/* regexpr.c
 *
 * Author: Tatu Ylonen <ylo@ngs.fi>
 *
 * Copyright (c) 1991 Tatu Ylonen, Espoo, Finland
 *
 * Permission to use, copy, modify, distribute, and sell this software
 * and its documentation for any purpose is hereby granted without
 * fee, provided that the above copyright notice appear in all copies.
 * This software is provided "as is" without express or implied
 * warranty.
 *
 * Created: Thu Sep 26 17:14:05 1991 ylo
 * Last modified: Mon Nov  4 17:06:48 1991 ylo
 * Ported to Think C: 19 Jan 1992 guido@cwi.nl
 *
 * This code draws many ideas from the regular expression packages by
 * Henry Spencer of the University of Toronto and Richard Stallman of
 * the Free Software Foundation.
 *
 * Emacs-specific code and syntax table code is almost directly borrowed
 * from GNU regexp.
 *
 * Bugs fixed and lots of reorganization by Jeffrey C. Ollie, April
 * 1997 Thanks for bug reports and ideas from Andrew Kuchling, Tim
 * Peters, Guido van Rossum, Ka-Ping Yee, Sjoerd Mullender, and
 * probably one or two others that I'm forgetting.
 *
 * $Id$ */

#include "config.h" /* For Win* specific redefinition of printf c.s. */

#include "myproto.h" /* For PROTO macro --Guido */

#include <stdio.h>

#ifndef NDEBUG
#define NDEBUG 1
#endif

#include <assert.h>
#include "regexpr.h"

#ifdef THINK_C
/* Think C on the Mac really needs these headers... --Guido */
#include <stdlib.h>
#include <string.h>
#else
#if defined(__STDC__) || defined(_MSC_VER)
/* Don't mess around, use the standard headers */
#include <stdlib.h>
#include <string.h>
#else
char *malloc();
void free();
char *realloc();
#endif /* __STDC__ */
#endif /* THINK_C */

/* The stack implementation is taken from an idea by Andrew Kuchling.
 * It's a doubly linked list of arrays. The advantages of this over a
 * simple linked list are that the number of mallocs required are
 * reduced. It also makes it possible to statically allocate enough
 * space so that small patterns don't ever need to call malloc.
 *
 * The advantages over a single array is that is periodically
 * realloced when more space is needed is that we avoid ever copying
 * the stack. */

/* item_t is the basic stack element.  Defined as a union of
 * structures so that both registers, failure points, and counters can
 * be pushed/popped from the stack.  There's nothing built into the
 * item to keep track of whether a certain stack item is a register, a
 * failure point, or a counter. */

typedef union item_t
{
      struct
      {
	    int num;
	    int level;
	    char *start;
	    char *end;
      } reg;
      struct
      {
	    int count;
	    int level;
	    int phantom;
	    char *code;
	    char *text;
      } fail;
      struct
      {
	    int num;
	    int level;
	    int count;
      } cntr;
} item_t;

#define STACK_PAGE_SIZE 256
#define NUM_REGISTERS 256

/* A 'page' of stack items. */

typedef struct item_page_t
{
      item_t items[STACK_PAGE_SIZE];
      struct item_page_t *prev;
      struct item_page_t *next;
} item_page_t;


typedef struct match_state
{
      /* Structure to encapsulate the stack. */
      struct
      {
	    /* index into the curent page.  If index == 0 and you need
	     * to pop and item, move to the previous page and set
	     * index = STACK_PAGE_SIZE - 1.  Otherwise decrement index
	     * to push a page. If index == STACK_PAGE_SIZE and you
	     * need to push a page move to the next page and set index
	     * = 0. If there is no new next page, allocate a new page
	     * and link it in. Otherwise, increment index to push a
	     * page. */
	    int index;
	    item_page_t *current; /* Pointer to the current page. */
	    item_page_t first; /* First page is statically allocated. */
      } stack;
      char *start[NUM_REGISTERS];
      char *end[NUM_REGISTERS];

      int changed[NUM_REGISTERS];
      /* The number of registers that have been pushed onto the stack
       * since the last failure point. */
      int count;
      /* Used to control when registers need to be pushed onto the
       * stack. */
      int level;
      /* The number of failure points on the stack. */
      int point;
} match_state;

/* Discard the top 'count' stack items. */

#define STACK_DISCARD(stack, count, on_error) \
stack.index -= count; \
while (stack.index < 0) \
{ \
   if (stack.current->prev == NULL) \
      on_error; \
   stack.current = stack.current->prev; \
   stack.index += STACK_PAGE_SIZE; \
}

/* Store a pointer to the previous item on the stack. Used to pop an
 * item off of the stack. */

#define STACK_PREV(stack, top, on_error) \
if (stack.index == 0) \
{ \
   if (stack.current->prev == NULL) \
      on_error; \
   stack.current = stack.current->prev; \
   stack.index = STACK_PAGE_SIZE - 1; \
} \
else \
   stack.index--; \
top = &(stack.current->items[stack.index])

/* Store a pointer to the next item on the stack. Used to push an item
 * on to the stack. */

#define STACK_NEXT(stack, top, on_error) \
if (stack.index == STACK_PAGE_SIZE) \
{ \
   if (stack.current->next == NULL) \
   { \
      stack.current->next = (item_page_t *)malloc(sizeof(item_page_t)); \
      if (stack.current->next == NULL) \
         on_error; \
      stack.current->next->prev = stack.current; \
      stack.current->next->next = NULL; \
   } \
   stack.current = stack.current->next; \
   stack.index = 0; \
} \
top = &(stack.current->items[stack.index++])

/* Store a pointer to the item that is 'count' items back in the
 * stack. STACK_BACK(stack, top, 1, on_error) is equivalent to
 * STACK_TOP(stack, top, on_error).  */

#define STACK_BACK(stack, top, count, on_error) \
{ \
   int index; \
   item_page_t *current; \
   current = stack.current; \
   index = stack.index - (count); \
   while (index < 0) \
   { \
      if (current->prev == NULL) \
	 on_error; \
      current = current->prev; \
      index += STACK_PAGE_SIZE; \
   } \
   top = &(current->items[index]); \
}

/* Store a pointer to the top item on the stack. Execute the
 * 'on_error' code if there are no items on the stack. */

#define STACK_TOP(stack, top, on_error) \
if (stack.index == 0) \
{ \
   if (stack.current->prev == NULL) \
      on_error; \
   top = &(stack.current->prev->items[STACK_PAGE_SIZE - 1]); \
} \
else \
   top = &(stack.current->items[stack.index - 1])

/* Test to see if the stack is empty */

#define STACK_EMPTY(stack) ((stack.index == 0) && \
			    (stack.current->prev == NULL))


/* Initialize a state object */

#define NEW_STATE(state) \
memset(&state, 0, sizeof(match_state)); \
state.stack.current = &state.stack.first; \
state.level = 1

/* Free any memory that might have been malloc'd */

#define FREE_STATE(state) \
while(state.stack.first.next != NULL) \
{ \
   state.stack.current = state.stack.first.next; \
   state.stack.first.next = state.stack.current->next; \
   free(state.stack.current); \
}

/* Return the start of register 'reg' */

#define GET_REG_START(state, reg) (state.start[reg])

/* Return the end of register 'reg' */

#define GET_REG_END(state, reg) (state.end[reg])

/* Set the start of register 'reg'. If the state of the register needs
 * saving, push it on the stack. */

#define SET_REG_START(state, reg, text, on_error) \
if(state.changed[reg] < state.level) \
{ \
   item_t *item; \
   STACK_NEXT(state.stack, item, on_error); \
   item->reg.num = reg; \
   item->reg.start = state.start[reg]; \
   item->reg.end = state.end[reg]; \
   item->reg.level = state.changed[reg]; \
   state.changed[reg] = state.level; \
   state.count++; \
} \
state.start[reg] = text

/* Set the end of register 'reg'. If the state of the register needs
 * saving, push it on the stack. */

#define SET_REG_END(state, reg, text, on_error) \
if(state.changed[reg] < state.level) \
{ \
   item_t *item; \
   STACK_NEXT(state.stack, item, on_error); \
   item->reg.num = reg; \
   item->reg.start = state.start[reg]; \
   item->reg.end = state.end[reg]; \
   item->reg.level = state.changed[reg]; \
   state.changed[reg] = state.level; \
   state.count++; \
} \
state.end[reg] = text

#define PUSH_FAILURE(state, xcode, xtext, on_error) \
{ \
   item_t *item; \
   STACK_NEXT(state.stack, item, on_error); \
   item->fail.code = xcode; \
   item->fail.text = xtext; \
   item->fail.count = state.count; \
   item->fail.level = state.level; \
   item->fail.phantom = 0; \
   state.count = 0; \
   state.level++; \
   state.point++; \
}

/* Update the last failure point with a new position in the text. */

/* #define UPDATE_FAILURE(state, xtext, on_error) \ */
/* { \ */
/*    item_t *item; \ */
/*    STACK_DISCARD(state.stack, state.count, on_error); \ */
/*    STACK_TOP(state.stack, item, on_error); \ */
/*    item->fail.text = xtext; \ */
/*    state.count = 0; \ */
/* } */

/* #define UPDATE_FAILURE(state, xtext, on_error) \ */
/* { \ */
/*    item_t *item; \ */
/*    STACK_BACK(state.stack, item, state.count + 1, on_error); \ */
/*    item->fail.text = xtext; \ */
/* } */

#define UPDATE_FAILURE(state, xtext, on_error) \
{ \
   item_t *item; \
   STACK_BACK(state.stack, item, state.count + 1, on_error); \
   if (!item->fail.phantom) \
   { \
      item_t *item2; \
      STACK_NEXT(state.stack, item2, on_error); \
      item2->fail.code = item->fail.code; \
      item2->fail.text = xtext; \
      item2->fail.count = state.count; \
      item2->fail.level = state.level; \
      item2->fail.phantom = 1; \
      state.count = 0; \
      state.level++; \
      state.point++; \
   } \
   else \
   { \
      STACK_DISCARD(state.stack, state.count, on_error); \
      STACK_TOP(state.stack, item, on_error); \
      item->fail.text = xtext; \
      state.count = 0; \
      state.level++; \
   } \
}

#define POP_FAILURE(state, xcode, xtext, on_empty, on_error) \
{ \
   item_t *item; \
   do \
   { \
      while(state.count > 0) \
      { \
         STACK_PREV(state.stack, item, on_error); \
         state.start[item->reg.num] = item->reg.start; \
         state.end[item->reg.num] = item->reg.end; \
         state.changed[item->reg.num] = item->reg.level; \
         state.count--; \
      } \
      STACK_PREV(state.stack, item, on_empty); \
      xcode = item->fail.code; \
      xtext = item->fail.text; \
      state.count = item->fail.count; \
      state.level = item->fail.level; \
      state.point--; \
   } \
   while (item->fail.text == NULL); \
}

enum regexp_compiled_ops /* opcodes for compiled regexp */
{
  Cend,			/* end of pattern reached */
  Cbol,			/* beginning of line */
  Ceol,			/* end of line */
  Cset,			/* character set.  Followed by 32 bytes of set. */
  Cexact,		/* followed by a byte to match */
  Canychar,		/* matches any character except newline */
  Cstart_memory,	/* set register start addr (followed by reg number) */
  Cend_memory,		/* set register end addr (followed by reg number) */
  Cmatch_memory,	/* match a duplicate of reg contents (regnum follows)*/
  Cjump,		/* followed by two bytes (lsb,msb) of displacement. */
  Cstar_jump,		/* will change to jump/update_failure_jump at runtime */
  Cfailure_jump,	/* jump to addr on failure */
  Cupdate_failure_jump,	/* update topmost failure point and jump */
  Cdummy_failure_jump,	/* push a dummy failure point and jump */
  Cbegbuf,		/* match at beginning of buffer */
  Cendbuf,		/* match at end of buffer */
  Cwordbeg,		/* match at beginning of word */
  Cwordend,		/* match at end of word */
  Cwordbound,		/* match if at word boundary */
  Cnotwordbound,	/* match if not at word boundary */
  Csyntaxspec,		/* matches syntax code (1 byte follows) */
  Cnotsyntaxspec	/* matches if syntax code does not match (1 byte foll)*/
};

enum regexp_syntax_op	/* syntax codes for plain and quoted characters */
{
  Rend,			/* special code for end of regexp */
  Rnormal,		/* normal character */
  Ranychar,		/* any character except newline */
  Rquote,		/* the quote character */
  Rbol,			/* match beginning of line */
  Reol,			/* match end of line */
  Roptional,		/* match preceding expression optionally */
  Rstar,		/* match preceding expr zero or more times */
  Rplus,		/* match preceding expr one or more times */
  Ror,			/* match either of alternatives */
  Ropenpar,		/* opening parenthesis */
  Rclosepar,		/* closing parenthesis */
  Rmemory,		/* match memory register */
  Rextended_memory,	/* \vnn to match registers 10-99 */
  Ropenset,		/* open set.  Internal syntax hard-coded below. */
  /* the following are gnu extensions to "normal" regexp syntax */
  Rbegbuf,		/* beginning of buffer */
  Rendbuf,		/* end of buffer */
  Rwordchar,		/* word character */
  Rnotwordchar,		/* not word character */
  Rwordbeg,		/* beginning of word */
  Rwordend,		/* end of word */
  Rwordbound,		/* word bound */
  Rnotwordbound,	/* not word bound */
  Rnum_ops
};

static int re_compile_initialized = 0;
static int regexp_syntax = 0;
int re_syntax = 0; /* Exported copy of regexp_syntax */
static unsigned char regexp_plain_ops[256];
static unsigned char regexp_quoted_ops[256];
static unsigned char regexp_precedences[Rnum_ops];
static int regexp_context_indep_ops;
static int regexp_ansi_sequences;

#define NUM_LEVELS  5    /* number of precedence levels in use */
#define MAX_NESTING 100  /* max nesting level of operators */

#define SYNTAX(ch) re_syntax_table[(unsigned char)(ch)]
#define Sword 1

static char re_syntax_table[256];

static void re_compile_initialize(void)
{
   int a;
  
   static int syntax_table_inited = 0;
   
   if (!syntax_table_inited)
   {
      syntax_table_inited = 1;
      memset(re_syntax_table, 0, 256);
      for (a = 'a'; a <= 'z'; a++)
	 re_syntax_table[a] = Sword;
      for (a = 'A'; a <= 'Z'; a++)
	 re_syntax_table[a] = Sword;
      for (a = '0'; a <= '9'; a++)
	 re_syntax_table[a] = Sword;
   }
   re_compile_initialized = 1;
   for (a = 0; a < 256; a++)
   {
      regexp_plain_ops[a] = Rnormal;
      regexp_quoted_ops[a] = Rnormal;
   }
   for (a = '0'; a <= '9'; a++)
      regexp_quoted_ops[a] = Rmemory;
   regexp_plain_ops['\134'] = Rquote;
   if (regexp_syntax & RE_NO_BK_PARENS)
   {
      regexp_plain_ops['('] = Ropenpar;
      regexp_plain_ops[')'] = Rclosepar;
   }
   else
   {
      regexp_quoted_ops['('] = Ropenpar;
      regexp_quoted_ops[')'] = Rclosepar;
   }
   if (regexp_syntax & RE_NO_BK_VBAR)
      regexp_plain_ops['\174'] = Ror;
   else
      regexp_quoted_ops['\174'] = Ror;
   regexp_plain_ops['*'] = Rstar;
   if (regexp_syntax & RE_BK_PLUS_QM)
   {
      regexp_quoted_ops['+'] = Rplus;
      regexp_quoted_ops['?'] = Roptional;
   }
   else
   {
      regexp_plain_ops['+'] = Rplus;
      regexp_plain_ops['?'] = Roptional;
   }
   if (regexp_syntax & RE_NEWLINE_OR)
      regexp_plain_ops['\n'] = Ror;
   regexp_plain_ops['\133'] = Ropenset;
   regexp_plain_ops['\136'] = Rbol;
   regexp_plain_ops['$'] = Reol;
   regexp_plain_ops['.'] = Ranychar;
   if (!(regexp_syntax & RE_NO_GNU_EXTENSIONS))
   {
      regexp_quoted_ops['w'] = Rwordchar;
      regexp_quoted_ops['W'] = Rnotwordchar;
      regexp_quoted_ops['<'] = Rwordbeg;
      regexp_quoted_ops['>'] = Rwordend;
      regexp_quoted_ops['b'] = Rwordbound;
      regexp_quoted_ops['B'] = Rnotwordbound;
      regexp_quoted_ops['`'] = Rbegbuf;
      regexp_quoted_ops['\''] = Rendbuf;
   }
   if (regexp_syntax & RE_ANSI_HEX)
      regexp_quoted_ops['v'] = Rextended_memory;
   for (a = 0; a < Rnum_ops; a++)
      regexp_precedences[a] = 4;
   if (regexp_syntax & RE_TIGHT_VBAR)
   {
      regexp_precedences[Ror] = 3;
      regexp_precedences[Rbol] = 2;
      regexp_precedences[Reol] = 2;
   }
   else
   {
      regexp_precedences[Ror] = 2;
      regexp_precedences[Rbol] = 3;
      regexp_precedences[Reol] = 3;
   }
   regexp_precedences[Rclosepar] = 1;
   regexp_precedences[Rend] = 0;
   regexp_context_indep_ops = (regexp_syntax & RE_CONTEXT_INDEP_OPS) != 0;
   regexp_ansi_sequences = (regexp_syntax & RE_ANSI_HEX) != 0;
}

int re_set_syntax(int syntax)
{
   int ret;
   
   ret = regexp_syntax;
   regexp_syntax = syntax;
   re_syntax = syntax; /* Exported copy */
   re_compile_initialize();
   return ret;
}

static int hex_char_to_decimal(int ch)
{
   if (ch >= '0' && ch <= '9')
      return ch - '0';
   if (ch >= 'a' && ch <= 'f')
      return ch - 'a' + 10;
   if (ch >= 'A' && ch <= 'F')
      return ch - 'A' + 10;
   return 16;
}

static void re_compile_fastmap_aux(char *code,
				   int pos,
				   char *visited,
				   char *can_be_null,
				   char *fastmap)
{
   int a;
   int b;
   int syntaxcode;
   
   if (visited[pos])
      return;  /* we have already been here */
   visited[pos] = 1;
   for (;;)
      switch (code[pos++])
      {
	 case Cend:
	 {
	    *can_be_null = 1;
	    return;
	 }
	 case Cbol:
	 case Cbegbuf:
	 case Cendbuf:
	 case Cwordbeg:
	 case Cwordend:
	 case Cwordbound:
	 case Cnotwordbound:
	 {
	    break;
	 }
	 case Csyntaxspec:
	 {
	    syntaxcode = code[pos++];
	    for (a = 0; a < 256; a++)
	       if (SYNTAX(a) == syntaxcode)
		  fastmap[a] = 1;
	    return;
	 }
	 case Cnotsyntaxspec:
	 {
	    syntaxcode = code[pos++];
	    for (a = 0; a < 256; a++)
	       if (SYNTAX(a) != syntaxcode)
		  fastmap[a] = 1;
	    return;
	 }
	 case Ceol:
	 {
	    fastmap['\n'] = 1;
	    if (*can_be_null == 0)
	       *can_be_null = 2; /* can match null, but only at end of buffer*/
	    return;
	 }
	 case Cset:
	 {
	    for (a = 0; a < 256/8; a++)
	       if (code[pos + a] != 0)
		  for (b = 0; b < 8; b++)
		     if (code[pos + a] & (1 << b))
			fastmap[(a << 3) + b] = 1;
	    pos += 256/8;
	    return;
	 }
	 case Cexact:
	 {
	    fastmap[(unsigned char)code[pos]] = 1;
	    return;
	 }
	 case Canychar:
	 {
	    for (a = 0; a < 256; a++)
	       if (a != '\n')
		  fastmap[a] = 1;
	    return;
	 }
	 case Cstart_memory:
	 case Cend_memory:
	 {
	    pos++;
	    break;
	 }
	 case Cmatch_memory:
	 {
	    for (a = 0; a < 256; a++)
	       fastmap[a] = 1;
	    *can_be_null = 1;
	    return;
	 }
	 case Cjump:
	 case Cdummy_failure_jump:
	 case Cupdate_failure_jump:
	 case Cstar_jump:
	 {
	    a = (unsigned char)code[pos++];
	    a |= (unsigned char)code[pos++] << 8;
	    pos += (int)(short)a;
	    if (visited[pos])
	    {
	       /* argh... the regexp contains empty loops.  This is not
		  good, as this may cause a failure stack overflow when
		  matching.  Oh well. */
	       /* this path leads nowhere; pursue other paths. */
	       return;
	    }
	    visited[pos] = 1;
	    break;
	 }
	 case Cfailure_jump:
	 {
	    a = (unsigned char)code[pos++];
	    a |= (unsigned char)code[pos++] << 8;
	    a = pos + (int)(short)a;
	    re_compile_fastmap_aux(code, a, visited, can_be_null, fastmap);
	    break;
	 }
	 default:
	 {
	    abort();  /* probably some opcode is missing from this switch */
	    /*NOTREACHED*/
	 }
      }
}

static int re_do_compile_fastmap(char *buffer,
				 int used,
				 int pos,
				 char *can_be_null,
				 char *fastmap)
{
   char small_visited[512], *visited;
   
   if (used <= sizeof(small_visited))
      visited = small_visited;
   else
   {
      visited = malloc(used);
      if (!visited)
	 return 0;
   }
   *can_be_null = 0;
   memset(fastmap, 0, 256);
   memset(visited, 0, used);
   re_compile_fastmap_aux(buffer, pos, visited, can_be_null, fastmap);
   if (visited != small_visited)
      free(visited);
   return 1;
}

void re_compile_fastmap(regexp_t bufp)
{
   if (!bufp->fastmap || bufp->fastmap_accurate)
      return;
   assert(bufp->used > 0);
   if (!re_do_compile_fastmap(bufp->buffer,
			      bufp->used,
			      0,
			      &bufp->can_be_null,
			      bufp->fastmap))
      return;
   if (bufp->buffer[0] == Cbol)
      bufp->anchor = 1;   /* begline */
   else
      if (bufp->buffer[0] == Cbegbuf)
	 bufp->anchor = 2; /* begbuf */
      else
	 bufp->anchor = 0; /* none */
   bufp->fastmap_accurate = 1;
}

/* 
 * star is coded as:
 * 1: failure_jump 2
 *    ... code for operand of star
 *    star_jump 1
 * 2: ... code after star
 *
 * We change the star_jump to update_failure_jump if we can determine
 * that it is safe to do so; otherwise we change it to an ordinary
 * jump.
 *
 * plus is coded as
 *
 *    jump 2
 * 1: failure_jump 3
 * 2: ... code for operand of plus
 *    star_jump 1
 * 3: ... code after plus
 *
 * For star_jump considerations this is processed identically to star.
 *
 */

static int re_optimize_star_jump(regexp_t bufp, char *code)
{
   char map[256];
   char can_be_null;
   char *p1;
   char *p2;
   char ch;
   int a;
   int b;

   a = (unsigned char)*code++;
   a |= (unsigned char)*code++ << 8;
   a = (int)(short)a;

   p1 = code + a + 3; /* skip the failure_jump */
   assert(p1[-3] == Cfailure_jump);
   p2 = code;
   /* p1 points inside loop, p2 points to after loop */
   if (!re_do_compile_fastmap(bufp->buffer, bufp->used,
			      p2 - bufp->buffer, &can_be_null, map))
      goto make_normal_jump;
   
   /* If we might introduce a new update point inside the
    * loop, we can't optimize because then update_jump would
    * update a wrong failure point.  Thus we have to be
    * quite careful here.
    */
      
   /* loop until we find something that consumes a character */
  loop_p1:
   switch (*p1++)
   {
      case Cbol:
      case Ceol:
      case Cbegbuf:
      case Cendbuf:
      case Cwordbeg:
      case Cwordend:
      case Cwordbound:
      case Cnotwordbound:
      {
	 goto loop_p1;
      }
      case Cstart_memory:
      case Cend_memory:
      {
	 p1++;
	 goto loop_p1;
      }
      case Cexact:
      {
	 ch = (unsigned char)*p1++;
	 if (map[(int)ch])
	    goto make_normal_jump;
	 break;
      }
      case Canychar:
      {
	 for (b = 0; b < 256; b++)
	    if (b != '\n' && map[b])
	       goto make_normal_jump;
	 break;
      }
      case Cset:
      {
	 for (b = 0; b < 256; b++)
	    if ((p1[b >> 3] & (1 << (b & 7))) && map[b])
	       goto make_normal_jump;
	 p1 += 256/8;
	 break;
      }
      default:
      {
	 goto make_normal_jump;
      }
   }
   /* now we know that we can't backtrack. */
   while (p1 != p2 - 3)
   {
      switch (*p1++)
      {
	 case Cend:
	 {
	    return 0;
	 }
	 case Cbol:
	 case Ceol:
	 case Canychar:
	 case Cbegbuf:
	 case Cendbuf:
	 case Cwordbeg:
	 case Cwordend:
	 case Cwordbound:
	 case Cnotwordbound:
	 {
	    break;
	 }
	 case Cset:
	 {
	    p1 += 256/8;
	    break;
	 }
	 case Cexact:
	 case Cstart_memory:
	 case Cend_memory:
	 case Cmatch_memory:
	 case Csyntaxspec:
	 case Cnotsyntaxspec:
	 {
	    p1++;
	    break;
	 }
	 case Cjump:
	 case Cstar_jump:
	 case Cfailure_jump:
	 case Cupdate_failure_jump:
	 case Cdummy_failure_jump:
	 {
	    goto make_normal_jump;
	 }
	 default:
	 {
	    return 0;
	    break;
	 }
      }
   }

   code -= 3;
   a += 3;  /* jump to after the Cfailure_jump */
   code[0] = Cupdate_failure_jump;
   code[1] = a & 0xff;
   code[2] = a >> 8;
   return 1;

  make_normal_jump:
   code -= 3;
   *code = Cjump;
   return 1;
}

static int re_optimize(regexp_t bufp)
{
   char *code;

   code = bufp->buffer;

   while(1)
   {
      switch (*code++)
      {
	 case Cend:
	 {
	    return 1;
	 }
	 case Canychar:
	 case Cbol:
	 case Ceol:
	 case Cbegbuf:
	 case Cendbuf:
	 case Cwordbeg:
	 case Cwordend:
	 case Cwordbound:
	 case Cnotwordbound:
	 {
	    break;
	 }
	 case Cset:
	 {
	    code += 256/8;
	    break;
	 }
	 case Cexact:
	 case Cstart_memory:
	 case Cend_memory:
	 case Cmatch_memory:
	 case Csyntaxspec:
	 case Cnotsyntaxspec:
	 {
	    code++;
	    break;
	 }
	 case Cstar_jump:
	 {
	    if (!re_optimize_star_jump(bufp, code))
	    {
	       return 0;
	    }
	    /* fall through */
	 }
	 case Cupdate_failure_jump:
	 case Cjump:
	 case Cdummy_failure_jump:
	 case Cfailure_jump:
	 {
	    code += 2;
	    break;
	 }
	 default:
	 {
	    return 0;
	 }
      }
   }
}

#define NEXTCHAR(var) \
{ \
   if (pos >= size) \
      goto ends_prematurely; \
   (var) = regex[pos]; \
   pos++; \
}

#define ALLOC(amount) \
{ \
   if (pattern_offset+(amount) > alloc) \
   { \
      alloc += 256 + (amount); \
      pattern = realloc(pattern, alloc); \
      if (!pattern) \
	 goto out_of_memory; \
   } \
}

#define STORE(ch) pattern[pattern_offset++] = (ch)

#define CURRENT_LEVEL_START (starts[starts_base + current_level])

#define SET_LEVEL_START starts[starts_base + current_level] = pattern_offset

#define PUSH_LEVEL_STARTS \
   if (starts_base < (MAX_NESTING-1)*NUM_LEVELS) \
      starts_base += NUM_LEVELS; \
   else \
      goto too_complex

#define POP_LEVEL_STARTS starts_base -= NUM_LEVELS

#define PUT_ADDR(offset,addr) \
{ \
   int disp = (addr) - (offset) - 2; \
   pattern[(offset)] = disp & 0xff; \
   pattern[(offset)+1] = (disp>>8) & 0xff; \
}

#define INSERT_JUMP(pos,type,addr) \
{ \
   int a, p = (pos), t = (type), ad = (addr); \
   for (a = pattern_offset - 1; a >= p; a--) \
      pattern[a + 3] = pattern[a]; \
   pattern[p] = t; \
   PUT_ADDR(p+1,ad); \
   pattern_offset += 3; \
}
#define SETBIT(buf,offset,bit) (buf)[(offset)+(bit)/8] |= (1<<((bit) & 7))

#define SET_FIELDS \
{ \
   bufp->allocated = alloc; \
   bufp->buffer = pattern; \
   bufp->used = pattern_offset; \
}
    
#define GETHEX(var) \
{ \
   char gethex_ch, gethex_value; \
   NEXTCHAR(gethex_ch); \
   gethex_value = hex_char_to_decimal(gethex_ch); \
   if (gethex_value == 16) \
      goto hex_error; \
   NEXTCHAR(gethex_ch); \
   gethex_ch = hex_char_to_decimal(gethex_ch); \
   if (gethex_ch == 16) \
      goto hex_error; \
   (var) = gethex_value * 16 + gethex_ch; \
}

#define ANSI_TRANSLATE(ch)  \
{ \
   switch (ch) \
   { \
      case 'a': \
      case 'A': \
      { \
	 ch = 7; /* audible bell */ \
	 break; \
      } \
      case 'b': \
      case 'B': \
      { \
	 ch = 8; /* backspace */ \
	 break; \
      } \
      case 'f': \
      case 'F': \
      { \
	 ch = 12; /* form feed */ \
	 break; \
      } \
      case 'n': \
      case 'N': \
      { \
	 ch = 10; /* line feed */ \
	 break; \
      } \
      case 'r': \
      case 'R': \
      { \
	 ch = 13; /* carriage return */ \
	 break; \
      } \
      case 't': \
      case 'T': \
      { \
	 ch = 9; /* tab */ \
	 break; \
      } \
      case 'v': \
      case 'V': \
      { \
	 ch = 11; /* vertical tab */ \
	 break; \
      } \
      case 'x': /* hex code */ \
      case 'X': \
      { \
	 GETHEX(ch); \
	 break; \
      } \
      default: \
      { \
	 /* other characters passed through */ \
	 if (translate) \
	    ch = translate[(unsigned char)ch]; \
	 break; \
      } \
   } \
}

char *re_compile_pattern(char *regex, int size, regexp_t bufp)
{
   int a;
   int pos;
   int op;
   int current_level;
   int level;
   int opcode;
   int pattern_offset = 0, alloc;
   int starts[NUM_LEVELS * MAX_NESTING];
   int starts_base;
   int future_jumps[MAX_NESTING];
   int num_jumps;
   unsigned char ch = '\0';
   char *pattern;
   char *translate;
   int next_register;
   int paren_depth;
   int num_open_registers;
   int open_registers[RE_NREGS];
   int beginning_context;

   if (!re_compile_initialized)
      re_compile_initialize();
   bufp->used = 0;
   bufp->fastmap_accurate = 0;
   bufp->uses_registers = 0;
   translate = bufp->translate;
   pattern = bufp->buffer;
   alloc = bufp->allocated;
   if (alloc == 0 || pattern == NULL)
   {
      alloc = 256;
      pattern = malloc(alloc);
      if (!pattern)
	 goto out_of_memory;
   }
   pattern_offset = 0;
   starts_base = 0;
   num_jumps = 0;
   current_level = 0;
   SET_LEVEL_START;
   num_open_registers = 0;
   next_register = 1;
   paren_depth = 0;
   beginning_context = 1;
   op = -1;
   /* we use Rend dummy to ensure that pending jumps are updated (due to
      low priority of Rend) before exiting the loop. */
   pos = 0;
   while (op != Rend)
   {
      if (pos >= size)
	 op = Rend;
      else
      {
	 NEXTCHAR(ch);
	 if (translate)
	    ch = translate[(unsigned char)ch];
	 op = regexp_plain_ops[(unsigned char)ch];
	 if (op == Rquote)
	 {
	    NEXTCHAR(ch);
	    op = regexp_quoted_ops[(unsigned char)ch];
	    if (op == Rnormal && regexp_ansi_sequences)
	       ANSI_TRANSLATE(ch);
	 }
      }
      level = regexp_precedences[op];
      /* printf("ch='%c' op=%d level=%d current_level=%d curlevstart=%d\n",
	 ch, op, level, current_level, CURRENT_LEVEL_START); */
      if (level > current_level)
      {
	 for (current_level++; current_level < level; current_level++)
	    SET_LEVEL_START;
	 SET_LEVEL_START;
      }
      else
	 if (level < current_level)
	 {
	    current_level = level;
	    for (;num_jumps > 0 &&
		    future_jumps[num_jumps-1] >= CURRENT_LEVEL_START;
		 num_jumps--)
	       PUT_ADDR(future_jumps[num_jumps-1], pattern_offset);
	 }
      switch (op)
      {
	 case Rend:
	 {
	    break;
	 }
	 case Rnormal:
	 {
	   normal_char:
	    opcode = Cexact;
	   store_opcode_and_arg: /* opcode & ch must be set */
	    SET_LEVEL_START;
	    ALLOC(2);
	    STORE(opcode);
	    STORE(ch);
	    break;
	 }
	 case Ranychar:
	 {
	    opcode = Canychar;
	   store_opcode:
	    SET_LEVEL_START;
	    ALLOC(1);
	    STORE(opcode);
	    break;
	 }
	 case Rquote:
	 {
	    abort();
	    /*NOTREACHED*/
	 }
	 case Rbol:
	 {
	    if (!beginning_context)
	       if (regexp_context_indep_ops)
		  goto op_error;
	       else
		  goto normal_char;
	    opcode = Cbol;
	    goto store_opcode;
	 }
	 case Reol:
	 {
	    if (!((pos >= size) ||
		  ((regexp_syntax & RE_NO_BK_VBAR) ?
		   (regex[pos] == '\174') :
		   (pos+1 < size && regex[pos] == '\134' &&
		    regex[pos+1] == '\174')) ||
		  ((regexp_syntax & RE_NO_BK_PARENS)?
		   (regex[pos] == ')'):
		   (pos+1 < size && regex[pos] == '\134' &&
		    regex[pos+1] == ')'))))
	       if (regexp_context_indep_ops)
		  goto op_error;
	       else
		  goto normal_char;
	    opcode = Ceol;
	    goto store_opcode;
	    /* NOTREACHED */
	    break;
	 }
	 case Roptional:
	 {
	    if (beginning_context)
	       if (regexp_context_indep_ops)
		  goto op_error;
	       else
		  goto normal_char;
	    if (CURRENT_LEVEL_START == pattern_offset)
	       break; /* ignore empty patterns for ? */
	    ALLOC(3);
	    INSERT_JUMP(CURRENT_LEVEL_START, Cfailure_jump,
			pattern_offset + 3);
	    break;
	 }
	 case Rstar:
	 case Rplus:
	 {
	    if (beginning_context)
	       if (regexp_context_indep_ops)
		  goto op_error;
	       else
		  goto normal_char;
	    if (CURRENT_LEVEL_START == pattern_offset)
	       break; /* ignore empty patterns for + and * */
	    ALLOC(9);
	    INSERT_JUMP(CURRENT_LEVEL_START, Cfailure_jump,
			pattern_offset + 6);
	    INSERT_JUMP(pattern_offset, Cstar_jump, CURRENT_LEVEL_START);
	    if (op == Rplus)  /* jump over initial failure_jump */
	       INSERT_JUMP(CURRENT_LEVEL_START, Cdummy_failure_jump,
			   CURRENT_LEVEL_START + 6);
	    break;
	 }
	 case Ror:
	 {
	    ALLOC(6);
	    INSERT_JUMP(CURRENT_LEVEL_START, Cfailure_jump,
			pattern_offset + 6);
	    if (num_jumps >= MAX_NESTING)
	       goto too_complex;
	    STORE(Cjump);
	    future_jumps[num_jumps++] = pattern_offset;
	    STORE(0);
	    STORE(0);
	    SET_LEVEL_START;
	    break;
	 }
	 case Ropenpar:
	 {
	    SET_LEVEL_START;
	    if (next_register < RE_NREGS)
	    {
	       bufp->uses_registers = 1;
	       ALLOC(2);
	       STORE(Cstart_memory);
	       STORE(next_register);
	       open_registers[num_open_registers++] = next_register;
	       next_register++;
	    }
	    paren_depth++;
	    PUSH_LEVEL_STARTS;
	    current_level = 0;
	    SET_LEVEL_START;
	    break;
	 }
	 case Rclosepar:
	 {
	    if (paren_depth <= 0)
	       goto parenthesis_error;
	    POP_LEVEL_STARTS;
	    current_level = regexp_precedences[Ropenpar];
	    paren_depth--;
	    if (paren_depth < num_open_registers)
	    {
	       bufp->uses_registers = 1;
	       ALLOC(2);
	       STORE(Cend_memory);
	       num_open_registers--;
	       STORE(open_registers[num_open_registers]);
	    }
	    break;
	 }
	 case Rmemory:
	 {
	    if (ch == '0')
	       goto bad_match_register;
	    assert(ch >= '0' && ch <= '9');
	    bufp->uses_registers = 1;
	    opcode = Cmatch_memory;
	    ch -= '0';
	    goto store_opcode_and_arg;
	 }
	 case Rextended_memory:
	 {
	    NEXTCHAR(ch);
	    if (ch < '0' || ch > '9')
	       goto bad_match_register;
	    NEXTCHAR(a);
	    if (a < '0' || a > '9')
	       goto bad_match_register;
	    ch = 10 * (a - '0') + ch - '0';
	    if (ch <= 0 || ch >= RE_NREGS)
	       goto bad_match_register;
	    bufp->uses_registers = 1;
	    opcode = Cmatch_memory;
	    goto store_opcode_and_arg;
	 }
	 case Ropenset:
	 {
	    int complement;
	    int prev;
	    int offset;
	    int range;
	    int firstchar;
	    
	    SET_LEVEL_START;
	    ALLOC(1+256/8);
	    STORE(Cset);
	    offset = pattern_offset;
	    for (a = 0; a < 256/8; a++)
	       STORE(0);
	    NEXTCHAR(ch);
	    if (translate)
	       ch = translate[(unsigned char)ch];
	    if (ch == '\136')
	    {
	       complement = 1;
	       NEXTCHAR(ch);
	       if (translate)
		  ch = translate[(unsigned char)ch];
	    }
	    else
	       complement = 0;
	    prev = -1;
	    range = 0;
	    firstchar = 1;
	    while (ch != '\135' || firstchar)
	    {
	       firstchar = 0;
	       if (regexp_ansi_sequences && ch == '\134')
	       {
		  NEXTCHAR(ch);
		  ANSI_TRANSLATE(ch);
	       }
	       if (range)
	       {
		  for (a = prev; a <= (int)ch; a++)
		     SETBIT(pattern, offset, a);
		  prev = -1;
		  range = 0;
	       }
	       else
		  if (prev != -1 && ch == '-')
		     range = 1;
		  else
		  {
		     SETBIT(pattern, offset, ch);
		     prev = ch;
		  }
	       NEXTCHAR(ch);
	       if (translate)
		  ch = translate[(unsigned char)ch];
	    }
	    if (range)
	       SETBIT(pattern, offset, '-');
	    if (complement)
	    {
	       for (a = 0; a < 256/8; a++)
		  pattern[offset+a] ^= 0xff;
	    }
	    break;
	 }
	 case Rbegbuf:
	 {
	    opcode = Cbegbuf;
	    goto store_opcode;
	 }
	 case Rendbuf:
	 {
	    opcode = Cendbuf;
	    goto store_opcode;
	 }
	 case Rwordchar:
	 {
	    opcode = Csyntaxspec;
	    ch = Sword;
	    goto store_opcode_and_arg;
	 }
	 case Rnotwordchar:
	 {
	    opcode = Cnotsyntaxspec;
	    ch = Sword;
	    goto store_opcode_and_arg;
	 }
	 case Rwordbeg:
	 {
	    opcode = Cwordbeg;
	    goto store_opcode;
	 }
	 case Rwordend:
	 {
	    opcode = Cwordend;
	    goto store_opcode;
	 }
	 case Rwordbound:
	 {
	    opcode = Cwordbound;
	    goto store_opcode;
	 }
	 case Rnotwordbound:
	 {
	    opcode = Cnotwordbound;
	    goto store_opcode;
	 }
	 default:
	 {
	    abort();
	 }
      }
      beginning_context = (op == Ropenpar || op == Ror);
   }
   if (starts_base != 0)
      goto parenthesis_error;
   assert(num_jumps == 0);
   ALLOC(1);
   STORE(Cend);
   SET_FIELDS;
   if(!re_optimize(bufp))
      return "Optimization error";
   return NULL;

  op_error:
   SET_FIELDS;
   return "Badly placed special character";

  bad_match_register:
   SET_FIELDS;
   return "Bad match register number";
   
  hex_error:
   SET_FIELDS;
   return "Bad hexadecimal number";
   
  parenthesis_error:
   SET_FIELDS;
   return "Badly placed parenthesis";
   
  out_of_memory:
   SET_FIELDS;
   return "Out of memory";
   
  ends_prematurely:
   SET_FIELDS;
   return "Regular expression ends prematurely";

  too_complex:
   SET_FIELDS;
   return "Regular expression too complex";
}

#undef CHARAT
#undef NEXTCHAR
#undef GETHEX
#undef ALLOC
#undef STORE
#undef CURRENT_LEVEL_START
#undef SET_LEVEL_START
#undef PUSH_LEVEL_STARTS
#undef POP_LEVEL_STARTS
#undef PUT_ADDR
#undef INSERT_JUMP
#undef SETBIT
#undef SET_FIELDS

#define PREFETCH if (text == textend) goto fail

#define NEXTCHAR(var) \
PREFETCH; \
var = (unsigned char)*text++; \
if (translate) \
   var = translate[var]

int re_match(regexp_t bufp,
	     char *string,
	     int size,
	     int pos,
	     regexp_registers_t old_regs)
{
  char *code;
  char *translate;
  char *text;
  char *textstart;
  char *textend;
  int a;
  int b;
  int ch;
  int reg;
  int match_end;
  char *regstart;
  char *regend;
  int regsize;
  match_state state;
  
  assert(pos >= 0 && size >= 0);
  assert(pos <= size);
  
  text = string + pos;
  textstart = string;
  textend = string + size;
  
  code = bufp->buffer;
  
  translate = bufp->translate;
/*   translated = NULL; */
/*   if (bufp->translate) */
/*   { */
/*      char *t1; */
/*      char *t2; */
     
/*      translated = malloc(size); */
/*      if (translated == NULL) */
/* 	goto error; */

/*      t1 = string; */
/*      t2 = translated; */
/*      while(t1 < textend) */
/* 	*t2++ = bufp->translate[*t1++]; */
     
/*      text = translated + pos; */
/*      textstart = translated; */
/*      textend = translated + size; */
/*   } */
  
  NEW_STATE(state);
  
  continue_matching:
  switch (*code++)
  {
     case Cend:
     {
	match_end = text - textstart;
	if (old_regs)
	{
	   old_regs->start[0] = pos;
	   old_regs->end[0] = match_end;
	   if (!bufp->uses_registers)
	   {
	      for (a = 1; a < RE_NREGS; a++)
	      {
		 old_regs->start[a] = -1;
		 old_regs->end[a] = -1;
	      }
	   }
	   else
	   {
	      for (a = 1; a < RE_NREGS; a++)
	      {
		 if ((GET_REG_START(state, a) == NULL) ||
		     (GET_REG_END(state, a) == NULL))
		 {
		    old_regs->start[a] = -1;
		    old_regs->end[a] = -1;
		    continue;
		 }
		 old_regs->start[a] = GET_REG_START(state, a) - textstart;
		 old_regs->end[a] = GET_REG_END(state, a) - textstart;
	      }
	   }
	}
/* 	if(translated) */
/* 	   free(translated); */
	FREE_STATE(state);
	return match_end - pos;
     }
     case Cbol:
     {
	if (text == textstart || text[-1] == '\n')
	   goto continue_matching;
	goto fail;
     }
     case Ceol:
     {
	if (text == textend || *text == '\n')
	   goto continue_matching;
	goto fail;
     }
     case Cset:
     {
	NEXTCHAR(ch);
	if (code[ch/8] & (1<<(ch & 7)))
	{
	   code += 256/8;
	   goto continue_matching;
	}
	goto fail;
     }
     case Cexact:
     {
	NEXTCHAR(ch);
	if (ch != (unsigned char)*code++)
	   goto fail;
/* 	{ */
/* 	   char *p1 = code - 2; */
/* 	   ch = *(code - 1); */
/* 	   POP_FAILURE(state, code, text, goto done_matching, goto error); */
/* 	   while ((code == p1) && (*text != ch)) */
/* 	      POP_FAILURE(state, code, text, goto done_matching, goto error); */
/* 	   if ((code == p1) && (*text == ch)) */
/* 	   { */
/* 	      code += 2; */
/* 	      text++; */
/* 	   } */
/* 	} */
	goto continue_matching;
     }
     case Canychar:
     {
	NEXTCHAR(ch);
	if (ch == '\n')
	   goto fail;
	goto continue_matching;
     }
     case Cstart_memory:
     {
	reg = *code++;
	SET_REG_START(state, reg, text, goto error);
	goto continue_matching;
     }
     case Cend_memory:
     {
	reg = *code++;
	SET_REG_END(state, reg, text, goto error);
	goto continue_matching;
     }
     case Cmatch_memory:
     {
	reg = *code++;
	regstart = GET_REG_START(state, reg);
	regend = GET_REG_END(state, reg);
	if ((regstart == NULL) || (regend == NULL))
	   goto fail;  /* or should we just match nothing? */
	regsize = regend - regstart;

	if (regsize > (textend - text))
	   goto fail;
	if(translate)
	{
	   for (; regstart < regend; regstart++, text++)
	      if (translate[*regstart] != translate[*text])
		 goto fail;
	}
	else
	   for (; regstart < regend; regstart++, text++)
	      if (*regstart != *text)
		 goto fail;
/*	if (memcmp(text, regstart, regsize) != 0)
	   goto fail;
	text += regsize; */
	goto continue_matching;
     }
     case Cupdate_failure_jump:
     {
	UPDATE_FAILURE(state, text, goto error);
	/* fall to next case */
     }
     /* treat Cstar_jump just like Cjump if it hasn't been optimized */
     case Cstar_jump:
     case Cjump:
     {
	a = (unsigned char)*code++;
	a |= (unsigned char)*code++ << 8;
	code += (int)(short)a;
	goto continue_matching;
     }
     case Cdummy_failure_jump:
     {
	a = (unsigned char)*code++;
	a |= (unsigned char)*code++ << 8;
	a = (int)(short)a;
	assert(*code == Cfailure_jump);
	b = (unsigned char)code[1];
	b |= (unsigned char)code[2] << 8;
	PUSH_FAILURE(state, code + (int)(short)b + 3, NULL, goto error);
	code += a;
	goto continue_matching;
     }
     case Cfailure_jump:
     {
	a = (unsigned char)*code++;
	a |= (unsigned char)*code++ << 8;
	a = (int)(short)a;
	PUSH_FAILURE(state, code + a, text, goto error);
	goto continue_matching;
     }
     case Cbegbuf:
     {
	if (text == textstart)
	   goto continue_matching;
	goto fail;
     }
     case Cendbuf:
     {
	if (text == textend)
	   goto continue_matching;
	goto fail;
     }
     case Cwordbeg:
     {
	if (text == textend)
	   goto fail;
	if (SYNTAX(*text) != Sword)
	   goto fail;
	if (text == textstart)
	   goto continue_matching;
	if (SYNTAX(text[-1]) != Sword)
	   goto continue_matching;
	goto fail;
     }
     case Cwordend:
     {
	if (text == textstart)
	   goto fail;
	if (SYNTAX(text[-1]) != Sword)
	   goto fail;
	if (text == textend)
	   goto continue_matching;
	if (SYNTAX(*text) == Sword)
	   goto fail;
	goto continue_matching;
     }
     case Cwordbound:
     {
	/* Note: as in gnu regexp, this also matches at the beginning
	 * and end of buffer.  */

	if (text == textstart || text == textend)
	   goto continue_matching;
	if ((SYNTAX(text[-1]) == Sword) ^ (SYNTAX(*text) == Sword))
	   goto continue_matching;
	goto fail;
     }
     case Cnotwordbound:
     {
	/* Note: as in gnu regexp, this never matches at the beginning
	 * and end of buffer.  */
	if (text == textstart || text == textend)
	   goto fail;
	if (!((SYNTAX(text[-1]) == Sword) ^ (SYNTAX(*text) == Sword)))
	   goto fail;
	goto continue_matching;
     }
     case Csyntaxspec:
     {
	NEXTCHAR(ch);
	if (SYNTAX(ch) != (unsigned char)*code++)
	   goto fail;
	   goto continue_matching;
     }
     case Cnotsyntaxspec:
     {
	NEXTCHAR(ch);
	if (SYNTAX(ch) != (unsigned char)*code++)
	   break;
	goto continue_matching;
     }
     default:
     {
	abort();
	/*NOTREACHED*/
     }
  }

#if 0 /* This line is never reached --Guido */
  abort();
#endif
  /*
   *NOTREACHED
   */
  
  fail:
  POP_FAILURE(state, code, text, goto done_matching, goto error);
  goto continue_matching;
  
  done_matching:
/*   if(translated != NULL) */
/*      free(translated); */
  FREE_STATE(state);
  return -1;

  error:
/*   if (translated != NULL) */
/*      free(translated); */
  FREE_STATE(state);
  return -2;
}

#undef PREFETCH
#undef NEXTCHAR

int re_search(regexp_t bufp,
	      char *string,
	      int size,
	      int pos,
	      int range,
	      regexp_registers_t regs)
{
  char *fastmap;
  char *translate;
  char *text;
  char *partstart;
  char *partend;
  int dir;
  int ret;
  char anchor;
  
  assert(size >= 0 && pos >= 0);
  assert(pos + range >= 0 && pos + range <= size); /* Bugfix by ylo */
  
  fastmap = bufp->fastmap;
  translate = bufp->translate;
  if (fastmap && !bufp->fastmap_accurate)
     re_compile_fastmap(bufp);
  anchor = bufp->anchor;
  if (bufp->can_be_null == 1) /* can_be_null == 2: can match null at eob */
     fastmap = NULL;

  if (range < 0)
  {
     dir = -1;
     range = -range;
  }
  else
     dir = 1;

  if (anchor == 2)
     if (pos != 0)
	return -1;
     else
	range = 0;

  for (; range >= 0; range--, pos += dir)
  {
     if (fastmap)
     {
	if (dir == 1)
	{ /* searching forwards */

	   text = string + pos;
	   partend = string + size;
	   partstart = text;
	   if (translate)
	      while (text != partend &&
		     !fastmap[(unsigned char) translate[(unsigned char)*text]])
		 text++;
	   else
	      while (text != partend && !fastmap[(unsigned char)*text])
		 text++;
	   pos += text - partstart;
	   range -= text - partstart;
	   if (pos == size && bufp->can_be_null == 0)
	      return -1;
	}
	else
	{ /* searching backwards */
	   text = string + pos;
	   partstart = string + pos - range;
	   partend = text;
	   if (translate)
	      while (text != partstart &&
		     !fastmap[(unsigned char)
			     translate[(unsigned char)*text]])
		 text--;
	   else
	      while (text != partstart &&
		     !fastmap[(unsigned char)*text])
		 text--;
	   pos -= partend - text;
	   range -= partend - text;
	}
     }
     if (anchor == 1)
     { /* anchored to begline */
	if (pos > 0 && (string[pos - 1] != '\n'))
	   continue;
     }
     assert(pos >= 0 && pos <= size);
     ret = re_match(bufp, string, size, pos, regs);
     if (ret >= 0)
	return pos;
     if (ret == -2)
	return -2;
  }
  return -1;
}