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);
2372 /* Test for forward reference or uncompleted reference. This is disabled
2373 when called to scan a completed pattern by setting cd->start_workspace to
2376 if (cd->start_workspace != NULL)
2378 const pcre_uchar *tcode;
2379 for (tcode = cd->start_workspace; tcode < cd->hwm; tcode += LINK_SIZE)
2380 if ((int)GET(tcode, 0) == (int)(code + 1 - cd->start_code)) return TRUE;
2381 if (GET(scode, 1) == 0) return TRUE; /* Unclosed */
2384 /* If we are scanning a completed pattern, there are no forward references
2385 and all groups are complete. We need to detect whether this is a recursive
2386 call, as otherwise there will be an infinite loop. If it is a recursion,
2387 just skip over it. Simple recursions are easily detected. For mutual
2388 recursions we keep a chain on the stack. */
2392 recurse_check *r = recurses;
2393 const pcre_uchar *endgroup = scode;
2395 do endgroup += GET(endgroup, 1); while (*endgroup == OP_ALT);
2396 if (code >= scode && code <= endgroup) continue; /* Simple recursion */
2398 for (r = recurses; r != NULL; r = r->prev)
2399 if (r->group == scode) break;
2400 if (r != NULL) continue; /* Mutual recursion */
2403 /* Completed reference; scan the referenced group, remembering it on the
2404 stack chain to detect mutual recursions. */
2406 empty_branch = FALSE;
2407 this_recurse.prev = recurses;
2408 this_recurse.group = scode;
2412 if (could_be_empty_branch(scode, endcode, utf, cd, &this_recurse))
2414 empty_branch = TRUE;
2417 scode += GET(scode, 1);
2419 while (*scode == OP_ALT);
2421 if (!empty_branch) return FALSE; /* All branches are non-empty */
2425 /* Groups with zero repeats can of course be empty; skip them. */
2427 if (c == OP_BRAZERO || c == OP_BRAMINZERO || c == OP_SKIPZERO ||
2430 code += PRIV(OP_lengths)[c];
2431 do code += GET(code, 1); while (*code == OP_ALT);
2436 /* A nested group that is already marked as "could be empty" can just be
2439 if (c == OP_SBRA || c == OP_SBRAPOS ||
2440 c == OP_SCBRA || c == OP_SCBRAPOS)
2442 do code += GET(code, 1); while (*code == OP_ALT);
2447 /* For other groups, scan the branches. */
2449 if (c == OP_BRA || c == OP_BRAPOS ||
2450 c == OP_CBRA || c == OP_CBRAPOS ||
2451 c == OP_ONCE || c == OP_ONCE_NC ||
2455 if (GET(code, 1) == 0) return TRUE; /* Hit unclosed bracket */
2457 /* If a conditional group has only one branch, there is a second, implied,
2458 empty branch, so just skip over the conditional, because it could be empty.
2459 Otherwise, scan the individual branches of the group. */
2461 if (c == OP_COND && code[GET(code, 1)] != OP_ALT)
2462 code += GET(code, 1);
2465 empty_branch = FALSE;
2468 if (!empty_branch && could_be_empty_branch(code, endcode, utf, cd, NULL))
2469 empty_branch = TRUE;
2470 code += GET(code, 1);
2472 while (*code == OP_ALT);
2473 if (!empty_branch) return FALSE; /* All branches are non-empty */
2480 /* Handle the other opcodes */
2484 /* Check for quantifiers after a class. XCLASS is used for classes that
2485 cannot be represented just by a bit map. This includes negated single
2486 high-valued characters. The length in PRIV(OP_lengths)[] is zero; the
2487 actual length is stored in the compiled code, so we must update "code"
2490 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
2492 ccode = code += GET(code, 1);
2493 goto CHECK_CLASS_REPEAT;
2498 ccode = code + PRIV(OP_lengths)[OP_CLASS];
2500 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
2506 case OP_CRSTAR: /* These could be empty; continue */
2514 default: /* Non-repeat => class must match */
2515 case OP_CRPLUS: /* These repeats aren't empty */
2523 if (GET2(ccode, 1) > 0) return FALSE; /* Minimum > 0 */
2528 /* Opcodes that must match a character */
2546 case OP_NOT_WHITESPACE:
2548 case OP_NOT_WORDCHAR:
2564 case OP_NOTMINPLUSI:
2569 case OP_NOTPOSPLUSI:
2577 case OP_TYPEMINPLUS:
2578 case OP_TYPEPOSPLUS:
2583 /* These are going to continue, as they may be empty, but we have to
2584 fudge the length for the \p and \P cases. */
2587 case OP_TYPEMINSTAR:
2588 case OP_TYPEPOSSTAR:
2590 case OP_TYPEMINQUERY:
2591 case OP_TYPEPOSQUERY:
2592 if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
2595 /* Same for these */
2598 case OP_TYPEMINUPTO:
2599 case OP_TYPEPOSUPTO:
2600 if (code[1 + IMM2_SIZE] == OP_PROP || code[1 + IMM2_SIZE] == OP_NOTPROP)
2613 /* In UTF-8 mode, STAR, MINSTAR, POSSTAR, QUERY, MINQUERY, POSQUERY, UPTO,
2614 MINUPTO, and POSUPTO and their caseless and negative versions may be
2615 followed by a multibyte character. */
2617 #if defined SUPPORT_UTF && !defined COMPILE_PCRE32
2626 case OP_NOTMINSTARI:
2631 case OP_NOTPOSSTARI:
2640 case OP_NOTMINQUERY:
2641 case OP_NOTMINQUERYI:
2645 case OP_NOTPOSQUERY:
2646 case OP_NOTPOSQUERYI:
2648 if (utf && HAS_EXTRALEN(code[1])) code += GET_EXTRALEN(code[1]);
2659 case OP_NOTMINUPTOI:
2664 case OP_NOTPOSUPTOI:
2666 if (utf && HAS_EXTRALEN(code[1 + IMM2_SIZE])) code += GET_EXTRALEN(code[1 + IMM2_SIZE]);
2670 /* MARK, and PRUNE/SKIP/THEN with an argument must skip over the argument
2680 /* None of the remaining opcodes are required to match a character. */
2692 /*************************************************
2693 * Scan compiled regex for non-emptiness *
2694 *************************************************/
2696 /* This function is called to check for left recursive calls. We want to check
2697 the current branch of the current pattern to see if it could match the empty
2698 string. If it could, we must look outwards for branches at other levels,
2699 stopping when we pass beyond the bracket which is the subject of the recursion.
2700 This function is called only during the real compile, not during the
2704 code points to start of the recursion
2705 endcode points to where to stop (current RECURSE item)
2706 bcptr points to the chain of current (unclosed) branch starts
2707 utf TRUE if in UTF-8 / UTF-16 / UTF-32 mode
2708 cd pointers to tables etc
2710 Returns: TRUE if what is matched could be empty
2714 could_be_empty(const pcre_uchar *code, const pcre_uchar *endcode,
2715 branch_chain *bcptr, BOOL utf, compile_data *cd)
2717 while (bcptr != NULL && bcptr->current_branch >= code)
2719 if (!could_be_empty_branch(bcptr->current_branch, endcode, utf, cd, NULL))
2721 bcptr = bcptr->outer;
2728 /*************************************************
2729 * Base opcode of repeated opcodes *
2730 *************************************************/
2732 /* Returns the base opcode for repeated single character type opcodes. If the
2733 opcode is not a repeated character type, it returns with the original value.
2736 Returns: base opcode for the type
2740 get_repeat_base(pcre_uchar c)
2742 return (c > OP_TYPEPOSUPTO)? c :
2743 (c >= OP_TYPESTAR)? OP_TYPESTAR :
2744 (c >= OP_NOTSTARI)? OP_NOTSTARI :
2745 (c >= OP_NOTSTAR)? OP_NOTSTAR :
2746 (c >= OP_STARI)? OP_STARI :
2753 /*************************************************
2754 * Check a character and a property *
2755 *************************************************/
2757 /* This function is called by check_auto_possessive() when a property item
2758 is adjacent to a fixed character.
2762 ptype the property type
2763 pdata the data for the type
2764 negated TRUE if it's a negated property (\P or \p{^)
2766 Returns: TRUE if auto-possessifying is OK
2770 check_char_prop(pcre_uint32 c, unsigned int ptype, unsigned int pdata,
2773 const pcre_uint32 *p;
2774 const ucd_record *prop = GET_UCD(c);
2779 return (prop->chartype == ucp_Lu ||
2780 prop->chartype == ucp_Ll ||
2781 prop->chartype == ucp_Lt) == negated;
2784 return (pdata == PRIV(ucp_gentype)[prop->chartype]) == negated;
2787 return (pdata == prop->chartype) == negated;
2790 return (pdata == prop->script) == negated;
2792 /* These are specials */
2795 return (PRIV(ucp_gentype)[prop->chartype] == ucp_L ||
2796 PRIV(ucp_gentype)[prop->chartype] == ucp_N) == negated;
2798 /* Perl space used to exclude VT, but from Perl 5.18 it is included, which
2799 means that Perl space and POSIX space are now identical. PCRE was changed
2802 case PT_SPACE: /* Perl space */
2803 case PT_PXSPACE: /* POSIX space */
2811 return (PRIV(ucp_gentype)[prop->chartype] == ucp_Z) == negated;
2813 break; /* Control never reaches here */
2816 return (PRIV(ucp_gentype)[prop->chartype] == ucp_L ||
2817 PRIV(ucp_gentype)[prop->chartype] == ucp_N ||
2818 c == CHAR_UNDERSCORE) == negated;
2821 p = PRIV(ucd_caseless_sets) + prop->caseset;
2824 if (c < *p) return !negated;
2825 if (c == *p++) return negated;
2827 break; /* Control never reaches here */
2832 #endif /* SUPPORT_UCP */
2836 /*************************************************
2837 * Fill the character property list *
2838 *************************************************/
2840 /* Checks whether the code points to an opcode that can take part in auto-
2841 possessification, and if so, fills a list with its properties.
2844 code points to start of expression
2845 utf TRUE if in UTF-8 / UTF-16 / UTF-32 mode
2846 fcc points to case-flipping table
2847 list points to output list
2848 list[0] will be filled with the opcode
2849 list[1] will be non-zero if this opcode
2850 can match an empty character string
2851 list[2..7] depends on the opcode
2853 Returns: points to the start of the next opcode if *code is accepted
2854 NULL if *code is not accepted
2857 static const pcre_uchar *
2858 get_chr_property_list(const pcre_uchar *code, BOOL utf,
2859 const pcre_uint8 *fcc, pcre_uint32 *list)
2861 pcre_uchar c = *code;
2863 const pcre_uchar *end;
2867 pcre_uint32 *clist_dest;
2868 const pcre_uint32 *clist_src;
2870 utf = utf; /* Suppress "unused parameter" compiler warning */
2877 if (c >= OP_STAR && c <= OP_TYPEPOSUPTO)
2879 base = get_repeat_base(c);
2880 c -= (base - OP_STAR);
2882 if (c == OP_UPTO || c == OP_MINUPTO || c == OP_EXACT || c == OP_POSUPTO)
2885 list[1] = (c != OP_PLUS && c != OP_MINPLUS && c != OP_EXACT && c != OP_POSPLUS);
2917 case OP_NOT_WHITESPACE:
2919 case OP_NOT_WORDCHAR:
2937 GETCHARINCTEST(chr, code);
2944 list[0] = (c == OP_CHARI) ? OP_CHAR : OP_NOT;
2945 GETCHARINCTEST(chr, code);
2949 if (chr < 128 || (chr < 256 && !utf))
2952 list[3] = UCD_OTHERCASE(chr);
2953 #elif defined SUPPORT_UTF || !defined COMPILE_PCRE8
2954 list[3] = (chr < 256) ? fcc[chr] : chr;
2959 /* The othercase might be the same value. */
2970 if (code[0] != PT_CLIST)
2977 /* Convert only if we have enough space. */
2979 clist_src = PRIV(ucd_caseless_sets) + code[1];
2980 clist_dest = list + 2;
2984 if (clist_dest >= list + 8)
2986 /* Early return if there is not enough space. This should never
2987 happen, since all clists are shorter than 5 character now. */
2992 *clist_dest++ = *clist_src;
2994 while(*clist_src++ != NOTACHAR);
2996 /* All characters are stored. The terminating NOTACHAR
2997 is copied form the clist itself. */
2999 list[0] = (c == OP_PROP) ? OP_CHAR : OP_NOT;
3005 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
3008 end = code + GET(code, 0) - 1;
3011 end = code + 32 / sizeof(pcre_uchar);
3034 list[1] = (GET2(end, 1) == 0);
3035 end += 1 + 2 * IMM2_SIZE;
3038 list[2] = end - code;
3041 return NULL; /* Opcode not accepted */
3046 /*************************************************
3047 * Scan further character sets for match *
3048 *************************************************/
3050 /* Checks whether the base and the current opcode have a common character, in
3051 which case the base cannot be possessified.
3054 code points to the byte code
3055 utf TRUE in UTF-8 / UTF-16 / UTF-32 mode
3056 cd static compile data
3057 base_list the data list of the base opcode
3059 Returns: TRUE if the auto-possessification is possible
3063 compare_opcodes(const pcre_uchar *code, BOOL utf, const compile_data *cd,
3064 const pcre_uint32 *base_list, const pcre_uchar *base_end)
3067 pcre_uint32 list[8];
3068 const pcre_uint32 *chr_ptr;
3069 const pcre_uint32 *ochr_ptr;
3070 const pcre_uint32 *list_ptr;
3071 const pcre_uchar *next_code;
3072 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
3073 const pcre_uchar *xclass_flags;
3075 const pcre_uint8 *class_bitset;
3076 const pcre_uint8 *set1, *set2, *set_end;
3078 BOOL accepted, invert_bits;
3080 /* Note: the base_list[1] contains whether the current opcode has greedy
3081 (represented by a non-zero value) quantifier. This is a different from
3082 other character type lists, which stores here that the character iterator
3083 matches to an empty string (also represented by a non-zero value). */
3087 /* All operations move the code pointer forward.
3088 Therefore infinite recursions are not possible. */
3092 /* Skip over callouts */
3094 if (c == OP_CALLOUT)
3096 code += PRIV(OP_lengths)[c];
3102 do code += GET(code, 1); while (*code == OP_ALT);
3110 /* TRUE only in greedy case. The non-greedy case could be replaced by
3111 an OP_EXACT, but it is probably not worth it. (And note that OP_EXACT
3112 uses more memory, which we cannot get at this stage.) */
3114 return base_list[1] != 0;
3117 /* If the bracket is capturing, and referenced by an OP_RECURSE, or
3118 it is an atomic sub-pattern (assert, once, etc.) the non-greedy case
3119 cannot be converted to a possessive form. */
3121 if (base_list[1] == 0) return FALSE;
3123 switch(*(code - GET(code, 1)))
3128 case OP_ASSERTBACK_NOT:
3131 /* Atomic sub-patterns and assertions can always auto-possessify their
3136 code += PRIV(OP_lengths)[c];
3143 next_code = code + GET(code, 1);
3144 code += PRIV(OP_lengths)[c];
3146 while (*next_code == OP_ALT)
3148 if (!compare_opcodes(code, utf, cd, base_list, base_end)) return FALSE;
3149 code = next_code + 1 + LINK_SIZE;
3150 next_code += GET(next_code, 1);
3157 next_code = code + 1;
3158 if (*next_code != OP_BRA && *next_code != OP_CBRA
3159 && *next_code != OP_ONCE && *next_code != OP_ONCE_NC) return FALSE;
3161 do next_code += GET(next_code, 1); while (*next_code == OP_ALT);
3163 /* The bracket content will be checked by the
3164 OP_BRA/OP_CBRA case above. */
3165 next_code += 1 + LINK_SIZE;
3166 if (!compare_opcodes(next_code, utf, cd, base_list, base_end))
3169 code += PRIV(OP_lengths)[c];
3173 /* Check for a supported opcode, and load its properties. */
3175 code = get_chr_property_list(code, utf, cd->fcc, list);
3176 if (code == NULL) return FALSE; /* Unsupported */
3178 /* If either opcode is a small character list, set pointers for comparing
3179 characters from that list with another list, or with a property. */
3181 if (base_list[0] == OP_CHAR)
3183 chr_ptr = base_list + 2;
3186 else if (list[0] == OP_CHAR)
3189 list_ptr = base_list;
3192 /* Character bitsets can also be compared to certain opcodes. */
3194 else if (base_list[0] == OP_CLASS || list[0] == OP_CLASS
3195 #ifdef COMPILE_PCRE8
3196 /* In 8 bit, non-UTF mode, OP_CLASS and OP_NCLASS are the same. */
3197 || (!utf && (base_list[0] == OP_NCLASS || list[0] == OP_NCLASS))
3201 #ifdef COMPILE_PCRE8
3202 if (base_list[0] == OP_CLASS || (!utf && base_list[0] == OP_NCLASS))
3204 if (base_list[0] == OP_CLASS)
3207 set1 = (pcre_uint8 *)(base_end - base_list[2]);
3212 set1 = (pcre_uint8 *)(code - list[2]);
3213 list_ptr = base_list;
3216 invert_bits = FALSE;
3221 set2 = (pcre_uint8 *)
3222 ((list_ptr == list ? code : base_end) - list_ptr[2]);
3225 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
3227 xclass_flags = (list_ptr == list ? code : base_end) - list_ptr[2] + LINK_SIZE;
3228 if ((*xclass_flags & XCL_HASPROP) != 0) return FALSE;
3229 if ((*xclass_flags & XCL_MAP) == 0)
3231 /* No bits are set for characters < 256. */
3232 if (list[1] == 0) return TRUE;
3233 /* Might be an empty repeat. */
3236 set2 = (pcre_uint8 *)(xclass_flags + 1);
3244 set2 = (pcre_uint8 *)(cd->cbits + cbit_digit);
3247 case OP_NOT_WHITESPACE:
3251 set2 = (pcre_uint8 *)(cd->cbits + cbit_space);
3254 case OP_NOT_WORDCHAR:
3258 set2 = (pcre_uint8 *)(cd->cbits + cbit_word);
3265 /* Because the sets are unaligned, we need
3266 to perform byte comparison here. */
3267 set_end = set1 + 32;
3272 if ((*set1++ & ~(*set2++)) != 0) return FALSE;
3274 while (set1 < set_end);
3280 if ((*set1++ & *set2++) != 0) return FALSE;
3282 while (set1 < set_end);
3285 if (list[1] == 0) return TRUE;
3286 /* Might be an empty repeat. */
3290 /* Some property combinations also acceptable. Unicode property opcodes are
3291 processed specially; the rest can be handled with a lookup table. */
3295 pcre_uint32 leftop, rightop;
3297 leftop = base_list[0];
3301 accepted = FALSE; /* Always set in non-unicode case. */
3302 if (leftop == OP_PROP || leftop == OP_NOTPROP)
3304 if (rightop == OP_EOD)
3306 else if (rightop == OP_PROP || rightop == OP_NOTPROP)
3309 const pcre_uint8 *p;
3310 BOOL same = leftop == rightop;
3311 BOOL lisprop = leftop == OP_PROP;
3312 BOOL risprop = rightop == OP_PROP;
3313 BOOL bothprop = lisprop && risprop;
3315 /* There's a table that specifies how each combination is to be
3317 0 Always return FALSE (never auto-possessify)
3318 1 Character groups are distinct (possessify if both are OP_PROP)
3319 2 Check character categories in the same group (general or particular)
3320 3 Return TRUE if the two opcodes are not the same
3321 ... see comments below
3324 n = propposstab[base_list[2]][list[2]];
3328 case 1: accepted = bothprop; break;
3329 case 2: accepted = (base_list[3] == list[3]) != same; break;
3330 case 3: accepted = !same; break;
3332 case 4: /* Left general category, right particular category */
3333 accepted = risprop && catposstab[base_list[3]][list[3]] == same;
3336 case 5: /* Right general category, left particular category */
3337 accepted = lisprop && catposstab[list[3]][base_list[3]] == same;
3340 /* This code is logically tricky. Think hard before fiddling with it.
3341 The posspropstab table has four entries per row. Each row relates to
3342 one of PCRE's special properties such as ALNUM or SPACE or WORD.
3343 Only WORD actually needs all four entries, but using repeats for the
3344 others means they can all use the same code below.
3346 The first two entries in each row are Unicode general categories, and
3347 apply always, because all the characters they include are part of the
3348 PCRE character set. The third and fourth entries are a general and a
3349 particular category, respectively, that include one or more relevant
3350 characters. One or the other is used, depending on whether the check
3351 is for a general or a particular category. However, in both cases the
3352 category contains more characters than the specials that are defined
3353 for the property being tested against. Therefore, it cannot be used
3356 Example: the row for WORD contains ucp_L, ucp_N, ucp_P, ucp_Po.
3357 Underscore is covered by ucp_P or ucp_Po. */
3359 case 6: /* Left alphanum vs right general category */
3360 case 7: /* Left space vs right general category */
3361 case 8: /* Left word vs right general category */
3362 p = posspropstab[n-6];
3363 accepted = risprop && lisprop ==
3366 (list[3] != p[2] || !lisprop));
3369 case 9: /* Right alphanum vs left general category */
3370 case 10: /* Right space vs left general category */
3371 case 11: /* Right word vs left general category */
3372 p = posspropstab[n-9];
3373 accepted = lisprop && risprop ==
3374 (base_list[3] != p[0] &&
3375 base_list[3] != p[1] &&
3376 (base_list[3] != p[2] || !risprop));
3379 case 12: /* Left alphanum vs right particular category */
3380 case 13: /* Left space vs right particular category */
3381 case 14: /* Left word vs right particular category */
3382 p = posspropstab[n-12];
3383 accepted = risprop && lisprop ==
3384 (catposstab[p[0]][list[3]] &&
3385 catposstab[p[1]][list[3]] &&
3386 (list[3] != p[3] || !lisprop));
3389 case 15: /* Right alphanum vs left particular category */
3390 case 16: /* Right space vs left particular category */
3391 case 17: /* Right word vs left particular category */
3392 p = posspropstab[n-15];
3393 accepted = lisprop && risprop ==
3394 (catposstab[p[0]][base_list[3]] &&
3395 catposstab[p[1]][base_list[3]] &&
3396 (base_list[3] != p[3] || !risprop));
3403 #endif /* SUPPORT_UCP */
3405 accepted = leftop >= FIRST_AUTOTAB_OP && leftop <= LAST_AUTOTAB_LEFT_OP &&
3406 rightop >= FIRST_AUTOTAB_OP && rightop <= LAST_AUTOTAB_RIGHT_OP &&
3407 autoposstab[leftop - FIRST_AUTOTAB_OP][rightop - FIRST_AUTOTAB_OP];
3412 if (list[1] == 0) return TRUE;
3413 /* Might be an empty repeat. */
3417 /* Control reaches here only if one of the items is a small character list.
3418 All characters are checked against the other side. */
3427 ochr_ptr = list_ptr + 2;
3430 if (chr == *ochr_ptr) return FALSE;
3433 while(*ochr_ptr != NOTACHAR);
3437 ochr_ptr = list_ptr + 2;
3440 if (chr == *ochr_ptr)
3444 while(*ochr_ptr != NOTACHAR);
3445 if (*ochr_ptr == NOTACHAR) return FALSE; /* Not found */
3448 /* Note that OP_DIGIT etc. are generated only when PCRE_UCP is *not*
3449 set. When it is set, \d etc. are converted into OP_(NOT_)PROP codes. */
3452 if (chr < 256 && (cd->ctypes[chr] & ctype_digit) != 0) return FALSE;
3456 if (chr > 255 || (cd->ctypes[chr] & ctype_digit) == 0) return FALSE;
3460 if (chr < 256 && (cd->ctypes[chr] & ctype_space) != 0) return FALSE;
3463 case OP_NOT_WHITESPACE:
3464 if (chr > 255 || (cd->ctypes[chr] & ctype_space) == 0) return FALSE;
3468 if (chr < 255 && (cd->ctypes[chr] & ctype_word) != 0) return FALSE;
3471 case OP_NOT_WORDCHAR:
3472 if (chr > 255 || (cd->ctypes[chr] & ctype_word) == 0) return FALSE;
3478 HSPACE_CASES: return FALSE;
3486 HSPACE_CASES: break;
3487 default: return FALSE;
3495 VSPACE_CASES: return FALSE;
3503 VSPACE_CASES: break;
3504 default: return FALSE;
3520 #endif /* Not EBCDIC */
3525 case OP_EOD: /* Can always possessify before \z */
3531 if (!check_char_prop(chr, list_ptr[2], list_ptr[3],
3532 list_ptr[0] == OP_NOTPROP))
3538 if (chr > 255) return FALSE;
3542 if (chr > 255) break;
3543 class_bitset = (pcre_uint8 *)
3544 ((list_ptr == list ? code : base_end) - list_ptr[2]);
3545 if ((class_bitset[chr >> 3] & (1 << (chr & 7))) != 0) return FALSE;
3548 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
3550 if (PRIV(xclass)(chr, (list_ptr == list ? code : base_end) -
3551 list_ptr[2] + LINK_SIZE, utf)) return FALSE;
3561 while(*chr_ptr != NOTACHAR);
3563 /* At least one character must be matched from this opcode. */
3565 if (list[1] == 0) return TRUE;
3568 /* Control never reaches here. There used to be a fail-save return FALSE; here,
3569 but some compilers complain about an unreachable statement. */
3575 /*************************************************
3576 * Scan compiled regex for auto-possession *
3577 *************************************************/
3579 /* Replaces single character iterations with their possessive alternatives
3580 if appropriate. This function modifies the compiled opcode!
3583 code points to start of the byte code
3584 utf TRUE in UTF-8 / UTF-16 / UTF-32 mode
3585 cd static compile data
3591 auto_possessify(pcre_uchar *code, BOOL utf, const compile_data *cd)
3593 register pcre_uchar c;
3594 const pcre_uchar *end;
3595 pcre_uchar *repeat_opcode;
3596 pcre_uint32 list[8];
3602 if (c >= OP_STAR && c <= OP_TYPEPOSUPTO)
3604 c -= get_repeat_base(c) - OP_STAR;
3605 end = (c <= OP_MINUPTO) ?
3606 get_chr_property_list(code, utf, cd->fcc, list) : NULL;
3607 list[1] = c == OP_STAR || c == OP_PLUS || c == OP_QUERY || c == OP_UPTO;
3609 if (end != NULL && compare_opcodes(end, utf, cd, list, end))
3614 *code += OP_POSSTAR - OP_STAR;
3618 *code += OP_POSSTAR - OP_MINSTAR;
3622 *code += OP_POSPLUS - OP_PLUS;
3626 *code += OP_POSPLUS - OP_MINPLUS;
3630 *code += OP_POSQUERY - OP_QUERY;
3634 *code += OP_POSQUERY - OP_MINQUERY;
3638 *code += OP_POSUPTO - OP_UPTO;
3642 *code += OP_POSUPTO - OP_MINUPTO;
3648 else if (c == OP_CLASS || c == OP_NCLASS || c == OP_XCLASS)
3650 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
3652 repeat_opcode = code + GET(code, 1);
3655 repeat_opcode = code + 1 + (32 / sizeof(pcre_uchar));
3658 if (c >= OP_CRSTAR && c <= OP_CRMINRANGE)
3660 /* end must not be NULL. */
3661 end = get_chr_property_list(code, utf, cd->fcc, list);
3663 list[1] = (c & 1) == 0;
3665 if (compare_opcodes(end, utf, cd, list, end))
3671 *repeat_opcode = OP_CRPOSSTAR;
3676 *repeat_opcode = OP_CRPOSPLUS;
3681 *repeat_opcode = OP_CRPOSQUERY;
3686 *repeat_opcode = OP_CRPOSRANGE;
3700 case OP_TYPEMINSTAR:
3702 case OP_TYPEMINPLUS:
3704 case OP_TYPEMINQUERY:
3705 case OP_TYPEPOSSTAR:
3706 case OP_TYPEPOSPLUS:
3707 case OP_TYPEPOSQUERY:
3708 if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
3712 case OP_TYPEMINUPTO:
3714 case OP_TYPEPOSUPTO:
3715 if (code[1 + IMM2_SIZE] == OP_PROP || code[1 + IMM2_SIZE] == OP_NOTPROP)
3719 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
3721 code += GET(code, 1);
3733 /* Add in the fixed length from the table */
3735 code += PRIV(OP_lengths)[c];
3737 /* In UTF-8 mode, opcodes that are followed by a character may be followed by
3738 a multi-byte character. The length in the table is a minimum, so we have to
3739 arrange to skip the extra bytes. */
3741 #if defined SUPPORT_UTF && !defined COMPILE_PCRE32
3779 case OP_NOTMINQUERY:
3785 case OP_NOTPOSQUERY:
3788 case OP_NOTMINSTARI:
3790 case OP_NOTMINPLUSI:
3792 case OP_NOTMINQUERYI:
3794 case OP_NOTMINUPTOI:
3796 case OP_NOTPOSSTARI:
3797 case OP_NOTPOSPLUSI:
3798 case OP_NOTPOSQUERYI:
3799 case OP_NOTPOSUPTOI:
3800 if (HAS_EXTRALEN(code[-1])) code += GET_EXTRALEN(code[-1]);
3804 (void)(utf); /* Keep compiler happy by referencing function argument */
3811 /*************************************************
3812 * Check for POSIX class syntax *
3813 *************************************************/
3815 /* This function is called when the sequence "[:" or "[." or "[=" is
3816 encountered in a character class. It checks whether this is followed by a
3817 sequence of characters terminated by a matching ":]" or ".]" or "=]". If we
3818 reach an unescaped ']' without the special preceding character, return FALSE.
3820 Originally, this function only recognized a sequence of letters between the
3821 terminators, but it seems that Perl recognizes any sequence of characters,
3822 though of course unknown POSIX names are subsequently rejected. Perl gives an
3823 "Unknown POSIX class" error for [:f\oo:] for example, where previously PCRE
3824 didn't consider this to be a POSIX class. Likewise for [:1234:].
3826 The problem in trying to be exactly like Perl is in the handling of escapes. We
3827 have to be sure that [abc[:x\]pqr] is *not* treated as containing a POSIX
3828 class, but [abc[:x\]pqr:]] is (so that an error can be generated). The code
3829 below handles the special case of \], but does not try to do any other escape
3830 processing. This makes it different from Perl for cases such as [:l\ower:]
3831 where Perl recognizes it as the POSIX class "lower" but PCRE does not recognize
3832 "l\ower". This is a lesser evil than not diagnosing bad classes when Perl does,
3835 A user pointed out that PCRE was rejecting [:a[:digit:]] whereas Perl was not.
3836 It seems that the appearance of a nested POSIX class supersedes an apparent
3837 external class. For example, [:a[:digit:]b:] matches "a", "b", ":", or
3840 In Perl, unescaped square brackets may also appear as part of class names. For
3841 example, [:a[:abc]b:] gives unknown POSIX class "[:abc]b:]". However, for
3842 [:a[:abc]b][b:] it gives unknown POSIX class "[:abc]b][b:]", which does not
3843 seem right at all. PCRE does not allow closing square brackets in POSIX class
3847 ptr pointer to the initial [
3848 endptr where to return the end pointer
3850 Returns: TRUE or FALSE
3854 check_posix_syntax(const pcre_uchar *ptr, const pcre_uchar **endptr)
3856 pcre_uchar terminator; /* Don't combine these lines; the Solaris cc */
3857 terminator = *(++ptr); /* compiler warns about "non-constant" initializer. */
3858 for (++ptr; *ptr != CHAR_NULL; ptr++)
3860 if (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET)
3862 else if (*ptr == CHAR_RIGHT_SQUARE_BRACKET) return FALSE;
3865 if (*ptr == terminator && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET)
3870 if (*ptr == CHAR_LEFT_SQUARE_BRACKET &&
3871 (ptr[1] == CHAR_COLON || ptr[1] == CHAR_DOT ||
3872 ptr[1] == CHAR_EQUALS_SIGN) &&
3873 check_posix_syntax(ptr, endptr))
3883 /*************************************************
3884 * Check POSIX class name *
3885 *************************************************/
3887 /* This function is called to check the name given in a POSIX-style class entry
3891 ptr points to the first letter
3892 len the length of the name
3894 Returns: a value representing the name, or -1 if unknown
3898 check_posix_name(const pcre_uchar *ptr, int len)
3900 const char *pn = posix_names;
3901 register int yield = 0;
3902 while (posix_name_lengths[yield] != 0)
3904 if (len == posix_name_lengths[yield] &&
3905 STRNCMP_UC_C8(ptr, pn, (unsigned int)len) == 0) return yield;
3906 pn += posix_name_lengths[yield] + 1;
3913 /*************************************************
3914 * Adjust OP_RECURSE items in repeated group *
3915 *************************************************/
3917 /* OP_RECURSE items contain an offset from the start of the regex to the group
3918 that is referenced. This means that groups can be replicated for fixed
3919 repetition simply by copying (because the recursion is allowed to refer to
3920 earlier groups that are outside the current group). However, when a group is
3921 optional (i.e. the minimum quantifier is zero), OP_BRAZERO or OP_SKIPZERO is
3922 inserted before it, after it has been compiled. This means that any OP_RECURSE
3923 items within it that refer to the group itself or any contained groups have to
3924 have their offsets adjusted. That one of the jobs of this function. Before it
3925 is called, the partially compiled regex must be temporarily terminated with
3928 This function has been extended with the possibility of forward references for
3929 recursions and subroutine calls. It must also check the list of such references
3930 for the group we are dealing with. If it finds that one of the recursions in
3931 the current group is on this list, it adjusts the offset in the list, not the
3932 value in the reference (which is a group number).
3935 group points to the start of the group
3936 adjust the amount by which the group is to be moved
3937 utf TRUE in UTF-8 / UTF-16 / UTF-32 mode
3938 cd contains pointers to tables etc.
3939 save_hwm the hwm forward reference pointer at the start of the group
3945 adjust_recurse(pcre_uchar *group, int adjust, BOOL utf, compile_data *cd,
3946 pcre_uchar *save_hwm)
3948 pcre_uchar *ptr = group;
3950 while ((ptr = (pcre_uchar *)find_recurse(ptr, utf)) != NULL)
3955 /* See if this recursion is on the forward reference list. If so, adjust the
3958 for (hc = save_hwm; hc < cd->hwm; hc += LINK_SIZE)
3960 offset = (int)GET(hc, 0);
3961 if (cd->start_code + offset == ptr + 1)
3963 PUT(hc, 0, offset + adjust);
3968 /* Otherwise, adjust the recursion offset if it's after the start of this
3973 offset = (int)GET(ptr, 1);
3974 if (cd->start_code + offset >= group) PUT(ptr, 1, offset + adjust);
3977 ptr += 1 + LINK_SIZE;
3983 /*************************************************
3984 * Insert an automatic callout point *
3985 *************************************************/
3987 /* This function is called when the PCRE_AUTO_CALLOUT option is set, to insert
3988 callout points before each pattern item.
3991 code current code pointer
3992 ptr current pattern pointer
3993 cd pointers to tables etc
3995 Returns: new code pointer
3999 auto_callout(pcre_uchar *code, const pcre_uchar *ptr, compile_data *cd)
4001 *code++ = OP_CALLOUT;
4003 PUT(code, 0, (int)(ptr - cd->start_pattern)); /* Pattern offset */
4004 PUT(code, LINK_SIZE, 0); /* Default length */
4005 return code + 2 * LINK_SIZE;
4010 /*************************************************
4011 * Complete a callout item *
4012 *************************************************/
4014 /* A callout item contains the length of the next item in the pattern, which
4015 we can't fill in till after we have reached the relevant point. This is used
4016 for both automatic and manual callouts.
4019 previous_callout points to previous callout item
4020 ptr current pattern pointer
4021 cd pointers to tables etc
4027 complete_callout(pcre_uchar *previous_callout, const pcre_uchar *ptr, compile_data *cd)
4029 int length = (int)(ptr - cd->start_pattern - GET(previous_callout, 2));
4030 PUT(previous_callout, 2 + LINK_SIZE, length);
4036 /*************************************************
4037 * Get othercase range *
4038 *************************************************/
4040 /* This function is passed the start and end of a class range, in UTF-8 mode
4041 with UCP support. It searches up the characters, looking for ranges of
4042 characters in the "other" case. Each call returns the next one, updating the
4043 start address. A character with multiple other cases is returned on its own
4044 with a special return value.
4047 cptr points to starting character value; updated
4049 ocptr where to put start of othercase range
4050 odptr where to put end of othercase range
4052 Yield: -1 when no more
4053 0 when a range is returned
4054 >0 the CASESET offset for char with multiple other cases
4055 in this case, ocptr contains the original
4059 get_othercase_range(pcre_uint32 *cptr, pcre_uint32 d, pcre_uint32 *ocptr,
4062 pcre_uint32 c, othercase, next;
4065 /* Find the first character that has an other case. If it has multiple other
4066 cases, return its case offset value. */
4068 for (c = *cptr; c <= d; c++)
4070 if ((co = UCD_CASESET(c)) != 0)
4072 *ocptr = c++; /* Character that has the set */
4073 *cptr = c; /* Rest of input range */
4076 if ((othercase = UCD_OTHERCASE(c)) != c) break;
4079 if (c > d) return -1; /* Reached end of range */
4081 /* Found a character that has a single other case. Search for the end of the
4082 range, which is either the end of the input range, or a character that has zero
4083 or more than one other cases. */
4086 next = othercase + 1;
4088 for (++c; c <= d; c++)
4090 if ((co = UCD_CASESET(c)) != 0 || UCD_OTHERCASE(c) != next) break;
4094 *odptr = next - 1; /* End of othercase range */
4095 *cptr = c; /* Rest of input range */
4098 #endif /* SUPPORT_UCP */
4102 /*************************************************
4103 * Add a character or range to a class *
4104 *************************************************/
4106 /* This function packages up the logic of adding a character or range of
4107 characters to a class. The character values in the arguments will be within the
4108 valid values for the current mode (8-bit, 16-bit, UTF, etc). This function is
4109 mutually recursive with the function immediately below.
4112 classbits the bit map for characters < 256
4113 uchardptr points to the pointer for extra data
4114 options the options word
4115 cd contains pointers to tables etc.
4116 start start of range character
4117 end end of range character
4119 Returns: the number of < 256 characters added
4120 the pointer to extra data is updated
4124 add_to_class(pcre_uint8 *classbits, pcre_uchar **uchardptr, int options,
4125 compile_data *cd, pcre_uint32 start, pcre_uint32 end)
4128 pcre_uint32 classbits_end = (end <= 0xff ? end : 0xff);
4131 /* If caseless matching is required, scan the range and process alternate
4132 cases. In Unicode, there are 8-bit characters that have alternate cases that
4133 are greater than 255 and vice-versa. Sometimes we can just extend the original
4136 if ((options & PCRE_CASELESS) != 0)
4139 if ((options & PCRE_UTF8) != 0)
4144 options &= ~PCRE_CASELESS; /* Remove for recursive calls */
4147 while ((rc = get_othercase_range(&c, end, &oc, &od)) >= 0)
4149 /* Handle a single character that has more than one other case. */
4151 if (rc > 0) n8 += add_list_to_class(classbits, uchardptr, options, cd,
4152 PRIV(ucd_caseless_sets) + rc, oc);
4154 /* Do nothing if the other case range is within the original range. */
4156 else if (oc >= start && od <= end) continue;
4158 /* Extend the original range if there is overlap, noting that if oc < c, we
4159 can't have od > end because a subrange is always shorter than the basic
4160 range. Otherwise, use a recursive call to add the additional range. */
4162 else if (oc < start && od >= start - 1) start = oc; /* Extend downwards */
4163 else if (od > end && oc <= end + 1) end = od; /* Extend upwards */
4164 else n8 += add_to_class(classbits, uchardptr, options, cd, oc, od);
4168 #endif /* SUPPORT_UCP */
4170 /* Not UTF-mode, or no UCP */
4172 for (c = start; c <= classbits_end; c++)
4174 SETBIT(classbits, cd->fcc[c]);
4179 /* Now handle the original range. Adjust the final value according to the bit
4180 length - this means that the same lists of (e.g.) horizontal spaces can be used
4183 #if defined COMPILE_PCRE8
4185 if ((options & PCRE_UTF8) == 0)
4187 if (end > 0xff) end = 0xff;
4189 #elif defined COMPILE_PCRE16
4191 if ((options & PCRE_UTF16) == 0)
4193 if (end > 0xffff) end = 0xffff;
4195 #endif /* COMPILE_PCRE[8|16] */
4197 /* Use the bitmap for characters < 256. Otherwise use extra data.*/
4199 for (c = start; c <= classbits_end; c++)
4201 /* Regardless of start, c will always be <= 255. */
4202 SETBIT(classbits, c);
4206 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
4207 if (start <= 0xff) start = 0xff + 1;
4211 pcre_uchar *uchardata = *uchardptr;
4213 if ((options & PCRE_UTF8) != 0) /* All UTFs use the same flag bit */
4217 *uchardata++ = XCL_RANGE;
4218 uchardata += PRIV(ord2utf)(start, uchardata);
4219 uchardata += PRIV(ord2utf)(end, uchardata);
4221 else if (start == end)
4223 *uchardata++ = XCL_SINGLE;
4224 uchardata += PRIV(ord2utf)(start, uchardata);
4228 #endif /* SUPPORT_UTF */
4230 /* Without UTF support, character values are constrained by the bit length,
4231 and can only be > 256 for 16-bit and 32-bit libraries. */
4233 #ifdef COMPILE_PCRE8
4238 *uchardata++ = XCL_RANGE;
4239 *uchardata++ = start;
4242 else if (start == end)
4244 *uchardata++ = XCL_SINGLE;
4245 *uchardata++ = start;
4249 *uchardptr = uchardata; /* Updata extra data pointer */
4251 #endif /* SUPPORT_UTF || !COMPILE_PCRE8 */
4253 return n8; /* Number of 8-bit characters */
4259 /*************************************************
4260 * Add a list of characters to a class *
4261 *************************************************/
4263 /* This function is used for adding a list of case-equivalent characters to a
4264 class, and also for adding a list of horizontal or vertical whitespace. If the
4265 list is in order (which it should be), ranges of characters are detected and
4266 handled appropriately. This function is mutually recursive with the function
4270 classbits the bit map for characters < 256
4271 uchardptr points to the pointer for extra data
4272 options the options word
4273 cd contains pointers to tables etc.
4274 p points to row of 32-bit values, terminated by NOTACHAR
4275 except character to omit; this is used when adding lists of
4276 case-equivalent characters to avoid including the one we
4279 Returns: the number of < 256 characters added
4280 the pointer to extra data is updated
4284 add_list_to_class(pcre_uint8 *classbits, pcre_uchar **uchardptr, int options,
4285 compile_data *cd, const pcre_uint32 *p, unsigned int except)
4288 while (p[0] < NOTACHAR)
4293 while(p[n+1] == p[0] + n + 1) n++;
4294 n8 += add_to_class(classbits, uchardptr, options, cd, p[0], p[n]);
4303 /*************************************************
4304 * Add characters not in a list to a class *
4305 *************************************************/
4307 /* This function is used for adding the complement of a list of horizontal or
4308 vertical whitespace to a class. The list must be in order.
4311 classbits the bit map for characters < 256
4312 uchardptr points to the pointer for extra data
4313 options the options word
4314 cd contains pointers to tables etc.
4315 p points to row of 32-bit values, terminated by NOTACHAR
4317 Returns: the number of < 256 characters added
4318 the pointer to extra data is updated
4322 add_not_list_to_class(pcre_uint8 *classbits, pcre_uchar **uchardptr,
4323 int options, compile_data *cd, const pcre_uint32 *p)
4325 BOOL utf = (options & PCRE_UTF8) != 0;
4328 n8 += add_to_class(classbits, uchardptr, options, cd, 0, p[0] - 1);
4329 while (p[0] < NOTACHAR)
4331 while (p[1] == p[0] + 1) p++;
4332 n8 += add_to_class(classbits, uchardptr, options, cd, p[0] + 1,
4333 (p[1] == NOTACHAR) ? (utf ? 0x10ffffu : 0xffffffffu) : p[1] - 1);
4341 /*************************************************
4342 * Compile one branch *
4343 *************************************************/
4345 /* Scan the pattern, compiling it into the a vector. If the options are
4346 changed during the branch, the pointer is used to change the external options
4347 bits. This function is used during the pre-compile phase when we are trying
4348 to find out the amount of memory needed, as well as during the real compile
4349 phase. The value of lengthptr distinguishes the two phases.
4352 optionsptr pointer to the option bits
4353 codeptr points to the pointer to the current code point
4354 ptrptr points to the current pattern pointer
4355 errorcodeptr points to error code variable
4356 firstcharptr place to put the first required character
4357 firstcharflagsptr place to put the first character flags, or a negative number
4358 reqcharptr place to put the last required character
4359 reqcharflagsptr place to put the last required character flags, or a negative number
4360 bcptr points to current branch chain
4361 cond_depth conditional nesting depth
4362 cd contains pointers to tables etc.
4363 lengthptr NULL during the real compile phase
4364 points to length accumulator during pre-compile phase
4366 Returns: TRUE on success
4367 FALSE, with *errorcodeptr set non-zero on error
4371 compile_branch(int *optionsptr, pcre_uchar **codeptr,
4372 const pcre_uchar **ptrptr, int *errorcodeptr,
4373 pcre_uint32 *firstcharptr, pcre_int32 *firstcharflagsptr,
4374 pcre_uint32 *reqcharptr, pcre_int32 *reqcharflagsptr,
4375 branch_chain *bcptr, int cond_depth,
4376 compile_data *cd, int *lengthptr)
4378 int repeat_type, op_type;
4379 int repeat_min = 0, repeat_max = 0; /* To please picky compilers */
4381 int greedy_default, greedy_non_default;
4382 pcre_uint32 firstchar, reqchar;
4383 pcre_int32 firstcharflags, reqcharflags;
4384 pcre_uint32 zeroreqchar, zerofirstchar;
4385 pcre_int32 zeroreqcharflags, zerofirstcharflags;
4386 pcre_int32 req_caseopt, reqvary, tempreqvary;
4387 int options = *optionsptr; /* May change dynamically */
4388 int after_manual_callout = 0;
4389 int length_prevgroup = 0;
4390 register pcre_uint32 c;
4392 register pcre_uchar *code = *codeptr;
4393 pcre_uchar *last_code = code;
4394 pcre_uchar *orig_code = code;
4395 pcre_uchar *tempcode;
4396 BOOL inescq = FALSE;
4397 BOOL groupsetfirstchar = FALSE;
4398 const pcre_uchar *ptr = *ptrptr;
4399 const pcre_uchar *tempptr;
4400 const pcre_uchar *nestptr = NULL;
4401 pcre_uchar *previous = NULL;
4402 pcre_uchar *previous_callout = NULL;
4403 pcre_uchar *save_hwm = NULL;
4404 pcre_uint8 classbits[32];
4406 /* We can fish out the UTF-8 setting once and for all into a BOOL, but we
4407 must not do this for other options (e.g. PCRE_EXTENDED) because they may change
4408 dynamically as we process the pattern. */
4411 /* PCRE_UTF[16|32] have the same value as PCRE_UTF8. */
4412 BOOL utf = (options & PCRE_UTF8) != 0;
4413 #ifndef COMPILE_PCRE32
4414 pcre_uchar utf_chars[6];
4420 /* Helper variables for OP_XCLASS opcode (for characters > 255). We define
4421 class_uchardata always so that it can be passed to add_to_class() always,
4422 though it will not be used in non-UTF 8-bit cases. This avoids having to supply
4423 alternative calls for the different cases. */
4425 pcre_uchar *class_uchardata;
4426 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
4428 pcre_uchar *class_uchardata_base;
4432 if (lengthptr != NULL) DPRINTF((">> start branch\n"));
4435 /* Set up the default and non-default settings for greediness */
4437 greedy_default = ((options & PCRE_UNGREEDY) != 0);
4438 greedy_non_default = greedy_default ^ 1;
4440 /* Initialize no first byte, no required byte. REQ_UNSET means "no char
4441 matching encountered yet". It gets changed to REQ_NONE if we hit something that
4442 matches a non-fixed char first char; reqchar just remains unset if we never
4445 When we hit a repeat whose minimum is zero, we may have to adjust these values
4446 to take the zero repeat into account. This is implemented by setting them to
4447 zerofirstbyte and zeroreqchar when such a repeat is encountered. The individual
4448 item types that can be repeated set these backoff variables appropriately. */
4450 firstchar = reqchar = zerofirstchar = zeroreqchar = 0;
4451 firstcharflags = reqcharflags = zerofirstcharflags = zeroreqcharflags = REQ_UNSET;
4453 /* The variable req_caseopt contains either the REQ_CASELESS value
4454 or zero, according to the current setting of the caseless flag. The
4455 REQ_CASELESS leaves the lower 28 bit empty. It is added into the
4456 firstchar or reqchar variables to record the case status of the
4457 value. This is used only for ASCII characters. */
4459 req_caseopt = ((options & PCRE_CASELESS) != 0)? REQ_CASELESS:0;
4461 /* Switch on next character until the end of the branch */
4466 BOOL should_flip_negation;
4467 BOOL possessive_quantifier;
4470 BOOL reset_bracount;
4471 int class_has_8bitchar;
4473 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
4474 BOOL xclass_has_prop;
4480 pcre_uint32 subreqchar, subfirstchar;
4481 pcre_int32 subreqcharflags, subfirstcharflags;
4483 unsigned int mclength;
4484 unsigned int tempbracount;
4486 pcre_uchar mcbuffer[8];
4488 /* Get next character in the pattern */
4492 /* If we are at the end of a nested substitution, revert to the outer level
4493 string. Nesting only happens one level deep. */
4495 if (c == CHAR_NULL && nestptr != NULL)
4502 /* If we are in the pre-compile phase, accumulate the length used for the
4503 previous cycle of this loop. */
4505 if (lengthptr != NULL)
4508 if (code > cd->hwm) cd->hwm = code; /* High water info */
4510 if (code > cd->start_workspace + cd->workspace_size -
4511 WORK_SIZE_SAFETY_MARGIN) /* Check for overrun */
4513 *errorcodeptr = ERR52;
4517 /* There is at least one situation where code goes backwards: this is the
4518 case of a zero quantifier after a class (e.g. [ab]{0}). At compile time,
4519 the class is simply eliminated. However, it is created first, so we have to
4520 allow memory for it. Therefore, don't ever reduce the length at this point.
4523 if (code < last_code) code = last_code;
4525 /* Paranoid check for integer overflow */
4527 if (OFLOW_MAX - *lengthptr < code - last_code)
4529 *errorcodeptr = ERR20;
4533 *lengthptr += (int)(code - last_code);
4534 DPRINTF(("length=%d added %d c=%c (0x%x)\n", *lengthptr,
4535 (int)(code - last_code), c, c));
4537 /* If "previous" is set and it is not at the start of the work space, move
4538 it back to there, in order to avoid filling up the work space. Otherwise,
4539 if "previous" is NULL, reset the current code pointer to the start. */
4541 if (previous != NULL)
4543 if (previous > orig_code)
4545 memmove(orig_code, previous, IN_UCHARS(code - previous));
4546 code -= previous - orig_code;
4547 previous = orig_code;
4550 else code = orig_code;
4552 /* Remember where this code item starts so we can pick up the length
4558 /* In the real compile phase, just check the workspace used by the forward
4561 else if (cd->hwm > cd->start_workspace + cd->workspace_size -
4562 WORK_SIZE_SAFETY_MARGIN)
4564 *errorcodeptr = ERR52;
4568 /* If in \Q...\E, check for the end; if not, we have a literal */
4570 if (inescq && c != CHAR_NULL)
4572 if (c == CHAR_BACKSLASH && ptr[1] == CHAR_E)
4580 if (previous_callout != NULL)
4582 if (lengthptr == NULL) /* Don't attempt in pre-compile phase */
4583 complete_callout(previous_callout, ptr, cd);
4584 previous_callout = NULL;
4586 if ((options & PCRE_AUTO_CALLOUT) != 0)
4588 previous_callout = code;
4589 code = auto_callout(code, ptr, cd);
4593 /* Control does not reach here. */
4596 /* In extended mode, skip white space and comments. We need a loop in order
4597 to check for more white space and more comments after a comment. */
4599 if ((options & PCRE_EXTENDED) != 0)
4603 while (MAX_255(c) && (cd->ctypes[c] & ctype_space) != 0) c = *(++ptr);
4604 if (c != CHAR_NUMBER_SIGN) break;
4606 while (*ptr != CHAR_NULL)
4608 if (IS_NEWLINE(ptr)) /* For non-fixed-length newline cases, */
4609 { /* IS_NEWLINE sets cd->nllen. */
4615 if (utf) FORWARDCHAR(ptr);
4618 c = *ptr; /* Either NULL or the char after a newline */
4622 /* See if the next thing is a quantifier. */
4625 c == CHAR_ASTERISK || c == CHAR_PLUS || c == CHAR_QUESTION_MARK ||
4626 (c == CHAR_LEFT_CURLY_BRACKET && is_counted_repeat(ptr+1));
4628 /* Fill in length of a previous callout, except when the next thing is a
4629 quantifier or when processing a property substitution string in UCP mode. */
4631 if (!is_quantifier && previous_callout != NULL && nestptr == NULL &&
4632 after_manual_callout-- <= 0)
4634 if (lengthptr == NULL) /* Don't attempt in pre-compile phase */
4635 complete_callout(previous_callout, ptr, cd);
4636 previous_callout = NULL;
4639 /* Create auto callout, except for quantifiers, or while processing property
4640 strings that are substituted for \w etc in UCP mode. */
4642 if ((options & PCRE_AUTO_CALLOUT) != 0 && !is_quantifier && nestptr == NULL)
4644 previous_callout = code;
4645 code = auto_callout(code, ptr, cd);
4648 /* Process the next pattern item. */
4652 /* ===================================================================*/
4653 case CHAR_NULL: /* The branch terminates at string end */
4654 case CHAR_VERTICAL_LINE: /* or | or ) */
4655 case CHAR_RIGHT_PARENTHESIS:
4656 *firstcharptr = firstchar;
4657 *firstcharflagsptr = firstcharflags;
4658 *reqcharptr = reqchar;
4659 *reqcharflagsptr = reqcharflags;
4662 if (lengthptr != NULL)
4664 if (OFLOW_MAX - *lengthptr < code - last_code)
4666 *errorcodeptr = ERR20;
4669 *lengthptr += (int)(code - last_code); /* To include callout length */
4670 DPRINTF((">> end branch\n"));
4675 /* ===================================================================*/
4676 /* Handle single-character metacharacters. In multiline mode, ^ disables
4677 the setting of any following char as a first character. */
4679 case CHAR_CIRCUMFLEX_ACCENT:
4681 if ((options & PCRE_MULTILINE) != 0)
4683 if (firstcharflags == REQ_UNSET) firstcharflags = REQ_NONE;
4686 else *code++ = OP_CIRC;
4689 case CHAR_DOLLAR_SIGN:
4691 *code++ = ((options & PCRE_MULTILINE) != 0)? OP_DOLLM : OP_DOLL;
4694 /* There can never be a first char if '.' is first, whatever happens about
4695 repeats. The value of reqchar doesn't change either. */
4698 if (firstcharflags == REQ_UNSET) firstcharflags = REQ_NONE;
4699 zerofirstchar = firstchar;
4700 zerofirstcharflags = firstcharflags;
4701 zeroreqchar = reqchar;
4702 zeroreqcharflags = reqcharflags;
4704 *code++ = ((options & PCRE_DOTALL) != 0)? OP_ALLANY: OP_ANY;
4708 /* ===================================================================*/
4709 /* Character classes. If the included characters are all < 256, we build a
4710 32-byte bitmap of the permitted characters, except in the special case
4711 where there is only one such character. For negated classes, we build the
4712 map as usual, then invert it at the end. However, we use a different opcode
4713 so that data characters > 255 can be handled correctly.
4715 If the class contains characters outside the 0-255 range, a different
4716 opcode is compiled. It may optionally have a bit map for characters < 256,
4717 but those above are are explicitly listed afterwards. A flag byte tells
4718 whether the bitmap is present, and whether this is a negated class or not.
4720 In JavaScript compatibility mode, an isolated ']' causes an error. In
4721 default (Perl) mode, it is treated as a data character. */
4723 case CHAR_RIGHT_SQUARE_BRACKET:
4724 if ((cd->external_options & PCRE_JAVASCRIPT_COMPAT) != 0)
4726 *errorcodeptr = ERR64;
4731 /* In another (POSIX) regex library, the ugly syntax [[:<:]] and [[:>:]] is
4732 used for "start of word" and "end of word". As these are otherwise illegal
4733 sequences, we don't break anything by recognizing them. They are replaced
4734 by \b(?=\w) and \b(?<=\w) respectively. Sequences like [a[:<:]] are
4735 erroneous and are handled by the normal code below. */
4737 case CHAR_LEFT_SQUARE_BRACKET:
4738 if (STRNCMP_UC_C8(ptr+1, STRING_WEIRD_STARTWORD, 6) == 0)
4741 ptr = sub_start_of_word - 1;
4745 if (STRNCMP_UC_C8(ptr+1, STRING_WEIRD_ENDWORD, 6) == 0)
4748 ptr = sub_end_of_word - 1;
4752 /* Handle a real character class. */
4756 /* PCRE supports POSIX class stuff inside a class. Perl gives an error if
4757 they are encountered at the top level, so we'll do that too. */
4759 if ((ptr[1] == CHAR_COLON || ptr[1] == CHAR_DOT ||
4760 ptr[1] == CHAR_EQUALS_SIGN) &&
4761 check_posix_syntax(ptr, &tempptr))
4763 *errorcodeptr = (ptr[1] == CHAR_COLON)? ERR13 : ERR31;
4767 /* If the first character is '^', set the negation flag and skip it. Also,
4768 if the first few characters (either before or after ^) are \Q\E or \E we
4769 skip them too. This makes for compatibility with Perl. */
4771 negate_class = FALSE;
4775 if (c == CHAR_BACKSLASH)
4777 if (ptr[1] == CHAR_E)
4779 else if (STRNCMP_UC_C8(ptr + 1, STR_Q STR_BACKSLASH STR_E, 3) == 0)
4784 else if (!negate_class && c == CHAR_CIRCUMFLEX_ACCENT)
4785 negate_class = TRUE;
4789 /* Empty classes are allowed in JavaScript compatibility mode. Otherwise,
4790 an initial ']' is taken as a data character -- the code below handles
4791 that. In JS mode, [] must always fail, so generate OP_FAIL, whereas
4792 [^] must match any character, so generate OP_ALLANY. */
4794 if (c == CHAR_RIGHT_SQUARE_BRACKET &&
4795 (cd->external_options & PCRE_JAVASCRIPT_COMPAT) != 0)
4797 *code++ = negate_class? OP_ALLANY : OP_FAIL;
4798 if (firstcharflags == REQ_UNSET) firstcharflags = REQ_NONE;
4799 zerofirstchar = firstchar;
4800 zerofirstcharflags = firstcharflags;
4804 /* If a class contains a negative special such as \S, we need to flip the
4805 negation flag at the end, so that support for characters > 255 works
4806 correctly (they are all included in the class). */
4808 should_flip_negation = FALSE;
4810 /* Extended class (xclass) will be used when characters > 255
4813 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
4815 class_uchardata = code + LINK_SIZE + 2; /* For XCLASS items */
4816 class_uchardata_base = class_uchardata; /* Save the start */
4819 /* For optimization purposes, we track some properties of the class:
4820 class_has_8bitchar will be non-zero if the class contains at least one <
4821 256 character; class_one_char will be 1 if the class contains just one
4822 character; xclass_has_prop will be TRUE if unicode property checks
4823 are present in the class. */
4825 class_has_8bitchar = 0;
4827 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
4828 xclass_has_prop = FALSE;
4831 /* Initialize the 32-char bit map to all zeros. We build the map in a
4832 temporary bit of memory, in case the class contains fewer than two
4833 8-bit characters because in that case the compiled code doesn't use the bit
4836 memset(classbits, 0, 32 * sizeof(pcre_uint8));
4838 /* Process characters until ] is reached. By writing this as a "do" it
4839 means that an initial ] is taken as a data character. At the start of the
4840 loop, c contains the first byte of the character. */
4842 if (c != CHAR_NULL) do
4844 const pcre_uchar *oldptr;
4847 if (utf && HAS_EXTRALEN(c))
4848 { /* Braces are required because the */
4849 GETCHARLEN(c, ptr, ptr); /* macro generates multiple statements */
4853 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
4854 /* In the pre-compile phase, accumulate the length of any extra
4855 data and reset the pointer. This is so that very large classes that
4856 contain a zillion > 255 characters no longer overwrite the work space
4857 (which is on the stack). We have to remember that there was XCLASS data,
4860 if (lengthptr != NULL && class_uchardata > class_uchardata_base)
4863 *lengthptr += class_uchardata - class_uchardata_base;
4864 class_uchardata = class_uchardata_base;
4868 /* Inside \Q...\E everything is literal except \E */
4872 if (c == CHAR_BACKSLASH && ptr[1] == CHAR_E) /* If we are at \E */
4874 inescq = FALSE; /* Reset literal state */
4875 ptr++; /* Skip the 'E' */
4876 continue; /* Carry on with next */
4878 goto CHECK_RANGE; /* Could be range if \E follows */
4881 /* Handle POSIX class names. Perl allows a negation extension of the
4882 form [:^name:]. A square bracket that doesn't match the syntax is
4883 treated as a literal. We also recognize the POSIX constructions
4884 [.ch.] and [=ch=] ("collating elements") and fault them, as Perl
4887 if (c == CHAR_LEFT_SQUARE_BRACKET &&
4888 (ptr[1] == CHAR_COLON || ptr[1] == CHAR_DOT ||
4889 ptr[1] == CHAR_EQUALS_SIGN) && check_posix_syntax(ptr, &tempptr))
4891 BOOL local_negate = FALSE;
4892 int posix_class, taboffset, tabopt;
4893 register const pcre_uint8 *cbits = cd->cbits;
4894 pcre_uint8 pbits[32];
4896 if (ptr[1] != CHAR_COLON)
4898 *errorcodeptr = ERR31;
4903 if (*ptr == CHAR_CIRCUMFLEX_ACCENT)
4905 local_negate = TRUE;
4906 should_flip_negation = TRUE; /* Note negative special */
4910 posix_class = check_posix_name(ptr, (int)(tempptr - ptr));
4911 if (posix_class < 0)
4913 *errorcodeptr = ERR30;
4917 /* If matching is caseless, upper and lower are converted to
4918 alpha. This relies on the fact that the class table starts with
4919 alpha, lower, upper as the first 3 entries. */
4921 if ((options & PCRE_CASELESS) != 0 && posix_class <= 2)
4924 /* When PCRE_UCP is set, some of the POSIX classes are converted to
4925 different escape sequences that use Unicode properties \p or \P. Others
4926 that are not available via \p or \P generate XCL_PROP/XCL_NOTPROP
4930 if ((options & PCRE_UCP) != 0)
4932 unsigned int ptype = 0;
4933 int pc = posix_class + ((local_negate)? POSIX_SUBSIZE/2 : 0);
4935 /* The posix_substitutes table specifies which POSIX classes can be
4936 converted to \p or \P items. */
4938 if (posix_substitutes[pc] != NULL)
4940 nestptr = tempptr + 1;
4941 ptr = posix_substitutes[pc] - 1;
4945 /* There are three other classes that generate special property calls
4946 that are recognized only in an XCLASS. */
4948 else switch(posix_class)
4954 if (ptype == 0) ptype = PT_PXPRINT;
4957 if (ptype == 0) ptype = PT_PXPUNCT;
4958 *class_uchardata++ = local_negate? XCL_NOTPROP : XCL_PROP;
4959 *class_uchardata++ = ptype;
4960 *class_uchardata++ = 0;
4961 xclass_has_prop = TRUE;
4965 /* For all other POSIX classes, no special action is taken in UCP
4966 mode. Fall through to the non_UCP case. */
4973 /* In the non-UCP case, or when UCP makes no difference, we build the
4974 bit map for the POSIX class in a chunk of local store because we may be
4975 adding and subtracting from it, and we don't want to subtract bits that
4976 may be in the main map already. At the end we or the result into the
4977 bit map that is being built. */
4981 /* Copy in the first table (always present) */
4983 memcpy(pbits, cbits + posix_class_maps[posix_class],
4984 32 * sizeof(pcre_uint8));
4986 /* If there is a second table, add or remove it as required. */
4988 taboffset = posix_class_maps[posix_class + 1];
4989 tabopt = posix_class_maps[posix_class + 2];
4994 for (c = 0; c < 32; c++) pbits[c] |= cbits[c + taboffset];
4996 for (c = 0; c < 32; c++) pbits[c] &= ~cbits[c + taboffset];
4999 /* Now see if we need to remove any special characters. An option
5000 value of 1 removes vertical space and 2 removes underscore. */
5002 if (tabopt < 0) tabopt = -tabopt;
5003 if (tabopt == 1) pbits[1] &= ~0x3c;
5004 else if (tabopt == 2) pbits[11] &= 0x7f;
5006 /* Add the POSIX table or its complement into the main table that is
5007 being built and we are done. */
5010 for (c = 0; c < 32; c++) classbits[c] |= ~pbits[c];
5012 for (c = 0; c < 32; c++) classbits[c] |= pbits[c];
5015 /* Every class contains at least one < 256 character. */
5016 class_has_8bitchar = 1;
5017 /* Every class contains at least two characters. */
5019 continue; /* End of POSIX syntax handling */
5022 /* Backslash may introduce a single character, or it may introduce one
5023 of the specials, which just set a flag. The sequence \b is a special
5024 case. Inside a class (and only there) it is treated as backspace. We
5025 assume that other escapes have more than one character in them, so
5026 speculatively set both class_has_8bitchar and class_one_char bigger
5027 than one. Unrecognized escapes fall through and are either treated
5028 as literal characters (by default), or are faulted if
5029 PCRE_EXTRA is set. */
5031 if (c == CHAR_BACKSLASH)
5033 escape = check_escape(&ptr, &ec, errorcodeptr, cd->bracount, options,
5035 if (*errorcodeptr != 0) goto FAILED;
5036 if (escape == 0) c = ec;
5037 else if (escape == ESC_b) c = CHAR_BS; /* \b is backspace in a class */
5038 else if (escape == ESC_N) /* \N is not supported in a class */
5040 *errorcodeptr = ERR71;
5043 else if (escape == ESC_Q) /* Handle start of quoted string */
5045 if (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
5047 ptr += 2; /* avoid empty string */
5052 else if (escape == ESC_E) continue; /* Ignore orphan \E */
5056 register const pcre_uint8 *cbits = cd->cbits;
5057 /* Every class contains at least two < 256 characters. */
5058 class_has_8bitchar++;
5059 /* Every class contains at least two characters. */
5060 class_one_char += 2;
5065 case ESC_du: /* These are the values given for \d etc */
5066 case ESC_DU: /* when PCRE_UCP is set. We replace the */
5067 case ESC_wu: /* escape sequence with an appropriate \p */
5068 case ESC_WU: /* or \P to test Unicode properties instead */
5069 case ESC_su: /* of the default ASCII testing. */
5072 ptr = substitutes[escape - ESC_DU] - 1; /* Just before substitute */
5073 class_has_8bitchar--; /* Undo! */
5077 for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_digit];
5081 should_flip_negation = TRUE;
5082 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_digit];
5086 for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_word];
5090 should_flip_negation = TRUE;
5091 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_word];
5094 /* Perl 5.004 onwards omitted VT from \s, but restored it at Perl
5095 5.18. Before PCRE 8.34, we had to preserve the VT bit if it was
5096 previously set by something earlier in the character class.
5097 Luckily, the value of CHAR_VT is 0x0b in both ASCII and EBCDIC, so
5098 we could just adjust the appropriate bit. From PCRE 8.34 we no
5099 longer treat \s and \S specially. */
5102 for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_space];
5106 should_flip_negation = TRUE;
5107 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_space];
5110 /* The rest apply in both UCP and non-UCP cases. */
5113 (void)add_list_to_class(classbits, &class_uchardata, options, cd,
5114 PRIV(hspace_list), NOTACHAR);
5118 (void)add_not_list_to_class(classbits, &class_uchardata, options,
5119 cd, PRIV(hspace_list));
5123 (void)add_list_to_class(classbits, &class_uchardata, options, cd,
5124 PRIV(vspace_list), NOTACHAR);
5128 (void)add_not_list_to_class(classbits, &class_uchardata, options,
5129 cd, PRIV(vspace_list));
5137 unsigned int ptype = 0, pdata = 0;
5138 if (!get_ucp(&ptr, &negated, &ptype, &pdata, errorcodeptr))
5140 *class_uchardata++ = ((escape == ESC_p) != negated)?
5141 XCL_PROP : XCL_NOTPROP;
5142 *class_uchardata++ = ptype;
5143 *class_uchardata++ = pdata;
5144 xclass_has_prop = TRUE;
5145 class_has_8bitchar--; /* Undo! */
5149 /* Unrecognized escapes are faulted if PCRE is running in its
5150 strict mode. By default, for compatibility with Perl, they are
5151 treated as literals. */
5154 if ((options & PCRE_EXTRA) != 0)
5156 *errorcodeptr = ERR7;
5159 class_has_8bitchar--; /* Undo the speculative increase. */
5160 class_one_char -= 2; /* Undo the speculative increase. */
5161 c = *ptr; /* Get the final character and fall through */
5166 /* Fall through if the escape just defined a single character (c >= 0).
5167 This may be greater than 256. */
5171 } /* End of backslash handling */
5173 /* A character may be followed by '-' to form a range. However, Perl does
5174 not permit ']' to be the end of the range. A '-' character at the end is
5175 treated as a literal. Perl ignores orphaned \E sequences entirely. The
5176 code for handling \Q and \E is messy. */
5179 while (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
5186 /* Remember if \r or \n were explicitly used */
5188 if (c == CHAR_CR || c == CHAR_NL) cd->external_flags |= PCRE_HASCRORLF;
5190 /* Check for range */
5192 if (!inescq && ptr[1] == CHAR_MINUS)
5196 while (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_E) ptr += 2;
5198 /* If we hit \Q (not followed by \E) at this point, go into escaped
5201 while (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_Q)
5204 if (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_E)
5205 { ptr += 2; continue; }
5210 /* Minus (hyphen) at the end of a class is treated as a literal, so put
5211 back the pointer and jump to handle the character that preceded it. */
5213 if (*ptr == CHAR_NULL || (!inescq && *ptr == CHAR_RIGHT_SQUARE_BRACKET))
5216 goto CLASS_SINGLE_CHARACTER;
5219 /* Otherwise, we have a potential range; pick up the next character */
5223 { /* Braces are required because the */
5224 GETCHARLEN(d, ptr, ptr); /* macro generates multiple statements */
5228 d = *ptr; /* Not UTF-8 mode */
5230 /* The second part of a range can be a single-character escape
5231 sequence, but not any of the other escapes. Perl treats a hyphen as a
5232 literal in such circumstances. However, in Perl's warning mode, a
5233 warning is given, so PCRE now faults it as it is almost certainly a
5234 mistake on the user's part. */
5238 if (d == CHAR_BACKSLASH)
5241 descape = check_escape(&ptr, &d, errorcodeptr, cd->bracount, options, TRUE);
5242 if (*errorcodeptr != 0) goto FAILED;
5244 /* 0 means a character was put into d; \b is backspace; any other
5245 special causes an error. */
5249 if (descape == ESC_b) d = CHAR_BS; else
5251 *errorcodeptr = ERR83;
5257 /* A hyphen followed by a POSIX class is treated in the same way. */
5259 else if (d == CHAR_LEFT_SQUARE_BRACKET &&
5260 (ptr[1] == CHAR_COLON || ptr[1] == CHAR_DOT ||
5261 ptr[1] == CHAR_EQUALS_SIGN) &&
5262 check_posix_syntax(ptr, &tempptr))
5264 *errorcodeptr = ERR83;
5269 /* Check that the two values are in the correct order. Optimize
5270 one-character ranges. */
5274 *errorcodeptr = ERR8;
5277 if (d == c) goto CLASS_SINGLE_CHARACTER; /* A few lines below */
5279 /* We have found a character range, so single character optimizations
5280 cannot be done anymore. Any value greater than 1 indicates that there
5281 is more than one character. */
5285 /* Remember an explicit \r or \n, and add the range to the class. */
5287 if (d == CHAR_CR || d == CHAR_NL) cd->external_flags |= PCRE_HASCRORLF;
5289 class_has_8bitchar +=
5290 add_to_class(classbits, &class_uchardata, options, cd, c, d);
5292 continue; /* Go get the next char in the class */
5295 /* Handle a single character - we can get here for a normal non-escape
5296 char, or after \ that introduces a single character or for an apparent
5297 range that isn't. Only the value 1 matters for class_one_char, so don't
5298 increase it if it is already 2 or more ... just in case there's a class
5299 with a zillion characters in it. */
5301 CLASS_SINGLE_CHARACTER:
5302 if (class_one_char < 2) class_one_char++;
5304 /* If class_one_char is 1, we have the first single character in the
5305 class, and there have been no prior ranges, or XCLASS items generated by
5306 escapes. If this is the final character in the class, we can optimize by
5307 turning the item into a 1-character OP_CHAR[I] if it's positive, or
5308 OP_NOT[I] if it's negative. In the positive case, it can cause firstchar
5309 to be set. Otherwise, there can be no first char if this item is first,
5310 whatever repeat count may follow. In the case of reqchar, save the
5311 previous value for reinstating. */
5313 if (class_one_char == 1 && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET)
5316 zeroreqchar = reqchar;
5317 zeroreqcharflags = reqcharflags;
5324 if (firstcharflags == REQ_UNSET) firstcharflags = REQ_NONE;
5325 zerofirstchar = firstchar;
5326 zerofirstcharflags = firstcharflags;
5328 /* For caseless UTF-8 mode when UCP support is available, check
5329 whether this character has more than one other case. If so, generate
5330 a special OP_NOTPROP item instead of OP_NOTI. */
5333 if (utf && (options & PCRE_CASELESS) != 0 &&
5334 (d = UCD_CASESET(c)) != 0)
5336 *code++ = OP_NOTPROP;
5342 /* Char has only one other case, or UCP not available */
5345 *code++ = ((options & PCRE_CASELESS) != 0)? OP_NOTI: OP_NOT;
5346 #if defined SUPPORT_UTF && !defined COMPILE_PCRE32
5347 if (utf && c > MAX_VALUE_FOR_SINGLE_CHAR)
5348 code += PRIV(ord2utf)(c, code);
5354 /* We are finished with this character class */
5359 /* For a single, positive character, get the value into mcbuffer, and
5360 then we can handle this with the normal one-character code. */
5362 #if defined SUPPORT_UTF && !defined COMPILE_PCRE32
5363 if (utf && c > MAX_VALUE_FOR_SINGLE_CHAR)
5364 mclength = PRIV(ord2utf)(c, mcbuffer);
5372 } /* End of 1-char optimization */
5374 /* There is more than one character in the class, or an XCLASS item
5375 has been generated. Add this character to the class. */
5377 class_has_8bitchar +=
5378 add_to_class(classbits, &class_uchardata, options, cd, c, c);
5381 /* Loop until ']' reached. This "while" is the end of the "do" far above.
5382 If we are at the end of an internal nested string, revert to the outer
5385 while (((c = *(++ptr)) != CHAR_NULL ||
5387 (ptr = nestptr, nestptr = NULL, c = *(++ptr)) != CHAR_NULL)) &&
5388 (c != CHAR_RIGHT_SQUARE_BRACKET || inescq));
5390 /* Check for missing terminating ']' */
5394 *errorcodeptr = ERR6;
5398 /* We will need an XCLASS if data has been placed in class_uchardata. In
5399 the second phase this is a sufficient test. However, in the pre-compile
5400 phase, class_uchardata gets emptied to prevent workspace overflow, so it
5401 only if the very last character in the class needs XCLASS will it contain
5402 anything at this point. For this reason, xclass gets set TRUE above when
5403 uchar_classdata is emptied, and that's why this code is the way it is here
5404 instead of just doing a test on class_uchardata below. */
5406 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
5407 if (class_uchardata > class_uchardata_base) xclass = TRUE;
5410 /* If this is the first thing in the branch, there can be no first char
5411 setting, whatever the repeat count. Any reqchar setting must remain
5412 unchanged after any kind of repeat. */
5414 if (firstcharflags == REQ_UNSET) firstcharflags = REQ_NONE;
5415 zerofirstchar = firstchar;
5416 zerofirstcharflags = firstcharflags;
5417 zeroreqchar = reqchar;
5418 zeroreqcharflags = reqcharflags;
5420 /* If there are characters with values > 255, we have to compile an
5421 extended class, with its own opcode, unless there was a negated special
5422 such as \S in the class, and PCRE_UCP is not set, because in that case all
5423 characters > 255 are in the class, so any that were explicitly given as
5424 well can be ignored. If (when there are explicit characters > 255 that must
5425 be listed) there are no characters < 256, we can omit the bitmap in the
5426 actual compiled code. */
5429 if (xclass && (!should_flip_negation || (options & PCRE_UCP) != 0))
5430 #elif !defined COMPILE_PCRE8
5431 if (xclass && !should_flip_negation)
5433 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
5435 *class_uchardata++ = XCL_END; /* Marks the end of extra data */
5436 *code++ = OP_XCLASS;
5438 *code = negate_class? XCL_NOT:0;
5439 if (xclass_has_prop) *code |= XCL_HASPROP;
5441 /* If the map is required, move up the extra data to make room for it;
5442 otherwise just move the code pointer to the end of the extra data. */
5444 if (class_has_8bitchar > 0)
5447 memmove(code + (32 / sizeof(pcre_uchar)), code,
5448 IN_UCHARS(class_uchardata - code));
5449 if (negate_class && !xclass_has_prop)
5450 for (c = 0; c < 32; c++) classbits[c] = ~classbits[c];
5451 memcpy(code, classbits, 32);
5452 code = class_uchardata + (32 / sizeof(pcre_uchar));
5454 else code = class_uchardata;
5456 /* Now fill in the complete length of the item */
5458 PUT(previous, 1, (int)(code - previous));
5459 break; /* End of class handling */
5463 /* If there are no characters > 255, or they are all to be included or
5464 excluded, set the opcode to OP_CLASS or OP_NCLASS, depending on whether the
5465 whole class was negated and whether there were negative specials such as \S
5466 (non-UCP) in the class. Then copy the 32-byte map into the code vector,
5467 negating it if necessary. */
5469 *code++ = (negate_class == should_flip_negation) ? OP_CLASS : OP_NCLASS;
5470 if (lengthptr == NULL) /* Save time in the pre-compile phase */
5473 for (c = 0; c < 32; c++) classbits[c] = ~classbits[c];
5474 memcpy(code, classbits, 32);
5476 code += 32 / sizeof(pcre_uchar);
5482 /* ===================================================================*/
5483 /* Various kinds of repeat; '{' is not necessarily a quantifier, but this
5484 has been tested above. */
5486 case CHAR_LEFT_CURLY_BRACKET:
5487 if (!is_quantifier) goto NORMAL_CHAR;
5488 ptr = read_repeat_counts(ptr+1, &repeat_min, &repeat_max, errorcodeptr);
5489 if (*errorcodeptr != 0) goto FAILED;
5502 case CHAR_QUESTION_MARK:
5507 if (previous == NULL)
5509 *errorcodeptr = ERR9;
5513 if (repeat_min == 0)
5515 firstchar = zerofirstchar; /* Adjust for zero repeat */
5516 firstcharflags = zerofirstcharflags;
5517 reqchar = zeroreqchar; /* Ditto */
5518 reqcharflags = zeroreqcharflags;
5521 /* Remember whether this is a variable length repeat */
5523 reqvary = (repeat_min == repeat_max)? 0 : REQ_VARY;
5525 op_type = 0; /* Default single-char op codes */
5526 possessive_quantifier = FALSE; /* Default not possessive quantifier */
5528 /* Save start of previous item, in case we have to move it up in order to
5529 insert something before it. */
5531 tempcode = previous;
5533 /* Before checking for a possessive quantifier, we must skip over
5534 whitespace and comments in extended mode because Perl allows white space at
5537 if ((options & PCRE_EXTENDED) != 0)
5539 const pcre_uchar *p = ptr + 1;
5542 while (MAX_255(*p) && (cd->ctypes[*p] & ctype_space) != 0) p++;
5543 if (*p != CHAR_NUMBER_SIGN) break;
5545 while (*p != CHAR_NULL)
5547 if (IS_NEWLINE(p)) /* For non-fixed-length newline cases, */
5548 { /* IS_NEWLINE sets cd->nllen. */
5554 if (utf) FORWARDCHAR(p);
5556 } /* Loop for comment characters */
5557 } /* Loop for multiple comments */
5558 ptr = p - 1; /* Character before the next significant one. */
5561 /* If the next character is '+', we have a possessive quantifier. This
5562 implies greediness, whatever the setting of the PCRE_UNGREEDY option.
5563 If the next character is '?' this is a minimizing repeat, by default,
5564 but if PCRE_UNGREEDY is set, it works the other way round. We change the
5565 repeat type to the non-default. */
5567 if (ptr[1] == CHAR_PLUS)
5569 repeat_type = 0; /* Force greedy */
5570 possessive_quantifier = TRUE;
5573 else if (ptr[1] == CHAR_QUESTION_MARK)
5575 repeat_type = greedy_non_default;
5578 else repeat_type = greedy_default;
5580 /* If previous was a recursion call, wrap it in atomic brackets so that
5581 previous becomes the atomic group. All recursions were so wrapped in the
5582 past, but it no longer happens for non-repeated recursions. In fact, the
5583 repeated ones could be re-implemented independently so as not to need this,
5584 but for the moment we rely on the code for repeating groups. */
5586 if (*previous == OP_RECURSE)
5588 memmove(previous + 1 + LINK_SIZE, previous, IN_UCHARS(1 + LINK_SIZE));
5589 *previous = OP_ONCE;
5590 PUT(previous, 1, 2 + 2*LINK_SIZE);
5591 previous[2 + 2*LINK_SIZE] = OP_KET;
5592 PUT(previous, 3 + 2*LINK_SIZE, 2 + 2*LINK_SIZE);
5593 code += 2 + 2 * LINK_SIZE;
5594 length_prevgroup = 3 + 3*LINK_SIZE;
5596 /* When actually compiling, we need to check whether this was a forward
5597 reference, and if so, adjust the offset. */
5599 if (lengthptr == NULL && cd->hwm >= cd->start_workspace + LINK_SIZE)
5601 int offset = GET(cd->hwm, -LINK_SIZE);
5602 if (offset == previous + 1 - cd->start_code)
5603 PUT(cd->hwm, -LINK_SIZE, offset + 1 + LINK_SIZE);
5607 /* Now handle repetition for the different types of item. */
5609 /* If previous was a character or negated character match, abolish the item
5610 and generate a repeat item instead. If a char item has a minimum of more
5611 than one, ensure that it is set in reqchar - it might not be if a sequence
5612 such as x{3} is the first thing in a branch because the x will have gone
5613 into firstchar instead. */
5615 if (*previous == OP_CHAR || *previous == OP_CHARI
5616 || *previous == OP_NOT || *previous == OP_NOTI)
5620 default: /* Make compiler happy. */
5621 case OP_CHAR: op_type = OP_STAR - OP_STAR; break;
5622 case OP_CHARI: op_type = OP_STARI - OP_STAR; break;
5623 case OP_NOT: op_type = OP_NOTSTAR - OP_STAR; break;
5624 case OP_NOTI: op_type = OP_NOTSTARI - OP_STAR; break;
5627 /* Deal with UTF characters that take up more than one character. It's
5628 easier to write this out separately than try to macrify it. Use c to
5629 hold the length of the character in bytes, plus UTF_LENGTH to flag that
5630 it's a length rather than a small character. */
5632 #if defined SUPPORT_UTF && !defined COMPILE_PCRE32
5633 if (utf && NOT_FIRSTCHAR(code[-1]))
5635 pcre_uchar *lastchar = code - 1;
5637 c = (int)(code - lastchar); /* Length of UTF-8 character */
5638 memcpy(utf_chars, lastchar, IN_UCHARS(c)); /* Save the char */
5639 c |= UTF_LENGTH; /* Flag c as a length */
5642 #endif /* SUPPORT_UTF */
5644 /* Handle the case of a single charater - either with no UTF support, or
5645 with UTF disabled, or for a single character UTF character. */
5648 if (*previous <= OP_CHARI && repeat_min > 1)
5651 reqcharflags = req_caseopt | cd->req_varyopt;
5655 goto OUTPUT_SINGLE_REPEAT; /* Code shared with single character types */
5658 /* If previous was a character type match (\d or similar), abolish it and
5659 create a suitable repeat item. The code is shared with single-character
5660 repeats by setting op_type to add a suitable offset into repeat_type. Note
5661 the the Unicode property types will be present only when SUPPORT_UCP is
5662 defined, but we don't wrap the little bits of code here because it just
5663 makes it horribly messy. */
5665 else if (*previous < OP_EODN)
5667 pcre_uchar *oldcode;
5668 int prop_type, prop_value;
5669 op_type = OP_TYPESTAR - OP_STAR; /* Use type opcodes */
5672 OUTPUT_SINGLE_REPEAT:
5673 if (*previous == OP_PROP || *previous == OP_NOTPROP)
5675 prop_type = previous[1];
5676 prop_value = previous[2];
5678 else prop_type = prop_value = -1;
5681 code = previous; /* Usually overwrite previous item */
5683 /* If the maximum is zero then the minimum must also be zero; Perl allows
5684 this case, so we do too - by simply omitting the item altogether. */
5686 if (repeat_max == 0) goto END_REPEAT;
5688 /* Combine the op_type with the repeat_type */
5690 repeat_type += op_type;
5692 /* A minimum of zero is handled either as the special case * or ?, or as
5693 an UPTO, with the maximum given. */
5695 if (repeat_min == 0)
5697 if (repeat_max == -1) *code++ = OP_STAR + repeat_type;
5698 else if (repeat_max == 1) *code++ = OP_QUERY + repeat_type;
5701 *code++ = OP_UPTO + repeat_type;
5702 PUT2INC(code, 0, repeat_max);
5706 /* A repeat minimum of 1 is optimized into some special cases. If the
5707 maximum is unlimited, we use OP_PLUS. Otherwise, the original item is
5708 left in place and, if the maximum is greater than 1, we use OP_UPTO with
5709 one less than the maximum. */
5711 else if (repeat_min == 1)
5713 if (repeat_max == -1)
5714 *code++ = OP_PLUS + repeat_type;
5717 code = oldcode; /* leave previous item in place */
5718 if (repeat_max == 1) goto END_REPEAT;
5719 *code++ = OP_UPTO + repeat_type;
5720 PUT2INC(code, 0, repeat_max - 1);
5724 /* The case {n,n} is just an EXACT, while the general case {n,m} is
5725 handled as an EXACT followed by an UPTO. */
5729 *code++ = OP_EXACT + op_type; /* NB EXACT doesn't have repeat_type */
5730 PUT2INC(code, 0, repeat_min);
5732 /* If the maximum is unlimited, insert an OP_STAR. Before doing so,
5733 we have to insert the character for the previous code. For a repeated
5734 Unicode property match, there are two extra bytes that define the
5735 required property. In UTF-8 mode, long characters have their length in
5736 c, with the UTF_LENGTH bit as a flag. */
5740 #if defined SUPPORT_UTF && !defined COMPILE_PCRE32
5741 if (utf && (c & UTF_LENGTH) != 0)
5743 memcpy(code, utf_chars, IN_UCHARS(c & 7));
5752 *code++ = prop_type;
5753 *code++ = prop_value;
5756 *code++ = OP_STAR + repeat_type;
5759 /* Else insert an UPTO if the max is greater than the min, again
5760 preceded by the character, for the previously inserted code. If the
5761 UPTO is just for 1 instance, we can use QUERY instead. */
5763 else if (repeat_max != repeat_min)
5765 #if defined SUPPORT_UTF && !defined COMPILE_PCRE32
5766 if (utf && (c & UTF_LENGTH) != 0)
5768 memcpy(code, utf_chars, IN_UCHARS(c & 7));
5776 *code++ = prop_type;
5777 *code++ = prop_value;
5779 repeat_max -= repeat_min;
5781 if (repeat_max == 1)
5783 *code++ = OP_QUERY + repeat_type;
5787 *code++ = OP_UPTO + repeat_type;
5788 PUT2INC(code, 0, repeat_max);
5793 /* The character or character type itself comes last in all cases. */
5795 #if defined SUPPORT_UTF && !defined COMPILE_PCRE32
5796 if (utf && (c & UTF_LENGTH) != 0)
5798 memcpy(code, utf_chars, IN_UCHARS(c & 7));
5805 /* For a repeated Unicode property match, there are two extra bytes that
5806 define the required property. */
5811 *code++ = prop_type;
5812 *code++ = prop_value;
5817 /* If previous was a character class or a back reference, we put the repeat
5818 stuff after it, but just skip the item if the repeat was {0,0}. */
5820 else if (*previous == OP_CLASS || *previous == OP_NCLASS ||
5821 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
5822 *previous == OP_XCLASS ||
5824 *previous == OP_REF || *previous == OP_REFI ||
5825 *previous == OP_DNREF || *previous == OP_DNREFI)
5827 if (repeat_max == 0)
5833 if (repeat_min == 0 && repeat_max == -1)
5834 *code++ = OP_CRSTAR + repeat_type;
5835 else if (repeat_min == 1 && repeat_max == -1)
5836 *code++ = OP_CRPLUS + repeat_type;
5837 else if (repeat_min == 0 && repeat_max == 1)
5838 *code++ = OP_CRQUERY + repeat_type;
5841 *code++ = OP_CRRANGE + repeat_type;
5842 PUT2INC(code, 0, repeat_min);
5843 if (repeat_max == -1) repeat_max = 0; /* 2-byte encoding for max */
5844 PUT2INC(code, 0, repeat_max);
5848 /* If previous was a bracket group, we may have to replicate it in certain
5849 cases. Note that at this point we can encounter only the "basic" bracket
5850 opcodes such as BRA and CBRA, as this is the place where they get converted
5851 into the more special varieties such as BRAPOS and SBRA. A test for >=
5852 OP_ASSERT and <= OP_COND includes ASSERT, ASSERT_NOT, ASSERTBACK,
5853 ASSERTBACK_NOT, ONCE, ONCE_NC, BRA, BRAPOS, CBRA, CBRAPOS, and COND.
5854 Originally, PCRE did not allow repetition of assertions, but now it does,
5855 for Perl compatibility. */
5857 else if (*previous >= OP_ASSERT && *previous <= OP_COND)
5860 int len = (int)(code - previous);
5861 pcre_uchar *bralink = NULL;
5862 pcre_uchar *brazeroptr = NULL;
5864 /* Repeating a DEFINE group is pointless, but Perl allows the syntax, so
5865 we just ignore the repeat. */
5867 if (*previous == OP_COND && previous[LINK_SIZE+1] == OP_DEF)
5870 /* There is no sense in actually repeating assertions. The only potential
5871 use of repetition is in cases when the assertion is optional. Therefore,
5872 if the minimum is greater than zero, just ignore the repeat. If the
5873 maximum is not zero or one, set it to 1. */
5875 if (*previous < OP_ONCE) /* Assertion */
5877 if (repeat_min > 0) goto END_REPEAT;
5878 if (repeat_max < 0 || repeat_max > 1) repeat_max = 1;
5881 /* The case of a zero minimum is special because of the need to stick
5882 OP_BRAZERO in front of it, and because the group appears once in the
5883 data, whereas in other cases it appears the minimum number of times. For
5884 this reason, it is simplest to treat this case separately, as otherwise
5885 the code gets far too messy. There are several special subcases when the
5888 if (repeat_min == 0)
5890 /* If the maximum is also zero, we used to just omit the group from the
5891 output altogether, like this:
5893 ** if (repeat_max == 0)
5899 However, that fails when a group or a subgroup within it is referenced
5900 as a subroutine from elsewhere in the pattern, so now we stick in
5901 OP_SKIPZERO in front of it so that it is skipped on execution. As we
5902 don't have a list of which groups are referenced, we cannot do this
5905 If the maximum is 1 or unlimited, we just have to stick in the BRAZERO
5906 and do no more at this point. However, we do need to adjust any
5907 OP_RECURSE calls inside the group that refer to the group itself or any
5908 internal or forward referenced group, because the offset is from the
5909 start of the whole regex. Temporarily terminate the pattern while doing
5912 if (repeat_max <= 1) /* Covers 0, 1, and unlimited */
5915 adjust_recurse(previous, 1, utf, cd, save_hwm);
5916 memmove(previous + 1, previous, IN_UCHARS(len));
5918 if (repeat_max == 0)
5920 *previous++ = OP_SKIPZERO;
5923 brazeroptr = previous; /* Save for possessive optimizing */
5924 *previous++ = OP_BRAZERO + repeat_type;
5927 /* If the maximum is greater than 1 and limited, we have to replicate
5928 in a nested fashion, sticking OP_BRAZERO before each set of brackets.
5929 The first one has to be handled carefully because it's the original
5930 copy, which has to be moved up. The remainder can be handled by code
5931 that is common with the non-zero minimum case below. We have to
5932 adjust the value or repeat_max, since one less copy is required. Once
5933 again, we may have to adjust any OP_RECURSE calls inside the group. */
5939 adjust_recurse(previous, 2 + LINK_SIZE, utf, cd, save_hwm);
5940 memmove(previous + 2 + LINK_SIZE, previous, IN_UCHARS(len));
5941 code += 2 + LINK_SIZE;
5942 *previous++ = OP_BRAZERO + repeat_type;
5943 *previous++ = OP_BRA;
5945 /* We chain together the bracket offset fields that have to be
5946 filled in later when the ends of the brackets are reached. */
5948 offset = (bralink == NULL)? 0 : (int)(previous - bralink);
5950 PUTINC(previous, 0, offset);
5956 /* If the minimum is greater than zero, replicate the group as many
5957 times as necessary, and adjust the maximum to the number of subsequent
5958 copies that we need. If we set a first char from the group, and didn't
5959 set a required char, copy the latter from the former. If there are any
5960 forward reference subroutine calls in the group, there will be entries on
5961 the workspace list; replicate these with an appropriate increment. */
5967 /* In the pre-compile phase, we don't actually do the replication. We
5968 just adjust the length as if we had. Do some paranoid checks for
5969 potential integer overflow. The INT64_OR_DOUBLE type is a 64-bit
5970 integer type when available, otherwise double. */
5972 if (lengthptr != NULL)
5974 int delta = (repeat_min - 1)*length_prevgroup;
5975 if ((INT64_OR_DOUBLE)(repeat_min - 1)*
5976 (INT64_OR_DOUBLE)length_prevgroup >
5977 (INT64_OR_DOUBLE)INT_MAX ||
5978 OFLOW_MAX - *lengthptr < delta)
5980 *errorcodeptr = ERR20;
5983 *lengthptr += delta;
5986 /* This is compiling for real. If there is a set first byte for
5987 the group, and we have not yet set a "required byte", set it. Make
5988 sure there is enough workspace for copying forward references before
5993 if (groupsetfirstchar && reqcharflags < 0)
5995 reqchar = firstchar;
5996 reqcharflags = firstcharflags;
5999 for (i = 1; i < repeat_min; i++)
6002 pcre_uchar *this_hwm = cd->hwm;
6003 memcpy(code, previous, IN_UCHARS(len));
6005 while (cd->hwm > cd->start_workspace + cd->workspace_size -
6006 WORK_SIZE_SAFETY_MARGIN - (this_hwm - save_hwm))
6008 int save_offset = save_hwm - cd->start_workspace;
6009 int this_offset = this_hwm - cd->start_workspace;
6010 *errorcodeptr = expand_workspace(cd);
6011 if (*errorcodeptr != 0) goto FAILED;
6012 save_hwm = (pcre_uchar *)cd->start_workspace + save_offset;
6013 this_hwm = (pcre_uchar *)cd->start_workspace + this_offset;
6016 for (hc = save_hwm; hc < this_hwm; hc += LINK_SIZE)
6018 PUT(cd->hwm, 0, GET(hc, 0) + len);
6019 cd->hwm += LINK_SIZE;
6021 save_hwm = this_hwm;
6027 if (repeat_max > 0) repeat_max -= repeat_min;
6030 /* This code is common to both the zero and non-zero minimum cases. If
6031 the maximum is limited, it replicates the group in a nested fashion,
6032 remembering the bracket starts on a stack. In the case of a zero minimum,
6033 the first one was set up above. In all cases the repeat_max now specifies
6034 the number of additional copies needed. Again, we must remember to
6035 replicate entries on the forward reference list. */
6037 if (repeat_max >= 0)
6039 /* In the pre-compile phase, we don't actually do the replication. We
6040 just adjust the length as if we had. For each repetition we must add 1
6041 to the length for BRAZERO and for all but the last repetition we must
6042 add 2 + 2*LINKSIZE to allow for the nesting that occurs. Do some
6043 paranoid checks to avoid integer overflow. The INT64_OR_DOUBLE type is
6044 a 64-bit integer type when available, otherwise double. */
6046 if (lengthptr != NULL && repeat_max > 0)
6048 int delta = repeat_max * (length_prevgroup + 1 + 2 + 2*LINK_SIZE) -
6049 2 - 2*LINK_SIZE; /* Last one doesn't nest */
6050 if ((INT64_OR_DOUBLE)repeat_max *
6051 (INT64_OR_DOUBLE)(length_prevgroup + 1 + 2 + 2*LINK_SIZE)
6052 > (INT64_OR_DOUBLE)INT_MAX ||
6053 OFLOW_MAX - *lengthptr < delta)
6055 *errorcodeptr = ERR20;
6058 *lengthptr += delta;
6061 /* This is compiling for real */
6063 else for (i = repeat_max - 1; i >= 0; i--)
6066 pcre_uchar *this_hwm = cd->hwm;
6068 *code++ = OP_BRAZERO + repeat_type;
6070 /* All but the final copy start a new nesting, maintaining the
6071 chain of brackets outstanding. */
6077 offset = (bralink == NULL)? 0 : (int)(code - bralink);
6079 PUTINC(code, 0, offset);
6082 memcpy(code, previous, IN_UCHARS(len));
6084 /* Ensure there is enough workspace for forward references before
6087 while (cd->hwm > cd->start_workspace + cd->workspace_size -
6088 WORK_SIZE_SAFETY_MARGIN - (this_hwm - save_hwm))
6090 int save_offset = save_hwm - cd->start_workspace;
6091 int this_offset = this_hwm - cd->start_workspace;
6092 *errorcodeptr = expand_workspace(cd);
6093 if (*errorcodeptr != 0) goto FAILED;
6094 save_hwm = (pcre_uchar *)cd->start_workspace + save_offset;
6095 this_hwm = (pcre_uchar *)cd->start_workspace + this_offset;
6098 for (hc = save_hwm; hc < this_hwm; hc += LINK_SIZE)
6100 PUT(cd->hwm, 0, GET(hc, 0) + len + ((i != 0)? 2+LINK_SIZE : 1));
6101 cd->hwm += LINK_SIZE;
6103 save_hwm = this_hwm;
6107 /* Now chain through the pending brackets, and fill in their length
6108 fields (which are holding the chain links pro tem). */
6110 while (bralink != NULL)
6113 int offset = (int)(code - bralink + 1);
6114 pcre_uchar *bra = code - offset;
6115 oldlinkoffset = GET(bra, 1);
6116 bralink = (oldlinkoffset == 0)? NULL : bralink - oldlinkoffset;
6118 PUTINC(code, 0, offset);
6119 PUT(bra, 1, offset);
6123 /* If the maximum is unlimited, set a repeater in the final copy. For
6124 ONCE brackets, that's all we need to do. However, possessively repeated
6125 ONCE brackets can be converted into non-capturing brackets, as the
6126 behaviour of (?:xx)++ is the same as (?>xx)++ and this saves having to
6127 deal with possessive ONCEs specially.
6129 Otherwise, when we are doing the actual compile phase, check to see
6130 whether this group is one that could match an empty string. If so,
6131 convert the initial operator to the S form (e.g. OP_BRA -> OP_SBRA) so
6132 that runtime checking can be done. [This check is also applied to ONCE
6133 groups at runtime, but in a different way.]
6135 Then, if the quantifier was possessive and the bracket is not a
6136 conditional, we convert the BRA code to the POS form, and the KET code to
6137 KETRPOS. (It turns out to be convenient at runtime to detect this kind of
6138 subpattern at both the start and at the end.) The use of special opcodes
6139 makes it possible to reduce greatly the stack usage in pcre_exec(). If
6140 the group is preceded by OP_BRAZERO, convert this to OP_BRAPOSZERO.
6142 Then, if the minimum number of matches is 1 or 0, cancel the possessive
6143 flag so that the default action below, of wrapping everything inside
6144 atomic brackets, does not happen. When the minimum is greater than 1,
6145 there will be earlier copies of the group, and so we still have to wrap
6150 pcre_uchar *ketcode = code - 1 - LINK_SIZE;
6151 pcre_uchar *bracode = ketcode - GET(ketcode, 1);
6153 /* Convert possessive ONCE brackets to non-capturing */
6155 if ((*bracode == OP_ONCE || *bracode == OP_ONCE_NC) &&
6156 possessive_quantifier) *bracode = OP_BRA;
6158 /* For non-possessive ONCE brackets, all we need to do is to
6161 if (*bracode == OP_ONCE || *bracode == OP_ONCE_NC)
6162 *ketcode = OP_KETRMAX + repeat_type;
6164 /* Handle non-ONCE brackets and possessive ONCEs (which have been
6165 converted to non-capturing above). */
6169 /* In the compile phase, check for empty string matching. */
6171 if (lengthptr == NULL)
6173 pcre_uchar *scode = bracode;
6176 if (could_be_empty_branch(scode, ketcode, utf, cd, NULL))
6178 *bracode += OP_SBRA - OP_BRA;
6181 scode += GET(scode, 1);
6183 while (*scode == OP_ALT);
6186 /* Handle possessive quantifiers. */
6188 if (possessive_quantifier)
6190 /* For COND brackets, we wrap the whole thing in a possessively
6191 repeated non-capturing bracket, because we have not invented POS
6192 versions of the COND opcodes. Because we are moving code along, we
6193 must ensure that any pending recursive references are updated. */
6195 if (*bracode == OP_COND || *bracode == OP_SCOND)
6197 int nlen = (int)(code - bracode);
6199 adjust_recurse(bracode, 1 + LINK_SIZE, utf, cd, save_hwm);
6200 memmove(bracode + 1 + LINK_SIZE, bracode, IN_UCHARS(nlen));
6201 code += 1 + LINK_SIZE;
6202 nlen += 1 + LINK_SIZE;
6203 *bracode = OP_BRAPOS;
6204 *code++ = OP_KETRPOS;
6205 PUTINC(code, 0, nlen);
6206 PUT(bracode, 1, nlen);
6209 /* For non-COND brackets, we modify the BRA code and use KETRPOS. */
6213 *bracode += 1; /* Switch to xxxPOS opcodes */
6214 *ketcode = OP_KETRPOS;
6217 /* If the minimum is zero, mark it as possessive, then unset the
6218 possessive flag when the minimum is 0 or 1. */
6220 if (brazeroptr != NULL) *brazeroptr = OP_BRAPOSZERO;
6221 if (repeat_min < 2) possessive_quantifier = FALSE;
6224 /* Non-possessive quantifier */
6226 else *ketcode = OP_KETRMAX + repeat_type;
6231 /* If previous is OP_FAIL, it was generated by an empty class [] in
6232 JavaScript mode. The other ways in which OP_FAIL can be generated, that is
6233 by (*FAIL) or (?!) set previous to NULL, which gives a "nothing to repeat"
6234 error above. We can just ignore the repeat in JS case. */
6236 else if (*previous == OP_FAIL) goto END_REPEAT;
6238 /* Else there's some kind of shambles */
6242 *errorcodeptr = ERR11;
6246 /* If the character following a repeat is '+', possessive_quantifier is
6247 TRUE. For some opcodes, there are special alternative opcodes for this
6248 case. For anything else, we wrap the entire repeated item inside OP_ONCE
6249 brackets. Logically, the '+' notation is just syntactic sugar, taken from
6250 Sun's Java package, but the special opcodes can optimize it.
6252 Some (but not all) possessively repeated subpatterns have already been
6253 completely handled in the code just above. For them, possessive_quantifier
6254 is always FALSE at this stage. Note that the repeated item starts at
6255 tempcode, not at previous, which might be the first part of a string whose
6256 (former) last char we repeated. */
6258 if (possessive_quantifier)
6262 /* Possessifying an EXACT quantifier has no effect, so we can ignore it.
6263 However, QUERY, STAR, or UPTO may follow (for quantifiers such as {5,6},
6264 {5,}, or {5,10}). We skip over an EXACT item; if the length of what
6265 remains is greater than zero, there's a further opcode that can be
6266 handled. If not, do nothing, leaving the EXACT alone. */
6271 tempcode += PRIV(OP_lengths)[*tempcode] +
6272 ((tempcode[1 + IMM2_SIZE] == OP_PROP
6273 || tempcode[1 + IMM2_SIZE] == OP_NOTPROP)? 2 : 0);
6276 /* CHAR opcodes are used for exacts whose count is 1. */
6286 tempcode += PRIV(OP_lengths)[*tempcode];
6288 if (utf && HAS_EXTRALEN(tempcode[-1]))
6289 tempcode += GET_EXTRALEN(tempcode[-1]);
6293 /* For the class opcodes, the repeat operator appears at the end;
6294 adjust tempcode to point to it. */
6298 tempcode += 1 + 32/sizeof(pcre_uchar);
6301 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
6303 tempcode += GET(tempcode, 1);
6308 /* If tempcode is equal to code (which points to the end of the repeated
6309 item), it means we have skipped an EXACT item but there is no following
6310 QUERY, STAR, or UPTO; the value of len will be 0, and we do nothing. In
6311 all other cases, tempcode will be pointing to the repeat opcode, and will
6312 be less than code, so the value of len will be greater than 0. */
6314 len = (int)(code - tempcode);
6317 unsigned int repcode = *tempcode;
6319 /* There is a table for possessifying opcodes, all of which are less
6320 than OP_CALLOUT. A zero entry means there is no possessified version.
6323 if (repcode < OP_CALLOUT && opcode_possessify[repcode] > 0)
6324 *tempcode = opcode_possessify[repcode];
6326 /* For opcode without a special possessified version, wrap the item in
6327 ONCE brackets. Because we are moving code along, we must ensure that any
6328 pending recursive references are updated. */
6333 adjust_recurse(tempcode, 1 + LINK_SIZE, utf, cd, save_hwm);
6334 memmove(tempcode + 1 + LINK_SIZE, tempcode, IN_UCHARS(len));
6335 code += 1 + LINK_SIZE;
6336 len += 1 + LINK_SIZE;
6337 tempcode[0] = OP_ONCE;
6339 PUTINC(code, 0, len);
6340 PUT(tempcode, 1, len);
6345 if (len > 0) switch (*tempcode)
6347 case OP_STAR: *tempcode = OP_POSSTAR; break;
6348 case OP_PLUS: *tempcode = OP_POSPLUS; break;
6349 case OP_QUERY: *tempcode = OP_POSQUERY; break;
6350 case OP_UPTO: *tempcode = OP_POSUPTO; break;
6352 case OP_STARI: *tempcode = OP_POSSTARI; break;
6353 case OP_PLUSI: *tempcode = OP_POSPLUSI; break;
6354 case OP_QUERYI: *tempcode = OP_POSQUERYI; break;
6355 case OP_UPTOI: *tempcode = OP_POSUPTOI; break;
6357 case OP_NOTSTAR: *tempcode = OP_NOTPOSSTAR; break;
6358 case OP_NOTPLUS: *tempcode = OP_NOTPOSPLUS; break;
6359 case OP_NOTQUERY: *tempcode = OP_NOTPOSQUERY; break;
6360 case OP_NOTUPTO: *tempcode = OP_NOTPOSUPTO; break;
6362 case OP_NOTSTARI: *tempcode = OP_NOTPOSSTARI; break;
6363 case OP_NOTPLUSI: *tempcode = OP_NOTPOSPLUSI; break;
6364 case OP_NOTQUERYI: *tempcode = OP_NOTPOSQUERYI; break;
6365 case OP_NOTUPTOI: *tempcode = OP_NOTPOSUPTOI; break;
6367 case OP_TYPESTAR: *tempcode = OP_TYPEPOSSTAR; break;
6368 case OP_TYPEPLUS: *tempcode = OP_TYPEPOSPLUS; break;
6369 case OP_TYPEQUERY: *tempcode = OP_TYPEPOSQUERY; break;
6370 case OP_TYPEUPTO: *tempcode = OP_TYPEPOSUPTO; break;
6372 case OP_CRSTAR: *tempcode = OP_CRPOSSTAR; break;
6373 case OP_CRPLUS: *tempcode = OP_CRPOSPLUS; break;
6374 case OP_CRQUERY: *tempcode = OP_CRPOSQUERY; break;
6375 case OP_CRRANGE: *tempcode = OP_CRPOSRANGE; break;
6377 /* Because we are moving code along, we must ensure that any
6378 pending recursive references are updated. */
6382 adjust_recurse(tempcode, 1 + LINK_SIZE, utf, cd, save_hwm);
6383 memmove(tempcode + 1 + LINK_SIZE, tempcode, IN_UCHARS(len));
6384 code += 1 + LINK_SIZE;
6385 len += 1 + LINK_SIZE;
6386 tempcode[0] = OP_ONCE;
6388 PUTINC(code, 0, len);
6389 PUT(tempcode, 1, len);
6395 /* In all case we no longer have a previous item. We also set the
6396 "follows varying string" flag for subsequently encountered reqchars if
6397 it isn't already set and we have just passed a varying length item. */
6401 cd->req_varyopt |= reqvary;
6405 /* ===================================================================*/
6406 /* Start of nested parenthesized sub-expression, or comment or lookahead or
6407 lookbehind or option setting or condition or all the other extended
6408 parenthesis forms. */
6410 case CHAR_LEFT_PARENTHESIS:
6411 newoptions = options;
6415 reset_bracount = FALSE;
6417 /* First deal with various "verbs" that can be introduced by '*'. */
6420 if (ptr[0] == CHAR_ASTERISK && (ptr[1] == ':'
6421 || (MAX_255(ptr[1]) && ((cd->ctypes[ptr[1]] & ctype_letter) != 0))))
6425 const char *vn = verbnames;
6426 const pcre_uchar *name = ptr + 1;
6427 const pcre_uchar *arg = NULL;
6430 while (MAX_255(*ptr) && (cd->ctypes[*ptr] & ctype_letter) != 0) ptr++;
6431 namelen = (int)(ptr - name);
6433 /* It appears that Perl allows any characters whatsoever, other than
6434 a closing parenthesis, to appear in arguments, so we no longer insist on
6435 letters, digits, and underscores. */
6437 if (*ptr == CHAR_COLON)
6440 while (*ptr != CHAR_NULL && *ptr != CHAR_RIGHT_PARENTHESIS) ptr++;
6441 arglen = (int)(ptr - arg);
6442 if ((unsigned int)arglen > MAX_MARK)
6444 *errorcodeptr = ERR75;
6449 if (*ptr != CHAR_RIGHT_PARENTHESIS)
6451 *errorcodeptr = ERR60;
6455 /* Scan the table of verb names */
6457 for (i = 0; i < verbcount; i++)
6459 if (namelen == verbs[i].len &&
6460 STRNCMP_UC_C8(name, vn, namelen) == 0)
6464 /* Check for open captures before ACCEPT and convert it to
6465 ASSERT_ACCEPT if in an assertion. */
6467 if (verbs[i].op == OP_ACCEPT)
6472 *errorcodeptr = ERR59;
6475 cd->had_accept = TRUE;
6476 for (oc = cd->open_caps; oc != NULL; oc = oc->next)
6479 PUT2INC(code, 0, oc->number);
6482 (cd->assert_depth > 0)? OP_ASSERT_ACCEPT : OP_ACCEPT;
6484 /* Do not set firstchar after *ACCEPT */
6485 if (firstcharflags == REQ_UNSET) firstcharflags = REQ_NONE;
6488 /* Handle other cases with/without an argument */
6490 else if (arglen == 0)
6492 if (verbs[i].op < 0) /* Argument is mandatory */
6494 *errorcodeptr = ERR66;
6497 setverb = *code++ = verbs[i].op;
6502 if (verbs[i].op_arg < 0) /* Argument is forbidden */
6504 *errorcodeptr = ERR59;
6507 setverb = *code++ = verbs[i].op_arg;
6509 memcpy(code, arg, IN_UCHARS(arglen));
6518 cd->external_flags |= PCRE_HASTHEN;
6525 cd->had_pruneorskip = TRUE;
6529 break; /* Found verb, exit loop */
6532 vn += verbs[i].len + 1;
6535 if (i < verbcount) continue; /* Successfully handled a verb */
6536 *errorcodeptr = ERR60; /* Verb not recognized */
6540 /* Deal with the extended parentheses; all are introduced by '?', and the
6541 appearance of any of them means that this is not a capturing group. */
6543 else if (*ptr == CHAR_QUESTION_MARK)
6545 int i, set, unset, namelen;
6547 const pcre_uchar *name;
6552 case CHAR_NUMBER_SIGN: /* Comment; skip to ket */
6554 while (*ptr != CHAR_NULL && *ptr != CHAR_RIGHT_PARENTHESIS) ptr++;
6555 if (*ptr == CHAR_NULL)
6557 *errorcodeptr = ERR18;
6563 /* ------------------------------------------------------------ */
6564 case CHAR_VERTICAL_LINE: /* Reset capture count for each branch */
6565 reset_bracount = TRUE;
6568 /* ------------------------------------------------------------ */
6569 case CHAR_COLON: /* Non-capturing bracket */
6575 /* ------------------------------------------------------------ */
6576 case CHAR_LEFT_PARENTHESIS:
6577 bravalue = OP_COND; /* Conditional group */
6580 /* A condition can be an assertion, a number (referring to a numbered
6581 group's having been set), a name (referring to a named group), or 'R',
6582 referring to recursion. R<digits> and R&name are also permitted for
6585 There are ways of testing a named group: (?(name)) is used by Python;
6586 Perl 5.10 onwards uses (?(<name>) or (?('name')).
6588 There is one unfortunate ambiguity, caused by history. 'R' can be the
6589 recursive thing or the name 'R' (and similarly for 'R' followed by
6590 digits). We look for a name first; if not found, we try the other case.
6592 For compatibility with auto-callouts, we allow a callout to be
6593 specified before a condition that is an assertion. First, check for the
6594 syntax of a callout; if found, adjust the temporary pointer that is
6595 used to check for an assertion condition. That's all that is needed! */
6597 if (ptr[1] == CHAR_QUESTION_MARK && ptr[2] == CHAR_C)
6599 for (i = 3;; i++) if (!IS_DIGIT(ptr[i])) break;
6600 if (ptr[i] == CHAR_RIGHT_PARENTHESIS)
6604 /* For conditions that are assertions, check the syntax, and then exit
6605 the switch. This will take control down to where bracketed groups,
6606 including assertions, are processed. */
6608 if (tempptr[1] == CHAR_QUESTION_MARK &&
6609 (tempptr[2] == CHAR_EQUALS_SIGN ||
6610 tempptr[2] == CHAR_EXCLAMATION_MARK ||
6611 tempptr[2] == CHAR_LESS_THAN_SIGN))
6614 /* Other conditions use OP_CREF/OP_DNCREF/OP_RREF/OP_DNRREF, and all
6615 need to skip at least 1+IMM2_SIZE bytes at the start of the group. */
6617 code[1+LINK_SIZE] = OP_CREF;
6618 skipbytes = 1+IMM2_SIZE;
6619 refsign = -1; /* => not a number */
6620 namelen = -1; /* => not a name; must set to avoid warning */
6621 name = NULL; /* Always set to avoid warning */
6622 recno = 0; /* Always set to avoid warning */
6624 /* Check for a test for recursion in a named group. */
6627 if (*ptr == CHAR_R && ptr[1] == CHAR_AMPERSAND)
6631 code[1+LINK_SIZE] = OP_RREF; /* Change the type of test */
6634 /* Check for a test for a named group's having been set, using the Perl
6635 syntax (?(<name>) or (?('name'), and also allow for the original PCRE
6636 syntax of (?(name) or for (?(+n), (?(-n), and just (?(n). */
6638 else if (*ptr == CHAR_LESS_THAN_SIGN)
6640 terminator = CHAR_GREATER_THAN_SIGN;
6643 else if (*ptr == CHAR_APOSTROPHE)
6645 terminator = CHAR_APOSTROPHE;
6650 terminator = CHAR_NULL;
6651 if (*ptr == CHAR_MINUS || *ptr == CHAR_PLUS) refsign = *ptr++;
6652 else if (IS_DIGIT(*ptr)) refsign = 0;
6655 /* Handle a number */
6659 while (IS_DIGIT(*ptr))
6661 recno = recno * 10 + (int)(*ptr - CHAR_0);
6666 /* Otherwise we expect to read a name; anything else is an error. When
6667 a name is one of a number of duplicates, a different opcode is used and
6668 it needs more memory. Unfortunately we cannot tell whether a name is a
6669 duplicate in the first pass, so we have to allow for more memory. */
6675 *errorcodeptr = ERR84;
6678 if (!MAX_255(*ptr) || (cd->ctypes[*ptr] & ctype_word) == 0)
6680 *errorcodeptr = ERR28; /* Assertion expected */
6684 while (MAX_255(*ptr) && (cd->ctypes[*ptr] & ctype_word) != 0)
6688 namelen = (int)(ptr - name);
6689 if (lengthptr != NULL) *lengthptr += IMM2_SIZE;
6692 /* Check the terminator */
6694 if ((terminator > 0 && *ptr++ != (pcre_uchar)terminator) ||
6695 *ptr++ != CHAR_RIGHT_PARENTHESIS)
6697 ptr--; /* Error offset */
6698 *errorcodeptr = ERR26; /* Malformed number or name */
6702 /* Do no further checking in the pre-compile phase. */
6704 if (lengthptr != NULL) break;
6706 /* In the real compile we do the work of looking for the actual
6707 reference. If refsign is not negative, it means we have a number in
6714 *errorcodeptr = ERR35;
6717 if (refsign != 0) recno = (refsign == CHAR_MINUS)?
6718 cd->bracount - recno + 1 : recno + cd->bracount;
6719 if (recno <= 0 || recno > cd->final_bracount)
6721 *errorcodeptr = ERR15;
6724 PUT2(code, 2+LINK_SIZE, recno);
6728 /* Otherwise look for the name. */
6730 slot = cd->name_table;
6731 for (i = 0; i < cd->names_found; i++)
6733 if (STRNCMP_UC_UC(name, slot+IMM2_SIZE, namelen) == 0) break;
6734 slot += cd->name_entry_size;
6737 /* Found the named subpattern. If the name is duplicated, add one to
6738 the opcode to change CREF/RREF into DNCREF/DNRREF and insert
6739 appropriate data values. Otherwise, just insert the unique subpattern
6742 if (i < cd->names_found)
6746 recno = GET2(slot, 0); /* Number from first found */
6747 for (; i < cd->names_found; i++)
6749 slot += cd->name_entry_size;
6750 if (STRNCMP_UC_UC(name, slot+IMM2_SIZE, namelen) != 0) break;
6755 PUT2(code, 2+LINK_SIZE, offset);
6756 PUT2(code, 2+LINK_SIZE+IMM2_SIZE, count);
6757 skipbytes += IMM2_SIZE;
6758 code[1+LINK_SIZE]++;
6760 else /* Not a duplicated name */
6762 PUT2(code, 2+LINK_SIZE, recno);
6766 /* If terminator == CHAR_NULL it means that the name followed directly
6767 after the opening parenthesis [e.g. (?(abc)...] and in this case there
6768 are some further alternatives to try. For the cases where terminator !=
6769 CHAR_NULL [things like (?(<name>... or (?('name')... or (?(R&name)... ]
6770 we have now checked all the possibilities, so give an error. */
6772 else if (terminator != CHAR_NULL)
6774 *errorcodeptr = ERR15;
6778 /* Check for (?(R) for recursion. Allow digits after R to specify a
6779 specific group number. */
6781 else if (*name == CHAR_R)
6784 for (i = 1; i < namelen; i++)
6786 if (!IS_DIGIT(name[i]))
6788 *errorcodeptr = ERR15;
6791 recno = recno * 10 + name[i] - CHAR_0;
6793 if (recno == 0) recno = RREF_ANY;
6794 code[1+LINK_SIZE] = OP_RREF; /* Change test type */
6795 PUT2(code, 2+LINK_SIZE, recno);
6798 /* Similarly, check for the (?(DEFINE) "condition", which is always
6801 else if (namelen == 6 && STRNCMP_UC_C8(name, STRING_DEFINE, 6) == 0)
6803 code[1+LINK_SIZE] = OP_DEF;
6807 /* Reference to an unidentified subpattern. */
6811 *errorcodeptr = ERR15;
6817 /* ------------------------------------------------------------ */
6818 case CHAR_EQUALS_SIGN: /* Positive lookahead */
6819 bravalue = OP_ASSERT;
6820 cd->assert_depth += 1;
6824 /* Optimize (?!) to (*FAIL) unless it is quantified - which is a weird
6825 thing to do, but Perl allows all assertions to be quantified, and when
6826 they contain capturing parentheses there may be a potential use for
6827 this feature. Not that that applies to a quantified (?!) but we allow
6828 it for uniformity. */
6830 /* ------------------------------------------------------------ */
6831 case CHAR_EXCLAMATION_MARK: /* Negative lookahead */
6833 if (*ptr == CHAR_RIGHT_PARENTHESIS && ptr[1] != CHAR_ASTERISK &&
6834 ptr[1] != CHAR_PLUS && ptr[1] != CHAR_QUESTION_MARK &&
6835 (ptr[1] != CHAR_LEFT_CURLY_BRACKET || !is_counted_repeat(ptr+2)))
6841 bravalue = OP_ASSERT_NOT;
6842 cd->assert_depth += 1;
6846 /* ------------------------------------------------------------ */
6847 case CHAR_LESS_THAN_SIGN: /* Lookbehind or named define */
6850 case CHAR_EQUALS_SIGN: /* Positive lookbehind */
6851 bravalue = OP_ASSERTBACK;
6852 cd->assert_depth += 1;
6856 case CHAR_EXCLAMATION_MARK: /* Negative lookbehind */
6857 bravalue = OP_ASSERTBACK_NOT;
6858 cd->assert_depth += 1;
6862 default: /* Could be name define, else bad */
6863 if (MAX_255(ptr[1]) && (cd->ctypes[ptr[1]] & ctype_word) != 0)
6865 ptr++; /* Correct offset for error */
6866 *errorcodeptr = ERR24;
6872 /* ------------------------------------------------------------ */
6873 case CHAR_GREATER_THAN_SIGN: /* One-time brackets */
6879 /* ------------------------------------------------------------ */
6880 case CHAR_C: /* Callout - may be followed by digits; */
6881 previous_callout = code; /* Save for later completion */
6882 after_manual_callout = 1; /* Skip one item before completing */
6883 *code++ = OP_CALLOUT;
6887 while(IS_DIGIT(*ptr))
6888 n = n * 10 + *ptr++ - CHAR_0;
6889 if (*ptr != CHAR_RIGHT_PARENTHESIS)
6891 *errorcodeptr = ERR39;
6896 *errorcodeptr = ERR38;
6900 PUT(code, 0, (int)(ptr - cd->start_pattern + 1)); /* Pattern offset */
6901 PUT(code, LINK_SIZE, 0); /* Default length */
6902 code += 2 * LINK_SIZE;
6908 /* ------------------------------------------------------------ */
6909 case CHAR_P: /* Python-style named subpattern handling */
6910 if (*(++ptr) == CHAR_EQUALS_SIGN ||
6911 *ptr == CHAR_GREATER_THAN_SIGN) /* Reference or recursion */
6913 is_recurse = *ptr == CHAR_GREATER_THAN_SIGN;
6914 terminator = CHAR_RIGHT_PARENTHESIS;
6915 goto NAMED_REF_OR_RECURSE;
6917 else if (*ptr != CHAR_LESS_THAN_SIGN) /* Test for Python-style defn */
6919 *errorcodeptr = ERR41;
6922 /* Fall through to handle (?P< as (?< is handled */
6925 /* ------------------------------------------------------------ */
6926 DEFINE_NAME: /* Come here from (?< handling */
6927 case CHAR_APOSTROPHE:
6928 terminator = (*ptr == CHAR_LESS_THAN_SIGN)?
6929 CHAR_GREATER_THAN_SIGN : CHAR_APOSTROPHE;
6933 *errorcodeptr = ERR84; /* Group name must start with non-digit */
6936 while (MAX_255(*ptr) && (cd->ctypes[*ptr] & ctype_word) != 0) ptr++;
6937 namelen = (int)(ptr - name);
6939 /* In the pre-compile phase, do a syntax check, remember the longest
6940 name, and then remember the group in a vector, expanding it if
6941 necessary. Duplicates for the same number are skipped; other duplicates
6942 are checked for validity. In the actual compile, there is nothing to
6945 if (lengthptr != NULL)
6948 pcre_uint32 number = cd->bracount + 1;
6950 if (*ptr != (pcre_uchar)terminator)
6952 *errorcodeptr = ERR42;
6956 if (cd->names_found >= MAX_NAME_COUNT)
6958 *errorcodeptr = ERR49;
6962 if (namelen + IMM2_SIZE + 1 > cd->name_entry_size)
6964 cd->name_entry_size = namelen + IMM2_SIZE + 1;
6965 if (namelen > MAX_NAME_SIZE)
6967 *errorcodeptr = ERR48;
6972 /* Scan the list to check for duplicates. For duplicate names, if the
6973 number is the same, break the loop, which causes the name to be
6974 discarded; otherwise, if DUPNAMES is not set, give an error.
6975 If it is set, allow the name with a different number, but continue
6976 scanning in case this is a duplicate with the same number. For
6977 non-duplicate names, give an error if the number is duplicated. */
6979 ng = cd->named_groups;
6980 for (i = 0; i < cd->names_found; i++, ng++)
6982 if (namelen == ng->length &&
6983 STRNCMP_UC_UC(name, ng->name, namelen) == 0)
6985 if (ng->number == number) break;
6986 if ((options & PCRE_DUPNAMES) == 0)
6988 *errorcodeptr = ERR43;
6991 cd->dupnames = TRUE; /* Duplicate names exist */
6993 else if (ng->number == number)
6995 *errorcodeptr = ERR65;
7000 if (i >= cd->names_found) /* Not a duplicate with same number */
7002 /* Increase the list size if necessary */
7004 if (cd->names_found >= cd->named_group_list_size)
7006 int newsize = cd->named_group_list_size * 2;
7007 named_group *newspace = (PUBL(malloc))
7008 (newsize * sizeof(named_group));
7010 if (newspace == NULL)
7012 *errorcodeptr = ERR21;
7016 memcpy(newspace, cd->named_groups,
7017 cd->named_group_list_size * sizeof(named_group));
7018 if (cd->named_group_list_size > NAMED_GROUP_LIST_SIZE)
7019 (PUBL(free))((void *)cd->named_groups);
7020 cd->named_groups = newspace;
7021 cd->named_group_list_size = newsize;
7024 cd->named_groups[cd->names_found].name = name;
7025 cd->named_groups[cd->names_found].length = namelen;
7026 cd->named_groups[cd->names_found].number = number;
7031 ptr++; /* Move past > or ' in both passes. */
7032 goto NUMBERED_GROUP;
7035 /* ------------------------------------------------------------ */
7036 case CHAR_AMPERSAND: /* Perl recursion/subroutine syntax */
7037 terminator = CHAR_RIGHT_PARENTHESIS;
7041 /* We come here from the Python syntax above that handles both
7042 references (?P=name) and recursion (?P>name), as well as falling
7043 through from the Perl recursion syntax (?&name). We also come here from
7044 the Perl \k<name> or \k'name' back reference syntax and the \k{name}
7045 .NET syntax, and the Oniguruma \g<...> and \g'...' subroutine syntax. */
7047 NAMED_REF_OR_RECURSE:
7051 *errorcodeptr = ERR84; /* Group name must start with non-digit */
7054 while (MAX_255(*ptr) && (cd->ctypes[*ptr] & ctype_word) != 0) ptr++;
7055 namelen = (int)(ptr - name);
7057 /* In the pre-compile phase, do a syntax check. We used to just set
7058 a dummy reference number, because it was not used in the first pass.
7059 However, with the change of recursive back references to be atomic,
7060 we have to look for the number so that this state can be identified, as
7061 otherwise the incorrect length is computed. If it's not a backwards
7062 reference, the dummy number will do. */
7064 if (lengthptr != NULL)
7070 *errorcodeptr = ERR62;
7073 if (*ptr != (pcre_uchar)terminator)
7075 *errorcodeptr = ERR42;
7078 if (namelen > MAX_NAME_SIZE)
7080 *errorcodeptr = ERR48;
7084 /* The name table does not exist in the first pass; instead we must
7085 scan the list of names encountered so far in order to get the
7086 number. If the name is not found, set the value to 0 for a forward
7089 ng = cd->named_groups;
7090 for (i = 0; i < cd->names_found; i++, ng++)
7092 if (namelen == ng->length &&
7093 STRNCMP_UC_UC(name, ng->name, namelen) == 0)
7096 recno = (i < cd->names_found)? ng->number : 0;
7098 /* Count named back references. */
7100 if (!is_recurse) cd->namedrefcount++;
7103 /* In the real compile, search the name table. We check the name
7104 first, and then check that we have reached the end of the name in the
7105 table. That way, if the name is longer than any in the table, the
7106 comparison will fail without reading beyond the table entry. */
7110 slot = cd->name_table;
7111 for (i = 0; i < cd->names_found; i++)
7113 if (STRNCMP_UC_UC(name, slot+IMM2_SIZE, namelen) == 0 &&
7114 slot[IMM2_SIZE+namelen] == 0)
7116 slot += cd->name_entry_size;
7119 if (i < cd->names_found)
7121 recno = GET2(slot, 0);
7125 *errorcodeptr = ERR15;
7130 /* In both phases, for recursions, we can now go to the code than
7131 handles numerical recursion. */
7133 if (is_recurse) goto HANDLE_RECURSION;
7135 /* In the second pass we must see if the name is duplicated. If so, we
7136 generate a different opcode. */
7138 if (lengthptr == NULL && cd->dupnames)
7141 unsigned int index = i;
7142 pcre_uchar *cslot = slot + cd->name_entry_size;
7144 for (i++; i < cd->names_found; i++)
7146 if (STRCMP_UC_UC(slot + IMM2_SIZE, cslot + IMM2_SIZE) != 0) break;
7148 cslot += cd->name_entry_size;
7153 if (firstcharflags == REQ_UNSET) firstcharflags = REQ_NONE;
7155 *code++ = ((options & PCRE_CASELESS) != 0)? OP_DNREFI : OP_DNREF;
7156 PUT2INC(code, 0, index);
7157 PUT2INC(code, 0, count);
7159 /* Process each potentially referenced group. */
7161 for (; slot < cslot; slot += cd->name_entry_size)
7164 recno = GET2(slot, 0);
7165 cd->backref_map |= (recno < 32)? (1 << recno) : 1;
7166 if (recno > cd->top_backref) cd->top_backref = recno;
7168 /* Check to see if this back reference is recursive, that it, it
7169 is inside the group that it references. A flag is set so that the
7170 group can be made atomic. */
7172 for (oc = cd->open_caps; oc != NULL; oc = oc->next)
7174 if (oc->number == recno)
7182 continue; /* End of back ref handling */
7186 /* First pass, or a non-duplicated name. */
7188 goto HANDLE_REFERENCE;
7191 /* ------------------------------------------------------------ */
7192 case CHAR_R: /* Recursion */
7193 ptr++; /* Same as (?0) */
7197 /* ------------------------------------------------------------ */
7198 case CHAR_MINUS: case CHAR_PLUS: /* Recursion or subroutine */
7199 case CHAR_0: case CHAR_1: case CHAR_2: case CHAR_3: case CHAR_4:
7200 case CHAR_5: case CHAR_6: case CHAR_7: case CHAR_8: case CHAR_9:
7202 const pcre_uchar *called;
7203 terminator = CHAR_RIGHT_PARENTHESIS;
7205 /* Come here from the \g<...> and \g'...' code (Oniguruma
7206 compatibility). However, the syntax has been checked to ensure that
7207 the ... are a (signed) number, so that neither ERR63 nor ERR29 will
7208 be called on this path, nor with the jump to OTHER_CHAR_AFTER_QUERY
7211 HANDLE_NUMERICAL_RECURSION:
7213 if ((refsign = *ptr) == CHAR_PLUS)
7216 if (!IS_DIGIT(*ptr))
7218 *errorcodeptr = ERR63;
7222 else if (refsign == CHAR_MINUS)
7224 if (!IS_DIGIT(ptr[1]))
7225 goto OTHER_CHAR_AFTER_QUERY;
7230 while(IS_DIGIT(*ptr))
7231 recno = recno * 10 + *ptr++ - CHAR_0;
7233 if (*ptr != (pcre_uchar)terminator)
7235 *errorcodeptr = ERR29;
7239 if (refsign == CHAR_MINUS)
7243 *errorcodeptr = ERR58;
7246 recno = cd->bracount - recno + 1;
7249 *errorcodeptr = ERR15;
7253 else if (refsign == CHAR_PLUS)
7257 *errorcodeptr = ERR58;
7260 recno += cd->bracount;
7263 /* Come here from code above that handles a named recursion */
7268 called = cd->start_code;
7270 /* When we are actually compiling, find the bracket that is being
7271 referenced. Temporarily end the regex in case it doesn't exist before
7272 this point. If we end up with a forward reference, first check that
7273 the bracket does occur later so we can give the error (and position)
7274 now. Then remember this forward reference in the workspace so it can
7275 be filled in at the end. */
7277 if (lengthptr == NULL)
7281 called = PRIV(find_bracket)(cd->start_code, utf, recno);
7283 /* Forward reference */
7287 if (recno > cd->final_bracount)
7289 *errorcodeptr = ERR15;
7293 /* Fudge the value of "called" so that when it is inserted as an
7294 offset below, what it actually inserted is the reference number
7295 of the group. Then remember the forward reference. */
7297 called = cd->start_code + recno;
7298 if (cd->hwm >= cd->start_workspace + cd->workspace_size -
7299 WORK_SIZE_SAFETY_MARGIN)
7301 *errorcodeptr = expand_workspace(cd);
7302 if (*errorcodeptr != 0) goto FAILED;
7304 PUTINC(cd->hwm, 0, (int)(code + 1 - cd->start_code));
7307 /* If not a forward reference, and the subpattern is still open,
7308 this is a recursive call. We check to see if this is a left
7309 recursion that could loop for ever, and diagnose that case. We
7310 must not, however, do this check if we are in a conditional
7311 subpattern because the condition might be testing for recursion in
7312 a pattern such as /(?(R)a+|(?R)b)/, which is perfectly valid.
7313 Forever loops are also detected at runtime, so those that occur in
7314 conditional subpatterns will be picked up then. */
7316 else if (GET(called, 1) == 0 && cond_depth <= 0 &&
7317 could_be_empty(called, code, bcptr, utf, cd))
7319 *errorcodeptr = ERR40;
7324 /* Insert the recursion/subroutine item. It does not have a set first
7325 character (relevant if it is repeated, because it will then be
7326 wrapped with ONCE brackets). */
7329 PUT(code, 1, (int)(called - cd->start_code));
7330 code += 1 + LINK_SIZE;
7331 groupsetfirstchar = FALSE;
7334 /* Can't determine a first byte now */
7336 if (firstcharflags == REQ_UNSET) firstcharflags = REQ_NONE;
7340 /* ------------------------------------------------------------ */
7341 default: /* Other characters: check option setting */
7342 OTHER_CHAR_AFTER_QUERY:
7346 while (*ptr != CHAR_RIGHT_PARENTHESIS && *ptr != CHAR_COLON)
7350 case CHAR_MINUS: optset = &unset; break;
7352 case CHAR_J: /* Record that it changed in the external options */
7353 *optset |= PCRE_DUPNAMES;
7354 cd->external_flags |= PCRE_JCHANGED;
7357 case CHAR_i: *optset |= PCRE_CASELESS; break;
7358 case CHAR_m: *optset |= PCRE_MULTILINE; break;
7359 case CHAR_s: *optset |= PCRE_DOTALL; break;
7360 case CHAR_x: *optset |= PCRE_EXTENDED; break;
7361 case CHAR_U: *optset |= PCRE_UNGREEDY; break;
7362 case CHAR_X: *optset |= PCRE_EXTRA; break;
7364 default: *errorcodeptr = ERR12;
7365 ptr--; /* Correct the offset */
7370 /* Set up the changed option bits, but don't change anything yet. */
7372 newoptions = (options | set) & (~unset);
7374 /* If the options ended with ')' this is not the start of a nested
7375 group with option changes, so the options change at this level. If this
7376 item is right at the start of the pattern, the options can be
7377 abstracted and made external in the pre-compile phase, and ignored in
7378 the compile phase. This can be helpful when matching -- for instance in
7379 caseless checking of required bytes.
7381 If the code pointer is not (cd->start_code + 1 + LINK_SIZE), we are
7382 definitely *not* at the start of the pattern because something has been
7383 compiled. In the pre-compile phase, however, the code pointer can have
7384 that value after the start, because it gets reset as code is discarded
7385 during the pre-compile. However, this can happen only at top level - if
7386 we are within parentheses, the starting BRA will still be present. At
7387 any parenthesis level, the length value can be used to test if anything
7388 has been compiled at that level. Thus, a test for both these conditions
7389 is necessary to ensure we correctly detect the start of the pattern in
7392 If we are not at the pattern start, reset the greedy defaults and the
7393 case value for firstchar and reqchar. */
7395 if (*ptr == CHAR_RIGHT_PARENTHESIS)
7397 if (code == cd->start_code + 1 + LINK_SIZE &&
7398 (lengthptr == NULL || *lengthptr == 2 + 2*LINK_SIZE))
7400 cd->external_options = newoptions;
7404 greedy_default = ((newoptions & PCRE_UNGREEDY) != 0);
7405 greedy_non_default = greedy_default ^ 1;
7406 req_caseopt = ((newoptions & PCRE_CASELESS) != 0)? REQ_CASELESS:0;
7409 /* Change options at this level, and pass them back for use
7410 in subsequent branches. */
7412 *optionsptr = options = newoptions;
7413 previous = NULL; /* This item can't be repeated */
7414 continue; /* It is complete */
7417 /* If the options ended with ':' we are heading into a nested group
7418 with possible change of options. Such groups are non-capturing and are
7419 not assertions of any kind. All we need to do is skip over the ':';
7420 the newoptions value is handled below. */
7424 } /* End of switch for character following (? */
7425 } /* End of (? handling */
7427 /* Opening parenthesis not followed by '*' or '?'. If PCRE_NO_AUTO_CAPTURE
7428 is set, all unadorned brackets become non-capturing and behave like (?:...)
7431 else if ((options & PCRE_NO_AUTO_CAPTURE) != 0)
7436 /* Else we have a capturing group. */
7442 PUT2(code, 1+LINK_SIZE, cd->bracount);
7443 skipbytes = IMM2_SIZE;
7446 /* Process nested bracketed regex. First check for parentheses nested too
7449 if ((cd->parens_depth += 1) > PARENS_NEST_LIMIT)
7451 *errorcodeptr = ERR82;
7455 /* Assertions used not to be repeatable, but this was changed for Perl
7456 compatibility, so all kinds can now be repeated. We copy code into a
7457 non-register variable (tempcode) in order to be able to pass its address
7458 because some compilers complain otherwise. */
7460 previous = code; /* For handling repetition */
7463 tempreqvary = cd->req_varyopt; /* Save value before bracket */
7464 tempbracount = cd->bracount; /* Save value before bracket */
7465 length_prevgroup = 0; /* Initialize for pre-compile phase */
7468 newoptions, /* The complete new option state */
7469 &tempcode, /* Where to put code (updated) */
7470 &ptr, /* Input pointer (updated) */
7471 errorcodeptr, /* Where to put an error message */
7472 (bravalue == OP_ASSERTBACK ||
7473 bravalue == OP_ASSERTBACK_NOT), /* TRUE if back assert */
7474 reset_bracount, /* True if (?| group */
7475 skipbytes, /* Skip over bracket number */
7477 ((bravalue == OP_COND)?1:0), /* Depth of condition subpatterns */
7478 &subfirstchar, /* For possible first char */
7480 &subreqchar, /* For possible last char */
7482 bcptr, /* Current branch chain */
7483 cd, /* Tables block */
7484 (lengthptr == NULL)? NULL : /* Actual compile phase */
7485 &length_prevgroup /* Pre-compile phase */
7489 cd->parens_depth -= 1;
7491 /* If this was an atomic group and there are no capturing groups within it,
7492 generate OP_ONCE_NC instead of OP_ONCE. */
7494 if (bravalue == OP_ONCE && cd->bracount <= tempbracount)
7497 if (bravalue >= OP_ASSERT && bravalue <= OP_ASSERTBACK_NOT)
7498 cd->assert_depth -= 1;
7500 /* At the end of compiling, code is still pointing to the start of the
7501 group, while tempcode has been updated to point past the end of the group.
7502 The pattern pointer (ptr) is on the bracket.
7504 If this is a conditional bracket, check that there are no more than
7505 two branches in the group, or just one if it's a DEFINE group. We do this
7506 in the real compile phase, not in the pre-pass, where the whole group may
7507 not be available. */
7509 if (bravalue == OP_COND && lengthptr == NULL)
7511 pcre_uchar *tc = code;
7518 while (*tc != OP_KET);
7520 /* A DEFINE group is never obeyed inline (the "condition" is always
7521 false). It must have only one branch. */
7523 if (code[LINK_SIZE+1] == OP_DEF)
7527 *errorcodeptr = ERR54;
7530 bravalue = OP_DEF; /* Just a flag to suppress char handling below */
7533 /* A "normal" conditional group. If there is just one branch, we must not
7534 make use of its firstchar or reqchar, because this is equivalent to an
7535 empty second branch. */
7541 *errorcodeptr = ERR27;
7544 if (condcount == 1) subfirstcharflags = subreqcharflags = REQ_NONE;
7548 /* Error if hit end of pattern */
7550 if (*ptr != CHAR_RIGHT_PARENTHESIS)
7552 *errorcodeptr = ERR14;
7556 /* In the pre-compile phase, update the length by the length of the group,
7557 less the brackets at either end. Then reduce the compiled code to just a
7558 set of non-capturing brackets so that it doesn't use much memory if it is
7559 duplicated by a quantifier.*/
7561 if (lengthptr != NULL)
7563 if (OFLOW_MAX - *lengthptr < length_prevgroup - 2 - 2*LINK_SIZE)
7565 *errorcodeptr = ERR20;
7568 *lengthptr += length_prevgroup - 2 - 2*LINK_SIZE;
7569 code++; /* This already contains bravalue */
7570 PUTINC(code, 0, 1 + LINK_SIZE);
7572 PUTINC(code, 0, 1 + LINK_SIZE);
7573 break; /* No need to waste time with special character handling */
7576 /* Otherwise update the main code pointer to the end of the group. */
7580 /* For a DEFINE group, required and first character settings are not
7583 if (bravalue == OP_DEF) break;
7585 /* Handle updating of the required and first characters for other types of
7586 group. Update for normal brackets of all kinds, and conditions with two
7587 branches (see code above). If the bracket is followed by a quantifier with
7588 zero repeat, we have to back off. Hence the definition of zeroreqchar and
7589 zerofirstchar outside the main loop so that they can be accessed for the
7592 zeroreqchar = reqchar;
7593 zeroreqcharflags = reqcharflags;
7594 zerofirstchar = firstchar;
7595 zerofirstcharflags = firstcharflags;
7596 groupsetfirstchar = FALSE;
7598 if (bravalue >= OP_ONCE)
7600 /* If we have not yet set a firstchar in this branch, take it from the
7601 subpattern, remembering that it was set here so that a repeat of more
7602 than one can replicate it as reqchar if necessary. If the subpattern has
7603 no firstchar, set "none" for the whole branch. In both cases, a zero
7604 repeat forces firstchar to "none". */
7606 if (firstcharflags == REQ_UNSET)
7608 if (subfirstcharflags >= 0)
7610 firstchar = subfirstchar;
7611 firstcharflags = subfirstcharflags;
7612 groupsetfirstchar = TRUE;
7614 else firstcharflags = REQ_NONE;
7615 zerofirstcharflags = REQ_NONE;
7618 /* If firstchar was previously set, convert the subpattern's firstchar
7619 into reqchar if there wasn't one, using the vary flag that was in
7620 existence beforehand. */
7622 else if (subfirstcharflags >= 0 && subreqcharflags < 0)
7624 subreqchar = subfirstchar;
7625 subreqcharflags = subfirstcharflags | tempreqvary;
7628 /* If the subpattern set a required byte (or set a first byte that isn't
7629 really the first byte - see above), set it. */
7631 if (subreqcharflags >= 0)
7633 reqchar = subreqchar;
7634 reqcharflags = subreqcharflags;
7638 /* For a forward assertion, we take the reqchar, if set. This can be
7639 helpful if the pattern that follows the assertion doesn't set a different
7640 char. For example, it's useful for /(?=abcde).+/. We can't set firstchar
7641 for an assertion, however because it leads to incorrect effect for patterns
7642 such as /(?=a)a.+/ when the "real" "a" would then become a reqchar instead
7643 of a firstchar. This is overcome by a scan at the end if there's no
7644 firstchar, looking for an asserted first char. */
7646 else if (bravalue == OP_ASSERT && subreqcharflags >= 0)
7648 reqchar = subreqchar;
7649 reqcharflags = subreqcharflags;
7651 break; /* End of processing '(' */
7654 /* ===================================================================*/
7655 /* Handle metasequences introduced by \. For ones like \d, the ESC_ values
7656 are arranged to be the negation of the corresponding OP_values in the
7657 default case when PCRE_UCP is not set. For the back references, the values
7658 are negative the reference number. Only back references and those types
7659 that consume a character may be repeated. We can test for values between
7660 ESC_b and ESC_Z for the latter; this may have to change if any new ones are
7663 case CHAR_BACKSLASH:
7665 escape = check_escape(&ptr, &ec, errorcodeptr, cd->bracount, options, FALSE);
7666 if (*errorcodeptr != 0) goto FAILED;
7668 if (escape == 0) /* The escape coded a single character */
7672 if (escape == ESC_Q) /* Handle start of quoted string */
7674 if (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
7675 ptr += 2; /* avoid empty string */
7680 if (escape == ESC_E) continue; /* Perl ignores an orphan \E */
7682 /* For metasequences that actually match a character, we disable the
7683 setting of a first character if it hasn't already been set. */
7685 if (firstcharflags == REQ_UNSET && escape > ESC_b && escape < ESC_Z)
7686 firstcharflags = REQ_NONE;
7688 /* Set values to reset to if this is followed by a zero repeat. */
7690 zerofirstchar = firstchar;
7691 zerofirstcharflags = firstcharflags;
7692 zeroreqchar = reqchar;
7693 zeroreqcharflags = reqcharflags;
7695 /* \g<name> or \g'name' is a subroutine call by name and \g<n> or \g'n'
7696 is a subroutine call by number (Oniguruma syntax). In fact, the value
7697 ESC_g is returned only for these cases. So we don't need to check for <
7698 or ' if the value is ESC_g. For the Perl syntax \g{n} the value is
7699 -n, and for the Perl syntax \g{name} the result is ESC_k (as
7700 that is a synonym for a named back reference). */
7702 if (escape == ESC_g)
7704 const pcre_uchar *p;
7707 save_hwm = cd->hwm; /* Normally this is set when '(' is read */
7708 terminator = (*(++ptr) == CHAR_LESS_THAN_SIGN)?
7709 CHAR_GREATER_THAN_SIGN : CHAR_APOSTROPHE;
7711 /* These two statements stop the compiler for warning about possibly
7712 unset variables caused by the jump to HANDLE_NUMERICAL_RECURSION. In
7713 fact, because we do the check for a number below, the paths that
7714 would actually be in error are never taken. */
7717 reset_bracount = FALSE;
7719 /* If it's not a signed or unsigned number, treat it as a name. */
7722 if (cf != CHAR_PLUS && cf != CHAR_MINUS && !IS_DIGIT(cf))
7725 goto NAMED_REF_OR_RECURSE;
7728 /* Signed or unsigned number (cf = ptr[1]) is known to be plus or minus
7732 while (IS_DIGIT(*p)) p++;
7733 if (*p != (pcre_uchar)terminator)
7735 *errorcodeptr = ERR57;
7739 goto HANDLE_NUMERICAL_RECURSION;
7742 /* \k<name> or \k'name' is a back reference by name (Perl syntax).
7743 We also support \k{name} (.NET syntax). */
7745 if (escape == ESC_k)
7747 if ((ptr[1] != CHAR_LESS_THAN_SIGN &&
7748 ptr[1] != CHAR_APOSTROPHE && ptr[1] != CHAR_LEFT_CURLY_BRACKET))
7750 *errorcodeptr = ERR69;
7754 terminator = (*(++ptr) == CHAR_LESS_THAN_SIGN)?
7755 CHAR_GREATER_THAN_SIGN : (*ptr == CHAR_APOSTROPHE)?
7756 CHAR_APOSTROPHE : CHAR_RIGHT_CURLY_BRACKET;
7757 goto NAMED_REF_OR_RECURSE;
7760 /* Back references are handled specially; must disable firstchar if
7761 not set to cope with cases like (?=(\w+))\1: which would otherwise set
7769 /* Come here from named backref handling when the reference is to a
7770 single group (i.e. not to a duplicated name. */
7773 if (firstcharflags == REQ_UNSET) firstcharflags = REQ_NONE;
7775 *code++ = ((options & PCRE_CASELESS) != 0)? OP_REFI : OP_REF;
7776 PUT2INC(code, 0, recno);
7777 cd->backref_map |= (recno < 32)? (1 << recno) : 1;
7778 if (recno > cd->top_backref) cd->top_backref = recno;
7780 /* Check to see if this back reference is recursive, that it, it
7781 is inside the group that it references. A flag is set so that the
7782 group can be made atomic. */
7784 for (oc = cd->open_caps; oc != NULL; oc = oc->next)
7786 if (oc->number == recno)
7794 /* So are Unicode property matches, if supported. */
7797 else if (escape == ESC_P || escape == ESC_p)
7800 unsigned int ptype = 0, pdata = 0;
7801 if (!get_ucp(&ptr, &negated, &ptype, &pdata, errorcodeptr))
7804 *code++ = ((escape == ESC_p) != negated)? OP_PROP : OP_NOTPROP;
7810 /* If Unicode properties are not supported, \X, \P, and \p are not
7813 else if (escape == ESC_X || escape == ESC_P || escape == ESC_p)
7815 *errorcodeptr = ERR45;
7820 /* For the rest (including \X when Unicode properties are supported), we
7821 can obtain the OP value by negating the escape value in the default
7822 situation when PCRE_UCP is not set. When it *is* set, we substitute
7823 Unicode property tests. Note that \b and \B do a one-character
7824 lookbehind, and \A also behaves as if it does. */
7828 if ((escape == ESC_b || escape == ESC_B || escape == ESC_A) &&
7829 cd->max_lookbehind == 0)
7830 cd->max_lookbehind = 1;
7832 if (escape >= ESC_DU && escape <= ESC_wu)
7834 nestptr = ptr + 1; /* Where to resume */
7835 ptr = substitutes[escape - ESC_DU] - 1; /* Just before substitute */
7839 /* In non-UTF-8 mode, we turn \C into OP_ALLANY instead of OP_ANYBYTE
7840 so that it works in DFA mode and in lookbehinds. */
7843 previous = (escape > ESC_b && escape < ESC_Z)? code : NULL;
7844 *code++ = (!utf && escape == ESC_C)? OP_ALLANY : escape;
7850 /* We have a data character whose value is in c. In UTF-8 mode it may have
7851 a value > 127. We set its representation in the length/buffer, and then
7852 handle it as a data character. */
7854 #if defined SUPPORT_UTF && !defined COMPILE_PCRE32
7855 if (utf && c > MAX_VALUE_FOR_SINGLE_CHAR)
7856 mclength = PRIV(ord2utf)(c, mcbuffer);
7867 /* ===================================================================*/
7868 /* Handle a literal character. It is guaranteed not to be whitespace or #
7869 when the extended flag is set. If we are in a UTF mode, it may be a
7870 multi-unit literal character. */
7878 if (utf && HAS_EXTRALEN(c))
7879 ACROSSCHAR(TRUE, ptr[1], mcbuffer[mclength++] = *(++ptr));
7882 /* At this point we have the character's bytes in mcbuffer, and the length
7883 in mclength. When not in UTF-8 mode, the length is always 1. */
7888 /* For caseless UTF-8 mode when UCP support is available, check whether
7889 this character has more than one other case. If so, generate a special
7890 OP_PROP item instead of OP_CHARI. */
7893 if (utf && (options & PCRE_CASELESS) != 0)
7895 GETCHAR(c, mcbuffer);
7896 if ((c = UCD_CASESET(c)) != 0)
7901 if (firstcharflags == REQ_UNSET)
7902 firstcharflags = zerofirstcharflags = REQ_NONE;
7908 /* Caseful matches, or not one of the multicase characters. */
7910 *code++ = ((options & PCRE_CASELESS) != 0)? OP_CHARI : OP_CHAR;
7911 for (c = 0; c < mclength; c++) *code++ = mcbuffer[c];
7913 /* Remember if \r or \n were seen */
7915 if (mcbuffer[0] == CHAR_CR || mcbuffer[0] == CHAR_NL)
7916 cd->external_flags |= PCRE_HASCRORLF;
7918 /* Set the first and required bytes appropriately. If no previous first
7919 byte, set it from this character, but revert to none on a zero repeat.
7920 Otherwise, leave the firstchar value alone, and don't change it on a zero
7923 if (firstcharflags == REQ_UNSET)
7925 zerofirstcharflags = REQ_NONE;
7926 zeroreqchar = reqchar;
7927 zeroreqcharflags = reqcharflags;
7929 /* If the character is more than one byte long, we can set firstchar
7930 only if it is not to be matched caselessly. */
7932 if (mclength == 1 || req_caseopt == 0)
7934 firstchar = mcbuffer[0] | req_caseopt;
7935 firstchar = mcbuffer[0];
7936 firstcharflags = req_caseopt;
7941 reqcharflags = cd->req_varyopt;
7944 else firstcharflags = reqcharflags = REQ_NONE;
7947 /* firstchar was previously set; we can set reqchar only if the length is
7948 1 or the matching is caseful. */
7952 zerofirstchar = firstchar;
7953 zerofirstcharflags = firstcharflags;
7954 zeroreqchar = reqchar;
7955 zeroreqcharflags = reqcharflags;
7956 if (mclength == 1 || req_caseopt == 0)
7959 reqcharflags = req_caseopt | cd->req_varyopt;
7963 break; /* End of literal character handling */
7965 } /* end of big loop */
7968 /* Control never reaches here by falling through, only by a goto for all the
7969 error states. Pass back the position in the pattern so that it can be displayed
7970 to the user for diagnosing the error. */
7979 /*************************************************
7980 * Compile sequence of alternatives *
7981 *************************************************/
7983 /* On entry, ptr is pointing past the bracket character, but on return it
7984 points to the closing bracket, or vertical bar, or end of string. The code
7985 variable is pointing at the byte into which the BRA operator has been stored.
7986 This function is used during the pre-compile phase when we are trying to find
7987 out the amount of memory needed, as well as during the real compile phase. The
7988 value of lengthptr distinguishes the two phases.
7991 options option bits, including any changes for this subpattern
7992 codeptr -> the address of the current code pointer
7993 ptrptr -> the address of the current pattern pointer
7994 errorcodeptr -> pointer to error code variable
7995 lookbehind TRUE if this is a lookbehind assertion
7996 reset_bracount TRUE to reset the count for each branch
7997 skipbytes skip this many bytes at start (for brackets and OP_COND)
7998 cond_depth depth of nesting for conditional subpatterns
7999 firstcharptr place to put the first required character
8000 firstcharflagsptr place to put the first character flags, or a negative number
8001 reqcharptr place to put the last required character
8002 reqcharflagsptr place to put the last required character flags, or a negative number
8003 bcptr pointer to the chain of currently open branches
8004 cd points to the data block with tables pointers etc.
8005 lengthptr NULL during the real compile phase
8006 points to length accumulator during pre-compile phase
8008 Returns: TRUE on success
8012 compile_regex(int options, pcre_uchar **codeptr, const pcre_uchar **ptrptr,
8013 int *errorcodeptr, BOOL lookbehind, BOOL reset_bracount, int skipbytes,
8015 pcre_uint32 *firstcharptr, pcre_int32 *firstcharflagsptr,
8016 pcre_uint32 *reqcharptr, pcre_int32 *reqcharflagsptr,
8017 branch_chain *bcptr, compile_data *cd, int *lengthptr)
8019 const pcre_uchar *ptr = *ptrptr;
8020 pcre_uchar *code = *codeptr;
8021 pcre_uchar *last_branch = code;
8022 pcre_uchar *start_bracket = code;
8023 pcre_uchar *reverse_count = NULL;
8024 open_capitem capitem;
8026 pcre_uint32 firstchar, reqchar;
8027 pcre_int32 firstcharflags, reqcharflags;
8028 pcre_uint32 branchfirstchar, branchreqchar;
8029 pcre_int32 branchfirstcharflags, branchreqcharflags;
8031 unsigned int orig_bracount;
8032 unsigned int max_bracount;
8035 /* If set, call the external function that checks for stack availability. */
8037 if (PUBL(stack_guard) != NULL && PUBL(stack_guard)())
8039 *errorcodeptr= ERR85;
8043 /* Miscellaneous initialization */
8046 bc.current_branch = code;
8048 firstchar = reqchar = 0;
8049 firstcharflags = reqcharflags = REQ_UNSET;
8051 /* Accumulate the length for use in the pre-compile phase. Start with the
8052 length of the BRA and KET and any extra bytes that are required at the
8053 beginning. We accumulate in a local variable to save frequent testing of
8054 lenthptr for NULL. We cannot do this by looking at the value of code at the
8055 start and end of each alternative, because compiled items are discarded during
8056 the pre-compile phase so that the work space is not exceeded. */
8058 length = 2 + 2*LINK_SIZE + skipbytes;
8060 /* WARNING: If the above line is changed for any reason, you must also change
8061 the code that abstracts option settings at the start of the pattern and makes
8062 them global. It tests the value of length for (2 + 2*LINK_SIZE) in the
8063 pre-compile phase to find out whether anything has yet been compiled or not. */
8065 /* If this is a capturing subpattern, add to the chain of open capturing items
8066 so that we can detect them if (*ACCEPT) is encountered. This is also used to
8067 detect groups that contain recursive back references to themselves. Note that
8068 only OP_CBRA need be tested here; changing this opcode to one of its variants,
8069 e.g. OP_SCBRAPOS, happens later, after the group has been compiled. */
8071 if (*code == OP_CBRA)
8073 capnumber = GET2(code, 1 + LINK_SIZE);
8074 capitem.number = capnumber;
8075 capitem.next = cd->open_caps;
8076 capitem.flag = FALSE;
8077 cd->open_caps = &capitem;
8080 /* Offset is set zero to mark that this bracket is still open */
8083 code += 1 + LINK_SIZE + skipbytes;
8085 /* Loop for each alternative branch */
8087 orig_bracount = max_bracount = cd->bracount;
8090 /* For a (?| group, reset the capturing bracket count so that each branch
8091 uses the same numbers. */
8093 if (reset_bracount) cd->bracount = orig_bracount;
8095 /* Set up dummy OP_REVERSE if lookbehind assertion */
8099 *code++ = OP_REVERSE;
8100 reverse_count = code;
8102 length += 1 + LINK_SIZE;
8105 /* Now compile the branch; in the pre-compile phase its length gets added
8108 if (!compile_branch(&options, &code, &ptr, errorcodeptr, &branchfirstchar,
8109 &branchfirstcharflags, &branchreqchar, &branchreqcharflags, &bc,
8110 cond_depth, cd, (lengthptr == NULL)? NULL : &length))
8116 /* Keep the highest bracket count in case (?| was used and some branch
8117 has fewer than the rest. */
8119 if (cd->bracount > max_bracount) max_bracount = cd->bracount;
8121 /* In the real compile phase, there is some post-processing to be done. */
8123 if (lengthptr == NULL)
8125 /* If this is the first branch, the firstchar and reqchar values for the
8126 branch become the values for the regex. */
8128 if (*last_branch != OP_ALT)
8130 firstchar = branchfirstchar;
8131 firstcharflags = branchfirstcharflags;
8132 reqchar = branchreqchar;
8133 reqcharflags = branchreqcharflags;
8136 /* If this is not the first branch, the first char and reqchar have to
8137 match the values from all the previous branches, except that if the
8138 previous value for reqchar didn't have REQ_VARY set, it can still match,
8139 and we set REQ_VARY for the regex. */
8143 /* If we previously had a firstchar, but it doesn't match the new branch,
8144 we have to abandon the firstchar for the regex, but if there was
8145 previously no reqchar, it takes on the value of the old firstchar. */
8147 if (firstcharflags >= 0 &&
8148 (firstcharflags != branchfirstcharflags || firstchar != branchfirstchar))
8150 if (reqcharflags < 0)
8152 reqchar = firstchar;
8153 reqcharflags = firstcharflags;
8155 firstcharflags = REQ_NONE;
8158 /* If we (now or from before) have no firstchar, a firstchar from the
8159 branch becomes a reqchar if there isn't a branch reqchar. */
8161 if (firstcharflags < 0 && branchfirstcharflags >= 0 && branchreqcharflags < 0)
8163 branchreqchar = branchfirstchar;
8164 branchreqcharflags = branchfirstcharflags;
8167 /* Now ensure that the reqchars match */
8169 if (((reqcharflags & ~REQ_VARY) != (branchreqcharflags & ~REQ_VARY)) ||
8170 reqchar != branchreqchar)
8171 reqcharflags = REQ_NONE;
8174 reqchar = branchreqchar;
8175 reqcharflags |= branchreqcharflags; /* To "or" REQ_VARY */
8179 /* If lookbehind, check that this branch matches a fixed-length string, and
8180 put the length into the OP_REVERSE item. Temporarily mark the end of the
8181 branch with OP_END. If the branch contains OP_RECURSE, the result is -3
8182 because there may be forward references that we can't check here. Set a
8183 flag to cause another lookbehind check at the end. Why not do it all at the
8184 end? Because common, erroneous checks are picked up here and the offset of
8185 the problem can be shown. */
8191 fixed_length = find_fixedlength(last_branch, (options & PCRE_UTF8) != 0,
8193 DPRINTF(("fixed length = %d\n", fixed_length));
8194 if (fixed_length == -3)
8196 cd->check_lookbehind = TRUE;
8198 else if (fixed_length < 0)
8200 *errorcodeptr = (fixed_length == -2)? ERR36 :
8201 (fixed_length == -4)? ERR70: ERR25;
8207 if (fixed_length > cd->max_lookbehind)
8208 cd->max_lookbehind = fixed_length;
8209 PUT(reverse_count, 0, fixed_length);
8214 /* Reached end of expression, either ')' or end of pattern. In the real
8215 compile phase, go back through the alternative branches and reverse the chain
8216 of offsets, with the field in the BRA item now becoming an offset to the
8217 first alternative. If there are no alternatives, it points to the end of the
8218 group. The length in the terminating ket is always the length of the whole
8219 bracketed item. Return leaving the pointer at the terminating char. */
8221 if (*ptr != CHAR_VERTICAL_LINE)
8223 if (lengthptr == NULL)
8225 int branch_length = (int)(code - last_branch);
8228 int prev_length = GET(last_branch, 1);
8229 PUT(last_branch, 1, branch_length);
8230 branch_length = prev_length;
8231 last_branch -= branch_length;
8233 while (branch_length > 0);
8236 /* Fill in the ket */
8239 PUT(code, 1, (int)(code - start_bracket));
8240 code += 1 + LINK_SIZE;
8242 /* If it was a capturing subpattern, check to see if it contained any
8243 recursive back references. If so, we must wrap it in atomic brackets.
8244 In any event, remove the block from the chain. */
8248 if (cd->open_caps->flag)
8250 memmove(start_bracket + 1 + LINK_SIZE, start_bracket,
8251 IN_UCHARS(code - start_bracket));
8252 *start_bracket = OP_ONCE;
8253 code += 1 + LINK_SIZE;
8254 PUT(start_bracket, 1, (int)(code - start_bracket));
8256 PUT(code, 1, (int)(code - start_bracket));
8257 code += 1 + LINK_SIZE;
8258 length += 2 + 2*LINK_SIZE;
8260 cd->open_caps = cd->open_caps->next;
8263 /* Retain the highest bracket number, in case resetting was used. */
8265 cd->bracount = max_bracount;
8267 /* Set values to pass back */
8271 *firstcharptr = firstchar;
8272 *firstcharflagsptr = firstcharflags;
8273 *reqcharptr = reqchar;
8274 *reqcharflagsptr = reqcharflags;
8275 if (lengthptr != NULL)
8277 if (OFLOW_MAX - *lengthptr < length)
8279 *errorcodeptr = ERR20;
8282 *lengthptr += length;
8287 /* Another branch follows. In the pre-compile phase, we can move the code
8288 pointer back to where it was for the start of the first branch. (That is,
8289 pretend that each branch is the only one.)
8291 In the real compile phase, insert an ALT node. Its length field points back
8292 to the previous branch while the bracket remains open. At the end the chain
8293 is reversed. It's done like this so that the start of the bracket has a
8294 zero offset until it is closed, making it possible to detect recursion. */
8296 if (lengthptr != NULL)
8298 code = *codeptr + 1 + LINK_SIZE + skipbytes;
8299 length += 1 + LINK_SIZE;
8304 PUT(code, 1, (int)(code - last_branch));
8305 bc.current_branch = last_branch = code;
8306 code += 1 + LINK_SIZE;
8311 /* Control never reaches here */
8317 /*************************************************
8318 * Check for anchored expression *
8319 *************************************************/
8321 /* Try to find out if this is an anchored regular expression. Consider each
8322 alternative branch. If they all start with OP_SOD or OP_CIRC, or with a bracket
8323 all of whose alternatives start with OP_SOD or OP_CIRC (recurse ad lib), then
8324 it's anchored. However, if this is a multiline pattern, then only OP_SOD will
8325 be found, because ^ generates OP_CIRCM in that mode.
8327 We can also consider a regex to be anchored if OP_SOM starts all its branches.
8328 This is the code for \G, which means "match at start of match position, taking
8329 into account the match offset".
8331 A branch is also implicitly anchored if it starts with .* and DOTALL is set,
8332 because that will try the rest of the pattern at all possible matching points,
8333 so there is no point trying again.... er ....
8335 .... except when the .* appears inside capturing parentheses, and there is a
8336 subsequent back reference to those parentheses. We haven't enough information
8337 to catch that case precisely.
8339 At first, the best we could do was to detect when .* was in capturing brackets
8340 and the highest back reference was greater than or equal to that level.
8341 However, by keeping a bitmap of the first 31 back references, we can catch some
8342 of the more common cases more precisely.
8344 ... A second exception is when the .* appears inside an atomic group, because
8345 this prevents the number of characters it matches from being adjusted.
8348 code points to start of expression (the bracket)
8349 bracket_map a bitmap of which brackets we are inside while testing; this
8350 handles up to substring 31; after that we just have to take
8351 the less precise approach
8352 cd points to the compile data block
8353 atomcount atomic group level
8355 Returns: TRUE or FALSE
8359 is_anchored(register const pcre_uchar *code, unsigned int bracket_map,
8360 compile_data *cd, int atomcount)
8363 const pcre_uchar *scode = first_significant_code(
8364 code + PRIV(OP_lengths)[*code], FALSE);
8365 register int op = *scode;
8367 /* Non-capturing brackets */
8369 if (op == OP_BRA || op == OP_BRAPOS ||
8370 op == OP_SBRA || op == OP_SBRAPOS)
8372 if (!is_anchored(scode, bracket_map, cd, atomcount)) return FALSE;
8375 /* Capturing brackets */
8377 else if (op == OP_CBRA || op == OP_CBRAPOS ||
8378 op == OP_SCBRA || op == OP_SCBRAPOS)
8380 int n = GET2(scode, 1+LINK_SIZE);
8381 int new_map = bracket_map | ((n < 32)? (1 << n) : 1);
8382 if (!is_anchored(scode, new_map, cd, atomcount)) return FALSE;
8385 /* Positive forward assertions and conditions */
8387 else if (op == OP_ASSERT || op == OP_COND)
8389 if (!is_anchored(scode, bracket_map, cd, atomcount)) return FALSE;
8394 else if (op == OP_ONCE || op == OP_ONCE_NC)
8396 if (!is_anchored(scode, bracket_map, cd, atomcount + 1))
8400 /* .* is not anchored unless DOTALL is set (which generates OP_ALLANY) and
8401 it isn't in brackets that are or may be referenced or inside an atomic
8404 else if ((op == OP_TYPESTAR || op == OP_TYPEMINSTAR ||
8405 op == OP_TYPEPOSSTAR))
8407 if (scode[1] != OP_ALLANY || (bracket_map & cd->backref_map) != 0 ||
8408 atomcount > 0 || cd->had_pruneorskip)
8412 /* Check for explicit anchoring */
8414 else if (op != OP_SOD && op != OP_SOM && op != OP_CIRC) return FALSE;
8416 code += GET(code, 1);
8418 while (*code == OP_ALT); /* Loop for each alternative */
8424 /*************************************************
8425 * Check for starting with ^ or .* *
8426 *************************************************/
8428 /* This is called to find out if every branch starts with ^ or .* so that
8429 "first char" processing can be done to speed things up in multiline
8430 matching and for non-DOTALL patterns that start with .* (which must start at
8431 the beginning or after \n). As in the case of is_anchored() (see above), we
8432 have to take account of back references to capturing brackets that contain .*
8433 because in that case we can't make the assumption. Also, the appearance of .*
8434 inside atomic brackets or in a pattern that contains *PRUNE or *SKIP does not
8435 count, because once again the assumption no longer holds.
8438 code points to start of expression (the bracket)
8439 bracket_map a bitmap of which brackets we are inside while testing; this
8440 handles up to substring 31; after that we just have to take
8441 the less precise approach
8442 cd points to the compile data
8443 atomcount atomic group level
8445 Returns: TRUE or FALSE
8449 is_startline(const pcre_uchar *code, unsigned int bracket_map,
8450 compile_data *cd, int atomcount)
8453 const pcre_uchar *scode = first_significant_code(
8454 code + PRIV(OP_lengths)[*code], FALSE);
8455 register int op = *scode;
8457 /* If we are at the start of a conditional assertion group, *both* the
8458 conditional assertion *and* what follows the condition must satisfy the test
8459 for start of line. Other kinds of condition fail. Note that there may be an
8460 auto-callout at the start of a condition. */
8464 scode += 1 + LINK_SIZE;
8465 if (*scode == OP_CALLOUT) scode += PRIV(OP_lengths)[OP_CALLOUT];
8475 default: /* Assertion */
8476 if (!is_startline(scode, bracket_map, cd, atomcount)) return FALSE;
8477 do scode += GET(scode, 1); while (*scode == OP_ALT);
8478 scode += 1 + LINK_SIZE;
8481 scode = first_significant_code(scode, FALSE);
8485 /* Non-capturing brackets */
8487 if (op == OP_BRA || op == OP_BRAPOS ||
8488 op == OP_SBRA || op == OP_SBRAPOS)
8490 if (!is_startline(scode, bracket_map, cd, atomcount)) return FALSE;
8493 /* Capturing brackets */
8495 else if (op == OP_CBRA || op == OP_CBRAPOS ||
8496 op == OP_SCBRA || op == OP_SCBRAPOS)
8498 int n = GET2(scode, 1+LINK_SIZE);
8499 int new_map = bracket_map | ((n < 32)? (1 << n) : 1);
8500 if (!is_startline(scode, new_map, cd, atomcount)) return FALSE;
8503 /* Positive forward assertions */
8505 else if (op == OP_ASSERT)
8507 if (!is_startline(scode, bracket_map, cd, atomcount)) return FALSE;
8510 /* Atomic brackets */
8512 else if (op == OP_ONCE || op == OP_ONCE_NC)
8514 if (!is_startline(scode, bracket_map, cd, atomcount + 1)) return FALSE;
8517 /* .* means "start at start or after \n" if it isn't in atomic brackets or
8518 brackets that may be referenced, as long as the pattern does not contain
8519 *PRUNE or *SKIP, because these break the feature. Consider, for example,
8520 /.*?a(*PRUNE)b/ with the subject "aab", which matches "ab", i.e. not at the
8523 else if (op == OP_TYPESTAR || op == OP_TYPEMINSTAR || op == OP_TYPEPOSSTAR)
8525 if (scode[1] != OP_ANY || (bracket_map & cd->backref_map) != 0 ||
8526 atomcount > 0 || cd->had_pruneorskip)
8530 /* Check for explicit circumflex; anything else gives a FALSE result. Note
8531 in particular that this includes atomic brackets OP_ONCE and OP_ONCE_NC
8532 because the number of characters matched by .* cannot be adjusted inside
8535 else if (op != OP_CIRC && op != OP_CIRCM) return FALSE;
8537 /* Move on to the next alternative */
8539 code += GET(code, 1);
8541 while (*code == OP_ALT); /* Loop for each alternative */
8547 /*************************************************
8548 * Check for asserted fixed first char *
8549 *************************************************/
8551 /* During compilation, the "first char" settings from forward assertions are
8552 discarded, because they can cause conflicts with actual literals that follow.
8553 However, if we end up without a first char setting for an unanchored pattern,
8554 it is worth scanning the regex to see if there is an initial asserted first
8555 char. If all branches start with the same asserted char, or with a
8556 non-conditional bracket all of whose alternatives start with the same asserted
8557 char (recurse ad lib), then we return that char, with the flags set to zero or
8558 REQ_CASELESS; otherwise return zero with REQ_NONE in the flags.
8561 code points to start of expression (the bracket)
8562 flags points to the first char flags, or to REQ_NONE
8563 inassert TRUE if in an assertion
8565 Returns: the fixed first char, or 0 with REQ_NONE in flags
8569 find_firstassertedchar(const pcre_uchar *code, pcre_int32 *flags,
8572 register pcre_uint32 c = 0;
8573 int cflags = REQ_NONE;
8579 int xl = (*code == OP_CBRA || *code == OP_SCBRA ||
8580 *code == OP_CBRAPOS || *code == OP_SCBRAPOS)? IMM2_SIZE:0;
8581 const pcre_uchar *scode = first_significant_code(code + 1+LINK_SIZE + xl,
8583 register pcre_uchar op = *scode;
8599 d = find_firstassertedchar(scode, &dflags, op == OP_ASSERT);
8602 if (cflags < 0) { c = d; cflags = dflags; } else if (c != d || cflags != dflags) return 0;
8613 if (!inassert) return 0;
8614 if (cflags < 0) { c = scode[1]; cflags = 0; }
8615 else if (c != scode[1]) return 0;
8626 if (!inassert) return 0;
8627 if (cflags < 0) { c = scode[1]; cflags = REQ_CASELESS; }
8628 else if (c != scode[1]) return 0;
8632 code += GET(code, 1);
8634 while (*code == OP_ALT);
8642 /*************************************************
8643 * Add an entry to the name/number table *
8644 *************************************************/
8646 /* This function is called between compiling passes to add an entry to the
8647 name/number table, maintaining alphabetical order. Checking for permitted
8648 and forbidden duplicates has already been done.
8651 cd the compile data block
8652 name the name to add
8653 length the length of the name
8654 groupno the group number
8660 add_name(compile_data *cd, const pcre_uchar *name, int length,
8661 unsigned int groupno)
8664 pcre_uchar *slot = cd->name_table;
8666 for (i = 0; i < cd->names_found; i++)
8668 int crc = memcmp(name, slot+IMM2_SIZE, IN_UCHARS(length));
8669 if (crc == 0 && slot[IMM2_SIZE+length] != 0)
8670 crc = -1; /* Current name is a substring */
8672 /* Make space in the table and break the loop for an earlier name. For a
8673 duplicate or later name, carry on. We do this for duplicates so that in the
8674 simple case (when ?(| is not used) they are in order of their numbers. In all
8675 cases they are in the order in which they appear in the pattern. */
8679 memmove(slot + cd->name_entry_size, slot,
8680 IN_UCHARS((cd->names_found - i) * cd->name_entry_size));
8684 /* Continue the loop for a later or duplicate name */
8686 slot += cd->name_entry_size;
8689 PUT2(slot, 0, groupno);
8690 memcpy(slot + IMM2_SIZE, name, IN_UCHARS(length));
8691 slot[IMM2_SIZE + length] = 0;
8697 /*************************************************
8698 * Compile a Regular Expression *
8699 *************************************************/
8701 /* This function takes a string and returns a pointer to a block of store
8702 holding a compiled version of the expression. The original API for this
8703 function had no error code return variable; it is retained for backwards
8704 compatibility. The new function is given a new name.
8707 pattern the regular expression
8708 options various option bits
8709 errorcodeptr pointer to error code variable (pcre_compile2() only)
8710 can be NULL if you don't want a code value
8711 errorptr pointer to pointer to error text
8712 erroroffset ptr offset in pattern where error was detected
8713 tables pointer to character tables or NULL
8715 Returns: pointer to compiled data block, or NULL on error,
8716 with errorptr and erroroffset set
8719 #if defined COMPILE_PCRE8
8720 PCRE_EXP_DEFN pcre * PCRE_CALL_CONVENTION
8721 pcre_compile(const char *pattern, int options, const char **errorptr,
8722 int *erroroffset, const unsigned char *tables)
8723 #elif defined COMPILE_PCRE16
8724 PCRE_EXP_DEFN pcre16 * PCRE_CALL_CONVENTION
8725 pcre16_compile(PCRE_SPTR16 pattern, int options, const char **errorptr,
8726 int *erroroffset, const unsigned char *tables)
8727 #elif defined COMPILE_PCRE32
8728 PCRE_EXP_DEFN pcre32 * PCRE_CALL_CONVENTION
8729 pcre32_compile(PCRE_SPTR32 pattern, int options, const char **errorptr,
8730 int *erroroffset, const unsigned char *tables)
8733 #if defined COMPILE_PCRE8
8734 return pcre_compile2(pattern, options, NULL, errorptr, erroroffset, tables);
8735 #elif defined COMPILE_PCRE16
8736 return pcre16_compile2(pattern, options, NULL, errorptr, erroroffset, tables);
8737 #elif defined COMPILE_PCRE32
8738 return pcre32_compile2(pattern, options, NULL, errorptr, erroroffset, tables);
8743 #if defined COMPILE_PCRE8
8744 PCRE_EXP_DEFN pcre * PCRE_CALL_CONVENTION
8745 pcre_compile2(const char *pattern, int options, int *errorcodeptr,
8746 const char **errorptr, int *erroroffset, const unsigned char *tables)
8747 #elif defined COMPILE_PCRE16
8748 PCRE_EXP_DEFN pcre16 * PCRE_CALL_CONVENTION
8749 pcre16_compile2(PCRE_SPTR16 pattern, int options, int *errorcodeptr,
8750 const char **errorptr, int *erroroffset, const unsigned char *tables)
8751 #elif defined COMPILE_PCRE32
8752 PCRE_EXP_DEFN pcre32 * PCRE_CALL_CONVENTION
8753 pcre32_compile2(PCRE_SPTR32 pattern, int options, int *errorcodeptr,
8754 const char **errorptr, int *erroroffset, const unsigned char *tables)
8758 int length = 1; /* For final END opcode */
8759 pcre_int32 firstcharflags, reqcharflags;
8760 pcre_uint32 firstchar, reqchar;
8761 pcre_uint32 limit_match = PCRE_UINT32_MAX;
8762 pcre_uint32 limit_recursion = PCRE_UINT32_MAX;
8765 int skipatstart = 0;
8767 BOOL never_utf = FALSE;
8770 const pcre_uchar *codestart;
8771 const pcre_uchar *ptr;
8772 compile_data compile_block;
8773 compile_data *cd = &compile_block;
8775 /* This space is used for "compiling" into during the first phase, when we are
8776 computing the amount of memory that is needed. Compiled items are thrown away
8777 as soon as possible, so that a fairly large buffer should be sufficient for
8778 this purpose. The same space is used in the second phase for remembering where
8779 to fill in forward references to subpatterns. That may overflow, in which case
8780 new memory is obtained from malloc(). */
8782 pcre_uchar cworkspace[COMPILE_WORK_SIZE];
8784 /* This vector is used for remembering name groups during the pre-compile. In a
8785 similar way to cworkspace, it can be expanded using malloc() if necessary. */
8787 named_group named_groups[NAMED_GROUP_LIST_SIZE];
8789 /* Set this early so that early errors get offset 0. */
8791 ptr = (const pcre_uchar *)pattern;
8793 /* We can't pass back an error message if errorptr is NULL; I guess the best we
8794 can do is just return NULL, but we can set a code value if there is a code
8797 if (errorptr == NULL)
8799 if (errorcodeptr != NULL) *errorcodeptr = 99;
8804 if (errorcodeptr != NULL) *errorcodeptr = ERR0;
8806 /* However, we can give a message for this error */
8808 if (erroroffset == NULL)
8811 goto PCRE_EARLY_ERROR_RETURN2;
8816 /* Set up pointers to the individual character tables */
8818 if (tables == NULL) tables = PRIV(default_tables);
8819 cd->lcc = tables + lcc_offset;
8820 cd->fcc = tables + fcc_offset;
8821 cd->cbits = tables + cbits_offset;
8822 cd->ctypes = tables + ctypes_offset;
8824 /* Check that all undefined public option bits are zero */
8826 if ((options & ~PUBLIC_COMPILE_OPTIONS) != 0)
8829 goto PCRE_EARLY_ERROR_RETURN;
8832 /* If PCRE_NEVER_UTF is set, remember it. */
8834 if ((options & PCRE_NEVER_UTF) != 0) never_utf = TRUE;
8836 /* Check for global one-time settings at the start of the pattern, and remember
8837 the offset for later. */
8839 cd->external_flags = 0; /* Initialize here for LIMIT_MATCH/RECURSION */
8841 while (ptr[skipatstart] == CHAR_LEFT_PARENTHESIS &&
8842 ptr[skipatstart+1] == CHAR_ASTERISK)
8847 /* For completeness and backward compatibility, (*UTFn) is supported in the
8848 relevant libraries, but (*UTF) is generic and always supported. Note that
8849 PCRE_UTF8 == PCRE_UTF16 == PCRE_UTF32. */
8851 #ifdef COMPILE_PCRE8
8852 if (STRNCMP_UC_C8(ptr+skipatstart+2, STRING_UTF8_RIGHTPAR, 5) == 0)
8853 { skipatstart += 7; options |= PCRE_UTF8; continue; }
8855 #ifdef COMPILE_PCRE16
8856 if (STRNCMP_UC_C8(ptr+skipatstart+2, STRING_UTF16_RIGHTPAR, 6) == 0)
8857 { skipatstart += 8; options |= PCRE_UTF16; continue; }
8859 #ifdef COMPILE_PCRE32
8860 if (STRNCMP_UC_C8(ptr+skipatstart+2, STRING_UTF32_RIGHTPAR, 6) == 0)
8861 { skipatstart += 8; options |= PCRE_UTF32; continue; }
8864 else if (STRNCMP_UC_C8(ptr+skipatstart+2, STRING_UTF_RIGHTPAR, 4) == 0)
8865 { skipatstart += 6; options |= PCRE_UTF8; continue; }
8866 else if (STRNCMP_UC_C8(ptr+skipatstart+2, STRING_UCP_RIGHTPAR, 4) == 0)
8867 { skipatstart += 6; options |= PCRE_UCP; continue; }
8868 else if (STRNCMP_UC_C8(ptr+skipatstart+2, STRING_NO_AUTO_POSSESS_RIGHTPAR, 16) == 0)
8869 { skipatstart += 18; options |= PCRE_NO_AUTO_POSSESS; continue; }
8870 else if (STRNCMP_UC_C8(ptr+skipatstart+2, STRING_NO_START_OPT_RIGHTPAR, 13) == 0)
8871 { skipatstart += 15; options |= PCRE_NO_START_OPTIMIZE; continue; }
8873 else if (STRNCMP_UC_C8(ptr+skipatstart+2, STRING_LIMIT_MATCH_EQ, 12) == 0)
8876 int p = skipatstart + 14;
8877 while (isdigit(ptr[p]))
8879 if (c > PCRE_UINT32_MAX / 10 - 1) break; /* Integer overflow */
8880 c = c*10 + ptr[p++] - CHAR_0;
8882 if (ptr[p++] != CHAR_RIGHT_PARENTHESIS) break;
8883 if (c < limit_match)
8886 cd->external_flags |= PCRE_MLSET;
8892 else if (STRNCMP_UC_C8(ptr+skipatstart+2, STRING_LIMIT_RECURSION_EQ, 16) == 0)
8895 int p = skipatstart + 18;
8896 while (isdigit(ptr[p]))
8898 if (c > PCRE_UINT32_MAX / 10 - 1) break; /* Integer overflow check */
8899 c = c*10 + ptr[p++] - CHAR_0;
8901 if (ptr[p++] != CHAR_RIGHT_PARENTHESIS) break;
8902 if (c < limit_recursion)
8904 limit_recursion = c;
8905 cd->external_flags |= PCRE_RLSET;
8911 if (STRNCMP_UC_C8(ptr+skipatstart+2, STRING_CR_RIGHTPAR, 3) == 0)
8912 { skipatstart += 5; newnl = PCRE_NEWLINE_CR; }
8913 else if (STRNCMP_UC_C8(ptr+skipatstart+2, STRING_LF_RIGHTPAR, 3) == 0)
8914 { skipatstart += 5; newnl = PCRE_NEWLINE_LF; }
8915 else if (STRNCMP_UC_C8(ptr+skipatstart+2, STRING_CRLF_RIGHTPAR, 5) == 0)
8916 { skipatstart += 7; newnl = PCRE_NEWLINE_CR + PCRE_NEWLINE_LF; }
8917 else if (STRNCMP_UC_C8(ptr+skipatstart+2, STRING_ANY_RIGHTPAR, 4) == 0)
8918 { skipatstart += 6; newnl = PCRE_NEWLINE_ANY; }
8919 else if (STRNCMP_UC_C8(ptr+skipatstart+2, STRING_ANYCRLF_RIGHTPAR, 8) == 0)
8920 { skipatstart += 10; newnl = PCRE_NEWLINE_ANYCRLF; }
8922 else if (STRNCMP_UC_C8(ptr+skipatstart+2, STRING_BSR_ANYCRLF_RIGHTPAR, 12) == 0)
8923 { skipatstart += 14; newbsr = PCRE_BSR_ANYCRLF; }
8924 else if (STRNCMP_UC_C8(ptr+skipatstart+2, STRING_BSR_UNICODE_RIGHTPAR, 12) == 0)
8925 { skipatstart += 14; newbsr = PCRE_BSR_UNICODE; }
8928 options = (options & ~PCRE_NEWLINE_BITS) | newnl;
8929 else if (newbsr != 0)
8930 options = (options & ~(PCRE_BSR_ANYCRLF|PCRE_BSR_UNICODE)) | newbsr;
8934 /* PCRE_UTF(16|32) have the same value as PCRE_UTF8. */
8935 utf = (options & PCRE_UTF8) != 0;
8936 if (utf && never_utf)
8939 goto PCRE_EARLY_ERROR_RETURN2;
8942 /* Can't support UTF unless PCRE has been compiled to include the code. The
8943 return of an error code from PRIV(valid_utf)() is a new feature, introduced in
8944 release 8.13. It is passed back from pcre_[dfa_]exec(), but at the moment is
8948 if (utf && (options & PCRE_NO_UTF8_CHECK) == 0 &&
8949 (errorcode = PRIV(valid_utf)((PCRE_PUCHAR)pattern, -1, erroroffset)) != 0)
8951 #if defined COMPILE_PCRE8
8953 #elif defined COMPILE_PCRE16
8955 #elif defined COMPILE_PCRE32
8958 goto PCRE_EARLY_ERROR_RETURN2;
8964 goto PCRE_EARLY_ERROR_RETURN;
8968 /* Can't support UCP unless PCRE has been compiled to include the code. */
8971 if ((options & PCRE_UCP) != 0)
8974 goto PCRE_EARLY_ERROR_RETURN;
8978 /* Check validity of \R options. */
8980 if ((options & (PCRE_BSR_ANYCRLF|PCRE_BSR_UNICODE)) ==
8981 (PCRE_BSR_ANYCRLF|PCRE_BSR_UNICODE))
8984 goto PCRE_EARLY_ERROR_RETURN;
8987 /* Handle different types of newline. The three bits give seven cases. The
8988 current code allows for fixed one- or two-byte sequences, plus "any" and
8991 switch (options & PCRE_NEWLINE_BITS)
8993 case 0: newline = NEWLINE; break; /* Build-time default */
8994 case PCRE_NEWLINE_CR: newline = CHAR_CR; break;
8995 case PCRE_NEWLINE_LF: newline = CHAR_NL; break;
8996 case PCRE_NEWLINE_CR+
8997 PCRE_NEWLINE_LF: newline = (CHAR_CR << 8) | CHAR_NL; break;
8998 case PCRE_NEWLINE_ANY: newline = -1; break;
8999 case PCRE_NEWLINE_ANYCRLF: newline = -2; break;
9000 default: errorcode = ERR56; goto PCRE_EARLY_ERROR_RETURN;
9005 cd->nltype = NLTYPE_ANYCRLF;
9007 else if (newline < 0)
9009 cd->nltype = NLTYPE_ANY;
9013 cd->nltype = NLTYPE_FIXED;
9017 cd->nl[0] = (newline >> 8) & 255;
9018 cd->nl[1] = newline & 255;
9023 cd->nl[0] = newline;
9027 /* Maximum back reference and backref bitmap. The bitmap records up to 31 back
9028 references to help in deciding whether (.*) can be treated as anchored or not.
9031 cd->top_backref = 0;
9032 cd->backref_map = 0;
9034 /* Reflect pattern for debugging output */
9036 DPRINTF(("------------------------------------------------------------------\n"));
9038 print_puchar(stdout, (PCRE_PUCHAR)pattern);
9042 /* Pretend to compile the pattern while actually just accumulating the length
9043 of memory required. This behaviour is triggered by passing a non-NULL final
9044 argument to compile_regex(). We pass a block of workspace (cworkspace) for it
9045 to compile parts of the pattern into; the compiled code is discarded when it is
9046 no longer needed, so hopefully this workspace will never overflow, though there
9047 is a test for its doing so. */
9049 cd->bracount = cd->final_bracount = 0;
9050 cd->names_found = 0;
9051 cd->name_entry_size = 0;
9052 cd->name_table = NULL;
9053 cd->dupnames = FALSE;
9054 cd->namedrefcount = 0;
9055 cd->start_code = cworkspace;
9056 cd->hwm = cworkspace;
9057 cd->start_workspace = cworkspace;
9058 cd->workspace_size = COMPILE_WORK_SIZE;
9059 cd->named_groups = named_groups;
9060 cd->named_group_list_size = NAMED_GROUP_LIST_SIZE;
9061 cd->start_pattern = (const pcre_uchar *)pattern;
9062 cd->end_pattern = (const pcre_uchar *)(pattern + STRLEN_UC((const pcre_uchar *)pattern));
9063 cd->req_varyopt = 0;
9064 cd->parens_depth = 0;
9065 cd->assert_depth = 0;
9066 cd->max_lookbehind = 0;
9067 cd->external_options = options;
9068 cd->open_caps = NULL;
9070 /* Now do the pre-compile. On error, errorcode will be set non-zero, so we
9071 don't need to look at the result of the function here. The initial options have
9072 been put into the cd block so that they can be changed if an option setting is
9073 found within the regex right at the beginning. Bringing initial option settings
9074 outside can help speed up starting point checks. */
9080 (void)compile_regex(cd->external_options, &code, &ptr, &errorcode, FALSE,
9081 FALSE, 0, 0, &firstchar, &firstcharflags, &reqchar, &reqcharflags, NULL,
9083 if (errorcode != 0) goto PCRE_EARLY_ERROR_RETURN;
9085 DPRINTF(("end pre-compile: length=%d workspace=%d\n", length,
9086 (int)(cd->hwm - cworkspace)));
9088 if (length > MAX_PATTERN_SIZE)
9091 goto PCRE_EARLY_ERROR_RETURN;
9094 /* If there are groups with duplicate names and there are also references by
9095 name, we must allow for the possibility of named references to duplicated
9096 groups. These require an extra data item each. */
9098 if (cd->dupnames && cd->namedrefcount > 0)
9099 length += cd->namedrefcount * IMM2_SIZE * sizeof(pcre_uchar);
9101 /* Compute the size of the data block for storing the compiled pattern. Integer
9102 overflow should no longer be possible because nowadays we limit the maximum
9103 value of cd->names_found and cd->name_entry_size. */
9105 size = sizeof(REAL_PCRE) +
9106 (length + cd->names_found * cd->name_entry_size) * sizeof(pcre_uchar);
9108 /* Get the memory. */
9110 re = (REAL_PCRE *)(PUBL(malloc))(size);
9114 goto PCRE_EARLY_ERROR_RETURN;
9117 /* Put in the magic number, and save the sizes, initial options, internal
9118 flags, and character table pointer. NULL is used for the default character
9119 tables. The nullpad field is at the end; it's there to help in the case when a
9120 regex compiled on a system with 4-byte pointers is run on another with 8-byte
9123 re->magic_number = MAGIC_NUMBER;
9124 re->size = (int)size;
9125 re->options = cd->external_options;
9126 re->flags = cd->external_flags;
9127 re->limit_match = limit_match;
9128 re->limit_recursion = limit_recursion;
9131 re->name_table_offset = sizeof(REAL_PCRE) / sizeof(pcre_uchar);
9132 re->name_entry_size = cd->name_entry_size;
9133 re->name_count = cd->names_found;
9135 re->tables = (tables == PRIV(default_tables))? NULL : tables;
9137 #ifdef COMPILE_PCRE32
9140 re->dummy1 = re->dummy2 = re->dummy3 = 0;
9143 /* The starting points of the name/number translation table and of the code are
9144 passed around in the compile data block. The start/end pattern and initial
9145 options are already set from the pre-compile phase, as is the name_entry_size
9146 field. Reset the bracket count and the names_found field. Also reset the hwm
9147 field; this time it's used for remembering forward references to subpatterns.
9150 cd->final_bracount = cd->bracount; /* Save for checking forward references */
9151 cd->parens_depth = 0;
9152 cd->assert_depth = 0;
9154 cd->max_lookbehind = 0;
9155 cd->name_table = (pcre_uchar *)re + re->name_table_offset;
9156 codestart = cd->name_table + re->name_entry_size * re->name_count;
9157 cd->start_code = codestart;
9158 cd->hwm = (pcre_uchar *)(cd->start_workspace);
9159 cd->req_varyopt = 0;
9160 cd->had_accept = FALSE;
9161 cd->had_pruneorskip = FALSE;
9162 cd->check_lookbehind = FALSE;
9163 cd->open_caps = NULL;
9165 /* If any named groups were found, create the name/number table from the list
9166 created in the first pass. */
9168 if (cd->names_found > 0)
9170 int i = cd->names_found;
9171 named_group *ng = cd->named_groups;
9172 cd->names_found = 0;
9173 for (; i > 0; i--, ng++)
9174 add_name(cd, ng->name, ng->length, ng->number);
9175 if (cd->named_group_list_size > NAMED_GROUP_LIST_SIZE)
9176 (PUBL(free))((void *)cd->named_groups);
9179 /* Set up a starting, non-extracting bracket, then compile the expression. On
9180 error, errorcode will be set non-zero, so we don't need to look at the result
9181 of the function here. */
9183 ptr = (const pcre_uchar *)pattern + skipatstart;
9184 code = (pcre_uchar *)codestart;
9186 (void)compile_regex(re->options, &code, &ptr, &errorcode, FALSE, FALSE, 0, 0,
9187 &firstchar, &firstcharflags, &reqchar, &reqcharflags, NULL, cd, NULL);
9188 re->top_bracket = cd->bracount;
9189 re->top_backref = cd->top_backref;
9190 re->max_lookbehind = cd->max_lookbehind;
9191 re->flags = cd->external_flags | PCRE_MODE;
9195 reqchar = 0; /* Must disable after (*ACCEPT) */
9196 reqcharflags = REQ_NONE;
9199 /* If not reached end of pattern on success, there's an excess bracket. */
9201 if (errorcode == 0 && *ptr != CHAR_NULL) errorcode = ERR22;
9203 /* Fill in the terminating state and check for disastrous overflow, but
9204 if debugging, leave the test till after things are printed out. */
9209 if (code - codestart > length) errorcode = ERR23;
9212 #ifdef SUPPORT_VALGRIND
9213 /* If the estimated length exceeds the really used length, mark the extra
9214 allocated memory as unaddressable, so that any out-of-bound reads can be
9216 VALGRIND_MAKE_MEM_NOACCESS(code, (length - (code - codestart)) * sizeof(pcre_uchar));
9219 /* Fill in any forward references that are required. There may be repeated
9220 references; optimize for them, as searching a large regex takes time. */
9222 if (cd->hwm > cd->start_workspace)
9224 int prev_recno = -1;
9225 const pcre_uchar *groupptr = NULL;
9226 while (errorcode == 0 && cd->hwm > cd->start_workspace)
9229 cd->hwm -= LINK_SIZE;
9230 offset = GET(cd->hwm, 0);
9231 recno = GET(codestart, offset);
9232 if (recno != prev_recno)
9234 groupptr = PRIV(find_bracket)(codestart, utf, recno);
9237 if (groupptr == NULL) errorcode = ERR53;
9238 else PUT(((pcre_uchar *)codestart), offset, (int)(groupptr - codestart));
9242 /* If the workspace had to be expanded, free the new memory. Set the pointer to
9243 NULL to indicate that forward references have been filled in. */
9245 if (cd->workspace_size > COMPILE_WORK_SIZE)
9246 (PUBL(free))((void *)cd->start_workspace);
9247 cd->start_workspace = NULL;
9249 /* Give an error if there's back reference to a non-existent capturing
9252 if (errorcode == 0 && re->top_backref > re->top_bracket) errorcode = ERR15;
9254 /* Unless disabled, check whether single character iterators can be
9255 auto-possessified. The function overwrites the appropriate opcode values. */
9257 if ((options & PCRE_NO_AUTO_POSSESS) == 0)
9258 auto_possessify((pcre_uchar *)codestart, utf, cd);
9260 /* If there were any lookbehind assertions that contained OP_RECURSE
9261 (recursions or subroutine calls), a flag is set for them to be checked here,
9262 because they may contain forward references. Actual recursions cannot be fixed
9263 length, but subroutine calls can. It is done like this so that those without
9264 OP_RECURSE that are not fixed length get a diagnosic with a useful offset. The
9265 exceptional ones forgo this. We scan the pattern to check that they are fixed
9266 length, and set their lengths. */
9268 if (cd->check_lookbehind)
9270 pcre_uchar *cc = (pcre_uchar *)codestart;
9272 /* Loop, searching for OP_REVERSE items, and process those that do not have
9273 their length set. (Actually, it will also re-process any that have a length
9274 of zero, but that is a pathological case, and it does no harm.) When we find
9275 one, we temporarily terminate the branch it is in while we scan it. */
9277 for (cc = (pcre_uchar *)PRIV(find_bracket)(codestart, utf, -1);
9279 cc = (pcre_uchar *)PRIV(find_bracket)(cc, utf, -1))
9281 if (GET(cc, 1) == 0)
9284 pcre_uchar *be = cc - 1 - LINK_SIZE + GET(cc, -LINK_SIZE);
9287 fixed_length = find_fixedlength(cc, (re->options & PCRE_UTF8) != 0, TRUE,
9290 DPRINTF(("fixed length = %d\n", fixed_length));
9291 if (fixed_length < 0)
9293 errorcode = (fixed_length == -2)? ERR36 :
9294 (fixed_length == -4)? ERR70 : ERR25;
9297 if (fixed_length > cd->max_lookbehind) cd->max_lookbehind = fixed_length;
9298 PUT(cc, 1, fixed_length);
9300 cc += 1 + LINK_SIZE;
9304 /* Failed to compile, or error while post-processing */
9309 PCRE_EARLY_ERROR_RETURN:
9310 *erroroffset = (int)(ptr - (const pcre_uchar *)pattern);
9311 PCRE_EARLY_ERROR_RETURN2:
9312 *errorptr = find_error_text(errorcode);
9313 if (errorcodeptr != NULL) *errorcodeptr = errorcode;
9317 /* If the anchored option was not passed, set the flag if we can determine that
9318 the pattern is anchored by virtue of ^ characters or \A or anything else, such
9319 as starting with non-atomic .* when DOTALL is set and there are no occurrences
9322 Otherwise, if we know what the first byte has to be, save it, because that
9323 speeds up unanchored matches no end. If not, see if we can set the
9324 PCRE_STARTLINE flag. This is helpful for multiline matches when all branches
9325 start with ^. and also when all branches start with non-atomic .* for
9326 non-DOTALL matches when *PRUNE and SKIP are not present. */
9328 if ((re->options & PCRE_ANCHORED) == 0)
9330 if (is_anchored(codestart, 0, cd, 0)) re->options |= PCRE_ANCHORED;
9333 if (firstcharflags < 0)
9334 firstchar = find_firstassertedchar(codestart, &firstcharflags, FALSE);
9335 if (firstcharflags >= 0) /* Remove caseless flag for non-caseable chars */
9337 #if defined COMPILE_PCRE8
9338 re->first_char = firstchar & 0xff;
9339 #elif defined COMPILE_PCRE16
9340 re->first_char = firstchar & 0xffff;
9341 #elif defined COMPILE_PCRE32
9342 re->first_char = firstchar;
9344 if ((firstcharflags & REQ_CASELESS) != 0)
9346 #if defined SUPPORT_UCP && !(defined COMPILE_PCRE8)
9347 /* We ignore non-ASCII first chars in 8 bit mode. */
9350 if (re->first_char < 128)
9352 if (cd->fcc[re->first_char] != re->first_char)
9353 re->flags |= PCRE_FCH_CASELESS;
9355 else if (UCD_OTHERCASE(re->first_char) != re->first_char)
9356 re->flags |= PCRE_FCH_CASELESS;
9360 if (MAX_255(re->first_char)
9361 && cd->fcc[re->first_char] != re->first_char)
9362 re->flags |= PCRE_FCH_CASELESS;
9365 re->flags |= PCRE_FIRSTSET;
9368 else if (is_startline(codestart, 0, cd, 0)) re->flags |= PCRE_STARTLINE;
9372 /* For an anchored pattern, we use the "required byte" only if it follows a
9373 variable length item in the regex. Remove the caseless flag for non-caseable
9376 if (reqcharflags >= 0 &&
9377 ((re->options & PCRE_ANCHORED) == 0 || (reqcharflags & REQ_VARY) != 0))
9379 #if defined COMPILE_PCRE8
9380 re->req_char = reqchar & 0xff;
9381 #elif defined COMPILE_PCRE16
9382 re->req_char = reqchar & 0xffff;
9383 #elif defined COMPILE_PCRE32
9384 re->req_char = reqchar;
9386 if ((reqcharflags & REQ_CASELESS) != 0)
9388 #if defined SUPPORT_UCP && !(defined COMPILE_PCRE8)
9389 /* We ignore non-ASCII first chars in 8 bit mode. */
9392 if (re->req_char < 128)
9394 if (cd->fcc[re->req_char] != re->req_char)
9395 re->flags |= PCRE_RCH_CASELESS;
9397 else if (UCD_OTHERCASE(re->req_char) != re->req_char)
9398 re->flags |= PCRE_RCH_CASELESS;
9402 if (MAX_255(re->req_char) && cd->fcc[re->req_char] != re->req_char)
9403 re->flags |= PCRE_RCH_CASELESS;
9406 re->flags |= PCRE_REQCHSET;
9409 /* Print out the compiled data if debugging is enabled. This is never the
9410 case when building a production library. */
9413 printf("Length = %d top_bracket = %d top_backref = %d\n",
9414 length, re->top_bracket, re->top_backref);
9416 printf("Options=%08x\n", re->options);
9418 if ((re->flags & PCRE_FIRSTSET) != 0)
9420 pcre_uchar ch = re->first_char;
9421 const char *caseless =
9422 ((re->flags & PCRE_FCH_CASELESS) == 0)? "" : " (caseless)";
9423 if (PRINTABLE(ch)) printf("First char = %c%s\n", ch, caseless);
9424 else printf("First char = \\x%02x%s\n", ch, caseless);
9427 if ((re->flags & PCRE_REQCHSET) != 0)
9429 pcre_uchar ch = re->req_char;
9430 const char *caseless =
9431 ((re->flags & PCRE_RCH_CASELESS) == 0)? "" : " (caseless)";
9432 if (PRINTABLE(ch)) printf("Req char = %c%s\n", ch, caseless);
9433 else printf("Req char = \\x%02x%s\n", ch, caseless);
9436 #if defined COMPILE_PCRE8
9437 pcre_printint((pcre *)re, stdout, TRUE);
9438 #elif defined COMPILE_PCRE16
9439 pcre16_printint((pcre *)re, stdout, TRUE);
9440 #elif defined COMPILE_PCRE32
9441 pcre32_printint((pcre *)re, stdout, TRUE);
9444 /* This check is done here in the debugging case so that the code that
9445 was compiled can be seen. */
9447 if (code - codestart > length)
9450 *errorptr = find_error_text(ERR23);
9451 *erroroffset = ptr - (pcre_uchar *)pattern;
9452 if (errorcodeptr != NULL) *errorcodeptr = ERR23;
9455 #endif /* PCRE_DEBUG */
9457 /* Check for a pattern than can match an empty string, so that this information
9458 can be provided to applications. */
9462 if (could_be_empty_branch(codestart, code, utf, cd, NULL))
9464 re->flags |= PCRE_MATCH_EMPTY;
9467 codestart += GET(codestart, 1);
9469 while (*codestart == OP_ALT);
9471 #if defined COMPILE_PCRE8
9473 #elif defined COMPILE_PCRE16
9474 return (pcre16 *)re;
9475 #elif defined COMPILE_PCRE32
9476 return (pcre32 *)re;
9480 /* End of pcre_compile.c */