Skip to content

HTTPS clone URL

Subversion checkout URL

You can clone with
or
.
Download ZIP
Branch: master
Fetching contributors…

Cannot retrieve contributors at this time

8157 lines (7679 sloc) 219.847 kB
/*
* regexp.c: generic and extensible Regular Expression engine
*
* Basically designed with the purpose of compiling regexps for
* the variety of validation/shemas mechanisms now available in
* XML related specifications these include:
* - XML-1.0 DTD validation
* - XML Schemas structure part 1
* - XML Schemas Datatypes part 2 especially Appendix F
* - RELAX-NG/TREX i.e. the counter proposal
*
* See Copyright for the status of this software.
*
* Daniel Veillard <veillard@redhat.com>
*/
#define IN_LIBXML
#include "libxml.h"
#ifdef LIBXML_REGEXP_ENABLED
/* #define DEBUG_ERR */
#include <stdio.h>
#include <string.h>
#ifdef HAVE_LIMITS_H
#include <limits.h>
#endif
#include <libxml/tree.h>
#include <libxml/parserInternals.h>
#include <libxml/xmlregexp.h>
#include <libxml/xmlautomata.h>
#include <libxml/xmlunicode.h>
#ifndef INT_MAX
#define INT_MAX 123456789 /* easy to flag and big enough for our needs */
#endif
/* #define DEBUG_REGEXP_GRAPH */
/* #define DEBUG_REGEXP_EXEC */
/* #define DEBUG_PUSH */
/* #define DEBUG_COMPACTION */
#define MAX_PUSH 10000000
#ifdef ERROR
#undef ERROR
#endif
#define ERROR(str) \
ctxt->error = XML_REGEXP_COMPILE_ERROR; \
xmlRegexpErrCompile(ctxt, str);
#define NEXT ctxt->cur++
#define CUR (*(ctxt->cur))
#define NXT(index) (ctxt->cur[index])
#define CUR_SCHAR(s, l) xmlStringCurrentChar(NULL, s, &l)
#define NEXTL(l) ctxt->cur += l;
#define XML_REG_STRING_SEPARATOR '|'
/*
* Need PREV to check on a '-' within a Character Group. May only be used
* when it's guaranteed that cur is not at the beginning of ctxt->string!
*/
#define PREV (ctxt->cur[-1])
/**
* TODO:
*
* macro to flag unimplemented blocks
*/
#define TODO \
xmlGenericError(xmlGenericErrorContext, \
"Unimplemented block at %s:%d\n", \
__FILE__, __LINE__);
/************************************************************************
* *
* Datatypes and structures *
* *
************************************************************************/
/*
* Note: the order of the enums below is significant, do not shuffle
*/
typedef enum {
XML_REGEXP_EPSILON = 1,
XML_REGEXP_CHARVAL,
XML_REGEXP_RANGES,
XML_REGEXP_SUBREG, /* used for () sub regexps */
XML_REGEXP_STRING,
XML_REGEXP_ANYCHAR, /* . */
XML_REGEXP_ANYSPACE, /* \s */
XML_REGEXP_NOTSPACE, /* \S */
XML_REGEXP_INITNAME, /* \l */
XML_REGEXP_NOTINITNAME, /* \L */
XML_REGEXP_NAMECHAR, /* \c */
XML_REGEXP_NOTNAMECHAR, /* \C */
XML_REGEXP_DECIMAL, /* \d */
XML_REGEXP_NOTDECIMAL, /* \D */
XML_REGEXP_REALCHAR, /* \w */
XML_REGEXP_NOTREALCHAR, /* \W */
XML_REGEXP_LETTER = 100,
XML_REGEXP_LETTER_UPPERCASE,
XML_REGEXP_LETTER_LOWERCASE,
XML_REGEXP_LETTER_TITLECASE,
XML_REGEXP_LETTER_MODIFIER,
XML_REGEXP_LETTER_OTHERS,
XML_REGEXP_MARK,
XML_REGEXP_MARK_NONSPACING,
XML_REGEXP_MARK_SPACECOMBINING,
XML_REGEXP_MARK_ENCLOSING,
XML_REGEXP_NUMBER,
XML_REGEXP_NUMBER_DECIMAL,
XML_REGEXP_NUMBER_LETTER,
XML_REGEXP_NUMBER_OTHERS,
XML_REGEXP_PUNCT,
XML_REGEXP_PUNCT_CONNECTOR,
XML_REGEXP_PUNCT_DASH,
XML_REGEXP_PUNCT_OPEN,
XML_REGEXP_PUNCT_CLOSE,
XML_REGEXP_PUNCT_INITQUOTE,
XML_REGEXP_PUNCT_FINQUOTE,
XML_REGEXP_PUNCT_OTHERS,
XML_REGEXP_SEPAR,
XML_REGEXP_SEPAR_SPACE,
XML_REGEXP_SEPAR_LINE,
XML_REGEXP_SEPAR_PARA,
XML_REGEXP_SYMBOL,
XML_REGEXP_SYMBOL_MATH,
XML_REGEXP_SYMBOL_CURRENCY,
XML_REGEXP_SYMBOL_MODIFIER,
XML_REGEXP_SYMBOL_OTHERS,
XML_REGEXP_OTHER,
XML_REGEXP_OTHER_CONTROL,
XML_REGEXP_OTHER_FORMAT,
XML_REGEXP_OTHER_PRIVATE,
XML_REGEXP_OTHER_NA,
XML_REGEXP_BLOCK_NAME
} xmlRegAtomType;
typedef enum {
XML_REGEXP_QUANT_EPSILON = 1,
XML_REGEXP_QUANT_ONCE,
XML_REGEXP_QUANT_OPT,
XML_REGEXP_QUANT_MULT,
XML_REGEXP_QUANT_PLUS,
XML_REGEXP_QUANT_ONCEONLY,
XML_REGEXP_QUANT_ALL,
XML_REGEXP_QUANT_RANGE
} xmlRegQuantType;
typedef enum {
XML_REGEXP_START_STATE = 1,
XML_REGEXP_FINAL_STATE,
XML_REGEXP_TRANS_STATE,
XML_REGEXP_SINK_STATE,
XML_REGEXP_UNREACH_STATE
} xmlRegStateType;
typedef enum {
XML_REGEXP_MARK_NORMAL = 0,
XML_REGEXP_MARK_START,
XML_REGEXP_MARK_VISITED
} xmlRegMarkedType;
typedef struct _xmlRegRange xmlRegRange;
typedef xmlRegRange *xmlRegRangePtr;
struct _xmlRegRange {
int neg; /* 0 normal, 1 not, 2 exclude */
xmlRegAtomType type;
int start;
int end;
xmlChar *blockName;
};
typedef struct _xmlRegAtom xmlRegAtom;
typedef xmlRegAtom *xmlRegAtomPtr;
typedef struct _xmlAutomataState xmlRegState;
typedef xmlRegState *xmlRegStatePtr;
struct _xmlRegAtom {
int no;
xmlRegAtomType type;
xmlRegQuantType quant;
int min;
int max;
void *valuep;
void *valuep2;
int neg;
int codepoint;
xmlRegStatePtr start;
xmlRegStatePtr start0;
xmlRegStatePtr stop;
int maxRanges;
int nbRanges;
xmlRegRangePtr *ranges;
void *data;
};
typedef struct _xmlRegCounter xmlRegCounter;
typedef xmlRegCounter *xmlRegCounterPtr;
struct _xmlRegCounter {
int min;
int max;
};
typedef struct _xmlRegTrans xmlRegTrans;
typedef xmlRegTrans *xmlRegTransPtr;
struct _xmlRegTrans {
xmlRegAtomPtr atom;
int to;
int counter;
int count;
int nd;
};
struct _xmlAutomataState {
xmlRegStateType type;
xmlRegMarkedType mark;
xmlRegMarkedType reached;
int no;
int maxTrans;
int nbTrans;
xmlRegTrans *trans;
/* knowing states ponting to us can speed things up */
int maxTransTo;
int nbTransTo;
int *transTo;
};
typedef struct _xmlAutomata xmlRegParserCtxt;
typedef xmlRegParserCtxt *xmlRegParserCtxtPtr;
#define AM_AUTOMATA_RNG 1
struct _xmlAutomata {
xmlChar *string;
xmlChar *cur;
int error;
int neg;
xmlRegStatePtr start;
xmlRegStatePtr end;
xmlRegStatePtr state;
xmlRegAtomPtr atom;
int maxAtoms;
int nbAtoms;
xmlRegAtomPtr *atoms;
int maxStates;
int nbStates;
xmlRegStatePtr *states;
int maxCounters;
int nbCounters;
xmlRegCounter *counters;
int determinist;
int negs;
int flags;
};
struct _xmlRegexp {
xmlChar *string;
int nbStates;
xmlRegStatePtr *states;
int nbAtoms;
xmlRegAtomPtr *atoms;
int nbCounters;
xmlRegCounter *counters;
int determinist;
int flags;
/*
* That's the compact form for determinists automatas
*/
int nbstates;
int *compact;
void **transdata;
int nbstrings;
xmlChar **stringMap;
};
typedef struct _xmlRegExecRollback xmlRegExecRollback;
typedef xmlRegExecRollback *xmlRegExecRollbackPtr;
struct _xmlRegExecRollback {
xmlRegStatePtr state;/* the current state */
int index; /* the index in the input stack */
int nextbranch; /* the next transition to explore in that state */
int *counts; /* save the automata state if it has some */
};
typedef struct _xmlRegInputToken xmlRegInputToken;
typedef xmlRegInputToken *xmlRegInputTokenPtr;
struct _xmlRegInputToken {
xmlChar *value;
void *data;
};
struct _xmlRegExecCtxt {
int status; /* execution status != 0 indicate an error */
int determinist; /* did we find an indeterministic behaviour */
xmlRegexpPtr comp; /* the compiled regexp */
xmlRegExecCallbacks callback;
void *data;
xmlRegStatePtr state;/* the current state */
int transno; /* the current transition on that state */
int transcount; /* the number of chars in char counted transitions */
/*
* A stack of rollback states
*/
int maxRollbacks;
int nbRollbacks;
xmlRegExecRollback *rollbacks;
/*
* The state of the automata if any
*/
int *counts;
/*
* The input stack
*/
int inputStackMax;
int inputStackNr;
int index;
int *charStack;
const xmlChar *inputString; /* when operating on characters */
xmlRegInputTokenPtr inputStack;/* when operating on strings */
/*
* error handling
*/
int errStateNo; /* the error state number */
xmlRegStatePtr errState; /* the error state */
xmlChar *errString; /* the string raising the error */
int *errCounts; /* counters at the error state */
int nbPush;
};
#define REGEXP_ALL_COUNTER 0x123456
#define REGEXP_ALL_LAX_COUNTER 0x123457
static void xmlFAParseRegExp(xmlRegParserCtxtPtr ctxt, int top);
static void xmlRegFreeState(xmlRegStatePtr state);
static void xmlRegFreeAtom(xmlRegAtomPtr atom);
static int xmlRegStrEqualWildcard(const xmlChar *expStr, const xmlChar *valStr);
static int xmlRegCheckCharacter(xmlRegAtomPtr atom, int codepoint);
static int xmlRegCheckCharacterRange(xmlRegAtomType type, int codepoint,
int neg, int start, int end, const xmlChar *blockName);
void xmlAutomataSetFlags(xmlAutomataPtr am, int flags);
/************************************************************************
* *
* Regexp memory error handler *
* *
************************************************************************/
/**
* xmlRegexpErrMemory:
* @extra: extra information
*
* Handle an out of memory condition
*/
static void
xmlRegexpErrMemory(xmlRegParserCtxtPtr ctxt, const char *extra)
{
const char *regexp = NULL;
if (ctxt != NULL) {
regexp = (const char *) ctxt->string;
ctxt->error = XML_ERR_NO_MEMORY;
}
__xmlRaiseError(NULL, NULL, NULL, NULL, NULL, XML_FROM_REGEXP,
XML_ERR_NO_MEMORY, XML_ERR_FATAL, NULL, 0, extra,
regexp, NULL, 0, 0,
"Memory allocation failed : %s\n", extra);
}
/**
* xmlRegexpErrCompile:
* @extra: extra information
*
* Handle a compilation failure
*/
static void
xmlRegexpErrCompile(xmlRegParserCtxtPtr ctxt, const char *extra)
{
const char *regexp = NULL;
int idx = 0;
if (ctxt != NULL) {
regexp = (const char *) ctxt->string;
idx = ctxt->cur - ctxt->string;
ctxt->error = XML_REGEXP_COMPILE_ERROR;
}
__xmlRaiseError(NULL, NULL, NULL, NULL, NULL, XML_FROM_REGEXP,
XML_REGEXP_COMPILE_ERROR, XML_ERR_FATAL, NULL, 0, extra,
regexp, NULL, idx, 0,
"failed to compile: %s\n", extra);
}
/************************************************************************
* *
* Allocation/Deallocation *
* *
************************************************************************/
static int xmlFAComputesDeterminism(xmlRegParserCtxtPtr ctxt);
/**
* xmlRegEpxFromParse:
* @ctxt: the parser context used to build it
*
* Allocate a new regexp and fill it with the result from the parser
*
* Returns the new regexp or NULL in case of error
*/
static xmlRegexpPtr
xmlRegEpxFromParse(xmlRegParserCtxtPtr ctxt) {
xmlRegexpPtr ret;
ret = (xmlRegexpPtr) xmlMalloc(sizeof(xmlRegexp));
if (ret == NULL) {
xmlRegexpErrMemory(ctxt, "compiling regexp");
return(NULL);
}
memset(ret, 0, sizeof(xmlRegexp));
ret->string = ctxt->string;
ret->nbStates = ctxt->nbStates;
ret->states = ctxt->states;
ret->nbAtoms = ctxt->nbAtoms;
ret->atoms = ctxt->atoms;
ret->nbCounters = ctxt->nbCounters;
ret->counters = ctxt->counters;
ret->determinist = ctxt->determinist;
ret->flags = ctxt->flags;
if (ret->determinist == -1) {
xmlRegexpIsDeterminist(ret);
}
if ((ret->determinist != 0) &&
(ret->nbCounters == 0) &&
(ctxt->negs == 0) &&
(ret->atoms != NULL) &&
(ret->atoms[0] != NULL) &&
(ret->atoms[0]->type == XML_REGEXP_STRING)) {
int i, j, nbstates = 0, nbatoms = 0;
int *stateRemap;
int *stringRemap;
int *transitions;
void **transdata;
xmlChar **stringMap;
xmlChar *value;
/*
* Switch to a compact representation
* 1/ counting the effective number of states left
* 2/ counting the unique number of atoms, and check that
* they are all of the string type
* 3/ build a table state x atom for the transitions
*/
stateRemap = xmlMalloc(ret->nbStates * sizeof(int));
if (stateRemap == NULL) {
xmlRegexpErrMemory(ctxt, "compiling regexp");
xmlFree(ret);
return(NULL);
}
for (i = 0;i < ret->nbStates;i++) {
if (ret->states[i] != NULL) {
stateRemap[i] = nbstates;
nbstates++;
} else {
stateRemap[i] = -1;
}
}
#ifdef DEBUG_COMPACTION
printf("Final: %d states\n", nbstates);
#endif
stringMap = xmlMalloc(ret->nbAtoms * sizeof(char *));
if (stringMap == NULL) {
xmlRegexpErrMemory(ctxt, "compiling regexp");
xmlFree(stateRemap);
xmlFree(ret);
return(NULL);
}
stringRemap = xmlMalloc(ret->nbAtoms * sizeof(int));
if (stringRemap == NULL) {
xmlRegexpErrMemory(ctxt, "compiling regexp");
xmlFree(stringMap);
xmlFree(stateRemap);
xmlFree(ret);
return(NULL);
}
for (i = 0;i < ret->nbAtoms;i++) {
if ((ret->atoms[i]->type == XML_REGEXP_STRING) &&
(ret->atoms[i]->quant == XML_REGEXP_QUANT_ONCE)) {
value = ret->atoms[i]->valuep;
for (j = 0;j < nbatoms;j++) {
if (xmlStrEqual(stringMap[j], value)) {
stringRemap[i] = j;
break;
}
}
if (j >= nbatoms) {
stringRemap[i] = nbatoms;
stringMap[nbatoms] = xmlStrdup(value);
if (stringMap[nbatoms] == NULL) {
for (i = 0;i < nbatoms;i++)
xmlFree(stringMap[i]);
xmlFree(stringRemap);
xmlFree(stringMap);
xmlFree(stateRemap);
xmlFree(ret);
return(NULL);
}
nbatoms++;
}
} else {
xmlFree(stateRemap);
xmlFree(stringRemap);
for (i = 0;i < nbatoms;i++)
xmlFree(stringMap[i]);
xmlFree(stringMap);
xmlFree(ret);
return(NULL);
}
}
#ifdef DEBUG_COMPACTION
printf("Final: %d atoms\n", nbatoms);
#endif
transitions = (int *) xmlMalloc((nbstates + 1) *
(nbatoms + 1) * sizeof(int));
if (transitions == NULL) {
xmlFree(stateRemap);
xmlFree(stringRemap);
xmlFree(stringMap);
xmlFree(ret);
return(NULL);
}
memset(transitions, 0, (nbstates + 1) * (nbatoms + 1) * sizeof(int));
/*
* Allocate the transition table. The first entry for each
* state corresponds to the state type.
*/
transdata = NULL;
for (i = 0;i < ret->nbStates;i++) {
int stateno, atomno, targetno, prev;
xmlRegStatePtr state;
xmlRegTransPtr trans;
stateno = stateRemap[i];
if (stateno == -1)
continue;
state = ret->states[i];
transitions[stateno * (nbatoms + 1)] = state->type;
for (j = 0;j < state->nbTrans;j++) {
trans = &(state->trans[j]);
if ((trans->to == -1) || (trans->atom == NULL))
continue;
atomno = stringRemap[trans->atom->no];
if ((trans->atom->data != NULL) && (transdata == NULL)) {
transdata = (void **) xmlMalloc(nbstates * nbatoms *
sizeof(void *));
if (transdata != NULL)
memset(transdata, 0,
nbstates * nbatoms * sizeof(void *));
else {
xmlRegexpErrMemory(ctxt, "compiling regexp");
break;
}
}
targetno = stateRemap[trans->to];
/*
* if the same atom can generate transitions to 2 different
* states then it means the automata is not determinist and
* the compact form can't be used !
*/
prev = transitions[stateno * (nbatoms + 1) + atomno + 1];
if (prev != 0) {
if (prev != targetno + 1) {
ret->determinist = 0;
#ifdef DEBUG_COMPACTION
printf("Indet: state %d trans %d, atom %d to %d : %d to %d\n",
i, j, trans->atom->no, trans->to, atomno, targetno);
printf(" previous to is %d\n", prev);
#endif
if (transdata != NULL)
xmlFree(transdata);
xmlFree(transitions);
xmlFree(stateRemap);
xmlFree(stringRemap);
for (i = 0;i < nbatoms;i++)
xmlFree(stringMap[i]);
xmlFree(stringMap);
goto not_determ;
}
} else {
#if 0
printf("State %d trans %d: atom %d to %d : %d to %d\n",
i, j, trans->atom->no, trans->to, atomno, targetno);
#endif
transitions[stateno * (nbatoms + 1) + atomno + 1] =
targetno + 1; /* to avoid 0 */
if (transdata != NULL)
transdata[stateno * nbatoms + atomno] =
trans->atom->data;
}
}
}
ret->determinist = 1;
#ifdef DEBUG_COMPACTION
/*
* Debug
*/
for (i = 0;i < nbstates;i++) {
for (j = 0;j < nbatoms + 1;j++) {
printf("%02d ", transitions[i * (nbatoms + 1) + j]);
}
printf("\n");
}
printf("\n");
#endif
/*
* Cleanup of the old data
*/
if (ret->states != NULL) {
for (i = 0;i < ret->nbStates;i++)
xmlRegFreeState(ret->states[i]);
xmlFree(ret->states);
}
ret->states = NULL;
ret->nbStates = 0;
if (ret->atoms != NULL) {
for (i = 0;i < ret->nbAtoms;i++)
xmlRegFreeAtom(ret->atoms[i]);
xmlFree(ret->atoms);
}
ret->atoms = NULL;
ret->nbAtoms = 0;
ret->compact = transitions;
ret->transdata = transdata;
ret->stringMap = stringMap;
ret->nbstrings = nbatoms;
ret->nbstates = nbstates;
xmlFree(stateRemap);
xmlFree(stringRemap);
}
not_determ:
ctxt->string = NULL;
ctxt->nbStates = 0;
ctxt->states = NULL;
ctxt->nbAtoms = 0;
ctxt->atoms = NULL;
ctxt->nbCounters = 0;
ctxt->counters = NULL;
return(ret);
}
/**
* xmlRegNewParserCtxt:
* @string: the string to parse
*
* Allocate a new regexp parser context
*
* Returns the new context or NULL in case of error
*/
static xmlRegParserCtxtPtr
xmlRegNewParserCtxt(const xmlChar *string) {
xmlRegParserCtxtPtr ret;
ret = (xmlRegParserCtxtPtr) xmlMalloc(sizeof(xmlRegParserCtxt));
if (ret == NULL)
return(NULL);
memset(ret, 0, sizeof(xmlRegParserCtxt));
if (string != NULL)
ret->string = xmlStrdup(string);
ret->cur = ret->string;
ret->neg = 0;
ret->negs = 0;
ret->error = 0;
ret->determinist = -1;
return(ret);
}
/**
* xmlRegNewRange:
* @ctxt: the regexp parser context
* @neg: is that negative
* @type: the type of range
* @start: the start codepoint
* @end: the end codepoint
*
* Allocate a new regexp range
*
* Returns the new range or NULL in case of error
*/
static xmlRegRangePtr
xmlRegNewRange(xmlRegParserCtxtPtr ctxt,
int neg, xmlRegAtomType type, int start, int end) {
xmlRegRangePtr ret;
ret = (xmlRegRangePtr) xmlMalloc(sizeof(xmlRegRange));
if (ret == NULL) {
xmlRegexpErrMemory(ctxt, "allocating range");
return(NULL);
}
ret->neg = neg;
ret->type = type;
ret->start = start;
ret->end = end;
return(ret);
}
/**
* xmlRegFreeRange:
* @range: the regexp range
*
* Free a regexp range
*/
static void
xmlRegFreeRange(xmlRegRangePtr range) {
if (range == NULL)
return;
if (range->blockName != NULL)
xmlFree(range->blockName);
xmlFree(range);
}
/**
* xmlRegCopyRange:
* @range: the regexp range
*
* Copy a regexp range
*
* Returns the new copy or NULL in case of error.
*/
static xmlRegRangePtr
xmlRegCopyRange(xmlRegParserCtxtPtr ctxt, xmlRegRangePtr range) {
xmlRegRangePtr ret;
if (range == NULL)
return(NULL);
ret = xmlRegNewRange(ctxt, range->neg, range->type, range->start,
range->end);
if (ret == NULL)
return(NULL);
if (range->blockName != NULL) {
ret->blockName = xmlStrdup(range->blockName);
if (ret->blockName == NULL) {
xmlRegexpErrMemory(ctxt, "allocating range");
xmlRegFreeRange(ret);
return(NULL);
}
}
return(ret);
}
/**
* xmlRegNewAtom:
* @ctxt: the regexp parser context
* @type: the type of atom
*
* Allocate a new atom
*
* Returns the new atom or NULL in case of error
*/
static xmlRegAtomPtr
xmlRegNewAtom(xmlRegParserCtxtPtr ctxt, xmlRegAtomType type) {
xmlRegAtomPtr ret;
ret = (xmlRegAtomPtr) xmlMalloc(sizeof(xmlRegAtom));
if (ret == NULL) {
xmlRegexpErrMemory(ctxt, "allocating atom");
return(NULL);
}
memset(ret, 0, sizeof(xmlRegAtom));
ret->type = type;
ret->quant = XML_REGEXP_QUANT_ONCE;
ret->min = 0;
ret->max = 0;
return(ret);
}
/**
* xmlRegFreeAtom:
* @atom: the regexp atom
*
* Free a regexp atom
*/
static void
xmlRegFreeAtom(xmlRegAtomPtr atom) {
int i;
if (atom == NULL)
return;
for (i = 0;i < atom->nbRanges;i++)
xmlRegFreeRange(atom->ranges[i]);
if (atom->ranges != NULL)
xmlFree(atom->ranges);
if ((atom->type == XML_REGEXP_STRING) && (atom->valuep != NULL))
xmlFree(atom->valuep);
if ((atom->type == XML_REGEXP_STRING) && (atom->valuep2 != NULL))
xmlFree(atom->valuep2);
if ((atom->type == XML_REGEXP_BLOCK_NAME) && (atom->valuep != NULL))
xmlFree(atom->valuep);
xmlFree(atom);
}
/**
* xmlRegCopyAtom:
* @ctxt: the regexp parser context
* @atom: the oiginal atom
*
* Allocate a new regexp range
*
* Returns the new atom or NULL in case of error
*/
static xmlRegAtomPtr
xmlRegCopyAtom(xmlRegParserCtxtPtr ctxt, xmlRegAtomPtr atom) {
xmlRegAtomPtr ret;
ret = (xmlRegAtomPtr) xmlMalloc(sizeof(xmlRegAtom));
if (ret == NULL) {
xmlRegexpErrMemory(ctxt, "copying atom");
return(NULL);
}
memset(ret, 0, sizeof(xmlRegAtom));
ret->type = atom->type;
ret->quant = atom->quant;
ret->min = atom->min;
ret->max = atom->max;
if (atom->nbRanges > 0) {
int i;
ret->ranges = (xmlRegRangePtr *) xmlMalloc(sizeof(xmlRegRangePtr) *
atom->nbRanges);
if (ret->ranges == NULL) {
xmlRegexpErrMemory(ctxt, "copying atom");
goto error;
}
for (i = 0;i < atom->nbRanges;i++) {
ret->ranges[i] = xmlRegCopyRange(ctxt, atom->ranges[i]);
if (ret->ranges[i] == NULL)
goto error;
ret->nbRanges = i + 1;
}
}
return(ret);
error:
xmlRegFreeAtom(ret);
return(NULL);
}
static xmlRegStatePtr
xmlRegNewState(xmlRegParserCtxtPtr ctxt) {
xmlRegStatePtr ret;
ret = (xmlRegStatePtr) xmlMalloc(sizeof(xmlRegState));
if (ret == NULL) {
xmlRegexpErrMemory(ctxt, "allocating state");
return(NULL);
}
memset(ret, 0, sizeof(xmlRegState));
ret->type = XML_REGEXP_TRANS_STATE;
ret->mark = XML_REGEXP_MARK_NORMAL;
return(ret);
}
/**
* xmlRegFreeState:
* @state: the regexp state
*
* Free a regexp state
*/
static void
xmlRegFreeState(xmlRegStatePtr state) {
if (state == NULL)
return;
if (state->trans != NULL)
xmlFree(state->trans);
if (state->transTo != NULL)
xmlFree(state->transTo);
xmlFree(state);
}
/**
* xmlRegFreeParserCtxt:
* @ctxt: the regexp parser context
*
* Free a regexp parser context
*/
static void
xmlRegFreeParserCtxt(xmlRegParserCtxtPtr ctxt) {
int i;
if (ctxt == NULL)
return;
if (ctxt->string != NULL)
xmlFree(ctxt->string);
if (ctxt->states != NULL) {
for (i = 0;i < ctxt->nbStates;i++)
xmlRegFreeState(ctxt->states[i]);
xmlFree(ctxt->states);
}
if (ctxt->atoms != NULL) {
for (i = 0;i < ctxt->nbAtoms;i++)
xmlRegFreeAtom(ctxt->atoms[i]);
xmlFree(ctxt->atoms);
}
if (ctxt->counters != NULL)
xmlFree(ctxt->counters);
xmlFree(ctxt);
}
/************************************************************************
* *
* Display of Data structures *
* *
************************************************************************/
static void
xmlRegPrintAtomType(FILE *output, xmlRegAtomType type) {
switch (type) {
case XML_REGEXP_EPSILON:
fprintf(output, "epsilon "); break;
case XML_REGEXP_CHARVAL:
fprintf(output, "charval "); break;
case XML_REGEXP_RANGES:
fprintf(output, "ranges "); break;
case XML_REGEXP_SUBREG:
fprintf(output, "subexpr "); break;
case XML_REGEXP_STRING:
fprintf(output, "string "); break;
case XML_REGEXP_ANYCHAR:
fprintf(output, "anychar "); break;
case XML_REGEXP_ANYSPACE:
fprintf(output, "anyspace "); break;
case XML_REGEXP_NOTSPACE:
fprintf(output, "notspace "); break;
case XML_REGEXP_INITNAME:
fprintf(output, "initname "); break;
case XML_REGEXP_NOTINITNAME:
fprintf(output, "notinitname "); break;
case XML_REGEXP_NAMECHAR:
fprintf(output, "namechar "); break;
case XML_REGEXP_NOTNAMECHAR:
fprintf(output, "notnamechar "); break;
case XML_REGEXP_DECIMAL:
fprintf(output, "decimal "); break;
case XML_REGEXP_NOTDECIMAL:
fprintf(output, "notdecimal "); break;
case XML_REGEXP_REALCHAR:
fprintf(output, "realchar "); break;
case XML_REGEXP_NOTREALCHAR:
fprintf(output, "notrealchar "); break;
case XML_REGEXP_LETTER:
fprintf(output, "LETTER "); break;
case XML_REGEXP_LETTER_UPPERCASE:
fprintf(output, "LETTER_UPPERCASE "); break;
case XML_REGEXP_LETTER_LOWERCASE:
fprintf(output, "LETTER_LOWERCASE "); break;
case XML_REGEXP_LETTER_TITLECASE:
fprintf(output, "LETTER_TITLECASE "); break;
case XML_REGEXP_LETTER_MODIFIER:
fprintf(output, "LETTER_MODIFIER "); break;
case XML_REGEXP_LETTER_OTHERS:
fprintf(output, "LETTER_OTHERS "); break;
case XML_REGEXP_MARK:
fprintf(output, "MARK "); break;
case XML_REGEXP_MARK_NONSPACING:
fprintf(output, "MARK_NONSPACING "); break;
case XML_REGEXP_MARK_SPACECOMBINING:
fprintf(output, "MARK_SPACECOMBINING "); break;
case XML_REGEXP_MARK_ENCLOSING:
fprintf(output, "MARK_ENCLOSING "); break;
case XML_REGEXP_NUMBER:
fprintf(output, "NUMBER "); break;
case XML_REGEXP_NUMBER_DECIMAL:
fprintf(output, "NUMBER_DECIMAL "); break;
case XML_REGEXP_NUMBER_LETTER:
fprintf(output, "NUMBER_LETTER "); break;
case XML_REGEXP_NUMBER_OTHERS:
fprintf(output, "NUMBER_OTHERS "); break;
case XML_REGEXP_PUNCT:
fprintf(output, "PUNCT "); break;
case XML_REGEXP_PUNCT_CONNECTOR:
fprintf(output, "PUNCT_CONNECTOR "); break;
case XML_REGEXP_PUNCT_DASH:
fprintf(output, "PUNCT_DASH "); break;
case XML_REGEXP_PUNCT_OPEN:
fprintf(output, "PUNCT_OPEN "); break;
case XML_REGEXP_PUNCT_CLOSE:
fprintf(output, "PUNCT_CLOSE "); break;
case XML_REGEXP_PUNCT_INITQUOTE:
fprintf(output, "PUNCT_INITQUOTE "); break;
case XML_REGEXP_PUNCT_FINQUOTE:
fprintf(output, "PUNCT_FINQUOTE "); break;
case XML_REGEXP_PUNCT_OTHERS:
fprintf(output, "PUNCT_OTHERS "); break;
case XML_REGEXP_SEPAR:
fprintf(output, "SEPAR "); break;
case XML_REGEXP_SEPAR_SPACE:
fprintf(output, "SEPAR_SPACE "); break;
case XML_REGEXP_SEPAR_LINE:
fprintf(output, "SEPAR_LINE "); break;
case XML_REGEXP_SEPAR_PARA:
fprintf(output, "SEPAR_PARA "); break;
case XML_REGEXP_SYMBOL:
fprintf(output, "SYMBOL "); break;
case XML_REGEXP_SYMBOL_MATH:
fprintf(output, "SYMBOL_MATH "); break;
case XML_REGEXP_SYMBOL_CURRENCY:
fprintf(output, "SYMBOL_CURRENCY "); break;
case XML_REGEXP_SYMBOL_MODIFIER:
fprintf(output, "SYMBOL_MODIFIER "); break;
case XML_REGEXP_SYMBOL_OTHERS:
fprintf(output, "SYMBOL_OTHERS "); break;
case XML_REGEXP_OTHER:
fprintf(output, "OTHER "); break;
case XML_REGEXP_OTHER_CONTROL:
fprintf(output, "OTHER_CONTROL "); break;
case XML_REGEXP_OTHER_FORMAT:
fprintf(output, "OTHER_FORMAT "); break;
case XML_REGEXP_OTHER_PRIVATE:
fprintf(output, "OTHER_PRIVATE "); break;
case XML_REGEXP_OTHER_NA:
fprintf(output, "OTHER_NA "); break;
case XML_REGEXP_BLOCK_NAME:
fprintf(output, "BLOCK "); break;
}
}
static void
xmlRegPrintQuantType(FILE *output, xmlRegQuantType type) {
switch (type) {
case XML_REGEXP_QUANT_EPSILON:
fprintf(output, "epsilon "); break;
case XML_REGEXP_QUANT_ONCE:
fprintf(output, "once "); break;
case XML_REGEXP_QUANT_OPT:
fprintf(output, "? "); break;
case XML_REGEXP_QUANT_MULT:
fprintf(output, "* "); break;
case XML_REGEXP_QUANT_PLUS:
fprintf(output, "+ "); break;
case XML_REGEXP_QUANT_RANGE:
fprintf(output, "range "); break;
case XML_REGEXP_QUANT_ONCEONLY:
fprintf(output, "onceonly "); break;
case XML_REGEXP_QUANT_ALL:
fprintf(output, "all "); break;
}
}
static void
xmlRegPrintRange(FILE *output, xmlRegRangePtr range) {
fprintf(output, " range: ");
if (range->neg)
fprintf(output, "negative ");
xmlRegPrintAtomType(output, range->type);
fprintf(output, "%c - %c\n", range->start, range->end);
}
static void
xmlRegPrintAtom(FILE *output, xmlRegAtomPtr atom) {
fprintf(output, " atom: ");
if (atom == NULL) {
fprintf(output, "NULL\n");
return;
}
if (atom->neg)
fprintf(output, "not ");
xmlRegPrintAtomType(output, atom->type);
xmlRegPrintQuantType(output, atom->quant);
if (atom->quant == XML_REGEXP_QUANT_RANGE)
fprintf(output, "%d-%d ", atom->min, atom->max);
if (atom->type == XML_REGEXP_STRING)
fprintf(output, "'%s' ", (char *) atom->valuep);
if (atom->type == XML_REGEXP_CHARVAL)
fprintf(output, "char %c\n", atom->codepoint);
else if (atom->type == XML_REGEXP_RANGES) {
int i;
fprintf(output, "%d entries\n", atom->nbRanges);
for (i = 0; i < atom->nbRanges;i++)
xmlRegPrintRange(output, atom->ranges[i]);
} else if (atom->type == XML_REGEXP_SUBREG) {
fprintf(output, "start %d end %d\n", atom->start->no, atom->stop->no);
} else {
fprintf(output, "\n");
}
}
static void
xmlRegPrintTrans(FILE *output, xmlRegTransPtr trans) {
fprintf(output, " trans: ");
if (trans == NULL) {
fprintf(output, "NULL\n");
return;
}
if (trans->to < 0) {
fprintf(output, "removed\n");
return;
}
if (trans->nd != 0) {
if (trans->nd == 2)
fprintf(output, "last not determinist, ");
else
fprintf(output, "not determinist, ");
}
if (trans->counter >= 0) {
fprintf(output, "counted %d, ", trans->counter);
}
if (trans->count == REGEXP_ALL_COUNTER) {
fprintf(output, "all transition, ");
} else if (trans->count >= 0) {
fprintf(output, "count based %d, ", trans->count);
}
if (trans->atom == NULL) {
fprintf(output, "epsilon to %d\n", trans->to);
return;
}
if (trans->atom->type == XML_REGEXP_CHARVAL)
fprintf(output, "char %c ", trans->atom->codepoint);
fprintf(output, "atom %d, to %d\n", trans->atom->no, trans->to);
}
static void
xmlRegPrintState(FILE *output, xmlRegStatePtr state) {
int i;
fprintf(output, " state: ");
if (state == NULL) {
fprintf(output, "NULL\n");
return;
}
if (state->type == XML_REGEXP_START_STATE)
fprintf(output, "START ");
if (state->type == XML_REGEXP_FINAL_STATE)
fprintf(output, "FINAL ");
fprintf(output, "%d, %d transitions:\n", state->no, state->nbTrans);
for (i = 0;i < state->nbTrans; i++) {
xmlRegPrintTrans(output, &(state->trans[i]));
}
}
#ifdef DEBUG_REGEXP_GRAPH
static void
xmlRegPrintCtxt(FILE *output, xmlRegParserCtxtPtr ctxt) {
int i;
fprintf(output, " ctxt: ");
if (ctxt == NULL) {
fprintf(output, "NULL\n");
return;
}
fprintf(output, "'%s' ", ctxt->string);
if (ctxt->error)
fprintf(output, "error ");
if (ctxt->neg)
fprintf(output, "neg ");
fprintf(output, "\n");
fprintf(output, "%d atoms:\n", ctxt->nbAtoms);
for (i = 0;i < ctxt->nbAtoms; i++) {
fprintf(output, " %02d ", i);
xmlRegPrintAtom(output, ctxt->atoms[i]);
}
if (ctxt->atom != NULL) {
fprintf(output, "current atom:\n");
xmlRegPrintAtom(output, ctxt->atom);
}
fprintf(output, "%d states:", ctxt->nbStates);
if (ctxt->start != NULL)
fprintf(output, " start: %d", ctxt->start->no);
if (ctxt->end != NULL)
fprintf(output, " end: %d", ctxt->end->no);
fprintf(output, "\n");
for (i = 0;i < ctxt->nbStates; i++) {
xmlRegPrintState(output, ctxt->states[i]);
}
fprintf(output, "%d counters:\n", ctxt->nbCounters);
for (i = 0;i < ctxt->nbCounters; i++) {
fprintf(output, " %d: min %d max %d\n", i, ctxt->counters[i].min,
ctxt->counters[i].max);
}
}
#endif
/************************************************************************
* *
* Finite Automata structures manipulations *
* *
************************************************************************/
static void
xmlRegAtomAddRange(xmlRegParserCtxtPtr ctxt, xmlRegAtomPtr atom,
int neg, xmlRegAtomType type, int start, int end,
xmlChar *blockName) {
xmlRegRangePtr range;
if (atom == NULL) {
ERROR("add range: atom is NULL");
return;
}
if (atom->type != XML_REGEXP_RANGES) {
ERROR("add range: atom is not ranges");
return;
}
if (atom->maxRanges == 0) {
atom->maxRanges = 4;
atom->ranges = (xmlRegRangePtr *) xmlMalloc(atom->maxRanges *
sizeof(xmlRegRangePtr));
if (atom->ranges == NULL) {
xmlRegexpErrMemory(ctxt, "adding ranges");
atom->maxRanges = 0;
return;
}
} else if (atom->nbRanges >= atom->maxRanges) {
xmlRegRangePtr *tmp;
atom->maxRanges *= 2;
tmp = (xmlRegRangePtr *) xmlRealloc(atom->ranges, atom->maxRanges *
sizeof(xmlRegRangePtr));
if (tmp == NULL) {
xmlRegexpErrMemory(ctxt, "adding ranges");
atom->maxRanges /= 2;
return;
}
atom->ranges = tmp;
}
range = xmlRegNewRange(ctxt, neg, type, start, end);
if (range == NULL)
return;
range->blockName = blockName;
atom->ranges[atom->nbRanges++] = range;
}
static int
xmlRegGetCounter(xmlRegParserCtxtPtr ctxt) {
if (ctxt->maxCounters == 0) {
ctxt->maxCounters = 4;
ctxt->counters = (xmlRegCounter *) xmlMalloc(ctxt->maxCounters *
sizeof(xmlRegCounter));
if (ctxt->counters == NULL) {
xmlRegexpErrMemory(ctxt, "allocating counter");
ctxt->maxCounters = 0;
return(-1);
}
} else if (ctxt->nbCounters >= ctxt->maxCounters) {
xmlRegCounter *tmp;
ctxt->maxCounters *= 2;
tmp = (xmlRegCounter *) xmlRealloc(ctxt->counters, ctxt->maxCounters *
sizeof(xmlRegCounter));
if (tmp == NULL) {
xmlRegexpErrMemory(ctxt, "allocating counter");
ctxt->maxCounters /= 2;
return(-1);
}
ctxt->counters = tmp;
}
ctxt->counters[ctxt->nbCounters].min = -1;
ctxt->counters[ctxt->nbCounters].max = -1;
return(ctxt->nbCounters++);
}
static int
xmlRegAtomPush(xmlRegParserCtxtPtr ctxt, xmlRegAtomPtr atom) {
if (atom == NULL) {
ERROR("atom push: atom is NULL");
return(-1);
}
if (ctxt->maxAtoms == 0) {
ctxt->maxAtoms = 4;
ctxt->atoms = (xmlRegAtomPtr *) xmlMalloc(ctxt->maxAtoms *
sizeof(xmlRegAtomPtr));
if (ctxt->atoms == NULL) {
xmlRegexpErrMemory(ctxt, "pushing atom");
ctxt->maxAtoms = 0;
return(-1);
}
} else if (ctxt->nbAtoms >= ctxt->maxAtoms) {
xmlRegAtomPtr *tmp;
ctxt->maxAtoms *= 2;
tmp = (xmlRegAtomPtr *) xmlRealloc(ctxt->atoms, ctxt->maxAtoms *
sizeof(xmlRegAtomPtr));
if (tmp == NULL) {
xmlRegexpErrMemory(ctxt, "allocating counter");
ctxt->maxAtoms /= 2;
return(-1);
}
ctxt->atoms = tmp;
}
atom->no = ctxt->nbAtoms;
ctxt->atoms[ctxt->nbAtoms++] = atom;
return(0);
}
static void
xmlRegStateAddTransTo(xmlRegParserCtxtPtr ctxt, xmlRegStatePtr target,
int from) {
if (target->maxTransTo == 0) {
target->maxTransTo = 8;
target->transTo = (int *) xmlMalloc(target->maxTransTo *
sizeof(int));
if (target->transTo == NULL) {
xmlRegexpErrMemory(ctxt, "adding transition");
target->maxTransTo = 0;
return;
}
} else if (target->nbTransTo >= target->maxTransTo) {
int *tmp;
target->maxTransTo *= 2;
tmp = (int *) xmlRealloc(target->transTo, target->maxTransTo *
sizeof(int));
if (tmp == NULL) {
xmlRegexpErrMemory(ctxt, "adding transition");
target->maxTransTo /= 2;
return;
}
target->transTo = tmp;
}
target->transTo[target->nbTransTo] = from;
target->nbTransTo++;
}
static void
xmlRegStateAddTrans(xmlRegParserCtxtPtr ctxt, xmlRegStatePtr state,
xmlRegAtomPtr atom, xmlRegStatePtr target,
int counter, int count) {
int nrtrans;
if (state == NULL) {
ERROR("add state: state is NULL");
return;
}
if (target == NULL) {
ERROR("add state: target is NULL");
return;
}
/*
* Other routines follow the philosophy 'When in doubt, add a transition'
* so we check here whether such a transition is already present and, if
* so, silently ignore this request.
*/
for (nrtrans = state->nbTrans - 1; nrtrans >= 0; nrtrans--) {
xmlRegTransPtr trans = &(state->trans[nrtrans]);
if ((trans->atom == atom) &&
(trans->to == target->no) &&
(trans->counter == counter) &&
(trans->count == count)) {
#ifdef DEBUG_REGEXP_GRAPH
printf("Ignoring duplicate transition from %d to %d\n",
state->no, target->no);
#endif
return;
}
}
if (state->maxTrans == 0) {
state->maxTrans = 8;
state->trans = (xmlRegTrans *) xmlMalloc(state->maxTrans *
sizeof(xmlRegTrans));
if (state->trans == NULL) {
xmlRegexpErrMemory(ctxt, "adding transition");
state->maxTrans = 0;
return;
}
} else if (state->nbTrans >= state->maxTrans) {
xmlRegTrans *tmp;
state->maxTrans *= 2;
tmp = (xmlRegTrans *) xmlRealloc(state->trans, state->maxTrans *
sizeof(xmlRegTrans));
if (tmp == NULL) {
xmlRegexpErrMemory(ctxt, "adding transition");
state->maxTrans /= 2;
return;
}
state->trans = tmp;
}
#ifdef DEBUG_REGEXP_GRAPH
printf("Add trans from %d to %d ", state->no, target->no);
if (count == REGEXP_ALL_COUNTER)
printf("all transition\n");
else if (count >= 0)
printf("count based %d\n", count);
else if (counter >= 0)
printf("counted %d\n", counter);
else if (atom == NULL)
printf("epsilon transition\n");
else if (atom != NULL)
xmlRegPrintAtom(stdout, atom);
#endif
state->trans[state->nbTrans].atom = atom;
state->trans[state->nbTrans].to = target->no;
state->trans[state->nbTrans].counter = counter;
state->trans[state->nbTrans].count = count;
state->trans[state->nbTrans].nd = 0;
state->nbTrans++;
xmlRegStateAddTransTo(ctxt, target, state->no);
}
static int
xmlRegStatePush(xmlRegParserCtxtPtr ctxt, xmlRegStatePtr state) {
if (state == NULL) return(-1);
if (ctxt->maxStates == 0) {
ctxt->maxStates = 4;
ctxt->states = (xmlRegStatePtr *) xmlMalloc(ctxt->maxStates *
sizeof(xmlRegStatePtr));
if (ctxt->states == NULL) {
xmlRegexpErrMemory(ctxt, "adding state");
ctxt->maxStates = 0;
return(-1);
}
} else if (ctxt->nbStates >= ctxt->maxStates) {
xmlRegStatePtr *tmp;
ctxt->maxStates *= 2;
tmp = (xmlRegStatePtr *) xmlRealloc(ctxt->states, ctxt->maxStates *
sizeof(xmlRegStatePtr));
if (tmp == NULL) {
xmlRegexpErrMemory(ctxt, "adding state");
ctxt->maxStates /= 2;
return(-1);
}
ctxt->states = tmp;
}
state->no = ctxt->nbStates;
ctxt->states[ctxt->nbStates++] = state;
return(0);
}
/**
* xmlFAGenerateAllTransition:
* @ctxt: a regexp parser context
* @from: the from state
* @to: the target state or NULL for building a new one
* @lax:
*
*/
static void
xmlFAGenerateAllTransition(xmlRegParserCtxtPtr ctxt,
xmlRegStatePtr from, xmlRegStatePtr to,
int lax) {
if (to == NULL) {
to = xmlRegNewState(ctxt);
xmlRegStatePush(ctxt, to);
ctxt->state = to;
}
if (lax)
xmlRegStateAddTrans(ctxt, from, NULL, to, -1, REGEXP_ALL_LAX_COUNTER);
else
xmlRegStateAddTrans(ctxt, from, NULL, to, -1, REGEXP_ALL_COUNTER);
}
/**
* xmlFAGenerateEpsilonTransition:
* @ctxt: a regexp parser context
* @from: the from state
* @to: the target state or NULL for building a new one
*
*/
static void
xmlFAGenerateEpsilonTransition(xmlRegParserCtxtPtr ctxt,
xmlRegStatePtr from, xmlRegStatePtr to) {
if (to == NULL) {
to = xmlRegNewState(ctxt);
xmlRegStatePush(ctxt, to);
ctxt->state = to;
}
xmlRegStateAddTrans(ctxt, from, NULL, to, -1, -1);
}
/**
* xmlFAGenerateCountedEpsilonTransition:
* @ctxt: a regexp parser context
* @from: the from state
* @to: the target state or NULL for building a new one
* counter: the counter for that transition
*
*/
static void
xmlFAGenerateCountedEpsilonTransition(xmlRegParserCtxtPtr ctxt,
xmlRegStatePtr from, xmlRegStatePtr to, int counter) {
if (to == NULL) {
to = xmlRegNewState(ctxt);
xmlRegStatePush(ctxt, to);
ctxt->state = to;
}
xmlRegStateAddTrans(ctxt, from, NULL, to, counter, -1);
}
/**
* xmlFAGenerateCountedTransition:
* @ctxt: a regexp parser context
* @from: the from state
* @to: the target state or NULL for building a new one
* counter: the counter for that transition
*
*/
static void
xmlFAGenerateCountedTransition(xmlRegParserCtxtPtr ctxt,
xmlRegStatePtr from, xmlRegStatePtr to, int counter) {
if (to == NULL) {
to = xmlRegNewState(ctxt);
xmlRegStatePush(ctxt, to);
ctxt->state = to;
}
xmlRegStateAddTrans(ctxt, from, NULL, to, -1, counter);
}
/**
* xmlFAGenerateTransitions:
* @ctxt: a regexp parser context
* @from: the from state
* @to: the target state or NULL for building a new one
* @atom: the atom generating the transition
*
* Returns 0 if success and -1 in case of error.
*/
static int
xmlFAGenerateTransitions(xmlRegParserCtxtPtr ctxt, xmlRegStatePtr from,
xmlRegStatePtr to, xmlRegAtomPtr atom) {
xmlRegStatePtr end;
if (atom == NULL) {
ERROR("genrate transition: atom == NULL");
return(-1);
}
if (atom->type == XML_REGEXP_SUBREG) {
/*
* this is a subexpression handling one should not need to
* create a new node except for XML_REGEXP_QUANT_RANGE.
*/
if (xmlRegAtomPush(ctxt, atom) < 0) {
return(-1);
}
if ((to != NULL) && (atom->stop != to) &&
(atom->quant != XML_REGEXP_QUANT_RANGE)) {
/*
* Generate an epsilon transition to link to the target
*/
xmlFAGenerateEpsilonTransition(ctxt, atom->stop, to);
#ifdef DV
} else if ((to == NULL) && (atom->quant != XML_REGEXP_QUANT_RANGE) &&
(atom->quant != XML_REGEXP_QUANT_ONCE)) {
to = xmlRegNewState(ctxt);
xmlRegStatePush(ctxt, to);
ctxt->state = to;
xmlFAGenerateEpsilonTransition(ctxt, atom->stop, to);
#endif
}
switch (atom->quant) {
case XML_REGEXP_QUANT_OPT:
atom->quant = XML_REGEXP_QUANT_ONCE;
/*
* transition done to the state after end of atom.
* 1. set transition from atom start to new state
* 2. set transition from atom end to this state.
*/
if (to == NULL) {
xmlFAGenerateEpsilonTransition(ctxt, atom->start, 0);
xmlFAGenerateEpsilonTransition(ctxt, atom->stop,
ctxt->state);
} else {
xmlFAGenerateEpsilonTransition(ctxt, atom->start, to);
}
break;
case XML_REGEXP_QUANT_MULT:
atom->quant = XML_REGEXP_QUANT_ONCE;
xmlFAGenerateEpsilonTransition(ctxt, atom->start, atom->stop);
xmlFAGenerateEpsilonTransition(ctxt, atom->stop, atom->start);
break;
case XML_REGEXP_QUANT_PLUS:
atom->quant = XML_REGEXP_QUANT_ONCE;
xmlFAGenerateEpsilonTransition(ctxt, atom->stop, atom->start);
break;
case XML_REGEXP_QUANT_RANGE: {
int counter;
xmlRegStatePtr inter, newstate;
/*
* create the final state now if needed
*/
if (to != NULL) {
newstate = to;
} else {
newstate = xmlRegNewState(ctxt);
xmlRegStatePush(ctxt, newstate);
}
/*
* The principle here is to use counted transition
* to avoid explosion in the number of states in the
* graph. This is clearly more complex but should not
* be exploitable at runtime.
*/
if ((atom->min == 0) && (atom->start0 == NULL)) {
xmlRegAtomPtr copy;
/*
* duplicate a transition based on atom to count next
* occurences after 1. We cannot loop to atom->start
* directly because we need an epsilon transition to
* newstate.
*/
/* ???? For some reason it seems we never reach that
case, I suppose this got optimized out before when
building the automata */
copy = xmlRegCopyAtom(ctxt, atom);
if (copy == NULL)
return(-1);
copy->quant = XML_REGEXP_QUANT_ONCE;
copy->min = 0;
copy->max = 0;
if (xmlFAGenerateTransitions(ctxt, atom->start, NULL, copy)
< 0)
return(-1);
inter = ctxt->state;
counter = xmlRegGetCounter(ctxt);
ctxt->counters[counter].min = atom->min - 1;
ctxt->counters[counter].max = atom->max - 1;
/* count the number of times we see it again */
xmlFAGenerateCountedEpsilonTransition(ctxt, inter,
atom->stop, counter);
/* allow a way out based on the count */
xmlFAGenerateCountedTransition(ctxt, inter,
newstate, counter);
/* and also allow a direct exit for 0 */
xmlFAGenerateEpsilonTransition(ctxt, atom->start,
newstate);
} else {
/*
* either we need the atom at least once or there
* is an atom->start0 allowing to easilly plug the
* epsilon transition.
*/
counter = xmlRegGetCounter(ctxt);
ctxt->counters[counter].min = atom->min - 1;
ctxt->counters[counter].max = atom->max - 1;
/* count the number of times we see it again */
xmlFAGenerateCountedEpsilonTransition(ctxt, atom->stop,
atom->start, counter);
/* allow a way out based on the count */
xmlFAGenerateCountedTransition(ctxt, atom->stop,
newstate, counter);
/* and if needed allow a direct exit for 0 */
if (atom->min == 0)
xmlFAGenerateEpsilonTransition(ctxt, atom->start0,
newstate);
}
atom->min = 0;
atom->max = 0;
atom->quant = XML_REGEXP_QUANT_ONCE;
ctxt->state = newstate;
}
default:
break;
}
return(0);
}
if ((atom->min == 0) && (atom->max == 0) &&
(atom->quant == XML_REGEXP_QUANT_RANGE)) {
/*
* we can discard the atom and generate an epsilon transition instead
*/
if (to == NULL) {
to = xmlRegNewState(ctxt);
if (to != NULL)
xmlRegStatePush(ctxt, to);
else {
return(-1);
}
}
xmlFAGenerateEpsilonTransition(ctxt, from, to);
ctxt->state = to;
xmlRegFreeAtom(atom);
return(0);
}
if (to == NULL) {
to = xmlRegNewState(ctxt);
if (to != NULL)
xmlRegStatePush(ctxt, to);
else {
return(-1);
}
}
end = to;
if ((atom->quant == XML_REGEXP_QUANT_MULT) ||
(atom->quant == XML_REGEXP_QUANT_PLUS)) {
/*
* Do not pollute the target state by adding transitions from
* it as it is likely to be the shared target of multiple branches.
* So isolate with an epsilon transition.
*/
xmlRegStatePtr tmp;
tmp = xmlRegNewState(ctxt);
if (tmp != NULL)
xmlRegStatePush(ctxt, tmp);
else {
return(-1);
}
xmlFAGenerateEpsilonTransition(ctxt, tmp, to);
to = tmp;
}
if (xmlRegAtomPush(ctxt, atom) < 0) {
return(-1);
}
xmlRegStateAddTrans(ctxt, from, atom, to, -1, -1);
ctxt->state = end;
switch (atom->quant) {
case XML_REGEXP_QUANT_OPT:
atom->quant = XML_REGEXP_QUANT_ONCE;
xmlFAGenerateEpsilonTransition(ctxt, from, to);
break;
case XML_REGEXP_QUANT_MULT:
atom->quant = XML_REGEXP_QUANT_ONCE;
xmlFAGenerateEpsilonTransition(ctxt, from, to);
xmlRegStateAddTrans(ctxt, to, atom, to, -1, -1);
break;
case XML_REGEXP_QUANT_PLUS:
atom->quant = XML_REGEXP_QUANT_ONCE;
xmlRegStateAddTrans(ctxt, to, atom, to, -1, -1);
break;
case XML_REGEXP_QUANT_RANGE:
#if DV_test
if (atom->min == 0) {
xmlFAGenerateEpsilonTransition(ctxt, from, to);
}
#endif
break;
default:
break;
}
return(0);
}
/**
* xmlFAReduceEpsilonTransitions:
* @ctxt: a regexp parser context
* @fromnr: the from state
* @tonr: the to state
* @counter: should that transition be associated to a counted
*
*/
static void
xmlFAReduceEpsilonTransitions(xmlRegParserCtxtPtr ctxt, int fromnr,
int tonr, int counter) {
int transnr;
xmlRegStatePtr from;
xmlRegStatePtr to;
#ifdef DEBUG_REGEXP_GRAPH
printf("xmlFAReduceEpsilonTransitions(%d, %d)\n", fromnr, tonr);
#endif
from = ctxt->states[fromnr];
if (from == NULL)
return;
to = ctxt->states[tonr];
if (to == NULL)
return;
if ((to->mark == XML_REGEXP_MARK_START) ||
(to->mark == XML_REGEXP_MARK_VISITED))
return;
to->mark = XML_REGEXP_MARK_VISITED;
if (to->type == XML_REGEXP_FINAL_STATE) {
#ifdef DEBUG_REGEXP_GRAPH
printf("State %d is final, so %d becomes final\n", tonr, fromnr);
#endif
from->type = XML_REGEXP_FINAL_STATE;
}
for (transnr = 0;transnr < to->nbTrans;transnr++) {
if (to->trans[transnr].to < 0)
continue;
if (to->trans[transnr].atom == NULL) {
/*
* Don't remove counted transitions
* Don't loop either
*/
if (to->trans[transnr].to != fromnr) {
if (to->trans[transnr].count >= 0) {
int newto = to->trans[transnr].to;
xmlRegStateAddTrans(ctxt, from, NULL,
ctxt->states[newto],
-1, to->trans[transnr].count);
} else {
#ifdef DEBUG_REGEXP_GRAPH
printf("Found epsilon trans %d from %d to %d\n",
transnr, tonr, to->trans[transnr].to);
#endif
if (to->trans[transnr].counter >= 0) {
xmlFAReduceEpsilonTransitions(ctxt, fromnr,
to->trans[transnr].to,
to->trans[transnr].counter);
} else {
xmlFAReduceEpsilonTransitions(ctxt, fromnr,
to->trans[transnr].to,
counter);
}
}
}
} else {
int newto = to->trans[transnr].to;
if (to->trans[transnr].counter >= 0) {
xmlRegStateAddTrans(ctxt, from, to->trans[transnr].atom,
ctxt->states[newto],
to->trans[transnr].counter, -1);
} else {
xmlRegStateAddTrans(ctxt, from, to->trans[transnr].atom,
ctxt->states[newto], counter, -1);
}
}
}
to->mark = XML_REGEXP_MARK_NORMAL;
}
/**
* xmlFAEliminateSimpleEpsilonTransitions:
* @ctxt: a regexp parser context
*
* Eliminating general epsilon transitions can get costly in the general
* algorithm due to the large amount of generated new transitions and
* associated comparisons. However for simple epsilon transition used just
* to separate building blocks when generating the automata this can be
* reduced to state elimination:
* - if there exists an epsilon from X to Y
* - if there is no other transition from X
* then X and Y are semantically equivalent and X can be eliminated
* If X is the start state then make Y the start state, else replace the
* target of all transitions to X by transitions to Y.
*/
static void
xmlFAEliminateSimpleEpsilonTransitions(xmlRegParserCtxtPtr ctxt) {
int statenr, i, j, newto;
xmlRegStatePtr state, tmp;
for (statenr = 0;statenr < ctxt->nbStates;statenr++) {
state = ctxt->states[statenr];
if (state == NULL)
continue;
if (state->nbTrans != 1)
continue;
if (state->type == XML_REGEXP_UNREACH_STATE)
continue;
/* is the only transition out a basic transition */
if ((state->trans[0].atom == NULL) &&
(state->trans[0].to >= 0) &&
(state->trans[0].to != statenr) &&
(state->trans[0].counter < 0) &&
(state->trans[0].count < 0)) {
newto = state->trans[0].to;
if (state->type == XML_REGEXP_START_STATE) {
#ifdef DEBUG_REGEXP_GRAPH
printf("Found simple epsilon trans from start %d to %d\n",
statenr, newto);
#endif
} else {
#ifdef DEBUG_REGEXP_GRAPH
printf("Found simple epsilon trans from %d to %d\n",
statenr, newto);
#endif
for (i = 0;i < state->nbTransTo;i++) {
tmp = ctxt->states[state->transTo[i]];
for (j = 0;j < tmp->nbTrans;j++) {
if (tmp->trans[j].to == statenr) {
#ifdef DEBUG_REGEXP_GRAPH
printf("Changed transition %d on %d to go to %d\n",
j, tmp->no, newto);
#endif
tmp->trans[j].to = -1;
xmlRegStateAddTrans(ctxt, tmp, tmp->trans[j].atom,
ctxt->states[newto],
tmp->trans[j].counter,
tmp->trans[j].count);
}
}
}
if (state->type == XML_REGEXP_FINAL_STATE)
ctxt->states[newto]->type = XML_REGEXP_FINAL_STATE;
/* eliminate the transition completely */
state->nbTrans = 0;
state->type = XML_REGEXP_UNREACH_STATE;
}
}
}
}
/**
* xmlFAEliminateEpsilonTransitions:
* @ctxt: a regexp parser context
*
*/
static void
xmlFAEliminateEpsilonTransitions(xmlRegParserCtxtPtr ctxt) {
int statenr, transnr;
xmlRegStatePtr state;
int has_epsilon;
if (ctxt->states == NULL) return;
/*
* Eliminate simple epsilon transition and the associated unreachable
* states.
*/
xmlFAEliminateSimpleEpsilonTransitions(ctxt);
for (statenr = 0;statenr < ctxt->nbStates;statenr++) {
state = ctxt->states[statenr];
if ((state != NULL) && (state->type == XML_REGEXP_UNREACH_STATE)) {
#ifdef DEBUG_REGEXP_GRAPH
printf("Removed unreachable state %d\n", statenr);
#endif
xmlRegFreeState(state);
ctxt->states[statenr] = NULL;
}
}
has_epsilon = 0;
/*
* Build the completed transitions bypassing the epsilons
* Use a marking algorithm to avoid loops
* Mark sink states too.
* Process from the latests states backward to the start when
* there is long cascading epsilon chains this minimize the
* recursions and transition compares when adding the new ones
*/
for (statenr = ctxt->nbStates - 1;statenr >= 0;statenr--) {
state = ctxt->states[statenr];
if (state == NULL)
continue;
if ((state->nbTrans == 0) &&
(state->type != XML_REGEXP_FINAL_STATE)) {
state->type = XML_REGEXP_SINK_STATE;
}
for (transnr = 0;transnr < state->nbTrans;transnr++) {
if ((state->trans[transnr].atom == NULL) &&
(state->trans[transnr].to >= 0)) {
if (state->trans[transnr].to == statenr) {
state->trans[transnr].to = -1;
#ifdef DEBUG_REGEXP_GRAPH
printf("Removed loopback epsilon trans %d on %d\n",
transnr, statenr);
#endif
} else if (state->trans[transnr].count < 0) {
int newto = state->trans[transnr].to;
#ifdef DEBUG_REGEXP_GRAPH
printf("Found epsilon trans %d from %d to %d\n",
transnr, statenr, newto);
#endif
has_epsilon = 1;
state->trans[transnr].to = -2;
state->mark = XML_REGEXP_MARK_START;
xmlFAReduceEpsilonTransitions(ctxt, statenr,
newto, state->trans[transnr].counter);
state->mark = XML_REGEXP_MARK_NORMAL;
#ifdef DEBUG_REGEXP_GRAPH
} else {
printf("Found counted transition %d on %d\n",
transnr, statenr);
#endif
}
}
}
}
/*
* Eliminate the epsilon transitions
*/
if (has_epsilon) {
for (statenr = 0;statenr < ctxt->nbStates;statenr++) {
state = ctxt->states[statenr];
if (state == NULL)
continue;
for (transnr = 0;transnr < state->nbTrans;transnr++) {
xmlRegTransPtr trans = &(state->trans[transnr]);
if ((trans->atom == NULL) &&
(trans->count < 0) &&
(trans->to >= 0)) {
trans->to = -1;
}
}
}
}
/*
* Use this pass to detect unreachable states too
*/
for (statenr = 0;statenr < ctxt->nbStates;statenr++) {
state = ctxt->states[statenr];
if (state != NULL)
state->reached = XML_REGEXP_MARK_NORMAL;
}
state = ctxt->states[0];
if (state != NULL)
state->reached = XML_REGEXP_MARK_START;
while (state != NULL) {
xmlRegStatePtr target = NULL;
state->reached = XML_REGEXP_MARK_VISITED;
/*
* Mark all states reachable from the current reachable state
*/
for (transnr = 0;transnr < state->nbTrans;transnr++) {
if ((state->trans[transnr].to >= 0) &&
((state->trans[transnr].atom != NULL) ||
(state->trans[transnr].count >= 0))) {
int newto = state->trans[transnr].to;
if (ctxt->states[newto] == NULL)
continue;
if (ctxt->states[newto]->reached == XML_REGEXP_MARK_NORMAL) {
ctxt->states[newto]->reached = XML_REGEXP_MARK_START;
target = ctxt->states[newto];
}
}
}
/*
* find the next accessible state not explored
*/
if (target == NULL) {
for (statenr = 1;statenr < ctxt->nbStates;statenr++) {
state = ctxt->states[statenr];
if ((state != NULL) && (state->reached ==
XML_REGEXP_MARK_START)) {
target = state;
break;
}
}
}
state = target;
}
for (statenr = 0;statenr < ctxt->nbStates;statenr++) {
state = ctxt->states[statenr];
if ((state != NULL) && (state->reached == XML_REGEXP_MARK_NORMAL)) {
#ifdef DEBUG_REGEXP_GRAPH
printf("Removed unreachable state %d\n", statenr);
#endif
xmlRegFreeState(state);
ctxt->states[statenr] = NULL;
}
}
}
static int
xmlFACompareRanges(xmlRegRangePtr range1, xmlRegRangePtr range2) {
int ret = 0;
if ((range1->type == XML_REGEXP_RANGES) ||
(range2->type == XML_REGEXP_RANGES) ||
(range2->type == XML_REGEXP_SUBREG) ||
(range1->type == XML_REGEXP_SUBREG) ||
(range1->type == XML_REGEXP_STRING) ||
(range2->type == XML_REGEXP_STRING))
return(-1);
/* put them in order */
if (range1->type > range2->type) {
xmlRegRangePtr tmp;
tmp = range1;
range1 = range2;
range2 = tmp;
}
if ((range1->type == XML_REGEXP_ANYCHAR) ||
(range2->type == XML_REGEXP_ANYCHAR)) {
ret = 1;
} else if ((range1->type == XML_REGEXP_EPSILON) ||
(range2->type == XML_REGEXP_EPSILON)) {
return(0);
} else if (range1->type == range2->type) {
if (range1->type != XML_REGEXP_CHARVAL)
ret = 1;
else if ((range1->end < range2->start) ||
(range2->end < range1->start))
ret = 0;
else
ret = 1;
} else if (range1->type == XML_REGEXP_CHARVAL) {
int codepoint;
int neg = 0;
/*
* just check all codepoints in the range for acceptance,
* this is usually way cheaper since done only once at
* compilation than testing over and over at runtime or
* pushing too many states when evaluating.
*/
if (((range1->neg == 0) && (range2->neg != 0)) ||
((range1->neg != 0) && (range2->neg == 0)))
neg = 1;
for (codepoint = range1->start;codepoint <= range1->end ;codepoint++) {
ret = xmlRegCheckCharacterRange(range2->type, codepoint,
0, range2->start, range2->end,
range2->blockName);
if (ret < 0)
return(-1);
if (((neg == 1) && (ret == 0)) ||
((neg == 0) && (ret == 1)))
return(1);
}
return(0);
} else if ((range1->type == XML_REGEXP_BLOCK_NAME) ||
(range2->type == XML_REGEXP_BLOCK_NAME)) {
if (range1->type == range2->type) {
ret = xmlStrEqual(range1->blockName, range2->blockName);
} else {
/*
* comparing a block range with anything else is way
* too costly, and maintining the table is like too much
* memory too, so let's force the automata to save state
* here.
*/
return(1);
}
} else if ((range1->type < XML_REGEXP_LETTER) ||
(range2->type < XML_REGEXP_LETTER)) {
if ((range1->type == XML_REGEXP_ANYSPACE) &&
(range2->type == XML_REGEXP_NOTSPACE))
ret = 0;
else if ((range1->type == XML_REGEXP_INITNAME) &&
(range2->type == XML_REGEXP_NOTINITNAME))
ret = 0;
else if ((range1->type == XML_REGEXP_NAMECHAR) &&
(range2->type == XML_REGEXP_NOTNAMECHAR))
ret = 0;
else if ((range1->type == XML_REGEXP_DECIMAL) &&
(range2->type == XML_REGEXP_NOTDECIMAL))
ret = 0;
else if ((range1->type == XML_REGEXP_REALCHAR) &&
(range2->type == XML_REGEXP_NOTREALCHAR))
ret = 0;
else {
/* same thing to limit complexity */
return(1);
}
} else {
ret = 0;
/* range1->type < range2->type here */
switch (range1->type) {
case XML_REGEXP_LETTER:
/* all disjoint except in the subgroups */
if ((range2->type == XML_REGEXP_LETTER_UPPERCASE) ||
(range2->type == XML_REGEXP_LETTER_LOWERCASE) ||
(range2->type == XML_REGEXP_LETTER_TITLECASE) ||
(range2->type == XML_REGEXP_LETTER_MODIFIER) ||
(range2->type == XML_REGEXP_LETTER_OTHERS))
ret = 1;
break;
case XML_REGEXP_MARK:
if ((range2->type == XML_REGEXP_MARK_NONSPACING) ||
(range2->type == XML_REGEXP_MARK_SPACECOMBINING) ||
(range2->type == XML_REGEXP_MARK_ENCLOSING))
ret = 1;
break;
case XML_REGEXP_NUMBER:
if ((range2->type == XML_REGEXP_NUMBER_DECIMAL) ||
(range2->type == XML_REGEXP_NUMBER_LETTER) ||
(range2->type == XML_REGEXP_NUMBER_OTHERS))
ret = 1;
break;
case XML_REGEXP_PUNCT:
if ((range2->type == XML_REGEXP_PUNCT_CONNECTOR) ||
(range2->type == XML_REGEXP_PUNCT_DASH) ||
(range2->type == XML_REGEXP_PUNCT_OPEN) ||
(range2->type == XML_REGEXP_PUNCT_CLOSE) ||
(range2->type == XML_REGEXP_PUNCT_INITQUOTE) ||
(range2->type == XML_REGEXP_PUNCT_FINQUOTE) ||
(range2->type == XML_REGEXP_PUNCT_OTHERS))
ret = 1;
break;
case XML_REGEXP_SEPAR:
if ((range2->type == XML_REGEXP_SEPAR_SPACE) ||
(range2->type == XML_REGEXP_SEPAR_LINE) ||
(range2->type == XML_REGEXP_SEPAR_PARA))
ret = 1;
break;
case XML_REGEXP_SYMBOL:
if ((range2->type == XML_REGEXP_SYMBOL_MATH) ||
(range2->type == XML_REGEXP_SYMBOL_CURRENCY) ||
(range2->type == XML_REGEXP_SYMBOL_MODIFIER) ||
(range2->type == XML_REGEXP_SYMBOL_OTHERS))
ret = 1;
break;
case XML_REGEXP_OTHER:
if ((range2->type == XML_REGEXP_OTHER_CONTROL) ||
(range2->type == XML_REGEXP_OTHER_FORMAT) ||
(range2->type == XML_REGEXP_OTHER_PRIVATE))
ret = 1;
break;
default:
if ((range2->type >= XML_REGEXP_LETTER) &&
(range2->type < XML_REGEXP_BLOCK_NAME))
ret = 0;
else {
/* safety net ! */
return(1);
}
}
}
if (((range1->neg == 0) && (range2->neg != 0)) ||
((range1->neg != 0) && (range2->neg == 0)))
ret = !ret;
return(ret);
}
/**
* xmlFACompareAtomTypes:
* @type1: an atom type
* @type2: an atom type
*
* Compares two atoms type to check whether they intersect in some ways,
* this is used by xmlFACompareAtoms only
*
* Returns 1 if they may intersect and 0 otherwise
*/
static int
xmlFACompareAtomTypes(xmlRegAtomType type1, xmlRegAtomType type2) {
if ((type1 == XML_REGEXP_EPSILON) ||
(type1 == XML_REGEXP_CHARVAL) ||
(type1 == XML_REGEXP_RANGES) ||
(type1 == XML_REGEXP_SUBREG) ||
(type1 == XML_REGEXP_STRING) ||
(type1 == XML_REGEXP_ANYCHAR))
return(1);
if ((type2 == XML_REGEXP_EPSILON) ||
(type2 == XML_REGEXP_CHARVAL) ||
(type2 == XML_REGEXP_RANGES) ||
(type2 == XML_REGEXP_SUBREG) ||
(type2 == XML_REGEXP_STRING) ||
(type2 == XML_REGEXP_ANYCHAR))
return(1);
if (type1 == type2) return(1);
/* simplify subsequent compares by making sure type1 < type2 */
if (type1 > type2) {
xmlRegAtomType tmp = type1;
type1 = type2;
type2 = tmp;
}
switch (type1) {
case XML_REGEXP_ANYSPACE: /* \s */
/* can't be a letter, number, mark, pontuation, symbol */
if ((type2 == XML_REGEXP_NOTSPACE) ||
((type2 >= XML_REGEXP_LETTER) &&
(type2 <= XML_REGEXP_LETTER_OTHERS)) ||
((type2 >= XML_REGEXP_NUMBER) &&
(type2 <= XML_REGEXP_NUMBER_OTHERS)) ||
((type2 >= XML_REGEXP_MARK) &&
(type2 <= XML_REGEXP_MARK_ENCLOSING)) ||
((type2 >= XML_REGEXP_PUNCT) &&
(type2 <= XML_REGEXP_PUNCT_OTHERS)) ||
((type2 >= XML_REGEXP_SYMBOL) &&
(type2 <= XML_REGEXP_SYMBOL_OTHERS))
) return(0);
break;
case XML_REGEXP_NOTSPACE: /* \S */
break;
case XML_REGEXP_INITNAME: /* \l */
/* can't be a number, mark, separator, pontuation, symbol or other */
if ((type2 == XML_REGEXP_NOTINITNAME) ||
((type2 >= XML_REGEXP_NUMBER) &&
(type2 <= XML_REGEXP_NUMBER_OTHERS)) ||
((type2 >= XML_REGEXP_MARK) &&
(type2 <= XML_REGEXP_MARK_ENCLOSING)) ||
((type2 >= XML_REGEXP_SEPAR) &&
(type2 <= XML_REGEXP_SEPAR_PARA)) ||
((type2 >= XML_REGEXP_PUNCT) &&
(type2 <= XML_REGEXP_PUNCT_OTHERS)) ||
((type2 >= XML_REGEXP_SYMBOL) &&
(type2 <= XML_REGEXP_SYMBOL_OTHERS)) ||
((type2 >= XML_REGEXP_OTHER) &&
(type2 <= XML_REGEXP_OTHER_NA))
) return(0);
break;
case XML_REGEXP_NOTINITNAME: /* \L */
break;
case XML_REGEXP_NAMECHAR: /* \c */
/* can't be a mark, separator, pontuation, symbol or other */
if ((type2 == XML_REGEXP_NOTNAMECHAR) ||
((type2 >= XML_REGEXP_MARK) &&
(type2 <= XML_REGEXP_MARK_ENCLOSING)) ||
((type2 >= XML_REGEXP_PUNCT) &&
(type2 <= XML_REGEXP_PUNCT_OTHERS)) ||
((type2 >= XML_REGEXP_SEPAR) &&
(type2 <= XML_REGEXP_SEPAR_PARA)) ||
((type2 >= XML_REGEXP_SYMBOL) &&
(type2 <= XML_REGEXP_SYMBOL_OTHERS)) ||
((type2 >= XML_REGEXP_OTHER) &&
(type2 <= XML_REGEXP_OTHER_NA))
) return(0);
break;
case XML_REGEXP_NOTNAMECHAR: /* \C */
break;
case XML_REGEXP_DECIMAL: /* \d */
/* can't be a letter, mark, separator, pontuation, symbol or other */
if ((type2 == XML_REGEXP_NOTDECIMAL) ||
(type2 == XML_REGEXP_REALCHAR) ||
((type2 >= XML_REGEXP_LETTER) &&
(type2 <= XML_REGEXP_LETTER_OTHERS)) ||
((type2 >= XML_REGEXP_MARK) &&
(type2 <= XML_REGEXP_MARK_ENCLOSING)) ||
((type2 >= XML_REGEXP_PUNCT) &&
(type2 <= XML_REGEXP_PUNCT_OTHERS)) ||
((type2 >= XML_REGEXP_SEPAR) &&
(type2 <= XML_REGEXP_SEPAR_PARA)) ||
((type2 >= XML_REGEXP_SYMBOL) &&
(type2 <= XML_REGEXP_SYMBOL_OTHERS)) ||
((type2 >= XML_REGEXP_OTHER) &&
(type2 <= XML_REGEXP_OTHER_NA))
)return(0);
break;
case XML_REGEXP_NOTDECIMAL: /* \D */
break;
case XML_REGEXP_REALCHAR: /* \w */
/* can't be a mark, separator, pontuation, symbol or other */
if ((type2 == XML_REGEXP_NOTDECIMAL) ||
((type2 >= XML_REGEXP_MARK) &&
(type2 <= XML_REGEXP_MARK_ENCLOSING)) ||
((type2 >= XML_REGEXP_PUNCT) &&
(type2 <= XML_REGEXP_PUNCT_OTHERS)) ||
((type2 >= XML_REGEXP_SEPAR) &&
(type2 <= XML_REGEXP_SEPAR_PARA)) ||
((type2 >= XML_REGEXP_SYMBOL) &&
(type2 <= XML_REGEXP_SYMBOL_OTHERS)) ||
((type2 >= XML_REGEXP_OTHER) &&
(type2 <= XML_REGEXP_OTHER_NA))
)return(0);
break;
case XML_REGEXP_NOTREALCHAR: /* \W */
break;
/*
* at that point we know both type 1 and type2 are from
* character categories are ordered and are different,
* it becomes simple because this is a partition
*/
case XML_REGEXP_LETTER:
if (type2 <= XML_REGEXP_LETTER_OTHERS)
return(1);
return(0);
case XML_REGEXP_LETTER_UPPERCASE:
case XML_REGEXP_LETTER_LOWERCASE:
case XML_REGEXP_LETTER_TITLECASE:
case XML_REGEXP_LETTER_MODIFIER:
case XML_REGEXP_LETTER_OTHERS:
return(0);
case XML_REGEXP_MARK:
if (type2 <= XML_REGEXP_MARK_ENCLOSING)
return(1);
return(0);
case XML_REGEXP_MARK_NONSPACING:
case XML_REGEXP_MARK_SPACECOMBINING:
case XML_REGEXP_MARK_ENCLOSING:
return(0);
case XML_REGEXP_NUMBER:
if (type2 <= XML_REGEXP_NUMBER_OTHERS)
return(1);
return(0);
case XML_REGEXP_NUMBER_DECIMAL:
case XML_REGEXP_NUMBER_LETTER:
case XML_REGEXP_NUMBER_OTHERS:
return(0);
case XML_REGEXP_PUNCT:
if (type2 <= XML_REGEXP_PUNCT_OTHERS)
return(1);
return(0);
case XML_REGEXP_PUNCT_CONNECTOR:
case XML_REGEXP_PUNCT_DASH:
case XML_REGEXP_PUNCT_OPEN:
case XML_REGEXP_PUNCT_CLOSE:
case XML_REGEXP_PUNCT_INITQUOTE:
case XML_REGEXP_PUNCT_FINQUOTE:
case XML_REGEXP_PUNCT_OTHERS:
return(0);
case XML_REGEXP_SEPAR:
if (type2 <= XML_REGEXP_SEPAR_PARA)
return(1);
return(0);
case XML_REGEXP_SEPAR_SPACE:
case XML_REGEXP_SEPAR_LINE:
case XML_REGEXP_SEPAR_PARA:
return(0);
case XML_REGEXP_SYMBOL:
if (type2 <= XML_REGEXP_SYMBOL_OTHERS)
return(1);
return(0);
case XML_REGEXP_SYMBOL_MATH:
case XML_REGEXP_SYMBOL_CURRENCY:
case XML_REGEXP_SYMBOL_MODIFIER:
case XML_REGEXP_SYMBOL_OTHERS:
return(0);
case XML_REGEXP_OTHER:
if (type2 <= XML_REGEXP_OTHER_NA)
return(1);
return(0);
case XML_REGEXP_OTHER_CONTROL:
case XML_REGEXP_OTHER_FORMAT:
case XML_REGEXP_OTHER_PRIVATE:
case XML_REGEXP_OTHER_NA:
return(0);
default:
break;
}
return(1);
}
/**
* xmlFAEqualAtoms:
* @atom1: an atom
* @atom2: an atom
* @deep: if not set only compare string pointers
*
* Compares two atoms to check whether they are the same exactly
* this is used to remove equivalent transitions
*
* Returns 1 if same and 0 otherwise
*/
static int
xmlFAEqualAtoms(xmlRegAtomPtr atom1, xmlRegAtomPtr atom2, int deep) {
int ret = 0;
if (atom1 == atom2)
return(1);
if ((atom1 == NULL) || (atom2 == NULL))
return(0);
if (atom1->type != atom2->type)
return(0);
switch (atom1->type) {
case XML_REGEXP_EPSILON:
ret = 0;
break;
case XML_REGEXP_STRING:
if (!deep)
ret = (atom1->valuep == atom2->valuep);
else
ret = xmlStrEqual((xmlChar *)atom1->valuep,
(xmlChar *)atom2->valuep);
break;
case XML_REGEXP_CHARVAL:
ret = (atom1->codepoint == atom2->codepoint);
break;
case XML_REGEXP_RANGES:
/* too hard to do in the general case */
ret = 0;
default:
break;
}
return(ret);
}
/**
* xmlFACompareAtoms:
* @atom1: an atom
* @atom2: an atom
* @deep: if not set only compare string pointers
*
* Compares two atoms to check whether they intersect in some ways,
* this is used by xmlFAComputesDeterminism and xmlFARecurseDeterminism only
*
* Returns 1 if yes and 0 otherwise
*/
static int
xmlFACompareAtoms(xmlRegAtomPtr atom1, xmlRegAtomPtr atom2, int deep) {
int ret = 1;
if (atom1 == atom2)
return(1);
if ((atom1 == NULL) || (atom2 == NULL))
return(0);
if ((atom1->type == XML_REGEXP_ANYCHAR) ||
(atom2->type == XML_REGEXP_ANYCHAR))
return(1);
if (atom1->type > atom2->type) {
xmlRegAtomPtr tmp;
tmp = atom1;
atom1 = atom2;
atom2 = tmp;
}
if (atom1->type != atom2->type) {
ret = xmlFACompareAtomTypes(atom1->type, atom2->type);
/* if they can't intersect at the type level break now */
if (ret == 0)
return(0);
}
switch (atom1->type) {
case XML_REGEXP_STRING:
if (!deep)
ret = (atom1->valuep != atom2->valuep);
else
ret = xmlRegStrEqualWildcard((xmlChar *)atom1->valuep,
(xmlChar *)atom2->valuep);
break;
case XML_REGEXP_EPSILON:
goto not_determinist;
case XML_REGEXP_CHARVAL:
if (atom2->type == XML_REGEXP_CHARVAL) {
ret = (atom1->codepoint == atom2->codepoint);
} else {
ret = xmlRegCheckCharacter(atom2, atom1->codepoint);
if (ret < 0)
ret = 1;
}
break;
case XML_REGEXP_RANGES:
if (atom2->type == XML_REGEXP_RANGES) {
int i, j, res;
xmlRegRangePtr r1, r2;
/*
* need to check that none of the ranges eventually matches
*/
for (i = 0;i < atom1->nbRanges;i++) {
for (j = 0;j < atom2->nbRanges;j++) {
r1 = atom1->ranges[i];
r2 = atom2->ranges[j];
res = xmlFACompareRanges(r1, r2);
if (res == 1) {
ret = 1;
goto done;
}
}
}
ret = 0;
}
break;
default:
goto not_determinist;
}
done:
if (atom1->neg != atom2->neg) {
ret = !ret;
}
if (ret == 0)
return(0);
not_determinist:
return(1);
}
/**
* xmlFARecurseDeterminism:
* @ctxt: a regexp parser context
*
* Check whether the associated regexp is determinist,
* should be called after xmlFAEliminateEpsilonTransitions()
*
*/
static int
xmlFARecurseDeterminism(xmlRegParserCtxtPtr ctxt, xmlRegStatePtr state,
int to, xmlRegAtomPtr atom) {
int ret = 1;
int res;
int transnr, nbTrans;
xmlRegTransPtr t1;
int deep = 1;
if (state == NULL)
return(ret);
if (ctxt->flags & AM_AUTOMATA_RNG)
deep = 0;
/*
* don't recurse on transitions potentially added in the course of
* the elimination.
*/
nbTrans = state->nbTrans;
for (transnr = 0;transnr < nbTrans;transnr++) {
t1 = &(state->trans[transnr]);
/*
* check transitions conflicting with the one looked at
*/
if (t1->atom == NULL) {
if (t1->to < 0)
continue;
res = xmlFARecurseDeterminism(ctxt, ctxt->states[t1->to],
to, atom);
if (res == 0) {
ret = 0;
/* t1->nd = 1; */
}
continue;
}
if (t1->to != to)
continue;
if (xmlFACompareAtoms(t1->atom, atom, deep)) {
ret = 0;
/* mark the transition as non-deterministic */
t1->nd = 1;
}
}
return(ret);
}
/**
* xmlFAComputesDeterminism:
* @ctxt: a regexp parser context
*
* Check whether the associated regexp is determinist,
* should be called after xmlFAEliminateEpsilonTransitions()
*
*/
static int
xmlFAComputesDeterminism(xmlRegParserCtxtPtr ctxt) {
int statenr, transnr;
xmlRegStatePtr state;
xmlRegTransPtr t1, t2, last;
int i;
int ret = 1;
int deep = 1;
#ifdef DEBUG_REGEXP_GRAPH
printf("xmlFAComputesDeterminism\n");
xmlRegPrintCtxt(stdout, ctxt);
#endif
if (ctxt->determinist != -1)
return(ctxt->determinist);
if (ctxt->flags & AM_AUTOMATA_RNG)
deep = 0;
/*
* First cleanup the automata removing cancelled transitions
*/
for (statenr = 0;statenr < ctxt->nbStates;statenr++) {
state = ctxt->states[statenr];
if (state == NULL)
continue;
if (state->nbTrans < 2)
continue;
for (transnr = 0;transnr < state->nbTrans;transnr++) {
t1 = &(state->trans[transnr]);
/*
* Determinism checks in case of counted or all transitions
* will have to be handled separately
*/
if (t1->atom == NULL) {
/* t1->nd = 1; */
continue;
}
if (t1->to == -1) /* eliminated */
continue;
for (i = 0;i < transnr;i++) {
t2 = &(state->trans[i]);
if (t2->to == -1) /* eliminated */
continue;
if (t2->atom != NULL) {
if (t1->to == t2->to) {
/*
* Here we use deep because we want to keep the
* transitions which indicate a conflict
*/
if (xmlFAEqualAtoms(t1->atom, t2->atom, deep) &&
(t1->counter == t2->counter) &&
(t1->count == t2->count))
t2->to = -1; /* eliminated */
}
}
}
}
}
/*
* Check for all states that there aren't 2 transitions
* with the same atom and a different target.
*/
for (statenr = 0;statenr < ctxt->nbStates;statenr++) {
state = ctxt->states[statenr];
if (state == NULL)
continue;
if (state->nbTrans < 2)
continue;
last = NULL;
for (transnr = 0;transnr < state->nbTrans;transnr++) {
t1 = &(state->trans[transnr]);
/*
* Determinism checks in case of counted or all transitions
* will have to be handled separately
*/
if (t1->atom == NULL) {
continue;
}
if (t1->to == -1) /* eliminated */
continue;
for (i = 0;i < transnr;i++) {
t2 = &(state->trans[i]);
if (t2->to == -1) /* eliminated */
continue;
if (t2->atom != NULL) {
/*
* But here we don't use deep because we want to
* find transitions which indicate a conflict
*/
if (xmlFACompareAtoms(t1->atom, t2->atom, 1)) {
ret = 0;
/* mark the transitions as non-deterministic ones */
t1->nd = 1;
t2->nd = 1;
last = t1;
}
} else if (t1->to != -1) {
/*
* do the closure in case of remaining specific
* epsilon transitions like choices or all
*/
ret = xmlFARecurseDeterminism(ctxt, ctxt->states[t1->to],
t2->to, t2->atom);
/* don't shortcut the computation so all non deterministic
transition get marked down
if (ret == 0)
return(0);
*/
if (ret == 0) {
t1->nd = 1;
/* t2->nd = 1; */
last = t1;
}
}
}
/* don't shortcut the computation so all non deterministic
transition get marked down
if (ret == 0)
break; */
}
/*
* mark specifically the last non-deterministic transition
* from a state since there is no need to set-up rollback
* from it
*/
if (last != NULL) {
last->nd = 2;
}
/* don't shortcut the computation so all non deterministic
transition get marked down
if (ret == 0)
break; */
}
ctxt->determinist = ret;
return(ret);
}
/************************************************************************
* *
* Routines to check input against transition atoms *
* *
************************************************************************/
static int
xmlRegCheckCharacterRange(xmlRegAtomType type, int codepoint, int neg,
int start, int end, const xmlChar *blockName) {
int ret = 0;
switch (type) {
case XML_REGEXP_STRING:
case XML_REGEXP_SUBREG:
case XML_REGEXP_RANGES:
case XML_REGEXP_EPSILON:
return(-1);
case XML_REGEXP_ANYCHAR:
ret = ((codepoint != '\n') && (codepoint != '\r'));
break;
case XML_REGEXP_CHARVAL:
ret = ((codepoint >= start) && (codepoint <= end));
break;
case XML_REGEXP_NOTSPACE:
neg = !neg;
case XML_REGEXP_ANYSPACE:
ret = ((codepoint == '\n') || (codepoint == '\r') ||
(codepoint == '\t') || (codepoint == ' '));
break;
case XML_REGEXP_NOTINITNAME:
neg = !neg;
case XML_REGEXP_INITNAME:
ret = (IS_LETTER(codepoint) ||
(codepoint == '_') || (codepoint == ':'));
break;
case XML_REGEXP_NOTNAMECHAR:
neg = !neg;
case XML_REGEXP_NAMECHAR:
ret = (IS_LETTER(codepoint) || IS_DIGIT(codepoint) ||
(codepoint == '.') || (codepoint == '-') ||
(codepoint == '_') || (codepoint == ':') ||
IS_COMBINING(codepoint) || IS_EXTENDER(codepoint));
break;
case XML_REGEXP_NOTDECIMAL:
neg = !neg;
case XML_REGEXP_DECIMAL:
ret = xmlUCSIsCatNd(codepoint);
break;
case XML_REGEXP_REALCHAR:
neg = !neg;
case XML_REGEXP_NOTREALCHAR:
ret = xmlUCSIsCatP(codepoint);
if (ret == 0)
ret = xmlUCSIsCatZ(codepoint);
if (ret == 0)
ret = xmlUCSIsCatC(codepoint);
break;
case XML_REGEXP_LETTER:
ret = xmlUCSIsCatL(codepoint);
break;
case XML_REGEXP_LETTER_UPPERCASE:
ret = xmlUCSIsCatLu(codepoint);
break;
case XML_REGEXP_LETTER_LOWERCASE:
ret = xmlUCSIsCatLl(codepoint);
break;
case XML_REGEXP_LETTER_TITLECASE:
ret = xmlUCSIsCatLt(codepoint);
break;
case XML_REGEXP_LETTER_MODIFIER:
ret = xmlUCSIsCatLm(codepoint);
break;
case XML_REGEXP_LETTER_OTHERS:
ret = xmlUCSIsCatLo(codepoint);
break;
case XML_REGEXP_MARK:
ret = xmlUCSIsCatM(codepoint);
break;
case XML_REGEXP_MARK_NONSPACING:
ret = xmlUCSIsCatMn(codepoint);
break;
case XML_REGEXP_MARK_SPACECOMBINING:
ret = xmlUCSIsCatMc(codepoint);
break;
case XML_REGEXP_MARK_ENCLOSING:
ret = xmlUCSIsCatMe(codepoint);
break;
case XML_REGEXP_NUMBER:
ret = xmlUCSIsCatN(codepoint);
break;
case XML_REGEXP_NUMBER_DECIMAL:
ret = xmlUCSIsCatNd(codepoint);
break;
case XML_REGEXP_NUMBER_LETTER:
ret = xmlUCSIsCatNl(codepoint);
break;
case XML_REGEXP_NUMBER_OTHERS:
ret = xmlUCSIsCatNo(codepoint);
break;
case XML_REGEXP_PUNCT:
ret = xmlUCSIsCatP(codepoint);
break;
case XML_REGEXP_PUNCT_CONNECTOR:
ret = xmlUCSIsCatPc(codepoint);
break;
case XML_REGEXP_PUNCT_DASH:
ret = xmlUCSIsCatPd(codepoint);
break;
case XML_REGEXP_PUNCT_OPEN:
ret = xmlUCSIsCatPs(codepoint);
break;
case XML_REGEXP_PUNCT_CLOSE:
ret = xmlUCSIsCatPe(codepoint);
break;
case XML_REGEXP_PUNCT_INITQUOTE:
ret = xmlUCSIsCatPi(codepoint);
break;
case XML_REGEXP_PUNCT_FINQUOTE:
ret = xmlUCSIsCatPf(codepoint);
break;
case XML_REGEXP_PUNCT_OTHERS:
ret = xmlUCSIsCatPo(codepoint);
break;
case XML_REGEXP_SEPAR:
ret = xmlUCSIsCatZ(codepoint);
break;
case XML_REGEXP_SEPAR_SPACE:
ret = xmlUCSIsCatZs(codepoint);
break;
case XML_REGEXP_SEPAR_LINE:
ret = xmlUCSIsCatZl(codepoint);
break;
case XML_REGEXP_SEPAR_PARA:
ret = xmlUCSIsCatZp(codepoint);
break;
case XML_REGEXP_SYMBOL:
ret = xmlUCSIsCatS(codepoint);
break;
case XML_REGEXP_SYMBOL_MATH:
ret = xmlUCSIsCatSm(codepoint);
break;
case XML_REGEXP_SYMBOL_CURRENCY:
ret = xmlUCSIsCatSc(codepoint);
break;
case XML_REGEXP_SYMBOL_MODIFIER:
ret = xmlUCSIsCatSk(codepoint);
break;
case XML_REGEXP_SYMBOL_OTHERS:
ret = xmlUCSIsCatSo(codepoint);
break;
case XML_REGEXP_OTHER:
ret = xmlUCSIsCatC(codepoint);
break;
case XML_REGEXP_OTHER_CONTROL:
ret = xmlUCSIsCatCc(codepoint);
break;
case XML_REGEXP_OTHER_FORMAT:
ret = xmlUCSIsCatCf(codepoint);
break;
case XML_REGEXP_OTHER_PRIVATE:
ret = xmlUCSIsCatCo(codepoint);
break;
case XML_REGEXP_OTHER_NA:
/* ret = xmlUCSIsCatCn(codepoint); */
/* Seems it doesn't exist anymore in recent Unicode releases */
ret = 0;
break;
case XML_REGEXP_BLOCK_NAME:
ret = xmlUCSIsBlock(codepoint, (const char *) blockName);
break;
}
if (neg)
return(!ret);
return(ret);
}
static int
xmlRegCheckCharacter(xmlRegAtomPtr atom, int codepoint) {
int i, ret = 0;
xmlRegRangePtr range;
if ((atom == NULL) || (!IS_CHAR(codepoint)))
return(-1);
switch (atom->type) {
case XML_REGEXP_SUBREG:
case XML_REGEXP_EPSILON:
return(-1);
case XML_REGEXP_CHARVAL:
return(codepoint == atom->codepoint);
case XML_REGEXP_RANGES: {
int accept = 0;
for (i = 0;i < atom->nbRanges;i++) {
range = atom->ranges[i];
if (range->neg == 2) {
ret = xmlRegCheckCharacterRange(range->type, codepoint,
0, range->start, range->end,
range->blockName);
if (ret != 0)
return(0); /* excluded char */
} else if (range->neg) {
ret = xmlRegCheckCharacterRange(range->type, codepoint,
0, range->start, range->end,
range->blockName);
if (ret == 0)
accept = 1;
else
return(0);
} else {
ret = xmlRegCheckCharacterRange(range->type, codepoint,
0, range->start, range->end,
range->blockName);
if (ret != 0)
accept = 1; /* might still be excluded */
}
}
return(accept);
}
case XML_REGEXP_STRING:
printf("TODO: XML_REGEXP_STRING\n");
return(-1);
case XML_REGEXP_ANYCHAR:
case XML_REGEXP_ANYSPACE:
case XML_REGEXP_NOTSPACE:
case XML_REGEXP_INITNAME:
case XML_REGEXP_NOTINITNAME:
case XML_REGEXP_NAMECHAR:
case XML_REGEXP_NOTNAMECHAR:
case XML_REGEXP_DECIMAL:
case XML_REGEXP_NOTDECIMAL:
case XML_REGEXP_REALCHAR:
case XML_REGEXP_NOTREALCHAR:
case XML_REGEXP_LETTER:
case XML_REGEXP_LETTER_UPPERCASE:
case XML_REGEXP_LETTER_LOWERCASE:
case XML_REGEXP_LETTER_TITLECASE:
case XML_REGEXP_LETTER_MODIFIER:
case XML_REGEXP_LETTER_OTHERS:
case XML_REGEXP_MARK:
case XML_REGEXP_MARK_NONSPACING:
case XML_REGEXP_MARK_SPACECOMBINING:
case XML_REGEXP_MARK_ENCLOSING:
case XML_REGEXP_NUMBER:
case XML_REGEXP_NUMBER_DECIMAL:
case XML_REGEXP_NUMBER_LETTER:
case XML_REGEXP_NUMBER_OTHERS:
case XML_REGEXP_PUNCT:
case XML_REGEXP_PUNCT_CONNECTOR:
case XML_REGEXP_PUNCT_DASH:
case XML_REGEXP_PUNCT_OPEN:
case XML_REGEXP_PUNCT_CLOSE:
case XML_REGEXP_PUNCT_INITQUOTE:
case XML_REGEXP_PUNCT_FINQUOTE:
case XML_REGEXP_PUNCT_OTHERS:
case XML_REGEXP_SEPAR:
case XML_REGEXP_SEPAR_SPACE:
case XML_REGEXP_SEPAR_LINE:
case XML_REGEXP_SEPAR_PARA:
case XML_REGEXP_SYMBOL:
case XML_REGEXP_SYMBOL_MATH:
case XML_REGEXP_SYMBOL_CURRENCY:
case XML_REGEXP_SYMBOL_MODIFIER:
case XML_REGEXP_SYMBOL_OTHERS:
case XML_REGEXP_OTHER:
case XML_REGEXP_OTHER_CONTROL:
case XML_REGEXP_OTHER_FORMAT:
case XML_REGEXP_OTHER_PRIVATE:
case XML_REGEXP_OTHER_NA:
case XML_REGEXP_BLOCK_NAME:
ret = xmlRegCheckCharacterRange(atom->type, codepoint, 0, 0, 0,
(const xmlChar *)atom->valuep);
if (atom->neg)
ret = !ret;
break;
}
return(ret);
}
/************************************************************************
* *
* Saving and restoring state of an execution context *
* *
************************************************************************/
#ifdef DEBUG_REGEXP_EXEC
static void
xmlFARegDebugExec(xmlRegExecCtxtPtr exec) {
printf("state: %d:%d:idx %d", exec->state->no, exec->transno, exec->index);
if (exec->inputStack != NULL) {
int i;
printf(": ");
for (i = 0;(i < 3) && (i < exec->inputStackNr);i++)
printf("%s ", (const char *)
exec->inputStack[exec->inputStackNr - (i + 1)].value);
} else {
printf(": %s", &(exec->inputString[exec->index]));
}
printf("\n");
}
#endif
static void
xmlFARegExecSave(xmlRegExecCtxtPtr exec) {
#ifdef DEBUG_REGEXP_EXEC
printf("saving ");
exec->transno++;
xmlFARegDebugExec(exec);
exec->transno--;
#endif
#ifdef MAX_PUSH
if (exec->nbPush > MAX_PUSH) {
return;
}
exec->nbPush++;
#endif
if (exec->maxRollbacks == 0) {
exec->maxRollbacks = 4;
exec->rollbacks = (xmlRegExecRollback *) xmlMalloc(exec->maxRollbacks *
sizeof(xmlRegExecRollback));
if (exec->rollbacks == NULL) {
xmlRegexpErrMemory(NULL, "saving regexp");
exec->maxRollbacks = 0;
return;
}
memset(exec->rollbacks, 0,
exec->maxRollbacks * sizeof(xmlRegExecRollback));
} else if (exec->nbRollbacks >= exec->maxRollbacks) {
xmlRegExecRollback *tmp;
int len = exec->maxRollbacks;
exec->maxRollbacks *= 2;
tmp = (xmlRegExecRollback *) xmlRealloc(exec->rollbacks,
exec->maxRollbacks * sizeof(xmlRegExecRollback));
if (tmp == NULL) {
xmlRegexpErrMemory(NULL, "saving regexp");
exec->maxRollbacks /= 2;
return;
}
exec->rollbacks = tmp;
tmp = &exec->rollbacks[len];
memset(tmp, 0, (exec->maxRollbacks - len) * sizeof(xmlRegExecRollback));
}
exec->rollbacks[exec->nbRollbacks].state = exec->state;
exec->rollbacks[exec->nbRollbacks].index = exec->index;
exec->rollbacks[exec->nbRollbacks].nextbranch = exec->transno + 1;
if (exec->comp->nbCounters > 0) {
if (exec->rollbacks[exec->nbRollbacks].counts == NULL) {
exec->rollbacks[exec->nbRollbacks].counts = (int *)
xmlMalloc(exec->comp->nbCounters * sizeof(int));
if (exec->rollbacks[exec->nbRollbacks].counts == NULL) {
xmlRegexpErrMemory(NULL, "saving regexp");
exec->status = -5;
return;
}
}
memcpy(exec->rollbacks[exec->nbRollbacks].counts, exec->counts,
exec->comp->nbCounters * sizeof(int));
}
exec->nbRollbacks++;
}
static void
xmlFARegExecRollBack(xmlRegExecCtxtPtr exec) {
if (exec->nbRollbacks <= 0) {
exec->status = -1;
#ifdef DEBUG_REGEXP_EXEC
printf("rollback failed on empty stack\n");
#endif
return;
}
exec->nbRollbacks--;
exec->state = exec->rollbacks[exec->nbRollbacks].state;
exec->index = exec->rollbacks[exec->nbRollbacks].index;
exec->transno = exec->rollbacks[exec->nbRollbacks].nextbranch;
if (exec->comp->nbCounters > 0) {
if (exec->rollbacks[exec->nbRollbacks].counts == NULL) {
fprintf(stderr, "exec save: allocation failed");
exec->status = -6;
return;
}
memcpy(exec->counts, exec->rollbacks[exec->nbRollbacks].counts,
exec->comp->nbCounters * sizeof(int));
}
#ifdef DEBUG_REGEXP_EXEC
printf("restored ");
xmlFARegDebugExec(exec);
#endif
}
/************************************************************************
* *
* Verifier, running an input against a compiled regexp *
* *
************************************************************************/
static int
xmlFARegExec(xmlRegexpPtr comp, const xmlChar *content) {
xmlRegExecCtxt execval;
xmlRegExecCtxtPtr exec = &execval;
int ret, codepoint = 0, len, deter;
exec->inputString = content;
exec->index = 0;
exec->nbPush = 0;
exec->determinist = 1;
exec->maxRollbacks = 0;
exec->nbRollbacks = 0;
exec->rollbacks = NULL;
exec->status = 0;
exec->comp = comp;
exec->state = comp->states[0];
exec->transno = 0;
exec->transcount = 0;
exec->inputStack = NULL;
exec->inputStackMax = 0;
if (comp->nbCounters > 0) {
exec->counts = (int *) xmlMalloc(comp->nbCounters * sizeof(int));
if (exec->counts == NULL) {
xmlRegexpErrMemory(NULL, "running regexp");
return(-1);
}
memset(exec->counts, 0, comp->nbCounters * sizeof(int));
} else
exec->counts = NULL;
while ((exec->status == 0) &&
((exec->inputString[exec->index] != 0) ||
((exec->state != NULL) &&
(exec->state->type != XML_REGEXP_FINAL_STATE)))) {
xmlRegTransPtr trans;
xmlRegAtomPtr atom;
/*
* If end of input on non-terminal state, rollback, however we may
* still have epsilon like transition for counted transitions
* on counters, in that case don't break too early. Additionally,
* if we are working on a range like "AB{0,2}", where B is not present,
* we don't want to break.
*/
len = 1;
if ((exec->inputString[exec->index] == 0) && (exec->counts == NULL)) {
/*
* if there is a transition, we must check if
* atom allows minOccurs of 0
*/
if (exec->transno < exec->state->nbTrans) {
trans = &exec->state->trans[exec->transno];
if (trans->to >=0) {
atom = trans->atom;
if (!((atom->min == 0) && (atom->max > 0)))
goto rollback;
}
} else
goto rollback;
}
exec->transcount = 0;
for (;exec->transno < exec->state->nbTrans;exec->transno++) {
trans = &exec->state->trans[exec->transno];
if (trans->to < 0)
continue;
atom = trans->atom;
ret = 0;
deter = 1;
if (trans->count >= 0) {
int count;
xmlRegCounterPtr counter;
if (exec->counts == NULL) {
exec->status = -1;
goto error;
}
/*
* A counted transition.
*/
count = exec->counts[trans->count];
counter = &exec->comp->counters[trans->count];
#ifdef DEBUG_REGEXP_EXEC
printf("testing count %d: val %d, min %d, max %d\n",
trans->count, count, counter->min, counter->max);
#endif
ret = ((count >= counter->min) && (count <= counter->max));
if ((ret) && (counter->min != counter->max))
deter = 0;
} else if (atom == NULL) {
fprintf(stderr, "epsilon transition left at runtime\n");
exec->status = -2;
break;
} else if (exec->inputString[exec->index] != 0) {
codepoint = CUR_SCHAR(&(exec->inputString[exec->index]), len);
ret = xmlRegCheckCharacter(atom, codepoint);
if ((ret == 1) && (atom->min >= 0) && (atom->max > 0)) {
xmlRegStatePtr to = comp->states[trans->to];
/*
* this is a multiple input sequence
* If there is a counter associated increment it now.
* before potentially saving and rollback
* do not increment if the counter is already over the
* maximum limit in which case get to next transition
*/
if (trans->counter >= 0) {
xmlRegCounterPtr counter;
if ((exec->counts == NULL) ||
(exec->comp == NULL) ||
(exec->comp->counters == NULL)) {
exec->status = -1;
goto error;
}
counter = &exec->comp->counters[trans->counter];
if (exec->counts[trans->counter] >= counter->max)
continue; /* for loop on transitions */
#ifdef DEBUG_REGEXP_EXEC
printf("Increasing count %d\n", trans->counter);
#endif
exec->counts[trans->counter]++;
}
if (exec->state->nbTrans > exec->transno + 1) {
xmlFARegExecSave(exec);
}
exec->transcount = 1;
do {
/*
* Try to progress as much as possible on the input
*/
if (exec->transcount == atom->max) {
break;
}
exec->index += len;
/*
* End of input: stop here
*/
if (exec->inputString[exec->index] == 0) {
exec->index -= len;
break;
}
if (exec->transcount >= atom->min) {
int transno = exec->transno;
xmlRegStatePtr state = exec->state;
/*
* The transition is acceptable save it
*/
exec->transno = -1; /* trick */
exec->state = to;
xmlFARegExecSave(exec);
exec->transno = transno;
exec->state = state;
}
codepoint = CUR_SCHAR(&(exec->inputString[exec->index]),
len);
ret = xmlRegCheckCharacter(atom, codepoint);
exec->transcount++;
} while (ret == 1);
if (exec->transcount < atom->min)
ret = 0;
/*
* If the last check failed but one transition was found
* possible, rollback
*/
if (ret < 0)
ret = 0;
if (ret == 0) {
goto rollback;
}
if (trans->counter >= 0) {
if (exec->counts == NULL) {
exec->status = -1;
goto error;
}
#ifdef DEBUG_REGEXP_EXEC
printf("Decreasing count %d\n", trans->counter);
#endif
exec->counts[trans->counter]--;
}
} else if ((ret == 0) && (atom->min == 0) && (atom->max > 0)) {
/*
* we don't match on the codepoint, but minOccurs of 0
* says that's ok. Setting len to 0 inhibits stepping
* over the codepoint.
*/
exec->transcount = 1;
len = 0;
ret = 1;
}
} else if ((atom->min == 0) && (atom->max > 0)) {
/* another spot to match when minOccurs is 0 */
exec->transcount = 1;
len = 0;
ret = 1;
}
if (ret == 1) {
if ((trans->nd == 1) ||
((trans->count >= 0) && (deter == 0) &&
(exec->state->nbTrans > exec->transno + 1))) {
#ifdef DEBUG_REGEXP_EXEC
if (trans->nd == 1)
printf("Saving on nd transition atom %d for %c at %d\n",
trans->atom->no, codepoint, exec->index);
else
printf("Saving on counted transition count %d for %c at %d\n",
trans->count, codepoint, exec->index);
#endif
xmlFARegExecSave(exec);
}
if (trans->counter >= 0) {
xmlRegCounterPtr counter;
/* make sure we don't go over the counter maximum value */
if ((exec->counts == NULL) ||
(exec->comp == NULL) ||
(exec->comp->counters == NULL)) {
exec->status = -1;
goto error;
}
counter = &exec->comp->counters[trans->counter];
if (exec->counts[trans->counter] >= counter->max)
continue; /* for loop on transitions */
#ifdef DEBUG_REGEXP_EXEC
printf("Increasing count %d\n", trans->counter);
#endif
exec->counts[trans->counter]++;
}
if ((trans->count >= 0) &&
(trans->count < REGEXP_ALL_COUNTER)) {
if (exec->counts == NULL) {
exec->status = -1;
goto error;
}
#ifdef DEBUG_REGEXP_EXEC
printf("resetting count %d on transition\n",
trans->count);
#endif
exec->counts[trans->count] = 0;
}
#ifdef DEBUG_REGEXP_EXEC
printf("entering state %d\n", trans->to);
#endif
exec->state = comp->states[trans->to];
exec->transno = 0;
if (trans->atom != NULL) {
exec->index += len;
}
goto progress;
} else if (ret < 0) {
exec->status = -4;
break;
}
}
if ((exec->transno != 0) || (exec->state->nbTrans == 0)) {
rollback:
/*
* Failed to find a way out
*/
exec->determinist = 0;
#ifdef DEBUG_REGEXP_EXEC
printf("rollback from state %d on %d:%c\n", exec->state->no,
codepoint,codepoint);
#endif
xmlFARegExecRollBack(exec);
}
progress:
continue;
}
error:
if (exec->rollbacks != NULL) {
if (exec->counts != NULL) {
int i;
for (i = 0;i < exec->maxRollbacks;i++)
if (exec->rollbacks[i].counts != NULL)
xmlFree(exec->rollbacks[i].counts);
}
xmlFree(exec->rollbacks);
}
if (exec->counts != NULL)
xmlFree(exec->counts);
if (exec->status == 0)
return(1);
if (exec->status == -1) {
if (exec->nbPush > MAX_PUSH)
return(-1);
return(0);
}
return(exec->status);
}
/************************************************************************
* *
* Progressive interface to the verifier one atom at a time *
* *
************************************************************************/
#ifdef DEBUG_ERR
static void testerr(xmlRegExecCtxtPtr exec);
#endif
/**
* xmlRegNewExecCtxt:
* @comp: a precompiled regular expression
* @callback: a callback function used for handling progresses in the
* automata matching phase
* @data: the context data associated to the callback in this context
*
* Build a context used for progressive evaluation of a regexp.
*
* Returns the new context
*/
xmlRegExecCtxtPtr
xmlRegNewExecCtxt(xmlRegexpPtr comp, xmlRegExecCallbacks callback, void *data) {
xmlRegExecCtxtPtr exec;
if (comp == NULL)
return(NULL);
if ((comp->compact == NULL) && (comp->states == NULL))
return(NULL);
exec = (xmlRegExecCtxtPtr) xmlMalloc(sizeof(xmlRegExecCtxt));
if (exec == NULL) {
xmlRegexpErrMemory(NULL, "creating execution context");
return(NULL);
}
memset(exec, 0, sizeof(xmlRegExecCtxt));
exec->inputString = NULL;
exec->index = 0;
exec->determinist = 1;
exec->maxRollbacks = 0;
exec->nbRollbacks = 0;
exec->rollbacks = NULL;
exec->status = 0;
exec->comp = comp;
if (comp->compact == NULL)
exec->state = comp->states[0];
exec->transno = 0;
exec->transcount = 0;
exec->callback = callback;
exec->data = data;
if (comp->nbCounters > 0) {
/*
* For error handling, exec->counts is allocated twice the size
* the second half is used to store the data in case of rollback
*/
exec->counts = (int *) xmlMalloc(comp->nbCounters * sizeof(int)
* 2);
if (exec->counts == NULL) {
xmlRegexpErrMemory(NULL, "creating execution context");
xmlFree(exec);
return(NULL);
}
memset(exec->counts, 0, comp->nbCounters * sizeof(int) * 2);
exec->errCounts = &exec->counts[comp->nbCounters];
} else {
exec->counts = NULL;
exec->errCounts = NULL;
}
exec->inputStackMax = 0;
exec->inputStackNr = 0;
exec->inputStack = NULL;
exec->errStateNo = -1;
exec->errString = NULL;
exec->nbPush = 0;
return(exec);
}
/**
* xmlRegFreeExecCtxt:
* @exec: a regular expression evaulation context
*
* Free the structures associated to a regular expression evaulation context.
*/
void
xmlRegFreeExecCtxt(xmlRegExecCtxtPtr exec) {
if (exec == NULL)
return;
if (exec->rollbacks != NULL) {
if (exec->counts != NULL) {
int i;
for (i = 0;i < exec->maxRollbacks;i++)
if (exec->rollbacks[i].counts != NULL)
xmlFree(exec->rollbacks[i].counts);
}
xmlFree(exec->rollbacks);
}
if (exec->counts != NULL)
xmlFree(exec->counts);
if (exec->inputStack != NULL) {
int i;
for (i = 0;i < exec->inputStackNr;i++) {
if (exec->inputStack[i].value != NULL)
xmlFree(exec->inputStack[i].value);
}
xmlFree(exec->inputStack);
}
if (exec->errString != NULL)
xmlFree(exec->errString);
xmlFree(exec);
}
static void
xmlFARegExecSaveInputString(xmlRegExecCtxtPtr exec, const xmlChar *value,
void *data) {
#ifdef DEBUG_PUSH
printf("saving value: %d:%s\n", exec->inputStackNr, value);
#endif
if (exec->inputStackMax == 0) {
exec->inputStackMax = 4;
exec->inputStack = (xmlRegInputTokenPtr)
xmlMalloc(exec->inputStackMax * sizeof(xmlRegInputToken));
if (exec->inputStack == NULL) {
xmlRegexpErrMemory(NULL, "pushing input string");
exec->inputStackMax = 0;
return;
}
} else if (exec->inputStackNr + 1 >= exec->inputStackMax) {
xmlRegInputTokenPtr tmp;
exec->inputStackMax *= 2;
tmp = (xmlRegInputTokenPtr) xmlRealloc(exec->inputStack,
exec->inputStackMax * sizeof(xmlRegInputToken));
if (tmp == NULL) {
xmlRegexpErrMemory(NULL, "pushing input string");
exec->inputStackMax /= 2;
return;
}
exec->inputStack = tmp;
}
exec->inputStack[exec->inputStackNr].value = xmlStrdup(value);
exec->inputStack[exec->inputStackNr].data = data;
exec->inputStackNr++;
exec->inputStack[exec->inputStackNr].value = NULL;
exec->inputStack[exec->inputStackNr].data = NULL;
}
/**
* xmlRegStrEqualWildcard:
* @expStr: the string to be evaluated
* @valStr: the validation string
*
* Checks if both strings are equal or have the same content. "*"
* can be used as a wildcard in @valStr; "|" is used as a seperator of
* substrings in both @expStr and @valStr.
*
* Returns 1 if the comparison is satisfied and the number of substrings
* is equal, 0 otherwise.
*/
static int
xmlRegStrEqualWildcard(const xmlChar *expStr, const xmlChar *valStr) {
if (expStr == valStr) return(1);
if (expStr == NULL) return(0);
if (valStr == NULL) return(0);
do {
/*
* Eval if we have a wildcard for the current item.
*/
if (*expStr != *valStr) {
/* if one of them starts with a wildcard make valStr be it */
if (*valStr == '*') {
const xmlChar *tmp;
tmp = valStr;
valStr = expStr;
expStr = tmp;
}
if ((*valStr != 0) && (*expStr != 0) && (*expStr++ == '*')) {
do {
if (*valStr == XML_REG_STRING_SEPARATOR)
break;
valStr++;
} while (*valStr != 0);
continue;
} else
return(0);
}
expStr++;
valStr++;
} while (*valStr != 0);
if (*expStr != 0)
return (0);
else
return (1);
}
/**
* xmlRegCompactPushString:
* @exec: a regexp execution context
* @comp: the precompiled exec with a compact table
* @value: a string token input
* @data: data associated to the token to reuse in callbacks
*
* Push one input token in the execution context
*
* Returns: 1 if the regexp reached a final state, 0 if non-final, and
* a negative value in case of error.
*/
static int
xmlRegCompactPushString(xmlRegExecCtxtPtr exec,
xmlRegexpPtr comp,
const xmlChar *value,
void *data) {
int state = exec->index;
int i, target;
if ((comp == NULL) || (comp->compact == NULL) || (comp->stringMap == NULL))
return(-1);
if (value == NULL) {
/*
* are we at a final state ?
*/
if (comp->compact[state * (comp->nbstrings + 1)] ==
XML_REGEXP_FINAL_STATE)
return(1);
return(0);
}
#ifdef DEBUG_PUSH
printf("value pushed: %s\n", value);
#endif
/*
* Examine all outside transitions from current state
*/
for (i = 0;i < comp->nbstrings;i++) {
target = comp->compact[state * (comp->nbstrings + 1) + i + 1];
if ((target > 0) && (target <= comp->nbstates)) {
target--; /* to avoid 0 */
if (xmlRegStrEqualWildcard(comp->stringMap[i], value)) {
exec->index = target;
if ((exec->callback != NULL) && (comp->transdata != NULL)) {
exec->callback(exec->data, value,
comp->transdata[state * comp->nbstrings + i], data);
}
#ifdef DEBUG_PUSH
printf("entering state %d\n", target);
#endif
if (comp->compact[target * (comp->nbstrings + 1)] ==
XML_REGEXP_SINK_STATE)
goto error;
if (comp->compact[target * (comp->nbstrings + 1)] ==
XML_REGEXP_FINAL_STATE)
return(1);
return(0);
}
}
}
/*
* Failed to find an exit transition out from current state for the
* current token
*/
#ifdef DEBUG_PUSH
printf("failed to find a transition for %s on state %d\n", value, state);
#endif
error:
if (exec->errString != NULL)
xmlFree(exec->errString);
exec->errString = xmlStrdup(value);
exec->errStateNo = state;
exec->status = -1;
#ifdef DEBUG_ERR
testerr(exec);
#endif
return(-1);
}
/**
* xmlRegExecPushStringInternal:
* @exec: a regexp execution context or NULL to indicate the end
* @value: a string token input
* @data: data associated to the token to reuse in callbacks
* @compound: value was assembled from 2 strings
*
* Push one input token in the execution context
*
* Returns: 1 if the regexp reached a final state, 0 if non-final, and
* a negative value in case of error.
*/
static int
xmlRegExecPushStringInternal(xmlRegExecCtxtPtr exec, const xmlChar *value,
void *data, int compound) {
xmlRegTransPtr trans;
xmlRegAtomPtr atom;
int ret;
int final = 0;
int progress = 1;
if (exec == NULL)
return(-1);
if (exec->comp == NULL)
return(-1);
if (exec->status != 0)
return(exec->status);
if (exec->comp->compact != NULL)
return(xmlRegCompactPushString(exec, exec->comp, value, data));
if (value == NULL) {
if (exec->state->type == XML_REGEXP_FINAL_STATE)
return(1);
final = 1;
}
#ifdef DEBUG_PUSH
printf("value pushed: %s\n", value);
#endif
/*
* If we have an active rollback stack push the new value there
* and get back to where we were left
*/
if ((value != NULL) && (exec->inputStackNr > 0)) {
xmlFARegExecSaveInputString(exec, value, data);
value = exec->inputStack[exec->index].value;
data = exec->inputStack[exec->index].data;
#ifdef DEBUG_PUSH
printf("value loaded: %s\n", value);
#endif
}
while ((exec->status == 0) &&
((value != NULL) ||
((final == 1) &&
(exec->state->type != XML_REGEXP_FINAL_STATE)))) {
/*
* End of input on non-terminal state, rollback, however we may
* still have epsilon like transition for counted transitions
* on counters, in that case don't break too early.
*/
if ((value == NULL) && (exec->counts == NULL))
goto rollback;
exec->transcount = 0;
for (;exec->transno < exec->state->nbTrans;exec->transno++) {
trans = &exec->state->trans[exec->transno];
if (trans->to < 0)
continue;
atom = trans->atom;
ret = 0;
if (trans->count == REGEXP_ALL_LAX_COUNTER) {
int i;
int count;
xmlRegTransPtr t;
xmlRegCounterPtr counter;
ret = 0;
#ifdef DEBUG_PUSH
printf("testing all lax %d\n", trans->count);
#endif
/*
* Check all counted transitions from the current state
*/
if ((value == NULL) && (final)) {
ret = 1;
} else if (value != NULL) {
for (i = 0;i < exec->state->nbTrans;i++) {
t = &exec->state->trans[i];
if ((t->counter < 0) || (t == trans))
continue;
counter = &exec->comp->counters[t->counter];
count = exec->counts[t->counter];
if ((count < counter->max) &&
(t->atom != NULL) &&
(xmlStrEqual(value, t->atom->valuep))) {
ret = 0;
break;
}
if ((count >= counter->min) &&
(count < counter->max) &&
(t->atom != NULL) &&
(xmlStrEqual(value, t->atom->valuep))) {
ret = 1;
break;
}
}
}
} else if (trans->count == REGEXP_ALL_COUNTER) {
int i;
int count;
xmlRegTransPtr t;
xmlRegCounterPtr counter;
ret = 1;
#ifdef DEBUG_PUSH
printf("testing all %d\n", trans->count);
#endif
/*
* Check all counted transitions from the current state
*/
for (i = 0;i < exec->state->nbTrans;i++) {
t = &exec->state->trans[i];
if ((t->counter < 0) || (t == trans))
continue;
counter = &exec->comp->counters[t->counter];
count = exec->counts[t->counter];
if ((count < counter->min) || (count > counter->max)) {
ret = 0;
break;
}
}
} else if (trans->count >= 0) {
int count;
xmlRegCounterPtr counter;
/*
* A counted transition.
*/
count = exec->counts[trans->count];
counter = &exec->comp->counters[trans->count];
#ifdef DEBUG_PUSH
printf("testing count %d: val %d, min %d, max %d\n",
trans->count, count, counter->min, counter->max);
#endif
ret = ((count >= counter->min) && (count <= counter->max));
} else if (atom == NULL) {
fprintf(stderr, "epsilon transition left at runtime\n");
exec->status = -2;
break;
} else if (value != NULL) {
ret = xmlRegStrEqualWildcard(atom->valuep, value);
if (atom->neg) {
ret = !ret;
if (!compound)
ret = 0;
}
if ((ret == 1) && (trans->counter >= 0)) {
xmlRegCounterPtr counter;
int count;
count = exec->counts[trans->counter];
counter = &exec->comp->counters[trans->counter];
if (count >= counter->max)
ret = 0;
}
if ((ret == 1) && (atom->min > 0) && (atom->max > 0)) {
xmlRegStatePtr to = exec->comp->states[trans->to];
/*
* this is a multiple input sequence
*/
if (exec->state->nbTrans > exec->transno + 1) {
if (exec->inputStackNr <= 0) {
xmlFARegExecSaveInputString(exec, value, data);
}
xmlFARegExecSave(exec);
}
exec->transcount = 1;
do {
/*
* Try to progress as much as possible on the input
*/
if (exec->transcount == atom->max) {
break;
}
exec->index++;
value = exec->inputStack[exec->index].value;
data = exec->inputStack[exec->index].data;
#ifdef DEBUG_PUSH
printf("value loaded: %s\n", value);
#endif
/*
* End of input: stop here
*/
if (value == NULL) {
exec->index --;
break;
}
if (exec->transcount >= atom->min) {
int transno = exec->transno;
xmlRegStatePtr state = exec->state;
/*
* The transition is acceptable save it
*/
exec->transno = -1; /* trick */
exec->state = to;
if (exec->inputStackNr <= 0) {
xmlFARegExecSaveInputString(exec, value, data);
}
xmlFARegExecSave(exec);
exec->transno = transno;
exec->state = state;
}
ret = xmlStrEqual(value, atom->valuep);
exec->transcount++;
} while (ret == 1);
if (exec->transcount < atom->min)
ret = 0;
/*
* If the last check failed but one transition was found
* possible, rollback
*/
if (ret < 0)
ret = 0;
if (ret == 0) {
goto rollback;
}
}
}
if (ret == 1) {
if ((exec->callback != NULL) && (atom != NULL) &&
(data != NULL)) {
exec->callback(exec->data, atom->valuep,
atom->data, data);
}
if (exec->state->nbTrans > exec->transno + 1) {
if (exec->inputStackNr <= 0) {
xmlFARegExecSaveInputString(exec, value, data);
}
xmlFARegExecSave(exec);
}
if (trans->counter >= 0) {
#ifdef DEBUG_PUSH
printf("Increasing count %d\n", trans->counter);
#endif
exec->counts[trans->counter]++;
}
if ((trans->count >= 0) &&
(trans->count < REGEXP_ALL_COUNTER)) {
#ifdef DEBUG_REGEXP_EXEC
printf("resetting count %d on transition\n",
trans->count);
#endif
exec->counts[trans->count] = 0;
}
#ifdef DEBUG_PUSH
printf("entering state %d\n", trans->to);
#endif
if ((exec->comp->states[trans->to] != NULL) &&
(exec->comp->states[trans->to]->type ==
XML_REGEXP_SINK_STATE)) {
/*
* entering a sink state, save the current state as error
* state.
*/
if (exec->errString != NULL)
xmlFree(exec->errString);
exec->errString = xmlStrdup(value);
exec->errState = exec->state;
memcpy(exec->errCounts, exec->counts,
exec->comp->nbCounters * sizeof(int));
}
exec->state = exec->comp->states[trans->to];
exec->transno = 0;
if (trans->atom != NULL) {
if (exec->inputStack != NULL) {
exec->index++;
if (exec->index < exec->inputStackNr) {
value = exec->inputStack[exec->index].value;
data = exec->inputStack[exec->index].data;
#ifdef DEBUG_PUSH
printf("value loaded: %s\n", value);
#endif
} else {
value = NULL;
data = NULL;
#ifdef DEBUG_PUSH
printf("end of input\n");
#endif
}
} else {
value = NULL;
data = NULL;
#ifdef DEBUG_PUSH
printf("end of input\n");
#endif
}
}
goto progress;
} else if (ret < 0) {
exec->status = -4;
break;
}
}
if ((exec->transno != 0) || (exec->state->nbTrans == 0)) {
rollback:
/*
* if we didn't yet rollback on the current input
* store the current state as the error state.
*/
if ((progress) && (exec->state != NULL) &&
(exec->state->type != XML_REGEXP_SINK_STATE)) {
progress = 0;
if (exec->errString != NULL)
xmlFree(exec->errString);
exec->errString = xmlStrdup(value);
exec->errState = exec->state;
memcpy(exec->errCounts, exec->counts,
exec->comp->nbCounters * sizeof(int));
}
/*
* Failed to find a way out
*/
exec->determinist = 0;
xmlFARegExecRollBack(exec);
if (exec->status == 0) {
value = exec->inputStack[exec->index].value;
data = exec->inputStack[exec->index].data;
#ifdef DEBUG_PUSH
printf("value loaded: %s\n", value);
#endif
}
}
continue;
progress:
progress = 1;
continue;
}
if (exec->status == 0) {
return(exec->state->type == XML_REGEXP_FINAL_STATE);
}
#ifdef DEBUG_ERR
if (exec->status < 0) {
testerr(exec);
}
#endif
return(exec->status);
}
/**
* xmlRegExecPushString:
* @exec: a regexp execution context or NULL to indicate the end
* @value: a string token input
* @data: data associated to the token to reuse in callbacks
*
* Push one input token in the execution context
*
* Returns: 1 if the regexp reached a final state, 0 if non-final, and
* a negative value in case of error.
*/
int
xmlRegExecPushString(xmlRegExecCtxtPtr exec, const xmlChar *value,
void *data) {
return(xmlRegExecPushStringInternal(exec, value, data, 0));
}
/**
* xmlRegExecPushString2:
* @exec: a regexp execution context or NULL to indicate the end
* @value: the first string token input
* @value2: the second string token input
* @data: data associated to the token to reuse in callbacks
*
* Push one input token in the execution context
*
* Returns: 1 if the regexp reached a final state, 0 if non-final, and
* a negative value in case of error.
*/
int
xmlRegExecPushString2(xmlRegExecCtxtPtr exec, const xmlChar *value,
const xmlChar *value2, void *data) {
xmlChar buf[150];
int lenn, lenp, ret;
xmlChar *str;
if (exec == NULL)
return(-1);
if (exec->comp == NULL)
return(-1);
if (exec->status != 0)
return(exec->status);
if (value2 == NULL)
return(xmlRegExecPushString(exec, value, data));
lenn = strlen((char *) value2);
lenp = strlen((char *) value);
if (150 < lenn + lenp + 2) {
str = (xmlChar *) xmlMallocAtomic(lenn + lenp + 2);
if (str == NULL) {
exec->status = -1;
return(-1);
}
} else {
str = buf;
}
memcpy(&str[0], value, lenp);
str[lenp] = XML_REG_STRING_SEPARATOR;
memcpy(&str[lenp + 1], value2, lenn);
str[lenn + lenp + 1] = 0;
if (exec->comp->compact != NULL)
ret = xmlRegCompactPushString(exec, exec->comp, str, data);
else
ret = xmlRegExecPushStringInternal(exec, str, data, 1);
if (str != buf)
xmlFree(str);
return(ret);
}
/**
* xmlRegExecGetValues:
* @exec: a regexp execution context
* @err: error extraction or normal one
* @nbval: pointer to the number of accepted values IN/OUT
* @nbneg: return number of negative transitions