/[pcre]/code/trunk/pcre_dfa_exec.c
ViewVC logotype

Diff of /code/trunk/pcre_dfa_exec.c

Parent Directory Parent Directory | Revision Log Revision Log | View Patch Patch

revision 361 by ph10, Thu Jul 10 16:03:28 2008 UTC revision 463 by ph10, Sun Oct 18 10:02:46 2009 UTC
# Line 3  Line 3 
3  *************************************************/  *************************************************/
4    
5  /* PCRE is a library of functions to support regular expressions whose syntax  /* PCRE is a library of functions to support regular expressions whose syntax
6  and semantics are as close as possible to those of the Perl 5 language.  and semantics are as close as possible to those of the Perl 5 language (but see
7    below for why this module is different).
8    
9                         Written by Philip Hazel                         Written by Philip Hazel
10             Copyright (c) 1997-2008 University of Cambridge             Copyright (c) 1997-2009 University of Cambridge
11    
12  -----------------------------------------------------------------------------  -----------------------------------------------------------------------------
13  Redistribution and use in source and binary forms, with or without  Redistribution and use in source and binary forms, with or without
# Line 44  FSM). This is NOT Perl- compatible, but Line 45  FSM). This is NOT Perl- compatible, but
45  applications. */  applications. */
46    
47    
48    /* NOTE ABOUT PERFORMANCE: A user of this function sent some code that improved
49    the performance of his patterns greatly. I could not use it as it stood, as it
50    was not thread safe, and made assumptions about pattern sizes. Also, it caused
51    test 7 to loop, and test 9 to crash with a segfault.
52    
53    The issue is the check for duplicate states, which is done by a simple linear
54    search up the state list. (Grep for "duplicate" below to find the code.) For
55    many patterns, there will never be many states active at one time, so a simple
56    linear search is fine. In patterns that have many active states, it might be a
57    bottleneck. The suggested code used an indexing scheme to remember which states
58    had previously been used for each character, and avoided the linear search when
59    it knew there was no chance of a duplicate. This was implemented when adding
60    states to the state lists.
61    
62    I wrote some thread-safe, not-limited code to try something similar at the time
63    of checking for duplicates (instead of when adding states), using index vectors
64    on the stack. It did give a 13% improvement with one specially constructed
65    pattern for certain subject strings, but on other strings and on many of the
66    simpler patterns in the test suite it did worse. The major problem, I think,
67    was the extra time to initialize the index. This had to be done for each call
68    of internal_dfa_exec(). (The supplied patch used a static vector, initialized
69    only once - I suspect this was the cause of the problems with the tests.)
70    
71    Overall, I concluded that the gains in some cases did not outweigh the losses
72    in others, so I abandoned this code. */
73    
74    
75    
76  #ifdef HAVE_CONFIG_H  #ifdef HAVE_CONFIG_H
77  #include "config.h"  #include "config.h"
78  #endif  #endif
# Line 60  applications. */ Line 89  applications. */
89  #define SP "                   "  #define SP "                   "
90    
91    
   
92  /*************************************************  /*************************************************
93  *      Code parameters and static tables         *  *      Code parameters and static tables         *
94  *************************************************/  *************************************************/
# Line 81  never stored, so we push them well clear Line 109  never stored, so we push them well clear
109  character that is to be tested in some way. This makes is possible to  character that is to be tested in some way. This makes is possible to
110  centralize the loading of these characters. In the case of Type * etc, the  centralize the loading of these characters. In the case of Type * etc, the
111  "character" is the opcode for \D, \d, \S, \s, \W, or \w, which will always be a  "character" is the opcode for \D, \d, \S, \s, \W, or \w, which will always be a
112  small value. ***NOTE*** If the start of this table is modified, the two tables  small value. Non-zero values in the table are the offsets from the opcode where
113  that follow must also be modified. */  the character is to be found. ***NOTE*** If the start of this table is
114    modified, the three tables that follow must also be modified. */
115    
116  static const uschar coptable[] = {  static const uschar coptable[] = {
117    0,                             /* End                                    */    0,                             /* End                                    */
# Line 132  static const uschar coptable[] = { Line 161  static const uschar coptable[] = {
161    0,                             /* DEF                                    */    0,                             /* DEF                                    */
162    0, 0,                          /* BRAZERO, BRAMINZERO                    */    0, 0,                          /* BRAZERO, BRAMINZERO                    */
163    0, 0, 0, 0,                    /* PRUNE, SKIP, THEN, COMMIT              */    0, 0, 0, 0,                    /* PRUNE, SKIP, THEN, COMMIT              */
164    0, 0, 0                        /* FAIL, ACCEPT, SKIPZERO                 */    0, 0, 0, 0                     /* FAIL, ACCEPT, CLOSE, SKIPZERO          */
165    };
166    
167    /* This table identifies those opcodes that inspect a character. It is used to
168    remember the fact that a character could have been inspected when the end of
169    the subject is reached. ***NOTE*** If the start of this table is modified, the
170    two tables that follow must also be modified. */
171    
172    static const uschar poptable[] = {
173      0,                             /* End                                    */
174      0, 0, 0, 1, 1,                 /* \A, \G, \K, \B, \b                     */
175      1, 1, 1, 1, 1, 1,              /* \D, \d, \S, \s, \W, \w                 */
176      1, 1, 1,                       /* Any, AllAny, Anybyte                   */
177      1, 1, 1,                       /* NOTPROP, PROP, EXTUNI                  */
178      1, 1, 1, 1, 1,                 /* \R, \H, \h, \V, \v                     */
179      0, 0, 0, 0, 0,                 /* \Z, \z, Opt, ^, $                      */
180      1,                             /* Char                                   */
181      1,                             /* Charnc                                 */
182      1,                             /* not                                    */
183      /* Positive single-char repeats                                          */
184      1, 1, 1, 1, 1, 1,              /* *, *?, +, +?, ?, ??                    */
185      1, 1, 1,                       /* upto, minupto, exact                   */
186      1, 1, 1, 1,                    /* *+, ++, ?+, upto+                      */
187      /* Negative single-char repeats - only for chars < 256                   */
188      1, 1, 1, 1, 1, 1,              /* NOT *, *?, +, +?, ?, ??                */
189      1, 1, 1,                       /* NOT upto, minupto, exact               */
190      1, 1, 1, 1,                    /* NOT *+, ++, ?+, upto+                  */
191      /* Positive type repeats                                                 */
192      1, 1, 1, 1, 1, 1,              /* Type *, *?, +, +?, ?, ??               */
193      1, 1, 1,                       /* Type upto, minupto, exact              */
194      1, 1, 1, 1,                    /* Type *+, ++, ?+, upto+                 */
195      /* Character class & ref repeats                                         */
196      1, 1, 1, 1, 1, 1,              /* *, *?, +, +?, ?, ??                    */
197      1, 1,                          /* CRRANGE, CRMINRANGE                    */
198      1,                             /* CLASS                                  */
199      1,                             /* NCLASS                                 */
200      1,                             /* XCLASS - variable length               */
201      0,                             /* REF                                    */
202      0,                             /* RECURSE                                */
203      0,                             /* CALLOUT                                */
204      0,                             /* Alt                                    */
205      0,                             /* Ket                                    */
206      0,                             /* KetRmax                                */
207      0,                             /* KetRmin                                */
208      0,                             /* Assert                                 */
209      0,                             /* Assert not                             */
210      0,                             /* Assert behind                          */
211      0,                             /* Assert behind not                      */
212      0,                             /* Reverse                                */
213      0, 0, 0, 0,                    /* ONCE, BRA, CBRA, COND                  */
214      0, 0, 0,                       /* SBRA, SCBRA, SCOND                     */
215      0,                             /* CREF                                   */
216      0,                             /* RREF                                   */
217      0,                             /* DEF                                    */
218      0, 0,                          /* BRAZERO, BRAMINZERO                    */
219      0, 0, 0, 0,                    /* PRUNE, SKIP, THEN, COMMIT              */
220      0, 0, 0, 0                     /* FAIL, ACCEPT, CLOSE, SKIPZERO          */
221  };  };
222    
223  /* These 2 tables allow for compact code for testing for \D, \d, \S, \s, \W,  /* These 2 tables allow for compact code for testing for \D, \d, \S, \s, \W,
# Line 390  if (*first_op == OP_REVERSE) Line 475  if (*first_op == OP_REVERSE)
475      current_subject -= gone_back;      current_subject -= gone_back;
476      }      }
477    
478      /* Save the earliest consulted character */
479    
480      if (current_subject < md->start_used_ptr)
481        md->start_used_ptr = current_subject;
482    
483    /* Now we can process the individual branches. */    /* Now we can process the individual branches. */
484    
485    end_code = this_start_code;    end_code = this_start_code;
# Line 454  for (;;) Line 544  for (;;)
544    int i, j;    int i, j;
545    int clen, dlen;    int clen, dlen;
546    unsigned int c, d;    unsigned int c, d;
547      int forced_fail = 0;
548      BOOL could_continue = FALSE;
549    
550    /* Make the new state list into the active state list and empty the    /* Make the new state list into the active state list and empty the
551    new state list. */    new state list. */
# Line 511  for (;;) Line 603  for (;;)
603      stateblock *current_state = active_states + i;      stateblock *current_state = active_states + i;
604      const uschar *code;      const uschar *code;
605      int state_offset = current_state->offset;      int state_offset = current_state->offset;
606      int count, codevalue;      int count, codevalue, rrc;
607    
608  #ifdef DEBUG  #ifdef DEBUG
609      printf ("%.*sProcessing state %d c=", rlevel*2-2, SP, state_offset);      printf ("%.*sProcessing state %d c=", rlevel*2-2, SP, state_offset);
# Line 543  for (;;) Line 635  for (;;)
635          }          }
636        }        }
637    
638      /* Check for a duplicate state with the same count, and skip if found. */      /* Check for a duplicate state with the same count, and skip if found.
639        See the note at the head of this module about the possibility of improving
640        performance here. */
641    
642      for (j = 0; j < i; j++)      for (j = 0; j < i; j++)
643        {        {
# Line 560  for (;;) Line 654  for (;;)
654      code = start_code + state_offset;      code = start_code + state_offset;
655      codevalue = *code;      codevalue = *code;
656    
657        /* If this opcode inspects a character, but we are at the end of the
658        subject, remember the fact for use when testing for a partial match. */
659    
660        if (clen == 0 && poptable[codevalue] != 0)
661          could_continue = TRUE;
662    
663      /* If this opcode is followed by an inline character, load it. It is      /* If this opcode is followed by an inline character, load it. It is
664      tempting to test for the presence of a subject character here, but that      tempting to test for the presence of a subject character here, but that
665      is wrong, because sometimes zero repetitions of the subject are      is wrong, because sometimes zero repetitions of the subject are
# Line 610  for (;;) Line 710  for (;;)
710  /* ========================================================================== */  /* ========================================================================== */
711        /* Reached a closing bracket. If not at the end of the pattern, carry        /* Reached a closing bracket. If not at the end of the pattern, carry
712        on with the next opcode. Otherwise, unless we have an empty string and        on with the next opcode. Otherwise, unless we have an empty string and
713        PCRE_NOTEMPTY is set, save the match data, shifting up all previous        PCRE_NOTEMPTY is set, or PCRE_NOTEMPTY_ATSTART is set and we are at the
714          start of the subject, save the match data, shifting up all previous
715        matches so we always have the longest first. */        matches so we always have the longest first. */
716    
717        case OP_KET:        case OP_KET:
# Line 624  for (;;) Line 725  for (;;)
725            ADD_ACTIVE(state_offset - GET(code, 1), 0);            ADD_ACTIVE(state_offset - GET(code, 1), 0);
726            }            }
727          }          }
728        else if (ptr > current_subject || (md->moptions & PCRE_NOTEMPTY) == 0)        else
729          {          {
730          if (match_count < 0) match_count = (offsetcount >= 2)? 1 : 0;          if (ptr > current_subject ||
731            else if (match_count > 0 && ++match_count * 2 >= offsetcount)              ((md->moptions & PCRE_NOTEMPTY) == 0 &&
732              match_count = 0;                ((md->moptions & PCRE_NOTEMPTY_ATSTART) == 0 ||
733          count = ((match_count == 0)? offsetcount : match_count * 2) - 2;                  current_subject > start_subject + md->start_offset)))
734          if (count > 0) memmove(offsets + 2, offsets, count * sizeof(int));            {
735          if (offsetcount >= 2)            if (match_count < 0) match_count = (offsetcount >= 2)? 1 : 0;
736            {              else if (match_count > 0 && ++match_count * 2 >= offsetcount)
737            offsets[0] = current_subject - start_subject;                match_count = 0;
738            offsets[1] = ptr - start_subject;            count = ((match_count == 0)? offsetcount : match_count * 2) - 2;
739            DPRINTF(("%.*sSet matched string = \"%.*s\"\n", rlevel*2-2, SP,            if (count > 0) memmove(offsets + 2, offsets, count * sizeof(int));
740              offsets[1] - offsets[0], current_subject));            if (offsetcount >= 2)
741            }              {
742          if ((md->moptions & PCRE_DFA_SHORTEST) != 0)              offsets[0] = current_subject - start_subject;
743            {              offsets[1] = ptr - start_subject;
744            DPRINTF(("%.*sEnd of internal_dfa_exec %d: returning %d\n"              DPRINTF(("%.*sSet matched string = \"%.*s\"\n", rlevel*2-2, SP,
745              "%.*s---------------------\n\n", rlevel*2-2, SP, rlevel,                offsets[1] - offsets[0], current_subject));
746              match_count, rlevel*2-2, SP));              }
747            return match_count;            if ((md->moptions & PCRE_DFA_SHORTEST) != 0)
748                {
749                DPRINTF(("%.*sEnd of internal_dfa_exec %d: returning %d\n"
750                  "%.*s---------------------\n\n", rlevel*2-2, SP, rlevel,
751                  match_count, rlevel*2-2, SP));
752                return match_count;
753                }
754            }            }
755          }          }
756        break;        break;
# Line 757  for (;;) Line 864  for (;;)
864        if ((md->moptions & PCRE_NOTEOL) == 0)        if ((md->moptions & PCRE_NOTEOL) == 0)
865          {          {
866          if (clen == 0 ||          if (clen == 0 ||
867              (IS_NEWLINE(ptr) &&              ((md->poptions & PCRE_DOLLAR_ENDONLY) == 0 && IS_NEWLINE(ptr) &&
868                 ((ims & PCRE_MULTILINE) != 0 || ptr == end_subject - md->nllen)                 ((ims & PCRE_MULTILINE) != 0 || ptr == end_subject - md->nllen)
869              ))              ))
870            { ADD_ACTIVE(state_offset + 1, 0); }            { ADD_ACTIVE(state_offset + 1, 0); }
# Line 794  for (;;) Line 901  for (;;)
901          if (ptr > start_subject)          if (ptr > start_subject)
902            {            {
903            const uschar *temp = ptr - 1;            const uschar *temp = ptr - 1;
904              if (temp < md->start_used_ptr) md->start_used_ptr = temp;
905  #ifdef SUPPORT_UTF8  #ifdef SUPPORT_UTF8
906            if (utf8) BACKCHAR(temp);            if (utf8) BACKCHAR(temp);
907  #endif  #endif
# Line 802  for (;;) Line 910  for (;;)
910            }            }
911          else left_word = 0;          else left_word = 0;
912    
913          if (clen > 0) right_word = c < 256 && (ctypes[c] & ctype_word) != 0;          if (clen > 0)
914            else right_word = 0;            right_word = c < 256 && (ctypes[c] & ctype_word) != 0;
915            else right_word = 0;
916    
917          if ((left_word == right_word) == (codevalue == OP_NOT_WORD_BOUNDARY))          if ((left_word == right_word) == (codevalue == OP_NOT_WORD_BOUNDARY))
918            { ADD_ACTIVE(state_offset + 1, 0); }            { ADD_ACTIVE(state_offset + 1, 0); }
# Line 2157  for (;;) Line 2266  for (;;)
2266    
2267  /* ========================================================================== */  /* ========================================================================== */
2268        /* These are the opcodes for fancy brackets of various kinds. We have        /* These are the opcodes for fancy brackets of various kinds. We have
2269        to use recursion in order to handle them. The "always failing" assersion        to use recursion in order to handle them. The "always failing" assertion
2270        (?!) is optimised when compiling to OP_FAIL, so we have to support that,        (?!) is optimised to OP_FAIL when compiling, so we have to support that,
2271        though the other "backtracking verbs" are not supported. */        though the other "backtracking verbs" are not supported. */
2272    
2273        case OP_FAIL:        case OP_FAIL:
2274          forced_fail++;    /* Count FAILs for multiple states */
2275        break;        break;
2276    
2277        case OP_ASSERT:        case OP_ASSERT:
# Line 2200  for (;;) Line 2310  for (;;)
2310          {          {
2311          int local_offsets[1000];          int local_offsets[1000];
2312          int local_workspace[1000];          int local_workspace[1000];
2313          int condcode = code[LINK_SIZE+1];          int codelink = GET(code, 1);
2314            int condcode;
2315    
2316            /* Because of the way auto-callout works during compile, a callout item
2317            is inserted between OP_COND and an assertion condition. This does not
2318            happen for the other conditions. */
2319    
2320            if (code[LINK_SIZE+1] == OP_CALLOUT)
2321              {
2322              rrc = 0;
2323              if (pcre_callout != NULL)
2324                {
2325                pcre_callout_block cb;
2326                cb.version          = 1;   /* Version 1 of the callout block */
2327                cb.callout_number   = code[LINK_SIZE+2];
2328                cb.offset_vector    = offsets;
2329                cb.subject          = (PCRE_SPTR)start_subject;
2330                cb.subject_length   = end_subject - start_subject;
2331                cb.start_match      = current_subject - start_subject;
2332                cb.current_position = ptr - start_subject;
2333                cb.pattern_position = GET(code, LINK_SIZE + 3);
2334                cb.next_item_length = GET(code, 3 + 2*LINK_SIZE);
2335                cb.capture_top      = 1;
2336                cb.capture_last     = -1;
2337                cb.callout_data     = md->callout_data;
2338                if ((rrc = (*pcre_callout)(&cb)) < 0) return rrc;   /* Abandon */
2339                }
2340              if (rrc > 0) break;                      /* Fail this thread */
2341              code += _pcre_OP_lengths[OP_CALLOUT];    /* Skip callout data */
2342              }
2343    
2344            condcode = code[LINK_SIZE+1];
2345    
2346          /* Back reference conditions are not supported */          /* Back reference conditions are not supported */
2347    
2348          if (condcode == OP_CREF) return PCRE_ERROR_DFA_UCOND;          if (condcode == OP_CREF || condcode == OP_NCREF)
2349              return PCRE_ERROR_DFA_UCOND;
2350    
2351          /* The DEFINE condition is always false */          /* The DEFINE condition is always false */
2352    
2353          if (condcode == OP_DEF)          if (condcode == OP_DEF)
2354            {            { ADD_ACTIVE(state_offset + codelink + LINK_SIZE + 1, 0); }
           ADD_ACTIVE(state_offset + GET(code, 1) + LINK_SIZE + 1, 0);  
           }  
2355    
2356          /* The only supported version of OP_RREF is for the value RREF_ANY,          /* The only supported version of OP_RREF is for the value RREF_ANY,
2357          which means "test if in any recursion". We can't test for specifically          which means "test if in any recursion". We can't test for specifically
2358          recursed groups. */          recursed groups. */
2359    
2360          else if (condcode == OP_RREF)          else if (condcode == OP_RREF || condcode == OP_NRREF)
2361            {            {
2362            int value = GET2(code, LINK_SIZE+2);            int value = GET2(code, LINK_SIZE+2);
2363            if (value != RREF_ANY) return PCRE_ERROR_DFA_UCOND;            if (value != RREF_ANY) return PCRE_ERROR_DFA_UCOND;
2364            if (recursing > 0) { ADD_ACTIVE(state_offset + LINK_SIZE + 4, 0); }            if (recursing > 0)
2365              else { ADD_ACTIVE(state_offset + GET(code, 1) + LINK_SIZE + 1, 0); }              { ADD_ACTIVE(state_offset + LINK_SIZE + 4, 0); }
2366              else { ADD_ACTIVE(state_offset + codelink + LINK_SIZE + 1, 0); }
2367            }            }
2368    
2369          /* Otherwise, the condition is an assertion */          /* Otherwise, the condition is an assertion */
# Line 2252  for (;;) Line 2393  for (;;)
2393                  (condcode == OP_ASSERT || condcode == OP_ASSERTBACK))                  (condcode == OP_ASSERT || condcode == OP_ASSERTBACK))
2394              { ADD_ACTIVE(endasscode + LINK_SIZE + 1 - start_code, 0); }              { ADD_ACTIVE(endasscode + LINK_SIZE + 1 - start_code, 0); }
2395            else            else
2396              { ADD_ACTIVE(state_offset + GET(code, 1) + LINK_SIZE + 1, 0); }              { ADD_ACTIVE(state_offset + codelink + LINK_SIZE + 1, 0); }
2397            }            }
2398          }          }
2399        break;        break;
# Line 2404  for (;;) Line 2545  for (;;)
2545        /* Handle callouts */        /* Handle callouts */
2546    
2547        case OP_CALLOUT:        case OP_CALLOUT:
2548          rrc = 0;
2549        if (pcre_callout != NULL)        if (pcre_callout != NULL)
2550          {          {
         int rrc;  
2551          pcre_callout_block cb;          pcre_callout_block cb;
2552          cb.version          = 1;   /* Version 1 of the callout block */          cb.version          = 1;   /* Version 1 of the callout block */
2553          cb.callout_number   = code[1];          cb.callout_number   = code[1];
# Line 2421  for (;;) Line 2562  for (;;)
2562          cb.capture_last     = -1;          cb.capture_last     = -1;
2563          cb.callout_data     = md->callout_data;          cb.callout_data     = md->callout_data;
2564          if ((rrc = (*pcre_callout)(&cb)) < 0) return rrc;   /* Abandon */          if ((rrc = (*pcre_callout)(&cb)) < 0) return rrc;   /* Abandon */
         if (rrc == 0) { ADD_ACTIVE(state_offset + 2 + 2*LINK_SIZE, 0); }  
2565          }          }
2566          if (rrc == 0)
2567            { ADD_ACTIVE(state_offset + _pcre_OP_lengths[OP_CALLOUT], 0); }
2568        break;        break;
2569    
2570    
# Line 2438  for (;;) Line 2580  for (;;)
2580    /* We have finished the processing at the current subject character. If no    /* We have finished the processing at the current subject character. If no
2581    new states have been set for the next character, we have found all the    new states have been set for the next character, we have found all the
2582    matches that we are going to find. If we are at the top level and partial    matches that we are going to find. If we are at the top level and partial
2583    matching has been requested, check for appropriate conditions. */    matching has been requested, check for appropriate conditions.
2584    
2585      The "forced_ fail" variable counts the number of (*F) encountered for the
2586      character. If it is equal to the original active_count (saved in
2587      workspace[1]) it means that (*F) was found on every active state. In this
2588      case we don't want to give a partial match.
2589    
2590      The "could_continue" variable is true if a state could have continued but
2591      for the fact that the end of the subject was reached. */
2592    
2593    if (new_count <= 0)    if (new_count <= 0)
2594      {      {
2595      if (match_count < 0 &&                     /* No matches found */      if (rlevel == 1 &&                               /* Top level, and */
2596          rlevel == 1 &&                         /* Top level match function */          could_continue &&                            /* Some could go on */
2597          (md->moptions & PCRE_PARTIAL) != 0 &&  /* Want partial matching */          forced_fail != workspace[1] &&               /* Not all forced fail & */
2598          ptr >= end_subject &&                  /* Reached end of subject */          (                                            /* either... */
2599          ptr > current_subject)                 /* Matched non-empty string */          (md->moptions & PCRE_PARTIAL_HARD) != 0      /* Hard partial */
2600            ||                                           /* or... */
2601            ((md->moptions & PCRE_PARTIAL_SOFT) != 0 &&  /* Soft partial and */
2602             match_count < 0)                            /* no matches */
2603            ) &&                                         /* And... */
2604            ptr >= end_subject &&                     /* Reached end of subject */
2605            ptr > current_subject)                    /* Matched non-empty string */
2606        {        {
2607        if (offsetcount >= 2)        if (offsetcount >= 2)
2608          {          {
2609          offsets[0] = current_subject - start_subject;          offsets[0] = md->start_used_ptr - start_subject;
2610          offsets[1] = end_subject - start_subject;          offsets[1] = end_subject - start_subject;
2611          }          }
2612        match_count = PCRE_ERROR_PARTIAL;        match_count = PCRE_ERROR_PARTIAL;
# Line 2592  md->start_code = (const uschar *)argumen Line 2748  md->start_code = (const uschar *)argumen
2748      re->name_table_offset + re->name_count * re->name_entry_size;      re->name_table_offset + re->name_count * re->name_entry_size;
2749  md->start_subject = (const unsigned char *)subject;  md->start_subject = (const unsigned char *)subject;
2750  md->end_subject = end_subject;  md->end_subject = end_subject;
2751    md->start_offset = start_offset;
2752  md->moptions = options;  md->moptions = options;
2753  md->poptions = re->options;  md->poptions = re->options;
2754    
# Line 2614  switch ((((options & PCRE_NEWLINE_BITS) Line 2771  switch ((((options & PCRE_NEWLINE_BITS)
2771           PCRE_NEWLINE_BITS)           PCRE_NEWLINE_BITS)
2772    {    {
2773    case 0: newline = NEWLINE; break;   /* Compile-time default */    case 0: newline = NEWLINE; break;   /* Compile-time default */
2774    case PCRE_NEWLINE_CR: newline = '\r'; break;    case PCRE_NEWLINE_CR: newline = CHAR_CR; break;
2775    case PCRE_NEWLINE_LF: newline = '\n'; break;    case PCRE_NEWLINE_LF: newline = CHAR_NL; break;
2776    case PCRE_NEWLINE_CR+    case PCRE_NEWLINE_CR+
2777         PCRE_NEWLINE_LF: newline = ('\r' << 8) | '\n'; break;         PCRE_NEWLINE_LF: newline = (CHAR_CR << 8) | CHAR_NL; break;
2778    case PCRE_NEWLINE_ANY: newline = -1; break;    case PCRE_NEWLINE_ANY: newline = -1; break;
2779    case PCRE_NEWLINE_ANYCRLF: newline = -2; break;    case PCRE_NEWLINE_ANYCRLF: newline = -2; break;
2780    default: return PCRE_ERROR_BADNEWLINE;    default: return PCRE_ERROR_BADNEWLINE;
# Line 2696  if (!anchored) Line 2853  if (!anchored)
2853      }      }
2854    else    else
2855      {      {
2856      if (startline && study != NULL &&      if (!startline && study != NULL &&
2857           (study->options & PCRE_STUDY_MAPPED) != 0)           (study->flags & PCRE_STUDY_MAPPED) != 0)
2858        start_bits = study->start_bits;        start_bits = study->start_bits;
2859      }      }
2860    }    }
# Line 2713  if ((re->flags & PCRE_REQCHSET) != 0) Line 2870  if ((re->flags & PCRE_REQCHSET) != 0)
2870    }    }
2871    
2872  /* Call the main matching function, looping for a non-anchored regex after a  /* Call the main matching function, looping for a non-anchored regex after a
2873  failed match. Unless restarting, optimize by moving to the first match  failed match. If not restarting, perform certain optimizations at the start of
2874  character if possible, when not anchored. Then unless wanting a partial match,  a match. */
 check for a required later character. */  
2875    
2876  for (;;)  for (;;)
2877    {    {
# Line 2725  for (;;) Line 2881  for (;;)
2881      {      {
2882      const uschar *save_end_subject = end_subject;      const uschar *save_end_subject = end_subject;
2883    
2884      /* Advance to a unique first char if possible. If firstline is TRUE, the      /* If firstline is TRUE, the start of the match is constrained to the first
2885      start of the match is constrained to the first line of a multiline string.      line of a multiline string. Implement this by temporarily adjusting
2886      Implement this by temporarily adjusting end_subject so that we stop      end_subject so that we stop scanning at a newline. If the match fails at
2887      scanning at a newline. If the match fails at the newline, later code breaks      the newline, later code breaks this loop. */
     this loop. */  
2888    
2889      if (firstline)      if (firstline)
2890        {        {
2891        const uschar *t = current_subject;        USPTR t = current_subject;
2892    #ifdef SUPPORT_UTF8
2893          if (utf8)
2894            {
2895            while (t < md->end_subject && !IS_NEWLINE(t))
2896              {
2897              t++;
2898              while (t < end_subject && (*t & 0xc0) == 0x80) t++;
2899              }
2900            }
2901          else
2902    #endif
2903        while (t < md->end_subject && !IS_NEWLINE(t)) t++;        while (t < md->end_subject && !IS_NEWLINE(t)) t++;
2904        end_subject = t;        end_subject = t;
2905        }        }
2906    
2907      if (first_byte >= 0)      /* There are some optimizations that avoid running the match if a known
2908        starting point is not found. However, there is an option that disables
2909        these, for testing and for ensuring that all callouts do actually occur. */
2910    
2911        if ((options & PCRE_NO_START_OPTIMIZE) == 0)
2912        {        {
2913        if (first_byte_caseless)        /* Advance to a known first byte. */
         while (current_subject < end_subject &&  
                lcc[*current_subject] != first_byte)  
           current_subject++;  
       else  
         while (current_subject < end_subject && *current_subject != first_byte)  
           current_subject++;  
       }  
2914    
2915      /* Or to just after a linebreak for a multiline match if possible */        if (first_byte >= 0)
2916            {
2917            if (first_byte_caseless)
2918              while (current_subject < end_subject &&
2919                     lcc[*current_subject] != first_byte)
2920                current_subject++;
2921            else
2922              while (current_subject < end_subject &&
2923                     *current_subject != first_byte)
2924                current_subject++;
2925            }
2926    
2927      else if (startline)        /* Or to just after a linebreak for a multiline match if possible */
2928        {  
2929        if (current_subject > md->start_subject + start_offset)        else if (startline)
2930          {          {
2931          while (current_subject < end_subject && !WAS_NEWLINE(current_subject))          if (current_subject > md->start_subject + start_offset)
2932            current_subject++;            {
2933    #ifdef SUPPORT_UTF8
2934              if (utf8)
2935                {
2936                while (current_subject < end_subject &&
2937                       !WAS_NEWLINE(current_subject))
2938                  {
2939                  current_subject++;
2940                  while(current_subject < end_subject &&
2941                        (*current_subject & 0xc0) == 0x80)
2942                    current_subject++;
2943                  }
2944                }
2945              else
2946    #endif
2947              while (current_subject < end_subject && !WAS_NEWLINE(current_subject))
2948                current_subject++;
2949    
2950          /* If we have just passed a CR and the newline option is ANY or            /* If we have just passed a CR and the newline option is ANY or
2951          ANYCRLF, and we are now at a LF, advance the match position by one more            ANYCRLF, and we are now at a LF, advance the match position by one
2952          character. */            more character. */
2953    
2954          if (current_subject[-1] == '\r' &&            if (current_subject[-1] == CHAR_CR &&
2955               (md->nltype == NLTYPE_ANY || md->nltype == NLTYPE_ANYCRLF) &&                 (md->nltype == NLTYPE_ANY || md->nltype == NLTYPE_ANYCRLF) &&
2956               current_subject < end_subject &&                 current_subject < end_subject &&
2957               *current_subject == '\n')                 *current_subject == CHAR_NL)
2958            current_subject++;              current_subject++;
2959              }
2960          }          }
       }  
2961    
2962      /* Or to a non-unique first char after study */        /* Or to a non-unique first char after study */
2963    
2964      else if (start_bits != NULL)        else if (start_bits != NULL)
       {  
       while (current_subject < end_subject)  
2965          {          {
2966          register unsigned int c = *current_subject;          while (current_subject < end_subject)
2967          if ((start_bits[c/8] & (1 << (c&7))) == 0) current_subject++;            {
2968            else break;            register unsigned int c = *current_subject;
2969              if ((start_bits[c/8] & (1 << (c&7))) == 0) current_subject++;
2970                else break;
2971              }
2972          }          }
2973        }        }
2974    
2975      /* Restore fudged end_subject */      /* Restore fudged end_subject */
2976    
2977      end_subject = save_end_subject;      end_subject = save_end_subject;
     }  
2978    
2979    /* If req_byte is set, we know that that character must appear in the subject      /* The following two optimizations are disabled for partial matching or if
2980    for the match to succeed. If the first character is set, req_byte must be      disabling is explicitly requested (and of course, by the test above, this
2981    later in the subject; otherwise the test starts at the match point. This      code is not obeyed when restarting after a partial match). */
   optimization can save a huge amount of work in patterns with nested unlimited  
   repeats that aren't going to match. Writing separate code for cased/caseless  
   versions makes it go faster, as does using an autoincrement and backing off  
   on a match.  
   
   HOWEVER: when the subject string is very, very long, searching to its end can  
   take a long time, and give bad performance on quite ordinary patterns. This  
   showed up when somebody was matching /^C/ on a 32-megabyte string... so we  
   don't do this when the string is sufficiently long.  
   
   ALSO: this processing is disabled when partial matching is requested.  
   */  
   
   if (req_byte >= 0 &&  
       end_subject - current_subject < REQ_BYTE_MAX &&  
       (options & PCRE_PARTIAL) == 0)  
     {  
     register const uschar *p = current_subject + ((first_byte >= 0)? 1 : 0);  
   
     /* We don't need to repeat the search if we haven't yet reached the  
     place we found it at last time. */  
2982    
2983      if (p > req_byte_ptr)      if ((options & PCRE_NO_START_OPTIMIZE) == 0 &&
2984            (options & (PCRE_PARTIAL_HARD|PCRE_PARTIAL_SOFT)) == 0)
2985        {        {
2986        if (req_byte_caseless)        /* If the pattern was studied, a minimum subject length may be set. This
2987          {        is a lower bound; no actual string of that length may actually match the
2988          while (p < end_subject)        pattern. Although the value is, strictly, in characters, we treat it as
2989            {        bytes to avoid spending too much time in this optimization. */
2990            register int pp = *p++;  
2991            if (pp == req_byte || pp == req_byte2) { p--; break; }        if (study != NULL && (study->flags & PCRE_STUDY_MINLEN) != 0 &&
2992            }            end_subject - current_subject < study->minlength)
2993          }          return PCRE_ERROR_NOMATCH;
2994        else  
2995          /* If req_byte is set, we know that that character must appear in the
2996          subject for the match to succeed. If the first character is set, req_byte
2997          must be later in the subject; otherwise the test starts at the match
2998          point. This optimization can save a huge amount of work in patterns with
2999          nested unlimited repeats that aren't going to match. Writing separate
3000          code for cased/caseless versions makes it go faster, as does using an
3001          autoincrement and backing off on a match.
3002    
3003          HOWEVER: when the subject string is very, very long, searching to its end
3004          can take a long time, and give bad performance on quite ordinary
3005          patterns. This showed up when somebody was matching /^C/ on a 32-megabyte
3006          string... so we don't do this when the string is sufficiently long. */
3007    
3008          if (req_byte >= 0 && end_subject - current_subject < REQ_BYTE_MAX)
3009          {          {
3010          while (p < end_subject)          register const uschar *p = current_subject + ((first_byte >= 0)? 1 : 0);
3011    
3012            /* We don't need to repeat the search if we haven't yet reached the
3013            place we found it at last time. */
3014    
3015            if (p > req_byte_ptr)
3016            {            {
3017            if (*p++ == req_byte) { p--; break; }            if (req_byte_caseless)
3018            }              {
3019          }              while (p < end_subject)
3020                  {
3021                  register int pp = *p++;
3022                  if (pp == req_byte || pp == req_byte2) { p--; break; }
3023                  }
3024                }
3025              else
3026                {
3027                while (p < end_subject)
3028                  {
3029                  if (*p++ == req_byte) { p--; break; }
3030                  }
3031                }
3032    
3033        /* If we can't find the required character, break the matching loop,            /* If we can't find the required character, break the matching loop,
3034        which will cause a return or PCRE_ERROR_NOMATCH. */            which will cause a return or PCRE_ERROR_NOMATCH. */
3035    
3036        if (p >= end_subject) break;            if (p >= end_subject) break;
3037    
3038        /* If we have found the required character, save the point where we            /* If we have found the required character, save the point where we
3039        found it, so that we don't search again next time round the loop if            found it, so that we don't search again next time round the loop if
3040        the start hasn't passed this character yet. */            the start hasn't passed this character yet. */
3041    
3042        req_byte_ptr = p;            req_byte_ptr = p;
3043              }
3044            }
3045        }        }
3046      }      }   /* End of optimizations that are done when not restarting */
3047    
3048    /* OK, now we can do the business */    /* OK, now we can do the business */
3049    
3050      md->start_used_ptr = current_subject;
3051    
3052    rc = internal_dfa_exec(    rc = internal_dfa_exec(
3053      md,                                /* fixed match data */      md,                                /* fixed match data */
3054      md->start_code,                    /* this subexpression's code */      md->start_code,                    /* this subexpression's code */
# Line 2879  for (;;) Line 3083  for (;;)
3083    not contain any explicit matches for \r or \n, and the newline option is CRLF    not contain any explicit matches for \r or \n, and the newline option is CRLF
3084    or ANY or ANYCRLF, advance the match position by one more character. */    or ANY or ANYCRLF, advance the match position by one more character. */
3085    
3086    if (current_subject[-1] == '\r' &&    if (current_subject[-1] == CHAR_CR &&
3087        current_subject < end_subject &&        current_subject < end_subject &&
3088        *current_subject == '\n' &&        *current_subject == CHAR_NL &&
3089        (re->flags & PCRE_HASCRORLF) == 0 &&        (re->flags & PCRE_HASCRORLF) == 0 &&
3090          (md->nltype == NLTYPE_ANY ||          (md->nltype == NLTYPE_ANY ||
3091           md->nltype == NLTYPE_ANYCRLF ||           md->nltype == NLTYPE_ANYCRLF ||

Legend:
Removed from v.361  
changed lines
  Added in v.463

  ViewVC Help
Powered by ViewVC 1.1.5