regcomp.c

Go to the documentation of this file.

/* Extended regular expression matching and search library.
   Copyright (C) 2002, 2003 Free Software Foundation, Inc.
   This file is part of the GNU C Library.
   Contributed by Isamu Hasegawa <isamu@yamato.ibm.com>.

   The GNU C Library is free software; you can redistribute it and/or
   modify it under the terms of the GNU Lesser General Public
   License as published by the Free Software Foundation; either
   version 2.1 of the License, or (at your option) any later version.

   The GNU C Library is distributed in the hope that it will be useful,
   but WITHOUT ANY WARRANTY; without even the implied warranty of
   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
   Lesser General Public License for more details.

   You should have received a copy of the GNU Lesser General Public
   License along with the GNU C Library; if not, write to the Free
   Software Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA
   02111-1307 USA.  */

static reg_errcode_t re_compile_internal _RE_ARGS((regex_t *preg, const char * pattern,
                                          int length, reg_syntax_t syntax));
static void re_compile_fastmap_iter _RE_ARGS((regex_t *bufp,
                                     const re_dfastate_t *init_state,
                                     char *fastmap));
static reg_errcode_t init_dfa _RE_ARGS((re_dfa_t *dfa, int pat_len));
static reg_errcode_t init_word_char _RE_ARGS((re_dfa_t *dfa));
#ifdef RE_ENABLE_I18N
static void free_charset _RE_ARGS((re_charset_t *cset));
#endif /* RE_ENABLE_I18N */
static void free_workarea_compile _RE_ARGS((regex_t *preg));
static reg_errcode_t create_initial_state _RE_ARGS((re_dfa_t *dfa));
static reg_errcode_t analyze _RE_ARGS((re_dfa_t *dfa));
static reg_errcode_t analyze_tree _RE_ARGS((re_dfa_t *dfa, bin_tree_t *node));
static void calc_first _RE_ARGS((re_dfa_t *dfa, bin_tree_t *node));
static void calc_next _RE_ARGS((re_dfa_t *dfa, bin_tree_t *node));
static void calc_epsdest _RE_ARGS((re_dfa_t *dfa, bin_tree_t *node));
static reg_errcode_t duplicate_node_closure _RE_ARGS((re_dfa_t *dfa, int top_org_node,
                                             int top_clone_node, int root_node,
                                             unsigned int constraint));
static reg_errcode_t duplicate_node _RE_ARGS((int *new_idx, re_dfa_t *dfa, int org_idx,
                                     unsigned int constraint));
static int search_duplicated_node _RE_ARGS((re_dfa_t *dfa, int org_node,
                                   unsigned int constraint));
static reg_errcode_t calc_eclosure _RE_ARGS((re_dfa_t *dfa));
static reg_errcode_t calc_eclosure_iter _RE_ARGS((re_node_set *new_set, re_dfa_t *dfa,
                                         int node, int root));
static void calc_inveclosure _RE_ARGS((re_dfa_t *dfa));
static int fetch_number _RE_ARGS((re_string_t *input, re_token_t *token,
                         reg_syntax_t syntax));
static re_token_t fetch_token _RE_ARGS((re_string_t *input, reg_syntax_t syntax));
static int peek_token _RE_ARGS((re_token_t *token, re_string_t *input,
                        reg_syntax_t syntax));
static int peek_token_bracket _RE_ARGS((re_token_t *token, re_string_t *input,
                               reg_syntax_t syntax));
static bin_tree_t *parse _RE_ARGS((re_string_t *regexp, regex_t *preg,
                          reg_syntax_t syntax, reg_errcode_t *err));
static bin_tree_t *parse_reg_exp _RE_ARGS((re_string_t *regexp, regex_t *preg,
                                  re_token_t *token, reg_syntax_t syntax,
                                  int nest, reg_errcode_t *err));
static bin_tree_t *parse_branch _RE_ARGS((re_string_t *regexp, regex_t *preg,
                                 re_token_t *token, reg_syntax_t syntax,
                                 int nest, reg_errcode_t *err));
static bin_tree_t *parse_expression _RE_ARGS((re_string_t *regexp, regex_t *preg,
                                     re_token_t *token, reg_syntax_t syntax,
                                     int nest, reg_errcode_t *err));
static bin_tree_t *parse_sub_exp _RE_ARGS((re_string_t *regexp, regex_t *preg,
                                  re_token_t *token, reg_syntax_t syntax,
                                  int nest, reg_errcode_t *err));
static bin_tree_t *parse_dup_op _RE_ARGS((bin_tree_t *dup_elem, re_string_t *regexp,
                                 re_dfa_t *dfa, re_token_t *token,
                                 reg_syntax_t syntax, reg_errcode_t *err));
static bin_tree_t *parse_bracket_exp _RE_ARGS((re_string_t *regexp, re_dfa_t *dfa,
                                      re_token_t *token, reg_syntax_t syntax,
                                      reg_errcode_t *err));
static reg_errcode_t parse_bracket_element _RE_ARGS((bracket_elem_t *elem,
                                            re_string_t *regexp,
                                            re_token_t *token, int token_len,
                                            re_dfa_t *dfa,
                                            reg_syntax_t syntax));
static reg_errcode_t parse_bracket_symbol _RE_ARGS((bracket_elem_t *elem,
                                          re_string_t *regexp,
                                          re_token_t *token));
#ifndef _LIBC
# ifdef RE_ENABLE_I18N
static reg_errcode_t build_range_exp _RE_ARGS((re_bitset_ptr_t sbcset,
                                      re_charset_t *mbcset, int *range_alloc,
                                      bracket_elem_t *start_elem,
                                      bracket_elem_t *end_elem));
static reg_errcode_t build_collating_symbol _RE_ARGS((re_bitset_ptr_t sbcset,
                                             re_charset_t *mbcset,
                                             int *coll_sym_alloc,
                                             const unsigned char *name));
# else /* not RE_ENABLE_I18N */
static reg_errcode_t build_range_exp _RE_ARGS((re_bitset_ptr_t sbcset,
                                      bracket_elem_t *start_elem,
                                      bracket_elem_t *end_elem));
static reg_errcode_t build_collating_symbol _RE_ARGS((re_bitset_ptr_t sbcset,
                                             const unsigned char *name));
# endif /* not RE_ENABLE_I18N */
#endif /* not _LIBC */
#ifdef RE_ENABLE_I18N
static reg_errcode_t build_equiv_class _RE_ARGS((re_bitset_ptr_t sbcset,
                                        re_charset_t *mbcset,
                                        int *equiv_class_alloc,
                                        const unsigned char *name));
static reg_errcode_t build_charclass _RE_ARGS((RE_TRANSLATE_TYPE trans,
                                      re_bitset_ptr_t sbcset,
                                      re_charset_t *mbcset,
                                      int *char_class_alloc,
                                      const unsigned char *class_name,
                                      reg_syntax_t syntax));
#else  /* not RE_ENABLE_I18N */
static reg_errcode_t build_equiv_class _RE_ARGS((re_bitset_ptr_t sbcset,
                                        const unsigned char *name));
static reg_errcode_t build_charclass _RE_ARGS((RE_TRANSLATE_TYPE trans,
                                      re_bitset_ptr_t sbcset,
                                      const unsigned char *class_name,
                                      reg_syntax_t syntax));
#endif /* not RE_ENABLE_I18N */
static bin_tree_t *build_word_op _RE_ARGS((re_dfa_t *dfa, RE_TRANSLATE_TYPE trans,
                                  int not, reg_errcode_t *err));
static void free_bin_tree _RE_ARGS((bin_tree_t *tree));
static bin_tree_t *create_tree _RE_ARGS((bin_tree_t *left, bin_tree_t *right,
                                re_token_type_t type, int index));
static bin_tree_t *duplicate_tree _RE_ARGS((const bin_tree_t *src, re_dfa_t *dfa));

/* This table gives an error message for each of the error codes listed
   in regex.h.  Obviously the order here has to be same as there.
   POSIX doesn't require that we do anything for REG_NOERROR,
   but why not be nice?  */

const char __re_error_msgid[] attribute_hidden =
  {
#define REG_NOERROR_IDX 0
    gettext_noop ("Success")    /* REG_NOERROR */
    "\0"
#define REG_NOMATCH_IDX (REG_NOERROR_IDX + sizeof "Success")
    gettext_noop ("No match")   /* REG_NOMATCH */
    "\0"
#define REG_BADPAT_IDX  (REG_NOMATCH_IDX + sizeof "No match")
    gettext_noop ("Invalid regular expression") /* REG_BADPAT */
    "\0"
#define REG_ECOLLATE_IDX (REG_BADPAT_IDX + sizeof "Invalid regular expression")
    gettext_noop ("Invalid collation character") /* REG_ECOLLATE */
    "\0"
#define REG_ECTYPE_IDX  (REG_ECOLLATE_IDX + sizeof "Invalid collation character")
    gettext_noop ("Invalid character class name") /* REG_ECTYPE */
    "\0"
#define REG_EESCAPE_IDX (REG_ECTYPE_IDX + sizeof "Invalid character class name")
    gettext_noop ("Trailing backslash") /* REG_EESCAPE */
    "\0"
#define REG_ESUBREG_IDX (REG_EESCAPE_IDX + sizeof "Trailing backslash")
    gettext_noop ("Invalid back reference") /* REG_ESUBREG */
    "\0"
#define REG_EBRACK_IDX  (REG_ESUBREG_IDX + sizeof "Invalid back reference")
    gettext_noop ("Unmatched [ or [^")  /* REG_EBRACK */
    "\0"
#define REG_EPAREN_IDX  (REG_EBRACK_IDX + sizeof "Unmatched [ or [^")
    gettext_noop ("Unmatched ( or \\(") /* REG_EPAREN */
    "\0"
#define REG_EBRACE_IDX  (REG_EPAREN_IDX + sizeof "Unmatched ( or \\(")
    gettext_noop ("Unmatched \\{") /* REG_EBRACE */
    "\0"
#define REG_BADBR_IDX   (REG_EBRACE_IDX + sizeof "Unmatched \\{")
    gettext_noop ("Invalid content of \\{\\}") /* REG_BADBR */
    "\0"
#define REG_ERANGE_IDX  (REG_BADBR_IDX + sizeof "Invalid content of \\{\\}")
    gettext_noop ("Invalid range end")  /* REG_ERANGE */
    "\0"
#define REG_ESPACE_IDX  (REG_ERANGE_IDX + sizeof "Invalid range end")
    gettext_noop ("Memory exhausted") /* REG_ESPACE */
    "\0"
#define REG_BADRPT_IDX  (REG_ESPACE_IDX + sizeof "Memory exhausted")
    gettext_noop ("Invalid preceding regular expression") /* REG_BADRPT */
    "\0"
#define REG_EEND_IDX    (REG_BADRPT_IDX + sizeof "Invalid preceding regular expression")
    gettext_noop ("Premature end of regular expression") /* REG_EEND */
    "\0"
#define REG_ESIZE_IDX   (REG_EEND_IDX + sizeof "Premature end of regular expression")
    gettext_noop ("Regular expression too big") /* REG_ESIZE */
    "\0"
#define REG_ERPAREN_IDX (REG_ESIZE_IDX + sizeof "Regular expression too big")
    gettext_noop ("Unmatched ) or \\)") /* REG_ERPAREN */
  };

const size_t __re_error_msgid_idx[] attribute_hidden =
  {
    REG_NOERROR_IDX,
    REG_NOMATCH_IDX,
    REG_BADPAT_IDX,
    REG_ECOLLATE_IDX,
    REG_ECTYPE_IDX,
    REG_EESCAPE_IDX,
    REG_ESUBREG_IDX,
    REG_EBRACK_IDX,
    REG_EPAREN_IDX,
    REG_EBRACE_IDX,
    REG_BADBR_IDX,
    REG_ERANGE_IDX,
    REG_ESPACE_IDX,
    REG_BADRPT_IDX,
    REG_EEND_IDX,
    REG_ESIZE_IDX,
    REG_ERPAREN_IDX
  };

/* Entry points for GNU code.  */

/* re_compile_pattern is the GNU regular expression compiler: it
   compiles PATTERN (of length LENGTH) and puts the result in BUFP.
   Returns 0 if the pattern was valid, otherwise an error string.

   Assumes the `allocated' (and perhaps `buffer') and `translate' fields
   are set in BUFP on entry.  */

const char *
re_compile_pattern (pattern, length, bufp)
    const char *pattern;
    size_t length;
    struct re_pattern_buffer *bufp;
{
  reg_errcode_t ret;

  /* And GNU code determines whether or not to get register information
     by passing null for the REGS argument to re_match, etc., not by
     setting no_sub.  */
  bufp->no_sub = 0;

  /* Match anchors at newline.  */
  bufp->newline_anchor = 1;

  ret = re_compile_internal (bufp, pattern, length, re_syntax_options);

  if (!ret)
    return NULL;
  return gettext (__re_error_msgid + __re_error_msgid_idx[(int) ret]);
}
#ifdef _LIBC
weak_alias (__re_compile_pattern, re_compile_pattern)
#endif

/* Set by `re_set_syntax' to the current regexp syntax to recognize.  Can
   also be assigned to arbitrarily: each pattern buffer stores its own
   syntax, so it can be changed between regex compilations.  */
/* This has no initializer because initialized variables in Emacs
   become read-only after dumping.  */
reg_syntax_t re_syntax_options;


/* Specify the precise syntax of regexps for compilation.  This provides
   for compatibility for various utilities which historically have
   different, incompatible syntaxes.

   The argument SYNTAX is a bit mask comprised of the various bits
   defined in regex.h.  We return the old syntax.  */

reg_syntax_t
re_set_syntax (syntax)
    reg_syntax_t syntax;
{
  reg_syntax_t ret = re_syntax_options;

  re_syntax_options = syntax;
#ifdef RE_ENABLE_I18N
  re_mb_cur_max = MB_CUR_MAX;
#endif
  return ret;
}
#ifdef _LIBC
weak_alias (__re_set_syntax, re_set_syntax)
#endif

int
re_compile_fastmap (bufp)
    struct re_pattern_buffer *bufp;
{
  re_dfa_t *dfa = (re_dfa_t *) bufp->buffer;
  char *fastmap = bufp->fastmap;

  memset (fastmap, '\0', sizeof (char) * SBC_MAX);
  re_compile_fastmap_iter (bufp, dfa->init_state, fastmap);
  if (dfa->init_state != dfa->init_state_word)
    re_compile_fastmap_iter (bufp, dfa->init_state_word, fastmap);
  if (dfa->init_state != dfa->init_state_nl)
    re_compile_fastmap_iter (bufp, dfa->init_state_nl, fastmap);
  if (dfa->init_state != dfa->init_state_begbuf)
    re_compile_fastmap_iter (bufp, dfa->init_state_begbuf, fastmap);
  bufp->fastmap_accurate = 1;
  return 0;
}
#ifdef _LIBC
weak_alias (__re_compile_fastmap, re_compile_fastmap)
#endif

static inline void
re_set_fastmap (char *fastmap, int icase, int ch)
{
  fastmap[ch] = 1;
  if (icase)
    fastmap[tolower (ch)] = 1;
}

/* Helper function for re_compile_fastmap.
   Compile fastmap for the initial_state INIT_STATE.  */

static void
re_compile_fastmap_iter (bufp, init_state, fastmap)
     regex_t *bufp;
     const re_dfastate_t *init_state;
     char *fastmap;
{
  re_dfa_t *dfa = (re_dfa_t *) bufp->buffer;
  int node_cnt;
#ifdef RE_ENABLE_I18N
  int icase = (re_mb_cur_max == 1 && (bufp->syntax & RE_ICASE));
#else
  int icase = (bufp->syntax & RE_ICASE);
#endif
  for (node_cnt = 0; node_cnt < init_state->nodes.nelem; ++node_cnt)
    {
      int node = init_state->nodes.elems[node_cnt];
      re_token_type_t type = dfa->nodes[node].type;

      if (type == CHARACTER)
        re_set_fastmap (fastmap, icase, dfa->nodes[node].opr.c);
      else if (type == SIMPLE_BRACKET)
        {
          int i, j, ch;
          for (i = 0, ch = 0; i < BITSET_UINTS; ++i)
            for (j = 0; j < UINT_BITS; ++j, ++ch)
              if (dfa->nodes[node].opr.sbcset[i] & (1UL << j))
                re_set_fastmap (fastmap, icase, ch);
        }
#ifdef RE_ENABLE_I18N
      else if (type == COMPLEX_BRACKET)
        {
          int i;
          re_charset_t *cset = dfa->nodes[node].opr.mbcset;
          if (cset->non_match || cset->ncoll_syms || cset->nequiv_classes
              || cset->nranges || cset->nchar_classes)
            {
# ifdef _LIBC
              if (_NL_CURRENT_WORD (LC_COLLATE, _NL_COLLATE_NRULES) != 0)
                {
                  /* In this case we want to catch the bytes which are
                     the first byte of any collation elements.
                     e.g. In da_DK, we want to catch 'a' since "aa"
                          is a valid collation element, and don't catch
                          'b' since 'b' is the only collation element
                          which starts from 'b'.  */
                  int j, ch;
                  const int32_t *table = (const int32_t *)
                    _NL_CURRENT (LC_COLLATE, _NL_COLLATE_TABLEMB);
                  for (i = 0, ch = 0; i < BITSET_UINTS; ++i)
                    for (j = 0; j < UINT_BITS; ++j, ++ch)
                      if (table[ch] < 0)
                        re_set_fastmap (fastmap, icase, ch);
                }
# else
              if (re_mb_cur_max > 1)
                for (i = 0; i < SBC_MAX; ++i)
                  if (__btowc (i) == WEOF)
                    re_set_fastmap (fastmap, icase, i);
# endif /* not _LIBC */
            }
          for (i = 0; i < cset->nmbchars; ++i)
            {
              char buf[256];
              mbstate_t state;
              memset (&state, '\0', sizeof (state));
              __wcrtomb (buf, cset->mbchars[i], &state);
              re_set_fastmap (fastmap, icase, *(unsigned char *) buf);
            }
        }
#endif /* RE_ENABLE_I18N */
      else if (type == END_OF_RE || type == OP_PERIOD)
        {
          memset (fastmap, '\1', sizeof (char) * SBC_MAX);
          if (type == END_OF_RE)
            bufp->can_be_null = 1;
          return;
        }
    }
}

/* Entry point for POSIX code.  */
/* regcomp takes a regular expression as a string and compiles it.

   PREG is a regex_t *.  We do not expect any fields to be initialized,
   since POSIX says we shouldn't.  Thus, we set

     `buffer' to the compiled pattern;
     `used' to the length of the compiled pattern;
     `syntax' to RE_SYNTAX_POSIX_EXTENDED if the
       REG_EXTENDED bit in CFLAGS is set; otherwise, to
       RE_SYNTAX_POSIX_BASIC;
     `newline_anchor' to REG_NEWLINE being set in CFLAGS;
     `fastmap' to an allocated space for the fastmap;
     `fastmap_accurate' to zero;
     `re_nsub' to the number of subexpressions in PATTERN.

   PATTERN is the address of the pattern string.

   CFLAGS is a series of bits which affect compilation.

     If REG_EXTENDED is set, we use POSIX extended syntax; otherwise, we
     use POSIX basic syntax.

     If REG_NEWLINE is set, then . and [^...] don't match newline.
     Also, regexec will try a match beginning after every newline.

     If REG_ICASE is set, then we considers upper- and lowercase
     versions of letters to be equivalent when matching.

     If REG_NOSUB is set, then when PREG is passed to regexec, that
     routine will report only success or failure, and nothing about the
     registers.

   It returns 0 if it succeeds, nonzero if it doesn't.  (See regex.h for
   the return codes and their meanings.)  */

int
regcomp (preg, pattern, cflags)
    regex_t *__restrict preg;
    const char *__restrict pattern;
    int cflags;
{
  reg_errcode_t ret;
  reg_syntax_t syntax = ((cflags & REG_EXTENDED) ? RE_SYNTAX_POSIX_EXTENDED
                         : RE_SYNTAX_POSIX_BASIC);

  preg->buffer = NULL;
  preg->allocated = 0;
  preg->used = 0;

  /* Try to allocate space for the fastmap.  */
  preg->fastmap = re_malloc (char, SBC_MAX);
  if (BE (preg->fastmap == NULL, 0))
    return REG_ESPACE;

  syntax |= (cflags & REG_ICASE) ? RE_ICASE : 0;

  /* If REG_NEWLINE is set, newlines are treated differently.  */
  if (cflags & REG_NEWLINE)
    { /* REG_NEWLINE implies neither . nor [^...] match newline.  */
      syntax &= ~RE_DOT_NEWLINE;
      syntax |= RE_HAT_LISTS_NOT_NEWLINE;
      /* It also changes the matching behavior.  */
      preg->newline_anchor = 1;
    }
  else
    preg->newline_anchor = 0;
  preg->no_sub = !!(cflags & REG_NOSUB);
  preg->translate = NULL;

  ret = re_compile_internal (preg, pattern, strlen (pattern), syntax);

  /* POSIX doesn't distinguish between an unmatched open-group and an
     unmatched close-group: both are REG_EPAREN.  */
  if (ret == REG_ERPAREN)
    ret = REG_EPAREN;

  /* We have already checked preg->fastmap != NULL.  */
  if (BE (ret == REG_NOERROR, 1))
    /* Compute the fastmap now, since regexec cannot modify the pattern
       buffer.  This function nevers fails in this implementation.  */
    (void) re_compile_fastmap (preg);
  else
    {
      /* Some error occurred while compiling the expression.  */
      re_free (preg->fastmap);
      preg->fastmap = NULL;
    }

  return (int) ret;
}
#ifdef _LIBC
weak_alias (__regcomp, regcomp)
#endif

/* Returns a message corresponding to an error code, ERRCODE, returned
   from either regcomp or regexec.   We don't use PREG here.  */

size_t
regerror (errcode, preg, errbuf, errbuf_size)
    int errcode;
    const regex_t *preg;
    char *errbuf;
    size_t errbuf_size;
{
  const char *msg;
  size_t msg_size;

  if (BE (errcode < 0
          || errcode >= (int) (sizeof (__re_error_msgid_idx)
                               / sizeof (__re_error_msgid_idx[0])), 0))
    /* Only error codes returned by the rest of the code should be passed
       to this routine.  If we are given anything else, or if other regex
       code generates an invalid error code, then the program has a bug.
       Dump core so we can fix it.  */
    abort ();

  msg = gettext (__re_error_msgid + __re_error_msgid_idx[errcode]);

  msg_size = strlen (msg) + 1; /* Includes the null.  */

  if (BE (errbuf_size != 0, 1))
    {
      if (BE (msg_size > errbuf_size, 0))
        {
          memcpy (errbuf, msg, errbuf_size - 1);
          errbuf[errbuf_size - 1] = 0;
        }
      else
        memcpy (errbuf, msg, msg_size);
    }

  return msg_size;
}
#ifdef _LIBC
weak_alias (__regerror, regerror)
#endif


static void
free_dfa_content (re_dfa_t *dfa)
{
  int i, j;

  re_free (dfa->subexps);

  for (i = 0; i < dfa->nodes_len; ++i)
    {
      re_token_t *node = dfa->nodes + i;
#ifdef RE_ENABLE_I18N
      if (node->type == COMPLEX_BRACKET && node->duplicated == 0)
        free_charset (node->opr.mbcset);
      else
#endif /* RE_ENABLE_I18N */
        if (node->type == SIMPLE_BRACKET && node->duplicated == 0)
          re_free (node->opr.sbcset);
    }
  re_free (dfa->nexts);
  for (i = 0; i < dfa->nodes_len; ++i)
    {
      if (dfa->eclosures != NULL)
        re_node_set_free (dfa->eclosures + i);
      if (dfa->inveclosures != NULL)
        re_node_set_free (dfa->inveclosures + i);
      if (dfa->edests != NULL)
        re_node_set_free (dfa->edests + i);
    }
  re_free (dfa->edests);
  re_free (dfa->eclosures);
  re_free (dfa->inveclosures);
  re_free (dfa->nodes);

  for (i = 0; i <= dfa->state_hash_mask; ++i)
    {
      struct re_state_table_entry *entry = dfa->state_table + i;
      for (j = 0; j < entry->num; ++j)
        {
          re_dfastate_t *state = entry->array[j];
          free_state (state);
        }
      re_free (entry->array);
    }
  re_free (dfa->state_table);

  if (dfa->word_char != NULL)
    re_free (dfa->word_char);
#ifdef DEBUG
  re_free (dfa->re_str);
#endif

  re_free (dfa);
}


/* Free dynamically allocated space used by PREG.  */

void
regfree (preg)
    regex_t *preg;
{
  re_dfa_t *dfa = (re_dfa_t *) preg->buffer;
  if (BE (dfa != NULL, 1))
    free_dfa_content (dfa);

  re_free (preg->fastmap);
}
#ifdef _LIBC
weak_alias (__regfree, regfree)
#endif

/* Entry points compatible with 4.2 BSD regex library.  We don't define
   them unless specifically requested.  */

#if defined _REGEX_RE_COMP || defined _LIBC

/* BSD has one and only one pattern buffer.  */
static struct re_pattern_buffer re_comp_buf;

char *
# ifdef _LIBC
/* Make these definitions weak in libc, so POSIX programs can redefine
   these names if they don't use our functions, and still use
   regcomp/regexec above without link errors.  */
weak_function
# endif
re_comp (s)
     const char *s;
{
  reg_errcode_t ret;
  char *fastmap;

  if (!s)
    {
      if (!re_comp_buf.buffer)
        return gettext ("No previous regular expression");
      return 0;
    }

  if (re_comp_buf.buffer)
    {
      fastmap = re_comp_buf.fastmap;
      re_comp_buf.fastmap = NULL;
      __regfree (&re_comp_buf);
      memset (&re_comp_buf, '\0', sizeof (re_comp_buf));
      re_comp_buf.fastmap = fastmap;
    }

  if (re_comp_buf.fastmap == NULL)
    {
      re_comp_buf.fastmap = (char *) malloc (SBC_MAX);
      if (re_comp_buf.fastmap == NULL)
        return (char *) gettext (__re_error_msgid
                                 + __re_error_msgid_idx[(int) REG_ESPACE]);
    }

  /* Since `re_exec' always passes NULL for the `regs' argument, we
     don't need to initialize the pattern buffer fields which affect it.  */

  /* Match anchors at newlines.  */
  re_comp_buf.newline_anchor = 1;

  ret = re_compile_internal (&re_comp_buf, s, strlen (s), re_syntax_options);

  if (!ret)
    return NULL;

  /* Yes, we're discarding `const' here if !HAVE_LIBINTL.  */
  return (char *) gettext (__re_error_msgid + __re_error_msgid_idx[(int) ret]);
}

#ifdef _LIBC
libc_freeres_fn (free_mem)
{
  __regfree (&re_comp_buf);
}
#endif

#endif /* _REGEX_RE_COMP */

/* Internal entry point.
   Compile the regular expression PATTERN, whose length is LENGTH.
   SYNTAX indicate regular expression's syntax.  */

static reg_errcode_t
re_compile_internal (preg, pattern, length, syntax)
     regex_t *preg;
     const char * pattern;
     int length;
     reg_syntax_t syntax;
{
  reg_errcode_t err = REG_NOERROR;
  re_dfa_t *dfa;
  re_string_t regexp;

  /* Initialize the pattern buffer.  */
  preg->fastmap_accurate = 0;
  preg->syntax = syntax;
  preg->not_bol = preg->not_eol = 0;
  preg->used = 0;
  preg->re_nsub = 0;
  preg->can_be_null = 0;
  preg->regs_allocated = REGS_UNALLOCATED;

  /* Initialize the dfa.  */
  dfa = (re_dfa_t *) preg->buffer;
  if (preg->allocated < sizeof (re_dfa_t))
    {
      /* If zero allocated, but buffer is non-null, try to realloc
         enough space.  This loses if buffer's address is bogus, but
         that is the user's responsibility.  If ->buffer is NULL this
         is a simple allocation.  */
      dfa = re_realloc (preg->buffer, re_dfa_t, 1);
      if (dfa == NULL)
        return REG_ESPACE;
      preg->allocated = sizeof (re_dfa_t);
    }
  preg->buffer = (unsigned char *) dfa;
  preg->used = sizeof (re_dfa_t);

  err = init_dfa (dfa, length);
  if (BE (err != REG_NOERROR, 0))
    {
      re_free (dfa);
      preg->buffer = NULL;
      preg->allocated = 0;
      return err;
    }
#ifdef DEBUG
  dfa->re_str = re_malloc (char, length + 1);
  strncpy (dfa->re_str, pattern, length + 1);
#endif

  err = re_string_construct (&regexp, pattern, length, preg->translate,
                             syntax & RE_ICASE);
  if (BE (err != REG_NOERROR, 0))
    {
      re_free (dfa);
      preg->buffer = NULL;
      preg->allocated = 0;
      return err;
    }

  /* Parse the regular expression, and build a structure tree.  */
  preg->re_nsub = 0;
  dfa->str_tree = parse (&regexp, preg, syntax, &err);
  if (BE (dfa->str_tree == NULL, 0))
    goto re_compile_internal_free_return;

  /* Analyze the tree and collect information which is necessary to
     create the dfa.  */
  err = analyze (dfa);
  if (BE (err != REG_NOERROR, 0))
    goto re_compile_internal_free_return;

  /* Then create the initial state of the dfa.  */
  err = create_initial_state (dfa);

  /* Release work areas.  */
  free_workarea_compile (preg);
  re_string_destruct (&regexp);

  if (BE (err != REG_NOERROR, 0))
    {
    re_compile_internal_free_return:
      free_dfa_content (dfa);
      preg->buffer = NULL;
      preg->allocated = 0;
    }

  return err;
}

/* Initialize DFA.  We use the length of the regular expression PAT_LEN
   as the initial length of some arrays.  */

static reg_errcode_t
init_dfa (dfa, pat_len)
     re_dfa_t *dfa;
     int pat_len;
{
  int table_size;

  memset (dfa, '\0', sizeof (re_dfa_t));

  dfa->nodes_alloc = pat_len + 1;
  dfa->nodes = re_malloc (re_token_t, dfa->nodes_alloc);

  dfa->states_alloc = pat_len + 1;

  /*  table_size = 2 ^ ceil(log pat_len) */
  for (table_size = 1; table_size > 0; table_size <<= 1)
    if (table_size > pat_len)
      break;

  dfa->state_table = calloc (sizeof (struct re_state_table_entry), table_size);
  dfa->state_hash_mask = table_size - 1;

  dfa->subexps_alloc = 1;
  dfa->subexps = re_malloc (re_subexp_t, dfa->subexps_alloc);
  dfa->word_char = NULL;

  if (BE (dfa->nodes == NULL || dfa->state_table == NULL
          || dfa->subexps == NULL, 0))
    {
      /* We don't bother to free anything which was allocated.  Very
         soon the process will go down anyway.  */
      dfa->subexps = NULL;
      dfa->state_table = NULL;
      dfa->nodes = NULL;
      return REG_ESPACE;
    }
  return REG_NOERROR;
}

/* Initialize WORD_CHAR table, which indicate which character is
   "word".  In this case "word" means that it is the word construction
   character used by some operators like "<", ">", etc.  */

static reg_errcode_t
init_word_char (dfa)
     re_dfa_t *dfa;
{
  int i, j, ch;
  dfa->word_char = (re_bitset_ptr_t) calloc (sizeof (bitset), 1);
  if (BE (dfa->word_char == NULL, 0))
    return REG_ESPACE;
  for (i = 0, ch = 0; i < BITSET_UINTS; ++i)
    for (j = 0; j < UINT_BITS; ++j, ++ch)
      if (isalnum (ch) || ch == '_')
        dfa->word_char[i] |= 1UL << j;
  return REG_NOERROR;
}

/* Free the work area which are only used while compiling.  */

static void
free_workarea_compile (preg)
     regex_t *preg;
{
  re_dfa_t *dfa = (re_dfa_t *) preg->buffer;
  free_bin_tree (dfa->str_tree);
  dfa->str_tree = NULL;
  re_free (dfa->org_indices);
  dfa->org_indices = NULL;
}

/* Create initial states for all contexts.  */

static reg_errcode_t
create_initial_state (dfa)
     re_dfa_t *dfa;
{
  int first, i;
  reg_errcode_t err;
  re_node_set init_nodes;

  /* Initial states have the epsilon closure of the node which is
     the first node of the regular expression.  */
  first = dfa->str_tree->first;
  dfa->init_node = first;
  err = re_node_set_init_copy (&init_nodes, dfa->eclosures + first);
  if (BE (err != REG_NOERROR, 0))
    return err;

  /* The back-references which are in initial states can epsilon transit,
     since in this case all of the subexpressions can be null.
     Then we add epsilon closures of the nodes which are the next nodes of
     the back-references.  */
  if (dfa->nbackref > 0)
    for (i = 0; i < init_nodes.nelem; ++i)
      {
        int node_idx = init_nodes.elems[i];
        re_token_type_t type = dfa->nodes[node_idx].type;

        int clexp_idx;
        if (type != OP_BACK_REF)
          continue;
        for (clexp_idx = 0; clexp_idx < init_nodes.nelem; ++clexp_idx)
          {
            re_token_t *clexp_node;
            clexp_node = dfa->nodes + init_nodes.elems[clexp_idx];
            if (clexp_node->type == OP_CLOSE_SUBEXP
                && clexp_node->opr.idx + 1 == dfa->nodes[node_idx].opr.idx)
              break;
          }
        if (clexp_idx == init_nodes.nelem)
          continue;

        if (type == OP_BACK_REF)
          {
            int dest_idx = dfa->edests[node_idx].elems[0];
            if (!re_node_set_contains (&init_nodes, dest_idx))
              {
                re_node_set_merge (&init_nodes, dfa->eclosures + dest_idx);
                i = 0;
              }
          }
      }

  /* It must be the first time to invoke acquire_state.  */
  dfa->init_state = re_acquire_state_context (&err, dfa, &init_nodes, 0);
  /* We don't check ERR here, since the initial state must not be NULL.  */
  if (BE (dfa->init_state == NULL, 0))
    return err;
  if (dfa->init_state->has_constraint)
    {
      dfa->init_state_word = re_acquire_state_context (&err, dfa, &init_nodes,
                                                       CONTEXT_WORD);
      dfa->init_state_nl = re_acquire_state_context (&err, dfa, &init_nodes,
                                                     CONTEXT_NEWLINE);
      dfa->init_state_begbuf = re_acquire_state_context (&err, dfa,
                                                         &init_nodes,
                                                         CONTEXT_NEWLINE
                                                         | CONTEXT_BEGBUF);
      if (BE (dfa->init_state_word == NULL || dfa->init_state_nl == NULL
              || dfa->init_state_begbuf == NULL, 0))
        return err;
    }
  else
    dfa->init_state_word = dfa->init_state_nl
      = dfa->init_state_begbuf = dfa->init_state;

  re_node_set_free (&init_nodes);
  return REG_NOERROR;
}

/* Analyze the structure tree, and calculate "first", "next", "edest",
   "eclosure", and "inveclosure".  */

static reg_errcode_t
analyze (dfa)
     re_dfa_t *dfa;
{
  int i;
  reg_errcode_t ret;

  /* Allocate arrays.  */
  dfa->nexts = re_malloc (int, dfa->nodes_alloc);
  dfa->org_indices = re_malloc (int, dfa->nodes_alloc);
  dfa->edests = re_malloc (re_node_set, dfa->nodes_alloc);
  dfa->eclosures = re_malloc (re_node_set, dfa->nodes_alloc);
  dfa->inveclosures = re_malloc (re_node_set, dfa->nodes_alloc);
  if (BE (dfa->nexts == NULL || dfa->org_indices == NULL || dfa->edests == NULL
          || dfa->eclosures == NULL || dfa->inveclosures == NULL, 0))
    return REG_ESPACE;
  /* Initialize them.  */
  for (i = 0; i < dfa->nodes_len; ++i)
    {
      dfa->nexts[i] = -1;
      re_node_set_init_empty (dfa->edests + i);
      re_node_set_init_empty (dfa->eclosures + i);
      re_node_set_init_empty (dfa->inveclosures + i);
    }

  ret = analyze_tree (dfa, dfa->str_tree);
  if (BE (ret == REG_NOERROR, 1))
    {
      ret = calc_eclosure (dfa);
      if (ret == REG_NOERROR)
        calc_inveclosure (dfa);
    }
  return ret;
}

/* Helper functions for analyze.
   This function calculate "first", "next", and "edest" for the subtree
   whose root is NODE.  */

static reg_errcode_t
analyze_tree (dfa, node)
     re_dfa_t *dfa;
     bin_tree_t *node;
{
  reg_errcode_t ret;
  if (node->first == -1)
    calc_first (dfa, node);
  if (node->next == -1)
    calc_next (dfa, node);
  if (node->eclosure.nelem == 0)
    calc_epsdest (dfa, node);
  /* Calculate "first" etc. for the left child.  */
  if (node->left != NULL)
    {
      ret = analyze_tree (dfa, node->left);
      if (BE (ret != REG_NOERROR, 0))
        return ret;
    }
  /* Calculate "first" etc. for the right child.  */
  if (node->right != NULL)
    {
      ret = analyze_tree (dfa, node->right);
      if (BE (ret != REG_NOERROR, 0))
        return ret;
    }
  return REG_NOERROR;
}

/* Calculate "first" for the node NODE.  */
static void
calc_first (dfa, node)
     re_dfa_t *dfa;
     bin_tree_t *node;
{
  int idx, type;
  idx = node->node_idx;
  type = (node->type == 0) ? dfa->nodes[idx].type : node->type;

  switch (type)
    {
#ifdef DEBUG
    case OP_OPEN_BRACKET:
    case OP_CLOSE_BRACKET:
    case OP_OPEN_DUP_NUM:
    case OP_CLOSE_DUP_NUM:
    case OP_NON_MATCH_LIST:
    case OP_OPEN_COLL_ELEM:
    case OP_CLOSE_COLL_ELEM:
    case OP_OPEN_EQUIV_CLASS:
    case OP_CLOSE_EQUIV_CLASS:
    case OP_OPEN_CHAR_CLASS:
    case OP_CLOSE_CHAR_CLASS:
      /* These must not be appeared here.  */
      assert (0);
#endif
    case END_OF_RE:
    case CHARACTER:
    case OP_PERIOD:
    case OP_DUP_ASTERISK:
    case OP_DUP_QUESTION:
#ifdef RE_ENABLE_I18N
    case COMPLEX_BRACKET:
#endif /* RE_ENABLE_I18N */
    case SIMPLE_BRACKET:
    case OP_BACK_REF:
    case ANCHOR:
    case OP_OPEN_SUBEXP:
    case OP_CLOSE_SUBEXP:
      node->first = idx;
      break;
    case OP_DUP_PLUS:
#ifdef DEBUG
      assert (node->left != NULL);
#endif
      if (node->left->first == -1)
        calc_first (dfa, node->left);
      node->first = node->left->first;
      break;
    case OP_ALT:
      node->first = idx;
      break;
      /* else fall through */
    default:
#ifdef DEBUG
      assert (node->left != NULL);
#endif
      if (node->left->first == -1)
        calc_first (dfa, node->left);
      node->first = node->left->first;
      break;
    }
}

/* Calculate "next" for the node NODE.  */

static void
calc_next (dfa, node)
     re_dfa_t *dfa;
     bin_tree_t *node;
{
  int idx, type;
  bin_tree_t *parent = node->parent;
  if (parent == NULL)
    {
      node->next = -1;
      idx = node->node_idx;
      if (node->type == 0)
        dfa->nexts[idx] = node->next;
      return;
    }

  idx = parent->node_idx;
  type = (parent->type == 0) ? dfa->nodes[idx].type : parent->type;

  switch (type)
    {
    case OP_DUP_ASTERISK:
    case OP_DUP_PLUS:
      node->next = idx;
      break;
    case CONCAT:
      if (parent->left == node)
        {
          if (parent->right->first == -1)
            calc_first (dfa, parent->right);
          node->next = parent->right->first;