing. */ goto done; } if (!(flags & SCAN_SUPPRESS)) { objPtr = Tcl_NewStringObj(string, end-string); Tcl_IncrRefCount(objPtr); objs[objIndex++] = objPtr; } string = end; break; } case 'c': /* * Scan a single Unicode character. */ offset = TclUtfToUCS4(string, &i); string += offset; if (!(flags & SCAN_SUPPRESS)) { objPtr = Tcl_NewIntObj(i); Tcl_IncrRefCount(objPtr); CLANG_ASSERT(objs); objs[objIndex++] = objPtr; } break; case 'i': /* * Scan an unsigned or signed integer. */ objPtr = Tcl_NewLongObj(0); Tcl_IncrRefCount(objPtr); if (width == 0) { width = ~0; } if (TCL_OK != TclParseNumber(NULL, objPtr, NULL, string, width, &end, TCL_PARSE_INTEGER_ONLY | parseFlag)) { Tcl_DecrRefCount(objPtr); if (width < 0) { if (*end == '\0') { underflow = 1; } } else { if (end == string + width) { underflow = 1; } } goto done; } string = end; if (flags & SCAN_SUPPRESS) { Tcl_DecrRefCount(objPtr); break; } if (flags & SCAN_LONGER) { if (Tcl_GetWideIntFromObj(NULL, objPtr, &wideValue) != TCL_OK) { wideValue = ~(Tcl_WideUInt)0 >> 1; /* WIDE_MAX */ if (TclGetString(objPtr)[0] == '-') { wideValue++; /* WIDE_MAX + 1 = WIDE_MIN */ } } if ((flags & SCAN_UNSIGNED) && (wideValue < 0)) { mp_int big; TclBNInitBignumFromWideUInt(&big, (Tcl_WideUInt)wideValue); Tcl_SetBignumObj(objPtr, &big); } else { Tcl_SetWideIntObj(objPtr, wideValue); } } else if (!(flags & SCAN_BIG)) { if (TclGetLongFromObj(NULL, objPtr, &value) != TCL_OK) { if (TclGetString(objPtr)[0] == '-') { value = LONG_MIN; } else { value = LONG_MAX; } } if ((flags & SCAN_UNSIGNED) && (value < 0)) { #ifdef TCL_WIDE_INT_IS_LONG mp_int big; TclBNInitBignumFromWideUInt(&big, (unsigned long)value); Tcl_SetBignumObj(objPtr, &big); #else Tcl_SetWideIntObj(objPtr, (unsigned long)value); #endif } else { Tcl_SetLongObj(objPtr, value); } } objs[objIndex++] = objPtr; break; case 'f': /* * Scan a floating point number */ objPtr = Tcl_NewDoubleObj(0.0); Tcl_IncrRefCount(objPtr); if (width == 0) { width = ~0; } if (TCL_OK != TclParseNumber(NULL, objPtr, NULL, string, width, &end, TCL_PARSE_DECIMAL_ONLY | TCL_PARSE_NO_WHITESPACE)) { Tcl_DecrRefCount(objPtr); if (width < 0) { if (*end == '\0') { underflow = 1; } } else { if (end == string + width) { underflow = 1; } } goto done; } else if (flags & SCAN_SUPPRESS) { Tcl_DecrRefCount(objPtr); string = end; } else { double dvalue; if (Tcl_GetDoubleFromObj(NULL, objPtr, &dvalue) != TCL_OK) { #ifdef ACCEPT_NAN if (objPtr->typePtr == &tclDoubleType) { dvalue = objPtr->internalRep.doubleValue; } else #endif { Tcl_DecrRefCount(objPtr); goto done; } } Tcl_SetDoubleObj(objPtr, dvalue); CLANG_ASSERT(objs); objs[objIndex++] = objPtr; string = end; } } nconversions++; } done: result = 0; code = TCL_OK; if (numVars) { /* * In this case, variables were specified (classic scan). */ for (i = 0; i < totalVars; i++) { if (objs[i] == NULL) { continue; } result++; /* * In case of multiple errors in setting variables, just report * the first one. */ if (Tcl_ObjSetVar2(interp, objv[i+3], NULL, objs[i], (code == TCL_OK) ? TCL_LEAVE_ERR_MSG : 0) == NULL) { code = TCL_ERROR; } Tcl_DecrRefCount(objs[i]); } } else { /* * Here no vars were specified, we want a list returned (inline scan) */ objPtr = Tcl_NewObj(); for (i = 0; i < totalVars; i++) { if (objs[i] != NULL) { Tcl_ListObjAppendElement(NULL, objPtr, objs[i]); Tcl_DecrRefCount(objs[i]); } else { /* * More %-specifiers than matching chars, so we just spit out * empty strings for these. */ Tcl_ListObjAppendElement(NULL, objPtr, Tcl_NewObj()); } } } if (objs != NULL) { ckfree(objs); } if (code == TCL_OK) { if (underflow && (nconversions == 0)) { if (numVars) { objPtr = Tcl_NewIntObj(-1); } else { if (objPtr) { Tcl_SetListObj(objPtr, 0, NULL); } else { objPtr = Tcl_NewObj(); } } } else if (numVars) { objPtr = Tcl_NewIntObj(result); } Tcl_SetObjResult(interp, objPtr); } return code; } /* * Local Variables: * mode: c * c-basic-offset: 4 * fill-column: 78 * End: */