1 /*************************************************
2 * Perl-Compatible Regular Expressions *
3 *************************************************/
5 /* 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.
8 Written by Philip Hazel
9 Copyright (c) 1997-2014 University of Cambridge
11 -----------------------------------------------------------------------------
12 Redistribution and use in source and binary forms, with or without
13 modification, are permitted provided that the following conditions are met:
15 * Redistributions of source code must retain the above copyright notice,
16 this list of conditions and the following disclaimer.
18 * Redistributions in binary form must reproduce the above copyright
19 notice, this list of conditions and the following disclaimer in the
20 documentation and/or other materials provided with the distribution.
22 * Neither the name of the University of Cambridge nor the names of its
23 contributors may be used to endorse or promote products derived from
24 this software without specific prior written permission.
26 THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
27 AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28 IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29 ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE
30 LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
31 CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
32 SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
33 INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
34 CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
35 ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
36 POSSIBILITY OF SUCH DAMAGE.
37 -----------------------------------------------------------------------------
41 /* This module contains the external function pcre_compile(), along with
42 supporting internal functions that are not used by other modules. */
49 #define NLBLOCK cd /* Block containing newline information */
50 #define PSSTART start_pattern /* Field containing processed string start */
51 #define PSEND end_pattern /* Field containing processed string end */
53 #include "pcre_internal.h"
56 /* When PCRE_DEBUG is defined, we need the pcre(16|32)_printint() function, which
57 is also used by pcretest. PCRE_DEBUG is not defined when building a production
58 library. We do not need to select pcre16_printint.c specially, because the
59 COMPILE_PCREx macro will already be appropriately set. */
62 /* pcre_printint.c should not include any headers */
64 #include "pcre_printint.c"
69 /* Macro for setting individual bits in class bitmaps. */
71 #define SETBIT(a,b) a[(b)/8] |= (1 << ((b)&7))
73 /* Maximum length value to check against when making sure that the integer that
74 holds the compiled pattern length does not overflow. We make it a bit less than
75 INT_MAX to allow for adding in group terminating bytes, so that we don't have
76 to check them every time. */
78 #define OFLOW_MAX (INT_MAX - 20)
80 /* Definitions to allow mutual recursion */
83 add_list_to_class(pcre_uint8 *, pcre_uchar **, int, compile_data *,
84 const pcre_uint32 *, unsigned int);
87 compile_regex(int, pcre_uchar **, const pcre_uchar **, int *, BOOL, BOOL, int, int,
88 pcre_uint32 *, pcre_int32 *, pcre_uint32 *, pcre_int32 *, branch_chain *,
89 compile_data *, int *);
93 /*************************************************
94 * Code parameters and static tables *
95 *************************************************/
97 /* This value specifies the size of stack workspace that is used during the
98 first pre-compile phase that determines how much memory is required. The regex
99 is partly compiled into this space, but the compiled parts are discarded as
100 soon as they can be, so that hopefully there will never be an overrun. The code
101 does, however, check for an overrun. The largest amount I've seen used is 218,
102 so this number is very generous.
104 The same workspace is used during the second, actual compile phase for
105 remembering forward references to groups so that they can be filled in at the
106 end. Each entry in this list occupies LINK_SIZE bytes, so even when LINK_SIZE
107 is 4 there is plenty of room for most patterns. However, the memory can get
108 filled up by repetitions of forward references, for example patterns like
109 /(?1){0,1999}(b)/, and one user did hit the limit. The code has been changed so
110 that the workspace is expanded using malloc() in this situation. The value
111 below is therefore a minimum, and we put a maximum on it for safety. The
112 minimum is now also defined in terms of LINK_SIZE so that the use of malloc()
113 kicks in at the same number of forward references in all cases. */
115 #define COMPILE_WORK_SIZE (2048*LINK_SIZE)
116 #define COMPILE_WORK_SIZE_MAX (100*COMPILE_WORK_SIZE)
118 /* This value determines the size of the initial vector that is used for
119 remembering named groups during the pre-compile. It is allocated on the stack,
120 but if it is too small, it is expanded using malloc(), in a similar way to the
121 workspace. The value is the number of slots in the list. */
123 #define NAMED_GROUP_LIST_SIZE 20
125 /* The overrun tests check for a slightly smaller size so that they detect the
126 overrun before it actually does run off the end of the data block. */
128 #define WORK_SIZE_SAFETY_MARGIN (100)
130 /* Private flags added to firstchar and reqchar. */
132 #define REQ_CASELESS (1 << 0) /* Indicates caselessness */
133 #define REQ_VARY (1 << 1) /* Reqchar followed non-literal item */
134 /* Negative values for the firstchar and reqchar flags */
135 #define REQ_UNSET (-2)
136 #define REQ_NONE (-1)
138 /* Repeated character flags. */
140 #define UTF_LENGTH 0x10000000l /* The char contains its length. */
142 /* Table for handling escaped characters in the range '0'-'z'. Positive returns
143 are simple data values; negative values are for special things like \d and so
144 on. Zero means further processing is needed (for things like \x), or the escape
149 /* This is the "normal" table for ASCII systems or for EBCDIC systems running
152 static const short int escapes[] = {
158 CHAR_COLON, CHAR_SEMICOLON,
159 CHAR_LESS_THAN_SIGN, CHAR_EQUALS_SIGN,
160 CHAR_GREATER_THAN_SIGN, CHAR_QUESTION_MARK,
161 CHAR_COMMERCIAL_AT, -ESC_A,
174 -ESC_Z, CHAR_LEFT_SQUARE_BRACKET,
175 CHAR_BACKSLASH, CHAR_RIGHT_SQUARE_BRACKET,
176 CHAR_CIRCUMFLEX_ACCENT, CHAR_UNDERSCORE,
177 CHAR_GRAVE_ACCENT, 7,
195 /* This is the "abnormal" table for EBCDIC systems without UTF-8 support. */
197 static const short int escapes[] = {
198 /* 48 */ 0, 0, 0, '.', '<', '(', '+', '|',
199 /* 50 */ '&', 0, 0, 0, 0, 0, 0, 0,
200 /* 58 */ 0, 0, '!', '$', '*', ')', ';', '~',
201 /* 60 */ '-', '/', 0, 0, 0, 0, 0, 0,
202 /* 68 */ 0, 0, '|', ',', '%', '_', '>', '?',
203 /* 70 */ 0, 0, 0, 0, 0, 0, 0, 0,
204 /* 78 */ 0, '`', ':', '#', '@', '\'', '=', '"',
205 /* 80 */ 0, 7, -ESC_b, 0, -ESC_d, ESC_e, ESC_f, 0,
206 /* 88 */-ESC_h, 0, 0, '{', 0, 0, 0, 0,
207 /* 90 */ 0, 0, -ESC_k, 'l', 0, ESC_n, 0, -ESC_p,
208 /* 98 */ 0, ESC_r, 0, '}', 0, 0, 0, 0,
209 /* A0 */ 0, '~', -ESC_s, ESC_tee, 0,-ESC_v, -ESC_w, 0,
210 /* A8 */ 0,-ESC_z, 0, 0, 0, '[', 0, 0,
211 /* B0 */ 0, 0, 0, 0, 0, 0, 0, 0,
212 /* B8 */ 0, 0, 0, 0, 0, ']', '=', '-',
213 /* C0 */ '{',-ESC_A, -ESC_B, -ESC_C, -ESC_D,-ESC_E, 0, -ESC_G,
214 /* C8 */-ESC_H, 0, 0, 0, 0, 0, 0, 0,
215 /* D0 */ '}', 0, -ESC_K, 0, 0,-ESC_N, 0, -ESC_P,
216 /* D8 */-ESC_Q,-ESC_R, 0, 0, 0, 0, 0, 0,
217 /* E0 */ '\\', 0, -ESC_S, 0, 0,-ESC_V, -ESC_W, -ESC_X,
218 /* E8 */ 0,-ESC_Z, 0, 0, 0, 0, 0, 0,
219 /* F0 */ 0, 0, 0, 0, 0, 0, 0, 0,
220 /* F8 */ 0, 0, 0, 0, 0, 0, 0, 0
225 /* Table of special "verbs" like (*PRUNE). This is a short table, so it is
226 searched linearly. Put all the names into a single string, in order to reduce
227 the number of relocations when a shared library is dynamically linked. The
228 string is built from string macros so that it works in UTF-8 mode on EBCDIC
231 typedef struct verbitem {
232 int len; /* Length of verb name */
233 int op; /* Op when no arg, or -1 if arg mandatory */
234 int op_arg; /* Op when arg present, or -1 if not allowed */
237 static const char verbnames[] =
238 "\0" /* Empty name is a shorthand for MARK */
248 static const verbitem verbs[] = {
251 { 6, OP_ACCEPT, -1 },
252 { 6, OP_COMMIT, -1 },
255 { 5, OP_PRUNE, OP_PRUNE_ARG },
256 { 4, OP_SKIP, OP_SKIP_ARG },
257 { 4, OP_THEN, OP_THEN_ARG }
260 static const int verbcount = sizeof(verbs)/sizeof(verbitem);
263 /* Substitutes for [[:<:]] and [[:>:]], which mean start and end of word in
264 another regex library. */
266 static const pcre_uchar sub_start_of_word[] = {
267 CHAR_BACKSLASH, CHAR_b, CHAR_LEFT_PARENTHESIS, CHAR_QUESTION_MARK,
268 CHAR_EQUALS_SIGN, CHAR_BACKSLASH, CHAR_w, CHAR_RIGHT_PARENTHESIS, '\0' };
270 static const pcre_uchar sub_end_of_word[] = {
271 CHAR_BACKSLASH, CHAR_b, CHAR_LEFT_PARENTHESIS, CHAR_QUESTION_MARK,
272 CHAR_LESS_THAN_SIGN, CHAR_EQUALS_SIGN, CHAR_BACKSLASH, CHAR_w,
273 CHAR_RIGHT_PARENTHESIS, '\0' };
276 /* Tables of names of POSIX character classes and their lengths. The names are
277 now all in a single string, to reduce the number of relocations when a shared
278 library is dynamically loaded. The list of lengths is terminated by a zero
279 length entry. The first three must be alpha, lower, upper, as this is assumed
280 for handling case independence. The indices for graph, print, and punct are
281 needed, so identify them. */
283 static const char posix_names[] =
284 STRING_alpha0 STRING_lower0 STRING_upper0 STRING_alnum0
285 STRING_ascii0 STRING_blank0 STRING_cntrl0 STRING_digit0
286 STRING_graph0 STRING_print0 STRING_punct0 STRING_space0
287 STRING_word0 STRING_xdigit;
289 static const pcre_uint8 posix_name_lengths[] = {
290 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 4, 6, 0 };
297 /* Table of class bit maps for each POSIX class. Each class is formed from a
298 base map, with an optional addition or removal of another map. Then, for some
299 classes, there is some additional tweaking: for [:blank:] the vertical space
300 characters are removed, and for [:alpha:] and [:alnum:] the underscore
301 character is removed. The triples in the table consist of the base map offset,
302 second map offset or -1 if no second map, and a non-negative value for map
303 addition or a negative value for map subtraction (if there are two maps). The
304 absolute value of the third field has these meanings: 0 => no tweaking, 1 =>
305 remove vertical space characters, 2 => remove underscore. */
307 static const int posix_class_maps[] = {
308 cbit_word, cbit_digit, -2, /* alpha */
309 cbit_lower, -1, 0, /* lower */
310 cbit_upper, -1, 0, /* upper */
311 cbit_word, -1, 2, /* alnum - word without underscore */
312 cbit_print, cbit_cntrl, 0, /* ascii */
313 cbit_space, -1, 1, /* blank - a GNU extension */
314 cbit_cntrl, -1, 0, /* cntrl */
315 cbit_digit, -1, 0, /* digit */
316 cbit_graph, -1, 0, /* graph */
317 cbit_print, -1, 0, /* print */
318 cbit_punct, -1, 0, /* punct */
319 cbit_space, -1, 0, /* space */
320 cbit_word, -1, 0, /* word - a Perl extension */
321 cbit_xdigit,-1, 0 /* xdigit */
324 /* Table of substitutes for \d etc when PCRE_UCP is set. They are replaced by
325 Unicode property escapes. */
328 static const pcre_uchar string_PNd[] = {
329 CHAR_BACKSLASH, CHAR_P, CHAR_LEFT_CURLY_BRACKET,
330 CHAR_N, CHAR_d, CHAR_RIGHT_CURLY_BRACKET, '\0' };
331 static const pcre_uchar string_pNd[] = {
332 CHAR_BACKSLASH, CHAR_p, CHAR_LEFT_CURLY_BRACKET,
333 CHAR_N, CHAR_d, CHAR_RIGHT_CURLY_BRACKET, '\0' };
334 static const pcre_uchar string_PXsp[] = {
335 CHAR_BACKSLASH, CHAR_P, CHAR_LEFT_CURLY_BRACKET,
336 CHAR_X, CHAR_s, CHAR_p, CHAR_RIGHT_CURLY_BRACKET, '\0' };
337 static const pcre_uchar string_pXsp[] = {
338 CHAR_BACKSLASH, CHAR_p, CHAR_LEFT_CURLY_BRACKET,
339 CHAR_X, CHAR_s, CHAR_p, CHAR_RIGHT_CURLY_BRACKET, '\0' };
340 static const pcre_uchar string_PXwd[] = {
341 CHAR_BACKSLASH, CHAR_P, CHAR_LEFT_CURLY_BRACKET,
342 CHAR_X, CHAR_w, CHAR_d, CHAR_RIGHT_CURLY_BRACKET, '\0' };
343 static const pcre_uchar string_pXwd[] = {
344 CHAR_BACKSLASH, CHAR_p, CHAR_LEFT_CURLY_BRACKET,
345 CHAR_X, CHAR_w, CHAR_d, CHAR_RIGHT_CURLY_BRACKET, '\0' };
347 static const pcre_uchar *substitutes[] = {
350 string_PXsp, /* \S */ /* Xsp is Perl space, but from 8.34, Perl */
351 string_pXsp, /* \s */ /* space and POSIX space are the same. */
352 string_PXwd, /* \W */
356 /* The POSIX class substitutes must be in the order of the POSIX class names,
357 defined above, and there are both positive and negative cases. NULL means no
358 general substitute of a Unicode property escape (\p or \P). However, for some
359 POSIX classes (e.g. graph, print, punct) a special property code is compiled
362 static const pcre_uchar string_pL[] = {
363 CHAR_BACKSLASH, CHAR_p, CHAR_LEFT_CURLY_BRACKET,
364 CHAR_L, CHAR_RIGHT_CURLY_BRACKET, '\0' };
365 static const pcre_uchar string_pLl[] = {
366 CHAR_BACKSLASH, CHAR_p, CHAR_LEFT_CURLY_BRACKET,
367 CHAR_L, CHAR_l, CHAR_RIGHT_CURLY_BRACKET, '\0' };
368 static const pcre_uchar string_pLu[] = {
369 CHAR_BACKSLASH, CHAR_p, CHAR_LEFT_CURLY_BRACKET,
370 CHAR_L, CHAR_u, CHAR_RIGHT_CURLY_BRACKET, '\0' };
371 static const pcre_uchar string_pXan[] = {
372 CHAR_BACKSLASH, CHAR_p, CHAR_LEFT_CURLY_BRACKET,
373 CHAR_X, CHAR_a, CHAR_n, CHAR_RIGHT_CURLY_BRACKET, '\0' };
374 static const pcre_uchar string_h[] = {
375 CHAR_BACKSLASH, CHAR_h, '\0' };
376 static const pcre_uchar string_pXps[] = {
377 CHAR_BACKSLASH, CHAR_p, CHAR_LEFT_CURLY_BRACKET,
378 CHAR_X, CHAR_p, CHAR_s, CHAR_RIGHT_CURLY_BRACKET, '\0' };
379 static const pcre_uchar string_PL[] = {
380 CHAR_BACKSLASH, CHAR_P, CHAR_LEFT_CURLY_BRACKET,
381 CHAR_L, CHAR_RIGHT_CURLY_BRACKET, '\0' };
382 static const pcre_uchar string_PLl[] = {
383 CHAR_BACKSLASH, CHAR_P, CHAR_LEFT_CURLY_BRACKET,
384 CHAR_L, CHAR_l, CHAR_RIGHT_CURLY_BRACKET, '\0' };
385 static const pcre_uchar string_PLu[] = {
386 CHAR_BACKSLASH, CHAR_P, CHAR_LEFT_CURLY_BRACKET,
387 CHAR_L, CHAR_u, CHAR_RIGHT_CURLY_BRACKET, '\0' };
388 static const pcre_uchar string_PXan[] = {
389 CHAR_BACKSLASH, CHAR_P, CHAR_LEFT_CURLY_BRACKET,
390 CHAR_X, CHAR_a, CHAR_n, CHAR_RIGHT_CURLY_BRACKET, '\0' };
391 static const pcre_uchar string_H[] = {
392 CHAR_BACKSLASH, CHAR_H, '\0' };
393 static const pcre_uchar string_PXps[] = {
394 CHAR_BACKSLASH, CHAR_P, CHAR_LEFT_CURLY_BRACKET,
395 CHAR_X, CHAR_p, CHAR_s, CHAR_RIGHT_CURLY_BRACKET, '\0' };
397 static const pcre_uchar *posix_substitutes[] = {
398 string_pL, /* alpha */
399 string_pLl, /* lower */
400 string_pLu, /* upper */
401 string_pXan, /* alnum */
403 string_h, /* blank */
405 string_pNd, /* digit */
409 string_pXps, /* space */ /* Xps is POSIX space, but from 8.34 */
410 string_pXwd, /* word */ /* Perl and POSIX space are the same */
413 string_PL, /* ^alpha */
414 string_PLl, /* ^lower */
415 string_PLu, /* ^upper */
416 string_PXan, /* ^alnum */
418 string_H, /* ^blank */
420 string_PNd, /* ^digit */
424 string_PXps, /* ^space */ /* Xps is POSIX space, but from 8.34 */
425 string_PXwd, /* ^word */ /* Perl and POSIX space are the same */
428 #define POSIX_SUBSIZE (sizeof(posix_substitutes) / sizeof(pcre_uchar *))
431 #define STRING(a) # a
432 #define XSTRING(s) STRING(s)
434 /* The texts of compile-time error messages. These are "char *" because they
435 are passed to the outside world. Do not ever re-use any error number, because
436 they are documented. Always add a new error instead. Messages marked DEAD below
437 are no longer used. This used to be a table of strings, but in order to reduce
438 the number of relocations needed when a shared library is loaded dynamically,
439 it is now one long string. We cannot use a table of offsets, because the
440 lengths of inserts such as XSTRING(MAX_NAME_SIZE) are not known. Instead, we
441 simply count through to the one we want - this isn't a performance issue
442 because these strings are used only when there is a compilation error.
444 Each substring ends with \0 to insert a null character. This includes the final
445 substring, so that the whole string ends with \0\0, which can be detected when
448 static const char error_texts[] =
450 "\\ at end of pattern\0"
451 "\\c at end of pattern\0"
452 "unrecognized character follows \\\0"
453 "numbers out of order in {} quantifier\0"
455 "number too big in {} quantifier\0"
456 "missing terminating ] for character class\0"
457 "invalid escape sequence in character class\0"
458 "range out of order in character class\0"
459 "nothing to repeat\0"
461 "operand of unlimited repeat could match the empty string\0" /** DEAD **/
462 "internal error: unexpected repeat\0"
463 "unrecognized character after (? or (?-\0"
464 "POSIX named classes are supported only within a class\0"
467 "reference to non-existent subpattern\0"
468 "erroffset passed as NULL\0"
469 "unknown option bit(s) set\0"
470 "missing ) after comment\0"
471 "parentheses nested too deeply\0" /** DEAD **/
473 "regular expression is too large\0"
474 "failed to get memory\0"
475 "unmatched parentheses\0"
476 "internal error: code overflow\0"
477 "unrecognized character after (?<\0"
479 "lookbehind assertion is not fixed length\0"
480 "malformed number or name after (?(\0"
481 "conditional group contains more than two branches\0"
482 "assertion expected after (?(\0"
483 "(?R or (?[+-]digits must be followed by )\0"
485 "unknown POSIX class name\0"
486 "POSIX collating elements are not supported\0"
487 "this version of PCRE is compiled without UTF support\0"
488 "spare error\0" /** DEAD **/
489 "character value in \\x{} or \\o{} is too large\0"
491 "invalid condition (?(0)\0"
492 "\\C not allowed in lookbehind assertion\0"
493 "PCRE does not support \\L, \\l, \\N{name}, \\U, or \\u\0"
494 "number after (?C is > 255\0"
495 "closing ) for (?C expected\0"
497 "recursive call could loop indefinitely\0"
498 "unrecognized character after (?P\0"
499 "syntax error in subpattern name (missing terminator)\0"
500 "two named subpatterns have the same name\0"
501 "invalid UTF-8 string\0"
503 "support for \\P, \\p, and \\X has not been compiled\0"
504 "malformed \\P or \\p sequence\0"
505 "unknown property name after \\P or \\p\0"
506 "subpattern name is too long (maximum " XSTRING(MAX_NAME_SIZE) " characters)\0"
507 "too many named subpatterns (maximum " XSTRING(MAX_NAME_COUNT) ")\0"
509 "repeated subpattern is too long\0" /** DEAD **/
510 "octal value is greater than \\377 in 8-bit non-UTF-8 mode\0"
511 "internal error: overran compiling workspace\0"
512 "internal error: previously-checked referenced subpattern not found\0"
513 "DEFINE group contains more than one branch\0"
515 "repeating a DEFINE group is not allowed\0" /** DEAD **/
516 "inconsistent NEWLINE options\0"
517 "\\g is not followed by a braced, angle-bracketed, or quoted name/number or by a plain number\0"
518 "a numbered reference must not be zero\0"
519 "an argument is not allowed for (*ACCEPT), (*FAIL), or (*COMMIT)\0"
521 "(*VERB) not recognized or malformed\0"
522 "number is too big\0"
523 "subpattern name expected\0"
524 "digit expected after (?+\0"
525 "] is an invalid data character in JavaScript compatibility mode\0"
527 "different names for subpatterns of the same number are not allowed\0"
528 "(*MARK) must have an argument\0"
529 "this version of PCRE is not compiled with Unicode property support\0"
530 "\\c must be followed by an ASCII character\0"
531 "\\k is not followed by a braced, angle-bracketed, or quoted name\0"
533 "internal error: unknown opcode in find_fixedlength()\0"
534 "\\N is not supported in a class\0"
535 "too many forward references\0"
536 "disallowed Unicode code point (>= 0xd800 && <= 0xdfff)\0"
537 "invalid UTF-16 string\0"
539 "name is too long in (*MARK), (*PRUNE), (*SKIP), or (*THEN)\0"
540 "character value in \\u.... sequence is too large\0"
541 "invalid UTF-32 string\0"
542 "setting UTF is disabled by the application\0"
543 "non-hex character in \\x{} (closing brace missing?)\0"
545 "non-octal character in \\o{} (closing brace missing?)\0"
546 "missing opening brace after \\o\0"
547 "parentheses are too deeply nested\0"
548 "invalid range in character class\0"
549 "group name must start with a non-digit\0"
551 "parentheses are too deeply nested (stack check)\0"
554 /* Table to identify digits and hex digits. This is used when compiling
555 patterns. Note that the tables in chartables are dependent on the locale, and
556 may mark arbitrary characters as digits - but the PCRE compiling code expects
557 to handle only 0-9, a-z, and A-Z as digits when compiling. That is why we have
558 a private table here. It costs 256 bytes, but it is a lot faster than doing
559 character value tests (at least in some simple cases I timed), and in some
560 applications one wants PCRE to compile efficiently as well as match
563 For convenience, we use the same bit definitions as in chartables:
566 0x08 hexadecimal digit
568 Then we can use ctype_digit and ctype_xdigit in the code. */
570 /* Using a simple comparison for decimal numbers rather than a memory read
571 is much faster, and the resulting code is simpler (the compiler turns it
572 into a subtraction and unsigned comparison). */
574 #define IS_DIGIT(x) ((x) >= CHAR_0 && (x) <= CHAR_9)
578 /* This is the "normal" case, for ASCII systems, and EBCDIC systems running in
581 static const pcre_uint8 digitab[] =
583 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 0- 7 */
584 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 8- 15 */
585 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 16- 23 */
586 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 24- 31 */
587 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - ' */
588 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* ( - / */
589 0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c, /* 0 - 7 */
590 0x0c,0x0c,0x00,0x00,0x00,0x00,0x00,0x00, /* 8 - ? */
591 0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* @ - G */
592 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* H - O */
593 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* P - W */
594 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* X - _ */
595 0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* ` - g */
596 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* h - o */
597 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* p - w */
598 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* x -127 */
599 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 128-135 */
600 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 136-143 */
601 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 144-151 */
602 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 152-159 */
603 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 160-167 */
604 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 168-175 */
605 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 176-183 */
606 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 184-191 */
607 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 192-199 */
608 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 200-207 */
609 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 208-215 */
610 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 216-223 */
611 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 224-231 */
612 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 232-239 */
613 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 240-247 */
614 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00};/* 248-255 */
618 /* This is the "abnormal" case, for EBCDIC systems not running in UTF-8 mode. */
620 static const pcre_uint8 digitab[] =
622 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 0- 7 0 */
623 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 8- 15 */
624 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 16- 23 10 */
625 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 24- 31 */
626 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 32- 39 20 */
627 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 40- 47 */
628 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 48- 55 30 */
629 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 56- 63 */
630 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - 71 40 */
631 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 72- | */
632 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* & - 87 50 */
633 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 88- 95 */
634 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - -103 60 */
635 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 104- ? */
636 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 112-119 70 */
637 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 120- " */
638 0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* 128- g 80 */
639 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* h -143 */
640 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 144- p 90 */
641 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* q -159 */
642 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 160- x A0 */
643 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* y -175 */
644 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* ^ -183 B0 */
645 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 184-191 */
646 0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* { - G C0 */
647 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* H -207 */
648 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* } - P D0 */
649 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* Q -223 */
650 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* \ - X E0 */
651 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* Y -239 */
652 0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c, /* 0 - 7 F0 */
653 0x0c,0x0c,0x00,0x00,0x00,0x00,0x00,0x00};/* 8 -255 */
655 static const pcre_uint8 ebcdic_chartab[] = { /* chartable partial dup */
656 0x80,0x00,0x00,0x00,0x00,0x01,0x00,0x00, /* 0- 7 */
657 0x00,0x00,0x00,0x00,0x01,0x01,0x00,0x00, /* 8- 15 */
658 0x00,0x00,0x00,0x00,0x00,0x01,0x00,0x00, /* 16- 23 */
659 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 24- 31 */
660 0x00,0x00,0x00,0x00,0x00,0x01,0x00,0x00, /* 32- 39 */
661 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 40- 47 */
662 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 48- 55 */
663 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 56- 63 */
664 0x01,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - 71 */
665 0x00,0x00,0x00,0x80,0x00,0x80,0x80,0x80, /* 72- | */
666 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* & - 87 */
667 0x00,0x00,0x00,0x80,0x80,0x80,0x00,0x00, /* 88- 95 */
668 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - -103 */
669 0x00,0x00,0x00,0x00,0x00,0x10,0x00,0x80, /* 104- ? */
670 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 112-119 */
671 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 120- " */
672 0x00,0x1a,0x1a,0x1a,0x1a,0x1a,0x1a,0x12, /* 128- g */
673 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* h -143 */
674 0x00,0x12,0x12,0x12,0x12,0x12,0x12,0x12, /* 144- p */
675 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* q -159 */
676 0x00,0x00,0x12,0x12,0x12,0x12,0x12,0x12, /* 160- x */
677 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* y -175 */
678 0x80,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* ^ -183 */
679 0x00,0x00,0x80,0x00,0x00,0x00,0x00,0x00, /* 184-191 */
680 0x80,0x1a,0x1a,0x1a,0x1a,0x1a,0x1a,0x12, /* { - G */
681 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* H -207 */
682 0x00,0x12,0x12,0x12,0x12,0x12,0x12,0x12, /* } - P */
683 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* Q -223 */
684 0x00,0x00,0x12,0x12,0x12,0x12,0x12,0x12, /* \ - X */
685 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* Y -239 */
686 0x1c,0x1c,0x1c,0x1c,0x1c,0x1c,0x1c,0x1c, /* 0 - 7 */
687 0x1c,0x1c,0x00,0x00,0x00,0x00,0x00,0x00};/* 8 -255 */
691 /* This table is used to check whether auto-possessification is possible
692 between adjacent character-type opcodes. The left-hand (repeated) opcode is
693 used to select the row, and the right-hand opcode is use to select the column.
694 A value of 1 means that auto-possessification is OK. For example, the second
695 value in the first row means that \D+\d can be turned into \D++\d.
697 The Unicode property types (\P and \p) have to be present to fill out the table
698 because of what their opcode values are, but the table values should always be
699 zero because property types are handled separately in the code. The last four
700 columns apply to items that cannot be repeated, so there is no need to have
701 rows for them. Note that OP_DIGIT etc. are generated only when PCRE_UCP is
702 *not* set. When it is set, \d etc. are converted into OP_(NOT_)PROP codes. */
704 #define APTROWS (LAST_AUTOTAB_LEFT_OP - FIRST_AUTOTAB_OP + 1)
705 #define APTCOLS (LAST_AUTOTAB_RIGHT_OP - FIRST_AUTOTAB_OP + 1)
707 static const pcre_uint8 autoposstab[APTROWS][APTCOLS] = {
708 /* \D \d \S \s \W \w . .+ \C \P \p \R \H \h \V \v \X \Z \z $ $M */
709 { 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0 }, /* \D */
710 { 1, 0, 0, 1, 1, 0, 0, 0, 0, 0, 0, 1, 0, 1, 0, 1, 0, 1, 1, 1, 1 }, /* \d */
711 { 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 1, 0, 1, 0, 1, 0, 1, 1, 1, 1 }, /* \S */
712 { 0, 1, 1, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0 }, /* \s */
713 { 0, 1, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0 }, /* \W */
714 { 0, 0, 0, 1, 1, 0, 0, 0, 0, 0, 0, 1, 0, 1, 0, 1, 0, 1, 1, 1, 1 }, /* \w */
715 { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 1, 0, 0 }, /* . */
716 { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0 }, /* .+ */
717 { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0 }, /* \C */
718 { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 }, /* \P */
719 { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 }, /* \p */
720 { 0, 1, 0, 1, 0, 1, 1, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 1, 0, 0 }, /* \R */
721 { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 1, 0, 0 }, /* \H */
722 { 0, 1, 1, 0, 0, 1, 0, 0, 0, 0, 0, 1, 1, 0, 0, 1, 0, 0, 1, 0, 0 }, /* \h */
723 { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 1, 0, 0, 1, 0, 0 }, /* \V */
724 { 0, 1, 1, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 1, 1, 0, 0, 0, 1, 0, 0 }, /* \v */
725 { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0 } /* \X */
729 /* This table is used to check whether auto-possessification is possible
730 between adjacent Unicode property opcodes (OP_PROP and OP_NOTPROP). The
731 left-hand (repeated) opcode is used to select the row, and the right-hand
732 opcode is used to select the column. The values are as follows:
734 0 Always return FALSE (never auto-possessify)
735 1 Character groups are distinct (possessify if both are OP_PROP)
736 2 Check character categories in the same group (general or particular)
737 3 TRUE if the two opcodes are not the same (PROP vs NOTPROP)
739 4 Check left general category vs right particular category
740 5 Check right general category vs left particular category
742 6 Left alphanum vs right general category
743 7 Left space vs right general category
744 8 Left word vs right general category
746 9 Right alphanum vs left general category
747 10 Right space vs left general category
748 11 Right word vs left general category
750 12 Left alphanum vs right particular category
751 13 Left space vs right particular category
752 14 Left word vs right particular category
754 15 Right alphanum vs left particular category
755 16 Right space vs left particular category
756 17 Right word vs left particular category
759 static const pcre_uint8 propposstab[PT_TABSIZE][PT_TABSIZE] = {
760 /* ANY LAMP GC PC SC ALNUM SPACE PXSPACE WORD CLIST UCNC */
761 { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 }, /* PT_ANY */
762 { 0, 3, 0, 0, 0, 3, 1, 1, 0, 0, 0 }, /* PT_LAMP */
763 { 0, 0, 2, 4, 0, 9, 10, 10, 11, 0, 0 }, /* PT_GC */
764 { 0, 0, 5, 2, 0, 15, 16, 16, 17, 0, 0 }, /* PT_PC */
765 { 0, 0, 0, 0, 2, 0, 0, 0, 0, 0, 0 }, /* PT_SC */
766 { 0, 3, 6, 12, 0, 3, 1, 1, 0, 0, 0 }, /* PT_ALNUM */
767 { 0, 1, 7, 13, 0, 1, 3, 3, 1, 0, 0 }, /* PT_SPACE */
768 { 0, 1, 7, 13, 0, 1, 3, 3, 1, 0, 0 }, /* PT_PXSPACE */
769 { 0, 0, 8, 14, 0, 0, 1, 1, 3, 0, 0 }, /* PT_WORD */
770 { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 }, /* PT_CLIST */
771 { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 3 } /* PT_UCNC */
774 /* This table is used to check whether auto-possessification is possible
775 between adjacent Unicode property opcodes (OP_PROP and OP_NOTPROP) when one
776 specifies a general category and the other specifies a particular category. The
777 row is selected by the general category and the column by the particular
778 category. The value is 1 if the particular category is not part of the general
781 static const pcre_uint8 catposstab[7][30] = {
782 /* Cc Cf Cn Co Cs Ll Lm Lo Lt Lu Mc Me Mn Nd Nl No Pc Pd Pe Pf Pi Po Ps Sc Sk Sm So Zl Zp Zs */
783 { 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1 }, /* C */
784 { 1, 1, 1, 1, 1, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1 }, /* L */
785 { 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1 }, /* M */
786 { 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1 }, /* N */
787 { 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 1 }, /* P */
788 { 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 1, 1, 1 }, /* S */
789 { 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0 } /* Z */
792 /* This table is used when checking ALNUM, (PX)SPACE, SPACE, and WORD against
793 a general or particular category. The properties in each row are those
794 that apply to the character set in question. Duplication means that a little
795 unnecessary work is done when checking, but this keeps things much simpler
796 because they can all use the same code. For more details see the comment where
799 Note: SPACE and PXSPACE used to be different because Perl excluded VT from
800 "space", but from Perl 5.18 it's included, so both categories are treated the
803 static const pcre_uint8 posspropstab[3][4] = {
804 { ucp_L, ucp_N, ucp_N, ucp_Nl }, /* ALNUM, 3rd and 4th values redundant */
805 { ucp_Z, ucp_Z, ucp_C, ucp_Cc }, /* SPACE and PXSPACE, 2nd value redundant */
806 { ucp_L, ucp_N, ucp_P, ucp_Po } /* WORD */
809 /* This table is used when converting repeating opcodes into possessified
810 versions as a result of an explicit possessive quantifier such as ++. A zero
811 value means there is no possessified version - in those cases the item in
812 question must be wrapped in ONCE brackets. The table is truncated at OP_CALLOUT
813 because all relevant opcodes are less than that. */
815 static const pcre_uint8 opcode_possessify[] = {
816 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, /* 0 - 15 */
817 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, /* 16 - 31 */
820 OP_POSSTAR, 0, /* STAR, MINSTAR */
821 OP_POSPLUS, 0, /* PLUS, MINPLUS */
822 OP_POSQUERY, 0, /* QUERY, MINQUERY */
823 OP_POSUPTO, 0, /* UPTO, MINUPTO */
825 0, 0, 0, 0, /* POS{STAR,PLUS,QUERY,UPTO} */
827 OP_POSSTARI, 0, /* STARI, MINSTARI */
828 OP_POSPLUSI, 0, /* PLUSI, MINPLUSI */
829 OP_POSQUERYI, 0, /* QUERYI, MINQUERYI */
830 OP_POSUPTOI, 0, /* UPTOI, MINUPTOI */
832 0, 0, 0, 0, /* POS{STARI,PLUSI,QUERYI,UPTOI} */
834 OP_NOTPOSSTAR, 0, /* NOTSTAR, NOTMINSTAR */
835 OP_NOTPOSPLUS, 0, /* NOTPLUS, NOTMINPLUS */
836 OP_NOTPOSQUERY, 0, /* NOTQUERY, NOTMINQUERY */
837 OP_NOTPOSUPTO, 0, /* NOTUPTO, NOTMINUPTO */
839 0, 0, 0, 0, /* NOTPOS{STAR,PLUS,QUERY,UPTO} */
841 OP_NOTPOSSTARI, 0, /* NOTSTARI, NOTMINSTARI */
842 OP_NOTPOSPLUSI, 0, /* NOTPLUSI, NOTMINPLUSI */
843 OP_NOTPOSQUERYI, 0, /* NOTQUERYI, NOTMINQUERYI */
844 OP_NOTPOSUPTOI, 0, /* NOTUPTOI, NOTMINUPTOI */
846 0, 0, 0, 0, /* NOTPOS{STARI,PLUSI,QUERYI,UPTOI} */
848 OP_TYPEPOSSTAR, 0, /* TYPESTAR, TYPEMINSTAR */
849 OP_TYPEPOSPLUS, 0, /* TYPEPLUS, TYPEMINPLUS */
850 OP_TYPEPOSQUERY, 0, /* TYPEQUERY, TYPEMINQUERY */
851 OP_TYPEPOSUPTO, 0, /* TYPEUPTO, TYPEMINUPTO */
853 0, 0, 0, 0, /* TYPEPOS{STAR,PLUS,QUERY,UPTO} */
855 OP_CRPOSSTAR, 0, /* CRSTAR, CRMINSTAR */
856 OP_CRPOSPLUS, 0, /* CRPLUS, CRMINPLUS */
857 OP_CRPOSQUERY, 0, /* CRQUERY, CRMINQUERY */
858 OP_CRPOSRANGE, 0, /* CRRANGE, CRMINRANGE */
859 0, 0, 0, 0, /* CRPOS{STAR,PLUS,QUERY,RANGE} */
861 0, 0, 0, /* CLASS, NCLASS, XCLASS */
862 0, 0, /* REF, REFI */
863 0, 0, /* DNREF, DNREFI */
864 0, 0 /* RECURSE, CALLOUT */
869 /*************************************************
870 * Find an error text *
871 *************************************************/
873 /* The error texts are now all in one long string, to save on relocations. As
874 some of the text is of unknown length, we can't use a table of offsets.
875 Instead, just count through the strings. This is not a performance issue
876 because it happens only when there has been a compilation error.
878 Argument: the error number
879 Returns: pointer to the error string
883 find_error_text(int n)
885 const char *s = error_texts;
888 while (*s++ != CHAR_NULL) {};
889 if (*s == CHAR_NULL) return "Error text not found (please report)";
896 /*************************************************
897 * Expand the workspace *
898 *************************************************/
900 /* This function is called during the second compiling phase, if the number of
901 forward references fills the existing workspace, which is originally a block on
902 the stack. A larger block is obtained from malloc() unless the ultimate limit
903 has been reached or the increase will be rather small.
905 Argument: pointer to the compile data block
906 Returns: 0 if all went well, else an error number
910 expand_workspace(compile_data *cd)
912 pcre_uchar *newspace;
913 int newsize = cd->workspace_size * 2;
915 if (newsize > COMPILE_WORK_SIZE_MAX) newsize = COMPILE_WORK_SIZE_MAX;
916 if (cd->workspace_size >= COMPILE_WORK_SIZE_MAX ||
917 newsize - cd->workspace_size < WORK_SIZE_SAFETY_MARGIN)
920 newspace = (PUBL(malloc))(IN_UCHARS(newsize));
921 if (newspace == NULL) return ERR21;
922 memcpy(newspace, cd->start_workspace, cd->workspace_size * sizeof(pcre_uchar));
923 cd->hwm = (pcre_uchar *)newspace + (cd->hwm - cd->start_workspace);
924 if (cd->workspace_size > COMPILE_WORK_SIZE)
925 (PUBL(free))((void *)cd->start_workspace);
926 cd->start_workspace = newspace;
927 cd->workspace_size = newsize;
933 /*************************************************
934 * Check for counted repeat *
935 *************************************************/
937 /* This function is called when a '{' is encountered in a place where it might
938 start a quantifier. It looks ahead to see if it really is a quantifier or not.
939 It is only a quantifier if it is one of the forms {ddd} {ddd,} or {ddd,ddd}
940 where the ddds are digits.
943 p pointer to the first char after '{'
945 Returns: TRUE or FALSE
949 is_counted_repeat(const pcre_uchar *p)
951 if (!IS_DIGIT(*p)) return FALSE;
953 while (IS_DIGIT(*p)) p++;
954 if (*p == CHAR_RIGHT_CURLY_BRACKET) return TRUE;
956 if (*p++ != CHAR_COMMA) return FALSE;
957 if (*p == CHAR_RIGHT_CURLY_BRACKET) return TRUE;
959 if (!IS_DIGIT(*p)) return FALSE;
961 while (IS_DIGIT(*p)) p++;
963 return (*p == CHAR_RIGHT_CURLY_BRACKET);
968 /*************************************************
970 *************************************************/
972 /* This function is called when a \ has been encountered. It either returns a
973 positive value for a simple escape such as \n, or 0 for a data character which
974 will be placed in chptr. A backreference to group n is returned as negative n.
975 When UTF-8 is enabled, a positive value greater than 255 may be returned in
976 chptr. On entry, ptr is pointing at the \. On exit, it is on the final
977 character of the escape sequence.
980 ptrptr points to the pattern position pointer
981 chptr points to a returned data character
982 errorcodeptr points to the errorcode variable
983 bracount number of previous extracting brackets
984 options the options bits
985 isclass TRUE if inside a character class
987 Returns: zero => a data character
988 positive => a special escape sequence
989 negative => a back reference
990 on error, errorcodeptr is set
994 check_escape(const pcre_uchar **ptrptr, pcre_uint32 *chptr, int *errorcodeptr,
995 int bracount, int options, BOOL isclass)
997 /* PCRE_UTF16 has the same value as PCRE_UTF8. */
998 BOOL utf = (options & PCRE_UTF8) != 0;
999 const pcre_uchar *ptr = *ptrptr + 1;
1004 GETCHARINCTEST(c, ptr); /* Get character value, increment pointer */
1005 ptr--; /* Set pointer back to the last byte */
1007 /* If backslash is at the end of the pattern, it's an error. */
1009 if (c == CHAR_NULL) *errorcodeptr = ERR1;
1011 /* Non-alphanumerics are literals. For digits or letters, do an initial lookup
1012 in a table. A non-zero result is something that can be returned immediately.
1013 Otherwise further processing may be required. */
1015 #ifndef EBCDIC /* ASCII/UTF-8 coding */
1016 /* Not alphanumeric */
1017 else if (c < CHAR_0 || c > CHAR_z) {}
1018 else if ((i = escapes[c - CHAR_0]) != 0)
1019 { if (i > 0) c = (pcre_uint32)i; else escape = -i; }
1021 #else /* EBCDIC coding */
1022 /* Not alphanumeric */
1023 else if (c < CHAR_a || (!MAX_255(c) || (ebcdic_chartab[c] & 0x0E) == 0)) {}
1024 else if ((i = escapes[c - 0x48]) != 0) { if (i > 0) c = (pcre_uint32)i; else escape = -i; }
1027 /* Escapes that need further processing, or are illegal. */
1031 const pcre_uchar *oldptr;
1032 BOOL braced, negated, overflow;
1037 /* A number of Perl escapes are not handled by PCRE. We give an explicit
1042 *errorcodeptr = ERR37;
1046 if ((options & PCRE_JAVASCRIPT_COMPAT) != 0)
1048 /* In JavaScript, \u must be followed by four hexadecimal numbers.
1049 Otherwise it is a lowercase u letter. */
1050 if (MAX_255(ptr[1]) && (digitab[ptr[1]] & ctype_xdigit) != 0
1051 && MAX_255(ptr[2]) && (digitab[ptr[2]] & ctype_xdigit) != 0
1052 && MAX_255(ptr[3]) && (digitab[ptr[3]] & ctype_xdigit) != 0
1053 && MAX_255(ptr[4]) && (digitab[ptr[4]] & ctype_xdigit) != 0)
1056 for (i = 0; i < 4; ++i)
1058 register pcre_uint32 cc = *(++ptr);
1059 #ifndef EBCDIC /* ASCII/UTF-8 coding */
1060 if (cc >= CHAR_a) cc -= 32; /* Convert to upper case */
1061 c = (c << 4) + cc - ((cc < CHAR_A)? CHAR_0 : (CHAR_A - 10));
1062 #else /* EBCDIC coding */
1063 if (cc >= CHAR_a && cc <= CHAR_z) cc += 64; /* Convert to upper case */
1064 c = (c << 4) + cc - ((cc >= CHAR_0)? CHAR_0 : (CHAR_A - 10));
1068 #if defined COMPILE_PCRE8
1069 if (c > (utf ? 0x10ffffU : 0xffU))
1070 #elif defined COMPILE_PCRE16
1071 if (c > (utf ? 0x10ffffU : 0xffffU))
1072 #elif defined COMPILE_PCRE32
1073 if (utf && c > 0x10ffffU)
1076 *errorcodeptr = ERR76;
1078 else if (utf && c >= 0xd800 && c <= 0xdfff) *errorcodeptr = ERR73;
1082 *errorcodeptr = ERR37;
1086 /* In JavaScript, \U is an uppercase U letter. */
1087 if ((options & PCRE_JAVASCRIPT_COMPAT) == 0) *errorcodeptr = ERR37;
1090 /* In a character class, \g is just a literal "g". Outside a character
1091 class, \g must be followed by one of a number of specific things:
1093 (1) A number, either plain or braced. If positive, it is an absolute
1094 backreference. If negative, it is a relative backreference. This is a Perl
1097 (2) Perl 5.10 also supports \g{name} as a reference to a named group. This
1098 is part of Perl's movement towards a unified syntax for back references. As
1099 this is synonymous with \k{name}, we fudge it up by pretending it really
1102 (3) For Oniguruma compatibility we also support \g followed by a name or a
1103 number either in angle brackets or in single quotes. However, these are
1104 (possibly recursive) subroutine calls, _not_ backreferences. Just return
1105 the ESC_g code (cf \k). */
1109 if (ptr[1] == CHAR_LESS_THAN_SIGN || ptr[1] == CHAR_APOSTROPHE)
1115 /* Handle the Perl-compatible cases */
1117 if (ptr[1] == CHAR_LEFT_CURLY_BRACKET)
1119 const pcre_uchar *p;
1120 for (p = ptr+2; *p != CHAR_NULL && *p != CHAR_RIGHT_CURLY_BRACKET; p++)
1121 if (*p != CHAR_MINUS && !IS_DIGIT(*p)) break;
1122 if (*p != CHAR_NULL && *p != CHAR_RIGHT_CURLY_BRACKET)
1130 else braced = FALSE;
1132 if (ptr[1] == CHAR_MINUS)
1137 else negated = FALSE;
1139 /* The integer range is limited by the machine's int representation. */
1142 while (IS_DIGIT(ptr[1]))
1144 if (s > INT_MAX / 10 - 1) /* Integer overflow */
1149 s = s * 10 + (int)(*(++ptr) - CHAR_0);
1151 if (overflow) /* Integer overflow */
1153 while (IS_DIGIT(ptr[1]))
1155 *errorcodeptr = ERR61;
1159 if (braced && *(++ptr) != CHAR_RIGHT_CURLY_BRACKET)
1161 *errorcodeptr = ERR57;
1167 *errorcodeptr = ERR58;
1175 *errorcodeptr = ERR15;
1178 s = bracount - (s - 1);
1184 /* The handling of escape sequences consisting of a string of digits
1185 starting with one that is not zero is not straightforward. Perl has changed
1186 over the years. Nowadays \g{} for backreferences and \o{} for octal are
1187 recommended to avoid the ambiguities in the old syntax.
1189 Outside a character class, the digits are read as a decimal number. If the
1190 number is less than 8 (used to be 10), or if there are that many previous
1191 extracting left brackets, then it is a back reference. Otherwise, up to
1192 three octal digits are read to form an escaped byte. Thus \123 is likely to
1193 be octal 123 (cf \0123, which is octal 012 followed by the literal 3). If
1194 the octal value is greater than 377, the least significant 8 bits are
1195 taken. \8 and \9 are treated as the literal characters 8 and 9.
1197 Inside a character class, \ followed by a digit is always either a literal
1198 8 or 9 or an octal number. */
1200 case CHAR_1: case CHAR_2: case CHAR_3: case CHAR_4: case CHAR_5:
1201 case CHAR_6: case CHAR_7: case CHAR_8: case CHAR_9:
1206 /* The integer range is limited by the machine's int representation. */
1207 s = (int)(c -CHAR_0);
1209 while (IS_DIGIT(ptr[1]))
1211 if (s > INT_MAX / 10 - 1) /* Integer overflow */
1216 s = s * 10 + (int)(*(++ptr) - CHAR_0);
1218 if (overflow) /* Integer overflow */
1220 while (IS_DIGIT(ptr[1]))
1222 *errorcodeptr = ERR61;
1225 if (s < 8 || s <= bracount) /* Check for back reference */
1230 ptr = oldptr; /* Put the pointer back and fall through */
1233 /* Handle a digit following \ when the number is not a back reference. If
1234 the first digit is 8 or 9, Perl used to generate a binary zero byte and
1235 then treat the digit as a following literal. At least by Perl 5.18 this
1236 changed so as not to insert the binary zero. */
1238 if ((c = *ptr) >= CHAR_8) break;
1240 /* Fall through with a digit less than 8 */
1242 /* \0 always starts an octal number, but we may drop through to here with a
1243 larger first octal digit. The original code used just to take the least
1244 significant 8 bits of octal numbers (I think this is what early Perls used
1245 to do). Nowadays we allow for larger numbers in UTF-8 mode and 16-bit mode,
1246 but no more than 3 octal digits. */
1250 while(i++ < 2 && ptr[1] >= CHAR_0 && ptr[1] <= CHAR_7)
1251 c = c * 8 + *(++ptr) - CHAR_0;
1252 #ifdef COMPILE_PCRE8
1253 if (!utf && c > 0xff) *errorcodeptr = ERR51;
1257 /* \o is a relatively new Perl feature, supporting a more general way of
1258 specifying character codes in octal. The only supported form is \o{ddd}. */
1261 if (ptr[1] != CHAR_LEFT_CURLY_BRACKET) *errorcodeptr = ERR81; else
1266 while (*ptr >= CHAR_0 && *ptr <= CHAR_7)
1268 register pcre_uint32 cc = *ptr++;
1269 if (c == 0 && cc == CHAR_0) continue; /* Leading zeroes */
1270 #ifdef COMPILE_PCRE32
1271 if (c >= 0x20000000l) { overflow = TRUE; break; }
1273 c = (c << 3) + cc - CHAR_0 ;
1274 #if defined COMPILE_PCRE8
1275 if (c > (utf ? 0x10ffffU : 0xffU)) { overflow = TRUE; break; }
1276 #elif defined COMPILE_PCRE16
1277 if (c > (utf ? 0x10ffffU : 0xffffU)) { overflow = TRUE; break; }
1278 #elif defined COMPILE_PCRE32
1279 if (utf && c > 0x10ffffU) { overflow = TRUE; break; }
1284 while (*ptr >= CHAR_0 && *ptr <= CHAR_7) ptr++;
1285 *errorcodeptr = ERR34;
1287 else if (*ptr == CHAR_RIGHT_CURLY_BRACKET)
1289 if (utf && c >= 0xd800 && c <= 0xdfff) *errorcodeptr = ERR73;
1291 else *errorcodeptr = ERR80;
1295 /* \x is complicated. In JavaScript, \x must be followed by two hexadecimal
1296 numbers. Otherwise it is a lowercase x letter. */
1299 if ((options & PCRE_JAVASCRIPT_COMPAT) != 0)
1301 if (MAX_255(ptr[1]) && (digitab[ptr[1]] & ctype_xdigit) != 0
1302 && MAX_255(ptr[2]) && (digitab[ptr[2]] & ctype_xdigit) != 0)
1305 for (i = 0; i < 2; ++i)
1307 register pcre_uint32 cc = *(++ptr);
1308 #ifndef EBCDIC /* ASCII/UTF-8 coding */
1309 if (cc >= CHAR_a) cc -= 32; /* Convert to upper case */
1310 c = (c << 4) + cc - ((cc < CHAR_A)? CHAR_0 : (CHAR_A - 10));
1311 #else /* EBCDIC coding */
1312 if (cc >= CHAR_a && cc <= CHAR_z) cc += 64; /* Convert to upper case */
1313 c = (c << 4) + cc - ((cc >= CHAR_0)? CHAR_0 : (CHAR_A - 10));
1317 } /* End JavaScript handling */
1319 /* Handle \x in Perl's style. \x{ddd} is a character number which can be
1320 greater than 0xff in utf or non-8bit mode, but only if the ddd are hex
1321 digits. If not, { used to be treated as a data character. However, Perl
1322 seems to read hex digits up to the first non-such, and ignore the rest, so
1323 that, for example \x{zz} matches a binary zero. This seems crazy, so PCRE
1324 now gives an error. */
1328 if (ptr[1] == CHAR_LEFT_CURLY_BRACKET)
1333 while (MAX_255(*ptr) && (digitab[*ptr] & ctype_xdigit) != 0)
1335 register pcre_uint32 cc = *ptr++;
1336 if (c == 0 && cc == CHAR_0) continue; /* Leading zeroes */
1338 #ifdef COMPILE_PCRE32
1339 if (c >= 0x10000000l) { overflow = TRUE; break; }
1342 #ifndef EBCDIC /* ASCII/UTF-8 coding */
1343 if (cc >= CHAR_a) cc -= 32; /* Convert to upper case */
1344 c = (c << 4) + cc - ((cc < CHAR_A)? CHAR_0 : (CHAR_A - 10));
1345 #else /* EBCDIC coding */
1346 if (cc >= CHAR_a && cc <= CHAR_z) cc += 64; /* Convert to upper case */
1347 c = (c << 4) + cc - ((cc >= CHAR_0)? CHAR_0 : (CHAR_A - 10));
1350 #if defined COMPILE_PCRE8
1351 if (c > (utf ? 0x10ffffU : 0xffU)) { overflow = TRUE; break; }
1352 #elif defined COMPILE_PCRE16
1353 if (c > (utf ? 0x10ffffU : 0xffffU)) { overflow = TRUE; break; }
1354 #elif defined COMPILE_PCRE32
1355 if (utf && c > 0x10ffffU) { overflow = TRUE; break; }
1361 while (MAX_255(*ptr) && (digitab[*ptr] & ctype_xdigit) != 0) ptr++;
1362 *errorcodeptr = ERR34;
1365 else if (*ptr == CHAR_RIGHT_CURLY_BRACKET)
1367 if (utf && c >= 0xd800 && c <= 0xdfff) *errorcodeptr = ERR73;
1370 /* If the sequence of hex digits does not end with '}', give an error.
1371 We used just to recognize this construct and fall through to the normal
1372 \x handling, but nowadays Perl gives an error, which seems much more
1373 sensible, so we do too. */
1375 else *errorcodeptr = ERR79;
1376 } /* End of \x{} processing */
1378 /* Read a single-byte hex-defined char (up to two hex digits after \x) */
1383 while (i++ < 2 && MAX_255(ptr[1]) && (digitab[ptr[1]] & ctype_xdigit) != 0)
1385 pcre_uint32 cc; /* Some compilers don't like */
1386 cc = *(++ptr); /* ++ in initializers */
1387 #ifndef EBCDIC /* ASCII/UTF-8 coding */
1388 if (cc >= CHAR_a) cc -= 32; /* Convert to upper case */
1389 c = c * 16 + cc - ((cc < CHAR_A)? CHAR_0 : (CHAR_A - 10));
1390 #else /* EBCDIC coding */
1391 if (cc <= CHAR_z) cc += 64; /* Convert to upper case */
1392 c = c * 16 + cc - ((cc >= CHAR_0)? CHAR_0 : (CHAR_A - 10));
1395 } /* End of \xdd handling */
1396 } /* End of Perl-style \x handling */
1399 /* For \c, a following letter is upper-cased; then the 0x40 bit is flipped.
1400 An error is given if the byte following \c is not an ASCII character. This
1401 coding is ASCII-specific, but then the whole concept of \cx is
1402 ASCII-specific. (However, an EBCDIC equivalent has now been added.) */
1408 *errorcodeptr = ERR2;
1411 #ifndef EBCDIC /* ASCII/UTF-8 coding */
1412 if (c > 127) /* Excludes all non-ASCII in either mode */
1414 *errorcodeptr = ERR68;
1417 if (c >= CHAR_a && c <= CHAR_z) c -= 32;
1419 #else /* EBCDIC coding */
1420 if (c >= CHAR_a && c <= CHAR_z) c += 64;
1425 /* PCRE_EXTRA enables extensions to Perl in the matter of escapes. Any
1426 other alphanumeric following \ is an error if PCRE_EXTRA was set;
1427 otherwise, for Perl compatibility, it is a literal. This code looks a bit
1428 odd, but there used to be some cases other than the default, and there may
1429 be again in future, so I haven't "optimized" it. */
1432 if ((options & PCRE_EXTRA) != 0) switch(c)
1435 *errorcodeptr = ERR3;
1442 /* Perl supports \N{name} for character names, as well as plain \N for "not
1443 newline". PCRE does not support \N{name}. However, it does support
1444 quantification such as \N{2,3}. */
1446 if (escape == ESC_N && ptr[1] == CHAR_LEFT_CURLY_BRACKET &&
1447 !is_counted_repeat(ptr+2))
1448 *errorcodeptr = ERR37;
1450 /* If PCRE_UCP is set, we change the values for \d etc. */
1452 if ((options & PCRE_UCP) != 0 && escape >= ESC_D && escape <= ESC_w)
1453 escape += (ESC_DU - ESC_D);
1455 /* Set the pointer to the final character before returning. */
1465 /*************************************************
1466 * Handle \P and \p *
1467 *************************************************/
1469 /* This function is called after \P or \p has been encountered, provided that
1470 PCRE is compiled with support for Unicode properties. On entry, ptrptr is
1471 pointing at the P or p. On exit, it is pointing at the final character of the
1475 ptrptr points to the pattern position pointer
1476 negptr points to a boolean that is set TRUE for negation else FALSE
1477 ptypeptr points to an unsigned int that is set to the type value
1478 pdataptr points to an unsigned int that is set to the detailed property value
1479 errorcodeptr points to the error code variable
1481 Returns: TRUE if the type value was found, or FALSE for an invalid type
1485 get_ucp(const pcre_uchar **ptrptr, BOOL *negptr, unsigned int *ptypeptr,
1486 unsigned int *pdataptr, int *errorcodeptr)
1490 const pcre_uchar *ptr = *ptrptr;
1491 pcre_uchar name[32];
1494 if (c == CHAR_NULL) goto ERROR_RETURN;
1498 /* \P or \p can be followed by a name in {}, optionally preceded by ^ for
1501 if (c == CHAR_LEFT_CURLY_BRACKET)
1503 if (ptr[1] == CHAR_CIRCUMFLEX_ACCENT)
1508 for (i = 0; i < (int)(sizeof(name) / sizeof(pcre_uchar)) - 1; i++)
1511 if (c == CHAR_NULL) goto ERROR_RETURN;
1512 if (c == CHAR_RIGHT_CURLY_BRACKET) break;
1515 if (c != CHAR_RIGHT_CURLY_BRACKET) goto ERROR_RETURN;
1519 /* Otherwise there is just one following character */
1529 /* Search for a recognized property name using binary chop */
1532 top = PRIV(utt_size);
1537 i = (bot + top) >> 1;
1538 r = STRCMP_UC_C8(name, PRIV(utt_names) + PRIV(utt)[i].name_offset);
1541 *ptypeptr = PRIV(utt)[i].type;
1542 *pdataptr = PRIV(utt)[i].value;
1545 if (r > 0) bot = i + 1; else top = i;
1548 *errorcodeptr = ERR47;
1553 *errorcodeptr = ERR46;
1561 /*************************************************
1562 * Read repeat counts *
1563 *************************************************/
1565 /* Read an item of the form {n,m} and return the values. This is called only
1566 after is_counted_repeat() has confirmed that a repeat-count quantifier exists,
1567 so the syntax is guaranteed to be correct, but we need to check the values.
1570 p pointer to first char after '{'
1571 minp pointer to int for min
1572 maxp pointer to int for max
1573 returned as -1 if no max
1574 errorcodeptr points to error code variable
1576 Returns: pointer to '}' on success;
1577 current ptr on error, with errorcodeptr set non-zero
1580 static const pcre_uchar *
1581 read_repeat_counts(const pcre_uchar *p, int *minp, int *maxp, int *errorcodeptr)
1586 while (IS_DIGIT(*p))
1588 min = min * 10 + (int)(*p++ - CHAR_0);
1591 *errorcodeptr = ERR5;
1596 if (*p == CHAR_RIGHT_CURLY_BRACKET) max = min; else
1598 if (*(++p) != CHAR_RIGHT_CURLY_BRACKET)
1603 max = max * 10 + (int)(*p++ - CHAR_0);
1606 *errorcodeptr = ERR5;
1612 *errorcodeptr = ERR4;
1625 /*************************************************
1626 * Find first significant op code *
1627 *************************************************/
1629 /* This is called by several functions that scan a compiled expression looking
1630 for a fixed first character, or an anchoring op code etc. It skips over things
1631 that do not influence this. For some calls, it makes sense to skip negative
1632 forward and all backward assertions, and also the \b assertion; for others it
1636 code pointer to the start of the group
1637 skipassert TRUE if certain assertions are to be skipped
1639 Returns: pointer to the first significant opcode
1642 static const pcre_uchar*
1643 first_significant_code(const pcre_uchar *code, BOOL skipassert)
1651 case OP_ASSERTBACK_NOT:
1652 if (!skipassert) return code;
1653 do code += GET(code, 1); while (*code == OP_ALT);
1654 code += PRIV(OP_lengths)[*code];
1657 case OP_WORD_BOUNDARY:
1658 case OP_NOT_WORD_BOUNDARY:
1659 if (!skipassert) return code;
1668 code += PRIV(OP_lengths)[*code];
1675 /* Control never reaches here */
1680 /*************************************************
1681 * Find the fixed length of a branch *
1682 *************************************************/
1684 /* Scan a branch and compute the fixed length of subject that will match it,
1685 if the length is fixed. This is needed for dealing with backward assertions.
1686 In UTF8 mode, the result is in characters rather than bytes. The branch is
1687 temporarily terminated with OP_END when this function is called.
1689 This function is called when a backward assertion is encountered, so that if it
1690 fails, the error message can point to the correct place in the pattern.
1691 However, we cannot do this when the assertion contains subroutine calls,
1692 because they can be forward references. We solve this by remembering this case
1693 and doing the check at the end; a flag specifies which mode we are running in.
1696 code points to the start of the pattern (the bracket)
1697 utf TRUE in UTF-8 / UTF-16 / UTF-32 mode
1698 atend TRUE if called when the pattern is complete
1699 cd the "compile data" structure
1701 Returns: the fixed length,
1702 or -1 if there is no fixed length,
1703 or -2 if \C was encountered (in UTF-8 mode only)
1704 or -3 if an OP_RECURSE item was encountered and atend is FALSE
1705 or -4 if an unknown opcode was encountered (internal error)
1709 find_fixedlength(pcre_uchar *code, BOOL utf, BOOL atend, compile_data *cd)
1713 register int branchlength = 0;
1714 register pcre_uchar *cc = code + 1 + LINK_SIZE;
1716 /* Scan along the opcodes for this branch. If we get to the end of the
1717 branch, check the length against that of the other branches. */
1722 pcre_uchar *ce, *cs;
1723 register pcre_uchar op = *cc;
1727 /* We only need to continue for OP_CBRA (normal capturing bracket) and
1728 OP_BRA (normal non-capturing bracket) because the other variants of these
1729 opcodes are all concerned with unlimited repeated groups, which of course
1730 are not of fixed length. */
1737 d = find_fixedlength(cc + ((op == OP_CBRA)? IMM2_SIZE : 0), utf, atend, cd);
1738 if (d < 0) return d;
1740 do cc += GET(cc, 1); while (*cc == OP_ALT);
1741 cc += 1 + LINK_SIZE;
1744 /* Reached end of a branch; if it's a ket it is the end of a nested call.
1745 If it's ALT it is an alternation in a nested call. An ACCEPT is effectively
1746 an ALT. If it is END it's the end of the outer call. All can be handled by
1747 the same code. Note that we must not include the OP_KETRxxx opcodes here,
1748 because they all imply an unlimited repeat. */
1754 case OP_ASSERT_ACCEPT:
1755 if (length < 0) length = branchlength;
1756 else if (length != branchlength) return -1;
1757 if (*cc != OP_ALT) return length;
1758 cc += 1 + LINK_SIZE;
1762 /* A true recursion implies not fixed length, but a subroutine call may
1763 be OK. If the subroutine is a forward reference, we can't deal with
1764 it until the end of the pattern, so return -3. */
1767 if (!atend) return -3;
1768 cs = ce = (pcre_uchar *)cd->start_code + GET(cc, 1); /* Start subpattern */
1769 do ce += GET(ce, 1); while (*ce == OP_ALT); /* End subpattern */
1770 if (cc > cs && cc < ce) return -1; /* Recursion */
1771 d = find_fixedlength(cs + IMM2_SIZE, utf, atend, cd);
1772 if (d < 0) return d;
1774 cc += 1 + LINK_SIZE;
1777 /* Skip over assertive subpatterns */
1782 case OP_ASSERTBACK_NOT:
1783 do cc += GET(cc, 1); while (*cc == OP_ALT);
1784 cc += PRIV(OP_lengths)[*cc];
1787 /* Skip over things that don't match chars */
1793 cc += cc[1] + PRIV(OP_lengths)[*cc];
1810 case OP_NOT_WORD_BOUNDARY:
1819 case OP_WORD_BOUNDARY:
1820 cc += PRIV(OP_lengths)[*cc];
1823 /* Handle literal characters */
1832 if (utf && HAS_EXTRALEN(cc[-1])) cc += GET_EXTRALEN(cc[-1]);
1836 /* Handle exact repetitions. The count is already in characters, but we
1837 need to skip over a multibyte character in UTF8 mode. */
1843 branchlength += (int)GET2(cc,1);
1844 cc += 2 + IMM2_SIZE;
1846 if (utf && HAS_EXTRALEN(cc[-1])) cc += GET_EXTRALEN(cc[-1]);
1851 branchlength += GET2(cc,1);
1852 if (cc[1 + IMM2_SIZE] == OP_PROP || cc[1 + IMM2_SIZE] == OP_NOTPROP)
1854 cc += 1 + IMM2_SIZE + 1;
1857 /* Handle single-char matchers */
1870 case OP_NOT_WHITESPACE:
1872 case OP_NOT_WORDCHAR:
1880 /* The single-byte matcher isn't allowed. This only happens in UTF-8 mode;
1881 otherwise \C is coded as OP_ALLANY. */
1886 /* Check a class for variable quantification */
1890 #if defined SUPPORT_UTF || defined COMPILE_PCRE16 || defined COMPILE_PCRE32
1892 /* The original code caused an unsigned overflow in 64 bit systems,
1893 so now we use a conditional statement. */
1894 if (op == OP_XCLASS)
1897 cc += PRIV(OP_lengths)[OP_CLASS];
1899 cc += PRIV(OP_lengths)[OP_CLASS];
1918 if (GET2(cc,1) != GET2(cc,1+IMM2_SIZE)) return -1;
1919 branchlength += (int)GET2(cc,1);
1920 cc += 1 + 2 * IMM2_SIZE;
1928 /* Anything else is variable length */
1949 case OP_NOTMINPLUSI:
1950 case OP_NOTMINQUERY:
1951 case OP_NOTMINQUERYI:
1953 case OP_NOTMINSTARI:
1955 case OP_NOTMINUPTOI:
1959 case OP_NOTPOSPLUSI:
1960 case OP_NOTPOSQUERY:
1961 case OP_NOTPOSQUERYI:
1963 case OP_NOTPOSSTARI:
1965 case OP_NOTPOSUPTOI:
1996 case OP_TYPEMINPLUS:
1997 case OP_TYPEMINQUERY:
1998 case OP_TYPEMINSTAR:
1999 case OP_TYPEMINUPTO:
2001 case OP_TYPEPOSPLUS:
2002 case OP_TYPEPOSQUERY:
2003 case OP_TYPEPOSSTAR:
2004 case OP_TYPEPOSUPTO:
2012 /* Catch unrecognized opcodes so that when new ones are added they
2013 are not forgotten, as has happened in the past. */
2019 /* Control never gets here */
2024 /*************************************************
2025 * Scan compiled regex for specific bracket *
2026 *************************************************/
2028 /* This little function scans through a compiled pattern until it finds a
2029 capturing bracket with the given number, or, if the number is negative, an
2030 instance of OP_REVERSE for a lookbehind. The function is global in the C sense
2031 so that it can be called from pcre_study() when finding the minimum matching
2035 code points to start of expression
2036 utf TRUE in UTF-8 / UTF-16 / UTF-32 mode
2037 number the required bracket number or negative to find a lookbehind
2039 Returns: pointer to the opcode for the bracket, or NULL if not found
2043 PRIV(find_bracket)(const pcre_uchar *code, BOOL utf, int number)
2047 register pcre_uchar c = *code;
2049 if (c == OP_END) return NULL;
2051 /* XCLASS is used for classes that cannot be represented just by a bit
2052 map. This includes negated single high-valued characters. The length in
2053 the table is zero; the actual length is stored in the compiled code. */
2055 if (c == OP_XCLASS) code += GET(code, 1);
2057 /* Handle recursion */
2059 else if (c == OP_REVERSE)
2061 if (number < 0) return (pcre_uchar *)code;
2062 code += PRIV(OP_lengths)[c];
2065 /* Handle capturing bracket */
2067 else if (c == OP_CBRA || c == OP_SCBRA ||
2068 c == OP_CBRAPOS || c == OP_SCBRAPOS)
2070 int n = (int)GET2(code, 1+LINK_SIZE);
2071 if (n == number) return (pcre_uchar *)code;
2072 code += PRIV(OP_lengths)[c];
2075 /* Otherwise, we can get the item's length from the table, except that for
2076 repeated character types, we have to test for \p and \P, which have an extra
2077 two bytes of parameters, and for MARK/PRUNE/SKIP/THEN with an argument, we
2078 must add in its length. */
2085 case OP_TYPEMINSTAR:
2087 case OP_TYPEMINPLUS:
2089 case OP_TYPEMINQUERY:
2090 case OP_TYPEPOSSTAR:
2091 case OP_TYPEPOSPLUS:
2092 case OP_TYPEPOSQUERY:
2093 if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
2097 case OP_TYPEMINUPTO:
2099 case OP_TYPEPOSUPTO:
2100 if (code[1 + IMM2_SIZE] == OP_PROP || code[1 + IMM2_SIZE] == OP_NOTPROP)
2112 /* Add in the fixed length from the table */
2114 code += PRIV(OP_lengths)[c];
2116 /* In UTF-8 mode, opcodes that are followed by a character may be followed by
2117 a multi-byte character. The length in the table is a minimum, so we have to
2118 arrange to skip the extra bytes. */
2120 #if defined SUPPORT_UTF && !defined COMPILE_PCRE32
2151 if (HAS_EXTRALEN(code[-1])) code += GET_EXTRALEN(code[-1]);
2155 (void)(utf); /* Keep compiler happy by referencing function argument */
2163 /*************************************************
2164 * Scan compiled regex for recursion reference *
2165 *************************************************/
2167 /* This little function scans through a compiled pattern until it finds an
2168 instance of OP_RECURSE.
2171 code points to start of expression
2172 utf TRUE in UTF-8 / UTF-16 / UTF-32 mode
2174 Returns: pointer to the opcode for OP_RECURSE, or NULL if not found
2177 static const pcre_uchar *
2178 find_recurse(const pcre_uchar *code, BOOL utf)
2182 register pcre_uchar c = *code;
2183 if (c == OP_END) return NULL;
2184 if (c == OP_RECURSE) return code;
2186 /* XCLASS is used for classes that cannot be represented just by a bit
2187 map. This includes negated single high-valued characters. The length in
2188 the table is zero; the actual length is stored in the compiled code. */
2190 if (c == OP_XCLASS) code += GET(code, 1);
2192 /* Otherwise, we can get the item's length from the table, except that for
2193 repeated character types, we have to test for \p and \P, which have an extra
2194 two bytes of parameters, and for MARK/PRUNE/SKIP/THEN with an argument, we
2195 must add in its length. */
2202 case OP_TYPEMINSTAR:
2204 case OP_TYPEMINPLUS:
2206 case OP_TYPEMINQUERY:
2207 case OP_TYPEPOSSTAR:
2208 case OP_TYPEPOSPLUS:
2209 case OP_TYPEPOSQUERY:
2210 if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
2213 case OP_TYPEPOSUPTO:
2215 case OP_TYPEMINUPTO:
2217 if (code[1 + IMM2_SIZE] == OP_PROP || code[1 + IMM2_SIZE] == OP_NOTPROP)
2229 /* Add in the fixed length from the table */
2231 code += PRIV(OP_lengths)[c];
2233 /* In UTF-8 mode, opcodes that are followed by a character may be followed
2234 by a multi-byte character. The length in the table is a minimum, so we have
2235 to arrange to skip the extra bytes. */
2237 #if defined SUPPORT_UTF && !defined COMPILE_PCRE32
2255 case OP_NOTMINUPTOI:
2259 case OP_NOTPOSUPTOI:
2267 case OP_NOTMINSTARI:
2271 case OP_NOTPOSSTARI:
2279 case OP_NOTMINPLUSI:
2283 case OP_NOTPOSPLUSI:
2290 case OP_NOTMINQUERY:
2291 case OP_NOTMINQUERYI:
2294 case OP_NOTPOSQUERY:
2295 case OP_NOTPOSQUERYI:
2296 if (HAS_EXTRALEN(code[-1])) code += GET_EXTRALEN(code[-1]);
2300 (void)(utf); /* Keep compiler happy by referencing function argument */
2308 /*************************************************
2309 * Scan compiled branch for non-emptiness *
2310 *************************************************/
2312 /* This function scans through a branch of a compiled pattern to see whether it
2313 can match the empty string or not. It is called from could_be_empty()
2314 below and from compile_branch() when checking for an unlimited repeat of a
2315 group that can match nothing. Note that first_significant_code() skips over
2316 backward and negative forward assertions when its final argument is TRUE. If we
2317 hit an unclosed bracket, we return "empty" - this means we've struck an inner
2318 bracket whose current branch will already have been scanned.
2321 code points to start of search
2322 endcode points to where to stop
2323 utf TRUE if in UTF-8 / UTF-16 / UTF-32 mode
2324 cd contains pointers to tables etc.
2325 recurses chain of recurse_check to catch mutual recursion
2327 Returns: TRUE if what is matched could be empty
2330 typedef struct recurse_check {
2331 struct recurse_check *prev;
2332 const pcre_uchar *group;
2336 could_be_empty_branch(const pcre_uchar *code, const pcre_uchar *endcode,
2337 BOOL utf, compile_data *cd, recurse_check *recurses)
2339 register pcre_uchar c;
2340 recurse_check this_recurse;
2342 for (code = first_significant_code(code + PRIV(OP_lengths)[*code], TRUE);
2344 code = first_significant_code(code + PRIV(OP_lengths)[c], TRUE))
2346 const pcre_uchar *ccode;
2350 /* Skip over forward assertions; the other assertions are skipped by
2351 first_significant_code() with a TRUE final argument. */
2355 do code += GET(code, 1); while (*code == OP_ALT);
2360 /* For a recursion/subroutine call, if its end has been reached, which
2361 implies a backward reference subroutine call, we can scan it. If it's a
2362 forward reference subroutine call, we can't. To detect forward reference
2363 we have to scan up the list that is kept in the workspace. This function is
2364 called only when doing the real compile, not during the pre-compile that
2365 measures the size of the compiled pattern. */
2367 if (c == OP_RECURSE)
2369 const pcre_uchar *scode = cd->start_code + GET(code, 1);
2370 const pcre_uchar *endgroup = scode;
2373 /* Test for forward reference or uncompleted reference. This is disabled
2374 when called to scan a completed pattern by setting cd->start_workspace to
2377 if (cd->start_workspace != NULL)
2379 const pcre_uchar *tcode;
2380 for (tcode = cd->start_workspace; tcode < cd->hwm; tcode += LINK_SIZE)
2381 if ((int)GET(tcode, 0) == (int)(code + 1 - cd->start_code)) return TRUE;
2382 if (GET(scode, 1) == 0) return TRUE; /* Unclosed */
2385 /* If the reference is to a completed group, we need to detect whether this
2386 is a recursive call, as otherwise there will be an infinite loop. If it is
2387 a recursion, just skip over it. Simple recursions are easily detected. For
2388 mutual recursions we keep a chain on the stack. */
2390 do endgroup += GET(endgroup, 1); while (*endgroup == OP_ALT);
2391 if (code >= scode && code <= endgroup) continue; /* Simple recursion */
2394 recurse_check *r = recurses;
2395 for (r = recurses; r != NULL; r = r->prev)
2396 if (r->group == scode) break;
2397 if (r != NULL) continue; /* Mutual recursion */
2400 /* Completed reference; scan the referenced group, remembering it on the
2401 stack chain to detect mutual recursions. */
2403 empty_branch = FALSE;
2404 this_recurse.prev = recurses;
2405 this_recurse.group = scode;
2409 if (could_be_empty_branch(scode, endcode, utf, cd, &this_recurse))
2411 empty_branch = TRUE;
2414 scode += GET(scode, 1);
2416 while (*scode == OP_ALT);
2418 if (!empty_branch) return FALSE; /* All branches are non-empty */
2422 /* Groups with zero repeats can of course be empty; skip them. */
2424 if (c == OP_BRAZERO || c == OP_BRAMINZERO || c == OP_SKIPZERO ||
2427 code += PRIV(OP_lengths)[c];
2428 do code += GET(code, 1); while (*code == OP_ALT);
2433 /* A nested group that is already marked as "could be empty" can just be
2436 if (c == OP_SBRA || c == OP_SBRAPOS ||
2437 c == OP_SCBRA || c == OP_SCBRAPOS)
2439 do code += GET(code, 1); while (*code == OP_ALT);
2444 /* For other groups, scan the branches. */
2446 if (c == OP_BRA || c == OP_BRAPOS ||
2447 c == OP_CBRA || c == OP_CBRAPOS ||
2448 c == OP_ONCE || c == OP_ONCE_NC ||
2452 if (GET(code, 1) == 0) return TRUE; /* Hit unclosed bracket */
2454 /* If a conditional group has only one branch, there is a second, implied,
2455 empty branch, so just skip over the conditional, because it could be empty.
2456 Otherwise, scan the individual branches of the group. */
2458 if (c == OP_COND && code[GET(code, 1)] != OP_ALT)
2459 code += GET(code, 1);
2462 empty_branch = FALSE;
2465 if (!empty_branch && could_be_empty_branch(code, endcode, utf, cd, NULL))
2466 empty_branch = TRUE;
2467 code += GET(code, 1);
2469 while (*code == OP_ALT);
2470 if (!empty_branch) return FALSE; /* All branches are non-empty */
2477 /* Handle the other opcodes */
2481 /* Check for quantifiers after a class. XCLASS is used for classes that
2482 cannot be represented just by a bit map. This includes negated single
2483 high-valued characters. The length in PRIV(OP_lengths)[] is zero; the
2484 actual length is stored in the compiled code, so we must update "code"
2487 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
2489 ccode = code += GET(code, 1);
2490 goto CHECK_CLASS_REPEAT;
2495 ccode = code + PRIV(OP_lengths)[OP_CLASS];
2497 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
2503 case OP_CRSTAR: /* These could be empty; continue */
2511 default: /* Non-repeat => class must match */
2512 case OP_CRPLUS: /* These repeats aren't empty */
2520 if (GET2(ccode, 1) > 0) return FALSE; /* Minimum > 0 */
2525 /* Opcodes that must match a character */
2543 case OP_NOT_WHITESPACE:
2545 case OP_NOT_WORDCHAR:
2561 case OP_NOTMINPLUSI:
2566 case OP_NOTPOSPLUSI:
2574 case OP_TYPEMINPLUS:
2575 case OP_TYPEPOSPLUS:
2580 /* These are going to continue, as they may be empty, but we have to
2581 fudge the length for the \p and \P cases. */
2584 case OP_TYPEMINSTAR:
2585 case OP_TYPEPOSSTAR:
2587 case OP_TYPEMINQUERY:
2588 case OP_TYPEPOSQUERY:
2589 if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
2592 /* Same for these */
2595 case OP_TYPEMINUPTO:
2596 case OP_TYPEPOSUPTO:
2597 if (code[1 + IMM2_SIZE] == OP_PROP || code[1 + IMM2_SIZE] == OP_NOTPROP)
2610 /* In UTF-8 mode, STAR, MINSTAR, POSSTAR, QUERY, MINQUERY, POSQUERY, UPTO,
2611 MINUPTO, and POSUPTO and their caseless and negative versions may be
2612 followed by a multibyte character. */
2614 #if defined SUPPORT_UTF && !defined COMPILE_PCRE32
2623 case OP_NOTMINSTARI:
2628 case OP_NOTPOSSTARI:
2637 case OP_NOTMINQUERY:
2638 case OP_NOTMINQUERYI:
2642 case OP_NOTPOSQUERY:
2643 case OP_NOTPOSQUERYI:
2645 if (utf && HAS_EXTRALEN(code[1])) code += GET_EXTRALEN(code[1]);
2656 case OP_NOTMINUPTOI:
2661 case OP_NOTPOSUPTOI:
2663 if (utf && HAS_EXTRALEN(code[1 + IMM2_SIZE])) code += GET_EXTRALEN(code[1 + IMM2_SIZE]);
2667 /* MARK, and PRUNE/SKIP/THEN with an argument must skip over the argument
2677 /* None of the remaining opcodes are required to match a character. */
2689 /*************************************************
2690 * Scan compiled regex for non-emptiness *
2691 *************************************************/
2693 /* This function is called to check for left recursive calls. We want to check
2694 the current branch of the current pattern to see if it could match the empty
2695 string. If it could, we must look outwards for branches at other levels,
2696 stopping when we pass beyond the bracket which is the subject of the recursion.
2697 This function is called only during the real compile, not during the
2701 code points to start of the recursion
2702 endcode points to where to stop (current RECURSE item)
2703 bcptr points to the chain of current (unclosed) branch starts
2704 utf TRUE if in UTF-8 / UTF-16 / UTF-32 mode
2705 cd pointers to tables etc
2707 Returns: TRUE if what is matched could be empty
2711 could_be_empty(const pcre_uchar *code, const pcre_uchar *endcode,
2712 branch_chain *bcptr, BOOL utf, compile_data *cd)
2714 while (bcptr != NULL && bcptr->current_branch >= code)
2716 if (!could_be_empty_branch(bcptr->current_branch, endcode, utf, cd, NULL))
2718 bcptr = bcptr->outer;
2725 /*************************************************
2726 * Base opcode of repeated opcodes *
2727 *************************************************/
2729 /* Returns the base opcode for repeated single character type opcodes. If the
2730 opcode is not a repeated character type, it returns with the original value.
2733 Returns: base opcode for the type
2737 get_repeat_base(pcre_uchar c)
2739 return (c > OP_TYPEPOSUPTO)? c :
2740 (c >= OP_TYPESTAR)? OP_TYPESTAR :
2741 (c >= OP_NOTSTARI)? OP_NOTSTARI :
2742 (c >= OP_NOTSTAR)? OP_NOTSTAR :
2743 (c >= OP_STARI)? OP_STARI :
2750 /*************************************************
2751 * Check a character and a property *
2752 *************************************************/
2754 /* This function is called by check_auto_possessive() when a property item
2755 is adjacent to a fixed character.
2759 ptype the property type
2760 pdata the data for the type
2761 negated TRUE if it's a negated property (\P or \p{^)
2763 Returns: TRUE if auto-possessifying is OK
2767 check_char_prop(pcre_uint32 c, unsigned int ptype, unsigned int pdata,
2770 const pcre_uint32 *p;
2771 const ucd_record *prop = GET_UCD(c);
2776 return (prop->chartype == ucp_Lu ||
2777 prop->chartype == ucp_Ll ||
2778 prop->chartype == ucp_Lt) == negated;
2781 return (pdata == PRIV(ucp_gentype)[prop->chartype]) == negated;
2784 return (pdata == prop->chartype) == negated;
2787 return (pdata == prop->script) == negated;
2789 /* These are specials */
2792 return (PRIV(ucp_gentype)[prop->chartype] == ucp_L ||
2793 PRIV(ucp_gentype)[prop->chartype] == ucp_N) == negated;
2795 /* Perl space used to exclude VT, but from Perl 5.18 it is included, which
2796 means that Perl space and POSIX space are now identical. PCRE was changed
2799 case PT_SPACE: /* Perl space */
2800 case PT_PXSPACE: /* POSIX space */
2808 return (PRIV(ucp_gentype)[prop->chartype] == ucp_Z) == negated;
2810 break; /* Control never reaches here */
2813 return (PRIV(ucp_gentype)[prop->chartype] == ucp_L ||
2814 PRIV(ucp_gentype)[prop->chartype] == ucp_N ||
2815 c == CHAR_UNDERSCORE) == negated;
2818 p = PRIV(ucd_caseless_sets) + prop->caseset;
2821 if (c < *p) return !negated;
2822 if (c == *p++) return negated;
2824 break; /* Control never reaches here */
2829 #endif /* SUPPORT_UCP */
2833 /*************************************************
2834 * Fill the character property list *
2835 *************************************************/
2837 /* Checks whether the code points to an opcode that can take part in auto-
2838 possessification, and if so, fills a list with its properties.
2841 code points to start of expression
2842 utf TRUE if in UTF-8 / UTF-16 / UTF-32 mode
2843 fcc points to case-flipping table
2844 list points to output list
2845 list[0] will be filled with the opcode
2846 list[1] will be non-zero if this opcode
2847 can match an empty character string
2848 list[2..7] depends on the opcode
2850 Returns: points to the start of the next opcode if *code is accepted
2851 NULL if *code is not accepted
2854 static const pcre_uchar *
2855 get_chr_property_list(const pcre_uchar *code, BOOL utf,
2856 const pcre_uint8 *fcc, pcre_uint32 *list)
2858 pcre_uchar c = *code;
2860 const pcre_uchar *end;
2864 pcre_uint32 *clist_dest;
2865 const pcre_uint32 *clist_src;
2867 utf = utf; /* Suppress "unused parameter" compiler warning */
2874 if (c >= OP_STAR && c <= OP_TYPEPOSUPTO)
2876 base = get_repeat_base(c);
2877 c -= (base - OP_STAR);
2879 if (c == OP_UPTO || c == OP_MINUPTO || c == OP_EXACT || c == OP_POSUPTO)
2882 list[1] = (c != OP_PLUS && c != OP_MINPLUS && c != OP_EXACT && c != OP_POSPLUS);
2914 case OP_NOT_WHITESPACE:
2916 case OP_NOT_WORDCHAR:
2934 GETCHARINCTEST(chr, code);
2941 list[0] = (c == OP_CHARI) ? OP_CHAR : OP_NOT;
2942 GETCHARINCTEST(chr, code);
2946 if (chr < 128 || (chr < 256 && !utf))
2949 list[3] = UCD_OTHERCASE(chr);
2950 #elif defined SUPPORT_UTF || !defined COMPILE_PCRE8
2951 list[3] = (chr < 256) ? fcc[chr] : chr;
2956 /* The othercase might be the same value. */
2967 if (code[0] != PT_CLIST)
2974 /* Convert only if we have enough space. */
2976 clist_src = PRIV(ucd_caseless_sets) + code[1];
2977 clist_dest = list + 2;
2981 if (clist_dest >= list + 8)
2983 /* Early return if there is not enough space. This should never
2984 happen, since all clists are shorter than 5 character now. */
2989 *clist_dest++ = *clist_src;
2991 while(*clist_src++ != NOTACHAR);
2993 /* All characters are stored. The terminating NOTACHAR
2994 is copied form the clist itself. */
2996 list[0] = (c == OP_PROP) ? OP_CHAR : OP_NOT;
3002 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
3005 end = code + GET(code, 0) - 1;
3008 end = code + 32 / sizeof(pcre_uchar);
3031 list[1] = (GET2(end, 1) == 0);
3032 end += 1 + 2 * IMM2_SIZE;
3035 list[2] = end - code;
3038 return NULL; /* Opcode not accepted */
3043 /*************************************************
3044 * Scan further character sets for match *
3045 *************************************************/
3047 /* Checks whether the base and the current opcode have a common character, in
3048 which case the base cannot be possessified.
3051 code points to the byte code
3052 utf TRUE in UTF-8 / UTF-16 / UTF-32 mode
3053 cd static compile data
3054 base_list the data list of the base opcode
3056 Returns: TRUE if the auto-possessification is possible
3060 compare_opcodes(const pcre_uchar *code, BOOL utf, const compile_data *cd,
3061 const pcre_uint32 *base_list, const pcre_uchar *base_end)
3064 pcre_uint32 list[8];
3065 const pcre_uint32 *chr_ptr;
3066 const pcre_uint32 *ochr_ptr;
3067 const pcre_uint32 *list_ptr;
3068 const pcre_uchar *next_code;
3069 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
3070 const pcre_uchar *xclass_flags;
3072 const pcre_uint8 *class_bitset;
3073 const pcre_uint8 *set1, *set2, *set_end;
3075 BOOL accepted, invert_bits;
3077 /* Note: the base_list[1] contains whether the current opcode has greedy
3078 (represented by a non-zero value) quantifier. This is a different from
3079 other character type lists, which stores here that the character iterator
3080 matches to an empty string (also represented by a non-zero value). */
3084 /* All operations move the code pointer forward.
3085 Therefore infinite recursions are not possible. */
3089 /* Skip over callouts */
3091 if (c == OP_CALLOUT)
3093 code += PRIV(OP_lengths)[c];
3099 do code += GET(code, 1); while (*code == OP_ALT);
3107 /* TRUE only in greedy case. The non-greedy case could be replaced by
3108 an OP_EXACT, but it is probably not worth it. (And note that OP_EXACT
3109 uses more memory, which we cannot get at this stage.) */
3111 return base_list[1] != 0;
3114 /* If the bracket is capturing, and referenced by an OP_RECURSE, or
3115 it is an atomic sub-pattern (assert, once, etc.) the non-greedy case
3116 cannot be converted to a possessive form. */
3118 if (base_list[1] == 0) return FALSE;
3120 switch(*(code - GET(code, 1)))
3125 case OP_ASSERTBACK_NOT:
3128 /* Atomic sub-patterns and assertions can always auto-possessify their
3133 code += PRIV(OP_lengths)[c];
3140 next_code = code + GET(code, 1);
3141 code += PRIV(OP_lengths)[c];
3143 while (*next_code == OP_ALT)
3145 if (!compare_opcodes(code, utf, cd, base_list, base_end)) return FALSE;
3146 code = next_code + 1 + LINK_SIZE;
3147 next_code += GET(next_code, 1);
3154 next_code = code + 1;
3155 if (*next_code != OP_BRA && *next_code != OP_CBRA
3156 && *next_code != OP_ONCE && *next_code != OP_ONCE_NC) return FALSE;
3158 do next_code += GET(next_code, 1); while (*next_code == OP_ALT);
3160 /* The bracket content will be checked by the
3161 OP_BRA/OP_CBRA case above. */
3162 next_code += 1 + LINK_SIZE;
3163 if (!compare_opcodes(next_code, utf, cd, base_list, base_end))
3166 code += PRIV(OP_lengths)[c];
3170 /* Check for a supported opcode, and load its properties. */
3172 code = get_chr_property_list(code, utf, cd->fcc, list);
3173 if (code == NULL) return FALSE; /* Unsupported */
3175 /* If either opcode is a small character list, set pointers for comparing
3176 characters from that list with another list, or with a property. */
3178 if (base_list[0] == OP_CHAR)
3180 chr_ptr = base_list + 2;
3183 else if (list[0] == OP_CHAR)
3186 list_ptr = base_list;
3189 /* Character bitsets can also be compared to certain opcodes. */
3191 else if (base_list[0] == OP_CLASS || list[0] == OP_CLASS
3192 #ifdef COMPILE_PCRE8
3193 /* In 8 bit, non-UTF mode, OP_CLASS and OP_NCLASS are the same. */
3194 || (!utf && (base_list[0] == OP_NCLASS || list[0] == OP_NCLASS))
3198 #ifdef COMPILE_PCRE8
3199 if (base_list[0] == OP_CLASS || (!utf && base_list[0] == OP_NCLASS))
3201 if (base_list[0] == OP_CLASS)
3204 set1 = (pcre_uint8 *)(base_end - base_list[2]);
3209 set1 = (pcre_uint8 *)(code - list[2]);
3210 list_ptr = base_list;
3213 invert_bits = FALSE;
3218 set2 = (pcre_uint8 *)
3219 ((list_ptr == list ? code : base_end) - list_ptr[2]);
3222 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
3224 xclass_flags = (list_ptr == list ? code : base_end) - list_ptr[2] + LINK_SIZE;
3225 if ((*xclass_flags & XCL_HASPROP) != 0) return FALSE;
3226 if ((*xclass_flags & XCL_MAP) == 0)
3228 /* No bits are set for characters < 256. */
3229 if (list[1] == 0) return TRUE;
3230 /* Might be an empty repeat. */
3233 set2 = (pcre_uint8 *)(xclass_flags + 1);
3241 set2 = (pcre_uint8 *)(cd->cbits + cbit_digit);
3244 case OP_NOT_WHITESPACE:
3248 set2 = (pcre_uint8 *)(cd->cbits + cbit_space);
3251 case OP_NOT_WORDCHAR:
3255 set2 = (pcre_uint8 *)(cd->cbits + cbit_word);
3262 /* Because the sets are unaligned, we need
3263 to perform byte comparison here. */
3264 set_end = set1 + 32;
3269 if ((*set1++ & ~(*set2++)) != 0) return FALSE;
3271 while (set1 < set_end);
3277 if ((*set1++ & *set2++) != 0) return FALSE;
3279 while (set1 < set_end);
3282 if (list[1] == 0) return TRUE;
3283 /* Might be an empty repeat. */
3287 /* Some property combinations also acceptable. Unicode property opcodes are
3288 processed specially; the rest can be handled with a lookup table. */
3292 pcre_uint32 leftop, rightop;
3294 leftop = base_list[0];
3298 accepted = FALSE; /* Always set in non-unicode case. */
3299 if (leftop == OP_PROP || leftop == OP_NOTPROP)
3301 if (rightop == OP_EOD)
3303 else if (rightop == OP_PROP || rightop == OP_NOTPROP)
3306 const pcre_uint8 *p;
3307 BOOL same = leftop == rightop;
3308 BOOL lisprop = leftop == OP_PROP;
3309 BOOL risprop = rightop == OP_PROP;
3310 BOOL bothprop = lisprop && risprop;
3312 /* There's a table that specifies how each combination is to be
3314 0 Always return FALSE (never auto-possessify)
3315 1 Character groups are distinct (possessify if both are OP_PROP)
3316 2 Check character categories in the same group (general or particular)
3317 3 Return TRUE if the two opcodes are not the same
3318 ... see comments below
3321 n = propposstab[base_list[2]][list[2]];
3325 case 1: accepted = bothprop; break;
3326 case 2: accepted = (base_list[3] == list[3]) != same; break;
3327 case 3: accepted = !same; break;
3329 case 4: /* Left general category, right particular category */
3330 accepted = risprop && catposstab[base_list[3]][list[3]] == same;
3333 case 5: /* Right general category, left particular category */
3334 accepted = lisprop && catposstab[list[3]][base_list[3]] == same;
3337 /* This code is logically tricky. Think hard before fiddling with it.
3338 The posspropstab table has four entries per row. Each row relates to
3339 one of PCRE's special properties such as ALNUM or SPACE or WORD.
3340 Only WORD actually needs all four entries, but using repeats for the
3341 others means they can all use the same code below.
3343 The first two entries in each row are Unicode general categories, and
3344 apply always, because all the characters they include are part of the
3345 PCRE character set. The third and fourth entries are a general and a
3346 particular category, respectively, that include one or more relevant
3347 characters. One or the other is used, depending on whether the check
3348 is for a general or a particular category. However, in both cases the
3349 category contains more characters than the specials that are defined
3350 for the property being tested against. Therefore, it cannot be used
3353 Example: the row for WORD contains ucp_L, ucp_N, ucp_P, ucp_Po.
3354 Underscore is covered by ucp_P or ucp_Po. */
3356 case 6: /* Left alphanum vs right general category */
3357 case 7: /* Left space vs right general category */
3358 case 8: /* Left word vs right general category */
3359 p = posspropstab[n-6];
3360 accepted = risprop && lisprop ==
3363 (list[3] != p[2] || !lisprop));
3366 case 9: /* Right alphanum vs left general category */
3367 case 10: /* Right space vs left general category */
3368 case 11: /* Right word vs left general category */
3369 p = posspropstab[n-9];
3370 accepted = lisprop && risprop ==
3371 (base_list[3] != p[0] &&
3372 base_list[3] != p[1] &&
3373 (base_list[3] != p[2] || !risprop));
3376 case 12: /* Left alphanum vs right particular category */
3377 case 13: /* Left space vs right particular category */
3378 case 14: /* Left word vs right particular category */
3379 p = posspropstab[n-12];
3380 accepted = risprop && lisprop ==
3381 (catposstab[p[0]][list[3]] &&
3382 catposstab[p[1]][list[3]] &&
3383 (list[3] != p[3] || !lisprop));
3386 case 15: /* Right alphanum vs left particular category */
3387 case 16: /* Right space vs left particular category */
3388 case 17: /* Right word vs left particular category */
3389 p = posspropstab[n-15];
3390 accepted = lisprop && risprop ==
3391 (catposstab[p[0]][base_list[3]] &&
3392 catposstab[p[1]][base_list[3]] &&
3393 (base_list[3] != p[3] || !risprop));
3400 #endif /* SUPPORT_UCP */
3402 accepted = leftop >= FIRST_AUTOTAB_OP && leftop <= LAST_AUTOTAB_LEFT_OP &&
3403 rightop >= FIRST_AUTOTAB_OP && rightop <= LAST_AUTOTAB_RIGHT_OP &&
3404 autoposstab[leftop - FIRST_AUTOTAB_OP][rightop - FIRST_AUTOTAB_OP];
3409 if (list[1] == 0) return TRUE;
3410 /* Might be an empty repeat. */
3414 /* Control reaches here only if one of the items is a small character list.
3415 All characters are checked against the other side. */
3424 ochr_ptr = list_ptr + 2;
3427 if (chr == *ochr_ptr) return FALSE;
3430 while(*ochr_ptr != NOTACHAR);
3434 ochr_ptr = list_ptr + 2;
3437 if (chr == *ochr_ptr)
3441 while(*ochr_ptr != NOTACHAR);
3442 if (*ochr_ptr == NOTACHAR) return FALSE; /* Not found */
3445 /* Note that OP_DIGIT etc. are generated only when PCRE_UCP is *not*
3446 set. When it is set, \d etc. are converted into OP_(NOT_)PROP codes. */
3449 if (chr < 256 && (cd->ctypes[chr] & ctype_digit) != 0) return FALSE;
3453 if (chr > 255 || (cd->ctypes[chr] & ctype_digit) == 0) return FALSE;
3457 if (chr < 256 && (cd->ctypes[chr] & ctype_space) != 0) return FALSE;
3460 case OP_NOT_WHITESPACE:
3461 if (chr > 255 || (cd->ctypes[chr] & ctype_space) == 0) return FALSE;
3465 if (chr < 255 && (cd->ctypes[chr] & ctype_word) != 0) return FALSE;
3468 case OP_NOT_WORDCHAR:
3469 if (chr > 255 || (cd->ctypes[chr] & ctype_word) == 0) return FALSE;
3475 HSPACE_CASES: return FALSE;
3483 HSPACE_CASES: break;
3484 default: return FALSE;
3492 VSPACE_CASES: return FALSE;
3500 VSPACE_CASES: break;
3501 default: return FALSE;
3517 #endif /* Not EBCDIC */
3522 case OP_EOD: /* Can always possessify before \z */
3528 if (!check_char_prop(chr, list_ptr[2], list_ptr[3],
3529 list_ptr[0] == OP_NOTPROP))
3535 if (chr > 255) return FALSE;
3539 if (chr > 255) break;
3540 class_bitset = (pcre_uint8 *)
3541 ((list_ptr == list ? code : base_end) - list_ptr[2]);
3542 if ((class_bitset[chr >> 3] & (1 << (chr & 7))) != 0) return FALSE;
3545 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
3547 if (PRIV(xclass)(chr, (list_ptr == list ? code : base_end) -
3548 list_ptr[2] + LINK_SIZE, utf)) return FALSE;
3558 while(*chr_ptr != NOTACHAR);
3560 /* At least one character must be matched from this opcode. */
3562 if (list[1] == 0) return TRUE;
3565 /* Control never reaches here. There used to be a fail-save return FALSE; here,
3566 but some compilers complain about an unreachable statement. */
3572 /*************************************************
3573 * Scan compiled regex for auto-possession *
3574 *************************************************/
3576 /* Replaces single character iterations with their possessive alternatives
3577 if appropriate. This function modifies the compiled opcode!
3580 code points to start of the byte code
3581 utf TRUE in UTF-8 / UTF-16 / UTF-32 mode
3582 cd static compile data
3588 auto_possessify(pcre_uchar *code, BOOL utf, const compile_data *cd)
3590 register pcre_uchar c;
3591 const pcre_uchar *end;
3592 pcre_uchar *repeat_opcode;
3593 pcre_uint32 list[8];
3599 if (c >= OP_STAR && c <= OP_TYPEPOSUPTO)
3601 c -= get_repeat_base(c) - OP_STAR;
3602 end = (c <= OP_MINUPTO) ?
3603 get_chr_property_list(code, utf, cd->fcc, list) : NULL;
3604 list[1] = c == OP_STAR || c == OP_PLUS || c == OP_QUERY || c == OP_UPTO;
3606 if (end != NULL && compare_opcodes(end, utf, cd, list, end))
3611 *code += OP_POSSTAR - OP_STAR;
3615 *code += OP_POSSTAR - OP_MINSTAR;
3619 *code += OP_POSPLUS - OP_PLUS;
3623 *code += OP_POSPLUS - OP_MINPLUS;
3627 *code += OP_POSQUERY - OP_QUERY;
3631 *code += OP_POSQUERY - OP_MINQUERY;
3635 *code += OP_POSUPTO - OP_UPTO;
3639 *code += OP_POSUPTO - OP_MINUPTO;
3645 else if (c == OP_CLASS || c == OP_NCLASS || c == OP_XCLASS)
3647 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
3649 repeat_opcode = code + GET(code, 1);
3652 repeat_opcode = code + 1 + (32 / sizeof(pcre_uchar));
3655 if (c >= OP_CRSTAR && c <= OP_CRMINRANGE)
3657 /* end must not be NULL. */
3658 end = get_chr_property_list(code, utf, cd->fcc, list);
3660 list[1] = (c & 1) == 0;
3662 if (compare_opcodes(end, utf, cd, list, end))
3668 *repeat_opcode = OP_CRPOSSTAR;
3673 *repeat_opcode = OP_CRPOSPLUS;
3678 *repeat_opcode = OP_CRPOSQUERY;
3683 *repeat_opcode = OP_CRPOSRANGE;
3697 case OP_TYPEMINSTAR:
3699 case OP_TYPEMINPLUS:
3701 case OP_TYPEMINQUERY:
3702 case OP_TYPEPOSSTAR:
3703 case OP_TYPEPOSPLUS:
3704 case OP_TYPEPOSQUERY:
3705 if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
3709 case OP_TYPEMINUPTO:
3711 case OP_TYPEPOSUPTO:
3712 if (code[1 + IMM2_SIZE] == OP_PROP || code[1 + IMM2_SIZE] == OP_NOTPROP)
3716 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
3718 code += GET(code, 1);
3730 /* Add in the fixed length from the table */
3732 code += PRIV(OP_lengths)[c];
3734 /* In UTF-8 mode, opcodes that are followed by a character may be followed by
3735 a multi-byte character. The length in the table is a minimum, so we have to
3736 arrange to skip the extra bytes. */
3738 #if defined SUPPORT_UTF && !defined COMPILE_PCRE32
3776 case OP_NOTMINQUERY:
3782 case OP_NOTPOSQUERY:
3785 case OP_NOTMINSTARI:
3787 case OP_NOTMINPLUSI:
3789 case OP_NOTMINQUERYI:
3791 case OP_NOTMINUPTOI:
3793 case OP_NOTPOSSTARI:
3794 case OP_NOTPOSPLUSI:
3795 case OP_NOTPOSQUERYI:
3796 case OP_NOTPOSUPTOI:
3797 if (HAS_EXTRALEN(code[-1])) code += GET_EXTRALEN(code[-1]);
3801 (void)(utf); /* Keep compiler happy by referencing function argument */
3808 /*************************************************
3809 * Check for POSIX class syntax *
3810 *************************************************/
3812 /* This function is called when the sequence "[:" or "[." or "[=" is
3813 encountered in a character class. It checks whether this is followed by a
3814 sequence of characters terminated by a matching ":]" or ".]" or "=]". If we
3815 reach an unescaped ']' without the special preceding character, return FALSE.
3817 Originally, this function only recognized a sequence of letters between the
3818 terminators, but it seems that Perl recognizes any sequence of characters,
3819 though of course unknown POSIX names are subsequently rejected. Perl gives an
3820 "Unknown POSIX class" error for [:f\oo:] for example, where previously PCRE
3821 didn't consider this to be a POSIX class. Likewise for [:1234:].
3823 The problem in trying to be exactly like Perl is in the handling of escapes. We
3824 have to be sure that [abc[:x\]pqr] is *not* treated as containing a POSIX
3825 class, but [abc[:x\]pqr:]] is (so that an error can be generated). The code
3826 below handles the special case of \], but does not try to do any other escape
3827 processing. This makes it different from Perl for cases such as [:l\ower:]
3828 where Perl recognizes it as the POSIX class "lower" but PCRE does not recognize
3829 "l\ower". This is a lesser evil than not diagnosing bad classes when Perl does,
3832 A user pointed out that PCRE was rejecting [:a[:digit:]] whereas Perl was not.
3833 It seems that the appearance of a nested POSIX class supersedes an apparent
3834 external class. For example, [:a[:digit:]b:] matches "a", "b", ":", or
3837 In Perl, unescaped square brackets may also appear as part of class names. For
3838 example, [:a[:abc]b:] gives unknown POSIX class "[:abc]b:]". However, for
3839 [:a[:abc]b][b:] it gives unknown POSIX class "[:abc]b][b:]", which does not
3840 seem right at all. PCRE does not allow closing square brackets in POSIX class
3844 ptr pointer to the initial [
3845 endptr where to return the end pointer
3847 Returns: TRUE or FALSE
3851 check_posix_syntax(const pcre_uchar *ptr, const pcre_uchar **endptr)
3853 pcre_uchar terminator; /* Don't combine these lines; the Solaris cc */
3854 terminator = *(++ptr); /* compiler warns about "non-constant" initializer. */
3855 for (++ptr; *ptr != CHAR_NULL; ptr++)
3857 if (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET)
3859 else if (*ptr == CHAR_RIGHT_SQUARE_BRACKET) return FALSE;
3862 if (*ptr == terminator && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET)
3867 if (*ptr == CHAR_LEFT_SQUARE_BRACKET &&
3868 (ptr[1] == CHAR_COLON || ptr[1] == CHAR_DOT ||
3869 ptr[1] == CHAR_EQUALS_SIGN) &&
3870 check_posix_syntax(ptr, endptr))
3880 /*************************************************
3881 * Check POSIX class name *
3882 *************************************************/
3884 /* This function is called to check the name given in a POSIX-style class entry
3888 ptr points to the first letter
3889 len the length of the name
3891 Returns: a value representing the name, or -1 if unknown
3895 check_posix_name(const pcre_uchar *ptr, int len)
3897 const char *pn = posix_names;
3898 register int yield = 0;
3899 while (posix_name_lengths[yield] != 0)
3901 if (len == posix_name_lengths[yield] &&
3902 STRNCMP_UC_C8(ptr, pn, (unsigned int)len) == 0) return yield;
3903 pn += posix_name_lengths[yield] + 1;
3910 /*************************************************
3911 * Adjust OP_RECURSE items in repeated group *
3912 *************************************************/
3914 /* OP_RECURSE items contain an offset from the start of the regex to the group
3915 that is referenced. This means that groups can be replicated for fixed
3916 repetition simply by copying (because the recursion is allowed to refer to
3917 earlier groups that are outside the current group). However, when a group is
3918 optional (i.e. the minimum quantifier is zero), OP_BRAZERO or OP_SKIPZERO is
3919 inserted before it, after it has been compiled. This means that any OP_RECURSE
3920 items within it that refer to the group itself or any contained groups have to
3921 have their offsets adjusted. That one of the jobs of this function. Before it
3922 is called, the partially compiled regex must be temporarily terminated with
3925 This function has been extended with the possibility of forward references for
3926 recursions and subroutine calls. It must also check the list of such references
3927 for the group we are dealing with. If it finds that one of the recursions in
3928 the current group is on this list, it adjusts the offset in the list, not the
3929 value in the reference (which is a group number).
3932 group points to the start of the group
3933 adjust the amount by which the group is to be moved
3934 utf TRUE in UTF-8 / UTF-16 / UTF-32 mode
3935 cd contains pointers to tables etc.
3936 save_hwm_offset the hwm forward reference offset at the start of the group
3942 adjust_recurse(pcre_uchar *group, int adjust, BOOL utf, compile_data *cd,
3943 size_t save_hwm_offset)
3945 pcre_uchar *ptr = group;
3947 while ((ptr = (pcre_uchar *)find_recurse(ptr, utf)) != NULL)
3952 /* See if this recursion is on the forward reference list. If so, adjust the
3955 for (hc = (pcre_uchar *)cd->start_workspace + save_hwm_offset; hc < cd->hwm;
3958 offset = (int)GET(hc, 0);
3959 if (cd->start_code + offset == ptr + 1)
3961 PUT(hc, 0, offset + adjust);
3966 /* Otherwise, adjust the recursion offset if it's after the start of this
3971 offset = (int)GET(ptr, 1);
3972 if (cd->start_code + offset >= group) PUT(ptr, 1, offset + adjust);
3975 ptr += 1 + LINK_SIZE;
3981 /*************************************************
3982 * Insert an automatic callout point *
3983 *************************************************/
3985 /* This function is called when the PCRE_AUTO_CALLOUT option is set, to insert
3986 callout points before each pattern item.
3989 code current code pointer
3990 ptr current pattern pointer
3991 cd pointers to tables etc
3993 Returns: new code pointer
3997 auto_callout(pcre_uchar *code, const pcre_uchar *ptr, compile_data *cd)
3999 *code++ = OP_CALLOUT;
4001 PUT(code, 0, (int)(ptr - cd->start_pattern)); /* Pattern offset */
4002 PUT(code, LINK_SIZE, 0); /* Default length */
4003 return code + 2 * LINK_SIZE;
4008 /*************************************************
4009 * Complete a callout item *
4010 *************************************************/
4012 /* A callout item contains the length of the next item in the pattern, which
4013 we can't fill in till after we have reached the relevant point. This is used
4014 for both automatic and manual callouts.
4017 previous_callout points to previous callout item
4018 ptr current pattern pointer
4019 cd pointers to tables etc
4025 complete_callout(pcre_uchar *previous_callout, const pcre_uchar *ptr, compile_data *cd)
4027 int length = (int)(ptr - cd->start_pattern - GET(previous_callout, 2));
4028 PUT(previous_callout, 2 + LINK_SIZE, length);
4034 /*************************************************
4035 * Get othercase range *
4036 *************************************************/
4038 /* This function is passed the start and end of a class range, in UTF-8 mode
4039 with UCP support. It searches up the characters, looking for ranges of
4040 characters in the "other" case. Each call returns the next one, updating the
4041 start address. A character with multiple other cases is returned on its own
4042 with a special return value.
4045 cptr points to starting character value; updated
4047 ocptr where to put start of othercase range
4048 odptr where to put end of othercase range
4050 Yield: -1 when no more
4051 0 when a range is returned
4052 >0 the CASESET offset for char with multiple other cases
4053 in this case, ocptr contains the original
4057 get_othercase_range(pcre_uint32 *cptr, pcre_uint32 d, pcre_uint32 *ocptr,
4060 pcre_uint32 c, othercase, next;
4063 /* Find the first character that has an other case. If it has multiple other
4064 cases, return its case offset value. */
4066 for (c = *cptr; c <= d; c++)
4068 if ((co = UCD_CASESET(c)) != 0)
4070 *ocptr = c++; /* Character that has the set */
4071 *cptr = c; /* Rest of input range */
4074 if ((othercase = UCD_OTHERCASE(c)) != c) break;
4077 if (c > d) return -1; /* Reached end of range */
4079 /* Found a character that has a single other case. Search for the end of the
4080 range, which is either the end of the input range, or a character that has zero
4081 or more than one other cases. */
4084 next = othercase + 1;
4086 for (++c; c <= d; c++)
4088 if ((co = UCD_CASESET(c)) != 0 || UCD_OTHERCASE(c) != next) break;
4092 *odptr = next - 1; /* End of othercase range */
4093 *cptr = c; /* Rest of input range */
4096 #endif /* SUPPORT_UCP */
4100 /*************************************************
4101 * Add a character or range to a class *
4102 *************************************************/
4104 /* This function packages up the logic of adding a character or range of
4105 characters to a class. The character values in the arguments will be within the
4106 valid values for the current mode (8-bit, 16-bit, UTF, etc). This function is
4107 mutually recursive with the function immediately below.
4110 classbits the bit map for characters < 256
4111 uchardptr points to the pointer for extra data
4112 options the options word
4113 cd contains pointers to tables etc.
4114 start start of range character
4115 end end of range character
4117 Returns: the number of < 256 characters added
4118 the pointer to extra data is updated
4122 add_to_class(pcre_uint8 *classbits, pcre_uchar **uchardptr, int options,
4123 compile_data *cd, pcre_uint32 start, pcre_uint32 end)
4126 pcre_uint32 classbits_end = (end <= 0xff ? end : 0xff);
4129 /* If caseless matching is required, scan the range and process alternate
4130 cases. In Unicode, there are 8-bit characters that have alternate cases that
4131 are greater than 255 and vice-versa. Sometimes we can just extend the original
4134 if ((options & PCRE_CASELESS) != 0)
4137 if ((options & PCRE_UTF8) != 0)
4142 options &= ~PCRE_CASELESS; /* Remove for recursive calls */
4145 while ((rc = get_othercase_range(&c, end, &oc, &od)) >= 0)
4147 /* Handle a single character that has more than one other case. */
4149 if (rc > 0) n8 += add_list_to_class(classbits, uchardptr, options, cd,
4150 PRIV(ucd_caseless_sets) + rc, oc);
4152 /* Do nothing if the other case range is within the original range. */
4154 else if (oc >= start && od <= end) continue;
4156 /* Extend the original range if there is overlap, noting that if oc < c, we
4157 can't have od > end because a subrange is always shorter than the basic
4158 range. Otherwise, use a recursive call to add the additional range. */
4160 else if (oc < start && od >= start - 1) start = oc; /* Extend downwards */
4161 else if (od > end && oc <= end + 1) end = od; /* Extend upwards */
4162 else n8 += add_to_class(classbits, uchardptr, options, cd, oc, od);
4166 #endif /* SUPPORT_UCP */
4168 /* Not UTF-mode, or no UCP */
4170 for (c = start; c <= classbits_end; c++)
4172 SETBIT(classbits, cd->fcc[c]);
4177 /* Now handle the original range. Adjust the final value according to the bit
4178 length - this means that the same lists of (e.g.) horizontal spaces can be used
4181 #if defined COMPILE_PCRE8
4183 if ((options & PCRE_UTF8) == 0)
4185 if (end > 0xff) end = 0xff;
4187 #elif defined COMPILE_PCRE16
4189 if ((options & PCRE_UTF16) == 0)
4191 if (end > 0xffff) end = 0xffff;
4193 #endif /* COMPILE_PCRE[8|16] */
4195 /* Use the bitmap for characters < 256. Otherwise use extra data.*/
4197 for (c = start; c <= classbits_end; c++)
4199 /* Regardless of start, c will always be <= 255. */
4200 SETBIT(classbits, c);
4204 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
4205 if (start <= 0xff) start = 0xff + 1;
4209 pcre_uchar *uchardata = *uchardptr;
4211 if ((options & PCRE_UTF8) != 0) /* All UTFs use the same flag bit */
4215 *uchardata++ = XCL_RANGE;
4216 uchardata += PRIV(ord2utf)(start, uchardata);
4217 uchardata += PRIV(ord2utf)(end, uchardata);
4219 else if (start == end)
4221 *uchardata++ = XCL_SINGLE;
4222 uchardata += PRIV(ord2utf)(start, uchardata);
4226 #endif /* SUPPORT_UTF */
4228 /* Without UTF support, character values are constrained by the bit length,
4229 and can only be > 256 for 16-bit and 32-bit libraries. */
4231 #ifdef COMPILE_PCRE8
4236 *uchardata++ = XCL_RANGE;
4237 *uchardata++ = start;
4240 else if (start == end)
4242 *uchardata++ = XCL_SINGLE;
4243 *uchardata++ = start;
4247 *uchardptr = uchardata; /* Updata extra data pointer */
4249 #endif /* SUPPORT_UTF || !COMPILE_PCRE8 */
4251 return n8; /* Number of 8-bit characters */
4257 /*************************************************
4258 * Add a list of characters to a class *
4259 *************************************************/
4261 /* This function is used for adding a list of case-equivalent characters to a
4262 class, and also for adding a list of horizontal or vertical whitespace. If the
4263 list is in order (which it should be), ranges of characters are detected and
4264 handled appropriately. This function is mutually recursive with the function
4268 classbits the bit map for characters < 256
4269 uchardptr points to the pointer for extra data
4270 options the options word
4271 cd contains pointers to tables etc.
4272 p points to row of 32-bit values, terminated by NOTACHAR
4273 except character to omit; this is used when adding lists of
4274 case-equivalent characters to avoid including the one we
4277 Returns: the number of < 256 characters added
4278 the pointer to extra data is updated
4282 add_list_to_class(pcre_uint8 *classbits, pcre_uchar **uchardptr, int options,
4283 compile_data *cd, const pcre_uint32 *p, unsigned int except)
4286 while (p[0] < NOTACHAR)
4291 while(p[n+1] == p[0] + n + 1) n++;
4292 n8 += add_to_class(classbits, uchardptr, options, cd, p[0], p[n]);
4301 /*************************************************
4302 * Add characters not in a list to a class *
4303 *************************************************/
4305 /* This function is used for adding the complement of a list of horizontal or
4306 vertical whitespace to a class. The list must be in order.
4309 classbits the bit map for characters < 256
4310 uchardptr points to the pointer for extra data
4311 options the options word
4312 cd contains pointers to tables etc.
4313 p points to row of 32-bit values, terminated by NOTACHAR
4315 Returns: the number of < 256 characters added
4316 the pointer to extra data is updated
4320 add_not_list_to_class(pcre_uint8 *classbits, pcre_uchar **uchardptr,
4321 int options, compile_data *cd, const pcre_uint32 *p)
4323 BOOL utf = (options & PCRE_UTF8) != 0;
4326 n8 += add_to_class(classbits, uchardptr, options, cd, 0, p[0] - 1);
4327 while (p[0] < NOTACHAR)
4329 while (p[1] == p[0] + 1) p++;
4330 n8 += add_to_class(classbits, uchardptr, options, cd, p[0] + 1,
4331 (p[1] == NOTACHAR) ? (utf ? 0x10ffffu : 0xffffffffu) : p[1] - 1);
4339 /*************************************************
4340 * Compile one branch *
4341 *************************************************/
4343 /* Scan the pattern, compiling it into the a vector. If the options are
4344 changed during the branch, the pointer is used to change the external options
4345 bits. This function is used during the pre-compile phase when we are trying
4346 to find out the amount of memory needed, as well as during the real compile
4347 phase. The value of lengthptr distinguishes the two phases.
4350 optionsptr pointer to the option bits
4351 codeptr points to the pointer to the current code point
4352 ptrptr points to the current pattern pointer
4353 errorcodeptr points to error code variable
4354 firstcharptr place to put the first required character
4355 firstcharflagsptr place to put the first character flags, or a negative number
4356 reqcharptr place to put the last required character
4357 reqcharflagsptr place to put the last required character flags, or a negative number
4358 bcptr points to current branch chain
4359 cond_depth conditional nesting depth
4360 cd contains pointers to tables etc.
4361 lengthptr NULL during the real compile phase
4362 points to length accumulator during pre-compile phase
4364 Returns: TRUE on success
4365 FALSE, with *errorcodeptr set non-zero on error
4369 compile_branch(int *optionsptr, pcre_uchar **codeptr,
4370 const pcre_uchar **ptrptr, int *errorcodeptr,
4371 pcre_uint32 *firstcharptr, pcre_int32 *firstcharflagsptr,
4372 pcre_uint32 *reqcharptr, pcre_int32 *reqcharflagsptr,
4373 branch_chain *bcptr, int cond_depth,
4374 compile_data *cd, int *lengthptr)
4376 int repeat_type, op_type;
4377 int repeat_min = 0, repeat_max = 0; /* To please picky compilers */
4379 int greedy_default, greedy_non_default;
4380 pcre_uint32 firstchar, reqchar;
4381 pcre_int32 firstcharflags, reqcharflags;
4382 pcre_uint32 zeroreqchar, zerofirstchar;
4383 pcre_int32 zeroreqcharflags, zerofirstcharflags;
4384 pcre_int32 req_caseopt, reqvary, tempreqvary;
4385 int options = *optionsptr; /* May change dynamically */
4386 int after_manual_callout = 0;
4387 int length_prevgroup = 0;
4388 register pcre_uint32 c;
4390 register pcre_uchar *code = *codeptr;
4391 pcre_uchar *last_code = code;
4392 pcre_uchar *orig_code = code;
4393 pcre_uchar *tempcode;
4394 BOOL inescq = FALSE;
4395 BOOL groupsetfirstchar = FALSE;
4396 const pcre_uchar *ptr = *ptrptr;
4397 const pcre_uchar *tempptr;
4398 const pcre_uchar *nestptr = NULL;
4399 pcre_uchar *previous = NULL;
4400 pcre_uchar *previous_callout = NULL;
4401 size_t save_hwm_offset = 0;
4402 pcre_uint8 classbits[32];
4404 /* We can fish out the UTF-8 setting once and for all into a BOOL, but we
4405 must not do this for other options (e.g. PCRE_EXTENDED) because they may change
4406 dynamically as we process the pattern. */
4409 /* PCRE_UTF[16|32] have the same value as PCRE_UTF8. */
4410 BOOL utf = (options & PCRE_UTF8) != 0;
4411 #ifndef COMPILE_PCRE32
4412 pcre_uchar utf_chars[6];
4418 /* Helper variables for OP_XCLASS opcode (for characters > 255). We define
4419 class_uchardata always so that it can be passed to add_to_class() always,
4420 though it will not be used in non-UTF 8-bit cases. This avoids having to supply
4421 alternative calls for the different cases. */
4423 pcre_uchar *class_uchardata;
4424 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
4426 pcre_uchar *class_uchardata_base;
4430 if (lengthptr != NULL) DPRINTF((">> start branch\n"));
4433 /* Set up the default and non-default settings for greediness */
4435 greedy_default = ((options & PCRE_UNGREEDY) != 0);
4436 greedy_non_default = greedy_default ^ 1;
4438 /* Initialize no first byte, no required byte. REQ_UNSET means "no char
4439 matching encountered yet". It gets changed to REQ_NONE if we hit something that
4440 matches a non-fixed char first char; reqchar just remains unset if we never
4443 When we hit a repeat whose minimum is zero, we may have to adjust these values
4444 to take the zero repeat into account. This is implemented by setting them to
4445 zerofirstbyte and zeroreqchar when such a repeat is encountered. The individual
4446 item types that can be repeated set these backoff variables appropriately. */
4448 firstchar = reqchar = zerofirstchar = zeroreqchar = 0;
4449 firstcharflags = reqcharflags = zerofirstcharflags = zeroreqcharflags = REQ_UNSET;
4451 /* The variable req_caseopt contains either the REQ_CASELESS value
4452 or zero, according to the current setting of the caseless flag. The
4453 REQ_CASELESS leaves the lower 28 bit empty. It is added into the
4454 firstchar or reqchar variables to record the case status of the
4455 value. This is used only for ASCII characters. */
4457 req_caseopt = ((options & PCRE_CASELESS) != 0)? REQ_CASELESS:0;
4459 /* Switch on next character until the end of the branch */
4464 BOOL should_flip_negation;
4465 BOOL possessive_quantifier;
4468 BOOL reset_bracount;
4469 int class_has_8bitchar;
4471 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
4472 BOOL xclass_has_prop;
4478 pcre_uint32 subreqchar, subfirstchar;
4479 pcre_int32 subreqcharflags, subfirstcharflags;
4481 unsigned int mclength;
4482 unsigned int tempbracount;
4484 pcre_uchar mcbuffer[8];
4486 /* Get next character in the pattern */
4490 /* If we are at the end of a nested substitution, revert to the outer level
4491 string. Nesting only happens one level deep. */
4493 if (c == CHAR_NULL && nestptr != NULL)
4500 /* If we are in the pre-compile phase, accumulate the length used for the
4501 previous cycle of this loop. */
4503 if (lengthptr != NULL)
4506 if (code > cd->hwm) cd->hwm = code; /* High water info */
4508 if (code > cd->start_workspace + cd->workspace_size -
4509 WORK_SIZE_SAFETY_MARGIN) /* Check for overrun */
4511 *errorcodeptr = ERR52;
4515 /* There is at least one situation where code goes backwards: this is the
4516 case of a zero quantifier after a class (e.g. [ab]{0}). At compile time,
4517 the class is simply eliminated. However, it is created first, so we have to
4518 allow memory for it. Therefore, don't ever reduce the length at this point.
4521 if (code < last_code) code = last_code;
4523 /* Paranoid check for integer overflow */
4525 if (OFLOW_MAX - *lengthptr < code - last_code)
4527 *errorcodeptr = ERR20;
4531 *lengthptr += (int)(code - last_code);
4532 DPRINTF(("length=%d added %d c=%c (0x%x)\n", *lengthptr,
4533 (int)(code - last_code), c, c));
4535 /* If "previous" is set and it is not at the start of the work space, move
4536 it back to there, in order to avoid filling up the work space. Otherwise,
4537 if "previous" is NULL, reset the current code pointer to the start. */
4539 if (previous != NULL)
4541 if (previous > orig_code)
4543 memmove(orig_code, previous, IN_UCHARS(code - previous));
4544 code -= previous - orig_code;
4545 previous = orig_code;
4548 else code = orig_code;
4550 /* Remember where this code item starts so we can pick up the length
4556 /* In the real compile phase, just check the workspace used by the forward
4559 else if (cd->hwm > cd->start_workspace + cd->workspace_size -
4560 WORK_SIZE_SAFETY_MARGIN)
4562 *errorcodeptr = ERR52;
4566 /* If in \Q...\E, check for the end; if not, we have a literal */
4568 if (inescq && c != CHAR_NULL)
4570 if (c == CHAR_BACKSLASH && ptr[1] == CHAR_E)
4578 if (previous_callout != NULL)
4580 if (lengthptr == NULL) /* Don't attempt in pre-compile phase */
4581 complete_callout(previous_callout, ptr, cd);
4582 previous_callout = NULL;
4584 if ((options & PCRE_AUTO_CALLOUT) != 0)
4586 previous_callout = code;
4587 code = auto_callout(code, ptr, cd);
4591 /* Control does not reach here. */
4594 /* In extended mode, skip white space and comments. We need a loop in order
4595 to check for more white space and more comments after a comment. */
4597 if ((options & PCRE_EXTENDED) != 0)
4601 while (MAX_255(c) && (cd->ctypes[c] & ctype_space) != 0) c = *(++ptr);
4602 if (c != CHAR_NUMBER_SIGN) break;
4604 while (*ptr != CHAR_NULL)
4606 if (IS_NEWLINE(ptr)) /* For non-fixed-length newline cases, */
4607 { /* IS_NEWLINE sets cd->nllen. */
4613 if (utf) FORWARDCHAR(ptr);
4616 c = *ptr; /* Either NULL or the char after a newline */
4620 /* See if the next thing is a quantifier. */
4623 c == CHAR_ASTERISK || c == CHAR_PLUS || c == CHAR_QUESTION_MARK ||
4624 (c == CHAR_LEFT_CURLY_BRACKET && is_counted_repeat(ptr+1));
4626 /* Fill in length of a previous callout, except when the next thing is a
4627 quantifier or when processing a property substitution string in UCP mode. */
4629 if (!is_quantifier && previous_callout != NULL && nestptr == NULL &&
4630 after_manual_callout-- <= 0)
4632 if (lengthptr == NULL) /* Don't attempt in pre-compile phase */
4633 complete_callout(previous_callout, ptr, cd);
4634 previous_callout = NULL;
4637 /* Create auto callout, except for quantifiers, or while processing property
4638 strings that are substituted for \w etc in UCP mode. */
4640 if ((options & PCRE_AUTO_CALLOUT) != 0 && !is_quantifier && nestptr == NULL)
4642 previous_callout = code;
4643 code = auto_callout(code, ptr, cd);
4646 /* Process the next pattern item. */
4650 /* ===================================================================*/
4651 case CHAR_NULL: /* The branch terminates at string end */
4652 case CHAR_VERTICAL_LINE: /* or | or ) */
4653 case CHAR_RIGHT_PARENTHESIS:
4654 *firstcharptr = firstchar;
4655 *firstcharflagsptr = firstcharflags;
4656 *reqcharptr = reqchar;
4657 *reqcharflagsptr = reqcharflags;
4660 if (lengthptr != NULL)
4662 if (OFLOW_MAX - *lengthptr < code - last_code)
4664 *errorcodeptr = ERR20;
4667 *lengthptr += (int)(code - last_code); /* To include callout length */
4668 DPRINTF((">> end branch\n"));
4673 /* ===================================================================*/
4674 /* Handle single-character metacharacters. In multiline mode, ^ disables
4675 the setting of any following char as a first character. */
4677 case CHAR_CIRCUMFLEX_ACCENT:
4679 if ((options & PCRE_MULTILINE) != 0)
4681 if (firstcharflags == REQ_UNSET) firstcharflags = REQ_NONE;
4684 else *code++ = OP_CIRC;
4687 case CHAR_DOLLAR_SIGN:
4689 *code++ = ((options & PCRE_MULTILINE) != 0)? OP_DOLLM : OP_DOLL;
4692 /* There can never be a first char if '.' is first, whatever happens about
4693 repeats. The value of reqchar doesn't change either. */
4696 if (firstcharflags == REQ_UNSET) firstcharflags = REQ_NONE;
4697 zerofirstchar = firstchar;
4698 zerofirstcharflags = firstcharflags;
4699 zeroreqchar = reqchar;
4700 zeroreqcharflags = reqcharflags;
4702 *code++ = ((options & PCRE_DOTALL) != 0)? OP_ALLANY: OP_ANY;
4706 /* ===================================================================*/
4707 /* Character classes. If the included characters are all < 256, we build a
4708 32-byte bitmap of the permitted characters, except in the special case
4709 where there is only one such character. For negated classes, we build the
4710 map as usual, then invert it at the end. However, we use a different opcode
4711 so that data characters > 255 can be handled correctly.
4713 If the class contains characters outside the 0-255 range, a different
4714 opcode is compiled. It may optionally have a bit map for characters < 256,
4715 but those above are are explicitly listed afterwards. A flag byte tells
4716 whether the bitmap is present, and whether this is a negated class or not.
4718 In JavaScript compatibility mode, an isolated ']' causes an error. In
4719 default (Perl) mode, it is treated as a data character. */
4721 case CHAR_RIGHT_SQUARE_BRACKET:
4722 if ((cd->external_options & PCRE_JAVASCRIPT_COMPAT) != 0)
4724 *errorcodeptr = ERR64;
4729 /* In another (POSIX) regex library, the ugly syntax [[:<:]] and [[:>:]] is
4730 used for "start of word" and "end of word". As these are otherwise illegal
4731 sequences, we don't break anything by recognizing them. They are replaced
4732 by \b(?=\w) and \b(?<=\w) respectively. Sequences like [a[:<:]] are
4733 erroneous and are handled by the normal code below. */
4735 case CHAR_LEFT_SQUARE_BRACKET:
4736 if (STRNCMP_UC_C8(ptr+1, STRING_WEIRD_STARTWORD, 6) == 0)
4739 ptr = sub_start_of_word - 1;
4743 if (STRNCMP_UC_C8(ptr+1, STRING_WEIRD_ENDWORD, 6) == 0)
4746 ptr = sub_end_of_word - 1;
4750 /* Handle a real character class. */
4754 /* PCRE supports POSIX class stuff inside a class. Perl gives an error if
4755 they are encountered at the top level, so we'll do that too. */
4757 if ((ptr[1] == CHAR_COLON || ptr[1] == CHAR_DOT ||
4758 ptr[1] == CHAR_EQUALS_SIGN) &&
4759 check_posix_syntax(ptr, &tempptr))
4761 *errorcodeptr = (ptr[1] == CHAR_COLON)? ERR13 : ERR31;
4765 /* If the first character is '^', set the negation flag and skip it. Also,
4766 if the first few characters (either before or after ^) are \Q\E or \E we
4767 skip them too. This makes for compatibility with Perl. */
4769 negate_class = FALSE;
4773 if (c == CHAR_BACKSLASH)
4775 if (ptr[1] == CHAR_E)
4777 else if (STRNCMP_UC_C8(ptr + 1, STR_Q STR_BACKSLASH STR_E, 3) == 0)
4782 else if (!negate_class && c == CHAR_CIRCUMFLEX_ACCENT)
4783 negate_class = TRUE;
4787 /* Empty classes are allowed in JavaScript compatibility mode. Otherwise,
4788 an initial ']' is taken as a data character -- the code below handles
4789 that. In JS mode, [] must always fail, so generate OP_FAIL, whereas
4790 [^] must match any character, so generate OP_ALLANY. */
4792 if (c == CHAR_RIGHT_SQUARE_BRACKET &&
4793 (cd->external_options & PCRE_JAVASCRIPT_COMPAT) != 0)
4795 *code++ = negate_class? OP_ALLANY : OP_FAIL;
4796 if (firstcharflags == REQ_UNSET) firstcharflags = REQ_NONE;
4797 zerofirstchar = firstchar;
4798 zerofirstcharflags = firstcharflags;
4802 /* If a class contains a negative special such as \S, we need to flip the
4803 negation flag at the end, so that support for characters > 255 works
4804 correctly (they are all included in the class). */
4806 should_flip_negation = FALSE;
4808 /* Extended class (xclass) will be used when characters > 255
4811 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
4813 class_uchardata = code + LINK_SIZE + 2; /* For XCLASS items */
4814 class_uchardata_base = class_uchardata; /* Save the start */
4817 /* For optimization purposes, we track some properties of the class:
4818 class_has_8bitchar will be non-zero if the class contains at least one <
4819 256 character; class_one_char will be 1 if the class contains just one
4820 character; xclass_has_prop will be TRUE if unicode property checks
4821 are present in the class. */
4823 class_has_8bitchar = 0;
4825 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
4826 xclass_has_prop = FALSE;
4829 /* Initialize the 32-char bit map to all zeros. We build the map in a
4830 temporary bit of memory, in case the class contains fewer than two
4831 8-bit characters because in that case the compiled code doesn't use the bit
4834 memset(classbits, 0, 32 * sizeof(pcre_uint8));
4836 /* Process characters until ] is reached. By writing this as a "do" it
4837 means that an initial ] is taken as a data character. At the start of the
4838 loop, c contains the first byte of the character. */
4840 if (c != CHAR_NULL) do
4842 const pcre_uchar *oldptr;
4845 if (utf && HAS_EXTRALEN(c))
4846 { /* Braces are required because the */
4847 GETCHARLEN(c, ptr, ptr); /* macro generates multiple statements */
4851 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
4852 /* In the pre-compile phase, accumulate the length of any extra
4853 data and reset the pointer. This is so that very large classes that
4854 contain a zillion > 255 characters no longer overwrite the work space
4855 (which is on the stack). We have to remember that there was XCLASS data,
4858 if (lengthptr != NULL && class_uchardata > class_uchardata_base)
4861 *lengthptr += class_uchardata - class_uchardata_base;
4862 class_uchardata = class_uchardata_base;
4866 /* Inside \Q...\E everything is literal except \E */
4870 if (c == CHAR_BACKSLASH && ptr[1] == CHAR_E) /* If we are at \E */
4872 inescq = FALSE; /* Reset literal state */
4873 ptr++; /* Skip the 'E' */
4874 continue; /* Carry on with next */
4876 goto CHECK_RANGE; /* Could be range if \E follows */
4879 /* Handle POSIX class names. Perl allows a negation extension of the
4880 form [:^name:]. A square bracket that doesn't match the syntax is
4881 treated as a literal. We also recognize the POSIX constructions
4882 [.ch.] and [=ch=] ("collating elements") and fault them, as Perl
4885 if (c == CHAR_LEFT_SQUARE_BRACKET &&
4886 (ptr[1] == CHAR_COLON || ptr[1] == CHAR_DOT ||
4887 ptr[1] == CHAR_EQUALS_SIGN) && check_posix_syntax(ptr, &tempptr))
4889 BOOL local_negate = FALSE;
4890 int posix_class, taboffset, tabopt;
4891 register const pcre_uint8 *cbits = cd->cbits;
4892 pcre_uint8 pbits[32];
4894 if (ptr[1] != CHAR_COLON)
4896 *errorcodeptr = ERR31;
4901 if (*ptr == CHAR_CIRCUMFLEX_ACCENT)
4903 local_negate = TRUE;
4904 should_flip_negation = TRUE; /* Note negative special */
4908 posix_class = check_posix_name(ptr, (int)(tempptr - ptr));
4909 if (posix_class < 0)
4911 *errorcodeptr = ERR30;
4915 /* If matching is caseless, upper and lower are converted to
4916 alpha. This relies on the fact that the class table starts with
4917 alpha, lower, upper as the first 3 entries. */
4919 if ((options & PCRE_CASELESS) != 0 && posix_class <= 2)
4922 /* When PCRE_UCP is set, some of the POSIX classes are converted to
4923 different escape sequences that use Unicode properties \p or \P. Others
4924 that are not available via \p or \P generate XCL_PROP/XCL_NOTPROP
4928 if ((options & PCRE_UCP) != 0)
4930 unsigned int ptype = 0;
4931 int pc = posix_class + ((local_negate)? POSIX_SUBSIZE/2 : 0);
4933 /* The posix_substitutes table specifies which POSIX classes can be
4934 converted to \p or \P items. */
4936 if (posix_substitutes[pc] != NULL)
4938 nestptr = tempptr + 1;
4939 ptr = posix_substitutes[pc] - 1;
4943 /* There are three other classes that generate special property calls
4944 that are recognized only in an XCLASS. */
4946 else switch(posix_class)
4952 if (ptype == 0) ptype = PT_PXPRINT;
4955 if (ptype == 0) ptype = PT_PXPUNCT;
4956 *class_uchardata++ = local_negate? XCL_NOTPROP : XCL_PROP;
4957 *class_uchardata++ = ptype;
4958 *class_uchardata++ = 0;
4959 xclass_has_prop = TRUE;
4963 /* For all other POSIX classes, no special action is taken in UCP
4964 mode. Fall through to the non_UCP case. */
4971 /* In the non-UCP case, or when UCP makes no difference, we build the
4972 bit map for the POSIX class in a chunk of local store because we may be
4973 adding and subtracting from it, and we don't want to subtract bits that
4974 may be in the main map already. At the end we or the result into the
4975 bit map that is being built. */
4979 /* Copy in the first table (always present) */
4981 memcpy(pbits, cbits + posix_class_maps[posix_class],
4982 32 * sizeof(pcre_uint8));
4984 /* If there is a second table, add or remove it as required. */
4986 taboffset = posix_class_maps[posix_class + 1];
4987 tabopt = posix_class_maps[posix_class + 2];
4992 for (c = 0; c < 32; c++) pbits[c] |= cbits[c + taboffset];
4994 for (c = 0; c < 32; c++) pbits[c] &= ~cbits[c + taboffset];
4997 /* Now see if we need to remove any special characters. An option
4998 value of 1 removes vertical space and 2 removes underscore. */
5000 if (tabopt < 0) tabopt = -tabopt;
5001 if (tabopt == 1) pbits[1] &= ~0x3c;
5002 else if (tabopt == 2) pbits[11] &= 0x7f;
5004 /* Add the POSIX table or its complement into the main table that is
5005 being built and we are done. */
5008 for (c = 0; c < 32; c++) classbits[c] |= ~pbits[c];
5010 for (c = 0; c < 32; c++) classbits[c] |= pbits[c];
5013 /* Every class contains at least one < 256 character. */
5014 class_has_8bitchar = 1;
5015 /* Every class contains at least two characters. */
5017 continue; /* End of POSIX syntax handling */
5020 /* Backslash may introduce a single character, or it may introduce one
5021 of the specials, which just set a flag. The sequence \b is a special
5022 case. Inside a class (and only there) it is treated as backspace. We
5023 assume that other escapes have more than one character in them, so
5024 speculatively set both class_has_8bitchar and class_one_char bigger
5025 than one. Unrecognized escapes fall through and are either treated
5026 as literal characters (by default), or are faulted if
5027 PCRE_EXTRA is set. */
5029 if (c == CHAR_BACKSLASH)
5031 escape = check_escape(&ptr, &ec, errorcodeptr, cd->bracount, options,
5033 if (*errorcodeptr != 0) goto FAILED;
5034 if (escape == 0) c = ec;
5035 else if (escape == ESC_b) c = CHAR_BS; /* \b is backspace in a class */
5036 else if (escape == ESC_N) /* \N is not supported in a class */
5038 *errorcodeptr = ERR71;
5041 else if (escape == ESC_Q) /* Handle start of quoted string */
5043 if (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
5045 ptr += 2; /* avoid empty string */
5050 else if (escape == ESC_E) continue; /* Ignore orphan \E */
5054 register const pcre_uint8 *cbits = cd->cbits;
5055 /* Every class contains at least two < 256 characters. */
5056 class_has_8bitchar++;
5057 /* Every class contains at least two characters. */
5058 class_one_char += 2;
5063 case ESC_du: /* These are the values given for \d etc */
5064 case ESC_DU: /* when PCRE_UCP is set. We replace the */
5065 case ESC_wu: /* escape sequence with an appropriate \p */
5066 case ESC_WU: /* or \P to test Unicode properties instead */
5067 case ESC_su: /* of the default ASCII testing. */
5070 ptr = substitutes[escape - ESC_DU] - 1; /* Just before substitute */
5071 class_has_8bitchar--; /* Undo! */
5075 for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_digit];
5079 should_flip_negation = TRUE;
5080 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_digit];
5084 for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_word];
5088 should_flip_negation = TRUE;
5089 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_word];
5092 /* Perl 5.004 onwards omitted VT from \s, but restored it at Perl
5093 5.18. Before PCRE 8.34, we had to preserve the VT bit if it was
5094 previously set by something earlier in the character class.
5095 Luckily, the value of CHAR_VT is 0x0b in both ASCII and EBCDIC, so
5096 we could just adjust the appropriate bit. From PCRE 8.34 we no
5097 longer treat \s and \S specially. */
5100 for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_space];
5104 should_flip_negation = TRUE;
5105 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_space];
5108 /* The rest apply in both UCP and non-UCP cases. */
5111 (void)add_list_to_class(classbits, &class_uchardata, options, cd,
5112 PRIV(hspace_list), NOTACHAR);
5116 (void)add_not_list_to_class(classbits, &class_uchardata, options,
5117 cd, PRIV(hspace_list));
5121 (void)add_list_to_class(classbits, &class_uchardata, options, cd,
5122 PRIV(vspace_list), NOTACHAR);
5126 (void)add_not_list_to_class(classbits, &class_uchardata, options,
5127 cd, PRIV(vspace_list));
5135 unsigned int ptype = 0, pdata = 0;
5136 if (!get_ucp(&ptr, &negated, &ptype, &pdata, errorcodeptr))
5138 *class_uchardata++ = ((escape == ESC_p) != negated)?
5139 XCL_PROP : XCL_NOTPROP;
5140 *class_uchardata++ = ptype;
5141 *class_uchardata++ = pdata;
5142 xclass_has_prop = TRUE;
5143 class_has_8bitchar--; /* Undo! */
5147 /* Unrecognized escapes are faulted if PCRE is running in its
5148 strict mode. By default, for compatibility with Perl, they are
5149 treated as literals. */
5152 if ((options & PCRE_EXTRA) != 0)
5154 *errorcodeptr = ERR7;
5157 class_has_8bitchar--; /* Undo the speculative increase. */
5158 class_one_char -= 2; /* Undo the speculative increase. */
5159 c = *ptr; /* Get the final character and fall through */
5164 /* Fall through if the escape just defined a single character (c >= 0).
5165 This may be greater than 256. */
5169 } /* End of backslash handling */
5171 /* A character may be followed by '-' to form a range. However, Perl does
5172 not permit ']' to be the end of the range. A '-' character at the end is
5173 treated as a literal. Perl ignores orphaned \E sequences entirely. The
5174 code for handling \Q and \E is messy. */
5177 while (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
5184 /* Remember if \r or \n were explicitly used */
5186 if (c == CHAR_CR || c == CHAR_NL) cd->external_flags |= PCRE_HASCRORLF;
5188 /* Check for range */
5190 if (!inescq && ptr[1] == CHAR_MINUS)
5194 while (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_E) ptr += 2;
5196 /* If we hit \Q (not followed by \E) at this point, go into escaped
5199 while (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_Q)
5202 if (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_E)
5203 { ptr += 2; continue; }
5208 /* Minus (hyphen) at the end of a class is treated as a literal, so put
5209 back the pointer and jump to handle the character that preceded it. */
5211 if (*ptr == CHAR_NULL || (!inescq && *ptr == CHAR_RIGHT_SQUARE_BRACKET))
5214 goto CLASS_SINGLE_CHARACTER;
5217 /* Otherwise, we have a potential range; pick up the next character */
5221 { /* Braces are required because the */
5222 GETCHARLEN(d, ptr, ptr); /* macro generates multiple statements */
5226 d = *ptr; /* Not UTF-8 mode */
5228 /* The second part of a range can be a single-character escape
5229 sequence, but not any of the other escapes. Perl treats a hyphen as a
5230 literal in such circumstances. However, in Perl's warning mode, a
5231 warning is given, so PCRE now faults it as it is almost certainly a
5232 mistake on the user's part. */
5236 if (d == CHAR_BACKSLASH)
5239 descape = check_escape(&ptr, &d, errorcodeptr, cd->bracount, options, TRUE);
5240 if (*errorcodeptr != 0) goto FAILED;
5242 /* 0 means a character was put into d; \b is backspace; any other
5243 special causes an error. */
5247 if (descape == ESC_b) d = CHAR_BS; else
5249 *errorcodeptr = ERR83;
5255 /* A hyphen followed by a POSIX class is treated in the same way. */
5257 else if (d == CHAR_LEFT_SQUARE_BRACKET &&
5258 (ptr[1] == CHAR_COLON || ptr[1] == CHAR_DOT ||
5259 ptr[1] == CHAR_EQUALS_SIGN) &&
5260 check_posix_syntax(ptr, &tempptr))
5262 *errorcodeptr = ERR83;
5267 /* Check that the two values are in the correct order. Optimize
5268 one-character ranges. */
5272 *errorcodeptr = ERR8;
5275 if (d == c) goto CLASS_SINGLE_CHARACTER; /* A few lines below */
5277 /* We have found a character range, so single character optimizations
5278 cannot be done anymore. Any value greater than 1 indicates that there
5279 is more than one character. */
5283 /* Remember an explicit \r or \n, and add the range to the class. */
5285 if (d == CHAR_CR || d == CHAR_NL) cd->external_flags |= PCRE_HASCRORLF;
5287 class_has_8bitchar +=
5288 add_to_class(classbits, &class_uchardata, options, cd, c, d);
5290 continue; /* Go get the next char in the class */
5293 /* Handle a single character - we can get here for a normal non-escape
5294 char, or after \ that introduces a single character or for an apparent
5295 range that isn't. Only the value 1 matters for class_one_char, so don't
5296 increase it if it is already 2 or more ... just in case there's a class
5297 with a zillion characters in it. */
5299 CLASS_SINGLE_CHARACTER:
5300 if (class_one_char < 2) class_one_char++;
5302 /* If class_one_char is 1, we have the first single character in the
5303 class, and there have been no prior ranges, or XCLASS items generated by
5304 escapes. If this is the final character in the class, we can optimize by
5305 turning the item into a 1-character OP_CHAR[I] if it's positive, or
5306 OP_NOT[I] if it's negative. In the positive case, it can cause firstchar
5307 to be set. Otherwise, there can be no first char if this item is first,
5308 whatever repeat count may follow. In the case of reqchar, save the
5309 previous value for reinstating. */
5311 if (class_one_char == 1 && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET)
5314 zeroreqchar = reqchar;
5315 zeroreqcharflags = reqcharflags;
5322 if (firstcharflags == REQ_UNSET) firstcharflags = REQ_NONE;
5323 zerofirstchar = firstchar;
5324 zerofirstcharflags = firstcharflags;
5326 /* For caseless UTF-8 mode when UCP support is available, check
5327 whether this character has more than one other case. If so, generate
5328 a special OP_NOTPROP item instead of OP_NOTI. */
5331 if (utf && (options & PCRE_CASELESS) != 0 &&
5332 (d = UCD_CASESET(c)) != 0)
5334 *code++ = OP_NOTPROP;
5340 /* Char has only one other case, or UCP not available */
5343 *code++ = ((options & PCRE_CASELESS) != 0)? OP_NOTI: OP_NOT;
5344 #if defined SUPPORT_UTF && !defined COMPILE_PCRE32
5345 if (utf && c > MAX_VALUE_FOR_SINGLE_CHAR)
5346 code += PRIV(ord2utf)(c, code);
5352 /* We are finished with this character class */
5357 /* For a single, positive character, get the value into mcbuffer, and
5358 then we can handle this with the normal one-character code. */
5360 #if defined SUPPORT_UTF && !defined COMPILE_PCRE32
5361 if (utf && c > MAX_VALUE_FOR_SINGLE_CHAR)
5362 mclength = PRIV(ord2utf)(c, mcbuffer);
5370 } /* End of 1-char optimization */
5372 /* There is more than one character in the class, or an XCLASS item
5373 has been generated. Add this character to the class. */
5375 class_has_8bitchar +=
5376 add_to_class(classbits, &class_uchardata, options, cd, c, c);
5379 /* Loop until ']' reached. This "while" is the end of the "do" far above.
5380 If we are at the end of an internal nested string, revert to the outer
5383 while (((c = *(++ptr)) != CHAR_NULL ||
5385 (ptr = nestptr, nestptr = NULL, c = *(++ptr)) != CHAR_NULL)) &&
5386 (c != CHAR_RIGHT_SQUARE_BRACKET || inescq));
5388 /* Check for missing terminating ']' */
5392 *errorcodeptr = ERR6;
5396 /* We will need an XCLASS if data has been placed in class_uchardata. In
5397 the second phase this is a sufficient test. However, in the pre-compile
5398 phase, class_uchardata gets emptied to prevent workspace overflow, so it
5399 only if the very last character in the class needs XCLASS will it contain
5400 anything at this point. For this reason, xclass gets set TRUE above when
5401 uchar_classdata is emptied, and that's why this code is the way it is here
5402 instead of just doing a test on class_uchardata below. */
5404 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
5405 if (class_uchardata > class_uchardata_base) xclass = TRUE;
5408 /* If this is the first thing in the branch, there can be no first char
5409 setting, whatever the repeat count. Any reqchar setting must remain
5410 unchanged after any kind of repeat. */
5412 if (firstcharflags == REQ_UNSET) firstcharflags = REQ_NONE;
5413 zerofirstchar = firstchar;
5414 zerofirstcharflags = firstcharflags;
5415 zeroreqchar = reqchar;
5416 zeroreqcharflags = reqcharflags;
5418 /* If there are characters with values > 255, we have to compile an
5419 extended class, with its own opcode, unless there was a negated special
5420 such as \S in the class, and PCRE_UCP is not set, because in that case all
5421 characters > 255 are in the class, so any that were explicitly given as
5422 well can be ignored. If (when there are explicit characters > 255 that must
5423 be listed) there are no characters < 256, we can omit the bitmap in the
5424 actual compiled code. */
5427 if (xclass && (!should_flip_negation || (options & PCRE_UCP) != 0))
5428 #elif !defined COMPILE_PCRE8
5429 if (xclass && !should_flip_negation)
5431 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
5433 *class_uchardata++ = XCL_END; /* Marks the end of extra data */
5434 *code++ = OP_XCLASS;
5436 *code = negate_class? XCL_NOT:0;
5437 if (xclass_has_prop) *code |= XCL_HASPROP;
5439 /* If the map is required, move up the extra data to make room for it;
5440 otherwise just move the code pointer to the end of the extra data. */
5442 if (class_has_8bitchar > 0)
5445 memmove(code + (32 / sizeof(pcre_uchar)), code,
5446 IN_UCHARS(class_uchardata - code));
5447 if (negate_class && !xclass_has_prop)
5448 for (c = 0; c < 32; c++) classbits[c] = ~classbits[c];
5449 memcpy(code, classbits, 32);
5450 code = class_uchardata + (32 / sizeof(pcre_uchar));
5452 else code = class_uchardata;
5454 /* Now fill in the complete length of the item */
5456 PUT(previous, 1, (int)(code - previous));
5457 break; /* End of class handling */
5461 /* If there are no characters > 255, or they are all to be included or
5462 excluded, set the opcode to OP_CLASS or OP_NCLASS, depending on whether the
5463 whole class was negated and whether there were negative specials such as \S
5464 (non-UCP) in the class. Then copy the 32-byte map into the code vector,
5465 negating it if necessary. */
5467 *code++ = (negate_class == should_flip_negation) ? OP_CLASS : OP_NCLASS;
5468 if (lengthptr == NULL) /* Save time in the pre-compile phase */
5471 for (c = 0; c < 32; c++) classbits[c] = ~classbits[c];
5472 memcpy(code, classbits, 32);
5474 code += 32 / sizeof(pcre_uchar);
5480 /* ===================================================================*/
5481 /* Various kinds of repeat; '{' is not necessarily a quantifier, but this
5482 has been tested above. */
5484 case CHAR_LEFT_CURLY_BRACKET:
5485 if (!is_quantifier) goto NORMAL_CHAR;
5486 ptr = read_repeat_counts(ptr+1, &repeat_min, &repeat_max, errorcodeptr);
5487 if (*errorcodeptr != 0) goto FAILED;
5500 case CHAR_QUESTION_MARK:
5505 if (previous == NULL)
5507 *errorcodeptr = ERR9;
5511 if (repeat_min == 0)
5513 firstchar = zerofirstchar; /* Adjust for zero repeat */
5514 firstcharflags = zerofirstcharflags;
5515 reqchar = zeroreqchar; /* Ditto */
5516 reqcharflags = zeroreqcharflags;
5519 /* Remember whether this is a variable length repeat */
5521 reqvary = (repeat_min == repeat_max)? 0 : REQ_VARY;
5523 op_type = 0; /* Default single-char op codes */
5524 possessive_quantifier = FALSE; /* Default not possessive quantifier */
5526 /* Save start of previous item, in case we have to move it up in order to
5527 insert something before it. */
5529 tempcode = previous;
5531 /* Before checking for a possessive quantifier, we must skip over
5532 whitespace and comments in extended mode because Perl allows white space at
5535 if ((options & PCRE_EXTENDED) != 0)
5537 const pcre_uchar *p = ptr + 1;
5540 while (MAX_255(*p) && (cd->ctypes[*p] & ctype_space) != 0) p++;
5541 if (*p != CHAR_NUMBER_SIGN) break;
5543 while (*p != CHAR_NULL)
5545 if (IS_NEWLINE(p)) /* For non-fixed-length newline cases, */
5546 { /* IS_NEWLINE sets cd->nllen. */
5552 if (utf) FORWARDCHAR(p);
5554 } /* Loop for comment characters */
5555 } /* Loop for multiple comments */
5556 ptr = p - 1; /* Character before the next significant one. */
5559 /* If the next character is '+', we have a possessive quantifier. This
5560 implies greediness, whatever the setting of the PCRE_UNGREEDY option.
5561 If the next character is '?' this is a minimizing repeat, by default,
5562 but if PCRE_UNGREEDY is set, it works the other way round. We change the
5563 repeat type to the non-default. */
5565 if (ptr[1] == CHAR_PLUS)
5567 repeat_type = 0; /* Force greedy */
5568 possessive_quantifier = TRUE;
5571 else if (ptr[1] == CHAR_QUESTION_MARK)
5573 repeat_type = greedy_non_default;
5576 else repeat_type = greedy_default;
5578 /* If previous was a recursion call, wrap it in atomic brackets so that
5579 previous becomes the atomic group. All recursions were so wrapped in the
5580 past, but it no longer happens for non-repeated recursions. In fact, the
5581 repeated ones could be re-implemented independently so as not to need this,
5582 but for the moment we rely on the code for repeating groups. */
5584 if (*previous == OP_RECURSE)
5586 memmove(previous + 1 + LINK_SIZE, previous, IN_UCHARS(1 + LINK_SIZE));
5587 *previous = OP_ONCE;
5588 PUT(previous, 1, 2 + 2*LINK_SIZE);
5589 previous[2 + 2*LINK_SIZE] = OP_KET;
5590 PUT(previous, 3 + 2*LINK_SIZE, 2 + 2*LINK_SIZE);
5591 code += 2 + 2 * LINK_SIZE;
5592 length_prevgroup = 3 + 3*LINK_SIZE;
5594 /* When actually compiling, we need to check whether this was a forward
5595 reference, and if so, adjust the offset. */
5597 if (lengthptr == NULL && cd->hwm >= cd->start_workspace + LINK_SIZE)
5599 int offset = GET(cd->hwm, -LINK_SIZE);
5600 if (offset == previous + 1 - cd->start_code)
5601 PUT(cd->hwm, -LINK_SIZE, offset + 1 + LINK_SIZE);
5605 /* Now handle repetition for the different types of item. */
5607 /* If previous was a character or negated character match, abolish the item
5608 and generate a repeat item instead. If a char item has a minimum of more
5609 than one, ensure that it is set in reqchar - it might not be if a sequence
5610 such as x{3} is the first thing in a branch because the x will have gone
5611 into firstchar instead. */
5613 if (*previous == OP_CHAR || *previous == OP_CHARI
5614 || *previous == OP_NOT || *previous == OP_NOTI)
5618 default: /* Make compiler happy. */
5619 case OP_CHAR: op_type = OP_STAR - OP_STAR; break;
5620 case OP_CHARI: op_type = OP_STARI - OP_STAR; break;
5621 case OP_NOT: op_type = OP_NOTSTAR - OP_STAR; break;
5622 case OP_NOTI: op_type = OP_NOTSTARI - OP_STAR; break;
5625 /* Deal with UTF characters that take up more than one character. It's
5626 easier to write this out separately than try to macrify it. Use c to
5627 hold the length of the character in bytes, plus UTF_LENGTH to flag that
5628 it's a length rather than a small character. */
5630 #if defined SUPPORT_UTF && !defined COMPILE_PCRE32
5631 if (utf && NOT_FIRSTCHAR(code[-1]))
5633 pcre_uchar *lastchar = code - 1;
5635 c = (int)(code - lastchar); /* Length of UTF-8 character */
5636 memcpy(utf_chars, lastchar, IN_UCHARS(c)); /* Save the char */
5637 c |= UTF_LENGTH; /* Flag c as a length */
5640 #endif /* SUPPORT_UTF */
5642 /* Handle the case of a single charater - either with no UTF support, or
5643 with UTF disabled, or for a single character UTF character. */
5646 if (*previous <= OP_CHARI && repeat_min > 1)
5649 reqcharflags = req_caseopt | cd->req_varyopt;
5653 goto OUTPUT_SINGLE_REPEAT; /* Code shared with single character types */
5656 /* If previous was a character type match (\d or similar), abolish it and
5657 create a suitable repeat item. The code is shared with single-character
5658 repeats by setting op_type to add a suitable offset into repeat_type. Note
5659 the the Unicode property types will be present only when SUPPORT_UCP is
5660 defined, but we don't wrap the little bits of code here because it just
5661 makes it horribly messy. */
5663 else if (*previous < OP_EODN)
5665 pcre_uchar *oldcode;
5666 int prop_type, prop_value;
5667 op_type = OP_TYPESTAR - OP_STAR; /* Use type opcodes */
5670 OUTPUT_SINGLE_REPEAT:
5671 if (*previous == OP_PROP || *previous == OP_NOTPROP)
5673 prop_type = previous[1];
5674 prop_value = previous[2];
5676 else prop_type = prop_value = -1;
5679 code = previous; /* Usually overwrite previous item */
5681 /* If the maximum is zero then the minimum must also be zero; Perl allows
5682 this case, so we do too - by simply omitting the item altogether. */
5684 if (repeat_max == 0) goto END_REPEAT;
5686 /* Combine the op_type with the repeat_type */
5688 repeat_type += op_type;
5690 /* A minimum of zero is handled either as the special case * or ?, or as
5691 an UPTO, with the maximum given. */
5693 if (repeat_min == 0)
5695 if (repeat_max == -1) *code++ = OP_STAR + repeat_type;
5696 else if (repeat_max == 1) *code++ = OP_QUERY + repeat_type;
5699 *code++ = OP_UPTO + repeat_type;
5700 PUT2INC(code, 0, repeat_max);
5704 /* A repeat minimum of 1 is optimized into some special cases. If the
5705 maximum is unlimited, we use OP_PLUS. Otherwise, the original item is
5706 left in place and, if the maximum is greater than 1, we use OP_UPTO with
5707 one less than the maximum. */
5709 else if (repeat_min == 1)
5711 if (repeat_max == -1)
5712 *code++ = OP_PLUS + repeat_type;
5715 code = oldcode; /* leave previous item in place */
5716 if (repeat_max == 1) goto END_REPEAT;
5717 *code++ = OP_UPTO + repeat_type;
5718 PUT2INC(code, 0, repeat_max - 1);
5722 /* The case {n,n} is just an EXACT, while the general case {n,m} is
5723 handled as an EXACT followed by an UPTO. */
5727 *code++ = OP_EXACT + op_type; /* NB EXACT doesn't have repeat_type */
5728 PUT2INC(code, 0, repeat_min);
5730 /* If the maximum is unlimited, insert an OP_STAR. Before doing so,
5731 we have to insert the character for the previous code. For a repeated
5732 Unicode property match, there are two extra bytes that define the
5733 required property. In UTF-8 mode, long characters have their length in
5734 c, with the UTF_LENGTH bit as a flag. */
5738 #if defined SUPPORT_UTF && !defined COMPILE_PCRE32
5739 if (utf && (c & UTF_LENGTH) != 0)
5741 memcpy(code, utf_chars, IN_UCHARS(c & 7));
5750 *code++ = prop_type;
5751 *code++ = prop_value;
5754 *code++ = OP_STAR + repeat_type;
5757 /* Else insert an UPTO if the max is greater than the min, again
5758 preceded by the character, for the previously inserted code. If the
5759 UPTO is just for 1 instance, we can use QUERY instead. */
5761 else if (repeat_max != repeat_min)
5763 #if defined SUPPORT_UTF && !defined COMPILE_PCRE32
5764 if (utf && (c & UTF_LENGTH) != 0)
5766 memcpy(code, utf_chars, IN_UCHARS(c & 7));
5774 *code++ = prop_type;
5775 *code++ = prop_value;
5777 repeat_max -= repeat_min;
5779 if (repeat_max == 1)
5781 *code++ = OP_QUERY + repeat_type;
5785 *code++ = OP_UPTO + repeat_type;
5786 PUT2INC(code, 0, repeat_max);
5791 /* The character or character type itself comes last in all cases. */
5793 #if defined SUPPORT_UTF && !defined COMPILE_PCRE32
5794 if (utf && (c & UTF_LENGTH) != 0)
5796 memcpy(code, utf_chars, IN_UCHARS(c & 7));
5803 /* For a repeated Unicode property match, there are two extra bytes that
5804 define the required property. */
5809 *code++ = prop_type;
5810 *code++ = prop_value;
5815 /* If previous was a character class or a back reference, we put the repeat
5816 stuff after it, but just skip the item if the repeat was {0,0}. */
5818 else if (*previous == OP_CLASS || *previous == OP_NCLASS ||
5819 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
5820 *previous == OP_XCLASS ||
5822 *previous == OP_REF || *previous == OP_REFI ||
5823 *previous == OP_DNREF || *previous == OP_DNREFI)
5825 if (repeat_max == 0)
5831 if (repeat_min == 0 && repeat_max == -1)
5832 *code++ = OP_CRSTAR + repeat_type;
5833 else if (repeat_min == 1 && repeat_max == -1)
5834 *code++ = OP_CRPLUS + repeat_type;
5835 else if (repeat_min == 0 && repeat_max == 1)
5836 *code++ = OP_CRQUERY + repeat_type;
5839 *code++ = OP_CRRANGE + repeat_type;
5840 PUT2INC(code, 0, repeat_min);
5841 if (repeat_max == -1) repeat_max = 0; /* 2-byte encoding for max */
5842 PUT2INC(code, 0, repeat_max);
5846 /* If previous was a bracket group, we may have to replicate it in certain
5847 cases. Note that at this point we can encounter only the "basic" bracket
5848 opcodes such as BRA and CBRA, as this is the place where they get converted
5849 into the more special varieties such as BRAPOS and SBRA. A test for >=
5850 OP_ASSERT and <= OP_COND includes ASSERT, ASSERT_NOT, ASSERTBACK,
5851 ASSERTBACK_NOT, ONCE, ONCE_NC, BRA, BRAPOS, CBRA, CBRAPOS, and COND.
5852 Originally, PCRE did not allow repetition of assertions, but now it does,
5853 for Perl compatibility. */
5855 else if (*previous >= OP_ASSERT && *previous <= OP_COND)
5858 int len = (int)(code - previous);
5859 pcre_uchar *bralink = NULL;
5860 pcre_uchar *brazeroptr = NULL;
5862 /* Repeating a DEFINE group is pointless, but Perl allows the syntax, so
5863 we just ignore the repeat. */
5865 if (*previous == OP_COND && previous[LINK_SIZE+1] == OP_DEF)
5868 /* There is no sense in actually repeating assertions. The only potential
5869 use of repetition is in cases when the assertion is optional. Therefore,
5870 if the minimum is greater than zero, just ignore the repeat. If the
5871 maximum is not zero or one, set it to 1. */
5873 if (*previous < OP_ONCE) /* Assertion */
5875 if (repeat_min > 0) goto END_REPEAT;
5876 if (repeat_max < 0 || repeat_max > 1) repeat_max = 1;
5879 /* The case of a zero minimum is special because of the need to stick
5880 OP_BRAZERO in front of it, and because the group appears once in the
5881 data, whereas in other cases it appears the minimum number of times. For
5882 this reason, it is simplest to treat this case separately, as otherwise
5883 the code gets far too messy. There are several special subcases when the
5886 if (repeat_min == 0)
5888 /* If the maximum is also zero, we used to just omit the group from the
5889 output altogether, like this:
5891 ** if (repeat_max == 0)
5897 However, that fails when a group or a subgroup within it is referenced
5898 as a subroutine from elsewhere in the pattern, so now we stick in
5899 OP_SKIPZERO in front of it so that it is skipped on execution. As we
5900 don't have a list of which groups are referenced, we cannot do this
5903 If the maximum is 1 or unlimited, we just have to stick in the BRAZERO
5904 and do no more at this point. However, we do need to adjust any
5905 OP_RECURSE calls inside the group that refer to the group itself or any
5906 internal or forward referenced group, because the offset is from the
5907 start of the whole regex. Temporarily terminate the pattern while doing
5910 if (repeat_max <= 1) /* Covers 0, 1, and unlimited */
5913 adjust_recurse(previous, 1, utf, cd, save_hwm_offset);
5914 memmove(previous + 1, previous, IN_UCHARS(len));
5916 if (repeat_max == 0)
5918 *previous++ = OP_SKIPZERO;
5921 brazeroptr = previous; /* Save for possessive optimizing */
5922 *previous++ = OP_BRAZERO + repeat_type;
5925 /* If the maximum is greater than 1 and limited, we have to replicate
5926 in a nested fashion, sticking OP_BRAZERO before each set of brackets.
5927 The first one has to be handled carefully because it's the original
5928 copy, which has to be moved up. The remainder can be handled by code
5929 that is common with the non-zero minimum case below. We have to
5930 adjust the value or repeat_max, since one less copy is required. Once
5931 again, we may have to adjust any OP_RECURSE calls inside the group. */
5937 adjust_recurse(previous, 2 + LINK_SIZE, utf, cd, save_hwm_offset);
5938 memmove(previous + 2 + LINK_SIZE, previous, IN_UCHARS(len));
5939 code += 2 + LINK_SIZE;
5940 *previous++ = OP_BRAZERO + repeat_type;
5941 *previous++ = OP_BRA;
5943 /* We chain together the bracket offset fields that have to be
5944 filled in later when the ends of the brackets are reached. */
5946 offset = (bralink == NULL)? 0 : (int)(previous - bralink);
5948 PUTINC(previous, 0, offset);
5954 /* If the minimum is greater than zero, replicate the group as many
5955 times as necessary, and adjust the maximum to the number of subsequent
5956 copies that we need. If we set a first char from the group, and didn't
5957 set a required char, copy the latter from the former. If there are any
5958 forward reference subroutine calls in the group, there will be entries on
5959 the workspace list; replicate these with an appropriate increment. */
5965 /* In the pre-compile phase, we don't actually do the replication. We
5966 just adjust the length as if we had. Do some paranoid checks for
5967 potential integer overflow. The INT64_OR_DOUBLE type is a 64-bit
5968 integer type when available, otherwise double. */
5970 if (lengthptr != NULL)
5972 int delta = (repeat_min - 1)*length_prevgroup;
5973 if ((INT64_OR_DOUBLE)(repeat_min - 1)*
5974 (INT64_OR_DOUBLE)length_prevgroup >
5975 (INT64_OR_DOUBLE)INT_MAX ||
5976 OFLOW_MAX - *lengthptr < delta)
5978 *errorcodeptr = ERR20;
5981 *lengthptr += delta;
5984 /* This is compiling for real. If there is a set first byte for
5985 the group, and we have not yet set a "required byte", set it. Make
5986 sure there is enough workspace for copying forward references before
5991 if (groupsetfirstchar && reqcharflags < 0)
5993 reqchar = firstchar;
5994 reqcharflags = firstcharflags;
5997 for (i = 1; i < repeat_min; i++)
6000 size_t this_hwm_offset = cd->hwm - cd->start_workspace;
6001 memcpy(code, previous, IN_UCHARS(len));
6003 while (cd->hwm > cd->start_workspace + cd->workspace_size -
6004 WORK_SIZE_SAFETY_MARGIN -
6005 (this_hwm_offset - save_hwm_offset))
6007 *errorcodeptr = expand_workspace(cd);
6008 if (*errorcodeptr != 0) goto FAILED;
6011 for (hc = (pcre_uchar *)cd->start_workspace + save_hwm_offset;
6012 hc < (pcre_uchar *)cd->start_workspace + this_hwm_offset;
6015 PUT(cd->hwm, 0, GET(hc, 0) + len);
6016 cd->hwm += LINK_SIZE;
6018 save_hwm_offset = this_hwm_offset;
6024 if (repeat_max > 0) repeat_max -= repeat_min;
6027 /* This code is common to both the zero and non-zero minimum cases. If
6028 the maximum is limited, it replicates the group in a nested fashion,
6029 remembering the bracket starts on a stack. In the case of a zero minimum,
6030 the first one was set up above. In all cases the repeat_max now specifies
6031 the number of additional copies needed. Again, we must remember to
6032 replicate entries on the forward reference list. */
6034 if (repeat_max >= 0)
6036 /* In the pre-compile phase, we don't actually do the replication. We
6037 just adjust the length as if we had. For each repetition we must add 1
6038 to the length for BRAZERO and for all but the last repetition we must
6039 add 2 + 2*LINKSIZE to allow for the nesting that occurs. Do some
6040 paranoid checks to avoid integer overflow. The INT64_OR_DOUBLE type is
6041 a 64-bit integer type when available, otherwise double. */
6043 if (lengthptr != NULL && repeat_max > 0)
6045 int delta = repeat_max * (length_prevgroup + 1 + 2 + 2*LINK_SIZE) -
6046 2 - 2*LINK_SIZE; /* Last one doesn't nest */
6047 if ((INT64_OR_DOUBLE)repeat_max *
6048 (INT64_OR_DOUBLE)(length_prevgroup + 1 + 2 + 2*LINK_SIZE)
6049 > (INT64_OR_DOUBLE)INT_MAX ||
6050 OFLOW_MAX - *lengthptr < delta)
6052 *errorcodeptr = ERR20;
6055 *lengthptr += delta;
6058 /* This is compiling for real */
6060 else for (i = repeat_max - 1; i >= 0; i--)
6063 size_t this_hwm_offset = cd->hwm - cd->start_workspace;
6065 *code++ = OP_BRAZERO + repeat_type;
6067 /* All but the final copy start a new nesting, maintaining the
6068 chain of brackets outstanding. */
6074 offset = (bralink == NULL)? 0 : (int)(code - bralink);
6076 PUTINC(code, 0, offset);
6079 memcpy(code, previous, IN_UCHARS(len));
6081 /* Ensure there is enough workspace for forward references before
6084 while (cd->hwm > cd->start_workspace + cd->workspace_size -
6085 WORK_SIZE_SAFETY_MARGIN -
6086 (this_hwm_offset - save_hwm_offset))
6088 *errorcodeptr = expand_workspace(cd);
6089 if (*errorcodeptr != 0) goto FAILED;
6092 for (hc = (pcre_uchar *)cd->start_workspace + save_hwm_offset;
6093 hc < (pcre_uchar *)cd->start_workspace + this_hwm_offset;
6096 PUT(cd->hwm, 0, GET(hc, 0) + len + ((i != 0)? 2+LINK_SIZE : 1));
6097 cd->hwm += LINK_SIZE;
6099 save_hwm_offset = this_hwm_offset;
6103 /* Now chain through the pending brackets, and fill in their length
6104 fields (which are holding the chain links pro tem). */
6106 while (bralink != NULL)
6109 int offset = (int)(code - bralink + 1);
6110 pcre_uchar *bra = code - offset;
6111 oldlinkoffset = GET(bra, 1);
6112 bralink = (oldlinkoffset == 0)? NULL : bralink - oldlinkoffset;
6114 PUTINC(code, 0, offset);
6115 PUT(bra, 1, offset);
6119 /* If the maximum is unlimited, set a repeater in the final copy. For
6120 ONCE brackets, that's all we need to do. However, possessively repeated
6121 ONCE brackets can be converted into non-capturing brackets, as the
6122 behaviour of (?:xx)++ is the same as (?>xx)++ and this saves having to
6123 deal with possessive ONCEs specially.
6125 Otherwise, when we are doing the actual compile phase, check to see
6126 whether this group is one that could match an empty string. If so,
6127 convert the initial operator to the S form (e.g. OP_BRA -> OP_SBRA) so
6128 that runtime checking can be done. [This check is also applied to ONCE
6129 groups at runtime, but in a different way.]
6131 Then, if the quantifier was possessive and the bracket is not a
6132 conditional, we convert the BRA code to the POS form, and the KET code to
6133 KETRPOS. (It turns out to be convenient at runtime to detect this kind of
6134 subpattern at both the start and at the end.) The use of special opcodes
6135 makes it possible to reduce greatly the stack usage in pcre_exec(). If
6136 the group is preceded by OP_BRAZERO, convert this to OP_BRAPOSZERO.
6138 Then, if the minimum number of matches is 1 or 0, cancel the possessive
6139 flag so that the default action below, of wrapping everything inside
6140 atomic brackets, does not happen. When the minimum is greater than 1,
6141 there will be earlier copies of the group, and so we still have to wrap
6146 pcre_uchar *ketcode = code - 1 - LINK_SIZE;
6147 pcre_uchar *bracode = ketcode - GET(ketcode, 1);
6149 /* Convert possessive ONCE brackets to non-capturing */
6151 if ((*bracode == OP_ONCE || *bracode == OP_ONCE_NC) &&
6152 possessive_quantifier) *bracode = OP_BRA;
6154 /* For non-possessive ONCE brackets, all we need to do is to
6157 if (*bracode == OP_ONCE || *bracode == OP_ONCE_NC)
6158 *ketcode = OP_KETRMAX + repeat_type;
6160 /* Handle non-ONCE brackets and possessive ONCEs (which have been
6161 converted to non-capturing above). */
6165 /* In the compile phase, check for empty string matching. */
6167 if (lengthptr == NULL)
6169 pcre_uchar *scode = bracode;
6172 if (could_be_empty_branch(scode, ketcode, utf, cd, NULL))
6174 *bracode += OP_SBRA - OP_BRA;
6177 scode += GET(scode, 1);
6179 while (*scode == OP_ALT);
6182 /* Handle possessive quantifiers. */
6184 if (possessive_quantifier)
6186 /* For COND brackets, we wrap the whole thing in a possessively
6187 repeated non-capturing bracket, because we have not invented POS
6188 versions of the COND opcodes. Because we are moving code along, we
6189 must ensure that any pending recursive references are updated. */
6191 if (*bracode == OP_COND || *bracode == OP_SCOND)
6193 int nlen = (int)(code - bracode);
6195 adjust_recurse(bracode, 1 + LINK_SIZE, utf, cd, save_hwm_offset);
6196 memmove(bracode + 1 + LINK_SIZE, bracode, IN_UCHARS(nlen));
6197 code += 1 + LINK_SIZE;
6198 nlen += 1 + LINK_SIZE;
6199 *bracode = OP_BRAPOS;
6200 *code++ = OP_KETRPOS;
6201 PUTINC(code, 0, nlen);
6202 PUT(bracode, 1, nlen);
6205 /* For non-COND brackets, we modify the BRA code and use KETRPOS. */
6209 *bracode += 1; /* Switch to xxxPOS opcodes */
6210 *ketcode = OP_KETRPOS;
6213 /* If the minimum is zero, mark it as possessive, then unset the
6214 possessive flag when the minimum is 0 or 1. */
6216 if (brazeroptr != NULL) *brazeroptr = OP_BRAPOSZERO;
6217 if (repeat_min < 2) possessive_quantifier = FALSE;
6220 /* Non-possessive quantifier */
6222 else *ketcode = OP_KETRMAX + repeat_type;
6227 /* If previous is OP_FAIL, it was generated by an empty class [] in
6228 JavaScript mode. The other ways in which OP_FAIL can be generated, that is
6229 by (*FAIL) or (?!) set previous to NULL, which gives a "nothing to repeat"
6230 error above. We can just ignore the repeat in JS case. */
6232 else if (*previous == OP_FAIL) goto END_REPEAT;
6234 /* Else there's some kind of shambles */
6238 *errorcodeptr = ERR11;
6242 /* If the character following a repeat is '+', possessive_quantifier is
6243 TRUE. For some opcodes, there are special alternative opcodes for this
6244 case. For anything else, we wrap the entire repeated item inside OP_ONCE
6245 brackets. Logically, the '+' notation is just syntactic sugar, taken from
6246 Sun's Java package, but the special opcodes can optimize it.
6248 Some (but not all) possessively repeated subpatterns have already been
6249 completely handled in the code just above. For them, possessive_quantifier
6250 is always FALSE at this stage. Note that the repeated item starts at
6251 tempcode, not at previous, which might be the first part of a string whose
6252 (former) last char we repeated. */
6254 if (possessive_quantifier)
6258 /* Possessifying an EXACT quantifier has no effect, so we can ignore it.
6259 However, QUERY, STAR, or UPTO may follow (for quantifiers such as {5,6},
6260 {5,}, or {5,10}). We skip over an EXACT item; if the length of what
6261 remains is greater than zero, there's a further opcode that can be
6262 handled. If not, do nothing, leaving the EXACT alone. */
6267 tempcode += PRIV(OP_lengths)[*tempcode] +
6268 ((tempcode[1 + IMM2_SIZE] == OP_PROP
6269 || tempcode[1 + IMM2_SIZE] == OP_NOTPROP)? 2 : 0);
6272 /* CHAR opcodes are used for exacts whose count is 1. */
6282 tempcode += PRIV(OP_lengths)[*tempcode];
6284 if (utf && HAS_EXTRALEN(tempcode[-1]))
6285 tempcode += GET_EXTRALEN(tempcode[-1]);
6289 /* For the class opcodes, the repeat operator appears at the end;
6290 adjust tempcode to point to it. */
6294 tempcode += 1 + 32/sizeof(pcre_uchar);
6297 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
6299 tempcode += GET(tempcode, 1);
6304 /* If tempcode is equal to code (which points to the end of the repeated
6305 item), it means we have skipped an EXACT item but there is no following
6306 QUERY, STAR, or UPTO; the value of len will be 0, and we do nothing. In
6307 all other cases, tempcode will be pointing to the repeat opcode, and will
6308 be less than code, so the value of len will be greater than 0. */
6310 len = (int)(code - tempcode);
6313 unsigned int repcode = *tempcode;
6315 /* There is a table for possessifying opcodes, all of which are less
6316 than OP_CALLOUT. A zero entry means there is no possessified version.
6319 if (repcode < OP_CALLOUT && opcode_possessify[repcode] > 0)
6320 *tempcode = opcode_possessify[repcode];
6322 /* For opcode without a special possessified version, wrap the item in
6323 ONCE brackets. Because we are moving code along, we must ensure that any
6324 pending recursive references are updated. */
6329 adjust_recurse(tempcode, 1 + LINK_SIZE, utf, cd, save_hwm_offset);
6330 memmove(tempcode + 1 + LINK_SIZE, tempcode, IN_UCHARS(len));
6331 code += 1 + LINK_SIZE;
6332 len += 1 + LINK_SIZE;
6333 tempcode[0] = OP_ONCE;
6335 PUTINC(code, 0, len);
6336 PUT(tempcode, 1, len);
6341 if (len > 0) switch (*tempcode)
6343 case OP_STAR: *tempcode = OP_POSSTAR; break;
6344 case OP_PLUS: *tempcode = OP_POSPLUS; break;
6345 case OP_QUERY: *tempcode = OP_POSQUERY; break;
6346 case OP_UPTO: *tempcode = OP_POSUPTO; break;
6348 case OP_STARI: *tempcode = OP_POSSTARI; break;
6349 case OP_PLUSI: *tempcode = OP_POSPLUSI; break;
6350 case OP_QUERYI: *tempcode = OP_POSQUERYI; break;
6351 case OP_UPTOI: *tempcode = OP_POSUPTOI; break;
6353 case OP_NOTSTAR: *tempcode = OP_NOTPOSSTAR; break;
6354 case OP_NOTPLUS: *tempcode = OP_NOTPOSPLUS; break;
6355 case OP_NOTQUERY: *tempcode = OP_NOTPOSQUERY; break;
6356 case OP_NOTUPTO: *tempcode = OP_NOTPOSUPTO; break;
6358 case OP_NOTSTARI: *tempcode = OP_NOTPOSSTARI; break;
6359 case OP_NOTPLUSI: *tempcode = OP_NOTPOSPLUSI; break;
6360 case OP_NOTQUERYI: *tempcode = OP_NOTPOSQUERYI; break;
6361 case OP_NOTUPTOI: *tempcode = OP_NOTPOSUPTOI; break;
6363 case OP_TYPESTAR: *tempcode = OP_TYPEPOSSTAR; break;
6364 case OP_TYPEPLUS: *tempcode = OP_TYPEPOSPLUS; break;
6365 case OP_TYPEQUERY: *tempcode = OP_TYPEPOSQUERY; break;
6366 case OP_TYPEUPTO: *tempcode = OP_TYPEPOSUPTO; break;
6368 case OP_CRSTAR: *tempcode = OP_CRPOSSTAR; break;
6369 case OP_CRPLUS: *tempcode = OP_CRPOSPLUS; break;
6370 case OP_CRQUERY: *tempcode = OP_CRPOSQUERY; break;
6371 case OP_CRRANGE: *tempcode = OP_CRPOSRANGE; break;
6373 /* Because we are moving code along, we must ensure that any
6374 pending recursive references are updated. */
6378 adjust_recurse(tempcode, 1 + LINK_SIZE, utf, cd, save_hwm_offset);
6379 memmove(tempcode + 1 + LINK_SIZE, tempcode, IN_UCHARS(len));
6380 code += 1 + LINK_SIZE;
6381 len += 1 + LINK_SIZE;
6382 tempcode[0] = OP_ONCE;
6384 PUTINC(code, 0, len);
6385 PUT(tempcode, 1, len);
6391 /* In all case we no longer have a previous item. We also set the
6392 "follows varying string" flag for subsequently encountered reqchars if
6393 it isn't already set and we have just passed a varying length item. */
6397 cd->req_varyopt |= reqvary;
6401 /* ===================================================================*/
6402 /* Start of nested parenthesized sub-expression, or comment or lookahead or
6403 lookbehind or option setting or condition or all the other extended
6404 parenthesis forms. */
6406 case CHAR_LEFT_PARENTHESIS:
6407 newoptions = options;
6410 save_hwm_offset = cd->hwm - cd->start_workspace;
6411 reset_bracount = FALSE;
6413 /* First deal with various "verbs" that can be introduced by '*'. */
6416 if (ptr[0] == CHAR_ASTERISK && (ptr[1] == ':'
6417 || (MAX_255(ptr[1]) && ((cd->ctypes[ptr[1]] & ctype_letter) != 0))))
6421 const char *vn = verbnames;
6422 const pcre_uchar *name = ptr + 1;
6423 const pcre_uchar *arg = NULL;
6426 while (MAX_255(*ptr) && (cd->ctypes[*ptr] & ctype_letter) != 0) ptr++;
6427 namelen = (int)(ptr - name);
6429 /* It appears that Perl allows any characters whatsoever, other than
6430 a closing parenthesis, to appear in arguments, so we no longer insist on
6431 letters, digits, and underscores. */
6433 if (*ptr == CHAR_COLON)
6436 while (*ptr != CHAR_NULL && *ptr != CHAR_RIGHT_PARENTHESIS) ptr++;
6437 arglen = (int)(ptr - arg);
6438 if ((unsigned int)arglen > MAX_MARK)
6440 *errorcodeptr = ERR75;
6445 if (*ptr != CHAR_RIGHT_PARENTHESIS)
6447 *errorcodeptr = ERR60;
6451 /* Scan the table of verb names */
6453 for (i = 0; i < verbcount; i++)
6455 if (namelen == verbs[i].len &&
6456 STRNCMP_UC_C8(name, vn, namelen) == 0)
6460 /* Check for open captures before ACCEPT and convert it to
6461 ASSERT_ACCEPT if in an assertion. */
6463 if (verbs[i].op == OP_ACCEPT)
6468 *errorcodeptr = ERR59;
6471 cd->had_accept = TRUE;
6472 for (oc = cd->open_caps; oc != NULL; oc = oc->next)
6475 PUT2INC(code, 0, oc->number);
6478 (cd->assert_depth > 0)? OP_ASSERT_ACCEPT : OP_ACCEPT;
6480 /* Do not set firstchar after *ACCEPT */
6481 if (firstcharflags == REQ_UNSET) firstcharflags = REQ_NONE;
6484 /* Handle other cases with/without an argument */
6486 else if (arglen == 0)
6488 if (verbs[i].op < 0) /* Argument is mandatory */
6490 *errorcodeptr = ERR66;
6493 setverb = *code++ = verbs[i].op;
6498 if (verbs[i].op_arg < 0) /* Argument is forbidden */
6500 *errorcodeptr = ERR59;
6503 setverb = *code++ = verbs[i].op_arg;
6505 memcpy(code, arg, IN_UCHARS(arglen));
6514 cd->external_flags |= PCRE_HASTHEN;
6521 cd->had_pruneorskip = TRUE;
6525 break; /* Found verb, exit loop */
6528 vn += verbs[i].len + 1;
6531 if (i < verbcount) continue; /* Successfully handled a verb */
6532 *errorcodeptr = ERR60; /* Verb not recognized */
6536 /* Deal with the extended parentheses; all are introduced by '?', and the
6537 appearance of any of them means that this is not a capturing group. */
6539 else if (*ptr == CHAR_QUESTION_MARK)
6541 int i, set, unset, namelen;
6543 const pcre_uchar *name;
6548 case CHAR_NUMBER_SIGN: /* Comment; skip to ket */
6550 while (*ptr != CHAR_NULL && *ptr != CHAR_RIGHT_PARENTHESIS) ptr++;
6551 if (*ptr == CHAR_NULL)
6553 *errorcodeptr = ERR18;
6559 /* ------------------------------------------------------------ */
6560 case CHAR_VERTICAL_LINE: /* Reset capture count for each branch */
6561 reset_bracount = TRUE;
6564 /* ------------------------------------------------------------ */
6565 case CHAR_COLON: /* Non-capturing bracket */
6571 /* ------------------------------------------------------------ */
6572 case CHAR_LEFT_PARENTHESIS:
6573 bravalue = OP_COND; /* Conditional group */
6576 /* A condition can be an assertion, a number (referring to a numbered
6577 group's having been set), a name (referring to a named group), or 'R',
6578 referring to recursion. R<digits> and R&name are also permitted for
6581 There are ways of testing a named group: (?(name)) is used by Python;
6582 Perl 5.10 onwards uses (?(<name>) or (?('name')).
6584 There is one unfortunate ambiguity, caused by history. 'R' can be the
6585 recursive thing or the name 'R' (and similarly for 'R' followed by
6586 digits). We look for a name first; if not found, we try the other case.
6588 For compatibility with auto-callouts, we allow a callout to be
6589 specified before a condition that is an assertion. First, check for the
6590 syntax of a callout; if found, adjust the temporary pointer that is
6591 used to check for an assertion condition. That's all that is needed! */
6593 if (ptr[1] == CHAR_QUESTION_MARK && ptr[2] == CHAR_C)
6595 for (i = 3;; i++) if (!IS_DIGIT(ptr[i])) break;
6596 if (ptr[i] == CHAR_RIGHT_PARENTHESIS)
6600 /* For conditions that are assertions, check the syntax, and then exit
6601 the switch. This will take control down to where bracketed groups,
6602 including assertions, are processed. */
6604 if (tempptr[1] == CHAR_QUESTION_MARK &&
6605 (tempptr[2] == CHAR_EQUALS_SIGN ||
6606 tempptr[2] == CHAR_EXCLAMATION_MARK ||
6607 tempptr[2] == CHAR_LESS_THAN_SIGN))
6610 /* Other conditions use OP_CREF/OP_DNCREF/OP_RREF/OP_DNRREF, and all
6611 need to skip at least 1+IMM2_SIZE bytes at the start of the group. */
6613 code[1+LINK_SIZE] = OP_CREF;
6614 skipbytes = 1+IMM2_SIZE;
6615 refsign = -1; /* => not a number */
6616 namelen = -1; /* => not a name; must set to avoid warning */
6617 name = NULL; /* Always set to avoid warning */
6618 recno = 0; /* Always set to avoid warning */
6620 /* Check for a test for recursion in a named group. */
6623 if (*ptr == CHAR_R && ptr[1] == CHAR_AMPERSAND)
6627 code[1+LINK_SIZE] = OP_RREF; /* Change the type of test */
6630 /* Check for a test for a named group's having been set, using the Perl
6631 syntax (?(<name>) or (?('name'), and also allow for the original PCRE
6632 syntax of (?(name) or for (?(+n), (?(-n), and just (?(n). */
6634 else if (*ptr == CHAR_LESS_THAN_SIGN)
6636 terminator = CHAR_GREATER_THAN_SIGN;
6639 else if (*ptr == CHAR_APOSTROPHE)
6641 terminator = CHAR_APOSTROPHE;
6646 terminator = CHAR_NULL;
6647 if (*ptr == CHAR_MINUS || *ptr == CHAR_PLUS) refsign = *ptr++;
6648 else if (IS_DIGIT(*ptr)) refsign = 0;
6651 /* Handle a number */
6655 while (IS_DIGIT(*ptr))
6657 recno = recno * 10 + (int)(*ptr - CHAR_0);
6662 /* Otherwise we expect to read a name; anything else is an error. When
6663 a name is one of a number of duplicates, a different opcode is used and
6664 it needs more memory. Unfortunately we cannot tell whether a name is a
6665 duplicate in the first pass, so we have to allow for more memory. */
6671 *errorcodeptr = ERR84;
6674 if (!MAX_255(*ptr) || (cd->ctypes[*ptr] & ctype_word) == 0)
6676 *errorcodeptr = ERR28; /* Assertion expected */
6680 while (MAX_255(*ptr) && (cd->ctypes[*ptr] & ctype_word) != 0)
6684 namelen = (int)(ptr - name);
6685 if (lengthptr != NULL) *lengthptr += IMM2_SIZE;
6688 /* Check the terminator */
6690 if ((terminator > 0 && *ptr++ != (pcre_uchar)terminator) ||
6691 *ptr++ != CHAR_RIGHT_PARENTHESIS)
6693 ptr--; /* Error offset */
6694 *errorcodeptr = ERR26; /* Malformed number or name */
6698 /* Do no further checking in the pre-compile phase. */
6700 if (lengthptr != NULL) break;
6702 /* In the real compile we do the work of looking for the actual
6703 reference. If refsign is not negative, it means we have a number in
6710 *errorcodeptr = ERR35;
6713 if (refsign != 0) recno = (refsign == CHAR_MINUS)?
6714 cd->bracount - recno + 1 : recno + cd->bracount;
6715 if (recno <= 0 || recno > cd->final_bracount)
6717 *errorcodeptr = ERR15;
6720 PUT2(code, 2+LINK_SIZE, recno);
6724 /* Otherwise look for the name. */
6726 slot = cd->name_table;
6727 for (i = 0; i < cd->names_found; i++)
6729 if (STRNCMP_UC_UC(name, slot+IMM2_SIZE, namelen) == 0) break;
6730 slot += cd->name_entry_size;
6733 /* Found the named subpattern. If the name is duplicated, add one to
6734 the opcode to change CREF/RREF into DNCREF/DNRREF and insert
6735 appropriate data values. Otherwise, just insert the unique subpattern
6738 if (i < cd->names_found)
6742 recno = GET2(slot, 0); /* Number from first found */
6743 for (; i < cd->names_found; i++)
6745 slot += cd->name_entry_size;
6746 if (STRNCMP_UC_UC(name, slot+IMM2_SIZE, namelen) != 0) break;
6751 PUT2(code, 2+LINK_SIZE, offset);
6752 PUT2(code, 2+LINK_SIZE+IMM2_SIZE, count);
6753 skipbytes += IMM2_SIZE;
6754 code[1+LINK_SIZE]++;
6756 else /* Not a duplicated name */
6758 PUT2(code, 2+LINK_SIZE, recno);
6762 /* If terminator == CHAR_NULL it means that the name followed directly
6763 after the opening parenthesis [e.g. (?(abc)...] and in this case there
6764 are some further alternatives to try. For the cases where terminator !=
6765 CHAR_NULL [things like (?(<name>... or (?('name')... or (?(R&name)... ]
6766 we have now checked all the possibilities, so give an error. */
6768 else if (terminator != CHAR_NULL)
6770 *errorcodeptr = ERR15;
6774 /* Check for (?(R) for recursion. Allow digits after R to specify a
6775 specific group number. */
6777 else if (*name == CHAR_R)
6780 for (i = 1; i < namelen; i++)
6782 if (!IS_DIGIT(name[i]))
6784 *errorcodeptr = ERR15;
6787 recno = recno * 10 + name[i] - CHAR_0;
6789 if (recno == 0) recno = RREF_ANY;
6790 code[1+LINK_SIZE] = OP_RREF; /* Change test type */
6791 PUT2(code, 2+LINK_SIZE, recno);
6794 /* Similarly, check for the (?(DEFINE) "condition", which is always
6797 else if (namelen == 6 && STRNCMP_UC_C8(name, STRING_DEFINE, 6) == 0)
6799 code[1+LINK_SIZE] = OP_DEF;
6803 /* Reference to an unidentified subpattern. */
6807 *errorcodeptr = ERR15;
6813 /* ------------------------------------------------------------ */
6814 case CHAR_EQUALS_SIGN: /* Positive lookahead */
6815 bravalue = OP_ASSERT;
6816 cd->assert_depth += 1;
6820 /* Optimize (?!) to (*FAIL) unless it is quantified - which is a weird
6821 thing to do, but Perl allows all assertions to be quantified, and when
6822 they contain capturing parentheses there may be a potential use for
6823 this feature. Not that that applies to a quantified (?!) but we allow
6824 it for uniformity. */
6826 /* ------------------------------------------------------------ */
6827 case CHAR_EXCLAMATION_MARK: /* Negative lookahead */
6829 if (*ptr == CHAR_RIGHT_PARENTHESIS && ptr[1] != CHAR_ASTERISK &&
6830 ptr[1] != CHAR_PLUS && ptr[1] != CHAR_QUESTION_MARK &&
6831 (ptr[1] != CHAR_LEFT_CURLY_BRACKET || !is_counted_repeat(ptr+2)))
6837 bravalue = OP_ASSERT_NOT;
6838 cd->assert_depth += 1;
6842 /* ------------------------------------------------------------ */
6843 case CHAR_LESS_THAN_SIGN: /* Lookbehind or named define */
6846 case CHAR_EQUALS_SIGN: /* Positive lookbehind */
6847 bravalue = OP_ASSERTBACK;
6848 cd->assert_depth += 1;
6852 case CHAR_EXCLAMATION_MARK: /* Negative lookbehind */
6853 bravalue = OP_ASSERTBACK_NOT;
6854 cd->assert_depth += 1;
6858 default: /* Could be name define, else bad */
6859 if (MAX_255(ptr[1]) && (cd->ctypes[ptr[1]] & ctype_word) != 0)
6861 ptr++; /* Correct offset for error */
6862 *errorcodeptr = ERR24;
6868 /* ------------------------------------------------------------ */
6869 case CHAR_GREATER_THAN_SIGN: /* One-time brackets */
6875 /* ------------------------------------------------------------ */
6876 case CHAR_C: /* Callout - may be followed by digits; */
6877 previous_callout = code; /* Save for later completion */
6878 after_manual_callout = 1; /* Skip one item before completing */
6879 *code++ = OP_CALLOUT;
6883 while(IS_DIGIT(*ptr))
6884 n = n * 10 + *ptr++ - CHAR_0;
6885 if (*ptr != CHAR_RIGHT_PARENTHESIS)
6887 *errorcodeptr = ERR39;
6892 *errorcodeptr = ERR38;
6896 PUT(code, 0, (int)(ptr - cd->start_pattern + 1)); /* Pattern offset */
6897 PUT(code, LINK_SIZE, 0); /* Default length */
6898 code += 2 * LINK_SIZE;
6904 /* ------------------------------------------------------------ */
6905 case CHAR_P: /* Python-style named subpattern handling */
6906 if (*(++ptr) == CHAR_EQUALS_SIGN ||
6907 *ptr == CHAR_GREATER_THAN_SIGN) /* Reference or recursion */
6909 is_recurse = *ptr == CHAR_GREATER_THAN_SIGN;
6910 terminator = CHAR_RIGHT_PARENTHESIS;
6911 goto NAMED_REF_OR_RECURSE;
6913 else if (*ptr != CHAR_LESS_THAN_SIGN) /* Test for Python-style defn */
6915 *errorcodeptr = ERR41;
6918 /* Fall through to handle (?P< as (?< is handled */
6921 /* ------------------------------------------------------------ */
6922 DEFINE_NAME: /* Come here from (?< handling */
6923 case CHAR_APOSTROPHE:
6924 terminator = (*ptr == CHAR_LESS_THAN_SIGN)?
6925 CHAR_GREATER_THAN_SIGN : CHAR_APOSTROPHE;
6929 *errorcodeptr = ERR84; /* Group name must start with non-digit */
6932 while (MAX_255(*ptr) && (cd->ctypes[*ptr] & ctype_word) != 0) ptr++;
6933 namelen = (int)(ptr - name);
6935 /* In the pre-compile phase, do a syntax check, remember the longest
6936 name, and then remember the group in a vector, expanding it if
6937 necessary. Duplicates for the same number are skipped; other duplicates
6938 are checked for validity. In the actual compile, there is nothing to
6941 if (lengthptr != NULL)
6944 pcre_uint32 number = cd->bracount + 1;
6946 if (*ptr != (pcre_uchar)terminator)
6948 *errorcodeptr = ERR42;
6952 if (cd->names_found >= MAX_NAME_COUNT)
6954 *errorcodeptr = ERR49;
6958 if (namelen + IMM2_SIZE + 1 > cd->name_entry_size)
6960 cd->name_entry_size = namelen + IMM2_SIZE + 1;
6961 if (namelen > MAX_NAME_SIZE)
6963 *errorcodeptr = ERR48;
6968 /* Scan the list to check for duplicates. For duplicate names, if the
6969 number is the same, break the loop, which causes the name to be
6970 discarded; otherwise, if DUPNAMES is not set, give an error.
6971 If it is set, allow the name with a different number, but continue
6972 scanning in case this is a duplicate with the same number. For
6973 non-duplicate names, give an error if the number is duplicated. */
6975 ng = cd->named_groups;
6976 for (i = 0; i < cd->names_found; i++, ng++)
6978 if (namelen == ng->length &&
6979 STRNCMP_UC_UC(name, ng->name, namelen) == 0)
6981 if (ng->number == number) break;
6982 if ((options & PCRE_DUPNAMES) == 0)
6984 *errorcodeptr = ERR43;
6987 cd->dupnames = TRUE; /* Duplicate names exist */
6989 else if (ng->number == number)
6991 *errorcodeptr = ERR65;
6996 if (i >= cd->names_found) /* Not a duplicate with same number */
6998 /* Increase the list size if necessary */
7000 if (cd->names_found >= cd->named_group_list_size)
7002 int newsize = cd->named_group_list_size * 2;
7003 named_group *newspace = (PUBL(malloc))
7004 (newsize * sizeof(named_group));
7006 if (newspace == NULL)
7008 *errorcodeptr = ERR21;
7012 memcpy(newspace, cd->named_groups,
7013 cd->named_group_list_size * sizeof(named_group));
7014 if (cd->named_group_list_size > NAMED_GROUP_LIST_SIZE)
7015 (PUBL(free))((void *)cd->named_groups);
7016 cd->named_groups = newspace;
7017 cd->named_group_list_size = newsize;
7020 cd->named_groups[cd->names_found].name = name;
7021 cd->named_groups[cd->names_found].length = namelen;
7022 cd->named_groups[cd->names_found].number = number;
7027 ptr++; /* Move past > or ' in both passes. */
7028 goto NUMBERED_GROUP;
7031 /* ------------------------------------------------------------ */
7032 case CHAR_AMPERSAND: /* Perl recursion/subroutine syntax */
7033 terminator = CHAR_RIGHT_PARENTHESIS;
7037 /* We come here from the Python syntax above that handles both
7038 references (?P=name) and recursion (?P>name), as well as falling
7039 through from the Perl recursion syntax (?&name). We also come here from
7040 the Perl \k<name> or \k'name' back reference syntax and the \k{name}
7041 .NET syntax, and the Oniguruma \g<...> and \g'...' subroutine syntax. */
7043 NAMED_REF_OR_RECURSE:
7047 *errorcodeptr = ERR84; /* Group name must start with non-digit */
7050 while (MAX_255(*ptr) && (cd->ctypes[*ptr] & ctype_word) != 0) ptr++;
7051 namelen = (int)(ptr - name);
7053 /* In the pre-compile phase, do a syntax check. We used to just set
7054 a dummy reference number, because it was not used in the first pass.
7055 However, with the change of recursive back references to be atomic,
7056 we have to look for the number so that this state can be identified, as
7057 otherwise the incorrect length is computed. If it's not a backwards
7058 reference, the dummy number will do. */
7060 if (lengthptr != NULL)
7066 *errorcodeptr = ERR62;
7069 if (*ptr != (pcre_uchar)terminator)
7071 *errorcodeptr = ERR42;
7074 if (namelen > MAX_NAME_SIZE)
7076 *errorcodeptr = ERR48;
7080 /* The name table does not exist in the first pass; instead we must
7081 scan the list of names encountered so far in order to get the
7082 number. If the name is not found, set the value to 0 for a forward
7086 ng = cd->named_groups;
7087 for (i = 0; i < cd->names_found; i++, ng++)
7089 if (namelen == ng->length &&
7090 STRNCMP_UC_UC(name, ng->name, namelen) == 0)
7094 if (is_recurse) break;
7095 for (oc = cd->open_caps; oc != NULL; oc = oc->next)
7097 if (oc->number == recno)
7106 /* Count named back references. */
7108 if (!is_recurse) cd->namedrefcount++;
7111 /* In the real compile, search the name table. We check the name
7112 first, and then check that we have reached the end of the name in the
7113 table. That way, if the name is longer than any in the table, the
7114 comparison will fail without reading beyond the table entry. */
7118 slot = cd->name_table;
7119 for (i = 0; i < cd->names_found; i++)
7121 if (STRNCMP_UC_UC(name, slot+IMM2_SIZE, namelen) == 0 &&
7122 slot[IMM2_SIZE+namelen] == 0)
7124 slot += cd->name_entry_size;
7127 if (i < cd->names_found)
7129 recno = GET2(slot, 0);
7133 *errorcodeptr = ERR15;
7138 /* In both phases, for recursions, we can now go to the code than
7139 handles numerical recursion. */
7141 if (is_recurse) goto HANDLE_RECURSION;
7143 /* In the second pass we must see if the name is duplicated. If so, we
7144 generate a different opcode. */
7146 if (lengthptr == NULL && cd->dupnames)
7149 unsigned int index = i;
7150 pcre_uchar *cslot = slot + cd->name_entry_size;
7152 for (i++; i < cd->names_found; i++)
7154 if (STRCMP_UC_UC(slot + IMM2_SIZE, cslot + IMM2_SIZE) != 0) break;
7156 cslot += cd->name_entry_size;
7161 if (firstcharflags == REQ_UNSET) firstcharflags = REQ_NONE;
7163 *code++ = ((options & PCRE_CASELESS) != 0)? OP_DNREFI : OP_DNREF;
7164 PUT2INC(code, 0, index);
7165 PUT2INC(code, 0, count);
7167 /* Process each potentially referenced group. */
7169 for (; slot < cslot; slot += cd->name_entry_size)
7172 recno = GET2(slot, 0);
7173 cd->backref_map |= (recno < 32)? (1 << recno) : 1;
7174 if (recno > cd->top_backref) cd->top_backref = recno;
7176 /* Check to see if this back reference is recursive, that it, it
7177 is inside the group that it references. A flag is set so that the
7178 group can be made atomic. */
7180 for (oc = cd->open_caps; oc != NULL; oc = oc->next)
7182 if (oc->number == recno)
7190 continue; /* End of back ref handling */
7194 /* First pass, or a non-duplicated name. */
7196 goto HANDLE_REFERENCE;
7199 /* ------------------------------------------------------------ */
7200 case CHAR_R: /* Recursion */
7201 ptr++; /* Same as (?0) */
7205 /* ------------------------------------------------------------ */
7206 case CHAR_MINUS: case CHAR_PLUS: /* Recursion or subroutine */
7207 case CHAR_0: case CHAR_1: case CHAR_2: case CHAR_3: case CHAR_4:
7208 case CHAR_5: case CHAR_6: case CHAR_7: case CHAR_8: case CHAR_9:
7210 const pcre_uchar *called;
7211 terminator = CHAR_RIGHT_PARENTHESIS;
7213 /* Come here from the \g<...> and \g'...' code (Oniguruma
7214 compatibility). However, the syntax has been checked to ensure that
7215 the ... are a (signed) number, so that neither ERR63 nor ERR29 will
7216 be called on this path, nor with the jump to OTHER_CHAR_AFTER_QUERY
7219 HANDLE_NUMERICAL_RECURSION:
7221 if ((refsign = *ptr) == CHAR_PLUS)
7224 if (!IS_DIGIT(*ptr))
7226 *errorcodeptr = ERR63;
7230 else if (refsign == CHAR_MINUS)
7232 if (!IS_DIGIT(ptr[1]))
7233 goto OTHER_CHAR_AFTER_QUERY;
7238 while(IS_DIGIT(*ptr))
7239 recno = recno * 10 + *ptr++ - CHAR_0;
7241 if (*ptr != (pcre_uchar)terminator)
7243 *errorcodeptr = ERR29;
7247 if (refsign == CHAR_MINUS)
7251 *errorcodeptr = ERR58;
7254 recno = cd->bracount - recno + 1;
7257 *errorcodeptr = ERR15;
7261 else if (refsign == CHAR_PLUS)
7265 *errorcodeptr = ERR58;
7268 recno += cd->bracount;
7271 /* Come here from code above that handles a named recursion */
7276 called = cd->start_code;
7278 /* When we are actually compiling, find the bracket that is being
7279 referenced. Temporarily end the regex in case it doesn't exist before
7280 this point. If we end up with a forward reference, first check that
7281 the bracket does occur later so we can give the error (and position)
7282 now. Then remember this forward reference in the workspace so it can
7283 be filled in at the end. */
7285 if (lengthptr == NULL)
7289 called = PRIV(find_bracket)(cd->start_code, utf, recno);
7291 /* Forward reference */
7295 if (recno > cd->final_bracount)
7297 *errorcodeptr = ERR15;
7301 /* Fudge the value of "called" so that when it is inserted as an
7302 offset below, what it actually inserted is the reference number
7303 of the group. Then remember the forward reference. */
7305 called = cd->start_code + recno;
7306 if (cd->hwm >= cd->start_workspace + cd->workspace_size -
7307 WORK_SIZE_SAFETY_MARGIN)
7309 *errorcodeptr = expand_workspace(cd);
7310 if (*errorcodeptr != 0) goto FAILED;
7312 PUTINC(cd->hwm, 0, (int)(code + 1 - cd->start_code));
7315 /* If not a forward reference, and the subpattern is still open,
7316 this is a recursive call. We check to see if this is a left
7317 recursion that could loop for ever, and diagnose that case. We
7318 must not, however, do this check if we are in a conditional
7319 subpattern because the condition might be testing for recursion in
7320 a pattern such as /(?(R)a+|(?R)b)/, which is perfectly valid.
7321 Forever loops are also detected at runtime, so those that occur in
7322 conditional subpatterns will be picked up then. */
7324 else if (GET(called, 1) == 0 && cond_depth <= 0 &&
7325 could_be_empty(called, code, bcptr, utf, cd))
7327 *errorcodeptr = ERR40;
7332 /* Insert the recursion/subroutine item. It does not have a set first
7333 character (relevant if it is repeated, because it will then be
7334 wrapped with ONCE brackets). */
7337 PUT(code, 1, (int)(called - cd->start_code));
7338 code += 1 + LINK_SIZE;
7339 groupsetfirstchar = FALSE;
7342 /* Can't determine a first byte now */
7344 if (firstcharflags == REQ_UNSET) firstcharflags = REQ_NONE;
7348 /* ------------------------------------------------------------ */
7349 default: /* Other characters: check option setting */
7350 OTHER_CHAR_AFTER_QUERY:
7354 while (*ptr != CHAR_RIGHT_PARENTHESIS && *ptr != CHAR_COLON)
7358 case CHAR_MINUS: optset = &unset; break;
7360 case CHAR_J: /* Record that it changed in the external options */
7361 *optset |= PCRE_DUPNAMES;
7362 cd->external_flags |= PCRE_JCHANGED;
7365 case CHAR_i: *optset |= PCRE_CASELESS; break;
7366 case CHAR_m: *optset |= PCRE_MULTILINE; break;
7367 case CHAR_s: *optset |= PCRE_DOTALL; break;
7368 case CHAR_x: *optset |= PCRE_EXTENDED; break;
7369 case CHAR_U: *optset |= PCRE_UNGREEDY; break;
7370 case CHAR_X: *optset |= PCRE_EXTRA; break;
7372 default: *errorcodeptr = ERR12;
7373 ptr--; /* Correct the offset */
7378 /* Set up the changed option bits, but don't change anything yet. */
7380 newoptions = (options | set) & (~unset);
7382 /* If the options ended with ')' this is not the start of a nested
7383 group with option changes, so the options change at this level. If this
7384 item is right at the start of the pattern, the options can be
7385 abstracted and made external in the pre-compile phase, and ignored in
7386 the compile phase. This can be helpful when matching -- for instance in
7387 caseless checking of required bytes.
7389 If the code pointer is not (cd->start_code + 1 + LINK_SIZE), we are
7390 definitely *not* at the start of the pattern because something has been
7391 compiled. In the pre-compile phase, however, the code pointer can have
7392 that value after the start, because it gets reset as code is discarded
7393 during the pre-compile. However, this can happen only at top level - if
7394 we are within parentheses, the starting BRA will still be present. At
7395 any parenthesis level, the length value can be used to test if anything
7396 has been compiled at that level. Thus, a test for both these conditions
7397 is necessary to ensure we correctly detect the start of the pattern in
7400 If we are not at the pattern start, reset the greedy defaults and the
7401 case value for firstchar and reqchar. */
7403 if (*ptr == CHAR_RIGHT_PARENTHESIS)
7405 if (code == cd->start_code + 1 + LINK_SIZE &&
7406 (lengthptr == NULL || *lengthptr == 2 + 2*LINK_SIZE))
7408 cd->external_options = newoptions;
7412 greedy_default = ((newoptions & PCRE_UNGREEDY) != 0);
7413 greedy_non_default = greedy_default ^ 1;
7414 req_caseopt = ((newoptions & PCRE_CASELESS) != 0)? REQ_CASELESS:0;
7417 /* Change options at this level, and pass them back for use
7418 in subsequent branches. */
7420 *optionsptr = options = newoptions;
7421 previous = NULL; /* This item can't be repeated */
7422 continue; /* It is complete */
7425 /* If the options ended with ':' we are heading into a nested group
7426 with possible change of options. Such groups are non-capturing and are
7427 not assertions of any kind. All we need to do is skip over the ':';
7428 the newoptions value is handled below. */
7432 } /* End of switch for character following (? */
7433 } /* End of (? handling */
7435 /* Opening parenthesis not followed by '*' or '?'. If PCRE_NO_AUTO_CAPTURE
7436 is set, all unadorned brackets become non-capturing and behave like (?:...)
7439 else if ((options & PCRE_NO_AUTO_CAPTURE) != 0)
7444 /* Else we have a capturing group. */
7450 PUT2(code, 1+LINK_SIZE, cd->bracount);
7451 skipbytes = IMM2_SIZE;
7454 /* Process nested bracketed regex. First check for parentheses nested too
7457 if ((cd->parens_depth += 1) > PARENS_NEST_LIMIT)
7459 *errorcodeptr = ERR82;
7463 /* Assertions used not to be repeatable, but this was changed for Perl
7464 compatibility, so all kinds can now be repeated. We copy code into a
7465 non-register variable (tempcode) in order to be able to pass its address
7466 because some compilers complain otherwise. */
7468 previous = code; /* For handling repetition */
7471 tempreqvary = cd->req_varyopt; /* Save value before bracket */
7472 tempbracount = cd->bracount; /* Save value before bracket */
7473 length_prevgroup = 0; /* Initialize for pre-compile phase */
7476 newoptions, /* The complete new option state */
7477 &tempcode, /* Where to put code (updated) */
7478 &ptr, /* Input pointer (updated) */
7479 errorcodeptr, /* Where to put an error message */
7480 (bravalue == OP_ASSERTBACK ||
7481 bravalue == OP_ASSERTBACK_NOT), /* TRUE if back assert */
7482 reset_bracount, /* True if (?| group */
7483 skipbytes, /* Skip over bracket number */
7485 ((bravalue == OP_COND)?1:0), /* Depth of condition subpatterns */
7486 &subfirstchar, /* For possible first char */
7488 &subreqchar, /* For possible last char */
7490 bcptr, /* Current branch chain */
7491 cd, /* Tables block */
7492 (lengthptr == NULL)? NULL : /* Actual compile phase */
7493 &length_prevgroup /* Pre-compile phase */
7497 cd->parens_depth -= 1;
7499 /* If this was an atomic group and there are no capturing groups within it,
7500 generate OP_ONCE_NC instead of OP_ONCE. */
7502 if (bravalue == OP_ONCE && cd->bracount <= tempbracount)
7505 if (bravalue >= OP_ASSERT && bravalue <= OP_ASSERTBACK_NOT)
7506 cd->assert_depth -= 1;
7508 /* At the end of compiling, code is still pointing to the start of the
7509 group, while tempcode has been updated to point past the end of the group.
7510 The pattern pointer (ptr) is on the bracket.
7512 If this is a conditional bracket, check that there are no more than
7513 two branches in the group, or just one if it's a DEFINE group. We do this
7514 in the real compile phase, not in the pre-pass, where the whole group may
7515 not be available. */
7517 if (bravalue == OP_COND && lengthptr == NULL)
7519 pcre_uchar *tc = code;
7526 while (*tc != OP_KET);
7528 /* A DEFINE group is never obeyed inline (the "condition" is always
7529 false). It must have only one branch. */
7531 if (code[LINK_SIZE+1] == OP_DEF)
7535 *errorcodeptr = ERR54;
7538 bravalue = OP_DEF; /* Just a flag to suppress char handling below */
7541 /* A "normal" conditional group. If there is just one branch, we must not
7542 make use of its firstchar or reqchar, because this is equivalent to an
7543 empty second branch. */
7549 *errorcodeptr = ERR27;
7552 if (condcount == 1) subfirstcharflags = subreqcharflags = REQ_NONE;
7556 /* Error if hit end of pattern */
7558 if (*ptr != CHAR_RIGHT_PARENTHESIS)
7560 *errorcodeptr = ERR14;
7564 /* In the pre-compile phase, update the length by the length of the group,
7565 less the brackets at either end. Then reduce the compiled code to just a
7566 set of non-capturing brackets so that it doesn't use much memory if it is
7567 duplicated by a quantifier.*/
7569 if (lengthptr != NULL)
7571 if (OFLOW_MAX - *lengthptr < length_prevgroup - 2 - 2*LINK_SIZE)
7573 *errorcodeptr = ERR20;
7576 *lengthptr += length_prevgroup - 2 - 2*LINK_SIZE;
7577 code++; /* This already contains bravalue */
7578 PUTINC(code, 0, 1 + LINK_SIZE);
7580 PUTINC(code, 0, 1 + LINK_SIZE);
7581 break; /* No need to waste time with special character handling */
7584 /* Otherwise update the main code pointer to the end of the group. */
7588 /* For a DEFINE group, required and first character settings are not
7591 if (bravalue == OP_DEF) break;
7593 /* Handle updating of the required and first characters for other types of
7594 group. Update for normal brackets of all kinds, and conditions with two
7595 branches (see code above). If the bracket is followed by a quantifier with
7596 zero repeat, we have to back off. Hence the definition of zeroreqchar and
7597 zerofirstchar outside the main loop so that they can be accessed for the
7600 zeroreqchar = reqchar;
7601 zeroreqcharflags = reqcharflags;
7602 zerofirstchar = firstchar;
7603 zerofirstcharflags = firstcharflags;
7604 groupsetfirstchar = FALSE;
7606 if (bravalue >= OP_ONCE)
7608 /* If we have not yet set a firstchar in this branch, take it from the
7609 subpattern, remembering that it was set here so that a repeat of more
7610 than one can replicate it as reqchar if necessary. If the subpattern has
7611 no firstchar, set "none" for the whole branch. In both cases, a zero
7612 repeat forces firstchar to "none". */
7614 if (firstcharflags == REQ_UNSET)
7616 if (subfirstcharflags >= 0)
7618 firstchar = subfirstchar;
7619 firstcharflags = subfirstcharflags;
7620 groupsetfirstchar = TRUE;
7622 else firstcharflags = REQ_NONE;
7623 zerofirstcharflags = REQ_NONE;
7626 /* If firstchar was previously set, convert the subpattern's firstchar
7627 into reqchar if there wasn't one, using the vary flag that was in
7628 existence beforehand. */
7630 else if (subfirstcharflags >= 0 && subreqcharflags < 0)
7632 subreqchar = subfirstchar;
7633 subreqcharflags = subfirstcharflags | tempreqvary;
7636 /* If the subpattern set a required byte (or set a first byte that isn't
7637 really the first byte - see above), set it. */
7639 if (subreqcharflags >= 0)
7641 reqchar = subreqchar;
7642 reqcharflags = subreqcharflags;
7646 /* For a forward assertion, we take the reqchar, if set. This can be
7647 helpful if the pattern that follows the assertion doesn't set a different
7648 char. For example, it's useful for /(?=abcde).+/. We can't set firstchar
7649 for an assertion, however because it leads to incorrect effect for patterns
7650 such as /(?=a)a.+/ when the "real" "a" would then become a reqchar instead
7651 of a firstchar. This is overcome by a scan at the end if there's no
7652 firstchar, looking for an asserted first char. */
7654 else if (bravalue == OP_ASSERT && subreqcharflags >= 0)
7656 reqchar = subreqchar;
7657 reqcharflags = subreqcharflags;
7659 break; /* End of processing '(' */
7662 /* ===================================================================*/
7663 /* Handle metasequences introduced by \. For ones like \d, the ESC_ values
7664 are arranged to be the negation of the corresponding OP_values in the
7665 default case when PCRE_UCP is not set. For the back references, the values
7666 are negative the reference number. Only back references and those types
7667 that consume a character may be repeated. We can test for values between
7668 ESC_b and ESC_Z for the latter; this may have to change if any new ones are
7671 case CHAR_BACKSLASH:
7673 escape = check_escape(&ptr, &ec, errorcodeptr, cd->bracount, options, FALSE);
7674 if (*errorcodeptr != 0) goto FAILED;
7676 if (escape == 0) /* The escape coded a single character */
7680 if (escape == ESC_Q) /* Handle start of quoted string */
7682 if (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
7683 ptr += 2; /* avoid empty string */
7688 if (escape == ESC_E) continue; /* Perl ignores an orphan \E */
7690 /* For metasequences that actually match a character, we disable the
7691 setting of a first character if it hasn't already been set. */
7693 if (firstcharflags == REQ_UNSET && escape > ESC_b && escape < ESC_Z)
7694 firstcharflags = REQ_NONE;
7696 /* Set values to reset to if this is followed by a zero repeat. */
7698 zerofirstchar = firstchar;
7699 zerofirstcharflags = firstcharflags;
7700 zeroreqchar = reqchar;
7701 zeroreqcharflags = reqcharflags;
7703 /* \g<name> or \g'name' is a subroutine call by name and \g<n> or \g'n'
7704 is a subroutine call by number (Oniguruma syntax). In fact, the value
7705 ESC_g is returned only for these cases. So we don't need to check for <
7706 or ' if the value is ESC_g. For the Perl syntax \g{n} the value is
7707 -n, and for the Perl syntax \g{name} the result is ESC_k (as
7708 that is a synonym for a named back reference). */
7710 if (escape == ESC_g)
7712 const pcre_uchar *p;
7715 save_hwm_offset = cd->hwm - cd->start_workspace; /* Normally this is set when '(' is read */
7716 terminator = (*(++ptr) == CHAR_LESS_THAN_SIGN)?
7717 CHAR_GREATER_THAN_SIGN : CHAR_APOSTROPHE;
7719 /* These two statements stop the compiler for warning about possibly
7720 unset variables caused by the jump to HANDLE_NUMERICAL_RECURSION. In
7721 fact, because we do the check for a number below, the paths that
7722 would actually be in error are never taken. */
7725 reset_bracount = FALSE;
7727 /* If it's not a signed or unsigned number, treat it as a name. */
7730 if (cf != CHAR_PLUS && cf != CHAR_MINUS && !IS_DIGIT(cf))
7733 goto NAMED_REF_OR_RECURSE;
7736 /* Signed or unsigned number (cf = ptr[1]) is known to be plus or minus
7740 while (IS_DIGIT(*p)) p++;
7741 if (*p != (pcre_uchar)terminator)
7743 *errorcodeptr = ERR57;
7747 goto HANDLE_NUMERICAL_RECURSION;
7750 /* \k<name> or \k'name' is a back reference by name (Perl syntax).
7751 We also support \k{name} (.NET syntax). */
7753 if (escape == ESC_k)
7755 if ((ptr[1] != CHAR_LESS_THAN_SIGN &&
7756 ptr[1] != CHAR_APOSTROPHE && ptr[1] != CHAR_LEFT_CURLY_BRACKET))
7758 *errorcodeptr = ERR69;
7762 terminator = (*(++ptr) == CHAR_LESS_THAN_SIGN)?
7763 CHAR_GREATER_THAN_SIGN : (*ptr == CHAR_APOSTROPHE)?
7764 CHAR_APOSTROPHE : CHAR_RIGHT_CURLY_BRACKET;
7765 goto NAMED_REF_OR_RECURSE;
7768 /* Back references are handled specially; must disable firstchar if
7769 not set to cope with cases like (?=(\w+))\1: which would otherwise set
7777 /* Come here from named backref handling when the reference is to a
7778 single group (i.e. not to a duplicated name. */
7781 if (firstcharflags == REQ_UNSET) firstcharflags = REQ_NONE;
7783 *code++ = ((options & PCRE_CASELESS) != 0)? OP_REFI : OP_REF;
7784 PUT2INC(code, 0, recno);
7785 cd->backref_map |= (recno < 32)? (1 << recno) : 1;
7786 if (recno > cd->top_backref) cd->top_backref = recno;
7788 /* Check to see if this back reference is recursive, that it, it
7789 is inside the group that it references. A flag is set so that the
7790 group can be made atomic. */
7792 for (oc = cd->open_caps; oc != NULL; oc = oc->next)
7794 if (oc->number == recno)
7802 /* So are Unicode property matches, if supported. */
7805 else if (escape == ESC_P || escape == ESC_p)
7808 unsigned int ptype = 0, pdata = 0;
7809 if (!get_ucp(&ptr, &negated, &ptype, &pdata, errorcodeptr))
7812 *code++ = ((escape == ESC_p) != negated)? OP_PROP : OP_NOTPROP;
7818 /* If Unicode properties are not supported, \X, \P, and \p are not
7821 else if (escape == ESC_X || escape == ESC_P || escape == ESC_p)
7823 *errorcodeptr = ERR45;
7828 /* For the rest (including \X when Unicode properties are supported), we
7829 can obtain the OP value by negating the escape value in the default
7830 situation when PCRE_UCP is not set. When it *is* set, we substitute
7831 Unicode property tests. Note that \b and \B do a one-character
7832 lookbehind, and \A also behaves as if it does. */
7836 if ((escape == ESC_b || escape == ESC_B || escape == ESC_A) &&
7837 cd->max_lookbehind == 0)
7838 cd->max_lookbehind = 1;
7840 if (escape >= ESC_DU && escape <= ESC_wu)
7842 nestptr = ptr + 1; /* Where to resume */
7843 ptr = substitutes[escape - ESC_DU] - 1; /* Just before substitute */
7847 /* In non-UTF-8 mode, we turn \C into OP_ALLANY instead of OP_ANYBYTE
7848 so that it works in DFA mode and in lookbehinds. */
7851 previous = (escape > ESC_b && escape < ESC_Z)? code : NULL;
7852 *code++ = (!utf && escape == ESC_C)? OP_ALLANY : escape;
7858 /* We have a data character whose value is in c. In UTF-8 mode it may have
7859 a value > 127. We set its representation in the length/buffer, and then
7860 handle it as a data character. */
7862 #if defined SUPPORT_UTF && !defined COMPILE_PCRE32
7863 if (utf && c > MAX_VALUE_FOR_SINGLE_CHAR)
7864 mclength = PRIV(ord2utf)(c, mcbuffer);
7875 /* ===================================================================*/
7876 /* Handle a literal character. It is guaranteed not to be whitespace or #
7877 when the extended flag is set. If we are in a UTF mode, it may be a
7878 multi-unit literal character. */
7886 if (utf && HAS_EXTRALEN(c))
7887 ACROSSCHAR(TRUE, ptr[1], mcbuffer[mclength++] = *(++ptr));
7890 /* At this point we have the character's bytes in mcbuffer, and the length
7891 in mclength. When not in UTF-8 mode, the length is always 1. */
7896 /* For caseless UTF-8 mode when UCP support is available, check whether
7897 this character has more than one other case. If so, generate a special
7898 OP_PROP item instead of OP_CHARI. */
7901 if (utf && (options & PCRE_CASELESS) != 0)
7903 GETCHAR(c, mcbuffer);
7904 if ((c = UCD_CASESET(c)) != 0)
7909 if (firstcharflags == REQ_UNSET)
7910 firstcharflags = zerofirstcharflags = REQ_NONE;
7916 /* Caseful matches, or not one of the multicase characters. */
7918 *code++ = ((options & PCRE_CASELESS) != 0)? OP_CHARI : OP_CHAR;
7919 for (c = 0; c < mclength; c++) *code++ = mcbuffer[c];
7921 /* Remember if \r or \n were seen */
7923 if (mcbuffer[0] == CHAR_CR || mcbuffer[0] == CHAR_NL)
7924 cd->external_flags |= PCRE_HASCRORLF;
7926 /* Set the first and required bytes appropriately. If no previous first
7927 byte, set it from this character, but revert to none on a zero repeat.
7928 Otherwise, leave the firstchar value alone, and don't change it on a zero
7931 if (firstcharflags == REQ_UNSET)
7933 zerofirstcharflags = REQ_NONE;
7934 zeroreqchar = reqchar;
7935 zeroreqcharflags = reqcharflags;
7937 /* If the character is more than one byte long, we can set firstchar
7938 only if it is not to be matched caselessly. */
7940 if (mclength == 1 || req_caseopt == 0)
7942 firstchar = mcbuffer[0] | req_caseopt;
7943 firstchar = mcbuffer[0];
7944 firstcharflags = req_caseopt;
7949 reqcharflags = cd->req_varyopt;
7952 else firstcharflags = reqcharflags = REQ_NONE;
7955 /* firstchar was previously set; we can set reqchar only if the length is
7956 1 or the matching is caseful. */
7960 zerofirstchar = firstchar;
7961 zerofirstcharflags = firstcharflags;
7962 zeroreqchar = reqchar;
7963 zeroreqcharflags = reqcharflags;
7964 if (mclength == 1 || req_caseopt == 0)
7967 reqcharflags = req_caseopt | cd->req_varyopt;
7971 break; /* End of literal character handling */
7973 } /* end of big loop */
7976 /* Control never reaches here by falling through, only by a goto for all the
7977 error states. Pass back the position in the pattern so that it can be displayed
7978 to the user for diagnosing the error. */
7987 /*************************************************
7988 * Compile sequence of alternatives *
7989 *************************************************/
7991 /* On entry, ptr is pointing past the bracket character, but on return it
7992 points to the closing bracket, or vertical bar, or end of string. The code
7993 variable is pointing at the byte into which the BRA operator has been stored.
7994 This function is used during the pre-compile phase when we are trying to find
7995 out the amount of memory needed, as well as during the real compile phase. The
7996 value of lengthptr distinguishes the two phases.
7999 options option bits, including any changes for this subpattern
8000 codeptr -> the address of the current code pointer
8001 ptrptr -> the address of the current pattern pointer
8002 errorcodeptr -> pointer to error code variable
8003 lookbehind TRUE if this is a lookbehind assertion
8004 reset_bracount TRUE to reset the count for each branch
8005 skipbytes skip this many bytes at start (for brackets and OP_COND)
8006 cond_depth depth of nesting for conditional subpatterns
8007 firstcharptr place to put the first required character
8008 firstcharflagsptr place to put the first character flags, or a negative number
8009 reqcharptr place to put the last required character
8010 reqcharflagsptr place to put the last required character flags, or a negative number
8011 bcptr pointer to the chain of currently open branches
8012 cd points to the data block with tables pointers etc.
8013 lengthptr NULL during the real compile phase
8014 points to length accumulator during pre-compile phase
8016 Returns: TRUE on success
8020 compile_regex(int options, pcre_uchar **codeptr, const pcre_uchar **ptrptr,
8021 int *errorcodeptr, BOOL lookbehind, BOOL reset_bracount, int skipbytes,
8023 pcre_uint32 *firstcharptr, pcre_int32 *firstcharflagsptr,
8024 pcre_uint32 *reqcharptr, pcre_int32 *reqcharflagsptr,
8025 branch_chain *bcptr, compile_data *cd, int *lengthptr)
8027 const pcre_uchar *ptr = *ptrptr;
8028 pcre_uchar *code = *codeptr;
8029 pcre_uchar *last_branch = code;
8030 pcre_uchar *start_bracket = code;
8031 pcre_uchar *reverse_count = NULL;
8032 open_capitem capitem;
8034 pcre_uint32 firstchar, reqchar;
8035 pcre_int32 firstcharflags, reqcharflags;
8036 pcre_uint32 branchfirstchar, branchreqchar;
8037 pcre_int32 branchfirstcharflags, branchreqcharflags;
8039 unsigned int orig_bracount;
8040 unsigned int max_bracount;
8042 size_t save_hwm_offset;
8044 /* If set, call the external function that checks for stack availability. */
8046 if (PUBL(stack_guard) != NULL && PUBL(stack_guard)())
8048 *errorcodeptr= ERR85;
8052 /* Miscellaneous initialization */
8055 bc.current_branch = code;
8057 firstchar = reqchar = 0;
8058 firstcharflags = reqcharflags = REQ_UNSET;
8060 save_hwm_offset = cd->hwm - cd->start_workspace;
8062 /* Accumulate the length for use in the pre-compile phase. Start with the
8063 length of the BRA and KET and any extra bytes that are required at the
8064 beginning. We accumulate in a local variable to save frequent testing of
8065 lenthptr for NULL. We cannot do this by looking at the value of code at the
8066 start and end of each alternative, because compiled items are discarded during
8067 the pre-compile phase so that the work space is not exceeded. */
8069 length = 2 + 2*LINK_SIZE + skipbytes;
8071 /* WARNING: If the above line is changed for any reason, you must also change
8072 the code that abstracts option settings at the start of the pattern and makes
8073 them global. It tests the value of length for (2 + 2*LINK_SIZE) in the
8074 pre-compile phase to find out whether anything has yet been compiled or not. */
8076 /* If this is a capturing subpattern, add to the chain of open capturing items
8077 so that we can detect them if (*ACCEPT) is encountered. This is also used to
8078 detect groups that contain recursive back references to themselves. Note that
8079 only OP_CBRA need be tested here; changing this opcode to one of its variants,
8080 e.g. OP_SCBRAPOS, happens later, after the group has been compiled. */
8082 if (*code == OP_CBRA)
8084 capnumber = GET2(code, 1 + LINK_SIZE);
8085 capitem.number = capnumber;
8086 capitem.next = cd->open_caps;
8087 capitem.flag = FALSE;
8088 cd->open_caps = &capitem;
8091 /* Offset is set zero to mark that this bracket is still open */
8094 code += 1 + LINK_SIZE + skipbytes;
8096 /* Loop for each alternative branch */
8098 orig_bracount = max_bracount = cd->bracount;
8101 /* For a (?| group, reset the capturing bracket count so that each branch
8102 uses the same numbers. */
8104 if (reset_bracount) cd->bracount = orig_bracount;
8106 /* Set up dummy OP_REVERSE if lookbehind assertion */
8110 *code++ = OP_REVERSE;
8111 reverse_count = code;
8113 length += 1 + LINK_SIZE;
8116 /* Now compile the branch; in the pre-compile phase its length gets added
8119 if (!compile_branch(&options, &code, &ptr, errorcodeptr, &branchfirstchar,
8120 &branchfirstcharflags, &branchreqchar, &branchreqcharflags, &bc,
8121 cond_depth, cd, (lengthptr == NULL)? NULL : &length))
8127 /* Keep the highest bracket count in case (?| was used and some branch
8128 has fewer than the rest. */
8130 if (cd->bracount > max_bracount) max_bracount = cd->bracount;
8132 /* In the real compile phase, there is some post-processing to be done. */
8134 if (lengthptr == NULL)
8136 /* If this is the first branch, the firstchar and reqchar values for the
8137 branch become the values for the regex. */
8139 if (*last_branch != OP_ALT)
8141 firstchar = branchfirstchar;
8142 firstcharflags = branchfirstcharflags;
8143 reqchar = branchreqchar;
8144 reqcharflags = branchreqcharflags;
8147 /* If this is not the first branch, the first char and reqchar have to
8148 match the values from all the previous branches, except that if the
8149 previous value for reqchar didn't have REQ_VARY set, it can still match,
8150 and we set REQ_VARY for the regex. */
8154 /* If we previously had a firstchar, but it doesn't match the new branch,
8155 we have to abandon the firstchar for the regex, but if there was
8156 previously no reqchar, it takes on the value of the old firstchar. */
8158 if (firstcharflags >= 0 &&
8159 (firstcharflags != branchfirstcharflags || firstchar != branchfirstchar))
8161 if (reqcharflags < 0)
8163 reqchar = firstchar;
8164 reqcharflags = firstcharflags;
8166 firstcharflags = REQ_NONE;
8169 /* If we (now or from before) have no firstchar, a firstchar from the
8170 branch becomes a reqchar if there isn't a branch reqchar. */
8172 if (firstcharflags < 0 && branchfirstcharflags >= 0 && branchreqcharflags < 0)
8174 branchreqchar = branchfirstchar;
8175 branchreqcharflags = branchfirstcharflags;
8178 /* Now ensure that the reqchars match */
8180 if (((reqcharflags & ~REQ_VARY) != (branchreqcharflags & ~REQ_VARY)) ||
8181 reqchar != branchreqchar)
8182 reqcharflags = REQ_NONE;
8185 reqchar = branchreqchar;
8186 reqcharflags |= branchreqcharflags; /* To "or" REQ_VARY */
8190 /* If lookbehind, check that this branch matches a fixed-length string, and
8191 put the length into the OP_REVERSE item. Temporarily mark the end of the
8192 branch with OP_END. If the branch contains OP_RECURSE, the result is -3
8193 because there may be forward references that we can't check here. Set a
8194 flag to cause another lookbehind check at the end. Why not do it all at the
8195 end? Because common, erroneous checks are picked up here and the offset of
8196 the problem can be shown. */
8202 fixed_length = find_fixedlength(last_branch, (options & PCRE_UTF8) != 0,
8204 DPRINTF(("fixed length = %d\n", fixed_length));
8205 if (fixed_length == -3)
8207 cd->check_lookbehind = TRUE;
8209 else if (fixed_length < 0)
8211 *errorcodeptr = (fixed_length == -2)? ERR36 :
8212 (fixed_length == -4)? ERR70: ERR25;
8218 if (fixed_length > cd->max_lookbehind)
8219 cd->max_lookbehind = fixed_length;
8220 PUT(reverse_count, 0, fixed_length);
8225 /* Reached end of expression, either ')' or end of pattern. In the real
8226 compile phase, go back through the alternative branches and reverse the chain
8227 of offsets, with the field in the BRA item now becoming an offset to the
8228 first alternative. If there are no alternatives, it points to the end of the
8229 group. The length in the terminating ket is always the length of the whole
8230 bracketed item. Return leaving the pointer at the terminating char. */
8232 if (*ptr != CHAR_VERTICAL_LINE)
8234 if (lengthptr == NULL)
8236 int branch_length = (int)(code - last_branch);
8239 int prev_length = GET(last_branch, 1);
8240 PUT(last_branch, 1, branch_length);
8241 branch_length = prev_length;
8242 last_branch -= branch_length;
8244 while (branch_length > 0);
8247 /* Fill in the ket */
8250 PUT(code, 1, (int)(code - start_bracket));
8251 code += 1 + LINK_SIZE;
8253 /* If it was a capturing subpattern, check to see if it contained any
8254 recursive back references. If so, we must wrap it in atomic brackets.
8255 Because we are moving code along, we must ensure that any pending recursive
8256 references are updated. In any event, remove the block from the chain. */
8260 if (cd->open_caps->flag)
8263 adjust_recurse(start_bracket, 1 + LINK_SIZE,
8264 (options & PCRE_UTF8) != 0, cd, save_hwm_offset);
8265 memmove(start_bracket + 1 + LINK_SIZE, start_bracket,
8266 IN_UCHARS(code - start_bracket));
8267 *start_bracket = OP_ONCE;
8268 code += 1 + LINK_SIZE;
8269 PUT(start_bracket, 1, (int)(code - start_bracket));
8271 PUT(code, 1, (int)(code - start_bracket));
8272 code += 1 + LINK_SIZE;
8273 length += 2 + 2*LINK_SIZE;
8275 cd->open_caps = cd->open_caps->next;
8278 /* Retain the highest bracket number, in case resetting was used. */
8280 cd->bracount = max_bracount;
8282 /* Set values to pass back */
8286 *firstcharptr = firstchar;
8287 *firstcharflagsptr = firstcharflags;
8288 *reqcharptr = reqchar;
8289 *reqcharflagsptr = reqcharflags;
8290 if (lengthptr != NULL)
8292 if (OFLOW_MAX - *lengthptr < length)
8294 *errorcodeptr = ERR20;
8297 *lengthptr += length;
8302 /* Another branch follows. In the pre-compile phase, we can move the code
8303 pointer back to where it was for the start of the first branch. (That is,
8304 pretend that each branch is the only one.)
8306 In the real compile phase, insert an ALT node. Its length field points back
8307 to the previous branch while the bracket remains open. At the end the chain
8308 is reversed. It's done like this so that the start of the bracket has a
8309 zero offset until it is closed, making it possible to detect recursion. */
8311 if (lengthptr != NULL)
8313 code = *codeptr + 1 + LINK_SIZE + skipbytes;
8314 length += 1 + LINK_SIZE;
8319 PUT(code, 1, (int)(code - last_branch));
8320 bc.current_branch = last_branch = code;
8321 code += 1 + LINK_SIZE;
8326 /* Control never reaches here */
8332 /*************************************************
8333 * Check for anchored expression *
8334 *************************************************/
8336 /* Try to find out if this is an anchored regular expression. Consider each
8337 alternative branch. If they all start with OP_SOD or OP_CIRC, or with a bracket
8338 all of whose alternatives start with OP_SOD or OP_CIRC (recurse ad lib), then
8339 it's anchored. However, if this is a multiline pattern, then only OP_SOD will
8340 be found, because ^ generates OP_CIRCM in that mode.
8342 We can also consider a regex to be anchored if OP_SOM starts all its branches.
8343 This is the code for \G, which means "match at start of match position, taking
8344 into account the match offset".
8346 A branch is also implicitly anchored if it starts with .* and DOTALL is set,
8347 because that will try the rest of the pattern at all possible matching points,
8348 so there is no point trying again.... er ....
8350 .... except when the .* appears inside capturing parentheses, and there is a
8351 subsequent back reference to those parentheses. We haven't enough information
8352 to catch that case precisely.
8354 At first, the best we could do was to detect when .* was in capturing brackets
8355 and the highest back reference was greater than or equal to that level.
8356 However, by keeping a bitmap of the first 31 back references, we can catch some
8357 of the more common cases more precisely.
8359 ... A second exception is when the .* appears inside an atomic group, because
8360 this prevents the number of characters it matches from being adjusted.
8363 code points to start of expression (the bracket)
8364 bracket_map a bitmap of which brackets we are inside while testing; this
8365 handles up to substring 31; after that we just have to take
8366 the less precise approach
8367 cd points to the compile data block
8368 atomcount atomic group level
8370 Returns: TRUE or FALSE
8374 is_anchored(register const pcre_uchar *code, unsigned int bracket_map,
8375 compile_data *cd, int atomcount)
8378 const pcre_uchar *scode = first_significant_code(
8379 code + PRIV(OP_lengths)[*code], FALSE);
8380 register int op = *scode;
8382 /* Non-capturing brackets */
8384 if (op == OP_BRA || op == OP_BRAPOS ||
8385 op == OP_SBRA || op == OP_SBRAPOS)
8387 if (!is_anchored(scode, bracket_map, cd, atomcount)) return FALSE;
8390 /* Capturing brackets */
8392 else if (op == OP_CBRA || op == OP_CBRAPOS ||
8393 op == OP_SCBRA || op == OP_SCBRAPOS)
8395 int n = GET2(scode, 1+LINK_SIZE);
8396 int new_map = bracket_map | ((n < 32)? (1 << n) : 1);
8397 if (!is_anchored(scode, new_map, cd, atomcount)) return FALSE;
8400 /* Positive forward assertions and conditions */
8402 else if (op == OP_ASSERT || op == OP_COND)
8404 if (!is_anchored(scode, bracket_map, cd, atomcount)) return FALSE;
8409 else if (op == OP_ONCE || op == OP_ONCE_NC)
8411 if (!is_anchored(scode, bracket_map, cd, atomcount + 1))
8415 /* .* is not anchored unless DOTALL is set (which generates OP_ALLANY) and
8416 it isn't in brackets that are or may be referenced or inside an atomic
8419 else if ((op == OP_TYPESTAR || op == OP_TYPEMINSTAR ||
8420 op == OP_TYPEPOSSTAR))
8422 if (scode[1] != OP_ALLANY || (bracket_map & cd->backref_map) != 0 ||
8423 atomcount > 0 || cd->had_pruneorskip)
8427 /* Check for explicit anchoring */
8429 else if (op != OP_SOD && op != OP_SOM && op != OP_CIRC) return FALSE;
8431 code += GET(code, 1);
8433 while (*code == OP_ALT); /* Loop for each alternative */
8439 /*************************************************
8440 * Check for starting with ^ or .* *
8441 *************************************************/
8443 /* This is called to find out if every branch starts with ^ or .* so that
8444 "first char" processing can be done to speed things up in multiline
8445 matching and for non-DOTALL patterns that start with .* (which must start at
8446 the beginning or after \n). As in the case of is_anchored() (see above), we
8447 have to take account of back references to capturing brackets that contain .*
8448 because in that case we can't make the assumption. Also, the appearance of .*
8449 inside atomic brackets or in a pattern that contains *PRUNE or *SKIP does not
8450 count, because once again the assumption no longer holds.
8453 code points to start of expression (the bracket)
8454 bracket_map a bitmap of which brackets we are inside while testing; this
8455 handles up to substring 31; after that we just have to take
8456 the less precise approach
8457 cd points to the compile data
8458 atomcount atomic group level
8460 Returns: TRUE or FALSE
8464 is_startline(const pcre_uchar *code, unsigned int bracket_map,
8465 compile_data *cd, int atomcount)
8468 const pcre_uchar *scode = first_significant_code(
8469 code + PRIV(OP_lengths)[*code], FALSE);
8470 register int op = *scode;
8472 /* If we are at the start of a conditional assertion group, *both* the
8473 conditional assertion *and* what follows the condition must satisfy the test
8474 for start of line. Other kinds of condition fail. Note that there may be an
8475 auto-callout at the start of a condition. */
8479 scode += 1 + LINK_SIZE;
8480 if (*scode == OP_CALLOUT) scode += PRIV(OP_lengths)[OP_CALLOUT];
8490 default: /* Assertion */
8491 if (!is_startline(scode, bracket_map, cd, atomcount)) return FALSE;
8492 do scode += GET(scode, 1); while (*scode == OP_ALT);
8493 scode += 1 + LINK_SIZE;
8496 scode = first_significant_code(scode, FALSE);
8500 /* Non-capturing brackets */
8502 if (op == OP_BRA || op == OP_BRAPOS ||
8503 op == OP_SBRA || op == OP_SBRAPOS)
8505 if (!is_startline(scode, bracket_map, cd, atomcount)) return FALSE;
8508 /* Capturing brackets */
8510 else if (op == OP_CBRA || op == OP_CBRAPOS ||
8511 op == OP_SCBRA || op == OP_SCBRAPOS)
8513 int n = GET2(scode, 1+LINK_SIZE);
8514 int new_map = bracket_map | ((n < 32)? (1 << n) : 1);
8515 if (!is_startline(scode, new_map, cd, atomcount)) return FALSE;
8518 /* Positive forward assertions */
8520 else if (op == OP_ASSERT)
8522 if (!is_startline(scode, bracket_map, cd, atomcount)) return FALSE;
8525 /* Atomic brackets */
8527 else if (op == OP_ONCE || op == OP_ONCE_NC)
8529 if (!is_startline(scode, bracket_map, cd, atomcount + 1)) return FALSE;
8532 /* .* means "start at start or after \n" if it isn't in atomic brackets or
8533 brackets that may be referenced, as long as the pattern does not contain
8534 *PRUNE or *SKIP, because these break the feature. Consider, for example,
8535 /.*?a(*PRUNE)b/ with the subject "aab", which matches "ab", i.e. not at the
8538 else if (op == OP_TYPESTAR || op == OP_TYPEMINSTAR || op == OP_TYPEPOSSTAR)
8540 if (scode[1] != OP_ANY || (bracket_map & cd->backref_map) != 0 ||
8541 atomcount > 0 || cd->had_pruneorskip)
8545 /* Check for explicit circumflex; anything else gives a FALSE result. Note
8546 in particular that this includes atomic brackets OP_ONCE and OP_ONCE_NC
8547 because the number of characters matched by .* cannot be adjusted inside
8550 else if (op != OP_CIRC && op != OP_CIRCM) return FALSE;
8552 /* Move on to the next alternative */
8554 code += GET(code, 1);
8556 while (*code == OP_ALT); /* Loop for each alternative */
8562 /*************************************************
8563 * Check for asserted fixed first char *
8564 *************************************************/
8566 /* During compilation, the "first char" settings from forward assertions are
8567 discarded, because they can cause conflicts with actual literals that follow.
8568 However, if we end up without a first char setting for an unanchored pattern,
8569 it is worth scanning the regex to see if there is an initial asserted first
8570 char. If all branches start with the same asserted char, or with a
8571 non-conditional bracket all of whose alternatives start with the same asserted
8572 char (recurse ad lib), then we return that char, with the flags set to zero or
8573 REQ_CASELESS; otherwise return zero with REQ_NONE in the flags.
8576 code points to start of expression (the bracket)
8577 flags points to the first char flags, or to REQ_NONE
8578 inassert TRUE if in an assertion
8580 Returns: the fixed first char, or 0 with REQ_NONE in flags
8584 find_firstassertedchar(const pcre_uchar *code, pcre_int32 *flags,
8587 register pcre_uint32 c = 0;
8588 int cflags = REQ_NONE;
8594 int xl = (*code == OP_CBRA || *code == OP_SCBRA ||
8595 *code == OP_CBRAPOS || *code == OP_SCBRAPOS)? IMM2_SIZE:0;
8596 const pcre_uchar *scode = first_significant_code(code + 1+LINK_SIZE + xl,
8598 register pcre_uchar op = *scode;
8614 d = find_firstassertedchar(scode, &dflags, op == OP_ASSERT);
8617 if (cflags < 0) { c = d; cflags = dflags; } else if (c != d || cflags != dflags) return 0;
8628 if (!inassert) return 0;
8629 if (cflags < 0) { c = scode[1]; cflags = 0; }
8630 else if (c != scode[1]) return 0;
8641 if (!inassert) return 0;
8642 if (cflags < 0) { c = scode[1]; cflags = REQ_CASELESS; }
8643 else if (c != scode[1]) return 0;
8647 code += GET(code, 1);
8649 while (*code == OP_ALT);
8657 /*************************************************
8658 * Add an entry to the name/number table *
8659 *************************************************/
8661 /* This function is called between compiling passes to add an entry to the
8662 name/number table, maintaining alphabetical order. Checking for permitted
8663 and forbidden duplicates has already been done.
8666 cd the compile data block
8667 name the name to add
8668 length the length of the name
8669 groupno the group number
8675 add_name(compile_data *cd, const pcre_uchar *name, int length,
8676 unsigned int groupno)
8679 pcre_uchar *slot = cd->name_table;
8681 for (i = 0; i < cd->names_found; i++)
8683 int crc = memcmp(name, slot+IMM2_SIZE, IN_UCHARS(length));
8684 if (crc == 0 && slot[IMM2_SIZE+length] != 0)
8685 crc = -1; /* Current name is a substring */
8687 /* Make space in the table and break the loop for an earlier name. For a
8688 duplicate or later name, carry on. We do this for duplicates so that in the
8689 simple case (when ?(| is not used) they are in order of their numbers. In all
8690 cases they are in the order in which they appear in the pattern. */
8694 memmove(slot + cd->name_entry_size, slot,
8695 IN_UCHARS((cd->names_found - i) * cd->name_entry_size));
8699 /* Continue the loop for a later or duplicate name */
8701 slot += cd->name_entry_size;
8704 PUT2(slot, 0, groupno);
8705 memcpy(slot + IMM2_SIZE, name, IN_UCHARS(length));
8706 slot[IMM2_SIZE + length] = 0;
8712 /*************************************************
8713 * Compile a Regular Expression *
8714 *************************************************/
8716 /* This function takes a string and returns a pointer to a block of store
8717 holding a compiled version of the expression. The original API for this
8718 function had no error code return variable; it is retained for backwards
8719 compatibility. The new function is given a new name.
8722 pattern the regular expression
8723 options various option bits
8724 errorcodeptr pointer to error code variable (pcre_compile2() only)
8725 can be NULL if you don't want a code value
8726 errorptr pointer to pointer to error text
8727 erroroffset ptr offset in pattern where error was detected
8728 tables pointer to character tables or NULL
8730 Returns: pointer to compiled data block, or NULL on error,
8731 with errorptr and erroroffset set
8734 #if defined COMPILE_PCRE8
8735 PCRE_EXP_DEFN pcre * PCRE_CALL_CONVENTION
8736 pcre_compile(const char *pattern, int options, const char **errorptr,
8737 int *erroroffset, const unsigned char *tables)
8738 #elif defined COMPILE_PCRE16
8739 PCRE_EXP_DEFN pcre16 * PCRE_CALL_CONVENTION
8740 pcre16_compile(PCRE_SPTR16 pattern, int options, const char **errorptr,
8741 int *erroroffset, const unsigned char *tables)
8742 #elif defined COMPILE_PCRE32
8743 PCRE_EXP_DEFN pcre32 * PCRE_CALL_CONVENTION
8744 pcre32_compile(PCRE_SPTR32 pattern, int options, const char **errorptr,
8745 int *erroroffset, const unsigned char *tables)
8748 #if defined COMPILE_PCRE8
8749 return pcre_compile2(pattern, options, NULL, errorptr, erroroffset, tables);
8750 #elif defined COMPILE_PCRE16
8751 return pcre16_compile2(pattern, options, NULL, errorptr, erroroffset, tables);
8752 #elif defined COMPILE_PCRE32
8753 return pcre32_compile2(pattern, options, NULL, errorptr, erroroffset, tables);
8758 #if defined COMPILE_PCRE8
8759 PCRE_EXP_DEFN pcre * PCRE_CALL_CONVENTION
8760 pcre_compile2(const char *pattern, int options, int *errorcodeptr,
8761 const char **errorptr, int *erroroffset, const unsigned char *tables)
8762 #elif defined COMPILE_PCRE16
8763 PCRE_EXP_DEFN pcre16 * PCRE_CALL_CONVENTION
8764 pcre16_compile2(PCRE_SPTR16 pattern, int options, int *errorcodeptr,
8765 const char **errorptr, int *erroroffset, const unsigned char *tables)
8766 #elif defined COMPILE_PCRE32
8767 PCRE_EXP_DEFN pcre32 * PCRE_CALL_CONVENTION
8768 pcre32_compile2(PCRE_SPTR32 pattern, int options, int *errorcodeptr,
8769 const char **errorptr, int *erroroffset, const unsigned char *tables)
8773 int length = 1; /* For final END opcode */
8774 pcre_int32 firstcharflags, reqcharflags;
8775 pcre_uint32 firstchar, reqchar;
8776 pcre_uint32 limit_match = PCRE_UINT32_MAX;
8777 pcre_uint32 limit_recursion = PCRE_UINT32_MAX;
8780 int skipatstart = 0;
8782 BOOL never_utf = FALSE;
8785 const pcre_uchar *codestart;
8786 const pcre_uchar *ptr;
8787 compile_data compile_block;
8788 compile_data *cd = &compile_block;
8790 /* This space is used for "compiling" into during the first phase, when we are
8791 computing the amount of memory that is needed. Compiled items are thrown away
8792 as soon as possible, so that a fairly large buffer should be sufficient for
8793 this purpose. The same space is used in the second phase for remembering where
8794 to fill in forward references to subpatterns. That may overflow, in which case
8795 new memory is obtained from malloc(). */
8797 pcre_uchar cworkspace[COMPILE_WORK_SIZE];
8799 /* This vector is used for remembering name groups during the pre-compile. In a
8800 similar way to cworkspace, it can be expanded using malloc() if necessary. */
8802 named_group named_groups[NAMED_GROUP_LIST_SIZE];
8804 /* Set this early so that early errors get offset 0. */
8806 ptr = (const pcre_uchar *)pattern;
8808 /* We can't pass back an error message if errorptr is NULL; I guess the best we
8809 can do is just return NULL, but we can set a code value if there is a code
8812 if (errorptr == NULL)
8814 if (errorcodeptr != NULL) *errorcodeptr = 99;
8819 if (errorcodeptr != NULL) *errorcodeptr = ERR0;
8821 /* However, we can give a message for this error */
8823 if (erroroffset == NULL)
8826 goto PCRE_EARLY_ERROR_RETURN2;
8831 /* Set up pointers to the individual character tables */
8833 if (tables == NULL) tables = PRIV(default_tables);
8834 cd->lcc = tables + lcc_offset;
8835 cd->fcc = tables + fcc_offset;
8836 cd->cbits = tables + cbits_offset;
8837 cd->ctypes = tables + ctypes_offset;
8839 /* Check that all undefined public option bits are zero */
8841 if ((options & ~PUBLIC_COMPILE_OPTIONS) != 0)
8844 goto PCRE_EARLY_ERROR_RETURN;
8847 /* If PCRE_NEVER_UTF is set, remember it. */
8849 if ((options & PCRE_NEVER_UTF) != 0) never_utf = TRUE;
8851 /* Check for global one-time settings at the start of the pattern, and remember
8852 the offset for later. */
8854 cd->external_flags = 0; /* Initialize here for LIMIT_MATCH/RECURSION */
8856 while (ptr[skipatstart] == CHAR_LEFT_PARENTHESIS &&
8857 ptr[skipatstart+1] == CHAR_ASTERISK)
8862 /* For completeness and backward compatibility, (*UTFn) is supported in the
8863 relevant libraries, but (*UTF) is generic and always supported. Note that
8864 PCRE_UTF8 == PCRE_UTF16 == PCRE_UTF32. */
8866 #ifdef COMPILE_PCRE8
8867 if (STRNCMP_UC_C8(ptr+skipatstart+2, STRING_UTF8_RIGHTPAR, 5) == 0)
8868 { skipatstart += 7; options |= PCRE_UTF8; continue; }
8870 #ifdef COMPILE_PCRE16
8871 if (STRNCMP_UC_C8(ptr+skipatstart+2, STRING_UTF16_RIGHTPAR, 6) == 0)
8872 { skipatstart += 8; options |= PCRE_UTF16; continue; }
8874 #ifdef COMPILE_PCRE32
8875 if (STRNCMP_UC_C8(ptr+skipatstart+2, STRING_UTF32_RIGHTPAR, 6) == 0)
8876 { skipatstart += 8; options |= PCRE_UTF32; continue; }
8879 else if (STRNCMP_UC_C8(ptr+skipatstart+2, STRING_UTF_RIGHTPAR, 4) == 0)
8880 { skipatstart += 6; options |= PCRE_UTF8; continue; }
8881 else if (STRNCMP_UC_C8(ptr+skipatstart+2, STRING_UCP_RIGHTPAR, 4) == 0)
8882 { skipatstart += 6; options |= PCRE_UCP; continue; }
8883 else if (STRNCMP_UC_C8(ptr+skipatstart+2, STRING_NO_AUTO_POSSESS_RIGHTPAR, 16) == 0)
8884 { skipatstart += 18; options |= PCRE_NO_AUTO_POSSESS; continue; }
8885 else if (STRNCMP_UC_C8(ptr+skipatstart+2, STRING_NO_START_OPT_RIGHTPAR, 13) == 0)
8886 { skipatstart += 15; options |= PCRE_NO_START_OPTIMIZE; continue; }
8888 else if (STRNCMP_UC_C8(ptr+skipatstart+2, STRING_LIMIT_MATCH_EQ, 12) == 0)
8891 int p = skipatstart + 14;
8892 while (isdigit(ptr[p]))
8894 if (c > PCRE_UINT32_MAX / 10 - 1) break; /* Integer overflow */
8895 c = c*10 + ptr[p++] - CHAR_0;
8897 if (ptr[p++] != CHAR_RIGHT_PARENTHESIS) break;
8898 if (c < limit_match)
8901 cd->external_flags |= PCRE_MLSET;
8907 else if (STRNCMP_UC_C8(ptr+skipatstart+2, STRING_LIMIT_RECURSION_EQ, 16) == 0)
8910 int p = skipatstart + 18;
8911 while (isdigit(ptr[p]))
8913 if (c > PCRE_UINT32_MAX / 10 - 1) break; /* Integer overflow check */
8914 c = c*10 + ptr[p++] - CHAR_0;
8916 if (ptr[p++] != CHAR_RIGHT_PARENTHESIS) break;
8917 if (c < limit_recursion)
8919 limit_recursion = c;
8920 cd->external_flags |= PCRE_RLSET;
8926 if (STRNCMP_UC_C8(ptr+skipatstart+2, STRING_CR_RIGHTPAR, 3) == 0)
8927 { skipatstart += 5; newnl = PCRE_NEWLINE_CR; }
8928 else if (STRNCMP_UC_C8(ptr+skipatstart+2, STRING_LF_RIGHTPAR, 3) == 0)
8929 { skipatstart += 5; newnl = PCRE_NEWLINE_LF; }
8930 else if (STRNCMP_UC_C8(ptr+skipatstart+2, STRING_CRLF_RIGHTPAR, 5) == 0)
8931 { skipatstart += 7; newnl = PCRE_NEWLINE_CR + PCRE_NEWLINE_LF; }
8932 else if (STRNCMP_UC_C8(ptr+skipatstart+2, STRING_ANY_RIGHTPAR, 4) == 0)
8933 { skipatstart += 6; newnl = PCRE_NEWLINE_ANY; }
8934 else if (STRNCMP_UC_C8(ptr+skipatstart+2, STRING_ANYCRLF_RIGHTPAR, 8) == 0)
8935 { skipatstart += 10; newnl = PCRE_NEWLINE_ANYCRLF; }
8937 else if (STRNCMP_UC_C8(ptr+skipatstart+2, STRING_BSR_ANYCRLF_RIGHTPAR, 12) == 0)
8938 { skipatstart += 14; newbsr = PCRE_BSR_ANYCRLF; }
8939 else if (STRNCMP_UC_C8(ptr+skipatstart+2, STRING_BSR_UNICODE_RIGHTPAR, 12) == 0)
8940 { skipatstart += 14; newbsr = PCRE_BSR_UNICODE; }
8943 options = (options & ~PCRE_NEWLINE_BITS) | newnl;
8944 else if (newbsr != 0)
8945 options = (options & ~(PCRE_BSR_ANYCRLF|PCRE_BSR_UNICODE)) | newbsr;
8949 /* PCRE_UTF(16|32) have the same value as PCRE_UTF8. */
8950 utf = (options & PCRE_UTF8) != 0;
8951 if (utf && never_utf)
8954 goto PCRE_EARLY_ERROR_RETURN2;
8957 /* Can't support UTF unless PCRE has been compiled to include the code. The
8958 return of an error code from PRIV(valid_utf)() is a new feature, introduced in
8959 release 8.13. It is passed back from pcre_[dfa_]exec(), but at the moment is
8963 if (utf && (options & PCRE_NO_UTF8_CHECK) == 0 &&
8964 (errorcode = PRIV(valid_utf)((PCRE_PUCHAR)pattern, -1, erroroffset)) != 0)
8966 #if defined COMPILE_PCRE8
8968 #elif defined COMPILE_PCRE16
8970 #elif defined COMPILE_PCRE32
8973 goto PCRE_EARLY_ERROR_RETURN2;
8979 goto PCRE_EARLY_ERROR_RETURN;
8983 /* Can't support UCP unless PCRE has been compiled to include the code. */
8986 if ((options & PCRE_UCP) != 0)
8989 goto PCRE_EARLY_ERROR_RETURN;
8993 /* Check validity of \R options. */
8995 if ((options & (PCRE_BSR_ANYCRLF|PCRE_BSR_UNICODE)) ==
8996 (PCRE_BSR_ANYCRLF|PCRE_BSR_UNICODE))
8999 goto PCRE_EARLY_ERROR_RETURN;
9002 /* Handle different types of newline. The three bits give seven cases. The
9003 current code allows for fixed one- or two-byte sequences, plus "any" and
9006 switch (options & PCRE_NEWLINE_BITS)
9008 case 0: newline = NEWLINE; break; /* Build-time default */
9009 case PCRE_NEWLINE_CR: newline = CHAR_CR; break;
9010 case PCRE_NEWLINE_LF: newline = CHAR_NL; break;
9011 case PCRE_NEWLINE_CR+
9012 PCRE_NEWLINE_LF: newline = (CHAR_CR << 8) | CHAR_NL; break;
9013 case PCRE_NEWLINE_ANY: newline = -1; break;
9014 case PCRE_NEWLINE_ANYCRLF: newline = -2; break;
9015 default: errorcode = ERR56; goto PCRE_EARLY_ERROR_RETURN;
9020 cd->nltype = NLTYPE_ANYCRLF;
9022 else if (newline < 0)
9024 cd->nltype = NLTYPE_ANY;
9028 cd->nltype = NLTYPE_FIXED;
9032 cd->nl[0] = (newline >> 8) & 255;
9033 cd->nl[1] = newline & 255;
9038 cd->nl[0] = newline;
9042 /* Maximum back reference and backref bitmap. The bitmap records up to 31 back
9043 references to help in deciding whether (.*) can be treated as anchored or not.
9046 cd->top_backref = 0;
9047 cd->backref_map = 0;
9049 /* Reflect pattern for debugging output */
9051 DPRINTF(("------------------------------------------------------------------\n"));
9053 print_puchar(stdout, (PCRE_PUCHAR)pattern);
9057 /* Pretend to compile the pattern while actually just accumulating the length
9058 of memory required. This behaviour is triggered by passing a non-NULL final
9059 argument to compile_regex(). We pass a block of workspace (cworkspace) for it
9060 to compile parts of the pattern into; the compiled code is discarded when it is
9061 no longer needed, so hopefully this workspace will never overflow, though there
9062 is a test for its doing so. */
9064 cd->bracount = cd->final_bracount = 0;
9065 cd->names_found = 0;
9066 cd->name_entry_size = 0;
9067 cd->name_table = NULL;
9068 cd->dupnames = FALSE;
9069 cd->namedrefcount = 0;
9070 cd->start_code = cworkspace;
9071 cd->hwm = cworkspace;
9072 cd->start_workspace = cworkspace;
9073 cd->workspace_size = COMPILE_WORK_SIZE;
9074 cd->named_groups = named_groups;
9075 cd->named_group_list_size = NAMED_GROUP_LIST_SIZE;
9076 cd->start_pattern = (const pcre_uchar *)pattern;
9077 cd->end_pattern = (const pcre_uchar *)(pattern + STRLEN_UC((const pcre_uchar *)pattern));
9078 cd->req_varyopt = 0;
9079 cd->parens_depth = 0;
9080 cd->assert_depth = 0;
9081 cd->max_lookbehind = 0;
9082 cd->external_options = options;
9083 cd->open_caps = NULL;
9085 /* Now do the pre-compile. On error, errorcode will be set non-zero, so we
9086 don't need to look at the result of the function here. The initial options have
9087 been put into the cd block so that they can be changed if an option setting is
9088 found within the regex right at the beginning. Bringing initial option settings
9089 outside can help speed up starting point checks. */
9095 (void)compile_regex(cd->external_options, &code, &ptr, &errorcode, FALSE,
9096 FALSE, 0, 0, &firstchar, &firstcharflags, &reqchar, &reqcharflags, NULL,
9098 if (errorcode != 0) goto PCRE_EARLY_ERROR_RETURN;
9100 DPRINTF(("end pre-compile: length=%d workspace=%d\n", length,
9101 (int)(cd->hwm - cworkspace)));
9103 if (length > MAX_PATTERN_SIZE)
9106 goto PCRE_EARLY_ERROR_RETURN;
9109 /* If there are groups with duplicate names and there are also references by
9110 name, we must allow for the possibility of named references to duplicated
9111 groups. These require an extra data item each. */
9113 if (cd->dupnames && cd->namedrefcount > 0)
9114 length += cd->namedrefcount * IMM2_SIZE * sizeof(pcre_uchar);
9116 /* Compute the size of the data block for storing the compiled pattern. Integer
9117 overflow should no longer be possible because nowadays we limit the maximum
9118 value of cd->names_found and cd->name_entry_size. */
9120 size = sizeof(REAL_PCRE) +
9121 (length + cd->names_found * cd->name_entry_size) * sizeof(pcre_uchar);
9123 /* Get the memory. */
9125 re = (REAL_PCRE *)(PUBL(malloc))(size);
9129 goto PCRE_EARLY_ERROR_RETURN;
9132 /* Put in the magic number, and save the sizes, initial options, internal
9133 flags, and character table pointer. NULL is used for the default character
9134 tables. The nullpad field is at the end; it's there to help in the case when a
9135 regex compiled on a system with 4-byte pointers is run on another with 8-byte
9138 re->magic_number = MAGIC_NUMBER;
9139 re->size = (int)size;
9140 re->options = cd->external_options;
9141 re->flags = cd->external_flags;
9142 re->limit_match = limit_match;
9143 re->limit_recursion = limit_recursion;
9146 re->name_table_offset = sizeof(REAL_PCRE) / sizeof(pcre_uchar);
9147 re->name_entry_size = cd->name_entry_size;
9148 re->name_count = cd->names_found;
9150 re->tables = (tables == PRIV(default_tables))? NULL : tables;
9152 #ifdef COMPILE_PCRE32
9155 re->dummy1 = re->dummy2 = re->dummy3 = 0;
9158 /* The starting points of the name/number translation table and of the code are
9159 passed around in the compile data block. The start/end pattern and initial
9160 options are already set from the pre-compile phase, as is the name_entry_size
9161 field. Reset the bracket count and the names_found field. Also reset the hwm
9162 field; this time it's used for remembering forward references to subpatterns.
9165 cd->final_bracount = cd->bracount; /* Save for checking forward references */
9166 cd->parens_depth = 0;
9167 cd->assert_depth = 0;
9169 cd->max_lookbehind = 0;
9170 cd->name_table = (pcre_uchar *)re + re->name_table_offset;
9171 codestart = cd->name_table + re->name_entry_size * re->name_count;
9172 cd->start_code = codestart;
9173 cd->hwm = (pcre_uchar *)(cd->start_workspace);
9174 cd->req_varyopt = 0;
9175 cd->had_accept = FALSE;
9176 cd->had_pruneorskip = FALSE;
9177 cd->check_lookbehind = FALSE;
9178 cd->open_caps = NULL;
9180 /* If any named groups were found, create the name/number table from the list
9181 created in the first pass. */
9183 if (cd->names_found > 0)
9185 int i = cd->names_found;
9186 named_group *ng = cd->named_groups;
9187 cd->names_found = 0;
9188 for (; i > 0; i--, ng++)
9189 add_name(cd, ng->name, ng->length, ng->number);
9190 if (cd->named_group_list_size > NAMED_GROUP_LIST_SIZE)
9191 (PUBL(free))((void *)cd->named_groups);
9194 /* Set up a starting, non-extracting bracket, then compile the expression. On
9195 error, errorcode will be set non-zero, so we don't need to look at the result
9196 of the function here. */
9198 ptr = (const pcre_uchar *)pattern + skipatstart;
9199 code = (pcre_uchar *)codestart;
9201 (void)compile_regex(re->options, &code, &ptr, &errorcode, FALSE, FALSE, 0, 0,
9202 &firstchar, &firstcharflags, &reqchar, &reqcharflags, NULL, cd, NULL);
9203 re->top_bracket = cd->bracount;
9204 re->top_backref = cd->top_backref;
9205 re->max_lookbehind = cd->max_lookbehind;
9206 re->flags = cd->external_flags | PCRE_MODE;
9210 reqchar = 0; /* Must disable after (*ACCEPT) */
9211 reqcharflags = REQ_NONE;
9214 /* If not reached end of pattern on success, there's an excess bracket. */
9216 if (errorcode == 0 && *ptr != CHAR_NULL) errorcode = ERR22;
9218 /* Fill in the terminating state and check for disastrous overflow, but
9219 if debugging, leave the test till after things are printed out. */
9224 if (code - codestart > length) errorcode = ERR23;
9227 #ifdef SUPPORT_VALGRIND
9228 /* If the estimated length exceeds the really used length, mark the extra
9229 allocated memory as unaddressable, so that any out-of-bound reads can be
9231 VALGRIND_MAKE_MEM_NOACCESS(code, (length - (code - codestart)) * sizeof(pcre_uchar));
9234 /* Fill in any forward references that are required. There may be repeated
9235 references; optimize for them, as searching a large regex takes time. */
9237 if (cd->hwm > cd->start_workspace)
9239 int prev_recno = -1;
9240 const pcre_uchar *groupptr = NULL;
9241 while (errorcode == 0 && cd->hwm > cd->start_workspace)
9244 cd->hwm -= LINK_SIZE;
9245 offset = GET(cd->hwm, 0);
9246 recno = GET(codestart, offset);
9247 if (recno != prev_recno)
9249 groupptr = PRIV(find_bracket)(codestart, utf, recno);
9252 if (groupptr == NULL) errorcode = ERR53;
9253 else PUT(((pcre_uchar *)codestart), offset, (int)(groupptr - codestart));
9257 /* If the workspace had to be expanded, free the new memory. Set the pointer to
9258 NULL to indicate that forward references have been filled in. */
9260 if (cd->workspace_size > COMPILE_WORK_SIZE)
9261 (PUBL(free))((void *)cd->start_workspace);
9262 cd->start_workspace = NULL;
9264 /* Give an error if there's back reference to a non-existent capturing
9267 if (errorcode == 0 && re->top_backref > re->top_bracket) errorcode = ERR15;
9269 /* Unless disabled, check whether single character iterators can be
9270 auto-possessified. The function overwrites the appropriate opcode values. */
9272 if ((options & PCRE_NO_AUTO_POSSESS) == 0)
9273 auto_possessify((pcre_uchar *)codestart, utf, cd);
9275 /* If there were any lookbehind assertions that contained OP_RECURSE
9276 (recursions or subroutine calls), a flag is set for them to be checked here,
9277 because they may contain forward references. Actual recursions cannot be fixed
9278 length, but subroutine calls can. It is done like this so that those without
9279 OP_RECURSE that are not fixed length get a diagnosic with a useful offset. The
9280 exceptional ones forgo this. We scan the pattern to check that they are fixed
9281 length, and set their lengths. */
9283 if (errorcode == 0 && cd->check_lookbehind)
9285 pcre_uchar *cc = (pcre_uchar *)codestart;
9287 /* Loop, searching for OP_REVERSE items, and process those that do not have
9288 their length set. (Actually, it will also re-process any that have a length
9289 of zero, but that is a pathological case, and it does no harm.) When we find
9290 one, we temporarily terminate the branch it is in while we scan it. */
9292 for (cc = (pcre_uchar *)PRIV(find_bracket)(codestart, utf, -1);
9294 cc = (pcre_uchar *)PRIV(find_bracket)(cc, utf, -1))
9296 if (GET(cc, 1) == 0)
9299 pcre_uchar *be = cc - 1 - LINK_SIZE + GET(cc, -LINK_SIZE);
9302 fixed_length = find_fixedlength(cc, (re->options & PCRE_UTF8) != 0, TRUE,
9305 DPRINTF(("fixed length = %d\n", fixed_length));
9306 if (fixed_length < 0)
9308 errorcode = (fixed_length == -2)? ERR36 :
9309 (fixed_length == -4)? ERR70 : ERR25;
9312 if (fixed_length > cd->max_lookbehind) cd->max_lookbehind = fixed_length;
9313 PUT(cc, 1, fixed_length);
9315 cc += 1 + LINK_SIZE;
9319 /* Failed to compile, or error while post-processing */
9324 PCRE_EARLY_ERROR_RETURN:
9325 *erroroffset = (int)(ptr - (const pcre_uchar *)pattern);
9326 PCRE_EARLY_ERROR_RETURN2:
9327 *errorptr = find_error_text(errorcode);
9328 if (errorcodeptr != NULL) *errorcodeptr = errorcode;
9332 /* If the anchored option was not passed, set the flag if we can determine that
9333 the pattern is anchored by virtue of ^ characters or \A or anything else, such
9334 as starting with non-atomic .* when DOTALL is set and there are no occurrences
9337 Otherwise, if we know what the first byte has to be, save it, because that
9338 speeds up unanchored matches no end. If not, see if we can set the
9339 PCRE_STARTLINE flag. This is helpful for multiline matches when all branches
9340 start with ^. and also when all branches start with non-atomic .* for
9341 non-DOTALL matches when *PRUNE and SKIP are not present. */
9343 if ((re->options & PCRE_ANCHORED) == 0)
9345 if (is_anchored(codestart, 0, cd, 0)) re->options |= PCRE_ANCHORED;
9348 if (firstcharflags < 0)
9349 firstchar = find_firstassertedchar(codestart, &firstcharflags, FALSE);
9350 if (firstcharflags >= 0) /* Remove caseless flag for non-caseable chars */
9352 #if defined COMPILE_PCRE8
9353 re->first_char = firstchar & 0xff;
9354 #elif defined COMPILE_PCRE16
9355 re->first_char = firstchar & 0xffff;
9356 #elif defined COMPILE_PCRE32
9357 re->first_char = firstchar;
9359 if ((firstcharflags & REQ_CASELESS) != 0)
9361 #if defined SUPPORT_UCP && !(defined COMPILE_PCRE8)
9362 /* We ignore non-ASCII first chars in 8 bit mode. */
9365 if (re->first_char < 128)
9367 if (cd->fcc[re->first_char] != re->first_char)
9368 re->flags |= PCRE_FCH_CASELESS;
9370 else if (UCD_OTHERCASE(re->first_char) != re->first_char)
9371 re->flags |= PCRE_FCH_CASELESS;
9375 if (MAX_255(re->first_char)
9376 && cd->fcc[re->first_char] != re->first_char)
9377 re->flags |= PCRE_FCH_CASELESS;
9380 re->flags |= PCRE_FIRSTSET;
9383 else if (is_startline(codestart, 0, cd, 0)) re->flags |= PCRE_STARTLINE;
9387 /* For an anchored pattern, we use the "required byte" only if it follows a
9388 variable length item in the regex. Remove the caseless flag for non-caseable
9391 if (reqcharflags >= 0 &&
9392 ((re->options & PCRE_ANCHORED) == 0 || (reqcharflags & REQ_VARY) != 0))
9394 #if defined COMPILE_PCRE8
9395 re->req_char = reqchar & 0xff;
9396 #elif defined COMPILE_PCRE16
9397 re->req_char = reqchar & 0xffff;
9398 #elif defined COMPILE_PCRE32
9399 re->req_char = reqchar;
9401 if ((reqcharflags & REQ_CASELESS) != 0)
9403 #if defined SUPPORT_UCP && !(defined COMPILE_PCRE8)
9404 /* We ignore non-ASCII first chars in 8 bit mode. */
9407 if (re->req_char < 128)
9409 if (cd->fcc[re->req_char] != re->req_char)
9410 re->flags |= PCRE_RCH_CASELESS;
9412 else if (UCD_OTHERCASE(re->req_char) != re->req_char)
9413 re->flags |= PCRE_RCH_CASELESS;
9417 if (MAX_255(re->req_char) && cd->fcc[re->req_char] != re->req_char)
9418 re->flags |= PCRE_RCH_CASELESS;
9421 re->flags |= PCRE_REQCHSET;
9424 /* Print out the compiled data if debugging is enabled. This is never the
9425 case when building a production library. */
9428 printf("Length = %d top_bracket = %d top_backref = %d\n",
9429 length, re->top_bracket, re->top_backref);
9431 printf("Options=%08x\n", re->options);
9433 if ((re->flags & PCRE_FIRSTSET) != 0)
9435 pcre_uchar ch = re->first_char;
9436 const char *caseless =
9437 ((re->flags & PCRE_FCH_CASELESS) == 0)? "" : " (caseless)";
9438 if (PRINTABLE(ch)) printf("First char = %c%s\n", ch, caseless);
9439 else printf("First char = \\x%02x%s\n", ch, caseless);
9442 if ((re->flags & PCRE_REQCHSET) != 0)
9444 pcre_uchar ch = re->req_char;
9445 const char *caseless =
9446 ((re->flags & PCRE_RCH_CASELESS) == 0)? "" : " (caseless)";
9447 if (PRINTABLE(ch)) printf("Req char = %c%s\n", ch, caseless);
9448 else printf("Req char = \\x%02x%s\n", ch, caseless);
9451 #if defined COMPILE_PCRE8
9452 pcre_printint((pcre *)re, stdout, TRUE);
9453 #elif defined COMPILE_PCRE16
9454 pcre16_printint((pcre *)re, stdout, TRUE);
9455 #elif defined COMPILE_PCRE32
9456 pcre32_printint((pcre *)re, stdout, TRUE);
9459 /* This check is done here in the debugging case so that the code that
9460 was compiled can be seen. */
9462 if (code - codestart > length)
9465 *errorptr = find_error_text(ERR23);
9466 *erroroffset = ptr - (pcre_uchar *)pattern;
9467 if (errorcodeptr != NULL) *errorcodeptr = ERR23;
9470 #endif /* PCRE_DEBUG */
9472 /* Check for a pattern than can match an empty string, so that this information
9473 can be provided to applications. */
9477 if (could_be_empty_branch(codestart, code, utf, cd, NULL))
9479 re->flags |= PCRE_MATCH_EMPTY;
9482 codestart += GET(codestart, 1);
9484 while (*codestart == OP_ALT);
9486 #if defined COMPILE_PCRE8
9488 #elif defined COMPILE_PCRE16
9489 return (pcre16 *)re;
9490 #elif defined COMPILE_PCRE32
9491 return (pcre32 *)re;
9495 /* End of pcre_compile.c */