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) 1997-2016 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 pattern start */
51 #define PSEND end_pattern /* Field containing pattern 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, ESC_a,
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, ESC_a, -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, 0, 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
223 /* We also need a table of characters that may follow \c in an EBCDIC
224 environment for characters 0-31. */
226 static unsigned char ebcdic_escape_c[] = "@ABCDEFGHIJKLMNOPQRSTUVWXYZ[\\]^_";
231 /* Table of special "verbs" like (*PRUNE). This is a short table, so it is
232 searched linearly. Put all the names into a single string, in order to reduce
233 the number of relocations when a shared library is dynamically linked. The
234 string is built from string macros so that it works in UTF-8 mode on EBCDIC
237 typedef struct verbitem {
238 int len; /* Length of verb name */
239 int op; /* Op when no arg, or -1 if arg mandatory */
240 int op_arg; /* Op when arg present, or -1 if not allowed */
243 static const char verbnames[] =
244 "\0" /* Empty name is a shorthand for MARK */
254 static const verbitem verbs[] = {
257 { 6, OP_ACCEPT, -1 },
258 { 6, OP_COMMIT, -1 },
261 { 5, OP_PRUNE, OP_PRUNE_ARG },
262 { 4, OP_SKIP, OP_SKIP_ARG },
263 { 4, OP_THEN, OP_THEN_ARG }
266 static const int verbcount = sizeof(verbs)/sizeof(verbitem);
269 /* Substitutes for [[:<:]] and [[:>:]], which mean start and end of word in
270 another regex library. */
272 static const pcre_uchar sub_start_of_word[] = {
273 CHAR_BACKSLASH, CHAR_b, CHAR_LEFT_PARENTHESIS, CHAR_QUESTION_MARK,
274 CHAR_EQUALS_SIGN, CHAR_BACKSLASH, CHAR_w, CHAR_RIGHT_PARENTHESIS, '\0' };
276 static const pcre_uchar sub_end_of_word[] = {
277 CHAR_BACKSLASH, CHAR_b, CHAR_LEFT_PARENTHESIS, CHAR_QUESTION_MARK,
278 CHAR_LESS_THAN_SIGN, CHAR_EQUALS_SIGN, CHAR_BACKSLASH, CHAR_w,
279 CHAR_RIGHT_PARENTHESIS, '\0' };
282 /* Tables of names of POSIX character classes and their lengths. The names are
283 now all in a single string, to reduce the number of relocations when a shared
284 library is dynamically loaded. The list of lengths is terminated by a zero
285 length entry. The first three must be alpha, lower, upper, as this is assumed
286 for handling case independence. The indices for graph, print, and punct are
287 needed, so identify them. */
289 static const char posix_names[] =
290 STRING_alpha0 STRING_lower0 STRING_upper0 STRING_alnum0
291 STRING_ascii0 STRING_blank0 STRING_cntrl0 STRING_digit0
292 STRING_graph0 STRING_print0 STRING_punct0 STRING_space0
293 STRING_word0 STRING_xdigit;
295 static const pcre_uint8 posix_name_lengths[] = {
296 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 4, 6, 0 };
303 /* Table of class bit maps for each POSIX class. Each class is formed from a
304 base map, with an optional addition or removal of another map. Then, for some
305 classes, there is some additional tweaking: for [:blank:] the vertical space
306 characters are removed, and for [:alpha:] and [:alnum:] the underscore
307 character is removed. The triples in the table consist of the base map offset,
308 second map offset or -1 if no second map, and a non-negative value for map
309 addition or a negative value for map subtraction (if there are two maps). The
310 absolute value of the third field has these meanings: 0 => no tweaking, 1 =>
311 remove vertical space characters, 2 => remove underscore. */
313 static const int posix_class_maps[] = {
314 cbit_word, cbit_digit, -2, /* alpha */
315 cbit_lower, -1, 0, /* lower */
316 cbit_upper, -1, 0, /* upper */
317 cbit_word, -1, 2, /* alnum - word without underscore */
318 cbit_print, cbit_cntrl, 0, /* ascii */
319 cbit_space, -1, 1, /* blank - a GNU extension */
320 cbit_cntrl, -1, 0, /* cntrl */
321 cbit_digit, -1, 0, /* digit */
322 cbit_graph, -1, 0, /* graph */
323 cbit_print, -1, 0, /* print */
324 cbit_punct, -1, 0, /* punct */
325 cbit_space, -1, 0, /* space */
326 cbit_word, -1, 0, /* word - a Perl extension */
327 cbit_xdigit,-1, 0 /* xdigit */
330 /* Table of substitutes for \d etc when PCRE_UCP is set. They are replaced by
331 Unicode property escapes. */
334 static const pcre_uchar string_PNd[] = {
335 CHAR_BACKSLASH, CHAR_P, CHAR_LEFT_CURLY_BRACKET,
336 CHAR_N, CHAR_d, CHAR_RIGHT_CURLY_BRACKET, '\0' };
337 static const pcre_uchar string_pNd[] = {
338 CHAR_BACKSLASH, CHAR_p, CHAR_LEFT_CURLY_BRACKET,
339 CHAR_N, CHAR_d, CHAR_RIGHT_CURLY_BRACKET, '\0' };
340 static const pcre_uchar string_PXsp[] = {
341 CHAR_BACKSLASH, CHAR_P, CHAR_LEFT_CURLY_BRACKET,
342 CHAR_X, CHAR_s, CHAR_p, CHAR_RIGHT_CURLY_BRACKET, '\0' };
343 static const pcre_uchar string_pXsp[] = {
344 CHAR_BACKSLASH, CHAR_p, CHAR_LEFT_CURLY_BRACKET,
345 CHAR_X, CHAR_s, CHAR_p, CHAR_RIGHT_CURLY_BRACKET, '\0' };
346 static const pcre_uchar string_PXwd[] = {
347 CHAR_BACKSLASH, CHAR_P, CHAR_LEFT_CURLY_BRACKET,
348 CHAR_X, CHAR_w, CHAR_d, CHAR_RIGHT_CURLY_BRACKET, '\0' };
349 static const pcre_uchar string_pXwd[] = {
350 CHAR_BACKSLASH, CHAR_p, CHAR_LEFT_CURLY_BRACKET,
351 CHAR_X, CHAR_w, CHAR_d, CHAR_RIGHT_CURLY_BRACKET, '\0' };
353 static const pcre_uchar *substitutes[] = {
356 string_PXsp, /* \S */ /* Xsp is Perl space, but from 8.34, Perl */
357 string_pXsp, /* \s */ /* space and POSIX space are the same. */
358 string_PXwd, /* \W */
362 /* The POSIX class substitutes must be in the order of the POSIX class names,
363 defined above, and there are both positive and negative cases. NULL means no
364 general substitute of a Unicode property escape (\p or \P). However, for some
365 POSIX classes (e.g. graph, print, punct) a special property code is compiled
368 static const pcre_uchar string_pL[] = {
369 CHAR_BACKSLASH, CHAR_p, CHAR_LEFT_CURLY_BRACKET,
370 CHAR_L, CHAR_RIGHT_CURLY_BRACKET, '\0' };
371 static const pcre_uchar string_pLl[] = {
372 CHAR_BACKSLASH, CHAR_p, CHAR_LEFT_CURLY_BRACKET,
373 CHAR_L, CHAR_l, CHAR_RIGHT_CURLY_BRACKET, '\0' };
374 static const pcre_uchar string_pLu[] = {
375 CHAR_BACKSLASH, CHAR_p, CHAR_LEFT_CURLY_BRACKET,
376 CHAR_L, CHAR_u, CHAR_RIGHT_CURLY_BRACKET, '\0' };
377 static const pcre_uchar string_pXan[] = {
378 CHAR_BACKSLASH, CHAR_p, CHAR_LEFT_CURLY_BRACKET,
379 CHAR_X, CHAR_a, CHAR_n, CHAR_RIGHT_CURLY_BRACKET, '\0' };
380 static const pcre_uchar string_h[] = {
381 CHAR_BACKSLASH, CHAR_h, '\0' };
382 static const pcre_uchar string_pXps[] = {
383 CHAR_BACKSLASH, CHAR_p, CHAR_LEFT_CURLY_BRACKET,
384 CHAR_X, CHAR_p, CHAR_s, CHAR_RIGHT_CURLY_BRACKET, '\0' };
385 static const pcre_uchar string_PL[] = {
386 CHAR_BACKSLASH, CHAR_P, CHAR_LEFT_CURLY_BRACKET,
387 CHAR_L, CHAR_RIGHT_CURLY_BRACKET, '\0' };
388 static const pcre_uchar string_PLl[] = {
389 CHAR_BACKSLASH, CHAR_P, CHAR_LEFT_CURLY_BRACKET,
390 CHAR_L, CHAR_l, CHAR_RIGHT_CURLY_BRACKET, '\0' };
391 static const pcre_uchar string_PLu[] = {
392 CHAR_BACKSLASH, CHAR_P, CHAR_LEFT_CURLY_BRACKET,
393 CHAR_L, CHAR_u, CHAR_RIGHT_CURLY_BRACKET, '\0' };
394 static const pcre_uchar string_PXan[] = {
395 CHAR_BACKSLASH, CHAR_P, CHAR_LEFT_CURLY_BRACKET,
396 CHAR_X, CHAR_a, CHAR_n, CHAR_RIGHT_CURLY_BRACKET, '\0' };
397 static const pcre_uchar string_H[] = {
398 CHAR_BACKSLASH, CHAR_H, '\0' };
399 static const pcre_uchar string_PXps[] = {
400 CHAR_BACKSLASH, CHAR_P, CHAR_LEFT_CURLY_BRACKET,
401 CHAR_X, CHAR_p, CHAR_s, CHAR_RIGHT_CURLY_BRACKET, '\0' };
403 static const pcre_uchar *posix_substitutes[] = {
404 string_pL, /* alpha */
405 string_pLl, /* lower */
406 string_pLu, /* upper */
407 string_pXan, /* alnum */
409 string_h, /* blank */
411 string_pNd, /* digit */
415 string_pXps, /* space */ /* Xps is POSIX space, but from 8.34 */
416 string_pXwd, /* word */ /* Perl and POSIX space are the same */
419 string_PL, /* ^alpha */
420 string_PLl, /* ^lower */
421 string_PLu, /* ^upper */
422 string_PXan, /* ^alnum */
424 string_H, /* ^blank */
426 string_PNd, /* ^digit */
430 string_PXps, /* ^space */ /* Xps is POSIX space, but from 8.34 */
431 string_PXwd, /* ^word */ /* Perl and POSIX space are the same */
434 #define POSIX_SUBSIZE (sizeof(posix_substitutes) / sizeof(pcre_uchar *))
437 #define STRING(a) # a
438 #define XSTRING(s) STRING(s)
440 /* The texts of compile-time error messages. These are "char *" because they
441 are passed to the outside world. Do not ever re-use any error number, because
442 they are documented. Always add a new error instead. Messages marked DEAD below
443 are no longer used. This used to be a table of strings, but in order to reduce
444 the number of relocations needed when a shared library is loaded dynamically,
445 it is now one long string. We cannot use a table of offsets, because the
446 lengths of inserts such as XSTRING(MAX_NAME_SIZE) are not known. Instead, we
447 simply count through to the one we want - this isn't a performance issue
448 because these strings are used only when there is a compilation error.
450 Each substring ends with \0 to insert a null character. This includes the final
451 substring, so that the whole string ends with \0\0, which can be detected when
454 static const char error_texts[] =
456 "\\ at end of pattern\0"
457 "\\c at end of pattern\0"
458 "unrecognized character follows \\\0"
459 "numbers out of order in {} quantifier\0"
461 "number too big in {} quantifier\0"
462 "missing terminating ] for character class\0"
463 "invalid escape sequence in character class\0"
464 "range out of order in character class\0"
465 "nothing to repeat\0"
467 "internal error: invalid forward reference offset\0"
468 "internal error: unexpected repeat\0"
469 "unrecognized character after (? or (?-\0"
470 "POSIX named classes are supported only within a class\0"
473 "reference to non-existent subpattern\0"
474 "erroffset passed as NULL\0"
475 "unknown option bit(s) set\0"
476 "missing ) after comment\0"
477 "parentheses nested too deeply\0" /** DEAD **/
479 "regular expression is too large\0"
480 "failed to get memory\0"
481 "unmatched parentheses\0"
482 "internal error: code overflow\0"
483 "unrecognized character after (?<\0"
485 "lookbehind assertion is not fixed length\0"
486 "malformed number or name after (?(\0"
487 "conditional group contains more than two branches\0"
488 "assertion expected after (?( or (?(?C)\0"
489 "(?R or (?[+-]digits must be followed by )\0"
491 "unknown POSIX class name\0"
492 "POSIX collating elements are not supported\0"
493 "this version of PCRE is compiled without UTF support\0"
494 "spare error\0" /** DEAD **/
495 "character value in \\x{} or \\o{} is too large\0"
497 "invalid condition (?(0)\0"
498 "\\C not allowed in lookbehind assertion\0"
499 "PCRE does not support \\L, \\l, \\N{name}, \\U, or \\u\0"
500 "number after (?C is > 255\0"
501 "closing ) for (?C expected\0"
503 "recursive call could loop indefinitely\0"
504 "unrecognized character after (?P\0"
505 "syntax error in subpattern name (missing terminator)\0"
506 "two named subpatterns have the same name\0"
507 "invalid UTF-8 string\0"
509 "support for \\P, \\p, and \\X has not been compiled\0"
510 "malformed \\P or \\p sequence\0"
511 "unknown property name after \\P or \\p\0"
512 "subpattern name is too long (maximum " XSTRING(MAX_NAME_SIZE) " characters)\0"
513 "too many named subpatterns (maximum " XSTRING(MAX_NAME_COUNT) ")\0"
515 "repeated subpattern is too long\0" /** DEAD **/
516 "octal value is greater than \\377 in 8-bit non-UTF-8 mode\0"
517 "internal error: overran compiling workspace\0"
518 "internal error: previously-checked referenced subpattern not found\0"
519 "DEFINE group contains more than one branch\0"
521 "repeating a DEFINE group is not allowed\0" /** DEAD **/
522 "inconsistent NEWLINE options\0"
523 "\\g is not followed by a braced, angle-bracketed, or quoted name/number or by a plain number\0"
524 "a numbered reference must not be zero\0"
525 "an argument is not allowed for (*ACCEPT), (*FAIL), or (*COMMIT)\0"
527 "(*VERB) not recognized or malformed\0"
528 "number is too big\0"
529 "subpattern name expected\0"
530 "digit expected after (?+\0"
531 "] is an invalid data character in JavaScript compatibility mode\0"
533 "different names for subpatterns of the same number are not allowed\0"
534 "(*MARK) must have an argument\0"
535 "this version of PCRE is not compiled with Unicode property support\0"
537 "\\c must be followed by an ASCII character\0"
539 "\\c must be followed by a letter or one of [\\]^_?\0"
541 "\\k is not followed by a braced, angle-bracketed, or quoted name\0"
543 "internal error: unknown opcode in find_fixedlength()\0"
544 "\\N is not supported in a class\0"
545 "too many forward references\0"
546 "disallowed Unicode code point (>= 0xd800 && <= 0xdfff)\0"
547 "invalid UTF-16 string\0"
549 "name is too long in (*MARK), (*PRUNE), (*SKIP), or (*THEN)\0"
550 "character value in \\u.... sequence is too large\0"
551 "invalid UTF-32 string\0"
552 "setting UTF is disabled by the application\0"
553 "non-hex character in \\x{} (closing brace missing?)\0"
555 "non-octal character in \\o{} (closing brace missing?)\0"
556 "missing opening brace after \\o\0"
557 "parentheses are too deeply nested\0"
558 "invalid range in character class\0"
559 "group name must start with a non-digit\0"
561 "parentheses are too deeply nested (stack check)\0"
562 "digits missing in \\x{} or \\o{}\0"
563 "regular expression is too complicated\0"
566 /* Table to identify digits and hex digits. This is used when compiling
567 patterns. Note that the tables in chartables are dependent on the locale, and
568 may mark arbitrary characters as digits - but the PCRE compiling code expects
569 to handle only 0-9, a-z, and A-Z as digits when compiling. That is why we have
570 a private table here. It costs 256 bytes, but it is a lot faster than doing
571 character value tests (at least in some simple cases I timed), and in some
572 applications one wants PCRE to compile efficiently as well as match
575 For convenience, we use the same bit definitions as in chartables:
578 0x08 hexadecimal digit
580 Then we can use ctype_digit and ctype_xdigit in the code. */
582 /* Using a simple comparison for decimal numbers rather than a memory read
583 is much faster, and the resulting code is simpler (the compiler turns it
584 into a subtraction and unsigned comparison). */
586 #define IS_DIGIT(x) ((x) >= CHAR_0 && (x) <= CHAR_9)
590 /* This is the "normal" case, for ASCII systems, and EBCDIC systems running in
593 static const pcre_uint8 digitab[] =
595 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 0- 7 */
596 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 8- 15 */
597 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 16- 23 */
598 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 24- 31 */
599 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - ' */
600 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* ( - / */
601 0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c, /* 0 - 7 */
602 0x0c,0x0c,0x00,0x00,0x00,0x00,0x00,0x00, /* 8 - ? */
603 0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* @ - G */
604 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* H - O */
605 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* P - W */
606 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* X - _ */
607 0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* ` - g */
608 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* h - o */
609 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* p - w */
610 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* x -127 */
611 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 128-135 */
612 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 136-143 */
613 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 144-151 */
614 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 152-159 */
615 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 160-167 */
616 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 168-175 */
617 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 176-183 */
618 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 184-191 */
619 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 192-199 */
620 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 200-207 */
621 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 208-215 */
622 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 216-223 */
623 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 224-231 */
624 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 232-239 */
625 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 240-247 */
626 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00};/* 248-255 */
630 /* This is the "abnormal" case, for EBCDIC systems not running in UTF-8 mode. */
632 static const pcre_uint8 digitab[] =
634 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 0- 7 0 */
635 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 8- 15 */
636 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 16- 23 10 */
637 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 24- 31 */
638 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 32- 39 20 */
639 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 40- 47 */
640 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 48- 55 30 */
641 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 56- 63 */
642 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - 71 40 */
643 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 72- | */
644 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* & - 87 50 */
645 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 88- 95 */
646 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - -103 60 */
647 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 104- ? */
648 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 112-119 70 */
649 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 120- " */
650 0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* 128- g 80 */
651 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* h -143 */
652 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 144- p 90 */
653 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* q -159 */
654 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 160- x A0 */
655 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* y -175 */
656 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* ^ -183 B0 */
657 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 184-191 */
658 0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* { - G C0 */
659 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* H -207 */
660 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* } - P D0 */
661 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* Q -223 */
662 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* \ - X E0 */
663 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* Y -239 */
664 0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c, /* 0 - 7 F0 */
665 0x0c,0x0c,0x00,0x00,0x00,0x00,0x00,0x00};/* 8 -255 */
667 static const pcre_uint8 ebcdic_chartab[] = { /* chartable partial dup */
668 0x80,0x00,0x00,0x00,0x00,0x01,0x00,0x00, /* 0- 7 */
669 0x00,0x00,0x00,0x00,0x01,0x01,0x00,0x00, /* 8- 15 */
670 0x00,0x00,0x00,0x00,0x00,0x01,0x00,0x00, /* 16- 23 */
671 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 24- 31 */
672 0x00,0x00,0x00,0x00,0x00,0x01,0x00,0x00, /* 32- 39 */
673 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 40- 47 */
674 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 48- 55 */
675 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 56- 63 */
676 0x01,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - 71 */
677 0x00,0x00,0x00,0x80,0x00,0x80,0x80,0x80, /* 72- | */
678 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* & - 87 */
679 0x00,0x00,0x00,0x80,0x80,0x80,0x00,0x00, /* 88- 95 */
680 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - -103 */
681 0x00,0x00,0x00,0x00,0x00,0x10,0x00,0x80, /* 104- ? */
682 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 112-119 */
683 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 120- " */
684 0x00,0x1a,0x1a,0x1a,0x1a,0x1a,0x1a,0x12, /* 128- g */
685 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* h -143 */
686 0x00,0x12,0x12,0x12,0x12,0x12,0x12,0x12, /* 144- p */
687 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* q -159 */
688 0x00,0x00,0x12,0x12,0x12,0x12,0x12,0x12, /* 160- x */
689 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* y -175 */
690 0x80,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* ^ -183 */
691 0x00,0x00,0x80,0x00,0x00,0x00,0x00,0x00, /* 184-191 */
692 0x80,0x1a,0x1a,0x1a,0x1a,0x1a,0x1a,0x12, /* { - G */
693 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* H -207 */
694 0x00,0x12,0x12,0x12,0x12,0x12,0x12,0x12, /* } - P */
695 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* Q -223 */
696 0x00,0x00,0x12,0x12,0x12,0x12,0x12,0x12, /* \ - X */
697 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* Y -239 */
698 0x1c,0x1c,0x1c,0x1c,0x1c,0x1c,0x1c,0x1c, /* 0 - 7 */
699 0x1c,0x1c,0x00,0x00,0x00,0x00,0x00,0x00};/* 8 -255 */
703 /* This table is used to check whether auto-possessification is possible
704 between adjacent character-type opcodes. The left-hand (repeated) opcode is
705 used to select the row, and the right-hand opcode is use to select the column.
706 A value of 1 means that auto-possessification is OK. For example, the second
707 value in the first row means that \D+\d can be turned into \D++\d.
709 The Unicode property types (\P and \p) have to be present to fill out the table
710 because of what their opcode values are, but the table values should always be
711 zero because property types are handled separately in the code. The last four
712 columns apply to items that cannot be repeated, so there is no need to have
713 rows for them. Note that OP_DIGIT etc. are generated only when PCRE_UCP is
714 *not* set. When it is set, \d etc. are converted into OP_(NOT_)PROP codes. */
716 #define APTROWS (LAST_AUTOTAB_LEFT_OP - FIRST_AUTOTAB_OP + 1)
717 #define APTCOLS (LAST_AUTOTAB_RIGHT_OP - FIRST_AUTOTAB_OP + 1)
719 static const pcre_uint8 autoposstab[APTROWS][APTCOLS] = {
720 /* \D \d \S \s \W \w . .+ \C \P \p \R \H \h \V \v \X \Z \z $ $M */
721 { 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0 }, /* \D */
722 { 1, 0, 0, 1, 1, 0, 0, 0, 0, 0, 0, 1, 0, 1, 0, 1, 0, 1, 1, 1, 1 }, /* \d */
723 { 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 1, 0, 1, 0, 1, 0, 1, 1, 1, 1 }, /* \S */
724 { 0, 1, 1, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0 }, /* \s */
725 { 0, 1, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0 }, /* \W */
726 { 0, 0, 0, 1, 1, 0, 0, 0, 0, 0, 0, 1, 0, 1, 0, 1, 0, 1, 1, 1, 1 }, /* \w */
727 { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 1, 0, 0 }, /* . */
728 { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0 }, /* .+ */
729 { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0 }, /* \C */
730 { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 }, /* \P */
731 { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 }, /* \p */
732 { 0, 1, 0, 1, 0, 1, 1, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 1, 0, 0 }, /* \R */
733 { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 1, 0, 0 }, /* \H */
734 { 0, 1, 1, 0, 0, 1, 0, 0, 0, 0, 0, 1, 1, 0, 0, 1, 0, 0, 1, 0, 0 }, /* \h */
735 { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 1, 0, 0, 1, 0, 0 }, /* \V */
736 { 0, 1, 1, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 1, 1, 0, 0, 0, 1, 0, 0 }, /* \v */
737 { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0 } /* \X */
741 /* This table is used to check whether auto-possessification is possible
742 between adjacent Unicode property opcodes (OP_PROP and OP_NOTPROP). The
743 left-hand (repeated) opcode is used to select the row, and the right-hand
744 opcode is used to select the column. The values are as follows:
746 0 Always return FALSE (never auto-possessify)
747 1 Character groups are distinct (possessify if both are OP_PROP)
748 2 Check character categories in the same group (general or particular)
749 3 TRUE if the two opcodes are not the same (PROP vs NOTPROP)
751 4 Check left general category vs right particular category
752 5 Check right general category vs left particular category
754 6 Left alphanum vs right general category
755 7 Left space vs right general category
756 8 Left word vs right general category
758 9 Right alphanum vs left general category
759 10 Right space vs left general category
760 11 Right word vs left general category
762 12 Left alphanum vs right particular category
763 13 Left space vs right particular category
764 14 Left word vs right particular category
766 15 Right alphanum vs left particular category
767 16 Right space vs left particular category
768 17 Right word vs left particular category
771 static const pcre_uint8 propposstab[PT_TABSIZE][PT_TABSIZE] = {
772 /* ANY LAMP GC PC SC ALNUM SPACE PXSPACE WORD CLIST UCNC */
773 { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 }, /* PT_ANY */
774 { 0, 3, 0, 0, 0, 3, 1, 1, 0, 0, 0 }, /* PT_LAMP */
775 { 0, 0, 2, 4, 0, 9, 10, 10, 11, 0, 0 }, /* PT_GC */
776 { 0, 0, 5, 2, 0, 15, 16, 16, 17, 0, 0 }, /* PT_PC */
777 { 0, 0, 0, 0, 2, 0, 0, 0, 0, 0, 0 }, /* PT_SC */
778 { 0, 3, 6, 12, 0, 3, 1, 1, 0, 0, 0 }, /* PT_ALNUM */
779 { 0, 1, 7, 13, 0, 1, 3, 3, 1, 0, 0 }, /* PT_SPACE */
780 { 0, 1, 7, 13, 0, 1, 3, 3, 1, 0, 0 }, /* PT_PXSPACE */
781 { 0, 0, 8, 14, 0, 0, 1, 1, 3, 0, 0 }, /* PT_WORD */
782 { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 }, /* PT_CLIST */
783 { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 3 } /* PT_UCNC */
786 /* This table is used to check whether auto-possessification is possible
787 between adjacent Unicode property opcodes (OP_PROP and OP_NOTPROP) when one
788 specifies a general category and the other specifies a particular category. The
789 row is selected by the general category and the column by the particular
790 category. The value is 1 if the particular category is not part of the general
793 static const pcre_uint8 catposstab[7][30] = {
794 /* 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 */
795 { 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 */
796 { 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 */
797 { 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 */
798 { 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 */
799 { 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 */
800 { 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 */
801 { 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 */
804 /* This table is used when checking ALNUM, (PX)SPACE, SPACE, and WORD against
805 a general or particular category. The properties in each row are those
806 that apply to the character set in question. Duplication means that a little
807 unnecessary work is done when checking, but this keeps things much simpler
808 because they can all use the same code. For more details see the comment where
811 Note: SPACE and PXSPACE used to be different because Perl excluded VT from
812 "space", but from Perl 5.18 it's included, so both categories are treated the
815 static const pcre_uint8 posspropstab[3][4] = {
816 { ucp_L, ucp_N, ucp_N, ucp_Nl }, /* ALNUM, 3rd and 4th values redundant */
817 { ucp_Z, ucp_Z, ucp_C, ucp_Cc }, /* SPACE and PXSPACE, 2nd value redundant */
818 { ucp_L, ucp_N, ucp_P, ucp_Po } /* WORD */
821 /* This table is used when converting repeating opcodes into possessified
822 versions as a result of an explicit possessive quantifier such as ++. A zero
823 value means there is no possessified version - in those cases the item in
824 question must be wrapped in ONCE brackets. The table is truncated at OP_CALLOUT
825 because all relevant opcodes are less than that. */
827 static const pcre_uint8 opcode_possessify[] = {
828 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, /* 0 - 15 */
829 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, /* 16 - 31 */
832 OP_POSSTAR, 0, /* STAR, MINSTAR */
833 OP_POSPLUS, 0, /* PLUS, MINPLUS */
834 OP_POSQUERY, 0, /* QUERY, MINQUERY */
835 OP_POSUPTO, 0, /* UPTO, MINUPTO */
837 0, 0, 0, 0, /* POS{STAR,PLUS,QUERY,UPTO} */
839 OP_POSSTARI, 0, /* STARI, MINSTARI */
840 OP_POSPLUSI, 0, /* PLUSI, MINPLUSI */
841 OP_POSQUERYI, 0, /* QUERYI, MINQUERYI */
842 OP_POSUPTOI, 0, /* UPTOI, MINUPTOI */
844 0, 0, 0, 0, /* POS{STARI,PLUSI,QUERYI,UPTOI} */
846 OP_NOTPOSSTAR, 0, /* NOTSTAR, NOTMINSTAR */
847 OP_NOTPOSPLUS, 0, /* NOTPLUS, NOTMINPLUS */
848 OP_NOTPOSQUERY, 0, /* NOTQUERY, NOTMINQUERY */
849 OP_NOTPOSUPTO, 0, /* NOTUPTO, NOTMINUPTO */
851 0, 0, 0, 0, /* NOTPOS{STAR,PLUS,QUERY,UPTO} */
853 OP_NOTPOSSTARI, 0, /* NOTSTARI, NOTMINSTARI */
854 OP_NOTPOSPLUSI, 0, /* NOTPLUSI, NOTMINPLUSI */
855 OP_NOTPOSQUERYI, 0, /* NOTQUERYI, NOTMINQUERYI */
856 OP_NOTPOSUPTOI, 0, /* NOTUPTOI, NOTMINUPTOI */
858 0, 0, 0, 0, /* NOTPOS{STARI,PLUSI,QUERYI,UPTOI} */
860 OP_TYPEPOSSTAR, 0, /* TYPESTAR, TYPEMINSTAR */
861 OP_TYPEPOSPLUS, 0, /* TYPEPLUS, TYPEMINPLUS */
862 OP_TYPEPOSQUERY, 0, /* TYPEQUERY, TYPEMINQUERY */
863 OP_TYPEPOSUPTO, 0, /* TYPEUPTO, TYPEMINUPTO */
865 0, 0, 0, 0, /* TYPEPOS{STAR,PLUS,QUERY,UPTO} */
867 OP_CRPOSSTAR, 0, /* CRSTAR, CRMINSTAR */
868 OP_CRPOSPLUS, 0, /* CRPLUS, CRMINPLUS */
869 OP_CRPOSQUERY, 0, /* CRQUERY, CRMINQUERY */
870 OP_CRPOSRANGE, 0, /* CRRANGE, CRMINRANGE */
871 0, 0, 0, 0, /* CRPOS{STAR,PLUS,QUERY,RANGE} */
873 0, 0, 0, /* CLASS, NCLASS, XCLASS */
874 0, 0, /* REF, REFI */
875 0, 0, /* DNREF, DNREFI */
876 0, 0 /* RECURSE, CALLOUT */
881 /*************************************************
882 * Find an error text *
883 *************************************************/
885 /* The error texts are now all in one long string, to save on relocations. As
886 some of the text is of unknown length, we can't use a table of offsets.
887 Instead, just count through the strings. This is not a performance issue
888 because it happens only when there has been a compilation error.
890 Argument: the error number
891 Returns: pointer to the error string
895 find_error_text(int n)
897 const char *s = error_texts;
900 while (*s++ != CHAR_NULL) {};
901 if (*s == CHAR_NULL) return "Error text not found (please report)";
908 /*************************************************
909 * Expand the workspace *
910 *************************************************/
912 /* This function is called during the second compiling phase, if the number of
913 forward references fills the existing workspace, which is originally a block on
914 the stack. A larger block is obtained from malloc() unless the ultimate limit
915 has been reached or the increase will be rather small.
917 Argument: pointer to the compile data block
918 Returns: 0 if all went well, else an error number
922 expand_workspace(compile_data *cd)
924 pcre_uchar *newspace;
925 int newsize = cd->workspace_size * 2;
927 if (newsize > COMPILE_WORK_SIZE_MAX) newsize = COMPILE_WORK_SIZE_MAX;
928 if (cd->workspace_size >= COMPILE_WORK_SIZE_MAX ||
929 newsize - cd->workspace_size < WORK_SIZE_SAFETY_MARGIN)
932 newspace = (PUBL(malloc))(IN_UCHARS(newsize));
933 if (newspace == NULL) return ERR21;
934 memcpy(newspace, cd->start_workspace, cd->workspace_size * sizeof(pcre_uchar));
935 cd->hwm = (pcre_uchar *)newspace + (cd->hwm - cd->start_workspace);
936 if (cd->workspace_size > COMPILE_WORK_SIZE)
937 (PUBL(free))((void *)cd->start_workspace);
938 cd->start_workspace = newspace;
939 cd->workspace_size = newsize;
945 /*************************************************
946 * Check for counted repeat *
947 *************************************************/
949 /* This function is called when a '{' is encountered in a place where it might
950 start a quantifier. It looks ahead to see if it really is a quantifier or not.
951 It is only a quantifier if it is one of the forms {ddd} {ddd,} or {ddd,ddd}
952 where the ddds are digits.
955 p pointer to the first char after '{'
957 Returns: TRUE or FALSE
961 is_counted_repeat(const pcre_uchar *p)
963 if (!IS_DIGIT(*p)) return FALSE;
965 while (IS_DIGIT(*p)) p++;
966 if (*p == CHAR_RIGHT_CURLY_BRACKET) return TRUE;
968 if (*p++ != CHAR_COMMA) return FALSE;
969 if (*p == CHAR_RIGHT_CURLY_BRACKET) return TRUE;
971 if (!IS_DIGIT(*p)) return FALSE;
973 while (IS_DIGIT(*p)) p++;
975 return (*p == CHAR_RIGHT_CURLY_BRACKET);
980 /*************************************************
982 *************************************************/
984 /* This function is called when a \ has been encountered. It either returns a
985 positive value for a simple escape such as \n, or 0 for a data character which
986 will be placed in chptr. A backreference to group n is returned as negative n.
987 When UTF-8 is enabled, a positive value greater than 255 may be returned in
988 chptr. On entry, ptr is pointing at the \. On exit, it is on the final
989 character of the escape sequence.
992 ptrptr points to the pattern position pointer
993 chptr points to a returned data character
994 errorcodeptr points to the errorcode variable
995 bracount number of previous extracting brackets
996 options the options bits
997 isclass TRUE if inside a character class
999 Returns: zero => a data character
1000 positive => a special escape sequence
1001 negative => a back reference
1002 on error, errorcodeptr is set
1006 check_escape(const pcre_uchar **ptrptr, pcre_uint32 *chptr, int *errorcodeptr,
1007 int bracount, int options, BOOL isclass)
1009 /* PCRE_UTF16 has the same value as PCRE_UTF8. */
1010 BOOL utf = (options & PCRE_UTF8) != 0;
1011 const pcre_uchar *ptr = *ptrptr + 1;
1016 GETCHARINCTEST(c, ptr); /* Get character value, increment pointer */
1017 ptr--; /* Set pointer back to the last byte */
1019 /* If backslash is at the end of the pattern, it's an error. */
1021 if (c == CHAR_NULL) *errorcodeptr = ERR1;
1023 /* Non-alphanumerics are literals. For digits or letters, do an initial lookup
1024 in a table. A non-zero result is something that can be returned immediately.
1025 Otherwise further processing may be required. */
1027 #ifndef EBCDIC /* ASCII/UTF-8 coding */
1028 /* Not alphanumeric */
1029 else if (c < CHAR_0 || c > CHAR_z) {}
1030 else if ((i = escapes[c - CHAR_0]) != 0)
1031 { if (i > 0) c = (pcre_uint32)i; else escape = -i; }
1033 #else /* EBCDIC coding */
1034 /* Not alphanumeric */
1035 else if (c < CHAR_a || (!MAX_255(c) || (ebcdic_chartab[c] & 0x0E) == 0)) {}
1036 else if ((i = escapes[c - 0x48]) != 0) { if (i > 0) c = (pcre_uint32)i; else escape = -i; }
1039 /* Escapes that need further processing, or are illegal. */
1043 const pcre_uchar *oldptr;
1044 BOOL braced, negated, overflow;
1049 /* A number of Perl escapes are not handled by PCRE. We give an explicit
1054 *errorcodeptr = ERR37;
1058 if ((options & PCRE_JAVASCRIPT_COMPAT) != 0)
1060 /* In JavaScript, \u must be followed by four hexadecimal numbers.
1061 Otherwise it is a lowercase u letter. */
1062 if (MAX_255(ptr[1]) && (digitab[ptr[1]] & ctype_xdigit) != 0
1063 && MAX_255(ptr[2]) && (digitab[ptr[2]] & ctype_xdigit) != 0
1064 && MAX_255(ptr[3]) && (digitab[ptr[3]] & ctype_xdigit) != 0
1065 && MAX_255(ptr[4]) && (digitab[ptr[4]] & ctype_xdigit) != 0)
1068 for (i = 0; i < 4; ++i)
1070 register pcre_uint32 cc = *(++ptr);
1071 #ifndef EBCDIC /* ASCII/UTF-8 coding */
1072 if (cc >= CHAR_a) cc -= 32; /* Convert to upper case */
1073 c = (c << 4) + cc - ((cc < CHAR_A)? CHAR_0 : (CHAR_A - 10));
1074 #else /* EBCDIC coding */
1075 if (cc >= CHAR_a && cc <= CHAR_z) cc += 64; /* Convert to upper case */
1076 c = (c << 4) + cc - ((cc >= CHAR_0)? CHAR_0 : (CHAR_A - 10));
1080 #if defined COMPILE_PCRE8
1081 if (c > (utf ? 0x10ffffU : 0xffU))
1082 #elif defined COMPILE_PCRE16
1083 if (c > (utf ? 0x10ffffU : 0xffffU))
1084 #elif defined COMPILE_PCRE32
1085 if (utf && c > 0x10ffffU)
1088 *errorcodeptr = ERR76;
1090 else if (utf && c >= 0xd800 && c <= 0xdfff) *errorcodeptr = ERR73;
1094 *errorcodeptr = ERR37;
1098 /* In JavaScript, \U is an uppercase U letter. */
1099 if ((options & PCRE_JAVASCRIPT_COMPAT) == 0) *errorcodeptr = ERR37;
1102 /* In a character class, \g is just a literal "g". Outside a character
1103 class, \g must be followed by one of a number of specific things:
1105 (1) A number, either plain or braced. If positive, it is an absolute
1106 backreference. If negative, it is a relative backreference. This is a Perl
1109 (2) Perl 5.10 also supports \g{name} as a reference to a named group. This
1110 is part of Perl's movement towards a unified syntax for back references. As
1111 this is synonymous with \k{name}, we fudge it up by pretending it really
1114 (3) For Oniguruma compatibility we also support \g followed by a name or a
1115 number either in angle brackets or in single quotes. However, these are
1116 (possibly recursive) subroutine calls, _not_ backreferences. Just return
1117 the ESC_g code (cf \k). */
1121 if (ptr[1] == CHAR_LESS_THAN_SIGN || ptr[1] == CHAR_APOSTROPHE)
1127 /* Handle the Perl-compatible cases */
1129 if (ptr[1] == CHAR_LEFT_CURLY_BRACKET)
1131 const pcre_uchar *p;
1132 for (p = ptr+2; *p != CHAR_NULL && *p != CHAR_RIGHT_CURLY_BRACKET; p++)
1133 if (*p != CHAR_MINUS && !IS_DIGIT(*p)) break;
1134 if (*p != CHAR_NULL && *p != CHAR_RIGHT_CURLY_BRACKET)
1142 else braced = FALSE;
1144 if (ptr[1] == CHAR_MINUS)
1149 else negated = FALSE;
1151 /* The integer range is limited by the machine's int representation. */
1154 while (IS_DIGIT(ptr[1]))
1156 if (s > INT_MAX / 10 - 1) /* Integer overflow */
1161 s = s * 10 + (int)(*(++ptr) - CHAR_0);
1163 if (overflow) /* Integer overflow */
1165 while (IS_DIGIT(ptr[1]))
1167 *errorcodeptr = ERR61;
1171 if (braced && *(++ptr) != CHAR_RIGHT_CURLY_BRACKET)
1173 *errorcodeptr = ERR57;
1179 *errorcodeptr = ERR58;
1187 *errorcodeptr = ERR15;
1190 s = bracount - (s - 1);
1196 /* The handling of escape sequences consisting of a string of digits
1197 starting with one that is not zero is not straightforward. Perl has changed
1198 over the years. Nowadays \g{} for backreferences and \o{} for octal are
1199 recommended to avoid the ambiguities in the old syntax.
1201 Outside a character class, the digits are read as a decimal number. If the
1202 number is less than 8 (used to be 10), or if there are that many previous
1203 extracting left brackets, then it is a back reference. Otherwise, up to
1204 three octal digits are read to form an escaped byte. Thus \123 is likely to
1205 be octal 123 (cf \0123, which is octal 012 followed by the literal 3). If
1206 the octal value is greater than 377, the least significant 8 bits are
1207 taken. \8 and \9 are treated as the literal characters 8 and 9.
1209 Inside a character class, \ followed by a digit is always either a literal
1210 8 or 9 or an octal number. */
1212 case CHAR_1: case CHAR_2: case CHAR_3: case CHAR_4: case CHAR_5:
1213 case CHAR_6: case CHAR_7: case CHAR_8: case CHAR_9:
1218 /* The integer range is limited by the machine's int representation. */
1219 s = (int)(c -CHAR_0);
1221 while (IS_DIGIT(ptr[1]))
1223 if (s > INT_MAX / 10 - 1) /* Integer overflow */
1228 s = s * 10 + (int)(*(++ptr) - CHAR_0);
1230 if (overflow) /* Integer overflow */
1232 while (IS_DIGIT(ptr[1]))
1234 *errorcodeptr = ERR61;
1237 if (s < 8 || s <= bracount) /* Check for back reference */
1242 ptr = oldptr; /* Put the pointer back and fall through */
1245 /* Handle a digit following \ when the number is not a back reference. If
1246 the first digit is 8 or 9, Perl used to generate a binary zero byte and
1247 then treat the digit as a following literal. At least by Perl 5.18 this
1248 changed so as not to insert the binary zero. */
1250 if ((c = *ptr) >= CHAR_8) break;
1252 /* Fall through with a digit less than 8 */
1254 /* \0 always starts an octal number, but we may drop through to here with a
1255 larger first octal digit. The original code used just to take the least
1256 significant 8 bits of octal numbers (I think this is what early Perls used
1257 to do). Nowadays we allow for larger numbers in UTF-8 mode and 16-bit mode,
1258 but no more than 3 octal digits. */
1262 while(i++ < 2 && ptr[1] >= CHAR_0 && ptr[1] <= CHAR_7)
1263 c = c * 8 + *(++ptr) - CHAR_0;
1264 #ifdef COMPILE_PCRE8
1265 if (!utf && c > 0xff) *errorcodeptr = ERR51;
1269 /* \o is a relatively new Perl feature, supporting a more general way of
1270 specifying character codes in octal. The only supported form is \o{ddd}. */
1273 if (ptr[1] != CHAR_LEFT_CURLY_BRACKET) *errorcodeptr = ERR81; else
1274 if (ptr[2] == CHAR_RIGHT_CURLY_BRACKET) *errorcodeptr = ERR86; else
1279 while (*ptr >= CHAR_0 && *ptr <= CHAR_7)
1281 register pcre_uint32 cc = *ptr++;
1282 if (c == 0 && cc == CHAR_0) continue; /* Leading zeroes */
1283 #ifdef COMPILE_PCRE32
1284 if (c >= 0x20000000l) { overflow = TRUE; break; }
1286 c = (c << 3) + cc - CHAR_0 ;
1287 #if defined COMPILE_PCRE8
1288 if (c > (utf ? 0x10ffffU : 0xffU)) { overflow = TRUE; break; }
1289 #elif defined COMPILE_PCRE16
1290 if (c > (utf ? 0x10ffffU : 0xffffU)) { overflow = TRUE; break; }
1291 #elif defined COMPILE_PCRE32
1292 if (utf && c > 0x10ffffU) { overflow = TRUE; break; }
1297 while (*ptr >= CHAR_0 && *ptr <= CHAR_7) ptr++;
1298 *errorcodeptr = ERR34;
1300 else if (*ptr == CHAR_RIGHT_CURLY_BRACKET)
1302 if (utf && c >= 0xd800 && c <= 0xdfff) *errorcodeptr = ERR73;
1304 else *errorcodeptr = ERR80;
1308 /* \x is complicated. In JavaScript, \x must be followed by two hexadecimal
1309 numbers. Otherwise it is a lowercase x letter. */
1312 if ((options & PCRE_JAVASCRIPT_COMPAT) != 0)
1314 if (MAX_255(ptr[1]) && (digitab[ptr[1]] & ctype_xdigit) != 0
1315 && MAX_255(ptr[2]) && (digitab[ptr[2]] & ctype_xdigit) != 0)
1318 for (i = 0; i < 2; ++i)
1320 register pcre_uint32 cc = *(++ptr);
1321 #ifndef EBCDIC /* ASCII/UTF-8 coding */
1322 if (cc >= CHAR_a) cc -= 32; /* Convert to upper case */
1323 c = (c << 4) + cc - ((cc < CHAR_A)? CHAR_0 : (CHAR_A - 10));
1324 #else /* EBCDIC coding */
1325 if (cc >= CHAR_a && cc <= CHAR_z) cc += 64; /* Convert to upper case */
1326 c = (c << 4) + cc - ((cc >= CHAR_0)? CHAR_0 : (CHAR_A - 10));
1330 } /* End JavaScript handling */
1332 /* Handle \x in Perl's style. \x{ddd} is a character number which can be
1333 greater than 0xff in utf or non-8bit mode, but only if the ddd are hex
1334 digits. If not, { used to be treated as a data character. However, Perl
1335 seems to read hex digits up to the first non-such, and ignore the rest, so
1336 that, for example \x{zz} matches a binary zero. This seems crazy, so PCRE
1337 now gives an error. */
1341 if (ptr[1] == CHAR_LEFT_CURLY_BRACKET)
1344 if (*ptr == CHAR_RIGHT_CURLY_BRACKET)
1346 *errorcodeptr = ERR86;
1351 while (MAX_255(*ptr) && (digitab[*ptr] & ctype_xdigit) != 0)
1353 register pcre_uint32 cc = *ptr++;
1354 if (c == 0 && cc == CHAR_0) continue; /* Leading zeroes */
1356 #ifdef COMPILE_PCRE32
1357 if (c >= 0x10000000l) { overflow = TRUE; break; }
1360 #ifndef EBCDIC /* ASCII/UTF-8 coding */
1361 if (cc >= CHAR_a) cc -= 32; /* Convert to upper case */
1362 c = (c << 4) + cc - ((cc < CHAR_A)? CHAR_0 : (CHAR_A - 10));
1363 #else /* EBCDIC coding */
1364 if (cc >= CHAR_a && cc <= CHAR_z) cc += 64; /* Convert to upper case */
1365 c = (c << 4) + cc - ((cc >= CHAR_0)? CHAR_0 : (CHAR_A - 10));
1368 #if defined COMPILE_PCRE8
1369 if (c > (utf ? 0x10ffffU : 0xffU)) { overflow = TRUE; break; }
1370 #elif defined COMPILE_PCRE16
1371 if (c > (utf ? 0x10ffffU : 0xffffU)) { overflow = TRUE; break; }
1372 #elif defined COMPILE_PCRE32
1373 if (utf && c > 0x10ffffU) { overflow = TRUE; break; }
1379 while (MAX_255(*ptr) && (digitab[*ptr] & ctype_xdigit) != 0) ptr++;
1380 *errorcodeptr = ERR34;
1383 else if (*ptr == CHAR_RIGHT_CURLY_BRACKET)
1385 if (utf && c >= 0xd800 && c <= 0xdfff) *errorcodeptr = ERR73;
1388 /* If the sequence of hex digits does not end with '}', give an error.
1389 We used just to recognize this construct and fall through to the normal
1390 \x handling, but nowadays Perl gives an error, which seems much more
1391 sensible, so we do too. */
1393 else *errorcodeptr = ERR79;
1394 } /* End of \x{} processing */
1396 /* Read a single-byte hex-defined char (up to two hex digits after \x) */
1401 while (i++ < 2 && MAX_255(ptr[1]) && (digitab[ptr[1]] & ctype_xdigit) != 0)
1403 pcre_uint32 cc; /* Some compilers don't like */
1404 cc = *(++ptr); /* ++ in initializers */
1405 #ifndef EBCDIC /* ASCII/UTF-8 coding */
1406 if (cc >= CHAR_a) cc -= 32; /* Convert to upper case */
1407 c = c * 16 + cc - ((cc < CHAR_A)? CHAR_0 : (CHAR_A - 10));
1408 #else /* EBCDIC coding */
1409 if (cc <= CHAR_z) cc += 64; /* Convert to upper case */
1410 c = c * 16 + cc - ((cc >= CHAR_0)? CHAR_0 : (CHAR_A - 10));
1413 } /* End of \xdd handling */
1414 } /* End of Perl-style \x handling */
1417 /* For \c, a following letter is upper-cased; then the 0x40 bit is flipped.
1418 An error is given if the byte following \c is not an ASCII character. This
1419 coding is ASCII-specific, but then the whole concept of \cx is
1420 ASCII-specific. (However, an EBCDIC equivalent has now been added.) */
1426 *errorcodeptr = ERR2;
1429 #ifndef EBCDIC /* ASCII/UTF-8 coding */
1430 if (c > 127) /* Excludes all non-ASCII in either mode */
1432 *errorcodeptr = ERR68;
1435 if (c >= CHAR_a && c <= CHAR_z) c -= 32;
1437 #else /* EBCDIC coding */
1438 if (c >= CHAR_a && c <= CHAR_z) c += 64;
1439 if (c == CHAR_QUESTION_MARK)
1440 c = ('\\' == 188 && '`' == 74)? 0x5f : 0xff;
1443 for (i = 0; i < 32; i++)
1445 if (c == ebcdic_escape_c[i]) break;
1447 if (i < 32) c = i; else *errorcodeptr = ERR68;
1452 /* PCRE_EXTRA enables extensions to Perl in the matter of escapes. Any
1453 other alphanumeric following \ is an error if PCRE_EXTRA was set;
1454 otherwise, for Perl compatibility, it is a literal. This code looks a bit
1455 odd, but there used to be some cases other than the default, and there may
1456 be again in future, so I haven't "optimized" it. */
1459 if ((options & PCRE_EXTRA) != 0) switch(c)
1462 *errorcodeptr = ERR3;
1469 /* Perl supports \N{name} for character names, as well as plain \N for "not
1470 newline". PCRE does not support \N{name}. However, it does support
1471 quantification such as \N{2,3}. */
1473 if (escape == ESC_N && ptr[1] == CHAR_LEFT_CURLY_BRACKET &&
1474 !is_counted_repeat(ptr+2))
1475 *errorcodeptr = ERR37;
1477 /* If PCRE_UCP is set, we change the values for \d etc. */
1479 if ((options & PCRE_UCP) != 0 && escape >= ESC_D && escape <= ESC_w)
1480 escape += (ESC_DU - ESC_D);
1482 /* Set the pointer to the final character before returning. */
1492 /*************************************************
1493 * Handle \P and \p *
1494 *************************************************/
1496 /* This function is called after \P or \p has been encountered, provided that
1497 PCRE is compiled with support for Unicode properties. On entry, ptrptr is
1498 pointing at the P or p. On exit, it is pointing at the final character of the
1502 ptrptr points to the pattern position pointer
1503 negptr points to a boolean that is set TRUE for negation else FALSE
1504 ptypeptr points to an unsigned int that is set to the type value
1505 pdataptr points to an unsigned int that is set to the detailed property value
1506 errorcodeptr points to the error code variable
1508 Returns: TRUE if the type value was found, or FALSE for an invalid type
1512 get_ucp(const pcre_uchar **ptrptr, BOOL *negptr, unsigned int *ptypeptr,
1513 unsigned int *pdataptr, int *errorcodeptr)
1517 const pcre_uchar *ptr = *ptrptr;
1518 pcre_uchar name[32];
1521 if (c == CHAR_NULL) goto ERROR_RETURN;
1525 /* \P or \p can be followed by a name in {}, optionally preceded by ^ for
1528 if (c == CHAR_LEFT_CURLY_BRACKET)
1530 if (ptr[1] == CHAR_CIRCUMFLEX_ACCENT)
1535 for (i = 0; i < (int)(sizeof(name) / sizeof(pcre_uchar)) - 1; i++)
1538 if (c == CHAR_NULL) goto ERROR_RETURN;
1539 if (c == CHAR_RIGHT_CURLY_BRACKET) break;
1542 if (c != CHAR_RIGHT_CURLY_BRACKET) goto ERROR_RETURN;
1546 /* Otherwise there is just one following character */
1556 /* Search for a recognized property name using binary chop */
1559 top = PRIV(utt_size);
1564 i = (bot + top) >> 1;
1565 r = STRCMP_UC_C8(name, PRIV(utt_names) + PRIV(utt)[i].name_offset);
1568 *ptypeptr = PRIV(utt)[i].type;
1569 *pdataptr = PRIV(utt)[i].value;
1572 if (r > 0) bot = i + 1; else top = i;
1575 *errorcodeptr = ERR47;
1580 *errorcodeptr = ERR46;
1588 /*************************************************
1589 * Read repeat counts *
1590 *************************************************/
1592 /* Read an item of the form {n,m} and return the values. This is called only
1593 after is_counted_repeat() has confirmed that a repeat-count quantifier exists,
1594 so the syntax is guaranteed to be correct, but we need to check the values.
1597 p pointer to first char after '{'
1598 minp pointer to int for min
1599 maxp pointer to int for max
1600 returned as -1 if no max
1601 errorcodeptr points to error code variable
1603 Returns: pointer to '}' on success;
1604 current ptr on error, with errorcodeptr set non-zero
1607 static const pcre_uchar *
1608 read_repeat_counts(const pcre_uchar *p, int *minp, int *maxp, int *errorcodeptr)
1613 while (IS_DIGIT(*p))
1615 min = min * 10 + (int)(*p++ - CHAR_0);
1618 *errorcodeptr = ERR5;
1623 if (*p == CHAR_RIGHT_CURLY_BRACKET) max = min; else
1625 if (*(++p) != CHAR_RIGHT_CURLY_BRACKET)
1630 max = max * 10 + (int)(*p++ - CHAR_0);
1633 *errorcodeptr = ERR5;
1639 *errorcodeptr = ERR4;
1652 /*************************************************
1653 * Find first significant op code *
1654 *************************************************/
1656 /* This is called by several functions that scan a compiled expression looking
1657 for a fixed first character, or an anchoring op code etc. It skips over things
1658 that do not influence this. For some calls, it makes sense to skip negative
1659 forward and all backward assertions, and also the \b assertion; for others it
1663 code pointer to the start of the group
1664 skipassert TRUE if certain assertions are to be skipped
1666 Returns: pointer to the first significant opcode
1669 static const pcre_uchar*
1670 first_significant_code(const pcre_uchar *code, BOOL skipassert)
1678 case OP_ASSERTBACK_NOT:
1679 if (!skipassert) return code;
1680 do code += GET(code, 1); while (*code == OP_ALT);
1681 code += PRIV(OP_lengths)[*code];
1684 case OP_WORD_BOUNDARY:
1685 case OP_NOT_WORD_BOUNDARY:
1686 if (!skipassert) return code;
1695 code += PRIV(OP_lengths)[*code];
1702 /* Control never reaches here */
1707 /*************************************************
1708 * Find the fixed length of a branch *
1709 *************************************************/
1711 /* Scan a branch and compute the fixed length of subject that will match it,
1712 if the length is fixed. This is needed for dealing with backward assertions.
1713 In UTF8 mode, the result is in characters rather than bytes. The branch is
1714 temporarily terminated with OP_END when this function is called.
1716 This function is called when a backward assertion is encountered, so that if it
1717 fails, the error message can point to the correct place in the pattern.
1718 However, we cannot do this when the assertion contains subroutine calls,
1719 because they can be forward references. We solve this by remembering this case
1720 and doing the check at the end; a flag specifies which mode we are running in.
1723 code points to the start of the pattern (the bracket)
1724 utf TRUE in UTF-8 / UTF-16 / UTF-32 mode
1725 atend TRUE if called when the pattern is complete
1726 cd the "compile data" structure
1727 recurses chain of recurse_check to catch mutual recursion
1729 Returns: the fixed length,
1730 or -1 if there is no fixed length,
1731 or -2 if \C was encountered (in UTF-8 mode only)
1732 or -3 if an OP_RECURSE item was encountered and atend is FALSE
1733 or -4 if an unknown opcode was encountered (internal error)
1737 find_fixedlength(pcre_uchar *code, BOOL utf, BOOL atend, compile_data *cd,
1738 recurse_check *recurses)
1741 recurse_check this_recurse;
1742 register int branchlength = 0;
1743 register pcre_uchar *cc = code + 1 + LINK_SIZE;
1745 /* Scan along the opcodes for this branch. If we get to the end of the
1746 branch, check the length against that of the other branches. */
1751 pcre_uchar *ce, *cs;
1752 register pcre_uchar op = *cc;
1756 /* We only need to continue for OP_CBRA (normal capturing bracket) and
1757 OP_BRA (normal non-capturing bracket) because the other variants of these
1758 opcodes are all concerned with unlimited repeated groups, which of course
1759 are not of fixed length. */
1766 d = find_fixedlength(cc + ((op == OP_CBRA)? IMM2_SIZE : 0), utf, atend, cd,
1768 if (d < 0) return d;
1770 do cc += GET(cc, 1); while (*cc == OP_ALT);
1771 cc += 1 + LINK_SIZE;
1774 /* Reached end of a branch; if it's a ket it is the end of a nested call.
1775 If it's ALT it is an alternation in a nested call. An ACCEPT is effectively
1776 an ALT. If it is END it's the end of the outer call. All can be handled by
1777 the same code. Note that we must not include the OP_KETRxxx opcodes here,
1778 because they all imply an unlimited repeat. */
1784 case OP_ASSERT_ACCEPT:
1785 if (length < 0) length = branchlength;
1786 else if (length != branchlength) return -1;
1787 if (*cc != OP_ALT) return length;
1788 cc += 1 + LINK_SIZE;
1792 /* A true recursion implies not fixed length, but a subroutine call may
1793 be OK. If the subroutine is a forward reference, we can't deal with
1794 it until the end of the pattern, so return -3. */
1797 if (!atend) return -3;
1798 cs = ce = (pcre_uchar *)cd->start_code + GET(cc, 1); /* Start subpattern */
1799 do ce += GET(ce, 1); while (*ce == OP_ALT); /* End subpattern */
1800 if (cc > cs && cc < ce) return -1; /* Recursion */
1801 else /* Check for mutual recursion */
1803 recurse_check *r = recurses;
1804 for (r = recurses; r != NULL; r = r->prev) if (r->group == cs) break;
1805 if (r != NULL) return -1; /* Mutual recursion */
1807 this_recurse.prev = recurses;
1808 this_recurse.group = cs;
1809 d = find_fixedlength(cs + IMM2_SIZE, utf, atend, cd, &this_recurse);
1810 if (d < 0) return d;
1812 cc += 1 + LINK_SIZE;
1815 /* Skip over assertive subpatterns */
1820 case OP_ASSERTBACK_NOT:
1821 do cc += GET(cc, 1); while (*cc == OP_ALT);
1822 cc += 1 + LINK_SIZE;
1825 /* Skip over things that don't match chars */
1831 cc += cc[1] + PRIV(OP_lengths)[*cc];
1848 case OP_NOT_WORD_BOUNDARY:
1857 case OP_WORD_BOUNDARY:
1858 cc += PRIV(OP_lengths)[*cc];
1861 /* Handle literal characters */
1870 if (utf && HAS_EXTRALEN(cc[-1])) cc += GET_EXTRALEN(cc[-1]);
1874 /* Handle exact repetitions. The count is already in characters, but we
1875 need to skip over a multibyte character in UTF8 mode. */
1881 branchlength += (int)GET2(cc,1);
1882 cc += 2 + IMM2_SIZE;
1884 if (utf && HAS_EXTRALEN(cc[-1])) cc += GET_EXTRALEN(cc[-1]);
1889 branchlength += GET2(cc,1);
1890 if (cc[1 + IMM2_SIZE] == OP_PROP || cc[1 + IMM2_SIZE] == OP_NOTPROP)
1892 cc += 1 + IMM2_SIZE + 1;
1895 /* Handle single-char matchers */
1908 case OP_NOT_WHITESPACE:
1910 case OP_NOT_WORDCHAR:
1918 /* The single-byte matcher isn't allowed. This only happens in UTF-8 mode;
1919 otherwise \C is coded as OP_ALLANY. */
1924 /* Check a class for variable quantification */
1928 #if defined SUPPORT_UTF || defined COMPILE_PCRE16 || defined COMPILE_PCRE32
1930 /* The original code caused an unsigned overflow in 64 bit systems,
1931 so now we use a conditional statement. */
1932 if (op == OP_XCLASS)
1935 cc += PRIV(OP_lengths)[OP_CLASS];
1937 cc += PRIV(OP_lengths)[OP_CLASS];
1956 if (GET2(cc,1) != GET2(cc,1+IMM2_SIZE)) return -1;
1957 branchlength += (int)GET2(cc,1);
1958 cc += 1 + 2 * IMM2_SIZE;
1966 /* Anything else is variable length */
1987 case OP_NOTMINPLUSI:
1988 case OP_NOTMINQUERY:
1989 case OP_NOTMINQUERYI:
1991 case OP_NOTMINSTARI:
1993 case OP_NOTMINUPTOI:
1997 case OP_NOTPOSPLUSI:
1998 case OP_NOTPOSQUERY:
1999 case OP_NOTPOSQUERYI:
2001 case OP_NOTPOSSTARI:
2003 case OP_NOTPOSUPTOI:
2034 case OP_TYPEMINPLUS:
2035 case OP_TYPEMINQUERY:
2036 case OP_TYPEMINSTAR:
2037 case OP_TYPEMINUPTO:
2039 case OP_TYPEPOSPLUS:
2040 case OP_TYPEPOSQUERY:
2041 case OP_TYPEPOSSTAR:
2042 case OP_TYPEPOSUPTO:
2050 /* Catch unrecognized opcodes so that when new ones are added they
2051 are not forgotten, as has happened in the past. */
2057 /* Control never gets here */
2062 /*************************************************
2063 * Scan compiled regex for specific bracket *
2064 *************************************************/
2066 /* This little function scans through a compiled pattern until it finds a
2067 capturing bracket with the given number, or, if the number is negative, an
2068 instance of OP_REVERSE for a lookbehind. The function is global in the C sense
2069 so that it can be called from pcre_study() when finding the minimum matching
2073 code points to start of expression
2074 utf TRUE in UTF-8 / UTF-16 / UTF-32 mode
2075 number the required bracket number or negative to find a lookbehind
2077 Returns: pointer to the opcode for the bracket, or NULL if not found
2081 PRIV(find_bracket)(const pcre_uchar *code, BOOL utf, int number)
2085 register pcre_uchar c = *code;
2087 if (c == OP_END) return NULL;
2089 /* XCLASS is used for classes that cannot be represented just by a bit
2090 map. This includes negated single high-valued characters. The length in
2091 the table is zero; the actual length is stored in the compiled code. */
2093 if (c == OP_XCLASS) code += GET(code, 1);
2095 /* Handle recursion */
2097 else if (c == OP_REVERSE)
2099 if (number < 0) return (pcre_uchar *)code;
2100 code += PRIV(OP_lengths)[c];
2103 /* Handle capturing bracket */
2105 else if (c == OP_CBRA || c == OP_SCBRA ||
2106 c == OP_CBRAPOS || c == OP_SCBRAPOS)
2108 int n = (int)GET2(code, 1+LINK_SIZE);
2109 if (n == number) return (pcre_uchar *)code;
2110 code += PRIV(OP_lengths)[c];
2113 /* Otherwise, we can get the item's length from the table, except that for
2114 repeated character types, we have to test for \p and \P, which have an extra
2115 two bytes of parameters, and for MARK/PRUNE/SKIP/THEN with an argument, we
2116 must add in its length. */
2123 case OP_TYPEMINSTAR:
2125 case OP_TYPEMINPLUS:
2127 case OP_TYPEMINQUERY:
2128 case OP_TYPEPOSSTAR:
2129 case OP_TYPEPOSPLUS:
2130 case OP_TYPEPOSQUERY:
2131 if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
2135 case OP_TYPEMINUPTO:
2137 case OP_TYPEPOSUPTO:
2138 if (code[1 + IMM2_SIZE] == OP_PROP || code[1 + IMM2_SIZE] == OP_NOTPROP)
2150 /* Add in the fixed length from the table */
2152 code += PRIV(OP_lengths)[c];
2154 /* In UTF-8 mode, opcodes that are followed by a character may be followed by
2155 a multi-byte character. The length in the table is a minimum, so we have to
2156 arrange to skip the extra bytes. */
2158 #if defined SUPPORT_UTF && !defined COMPILE_PCRE32
2176 case OP_NOTMINUPTOI:
2180 case OP_NOTPOSUPTOI:
2188 case OP_NOTMINSTARI:
2192 case OP_NOTPOSSTARI:
2200 case OP_NOTMINPLUSI:
2204 case OP_NOTPOSPLUSI:
2211 case OP_NOTMINQUERY:
2212 case OP_NOTMINQUERYI:
2215 case OP_NOTPOSQUERY:
2216 case OP_NOTPOSQUERYI:
2217 if (HAS_EXTRALEN(code[-1])) code += GET_EXTRALEN(code[-1]);
2221 (void)(utf); /* Keep compiler happy by referencing function argument */
2229 /*************************************************
2230 * Scan compiled regex for recursion reference *
2231 *************************************************/
2233 /* This little function scans through a compiled pattern until it finds an
2234 instance of OP_RECURSE.
2237 code points to start of expression
2238 utf TRUE in UTF-8 / UTF-16 / UTF-32 mode
2240 Returns: pointer to the opcode for OP_RECURSE, or NULL if not found
2243 static const pcre_uchar *
2244 find_recurse(const pcre_uchar *code, BOOL utf)
2248 register pcre_uchar c = *code;
2249 if (c == OP_END) return NULL;
2250 if (c == OP_RECURSE) return code;
2252 /* XCLASS is used for classes that cannot be represented just by a bit
2253 map. This includes negated single high-valued characters. The length in
2254 the table is zero; the actual length is stored in the compiled code. */
2256 if (c == OP_XCLASS) code += GET(code, 1);
2258 /* Otherwise, we can get the item's length from the table, except that for
2259 repeated character types, we have to test for \p and \P, which have an extra
2260 two bytes of parameters, and for MARK/PRUNE/SKIP/THEN with an argument, we
2261 must add in its length. */
2268 case OP_TYPEMINSTAR:
2270 case OP_TYPEMINPLUS:
2272 case OP_TYPEMINQUERY:
2273 case OP_TYPEPOSSTAR:
2274 case OP_TYPEPOSPLUS:
2275 case OP_TYPEPOSQUERY:
2276 if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
2279 case OP_TYPEPOSUPTO:
2281 case OP_TYPEMINUPTO:
2283 if (code[1 + IMM2_SIZE] == OP_PROP || code[1 + IMM2_SIZE] == OP_NOTPROP)
2295 /* Add in the fixed length from the table */
2297 code += PRIV(OP_lengths)[c];
2299 /* In UTF-8 mode, opcodes that are followed by a character may be followed
2300 by a multi-byte character. The length in the table is a minimum, so we have
2301 to arrange to skip the extra bytes. */
2303 #if defined SUPPORT_UTF && !defined COMPILE_PCRE32
2321 case OP_NOTMINUPTOI:
2325 case OP_NOTPOSUPTOI:
2333 case OP_NOTMINSTARI:
2337 case OP_NOTPOSSTARI:
2345 case OP_NOTMINPLUSI:
2349 case OP_NOTPOSPLUSI:
2356 case OP_NOTMINQUERY:
2357 case OP_NOTMINQUERYI:
2360 case OP_NOTPOSQUERY:
2361 case OP_NOTPOSQUERYI:
2362 if (HAS_EXTRALEN(code[-1])) code += GET_EXTRALEN(code[-1]);
2366 (void)(utf); /* Keep compiler happy by referencing function argument */
2374 /*************************************************
2375 * Scan compiled branch for non-emptiness *
2376 *************************************************/
2378 /* This function scans through a branch of a compiled pattern to see whether it
2379 can match the empty string or not. It is called from could_be_empty()
2380 below and from compile_branch() when checking for an unlimited repeat of a
2381 group that can match nothing. Note that first_significant_code() skips over
2382 backward and negative forward assertions when its final argument is TRUE. If we
2383 hit an unclosed bracket, we return "empty" - this means we've struck an inner
2384 bracket whose current branch will already have been scanned.
2387 code points to start of search
2388 endcode points to where to stop
2389 utf TRUE if in UTF-8 / UTF-16 / UTF-32 mode
2390 cd contains pointers to tables etc.
2391 recurses chain of recurse_check to catch mutual recursion
2393 Returns: TRUE if what is matched could be empty
2397 could_be_empty_branch(const pcre_uchar *code, const pcre_uchar *endcode,
2398 BOOL utf, compile_data *cd, recurse_check *recurses)
2400 register pcre_uchar c;
2401 recurse_check this_recurse;
2403 for (code = first_significant_code(code + PRIV(OP_lengths)[*code], TRUE);
2405 code = first_significant_code(code + PRIV(OP_lengths)[c], TRUE))
2407 const pcre_uchar *ccode;
2411 /* Skip over forward assertions; the other assertions are skipped by
2412 first_significant_code() with a TRUE final argument. */
2416 do code += GET(code, 1); while (*code == OP_ALT);
2421 /* For a recursion/subroutine call, if its end has been reached, which
2422 implies a backward reference subroutine call, we can scan it. If it's a
2423 forward reference subroutine call, we can't. To detect forward reference
2424 we have to scan up the list that is kept in the workspace. This function is
2425 called only when doing the real compile, not during the pre-compile that
2426 measures the size of the compiled pattern. */
2428 if (c == OP_RECURSE)
2430 const pcre_uchar *scode = cd->start_code + GET(code, 1);
2431 const pcre_uchar *endgroup = scode;
2434 /* Test for forward reference or uncompleted reference. This is disabled
2435 when called to scan a completed pattern by setting cd->start_workspace to
2438 if (cd->start_workspace != NULL)
2440 const pcre_uchar *tcode;
2441 for (tcode = cd->start_workspace; tcode < cd->hwm; tcode += LINK_SIZE)
2442 if ((int)GET(tcode, 0) == (int)(code + 1 - cd->start_code)) return TRUE;
2443 if (GET(scode, 1) == 0) return TRUE; /* Unclosed */
2446 /* If the reference is to a completed group, we need to detect whether this
2447 is a recursive call, as otherwise there will be an infinite loop. If it is
2448 a recursion, just skip over it. Simple recursions are easily detected. For
2449 mutual recursions we keep a chain on the stack. */
2451 do endgroup += GET(endgroup, 1); while (*endgroup == OP_ALT);
2452 if (code >= scode && code <= endgroup) continue; /* Simple recursion */
2455 recurse_check *r = recurses;
2456 for (r = recurses; r != NULL; r = r->prev)
2457 if (r->group == scode) break;
2458 if (r != NULL) continue; /* Mutual recursion */
2461 /* Completed reference; scan the referenced group, remembering it on the
2462 stack chain to detect mutual recursions. */
2464 empty_branch = FALSE;
2465 this_recurse.prev = recurses;
2466 this_recurse.group = scode;
2470 if (could_be_empty_branch(scode, endcode, utf, cd, &this_recurse))
2472 empty_branch = TRUE;
2475 scode += GET(scode, 1);
2477 while (*scode == OP_ALT);
2479 if (!empty_branch) return FALSE; /* All branches are non-empty */
2483 /* Groups with zero repeats can of course be empty; skip them. */
2485 if (c == OP_BRAZERO || c == OP_BRAMINZERO || c == OP_SKIPZERO ||
2488 code += PRIV(OP_lengths)[c];
2489 do code += GET(code, 1); while (*code == OP_ALT);
2494 /* A nested group that is already marked as "could be empty" can just be
2497 if (c == OP_SBRA || c == OP_SBRAPOS ||
2498 c == OP_SCBRA || c == OP_SCBRAPOS)
2500 do code += GET(code, 1); while (*code == OP_ALT);
2505 /* For other groups, scan the branches. */
2507 if (c == OP_BRA || c == OP_BRAPOS ||
2508 c == OP_CBRA || c == OP_CBRAPOS ||
2509 c == OP_ONCE || c == OP_ONCE_NC ||
2510 c == OP_COND || c == OP_SCOND)
2513 if (GET(code, 1) == 0) return TRUE; /* Hit unclosed bracket */
2515 /* If a conditional group has only one branch, there is a second, implied,
2516 empty branch, so just skip over the conditional, because it could be empty.
2517 Otherwise, scan the individual branches of the group. */
2519 if (c == OP_COND && code[GET(code, 1)] != OP_ALT)
2520 code += GET(code, 1);
2523 empty_branch = FALSE;
2526 if (!empty_branch && could_be_empty_branch(code, endcode, utf, cd,
2527 recurses)) empty_branch = TRUE;
2528 code += GET(code, 1);
2530 while (*code == OP_ALT);
2531 if (!empty_branch) return FALSE; /* All branches are non-empty */
2538 /* Handle the other opcodes */
2542 /* Check for quantifiers after a class. XCLASS is used for classes that
2543 cannot be represented just by a bit map. This includes negated single
2544 high-valued characters. The length in PRIV(OP_lengths)[] is zero; the
2545 actual length is stored in the compiled code, so we must update "code"
2548 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
2550 ccode = code += GET(code, 1);
2551 goto CHECK_CLASS_REPEAT;
2556 ccode = code + PRIV(OP_lengths)[OP_CLASS];
2558 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
2564 case OP_CRSTAR: /* These could be empty; continue */
2572 default: /* Non-repeat => class must match */
2573 case OP_CRPLUS: /* These repeats aren't empty */
2581 if (GET2(ccode, 1) > 0) return FALSE; /* Minimum > 0 */
2586 /* Opcodes that must match a character */
2604 case OP_NOT_WHITESPACE:
2606 case OP_NOT_WORDCHAR:
2622 case OP_NOTMINPLUSI:
2627 case OP_NOTPOSPLUSI:
2635 case OP_TYPEMINPLUS:
2636 case OP_TYPEPOSPLUS:
2641 /* These are going to continue, as they may be empty, but we have to
2642 fudge the length for the \p and \P cases. */
2645 case OP_TYPEMINSTAR:
2646 case OP_TYPEPOSSTAR:
2648 case OP_TYPEMINQUERY:
2649 case OP_TYPEPOSQUERY:
2650 if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
2653 /* Same for these */
2656 case OP_TYPEMINUPTO:
2657 case OP_TYPEPOSUPTO:
2658 if (code[1 + IMM2_SIZE] == OP_PROP || code[1 + IMM2_SIZE] == OP_NOTPROP)
2671 /* In UTF-8 mode, STAR, MINSTAR, POSSTAR, QUERY, MINQUERY, POSQUERY, UPTO,
2672 MINUPTO, and POSUPTO and their caseless and negative versions may be
2673 followed by a multibyte character. */
2675 #if defined SUPPORT_UTF && !defined COMPILE_PCRE32
2684 case OP_NOTMINSTARI:
2689 case OP_NOTPOSSTARI:
2698 case OP_NOTMINQUERY:
2699 case OP_NOTMINQUERYI:
2703 case OP_NOTPOSQUERY:
2704 case OP_NOTPOSQUERYI:
2706 if (utf && HAS_EXTRALEN(code[1])) code += GET_EXTRALEN(code[1]);
2717 case OP_NOTMINUPTOI:
2722 case OP_NOTPOSUPTOI:
2724 if (utf && HAS_EXTRALEN(code[1 + IMM2_SIZE])) code += GET_EXTRALEN(code[1 + IMM2_SIZE]);
2728 /* MARK, and PRUNE/SKIP/THEN with an argument must skip over the argument
2738 /* None of the remaining opcodes are required to match a character. */
2750 /*************************************************
2751 * Scan compiled regex for non-emptiness *
2752 *************************************************/
2754 /* This function is called to check for left recursive calls. We want to check
2755 the current branch of the current pattern to see if it could match the empty
2756 string. If it could, we must look outwards for branches at other levels,
2757 stopping when we pass beyond the bracket which is the subject of the recursion.
2758 This function is called only during the real compile, not during the
2762 code points to start of the recursion
2763 endcode points to where to stop (current RECURSE item)
2764 bcptr points to the chain of current (unclosed) branch starts
2765 utf TRUE if in UTF-8 / UTF-16 / UTF-32 mode
2766 cd pointers to tables etc
2768 Returns: TRUE if what is matched could be empty
2772 could_be_empty(const pcre_uchar *code, const pcre_uchar *endcode,
2773 branch_chain *bcptr, BOOL utf, compile_data *cd)
2775 while (bcptr != NULL && bcptr->current_branch >= code)
2777 if (!could_be_empty_branch(bcptr->current_branch, endcode, utf, cd, NULL))
2779 bcptr = bcptr->outer;
2786 /*************************************************
2787 * Base opcode of repeated opcodes *
2788 *************************************************/
2790 /* Returns the base opcode for repeated single character type opcodes. If the
2791 opcode is not a repeated character type, it returns with the original value.
2794 Returns: base opcode for the type
2798 get_repeat_base(pcre_uchar c)
2800 return (c > OP_TYPEPOSUPTO)? c :
2801 (c >= OP_TYPESTAR)? OP_TYPESTAR :
2802 (c >= OP_NOTSTARI)? OP_NOTSTARI :
2803 (c >= OP_NOTSTAR)? OP_NOTSTAR :
2804 (c >= OP_STARI)? OP_STARI :
2811 /*************************************************
2812 * Check a character and a property *
2813 *************************************************/
2815 /* This function is called by check_auto_possessive() when a property item
2816 is adjacent to a fixed character.
2820 ptype the property type
2821 pdata the data for the type
2822 negated TRUE if it's a negated property (\P or \p{^)
2824 Returns: TRUE if auto-possessifying is OK
2828 check_char_prop(pcre_uint32 c, unsigned int ptype, unsigned int pdata,
2831 const pcre_uint32 *p;
2832 const ucd_record *prop = GET_UCD(c);
2837 return (prop->chartype == ucp_Lu ||
2838 prop->chartype == ucp_Ll ||
2839 prop->chartype == ucp_Lt) == negated;
2842 return (pdata == PRIV(ucp_gentype)[prop->chartype]) == negated;
2845 return (pdata == prop->chartype) == negated;
2848 return (pdata == prop->script) == negated;
2850 /* These are specials */
2853 return (PRIV(ucp_gentype)[prop->chartype] == ucp_L ||
2854 PRIV(ucp_gentype)[prop->chartype] == ucp_N) == negated;
2856 /* Perl space used to exclude VT, but from Perl 5.18 it is included, which
2857 means that Perl space and POSIX space are now identical. PCRE was changed
2860 case PT_SPACE: /* Perl space */
2861 case PT_PXSPACE: /* POSIX space */
2869 return (PRIV(ucp_gentype)[prop->chartype] == ucp_Z) == negated;
2871 break; /* Control never reaches here */
2874 return (PRIV(ucp_gentype)[prop->chartype] == ucp_L ||
2875 PRIV(ucp_gentype)[prop->chartype] == ucp_N ||
2876 c == CHAR_UNDERSCORE) == negated;
2879 p = PRIV(ucd_caseless_sets) + prop->caseset;
2882 if (c < *p) return !negated;
2883 if (c == *p++) return negated;
2885 break; /* Control never reaches here */
2890 #endif /* SUPPORT_UCP */
2894 /*************************************************
2895 * Fill the character property list *
2896 *************************************************/
2898 /* Checks whether the code points to an opcode that can take part in auto-
2899 possessification, and if so, fills a list with its properties.
2902 code points to start of expression
2903 utf TRUE if in UTF-8 / UTF-16 / UTF-32 mode
2904 fcc points to case-flipping table
2905 list points to output list
2906 list[0] will be filled with the opcode
2907 list[1] will be non-zero if this opcode
2908 can match an empty character string
2909 list[2..7] depends on the opcode
2911 Returns: points to the start of the next opcode if *code is accepted
2912 NULL if *code is not accepted
2915 static const pcre_uchar *
2916 get_chr_property_list(const pcre_uchar *code, BOOL utf,
2917 const pcre_uint8 *fcc, pcre_uint32 *list)
2919 pcre_uchar c = *code;
2921 const pcre_uchar *end;
2925 pcre_uint32 *clist_dest;
2926 const pcre_uint32 *clist_src;
2928 utf = utf; /* Suppress "unused parameter" compiler warning */
2935 if (c >= OP_STAR && c <= OP_TYPEPOSUPTO)
2937 base = get_repeat_base(c);
2938 c -= (base - OP_STAR);
2940 if (c == OP_UPTO || c == OP_MINUPTO || c == OP_EXACT || c == OP_POSUPTO)
2943 list[1] = (c != OP_PLUS && c != OP_MINPLUS && c != OP_EXACT && c != OP_POSPLUS);
2975 case OP_NOT_WHITESPACE:
2977 case OP_NOT_WORDCHAR:
2995 GETCHARINCTEST(chr, code);
3002 list[0] = (c == OP_CHARI) ? OP_CHAR : OP_NOT;
3003 GETCHARINCTEST(chr, code);
3007 if (chr < 128 || (chr < 256 && !utf))
3010 list[3] = UCD_OTHERCASE(chr);
3011 #elif defined SUPPORT_UTF || !defined COMPILE_PCRE8
3012 list[3] = (chr < 256) ? fcc[chr] : chr;
3017 /* The othercase might be the same value. */
3028 if (code[0] != PT_CLIST)
3035 /* Convert only if we have enough space. */
3037 clist_src = PRIV(ucd_caseless_sets) + code[1];
3038 clist_dest = list + 2;
3042 if (clist_dest >= list + 8)
3044 /* Early return if there is not enough space. This should never
3045 happen, since all clists are shorter than 5 character now. */
3050 *clist_dest++ = *clist_src;
3052 while(*clist_src++ != NOTACHAR);
3054 /* All characters are stored. The terminating NOTACHAR
3055 is copied form the clist itself. */
3057 list[0] = (c == OP_PROP) ? OP_CHAR : OP_NOT;
3063 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
3066 end = code + GET(code, 0) - 1;
3069 end = code + 32 / sizeof(pcre_uchar);
3092 list[1] = (GET2(end, 1) == 0);
3093 end += 1 + 2 * IMM2_SIZE;
3096 list[2] = (pcre_uint32)(end - code);
3099 return NULL; /* Opcode not accepted */
3104 /*************************************************
3105 * Scan further character sets for match *
3106 *************************************************/
3108 /* Checks whether the base and the current opcode have a common character, in
3109 which case the base cannot be possessified.
3112 code points to the byte code
3113 utf TRUE in UTF-8 / UTF-16 / UTF-32 mode
3114 cd static compile data
3115 base_list the data list of the base opcode
3117 Returns: TRUE if the auto-possessification is possible
3121 compare_opcodes(const pcre_uchar *code, BOOL utf, const compile_data *cd,
3122 const pcre_uint32 *base_list, const pcre_uchar *base_end, int *rec_limit)
3125 pcre_uint32 list[8];
3126 const pcre_uint32 *chr_ptr;
3127 const pcre_uint32 *ochr_ptr;
3128 const pcre_uint32 *list_ptr;
3129 const pcre_uchar *next_code;
3130 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
3131 const pcre_uchar *xclass_flags;
3133 const pcre_uint8 *class_bitset;
3134 const pcre_uint8 *set1, *set2, *set_end;
3136 BOOL accepted, invert_bits;
3137 BOOL entered_a_group = FALSE;
3139 if (*rec_limit == 0) return FALSE;
3142 /* Note: the base_list[1] contains whether the current opcode has greedy
3143 (represented by a non-zero value) quantifier. This is a different from
3144 other character type lists, which stores here that the character iterator
3145 matches to an empty string (also represented by a non-zero value). */
3149 /* All operations move the code pointer forward.
3150 Therefore infinite recursions are not possible. */
3154 /* Skip over callouts */
3156 if (c == OP_CALLOUT)
3158 code += PRIV(OP_lengths)[c];
3164 do code += GET(code, 1); while (*code == OP_ALT);
3172 /* TRUE only in greedy case. The non-greedy case could be replaced by
3173 an OP_EXACT, but it is probably not worth it. (And note that OP_EXACT
3174 uses more memory, which we cannot get at this stage.) */
3176 return base_list[1] != 0;
3179 /* If the bracket is capturing, and referenced by an OP_RECURSE, or
3180 it is an atomic sub-pattern (assert, once, etc.) the non-greedy case
3181 cannot be converted to a possessive form. */
3183 if (base_list[1] == 0) return FALSE;
3185 switch(*(code - GET(code, 1)))
3190 case OP_ASSERTBACK_NOT:
3193 /* Atomic sub-patterns and assertions can always auto-possessify their
3194 last iterator. However, if the group was entered as a result of checking
3195 a previous iterator, this is not possible. */
3197 return !entered_a_group;
3200 code += PRIV(OP_lengths)[c];
3207 next_code = code + GET(code, 1);
3208 code += PRIV(OP_lengths)[c];
3210 while (*next_code == OP_ALT)
3212 if (!compare_opcodes(code, utf, cd, base_list, base_end, rec_limit))
3214 code = next_code + 1 + LINK_SIZE;
3215 next_code += GET(next_code, 1);
3218 entered_a_group = TRUE;
3224 next_code = code + 1;
3225 if (*next_code != OP_BRA && *next_code != OP_CBRA
3226 && *next_code != OP_ONCE && *next_code != OP_ONCE_NC) return FALSE;
3228 do next_code += GET(next_code, 1); while (*next_code == OP_ALT);
3230 /* The bracket content will be checked by the
3231 OP_BRA/OP_CBRA case above. */
3232 next_code += 1 + LINK_SIZE;
3233 if (!compare_opcodes(next_code, utf, cd, base_list, base_end, rec_limit))
3236 code += PRIV(OP_lengths)[c];
3243 /* Check for a supported opcode, and load its properties. */
3245 code = get_chr_property_list(code, utf, cd->fcc, list);
3246 if (code == NULL) return FALSE; /* Unsupported */
3248 /* If either opcode is a small character list, set pointers for comparing
3249 characters from that list with another list, or with a property. */
3251 if (base_list[0] == OP_CHAR)
3253 chr_ptr = base_list + 2;
3256 else if (list[0] == OP_CHAR)
3259 list_ptr = base_list;
3262 /* Character bitsets can also be compared to certain opcodes. */
3264 else if (base_list[0] == OP_CLASS || list[0] == OP_CLASS
3265 #ifdef COMPILE_PCRE8
3266 /* In 8 bit, non-UTF mode, OP_CLASS and OP_NCLASS are the same. */
3267 || (!utf && (base_list[0] == OP_NCLASS || list[0] == OP_NCLASS))
3271 #ifdef COMPILE_PCRE8
3272 if (base_list[0] == OP_CLASS || (!utf && base_list[0] == OP_NCLASS))
3274 if (base_list[0] == OP_CLASS)
3277 set1 = (pcre_uint8 *)(base_end - base_list[2]);
3282 set1 = (pcre_uint8 *)(code - list[2]);
3283 list_ptr = base_list;
3286 invert_bits = FALSE;
3291 set2 = (pcre_uint8 *)
3292 ((list_ptr == list ? code : base_end) - list_ptr[2]);
3295 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
3297 xclass_flags = (list_ptr == list ? code : base_end) - list_ptr[2] + LINK_SIZE;
3298 if ((*xclass_flags & XCL_HASPROP) != 0) return FALSE;
3299 if ((*xclass_flags & XCL_MAP) == 0)
3301 /* No bits are set for characters < 256. */
3302 if (list[1] == 0) return TRUE;
3303 /* Might be an empty repeat. */
3306 set2 = (pcre_uint8 *)(xclass_flags + 1);
3314 set2 = (pcre_uint8 *)(cd->cbits + cbit_digit);
3317 case OP_NOT_WHITESPACE:
3321 set2 = (pcre_uint8 *)(cd->cbits + cbit_space);
3324 case OP_NOT_WORDCHAR:
3328 set2 = (pcre_uint8 *)(cd->cbits + cbit_word);
3335 /* Because the sets are unaligned, we need
3336 to perform byte comparison here. */
3337 set_end = set1 + 32;
3342 if ((*set1++ & ~(*set2++)) != 0) return FALSE;
3344 while (set1 < set_end);
3350 if ((*set1++ & *set2++) != 0) return FALSE;
3352 while (set1 < set_end);
3355 if (list[1] == 0) return TRUE;
3356 /* Might be an empty repeat. */
3360 /* Some property combinations also acceptable. Unicode property opcodes are
3361 processed specially; the rest can be handled with a lookup table. */
3365 pcre_uint32 leftop, rightop;
3367 leftop = base_list[0];
3371 accepted = FALSE; /* Always set in non-unicode case. */
3372 if (leftop == OP_PROP || leftop == OP_NOTPROP)
3374 if (rightop == OP_EOD)
3376 else if (rightop == OP_PROP || rightop == OP_NOTPROP)
3379 const pcre_uint8 *p;
3380 BOOL same = leftop == rightop;
3381 BOOL lisprop = leftop == OP_PROP;
3382 BOOL risprop = rightop == OP_PROP;
3383 BOOL bothprop = lisprop && risprop;
3385 /* There's a table that specifies how each combination is to be
3387 0 Always return FALSE (never auto-possessify)
3388 1 Character groups are distinct (possessify if both are OP_PROP)
3389 2 Check character categories in the same group (general or particular)
3390 3 Return TRUE if the two opcodes are not the same
3391 ... see comments below
3394 n = propposstab[base_list[2]][list[2]];
3398 case 1: accepted = bothprop; break;
3399 case 2: accepted = (base_list[3] == list[3]) != same; break;
3400 case 3: accepted = !same; break;
3402 case 4: /* Left general category, right particular category */
3403 accepted = risprop && catposstab[base_list[3]][list[3]] == same;
3406 case 5: /* Right general category, left particular category */
3407 accepted = lisprop && catposstab[list[3]][base_list[3]] == same;
3410 /* This code is logically tricky. Think hard before fiddling with it.
3411 The posspropstab table has four entries per row. Each row relates to
3412 one of PCRE's special properties such as ALNUM or SPACE or WORD.
3413 Only WORD actually needs all four entries, but using repeats for the
3414 others means they can all use the same code below.
3416 The first two entries in each row are Unicode general categories, and
3417 apply always, because all the characters they include are part of the
3418 PCRE character set. The third and fourth entries are a general and a
3419 particular category, respectively, that include one or more relevant
3420 characters. One or the other is used, depending on whether the check
3421 is for a general or a particular category. However, in both cases the
3422 category contains more characters than the specials that are defined
3423 for the property being tested against. Therefore, it cannot be used
3426 Example: the row for WORD contains ucp_L, ucp_N, ucp_P, ucp_Po.
3427 Underscore is covered by ucp_P or ucp_Po. */
3429 case 6: /* Left alphanum vs right general category */
3430 case 7: /* Left space vs right general category */
3431 case 8: /* Left word vs right general category */
3432 p = posspropstab[n-6];
3433 accepted = risprop && lisprop ==
3436 (list[3] != p[2] || !lisprop));
3439 case 9: /* Right alphanum vs left general category */
3440 case 10: /* Right space vs left general category */
3441 case 11: /* Right word vs left general category */
3442 p = posspropstab[n-9];
3443 accepted = lisprop && risprop ==
3444 (base_list[3] != p[0] &&
3445 base_list[3] != p[1] &&
3446 (base_list[3] != p[2] || !risprop));
3449 case 12: /* Left alphanum vs right particular category */
3450 case 13: /* Left space vs right particular category */
3451 case 14: /* Left word vs right particular category */
3452 p = posspropstab[n-12];
3453 accepted = risprop && lisprop ==
3454 (catposstab[p[0]][list[3]] &&
3455 catposstab[p[1]][list[3]] &&
3456 (list[3] != p[3] || !lisprop));
3459 case 15: /* Right alphanum vs left particular category */
3460 case 16: /* Right space vs left particular category */
3461 case 17: /* Right word vs left particular category */
3462 p = posspropstab[n-15];
3463 accepted = lisprop && risprop ==
3464 (catposstab[p[0]][base_list[3]] &&
3465 catposstab[p[1]][base_list[3]] &&
3466 (base_list[3] != p[3] || !risprop));
3473 #endif /* SUPPORT_UCP */
3475 accepted = leftop >= FIRST_AUTOTAB_OP && leftop <= LAST_AUTOTAB_LEFT_OP &&
3476 rightop >= FIRST_AUTOTAB_OP && rightop <= LAST_AUTOTAB_RIGHT_OP &&
3477 autoposstab[leftop - FIRST_AUTOTAB_OP][rightop - FIRST_AUTOTAB_OP];
3479 if (!accepted) return FALSE;
3481 if (list[1] == 0) return TRUE;
3482 /* Might be an empty repeat. */
3486 /* Control reaches here only if one of the items is a small character list.
3487 All characters are checked against the other side. */
3496 ochr_ptr = list_ptr + 2;
3499 if (chr == *ochr_ptr) return FALSE;
3502 while(*ochr_ptr != NOTACHAR);
3506 ochr_ptr = list_ptr + 2;
3509 if (chr == *ochr_ptr)
3513 while(*ochr_ptr != NOTACHAR);
3514 if (*ochr_ptr == NOTACHAR) return FALSE; /* Not found */
3517 /* Note that OP_DIGIT etc. are generated only when PCRE_UCP is *not*
3518 set. When it is set, \d etc. are converted into OP_(NOT_)PROP codes. */
3521 if (chr < 256 && (cd->ctypes[chr] & ctype_digit) != 0) return FALSE;
3525 if (chr > 255 || (cd->ctypes[chr] & ctype_digit) == 0) return FALSE;
3529 if (chr < 256 && (cd->ctypes[chr] & ctype_space) != 0) return FALSE;
3532 case OP_NOT_WHITESPACE:
3533 if (chr > 255 || (cd->ctypes[chr] & ctype_space) == 0) return FALSE;
3537 if (chr < 255 && (cd->ctypes[chr] & ctype_word) != 0) return FALSE;
3540 case OP_NOT_WORDCHAR:
3541 if (chr > 255 || (cd->ctypes[chr] & ctype_word) == 0) return FALSE;
3547 HSPACE_CASES: return FALSE;
3555 HSPACE_CASES: break;
3556 default: return FALSE;
3564 VSPACE_CASES: return FALSE;
3572 VSPACE_CASES: break;
3573 default: return FALSE;
3589 #endif /* Not EBCDIC */
3594 case OP_EOD: /* Can always possessify before \z */
3600 if (!check_char_prop(chr, list_ptr[2], list_ptr[3],
3601 list_ptr[0] == OP_NOTPROP))
3607 if (chr > 255) return FALSE;
3611 if (chr > 255) break;
3612 class_bitset = (pcre_uint8 *)
3613 ((list_ptr == list ? code : base_end) - list_ptr[2]);
3614 if ((class_bitset[chr >> 3] & (1 << (chr & 7))) != 0) return FALSE;
3617 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
3619 if (PRIV(xclass)(chr, (list_ptr == list ? code : base_end) -
3620 list_ptr[2] + LINK_SIZE, utf)) return FALSE;
3630 while(*chr_ptr != NOTACHAR);
3632 /* At least one character must be matched from this opcode. */
3634 if (list[1] == 0) return TRUE;
3637 /* Control never reaches here. There used to be a fail-save return FALSE; here,
3638 but some compilers complain about an unreachable statement. */
3644 /*************************************************
3645 * Scan compiled regex for auto-possession *
3646 *************************************************/
3648 /* Replaces single character iterations with their possessive alternatives
3649 if appropriate. This function modifies the compiled opcode!
3652 code points to start of the byte code
3653 utf TRUE in UTF-8 / UTF-16 / UTF-32 mode
3654 cd static compile data
3660 auto_possessify(pcre_uchar *code, BOOL utf, const compile_data *cd)
3662 register pcre_uchar c;
3663 const pcre_uchar *end;
3664 pcre_uchar *repeat_opcode;
3665 pcre_uint32 list[8];
3672 /* When a pattern with bad UTF-8 encoding is compiled with NO_UTF_CHECK,
3673 it may compile without complaining, but may get into a loop here if the code
3674 pointer points to a bad value. This is, of course a documentated possibility,
3675 when NO_UTF_CHECK is set, so it isn't a bug, but we can detect this case and
3676 just give up on this optimization. */
3678 if (c >= OP_TABLE_LENGTH) return;
3680 if (c >= OP_STAR && c <= OP_TYPEPOSUPTO)
3682 c -= get_repeat_base(c) - OP_STAR;
3683 end = (c <= OP_MINUPTO) ?
3684 get_chr_property_list(code, utf, cd->fcc, list) : NULL;
3685 list[1] = c == OP_STAR || c == OP_PLUS || c == OP_QUERY || c == OP_UPTO;
3688 if (end != NULL && compare_opcodes(end, utf, cd, list, end, &rec_limit))
3693 *code += OP_POSSTAR - OP_STAR;
3697 *code += OP_POSSTAR - OP_MINSTAR;
3701 *code += OP_POSPLUS - OP_PLUS;
3705 *code += OP_POSPLUS - OP_MINPLUS;
3709 *code += OP_POSQUERY - OP_QUERY;
3713 *code += OP_POSQUERY - OP_MINQUERY;
3717 *code += OP_POSUPTO - OP_UPTO;
3721 *code += OP_POSUPTO - OP_MINUPTO;
3727 else if (c == OP_CLASS || c == OP_NCLASS || c == OP_XCLASS)
3729 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
3731 repeat_opcode = code + GET(code, 1);
3734 repeat_opcode = code + 1 + (32 / sizeof(pcre_uchar));
3737 if (c >= OP_CRSTAR && c <= OP_CRMINRANGE)
3739 /* end must not be NULL. */
3740 end = get_chr_property_list(code, utf, cd->fcc, list);
3742 list[1] = (c & 1) == 0;
3745 if (compare_opcodes(end, utf, cd, list, end, &rec_limit))
3751 *repeat_opcode = OP_CRPOSSTAR;
3756 *repeat_opcode = OP_CRPOSPLUS;
3761 *repeat_opcode = OP_CRPOSQUERY;
3766 *repeat_opcode = OP_CRPOSRANGE;
3780 case OP_TYPEMINSTAR:
3782 case OP_TYPEMINPLUS:
3784 case OP_TYPEMINQUERY:
3785 case OP_TYPEPOSSTAR:
3786 case OP_TYPEPOSPLUS:
3787 case OP_TYPEPOSQUERY:
3788 if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
3792 case OP_TYPEMINUPTO:
3794 case OP_TYPEPOSUPTO:
3795 if (code[1 + IMM2_SIZE] == OP_PROP || code[1 + IMM2_SIZE] == OP_NOTPROP)
3799 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
3801 code += GET(code, 1);
3813 /* Add in the fixed length from the table */
3815 code += PRIV(OP_lengths)[c];
3817 /* In UTF-8 mode, opcodes that are followed by a character may be followed by
3818 a multi-byte character. The length in the table is a minimum, so we have to
3819 arrange to skip the extra bytes. */
3821 #if defined SUPPORT_UTF && !defined COMPILE_PCRE32
3859 case OP_NOTMINQUERY:
3865 case OP_NOTPOSQUERY:
3868 case OP_NOTMINSTARI:
3870 case OP_NOTMINPLUSI:
3872 case OP_NOTMINQUERYI:
3874 case OP_NOTMINUPTOI:
3876 case OP_NOTPOSSTARI:
3877 case OP_NOTPOSPLUSI:
3878 case OP_NOTPOSQUERYI:
3879 case OP_NOTPOSUPTOI:
3880 if (HAS_EXTRALEN(code[-1])) code += GET_EXTRALEN(code[-1]);
3884 (void)(utf); /* Keep compiler happy by referencing function argument */
3891 /*************************************************
3892 * Check for POSIX class syntax *
3893 *************************************************/
3895 /* This function is called when the sequence "[:" or "[." or "[=" is
3896 encountered in a character class. It checks whether this is followed by a
3897 sequence of characters terminated by a matching ":]" or ".]" or "=]". If we
3898 reach an unescaped ']' without the special preceding character, return FALSE.
3900 Originally, this function only recognized a sequence of letters between the
3901 terminators, but it seems that Perl recognizes any sequence of characters,
3902 though of course unknown POSIX names are subsequently rejected. Perl gives an
3903 "Unknown POSIX class" error for [:f\oo:] for example, where previously PCRE
3904 didn't consider this to be a POSIX class. Likewise for [:1234:].
3906 The problem in trying to be exactly like Perl is in the handling of escapes. We
3907 have to be sure that [abc[:x\]pqr] is *not* treated as containing a POSIX
3908 class, but [abc[:x\]pqr:]] is (so that an error can be generated). The code
3909 below handles the special cases \\ and \], but does not try to do any other
3910 escape processing. This makes it different from Perl for cases such as
3911 [:l\ower:] where Perl recognizes it as the POSIX class "lower" but PCRE does
3912 not recognize "l\ower". This is a lesser evil than not diagnosing bad classes
3913 when Perl does, I think.
3915 A user pointed out that PCRE was rejecting [:a[:digit:]] whereas Perl was not.
3916 It seems that the appearance of a nested POSIX class supersedes an apparent
3917 external class. For example, [:a[:digit:]b:] matches "a", "b", ":", or
3920 In Perl, unescaped square brackets may also appear as part of class names. For
3921 example, [:a[:abc]b:] gives unknown POSIX class "[:abc]b:]". However, for
3922 [:a[:abc]b][b:] it gives unknown POSIX class "[:abc]b][b:]", which does not
3923 seem right at all. PCRE does not allow closing square brackets in POSIX class
3927 ptr pointer to the initial [
3928 endptr where to return the end pointer
3930 Returns: TRUE or FALSE
3934 check_posix_syntax(const pcre_uchar *ptr, const pcre_uchar **endptr)
3936 pcre_uchar terminator; /* Don't combine these lines; the Solaris cc */
3937 terminator = *(++ptr); /* compiler warns about "non-constant" initializer. */
3938 for (++ptr; *ptr != CHAR_NULL; ptr++)
3940 if (*ptr == CHAR_BACKSLASH &&
3941 (ptr[1] == CHAR_RIGHT_SQUARE_BRACKET ||
3942 ptr[1] == CHAR_BACKSLASH))
3944 else if ((*ptr == CHAR_LEFT_SQUARE_BRACKET && ptr[1] == terminator) ||
3945 *ptr == CHAR_RIGHT_SQUARE_BRACKET) return FALSE;
3946 else if (*ptr == terminator && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET)
3958 /*************************************************
3959 * Check POSIX class name *
3960 *************************************************/
3962 /* This function is called to check the name given in a POSIX-style class entry
3966 ptr points to the first letter
3967 len the length of the name
3969 Returns: a value representing the name, or -1 if unknown
3973 check_posix_name(const pcre_uchar *ptr, int len)
3975 const char *pn = posix_names;
3976 register int yield = 0;
3977 while (posix_name_lengths[yield] != 0)
3979 if (len == posix_name_lengths[yield] &&
3980 STRNCMP_UC_C8(ptr, pn, (unsigned int)len) == 0) return yield;
3981 pn += posix_name_lengths[yield] + 1;
3988 /*************************************************
3989 * Adjust OP_RECURSE items in repeated group *
3990 *************************************************/
3992 /* OP_RECURSE items contain an offset from the start of the regex to the group
3993 that is referenced. This means that groups can be replicated for fixed
3994 repetition simply by copying (because the recursion is allowed to refer to
3995 earlier groups that are outside the current group). However, when a group is
3996 optional (i.e. the minimum quantifier is zero), OP_BRAZERO or OP_SKIPZERO is
3997 inserted before it, after it has been compiled. This means that any OP_RECURSE
3998 items within it that refer to the group itself or any contained groups have to
3999 have their offsets adjusted. That one of the jobs of this function. Before it
4000 is called, the partially compiled regex must be temporarily terminated with
4003 This function has been extended to cope with forward references for recursions
4004 and subroutine calls. It must check the list of such references for the
4005 group we are dealing with. If it finds that one of the recursions in the
4006 current group is on this list, it does not adjust the value in the reference
4007 (which is a group number). After the group has been scanned, all the offsets in
4008 the forward reference list for the group are adjusted.
4011 group points to the start of the group
4012 adjust the amount by which the group is to be moved
4013 utf TRUE in UTF-8 / UTF-16 / UTF-32 mode
4014 cd contains pointers to tables etc.
4015 save_hwm_offset the hwm forward reference offset at the start of the group
4021 adjust_recurse(pcre_uchar *group, int adjust, BOOL utf, compile_data *cd,
4022 size_t save_hwm_offset)
4026 pcre_uchar *ptr = group;
4028 while ((ptr = (pcre_uchar *)find_recurse(ptr, utf)) != NULL)
4030 for (hc = (pcre_uchar *)cd->start_workspace + save_hwm_offset; hc < cd->hwm;
4033 offset = (int)GET(hc, 0);
4034 if (cd->start_code + offset == ptr + 1) break;
4037 /* If we have not found this recursion on the forward reference list, adjust
4038 the recursion's offset if it's after the start of this group. */
4042 offset = (int)GET(ptr, 1);
4043 if (cd->start_code + offset >= group) PUT(ptr, 1, offset + adjust);
4046 ptr += 1 + LINK_SIZE;
4049 /* Now adjust all forward reference offsets for the group. */
4051 for (hc = (pcre_uchar *)cd->start_workspace + save_hwm_offset; hc < cd->hwm;
4054 offset = (int)GET(hc, 0);
4055 PUT(hc, 0, offset + adjust);
4061 /*************************************************
4062 * Insert an automatic callout point *
4063 *************************************************/
4065 /* This function is called when the PCRE_AUTO_CALLOUT option is set, to insert
4066 callout points before each pattern item.
4069 code current code pointer
4070 ptr current pattern pointer
4071 cd pointers to tables etc
4073 Returns: new code pointer
4077 auto_callout(pcre_uchar *code, const pcre_uchar *ptr, compile_data *cd)
4079 *code++ = OP_CALLOUT;
4081 PUT(code, 0, (int)(ptr - cd->start_pattern)); /* Pattern offset */
4082 PUT(code, LINK_SIZE, 0); /* Default length */
4083 return code + 2 * LINK_SIZE;
4088 /*************************************************
4089 * Complete a callout item *
4090 *************************************************/
4092 /* A callout item contains the length of the next item in the pattern, which
4093 we can't fill in till after we have reached the relevant point. This is used
4094 for both automatic and manual callouts.
4097 previous_callout points to previous callout item
4098 ptr current pattern pointer
4099 cd pointers to tables etc
4105 complete_callout(pcre_uchar *previous_callout, const pcre_uchar *ptr, compile_data *cd)
4107 int length = (int)(ptr - cd->start_pattern - GET(previous_callout, 2));
4108 PUT(previous_callout, 2 + LINK_SIZE, length);
4114 /*************************************************
4115 * Get othercase range *
4116 *************************************************/
4118 /* This function is passed the start and end of a class range, in UTF-8 mode
4119 with UCP support. It searches up the characters, looking for ranges of
4120 characters in the "other" case. Each call returns the next one, updating the
4121 start address. A character with multiple other cases is returned on its own
4122 with a special return value.
4125 cptr points to starting character value; updated
4127 ocptr where to put start of othercase range
4128 odptr where to put end of othercase range
4130 Yield: -1 when no more
4131 0 when a range is returned
4132 >0 the CASESET offset for char with multiple other cases
4133 in this case, ocptr contains the original
4137 get_othercase_range(pcre_uint32 *cptr, pcre_uint32 d, pcre_uint32 *ocptr,
4140 pcre_uint32 c, othercase, next;
4143 /* Find the first character that has an other case. If it has multiple other
4144 cases, return its case offset value. */
4146 for (c = *cptr; c <= d; c++)
4148 if ((co = UCD_CASESET(c)) != 0)
4150 *ocptr = c++; /* Character that has the set */
4151 *cptr = c; /* Rest of input range */
4154 if ((othercase = UCD_OTHERCASE(c)) != c) break;
4157 if (c > d) return -1; /* Reached end of range */
4159 /* Found a character that has a single other case. Search for the end of the
4160 range, which is either the end of the input range, or a character that has zero
4161 or more than one other cases. */
4164 next = othercase + 1;
4166 for (++c; c <= d; c++)
4168 if ((co = UCD_CASESET(c)) != 0 || UCD_OTHERCASE(c) != next) break;
4172 *odptr = next - 1; /* End of othercase range */
4173 *cptr = c; /* Rest of input range */
4176 #endif /* SUPPORT_UCP */
4180 /*************************************************
4181 * Add a character or range to a class *
4182 *************************************************/
4184 /* This function packages up the logic of adding a character or range of
4185 characters to a class. The character values in the arguments will be within the
4186 valid values for the current mode (8-bit, 16-bit, UTF, etc). This function is
4187 mutually recursive with the function immediately below.
4190 classbits the bit map for characters < 256
4191 uchardptr points to the pointer for extra data
4192 options the options word
4193 cd contains pointers to tables etc.
4194 start start of range character
4195 end end of range character
4197 Returns: the number of < 256 characters added
4198 the pointer to extra data is updated
4202 add_to_class(pcre_uint8 *classbits, pcre_uchar **uchardptr, int options,
4203 compile_data *cd, pcre_uint32 start, pcre_uint32 end)
4206 pcre_uint32 classbits_end = (end <= 0xff ? end : 0xff);
4209 /* If caseless matching is required, scan the range and process alternate
4210 cases. In Unicode, there are 8-bit characters that have alternate cases that
4211 are greater than 255 and vice-versa. Sometimes we can just extend the original
4214 if ((options & PCRE_CASELESS) != 0)
4217 if ((options & PCRE_UTF8) != 0)
4222 options &= ~PCRE_CASELESS; /* Remove for recursive calls */
4225 while ((rc = get_othercase_range(&c, end, &oc, &od)) >= 0)
4227 /* Handle a single character that has more than one other case. */
4229 if (rc > 0) n8 += add_list_to_class(classbits, uchardptr, options, cd,
4230 PRIV(ucd_caseless_sets) + rc, oc);
4232 /* Do nothing if the other case range is within the original range. */
4234 else if (oc >= start && od <= end) continue;
4236 /* Extend the original range if there is overlap, noting that if oc < c, we
4237 can't have od > end because a subrange is always shorter than the basic
4238 range. Otherwise, use a recursive call to add the additional range. */
4240 else if (oc < start && od >= start - 1) start = oc; /* Extend downwards */
4241 else if (od > end && oc <= end + 1)
4243 end = od; /* Extend upwards */
4244 if (end > classbits_end) classbits_end = (end <= 0xff ? end : 0xff);
4246 else n8 += add_to_class(classbits, uchardptr, options, cd, oc, od);
4250 #endif /* SUPPORT_UCP */
4252 /* Not UTF-mode, or no UCP */
4254 for (c = start; c <= classbits_end; c++)
4256 SETBIT(classbits, cd->fcc[c]);
4261 /* Now handle the original range. Adjust the final value according to the bit
4262 length - this means that the same lists of (e.g.) horizontal spaces can be used
4265 #if defined COMPILE_PCRE8
4267 if ((options & PCRE_UTF8) == 0)
4269 if (end > 0xff) end = 0xff;
4271 #elif defined COMPILE_PCRE16
4273 if ((options & PCRE_UTF16) == 0)
4275 if (end > 0xffff) end = 0xffff;
4277 #endif /* COMPILE_PCRE[8|16] */
4279 /* Use the bitmap for characters < 256. Otherwise use extra data.*/
4281 for (c = start; c <= classbits_end; c++)
4283 /* Regardless of start, c will always be <= 255. */
4284 SETBIT(classbits, c);
4288 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
4289 if (start <= 0xff) start = 0xff + 1;
4293 pcre_uchar *uchardata = *uchardptr;
4295 if ((options & PCRE_UTF8) != 0) /* All UTFs use the same flag bit */
4299 *uchardata++ = XCL_RANGE;
4300 uchardata += PRIV(ord2utf)(start, uchardata);
4301 uchardata += PRIV(ord2utf)(end, uchardata);
4303 else if (start == end)
4305 *uchardata++ = XCL_SINGLE;
4306 uchardata += PRIV(ord2utf)(start, uchardata);
4310 #endif /* SUPPORT_UTF */
4312 /* Without UTF support, character values are constrained by the bit length,
4313 and can only be > 256 for 16-bit and 32-bit libraries. */
4315 #ifdef COMPILE_PCRE8
4320 *uchardata++ = XCL_RANGE;
4321 *uchardata++ = start;
4324 else if (start == end)
4326 *uchardata++ = XCL_SINGLE;
4327 *uchardata++ = start;
4331 *uchardptr = uchardata; /* Updata extra data pointer */
4333 #endif /* SUPPORT_UTF || !COMPILE_PCRE8 */
4335 return n8; /* Number of 8-bit characters */
4341 /*************************************************
4342 * Add a list of characters to a class *
4343 *************************************************/
4345 /* This function is used for adding a list of case-equivalent characters to a
4346 class, and also for adding a list of horizontal or vertical whitespace. If the
4347 list is in order (which it should be), ranges of characters are detected and
4348 handled appropriately. This function is mutually recursive with the function
4352 classbits the bit map for characters < 256
4353 uchardptr points to the pointer for extra data
4354 options the options word
4355 cd contains pointers to tables etc.
4356 p points to row of 32-bit values, terminated by NOTACHAR
4357 except character to omit; this is used when adding lists of
4358 case-equivalent characters to avoid including the one we
4361 Returns: the number of < 256 characters added
4362 the pointer to extra data is updated
4366 add_list_to_class(pcre_uint8 *classbits, pcre_uchar **uchardptr, int options,
4367 compile_data *cd, const pcre_uint32 *p, unsigned int except)
4370 while (p[0] < NOTACHAR)
4375 while(p[n+1] == p[0] + n + 1) n++;
4376 n8 += add_to_class(classbits, uchardptr, options, cd, p[0], p[n]);
4385 /*************************************************
4386 * Add characters not in a list to a class *
4387 *************************************************/
4389 /* This function is used for adding the complement of a list of horizontal or
4390 vertical whitespace to a class. The list must be in order.
4393 classbits the bit map for characters < 256
4394 uchardptr points to the pointer for extra data
4395 options the options word
4396 cd contains pointers to tables etc.
4397 p points to row of 32-bit values, terminated by NOTACHAR
4399 Returns: the number of < 256 characters added
4400 the pointer to extra data is updated
4404 add_not_list_to_class(pcre_uint8 *classbits, pcre_uchar **uchardptr,
4405 int options, compile_data *cd, const pcre_uint32 *p)
4407 BOOL utf = (options & PCRE_UTF8) != 0;
4410 n8 += add_to_class(classbits, uchardptr, options, cd, 0, p[0] - 1);
4411 while (p[0] < NOTACHAR)
4413 while (p[1] == p[0] + 1) p++;
4414 n8 += add_to_class(classbits, uchardptr, options, cd, p[0] + 1,
4415 (p[1] == NOTACHAR) ? (utf ? 0x10ffffu : 0xffffffffu) : p[1] - 1);
4423 /*************************************************
4424 * Compile one branch *
4425 *************************************************/
4427 /* Scan the pattern, compiling it into the a vector. If the options are
4428 changed during the branch, the pointer is used to change the external options
4429 bits. This function is used during the pre-compile phase when we are trying
4430 to find out the amount of memory needed, as well as during the real compile
4431 phase. The value of lengthptr distinguishes the two phases.
4434 optionsptr pointer to the option bits
4435 codeptr points to the pointer to the current code point
4436 ptrptr points to the current pattern pointer
4437 errorcodeptr points to error code variable
4438 firstcharptr place to put the first required character
4439 firstcharflagsptr place to put the first character flags, or a negative number
4440 reqcharptr place to put the last required character
4441 reqcharflagsptr place to put the last required character flags, or a negative number
4442 bcptr points to current branch chain
4443 cond_depth conditional nesting depth
4444 cd contains pointers to tables etc.
4445 lengthptr NULL during the real compile phase
4446 points to length accumulator during pre-compile phase
4448 Returns: TRUE on success
4449 FALSE, with *errorcodeptr set non-zero on error
4453 compile_branch(int *optionsptr, pcre_uchar **codeptr,
4454 const pcre_uchar **ptrptr, int *errorcodeptr,
4455 pcre_uint32 *firstcharptr, pcre_int32 *firstcharflagsptr,
4456 pcre_uint32 *reqcharptr, pcre_int32 *reqcharflagsptr,
4457 branch_chain *bcptr, int cond_depth,
4458 compile_data *cd, int *lengthptr)
4460 int repeat_type, op_type;
4461 int repeat_min = 0, repeat_max = 0; /* To please picky compilers */
4463 int greedy_default, greedy_non_default;
4464 pcre_uint32 firstchar, reqchar;
4465 pcre_int32 firstcharflags, reqcharflags;
4466 pcre_uint32 zeroreqchar, zerofirstchar;
4467 pcre_int32 zeroreqcharflags, zerofirstcharflags;
4468 pcre_int32 req_caseopt, reqvary, tempreqvary;
4469 int options = *optionsptr; /* May change dynamically */
4470 int after_manual_callout = 0;
4471 int length_prevgroup = 0;
4472 register pcre_uint32 c;
4474 register pcre_uchar *code = *codeptr;
4475 pcre_uchar *last_code = code;
4476 pcre_uchar *orig_code = code;
4477 pcre_uchar *tempcode;
4478 BOOL inescq = FALSE;
4479 BOOL groupsetfirstchar = FALSE;
4480 const pcre_uchar *ptr = *ptrptr;
4481 const pcre_uchar *tempptr;
4482 const pcre_uchar *nestptr = NULL;
4483 pcre_uchar *previous = NULL;
4484 pcre_uchar *previous_callout = NULL;
4485 size_t item_hwm_offset = 0;
4486 pcre_uint8 classbits[32];
4488 /* We can fish out the UTF-8 setting once and for all into a BOOL, but we
4489 must not do this for other options (e.g. PCRE_EXTENDED) because they may change
4490 dynamically as we process the pattern. */
4493 /* PCRE_UTF[16|32] have the same value as PCRE_UTF8. */
4494 BOOL utf = (options & PCRE_UTF8) != 0;
4495 #ifndef COMPILE_PCRE32
4496 pcre_uchar utf_chars[6];
4502 /* Helper variables for OP_XCLASS opcode (for characters > 255). We define
4503 class_uchardata always so that it can be passed to add_to_class() always,
4504 though it will not be used in non-UTF 8-bit cases. This avoids having to supply
4505 alternative calls for the different cases. */
4507 pcre_uchar *class_uchardata;
4508 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
4510 pcre_uchar *class_uchardata_base;
4514 if (lengthptr != NULL) DPRINTF((">> start branch\n"));
4517 /* Set up the default and non-default settings for greediness */
4519 greedy_default = ((options & PCRE_UNGREEDY) != 0);
4520 greedy_non_default = greedy_default ^ 1;
4522 /* Initialize no first byte, no required byte. REQ_UNSET means "no char
4523 matching encountered yet". It gets changed to REQ_NONE if we hit something that
4524 matches a non-fixed char first char; reqchar just remains unset if we never
4527 When we hit a repeat whose minimum is zero, we may have to adjust these values
4528 to take the zero repeat into account. This is implemented by setting them to
4529 zerofirstbyte and zeroreqchar when such a repeat is encountered. The individual
4530 item types that can be repeated set these backoff variables appropriately. */
4532 firstchar = reqchar = zerofirstchar = zeroreqchar = 0;
4533 firstcharflags = reqcharflags = zerofirstcharflags = zeroreqcharflags = REQ_UNSET;
4535 /* The variable req_caseopt contains either the REQ_CASELESS value
4536 or zero, according to the current setting of the caseless flag. The
4537 REQ_CASELESS leaves the lower 28 bit empty. It is added into the
4538 firstchar or reqchar variables to record the case status of the
4539 value. This is used only for ASCII characters. */
4541 req_caseopt = ((options & PCRE_CASELESS) != 0)? REQ_CASELESS:0;
4543 /* Switch on next character until the end of the branch */
4548 BOOL should_flip_negation;
4549 BOOL possessive_quantifier;
4552 BOOL reset_bracount;
4553 int class_has_8bitchar;
4555 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
4556 BOOL xclass_has_prop;
4562 pcre_uint32 subreqchar, subfirstchar;
4563 pcre_int32 subreqcharflags, subfirstcharflags;
4565 unsigned int mclength;
4566 unsigned int tempbracount;
4568 pcre_uchar mcbuffer[8];
4570 /* Come here to restart the loop without advancing the pointer. */
4574 /* Get next character in the pattern */
4578 /* If we are at the end of a nested substitution, revert to the outer level
4579 string. Nesting only happens one level deep. */
4581 if (c == CHAR_NULL && nestptr != NULL)
4588 /* If we are in the pre-compile phase, accumulate the length used for the
4589 previous cycle of this loop. */
4591 if (lengthptr != NULL)
4594 if (code > cd->hwm) cd->hwm = code; /* High water info */
4596 if (code > cd->start_workspace + cd->workspace_size -
4597 WORK_SIZE_SAFETY_MARGIN) /* Check for overrun */
4599 *errorcodeptr = (code >= cd->start_workspace + cd->workspace_size)?
4604 /* There is at least one situation where code goes backwards: this is the
4605 case of a zero quantifier after a class (e.g. [ab]{0}). At compile time,
4606 the class is simply eliminated. However, it is created first, so we have to
4607 allow memory for it. Therefore, don't ever reduce the length at this point.
4610 if (code < last_code) code = last_code;
4612 /* Paranoid check for integer overflow */
4614 if (OFLOW_MAX - *lengthptr < code - last_code)
4616 *errorcodeptr = ERR20;
4620 *lengthptr += (int)(code - last_code);
4621 DPRINTF(("length=%d added %d c=%c (0x%x)\n", *lengthptr,
4622 (int)(code - last_code), c, c));
4624 /* If "previous" is set and it is not at the start of the work space, move
4625 it back to there, in order to avoid filling up the work space. Otherwise,
4626 if "previous" is NULL, reset the current code pointer to the start. */
4628 if (previous != NULL)
4630 if (previous > orig_code)
4632 memmove(orig_code, previous, IN_UCHARS(code - previous));
4633 code -= previous - orig_code;
4634 previous = orig_code;
4637 else code = orig_code;
4639 /* Remember where this code item starts so we can pick up the length
4645 /* In the real compile phase, just check the workspace used by the forward
4648 else if (cd->hwm > cd->start_workspace + cd->workspace_size)
4650 *errorcodeptr = ERR52;
4654 /* If in \Q...\E, check for the end; if not, we have a literal. Otherwise an
4655 isolated \E is ignored. */
4659 if (c == CHAR_BACKSLASH && ptr[1] == CHAR_E)
4667 if (previous_callout != NULL)
4669 if (lengthptr == NULL) /* Don't attempt in pre-compile phase */
4670 complete_callout(previous_callout, ptr, cd);
4671 previous_callout = NULL;
4673 if ((options & PCRE_AUTO_CALLOUT) != 0)
4675 previous_callout = code;
4676 code = auto_callout(code, ptr, cd);
4681 /* Check for the start of a \Q...\E sequence. We must do this here rather
4682 than later in case it is immediately followed by \E, which turns it into a
4683 "do nothing" sequence. */
4685 if (c == CHAR_BACKSLASH && ptr[1] == CHAR_Q)
4693 /* In extended mode, skip white space and comments. */
4695 if ((options & PCRE_EXTENDED) != 0)
4697 const pcre_uchar *wscptr = ptr;
4698 while (MAX_255(c) && (cd->ctypes[c] & ctype_space) != 0) c = *(++ptr);
4699 if (c == CHAR_NUMBER_SIGN)
4702 while (*ptr != CHAR_NULL)
4704 if (IS_NEWLINE(ptr)) /* For non-fixed-length newline cases, */
4705 { /* IS_NEWLINE sets cd->nllen. */
4711 if (utf) FORWARDCHAR(ptr);
4716 /* If we skipped any characters, restart the loop. Otherwise, we didn't see
4719 if (ptr > wscptr) goto REDO_LOOP;
4722 /* Skip over (?# comments. We need to do this here because we want to know if
4723 the next thing is a quantifier, and these comments may come between an item
4724 and its quantifier. */
4726 if (c == CHAR_LEFT_PARENTHESIS && ptr[1] == CHAR_QUESTION_MARK &&
4727 ptr[2] == CHAR_NUMBER_SIGN)
4730 while (*ptr != CHAR_NULL && *ptr != CHAR_RIGHT_PARENTHESIS) ptr++;
4731 if (*ptr == CHAR_NULL)
4733 *errorcodeptr = ERR18;
4739 /* See if the next thing is a quantifier. */
4742 c == CHAR_ASTERISK || c == CHAR_PLUS || c == CHAR_QUESTION_MARK ||
4743 (c == CHAR_LEFT_CURLY_BRACKET && is_counted_repeat(ptr+1));
4745 /* Fill in length of a previous callout, except when the next thing is a
4746 quantifier or when processing a property substitution string in UCP mode. */
4748 if (!is_quantifier && previous_callout != NULL && nestptr == NULL &&
4749 after_manual_callout-- <= 0)
4751 if (lengthptr == NULL) /* Don't attempt in pre-compile phase */
4752 complete_callout(previous_callout, ptr, cd);
4753 previous_callout = NULL;
4756 /* Create auto callout, except for quantifiers, or while processing property
4757 strings that are substituted for \w etc in UCP mode. */
4759 if ((options & PCRE_AUTO_CALLOUT) != 0 && !is_quantifier && nestptr == NULL)
4761 previous_callout = code;
4762 code = auto_callout(code, ptr, cd);
4765 /* Process the next pattern item. */
4769 /* ===================================================================*/
4770 case CHAR_NULL: /* The branch terminates at string end */
4771 case CHAR_VERTICAL_LINE: /* or | or ) */
4772 case CHAR_RIGHT_PARENTHESIS:
4773 *firstcharptr = firstchar;
4774 *firstcharflagsptr = firstcharflags;
4775 *reqcharptr = reqchar;
4776 *reqcharflagsptr = reqcharflags;
4779 if (lengthptr != NULL)
4781 if (OFLOW_MAX - *lengthptr < code - last_code)
4783 *errorcodeptr = ERR20;
4786 *lengthptr += (int)(code - last_code); /* To include callout length */
4787 DPRINTF((">> end branch\n"));
4792 /* ===================================================================*/
4793 /* Handle single-character metacharacters. In multiline mode, ^ disables
4794 the setting of any following char as a first character. */
4796 case CHAR_CIRCUMFLEX_ACCENT:
4798 if ((options & PCRE_MULTILINE) != 0)
4800 if (firstcharflags == REQ_UNSET)
4801 zerofirstcharflags = firstcharflags = REQ_NONE;
4804 else *code++ = OP_CIRC;
4807 case CHAR_DOLLAR_SIGN:
4809 *code++ = ((options & PCRE_MULTILINE) != 0)? OP_DOLLM : OP_DOLL;
4812 /* There can never be a first char if '.' is first, whatever happens about
4813 repeats. The value of reqchar doesn't change either. */
4816 if (firstcharflags == REQ_UNSET) firstcharflags = REQ_NONE;
4817 zerofirstchar = firstchar;
4818 zerofirstcharflags = firstcharflags;
4819 zeroreqchar = reqchar;
4820 zeroreqcharflags = reqcharflags;
4822 item_hwm_offset = cd->hwm - cd->start_workspace;
4823 *code++ = ((options & PCRE_DOTALL) != 0)? OP_ALLANY: OP_ANY;
4827 /* ===================================================================*/
4828 /* Character classes. If the included characters are all < 256, we build a
4829 32-byte bitmap of the permitted characters, except in the special case
4830 where there is only one such character. For negated classes, we build the
4831 map as usual, then invert it at the end. However, we use a different opcode
4832 so that data characters > 255 can be handled correctly.
4834 If the class contains characters outside the 0-255 range, a different
4835 opcode is compiled. It may optionally have a bit map for characters < 256,
4836 but those above are are explicitly listed afterwards. A flag byte tells
4837 whether the bitmap is present, and whether this is a negated class or not.
4839 In JavaScript compatibility mode, an isolated ']' causes an error. In
4840 default (Perl) mode, it is treated as a data character. */
4842 case CHAR_RIGHT_SQUARE_BRACKET:
4843 if ((cd->external_options & PCRE_JAVASCRIPT_COMPAT) != 0)
4845 *errorcodeptr = ERR64;
4850 /* In another (POSIX) regex library, the ugly syntax [[:<:]] and [[:>:]] is
4851 used for "start of word" and "end of word". As these are otherwise illegal
4852 sequences, we don't break anything by recognizing them. They are replaced
4853 by \b(?=\w) and \b(?<=\w) respectively. Sequences like [a[:<:]] are
4854 erroneous and are handled by the normal code below. */
4856 case CHAR_LEFT_SQUARE_BRACKET:
4857 if (STRNCMP_UC_C8(ptr+1, STRING_WEIRD_STARTWORD, 6) == 0)
4860 ptr = sub_start_of_word;
4864 if (STRNCMP_UC_C8(ptr+1, STRING_WEIRD_ENDWORD, 6) == 0)
4867 ptr = sub_end_of_word;
4871 /* Handle a real character class. */
4874 item_hwm_offset = cd->hwm - cd->start_workspace;
4876 /* PCRE supports POSIX class stuff inside a class. Perl gives an error if
4877 they are encountered at the top level, so we'll do that too. */
4879 if ((ptr[1] == CHAR_COLON || ptr[1] == CHAR_DOT ||
4880 ptr[1] == CHAR_EQUALS_SIGN) &&
4881 check_posix_syntax(ptr, &tempptr))
4883 *errorcodeptr = (ptr[1] == CHAR_COLON)? ERR13 : ERR31;
4887 /* If the first character is '^', set the negation flag and skip it. Also,
4888 if the first few characters (either before or after ^) are \Q\E or \E we
4889 skip them too. This makes for compatibility with Perl. */
4891 negate_class = FALSE;
4895 if (c == CHAR_BACKSLASH)
4897 if (ptr[1] == CHAR_E)
4899 else if (STRNCMP_UC_C8(ptr + 1, STR_Q STR_BACKSLASH STR_E, 3) == 0)
4904 else if (!negate_class && c == CHAR_CIRCUMFLEX_ACCENT)
4905 negate_class = TRUE;
4909 /* Empty classes are allowed in JavaScript compatibility mode. Otherwise,
4910 an initial ']' is taken as a data character -- the code below handles
4911 that. In JS mode, [] must always fail, so generate OP_FAIL, whereas
4912 [^] must match any character, so generate OP_ALLANY. */
4914 if (c == CHAR_RIGHT_SQUARE_BRACKET &&
4915 (cd->external_options & PCRE_JAVASCRIPT_COMPAT) != 0)
4917 *code++ = negate_class? OP_ALLANY : OP_FAIL;
4918 if (firstcharflags == REQ_UNSET) firstcharflags = REQ_NONE;
4919 zerofirstchar = firstchar;
4920 zerofirstcharflags = firstcharflags;
4924 /* If a class contains a negative special such as \S, we need to flip the
4925 negation flag at the end, so that support for characters > 255 works
4926 correctly (they are all included in the class). */
4928 should_flip_negation = FALSE;
4930 /* Extended class (xclass) will be used when characters > 255
4933 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
4935 class_uchardata = code + LINK_SIZE + 2; /* For XCLASS items */
4936 class_uchardata_base = class_uchardata; /* Save the start */
4939 /* For optimization purposes, we track some properties of the class:
4940 class_has_8bitchar will be non-zero if the class contains at least one <
4941 256 character; class_one_char will be 1 if the class contains just one
4942 character; xclass_has_prop will be TRUE if unicode property checks
4943 are present in the class. */
4945 class_has_8bitchar = 0;
4947 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
4948 xclass_has_prop = FALSE;
4951 /* Initialize the 32-char bit map to all zeros. We build the map in a
4952 temporary bit of memory, in case the class contains fewer than two
4953 8-bit characters because in that case the compiled code doesn't use the bit
4956 memset(classbits, 0, 32 * sizeof(pcre_uint8));
4958 /* Process characters until ] is reached. By writing this as a "do" it
4959 means that an initial ] is taken as a data character. At the start of the
4960 loop, c contains the first byte of the character. */
4962 if (c != CHAR_NULL) do
4964 const pcre_uchar *oldptr;
4967 if (utf && HAS_EXTRALEN(c))
4968 { /* Braces are required because the */
4969 GETCHARLEN(c, ptr, ptr); /* macro generates multiple statements */
4973 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
4974 /* In the pre-compile phase, accumulate the length of any extra
4975 data and reset the pointer. This is so that very large classes that
4976 contain a zillion > 255 characters no longer overwrite the work space
4977 (which is on the stack). We have to remember that there was XCLASS data,
4980 if (class_uchardata > class_uchardata_base) xclass = TRUE;
4982 if (lengthptr != NULL && class_uchardata > class_uchardata_base)
4984 *lengthptr += (int)(class_uchardata - class_uchardata_base);
4985 class_uchardata = class_uchardata_base;
4989 /* Inside \Q...\E everything is literal except \E */
4993 if (c == CHAR_BACKSLASH && ptr[1] == CHAR_E) /* If we are at \E */
4995 inescq = FALSE; /* Reset literal state */
4996 ptr++; /* Skip the 'E' */
4997 continue; /* Carry on with next */
4999 goto CHECK_RANGE; /* Could be range if \E follows */
5002 /* Handle POSIX class names. Perl allows a negation extension of the
5003 form [:^name:]. A square bracket that doesn't match the syntax is
5004 treated as a literal. We also recognize the POSIX constructions
5005 [.ch.] and [=ch=] ("collating elements") and fault them, as Perl
5008 if (c == CHAR_LEFT_SQUARE_BRACKET &&
5009 (ptr[1] == CHAR_COLON || ptr[1] == CHAR_DOT ||
5010 ptr[1] == CHAR_EQUALS_SIGN) && check_posix_syntax(ptr, &tempptr))
5012 BOOL local_negate = FALSE;
5013 int posix_class, taboffset, tabopt;
5014 register const pcre_uint8 *cbits = cd->cbits;
5015 pcre_uint8 pbits[32];
5017 if (ptr[1] != CHAR_COLON)
5019 *errorcodeptr = ERR31;
5024 if (*ptr == CHAR_CIRCUMFLEX_ACCENT)
5026 local_negate = TRUE;
5027 should_flip_negation = TRUE; /* Note negative special */
5031 posix_class = check_posix_name(ptr, (int)(tempptr - ptr));
5032 if (posix_class < 0)
5034 *errorcodeptr = ERR30;
5038 /* If matching is caseless, upper and lower are converted to
5039 alpha. This relies on the fact that the class table starts with
5040 alpha, lower, upper as the first 3 entries. */
5042 if ((options & PCRE_CASELESS) != 0 && posix_class <= 2)
5045 /* When PCRE_UCP is set, some of the POSIX classes are converted to
5046 different escape sequences that use Unicode properties \p or \P. Others
5047 that are not available via \p or \P generate XCL_PROP/XCL_NOTPROP
5051 if ((options & PCRE_UCP) != 0)
5053 unsigned int ptype = 0;
5054 int pc = posix_class + ((local_negate)? POSIX_SUBSIZE/2 : 0);
5056 /* The posix_substitutes table specifies which POSIX classes can be
5057 converted to \p or \P items. */
5059 if (posix_substitutes[pc] != NULL)
5061 nestptr = tempptr + 1;
5062 ptr = posix_substitutes[pc] - 1;
5066 /* There are three other classes that generate special property calls
5067 that are recognized only in an XCLASS. */
5069 else switch(posix_class)
5075 if (ptype == 0) ptype = PT_PXPRINT;
5078 if (ptype == 0) ptype = PT_PXPUNCT;
5079 *class_uchardata++ = local_negate? XCL_NOTPROP : XCL_PROP;
5080 *class_uchardata++ = ptype;
5081 *class_uchardata++ = 0;
5082 xclass_has_prop = TRUE;
5086 /* For the other POSIX classes (ascii, cntrl, xdigit) we are going
5087 to fall through to the non-UCP case and build a bit map for
5088 characters with code points less than 256. If we are in a negated
5089 POSIX class, characters with code points greater than 255 must
5090 either all match or all not match. In the special case where we
5091 have not yet generated any xclass data, and this is the final item
5092 in the overall class, we need do nothing: later on, the opcode
5093 OP_NCLASS will be used to indicate that characters greater than 255
5094 are acceptable. If we have already seen an xclass item or one may
5095 follow (we have to assume that it might if this is not the end of
5096 the class), explicitly list all wide codepoints, which will then
5097 either not match or match, depending on whether the class is or is
5102 (xclass || tempptr[2] != CHAR_RIGHT_SQUARE_BRACKET))
5104 *class_uchardata++ = XCL_RANGE;
5105 class_uchardata += PRIV(ord2utf)(0x100, class_uchardata);
5106 class_uchardata += PRIV(ord2utf)(0x10ffff, class_uchardata);
5112 /* In the non-UCP case, or when UCP makes no difference, we build the
5113 bit map for the POSIX class in a chunk of local store because we may be
5114 adding and subtracting from it, and we don't want to subtract bits that
5115 may be in the main map already. At the end we or the result into the
5116 bit map that is being built. */
5120 /* Copy in the first table (always present) */
5122 memcpy(pbits, cbits + posix_class_maps[posix_class],
5123 32 * sizeof(pcre_uint8));
5125 /* If there is a second table, add or remove it as required. */
5127 taboffset = posix_class_maps[posix_class + 1];
5128 tabopt = posix_class_maps[posix_class + 2];
5133 for (c = 0; c < 32; c++) pbits[c] |= cbits[c + taboffset];
5135 for (c = 0; c < 32; c++) pbits[c] &= ~cbits[c + taboffset];
5138 /* Now see if we need to remove any special characters. An option
5139 value of 1 removes vertical space and 2 removes underscore. */
5141 if (tabopt < 0) tabopt = -tabopt;
5142 if (tabopt == 1) pbits[1] &= ~0x3c;
5143 else if (tabopt == 2) pbits[11] &= 0x7f;
5145 /* Add the POSIX table or its complement into the main table that is
5146 being built and we are done. */
5149 for (c = 0; c < 32; c++) classbits[c] |= ~pbits[c];
5151 for (c = 0; c < 32; c++) classbits[c] |= pbits[c];
5154 /* Every class contains at least one < 256 character. */
5155 class_has_8bitchar = 1;
5156 /* Every class contains at least two characters. */
5158 continue; /* End of POSIX syntax handling */
5161 /* Backslash may introduce a single character, or it may introduce one
5162 of the specials, which just set a flag. The sequence \b is a special
5163 case. Inside a class (and only there) it is treated as backspace. We
5164 assume that other escapes have more than one character in them, so
5165 speculatively set both class_has_8bitchar and class_one_char bigger
5166 than one. Unrecognized escapes fall through and are either treated
5167 as literal characters (by default), or are faulted if
5168 PCRE_EXTRA is set. */
5170 if (c == CHAR_BACKSLASH)
5172 escape = check_escape(&ptr, &ec, errorcodeptr, cd->bracount, options,
5174 if (*errorcodeptr != 0) goto FAILED;
5175 if (escape == 0) c = ec;
5176 else if (escape == ESC_b) c = CHAR_BS; /* \b is backspace in a class */
5177 else if (escape == ESC_N) /* \N is not supported in a class */
5179 *errorcodeptr = ERR71;
5182 else if (escape == ESC_Q) /* Handle start of quoted string */
5184 if (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
5186 ptr += 2; /* avoid empty string */
5191 else if (escape == ESC_E) continue; /* Ignore orphan \E */
5195 register const pcre_uint8 *cbits = cd->cbits;
5196 /* Every class contains at least two < 256 characters. */
5197 class_has_8bitchar++;
5198 /* Every class contains at least two characters. */
5199 class_one_char += 2;
5204 case ESC_du: /* These are the values given for \d etc */
5205 case ESC_DU: /* when PCRE_UCP is set. We replace the */
5206 case ESC_wu: /* escape sequence with an appropriate \p */
5207 case ESC_WU: /* or \P to test Unicode properties instead */
5208 case ESC_su: /* of the default ASCII testing. */
5211 ptr = substitutes[escape - ESC_DU] - 1; /* Just before substitute */
5212 class_has_8bitchar--; /* Undo! */
5216 for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_digit];
5220 should_flip_negation = TRUE;
5221 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_digit];
5225 for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_word];
5229 should_flip_negation = TRUE;
5230 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_word];
5233 /* Perl 5.004 onwards omitted VT from \s, but restored it at Perl
5234 5.18. Before PCRE 8.34, we had to preserve the VT bit if it was
5235 previously set by something earlier in the character class.
5236 Luckily, the value of CHAR_VT is 0x0b in both ASCII and EBCDIC, so
5237 we could just adjust the appropriate bit. From PCRE 8.34 we no
5238 longer treat \s and \S specially. */
5241 for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_space];
5245 should_flip_negation = TRUE;
5246 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_space];
5249 /* The rest apply in both UCP and non-UCP cases. */
5252 (void)add_list_to_class(classbits, &class_uchardata, options, cd,
5253 PRIV(hspace_list), NOTACHAR);
5257 (void)add_not_list_to_class(classbits, &class_uchardata, options,
5258 cd, PRIV(hspace_list));
5262 (void)add_list_to_class(classbits, &class_uchardata, options, cd,
5263 PRIV(vspace_list), NOTACHAR);
5267 (void)add_not_list_to_class(classbits, &class_uchardata, options,
5268 cd, PRIV(vspace_list));
5276 unsigned int ptype = 0, pdata = 0;
5277 if (!get_ucp(&ptr, &negated, &ptype, &pdata, errorcodeptr))
5279 *class_uchardata++ = ((escape == ESC_p) != negated)?
5280 XCL_PROP : XCL_NOTPROP;
5281 *class_uchardata++ = ptype;
5282 *class_uchardata++ = pdata;
5283 xclass_has_prop = TRUE;
5284 class_has_8bitchar--; /* Undo! */
5288 *errorcodeptr = ERR45;
5291 /* Unrecognized escapes are faulted if PCRE is running in its
5292 strict mode. By default, for compatibility with Perl, they are
5293 treated as literals. */
5296 if ((options & PCRE_EXTRA) != 0)
5298 *errorcodeptr = ERR7;
5301 class_has_8bitchar--; /* Undo the speculative increase. */
5302 class_one_char -= 2; /* Undo the speculative increase. */
5303 c = *ptr; /* Get the final character and fall through */
5308 /* Fall through if the escape just defined a single character (c >= 0).
5309 This may be greater than 256. */
5313 } /* End of backslash handling */
5315 /* A character may be followed by '-' to form a range. However, Perl does
5316 not permit ']' to be the end of the range. A '-' character at the end is
5317 treated as a literal. Perl ignores orphaned \E sequences entirely. The
5318 code for handling \Q and \E is messy. */
5321 while (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
5328 /* Remember if \r or \n were explicitly used */
5330 if (c == CHAR_CR || c == CHAR_NL) cd->external_flags |= PCRE_HASCRORLF;
5332 /* Check for range */
5334 if (!inescq && ptr[1] == CHAR_MINUS)
5338 while (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_E) ptr += 2;
5340 /* If we hit \Q (not followed by \E) at this point, go into escaped
5343 while (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_Q)
5346 if (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_E)
5347 { ptr += 2; continue; }
5352 /* Minus (hyphen) at the end of a class is treated as a literal, so put
5353 back the pointer and jump to handle the character that preceded it. */
5355 if (*ptr == CHAR_NULL || (!inescq && *ptr == CHAR_RIGHT_SQUARE_BRACKET))
5358 goto CLASS_SINGLE_CHARACTER;
5361 /* Otherwise, we have a potential range; pick up the next character */
5365 { /* Braces are required because the */
5366 GETCHARLEN(d, ptr, ptr); /* macro generates multiple statements */
5370 d = *ptr; /* Not UTF-8 mode */
5372 /* The second part of a range can be a single-character escape
5373 sequence, but not any of the other escapes. Perl treats a hyphen as a
5374 literal in such circumstances. However, in Perl's warning mode, a
5375 warning is given, so PCRE now faults it as it is almost certainly a
5376 mistake on the user's part. */
5380 if (d == CHAR_BACKSLASH)
5383 descape = check_escape(&ptr, &d, errorcodeptr, cd->bracount, options, TRUE);
5384 if (*errorcodeptr != 0) goto FAILED;
5386 /* 0 means a character was put into d; \b is backspace; any other
5387 special causes an error. */
5391 if (descape == ESC_b) d = CHAR_BS; else
5393 *errorcodeptr = ERR83;
5399 /* A hyphen followed by a POSIX class is treated in the same way. */
5401 else if (d == CHAR_LEFT_SQUARE_BRACKET &&
5402 (ptr[1] == CHAR_COLON || ptr[1] == CHAR_DOT ||
5403 ptr[1] == CHAR_EQUALS_SIGN) &&
5404 check_posix_syntax(ptr, &tempptr))
5406 *errorcodeptr = ERR83;
5411 /* Check that the two values are in the correct order. Optimize
5412 one-character ranges. */
5416 *errorcodeptr = ERR8;
5419 if (d == c) goto CLASS_SINGLE_CHARACTER; /* A few lines below */
5421 /* We have found a character range, so single character optimizations
5422 cannot be done anymore. Any value greater than 1 indicates that there
5423 is more than one character. */
5427 /* Remember an explicit \r or \n, and add the range to the class. */
5429 if (d == CHAR_CR || d == CHAR_NL) cd->external_flags |= PCRE_HASCRORLF;
5431 class_has_8bitchar +=
5432 add_to_class(classbits, &class_uchardata, options, cd, c, d);
5434 continue; /* Go get the next char in the class */
5437 /* Handle a single character - we can get here for a normal non-escape
5438 char, or after \ that introduces a single character or for an apparent
5439 range that isn't. Only the value 1 matters for class_one_char, so don't
5440 increase it if it is already 2 or more ... just in case there's a class
5441 with a zillion characters in it. */
5443 CLASS_SINGLE_CHARACTER:
5444 if (class_one_char < 2) class_one_char++;
5446 /* If xclass_has_prop is false and class_one_char is 1, we have the first
5447 single character in the class, and there have been no prior ranges, or
5448 XCLASS items generated by escapes. If this is the final character in the
5449 class, we can optimize by turning the item into a 1-character OP_CHAR[I]
5450 if it's positive, or OP_NOT[I] if it's negative. In the positive case, it
5451 can cause firstchar to be set. Otherwise, there can be no first char if
5452 this item is first, whatever repeat count may follow. In the case of
5453 reqchar, save the previous value for reinstating. */
5459 class_one_char == 1 && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET)
5462 zeroreqchar = reqchar;
5463 zeroreqcharflags = reqcharflags;
5470 if (firstcharflags == REQ_UNSET) firstcharflags = REQ_NONE;
5471 zerofirstchar = firstchar;
5472 zerofirstcharflags = firstcharflags;
5474 /* For caseless UTF-8 mode when UCP support is available, check
5475 whether this character has more than one other case. If so, generate
5476 a special OP_NOTPROP item instead of OP_NOTI. */
5479 if (utf && (options & PCRE_CASELESS) != 0 &&
5480 (d = UCD_CASESET(c)) != 0)
5482 *code++ = OP_NOTPROP;
5488 /* Char has only one other case, or UCP not available */
5491 *code++ = ((options & PCRE_CASELESS) != 0)? OP_NOTI: OP_NOT;
5492 #if defined SUPPORT_UTF && !defined COMPILE_PCRE32
5493 if (utf && c > MAX_VALUE_FOR_SINGLE_CHAR)
5494 code += PRIV(ord2utf)(c, code);
5500 /* We are finished with this character class */
5505 /* For a single, positive character, get the value into mcbuffer, and
5506 then we can handle this with the normal one-character code. */
5508 #if defined SUPPORT_UTF && !defined COMPILE_PCRE32
5509 if (utf && c > MAX_VALUE_FOR_SINGLE_CHAR)
5510 mclength = PRIV(ord2utf)(c, mcbuffer);
5518 } /* End of 1-char optimization */
5520 /* There is more than one character in the class, or an XCLASS item
5521 has been generated. Add this character to the class. */
5523 class_has_8bitchar +=
5524 add_to_class(classbits, &class_uchardata, options, cd, c, c);
5527 /* Loop until ']' reached. This "while" is the end of the "do" far above.
5528 If we are at the end of an internal nested string, revert to the outer
5531 while (((c = *(++ptr)) != CHAR_NULL ||
5533 (ptr = nestptr, nestptr = NULL, c = *(++ptr)) != CHAR_NULL)) &&
5534 (c != CHAR_RIGHT_SQUARE_BRACKET || inescq));
5536 /* Check for missing terminating ']' */
5540 *errorcodeptr = ERR6;
5544 /* We will need an XCLASS if data has been placed in class_uchardata. In
5545 the second phase this is a sufficient test. However, in the pre-compile
5546 phase, class_uchardata gets emptied to prevent workspace overflow, so it
5547 only if the very last character in the class needs XCLASS will it contain
5548 anything at this point. For this reason, xclass gets set TRUE above when
5549 uchar_classdata is emptied, and that's why this code is the way it is here
5550 instead of just doing a test on class_uchardata below. */
5552 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
5553 if (class_uchardata > class_uchardata_base) xclass = TRUE;
5556 /* If this is the first thing in the branch, there can be no first char
5557 setting, whatever the repeat count. Any reqchar setting must remain
5558 unchanged after any kind of repeat. */
5560 if (firstcharflags == REQ_UNSET) firstcharflags = REQ_NONE;
5561 zerofirstchar = firstchar;
5562 zerofirstcharflags = firstcharflags;
5563 zeroreqchar = reqchar;
5564 zeroreqcharflags = reqcharflags;
5566 /* If there are characters with values > 255, we have to compile an
5567 extended class, with its own opcode, unless there was a negated special
5568 such as \S in the class, and PCRE_UCP is not set, because in that case all
5569 characters > 255 are in the class, so any that were explicitly given as
5570 well can be ignored. If (when there are explicit characters > 255 that must
5571 be listed) there are no characters < 256, we can omit the bitmap in the
5572 actual compiled code. */
5575 if (xclass && (xclass_has_prop || !should_flip_negation ||
5576 (options & PCRE_UCP) != 0))
5577 #elif !defined COMPILE_PCRE8
5578 if (xclass && (xclass_has_prop || !should_flip_negation))
5580 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
5582 *class_uchardata++ = XCL_END; /* Marks the end of extra data */
5583 *code++ = OP_XCLASS;
5585 *code = negate_class? XCL_NOT:0;
5586 if (xclass_has_prop) *code |= XCL_HASPROP;
5588 /* If the map is required, move up the extra data to make room for it;
5589 otherwise just move the code pointer to the end of the extra data. */
5591 if (class_has_8bitchar > 0)
5594 memmove(code + (32 / sizeof(pcre_uchar)), code,
5595 IN_UCHARS(class_uchardata - code));
5596 if (negate_class && !xclass_has_prop)
5597 for (c = 0; c < 32; c++) classbits[c] = ~classbits[c];
5598 memcpy(code, classbits, 32);
5599 code = class_uchardata + (32 / sizeof(pcre_uchar));
5601 else code = class_uchardata;
5603 /* Now fill in the complete length of the item */
5605 PUT(previous, 1, (int)(code - previous));
5606 break; /* End of class handling */
5609 /* Even though any XCLASS list is now discarded, we must allow for
5612 if (lengthptr != NULL)
5613 *lengthptr += (int)(class_uchardata - class_uchardata_base);
5616 /* If there are no characters > 255, or they are all to be included or
5617 excluded, set the opcode to OP_CLASS or OP_NCLASS, depending on whether the
5618 whole class was negated and whether there were negative specials such as \S
5619 (non-UCP) in the class. Then copy the 32-byte map into the code vector,
5620 negating it if necessary. */
5622 *code++ = (negate_class == should_flip_negation) ? OP_CLASS : OP_NCLASS;
5623 if (lengthptr == NULL) /* Save time in the pre-compile phase */
5626 for (c = 0; c < 32; c++) classbits[c] = ~classbits[c];
5627 memcpy(code, classbits, 32);
5629 code += 32 / sizeof(pcre_uchar);
5635 /* ===================================================================*/
5636 /* Various kinds of repeat; '{' is not necessarily a quantifier, but this
5637 has been tested above. */
5639 case CHAR_LEFT_CURLY_BRACKET:
5640 if (!is_quantifier) goto NORMAL_CHAR;
5641 ptr = read_repeat_counts(ptr+1, &repeat_min, &repeat_max, errorcodeptr);
5642 if (*errorcodeptr != 0) goto FAILED;
5655 case CHAR_QUESTION_MARK:
5660 if (previous == NULL)
5662 *errorcodeptr = ERR9;
5666 if (repeat_min == 0)
5668 firstchar = zerofirstchar; /* Adjust for zero repeat */
5669 firstcharflags = zerofirstcharflags;
5670 reqchar = zeroreqchar; /* Ditto */
5671 reqcharflags = zeroreqcharflags;
5674 /* Remember whether this is a variable length repeat */
5676 reqvary = (repeat_min == repeat_max)? 0 : REQ_VARY;
5678 op_type = 0; /* Default single-char op codes */
5679 possessive_quantifier = FALSE; /* Default not possessive quantifier */
5681 /* Save start of previous item, in case we have to move it up in order to
5682 insert something before it. */
5684 tempcode = previous;
5686 /* Before checking for a possessive quantifier, we must skip over
5687 whitespace and comments in extended mode because Perl allows white space at
5690 if ((options & PCRE_EXTENDED) != 0)
5692 const pcre_uchar *p = ptr + 1;
5695 while (MAX_255(*p) && (cd->ctypes[*p] & ctype_space) != 0) p++;
5696 if (*p != CHAR_NUMBER_SIGN) break;
5698 while (*p != CHAR_NULL)
5700 if (IS_NEWLINE(p)) /* For non-fixed-length newline cases, */
5701 { /* IS_NEWLINE sets cd->nllen. */
5707 if (utf) FORWARDCHAR(p);
5709 } /* Loop for comment characters */
5710 } /* Loop for multiple comments */
5711 ptr = p - 1; /* Character before the next significant one. */
5714 /* If the next character is '+', we have a possessive quantifier. This
5715 implies greediness, whatever the setting of the PCRE_UNGREEDY option.
5716 If the next character is '?' this is a minimizing repeat, by default,
5717 but if PCRE_UNGREEDY is set, it works the other way round. We change the
5718 repeat type to the non-default. */
5720 if (ptr[1] == CHAR_PLUS)
5722 repeat_type = 0; /* Force greedy */
5723 possessive_quantifier = TRUE;
5726 else if (ptr[1] == CHAR_QUESTION_MARK)
5728 repeat_type = greedy_non_default;
5731 else repeat_type = greedy_default;
5733 /* If previous was a recursion call, wrap it in atomic brackets so that
5734 previous becomes the atomic group. All recursions were so wrapped in the
5735 past, but it no longer happens for non-repeated recursions. In fact, the
5736 repeated ones could be re-implemented independently so as not to need this,
5737 but for the moment we rely on the code for repeating groups. */
5739 if (*previous == OP_RECURSE)
5741 memmove(previous + 1 + LINK_SIZE, previous, IN_UCHARS(1 + LINK_SIZE));
5742 *previous = OP_ONCE;
5743 PUT(previous, 1, 2 + 2*LINK_SIZE);
5744 previous[2 + 2*LINK_SIZE] = OP_KET;
5745 PUT(previous, 3 + 2*LINK_SIZE, 2 + 2*LINK_SIZE);
5746 code += 2 + 2 * LINK_SIZE;
5747 length_prevgroup = 3 + 3*LINK_SIZE;
5749 /* When actually compiling, we need to check whether this was a forward
5750 reference, and if so, adjust the offset. */
5752 if (lengthptr == NULL && cd->hwm >= cd->start_workspace + LINK_SIZE)
5754 int offset = GET(cd->hwm, -LINK_SIZE);
5755 if (offset == previous + 1 - cd->start_code)
5756 PUT(cd->hwm, -LINK_SIZE, offset + 1 + LINK_SIZE);
5760 /* Now handle repetition for the different types of item. */
5762 /* If previous was a character or negated character match, abolish the item
5763 and generate a repeat item instead. If a char item has a minimum of more
5764 than one, ensure that it is set in reqchar - it might not be if a sequence
5765 such as x{3} is the first thing in a branch because the x will have gone
5766 into firstchar instead. */
5768 if (*previous == OP_CHAR || *previous == OP_CHARI
5769 || *previous == OP_NOT || *previous == OP_NOTI)
5773 default: /* Make compiler happy. */
5774 case OP_CHAR: op_type = OP_STAR - OP_STAR; break;
5775 case OP_CHARI: op_type = OP_STARI - OP_STAR; break;
5776 case OP_NOT: op_type = OP_NOTSTAR - OP_STAR; break;
5777 case OP_NOTI: op_type = OP_NOTSTARI - OP_STAR; break;
5780 /* Deal with UTF characters that take up more than one character. It's
5781 easier to write this out separately than try to macrify it. Use c to
5782 hold the length of the character in bytes, plus UTF_LENGTH to flag that
5783 it's a length rather than a small character. */
5785 #if defined SUPPORT_UTF && !defined COMPILE_PCRE32
5786 if (utf && NOT_FIRSTCHAR(code[-1]))
5788 pcre_uchar *lastchar = code - 1;
5790 c = (int)(code - lastchar); /* Length of UTF-8 character */
5791 memcpy(utf_chars, lastchar, IN_UCHARS(c)); /* Save the char */
5792 c |= UTF_LENGTH; /* Flag c as a length */
5795 #endif /* SUPPORT_UTF */
5797 /* Handle the case of a single charater - either with no UTF support, or
5798 with UTF disabled, or for a single character UTF character. */
5801 if (*previous <= OP_CHARI && repeat_min > 1)
5804 reqcharflags = req_caseopt | cd->req_varyopt;
5808 goto OUTPUT_SINGLE_REPEAT; /* Code shared with single character types */
5811 /* If previous was a character type match (\d or similar), abolish it and
5812 create a suitable repeat item. The code is shared with single-character
5813 repeats by setting op_type to add a suitable offset into repeat_type. Note
5814 the the Unicode property types will be present only when SUPPORT_UCP is
5815 defined, but we don't wrap the little bits of code here because it just
5816 makes it horribly messy. */
5818 else if (*previous < OP_EODN)
5820 pcre_uchar *oldcode;
5821 int prop_type, prop_value;
5822 op_type = OP_TYPESTAR - OP_STAR; /* Use type opcodes */
5825 OUTPUT_SINGLE_REPEAT:
5826 if (*previous == OP_PROP || *previous == OP_NOTPROP)
5828 prop_type = previous[1];
5829 prop_value = previous[2];
5831 else prop_type = prop_value = -1;
5834 code = previous; /* Usually overwrite previous item */
5836 /* If the maximum is zero then the minimum must also be zero; Perl allows
5837 this case, so we do too - by simply omitting the item altogether. */
5839 if (repeat_max == 0) goto END_REPEAT;
5841 /* Combine the op_type with the repeat_type */
5843 repeat_type += op_type;
5845 /* A minimum of zero is handled either as the special case * or ?, or as
5846 an UPTO, with the maximum given. */
5848 if (repeat_min == 0)
5850 if (repeat_max == -1) *code++ = OP_STAR + repeat_type;
5851 else if (repeat_max == 1) *code++ = OP_QUERY + repeat_type;
5854 *code++ = OP_UPTO + repeat_type;
5855 PUT2INC(code, 0, repeat_max);
5859 /* A repeat minimum of 1 is optimized into some special cases. If the
5860 maximum is unlimited, we use OP_PLUS. Otherwise, the original item is
5861 left in place and, if the maximum is greater than 1, we use OP_UPTO with
5862 one less than the maximum. */
5864 else if (repeat_min == 1)
5866 if (repeat_max == -1)
5867 *code++ = OP_PLUS + repeat_type;
5870 code = oldcode; /* leave previous item in place */
5871 if (repeat_max == 1) goto END_REPEAT;
5872 *code++ = OP_UPTO + repeat_type;
5873 PUT2INC(code, 0, repeat_max - 1);
5877 /* The case {n,n} is just an EXACT, while the general case {n,m} is
5878 handled as an EXACT followed by an UPTO. */
5882 *code++ = OP_EXACT + op_type; /* NB EXACT doesn't have repeat_type */
5883 PUT2INC(code, 0, repeat_min);
5885 /* If the maximum is unlimited, insert an OP_STAR. Before doing so,
5886 we have to insert the character for the previous code. For a repeated
5887 Unicode property match, there are two extra bytes that define the
5888 required property. In UTF-8 mode, long characters have their length in
5889 c, with the UTF_LENGTH bit as a flag. */
5893 #if defined SUPPORT_UTF && !defined COMPILE_PCRE32
5894 if (utf && (c & UTF_LENGTH) != 0)
5896 memcpy(code, utf_chars, IN_UCHARS(c & 7));
5905 *code++ = prop_type;
5906 *code++ = prop_value;
5909 *code++ = OP_STAR + repeat_type;
5912 /* Else insert an UPTO if the max is greater than the min, again
5913 preceded by the character, for the previously inserted code. If the
5914 UPTO is just for 1 instance, we can use QUERY instead. */
5916 else if (repeat_max != repeat_min)
5918 #if defined SUPPORT_UTF && !defined COMPILE_PCRE32
5919 if (utf && (c & UTF_LENGTH) != 0)
5921 memcpy(code, utf_chars, IN_UCHARS(c & 7));
5929 *code++ = prop_type;
5930 *code++ = prop_value;
5932 repeat_max -= repeat_min;
5934 if (repeat_max == 1)
5936 *code++ = OP_QUERY + repeat_type;
5940 *code++ = OP_UPTO + repeat_type;
5941 PUT2INC(code, 0, repeat_max);
5946 /* The character or character type itself comes last in all cases. */
5948 #if defined SUPPORT_UTF && !defined COMPILE_PCRE32
5949 if (utf && (c & UTF_LENGTH) != 0)
5951 memcpy(code, utf_chars, IN_UCHARS(c & 7));
5958 /* For a repeated Unicode property match, there are two extra bytes that
5959 define the required property. */
5964 *code++ = prop_type;
5965 *code++ = prop_value;
5970 /* If previous was a character class or a back reference, we put the repeat
5971 stuff after it, but just skip the item if the repeat was {0,0}. */
5973 else if (*previous == OP_CLASS || *previous == OP_NCLASS ||
5974 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
5975 *previous == OP_XCLASS ||
5977 *previous == OP_REF || *previous == OP_REFI ||
5978 *previous == OP_DNREF || *previous == OP_DNREFI)
5980 if (repeat_max == 0)
5986 if (repeat_min == 0 && repeat_max == -1)
5987 *code++ = OP_CRSTAR + repeat_type;
5988 else if (repeat_min == 1 && repeat_max == -1)
5989 *code++ = OP_CRPLUS + repeat_type;
5990 else if (repeat_min == 0 && repeat_max == 1)
5991 *code++ = OP_CRQUERY + repeat_type;
5994 *code++ = OP_CRRANGE + repeat_type;
5995 PUT2INC(code, 0, repeat_min);
5996 if (repeat_max == -1) repeat_max = 0; /* 2-byte encoding for max */
5997 PUT2INC(code, 0, repeat_max);
6001 /* If previous was a bracket group, we may have to replicate it in certain
6002 cases. Note that at this point we can encounter only the "basic" bracket
6003 opcodes such as BRA and CBRA, as this is the place where they get converted
6004 into the more special varieties such as BRAPOS and SBRA. A test for >=
6005 OP_ASSERT and <= OP_COND includes ASSERT, ASSERT_NOT, ASSERTBACK,
6006 ASSERTBACK_NOT, ONCE, ONCE_NC, BRA, BRAPOS, CBRA, CBRAPOS, and COND.
6007 Originally, PCRE did not allow repetition of assertions, but now it does,
6008 for Perl compatibility. */
6010 else if (*previous >= OP_ASSERT && *previous <= OP_COND)
6013 int len = (int)(code - previous);
6014 size_t base_hwm_offset = item_hwm_offset;
6015 pcre_uchar *bralink = NULL;
6016 pcre_uchar *brazeroptr = NULL;
6018 /* Repeating a DEFINE group is pointless, but Perl allows the syntax, so
6019 we just ignore the repeat. */
6021 if (*previous == OP_COND && previous[LINK_SIZE+1] == OP_DEF)
6024 /* There is no sense in actually repeating assertions. The only potential
6025 use of repetition is in cases when the assertion is optional. Therefore,
6026 if the minimum is greater than zero, just ignore the repeat. If the
6027 maximum is not zero or one, set it to 1. */
6029 if (*previous < OP_ONCE) /* Assertion */
6031 if (repeat_min > 0) goto END_REPEAT;
6032 if (repeat_max < 0 || repeat_max > 1) repeat_max = 1;
6035 /* The case of a zero minimum is special because of the need to stick
6036 OP_BRAZERO in front of it, and because the group appears once in the
6037 data, whereas in other cases it appears the minimum number of times. For
6038 this reason, it is simplest to treat this case separately, as otherwise
6039 the code gets far too messy. There are several special subcases when the
6042 if (repeat_min == 0)
6044 /* If the maximum is also zero, we used to just omit the group from the
6045 output altogether, like this:
6047 ** if (repeat_max == 0)
6053 However, that fails when a group or a subgroup within it is referenced
6054 as a subroutine from elsewhere in the pattern, so now we stick in
6055 OP_SKIPZERO in front of it so that it is skipped on execution. As we
6056 don't have a list of which groups are referenced, we cannot do this
6059 If the maximum is 1 or unlimited, we just have to stick in the BRAZERO
6060 and do no more at this point. However, we do need to adjust any
6061 OP_RECURSE calls inside the group that refer to the group itself or any
6062 internal or forward referenced group, because the offset is from the
6063 start of the whole regex. Temporarily terminate the pattern while doing
6066 if (repeat_max <= 1) /* Covers 0, 1, and unlimited */
6069 adjust_recurse(previous, 1, utf, cd, item_hwm_offset);
6070 memmove(previous + 1, previous, IN_UCHARS(len));
6072 if (repeat_max == 0)
6074 *previous++ = OP_SKIPZERO;
6077 brazeroptr = previous; /* Save for possessive optimizing */
6078 *previous++ = OP_BRAZERO + repeat_type;
6081 /* If the maximum is greater than 1 and limited, we have to replicate
6082 in a nested fashion, sticking OP_BRAZERO before each set of brackets.
6083 The first one has to be handled carefully because it's the original
6084 copy, which has to be moved up. The remainder can be handled by code
6085 that is common with the non-zero minimum case below. We have to
6086 adjust the value or repeat_max, since one less copy is required. Once
6087 again, we may have to adjust any OP_RECURSE calls inside the group. */
6093 adjust_recurse(previous, 2 + LINK_SIZE, utf, cd, item_hwm_offset);
6094 memmove(previous + 2 + LINK_SIZE, previous, IN_UCHARS(len));
6095 code += 2 + LINK_SIZE;
6096 *previous++ = OP_BRAZERO + repeat_type;
6097 *previous++ = OP_BRA;
6099 /* We chain together the bracket offset fields that have to be
6100 filled in later when the ends of the brackets are reached. */
6102 offset = (bralink == NULL)? 0 : (int)(previous - bralink);
6104 PUTINC(previous, 0, offset);
6110 /* If the minimum is greater than zero, replicate the group as many
6111 times as necessary, and adjust the maximum to the number of subsequent
6112 copies that we need. If we set a first char from the group, and didn't
6113 set a required char, copy the latter from the former. If there are any
6114 forward reference subroutine calls in the group, there will be entries on
6115 the workspace list; replicate these with an appropriate increment. */
6121 /* In the pre-compile phase, we don't actually do the replication. We
6122 just adjust the length as if we had. Do some paranoid checks for
6123 potential integer overflow. The INT64_OR_DOUBLE type is a 64-bit
6124 integer type when available, otherwise double. */
6126 if (lengthptr != NULL)
6128 int delta = (repeat_min - 1)*length_prevgroup;
6129 if ((INT64_OR_DOUBLE)(repeat_min - 1)*
6130 (INT64_OR_DOUBLE)length_prevgroup >
6131 (INT64_OR_DOUBLE)INT_MAX ||
6132 OFLOW_MAX - *lengthptr < delta)
6134 *errorcodeptr = ERR20;
6137 *lengthptr += delta;
6140 /* This is compiling for real. If there is a set first byte for
6141 the group, and we have not yet set a "required byte", set it. Make
6142 sure there is enough workspace for copying forward references before
6147 if (groupsetfirstchar && reqcharflags < 0)
6149 reqchar = firstchar;
6150 reqcharflags = firstcharflags;
6153 for (i = 1; i < repeat_min; i++)
6156 size_t this_hwm_offset = cd->hwm - cd->start_workspace;
6157 memcpy(code, previous, IN_UCHARS(len));
6159 while (cd->hwm > cd->start_workspace + cd->workspace_size -
6160 WORK_SIZE_SAFETY_MARGIN -
6161 (this_hwm_offset - base_hwm_offset))
6163 *errorcodeptr = expand_workspace(cd);
6164 if (*errorcodeptr != 0) goto FAILED;
6167 for (hc = (pcre_uchar *)cd->start_workspace + base_hwm_offset;
6168 hc < (pcre_uchar *)cd->start_workspace + this_hwm_offset;
6171 PUT(cd->hwm, 0, GET(hc, 0) + len);
6172 cd->hwm += LINK_SIZE;
6174 base_hwm_offset = this_hwm_offset;
6180 if (repeat_max > 0) repeat_max -= repeat_min;
6183 /* This code is common to both the zero and non-zero minimum cases. If
6184 the maximum is limited, it replicates the group in a nested fashion,
6185 remembering the bracket starts on a stack. In the case of a zero minimum,
6186 the first one was set up above. In all cases the repeat_max now specifies
6187 the number of additional copies needed. Again, we must remember to
6188 replicate entries on the forward reference list. */
6190 if (repeat_max >= 0)
6192 /* In the pre-compile phase, we don't actually do the replication. We
6193 just adjust the length as if we had. For each repetition we must add 1
6194 to the length for BRAZERO and for all but the last repetition we must
6195 add 2 + 2*LINKSIZE to allow for the nesting that occurs. Do some
6196 paranoid checks to avoid integer overflow. The INT64_OR_DOUBLE type is
6197 a 64-bit integer type when available, otherwise double. */
6199 if (lengthptr != NULL && repeat_max > 0)
6201 int delta = repeat_max * (length_prevgroup + 1 + 2 + 2*LINK_SIZE) -
6202 2 - 2*LINK_SIZE; /* Last one doesn't nest */
6203 if ((INT64_OR_DOUBLE)repeat_max *
6204 (INT64_OR_DOUBLE)(length_prevgroup + 1 + 2 + 2*LINK_SIZE)
6205 > (INT64_OR_DOUBLE)INT_MAX ||
6206 OFLOW_MAX - *lengthptr < delta)
6208 *errorcodeptr = ERR20;
6211 *lengthptr += delta;
6214 /* This is compiling for real */
6216 else for (i = repeat_max - 1; i >= 0; i--)
6219 size_t this_hwm_offset = cd->hwm - cd->start_workspace;
6221 *code++ = OP_BRAZERO + repeat_type;
6223 /* All but the final copy start a new nesting, maintaining the
6224 chain of brackets outstanding. */
6230 offset = (bralink == NULL)? 0 : (int)(code - bralink);
6232 PUTINC(code, 0, offset);
6235 memcpy(code, previous, IN_UCHARS(len));
6237 /* Ensure there is enough workspace for forward references before
6240 while (cd->hwm > cd->start_workspace + cd->workspace_size -
6241 WORK_SIZE_SAFETY_MARGIN -
6242 (this_hwm_offset - base_hwm_offset))
6244 *errorcodeptr = expand_workspace(cd);
6245 if (*errorcodeptr != 0) goto FAILED;
6248 for (hc = (pcre_uchar *)cd->start_workspace + base_hwm_offset;
6249 hc < (pcre_uchar *)cd->start_workspace + this_hwm_offset;
6252 PUT(cd->hwm, 0, GET(hc, 0) + len + ((i != 0)? 2+LINK_SIZE : 1));
6253 cd->hwm += LINK_SIZE;
6255 base_hwm_offset = this_hwm_offset;
6259 /* Now chain through the pending brackets, and fill in their length
6260 fields (which are holding the chain links pro tem). */
6262 while (bralink != NULL)
6265 int offset = (int)(code - bralink + 1);
6266 pcre_uchar *bra = code - offset;
6267 oldlinkoffset = GET(bra, 1);
6268 bralink = (oldlinkoffset == 0)? NULL : bralink - oldlinkoffset;
6270 PUTINC(code, 0, offset);
6271 PUT(bra, 1, offset);
6275 /* If the maximum is unlimited, set a repeater in the final copy. For
6276 ONCE brackets, that's all we need to do. However, possessively repeated
6277 ONCE brackets can be converted into non-capturing brackets, as the
6278 behaviour of (?:xx)++ is the same as (?>xx)++ and this saves having to
6279 deal with possessive ONCEs specially.
6281 Otherwise, when we are doing the actual compile phase, check to see
6282 whether this group is one that could match an empty string. If so,
6283 convert the initial operator to the S form (e.g. OP_BRA -> OP_SBRA) so
6284 that runtime checking can be done. [This check is also applied to ONCE
6285 groups at runtime, but in a different way.]
6287 Then, if the quantifier was possessive and the bracket is not a
6288 conditional, we convert the BRA code to the POS form, and the KET code to
6289 KETRPOS. (It turns out to be convenient at runtime to detect this kind of
6290 subpattern at both the start and at the end.) The use of special opcodes
6291 makes it possible to reduce greatly the stack usage in pcre_exec(). If
6292 the group is preceded by OP_BRAZERO, convert this to OP_BRAPOSZERO.
6294 Then, if the minimum number of matches is 1 or 0, cancel the possessive
6295 flag so that the default action below, of wrapping everything inside
6296 atomic brackets, does not happen. When the minimum is greater than 1,
6297 there will be earlier copies of the group, and so we still have to wrap
6302 pcre_uchar *ketcode = code - 1 - LINK_SIZE;
6303 pcre_uchar *bracode = ketcode - GET(ketcode, 1);
6305 /* Convert possessive ONCE brackets to non-capturing */
6307 if ((*bracode == OP_ONCE || *bracode == OP_ONCE_NC) &&
6308 possessive_quantifier) *bracode = OP_BRA;
6310 /* For non-possessive ONCE brackets, all we need to do is to
6313 if (*bracode == OP_ONCE || *bracode == OP_ONCE_NC)
6314 *ketcode = OP_KETRMAX + repeat_type;
6316 /* Handle non-ONCE brackets and possessive ONCEs (which have been
6317 converted to non-capturing above). */
6321 /* In the compile phase, check for empty string matching. */
6323 if (lengthptr == NULL)
6325 pcre_uchar *scode = bracode;
6328 if (could_be_empty_branch(scode, ketcode, utf, cd, NULL))
6330 *bracode += OP_SBRA - OP_BRA;
6333 scode += GET(scode, 1);
6335 while (*scode == OP_ALT);
6338 /* A conditional group with only one branch has an implicit empty
6339 alternative branch. */
6341 if (*bracode == OP_COND && bracode[GET(bracode,1)] != OP_ALT)
6342 *bracode = OP_SCOND;
6344 /* Handle possessive quantifiers. */
6346 if (possessive_quantifier)
6348 /* For COND brackets, we wrap the whole thing in a possessively
6349 repeated non-capturing bracket, because we have not invented POS
6350 versions of the COND opcodes. Because we are moving code along, we
6351 must ensure that any pending recursive references are updated. */
6353 if (*bracode == OP_COND || *bracode == OP_SCOND)
6355 int nlen = (int)(code - bracode);
6357 adjust_recurse(bracode, 1 + LINK_SIZE, utf, cd, item_hwm_offset);
6358 memmove(bracode + 1 + LINK_SIZE, bracode, IN_UCHARS(nlen));
6359 code += 1 + LINK_SIZE;
6360 nlen += 1 + LINK_SIZE;
6361 *bracode = (*bracode == OP_COND)? OP_BRAPOS : OP_SBRAPOS;
6362 *code++ = OP_KETRPOS;
6363 PUTINC(code, 0, nlen);
6364 PUT(bracode, 1, nlen);
6367 /* For non-COND brackets, we modify the BRA code and use KETRPOS. */
6371 *bracode += 1; /* Switch to xxxPOS opcodes */
6372 *ketcode = OP_KETRPOS;
6375 /* If the minimum is zero, mark it as possessive, then unset the
6376 possessive flag when the minimum is 0 or 1. */
6378 if (brazeroptr != NULL) *brazeroptr = OP_BRAPOSZERO;
6379 if (repeat_min < 2) possessive_quantifier = FALSE;
6382 /* Non-possessive quantifier */
6384 else *ketcode = OP_KETRMAX + repeat_type;
6389 /* If previous is OP_FAIL, it was generated by an empty class [] in
6390 JavaScript mode. The other ways in which OP_FAIL can be generated, that is
6391 by (*FAIL) or (?!) set previous to NULL, which gives a "nothing to repeat"
6392 error above. We can just ignore the repeat in JS case. */
6394 else if (*previous == OP_FAIL) goto END_REPEAT;
6396 /* Else there's some kind of shambles */
6400 *errorcodeptr = ERR11;
6404 /* If the character following a repeat is '+', possessive_quantifier is
6405 TRUE. For some opcodes, there are special alternative opcodes for this
6406 case. For anything else, we wrap the entire repeated item inside OP_ONCE
6407 brackets. Logically, the '+' notation is just syntactic sugar, taken from
6408 Sun's Java package, but the special opcodes can optimize it.
6410 Some (but not all) possessively repeated subpatterns have already been
6411 completely handled in the code just above. For them, possessive_quantifier
6412 is always FALSE at this stage. Note that the repeated item starts at
6413 tempcode, not at previous, which might be the first part of a string whose
6414 (former) last char we repeated. */
6416 if (possessive_quantifier)
6420 /* Possessifying an EXACT quantifier has no effect, so we can ignore it.
6421 However, QUERY, STAR, or UPTO may follow (for quantifiers such as {5,6},
6422 {5,}, or {5,10}). We skip over an EXACT item; if the length of what
6423 remains is greater than zero, there's a further opcode that can be
6424 handled. If not, do nothing, leaving the EXACT alone. */
6429 tempcode += PRIV(OP_lengths)[*tempcode] +
6430 ((tempcode[1 + IMM2_SIZE] == OP_PROP
6431 || tempcode[1 + IMM2_SIZE] == OP_NOTPROP)? 2 : 0);
6434 /* CHAR opcodes are used for exacts whose count is 1. */
6444 tempcode += PRIV(OP_lengths)[*tempcode];
6446 if (utf && HAS_EXTRALEN(tempcode[-1]))
6447 tempcode += GET_EXTRALEN(tempcode[-1]);
6451 /* For the class opcodes, the repeat operator appears at the end;
6452 adjust tempcode to point to it. */
6456 tempcode += 1 + 32/sizeof(pcre_uchar);
6459 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
6461 tempcode += GET(tempcode, 1);
6466 /* If tempcode is equal to code (which points to the end of the repeated
6467 item), it means we have skipped an EXACT item but there is no following
6468 QUERY, STAR, or UPTO; the value of len will be 0, and we do nothing. In
6469 all other cases, tempcode will be pointing to the repeat opcode, and will
6470 be less than code, so the value of len will be greater than 0. */
6472 len = (int)(code - tempcode);
6475 unsigned int repcode = *tempcode;
6477 /* There is a table for possessifying opcodes, all of which are less
6478 than OP_CALLOUT. A zero entry means there is no possessified version.
6481 if (repcode < OP_CALLOUT && opcode_possessify[repcode] > 0)
6482 *tempcode = opcode_possessify[repcode];
6484 /* For opcode without a special possessified version, wrap the item in
6485 ONCE brackets. Because we are moving code along, we must ensure that any
6486 pending recursive references are updated. */
6491 adjust_recurse(tempcode, 1 + LINK_SIZE, utf, cd, item_hwm_offset);
6492 memmove(tempcode + 1 + LINK_SIZE, tempcode, IN_UCHARS(len));
6493 code += 1 + LINK_SIZE;
6494 len += 1 + LINK_SIZE;
6495 tempcode[0] = OP_ONCE;
6497 PUTINC(code, 0, len);
6498 PUT(tempcode, 1, len);
6503 if (len > 0) switch (*tempcode)
6505 case OP_STAR: *tempcode = OP_POSSTAR; break;
6506 case OP_PLUS: *tempcode = OP_POSPLUS; break;
6507 case OP_QUERY: *tempcode = OP_POSQUERY; break;
6508 case OP_UPTO: *tempcode = OP_POSUPTO; break;
6510 case OP_STARI: *tempcode = OP_POSSTARI; break;
6511 case OP_PLUSI: *tempcode = OP_POSPLUSI; break;
6512 case OP_QUERYI: *tempcode = OP_POSQUERYI; break;
6513 case OP_UPTOI: *tempcode = OP_POSUPTOI; break;
6515 case OP_NOTSTAR: *tempcode = OP_NOTPOSSTAR; break;
6516 case OP_NOTPLUS: *tempcode = OP_NOTPOSPLUS; break;
6517 case OP_NOTQUERY: *tempcode = OP_NOTPOSQUERY; break;
6518 case OP_NOTUPTO: *tempcode = OP_NOTPOSUPTO; break;
6520 case OP_NOTSTARI: *tempcode = OP_NOTPOSSTARI; break;
6521 case OP_NOTPLUSI: *tempcode = OP_NOTPOSPLUSI; break;
6522 case OP_NOTQUERYI: *tempcode = OP_NOTPOSQUERYI; break;
6523 case OP_NOTUPTOI: *tempcode = OP_NOTPOSUPTOI; break;
6525 case OP_TYPESTAR: *tempcode = OP_TYPEPOSSTAR; break;
6526 case OP_TYPEPLUS: *tempcode = OP_TYPEPOSPLUS; break;
6527 case OP_TYPEQUERY: *tempcode = OP_TYPEPOSQUERY; break;
6528 case OP_TYPEUPTO: *tempcode = OP_TYPEPOSUPTO; break;
6530 case OP_CRSTAR: *tempcode = OP_CRPOSSTAR; break;
6531 case OP_CRPLUS: *tempcode = OP_CRPOSPLUS; break;
6532 case OP_CRQUERY: *tempcode = OP_CRPOSQUERY; break;
6533 case OP_CRRANGE: *tempcode = OP_CRPOSRANGE; break;
6535 /* Because we are moving code along, we must ensure that any
6536 pending recursive references are updated. */
6540 adjust_recurse(tempcode, 1 + LINK_SIZE, utf, cd, item_hwm_offset);
6541 memmove(tempcode + 1 + LINK_SIZE, tempcode, IN_UCHARS(len));
6542 code += 1 + LINK_SIZE;
6543 len += 1 + LINK_SIZE;
6544 tempcode[0] = OP_ONCE;
6546 PUTINC(code, 0, len);
6547 PUT(tempcode, 1, len);
6553 /* In all case we no longer have a previous item. We also set the
6554 "follows varying string" flag for subsequently encountered reqchars if
6555 it isn't already set and we have just passed a varying length item. */
6559 cd->req_varyopt |= reqvary;
6563 /* ===================================================================*/
6564 /* Start of nested parenthesized sub-expression, or comment or lookahead or
6565 lookbehind or option setting or condition or all the other extended
6566 parenthesis forms. */
6568 case CHAR_LEFT_PARENTHESIS:
6571 /* Now deal with various "verbs" that can be introduced by '*'. */
6573 if (ptr[0] == CHAR_ASTERISK && (ptr[1] == ':'
6574 || (MAX_255(ptr[1]) && ((cd->ctypes[ptr[1]] & ctype_letter) != 0))))
6578 const char *vn = verbnames;
6579 const pcre_uchar *name = ptr + 1;
6580 const pcre_uchar *arg = NULL;
6583 while (MAX_255(*ptr) && (cd->ctypes[*ptr] & ctype_letter) != 0) ptr++;
6584 namelen = (int)(ptr - name);
6586 /* It appears that Perl allows any characters whatsoever, other than
6587 a closing parenthesis, to appear in arguments, so we no longer insist on
6588 letters, digits, and underscores. */
6590 if (*ptr == CHAR_COLON)
6593 while (*ptr != CHAR_NULL && *ptr != CHAR_RIGHT_PARENTHESIS) ptr++;
6594 arglen = (int)(ptr - arg);
6595 if ((unsigned int)arglen > MAX_MARK)
6597 *errorcodeptr = ERR75;
6602 if (*ptr != CHAR_RIGHT_PARENTHESIS)
6604 *errorcodeptr = ERR60;
6608 /* Scan the table of verb names */
6610 for (i = 0; i < verbcount; i++)
6612 if (namelen == verbs[i].len &&
6613 STRNCMP_UC_C8(name, vn, namelen) == 0)
6617 /* Check for open captures before ACCEPT and convert it to
6618 ASSERT_ACCEPT if in an assertion. */
6620 if (verbs[i].op == OP_ACCEPT)
6625 *errorcodeptr = ERR59;
6628 cd->had_accept = TRUE;
6629 for (oc = cd->open_caps; oc != NULL; oc = oc->next)
6631 if (lengthptr != NULL)
6633 #ifdef COMPILE_PCRE8
6634 *lengthptr += 1 + IMM2_SIZE;
6635 #elif defined COMPILE_PCRE16
6636 *lengthptr += 2 + IMM2_SIZE;
6637 #elif defined COMPILE_PCRE32
6638 *lengthptr += 4 + IMM2_SIZE;
6644 PUT2INC(code, 0, oc->number);
6648 (cd->assert_depth > 0)? OP_ASSERT_ACCEPT : OP_ACCEPT;
6650 /* Do not set firstchar after *ACCEPT */
6651 if (firstcharflags == REQ_UNSET) firstcharflags = REQ_NONE;
6654 /* Handle other cases with/without an argument */
6656 else if (arglen == 0)
6658 if (verbs[i].op < 0) /* Argument is mandatory */
6660 *errorcodeptr = ERR66;
6663 setverb = *code++ = verbs[i].op;
6668 if (verbs[i].op_arg < 0) /* Argument is forbidden */
6670 *errorcodeptr = ERR59;
6673 setverb = *code++ = verbs[i].op_arg;
6674 if (lengthptr != NULL) /* In pass 1 just add in the length */
6675 { /* to avoid potential workspace */
6676 *lengthptr += arglen; /* overflow. */
6682 memcpy(code, arg, IN_UCHARS(arglen));
6692 cd->external_flags |= PCRE_HASTHEN;
6699 cd->had_pruneorskip = TRUE;
6703 break; /* Found verb, exit loop */
6706 vn += verbs[i].len + 1;
6709 if (i < verbcount) continue; /* Successfully handled a verb */
6710 *errorcodeptr = ERR60; /* Verb not recognized */
6714 /* Initialize for "real" parentheses */
6716 newoptions = options;
6719 item_hwm_offset = cd->hwm - cd->start_workspace;
6720 reset_bracount = FALSE;
6722 /* Deal with the extended parentheses; all are introduced by '?', and the
6723 appearance of any of them means that this is not a capturing group. */
6725 if (*ptr == CHAR_QUESTION_MARK)
6727 int i, set, unset, namelen;
6729 const pcre_uchar *name;
6734 /* ------------------------------------------------------------ */
6735 case CHAR_VERTICAL_LINE: /* Reset capture count for each branch */
6736 reset_bracount = TRUE;
6737 cd->dupgroups = TRUE; /* Record (?| encountered */
6740 /* ------------------------------------------------------------ */
6741 case CHAR_COLON: /* Non-capturing bracket */
6747 /* ------------------------------------------------------------ */
6748 case CHAR_LEFT_PARENTHESIS:
6749 bravalue = OP_COND; /* Conditional group */
6752 /* A condition can be an assertion, a number (referring to a numbered
6753 group's having been set), a name (referring to a named group), or 'R',
6754 referring to recursion. R<digits> and R&name are also permitted for
6757 There are ways of testing a named group: (?(name)) is used by Python;
6758 Perl 5.10 onwards uses (?(<name>) or (?('name')).
6760 There is one unfortunate ambiguity, caused by history. 'R' can be the
6761 recursive thing or the name 'R' (and similarly for 'R' followed by
6762 digits). We look for a name first; if not found, we try the other case.
6764 For compatibility with auto-callouts, we allow a callout to be
6765 specified before a condition that is an assertion. First, check for the
6766 syntax of a callout; if found, adjust the temporary pointer that is
6767 used to check for an assertion condition. That's all that is needed! */
6769 if (ptr[1] == CHAR_QUESTION_MARK && ptr[2] == CHAR_C)
6771 for (i = 3;; i++) if (!IS_DIGIT(ptr[i])) break;
6772 if (ptr[i] == CHAR_RIGHT_PARENTHESIS)
6775 /* tempptr should now be pointing to the opening parenthesis of the
6776 assertion condition. */
6778 if (*tempptr != CHAR_LEFT_PARENTHESIS)
6780 *errorcodeptr = ERR28;
6785 /* For conditions that are assertions, check the syntax, and then exit
6786 the switch. This will take control down to where bracketed groups,
6787 including assertions, are processed. */
6789 if (tempptr[1] == CHAR_QUESTION_MARK &&
6790 (tempptr[2] == CHAR_EQUALS_SIGN ||
6791 tempptr[2] == CHAR_EXCLAMATION_MARK ||
6792 (tempptr[2] == CHAR_LESS_THAN_SIGN &&
6793 (tempptr[3] == CHAR_EQUALS_SIGN ||
6794 tempptr[3] == CHAR_EXCLAMATION_MARK))))
6796 cd->iscondassert = TRUE;
6800 /* Other conditions use OP_CREF/OP_DNCREF/OP_RREF/OP_DNRREF, and all
6801 need to skip at least 1+IMM2_SIZE bytes at the start of the group. */
6803 code[1+LINK_SIZE] = OP_CREF;
6804 skipbytes = 1+IMM2_SIZE;
6805 refsign = -1; /* => not a number */
6806 namelen = -1; /* => not a name; must set to avoid warning */
6807 name = NULL; /* Always set to avoid warning */
6808 recno = 0; /* Always set to avoid warning */
6810 /* Check for a test for recursion in a named group. */
6813 if (*ptr == CHAR_R && ptr[1] == CHAR_AMPERSAND)
6817 code[1+LINK_SIZE] = OP_RREF; /* Change the type of test */
6820 /* Check for a test for a named group's having been set, using the Perl
6821 syntax (?(<name>) or (?('name'), and also allow for the original PCRE
6822 syntax of (?(name) or for (?(+n), (?(-n), and just (?(n). */
6824 else if (*ptr == CHAR_LESS_THAN_SIGN)
6826 terminator = CHAR_GREATER_THAN_SIGN;
6829 else if (*ptr == CHAR_APOSTROPHE)
6831 terminator = CHAR_APOSTROPHE;
6836 terminator = CHAR_NULL;
6837 if (*ptr == CHAR_MINUS || *ptr == CHAR_PLUS) refsign = *ptr++;
6838 else if (IS_DIGIT(*ptr)) refsign = 0;
6841 /* Handle a number */
6845 while (IS_DIGIT(*ptr))
6847 if (recno > INT_MAX / 10 - 1) /* Integer overflow */
6849 while (IS_DIGIT(*ptr)) ptr++;
6850 *errorcodeptr = ERR61;
6853 recno = recno * 10 + (int)(*ptr - CHAR_0);
6858 /* Otherwise we expect to read a name; anything else is an error. When
6859 a name is one of a number of duplicates, a different opcode is used and
6860 it needs more memory. Unfortunately we cannot tell whether a name is a
6861 duplicate in the first pass, so we have to allow for more memory. */
6867 *errorcodeptr = ERR84;
6870 if (!MAX_255(*ptr) || (cd->ctypes[*ptr] & ctype_word) == 0)
6872 *errorcodeptr = ERR28; /* Assertion expected */
6876 while (MAX_255(*ptr) && (cd->ctypes[*ptr] & ctype_word) != 0)
6880 namelen = (int)(ptr - name);
6881 if (lengthptr != NULL) skipbytes += IMM2_SIZE;
6884 /* Check the terminator */
6886 if ((terminator > 0 && *ptr++ != (pcre_uchar)terminator) ||
6887 *ptr++ != CHAR_RIGHT_PARENTHESIS)
6889 ptr--; /* Error offset */
6890 *errorcodeptr = ERR26; /* Malformed number or name */
6894 /* Do no further checking in the pre-compile phase. */
6896 if (lengthptr != NULL) break;
6898 /* In the real compile we do the work of looking for the actual
6899 reference. If refsign is not negative, it means we have a number in
6906 *errorcodeptr = ERR35;
6909 if (refsign != 0) recno = (refsign == CHAR_MINUS)?
6910 cd->bracount - recno + 1 : recno + cd->bracount;
6911 if (recno <= 0 || recno > cd->final_bracount)
6913 *errorcodeptr = ERR15;
6916 PUT2(code, 2+LINK_SIZE, recno);
6917 if (recno > cd->top_backref) cd->top_backref = recno;
6921 /* Otherwise look for the name. */
6923 slot = cd->name_table;
6924 for (i = 0; i < cd->names_found; i++)
6926 if (STRNCMP_UC_UC(name, slot+IMM2_SIZE, namelen) == 0) break;
6927 slot += cd->name_entry_size;
6930 /* Found the named subpattern. If the name is duplicated, add one to
6931 the opcode to change CREF/RREF into DNCREF/DNRREF and insert
6932 appropriate data values. Otherwise, just insert the unique subpattern
6935 if (i < cd->names_found)
6939 recno = GET2(slot, 0); /* Number from first found */
6940 if (recno > cd->top_backref) cd->top_backref = recno;
6941 for (; i < cd->names_found; i++)
6943 slot += cd->name_entry_size;
6944 if (STRNCMP_UC_UC(name, slot+IMM2_SIZE, namelen) != 0 ||
6945 (slot+IMM2_SIZE)[namelen] != 0) break;
6951 PUT2(code, 2+LINK_SIZE, offset);
6952 PUT2(code, 2+LINK_SIZE+IMM2_SIZE, count);
6953 skipbytes += IMM2_SIZE;
6954 code[1+LINK_SIZE]++;
6956 else /* Not a duplicated name */
6958 PUT2(code, 2+LINK_SIZE, recno);
6962 /* If terminator == CHAR_NULL it means that the name followed directly
6963 after the opening parenthesis [e.g. (?(abc)...] and in this case there
6964 are some further alternatives to try. For the cases where terminator !=
6965 CHAR_NULL [things like (?(<name>... or (?('name')... or (?(R&name)... ]
6966 we have now checked all the possibilities, so give an error. */
6968 else if (terminator != CHAR_NULL)
6970 *errorcodeptr = ERR15;
6974 /* Check for (?(R) for recursion. Allow digits after R to specify a
6975 specific group number. */
6977 else if (*name == CHAR_R)
6980 for (i = 1; i < namelen; i++)
6982 if (!IS_DIGIT(name[i]))
6984 *errorcodeptr = ERR15;
6987 if (recno > INT_MAX / 10 - 1) /* Integer overflow */
6989 *errorcodeptr = ERR61;
6992 recno = recno * 10 + name[i] - CHAR_0;
6994 if (recno == 0) recno = RREF_ANY;
6995 code[1+LINK_SIZE] = OP_RREF; /* Change test type */
6996 PUT2(code, 2+LINK_SIZE, recno);
6999 /* Similarly, check for the (?(DEFINE) "condition", which is always
7002 else if (namelen == 6 && STRNCMP_UC_C8(name, STRING_DEFINE, 6) == 0)
7004 code[1+LINK_SIZE] = OP_DEF;
7008 /* Reference to an unidentified subpattern. */
7012 *errorcodeptr = ERR15;
7018 /* ------------------------------------------------------------ */
7019 case CHAR_EQUALS_SIGN: /* Positive lookahead */
7020 bravalue = OP_ASSERT;
7021 cd->assert_depth += 1;
7025 /* Optimize (?!) to (*FAIL) unless it is quantified - which is a weird
7026 thing to do, but Perl allows all assertions to be quantified, and when
7027 they contain capturing parentheses there may be a potential use for
7028 this feature. Not that that applies to a quantified (?!) but we allow
7029 it for uniformity. */
7031 /* ------------------------------------------------------------ */
7032 case CHAR_EXCLAMATION_MARK: /* Negative lookahead */
7034 if (*ptr == CHAR_RIGHT_PARENTHESIS && ptr[1] != CHAR_ASTERISK &&
7035 ptr[1] != CHAR_PLUS && ptr[1] != CHAR_QUESTION_MARK &&
7036 (ptr[1] != CHAR_LEFT_CURLY_BRACKET || !is_counted_repeat(ptr+2)))
7042 bravalue = OP_ASSERT_NOT;
7043 cd->assert_depth += 1;
7047 /* ------------------------------------------------------------ */
7048 case CHAR_LESS_THAN_SIGN: /* Lookbehind or named define */
7051 case CHAR_EQUALS_SIGN: /* Positive lookbehind */
7052 bravalue = OP_ASSERTBACK;
7053 cd->assert_depth += 1;
7057 case CHAR_EXCLAMATION_MARK: /* Negative lookbehind */
7058 bravalue = OP_ASSERTBACK_NOT;
7059 cd->assert_depth += 1;
7063 default: /* Could be name define, else bad */
7064 if (MAX_255(ptr[1]) && (cd->ctypes[ptr[1]] & ctype_word) != 0)
7066 ptr++; /* Correct offset for error */
7067 *errorcodeptr = ERR24;
7073 /* ------------------------------------------------------------ */
7074 case CHAR_GREATER_THAN_SIGN: /* One-time brackets */
7080 /* ------------------------------------------------------------ */
7081 case CHAR_C: /* Callout - may be followed by digits; */
7082 previous_callout = code; /* Save for later completion */
7083 after_manual_callout = 1; /* Skip one item before completing */
7084 *code++ = OP_CALLOUT;
7088 while(IS_DIGIT(*ptr))
7089 n = n * 10 + *ptr++ - CHAR_0;
7090 if (*ptr != CHAR_RIGHT_PARENTHESIS)
7092 *errorcodeptr = ERR39;
7097 *errorcodeptr = ERR38;
7101 PUT(code, 0, (int)(ptr - cd->start_pattern + 1)); /* Pattern offset */
7102 PUT(code, LINK_SIZE, 0); /* Default length */
7103 code += 2 * LINK_SIZE;
7109 /* ------------------------------------------------------------ */
7110 case CHAR_P: /* Python-style named subpattern handling */
7111 if (*(++ptr) == CHAR_EQUALS_SIGN ||
7112 *ptr == CHAR_GREATER_THAN_SIGN) /* Reference or recursion */
7114 is_recurse = *ptr == CHAR_GREATER_THAN_SIGN;
7115 terminator = CHAR_RIGHT_PARENTHESIS;
7116 goto NAMED_REF_OR_RECURSE;
7118 else if (*ptr != CHAR_LESS_THAN_SIGN) /* Test for Python-style defn */
7120 *errorcodeptr = ERR41;
7123 /* Fall through to handle (?P< as (?< is handled */
7126 /* ------------------------------------------------------------ */
7127 DEFINE_NAME: /* Come here from (?< handling */
7128 case CHAR_APOSTROPHE:
7129 terminator = (*ptr == CHAR_LESS_THAN_SIGN)?
7130 CHAR_GREATER_THAN_SIGN : CHAR_APOSTROPHE;
7134 *errorcodeptr = ERR84; /* Group name must start with non-digit */
7137 while (MAX_255(*ptr) && (cd->ctypes[*ptr] & ctype_word) != 0) ptr++;
7138 namelen = (int)(ptr - name);
7140 /* In the pre-compile phase, do a syntax check, remember the longest
7141 name, and then remember the group in a vector, expanding it if
7142 necessary. Duplicates for the same number are skipped; other duplicates
7143 are checked for validity. In the actual compile, there is nothing to
7146 if (lengthptr != NULL)
7149 pcre_uint32 number = cd->bracount + 1;
7151 if (*ptr != (pcre_uchar)terminator)
7153 *errorcodeptr = ERR42;
7157 if (cd->names_found >= MAX_NAME_COUNT)
7159 *errorcodeptr = ERR49;
7163 if (namelen + IMM2_SIZE + 1 > cd->name_entry_size)
7165 cd->name_entry_size = namelen + IMM2_SIZE + 1;
7166 if (namelen > MAX_NAME_SIZE)
7168 *errorcodeptr = ERR48;
7173 /* Scan the list to check for duplicates. For duplicate names, if the
7174 number is the same, break the loop, which causes the name to be
7175 discarded; otherwise, if DUPNAMES is not set, give an error.
7176 If it is set, allow the name with a different number, but continue
7177 scanning in case this is a duplicate with the same number. For
7178 non-duplicate names, give an error if the number is duplicated. */
7180 ng = cd->named_groups;
7181 for (i = 0; i < cd->names_found; i++, ng++)
7183 if (namelen == ng->length &&
7184 STRNCMP_UC_UC(name, ng->name, namelen) == 0)
7186 if (ng->number == number) break;
7187 if ((options & PCRE_DUPNAMES) == 0)
7189 *errorcodeptr = ERR43;
7192 cd->dupnames = TRUE; /* Duplicate names exist */
7194 else if (ng->number == number)
7196 *errorcodeptr = ERR65;
7201 if (i >= cd->names_found) /* Not a duplicate with same number */
7203 /* Increase the list size if necessary */
7205 if (cd->names_found >= cd->named_group_list_size)
7207 int newsize = cd->named_group_list_size * 2;
7208 named_group *newspace = (PUBL(malloc))
7209 (newsize * sizeof(named_group));
7211 if (newspace == NULL)
7213 *errorcodeptr = ERR21;
7217 memcpy(newspace, cd->named_groups,
7218 cd->named_group_list_size * sizeof(named_group));
7219 if (cd->named_group_list_size > NAMED_GROUP_LIST_SIZE)
7220 (PUBL(free))((void *)cd->named_groups);
7221 cd->named_groups = newspace;
7222 cd->named_group_list_size = newsize;
7225 cd->named_groups[cd->names_found].name = name;
7226 cd->named_groups[cd->names_found].length = namelen;
7227 cd->named_groups[cd->names_found].number = number;
7232 ptr++; /* Move past > or ' in both passes. */
7233 goto NUMBERED_GROUP;
7236 /* ------------------------------------------------------------ */
7237 case CHAR_AMPERSAND: /* Perl recursion/subroutine syntax */
7238 terminator = CHAR_RIGHT_PARENTHESIS;
7242 /* We come here from the Python syntax above that handles both
7243 references (?P=name) and recursion (?P>name), as well as falling
7244 through from the Perl recursion syntax (?&name). We also come here from
7245 the Perl \k<name> or \k'name' back reference syntax and the \k{name}
7246 .NET syntax, and the Oniguruma \g<...> and \g'...' subroutine syntax. */
7248 NAMED_REF_OR_RECURSE:
7252 *errorcodeptr = ERR84; /* Group name must start with non-digit */
7255 while (MAX_255(*ptr) && (cd->ctypes[*ptr] & ctype_word) != 0) ptr++;
7256 namelen = (int)(ptr - name);
7258 /* In the pre-compile phase, do a syntax check. We used to just set
7259 a dummy reference number, because it was not used in the first pass.
7260 However, with the change of recursive back references to be atomic,
7261 we have to look for the number so that this state can be identified, as
7262 otherwise the incorrect length is computed. If it's not a backwards
7263 reference, the dummy number will do. */
7265 if (lengthptr != NULL)
7272 *errorcodeptr = ERR62;
7275 if (*ptr != (pcre_uchar)terminator)
7277 *errorcodeptr = ERR42;
7280 if (namelen > MAX_NAME_SIZE)
7282 *errorcodeptr = ERR48;
7286 /* Count named back references. */
7288 if (!is_recurse) cd->namedrefcount++;
7290 /* We have to allow for a named reference to a duplicated name (this
7291 cannot be determined until the second pass). This needs an extra
7292 16-bit data item. */
7294 *lengthptr += IMM2_SIZE;
7296 /* If this is a forward reference and we are within a (?|...) group,
7297 the reference may end up as the number of a group which we are
7298 currently inside, that is, it could be a recursive reference. In the
7299 real compile this will be picked up and the reference wrapped with
7300 OP_ONCE to make it atomic, so we must space in case this occurs. */
7302 /* In fact, this can happen for a non-forward reference because
7303 another group with the same number might be created later. This
7304 issue is fixed "properly" in PCRE2. As PCRE1 is now in maintenance
7305 only mode, we finesse the bug by allowing more memory always. */
7307 *lengthptr += 4 + 4*LINK_SIZE;
7309 /* It is even worse than that. The current reference may be to an
7310 existing named group with a different number (so apparently not
7311 recursive) but which later on is also attached to a group with the
7312 current number. This can only happen if $(| has been previous
7313 encountered. In that case, we allow yet more memory, just in case.
7314 (Again, this is fixed "properly" in PCRE2. */
7316 if (cd->dupgroups) *lengthptr += 4 + 4*LINK_SIZE;
7318 /* Otherwise, check for recursion here. The name table does not exist
7319 in the first pass; instead we must scan the list of names encountered
7320 so far in order to get the number. If the name is not found, leave
7321 the value of recno as 0 for a forward reference. */
7323 /* This patch (removing "else") fixes a problem when a reference is
7324 to multiple identically named nested groups from within the nest.
7325 Once again, it is not the "proper" fix, and it results in an
7326 over-allocation of memory. */
7330 ng = cd->named_groups;
7331 for (i = 0; i < cd->names_found; i++, ng++)
7333 if (namelen == ng->length &&
7334 STRNCMP_UC_UC(name, ng->name, namelen) == 0)
7338 if (is_recurse) break;
7339 for (oc = cd->open_caps; oc != NULL; oc = oc->next)
7341 if (oc->number == recno)
7352 /* In the real compile, search the name table. We check the name
7353 first, and then check that we have reached the end of the name in the
7354 table. That way, if the name is longer than any in the table, the
7355 comparison will fail without reading beyond the table entry. */
7359 slot = cd->name_table;
7360 for (i = 0; i < cd->names_found; i++)
7362 if (STRNCMP_UC_UC(name, slot+IMM2_SIZE, namelen) == 0 &&
7363 slot[IMM2_SIZE+namelen] == 0)
7365 slot += cd->name_entry_size;
7368 if (i < cd->names_found)
7370 recno = GET2(slot, 0);
7374 *errorcodeptr = ERR15;
7379 /* In both phases, for recursions, we can now go to the code than
7380 handles numerical recursion. */
7382 if (is_recurse) goto HANDLE_RECURSION;
7384 /* In the second pass we must see if the name is duplicated. If so, we
7385 generate a different opcode. */
7387 if (lengthptr == NULL && cd->dupnames)
7390 unsigned int index = i;
7391 pcre_uchar *cslot = slot + cd->name_entry_size;
7393 for (i++; i < cd->names_found; i++)
7395 if (STRCMP_UC_UC(slot + IMM2_SIZE, cslot + IMM2_SIZE) != 0) break;
7397 cslot += cd->name_entry_size;
7402 if (firstcharflags == REQ_UNSET) firstcharflags = REQ_NONE;
7404 item_hwm_offset = cd->hwm - cd->start_workspace;
7405 *code++ = ((options & PCRE_CASELESS) != 0)? OP_DNREFI : OP_DNREF;
7406 PUT2INC(code, 0, index);
7407 PUT2INC(code, 0, count);
7409 /* Process each potentially referenced group. */
7411 for (; slot < cslot; slot += cd->name_entry_size)
7414 recno = GET2(slot, 0);
7415 cd->backref_map |= (recno < 32)? (1 << recno) : 1;
7416 if (recno > cd->top_backref) cd->top_backref = recno;
7418 /* Check to see if this back reference is recursive, that it, it
7419 is inside the group that it references. A flag is set so that the
7420 group can be made atomic. */
7422 for (oc = cd->open_caps; oc != NULL; oc = oc->next)
7424 if (oc->number == recno)
7432 continue; /* End of back ref handling */
7436 /* First pass, or a non-duplicated name. */
7438 goto HANDLE_REFERENCE;
7441 /* ------------------------------------------------------------ */
7442 case CHAR_R: /* Recursion, same as (?0) */
7444 if (*(++ptr) != CHAR_RIGHT_PARENTHESIS)
7446 *errorcodeptr = ERR29;
7449 goto HANDLE_RECURSION;
7452 /* ------------------------------------------------------------ */
7453 case CHAR_MINUS: case CHAR_PLUS: /* Recursion or subroutine */
7454 case CHAR_0: case CHAR_1: case CHAR_2: case CHAR_3: case CHAR_4:
7455 case CHAR_5: case CHAR_6: case CHAR_7: case CHAR_8: case CHAR_9:
7457 const pcre_uchar *called;
7458 terminator = CHAR_RIGHT_PARENTHESIS;
7460 /* Come here from the \g<...> and \g'...' code (Oniguruma
7461 compatibility). However, the syntax has been checked to ensure that
7462 the ... are a (signed) number, so that neither ERR63 nor ERR29 will
7463 be called on this path, nor with the jump to OTHER_CHAR_AFTER_QUERY
7466 HANDLE_NUMERICAL_RECURSION:
7468 if ((refsign = *ptr) == CHAR_PLUS)
7471 if (!IS_DIGIT(*ptr))
7473 *errorcodeptr = ERR63;
7477 else if (refsign == CHAR_MINUS)
7479 if (!IS_DIGIT(ptr[1]))
7480 goto OTHER_CHAR_AFTER_QUERY;
7485 while(IS_DIGIT(*ptr))
7487 if (recno > INT_MAX / 10 - 1) /* Integer overflow */
7489 while (IS_DIGIT(*ptr)) ptr++;
7490 *errorcodeptr = ERR61;
7493 recno = recno * 10 + *ptr++ - CHAR_0;
7496 if (*ptr != (pcre_uchar)terminator)
7498 *errorcodeptr = ERR29;
7502 if (refsign == CHAR_MINUS)
7506 *errorcodeptr = ERR58;
7509 recno = cd->bracount - recno + 1;
7512 *errorcodeptr = ERR15;
7516 else if (refsign == CHAR_PLUS)
7520 *errorcodeptr = ERR58;
7523 recno += cd->bracount;
7526 /* Come here from code above that handles a named recursion */
7531 item_hwm_offset = cd->hwm - cd->start_workspace;
7532 called = cd->start_code;
7534 /* When we are actually compiling, find the bracket that is being
7535 referenced. Temporarily end the regex in case it doesn't exist before
7536 this point. If we end up with a forward reference, first check that
7537 the bracket does occur later so we can give the error (and position)
7538 now. Then remember this forward reference in the workspace so it can
7539 be filled in at the end. */
7541 if (lengthptr == NULL)
7545 called = PRIV(find_bracket)(cd->start_code, utf, recno);
7547 /* Forward reference */
7551 if (recno > cd->final_bracount)
7553 *errorcodeptr = ERR15;
7557 /* Fudge the value of "called" so that when it is inserted as an
7558 offset below, what it actually inserted is the reference number
7559 of the group. Then remember the forward reference. */
7561 called = cd->start_code + recno;
7562 if (cd->hwm >= cd->start_workspace + cd->workspace_size -
7563 WORK_SIZE_SAFETY_MARGIN)
7565 *errorcodeptr = expand_workspace(cd);
7566 if (*errorcodeptr != 0) goto FAILED;
7568 PUTINC(cd->hwm, 0, (int)(code + 1 - cd->start_code));
7571 /* If not a forward reference, and the subpattern is still open,
7572 this is a recursive call. We check to see if this is a left
7573 recursion that could loop for ever, and diagnose that case. We
7574 must not, however, do this check if we are in a conditional
7575 subpattern because the condition might be testing for recursion in
7576 a pattern such as /(?(R)a+|(?R)b)/, which is perfectly valid.
7577 Forever loops are also detected at runtime, so those that occur in
7578 conditional subpatterns will be picked up then. */
7580 else if (GET(called, 1) == 0 && cond_depth <= 0 &&
7581 could_be_empty(called, code, bcptr, utf, cd))
7583 *errorcodeptr = ERR40;
7588 /* Insert the recursion/subroutine item. It does not have a set first
7589 character (relevant if it is repeated, because it will then be
7590 wrapped with ONCE brackets). */
7593 PUT(code, 1, (int)(called - cd->start_code));
7594 code += 1 + LINK_SIZE;
7595 groupsetfirstchar = FALSE;
7598 /* Can't determine a first byte now */
7600 if (firstcharflags == REQ_UNSET) firstcharflags = REQ_NONE;
7604 /* ------------------------------------------------------------ */
7605 default: /* Other characters: check option setting */
7606 OTHER_CHAR_AFTER_QUERY:
7610 while (*ptr != CHAR_RIGHT_PARENTHESIS && *ptr != CHAR_COLON)
7614 case CHAR_MINUS: optset = &unset; break;
7616 case CHAR_J: /* Record that it changed in the external options */
7617 *optset |= PCRE_DUPNAMES;
7618 cd->external_flags |= PCRE_JCHANGED;
7621 case CHAR_i: *optset |= PCRE_CASELESS; break;
7622 case CHAR_m: *optset |= PCRE_MULTILINE; break;
7623 case CHAR_s: *optset |= PCRE_DOTALL; break;
7624 case CHAR_x: *optset |= PCRE_EXTENDED; break;
7625 case CHAR_U: *optset |= PCRE_UNGREEDY; break;
7626 case CHAR_X: *optset |= PCRE_EXTRA; break;
7628 default: *errorcodeptr = ERR12;
7629 ptr--; /* Correct the offset */
7634 /* Set up the changed option bits, but don't change anything yet. */
7636 newoptions = (options | set) & (~unset);
7638 /* If the options ended with ')' this is not the start of a nested
7639 group with option changes, so the options change at this level.
7640 If we are not at the pattern start, reset the greedy defaults and the
7641 case value for firstchar and reqchar. */
7643 if (*ptr == CHAR_RIGHT_PARENTHESIS)
7645 greedy_default = ((newoptions & PCRE_UNGREEDY) != 0);
7646 greedy_non_default = greedy_default ^ 1;
7647 req_caseopt = ((newoptions & PCRE_CASELESS) != 0)? REQ_CASELESS:0;
7649 /* Change options at this level, and pass them back for use
7650 in subsequent branches. */
7652 *optionsptr = options = newoptions;
7653 previous = NULL; /* This item can't be repeated */
7654 continue; /* It is complete */
7657 /* If the options ended with ':' we are heading into a nested group
7658 with possible change of options. Such groups are non-capturing and are
7659 not assertions of any kind. All we need to do is skip over the ':';
7660 the newoptions value is handled below. */
7664 } /* End of switch for character following (? */
7665 } /* End of (? handling */
7667 /* Opening parenthesis not followed by '*' or '?'. If PCRE_NO_AUTO_CAPTURE
7668 is set, all unadorned brackets become non-capturing and behave like (?:...)
7671 else if ((options & PCRE_NO_AUTO_CAPTURE) != 0)
7676 /* Else we have a capturing group. */
7682 PUT2(code, 1+LINK_SIZE, cd->bracount);
7683 skipbytes = IMM2_SIZE;
7686 /* Process nested bracketed regex. First check for parentheses nested too
7689 if ((cd->parens_depth += 1) > PARENS_NEST_LIMIT)
7691 *errorcodeptr = ERR82;
7695 /* All assertions used not to be repeatable, but this was changed for Perl
7696 compatibility. All kinds can now be repeated except for assertions that are
7697 conditions (Perl also forbids these to be repeated). We copy code into a
7698 non-register variable (tempcode) in order to be able to pass its address
7699 because some compilers complain otherwise. At the start of a conditional
7700 group whose condition is an assertion, cd->iscondassert is set. We unset it
7701 here so as to allow assertions later in the group to be quantified. */
7703 if (bravalue >= OP_ASSERT && bravalue <= OP_ASSERTBACK_NOT &&
7707 cd->iscondassert = FALSE;
7712 item_hwm_offset = cd->hwm - cd->start_workspace;
7717 tempreqvary = cd->req_varyopt; /* Save value before bracket */
7718 tempbracount = cd->bracount; /* Save value before bracket */
7719 length_prevgroup = 0; /* Initialize for pre-compile phase */
7722 newoptions, /* The complete new option state */
7723 &tempcode, /* Where to put code (updated) */
7724 &ptr, /* Input pointer (updated) */
7725 errorcodeptr, /* Where to put an error message */
7726 (bravalue == OP_ASSERTBACK ||
7727 bravalue == OP_ASSERTBACK_NOT), /* TRUE if back assert */
7728 reset_bracount, /* True if (?| group */
7729 skipbytes, /* Skip over bracket number */
7731 ((bravalue == OP_COND)?1:0), /* Depth of condition subpatterns */
7732 &subfirstchar, /* For possible first char */
7734 &subreqchar, /* For possible last char */
7736 bcptr, /* Current branch chain */
7737 cd, /* Tables block */
7738 (lengthptr == NULL)? NULL : /* Actual compile phase */
7739 &length_prevgroup /* Pre-compile phase */
7743 cd->parens_depth -= 1;
7745 /* If this was an atomic group and there are no capturing groups within it,
7746 generate OP_ONCE_NC instead of OP_ONCE. */
7748 if (bravalue == OP_ONCE && cd->bracount <= tempbracount)
7751 if (bravalue >= OP_ASSERT && bravalue <= OP_ASSERTBACK_NOT)
7752 cd->assert_depth -= 1;
7754 /* At the end of compiling, code is still pointing to the start of the
7755 group, while tempcode has been updated to point past the end of the group.
7756 The pattern pointer (ptr) is on the bracket.
7758 If this is a conditional bracket, check that there are no more than
7759 two branches in the group, or just one if it's a DEFINE group. We do this
7760 in the real compile phase, not in the pre-pass, where the whole group may
7761 not be available. */
7763 if (bravalue == OP_COND && lengthptr == NULL)
7765 pcre_uchar *tc = code;
7772 while (*tc != OP_KET);
7774 /* A DEFINE group is never obeyed inline (the "condition" is always
7775 false). It must have only one branch. */
7777 if (code[LINK_SIZE+1] == OP_DEF)
7781 *errorcodeptr = ERR54;
7784 bravalue = OP_DEF; /* Just a flag to suppress char handling below */
7787 /* A "normal" conditional group. If there is just one branch, we must not
7788 make use of its firstchar or reqchar, because this is equivalent to an
7789 empty second branch. */
7795 *errorcodeptr = ERR27;
7798 if (condcount == 1) subfirstcharflags = subreqcharflags = REQ_NONE;
7802 /* Error if hit end of pattern */
7804 if (*ptr != CHAR_RIGHT_PARENTHESIS)
7806 *errorcodeptr = ERR14;
7810 /* In the pre-compile phase, update the length by the length of the group,
7811 less the brackets at either end. Then reduce the compiled code to just a
7812 set of non-capturing brackets so that it doesn't use much memory if it is
7813 duplicated by a quantifier.*/
7815 if (lengthptr != NULL)
7817 if (OFLOW_MAX - *lengthptr < length_prevgroup - 2 - 2*LINK_SIZE)
7819 *errorcodeptr = ERR20;
7822 *lengthptr += length_prevgroup - 2 - 2*LINK_SIZE;
7823 code++; /* This already contains bravalue */
7824 PUTINC(code, 0, 1 + LINK_SIZE);
7826 PUTINC(code, 0, 1 + LINK_SIZE);
7827 break; /* No need to waste time with special character handling */
7830 /* Otherwise update the main code pointer to the end of the group. */
7834 /* For a DEFINE group, required and first character settings are not
7837 if (bravalue == OP_DEF) break;
7839 /* Handle updating of the required and first characters for other types of
7840 group. Update for normal brackets of all kinds, and conditions with two
7841 branches (see code above). If the bracket is followed by a quantifier with
7842 zero repeat, we have to back off. Hence the definition of zeroreqchar and
7843 zerofirstchar outside the main loop so that they can be accessed for the
7846 zeroreqchar = reqchar;
7847 zeroreqcharflags = reqcharflags;
7848 zerofirstchar = firstchar;
7849 zerofirstcharflags = firstcharflags;
7850 groupsetfirstchar = FALSE;
7852 if (bravalue >= OP_ONCE)
7854 /* If we have not yet set a firstchar in this branch, take it from the
7855 subpattern, remembering that it was set here so that a repeat of more
7856 than one can replicate it as reqchar if necessary. If the subpattern has
7857 no firstchar, set "none" for the whole branch. In both cases, a zero
7858 repeat forces firstchar to "none". */
7860 if (firstcharflags == REQ_UNSET)
7862 if (subfirstcharflags >= 0)
7864 firstchar = subfirstchar;
7865 firstcharflags = subfirstcharflags;
7866 groupsetfirstchar = TRUE;
7868 else firstcharflags = REQ_NONE;
7869 zerofirstcharflags = REQ_NONE;
7872 /* If firstchar was previously set, convert the subpattern's firstchar
7873 into reqchar if there wasn't one, using the vary flag that was in
7874 existence beforehand. */
7876 else if (subfirstcharflags >= 0 && subreqcharflags < 0)
7878 subreqchar = subfirstchar;
7879 subreqcharflags = subfirstcharflags | tempreqvary;
7882 /* If the subpattern set a required byte (or set a first byte that isn't
7883 really the first byte - see above), set it. */
7885 if (subreqcharflags >= 0)
7887 reqchar = subreqchar;
7888 reqcharflags = subreqcharflags;
7892 /* For a forward assertion, we take the reqchar, if set. This can be
7893 helpful if the pattern that follows the assertion doesn't set a different
7894 char. For example, it's useful for /(?=abcde).+/. We can't set firstchar
7895 for an assertion, however because it leads to incorrect effect for patterns
7896 such as /(?=a)a.+/ when the "real" "a" would then become a reqchar instead
7897 of a firstchar. This is overcome by a scan at the end if there's no
7898 firstchar, looking for an asserted first char. */
7900 else if (bravalue == OP_ASSERT && subreqcharflags >= 0)
7902 reqchar = subreqchar;
7903 reqcharflags = subreqcharflags;
7905 break; /* End of processing '(' */
7908 /* ===================================================================*/
7909 /* Handle metasequences introduced by \. For ones like \d, the ESC_ values
7910 are arranged to be the negation of the corresponding OP_values in the
7911 default case when PCRE_UCP is not set. For the back references, the values
7912 are negative the reference number. Only back references and those types
7913 that consume a character may be repeated. We can test for values between
7914 ESC_b and ESC_Z for the latter; this may have to change if any new ones are
7917 case CHAR_BACKSLASH:
7919 escape = check_escape(&ptr, &ec, errorcodeptr, cd->bracount, options, FALSE);
7920 if (*errorcodeptr != 0) goto FAILED;
7922 if (escape == 0) /* The escape coded a single character */
7926 /* For metasequences that actually match a character, we disable the
7927 setting of a first character if it hasn't already been set. */
7929 if (firstcharflags == REQ_UNSET && escape > ESC_b && escape < ESC_Z)
7930 firstcharflags = REQ_NONE;
7932 /* Set values to reset to if this is followed by a zero repeat. */
7934 zerofirstchar = firstchar;
7935 zerofirstcharflags = firstcharflags;
7936 zeroreqchar = reqchar;
7937 zeroreqcharflags = reqcharflags;
7939 /* \g<name> or \g'name' is a subroutine call by name and \g<n> or \g'n'
7940 is a subroutine call by number (Oniguruma syntax). In fact, the value
7941 ESC_g is returned only for these cases. So we don't need to check for <
7942 or ' if the value is ESC_g. For the Perl syntax \g{n} the value is
7943 -n, and for the Perl syntax \g{name} the result is ESC_k (as
7944 that is a synonym for a named back reference). */
7946 if (escape == ESC_g)
7948 const pcre_uchar *p;
7951 item_hwm_offset = cd->hwm - cd->start_workspace; /* Normally this is set when '(' is read */
7952 terminator = (*(++ptr) == CHAR_LESS_THAN_SIGN)?
7953 CHAR_GREATER_THAN_SIGN : CHAR_APOSTROPHE;
7955 /* These two statements stop the compiler for warning about possibly
7956 unset variables caused by the jump to HANDLE_NUMERICAL_RECURSION. In
7957 fact, because we do the check for a number below, the paths that
7958 would actually be in error are never taken. */
7961 reset_bracount = FALSE;
7963 /* If it's not a signed or unsigned number, treat it as a name. */
7966 if (cf != CHAR_PLUS && cf != CHAR_MINUS && !IS_DIGIT(cf))
7969 goto NAMED_REF_OR_RECURSE;
7972 /* Signed or unsigned number (cf = ptr[1]) is known to be plus or minus
7976 while (IS_DIGIT(*p)) p++;
7977 if (*p != (pcre_uchar)terminator)
7979 *errorcodeptr = ERR57;
7983 goto HANDLE_NUMERICAL_RECURSION;
7986 /* \k<name> or \k'name' is a back reference by name (Perl syntax).
7987 We also support \k{name} (.NET syntax). */
7989 if (escape == ESC_k)
7991 if ((ptr[1] != CHAR_LESS_THAN_SIGN &&
7992 ptr[1] != CHAR_APOSTROPHE && ptr[1] != CHAR_LEFT_CURLY_BRACKET))
7994 *errorcodeptr = ERR69;
7998 terminator = (*(++ptr) == CHAR_LESS_THAN_SIGN)?
7999 CHAR_GREATER_THAN_SIGN : (*ptr == CHAR_APOSTROPHE)?
8000 CHAR_APOSTROPHE : CHAR_RIGHT_CURLY_BRACKET;
8001 goto NAMED_REF_OR_RECURSE;
8004 /* Back references are handled specially; must disable firstchar if
8005 not set to cope with cases like (?=(\w+))\1: which would otherwise set
8013 /* Come here from named backref handling when the reference is to a
8014 single group (i.e. not to a duplicated name. */
8017 if (firstcharflags == REQ_UNSET) firstcharflags = REQ_NONE;
8019 item_hwm_offset = cd->hwm - cd->start_workspace;
8020 *code++ = ((options & PCRE_CASELESS) != 0)? OP_REFI : OP_REF;
8021 PUT2INC(code, 0, recno);
8022 cd->backref_map |= (recno < 32)? (1 << recno) : 1;
8023 if (recno > cd->top_backref) cd->top_backref = recno;
8025 /* Check to see if this back reference is recursive, that it, it
8026 is inside the group that it references. A flag is set so that the
8027 group can be made atomic. */
8029 for (oc = cd->open_caps; oc != NULL; oc = oc->next)
8031 if (oc->number == recno)
8039 /* So are Unicode property matches, if supported. */
8042 else if (escape == ESC_P || escape == ESC_p)
8045 unsigned int ptype = 0, pdata = 0;
8046 if (!get_ucp(&ptr, &negated, &ptype, &pdata, errorcodeptr))
8049 item_hwm_offset = cd->hwm - cd->start_workspace;
8050 *code++ = ((escape == ESC_p) != negated)? OP_PROP : OP_NOTPROP;
8056 /* If Unicode properties are not supported, \X, \P, and \p are not
8059 else if (escape == ESC_X || escape == ESC_P || escape == ESC_p)
8061 *errorcodeptr = ERR45;
8066 /* For the rest (including \X when Unicode properties are supported), we
8067 can obtain the OP value by negating the escape value in the default
8068 situation when PCRE_UCP is not set. When it *is* set, we substitute
8069 Unicode property tests. Note that \b and \B do a one-character
8070 lookbehind, and \A also behaves as if it does. */
8074 if ((escape == ESC_b || escape == ESC_B || escape == ESC_A) &&
8075 cd->max_lookbehind == 0)
8076 cd->max_lookbehind = 1;
8078 if (escape >= ESC_DU && escape <= ESC_wu)
8080 nestptr = ptr + 1; /* Where to resume */
8081 ptr = substitutes[escape - ESC_DU] - 1; /* Just before substitute */
8085 /* In non-UTF-8 mode, we turn \C into OP_ALLANY instead of OP_ANYBYTE
8086 so that it works in DFA mode and in lookbehinds. */
8089 previous = (escape > ESC_b && escape < ESC_Z)? code : NULL;
8090 item_hwm_offset = cd->hwm - cd->start_workspace;
8091 *code++ = (!utf && escape == ESC_C)? OP_ALLANY : escape;
8097 /* We have a data character whose value is in c. In UTF-8 mode it may have
8098 a value > 127. We set its representation in the length/buffer, and then
8099 handle it as a data character. */
8101 #if defined SUPPORT_UTF && !defined COMPILE_PCRE32
8102 if (utf && c > MAX_VALUE_FOR_SINGLE_CHAR)
8103 mclength = PRIV(ord2utf)(c, mcbuffer);
8114 /* ===================================================================*/
8115 /* Handle a literal character. It is guaranteed not to be whitespace or #
8116 when the extended flag is set. If we are in a UTF mode, it may be a
8117 multi-unit literal character. */
8125 if (utf && HAS_EXTRALEN(c))
8126 ACROSSCHAR(TRUE, ptr[1], mcbuffer[mclength++] = *(++ptr));
8129 /* At this point we have the character's bytes in mcbuffer, and the length
8130 in mclength. When not in UTF-8 mode, the length is always 1. */
8134 item_hwm_offset = cd->hwm - cd->start_workspace;
8136 /* For caseless UTF-8 mode when UCP support is available, check whether
8137 this character has more than one other case. If so, generate a special
8138 OP_PROP item instead of OP_CHARI. */
8141 if (utf && (options & PCRE_CASELESS) != 0)
8143 GETCHAR(c, mcbuffer);
8144 if ((c = UCD_CASESET(c)) != 0)
8149 if (firstcharflags == REQ_UNSET)
8150 firstcharflags = zerofirstcharflags = REQ_NONE;
8156 /* Caseful matches, or not one of the multicase characters. */
8158 *code++ = ((options & PCRE_CASELESS) != 0)? OP_CHARI : OP_CHAR;
8159 for (c = 0; c < mclength; c++) *code++ = mcbuffer[c];
8161 /* Remember if \r or \n were seen */
8163 if (mcbuffer[0] == CHAR_CR || mcbuffer[0] == CHAR_NL)
8164 cd->external_flags |= PCRE_HASCRORLF;
8166 /* Set the first and required bytes appropriately. If no previous first
8167 byte, set it from this character, but revert to none on a zero repeat.
8168 Otherwise, leave the firstchar value alone, and don't change it on a zero
8171 if (firstcharflags == REQ_UNSET)
8173 zerofirstcharflags = REQ_NONE;
8174 zeroreqchar = reqchar;
8175 zeroreqcharflags = reqcharflags;
8177 /* If the character is more than one byte long, we can set firstchar
8178 only if it is not to be matched caselessly. */
8180 if (mclength == 1 || req_caseopt == 0)
8182 firstchar = mcbuffer[0] | req_caseopt;
8183 firstchar = mcbuffer[0];
8184 firstcharflags = req_caseopt;
8189 reqcharflags = cd->req_varyopt;
8192 else firstcharflags = reqcharflags = REQ_NONE;
8195 /* firstchar was previously set; we can set reqchar only if the length is
8196 1 or the matching is caseful. */
8200 zerofirstchar = firstchar;
8201 zerofirstcharflags = firstcharflags;
8202 zeroreqchar = reqchar;
8203 zeroreqcharflags = reqcharflags;
8204 if (mclength == 1 || req_caseopt == 0)
8207 reqcharflags = req_caseopt | cd->req_varyopt;
8211 break; /* End of literal character handling */
8213 } /* end of big loop */
8216 /* Control never reaches here by falling through, only by a goto for all the
8217 error states. Pass back the position in the pattern so that it can be displayed
8218 to the user for diagnosing the error. */
8227 /*************************************************
8228 * Compile sequence of alternatives *
8229 *************************************************/
8231 /* On entry, ptr is pointing past the bracket character, but on return it
8232 points to the closing bracket, or vertical bar, or end of string. The code
8233 variable is pointing at the byte into which the BRA operator has been stored.
8234 This function is used during the pre-compile phase when we are trying to find
8235 out the amount of memory needed, as well as during the real compile phase. The
8236 value of lengthptr distinguishes the two phases.
8239 options option bits, including any changes for this subpattern
8240 codeptr -> the address of the current code pointer
8241 ptrptr -> the address of the current pattern pointer
8242 errorcodeptr -> pointer to error code variable
8243 lookbehind TRUE if this is a lookbehind assertion
8244 reset_bracount TRUE to reset the count for each branch
8245 skipbytes skip this many bytes at start (for brackets and OP_COND)
8246 cond_depth depth of nesting for conditional subpatterns
8247 firstcharptr place to put the first required character
8248 firstcharflagsptr place to put the first character flags, or a negative number
8249 reqcharptr place to put the last required character
8250 reqcharflagsptr place to put the last required character flags, or a negative number
8251 bcptr pointer to the chain of currently open branches
8252 cd points to the data block with tables pointers etc.
8253 lengthptr NULL during the real compile phase
8254 points to length accumulator during pre-compile phase
8256 Returns: TRUE on success
8260 compile_regex(int options, pcre_uchar **codeptr, const pcre_uchar **ptrptr,
8261 int *errorcodeptr, BOOL lookbehind, BOOL reset_bracount, int skipbytes,
8263 pcre_uint32 *firstcharptr, pcre_int32 *firstcharflagsptr,
8264 pcre_uint32 *reqcharptr, pcre_int32 *reqcharflagsptr,
8265 branch_chain *bcptr, compile_data *cd, int *lengthptr)
8267 const pcre_uchar *ptr = *ptrptr;
8268 pcre_uchar *code = *codeptr;
8269 pcre_uchar *last_branch = code;
8270 pcre_uchar *start_bracket = code;
8271 pcre_uchar *reverse_count = NULL;
8272 open_capitem capitem;
8274 pcre_uint32 firstchar, reqchar;
8275 pcre_int32 firstcharflags, reqcharflags;
8276 pcre_uint32 branchfirstchar, branchreqchar;
8277 pcre_int32 branchfirstcharflags, branchreqcharflags;
8279 unsigned int orig_bracount;
8280 unsigned int max_bracount;
8282 size_t save_hwm_offset;
8284 /* If set, call the external function that checks for stack availability. */
8286 if (PUBL(stack_guard) != NULL && PUBL(stack_guard)())
8288 *errorcodeptr= ERR85;
8292 /* Miscellaneous initialization */
8295 bc.current_branch = code;
8297 firstchar = reqchar = 0;
8298 firstcharflags = reqcharflags = REQ_UNSET;
8300 save_hwm_offset = cd->hwm - cd->start_workspace;
8302 /* Accumulate the length for use in the pre-compile phase. Start with the
8303 length of the BRA and KET and any extra bytes that are required at the
8304 beginning. We accumulate in a local variable to save frequent testing of
8305 lenthptr for NULL. We cannot do this by looking at the value of code at the
8306 start and end of each alternative, because compiled items are discarded during
8307 the pre-compile phase so that the work space is not exceeded. */
8309 length = 2 + 2*LINK_SIZE + skipbytes;
8311 /* WARNING: If the above line is changed for any reason, you must also change
8312 the code that abstracts option settings at the start of the pattern and makes
8313 them global. It tests the value of length for (2 + 2*LINK_SIZE) in the
8314 pre-compile phase to find out whether anything has yet been compiled or not. */
8316 /* If this is a capturing subpattern, add to the chain of open capturing items
8317 so that we can detect them if (*ACCEPT) is encountered. This is also used to
8318 detect groups that contain recursive back references to themselves. Note that
8319 only OP_CBRA need be tested here; changing this opcode to one of its variants,
8320 e.g. OP_SCBRAPOS, happens later, after the group has been compiled. */
8322 if (*code == OP_CBRA)
8324 capnumber = GET2(code, 1 + LINK_SIZE);
8325 capitem.number = capnumber;
8326 capitem.next = cd->open_caps;
8327 capitem.flag = FALSE;
8328 cd->open_caps = &capitem;
8331 /* Offset is set zero to mark that this bracket is still open */
8334 code += 1 + LINK_SIZE + skipbytes;
8336 /* Loop for each alternative branch */
8338 orig_bracount = max_bracount = cd->bracount;
8341 /* For a (?| group, reset the capturing bracket count so that each branch
8342 uses the same numbers. */
8344 if (reset_bracount) cd->bracount = orig_bracount;
8346 /* Set up dummy OP_REVERSE if lookbehind assertion */
8350 *code++ = OP_REVERSE;
8351 reverse_count = code;
8353 length += 1 + LINK_SIZE;
8356 /* Now compile the branch; in the pre-compile phase its length gets added
8359 if (!compile_branch(&options, &code, &ptr, errorcodeptr, &branchfirstchar,
8360 &branchfirstcharflags, &branchreqchar, &branchreqcharflags, &bc,
8361 cond_depth, cd, (lengthptr == NULL)? NULL : &length))
8367 /* Keep the highest bracket count in case (?| was used and some branch
8368 has fewer than the rest. */
8370 if (cd->bracount > max_bracount) max_bracount = cd->bracount;
8372 /* In the real compile phase, there is some post-processing to be done. */
8374 if (lengthptr == NULL)
8376 /* If this is the first branch, the firstchar and reqchar values for the
8377 branch become the values for the regex. */
8379 if (*last_branch != OP_ALT)
8381 firstchar = branchfirstchar;
8382 firstcharflags = branchfirstcharflags;
8383 reqchar = branchreqchar;
8384 reqcharflags = branchreqcharflags;
8387 /* If this is not the first branch, the first char and reqchar have to
8388 match the values from all the previous branches, except that if the
8389 previous value for reqchar didn't have REQ_VARY set, it can still match,
8390 and we set REQ_VARY for the regex. */
8394 /* If we previously had a firstchar, but it doesn't match the new branch,
8395 we have to abandon the firstchar for the regex, but if there was
8396 previously no reqchar, it takes on the value of the old firstchar. */
8398 if (firstcharflags >= 0 &&
8399 (firstcharflags != branchfirstcharflags || firstchar != branchfirstchar))
8401 if (reqcharflags < 0)
8403 reqchar = firstchar;
8404 reqcharflags = firstcharflags;
8406 firstcharflags = REQ_NONE;
8409 /* If we (now or from before) have no firstchar, a firstchar from the
8410 branch becomes a reqchar if there isn't a branch reqchar. */
8412 if (firstcharflags < 0 && branchfirstcharflags >= 0 && branchreqcharflags < 0)
8414 branchreqchar = branchfirstchar;
8415 branchreqcharflags = branchfirstcharflags;
8418 /* Now ensure that the reqchars match */
8420 if (((reqcharflags & ~REQ_VARY) != (branchreqcharflags & ~REQ_VARY)) ||
8421 reqchar != branchreqchar)
8422 reqcharflags = REQ_NONE;
8425 reqchar = branchreqchar;
8426 reqcharflags |= branchreqcharflags; /* To "or" REQ_VARY */
8430 /* If lookbehind, check that this branch matches a fixed-length string, and
8431 put the length into the OP_REVERSE item. Temporarily mark the end of the
8432 branch with OP_END. If the branch contains OP_RECURSE, the result is -3
8433 because there may be forward references that we can't check here. Set a
8434 flag to cause another lookbehind check at the end. Why not do it all at the
8435 end? Because common, erroneous checks are picked up here and the offset of
8436 the problem can be shown. */
8442 fixed_length = find_fixedlength(last_branch, (options & PCRE_UTF8) != 0,
8444 DPRINTF(("fixed length = %d\n", fixed_length));
8445 if (fixed_length == -3)
8447 cd->check_lookbehind = TRUE;
8449 else if (fixed_length < 0)
8451 *errorcodeptr = (fixed_length == -2)? ERR36 :
8452 (fixed_length == -4)? ERR70: ERR25;
8458 if (fixed_length > cd->max_lookbehind)
8459 cd->max_lookbehind = fixed_length;
8460 PUT(reverse_count, 0, fixed_length);
8465 /* Reached end of expression, either ')' or end of pattern. In the real
8466 compile phase, go back through the alternative branches and reverse the chain
8467 of offsets, with the field in the BRA item now becoming an offset to the
8468 first alternative. If there are no alternatives, it points to the end of the
8469 group. The length in the terminating ket is always the length of the whole
8470 bracketed item. Return leaving the pointer at the terminating char. */
8472 if (*ptr != CHAR_VERTICAL_LINE)
8474 if (lengthptr == NULL)
8476 int branch_length = (int)(code - last_branch);
8479 int prev_length = GET(last_branch, 1);
8480 PUT(last_branch, 1, branch_length);
8481 branch_length = prev_length;
8482 last_branch -= branch_length;
8484 while (branch_length > 0);
8487 /* Fill in the ket */
8490 PUT(code, 1, (int)(code - start_bracket));
8491 code += 1 + LINK_SIZE;
8493 /* If it was a capturing subpattern, check to see if it contained any
8494 recursive back references. If so, we must wrap it in atomic brackets.
8495 Because we are moving code along, we must ensure that any pending recursive
8496 references are updated. In any event, remove the block from the chain. */
8500 if (cd->open_caps->flag)
8503 adjust_recurse(start_bracket, 1 + LINK_SIZE,
8504 (options & PCRE_UTF8) != 0, cd, save_hwm_offset);
8505 memmove(start_bracket + 1 + LINK_SIZE, start_bracket,
8506 IN_UCHARS(code - start_bracket));
8507 *start_bracket = OP_ONCE;
8508 code += 1 + LINK_SIZE;
8509 PUT(start_bracket, 1, (int)(code - start_bracket));
8511 PUT(code, 1, (int)(code - start_bracket));
8512 code += 1 + LINK_SIZE;
8513 length += 2 + 2*LINK_SIZE;
8515 cd->open_caps = cd->open_caps->next;
8518 /* Retain the highest bracket number, in case resetting was used. */
8520 cd->bracount = max_bracount;
8522 /* Set values to pass back */
8526 *firstcharptr = firstchar;
8527 *firstcharflagsptr = firstcharflags;
8528 *reqcharptr = reqchar;
8529 *reqcharflagsptr = reqcharflags;
8530 if (lengthptr != NULL)
8532 if (OFLOW_MAX - *lengthptr < length)
8534 *errorcodeptr = ERR20;
8537 *lengthptr += length;
8542 /* Another branch follows. In the pre-compile phase, we can move the code
8543 pointer back to where it was for the start of the first branch. (That is,
8544 pretend that each branch is the only one.)
8546 In the real compile phase, insert an ALT node. Its length field points back
8547 to the previous branch while the bracket remains open. At the end the chain
8548 is reversed. It's done like this so that the start of the bracket has a
8549 zero offset until it is closed, making it possible to detect recursion. */
8551 if (lengthptr != NULL)
8553 code = *codeptr + 1 + LINK_SIZE + skipbytes;
8554 length += 1 + LINK_SIZE;
8559 PUT(code, 1, (int)(code - last_branch));
8560 bc.current_branch = last_branch = code;
8561 code += 1 + LINK_SIZE;
8566 /* Control never reaches here */
8572 /*************************************************
8573 * Check for anchored expression *
8574 *************************************************/
8576 /* Try to find out if this is an anchored regular expression. Consider each
8577 alternative branch. If they all start with OP_SOD or OP_CIRC, or with a bracket
8578 all of whose alternatives start with OP_SOD or OP_CIRC (recurse ad lib), then
8579 it's anchored. However, if this is a multiline pattern, then only OP_SOD will
8580 be found, because ^ generates OP_CIRCM in that mode.
8582 We can also consider a regex to be anchored if OP_SOM starts all its branches.
8583 This is the code for \G, which means "match at start of match position, taking
8584 into account the match offset".
8586 A branch is also implicitly anchored if it starts with .* and DOTALL is set,
8587 because that will try the rest of the pattern at all possible matching points,
8588 so there is no point trying again.... er ....
8590 .... except when the .* appears inside capturing parentheses, and there is a
8591 subsequent back reference to those parentheses. We haven't enough information
8592 to catch that case precisely.
8594 At first, the best we could do was to detect when .* was in capturing brackets
8595 and the highest back reference was greater than or equal to that level.
8596 However, by keeping a bitmap of the first 31 back references, we can catch some
8597 of the more common cases more precisely.
8599 ... A second exception is when the .* appears inside an atomic group, because
8600 this prevents the number of characters it matches from being adjusted.
8603 code points to start of expression (the bracket)
8604 bracket_map a bitmap of which brackets we are inside while testing; this
8605 handles up to substring 31; after that we just have to take
8606 the less precise approach
8607 cd points to the compile data block
8608 atomcount atomic group level
8610 Returns: TRUE or FALSE
8614 is_anchored(register const pcre_uchar *code, unsigned int bracket_map,
8615 compile_data *cd, int atomcount)
8618 const pcre_uchar *scode = first_significant_code(
8619 code + PRIV(OP_lengths)[*code], FALSE);
8620 register int op = *scode;
8622 /* Non-capturing brackets */
8624 if (op == OP_BRA || op == OP_BRAPOS ||
8625 op == OP_SBRA || op == OP_SBRAPOS)
8627 if (!is_anchored(scode, bracket_map, cd, atomcount)) return FALSE;
8630 /* Capturing brackets */
8632 else if (op == OP_CBRA || op == OP_CBRAPOS ||
8633 op == OP_SCBRA || op == OP_SCBRAPOS)
8635 int n = GET2(scode, 1+LINK_SIZE);
8636 int new_map = bracket_map | ((n < 32)? (1 << n) : 1);
8637 if (!is_anchored(scode, new_map, cd, atomcount)) return FALSE;
8640 /* Positive forward assertions and conditions */
8642 else if (op == OP_ASSERT || op == OP_COND)
8644 if (!is_anchored(scode, bracket_map, cd, atomcount)) return FALSE;
8649 else if (op == OP_ONCE || op == OP_ONCE_NC)
8651 if (!is_anchored(scode, bracket_map, cd, atomcount + 1))
8655 /* .* is not anchored unless DOTALL is set (which generates OP_ALLANY) and
8656 it isn't in brackets that are or may be referenced or inside an atomic
8659 else if ((op == OP_TYPESTAR || op == OP_TYPEMINSTAR ||
8660 op == OP_TYPEPOSSTAR))
8662 if (scode[1] != OP_ALLANY || (bracket_map & cd->backref_map) != 0 ||
8663 atomcount > 0 || cd->had_pruneorskip)
8667 /* Check for explicit anchoring */
8669 else if (op != OP_SOD && op != OP_SOM && op != OP_CIRC) return FALSE;
8671 code += GET(code, 1);
8673 while (*code == OP_ALT); /* Loop for each alternative */
8679 /*************************************************
8680 * Check for starting with ^ or .* *
8681 *************************************************/
8683 /* This is called to find out if every branch starts with ^ or .* so that
8684 "first char" processing can be done to speed things up in multiline
8685 matching and for non-DOTALL patterns that start with .* (which must start at
8686 the beginning or after \n). As in the case of is_anchored() (see above), we
8687 have to take account of back references to capturing brackets that contain .*
8688 because in that case we can't make the assumption. Also, the appearance of .*
8689 inside atomic brackets or in a pattern that contains *PRUNE or *SKIP does not
8690 count, because once again the assumption no longer holds.
8693 code points to start of expression (the bracket)
8694 bracket_map a bitmap of which brackets we are inside while testing; this
8695 handles up to substring 31; after that we just have to take
8696 the less precise approach
8697 cd points to the compile data
8698 atomcount atomic group level
8700 Returns: TRUE or FALSE
8704 is_startline(const pcre_uchar *code, unsigned int bracket_map,
8705 compile_data *cd, int atomcount)
8708 const pcre_uchar *scode = first_significant_code(
8709 code + PRIV(OP_lengths)[*code], FALSE);
8710 register int op = *scode;
8712 /* If we are at the start of a conditional assertion group, *both* the
8713 conditional assertion *and* what follows the condition must satisfy the test
8714 for start of line. Other kinds of condition fail. Note that there may be an
8715 auto-callout at the start of a condition. */
8719 scode += 1 + LINK_SIZE;
8720 if (*scode == OP_CALLOUT) scode += PRIV(OP_lengths)[OP_CALLOUT];
8731 default: /* Assertion */
8732 if (!is_startline(scode, bracket_map, cd, atomcount)) return FALSE;
8733 do scode += GET(scode, 1); while (*scode == OP_ALT);
8734 scode += 1 + LINK_SIZE;
8737 scode = first_significant_code(scode, FALSE);
8741 /* Non-capturing brackets */
8743 if (op == OP_BRA || op == OP_BRAPOS ||
8744 op == OP_SBRA || op == OP_SBRAPOS)
8746 if (!is_startline(scode, bracket_map, cd, atomcount)) return FALSE;
8749 /* Capturing brackets */
8751 else if (op == OP_CBRA || op == OP_CBRAPOS ||
8752 op == OP_SCBRA || op == OP_SCBRAPOS)
8754 int n = GET2(scode, 1+LINK_SIZE);
8755 int new_map = bracket_map | ((n < 32)? (1 << n) : 1);
8756 if (!is_startline(scode, new_map, cd, atomcount)) return FALSE;
8759 /* Positive forward assertions */
8761 else if (op == OP_ASSERT)
8763 if (!is_startline(scode, bracket_map, cd, atomcount)) return FALSE;
8766 /* Atomic brackets */
8768 else if (op == OP_ONCE || op == OP_ONCE_NC)
8770 if (!is_startline(scode, bracket_map, cd, atomcount + 1)) return FALSE;
8773 /* .* means "start at start or after \n" if it isn't in atomic brackets or
8774 brackets that may be referenced, as long as the pattern does not contain
8775 *PRUNE or *SKIP, because these break the feature. Consider, for example,
8776 /.*?a(*PRUNE)b/ with the subject "aab", which matches "ab", i.e. not at the
8779 else if (op == OP_TYPESTAR || op == OP_TYPEMINSTAR || op == OP_TYPEPOSSTAR)
8781 if (scode[1] != OP_ANY || (bracket_map & cd->backref_map) != 0 ||
8782 atomcount > 0 || cd->had_pruneorskip)
8786 /* Check for explicit circumflex; anything else gives a FALSE result. Note
8787 in particular that this includes atomic brackets OP_ONCE and OP_ONCE_NC
8788 because the number of characters matched by .* cannot be adjusted inside
8791 else if (op != OP_CIRC && op != OP_CIRCM) return FALSE;
8793 /* Move on to the next alternative */
8795 code += GET(code, 1);
8797 while (*code == OP_ALT); /* Loop for each alternative */
8803 /*************************************************
8804 * Check for asserted fixed first char *
8805 *************************************************/
8807 /* During compilation, the "first char" settings from forward assertions are
8808 discarded, because they can cause conflicts with actual literals that follow.
8809 However, if we end up without a first char setting for an unanchored pattern,
8810 it is worth scanning the regex to see if there is an initial asserted first
8811 char. If all branches start with the same asserted char, or with a
8812 non-conditional bracket all of whose alternatives start with the same asserted
8813 char (recurse ad lib), then we return that char, with the flags set to zero or
8814 REQ_CASELESS; otherwise return zero with REQ_NONE in the flags.
8817 code points to start of expression (the bracket)
8818 flags points to the first char flags, or to REQ_NONE
8819 inassert TRUE if in an assertion
8821 Returns: the fixed first char, or 0 with REQ_NONE in flags
8825 find_firstassertedchar(const pcre_uchar *code, pcre_int32 *flags,
8828 register pcre_uint32 c = 0;
8829 int cflags = REQ_NONE;
8835 int xl = (*code == OP_CBRA || *code == OP_SCBRA ||
8836 *code == OP_CBRAPOS || *code == OP_SCBRAPOS)? IMM2_SIZE:0;
8837 const pcre_uchar *scode = first_significant_code(code + 1+LINK_SIZE + xl,
8839 register pcre_uchar op = *scode;
8855 d = find_firstassertedchar(scode, &dflags, op == OP_ASSERT);
8858 if (cflags < 0) { c = d; cflags = dflags; } else if (c != d || cflags != dflags) return 0;
8869 if (!inassert) return 0;
8870 if (cflags < 0) { c = scode[1]; cflags = 0; }
8871 else if (c != scode[1]) return 0;
8882 if (!inassert) return 0;
8883 if (cflags < 0) { c = scode[1]; cflags = REQ_CASELESS; }
8884 else if (c != scode[1]) return 0;
8888 code += GET(code, 1);
8890 while (*code == OP_ALT);
8898 /*************************************************
8899 * Add an entry to the name/number table *
8900 *************************************************/
8902 /* This function is called between compiling passes to add an entry to the
8903 name/number table, maintaining alphabetical order. Checking for permitted
8904 and forbidden duplicates has already been done.
8907 cd the compile data block
8908 name the name to add
8909 length the length of the name
8910 groupno the group number
8916 add_name(compile_data *cd, const pcre_uchar *name, int length,
8917 unsigned int groupno)
8920 pcre_uchar *slot = cd->name_table;
8922 for (i = 0; i < cd->names_found; i++)
8924 int crc = memcmp(name, slot+IMM2_SIZE, IN_UCHARS(length));
8925 if (crc == 0 && slot[IMM2_SIZE+length] != 0)
8926 crc = -1; /* Current name is a substring */
8928 /* Make space in the table and break the loop for an earlier name. For a
8929 duplicate or later name, carry on. We do this for duplicates so that in the
8930 simple case (when ?(| is not used) they are in order of their numbers. In all
8931 cases they are in the order in which they appear in the pattern. */
8935 memmove(slot + cd->name_entry_size, slot,
8936 IN_UCHARS((cd->names_found - i) * cd->name_entry_size));
8940 /* Continue the loop for a later or duplicate name */
8942 slot += cd->name_entry_size;
8945 PUT2(slot, 0, groupno);
8946 memcpy(slot + IMM2_SIZE, name, IN_UCHARS(length));
8947 slot[IMM2_SIZE + length] = 0;
8953 /*************************************************
8954 * Compile a Regular Expression *
8955 *************************************************/
8957 /* This function takes a string and returns a pointer to a block of store
8958 holding a compiled version of the expression. The original API for this
8959 function had no error code return variable; it is retained for backwards
8960 compatibility. The new function is given a new name.
8963 pattern the regular expression
8964 options various option bits
8965 errorcodeptr pointer to error code variable (pcre_compile2() only)
8966 can be NULL if you don't want a code value
8967 errorptr pointer to pointer to error text
8968 erroroffset ptr offset in pattern where error was detected
8969 tables pointer to character tables or NULL
8971 Returns: pointer to compiled data block, or NULL on error,
8972 with errorptr and erroroffset set
8975 #if defined COMPILE_PCRE8
8976 PCRE_EXP_DEFN pcre * PCRE_CALL_CONVENTION
8977 pcre_compile(const char *pattern, int options, const char **errorptr,
8978 int *erroroffset, const unsigned char *tables)
8979 #elif defined COMPILE_PCRE16
8980 PCRE_EXP_DEFN pcre16 * PCRE_CALL_CONVENTION
8981 pcre16_compile(PCRE_SPTR16 pattern, int options, const char **errorptr,
8982 int *erroroffset, const unsigned char *tables)
8983 #elif defined COMPILE_PCRE32
8984 PCRE_EXP_DEFN pcre32 * PCRE_CALL_CONVENTION
8985 pcre32_compile(PCRE_SPTR32 pattern, int options, const char **errorptr,
8986 int *erroroffset, const unsigned char *tables)
8989 #if defined COMPILE_PCRE8
8990 return pcre_compile2(pattern, options, NULL, errorptr, erroroffset, tables);
8991 #elif defined COMPILE_PCRE16
8992 return pcre16_compile2(pattern, options, NULL, errorptr, erroroffset, tables);
8993 #elif defined COMPILE_PCRE32
8994 return pcre32_compile2(pattern, options, NULL, errorptr, erroroffset, tables);
8999 #if defined COMPILE_PCRE8
9000 PCRE_EXP_DEFN pcre * PCRE_CALL_CONVENTION
9001 pcre_compile2(const char *pattern, int options, int *errorcodeptr,
9002 const char **errorptr, int *erroroffset, const unsigned char *tables)
9003 #elif defined COMPILE_PCRE16
9004 PCRE_EXP_DEFN pcre16 * PCRE_CALL_CONVENTION
9005 pcre16_compile2(PCRE_SPTR16 pattern, int options, int *errorcodeptr,
9006 const char **errorptr, int *erroroffset, const unsigned char *tables)
9007 #elif defined COMPILE_PCRE32
9008 PCRE_EXP_DEFN pcre32 * PCRE_CALL_CONVENTION
9009 pcre32_compile2(PCRE_SPTR32 pattern, int options, int *errorcodeptr,
9010 const char **errorptr, int *erroroffset, const unsigned char *tables)
9014 int length = 1; /* For final END opcode */
9015 pcre_int32 firstcharflags, reqcharflags;
9016 pcre_uint32 firstchar, reqchar;
9017 pcre_uint32 limit_match = PCRE_UINT32_MAX;
9018 pcre_uint32 limit_recursion = PCRE_UINT32_MAX;
9021 int skipatstart = 0;
9023 BOOL never_utf = FALSE;
9026 const pcre_uchar *codestart;
9027 const pcre_uchar *ptr;
9028 compile_data compile_block;
9029 compile_data *cd = &compile_block;
9031 /* This space is used for "compiling" into during the first phase, when we are
9032 computing the amount of memory that is needed. Compiled items are thrown away
9033 as soon as possible, so that a fairly large buffer should be sufficient for
9034 this purpose. The same space is used in the second phase for remembering where
9035 to fill in forward references to subpatterns. That may overflow, in which case
9036 new memory is obtained from malloc(). */
9038 pcre_uchar cworkspace[COMPILE_WORK_SIZE];
9040 /* This vector is used for remembering name groups during the pre-compile. In a
9041 similar way to cworkspace, it can be expanded using malloc() if necessary. */
9043 named_group named_groups[NAMED_GROUP_LIST_SIZE];
9045 /* Set this early so that early errors get offset 0. */
9047 ptr = (const pcre_uchar *)pattern;
9049 /* We can't pass back an error message if errorptr is NULL; I guess the best we
9050 can do is just return NULL, but we can set a code value if there is a code
9053 if (errorptr == NULL)
9055 if (errorcodeptr != NULL) *errorcodeptr = 99;
9060 if (errorcodeptr != NULL) *errorcodeptr = ERR0;
9062 /* However, we can give a message for this error */
9064 if (erroroffset == NULL)
9067 goto PCRE_EARLY_ERROR_RETURN2;
9072 /* Set up pointers to the individual character tables */
9074 if (tables == NULL) tables = PRIV(default_tables);
9075 cd->lcc = tables + lcc_offset;
9076 cd->fcc = tables + fcc_offset;
9077 cd->cbits = tables + cbits_offset;
9078 cd->ctypes = tables + ctypes_offset;
9080 /* Check that all undefined public option bits are zero */
9082 if ((options & ~PUBLIC_COMPILE_OPTIONS) != 0)
9085 goto PCRE_EARLY_ERROR_RETURN;
9088 /* If PCRE_NEVER_UTF is set, remember it. */
9090 if ((options & PCRE_NEVER_UTF) != 0) never_utf = TRUE;
9092 /* Check for global one-time settings at the start of the pattern, and remember
9093 the offset for later. */
9095 cd->external_flags = 0; /* Initialize here for LIMIT_MATCH/RECURSION */
9097 while (ptr[skipatstart] == CHAR_LEFT_PARENTHESIS &&
9098 ptr[skipatstart+1] == CHAR_ASTERISK)
9103 /* For completeness and backward compatibility, (*UTFn) is supported in the
9104 relevant libraries, but (*UTF) is generic and always supported. Note that
9105 PCRE_UTF8 == PCRE_UTF16 == PCRE_UTF32. */
9107 #ifdef COMPILE_PCRE8
9108 if (STRNCMP_UC_C8(ptr+skipatstart+2, STRING_UTF8_RIGHTPAR, 5) == 0)
9109 { skipatstart += 7; options |= PCRE_UTF8; continue; }
9111 #ifdef COMPILE_PCRE16
9112 if (STRNCMP_UC_C8(ptr+skipatstart+2, STRING_UTF16_RIGHTPAR, 6) == 0)
9113 { skipatstart += 8; options |= PCRE_UTF16; continue; }
9115 #ifdef COMPILE_PCRE32
9116 if (STRNCMP_UC_C8(ptr+skipatstart+2, STRING_UTF32_RIGHTPAR, 6) == 0)
9117 { skipatstart += 8; options |= PCRE_UTF32; continue; }
9120 else if (STRNCMP_UC_C8(ptr+skipatstart+2, STRING_UTF_RIGHTPAR, 4) == 0)
9121 { skipatstart += 6; options |= PCRE_UTF8; continue; }
9122 else if (STRNCMP_UC_C8(ptr+skipatstart+2, STRING_UCP_RIGHTPAR, 4) == 0)
9123 { skipatstart += 6; options |= PCRE_UCP; continue; }
9124 else if (STRNCMP_UC_C8(ptr+skipatstart+2, STRING_NO_AUTO_POSSESS_RIGHTPAR, 16) == 0)
9125 { skipatstart += 18; options |= PCRE_NO_AUTO_POSSESS; continue; }
9126 else if (STRNCMP_UC_C8(ptr+skipatstart+2, STRING_NO_START_OPT_RIGHTPAR, 13) == 0)
9127 { skipatstart += 15; options |= PCRE_NO_START_OPTIMIZE; continue; }
9129 else if (STRNCMP_UC_C8(ptr+skipatstart+2, STRING_LIMIT_MATCH_EQ, 12) == 0)
9132 int p = skipatstart + 14;
9133 while (isdigit(ptr[p]))
9135 if (c > PCRE_UINT32_MAX / 10 - 1) break; /* Integer overflow */
9136 c = c*10 + ptr[p++] - CHAR_0;
9138 if (ptr[p++] != CHAR_RIGHT_PARENTHESIS) break;
9139 if (c < limit_match)
9142 cd->external_flags |= PCRE_MLSET;
9148 else if (STRNCMP_UC_C8(ptr+skipatstart+2, STRING_LIMIT_RECURSION_EQ, 16) == 0)
9151 int p = skipatstart + 18;
9152 while (isdigit(ptr[p]))
9154 if (c > PCRE_UINT32_MAX / 10 - 1) break; /* Integer overflow check */
9155 c = c*10 + ptr[p++] - CHAR_0;
9157 if (ptr[p++] != CHAR_RIGHT_PARENTHESIS) break;
9158 if (c < limit_recursion)
9160 limit_recursion = c;
9161 cd->external_flags |= PCRE_RLSET;
9167 if (STRNCMP_UC_C8(ptr+skipatstart+2, STRING_CR_RIGHTPAR, 3) == 0)
9168 { skipatstart += 5; newnl = PCRE_NEWLINE_CR; }
9169 else if (STRNCMP_UC_C8(ptr+skipatstart+2, STRING_LF_RIGHTPAR, 3) == 0)
9170 { skipatstart += 5; newnl = PCRE_NEWLINE_LF; }
9171 else if (STRNCMP_UC_C8(ptr+skipatstart+2, STRING_CRLF_RIGHTPAR, 5) == 0)
9172 { skipatstart += 7; newnl = PCRE_NEWLINE_CR + PCRE_NEWLINE_LF; }
9173 else if (STRNCMP_UC_C8(ptr+skipatstart+2, STRING_ANY_RIGHTPAR, 4) == 0)
9174 { skipatstart += 6; newnl = PCRE_NEWLINE_ANY; }
9175 else if (STRNCMP_UC_C8(ptr+skipatstart+2, STRING_ANYCRLF_RIGHTPAR, 8) == 0)
9176 { skipatstart += 10; newnl = PCRE_NEWLINE_ANYCRLF; }
9178 else if (STRNCMP_UC_C8(ptr+skipatstart+2, STRING_BSR_ANYCRLF_RIGHTPAR, 12) == 0)
9179 { skipatstart += 14; newbsr = PCRE_BSR_ANYCRLF; }
9180 else if (STRNCMP_UC_C8(ptr+skipatstart+2, STRING_BSR_UNICODE_RIGHTPAR, 12) == 0)
9181 { skipatstart += 14; newbsr = PCRE_BSR_UNICODE; }
9184 options = (options & ~PCRE_NEWLINE_BITS) | newnl;
9185 else if (newbsr != 0)
9186 options = (options & ~(PCRE_BSR_ANYCRLF|PCRE_BSR_UNICODE)) | newbsr;
9190 /* PCRE_UTF(16|32) have the same value as PCRE_UTF8. */
9191 utf = (options & PCRE_UTF8) != 0;
9192 if (utf && never_utf)
9195 goto PCRE_EARLY_ERROR_RETURN2;
9198 /* Can't support UTF unless PCRE has been compiled to include the code. The
9199 return of an error code from PRIV(valid_utf)() is a new feature, introduced in
9200 release 8.13. It is passed back from pcre_[dfa_]exec(), but at the moment is
9204 if (utf && (options & PCRE_NO_UTF8_CHECK) == 0 &&
9205 (errorcode = PRIV(valid_utf)((PCRE_PUCHAR)pattern, -1, erroroffset)) != 0)
9207 #if defined COMPILE_PCRE8
9209 #elif defined COMPILE_PCRE16
9211 #elif defined COMPILE_PCRE32
9214 goto PCRE_EARLY_ERROR_RETURN2;
9220 goto PCRE_EARLY_ERROR_RETURN;
9224 /* Can't support UCP unless PCRE has been compiled to include the code. */
9227 if ((options & PCRE_UCP) != 0)
9230 goto PCRE_EARLY_ERROR_RETURN;
9234 /* Check validity of \R options. */
9236 if ((options & (PCRE_BSR_ANYCRLF|PCRE_BSR_UNICODE)) ==
9237 (PCRE_BSR_ANYCRLF|PCRE_BSR_UNICODE))
9240 goto PCRE_EARLY_ERROR_RETURN;
9243 /* Handle different types of newline. The three bits give seven cases. The
9244 current code allows for fixed one- or two-byte sequences, plus "any" and
9247 switch (options & PCRE_NEWLINE_BITS)
9249 case 0: newline = NEWLINE; break; /* Build-time default */
9250 case PCRE_NEWLINE_CR: newline = CHAR_CR; break;
9251 case PCRE_NEWLINE_LF: newline = CHAR_NL; break;
9252 case PCRE_NEWLINE_CR+
9253 PCRE_NEWLINE_LF: newline = (CHAR_CR << 8) | CHAR_NL; break;
9254 case PCRE_NEWLINE_ANY: newline = -1; break;
9255 case PCRE_NEWLINE_ANYCRLF: newline = -2; break;
9256 default: errorcode = ERR56; goto PCRE_EARLY_ERROR_RETURN;
9261 cd->nltype = NLTYPE_ANYCRLF;
9263 else if (newline < 0)
9265 cd->nltype = NLTYPE_ANY;
9269 cd->nltype = NLTYPE_FIXED;
9273 cd->nl[0] = (newline >> 8) & 255;
9274 cd->nl[1] = newline & 255;
9279 cd->nl[0] = newline;
9283 /* Maximum back reference and backref bitmap. The bitmap records up to 31 back
9284 references to help in deciding whether (.*) can be treated as anchored or not.
9287 cd->top_backref = 0;
9288 cd->backref_map = 0;
9290 /* Reflect pattern for debugging output */
9292 DPRINTF(("------------------------------------------------------------------\n"));
9294 print_puchar(stdout, (PCRE_PUCHAR)pattern);
9298 /* Pretend to compile the pattern while actually just accumulating the length
9299 of memory required. This behaviour is triggered by passing a non-NULL final
9300 argument to compile_regex(). We pass a block of workspace (cworkspace) for it
9301 to compile parts of the pattern into; the compiled code is discarded when it is
9302 no longer needed, so hopefully this workspace will never overflow, though there
9303 is a test for its doing so. */
9305 cd->bracount = cd->final_bracount = 0;
9306 cd->names_found = 0;
9307 cd->name_entry_size = 0;
9308 cd->name_table = NULL;
9309 cd->dupnames = FALSE;
9310 cd->dupgroups = FALSE;
9311 cd->namedrefcount = 0;
9312 cd->start_code = cworkspace;
9313 cd->hwm = cworkspace;
9314 cd->iscondassert = FALSE;
9315 cd->start_workspace = cworkspace;
9316 cd->workspace_size = COMPILE_WORK_SIZE;
9317 cd->named_groups = named_groups;
9318 cd->named_group_list_size = NAMED_GROUP_LIST_SIZE;
9319 cd->start_pattern = (const pcre_uchar *)pattern;
9320 cd->end_pattern = (const pcre_uchar *)(pattern + STRLEN_UC((const pcre_uchar *)pattern));
9321 cd->req_varyopt = 0;
9322 cd->parens_depth = 0;
9323 cd->assert_depth = 0;
9324 cd->max_lookbehind = 0;
9325 cd->external_options = options;
9326 cd->open_caps = NULL;
9328 /* Now do the pre-compile. On error, errorcode will be set non-zero, so we
9329 don't need to look at the result of the function here. The initial options have
9330 been put into the cd block so that they can be changed if an option setting is
9331 found within the regex right at the beginning. Bringing initial option settings
9332 outside can help speed up starting point checks. */
9338 (void)compile_regex(cd->external_options, &code, &ptr, &errorcode, FALSE,
9339 FALSE, 0, 0, &firstchar, &firstcharflags, &reqchar, &reqcharflags, NULL,
9341 if (errorcode != 0) goto PCRE_EARLY_ERROR_RETURN;
9343 DPRINTF(("end pre-compile: length=%d workspace=%d\n", length,
9344 (int)(cd->hwm - cworkspace)));
9346 if (length > MAX_PATTERN_SIZE)
9349 goto PCRE_EARLY_ERROR_RETURN;
9352 /* Compute the size of the data block for storing the compiled pattern. Integer
9353 overflow should no longer be possible because nowadays we limit the maximum
9354 value of cd->names_found and cd->name_entry_size. */
9356 size = sizeof(REAL_PCRE) +
9357 (length + cd->names_found * cd->name_entry_size) * sizeof(pcre_uchar);
9359 /* Get the memory. */
9361 re = (REAL_PCRE *)(PUBL(malloc))(size);
9365 goto PCRE_EARLY_ERROR_RETURN;
9368 /* Put in the magic number, and save the sizes, initial options, internal
9369 flags, and character table pointer. NULL is used for the default character
9370 tables. The nullpad field is at the end; it's there to help in the case when a
9371 regex compiled on a system with 4-byte pointers is run on another with 8-byte
9374 re->magic_number = MAGIC_NUMBER;
9375 re->size = (int)size;
9376 re->options = cd->external_options;
9377 re->flags = cd->external_flags;
9378 re->limit_match = limit_match;
9379 re->limit_recursion = limit_recursion;
9382 re->name_table_offset = sizeof(REAL_PCRE) / sizeof(pcre_uchar);
9383 re->name_entry_size = cd->name_entry_size;
9384 re->name_count = cd->names_found;
9386 re->tables = (tables == PRIV(default_tables))? NULL : tables;
9388 #ifdef COMPILE_PCRE32
9391 re->dummy1 = re->dummy2 = re->dummy3 = 0;
9394 /* The starting points of the name/number translation table and of the code are
9395 passed around in the compile data block. The start/end pattern and initial
9396 options are already set from the pre-compile phase, as is the name_entry_size
9397 field. Reset the bracket count and the names_found field. Also reset the hwm
9398 field; this time it's used for remembering forward references to subpatterns.
9401 cd->final_bracount = cd->bracount; /* Save for checking forward references */
9402 cd->parens_depth = 0;
9403 cd->assert_depth = 0;
9405 cd->max_lookbehind = 0;
9406 cd->name_table = (pcre_uchar *)re + re->name_table_offset;
9407 codestart = cd->name_table + re->name_entry_size * re->name_count;
9408 cd->start_code = codestart;
9409 cd->hwm = (pcre_uchar *)(cd->start_workspace);
9410 cd->iscondassert = FALSE;
9411 cd->req_varyopt = 0;
9412 cd->had_accept = FALSE;
9413 cd->had_pruneorskip = FALSE;
9414 cd->check_lookbehind = FALSE;
9415 cd->open_caps = NULL;
9417 /* If any named groups were found, create the name/number table from the list
9418 created in the first pass. */
9420 if (cd->names_found > 0)
9422 int i = cd->names_found;
9423 named_group *ng = cd->named_groups;
9424 cd->names_found = 0;
9425 for (; i > 0; i--, ng++)
9426 add_name(cd, ng->name, ng->length, ng->number);
9427 if (cd->named_group_list_size > NAMED_GROUP_LIST_SIZE)
9428 (PUBL(free))((void *)cd->named_groups);
9431 /* Set up a starting, non-extracting bracket, then compile the expression. On
9432 error, errorcode will be set non-zero, so we don't need to look at the result
9433 of the function here. */
9435 ptr = (const pcre_uchar *)pattern + skipatstart;
9436 code = (pcre_uchar *)codestart;
9438 (void)compile_regex(re->options, &code, &ptr, &errorcode, FALSE, FALSE, 0, 0,
9439 &firstchar, &firstcharflags, &reqchar, &reqcharflags, NULL, cd, NULL);
9440 re->top_bracket = cd->bracount;
9441 re->top_backref = cd->top_backref;
9442 re->max_lookbehind = cd->max_lookbehind;
9443 re->flags = cd->external_flags | PCRE_MODE;
9447 reqchar = 0; /* Must disable after (*ACCEPT) */
9448 reqcharflags = REQ_NONE;
9451 /* If not reached end of pattern on success, there's an excess bracket. */
9453 if (errorcode == 0 && *ptr != CHAR_NULL) errorcode = ERR22;
9455 /* Fill in the terminating state and check for disastrous overflow, but
9456 if debugging, leave the test till after things are printed out. */
9461 if (code - codestart > length) errorcode = ERR23;
9464 #ifdef SUPPORT_VALGRIND
9465 /* If the estimated length exceeds the really used length, mark the extra
9466 allocated memory as unaddressable, so that any out-of-bound reads can be
9468 VALGRIND_MAKE_MEM_NOACCESS(code, (length - (code - codestart)) * sizeof(pcre_uchar));
9471 /* Fill in any forward references that are required. There may be repeated
9472 references; optimize for them, as searching a large regex takes time. */
9474 if (cd->hwm > cd->start_workspace)
9476 int prev_recno = -1;
9477 const pcre_uchar *groupptr = NULL;
9478 while (errorcode == 0 && cd->hwm > cd->start_workspace)
9481 cd->hwm -= LINK_SIZE;
9482 offset = GET(cd->hwm, 0);
9484 /* Check that the hwm handling hasn't gone wrong. This whole area is
9485 rewritten in PCRE2 because there are some obscure cases. */
9487 if (offset == 0 || codestart[offset-1] != OP_RECURSE)
9493 recno = GET(codestart, offset);
9494 if (recno != prev_recno)
9496 groupptr = PRIV(find_bracket)(codestart, utf, recno);
9499 if (groupptr == NULL) errorcode = ERR53;
9500 else PUT(((pcre_uchar *)codestart), offset, (int)(groupptr - codestart));
9504 /* If the workspace had to be expanded, free the new memory. Set the pointer to
9505 NULL to indicate that forward references have been filled in. */
9507 if (cd->workspace_size > COMPILE_WORK_SIZE)
9508 (PUBL(free))((void *)cd->start_workspace);
9509 cd->start_workspace = NULL;
9511 /* Give an error if there's back reference to a non-existent capturing
9514 if (errorcode == 0 && re->top_backref > re->top_bracket) errorcode = ERR15;
9516 /* Unless disabled, check whether any single character iterators can be
9517 auto-possessified. The function overwrites the appropriate opcode values, so
9518 the type of the pointer must be cast. NOTE: the intermediate variable "temp" is
9519 used in this code because at least one compiler gives a warning about loss of
9520 "const" attribute if the cast (pcre_uchar *)codestart is used directly in the
9523 if (errorcode == 0 && (options & PCRE_NO_AUTO_POSSESS) == 0)
9525 pcre_uchar *temp = (pcre_uchar *)codestart;
9526 auto_possessify(temp, utf, cd);
9529 /* If there were any lookbehind assertions that contained OP_RECURSE
9530 (recursions or subroutine calls), a flag is set for them to be checked here,
9531 because they may contain forward references. Actual recursions cannot be fixed
9532 length, but subroutine calls can. It is done like this so that those without
9533 OP_RECURSE that are not fixed length get a diagnosic with a useful offset. The
9534 exceptional ones forgo this. We scan the pattern to check that they are fixed
9535 length, and set their lengths. */
9537 if (errorcode == 0 && cd->check_lookbehind)
9539 pcre_uchar *cc = (pcre_uchar *)codestart;
9541 /* Loop, searching for OP_REVERSE items, and process those that do not have
9542 their length set. (Actually, it will also re-process any that have a length
9543 of zero, but that is a pathological case, and it does no harm.) When we find
9544 one, we temporarily terminate the branch it is in while we scan it. */
9546 for (cc = (pcre_uchar *)PRIV(find_bracket)(codestart, utf, -1);
9548 cc = (pcre_uchar *)PRIV(find_bracket)(cc, utf, -1))
9550 if (GET(cc, 1) == 0)
9553 pcre_uchar *be = cc - 1 - LINK_SIZE + GET(cc, -LINK_SIZE);
9556 fixed_length = find_fixedlength(cc, (re->options & PCRE_UTF8) != 0, TRUE,
9559 DPRINTF(("fixed length = %d\n", fixed_length));
9560 if (fixed_length < 0)
9562 errorcode = (fixed_length == -2)? ERR36 :
9563 (fixed_length == -4)? ERR70 : ERR25;
9566 if (fixed_length > cd->max_lookbehind) cd->max_lookbehind = fixed_length;
9567 PUT(cc, 1, fixed_length);
9569 cc += 1 + LINK_SIZE;
9573 /* Failed to compile, or error while post-processing */
9578 PCRE_EARLY_ERROR_RETURN:
9579 *erroroffset = (int)(ptr - (const pcre_uchar *)pattern);
9580 PCRE_EARLY_ERROR_RETURN2:
9581 *errorptr = find_error_text(errorcode);
9582 if (errorcodeptr != NULL) *errorcodeptr = errorcode;
9586 /* If the anchored option was not passed, set the flag if we can determine that
9587 the pattern is anchored by virtue of ^ characters or \A or anything else, such
9588 as starting with non-atomic .* when DOTALL is set and there are no occurrences
9591 Otherwise, if we know what the first byte has to be, save it, because that
9592 speeds up unanchored matches no end. If not, see if we can set the
9593 PCRE_STARTLINE flag. This is helpful for multiline matches when all branches
9594 start with ^. and also when all branches start with non-atomic .* for
9595 non-DOTALL matches when *PRUNE and SKIP are not present. */
9597 if ((re->options & PCRE_ANCHORED) == 0)
9599 if (is_anchored(codestart, 0, cd, 0)) re->options |= PCRE_ANCHORED;
9602 if (firstcharflags < 0)
9603 firstchar = find_firstassertedchar(codestart, &firstcharflags, FALSE);
9604 if (firstcharflags >= 0) /* Remove caseless flag for non-caseable chars */
9606 #if defined COMPILE_PCRE8
9607 re->first_char = firstchar & 0xff;
9608 #elif defined COMPILE_PCRE16
9609 re->first_char = firstchar & 0xffff;
9610 #elif defined COMPILE_PCRE32
9611 re->first_char = firstchar;
9613 if ((firstcharflags & REQ_CASELESS) != 0)
9615 #if defined SUPPORT_UCP && !(defined COMPILE_PCRE8)
9616 /* We ignore non-ASCII first chars in 8 bit mode. */
9619 if (re->first_char < 128)
9621 if (cd->fcc[re->first_char] != re->first_char)
9622 re->flags |= PCRE_FCH_CASELESS;
9624 else if (UCD_OTHERCASE(re->first_char) != re->first_char)
9625 re->flags |= PCRE_FCH_CASELESS;
9629 if (MAX_255(re->first_char)
9630 && cd->fcc[re->first_char] != re->first_char)
9631 re->flags |= PCRE_FCH_CASELESS;
9634 re->flags |= PCRE_FIRSTSET;
9637 else if (is_startline(codestart, 0, cd, 0)) re->flags |= PCRE_STARTLINE;
9641 /* For an anchored pattern, we use the "required byte" only if it follows a
9642 variable length item in the regex. Remove the caseless flag for non-caseable
9645 if (reqcharflags >= 0 &&
9646 ((re->options & PCRE_ANCHORED) == 0 || (reqcharflags & REQ_VARY) != 0))
9648 #if defined COMPILE_PCRE8
9649 re->req_char = reqchar & 0xff;
9650 #elif defined COMPILE_PCRE16
9651 re->req_char = reqchar & 0xffff;
9652 #elif defined COMPILE_PCRE32
9653 re->req_char = reqchar;
9655 if ((reqcharflags & REQ_CASELESS) != 0)
9657 #if defined SUPPORT_UCP && !(defined COMPILE_PCRE8)
9658 /* We ignore non-ASCII first chars in 8 bit mode. */
9661 if (re->req_char < 128)
9663 if (cd->fcc[re->req_char] != re->req_char)
9664 re->flags |= PCRE_RCH_CASELESS;
9666 else if (UCD_OTHERCASE(re->req_char) != re->req_char)
9667 re->flags |= PCRE_RCH_CASELESS;
9671 if (MAX_255(re->req_char) && cd->fcc[re->req_char] != re->req_char)
9672 re->flags |= PCRE_RCH_CASELESS;
9675 re->flags |= PCRE_REQCHSET;
9678 /* Print out the compiled data if debugging is enabled. This is never the
9679 case when building a production library. */
9682 printf("Length = %d top_bracket = %d top_backref = %d\n",
9683 length, re->top_bracket, re->top_backref);
9685 printf("Options=%08x\n", re->options);
9687 if ((re->flags & PCRE_FIRSTSET) != 0)
9689 pcre_uchar ch = re->first_char;
9690 const char *caseless =
9691 ((re->flags & PCRE_FCH_CASELESS) == 0)? "" : " (caseless)";
9692 if (PRINTABLE(ch)) printf("First char = %c%s\n", ch, caseless);
9693 else printf("First char = \\x%02x%s\n", ch, caseless);
9696 if ((re->flags & PCRE_REQCHSET) != 0)
9698 pcre_uchar ch = re->req_char;
9699 const char *caseless =
9700 ((re->flags & PCRE_RCH_CASELESS) == 0)? "" : " (caseless)";
9701 if (PRINTABLE(ch)) printf("Req char = %c%s\n", ch, caseless);
9702 else printf("Req char = \\x%02x%s\n", ch, caseless);
9705 #if defined COMPILE_PCRE8
9706 pcre_printint((pcre *)re, stdout, TRUE);
9707 #elif defined COMPILE_PCRE16
9708 pcre16_printint((pcre *)re, stdout, TRUE);
9709 #elif defined COMPILE_PCRE32
9710 pcre32_printint((pcre *)re, stdout, TRUE);
9713 /* This check is done here in the debugging case so that the code that
9714 was compiled can be seen. */
9716 if (code - codestart > length)
9719 *errorptr = find_error_text(ERR23);
9720 *erroroffset = ptr - (pcre_uchar *)pattern;
9721 if (errorcodeptr != NULL) *errorcodeptr = ERR23;
9724 #endif /* PCRE_DEBUG */
9726 /* Check for a pattern than can match an empty string, so that this information
9727 can be provided to applications. */
9731 if (could_be_empty_branch(codestart, code, utf, cd, NULL))
9733 re->flags |= PCRE_MATCH_EMPTY;
9736 codestart += GET(codestart, 1);
9738 while (*codestart == OP_ALT);
9740 #if defined COMPILE_PCRE8
9742 #elif defined COMPILE_PCRE16
9743 return (pcre16 *)re;
9744 #elif defined COMPILE_PCRE32
9745 return (pcre32 *)re;
9749 /* End of pcre_compile.c */