1 /*************************************************
2 * Perl-Compatible Regular Expressions *
3 *************************************************/
4
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.
7
8 Written by Philip Hazel
9 Copyright (c) 1997-2020 University of Cambridge
10
11 -----------------------------------------------------------------------------
12 Redistribution and use in source and binary forms, with or without
13 modification, are permitted provided that the following conditions are met:
14
15 * Redistributions of source code must retain the above copyright notice,
16 this list of conditions and the following disclaimer.
17
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.
21
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.
25
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 -----------------------------------------------------------------------------
38 */
39
40
41 /* This module contains the external function pcre_compile(), along with
42 supporting internal functions that are not used by other modules. */
43
44
45 #ifdef HAVE_CONFIG_H
46 #include "config.h"
47 #endif
48
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 */
52
53 #include "pcre_internal.h"
54
55
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. */
60
61 #ifdef PCRE_DEBUG
62 /* pcre_printint.c should not include any headers */
63 #define PCRE_INCLUDED
64 #include "pcre_printint.c"
65 #undef PCRE_INCLUDED
66 #endif
67
68
69 /* Macro for setting individual bits in class bitmaps. */
70
71 #define SETBIT(a,b) a[(b)/8] |= (1U << ((b)&7))
72
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. */
77
78 #define OFLOW_MAX (INT_MAX - 20)
79
80 /* Definitions to allow mutual recursion */
81
82 static int
83 add_list_to_class(pcre_uint8 *, pcre_uchar **, int, compile_data *,
84 const pcre_uint32 *, unsigned int);
85
86 static BOOL
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 *);
90
91
92
93 /*************************************************
94 * Code parameters and static tables *
95 *************************************************/
96
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.
103
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. */
114
115 #define COMPILE_WORK_SIZE (2048*LINK_SIZE)
116 #define COMPILE_WORK_SIZE_MAX (100*COMPILE_WORK_SIZE)
117
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. */
122
123 #define NAMED_GROUP_LIST_SIZE 20
124
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. */
127
128 #define WORK_SIZE_SAFETY_MARGIN (100)
129
130 /* Private flags added to firstchar and reqchar. */
131
132 #define REQ_CASELESS (1U << 0) /* Indicates caselessness */
133 #define REQ_VARY (1U << 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)
137
138 /* Repeated character flags. */
139
140 #define UTF_LENGTH 0x10000000l /* The char contains its length. */
141
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
145 is invalid. */
146
147 #ifndef EBCDIC
148
149 /* This is the "normal" table for ASCII systems or for EBCDIC systems running
150 in UTF-8 mode. */
151
152 static const short int escapes[] = {
153 0, 0,
154 0, 0,
155 0, 0,
156 0, 0,
157 0, 0,
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,
162 -ESC_B, -ESC_C,
163 -ESC_D, -ESC_E,
164 0, -ESC_G,
165 -ESC_H, 0,
166 0, -ESC_K,
167 0, 0,
168 -ESC_N, 0,
169 -ESC_P, -ESC_Q,
170 -ESC_R, -ESC_S,
171 0, 0,
172 -ESC_V, -ESC_W,
173 -ESC_X, 0,
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,
178 -ESC_b, 0,
179 -ESC_d, ESC_e,
180 ESC_f, 0,
181 -ESC_h, 0,
182 0, -ESC_k,
183 0, 0,
184 ESC_n, 0,
185 -ESC_p, 0,
186 ESC_r, -ESC_s,
187 ESC_tee, 0,
188 -ESC_v, -ESC_w,
189 0, 0,
190 -ESC_z
191 };
192
193 #else
194
195 /* This is the "abnormal" table for EBCDIC systems without UTF-8 support. */
196
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
221 };
222
223 /* We also need a table of characters that may follow \c in an EBCDIC
224 environment for characters 0-31. */
225
226 static unsigned char ebcdic_escape_c[] = "@ABCDEFGHIJKLMNOPQRSTUVWXYZ[\\]^_";
227
228 #endif
229
230
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
235 platforms. */
236
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 */
241 } verbitem;
242
243 static const char verbnames[] =
244 "\0" /* Empty name is a shorthand for MARK */
245 STRING_MARK0
246 STRING_ACCEPT0
247 STRING_COMMIT0
248 STRING_F0
249 STRING_FAIL0
250 STRING_PRUNE0
251 STRING_SKIP0
252 STRING_THEN;
253
254 static const verbitem verbs[] = {
255 { 0, -1, OP_MARK },
256 { 4, -1, OP_MARK },
257 { 6, OP_ACCEPT, -1 },
258 { 6, OP_COMMIT, -1 },
259 { 1, OP_FAIL, -1 },
260 { 4, OP_FAIL, -1 },
261 { 5, OP_PRUNE, OP_PRUNE_ARG },
262 { 4, OP_SKIP, OP_SKIP_ARG },
263 { 4, OP_THEN, OP_THEN_ARG }
264 };
265
266 static const int verbcount = sizeof(verbs)/sizeof(verbitem);
267
268
269 /* Substitutes for [[:<:]] and [[:>:]], which mean start and end of word in
270 another regex library. */
271
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' };
275
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' };
280
281
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. */
288
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;
294
295 static const pcre_uint8 posix_name_lengths[] = {
296 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 4, 6, 0 };
297
298 #define PC_GRAPH 8
299 #define PC_PRINT 9
300 #define PC_PUNCT 10
301
302
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. */
312
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 */
328 };
329
330 /* Table of substitutes for \d etc when PCRE_UCP is set. They are replaced by
331 Unicode property escapes. */
332
333 #ifdef SUPPORT_UCP
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' };
352
353 static const pcre_uchar *substitutes[] = {
354 string_PNd, /* \D */
355 string_pNd, /* \d */
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 */
359 string_pXwd /* \w */
360 };
361
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
366 directly. */
367
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' };
402
403 static const pcre_uchar *posix_substitutes[] = {
404 string_pL, /* alpha */
405 string_pLl, /* lower */
406 string_pLu, /* upper */
407 string_pXan, /* alnum */
408 NULL, /* ascii */
409 string_h, /* blank */
410 NULL, /* cntrl */
411 string_pNd, /* digit */
412 NULL, /* graph */
413 NULL, /* print */
414 NULL, /* punct */
415 string_pXps, /* space */ /* Xps is POSIX space, but from 8.34 */
416 string_pXwd, /* word */ /* Perl and POSIX space are the same */
417 NULL, /* xdigit */
418 /* Negated cases */
419 string_PL, /* ^alpha */
420 string_PLl, /* ^lower */
421 string_PLu, /* ^upper */
422 string_PXan, /* ^alnum */
423 NULL, /* ^ascii */
424 string_H, /* ^blank */
425 NULL, /* ^cntrl */
426 string_PNd, /* ^digit */
427 NULL, /* ^graph */
428 NULL, /* ^print */
429 NULL, /* ^punct */
430 string_PXps, /* ^space */ /* Xps is POSIX space, but from 8.34 */
431 string_PXwd, /* ^word */ /* Perl and POSIX space are the same */
432 NULL /* ^xdigit */
433 };
434 #define POSIX_SUBSIZE (sizeof(posix_substitutes) / sizeof(pcre_uchar *))
435 #endif
436
437 #define STRING(a) # a
438 #define XSTRING(s) STRING(s)
439
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.
449
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
452 counting through. */
453
454 static const char error_texts[] =
455 "no error\0"
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"
460 /* 5 */
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"
466 /* 10 */
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"
471 "missing )\0"
472 /* 15 */
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 **/
478 /* 20 */
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"
484 /* 25 */
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"
490 /* 30 */
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"
496 /* 35 */
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"
502 /* 40 */
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"
508 /* 45 */
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"
514 /* 50 */
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"
520 /* 55 */
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"
526 /* 60 */
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"
532 /* 65 */
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"
536 #ifndef EBCDIC
537 "\\c must be followed by an ASCII character\0"
538 #else
539 "\\c must be followed by a letter or one of [\\]^_?\0"
540 #endif
541 "\\k is not followed by a braced, angle-bracketed, or quoted name\0"
542 /* 70 */
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"
548 /* 75 */
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"
554 /* 80 */
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"
560 /* 85 */
561 "parentheses are too deeply nested (stack check)\0"
562 "digits missing in \\x{} or \\o{}\0"
563 "regular expression is too complicated\0"
564 ;
565
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
573 efficiently.
574
575 For convenience, we use the same bit definitions as in chartables:
576
577 0x04 decimal digit
578 0x08 hexadecimal digit
579
580 Then we can use ctype_digit and ctype_xdigit in the code. */
581
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). */
585
586 #define IS_DIGIT(x) ((x) >= CHAR_0 && (x) <= CHAR_9)
587
588 #ifndef EBCDIC
589
590 /* This is the "normal" case, for ASCII systems, and EBCDIC systems running in
591 UTF-8 mode. */
592
593 static const pcre_uint8 digitab[] =
594 {
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 */
627
628 #else
629
630 /* This is the "abnormal" case, for EBCDIC systems not running in UTF-8 mode. */
631
632 static const pcre_uint8 digitab[] =
633 {
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 */
666
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 */
700 #endif
701
702
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.
708
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. */
715
716 #define APTROWS (LAST_AUTOTAB_LEFT_OP - FIRST_AUTOTAB_OP + 1)
717 #define APTCOLS (LAST_AUTOTAB_RIGHT_OP - FIRST_AUTOTAB_OP + 1)
718
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 */
738 };
739
740
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:
745
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)
750
751 4 Check left general category vs right particular category
752 5 Check right general category vs left particular category
753
754 6 Left alphanum vs right general category
755 7 Left space vs right general category
756 8 Left word vs right general category
757
758 9 Right alphanum vs left general category
759 10 Right space vs left general category
760 11 Right word vs left general category
761
762 12 Left alphanum vs right particular category
763 13 Left space vs right particular category
764 14 Left word vs right particular category
765
766 15 Right alphanum vs left particular category
767 16 Right space vs left particular category
768 17 Right word vs left particular category
769 */
770
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 */
784 };
785
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
791 category. */
792
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 */
802 };
803
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
809 this table is used.
810
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
813 same here. */
814
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 */
819 };
820
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. */
826
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 */
830
831 0, /* NOTI */
832 OP_POSSTAR, 0, /* STAR, MINSTAR */
833 OP_POSPLUS, 0, /* PLUS, MINPLUS */
834 OP_POSQUERY, 0, /* QUERY, MINQUERY */
835 OP_POSUPTO, 0, /* UPTO, MINUPTO */
836 0, /* EXACT */
837 0, 0, 0, 0, /* POS{STAR,PLUS,QUERY,UPTO} */
838
839 OP_POSSTARI, 0, /* STARI, MINSTARI */
840 OP_POSPLUSI, 0, /* PLUSI, MINPLUSI */
841 OP_POSQUERYI, 0, /* QUERYI, MINQUERYI */
842 OP_POSUPTOI, 0, /* UPTOI, MINUPTOI */
843 0, /* EXACTI */
844 0, 0, 0, 0, /* POS{STARI,PLUSI,QUERYI,UPTOI} */
845
846 OP_NOTPOSSTAR, 0, /* NOTSTAR, NOTMINSTAR */
847 OP_NOTPOSPLUS, 0, /* NOTPLUS, NOTMINPLUS */
848 OP_NOTPOSQUERY, 0, /* NOTQUERY, NOTMINQUERY */
849 OP_NOTPOSUPTO, 0, /* NOTUPTO, NOTMINUPTO */
850 0, /* NOTEXACT */
851 0, 0, 0, 0, /* NOTPOS{STAR,PLUS,QUERY,UPTO} */
852
853 OP_NOTPOSSTARI, 0, /* NOTSTARI, NOTMINSTARI */
854 OP_NOTPOSPLUSI, 0, /* NOTPLUSI, NOTMINPLUSI */
855 OP_NOTPOSQUERYI, 0, /* NOTQUERYI, NOTMINQUERYI */
856 OP_NOTPOSUPTOI, 0, /* NOTUPTOI, NOTMINUPTOI */
857 0, /* NOTEXACTI */
858 0, 0, 0, 0, /* NOTPOS{STARI,PLUSI,QUERYI,UPTOI} */
859
860 OP_TYPEPOSSTAR, 0, /* TYPESTAR, TYPEMINSTAR */
861 OP_TYPEPOSPLUS, 0, /* TYPEPLUS, TYPEMINPLUS */
862 OP_TYPEPOSQUERY, 0, /* TYPEQUERY, TYPEMINQUERY */
863 OP_TYPEPOSUPTO, 0, /* TYPEUPTO, TYPEMINUPTO */
864 0, /* TYPEEXACT */
865 0, 0, 0, 0, /* TYPEPOS{STAR,PLUS,QUERY,UPTO} */
866
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} */
872
873 0, 0, 0, /* CLASS, NCLASS, XCLASS */
874 0, 0, /* REF, REFI */
875 0, 0, /* DNREF, DNREFI */
876 0, 0 /* RECURSE, CALLOUT */
877 };
878
879
880
881 /*************************************************
882 * Find an error text *
883 *************************************************/
884
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.
889
890 Argument: the error number
891 Returns: pointer to the error string
892 */
893
894 static const char *
find_error_text(int n)895 find_error_text(int n)
896 {
897 const char *s = error_texts;
898 for (; n > 0; n--)
899 {
900 while (*s++ != CHAR_NULL) {};
901 if (*s == CHAR_NULL) return "Error text not found (please report)";
902 }
903 return s;
904 }
905
906
907
908 /*************************************************
909 * Expand the workspace *
910 *************************************************/
911
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.
916
917 Argument: pointer to the compile data block
918 Returns: 0 if all went well, else an error number
919 */
920
921 static int
expand_workspace(compile_data * cd)922 expand_workspace(compile_data *cd)
923 {
924 pcre_uchar *newspace;
925 int newsize = cd->workspace_size * 2;
926
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)
930 return ERR72;
931
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;
940 return 0;
941 }
942
943
944
945 /*************************************************
946 * Check for counted repeat *
947 *************************************************/
948
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.
953
954 Arguments:
955 p pointer to the first char after '{'
956
957 Returns: TRUE or FALSE
958 */
959
960 static BOOL
is_counted_repeat(const pcre_uchar * p)961 is_counted_repeat(const pcre_uchar *p)
962 {
963 if (!IS_DIGIT(*p)) return FALSE;
964 p++;
965 while (IS_DIGIT(*p)) p++;
966 if (*p == CHAR_RIGHT_CURLY_BRACKET) return TRUE;
967
968 if (*p++ != CHAR_COMMA) return FALSE;
969 if (*p == CHAR_RIGHT_CURLY_BRACKET) return TRUE;
970
971 if (!IS_DIGIT(*p)) return FALSE;
972 p++;
973 while (IS_DIGIT(*p)) p++;
974
975 return (*p == CHAR_RIGHT_CURLY_BRACKET);
976 }
977
978
979
980 /*************************************************
981 * Handle escapes *
982 *************************************************/
983
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.
990
991 Arguments:
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
998
999 Returns: zero => a data character
1000 positive => a special escape sequence
1001 negative => a back reference
1002 on error, errorcodeptr is set
1003 */
1004
1005 static int
check_escape(const pcre_uchar ** ptrptr,pcre_uint32 * chptr,int * errorcodeptr,int bracount,int options,BOOL isclass)1006 check_escape(const pcre_uchar **ptrptr, pcre_uint32 *chptr, int *errorcodeptr,
1007 int bracount, int options, BOOL isclass)
1008 {
1009 /* PCRE_UTF16 has the same value as PCRE_UTF8. */
1010 BOOL utf = (options & PCRE_UTF8) != 0;
1011 const pcre_uchar *ptr = *ptrptr + 1;
1012 pcre_uint32 c;
1013 int escape = 0;
1014 int i;
1015
1016 GETCHARINCTEST(c, ptr); /* Get character value, increment pointer */
1017 ptr--; /* Set pointer back to the last byte */
1018
1019 /* If backslash is at the end of the pattern, it's an error. */
1020
1021 if (c == CHAR_NULL) *errorcodeptr = ERR1;
1022
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. */
1026
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; }
1032
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; }
1037 #endif
1038
1039 /* Escapes that need further processing, or are illegal. */
1040
1041 else
1042 {
1043 const pcre_uchar *oldptr;
1044 BOOL braced, negated, overflow;
1045 int s;
1046
1047 switch (c)
1048 {
1049 /* A number of Perl escapes are not handled by PCRE. We give an explicit
1050 error. */
1051
1052 case CHAR_l:
1053 case CHAR_L:
1054 *errorcodeptr = ERR37;
1055 break;
1056
1057 case CHAR_u:
1058 if ((options & PCRE_JAVASCRIPT_COMPAT) != 0)
1059 {
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)
1066 {
1067 c = 0;
1068 for (i = 0; i < 4; ++i)
1069 {
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));
1077 #endif
1078 }
1079
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)
1086 #endif
1087 {
1088 *errorcodeptr = ERR76;
1089 }
1090 else if (utf && c >= 0xd800 && c <= 0xdfff) *errorcodeptr = ERR73;
1091 }
1092 }
1093 else
1094 *errorcodeptr = ERR37;
1095 break;
1096
1097 case CHAR_U:
1098 /* In JavaScript, \U is an uppercase U letter. */
1099 if ((options & PCRE_JAVASCRIPT_COMPAT) == 0) *errorcodeptr = ERR37;
1100 break;
1101
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:
1104
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
1107 5.10 feature.
1108
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
1112 was \k.
1113
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). */
1118
1119 case CHAR_g:
1120 if (isclass) break;
1121 if (ptr[1] == CHAR_LESS_THAN_SIGN || ptr[1] == CHAR_APOSTROPHE)
1122 {
1123 escape = ESC_g;
1124 break;
1125 }
1126
1127 /* Handle the Perl-compatible cases */
1128
1129 if (ptr[1] == CHAR_LEFT_CURLY_BRACKET)
1130 {
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)
1135 {
1136 escape = ESC_k;
1137 break;
1138 }
1139 braced = TRUE;
1140 ptr++;
1141 }
1142 else braced = FALSE;
1143
1144 if (ptr[1] == CHAR_MINUS)
1145 {
1146 negated = TRUE;
1147 ptr++;
1148 }
1149 else negated = FALSE;
1150
1151 /* The integer range is limited by the machine's int representation. */
1152 s = 0;
1153 overflow = FALSE;
1154 while (IS_DIGIT(ptr[1]))
1155 {
1156 if (s > INT_MAX / 10 - 1) /* Integer overflow */
1157 {
1158 overflow = TRUE;
1159 break;
1160 }
1161 s = s * 10 + (int)(*(++ptr) - CHAR_0);
1162 }
1163 if (overflow) /* Integer overflow */
1164 {
1165 while (IS_DIGIT(ptr[1]))
1166 ptr++;
1167 *errorcodeptr = ERR61;
1168 break;
1169 }
1170
1171 if (braced && *(++ptr) != CHAR_RIGHT_CURLY_BRACKET)
1172 {
1173 *errorcodeptr = ERR57;
1174 break;
1175 }
1176
1177 if (s == 0)
1178 {
1179 *errorcodeptr = ERR58;
1180 break;
1181 }
1182
1183 if (negated)
1184 {
1185 if (s > bracount)
1186 {
1187 *errorcodeptr = ERR15;
1188 break;
1189 }
1190 s = bracount - (s - 1);
1191 }
1192
1193 escape = -s;
1194 break;
1195
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.
1200
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.
1208
1209 Inside a character class, \ followed by a digit is always either a literal
1210 8 or 9 or an octal number. */
1211
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:
1214
1215 if (!isclass)
1216 {
1217 oldptr = ptr;
1218 /* The integer range is limited by the machine's int representation. */
1219 s = (int)(c -CHAR_0);
1220 overflow = FALSE;
1221 while (IS_DIGIT(ptr[1]))
1222 {
1223 if (s > INT_MAX / 10 - 1) /* Integer overflow */
1224 {
1225 overflow = TRUE;
1226 break;
1227 }
1228 s = s * 10 + (int)(*(++ptr) - CHAR_0);
1229 }
1230 if (overflow) /* Integer overflow */
1231 {
1232 while (IS_DIGIT(ptr[1]))
1233 ptr++;
1234 *errorcodeptr = ERR61;
1235 break;
1236 }
1237 if (s < 8 || s <= bracount) /* Check for back reference */
1238 {
1239 escape = -s;
1240 break;
1241 }
1242 ptr = oldptr; /* Put the pointer back and fall through */
1243 }
1244
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. */
1249
1250 if ((c = *ptr) >= CHAR_8) break;
1251
1252 /* Fall through with a digit less than 8 */
1253
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. */
1259
1260 case CHAR_0:
1261 c -= CHAR_0;
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;
1266 #endif
1267 break;
1268
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}. */
1271
1272 case CHAR_o:
1273 if (ptr[1] != CHAR_LEFT_CURLY_BRACKET) *errorcodeptr = ERR81; else
1274 if (ptr[2] == CHAR_RIGHT_CURLY_BRACKET) *errorcodeptr = ERR86; else
1275 {
1276 ptr += 2;
1277 c = 0;
1278 overflow = FALSE;
1279 while (*ptr >= CHAR_0 && *ptr <= CHAR_7)
1280 {
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; }
1285 #endif
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; }
1293 #endif
1294 }
1295 if (overflow)
1296 {
1297 while (*ptr >= CHAR_0 && *ptr <= CHAR_7) ptr++;
1298 *errorcodeptr = ERR34;
1299 }
1300 else if (*ptr == CHAR_RIGHT_CURLY_BRACKET)
1301 {
1302 if (utf && c >= 0xd800 && c <= 0xdfff) *errorcodeptr = ERR73;
1303 }
1304 else *errorcodeptr = ERR80;
1305 }
1306 break;
1307
1308 /* \x is complicated. In JavaScript, \x must be followed by two hexadecimal
1309 numbers. Otherwise it is a lowercase x letter. */
1310
1311 case CHAR_x:
1312 if ((options & PCRE_JAVASCRIPT_COMPAT) != 0)
1313 {
1314 if (MAX_255(ptr[1]) && (digitab[ptr[1]] & ctype_xdigit) != 0
1315 && MAX_255(ptr[2]) && (digitab[ptr[2]] & ctype_xdigit) != 0)
1316 {
1317 c = 0;
1318 for (i = 0; i < 2; ++i)
1319 {
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));
1327 #endif
1328 }
1329 }
1330 } /* End JavaScript handling */
1331
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. */
1338
1339 else
1340 {
1341 if (ptr[1] == CHAR_LEFT_CURLY_BRACKET)
1342 {
1343 ptr += 2;
1344 if (*ptr == CHAR_RIGHT_CURLY_BRACKET)
1345 {
1346 *errorcodeptr = ERR86;
1347 break;
1348 }
1349 c = 0;
1350 overflow = FALSE;
1351 while (MAX_255(*ptr) && (digitab[*ptr] & ctype_xdigit) != 0)
1352 {
1353 register pcre_uint32 cc = *ptr++;
1354 if (c == 0 && cc == CHAR_0) continue; /* Leading zeroes */
1355
1356 #ifdef COMPILE_PCRE32
1357 if (c >= 0x10000000l) { overflow = TRUE; break; }
1358 #endif
1359
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));
1366 #endif
1367
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; }
1374 #endif
1375 }
1376
1377 if (overflow)
1378 {
1379 while (MAX_255(*ptr) && (digitab[*ptr] & ctype_xdigit) != 0) ptr++;
1380 *errorcodeptr = ERR34;
1381 }
1382
1383 else if (*ptr == CHAR_RIGHT_CURLY_BRACKET)
1384 {
1385 if (utf && c >= 0xd800 && c <= 0xdfff) *errorcodeptr = ERR73;
1386 }
1387
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. */
1392
1393 else *errorcodeptr = ERR79;
1394 } /* End of \x{} processing */
1395
1396 /* Read a single-byte hex-defined char (up to two hex digits after \x) */
1397
1398 else
1399 {
1400 c = 0;
1401 while (i++ < 2 && MAX_255(ptr[1]) && (digitab[ptr[1]] & ctype_xdigit) != 0)
1402 {
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));
1411 #endif
1412 }
1413 } /* End of \xdd handling */
1414 } /* End of Perl-style \x handling */
1415 break;
1416
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.) */
1421
1422 case CHAR_c:
1423 c = *(++ptr);
1424 if (c == CHAR_NULL)
1425 {
1426 *errorcodeptr = ERR2;
1427 break;
1428 }
1429 #ifndef EBCDIC /* ASCII/UTF-8 coding */
1430 if (c > 127) /* Excludes all non-ASCII in either mode */
1431 {
1432 *errorcodeptr = ERR68;
1433 break;
1434 }
1435 if (c >= CHAR_a && c <= CHAR_z) c -= 32;
1436 c ^= 0x40;
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;
1441 else
1442 {
1443 for (i = 0; i < 32; i++)
1444 {
1445 if (c == ebcdic_escape_c[i]) break;
1446 }
1447 if (i < 32) c = i; else *errorcodeptr = ERR68;
1448 }
1449 #endif
1450 break;
1451
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. */
1457
1458 default:
1459 if ((options & PCRE_EXTRA) != 0) switch(c)
1460 {
1461 default:
1462 *errorcodeptr = ERR3;
1463 break;
1464 }
1465 break;
1466 }
1467 }
1468
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}. */
1472
1473 if (escape == ESC_N && ptr[1] == CHAR_LEFT_CURLY_BRACKET &&
1474 !is_counted_repeat(ptr+2))
1475 *errorcodeptr = ERR37;
1476
1477 /* If PCRE_UCP is set, we change the values for \d etc. */
1478
1479 if ((options & PCRE_UCP) != 0 && escape >= ESC_D && escape <= ESC_w)
1480 escape += (ESC_DU - ESC_D);
1481
1482 /* Set the pointer to the final character before returning. */
1483
1484 *ptrptr = ptr;
1485 *chptr = c;
1486 return escape;
1487 }
1488
1489
1490
1491 #ifdef SUPPORT_UCP
1492 /*************************************************
1493 * Handle \P and \p *
1494 *************************************************/
1495
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
1499 escape sequence.
1500
1501 Argument:
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
1507
1508 Returns: TRUE if the type value was found, or FALSE for an invalid type
1509 */
1510
1511 static BOOL
get_ucp(const pcre_uchar ** ptrptr,BOOL * negptr,unsigned int * ptypeptr,unsigned int * pdataptr,int * errorcodeptr)1512 get_ucp(const pcre_uchar **ptrptr, BOOL *negptr, unsigned int *ptypeptr,
1513 unsigned int *pdataptr, int *errorcodeptr)
1514 {
1515 pcre_uchar c;
1516 int i, bot, top;
1517 const pcre_uchar *ptr = *ptrptr;
1518 pcre_uchar name[32];
1519
1520 c = *(++ptr);
1521 if (c == CHAR_NULL) goto ERROR_RETURN;
1522
1523 *negptr = FALSE;
1524
1525 /* \P or \p can be followed by a name in {}, optionally preceded by ^ for
1526 negation. */
1527
1528 if (c == CHAR_LEFT_CURLY_BRACKET)
1529 {
1530 if (ptr[1] == CHAR_CIRCUMFLEX_ACCENT)
1531 {
1532 *negptr = TRUE;
1533 ptr++;
1534 }
1535 for (i = 0; i < (int)(sizeof(name) / sizeof(pcre_uchar)) - 1; i++)
1536 {
1537 c = *(++ptr);
1538 if (c == CHAR_NULL) goto ERROR_RETURN;
1539 if (c == CHAR_RIGHT_CURLY_BRACKET) break;
1540 name[i] = c;
1541 }
1542 if (c != CHAR_RIGHT_CURLY_BRACKET) goto ERROR_RETURN;
1543 name[i] = 0;
1544 }
1545
1546 /* Otherwise there is just one following character */
1547
1548 else
1549 {
1550 name[0] = c;
1551 name[1] = 0;
1552 }
1553
1554 *ptrptr = ptr;
1555
1556 /* Search for a recognized property name using binary chop */
1557
1558 bot = 0;
1559 top = PRIV(utt_size);
1560
1561 while (bot < top)
1562 {
1563 int r;
1564 i = (bot + top) >> 1;
1565 r = STRCMP_UC_C8(name, PRIV(utt_names) + PRIV(utt)[i].name_offset);
1566 if (r == 0)
1567 {
1568 *ptypeptr = PRIV(utt)[i].type;
1569 *pdataptr = PRIV(utt)[i].value;
1570 return TRUE;
1571 }
1572 if (r > 0) bot = i + 1; else top = i;
1573 }
1574
1575 *errorcodeptr = ERR47;
1576 *ptrptr = ptr;
1577 return FALSE;
1578
1579 ERROR_RETURN:
1580 *errorcodeptr = ERR46;
1581 *ptrptr = ptr;
1582 return FALSE;
1583 }
1584 #endif
1585
1586
1587
1588 /*************************************************
1589 * Read repeat counts *
1590 *************************************************/
1591
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.
1595
1596 Arguments:
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
1602
1603 Returns: pointer to '}' on success;
1604 current ptr on error, with errorcodeptr set non-zero
1605 */
1606
1607 static const pcre_uchar *
read_repeat_counts(const pcre_uchar * p,int * minp,int * maxp,int * errorcodeptr)1608 read_repeat_counts(const pcre_uchar *p, int *minp, int *maxp, int *errorcodeptr)
1609 {
1610 int min = 0;
1611 int max = -1;
1612
1613 while (IS_DIGIT(*p))
1614 {
1615 min = min * 10 + (int)(*p++ - CHAR_0);
1616 if (min > 65535)
1617 {
1618 *errorcodeptr = ERR5;
1619 return p;
1620 }
1621 }
1622
1623 if (*p == CHAR_RIGHT_CURLY_BRACKET) max = min; else
1624 {
1625 if (*(++p) != CHAR_RIGHT_CURLY_BRACKET)
1626 {
1627 max = 0;
1628 while(IS_DIGIT(*p))
1629 {
1630 max = max * 10 + (int)(*p++ - CHAR_0);
1631 if (max > 65535)
1632 {
1633 *errorcodeptr = ERR5;
1634 return p;
1635 }
1636 }
1637 if (max < min)
1638 {
1639 *errorcodeptr = ERR4;
1640 return p;
1641 }
1642 }
1643 }
1644
1645 *minp = min;
1646 *maxp = max;
1647 return p;
1648 }
1649
1650
1651
1652 /*************************************************
1653 * Find first significant op code *
1654 *************************************************/
1655
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
1660 does not.
1661
1662 Arguments:
1663 code pointer to the start of the group
1664 skipassert TRUE if certain assertions are to be skipped
1665
1666 Returns: pointer to the first significant opcode
1667 */
1668
1669 static const pcre_uchar*
first_significant_code(const pcre_uchar * code,BOOL skipassert)1670 first_significant_code(const pcre_uchar *code, BOOL skipassert)
1671 {
1672 for (;;)
1673 {
1674 switch ((int)*code)
1675 {
1676 case OP_ASSERT_NOT:
1677 case OP_ASSERTBACK:
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];
1682 break;
1683
1684 case OP_WORD_BOUNDARY:
1685 case OP_NOT_WORD_BOUNDARY:
1686 if (!skipassert) return code;
1687 /* Fall through */
1688
1689 case OP_CALLOUT:
1690 case OP_CREF:
1691 case OP_DNCREF:
1692 case OP_RREF:
1693 case OP_DNRREF:
1694 case OP_DEF:
1695 code += PRIV(OP_lengths)[*code];
1696 break;
1697
1698 default:
1699 return code;
1700 }
1701 }
1702 /* Control never reaches here */
1703 }
1704
1705
1706
1707 /*************************************************
1708 * Find the fixed length of a branch *
1709 *************************************************/
1710
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.
1715
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.
1721
1722 Arguments:
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
1728
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)
1734 */
1735
1736 static int
find_fixedlength(pcre_uchar * code,BOOL utf,BOOL atend,compile_data * cd,recurse_check * recurses)1737 find_fixedlength(pcre_uchar *code, BOOL utf, BOOL atend, compile_data *cd,
1738 recurse_check *recurses)
1739 {
1740 int length = -1;
1741 recurse_check this_recurse;
1742 register int branchlength = 0;
1743 register pcre_uchar *cc = code + 1 + LINK_SIZE;
1744
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. */
1747
1748 for (;;)
1749 {
1750 int d;
1751 pcre_uchar *ce, *cs;
1752 register pcre_uchar op = *cc;
1753
1754 switch (op)
1755 {
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. */
1760
1761 case OP_CBRA:
1762 case OP_BRA:
1763 case OP_ONCE:
1764 case OP_ONCE_NC:
1765 case OP_COND:
1766 d = find_fixedlength(cc + ((op == OP_CBRA)? IMM2_SIZE : 0), utf, atend, cd,
1767 recurses);
1768 if (d < 0) return d;
1769 branchlength += d;
1770 do cc += GET(cc, 1); while (*cc == OP_ALT);
1771 cc += 1 + LINK_SIZE;
1772 break;
1773
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. */
1779
1780 case OP_ALT:
1781 case OP_KET:
1782 case OP_END:
1783 case OP_ACCEPT:
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;
1789 branchlength = 0;
1790 break;
1791
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. */
1795
1796 case OP_RECURSE:
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 */
1802 {
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 */
1806 }
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;
1811 branchlength += d;
1812 cc += 1 + LINK_SIZE;
1813 break;
1814
1815 /* Skip over assertive subpatterns */
1816
1817 case OP_ASSERT:
1818 case OP_ASSERT_NOT:
1819 case OP_ASSERTBACK:
1820 case OP_ASSERTBACK_NOT:
1821 do cc += GET(cc, 1); while (*cc == OP_ALT);
1822 cc += 1 + LINK_SIZE;
1823 break;
1824
1825 /* Skip over things that don't match chars */
1826
1827 case OP_MARK:
1828 case OP_PRUNE_ARG:
1829 case OP_SKIP_ARG:
1830 case OP_THEN_ARG:
1831 cc += cc[1] + PRIV(OP_lengths)[*cc];
1832 break;
1833
1834 case OP_CALLOUT:
1835 case OP_CIRC:
1836 case OP_CIRCM:
1837 case OP_CLOSE:
1838 case OP_COMMIT:
1839 case OP_CREF:
1840 case OP_DEF:
1841 case OP_DNCREF:
1842 case OP_DNRREF:
1843 case OP_DOLL:
1844 case OP_DOLLM:
1845 case OP_EOD:
1846 case OP_EODN:
1847 case OP_FAIL:
1848 case OP_NOT_WORD_BOUNDARY:
1849 case OP_PRUNE:
1850 case OP_REVERSE:
1851 case OP_RREF:
1852 case OP_SET_SOM:
1853 case OP_SKIP:
1854 case OP_SOD:
1855 case OP_SOM:
1856 case OP_THEN:
1857 case OP_WORD_BOUNDARY:
1858 cc += PRIV(OP_lengths)[*cc];
1859 break;
1860
1861 /* Handle literal characters */
1862
1863 case OP_CHAR:
1864 case OP_CHARI:
1865 case OP_NOT:
1866 case OP_NOTI:
1867 branchlength++;
1868 cc += 2;
1869 #ifdef SUPPORT_UTF
1870 if (utf && HAS_EXTRALEN(cc[-1])) cc += GET_EXTRALEN(cc[-1]);
1871 #endif
1872 break;
1873
1874 /* Handle exact repetitions. The count is already in characters, but we
1875 need to skip over a multibyte character in UTF8 mode. */
1876
1877 case OP_EXACT:
1878 case OP_EXACTI:
1879 case OP_NOTEXACT:
1880 case OP_NOTEXACTI:
1881 branchlength += (int)GET2(cc,1);
1882 cc += 2 + IMM2_SIZE;
1883 #ifdef SUPPORT_UTF
1884 if (utf && HAS_EXTRALEN(cc[-1])) cc += GET_EXTRALEN(cc[-1]);
1885 #endif
1886 break;
1887
1888 case OP_TYPEEXACT:
1889 branchlength += GET2(cc,1);
1890 if (cc[1 + IMM2_SIZE] == OP_PROP || cc[1 + IMM2_SIZE] == OP_NOTPROP)
1891 cc += 2;
1892 cc += 1 + IMM2_SIZE + 1;
1893 break;
1894
1895 /* Handle single-char matchers */
1896
1897 case OP_PROP:
1898 case OP_NOTPROP:
1899 cc += 2;
1900 /* Fall through */
1901
1902 case OP_HSPACE:
1903 case OP_VSPACE:
1904 case OP_NOT_HSPACE:
1905 case OP_NOT_VSPACE:
1906 case OP_NOT_DIGIT:
1907 case OP_DIGIT:
1908 case OP_NOT_WHITESPACE:
1909 case OP_WHITESPACE:
1910 case OP_NOT_WORDCHAR:
1911 case OP_WORDCHAR:
1912 case OP_ANY:
1913 case OP_ALLANY:
1914 branchlength++;
1915 cc++;
1916 break;
1917
1918 /* The single-byte matcher isn't allowed. This only happens in UTF-8 mode;
1919 otherwise \C is coded as OP_ALLANY. */
1920
1921 case OP_ANYBYTE:
1922 return -2;
1923
1924 /* Check a class for variable quantification */
1925
1926 case OP_CLASS:
1927 case OP_NCLASS:
1928 #if defined SUPPORT_UTF || defined COMPILE_PCRE16 || defined COMPILE_PCRE32
1929 case OP_XCLASS:
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)
1933 cc += GET(cc, 1);
1934 else
1935 cc += PRIV(OP_lengths)[OP_CLASS];
1936 #else
1937 cc += PRIV(OP_lengths)[OP_CLASS];
1938 #endif
1939
1940 switch (*cc)
1941 {
1942 case OP_CRSTAR:
1943 case OP_CRMINSTAR:
1944 case OP_CRPLUS:
1945 case OP_CRMINPLUS:
1946 case OP_CRQUERY:
1947 case OP_CRMINQUERY:
1948 case OP_CRPOSSTAR:
1949 case OP_CRPOSPLUS:
1950 case OP_CRPOSQUERY:
1951 return -1;
1952
1953 case OP_CRRANGE:
1954 case OP_CRMINRANGE:
1955 case OP_CRPOSRANGE:
1956 if (GET2(cc,1) != GET2(cc,1+IMM2_SIZE)) return -1;
1957 branchlength += (int)GET2(cc,1);
1958 cc += 1 + 2 * IMM2_SIZE;
1959 break;
1960
1961 default:
1962 branchlength++;
1963 }
1964 break;
1965
1966 /* Anything else is variable length */
1967
1968 case OP_ANYNL:
1969 case OP_BRAMINZERO:
1970 case OP_BRAPOS:
1971 case OP_BRAPOSZERO:
1972 case OP_BRAZERO:
1973 case OP_CBRAPOS:
1974 case OP_EXTUNI:
1975 case OP_KETRMAX:
1976 case OP_KETRMIN:
1977 case OP_KETRPOS:
1978 case OP_MINPLUS:
1979 case OP_MINPLUSI:
1980 case OP_MINQUERY:
1981 case OP_MINQUERYI:
1982 case OP_MINSTAR:
1983 case OP_MINSTARI:
1984 case OP_MINUPTO:
1985 case OP_MINUPTOI:
1986 case OP_NOTMINPLUS:
1987 case OP_NOTMINPLUSI:
1988 case OP_NOTMINQUERY:
1989 case OP_NOTMINQUERYI:
1990 case OP_NOTMINSTAR:
1991 case OP_NOTMINSTARI:
1992 case OP_NOTMINUPTO:
1993 case OP_NOTMINUPTOI:
1994 case OP_NOTPLUS:
1995 case OP_NOTPLUSI:
1996 case OP_NOTPOSPLUS:
1997 case OP_NOTPOSPLUSI:
1998 case OP_NOTPOSQUERY:
1999 case OP_NOTPOSQUERYI:
2000 case OP_NOTPOSSTAR:
2001 case OP_NOTPOSSTARI:
2002 case OP_NOTPOSUPTO:
2003 case OP_NOTPOSUPTOI:
2004 case OP_NOTQUERY:
2005 case OP_NOTQUERYI:
2006 case OP_NOTSTAR:
2007 case OP_NOTSTARI:
2008 case OP_NOTUPTO:
2009 case OP_NOTUPTOI:
2010 case OP_PLUS:
2011 case OP_PLUSI:
2012 case OP_POSPLUS:
2013 case OP_POSPLUSI:
2014 case OP_POSQUERY:
2015 case OP_POSQUERYI:
2016 case OP_POSSTAR:
2017 case OP_POSSTARI:
2018 case OP_POSUPTO:
2019 case OP_POSUPTOI:
2020 case OP_QUERY:
2021 case OP_QUERYI:
2022 case OP_REF:
2023 case OP_REFI:
2024 case OP_DNREF:
2025 case OP_DNREFI:
2026 case OP_SBRA:
2027 case OP_SBRAPOS:
2028 case OP_SCBRA:
2029 case OP_SCBRAPOS:
2030 case OP_SCOND:
2031 case OP_SKIPZERO:
2032 case OP_STAR:
2033 case OP_STARI:
2034 case OP_TYPEMINPLUS:
2035 case OP_TYPEMINQUERY:
2036 case OP_TYPEMINSTAR:
2037 case OP_TYPEMINUPTO:
2038 case OP_TYPEPLUS:
2039 case OP_TYPEPOSPLUS:
2040 case OP_TYPEPOSQUERY:
2041 case OP_TYPEPOSSTAR:
2042 case OP_TYPEPOSUPTO:
2043 case OP_TYPEQUERY:
2044 case OP_TYPESTAR:
2045 case OP_TYPEUPTO:
2046 case OP_UPTO:
2047 case OP_UPTOI:
2048 return -1;
2049
2050 /* Catch unrecognized opcodes so that when new ones are added they
2051 are not forgotten, as has happened in the past. */
2052
2053 default:
2054 return -4;
2055 }
2056 }
2057 /* Control never gets here */
2058 }
2059
2060
2061
2062 /*************************************************
2063 * Scan compiled regex for specific bracket *
2064 *************************************************/
2065
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
2070 length.
2071
2072 Arguments:
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
2076
2077 Returns: pointer to the opcode for the bracket, or NULL if not found
2078 */
2079
2080 const pcre_uchar *
PRIV(find_bracket)2081 PRIV(find_bracket)(const pcre_uchar *code, BOOL utf, int number)
2082 {
2083 for (;;)
2084 {
2085 register pcre_uchar c = *code;
2086
2087 if (c == OP_END) return NULL;
2088
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. */
2092
2093 if (c == OP_XCLASS) code += GET(code, 1);
2094
2095 /* Handle recursion */
2096
2097 else if (c == OP_REVERSE)
2098 {
2099 if (number < 0) return (pcre_uchar *)code;
2100 code += PRIV(OP_lengths)[c];
2101 }
2102
2103 /* Handle capturing bracket */
2104
2105 else if (c == OP_CBRA || c == OP_SCBRA ||
2106 c == OP_CBRAPOS || c == OP_SCBRAPOS)
2107 {
2108 int n = (int)GET2(code, 1+LINK_SIZE);
2109 if (n == number) return (pcre_uchar *)code;
2110 code += PRIV(OP_lengths)[c];
2111 }
2112
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. */
2117
2118 else
2119 {
2120 switch(c)
2121 {
2122 case OP_TYPESTAR:
2123 case OP_TYPEMINSTAR:
2124 case OP_TYPEPLUS:
2125 case OP_TYPEMINPLUS:
2126 case OP_TYPEQUERY:
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;
2132 break;
2133
2134 case OP_TYPEUPTO:
2135 case OP_TYPEMINUPTO:
2136 case OP_TYPEEXACT:
2137 case OP_TYPEPOSUPTO:
2138 if (code[1 + IMM2_SIZE] == OP_PROP || code[1 + IMM2_SIZE] == OP_NOTPROP)
2139 code += 2;
2140 break;
2141
2142 case OP_MARK:
2143 case OP_PRUNE_ARG:
2144 case OP_SKIP_ARG:
2145 case OP_THEN_ARG:
2146 code += code[1];
2147 break;
2148 }
2149
2150 /* Add in the fixed length from the table */
2151
2152 code += PRIV(OP_lengths)[c];
2153
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. */
2157
2158 #if defined SUPPORT_UTF && !defined COMPILE_PCRE32
2159 if (utf) switch(c)
2160 {
2161 case OP_CHAR:
2162 case OP_CHARI:
2163 case OP_NOT:
2164 case OP_NOTI:
2165 case OP_EXACT:
2166 case OP_EXACTI:
2167 case OP_NOTEXACT:
2168 case OP_NOTEXACTI:
2169 case OP_UPTO:
2170 case OP_UPTOI:
2171 case OP_NOTUPTO:
2172 case OP_NOTUPTOI:
2173 case OP_MINUPTO:
2174 case OP_MINUPTOI:
2175 case OP_NOTMINUPTO:
2176 case OP_NOTMINUPTOI:
2177 case OP_POSUPTO:
2178 case OP_POSUPTOI:
2179 case OP_NOTPOSUPTO:
2180 case OP_NOTPOSUPTOI:
2181 case OP_STAR:
2182 case OP_STARI:
2183 case OP_NOTSTAR:
2184 case OP_NOTSTARI:
2185 case OP_MINSTAR:
2186 case OP_MINSTARI:
2187 case OP_NOTMINSTAR:
2188 case OP_NOTMINSTARI:
2189 case OP_POSSTAR:
2190 case OP_POSSTARI:
2191 case OP_NOTPOSSTAR:
2192 case OP_NOTPOSSTARI:
2193 case OP_PLUS:
2194 case OP_PLUSI:
2195 case OP_NOTPLUS:
2196 case OP_NOTPLUSI:
2197 case OP_MINPLUS:
2198 case OP_MINPLUSI:
2199 case OP_NOTMINPLUS:
2200 case OP_NOTMINPLUSI:
2201 case OP_POSPLUS:
2202 case OP_POSPLUSI:
2203 case OP_NOTPOSPLUS:
2204 case OP_NOTPOSPLUSI:
2205 case OP_QUERY:
2206 case OP_QUERYI:
2207 case OP_NOTQUERY:
2208 case OP_NOTQUERYI:
2209 case OP_MINQUERY:
2210 case OP_MINQUERYI:
2211 case OP_NOTMINQUERY:
2212 case OP_NOTMINQUERYI:
2213 case OP_POSQUERY:
2214 case OP_POSQUERYI:
2215 case OP_NOTPOSQUERY:
2216 case OP_NOTPOSQUERYI:
2217 if (HAS_EXTRALEN(code[-1])) code += GET_EXTRALEN(code[-1]);
2218 break;
2219 }
2220 #else
2221 (void)(utf); /* Keep compiler happy by referencing function argument */
2222 #endif
2223 }
2224 }
2225 }
2226
2227
2228
2229 /*************************************************
2230 * Scan compiled regex for recursion reference *
2231 *************************************************/
2232
2233 /* This little function scans through a compiled pattern until it finds an
2234 instance of OP_RECURSE.
2235
2236 Arguments:
2237 code points to start of expression
2238 utf TRUE in UTF-8 / UTF-16 / UTF-32 mode
2239
2240 Returns: pointer to the opcode for OP_RECURSE, or NULL if not found
2241 */
2242
2243 static const pcre_uchar *
find_recurse(const pcre_uchar * code,BOOL utf)2244 find_recurse(const pcre_uchar *code, BOOL utf)
2245 {
2246 for (;;)
2247 {
2248 register pcre_uchar c = *code;
2249 if (c == OP_END) return NULL;
2250 if (c == OP_RECURSE) return code;
2251
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. */
2255
2256 if (c == OP_XCLASS) code += GET(code, 1);
2257
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. */
2262
2263 else
2264 {
2265 switch(c)
2266 {
2267 case OP_TYPESTAR:
2268 case OP_TYPEMINSTAR:
2269 case OP_TYPEPLUS:
2270 case OP_TYPEMINPLUS:
2271 case OP_TYPEQUERY:
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;
2277 break;
2278
2279 case OP_TYPEPOSUPTO:
2280 case OP_TYPEUPTO:
2281 case OP_TYPEMINUPTO:
2282 case OP_TYPEEXACT:
2283 if (code[1 + IMM2_SIZE] == OP_PROP || code[1 + IMM2_SIZE] == OP_NOTPROP)
2284 code += 2;
2285 break;
2286
2287 case OP_MARK:
2288 case OP_PRUNE_ARG:
2289 case OP_SKIP_ARG:
2290 case OP_THEN_ARG:
2291 code += code[1];
2292 break;
2293 }
2294
2295 /* Add in the fixed length from the table */
2296
2297 code += PRIV(OP_lengths)[c];
2298
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. */
2302
2303 #if defined SUPPORT_UTF && !defined COMPILE_PCRE32
2304 if (utf) switch(c)
2305 {
2306 case OP_CHAR:
2307 case OP_CHARI:
2308 case OP_NOT:
2309 case OP_NOTI:
2310 case OP_EXACT:
2311 case OP_EXACTI:
2312 case OP_NOTEXACT:
2313 case OP_NOTEXACTI:
2314 case OP_UPTO:
2315 case OP_UPTOI:
2316 case OP_NOTUPTO:
2317 case OP_NOTUPTOI:
2318 case OP_MINUPTO:
2319 case OP_MINUPTOI:
2320 case OP_NOTMINUPTO:
2321 case OP_NOTMINUPTOI:
2322 case OP_POSUPTO:
2323 case OP_POSUPTOI:
2324 case OP_NOTPOSUPTO:
2325 case OP_NOTPOSUPTOI:
2326 case OP_STAR:
2327 case OP_STARI:
2328 case OP_NOTSTAR:
2329 case OP_NOTSTARI:
2330 case OP_MINSTAR:
2331 case OP_MINSTARI:
2332 case OP_NOTMINSTAR:
2333 case OP_NOTMINSTARI:
2334 case OP_POSSTAR:
2335 case OP_POSSTARI:
2336 case OP_NOTPOSSTAR:
2337 case OP_NOTPOSSTARI:
2338 case OP_PLUS:
2339 case OP_PLUSI:
2340 case OP_NOTPLUS:
2341 case OP_NOTPLUSI:
2342 case OP_MINPLUS:
2343 case OP_MINPLUSI:
2344 case OP_NOTMINPLUS:
2345 case OP_NOTMINPLUSI:
2346 case OP_POSPLUS:
2347 case OP_POSPLUSI:
2348 case OP_NOTPOSPLUS:
2349 case OP_NOTPOSPLUSI:
2350 case OP_QUERY:
2351 case OP_QUERYI:
2352 case OP_NOTQUERY:
2353 case OP_NOTQUERYI:
2354 case OP_MINQUERY:
2355 case OP_MINQUERYI:
2356 case OP_NOTMINQUERY:
2357 case OP_NOTMINQUERYI:
2358 case OP_POSQUERY:
2359 case OP_POSQUERYI:
2360 case OP_NOTPOSQUERY:
2361 case OP_NOTPOSQUERYI:
2362 if (HAS_EXTRALEN(code[-1])) code += GET_EXTRALEN(code[-1]);
2363 break;
2364 }
2365 #else
2366 (void)(utf); /* Keep compiler happy by referencing function argument */
2367 #endif
2368 }
2369 }
2370 }
2371
2372
2373
2374 /*************************************************
2375 * Scan compiled branch for non-emptiness *
2376 *************************************************/
2377
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.
2385
2386 Arguments:
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
2392
2393 Returns: TRUE if what is matched could be empty
2394 */
2395
2396 static BOOL
could_be_empty_branch(const pcre_uchar * code,const pcre_uchar * endcode,BOOL utf,compile_data * cd,recurse_check * recurses)2397 could_be_empty_branch(const pcre_uchar *code, const pcre_uchar *endcode,
2398 BOOL utf, compile_data *cd, recurse_check *recurses)
2399 {
2400 register pcre_uchar c;
2401 recurse_check this_recurse;
2402
2403 for (code = first_significant_code(code + PRIV(OP_lengths)[*code], TRUE);
2404 code < endcode;
2405 code = first_significant_code(code + PRIV(OP_lengths)[c], TRUE))
2406 {
2407 const pcre_uchar *ccode;
2408
2409 c = *code;
2410
2411 /* Skip over forward assertions; the other assertions are skipped by
2412 first_significant_code() with a TRUE final argument. */
2413
2414 if (c == OP_ASSERT)
2415 {
2416 do code += GET(code, 1); while (*code == OP_ALT);
2417 c = *code;
2418 continue;
2419 }
2420
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. */
2427
2428 if (c == OP_RECURSE)
2429 {
2430 const pcre_uchar *scode = cd->start_code + GET(code, 1);
2431 const pcre_uchar *endgroup = scode;
2432 BOOL empty_branch;
2433
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
2436 NULL. */
2437
2438 if (cd->start_workspace != NULL)
2439 {
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 */
2444 }
2445
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. */
2450
2451 do endgroup += GET(endgroup, 1); while (*endgroup == OP_ALT);
2452 if (code >= scode && code <= endgroup) continue; /* Simple recursion */
2453 else
2454 {
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 */
2459 }
2460
2461 /* Completed reference; scan the referenced group, remembering it on the
2462 stack chain to detect mutual recursions. */
2463
2464 empty_branch = FALSE;
2465 this_recurse.prev = recurses;
2466 this_recurse.group = scode;
2467
2468 do
2469 {
2470 if (could_be_empty_branch(scode, endcode, utf, cd, &this_recurse))
2471 {
2472 empty_branch = TRUE;
2473 break;
2474 }
2475 scode += GET(scode, 1);
2476 }
2477 while (*scode == OP_ALT);
2478
2479 if (!empty_branch) return FALSE; /* All branches are non-empty */
2480 continue;
2481 }
2482
2483 /* Groups with zero repeats can of course be empty; skip them. */
2484
2485 if (c == OP_BRAZERO || c == OP_BRAMINZERO || c == OP_SKIPZERO ||
2486 c == OP_BRAPOSZERO)
2487 {
2488 code += PRIV(OP_lengths)[c];
2489 do code += GET(code, 1); while (*code == OP_ALT);
2490 c = *code;
2491 continue;
2492 }
2493
2494 /* A nested group that is already marked as "could be empty" can just be
2495 skipped. */
2496
2497 if (c == OP_SBRA || c == OP_SBRAPOS ||
2498 c == OP_SCBRA || c == OP_SCBRAPOS)
2499 {
2500 do code += GET(code, 1); while (*code == OP_ALT);
2501 c = *code;
2502 continue;
2503 }
2504
2505 /* For other groups, scan the branches. */
2506
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)
2511 {
2512 BOOL empty_branch;
2513 if (GET(code, 1) == 0) return TRUE; /* Hit unclosed bracket */
2514
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. */
2518
2519 if (c == OP_COND && code[GET(code, 1)] != OP_ALT)
2520 code += GET(code, 1);
2521 else
2522 {
2523 empty_branch = FALSE;
2524 do
2525 {
2526 if (!empty_branch && could_be_empty_branch(code, endcode, utf, cd,
2527 recurses)) empty_branch = TRUE;
2528 code += GET(code, 1);
2529 }
2530 while (*code == OP_ALT);
2531 if (!empty_branch) return FALSE; /* All branches are non-empty */
2532 }
2533
2534 c = *code;
2535 continue;
2536 }
2537
2538 /* Handle the other opcodes */
2539
2540 switch (c)
2541 {
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"
2546 here. */
2547
2548 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
2549 case OP_XCLASS:
2550 ccode = code += GET(code, 1);
2551 goto CHECK_CLASS_REPEAT;
2552 #endif
2553
2554 case OP_CLASS:
2555 case OP_NCLASS:
2556 ccode = code + PRIV(OP_lengths)[OP_CLASS];
2557
2558 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
2559 CHECK_CLASS_REPEAT:
2560 #endif
2561
2562 switch (*ccode)
2563 {
2564 case OP_CRSTAR: /* These could be empty; continue */
2565 case OP_CRMINSTAR:
2566 case OP_CRQUERY:
2567 case OP_CRMINQUERY:
2568 case OP_CRPOSSTAR:
2569 case OP_CRPOSQUERY:
2570 break;
2571
2572 default: /* Non-repeat => class must match */
2573 case OP_CRPLUS: /* These repeats aren't empty */
2574 case OP_CRMINPLUS:
2575 case OP_CRPOSPLUS:
2576 return FALSE;
2577
2578 case OP_CRRANGE:
2579 case OP_CRMINRANGE:
2580 case OP_CRPOSRANGE:
2581 if (GET2(ccode, 1) > 0) return FALSE; /* Minimum > 0 */
2582 break;
2583 }
2584 break;
2585
2586 /* Opcodes that must match a character */
2587
2588 case OP_ANY:
2589 case OP_ALLANY:
2590 case OP_ANYBYTE:
2591
2592 case OP_PROP:
2593 case OP_NOTPROP:
2594 case OP_ANYNL:
2595
2596 case OP_NOT_HSPACE:
2597 case OP_HSPACE:
2598 case OP_NOT_VSPACE:
2599 case OP_VSPACE:
2600 case OP_EXTUNI:
2601
2602 case OP_NOT_DIGIT:
2603 case OP_DIGIT:
2604 case OP_NOT_WHITESPACE:
2605 case OP_WHITESPACE:
2606 case OP_NOT_WORDCHAR:
2607 case OP_WORDCHAR:
2608
2609 case OP_CHAR:
2610 case OP_CHARI:
2611 case OP_NOT:
2612 case OP_NOTI:
2613
2614 case OP_PLUS:
2615 case OP_PLUSI:
2616 case OP_MINPLUS:
2617 case OP_MINPLUSI:
2618
2619 case OP_NOTPLUS:
2620 case OP_NOTPLUSI:
2621 case OP_NOTMINPLUS:
2622 case OP_NOTMINPLUSI:
2623
2624 case OP_POSPLUS:
2625 case OP_POSPLUSI:
2626 case OP_NOTPOSPLUS:
2627 case OP_NOTPOSPLUSI:
2628
2629 case OP_EXACT:
2630 case OP_EXACTI:
2631 case OP_NOTEXACT:
2632 case OP_NOTEXACTI:
2633
2634 case OP_TYPEPLUS:
2635 case OP_TYPEMINPLUS:
2636 case OP_TYPEPOSPLUS:
2637 case OP_TYPEEXACT:
2638
2639 return FALSE;
2640
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. */
2643
2644 case OP_TYPESTAR:
2645 case OP_TYPEMINSTAR:
2646 case OP_TYPEPOSSTAR:
2647 case OP_TYPEQUERY:
2648 case OP_TYPEMINQUERY:
2649 case OP_TYPEPOSQUERY:
2650 if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
2651 break;
2652
2653 /* Same for these */
2654
2655 case OP_TYPEUPTO:
2656 case OP_TYPEMINUPTO:
2657 case OP_TYPEPOSUPTO:
2658 if (code[1 + IMM2_SIZE] == OP_PROP || code[1 + IMM2_SIZE] == OP_NOTPROP)
2659 code += 2;
2660 break;
2661
2662 /* End of branch */
2663
2664 case OP_KET:
2665 case OP_KETRMAX:
2666 case OP_KETRMIN:
2667 case OP_KETRPOS:
2668 case OP_ALT:
2669 return TRUE;
2670
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. */
2674
2675 #if defined SUPPORT_UTF && !defined COMPILE_PCRE32
2676 case OP_STAR:
2677 case OP_STARI:
2678 case OP_NOTSTAR:
2679 case OP_NOTSTARI:
2680
2681 case OP_MINSTAR:
2682 case OP_MINSTARI:
2683 case OP_NOTMINSTAR:
2684 case OP_NOTMINSTARI:
2685
2686 case OP_POSSTAR:
2687 case OP_POSSTARI:
2688 case OP_NOTPOSSTAR:
2689 case OP_NOTPOSSTARI:
2690
2691 case OP_QUERY:
2692 case OP_QUERYI:
2693 case OP_NOTQUERY:
2694 case OP_NOTQUERYI:
2695
2696 case OP_MINQUERY:
2697 case OP_MINQUERYI:
2698 case OP_NOTMINQUERY:
2699 case OP_NOTMINQUERYI:
2700
2701 case OP_POSQUERY:
2702 case OP_POSQUERYI:
2703 case OP_NOTPOSQUERY:
2704 case OP_NOTPOSQUERYI:
2705
2706 if (utf && HAS_EXTRALEN(code[1])) code += GET_EXTRALEN(code[1]);
2707 break;
2708
2709 case OP_UPTO:
2710 case OP_UPTOI:
2711 case OP_NOTUPTO:
2712 case OP_NOTUPTOI:
2713
2714 case OP_MINUPTO:
2715 case OP_MINUPTOI:
2716 case OP_NOTMINUPTO:
2717 case OP_NOTMINUPTOI:
2718
2719 case OP_POSUPTO:
2720 case OP_POSUPTOI:
2721 case OP_NOTPOSUPTO:
2722 case OP_NOTPOSUPTOI:
2723
2724 if (utf && HAS_EXTRALEN(code[1 + IMM2_SIZE])) code += GET_EXTRALEN(code[1 + IMM2_SIZE]);
2725 break;
2726 #endif
2727
2728 /* MARK, and PRUNE/SKIP/THEN with an argument must skip over the argument
2729 string. */
2730
2731 case OP_MARK:
2732 case OP_PRUNE_ARG:
2733 case OP_SKIP_ARG:
2734 case OP_THEN_ARG:
2735 code += code[1];
2736 break;
2737
2738 /* None of the remaining opcodes are required to match a character. */
2739
2740 default:
2741 break;
2742 }
2743 }
2744
2745 return TRUE;
2746 }
2747
2748
2749
2750 /*************************************************
2751 * Scan compiled regex for non-emptiness *
2752 *************************************************/
2753
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
2759 pre-compile.
2760
2761 Arguments:
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
2767
2768 Returns: TRUE if what is matched could be empty
2769 */
2770
2771 static BOOL
could_be_empty(const pcre_uchar * code,const pcre_uchar * endcode,branch_chain * bcptr,BOOL utf,compile_data * cd)2772 could_be_empty(const pcre_uchar *code, const pcre_uchar *endcode,
2773 branch_chain *bcptr, BOOL utf, compile_data *cd)
2774 {
2775 while (bcptr != NULL && bcptr->current_branch >= code)
2776 {
2777 if (!could_be_empty_branch(bcptr->current_branch, endcode, utf, cd, NULL))
2778 return FALSE;
2779 bcptr = bcptr->outer;
2780 }
2781 return TRUE;
2782 }
2783
2784
2785
2786 /*************************************************
2787 * Base opcode of repeated opcodes *
2788 *************************************************/
2789
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.
2792
2793 Arguments: c opcode
2794 Returns: base opcode for the type
2795 */
2796
2797 static pcre_uchar
get_repeat_base(pcre_uchar c)2798 get_repeat_base(pcre_uchar c)
2799 {
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 :
2805 OP_STAR;
2806 }
2807
2808
2809
2810 #ifdef SUPPORT_UCP
2811 /*************************************************
2812 * Check a character and a property *
2813 *************************************************/
2814
2815 /* This function is called by check_auto_possessive() when a property item
2816 is adjacent to a fixed character.
2817
2818 Arguments:
2819 c the 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{^)
2823
2824 Returns: TRUE if auto-possessifying is OK
2825 */
2826
2827 static BOOL
check_char_prop(pcre_uint32 c,unsigned int ptype,unsigned int pdata,BOOL negated)2828 check_char_prop(pcre_uint32 c, unsigned int ptype, unsigned int pdata,
2829 BOOL negated)
2830 {
2831 const pcre_uint32 *p;
2832 const ucd_record *prop = GET_UCD(c);
2833
2834 switch(ptype)
2835 {
2836 case PT_LAMP:
2837 return (prop->chartype == ucp_Lu ||
2838 prop->chartype == ucp_Ll ||
2839 prop->chartype == ucp_Lt) == negated;
2840
2841 case PT_GC:
2842 return (pdata == PRIV(ucp_gentype)[prop->chartype]) == negated;
2843
2844 case PT_PC:
2845 return (pdata == prop->chartype) == negated;
2846
2847 case PT_SC:
2848 return (pdata == prop->script) == negated;
2849
2850 /* These are specials */
2851
2852 case PT_ALNUM:
2853 return (PRIV(ucp_gentype)[prop->chartype] == ucp_L ||
2854 PRIV(ucp_gentype)[prop->chartype] == ucp_N) == negated;
2855
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
2858 at release 8.34. */
2859
2860 case PT_SPACE: /* Perl space */
2861 case PT_PXSPACE: /* POSIX space */
2862 switch(c)
2863 {
2864 HSPACE_CASES:
2865 VSPACE_CASES:
2866 return negated;
2867
2868 default:
2869 return (PRIV(ucp_gentype)[prop->chartype] == ucp_Z) == negated;
2870 }
2871 break; /* Control never reaches here */
2872
2873 case PT_WORD:
2874 return (PRIV(ucp_gentype)[prop->chartype] == ucp_L ||
2875 PRIV(ucp_gentype)[prop->chartype] == ucp_N ||
2876 c == CHAR_UNDERSCORE) == negated;
2877
2878 case PT_CLIST:
2879 p = PRIV(ucd_caseless_sets) + prop->caseset;
2880 for (;;)
2881 {
2882 if (c < *p) return !negated;
2883 if (c == *p++) return negated;
2884 }
2885 break; /* Control never reaches here */
2886 }
2887
2888 return FALSE;
2889 }
2890 #endif /* SUPPORT_UCP */
2891
2892
2893
2894 /*************************************************
2895 * Fill the character property list *
2896 *************************************************/
2897
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.
2900
2901 Arguments:
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
2910
2911 Returns: points to the start of the next opcode if *code is accepted
2912 NULL if *code is not accepted
2913 */
2914
2915 static const pcre_uchar *
get_chr_property_list(const pcre_uchar * code,BOOL utf,const pcre_uint8 * fcc,pcre_uint32 * list)2916 get_chr_property_list(const pcre_uchar *code, BOOL utf,
2917 const pcre_uint8 *fcc, pcre_uint32 *list)
2918 {
2919 pcre_uchar c = *code;
2920 pcre_uchar base;
2921 const pcre_uchar *end;
2922 pcre_uint32 chr;
2923
2924 #ifdef SUPPORT_UCP
2925 pcre_uint32 *clist_dest;
2926 const pcre_uint32 *clist_src;
2927 #else
2928 ((void)utf); /* Suppress "unused parameter" compiler warning */
2929 #endif
2930
2931 list[0] = c;
2932 list[1] = FALSE;
2933 code++;
2934
2935 if (c >= OP_STAR && c <= OP_TYPEPOSUPTO)
2936 {
2937 base = get_repeat_base(c);
2938 c -= (base - OP_STAR);
2939
2940 if (c == OP_UPTO || c == OP_MINUPTO || c == OP_EXACT || c == OP_POSUPTO)
2941 code += IMM2_SIZE;
2942
2943 list[1] = (c != OP_PLUS && c != OP_MINPLUS && c != OP_EXACT && c != OP_POSPLUS);
2944
2945 switch(base)
2946 {
2947 case OP_STAR:
2948 list[0] = OP_CHAR;
2949 break;
2950
2951 case OP_STARI:
2952 list[0] = OP_CHARI;
2953 break;
2954
2955 case OP_NOTSTAR:
2956 list[0] = OP_NOT;
2957 break;
2958
2959 case OP_NOTSTARI:
2960 list[0] = OP_NOTI;
2961 break;
2962
2963 case OP_TYPESTAR:
2964 list[0] = *code;
2965 code++;
2966 break;
2967 }
2968 c = list[0];
2969 }
2970
2971 switch(c)
2972 {
2973 case OP_NOT_DIGIT:
2974 case OP_DIGIT:
2975 case OP_NOT_WHITESPACE:
2976 case OP_WHITESPACE:
2977 case OP_NOT_WORDCHAR:
2978 case OP_WORDCHAR:
2979 case OP_ANY:
2980 case OP_ALLANY:
2981 case OP_ANYNL:
2982 case OP_NOT_HSPACE:
2983 case OP_HSPACE:
2984 case OP_NOT_VSPACE:
2985 case OP_VSPACE:
2986 case OP_EXTUNI:
2987 case OP_EODN:
2988 case OP_EOD:
2989 case OP_DOLL:
2990 case OP_DOLLM:
2991 return code;
2992
2993 case OP_CHAR:
2994 case OP_NOT:
2995 GETCHARINCTEST(chr, code);
2996 list[2] = chr;
2997 list[3] = NOTACHAR;
2998 return code;
2999
3000 case OP_CHARI:
3001 case OP_NOTI:
3002 list[0] = (c == OP_CHARI) ? OP_CHAR : OP_NOT;
3003 GETCHARINCTEST(chr, code);
3004 list[2] = chr;
3005
3006 #ifdef SUPPORT_UCP
3007 if (chr < 128 || (chr < 256 && !utf))
3008 list[3] = fcc[chr];
3009 else
3010 list[3] = UCD_OTHERCASE(chr);
3011 #elif defined SUPPORT_UTF || !defined COMPILE_PCRE8
3012 list[3] = (chr < 256) ? fcc[chr] : chr;
3013 #else
3014 list[3] = fcc[chr];
3015 #endif
3016
3017 /* The othercase might be the same value. */
3018
3019 if (chr == list[3])
3020 list[3] = NOTACHAR;
3021 else
3022 list[4] = NOTACHAR;
3023 return code;
3024
3025 #ifdef SUPPORT_UCP
3026 case OP_PROP:
3027 case OP_NOTPROP:
3028 if (code[0] != PT_CLIST)
3029 {
3030 list[2] = code[0];
3031 list[3] = code[1];
3032 return code + 2;
3033 }
3034
3035 /* Convert only if we have enough space. */
3036
3037 clist_src = PRIV(ucd_caseless_sets) + code[1];
3038 clist_dest = list + 2;
3039 code += 2;
3040
3041 do {
3042 if (clist_dest >= list + 8)
3043 {
3044 /* Early return if there is not enough space. This should never
3045 happen, since all clists are shorter than 5 character now. */
3046 list[2] = code[0];
3047 list[3] = code[1];
3048 return code;
3049 }
3050 *clist_dest++ = *clist_src;
3051 }
3052 while(*clist_src++ != NOTACHAR);
3053
3054 /* All characters are stored. The terminating NOTACHAR
3055 is copied form the clist itself. */
3056
3057 list[0] = (c == OP_PROP) ? OP_CHAR : OP_NOT;
3058 return code;
3059 #endif
3060
3061 case OP_NCLASS:
3062 case OP_CLASS:
3063 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
3064 case OP_XCLASS:
3065 if (c == OP_XCLASS)
3066 end = code + GET(code, 0) - 1;
3067 else
3068 #endif
3069 end = code + 32 / sizeof(pcre_uchar);
3070
3071 switch(*end)
3072 {
3073 case OP_CRSTAR:
3074 case OP_CRMINSTAR:
3075 case OP_CRQUERY:
3076 case OP_CRMINQUERY:
3077 case OP_CRPOSSTAR:
3078 case OP_CRPOSQUERY:
3079 list[1] = TRUE;
3080 end++;
3081 break;
3082
3083 case OP_CRPLUS:
3084 case OP_CRMINPLUS:
3085 case OP_CRPOSPLUS:
3086 end++;
3087 break;
3088
3089 case OP_CRRANGE:
3090 case OP_CRMINRANGE:
3091 case OP_CRPOSRANGE:
3092 list[1] = (GET2(end, 1) == 0);
3093 end += 1 + 2 * IMM2_SIZE;
3094 break;
3095 }
3096 list[2] = (pcre_uint32)(end - code);
3097 return end;
3098 }
3099 return NULL; /* Opcode not accepted */
3100 }
3101
3102
3103
3104 /*************************************************
3105 * Scan further character sets for match *
3106 *************************************************/
3107
3108 /* Checks whether the base and the current opcode have a common character, in
3109 which case the base cannot be possessified.
3110
3111 Arguments:
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
3116
3117 Returns: TRUE if the auto-possessification is possible
3118 */
3119
3120 static BOOL
compare_opcodes(const pcre_uchar * code,BOOL utf,const compile_data * cd,const pcre_uint32 * base_list,const pcre_uchar * base_end,int * rec_limit)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)
3123 {
3124 pcre_uchar c;
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;
3132 #endif
3133 const pcre_uint8 *class_bitset;
3134 const pcre_uint8 *set1, *set2, *set_end;
3135 pcre_uint32 chr;
3136 BOOL accepted, invert_bits;
3137 BOOL entered_a_group = FALSE;
3138
3139 if (*rec_limit == 0) return FALSE;
3140 --(*rec_limit);
3141
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). */
3146
3147 for(;;)
3148 {
3149 /* All operations move the code pointer forward.
3150 Therefore infinite recursions are not possible. */
3151
3152 c = *code;
3153
3154 /* Skip over callouts */
3155
3156 if (c == OP_CALLOUT)
3157 {
3158 code += PRIV(OP_lengths)[c];
3159 continue;
3160 }
3161
3162 if (c == OP_ALT)
3163 {
3164 do code += GET(code, 1); while (*code == OP_ALT);
3165 c = *code;
3166 }
3167
3168 switch(c)
3169 {
3170 case OP_END:
3171 case OP_KETRPOS:
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.) */
3175
3176 return base_list[1] != 0;
3177
3178 case OP_KET:
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. */
3182
3183 if (base_list[1] == 0) return FALSE;
3184
3185 switch(*(code - GET(code, 1)))
3186 {
3187 case OP_ASSERT:
3188 case OP_ASSERT_NOT:
3189 case OP_ASSERTBACK:
3190 case OP_ASSERTBACK_NOT:
3191 case OP_ONCE:
3192 case OP_ONCE_NC:
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. */
3196
3197 return !entered_a_group;
3198 }
3199
3200 code += PRIV(OP_lengths)[c];
3201 continue;
3202
3203 case OP_ONCE:
3204 case OP_ONCE_NC:
3205 case OP_BRA:
3206 case OP_CBRA:
3207 next_code = code + GET(code, 1);
3208 code += PRIV(OP_lengths)[c];
3209
3210 while (*next_code == OP_ALT)
3211 {
3212 if (!compare_opcodes(code, utf, cd, base_list, base_end, rec_limit))
3213 return FALSE;
3214 code = next_code + 1 + LINK_SIZE;
3215 next_code += GET(next_code, 1);
3216 }
3217
3218 entered_a_group = TRUE;
3219 continue;
3220
3221 case OP_BRAZERO:
3222 case OP_BRAMINZERO:
3223
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;
3227
3228 do next_code += GET(next_code, 1); while (*next_code == OP_ALT);
3229
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))
3234 return FALSE;
3235
3236 code += PRIV(OP_lengths)[c];
3237 continue;
3238
3239 default:
3240 break;
3241 }
3242
3243 /* Check for a supported opcode, and load its properties. */
3244
3245 code = get_chr_property_list(code, utf, cd->fcc, list);
3246 if (code == NULL) return FALSE; /* Unsupported */
3247
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. */
3250
3251 if (base_list[0] == OP_CHAR)
3252 {
3253 chr_ptr = base_list + 2;
3254 list_ptr = list;
3255 }
3256 else if (list[0] == OP_CHAR)
3257 {
3258 chr_ptr = list + 2;
3259 list_ptr = base_list;
3260 }
3261
3262 /* Character bitsets can also be compared to certain opcodes. */
3263
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))
3268 #endif
3269 )
3270 {
3271 #ifdef COMPILE_PCRE8
3272 if (base_list[0] == OP_CLASS || (!utf && base_list[0] == OP_NCLASS))
3273 #else
3274 if (base_list[0] == OP_CLASS)
3275 #endif
3276 {
3277 set1 = (pcre_uint8 *)(base_end - base_list[2]);
3278 list_ptr = list;
3279 }
3280 else
3281 {
3282 set1 = (pcre_uint8 *)(code - list[2]);
3283 list_ptr = base_list;
3284 }
3285
3286 invert_bits = FALSE;
3287 switch(list_ptr[0])
3288 {
3289 case OP_CLASS:
3290 case OP_NCLASS:
3291 set2 = (pcre_uint8 *)
3292 ((list_ptr == list ? code : base_end) - list_ptr[2]);
3293 break;
3294
3295 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
3296 case OP_XCLASS:
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)
3300 {
3301 /* No bits are set for characters < 256. */
3302 if (list[1] == 0) return (*xclass_flags & XCL_NOT) == 0;
3303 /* Might be an empty repeat. */
3304 continue;
3305 }
3306 set2 = (pcre_uint8 *)(xclass_flags + 1);
3307 break;
3308 #endif
3309
3310 case OP_NOT_DIGIT:
3311 invert_bits = TRUE;
3312 /* Fall through */
3313 case OP_DIGIT:
3314 set2 = (pcre_uint8 *)(cd->cbits + cbit_digit);
3315 break;
3316
3317 case OP_NOT_WHITESPACE:
3318 invert_bits = TRUE;
3319 /* Fall through */
3320 case OP_WHITESPACE:
3321 set2 = (pcre_uint8 *)(cd->cbits + cbit_space);
3322 break;
3323
3324 case OP_NOT_WORDCHAR:
3325 invert_bits = TRUE;
3326 /* Fall through */
3327 case OP_WORDCHAR:
3328 set2 = (pcre_uint8 *)(cd->cbits + cbit_word);
3329 break;
3330
3331 default:
3332 return FALSE;
3333 }
3334
3335 /* Because the sets are unaligned, we need
3336 to perform byte comparison here. */
3337 set_end = set1 + 32;
3338 if (invert_bits)
3339 {
3340 do
3341 {
3342 if ((*set1++ & ~(*set2++)) != 0) return FALSE;
3343 }
3344 while (set1 < set_end);
3345 }
3346 else
3347 {
3348 do
3349 {
3350 if ((*set1++ & *set2++) != 0) return FALSE;
3351 }
3352 while (set1 < set_end);
3353 }
3354
3355 if (list[1] == 0) return TRUE;
3356 /* Might be an empty repeat. */
3357 continue;
3358 }
3359
3360 /* Some property combinations also acceptable. Unicode property opcodes are
3361 processed specially; the rest can be handled with a lookup table. */
3362
3363 else
3364 {
3365 pcre_uint32 leftop, rightop;
3366
3367 leftop = base_list[0];
3368 rightop = list[0];
3369
3370 #ifdef SUPPORT_UCP
3371 accepted = FALSE; /* Always set in non-unicode case. */
3372 if (leftop == OP_PROP || leftop == OP_NOTPROP)
3373 {
3374 if (rightop == OP_EOD)
3375 accepted = TRUE;
3376 else if (rightop == OP_PROP || rightop == OP_NOTPROP)
3377 {
3378 int n;
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;
3384
3385 /* There's a table that specifies how each combination is to be
3386 processed:
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
3392 */
3393
3394 n = propposstab[base_list[2]][list[2]];
3395 switch(n)
3396 {
3397 case 0: break;
3398 case 1: accepted = bothprop; break;
3399 case 2: accepted = (base_list[3] == list[3]) != same; break;
3400 case 3: accepted = !same; break;
3401
3402 case 4: /* Left general category, right particular category */
3403 accepted = risprop && catposstab[base_list[3]][list[3]] == same;
3404 break;
3405
3406 case 5: /* Right general category, left particular category */
3407 accepted = lisprop && catposstab[list[3]][base_list[3]] == same;
3408 break;
3409
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.
3415
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
3424 in a NOTPROP case.
3425
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. */
3428
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 ==
3434 (list[3] != p[0] &&
3435 list[3] != p[1] &&
3436 (list[3] != p[2] || !lisprop));
3437 break;
3438
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));
3447 break;
3448
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));
3457 break;
3458
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));
3467 break;
3468 }
3469 }
3470 }
3471
3472 else
3473 #endif /* SUPPORT_UCP */
3474
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];
3478
3479 if (!accepted) return FALSE;
3480
3481 if (list[1] == 0) return TRUE;
3482 /* Might be an empty repeat. */
3483 continue;
3484 }
3485
3486 /* Control reaches here only if one of the items is a small character list.
3487 All characters are checked against the other side. */
3488
3489 do
3490 {
3491 chr = *chr_ptr;
3492
3493 switch(list_ptr[0])
3494 {
3495 case OP_CHAR:
3496 ochr_ptr = list_ptr + 2;
3497 do
3498 {
3499 if (chr == *ochr_ptr) return FALSE;
3500 ochr_ptr++;
3501 }
3502 while(*ochr_ptr != NOTACHAR);
3503 break;
3504
3505 case OP_NOT:
3506 ochr_ptr = list_ptr + 2;
3507 do
3508 {
3509 if (chr == *ochr_ptr)
3510 break;
3511 ochr_ptr++;
3512 }
3513 while(*ochr_ptr != NOTACHAR);
3514 if (*ochr_ptr == NOTACHAR) return FALSE; /* Not found */
3515 break;
3516
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. */
3519
3520 case OP_DIGIT:
3521 if (chr < 256 && (cd->ctypes[chr] & ctype_digit) != 0) return FALSE;
3522 break;
3523
3524 case OP_NOT_DIGIT:
3525 if (chr > 255 || (cd->ctypes[chr] & ctype_digit) == 0) return FALSE;
3526 break;
3527
3528 case OP_WHITESPACE:
3529 if (chr < 256 && (cd->ctypes[chr] & ctype_space) != 0) return FALSE;
3530 break;
3531
3532 case OP_NOT_WHITESPACE:
3533 if (chr > 255 || (cd->ctypes[chr] & ctype_space) == 0) return FALSE;
3534 break;
3535
3536 case OP_WORDCHAR:
3537 if (chr < 255 && (cd->ctypes[chr] & ctype_word) != 0) return FALSE;
3538 break;
3539
3540 case OP_NOT_WORDCHAR:
3541 if (chr > 255 || (cd->ctypes[chr] & ctype_word) == 0) return FALSE;
3542 break;
3543
3544 case OP_HSPACE:
3545 switch(chr)
3546 {
3547 HSPACE_CASES: return FALSE;
3548 default: break;
3549 }
3550 break;
3551
3552 case OP_NOT_HSPACE:
3553 switch(chr)
3554 {
3555 HSPACE_CASES: break;
3556 default: return FALSE;
3557 }
3558 break;
3559
3560 case OP_ANYNL:
3561 case OP_VSPACE:
3562 switch(chr)
3563 {
3564 VSPACE_CASES: return FALSE;
3565 default: break;
3566 }
3567 break;
3568
3569 case OP_NOT_VSPACE:
3570 switch(chr)
3571 {
3572 VSPACE_CASES: break;
3573 default: return FALSE;
3574 }
3575 break;
3576
3577 case OP_DOLL:
3578 case OP_EODN:
3579 switch (chr)
3580 {
3581 case CHAR_CR:
3582 case CHAR_LF:
3583 case CHAR_VT:
3584 case CHAR_FF:
3585 case CHAR_NEL:
3586 #ifndef EBCDIC
3587 case 0x2028:
3588 case 0x2029:
3589 #endif /* Not EBCDIC */
3590 return FALSE;
3591 }
3592 break;
3593
3594 case OP_EOD: /* Can always possessify before \z */
3595 break;
3596
3597 #ifdef SUPPORT_UCP
3598 case OP_PROP:
3599 case OP_NOTPROP:
3600 if (!check_char_prop(chr, list_ptr[2], list_ptr[3],
3601 list_ptr[0] == OP_NOTPROP))
3602 return FALSE;
3603 break;
3604 #endif
3605
3606 case OP_NCLASS:
3607 if (chr > 255) return FALSE;
3608 /* Fall through */
3609
3610 case OP_CLASS:
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] & (1U << (chr & 7))) != 0) return FALSE;
3615 break;
3616
3617 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
3618 case OP_XCLASS:
3619 if (PRIV(xclass)(chr, (list_ptr == list ? code : base_end) -
3620 list_ptr[2] + LINK_SIZE, utf)) return FALSE;
3621 break;
3622 #endif
3623
3624 default:
3625 return FALSE;
3626 }
3627
3628 chr_ptr++;
3629 }
3630 while(*chr_ptr != NOTACHAR);
3631
3632 /* At least one character must be matched from this opcode. */
3633
3634 if (list[1] == 0) return TRUE;
3635 }
3636
3637 /* Control never reaches here. There used to be a fail-save return FALSE; here,
3638 but some compilers complain about an unreachable statement. */
3639
3640 }
3641
3642
3643
3644 /*************************************************
3645 * Scan compiled regex for auto-possession *
3646 *************************************************/
3647
3648 /* Replaces single character iterations with their possessive alternatives
3649 if appropriate. This function modifies the compiled opcode!
3650
3651 Arguments:
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
3655
3656 Returns: nothing
3657 */
3658
3659 static void
auto_possessify(pcre_uchar * code,BOOL utf,const compile_data * cd)3660 auto_possessify(pcre_uchar *code, BOOL utf, const compile_data *cd)
3661 {
3662 register pcre_uchar c;
3663 const pcre_uchar *end;
3664 pcre_uchar *repeat_opcode;
3665 pcre_uint32 list[8];
3666 int rec_limit;
3667
3668 for (;;)
3669 {
3670 c = *code;
3671
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. */
3677
3678 if (c >= OP_TABLE_LENGTH) return;
3679
3680 if (c >= OP_STAR && c <= OP_TYPEPOSUPTO)
3681 {
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;
3686
3687 rec_limit = 1000;
3688 if (end != NULL && compare_opcodes(end, utf, cd, list, end, &rec_limit))
3689 {
3690 switch(c)
3691 {
3692 case OP_STAR:
3693 *code += OP_POSSTAR - OP_STAR;
3694 break;
3695
3696 case OP_MINSTAR:
3697 *code += OP_POSSTAR - OP_MINSTAR;
3698 break;
3699
3700 case OP_PLUS:
3701 *code += OP_POSPLUS - OP_PLUS;
3702 break;
3703
3704 case OP_MINPLUS:
3705 *code += OP_POSPLUS - OP_MINPLUS;
3706 break;
3707
3708 case OP_QUERY:
3709 *code += OP_POSQUERY - OP_QUERY;
3710 break;
3711
3712 case OP_MINQUERY:
3713 *code += OP_POSQUERY - OP_MINQUERY;
3714 break;
3715
3716 case OP_UPTO:
3717 *code += OP_POSUPTO - OP_UPTO;
3718 break;
3719
3720 case OP_MINUPTO:
3721 *code += OP_POSUPTO - OP_MINUPTO;
3722 break;
3723 }
3724 }
3725 c = *code;
3726 }
3727 else if (c == OP_CLASS || c == OP_NCLASS || c == OP_XCLASS)
3728 {
3729 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
3730 if (c == OP_XCLASS)
3731 repeat_opcode = code + GET(code, 1);
3732 else
3733 #endif
3734 repeat_opcode = code + 1 + (32 / sizeof(pcre_uchar));
3735
3736 c = *repeat_opcode;
3737 if (c >= OP_CRSTAR && c <= OP_CRMINRANGE)
3738 {
3739 /* end must not be NULL. */
3740 end = get_chr_property_list(code, utf, cd->fcc, list);
3741
3742 list[1] = (c & 1) == 0;
3743
3744 rec_limit = 1000;
3745 if (compare_opcodes(end, utf, cd, list, end, &rec_limit))
3746 {
3747 switch (c)
3748 {
3749 case OP_CRSTAR:
3750 case OP_CRMINSTAR:
3751 *repeat_opcode = OP_CRPOSSTAR;
3752 break;
3753
3754 case OP_CRPLUS:
3755 case OP_CRMINPLUS:
3756 *repeat_opcode = OP_CRPOSPLUS;
3757 break;
3758
3759 case OP_CRQUERY:
3760 case OP_CRMINQUERY:
3761 *repeat_opcode = OP_CRPOSQUERY;
3762 break;
3763
3764 case OP_CRRANGE:
3765 case OP_CRMINRANGE:
3766 *repeat_opcode = OP_CRPOSRANGE;
3767 break;
3768 }
3769 }
3770 }
3771 c = *code;
3772 }
3773
3774 switch(c)
3775 {
3776 case OP_END:
3777 return;
3778
3779 case OP_TYPESTAR:
3780 case OP_TYPEMINSTAR:
3781 case OP_TYPEPLUS:
3782 case OP_TYPEMINPLUS:
3783 case OP_TYPEQUERY:
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;
3789 break;
3790
3791 case OP_TYPEUPTO:
3792 case OP_TYPEMINUPTO:
3793 case OP_TYPEEXACT:
3794 case OP_TYPEPOSUPTO:
3795 if (code[1 + IMM2_SIZE] == OP_PROP || code[1 + IMM2_SIZE] == OP_NOTPROP)
3796 code += 2;
3797 break;
3798
3799 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
3800 case OP_XCLASS:
3801 code += GET(code, 1);
3802 break;
3803 #endif
3804
3805 case OP_MARK:
3806 case OP_PRUNE_ARG:
3807 case OP_SKIP_ARG:
3808 case OP_THEN_ARG:
3809 code += code[1];
3810 break;
3811 }
3812
3813 /* Add in the fixed length from the table */
3814
3815 code += PRIV(OP_lengths)[c];
3816
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. */
3820
3821 #if defined SUPPORT_UTF && !defined COMPILE_PCRE32
3822 if (utf) switch(c)
3823 {
3824 case OP_CHAR:
3825 case OP_CHARI:
3826 case OP_NOT:
3827 case OP_NOTI:
3828 case OP_STAR:
3829 case OP_MINSTAR:
3830 case OP_PLUS:
3831 case OP_MINPLUS:
3832 case OP_QUERY:
3833 case OP_MINQUERY:
3834 case OP_UPTO:
3835 case OP_MINUPTO:
3836 case OP_EXACT:
3837 case OP_POSSTAR:
3838 case OP_POSPLUS:
3839 case OP_POSQUERY:
3840 case OP_POSUPTO:
3841 case OP_STARI:
3842 case OP_MINSTARI:
3843 case OP_PLUSI:
3844 case OP_MINPLUSI:
3845 case OP_QUERYI:
3846 case OP_MINQUERYI:
3847 case OP_UPTOI:
3848 case OP_MINUPTOI:
3849 case OP_EXACTI:
3850 case OP_POSSTARI:
3851 case OP_POSPLUSI:
3852 case OP_POSQUERYI:
3853 case OP_POSUPTOI:
3854 case OP_NOTSTAR:
3855 case OP_NOTMINSTAR:
3856 case OP_NOTPLUS:
3857 case OP_NOTMINPLUS:
3858 case OP_NOTQUERY:
3859 case OP_NOTMINQUERY:
3860 case OP_NOTUPTO:
3861 case OP_NOTMINUPTO:
3862 case OP_NOTEXACT:
3863 case OP_NOTPOSSTAR:
3864 case OP_NOTPOSPLUS:
3865 case OP_NOTPOSQUERY:
3866 case OP_NOTPOSUPTO:
3867 case OP_NOTSTARI:
3868 case OP_NOTMINSTARI:
3869 case OP_NOTPLUSI:
3870 case OP_NOTMINPLUSI:
3871 case OP_NOTQUERYI:
3872 case OP_NOTMINQUERYI:
3873 case OP_NOTUPTOI:
3874 case OP_NOTMINUPTOI:
3875 case OP_NOTEXACTI:
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]);
3881 break;
3882 }
3883 #else
3884 (void)(utf); /* Keep compiler happy by referencing function argument */
3885 #endif
3886 }
3887 }
3888
3889
3890
3891 /*************************************************
3892 * Check for POSIX class syntax *
3893 *************************************************/
3894
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.
3899
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:].
3905
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.
3914
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
3918 a digit.
3919
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
3924 names.
3925
3926 Arguments:
3927 ptr pointer to the initial [
3928 endptr where to return the end pointer
3929
3930 Returns: TRUE or FALSE
3931 */
3932
3933 static BOOL
check_posix_syntax(const pcre_uchar * ptr,const pcre_uchar ** endptr)3934 check_posix_syntax(const pcre_uchar *ptr, const pcre_uchar **endptr)
3935 {
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++)
3939 {
3940 if (*ptr == CHAR_BACKSLASH &&
3941 (ptr[1] == CHAR_RIGHT_SQUARE_BRACKET ||
3942 ptr[1] == CHAR_BACKSLASH))
3943 ptr++;
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)
3947 {
3948 *endptr = ptr;
3949 return TRUE;
3950 }
3951 }
3952 return FALSE;
3953 }
3954
3955
3956
3957
3958 /*************************************************
3959 * Check POSIX class name *
3960 *************************************************/
3961
3962 /* This function is called to check the name given in a POSIX-style class entry
3963 such as [:alnum:].
3964
3965 Arguments:
3966 ptr points to the first letter
3967 len the length of the name
3968
3969 Returns: a value representing the name, or -1 if unknown
3970 */
3971
3972 static int
check_posix_name(const pcre_uchar * ptr,int len)3973 check_posix_name(const pcre_uchar *ptr, int len)
3974 {
3975 const char *pn = posix_names;
3976 register int yield = 0;
3977 while (posix_name_lengths[yield] != 0)
3978 {
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;
3982 yield++;
3983 }
3984 return -1;
3985 }
3986
3987
3988 /*************************************************
3989 * Adjust OP_RECURSE items in repeated group *
3990 *************************************************/
3991
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
4001 OP_END.
4002
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.
4009
4010 Arguments:
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
4016
4017 Returns: nothing
4018 */
4019
4020 static void
adjust_recurse(pcre_uchar * group,int adjust,BOOL utf,compile_data * cd,size_t save_hwm_offset)4021 adjust_recurse(pcre_uchar *group, int adjust, BOOL utf, compile_data *cd,
4022 size_t save_hwm_offset)
4023 {
4024 int offset;
4025 pcre_uchar *hc;
4026 pcre_uchar *ptr = group;
4027
4028 while ((ptr = (pcre_uchar *)find_recurse(ptr, utf)) != NULL)
4029 {
4030 for (hc = (pcre_uchar *)cd->start_workspace + save_hwm_offset; hc < cd->hwm;
4031 hc += LINK_SIZE)
4032 {
4033 offset = (int)GET(hc, 0);
4034 if (cd->start_code + offset == ptr + 1) break;
4035 }
4036
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. */
4039
4040 if (hc >= cd->hwm)
4041 {
4042 offset = (int)GET(ptr, 1);
4043 if (cd->start_code + offset >= group) PUT(ptr, 1, offset + adjust);
4044 }
4045
4046 ptr += 1 + LINK_SIZE;
4047 }
4048
4049 /* Now adjust all forward reference offsets for the group. */
4050
4051 for (hc = (pcre_uchar *)cd->start_workspace + save_hwm_offset; hc < cd->hwm;
4052 hc += LINK_SIZE)
4053 {
4054 offset = (int)GET(hc, 0);
4055 PUT(hc, 0, offset + adjust);
4056 }
4057 }
4058
4059
4060
4061 /*************************************************
4062 * Insert an automatic callout point *
4063 *************************************************/
4064
4065 /* This function is called when the PCRE_AUTO_CALLOUT option is set, to insert
4066 callout points before each pattern item.
4067
4068 Arguments:
4069 code current code pointer
4070 ptr current pattern pointer
4071 cd pointers to tables etc
4072
4073 Returns: new code pointer
4074 */
4075
4076 static pcre_uchar *
auto_callout(pcre_uchar * code,const pcre_uchar * ptr,compile_data * cd)4077 auto_callout(pcre_uchar *code, const pcre_uchar *ptr, compile_data *cd)
4078 {
4079 *code++ = OP_CALLOUT;
4080 *code++ = 255;
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;
4084 }
4085
4086
4087
4088 /*************************************************
4089 * Complete a callout item *
4090 *************************************************/
4091
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.
4095
4096 Arguments:
4097 previous_callout points to previous callout item
4098 ptr current pattern pointer
4099 cd pointers to tables etc
4100
4101 Returns: nothing
4102 */
4103
4104 static void
complete_callout(pcre_uchar * previous_callout,const pcre_uchar * ptr,compile_data * cd)4105 complete_callout(pcre_uchar *previous_callout, const pcre_uchar *ptr, compile_data *cd)
4106 {
4107 int length = (int)(ptr - cd->start_pattern - GET(previous_callout, 2));
4108 PUT(previous_callout, 2 + LINK_SIZE, length);
4109 }
4110
4111
4112
4113 #ifdef SUPPORT_UCP
4114 /*************************************************
4115 * Get othercase range *
4116 *************************************************/
4117
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.
4123
4124 Arguments:
4125 cptr points to starting character value; updated
4126 d end value
4127 ocptr where to put start of othercase range
4128 odptr where to put end of othercase range
4129
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
4134 */
4135
4136 static int
get_othercase_range(pcre_uint32 * cptr,pcre_uint32 d,pcre_uint32 * ocptr,pcre_uint32 * odptr)4137 get_othercase_range(pcre_uint32 *cptr, pcre_uint32 d, pcre_uint32 *ocptr,
4138 pcre_uint32 *odptr)
4139 {
4140 pcre_uint32 c, othercase, next;
4141 unsigned int co;
4142
4143 /* Find the first character that has an other case. If it has multiple other
4144 cases, return its case offset value. */
4145
4146 for (c = *cptr; c <= d; c++)
4147 {
4148 if ((co = UCD_CASESET(c)) != 0)
4149 {
4150 *ocptr = c++; /* Character that has the set */
4151 *cptr = c; /* Rest of input range */
4152 return (int)co;
4153 }
4154 if ((othercase = UCD_OTHERCASE(c)) != c) break;
4155 }
4156
4157 if (c > d) return -1; /* Reached end of range */
4158
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. */
4162
4163 *ocptr = othercase;
4164 next = othercase + 1;
4165
4166 for (++c; c <= d; c++)
4167 {
4168 if ((co = UCD_CASESET(c)) != 0 || UCD_OTHERCASE(c) != next) break;
4169 next++;
4170 }
4171
4172 *odptr = next - 1; /* End of othercase range */
4173 *cptr = c; /* Rest of input range */
4174 return 0;
4175 }
4176 #endif /* SUPPORT_UCP */
4177
4178
4179
4180 /*************************************************
4181 * Add a character or range to a class *
4182 *************************************************/
4183
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.
4188
4189 Arguments:
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
4196
4197 Returns: the number of < 256 characters added
4198 the pointer to extra data is updated
4199 */
4200
4201 static int
add_to_class(pcre_uint8 * classbits,pcre_uchar ** uchardptr,int options,compile_data * cd,pcre_uint32 start,pcre_uint32 end)4202 add_to_class(pcre_uint8 *classbits, pcre_uchar **uchardptr, int options,
4203 compile_data *cd, pcre_uint32 start, pcre_uint32 end)
4204 {
4205 pcre_uint32 c;
4206 pcre_uint32 classbits_end = (end <= 0xff ? end : 0xff);
4207 int n8 = 0;
4208
4209 ((void)uchardptr);
4210 ((void)propposstab);
4211 ((void)catposstab);
4212 ((void)posspropstab);
4213
4214 /* If caseless matching is required, scan the range and process alternate
4215 cases. In Unicode, there are 8-bit characters that have alternate cases that
4216 are greater than 255 and vice-versa. Sometimes we can just extend the original
4217 range. */
4218
4219 if ((options & PCRE_CASELESS) != 0)
4220 {
4221 #ifdef SUPPORT_UCP
4222 if ((options & PCRE_UTF8) != 0)
4223 {
4224 int rc;
4225 pcre_uint32 oc, od;
4226
4227 options &= ~PCRE_CASELESS; /* Remove for recursive calls */
4228 c = start;
4229
4230 while ((rc = get_othercase_range(&c, end, &oc, &od)) >= 0)
4231 {
4232 /* Handle a single character that has more than one other case. */
4233
4234 if (rc > 0) n8 += add_list_to_class(classbits, uchardptr, options, cd,
4235 PRIV(ucd_caseless_sets) + rc, oc);
4236
4237 /* Do nothing if the other case range is within the original range. */
4238
4239 else if (oc >= start && od <= end) continue;
4240
4241 /* Extend the original range if there is overlap, noting that if oc < c, we
4242 can't have od > end because a subrange is always shorter than the basic
4243 range. Otherwise, use a recursive call to add the additional range. */
4244
4245 else if (oc < start && od >= start - 1) start = oc; /* Extend downwards */
4246 else if (od > end && oc <= end + 1)
4247 {
4248 end = od; /* Extend upwards */
4249 if (end > classbits_end) classbits_end = (end <= 0xff ? end : 0xff);
4250 }
4251 else n8 += add_to_class(classbits, uchardptr, options, cd, oc, od);
4252 }
4253 }
4254 else
4255 #endif /* SUPPORT_UCP */
4256
4257 /* Not UTF-mode, or no UCP */
4258
4259 for (c = start; c <= classbits_end; c++)
4260 {
4261 SETBIT(classbits, cd->fcc[c]);
4262 n8++;
4263 }
4264 }
4265
4266 /* Now handle the original range. Adjust the final value according to the bit
4267 length - this means that the same lists of (e.g.) horizontal spaces can be used
4268 in all cases. */
4269
4270 #if defined COMPILE_PCRE8
4271 #ifdef SUPPORT_UTF
4272 if ((options & PCRE_UTF8) == 0)
4273 #endif
4274 if (end > 0xff) end = 0xff;
4275
4276 #elif defined COMPILE_PCRE16
4277 #ifdef SUPPORT_UTF
4278 if ((options & PCRE_UTF16) == 0)
4279 #endif
4280 if (end > 0xffff) end = 0xffff;
4281
4282 #endif /* COMPILE_PCRE[8|16] */
4283
4284 /* Use the bitmap for characters < 256. Otherwise use extra data.*/
4285
4286 for (c = start; c <= classbits_end; c++)
4287 {
4288 /* Regardless of start, c will always be <= 255. */
4289 SETBIT(classbits, c);
4290 n8++;
4291 }
4292
4293 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
4294 if (start <= 0xff) start = 0xff + 1;
4295
4296 if (end >= start)
4297 {
4298 pcre_uchar *uchardata = *uchardptr;
4299 #ifdef SUPPORT_UTF
4300 if ((options & PCRE_UTF8) != 0) /* All UTFs use the same flag bit */
4301 {
4302 if (start < end)
4303 {
4304 *uchardata++ = XCL_RANGE;
4305 uchardata += PRIV(ord2utf)(start, uchardata);
4306 uchardata += PRIV(ord2utf)(end, uchardata);
4307 }
4308 else if (start == end)
4309 {
4310 *uchardata++ = XCL_SINGLE;
4311 uchardata += PRIV(ord2utf)(start, uchardata);
4312 }
4313 }
4314 else
4315 #endif /* SUPPORT_UTF */
4316
4317 /* Without UTF support, character values are constrained by the bit length,
4318 and can only be > 256 for 16-bit and 32-bit libraries. */
4319
4320 #ifdef COMPILE_PCRE8
4321 {}
4322 #else
4323 if (start < end)
4324 {
4325 *uchardata++ = XCL_RANGE;
4326 *uchardata++ = start;
4327 *uchardata++ = end;
4328 }
4329 else if (start == end)
4330 {
4331 *uchardata++ = XCL_SINGLE;
4332 *uchardata++ = start;
4333 }
4334 #endif
4335
4336 *uchardptr = uchardata; /* Updata extra data pointer */
4337 }
4338 #endif /* SUPPORT_UTF || !COMPILE_PCRE8 */
4339
4340 return n8; /* Number of 8-bit characters */
4341 }
4342
4343
4344
4345
4346 /*************************************************
4347 * Add a list of characters to a class *
4348 *************************************************/
4349
4350 /* This function is used for adding a list of case-equivalent characters to a
4351 class, and also for adding a list of horizontal or vertical whitespace. If the
4352 list is in order (which it should be), ranges of characters are detected and
4353 handled appropriately. This function is mutually recursive with the function
4354 above.
4355
4356 Arguments:
4357 classbits the bit map for characters < 256
4358 uchardptr points to the pointer for extra data
4359 options the options word
4360 cd contains pointers to tables etc.
4361 p points to row of 32-bit values, terminated by NOTACHAR
4362 except character to omit; this is used when adding lists of
4363 case-equivalent characters to avoid including the one we
4364 already know about
4365
4366 Returns: the number of < 256 characters added
4367 the pointer to extra data is updated
4368 */
4369
4370 static int
add_list_to_class(pcre_uint8 * classbits,pcre_uchar ** uchardptr,int options,compile_data * cd,const pcre_uint32 * p,unsigned int except)4371 add_list_to_class(pcre_uint8 *classbits, pcre_uchar **uchardptr, int options,
4372 compile_data *cd, const pcre_uint32 *p, unsigned int except)
4373 {
4374 int n8 = 0;
4375 while (p[0] < NOTACHAR)
4376 {
4377 int n = 0;
4378 if (p[0] != except)
4379 {
4380 while(p[n+1] == p[0] + n + 1) n++;
4381 n8 += add_to_class(classbits, uchardptr, options, cd, p[0], p[n]);
4382 }
4383 p += n + 1;
4384 }
4385 return n8;
4386 }
4387
4388
4389
4390 /*************************************************
4391 * Add characters not in a list to a class *
4392 *************************************************/
4393
4394 /* This function is used for adding the complement of a list of horizontal or
4395 vertical whitespace to a class. The list must be in order.
4396
4397 Arguments:
4398 classbits the bit map for characters < 256
4399 uchardptr points to the pointer for extra data
4400 options the options word
4401 cd contains pointers to tables etc.
4402 p points to row of 32-bit values, terminated by NOTACHAR
4403
4404 Returns: the number of < 256 characters added
4405 the pointer to extra data is updated
4406 */
4407
4408 static int
add_not_list_to_class(pcre_uint8 * classbits,pcre_uchar ** uchardptr,int options,compile_data * cd,const pcre_uint32 * p)4409 add_not_list_to_class(pcre_uint8 *classbits, pcre_uchar **uchardptr,
4410 int options, compile_data *cd, const pcre_uint32 *p)
4411 {
4412 BOOL utf = (options & PCRE_UTF8) != 0;
4413 int n8 = 0;
4414 if (p[0] > 0)
4415 n8 += add_to_class(classbits, uchardptr, options, cd, 0, p[0] - 1);
4416 while (p[0] < NOTACHAR)
4417 {
4418 while (p[1] == p[0] + 1) p++;
4419 n8 += add_to_class(classbits, uchardptr, options, cd, p[0] + 1,
4420 (p[1] == NOTACHAR) ? (utf ? 0x10ffffu : 0xffffffffu) : p[1] - 1);
4421 p++;
4422 }
4423 return n8;
4424 }
4425
4426
4427
4428 /*************************************************
4429 * Compile one branch *
4430 *************************************************/
4431
4432 /* Scan the pattern, compiling it into the a vector. If the options are
4433 changed during the branch, the pointer is used to change the external options
4434 bits. This function is used during the pre-compile phase when we are trying
4435 to find out the amount of memory needed, as well as during the real compile
4436 phase. The value of lengthptr distinguishes the two phases.
4437
4438 Arguments:
4439 optionsptr pointer to the option bits
4440 codeptr points to the pointer to the current code point
4441 ptrptr points to the current pattern pointer
4442 errorcodeptr points to error code variable
4443 firstcharptr place to put the first required character
4444 firstcharflagsptr place to put the first character flags, or a negative number
4445 reqcharptr place to put the last required character
4446 reqcharflagsptr place to put the last required character flags, or a negative number
4447 bcptr points to current branch chain
4448 cond_depth conditional nesting depth
4449 cd contains pointers to tables etc.
4450 lengthptr NULL during the real compile phase
4451 points to length accumulator during pre-compile phase
4452
4453 Returns: TRUE on success
4454 FALSE, with *errorcodeptr set non-zero on error
4455 */
4456
4457 static BOOL
compile_branch(int * optionsptr,pcre_uchar ** codeptr,const pcre_uchar ** ptrptr,int * errorcodeptr,pcre_uint32 * firstcharptr,pcre_int32 * firstcharflagsptr,pcre_uint32 * reqcharptr,pcre_int32 * reqcharflagsptr,branch_chain * bcptr,int cond_depth,compile_data * cd,int * lengthptr)4458 compile_branch(int *optionsptr, pcre_uchar **codeptr,
4459 const pcre_uchar **ptrptr, int *errorcodeptr,
4460 pcre_uint32 *firstcharptr, pcre_int32 *firstcharflagsptr,
4461 pcre_uint32 *reqcharptr, pcre_int32 *reqcharflagsptr,
4462 branch_chain *bcptr, int cond_depth,
4463 compile_data *cd, int *lengthptr)
4464 {
4465 int repeat_type, op_type;
4466 int repeat_min = 0, repeat_max = 0; /* To please picky compilers */
4467 int bravalue = 0;
4468 int greedy_default, greedy_non_default;
4469 pcre_uint32 firstchar, reqchar;
4470 pcre_int32 firstcharflags, reqcharflags;
4471 pcre_uint32 zeroreqchar, zerofirstchar;
4472 pcre_int32 zeroreqcharflags, zerofirstcharflags;
4473 pcre_int32 req_caseopt, reqvary, tempreqvary;
4474 int options = *optionsptr; /* May change dynamically */
4475 int after_manual_callout = 0;
4476 int length_prevgroup = 0;
4477 register pcre_uint32 c;
4478 int escape;
4479 register pcre_uchar *code = *codeptr;
4480 pcre_uchar *last_code = code;
4481 pcre_uchar *orig_code = code;
4482 pcre_uchar *tempcode;
4483 BOOL inescq = FALSE;
4484 BOOL groupsetfirstchar = FALSE;
4485 const pcre_uchar *ptr = *ptrptr;
4486 const pcre_uchar *tempptr;
4487 const pcre_uchar *nestptr = NULL;
4488 pcre_uchar *previous = NULL;
4489 pcre_uchar *previous_callout = NULL;
4490 size_t item_hwm_offset = 0;
4491 pcre_uint8 classbits[32];
4492
4493 /* We can fish out the UTF-8 setting once and for all into a BOOL, but we
4494 must not do this for other options (e.g. PCRE_EXTENDED) because they may change
4495 dynamically as we process the pattern. */
4496
4497 #ifdef SUPPORT_UTF
4498 /* PCRE_UTF[16|32] have the same value as PCRE_UTF8. */
4499 BOOL utf = (options & PCRE_UTF8) != 0;
4500 #ifndef COMPILE_PCRE32
4501 pcre_uchar utf_chars[6];
4502 #endif
4503 #else
4504 BOOL utf = FALSE;
4505 #endif
4506
4507 /* Helper variables for OP_XCLASS opcode (for characters > 255). We define
4508 class_uchardata always so that it can be passed to add_to_class() always,
4509 though it will not be used in non-UTF 8-bit cases. This avoids having to supply
4510 alternative calls for the different cases. */
4511
4512 pcre_uchar *class_uchardata;
4513 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
4514 BOOL xclass;
4515 pcre_uchar *class_uchardata_base;
4516 #endif
4517
4518 #ifdef PCRE_DEBUG
4519 if (lengthptr != NULL) DPRINTF((">> start branch\n"));
4520 #endif
4521
4522 /* Set up the default and non-default settings for greediness */
4523
4524 greedy_default = ((options & PCRE_UNGREEDY) != 0);
4525 greedy_non_default = greedy_default ^ 1;
4526
4527 /* Initialize no first byte, no required byte. REQ_UNSET means "no char
4528 matching encountered yet". It gets changed to REQ_NONE if we hit something that
4529 matches a non-fixed char first char; reqchar just remains unset if we never
4530 find one.
4531
4532 When we hit a repeat whose minimum is zero, we may have to adjust these values
4533 to take the zero repeat into account. This is implemented by setting them to
4534 zerofirstbyte and zeroreqchar when such a repeat is encountered. The individual
4535 item types that can be repeated set these backoff variables appropriately. */
4536
4537 firstchar = reqchar = zerofirstchar = zeroreqchar = 0;
4538 firstcharflags = reqcharflags = zerofirstcharflags = zeroreqcharflags = REQ_UNSET;
4539
4540 /* The variable req_caseopt contains either the REQ_CASELESS value
4541 or zero, according to the current setting of the caseless flag. The
4542 REQ_CASELESS leaves the lower 28 bit empty. It is added into the
4543 firstchar or reqchar variables to record the case status of the
4544 value. This is used only for ASCII characters. */
4545
4546 req_caseopt = ((options & PCRE_CASELESS) != 0)? REQ_CASELESS:0;
4547
4548 /* Switch on next character until the end of the branch */
4549
4550 for (;; ptr++)
4551 {
4552 BOOL negate_class;
4553 BOOL should_flip_negation;
4554 BOOL possessive_quantifier;
4555 BOOL is_quantifier;
4556 BOOL is_recurse;
4557 BOOL reset_bracount;
4558 int class_has_8bitchar;
4559 int class_one_char;
4560 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
4561 BOOL xclass_has_prop;
4562 #endif
4563 int newoptions;
4564 int recno;
4565 int refsign;
4566 int skipbytes;
4567 pcre_uint32 subreqchar, subfirstchar;
4568 pcre_int32 subreqcharflags, subfirstcharflags;
4569 int terminator;
4570 unsigned int mclength;
4571 unsigned int tempbracount;
4572 pcre_uint32 ec;
4573 pcre_uchar mcbuffer[8];
4574
4575 /* Come here to restart the loop without advancing the pointer. */
4576
4577 REDO_LOOP:
4578
4579 /* Get next character in the pattern */
4580
4581 c = *ptr;
4582
4583 /* If we are at the end of a nested substitution, revert to the outer level
4584 string. Nesting only happens one level deep. */
4585
4586 if (c == CHAR_NULL && nestptr != NULL)
4587 {
4588 ptr = nestptr;
4589 nestptr = NULL;
4590 c = *ptr;
4591 }
4592
4593 /* If we are in the pre-compile phase, accumulate the length used for the
4594 previous cycle of this loop. */
4595
4596 if (lengthptr != NULL)
4597 {
4598 #ifdef PCRE_DEBUG
4599 if (code > cd->hwm) cd->hwm = code; /* High water info */
4600 #endif
4601 if (code > cd->start_workspace + cd->workspace_size -
4602 WORK_SIZE_SAFETY_MARGIN) /* Check for overrun */
4603 {
4604 *errorcodeptr = (code >= cd->start_workspace + cd->workspace_size)?
4605 ERR52 : ERR87;
4606 goto FAILED;
4607 }
4608
4609 /* There is at least one situation where code goes backwards: this is the
4610 case of a zero quantifier after a class (e.g. [ab]{0}). At compile time,
4611 the class is simply eliminated. However, it is created first, so we have to
4612 allow memory for it. Therefore, don't ever reduce the length at this point.
4613 */
4614
4615 if (code < last_code) code = last_code;
4616
4617 /* Paranoid check for integer overflow */
4618
4619 if (OFLOW_MAX - *lengthptr < code - last_code)
4620 {
4621 *errorcodeptr = ERR20;
4622 goto FAILED;
4623 }
4624
4625 *lengthptr += (int)(code - last_code);
4626 DPRINTF(("length=%d added %d c=%c (0x%x)\n", *lengthptr,
4627 (int)(code - last_code), c, c));
4628
4629 /* If "previous" is set and it is not at the start of the work space, move
4630 it back to there, in order to avoid filling up the work space. Otherwise,
4631 if "previous" is NULL, reset the current code pointer to the start. */
4632
4633 if (previous != NULL)
4634 {
4635 if (previous > orig_code)
4636 {
4637 memmove(orig_code, previous, IN_UCHARS(code - previous));
4638 code -= previous - orig_code;
4639 previous = orig_code;
4640 }
4641 }
4642 else code = orig_code;
4643
4644 /* Remember where this code item starts so we can pick up the length
4645 next time round. */
4646
4647 last_code = code;
4648 }
4649
4650 /* In the real compile phase, just check the workspace used by the forward
4651 reference list. */
4652
4653 else if (cd->hwm > cd->start_workspace + cd->workspace_size)
4654 {
4655 *errorcodeptr = ERR52;
4656 goto FAILED;
4657 }
4658
4659 /* If in \Q...\E, check for the end; if not, we have a literal. Otherwise an
4660 isolated \E is ignored. */
4661
4662 if (c != CHAR_NULL)
4663 {
4664 if (c == CHAR_BACKSLASH && ptr[1] == CHAR_E)
4665 {
4666 inescq = FALSE;
4667 ptr++;
4668 continue;
4669 }
4670 else if (inescq)
4671 {
4672 if (previous_callout != NULL)
4673 {
4674 if (lengthptr == NULL) /* Don't attempt in pre-compile phase */
4675 complete_callout(previous_callout, ptr, cd);
4676 previous_callout = NULL;
4677 }
4678 if ((options & PCRE_AUTO_CALLOUT) != 0)
4679 {
4680 previous_callout = code;
4681 code = auto_callout(code, ptr, cd);
4682 }
4683 goto NORMAL_CHAR;
4684 }
4685
4686 /* Check for the start of a \Q...\E sequence. We must do this here rather
4687 than later in case it is immediately followed by \E, which turns it into a
4688 "do nothing" sequence. */
4689
4690 if (c == CHAR_BACKSLASH && ptr[1] == CHAR_Q)
4691 {
4692 inescq = TRUE;
4693 ptr++;
4694 continue;
4695 }
4696 }
4697
4698 /* In extended mode, skip white space and comments. */
4699
4700 if ((options & PCRE_EXTENDED) != 0)
4701 {
4702 const pcre_uchar *wscptr = ptr;
4703 while (MAX_255(c) && (cd->ctypes[c] & ctype_space) != 0) c = *(++ptr);
4704 if (c == CHAR_NUMBER_SIGN)
4705 {
4706 ptr++;
4707 while (*ptr != CHAR_NULL)
4708 {
4709 if (IS_NEWLINE(ptr)) /* For non-fixed-length newline cases, */
4710 { /* IS_NEWLINE sets cd->nllen. */
4711 ptr += cd->nllen;
4712 break;
4713 }
4714 ptr++;
4715 #ifdef SUPPORT_UTF
4716 if (utf) FORWARDCHAR(ptr);
4717 #endif
4718 }
4719 }
4720
4721 /* If we skipped any characters, restart the loop. Otherwise, we didn't see
4722 a comment. */
4723
4724 if (ptr > wscptr) goto REDO_LOOP;
4725 }
4726
4727 /* Skip over (?# comments. We need to do this here because we want to know if
4728 the next thing is a quantifier, and these comments may come between an item
4729 and its quantifier. */
4730
4731 if (c == CHAR_LEFT_PARENTHESIS && ptr[1] == CHAR_QUESTION_MARK &&
4732 ptr[2] == CHAR_NUMBER_SIGN)
4733 {
4734 ptr += 3;
4735 while (*ptr != CHAR_NULL && *ptr != CHAR_RIGHT_PARENTHESIS) ptr++;
4736 if (*ptr == CHAR_NULL)
4737 {
4738 *errorcodeptr = ERR18;
4739 goto FAILED;
4740 }
4741 continue;
4742 }
4743
4744 /* See if the next thing is a quantifier. */
4745
4746 is_quantifier =
4747 c == CHAR_ASTERISK || c == CHAR_PLUS || c == CHAR_QUESTION_MARK ||
4748 (c == CHAR_LEFT_CURLY_BRACKET && is_counted_repeat(ptr+1));
4749
4750 /* Fill in length of a previous callout, except when the next thing is a
4751 quantifier or when processing a property substitution string in UCP mode. */
4752
4753 if (!is_quantifier && previous_callout != NULL && nestptr == NULL &&
4754 after_manual_callout-- <= 0)
4755 {
4756 if (lengthptr == NULL) /* Don't attempt in pre-compile phase */
4757 complete_callout(previous_callout, ptr, cd);
4758 previous_callout = NULL;
4759 }
4760
4761 /* Create auto callout, except for quantifiers, or while processing property
4762 strings that are substituted for \w etc in UCP mode. */
4763
4764 if ((options & PCRE_AUTO_CALLOUT) != 0 && !is_quantifier && nestptr == NULL)
4765 {
4766 previous_callout = code;
4767 code = auto_callout(code, ptr, cd);
4768 }
4769
4770 /* Process the next pattern item. */
4771
4772 switch(c)
4773 {
4774 /* ===================================================================*/
4775 case CHAR_NULL: /* The branch terminates at string end */
4776 case CHAR_VERTICAL_LINE: /* or | or ) */
4777 case CHAR_RIGHT_PARENTHESIS:
4778 *firstcharptr = firstchar;
4779 *firstcharflagsptr = firstcharflags;
4780 *reqcharptr = reqchar;
4781 *reqcharflagsptr = reqcharflags;
4782 *codeptr = code;
4783 *ptrptr = ptr;
4784 if (lengthptr != NULL)
4785 {
4786 if (OFLOW_MAX - *lengthptr < code - last_code)
4787 {
4788 *errorcodeptr = ERR20;
4789 goto FAILED;
4790 }
4791 *lengthptr += (int)(code - last_code); /* To include callout length */
4792 DPRINTF((">> end branch\n"));
4793 }
4794 return TRUE;
4795
4796
4797 /* ===================================================================*/
4798 /* Handle single-character metacharacters. In multiline mode, ^ disables
4799 the setting of any following char as a first character. */
4800
4801 case CHAR_CIRCUMFLEX_ACCENT:
4802 previous = NULL;
4803 if ((options & PCRE_MULTILINE) != 0)
4804 {
4805 if (firstcharflags == REQ_UNSET)
4806 zerofirstcharflags = firstcharflags = REQ_NONE;
4807 *code++ = OP_CIRCM;
4808 }
4809 else *code++ = OP_CIRC;
4810 break;
4811
4812 case CHAR_DOLLAR_SIGN:
4813 previous = NULL;
4814 *code++ = ((options & PCRE_MULTILINE) != 0)? OP_DOLLM : OP_DOLL;
4815 break;
4816
4817 /* There can never be a first char if '.' is first, whatever happens about
4818 repeats. The value of reqchar doesn't change either. */
4819
4820 case CHAR_DOT:
4821 if (firstcharflags == REQ_UNSET) firstcharflags = REQ_NONE;
4822 zerofirstchar = firstchar;
4823 zerofirstcharflags = firstcharflags;
4824 zeroreqchar = reqchar;
4825 zeroreqcharflags = reqcharflags;
4826 previous = code;
4827 item_hwm_offset = cd->hwm - cd->start_workspace;
4828 *code++ = ((options & PCRE_DOTALL) != 0)? OP_ALLANY: OP_ANY;
4829 break;
4830
4831
4832 /* ===================================================================*/
4833 /* Character classes. If the included characters are all < 256, we build a
4834 32-byte bitmap of the permitted characters, except in the special case
4835 where there is only one such character. For negated classes, we build the
4836 map as usual, then invert it at the end. However, we use a different opcode
4837 so that data characters > 255 can be handled correctly.
4838
4839 If the class contains characters outside the 0-255 range, a different
4840 opcode is compiled. It may optionally have a bit map for characters < 256,
4841 but those above are are explicitly listed afterwards. A flag byte tells
4842 whether the bitmap is present, and whether this is a negated class or not.
4843
4844 In JavaScript compatibility mode, an isolated ']' causes an error. In
4845 default (Perl) mode, it is treated as a data character. */
4846
4847 case CHAR_RIGHT_SQUARE_BRACKET:
4848 if ((cd->external_options & PCRE_JAVASCRIPT_COMPAT) != 0)
4849 {
4850 *errorcodeptr = ERR64;
4851 goto FAILED;
4852 }
4853 goto NORMAL_CHAR;
4854
4855 /* In another (POSIX) regex library, the ugly syntax [[:<:]] and [[:>:]] is
4856 used for "start of word" and "end of word". As these are otherwise illegal
4857 sequences, we don't break anything by recognizing them. They are replaced
4858 by \b(?=\w) and \b(?<=\w) respectively. Sequences like [a[:<:]] are
4859 erroneous and are handled by the normal code below. */
4860
4861 case CHAR_LEFT_SQUARE_BRACKET:
4862 if (STRNCMP_UC_C8(ptr+1, STRING_WEIRD_STARTWORD, 6) == 0)
4863 {
4864 nestptr = ptr + 7;
4865 ptr = sub_start_of_word;
4866 goto REDO_LOOP;
4867 }
4868
4869 if (STRNCMP_UC_C8(ptr+1, STRING_WEIRD_ENDWORD, 6) == 0)
4870 {
4871 nestptr = ptr + 7;
4872 ptr = sub_end_of_word;
4873 goto REDO_LOOP;
4874 }
4875
4876 /* Handle a real character class. */
4877
4878 previous = code;
4879 item_hwm_offset = cd->hwm - cd->start_workspace;
4880
4881 /* PCRE supports POSIX class stuff inside a class. Perl gives an error if
4882 they are encountered at the top level, so we'll do that too. */
4883
4884 if ((ptr[1] == CHAR_COLON || ptr[1] == CHAR_DOT ||
4885 ptr[1] == CHAR_EQUALS_SIGN) &&
4886 check_posix_syntax(ptr, &tempptr))
4887 {
4888 *errorcodeptr = (ptr[1] == CHAR_COLON)? ERR13 : ERR31;
4889 goto FAILED;
4890 }
4891
4892 /* If the first character is '^', set the negation flag and skip it. Also,
4893 if the first few characters (either before or after ^) are \Q\E or \E we
4894 skip them too. This makes for compatibility with Perl. */
4895
4896 negate_class = FALSE;
4897 for (;;)
4898 {
4899 c = *(++ptr);
4900 if (c == CHAR_BACKSLASH)
4901 {
4902 if (ptr[1] == CHAR_E)
4903 ptr++;
4904 else if (STRNCMP_UC_C8(ptr + 1, STR_Q STR_BACKSLASH STR_E, 3) == 0)
4905 ptr += 3;
4906 else
4907 break;
4908 }
4909 else if (!negate_class && c == CHAR_CIRCUMFLEX_ACCENT)
4910 negate_class = TRUE;
4911 else break;
4912 }
4913
4914 /* Empty classes are allowed in JavaScript compatibility mode. Otherwise,
4915 an initial ']' is taken as a data character -- the code below handles
4916 that. In JS mode, [] must always fail, so generate OP_FAIL, whereas
4917 [^] must match any character, so generate OP_ALLANY. */
4918
4919 if (c == CHAR_RIGHT_SQUARE_BRACKET &&
4920 (cd->external_options & PCRE_JAVASCRIPT_COMPAT) != 0)
4921 {
4922 *code++ = negate_class? OP_ALLANY : OP_FAIL;
4923 if (firstcharflags == REQ_UNSET) firstcharflags = REQ_NONE;
4924 zerofirstchar = firstchar;
4925 zerofirstcharflags = firstcharflags;
4926 break;
4927 }
4928
4929 /* If a class contains a negative special such as \S, we need to flip the
4930 negation flag at the end, so that support for characters > 255 works
4931 correctly (they are all included in the class). */
4932
4933 should_flip_negation = FALSE;
4934
4935 /* Extended class (xclass) will be used when characters > 255
4936 might match. */
4937
4938 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
4939 xclass = FALSE;
4940 class_uchardata = code + LINK_SIZE + 2; /* For XCLASS items */
4941 class_uchardata_base = class_uchardata; /* Save the start */
4942 #endif
4943
4944 /* For optimization purposes, we track some properties of the class:
4945 class_has_8bitchar will be non-zero if the class contains at least one <
4946 256 character; class_one_char will be 1 if the class contains just one
4947 character; xclass_has_prop will be TRUE if unicode property checks
4948 are present in the class. */
4949
4950 class_has_8bitchar = 0;
4951 class_one_char = 0;
4952 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
4953 xclass_has_prop = FALSE;
4954 #endif
4955
4956 /* Initialize the 32-char bit map to all zeros. We build the map in a
4957 temporary bit of memory, in case the class contains fewer than two
4958 8-bit characters because in that case the compiled code doesn't use the bit
4959 map. */
4960
4961 memset(classbits, 0, 32 * sizeof(pcre_uint8));
4962
4963 /* Process characters until ] is reached. By writing this as a "do" it
4964 means that an initial ] is taken as a data character. At the start of the
4965 loop, c contains the first byte of the character. */
4966
4967 if (c != CHAR_NULL) do
4968 {
4969 const pcre_uchar *oldptr;
4970
4971 #ifdef SUPPORT_UTF
4972 if (utf && HAS_EXTRALEN(c))
4973 { /* Braces are required because the */
4974 GETCHARLEN(c, ptr, ptr); /* macro generates multiple statements */
4975 }
4976 #endif
4977
4978 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
4979 /* In the pre-compile phase, accumulate the length of any extra
4980 data and reset the pointer. This is so that very large classes that
4981 contain a zillion > 255 characters no longer overwrite the work space
4982 (which is on the stack). We have to remember that there was XCLASS data,
4983 however. */
4984
4985 if (class_uchardata > class_uchardata_base) xclass = TRUE;
4986
4987 if (lengthptr != NULL && class_uchardata > class_uchardata_base)
4988 {
4989 *lengthptr += (int)(class_uchardata - class_uchardata_base);
4990 class_uchardata = class_uchardata_base;
4991 }
4992 #endif
4993
4994 /* Inside \Q...\E everything is literal except \E */
4995
4996 if (inescq)
4997 {
4998 if (c == CHAR_BACKSLASH && ptr[1] == CHAR_E) /* If we are at \E */
4999 {
5000 inescq = FALSE; /* Reset literal state */
5001 ptr++; /* Skip the 'E' */
5002 continue; /* Carry on with next */
5003 }
5004 goto CHECK_RANGE; /* Could be range if \E follows */
5005 }
5006
5007 /* Handle POSIX class names. Perl allows a negation extension of the
5008 form [:^name:]. A square bracket that doesn't match the syntax is
5009 treated as a literal. We also recognize the POSIX constructions
5010 [.ch.] and [=ch=] ("collating elements") and fault them, as Perl
5011 5.6 and 5.8 do. */
5012
5013 if (c == CHAR_LEFT_SQUARE_BRACKET &&
5014 (ptr[1] == CHAR_COLON || ptr[1] == CHAR_DOT ||
5015 ptr[1] == CHAR_EQUALS_SIGN) && check_posix_syntax(ptr, &tempptr))
5016 {
5017 BOOL local_negate = FALSE;
5018 int posix_class, taboffset, tabopt;
5019 register const pcre_uint8 *cbits = cd->cbits;
5020 pcre_uint8 pbits[32];
5021
5022 if (ptr[1] != CHAR_COLON)
5023 {
5024 *errorcodeptr = ERR31;
5025 goto FAILED;
5026 }
5027
5028 ptr += 2;
5029 if (*ptr == CHAR_CIRCUMFLEX_ACCENT)
5030 {
5031 local_negate = TRUE;
5032 should_flip_negation = TRUE; /* Note negative special */
5033 ptr++;
5034 }
5035
5036 posix_class = check_posix_name(ptr, (int)(tempptr - ptr));
5037 if (posix_class < 0)
5038 {
5039 *errorcodeptr = ERR30;
5040 goto FAILED;
5041 }
5042
5043 /* If matching is caseless, upper and lower are converted to
5044 alpha. This relies on the fact that the class table starts with
5045 alpha, lower, upper as the first 3 entries. */
5046
5047 if ((options & PCRE_CASELESS) != 0 && posix_class <= 2)
5048 posix_class = 0;
5049
5050 /* When PCRE_UCP is set, some of the POSIX classes are converted to
5051 different escape sequences that use Unicode properties \p or \P. Others
5052 that are not available via \p or \P generate XCL_PROP/XCL_NOTPROP
5053 directly. */
5054
5055 #ifdef SUPPORT_UCP
5056 if ((options & PCRE_UCP) != 0)
5057 {
5058 unsigned int ptype = 0;
5059 int pc = posix_class + ((local_negate)? POSIX_SUBSIZE/2 : 0);
5060
5061 /* The posix_substitutes table specifies which POSIX classes can be
5062 converted to \p or \P items. */
5063
5064 if (posix_substitutes[pc] != NULL)
5065 {
5066 nestptr = tempptr + 1;
5067 ptr = posix_substitutes[pc] - 1;
5068 continue;
5069 }
5070
5071 /* There are three other classes that generate special property calls
5072 that are recognized only in an XCLASS. */
5073
5074 else switch(posix_class)
5075 {
5076 case PC_GRAPH:
5077 ptype = PT_PXGRAPH;
5078 /* Fall through */
5079 case PC_PRINT:
5080 if (ptype == 0) ptype = PT_PXPRINT;
5081 /* Fall through */
5082 case PC_PUNCT:
5083 if (ptype == 0) ptype = PT_PXPUNCT;
5084 *class_uchardata++ = local_negate? XCL_NOTPROP : XCL_PROP;
5085 *class_uchardata++ = ptype;
5086 *class_uchardata++ = 0;
5087 xclass_has_prop = TRUE;
5088 ptr = tempptr + 1;
5089 continue;
5090
5091 /* For the other POSIX classes (ascii, cntrl, xdigit) we are going
5092 to fall through to the non-UCP case and build a bit map for
5093 characters with code points less than 256. If we are in a negated
5094 POSIX class, characters with code points greater than 255 must
5095 either all match or all not match. In the special case where we
5096 have not yet generated any xclass data, and this is the final item
5097 in the overall class, we need do nothing: later on, the opcode
5098 OP_NCLASS will be used to indicate that characters greater than 255
5099 are acceptable. If we have already seen an xclass item or one may
5100 follow (we have to assume that it might if this is not the end of
5101 the class), explicitly list all wide codepoints, which will then
5102 either not match or match, depending on whether the class is or is
5103 not negated. */
5104
5105 default:
5106 if (local_negate &&
5107 (xclass || tempptr[2] != CHAR_RIGHT_SQUARE_BRACKET))
5108 {
5109 *class_uchardata++ = XCL_RANGE;
5110 class_uchardata += PRIV(ord2utf)(0x100, class_uchardata);
5111 class_uchardata += PRIV(ord2utf)(0x10ffff, class_uchardata);
5112 }
5113 break;
5114 }
5115 }
5116 #endif
5117 /* In the non-UCP case, or when UCP makes no difference, we build the
5118 bit map for the POSIX class in a chunk of local store because we may be
5119 adding and subtracting from it, and we don't want to subtract bits that
5120 may be in the main map already. At the end we or the result into the
5121 bit map that is being built. */
5122
5123 posix_class *= 3;
5124
5125 /* Copy in the first table (always present) */
5126
5127 memcpy(pbits, cbits + posix_class_maps[posix_class],
5128 32 * sizeof(pcre_uint8));
5129
5130 /* If there is a second table, add or remove it as required. */
5131
5132 taboffset = posix_class_maps[posix_class + 1];
5133 tabopt = posix_class_maps[posix_class + 2];
5134
5135 if (taboffset >= 0)
5136 {
5137 if (tabopt >= 0)
5138 for (c = 0; c < 32; c++) pbits[c] |= cbits[c + taboffset];
5139 else
5140 for (c = 0; c < 32; c++) pbits[c] &= ~cbits[c + taboffset];
5141 }
5142
5143 /* Now see if we need to remove any special characters. An option
5144 value of 1 removes vertical space and 2 removes underscore. */
5145
5146 if (tabopt < 0) tabopt = -tabopt;
5147 if (tabopt == 1) pbits[1] &= ~0x3c;
5148 else if (tabopt == 2) pbits[11] &= 0x7f;
5149
5150 /* Add the POSIX table or its complement into the main table that is
5151 being built and we are done. */
5152
5153 if (local_negate)
5154 for (c = 0; c < 32; c++) classbits[c] |= ~pbits[c];
5155 else
5156 for (c = 0; c < 32; c++) classbits[c] |= pbits[c];
5157
5158 ptr = tempptr + 1;
5159 /* Every class contains at least one < 256 character. */
5160 class_has_8bitchar = 1;
5161 /* Every class contains at least two characters. */
5162 class_one_char = 2;
5163 continue; /* End of POSIX syntax handling */
5164 }
5165
5166 /* Backslash may introduce a single character, or it may introduce one
5167 of the specials, which just set a flag. The sequence \b is a special
5168 case. Inside a class (and only there) it is treated as backspace. We
5169 assume that other escapes have more than one character in them, so
5170 speculatively set both class_has_8bitchar and class_one_char bigger
5171 than one. Unrecognized escapes fall through and are either treated
5172 as literal characters (by default), or are faulted if
5173 PCRE_EXTRA is set. */
5174
5175 if (c == CHAR_BACKSLASH)
5176 {
5177 escape = check_escape(&ptr, &ec, errorcodeptr, cd->bracount, options,
5178 TRUE);
5179 if (*errorcodeptr != 0) goto FAILED;
5180 if (escape == 0) c = ec;
5181 else if (escape == ESC_b) c = CHAR_BS; /* \b is backspace in a class */
5182 else if (escape == ESC_N) /* \N is not supported in a class */
5183 {
5184 *errorcodeptr = ERR71;
5185 goto FAILED;
5186 }
5187 else if (escape == ESC_Q) /* Handle start of quoted string */
5188 {
5189 if (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
5190 {
5191 ptr += 2; /* avoid empty string */
5192 }
5193 else inescq = TRUE;
5194 continue;
5195 }
5196 else if (escape == ESC_E) continue; /* Ignore orphan \E */
5197
5198 else
5199 {
5200 register const pcre_uint8 *cbits = cd->cbits;
5201 /* Every class contains at least two < 256 characters. */
5202 class_has_8bitchar++;
5203 /* Every class contains at least two characters. */
5204 class_one_char += 2;
5205
5206 switch (escape)
5207 {
5208 #ifdef SUPPORT_UCP
5209 case ESC_du: /* These are the values given for \d etc */
5210 case ESC_DU: /* when PCRE_UCP is set. We replace the */
5211 case ESC_wu: /* escape sequence with an appropriate \p */
5212 case ESC_WU: /* or \P to test Unicode properties instead */
5213 case ESC_su: /* of the default ASCII testing. */
5214 case ESC_SU:
5215 nestptr = ptr;
5216 ptr = substitutes[escape - ESC_DU] - 1; /* Just before substitute */
5217 class_has_8bitchar--; /* Undo! */
5218 continue;
5219 #endif
5220 case ESC_d:
5221 for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_digit];
5222 continue;
5223
5224 case ESC_D:
5225 should_flip_negation = TRUE;
5226 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_digit];
5227 continue;
5228
5229 case ESC_w:
5230 for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_word];
5231 continue;
5232
5233 case ESC_W:
5234 should_flip_negation = TRUE;
5235 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_word];
5236 continue;
5237
5238 /* Perl 5.004 onwards omitted VT from \s, but restored it at Perl
5239 5.18. Before PCRE 8.34, we had to preserve the VT bit if it was
5240 previously set by something earlier in the character class.
5241 Luckily, the value of CHAR_VT is 0x0b in both ASCII and EBCDIC, so
5242 we could just adjust the appropriate bit. From PCRE 8.34 we no
5243 longer treat \s and \S specially. */
5244
5245 case ESC_s:
5246 for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_space];
5247 continue;
5248
5249 case ESC_S:
5250 should_flip_negation = TRUE;
5251 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_space];
5252 continue;
5253
5254 /* The rest apply in both UCP and non-UCP cases. */
5255
5256 case ESC_h:
5257 (void)add_list_to_class(classbits, &class_uchardata, options, cd,
5258 PRIV(hspace_list), NOTACHAR);
5259 continue;
5260
5261 case ESC_H:
5262 (void)add_not_list_to_class(classbits, &class_uchardata, options,
5263 cd, PRIV(hspace_list));
5264 continue;
5265
5266 case ESC_v:
5267 (void)add_list_to_class(classbits, &class_uchardata, options, cd,
5268 PRIV(vspace_list), NOTACHAR);
5269 continue;
5270
5271 case ESC_V:
5272 (void)add_not_list_to_class(classbits, &class_uchardata, options,
5273 cd, PRIV(vspace_list));
5274 continue;
5275
5276 case ESC_p:
5277 case ESC_P:
5278 #ifdef SUPPORT_UCP
5279 {
5280 BOOL negated;
5281 unsigned int ptype = 0, pdata = 0;
5282 if (!get_ucp(&ptr, &negated, &ptype, &pdata, errorcodeptr))
5283 goto FAILED;
5284 *class_uchardata++ = ((escape == ESC_p) != negated)?
5285 XCL_PROP : XCL_NOTPROP;
5286 *class_uchardata++ = ptype;
5287 *class_uchardata++ = pdata;
5288 xclass_has_prop = TRUE;
5289 class_has_8bitchar--; /* Undo! */
5290 continue;
5291 }
5292 #else
5293 *errorcodeptr = ERR45;
5294 goto FAILED;
5295 #endif
5296 /* Unrecognized escapes are faulted if PCRE is running in its
5297 strict mode. By default, for compatibility with Perl, they are
5298 treated as literals. */
5299
5300 default:
5301 if ((options & PCRE_EXTRA) != 0)
5302 {
5303 *errorcodeptr = ERR7;
5304 goto FAILED;
5305 }
5306 class_has_8bitchar--; /* Undo the speculative increase. */
5307 class_one_char -= 2; /* Undo the speculative increase. */
5308 c = *ptr; /* Get the final character and fall through */
5309 break;
5310 }
5311 }
5312
5313 /* Fall through if the escape just defined a single character (c >= 0).
5314 This may be greater than 256. */
5315
5316 escape = 0;
5317
5318 } /* End of backslash handling */
5319
5320 /* A character may be followed by '-' to form a range. However, Perl does
5321 not permit ']' to be the end of the range. A '-' character at the end is
5322 treated as a literal. Perl ignores orphaned \E sequences entirely. The
5323 code for handling \Q and \E is messy. */
5324
5325 CHECK_RANGE:
5326 while (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
5327 {
5328 inescq = FALSE;
5329 ptr += 2;
5330 }
5331 oldptr = ptr;
5332
5333 /* Remember if \r or \n were explicitly used */
5334
5335 if (c == CHAR_CR || c == CHAR_NL) cd->external_flags |= PCRE_HASCRORLF;
5336
5337 /* Check for range */
5338
5339 if (!inescq && ptr[1] == CHAR_MINUS)
5340 {
5341 pcre_uint32 d;
5342 ptr += 2;
5343 while (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_E) ptr += 2;
5344
5345 /* If we hit \Q (not followed by \E) at this point, go into escaped
5346 mode. */
5347
5348 while (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_Q)
5349 {
5350 ptr += 2;
5351 if (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_E)
5352 { ptr += 2; continue; }
5353 inescq = TRUE;
5354 break;
5355 }
5356
5357 /* Minus (hyphen) at the end of a class is treated as a literal, so put
5358 back the pointer and jump to handle the character that preceded it. */
5359
5360 if (*ptr == CHAR_NULL || (!inescq && *ptr == CHAR_RIGHT_SQUARE_BRACKET))
5361 {
5362 ptr = oldptr;
5363 goto CLASS_SINGLE_CHARACTER;
5364 }
5365
5366 /* Otherwise, we have a potential range; pick up the next character */
5367
5368 #ifdef SUPPORT_UTF
5369 if (utf)
5370 { /* Braces are required because the */
5371 GETCHARLEN(d, ptr, ptr); /* macro generates multiple statements */
5372 }
5373 else
5374 #endif
5375 d = *ptr; /* Not UTF-8 mode */
5376
5377 /* The second part of a range can be a single-character escape
5378 sequence, but not any of the other escapes. Perl treats a hyphen as a
5379 literal in such circumstances. However, in Perl's warning mode, a
5380 warning is given, so PCRE now faults it as it is almost certainly a
5381 mistake on the user's part. */
5382
5383 if (!inescq)
5384 {
5385 if (d == CHAR_BACKSLASH)
5386 {
5387 int descape;
5388 descape = check_escape(&ptr, &d, errorcodeptr, cd->bracount, options, TRUE);
5389 if (*errorcodeptr != 0) goto FAILED;
5390
5391 /* 0 means a character was put into d; \b is backspace; any other
5392 special causes an error. */
5393
5394 if (descape != 0)
5395 {
5396 if (descape == ESC_b) d = CHAR_BS; else
5397 {
5398 *errorcodeptr = ERR83;
5399 goto FAILED;
5400 }
5401 }
5402 }
5403
5404 /* A hyphen followed by a POSIX class is treated in the same way. */
5405
5406 else if (d == CHAR_LEFT_SQUARE_BRACKET &&
5407 (ptr[1] == CHAR_COLON || ptr[1] == CHAR_DOT ||
5408 ptr[1] == CHAR_EQUALS_SIGN) &&
5409 check_posix_syntax(ptr, &tempptr))
5410 {
5411 *errorcodeptr = ERR83;
5412 goto FAILED;
5413 }
5414 }
5415
5416 /* Check that the two values are in the correct order. Optimize
5417 one-character ranges. */
5418
5419 if (d < c)
5420 {
5421 *errorcodeptr = ERR8;
5422 goto FAILED;
5423 }
5424 if (d == c) goto CLASS_SINGLE_CHARACTER; /* A few lines below */
5425
5426 /* We have found a character range, so single character optimizations
5427 cannot be done anymore. Any value greater than 1 indicates that there
5428 is more than one character. */
5429
5430 class_one_char = 2;
5431
5432 /* Remember an explicit \r or \n, and add the range to the class. */
5433
5434 if (d == CHAR_CR || d == CHAR_NL) cd->external_flags |= PCRE_HASCRORLF;
5435
5436 class_has_8bitchar +=
5437 add_to_class(classbits, &class_uchardata, options, cd, c, d);
5438
5439 continue; /* Go get the next char in the class */
5440 }
5441
5442 /* Handle a single character - we can get here for a normal non-escape
5443 char, or after \ that introduces a single character or for an apparent
5444 range that isn't. Only the value 1 matters for class_one_char, so don't
5445 increase it if it is already 2 or more ... just in case there's a class
5446 with a zillion characters in it. */
5447
5448 CLASS_SINGLE_CHARACTER:
5449 if (class_one_char < 2) class_one_char++;
5450
5451 /* If xclass_has_prop is false and class_one_char is 1, we have the first
5452 single character in the class, and there have been no prior ranges, or
5453 XCLASS items generated by escapes. If this is the final character in the
5454 class, we can optimize by turning the item into a 1-character OP_CHAR[I]
5455 if it's positive, or OP_NOT[I] if it's negative. In the positive case, it
5456 can cause firstchar to be set. Otherwise, there can be no first char if
5457 this item is first, whatever repeat count may follow. In the case of
5458 reqchar, save the previous value for reinstating. */
5459
5460 if (!inescq &&
5461 #ifdef SUPPORT_UCP
5462 !xclass_has_prop &&
5463 #endif
5464 class_one_char == 1 && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET)
5465 {
5466 ptr++;
5467 zeroreqchar = reqchar;
5468 zeroreqcharflags = reqcharflags;
5469
5470 if (negate_class)
5471 {
5472 #ifdef SUPPORT_UCP
5473 int d;
5474 #endif
5475 if (firstcharflags == REQ_UNSET) firstcharflags = REQ_NONE;
5476 zerofirstchar = firstchar;
5477 zerofirstcharflags = firstcharflags;
5478
5479 /* For caseless UTF-8 mode when UCP support is available, check
5480 whether this character has more than one other case. If so, generate
5481 a special OP_NOTPROP item instead of OP_NOTI. */
5482
5483 #ifdef SUPPORT_UCP
5484 if (utf && (options & PCRE_CASELESS) != 0 &&
5485 (d = UCD_CASESET(c)) != 0)
5486 {
5487 *code++ = OP_NOTPROP;
5488 *code++ = PT_CLIST;
5489 *code++ = d;
5490 }
5491 else
5492 #endif
5493 /* Char has only one other case, or UCP not available */
5494
5495 {
5496 *code++ = ((options & PCRE_CASELESS) != 0)? OP_NOTI: OP_NOT;
5497 #if defined SUPPORT_UTF && !defined COMPILE_PCRE32
5498 if (utf && c > MAX_VALUE_FOR_SINGLE_CHAR)
5499 code += PRIV(ord2utf)(c, code);
5500 else
5501 #endif
5502 *code++ = c;
5503 }
5504
5505 /* We are finished with this character class */
5506
5507 goto END_CLASS;
5508 }
5509
5510 /* For a single, positive character, get the value into mcbuffer, and
5511 then we can handle this with the normal one-character code. */
5512
5513 #if defined SUPPORT_UTF && !defined COMPILE_PCRE32
5514 if (utf && c > MAX_VALUE_FOR_SINGLE_CHAR)
5515 mclength = PRIV(ord2utf)(c, mcbuffer);
5516 else
5517 #endif
5518 {
5519 mcbuffer[0] = c;
5520 mclength = 1;
5521 }
5522 goto ONE_CHAR;
5523 } /* End of 1-char optimization */
5524
5525 /* There is more than one character in the class, or an XCLASS item
5526 has been generated. Add this character to the class. */
5527
5528 class_has_8bitchar +=
5529 add_to_class(classbits, &class_uchardata, options, cd, c, c);
5530 }
5531
5532 /* Loop until ']' reached. This "while" is the end of the "do" far above.
5533 If we are at the end of an internal nested string, revert to the outer
5534 string. */
5535
5536 while (((c = *(++ptr)) != CHAR_NULL ||
5537 (nestptr != NULL &&
5538 (ptr = nestptr, nestptr = NULL, c = *(++ptr)) != CHAR_NULL)) &&
5539 (c != CHAR_RIGHT_SQUARE_BRACKET || inescq));
5540
5541 /* Check for missing terminating ']' */
5542
5543 if (c == CHAR_NULL)
5544 {
5545 *errorcodeptr = ERR6;
5546 goto FAILED;
5547 }
5548
5549 /* We will need an XCLASS if data has been placed in class_uchardata. In
5550 the second phase this is a sufficient test. However, in the pre-compile
5551 phase, class_uchardata gets emptied to prevent workspace overflow, so it
5552 only if the very last character in the class needs XCLASS will it contain
5553 anything at this point. For this reason, xclass gets set TRUE above when
5554 uchar_classdata is emptied, and that's why this code is the way it is here
5555 instead of just doing a test on class_uchardata below. */
5556
5557 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
5558 if (class_uchardata > class_uchardata_base) xclass = TRUE;
5559 #endif
5560
5561 /* If this is the first thing in the branch, there can be no first char
5562 setting, whatever the repeat count. Any reqchar setting must remain
5563 unchanged after any kind of repeat. */
5564
5565 if (firstcharflags == REQ_UNSET) firstcharflags = REQ_NONE;
5566 zerofirstchar = firstchar;
5567 zerofirstcharflags = firstcharflags;
5568 zeroreqchar = reqchar;
5569 zeroreqcharflags = reqcharflags;
5570
5571 /* If there are characters with values > 255, we have to compile an
5572 extended class, with its own opcode, unless there was a negated special
5573 such as \S in the class, and PCRE_UCP is not set, because in that case all
5574 characters > 255 are in the class, so any that were explicitly given as
5575 well can be ignored. If (when there are explicit characters > 255 that must
5576 be listed) there are no characters < 256, we can omit the bitmap in the
5577 actual compiled code. */
5578
5579 #ifdef SUPPORT_UTF
5580 if (xclass && (xclass_has_prop || !should_flip_negation ||
5581 (options & PCRE_UCP) != 0))
5582 #elif !defined COMPILE_PCRE8
5583 if (xclass && (xclass_has_prop || !should_flip_negation))
5584 #endif
5585 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
5586 {
5587 /* For non-UCP wide characters, in a non-negative class containing \S or
5588 similar (should_flip_negation is set), all characters greater than 255
5589 must be in the class. */
5590
5591 if (
5592 #if defined COMPILE_PCRE8
5593 utf &&
5594 #endif
5595 should_flip_negation && !negate_class && (options & PCRE_UCP) == 0)
5596 {
5597 *class_uchardata++ = XCL_RANGE;
5598 if (utf) /* Will always be utf in the 8-bit library */
5599 {
5600 class_uchardata += PRIV(ord2utf)(0x100, class_uchardata);
5601 class_uchardata += PRIV(ord2utf)(0x10ffff, class_uchardata);
5602 }
5603 else /* Can only happen for the 16-bit & 32-bit libraries */
5604 {
5605 #if defined COMPILE_PCRE16
5606 *class_uchardata++ = 0x100;
5607 *class_uchardata++ = 0xffffu;
5608 #elif defined COMPILE_PCRE32
5609 *class_uchardata++ = 0x100;
5610 *class_uchardata++ = 0xffffffffu;
5611 #endif
5612 }
5613 }
5614
5615 *class_uchardata++ = XCL_END; /* Marks the end of extra data */
5616 *code++ = OP_XCLASS;
5617 code += LINK_SIZE;
5618 *code = negate_class? XCL_NOT:0;
5619 if (xclass_has_prop) *code |= XCL_HASPROP;
5620
5621 /* If the map is required, move up the extra data to make room for it;
5622 otherwise just move the code pointer to the end of the extra data. */
5623
5624 if (class_has_8bitchar > 0)
5625 {
5626 *code++ |= XCL_MAP;
5627 memmove(code + (32 / sizeof(pcre_uchar)), code,
5628 IN_UCHARS(class_uchardata - code));
5629 if (negate_class && !xclass_has_prop)
5630 for (c = 0; c < 32; c++) classbits[c] = ~classbits[c];
5631 memcpy(code, classbits, 32);
5632 code = class_uchardata + (32 / sizeof(pcre_uchar));
5633 }
5634 else code = class_uchardata;
5635
5636 /* Now fill in the complete length of the item */
5637
5638 PUT(previous, 1, (int)(code - previous));
5639 break; /* End of class handling */
5640 }
5641
5642 /* Even though any XCLASS list is now discarded, we must allow for
5643 its memory. */
5644
5645 if (lengthptr != NULL)
5646 *lengthptr += (int)(class_uchardata - class_uchardata_base);
5647 #endif
5648
5649 /* If there are no characters > 255, or they are all to be included or
5650 excluded, set the opcode to OP_CLASS or OP_NCLASS, depending on whether the
5651 whole class was negated and whether there were negative specials such as \S
5652 (non-UCP) in the class. Then copy the 32-byte map into the code vector,
5653 negating it if necessary. */
5654
5655 *code++ = (negate_class == should_flip_negation) ? OP_CLASS : OP_NCLASS;
5656 if (lengthptr == NULL) /* Save time in the pre-compile phase */
5657 {
5658 if (negate_class)
5659 for (c = 0; c < 32; c++) classbits[c] = ~classbits[c];
5660 memcpy(code, classbits, 32);
5661 }
5662 code += 32 / sizeof(pcre_uchar);
5663
5664 END_CLASS:
5665 break;
5666
5667
5668 /* ===================================================================*/
5669 /* Various kinds of repeat; '{' is not necessarily a quantifier, but this
5670 has been tested above. */
5671
5672 case CHAR_LEFT_CURLY_BRACKET:
5673 if (!is_quantifier) goto NORMAL_CHAR;
5674 ptr = read_repeat_counts(ptr+1, &repeat_min, &repeat_max, errorcodeptr);
5675 if (*errorcodeptr != 0) goto FAILED;
5676 goto REPEAT;
5677
5678 case CHAR_ASTERISK:
5679 repeat_min = 0;
5680 repeat_max = -1;
5681 goto REPEAT;
5682
5683 case CHAR_PLUS:
5684 repeat_min = 1;
5685 repeat_max = -1;
5686 goto REPEAT;
5687
5688 case CHAR_QUESTION_MARK:
5689 repeat_min = 0;
5690 repeat_max = 1;
5691
5692 REPEAT:
5693 if (previous == NULL)
5694 {
5695 *errorcodeptr = ERR9;
5696 goto FAILED;
5697 }
5698
5699 if (repeat_min == 0)
5700 {
5701 firstchar = zerofirstchar; /* Adjust for zero repeat */
5702 firstcharflags = zerofirstcharflags;
5703 reqchar = zeroreqchar; /* Ditto */
5704 reqcharflags = zeroreqcharflags;
5705 }
5706
5707 /* Remember whether this is a variable length repeat */
5708
5709 reqvary = (repeat_min == repeat_max)? 0 : REQ_VARY;
5710
5711 op_type = 0; /* Default single-char op codes */
5712 possessive_quantifier = FALSE; /* Default not possessive quantifier */
5713
5714 /* Save start of previous item, in case we have to move it up in order to
5715 insert something before it. */
5716
5717 tempcode = previous;
5718
5719 /* Before checking for a possessive quantifier, we must skip over
5720 whitespace and comments in extended mode because Perl allows white space at
5721 this point. */
5722
5723 if ((options & PCRE_EXTENDED) != 0)
5724 {
5725 const pcre_uchar *p = ptr + 1;
5726 for (;;)
5727 {
5728 while (MAX_255(*p) && (cd->ctypes[*p] & ctype_space) != 0) p++;
5729 if (*p != CHAR_NUMBER_SIGN) break;
5730 p++;
5731 while (*p != CHAR_NULL)
5732 {
5733 if (IS_NEWLINE(p)) /* For non-fixed-length newline cases, */
5734 { /* IS_NEWLINE sets cd->nllen. */
5735 p += cd->nllen;
5736 break;
5737 }
5738 p++;
5739 #ifdef SUPPORT_UTF
5740 if (utf) FORWARDCHAR(p);
5741 #endif
5742 } /* Loop for comment characters */
5743 } /* Loop for multiple comments */
5744 ptr = p - 1; /* Character before the next significant one. */
5745 }
5746
5747 /* We also need to skip over (?# comments, which are not dependent on
5748 extended mode. */
5749
5750 if (ptr[1] == CHAR_LEFT_PARENTHESIS && ptr[2] == CHAR_QUESTION_MARK &&
5751 ptr[3] == CHAR_NUMBER_SIGN)
5752 {
5753 ptr += 4;
5754 while (*ptr != CHAR_NULL && *ptr != CHAR_RIGHT_PARENTHESIS) ptr++;
5755 if (*ptr == CHAR_NULL)
5756 {
5757 *errorcodeptr = ERR18;
5758 goto FAILED;
5759 }
5760 }
5761
5762 /* If the next character is '+', we have a possessive quantifier. This
5763 implies greediness, whatever the setting of the PCRE_UNGREEDY option.
5764 If the next character is '?' this is a minimizing repeat, by default,
5765 but if PCRE_UNGREEDY is set, it works the other way round. We change the
5766 repeat type to the non-default. */
5767
5768 if (ptr[1] == CHAR_PLUS)
5769 {
5770 repeat_type = 0; /* Force greedy */
5771 possessive_quantifier = TRUE;
5772 ptr++;
5773 }
5774 else if (ptr[1] == CHAR_QUESTION_MARK)
5775 {
5776 repeat_type = greedy_non_default;
5777 ptr++;
5778 }
5779 else repeat_type = greedy_default;
5780
5781 /* If previous was a recursion call, wrap it in atomic brackets so that
5782 previous becomes the atomic group. All recursions were so wrapped in the
5783 past, but it no longer happens for non-repeated recursions. In fact, the
5784 repeated ones could be re-implemented independently so as not to need this,
5785 but for the moment we rely on the code for repeating groups. */
5786
5787 if (*previous == OP_RECURSE)
5788 {
5789 memmove(previous + 1 + LINK_SIZE, previous, IN_UCHARS(1 + LINK_SIZE));
5790 *previous = OP_ONCE;
5791 PUT(previous, 1, 2 + 2*LINK_SIZE);
5792 previous[2 + 2*LINK_SIZE] = OP_KET;
5793 PUT(previous, 3 + 2*LINK_SIZE, 2 + 2*LINK_SIZE);
5794 code += 2 + 2 * LINK_SIZE;
5795 length_prevgroup = 3 + 3*LINK_SIZE;
5796
5797 /* When actually compiling, we need to check whether this was a forward
5798 reference, and if so, adjust the offset. */
5799
5800 if (lengthptr == NULL && cd->hwm >= cd->start_workspace + LINK_SIZE)
5801 {
5802 int offset = GET(cd->hwm, -LINK_SIZE);
5803 if (offset == previous + 1 - cd->start_code)
5804 PUT(cd->hwm, -LINK_SIZE, offset + 1 + LINK_SIZE);
5805 }
5806 }
5807
5808 /* Now handle repetition for the different types of item. */
5809
5810 /* If previous was a character or negated character match, abolish the item
5811 and generate a repeat item instead. If a char item has a minimum of more
5812 than one, ensure that it is set in reqchar - it might not be if a sequence
5813 such as x{3} is the first thing in a branch because the x will have gone
5814 into firstchar instead. */
5815
5816 if (*previous == OP_CHAR || *previous == OP_CHARI
5817 || *previous == OP_NOT || *previous == OP_NOTI)
5818 {
5819 switch (*previous)
5820 {
5821 default: /* Make compiler happy. */
5822 case OP_CHAR: op_type = OP_STAR - OP_STAR; break;
5823 case OP_CHARI: op_type = OP_STARI - OP_STAR; break;
5824 case OP_NOT: op_type = OP_NOTSTAR - OP_STAR; break;
5825 case OP_NOTI: op_type = OP_NOTSTARI - OP_STAR; break;
5826 }
5827
5828 /* Deal with UTF characters that take up more than one character. It's
5829 easier to write this out separately than try to macrify it. Use c to
5830 hold the length of the character in bytes, plus UTF_LENGTH to flag that
5831 it's a length rather than a small character. */
5832
5833 #if defined SUPPORT_UTF && !defined COMPILE_PCRE32
5834 if (utf && NOT_FIRSTCHAR(code[-1]))
5835 {
5836 pcre_uchar *lastchar = code - 1;
5837 BACKCHAR(lastchar);
5838 c = (int)(code - lastchar); /* Length of UTF-8 character */
5839 memcpy(utf_chars, lastchar, IN_UCHARS(c)); /* Save the char */
5840 c |= UTF_LENGTH; /* Flag c as a length */
5841 }
5842 else
5843 #endif /* SUPPORT_UTF */
5844
5845 /* Handle the case of a single charater - either with no UTF support, or
5846 with UTF disabled, or for a single character UTF character. */
5847 {
5848 c = code[-1];
5849 if (*previous <= OP_CHARI && repeat_min > 1)
5850 {
5851 reqchar = c;
5852 reqcharflags = req_caseopt | cd->req_varyopt;
5853 }
5854 }
5855
5856 goto OUTPUT_SINGLE_REPEAT; /* Code shared with single character types */
5857 }
5858
5859 /* If previous was a character type match (\d or similar), abolish it and
5860 create a suitable repeat item. The code is shared with single-character
5861 repeats by setting op_type to add a suitable offset into repeat_type. Note
5862 the the Unicode property types will be present only when SUPPORT_UCP is
5863 defined, but we don't wrap the little bits of code here because it just
5864 makes it horribly messy. */
5865
5866 else if (*previous < OP_EODN)
5867 {
5868 pcre_uchar *oldcode;
5869 int prop_type, prop_value;
5870 op_type = OP_TYPESTAR - OP_STAR; /* Use type opcodes */
5871 c = *previous;
5872
5873 OUTPUT_SINGLE_REPEAT:
5874 if (*previous == OP_PROP || *previous == OP_NOTPROP)
5875 {
5876 prop_type = previous[1];
5877 prop_value = previous[2];
5878 }
5879 else prop_type = prop_value = -1;
5880
5881 oldcode = code;
5882 code = previous; /* Usually overwrite previous item */
5883
5884 /* If the maximum is zero then the minimum must also be zero; Perl allows
5885 this case, so we do too - by simply omitting the item altogether. */
5886
5887 if (repeat_max == 0) goto END_REPEAT;
5888
5889 /* Combine the op_type with the repeat_type */
5890
5891 repeat_type += op_type;
5892
5893 /* A minimum of zero is handled either as the special case * or ?, or as
5894 an UPTO, with the maximum given. */
5895
5896 if (repeat_min == 0)
5897 {
5898 if (repeat_max == -1) *code++ = OP_STAR + repeat_type;
5899 else if (repeat_max == 1) *code++ = OP_QUERY + repeat_type;
5900 else
5901 {
5902 *code++ = OP_UPTO + repeat_type;
5903 PUT2INC(code, 0, repeat_max);
5904 }
5905 }
5906
5907 /* A repeat minimum of 1 is optimized into some special cases. If the
5908 maximum is unlimited, we use OP_PLUS. Otherwise, the original item is
5909 left in place and, if the maximum is greater than 1, we use OP_UPTO with
5910 one less than the maximum. */
5911
5912 else if (repeat_min == 1)
5913 {
5914 if (repeat_max == -1)
5915 *code++ = OP_PLUS + repeat_type;
5916 else
5917 {
5918 code = oldcode; /* leave previous item in place */
5919 if (repeat_max == 1) goto END_REPEAT;
5920 *code++ = OP_UPTO + repeat_type;
5921 PUT2INC(code, 0, repeat_max - 1);
5922 }
5923 }
5924
5925 /* The case {n,n} is just an EXACT, while the general case {n,m} is
5926 handled as an EXACT followed by an UPTO. */
5927
5928 else
5929 {
5930 *code++ = OP_EXACT + op_type; /* NB EXACT doesn't have repeat_type */
5931 PUT2INC(code, 0, repeat_min);
5932
5933 /* If the maximum is unlimited, insert an OP_STAR. Before doing so,
5934 we have to insert the character for the previous code. For a repeated
5935 Unicode property match, there are two extra bytes that define the
5936 required property. In UTF-8 mode, long characters have their length in
5937 c, with the UTF_LENGTH bit as a flag. */
5938
5939 if (repeat_max < 0)
5940 {
5941 #if defined SUPPORT_UTF && !defined COMPILE_PCRE32
5942 if (utf && (c & UTF_LENGTH) != 0)
5943 {
5944 memcpy(code, utf_chars, IN_UCHARS(c & 7));
5945 code += c & 7;
5946 }
5947 else
5948 #endif
5949 {
5950 *code++ = c;
5951 if (prop_type >= 0)
5952 {
5953 *code++ = prop_type;
5954 *code++ = prop_value;
5955 }
5956 }
5957 *code++ = OP_STAR + repeat_type;
5958 }
5959
5960 /* Else insert an UPTO if the max is greater than the min, again
5961 preceded by the character, for the previously inserted code. If the
5962 UPTO is just for 1 instance, we can use QUERY instead. */
5963
5964 else if (repeat_max != repeat_min)
5965 {
5966 #if defined SUPPORT_UTF && !defined COMPILE_PCRE32
5967 if (utf && (c & UTF_LENGTH) != 0)
5968 {
5969 memcpy(code, utf_chars, IN_UCHARS(c & 7));
5970 code += c & 7;
5971 }
5972 else
5973 #endif
5974 *code++ = c;
5975 if (prop_type >= 0)
5976 {
5977 *code++ = prop_type;
5978 *code++ = prop_value;
5979 }
5980 repeat_max -= repeat_min;
5981
5982 if (repeat_max == 1)
5983 {
5984 *code++ = OP_QUERY + repeat_type;
5985 }
5986 else
5987 {
5988 *code++ = OP_UPTO + repeat_type;
5989 PUT2INC(code, 0, repeat_max);
5990 }
5991 }
5992 }
5993
5994 /* The character or character type itself comes last in all cases. */
5995
5996 #if defined SUPPORT_UTF && !defined COMPILE_PCRE32
5997 if (utf && (c & UTF_LENGTH) != 0)
5998 {
5999 memcpy(code, utf_chars, IN_UCHARS(c & 7));
6000 code += c & 7;
6001 }
6002 else
6003 #endif
6004 *code++ = c;
6005
6006 /* For a repeated Unicode property match, there are two extra bytes that
6007 define the required property. */
6008
6009 #ifdef SUPPORT_UCP
6010 if (prop_type >= 0)
6011 {
6012 *code++ = prop_type;
6013 *code++ = prop_value;
6014 }
6015 #endif
6016 }
6017
6018 /* If previous was a character class or a back reference, we put the repeat
6019 stuff after it, but just skip the item if the repeat was {0,0}. */
6020
6021 else if (*previous == OP_CLASS || *previous == OP_NCLASS ||
6022 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
6023 *previous == OP_XCLASS ||
6024 #endif
6025 *previous == OP_REF || *previous == OP_REFI ||
6026 *previous == OP_DNREF || *previous == OP_DNREFI)
6027 {
6028 if (repeat_max == 0)
6029 {
6030 code = previous;
6031 goto END_REPEAT;
6032 }
6033
6034 if (repeat_min == 0 && repeat_max == -1)
6035 *code++ = OP_CRSTAR + repeat_type;
6036 else if (repeat_min == 1 && repeat_max == -1)
6037 *code++ = OP_CRPLUS + repeat_type;
6038 else if (repeat_min == 0 && repeat_max == 1)
6039 *code++ = OP_CRQUERY + repeat_type;
6040 else
6041 {
6042 *code++ = OP_CRRANGE + repeat_type;
6043 PUT2INC(code, 0, repeat_min);
6044 if (repeat_max == -1) repeat_max = 0; /* 2-byte encoding for max */
6045 PUT2INC(code, 0, repeat_max);
6046 }
6047 }
6048
6049 /* If previous was a bracket group, we may have to replicate it in certain
6050 cases. Note that at this point we can encounter only the "basic" bracket
6051 opcodes such as BRA and CBRA, as this is the place where they get converted
6052 into the more special varieties such as BRAPOS and SBRA. A test for >=
6053 OP_ASSERT and <= OP_COND includes ASSERT, ASSERT_NOT, ASSERTBACK,
6054 ASSERTBACK_NOT, ONCE, ONCE_NC, BRA, BRAPOS, CBRA, CBRAPOS, and COND.
6055 Originally, PCRE did not allow repetition of assertions, but now it does,
6056 for Perl compatibility. */
6057
6058 else if (*previous >= OP_ASSERT && *previous <= OP_COND)
6059 {
6060 register int i;
6061 int len = (int)(code - previous);
6062 size_t base_hwm_offset = item_hwm_offset;
6063 pcre_uchar *bralink = NULL;
6064 pcre_uchar *brazeroptr = NULL;
6065
6066 /* Repeating a DEFINE group is pointless, but Perl allows the syntax, so
6067 we just ignore the repeat. */
6068
6069 if (*previous == OP_COND && previous[LINK_SIZE+1] == OP_DEF)
6070 goto END_REPEAT;
6071
6072 /* There is no sense in actually repeating assertions. The only potential
6073 use of repetition is in cases when the assertion is optional. Therefore,
6074 if the minimum is greater than zero, just ignore the repeat. If the
6075 maximum is not zero or one, set it to 1. */
6076
6077 if (*previous < OP_ONCE) /* Assertion */
6078 {
6079 if (repeat_min > 0) goto END_REPEAT;
6080 if (repeat_max < 0 || repeat_max > 1) repeat_max = 1;
6081 }
6082
6083 /* The case of a zero minimum is special because of the need to stick
6084 OP_BRAZERO in front of it, and because the group appears once in the
6085 data, whereas in other cases it appears the minimum number of times. For
6086 this reason, it is simplest to treat this case separately, as otherwise
6087 the code gets far too messy. There are several special subcases when the
6088 minimum is zero. */
6089
6090 if (repeat_min == 0)
6091 {
6092 /* If the maximum is also zero, we used to just omit the group from the
6093 output altogether, like this:
6094
6095 ** if (repeat_max == 0)
6096 ** {
6097 ** code = previous;
6098 ** goto END_REPEAT;
6099 ** }
6100
6101 However, that fails when a group or a subgroup within it is referenced
6102 as a subroutine from elsewhere in the pattern, so now we stick in
6103 OP_SKIPZERO in front of it so that it is skipped on execution. As we
6104 don't have a list of which groups are referenced, we cannot do this
6105 selectively.
6106
6107 If the maximum is 1 or unlimited, we just have to stick in the BRAZERO
6108 and do no more at this point. However, we do need to adjust any
6109 OP_RECURSE calls inside the group that refer to the group itself or any
6110 internal or forward referenced group, because the offset is from the
6111 start of the whole regex. Temporarily terminate the pattern while doing
6112 this. */
6113
6114 if (repeat_max <= 1) /* Covers 0, 1, and unlimited */
6115 {
6116 *code = OP_END;
6117 adjust_recurse(previous, 1, utf, cd, item_hwm_offset);
6118 memmove(previous + 1, previous, IN_UCHARS(len));
6119 code++;
6120 if (repeat_max == 0)
6121 {
6122 *previous++ = OP_SKIPZERO;
6123 goto END_REPEAT;
6124 }
6125 brazeroptr = previous; /* Save for possessive optimizing */
6126 *previous++ = OP_BRAZERO + repeat_type;
6127 }
6128
6129 /* If the maximum is greater than 1 and limited, we have to replicate
6130 in a nested fashion, sticking OP_BRAZERO before each set of brackets.
6131 The first one has to be handled carefully because it's the original
6132 copy, which has to be moved up. The remainder can be handled by code
6133 that is common with the non-zero minimum case below. We have to
6134 adjust the value or repeat_max, since one less copy is required. Once
6135 again, we may have to adjust any OP_RECURSE calls inside the group. */
6136
6137 else
6138 {
6139 int offset;
6140 *code = OP_END;
6141 adjust_recurse(previous, 2 + LINK_SIZE, utf, cd, item_hwm_offset);
6142 memmove(previous + 2 + LINK_SIZE, previous, IN_UCHARS(len));
6143 code += 2 + LINK_SIZE;
6144 *previous++ = OP_BRAZERO + repeat_type;
6145 *previous++ = OP_BRA;
6146
6147 /* We chain together the bracket offset fields that have to be
6148 filled in later when the ends of the brackets are reached. */
6149
6150 offset = (bralink == NULL)? 0 : (int)(previous - bralink);
6151 bralink = previous;
6152 PUTINC(previous, 0, offset);
6153 }
6154
6155 repeat_max--;
6156 }
6157
6158 /* If the minimum is greater than zero, replicate the group as many
6159 times as necessary, and adjust the maximum to the number of subsequent
6160 copies that we need. If we set a first char from the group, and didn't
6161 set a required char, copy the latter from the former. If there are any
6162 forward reference subroutine calls in the group, there will be entries on
6163 the workspace list; replicate these with an appropriate increment. */
6164
6165 else
6166 {
6167 if (repeat_min > 1)
6168 {
6169 /* In the pre-compile phase, we don't actually do the replication. We
6170 just adjust the length as if we had. Do some paranoid checks for
6171 potential integer overflow. The INT64_OR_DOUBLE type is a 64-bit
6172 integer type when available, otherwise double. */
6173
6174 if (lengthptr != NULL)
6175 {
6176 int delta = (repeat_min - 1)*length_prevgroup;
6177 if ((INT64_OR_DOUBLE)(repeat_min - 1)*
6178 (INT64_OR_DOUBLE)length_prevgroup >
6179 (INT64_OR_DOUBLE)INT_MAX ||
6180 OFLOW_MAX - *lengthptr < delta)
6181 {
6182 *errorcodeptr = ERR20;
6183 goto FAILED;
6184 }
6185 *lengthptr += delta;
6186 }
6187
6188 /* This is compiling for real. If there is a set first byte for
6189 the group, and we have not yet set a "required byte", set it. Make
6190 sure there is enough workspace for copying forward references before
6191 doing the copy. */
6192
6193 else
6194 {
6195 if (groupsetfirstchar && reqcharflags < 0)
6196 {
6197 reqchar = firstchar;
6198 reqcharflags = firstcharflags;
6199 }
6200
6201 for (i = 1; i < repeat_min; i++)
6202 {
6203 pcre_uchar *hc;
6204 size_t this_hwm_offset = cd->hwm - cd->start_workspace;
6205 memcpy(code, previous, IN_UCHARS(len));
6206
6207 while (cd->hwm > cd->start_workspace + cd->workspace_size -
6208 WORK_SIZE_SAFETY_MARGIN -
6209 (this_hwm_offset - base_hwm_offset))
6210 {
6211 *errorcodeptr = expand_workspace(cd);
6212 if (*errorcodeptr != 0) goto FAILED;
6213 }
6214
6215 for (hc = (pcre_uchar *)cd->start_workspace + base_hwm_offset;
6216 hc < (pcre_uchar *)cd->start_workspace + this_hwm_offset;
6217 hc += LINK_SIZE)
6218 {
6219 PUT(cd->hwm, 0, GET(hc, 0) + len);
6220 cd->hwm += LINK_SIZE;
6221 }
6222 base_hwm_offset = this_hwm_offset;
6223 code += len;
6224 }
6225 }
6226 }
6227
6228 if (repeat_max > 0) repeat_max -= repeat_min;
6229 }
6230
6231 /* This code is common to both the zero and non-zero minimum cases. If
6232 the maximum is limited, it replicates the group in a nested fashion,
6233 remembering the bracket starts on a stack. In the case of a zero minimum,
6234 the first one was set up above. In all cases the repeat_max now specifies
6235 the number of additional copies needed. Again, we must remember to
6236 replicate entries on the forward reference list. */
6237
6238 if (repeat_max >= 0)
6239 {
6240 /* In the pre-compile phase, we don't actually do the replication. We
6241 just adjust the length as if we had. For each repetition we must add 1
6242 to the length for BRAZERO and for all but the last repetition we must
6243 add 2 + 2*LINKSIZE to allow for the nesting that occurs. Do some
6244 paranoid checks to avoid integer overflow. The INT64_OR_DOUBLE type is
6245 a 64-bit integer type when available, otherwise double. */
6246
6247 if (lengthptr != NULL && repeat_max > 0)
6248 {
6249 int delta = repeat_max * (length_prevgroup + 1 + 2 + 2*LINK_SIZE) -
6250 2 - 2*LINK_SIZE; /* Last one doesn't nest */
6251 if ((INT64_OR_DOUBLE)repeat_max *
6252 (INT64_OR_DOUBLE)(length_prevgroup + 1 + 2 + 2*LINK_SIZE)
6253 > (INT64_OR_DOUBLE)INT_MAX ||
6254 OFLOW_MAX - *lengthptr < delta)
6255 {
6256 *errorcodeptr = ERR20;
6257 goto FAILED;
6258 }
6259 *lengthptr += delta;
6260 }
6261
6262 /* This is compiling for real */
6263
6264 else for (i = repeat_max - 1; i >= 0; i--)
6265 {
6266 pcre_uchar *hc;
6267 size_t this_hwm_offset = cd->hwm - cd->start_workspace;
6268
6269 *code++ = OP_BRAZERO + repeat_type;
6270
6271 /* All but the final copy start a new nesting, maintaining the
6272 chain of brackets outstanding. */
6273
6274 if (i != 0)
6275 {
6276 int offset;
6277 *code++ = OP_BRA;
6278 offset = (bralink == NULL)? 0 : (int)(code - bralink);
6279 bralink = code;
6280 PUTINC(code, 0, offset);
6281 }
6282
6283 memcpy(code, previous, IN_UCHARS(len));
6284
6285 /* Ensure there is enough workspace for forward references before
6286 copying them. */
6287
6288 while (cd->hwm > cd->start_workspace + cd->workspace_size -
6289 WORK_SIZE_SAFETY_MARGIN -
6290 (this_hwm_offset - base_hwm_offset))
6291 {
6292 *errorcodeptr = expand_workspace(cd);
6293 if (*errorcodeptr != 0) goto FAILED;
6294 }
6295
6296 for (hc = (pcre_uchar *)cd->start_workspace + base_hwm_offset;
6297 hc < (pcre_uchar *)cd->start_workspace + this_hwm_offset;
6298 hc += LINK_SIZE)
6299 {
6300 PUT(cd->hwm, 0, GET(hc, 0) + len + ((i != 0)? 2+LINK_SIZE : 1));
6301 cd->hwm += LINK_SIZE;
6302 }
6303 base_hwm_offset = this_hwm_offset;
6304 code += len;
6305 }
6306
6307 /* Now chain through the pending brackets, and fill in their length
6308 fields (which are holding the chain links pro tem). */
6309
6310 while (bralink != NULL)
6311 {
6312 int oldlinkoffset;
6313 int offset = (int)(code - bralink + 1);
6314 pcre_uchar *bra = code - offset;
6315 oldlinkoffset = GET(bra, 1);
6316 bralink = (oldlinkoffset == 0)? NULL : bralink - oldlinkoffset;
6317 *code++ = OP_KET;
6318 PUTINC(code, 0, offset);
6319 PUT(bra, 1, offset);
6320 }
6321 }
6322
6323 /* If the maximum is unlimited, set a repeater in the final copy. For
6324 ONCE brackets, that's all we need to do. However, possessively repeated
6325 ONCE brackets can be converted into non-capturing brackets, as the
6326 behaviour of (?:xx)++ is the same as (?>xx)++ and this saves having to
6327 deal with possessive ONCEs specially.
6328
6329 Otherwise, when we are doing the actual compile phase, check to see
6330 whether this group is one that could match an empty string. If so,
6331 convert the initial operator to the S form (e.g. OP_BRA -> OP_SBRA) so
6332 that runtime checking can be done. [This check is also applied to ONCE
6333 groups at runtime, but in a different way.]
6334
6335 Then, if the quantifier was possessive and the bracket is not a
6336 conditional, we convert the BRA code to the POS form, and the KET code to
6337 KETRPOS. (It turns out to be convenient at runtime to detect this kind of
6338 subpattern at both the start and at the end.) The use of special opcodes
6339 makes it possible to reduce greatly the stack usage in pcre_exec(). If
6340 the group is preceded by OP_BRAZERO, convert this to OP_BRAPOSZERO.
6341
6342 Then, if the minimum number of matches is 1 or 0, cancel the possessive
6343 flag so that the default action below, of wrapping everything inside
6344 atomic brackets, does not happen. When the minimum is greater than 1,
6345 there will be earlier copies of the group, and so we still have to wrap
6346 the whole thing. */
6347
6348 else
6349 {
6350 pcre_uchar *ketcode = code - 1 - LINK_SIZE;
6351 pcre_uchar *bracode = ketcode - GET(ketcode, 1);
6352
6353 /* Convert possessive ONCE brackets to non-capturing */
6354
6355 if ((*bracode == OP_ONCE || *bracode == OP_ONCE_NC) &&
6356 possessive_quantifier) *bracode = OP_BRA;
6357
6358 /* For non-possessive ONCE brackets, all we need to do is to
6359 set the KET. */
6360
6361 if (*bracode == OP_ONCE || *bracode == OP_ONCE_NC)
6362 *ketcode = OP_KETRMAX + repeat_type;
6363
6364 /* Handle non-ONCE brackets and possessive ONCEs (which have been
6365 converted to non-capturing above). */
6366
6367 else
6368 {
6369 /* In the compile phase, check for empty string matching. */
6370
6371 if (lengthptr == NULL)
6372 {
6373 pcre_uchar *scode = bracode;
6374 do
6375 {
6376 if (could_be_empty_branch(scode, ketcode, utf, cd, NULL))
6377 {
6378 *bracode += OP_SBRA - OP_BRA;
6379 break;
6380 }
6381 scode += GET(scode, 1);
6382 }
6383 while (*scode == OP_ALT);
6384 }
6385
6386 /* A conditional group with only one branch has an implicit empty
6387 alternative branch. */
6388
6389 if (*bracode == OP_COND && bracode[GET(bracode,1)] != OP_ALT)
6390 *bracode = OP_SCOND;
6391
6392 /* Handle possessive quantifiers. */
6393
6394 if (possessive_quantifier)
6395 {
6396 /* For COND brackets, we wrap the whole thing in a possessively
6397 repeated non-capturing bracket, because we have not invented POS
6398 versions of the COND opcodes. Because we are moving code along, we
6399 must ensure that any pending recursive references are updated. */
6400
6401 if (*bracode == OP_COND || *bracode == OP_SCOND)
6402 {
6403 int nlen = (int)(code - bracode);
6404 *code = OP_END;
6405 adjust_recurse(bracode, 1 + LINK_SIZE, utf, cd, item_hwm_offset);
6406 memmove(bracode + 1 + LINK_SIZE, bracode, IN_UCHARS(nlen));
6407 code += 1 + LINK_SIZE;
6408 nlen += 1 + LINK_SIZE;
6409 *bracode = (*bracode == OP_COND)? OP_BRAPOS : OP_SBRAPOS;
6410 *code++ = OP_KETRPOS;
6411 PUTINC(code, 0, nlen);
6412 PUT(bracode, 1, nlen);
6413 }
6414
6415 /* For non-COND brackets, we modify the BRA code and use KETRPOS. */
6416
6417 else
6418 {
6419 *bracode += 1; /* Switch to xxxPOS opcodes */
6420 *ketcode = OP_KETRPOS;
6421 }
6422
6423 /* If the minimum is zero, mark it as possessive, then unset the
6424 possessive flag when the minimum is 0 or 1. */
6425
6426 if (brazeroptr != NULL) *brazeroptr = OP_BRAPOSZERO;
6427 if (repeat_min < 2) possessive_quantifier = FALSE;
6428 }
6429
6430 /* Non-possessive quantifier */
6431
6432 else *ketcode = OP_KETRMAX + repeat_type;
6433 }
6434 }
6435 }
6436
6437 /* If previous is OP_FAIL, it was generated by an empty class [] in
6438 JavaScript mode. The other ways in which OP_FAIL can be generated, that is
6439 by (*FAIL) or (?!) set previous to NULL, which gives a "nothing to repeat"
6440 error above. We can just ignore the repeat in JS case. */
6441
6442 else if (*previous == OP_FAIL) goto END_REPEAT;
6443
6444 /* Else there's some kind of shambles */
6445
6446 else
6447 {
6448 *errorcodeptr = ERR11;
6449 goto FAILED;
6450 }
6451
6452 /* If the character following a repeat is '+', possessive_quantifier is
6453 TRUE. For some opcodes, there are special alternative opcodes for this
6454 case. For anything else, we wrap the entire repeated item inside OP_ONCE
6455 brackets. Logically, the '+' notation is just syntactic sugar, taken from
6456 Sun's Java package, but the special opcodes can optimize it.
6457
6458 Some (but not all) possessively repeated subpatterns have already been
6459 completely handled in the code just above. For them, possessive_quantifier
6460 is always FALSE at this stage. Note that the repeated item starts at
6461 tempcode, not at previous, which might be the first part of a string whose
6462 (former) last char we repeated. */
6463
6464 if (possessive_quantifier)
6465 {
6466 int len;
6467
6468 /* Possessifying an EXACT quantifier has no effect, so we can ignore it.
6469 However, QUERY, STAR, or UPTO may follow (for quantifiers such as {5,6},
6470 {5,}, or {5,10}). We skip over an EXACT item; if the length of what
6471 remains is greater than zero, there's a further opcode that can be
6472 handled. If not, do nothing, leaving the EXACT alone. */
6473
6474 switch(*tempcode)
6475 {
6476 case OP_TYPEEXACT:
6477 tempcode += PRIV(OP_lengths)[*tempcode] +
6478 ((tempcode[1 + IMM2_SIZE] == OP_PROP
6479 || tempcode[1 + IMM2_SIZE] == OP_NOTPROP)? 2 : 0);
6480 break;
6481
6482 /* CHAR opcodes are used for exacts whose count is 1. */
6483
6484 case OP_CHAR:
6485 case OP_CHARI:
6486 case OP_NOT:
6487 case OP_NOTI:
6488 case OP_EXACT:
6489 case OP_EXACTI:
6490 case OP_NOTEXACT:
6491 case OP_NOTEXACTI:
6492 tempcode += PRIV(OP_lengths)[*tempcode];
6493 #ifdef SUPPORT_UTF
6494 if (utf && HAS_EXTRALEN(tempcode[-1]))
6495 tempcode += GET_EXTRALEN(tempcode[-1]);
6496 #endif
6497 break;
6498
6499 /* For the class opcodes, the repeat operator appears at the end;
6500 adjust tempcode to point to it. */
6501
6502 case OP_CLASS:
6503 case OP_NCLASS:
6504 tempcode += 1 + 32/sizeof(pcre_uchar);
6505 break;
6506
6507 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
6508 case OP_XCLASS:
6509 tempcode += GET(tempcode, 1);
6510 break;
6511 #endif
6512 }
6513
6514 /* If tempcode is equal to code (which points to the end of the repeated
6515 item), it means we have skipped an EXACT item but there is no following
6516 QUERY, STAR, or UPTO; the value of len will be 0, and we do nothing. In
6517 all other cases, tempcode will be pointing to the repeat opcode, and will
6518 be less than code, so the value of len will be greater than 0. */
6519
6520 len = (int)(code - tempcode);
6521 if (len > 0)
6522 {
6523 unsigned int repcode = *tempcode;
6524
6525 /* There is a table for possessifying opcodes, all of which are less
6526 than OP_CALLOUT. A zero entry means there is no possessified version.
6527 */
6528
6529 if (repcode < OP_CALLOUT && opcode_possessify[repcode] > 0)
6530 *tempcode = opcode_possessify[repcode];
6531
6532 /* For opcode without a special possessified version, wrap the item in
6533 ONCE brackets. Because we are moving code along, we must ensure that any
6534 pending recursive references are updated. */
6535
6536 else
6537 {
6538 *code = OP_END;
6539 adjust_recurse(tempcode, 1 + LINK_SIZE, utf, cd, item_hwm_offset);
6540 memmove(tempcode + 1 + LINK_SIZE, tempcode, IN_UCHARS(len));
6541 code += 1 + LINK_SIZE;
6542 len += 1 + LINK_SIZE;
6543 tempcode[0] = OP_ONCE;
6544 *code++ = OP_KET;
6545 PUTINC(code, 0, len);
6546 PUT(tempcode, 1, len);
6547 }
6548 }
6549
6550 #ifdef NEVER
6551 if (len > 0) switch (*tempcode)
6552 {
6553 case OP_STAR: *tempcode = OP_POSSTAR; break;
6554 case OP_PLUS: *tempcode = OP_POSPLUS; break;
6555 case OP_QUERY: *tempcode = OP_POSQUERY; break;
6556 case OP_UPTO: *tempcode = OP_POSUPTO; break;
6557
6558 case OP_STARI: *tempcode = OP_POSSTARI; break;
6559 case OP_PLUSI: *tempcode = OP_POSPLUSI; break;
6560 case OP_QUERYI: *tempcode = OP_POSQUERYI; break;
6561 case OP_UPTOI: *tempcode = OP_POSUPTOI; break;
6562
6563 case OP_NOTSTAR: *tempcode = OP_NOTPOSSTAR; break;
6564 case OP_NOTPLUS: *tempcode = OP_NOTPOSPLUS; break;
6565 case OP_NOTQUERY: *tempcode = OP_NOTPOSQUERY; break;
6566 case OP_NOTUPTO: *tempcode = OP_NOTPOSUPTO; break;
6567
6568 case OP_NOTSTARI: *tempcode = OP_NOTPOSSTARI; break;
6569 case OP_NOTPLUSI: *tempcode = OP_NOTPOSPLUSI; break;
6570 case OP_NOTQUERYI: *tempcode = OP_NOTPOSQUERYI; break;
6571 case OP_NOTUPTOI: *tempcode = OP_NOTPOSUPTOI; break;
6572
6573 case OP_TYPESTAR: *tempcode = OP_TYPEPOSSTAR; break;
6574 case OP_TYPEPLUS: *tempcode = OP_TYPEPOSPLUS; break;
6575 case OP_TYPEQUERY: *tempcode = OP_TYPEPOSQUERY; break;
6576 case OP_TYPEUPTO: *tempcode = OP_TYPEPOSUPTO; break;
6577
6578 case OP_CRSTAR: *tempcode = OP_CRPOSSTAR; break;
6579 case OP_CRPLUS: *tempcode = OP_CRPOSPLUS; break;
6580 case OP_CRQUERY: *tempcode = OP_CRPOSQUERY; break;
6581 case OP_CRRANGE: *tempcode = OP_CRPOSRANGE; break;
6582
6583 /* Because we are moving code along, we must ensure that any
6584 pending recursive references are updated. */
6585
6586 default:
6587 *code = OP_END;
6588 adjust_recurse(tempcode, 1 + LINK_SIZE, utf, cd, item_hwm_offset);
6589 memmove(tempcode + 1 + LINK_SIZE, tempcode, IN_UCHARS(len));
6590 code += 1 + LINK_SIZE;
6591 len += 1 + LINK_SIZE;
6592 tempcode[0] = OP_ONCE;
6593 *code++ = OP_KET;
6594 PUTINC(code, 0, len);
6595 PUT(tempcode, 1, len);
6596 break;
6597 }
6598 #endif
6599 }
6600
6601 /* In all case we no longer have a previous item. We also set the
6602 "follows varying string" flag for subsequently encountered reqchars if
6603 it isn't already set and we have just passed a varying length item. */
6604
6605 END_REPEAT:
6606 previous = NULL;
6607 cd->req_varyopt |= reqvary;
6608 break;
6609
6610
6611 /* ===================================================================*/
6612 /* Start of nested parenthesized sub-expression, or comment or lookahead or
6613 lookbehind or option setting or condition or all the other extended
6614 parenthesis forms. */
6615
6616 case CHAR_LEFT_PARENTHESIS:
6617 ptr++;
6618
6619 /* Now deal with various "verbs" that can be introduced by '*'. */
6620
6621 if (ptr[0] == CHAR_ASTERISK && (ptr[1] == ':'
6622 || (MAX_255(ptr[1]) && ((cd->ctypes[ptr[1]] & ctype_letter) != 0))))
6623 {
6624 int i, namelen;
6625 int arglen = 0;
6626 const char *vn = verbnames;
6627 const pcre_uchar *name = ptr + 1;
6628 const pcre_uchar *arg = NULL;
6629 previous = NULL;
6630 ptr++;
6631 while (MAX_255(*ptr) && (cd->ctypes[*ptr] & ctype_letter) != 0) ptr++;
6632 namelen = (int)(ptr - name);
6633
6634 /* It appears that Perl allows any characters whatsoever, other than
6635 a closing parenthesis, to appear in arguments, so we no longer insist on
6636 letters, digits, and underscores. */
6637
6638 if (*ptr == CHAR_COLON)
6639 {
6640 arg = ++ptr;
6641 while (*ptr != CHAR_NULL && *ptr != CHAR_RIGHT_PARENTHESIS) ptr++;
6642 arglen = (int)(ptr - arg);
6643 if ((unsigned int)arglen > MAX_MARK)
6644 {
6645 *errorcodeptr = ERR75;
6646 goto FAILED;
6647 }
6648 }
6649
6650 if (*ptr != CHAR_RIGHT_PARENTHESIS)
6651 {
6652 *errorcodeptr = ERR60;
6653 goto FAILED;
6654 }
6655
6656 /* Scan the table of verb names */
6657
6658 for (i = 0; i < verbcount; i++)
6659 {
6660 if (namelen == verbs[i].len &&
6661 STRNCMP_UC_C8(name, vn, namelen) == 0)
6662 {
6663 int setverb;
6664
6665 /* Check for open captures before ACCEPT and convert it to
6666 ASSERT_ACCEPT if in an assertion. */
6667
6668 if (verbs[i].op == OP_ACCEPT)
6669 {
6670 open_capitem *oc;
6671 if (arglen != 0)
6672 {
6673 *errorcodeptr = ERR59;
6674 goto FAILED;
6675 }
6676 cd->had_accept = TRUE;
6677 for (oc = cd->open_caps; oc != NULL; oc = oc->next)
6678 {
6679 if (lengthptr != NULL)
6680 {
6681 #ifdef COMPILE_PCRE8
6682 *lengthptr += 1 + IMM2_SIZE;
6683 #elif defined COMPILE_PCRE16
6684 *lengthptr += 2 + IMM2_SIZE;
6685 #elif defined COMPILE_PCRE32
6686 *lengthptr += 4 + IMM2_SIZE;
6687 #endif
6688 }
6689 else
6690 {
6691 *code++ = OP_CLOSE;
6692 PUT2INC(code, 0, oc->number);
6693 }
6694 }
6695 setverb = *code++ =
6696 (cd->assert_depth > 0)? OP_ASSERT_ACCEPT : OP_ACCEPT;
6697
6698 /* Do not set firstchar after *ACCEPT */
6699 if (firstcharflags == REQ_UNSET) firstcharflags = REQ_NONE;
6700 }
6701
6702 /* Handle other cases with/without an argument */
6703
6704 else if (arglen == 0)
6705 {
6706 if (verbs[i].op < 0) /* Argument is mandatory */
6707 {
6708 *errorcodeptr = ERR66;
6709 goto FAILED;
6710 }
6711 setverb = *code++ = verbs[i].op;
6712 }
6713
6714 else
6715 {
6716 if (verbs[i].op_arg < 0) /* Argument is forbidden */
6717 {
6718 *errorcodeptr = ERR59;
6719 goto FAILED;
6720 }
6721 setverb = *code++ = verbs[i].op_arg;
6722 if (lengthptr != NULL) /* In pass 1 just add in the length */
6723 { /* to avoid potential workspace */
6724 *lengthptr += arglen; /* overflow. */
6725 *code++ = 0;
6726 }
6727 else
6728 {
6729 *code++ = arglen;
6730 memcpy(code, arg, IN_UCHARS(arglen));
6731 code += arglen;
6732 }
6733 *code++ = 0;
6734 }
6735
6736 switch (setverb)
6737 {
6738 case OP_THEN:
6739 case OP_THEN_ARG:
6740 cd->external_flags |= PCRE_HASTHEN;
6741 break;
6742
6743 case OP_PRUNE:
6744 case OP_PRUNE_ARG:
6745 case OP_SKIP:
6746 case OP_SKIP_ARG:
6747 cd->had_pruneorskip = TRUE;
6748 break;
6749 }
6750
6751 break; /* Found verb, exit loop */
6752 }
6753
6754 vn += verbs[i].len + 1;
6755 }
6756
6757 if (i < verbcount) continue; /* Successfully handled a verb */
6758 *errorcodeptr = ERR60; /* Verb not recognized */
6759 goto FAILED;
6760 }
6761
6762 /* Initialize for "real" parentheses */
6763
6764 newoptions = options;
6765 skipbytes = 0;
6766 bravalue = OP_CBRA;
6767 item_hwm_offset = cd->hwm - cd->start_workspace;
6768 reset_bracount = FALSE;
6769
6770 /* Deal with the extended parentheses; all are introduced by '?', and the
6771 appearance of any of them means that this is not a capturing group. */
6772
6773 if (*ptr == CHAR_QUESTION_MARK)
6774 {
6775 int i, set, unset, namelen;
6776 int *optset;
6777 const pcre_uchar *name;
6778 pcre_uchar *slot;
6779
6780 switch (*(++ptr))
6781 {
6782 /* ------------------------------------------------------------ */
6783 case CHAR_VERTICAL_LINE: /* Reset capture count for each branch */
6784 reset_bracount = TRUE;
6785 cd->dupgroups = TRUE; /* Record (?| encountered */
6786 /* Fall through */
6787
6788 /* ------------------------------------------------------------ */
6789 case CHAR_COLON: /* Non-capturing bracket */
6790 bravalue = OP_BRA;
6791 ptr++;
6792 break;
6793
6794
6795 /* ------------------------------------------------------------ */
6796 case CHAR_LEFT_PARENTHESIS:
6797 bravalue = OP_COND; /* Conditional group */
6798 tempptr = ptr;
6799
6800 /* A condition can be an assertion, a number (referring to a numbered
6801 group's having been set), a name (referring to a named group), or 'R',
6802 referring to recursion. R<digits> and R&name are also permitted for
6803 recursion tests.
6804
6805 There are ways of testing a named group: (?(name)) is used by Python;
6806 Perl 5.10 onwards uses (?(<name>) or (?('name')).
6807
6808 There is one unfortunate ambiguity, caused by history. 'R' can be the
6809 recursive thing or the name 'R' (and similarly for 'R' followed by
6810 digits). We look for a name first; if not found, we try the other case.
6811
6812 For compatibility with auto-callouts, we allow a callout to be
6813 specified before a condition that is an assertion. First, check for the
6814 syntax of a callout; if found, adjust the temporary pointer that is
6815 used to check for an assertion condition. That's all that is needed! */
6816
6817 if (ptr[1] == CHAR_QUESTION_MARK && ptr[2] == CHAR_C)
6818 {
6819 for (i = 3;; i++) if (!IS_DIGIT(ptr[i])) break;
6820 if (ptr[i] == CHAR_RIGHT_PARENTHESIS)
6821 tempptr += i + 1;
6822
6823 /* tempptr should now be pointing to the opening parenthesis of the
6824 assertion condition. */
6825
6826 if (*tempptr != CHAR_LEFT_PARENTHESIS)
6827 {
6828 *errorcodeptr = ERR28;
6829 goto FAILED;
6830 }
6831 }
6832
6833 /* For conditions that are assertions, check the syntax, and then exit
6834 the switch. This will take control down to where bracketed groups,
6835 including assertions, are processed. */
6836
6837 if (tempptr[1] == CHAR_QUESTION_MARK &&
6838 (tempptr[2] == CHAR_EQUALS_SIGN ||
6839 tempptr[2] == CHAR_EXCLAMATION_MARK ||
6840 (tempptr[2] == CHAR_LESS_THAN_SIGN &&
6841 (tempptr[3] == CHAR_EQUALS_SIGN ||
6842 tempptr[3] == CHAR_EXCLAMATION_MARK))))
6843 {
6844 cd->iscondassert = TRUE;
6845 break;
6846 }
6847
6848 /* Other conditions use OP_CREF/OP_DNCREF/OP_RREF/OP_DNRREF, and all
6849 need to skip at least 1+IMM2_SIZE bytes at the start of the group. */
6850
6851 code[1+LINK_SIZE] = OP_CREF;
6852 skipbytes = 1+IMM2_SIZE;
6853 refsign = -1; /* => not a number */
6854 namelen = -1; /* => not a name; must set to avoid warning */
6855 name = NULL; /* Always set to avoid warning */
6856 recno = 0; /* Always set to avoid warning */
6857
6858 /* Check for a test for recursion in a named group. */
6859
6860 ptr++;
6861 if (*ptr == CHAR_R && ptr[1] == CHAR_AMPERSAND)
6862 {
6863 terminator = -1;
6864 ptr += 2;
6865 code[1+LINK_SIZE] = OP_RREF; /* Change the type of test */
6866 }
6867
6868 /* Check for a test for a named group's having been set, using the Perl
6869 syntax (?(<name>) or (?('name'), and also allow for the original PCRE
6870 syntax of (?(name) or for (?(+n), (?(-n), and just (?(n). */
6871
6872 else if (*ptr == CHAR_LESS_THAN_SIGN)
6873 {
6874 terminator = CHAR_GREATER_THAN_SIGN;
6875 ptr++;
6876 }
6877 else if (*ptr == CHAR_APOSTROPHE)
6878 {
6879 terminator = CHAR_APOSTROPHE;
6880 ptr++;
6881 }
6882 else
6883 {
6884 terminator = CHAR_NULL;
6885 if (*ptr == CHAR_MINUS || *ptr == CHAR_PLUS) refsign = *ptr++;
6886 else if (IS_DIGIT(*ptr)) refsign = 0;
6887 }
6888
6889 /* Handle a number */
6890
6891 if (refsign >= 0)
6892 {
6893 while (IS_DIGIT(*ptr))
6894 {
6895 if (recno > INT_MAX / 10 - 1) /* Integer overflow */
6896 {
6897 while (IS_DIGIT(*ptr)) ptr++;
6898 *errorcodeptr = ERR61;
6899 goto FAILED;
6900 }
6901 recno = recno * 10 + (int)(*ptr - CHAR_0);
6902 ptr++;
6903 }
6904 }
6905
6906 /* Otherwise we expect to read a name; anything else is an error. When
6907 a name is one of a number of duplicates, a different opcode is used and
6908 it needs more memory. Unfortunately we cannot tell whether a name is a
6909 duplicate in the first pass, so we have to allow for more memory. */
6910
6911 else
6912 {
6913 if (IS_DIGIT(*ptr))
6914 {
6915 *errorcodeptr = ERR84;
6916 goto FAILED;
6917 }
6918 if (!MAX_255(*ptr) || (cd->ctypes[*ptr] & ctype_word) == 0)
6919 {
6920 *errorcodeptr = ERR28; /* Assertion expected */
6921 goto FAILED;
6922 }
6923 name = ptr++;
6924 while (MAX_255(*ptr) && (cd->ctypes[*ptr] & ctype_word) != 0)
6925 {
6926 ptr++;
6927 }
6928 namelen = (int)(ptr - name);
6929 if (lengthptr != NULL) skipbytes += IMM2_SIZE;
6930 }
6931
6932 /* Check the terminator */
6933
6934 if ((terminator > 0 && *ptr++ != (pcre_uchar)terminator) ||
6935 *ptr++ != CHAR_RIGHT_PARENTHESIS)
6936 {
6937 ptr--; /* Error offset */
6938 *errorcodeptr = ERR26; /* Malformed number or name */
6939 goto FAILED;
6940 }
6941
6942 /* Do no further checking in the pre-compile phase. */
6943
6944 if (lengthptr != NULL) break;
6945
6946 /* In the real compile we do the work of looking for the actual
6947 reference. If refsign is not negative, it means we have a number in
6948 recno. */
6949
6950 if (refsign >= 0)
6951 {
6952 if (recno <= 0)
6953 {
6954 *errorcodeptr = ERR35;
6955 goto FAILED;
6956 }
6957 if (refsign != 0) recno = (refsign == CHAR_MINUS)?
6958 cd->bracount - recno + 1 : recno + cd->bracount;
6959 if (recno <= 0 || recno > cd->final_bracount)
6960 {
6961 *errorcodeptr = ERR15;
6962 goto FAILED;
6963 }
6964 PUT2(code, 2+LINK_SIZE, recno);
6965 if (recno > cd->top_backref) cd->top_backref = recno;
6966 break;
6967 }
6968
6969 /* Otherwise look for the name. */
6970
6971 slot = cd->name_table;
6972 for (i = 0; i < cd->names_found; i++)
6973 {
6974 if (STRNCMP_UC_UC(name, slot+IMM2_SIZE, namelen) == 0 &&
6975 slot[IMM2_SIZE+namelen] == 0) break;
6976 slot += cd->name_entry_size;
6977 }
6978
6979 /* Found the named subpattern. If the name is duplicated, add one to
6980 the opcode to change CREF/RREF into DNCREF/DNRREF and insert
6981 appropriate data values. Otherwise, just insert the unique subpattern
6982 number. */
6983
6984 if (i < cd->names_found)
6985 {
6986 int offset = i++;
6987 int count = 1;
6988 recno = GET2(slot, 0); /* Number from first found */
6989 if (recno > cd->top_backref) cd->top_backref = recno;
6990 for (; i < cd->names_found; i++)
6991 {
6992 slot += cd->name_entry_size;
6993 if (STRNCMP_UC_UC(name, slot+IMM2_SIZE, namelen) != 0 ||
6994 (slot+IMM2_SIZE)[namelen] != 0) break;
6995 count++;
6996 }
6997
6998 if (count > 1)
6999 {
7000 PUT2(code, 2+LINK_SIZE, offset);
7001 PUT2(code, 2+LINK_SIZE+IMM2_SIZE, count);
7002 skipbytes += IMM2_SIZE;
7003 code[1+LINK_SIZE]++;
7004 }
7005 else /* Not a duplicated name */
7006 {
7007 PUT2(code, 2+LINK_SIZE, recno);
7008 }
7009 }
7010
7011 /* If terminator == CHAR_NULL it means that the name followed directly
7012 after the opening parenthesis [e.g. (?(abc)...] and in this case there
7013 are some further alternatives to try. For the cases where terminator !=
7014 CHAR_NULL [things like (?(<name>... or (?('name')... or (?(R&name)... ]
7015 we have now checked all the possibilities, so give an error. */
7016
7017 else if (terminator != CHAR_NULL)
7018 {
7019 *errorcodeptr = ERR15;
7020 goto FAILED;
7021 }
7022
7023 /* Check for (?(R) for recursion. Allow digits after R to specify a
7024 specific group number. */
7025
7026 else if (*name == CHAR_R)
7027 {
7028 recno = 0;
7029 for (i = 1; i < namelen; i++)
7030 {
7031 if (!IS_DIGIT(name[i]))
7032 {
7033 *errorcodeptr = ERR15;
7034 goto FAILED;
7035 }
7036 if (recno > INT_MAX / 10 - 1) /* Integer overflow */
7037 {
7038 *errorcodeptr = ERR61;
7039 goto FAILED;
7040 }
7041 recno = recno * 10 + name[i] - CHAR_0;
7042 }
7043 if (recno == 0) recno = RREF_ANY;
7044 code[1+LINK_SIZE] = OP_RREF; /* Change test type */
7045 PUT2(code, 2+LINK_SIZE, recno);
7046 }
7047
7048 /* Similarly, check for the (?(DEFINE) "condition", which is always
7049 false. */
7050
7051 else if (namelen == 6 && STRNCMP_UC_C8(name, STRING_DEFINE, 6) == 0)
7052 {
7053 code[1+LINK_SIZE] = OP_DEF;
7054 skipbytes = 1;
7055 }
7056
7057 /* Reference to an unidentified subpattern. */
7058
7059 else
7060 {
7061 *errorcodeptr = ERR15;
7062 goto FAILED;
7063 }
7064 break;
7065
7066
7067 /* ------------------------------------------------------------ */
7068 case CHAR_EQUALS_SIGN: /* Positive lookahead */
7069 bravalue = OP_ASSERT;
7070 cd->assert_depth += 1;
7071 ptr++;
7072 break;
7073
7074 /* Optimize (?!) to (*FAIL) unless it is quantified - which is a weird
7075 thing to do, but Perl allows all assertions to be quantified, and when
7076 they contain capturing parentheses there may be a potential use for
7077 this feature. Not that that applies to a quantified (?!) but we allow
7078 it for uniformity. */
7079
7080 /* ------------------------------------------------------------ */
7081 case CHAR_EXCLAMATION_MARK: /* Negative lookahead */
7082 ptr++;
7083 if (*ptr == CHAR_RIGHT_PARENTHESIS && ptr[1] != CHAR_ASTERISK &&
7084 ptr[1] != CHAR_PLUS && ptr[1] != CHAR_QUESTION_MARK &&
7085 (ptr[1] != CHAR_LEFT_CURLY_BRACKET || !is_counted_repeat(ptr+2)))
7086 {
7087 *code++ = OP_FAIL;
7088 previous = NULL;
7089 continue;
7090 }
7091 bravalue = OP_ASSERT_NOT;
7092 cd->assert_depth += 1;
7093 break;
7094
7095
7096 /* ------------------------------------------------------------ */
7097 case CHAR_LESS_THAN_SIGN: /* Lookbehind or named define */
7098 switch (ptr[1])
7099 {
7100 case CHAR_EQUALS_SIGN: /* Positive lookbehind */
7101 bravalue = OP_ASSERTBACK;
7102 cd->assert_depth += 1;
7103 ptr += 2;
7104 break;
7105
7106 case CHAR_EXCLAMATION_MARK: /* Negative lookbehind */
7107 bravalue = OP_ASSERTBACK_NOT;
7108 cd->assert_depth += 1;
7109 ptr += 2;
7110 break;
7111
7112 default: /* Could be name define, else bad */
7113 if (MAX_255(ptr[1]) && (cd->ctypes[ptr[1]] & ctype_word) != 0)
7114 goto DEFINE_NAME;
7115 ptr++; /* Correct offset for error */
7116 *errorcodeptr = ERR24;
7117 goto FAILED;
7118 }
7119 break;
7120
7121
7122 /* ------------------------------------------------------------ */
7123 case CHAR_GREATER_THAN_SIGN: /* One-time brackets */
7124 bravalue = OP_ONCE;
7125 ptr++;
7126 break;
7127
7128
7129 /* ------------------------------------------------------------ */
7130 case CHAR_C: /* Callout - may be followed by digits; */
7131 previous_callout = code; /* Save for later completion */
7132 after_manual_callout = 1; /* Skip one item before completing */
7133 *code++ = OP_CALLOUT;
7134 {
7135 int n = 0;
7136 ptr++;
7137 while(IS_DIGIT(*ptr))
7138 {
7139 n = n * 10 + *ptr++ - CHAR_0;
7140 if (n > 255)
7141 {
7142 *errorcodeptr = ERR38;
7143 goto FAILED;
7144 }
7145 }
7146 if (*ptr != CHAR_RIGHT_PARENTHESIS)
7147 {
7148 *errorcodeptr = ERR39;
7149 goto FAILED;
7150 }
7151 *code++ = n;
7152 PUT(code, 0, (int)(ptr - cd->start_pattern + 1)); /* Pattern offset */
7153 PUT(code, LINK_SIZE, 0); /* Default length */
7154 code += 2 * LINK_SIZE;
7155 }
7156 previous = NULL;
7157 continue;
7158
7159
7160 /* ------------------------------------------------------------ */
7161 case CHAR_P: /* Python-style named subpattern handling */
7162 if (*(++ptr) == CHAR_EQUALS_SIGN ||
7163 *ptr == CHAR_GREATER_THAN_SIGN) /* Reference or recursion */
7164 {
7165 is_recurse = *ptr == CHAR_GREATER_THAN_SIGN;
7166 terminator = CHAR_RIGHT_PARENTHESIS;
7167 goto NAMED_REF_OR_RECURSE;
7168 }
7169 else if (*ptr != CHAR_LESS_THAN_SIGN) /* Test for Python-style defn */
7170 {
7171 *errorcodeptr = ERR41;
7172 goto FAILED;
7173 }
7174 /* Fall through to handle (?P< as (?< is handled */
7175
7176
7177 /* ------------------------------------------------------------ */
7178 DEFINE_NAME: /* Come here from (?< handling */
7179 case CHAR_APOSTROPHE:
7180 terminator = (*ptr == CHAR_LESS_THAN_SIGN)?
7181 CHAR_GREATER_THAN_SIGN : CHAR_APOSTROPHE;
7182 name = ++ptr;
7183 if (IS_DIGIT(*ptr))
7184 {
7185 *errorcodeptr = ERR84; /* Group name must start with non-digit */
7186 goto FAILED;
7187 }
7188 while (MAX_255(*ptr) && (cd->ctypes[*ptr] & ctype_word) != 0) ptr++;
7189 namelen = (int)(ptr - name);
7190
7191 /* In the pre-compile phase, do a syntax check, remember the longest
7192 name, and then remember the group in a vector, expanding it if
7193 necessary. Duplicates for the same number are skipped; other duplicates
7194 are checked for validity. In the actual compile, there is nothing to
7195 do. */
7196
7197 if (lengthptr != NULL)
7198 {
7199 named_group *ng;
7200 pcre_uint32 number = cd->bracount + 1;
7201
7202 if (*ptr != (pcre_uchar)terminator)
7203 {
7204 *errorcodeptr = ERR42;
7205 goto FAILED;
7206 }
7207
7208 if (cd->names_found >= MAX_NAME_COUNT)
7209 {
7210 *errorcodeptr = ERR49;
7211 goto FAILED;
7212 }
7213
7214 if (namelen + IMM2_SIZE + 1 > cd->name_entry_size)
7215 {
7216 cd->name_entry_size = namelen + IMM2_SIZE + 1;
7217 if (namelen > MAX_NAME_SIZE)
7218 {
7219 *errorcodeptr = ERR48;
7220 goto FAILED;
7221 }
7222 }
7223
7224 /* Scan the list to check for duplicates. For duplicate names, if the
7225 number is the same, break the loop, which causes the name to be
7226 discarded; otherwise, if DUPNAMES is not set, give an error.
7227 If it is set, allow the name with a different number, but continue
7228 scanning in case this is a duplicate with the same number. For
7229 non-duplicate names, give an error if the number is duplicated. */
7230
7231 ng = cd->named_groups;
7232 for (i = 0; i < cd->names_found; i++, ng++)
7233 {
7234 if (namelen == ng->length &&
7235 STRNCMP_UC_UC(name, ng->name, namelen) == 0)
7236 {
7237 if (ng->number == number) break;
7238 if ((options & PCRE_DUPNAMES) == 0)
7239 {
7240 *errorcodeptr = ERR43;
7241 goto FAILED;
7242 }
7243 cd->dupnames = TRUE; /* Duplicate names exist */
7244 }
7245 else if (ng->number == number)
7246 {
7247 *errorcodeptr = ERR65;
7248 goto FAILED;
7249 }
7250 }
7251
7252 if (i >= cd->names_found) /* Not a duplicate with same number */
7253 {
7254 /* Increase the list size if necessary */
7255
7256 if (cd->names_found >= cd->named_group_list_size)
7257 {
7258 int newsize = cd->named_group_list_size * 2;
7259 named_group *newspace = (PUBL(malloc))
7260 (newsize * sizeof(named_group));
7261
7262 if (newspace == NULL)
7263 {
7264 *errorcodeptr = ERR21;
7265 goto FAILED;
7266 }
7267
7268 memcpy(newspace, cd->named_groups,
7269 cd->named_group_list_size * sizeof(named_group));
7270 if (cd->named_group_list_size > NAMED_GROUP_LIST_SIZE)
7271 (PUBL(free))((void *)cd->named_groups);
7272 cd->named_groups = newspace;
7273 cd->named_group_list_size = newsize;
7274 }
7275
7276 cd->named_groups[cd->names_found].name = name;
7277 cd->named_groups[cd->names_found].length = namelen;
7278 cd->named_groups[cd->names_found].number = number;
7279 cd->names_found++;
7280 }
7281 }
7282
7283 ptr++; /* Move past > or ' in both passes. */
7284 goto NUMBERED_GROUP;
7285
7286
7287 /* ------------------------------------------------------------ */
7288 case CHAR_AMPERSAND: /* Perl recursion/subroutine syntax */
7289 terminator = CHAR_RIGHT_PARENTHESIS;
7290 is_recurse = TRUE;
7291 /* Fall through */
7292
7293 /* We come here from the Python syntax above that handles both
7294 references (?P=name) and recursion (?P>name), as well as falling
7295 through from the Perl recursion syntax (?&name). We also come here from
7296 the Perl \k<name> or \k'name' back reference syntax and the \k{name}
7297 .NET syntax, and the Oniguruma \g<...> and \g'...' subroutine syntax. */
7298
7299 NAMED_REF_OR_RECURSE:
7300 name = ++ptr;
7301 if (IS_DIGIT(*ptr))
7302 {
7303 *errorcodeptr = ERR84; /* Group name must start with non-digit */
7304 goto FAILED;
7305 }
7306 while (MAX_255(*ptr) && (cd->ctypes[*ptr] & ctype_word) != 0) ptr++;
7307 namelen = (int)(ptr - name);
7308
7309 /* In the pre-compile phase, do a syntax check. We used to just set
7310 a dummy reference number, because it was not used in the first pass.
7311 However, with the change of recursive back references to be atomic,
7312 we have to look for the number so that this state can be identified, as
7313 otherwise the incorrect length is computed. If it's not a backwards
7314 reference, the dummy number will do. */
7315
7316 if (lengthptr != NULL)
7317 {
7318 named_group *ng;
7319 recno = 0;
7320
7321 if (namelen == 0)
7322 {
7323 *errorcodeptr = ERR62;
7324 goto FAILED;
7325 }
7326 if (*ptr != (pcre_uchar)terminator)
7327 {
7328 *errorcodeptr = ERR42;
7329 goto FAILED;
7330 }
7331 if (namelen > MAX_NAME_SIZE)
7332 {
7333 *errorcodeptr = ERR48;
7334 goto FAILED;
7335 }
7336
7337 /* Count named back references. */
7338
7339 if (!is_recurse) cd->namedrefcount++;
7340
7341 /* We have to allow for a named reference to a duplicated name (this
7342 cannot be determined until the second pass). This needs an extra
7343 16-bit data item. */
7344
7345 *lengthptr += IMM2_SIZE;
7346
7347 /* If this is a forward reference and we are within a (?|...) group,
7348 the reference may end up as the number of a group which we are
7349 currently inside, that is, it could be a recursive reference. In the
7350 real compile this will be picked up and the reference wrapped with
7351 OP_ONCE to make it atomic, so we must space in case this occurs. */
7352
7353 /* In fact, this can happen for a non-forward reference because
7354 another group with the same number might be created later. This
7355 issue is fixed "properly" in PCRE2. As PCRE1 is now in maintenance
7356 only mode, we finesse the bug by allowing more memory always. */
7357
7358 *lengthptr += 4 + 4*LINK_SIZE;
7359
7360 /* It is even worse than that. The current reference may be to an
7361 existing named group with a different number (so apparently not
7362 recursive) but which later on is also attached to a group with the
7363 current number. This can only happen if $(| has been previous
7364 encountered. In that case, we allow yet more memory, just in case.
7365 (Again, this is fixed "properly" in PCRE2. */
7366
7367 if (cd->dupgroups) *lengthptr += 4 + 4*LINK_SIZE;
7368
7369 /* Otherwise, check for recursion here. The name table does not exist
7370 in the first pass; instead we must scan the list of names encountered
7371 so far in order to get the number. If the name is not found, leave
7372 the value of recno as 0 for a forward reference. */
7373
7374 /* This patch (removing "else") fixes a problem when a reference is
7375 to multiple identically named nested groups from within the nest.
7376 Once again, it is not the "proper" fix, and it results in an
7377 over-allocation of memory. */
7378
7379 /* else */
7380 {
7381 ng = cd->named_groups;
7382 for (i = 0; i < cd->names_found; i++, ng++)
7383 {
7384 if (namelen == ng->length &&
7385 STRNCMP_UC_UC(name, ng->name, namelen) == 0)
7386 {
7387 open_capitem *oc;
7388 recno = ng->number;
7389 if (is_recurse) break;
7390 for (oc = cd->open_caps; oc != NULL; oc = oc->next)
7391 {
7392 if (oc->number == recno)
7393 {
7394 oc->flag = TRUE;
7395 break;
7396 }
7397 }
7398 }
7399 }
7400 }
7401 }
7402
7403 /* In the real compile, search the name table. We check the name
7404 first, and then check that we have reached the end of the name in the
7405 table. That way, if the name is longer than any in the table, the
7406 comparison will fail without reading beyond the table entry. */
7407
7408 else
7409 {
7410 slot = cd->name_table;
7411 for (i = 0; i < cd->names_found; i++)
7412 {
7413 if (STRNCMP_UC_UC(name, slot+IMM2_SIZE, namelen) == 0 &&
7414 slot[IMM2_SIZE+namelen] == 0)
7415 break;
7416 slot += cd->name_entry_size;
7417 }
7418
7419 if (i < cd->names_found)
7420 {
7421 recno = GET2(slot, 0);
7422 }
7423 else
7424 {
7425 *errorcodeptr = ERR15;
7426 goto FAILED;
7427 }
7428 }
7429
7430 /* In both phases, for recursions, we can now go to the code than
7431 handles numerical recursion. */
7432
7433 if (is_recurse) goto HANDLE_RECURSION;
7434
7435 /* In the second pass we must see if the name is duplicated. If so, we
7436 generate a different opcode. */
7437
7438 if (lengthptr == NULL && cd->dupnames)
7439 {
7440 int count = 1;
7441 unsigned int index = i;
7442 pcre_uchar *cslot = slot + cd->name_entry_size;
7443
7444 for (i++; i < cd->names_found; i++)
7445 {
7446 if (STRCMP_UC_UC(slot + IMM2_SIZE, cslot + IMM2_SIZE) != 0) break;
7447 count++;
7448 cslot += cd->name_entry_size;
7449 }
7450
7451 if (count > 1)
7452 {
7453 if (firstcharflags == REQ_UNSET) firstcharflags = REQ_NONE;
7454 previous = code;
7455 item_hwm_offset = cd->hwm - cd->start_workspace;
7456 *code++ = ((options & PCRE_CASELESS) != 0)? OP_DNREFI : OP_DNREF;
7457 PUT2INC(code, 0, index);
7458 PUT2INC(code, 0, count);
7459
7460 /* Process each potentially referenced group. */
7461
7462 for (; slot < cslot; slot += cd->name_entry_size)
7463 {
7464 open_capitem *oc;
7465 recno = GET2(slot, 0);
7466 cd->backref_map |= (recno < 32)? (1U << recno) : 1;
7467 if (recno > cd->top_backref) cd->top_backref = recno;
7468
7469 /* Check to see if this back reference is recursive, that it, it
7470 is inside the group that it references. A flag is set so that the
7471 group can be made atomic. */
7472
7473 for (oc = cd->open_caps; oc != NULL; oc = oc->next)
7474 {
7475 if (oc->number == recno)
7476 {
7477 oc->flag = TRUE;
7478 break;
7479 }
7480 }
7481 }
7482
7483 continue; /* End of back ref handling */
7484 }
7485 }
7486
7487 /* First pass, or a non-duplicated name. */
7488
7489 goto HANDLE_REFERENCE;
7490
7491
7492 /* ------------------------------------------------------------ */
7493 case CHAR_R: /* Recursion, same as (?0) */
7494 recno = 0;
7495 if (*(++ptr) != CHAR_RIGHT_PARENTHESIS)
7496 {
7497 *errorcodeptr = ERR29;
7498 goto FAILED;
7499 }
7500 goto HANDLE_RECURSION;
7501
7502
7503 /* ------------------------------------------------------------ */
7504 case CHAR_MINUS: case CHAR_PLUS: /* Recursion or subroutine */
7505 case CHAR_0: case CHAR_1: case CHAR_2: case CHAR_3: case CHAR_4:
7506 case CHAR_5: case CHAR_6: case CHAR_7: case CHAR_8: case CHAR_9:
7507 {
7508 const pcre_uchar *called;
7509 terminator = CHAR_RIGHT_PARENTHESIS;
7510
7511 /* Come here from the \g<...> and \g'...' code (Oniguruma
7512 compatibility). However, the syntax has been checked to ensure that
7513 the ... are a (signed) number, so that neither ERR63 nor ERR29 will
7514 be called on this path, nor with the jump to OTHER_CHAR_AFTER_QUERY
7515 ever be taken. */
7516
7517 HANDLE_NUMERICAL_RECURSION:
7518
7519 if ((refsign = *ptr) == CHAR_PLUS)
7520 {
7521 ptr++;
7522 if (!IS_DIGIT(*ptr))
7523 {
7524 *errorcodeptr = ERR63;
7525 goto FAILED;
7526 }
7527 }
7528 else if (refsign == CHAR_MINUS)
7529 {
7530 if (!IS_DIGIT(ptr[1]))
7531 goto OTHER_CHAR_AFTER_QUERY;
7532 ptr++;
7533 }
7534
7535 recno = 0;
7536 while(IS_DIGIT(*ptr))
7537 {
7538 if (recno > INT_MAX / 10 - 1) /* Integer overflow */
7539 {
7540 while (IS_DIGIT(*ptr)) ptr++;
7541 *errorcodeptr = ERR61;
7542 goto FAILED;
7543 }
7544 recno = recno * 10 + *ptr++ - CHAR_0;
7545 }
7546
7547 if (*ptr != (pcre_uchar)terminator)
7548 {
7549 *errorcodeptr = ERR29;
7550 goto FAILED;
7551 }
7552
7553 if (refsign == CHAR_MINUS)
7554 {
7555 if (recno == 0)
7556 {
7557 *errorcodeptr = ERR58;
7558 goto FAILED;
7559 }
7560 recno = cd->bracount - recno + 1;
7561 if (recno <= 0)
7562 {
7563 *errorcodeptr = ERR15;
7564 goto FAILED;
7565 }
7566 }
7567 else if (refsign == CHAR_PLUS)
7568 {
7569 if (recno == 0)
7570 {
7571 *errorcodeptr = ERR58;
7572 goto FAILED;
7573 }
7574 recno += cd->bracount;
7575 }
7576
7577 /* Come here from code above that handles a named recursion */
7578
7579 HANDLE_RECURSION:
7580
7581 previous = code;
7582 item_hwm_offset = cd->hwm - cd->start_workspace;
7583 called = cd->start_code;
7584
7585 /* When we are actually compiling, find the bracket that is being
7586 referenced. Temporarily end the regex in case it doesn't exist before
7587 this point. If we end up with a forward reference, first check that
7588 the bracket does occur later so we can give the error (and position)
7589 now. Then remember this forward reference in the workspace so it can
7590 be filled in at the end. */
7591
7592 if (lengthptr == NULL)
7593 {
7594 *code = OP_END;
7595 if (recno != 0)
7596 called = PRIV(find_bracket)(cd->start_code, utf, recno);
7597
7598 /* Forward reference */
7599
7600 if (called == NULL)
7601 {
7602 if (recno > cd->final_bracount)
7603 {
7604 *errorcodeptr = ERR15;
7605 goto FAILED;
7606 }
7607
7608 /* Fudge the value of "called" so that when it is inserted as an
7609 offset below, what it actually inserted is the reference number
7610 of the group. Then remember the forward reference. */
7611
7612 called = cd->start_code + recno;
7613 if (cd->hwm >= cd->start_workspace + cd->workspace_size -
7614 WORK_SIZE_SAFETY_MARGIN)
7615 {
7616 *errorcodeptr = expand_workspace(cd);
7617 if (*errorcodeptr != 0) goto FAILED;
7618 }
7619 PUTINC(cd->hwm, 0, (int)(code + 1 - cd->start_code));
7620 }
7621
7622 /* If not a forward reference, and the subpattern is still open,
7623 this is a recursive call. We check to see if this is a left
7624 recursion that could loop for ever, and diagnose that case. We
7625 must not, however, do this check if we are in a conditional
7626 subpattern because the condition might be testing for recursion in
7627 a pattern such as /(?(R)a+|(?R)b)/, which is perfectly valid.
7628 Forever loops are also detected at runtime, so those that occur in
7629 conditional subpatterns will be picked up then. */
7630
7631 else if (GET(called, 1) == 0 && cond_depth <= 0 &&
7632 could_be_empty(called, code, bcptr, utf, cd))
7633 {
7634 *errorcodeptr = ERR40;
7635 goto FAILED;
7636 }
7637 }
7638
7639 /* Insert the recursion/subroutine item. It does not have a set first
7640 character (relevant if it is repeated, because it will then be
7641 wrapped with ONCE brackets). */
7642
7643 *code = OP_RECURSE;
7644 PUT(code, 1, (int)(called - cd->start_code));
7645 code += 1 + LINK_SIZE;
7646 groupsetfirstchar = FALSE;
7647 }
7648
7649 /* Can't determine a first byte now */
7650
7651 if (firstcharflags == REQ_UNSET) firstcharflags = REQ_NONE;
7652 zerofirstchar = firstchar;
7653 zerofirstcharflags = firstcharflags;
7654 continue;
7655
7656
7657 /* ------------------------------------------------------------ */
7658 default: /* Other characters: check option setting */
7659 OTHER_CHAR_AFTER_QUERY:
7660 set = unset = 0;
7661 optset = &set;
7662
7663 while (*ptr != CHAR_RIGHT_PARENTHESIS && *ptr != CHAR_COLON)
7664 {
7665 switch (*ptr++)
7666 {
7667 case CHAR_MINUS: optset = &unset; break;
7668
7669 case CHAR_J: /* Record that it changed in the external options */
7670 *optset |= PCRE_DUPNAMES;
7671 cd->external_flags |= PCRE_JCHANGED;
7672 break;
7673
7674 case CHAR_i: *optset |= PCRE_CASELESS; break;
7675 case CHAR_m: *optset |= PCRE_MULTILINE; break;
7676 case CHAR_s: *optset |= PCRE_DOTALL; break;
7677 case CHAR_x: *optset |= PCRE_EXTENDED; break;
7678 case CHAR_U: *optset |= PCRE_UNGREEDY; break;
7679 case CHAR_X: *optset |= PCRE_EXTRA; break;
7680
7681 default: *errorcodeptr = ERR12;
7682 ptr--; /* Correct the offset */
7683 goto FAILED;
7684 }
7685 }
7686
7687 /* Set up the changed option bits, but don't change anything yet. */
7688
7689 newoptions = (options | set) & (~unset);
7690
7691 /* If the options ended with ')' this is not the start of a nested
7692 group with option changes, so the options change at this level.
7693 If we are not at the pattern start, reset the greedy defaults and the
7694 case value for firstchar and reqchar. */
7695
7696 if (*ptr == CHAR_RIGHT_PARENTHESIS)
7697 {
7698 greedy_default = ((newoptions & PCRE_UNGREEDY) != 0);
7699 greedy_non_default = greedy_default ^ 1;
7700 req_caseopt = ((newoptions & PCRE_CASELESS) != 0)? REQ_CASELESS:0;
7701
7702 /* Change options at this level, and pass them back for use
7703 in subsequent branches. */
7704
7705 *optionsptr = options = newoptions;
7706 previous = NULL; /* This item can't be repeated */
7707 continue; /* It is complete */
7708 }
7709
7710 /* If the options ended with ':' we are heading into a nested group
7711 with possible change of options. Such groups are non-capturing and are
7712 not assertions of any kind. All we need to do is skip over the ':';
7713 the newoptions value is handled below. */
7714
7715 bravalue = OP_BRA;
7716 ptr++;
7717 } /* End of switch for character following (? */
7718 } /* End of (? handling */
7719
7720 /* Opening parenthesis not followed by '*' or '?'. If PCRE_NO_AUTO_CAPTURE
7721 is set, all unadorned brackets become non-capturing and behave like (?:...)
7722 brackets. */
7723
7724 else if ((options & PCRE_NO_AUTO_CAPTURE) != 0)
7725 {
7726 bravalue = OP_BRA;
7727 }
7728
7729 /* Else we have a capturing group. */
7730
7731 else
7732 {
7733 NUMBERED_GROUP:
7734 cd->bracount += 1;
7735 PUT2(code, 1+LINK_SIZE, cd->bracount);
7736 skipbytes = IMM2_SIZE;
7737 }
7738
7739 /* Process nested bracketed regex. First check for parentheses nested too
7740 deeply. */
7741
7742 if ((cd->parens_depth += 1) > PARENS_NEST_LIMIT)
7743 {
7744 *errorcodeptr = ERR82;
7745 goto FAILED;
7746 }
7747
7748 /* All assertions used not to be repeatable, but this was changed for Perl
7749 compatibility. All kinds can now be repeated except for assertions that are
7750 conditions (Perl also forbids these to be repeated). We copy code into a
7751 non-register variable (tempcode) in order to be able to pass its address
7752 because some compilers complain otherwise. At the start of a conditional
7753 group whose condition is an assertion, cd->iscondassert is set. We unset it
7754 here so as to allow assertions later in the group to be quantified. */
7755
7756 if (bravalue >= OP_ASSERT && bravalue <= OP_ASSERTBACK_NOT &&
7757 cd->iscondassert)
7758 {
7759 previous = NULL;
7760 cd->iscondassert = FALSE;
7761 }
7762 else
7763 {
7764 previous = code;
7765 item_hwm_offset = cd->hwm - cd->start_workspace;
7766 }
7767
7768 *code = bravalue;
7769 tempcode = code;
7770 tempreqvary = cd->req_varyopt; /* Save value before bracket */
7771 tempbracount = cd->bracount; /* Save value before bracket */
7772 length_prevgroup = 0; /* Initialize for pre-compile phase */
7773
7774 if (!compile_regex(
7775 newoptions, /* The complete new option state */
7776 &tempcode, /* Where to put code (updated) */
7777 &ptr, /* Input pointer (updated) */
7778 errorcodeptr, /* Where to put an error message */
7779 (bravalue == OP_ASSERTBACK ||
7780 bravalue == OP_ASSERTBACK_NOT), /* TRUE if back assert */
7781 reset_bracount, /* True if (?| group */
7782 skipbytes, /* Skip over bracket number */
7783 cond_depth +
7784 ((bravalue == OP_COND)?1:0), /* Depth of condition subpatterns */
7785 &subfirstchar, /* For possible first char */
7786 &subfirstcharflags,
7787 &subreqchar, /* For possible last char */
7788 &subreqcharflags,
7789 bcptr, /* Current branch chain */
7790 cd, /* Tables block */
7791 (lengthptr == NULL)? NULL : /* Actual compile phase */
7792 &length_prevgroup /* Pre-compile phase */
7793 ))
7794 goto FAILED;
7795
7796 cd->parens_depth -= 1;
7797
7798 /* If this was an atomic group and there are no capturing groups within it,
7799 generate OP_ONCE_NC instead of OP_ONCE. */
7800
7801 if (bravalue == OP_ONCE && cd->bracount <= tempbracount)
7802 *code = OP_ONCE_NC;
7803
7804 if (bravalue >= OP_ASSERT && bravalue <= OP_ASSERTBACK_NOT)
7805 cd->assert_depth -= 1;
7806
7807 /* At the end of compiling, code is still pointing to the start of the
7808 group, while tempcode has been updated to point past the end of the group.
7809 The pattern pointer (ptr) is on the bracket.
7810
7811 If this is a conditional bracket, check that there are no more than
7812 two branches in the group, or just one if it's a DEFINE group. We do this
7813 in the real compile phase, not in the pre-pass, where the whole group may
7814 not be available. */
7815
7816 if (bravalue == OP_COND && lengthptr == NULL)
7817 {
7818 pcre_uchar *tc = code;
7819 int condcount = 0;
7820
7821 do {
7822 condcount++;
7823 tc += GET(tc,1);
7824 }
7825 while (*tc != OP_KET);
7826
7827 /* A DEFINE group is never obeyed inline (the "condition" is always
7828 false). It must have only one branch. */
7829
7830 if (code[LINK_SIZE+1] == OP_DEF)
7831 {
7832 if (condcount > 1)
7833 {
7834 *errorcodeptr = ERR54;
7835 goto FAILED;
7836 }
7837 bravalue = OP_DEF; /* Just a flag to suppress char handling below */
7838 }
7839
7840 /* A "normal" conditional group. If there is just one branch, we must not
7841 make use of its firstchar or reqchar, because this is equivalent to an
7842 empty second branch. */
7843
7844 else
7845 {
7846 if (condcount > 2)
7847 {
7848 *errorcodeptr = ERR27;
7849 goto FAILED;
7850 }
7851 if (condcount == 1) subfirstcharflags = subreqcharflags = REQ_NONE;
7852 }
7853 }
7854
7855 /* Error if hit end of pattern */
7856
7857 if (*ptr != CHAR_RIGHT_PARENTHESIS)
7858 {
7859 *errorcodeptr = ERR14;
7860 goto FAILED;
7861 }
7862
7863 /* In the pre-compile phase, update the length by the length of the group,
7864 less the brackets at either end. Then reduce the compiled code to just a
7865 set of non-capturing brackets so that it doesn't use much memory if it is
7866 duplicated by a quantifier.*/
7867
7868 if (lengthptr != NULL)
7869 {
7870 if (OFLOW_MAX - *lengthptr < length_prevgroup - 2 - 2*LINK_SIZE)
7871 {
7872 *errorcodeptr = ERR20;
7873 goto FAILED;
7874 }
7875 *lengthptr += length_prevgroup - 2 - 2*LINK_SIZE;
7876 code++; /* This already contains bravalue */
7877 PUTINC(code, 0, 1 + LINK_SIZE);
7878 *code++ = OP_KET;
7879 PUTINC(code, 0, 1 + LINK_SIZE);
7880 break; /* No need to waste time with special character handling */
7881 }
7882
7883 /* Otherwise update the main code pointer to the end of the group. */
7884
7885 code = tempcode;
7886
7887 /* For a DEFINE group, required and first character settings are not
7888 relevant. */
7889
7890 if (bravalue == OP_DEF) break;
7891
7892 /* Handle updating of the required and first characters for other types of
7893 group. Update for normal brackets of all kinds, and conditions with two
7894 branches (see code above). If the bracket is followed by a quantifier with
7895 zero repeat, we have to back off. Hence the definition of zeroreqchar and
7896 zerofirstchar outside the main loop so that they can be accessed for the
7897 back off. */
7898
7899 zeroreqchar = reqchar;
7900 zeroreqcharflags = reqcharflags;
7901 zerofirstchar = firstchar;
7902 zerofirstcharflags = firstcharflags;
7903 groupsetfirstchar = FALSE;
7904
7905 if (bravalue >= OP_ONCE)
7906 {
7907 /* If we have not yet set a firstchar in this branch, take it from the
7908 subpattern, remembering that it was set here so that a repeat of more
7909 than one can replicate it as reqchar if necessary. If the subpattern has
7910 no firstchar, set "none" for the whole branch. In both cases, a zero
7911 repeat forces firstchar to "none". */
7912
7913 if (firstcharflags == REQ_UNSET)
7914 {
7915 if (subfirstcharflags >= 0)
7916 {
7917 firstchar = subfirstchar;
7918 firstcharflags = subfirstcharflags;
7919 groupsetfirstchar = TRUE;
7920 }
7921 else firstcharflags = REQ_NONE;
7922 zerofirstcharflags = REQ_NONE;
7923 }
7924
7925 /* If firstchar was previously set, convert the subpattern's firstchar
7926 into reqchar if there wasn't one, using the vary flag that was in
7927 existence beforehand. */
7928
7929 else if (subfirstcharflags >= 0 && subreqcharflags < 0)
7930 {
7931 subreqchar = subfirstchar;
7932 subreqcharflags = subfirstcharflags | tempreqvary;
7933 }
7934
7935 /* If the subpattern set a required byte (or set a first byte that isn't
7936 really the first byte - see above), set it. */
7937
7938 if (subreqcharflags >= 0)
7939 {
7940 reqchar = subreqchar;
7941 reqcharflags = subreqcharflags;
7942 }
7943 }
7944
7945 /* For a forward assertion, we take the reqchar, if set, provided that the
7946 group has also set a first char. This can be helpful if the pattern that
7947 follows the assertion doesn't set a different char. For example, it's
7948 useful for /(?=abcde).+/. We can't set firstchar for an assertion, however
7949 because it leads to incorrect effect for patterns such as /(?=a)a.+/ when
7950 the "real" "a" would then become a reqchar instead of a firstchar. This is
7951 overcome by a scan at the end if there's no firstchar, looking for an
7952 asserted first char. */
7953
7954 else if (bravalue == OP_ASSERT && subreqcharflags >= 0 &&
7955 subfirstcharflags >= 0)
7956 {
7957 reqchar = subreqchar;
7958 reqcharflags = subreqcharflags;
7959 }
7960 break; /* End of processing '(' */
7961
7962
7963 /* ===================================================================*/
7964 /* Handle metasequences introduced by \. For ones like \d, the ESC_ values
7965 are arranged to be the negation of the corresponding OP_values in the
7966 default case when PCRE_UCP is not set. For the back references, the values
7967 are negative the reference number. Only back references and those types
7968 that consume a character may be repeated. We can test for values between
7969 ESC_b and ESC_Z for the latter; this may have to change if any new ones are
7970 ever created. */
7971
7972 case CHAR_BACKSLASH:
7973 tempptr = ptr;
7974 escape = check_escape(&ptr, &ec, errorcodeptr, cd->bracount, options, FALSE);
7975 if (*errorcodeptr != 0) goto FAILED;
7976
7977 if (escape == 0) /* The escape coded a single character */
7978 c = ec;
7979 else
7980 {
7981 /* For metasequences that actually match a character, we disable the
7982 setting of a first character if it hasn't already been set. */
7983
7984 if (firstcharflags == REQ_UNSET && escape > ESC_b && escape < ESC_Z)
7985 firstcharflags = REQ_NONE;
7986
7987 /* Set values to reset to if this is followed by a zero repeat. */
7988
7989 zerofirstchar = firstchar;
7990 zerofirstcharflags = firstcharflags;
7991 zeroreqchar = reqchar;
7992 zeroreqcharflags = reqcharflags;
7993
7994 /* \g<name> or \g'name' is a subroutine call by name and \g<n> or \g'n'
7995 is a subroutine call by number (Oniguruma syntax). In fact, the value
7996 ESC_g is returned only for these cases. So we don't need to check for <
7997 or ' if the value is ESC_g. For the Perl syntax \g{n} the value is
7998 -n, and for the Perl syntax \g{name} the result is ESC_k (as
7999 that is a synonym for a named back reference). */
8000
8001 if (escape == ESC_g)
8002 {
8003 const pcre_uchar *p;
8004 pcre_uint32 cf;
8005
8006 item_hwm_offset = cd->hwm - cd->start_workspace; /* Normally this is set when '(' is read */
8007 terminator = (*(++ptr) == CHAR_LESS_THAN_SIGN)?
8008 CHAR_GREATER_THAN_SIGN : CHAR_APOSTROPHE;
8009
8010 /* These two statements stop the compiler for warning about possibly
8011 unset variables caused by the jump to HANDLE_NUMERICAL_RECURSION. In
8012 fact, because we do the check for a number below, the paths that
8013 would actually be in error are never taken. */
8014
8015 skipbytes = 0;
8016 reset_bracount = FALSE;
8017
8018 /* If it's not a signed or unsigned number, treat it as a name. */
8019
8020 cf = ptr[1];
8021 if (cf != CHAR_PLUS && cf != CHAR_MINUS && !IS_DIGIT(cf))
8022 {
8023 is_recurse = TRUE;
8024 goto NAMED_REF_OR_RECURSE;
8025 }
8026
8027 /* Signed or unsigned number (cf = ptr[1]) is known to be plus or minus
8028 or a digit. */
8029
8030 p = ptr + 2;
8031 while (IS_DIGIT(*p)) p++;
8032 if (*p != (pcre_uchar)terminator)
8033 {
8034 *errorcodeptr = ERR57;
8035 goto FAILED;
8036 }
8037 ptr++;
8038 goto HANDLE_NUMERICAL_RECURSION;
8039 }
8040
8041 /* \k<name> or \k'name' is a back reference by name (Perl syntax).
8042 We also support \k{name} (.NET syntax). */
8043
8044 if (escape == ESC_k)
8045 {
8046 if ((ptr[1] != CHAR_LESS_THAN_SIGN &&
8047 ptr[1] != CHAR_APOSTROPHE && ptr[1] != CHAR_LEFT_CURLY_BRACKET))
8048 {
8049 *errorcodeptr = ERR69;
8050 goto FAILED;
8051 }
8052 is_recurse = FALSE;
8053 terminator = (*(++ptr) == CHAR_LESS_THAN_SIGN)?
8054 CHAR_GREATER_THAN_SIGN : (*ptr == CHAR_APOSTROPHE)?
8055 CHAR_APOSTROPHE : CHAR_RIGHT_CURLY_BRACKET;
8056 goto NAMED_REF_OR_RECURSE;
8057 }
8058
8059 /* Back references are handled specially; must disable firstchar if
8060 not set to cope with cases like (?=(\w+))\1: which would otherwise set
8061 ':' later. */
8062
8063 if (escape < 0)
8064 {
8065 open_capitem *oc;
8066 recno = -escape;
8067
8068 /* Come here from named backref handling when the reference is to a
8069 single group (i.e. not to a duplicated name. */
8070
8071 HANDLE_REFERENCE:
8072 if (firstcharflags == REQ_UNSET) zerofirstcharflags = firstcharflags = REQ_NONE;
8073 previous = code;
8074 item_hwm_offset = cd->hwm - cd->start_workspace;
8075 *code++ = ((options & PCRE_CASELESS) != 0)? OP_REFI : OP_REF;
8076 PUT2INC(code, 0, recno);
8077 cd->backref_map |= (recno < 32)? (1U << recno) : 1;
8078 if (recno > cd->top_backref) cd->top_backref = recno;
8079
8080 /* Check to see if this back reference is recursive, that it, it
8081 is inside the group that it references. A flag is set so that the
8082 group can be made atomic. */
8083
8084 for (oc = cd->open_caps; oc != NULL; oc = oc->next)
8085 {
8086 if (oc->number == recno)
8087 {
8088 oc->flag = TRUE;
8089 break;
8090 }
8091 }
8092 }
8093
8094 /* So are Unicode property matches, if supported. */
8095
8096 #ifdef SUPPORT_UCP
8097 else if (escape == ESC_P || escape == ESC_p)
8098 {
8099 BOOL negated;
8100 unsigned int ptype = 0, pdata = 0;
8101 if (!get_ucp(&ptr, &negated, &ptype, &pdata, errorcodeptr))
8102 goto FAILED;
8103 previous = code;
8104 item_hwm_offset = cd->hwm - cd->start_workspace;
8105 *code++ = ((escape == ESC_p) != negated)? OP_PROP : OP_NOTPROP;
8106 *code++ = ptype;
8107 *code++ = pdata;
8108 }
8109 #else
8110
8111 /* If Unicode properties are not supported, \X, \P, and \p are not
8112 allowed. */
8113
8114 else if (escape == ESC_X || escape == ESC_P || escape == ESC_p)
8115 {
8116 *errorcodeptr = ERR45;
8117 goto FAILED;
8118 }
8119 #endif
8120
8121 /* For the rest (including \X when Unicode properties are supported), we
8122 can obtain the OP value by negating the escape value in the default
8123 situation when PCRE_UCP is not set. When it *is* set, we substitute
8124 Unicode property tests. Note that \b and \B do a one-character
8125 lookbehind, and \A also behaves as if it does. */
8126
8127 else
8128 {
8129 if ((escape == ESC_b || escape == ESC_B || escape == ESC_A) &&
8130 cd->max_lookbehind == 0)
8131 cd->max_lookbehind = 1;
8132 #ifdef SUPPORT_UCP
8133 if (escape >= ESC_DU && escape <= ESC_wu)
8134 {
8135 nestptr = ptr + 1; /* Where to resume */
8136 ptr = substitutes[escape - ESC_DU] - 1; /* Just before substitute */
8137 }
8138 else
8139 #endif
8140 /* In non-UTF-8 mode, we turn \C into OP_ALLANY instead of OP_ANYBYTE
8141 so that it works in DFA mode and in lookbehinds. */
8142
8143 {
8144 previous = (escape > ESC_b && escape < ESC_Z)? code : NULL;
8145 item_hwm_offset = cd->hwm - cd->start_workspace;
8146 *code++ = (!utf && escape == ESC_C)? OP_ALLANY : escape;
8147 }
8148 }
8149 continue;
8150 }
8151
8152 /* We have a data character whose value is in c. In UTF-8 mode it may have
8153 a value > 127. We set its representation in the length/buffer, and then
8154 handle it as a data character. */
8155
8156 #if defined SUPPORT_UTF && !defined COMPILE_PCRE32
8157 if (utf && c > MAX_VALUE_FOR_SINGLE_CHAR)
8158 mclength = PRIV(ord2utf)(c, mcbuffer);
8159 else
8160 #endif
8161
8162 {
8163 mcbuffer[0] = c;
8164 mclength = 1;
8165 }
8166 goto ONE_CHAR;
8167
8168
8169 /* ===================================================================*/
8170 /* Handle a literal character. It is guaranteed not to be whitespace or #
8171 when the extended flag is set. If we are in a UTF mode, it may be a
8172 multi-unit literal character. */
8173
8174 default:
8175 NORMAL_CHAR:
8176 mclength = 1;
8177 mcbuffer[0] = c;
8178
8179 #ifdef SUPPORT_UTF
8180 if (utf && HAS_EXTRALEN(c))
8181 ACROSSCHAR(TRUE, ptr[1], mcbuffer[mclength++] = *(++ptr));
8182 #endif
8183
8184 /* At this point we have the character's bytes in mcbuffer, and the length
8185 in mclength. When not in UTF-8 mode, the length is always 1. */
8186
8187 ONE_CHAR:
8188 previous = code;
8189 item_hwm_offset = cd->hwm - cd->start_workspace;
8190
8191 /* For caseless UTF-8 mode when UCP support is available, check whether
8192 this character has more than one other case. If so, generate a special
8193 OP_PROP item instead of OP_CHARI. */
8194
8195 #ifdef SUPPORT_UCP
8196 if (utf && (options & PCRE_CASELESS) != 0)
8197 {
8198 GETCHAR(c, mcbuffer);
8199 if ((c = UCD_CASESET(c)) != 0)
8200 {
8201 *code++ = OP_PROP;
8202 *code++ = PT_CLIST;
8203 *code++ = c;
8204 if (firstcharflags == REQ_UNSET)
8205 firstcharflags = zerofirstcharflags = REQ_NONE;
8206 break;
8207 }
8208 }
8209 #endif
8210
8211 /* Caseful matches, or not one of the multicase characters. */
8212
8213 *code++ = ((options & PCRE_CASELESS) != 0)? OP_CHARI : OP_CHAR;
8214 for (c = 0; c < mclength; c++) *code++ = mcbuffer[c];
8215
8216 /* Remember if \r or \n were seen */
8217
8218 if (mcbuffer[0] == CHAR_CR || mcbuffer[0] == CHAR_NL)
8219 cd->external_flags |= PCRE_HASCRORLF;
8220
8221 /* Set the first and required bytes appropriately. If no previous first
8222 byte, set it from this character, but revert to none on a zero repeat.
8223 Otherwise, leave the firstchar value alone, and don't change it on a zero
8224 repeat. */
8225
8226 if (firstcharflags == REQ_UNSET)
8227 {
8228 zerofirstcharflags = REQ_NONE;
8229 zeroreqchar = reqchar;
8230 zeroreqcharflags = reqcharflags;
8231
8232 /* If the character is more than one byte long, we can set firstchar
8233 only if it is not to be matched caselessly. */
8234
8235 if (mclength == 1 || req_caseopt == 0)
8236 {
8237 firstchar = mcbuffer[0];
8238 firstcharflags = req_caseopt;
8239
8240 if (mclength != 1)
8241 {
8242 reqchar = code[-1];
8243 reqcharflags = cd->req_varyopt;
8244 }
8245 }
8246 else firstcharflags = reqcharflags = REQ_NONE;
8247 }
8248
8249 /* firstchar was previously set; we can set reqchar only if the length is
8250 1 or the matching is caseful. */
8251
8252 else
8253 {
8254 zerofirstchar = firstchar;
8255 zerofirstcharflags = firstcharflags;
8256 zeroreqchar = reqchar;
8257 zeroreqcharflags = reqcharflags;
8258 if (mclength == 1 || req_caseopt == 0)
8259 {
8260 reqchar = code[-1];
8261 reqcharflags = req_caseopt | cd->req_varyopt;
8262 }
8263 }
8264
8265 break; /* End of literal character handling */
8266 }
8267 } /* end of big loop */
8268
8269
8270 /* Control never reaches here by falling through, only by a goto for all the
8271 error states. Pass back the position in the pattern so that it can be displayed
8272 to the user for diagnosing the error. */
8273
8274 FAILED:
8275 *ptrptr = ptr;
8276 return FALSE;
8277 }
8278
8279
8280
8281 /*************************************************
8282 * Compile sequence of alternatives *
8283 *************************************************/
8284
8285 /* On entry, ptr is pointing past the bracket character, but on return it
8286 points to the closing bracket, or vertical bar, or end of string. The code
8287 variable is pointing at the byte into which the BRA operator has been stored.
8288 This function is used during the pre-compile phase when we are trying to find
8289 out the amount of memory needed, as well as during the real compile phase. The
8290 value of lengthptr distinguishes the two phases.
8291
8292 Arguments:
8293 options option bits, including any changes for this subpattern
8294 codeptr -> the address of the current code pointer
8295 ptrptr -> the address of the current pattern pointer
8296 errorcodeptr -> pointer to error code variable
8297 lookbehind TRUE if this is a lookbehind assertion
8298 reset_bracount TRUE to reset the count for each branch
8299 skipbytes skip this many bytes at start (for brackets and OP_COND)
8300 cond_depth depth of nesting for conditional subpatterns
8301 firstcharptr place to put the first required character
8302 firstcharflagsptr place to put the first character flags, or a negative number
8303 reqcharptr place to put the last required character
8304 reqcharflagsptr place to put the last required character flags, or a negative number
8305 bcptr pointer to the chain of currently open branches
8306 cd points to the data block with tables pointers etc.
8307 lengthptr NULL during the real compile phase
8308 points to length accumulator during pre-compile phase
8309
8310 Returns: TRUE on success
8311 */
8312
8313 static BOOL
compile_regex(int options,pcre_uchar ** codeptr,const pcre_uchar ** ptrptr,int * errorcodeptr,BOOL lookbehind,BOOL reset_bracount,int skipbytes,int cond_depth,pcre_uint32 * firstcharptr,pcre_int32 * firstcharflagsptr,pcre_uint32 * reqcharptr,pcre_int32 * reqcharflagsptr,branch_chain * bcptr,compile_data * cd,int * lengthptr)8314 compile_regex(int options, pcre_uchar **codeptr, const pcre_uchar **ptrptr,
8315 int *errorcodeptr, BOOL lookbehind, BOOL reset_bracount, int skipbytes,
8316 int cond_depth,
8317 pcre_uint32 *firstcharptr, pcre_int32 *firstcharflagsptr,
8318 pcre_uint32 *reqcharptr, pcre_int32 *reqcharflagsptr,
8319 branch_chain *bcptr, compile_data *cd, int *lengthptr)
8320 {
8321 const pcre_uchar *ptr = *ptrptr;
8322 pcre_uchar *code = *codeptr;
8323 pcre_uchar *last_branch = code;
8324 pcre_uchar *start_bracket = code;
8325 pcre_uchar *reverse_count = NULL;
8326 open_capitem capitem;
8327 int capnumber = 0;
8328 pcre_uint32 firstchar, reqchar;
8329 pcre_int32 firstcharflags, reqcharflags;
8330 pcre_uint32 branchfirstchar, branchreqchar;
8331 pcre_int32 branchfirstcharflags, branchreqcharflags;
8332 int length;
8333 unsigned int orig_bracount;
8334 unsigned int max_bracount;
8335 branch_chain bc;
8336 size_t save_hwm_offset;
8337
8338 /* If set, call the external function that checks for stack availability. */
8339
8340 if (PUBL(stack_guard) != NULL && PUBL(stack_guard)())
8341 {
8342 *errorcodeptr= ERR85;
8343 return FALSE;
8344 }
8345
8346 /* Miscellaneous initialization */
8347
8348 bc.outer = bcptr;
8349 bc.current_branch = code;
8350
8351 firstchar = reqchar = 0;
8352 firstcharflags = reqcharflags = REQ_UNSET;
8353
8354 save_hwm_offset = cd->hwm - cd->start_workspace;
8355
8356 /* Accumulate the length for use in the pre-compile phase. Start with the
8357 length of the BRA and KET and any extra bytes that are required at the
8358 beginning. We accumulate in a local variable to save frequent testing of
8359 lenthptr for NULL. We cannot do this by looking at the value of code at the
8360 start and end of each alternative, because compiled items are discarded during
8361 the pre-compile phase so that the work space is not exceeded. */
8362
8363 length = 2 + 2*LINK_SIZE + skipbytes;
8364
8365 /* WARNING: If the above line is changed for any reason, you must also change
8366 the code that abstracts option settings at the start of the pattern and makes
8367 them global. It tests the value of length for (2 + 2*LINK_SIZE) in the
8368 pre-compile phase to find out whether anything has yet been compiled or not. */
8369
8370 /* If this is a capturing subpattern, add to the chain of open capturing items
8371 so that we can detect them if (*ACCEPT) is encountered. This is also used to
8372 detect groups that contain recursive back references to themselves. Note that
8373 only OP_CBRA need be tested here; changing this opcode to one of its variants,
8374 e.g. OP_SCBRAPOS, happens later, after the group has been compiled. */
8375
8376 if (*code == OP_CBRA)
8377 {
8378 capnumber = GET2(code, 1 + LINK_SIZE);
8379 capitem.number = capnumber;
8380 capitem.next = cd->open_caps;
8381 capitem.flag = FALSE;
8382 cd->open_caps = &capitem;
8383 }
8384
8385 /* Offset is set zero to mark that this bracket is still open */
8386
8387 PUT(code, 1, 0);
8388 code += 1 + LINK_SIZE + skipbytes;
8389
8390 /* Loop for each alternative branch */
8391
8392 orig_bracount = max_bracount = cd->bracount;
8393 for (;;)
8394 {
8395 /* For a (?| group, reset the capturing bracket count so that each branch
8396 uses the same numbers. */
8397
8398 if (reset_bracount) cd->bracount = orig_bracount;
8399
8400 /* Set up dummy OP_REVERSE if lookbehind assertion */
8401
8402 if (lookbehind)
8403 {
8404 *code++ = OP_REVERSE;
8405 reverse_count = code;
8406 PUTINC(code, 0, 0);
8407 length += 1 + LINK_SIZE;
8408 }
8409
8410 /* Now compile the branch; in the pre-compile phase its length gets added
8411 into the length. */
8412
8413 if (!compile_branch(&options, &code, &ptr, errorcodeptr, &branchfirstchar,
8414 &branchfirstcharflags, &branchreqchar, &branchreqcharflags, &bc,
8415 cond_depth, cd, (lengthptr == NULL)? NULL : &length))
8416 {
8417 *ptrptr = ptr;
8418 return FALSE;
8419 }
8420
8421 /* Keep the highest bracket count in case (?| was used and some branch
8422 has fewer than the rest. */
8423
8424 if (cd->bracount > max_bracount) max_bracount = cd->bracount;
8425
8426 /* In the real compile phase, there is some post-processing to be done. */
8427
8428 if (lengthptr == NULL)
8429 {
8430 /* If this is the first branch, the firstchar and reqchar values for the
8431 branch become the values for the regex. */
8432
8433 if (*last_branch != OP_ALT)
8434 {
8435 firstchar = branchfirstchar;
8436 firstcharflags = branchfirstcharflags;
8437 reqchar = branchreqchar;
8438 reqcharflags = branchreqcharflags;
8439 }
8440
8441 /* If this is not the first branch, the first char and reqchar have to
8442 match the values from all the previous branches, except that if the
8443 previous value for reqchar didn't have REQ_VARY set, it can still match,
8444 and we set REQ_VARY for the regex. */
8445
8446 else
8447 {
8448 /* If we previously had a firstchar, but it doesn't match the new branch,
8449 we have to abandon the firstchar for the regex, but if there was
8450 previously no reqchar, it takes on the value of the old firstchar. */
8451
8452 if (firstcharflags >= 0 &&
8453 (firstcharflags != branchfirstcharflags || firstchar != branchfirstchar))
8454 {
8455 if (reqcharflags < 0)
8456 {
8457 reqchar = firstchar;
8458 reqcharflags = firstcharflags;
8459 }
8460 firstcharflags = REQ_NONE;
8461 }
8462
8463 /* If we (now or from before) have no firstchar, a firstchar from the
8464 branch becomes a reqchar if there isn't a branch reqchar. */
8465
8466 if (firstcharflags < 0 && branchfirstcharflags >= 0 && branchreqcharflags < 0)
8467 {
8468 branchreqchar = branchfirstchar;
8469 branchreqcharflags = branchfirstcharflags;
8470 }
8471
8472 /* Now ensure that the reqchars match */
8473
8474 if (((reqcharflags & ~REQ_VARY) != (branchreqcharflags & ~REQ_VARY)) ||
8475 reqchar != branchreqchar)
8476 reqcharflags = REQ_NONE;
8477 else
8478 {
8479 reqchar = branchreqchar;
8480 reqcharflags |= branchreqcharflags; /* To "or" REQ_VARY */
8481 }
8482 }
8483
8484 /* If lookbehind, check that this branch matches a fixed-length string, and
8485 put the length into the OP_REVERSE item. Temporarily mark the end of the
8486 branch with OP_END. If the branch contains OP_RECURSE, the result is -3
8487 because there may be forward references that we can't check here. Set a
8488 flag to cause another lookbehind check at the end. Why not do it all at the
8489 end? Because common, erroneous checks are picked up here and the offset of
8490 the problem can be shown. */
8491
8492 if (lookbehind)
8493 {
8494 int fixed_length;
8495 *code = OP_END;
8496 fixed_length = find_fixedlength(last_branch, (options & PCRE_UTF8) != 0,
8497 FALSE, cd, NULL);
8498 DPRINTF(("fixed length = %d\n", fixed_length));
8499 if (fixed_length == -3)
8500 {
8501 cd->check_lookbehind = TRUE;
8502 }
8503 else if (fixed_length < 0)
8504 {
8505 *errorcodeptr = (fixed_length == -2)? ERR36 :
8506 (fixed_length == -4)? ERR70: ERR25;
8507 *ptrptr = ptr;
8508 return FALSE;
8509 }
8510 else
8511 {
8512 if (fixed_length > cd->max_lookbehind)
8513 cd->max_lookbehind = fixed_length;
8514 PUT(reverse_count, 0, fixed_length);
8515 }
8516 }
8517 }
8518
8519 /* Reached end of expression, either ')' or end of pattern. In the real
8520 compile phase, go back through the alternative branches and reverse the chain
8521 of offsets, with the field in the BRA item now becoming an offset to the
8522 first alternative. If there are no alternatives, it points to the end of the
8523 group. The length in the terminating ket is always the length of the whole
8524 bracketed item. Return leaving the pointer at the terminating char. */
8525
8526 if (*ptr != CHAR_VERTICAL_LINE)
8527 {
8528 if (lengthptr == NULL)
8529 {
8530 int branch_length = (int)(code - last_branch);
8531 do
8532 {
8533 int prev_length = GET(last_branch, 1);
8534 PUT(last_branch, 1, branch_length);
8535 branch_length = prev_length;
8536 last_branch -= branch_length;
8537 }
8538 while (branch_length > 0);
8539 }
8540
8541 /* Fill in the ket */
8542
8543 *code = OP_KET;
8544 PUT(code, 1, (int)(code - start_bracket));
8545 code += 1 + LINK_SIZE;
8546
8547 /* If it was a capturing subpattern, check to see if it contained any
8548 recursive back references. If so, we must wrap it in atomic brackets.
8549 Because we are moving code along, we must ensure that any pending recursive
8550 references are updated. In any event, remove the block from the chain. */
8551
8552 if (capnumber > 0)
8553 {
8554 if (cd->open_caps->flag)
8555 {
8556 *code = OP_END;
8557 adjust_recurse(start_bracket, 1 + LINK_SIZE,
8558 (options & PCRE_UTF8) != 0, cd, save_hwm_offset);
8559 memmove(start_bracket + 1 + LINK_SIZE, start_bracket,
8560 IN_UCHARS(code - start_bracket));
8561 *start_bracket = OP_ONCE;
8562 code += 1 + LINK_SIZE;
8563 PUT(start_bracket, 1, (int)(code - start_bracket));
8564 *code = OP_KET;
8565 PUT(code, 1, (int)(code - start_bracket));
8566 code += 1 + LINK_SIZE;
8567 length += 2 + 2*LINK_SIZE;
8568 }
8569 cd->open_caps = cd->open_caps->next;
8570 }
8571
8572 /* Retain the highest bracket number, in case resetting was used. */
8573
8574 cd->bracount = max_bracount;
8575
8576 /* Set values to pass back */
8577
8578 *codeptr = code;
8579 *ptrptr = ptr;
8580 *firstcharptr = firstchar;
8581 *firstcharflagsptr = firstcharflags;
8582 *reqcharptr = reqchar;
8583 *reqcharflagsptr = reqcharflags;
8584 if (lengthptr != NULL)
8585 {
8586 if (OFLOW_MAX - *lengthptr < length)
8587 {
8588 *errorcodeptr = ERR20;
8589 return FALSE;
8590 }
8591 *lengthptr += length;
8592 }
8593 return TRUE;
8594 }
8595
8596 /* Another branch follows. In the pre-compile phase, we can move the code
8597 pointer back to where it was for the start of the first branch. (That is,
8598 pretend that each branch is the only one.)
8599
8600 In the real compile phase, insert an ALT node. Its length field points back
8601 to the previous branch while the bracket remains open. At the end the chain
8602 is reversed. It's done like this so that the start of the bracket has a
8603 zero offset until it is closed, making it possible to detect recursion. */
8604
8605 if (lengthptr != NULL)
8606 {
8607 code = *codeptr + 1 + LINK_SIZE + skipbytes;
8608 length += 1 + LINK_SIZE;
8609 }
8610 else
8611 {
8612 *code = OP_ALT;
8613 PUT(code, 1, (int)(code - last_branch));
8614 bc.current_branch = last_branch = code;
8615 code += 1 + LINK_SIZE;
8616 }
8617
8618 ptr++;
8619 }
8620 /* Control never reaches here */
8621 }
8622
8623
8624
8625
8626 /*************************************************
8627 * Check for anchored expression *
8628 *************************************************/
8629
8630 /* Try to find out if this is an anchored regular expression. Consider each
8631 alternative branch. If they all start with OP_SOD or OP_CIRC, or with a bracket
8632 all of whose alternatives start with OP_SOD or OP_CIRC (recurse ad lib), then
8633 it's anchored. However, if this is a multiline pattern, then only OP_SOD will
8634 be found, because ^ generates OP_CIRCM in that mode.
8635
8636 We can also consider a regex to be anchored if OP_SOM starts all its branches.
8637 This is the code for \G, which means "match at start of match position, taking
8638 into account the match offset".
8639
8640 A branch is also implicitly anchored if it starts with .* and DOTALL is set,
8641 because that will try the rest of the pattern at all possible matching points,
8642 so there is no point trying again.... er ....
8643
8644 .... except when the .* appears inside capturing parentheses, and there is a
8645 subsequent back reference to those parentheses. We haven't enough information
8646 to catch that case precisely.
8647
8648 At first, the best we could do was to detect when .* was in capturing brackets
8649 and the highest back reference was greater than or equal to that level.
8650 However, by keeping a bitmap of the first 31 back references, we can catch some
8651 of the more common cases more precisely.
8652
8653 ... A second exception is when the .* appears inside an atomic group, because
8654 this prevents the number of characters it matches from being adjusted.
8655
8656 Arguments:
8657 code points to start of expression (the bracket)
8658 bracket_map a bitmap of which brackets we are inside while testing; this
8659 handles up to substring 31; after that we just have to take
8660 the less precise approach
8661 cd points to the compile data block
8662 atomcount atomic group level
8663
8664 Returns: TRUE or FALSE
8665 */
8666
8667 static BOOL
is_anchored(register const pcre_uchar * code,unsigned int bracket_map,compile_data * cd,int atomcount)8668 is_anchored(register const pcre_uchar *code, unsigned int bracket_map,
8669 compile_data *cd, int atomcount)
8670 {
8671 do {
8672 const pcre_uchar *scode = first_significant_code(
8673 code + PRIV(OP_lengths)[*code], FALSE);
8674 register int op = *scode;
8675
8676 /* Non-capturing brackets */
8677
8678 if (op == OP_BRA || op == OP_BRAPOS ||
8679 op == OP_SBRA || op == OP_SBRAPOS)
8680 {
8681 if (!is_anchored(scode, bracket_map, cd, atomcount)) return FALSE;
8682 }
8683
8684 /* Capturing brackets */
8685
8686 else if (op == OP_CBRA || op == OP_CBRAPOS ||
8687 op == OP_SCBRA || op == OP_SCBRAPOS)
8688 {
8689 int n = GET2(scode, 1+LINK_SIZE);
8690 int new_map = bracket_map | ((n < 32)? (1U << n) : 1);
8691 if (!is_anchored(scode, new_map, cd, atomcount)) return FALSE;
8692 }
8693
8694 /* Positive forward assertion */
8695
8696 else if (op == OP_ASSERT)
8697 {
8698 if (!is_anchored(scode, bracket_map, cd, atomcount)) return FALSE;
8699 }
8700
8701 /* Condition; not anchored if no second branch */
8702
8703 else if (op == OP_COND)
8704 {
8705 if (scode[GET(scode,1)] != OP_ALT) return FALSE;
8706 if (!is_anchored(scode, bracket_map, cd, atomcount)) return FALSE;
8707 }
8708
8709 /* Atomic groups */
8710
8711 else if (op == OP_ONCE || op == OP_ONCE_NC)
8712 {
8713 if (!is_anchored(scode, bracket_map, cd, atomcount + 1))
8714 return FALSE;
8715 }
8716
8717 /* .* is not anchored unless DOTALL is set (which generates OP_ALLANY) and
8718 it isn't in brackets that are or may be referenced or inside an atomic
8719 group. */
8720
8721 else if ((op == OP_TYPESTAR || op == OP_TYPEMINSTAR ||
8722 op == OP_TYPEPOSSTAR))
8723 {
8724 if (scode[1] != OP_ALLANY || (bracket_map & cd->backref_map) != 0 ||
8725 atomcount > 0 || cd->had_pruneorskip)
8726 return FALSE;
8727 }
8728
8729 /* Check for explicit anchoring */
8730
8731 else if (op != OP_SOD && op != OP_SOM && op != OP_CIRC) return FALSE;
8732
8733 code += GET(code, 1);
8734 }
8735 while (*code == OP_ALT); /* Loop for each alternative */
8736 return TRUE;
8737 }
8738
8739
8740
8741 /*************************************************
8742 * Check for starting with ^ or .* *
8743 *************************************************/
8744
8745 /* This is called to find out if every branch starts with ^ or .* so that
8746 "first char" processing can be done to speed things up in multiline
8747 matching and for non-DOTALL patterns that start with .* (which must start at
8748 the beginning or after \n). As in the case of is_anchored() (see above), we
8749 have to take account of back references to capturing brackets that contain .*
8750 because in that case we can't make the assumption. Also, the appearance of .*
8751 inside atomic brackets or in an assertion, or in a pattern that contains *PRUNE
8752 or *SKIP does not count, because once again the assumption no longer holds.
8753
8754 Arguments:
8755 code points to start of expression (the bracket)
8756 bracket_map a bitmap of which brackets we are inside while testing; this
8757 handles up to substring 31; after that we just have to take
8758 the less precise approach
8759 cd points to the compile data
8760 atomcount atomic group level
8761 inassert TRUE if in an assertion
8762
8763 Returns: TRUE or FALSE
8764 */
8765
8766 static BOOL
is_startline(const pcre_uchar * code,unsigned int bracket_map,compile_data * cd,int atomcount,BOOL inassert)8767 is_startline(const pcre_uchar *code, unsigned int bracket_map,
8768 compile_data *cd, int atomcount, BOOL inassert)
8769 {
8770 do {
8771 const pcre_uchar *scode = first_significant_code(
8772 code + PRIV(OP_lengths)[*code], FALSE);
8773 register int op = *scode;
8774
8775 /* If we are at the start of a conditional assertion group, *both* the
8776 conditional assertion *and* what follows the condition must satisfy the test
8777 for start of line. Other kinds of condition fail. Note that there may be an
8778 auto-callout at the start of a condition. */
8779
8780 if (op == OP_COND)
8781 {
8782 scode += 1 + LINK_SIZE;
8783 if (*scode == OP_CALLOUT) scode += PRIV(OP_lengths)[OP_CALLOUT];
8784 switch (*scode)
8785 {
8786 case OP_CREF:
8787 case OP_DNCREF:
8788 case OP_RREF:
8789 case OP_DNRREF:
8790 case OP_DEF:
8791 case OP_FAIL:
8792 return FALSE;
8793
8794 default: /* Assertion */
8795 if (!is_startline(scode, bracket_map, cd, atomcount, TRUE)) return FALSE;
8796 do scode += GET(scode, 1); while (*scode == OP_ALT);
8797 scode += 1 + LINK_SIZE;
8798 break;
8799 }
8800 scode = first_significant_code(scode, FALSE);
8801 op = *scode;
8802 }
8803
8804 /* Non-capturing brackets */
8805
8806 if (op == OP_BRA || op == OP_BRAPOS ||
8807 op == OP_SBRA || op == OP_SBRAPOS)
8808 {
8809 if (!is_startline(scode, bracket_map, cd, atomcount, inassert)) return FALSE;
8810 }
8811
8812 /* Capturing brackets */
8813
8814 else if (op == OP_CBRA || op == OP_CBRAPOS ||
8815 op == OP_SCBRA || op == OP_SCBRAPOS)
8816 {
8817 int n = GET2(scode, 1+LINK_SIZE);
8818 int new_map = bracket_map | ((n < 32)? (1U << n) : 1);
8819 if (!is_startline(scode, new_map, cd, atomcount, inassert)) return FALSE;
8820 }
8821
8822 /* Positive forward assertions */
8823
8824 else if (op == OP_ASSERT)
8825 {
8826 if (!is_startline(scode, bracket_map, cd, atomcount, TRUE)) return FALSE;
8827 }
8828
8829 /* Atomic brackets */
8830
8831 else if (op == OP_ONCE || op == OP_ONCE_NC)
8832 {
8833 if (!is_startline(scode, bracket_map, cd, atomcount + 1, inassert)) return FALSE;
8834 }
8835
8836 /* .* means "start at start or after \n" if it isn't in atomic brackets or
8837 brackets that may be referenced or an assertion, as long as the pattern does
8838 not contain *PRUNE or *SKIP, because these break the feature. Consider, for
8839 example, /.*?a(*PRUNE)b/ with the subject "aab", which matches "ab", i.e.
8840 not at the start of a line. */
8841
8842 else if (op == OP_TYPESTAR || op == OP_TYPEMINSTAR || op == OP_TYPEPOSSTAR)
8843 {
8844 if (scode[1] != OP_ANY || (bracket_map & cd->backref_map) != 0 ||
8845 atomcount > 0 || cd->had_pruneorskip || inassert)
8846 return FALSE;
8847 }
8848
8849 /* Check for explicit circumflex; anything else gives a FALSE result. Note
8850 in particular that this includes atomic brackets OP_ONCE and OP_ONCE_NC
8851 because the number of characters matched by .* cannot be adjusted inside
8852 them. */
8853
8854 else if (op != OP_CIRC && op != OP_CIRCM) return FALSE;
8855
8856 /* Move on to the next alternative */
8857
8858 code += GET(code, 1);
8859 }
8860 while (*code == OP_ALT); /* Loop for each alternative */
8861 return TRUE;
8862 }
8863
8864
8865
8866 /*************************************************
8867 * Check for asserted fixed first char *
8868 *************************************************/
8869
8870 /* During compilation, the "first char" settings from forward assertions are
8871 discarded, because they can cause conflicts with actual literals that follow.
8872 However, if we end up without a first char setting for an unanchored pattern,
8873 it is worth scanning the regex to see if there is an initial asserted first
8874 char. If all branches start with the same asserted char, or with a
8875 non-conditional bracket all of whose alternatives start with the same asserted
8876 char (recurse ad lib), then we return that char, with the flags set to zero or
8877 REQ_CASELESS; otherwise return zero with REQ_NONE in the flags.
8878
8879 Arguments:
8880 code points to start of expression (the bracket)
8881 flags points to the first char flags, or to REQ_NONE
8882 inassert TRUE if in an assertion
8883
8884 Returns: the fixed first char, or 0 with REQ_NONE in flags
8885 */
8886
8887 static pcre_uint32
find_firstassertedchar(const pcre_uchar * code,pcre_int32 * flags,BOOL inassert)8888 find_firstassertedchar(const pcre_uchar *code, pcre_int32 *flags,
8889 BOOL inassert)
8890 {
8891 register pcre_uint32 c = 0;
8892 int cflags = REQ_NONE;
8893
8894 *flags = REQ_NONE;
8895 do {
8896 pcre_uint32 d;
8897 int dflags;
8898 int xl = (*code == OP_CBRA || *code == OP_SCBRA ||
8899 *code == OP_CBRAPOS || *code == OP_SCBRAPOS)? IMM2_SIZE:0;
8900 const pcre_uchar *scode = first_significant_code(code + 1+LINK_SIZE + xl,
8901 TRUE);
8902 register pcre_uchar op = *scode;
8903
8904 switch(op)
8905 {
8906 default:
8907 return 0;
8908
8909 case OP_BRA:
8910 case OP_BRAPOS:
8911 case OP_CBRA:
8912 case OP_SCBRA:
8913 case OP_CBRAPOS:
8914 case OP_SCBRAPOS:
8915 case OP_ASSERT:
8916 case OP_ONCE:
8917 case OP_ONCE_NC:
8918 d = find_firstassertedchar(scode, &dflags, op == OP_ASSERT);
8919 if (dflags < 0)
8920 return 0;
8921 if (cflags < 0) { c = d; cflags = dflags; } else if (c != d || cflags != dflags) return 0;
8922 break;
8923
8924 case OP_EXACT:
8925 scode += IMM2_SIZE;
8926 /* Fall through */
8927
8928 case OP_CHAR:
8929 case OP_PLUS:
8930 case OP_MINPLUS:
8931 case OP_POSPLUS:
8932 if (!inassert) return 0;
8933 if (cflags < 0) { c = scode[1]; cflags = 0; }
8934 else if (c != scode[1]) return 0;
8935 break;
8936
8937 case OP_EXACTI:
8938 scode += IMM2_SIZE;
8939 /* Fall through */
8940
8941 case OP_CHARI:
8942 case OP_PLUSI:
8943 case OP_MINPLUSI:
8944 case OP_POSPLUSI:
8945 if (!inassert) return 0;
8946 if (cflags < 0) { c = scode[1]; cflags = REQ_CASELESS; }
8947 else if (c != scode[1]) return 0;
8948 break;
8949 }
8950
8951 code += GET(code, 1);
8952 }
8953 while (*code == OP_ALT);
8954
8955 *flags = cflags;
8956 return c;
8957 }
8958
8959
8960
8961 /*************************************************
8962 * Add an entry to the name/number table *
8963 *************************************************/
8964
8965 /* This function is called between compiling passes to add an entry to the
8966 name/number table, maintaining alphabetical order. Checking for permitted
8967 and forbidden duplicates has already been done.
8968
8969 Arguments:
8970 cd the compile data block
8971 name the name to add
8972 length the length of the name
8973 groupno the group number
8974
8975 Returns: nothing
8976 */
8977
8978 static void
add_name(compile_data * cd,const pcre_uchar * name,int length,unsigned int groupno)8979 add_name(compile_data *cd, const pcre_uchar *name, int length,
8980 unsigned int groupno)
8981 {
8982 int i;
8983 pcre_uchar *slot = cd->name_table;
8984
8985 for (i = 0; i < cd->names_found; i++)
8986 {
8987 int crc = memcmp(name, slot+IMM2_SIZE, IN_UCHARS(length));
8988 if (crc == 0 && slot[IMM2_SIZE+length] != 0)
8989 crc = -1; /* Current name is a substring */
8990
8991 /* Make space in the table and break the loop for an earlier name. For a
8992 duplicate or later name, carry on. We do this for duplicates so that in the
8993 simple case (when ?(| is not used) they are in order of their numbers. In all
8994 cases they are in the order in which they appear in the pattern. */
8995
8996 if (crc < 0)
8997 {
8998 memmove(slot + cd->name_entry_size, slot,
8999 IN_UCHARS((cd->names_found - i) * cd->name_entry_size));
9000 break;
9001 }
9002
9003 /* Continue the loop for a later or duplicate name */
9004
9005 slot += cd->name_entry_size;
9006 }
9007
9008 PUT2(slot, 0, groupno);
9009 memcpy(slot + IMM2_SIZE, name, IN_UCHARS(length));
9010 slot[IMM2_SIZE + length] = 0;
9011 cd->names_found++;
9012 }
9013
9014
9015
9016 /*************************************************
9017 * Compile a Regular Expression *
9018 *************************************************/
9019
9020 /* This function takes a string and returns a pointer to a block of store
9021 holding a compiled version of the expression. The original API for this
9022 function had no error code return variable; it is retained for backwards
9023 compatibility. The new function is given a new name.
9024
9025 Arguments:
9026 pattern the regular expression
9027 options various option bits
9028 errorcodeptr pointer to error code variable (pcre_compile2() only)
9029 can be NULL if you don't want a code value
9030 errorptr pointer to pointer to error text
9031 erroroffset ptr offset in pattern where error was detected
9032 tables pointer to character tables or NULL
9033
9034 Returns: pointer to compiled data block, or NULL on error,
9035 with errorptr and erroroffset set
9036 */
9037
9038 #if defined COMPILE_PCRE8
9039 PCRE_EXP_DEFN pcre * PCRE_CALL_CONVENTION
pcre_compile(const char * pattern,int options,const char ** errorptr,int * erroroffset,const unsigned char * tables)9040 pcre_compile(const char *pattern, int options, const char **errorptr,
9041 int *erroroffset, const unsigned char *tables)
9042 #elif defined COMPILE_PCRE16
9043 PCRE_EXP_DEFN pcre16 * PCRE_CALL_CONVENTION
9044 pcre16_compile(PCRE_SPTR16 pattern, int options, const char **errorptr,
9045 int *erroroffset, const unsigned char *tables)
9046 #elif defined COMPILE_PCRE32
9047 PCRE_EXP_DEFN pcre32 * PCRE_CALL_CONVENTION
9048 pcre32_compile(PCRE_SPTR32 pattern, int options, const char **errorptr,
9049 int *erroroffset, const unsigned char *tables)
9050 #endif
9051 {
9052 #if defined COMPILE_PCRE8
9053 return pcre_compile2(pattern, options, NULL, errorptr, erroroffset, tables);
9054 #elif defined COMPILE_PCRE16
9055 return pcre16_compile2(pattern, options, NULL, errorptr, erroroffset, tables);
9056 #elif defined COMPILE_PCRE32
9057 return pcre32_compile2(pattern, options, NULL, errorptr, erroroffset, tables);
9058 #endif
9059 }
9060
9061
9062 #if defined COMPILE_PCRE8
9063 PCRE_EXP_DEFN pcre * PCRE_CALL_CONVENTION
pcre_compile2(const char * pattern,int options,int * errorcodeptr,const char ** errorptr,int * erroroffset,const unsigned char * tables)9064 pcre_compile2(const char *pattern, int options, int *errorcodeptr,
9065 const char **errorptr, int *erroroffset, const unsigned char *tables)
9066 #elif defined COMPILE_PCRE16
9067 PCRE_EXP_DEFN pcre16 * PCRE_CALL_CONVENTION
9068 pcre16_compile2(PCRE_SPTR16 pattern, int options, int *errorcodeptr,
9069 const char **errorptr, int *erroroffset, const unsigned char *tables)
9070 #elif defined COMPILE_PCRE32
9071 PCRE_EXP_DEFN pcre32 * PCRE_CALL_CONVENTION
9072 pcre32_compile2(PCRE_SPTR32 pattern, int options, int *errorcodeptr,
9073 const char **errorptr, int *erroroffset, const unsigned char *tables)
9074 #endif
9075 {
9076 REAL_PCRE *re;
9077 int length = 1; /* For final END opcode */
9078 pcre_int32 firstcharflags, reqcharflags;
9079 pcre_uint32 firstchar, reqchar;
9080 pcre_uint32 limit_match = PCRE_UINT32_MAX;
9081 pcre_uint32 limit_recursion = PCRE_UINT32_MAX;
9082 int newline;
9083 int errorcode = 0;
9084 int skipatstart = 0;
9085 BOOL utf;
9086 BOOL never_utf = FALSE;
9087 size_t size;
9088 pcre_uchar *code;
9089 const pcre_uchar *codestart;
9090 const pcre_uchar *ptr;
9091 compile_data compile_block;
9092 compile_data *cd = &compile_block;
9093
9094 /* This space is used for "compiling" into during the first phase, when we are
9095 computing the amount of memory that is needed. Compiled items are thrown away
9096 as soon as possible, so that a fairly large buffer should be sufficient for
9097 this purpose. The same space is used in the second phase for remembering where
9098 to fill in forward references to subpatterns. That may overflow, in which case
9099 new memory is obtained from malloc(). */
9100
9101 pcre_uchar cworkspace[COMPILE_WORK_SIZE];
9102
9103 /* This vector is used for remembering name groups during the pre-compile. In a
9104 similar way to cworkspace, it can be expanded using malloc() if necessary. */
9105
9106 named_group named_groups[NAMED_GROUP_LIST_SIZE];
9107
9108 /* Set this early so that early errors get offset 0. */
9109
9110 ptr = (const pcre_uchar *)pattern;
9111
9112 /* We can't pass back an error message if errorptr is NULL; I guess the best we
9113 can do is just return NULL, but we can set a code value if there is a code
9114 pointer. */
9115
9116 if (errorptr == NULL)
9117 {
9118 if (errorcodeptr != NULL) *errorcodeptr = 99;
9119 return NULL;
9120 }
9121
9122 *errorptr = NULL;
9123 if (errorcodeptr != NULL) *errorcodeptr = ERR0;
9124
9125 /* However, we can give a message for this error */
9126
9127 if (erroroffset == NULL)
9128 {
9129 errorcode = ERR16;
9130 goto PCRE_EARLY_ERROR_RETURN2;
9131 }
9132
9133 *erroroffset = 0;
9134
9135 /* Set up pointers to the individual character tables */
9136
9137 if (tables == NULL) tables = PRIV(default_tables);
9138 cd->lcc = tables + lcc_offset;
9139 cd->fcc = tables + fcc_offset;
9140 cd->cbits = tables + cbits_offset;
9141 cd->ctypes = tables + ctypes_offset;
9142
9143 /* Check that all undefined public option bits are zero */
9144
9145 if ((options & ~PUBLIC_COMPILE_OPTIONS) != 0)
9146 {
9147 errorcode = ERR17;
9148 goto PCRE_EARLY_ERROR_RETURN;
9149 }
9150
9151 /* If PCRE_NEVER_UTF is set, remember it. */
9152
9153 if ((options & PCRE_NEVER_UTF) != 0) never_utf = TRUE;
9154
9155 /* Check for global one-time settings at the start of the pattern, and remember
9156 the offset for later. */
9157
9158 cd->external_flags = 0; /* Initialize here for LIMIT_MATCH/RECURSION */
9159
9160 while (ptr[skipatstart] == CHAR_LEFT_PARENTHESIS &&
9161 ptr[skipatstart+1] == CHAR_ASTERISK)
9162 {
9163 int newnl = 0;
9164 int newbsr = 0;
9165
9166 /* For completeness and backward compatibility, (*UTFn) is supported in the
9167 relevant libraries, but (*UTF) is generic and always supported. Note that
9168 PCRE_UTF8 == PCRE_UTF16 == PCRE_UTF32. */
9169
9170 #ifdef COMPILE_PCRE8
9171 if (STRNCMP_UC_C8(ptr+skipatstart+2, STRING_UTF8_RIGHTPAR, 5) == 0)
9172 { skipatstart += 7; options |= PCRE_UTF8; continue; }
9173 #endif
9174 #ifdef COMPILE_PCRE16
9175 if (STRNCMP_UC_C8(ptr+skipatstart+2, STRING_UTF16_RIGHTPAR, 6) == 0)
9176 { skipatstart += 8; options |= PCRE_UTF16; continue; }
9177 #endif
9178 #ifdef COMPILE_PCRE32
9179 if (STRNCMP_UC_C8(ptr+skipatstart+2, STRING_UTF32_RIGHTPAR, 6) == 0)
9180 { skipatstart += 8; options |= PCRE_UTF32; continue; }
9181 #endif
9182
9183 else if (STRNCMP_UC_C8(ptr+skipatstart+2, STRING_UTF_RIGHTPAR, 4) == 0)
9184 { skipatstart += 6; options |= PCRE_UTF8; continue; }
9185 else if (STRNCMP_UC_C8(ptr+skipatstart+2, STRING_UCP_RIGHTPAR, 4) == 0)
9186 { skipatstart += 6; options |= PCRE_UCP; continue; }
9187 else if (STRNCMP_UC_C8(ptr+skipatstart+2, STRING_NO_AUTO_POSSESS_RIGHTPAR, 16) == 0)
9188 { skipatstart += 18; options |= PCRE_NO_AUTO_POSSESS; continue; }
9189 else if (STRNCMP_UC_C8(ptr+skipatstart+2, STRING_NO_START_OPT_RIGHTPAR, 13) == 0)
9190 { skipatstart += 15; options |= PCRE_NO_START_OPTIMIZE; continue; }
9191
9192 else if (STRNCMP_UC_C8(ptr+skipatstart+2, STRING_LIMIT_MATCH_EQ, 12) == 0)
9193 {
9194 pcre_uint32 c = 0;
9195 int p = skipatstart + 14;
9196 while (isdigit(ptr[p]))
9197 {
9198 if (c > PCRE_UINT32_MAX / 10 - 1) break; /* Integer overflow */
9199 c = c*10 + ptr[p++] - CHAR_0;
9200 }
9201 if (ptr[p++] != CHAR_RIGHT_PARENTHESIS) break;
9202 if (c < limit_match)
9203 {
9204 limit_match = c;
9205 cd->external_flags |= PCRE_MLSET;
9206 }
9207 skipatstart = p;
9208 continue;
9209 }
9210
9211 else if (STRNCMP_UC_C8(ptr+skipatstart+2, STRING_LIMIT_RECURSION_EQ, 16) == 0)
9212 {
9213 pcre_uint32 c = 0;
9214 int p = skipatstart + 18;
9215 while (isdigit(ptr[p]))
9216 {
9217 if (c > PCRE_UINT32_MAX / 10 - 1) break; /* Integer overflow check */
9218 c = c*10 + ptr[p++] - CHAR_0;
9219 }
9220 if (ptr[p++] != CHAR_RIGHT_PARENTHESIS) break;
9221 if (c < limit_recursion)
9222 {
9223 limit_recursion = c;
9224 cd->external_flags |= PCRE_RLSET;
9225 }
9226 skipatstart = p;
9227 continue;
9228 }
9229
9230 if (STRNCMP_UC_C8(ptr+skipatstart+2, STRING_CR_RIGHTPAR, 3) == 0)
9231 { skipatstart += 5; newnl = PCRE_NEWLINE_CR; }
9232 else if (STRNCMP_UC_C8(ptr+skipatstart+2, STRING_LF_RIGHTPAR, 3) == 0)
9233 { skipatstart += 5; newnl = PCRE_NEWLINE_LF; }
9234 else if (STRNCMP_UC_C8(ptr+skipatstart+2, STRING_CRLF_RIGHTPAR, 5) == 0)
9235 { skipatstart += 7; newnl = PCRE_NEWLINE_CR + PCRE_NEWLINE_LF; }
9236 else if (STRNCMP_UC_C8(ptr+skipatstart+2, STRING_ANY_RIGHTPAR, 4) == 0)
9237 { skipatstart += 6; newnl = PCRE_NEWLINE_ANY; }
9238 else if (STRNCMP_UC_C8(ptr+skipatstart+2, STRING_ANYCRLF_RIGHTPAR, 8) == 0)
9239 { skipatstart += 10; newnl = PCRE_NEWLINE_ANYCRLF; }
9240
9241 else if (STRNCMP_UC_C8(ptr+skipatstart+2, STRING_BSR_ANYCRLF_RIGHTPAR, 12) == 0)
9242 { skipatstart += 14; newbsr = PCRE_BSR_ANYCRLF; }
9243 else if (STRNCMP_UC_C8(ptr+skipatstart+2, STRING_BSR_UNICODE_RIGHTPAR, 12) == 0)
9244 { skipatstart += 14; newbsr = PCRE_BSR_UNICODE; }
9245
9246 if (newnl != 0)
9247 options = (options & ~PCRE_NEWLINE_BITS) | newnl;
9248 else if (newbsr != 0)
9249 options = (options & ~(PCRE_BSR_ANYCRLF|PCRE_BSR_UNICODE)) | newbsr;
9250 else break;
9251 }
9252
9253 /* PCRE_UTF(16|32) have the same value as PCRE_UTF8. */
9254 utf = (options & PCRE_UTF8) != 0;
9255 if (utf && never_utf)
9256 {
9257 errorcode = ERR78;
9258 goto PCRE_EARLY_ERROR_RETURN2;
9259 }
9260
9261 /* Can't support UTF unless PCRE has been compiled to include the code. The
9262 return of an error code from PRIV(valid_utf)() is a new feature, introduced in
9263 release 8.13. It is passed back from pcre_[dfa_]exec(), but at the moment is
9264 not used here. */
9265
9266 #ifdef SUPPORT_UTF
9267 if (utf && (options & PCRE_NO_UTF8_CHECK) == 0 &&
9268 (errorcode = PRIV(valid_utf)((PCRE_PUCHAR)pattern, -1, erroroffset)) != 0)
9269 {
9270 #if defined COMPILE_PCRE8
9271 errorcode = ERR44;
9272 #elif defined COMPILE_PCRE16
9273 errorcode = ERR74;
9274 #elif defined COMPILE_PCRE32
9275 errorcode = ERR77;
9276 #endif
9277 goto PCRE_EARLY_ERROR_RETURN2;
9278 }
9279 #else
9280 if (utf)
9281 {
9282 errorcode = ERR32;
9283 goto PCRE_EARLY_ERROR_RETURN;
9284 }
9285 #endif
9286
9287 /* Can't support UCP unless PCRE has been compiled to include the code. */
9288
9289 #ifndef SUPPORT_UCP
9290 if ((options & PCRE_UCP) != 0)
9291 {
9292 errorcode = ERR67;
9293 goto PCRE_EARLY_ERROR_RETURN;
9294 }
9295 #endif
9296
9297 /* Check validity of \R options. */
9298
9299 if ((options & (PCRE_BSR_ANYCRLF|PCRE_BSR_UNICODE)) ==
9300 (PCRE_BSR_ANYCRLF|PCRE_BSR_UNICODE))
9301 {
9302 errorcode = ERR56;
9303 goto PCRE_EARLY_ERROR_RETURN;
9304 }
9305
9306 /* Handle different types of newline. The three bits give seven cases. The
9307 current code allows for fixed one- or two-byte sequences, plus "any" and
9308 "anycrlf". */
9309
9310 switch (options & PCRE_NEWLINE_BITS)
9311 {
9312 case 0: newline = NEWLINE; break; /* Build-time default */
9313 case PCRE_NEWLINE_CR: newline = CHAR_CR; break;
9314 case PCRE_NEWLINE_LF: newline = CHAR_NL; break;
9315 case PCRE_NEWLINE_CR+
9316 PCRE_NEWLINE_LF: newline = (CHAR_CR << 8) | CHAR_NL; break;
9317 case PCRE_NEWLINE_ANY: newline = -1; break;
9318 case PCRE_NEWLINE_ANYCRLF: newline = -2; break;
9319 default: errorcode = ERR56; goto PCRE_EARLY_ERROR_RETURN;
9320 }
9321
9322 if (newline == -2)
9323 {
9324 cd->nltype = NLTYPE_ANYCRLF;
9325 }
9326 else if (newline < 0)
9327 {
9328 cd->nltype = NLTYPE_ANY;
9329 }
9330 else
9331 {
9332 cd->nltype = NLTYPE_FIXED;
9333 if (newline > 255)
9334 {
9335 cd->nllen = 2;
9336 cd->nl[0] = (newline >> 8) & 255;
9337 cd->nl[1] = newline & 255;
9338 }
9339 else
9340 {
9341 cd->nllen = 1;
9342 cd->nl[0] = newline;
9343 }
9344 }
9345
9346 /* Maximum back reference and backref bitmap. The bitmap records up to 31 back
9347 references to help in deciding whether (.*) can be treated as anchored or not.
9348 */
9349
9350 cd->top_backref = 0;
9351 cd->backref_map = 0;
9352
9353 /* Reflect pattern for debugging output */
9354
9355 DPRINTF(("------------------------------------------------------------------\n"));
9356 #ifdef PCRE_DEBUG
9357 print_puchar(stdout, (PCRE_PUCHAR)pattern);
9358 #endif
9359 DPRINTF(("\n"));
9360
9361 /* Pretend to compile the pattern while actually just accumulating the length
9362 of memory required. This behaviour is triggered by passing a non-NULL final
9363 argument to compile_regex(). We pass a block of workspace (cworkspace) for it
9364 to compile parts of the pattern into; the compiled code is discarded when it is
9365 no longer needed, so hopefully this workspace will never overflow, though there
9366 is a test for its doing so. */
9367
9368 cd->bracount = cd->final_bracount = 0;
9369 cd->names_found = 0;
9370 cd->name_entry_size = 0;
9371 cd->name_table = NULL;
9372 cd->dupnames = FALSE;
9373 cd->dupgroups = FALSE;
9374 cd->namedrefcount = 0;
9375 cd->start_code = cworkspace;
9376 cd->hwm = cworkspace;
9377 cd->iscondassert = FALSE;
9378 cd->start_workspace = cworkspace;
9379 cd->workspace_size = COMPILE_WORK_SIZE;
9380 cd->named_groups = named_groups;
9381 cd->named_group_list_size = NAMED_GROUP_LIST_SIZE;
9382 cd->start_pattern = (const pcre_uchar *)pattern;
9383 cd->end_pattern = (const pcre_uchar *)(pattern + STRLEN_UC((const pcre_uchar *)pattern));
9384 cd->req_varyopt = 0;
9385 cd->parens_depth = 0;
9386 cd->assert_depth = 0;
9387 cd->max_lookbehind = 0;
9388 cd->external_options = options;
9389 cd->open_caps = NULL;
9390
9391 /* Now do the pre-compile. On error, errorcode will be set non-zero, so we
9392 don't need to look at the result of the function here. The initial options have
9393 been put into the cd block so that they can be changed if an option setting is
9394 found within the regex right at the beginning. Bringing initial option settings
9395 outside can help speed up starting point checks. */
9396
9397 ptr += skipatstart;
9398 code = cworkspace;
9399 *code = OP_BRA;
9400
9401 (void)compile_regex(cd->external_options, &code, &ptr, &errorcode, FALSE,
9402 FALSE, 0, 0, &firstchar, &firstcharflags, &reqchar, &reqcharflags, NULL,
9403 cd, &length);
9404 if (errorcode != 0) goto PCRE_EARLY_ERROR_RETURN;
9405
9406 DPRINTF(("end pre-compile: length=%d workspace=%d\n", length,
9407 (int)(cd->hwm - cworkspace)));
9408
9409 if (length > MAX_PATTERN_SIZE)
9410 {
9411 errorcode = ERR20;
9412 goto PCRE_EARLY_ERROR_RETURN;
9413 }
9414
9415 /* Compute the size of the data block for storing the compiled pattern. Integer
9416 overflow should no longer be possible because nowadays we limit the maximum
9417 value of cd->names_found and cd->name_entry_size. */
9418
9419 size = sizeof(REAL_PCRE) +
9420 (length + cd->names_found * cd->name_entry_size) * sizeof(pcre_uchar);
9421
9422 /* Get the memory. */
9423
9424 re = (REAL_PCRE *)(PUBL(malloc))(size);
9425 if (re == NULL)
9426 {
9427 errorcode = ERR21;
9428 goto PCRE_EARLY_ERROR_RETURN;
9429 }
9430
9431 /* Put in the magic number, and save the sizes, initial options, internal
9432 flags, and character table pointer. NULL is used for the default character
9433 tables. The nullpad field is at the end; it's there to help in the case when a
9434 regex compiled on a system with 4-byte pointers is run on another with 8-byte
9435 pointers. */
9436
9437 re->magic_number = MAGIC_NUMBER;
9438 re->size = (int)size;
9439 re->options = cd->external_options;
9440 re->flags = cd->external_flags;
9441 re->limit_match = limit_match;
9442 re->limit_recursion = limit_recursion;
9443 re->first_char = 0;
9444 re->req_char = 0;
9445 re->name_table_offset = sizeof(REAL_PCRE) / sizeof(pcre_uchar);
9446 re->name_entry_size = cd->name_entry_size;
9447 re->name_count = cd->names_found;
9448 re->ref_count = 0;
9449 re->tables = (tables == PRIV(default_tables))? NULL : tables;
9450 re->nullpad = NULL;
9451 #ifdef COMPILE_PCRE32
9452 re->dummy = 0;
9453 #else
9454 re->dummy1 = re->dummy2 = re->dummy3 = 0;
9455 #endif
9456
9457 /* The starting points of the name/number translation table and of the code are
9458 passed around in the compile data block. The start/end pattern and initial
9459 options are already set from the pre-compile phase, as is the name_entry_size
9460 field. Reset the bracket count and the names_found field. Also reset the hwm
9461 field; this time it's used for remembering forward references to subpatterns.
9462 */
9463
9464 cd->final_bracount = cd->bracount; /* Save for checking forward references */
9465 cd->parens_depth = 0;
9466 cd->assert_depth = 0;
9467 cd->bracount = 0;
9468 cd->max_lookbehind = 0;
9469 cd->name_table = (pcre_uchar *)re + re->name_table_offset;
9470 codestart = cd->name_table + re->name_entry_size * re->name_count;
9471 cd->start_code = codestart;
9472 cd->hwm = (pcre_uchar *)(cd->start_workspace);
9473 cd->iscondassert = FALSE;
9474 cd->req_varyopt = 0;
9475 cd->had_accept = FALSE;
9476 cd->had_pruneorskip = FALSE;
9477 cd->check_lookbehind = FALSE;
9478 cd->open_caps = NULL;
9479
9480 /* If any named groups were found, create the name/number table from the list
9481 created in the first pass. */
9482
9483 if (cd->names_found > 0)
9484 {
9485 int i = cd->names_found;
9486 named_group *ng = cd->named_groups;
9487 cd->names_found = 0;
9488 for (; i > 0; i--, ng++)
9489 add_name(cd, ng->name, ng->length, ng->number);
9490 if (cd->named_group_list_size > NAMED_GROUP_LIST_SIZE)
9491 (PUBL(free))((void *)cd->named_groups);
9492 }
9493
9494 /* Set up a starting, non-extracting bracket, then compile the expression. On
9495 error, errorcode will be set non-zero, so we don't need to look at the result
9496 of the function here. */
9497
9498 ptr = (const pcre_uchar *)pattern + skipatstart;
9499 code = (pcre_uchar *)codestart;
9500 *code = OP_BRA;
9501 (void)compile_regex(re->options, &code, &ptr, &errorcode, FALSE, FALSE, 0, 0,
9502 &firstchar, &firstcharflags, &reqchar, &reqcharflags, NULL, cd, NULL);
9503 re->top_bracket = cd->bracount;
9504 re->top_backref = cd->top_backref;
9505 re->max_lookbehind = cd->max_lookbehind;
9506 re->flags = cd->external_flags | PCRE_MODE;
9507
9508 if (cd->had_accept)
9509 {
9510 reqchar = 0; /* Must disable after (*ACCEPT) */
9511 reqcharflags = REQ_NONE;
9512 }
9513
9514 /* If not reached end of pattern on success, there's an excess bracket. */
9515
9516 if (errorcode == 0 && *ptr != CHAR_NULL) errorcode = ERR22;
9517
9518 /* Fill in the terminating state and check for disastrous overflow, but
9519 if debugging, leave the test till after things are printed out. */
9520
9521 *code++ = OP_END;
9522
9523 #ifndef PCRE_DEBUG
9524 if (code - codestart > length) errorcode = ERR23;
9525 #endif
9526
9527 #ifdef SUPPORT_VALGRIND
9528 /* If the estimated length exceeds the really used length, mark the extra
9529 allocated memory as unaddressable, so that any out-of-bound reads can be
9530 detected. */
9531 VALGRIND_MAKE_MEM_NOACCESS(code, (length - (code - codestart)) * sizeof(pcre_uchar));
9532 #endif
9533
9534 /* Fill in any forward references that are required. There may be repeated
9535 references; optimize for them, as searching a large regex takes time. */
9536
9537 if (cd->hwm > cd->start_workspace)
9538 {
9539 int prev_recno = -1;
9540 const pcre_uchar *groupptr = NULL;
9541 while (errorcode == 0 && cd->hwm > cd->start_workspace)
9542 {
9543 int offset, recno;
9544 cd->hwm -= LINK_SIZE;
9545 offset = GET(cd->hwm, 0);
9546
9547 /* Check that the hwm handling hasn't gone wrong. This whole area is
9548 rewritten in PCRE2 because there are some obscure cases. */
9549
9550 if (offset == 0 || codestart[offset-1] != OP_RECURSE)
9551 {
9552 errorcode = ERR10;
9553 break;
9554 }
9555
9556 recno = GET(codestart, offset);
9557 if (recno != prev_recno)
9558 {
9559 groupptr = PRIV(find_bracket)(codestart, utf, recno);
9560 prev_recno = recno;
9561 }
9562 if (groupptr == NULL) errorcode = ERR53;
9563 else PUT(((pcre_uchar *)codestart), offset, (int)(groupptr - codestart));
9564 }
9565 }
9566
9567 /* If the workspace had to be expanded, free the new memory. Set the pointer to
9568 NULL to indicate that forward references have been filled in. */
9569
9570 if (cd->workspace_size > COMPILE_WORK_SIZE)
9571 (PUBL(free))((void *)cd->start_workspace);
9572 cd->start_workspace = NULL;
9573
9574 /* Give an error if there's back reference to a non-existent capturing
9575 subpattern. */
9576
9577 if (errorcode == 0 && re->top_backref > re->top_bracket) errorcode = ERR15;
9578
9579 /* Unless disabled, check whether any single character iterators can be
9580 auto-possessified. The function overwrites the appropriate opcode values, so
9581 the type of the pointer must be cast. NOTE: the intermediate variable "temp" is
9582 used in this code because at least one compiler gives a warning about loss of
9583 "const" attribute if the cast (pcre_uchar *)codestart is used directly in the
9584 function call. */
9585
9586 if (errorcode == 0 && (options & PCRE_NO_AUTO_POSSESS) == 0)
9587 {
9588 pcre_uchar *temp = (pcre_uchar *)codestart;
9589 auto_possessify(temp, utf, cd);
9590 }
9591
9592 /* If there were any lookbehind assertions that contained OP_RECURSE
9593 (recursions or subroutine calls), a flag is set for them to be checked here,
9594 because they may contain forward references. Actual recursions cannot be fixed
9595 length, but subroutine calls can. It is done like this so that those without
9596 OP_RECURSE that are not fixed length get a diagnosic with a useful offset. The
9597 exceptional ones forgo this. We scan the pattern to check that they are fixed
9598 length, and set their lengths. */
9599
9600 if (errorcode == 0 && cd->check_lookbehind)
9601 {
9602 pcre_uchar *cc = (pcre_uchar *)codestart;
9603
9604 /* Loop, searching for OP_REVERSE items, and process those that do not have
9605 their length set. (Actually, it will also re-process any that have a length
9606 of zero, but that is a pathological case, and it does no harm.) When we find
9607 one, we temporarily terminate the branch it is in while we scan it. */
9608
9609 for (cc = (pcre_uchar *)PRIV(find_bracket)(codestart, utf, -1);
9610 cc != NULL;
9611 cc = (pcre_uchar *)PRIV(find_bracket)(cc, utf, -1))
9612 {
9613 if (GET(cc, 1) == 0)
9614 {
9615 int fixed_length;
9616 pcre_uchar *be = cc - 1 - LINK_SIZE + GET(cc, -LINK_SIZE);
9617 int end_op = *be;
9618 *be = OP_END;
9619 fixed_length = find_fixedlength(cc, (re->options & PCRE_UTF8) != 0, TRUE,
9620 cd, NULL);
9621 *be = end_op;
9622 DPRINTF(("fixed length = %d\n", fixed_length));
9623 if (fixed_length < 0)
9624 {
9625 errorcode = (fixed_length == -2)? ERR36 :
9626 (fixed_length == -4)? ERR70 : ERR25;
9627 break;
9628 }
9629 if (fixed_length > cd->max_lookbehind) cd->max_lookbehind = fixed_length;
9630 PUT(cc, 1, fixed_length);
9631 }
9632 cc += 1 + LINK_SIZE;
9633 }
9634 }
9635
9636 /* Failed to compile, or error while post-processing */
9637
9638 if (errorcode != 0)
9639 {
9640 (PUBL(free))(re);
9641 PCRE_EARLY_ERROR_RETURN:
9642 *erroroffset = (int)(ptr - (const pcre_uchar *)pattern);
9643 PCRE_EARLY_ERROR_RETURN2:
9644 *errorptr = find_error_text(errorcode);
9645 if (errorcodeptr != NULL) *errorcodeptr = errorcode;
9646 return NULL;
9647 }
9648
9649 /* If the anchored option was not passed, set the flag if we can determine that
9650 the pattern is anchored by virtue of ^ characters or \A or anything else, such
9651 as starting with non-atomic .* when DOTALL is set and there are no occurrences
9652 of *PRUNE or *SKIP.
9653
9654 Otherwise, if we know what the first byte has to be, save it, because that
9655 speeds up unanchored matches no end. If not, see if we can set the
9656 PCRE_STARTLINE flag. This is helpful for multiline matches when all branches
9657 start with ^. and also when all branches start with non-atomic .* for
9658 non-DOTALL matches when *PRUNE and SKIP are not present. */
9659
9660 if ((re->options & PCRE_ANCHORED) == 0)
9661 {
9662 if (is_anchored(codestart, 0, cd, 0)) re->options |= PCRE_ANCHORED;
9663 else
9664 {
9665 if (firstcharflags < 0)
9666 firstchar = find_firstassertedchar(codestart, &firstcharflags, FALSE);
9667 if (firstcharflags >= 0) /* Remove caseless flag for non-caseable chars */
9668 {
9669 #if defined COMPILE_PCRE8
9670 re->first_char = firstchar & 0xff;
9671 #elif defined COMPILE_PCRE16
9672 re->first_char = firstchar & 0xffff;
9673 #elif defined COMPILE_PCRE32
9674 re->first_char = firstchar;
9675 #endif
9676 if ((firstcharflags & REQ_CASELESS) != 0)
9677 {
9678 #if defined SUPPORT_UCP && !(defined COMPILE_PCRE8)
9679 /* We ignore non-ASCII first chars in 8 bit mode. */
9680 if (utf)
9681 {
9682 if (re->first_char < 128)
9683 {
9684 if (cd->fcc[re->first_char] != re->first_char)
9685 re->flags |= PCRE_FCH_CASELESS;
9686 }
9687 else if (UCD_OTHERCASE(re->first_char) != re->first_char)
9688 re->flags |= PCRE_FCH_CASELESS;
9689 }
9690 else
9691 #endif
9692 if (MAX_255(re->first_char)
9693 && cd->fcc[re->first_char] != re->first_char)
9694 re->flags |= PCRE_FCH_CASELESS;
9695 }
9696
9697 re->flags |= PCRE_FIRSTSET;
9698 }
9699
9700 else if (is_startline(codestart, 0, cd, 0, FALSE)) re->flags |= PCRE_STARTLINE;
9701 }
9702 }
9703
9704 /* For an anchored pattern, we use the "required byte" only if it follows a
9705 variable length item in the regex. Remove the caseless flag for non-caseable
9706 bytes. */
9707
9708 if (reqcharflags >= 0 &&
9709 ((re->options & PCRE_ANCHORED) == 0 || (reqcharflags & REQ_VARY) != 0))
9710 {
9711 #if defined COMPILE_PCRE8
9712 re->req_char = reqchar & 0xff;
9713 #elif defined COMPILE_PCRE16
9714 re->req_char = reqchar & 0xffff;
9715 #elif defined COMPILE_PCRE32
9716 re->req_char = reqchar;
9717 #endif
9718 if ((reqcharflags & REQ_CASELESS) != 0)
9719 {
9720 #if defined SUPPORT_UCP && !(defined COMPILE_PCRE8)
9721 /* We ignore non-ASCII first chars in 8 bit mode. */
9722 if (utf)
9723 {
9724 if (re->req_char < 128)
9725 {
9726 if (cd->fcc[re->req_char] != re->req_char)
9727 re->flags |= PCRE_RCH_CASELESS;
9728 }
9729 else if (UCD_OTHERCASE(re->req_char) != re->req_char)
9730 re->flags |= PCRE_RCH_CASELESS;
9731 }
9732 else
9733 #endif
9734 if (MAX_255(re->req_char) && cd->fcc[re->req_char] != re->req_char)
9735 re->flags |= PCRE_RCH_CASELESS;
9736 }
9737
9738 re->flags |= PCRE_REQCHSET;
9739 }
9740
9741 /* Print out the compiled data if debugging is enabled. This is never the
9742 case when building a production library. */
9743
9744 #ifdef PCRE_DEBUG
9745 printf("Length = %d top_bracket = %d top_backref = %d\n",
9746 length, re->top_bracket, re->top_backref);
9747
9748 printf("Options=%08x\n", re->options);
9749
9750 if ((re->flags & PCRE_FIRSTSET) != 0)
9751 {
9752 pcre_uchar ch = re->first_char;
9753 const char *caseless =
9754 ((re->flags & PCRE_FCH_CASELESS) == 0)? "" : " (caseless)";
9755 if (PRINTABLE(ch)) printf("First char = %c%s\n", ch, caseless);
9756 else printf("First char = \\x%02x%s\n", ch, caseless);
9757 }
9758
9759 if ((re->flags & PCRE_REQCHSET) != 0)
9760 {
9761 pcre_uchar ch = re->req_char;
9762 const char *caseless =
9763 ((re->flags & PCRE_RCH_CASELESS) == 0)? "" : " (caseless)";
9764 if (PRINTABLE(ch)) printf("Req char = %c%s\n", ch, caseless);
9765 else printf("Req char = \\x%02x%s\n", ch, caseless);
9766 }
9767
9768 #if defined COMPILE_PCRE8
9769 pcre_printint((pcre *)re, stdout, TRUE);
9770 #elif defined COMPILE_PCRE16
9771 pcre16_printint((pcre *)re, stdout, TRUE);
9772 #elif defined COMPILE_PCRE32
9773 pcre32_printint((pcre *)re, stdout, TRUE);
9774 #endif
9775
9776 /* This check is done here in the debugging case so that the code that
9777 was compiled can be seen. */
9778
9779 if (code - codestart > length)
9780 {
9781 (PUBL(free))(re);
9782 *errorptr = find_error_text(ERR23);
9783 *erroroffset = ptr - (pcre_uchar *)pattern;
9784 if (errorcodeptr != NULL) *errorcodeptr = ERR23;
9785 return NULL;
9786 }
9787 #endif /* PCRE_DEBUG */
9788
9789 /* Check for a pattern than can match an empty string, so that this information
9790 can be provided to applications. */
9791
9792 do
9793 {
9794 if (could_be_empty_branch(codestart, code, utf, cd, NULL))
9795 {
9796 re->flags |= PCRE_MATCH_EMPTY;
9797 break;
9798 }
9799 codestart += GET(codestart, 1);
9800 }
9801 while (*codestart == OP_ALT);
9802
9803 #if defined COMPILE_PCRE8
9804 return (pcre *)re;
9805 #elif defined COMPILE_PCRE16
9806 return (pcre16 *)re;
9807 #elif defined COMPILE_PCRE32
9808 return (pcre32 *)re;
9809 #endif
9810 }
9811
9812 /* End of pcre_compile.c */
9813