chore: import upstream snapshot with attribution

vendored/tre/LICENSE

@@ -0,0 +1,29 @@
+This is the license, copyright notice, and disclaimer for TRE, a regex
+matching package (library and tools) with support for approximate
+matching.
+
+Copyright (c) 2001-2009 Ville Laurikari <vl@iki.fi>
+All rights reserved.
+
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions
+are met:
+
+  1. Redistributions of source code must retain the above copyright
+     notice, this list of conditions and the following disclaimer.
+
+  2. Redistributions in binary form must reproduce the above copyright
+     notice, this list of conditions and the following disclaimer in the
+     documentation and/or other materials provided with the distribution.
+
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDER AND CONTRIBUTORS
+``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+A PARTICULAR PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE COPYRIGHT
+HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

vendored/tre/regcomp.c

@@ -0,0 +1,188 @@
+/*
+  tre_regcomp.c - TRE POSIX compatible regex compilation functions.
+
+  This software is released under a BSD-style license.
+  See the file LICENSE for details and copyright.
+
+*/
+
+#ifdef HAVE_CONFIG_H
+#include <config.h>
+#endif /* HAVE_CONFIG_H */
+
+#include <string.h>
+#include <errno.h>
+#include <stdlib.h>
+
+#include "tre-internal.h"
+#include "xmalloc.h"
+
+int
+tre_regncomp(regex_t *preg, const char *regex, size_t n, int cflags)
+{
+  int ret;
+  if (n > TRE_MAX_RE)
+    return REG_ESPACE;
+#if TRE_WCHAR
+  tre_char_t *wregex;
+  size_t wlen;
+
+  wregex = xmalloc(sizeof(tre_char_t) * (n + 1));
+  if (wregex == NULL)
+    return REG_ESPACE;
+
+  /* If the current locale uses the standard single byte encoding of
+     characters, we don't do a multibyte string conversion.  If we did,
+     many applications which use the default locale would break since
+     the default "C" locale uses the 7-bit ASCII character set, and
+     all characters with the eighth bit set would be considered invalid. */
+#if TRE_MULTIBYTE
+  if (TRE_MB_CUR_MAX == 1)
+#endif /* TRE_MULTIBYTE */
+    {
+      size_t i;
+      const unsigned char *str = (const unsigned char *)regex;
+      tre_char_t *wstr = wregex;
+
+      for (i = 0; i < n; i++)
+	*(wstr++) = *(str++);
+      wlen = n;
+    }
+#if TRE_MULTIBYTE
+  else
+    {
+      size_t consumed;
+      tre_char_t *wcptr = wregex;
+#ifdef HAVE_MBSTATE_T
+      mbstate_t state;
+      memset(&state, '\0', sizeof(state));
+#endif /* HAVE_MBSTATE_T */
+      while (n > 0)
+	{
+	  consumed = tre_mbrtowc(wcptr, regex, n, &state);
+
+	  switch (consumed)
+	    {
+	    case 0:
+	      if (*regex == '\0')
+		consumed = 1;
+	      else
+		{
+		  xfree(wregex);
+		  return REG_BADPAT;
+		}
+	      break;
+	    case -1:
+	      DPRINT(("mbrtowc: error %d: %s.\n", errno, strerror(errno)));
+	      xfree(wregex);
+	      return REG_BADPAT;
+	    case -2:
+	      /* The last character wasn't complete.  Let's not call it a
+		 fatal error. */
+	      consumed = n;
+	      break;
+	    }
+	  regex += consumed;
+	  n -= consumed;
+	  wcptr++;
+	}
+      wlen = wcptr - wregex;
+    }
+#endif /* TRE_MULTIBYTE */
+
+  wregex[wlen] = L'\0';
+  ret = tre_compile(preg, wregex, wlen, cflags);
+  xfree(wregex);
+#else /* !TRE_WCHAR */
+  ret = tre_compile(preg, (const tre_char_t *)regex, n, cflags);
+#endif /* !TRE_WCHAR */
+
+  return ret;
+}
+
+/* this version takes bytes literally, to be used with raw vectors */
+int
+tre_regncompb(regex_t *preg, const char *regex, size_t n, int cflags)
+{
+  int ret;
+  if (n > TRE_MAX_RE)
+    return REG_ESPACE;
+#if TRE_WCHAR /* wide chars = we need to convert it all to the wide format */
+  tre_char_t *wregex;
+  size_t i;
+
+  wregex = xmalloc(sizeof(tre_char_t) * n);
+  if (wregex == NULL)
+    return REG_ESPACE;
+
+  for (i = 0; i < n; i++)
+    wregex[i] = (tre_char_t) ((unsigned char) regex[i]);
+
+  ret = tre_compile(preg, wregex, n, cflags | REG_USEBYTES);
+  xfree(wregex);
+#else /* !TRE_WCHAR */
+  ret = tre_compile(preg, (const tre_char_t *)regex, n, cflags | REG_USEBYTES);
+#endif /* !TRE_WCHAR */
+
+  return ret;
+}
+
+int
+tre_regcomp(regex_t *preg, const char *regex, int cflags)
+{
+  size_t n = regex ? strlen(regex) : 0;
+  if (n > TRE_MAX_RE)
+    return REG_ESPACE;
+  return tre_regncomp(preg, regex, n, cflags);
+}
+
+int
+tre_regcompb(regex_t *preg, const char *regex, int cflags)
+{
+  int ret;
+  tre_char_t *wregex;
+  size_t i, n = regex ? strlen(regex) : 0;
+  const unsigned char *str = (const unsigned char *)regex;
+  tre_char_t *wstr;
+
+  if (n > TRE_MAX_RE)
+    return REG_ESPACE;
+  wregex = xmalloc(sizeof(tre_char_t) * (n + 1));
+  if (wregex == NULL) return REG_ESPACE;
+  wstr = wregex;
+
+  for (i = 0; i < n; i++)
+    *(wstr++) = *(str++);
+  wregex[n] = L'\0';
+  ret = tre_compile(preg, wregex, n, cflags | REG_USEBYTES);
+  xfree(wregex);
+  return ret;
+}
+
+
+#ifdef TRE_WCHAR
+int
+tre_regwncomp(regex_t *preg, const wchar_t *regex, size_t n, int cflags)
+{
+  if (n > TRE_MAX_RE)
+    return REG_ESPACE;
+  return tre_compile(preg, regex, n, cflags);
+}
+
+int
+tre_regwcomp(regex_t *preg, const wchar_t *regex, int cflags)
+{
+  size_t n = regex ? wcslen(regex) : 0;
+  if (n > TRE_MAX_RE)
+    return REG_ESPACE;
+  return tre_compile(preg, regex, n, cflags);
+}
+#endif /* TRE_WCHAR */
+
+void
+tre_regfree(regex_t *preg)
+{
+  tre_free(preg);
+}
+
+/* EOF */

vendored/tre/regerror.c

@@ -0,0 +1,86 @@
+/*
+  tre_regerror.c - POSIX tre_regerror() implementation for TRE.
+
+  This software is released under a BSD-style license.
+  See the file LICENSE for details and copyright.
+
+*/
+
+#ifdef HAVE_CONFIG_H
+#include <config.h>
+#endif /* HAVE_CONFIG_H */
+
+#include <string.h>
+#ifdef HAVE_WCHAR_H
+#include <wchar.h>
+#endif /* HAVE_WCHAR_H */
+#ifdef HAVE_WCTYPE_H
+#include <wctype.h>
+#endif /* HAVE_WCTYPE_H */
+
+#include "tre-internal.h"
+
+#ifdef HAVE_GETTEXT
+#include <libintl.h>
+#else
+#define dgettext(p, s) s
+#define gettext(s) s
+#endif
+
+#define _(String) dgettext(PACKAGE, String)
+#define gettext_noop(String) String
+
+#define xstr(s) str(s)
+#define str(s) #s
+
+/* Error message strings for error codes listed in `tre.h'.  This list
+   needs to be in sync with the codes listed there, naturally. */
+static const char *tre_error_messages[] =
+  { gettext_noop("No error"),							/* REG_OK */
+    gettext_noop("No match"),							/* REG_NOMATCH */
+    gettext_noop("Invalid regexp"),						/* REG_BADPAT */
+    gettext_noop("Unknown collating element"),					/* REG_ECOLLATE */
+    gettext_noop("Unknown character class name"),				/* REG_ECTYPE */
+    gettext_noop("Trailing backslash"),						/* REG_EESCAPE */
+    gettext_noop("Invalid back reference"),					/* REG_ESUBREG */
+    gettext_noop("Missing ']'"),						/* REG_EBRACK */
+    gettext_noop("Missing ')'"),						/* REG_EPAREN */
+    gettext_noop("Missing '}'"),						/* REG_EBRACE */
+    gettext_noop("Invalid contents of {}"),					/* REG_BADBR */
+    gettext_noop("Invalid character range"),					/* REG_ERANGE */
+    gettext_noop("Out of memory"),						/* REG_ESPACE */
+    gettext_noop("Invalid use of repetition operators"),			/* REG_BADRPT */
+    gettext_noop("Maximum repetition in {} larger than " xstr(RE_DUP_MAX)),	/* REG_BADMAX */
+  };
+
+size_t
+tre_regerror(int errcode, const regex_t *preg, char *errbuf, size_t errbuf_size)
+{
+  const char *err;
+  size_t err_len;
+
+  /*LINTED*/(void)&preg;
+  if (errcode >= 0
+      && errcode < (int)(sizeof(tre_error_messages)
+			 / sizeof(*tre_error_messages)))
+    err = gettext(tre_error_messages[errcode]);
+  else
+    err = gettext("Unknown error");
+
+  err_len = strlen(err) + 1;
+  if (errbuf_size > 0 && errbuf != NULL)
+    {
+      if (err_len > errbuf_size)
+	{
+	  strncpy(errbuf, err, errbuf_size - 1);
+	  errbuf[errbuf_size - 1] = '\0';
+	}
+      else
+	{
+	  strcpy(errbuf, err);
+	}
+    }
+  return err_len;
+}
+
+/* EOF */

vendored/tre/regex.h

@@ -0,0 +1,48 @@
+/*
+  regex.h - TRE legacy API
+
+  This software is released under a BSD-style license.
+  See the file LICENSE for details and copyright.
+
+  This header is for source level compatibility with old code using
+  the <tre/regex.h> header which defined the TRE API functions without
+  a prefix.  New code should include <tre/tre.h> instead.
+
+*/
+
+#ifndef TRE_REXEX_H
+#define TRE_REGEX_H 1
+
+#ifdef USE_LOCAL_TRE_H
+/* Use the header(s) from the TRE package that this file is part of.
+   (Yes, this file is in local_include too, but the explict path
+   means there is no way to get a system tre.h by accident.) */
+#include "tre.h"
+#else
+/* Use the header(s) from an installed version of the TRE package
+   (so that this application matches the installed libtre),
+   not the one(s) in the local_includes directory. */
+#include <tre/tre.h>
+#endif
+
+#ifndef TRE_USE_SYSTEM_REGEX_H
+#define regcomp	   tre_regcomp
+#define regerror   tre_regerror
+#define regexec	   tre_regexec
+#define regfree	   tre_regfree
+#endif /* TRE_USE_SYSTEM_REGEX_H */
+
+#define regacomp   tre_regacomp
+#define regaexec   tre_regaexec
+#define regancomp  tre_regancomp
+#define reganexec  tre_reganexec
+#define regawncomp tre_regawncomp
+#define regawnexec tre_regawnexec
+#define regncomp   tre_regncomp
+#define regnexec   tre_regnexec
+#define regwcomp   tre_regwcomp
+#define regwexec   tre_regwexec
+#define regwncomp  tre_regwncomp
+#define regwnexec  tre_regwnexec
+
+#endif /* TRE_REGEX_H */

vendored/tre/regexec.c

@@ -0,0 +1,387 @@
+/*
+  tre_regexec.c - TRE POSIX compatible matching functions (and more).
+
+  This software is released under a BSD-style license.
+  See the file LICENSE for details and copyright.
+
+*/
+
+#ifdef HAVE_CONFIG_H
+#include <config.h>
+#endif /* HAVE_CONFIG_H */
+
+#ifdef TRE_USE_ALLOCA
+/* AIX requires this to be the first thing in the file.	 */
+#ifndef __GNUC__
+# if HAVE_ALLOCA_H
+#  include <alloca.h>
+# else
+#  ifdef _AIX
+ #pragma alloca
+#  else
+#   ifndef alloca /* predefined by HP cc +Olibcalls */
+char *alloca ();
+#   endif
+#  endif
+# endif
+#endif
+#endif /* TRE_USE_ALLOCA */
+
+#include <assert.h>
+#include <stdlib.h>
+#include <string.h>
+#ifdef HAVE_WCHAR_H
+#include <wchar.h>
+#endif /* HAVE_WCHAR_H */
+#ifdef HAVE_WCTYPE_H
+#include <wctype.h>
+#endif /* HAVE_WCTYPE_H */
+#ifndef TRE_WCHAR
+#include <ctype.h>
+#endif /* !TRE_WCHAR */
+#ifdef HAVE_MALLOC_H
+#include <malloc.h>
+#endif /* HAVE_MALLOC_H */
+#include <limits.h>
+
+#include "tre-internal.h"
+#include "xmalloc.h"
+
+
+/* Fills the POSIX.2 regmatch_t array according to the TNFA tag and match
+   endpoint values. */
+void
+tre_fill_pmatch(size_t nmatch, regmatch_t pmatch[], int cflags,
+		const tre_tnfa_t *tnfa, int *tags, int match_eo)
+{
+  tre_submatch_data_t *submatch_data;
+  unsigned int i, j;
+  int *parents;
+
+  i = 0;
+  if (match_eo >= 0 && !(cflags & REG_NOSUB))
+    {
+      /* Construct submatch offsets from the tags. */
+      DPRINT(("end tag = t%d = %d\n", tnfa->end_tag, match_eo));
+      submatch_data = tnfa->submatch_data;
+      while (i < tnfa->num_submatches && i < nmatch)
+	{
+	  if (submatch_data[i].so_tag == tnfa->end_tag)
+	    pmatch[i].rm_so = match_eo;
+	  else
+	    pmatch[i].rm_so = tags[submatch_data[i].so_tag];
+
+	  if (submatch_data[i].eo_tag == tnfa->end_tag)
+	    pmatch[i].rm_eo = match_eo;
+	  else
+	    pmatch[i].rm_eo = tags[submatch_data[i].eo_tag];
+
+	  /* If either of the endpoints were not used, this submatch
+	     was not part of the match. */
+	  if (pmatch[i].rm_so == -1 || pmatch[i].rm_eo == -1)
+	    pmatch[i].rm_so = pmatch[i].rm_eo = -1;
+
+	  DPRINT(("pmatch[%d] = {t%d = %d, t%d = %d}\n", i,
+		  submatch_data[i].so_tag, pmatch[i].rm_so,
+		  submatch_data[i].eo_tag, pmatch[i].rm_eo));
+	  i++;
+	}
+      /* Reset all submatches that are not within all of their parent
+	 submatches. */
+      i = 0;
+      while (i < tnfa->num_submatches && i < nmatch)
+	{
+	  if (pmatch[i].rm_eo == -1)
+	    assert(pmatch[i].rm_so == -1);
+	  assert(pmatch[i].rm_so <= pmatch[i].rm_eo);
+
+	  parents = submatch_data[i].parents;
+	  if (parents != NULL)
+	    for (j = 0; parents[j] >= 0; j++)
+	      {
+		DPRINT(("pmatch[%d] parent %d\n", i, parents[j]));
+		if (pmatch[i].rm_so < pmatch[parents[j]].rm_so
+		    || pmatch[i].rm_eo > pmatch[parents[j]].rm_eo)
+		  pmatch[i].rm_so = pmatch[i].rm_eo = -1;
+	      }
+	  i++;
+	}
+    }
+
+  while (i < nmatch)
+    {
+      pmatch[i].rm_so = -1;
+      pmatch[i].rm_eo = -1;
+      i++;
+    }
+}
+
+
+/*
+  Wrapper functions for POSIX compatible regexp matching.
+*/
+
+int
+tre_have_backrefs(const regex_t *preg)
+{
+  tre_tnfa_t *tnfa = (void *)preg->TRE_REGEX_T_FIELD;
+  return tnfa->have_backrefs;
+}
+
+int
+tre_have_approx(const regex_t *preg)
+{
+  tre_tnfa_t *tnfa = (void *)preg->TRE_REGEX_T_FIELD;
+  return tnfa->have_approx;
+}
+
+static int
+tre_match(const tre_tnfa_t *tnfa, const void *string, ssize_t len,
+	  tre_str_type_t type, size_t nmatch, regmatch_t pmatch[],
+	  int eflags)
+{
+  reg_errcode_t status;
+  int *tags = NULL, eo;
+  if (tnfa->num_tags > 0 && nmatch > 0)
+    {
+#ifdef TRE_USE_ALLOCA
+      tags = alloca(sizeof(*tags) * tnfa->num_tags);
+#else /* !TRE_USE_ALLOCA */
+      tags = xmalloc(sizeof(*tags) * tnfa->num_tags);
+#endif /* !TRE_USE_ALLOCA */
+      if (tags == NULL)
+	return REG_ESPACE;
+    }
+
+  /* Dispatch to the appropriate matcher. */
+  if (tnfa->have_backrefs || eflags & REG_BACKTRACKING_MATCHER)
+    {
+      /* The regex has back references, use the backtracking matcher. */
+      if (type == STR_USER)
+	{
+	  const tre_str_source *source = string;
+	  if (source->rewind == NULL || source->compare == NULL)
+	    {
+	      /* The backtracking matcher requires rewind and compare
+		 capabilities from the input stream. */
+#ifndef TRE_USE_ALLOCA
+	      if (tags)
+		xfree(tags);
+#endif /* !TRE_USE_ALLOCA */
+	      return REG_BADPAT;
+	    }
+	}
+      status = tre_tnfa_run_backtrack(tnfa, string, len, type,
+				      tags, eflags, &eo);
+    }
+#ifdef TRE_APPROX
+  else if (tnfa->have_approx || eflags & REG_APPROX_MATCHER)
+    {
+      /* The regex uses approximate matching, use the approximate matcher. */
+      regamatch_t match;
+      regaparams_t params;
+      tre_regaparams_default(&params);
+      params.max_err = 0;
+      params.max_cost = 0;
+      status = tre_tnfa_run_approx(tnfa, string, len, type, tags,
+				   &match, params, eflags, &eo);
+    }
+#endif /* TRE_APPROX */
+  else
+    {
+      /* Exact matching, no back references, use the parallel matcher. */
+      status = tre_tnfa_run_parallel(tnfa, string, len, type,
+				     tags, eflags, &eo);
+    }
+
+  if (status == REG_OK)
+    /* A match was found, so fill the submatch registers. */
+    tre_fill_pmatch(nmatch, pmatch, tnfa->cflags, tnfa, tags, eo);
+#ifndef TRE_USE_ALLOCA
+  if (tags)
+    xfree(tags);
+#endif /* !TRE_USE_ALLOCA */
+  return status;
+}
+
+int
+tre_regnexec(const regex_t *preg, const char *str, size_t len,
+	 size_t nmatch, regmatch_t pmatch[], int eflags)
+{
+  tre_tnfa_t *tnfa = (void *)preg->TRE_REGEX_T_FIELD;
+  tre_str_type_t type = (TRE_MB_CUR_MAX == 1) ? STR_BYTE : STR_MBS;
+
+  return tre_match(tnfa, str, len, type, nmatch, pmatch, eflags);
+}
+
+#ifdef TRE_USE_GNUC_REGEXEC_FPL
+int
+tre_regexec(const regex_t *preg, const char *str,
+	size_t nmatch, regmatch_t pmatch[_Restrict_arr_ _REGEX_NELTS (nmatch)],
+	int eflags)
+#else
+int
+tre_regexec(const regex_t *preg, const char *str,
+	size_t nmatch, regmatch_t pmatch[], int eflags)
+#endif
+{
+  return tre_regnexec(preg, str, -1, nmatch, pmatch, eflags);
+}
+
+int
+tre_regexecb(const regex_t *preg, const char *str,
+        size_t nmatch, regmatch_t pmatch[], int eflags)
+{
+  tre_tnfa_t *tnfa = (void *)preg->TRE_REGEX_T_FIELD;
+
+  return tre_match(tnfa, str, -1, STR_BYTE, nmatch, pmatch, eflags);
+}
+
+int
+tre_regnexecb(const regex_t *preg, const char *str, size_t len,
+        size_t nmatch, regmatch_t pmatch[], int eflags)
+{
+  tre_tnfa_t *tnfa = (void *)preg->TRE_REGEX_T_FIELD;
+
+  return tre_match(tnfa, str, len, STR_BYTE, nmatch, pmatch, eflags);
+}
+
+
+#ifdef TRE_WCHAR
+
+int
+tre_regwnexec(const regex_t *preg, const wchar_t *str, size_t len,
+	  size_t nmatch, regmatch_t pmatch[], int eflags)
+{
+  tre_tnfa_t *tnfa = (void *)preg->TRE_REGEX_T_FIELD;
+  return tre_match(tnfa, str, len, STR_WIDE, nmatch, pmatch, eflags);
+}
+
+int
+tre_regwexec(const regex_t *preg, const wchar_t *str,
+	 size_t nmatch, regmatch_t pmatch[], int eflags)
+{
+  return tre_regwnexec(preg, str, -1, nmatch, pmatch, eflags);
+}
+
+#endif /* TRE_WCHAR */
+
+int
+tre_reguexec(const regex_t *preg, const tre_str_source *str,
+	 size_t nmatch, regmatch_t pmatch[], int eflags)
+{
+  tre_tnfa_t *tnfa = (void *)preg->TRE_REGEX_T_FIELD;
+  return tre_match(tnfa, str, -1, STR_USER, nmatch, pmatch, eflags);
+}
+
+
+#ifdef TRE_APPROX
+
+/*
+  Wrapper functions for approximate regexp matching.
+*/
+
+static int
+tre_match_approx(const tre_tnfa_t *tnfa, const void *string, ssize_t len,
+		 tre_str_type_t type, regamatch_t *match, regaparams_t params,
+		 int eflags)
+{
+  reg_errcode_t status;
+  int *tags = NULL, eo;
+
+  /* If the regexp does not use approximate matching features, the
+     maximum cost is zero, and the approximate matcher isn't forced,
+     use the exact matcher instead. */
+  if (params.max_cost == 0 && !tnfa->have_approx
+      && !(eflags & REG_APPROX_MATCHER))
+    return tre_match(tnfa, string, len, type, match->nmatch, match->pmatch,
+		     eflags);
+
+  /* Back references are not supported by the approximate matcher. */
+  if (tnfa->have_backrefs)
+    return REG_BADPAT;
+
+  if (tnfa->num_tags > 0 && match->nmatch > 0)
+    {
+#if TRE_USE_ALLOCA
+      tags = alloca(sizeof(*tags) * tnfa->num_tags);
+#else /* !TRE_USE_ALLOCA */
+      tags = xmalloc(sizeof(*tags) * tnfa->num_tags);
+#endif /* !TRE_USE_ALLOCA */
+      if (tags == NULL)
+	return REG_ESPACE;
+    }
+  status = tre_tnfa_run_approx(tnfa, string, len, type, tags,
+			       match, params, eflags, &eo);
+  if (status == REG_OK)
+    tre_fill_pmatch(match->nmatch, match->pmatch, tnfa->cflags, tnfa, tags, eo);
+#ifndef TRE_USE_ALLOCA
+  if (tags)
+    xfree(tags);
+#endif /* !TRE_USE_ALLOCA */
+  return status;
+}
+
+int
+tre_reganexec(const regex_t *preg, const char *str, size_t len,
+	  regamatch_t *match, regaparams_t params, int eflags)
+{
+  tre_tnfa_t *tnfa = (void *)preg->TRE_REGEX_T_FIELD;
+  tre_str_type_t type = (TRE_MB_CUR_MAX == 1) ? STR_BYTE : STR_MBS;
+
+  return tre_match_approx(tnfa, str, len, type, match, params, eflags);
+}
+
+int
+tre_regaexec(const regex_t *preg, const char *str,
+	 regamatch_t *match, regaparams_t params, int eflags)
+{
+  return tre_reganexec(preg, str, -1, match, params, eflags);
+}
+
+int
+tre_regaexecb(const regex_t *preg, const char *str,
+          regamatch_t *match, regaparams_t params, int eflags)
+{
+  tre_tnfa_t *tnfa = (void *)preg->TRE_REGEX_T_FIELD;
+
+  return tre_match_approx(tnfa, str, -1, STR_BYTE, match, params, eflags);
+}
+
+#ifdef TRE_WCHAR
+
+int
+tre_regawnexec(const regex_t *preg, const wchar_t *str, size_t len,
+	   regamatch_t *match, regaparams_t params, int eflags)
+{
+  tre_tnfa_t *tnfa = (void *)preg->TRE_REGEX_T_FIELD;
+  return tre_match_approx(tnfa, str, len, STR_WIDE,
+			  match, params, eflags);
+}
+
+int
+tre_regawexec(const regex_t *preg, const wchar_t *str,
+	  regamatch_t *match, regaparams_t params, int eflags)
+{
+  return tre_regawnexec(preg, str, -1, match, params, eflags);
+}
+
+#endif /* TRE_WCHAR */
+
+void
+tre_regaparams_default(regaparams_t *params)
+{
+  memset(params, 0, sizeof(*params));
+  params->cost_ins = 1;
+  params->cost_del = 1;
+  params->cost_subst = 1;
+  params->max_cost = INT_MAX;
+  params->max_ins = INT_MAX;
+  params->max_del = INT_MAX;
+  params->max_subst = INT_MAX;
+  params->max_err = INT_MAX;
+}
+
+#endif /* TRE_APPROX */
+
+/* EOF */

vendored/tre/tre-ast.c

@@ -0,0 +1,226 @@
+/*
+  tre-ast.c - Abstract syntax tree (AST) routines
+
+  This software is released under a BSD-style license.
+  See the file LICENSE for details and copyright.
+
+*/
+
+#ifdef HAVE_CONFIG_H
+#include <config.h>
+#endif /* HAVE_CONFIG_H */
+#include <assert.h>
+
+#include "tre-ast.h"
+#include "tre-mem.h"
+
+tre_ast_node_t *
+tre_ast_new_node(tre_mem_t mem, tre_ast_type_t type, size_t size)
+{
+  tre_ast_node_t *node;
+
+  node = tre_mem_calloc(mem, sizeof(*node));
+  if (!node)
+    return NULL;
+  node->obj = tre_mem_calloc(mem, size);
+  if (!node->obj)
+    return NULL;
+  node->type = type;
+  node->nullable = -1;
+  node->submatch_id = -1;
+
+  return node;
+}
+
+tre_ast_node_t *
+tre_ast_new_literal(tre_mem_t mem, int code_min, int code_max)
+{
+  tre_ast_node_t *node;
+  tre_literal_t *lit;
+
+  node = tre_ast_new_node(mem, LITERAL, sizeof(tre_literal_t));
+  if (!node)
+    return NULL;
+  lit = node->obj;
+  lit->code_min = code_min;
+  lit->code_max = code_max;
+  lit->position = -1;
+
+  return node;
+}
+
+tre_ast_node_t *
+tre_ast_new_iter(tre_mem_t mem, tre_ast_node_t *arg, int min, int max,
+		 int minimal)
+{
+  tre_ast_node_t *node;
+  tre_iteration_t *iter;
+
+  node = tre_ast_new_node(mem, ITERATION, sizeof(tre_iteration_t));
+  if (!node)
+    return NULL;
+  iter = node->obj;
+  iter->arg = arg;
+  iter->min = min;
+  iter->max = max;
+  iter->minimal = minimal;
+  node->num_submatches = arg->num_submatches;
+
+  return node;
+}
+
+tre_ast_node_t *
+tre_ast_new_union(tre_mem_t mem, tre_ast_node_t *left, tre_ast_node_t *right)
+{
+  tre_ast_node_t *node;
+
+  node = tre_ast_new_node(mem, UNION, sizeof(tre_union_t));
+  if (node == NULL)
+    return NULL;
+  ((tre_union_t *)node->obj)->left = left;
+  ((tre_union_t *)node->obj)->right = right;
+  node->num_submatches = left->num_submatches + right->num_submatches;
+
+  return node;
+}
+
+tre_ast_node_t *
+tre_ast_new_catenation(tre_mem_t mem, tre_ast_node_t *left,
+		       tre_ast_node_t *right)
+{
+  tre_ast_node_t *node;
+
+  node = tre_ast_new_node(mem, CATENATION, sizeof(tre_catenation_t));
+  if (node == NULL)
+    return NULL;
+  ((tre_catenation_t *)node->obj)->left = left;
+  ((tre_catenation_t *)node->obj)->right = right;
+  node->num_submatches = left->num_submatches + right->num_submatches;
+
+  return node;
+}
+
+#ifdef TRE_DEBUG
+
+static void
+tre_findent(FILE *stream, int i)
+{
+  while (i-- > 0)
+    fputc(' ', stream);
+}
+
+void
+tre_print_params(int *params)
+{
+  int i;
+  if (params)
+    {
+      DPRINT(("params ["));
+      for (i = 0; i < TRE_PARAM_LAST; i++)
+	{
+	  if (params[i] == TRE_PARAM_UNSET)
+	    DPRINT(("unset"));
+	  else if (params[i] == TRE_PARAM_DEFAULT)
+	    DPRINT(("default"));
+	  else
+	    DPRINT(("%d", params[i]));
+	  if (i < TRE_PARAM_LAST - 1)
+	    DPRINT((", "));
+	}
+      DPRINT(("]"));
+    }
+}
+
+static void
+tre_do_print(FILE *stream, tre_ast_node_t *ast, int indent)
+{
+  int code_min, code_max, pos;
+  int num_tags = ast->num_tags;
+  tre_literal_t *lit;
+  tre_iteration_t *iter;
+
+  tre_findent(stream, indent);
+  switch (ast->type)
+    {
+    case LITERAL:
+      lit = ast->obj;
+      code_min = lit->code_min;
+      code_max = lit->code_max;
+      pos = lit->position;
+      if (IS_EMPTY(lit))
+	{
+	  fprintf(stream, "literal empty\n");
+	}
+      else if (IS_ASSERTION(lit))
+	{
+	  int i;
+	  char *assertions[] = { "bol", "eol", "ctype", "!ctype",
+				 "bow", "eow", "wb", "!wb" };
+	  if (code_max >= ASSERT_LAST << 1)
+	    assert(0);
+	  fprintf(stream, "assertions: ");
+	  for (i = 0; (1 << i) <= ASSERT_LAST; i++)
+	    if (code_max & (1 << i))
+	      fprintf(stream, "%s ", assertions[i]);
+	  fprintf(stream, "\n");
+	}
+      else if (IS_TAG(lit))
+	{
+	  fprintf(stream, "tag %d\n", code_max);
+	}
+      else if (IS_BACKREF(lit))
+	{
+	  fprintf(stream, "backref %d, pos %d\n", code_max, pos);
+	}
+      else if (IS_PARAMETER(lit))
+	{
+	  tre_print_params(lit->u.params);
+	  fprintf(stream, "\n");
+	}
+      else
+	{
+	  fprintf(stream, "literal (%c, %c) (%d, %d), pos %d, sub %d, "
+		  "%d tags\n", code_min, code_max, code_min, code_max, pos,
+		  ast->submatch_id, num_tags);
+	}
+      break;
+    case ITERATION:
+      iter = ast->obj;
+      fprintf(stream, "iteration {%d, %d}, sub %d, %d tags, %s\n",
+	      iter->min, iter->max, ast->submatch_id, num_tags,
+	      iter->minimal ? "minimal" : "greedy");
+      tre_do_print(stream, iter->arg, indent + 2);
+      break;
+    case UNION:
+      fprintf(stream, "union, sub %d, %d tags\n", ast->submatch_id, num_tags);
+      tre_do_print(stream, ((tre_union_t *)ast->obj)->left, indent + 2);
+      tre_do_print(stream, ((tre_union_t *)ast->obj)->right, indent + 2);
+      break;
+    case CATENATION:
+      fprintf(stream, "catenation, sub %d, %d tags\n", ast->submatch_id,
+	      num_tags);
+      tre_do_print(stream, ((tre_catenation_t *)ast->obj)->left, indent + 2);
+      tre_do_print(stream, ((tre_catenation_t *)ast->obj)->right, indent + 2);
+      break;
+    default:
+      assert(0);
+      break;
+    }
+}
+
+static void
+tre_ast_fprint(FILE *stream, tre_ast_node_t *ast)
+{
+  tre_do_print(stream, ast, 0);
+}
+
+void
+tre_ast_print(tre_ast_node_t *tree)
+{
+  printf("AST:\n");
+  tre_ast_fprint(stdout, tree);
+}
+
+#endif /* TRE_DEBUG */
+
+/* EOF */

vendored/tre/tre-ast.h

@@ -0,0 +1,128 @@
+/*
+  tre-ast.h - Abstract syntax tree (AST) definitions
+
+  This software is released under a BSD-style license.
+  See the file LICENSE for details and copyright.
+
+*/
+
+
+#ifndef TRE_AST_H
+#define TRE_AST_H 1
+
+#include "tre-mem.h"
+#include "tre-internal.h"
+#include "tre-compile.h"
+
+/* The different AST node types. */
+typedef enum {
+  LITERAL,
+  CATENATION,
+  ITERATION,
+  UNION
+} tre_ast_type_t;
+
+/* Special subtypes of TRE_LITERAL. */
+#define EMPTY	  -1   /* Empty leaf (denotes empty string). */
+#define ASSERTION -2   /* Assertion leaf. */
+#define TAG	  -3   /* Tag leaf. */
+#define BACKREF	  -4   /* Back reference leaf. */
+#define PARAMETER -5   /* Parameter. */
+
+#define IS_SPECIAL(x)	((x)->code_min < 0)
+#define IS_EMPTY(x)	((x)->code_min == EMPTY)
+#define IS_ASSERTION(x) ((x)->code_min == ASSERTION)
+#define IS_TAG(x)	((x)->code_min == TAG)
+#define IS_BACKREF(x)	((x)->code_min == BACKREF)
+#define IS_PARAMETER(x) ((x)->code_min == PARAMETER)
+
+
+/* A generic AST node.  All AST nodes consist of this node on the top
+   level with `obj' pointing to the actual content. */
+typedef struct {
+  tre_ast_type_t type;   /* Type of the node. */
+  void *obj;             /* Pointer to actual node. */
+  int nullable;
+  int submatch_id;
+  unsigned int num_submatches;
+  unsigned int num_tags;
+  tre_pos_and_tags_t *firstpos;
+  tre_pos_and_tags_t *lastpos;
+} tre_ast_node_t;
+
+
+/* A "literal" node.  These are created for assertions, back references,
+   tags, matching parameter settings, and all expressions that match one
+   character. */
+typedef struct {
+  long code_min;
+  long code_max;
+  int position;
+  union {
+    tre_ctype_t class;
+    int *params;
+  } u;
+  tre_ctype_t *neg_classes;
+} tre_literal_t;
+
+/* A "catenation" node.	 These are created when two regexps are concatenated.
+   If there are more than one subexpressions in sequence, the `left' part
+   holds all but the last, and `right' part holds the last subexpression
+   (catenation is left associative). */
+typedef struct {
+  tre_ast_node_t *left;
+  tre_ast_node_t *right;
+} tre_catenation_t;
+
+/* An "iteration" node.	 These are created for the "*", "+", "?", and "{m,n}"
+   operators. */
+typedef struct {
+  /* Subexpression to match. */
+  tre_ast_node_t *arg;
+  /* Minimum number of consecutive matches. */
+  int min;
+  /* Maximum number of consecutive matches. */
+  int max;
+  /* If 0, match as many characters as possible, if 1 match as few as
+     possible.	Note that this does not always mean the same thing as
+     matching as many/few repetitions as possible. */
+  unsigned int minimal:1;
+  /* Approximate matching parameters (or NULL). */
+  int *params;
+} tre_iteration_t;
+
+/* An "union" node.  These are created for the "|" operator. */
+typedef struct {
+  tre_ast_node_t *left;
+  tre_ast_node_t *right;
+} tre_union_t;
+
+tre_ast_node_t *
+tre_ast_new_node(tre_mem_t mem, tre_ast_type_t type, size_t size);
+
+tre_ast_node_t *
+tre_ast_new_literal(tre_mem_t mem, int code_min, int code_max);
+
+tre_ast_node_t *
+tre_ast_new_iter(tre_mem_t mem, tre_ast_node_t *arg, int min, int max,
+		 int minimal);
+
+tre_ast_node_t *
+tre_ast_new_union(tre_mem_t mem, tre_ast_node_t *left, tre_ast_node_t *right);
+
+tre_ast_node_t *
+tre_ast_new_catenation(tre_mem_t mem, tre_ast_node_t *left,
+		       tre_ast_node_t *right);
+
+#ifdef TRE_DEBUG
+void
+tre_ast_print(tre_ast_node_t *tree);
+
+/* XXX - rethink AST printing API */
+void
+tre_print_params(int *params);
+#endif /* TRE_DEBUG */
+
+#endif /* TRE_AST_H */
+
+/* EOF */

vendored/tre/tre-compile.h

@@ -0,0 +1,27 @@
+/*
+  tre-compile.h: Regex compilation definitions
+
+  This software is released under a BSD-style license.
+  See the file LICENSE for details and copyright.
+
+*/
+
+
+#ifndef TRE_COMPILE_H
+#define TRE_COMPILE_H 1
+
+typedef struct {
+  int position;
+  int code_min;
+  int code_max;
+  int *tags;
+  int assertions;
+  tre_ctype_t class;
+  tre_ctype_t *neg_classes;
+  int backref;
+  int *params;
+} tre_pos_and_tags_t;
+
+#endif /* TRE_COMPILE_H */
+
+/* EOF */

vendored/tre/tre-config.h

@@ -0,0 +1,18 @@
+/* Minimal TRE config for vendored use — no autoconf needed */
+#define TRE_VERSION "0.9.0"
+#define HAVE_STDLIB_H 1
+#define HAVE_STRING_H 1
+#define HAVE_WCHAR_H 1
+#define HAVE_WCTYPE_H 1
+#define HAVE_MBSTATE_T 1
+/* Do NOT define TRE_MULTIBYTE or TRE_WCHAR — we use single-byte char only.
+ * #ifdef checks presence, not value, so defining as 0 still enables them. */
+/* #undef TRE_MULTIBYTE */
+/* #undef TRE_WCHAR */
+#define TRE_APPROX 0
+#define TRE_REGEX_T_FIELD value
+#define HAVE_MBRTOWC 1
+#define HAVE_MBTOWC 1
+#define HAVE_WCSLEN 1
+#define HAVE_ISASCII 1
+#define NDEBUG 1

vendored/tre/tre-filter.c

@@ -0,0 +1,73 @@
+/*
+  tre-filter.c: Histogram filter to quickly find regexp match candidates
+
+  This software is released under a BSD-style license.
+  See the file LICENSE for details and copyright.
+
+*/
+
+/* The idea of this filter is quite simple.  First, let's assume the
+   search pattern is a simple string.  In order for a substring of a
+   longer string to match the search pattern, it must have the same
+   numbers of different characters as the pattern, and those
+   characters must occur in the same order as they occur in pattern. */
+
+#ifdef HAVE_CONFIG_H
+#include <config.h>
+#endif /* HAVE_CONFIG_H */
+#include <stdio.h>
+#include "tre-internal.h"
+#include "tre-filter.h"
+
+int
+tre_filter_find(const unsigned char *str, size_t len, tre_filter_t *filter)
+{
+  unsigned short counts[256];
+  unsigned int i;
+  unsigned int window_len = filter->window_len;
+  tre_filter_profile_t *profile = filter->profile;
+  const unsigned char *str_orig = str;
+
+  DPRINT(("tre_filter_find: %.*s\n", len, str));
+
+  for (i = 0; i < elementsof(counts); i++)
+    counts[i] = 0;
+
+  i = 0;
+  while (*str && i < window_len && i < len)
+    {
+      counts[*str]++;
+      i++;
+      str++;
+      len--;
+    }
+
+  while (len > 0)
+    {
+      tre_filter_profile_t *p;
+      counts[*str]++;
+      counts[*(str - window_len)]--;
+
+      p = profile;
+      while (p->ch)
+	{
+	  if (counts[p->ch] < p->count)
+	    break;
+	  p++;
+	}
+      if (!p->ch)
+	{
+	  DPRINT(("Found possible match at %d\n",
+		  str - str_orig));
+	  return str - str_orig;
+	}
+      else
+	{
+	  DPRINT(("No match so far...\n"));
+	}
+      len--;
+      str++;
+    }
+  DPRINT(("This string cannot match.\n"));
+  return -1;
+}

vendored/tre/tre-filter.h

@@ -0,0 +1,19 @@
+
+
+
+
+typedef struct {
+  unsigned char ch;
+  unsigned char count;
+} tre_filter_profile_t;
+
+typedef struct {
+  /* Length of the window where the character counts are kept. */
+  int window_len;
+  /* Required character counts table. */
+  tre_filter_profile_t *profile;
+} tre_filter_t;
+
+
+int
+tre_filter_find(const unsigned char *str, size_t len, tre_filter_t *filter);

vendored/tre/tre-internal.h

@@ -0,0 +1,306 @@
+/*
+  tre-internal.h - TRE internal definitions
+
+  This software is released under a BSD-style license.
+  See the file LICENSE for details and copyright.
+
+*/
+
+#ifndef TRE_INTERNAL_H
+#define TRE_INTERNAL_H 1
+
+#ifdef _WIN32
+#include <basetsd.h>
+typedef SSIZE_T ssize_t;
+#else
+#include <sys/types.h> /* ssize_t */
+#endif
+
+#ifdef HAVE_WCHAR_H
+#include <wchar.h>
+#endif /* HAVE_WCHAR_H */
+
+#ifdef HAVE_WCTYPE_H
+#include <wctype.h>
+#endif /* HAVE_WCTYPE_H */
+
+#ifdef HAVE_SYS_TYPES_H
+#include <sys/types.h>
+#endif /* HAVE_SYS_TYPES_H */
+
+#include <limits.h>
+#include <ctype.h>
+#include "tre.h"
+
+#define TRE_MAX_RE	65536
+#define TRE_MAX_STRING	INT_MAX
+#define TRE_MAX_STACK	1048576
+
+#ifdef TRE_DEBUG
+#include <stdio.h>
+#define DPRINT(msg) do {printf msg; fflush(stdout);} while(/*CONSTCOND*/(void)0,0)
+#else /* !TRE_DEBUG */
+#define DPRINT(msg) do { } while(/*CONSTCOND*/(void)0,0)
+#endif /* !TRE_DEBUG */
+
+#define elementsof(x)	( sizeof(x) / sizeof(x[0]) )
+
+#ifdef HAVE_MBRTOWC
+#define tre_mbrtowc(pwc, s, n, ps) (mbrtowc((pwc), (s), (n), (ps)))
+#else /* !HAVE_MBRTOWC */
+#ifdef HAVE_MBTOWC
+#define tre_mbrtowc(pwc, s, n, ps) (mbtowc((pwc), (s), (n)))
+#endif /* HAVE_MBTOWC */
+#endif /* !HAVE_MBRTOWC */
+
+#ifdef TRE_MULTIBYTE
+#ifdef HAVE_MBSTATE_T
+#define TRE_MBSTATE
+#endif /* TRE_MULTIBYTE */
+#endif /* HAVE_MBSTATE_T */
+
+/* Define the character types and functions. */
+#ifdef TRE_WCHAR
+
+/* Wide characters. */
+typedef wint_t tre_cint_t;
+#if WCHAR_MAX <= INT_MAX
+#define TRE_CHAR_MAX WCHAR_MAX
+#else /* WCHAR_MAX > INT_MAX */
+#define TRE_CHAR_MAX INT_MAX
+#endif
+
+#ifdef TRE_MULTIBYTE
+#define TRE_MB_CUR_MAX MB_CUR_MAX
+#else /* !TRE_MULTIBYTE */
+#define TRE_MB_CUR_MAX 1
+#endif /* !TRE_MULTIBYTE */
+
+#define tre_isalnum iswalnum
+#define tre_isalpha iswalpha
+#ifdef HAVE_ISWBLANK
+#define tre_isblank iswblank
+#endif /* HAVE_ISWBLANK */
+#define tre_iscntrl iswcntrl
+#define tre_isdigit iswdigit
+#define tre_isgraph iswgraph
+#define tre_islower iswlower
+#define tre_isprint iswprint
+#define tre_ispunct iswpunct
+#define tre_isspace iswspace
+#define tre_isupper iswupper
+#define tre_isxdigit iswxdigit
+
+#define tre_tolower towlower
+#define tre_toupper towupper
+#define tre_strlen  wcslen
+
+#else /* !TRE_WCHAR */
+
+/* 8 bit characters. */
+typedef short tre_cint_t;
+#define TRE_CHAR_MAX 255
+#define TRE_MB_CUR_MAX 1
+
+#define tre_isalnum isalnum
+#define tre_isalpha isalpha
+#ifdef HAVE_ISASCII
+#define tre_isascii isascii
+#endif /* HAVE_ISASCII */
+#ifdef HAVE_ISBLANK
+#define tre_isblank isblank
+#endif /* HAVE_ISBLANK */
+#define tre_iscntrl iscntrl
+#define tre_isdigit isdigit
+#define tre_isgraph isgraph
+#define tre_islower islower
+#define tre_isprint isprint
+#define tre_ispunct ispunct
+#define tre_isspace isspace
+#define tre_isupper isupper
+#define tre_isxdigit isxdigit
+
+#define tre_tolower(c) (tre_cint_t)(tolower(c))
+#define tre_toupper(c) (tre_cint_t)(toupper(c))
+#define tre_strlen(s)  (strlen((const char*)s))
+
+#endif /* !TRE_WCHAR */
+
+#if defined(TRE_WCHAR) && defined(HAVE_ISWCTYPE) && defined(HAVE_WCTYPE)
+#define TRE_USE_SYSTEM_WCTYPE 1
+#endif
+
+#ifdef TRE_USE_SYSTEM_WCTYPE
+/* Use system provided iswctype() and wctype(). */
+typedef wctype_t tre_ctype_t;
+#define tre_isctype iswctype
+#define tre_ctype   wctype
+#else /* !TRE_USE_SYSTEM_WCTYPE */
+/* Define our own versions of iswctype() and wctype(). */
+typedef int (*tre_ctype_t)(tre_cint_t);
+#define tre_isctype(c, type) ( (type)(c) )
+tre_ctype_t tre_ctype(const char *name);
+#endif /* !TRE_USE_SYSTEM_WCTYPE */
+
+typedef enum { STR_WIDE, STR_BYTE, STR_MBS, STR_USER } tre_str_type_t;
+
+/* Returns number of bytes to add to (char *)ptr to make it
+   properly aligned for the type. */
+#define ALIGN(ptr, type) \
+  ((((long)ptr) % sizeof(type)) \
+   ? (sizeof(type) - (((long)ptr) % sizeof(type))) \
+   : 0)
+
+#undef MAX
+#undef MIN
+#define MAX(a, b) (((a) >= (b)) ? (a) : (b))
+#define MIN(a, b) (((a) <= (b)) ? (a) : (b))
+
+/* Define STRF to the correct printf formatter for strings. */
+#ifdef TRE_WCHAR
+#define STRF "ls"
+#else /* !TRE_WCHAR */
+#define STRF "s"
+#endif /* !TRE_WCHAR */
+
+/* TNFA transition type. A TNFA state is an array of transitions,
+   the terminator is a transition with NULL `state'. */
+typedef struct tnfa_transition tre_tnfa_transition_t;
+
+struct tnfa_transition {
+  /* Range of accepted characters. */
+  tre_cint_t code_min;
+  tre_cint_t code_max;
+  /* Pointer to the destination state. */
+  tre_tnfa_transition_t *state;
+  /* ID number of the destination state. */
+  int state_id;
+  /* -1 terminated array of tags (or NULL). */
+  int *tags;
+  /* Matching parameters settings (or NULL). */
+  int *params;
+  /* Assertion bitmap. */
+  int assertions;
+  /* Assertion parameters. */
+  union {
+    /* Character class assertion. */
+    tre_ctype_t class;
+    /* Back reference assertion. */
+    int backref;
+  } u;
+  /* Negative character class assertions. */
+  tre_ctype_t *neg_classes;
+};
+
+
+/* Assertions. */
+#define ASSERT_AT_BOL		  1   /* Beginning of line. */
+#define ASSERT_AT_EOL		  2   /* End of line. */
+#define ASSERT_CHAR_CLASS	  4   /* Character class in `class'. */
+#define ASSERT_CHAR_CLASS_NEG	  8   /* Character classes in `neg_classes'. */
+#define ASSERT_AT_BOW		 16   /* Beginning of word. */
+#define ASSERT_AT_EOW		 32   /* End of word. */
+#define ASSERT_AT_WB		 64   /* Word boundary. */
+#define ASSERT_AT_WB_NEG	128   /* Not a word boundary. */
+#define ASSERT_BACKREF		256   /* A back reference in `backref'. */
+#define ASSERT_LAST		256
+
+/* Tag directions. */
+typedef enum {
+  TRE_TAG_MINIMIZE = 0,
+  TRE_TAG_MAXIMIZE = 1
+} tre_tag_direction_t;
+
+/* Parameters that can be changed dynamically while matching. */
+typedef enum {
+  TRE_PARAM_COST_INS	    = 0,
+  TRE_PARAM_COST_DEL	    = 1,
+  TRE_PARAM_COST_SUBST	    = 2,
+  TRE_PARAM_COST_MAX	    = 3,
+  TRE_PARAM_MAX_INS	    = 4,
+  TRE_PARAM_MAX_DEL	    = 5,
+  TRE_PARAM_MAX_SUBST	    = 6,
+  TRE_PARAM_MAX_ERR	    = 7,
+  TRE_PARAM_DEPTH	    = 8,
+  TRE_PARAM_LAST	    = 9
+} tre_param_t;
+
+/* Unset matching parameter */
+#define TRE_PARAM_UNSET -1
+
+/* Signifies the default matching parameter value. */
+#define TRE_PARAM_DEFAULT -2
+
+/* Instructions to compute submatch register values from tag values
+   after a successful match.  */
+struct tre_submatch_data {
+  /* Tag that gives the value for rm_so (submatch start offset). */
+  int so_tag;
+  /* Tag that gives the value for rm_eo (submatch end offset). */
+  int eo_tag;
+  /* List of submatches this submatch is contained in. */
+  int *parents;
+};
+
+typedef struct tre_submatch_data tre_submatch_data_t;
+
+
+/* TNFA definition. */
+typedef struct tnfa tre_tnfa_t;
+
+struct tnfa {
+  tre_tnfa_transition_t *transitions;
+  unsigned int num_transitions;
+  tre_tnfa_transition_t *initial;
+  tre_tnfa_transition_t *final;
+  tre_submatch_data_t *submatch_data;
+  char *firstpos_chars;
+  int first_char;
+  unsigned int num_submatches;
+  tre_tag_direction_t *tag_directions;
+  int *minimal_tags;
+  int num_tags;
+  int num_minimals;
+  int end_tag;
+  int num_states;
+  int cflags;
+  int have_backrefs;
+  int have_approx;
+  int params_depth;
+};
+
+int
+tre_compile(regex_t *preg, const tre_char_t *regex, size_t n, int cflags);
+
+void
+tre_free(regex_t *preg);
+
+void
+tre_fill_pmatch(size_t nmatch, regmatch_t pmatch[], int cflags,
+		const tre_tnfa_t *tnfa, int *tags, int match_eo);
+
+reg_errcode_t
+tre_tnfa_run_parallel(const tre_tnfa_t *tnfa, const void *string, ssize_t len,
+		      tre_str_type_t type, int *match_tags, int eflags,
+		      int *match_end_ofs);
+
+reg_errcode_t
+tre_tnfa_run_parallel(const tre_tnfa_t *tnfa, const void *string, ssize_t len,
+		      tre_str_type_t type, int *match_tags, int eflags,
+		      int *match_end_ofs);
+
+reg_errcode_t
+tre_tnfa_run_backtrack(const tre_tnfa_t *tnfa, const void *string, ssize_t len,
+		       tre_str_type_t type, int *match_tags, int eflags,
+		       int *match_end_ofs);
+
+#ifdef TRE_APPROX
+reg_errcode_t
+tre_tnfa_run_approx(const tre_tnfa_t *tnfa, const void *string, ssize_t len,
+		    tre_str_type_t type, int *match_tags, regamatch_t *match,
+		    regaparams_t params, int eflags, int *match_end_ofs);
+#endif /* TRE_APPROX */
+
+#endif /* TRE_INTERNAL_H */
+
+/* EOF */

vendored/tre/tre-match-approx.c

@@ -0,0 +1,819 @@
+/*
+  tre-match-approx.c - TRE approximate regex matching engine
+
+  This software is released under a BSD-style license.
+  See the file LICENSE for details and copyright.
+
+*/
+
+#ifdef HAVE_CONFIG_H
+#include <config.h>
+#endif /* HAVE_CONFIG_H */
+
+/* AIX requires this to be the first thing in the file.	 */
+#ifdef TRE_USE_ALLOCA
+#ifndef __GNUC__
+# if HAVE_ALLOCA_H
+#  include <alloca.h>
+# else
+#  ifdef _AIX
+ #pragma alloca
+#  else
+#   ifndef alloca /* predefined by HP cc +Olibcalls */
+char *alloca ();
+#   endif
+#  endif
+# endif
+#endif
+#endif /* TRE_USE_ALLOCA */
+
+#include <assert.h>
+#include <stdlib.h>
+#include <string.h>
+#include <limits.h>
+#ifdef HAVE_WCHAR_H
+#include <wchar.h>
+#endif /* HAVE_WCHAR_H */
+#ifdef HAVE_WCTYPE_H
+#include <wctype.h>
+#endif /* HAVE_WCTYPE_H */
+#ifndef TRE_WCHAR
+#include <ctype.h>
+#endif /* !TRE_WCHAR */
+#ifdef HAVE_MALLOC_H
+#include <malloc.h>
+#endif /* HAVE_MALLOC_H */
+
+#include "tre-internal.h"
+#include "tre-match-utils.h"
+#include "xmalloc.h"
+
+#define TRE_M_COST	0
+#define TRE_M_NUM_INS	1
+#define TRE_M_NUM_DEL	2
+#define TRE_M_NUM_SUBST 3
+#define TRE_M_NUM_ERR	4
+#define TRE_M_LAST	5
+
+#define TRE_M_MAX_DEPTH 3
+
+typedef struct {
+  /* State in the TNFA transition table. */
+  tre_tnfa_transition_t *state;
+  /* Position in input string. */
+  int pos;
+  /* Tag values. */
+  int *tags;
+  /* Matching parameters. */
+  regaparams_t params;
+  /* Nesting depth of parameters.  This is used as an index in
+     the `costs' array. */
+  int depth;
+  /* Costs and counter values for different parameter nesting depths. */
+  int costs[TRE_M_MAX_DEPTH + 1][TRE_M_LAST];
+} tre_tnfa_approx_reach_t;
+
+
+#ifdef TRE_DEBUG
+/* Prints the `reach' array in a readable fashion with DPRINT. */
+static void
+tre_print_reach(const tre_tnfa_t *tnfa, tre_tnfa_approx_reach_t *reach,
+		int pos, int num_tags)
+{
+  int id;
+
+  /* Print each state on one line. */
+  DPRINT(("  reach:\n"));
+  for (id = 0; id < tnfa->num_states; id++)
+    {
+      int i, j;
+      if (reach[id].pos < pos)
+	continue;  /* Not reached. */
+      DPRINT(("	 %03d, costs ", id));
+      for (i = 0; i <= reach[id].depth; i++)
+	{
+	  DPRINT(("["));
+	  for (j = 0; j < TRE_M_LAST; j++)
+	    {
+	      DPRINT(("%2d", reach[id].costs[i][j]));
+	      if (j + 1 < TRE_M_LAST)
+		DPRINT((","));
+	    }
+	  DPRINT(("]"));
+	  if (i + 1 <= reach[id].depth)
+	    DPRINT((", "));
+	}
+      DPRINT(("\n	tags "));
+      for (i = 0; i < num_tags; i++)
+	{
+	  DPRINT(("%02d", reach[id].tags[i]));
+	  if (i + 1 < num_tags)
+	    DPRINT((","));
+	}
+      DPRINT(("\n"));
+    }
+  DPRINT(("\n"));
+}
+#endif /* TRE_DEBUG */
+
+
+/* Sets the matching parameters in `reach' to the ones defined in the `pa'
+   array.  If `pa' specifies default values, they are taken from
+   `default_params'. */
+inline static void
+tre_set_params(tre_tnfa_approx_reach_t *reach,
+	       int *pa, regaparams_t default_params)
+{
+  int value;
+
+  /* If depth is increased reset costs and counters to zero for the
+     new levels. */
+  value = pa[TRE_PARAM_DEPTH];
+  assert(value <= TRE_M_MAX_DEPTH);
+  if (value > reach->depth)
+    {
+      int i, j;
+      for (i = reach->depth + 1; i <= value; i++)
+	for (j = 0; j < TRE_M_LAST; j++)
+	  reach->costs[i][j] = 0;
+    }
+  reach->depth = value;
+
+  /* Set insert cost. */
+  value = pa[TRE_PARAM_COST_INS];
+  if (value == TRE_PARAM_DEFAULT)
+    reach->params.cost_ins = default_params.cost_ins;
+  else if (value != TRE_PARAM_UNSET)
+    reach->params.cost_ins = value;
+
+  /* Set delete cost. */
+  value = pa[TRE_PARAM_COST_DEL];
+  if (value == TRE_PARAM_DEFAULT)
+    reach->params.cost_del = default_params.cost_del;
+  else if (value != TRE_PARAM_UNSET)
+    reach->params.cost_del = value;
+
+  /* Set substitute cost. */
+  value = pa[TRE_PARAM_COST_SUBST];
+  if (value == TRE_PARAM_DEFAULT)
+    reach->params.cost_subst = default_params.cost_subst;
+  else
+    reach->params.cost_subst = value;
+
+  /* Set maximum cost. */
+  value = pa[TRE_PARAM_COST_MAX];
+  if (value == TRE_PARAM_DEFAULT)
+    reach->params.max_cost = default_params.max_cost;
+  else if (value != TRE_PARAM_UNSET)
+    reach->params.max_cost = value;
+
+  /* Set maximum inserts. */
+  value = pa[TRE_PARAM_MAX_INS];
+  if (value == TRE_PARAM_DEFAULT)
+    reach->params.max_ins = default_params.max_ins;
+  else if (value != TRE_PARAM_UNSET)
+    reach->params.max_ins = value;
+
+  /* Set maximum deletes. */
+  value = pa[TRE_PARAM_MAX_DEL];
+  if (value == TRE_PARAM_DEFAULT)
+    reach->params.max_del = default_params.max_del;
+  else if (value != TRE_PARAM_UNSET)
+    reach->params.max_del = value;
+
+  /* Set maximum substitutes. */
+  value = pa[TRE_PARAM_MAX_SUBST];
+  if (value == TRE_PARAM_DEFAULT)
+    reach->params.max_subst = default_params.max_subst;
+  else if (value != TRE_PARAM_UNSET)
+    reach->params.max_subst = value;
+
+  /* Set maximum number of errors. */
+  value = pa[TRE_PARAM_MAX_ERR];
+  if (value == TRE_PARAM_DEFAULT)
+    reach->params.max_err = default_params.max_err;
+  else if (value != TRE_PARAM_UNSET)
+    reach->params.max_err = value;
+}
+
+reg_errcode_t
+tre_tnfa_run_approx(const tre_tnfa_t *tnfa, const void *string, ssize_t len,
+		    tre_str_type_t type, int *match_tags,
+		    regamatch_t *match, regaparams_t default_params,
+		    int eflags, int *match_end_ofs)
+{
+  /* State variables required by GET_NEXT_WCHAR. */
+  tre_char_t prev_c = 0, next_c = 0;
+  const char *str_byte = string;
+  ssize_t pos = -1;
+  unsigned int pos_add_next = 1;
+#ifdef TRE_WCHAR
+  const wchar_t *str_wide = string;
+#ifdef TRE_MBSTATE
+  mbstate_t mbstate;
+#endif /* !TRE_WCHAR */
+#endif /* TRE_WCHAR */
+  reg_errcode_t ret;
+  int reg_notbol = eflags & REG_NOTBOL;
+  int reg_noteol = eflags & REG_NOTEOL;
+  int reg_newline = tnfa->cflags & REG_NEWLINE;
+  int str_user_end = 0;
+
+  int prev_pos;
+
+  /* Number of tags. */
+  int num_tags;
+  /* The reach tables. */
+  tre_tnfa_approx_reach_t *reach, *reach_next;
+  /* Tag array for temporary use. */
+  int *tmp_tags;
+
+  /* End offset of best match so far, or -1 if no match found yet. */
+  int match_eo = -1;
+  /* Costs of the match. */
+  int match_costs[TRE_M_LAST];
+
+  /* Space for temporary data required for matching. */
+  unsigned char *buf;
+
+  int i, id;
+
+  /*
+   * TRE internals tend to use int instead of size_t for positions or
+   * lengths and don't check for overflow.  This will take time to fix
+   * properly.  In the meantime, simply limit the input to what we can
+   * handle.
+   */
+  if (len > TRE_MAX_STRING)
+    len = TRE_MAX_STRING;
+
+#ifdef TRE_MBSTATE
+  memset(&mbstate, '\0', sizeof(mbstate));
+#endif /* TRE_MBSTATE */
+
+  DPRINT(("tre_tnfa_run_approx, input type %d, len %zd, eflags %d, "
+	  "match_tags %p\n",
+	  type, len, eflags,
+	  match_tags));
+  DPRINT(("max cost %d, ins %d, del %d, subst %d\n",
+	  default_params.max_cost,
+	  default_params.cost_ins,
+	  default_params.cost_del,
+	  default_params.cost_subst));
+
+  if (!match_tags)
+    num_tags = 0;
+  else
+    num_tags = tnfa->num_tags;
+
+  /* Allocate memory for temporary data required for matching.	This needs to
+     be done for every matching operation to be thread safe.  This allocates
+     everything in a single large block from the stack frame using alloca()
+     or with malloc() if alloca is unavailable. */
+  {
+    unsigned char *buf_cursor;
+    /* Space needed for one array of tags. */
+    int tag_bytes = sizeof(*tmp_tags) * num_tags;
+    /* Space needed for one reach table. */
+    int reach_bytes = sizeof(*reach_next) * tnfa->num_states;
+    /* Total space needed. */
+    int total_bytes = reach_bytes * 2 + (tnfa->num_states * 2 + 1 ) * tag_bytes;
+    /* Add some extra to make sure we can align the pointers.  The multiplier
+       used here must be equal to the number of ALIGN calls below. */
+    total_bytes += (sizeof(long) - 1) * 3;
+
+    /* Allocate the memory. */
+#ifdef TRE_USE_ALLOCA
+    buf = alloca(total_bytes);
+#else /* !TRE_USE_ALLOCA */
+    buf = xmalloc((unsigned)total_bytes);
+#endif /* !TRE_USE_ALLOCA */
+    if (!buf)
+      return REG_ESPACE;
+    memset(buf, 0, (size_t)total_bytes);
+
+    /* Allocate `tmp_tags' from `buf'. */
+    tmp_tags = (void *)buf;
+    buf_cursor = buf + tag_bytes;
+    buf_cursor += ALIGN(buf_cursor, long);
+
+    /* Allocate `reach' from `buf'. */
+    reach = (void *)buf_cursor;
+    buf_cursor += reach_bytes;
+    buf_cursor += ALIGN(buf_cursor, long);
+
+    /* Allocate `reach_next' from `buf'. */
+    reach_next = (void *)buf_cursor;
+    buf_cursor += reach_bytes;
+    buf_cursor += ALIGN(buf_cursor, long);
+
+    /* Allocate tag arrays for `reach' and `reach_next' from `buf'. */
+    for (i = 0; i < tnfa->num_states; i++)
+      {
+	reach[i].tags = (void *)buf_cursor;
+	buf_cursor += tag_bytes;
+	reach_next[i].tags = (void *)buf_cursor;
+	buf_cursor += tag_bytes;
+      }
+    assert(buf_cursor <= buf + total_bytes);
+  }
+
+  for (i = 0; i < TRE_M_LAST; i++)
+    match_costs[i] = INT_MAX;
+
+  /* Mark the reach arrays empty. */
+  for (i = 0; i < tnfa->num_states; i++)
+    reach[i].pos = reach_next[i].pos = -2;
+
+  prev_pos = pos;
+  GET_NEXT_WCHAR();
+  pos = 0;
+
+  while (/*CONSTCOND*/(void)1,1)
+    {
+      DPRINT(("%03zd:%2lc/%05d\n", pos, (tre_cint_t)next_c, (int)next_c));
+
+      /* Add initial states to `reach_next' if an exact match has not yet
+	 been found. */
+      if (match_costs[TRE_M_COST] > 0)
+	{
+	  tre_tnfa_transition_t *trans;
+	  DPRINT(("  init"));
+	  for (trans = tnfa->initial; trans->state; trans++)
+	    {
+	      int stateid = trans->state_id;
+
+	      /* If this state is not currently in `reach_next', add it
+		 there. */
+	      if (reach_next[stateid].pos < pos)
+		{
+		  if (trans->assertions && CHECK_ASSERTIONS(trans->assertions))
+		    {
+		      /* Assertions failed, don't add this state. */
+		      DPRINT((" !%d (assert)", stateid));
+		      continue;
+		    }
+		  DPRINT((" %d", stateid));
+		  reach_next[stateid].state = trans->state;
+		  reach_next[stateid].pos = pos;
+
+		  /* Compute tag values after this transition. */
+		  for (i = 0; i < num_tags; i++)
+		    reach_next[stateid].tags[i] = -1;
+
+		  if (trans->tags)
+		    for (i = 0; trans->tags[i] >= 0; i++)
+		      if (trans->tags[i] < num_tags)
+			reach_next[stateid].tags[trans->tags[i]] = pos;
+
+		  /* Set the parameters, depth, and costs. */
+		  reach_next[stateid].params = default_params;
+		  reach_next[stateid].depth = 0;
+		  for (i = 0; i < TRE_M_LAST; i++)
+		    reach_next[stateid].costs[0][i] = 0;
+		  if (trans->params)
+		    tre_set_params(&reach_next[stateid], trans->params,
+				   default_params);
+
+		  /* If this is the final state, mark the exact match. */
+		  if (trans->state == tnfa->final)
+		    {
+		      match_eo = pos;
+		      for (i = 0; i < num_tags; i++)
+			match_tags[i] = reach_next[stateid].tags[i];
+		      for (i = 0; i < TRE_M_LAST; i++)
+			match_costs[i] = 0;
+		    }
+		}
+	    }
+	    DPRINT(("\n"));
+	}
+
+
+      /* Handle inserts.  This is done by pretending there's an epsilon
+	 transition from each state in `reach' back to the same state.
+	 We don't need to worry about the final state here; this will never
+	 give a better match than what we already have. */
+      for (id = 0; id < tnfa->num_states; id++)
+	{
+	  int depth;
+	  int cost, cost0;
+
+	  if (reach[id].pos != prev_pos)
+	    {
+	      DPRINT(("	 insert: %d not reached\n", id));
+	      continue;	 /* Not reached. */
+	    }
+
+	  depth = reach[id].depth;
+
+	  /* Compute and check cost at current depth. */
+	  cost = reach[id].costs[depth][TRE_M_COST];
+	  if (reach[id].params.cost_ins != TRE_PARAM_UNSET)
+	    cost += reach[id].params.cost_ins;
+	  if (cost > reach[id].params.max_cost)
+	    continue;  /* Cost too large. */
+
+	  /* Check number of inserts at current depth. */
+	  if (reach[id].costs[depth][TRE_M_NUM_INS] + 1
+	      > reach[id].params.max_ins)
+	    continue;  /* Too many inserts. */
+
+	  /* Check total number of errors at current depth. */
+	  if (reach[id].costs[depth][TRE_M_NUM_ERR] + 1
+	      > reach[id].params.max_err)
+	    continue;  /* Too many errors. */
+
+	  /* Compute overall cost. */
+	  cost0 = cost;
+	  if (depth > 0)
+	    {
+	      cost0 = reach[id].costs[0][TRE_M_COST];
+	      if (reach[id].params.cost_ins != TRE_PARAM_UNSET)
+		cost0 += reach[id].params.cost_ins;
+	      else
+		cost0 += default_params.cost_ins;
+	    }
+
+	  DPRINT(("  insert: from %d to %d, cost %d: ", id, id,
+		  reach[id].costs[depth][TRE_M_COST]));
+	  if (reach_next[id].pos == pos
+	      && (cost0 >= reach_next[id].costs[0][TRE_M_COST]))
+	    {
+	      DPRINT(("lose\n"));
+	      continue;
+	    }
+	  DPRINT(("win\n"));
+
+	  /* Copy state, position, tags, parameters, and depth. */
+	  reach_next[id].state = reach[id].state;
+	  reach_next[id].pos = pos;
+	  for (i = 0; i < num_tags; i++)
+	    reach_next[id].tags[i] = reach[id].tags[i];
+	  reach_next[id].params = reach[id].params;
+	  reach_next[id].depth = reach[id].depth;
+
+	  /* Set the costs after this transition. */
+	  memcpy(reach_next[id].costs, reach[id].costs,
+		 sizeof(reach_next[id].costs[0][0])
+		 * TRE_M_LAST * (depth + 1));
+	  reach_next[id].costs[depth][TRE_M_COST] = cost;
+	  reach_next[id].costs[depth][TRE_M_NUM_INS]++;
+	  reach_next[id].costs[depth][TRE_M_NUM_ERR]++;
+	  if (depth > 0)
+	    {
+	      reach_next[id].costs[0][TRE_M_COST] = cost0;
+	      reach_next[id].costs[0][TRE_M_NUM_INS]++;
+	      reach_next[id].costs[0][TRE_M_NUM_ERR]++;
+	    }
+
+	}
+
+
+      /* Handle deletes.  This is done by traversing through the whole TNFA
+	 pretending that all transitions are epsilon transitions, until
+	 no more states can be reached with better costs. */
+      {
+	/* XXX - dynamic ringbuffer size */
+	tre_tnfa_approx_reach_t *ringbuffer[512];
+	tre_tnfa_approx_reach_t **deque_start, **deque_end;
+
+	deque_start = deque_end = ringbuffer;
+
+	/* Add all states in `reach_next' to the deque. */
+	for (id = 0; id < tnfa->num_states; id++)
+	  {
+	    if (reach_next[id].pos != pos)
+	      continue;
+	    *deque_end = &reach_next[id];
+	    deque_end++;
+	    assert(deque_end != deque_start);
+	  }
+
+	/* Repeat until the deque is empty. */
+	while (deque_end != deque_start)
+	  {
+	    tre_tnfa_approx_reach_t *reach_p;
+	    int depth;
+	    int cost, cost0;
+	    tre_tnfa_transition_t *trans;
+
+	    /* Pop the first item off the deque. */
+	    reach_p = *deque_start;
+	    id = reach_p - reach_next;
+	    depth = reach_p->depth;
+
+	    /* Compute cost at current depth. */
+	    cost = reach_p->costs[depth][TRE_M_COST];
+	    if (reach_p->params.cost_del != TRE_PARAM_UNSET)
+	      cost += reach_p->params.cost_del;
+
+	    /* Check cost, number of deletes, and total number of errors
+	       at current depth. */
+	    if (cost > reach_p->params.max_cost
+		|| (reach_p->costs[depth][TRE_M_NUM_DEL] + 1
+		    > reach_p->params.max_del)
+		|| (reach_p->costs[depth][TRE_M_NUM_ERR] + 1
+		    > reach_p->params.max_err))
+	      {
+		/* Too many errors or cost too large. */
+		DPRINT(("  delete: from %03d: cost too large\n", id));
+		deque_start++;
+		if (deque_start >= (ringbuffer + 512))
+		  deque_start = ringbuffer;
+		continue;
+	      }
+
+	    /* Compute overall cost. */
+	    cost0 = cost;
+	    if (depth > 0)
+	      {
+		cost0 = reach_p->costs[0][TRE_M_COST];
+		if (reach_p->params.cost_del != TRE_PARAM_UNSET)
+		  cost0 += reach_p->params.cost_del;
+		else
+		  cost0 += default_params.cost_del;
+	      }
+
+	    for (trans = reach_p->state; trans->state; trans++)
+	      {
+		int dest_id = trans->state_id;
+		DPRINT(("  delete: from %03d to %03d, cost %d (%d): ",
+			id, dest_id, cost0, reach_p->params.max_cost));
+
+		if (trans->assertions && CHECK_ASSERTIONS(trans->assertions))
+		  {
+		    DPRINT(("assertion failed\n"));
+		    continue;
+		  }
+
+		/* Compute tag values after this transition. */
+		for (i = 0; i < num_tags; i++)
+		  tmp_tags[i] = reach_p->tags[i];
+		if (trans->tags)
+		  for (i = 0; trans->tags[i] >= 0; i++)
+		    if (trans->tags[i] < num_tags)
+		      tmp_tags[trans->tags[i]] = pos;
+
+		/* If another path has also reached this state, choose the one
+		   with the smallest cost or best tags if costs are equal. */
+		if (reach_next[dest_id].pos == pos
+		    && (cost0 > reach_next[dest_id].costs[0][TRE_M_COST]
+			|| (cost0 == reach_next[dest_id].costs[0][TRE_M_COST]
+			    && (!match_tags
+				|| !tre_tag_order(num_tags,
+						  tnfa->tag_directions,
+						  tmp_tags,
+						  reach_next[dest_id].tags)))))
+		  {
+		    DPRINT(("lose, cost0 %d, have %d\n",
+			    cost0, reach_next[dest_id].costs[0][TRE_M_COST]));
+		    continue;
+		  }
+		DPRINT(("win\n"));
+
+		/* Set state, position, tags, parameters, depth, and costs. */
+		reach_next[dest_id].state = trans->state;
+		reach_next[dest_id].pos = pos;
+		for (i = 0; i < num_tags; i++)
+		  reach_next[dest_id].tags[i] = tmp_tags[i];
+
+		reach_next[dest_id].params = reach_p->params;
+		if (trans->params)
+		  tre_set_params(&reach_next[dest_id], trans->params,
+				 default_params);
+
+		reach_next[dest_id].depth = reach_p->depth;
+		memcpy(&reach_next[dest_id].costs,
+		       reach_p->costs,
+		       sizeof(reach_p->costs[0][0])
+		       * TRE_M_LAST * (depth + 1));
+		reach_next[dest_id].costs[depth][TRE_M_COST] = cost;
+		reach_next[dest_id].costs[depth][TRE_M_NUM_DEL]++;
+		reach_next[dest_id].costs[depth][TRE_M_NUM_ERR]++;
+		if (depth > 0)
+		  {
+		    reach_next[dest_id].costs[0][TRE_M_COST] = cost0;
+		    reach_next[dest_id].costs[0][TRE_M_NUM_DEL]++;
+		    reach_next[dest_id].costs[0][TRE_M_NUM_ERR]++;
+		  }
+
+		if (trans->state == tnfa->final
+		    && (match_eo < 0
+			|| match_costs[TRE_M_COST] > cost0
+			|| (match_costs[TRE_M_COST] == cost0
+			    && (num_tags > 0
+				&& tmp_tags[0] <= match_tags[0]))))
+		  {
+		    DPRINT(("	 setting new match at %zd, cost %d\n",
+			    pos, cost0));
+		    match_eo = pos;
+		    memcpy(match_costs, reach_next[dest_id].costs[0],
+			   sizeof(match_costs[0]) * TRE_M_LAST);
+		    for (i = 0; i < num_tags; i++)
+		      match_tags[i] = tmp_tags[i];
+		  }
+
+		/* Add to the end of the deque. */
+		*deque_end = &reach_next[dest_id];
+		deque_end++;
+		if (deque_end >= (ringbuffer + 512))
+		  deque_end = ringbuffer;
+		assert(deque_end != deque_start);
+	      }
+	    deque_start++;
+	    if (deque_start >= (ringbuffer + 512))
+	      deque_start = ringbuffer;
+	  }
+
+      }
+
+#ifdef TRE_DEBUG
+      tre_print_reach(tnfa, reach_next, pos, num_tags);
+#endif /* TRE_DEBUG */
+
+      /* Check for end of string. */
+      if (len < 0)
+	{
+	  if (type == STR_USER)
+	    {
+	      if (str_user_end)
+		break;
+	    }
+	  else if (next_c == L'\0' || pos >= TRE_MAX_STRING)
+	    break;
+	}
+      else
+	{
+	  if (pos >= len)
+	    break;
+	}
+
+      prev_pos = pos;
+      GET_NEXT_WCHAR();
+
+      /* Swap `reach' and `reach_next'. */
+      {
+	tre_tnfa_approx_reach_t *tmp;
+	tmp = reach;
+	reach = reach_next;
+	reach_next = tmp;
+      }
+
+      /* Handle exact matches and substitutions. */
+      for (id = 0; id < tnfa->num_states; id++)
+	{
+	  tre_tnfa_transition_t *trans;
+
+	  if (reach[id].pos < prev_pos)
+	    continue;  /* Not reached. */
+	  for (trans = reach[id].state; trans->state; trans++)
+	    {
+	      int dest_id;
+	      int depth;
+	      int cost, cost0, err;
+
+	      if (trans->assertions
+		  && (CHECK_ASSERTIONS(trans->assertions)
+		      || CHECK_CHAR_CLASSES(trans, tnfa, eflags)))
+		{
+		  DPRINT(("  exact,  from %d: assert failed\n", id));
+		  continue;
+		}
+
+	      depth = reach[id].depth;
+	      dest_id = trans->state_id;
+
+	      cost = reach[id].costs[depth][TRE_M_COST];
+	      cost0 = reach[id].costs[0][TRE_M_COST];
+	      err = 0;
+
+	      if (trans->code_min > (tre_cint_t)prev_c
+		  || trans->code_max < (tre_cint_t)prev_c)
+		{
+		  /* Handle substitutes.  The required character was not in
+		     the string, so match it in place of whatever was supposed
+		     to be there and increase costs accordingly. */
+		  err = 1;
+
+		  /* Compute and check cost at current depth. */
+		  cost = reach[id].costs[depth][TRE_M_COST];
+		  if (reach[id].params.cost_subst != TRE_PARAM_UNSET)
+		    cost += reach[id].params.cost_subst;
+		  if (cost > reach[id].params.max_cost)
+		    continue; /* Cost too large. */
+
+		  /* Check number of substitutes at current depth. */
+		  if (reach[id].costs[depth][TRE_M_NUM_SUBST] + 1
+		      > reach[id].params.max_subst)
+		    continue; /* Too many substitutes. */
+
+		  /* Check total number of errors at current depth. */
+		  if (reach[id].costs[depth][TRE_M_NUM_ERR] + 1
+		      > reach[id].params.max_err)
+		    continue; /* Too many errors. */
+
+		  /* Compute overall cost. */
+		  cost0 = cost;
+		  if (depth > 0)
+		    {
+		      cost0 = reach[id].costs[0][TRE_M_COST];
+		      if (reach[id].params.cost_subst != TRE_PARAM_UNSET)
+			cost0 += reach[id].params.cost_subst;
+		      else
+			cost0 += default_params.cost_subst;
+		    }
+		  DPRINT(("  subst,  from %03d to %03d, cost %d: ",
+			  id, dest_id, cost0));
+		}
+	      else
+		DPRINT(("  exact,  from %03d to %03d, cost %d: ",
+			id, dest_id, cost0));
+
+	      /* Compute tag values after this transition. */
+	      for (i = 0; i < num_tags; i++)
+		tmp_tags[i] = reach[id].tags[i];
+	      if (trans->tags)
+		for (i = 0; trans->tags[i] >= 0; i++)
+		  if (trans->tags[i] < num_tags)
+		    tmp_tags[trans->tags[i]] = pos;
+
+	      /* If another path has also reached this state, choose the
+		 one with the smallest cost or best tags if costs are equal. */
+	      if (reach_next[dest_id].pos == pos
+		  && (cost0 > reach_next[dest_id].costs[0][TRE_M_COST]
+		      || (cost0 == reach_next[dest_id].costs[0][TRE_M_COST]
+			  && !tre_tag_order(num_tags, tnfa->tag_directions,
+					    tmp_tags,
+					    reach_next[dest_id].tags))))
+		{
+		  DPRINT(("lose\n"));
+		  continue;
+		}
+	      DPRINT(("win %d %d\n",
+		      reach_next[dest_id].pos,
+		      reach_next[dest_id].costs[0][TRE_M_COST]));
+
+	      /* Set state, position, tags, and depth. */
+	      reach_next[dest_id].state = trans->state;
+	      reach_next[dest_id].pos = pos;
+	      for (i = 0; i < num_tags; i++)
+		reach_next[dest_id].tags[i] = tmp_tags[i];
+	      reach_next[dest_id].depth = reach[id].depth;
+
+	      /* Set parameters. */
+	      reach_next[dest_id].params = reach[id].params;
+	      if (trans->params)
+		tre_set_params(&reach_next[dest_id], trans->params,
+			       default_params);
+
+	      /* Set the costs after this transition. */
+	      memcpy(&reach_next[dest_id].costs,
+		     reach[id].costs,
+		     sizeof(reach[id].costs[0][0])
+		     * TRE_M_LAST * (depth + 1));
+	      reach_next[dest_id].costs[depth][TRE_M_COST] = cost;
+	      reach_next[dest_id].costs[depth][TRE_M_NUM_SUBST] += err;
+	      reach_next[dest_id].costs[depth][TRE_M_NUM_ERR] += err;
+	      if (depth > 0)
+		{
+		  reach_next[dest_id].costs[0][TRE_M_COST] = cost0;
+		  reach_next[dest_id].costs[0][TRE_M_NUM_SUBST] += err;
+		  reach_next[dest_id].costs[0][TRE_M_NUM_ERR] += err;
+		}
+
+	      if (trans->state == tnfa->final
+		  && (match_eo < 0
+		      || cost0 < match_costs[TRE_M_COST]
+		      || (cost0 == match_costs[TRE_M_COST]
+			  && num_tags > 0 && tmp_tags[0] <= match_tags[0])))
+		{
+		  DPRINT(("    setting new match at %zd, cost %d\n",
+			  pos, cost0));
+		  match_eo = pos;
+		  for (i = 0; i < TRE_M_LAST; i++)
+		    match_costs[i] = reach_next[dest_id].costs[0][i];
+		  for (i = 0; i < num_tags; i++)
+		    match_tags[i] = tmp_tags[i];
+		}
+	    }
+	}
+    }
+
+  DPRINT(("match end offset = %d, match cost = %d\n", match_eo,
+	  match_costs[TRE_M_COST]));
+
+  match->cost = match_costs[TRE_M_COST];
+  match->num_ins = match_costs[TRE_M_NUM_INS];
+  match->num_del = match_costs[TRE_M_NUM_DEL];
+  match->num_subst = match_costs[TRE_M_NUM_SUBST];
+  *match_end_ofs = match_eo;
+
+  ret = match_eo >= 0 ? REG_OK : REG_NOMATCH;
+
+#ifndef TRE_USE_ALLOCA
+  if (buf)
+    xfree(buf);
+#endif /* !TRE_USE_ALLOCA */
+  return ret;
+}

vendored/tre/tre-match-backtrack.c

@@ -0,0 +1,676 @@
+/*
+  tre-match-backtrack.c - TRE backtracking regex matching engine
+
+  This software is released under a BSD-style license.
+  See the file LICENSE for details and copyright.
+
+*/
+
+/*
+  This matcher is for regexps that use back referencing.  Regexp matching
+  with back referencing is an NP-complete problem on the number of back
+  references.  The easiest way to match them is to use a backtracking
+  routine which basically goes through all possible paths in the TNFA
+  and chooses the one which results in the best (leftmost and longest)
+  match.  This can be spectacularly expensive and may run out of stack
+  space, but there really is no better known generic algorithm.	 Quoting
+  Henry Spencer from comp.compilers:
+  <URL: http://compilers.iecc.com/comparch/article/93-03-102>
+
+    POSIX.2 REs require longest match, which is really exciting to
+    implement since the obsolete ("basic") variant also includes
+    \<digit>.  I haven't found a better way of tackling this than doing
+    a preliminary match using a DFA (or simulation) on a modified RE
+    that just replicates subREs for \<digit>, and then doing a
+    backtracking match to determine whether the subRE matches were
+    right.  This can be rather slow, but I console myself with the
+    thought that people who use \<digit> deserve very slow execution.
+    (Pun unintentional but very appropriate.)
+
+*/
+
+
+#ifdef HAVE_CONFIG_H
+#include <config.h>
+#endif /* HAVE_CONFIG_H */
+
+#ifdef TRE_USE_ALLOCA
+/* AIX requires this to be the first thing in the file.	 */
+#ifndef __GNUC__
+# if HAVE_ALLOCA_H
+#  include <alloca.h>
+# else
+#  ifdef _AIX
+ #pragma alloca
+#  else
+#   ifndef alloca /* predefined by HP cc +Olibcalls */
+char *alloca ();
+#   endif
+#  endif
+# endif
+#endif
+#endif /* TRE_USE_ALLOCA */
+
+#include <assert.h>
+#include <stdlib.h>
+#include <string.h>
+#ifdef HAVE_WCHAR_H
+#include <wchar.h>
+#endif /* HAVE_WCHAR_H */
+#ifdef HAVE_WCTYPE_H
+#include <wctype.h>
+#endif /* HAVE_WCTYPE_H */
+#ifndef TRE_WCHAR
+#include <ctype.h>
+#endif /* !TRE_WCHAR */
+#ifdef HAVE_MALLOC_H
+#include <malloc.h>
+#endif /* HAVE_MALLOC_H */
+
+#include "tre-internal.h"
+#include "tre-mem.h"
+#include "tre-match-utils.h"
+#include "xmalloc.h"
+
+typedef struct {
+  int pos;
+  const char *str_byte;
+#ifdef TRE_WCHAR
+  const wchar_t *str_wide;
+#endif /* TRE_WCHAR */
+  tre_tnfa_transition_t *state;
+  int state_id;
+  int next_c;
+  int *tags;
+#ifdef TRE_MBSTATE
+  mbstate_t mbstate;
+#endif /* TRE_MBSTATE */
+} tre_backtrack_item_t;
+
+typedef struct tre_backtrack_struct {
+  tre_backtrack_item_t item;
+  struct tre_backtrack_struct *prev;
+  struct tre_backtrack_struct *next;
+} *tre_backtrack_t;
+
+#ifdef TRE_WCHAR
+#define BT_STACK_WIDE_IN(_str_wide)	stack->item.str_wide = (_str_wide)
+#define BT_STACK_WIDE_OUT		(str_wide) = stack->item.str_wide
+#else /* !TRE_WCHAR */
+#define BT_STACK_WIDE_IN(_str_wide)
+#define BT_STACK_WIDE_OUT
+#endif /* !TRE_WCHAR */
+
+#ifdef TRE_MBSTATE
+#define BT_STACK_MBSTATE_IN  stack->item.mbstate = (mbstate)
+#define BT_STACK_MBSTATE_OUT (mbstate) = stack->item.mbstate
+#else /* !TRE_MBSTATE */
+#define BT_STACK_MBSTATE_IN
+#define BT_STACK_MBSTATE_OUT
+#endif /* !TRE_MBSTATE */
+
+
+#ifdef TRE_USE_ALLOCA
+#define tre_bt_mem_new		  tre_mem_newa
+#define tre_bt_mem_alloc	  tre_mem_alloca
+#define tre_bt_mem_destroy(obj)	  do { } while (0)
+#define xafree(obj)		  do { } while (0) /* do nothing, obj was obtained with alloca() */
+#else /* !TRE_USE_ALLOCA */
+#define tre_bt_mem_new		  tre_mem_new
+#define tre_bt_mem_alloc	  tre_mem_alloc
+#define tre_bt_mem_destroy	  tre_mem_destroy
+#define xafree(obj)		  xfree(obj)
+#endif /* !TRE_USE_ALLOCA */
+
+
+#define BT_STACK_PUSH(_pos, _str_byte, _str_wide, _state, _state_id, _next_c, _tags, _mbstate) \
+  do									      \
+    {									      \
+      int i;								      \
+      if (!stack->next)							      \
+	{								      \
+	  tre_backtrack_t s;						      \
+	  s = tre_bt_mem_alloc(mem, sizeof(*s));			      \
+	  if (!s)							      \
+	    {								      \
+	      tre_bt_mem_destroy(mem);					      \
+	      if (tags)							      \
+		xafree(tags);						      \
+	      if (pmatch)						      \
+		xafree(pmatch);						      \
+	      if (states_seen)						      \
+		xafree(states_seen);					      \
+	      return REG_ESPACE;					      \
+	    }								      \
+	  s->prev = stack;						      \
+	  s->next = NULL;						      \
+	  s->item.tags = tre_bt_mem_alloc(mem,				      \
+					  sizeof(*tags) * tnfa->num_tags);    \
+	  if (!s->item.tags)						      \
+	    {								      \
+	      tre_bt_mem_destroy(mem);					      \
+	      if (tags)							      \
+		xafree(tags);						      \
+	      if (pmatch)						      \
+		xafree(pmatch);						      \
+	      if (states_seen)						      \
+		xafree(states_seen);					      \
+	      return REG_ESPACE;					      \
+	    }								      \
+	  stack->next = s;						      \
+	  stack = s;							      \
+	}								      \
+      else								      \
+	stack = stack->next;						      \
+      stack->item.pos = (_pos);						      \
+      stack->item.str_byte = (_str_byte);				      \
+      BT_STACK_WIDE_IN(_str_wide);					      \
+      stack->item.state = (_state);					      \
+      stack->item.state_id = (_state_id);				      \
+      stack->item.next_c = (_next_c);					      \
+      for (i = 0; i < tnfa->num_tags; i++)				      \
+	stack->item.tags[i] = (_tags)[i];				      \
+      BT_STACK_MBSTATE_IN;						      \
+    }									      \
+  while (/*CONSTCOND*/(void)0,0)
+
+#define BT_STACK_POP()							      \
+  do									      \
+    {									      \
+      int i;								      \
+      assert(stack->prev);						      \
+      pos = stack->item.pos;						      \
+      if (type == STR_USER)                                                   \
+        str_source->rewind(pos + pos_add_next, str_source->context);          \
+      str_byte = stack->item.str_byte;					      \
+      BT_STACK_WIDE_OUT;						      \
+      state = stack->item.state;					      \
+      next_c = (tre_char_t) stack->item.next_c;					      \
+      for (i = 0; i < tnfa->num_tags; i++)				      \
+	tags[i] = stack->item.tags[i];					      \
+      BT_STACK_MBSTATE_OUT;						      \
+      stack = stack->prev;						      \
+    }									      \
+  while (/*CONSTCOND*/(void)0,0)
+
+#undef MIN
+#define MIN(a, b) ((a) <= (b) ? (a) : (b))
+
+reg_errcode_t
+tre_tnfa_run_backtrack(const tre_tnfa_t *tnfa, const void *string,
+		       ssize_t len, tre_str_type_t type, int *match_tags,
+		       int eflags, int *match_end_ofs)
+{
+  /* State variables required by GET_NEXT_WCHAR. */
+  tre_char_t prev_c = 0, next_c = 0;
+  const char *str_byte = string;
+  ssize_t pos = 0;
+  unsigned int pos_add_next = 1;
+#ifdef TRE_WCHAR
+  const wchar_t *str_wide = string;
+#ifdef TRE_MBSTATE
+  mbstate_t mbstate;
+#endif /* TRE_MBSTATE */
+#endif /* TRE_WCHAR */
+  int reg_notbol = eflags & REG_NOTBOL;
+  int reg_noteol = eflags & REG_NOTEOL;
+  int reg_newline = tnfa->cflags & REG_NEWLINE;
+  int str_user_end = 0;
+
+  /* These are used to remember the necessary values of the above
+     variables to return to the position where the current search
+     started from. */
+  int next_c_start;
+  const char *str_byte_start;
+  int pos_start = -1;
+#ifdef TRE_WCHAR
+  const wchar_t *str_wide_start;
+#endif /* TRE_WCHAR */
+#ifdef TRE_MBSTATE
+  mbstate_t mbstate_start;
+#endif /* TRE_MBSTATE */
+  reg_errcode_t ret;
+
+  /* End offset of best match so far, or -1 if no match found yet. */
+  int match_eo = -1;
+  /* Tag arrays. */
+  int *next_tags, *tags = NULL;
+  /* Current TNFA state. */
+  tre_tnfa_transition_t *state;
+  int *states_seen = NULL;
+
+  /* Memory allocator to for allocating the backtracking stack. */
+  tre_mem_t mem = tre_bt_mem_new();
+
+  /* The backtracking stack. */
+  tre_backtrack_t stack;
+
+  tre_tnfa_transition_t *trans_i;
+  regmatch_t *pmatch = NULL;
+
+  /*
+   * TRE internals tend to use int instead of size_t for positions or
+   * lengths and don't check for overflow.  This will take time to fix
+   * properly.  In the meantime, simply limit the input to what we can
+   * handle.
+   */
+  if (len > TRE_MAX_STRING)
+    len = TRE_MAX_STRING;
+
+#ifdef TRE_MBSTATE
+  memset(&mbstate, '\0', sizeof(mbstate));
+#endif /* TRE_MBSTATE */
+
+  if (!mem)
+    return REG_ESPACE;
+  stack = tre_bt_mem_alloc(mem, sizeof(*stack));
+  if (!stack)
+    {
+      ret = REG_ESPACE;
+      goto error_exit;
+    }
+  stack->prev = NULL;
+  stack->next = NULL;
+
+  DPRINT(("tnfa_execute_backtrack, input type %d\n", type));
+  DPRINT(("len = %zd\n", len));
+
+#ifdef TRE_USE_ALLOCA
+  tags = alloca(sizeof(*tags) * tnfa->num_tags);
+  pmatch = alloca(sizeof(*pmatch) * tnfa->num_submatches);
+  states_seen = alloca(sizeof(*states_seen) * tnfa->num_states);
+#else /* !TRE_USE_ALLOCA */
+  if (tnfa->num_tags)
+    {
+      tags = xmalloc(sizeof(*tags) * tnfa->num_tags);
+      if (!tags)
+	{
+	  ret = REG_ESPACE;
+	  goto error_exit;
+	}
+    }
+  if (tnfa->num_submatches)
+    {
+      pmatch = xmalloc(sizeof(*pmatch) * tnfa->num_submatches);
+      if (!pmatch)
+	{
+	  ret = REG_ESPACE;
+	  goto error_exit;
+	}
+    }
+  if (tnfa->num_states)
+    {
+      states_seen = xmalloc(sizeof(*states_seen) * tnfa->num_states);
+      if (!states_seen)
+	{
+	  ret = REG_ESPACE;
+	  goto error_exit;
+	}
+    }
+#endif /* !TRE_USE_ALLOCA */
+
+ retry:
+  {
+    int i;
+    for (i = 0; i < tnfa->num_tags; i++)
+      {
+	tags[i] = -1;
+	if (match_tags)
+	  match_tags[i] = -1;
+      }
+    for (i = 0; i < tnfa->num_states; i++)
+      states_seen[i] = 0;
+  }
+
+  state = NULL;
+  pos = pos_start;
+  if (type == STR_USER)
+    str_source->rewind(pos + pos_add_next, str_source->context);
+  GET_NEXT_WCHAR();
+  pos_start = pos;
+  next_c_start = next_c;
+  str_byte_start = str_byte;
+#ifdef TRE_WCHAR
+  str_wide_start = str_wide;
+#endif /* TRE_WCHAR */
+#ifdef TRE_MBSTATE
+  mbstate_start = mbstate;
+#endif /* TRE_MBSTATE */
+
+  /* Handle initial states. */
+  next_tags = NULL;
+  for (trans_i = tnfa->initial; trans_i->state; trans_i++)
+    {
+      DPRINT(("> init %p, prev_c %lc\n", trans_i->state, (tre_cint_t)prev_c));
+      if (trans_i->assertions && CHECK_ASSERTIONS(trans_i->assertions))
+	{
+	  DPRINT(("assert failed\n"));
+	  continue;
+	}
+      if (state == NULL)
+	{
+	  /* Start from this state. */
+	  state = trans_i->state;
+	  next_tags = trans_i->tags;
+	}
+      else
+	{
+	  /* Backtrack to this state. */
+	  DPRINT(("saving state %d for backtracking\n", trans_i->state_id));
+	  BT_STACK_PUSH(pos, str_byte, str_wide, trans_i->state,
+			trans_i->state_id, next_c, tags, mbstate);
+	  {
+	    int *tmp = trans_i->tags;
+	    if (tmp)
+	      while (*tmp >= 0)
+		stack->item.tags[*tmp++] = pos;
+	  }
+	}
+    }
+
+  if (next_tags)
+    for (; *next_tags >= 0; next_tags++)
+      tags[*next_tags] = pos;
+
+
+  DPRINT(("entering match loop, pos %zd, str_byte %p\n", pos, str_byte));
+  DPRINT(("pos:chr/code | state and tags\n"));
+  DPRINT(("-------------+------------------------------------------------\n"));
+
+  if (state == NULL)
+    goto backtrack;
+
+  while (/*CONSTCOND*/(void)1,1)
+    {
+      tre_tnfa_transition_t *next_state;
+      int empty_br_match;
+
+      DPRINT(("start loop\n"));
+      if (state == tnfa->final)
+	{
+	  DPRINT(("  match found, %d %zd\n", match_eo, pos));
+	  if (match_eo < pos
+	      || (match_eo == pos
+		  && match_tags
+		  && tre_tag_order(tnfa->num_tags, tnfa->tag_directions,
+				   tags, match_tags)))
+	    {
+	      int i;
+	      /* This match wins the previous match. */
+	      DPRINT(("	 win previous\n"));
+	      match_eo = pos;
+	      if (match_tags)
+		for (i = 0; i < tnfa->num_tags; i++)
+		  match_tags[i] = tags[i];
+	    }
+	  /* Our TNFAs never have transitions leaving from the final state,
+	     so we jump right to backtracking. */
+	  goto backtrack;
+	}
+
+#ifdef TRE_DEBUG
+      DPRINT(("%3zd:%2lc/%05d | %p ", pos, (tre_cint_t)next_c, (int)next_c,
+	      state));
+      {
+	int i;
+	for (i = 0; i < tnfa->num_tags; i++)
+	  DPRINT(("%d%s", tags[i], i < tnfa->num_tags - 1 ? ", " : ""));
+	DPRINT(("\n"));
+      }
+#endif /* TRE_DEBUG */
+
+      /* Go to the next character in the input string. */
+      empty_br_match = 0;
+      trans_i = state;
+      if (trans_i->state && trans_i->assertions & ASSERT_BACKREF)
+	{
+	  /* This is a back reference state.  All transitions leaving from
+	     this state have the same back reference "assertion".  Instead
+	     of reading the next character, we match the back reference. */
+	  int so, eo, bt = trans_i->u.backref;
+	  int bt_len;
+	  int result;
+
+	  DPRINT(("  should match back reference %d\n", bt));
+	  /* Get the substring we need to match against.  Remember to
+	     turn off REG_NOSUB temporarily. */
+	  tre_fill_pmatch(bt + 1, pmatch, tnfa->cflags & ~REG_NOSUB,
+			  tnfa, tags, pos);
+	  so = pmatch[bt].rm_so;
+	  eo = pmatch[bt].rm_eo;
+	  bt_len = eo - so;
+
+#ifdef TRE_DEBUG
+	  {
+	    int slen;
+	    if (len < 0)
+	      slen = bt_len;
+	    else
+	      slen = MIN(bt_len, len - pos);
+
+	    if (type == STR_BYTE)
+	      {
+		DPRINT(("  substring (len %d) is [%d, %d[: '%.*s'\n",
+			bt_len, so, eo, bt_len, (char*)string + so));
+		DPRINT(("  current string is '%.*s'\n", slen, str_byte - 1));
+	      }
+#ifdef TRE_WCHAR
+	    else if (type == STR_WIDE)
+	      {
+		DPRINT(("  substring (len %d) is [%d, %d[: '%.*" STRF "'\n",
+			bt_len, so, eo, bt_len, (wchar_t*)string + so));
+		DPRINT(("  current string is '%.*" STRF "'\n",
+			slen, str_wide - 1));
+	      }
+#endif /* TRE_WCHAR */
+	  }
+#endif
+
+	  if (len < 0)
+	    {
+	      if (type == STR_USER)
+		result = str_source->compare((unsigned)so, (unsigned)pos,
+					     (unsigned)bt_len,
+					     str_source->context);
+#ifdef TRE_WCHAR
+	      else if (type == STR_WIDE)
+		result = wcsncmp((const wchar_t*)string + so, str_wide - 1,
+				 (size_t)bt_len);
+#endif /* TRE_WCHAR */
+	      else
+		result = strncmp((const char*)string + so, str_byte - 1,
+				 (size_t)bt_len);
+	    }
+	  else if (len - pos < bt_len)
+	    result = 1;
+#ifdef TRE_WCHAR
+	  else if (type == STR_WIDE)
+	    result = wmemcmp((const wchar_t*)string + so, str_wide - 1,
+			     (size_t)bt_len);
+#endif /* TRE_WCHAR */
+	  else
+	    result = memcmp((const char*)string + so, str_byte - 1,
+			    (size_t)bt_len);
+
+	  if (result == 0)
+	    {
+	      /* Back reference matched.  Check for infinite loop. */
+	      if (bt_len == 0)
+		empty_br_match = 1;
+	      if (empty_br_match && states_seen[trans_i->state_id])
+		{
+		  DPRINT(("  avoid loop\n"));
+		  goto backtrack;
+		}
+
+	      states_seen[trans_i->state_id] = empty_br_match;
+
+	      /* Advance in input string and resync `prev_c', `next_c'
+		 and pos. */
+	      DPRINT(("	 back reference matched\n"));
+	      str_byte += bt_len - 1;
+#ifdef TRE_WCHAR
+	      str_wide += bt_len - 1;
+#endif /* TRE_WCHAR */
+	      pos += bt_len - 1;
+	      GET_NEXT_WCHAR();
+	      DPRINT(("	 pos now %zd\n", pos));
+	    }
+	  else
+	    {
+	      DPRINT(("	 back reference did not match\n"));
+	      goto backtrack;
+	    }
+	}
+      else
+	{
+	  /* Check for end of string. */
+	  if (len < 0)
+	    {
+	      if (type == STR_USER)
+		{
+		  if (str_user_end)
+		    goto backtrack;
+		}
+	      else if (next_c == L'\0' || pos >= TRE_MAX_STRING)
+		goto backtrack;
+	    }
+	  else
+	    {
+	      if (pos >= len)
+		goto backtrack;
+	    }
+
+	  /* Read the next character. */
+	  GET_NEXT_WCHAR();
+	}
+
+      next_state = NULL;
+      for (trans_i = state; trans_i->state; trans_i++)
+	{
+	  DPRINT(("  transition %d-%d (%c-%c) %d to %d\n",
+		  trans_i->code_min, trans_i->code_max,
+		  trans_i->code_min, trans_i->code_max,
+		  trans_i->assertions, trans_i->state_id));
+	  if (trans_i->code_min <= (tre_cint_t)prev_c
+	      && trans_i->code_max >= (tre_cint_t)prev_c)
+	    {
+	      if (trans_i->assertions
+		  && (CHECK_ASSERTIONS(trans_i->assertions)
+		      || CHECK_CHAR_CLASSES(trans_i, tnfa, eflags)))
+		{
+		  DPRINT(("  assertion failed\n"));
+		  continue;
+		}
+
+	      if (next_state == NULL)
+		{
+		  /* First matching transition. */
+		  DPRINT(("  Next state is %d\n", trans_i->state_id));
+		  next_state = trans_i->state;
+		  next_tags = trans_i->tags;
+		}
+	      else
+		{
+		  /* Second matching transition.  We may need to backtrack here
+		     to take this transition instead of the first one, so we
+		     push this transition in the backtracking stack so we can
+		     jump back here if needed. */
+		  DPRINT(("  saving state %d for backtracking\n",
+			  trans_i->state_id));
+		  BT_STACK_PUSH(pos, str_byte, str_wide, trans_i->state,
+				trans_i->state_id, next_c, tags, mbstate);
+		  {
+		    int *tmp;
+		    for (tmp = trans_i->tags; tmp && *tmp >= 0; tmp++)
+		      stack->item.tags[*tmp] = pos;
+		  }
+#if 0 /* XXX - it's important not to look at all transitions here to keep
+	 the stack small! */
+		  break;
+#endif
+		}
+	    }
+	}
+
+      if (next_state != NULL)
+	{
+	  /* Matching transitions were found.  Take the first one. */
+	  state = next_state;
+
+	  /* Update the tag values. */
+	  if (next_tags)
+	    while (*next_tags >= 0)
+	      tags[*next_tags++] = pos;
+	}
+      else
+	{
+	backtrack:
+	  /* A matching transition was not found.  Try to backtrack. */
+	  if (stack->prev)
+	    {
+	      DPRINT(("	 backtracking\n"));
+	      if (stack->item.state->assertions & ASSERT_BACKREF)
+		{
+		  DPRINT(("  states_seen[%d] = 0\n",
+			  stack->item.state_id));
+		  states_seen[stack->item.state_id] = 0;
+		}
+
+	      BT_STACK_POP();
+	    }
+	  else if (match_eo < 0)
+	    {
+	      /* Try starting from a later position in the input string. */
+	      /* Check for end of string. */
+	      if (len < 0)
+		{
+		  if (next_c_start == L'\0' || pos_start >= TRE_MAX_STRING)
+		    {
+		      DPRINT(("end of string.\n"));
+		      break;
+		    }
+		}
+	      else
+		{
+		  if (pos_start >= len)
+		    {
+		      DPRINT(("end of string.\n"));
+		      break;
+		    }
+		}
+	      DPRINT(("restarting from next start position\n"));
+	      next_c = (tre_char_t) next_c_start;
+#ifdef TRE_MBSTATE
+	      mbstate = mbstate_start;
+#endif /* TRE_MBSTATE */
+	      str_byte = str_byte_start;
+#ifdef TRE_WCHAR
+	      str_wide = str_wide_start;
+#endif /* TRE_WCHAR */
+	      goto retry;
+	    }
+	  else
+	    {
+	      DPRINT(("finished\n"));
+	      break;
+	    }
+	}
+    }
+
+  ret = match_eo >= 0 ? REG_OK : REG_NOMATCH;
+  *match_end_ofs = match_eo;
+
+ error_exit:
+  tre_bt_mem_destroy(mem);
+#ifndef TRE_USE_ALLOCA
+  if (tags)
+    xafree(tags);
+  if (pmatch)
+    xafree(pmatch);
+  if (states_seen)
+    xafree(states_seen);
+#endif /* !TRE_USE_ALLOCA */
+
+  return ret;
+}

vendored/tre/tre-match-parallel.c

@@ -0,0 +1,511 @@
+/*
+  tre-match-parallel.c - TRE parallel regex matching engine
+
+  This software is released under a BSD-style license.
+  See the file LICENSE for details and copyright.
+
+*/
+
+/*
+  This algorithm searches for matches basically by reading characters
+  in the searched string one by one, starting at the beginning.	 All
+  matching paths in the TNFA are traversed in parallel.	 When two or
+  more paths reach the same state, exactly one is chosen according to
+  tag ordering rules; if returning submatches is not required it does
+  not matter which path is chosen.
+
+  The worst case time required for finding the leftmost and longest
+  match, or determining that there is no match, is always linearly
+  dependent on the length of the text being searched.
+
+  This algorithm cannot handle TNFAs with back referencing nodes.
+  See `tre-match-backtrack.c'.
+*/
+
+
+#ifdef HAVE_CONFIG_H
+#include <config.h>
+#endif /* HAVE_CONFIG_H */
+
+#ifdef TRE_USE_ALLOCA
+/* AIX requires this to be the first thing in the file.	 */
+#ifndef __GNUC__
+# if HAVE_ALLOCA_H
+#  include <alloca.h>
+# else
+#  ifdef _AIX
+ #pragma alloca
+#  else
+#   ifndef alloca /* predefined by HP cc +Olibcalls */
+char *alloca ();
+#   endif
+#  endif
+# endif
+#endif
+#endif /* TRE_USE_ALLOCA */
+
+#include <assert.h>
+#include <stdlib.h>
+#include <string.h>
+#ifdef HAVE_WCHAR_H
+#include <wchar.h>
+#endif /* HAVE_WCHAR_H */
+#ifdef HAVE_WCTYPE_H
+#include <wctype.h>
+#endif /* HAVE_WCTYPE_H */
+#ifndef TRE_WCHAR
+#include <ctype.h>
+#endif /* !TRE_WCHAR */
+#ifdef HAVE_MALLOC_H
+#include <malloc.h>
+#endif /* HAVE_MALLOC_H */
+
+#include "tre-internal.h"
+#include "tre-match-utils.h"
+#include "xmalloc.h"
+
+
+
+typedef struct {
+  tre_tnfa_transition_t *state;
+  int *tags;
+} tre_tnfa_reach_t;
+
+typedef struct {
+  int pos;
+  int **tags;
+} tre_reach_pos_t;
+
+
+#ifdef TRE_DEBUG
+static void
+tre_print_reach(const tre_tnfa_reach_t *reach, int num_tags)
+{
+  int i;
+
+  while (reach->state != NULL)
+    {
+      DPRINT((" %p", (void *)reach->state));
+      if (num_tags > 0)
+	{
+	  DPRINT(("/"));
+	  for (i = 0; i < num_tags; i++)
+	    {
+	      DPRINT(("%d:%d", i, reach->tags[i]));
+	      if (i < (num_tags-1))
+		DPRINT((","));
+	    }
+	}
+      reach++;
+    }
+  DPRINT(("\n"));
+
+}
+#endif /* TRE_DEBUG */
+
+reg_errcode_t
+tre_tnfa_run_parallel(const tre_tnfa_t *tnfa, const void *string, ssize_t len,
+		      tre_str_type_t type, int *match_tags, int eflags,
+		      int *match_end_ofs)
+{
+  /* State variables required by GET_NEXT_WCHAR. */
+  tre_char_t prev_c = 0, next_c = 0;
+  const char *str_byte = string;
+  ssize_t pos = -1;
+  unsigned int pos_add_next = 1;
+#ifdef TRE_WCHAR
+  const wchar_t *str_wide = string;
+#ifdef TRE_MBSTATE
+  mbstate_t mbstate;
+#endif /* TRE_MBSTATE */
+#endif /* TRE_WCHAR */
+  reg_errcode_t ret;
+  int reg_notbol = eflags & REG_NOTBOL;
+  int reg_noteol = eflags & REG_NOTEOL;
+  int reg_newline = tnfa->cflags & REG_NEWLINE;
+  int str_user_end = 0;
+
+  char *buf;
+  tre_tnfa_transition_t *trans_i;
+  tre_tnfa_reach_t *reach, *reach_next, *reach_i, *reach_next_i;
+  tre_reach_pos_t *reach_pos;
+  int *tag_i;
+  int num_tags, i;
+
+  int match_eo = -1;	   /* end offset of match (-1 if no match found yet) */
+  int new_match = 0;
+  int *tmp_tags = NULL;
+  int *tmp_iptr;
+
+  /*
+   * TRE internals tend to use int instead of size_t for positions or
+   * lengths and don't check for overflow.  This will take time to fix
+   * properly.  In the meantime, simply limit the input to what we can
+   * handle.
+   */
+  if (len > TRE_MAX_STRING)
+    len = TRE_MAX_STRING;
+
+#ifdef TRE_MBSTATE
+  memset(&mbstate, '\0', sizeof(mbstate));
+#endif /* TRE_MBSTATE */
+
+  DPRINT(("tre_tnfa_run_parallel, input type %d\n", type));
+
+  if (!match_tags)
+    num_tags = 0;
+  else
+    num_tags = tnfa->num_tags;
+
+  /* Allocate memory for temporary data required for matching.	This needs to
+     be done for every matching operation to be thread safe.  This allocates
+     everything in a single large block from the stack frame using alloca()
+     or with malloc() if alloca is unavailable. */
+  {
+    size_t tbytes, rbytes, pbytes, xbytes, total_bytes;
+    char *tmp_buf;
+    /* Compute the length of the block we need. */
+    tbytes = sizeof(*tmp_tags) * num_tags;
+    rbytes = sizeof(*reach_next) * (tnfa->num_states + 1);
+    pbytes = sizeof(*reach_pos) * tnfa->num_states;
+    xbytes = sizeof(int) * num_tags;
+    total_bytes =
+      (sizeof(long) - 1) * 4 /* for alignment paddings */
+      + (rbytes + xbytes * tnfa->num_states) * 2 + tbytes + pbytes;
+
+    /* Allocate the memory. */
+#ifdef TRE_USE_ALLOCA
+    buf = alloca(total_bytes);
+#else /* !TRE_USE_ALLOCA */
+    buf = xmalloc(total_bytes);
+#endif /* !TRE_USE_ALLOCA */
+    if (buf == NULL)
+      return REG_ESPACE;
+    memset(buf, 0, total_bytes);
+
+    /* Get the various pointers within tmp_buf (properly aligned). */
+    tmp_tags = (void *)buf;
+    tmp_buf = buf + tbytes;
+    tmp_buf += ALIGN(tmp_buf, long);
+    reach_next = (void *)tmp_buf;
+    tmp_buf += rbytes;
+    tmp_buf += ALIGN(tmp_buf, long);
+    reach = (void *)tmp_buf;
+    tmp_buf += rbytes;
+    tmp_buf += ALIGN(tmp_buf, long);
+    reach_pos = (void *)tmp_buf;
+    tmp_buf += pbytes;
+    tmp_buf += ALIGN(tmp_buf, long);
+    for (i = 0; i < tnfa->num_states; i++)
+      {
+	reach[i].tags = (void *)tmp_buf;
+	tmp_buf += xbytes;
+	reach_next[i].tags = (void *)tmp_buf;
+	tmp_buf += xbytes;
+      }
+  }
+
+  for (i = 0; i < tnfa->num_states; i++)
+    reach_pos[i].pos = -1;
+
+  /* If only one character can start a match, find it first. */
+  if (tnfa->first_char >= 0 && type == STR_BYTE && str_byte)
+    {
+      const char *orig_str = str_byte;
+      int first = tnfa->first_char;
+
+      if (len >= 0)
+	str_byte = memchr(orig_str, first, (size_t)len);
+      else
+	str_byte = strchr(orig_str, first);
+      if (str_byte == NULL)
+	{
+#ifndef TRE_USE_ALLOCA
+	  if (buf)
+	    xfree(buf);
+#endif /* !TRE_USE_ALLOCA */
+	  return REG_NOMATCH;
+	}
+      DPRINT(("skipped %lu chars\n", (unsigned long)(str_byte - orig_str)));
+      if (str_byte >= orig_str + 1)
+	prev_c = (unsigned char)*(str_byte - 1);
+      next_c = (unsigned char)*str_byte;
+      pos = str_byte - orig_str;
+      if (len < 0 || pos < len)
+	str_byte++;
+    }
+  else
+    {
+      GET_NEXT_WCHAR();
+      pos = 0;
+    }
+
+#if 0
+  /* Skip over characters that cannot possibly be the first character
+     of a match. */
+  if (tnfa->firstpos_chars != NULL)
+    {
+      char *chars = tnfa->firstpos_chars;
+
+      if (len < 0)
+	{
+	  const char *orig_str = str_byte;
+	  /* XXX - use strpbrk() and wcspbrk() because they might be
+	     optimized for the target architecture.  Try also strcspn()
+	     and wcscspn() and compare the speeds. */
+	  while (next_c != L'\0' && !chars[next_c])
+	    {
+	      next_c = *str_byte++;
+	    }
+	  prev_c = *(str_byte - 2);
+	  pos += str_byte - orig_str;
+	  DPRINT(("skipped %d chars\n", str_byte - orig_str));
+	}
+      else
+	{
+	  while (pos <= len && !chars[next_c])
+	    {
+	      prev_c = next_c;
+	      next_c = (unsigned char)(*str_byte++);
+	      pos++;
+	    }
+	}
+    }
+#endif
+
+  DPRINT(("length: %zd\n", len));
+  DPRINT(("pos:chr/code | states and tags\n"));
+  DPRINT(("-------------+------------------------------------------------\n"));
+
+  reach_next_i = reach_next;
+  while (/*CONSTCOND*/(void)1,1)
+    {
+      /* If no match found yet, add the initial states to `reach_next'. */
+      if (match_eo < 0)
+	{
+	  DPRINT((" init >"));
+	  trans_i = tnfa->initial;
+	  while (trans_i->state != NULL)
+	    {
+	      if (reach_pos[trans_i->state_id].pos < pos)
+		{
+		  if (trans_i->assertions
+		      && CHECK_ASSERTIONS(trans_i->assertions))
+		    {
+		      DPRINT(("assertion failed\n"));
+		      trans_i++;
+		      continue;
+		    }
+
+		  DPRINT((" %p", (void *)trans_i->state));
+		  reach_next_i->state = trans_i->state;
+		  for (i = 0; i < num_tags; i++)
+		    reach_next_i->tags[i] = -1;
+		  tag_i = trans_i->tags;
+		  if (tag_i)
+		    while (*tag_i >= 0)
+		      {
+			if (*tag_i < num_tags)
+			  reach_next_i->tags[*tag_i] = pos;
+			tag_i++;
+		      }
+		  if (reach_next_i->state == tnfa->final)
+		    {
+		      DPRINT(("	 found empty match\n"));
+		      match_eo = pos;
+		      new_match = 1;
+		      for (i = 0; i < num_tags; i++)
+			match_tags[i] = reach_next_i->tags[i];
+		    }
+		  reach_pos[trans_i->state_id].pos = pos;
+		  reach_pos[trans_i->state_id].tags = &reach_next_i->tags;
+		  reach_next_i++;
+		}
+	      trans_i++;
+	    }
+	  DPRINT(("\n"));
+	  reach_next_i->state = NULL;
+	}
+      else
+	{
+	  if (num_tags == 0 || reach_next_i == reach_next)
+	    /* We have found a match. */
+	    break;
+	}
+
+      /* Check for end of string. */
+      if (len < 0)
+	{
+	  if (type == STR_USER)
+	    {
+	      if (str_user_end)
+		break;
+	    }
+	  else if (next_c == L'\0' || pos >= TRE_MAX_STRING)
+	    break;
+	}
+      else
+	{
+	  if (pos >= len)
+	    break;
+	}
+
+      GET_NEXT_WCHAR();
+
+#ifdef TRE_DEBUG
+      DPRINT(("%3zd:%2lc/%05d |", pos - 1, (tre_cint_t)prev_c, (int)prev_c));
+      tre_print_reach(reach_next, num_tags);
+      DPRINT(("%3zd:%2lc/%05d |", pos, (tre_cint_t)next_c, (int)next_c));
+      tre_print_reach(reach_next, num_tags);
+#endif /* TRE_DEBUG */
+
+      /* Swap `reach' and `reach_next'. */
+      reach_i = reach;
+      reach = reach_next;
+      reach_next = reach_i;
+
+      /* For each state in `reach', weed out states that don't fulfill the
+	 minimal matching conditions. */
+      if (tnfa->num_minimals && new_match)
+	{
+	  new_match = 0;
+	  reach_next_i = reach_next;
+	  for (reach_i = reach; reach_i->state; reach_i++)
+	    {
+	      int skip = 0;
+	      for (i = 0; tnfa->minimal_tags[i] >= 0; i += 2)
+		{
+		  int end = tnfa->minimal_tags[i];
+		  int start = tnfa->minimal_tags[i + 1];
+		  DPRINT(("  Minimal start %d, end %d\n", start, end));
+		  if (end >= num_tags)
+		    {
+		      DPRINT(("	 Throwing %p out.\n", reach_i->state));
+		      skip = 1;
+		      break;
+		    }
+		  else if (reach_i->tags[start] == match_tags[start]
+			   && reach_i->tags[end] < match_tags[end])
+		    {
+		      DPRINT(("	 Throwing %p out because t%d < %d\n",
+			      reach_i->state, end, match_tags[end]));
+		      skip = 1;
+		      break;
+		    }
+		}
+	      if (!skip)
+		{
+		  reach_next_i->state = reach_i->state;
+		  tmp_iptr = reach_next_i->tags;
+		  reach_next_i->tags = reach_i->tags;
+		  reach_i->tags = tmp_iptr;
+		  reach_next_i++;
+		}
+	    }
+	  reach_next_i->state = NULL;
+
+	  /* Swap `reach' and `reach_next'. */
+	  reach_i = reach;
+	  reach = reach_next;
+	  reach_next = reach_i;
+	}
+
+      /* For each state in `reach' see if there is a transition leaving with
+	 the current input symbol to a state not yet in `reach_next', and
+	 add the destination states to `reach_next'. */
+      reach_next_i = reach_next;
+      for (reach_i = reach; reach_i->state; reach_i++)
+	{
+	  for (trans_i = reach_i->state; trans_i->state; trans_i++)
+	    {
+	      /* Does this transition match the input symbol? */
+	      if (trans_i->code_min <= (tre_cint_t)prev_c &&
+		  trans_i->code_max >= (tre_cint_t)prev_c)
+		{
+		  if (trans_i->assertions
+		      && (CHECK_ASSERTIONS(trans_i->assertions)
+			  || CHECK_CHAR_CLASSES(trans_i, tnfa, eflags)))
+		    {
+		      DPRINT(("assertion failed\n"));
+		      continue;
+		    }
+
+		  /* Compute the tags after this transition. */
+		  for (i = 0; i < num_tags; i++)
+		    tmp_tags[i] = reach_i->tags[i];
+		  tag_i = trans_i->tags;
+		  if (tag_i != NULL)
+		    while (*tag_i >= 0)
+		      {
+			if (*tag_i < num_tags)
+			  tmp_tags[*tag_i] = pos;
+			tag_i++;
+		      }
+
+		  if (reach_pos[trans_i->state_id].pos < pos)
+		    {
+		      /* Found an unvisited node. */
+		      reach_next_i->state = trans_i->state;
+		      tmp_iptr = reach_next_i->tags;
+		      reach_next_i->tags = tmp_tags;
+		      tmp_tags = tmp_iptr;
+		      reach_pos[trans_i->state_id].pos = pos;
+		      reach_pos[trans_i->state_id].tags = &reach_next_i->tags;
+
+		      if (reach_next_i->state == tnfa->final
+			  && (match_eo == -1
+			      || (num_tags > 0
+				  && reach_next_i->tags[0] <= match_tags[0])))
+			{
+			  DPRINT(("  found match %p\n", trans_i->state));
+			  match_eo = pos;
+			  new_match = 1;
+			  for (i = 0; i < num_tags; i++)
+			    match_tags[i] = reach_next_i->tags[i];
+			}
+		      reach_next_i++;
+
+		    }
+		  else
+		    {
+		      assert(reach_pos[trans_i->state_id].pos == pos);
+		      /* Another path has also reached this state.  We choose
+			 the winner by examining the tag values for both
+			 paths. */
+		      if (tre_tag_order(num_tags, tnfa->tag_directions,
+					tmp_tags,
+					*reach_pos[trans_i->state_id].tags))
+			{
+			  /* The new path wins. */
+			  tmp_iptr = *reach_pos[trans_i->state_id].tags;
+			  *reach_pos[trans_i->state_id].tags = tmp_tags;
+			  if (trans_i->state == tnfa->final)
+			    {
+			      DPRINT(("	 found better match\n"));
+			      match_eo = pos;
+			      new_match = 1;
+			      for (i = 0; i < num_tags; i++)
+				match_tags[i] = tmp_tags[i];
+			    }
+			  tmp_tags = tmp_iptr;
+			}
+		    }
+		}
+	    }
+	}
+      reach_next_i->state = NULL;
+    }
+
+  DPRINT(("match end offset = %d\n", match_eo));
+
+  *match_end_ofs = match_eo;
+  ret = match_eo >= 0 ? REG_OK : REG_NOMATCH;
+
+#ifndef TRE_USE_ALLOCA
+  if (buf)
+    xfree(buf);
+#endif /* !TRE_USE_ALLOCA */
+  return ret;
+}
+
+/* EOF */

vendored/tre/tre-match-utils.h

@@ -0,0 +1,219 @@
+/*
+  tre-match-utils.h - TRE matcher helper definitions
+
+  This software is released under a BSD-style license.
+  See the file LICENSE for details and copyright.
+
+*/
+#ifndef TRE_MATCH_UTILS_H
+#define TRE_MATCH_UTILS_H
+
+#define str_source ((const tre_str_source*)string)
+
+#ifdef TRE_WCHAR
+
+#ifdef TRE_MULTIBYTE
+
+/* Wide character and multibyte support. */
+
+#define GET_NEXT_WCHAR()						      \
+  do {									      \
+    prev_c = next_c;							      \
+    if (type == STR_BYTE)						      \
+      {									      \
+	pos++;								      \
+	if (len >= 0 && pos >= len)					      \
+	  next_c = '\0';						      \
+	else								      \
+	  next_c = (unsigned char)(*str_byte++);			      \
+      }									      \
+    else if (type == STR_WIDE)						      \
+      {									      \
+	pos++;								      \
+	if (len >= 0 && pos >= len)					      \
+	  next_c = L'\0';						      \
+	else								      \
+	  next_c = *str_wide++;						      \
+      }									      \
+    else if (type == STR_MBS)						      \
+      {									      \
+        pos += pos_add_next;					      	      \
+	if (str_byte == NULL)						      \
+	  next_c = L'\0';						      \
+	else								      \
+	  {								      \
+	    size_t w;							      \
+	    size_t max;							      \
+	    if (len >= 0)						      \
+	      max = len - pos;						      \
+	    else							      \
+	      max = 32;							      \
+	    if (max <= 0)						      \
+	      {								      \
+		next_c = L'\0';						      \
+		pos_add_next = 1;					      \
+	      }								      \
+	    else							      \
+	      {								      \
+		w = tre_mbrtowc(&next_c, str_byte, (size_t)max, &mbstate);    \
+		if (w == (size_t)-1 || w == (size_t)-2)			      \
+		  return REG_NOMATCH;					      \
+		if (w == 0 && len >= 0)					      \
+		  {							      \
+		    pos_add_next = 1;					      \
+		    next_c = 0;						      \
+		    str_byte++;						      \
+		  }							      \
+		else							      \
+		  {							      \
+		    pos_add_next = w;					      \
+		    str_byte += w;					      \
+		  }							      \
+	      }								      \
+	  }								      \
+      }									      \
+    else if (type == STR_USER)						      \
+      {									      \
+        pos += pos_add_next;					      	      \
+	str_user_end = str_source->get_next_char(&next_c, &pos_add_next,      \
+                                                 str_source->context);	      \
+      }									      \
+  } while(/*CONSTCOND*/(void)0,0)
+
+#else /* !TRE_MULTIBYTE */
+
+/* Wide character support, no multibyte support. */
+
+#define GET_NEXT_WCHAR()						      \
+  do {									      \
+    prev_c = next_c;							      \
+    if (type == STR_BYTE)						      \
+      {									      \
+	pos++;								      \
+	if (len >= 0 && pos >= len)					      \
+	  next_c = '\0';						      \
+	else								      \
+	  next_c = (unsigned char)(*str_byte++);			      \
+      }									      \
+    else if (type == STR_WIDE)						      \
+      {									      \
+	pos++;								      \
+	if (len >= 0 && pos >= len)					      \
+	  next_c = L'\0';						      \
+	else								      \
+	  next_c = *str_wide++;						      \
+      }									      \
+    else if (type == STR_USER)						      \
+      {									      \
+        pos += pos_add_next;					      	      \
+	str_user_end = str_source->get_next_char(&next_c, &pos_add_next,      \
+                                                 str_source->context);	      \
+      }									      \
+  } while(/*CONSTCOND*/(void)0,0)
+
+#endif /* !TRE_MULTIBYTE */
+
+#else /* !TRE_WCHAR */
+
+/* No wide character or multibyte support. */
+
+#define GET_NEXT_WCHAR()						      \
+  do {									      \
+    prev_c = next_c;							      \
+    if (type == STR_BYTE)						      \
+      {									      \
+	pos++;								      \
+	if (len >= 0 && pos >= len)					      \
+	  next_c = '\0';						      \
+	else								      \
+	  next_c = (unsigned char)(*str_byte++);			      \
+      }									      \
+    else if (type == STR_USER)						      \
+      {									      \
+	pos += pos_add_next;						      \
+	str_user_end = str_source->get_next_char(&next_c, &pos_add_next,      \
+						 str_source->context);	      \
+      }									      \
+  } while(/*CONSTCOND*/(void)0,0)
+
+#endif /* !TRE_WCHAR */
+
+
+
+#define IS_WORD_CHAR(c)	 ((c) == L'_' || tre_isalnum(c))
+
+#define CHECK_ASSERTIONS(assertions)					      \
+  (((assertions & ASSERT_AT_BOL)					      \
+    && (pos > 0 || reg_notbol)						      \
+    && (prev_c != L'\n' || !reg_newline))				      \
+   || ((assertions & ASSERT_AT_EOL)					      \
+       && (next_c != L'\0' || reg_noteol)				      \
+       && (next_c != L'\n' || !reg_newline))				      \
+   || ((assertions & ASSERT_AT_BOW)					      \
+       && (IS_WORD_CHAR(prev_c) || !IS_WORD_CHAR(next_c)))	              \
+   || ((assertions & ASSERT_AT_EOW)					      \
+       && (!IS_WORD_CHAR(prev_c) || IS_WORD_CHAR(next_c)))		      \
+   || ((assertions & ASSERT_AT_WB)					      \
+       && (pos != 0 && next_c != L'\0'					      \
+	   && IS_WORD_CHAR(prev_c) == IS_WORD_CHAR(next_c)))		      \
+   || ((assertions & ASSERT_AT_WB_NEG)					      \
+       && (pos == 0 || next_c == L'\0'					      \
+	   || IS_WORD_CHAR(prev_c) != IS_WORD_CHAR(next_c))))
+
+#define CHECK_CHAR_CLASSES(trans_i, tnfa, eflags)                             \
+  (((trans_i->assertions & ASSERT_CHAR_CLASS)                                 \
+       && !(tnfa->cflags & REG_ICASE)                                         \
+       && !tre_isctype((tre_cint_t)prev_c, trans_i->u.class))                 \
+    || ((trans_i->assertions & ASSERT_CHAR_CLASS)                             \
+        && (tnfa->cflags & REG_ICASE)                                         \
+        && !tre_isctype(tre_tolower((tre_cint_t)prev_c),trans_i->u.class)     \
+	&& !tre_isctype(tre_toupper((tre_cint_t)prev_c),trans_i->u.class))    \
+    || ((trans_i->assertions & ASSERT_CHAR_CLASS_NEG)                         \
+        && tre_neg_char_classes_match(trans_i->neg_classes,(tre_cint_t)prev_c,\
+                                      tnfa->cflags & REG_ICASE)))
+
+
+
+
+/* Returns 1 if `t1' wins `t2', 0 otherwise. */
+inline static int
+tre_tag_order(int num_tags, tre_tag_direction_t *tag_directions,
+	      int *t1, int *t2)
+{
+  int i;
+  for (i = 0; i < num_tags; i++)
+    {
+      if (tag_directions[i] == TRE_TAG_MINIMIZE)
+	{
+	  if (t1[i] < t2[i])
+	    return 1;
+	  if (t1[i] > t2[i])
+	    return 0;
+	}
+      else
+	{
+	  if (t1[i] > t2[i])
+	    return 1;
+	  if (t1[i] < t2[i])
+	    return 0;
+	}
+    }
+  /*  assert(0);*/
+  return 0;
+}
+
+inline static int
+tre_neg_char_classes_match(tre_ctype_t *classes, tre_cint_t wc, int icase)
+{
+  DPRINT(("neg_char_classes_test: %p, %d, %d\n", classes, wc, icase));
+  while (*classes != (tre_ctype_t)0)
+    if ((!icase && tre_isctype(wc, *classes))
+	|| (icase && (tre_isctype(tre_toupper(wc), *classes)
+		      || tre_isctype(tre_tolower(wc), *classes))))
+      return 1; /* Match. */
+    else
+      classes++;
+  return 0; /* No match. */
+}
+
+#endif /* TRE_MATCH_UTILS_H */

vendored/tre/tre-mem.c

@@ -0,0 +1,155 @@
+/*
+  tre-mem.c - TRE memory allocator
+
+  This software is released under a BSD-style license.
+  See the file LICENSE for details and copyright.
+
+*/
+
+/*
+  This memory allocator is for allocating small memory blocks efficiently
+  in terms of memory overhead and execution speed.  The allocated blocks
+  cannot be freed individually, only all at once.  There can be multiple
+  allocators, though.
+*/
+
+#ifdef HAVE_CONFIG_H
+#include <config.h>
+#endif /* HAVE_CONFIG_H */
+#include <stdlib.h>
+#include <string.h>
+
+#include "tre-internal.h"
+#include "tre-mem.h"
+#include "xmalloc.h"
+
+
+/* Returns a new memory allocator or NULL if out of memory. */
+tre_mem_t
+tre_mem_new_impl(int provided, void *provided_block)
+{
+  tre_mem_t mem;
+  if (provided)
+    {
+      mem = provided_block;
+      memset(mem, 0, sizeof(*mem));
+    }
+  else
+    mem = xcalloc(1, sizeof(*mem));
+  if (mem == NULL)
+    return NULL;
+  return mem;
+}
+
+
+/* Frees the memory allocator and all memory allocated with it. */
+void
+tre_mem_destroy(tre_mem_t mem)
+{
+  tre_list_t *tmp, *l = mem->blocks;
+
+  while (l != NULL)
+    {
+      xfree(l->data);
+      tmp = l->next;
+      xfree(l);
+      l = tmp;
+    }
+  xfree(mem);
+}
+
+
+/* Allocates a block of `size' bytes from `mem'.  Returns a pointer to the
+   allocated block or NULL if an underlying malloc() failed. */
+void *
+tre_mem_alloc_impl(tre_mem_t mem, int provided, void *provided_block,
+		   int zero, size_t size)
+{
+  void *ptr;
+
+  if (mem->failed)
+    {
+      DPRINT(("tre_mem_alloc: oops, called after failure?!\n"));
+      return NULL;
+    }
+
+#ifdef MALLOC_DEBUGGING
+  if (!provided)
+    {
+      ptr = xmalloc(1);
+      if (ptr == NULL)
+	{
+	  DPRINT(("tre_mem_alloc: xmalloc forced failure\n"));
+	  mem->failed = 1;
+	  return NULL;
+	}
+      xfree(ptr);
+    }
+#endif /* MALLOC_DEBUGGING */
+
+  if (mem->n < size)
+    {
+      /* We need more memory than is available in the current block.
+	 Allocate a new block. */
+      tre_list_t *l;
+      if (provided)
+	{
+	  DPRINT(("tre_mem_alloc: using provided block\n"));
+	  if (provided_block == NULL)
+	    {
+	      DPRINT(("tre_mem_alloc: provided block was NULL\n"));
+	      mem->failed = 1;
+	      return NULL;
+	    }
+	  mem->ptr = provided_block;
+	  mem->n = TRE_MEM_BLOCK_SIZE;
+	}
+      else
+	{
+	  size_t block_size;
+	  if (size * 8 > TRE_MEM_BLOCK_SIZE)
+	    block_size = size * 8;
+	  else
+	    block_size = TRE_MEM_BLOCK_SIZE;
+	  DPRINT(("tre_mem_alloc: allocating new %zu byte block\n",
+		  block_size));
+	  l = xmalloc(sizeof(*l));
+	  if (l == NULL)
+	    {
+	      mem->failed = 1;
+	      return NULL;
+	    }
+	  l->data = xmalloc(block_size);
+	  if (l->data == NULL)
+	    {
+	      xfree(l);
+	      mem->failed = 1;
+	      return NULL;
+	    }
+	  l->next = NULL;
+	  if (mem->current != NULL)
+	    mem->current->next = l;
+	  if (mem->blocks == NULL)
+	    mem->blocks = l;
+	  mem->current = l;
+	  mem->ptr = l->data;
+	  mem->n = block_size;
+	}
+    }
+
+  /* Make sure the next pointer will be aligned. */
+  size += ALIGN(mem->ptr + size, long);
+
+  /* Allocate from current block. */
+  ptr = mem->ptr;
+  mem->ptr += size;
+  mem->n -= size;
+
+  /* Set to zero if needed. */
+  if (zero)
+    memset(ptr, 0, size);
+
+  return ptr;
+}
+
+/* EOF */

vendored/tre/tre-mem.h

@@ -0,0 +1,66 @@
+/*
+  tre-mem.h - TRE memory allocator interface
+
+  This software is released under a BSD-style license.
+  See the file LICENSE for details and copyright.
+
+*/
+
+#ifndef TRE_MEM_H
+#define TRE_MEM_H 1
+
+#include <stdlib.h>
+
+#define TRE_MEM_BLOCK_SIZE 1024
+
+typedef struct tre_list {
+  void *data;
+  struct tre_list *next;
+} tre_list_t;
+
+typedef struct tre_mem_struct {
+  tre_list_t *blocks;
+  tre_list_t *current;
+  char *ptr;
+  size_t n;
+  int failed;
+  void **provided;
+} *tre_mem_t;
+
+
+tre_mem_t tre_mem_new_impl(int provided, void *provided_block);
+void *tre_mem_alloc_impl(tre_mem_t mem, int provided, void *provided_block,
+			 int zero, size_t size);
+
+/* Returns a new memory allocator or NULL if out of memory. */
+#define tre_mem_new()  tre_mem_new_impl(0, NULL)
+
+/* Allocates a block of `size' bytes from `mem'.  Returns a pointer to the
+   allocated block or NULL if an underlying malloc() failed. */
+#define tre_mem_alloc(mem, size) tre_mem_alloc_impl(mem, 0, NULL, 0, size)
+
+/* Allocates a block of `size' bytes from `mem'.  Returns a pointer to the
+   allocated block or NULL if an underlying malloc() failed.  The memory
+   is set to zero. */
+#define tre_mem_calloc(mem, size) tre_mem_alloc_impl(mem, 0, NULL, 1, size)
+
+#ifdef TRE_USE_ALLOCA
+/* alloca() versions.  Like above, but memory is allocated with alloca()
+   instead of malloc(). */
+
+#define tre_mem_newa() \
+  tre_mem_new_impl(1, alloca(sizeof(struct tre_mem_struct)))
+
+#define tre_mem_alloca(mem, size)					      \
+  ((mem)->n >= (size)							      \
+   ? tre_mem_alloc_impl((mem), 1, NULL, 0, (size))			      \
+   : tre_mem_alloc_impl((mem), 1, alloca(TRE_MEM_BLOCK_SIZE), 0, (size)))
+#endif /* TRE_USE_ALLOCA */
+
+
+/* Frees the memory allocator and all memory allocated with it. */
+void tre_mem_destroy(tre_mem_t mem);
+
+#endif /* TRE_MEM_H */
+
+/* EOF */

vendored/tre/tre-parse.h

@@ -0,0 +1,50 @@
+/*
+  tre-parse.c - Regexp parser definitions
+
+  This software is released under a BSD-style license.
+  See the file LICENSE for details and copyright.
+
+*/
+
+#ifndef TRE_PARSE_H
+#define TRE_PARSE_H 1
+
+/* Parse context. */
+typedef struct {
+  /* Memory allocator.	The AST is allocated using this. */
+  tre_mem_t mem;
+  /* Stack used for keeping track of regexp syntax. */
+  tre_stack_t *stack;
+  /* The parse result. */
+  tre_ast_node_t *result;
+  /* The regexp to parse and its length. */
+  const tre_char_t *re;
+  /* The first character of the entire regexp. */
+  const tre_char_t *re_start;
+  /* The first character after the end of the regexp. */
+  const tre_char_t *re_end;
+  size_t len;
+  /* Current submatch ID. */
+  int submatch_id;
+  /* The highest back reference or -1 if none seen so far. */
+  int max_backref;
+  /* This flag is set if the regexp uses approximate matching. */
+  int have_approx;
+  /* Compilation flags. */
+  int cflags;
+  /* If this flag is set the top-level submatch is not captured. */
+  int nofirstsub;
+  /* The currently set approximate matching parameters. */
+  int params[TRE_PARAM_LAST];
+  /* the MB_CUR_MAX in use */
+  int mb_cur_max;
+} tre_parse_ctx_t;
+
+/* Parses a wide character regexp pattern into a syntax tree.  This parser
+   handles both syntaxes (BRE and ERE), including the TRE extensions. */
+reg_errcode_t
+tre_parse(tre_parse_ctx_t *ctx);
+
+#endif /* TRE_PARSE_H */
+
+/* EOF */

vendored/tre/tre-stack.c

@@ -0,0 +1,123 @@
+/*
+  tre-stack.c - Simple stack implementation
+
+  This software is released under a BSD-style license.
+  See the file LICENSE for details and copyright.
+
+*/
+
+#ifdef HAVE_CONFIG_H
+#include <config.h>
+#endif /* HAVE_CONFIG_H */
+#include <stdlib.h>
+#include <assert.h>
+
+#include "tre-internal.h"
+#include "tre-stack.h"
+#include "xmalloc.h"
+
+union tre_stack_item {
+  void *voidptr_value;
+  int int_value;
+};
+
+struct tre_stack_rec {
+  size_t size;
+  size_t max_size;
+  size_t ptr;
+  union tre_stack_item *stack;
+};
+
+
+tre_stack_t *
+tre_stack_new(size_t size, size_t max_size)
+{
+  tre_stack_t *s;
+
+  s = xmalloc(sizeof(*s));
+  if (s != NULL)
+    {
+      s->stack = xmalloc(sizeof(*s->stack) * size);
+      if (s->stack == NULL)
+	{
+	  xfree(s);
+	  return NULL;
+	}
+      s->size = size;
+      s->max_size = max_size;
+      s->ptr = 0;
+    }
+  return s;
+}
+
+void
+tre_stack_destroy(tre_stack_t *s)
+{
+  xfree(s->stack);
+  xfree(s);
+}
+
+size_t
+tre_stack_num_items(tre_stack_t *s)
+{
+  return s->ptr;
+}
+
+static reg_errcode_t
+tre_stack_push(tre_stack_t *s, union tre_stack_item value)
+{
+  if (s->ptr < s->size)
+    {
+      s->stack[s->ptr] = value;
+      s->ptr++;
+    }
+  else
+    {
+      if (s->size >= s->max_size)
+	{
+	  DPRINT(("tre_stack_push: stack full\n"));
+	  return REG_ESPACE;
+	}
+      else
+	{
+	  union tre_stack_item *new_buffer;
+	  size_t new_size;
+	  DPRINT(("tre_stack_push: trying to realloc more space\n"));
+	  new_size = s->size + s->size;
+	  if (new_size > s->max_size)
+	    new_size = s->max_size;
+	  new_buffer = xrealloc(s->stack, sizeof(*new_buffer) * new_size);
+	  if (new_buffer == NULL)
+	    {
+	      DPRINT(("tre_stack_push: realloc failed.\n"));
+	      return REG_ESPACE;
+	    }
+	  DPRINT(("tre_stack_push: realloc succeeded.\n"));
+	  assert(new_size > s->size);
+	  s->size = new_size;
+	  s->stack = new_buffer;
+	  tre_stack_push(s, value);
+	}
+    }
+  return REG_OK;
+}
+
+#define define_pushf(typetag, type)  \
+  declare_pushf(typetag, type) {     \
+    union tre_stack_item item;	     \
+    item.typetag ## _value = value;  \
+    return tre_stack_push(s, item);  \
+}
+
+define_pushf(int, int)
+define_pushf(voidptr, void *)
+
+#define define_popf(typetag, type)		    \
+  declare_popf(typetag, type) {			    \
+    return s->stack[--s->ptr].typetag ## _value;    \
+  }
+
+define_popf(int, int)
+define_popf(voidptr, void *)
+
+/* EOF */

vendored/tre/tre-stack.h

@@ -0,0 +1,76 @@
+/*
+  tre-stack.h: Stack definitions
+
+  This software is released under a BSD-style license.
+  See the file LICENSE for details and copyright.
+
+*/
+
+
+#ifndef TRE_STACK_H
+#define TRE_STACK_H 1
+
+#include "tre.h"
+
+typedef struct tre_stack_rec tre_stack_t;
+
+/* Creates a new stack object with initial size `size' and maximum size
+   `max_size'. Pushing an additional item onto a full stack will resize
+   the stack to double its capacity until the maximum is reached. Returns
+   the stack object or NULL if out of memory. */
+tre_stack_t *
+tre_stack_new(size_t size, size_t max_size);
+
+/* Frees the stack object. */
+void
+tre_stack_destroy(tre_stack_t *s);
+
+/* Returns the current number of items on the stack. */
+size_t
+tre_stack_num_items(tre_stack_t *s);
+
+/* Each tre_stack_push_*(tre_stack_t *s, <type> value) function pushes
+   `value' on top of stack `s'.  Returns REG_ESPACE if out of memory.
+   This tries to realloc() more space before failing if maximum size
+   has not yet been reached.  Returns REG_OK if successful. */
+#define declare_pushf(typetag, type)					      \
+  reg_errcode_t tre_stack_push_ ## typetag(tre_stack_t *s, type value)
+
+declare_pushf(voidptr, void *);
+declare_pushf(int, int);
+
+/* Each tre_stack_pop_*(tre_stack_t *s) function pops the topmost
+   element off of stack `s' and returns it.  The stack must not be
+   empty. */
+#define declare_popf(typetag, type)		  \
+  type tre_stack_pop_ ## typetag(tre_stack_t *s)
+
+declare_popf(voidptr, void *);
+declare_popf(int, int);
+
+/* Just to save some typing. */
+#define STACK_PUSH(s, typetag, value)					      \
+  do									      \
+    {									      \
+      status = tre_stack_push_ ## typetag(s, value);			      \
+    }									      \
+  while (/*CONSTCOND*/(void)0,0)
+
+#define STACK_PUSHX(s, typetag, value)					      \
+  {									      \
+    status = tre_stack_push_ ## typetag(s, value);			      \
+    if (status != REG_OK)						      \
+      break;								      \
+  }
+
+#define STACK_PUSHR(s, typetag, value)					      \
+  {									      \
+    reg_errcode_t _status;						      \
+    _status = tre_stack_push_ ## typetag(s, value);			      \
+    if (_status != REG_OK)						      \
+      return _status;							      \
+  }
+
+#endif /* TRE_STACK_H */
+
+/* EOF */

vendored/tre/tre.h

@@ -0,0 +1,344 @@
+/*
+  tre.h - TRE public API definitions
+
+  This software is released under a BSD-style license.
+  See the file LICENSE for details and copyright.
+
+*/
+
+#ifndef TRE_H
+#define TRE_H 1
+
+#ifdef USE_LOCAL_TRE_H
+/* Make certain to use the header(s) from the TRE package that this
+   file is part of by giving the full path to the header from this directory. */
+#include "tre-config.h"
+#else
+/* Use the header in the same directory as this file if there is one. */
+#include "tre-config.h"
+#endif
+
+#ifdef HAVE_SYS_TYPES_H
+#include <sys/types.h>
+#endif /* HAVE_SYS_TYPES_H */
+
+#ifdef HAVE_LIBUTF8_H
+#include <libutf8.h>
+#endif /* HAVE_LIBUTF8_H */
+
+#ifdef TRE_USE_SYSTEM_REGEX_H
+/* Include the system regex.h to make TRE ABI compatible with the
+   system regex. */
+#include TRE_SYSTEM_REGEX_H_PATH
+#define tre_regcomp  regcomp
+#define tre_regexec  regexec
+#define tre_regerror regerror
+#define tre_regfree  regfree
+/* The GNU C regex has a number of refinements to the POSIX standard for the
+   formal parameter list of the regexec() function, and some of those fail to
+   compile when using LLVM.  The refinements seem to be opt-out rather than
+   opt-in when using a recent gcc, and they produce a warning when TRE tries
+   to mimic the API without the refinements.  The TRE code still works but
+   the warnings are distracting, so try to #define a flag to indicate when to 
+   add the refinements to TRE's parameter list too. */
+#ifdef __GNUC__
+/* Try to test something that looks pretty REGEX specific and hope we don't
+   need a zillion different platform+compiler specific tests to deal with this. */
+#ifdef _REGEX_NELTS
+/* Define a TRE specific flag here so that:
+   1) there is only one place where code has to be changed if the test above is not adequate, and
+   2) the flag can be used in any other parts of the TRE source that might be affected by the
+      GNUC refinements.
+   Note that this flag is only defined when all of TRE_USE_SYSTEM_REGEX_H, __GNUC__, and _REGEX_NELTS are defined. */
+#define TRE_USE_GNUC_REGEXEC_FPL 1
+#endif
+#endif
+#endif /* TRE_USE_SYSTEM_REGEX_H */
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+#ifdef TRE_USE_SYSTEM_REGEX_H
+
+#ifndef REG_OK
+#define REG_OK 0
+#endif /* !REG_OK */
+
+#ifndef HAVE_REG_ERRCODE_T
+typedef int reg_errcode_t;
+#endif /* !HAVE_REG_ERRCODE_T */
+
+#if !defined(REG_NOSPEC) && !defined(REG_LITERAL)
+#define REG_LITERAL 0x1000
+#endif
+
+/* Extra tre_regcomp() return error codes. */
+#define REG_BADMAX REG_BADBR
+
+/* Extra tre_regcomp() flags. */
+#ifndef REG_BASIC
+#define REG_BASIC	0
+#endif /* !REG_BASIC */
+#define REG_RIGHT_ASSOC (REG_LITERAL << 1)
+#ifdef REG_UNGREEDY
+/* We're going to use TRE code, so we need the TRE define (dodge problem in MacOS). */
+#undef REG_UNGREEDY
+#endif
+#define REG_UNGREEDY    (REG_RIGHT_ASSOC << 1)
+
+#define REG_USEBYTES    (REG_UNGREEDY << 1)
+
+/* Extra tre_regexec() flags. */
+#define REG_APPROX_MATCHER	 0x1000
+#ifdef REG_BACKTRACKING_MATCHER
+/* We're going to use TRE code, so we need the TRE define (dodge problem in MacOS). */
+#undef REG_BACKTRACKING_MATCHER
+#endif
+#define REG_BACKTRACKING_MATCHER (REG_APPROX_MATCHER << 1)
+
+#else /* !TRE_USE_SYSTEM_REGEX_H */
+
+/* If the we're not using system regex.h, we need to define the
+   structs and enums ourselves. */
+
+typedef int regoff_t;
+typedef struct {
+  size_t re_nsub;  /* Number of parenthesized subexpressions. */
+  void *value;	   /* For internal use only. */
+} regex_t;
+
+typedef struct {
+  regoff_t rm_so;
+  regoff_t rm_eo;
+} regmatch_t;
+
+
+typedef enum {
+  REG_OK = 0,		/* No error. */
+  /* POSIX tre_regcomp() return error codes.  (In the order listed in the
+     standard.)	 */
+  REG_NOMATCH,		/* No match. */
+  REG_BADPAT,		/* Invalid regexp. */
+  REG_ECOLLATE,		/* Unknown collating element. */
+  REG_ECTYPE,		/* Unknown character class name. */
+  REG_EESCAPE,		/* Trailing backslash. */
+  REG_ESUBREG,		/* Invalid back reference. */
+  REG_EBRACK,		/* "[]" imbalance */
+  REG_EPAREN,		/* "\(\)" or "()" imbalance */
+  REG_EBRACE,		/* "\{\}" or "{}" imbalance */
+  REG_BADBR,		/* Invalid content of {} */
+  REG_ERANGE,		/* Invalid use of range operator */
+  REG_ESPACE,		/* Out of memory.  */
+  REG_BADRPT,		/* Invalid use of repetition operators. */
+  REG_BADMAX,		/* Maximum repetition in {} too large */
+} reg_errcode_t;
+
+/* POSIX tre_regcomp() flags. */
+#define REG_EXTENDED	1
+#define REG_ICASE	(REG_EXTENDED << 1)
+#define REG_NEWLINE	(REG_ICASE << 1)
+#define REG_NOSUB	(REG_NEWLINE << 1)
+
+/* Extra tre_regcomp() flags. */
+#define REG_BASIC	0
+#define REG_LITERAL	(REG_NOSUB << 1)
+#define REG_RIGHT_ASSOC (REG_LITERAL << 1)
+#define REG_UNGREEDY    (REG_RIGHT_ASSOC << 1)
+#define REG_USEBYTES    (REG_UNGREEDY << 1)
+
+/* POSIX tre_regexec() flags. */
+#define REG_NOTBOL 1
+#define REG_NOTEOL (REG_NOTBOL << 1)
+
+/* Extra tre_regexec() flags. */
+#define REG_APPROX_MATCHER	 (REG_NOTEOL << 1)
+#define REG_BACKTRACKING_MATCHER (REG_APPROX_MATCHER << 1)
+
+#endif /* !TRE_USE_SYSTEM_REGEX_H */
+
+/* REG_NOSPEC and REG_LITERAL mean the same thing. */
+#if defined(REG_LITERAL) && !defined(REG_NOSPEC)
+#define REG_NOSPEC	REG_LITERAL
+#elif defined(REG_NOSPEC) && !defined(REG_LITERAL)
+#define REG_LITERAL	REG_NOSPEC
+#endif /* defined(REG_NOSPEC) */
+
+/* The maximum number of iterations in a bound expression. */
+#undef RE_DUP_MAX
+#define RE_DUP_MAX 255
+
+/* The POSIX.2 regexp functions */
+extern int
+tre_regcomp(regex_t *preg, const char *regex, int cflags);
+
+#ifdef TRE_USE_GNUC_REGEXEC_FPL
+extern int
+tre_regexec(const regex_t *preg, const char *string,
+	size_t nmatch, regmatch_t pmatch[_Restrict_arr_ _REGEX_NELTS (nmatch)],
+	int eflags);
+#else
+extern int
+tre_regexec(const regex_t *preg, const char *string, size_t nmatch,
+	regmatch_t pmatch[], int eflags);
+#endif
+
+extern int
+tre_regcompb(regex_t *preg, const char *regex, int cflags);
+
+extern int
+tre_regexecb(const regex_t *preg, const char *string, size_t nmatch,
+	regmatch_t pmatch[], int eflags);
+
+extern size_t
+tre_regerror(int errcode, const regex_t *preg, char *errbuf,
+	 size_t errbuf_size);
+
+extern void
+tre_regfree(regex_t *preg);
+
+#ifdef TRE_WCHAR
+#ifdef HAVE_WCHAR_H
+#include <wchar.h>
+#endif /* HAVE_WCHAR_H */
+
+/* Wide character versions (not in POSIX.2). */
+extern int
+tre_regwcomp(regex_t *preg, const wchar_t *regex, int cflags);
+
+extern int
+tre_regwexec(const regex_t *preg, const wchar_t *string,
+	 size_t nmatch, regmatch_t pmatch[], int eflags);
+#endif /* TRE_WCHAR */
+
+/* Versions with a maximum length argument and therefore the capability to
+   handle null characters in the middle of the strings (not in POSIX.2). */
+extern int
+tre_regncomp(regex_t *preg, const char *regex, size_t len, int cflags);
+
+extern int
+tre_regnexec(const regex_t *preg, const char *string, size_t len,
+	 size_t nmatch, regmatch_t pmatch[], int eflags);
+
+/* regn*b versions take byte literally as 8-bit values */
+extern int
+tre_regncompb(regex_t *preg, const char *regex, size_t n, int cflags);
+
+extern int
+tre_regnexecb(const regex_t *preg, const char *str, size_t len,
+	  size_t nmatch, regmatch_t pmatch[], int eflags);
+
+#ifdef TRE_WCHAR
+extern int
+tre_regwncomp(regex_t *preg, const wchar_t *regex, size_t len, int cflags);
+
+extern int
+tre_regwnexec(const regex_t *preg, const wchar_t *string, size_t len,
+	  size_t nmatch, regmatch_t pmatch[], int eflags);
+#endif /* TRE_WCHAR */
+
+#ifdef TRE_APPROX
+
+/* Approximate matching parameter struct. */
+typedef struct {
+  int cost_ins;	       /* Default cost of an inserted character. */
+  int cost_del;	       /* Default cost of a deleted character. */
+  int cost_subst;      /* Default cost of a substituted character. */
+  int max_cost;	       /* Maximum allowed cost of a match. */
+
+  int max_ins;	       /* Maximum allowed number of inserts. */
+  int max_del;	       /* Maximum allowed number of deletes. */
+  int max_subst;       /* Maximum allowed number of substitutes. */
+  int max_err;	       /* Maximum allowed number of errors total. */
+} regaparams_t;
+
+/* Approximate matching result struct. */
+typedef struct {
+  size_t nmatch;       /* Length of pmatch[] array. */
+  regmatch_t *pmatch;  /* Submatch data. */
+  int cost;	       /* Cost of the match. */
+  int num_ins;	       /* Number of inserts in the match. */
+  int num_del;	       /* Number of deletes in the match. */
+  int num_subst;       /* Number of substitutes in the match. */
+} regamatch_t;
+
+
+/* Approximate matching functions. */
+extern int
+tre_regaexec(const regex_t *preg, const char *string,
+	 regamatch_t *match, regaparams_t params, int eflags);
+
+extern int
+tre_reganexec(const regex_t *preg, const char *string, size_t len,
+	  regamatch_t *match, regaparams_t params, int eflags);
+
+extern int
+tre_regaexecb(const regex_t *preg, const char *string,
+	  regamatch_t *match, regaparams_t params, int eflags);
+
+#ifdef TRE_WCHAR
+/* Wide character approximate matching. */
+extern int
+tre_regawexec(const regex_t *preg, const wchar_t *string,
+	  regamatch_t *match, regaparams_t params, int eflags);
+
+extern int
+tre_regawnexec(const regex_t *preg, const wchar_t *string, size_t len,
+	   regamatch_t *match, regaparams_t params, int eflags);
+#endif /* TRE_WCHAR */
+
+/* Sets the parameters to default values. */
+extern void
+tre_regaparams_default(regaparams_t *params);
+#endif /* TRE_APPROX */
+
+#ifdef TRE_WCHAR
+typedef wchar_t tre_char_t;
+#else /* !TRE_WCHAR */
+typedef unsigned char tre_char_t;
+#endif /* !TRE_WCHAR */
+
+typedef struct {
+  int (*get_next_char)(tre_char_t *c, unsigned int *pos_add, void *context);
+  void (*rewind)(size_t pos, void *context);
+  int (*compare)(size_t pos1, size_t pos2, size_t len, void *context);
+  void *context;
+} tre_str_source;
+
+extern int
+tre_reguexec(const regex_t *preg, const tre_str_source *string,
+	 size_t nmatch, regmatch_t pmatch[], int eflags);
+
+/* Returns the version string.	The returned string is static. */
+extern char *
+tre_version(void);
+
+/* Returns the value for a config parameter.  The type to which `result'
+   must point to depends of the value of `query', see documentation for
+   more details. */
+extern int
+tre_config(int query, void *result);
+
+enum {
+  TRE_CONFIG_APPROX,
+  TRE_CONFIG_WCHAR,
+  TRE_CONFIG_MULTIBYTE,
+  TRE_CONFIG_SYSTEM_ABI,
+  TRE_CONFIG_VERSION
+};
+
+/* Returns 1 if the compiled pattern has back references, 0 if not. */
+extern int
+tre_have_backrefs(const regex_t *preg);
+
+/* Returns 1 if the compiled pattern uses approximate matching features,
+   0 if not. */
+extern int
+tre_have_approx(const regex_t *preg);
+
+#ifdef __cplusplus
+}
+#endif
+#endif				/* TRE_H */
+
+/* EOF */

vendored/tre/tre_all.c

@@ -0,0 +1,30 @@
+/*
+ * tre_all.c — Single compilation unit for vendored TRE regex library.
+ * Only compiled on Windows (POSIX systems use system <regex.h>).
+ */
+#ifdef _WIN32
+
+#include "tre-config.h"
+
+/* Core library sources */
+#include "tre-ast.c"
+#include "tre-compile.c"
+#include "tre-mem.c"
+#include "tre-parse.c"
+#include "tre-stack.c"
+
+/* Matchers (tre-match-utils.h has include guard to handle being
+ * included by both matcher .c files in same translation unit) */
+#include "tre-match-parallel.c"
+#include "tre-match-backtrack.c"
+
+/* Approximate matching + filter */
+#include "tre-match-approx.c"
+#include "tre-filter.c"
+
+/* POSIX API */
+#include "regcomp.c"
+#include "regexec.c"
+#include "regerror.c"
+
+#endif /* _WIN32 */

vendored/tre/xmalloc.c

@@ -0,0 +1,355 @@
+/*
+  xmalloc.c - Simple malloc debugging library implementation
+
+  This software is released under a BSD-style license.
+  See the file LICENSE for details and copyright.
+
+*/
+
+/*
+  TODO:
+   - red zones
+   - group dumps by source location
+*/
+
+#ifdef HAVE_CONFIG_H
+#include <config.h>
+#endif /* HAVE_CONFIG_H */
+
+#include <stdint.h>
+#include <stdlib.h>
+#include <assert.h>
+#include <stdio.h>
+#define XMALLOC_INTERNAL 1
+#include "xmalloc.h"
+
+
+/*
+  Internal stuff.
+*/
+
+typedef struct hashTableItemRec {
+  void *ptr;
+  size_t bytes;
+  const char *file;
+  int line;
+  const char *func;
+  struct hashTableItemRec *next;
+} hashTableItem;
+
+typedef struct {
+  hashTableItem **table;
+} hashTable;
+
+static int xmalloc_peak;
+int xmalloc_current;
+static int xmalloc_peak_blocks;
+int xmalloc_current_blocks;
+static int xmalloc_fail_after;
+
+#define TABLE_BITS 8
+#define TABLE_MASK ((1 << TABLE_BITS) - 1)
+#define TABLE_SIZE (1 << TABLE_BITS)
+
+static hashTable *
+hash_table_new(void)
+{
+  hashTable *tbl;
+
+  tbl = malloc(sizeof(*tbl));
+
+  if (tbl != NULL)
+    {
+      tbl->table = calloc(TABLE_SIZE, sizeof(*tbl->table));
+
+      if (tbl->table == NULL)
+	{
+	  free(tbl);
+	  return NULL;
+	}
+    }
+
+  return tbl;
+}
+
+static unsigned int
+hash_void_ptr(void *ptr)
+{
+  unsigned int hash;
+  unsigned int i;
+
+  /* I took this hash function just off the top of my head, I have
+     no idea whether it is bad or very bad. */
+  hash = 0;
+  for (i = 0; i < sizeof(ptr) * 8 / TABLE_BITS; i++)
+    {
+      hash ^= (uintptr_t)ptr >> i * 8;
+      hash += i * 17;
+      hash &= TABLE_MASK;
+    }
+  return hash;
+}
+
+static void
+hash_table_add(hashTable *tbl, void *ptr, size_t bytes,
+	       const char *file, int line, const char *func)
+{
+  unsigned int i;
+  hashTableItem *item, *new;
+
+  i = hash_void_ptr(ptr);
+
+  item = tbl->table[i];
+  if (item != NULL)
+    while (item->next != NULL)
+      item = item->next;
+
+  new = malloc(sizeof(*new));
+  assert(new != NULL);
+  new->ptr = ptr;
+  new->bytes = bytes;
+  new->file = file;
+  new->line = line;
+  new->func = func;
+  new->next = NULL;
+  if (item != NULL)
+    item->next = new;
+  else
+    tbl->table[i] = new;
+
+  xmalloc_current += bytes;
+  if (xmalloc_current > xmalloc_peak)
+    xmalloc_peak = xmalloc_current;
+  xmalloc_current_blocks++;
+  if (xmalloc_current_blocks > xmalloc_peak_blocks)
+    xmalloc_peak_blocks = xmalloc_current_blocks;
+}
+
+static void
+#if defined(__GNUC__) && __GNUC__ >= 10
+__attribute__((access(none, 2)))
+#endif
+hash_table_del(hashTable *tbl, void *ptr)
+{
+  int i;
+  hashTableItem *item, *prev;
+
+  i = hash_void_ptr(ptr);
+
+  item = tbl->table[i];
+  if (item == NULL)
+    {
+      printf("xfree: invalid ptr %p\n", ptr);
+      abort();
+    }
+  prev = NULL;
+  while (item->ptr != ptr)
+    {
+      prev = item;
+      item = item->next;
+    }
+  if (item->ptr != ptr)
+    {
+      printf("xfree: invalid ptr %p\n", ptr);
+      abort();
+    }
+
+  xmalloc_current -= item->bytes;
+  xmalloc_current_blocks--;
+
+  if (prev != NULL)
+    {
+      prev->next = item->next;
+      free(item);
+    }
+  else
+    {
+      tbl->table[i] = item->next;
+      free(item);
+    }
+}
+
+static hashTable *xmalloc_table = NULL;
+
+static void
+xmalloc_init(void)
+{
+  if (xmalloc_table == NULL)
+    {
+      xmalloc_table = hash_table_new();
+      xmalloc_peak = 0;
+      xmalloc_peak_blocks = 0;
+      xmalloc_current = 0;
+      xmalloc_current_blocks = 0;
+      xmalloc_fail_after = -1;
+    }
+  assert(xmalloc_table != NULL);
+  assert(xmalloc_table->table != NULL);
+}
+
+
+
+/*
+  Public API.
+*/
+
+void
+xmalloc_configure(int fail_after)
+{
+  xmalloc_init();
+  xmalloc_fail_after = fail_after;
+}
+
+int
+xmalloc_dump_leaks(void)
+{
+  unsigned int i;
+  unsigned int num_leaks = 0;
+  size_t leaked_bytes = 0;
+  hashTableItem *item;
+
+  xmalloc_init();
+
+  for (i = 0; i < TABLE_SIZE; i++)
+    {
+      item = xmalloc_table->table[i];
+      while (item != NULL)
+	{
+	  printf("%s:%d: %s: %zu bytes at %p not freed\n",
+		 item->file, item->line, item->func, item->bytes, item->ptr);
+	  num_leaks++;
+	  leaked_bytes += item->bytes;
+	  item = item->next;
+	}
+    }
+  if (num_leaks == 0)
+    printf("No memory leaks.\n");
+  else
+    printf("%u unfreed memory chuncks, total %zu unfreed bytes.\n",
+	   num_leaks, leaked_bytes);
+  printf("Peak memory consumption %d bytes (%.1f kB, %.1f MB) in %d blocks ",
+	 xmalloc_peak, (double)xmalloc_peak / 1024,
+	 (double)xmalloc_peak / (1024*1024), xmalloc_peak_blocks);
+  printf("(average ");
+  if (xmalloc_peak_blocks)
+    printf("%d", ((xmalloc_peak + xmalloc_peak_blocks / 2)
+		  / xmalloc_peak_blocks));
+  else
+    printf("N/A");
+  printf(" bytes per block).\n");
+
+  return num_leaks;
+}
+
+void *
+xmalloc_impl(size_t size, const char *file, int line, const char *func)
+{
+  void *ptr;
+
+  xmalloc_init();
+  assert(size > 0);
+
+  if (xmalloc_fail_after == 0)
+    {
+      xmalloc_fail_after = -2;
+#if 0
+      printf("xmalloc: forced failure %s:%d: %s\n", file, line, func);
+#endif
+      return NULL;
+    }
+  else if (xmalloc_fail_after == -2)
+    {
+      printf("xmalloc: called after failure from %s:%d: %s\n",
+	     file, line, func);
+      assert(0);
+    }
+  else if (xmalloc_fail_after > 0)
+    xmalloc_fail_after--;
+
+  ptr = malloc(size);
+  if (ptr != NULL)
+    hash_table_add(xmalloc_table, ptr, (int)size, file, line, func);
+  return ptr;
+}
+
+void *
+xcalloc_impl(size_t nmemb, size_t size, const char *file, int line,
+	     const char *func)
+{
+  void *ptr;
+
+  xmalloc_init();
+  assert(size > 0);
+
+  if (xmalloc_fail_after == 0)
+    {
+      xmalloc_fail_after = -2;
+#if 0
+      printf("xcalloc: forced failure %s:%d: %s\n", file, line, func);
+#endif
+      return NULL;
+    }
+  else if (xmalloc_fail_after == -2)
+    {
+      printf("xcalloc: called after failure from %s:%d: %s\n",
+	     file, line, func);
+      assert(0);
+    }
+  else if (xmalloc_fail_after > 0)
+    xmalloc_fail_after--;
+
+  ptr = calloc(nmemb, size);
+  if (ptr != NULL)
+    hash_table_add(xmalloc_table, ptr, (int)(nmemb * size), file, line, func);
+  return ptr;
+}
+
+void
+xfree_impl(void *ptr, const char *file, int line, const char *func)
+{
+  /*LINTED*/(void)&file;
+  /*LINTED*/(void)&line;
+  /*LINTED*/(void)&func;
+  xmalloc_init();
+
+  if (ptr != NULL)
+    hash_table_del(xmalloc_table, ptr);
+  free(ptr);
+}
+
+void *
+xrealloc_impl(void *ptr, size_t new_size, const char *file, int line,
+	      const char *func)
+{
+  void *new_ptr;
+
+  xmalloc_init();
+  assert(ptr != NULL);
+  assert(new_size > 0);
+
+  if (xmalloc_fail_after == 0)
+    {
+      xmalloc_fail_after = -2;
+      return NULL;
+    }
+  else if (xmalloc_fail_after == -2)
+    {
+      printf("xrealloc: called after failure from %s:%d: %s\n",
+	     file, line, func);
+      assert(0);
+    }
+  else if (xmalloc_fail_after > 0)
+    xmalloc_fail_after--;
+
+  new_ptr = realloc(ptr, new_size);
+  if (new_ptr != NULL && new_ptr != ptr)
+    {
+      hash_table_del(xmalloc_table, ptr);
+      hash_table_add(xmalloc_table, new_ptr, (int)new_size, file, line, func);
+    }
+  return new_ptr;
+}
+
+
+
+/* EOF */

vendored/tre/xmalloc.h

@@ -0,0 +1,77 @@
+/*
+  xmalloc.h - Simple malloc debugging library API
+
+  This software is released under a BSD-style license.
+  See the file LICENSE for details and copyright.
+
+*/
+
+#ifndef _XMALLOC_H
+#define _XMALLOC_H 1
+
+void *xmalloc_impl(size_t size, const char *file, int line, const char *func);
+void *xcalloc_impl(size_t nmemb, size_t size, const char *file, int line,
+		   const char *func);
+void xfree_impl(void *ptr, const char *file, int line, const char *func);
+void *xrealloc_impl(void *ptr, size_t new_size, const char *file, int line,
+		    const char *func);
+int xmalloc_dump_leaks(void);
+void xmalloc_configure(int fail_after);
+
+
+#ifndef XMALLOC_INTERNAL
+#ifdef MALLOC_DEBUGGING
+
+/* Version 2.4 and later of GCC define a magical variable `__PRETTY_FUNCTION__'
+   which contains the name of the function currently being defined.
+#  define __XMALLOC_FUNCTION	 __PRETTY_FUNCTION__
+   This is broken in G++ before version 2.6.
+   C9x has a similar variable called __func__, but prefer the GCC one since
+   it demangles C++ function names.  */
+# ifdef __GNUC__
+#  if __GNUC__ > 2 || (__GNUC__ == 2 \
+		       && __GNUC_MINOR__ >= (defined __cplusplus ? 6 : 4))
+#   define __XMALLOC_FUNCTION	 __PRETTY_FUNCTION__
+#  else
+#   define __XMALLOC_FUNCTION	 ((const char *) 0)
+#  endif
+# else
+#  if defined __STDC_VERSION__ && __STDC_VERSION__ >= 199901L
+#   define __XMALLOC_FUNCTION	 __func__
+#  else
+#   define __XMALLOC_FUNCTION	 ((const char *) 0)
+#  endif
+# endif
+
+#define xmalloc(size) xmalloc_impl(size, __FILE__, __LINE__, \
+				   __XMALLOC_FUNCTION)
+#define xcalloc(nmemb, size) xcalloc_impl(nmemb, size, __FILE__, __LINE__, \
+					  __XMALLOC_FUNCTION)
+#define xfree(ptr) xfree_impl(ptr, __FILE__, __LINE__, __XMALLOC_FUNCTION)
+#define xrealloc(ptr, new_size) xrealloc_impl(ptr, new_size, __FILE__, \
+					      __LINE__, __XMALLOC_FUNCTION)
+#undef malloc
+#undef calloc
+#undef free
+#undef realloc
+
+#define malloc	USE_XMALLOC_INSTEAD_OF_MALLOC
+#define calloc	USE_XCALLOC_INSTEAD_OF_CALLOC
+#define free	USE_XFREE_INSTEAD_OF_FREE
+#define realloc USE_XREALLOC_INSTEAD_OF_REALLOC
+
+#else /* !MALLOC_DEBUGGING */
+
+#include <stdlib.h>
+
+#define xmalloc(size) malloc(size)
+#define xcalloc(nmemb, size) calloc(nmemb, size)
+#define xfree(ptr) free(ptr)
+#define xrealloc(ptr, new_size) realloc(ptr, new_size)
+
+#endif /* !MALLOC_DEBUGGING */
+#endif /* !XMALLOC_INTERNAL */
+
+#endif /* _XMALLOC_H */
+
+/* EOF */