/[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 365 by ph10, Fri Jul 11 17:06:55 2008 UTC revision 462 by ph10, Sat Oct 17 19:55:02 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, in order to support PCRE_PARTIAL_HARD behaviour.
170    ***NOTE*** If the start of this table is modified, the two tables that follow
171    must also be modified. */
172    
173    static const uschar poptable[] = {
174      0,                             /* End                                    */
175      0, 0, 0, 0, 0,                 /* \A, \G, \K, \B, \b                     */
176      1, 1, 1, 1, 1, 1,              /* \D, \d, \S, \s, \W, \w                 */
177      1, 1, 1,                       /* Any, AllAny, Anybyte                   */
178      1, 1, 1,                       /* NOTPROP, PROP, EXTUNI                  */
179      1, 1, 1, 1, 1,                 /* \R, \H, \h, \V, \v                     */
180      0, 0, 0, 0, 0,                 /* \Z, \z, Opt, ^, $                      */
181      1,                             /* Char                                   */
182      1,                             /* Charnc                                 */
183      1,                             /* not                                    */
184      /* Positive single-char repeats                                          */
185      1, 1, 1, 1, 1, 1,              /* *, *?, +, +?, ?, ??                    */
186      1, 1, 1,                       /* upto, minupto, exact                   */
187      1, 1, 1, 1,                    /* *+, ++, ?+, upto+                      */
188      /* Negative single-char repeats - only for chars < 256                   */
189      1, 1, 1, 1, 1, 1,              /* NOT *, *?, +, +?, ?, ??                */
190      1, 1, 1,                       /* NOT upto, minupto, exact               */
191      1, 1, 1, 1,                    /* NOT *+, ++, ?+, upto+                  */
192      /* Positive type repeats                                                 */
193      1, 1, 1, 1, 1, 1,              /* Type *, *?, +, +?, ?, ??               */
194      1, 1, 1,                       /* Type upto, minupto, exact              */
195      1, 1, 1, 1,                    /* Type *+, ++, ?+, upto+                 */
196      /* Character class & ref repeats                                         */
197      1, 1, 1, 1, 1, 1,              /* *, *?, +, +?, ?, ??                    */
198      1, 1,                          /* CRRANGE, CRMINRANGE                    */
199      1,                             /* CLASS                                  */
200      1,                             /* NCLASS                                 */
201      1,                             /* XCLASS - variable length               */
202      0,                             /* REF                                    */
203      0,                             /* RECURSE                                */
204      0,                             /* CALLOUT                                */
205      0,                             /* Alt                                    */
206      0,                             /* Ket                                    */
207      0,                             /* KetRmax                                */
208      0,                             /* KetRmin                                */
209      0,                             /* Assert                                 */
210      0,                             /* Assert not                             */
211      0,                             /* Assert behind                          */
212      0,                             /* Assert behind not                      */
213      0,                             /* Reverse                                */
214      0, 0, 0, 0,                    /* ONCE, BRA, CBRA, COND                  */
215      0, 0, 0,                       /* SBRA, SCBRA, SCOND                     */
216      0,                             /* CREF                                   */
217      0,                             /* RREF                                   */
218      0,                             /* DEF                                    */
219      0, 0,                          /* BRAZERO, BRAMINZERO                    */
220      0, 0, 0, 0,                    /* PRUNE, SKIP, THEN, COMMIT              */
221      0, 0, 0, 0                     /* FAIL, ACCEPT, CLOSE, SKIPZERO          */
222  };  };
223    
224  /* 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 476  if (*first_op == OP_REVERSE)
476      current_subject -= gone_back;      current_subject -= gone_back;
477      }      }
478    
479      /* Save the earliest consulted character */
480    
481      if (current_subject < md->start_used_ptr)
482        md->start_used_ptr = current_subject;
483    
484    /* Now we can process the individual branches. */    /* Now we can process the individual branches. */
485    
486    end_code = this_start_code;    end_code = this_start_code;
# Line 454  for (;;) Line 545  for (;;)
545    int i, j;    int i, j;
546    int clen, dlen;    int clen, dlen;
547    unsigned int c, d;    unsigned int c, d;
548      int forced_fail = 0;
549      int reached_end = 0;
550      BOOL could_continue = FALSE;
551    
552    /* 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
553    new state list. */    new state list. */
# Line 511  for (;;) Line 605  for (;;)
605      stateblock *current_state = active_states + i;      stateblock *current_state = active_states + i;
606      const uschar *code;      const uschar *code;
607      int state_offset = current_state->offset;      int state_offset = current_state->offset;
608      int count, codevalue;      int count, codevalue, rrc;
609    
610  #ifdef DEBUG  #ifdef DEBUG
611      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 637  for (;;)
637          }          }
638        }        }
639    
640      /* 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.
641        See the note at the head of this module about the possibility of improving
642        performance here. */
643    
644      for (j = 0; j < i; j++)      for (j = 0; j < i; j++)
645        {        {
# Line 559  for (;;) Line 655  for (;;)
655    
656      code = start_code + state_offset;      code = start_code + state_offset;
657      codevalue = *code;      codevalue = *code;
658    
659        /* If this opcode inspects a character, but we are at the end of the
660        subject, remember the fact so that we can support PCRE_PARTIAL_HARD. */
661    
662        if (clen == 0 && poptable[codevalue] != 0)
663          could_continue = TRUE;
664    
665      /* 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
666      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
# Line 610  for (;;) Line 712  for (;;)
712  /* ========================================================================== */  /* ========================================================================== */
713        /* 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
714        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
715        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
716          start of the subject, save the match data, shifting up all previous
717        matches so we always have the longest first. */        matches so we always have the longest first. */
718    
719        case OP_KET:        case OP_KET:
# Line 624  for (;;) Line 727  for (;;)
727            ADD_ACTIVE(state_offset - GET(code, 1), 0);            ADD_ACTIVE(state_offset - GET(code, 1), 0);
728            }            }
729          }          }
730        else if (ptr > current_subject || (md->moptions & PCRE_NOTEMPTY) == 0)        else
731          {          {
732          if (match_count < 0) match_count = (offsetcount >= 2)? 1 : 0;          reached_end++;    /* Count branches that reach the end */
733            else if (match_count > 0 && ++match_count * 2 >= offsetcount)          if (ptr > current_subject ||
734              match_count = 0;              ((md->moptions & PCRE_NOTEMPTY) == 0 &&
735          count = ((match_count == 0)? offsetcount : match_count * 2) - 2;                ((md->moptions & PCRE_NOTEMPTY_ATSTART) == 0 ||
736          if (count > 0) memmove(offsets + 2, offsets, count * sizeof(int));                  current_subject > start_subject + md->start_offset)))
737          if (offsetcount >= 2)            {
738            {            if (match_count < 0) match_count = (offsetcount >= 2)? 1 : 0;
739            offsets[0] = current_subject - start_subject;              else if (match_count > 0 && ++match_count * 2 >= offsetcount)
740            offsets[1] = ptr - start_subject;                match_count = 0;
741            DPRINTF(("%.*sSet matched string = \"%.*s\"\n", rlevel*2-2, SP,            count = ((match_count == 0)? offsetcount : match_count * 2) - 2;
742              offsets[1] - offsets[0], current_subject));            if (count > 0) memmove(offsets + 2, offsets, count * sizeof(int));
743            }            if (offsetcount >= 2)
744          if ((md->moptions & PCRE_DFA_SHORTEST) != 0)              {
745            {              offsets[0] = current_subject - start_subject;
746            DPRINTF(("%.*sEnd of internal_dfa_exec %d: returning %d\n"              offsets[1] = ptr - start_subject;
747              "%.*s---------------------\n\n", rlevel*2-2, SP, rlevel,              DPRINTF(("%.*sSet matched string = \"%.*s\"\n", rlevel*2-2, SP,
748              match_count, rlevel*2-2, SP));                offsets[1] - offsets[0], current_subject));
749            return match_count;              }
750              if ((md->moptions & PCRE_DFA_SHORTEST) != 0)
751                {
752                DPRINTF(("%.*sEnd of internal_dfa_exec %d: returning %d\n"
753                  "%.*s---------------------\n\n", rlevel*2-2, SP, rlevel,
754                  match_count, rlevel*2-2, SP));
755                return match_count;
756                }
757            }            }
758          }          }
759        break;        break;
# Line 757  for (;;) Line 867  for (;;)
867        if ((md->moptions & PCRE_NOTEOL) == 0)        if ((md->moptions & PCRE_NOTEOL) == 0)
868          {          {
869          if (clen == 0 ||          if (clen == 0 ||
870              (IS_NEWLINE(ptr) &&              ((md->poptions & PCRE_DOLLAR_ENDONLY) == 0 && IS_NEWLINE(ptr) &&
871                 ((ims & PCRE_MULTILINE) != 0 || ptr == end_subject - md->nllen)                 ((ims & PCRE_MULTILINE) != 0 || ptr == end_subject - md->nllen)
872              ))              ))
873            { ADD_ACTIVE(state_offset + 1, 0); }            { ADD_ACTIVE(state_offset + 1, 0); }
# Line 794  for (;;) Line 904  for (;;)
904          if (ptr > start_subject)          if (ptr > start_subject)
905            {            {
906            const uschar *temp = ptr - 1;            const uschar *temp = ptr - 1;
907              if (temp < md->start_used_ptr) md->start_used_ptr = temp;
908  #ifdef SUPPORT_UTF8  #ifdef SUPPORT_UTF8
909            if (utf8) BACKCHAR(temp);            if (utf8) BACKCHAR(temp);
910  #endif  #endif
# Line 802  for (;;) Line 913  for (;;)
913            }            }
914          else left_word = 0;          else left_word = 0;
915    
916          if (clen > 0) right_word = c < 256 && (ctypes[c] & ctype_word) != 0;          if (clen > 0)
917            else right_word = 0;            right_word = c < 256 && (ctypes[c] & ctype_word) != 0;
918            else              /* This is a fudge to ensure that if this is the */
919              {               /* last item in the pattern, we don't count it as */
920              reached_end--;  /* reached, thus disabling a partial match. */
921              right_word = 0;
922              }
923    
924          if ((left_word == right_word) == (codevalue == OP_NOT_WORD_BOUNDARY))          if ((left_word == right_word) == (codevalue == OP_NOT_WORD_BOUNDARY))
925            { ADD_ACTIVE(state_offset + 1, 0); }            { ADD_ACTIVE(state_offset + 1, 0); }
# Line 2157  for (;;) Line 2273  for (;;)
2273    
2274  /* ========================================================================== */  /* ========================================================================== */
2275        /* These are the opcodes for fancy brackets of various kinds. We have        /* These are the opcodes for fancy brackets of various kinds. We have
2276        to use recursion in order to handle them. The "always failing" assersion        to use recursion in order to handle them. The "always failing" assertion
2277        (?!) 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,
2278        though the other "backtracking verbs" are not supported. */        though the other "backtracking verbs" are not supported. */
2279    
2280        case OP_FAIL:        case OP_FAIL:
2281          forced_fail++;    /* Count FAILs for multiple states */
2282        break;        break;
2283    
2284        case OP_ASSERT:        case OP_ASSERT:
# Line 2200  for (;;) Line 2317  for (;;)
2317          {          {
2318          int local_offsets[1000];          int local_offsets[1000];
2319          int local_workspace[1000];          int local_workspace[1000];
2320          int condcode = code[LINK_SIZE+1];          int codelink = GET(code, 1);
2321            int condcode;
2322    
2323            /* Because of the way auto-callout works during compile, a callout item
2324            is inserted between OP_COND and an assertion condition. This does not
2325            happen for the other conditions. */
2326    
2327            if (code[LINK_SIZE+1] == OP_CALLOUT)
2328              {
2329              rrc = 0;
2330              if (pcre_callout != NULL)
2331                {
2332                pcre_callout_block cb;
2333                cb.version          = 1;   /* Version 1 of the callout block */
2334                cb.callout_number   = code[LINK_SIZE+2];
2335                cb.offset_vector    = offsets;
2336                cb.subject          = (PCRE_SPTR)start_subject;
2337                cb.subject_length   = end_subject - start_subject;
2338                cb.start_match      = current_subject - start_subject;
2339                cb.current_position = ptr - start_subject;
2340                cb.pattern_position = GET(code, LINK_SIZE + 3);
2341                cb.next_item_length = GET(code, 3 + 2*LINK_SIZE);
2342                cb.capture_top      = 1;
2343                cb.capture_last     = -1;
2344                cb.callout_data     = md->callout_data;
2345                if ((rrc = (*pcre_callout)(&cb)) < 0) return rrc;   /* Abandon */
2346                }
2347              if (rrc > 0) break;                      /* Fail this thread */
2348              code += _pcre_OP_lengths[OP_CALLOUT];    /* Skip callout data */
2349              }
2350    
2351            condcode = code[LINK_SIZE+1];
2352    
2353          /* Back reference conditions are not supported */          /* Back reference conditions are not supported */
2354    
2355          if (condcode == OP_CREF) return PCRE_ERROR_DFA_UCOND;          if (condcode == OP_CREF || condcode == OP_NCREF)
2356              return PCRE_ERROR_DFA_UCOND;
2357    
2358          /* The DEFINE condition is always false */          /* The DEFINE condition is always false */
2359    
2360          if (condcode == OP_DEF)          if (condcode == OP_DEF)
2361            {            { ADD_ACTIVE(state_offset + codelink + LINK_SIZE + 1, 0); }
           ADD_ACTIVE(state_offset + GET(code, 1) + LINK_SIZE + 1, 0);  
           }  
2362    
2363          /* 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,
2364          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
2365          recursed groups. */          recursed groups. */
2366    
2367          else if (condcode == OP_RREF)          else if (condcode == OP_RREF || condcode == OP_NRREF)
2368            {            {
2369            int value = GET2(code, LINK_SIZE+2);            int value = GET2(code, LINK_SIZE+2);
2370            if (value != RREF_ANY) return PCRE_ERROR_DFA_UCOND;            if (value != RREF_ANY) return PCRE_ERROR_DFA_UCOND;
2371            if (recursing > 0) { ADD_ACTIVE(state_offset + LINK_SIZE + 4, 0); }            if (recursing > 0)
2372              else { ADD_ACTIVE(state_offset + GET(code, 1) + LINK_SIZE + 1, 0); }              { ADD_ACTIVE(state_offset + LINK_SIZE + 4, 0); }
2373              else { ADD_ACTIVE(state_offset + codelink + LINK_SIZE + 1, 0); }
2374            }            }
2375    
2376          /* Otherwise, the condition is an assertion */          /* Otherwise, the condition is an assertion */
# Line 2252  for (;;) Line 2400  for (;;)
2400                  (condcode == OP_ASSERT || condcode == OP_ASSERTBACK))                  (condcode == OP_ASSERT || condcode == OP_ASSERTBACK))
2401              { ADD_ACTIVE(endasscode + LINK_SIZE + 1 - start_code, 0); }              { ADD_ACTIVE(endasscode + LINK_SIZE + 1 - start_code, 0); }
2402            else            else
2403              { ADD_ACTIVE(state_offset + GET(code, 1) + LINK_SIZE + 1, 0); }              { ADD_ACTIVE(state_offset + codelink + LINK_SIZE + 1, 0); }
2404            }            }
2405          }          }
2406        break;        break;
# Line 2404  for (;;) Line 2552  for (;;)
2552        /* Handle callouts */        /* Handle callouts */
2553    
2554        case OP_CALLOUT:        case OP_CALLOUT:
2555          rrc = 0;
2556        if (pcre_callout != NULL)        if (pcre_callout != NULL)
2557          {          {
         int rrc;  
2558          pcre_callout_block cb;          pcre_callout_block cb;
2559          cb.version          = 1;   /* Version 1 of the callout block */          cb.version          = 1;   /* Version 1 of the callout block */
2560          cb.callout_number   = code[1];          cb.callout_number   = code[1];
# Line 2421  for (;;) Line 2569  for (;;)
2569          cb.capture_last     = -1;          cb.capture_last     = -1;
2570          cb.callout_data     = md->callout_data;          cb.callout_data     = md->callout_data;
2571          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); }  
2572          }          }
2573          if (rrc == 0)
2574            { ADD_ACTIVE(state_offset + _pcre_OP_lengths[OP_CALLOUT], 0); }
2575        break;        break;
2576    
2577    
# Line 2438  for (;;) Line 2587  for (;;)
2587    /* We have finished the processing at the current subject character. If no    /* We have finished the processing at the current subject character. If no
2588    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
2589    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
2590    matching has been requested, check for appropriate conditions. */    matching has been requested, check for appropriate conditions.
2591    
2592      The "forced_ fail" variable counts the number of (*F) encountered for the
2593      character. If it is equal to the original active_count (saved in
2594      workspace[1]) it means that (*F) was found on every active state. In this
2595      case we don't want to give a partial match.
2596    
2597      The "reached_end" variable counts the number of threads that have reached the
2598      end of the pattern. The "could_continue" variable is true if a thread could
2599      have continued but for the fact that the end of the subject was reached. */
2600    
2601    if (new_count <= 0)    if (new_count <= 0)
2602      {      {
2603      if (match_count < 0 &&                     /* No matches found */      if (rlevel == 1 &&                               /* Top level, and */
2604          rlevel == 1 &&                         /* Top level match function */          (                                            /* either... */
2605          (md->moptions & PCRE_PARTIAL) != 0 &&  /* Want partial matching */          reached_end != workspace[1] ||               /* Not all reached end */
2606          ptr >= end_subject &&                  /* Reached end of subject */            could_continue                             /* or some could go on */
2607          ptr > current_subject)                 /* Matched non-empty string */          ) &&                                         /* and... */
2608            forced_fail != workspace[1] &&               /* Not all forced fail & */
2609            (                                            /* either... */
2610            (md->moptions & PCRE_PARTIAL_HARD) != 0      /* Hard partial */
2611            ||                                           /* or... */
2612            ((md->moptions & PCRE_PARTIAL_SOFT) != 0 &&  /* Soft partial and */
2613             match_count < 0)                            /* no matches */
2614            ) &&                                         /* And... */
2615            ptr >= end_subject &&                     /* Reached end of subject */
2616            ptr > current_subject)                    /* Matched non-empty string */
2617        {        {
2618        if (offsetcount >= 2)        if (offsetcount >= 2)
2619          {          {
2620          offsets[0] = current_subject - start_subject;          offsets[0] = md->start_used_ptr - start_subject;
2621          offsets[1] = end_subject - start_subject;          offsets[1] = end_subject - start_subject;
2622          }          }
2623        match_count = PCRE_ERROR_PARTIAL;        match_count = PCRE_ERROR_PARTIAL;
# Line 2592  md->start_code = (const uschar *)argumen Line 2759  md->start_code = (const uschar *)argumen
2759      re->name_table_offset + re->name_count * re->name_entry_size;      re->name_table_offset + re->name_count * re->name_entry_size;
2760  md->start_subject = (const unsigned char *)subject;  md->start_subject = (const unsigned char *)subject;
2761  md->end_subject = end_subject;  md->end_subject = end_subject;
2762    md->start_offset = start_offset;
2763  md->moptions = options;  md->moptions = options;
2764  md->poptions = re->options;  md->poptions = re->options;
2765    
# Line 2614  switch ((((options & PCRE_NEWLINE_BITS) Line 2782  switch ((((options & PCRE_NEWLINE_BITS)
2782           PCRE_NEWLINE_BITS)           PCRE_NEWLINE_BITS)
2783    {    {
2784    case 0: newline = NEWLINE; break;   /* Compile-time default */    case 0: newline = NEWLINE; break;   /* Compile-time default */
2785    case PCRE_NEWLINE_CR: newline = '\r'; break;    case PCRE_NEWLINE_CR: newline = CHAR_CR; break;
2786    case PCRE_NEWLINE_LF: newline = '\n'; break;    case PCRE_NEWLINE_LF: newline = CHAR_NL; break;
2787    case PCRE_NEWLINE_CR+    case PCRE_NEWLINE_CR+
2788         PCRE_NEWLINE_LF: newline = ('\r' << 8) | '\n'; break;         PCRE_NEWLINE_LF: newline = (CHAR_CR << 8) | CHAR_NL; break;
2789    case PCRE_NEWLINE_ANY: newline = -1; break;    case PCRE_NEWLINE_ANY: newline = -1; break;
2790    case PCRE_NEWLINE_ANYCRLF: newline = -2; break;    case PCRE_NEWLINE_ANYCRLF: newline = -2; break;
2791    default: return PCRE_ERROR_BADNEWLINE;    default: return PCRE_ERROR_BADNEWLINE;
# Line 2696  if (!anchored) Line 2864  if (!anchored)
2864      }      }
2865    else    else
2866      {      {
2867      if (startline && study != NULL &&      if (!startline && study != NULL &&
2868           (study->options & PCRE_STUDY_MAPPED) != 0)           (study->flags & PCRE_STUDY_MAPPED) != 0)
2869        start_bits = study->start_bits;        start_bits = study->start_bits;
2870      }      }
2871    }    }
# Line 2713  if ((re->flags & PCRE_REQCHSET) != 0) Line 2881  if ((re->flags & PCRE_REQCHSET) != 0)
2881    }    }
2882    
2883  /* 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
2884  failed match. Unless restarting, optimize by moving to the first match  failed match. If not restarting, perform certain optimizations at the start of
2885  character if possible, when not anchored. Then unless wanting a partial match,  a match. */
 check for a required later character. */  
2886    
2887  for (;;)  for (;;)
2888    {    {
# Line 2725  for (;;) Line 2892  for (;;)
2892      {      {
2893      const uschar *save_end_subject = end_subject;      const uschar *save_end_subject = end_subject;
2894    
2895      /* 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
2896      start of the match is constrained to the first line of a multiline string.      line of a multiline string. Implement this by temporarily adjusting
2897      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
2898      scanning at a newline. If the match fails at the newline, later code breaks      the newline, later code breaks this loop. */
     this loop. */  
2899    
2900      if (firstline)      if (firstline)
2901        {        {
2902        USPTR t = current_subject;        USPTR t = current_subject;
2903  #ifdef SUPPORT_UTF8  #ifdef SUPPORT_UTF8
2904        if (utf8)        if (utf8)
2905          {          {
2906          while (t < md->end_subject && !IS_NEWLINE(t))          while (t < md->end_subject && !IS_NEWLINE(t))
2907            {            {
2908            t++;            t++;
2909            while (t < end_subject && (*t & 0xc0) == 0x80) t++;            while (t < end_subject && (*t & 0xc0) == 0x80) t++;
2910            }            }
2911          }          }
2912        else        else
2913  #endif  #endif
2914        while (t < md->end_subject && !IS_NEWLINE(t)) t++;        while (t < md->end_subject && !IS_NEWLINE(t)) t++;
2915        end_subject = t;        end_subject = t;
2916        }        }
2917    
2918      if (first_byte >= 0)      /* There are some optimizations that avoid running the match if a known
2919        starting point is not found. However, there is an option that disables
2920        these, for testing and for ensuring that all callouts do actually occur. */
2921    
2922        if ((options & PCRE_NO_START_OPTIMIZE) == 0)
2923        {        {
2924        if (first_byte_caseless)        /* Advance to a known first byte. */
2925          while (current_subject < end_subject &&  
2926                 lcc[*current_subject] != first_byte)        if (first_byte >= 0)
2927            current_subject++;          {
2928        else          if (first_byte_caseless)
2929          while (current_subject < end_subject && *current_subject != first_byte)            while (current_subject < end_subject &&
2930            current_subject++;                   lcc[*current_subject] != first_byte)
2931        }              current_subject++;
2932            else
2933              while (current_subject < end_subject &&
2934                     *current_subject != first_byte)
2935                current_subject++;
2936            }
2937    
2938      /* Or to just after a linebreak for a multiline match if possible */        /* Or to just after a linebreak for a multiline match if possible */
2939    
2940      else if (startline)        else if (startline)
       {  
       if (current_subject > md->start_subject + start_offset)  
2941          {          {
2942  #ifdef SUPPORT_UTF8          if (current_subject > md->start_subject + start_offset)
         if (utf8)  
2943            {            {
2944            while (current_subject < end_subject && !WAS_NEWLINE(current_subject))  #ifdef SUPPORT_UTF8
2945              if (utf8)
2946              {              {
2947              current_subject++;              while (current_subject < end_subject &&
2948              while(current_subject < end_subject &&                     !WAS_NEWLINE(current_subject))
2949                    (*current_subject & 0xc0) == 0x80)                {
2950                current_subject++;                current_subject++;
2951              }                while(current_subject < end_subject &&
2952                        (*current_subject & 0xc0) == 0x80)
2953                    current_subject++;
2954                  }
2955                }
2956              else
2957    #endif
2958              while (current_subject < end_subject && !WAS_NEWLINE(current_subject))
2959                current_subject++;
2960    
2961              /* If we have just passed a CR and the newline option is ANY or
2962              ANYCRLF, and we are now at a LF, advance the match position by one
2963              more character. */
2964    
2965              if (current_subject[-1] == CHAR_CR &&
2966                   (md->nltype == NLTYPE_ANY || md->nltype == NLTYPE_ANYCRLF) &&
2967                   current_subject < end_subject &&
2968                   *current_subject == CHAR_NL)
2969                current_subject++;
2970            }            }
         else  
 #endif  
         while (current_subject < end_subject && !WAS_NEWLINE(current_subject))  
           current_subject++;  
   
         /* If we have just passed a CR and the newline option is ANY or  
         ANYCRLF, and we are now at a LF, advance the match position by one more  
         character. */  
   
         if (current_subject[-1] == '\r' &&  
              (md->nltype == NLTYPE_ANY || md->nltype == NLTYPE_ANYCRLF) &&  
              current_subject < end_subject &&  
              *current_subject == '\n')  
           current_subject++;  
2971          }          }
       }  
2972    
2973      /* Or to a non-unique first char after study */        /* Or to a non-unique first char after study */
2974    
2975      else if (start_bits != NULL)        else if (start_bits != NULL)
       {  
       while (current_subject < end_subject)  
2976          {          {
2977          register unsigned int c = *current_subject;          while (current_subject < end_subject)
2978          if ((start_bits[c/8] & (1 << (c&7))) == 0) current_subject++;            {
2979            else break;            register unsigned int c = *current_subject;
2980              if ((start_bits[c/8] & (1 << (c&7))) == 0) current_subject++;
2981                else break;
2982              }
2983          }          }
2984        }        }
2985    
2986      /* Restore fudged end_subject */      /* Restore fudged end_subject */
2987    
2988      end_subject = save_end_subject;      end_subject = save_end_subject;
     }  
   
   /* If req_byte is set, we know that that character must appear in the subject  
   for the match to succeed. If the first character is set, req_byte must be  
   later in the subject; otherwise the test starts at the match point. This  
   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);  
2989    
2990      /* We don't need to repeat the search if we haven't yet reached the      /* The following two optimizations are disabled for partial matching or if
2991      place we found it at last time. */      disabling is explicitly requested (and of course, by the test above, this
2992        code is not obeyed when restarting after a partial match). */
2993    
2994      if (p > req_byte_ptr)      if ((options & PCRE_NO_START_OPTIMIZE) == 0 &&
2995            (options & (PCRE_PARTIAL_HARD|PCRE_PARTIAL_SOFT)) == 0)
2996        {        {
2997        if (req_byte_caseless)        /* If the pattern was studied, a minimum subject length may be set. This
2998          {        is a lower bound; no actual string of that length may actually match the
2999          while (p < end_subject)        pattern. Although the value is, strictly, in characters, we treat it as
3000            {        bytes to avoid spending too much time in this optimization. */
3001            register int pp = *p++;  
3002            if (pp == req_byte || pp == req_byte2) { p--; break; }        if (study != NULL && (study->flags & PCRE_STUDY_MINLEN) != 0 &&
3003            }            end_subject - current_subject < study->minlength)
3004          }          return PCRE_ERROR_NOMATCH;
3005        else  
3006          /* If req_byte is set, we know that that character must appear in the
3007          subject for the match to succeed. If the first character is set, req_byte
3008          must be later in the subject; otherwise the test starts at the match
3009          point. This optimization can save a huge amount of work in patterns with
3010          nested unlimited repeats that aren't going to match. Writing separate
3011          code for cased/caseless versions makes it go faster, as does using an
3012          autoincrement and backing off on a match.
3013    
3014          HOWEVER: when the subject string is very, very long, searching to its end
3015          can take a long time, and give bad performance on quite ordinary
3016          patterns. This showed up when somebody was matching /^C/ on a 32-megabyte
3017          string... so we don't do this when the string is sufficiently long. */
3018    
3019          if (req_byte >= 0 && end_subject - current_subject < REQ_BYTE_MAX)
3020          {          {
3021          while (p < end_subject)          register const uschar *p = current_subject + ((first_byte >= 0)? 1 : 0);
3022    
3023            /* We don't need to repeat the search if we haven't yet reached the
3024            place we found it at last time. */
3025    
3026            if (p > req_byte_ptr)
3027            {            {
3028            if (*p++ == req_byte) { p--; break; }            if (req_byte_caseless)
3029            }              {
3030          }              while (p < end_subject)
3031                  {
3032                  register int pp = *p++;
3033                  if (pp == req_byte || pp == req_byte2) { p--; break; }
3034                  }
3035                }
3036              else
3037                {
3038                while (p < end_subject)
3039                  {
3040                  if (*p++ == req_byte) { p--; break; }
3041                  }
3042                }
3043    
3044        /* If we can't find the required character, break the matching loop,            /* If we can't find the required character, break the matching loop,
3045        which will cause a return or PCRE_ERROR_NOMATCH. */            which will cause a return or PCRE_ERROR_NOMATCH. */
3046    
3047        if (p >= end_subject) break;            if (p >= end_subject) break;
3048    
3049        /* If we have found the required character, save the point where we            /* If we have found the required character, save the point where we
3050        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
3051        the start hasn't passed this character yet. */            the start hasn't passed this character yet. */
3052    
3053        req_byte_ptr = p;            req_byte_ptr = p;
3054              }
3055            }
3056        }        }
3057      }      }   /* End of optimizations that are done when not restarting */
3058    
3059    /* OK, now we can do the business */    /* OK, now we can do the business */
3060    
3061      md->start_used_ptr = current_subject;
3062    
3063    rc = internal_dfa_exec(    rc = internal_dfa_exec(
3064      md,                                /* fixed match data */      md,                                /* fixed match data */
3065      md->start_code,                    /* this subexpression's code */      md->start_code,                    /* this subexpression's code */
# Line 2903  for (;;) Line 3094  for (;;)
3094    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
3095    or ANY or ANYCRLF, advance the match position by one more character. */    or ANY or ANYCRLF, advance the match position by one more character. */
3096    
3097    if (current_subject[-1] == '\r' &&    if (current_subject[-1] == CHAR_CR &&
3098        current_subject < end_subject &&        current_subject < end_subject &&
3099        *current_subject == '\n' &&        *current_subject == CHAR_NL &&
3100        (re->flags & PCRE_HASCRORLF) == 0 &&        (re->flags & PCRE_HASCRORLF) == 0 &&
3101          (md->nltype == NLTYPE_ANY ||          (md->nltype == NLTYPE_ANY ||
3102           md->nltype == NLTYPE_ANYCRLF ||           md->nltype == NLTYPE_ANYCRLF ||

Legend:
Removed from v.365  
changed lines
  Added in v.462

  ViewVC Help
Powered by ViewVC 1.1.5