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 /* %ExternalCopyright% */
45
46 #ifdef HAVE_CONFIG_H
47 #include "config.h"
48 #endif
49
50 #define NLBLOCK cd /* Block containing newline information */
51 #define PSSTART start_pattern /* Field containing pattern start */
52 #define PSEND end_pattern /* Field containing pattern end */
53
54 #include "pcre_internal.h"
55
56
57 /* When PCRE_DEBUG is defined, we need the pcre(16|32)_printint() function, which
58 is also used by pcretest. PCRE_DEBUG is not defined when building a production
59 library. We do not need to select pcre16_printint.c specially, because the
60 COMPILE_PCREx macro will already be appropriately set. */
61
62 #ifdef PCRE_DEBUG
63 /* pcre_printint.c should not include any headers */
64 #define PCRE_INCLUDED
65 #include "pcre_printint.c"
66 #undef PCRE_INCLUDED
67 #endif
68
69
70 /* Macro for setting individual bits in class bitmaps. */
71
72 #define SETBIT(a,b) a[(b)/8] |= (1U << ((b)&7))
73
74 /* Maximum length value to check against when making sure that the integer that
75 holds the compiled pattern length does not overflow. We make it a bit less than
76 INT_MAX to allow for adding in group terminating bytes, so that we don't have
77 to check them every time. */
78
79 #define OFLOW_MAX (INT_MAX - 20)
80
81 /* Definitions to allow mutual recursion */
82
83 static int
84 add_list_to_class(pcre_uint8 *, pcre_uchar **, int, compile_data *,
85 const pcre_uint32 *, unsigned int);
86
87 static BOOL
88 compile_regex(int, pcre_uchar **, const pcre_uchar **, int *, BOOL, BOOL, int, int,
89 pcre_uint32 *, pcre_int32 *, pcre_uint32 *, pcre_int32 *, branch_chain *,
90 compile_data *, int *);
91
92
93
94 /*************************************************
95 * Code parameters and static tables *
96 *************************************************/
97
98 /* This value specifies the size of stack workspace that is used during the
99 first pre-compile phase that determines how much memory is required. The regex
100 is partly compiled into this space, but the compiled parts are discarded as
101 soon as they can be, so that hopefully there will never be an overrun. The code
102 does, however, check for an overrun. The largest amount I've seen used is 218,
103 so this number is very generous.
104
105 The same workspace is used during the second, actual compile phase for
106 remembering forward references to groups so that they can be filled in at the
107 end. Each entry in this list occupies LINK_SIZE bytes, so even when LINK_SIZE
108 is 4 there is plenty of room for most patterns. However, the memory can get
109 filled up by repetitions of forward references, for example patterns like
110 /(?1){0,1999}(b)/, and one user did hit the limit. The code has been changed so
111 that the workspace is expanded using malloc() in this situation. The value
112 below is therefore a minimum, and we put a maximum on it for safety. The
113 minimum is now also defined in terms of LINK_SIZE so that the use of malloc()
114 kicks in at the same number of forward references in all cases. */
115
116 #define COMPILE_WORK_SIZE (2048*LINK_SIZE)
117 #define COMPILE_WORK_SIZE_MAX (100*COMPILE_WORK_SIZE)
118
119 /* This value determines the size of the initial vector that is used for
120 remembering named groups during the pre-compile. It is allocated on the stack,
121 but if it is too small, it is expanded using malloc(), in a similar way to the
122 workspace. The value is the number of slots in the list. */
123
124 #define NAMED_GROUP_LIST_SIZE 20
125
126 /* The overrun tests check for a slightly smaller size so that they detect the
127 overrun before it actually does run off the end of the data block. */
128
129 #define WORK_SIZE_SAFETY_MARGIN (100)
130
131 /* Private flags added to firstchar and reqchar. */
132
133 #define REQ_CASELESS (1U << 0) /* Indicates caselessness */
134 #define REQ_VARY (1U << 1) /* Reqchar followed non-literal item */
135 /* Negative values for the firstchar and reqchar flags */
136 #define REQ_UNSET (-2)
137 #define REQ_NONE (-1)
138
139 /* Repeated character flags. */
140
141 #define UTF_LENGTH 0x10000000l /* The char contains its length. */
142
143 /* Table for handling escaped characters in the range '0'-'z'. Positive returns
144 are simple data values; negative values are for special things like \d and so
145 on. Zero means further processing is needed (for things like \x), or the escape
146 is invalid. */
147
148 #ifndef EBCDIC
149
150 /* This is the "normal" table for ASCII systems or for EBCDIC systems running
151 in UTF-8 mode. */
152
153 static const short int escapes[] = {
154 0, 0,
155 0, 0,
156 0, 0,
157 0, 0,
158 0, 0,
159 CHAR_COLON, CHAR_SEMICOLON,
160 CHAR_LESS_THAN_SIGN, CHAR_EQUALS_SIGN,
161 CHAR_GREATER_THAN_SIGN, CHAR_QUESTION_MARK,
162 CHAR_COMMERCIAL_AT, -ESC_A,
163 -ESC_B, -ESC_C,
164 -ESC_D, -ESC_E,
165 0, -ESC_G,
166 -ESC_H, 0,
167 0, -ESC_K,
168 0, 0,
169 -ESC_N, 0,
170 -ESC_P, -ESC_Q,
171 -ESC_R, -ESC_S,
172 0, 0,
173 -ESC_V, -ESC_W,
174 -ESC_X, 0,
175 -ESC_Z, CHAR_LEFT_SQUARE_BRACKET,
176 CHAR_BACKSLASH, CHAR_RIGHT_SQUARE_BRACKET,
177 CHAR_CIRCUMFLEX_ACCENT, CHAR_UNDERSCORE,
178 CHAR_GRAVE_ACCENT, ESC_a,
179 -ESC_b, 0,
180 -ESC_d, ESC_e,
181 ESC_f, 0,
182 -ESC_h, 0,
183 0, -ESC_k,
184 0, 0,
185 ESC_n, 0,
186 -ESC_p, 0,
187 ESC_r, -ESC_s,
188 ESC_tee, 0,
189 -ESC_v, -ESC_w,
190 0, 0,
191 -ESC_z
192 };
193
194 #else
195
196 /* This is the "abnormal" table for EBCDIC systems without UTF-8 support. */
197
198 static const short int escapes[] = {
199 /* 48 */ 0, 0, 0, '.', '<', '(', '+', '|',
200 /* 50 */ '&', 0, 0, 0, 0, 0, 0, 0,
201 /* 58 */ 0, 0, '!', '$', '*', ')', ';', '~',
202 /* 60 */ '-', '/', 0, 0, 0, 0, 0, 0,
203 /* 68 */ 0, 0, '|', ',', '%', '_', '>', '?',
204 /* 70 */ 0, 0, 0, 0, 0, 0, 0, 0,
205 /* 78 */ 0, '`', ':', '#', '@', '\'', '=', '"',
206 /* 80 */ 0, ESC_a, -ESC_b, 0, -ESC_d, ESC_e, ESC_f, 0,
207 /* 88 */-ESC_h, 0, 0, '{', 0, 0, 0, 0,
208 /* 90 */ 0, 0, -ESC_k, 0, 0, ESC_n, 0, -ESC_p,
209 /* 98 */ 0, ESC_r, 0, '}', 0, 0, 0, 0,
210 /* A0 */ 0, '~', -ESC_s, ESC_tee, 0,-ESC_v, -ESC_w, 0,
211 /* A8 */ 0,-ESC_z, 0, 0, 0, '[', 0, 0,
212 /* B0 */ 0, 0, 0, 0, 0, 0, 0, 0,
213 /* B8 */ 0, 0, 0, 0, 0, ']', '=', '-',
214 /* C0 */ '{',-ESC_A, -ESC_B, -ESC_C, -ESC_D,-ESC_E, 0, -ESC_G,
215 /* C8 */-ESC_H, 0, 0, 0, 0, 0, 0, 0,
216 /* D0 */ '}', 0, -ESC_K, 0, 0,-ESC_N, 0, -ESC_P,
217 /* D8 */-ESC_Q,-ESC_R, 0, 0, 0, 0, 0, 0,
218 /* E0 */ '\\', 0, -ESC_S, 0, 0,-ESC_V, -ESC_W, -ESC_X,
219 /* E8 */ 0,-ESC_Z, 0, 0, 0, 0, 0, 0,
220 /* F0 */ 0, 0, 0, 0, 0, 0, 0, 0,
221 /* F8 */ 0, 0, 0, 0, 0, 0, 0, 0
222 };
223
224 /* We also need a table of characters that may follow \c in an EBCDIC
225 environment for characters 0-31. */
226
227 static unsigned char ebcdic_escape_c[] = "@ABCDEFGHIJKLMNOPQRSTUVWXYZ[\\]^_";
228
229 #endif
230
231
232 /* Table of special "verbs" like (*PRUNE). This is a short table, so it is
233 searched linearly. Put all the names into a single string, in order to reduce
234 the number of relocations when a shared library is dynamically linked. The
235 string is built from string macros so that it works in UTF-8 mode on EBCDIC
236 platforms. */
237
238 typedef struct verbitem {
239 int len; /* Length of verb name */
240 int op; /* Op when no arg, or -1 if arg mandatory */
241 int op_arg; /* Op when arg present, or -1 if not allowed */
242 } verbitem;
243
244 static const char verbnames[] =
245 "\0" /* Empty name is a shorthand for MARK */
246 STRING_MARK0
247 STRING_ACCEPT0
248 STRING_COMMIT0
249 STRING_F0
250 STRING_FAIL0
251 STRING_PRUNE0
252 STRING_SKIP0
253 STRING_THEN;
254
255 static const verbitem verbs[] = {
256 { 0, -1, OP_MARK },
257 { 4, -1, OP_MARK },
258 { 6, OP_ACCEPT, -1 },
259 { 6, OP_COMMIT, -1 },
260 { 1, OP_FAIL, -1 },
261 { 4, OP_FAIL, -1 },
262 { 5, OP_PRUNE, OP_PRUNE_ARG },
263 { 4, OP_SKIP, OP_SKIP_ARG },
264 { 4, OP_THEN, OP_THEN_ARG }
265 };
266
267 static const int verbcount = sizeof(verbs)/sizeof(verbitem);
268
269
270 /* Substitutes for [[:<:]] and [[:>:]], which mean start and end of word in
271 another regex library. */
272
273 static const pcre_uchar sub_start_of_word[] = {
274 CHAR_BACKSLASH, CHAR_b, CHAR_LEFT_PARENTHESIS, CHAR_QUESTION_MARK,
275 CHAR_EQUALS_SIGN, CHAR_BACKSLASH, CHAR_w, CHAR_RIGHT_PARENTHESIS, '\0' };
276
277 static const pcre_uchar sub_end_of_word[] = {
278 CHAR_BACKSLASH, CHAR_b, CHAR_LEFT_PARENTHESIS, CHAR_QUESTION_MARK,
279 CHAR_LESS_THAN_SIGN, CHAR_EQUALS_SIGN, CHAR_BACKSLASH, CHAR_w,
280 CHAR_RIGHT_PARENTHESIS, '\0' };
281
282
283 /* Tables of names of POSIX character classes and their lengths. The names are
284 now all in a single string, to reduce the number of relocations when a shared
285 library is dynamically loaded. The list of lengths is terminated by a zero
286 length entry. The first three must be alpha, lower, upper, as this is assumed
287 for handling case independence. The indices for graph, print, and punct are
288 needed, so identify them. */
289
290 static const char posix_names[] =
291 STRING_alpha0 STRING_lower0 STRING_upper0 STRING_alnum0
292 STRING_ascii0 STRING_blank0 STRING_cntrl0 STRING_digit0
293 STRING_graph0 STRING_print0 STRING_punct0 STRING_space0
294 STRING_word0 STRING_xdigit;
295
296 static const pcre_uint8 posix_name_lengths[] = {
297 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 4, 6, 0 };
298
299 #define PC_GRAPH 8
300 #define PC_PRINT 9
301 #define PC_PUNCT 10
302
303
304 /* Table of class bit maps for each POSIX class. Each class is formed from a
305 base map, with an optional addition or removal of another map. Then, for some
306 classes, there is some additional tweaking: for [:blank:] the vertical space
307 characters are removed, and for [:alpha:] and [:alnum:] the underscore
308 character is removed. The triples in the table consist of the base map offset,
309 second map offset or -1 if no second map, and a non-negative value for map
310 addition or a negative value for map subtraction (if there are two maps). The
311 absolute value of the third field has these meanings: 0 => no tweaking, 1 =>
312 remove vertical space characters, 2 => remove underscore. */
313
314 static const int posix_class_maps[] = {
315 cbit_word, cbit_digit, -2, /* alpha */
316 cbit_lower, -1, 0, /* lower */
317 cbit_upper, -1, 0, /* upper */
318 cbit_word, -1, 2, /* alnum - word without underscore */
319 cbit_print, cbit_cntrl, 0, /* ascii */
320 cbit_space, -1, 1, /* blank - a GNU extension */
321 cbit_cntrl, -1, 0, /* cntrl */
322 cbit_digit, -1, 0, /* digit */
323 cbit_graph, -1, 0, /* graph */
324 cbit_print, -1, 0, /* print */
325 cbit_punct, -1, 0, /* punct */
326 cbit_space, -1, 0, /* space */
327 cbit_word, -1, 0, /* word - a Perl extension */
328 cbit_xdigit,-1, 0 /* xdigit */
329 };
330
331 /* Table of substitutes for \d etc when PCRE_UCP is set. They are replaced by
332 Unicode property escapes. */
333
334 #ifdef SUPPORT_UCP
335 static const pcre_uchar string_PNd[] = {
336 CHAR_BACKSLASH, CHAR_P, CHAR_LEFT_CURLY_BRACKET,
337 CHAR_N, CHAR_d, CHAR_RIGHT_CURLY_BRACKET, '\0' };
338 static const pcre_uchar string_pNd[] = {
339 CHAR_BACKSLASH, CHAR_p, CHAR_LEFT_CURLY_BRACKET,
340 CHAR_N, CHAR_d, CHAR_RIGHT_CURLY_BRACKET, '\0' };
341 static const pcre_uchar string_PXsp[] = {
342 CHAR_BACKSLASH, CHAR_P, CHAR_LEFT_CURLY_BRACKET,
343 CHAR_X, CHAR_s, CHAR_p, CHAR_RIGHT_CURLY_BRACKET, '\0' };
344 static const pcre_uchar string_pXsp[] = {
345 CHAR_BACKSLASH, CHAR_p, CHAR_LEFT_CURLY_BRACKET,
346 CHAR_X, CHAR_s, CHAR_p, CHAR_RIGHT_CURLY_BRACKET, '\0' };
347 static const pcre_uchar string_PXwd[] = {
348 CHAR_BACKSLASH, CHAR_P, CHAR_LEFT_CURLY_BRACKET,
349 CHAR_X, CHAR_w, CHAR_d, CHAR_RIGHT_CURLY_BRACKET, '\0' };
350 static const pcre_uchar string_pXwd[] = {
351 CHAR_BACKSLASH, CHAR_p, CHAR_LEFT_CURLY_BRACKET,
352 CHAR_X, CHAR_w, CHAR_d, CHAR_RIGHT_CURLY_BRACKET, '\0' };
353
354 static const pcre_uchar *substitutes[] = {
355 string_PNd, /* \D */
356 string_pNd, /* \d */
357 string_PXsp, /* \S */ /* Xsp is Perl space, but from 8.34, Perl */
358 string_pXsp, /* \s */ /* space and POSIX space are the same. */
359 string_PXwd, /* \W */
360 string_pXwd /* \w */
361 };
362
363 /* The POSIX class substitutes must be in the order of the POSIX class names,
364 defined above, and there are both positive and negative cases. NULL means no
365 general substitute of a Unicode property escape (\p or \P). However, for some
366 POSIX classes (e.g. graph, print, punct) a special property code is compiled
367 directly. */
368
369 static const pcre_uchar string_pL[] = {
370 CHAR_BACKSLASH, CHAR_p, CHAR_LEFT_CURLY_BRACKET,
371 CHAR_L, CHAR_RIGHT_CURLY_BRACKET, '\0' };
372 static const pcre_uchar string_pLl[] = {
373 CHAR_BACKSLASH, CHAR_p, CHAR_LEFT_CURLY_BRACKET,
374 CHAR_L, CHAR_l, CHAR_RIGHT_CURLY_BRACKET, '\0' };
375 static const pcre_uchar string_pLu[] = {
376 CHAR_BACKSLASH, CHAR_p, CHAR_LEFT_CURLY_BRACKET,
377 CHAR_L, CHAR_u, CHAR_RIGHT_CURLY_BRACKET, '\0' };
378 static const pcre_uchar string_pXan[] = {
379 CHAR_BACKSLASH, CHAR_p, CHAR_LEFT_CURLY_BRACKET,
380 CHAR_X, CHAR_a, CHAR_n, CHAR_RIGHT_CURLY_BRACKET, '\0' };
381 static const pcre_uchar string_h[] = {
382 CHAR_BACKSLASH, CHAR_h, '\0' };
383 static const pcre_uchar string_pXps[] = {
384 CHAR_BACKSLASH, CHAR_p, CHAR_LEFT_CURLY_BRACKET,
385 CHAR_X, CHAR_p, CHAR_s, CHAR_RIGHT_CURLY_BRACKET, '\0' };
386 static const pcre_uchar string_PL[] = {
387 CHAR_BACKSLASH, CHAR_P, CHAR_LEFT_CURLY_BRACKET,
388 CHAR_L, CHAR_RIGHT_CURLY_BRACKET, '\0' };
389 static const pcre_uchar string_PLl[] = {
390 CHAR_BACKSLASH, CHAR_P, CHAR_LEFT_CURLY_BRACKET,
391 CHAR_L, CHAR_l, CHAR_RIGHT_CURLY_BRACKET, '\0' };
392 static const pcre_uchar string_PLu[] = {
393 CHAR_BACKSLASH, CHAR_P, CHAR_LEFT_CURLY_BRACKET,
394 CHAR_L, CHAR_u, CHAR_RIGHT_CURLY_BRACKET, '\0' };
395 static const pcre_uchar string_PXan[] = {
396 CHAR_BACKSLASH, CHAR_P, CHAR_LEFT_CURLY_BRACKET,
397 CHAR_X, CHAR_a, CHAR_n, CHAR_RIGHT_CURLY_BRACKET, '\0' };
398 static const pcre_uchar string_H[] = {
399 CHAR_BACKSLASH, CHAR_H, '\0' };
400 static const pcre_uchar string_PXps[] = {
401 CHAR_BACKSLASH, CHAR_P, CHAR_LEFT_CURLY_BRACKET,
402 CHAR_X, CHAR_p, CHAR_s, CHAR_RIGHT_CURLY_BRACKET, '\0' };
403
404 static const pcre_uchar *posix_substitutes[] = {
405 string_pL, /* alpha */
406 string_pLl, /* lower */
407 string_pLu, /* upper */
408 string_pXan, /* alnum */
409 NULL, /* ascii */
410 string_h, /* blank */
411 NULL, /* cntrl */
412 string_pNd, /* digit */
413 NULL, /* graph */
414 NULL, /* print */
415 NULL, /* punct */
416 string_pXps, /* space */ /* Xps is POSIX space, but from 8.34 */
417 string_pXwd, /* word */ /* Perl and POSIX space are the same */
418 NULL, /* xdigit */
419 /* Negated cases */
420 string_PL, /* ^alpha */
421 string_PLl, /* ^lower */
422 string_PLu, /* ^upper */
423 string_PXan, /* ^alnum */
424 NULL, /* ^ascii */
425 string_H, /* ^blank */
426 NULL, /* ^cntrl */
427 string_PNd, /* ^digit */
428 NULL, /* ^graph */
429 NULL, /* ^print */
430 NULL, /* ^punct */
431 string_PXps, /* ^space */ /* Xps is POSIX space, but from 8.34 */
432 string_PXwd, /* ^word */ /* Perl and POSIX space are the same */
433 NULL /* ^xdigit */
434 };
435 #define POSIX_SUBSIZE (sizeof(posix_substitutes) / sizeof(pcre_uchar *))
436 #endif
437
438 #define STRING(a) # a
439 #define XSTRING(s) STRING(s)
440
441 /* The texts of compile-time error messages. These are "char *" because they
442 are passed to the outside world. Do not ever re-use any error number, because
443 they are documented. Always add a new error instead. Messages marked DEAD below
444 are no longer used. This used to be a table of strings, but in order to reduce
445 the number of relocations needed when a shared library is loaded dynamically,
446 it is now one long string. We cannot use a table of offsets, because the
447 lengths of inserts such as XSTRING(MAX_NAME_SIZE) are not known. Instead, we
448 simply count through to the one we want - this isn't a performance issue
449 because these strings are used only when there is a compilation error.
450
451 Each substring ends with \0 to insert a null character. This includes the final
452 substring, so that the whole string ends with \0\0, which can be detected when
453 counting through. */
454
455 static const char error_texts[] =
456 "no error\0"
457 "\\ at end of pattern\0"
458 "\\c at end of pattern\0"
459 "unrecognized character follows \\\0"
460 "numbers out of order in {} quantifier\0"
461 /* 5 */
462 "number too big in {} quantifier\0"
463 "missing terminating ] for character class\0"
464 "invalid escape sequence in character class\0"
465 "range out of order in character class\0"
466 "nothing to repeat\0"
467 /* 10 */
468 "internal error: invalid forward reference offset\0"
469 "internal error: unexpected repeat\0"
470 "unrecognized character after (? or (?-\0"
471 "POSIX named classes are supported only within a class\0"
472 "missing )\0"
473 /* 15 */
474 "reference to non-existent subpattern\0"
475 "erroffset passed as NULL\0"
476 "unknown option bit(s) set\0"
477 "missing ) after comment\0"
478 "parentheses nested too deeply\0" /** DEAD **/
479 /* 20 */
480 "regular expression is too large\0"
481 "failed to get memory\0"
482 "unmatched parentheses\0"
483 "internal error: code overflow\0"
484 "unrecognized character after (?<\0"
485 /* 25 */
486 "lookbehind assertion is not fixed length\0"
487 "malformed number or name after (?(\0"
488 "conditional group contains more than two branches\0"
489 "assertion expected after (?( or (?(?C)\0"
490 "(?R or (?[+-]digits must be followed by )\0"
491 /* 30 */
492 "unknown POSIX class name\0"
493 "POSIX collating elements are not supported\0"
494 "this version of PCRE is compiled without UTF support\0"
495 "spare error\0" /** DEAD **/
496 "character value in \\x{} or \\o{} is too large\0"
497 /* 35 */
498 "invalid condition (?(0)\0"
499 "\\C not allowed in lookbehind assertion\0"
500 "PCRE does not support \\L, \\l, \\N{name}, \\U, or \\u\0"
501 "number after (?C is > 255\0"
502 "closing ) for (?C expected\0"
503 /* 40 */
504 "recursive call could loop indefinitely\0"
505 "unrecognized character after (?P\0"
506 "syntax error in subpattern name (missing terminator)\0"
507 "two named subpatterns have the same name\0"
508 "invalid UTF-8 string\0"
509 /* 45 */
510 "support for \\P, \\p, and \\X has not been compiled\0"
511 "malformed \\P or \\p sequence\0"
512 "unknown property name after \\P or \\p\0"
513 "subpattern name is too long (maximum " XSTRING(MAX_NAME_SIZE) " characters)\0"
514 "too many named subpatterns (maximum " XSTRING(MAX_NAME_COUNT) ")\0"
515 /* 50 */
516 "repeated subpattern is too long\0" /** DEAD **/
517 "octal value is greater than \\377 in 8-bit non-UTF-8 mode\0"
518 "internal error: overran compiling workspace\0"
519 "internal error: previously-checked referenced subpattern not found\0"
520 "DEFINE group contains more than one branch\0"
521 /* 55 */
522 "repeating a DEFINE group is not allowed\0" /** DEAD **/
523 "inconsistent NEWLINE options\0"
524 "\\g is not followed by a braced, angle-bracketed, or quoted name/number or by a plain number\0"
525 "a numbered reference must not be zero\0"
526 "an argument is not allowed for (*ACCEPT), (*FAIL), or (*COMMIT)\0"
527 /* 60 */
528 "(*VERB) not recognized or malformed\0"
529 "number is too big\0"
530 "subpattern name expected\0"
531 "digit expected after (?+\0"
532 "] is an invalid data character in JavaScript compatibility mode\0"
533 /* 65 */
534 "different names for subpatterns of the same number are not allowed\0"
535 "(*MARK) must have an argument\0"
536 "this version of PCRE is not compiled with Unicode property support\0"
537 #ifndef EBCDIC
538 "\\c must be followed by an ASCII character\0"
539 #else
540 "\\c must be followed by a letter or one of [\\]^_?\0"
541 #endif
542 "\\k is not followed by a braced, angle-bracketed, or quoted name\0"
543 /* 70 */
544 "internal error: unknown opcode in find_fixedlength()\0"
545 "\\N is not supported in a class\0"
546 "too many forward references\0"
547 "disallowed Unicode code point (>= 0xd800 && <= 0xdfff)\0"
548 "invalid UTF-16 string\0"
549 /* 75 */
550 "name is too long in (*MARK), (*PRUNE), (*SKIP), or (*THEN)\0"
551 "character value in \\u.... sequence is too large\0"
552 "invalid UTF-32 string\0"
553 "setting UTF is disabled by the application\0"
554 "non-hex character in \\x{} (closing brace missing?)\0"
555 /* 80 */
556 "non-octal character in \\o{} (closing brace missing?)\0"
557 "missing opening brace after \\o\0"
558 "parentheses are too deeply nested\0"
559 "invalid range in character class\0"
560 "group name must start with a non-digit\0"
561 /* 85 */
562 "parentheses are too deeply nested (stack check)\0"
563 "digits missing in \\x{} or \\o{}\0"
564 "regular expression is too complicated\0"
565 ;
566
567 /* Table to identify digits and hex digits. This is used when compiling
568 patterns. Note that the tables in chartables are dependent on the locale, and
569 may mark arbitrary characters as digits - but the PCRE compiling code expects
570 to handle only 0-9, a-z, and A-Z as digits when compiling. That is why we have
571 a private table here. It costs 256 bytes, but it is a lot faster than doing
572 character value tests (at least in some simple cases I timed), and in some
573 applications one wants PCRE to compile efficiently as well as match
574 efficiently.
575
576 For convenience, we use the same bit definitions as in chartables:
577
578 0x04 decimal digit
579 0x08 hexadecimal digit
580
581 Then we can use ctype_digit and ctype_xdigit in the code. */
582
583 /* Using a simple comparison for decimal numbers rather than a memory read
584 is much faster, and the resulting code is simpler (the compiler turns it
585 into a subtraction and unsigned comparison). */
586
587 #define IS_DIGIT(x) ((x) >= CHAR_0 && (x) <= CHAR_9)
588
589 #ifndef EBCDIC
590
591 /* This is the "normal" case, for ASCII systems, and EBCDIC systems running in
592 UTF-8 mode. */
593
594 static const pcre_uint8 digitab[] =
595 {
596 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 0- 7 */
597 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 8- 15 */
598 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 16- 23 */
599 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 24- 31 */
600 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - ' */
601 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* ( - / */
602 0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c, /* 0 - 7 */
603 0x0c,0x0c,0x00,0x00,0x00,0x00,0x00,0x00, /* 8 - ? */
604 0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* @ - G */
605 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* H - O */
606 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* P - W */
607 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* X - _ */
608 0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* ` - g */
609 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* h - o */
610 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* p - w */
611 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* x -127 */
612 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 128-135 */
613 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 136-143 */
614 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 144-151 */
615 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 152-159 */
616 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 160-167 */
617 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 168-175 */
618 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 176-183 */
619 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 184-191 */
620 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 192-199 */
621 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 200-207 */
622 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 208-215 */
623 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 216-223 */
624 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 224-231 */
625 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 232-239 */
626 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 240-247 */
627 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00};/* 248-255 */
628
629 #else
630
631 /* This is the "abnormal" case, for EBCDIC systems not running in UTF-8 mode. */
632
633 static const pcre_uint8 digitab[] =
634 {
635 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 0- 7 0 */
636 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 8- 15 */
637 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 16- 23 10 */
638 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 24- 31 */
639 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 32- 39 20 */
640 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 40- 47 */
641 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 48- 55 30 */
642 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 56- 63 */
643 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - 71 40 */
644 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 72- | */
645 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* & - 87 50 */
646 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 88- 95 */
647 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - -103 60 */
648 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 104- ? */
649 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 112-119 70 */
650 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 120- " */
651 0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* 128- g 80 */
652 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* h -143 */
653 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 144- p 90 */
654 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* q -159 */
655 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 160- x A0 */
656 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* y -175 */
657 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* ^ -183 B0 */
658 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 184-191 */
659 0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* { - G C0 */
660 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* H -207 */
661 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* } - P D0 */
662 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* Q -223 */
663 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* \ - X E0 */
664 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* Y -239 */
665 0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c, /* 0 - 7 F0 */
666 0x0c,0x0c,0x00,0x00,0x00,0x00,0x00,0x00};/* 8 -255 */
667
668 static const pcre_uint8 ebcdic_chartab[] = { /* chartable partial dup */
669 0x80,0x00,0x00,0x00,0x00,0x01,0x00,0x00, /* 0- 7 */
670 0x00,0x00,0x00,0x00,0x01,0x01,0x00,0x00, /* 8- 15 */
671 0x00,0x00,0x00,0x00,0x00,0x01,0x00,0x00, /* 16- 23 */
672 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 24- 31 */
673 0x00,0x00,0x00,0x00,0x00,0x01,0x00,0x00, /* 32- 39 */
674 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 40- 47 */
675 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 48- 55 */
676 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 56- 63 */
677 0x01,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - 71 */
678 0x00,0x00,0x00,0x80,0x00,0x80,0x80,0x80, /* 72- | */
679 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* & - 87 */
680 0x00,0x00,0x00,0x80,0x80,0x80,0x00,0x00, /* 88- 95 */
681 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - -103 */
682 0x00,0x00,0x00,0x00,0x00,0x10,0x00,0x80, /* 104- ? */
683 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 112-119 */
684 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 120- " */
685 0x00,0x1a,0x1a,0x1a,0x1a,0x1a,0x1a,0x12, /* 128- g */
686 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* h -143 */
687 0x00,0x12,0x12,0x12,0x12,0x12,0x12,0x12, /* 144- p */
688 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* q -159 */
689 0x00,0x00,0x12,0x12,0x12,0x12,0x12,0x12, /* 160- x */
690 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* y -175 */
691 0x80,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* ^ -183 */
692 0x00,0x00,0x80,0x00,0x00,0x00,0x00,0x00, /* 184-191 */
693 0x80,0x1a,0x1a,0x1a,0x1a,0x1a,0x1a,0x12, /* { - G */
694 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* H -207 */
695 0x00,0x12,0x12,0x12,0x12,0x12,0x12,0x12, /* } - P */
696 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* Q -223 */
697 0x00,0x00,0x12,0x12,0x12,0x12,0x12,0x12, /* \ - X */
698 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* Y -239 */
699 0x1c,0x1c,0x1c,0x1c,0x1c,0x1c,0x1c,0x1c, /* 0 - 7 */
700 0x1c,0x1c,0x00,0x00,0x00,0x00,0x00,0x00};/* 8 -255 */
701 #endif
702
703
704 /* This table is used to check whether auto-possessification is possible
705 between adjacent character-type opcodes. The left-hand (repeated) opcode is
706 used to select the row, and the right-hand opcode is use to select the column.
707 A value of 1 means that auto-possessification is OK. For example, the second
708 value in the first row means that \D+\d can be turned into \D++\d.
709
710 The Unicode property types (\P and \p) have to be present to fill out the table
711 because of what their opcode values are, but the table values should always be
712 zero because property types are handled separately in the code. The last four
713 columns apply to items that cannot be repeated, so there is no need to have
714 rows for them. Note that OP_DIGIT etc. are generated only when PCRE_UCP is
715 *not* set. When it is set, \d etc. are converted into OP_(NOT_)PROP codes. */
716
717 #define APTROWS (LAST_AUTOTAB_LEFT_OP - FIRST_AUTOTAB_OP + 1)
718 #define APTCOLS (LAST_AUTOTAB_RIGHT_OP - FIRST_AUTOTAB_OP + 1)
719
720 static const pcre_uint8 autoposstab[APTROWS][APTCOLS] = {
721 /* \D \d \S \s \W \w . .+ \C \P \p \R \H \h \V \v \X \Z \z $ $M */
722 { 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0 }, /* \D */
723 { 1, 0, 0, 1, 1, 0, 0, 0, 0, 0, 0, 1, 0, 1, 0, 1, 0, 1, 1, 1, 1 }, /* \d */
724 { 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 1, 0, 1, 0, 1, 0, 1, 1, 1, 1 }, /* \S */
725 { 0, 1, 1, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0 }, /* \s */
726 { 0, 1, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0 }, /* \W */
727 { 0, 0, 0, 1, 1, 0, 0, 0, 0, 0, 0, 1, 0, 1, 0, 1, 0, 1, 1, 1, 1 }, /* \w */
728 { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 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 }, /* .+ */
730 { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0 }, /* \C */
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, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 }, /* \p */
733 { 0, 1, 0, 1, 0, 1, 1, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 1, 0, 0 }, /* \R */
734 { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 1, 0, 0 }, /* \H */
735 { 0, 1, 1, 0, 0, 1, 0, 0, 0, 0, 0, 1, 1, 0, 0, 1, 0, 0, 1, 0, 0 }, /* \h */
736 { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 1, 0, 0, 1, 0, 0 }, /* \V */
737 { 0, 1, 1, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 1, 1, 0, 0, 0, 1, 0, 0 }, /* \v */
738 { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0 } /* \X */
739 };
740
741
742 /* This table is used to check whether auto-possessification is possible
743 between adjacent Unicode property opcodes (OP_PROP and OP_NOTPROP). The
744 left-hand (repeated) opcode is used to select the row, and the right-hand
745 opcode is used to select the column. The values are as follows:
746
747 0 Always return FALSE (never auto-possessify)
748 1 Character groups are distinct (possessify if both are OP_PROP)
749 2 Check character categories in the same group (general or particular)
750 3 TRUE if the two opcodes are not the same (PROP vs NOTPROP)
751
752 4 Check left general category vs right particular category
753 5 Check right general category vs left particular category
754
755 6 Left alphanum vs right general category
756 7 Left space vs right general category
757 8 Left word vs right general category
758
759 9 Right alphanum vs left general category
760 10 Right space vs left general category
761 11 Right word vs left general category
762
763 12 Left alphanum vs right particular category
764 13 Left space vs right particular category
765 14 Left word vs right particular category
766
767 15 Right alphanum vs left particular category
768 16 Right space vs left particular category
769 17 Right word vs left particular category
770 */
771
772 static const pcre_uint8 propposstab[PT_TABSIZE][PT_TABSIZE] = {
773 /* ANY LAMP GC PC SC ALNUM SPACE PXSPACE WORD CLIST UCNC */
774 { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 }, /* PT_ANY */
775 { 0, 3, 0, 0, 0, 3, 1, 1, 0, 0, 0 }, /* PT_LAMP */
776 { 0, 0, 2, 4, 0, 9, 10, 10, 11, 0, 0 }, /* PT_GC */
777 { 0, 0, 5, 2, 0, 15, 16, 16, 17, 0, 0 }, /* PT_PC */
778 { 0, 0, 0, 0, 2, 0, 0, 0, 0, 0, 0 }, /* PT_SC */
779 { 0, 3, 6, 12, 0, 3, 1, 1, 0, 0, 0 }, /* PT_ALNUM */
780 { 0, 1, 7, 13, 0, 1, 3, 3, 1, 0, 0 }, /* PT_SPACE */
781 { 0, 1, 7, 13, 0, 1, 3, 3, 1, 0, 0 }, /* PT_PXSPACE */
782 { 0, 0, 8, 14, 0, 0, 1, 1, 3, 0, 0 }, /* PT_WORD */
783 { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 }, /* PT_CLIST */
784 { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 3 } /* PT_UCNC */
785 };
786
787 /* This table is used to check whether auto-possessification is possible
788 between adjacent Unicode property opcodes (OP_PROP and OP_NOTPROP) when one
789 specifies a general category and the other specifies a particular category. The
790 row is selected by the general category and the column by the particular
791 category. The value is 1 if the particular category is not part of the general
792 category. */
793
794 static const pcre_uint8 catposstab[7][30] = {
795 /* 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 */
796 { 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 */
797 { 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 */
798 { 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 */
799 { 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 */
800 { 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 */
801 { 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 */
802 { 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 */
803 };
804
805 /* This table is used when checking ALNUM, (PX)SPACE, SPACE, and WORD against
806 a general or particular category. The properties in each row are those
807 that apply to the character set in question. Duplication means that a little
808 unnecessary work is done when checking, but this keeps things much simpler
809 because they can all use the same code. For more details see the comment where
810 this table is used.
811
812 Note: SPACE and PXSPACE used to be different because Perl excluded VT from
813 "space", but from Perl 5.18 it's included, so both categories are treated the
814 same here. */
815
816 static const pcre_uint8 posspropstab[3][4] = {
817 { ucp_L, ucp_N, ucp_N, ucp_Nl }, /* ALNUM, 3rd and 4th values redundant */
818 { ucp_Z, ucp_Z, ucp_C, ucp_Cc }, /* SPACE and PXSPACE, 2nd value redundant */
819 { ucp_L, ucp_N, ucp_P, ucp_Po } /* WORD */
820 };
821
822 /* This table is used when converting repeating opcodes into possessified
823 versions as a result of an explicit possessive quantifier such as ++. A zero
824 value means there is no possessified version - in those cases the item in
825 question must be wrapped in ONCE brackets. The table is truncated at OP_CALLOUT
826 because all relevant opcodes are less than that. */
827
828 static const pcre_uint8 opcode_possessify[] = {
829 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, /* 0 - 15 */
830 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, /* 16 - 31 */
831
832 0, /* NOTI */
833 OP_POSSTAR, 0, /* STAR, MINSTAR */
834 OP_POSPLUS, 0, /* PLUS, MINPLUS */
835 OP_POSQUERY, 0, /* QUERY, MINQUERY */
836 OP_POSUPTO, 0, /* UPTO, MINUPTO */
837 0, /* EXACT */
838 0, 0, 0, 0, /* POS{STAR,PLUS,QUERY,UPTO} */
839
840 OP_POSSTARI, 0, /* STARI, MINSTARI */
841 OP_POSPLUSI, 0, /* PLUSI, MINPLUSI */
842 OP_POSQUERYI, 0, /* QUERYI, MINQUERYI */
843 OP_POSUPTOI, 0, /* UPTOI, MINUPTOI */
844 0, /* EXACTI */
845 0, 0, 0, 0, /* POS{STARI,PLUSI,QUERYI,UPTOI} */
846
847 OP_NOTPOSSTAR, 0, /* NOTSTAR, NOTMINSTAR */
848 OP_NOTPOSPLUS, 0, /* NOTPLUS, NOTMINPLUS */
849 OP_NOTPOSQUERY, 0, /* NOTQUERY, NOTMINQUERY */
850 OP_NOTPOSUPTO, 0, /* NOTUPTO, NOTMINUPTO */
851 0, /* NOTEXACT */
852 0, 0, 0, 0, /* NOTPOS{STAR,PLUS,QUERY,UPTO} */
853
854 OP_NOTPOSSTARI, 0, /* NOTSTARI, NOTMINSTARI */
855 OP_NOTPOSPLUSI, 0, /* NOTPLUSI, NOTMINPLUSI */
856 OP_NOTPOSQUERYI, 0, /* NOTQUERYI, NOTMINQUERYI */
857 OP_NOTPOSUPTOI, 0, /* NOTUPTOI, NOTMINUPTOI */
858 0, /* NOTEXACTI */
859 0, 0, 0, 0, /* NOTPOS{STARI,PLUSI,QUERYI,UPTOI} */
860
861 OP_TYPEPOSSTAR, 0, /* TYPESTAR, TYPEMINSTAR */
862 OP_TYPEPOSPLUS, 0, /* TYPEPLUS, TYPEMINPLUS */
863 OP_TYPEPOSQUERY, 0, /* TYPEQUERY, TYPEMINQUERY */
864 OP_TYPEPOSUPTO, 0, /* TYPEUPTO, TYPEMINUPTO */
865 0, /* TYPEEXACT */
866 0, 0, 0, 0, /* TYPEPOS{STAR,PLUS,QUERY,UPTO} */
867
868 OP_CRPOSSTAR, 0, /* CRSTAR, CRMINSTAR */
869 OP_CRPOSPLUS, 0, /* CRPLUS, CRMINPLUS */
870 OP_CRPOSQUERY, 0, /* CRQUERY, CRMINQUERY */
871 OP_CRPOSRANGE, 0, /* CRRANGE, CRMINRANGE */
872 0, 0, 0, 0, /* CRPOS{STAR,PLUS,QUERY,RANGE} */
873
874 0, 0, 0, /* CLASS, NCLASS, XCLASS */
875 0, 0, /* REF, REFI */
876 0, 0, /* DNREF, DNREFI */
877 0, 0 /* RECURSE, CALLOUT */
878 };
879
880
881
882 /*************************************************
883 * Find an error text *
884 *************************************************/
885
886 /* The error texts are now all in one long string, to save on relocations. As
887 some of the text is of unknown length, we can't use a table of offsets.
888 Instead, just count through the strings. This is not a performance issue
889 because it happens only when there has been a compilation error.
890
891 Argument: the error number
892 Returns: pointer to the error string
893 */
894
895 static const char *
find_error_text(int n)896 find_error_text(int n)
897 {
898 const char *s = error_texts;
899 for (; n > 0; n--)
900 {
901 while (*s++ != CHAR_NULL) {};
902 if (*s == CHAR_NULL) return "Error text not found (please report)";
903 }
904 return s;
905 }
906
907
908
909 /*************************************************
910 * Expand the workspace *
911 *************************************************/
912
913 /* This function is called during the second compiling phase, if the number of
914 forward references fills the existing workspace, which is originally a block on
915 the stack. A larger block is obtained from malloc() unless the ultimate limit
916 has been reached or the increase will be rather small.
917
918 Argument: pointer to the compile data block
919 Returns: 0 if all went well, else an error number
920 */
921
922 static int
expand_workspace(compile_data * cd)923 expand_workspace(compile_data *cd)
924 {
925 pcre_uchar *newspace;
926 int newsize = cd->workspace_size * 2;
927
928 if (newsize > COMPILE_WORK_SIZE_MAX) newsize = COMPILE_WORK_SIZE_MAX;
929 if (cd->workspace_size >= COMPILE_WORK_SIZE_MAX ||
930 newsize - cd->workspace_size < WORK_SIZE_SAFETY_MARGIN)
931 return ERR72;
932
933 newspace = (PUBL(malloc))(IN_UCHARS(newsize));
934 if (newspace == NULL) return ERR21;
935 memcpy(newspace, cd->start_workspace, cd->workspace_size * sizeof(pcre_uchar));
936 cd->hwm = (pcre_uchar *)newspace + (cd->hwm - cd->start_workspace);
937 if (cd->workspace_size > COMPILE_WORK_SIZE)
938 (PUBL(free))((void *)cd->start_workspace);
939 cd->start_workspace = newspace;
940 cd->workspace_size = newsize;
941 return 0;
942 }
943
944
945
946 /*************************************************
947 * Check for counted repeat *
948 *************************************************/
949
950 /* This function is called when a '{' is encountered in a place where it might
951 start a quantifier. It looks ahead to see if it really is a quantifier or not.
952 It is only a quantifier if it is one of the forms {ddd} {ddd,} or {ddd,ddd}
953 where the ddds are digits.
954
955 Arguments:
956 p pointer to the first char after '{'
957
958 Returns: TRUE or FALSE
959 */
960
961 static BOOL
is_counted_repeat(const pcre_uchar * p)962 is_counted_repeat(const pcre_uchar *p)
963 {
964 if (!IS_DIGIT(*p)) return FALSE;
965 p++;
966 while (IS_DIGIT(*p)) p++;
967 if (*p == CHAR_RIGHT_CURLY_BRACKET) return TRUE;
968
969 if (*p++ != CHAR_COMMA) return FALSE;
970 if (*p == CHAR_RIGHT_CURLY_BRACKET) return TRUE;
971
972 if (!IS_DIGIT(*p)) return FALSE;
973 p++;
974 while (IS_DIGIT(*p)) p++;
975
976 return (*p == CHAR_RIGHT_CURLY_BRACKET);
977 }
978
979
980
981 /*************************************************
982 * Handle escapes *
983 *************************************************/
984
985 /* This function is called when a \ has been encountered. It either returns a
986 positive value for a simple escape such as \n, or 0 for a data character which
987 will be placed in chptr. A backreference to group n is returned as negative n.
988 When UTF-8 is enabled, a positive value greater than 255 may be returned in
989 chptr. On entry, ptr is pointing at the \. On exit, it is on the final
990 character of the escape sequence.
991
992 Arguments:
993 ptrptr points to the pattern position pointer
994 chptr points to a returned data character
995 errorcodeptr points to the errorcode variable
996 bracount number of previous extracting brackets
997 options the options bits
998 isclass TRUE if inside a character class
999
1000 Returns: zero => a data character
1001 positive => a special escape sequence
1002 negative => a back reference
1003 on error, errorcodeptr is set
1004 */
1005
1006 static int
check_escape(const pcre_uchar ** ptrptr,pcre_uint32 * chptr,int * errorcodeptr,int bracount,int options,BOOL isclass)1007 check_escape(const pcre_uchar **ptrptr, pcre_uint32 *chptr, int *errorcodeptr,
1008 int bracount, int options, BOOL isclass)
1009 {
1010 /* PCRE_UTF16 has the same value as PCRE_UTF8. */
1011 BOOL utf = (options & PCRE_UTF8) != 0;
1012 const pcre_uchar *ptr = *ptrptr + 1;
1013 pcre_uint32 c;
1014 int escape = 0;
1015 int i;
1016
1017 GETCHARINCTEST(c, ptr); /* Get character value, increment pointer */
1018 ptr--; /* Set pointer back to the last byte */
1019
1020 /* If backslash is at the end of the pattern, it's an error. */
1021
1022 if (c == CHAR_NULL) *errorcodeptr = ERR1;
1023
1024 /* Non-alphanumerics are literals. For digits or letters, do an initial lookup
1025 in a table. A non-zero result is something that can be returned immediately.
1026 Otherwise further processing may be required. */
1027
1028 #ifndef EBCDIC /* ASCII/UTF-8 coding */
1029 /* Not alphanumeric */
1030 else if (c < CHAR_0 || c > CHAR_z) {}
1031 else if ((i = escapes[c - CHAR_0]) != 0)
1032 { if (i > 0) c = (pcre_uint32)i; else escape = -i; }
1033
1034 #else /* EBCDIC coding */
1035 /* Not alphanumeric */
1036 else if (c < CHAR_a || (!MAX_255(c) || (ebcdic_chartab[c] & 0x0E) == 0)) {}
1037 else if ((i = escapes[c - 0x48]) != 0) { if (i > 0) c = (pcre_uint32)i; else escape = -i; }
1038 #endif
1039
1040 /* Escapes that need further processing, or are illegal. */
1041
1042 else
1043 {
1044 const pcre_uchar *oldptr;
1045 BOOL braced, negated, overflow;
1046 int s;
1047
1048 switch (c)
1049 {
1050 /* A number of Perl escapes are not handled by PCRE. We give an explicit
1051 error. */
1052
1053 case CHAR_l:
1054 case CHAR_L:
1055 *errorcodeptr = ERR37;
1056 break;
1057
1058 case CHAR_u:
1059 if ((options & PCRE_JAVASCRIPT_COMPAT) != 0)
1060 {
1061 /* In JavaScript, \u must be followed by four hexadecimal numbers.
1062 Otherwise it is a lowercase u letter. */
1063 if (MAX_255(ptr[1]) && (digitab[ptr[1]] & ctype_xdigit) != 0
1064 && MAX_255(ptr[2]) && (digitab[ptr[2]] & ctype_xdigit) != 0
1065 && MAX_255(ptr[3]) && (digitab[ptr[3]] & ctype_xdigit) != 0
1066 && MAX_255(ptr[4]) && (digitab[ptr[4]] & ctype_xdigit) != 0)
1067 {
1068 c = 0;
1069 for (i = 0; i < 4; ++i)
1070 {
1071 register pcre_uint32 cc = *(++ptr);
1072 #ifndef EBCDIC /* ASCII/UTF-8 coding */
1073 if (cc >= CHAR_a) cc -= 32; /* Convert to upper case */
1074 c = (c << 4) + cc - ((cc < CHAR_A)? CHAR_0 : (CHAR_A - 10));
1075 #else /* EBCDIC coding */
1076 if (cc >= CHAR_a && cc <= CHAR_z) cc += 64; /* Convert to upper case */
1077 c = (c << 4) + cc - ((cc >= CHAR_0)? CHAR_0 : (CHAR_A - 10));
1078 #endif
1079 }
1080
1081 #if defined COMPILE_PCRE8
1082 if (c > (utf ? 0x10ffffU : 0xffU))
1083 #elif defined COMPILE_PCRE16
1084 if (c > (utf ? 0x10ffffU : 0xffffU))
1085 #elif defined COMPILE_PCRE32
1086 if (utf && c > 0x10ffffU)
1087 #endif
1088 {
1089 *errorcodeptr = ERR76;
1090 }
1091 else if (utf && c >= 0xd800 && c <= 0xdfff) *errorcodeptr = ERR73;
1092 }
1093 }
1094 else
1095 *errorcodeptr = ERR37;
1096 break;
1097
1098 case CHAR_U:
1099 /* In JavaScript, \U is an uppercase U letter. */
1100 if ((options & PCRE_JAVASCRIPT_COMPAT) == 0) *errorcodeptr = ERR37;
1101 break;
1102
1103 /* In a character class, \g is just a literal "g". Outside a character
1104 class, \g must be followed by one of a number of specific things:
1105
1106 (1) A number, either plain or braced. If positive, it is an absolute
1107 backreference. If negative, it is a relative backreference. This is a Perl
1108 5.10 feature.
1109
1110 (2) Perl 5.10 also supports \g{name} as a reference to a named group. This
1111 is part of Perl's movement towards a unified syntax for back references. As
1112 this is synonymous with \k{name}, we fudge it up by pretending it really
1113 was \k.
1114
1115 (3) For Oniguruma compatibility we also support \g followed by a name or a
1116 number either in angle brackets or in single quotes. However, these are
1117 (possibly recursive) subroutine calls, _not_ backreferences. Just return
1118 the ESC_g code (cf \k). */
1119
1120 case CHAR_g:
1121 if (isclass) break;
1122 if (ptr[1] == CHAR_LESS_THAN_SIGN || ptr[1] == CHAR_APOSTROPHE)
1123 {
1124 escape = ESC_g;
1125 break;
1126 }
1127
1128 /* Handle the Perl-compatible cases */
1129
1130 if (ptr[1] == CHAR_LEFT_CURLY_BRACKET)
1131 {
1132 const pcre_uchar *p;
1133 for (p = ptr+2; *p != CHAR_NULL && *p != CHAR_RIGHT_CURLY_BRACKET; p++)
1134 if (*p != CHAR_MINUS && !IS_DIGIT(*p)) break;
1135 if (*p != CHAR_NULL && *p != CHAR_RIGHT_CURLY_BRACKET)
1136 {
1137 escape = ESC_k;
1138 break;
1139 }
1140 braced = TRUE;
1141 ptr++;
1142 }
1143 else braced = FALSE;
1144
1145 if (ptr[1] == CHAR_MINUS)
1146 {
1147 negated = TRUE;
1148 ptr++;
1149 }
1150 else negated = FALSE;
1151
1152 /* The integer range is limited by the machine's int representation. */
1153 s = 0;
1154 overflow = FALSE;
1155 while (IS_DIGIT(ptr[1]))
1156 {
1157 if (s > INT_MAX / 10 - 1) /* Integer overflow */
1158 {
1159 overflow = TRUE;
1160 break;
1161 }
1162 s = s * 10 + (int)(*(++ptr) - CHAR_0);
1163 }
1164 if (overflow) /* Integer overflow */
1165 {
1166 while (IS_DIGIT(ptr[1]))
1167 ptr++;
1168 *errorcodeptr = ERR61;
1169 break;
1170 }
1171
1172 if (braced && *(++ptr) != CHAR_RIGHT_CURLY_BRACKET)
1173 {
1174 *errorcodeptr = ERR57;
1175 break;
1176 }
1177
1178 if (s == 0)
1179 {
1180 *errorcodeptr = ERR58;
1181 break;
1182 }
1183
1184 if (negated)
1185 {
1186 if (s > bracount)
1187 {
1188 *errorcodeptr = ERR15;
1189 break;
1190 }
1191 s = bracount - (s - 1);
1192 }
1193
1194 escape = -s;
1195 break;
1196
1197 /* The handling of escape sequences consisting of a string of digits
1198 starting with one that is not zero is not straightforward. Perl has changed
1199 over the years. Nowadays \g{} for backreferences and \o{} for octal are
1200 recommended to avoid the ambiguities in the old syntax.
1201
1202 Outside a character class, the digits are read as a decimal number. If the
1203 number is less than 8 (used to be 10), or if there are that many previous
1204 extracting left brackets, then it is a back reference. Otherwise, up to
1205 three octal digits are read to form an escaped byte. Thus \123 is likely to
1206 be octal 123 (cf \0123, which is octal 012 followed by the literal 3). If
1207 the octal value is greater than 377, the least significant 8 bits are
1208 taken. \8 and \9 are treated as the literal characters 8 and 9.
1209
1210 Inside a character class, \ followed by a digit is always either a literal
1211 8 or 9 or an octal number. */
1212
1213 case CHAR_1: case CHAR_2: case CHAR_3: case CHAR_4: case CHAR_5:
1214 case CHAR_6: case CHAR_7: case CHAR_8: case CHAR_9:
1215
1216 if (!isclass)
1217 {
1218 oldptr = ptr;
1219 /* The integer range is limited by the machine's int representation. */
1220 s = (int)(c -CHAR_0);
1221 overflow = FALSE;
1222 while (IS_DIGIT(ptr[1]))
1223 {
1224 if (s > INT_MAX / 10 - 1) /* Integer overflow */
1225 {
1226 overflow = TRUE;
1227 break;
1228 }
1229 s = s * 10 + (int)(*(++ptr) - CHAR_0);
1230 }
1231 if (overflow) /* Integer overflow */
1232 {
1233 while (IS_DIGIT(ptr[1]))
1234 ptr++;
1235 *errorcodeptr = ERR61;
1236 break;
1237 }
1238 if (s < 8 || s <= bracount) /* Check for back reference */
1239 {
1240 escape = -s;
1241 break;
1242 }
1243 ptr = oldptr; /* Put the pointer back and fall through */
1244 }
1245
1246 /* Handle a digit following \ when the number is not a back reference. If
1247 the first digit is 8 or 9, Perl used to generate a binary zero byte and
1248 then treat the digit as a following literal. At least by Perl 5.18 this
1249 changed so as not to insert the binary zero. */
1250
1251 if ((c = *ptr) >= CHAR_8) break;
1252
1253 /* Fall through with a digit less than 8 */
1254
1255 /* \0 always starts an octal number, but we may drop through to here with a
1256 larger first octal digit. The original code used just to take the least
1257 significant 8 bits of octal numbers (I think this is what early Perls used
1258 to do). Nowadays we allow for larger numbers in UTF-8 mode and 16-bit mode,
1259 but no more than 3 octal digits. */
1260
1261 case CHAR_0:
1262 c -= CHAR_0;
1263 while(i++ < 2 && ptr[1] >= CHAR_0 && ptr[1] <= CHAR_7)
1264 c = c * 8 + *(++ptr) - CHAR_0;
1265 #ifdef COMPILE_PCRE8
1266 if (!utf && c > 0xff) *errorcodeptr = ERR51;
1267 #endif
1268 break;
1269
1270 /* \o is a relatively new Perl feature, supporting a more general way of
1271 specifying character codes in octal. The only supported form is \o{ddd}. */
1272
1273 case CHAR_o:
1274 if (ptr[1] != CHAR_LEFT_CURLY_BRACKET) *errorcodeptr = ERR81; else
1275 if (ptr[2] == CHAR_RIGHT_CURLY_BRACKET) *errorcodeptr = ERR86; else
1276 {
1277 ptr += 2;
1278 c = 0;
1279 overflow = FALSE;
1280 while (*ptr >= CHAR_0 && *ptr <= CHAR_7)
1281 {
1282 register pcre_uint32 cc = *ptr++;
1283 if (c == 0 && cc == CHAR_0) continue; /* Leading zeroes */
1284 #ifdef COMPILE_PCRE32
1285 if (c >= 0x20000000l) { overflow = TRUE; break; }
1286 #endif
1287 c = (c << 3) + cc - CHAR_0 ;
1288 #if defined COMPILE_PCRE8
1289 if (c > (utf ? 0x10ffffU : 0xffU)) { overflow = TRUE; break; }
1290 #elif defined COMPILE_PCRE16
1291 if (c > (utf ? 0x10ffffU : 0xffffU)) { overflow = TRUE; break; }
1292 #elif defined COMPILE_PCRE32
1293 if (utf && c > 0x10ffffU) { overflow = TRUE; break; }
1294 #endif
1295 }
1296 if (overflow)
1297 {
1298 while (*ptr >= CHAR_0 && *ptr <= CHAR_7) ptr++;
1299 *errorcodeptr = ERR34;
1300 }
1301 else if (*ptr == CHAR_RIGHT_CURLY_BRACKET)
1302 {
1303 if (utf && c >= 0xd800 && c <= 0xdfff) *errorcodeptr = ERR73;
1304 }
1305 else *errorcodeptr = ERR80;
1306 }
1307 break;
1308
1309 /* \x is complicated. In JavaScript, \x must be followed by two hexadecimal
1310 numbers. Otherwise it is a lowercase x letter. */
1311
1312 case CHAR_x:
1313 if ((options & PCRE_JAVASCRIPT_COMPAT) != 0)
1314 {
1315 if (MAX_255(ptr[1]) && (digitab[ptr[1]] & ctype_xdigit) != 0
1316 && MAX_255(ptr[2]) && (digitab[ptr[2]] & ctype_xdigit) != 0)
1317 {
1318 c = 0;
1319 for (i = 0; i < 2; ++i)
1320 {
1321 register pcre_uint32 cc = *(++ptr);
1322 #ifndef EBCDIC /* ASCII/UTF-8 coding */
1323 if (cc >= CHAR_a) cc -= 32; /* Convert to upper case */
1324 c = (c << 4) + cc - ((cc < CHAR_A)? CHAR_0 : (CHAR_A - 10));
1325 #else /* EBCDIC coding */
1326 if (cc >= CHAR_a && cc <= CHAR_z) cc += 64; /* Convert to upper case */
1327 c = (c << 4) + cc - ((cc >= CHAR_0)? CHAR_0 : (CHAR_A - 10));
1328 #endif
1329 }
1330 }
1331 } /* End JavaScript handling */
1332
1333 /* Handle \x in Perl's style. \x{ddd} is a character number which can be
1334 greater than 0xff in utf or non-8bit mode, but only if the ddd are hex
1335 digits. If not, { used to be treated as a data character. However, Perl
1336 seems to read hex digits up to the first non-such, and ignore the rest, so
1337 that, for example \x{zz} matches a binary zero. This seems crazy, so PCRE
1338 now gives an error. */
1339
1340 else
1341 {
1342 if (ptr[1] == CHAR_LEFT_CURLY_BRACKET)
1343 {
1344 ptr += 2;
1345 if (*ptr == CHAR_RIGHT_CURLY_BRACKET)
1346 {
1347 *errorcodeptr = ERR86;
1348 break;
1349 }
1350 c = 0;
1351 overflow = FALSE;
1352 while (MAX_255(*ptr) && (digitab[*ptr] & ctype_xdigit) != 0)
1353 {
1354 register pcre_uint32 cc = *ptr++;
1355 if (c == 0 && cc == CHAR_0) continue; /* Leading zeroes */
1356
1357 #ifdef COMPILE_PCRE32
1358 if (c >= 0x10000000l) { overflow = TRUE; break; }
1359 #endif
1360
1361 #ifndef EBCDIC /* ASCII/UTF-8 coding */
1362 if (cc >= CHAR_a) cc -= 32; /* Convert to upper case */
1363 c = (c << 4) + cc - ((cc < CHAR_A)? CHAR_0 : (CHAR_A - 10));
1364 #else /* EBCDIC coding */
1365 if (cc >= CHAR_a && cc <= CHAR_z) cc += 64; /* Convert to upper case */
1366 c = (c << 4) + cc - ((cc >= CHAR_0)? CHAR_0 : (CHAR_A - 10));
1367 #endif
1368
1369 #if defined COMPILE_PCRE8
1370 if (c > (utf ? 0x10ffffU : 0xffU)) { overflow = TRUE; break; }
1371 #elif defined COMPILE_PCRE16
1372 if (c > (utf ? 0x10ffffU : 0xffffU)) { overflow = TRUE; break; }
1373 #elif defined COMPILE_PCRE32
1374 if (utf && c > 0x10ffffU) { overflow = TRUE; break; }
1375 #endif
1376 }
1377
1378 if (overflow)
1379 {
1380 while (MAX_255(*ptr) && (digitab[*ptr] & ctype_xdigit) != 0) ptr++;
1381 *errorcodeptr = ERR34;
1382 }
1383
1384 else if (*ptr == CHAR_RIGHT_CURLY_BRACKET)
1385 {
1386 if (utf && c >= 0xd800 && c <= 0xdfff) *errorcodeptr = ERR73;
1387 }
1388
1389 /* If the sequence of hex digits does not end with '}', give an error.
1390 We used just to recognize this construct and fall through to the normal
1391 \x handling, but nowadays Perl gives an error, which seems much more
1392 sensible, so we do too. */
1393
1394 else *errorcodeptr = ERR79;
1395 } /* End of \x{} processing */
1396
1397 /* Read a single-byte hex-defined char (up to two hex digits after \x) */
1398
1399 else
1400 {
1401 c = 0;
1402 while (i++ < 2 && MAX_255(ptr[1]) && (digitab[ptr[1]] & ctype_xdigit) != 0)
1403 {
1404 pcre_uint32 cc; /* Some compilers don't like */
1405 cc = *(++ptr); /* ++ in initializers */
1406 #ifndef EBCDIC /* ASCII/UTF-8 coding */
1407 if (cc >= CHAR_a) cc -= 32; /* Convert to upper case */
1408 c = c * 16 + cc - ((cc < CHAR_A)? CHAR_0 : (CHAR_A - 10));
1409 #else /* EBCDIC coding */
1410 if (cc <= CHAR_z) cc += 64; /* Convert to upper case */
1411 c = c * 16 + cc - ((cc >= CHAR_0)? CHAR_0 : (CHAR_A - 10));
1412 #endif
1413 }
1414 } /* End of \xdd handling */
1415 } /* End of Perl-style \x handling */
1416 break;
1417
1418 /* For \c, a following letter is upper-cased; then the 0x40 bit is flipped.
1419 An error is given if the byte following \c is not an ASCII character. This
1420 coding is ASCII-specific, but then the whole concept of \cx is
1421 ASCII-specific. (However, an EBCDIC equivalent has now been added.) */
1422
1423 case CHAR_c:
1424 c = *(++ptr);
1425 if (c == CHAR_NULL)
1426 {
1427 *errorcodeptr = ERR2;
1428 break;
1429 }
1430 #ifndef EBCDIC /* ASCII/UTF-8 coding */
1431 if (c > 127) /* Excludes all non-ASCII in either mode */
1432 {
1433 *errorcodeptr = ERR68;
1434 break;
1435 }
1436 if (c >= CHAR_a && c <= CHAR_z) c -= 32;
1437 c ^= 0x40;
1438 #else /* EBCDIC coding */
1439 if (c >= CHAR_a && c <= CHAR_z) c += 64;
1440 if (c == CHAR_QUESTION_MARK)
1441 c = ('\\' == 188 && '`' == 74)? 0x5f : 0xff;
1442 else
1443 {
1444 for (i = 0; i < 32; i++)
1445 {
1446 if (c == ebcdic_escape_c[i]) break;
1447 }
1448 if (i < 32) c = i; else *errorcodeptr = ERR68;
1449 }
1450 #endif
1451 break;
1452
1453 /* PCRE_EXTRA enables extensions to Perl in the matter of escapes. Any
1454 other alphanumeric following \ is an error if PCRE_EXTRA was set;
1455 otherwise, for Perl compatibility, it is a literal. This code looks a bit
1456 odd, but there used to be some cases other than the default, and there may
1457 be again in future, so I haven't "optimized" it. */
1458
1459 default:
1460 if ((options & PCRE_EXTRA) != 0) switch(c)
1461 {
1462 default:
1463 *errorcodeptr = ERR3;
1464 break;
1465 }
1466 break;
1467 }
1468 }
1469
1470 /* Perl supports \N{name} for character names, as well as plain \N for "not
1471 newline". PCRE does not support \N{name}. However, it does support
1472 quantification such as \N{2,3}. */
1473
1474 if (escape == ESC_N && ptr[1] == CHAR_LEFT_CURLY_BRACKET &&
1475 !is_counted_repeat(ptr+2))
1476 *errorcodeptr = ERR37;
1477
1478 /* If PCRE_UCP is set, we change the values for \d etc. */
1479
1480 if ((options & PCRE_UCP) != 0 && escape >= ESC_D && escape <= ESC_w)
1481 escape += (ESC_DU - ESC_D);
1482
1483 /* Set the pointer to the final character before returning. */
1484
1485 *ptrptr = ptr;
1486 *chptr = c;
1487 return escape;
1488 }
1489
1490
1491
1492 #ifdef SUPPORT_UCP
1493 /*************************************************
1494 * Handle \P and \p *
1495 *************************************************/
1496
1497 /* This function is called after \P or \p has been encountered, provided that
1498 PCRE is compiled with support for Unicode properties. On entry, ptrptr is
1499 pointing at the P or p. On exit, it is pointing at the final character of the
1500 escape sequence.
1501
1502 Argument:
1503 ptrptr points to the pattern position pointer
1504 negptr points to a boolean that is set TRUE for negation else FALSE
1505 ptypeptr points to an unsigned int that is set to the type value
1506 pdataptr points to an unsigned int that is set to the detailed property value
1507 errorcodeptr points to the error code variable
1508
1509 Returns: TRUE if the type value was found, or FALSE for an invalid type
1510 */
1511
1512 static BOOL
get_ucp(const pcre_uchar ** ptrptr,BOOL * negptr,unsigned int * ptypeptr,unsigned int * pdataptr,int * errorcodeptr)1513 get_ucp(const pcre_uchar **ptrptr, BOOL *negptr, unsigned int *ptypeptr,
1514 unsigned int *pdataptr, int *errorcodeptr)
1515 {
1516 pcre_uchar c;
1517 int i, bot, top;
1518 const pcre_uchar *ptr = *ptrptr;
1519 pcre_uchar name[32];
1520
1521 c = *(++ptr);
1522 if (c == CHAR_NULL) goto ERROR_RETURN;
1523
1524 *negptr = FALSE;
1525
1526 /* \P or \p can be followed by a name in {}, optionally preceded by ^ for
1527 negation. */
1528
1529 if (c == CHAR_LEFT_CURLY_BRACKET)
1530 {
1531 if (ptr[1] == CHAR_CIRCUMFLEX_ACCENT)
1532 {
1533 *negptr = TRUE;
1534 ptr++;
1535 }
1536 for (i = 0; i < (int)(sizeof(name) / sizeof(pcre_uchar)) - 1; i++)
1537 {
1538 c = *(++ptr);
1539 if (c == CHAR_NULL) goto ERROR_RETURN;
1540 if (c == CHAR_RIGHT_CURLY_BRACKET) break;
1541 name[i] = c;
1542 }
1543 if (c != CHAR_RIGHT_CURLY_BRACKET) goto ERROR_RETURN;
1544 name[i] = 0;
1545 }
1546
1547 /* Otherwise there is just one following character */
1548
1549 else
1550 {
1551 name[0] = c;
1552 name[1] = 0;
1553 }
1554
1555 *ptrptr = ptr;
1556
1557 /* Search for a recognized property name using binary chop */
1558
1559 bot = 0;
1560 top = PRIV(utt_size);
1561
1562 while (bot < top)
1563 {
1564 int r;
1565 i = (bot + top) >> 1;
1566 r = STRCMP_UC_C8(name, PRIV(utt_names) + PRIV(utt)[i].name_offset);
1567 if (r == 0)
1568 {
1569 *ptypeptr = PRIV(utt)[i].type;
1570 *pdataptr = PRIV(utt)[i].value;
1571 return TRUE;
1572 }
1573 if (r > 0) bot = i + 1; else top = i;
1574 }
1575
1576 *errorcodeptr = ERR47;
1577 *ptrptr = ptr;
1578 return FALSE;
1579
1580 ERROR_RETURN:
1581 *errorcodeptr = ERR46;
1582 *ptrptr = ptr;
1583 return FALSE;
1584 }
1585 #endif
1586
1587
1588
1589 /*************************************************
1590 * Read repeat counts *
1591 *************************************************/
1592
1593 /* Read an item of the form {n,m} and return the values. This is called only
1594 after is_counted_repeat() has confirmed that a repeat-count quantifier exists,
1595 so the syntax is guaranteed to be correct, but we need to check the values.
1596
1597 Arguments:
1598 p pointer to first char after '{'
1599 minp pointer to int for min
1600 maxp pointer to int for max
1601 returned as -1 if no max
1602 errorcodeptr points to error code variable
1603
1604 Returns: pointer to '}' on success;
1605 current ptr on error, with errorcodeptr set non-zero
1606 */
1607
1608 static const pcre_uchar *
read_repeat_counts(const pcre_uchar * p,int * minp,int * maxp,int * errorcodeptr)1609 read_repeat_counts(const pcre_uchar *p, int *minp, int *maxp, int *errorcodeptr)
1610 {
1611 int min = 0;
1612 int max = -1;
1613
1614 while (IS_DIGIT(*p))
1615 {
1616 min = min * 10 + (int)(*p++ - CHAR_0);
1617 if (min > 65535)
1618 {
1619 *errorcodeptr = ERR5;
1620 return p;
1621 }
1622 }
1623
1624 if (*p == CHAR_RIGHT_CURLY_BRACKET) max = min; else
1625 {
1626 if (*(++p) != CHAR_RIGHT_CURLY_BRACKET)
1627 {
1628 max = 0;
1629 while(IS_DIGIT(*p))
1630 {
1631 max = max * 10 + (int)(*p++ - CHAR_0);
1632 if (max > 65535)
1633 {
1634 *errorcodeptr = ERR5;
1635 return p;
1636 }
1637 }
1638 if (max < min)
1639 {
1640 *errorcodeptr = ERR4;
1641 return p;
1642 }
1643 }
1644 }
1645
1646 *minp = min;
1647 *maxp = max;
1648 return p;
1649 }
1650
1651
1652
1653 /*************************************************
1654 * Find first significant op code *
1655 *************************************************/
1656
1657 /* This is called by several functions that scan a compiled expression looking
1658 for a fixed first character, or an anchoring op code etc. It skips over things
1659 that do not influence this. For some calls, it makes sense to skip negative
1660 forward and all backward assertions, and also the \b assertion; for others it
1661 does not.
1662
1663 Arguments:
1664 code pointer to the start of the group
1665 skipassert TRUE if certain assertions are to be skipped
1666
1667 Returns: pointer to the first significant opcode
1668 */
1669
1670 static const pcre_uchar*
first_significant_code(const pcre_uchar * code,BOOL skipassert)1671 first_significant_code(const pcre_uchar *code, BOOL skipassert)
1672 {
1673 for (;;)
1674 {
1675 switch ((int)*code)
1676 {
1677 case OP_ASSERT_NOT:
1678 case OP_ASSERTBACK:
1679 case OP_ASSERTBACK_NOT:
1680 if (!skipassert) return code;
1681 do code += GET(code, 1); while (*code == OP_ALT);
1682 code += PRIV(OP_lengths)[*code];
1683 break;
1684
1685 case OP_WORD_BOUNDARY:
1686 case OP_NOT_WORD_BOUNDARY:
1687 if (!skipassert) return code;
1688 /* Fall through */
1689
1690 case OP_CALLOUT:
1691 case OP_CREF:
1692 case OP_DNCREF:
1693 case OP_RREF:
1694 case OP_DNRREF:
1695 case OP_DEF:
1696 code += PRIV(OP_lengths)[*code];
1697 break;
1698
1699 default:
1700 return code;
1701 }
1702 }
1703 /* Control never reaches here */
1704 }
1705
1706
1707
1708 /*************************************************
1709 * Find the fixed length of a branch *
1710 *************************************************/
1711
1712 /* Scan a branch and compute the fixed length of subject that will match it,
1713 if the length is fixed. This is needed for dealing with backward assertions.
1714 In UTF8 mode, the result is in characters rather than bytes. The branch is
1715 temporarily terminated with OP_END when this function is called.
1716
1717 This function is called when a backward assertion is encountered, so that if it
1718 fails, the error message can point to the correct place in the pattern.
1719 However, we cannot do this when the assertion contains subroutine calls,
1720 because they can be forward references. We solve this by remembering this case
1721 and doing the check at the end; a flag specifies which mode we are running in.
1722
1723 Arguments:
1724 code points to the start of the pattern (the bracket)
1725 utf TRUE in UTF-8 / UTF-16 / UTF-32 mode
1726 atend TRUE if called when the pattern is complete
1727 cd the "compile data" structure
1728 recurses chain of recurse_check to catch mutual recursion
1729
1730 Returns: the fixed length,
1731 or -1 if there is no fixed length,
1732 or -2 if \C was encountered (in UTF-8 mode only)
1733 or -3 if an OP_RECURSE item was encountered and atend is FALSE
1734 or -4 if an unknown opcode was encountered (internal error)
1735 */
1736
1737 static int
find_fixedlength(pcre_uchar * code,BOOL utf,BOOL atend,compile_data * cd,recurse_check * recurses)1738 find_fixedlength(pcre_uchar *code, BOOL utf, BOOL atend, compile_data *cd,
1739 recurse_check *recurses)
1740 {
1741 int length = -1;
1742 recurse_check this_recurse;
1743 register int branchlength = 0;
1744 register pcre_uchar *cc = code + 1 + LINK_SIZE;
1745
1746 /* Scan along the opcodes for this branch. If we get to the end of the
1747 branch, check the length against that of the other branches. */
1748
1749 for (;;)
1750 {
1751 int d;
1752 pcre_uchar *ce, *cs;
1753 register pcre_uchar op = *cc;
1754
1755 switch (op)
1756 {
1757 /* We only need to continue for OP_CBRA (normal capturing bracket) and
1758 OP_BRA (normal non-capturing bracket) because the other variants of these
1759 opcodes are all concerned with unlimited repeated groups, which of course
1760 are not of fixed length. */
1761
1762 case OP_CBRA:
1763 case OP_BRA:
1764 case OP_ONCE:
1765 case OP_ONCE_NC:
1766 case OP_COND:
1767 d = find_fixedlength(cc + ((op == OP_CBRA)? IMM2_SIZE : 0), utf, atend, cd,
1768 recurses);
1769 if (d < 0) return d;
1770 branchlength += d;
1771 do cc += GET(cc, 1); while (*cc == OP_ALT);
1772 cc += 1 + LINK_SIZE;
1773 break;
1774
1775 /* Reached end of a branch; if it's a ket it is the end of a nested call.
1776 If it's ALT it is an alternation in a nested call. An ACCEPT is effectively
1777 an ALT. If it is END it's the end of the outer call. All can be handled by
1778 the same code. Note that we must not include the OP_KETRxxx opcodes here,
1779 because they all imply an unlimited repeat. */
1780
1781 case OP_ALT:
1782 case OP_KET:
1783 case OP_END:
1784 case OP_ACCEPT:
1785 case OP_ASSERT_ACCEPT:
1786 if (length < 0) length = branchlength;
1787 else if (length != branchlength) return -1;
1788 if (*cc != OP_ALT) return length;
1789 cc += 1 + LINK_SIZE;
1790 branchlength = 0;
1791 break;
1792
1793 /* A true recursion implies not fixed length, but a subroutine call may
1794 be OK. If the subroutine is a forward reference, we can't deal with
1795 it until the end of the pattern, so return -3. */
1796
1797 case OP_RECURSE:
1798 if (!atend) return -3;
1799 cs = ce = (pcre_uchar *)cd->start_code + GET(cc, 1); /* Start subpattern */
1800 do ce += GET(ce, 1); while (*ce == OP_ALT); /* End subpattern */
1801 if (cc > cs && cc < ce) return -1; /* Recursion */
1802 else /* Check for mutual recursion */
1803 {
1804 recurse_check *r = recurses;
1805 for (r = recurses; r != NULL; r = r->prev) if (r->group == cs) break;
1806 if (r != NULL) return -1; /* Mutual recursion */
1807 }
1808 this_recurse.prev = recurses;
1809 this_recurse.group = cs;
1810 d = find_fixedlength(cs + IMM2_SIZE, utf, atend, cd, &this_recurse);
1811 if (d < 0) return d;
1812 branchlength += d;
1813 cc += 1 + LINK_SIZE;
1814 break;
1815
1816 /* Skip over assertive subpatterns */
1817
1818 case OP_ASSERT:
1819 case OP_ASSERT_NOT:
1820 case OP_ASSERTBACK:
1821 case OP_ASSERTBACK_NOT:
1822 do cc += GET(cc, 1); while (*cc == OP_ALT);
1823 cc += 1 + LINK_SIZE;
1824 break;
1825
1826 /* Skip over things that don't match chars */
1827
1828 case OP_MARK:
1829 case OP_PRUNE_ARG:
1830 case OP_SKIP_ARG:
1831 case OP_THEN_ARG:
1832 cc += cc[1] + PRIV(OP_lengths)[*cc];
1833 break;
1834
1835 case OP_CALLOUT:
1836 case OP_CIRC:
1837 case OP_CIRCM:
1838 case OP_CLOSE:
1839 case OP_COMMIT:
1840 case OP_CREF:
1841 case OP_DEF:
1842 case OP_DNCREF:
1843 case OP_DNRREF:
1844 case OP_DOLL:
1845 case OP_DOLLM:
1846 case OP_EOD:
1847 case OP_EODN:
1848 case OP_FAIL:
1849 case OP_NOT_WORD_BOUNDARY:
1850 case OP_PRUNE:
1851 case OP_REVERSE:
1852 case OP_RREF:
1853 case OP_SET_SOM:
1854 case OP_SKIP:
1855 case OP_SOD:
1856 case OP_SOM:
1857 case OP_THEN:
1858 case OP_WORD_BOUNDARY:
1859 cc += PRIV(OP_lengths)[*cc];
1860 break;
1861
1862 /* Handle literal characters */
1863
1864 case OP_CHAR:
1865 case OP_CHARI:
1866 case OP_NOT:
1867 case OP_NOTI:
1868 branchlength++;
1869 cc += 2;
1870 #ifdef SUPPORT_UTF
1871 if (utf && HAS_EXTRALEN(cc[-1])) cc += GET_EXTRALEN(cc[-1]);
1872 #endif
1873 break;
1874
1875 /* Handle exact repetitions. The count is already in characters, but we
1876 need to skip over a multibyte character in UTF8 mode. */
1877
1878 case OP_EXACT:
1879 case OP_EXACTI:
1880 case OP_NOTEXACT:
1881 case OP_NOTEXACTI:
1882 branchlength += (int)GET2(cc,1);
1883 cc += 2 + IMM2_SIZE;
1884 #ifdef SUPPORT_UTF
1885 if (utf && HAS_EXTRALEN(cc[-1])) cc += GET_EXTRALEN(cc[-1]);
1886 #endif
1887 break;
1888
1889 case OP_TYPEEXACT:
1890 branchlength += GET2(cc,1);
1891 if (cc[1 + IMM2_SIZE] == OP_PROP || cc[1 + IMM2_SIZE] == OP_NOTPROP)
1892 cc += 2;
1893 cc += 1 + IMM2_SIZE + 1;
1894 break;
1895
1896 /* Handle single-char matchers */
1897
1898 case OP_PROP:
1899 case OP_NOTPROP:
1900 cc += 2;
1901 /* Fall through */
1902
1903 case OP_HSPACE:
1904 case OP_VSPACE:
1905 case OP_NOT_HSPACE:
1906 case OP_NOT_VSPACE:
1907 case OP_NOT_DIGIT:
1908 case OP_DIGIT:
1909 case OP_NOT_WHITESPACE:
1910 case OP_WHITESPACE:
1911 case OP_NOT_WORDCHAR:
1912 case OP_WORDCHAR:
1913 case OP_ANY:
1914 case OP_ALLANY:
1915 branchlength++;
1916 cc++;
1917 break;
1918
1919 /* The single-byte matcher isn't allowed. This only happens in UTF-8 mode;
1920 otherwise \C is coded as OP_ALLANY. */
1921
1922 case OP_ANYBYTE:
1923 return -2;
1924
1925 /* Check a class for variable quantification */
1926
1927 case OP_CLASS:
1928 case OP_NCLASS:
1929 #if defined SUPPORT_UTF || defined COMPILE_PCRE16 || defined COMPILE_PCRE32
1930 case OP_XCLASS:
1931 /* The original code caused an unsigned overflow in 64 bit systems,
1932 so now we use a conditional statement. */
1933 if (op == OP_XCLASS)
1934 cc += GET(cc, 1);
1935 else
1936 cc += PRIV(OP_lengths)[OP_CLASS];
1937 #else
1938 cc += PRIV(OP_lengths)[OP_CLASS];
1939 #endif
1940
1941 switch (*cc)
1942 {
1943 case OP_CRSTAR:
1944 case OP_CRMINSTAR:
1945 case OP_CRPLUS:
1946 case OP_CRMINPLUS:
1947 case OP_CRQUERY:
1948 case OP_CRMINQUERY:
1949 case OP_CRPOSSTAR:
1950 case OP_CRPOSPLUS:
1951 case OP_CRPOSQUERY:
1952 return -1;
1953
1954 case OP_CRRANGE:
1955 case OP_CRMINRANGE:
1956 case OP_CRPOSRANGE:
1957 if (GET2(cc,1) != GET2(cc,1+IMM2_SIZE)) return -1;
1958 branchlength += (int)GET2(cc,1);
1959 cc += 1 + 2 * IMM2_SIZE;
1960 break;
1961
1962 default:
1963 branchlength++;
1964 }
1965 break;
1966
1967 /* Anything else is variable length */
1968
1969 case OP_ANYNL:
1970 case OP_BRAMINZERO:
1971 case OP_BRAPOS:
1972 case OP_BRAPOSZERO:
1973 case OP_BRAZERO:
1974 case OP_CBRAPOS:
1975 case OP_EXTUNI:
1976 case OP_KETRMAX:
1977 case OP_KETRMIN:
1978 case OP_KETRPOS:
1979 case OP_MINPLUS:
1980 case OP_MINPLUSI:
1981 case OP_MINQUERY:
1982 case OP_MINQUERYI:
1983 case OP_MINSTAR:
1984 case OP_MINSTARI:
1985 case OP_MINUPTO:
1986 case OP_MINUPTOI:
1987 case OP_NOTMINPLUS:
1988 case OP_NOTMINPLUSI:
1989 case OP_NOTMINQUERY:
1990 case OP_NOTMINQUERYI:
1991 case OP_NOTMINSTAR:
1992 case OP_NOTMINSTARI:
1993 case OP_NOTMINUPTO:
1994 case OP_NOTMINUPTOI:
1995 case OP_NOTPLUS:
1996 case OP_NOTPLUSI:
1997 case OP_NOTPOSPLUS:
1998 case OP_NOTPOSPLUSI:
1999 case OP_NOTPOSQUERY:
2000 case OP_NOTPOSQUERYI:
2001 case OP_NOTPOSSTAR:
2002 case OP_NOTPOSSTARI:
2003 case OP_NOTPOSUPTO:
2004 case OP_NOTPOSUPTOI:
2005 case OP_NOTQUERY:
2006 case OP_NOTQUERYI:
2007 case OP_NOTSTAR:
2008 case OP_NOTSTARI:
2009 case OP_NOTUPTO:
2010 case OP_NOTUPTOI:
2011 case OP_PLUS:
2012 case OP_PLUSI:
2013 case OP_POSPLUS:
2014 case OP_POSPLUSI:
2015 case OP_POSQUERY:
2016 case OP_POSQUERYI:
2017 case OP_POSSTAR:
2018 case OP_POSSTARI:
2019 case OP_POSUPTO:
2020 case OP_POSUPTOI:
2021 case OP_QUERY:
2022 case OP_QUERYI:
2023 case OP_REF:
2024 case OP_REFI:
2025 case OP_DNREF:
2026 case OP_DNREFI:
2027 case OP_SBRA:
2028 case OP_SBRAPOS:
2029 case OP_SCBRA:
2030 case OP_SCBRAPOS:
2031 case OP_SCOND:
2032 case OP_SKIPZERO:
2033 case OP_STAR:
2034 case OP_STARI:
2035 case OP_TYPEMINPLUS:
2036 case OP_TYPEMINQUERY:
2037 case OP_TYPEMINSTAR:
2038 case OP_TYPEMINUPTO:
2039 case OP_TYPEPLUS:
2040 case OP_TYPEPOSPLUS:
2041 case OP_TYPEPOSQUERY:
2042 case OP_TYPEPOSSTAR:
2043 case OP_TYPEPOSUPTO:
2044 case OP_TYPEQUERY:
2045 case OP_TYPESTAR:
2046 case OP_TYPEUPTO:
2047 case OP_UPTO:
2048 case OP_UPTOI:
2049 return -1;
2050
2051 /* Catch unrecognized opcodes so that when new ones are added they
2052 are not forgotten, as has happened in the past. */
2053
2054 default:
2055 return -4;
2056 }
2057 }
2058 /* Control never gets here */
2059 }
2060
2061
2062
2063 /*************************************************
2064 * Scan compiled regex for specific bracket *
2065 *************************************************/
2066
2067 /* This little function scans through a compiled pattern until it finds a
2068 capturing bracket with the given number, or, if the number is negative, an
2069 instance of OP_REVERSE for a lookbehind. The function is global in the C sense
2070 so that it can be called from pcre_study() when finding the minimum matching
2071 length.
2072
2073 Arguments:
2074 code points to start of expression
2075 utf TRUE in UTF-8 / UTF-16 / UTF-32 mode
2076 number the required bracket number or negative to find a lookbehind
2077
2078 Returns: pointer to the opcode for the bracket, or NULL if not found
2079 */
2080
2081 const pcre_uchar *
PRIV(find_bracket)2082 PRIV(find_bracket)(const pcre_uchar *code, BOOL utf, int number)
2083 {
2084 for (;;)
2085 {
2086 register pcre_uchar c = *code;
2087
2088 if (c == OP_END) return NULL;
2089
2090 /* XCLASS is used for classes that cannot be represented just by a bit
2091 map. This includes negated single high-valued characters. The length in
2092 the table is zero; the actual length is stored in the compiled code. */
2093
2094 if (c == OP_XCLASS) code += GET(code, 1);
2095
2096 /* Handle recursion */
2097
2098 else if (c == OP_REVERSE)
2099 {
2100 if (number < 0) return (pcre_uchar *)code;
2101 code += PRIV(OP_lengths)[c];
2102 }
2103
2104 /* Handle capturing bracket */
2105
2106 else if (c == OP_CBRA || c == OP_SCBRA ||
2107 c == OP_CBRAPOS || c == OP_SCBRAPOS)
2108 {
2109 int n = (int)GET2(code, 1+LINK_SIZE);
2110 if (n == number) return (pcre_uchar *)code;
2111 code += PRIV(OP_lengths)[c];
2112 }
2113
2114 /* Otherwise, we can get the item's length from the table, except that for
2115 repeated character types, we have to test for \p and \P, which have an extra
2116 two bytes of parameters, and for MARK/PRUNE/SKIP/THEN with an argument, we
2117 must add in its length. */
2118
2119 else
2120 {
2121 switch(c)
2122 {
2123 case OP_TYPESTAR:
2124 case OP_TYPEMINSTAR:
2125 case OP_TYPEPLUS:
2126 case OP_TYPEMINPLUS:
2127 case OP_TYPEQUERY:
2128 case OP_TYPEMINQUERY:
2129 case OP_TYPEPOSSTAR:
2130 case OP_TYPEPOSPLUS:
2131 case OP_TYPEPOSQUERY:
2132 if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
2133 break;
2134
2135 case OP_TYPEUPTO:
2136 case OP_TYPEMINUPTO:
2137 case OP_TYPEEXACT:
2138 case OP_TYPEPOSUPTO:
2139 if (code[1 + IMM2_SIZE] == OP_PROP || code[1 + IMM2_SIZE] == OP_NOTPROP)
2140 code += 2;
2141 break;
2142
2143 case OP_MARK:
2144 case OP_PRUNE_ARG:
2145 case OP_SKIP_ARG:
2146 case OP_THEN_ARG:
2147 code += code[1];
2148 break;
2149 }
2150
2151 /* Add in the fixed length from the table */
2152
2153 code += PRIV(OP_lengths)[c];
2154
2155 /* In UTF-8 mode, opcodes that are followed by a character may be followed by
2156 a multi-byte character. The length in the table is a minimum, so we have to
2157 arrange to skip the extra bytes. */
2158
2159 #if defined SUPPORT_UTF && !defined COMPILE_PCRE32
2160 if (utf) switch(c)
2161 {
2162 case OP_CHAR:
2163 case OP_CHARI:
2164 case OP_NOT:
2165 case OP_NOTI:
2166 case OP_EXACT:
2167 case OP_EXACTI:
2168 case OP_NOTEXACT:
2169 case OP_NOTEXACTI:
2170 case OP_UPTO:
2171 case OP_UPTOI:
2172 case OP_NOTUPTO:
2173 case OP_NOTUPTOI:
2174 case OP_MINUPTO:
2175 case OP_MINUPTOI:
2176 case OP_NOTMINUPTO:
2177 case OP_NOTMINUPTOI:
2178 case OP_POSUPTO:
2179 case OP_POSUPTOI:
2180 case OP_NOTPOSUPTO:
2181 case OP_NOTPOSUPTOI:
2182 case OP_STAR:
2183 case OP_STARI:
2184 case OP_NOTSTAR:
2185 case OP_NOTSTARI:
2186 case OP_MINSTAR:
2187 case OP_MINSTARI:
2188 case OP_NOTMINSTAR:
2189 case OP_NOTMINSTARI:
2190 case OP_POSSTAR:
2191 case OP_POSSTARI:
2192 case OP_NOTPOSSTAR:
2193 case OP_NOTPOSSTARI:
2194 case OP_PLUS:
2195 case OP_PLUSI:
2196 case OP_NOTPLUS:
2197 case OP_NOTPLUSI:
2198 case OP_MINPLUS:
2199 case OP_MINPLUSI:
2200 case OP_NOTMINPLUS:
2201 case OP_NOTMINPLUSI:
2202 case OP_POSPLUS:
2203 case OP_POSPLUSI:
2204 case OP_NOTPOSPLUS:
2205 case OP_NOTPOSPLUSI:
2206 case OP_QUERY:
2207 case OP_QUERYI:
2208 case OP_NOTQUERY:
2209 case OP_NOTQUERYI:
2210 case OP_MINQUERY:
2211 case OP_MINQUERYI:
2212 case OP_NOTMINQUERY:
2213 case OP_NOTMINQUERYI:
2214 case OP_POSQUERY:
2215 case OP_POSQUERYI:
2216 case OP_NOTPOSQUERY:
2217 case OP_NOTPOSQUERYI:
2218 if (HAS_EXTRALEN(code[-1])) code += GET_EXTRALEN(code[-1]);
2219 break;
2220 }
2221 #else
2222 (void)(utf); /* Keep compiler happy by referencing function argument */
2223 #endif
2224 }
2225 }
2226 }
2227
2228
2229
2230 /*************************************************
2231 * Scan compiled regex for recursion reference *
2232 *************************************************/
2233
2234 /* This little function scans through a compiled pattern until it finds an
2235 instance of OP_RECURSE.
2236
2237 Arguments:
2238 code points to start of expression
2239 utf TRUE in UTF-8 / UTF-16 / UTF-32 mode
2240
2241 Returns: pointer to the opcode for OP_RECURSE, or NULL if not found
2242 */
2243
2244 static const pcre_uchar *
find_recurse(const pcre_uchar * code,BOOL utf)2245 find_recurse(const pcre_uchar *code, BOOL utf)
2246 {
2247 for (;;)
2248 {
2249 register pcre_uchar c = *code;
2250 if (c == OP_END) return NULL;
2251 if (c == OP_RECURSE) return code;
2252
2253 /* XCLASS is used for classes that cannot be represented just by a bit
2254 map. This includes negated single high-valued characters. The length in
2255 the table is zero; the actual length is stored in the compiled code. */
2256
2257 if (c == OP_XCLASS) code += GET(code, 1);
2258
2259 /* Otherwise, we can get the item's length from the table, except that for
2260 repeated character types, we have to test for \p and \P, which have an extra
2261 two bytes of parameters, and for MARK/PRUNE/SKIP/THEN with an argument, we
2262 must add in its length. */
2263
2264 else
2265 {
2266 switch(c)
2267 {
2268 case OP_TYPESTAR:
2269 case OP_TYPEMINSTAR:
2270 case OP_TYPEPLUS:
2271 case OP_TYPEMINPLUS:
2272 case OP_TYPEQUERY:
2273 case OP_TYPEMINQUERY:
2274 case OP_TYPEPOSSTAR:
2275 case OP_TYPEPOSPLUS:
2276 case OP_TYPEPOSQUERY:
2277 if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
2278 break;
2279
2280 case OP_TYPEPOSUPTO:
2281 case OP_TYPEUPTO:
2282 case OP_TYPEMINUPTO:
2283 case OP_TYPEEXACT:
2284 if (code[1 + IMM2_SIZE] == OP_PROP || code[1 + IMM2_SIZE] == OP_NOTPROP)
2285 code += 2;
2286 break;
2287
2288 case OP_MARK:
2289 case OP_PRUNE_ARG:
2290 case OP_SKIP_ARG:
2291 case OP_THEN_ARG:
2292 code += code[1];
2293 break;
2294 }
2295
2296 /* Add in the fixed length from the table */
2297
2298 code += PRIV(OP_lengths)[c];
2299
2300 /* In UTF-8 mode, opcodes that are followed by a character may be followed
2301 by a multi-byte character. The length in the table is a minimum, so we have
2302 to arrange to skip the extra bytes. */
2303
2304 #if defined SUPPORT_UTF && !defined COMPILE_PCRE32
2305 if (utf) switch(c)
2306 {
2307 case OP_CHAR:
2308 case OP_CHARI:
2309 case OP_NOT:
2310 case OP_NOTI:
2311 case OP_EXACT:
2312 case OP_EXACTI:
2313 case OP_NOTEXACT:
2314 case OP_NOTEXACTI:
2315 case OP_UPTO:
2316 case OP_UPTOI:
2317 case OP_NOTUPTO:
2318 case OP_NOTUPTOI:
2319 case OP_MINUPTO:
2320 case OP_MINUPTOI:
2321 case OP_NOTMINUPTO:
2322 case OP_NOTMINUPTOI:
2323 case OP_POSUPTO:
2324 case OP_POSUPTOI:
2325 case OP_NOTPOSUPTO:
2326 case OP_NOTPOSUPTOI:
2327 case OP_STAR:
2328 case OP_STARI:
2329 case OP_NOTSTAR:
2330 case OP_NOTSTARI:
2331 case OP_MINSTAR:
2332 case OP_MINSTARI:
2333 case OP_NOTMINSTAR:
2334 case OP_NOTMINSTARI:
2335 case OP_POSSTAR:
2336 case OP_POSSTARI:
2337 case OP_NOTPOSSTAR:
2338 case OP_NOTPOSSTARI:
2339 case OP_PLUS:
2340 case OP_PLUSI:
2341 case OP_NOTPLUS:
2342 case OP_NOTPLUSI:
2343 case OP_MINPLUS:
2344 case OP_MINPLUSI:
2345 case OP_NOTMINPLUS:
2346 case OP_NOTMINPLUSI:
2347 case OP_POSPLUS:
2348 case OP_POSPLUSI:
2349 case OP_NOTPOSPLUS:
2350 case OP_NOTPOSPLUSI:
2351 case OP_QUERY:
2352 case OP_QUERYI:
2353 case OP_NOTQUERY:
2354 case OP_NOTQUERYI:
2355 case OP_MINQUERY:
2356 case OP_MINQUERYI:
2357 case OP_NOTMINQUERY:
2358 case OP_NOTMINQUERYI:
2359 case OP_POSQUERY:
2360 case OP_POSQUERYI:
2361 case OP_NOTPOSQUERY:
2362 case OP_NOTPOSQUERYI:
2363 if (HAS_EXTRALEN(code[-1])) code += GET_EXTRALEN(code[-1]);
2364 break;
2365 }
2366 #else
2367 (void)(utf); /* Keep compiler happy by referencing function argument */
2368 #endif
2369 }
2370 }
2371 }
2372
2373
2374
2375 /*************************************************
2376 * Scan compiled branch for non-emptiness *
2377 *************************************************/
2378
2379 /* This function scans through a branch of a compiled pattern to see whether it
2380 can match the empty string or not. It is called from could_be_empty()
2381 below and from compile_branch() when checking for an unlimited repeat of a
2382 group that can match nothing. Note that first_significant_code() skips over
2383 backward and negative forward assertions when its final argument is TRUE. If we
2384 hit an unclosed bracket, we return "empty" - this means we've struck an inner
2385 bracket whose current branch will already have been scanned.
2386
2387 Arguments:
2388 code points to start of search
2389 endcode points to where to stop
2390 utf TRUE if in UTF-8 / UTF-16 / UTF-32 mode
2391 cd contains pointers to tables etc.
2392 recurses chain of recurse_check to catch mutual recursion
2393
2394 Returns: TRUE if what is matched could be empty
2395 */
2396
2397 static BOOL
could_be_empty_branch(const pcre_uchar * code,const pcre_uchar * endcode,BOOL utf,compile_data * cd,recurse_check * recurses)2398 could_be_empty_branch(const pcre_uchar *code, const pcre_uchar *endcode,
2399 BOOL utf, compile_data *cd, recurse_check *recurses)
2400 {
2401 register pcre_uchar c;
2402 recurse_check this_recurse;
2403
2404 for (code = first_significant_code(code + PRIV(OP_lengths)[*code], TRUE);
2405 code < endcode;
2406 code = first_significant_code(code + PRIV(OP_lengths)[c], TRUE))
2407 {
2408 const pcre_uchar *ccode;
2409
2410 c = *code;
2411
2412 /* Skip over forward assertions; the other assertions are skipped by
2413 first_significant_code() with a TRUE final argument. */
2414
2415 if (c == OP_ASSERT)
2416 {
2417 do code += GET(code, 1); while (*code == OP_ALT);
2418 c = *code;
2419 continue;
2420 }
2421
2422 /* For a recursion/subroutine call, if its end has been reached, which
2423 implies a backward reference subroutine call, we can scan it. If it's a
2424 forward reference subroutine call, we can't. To detect forward reference
2425 we have to scan up the list that is kept in the workspace. This function is
2426 called only when doing the real compile, not during the pre-compile that
2427 measures the size of the compiled pattern. */
2428
2429 if (c == OP_RECURSE)
2430 {
2431 const pcre_uchar *scode = cd->start_code + GET(code, 1);
2432 const pcre_uchar *endgroup = scode;
2433 BOOL empty_branch;
2434
2435 /* Test for forward reference or uncompleted reference. This is disabled
2436 when called to scan a completed pattern by setting cd->start_workspace to
2437 NULL. */
2438
2439 if (cd->start_workspace != NULL)
2440 {
2441 const pcre_uchar *tcode;
2442 for (tcode = cd->start_workspace; tcode < cd->hwm; tcode += LINK_SIZE)
2443 if ((int)GET(tcode, 0) == (int)(code + 1 - cd->start_code)) return TRUE;
2444 if (GET(scode, 1) == 0) return TRUE; /* Unclosed */
2445 }
2446
2447 /* If the reference is to a completed group, we need to detect whether this
2448 is a recursive call, as otherwise there will be an infinite loop. If it is
2449 a recursion, just skip over it. Simple recursions are easily detected. For
2450 mutual recursions we keep a chain on the stack. */
2451
2452 do endgroup += GET(endgroup, 1); while (*endgroup == OP_ALT);
2453 if (code >= scode && code <= endgroup) continue; /* Simple recursion */
2454 else
2455 {
2456 recurse_check *r = recurses;
2457 for (r = recurses; r != NULL; r = r->prev)
2458 if (r->group == scode) break;
2459 if (r != NULL) continue; /* Mutual recursion */
2460 }
2461
2462 /* Completed reference; scan the referenced group, remembering it on the
2463 stack chain to detect mutual recursions. */
2464
2465 empty_branch = FALSE;
2466 this_recurse.prev = recurses;
2467 this_recurse.group = scode;
2468
2469 do
2470 {
2471 if (could_be_empty_branch(scode, endcode, utf, cd, &this_recurse))
2472 {
2473 empty_branch = TRUE;
2474 break;
2475 }
2476 scode += GET(scode, 1);
2477 }
2478 while (*scode == OP_ALT);
2479
2480 if (!empty_branch) return FALSE; /* All branches are non-empty */
2481 continue;
2482 }
2483
2484 /* Groups with zero repeats can of course be empty; skip them. */
2485
2486 if (c == OP_BRAZERO || c == OP_BRAMINZERO || c == OP_SKIPZERO ||
2487 c == OP_BRAPOSZERO)
2488 {
2489 code += PRIV(OP_lengths)[c];
2490 do code += GET(code, 1); while (*code == OP_ALT);
2491 c = *code;
2492 continue;
2493 }
2494
2495 /* A nested group that is already marked as "could be empty" can just be
2496 skipped. */
2497
2498 if (c == OP_SBRA || c == OP_SBRAPOS ||
2499 c == OP_SCBRA || c == OP_SCBRAPOS)
2500 {
2501 do code += GET(code, 1); while (*code == OP_ALT);
2502 c = *code;
2503 continue;
2504 }
2505
2506 /* For other groups, scan the branches. */
2507
2508 if (c == OP_BRA || c == OP_BRAPOS ||
2509 c == OP_CBRA || c == OP_CBRAPOS ||
2510 c == OP_ONCE || c == OP_ONCE_NC ||
2511 c == OP_COND || c == OP_SCOND)
2512 {
2513 BOOL empty_branch;
2514 if (GET(code, 1) == 0) return TRUE; /* Hit unclosed bracket */
2515
2516 /* If a conditional group has only one branch, there is a second, implied,
2517 empty branch, so just skip over the conditional, because it could be empty.
2518 Otherwise, scan the individual branches of the group. */
2519
2520 if (c == OP_COND && code[GET(code, 1)] != OP_ALT)
2521 code += GET(code, 1);
2522 else
2523 {
2524 empty_branch = FALSE;
2525 do
2526 {
2527 if (!empty_branch && could_be_empty_branch(code, endcode, utf, cd,
2528 recurses)) empty_branch = TRUE;
2529 code += GET(code, 1);
2530 }
2531 while (*code == OP_ALT);
2532 if (!empty_branch) return FALSE; /* All branches are non-empty */
2533 }
2534
2535 c = *code;
2536 continue;
2537 }
2538
2539 /* Handle the other opcodes */
2540
2541 switch (c)
2542 {
2543 /* Check for quantifiers after a class. XCLASS is used for classes that
2544 cannot be represented just by a bit map. This includes negated single
2545 high-valued characters. The length in PRIV(OP_lengths)[] is zero; the
2546 actual length is stored in the compiled code, so we must update "code"
2547 here. */
2548
2549 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
2550 case OP_XCLASS:
2551 ccode = code += GET(code, 1);
2552 goto CHECK_CLASS_REPEAT;
2553 #endif
2554
2555 case OP_CLASS:
2556 case OP_NCLASS:
2557 ccode = code + PRIV(OP_lengths)[OP_CLASS];
2558
2559 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
2560 CHECK_CLASS_REPEAT:
2561 #endif
2562
2563 switch (*ccode)
2564 {
2565 case OP_CRSTAR: /* These could be empty; continue */
2566 case OP_CRMINSTAR:
2567 case OP_CRQUERY:
2568 case OP_CRMINQUERY:
2569 case OP_CRPOSSTAR:
2570 case OP_CRPOSQUERY:
2571 break;
2572
2573 default: /* Non-repeat => class must match */
2574 case OP_CRPLUS: /* These repeats aren't empty */
2575 case OP_CRMINPLUS:
2576 case OP_CRPOSPLUS:
2577 return FALSE;
2578
2579 case OP_CRRANGE:
2580 case OP_CRMINRANGE:
2581 case OP_CRPOSRANGE:
2582 if (GET2(ccode, 1) > 0) return FALSE; /* Minimum > 0 */
2583 break;
2584 }
2585 break;
2586
2587 /* Opcodes that must match a character */
2588
2589 case OP_ANY:
2590 case OP_ALLANY:
2591 case OP_ANYBYTE:
2592
2593 case OP_PROP:
2594 case OP_NOTPROP:
2595 case OP_ANYNL:
2596
2597 case OP_NOT_HSPACE:
2598 case OP_HSPACE:
2599 case OP_NOT_VSPACE:
2600 case OP_VSPACE:
2601 case OP_EXTUNI:
2602
2603 case OP_NOT_DIGIT:
2604 case OP_DIGIT:
2605 case OP_NOT_WHITESPACE:
2606 case OP_WHITESPACE:
2607 case OP_NOT_WORDCHAR:
2608 case OP_WORDCHAR:
2609
2610 case OP_CHAR:
2611 case OP_CHARI:
2612 case OP_NOT:
2613 case OP_NOTI:
2614
2615 case OP_PLUS:
2616 case OP_PLUSI:
2617 case OP_MINPLUS:
2618 case OP_MINPLUSI:
2619
2620 case OP_NOTPLUS:
2621 case OP_NOTPLUSI:
2622 case OP_NOTMINPLUS:
2623 case OP_NOTMINPLUSI:
2624
2625 case OP_POSPLUS:
2626 case OP_POSPLUSI:
2627 case OP_NOTPOSPLUS:
2628 case OP_NOTPOSPLUSI:
2629
2630 case OP_EXACT:
2631 case OP_EXACTI:
2632 case OP_NOTEXACT:
2633 case OP_NOTEXACTI:
2634
2635 case OP_TYPEPLUS:
2636 case OP_TYPEMINPLUS:
2637 case OP_TYPEPOSPLUS:
2638 case OP_TYPEEXACT:
2639
2640 return FALSE;
2641
2642 /* These are going to continue, as they may be empty, but we have to
2643 fudge the length for the \p and \P cases. */
2644
2645 case OP_TYPESTAR:
2646 case OP_TYPEMINSTAR:
2647 case OP_TYPEPOSSTAR:
2648 case OP_TYPEQUERY:
2649 case OP_TYPEMINQUERY:
2650 case OP_TYPEPOSQUERY:
2651 if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
2652 break;
2653
2654 /* Same for these */
2655
2656 case OP_TYPEUPTO:
2657 case OP_TYPEMINUPTO:
2658 case OP_TYPEPOSUPTO:
2659 if (code[1 + IMM2_SIZE] == OP_PROP || code[1 + IMM2_SIZE] == OP_NOTPROP)
2660 code += 2;
2661 break;
2662
2663 /* End of branch */
2664
2665 case OP_KET:
2666 case OP_KETRMAX:
2667 case OP_KETRMIN:
2668 case OP_KETRPOS:
2669 case OP_ALT:
2670 return TRUE;
2671
2672 /* In UTF-8 mode, STAR, MINSTAR, POSSTAR, QUERY, MINQUERY, POSQUERY, UPTO,
2673 MINUPTO, and POSUPTO and their caseless and negative versions may be
2674 followed by a multibyte character. */
2675
2676 #if defined SUPPORT_UTF && !defined COMPILE_PCRE32
2677 case OP_STAR:
2678 case OP_STARI:
2679 case OP_NOTSTAR:
2680 case OP_NOTSTARI:
2681
2682 case OP_MINSTAR:
2683 case OP_MINSTARI:
2684 case OP_NOTMINSTAR:
2685 case OP_NOTMINSTARI:
2686
2687 case OP_POSSTAR:
2688 case OP_POSSTARI:
2689 case OP_NOTPOSSTAR:
2690 case OP_NOTPOSSTARI:
2691
2692 case OP_QUERY:
2693 case OP_QUERYI:
2694 case OP_NOTQUERY:
2695 case OP_NOTQUERYI:
2696
2697 case OP_MINQUERY:
2698 case OP_MINQUERYI:
2699 case OP_NOTMINQUERY:
2700 case OP_NOTMINQUERYI:
2701
2702 case OP_POSQUERY:
2703 case OP_POSQUERYI:
2704 case OP_NOTPOSQUERY:
2705 case OP_NOTPOSQUERYI:
2706
2707 if (utf && HAS_EXTRALEN(code[1])) code += GET_EXTRALEN(code[1]);
2708 break;
2709
2710 case OP_UPTO:
2711 case OP_UPTOI:
2712 case OP_NOTUPTO:
2713 case OP_NOTUPTOI:
2714
2715 case OP_MINUPTO:
2716 case OP_MINUPTOI:
2717 case OP_NOTMINUPTO:
2718 case OP_NOTMINUPTOI:
2719
2720 case OP_POSUPTO:
2721 case OP_POSUPTOI:
2722 case OP_NOTPOSUPTO:
2723 case OP_NOTPOSUPTOI:
2724
2725 if (utf && HAS_EXTRALEN(code[1 + IMM2_SIZE])) code += GET_EXTRALEN(code[1 + IMM2_SIZE]);
2726 break;
2727 #endif
2728
2729 /* MARK, and PRUNE/SKIP/THEN with an argument must skip over the argument
2730 string. */
2731
2732 case OP_MARK:
2733 case OP_PRUNE_ARG:
2734 case OP_SKIP_ARG:
2735 case OP_THEN_ARG:
2736 code += code[1];
2737 break;
2738
2739 /* None of the remaining opcodes are required to match a character. */
2740
2741 default:
2742 break;
2743 }
2744 }
2745
2746 return TRUE;
2747 }
2748
2749
2750
2751 /*************************************************
2752 * Scan compiled regex for non-emptiness *
2753 *************************************************/
2754
2755 /* This function is called to check for left recursive calls. We want to check
2756 the current branch of the current pattern to see if it could match the empty
2757 string. If it could, we must look outwards for branches at other levels,
2758 stopping when we pass beyond the bracket which is the subject of the recursion.
2759 This function is called only during the real compile, not during the
2760 pre-compile.
2761
2762 Arguments:
2763 code points to start of the recursion
2764 endcode points to where to stop (current RECURSE item)
2765 bcptr points to the chain of current (unclosed) branch starts
2766 utf TRUE if in UTF-8 / UTF-16 / UTF-32 mode
2767 cd pointers to tables etc
2768
2769 Returns: TRUE if what is matched could be empty
2770 */
2771
2772 static BOOL
could_be_empty(const pcre_uchar * code,const pcre_uchar * endcode,branch_chain * bcptr,BOOL utf,compile_data * cd)2773 could_be_empty(const pcre_uchar *code, const pcre_uchar *endcode,
2774 branch_chain *bcptr, BOOL utf, compile_data *cd)
2775 {
2776 while (bcptr != NULL && bcptr->current_branch >= code)
2777 {
2778 if (!could_be_empty_branch(bcptr->current_branch, endcode, utf, cd, NULL))
2779 return FALSE;
2780 bcptr = bcptr->outer;
2781 }
2782 return TRUE;
2783 }
2784
2785
2786
2787 /*************************************************
2788 * Base opcode of repeated opcodes *
2789 *************************************************/
2790
2791 /* Returns the base opcode for repeated single character type opcodes. If the
2792 opcode is not a repeated character type, it returns with the original value.
2793
2794 Arguments: c opcode
2795 Returns: base opcode for the type
2796 */
2797
2798 static pcre_uchar
get_repeat_base(pcre_uchar c)2799 get_repeat_base(pcre_uchar c)
2800 {
2801 return (c > OP_TYPEPOSUPTO)? c :
2802 (c >= OP_TYPESTAR)? OP_TYPESTAR :
2803 (c >= OP_NOTSTARI)? OP_NOTSTARI :
2804 (c >= OP_NOTSTAR)? OP_NOTSTAR :
2805 (c >= OP_STARI)? OP_STARI :
2806 OP_STAR;
2807 }
2808
2809
2810
2811 #ifdef SUPPORT_UCP
2812 /*************************************************
2813 * Check a character and a property *
2814 *************************************************/
2815
2816 /* This function is called by check_auto_possessive() when a property item
2817 is adjacent to a fixed character.
2818
2819 Arguments:
2820 c the character
2821 ptype the property type
2822 pdata the data for the type
2823 negated TRUE if it's a negated property (\P or \p{^)
2824
2825 Returns: TRUE if auto-possessifying is OK
2826 */
2827
2828 static BOOL
check_char_prop(pcre_uint32 c,unsigned int ptype,unsigned int pdata,BOOL negated)2829 check_char_prop(pcre_uint32 c, unsigned int ptype, unsigned int pdata,
2830 BOOL negated)
2831 {
2832 const pcre_uint32 *p;
2833 const ucd_record *prop = GET_UCD(c);
2834
2835 switch(ptype)
2836 {
2837 case PT_LAMP:
2838 return (prop->chartype == ucp_Lu ||
2839 prop->chartype == ucp_Ll ||
2840 prop->chartype == ucp_Lt) == negated;
2841
2842 case PT_GC:
2843 return (pdata == PRIV(ucp_gentype)[prop->chartype]) == negated;
2844
2845 case PT_PC:
2846 return (pdata == prop->chartype) == negated;
2847
2848 case PT_SC:
2849 return (pdata == prop->script) == negated;
2850
2851 /* These are specials */
2852
2853 case PT_ALNUM:
2854 return (PRIV(ucp_gentype)[prop->chartype] == ucp_L ||
2855 PRIV(ucp_gentype)[prop->chartype] == ucp_N) == negated;
2856
2857 /* Perl space used to exclude VT, but from Perl 5.18 it is included, which
2858 means that Perl space and POSIX space are now identical. PCRE was changed
2859 at release 8.34. */
2860
2861 case PT_SPACE: /* Perl space */
2862 case PT_PXSPACE: /* POSIX space */
2863 switch(c)
2864 {
2865 HSPACE_CASES:
2866 VSPACE_CASES:
2867 return negated;
2868
2869 default:
2870 return (PRIV(ucp_gentype)[prop->chartype] == ucp_Z) == negated;
2871 }
2872 break; /* Control never reaches here */
2873
2874 case PT_WORD:
2875 return (PRIV(ucp_gentype)[prop->chartype] == ucp_L ||
2876 PRIV(ucp_gentype)[prop->chartype] == ucp_N ||
2877 c == CHAR_UNDERSCORE) == negated;
2878
2879 case PT_CLIST:
2880 p = PRIV(ucd_caseless_sets) + prop->caseset;
2881 for (;;)
2882 {
2883 if (c < *p) return !negated;
2884 if (c == *p++) return negated;
2885 }
2886 break; /* Control never reaches here */
2887 }
2888
2889 return FALSE;
2890 }
2891 #endif /* SUPPORT_UCP */
2892
2893
2894
2895 /*************************************************
2896 * Fill the character property list *
2897 *************************************************/
2898
2899 /* Checks whether the code points to an opcode that can take part in auto-
2900 possessification, and if so, fills a list with its properties.
2901
2902 Arguments:
2903 code points to start of expression
2904 utf TRUE if in UTF-8 / UTF-16 / UTF-32 mode
2905 fcc points to case-flipping table
2906 list points to output list
2907 list[0] will be filled with the opcode
2908 list[1] will be non-zero if this opcode
2909 can match an empty character string
2910 list[2..7] depends on the opcode
2911
2912 Returns: points to the start of the next opcode if *code is accepted
2913 NULL if *code is not accepted
2914 */
2915
2916 static const pcre_uchar *
get_chr_property_list(const pcre_uchar * code,BOOL utf,const pcre_uint8 * fcc,pcre_uint32 * list)2917 get_chr_property_list(const pcre_uchar *code, BOOL utf,
2918 const pcre_uint8 *fcc, pcre_uint32 *list)
2919 {
2920 pcre_uchar c = *code;
2921 pcre_uchar base;
2922 const pcre_uchar *end;
2923 pcre_uint32 chr;
2924
2925 #ifdef SUPPORT_UCP
2926 pcre_uint32 *clist_dest;
2927 const pcre_uint32 *clist_src;
2928 #else
2929 utf = utf; /* Suppress "unused parameter" compiler warning */
2930 #endif
2931
2932 list[0] = c;
2933 list[1] = FALSE;
2934 code++;
2935
2936 if (c >= OP_STAR && c <= OP_TYPEPOSUPTO)
2937 {
2938 base = get_repeat_base(c);
2939 c -= (base - OP_STAR);
2940
2941 if (c == OP_UPTO || c == OP_MINUPTO || c == OP_EXACT || c == OP_POSUPTO)
2942 code += IMM2_SIZE;
2943
2944 list[1] = (c != OP_PLUS && c != OP_MINPLUS && c != OP_EXACT && c != OP_POSPLUS);
2945
2946 switch(base)
2947 {
2948 case OP_STAR:
2949 list[0] = OP_CHAR;
2950 break;
2951
2952 case OP_STARI:
2953 list[0] = OP_CHARI;
2954 break;
2955
2956 case OP_NOTSTAR:
2957 list[0] = OP_NOT;
2958 break;
2959
2960 case OP_NOTSTARI:
2961 list[0] = OP_NOTI;
2962 break;
2963
2964 case OP_TYPESTAR:
2965 list[0] = *code;
2966 code++;
2967 break;
2968 }
2969 c = list[0];
2970 }
2971
2972 switch(c)
2973 {
2974 case OP_NOT_DIGIT:
2975 case OP_DIGIT:
2976 case OP_NOT_WHITESPACE:
2977 case OP_WHITESPACE:
2978 case OP_NOT_WORDCHAR:
2979 case OP_WORDCHAR:
2980 case OP_ANY:
2981 case OP_ALLANY:
2982 case OP_ANYNL:
2983 case OP_NOT_HSPACE:
2984 case OP_HSPACE:
2985 case OP_NOT_VSPACE:
2986 case OP_VSPACE:
2987 case OP_EXTUNI:
2988 case OP_EODN:
2989 case OP_EOD:
2990 case OP_DOLL:
2991 case OP_DOLLM:
2992 return code;
2993
2994 case OP_CHAR:
2995 case OP_NOT:
2996 GETCHARINCTEST(chr, code);
2997 list[2] = chr;
2998 list[3] = NOTACHAR;
2999 return code;
3000
3001 case OP_CHARI:
3002 case OP_NOTI:
3003 list[0] = (c == OP_CHARI) ? OP_CHAR : OP_NOT;
3004 GETCHARINCTEST(chr, code);
3005 list[2] = chr;
3006
3007 #ifdef SUPPORT_UCP
3008 if (chr < 128 || (chr < 256 && !utf))
3009 list[3] = fcc[chr];
3010 else
3011 list[3] = UCD_OTHERCASE(chr);
3012 #elif defined SUPPORT_UTF || !defined COMPILE_PCRE8
3013 list[3] = (chr < 256) ? fcc[chr] : chr;
3014 #else
3015 list[3] = fcc[chr];
3016 #endif
3017
3018 /* The othercase might be the same value. */
3019
3020 if (chr == list[3])
3021 list[3] = NOTACHAR;
3022 else
3023 list[4] = NOTACHAR;
3024 return code;
3025
3026 #ifdef SUPPORT_UCP
3027 case OP_PROP:
3028 case OP_NOTPROP:
3029 if (code[0] != PT_CLIST)
3030 {
3031 list[2] = code[0];
3032 list[3] = code[1];
3033 return code + 2;
3034 }
3035
3036 /* Convert only if we have enough space. */
3037
3038 clist_src = PRIV(ucd_caseless_sets) + code[1];
3039 clist_dest = list + 2;
3040 code += 2;
3041
3042 do {
3043 if (clist_dest >= list + 8)
3044 {
3045 /* Early return if there is not enough space. This should never
3046 happen, since all clists are shorter than 5 character now. */
3047 list[2] = code[0];
3048 list[3] = code[1];
3049 return code;
3050 }
3051 *clist_dest++ = *clist_src;
3052 }
3053 while(*clist_src++ != NOTACHAR);
3054
3055 /* All characters are stored. The terminating NOTACHAR
3056 is copied form the clist itself. */
3057
3058 list[0] = (c == OP_PROP) ? OP_CHAR : OP_NOT;
3059 return code;
3060 #endif
3061
3062 case OP_NCLASS:
3063 case OP_CLASS:
3064 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
3065 case OP_XCLASS:
3066 if (c == OP_XCLASS)
3067 end = code + GET(code, 0) - 1;
3068 else
3069 #endif
3070 end = code + 32 / sizeof(pcre_uchar);
3071
3072 switch(*end)
3073 {
3074 case OP_CRSTAR:
3075 case OP_CRMINSTAR:
3076 case OP_CRQUERY:
3077 case OP_CRMINQUERY:
3078 case OP_CRPOSSTAR:
3079 case OP_CRPOSQUERY:
3080 list[1] = TRUE;
3081 end++;
3082 break;
3083
3084 case OP_CRPLUS:
3085 case OP_CRMINPLUS:
3086 case OP_CRPOSPLUS:
3087 end++;
3088 break;
3089
3090 case OP_CRRANGE:
3091 case OP_CRMINRANGE:
3092 case OP_CRPOSRANGE:
3093 list[1] = (GET2(end, 1) == 0);
3094 end += 1 + 2 * IMM2_SIZE;
3095 break;
3096 }
3097 list[2] = (pcre_uint32)(end - code);
3098 return end;
3099 }
3100 return NULL; /* Opcode not accepted */
3101 }
3102
3103
3104
3105 /*************************************************
3106 * Scan further character sets for match *
3107 *************************************************/
3108
3109 /* Checks whether the base and the current opcode have a common character, in
3110 which case the base cannot be possessified.
3111
3112 Arguments:
3113 code points to the byte code
3114 utf TRUE in UTF-8 / UTF-16 / UTF-32 mode
3115 cd static compile data
3116 base_list the data list of the base opcode
3117
3118 Returns: TRUE if the auto-possessification is possible
3119 */
3120
3121 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)3122 compare_opcodes(const pcre_uchar *code, BOOL utf, const compile_data *cd,
3123 const pcre_uint32 *base_list, const pcre_uchar *base_end, int *rec_limit)
3124 {
3125 pcre_uchar c;
3126 pcre_uint32 list[8];
3127 const pcre_uint32 *chr_ptr;
3128 const pcre_uint32 *ochr_ptr;
3129 const pcre_uint32 *list_ptr;
3130 const pcre_uchar *next_code;
3131 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
3132 const pcre_uchar *xclass_flags;
3133 #endif
3134 const pcre_uint8 *class_bitset;
3135 const pcre_uint8 *set1, *set2, *set_end;
3136 pcre_uint32 chr;
3137 BOOL accepted, invert_bits;
3138 BOOL entered_a_group = FALSE;
3139
3140 if (*rec_limit == 0) return FALSE;
3141 --(*rec_limit);
3142
3143 /* Note: the base_list[1] contains whether the current opcode has greedy
3144 (represented by a non-zero value) quantifier. This is a different from
3145 other character type lists, which stores here that the character iterator
3146 matches to an empty string (also represented by a non-zero value). */
3147
3148 for(;;)
3149 {
3150 /* All operations move the code pointer forward.
3151 Therefore infinite recursions are not possible. */
3152
3153 c = *code;
3154
3155 /* Skip over callouts */
3156
3157 if (c == OP_CALLOUT)
3158 {
3159 code += PRIV(OP_lengths)[c];
3160 continue;
3161 }
3162
3163 if (c == OP_ALT)
3164 {
3165 do code += GET(code, 1); while (*code == OP_ALT);
3166 c = *code;
3167 }
3168
3169 switch(c)
3170 {
3171 case OP_END:
3172 case OP_KETRPOS:
3173 /* TRUE only in greedy case. The non-greedy case could be replaced by
3174 an OP_EXACT, but it is probably not worth it. (And note that OP_EXACT
3175 uses more memory, which we cannot get at this stage.) */
3176
3177 return base_list[1] != 0;
3178
3179 case OP_KET:
3180 /* If the bracket is capturing, and referenced by an OP_RECURSE, or
3181 it is an atomic sub-pattern (assert, once, etc.) the non-greedy case
3182 cannot be converted to a possessive form. */
3183
3184 if (base_list[1] == 0) return FALSE;
3185
3186 switch(*(code - GET(code, 1)))
3187 {
3188 case OP_ASSERT:
3189 case OP_ASSERT_NOT:
3190 case OP_ASSERTBACK:
3191 case OP_ASSERTBACK_NOT:
3192 case OP_ONCE:
3193 case OP_ONCE_NC:
3194 /* Atomic sub-patterns and assertions can always auto-possessify their
3195 last iterator. However, if the group was entered as a result of checking
3196 a previous iterator, this is not possible. */
3197
3198 return !entered_a_group;
3199 }
3200
3201 code += PRIV(OP_lengths)[c];
3202 continue;
3203
3204 case OP_ONCE:
3205 case OP_ONCE_NC:
3206 case OP_BRA:
3207 case OP_CBRA:
3208 next_code = code + GET(code, 1);
3209 code += PRIV(OP_lengths)[c];
3210
3211 while (*next_code == OP_ALT)
3212 {
3213 if (!compare_opcodes(code, utf, cd, base_list, base_end, rec_limit))
3214 return FALSE;
3215 code = next_code + 1 + LINK_SIZE;
3216 next_code += GET(next_code, 1);
3217 }
3218
3219 entered_a_group = TRUE;
3220 continue;
3221
3222 case OP_BRAZERO:
3223 case OP_BRAMINZERO:
3224
3225 next_code = code + 1;
3226 if (*next_code != OP_BRA && *next_code != OP_CBRA
3227 && *next_code != OP_ONCE && *next_code != OP_ONCE_NC) return FALSE;
3228
3229 do next_code += GET(next_code, 1); while (*next_code == OP_ALT);
3230
3231 /* The bracket content will be checked by the
3232 OP_BRA/OP_CBRA case above. */
3233 next_code += 1 + LINK_SIZE;
3234 if (!compare_opcodes(next_code, utf, cd, base_list, base_end, rec_limit))
3235 return FALSE;
3236
3237 code += PRIV(OP_lengths)[c];
3238 continue;
3239
3240 default:
3241 break;
3242 }
3243
3244 /* Check for a supported opcode, and load its properties. */
3245
3246 code = get_chr_property_list(code, utf, cd->fcc, list);
3247 if (code == NULL) return FALSE; /* Unsupported */
3248
3249 /* If either opcode is a small character list, set pointers for comparing
3250 characters from that list with another list, or with a property. */
3251
3252 if (base_list[0] == OP_CHAR)
3253 {
3254 chr_ptr = base_list + 2;
3255 list_ptr = list;
3256 }
3257 else if (list[0] == OP_CHAR)
3258 {
3259 chr_ptr = list + 2;
3260 list_ptr = base_list;
3261 }
3262
3263 /* Character bitsets can also be compared to certain opcodes. */
3264
3265 else if (base_list[0] == OP_CLASS || list[0] == OP_CLASS
3266 #ifdef COMPILE_PCRE8
3267 /* In 8 bit, non-UTF mode, OP_CLASS and OP_NCLASS are the same. */
3268 || (!utf && (base_list[0] == OP_NCLASS || list[0] == OP_NCLASS))
3269 #endif
3270 )
3271 {
3272 #ifdef COMPILE_PCRE8
3273 if (base_list[0] == OP_CLASS || (!utf && base_list[0] == OP_NCLASS))
3274 #else
3275 if (base_list[0] == OP_CLASS)
3276 #endif
3277 {
3278 set1 = (pcre_uint8 *)(base_end - base_list[2]);
3279 list_ptr = list;
3280 }
3281 else
3282 {
3283 set1 = (pcre_uint8 *)(code - list[2]);
3284 list_ptr = base_list;
3285 }
3286
3287 invert_bits = FALSE;
3288 switch(list_ptr[0])
3289 {
3290 case OP_CLASS:
3291 case OP_NCLASS:
3292 set2 = (pcre_uint8 *)
3293 ((list_ptr == list ? code : base_end) - list_ptr[2]);
3294 break;
3295
3296 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
3297 case OP_XCLASS:
3298 xclass_flags = (list_ptr == list ? code : base_end) - list_ptr[2] + LINK_SIZE;
3299 if ((*xclass_flags & XCL_HASPROP) != 0) return FALSE;
3300 if ((*xclass_flags & XCL_MAP) == 0)
3301 {
3302 /* No bits are set for characters < 256. */
3303 if (list[1] == 0) return (*xclass_flags & XCL_NOT) == 0;
3304 /* Might be an empty repeat. */
3305 continue;
3306 }
3307 set2 = (pcre_uint8 *)(xclass_flags + 1);
3308 break;
3309 #endif
3310
3311 case OP_NOT_DIGIT:
3312 invert_bits = TRUE;
3313 /* Fall through */
3314 case OP_DIGIT:
3315 set2 = (pcre_uint8 *)(cd->cbits + cbit_digit);
3316 break;
3317
3318 case OP_NOT_WHITESPACE:
3319 invert_bits = TRUE;
3320 /* Fall through */
3321 case OP_WHITESPACE:
3322 set2 = (pcre_uint8 *)(cd->cbits + cbit_space);
3323 break;
3324
3325 case OP_NOT_WORDCHAR:
3326 invert_bits = TRUE;
3327 /* Fall through */
3328 case OP_WORDCHAR:
3329 set2 = (pcre_uint8 *)(cd->cbits + cbit_word);
3330 break;
3331
3332 default:
3333 return FALSE;
3334 }
3335
3336 /* Because the sets are unaligned, we need
3337 to perform byte comparison here. */
3338 set_end = set1 + 32;
3339 if (invert_bits)
3340 {
3341 do
3342 {
3343 if ((*set1++ & ~(*set2++)) != 0) return FALSE;
3344 }
3345 while (set1 < set_end);
3346 }
3347 else
3348 {
3349 do
3350 {
3351 if ((*set1++ & *set2++) != 0) return FALSE;
3352 }
3353 while (set1 < set_end);
3354 }
3355
3356 if (list[1] == 0) return TRUE;
3357 /* Might be an empty repeat. */
3358 continue;
3359 }
3360
3361 /* Some property combinations also acceptable. Unicode property opcodes are
3362 processed specially; the rest can be handled with a lookup table. */
3363
3364 else
3365 {
3366 pcre_uint32 leftop, rightop;
3367
3368 leftop = base_list[0];
3369 rightop = list[0];
3370
3371 #ifdef SUPPORT_UCP
3372 accepted = FALSE; /* Always set in non-unicode case. */
3373 if (leftop == OP_PROP || leftop == OP_NOTPROP)
3374 {
3375 if (rightop == OP_EOD)
3376 accepted = TRUE;
3377 else if (rightop == OP_PROP || rightop == OP_NOTPROP)
3378 {
3379 int n;
3380 const pcre_uint8 *p;
3381 BOOL same = leftop == rightop;
3382 BOOL lisprop = leftop == OP_PROP;
3383 BOOL risprop = rightop == OP_PROP;
3384 BOOL bothprop = lisprop && risprop;
3385
3386 /* There's a table that specifies how each combination is to be
3387 processed:
3388 0 Always return FALSE (never auto-possessify)
3389 1 Character groups are distinct (possessify if both are OP_PROP)
3390 2 Check character categories in the same group (general or particular)
3391 3 Return TRUE if the two opcodes are not the same
3392 ... see comments below
3393 */
3394
3395 n = propposstab[base_list[2]][list[2]];
3396 switch(n)
3397 {
3398 case 0: break;
3399 case 1: accepted = bothprop; break;
3400 case 2: accepted = (base_list[3] == list[3]) != same; break;
3401 case 3: accepted = !same; break;
3402
3403 case 4: /* Left general category, right particular category */
3404 accepted = risprop && catposstab[base_list[3]][list[3]] == same;
3405 break;
3406
3407 case 5: /* Right general category, left particular category */
3408 accepted = lisprop && catposstab[list[3]][base_list[3]] == same;
3409 break;
3410
3411 /* This code is logically tricky. Think hard before fiddling with it.
3412 The posspropstab table has four entries per row. Each row relates to
3413 one of PCRE's special properties such as ALNUM or SPACE or WORD.
3414 Only WORD actually needs all four entries, but using repeats for the
3415 others means they can all use the same code below.
3416
3417 The first two entries in each row are Unicode general categories, and
3418 apply always, because all the characters they include are part of the
3419 PCRE character set. The third and fourth entries are a general and a
3420 particular category, respectively, that include one or more relevant
3421 characters. One or the other is used, depending on whether the check
3422 is for a general or a particular category. However, in both cases the
3423 category contains more characters than the specials that are defined
3424 for the property being tested against. Therefore, it cannot be used
3425 in a NOTPROP case.
3426
3427 Example: the row for WORD contains ucp_L, ucp_N, ucp_P, ucp_Po.
3428 Underscore is covered by ucp_P or ucp_Po. */
3429
3430 case 6: /* Left alphanum vs right general category */
3431 case 7: /* Left space vs right general category */
3432 case 8: /* Left word vs right general category */
3433 p = posspropstab[n-6];
3434 accepted = risprop && lisprop ==
3435 (list[3] != p[0] &&
3436 list[3] != p[1] &&
3437 (list[3] != p[2] || !lisprop));
3438 break;
3439
3440 case 9: /* Right alphanum vs left general category */
3441 case 10: /* Right space vs left general category */
3442 case 11: /* Right word vs left general category */
3443 p = posspropstab[n-9];
3444 accepted = lisprop && risprop ==
3445 (base_list[3] != p[0] &&
3446 base_list[3] != p[1] &&
3447 (base_list[3] != p[2] || !risprop));
3448 break;
3449
3450 case 12: /* Left alphanum vs right particular category */
3451 case 13: /* Left space vs right particular category */
3452 case 14: /* Left word vs right particular category */
3453 p = posspropstab[n-12];
3454 accepted = risprop && lisprop ==
3455 (catposstab[p[0]][list[3]] &&
3456 catposstab[p[1]][list[3]] &&
3457 (list[3] != p[3] || !lisprop));
3458 break;
3459
3460 case 15: /* Right alphanum vs left particular category */
3461 case 16: /* Right space vs left particular category */
3462 case 17: /* Right word vs left particular category */
3463 p = posspropstab[n-15];
3464 accepted = lisprop && risprop ==
3465 (catposstab[p[0]][base_list[3]] &&
3466 catposstab[p[1]][base_list[3]] &&
3467 (base_list[3] != p[3] || !risprop));
3468 break;
3469 }
3470 }
3471 }
3472
3473 else
3474 #endif /* SUPPORT_UCP */
3475
3476 accepted = leftop >= FIRST_AUTOTAB_OP && leftop <= LAST_AUTOTAB_LEFT_OP &&
3477 rightop >= FIRST_AUTOTAB_OP && rightop <= LAST_AUTOTAB_RIGHT_OP &&
3478 autoposstab[leftop - FIRST_AUTOTAB_OP][rightop - FIRST_AUTOTAB_OP];
3479
3480 if (!accepted) return FALSE;
3481
3482 if (list[1] == 0) return TRUE;
3483 /* Might be an empty repeat. */
3484 continue;
3485 }
3486
3487 /* Control reaches here only if one of the items is a small character list.
3488 All characters are checked against the other side. */
3489
3490 do
3491 {
3492 chr = *chr_ptr;
3493
3494 switch(list_ptr[0])
3495 {
3496 case OP_CHAR:
3497 ochr_ptr = list_ptr + 2;
3498 do
3499 {
3500 if (chr == *ochr_ptr) return FALSE;
3501 ochr_ptr++;
3502 }
3503 while(*ochr_ptr != NOTACHAR);
3504 break;
3505
3506 case OP_NOT:
3507 ochr_ptr = list_ptr + 2;
3508 do
3509 {
3510 if (chr == *ochr_ptr)
3511 break;
3512 ochr_ptr++;
3513 }
3514 while(*ochr_ptr != NOTACHAR);
3515 if (*ochr_ptr == NOTACHAR) return FALSE; /* Not found */
3516 break;
3517
3518 /* Note that OP_DIGIT etc. are generated only when PCRE_UCP is *not*
3519 set. When it is set, \d etc. are converted into OP_(NOT_)PROP codes. */
3520
3521 case OP_DIGIT:
3522 if (chr < 256 && (cd->ctypes[chr] & ctype_digit) != 0) return FALSE;
3523 break;
3524
3525 case OP_NOT_DIGIT:
3526 if (chr > 255 || (cd->ctypes[chr] & ctype_digit) == 0) return FALSE;
3527 break;
3528
3529 case OP_WHITESPACE:
3530 if (chr < 256 && (cd->ctypes[chr] & ctype_space) != 0) return FALSE;
3531 break;
3532
3533 case OP_NOT_WHITESPACE:
3534 if (chr > 255 || (cd->ctypes[chr] & ctype_space) == 0) return FALSE;
3535 break;
3536
3537 case OP_WORDCHAR:
3538 if (chr < 255 && (cd->ctypes[chr] & ctype_word) != 0) return FALSE;
3539 break;
3540
3541 case OP_NOT_WORDCHAR:
3542 if (chr > 255 || (cd->ctypes[chr] & ctype_word) == 0) return FALSE;
3543 break;
3544
3545 case OP_HSPACE:
3546 switch(chr)
3547 {
3548 HSPACE_CASES: return FALSE;
3549 default: break;
3550 }
3551 break;
3552
3553 case OP_NOT_HSPACE:
3554 switch(chr)
3555 {
3556 HSPACE_CASES: break;
3557 default: return FALSE;
3558 }
3559 break;
3560
3561 case OP_ANYNL:
3562 case OP_VSPACE:
3563 switch(chr)
3564 {
3565 VSPACE_CASES: return FALSE;
3566 default: break;
3567 }
3568 break;
3569
3570 case OP_NOT_VSPACE:
3571 switch(chr)
3572 {
3573 VSPACE_CASES: break;
3574 default: return FALSE;
3575 }
3576 break;
3577
3578 case OP_DOLL:
3579 case OP_EODN:
3580 switch (chr)
3581 {
3582 case CHAR_CR:
3583 case CHAR_LF:
3584 case CHAR_VT:
3585 case CHAR_FF:
3586 case CHAR_NEL:
3587 #ifndef EBCDIC
3588 case 0x2028:
3589 case 0x2029:
3590 #endif /* Not EBCDIC */
3591 return FALSE;
3592 }
3593 break;
3594
3595 case OP_EOD: /* Can always possessify before \z */
3596 break;
3597
3598 #ifdef SUPPORT_UCP
3599 case OP_PROP:
3600 case OP_NOTPROP:
3601 if (!check_char_prop(chr, list_ptr[2], list_ptr[3],
3602 list_ptr[0] == OP_NOTPROP))
3603 return FALSE;
3604 break;
3605 #endif
3606
3607 case OP_NCLASS:
3608 if (chr > 255) return FALSE;
3609 /* Fall through */
3610
3611 case OP_CLASS:
3612 if (chr > 255) break;
3613 class_bitset = (pcre_uint8 *)
3614 ((list_ptr == list ? code : base_end) - list_ptr[2]);
3615 if ((class_bitset[chr >> 3] & (1U << (chr & 7))) != 0) return FALSE;
3616 break;
3617
3618 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
3619 case OP_XCLASS:
3620 if (PRIV(xclass)(chr, (list_ptr == list ? code : base_end) -
3621 list_ptr[2] + LINK_SIZE, utf)) return FALSE;
3622 break;
3623 #endif
3624
3625 default:
3626 return FALSE;
3627 }
3628
3629 chr_ptr++;
3630 }
3631 while(*chr_ptr != NOTACHAR);
3632
3633 /* At least one character must be matched from this opcode. */
3634
3635 if (list[1] == 0) return TRUE;
3636 }
3637
3638 /* Control never reaches here. There used to be a fail-save return FALSE; here,
3639 but some compilers complain about an unreachable statement. */
3640
3641 }
3642
3643
3644
3645 /*************************************************
3646 * Scan compiled regex for auto-possession *
3647 *************************************************/
3648
3649 /* Replaces single character iterations with their possessive alternatives
3650 if appropriate. This function modifies the compiled opcode!
3651
3652 Arguments:
3653 code points to start of the byte code
3654 utf TRUE in UTF-8 / UTF-16 / UTF-32 mode
3655 cd static compile data
3656
3657 Returns: nothing
3658 */
3659
3660 static void
auto_possessify(pcre_uchar * code,BOOL utf,const compile_data * cd)3661 auto_possessify(pcre_uchar *code, BOOL utf, const compile_data *cd)
3662 {
3663 register pcre_uchar c;
3664 const pcre_uchar *end;
3665 pcre_uchar *repeat_opcode;
3666 pcre_uint32 list[8];
3667 int rec_limit;
3668
3669 for (;;)
3670 {
3671 c = *code;
3672
3673 /* When a pattern with bad UTF-8 encoding is compiled with NO_UTF_CHECK,
3674 it may compile without complaining, but may get into a loop here if the code
3675 pointer points to a bad value. This is, of course a documentated possibility,
3676 when NO_UTF_CHECK is set, so it isn't a bug, but we can detect this case and
3677 just give up on this optimization. */
3678
3679 if (c >= OP_TABLE_LENGTH) return;
3680
3681 if (c >= OP_STAR && c <= OP_TYPEPOSUPTO)
3682 {
3683 c -= get_repeat_base(c) - OP_STAR;
3684 end = (c <= OP_MINUPTO) ?
3685 get_chr_property_list(code, utf, cd->fcc, list) : NULL;
3686 list[1] = c == OP_STAR || c == OP_PLUS || c == OP_QUERY || c == OP_UPTO;
3687
3688 rec_limit = 1000;
3689 if (end != NULL && compare_opcodes(end, utf, cd, list, end, &rec_limit))
3690 {
3691 switch(c)
3692 {
3693 case OP_STAR:
3694 *code += OP_POSSTAR - OP_STAR;
3695 break;
3696
3697 case OP_MINSTAR:
3698 *code += OP_POSSTAR - OP_MINSTAR;
3699 break;
3700
3701 case OP_PLUS:
3702 *code += OP_POSPLUS - OP_PLUS;
3703 break;
3704
3705 case OP_MINPLUS:
3706 *code += OP_POSPLUS - OP_MINPLUS;
3707 break;
3708
3709 case OP_QUERY:
3710 *code += OP_POSQUERY - OP_QUERY;
3711 break;
3712
3713 case OP_MINQUERY:
3714 *code += OP_POSQUERY - OP_MINQUERY;
3715 break;
3716
3717 case OP_UPTO:
3718 *code += OP_POSUPTO - OP_UPTO;
3719 break;
3720
3721 case OP_MINUPTO:
3722 *code += OP_POSUPTO - OP_MINUPTO;
3723 break;
3724 }
3725 }
3726 c = *code;
3727 }
3728 else if (c == OP_CLASS || c == OP_NCLASS || c == OP_XCLASS)
3729 {
3730 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
3731 if (c == OP_XCLASS)
3732 repeat_opcode = code + GET(code, 1);
3733 else
3734 #endif
3735 repeat_opcode = code + 1 + (32 / sizeof(pcre_uchar));
3736
3737 c = *repeat_opcode;
3738 if (c >= OP_CRSTAR && c <= OP_CRMINRANGE)
3739 {
3740 /* end must not be NULL. */
3741 end = get_chr_property_list(code, utf, cd->fcc, list);
3742
3743 list[1] = (c & 1) == 0;
3744
3745 rec_limit = 1000;
3746 if (compare_opcodes(end, utf, cd, list, end, &rec_limit))
3747 {
3748 switch (c)
3749 {
3750 case OP_CRSTAR:
3751 case OP_CRMINSTAR:
3752 *repeat_opcode = OP_CRPOSSTAR;
3753 break;
3754
3755 case OP_CRPLUS:
3756 case OP_CRMINPLUS:
3757 *repeat_opcode = OP_CRPOSPLUS;
3758 break;
3759
3760 case OP_CRQUERY:
3761 case OP_CRMINQUERY:
3762 *repeat_opcode = OP_CRPOSQUERY;
3763 break;
3764
3765 case OP_CRRANGE:
3766 case OP_CRMINRANGE:
3767 *repeat_opcode = OP_CRPOSRANGE;
3768 break;
3769 }
3770 }
3771 }
3772 c = *code;
3773 }
3774
3775 switch(c)
3776 {
3777 case OP_END:
3778 return;
3779
3780 case OP_TYPESTAR:
3781 case OP_TYPEMINSTAR:
3782 case OP_TYPEPLUS:
3783 case OP_TYPEMINPLUS:
3784 case OP_TYPEQUERY:
3785 case OP_TYPEMINQUERY:
3786 case OP_TYPEPOSSTAR:
3787 case OP_TYPEPOSPLUS:
3788 case OP_TYPEPOSQUERY:
3789 if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
3790 break;
3791
3792 case OP_TYPEUPTO:
3793 case OP_TYPEMINUPTO:
3794 case OP_TYPEEXACT:
3795 case OP_TYPEPOSUPTO:
3796 if (code[1 + IMM2_SIZE] == OP_PROP || code[1 + IMM2_SIZE] == OP_NOTPROP)
3797 code += 2;
3798 break;
3799
3800 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
3801 case OP_XCLASS:
3802 code += GET(code, 1);
3803 break;
3804 #endif
3805
3806 case OP_MARK:
3807 case OP_PRUNE_ARG:
3808 case OP_SKIP_ARG:
3809 case OP_THEN_ARG:
3810 code += code[1];
3811 break;
3812 }
3813
3814 /* Add in the fixed length from the table */
3815
3816 code += PRIV(OP_lengths)[c];
3817
3818 /* In UTF-8 mode, opcodes that are followed by a character may be followed by
3819 a multi-byte character. The length in the table is a minimum, so we have to
3820 arrange to skip the extra bytes. */
3821
3822 #if defined SUPPORT_UTF && !defined COMPILE_PCRE32
3823 if (utf) switch(c)
3824 {
3825 case OP_CHAR:
3826 case OP_CHARI:
3827 case OP_NOT:
3828 case OP_NOTI:
3829 case OP_STAR:
3830 case OP_MINSTAR:
3831 case OP_PLUS:
3832 case OP_MINPLUS:
3833 case OP_QUERY:
3834 case OP_MINQUERY:
3835 case OP_UPTO:
3836 case OP_MINUPTO:
3837 case OP_EXACT:
3838 case OP_POSSTAR:
3839 case OP_POSPLUS:
3840 case OP_POSQUERY:
3841 case OP_POSUPTO:
3842 case OP_STARI:
3843 case OP_MINSTARI:
3844 case OP_PLUSI:
3845 case OP_MINPLUSI:
3846 case OP_QUERYI:
3847 case OP_MINQUERYI:
3848 case OP_UPTOI:
3849 case OP_MINUPTOI:
3850 case OP_EXACTI:
3851 case OP_POSSTARI:
3852 case OP_POSPLUSI:
3853 case OP_POSQUERYI:
3854 case OP_POSUPTOI:
3855 case OP_NOTSTAR:
3856 case OP_NOTMINSTAR:
3857 case OP_NOTPLUS:
3858 case OP_NOTMINPLUS:
3859 case OP_NOTQUERY:
3860 case OP_NOTMINQUERY:
3861 case OP_NOTUPTO:
3862 case OP_NOTMINUPTO:
3863 case OP_NOTEXACT:
3864 case OP_NOTPOSSTAR:
3865 case OP_NOTPOSPLUS:
3866 case OP_NOTPOSQUERY:
3867 case OP_NOTPOSUPTO:
3868 case OP_NOTSTARI:
3869 case OP_NOTMINSTARI:
3870 case OP_NOTPLUSI:
3871 case OP_NOTMINPLUSI:
3872 case OP_NOTQUERYI:
3873 case OP_NOTMINQUERYI:
3874 case OP_NOTUPTOI:
3875 case OP_NOTMINUPTOI:
3876 case OP_NOTEXACTI:
3877 case OP_NOTPOSSTARI:
3878 case OP_NOTPOSPLUSI:
3879 case OP_NOTPOSQUERYI:
3880 case OP_NOTPOSUPTOI:
3881 if (HAS_EXTRALEN(code[-1])) code += GET_EXTRALEN(code[-1]);
3882 break;
3883 }
3884 #else
3885 (void)(utf); /* Keep compiler happy by referencing function argument */
3886 #endif
3887 }
3888 }
3889
3890
3891
3892 /*************************************************
3893 * Check for POSIX class syntax *
3894 *************************************************/
3895
3896 /* This function is called when the sequence "[:" or "[." or "[=" is
3897 encountered in a character class. It checks whether this is followed by a
3898 sequence of characters terminated by a matching ":]" or ".]" or "=]". If we
3899 reach an unescaped ']' without the special preceding character, return FALSE.
3900
3901 Originally, this function only recognized a sequence of letters between the
3902 terminators, but it seems that Perl recognizes any sequence of characters,
3903 though of course unknown POSIX names are subsequently rejected. Perl gives an
3904 "Unknown POSIX class" error for [:f\oo:] for example, where previously PCRE
3905 didn't consider this to be a POSIX class. Likewise for [:1234:].
3906
3907 The problem in trying to be exactly like Perl is in the handling of escapes. We
3908 have to be sure that [abc[:x\]pqr] is *not* treated as containing a POSIX
3909 class, but [abc[:x\]pqr:]] is (so that an error can be generated). The code
3910 below handles the special cases \\ and \], but does not try to do any other
3911 escape processing. This makes it different from Perl for cases such as
3912 [:l\ower:] where Perl recognizes it as the POSIX class "lower" but PCRE does
3913 not recognize "l\ower". This is a lesser evil than not diagnosing bad classes
3914 when Perl does, I think.
3915
3916 A user pointed out that PCRE was rejecting [:a[:digit:]] whereas Perl was not.
3917 It seems that the appearance of a nested POSIX class supersedes an apparent
3918 external class. For example, [:a[:digit:]b:] matches "a", "b", ":", or
3919 a digit.
3920
3921 In Perl, unescaped square brackets may also appear as part of class names. For
3922 example, [:a[:abc]b:] gives unknown POSIX class "[:abc]b:]". However, for
3923 [:a[:abc]b][b:] it gives unknown POSIX class "[:abc]b][b:]", which does not
3924 seem right at all. PCRE does not allow closing square brackets in POSIX class
3925 names.
3926
3927 Arguments:
3928 ptr pointer to the initial [
3929 endptr where to return the end pointer
3930
3931 Returns: TRUE or FALSE
3932 */
3933
3934 static BOOL
check_posix_syntax(const pcre_uchar * ptr,const pcre_uchar ** endptr)3935 check_posix_syntax(const pcre_uchar *ptr, const pcre_uchar **endptr)
3936 {
3937 pcre_uchar terminator; /* Don't combine these lines; the Solaris cc */
3938 terminator = *(++ptr); /* compiler warns about "non-constant" initializer. */
3939 for (++ptr; *ptr != CHAR_NULL; ptr++)
3940 {
3941 if (*ptr == CHAR_BACKSLASH &&
3942 (ptr[1] == CHAR_RIGHT_SQUARE_BRACKET ||
3943 ptr[1] == CHAR_BACKSLASH))
3944 ptr++;
3945 else if ((*ptr == CHAR_LEFT_SQUARE_BRACKET && ptr[1] == terminator) ||
3946 *ptr == CHAR_RIGHT_SQUARE_BRACKET) return FALSE;
3947 else if (*ptr == terminator && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET)
3948 {
3949 *endptr = ptr;
3950 return TRUE;
3951 }
3952 }
3953 return FALSE;
3954 }
3955
3956
3957
3958
3959 /*************************************************
3960 * Check POSIX class name *
3961 *************************************************/
3962
3963 /* This function is called to check the name given in a POSIX-style class entry
3964 such as [:alnum:].
3965
3966 Arguments:
3967 ptr points to the first letter
3968 len the length of the name
3969
3970 Returns: a value representing the name, or -1 if unknown
3971 */
3972
3973 static int
check_posix_name(const pcre_uchar * ptr,int len)3974 check_posix_name(const pcre_uchar *ptr, int len)
3975 {
3976 const char *pn = posix_names;
3977 register int yield = 0;
3978 while (posix_name_lengths[yield] != 0)
3979 {
3980 if (len == posix_name_lengths[yield] &&
3981 STRNCMP_UC_C8(ptr, pn, (unsigned int)len) == 0) return yield;
3982 pn += posix_name_lengths[yield] + 1;
3983 yield++;
3984 }
3985 return -1;
3986 }
3987
3988
3989 /*************************************************
3990 * Adjust OP_RECURSE items in repeated group *
3991 *************************************************/
3992
3993 /* OP_RECURSE items contain an offset from the start of the regex to the group
3994 that is referenced. This means that groups can be replicated for fixed
3995 repetition simply by copying (because the recursion is allowed to refer to
3996 earlier groups that are outside the current group). However, when a group is
3997 optional (i.e. the minimum quantifier is zero), OP_BRAZERO or OP_SKIPZERO is
3998 inserted before it, after it has been compiled. This means that any OP_RECURSE
3999 items within it that refer to the group itself or any contained groups have to
4000 have their offsets adjusted. That one of the jobs of this function. Before it
4001 is called, the partially compiled regex must be temporarily terminated with
4002 OP_END.
4003
4004 This function has been extended to cope with forward references for recursions
4005 and subroutine calls. It must check the list of such references for the
4006 group we are dealing with. If it finds that one of the recursions in the
4007 current group is on this list, it does not adjust the value in the reference
4008 (which is a group number). After the group has been scanned, all the offsets in
4009 the forward reference list for the group are adjusted.
4010
4011 Arguments:
4012 group points to the start of the group
4013 adjust the amount by which the group is to be moved
4014 utf TRUE in UTF-8 / UTF-16 / UTF-32 mode
4015 cd contains pointers to tables etc.
4016 save_hwm_offset the hwm forward reference offset at the start of the group
4017
4018 Returns: nothing
4019 */
4020
4021 static void
adjust_recurse(pcre_uchar * group,int adjust,BOOL utf,compile_data * cd,size_t save_hwm_offset)4022 adjust_recurse(pcre_uchar *group, int adjust, BOOL utf, compile_data *cd,
4023 size_t save_hwm_offset)
4024 {
4025 int offset;
4026 pcre_uchar *hc;
4027 pcre_uchar *ptr = group;
4028
4029 while ((ptr = (pcre_uchar *)find_recurse(ptr, utf)) != NULL)
4030 {
4031 for (hc = (pcre_uchar *)cd->start_workspace + save_hwm_offset; hc < cd->hwm;
4032 hc += LINK_SIZE)
4033 {
4034 offset = (int)GET(hc, 0);
4035 if (cd->start_code + offset == ptr + 1) break;
4036 }
4037
4038 /* If we have not found this recursion on the forward reference list, adjust
4039 the recursion's offset if it's after the start of this group. */
4040
4041 if (hc >= cd->hwm)
4042 {
4043 offset = (int)GET(ptr, 1);
4044 if (cd->start_code + offset >= group) PUT(ptr, 1, offset + adjust);
4045 }
4046
4047 ptr += 1 + LINK_SIZE;
4048 }
4049
4050 /* Now adjust all forward reference offsets for the group. */
4051
4052 for (hc = (pcre_uchar *)cd->start_workspace + save_hwm_offset; hc < cd->hwm;
4053 hc += LINK_SIZE)
4054 {
4055 offset = (int)GET(hc, 0);
4056 PUT(hc, 0, offset + adjust);
4057 }
4058 }
4059
4060
4061
4062 /*************************************************
4063 * Insert an automatic callout point *
4064 *************************************************/
4065
4066 /* This function is called when the PCRE_AUTO_CALLOUT option is set, to insert
4067 callout points before each pattern item.
4068
4069 Arguments:
4070 code current code pointer
4071 ptr current pattern pointer
4072 cd pointers to tables etc
4073
4074 Returns: new code pointer
4075 */
4076
4077 static pcre_uchar *
auto_callout(pcre_uchar * code,const pcre_uchar * ptr,compile_data * cd)4078 auto_callout(pcre_uchar *code, const pcre_uchar *ptr, compile_data *cd)
4079 {
4080 *code++ = OP_CALLOUT;
4081 *code++ = 255;
4082 PUT(code, 0, (int)(ptr - cd->start_pattern)); /* Pattern offset */
4083 PUT(code, LINK_SIZE, 0); /* Default length */
4084 return code + 2 * LINK_SIZE;
4085 }
4086
4087
4088
4089 /*************************************************
4090 * Complete a callout item *
4091 *************************************************/
4092
4093 /* A callout item contains the length of the next item in the pattern, which
4094 we can't fill in till after we have reached the relevant point. This is used
4095 for both automatic and manual callouts.
4096
4097 Arguments:
4098 previous_callout points to previous callout item
4099 ptr current pattern pointer
4100 cd pointers to tables etc
4101
4102 Returns: nothing
4103 */
4104
4105 static void
complete_callout(pcre_uchar * previous_callout,const pcre_uchar * ptr,compile_data * cd)4106 complete_callout(pcre_uchar *previous_callout, const pcre_uchar *ptr, compile_data *cd)
4107 {
4108 int length = (int)(ptr - cd->start_pattern - GET(previous_callout, 2));
4109 PUT(previous_callout, 2 + LINK_SIZE, length);
4110 }
4111
4112
4113
4114 #ifdef SUPPORT_UCP
4115 /*************************************************
4116 * Get othercase range *
4117 *************************************************/
4118
4119 /* This function is passed the start and end of a class range, in UTF-8 mode
4120 with UCP support. It searches up the characters, looking for ranges of
4121 characters in the "other" case. Each call returns the next one, updating the
4122 start address. A character with multiple other cases is returned on its own
4123 with a special return value.
4124
4125 Arguments:
4126 cptr points to starting character value; updated
4127 d end value
4128 ocptr where to put start of othercase range
4129 odptr where to put end of othercase range
4130
4131 Yield: -1 when no more
4132 0 when a range is returned
4133 >0 the CASESET offset for char with multiple other cases
4134 in this case, ocptr contains the original
4135 */
4136
4137 static int
get_othercase_range(pcre_uint32 * cptr,pcre_uint32 d,pcre_uint32 * ocptr,pcre_uint32 * odptr)4138 get_othercase_range(pcre_uint32 *cptr, pcre_uint32 d, pcre_uint32 *ocptr,
4139 pcre_uint32 *odptr)
4140 {
4141 pcre_uint32 c, othercase, next;
4142 unsigned int co;
4143
4144 /* Find the first character that has an other case. If it has multiple other
4145 cases, return its case offset value. */
4146
4147 for (c = *cptr; c <= d; c++)
4148 {
4149 if ((co = UCD_CASESET(c)) != 0)
4150 {
4151 *ocptr = c++; /* Character that has the set */
4152 *cptr = c; /* Rest of input range */
4153 return (int)co;
4154 }
4155 if ((othercase = UCD_OTHERCASE(c)) != c) break;
4156 }
4157
4158 if (c > d) return -1; /* Reached end of range */
4159
4160 /* Found a character that has a single other case. Search for the end of the
4161 range, which is either the end of the input range, or a character that has zero
4162 or more than one other cases. */
4163
4164 *ocptr = othercase;
4165 next = othercase + 1;
4166
4167 for (++c; c <= d; c++)
4168 {
4169 if ((co = UCD_CASESET(c)) != 0 || UCD_OTHERCASE(c) != next) break;
4170 next++;
4171 }
4172
4173 *odptr = next - 1; /* End of othercase range */
4174 *cptr = c; /* Rest of input range */
4175 return 0;
4176 }
4177 #endif /* SUPPORT_UCP */
4178
4179
4180
4181 /*************************************************
4182 * Add a character or range to a class *
4183 *************************************************/
4184
4185 /* This function packages up the logic of adding a character or range of
4186 characters to a class. The character values in the arguments will be within the
4187 valid values for the current mode (8-bit, 16-bit, UTF, etc). This function is
4188 mutually recursive with the function immediately below.
4189
4190 Arguments:
4191 classbits the bit map for characters < 256
4192 uchardptr points to the pointer for extra data
4193 options the options word
4194 cd contains pointers to tables etc.
4195 start start of range character
4196 end end of range character
4197
4198 Returns: the number of < 256 characters added
4199 the pointer to extra data is updated
4200 */
4201
4202 static int
add_to_class(pcre_uint8 * classbits,pcre_uchar ** uchardptr,int options,compile_data * cd,pcre_uint32 start,pcre_uint32 end)4203 add_to_class(pcre_uint8 *classbits, pcre_uchar **uchardptr, int options,
4204 compile_data *cd, pcre_uint32 start, pcre_uint32 end)
4205 {
4206 pcre_uint32 c;
4207 pcre_uint32 classbits_end = (end <= 0xff ? end : 0xff);
4208 int n8 = 0;
4209
4210 /* If caseless matching is required, scan the range and process alternate
4211 cases. In Unicode, there are 8-bit characters that have alternate cases that
4212 are greater than 255 and vice-versa. Sometimes we can just extend the original
4213 range. */
4214
4215 if ((options & PCRE_CASELESS) != 0)
4216 {
4217 #ifdef SUPPORT_UCP
4218 if ((options & PCRE_UTF8) != 0)
4219 {
4220 int rc;
4221 pcre_uint32 oc, od;
4222
4223 options &= ~PCRE_CASELESS; /* Remove for recursive calls */
4224 c = start;
4225
4226 while ((rc = get_othercase_range(&c, end, &oc, &od)) >= 0)
4227 {
4228 /* Handle a single character that has more than one other case. */
4229
4230 if (rc > 0) n8 += add_list_to_class(classbits, uchardptr, options, cd,
4231 PRIV(ucd_caseless_sets) + rc, oc);
4232
4233 /* Do nothing if the other case range is within the original range. */
4234
4235 else if (oc >= start && od <= end) continue;
4236
4237 /* Extend the original range if there is overlap, noting that if oc < c, we
4238 can't have od > end because a subrange is always shorter than the basic
4239 range. Otherwise, use a recursive call to add the additional range. */
4240
4241 else if (oc < start && od >= start - 1) start = oc; /* Extend downwards */
4242 else if (od > end && oc <= end + 1)
4243 {
4244 end = od; /* Extend upwards */
4245 if (end > classbits_end) classbits_end = (end <= 0xff ? end : 0xff);
4246 }
4247 else n8 += add_to_class(classbits, uchardptr, options, cd, oc, od);
4248 }
4249 }
4250 else
4251 #endif /* SUPPORT_UCP */
4252
4253 /* Not UTF-mode, or no UCP */
4254
4255 for (c = start; c <= classbits_end; c++)
4256 {
4257 SETBIT(classbits, cd->fcc[c]);
4258 n8++;
4259 }
4260 }
4261
4262 /* Now handle the original range. Adjust the final value according to the bit
4263 length - this means that the same lists of (e.g.) horizontal spaces can be used
4264 in all cases. */
4265
4266 #if defined COMPILE_PCRE8
4267 #ifdef SUPPORT_UTF
4268 if ((options & PCRE_UTF8) == 0)
4269 #endif
4270 if (end > 0xff) end = 0xff;
4271
4272 #elif defined COMPILE_PCRE16
4273 #ifdef SUPPORT_UTF
4274 if ((options & PCRE_UTF16) == 0)
4275 #endif
4276 if (end > 0xffff) end = 0xffff;
4277
4278 #endif /* COMPILE_PCRE[8|16] */
4279
4280 /* Use the bitmap for characters < 256. Otherwise use extra data.*/
4281
4282 for (c = start; c <= classbits_end; c++)
4283 {
4284 /* Regardless of start, c will always be <= 255. */
4285 SETBIT(classbits, c);
4286 n8++;
4287 }
4288
4289 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
4290 if (start <= 0xff) start = 0xff + 1;
4291
4292 if (end >= start)
4293 {
4294 pcre_uchar *uchardata = *uchardptr;
4295 #ifdef SUPPORT_UTF
4296 if ((options & PCRE_UTF8) != 0) /* All UTFs use the same flag bit */
4297 {
4298 if (start < end)
4299 {
4300 *uchardata++ = XCL_RANGE;
4301 uchardata += PRIV(ord2utf)(start, uchardata);
4302 uchardata += PRIV(ord2utf)(end, uchardata);
4303 }
4304 else if (start == end)
4305 {
4306 *uchardata++ = XCL_SINGLE;
4307 uchardata += PRIV(ord2utf)(start, uchardata);
4308 }
4309 }
4310 else
4311 #endif /* SUPPORT_UTF */
4312
4313 /* Without UTF support, character values are constrained by the bit length,
4314 and can only be > 256 for 16-bit and 32-bit libraries. */
4315
4316 #ifdef COMPILE_PCRE8
4317 {}
4318 #else
4319 if (start < end)
4320 {
4321 *uchardata++ = XCL_RANGE;
4322 *uchardata++ = start;
4323 *uchardata++ = end;
4324 }
4325 else if (start == end)
4326 {
4327 *uchardata++ = XCL_SINGLE;
4328 *uchardata++ = start;
4329 }
4330 #endif
4331
4332 *uchardptr = uchardata; /* Updata extra data pointer */
4333 }
4334 #endif /* SUPPORT_UTF || !COMPILE_PCRE8 */
4335
4336 return n8; /* Number of 8-bit characters */
4337 }
4338
4339
4340
4341
4342 /*************************************************
4343 * Add a list of characters to a class *
4344 *************************************************/
4345
4346 /* This function is used for adding a list of case-equivalent characters to a
4347 class, and also for adding a list of horizontal or vertical whitespace. If the
4348 list is in order (which it should be), ranges of characters are detected and
4349 handled appropriately. This function is mutually recursive with the function
4350 above.
4351
4352 Arguments:
4353 classbits the bit map for characters < 256
4354 uchardptr points to the pointer for extra data
4355 options the options word
4356 cd contains pointers to tables etc.
4357 p points to row of 32-bit values, terminated by NOTACHAR
4358 except character to omit; this is used when adding lists of
4359 case-equivalent characters to avoid including the one we
4360 already know about
4361
4362 Returns: the number of < 256 characters added
4363 the pointer to extra data is updated
4364 */
4365
4366 static int
add_list_to_class(pcre_uint8 * classbits,pcre_uchar ** uchardptr,int options,compile_data * cd,const pcre_uint32 * p,unsigned int except)4367 add_list_to_class(pcre_uint8 *classbits, pcre_uchar **uchardptr, int options,
4368 compile_data *cd, const pcre_uint32 *p, unsigned int except)
4369 {
4370 int n8 = 0;
4371 while (p[0] < NOTACHAR)
4372 {
4373 int n = 0;
4374 if (p[0] != except)
4375 {
4376 while(p[n+1] == p[0] + n + 1) n++;
4377 n8 += add_to_class(classbits, uchardptr, options, cd, p[0], p[n]);
4378 }
4379 p += n + 1;
4380 }
4381 return n8;
4382 }
4383
4384
4385
4386 /*************************************************
4387 * Add characters not in a list to a class *
4388 *************************************************/
4389
4390 /* This function is used for adding the complement of a list of horizontal or
4391 vertical whitespace to a class. The list must be in order.
4392
4393 Arguments:
4394 classbits the bit map for characters < 256
4395 uchardptr points to the pointer for extra data
4396 options the options word
4397 cd contains pointers to tables etc.
4398 p points to row of 32-bit values, terminated by NOTACHAR
4399
4400 Returns: the number of < 256 characters added
4401 the pointer to extra data is updated
4402 */
4403
4404 static int
add_not_list_to_class(pcre_uint8 * classbits,pcre_uchar ** uchardptr,int options,compile_data * cd,const pcre_uint32 * p)4405 add_not_list_to_class(pcre_uint8 *classbits, pcre_uchar **uchardptr,
4406 int options, compile_data *cd, const pcre_uint32 *p)
4407 {
4408 BOOL utf = (options & PCRE_UTF8) != 0;
4409 int n8 = 0;
4410 if (p[0] > 0)
4411 n8 += add_to_class(classbits, uchardptr, options, cd, 0, p[0] - 1);
4412 while (p[0] < NOTACHAR)
4413 {
4414 while (p[1] == p[0] + 1) p++;
4415 n8 += add_to_class(classbits, uchardptr, options, cd, p[0] + 1,
4416 (p[1] == NOTACHAR) ? (utf ? 0x10ffffu : 0xffffffffu) : p[1] - 1);
4417 p++;
4418 }
4419 return n8;
4420 }
4421
4422
4423
4424 /*************************************************
4425 * Compile one branch *
4426 *************************************************/
4427
4428 /* Scan the pattern, compiling it into the a vector. If the options are
4429 changed during the branch, the pointer is used to change the external options
4430 bits. This function is used during the pre-compile phase when we are trying
4431 to find out the amount of memory needed, as well as during the real compile
4432 phase. The value of lengthptr distinguishes the two phases.
4433
4434 Arguments:
4435 optionsptr pointer to the option bits
4436 codeptr points to the pointer to the current code point
4437 ptrptr points to the current pattern pointer
4438 errorcodeptr points to error code variable
4439 firstcharptr place to put the first required character
4440 firstcharflagsptr place to put the first character flags, or a negative number
4441 reqcharptr place to put the last required character
4442 reqcharflagsptr place to put the last required character flags, or a negative number
4443 bcptr points to current branch chain
4444 cond_depth conditional nesting depth
4445 cd contains pointers to tables etc.
4446 lengthptr NULL during the real compile phase
4447 points to length accumulator during pre-compile phase
4448
4449 Returns: TRUE on success
4450 FALSE, with *errorcodeptr set non-zero on error
4451 */
4452
4453 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)4454 compile_branch(int *optionsptr, pcre_uchar **codeptr,
4455 const pcre_uchar **ptrptr, int *errorcodeptr,
4456 pcre_uint32 *firstcharptr, pcre_int32 *firstcharflagsptr,
4457 pcre_uint32 *reqcharptr, pcre_int32 *reqcharflagsptr,
4458 branch_chain *bcptr, int cond_depth,
4459 compile_data *cd, int *lengthptr)
4460 {
4461 int repeat_type, op_type;
4462 int repeat_min = 0, repeat_max = 0; /* To please picky compilers */
4463 int bravalue = 0;
4464 int greedy_default, greedy_non_default;
4465 pcre_uint32 firstchar, reqchar;
4466 pcre_int32 firstcharflags, reqcharflags;
4467 pcre_uint32 zeroreqchar, zerofirstchar;
4468 pcre_int32 zeroreqcharflags, zerofirstcharflags;
4469 pcre_int32 req_caseopt, reqvary, tempreqvary;
4470 int options = *optionsptr; /* May change dynamically */
4471 int after_manual_callout = 0;
4472 int length_prevgroup = 0;
4473 register pcre_uint32 c;
4474 int escape;
4475 register pcre_uchar *code = *codeptr;
4476 pcre_uchar *last_code = code;
4477 pcre_uchar *orig_code = code;
4478 pcre_uchar *tempcode;
4479 BOOL inescq = FALSE;
4480 BOOL groupsetfirstchar = FALSE;
4481 const pcre_uchar *ptr = *ptrptr;
4482 const pcre_uchar *tempptr;
4483 const pcre_uchar *nestptr = NULL;
4484 pcre_uchar *previous = NULL;
4485 pcre_uchar *previous_callout = NULL;
4486 size_t item_hwm_offset = 0;
4487 pcre_uint8 classbits[32];
4488
4489 /* We can fish out the UTF-8 setting once and for all into a BOOL, but we
4490 must not do this for other options (e.g. PCRE_EXTENDED) because they may change
4491 dynamically as we process the pattern. */
4492
4493 #ifdef SUPPORT_UTF
4494 /* PCRE_UTF[16|32] have the same value as PCRE_UTF8. */
4495 BOOL utf = (options & PCRE_UTF8) != 0;
4496 #ifndef COMPILE_PCRE32
4497 pcre_uchar utf_chars[6];
4498 #endif
4499 #else
4500 BOOL utf = FALSE;
4501 #endif
4502
4503 /* Helper variables for OP_XCLASS opcode (for characters > 255). We define
4504 class_uchardata always so that it can be passed to add_to_class() always,
4505 though it will not be used in non-UTF 8-bit cases. This avoids having to supply
4506 alternative calls for the different cases. */
4507
4508 pcre_uchar *class_uchardata;
4509 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
4510 BOOL xclass;
4511 pcre_uchar *class_uchardata_base;
4512 #endif
4513
4514 #ifdef PCRE_DEBUG
4515 if (lengthptr != NULL) DPRINTF((">> start branch\n"));
4516 #endif
4517
4518 /* Set up the default and non-default settings for greediness */
4519
4520 greedy_default = ((options & PCRE_UNGREEDY) != 0);
4521 greedy_non_default = greedy_default ^ 1;
4522
4523 /* Initialize no first byte, no required byte. REQ_UNSET means "no char
4524 matching encountered yet". It gets changed to REQ_NONE if we hit something that
4525 matches a non-fixed char first char; reqchar just remains unset if we never
4526 find one.
4527
4528 When we hit a repeat whose minimum is zero, we may have to adjust these values
4529 to take the zero repeat into account. This is implemented by setting them to
4530 zerofirstbyte and zeroreqchar when such a repeat is encountered. The individual
4531 item types that can be repeated set these backoff variables appropriately. */
4532
4533 firstchar = reqchar = zerofirstchar = zeroreqchar = 0;
4534 firstcharflags = reqcharflags = zerofirstcharflags = zeroreqcharflags = REQ_UNSET;
4535
4536 /* The variable req_caseopt contains either the REQ_CASELESS value
4537 or zero, according to the current setting of the caseless flag. The
4538 REQ_CASELESS leaves the lower 28 bit empty. It is added into the
4539 firstchar or reqchar variables to record the case status of the
4540 value. This is used only for ASCII characters. */
4541
4542 req_caseopt = ((options & PCRE_CASELESS) != 0)? REQ_CASELESS:0;
4543
4544 /* Switch on next character until the end of the branch */
4545
4546 for (;; ptr++)
4547 {
4548 BOOL negate_class;
4549 BOOL should_flip_negation;
4550 BOOL possessive_quantifier;
4551 BOOL is_quantifier;
4552 BOOL is_recurse;
4553 BOOL reset_bracount;
4554 int class_has_8bitchar;
4555 int class_one_char;
4556 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
4557 BOOL xclass_has_prop;
4558 #endif
4559 int newoptions;
4560 int recno;
4561 int refsign;
4562 int skipbytes;
4563 pcre_uint32 subreqchar, subfirstchar;
4564 pcre_int32 subreqcharflags, subfirstcharflags;
4565 int terminator;
4566 unsigned int mclength;
4567 unsigned int tempbracount;
4568 pcre_uint32 ec;
4569 pcre_uchar mcbuffer[8];
4570
4571 /* Come here to restart the loop without advancing the pointer. */
4572
4573 REDO_LOOP:
4574
4575 /* Get next character in the pattern */
4576
4577 c = *ptr;
4578
4579 /* If we are at the end of a nested substitution, revert to the outer level
4580 string. Nesting only happens one level deep. */
4581
4582 if (c == CHAR_NULL && nestptr != NULL)
4583 {
4584 ptr = nestptr;
4585 nestptr = NULL;
4586 c = *ptr;
4587 }
4588
4589 /* If we are in the pre-compile phase, accumulate the length used for the
4590 previous cycle of this loop. */
4591
4592 if (lengthptr != NULL)
4593 {
4594 #ifdef PCRE_DEBUG
4595 if (code > cd->hwm) cd->hwm = code; /* High water info */
4596 #endif
4597 if (code > cd->start_workspace + cd->workspace_size -
4598 WORK_SIZE_SAFETY_MARGIN) /* Check for overrun */
4599 {
4600 *errorcodeptr = (code >= cd->start_workspace + cd->workspace_size)?
4601 ERR52 : ERR87;
4602 goto FAILED;
4603 }
4604
4605 /* There is at least one situation where code goes backwards: this is the
4606 case of a zero quantifier after a class (e.g. [ab]{0}). At compile time,
4607 the class is simply eliminated. However, it is created first, so we have to
4608 allow memory for it. Therefore, don't ever reduce the length at this point.
4609 */
4610
4611 if (code < last_code) code = last_code;
4612
4613 /* Paranoid check for integer overflow */
4614
4615 if (OFLOW_MAX - *lengthptr < code - last_code)
4616 {
4617 *errorcodeptr = ERR20;
4618 goto FAILED;
4619 }
4620
4621 *lengthptr += (int)(code - last_code);
4622 DPRINTF(("length=%d added %d c=%c (0x%x)\n", *lengthptr,
4623 (int)(code - last_code), c, c));
4624
4625 /* If "previous" is set and it is not at the start of the work space, move
4626 it back to there, in order to avoid filling up the work space. Otherwise,
4627 if "previous" is NULL, reset the current code pointer to the start. */
4628
4629 if (previous != NULL)
4630 {
4631 if (previous > orig_code)
4632 {
4633 memmove(orig_code, previous, IN_UCHARS(code - previous));
4634 code -= previous - orig_code;
4635 previous = orig_code;
4636 }
4637 }
4638 else code = orig_code;
4639
4640 /* Remember where this code item starts so we can pick up the length
4641 next time round. */
4642
4643 last_code = code;
4644 }
4645
4646 /* In the real compile phase, just check the workspace used by the forward
4647 reference list. */
4648
4649 else if (cd->hwm > cd->start_workspace + cd->workspace_size)
4650 {
4651 *errorcodeptr = ERR52;
4652 goto FAILED;
4653 }
4654
4655 /* If in \Q...\E, check for the end; if not, we have a literal. Otherwise an
4656 isolated \E is ignored. */
4657
4658 if (c != CHAR_NULL)
4659 {
4660 if (c == CHAR_BACKSLASH && ptr[1] == CHAR_E)
4661 {
4662 inescq = FALSE;
4663 ptr++;
4664 continue;
4665 }
4666 else if (inescq)
4667 {
4668 if (previous_callout != NULL)
4669 {
4670 if (lengthptr == NULL) /* Don't attempt in pre-compile phase */
4671 complete_callout(previous_callout, ptr, cd);
4672 previous_callout = NULL;
4673 }
4674 if ((options & PCRE_AUTO_CALLOUT) != 0)
4675 {
4676 previous_callout = code;
4677 code = auto_callout(code, ptr, cd);
4678 }
4679 goto NORMAL_CHAR;
4680 }
4681
4682 /* Check for the start of a \Q...\E sequence. We must do this here rather
4683 than later in case it is immediately followed by \E, which turns it into a
4684 "do nothing" sequence. */
4685
4686 if (c == CHAR_BACKSLASH && ptr[1] == CHAR_Q)
4687 {
4688 inescq = TRUE;
4689 ptr++;
4690 continue;
4691 }
4692 }
4693
4694 /* In extended mode, skip white space and comments. */
4695
4696 if ((options & PCRE_EXTENDED) != 0)
4697 {
4698 const pcre_uchar *wscptr = ptr;
4699 while (MAX_255(c) && (cd->ctypes[c] & ctype_space) != 0) c = *(++ptr);
4700 if (c == CHAR_NUMBER_SIGN)
4701 {
4702 ptr++;
4703 while (*ptr != CHAR_NULL)
4704 {
4705 if (IS_NEWLINE(ptr)) /* For non-fixed-length newline cases, */
4706 { /* IS_NEWLINE sets cd->nllen. */
4707 ptr += cd->nllen;
4708 break;
4709 }
4710 ptr++;
4711 #ifdef SUPPORT_UTF
4712 if (utf) FORWARDCHAR(ptr);
4713 #endif
4714 }
4715 }
4716
4717 /* If we skipped any characters, restart the loop. Otherwise, we didn't see
4718 a comment. */
4719
4720 if (ptr > wscptr) goto REDO_LOOP;
4721 }
4722
4723 /* Skip over (?# comments. We need to do this here because we want to know if
4724 the next thing is a quantifier, and these comments may come between an item
4725 and its quantifier. */
4726
4727 if (c == CHAR_LEFT_PARENTHESIS && ptr[1] == CHAR_QUESTION_MARK &&
4728 ptr[2] == CHAR_NUMBER_SIGN)
4729 {
4730 ptr += 3;
4731 while (*ptr != CHAR_NULL && *ptr != CHAR_RIGHT_PARENTHESIS) ptr++;
4732 if (*ptr == CHAR_NULL)
4733 {
4734 *errorcodeptr = ERR18;
4735 goto FAILED;
4736 }
4737 continue;
4738 }
4739
4740 /* See if the next thing is a quantifier. */
4741
4742 is_quantifier =
4743 c == CHAR_ASTERISK || c == CHAR_PLUS || c == CHAR_QUESTION_MARK ||
4744 (c == CHAR_LEFT_CURLY_BRACKET && is_counted_repeat(ptr+1));
4745
4746 /* Fill in length of a previous callout, except when the next thing is a
4747 quantifier or when processing a property substitution string in UCP mode. */
4748
4749 if (!is_quantifier && previous_callout != NULL && nestptr == NULL &&
4750 after_manual_callout-- <= 0)
4751 {
4752 if (lengthptr == NULL) /* Don't attempt in pre-compile phase */
4753 complete_callout(previous_callout, ptr, cd);
4754 previous_callout = NULL;
4755 }
4756
4757 /* Create auto callout, except for quantifiers, or while processing property
4758 strings that are substituted for \w etc in UCP mode. */
4759
4760 if ((options & PCRE_AUTO_CALLOUT) != 0 && !is_quantifier && nestptr == NULL)
4761 {
4762 previous_callout = code;
4763 code = auto_callout(code, ptr, cd);
4764 }
4765
4766 /* Process the next pattern item. */
4767
4768 switch(c)
4769 {
4770 /* ===================================================================*/
4771 case CHAR_NULL: /* The branch terminates at string end */
4772 case CHAR_VERTICAL_LINE: /* or | or ) */
4773 case CHAR_RIGHT_PARENTHESIS:
4774 *firstcharptr = firstchar;
4775 *firstcharflagsptr = firstcharflags;
4776 *reqcharptr = reqchar;
4777 *reqcharflagsptr = reqcharflags;
4778 *codeptr = code;
4779 *ptrptr = ptr;
4780 if (lengthptr != NULL)
4781 {
4782 if (OFLOW_MAX - *lengthptr < code - last_code)
4783 {
4784 *errorcodeptr = ERR20;
4785 goto FAILED;
4786 }
4787 *lengthptr += (int)(code - last_code); /* To include callout length */
4788 DPRINTF((">> end branch\n"));
4789 }
4790 return TRUE;
4791
4792
4793 /* ===================================================================*/
4794 /* Handle single-character metacharacters. In multiline mode, ^ disables
4795 the setting of any following char as a first character. */
4796
4797 case CHAR_CIRCUMFLEX_ACCENT:
4798 previous = NULL;
4799 if ((options & PCRE_MULTILINE) != 0)
4800 {
4801 if (firstcharflags == REQ_UNSET)
4802 zerofirstcharflags = firstcharflags = REQ_NONE;
4803 *code++ = OP_CIRCM;
4804 }
4805 else *code++ = OP_CIRC;
4806 break;
4807
4808 case CHAR_DOLLAR_SIGN:
4809 previous = NULL;
4810 *code++ = ((options & PCRE_MULTILINE) != 0)? OP_DOLLM : OP_DOLL;
4811 break;
4812
4813 /* There can never be a first char if '.' is first, whatever happens about
4814 repeats. The value of reqchar doesn't change either. */
4815
4816 case CHAR_DOT:
4817 if (firstcharflags == REQ_UNSET) firstcharflags = REQ_NONE;
4818 zerofirstchar = firstchar;
4819 zerofirstcharflags = firstcharflags;
4820 zeroreqchar = reqchar;
4821 zeroreqcharflags = reqcharflags;
4822 previous = code;
4823 item_hwm_offset = cd->hwm - cd->start_workspace;
4824 *code++ = ((options & PCRE_DOTALL) != 0)? OP_ALLANY: OP_ANY;
4825 break;
4826
4827
4828 /* ===================================================================*/
4829 /* Character classes. If the included characters are all < 256, we build a
4830 32-byte bitmap of the permitted characters, except in the special case
4831 where there is only one such character. For negated classes, we build the
4832 map as usual, then invert it at the end. However, we use a different opcode
4833 so that data characters > 255 can be handled correctly.
4834
4835 If the class contains characters outside the 0-255 range, a different
4836 opcode is compiled. It may optionally have a bit map for characters < 256,
4837 but those above are are explicitly listed afterwards. A flag byte tells
4838 whether the bitmap is present, and whether this is a negated class or not.
4839
4840 In JavaScript compatibility mode, an isolated ']' causes an error. In
4841 default (Perl) mode, it is treated as a data character. */
4842
4843 case CHAR_RIGHT_SQUARE_BRACKET:
4844 if ((cd->external_options & PCRE_JAVASCRIPT_COMPAT) != 0)
4845 {
4846 *errorcodeptr = ERR64;
4847 goto FAILED;
4848 }
4849 goto NORMAL_CHAR;
4850
4851 /* In another (POSIX) regex library, the ugly syntax [[:<:]] and [[:>:]] is
4852 used for "start of word" and "end of word". As these are otherwise illegal
4853 sequences, we don't break anything by recognizing them. They are replaced
4854 by \b(?=\w) and \b(?<=\w) respectively. Sequences like [a[:<:]] are
4855 erroneous and are handled by the normal code below. */
4856
4857 case CHAR_LEFT_SQUARE_BRACKET:
4858 if (STRNCMP_UC_C8(ptr+1, STRING_WEIRD_STARTWORD, 6) == 0)
4859 {
4860 nestptr = ptr + 7;
4861 ptr = sub_start_of_word;
4862 goto REDO_LOOP;
4863 }
4864
4865 if (STRNCMP_UC_C8(ptr+1, STRING_WEIRD_ENDWORD, 6) == 0)
4866 {
4867 nestptr = ptr + 7;
4868 ptr = sub_end_of_word;
4869 goto REDO_LOOP;
4870 }
4871
4872 /* Handle a real character class. */
4873
4874 previous = code;
4875 item_hwm_offset = cd->hwm - cd->start_workspace;
4876
4877 /* PCRE supports POSIX class stuff inside a class. Perl gives an error if
4878 they are encountered at the top level, so we'll do that too. */
4879
4880 if ((ptr[1] == CHAR_COLON || ptr[1] == CHAR_DOT ||
4881 ptr[1] == CHAR_EQUALS_SIGN) &&
4882 check_posix_syntax(ptr, &tempptr))
4883 {
4884 *errorcodeptr = (ptr[1] == CHAR_COLON)? ERR13 : ERR31;
4885 goto FAILED;
4886 }
4887
4888 /* If the first character is '^', set the negation flag and skip it. Also,
4889 if the first few characters (either before or after ^) are \Q\E or \E we
4890 skip them too. This makes for compatibility with Perl. */
4891
4892 negate_class = FALSE;
4893 for (;;)
4894 {
4895 c = *(++ptr);
4896 if (c == CHAR_BACKSLASH)
4897 {
4898 if (ptr[1] == CHAR_E)
4899 ptr++;
4900 else if (STRNCMP_UC_C8(ptr + 1, STR_Q STR_BACKSLASH STR_E, 3) == 0)
4901 ptr += 3;
4902 else
4903 break;
4904 }
4905 else if (!negate_class && c == CHAR_CIRCUMFLEX_ACCENT)
4906 negate_class = TRUE;
4907 else break;
4908 }
4909
4910 /* Empty classes are allowed in JavaScript compatibility mode. Otherwise,
4911 an initial ']' is taken as a data character -- the code below handles
4912 that. In JS mode, [] must always fail, so generate OP_FAIL, whereas
4913 [^] must match any character, so generate OP_ALLANY. */
4914
4915 if (c == CHAR_RIGHT_SQUARE_BRACKET &&
4916 (cd->external_options & PCRE_JAVASCRIPT_COMPAT) != 0)
4917 {
4918 *code++ = negate_class? OP_ALLANY : OP_FAIL;
4919 if (firstcharflags == REQ_UNSET) firstcharflags = REQ_NONE;
4920 zerofirstchar = firstchar;
4921 zerofirstcharflags = firstcharflags;
4922 break;
4923 }
4924
4925 /* If a class contains a negative special such as \S, we need to flip the
4926 negation flag at the end, so that support for characters > 255 works
4927 correctly (they are all included in the class). */
4928
4929 should_flip_negation = FALSE;
4930
4931 /* Extended class (xclass) will be used when characters > 255
4932 might match. */
4933
4934 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
4935 xclass = FALSE;
4936 class_uchardata = code + LINK_SIZE + 2; /* For XCLASS items */
4937 class_uchardata_base = class_uchardata; /* Save the start */
4938 #endif
4939
4940 /* For optimization purposes, we track some properties of the class:
4941 class_has_8bitchar will be non-zero if the class contains at least one <
4942 256 character; class_one_char will be 1 if the class contains just one
4943 character; xclass_has_prop will be TRUE if unicode property checks
4944 are present in the class. */
4945
4946 class_has_8bitchar = 0;
4947 class_one_char = 0;
4948 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
4949 xclass_has_prop = FALSE;
4950 #endif
4951
4952 /* Initialize the 32-char bit map to all zeros. We build the map in a
4953 temporary bit of memory, in case the class contains fewer than two
4954 8-bit characters because in that case the compiled code doesn't use the bit
4955 map. */
4956
4957 memset(classbits, 0, 32 * sizeof(pcre_uint8));
4958
4959 /* Process characters until ] is reached. By writing this as a "do" it
4960 means that an initial ] is taken as a data character. At the start of the
4961 loop, c contains the first byte of the character. */
4962
4963 if (c != CHAR_NULL) do
4964 {
4965 const pcre_uchar *oldptr;
4966
4967 #ifdef SUPPORT_UTF
4968 if (utf && HAS_EXTRALEN(c))
4969 { /* Braces are required because the */
4970 GETCHARLEN(c, ptr, ptr); /* macro generates multiple statements */
4971 }
4972 #endif
4973
4974 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
4975 /* In the pre-compile phase, accumulate the length of any extra
4976 data and reset the pointer. This is so that very large classes that
4977 contain a zillion > 255 characters no longer overwrite the work space
4978 (which is on the stack). We have to remember that there was XCLASS data,
4979 however. */
4980
4981 if (class_uchardata > class_uchardata_base) xclass = TRUE;
4982
4983 if (lengthptr != NULL && class_uchardata > class_uchardata_base)
4984 {
4985 *lengthptr += (int)(class_uchardata - class_uchardata_base);
4986 class_uchardata = class_uchardata_base;
4987 }
4988 #endif
4989
4990 /* Inside \Q...\E everything is literal except \E */
4991
4992 if (inescq)
4993 {
4994 if (c == CHAR_BACKSLASH && ptr[1] == CHAR_E) /* If we are at \E */
4995 {
4996 inescq = FALSE; /* Reset literal state */
4997 ptr++; /* Skip the 'E' */
4998 continue; /* Carry on with next */
4999 }
5000 goto CHECK_RANGE; /* Could be range if \E follows */
5001 }
5002
5003 /* Handle POSIX class names. Perl allows a negation extension of the
5004 form [:^name:]. A square bracket that doesn't match the syntax is
5005 treated as a literal. We also recognize the POSIX constructions
5006 [.ch.] and [=ch=] ("collating elements") and fault them, as Perl
5007 5.6 and 5.8 do. */
5008
5009 if (c == CHAR_LEFT_SQUARE_BRACKET &&
5010 (ptr[1] == CHAR_COLON || ptr[1] == CHAR_DOT ||
5011 ptr[1] == CHAR_EQUALS_SIGN) && check_posix_syntax(ptr, &tempptr))
5012 {
5013 BOOL local_negate = FALSE;
5014 int posix_class, taboffset, tabopt;
5015 register const pcre_uint8 *cbits = cd->cbits;
5016 pcre_uint8 pbits[32];
5017
5018 if (ptr[1] != CHAR_COLON)
5019 {
5020 *errorcodeptr = ERR31;
5021 goto FAILED;
5022 }
5023
5024 ptr += 2;
5025 if (*ptr == CHAR_CIRCUMFLEX_ACCENT)
5026 {
5027 local_negate = TRUE;
5028 should_flip_negation = TRUE; /* Note negative special */
5029 ptr++;
5030 }
5031
5032 posix_class = check_posix_name(ptr, (int)(tempptr - ptr));
5033 if (posix_class < 0)
5034 {
5035 *errorcodeptr = ERR30;
5036 goto FAILED;
5037 }
5038
5039 /* If matching is caseless, upper and lower are converted to
5040 alpha. This relies on the fact that the class table starts with
5041 alpha, lower, upper as the first 3 entries. */
5042
5043 if ((options & PCRE_CASELESS) != 0 && posix_class <= 2)
5044 posix_class = 0;
5045
5046 /* When PCRE_UCP is set, some of the POSIX classes are converted to
5047 different escape sequences that use Unicode properties \p or \P. Others
5048 that are not available via \p or \P generate XCL_PROP/XCL_NOTPROP
5049 directly. */
5050
5051 #ifdef SUPPORT_UCP
5052 if ((options & PCRE_UCP) != 0)
5053 {
5054 unsigned int ptype = 0;
5055 int pc = posix_class + ((local_negate)? POSIX_SUBSIZE/2 : 0);
5056
5057 /* The posix_substitutes table specifies which POSIX classes can be
5058 converted to \p or \P items. */
5059
5060 if (posix_substitutes[pc] != NULL)
5061 {
5062 nestptr = tempptr + 1;
5063 ptr = posix_substitutes[pc] - 1;
5064 continue;
5065 }
5066
5067 /* There are three other classes that generate special property calls
5068 that are recognized only in an XCLASS. */
5069
5070 else switch(posix_class)
5071 {
5072 case PC_GRAPH:
5073 ptype = PT_PXGRAPH;
5074 /* Fall through */
5075 case PC_PRINT:
5076 if (ptype == 0) ptype = PT_PXPRINT;
5077 /* Fall through */
5078 case PC_PUNCT:
5079 if (ptype == 0) ptype = PT_PXPUNCT;
5080 *class_uchardata++ = local_negate? XCL_NOTPROP : XCL_PROP;
5081 *class_uchardata++ = ptype;
5082 *class_uchardata++ = 0;
5083 xclass_has_prop = TRUE;
5084 ptr = tempptr + 1;
5085 continue;
5086
5087 /* For the other POSIX classes (ascii, cntrl, xdigit) we are going
5088 to fall through to the non-UCP case and build a bit map for
5089 characters with code points less than 256. If we are in a negated
5090 POSIX class, characters with code points greater than 255 must
5091 either all match or all not match. In the special case where we
5092 have not yet generated any xclass data, and this is the final item
5093 in the overall class, we need do nothing: later on, the opcode
5094 OP_NCLASS will be used to indicate that characters greater than 255
5095 are acceptable. If we have already seen an xclass item or one may
5096 follow (we have to assume that it might if this is not the end of
5097 the class), explicitly list all wide codepoints, which will then
5098 either not match or match, depending on whether the class is or is
5099 not negated. */
5100
5101 default:
5102 if (local_negate &&
5103 (xclass || tempptr[2] != CHAR_RIGHT_SQUARE_BRACKET))
5104 {
5105 *class_uchardata++ = XCL_RANGE;
5106 class_uchardata += PRIV(ord2utf)(0x100, class_uchardata);
5107 class_uchardata += PRIV(ord2utf)(0x10ffff, class_uchardata);
5108 }
5109 break;
5110 }
5111 }
5112 #endif
5113 /* In the non-UCP case, or when UCP makes no difference, we build the
5114 bit map for the POSIX class in a chunk of local store because we may be
5115 adding and subtracting from it, and we don't want to subtract bits that
5116 may be in the main map already. At the end we or the result into the
5117 bit map that is being built. */
5118
5119 posix_class *= 3;
5120
5121 /* Copy in the first table (always present) */
5122
5123 memcpy(pbits, cbits + posix_class_maps[posix_class],
5124 32 * sizeof(pcre_uint8));
5125
5126 /* If there is a second table, add or remove it as required. */
5127
5128 taboffset = posix_class_maps[posix_class + 1];
5129 tabopt = posix_class_maps[posix_class + 2];
5130
5131 if (taboffset >= 0)
5132 {
5133 if (tabopt >= 0)
5134 for (c = 0; c < 32; c++) pbits[c] |= cbits[c + taboffset];
5135 else
5136 for (c = 0; c < 32; c++) pbits[c] &= ~cbits[c + taboffset];
5137 }
5138
5139 /* Now see if we need to remove any special characters. An option
5140 value of 1 removes vertical space and 2 removes underscore. */
5141
5142 if (tabopt < 0) tabopt = -tabopt;
5143 if (tabopt == 1) pbits[1] &= ~0x3c;
5144 else if (tabopt == 2) pbits[11] &= 0x7f;
5145
5146 /* Add the POSIX table or its complement into the main table that is
5147 being built and we are done. */
5148
5149 if (local_negate)
5150 for (c = 0; c < 32; c++) classbits[c] |= ~pbits[c];
5151 else
5152 for (c = 0; c < 32; c++) classbits[c] |= pbits[c];
5153
5154 ptr = tempptr + 1;
5155 /* Every class contains at least one < 256 character. */
5156 class_has_8bitchar = 1;
5157 /* Every class contains at least two characters. */
5158 class_one_char = 2;
5159 continue; /* End of POSIX syntax handling */
5160 }
5161
5162 /* Backslash may introduce a single character, or it may introduce one
5163 of the specials, which just set a flag. The sequence \b is a special
5164 case. Inside a class (and only there) it is treated as backspace. We
5165 assume that other escapes have more than one character in them, so
5166 speculatively set both class_has_8bitchar and class_one_char bigger
5167 than one. Unrecognized escapes fall through and are either treated
5168 as literal characters (by default), or are faulted if
5169 PCRE_EXTRA is set. */
5170
5171 if (c == CHAR_BACKSLASH)
5172 {
5173 escape = check_escape(&ptr, &ec, errorcodeptr, cd->bracount, options,
5174 TRUE);
5175 if (*errorcodeptr != 0) goto FAILED;
5176 if (escape == 0) c = ec;
5177 else if (escape == ESC_b) c = CHAR_BS; /* \b is backspace in a class */
5178 else if (escape == ESC_N) /* \N is not supported in a class */
5179 {
5180 *errorcodeptr = ERR71;
5181 goto FAILED;
5182 }
5183 else if (escape == ESC_Q) /* Handle start of quoted string */
5184 {
5185 if (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
5186 {
5187 ptr += 2; /* avoid empty string */
5188 }
5189 else inescq = TRUE;
5190 continue;
5191 }
5192 else if (escape == ESC_E) continue; /* Ignore orphan \E */
5193
5194 else
5195 {
5196 register const pcre_uint8 *cbits = cd->cbits;
5197 /* Every class contains at least two < 256 characters. */
5198 class_has_8bitchar++;
5199 /* Every class contains at least two characters. */
5200 class_one_char += 2;
5201
5202 switch (escape)
5203 {
5204 #ifdef SUPPORT_UCP
5205 case ESC_du: /* These are the values given for \d etc */
5206 case ESC_DU: /* when PCRE_UCP is set. We replace the */
5207 case ESC_wu: /* escape sequence with an appropriate \p */
5208 case ESC_WU: /* or \P to test Unicode properties instead */
5209 case ESC_su: /* of the default ASCII testing. */
5210 case ESC_SU:
5211 nestptr = ptr;
5212 ptr = substitutes[escape - ESC_DU] - 1; /* Just before substitute */
5213 class_has_8bitchar--; /* Undo! */
5214 continue;
5215 #endif
5216 case ESC_d:
5217 for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_digit];
5218 continue;
5219
5220 case ESC_D:
5221 should_flip_negation = TRUE;
5222 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_digit];
5223 continue;
5224
5225 case ESC_w:
5226 for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_word];
5227 continue;
5228
5229 case ESC_W:
5230 should_flip_negation = TRUE;
5231 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_word];
5232 continue;
5233
5234 /* Perl 5.004 onwards omitted VT from \s, but restored it at Perl
5235 5.18. Before PCRE 8.34, we had to preserve the VT bit if it was
5236 previously set by something earlier in the character class.
5237 Luckily, the value of CHAR_VT is 0x0b in both ASCII and EBCDIC, so
5238 we could just adjust the appropriate bit. From PCRE 8.34 we no
5239 longer treat \s and \S specially. */
5240
5241 case ESC_s:
5242 for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_space];
5243 continue;
5244
5245 case ESC_S:
5246 should_flip_negation = TRUE;
5247 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_space];
5248 continue;
5249
5250 /* The rest apply in both UCP and non-UCP cases. */
5251
5252 case ESC_h:
5253 (void)add_list_to_class(classbits, &class_uchardata, options, cd,
5254 PRIV(hspace_list), NOTACHAR);
5255 continue;
5256
5257 case ESC_H:
5258 (void)add_not_list_to_class(classbits, &class_uchardata, options,
5259 cd, PRIV(hspace_list));
5260 continue;
5261
5262 case ESC_v:
5263 (void)add_list_to_class(classbits, &class_uchardata, options, cd,
5264 PRIV(vspace_list), NOTACHAR);
5265 continue;
5266
5267 case ESC_V:
5268 (void)add_not_list_to_class(classbits, &class_uchardata, options,
5269 cd, PRIV(vspace_list));
5270 continue;
5271
5272 case ESC_p:
5273 case ESC_P:
5274 #ifdef SUPPORT_UCP
5275 {
5276 BOOL negated;
5277 unsigned int ptype = 0, pdata = 0;
5278 if (!get_ucp(&ptr, &negated, &ptype, &pdata, errorcodeptr))
5279 goto FAILED;
5280 *class_uchardata++ = ((escape == ESC_p) != negated)?
5281 XCL_PROP : XCL_NOTPROP;
5282 *class_uchardata++ = ptype;
5283 *class_uchardata++ = pdata;
5284 xclass_has_prop = TRUE;
5285 class_has_8bitchar--; /* Undo! */
5286 continue;
5287 }
5288 #else
5289 *errorcodeptr = ERR45;
5290 goto FAILED;
5291 #endif
5292 /* Unrecognized escapes are faulted if PCRE is running in its
5293 strict mode. By default, for compatibility with Perl, they are
5294 treated as literals. */
5295
5296 default:
5297 if ((options & PCRE_EXTRA) != 0)
5298 {
5299 *errorcodeptr = ERR7;
5300 goto FAILED;
5301 }
5302 class_has_8bitchar--; /* Undo the speculative increase. */
5303 class_one_char -= 2; /* Undo the speculative increase. */
5304 c = *ptr; /* Get the final character and fall through */
5305 break;
5306 }
5307 }
5308
5309 /* Fall through if the escape just defined a single character (c >= 0).
5310 This may be greater than 256. */
5311
5312 escape = 0;
5313
5314 } /* End of backslash handling */
5315
5316 /* A character may be followed by '-' to form a range. However, Perl does
5317 not permit ']' to be the end of the range. A '-' character at the end is
5318 treated as a literal. Perl ignores orphaned \E sequences entirely. The
5319 code for handling \Q and \E is messy. */
5320
5321 CHECK_RANGE:
5322 while (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
5323 {
5324 inescq = FALSE;
5325 ptr += 2;
5326 }
5327 oldptr = ptr;
5328
5329 /* Remember if \r or \n were explicitly used */
5330
5331 if (c == CHAR_CR || c == CHAR_NL) cd->external_flags |= PCRE_HASCRORLF;
5332
5333 /* Check for range */
5334
5335 if (!inescq && ptr[1] == CHAR_MINUS)
5336 {
5337 pcre_uint32 d;
5338 ptr += 2;
5339 while (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_E) ptr += 2;
5340
5341 /* If we hit \Q (not followed by \E) at this point, go into escaped
5342 mode. */
5343
5344 while (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_Q)
5345 {
5346 ptr += 2;
5347 if (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_E)
5348 { ptr += 2; continue; }
5349 inescq = TRUE;
5350 break;
5351 }
5352
5353 /* Minus (hyphen) at the end of a class is treated as a literal, so put
5354 back the pointer and jump to handle the character that preceded it. */
5355
5356 if (*ptr == CHAR_NULL || (!inescq && *ptr == CHAR_RIGHT_SQUARE_BRACKET))
5357 {
5358 ptr = oldptr;
5359 goto CLASS_SINGLE_CHARACTER;
5360 }
5361
5362 /* Otherwise, we have a potential range; pick up the next character */
5363
5364 #ifdef SUPPORT_UTF
5365 if (utf)
5366 { /* Braces are required because the */
5367 GETCHARLEN(d, ptr, ptr); /* macro generates multiple statements */
5368 }
5369 else
5370 #endif
5371 d = *ptr; /* Not UTF-8 mode */
5372
5373 /* The second part of a range can be a single-character escape
5374 sequence, but not any of the other escapes. Perl treats a hyphen as a
5375 literal in such circumstances. However, in Perl's warning mode, a
5376 warning is given, so PCRE now faults it as it is almost certainly a
5377 mistake on the user's part. */
5378
5379 if (!inescq)
5380 {
5381 if (d == CHAR_BACKSLASH)
5382 {
5383 int descape;
5384 descape = check_escape(&ptr, &d, errorcodeptr, cd->bracount, options, TRUE);
5385 if (*errorcodeptr != 0) goto FAILED;
5386
5387 /* 0 means a character was put into d; \b is backspace; any other
5388 special causes an error. */
5389
5390 if (descape != 0)
5391 {
5392 if (descape == ESC_b) d = CHAR_BS; else
5393 {
5394 *errorcodeptr = ERR83;
5395 goto FAILED;
5396 }
5397 }
5398 }
5399
5400 /* A hyphen followed by a POSIX class is treated in the same way. */
5401
5402 else if (d == CHAR_LEFT_SQUARE_BRACKET &&
5403 (ptr[1] == CHAR_COLON || ptr[1] == CHAR_DOT ||
5404 ptr[1] == CHAR_EQUALS_SIGN) &&
5405 check_posix_syntax(ptr, &tempptr))
5406 {
5407 *errorcodeptr = ERR83;
5408 goto FAILED;
5409 }
5410 }
5411
5412 /* Check that the two values are in the correct order. Optimize
5413 one-character ranges. */
5414
5415 if (d < c)
5416 {
5417 *errorcodeptr = ERR8;
5418 goto FAILED;
5419 }
5420 if (d == c) goto CLASS_SINGLE_CHARACTER; /* A few lines below */
5421
5422 /* We have found a character range, so single character optimizations
5423 cannot be done anymore. Any value greater than 1 indicates that there
5424 is more than one character. */
5425
5426 class_one_char = 2;
5427
5428 /* Remember an explicit \r or \n, and add the range to the class. */
5429
5430 if (d == CHAR_CR || d == CHAR_NL) cd->external_flags |= PCRE_HASCRORLF;
5431
5432 class_has_8bitchar +=
5433 add_to_class(classbits, &class_uchardata, options, cd, c, d);
5434
5435 continue; /* Go get the next char in the class */
5436 }
5437
5438 /* Handle a single character - we can get here for a normal non-escape
5439 char, or after \ that introduces a single character or for an apparent
5440 range that isn't. Only the value 1 matters for class_one_char, so don't
5441 increase it if it is already 2 or more ... just in case there's a class
5442 with a zillion characters in it. */
5443
5444 CLASS_SINGLE_CHARACTER:
5445 if (class_one_char < 2) class_one_char++;
5446
5447 /* If xclass_has_prop is false and class_one_char is 1, we have the first
5448 single character in the class, and there have been no prior ranges, or
5449 XCLASS items generated by escapes. If this is the final character in the
5450 class, we can optimize by turning the item into a 1-character OP_CHAR[I]
5451 if it's positive, or OP_NOT[I] if it's negative. In the positive case, it
5452 can cause firstchar to be set. Otherwise, there can be no first char if
5453 this item is first, whatever repeat count may follow. In the case of
5454 reqchar, save the previous value for reinstating. */
5455
5456 if (!inescq &&
5457 #ifdef SUPPORT_UCP
5458 !xclass_has_prop &&
5459 #endif
5460 class_one_char == 1 && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET)
5461 {
5462 ptr++;
5463 zeroreqchar = reqchar;
5464 zeroreqcharflags = reqcharflags;
5465
5466 if (negate_class)
5467 {
5468 #ifdef SUPPORT_UCP
5469 int d;
5470 #endif
5471 if (firstcharflags == REQ_UNSET) firstcharflags = REQ_NONE;
5472 zerofirstchar = firstchar;
5473 zerofirstcharflags = firstcharflags;
5474
5475 /* For caseless UTF-8 mode when UCP support is available, check
5476 whether this character has more than one other case. If so, generate
5477 a special OP_NOTPROP item instead of OP_NOTI. */
5478
5479 #ifdef SUPPORT_UCP
5480 if (utf && (options & PCRE_CASELESS) != 0 &&
5481 (d = UCD_CASESET(c)) != 0)
5482 {
5483 *code++ = OP_NOTPROP;
5484 *code++ = PT_CLIST;
5485 *code++ = d;
5486 }
5487 else
5488 #endif
5489 /* Char has only one other case, or UCP not available */
5490
5491 {
5492 *code++ = ((options & PCRE_CASELESS) != 0)? OP_NOTI: OP_NOT;
5493 #if defined SUPPORT_UTF && !defined COMPILE_PCRE32
5494 if (utf && c > MAX_VALUE_FOR_SINGLE_CHAR)
5495 code += PRIV(ord2utf)(c, code);
5496 else
5497 #endif
5498 *code++ = c;
5499 }
5500
5501 /* We are finished with this character class */
5502
5503 goto END_CLASS;
5504 }
5505
5506 /* For a single, positive character, get the value into mcbuffer, and
5507 then we can handle this with the normal one-character code. */
5508
5509 #if defined SUPPORT_UTF && !defined COMPILE_PCRE32
5510 if (utf && c > MAX_VALUE_FOR_SINGLE_CHAR)
5511 mclength = PRIV(ord2utf)(c, mcbuffer);
5512 else
5513 #endif
5514 {
5515 mcbuffer[0] = c;
5516 mclength = 1;
5517 }
5518 goto ONE_CHAR;
5519 } /* End of 1-char optimization */
5520
5521 /* There is more than one character in the class, or an XCLASS item
5522 has been generated. Add this character to the class. */
5523
5524 class_has_8bitchar +=
5525 add_to_class(classbits, &class_uchardata, options, cd, c, c);
5526 }
5527
5528 /* Loop until ']' reached. This "while" is the end of the "do" far above.
5529 If we are at the end of an internal nested string, revert to the outer
5530 string. */
5531
5532 while (((c = *(++ptr)) != CHAR_NULL ||
5533 (nestptr != NULL &&
5534 (ptr = nestptr, nestptr = NULL, c = *(++ptr)) != CHAR_NULL)) &&
5535 (c != CHAR_RIGHT_SQUARE_BRACKET || inescq));
5536
5537 /* Check for missing terminating ']' */
5538
5539 if (c == CHAR_NULL)
5540 {
5541 *errorcodeptr = ERR6;
5542 goto FAILED;
5543 }
5544
5545 /* We will need an XCLASS if data has been placed in class_uchardata. In
5546 the second phase this is a sufficient test. However, in the pre-compile
5547 phase, class_uchardata gets emptied to prevent workspace overflow, so it
5548 only if the very last character in the class needs XCLASS will it contain
5549 anything at this point. For this reason, xclass gets set TRUE above when
5550 uchar_classdata is emptied, and that's why this code is the way it is here
5551 instead of just doing a test on class_uchardata below. */
5552
5553 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
5554 if (class_uchardata > class_uchardata_base) xclass = TRUE;
5555 #endif
5556
5557 /* If this is the first thing in the branch, there can be no first char
5558 setting, whatever the repeat count. Any reqchar setting must remain
5559 unchanged after any kind of repeat. */
5560
5561 if (firstcharflags == REQ_UNSET) firstcharflags = REQ_NONE;
5562 zerofirstchar = firstchar;
5563 zerofirstcharflags = firstcharflags;
5564 zeroreqchar = reqchar;
5565 zeroreqcharflags = reqcharflags;
5566
5567 /* If there are characters with values > 255, we have to compile an
5568 extended class, with its own opcode, unless there was a negated special
5569 such as \S in the class, and PCRE_UCP is not set, because in that case all
5570 characters > 255 are in the class, so any that were explicitly given as
5571 well can be ignored. If (when there are explicit characters > 255 that must
5572 be listed) there are no characters < 256, we can omit the bitmap in the
5573 actual compiled code. */
5574
5575 #ifdef SUPPORT_UTF
5576 if (xclass && (xclass_has_prop || !should_flip_negation ||
5577 (options & PCRE_UCP) != 0))
5578 #elif !defined COMPILE_PCRE8
5579 if (xclass && (xclass_has_prop || !should_flip_negation))
5580 #endif
5581 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
5582 {
5583 /* For non-UCP wide characters, in a non-negative class containing \S or
5584 similar (should_flip_negation is set), all characters greater than 255
5585 must be in the class. */
5586
5587 if (
5588 #if defined COMPILE_PCRE8
5589 utf &&
5590 #endif
5591 should_flip_negation && !negate_class && (options & PCRE_UCP) == 0)
5592 {
5593 *class_uchardata++ = XCL_RANGE;
5594 if (utf) /* Will always be utf in the 8-bit library */
5595 {
5596 class_uchardata += PRIV(ord2utf)(0x100, class_uchardata);
5597 class_uchardata += PRIV(ord2utf)(0x10ffff, class_uchardata);
5598 }
5599 else /* Can only happen for the 16-bit & 32-bit libraries */
5600 {
5601 #if defined COMPILE_PCRE16
5602 *class_uchardata++ = 0x100;
5603 *class_uchardata++ = 0xffffu;
5604 #elif defined COMPILE_PCRE32
5605 *class_uchardata++ = 0x100;
5606 *class_uchardata++ = 0xffffffffu;
5607 #endif
5608 }
5609 }
5610
5611 *class_uchardata++ = XCL_END; /* Marks the end of extra data */
5612 *code++ = OP_XCLASS;
5613 code += LINK_SIZE;
5614 *code = negate_class? XCL_NOT:0;
5615 if (xclass_has_prop) *code |= XCL_HASPROP;
5616
5617 /* If the map is required, move up the extra data to make room for it;
5618 otherwise just move the code pointer to the end of the extra data. */
5619
5620 if (class_has_8bitchar > 0)
5621 {
5622 *code++ |= XCL_MAP;
5623 memmove(code + (32 / sizeof(pcre_uchar)), code,
5624 IN_UCHARS(class_uchardata - code));
5625 if (negate_class && !xclass_has_prop)
5626 for (c = 0; c < 32; c++) classbits[c] = ~classbits[c];
5627 memcpy(code, classbits, 32);
5628 code = class_uchardata + (32 / sizeof(pcre_uchar));
5629 }
5630 else code = class_uchardata;
5631
5632 /* Now fill in the complete length of the item */
5633
5634 PUT(previous, 1, (int)(code - previous));
5635 break; /* End of class handling */
5636 }
5637
5638 /* Even though any XCLASS list is now discarded, we must allow for
5639 its memory. */
5640
5641 if (lengthptr != NULL)
5642 *lengthptr += (int)(class_uchardata - class_uchardata_base);
5643 #endif
5644
5645 /* If there are no characters > 255, or they are all to be included or
5646 excluded, set the opcode to OP_CLASS or OP_NCLASS, depending on whether the
5647 whole class was negated and whether there were negative specials such as \S
5648 (non-UCP) in the class. Then copy the 32-byte map into the code vector,
5649 negating it if necessary. */
5650
5651 *code++ = (negate_class == should_flip_negation) ? OP_CLASS : OP_NCLASS;
5652 if (lengthptr == NULL) /* Save time in the pre-compile phase */
5653 {
5654 if (negate_class)
5655 for (c = 0; c < 32; c++) classbits[c] = ~classbits[c];
5656 memcpy(code, classbits, 32);
5657 }
5658 code += 32 / sizeof(pcre_uchar);
5659
5660 END_CLASS:
5661 break;
5662
5663
5664 /* ===================================================================*/
5665 /* Various kinds of repeat; '{' is not necessarily a quantifier, but this
5666 has been tested above. */
5667
5668 case CHAR_LEFT_CURLY_BRACKET:
5669 if (!is_quantifier) goto NORMAL_CHAR;
5670 ptr = read_repeat_counts(ptr+1, &repeat_min, &repeat_max, errorcodeptr);
5671 if (*errorcodeptr != 0) goto FAILED;
5672 goto REPEAT;
5673
5674 case CHAR_ASTERISK:
5675 repeat_min = 0;
5676 repeat_max = -1;
5677 goto REPEAT;
5678
5679 case CHAR_PLUS:
5680 repeat_min = 1;
5681 repeat_max = -1;
5682 goto REPEAT;
5683
5684 case CHAR_QUESTION_MARK:
5685 repeat_min = 0;
5686 repeat_max = 1;
5687
5688 REPEAT:
5689 if (previous == NULL)
5690 {
5691 *errorcodeptr = ERR9;
5692 goto FAILED;
5693 }
5694
5695 if (repeat_min == 0)
5696 {
5697 firstchar = zerofirstchar; /* Adjust for zero repeat */
5698 firstcharflags = zerofirstcharflags;
5699 reqchar = zeroreqchar; /* Ditto */
5700 reqcharflags = zeroreqcharflags;
5701 }
5702
5703 /* Remember whether this is a variable length repeat */
5704
5705 reqvary = (repeat_min == repeat_max)? 0 : REQ_VARY;
5706
5707 op_type = 0; /* Default single-char op codes */
5708 possessive_quantifier = FALSE; /* Default not possessive quantifier */
5709
5710 /* Save start of previous item, in case we have to move it up in order to
5711 insert something before it. */
5712
5713 tempcode = previous;
5714
5715 /* Before checking for a possessive quantifier, we must skip over
5716 whitespace and comments in extended mode because Perl allows white space at
5717 this point. */
5718
5719 if ((options & PCRE_EXTENDED) != 0)
5720 {
5721 const pcre_uchar *p = ptr + 1;
5722 for (;;)
5723 {
5724 while (MAX_255(*p) && (cd->ctypes[*p] & ctype_space) != 0) p++;
5725 if (*p != CHAR_NUMBER_SIGN) break;
5726 p++;
5727 while (*p != CHAR_NULL)
5728 {
5729 if (IS_NEWLINE(p)) /* For non-fixed-length newline cases, */
5730 { /* IS_NEWLINE sets cd->nllen. */
5731 p += cd->nllen;
5732 break;
5733 }
5734 p++;
5735 #ifdef SUPPORT_UTF
5736 if (utf) FORWARDCHAR(p);
5737 #endif
5738 } /* Loop for comment characters */
5739 } /* Loop for multiple comments */
5740 ptr = p - 1; /* Character before the next significant one. */
5741 }
5742
5743 /* We also need to skip over (?# comments, which are not dependent on
5744 extended mode. */
5745
5746 if (ptr[1] == CHAR_LEFT_PARENTHESIS && ptr[2] == CHAR_QUESTION_MARK &&
5747 ptr[3] == CHAR_NUMBER_SIGN)
5748 {
5749 ptr += 4;
5750 while (*ptr != CHAR_NULL && *ptr != CHAR_RIGHT_PARENTHESIS) ptr++;
5751 if (*ptr == CHAR_NULL)
5752 {
5753 *errorcodeptr = ERR18;
5754 goto FAILED;
5755 }
5756 }
5757
5758 /* If the next character is '+', we have a possessive quantifier. This
5759 implies greediness, whatever the setting of the PCRE_UNGREEDY option.
5760 If the next character is '?' this is a minimizing repeat, by default,
5761 but if PCRE_UNGREEDY is set, it works the other way round. We change the
5762 repeat type to the non-default. */
5763
5764 if (ptr[1] == CHAR_PLUS)
5765 {
5766 repeat_type = 0; /* Force greedy */
5767 possessive_quantifier = TRUE;
5768 ptr++;
5769 }
5770 else if (ptr[1] == CHAR_QUESTION_MARK)
5771 {
5772 repeat_type = greedy_non_default;
5773 ptr++;
5774 }
5775 else repeat_type = greedy_default;
5776
5777 /* If previous was a recursion call, wrap it in atomic brackets so that
5778 previous becomes the atomic group. All recursions were so wrapped in the
5779 past, but it no longer happens for non-repeated recursions. In fact, the
5780 repeated ones could be re-implemented independently so as not to need this,
5781 but for the moment we rely on the code for repeating groups. */
5782
5783 if (*previous == OP_RECURSE)
5784 {
5785 memmove(previous + 1 + LINK_SIZE, previous, IN_UCHARS(1 + LINK_SIZE));
5786 *previous = OP_ONCE;
5787 PUT(previous, 1, 2 + 2*LINK_SIZE);
5788 previous[2 + 2*LINK_SIZE] = OP_KET;
5789 PUT(previous, 3 + 2*LINK_SIZE, 2 + 2*LINK_SIZE);
5790 code += 2 + 2 * LINK_SIZE;
5791 length_prevgroup = 3 + 3*LINK_SIZE;
5792
5793 /* When actually compiling, we need to check whether this was a forward
5794 reference, and if so, adjust the offset. */
5795
5796 if (lengthptr == NULL && cd->hwm >= cd->start_workspace + LINK_SIZE)
5797 {
5798 int offset = GET(cd->hwm, -LINK_SIZE);
5799 if (offset == previous + 1 - cd->start_code)
5800 PUT(cd->hwm, -LINK_SIZE, offset + 1 + LINK_SIZE);
5801 }
5802 }
5803
5804 /* Now handle repetition for the different types of item. */
5805
5806 /* If previous was a character or negated character match, abolish the item
5807 and generate a repeat item instead. If a char item has a minimum of more
5808 than one, ensure that it is set in reqchar - it might not be if a sequence
5809 such as x{3} is the first thing in a branch because the x will have gone
5810 into firstchar instead. */
5811
5812 if (*previous == OP_CHAR || *previous == OP_CHARI
5813 || *previous == OP_NOT || *previous == OP_NOTI)
5814 {
5815 switch (*previous)
5816 {
5817 default: /* Make compiler happy. */
5818 case OP_CHAR: op_type = OP_STAR - OP_STAR; break;
5819 case OP_CHARI: op_type = OP_STARI - OP_STAR; break;
5820 case OP_NOT: op_type = OP_NOTSTAR - OP_STAR; break;
5821 case OP_NOTI: op_type = OP_NOTSTARI - OP_STAR; break;
5822 }
5823
5824 /* Deal with UTF characters that take up more than one character. It's
5825 easier to write this out separately than try to macrify it. Use c to
5826 hold the length of the character in bytes, plus UTF_LENGTH to flag that
5827 it's a length rather than a small character. */
5828
5829 #if defined SUPPORT_UTF && !defined COMPILE_PCRE32
5830 if (utf && NOT_FIRSTCHAR(code[-1]))
5831 {
5832 pcre_uchar *lastchar = code - 1;
5833 BACKCHAR(lastchar);
5834 c = (int)(code - lastchar); /* Length of UTF-8 character */
5835 memcpy(utf_chars, lastchar, IN_UCHARS(c)); /* Save the char */
5836 c |= UTF_LENGTH; /* Flag c as a length */
5837 }
5838 else
5839 #endif /* SUPPORT_UTF */
5840
5841 /* Handle the case of a single charater - either with no UTF support, or
5842 with UTF disabled, or for a single character UTF character. */
5843 {
5844 c = code[-1];
5845 if (*previous <= OP_CHARI && repeat_min > 1)
5846 {
5847 reqchar = c;
5848 reqcharflags = req_caseopt | cd->req_varyopt;
5849 }
5850 }
5851
5852 goto OUTPUT_SINGLE_REPEAT; /* Code shared with single character types */
5853 }
5854
5855 /* If previous was a character type match (\d or similar), abolish it and
5856 create a suitable repeat item. The code is shared with single-character
5857 repeats by setting op_type to add a suitable offset into repeat_type. Note
5858 the the Unicode property types will be present only when SUPPORT_UCP is
5859 defined, but we don't wrap the little bits of code here because it just
5860 makes it horribly messy. */
5861
5862 else if (*previous < OP_EODN)
5863 {
5864 pcre_uchar *oldcode;
5865 int prop_type, prop_value;
5866 op_type = OP_TYPESTAR - OP_STAR; /* Use type opcodes */
5867 c = *previous;
5868
5869 OUTPUT_SINGLE_REPEAT:
5870 if (*previous == OP_PROP || *previous == OP_NOTPROP)
5871 {
5872 prop_type = previous[1];
5873 prop_value = previous[2];
5874 }
5875 else prop_type = prop_value = -1;
5876
5877 oldcode = code;
5878 code = previous; /* Usually overwrite previous item */
5879
5880 /* If the maximum is zero then the minimum must also be zero; Perl allows
5881 this case, so we do too - by simply omitting the item altogether. */
5882
5883 if (repeat_max == 0) goto END_REPEAT;
5884
5885 /* Combine the op_type with the repeat_type */
5886
5887 repeat_type += op_type;
5888
5889 /* A minimum of zero is handled either as the special case * or ?, or as
5890 an UPTO, with the maximum given. */
5891
5892 if (repeat_min == 0)
5893 {
5894 if (repeat_max == -1) *code++ = OP_STAR + repeat_type;
5895 else if (repeat_max == 1) *code++ = OP_QUERY + repeat_type;
5896 else
5897 {
5898 *code++ = OP_UPTO + repeat_type;
5899 PUT2INC(code, 0, repeat_max);
5900 }
5901 }
5902
5903 /* A repeat minimum of 1 is optimized into some special cases. If the
5904 maximum is unlimited, we use OP_PLUS. Otherwise, the original item is
5905 left in place and, if the maximum is greater than 1, we use OP_UPTO with
5906 one less than the maximum. */
5907
5908 else if (repeat_min == 1)
5909 {
5910 if (repeat_max == -1)
5911 *code++ = OP_PLUS + repeat_type;
5912 else
5913 {
5914 code = oldcode; /* leave previous item in place */
5915 if (repeat_max == 1) goto END_REPEAT;
5916 *code++ = OP_UPTO + repeat_type;
5917 PUT2INC(code, 0, repeat_max - 1);
5918 }
5919 }
5920
5921 /* The case {n,n} is just an EXACT, while the general case {n,m} is
5922 handled as an EXACT followed by an UPTO. */
5923
5924 else
5925 {
5926 *code++ = OP_EXACT + op_type; /* NB EXACT doesn't have repeat_type */
5927 PUT2INC(code, 0, repeat_min);
5928
5929 /* If the maximum is unlimited, insert an OP_STAR. Before doing so,
5930 we have to insert the character for the previous code. For a repeated
5931 Unicode property match, there are two extra bytes that define the
5932 required property. In UTF-8 mode, long characters have their length in
5933 c, with the UTF_LENGTH bit as a flag. */
5934
5935 if (repeat_max < 0)
5936 {
5937 #if defined SUPPORT_UTF && !defined COMPILE_PCRE32
5938 if (utf && (c & UTF_LENGTH) != 0)
5939 {
5940 memcpy(code, utf_chars, IN_UCHARS(c & 7));
5941 code += c & 7;
5942 }
5943 else
5944 #endif
5945 {
5946 *code++ = c;
5947 if (prop_type >= 0)
5948 {
5949 *code++ = prop_type;
5950 *code++ = prop_value;
5951 }
5952 }
5953 *code++ = OP_STAR + repeat_type;
5954 }
5955
5956 /* Else insert an UPTO if the max is greater than the min, again
5957 preceded by the character, for the previously inserted code. If the
5958 UPTO is just for 1 instance, we can use QUERY instead. */
5959
5960 else if (repeat_max != repeat_min)
5961 {
5962 #if defined SUPPORT_UTF && !defined COMPILE_PCRE32
5963 if (utf && (c & UTF_LENGTH) != 0)
5964 {
5965 memcpy(code, utf_chars, IN_UCHARS(c & 7));
5966 code += c & 7;
5967 }
5968 else
5969 #endif
5970 *code++ = c;
5971 if (prop_type >= 0)
5972 {
5973 *code++ = prop_type;
5974 *code++ = prop_value;
5975 }
5976 repeat_max -= repeat_min;
5977
5978 if (repeat_max == 1)
5979 {
5980 *code++ = OP_QUERY + repeat_type;
5981 }
5982 else
5983 {
5984 *code++ = OP_UPTO + repeat_type;
5985 PUT2INC(code, 0, repeat_max);
5986 }
5987 }
5988 }
5989
5990 /* The character or character type itself comes last in all cases. */
5991
5992 #if defined SUPPORT_UTF && !defined COMPILE_PCRE32
5993 if (utf && (c & UTF_LENGTH) != 0)
5994 {
5995 memcpy(code, utf_chars, IN_UCHARS(c & 7));
5996 code += c & 7;
5997 }
5998 else
5999 #endif
6000 *code++ = c;
6001
6002 /* For a repeated Unicode property match, there are two extra bytes that
6003 define the required property. */
6004
6005 #ifdef SUPPORT_UCP
6006 if (prop_type >= 0)
6007 {
6008 *code++ = prop_type;
6009 *code++ = prop_value;
6010 }
6011 #endif
6012 }
6013
6014 /* If previous was a character class or a back reference, we put the repeat
6015 stuff after it, but just skip the item if the repeat was {0,0}. */
6016
6017 else if (*previous == OP_CLASS || *previous == OP_NCLASS ||
6018 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
6019 *previous == OP_XCLASS ||
6020 #endif
6021 *previous == OP_REF || *previous == OP_REFI ||
6022 *previous == OP_DNREF || *previous == OP_DNREFI)
6023 {
6024 if (repeat_max == 0)
6025 {
6026 code = previous;
6027 goto END_REPEAT;
6028 }
6029
6030 if (repeat_min == 0 && repeat_max == -1)
6031 *code++ = OP_CRSTAR + repeat_type;
6032 else if (repeat_min == 1 && repeat_max == -1)
6033 *code++ = OP_CRPLUS + repeat_type;
6034 else if (repeat_min == 0 && repeat_max == 1)
6035 *code++ = OP_CRQUERY + repeat_type;
6036 else
6037 {
6038 *code++ = OP_CRRANGE + repeat_type;
6039 PUT2INC(code, 0, repeat_min);
6040 if (repeat_max == -1) repeat_max = 0; /* 2-byte encoding for max */
6041 PUT2INC(code, 0, repeat_max);
6042 }
6043 }
6044
6045 /* If previous was a bracket group, we may have to replicate it in certain
6046 cases. Note that at this point we can encounter only the "basic" bracket
6047 opcodes such as BRA and CBRA, as this is the place where they get converted
6048 into the more special varieties such as BRAPOS and SBRA. A test for >=
6049 OP_ASSERT and <= OP_COND includes ASSERT, ASSERT_NOT, ASSERTBACK,
6050 ASSERTBACK_NOT, ONCE, ONCE_NC, BRA, BRAPOS, CBRA, CBRAPOS, and COND.
6051 Originally, PCRE did not allow repetition of assertions, but now it does,
6052 for Perl compatibility. */
6053
6054 else if (*previous >= OP_ASSERT && *previous <= OP_COND)
6055 {
6056 register int i;
6057 int len = (int)(code - previous);
6058 size_t base_hwm_offset = item_hwm_offset;
6059 pcre_uchar *bralink = NULL;
6060 pcre_uchar *brazeroptr = NULL;
6061
6062 /* Repeating a DEFINE group is pointless, but Perl allows the syntax, so
6063 we just ignore the repeat. */
6064
6065 if (*previous == OP_COND && previous[LINK_SIZE+1] == OP_DEF)
6066 goto END_REPEAT;
6067
6068 /* There is no sense in actually repeating assertions. The only potential
6069 use of repetition is in cases when the assertion is optional. Therefore,
6070 if the minimum is greater than zero, just ignore the repeat. If the
6071 maximum is not zero or one, set it to 1. */
6072
6073 if (*previous < OP_ONCE) /* Assertion */
6074 {
6075 if (repeat_min > 0) goto END_REPEAT;
6076 if (repeat_max < 0 || repeat_max > 1) repeat_max = 1;
6077 }
6078
6079 /* The case of a zero minimum is special because of the need to stick
6080 OP_BRAZERO in front of it, and because the group appears once in the
6081 data, whereas in other cases it appears the minimum number of times. For
6082 this reason, it is simplest to treat this case separately, as otherwise
6083 the code gets far too messy. There are several special subcases when the
6084 minimum is zero. */
6085
6086 if (repeat_min == 0)
6087 {
6088 /* If the maximum is also zero, we used to just omit the group from the
6089 output altogether, like this:
6090
6091 ** if (repeat_max == 0)
6092 ** {
6093 ** code = previous;
6094 ** goto END_REPEAT;
6095 ** }
6096
6097 However, that fails when a group or a subgroup within it is referenced
6098 as a subroutine from elsewhere in the pattern, so now we stick in
6099 OP_SKIPZERO in front of it so that it is skipped on execution. As we
6100 don't have a list of which groups are referenced, we cannot do this
6101 selectively.
6102
6103 If the maximum is 1 or unlimited, we just have to stick in the BRAZERO
6104 and do no more at this point. However, we do need to adjust any
6105 OP_RECURSE calls inside the group that refer to the group itself or any
6106 internal or forward referenced group, because the offset is from the
6107 start of the whole regex. Temporarily terminate the pattern while doing
6108 this. */
6109
6110 if (repeat_max <= 1) /* Covers 0, 1, and unlimited */
6111 {
6112 *code = OP_END;
6113 adjust_recurse(previous, 1, utf, cd, item_hwm_offset);
6114 memmove(previous + 1, previous, IN_UCHARS(len));
6115 code++;
6116 if (repeat_max == 0)
6117 {
6118 *previous++ = OP_SKIPZERO;
6119 goto END_REPEAT;
6120 }
6121 brazeroptr = previous; /* Save for possessive optimizing */
6122 *previous++ = OP_BRAZERO + repeat_type;
6123 }
6124
6125 /* If the maximum is greater than 1 and limited, we have to replicate
6126 in a nested fashion, sticking OP_BRAZERO before each set of brackets.
6127 The first one has to be handled carefully because it's the original
6128 copy, which has to be moved up. The remainder can be handled by code
6129 that is common with the non-zero minimum case below. We have to
6130 adjust the value or repeat_max, since one less copy is required. Once
6131 again, we may have to adjust any OP_RECURSE calls inside the group. */
6132
6133 else
6134 {
6135 int offset;
6136 *code = OP_END;
6137 adjust_recurse(previous, 2 + LINK_SIZE, utf, cd, item_hwm_offset);
6138 memmove(previous + 2 + LINK_SIZE, previous, IN_UCHARS(len));
6139 code += 2 + LINK_SIZE;
6140 *previous++ = OP_BRAZERO + repeat_type;
6141 *previous++ = OP_BRA;
6142
6143 /* We chain together the bracket offset fields that have to be
6144 filled in later when the ends of the brackets are reached. */
6145
6146 offset = (bralink == NULL)? 0 : (int)(previous - bralink);
6147 bralink = previous;
6148 PUTINC(previous, 0, offset);
6149 }
6150
6151 repeat_max--;
6152 }
6153
6154 /* If the minimum is greater than zero, replicate the group as many
6155 times as necessary, and adjust the maximum to the number of subsequent
6156 copies that we need. If we set a first char from the group, and didn't
6157 set a required char, copy the latter from the former. If there are any
6158 forward reference subroutine calls in the group, there will be entries on
6159 the workspace list; replicate these with an appropriate increment. */
6160
6161 else
6162 {
6163 if (repeat_min > 1)
6164 {
6165 /* In the pre-compile phase, we don't actually do the replication. We
6166 just adjust the length as if we had. Do some paranoid checks for
6167 potential integer overflow. The INT64_OR_DOUBLE type is a 64-bit
6168 integer type when available, otherwise double. */
6169
6170 if (lengthptr != NULL)
6171 {
6172 int delta = (repeat_min - 1)*length_prevgroup;
6173 if ((INT64_OR_DOUBLE)(repeat_min - 1)*
6174 (INT64_OR_DOUBLE)length_prevgroup >
6175 (INT64_OR_DOUBLE)INT_MAX ||
6176 OFLOW_MAX - *lengthptr < delta)
6177 {
6178 *errorcodeptr = ERR20;
6179 goto FAILED;
6180 }
6181 *lengthptr += delta;
6182 }
6183
6184 /* This is compiling for real. If there is a set first byte for
6185 the group, and we have not yet set a "required byte", set it. Make
6186 sure there is enough workspace for copying forward references before
6187 doing the copy. */
6188
6189 else
6190 {
6191 if (groupsetfirstchar && reqcharflags < 0)
6192 {
6193 reqchar = firstchar;
6194 reqcharflags = firstcharflags;
6195 }
6196
6197 for (i = 1; i < repeat_min; i++)
6198 {
6199 pcre_uchar *hc;
6200 size_t this_hwm_offset = cd->hwm - cd->start_workspace;
6201 memcpy(code, previous, IN_UCHARS(len));
6202
6203 while (cd->hwm > cd->start_workspace + cd->workspace_size -
6204 WORK_SIZE_SAFETY_MARGIN -
6205 (this_hwm_offset - base_hwm_offset))
6206 {
6207 *errorcodeptr = expand_workspace(cd);
6208 if (*errorcodeptr != 0) goto FAILED;
6209 }
6210
6211 for (hc = (pcre_uchar *)cd->start_workspace + base_hwm_offset;
6212 hc < (pcre_uchar *)cd->start_workspace + this_hwm_offset;
6213 hc += LINK_SIZE)
6214 {
6215 PUT(cd->hwm, 0, GET(hc, 0) + len);
6216 cd->hwm += LINK_SIZE;
6217 }
6218 base_hwm_offset = this_hwm_offset;
6219 code += len;
6220 }
6221 }
6222 }
6223
6224 if (repeat_max > 0) repeat_max -= repeat_min;
6225 }
6226
6227 /* This code is common to both the zero and non-zero minimum cases. If
6228 the maximum is limited, it replicates the group in a nested fashion,
6229 remembering the bracket starts on a stack. In the case of a zero minimum,
6230 the first one was set up above. In all cases the repeat_max now specifies
6231 the number of additional copies needed. Again, we must remember to
6232 replicate entries on the forward reference list. */
6233
6234 if (repeat_max >= 0)
6235 {
6236 /* In the pre-compile phase, we don't actually do the replication. We
6237 just adjust the length as if we had. For each repetition we must add 1
6238 to the length for BRAZERO and for all but the last repetition we must
6239 add 2 + 2*LINKSIZE to allow for the nesting that occurs. Do some
6240 paranoid checks to avoid integer overflow. The INT64_OR_DOUBLE type is
6241 a 64-bit integer type when available, otherwise double. */
6242
6243 if (lengthptr != NULL && repeat_max > 0)
6244 {
6245 int delta = repeat_max * (length_prevgroup + 1 + 2 + 2*LINK_SIZE) -
6246 2 - 2*LINK_SIZE; /* Last one doesn't nest */
6247 if ((INT64_OR_DOUBLE)repeat_max *
6248 (INT64_OR_DOUBLE)(length_prevgroup + 1 + 2 + 2*LINK_SIZE)
6249 > (INT64_OR_DOUBLE)INT_MAX ||
6250 OFLOW_MAX - *lengthptr < delta)
6251 {
6252 *errorcodeptr = ERR20;
6253 goto FAILED;
6254 }
6255 *lengthptr += delta;
6256 }
6257
6258 /* This is compiling for real */
6259
6260 else for (i = repeat_max - 1; i >= 0; i--)
6261 {
6262 pcre_uchar *hc;
6263 size_t this_hwm_offset = cd->hwm - cd->start_workspace;
6264
6265 *code++ = OP_BRAZERO + repeat_type;
6266
6267 /* All but the final copy start a new nesting, maintaining the
6268 chain of brackets outstanding. */
6269
6270 if (i != 0)
6271 {
6272 int offset;
6273 *code++ = OP_BRA;
6274 offset = (bralink == NULL)? 0 : (int)(code - bralink);
6275 bralink = code;
6276 PUTINC(code, 0, offset);
6277 }
6278
6279 memcpy(code, previous, IN_UCHARS(len));
6280
6281 /* Ensure there is enough workspace for forward references before
6282 copying them. */
6283
6284 while (cd->hwm > cd->start_workspace + cd->workspace_size -
6285 WORK_SIZE_SAFETY_MARGIN -
6286 (this_hwm_offset - base_hwm_offset))
6287 {
6288 *errorcodeptr = expand_workspace(cd);
6289 if (*errorcodeptr != 0) goto FAILED;
6290 }
6291
6292 for (hc = (pcre_uchar *)cd->start_workspace + base_hwm_offset;
6293 hc < (pcre_uchar *)cd->start_workspace + this_hwm_offset;
6294 hc += LINK_SIZE)
6295 {
6296 PUT(cd->hwm, 0, GET(hc, 0) + len + ((i != 0)? 2+LINK_SIZE : 1));
6297 cd->hwm += LINK_SIZE;
6298 }
6299 base_hwm_offset = this_hwm_offset;
6300 code += len;
6301 }
6302
6303 /* Now chain through the pending brackets, and fill in their length
6304 fields (which are holding the chain links pro tem). */
6305
6306 while (bralink != NULL)
6307 {
6308 int oldlinkoffset;
6309 int offset = (int)(code - bralink + 1);
6310 pcre_uchar *bra = code - offset;
6311 oldlinkoffset = GET(bra, 1);
6312 bralink = (oldlinkoffset == 0)? NULL : bralink - oldlinkoffset;
6313 *code++ = OP_KET;
6314 PUTINC(code, 0, offset);
6315 PUT(bra, 1, offset);
6316 }
6317 }
6318
6319 /* If the maximum is unlimited, set a repeater in the final copy. For
6320 ONCE brackets, that's all we need to do. However, possessively repeated
6321 ONCE brackets can be converted into non-capturing brackets, as the
6322 behaviour of (?:xx)++ is the same as (?>xx)++ and this saves having to
6323 deal with possessive ONCEs specially.
6324
6325 Otherwise, when we are doing the actual compile phase, check to see
6326 whether this group is one that could match an empty string. If so,
6327 convert the initial operator to the S form (e.g. OP_BRA -> OP_SBRA) so
6328 that runtime checking can be done. [This check is also applied to ONCE
6329 groups at runtime, but in a different way.]
6330
6331 Then, if the quantifier was possessive and the bracket is not a
6332 conditional, we convert the BRA code to the POS form, and the KET code to
6333 KETRPOS. (It turns out to be convenient at runtime to detect this kind of
6334 subpattern at both the start and at the end.) The use of special opcodes
6335 makes it possible to reduce greatly the stack usage in pcre_exec(). If
6336 the group is preceded by OP_BRAZERO, convert this to OP_BRAPOSZERO.
6337
6338 Then, if the minimum number of matches is 1 or 0, cancel the possessive
6339 flag so that the default action below, of wrapping everything inside
6340 atomic brackets, does not happen. When the minimum is greater than 1,
6341 there will be earlier copies of the group, and so we still have to wrap
6342 the whole thing. */
6343
6344 else
6345 {
6346 pcre_uchar *ketcode = code - 1 - LINK_SIZE;
6347 pcre_uchar *bracode = ketcode - GET(ketcode, 1);
6348
6349 /* Convert possessive ONCE brackets to non-capturing */
6350
6351 if ((*bracode == OP_ONCE || *bracode == OP_ONCE_NC) &&
6352 possessive_quantifier) *bracode = OP_BRA;
6353
6354 /* For non-possessive ONCE brackets, all we need to do is to
6355 set the KET. */
6356
6357 if (*bracode == OP_ONCE || *bracode == OP_ONCE_NC)
6358 *ketcode = OP_KETRMAX + repeat_type;
6359
6360 /* Handle non-ONCE brackets and possessive ONCEs (which have been
6361 converted to non-capturing above). */
6362
6363 else
6364 {
6365 /* In the compile phase, check for empty string matching. */
6366
6367 if (lengthptr == NULL)
6368 {
6369 pcre_uchar *scode = bracode;
6370 do
6371 {
6372 if (could_be_empty_branch(scode, ketcode, utf, cd, NULL))
6373 {
6374 *bracode += OP_SBRA - OP_BRA;
6375 break;
6376 }
6377 scode += GET(scode, 1);
6378 }
6379 while (*scode == OP_ALT);
6380 }
6381
6382 /* A conditional group with only one branch has an implicit empty
6383 alternative branch. */
6384
6385 if (*bracode == OP_COND && bracode[GET(bracode,1)] != OP_ALT)
6386 *bracode = OP_SCOND;
6387
6388 /* Handle possessive quantifiers. */
6389
6390 if (possessive_quantifier)
6391 {
6392 /* For COND brackets, we wrap the whole thing in a possessively
6393 repeated non-capturing bracket, because we have not invented POS
6394 versions of the COND opcodes. Because we are moving code along, we
6395 must ensure that any pending recursive references are updated. */
6396
6397 if (*bracode == OP_COND || *bracode == OP_SCOND)
6398 {
6399 int nlen = (int)(code - bracode);
6400 *code = OP_END;
6401 adjust_recurse(bracode, 1 + LINK_SIZE, utf, cd, item_hwm_offset);
6402 memmove(bracode + 1 + LINK_SIZE, bracode, IN_UCHARS(nlen));
6403 code += 1 + LINK_SIZE;
6404 nlen += 1 + LINK_SIZE;
6405 *bracode = (*bracode == OP_COND)? OP_BRAPOS : OP_SBRAPOS;
6406 *code++ = OP_KETRPOS;
6407 PUTINC(code, 0, nlen);
6408 PUT(bracode, 1, nlen);
6409 }
6410
6411 /* For non-COND brackets, we modify the BRA code and use KETRPOS. */
6412
6413 else
6414 {
6415 *bracode += 1; /* Switch to xxxPOS opcodes */
6416 *ketcode = OP_KETRPOS;
6417 }
6418
6419 /* If the minimum is zero, mark it as possessive, then unset the
6420 possessive flag when the minimum is 0 or 1. */
6421
6422 if (brazeroptr != NULL) *brazeroptr = OP_BRAPOSZERO;
6423 if (repeat_min < 2) possessive_quantifier = FALSE;
6424 }
6425
6426 /* Non-possessive quantifier */
6427
6428 else *ketcode = OP_KETRMAX + repeat_type;
6429 }
6430 }
6431 }
6432
6433 /* If previous is OP_FAIL, it was generated by an empty class [] in
6434 JavaScript mode. The other ways in which OP_FAIL can be generated, that is
6435 by (*FAIL) or (?!) set previous to NULL, which gives a "nothing to repeat"
6436 error above. We can just ignore the repeat in JS case. */
6437
6438 else if (*previous == OP_FAIL) goto END_REPEAT;
6439
6440 /* Else there's some kind of shambles */
6441
6442 else
6443 {
6444 *errorcodeptr = ERR11;
6445 goto FAILED;
6446 }
6447
6448 /* If the character following a repeat is '+', possessive_quantifier is
6449 TRUE. For some opcodes, there are special alternative opcodes for this
6450 case. For anything else, we wrap the entire repeated item inside OP_ONCE
6451 brackets. Logically, the '+' notation is just syntactic sugar, taken from
6452 Sun's Java package, but the special opcodes can optimize it.
6453
6454 Some (but not all) possessively repeated subpatterns have already been
6455 completely handled in the code just above. For them, possessive_quantifier
6456 is always FALSE at this stage. Note that the repeated item starts at
6457 tempcode, not at previous, which might be the first part of a string whose
6458 (former) last char we repeated. */
6459
6460 if (possessive_quantifier)
6461 {
6462 int len;
6463
6464 /* Possessifying an EXACT quantifier has no effect, so we can ignore it.
6465 However, QUERY, STAR, or UPTO may follow (for quantifiers such as {5,6},
6466 {5,}, or {5,10}). We skip over an EXACT item; if the length of what
6467 remains is greater than zero, there's a further opcode that can be
6468 handled. If not, do nothing, leaving the EXACT alone. */
6469
6470 switch(*tempcode)
6471 {
6472 case OP_TYPEEXACT:
6473 tempcode += PRIV(OP_lengths)[*tempcode] +
6474 ((tempcode[1 + IMM2_SIZE] == OP_PROP
6475 || tempcode[1 + IMM2_SIZE] == OP_NOTPROP)? 2 : 0);
6476 break;
6477
6478 /* CHAR opcodes are used for exacts whose count is 1. */
6479
6480 case OP_CHAR:
6481 case OP_CHARI:
6482 case OP_NOT:
6483 case OP_NOTI:
6484 case OP_EXACT:
6485 case OP_EXACTI:
6486 case OP_NOTEXACT:
6487 case OP_NOTEXACTI:
6488 tempcode += PRIV(OP_lengths)[*tempcode];
6489 #ifdef SUPPORT_UTF
6490 if (utf && HAS_EXTRALEN(tempcode[-1]))
6491 tempcode += GET_EXTRALEN(tempcode[-1]);
6492 #endif
6493 break;
6494
6495 /* For the class opcodes, the repeat operator appears at the end;
6496 adjust tempcode to point to it. */
6497
6498 case OP_CLASS:
6499 case OP_NCLASS:
6500 tempcode += 1 + 32/sizeof(pcre_uchar);
6501 break;
6502
6503 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
6504 case OP_XCLASS:
6505 tempcode += GET(tempcode, 1);
6506 break;
6507 #endif
6508 }
6509
6510 /* If tempcode is equal to code (which points to the end of the repeated
6511 item), it means we have skipped an EXACT item but there is no following
6512 QUERY, STAR, or UPTO; the value of len will be 0, and we do nothing. In
6513 all other cases, tempcode will be pointing to the repeat opcode, and will
6514 be less than code, so the value of len will be greater than 0. */
6515
6516 len = (int)(code - tempcode);
6517 if (len > 0)
6518 {
6519 unsigned int repcode = *tempcode;
6520
6521 /* There is a table for possessifying opcodes, all of which are less
6522 than OP_CALLOUT. A zero entry means there is no possessified version.
6523 */
6524
6525 if (repcode < OP_CALLOUT && opcode_possessify[repcode] > 0)
6526 *tempcode = opcode_possessify[repcode];
6527
6528 /* For opcode without a special possessified version, wrap the item in
6529 ONCE brackets. Because we are moving code along, we must ensure that any
6530 pending recursive references are updated. */
6531
6532 else
6533 {
6534 *code = OP_END;
6535 adjust_recurse(tempcode, 1 + LINK_SIZE, utf, cd, item_hwm_offset);
6536 memmove(tempcode + 1 + LINK_SIZE, tempcode, IN_UCHARS(len));
6537 code += 1 + LINK_SIZE;
6538 len += 1 + LINK_SIZE;
6539 tempcode[0] = OP_ONCE;
6540 *code++ = OP_KET;
6541 PUTINC(code, 0, len);
6542 PUT(tempcode, 1, len);
6543 }
6544 }
6545
6546 #ifdef NEVER
6547 if (len > 0) switch (*tempcode)
6548 {
6549 case OP_STAR: *tempcode = OP_POSSTAR; break;
6550 case OP_PLUS: *tempcode = OP_POSPLUS; break;
6551 case OP_QUERY: *tempcode = OP_POSQUERY; break;
6552 case OP_UPTO: *tempcode = OP_POSUPTO; break;
6553
6554 case OP_STARI: *tempcode = OP_POSSTARI; break;
6555 case OP_PLUSI: *tempcode = OP_POSPLUSI; break;
6556 case OP_QUERYI: *tempcode = OP_POSQUERYI; break;
6557 case OP_UPTOI: *tempcode = OP_POSUPTOI; break;
6558
6559 case OP_NOTSTAR: *tempcode = OP_NOTPOSSTAR; break;
6560 case OP_NOTPLUS: *tempcode = OP_NOTPOSPLUS; break;
6561 case OP_NOTQUERY: *tempcode = OP_NOTPOSQUERY; break;
6562 case OP_NOTUPTO: *tempcode = OP_NOTPOSUPTO; break;
6563
6564 case OP_NOTSTARI: *tempcode = OP_NOTPOSSTARI; break;
6565 case OP_NOTPLUSI: *tempcode = OP_NOTPOSPLUSI; break;
6566 case OP_NOTQUERYI: *tempcode = OP_NOTPOSQUERYI; break;
6567 case OP_NOTUPTOI: *tempcode = OP_NOTPOSUPTOI; break;
6568
6569 case OP_TYPESTAR: *tempcode = OP_TYPEPOSSTAR; break;
6570 case OP_TYPEPLUS: *tempcode = OP_TYPEPOSPLUS; break;
6571 case OP_TYPEQUERY: *tempcode = OP_TYPEPOSQUERY; break;
6572 case OP_TYPEUPTO: *tempcode = OP_TYPEPOSUPTO; break;
6573
6574 case OP_CRSTAR: *tempcode = OP_CRPOSSTAR; break;
6575 case OP_CRPLUS: *tempcode = OP_CRPOSPLUS; break;
6576 case OP_CRQUERY: *tempcode = OP_CRPOSQUERY; break;
6577 case OP_CRRANGE: *tempcode = OP_CRPOSRANGE; break;
6578
6579 /* Because we are moving code along, we must ensure that any
6580 pending recursive references are updated. */
6581
6582 default:
6583 *code = OP_END;
6584 adjust_recurse(tempcode, 1 + LINK_SIZE, utf, cd, item_hwm_offset);
6585 memmove(tempcode + 1 + LINK_SIZE, tempcode, IN_UCHARS(len));
6586 code += 1 + LINK_SIZE;
6587 len += 1 + LINK_SIZE;
6588 tempcode[0] = OP_ONCE;
6589 *code++ = OP_KET;
6590 PUTINC(code, 0, len);
6591 PUT(tempcode, 1, len);
6592 break;
6593 }
6594 #endif
6595 }
6596
6597 /* In all case we no longer have a previous item. We also set the
6598 "follows varying string" flag for subsequently encountered reqchars if
6599 it isn't already set and we have just passed a varying length item. */
6600
6601 END_REPEAT:
6602 previous = NULL;
6603 cd->req_varyopt |= reqvary;
6604 break;
6605
6606
6607 /* ===================================================================*/
6608 /* Start of nested parenthesized sub-expression, or comment or lookahead or
6609 lookbehind or option setting or condition or all the other extended
6610 parenthesis forms. */
6611
6612 case CHAR_LEFT_PARENTHESIS:
6613 ptr++;
6614
6615 /* Now deal with various "verbs" that can be introduced by '*'. */
6616
6617 if (ptr[0] == CHAR_ASTERISK && (ptr[1] == ':'
6618 || (MAX_255(ptr[1]) && ((cd->ctypes[ptr[1]] & ctype_letter) != 0))))
6619 {
6620 int i, namelen;
6621 int arglen = 0;
6622 const char *vn = verbnames;
6623 const pcre_uchar *name = ptr + 1;
6624 const pcre_uchar *arg = NULL;
6625 previous = NULL;
6626 ptr++;
6627 while (MAX_255(*ptr) && (cd->ctypes[*ptr] & ctype_letter) != 0) ptr++;
6628 namelen = (int)(ptr - name);
6629
6630 /* It appears that Perl allows any characters whatsoever, other than
6631 a closing parenthesis, to appear in arguments, so we no longer insist on
6632 letters, digits, and underscores. */
6633
6634 if (*ptr == CHAR_COLON)
6635 {
6636 arg = ++ptr;
6637 while (*ptr != CHAR_NULL && *ptr != CHAR_RIGHT_PARENTHESIS) ptr++;
6638 arglen = (int)(ptr - arg);
6639 if ((unsigned int)arglen > MAX_MARK)
6640 {
6641 *errorcodeptr = ERR75;
6642 goto FAILED;
6643 }
6644 }
6645
6646 if (*ptr != CHAR_RIGHT_PARENTHESIS)
6647 {
6648 *errorcodeptr = ERR60;
6649 goto FAILED;
6650 }
6651
6652 /* Scan the table of verb names */
6653
6654 for (i = 0; i < verbcount; i++)
6655 {
6656 if (namelen == verbs[i].len &&
6657 STRNCMP_UC_C8(name, vn, namelen) == 0)
6658 {
6659 int setverb;
6660
6661 /* Check for open captures before ACCEPT and convert it to
6662 ASSERT_ACCEPT if in an assertion. */
6663
6664 if (verbs[i].op == OP_ACCEPT)
6665 {
6666 open_capitem *oc;
6667 if (arglen != 0)
6668 {
6669 *errorcodeptr = ERR59;
6670 goto FAILED;
6671 }
6672 cd->had_accept = TRUE;
6673 for (oc = cd->open_caps; oc != NULL; oc = oc->next)
6674 {
6675 if (lengthptr != NULL)
6676 {
6677 #ifdef COMPILE_PCRE8
6678 *lengthptr += 1 + IMM2_SIZE;
6679 #elif defined COMPILE_PCRE16
6680 *lengthptr += 2 + IMM2_SIZE;
6681 #elif defined COMPILE_PCRE32
6682 *lengthptr += 4 + IMM2_SIZE;
6683 #endif
6684 }
6685 else
6686 {
6687 *code++ = OP_CLOSE;
6688 PUT2INC(code, 0, oc->number);
6689 }
6690 }
6691 setverb = *code++ =
6692 (cd->assert_depth > 0)? OP_ASSERT_ACCEPT : OP_ACCEPT;
6693
6694 /* Do not set firstchar after *ACCEPT */
6695 if (firstcharflags == REQ_UNSET) firstcharflags = REQ_NONE;
6696 }
6697
6698 /* Handle other cases with/without an argument */
6699
6700 else if (arglen == 0)
6701 {
6702 if (verbs[i].op < 0) /* Argument is mandatory */
6703 {
6704 *errorcodeptr = ERR66;
6705 goto FAILED;
6706 }
6707 setverb = *code++ = verbs[i].op;
6708 }
6709
6710 else
6711 {
6712 if (verbs[i].op_arg < 0) /* Argument is forbidden */
6713 {
6714 *errorcodeptr = ERR59;
6715 goto FAILED;
6716 }
6717 setverb = *code++ = verbs[i].op_arg;
6718 if (lengthptr != NULL) /* In pass 1 just add in the length */
6719 { /* to avoid potential workspace */
6720 *lengthptr += arglen; /* overflow. */
6721 *code++ = 0;
6722 }
6723 else
6724 {
6725 *code++ = arglen;
6726 memcpy(code, arg, IN_UCHARS(arglen));
6727 code += arglen;
6728 }
6729 *code++ = 0;
6730 }
6731
6732 switch (setverb)
6733 {
6734 case OP_THEN:
6735 case OP_THEN_ARG:
6736 cd->external_flags |= PCRE_HASTHEN;
6737 break;
6738
6739 case OP_PRUNE:
6740 case OP_PRUNE_ARG:
6741 case OP_SKIP:
6742 case OP_SKIP_ARG:
6743 cd->had_pruneorskip = TRUE;
6744 break;
6745 }
6746
6747 break; /* Found verb, exit loop */
6748 }
6749
6750 vn += verbs[i].len + 1;
6751 }
6752
6753 if (i < verbcount) continue; /* Successfully handled a verb */
6754 *errorcodeptr = ERR60; /* Verb not recognized */
6755 goto FAILED;
6756 }
6757
6758 /* Initialize for "real" parentheses */
6759
6760 newoptions = options;
6761 skipbytes = 0;
6762 bravalue = OP_CBRA;
6763 item_hwm_offset = cd->hwm - cd->start_workspace;
6764 reset_bracount = FALSE;
6765
6766 /* Deal with the extended parentheses; all are introduced by '?', and the
6767 appearance of any of them means that this is not a capturing group. */
6768
6769 if (*ptr == CHAR_QUESTION_MARK)
6770 {
6771 int i, set, unset, namelen;
6772 int *optset;
6773 const pcre_uchar *name;
6774 pcre_uchar *slot;
6775
6776 switch (*(++ptr))
6777 {
6778 /* ------------------------------------------------------------ */
6779 case CHAR_VERTICAL_LINE: /* Reset capture count for each branch */
6780 reset_bracount = TRUE;
6781 cd->dupgroups = TRUE; /* Record (?| encountered */
6782 /* Fall through */
6783
6784 /* ------------------------------------------------------------ */
6785 case CHAR_COLON: /* Non-capturing bracket */
6786 bravalue = OP_BRA;
6787 ptr++;
6788 break;
6789
6790
6791 /* ------------------------------------------------------------ */
6792 case CHAR_LEFT_PARENTHESIS:
6793 bravalue = OP_COND; /* Conditional group */
6794 tempptr = ptr;
6795
6796 /* A condition can be an assertion, a number (referring to a numbered
6797 group's having been set), a name (referring to a named group), or 'R',
6798 referring to recursion. R<digits> and R&name are also permitted for
6799 recursion tests.
6800
6801 There are ways of testing a named group: (?(name)) is used by Python;
6802 Perl 5.10 onwards uses (?(<name>) or (?('name')).
6803
6804 There is one unfortunate ambiguity, caused by history. 'R' can be the
6805 recursive thing or the name 'R' (and similarly for 'R' followed by
6806 digits). We look for a name first; if not found, we try the other case.
6807
6808 For compatibility with auto-callouts, we allow a callout to be
6809 specified before a condition that is an assertion. First, check for the
6810 syntax of a callout; if found, adjust the temporary pointer that is
6811 used to check for an assertion condition. That's all that is needed! */
6812
6813 if (ptr[1] == CHAR_QUESTION_MARK && ptr[2] == CHAR_C)
6814 {
6815 for (i = 3;; i++) if (!IS_DIGIT(ptr[i])) break;
6816 if (ptr[i] == CHAR_RIGHT_PARENTHESIS)
6817 tempptr += i + 1;
6818
6819 /* tempptr should now be pointing to the opening parenthesis of the
6820 assertion condition. */
6821
6822 if (*tempptr != CHAR_LEFT_PARENTHESIS)
6823 {
6824 *errorcodeptr = ERR28;
6825 goto FAILED;
6826 }
6827 }
6828
6829 /* For conditions that are assertions, check the syntax, and then exit
6830 the switch. This will take control down to where bracketed groups,
6831 including assertions, are processed. */
6832
6833 if (tempptr[1] == CHAR_QUESTION_MARK &&
6834 (tempptr[2] == CHAR_EQUALS_SIGN ||
6835 tempptr[2] == CHAR_EXCLAMATION_MARK ||
6836 (tempptr[2] == CHAR_LESS_THAN_SIGN &&
6837 (tempptr[3] == CHAR_EQUALS_SIGN ||
6838 tempptr[3] == CHAR_EXCLAMATION_MARK))))
6839 {
6840 cd->iscondassert = TRUE;
6841 break;
6842 }
6843
6844 /* Other conditions use OP_CREF/OP_DNCREF/OP_RREF/OP_DNRREF, and all
6845 need to skip at least 1+IMM2_SIZE bytes at the start of the group. */
6846
6847 code[1+LINK_SIZE] = OP_CREF;
6848 skipbytes = 1+IMM2_SIZE;
6849 refsign = -1; /* => not a number */
6850 namelen = -1; /* => not a name; must set to avoid warning */
6851 name = NULL; /* Always set to avoid warning */
6852 recno = 0; /* Always set to avoid warning */
6853
6854 /* Check for a test for recursion in a named group. */
6855
6856 ptr++;
6857 if (*ptr == CHAR_R && ptr[1] == CHAR_AMPERSAND)
6858 {
6859 terminator = -1;
6860 ptr += 2;
6861 code[1+LINK_SIZE] = OP_RREF; /* Change the type of test */
6862 }
6863
6864 /* Check for a test for a named group's having been set, using the Perl
6865 syntax (?(<name>) or (?('name'), and also allow for the original PCRE
6866 syntax of (?(name) or for (?(+n), (?(-n), and just (?(n). */
6867
6868 else if (*ptr == CHAR_LESS_THAN_SIGN)
6869 {
6870 terminator = CHAR_GREATER_THAN_SIGN;
6871 ptr++;
6872 }
6873 else if (*ptr == CHAR_APOSTROPHE)
6874 {
6875 terminator = CHAR_APOSTROPHE;
6876 ptr++;
6877 }
6878 else
6879 {
6880 terminator = CHAR_NULL;
6881 if (*ptr == CHAR_MINUS || *ptr == CHAR_PLUS) refsign = *ptr++;
6882 else if (IS_DIGIT(*ptr)) refsign = 0;
6883 }
6884
6885 /* Handle a number */
6886
6887 if (refsign >= 0)
6888 {
6889 while (IS_DIGIT(*ptr))
6890 {
6891 if (recno > INT_MAX / 10 - 1) /* Integer overflow */
6892 {
6893 while (IS_DIGIT(*ptr)) ptr++;
6894 *errorcodeptr = ERR61;
6895 goto FAILED;
6896 }
6897 recno = recno * 10 + (int)(*ptr - CHAR_0);
6898 ptr++;
6899 }
6900 }
6901
6902 /* Otherwise we expect to read a name; anything else is an error. When
6903 a name is one of a number of duplicates, a different opcode is used and
6904 it needs more memory. Unfortunately we cannot tell whether a name is a
6905 duplicate in the first pass, so we have to allow for more memory. */
6906
6907 else
6908 {
6909 if (IS_DIGIT(*ptr))
6910 {
6911 *errorcodeptr = ERR84;
6912 goto FAILED;
6913 }
6914 if (!MAX_255(*ptr) || (cd->ctypes[*ptr] & ctype_word) == 0)
6915 {
6916 *errorcodeptr = ERR28; /* Assertion expected */
6917 goto FAILED;
6918 }
6919 name = ptr++;
6920 while (MAX_255(*ptr) && (cd->ctypes[*ptr] & ctype_word) != 0)
6921 {
6922 ptr++;
6923 }
6924 namelen = (int)(ptr - name);
6925 if (lengthptr != NULL) skipbytes += IMM2_SIZE;
6926 }
6927
6928 /* Check the terminator */
6929
6930 if ((terminator > 0 && *ptr++ != (pcre_uchar)terminator) ||
6931 *ptr++ != CHAR_RIGHT_PARENTHESIS)
6932 {
6933 ptr--; /* Error offset */
6934 *errorcodeptr = ERR26; /* Malformed number or name */
6935 goto FAILED;
6936 }
6937
6938 /* Do no further checking in the pre-compile phase. */
6939
6940 if (lengthptr != NULL) break;
6941
6942 /* In the real compile we do the work of looking for the actual
6943 reference. If refsign is not negative, it means we have a number in
6944 recno. */
6945
6946 if (refsign >= 0)
6947 {
6948 if (recno <= 0)
6949 {
6950 *errorcodeptr = ERR35;
6951 goto FAILED;
6952 }
6953 if (refsign != 0) recno = (refsign == CHAR_MINUS)?
6954 cd->bracount - recno + 1 : recno + cd->bracount;
6955 if (recno <= 0 || recno > cd->final_bracount)
6956 {
6957 *errorcodeptr = ERR15;
6958 goto FAILED;
6959 }
6960 PUT2(code, 2+LINK_SIZE, recno);
6961 if (recno > cd->top_backref) cd->top_backref = recno;
6962 break;
6963 }
6964
6965 /* Otherwise look for the name. */
6966
6967 slot = cd->name_table;
6968 for (i = 0; i < cd->names_found; i++)
6969 {
6970 if (STRNCMP_UC_UC(name, slot+IMM2_SIZE, namelen) == 0 &&
6971 slot[IMM2_SIZE+namelen] == 0) break;
6972 slot += cd->name_entry_size;
6973 }
6974
6975 /* Found the named subpattern. If the name is duplicated, add one to
6976 the opcode to change CREF/RREF into DNCREF/DNRREF and insert
6977 appropriate data values. Otherwise, just insert the unique subpattern
6978 number. */
6979
6980 if (i < cd->names_found)
6981 {
6982 int offset = i++;
6983 int count = 1;
6984 recno = GET2(slot, 0); /* Number from first found */
6985 if (recno > cd->top_backref) cd->top_backref = recno;
6986 for (; i < cd->names_found; i++)
6987 {
6988 slot += cd->name_entry_size;
6989 if (STRNCMP_UC_UC(name, slot+IMM2_SIZE, namelen) != 0 ||
6990 (slot+IMM2_SIZE)[namelen] != 0) break;
6991 count++;
6992 }
6993
6994 if (count > 1)
6995 {
6996 PUT2(code, 2+LINK_SIZE, offset);
6997 PUT2(code, 2+LINK_SIZE+IMM2_SIZE, count);
6998 skipbytes += IMM2_SIZE;
6999 code[1+LINK_SIZE]++;
7000 }
7001 else /* Not a duplicated name */
7002 {
7003 PUT2(code, 2+LINK_SIZE, recno);
7004 }
7005 }
7006
7007 /* If terminator == CHAR_NULL it means that the name followed directly
7008 after the opening parenthesis [e.g. (?(abc)...] and in this case there
7009 are some further alternatives to try. For the cases where terminator !=
7010 CHAR_NULL [things like (?(<name>... or (?('name')... or (?(R&name)... ]
7011 we have now checked all the possibilities, so give an error. */
7012
7013 else if (terminator != CHAR_NULL)
7014 {
7015 *errorcodeptr = ERR15;
7016 goto FAILED;
7017 }
7018
7019 /* Check for (?(R) for recursion. Allow digits after R to specify a
7020 specific group number. */
7021
7022 else if (*name == CHAR_R)
7023 {
7024 recno = 0;
7025 for (i = 1; i < namelen; i++)
7026 {
7027 if (!IS_DIGIT(name[i]))
7028 {
7029 *errorcodeptr = ERR15;
7030 goto FAILED;
7031 }
7032 if (recno > INT_MAX / 10 - 1) /* Integer overflow */
7033 {
7034 *errorcodeptr = ERR61;
7035 goto FAILED;
7036 }
7037 recno = recno * 10 + name[i] - CHAR_0;
7038 }
7039 if (recno == 0) recno = RREF_ANY;
7040 code[1+LINK_SIZE] = OP_RREF; /* Change test type */
7041 PUT2(code, 2+LINK_SIZE, recno);
7042 }
7043
7044 /* Similarly, check for the (?(DEFINE) "condition", which is always
7045 false. */
7046
7047 else if (namelen == 6 && STRNCMP_UC_C8(name, STRING_DEFINE, 6) == 0)
7048 {
7049 code[1+LINK_SIZE] = OP_DEF;
7050 skipbytes = 1;
7051 }
7052
7053 /* Reference to an unidentified subpattern. */
7054
7055 else
7056 {
7057 *errorcodeptr = ERR15;
7058 goto FAILED;
7059 }
7060 break;
7061
7062
7063 /* ------------------------------------------------------------ */
7064 case CHAR_EQUALS_SIGN: /* Positive lookahead */
7065 bravalue = OP_ASSERT;
7066 cd->assert_depth += 1;
7067 ptr++;
7068 break;
7069
7070 /* Optimize (?!) to (*FAIL) unless it is quantified - which is a weird
7071 thing to do, but Perl allows all assertions to be quantified, and when
7072 they contain capturing parentheses there may be a potential use for
7073 this feature. Not that that applies to a quantified (?!) but we allow
7074 it for uniformity. */
7075
7076 /* ------------------------------------------------------------ */
7077 case CHAR_EXCLAMATION_MARK: /* Negative lookahead */
7078 ptr++;
7079 if (*ptr == CHAR_RIGHT_PARENTHESIS && ptr[1] != CHAR_ASTERISK &&
7080 ptr[1] != CHAR_PLUS && ptr[1] != CHAR_QUESTION_MARK &&
7081 (ptr[1] != CHAR_LEFT_CURLY_BRACKET || !is_counted_repeat(ptr+2)))
7082 {
7083 *code++ = OP_FAIL;
7084 previous = NULL;
7085 continue;
7086 }
7087 bravalue = OP_ASSERT_NOT;
7088 cd->assert_depth += 1;
7089 break;
7090
7091
7092 /* ------------------------------------------------------------ */
7093 case CHAR_LESS_THAN_SIGN: /* Lookbehind or named define */
7094 switch (ptr[1])
7095 {
7096 case CHAR_EQUALS_SIGN: /* Positive lookbehind */
7097 bravalue = OP_ASSERTBACK;
7098 cd->assert_depth += 1;
7099 ptr += 2;
7100 break;
7101
7102 case CHAR_EXCLAMATION_MARK: /* Negative lookbehind */
7103 bravalue = OP_ASSERTBACK_NOT;
7104 cd->assert_depth += 1;
7105 ptr += 2;
7106 break;
7107
7108 default: /* Could be name define, else bad */
7109 if (MAX_255(ptr[1]) && (cd->ctypes[ptr[1]] & ctype_word) != 0)
7110 goto DEFINE_NAME;
7111 ptr++; /* Correct offset for error */
7112 *errorcodeptr = ERR24;
7113 goto FAILED;
7114 }
7115 break;
7116
7117
7118 /* ------------------------------------------------------------ */
7119 case CHAR_GREATER_THAN_SIGN: /* One-time brackets */
7120 bravalue = OP_ONCE;
7121 ptr++;
7122 break;
7123
7124
7125 /* ------------------------------------------------------------ */
7126 case CHAR_C: /* Callout - may be followed by digits; */
7127 previous_callout = code; /* Save for later completion */
7128 after_manual_callout = 1; /* Skip one item before completing */
7129 *code++ = OP_CALLOUT;
7130 {
7131 int n = 0;
7132 ptr++;
7133 while(IS_DIGIT(*ptr))
7134 {
7135 n = n * 10 + *ptr++ - CHAR_0;
7136 if (n > 255)
7137 {
7138 *errorcodeptr = ERR38;
7139 goto FAILED;
7140 }
7141 }
7142 if (*ptr != CHAR_RIGHT_PARENTHESIS)
7143 {
7144 *errorcodeptr = ERR39;
7145 goto FAILED;
7146 }
7147 *code++ = n;
7148 PUT(code, 0, (int)(ptr - cd->start_pattern + 1)); /* Pattern offset */
7149 PUT(code, LINK_SIZE, 0); /* Default length */
7150 code += 2 * LINK_SIZE;
7151 }
7152 previous = NULL;
7153 continue;
7154
7155
7156 /* ------------------------------------------------------------ */
7157 case CHAR_P: /* Python-style named subpattern handling */
7158 if (*(++ptr) == CHAR_EQUALS_SIGN ||
7159 *ptr == CHAR_GREATER_THAN_SIGN) /* Reference or recursion */
7160 {
7161 is_recurse = *ptr == CHAR_GREATER_THAN_SIGN;
7162 terminator = CHAR_RIGHT_PARENTHESIS;
7163 goto NAMED_REF_OR_RECURSE;
7164 }
7165 else if (*ptr != CHAR_LESS_THAN_SIGN) /* Test for Python-style defn */
7166 {
7167 *errorcodeptr = ERR41;
7168 goto FAILED;
7169 }
7170 /* Fall through to handle (?P< as (?< is handled */
7171
7172
7173 /* ------------------------------------------------------------ */
7174 DEFINE_NAME: /* Come here from (?< handling */
7175 case CHAR_APOSTROPHE:
7176 terminator = (*ptr == CHAR_LESS_THAN_SIGN)?
7177 CHAR_GREATER_THAN_SIGN : CHAR_APOSTROPHE;
7178 name = ++ptr;
7179 if (IS_DIGIT(*ptr))
7180 {
7181 *errorcodeptr = ERR84; /* Group name must start with non-digit */
7182 goto FAILED;
7183 }
7184 while (MAX_255(*ptr) && (cd->ctypes[*ptr] & ctype_word) != 0) ptr++;
7185 namelen = (int)(ptr - name);
7186
7187 /* In the pre-compile phase, do a syntax check, remember the longest
7188 name, and then remember the group in a vector, expanding it if
7189 necessary. Duplicates for the same number are skipped; other duplicates
7190 are checked for validity. In the actual compile, there is nothing to
7191 do. */
7192
7193 if (lengthptr != NULL)
7194 {
7195 named_group *ng;
7196 pcre_uint32 number = cd->bracount + 1;
7197
7198 if (*ptr != (pcre_uchar)terminator)
7199 {
7200 *errorcodeptr = ERR42;
7201 goto FAILED;
7202 }
7203
7204 if (cd->names_found >= MAX_NAME_COUNT)
7205 {
7206 *errorcodeptr = ERR49;
7207 goto FAILED;
7208 }
7209
7210 if (namelen + IMM2_SIZE + 1 > cd->name_entry_size)
7211 {
7212 cd->name_entry_size = namelen + IMM2_SIZE + 1;
7213 if (namelen > MAX_NAME_SIZE)
7214 {
7215 *errorcodeptr = ERR48;
7216 goto FAILED;
7217 }
7218 }
7219
7220 /* Scan the list to check for duplicates. For duplicate names, if the
7221 number is the same, break the loop, which causes the name to be
7222 discarded; otherwise, if DUPNAMES is not set, give an error.
7223 If it is set, allow the name with a different number, but continue
7224 scanning in case this is a duplicate with the same number. For
7225 non-duplicate names, give an error if the number is duplicated. */
7226
7227 ng = cd->named_groups;
7228 for (i = 0; i < cd->names_found; i++, ng++)
7229 {
7230 if (namelen == ng->length &&
7231 STRNCMP_UC_UC(name, ng->name, namelen) == 0)
7232 {
7233 if (ng->number == number) break;
7234 if ((options & PCRE_DUPNAMES) == 0)
7235 {
7236 *errorcodeptr = ERR43;
7237 goto FAILED;
7238 }
7239 cd->dupnames = TRUE; /* Duplicate names exist */
7240 }
7241 else if (ng->number == number)
7242 {
7243 *errorcodeptr = ERR65;
7244 goto FAILED;
7245 }
7246 }
7247
7248 if (i >= cd->names_found) /* Not a duplicate with same number */
7249 {
7250 /* Increase the list size if necessary */
7251
7252 if (cd->names_found >= cd->named_group_list_size)
7253 {
7254 int newsize = cd->named_group_list_size * 2;
7255 named_group *newspace = (PUBL(malloc))
7256 (newsize * sizeof(named_group));
7257
7258 if (newspace == NULL)
7259 {
7260 *errorcodeptr = ERR21;
7261 goto FAILED;
7262 }
7263
7264 memcpy(newspace, cd->named_groups,
7265 cd->named_group_list_size * sizeof(named_group));
7266 if (cd->named_group_list_size > NAMED_GROUP_LIST_SIZE)
7267 (PUBL(free))((void *)cd->named_groups);
7268 cd->named_groups = newspace;
7269 cd->named_group_list_size = newsize;
7270 }
7271
7272 cd->named_groups[cd->names_found].name = name;
7273 cd->named_groups[cd->names_found].length = namelen;
7274 cd->named_groups[cd->names_found].number = number;
7275 cd->names_found++;
7276 }
7277 }
7278
7279 ptr++; /* Move past > or ' in both passes. */
7280 goto NUMBERED_GROUP;
7281
7282
7283 /* ------------------------------------------------------------ */
7284 case CHAR_AMPERSAND: /* Perl recursion/subroutine syntax */
7285 terminator = CHAR_RIGHT_PARENTHESIS;
7286 is_recurse = TRUE;
7287 /* Fall through */
7288
7289 /* We come here from the Python syntax above that handles both
7290 references (?P=name) and recursion (?P>name), as well as falling
7291 through from the Perl recursion syntax (?&name). We also come here from
7292 the Perl \k<name> or \k'name' back reference syntax and the \k{name}
7293 .NET syntax, and the Oniguruma \g<...> and \g'...' subroutine syntax. */
7294
7295 NAMED_REF_OR_RECURSE:
7296 name = ++ptr;
7297 if (IS_DIGIT(*ptr))
7298 {
7299 *errorcodeptr = ERR84; /* Group name must start with non-digit */
7300 goto FAILED;
7301 }
7302 while (MAX_255(*ptr) && (cd->ctypes[*ptr] & ctype_word) != 0) ptr++;
7303 namelen = (int)(ptr - name);
7304
7305 /* In the pre-compile phase, do a syntax check. We used to just set
7306 a dummy reference number, because it was not used in the first pass.
7307 However, with the change of recursive back references to be atomic,
7308 we have to look for the number so that this state can be identified, as
7309 otherwise the incorrect length is computed. If it's not a backwards
7310 reference, the dummy number will do. */
7311
7312 if (lengthptr != NULL)
7313 {
7314 named_group *ng;
7315 recno = 0;
7316
7317 if (namelen == 0)
7318 {
7319 *errorcodeptr = ERR62;
7320 goto FAILED;
7321 }
7322 if (*ptr != (pcre_uchar)terminator)
7323 {
7324 *errorcodeptr = ERR42;
7325 goto FAILED;
7326 }
7327 if (namelen > MAX_NAME_SIZE)
7328 {
7329 *errorcodeptr = ERR48;
7330 goto FAILED;
7331 }
7332
7333 /* Count named back references. */
7334
7335 if (!is_recurse) cd->namedrefcount++;
7336
7337 /* We have to allow for a named reference to a duplicated name (this
7338 cannot be determined until the second pass). This needs an extra
7339 16-bit data item. */
7340
7341 *lengthptr += IMM2_SIZE;
7342
7343 /* If this is a forward reference and we are within a (?|...) group,
7344 the reference may end up as the number of a group which we are
7345 currently inside, that is, it could be a recursive reference. In the
7346 real compile this will be picked up and the reference wrapped with
7347 OP_ONCE to make it atomic, so we must space in case this occurs. */
7348
7349 /* In fact, this can happen for a non-forward reference because
7350 another group with the same number might be created later. This
7351 issue is fixed "properly" in PCRE2. As PCRE1 is now in maintenance
7352 only mode, we finesse the bug by allowing more memory always. */
7353
7354 *lengthptr += 4 + 4*LINK_SIZE;
7355
7356 /* It is even worse than that. The current reference may be to an
7357 existing named group with a different number (so apparently not
7358 recursive) but which later on is also attached to a group with the
7359 current number. This can only happen if $(| has been previous
7360 encountered. In that case, we allow yet more memory, just in case.
7361 (Again, this is fixed "properly" in PCRE2. */
7362
7363 if (cd->dupgroups) *lengthptr += 4 + 4*LINK_SIZE;
7364
7365 /* Otherwise, check for recursion here. The name table does not exist
7366 in the first pass; instead we must scan the list of names encountered
7367 so far in order to get the number. If the name is not found, leave
7368 the value of recno as 0 for a forward reference. */
7369
7370 /* This patch (removing "else") fixes a problem when a reference is
7371 to multiple identically named nested groups from within the nest.
7372 Once again, it is not the "proper" fix, and it results in an
7373 over-allocation of memory. */
7374
7375 /* else */
7376 {
7377 ng = cd->named_groups;
7378 for (i = 0; i < cd->names_found; i++, ng++)
7379 {
7380 if (namelen == ng->length &&
7381 STRNCMP_UC_UC(name, ng->name, namelen) == 0)
7382 {
7383 open_capitem *oc;
7384 recno = ng->number;
7385 if (is_recurse) break;
7386 for (oc = cd->open_caps; oc != NULL; oc = oc->next)
7387 {
7388 if (oc->number == recno)
7389 {
7390 oc->flag = TRUE;
7391 break;
7392 }
7393 }
7394 }
7395 }
7396 }
7397 }
7398
7399 /* In the real compile, search the name table. We check the name
7400 first, and then check that we have reached the end of the name in the
7401 table. That way, if the name is longer than any in the table, the
7402 comparison will fail without reading beyond the table entry. */
7403
7404 else
7405 {
7406 slot = cd->name_table;
7407 for (i = 0; i < cd->names_found; i++)
7408 {
7409 if (STRNCMP_UC_UC(name, slot+IMM2_SIZE, namelen) == 0 &&
7410 slot[IMM2_SIZE+namelen] == 0)
7411 break;
7412 slot += cd->name_entry_size;
7413 }
7414
7415 if (i < cd->names_found)
7416 {
7417 recno = GET2(slot, 0);
7418 }
7419 else
7420 {
7421 *errorcodeptr = ERR15;
7422 goto FAILED;
7423 }
7424 }
7425
7426 /* In both phases, for recursions, we can now go to the code than
7427 handles numerical recursion. */
7428
7429 if (is_recurse) goto HANDLE_RECURSION;
7430
7431 /* In the second pass we must see if the name is duplicated. If so, we
7432 generate a different opcode. */
7433
7434 if (lengthptr == NULL && cd->dupnames)
7435 {
7436 int count = 1;
7437 unsigned int index = i;
7438 pcre_uchar *cslot = slot + cd->name_entry_size;
7439
7440 for (i++; i < cd->names_found; i++)
7441 {
7442 if (STRCMP_UC_UC(slot + IMM2_SIZE, cslot + IMM2_SIZE) != 0) break;
7443 count++;
7444 cslot += cd->name_entry_size;
7445 }
7446
7447 if (count > 1)
7448 {
7449 if (firstcharflags == REQ_UNSET) firstcharflags = REQ_NONE;
7450 previous = code;
7451 item_hwm_offset = cd->hwm - cd->start_workspace;
7452 *code++ = ((options & PCRE_CASELESS) != 0)? OP_DNREFI : OP_DNREF;
7453 PUT2INC(code, 0, index);
7454 PUT2INC(code, 0, count);
7455
7456 /* Process each potentially referenced group. */
7457
7458 for (; slot < cslot; slot += cd->name_entry_size)
7459 {
7460 open_capitem *oc;
7461 recno = GET2(slot, 0);
7462 cd->backref_map |= (recno < 32)? (1U << recno) : 1;
7463 if (recno > cd->top_backref) cd->top_backref = recno;
7464
7465 /* Check to see if this back reference is recursive, that it, it
7466 is inside the group that it references. A flag is set so that the
7467 group can be made atomic. */
7468
7469 for (oc = cd->open_caps; oc != NULL; oc = oc->next)
7470 {
7471 if (oc->number == recno)
7472 {
7473 oc->flag = TRUE;
7474 break;
7475 }
7476 }
7477 }
7478
7479 continue; /* End of back ref handling */
7480 }
7481 }
7482
7483 /* First pass, or a non-duplicated name. */
7484
7485 goto HANDLE_REFERENCE;
7486
7487
7488 /* ------------------------------------------------------------ */
7489 case CHAR_R: /* Recursion, same as (?0) */
7490 recno = 0;
7491 if (*(++ptr) != CHAR_RIGHT_PARENTHESIS)
7492 {
7493 *errorcodeptr = ERR29;
7494 goto FAILED;
7495 }
7496 goto HANDLE_RECURSION;
7497
7498
7499 /* ------------------------------------------------------------ */
7500 case CHAR_MINUS: case CHAR_PLUS: /* Recursion or subroutine */
7501 case CHAR_0: case CHAR_1: case CHAR_2: case CHAR_3: case CHAR_4:
7502 case CHAR_5: case CHAR_6: case CHAR_7: case CHAR_8: case CHAR_9:
7503 {
7504 const pcre_uchar *called;
7505 terminator = CHAR_RIGHT_PARENTHESIS;
7506
7507 /* Come here from the \g<...> and \g'...' code (Oniguruma
7508 compatibility). However, the syntax has been checked to ensure that
7509 the ... are a (signed) number, so that neither ERR63 nor ERR29 will
7510 be called on this path, nor with the jump to OTHER_CHAR_AFTER_QUERY
7511 ever be taken. */
7512
7513 HANDLE_NUMERICAL_RECURSION:
7514
7515 if ((refsign = *ptr) == CHAR_PLUS)
7516 {
7517 ptr++;
7518 if (!IS_DIGIT(*ptr))
7519 {
7520 *errorcodeptr = ERR63;
7521 goto FAILED;
7522 }
7523 }
7524 else if (refsign == CHAR_MINUS)
7525 {
7526 if (!IS_DIGIT(ptr[1]))
7527 goto OTHER_CHAR_AFTER_QUERY;
7528 ptr++;
7529 }
7530
7531 recno = 0;
7532 while(IS_DIGIT(*ptr))
7533 {
7534 if (recno > INT_MAX / 10 - 1) /* Integer overflow */
7535 {
7536 while (IS_DIGIT(*ptr)) ptr++;
7537 *errorcodeptr = ERR61;
7538 goto FAILED;
7539 }
7540 recno = recno * 10 + *ptr++ - CHAR_0;
7541 }
7542
7543 if (*ptr != (pcre_uchar)terminator)
7544 {
7545 *errorcodeptr = ERR29;
7546 goto FAILED;
7547 }
7548
7549 if (refsign == CHAR_MINUS)
7550 {
7551 if (recno == 0)
7552 {
7553 *errorcodeptr = ERR58;
7554 goto FAILED;
7555 }
7556 recno = cd->bracount - recno + 1;
7557 if (recno <= 0)
7558 {
7559 *errorcodeptr = ERR15;
7560 goto FAILED;
7561 }
7562 }
7563 else if (refsign == CHAR_PLUS)
7564 {
7565 if (recno == 0)
7566 {
7567 *errorcodeptr = ERR58;
7568 goto FAILED;
7569 }
7570 recno += cd->bracount;
7571 }
7572
7573 /* Come here from code above that handles a named recursion */
7574
7575 HANDLE_RECURSION:
7576
7577 previous = code;
7578 item_hwm_offset = cd->hwm - cd->start_workspace;
7579 called = cd->start_code;
7580
7581 /* When we are actually compiling, find the bracket that is being
7582 referenced. Temporarily end the regex in case it doesn't exist before
7583 this point. If we end up with a forward reference, first check that
7584 the bracket does occur later so we can give the error (and position)
7585 now. Then remember this forward reference in the workspace so it can
7586 be filled in at the end. */
7587
7588 if (lengthptr == NULL)
7589 {
7590 *code = OP_END;
7591 if (recno != 0)
7592 called = PRIV(find_bracket)(cd->start_code, utf, recno);
7593
7594 /* Forward reference */
7595
7596 if (called == NULL)
7597 {
7598 if (recno > cd->final_bracount)
7599 {
7600 *errorcodeptr = ERR15;
7601 goto FAILED;
7602 }
7603
7604 /* Fudge the value of "called" so that when it is inserted as an
7605 offset below, what it actually inserted is the reference number
7606 of the group. Then remember the forward reference. */
7607
7608 called = cd->start_code + recno;
7609 if (cd->hwm >= cd->start_workspace + cd->workspace_size -
7610 WORK_SIZE_SAFETY_MARGIN)
7611 {
7612 *errorcodeptr = expand_workspace(cd);
7613 if (*errorcodeptr != 0) goto FAILED;
7614 }
7615 PUTINC(cd->hwm, 0, (int)(code + 1 - cd->start_code));
7616 }
7617
7618 /* If not a forward reference, and the subpattern is still open,
7619 this is a recursive call. We check to see if this is a left
7620 recursion that could loop for ever, and diagnose that case. We
7621 must not, however, do this check if we are in a conditional
7622 subpattern because the condition might be testing for recursion in
7623 a pattern such as /(?(R)a+|(?R)b)/, which is perfectly valid.
7624 Forever loops are also detected at runtime, so those that occur in
7625 conditional subpatterns will be picked up then. */
7626
7627 else if (GET(called, 1) == 0 && cond_depth <= 0 &&
7628 could_be_empty(called, code, bcptr, utf, cd))
7629 {
7630 *errorcodeptr = ERR40;
7631 goto FAILED;
7632 }
7633 }
7634
7635 /* Insert the recursion/subroutine item. It does not have a set first
7636 character (relevant if it is repeated, because it will then be
7637 wrapped with ONCE brackets). */
7638
7639 *code = OP_RECURSE;
7640 PUT(code, 1, (int)(called - cd->start_code));
7641 code += 1 + LINK_SIZE;
7642 groupsetfirstchar = FALSE;
7643 }
7644
7645 /* Can't determine a first byte now */
7646
7647 if (firstcharflags == REQ_UNSET) firstcharflags = REQ_NONE;
7648 zerofirstchar = firstchar;
7649 zerofirstcharflags = firstcharflags;
7650 continue;
7651
7652
7653 /* ------------------------------------------------------------ */
7654 default: /* Other characters: check option setting */
7655 OTHER_CHAR_AFTER_QUERY:
7656 set = unset = 0;
7657 optset = &set;
7658
7659 while (*ptr != CHAR_RIGHT_PARENTHESIS && *ptr != CHAR_COLON)
7660 {
7661 switch (*ptr++)
7662 {
7663 case CHAR_MINUS: optset = &unset; break;
7664
7665 case CHAR_J: /* Record that it changed in the external options */
7666 *optset |= PCRE_DUPNAMES;
7667 cd->external_flags |= PCRE_JCHANGED;
7668 break;
7669
7670 case CHAR_i: *optset |= PCRE_CASELESS; break;
7671 case CHAR_m: *optset |= PCRE_MULTILINE; break;
7672 case CHAR_s: *optset |= PCRE_DOTALL; break;
7673 case CHAR_x: *optset |= PCRE_EXTENDED; break;
7674 case CHAR_U: *optset |= PCRE_UNGREEDY; break;
7675 case CHAR_X: *optset |= PCRE_EXTRA; break;
7676
7677 default: *errorcodeptr = ERR12;
7678 ptr--; /* Correct the offset */
7679 goto FAILED;
7680 }
7681 }
7682
7683 /* Set up the changed option bits, but don't change anything yet. */
7684
7685 newoptions = (options | set) & (~unset);
7686
7687 /* If the options ended with ')' this is not the start of a nested
7688 group with option changes, so the options change at this level.
7689 If we are not at the pattern start, reset the greedy defaults and the
7690 case value for firstchar and reqchar. */
7691
7692 if (*ptr == CHAR_RIGHT_PARENTHESIS)
7693 {
7694 greedy_default = ((newoptions & PCRE_UNGREEDY) != 0);
7695 greedy_non_default = greedy_default ^ 1;
7696 req_caseopt = ((newoptions & PCRE_CASELESS) != 0)? REQ_CASELESS:0;
7697
7698 /* Change options at this level, and pass them back for use
7699 in subsequent branches. */
7700
7701 *optionsptr = options = newoptions;
7702 previous = NULL; /* This item can't be repeated */
7703 continue; /* It is complete */
7704 }
7705
7706 /* If the options ended with ':' we are heading into a nested group
7707 with possible change of options. Such groups are non-capturing and are
7708 not assertions of any kind. All we need to do is skip over the ':';
7709 the newoptions value is handled below. */
7710
7711 bravalue = OP_BRA;
7712 ptr++;
7713 } /* End of switch for character following (? */
7714 } /* End of (? handling */
7715
7716 /* Opening parenthesis not followed by '*' or '?'. If PCRE_NO_AUTO_CAPTURE
7717 is set, all unadorned brackets become non-capturing and behave like (?:...)
7718 brackets. */
7719
7720 else if ((options & PCRE_NO_AUTO_CAPTURE) != 0)
7721 {
7722 bravalue = OP_BRA;
7723 }
7724
7725 /* Else we have a capturing group. */
7726
7727 else
7728 {
7729 NUMBERED_GROUP:
7730 cd->bracount += 1;
7731 PUT2(code, 1+LINK_SIZE, cd->bracount);
7732 skipbytes = IMM2_SIZE;
7733 }
7734
7735 /* Process nested bracketed regex. First check for parentheses nested too
7736 deeply. */
7737
7738 if ((cd->parens_depth += 1) > PARENS_NEST_LIMIT)
7739 {
7740 *errorcodeptr = ERR82;
7741 goto FAILED;
7742 }
7743
7744 /* All assertions used not to be repeatable, but this was changed for Perl
7745 compatibility. All kinds can now be repeated except for assertions that are
7746 conditions (Perl also forbids these to be repeated). We copy code into a
7747 non-register variable (tempcode) in order to be able to pass its address
7748 because some compilers complain otherwise. At the start of a conditional
7749 group whose condition is an assertion, cd->iscondassert is set. We unset it
7750 here so as to allow assertions later in the group to be quantified. */
7751
7752 if (bravalue >= OP_ASSERT && bravalue <= OP_ASSERTBACK_NOT &&
7753 cd->iscondassert)
7754 {
7755 previous = NULL;
7756 cd->iscondassert = FALSE;
7757 }
7758 else
7759 {
7760 previous = code;
7761 item_hwm_offset = cd->hwm - cd->start_workspace;
7762 }
7763
7764 *code = bravalue;
7765 tempcode = code;
7766 tempreqvary = cd->req_varyopt; /* Save value before bracket */
7767 tempbracount = cd->bracount; /* Save value before bracket */
7768 length_prevgroup = 0; /* Initialize for pre-compile phase */
7769
7770 if (!compile_regex(
7771 newoptions, /* The complete new option state */
7772 &tempcode, /* Where to put code (updated) */
7773 &ptr, /* Input pointer (updated) */
7774 errorcodeptr, /* Where to put an error message */
7775 (bravalue == OP_ASSERTBACK ||
7776 bravalue == OP_ASSERTBACK_NOT), /* TRUE if back assert */
7777 reset_bracount, /* True if (?| group */
7778 skipbytes, /* Skip over bracket number */
7779 cond_depth +
7780 ((bravalue == OP_COND)?1:0), /* Depth of condition subpatterns */
7781 &subfirstchar, /* For possible first char */
7782 &subfirstcharflags,
7783 &subreqchar, /* For possible last char */
7784 &subreqcharflags,
7785 bcptr, /* Current branch chain */
7786 cd, /* Tables block */
7787 (lengthptr == NULL)? NULL : /* Actual compile phase */
7788 &length_prevgroup /* Pre-compile phase */
7789 ))
7790 goto FAILED;
7791
7792 cd->parens_depth -= 1;
7793
7794 /* If this was an atomic group and there are no capturing groups within it,
7795 generate OP_ONCE_NC instead of OP_ONCE. */
7796
7797 if (bravalue == OP_ONCE && cd->bracount <= tempbracount)
7798 *code = OP_ONCE_NC;
7799
7800 if (bravalue >= OP_ASSERT && bravalue <= OP_ASSERTBACK_NOT)
7801 cd->assert_depth -= 1;
7802
7803 /* At the end of compiling, code is still pointing to the start of the
7804 group, while tempcode has been updated to point past the end of the group.
7805 The pattern pointer (ptr) is on the bracket.
7806
7807 If this is a conditional bracket, check that there are no more than
7808 two branches in the group, or just one if it's a DEFINE group. We do this
7809 in the real compile phase, not in the pre-pass, where the whole group may
7810 not be available. */
7811
7812 if (bravalue == OP_COND && lengthptr == NULL)
7813 {
7814 pcre_uchar *tc = code;
7815 int condcount = 0;
7816
7817 do {
7818 condcount++;
7819 tc += GET(tc,1);
7820 }
7821 while (*tc != OP_KET);
7822
7823 /* A DEFINE group is never obeyed inline (the "condition" is always
7824 false). It must have only one branch. */
7825
7826 if (code[LINK_SIZE+1] == OP_DEF)
7827 {
7828 if (condcount > 1)
7829 {
7830 *errorcodeptr = ERR54;
7831 goto FAILED;
7832 }
7833 bravalue = OP_DEF; /* Just a flag to suppress char handling below */
7834 }
7835
7836 /* A "normal" conditional group. If there is just one branch, we must not
7837 make use of its firstchar or reqchar, because this is equivalent to an
7838 empty second branch. */
7839
7840 else
7841 {
7842 if (condcount > 2)
7843 {
7844 *errorcodeptr = ERR27;
7845 goto FAILED;
7846 }
7847 if (condcount == 1) subfirstcharflags = subreqcharflags = REQ_NONE;
7848 }
7849 }
7850
7851 /* Error if hit end of pattern */
7852
7853 if (*ptr != CHAR_RIGHT_PARENTHESIS)
7854 {
7855 *errorcodeptr = ERR14;
7856 goto FAILED;
7857 }
7858
7859 /* In the pre-compile phase, update the length by the length of the group,
7860 less the brackets at either end. Then reduce the compiled code to just a
7861 set of non-capturing brackets so that it doesn't use much memory if it is
7862 duplicated by a quantifier.*/
7863
7864 if (lengthptr != NULL)
7865 {
7866 if (OFLOW_MAX - *lengthptr < length_prevgroup - 2 - 2*LINK_SIZE)
7867 {
7868 *errorcodeptr = ERR20;
7869 goto FAILED;
7870 }
7871 *lengthptr += length_prevgroup - 2 - 2*LINK_SIZE;
7872 code++; /* This already contains bravalue */
7873 PUTINC(code, 0, 1 + LINK_SIZE);
7874 *code++ = OP_KET;
7875 PUTINC(code, 0, 1 + LINK_SIZE);
7876 break; /* No need to waste time with special character handling */
7877 }
7878
7879 /* Otherwise update the main code pointer to the end of the group. */
7880
7881 code = tempcode;
7882
7883 /* For a DEFINE group, required and first character settings are not
7884 relevant. */
7885
7886 if (bravalue == OP_DEF) break;
7887
7888 /* Handle updating of the required and first characters for other types of
7889 group. Update for normal brackets of all kinds, and conditions with two
7890 branches (see code above). If the bracket is followed by a quantifier with
7891 zero repeat, we have to back off. Hence the definition of zeroreqchar and
7892 zerofirstchar outside the main loop so that they can be accessed for the
7893 back off. */
7894
7895 zeroreqchar = reqchar;
7896 zeroreqcharflags = reqcharflags;
7897 zerofirstchar = firstchar;
7898 zerofirstcharflags = firstcharflags;
7899 groupsetfirstchar = FALSE;
7900
7901 if (bravalue >= OP_ONCE)
7902 {
7903 /* If we have not yet set a firstchar in this branch, take it from the
7904 subpattern, remembering that it was set here so that a repeat of more
7905 than one can replicate it as reqchar if necessary. If the subpattern has
7906 no firstchar, set "none" for the whole branch. In both cases, a zero
7907 repeat forces firstchar to "none". */
7908
7909 if (firstcharflags == REQ_UNSET)
7910 {
7911 if (subfirstcharflags >= 0)
7912 {
7913 firstchar = subfirstchar;
7914 firstcharflags = subfirstcharflags;
7915 groupsetfirstchar = TRUE;
7916 }
7917 else firstcharflags = REQ_NONE;
7918 zerofirstcharflags = REQ_NONE;
7919 }
7920
7921 /* If firstchar was previously set, convert the subpattern's firstchar
7922 into reqchar if there wasn't one, using the vary flag that was in
7923 existence beforehand. */
7924
7925 else if (subfirstcharflags >= 0 && subreqcharflags < 0)
7926 {
7927 subreqchar = subfirstchar;
7928 subreqcharflags = subfirstcharflags | tempreqvary;
7929 }
7930
7931 /* If the subpattern set a required byte (or set a first byte that isn't
7932 really the first byte - see above), set it. */
7933
7934 if (subreqcharflags >= 0)
7935 {
7936 reqchar = subreqchar;
7937 reqcharflags = subreqcharflags;
7938 }
7939 }
7940
7941 /* For a forward assertion, we take the reqchar, if set, provided that the
7942 group has also set a first char. This can be helpful if the pattern that
7943 follows the assertion doesn't set a different char. For example, it's
7944 useful for /(?=abcde).+/. We can't set firstchar for an assertion, however
7945 because it leads to incorrect effect for patterns such as /(?=a)a.+/ when
7946 the "real" "a" would then become a reqchar instead of a firstchar. This is
7947 overcome by a scan at the end if there's no firstchar, looking for an
7948 asserted first char. */
7949
7950 else if (bravalue == OP_ASSERT && subreqcharflags >= 0 &&
7951 subfirstcharflags >= 0)
7952 {
7953 reqchar = subreqchar;
7954 reqcharflags = subreqcharflags;
7955 }
7956 break; /* End of processing '(' */
7957
7958
7959 /* ===================================================================*/
7960 /* Handle metasequences introduced by \. For ones like \d, the ESC_ values
7961 are arranged to be the negation of the corresponding OP_values in the
7962 default case when PCRE_UCP is not set. For the back references, the values
7963 are negative the reference number. Only back references and those types
7964 that consume a character may be repeated. We can test for values between
7965 ESC_b and ESC_Z for the latter; this may have to change if any new ones are
7966 ever created. */
7967
7968 case CHAR_BACKSLASH:
7969 tempptr = ptr;
7970 escape = check_escape(&ptr, &ec, errorcodeptr, cd->bracount, options, FALSE);
7971 if (*errorcodeptr != 0) goto FAILED;
7972
7973 if (escape == 0) /* The escape coded a single character */
7974 c = ec;
7975 else
7976 {
7977 /* For metasequences that actually match a character, we disable the
7978 setting of a first character if it hasn't already been set. */
7979
7980 if (firstcharflags == REQ_UNSET && escape > ESC_b && escape < ESC_Z)
7981 firstcharflags = REQ_NONE;
7982
7983 /* Set values to reset to if this is followed by a zero repeat. */
7984
7985 zerofirstchar = firstchar;
7986 zerofirstcharflags = firstcharflags;
7987 zeroreqchar = reqchar;
7988 zeroreqcharflags = reqcharflags;
7989
7990 /* \g<name> or \g'name' is a subroutine call by name and \g<n> or \g'n'
7991 is a subroutine call by number (Oniguruma syntax). In fact, the value
7992 ESC_g is returned only for these cases. So we don't need to check for <
7993 or ' if the value is ESC_g. For the Perl syntax \g{n} the value is
7994 -n, and for the Perl syntax \g{name} the result is ESC_k (as
7995 that is a synonym for a named back reference). */
7996
7997 if (escape == ESC_g)
7998 {
7999 const pcre_uchar *p;
8000 pcre_uint32 cf;
8001
8002 item_hwm_offset = cd->hwm - cd->start_workspace; /* Normally this is set when '(' is read */
8003 terminator = (*(++ptr) == CHAR_LESS_THAN_SIGN)?
8004 CHAR_GREATER_THAN_SIGN : CHAR_APOSTROPHE;
8005
8006 /* These two statements stop the compiler for warning about possibly
8007 unset variables caused by the jump to HANDLE_NUMERICAL_RECURSION. In
8008 fact, because we do the check for a number below, the paths that
8009 would actually be in error are never taken. */
8010
8011 skipbytes = 0;
8012 reset_bracount = FALSE;
8013
8014 /* If it's not a signed or unsigned number, treat it as a name. */
8015
8016 cf = ptr[1];
8017 if (cf != CHAR_PLUS && cf != CHAR_MINUS && !IS_DIGIT(cf))
8018 {
8019 is_recurse = TRUE;
8020 goto NAMED_REF_OR_RECURSE;
8021 }
8022
8023 /* Signed or unsigned number (cf = ptr[1]) is known to be plus or minus
8024 or a digit. */
8025
8026 p = ptr + 2;
8027 while (IS_DIGIT(*p)) p++;
8028 if (*p != (pcre_uchar)terminator)
8029 {
8030 *errorcodeptr = ERR57;
8031 goto FAILED;
8032 }
8033 ptr++;
8034 goto HANDLE_NUMERICAL_RECURSION;
8035 }
8036
8037 /* \k<name> or \k'name' is a back reference by name (Perl syntax).
8038 We also support \k{name} (.NET syntax). */
8039
8040 if (escape == ESC_k)
8041 {
8042 if ((ptr[1] != CHAR_LESS_THAN_SIGN &&
8043 ptr[1] != CHAR_APOSTROPHE && ptr[1] != CHAR_LEFT_CURLY_BRACKET))
8044 {
8045 *errorcodeptr = ERR69;
8046 goto FAILED;
8047 }
8048 is_recurse = FALSE;
8049 terminator = (*(++ptr) == CHAR_LESS_THAN_SIGN)?
8050 CHAR_GREATER_THAN_SIGN : (*ptr == CHAR_APOSTROPHE)?
8051 CHAR_APOSTROPHE : CHAR_RIGHT_CURLY_BRACKET;
8052 goto NAMED_REF_OR_RECURSE;
8053 }
8054
8055 /* Back references are handled specially; must disable firstchar if
8056 not set to cope with cases like (?=(\w+))\1: which would otherwise set
8057 ':' later. */
8058
8059 if (escape < 0)
8060 {
8061 open_capitem *oc;
8062 recno = -escape;
8063
8064 /* Come here from named backref handling when the reference is to a
8065 single group (i.e. not to a duplicated name. */
8066
8067 HANDLE_REFERENCE:
8068 if (firstcharflags == REQ_UNSET) zerofirstcharflags = firstcharflags = REQ_NONE;
8069 previous = code;
8070 item_hwm_offset = cd->hwm - cd->start_workspace;
8071 *code++ = ((options & PCRE_CASELESS) != 0)? OP_REFI : OP_REF;
8072 PUT2INC(code, 0, recno);
8073 cd->backref_map |= (recno < 32)? (1U << recno) : 1;
8074 if (recno > cd->top_backref) cd->top_backref = recno;
8075
8076 /* Check to see if this back reference is recursive, that it, it
8077 is inside the group that it references. A flag is set so that the
8078 group can be made atomic. */
8079
8080 for (oc = cd->open_caps; oc != NULL; oc = oc->next)
8081 {
8082 if (oc->number == recno)
8083 {
8084 oc->flag = TRUE;
8085 break;
8086 }
8087 }
8088 }
8089
8090 /* So are Unicode property matches, if supported. */
8091
8092 #ifdef SUPPORT_UCP
8093 else if (escape == ESC_P || escape == ESC_p)
8094 {
8095 BOOL negated;
8096 unsigned int ptype = 0, pdata = 0;
8097 if (!get_ucp(&ptr, &negated, &ptype, &pdata, errorcodeptr))
8098 goto FAILED;
8099 previous = code;
8100 item_hwm_offset = cd->hwm - cd->start_workspace;
8101 *code++ = ((escape == ESC_p) != negated)? OP_PROP : OP_NOTPROP;
8102 *code++ = ptype;
8103 *code++ = pdata;
8104 }
8105 #else
8106
8107 /* If Unicode properties are not supported, \X, \P, and \p are not
8108 allowed. */
8109
8110 else if (escape == ESC_X || escape == ESC_P || escape == ESC_p)
8111 {
8112 *errorcodeptr = ERR45;
8113 goto FAILED;
8114 }
8115 #endif
8116
8117 /* For the rest (including \X when Unicode properties are supported), we
8118 can obtain the OP value by negating the escape value in the default
8119 situation when PCRE_UCP is not set. When it *is* set, we substitute
8120 Unicode property tests. Note that \b and \B do a one-character
8121 lookbehind, and \A also behaves as if it does. */
8122
8123 else
8124 {
8125 if ((escape == ESC_b || escape == ESC_B || escape == ESC_A) &&
8126 cd->max_lookbehind == 0)
8127 cd->max_lookbehind = 1;
8128 #ifdef SUPPORT_UCP
8129 if (escape >= ESC_DU && escape <= ESC_wu)
8130 {
8131 nestptr = ptr + 1; /* Where to resume */
8132 ptr = substitutes[escape - ESC_DU] - 1; /* Just before substitute */
8133 }
8134 else
8135 #endif
8136 /* In non-UTF-8 mode, we turn \C into OP_ALLANY instead of OP_ANYBYTE
8137 so that it works in DFA mode and in lookbehinds. */
8138
8139 {
8140 previous = (escape > ESC_b && escape < ESC_Z)? code : NULL;
8141 item_hwm_offset = cd->hwm - cd->start_workspace;
8142 *code++ = (!utf && escape == ESC_C)? OP_ALLANY : escape;
8143 }
8144 }
8145 continue;
8146 }
8147
8148 /* We have a data character whose value is in c. In UTF-8 mode it may have
8149 a value > 127. We set its representation in the length/buffer, and then
8150 handle it as a data character. */
8151
8152 #if defined SUPPORT_UTF && !defined COMPILE_PCRE32
8153 if (utf && c > MAX_VALUE_FOR_SINGLE_CHAR)
8154 mclength = PRIV(ord2utf)(c, mcbuffer);
8155 else
8156 #endif
8157
8158 {
8159 mcbuffer[0] = c;
8160 mclength = 1;
8161 }
8162 goto ONE_CHAR;
8163
8164
8165 /* ===================================================================*/
8166 /* Handle a literal character. It is guaranteed not to be whitespace or #
8167 when the extended flag is set. If we are in a UTF mode, it may be a
8168 multi-unit literal character. */
8169
8170 default:
8171 NORMAL_CHAR:
8172 mclength = 1;
8173 mcbuffer[0] = c;
8174
8175 #ifdef SUPPORT_UTF
8176 if (utf && HAS_EXTRALEN(c))
8177 ACROSSCHAR(TRUE, ptr[1], mcbuffer[mclength++] = *(++ptr));
8178 #endif
8179
8180 /* At this point we have the character's bytes in mcbuffer, and the length
8181 in mclength. When not in UTF-8 mode, the length is always 1. */
8182
8183 ONE_CHAR:
8184 previous = code;
8185 item_hwm_offset = cd->hwm - cd->start_workspace;
8186
8187 /* For caseless UTF-8 mode when UCP support is available, check whether
8188 this character has more than one other case. If so, generate a special
8189 OP_PROP item instead of OP_CHARI. */
8190
8191 #ifdef SUPPORT_UCP
8192 if (utf && (options & PCRE_CASELESS) != 0)
8193 {
8194 GETCHAR(c, mcbuffer);
8195 if ((c = UCD_CASESET(c)) != 0)
8196 {
8197 *code++ = OP_PROP;
8198 *code++ = PT_CLIST;
8199 *code++ = c;
8200 if (firstcharflags == REQ_UNSET)
8201 firstcharflags = zerofirstcharflags = REQ_NONE;
8202 break;
8203 }
8204 }
8205 #endif
8206
8207 /* Caseful matches, or not one of the multicase characters. */
8208
8209 *code++ = ((options & PCRE_CASELESS) != 0)? OP_CHARI : OP_CHAR;
8210 for (c = 0; c < mclength; c++) *code++ = mcbuffer[c];
8211
8212 /* Remember if \r or \n were seen */
8213
8214 if (mcbuffer[0] == CHAR_CR || mcbuffer[0] == CHAR_NL)
8215 cd->external_flags |= PCRE_HASCRORLF;
8216
8217 /* Set the first and required bytes appropriately. If no previous first
8218 byte, set it from this character, but revert to none on a zero repeat.
8219 Otherwise, leave the firstchar value alone, and don't change it on a zero
8220 repeat. */
8221
8222 if (firstcharflags == REQ_UNSET)
8223 {
8224 zerofirstcharflags = REQ_NONE;
8225 zeroreqchar = reqchar;
8226 zeroreqcharflags = reqcharflags;
8227
8228 /* If the character is more than one byte long, we can set firstchar
8229 only if it is not to be matched caselessly. */
8230
8231 if (mclength == 1 || req_caseopt == 0)
8232 {
8233 firstchar = mcbuffer[0];
8234 firstcharflags = req_caseopt;
8235
8236 if (mclength != 1)
8237 {
8238 reqchar = code[-1];
8239 reqcharflags = cd->req_varyopt;
8240 }
8241 }
8242 else firstcharflags = reqcharflags = REQ_NONE;
8243 }
8244
8245 /* firstchar was previously set; we can set reqchar only if the length is
8246 1 or the matching is caseful. */
8247
8248 else
8249 {
8250 zerofirstchar = firstchar;
8251 zerofirstcharflags = firstcharflags;
8252 zeroreqchar = reqchar;
8253 zeroreqcharflags = reqcharflags;
8254 if (mclength == 1 || req_caseopt == 0)
8255 {
8256 reqchar = code[-1];
8257 reqcharflags = req_caseopt | cd->req_varyopt;
8258 }
8259 }
8260
8261 break; /* End of literal character handling */
8262 }
8263 } /* end of big loop */
8264
8265
8266 /* Control never reaches here by falling through, only by a goto for all the
8267 error states. Pass back the position in the pattern so that it can be displayed
8268 to the user for diagnosing the error. */
8269
8270 FAILED:
8271 *ptrptr = ptr;
8272 return FALSE;
8273 }
8274
8275
8276
8277 /*************************************************
8278 * Compile sequence of alternatives *
8279 *************************************************/
8280
8281 /* On entry, ptr is pointing past the bracket character, but on return it
8282 points to the closing bracket, or vertical bar, or end of string. The code
8283 variable is pointing at the byte into which the BRA operator has been stored.
8284 This function is used during the pre-compile phase when we are trying to find
8285 out the amount of memory needed, as well as during the real compile phase. The
8286 value of lengthptr distinguishes the two phases.
8287
8288 Arguments:
8289 options option bits, including any changes for this subpattern
8290 codeptr -> the address of the current code pointer
8291 ptrptr -> the address of the current pattern pointer
8292 errorcodeptr -> pointer to error code variable
8293 lookbehind TRUE if this is a lookbehind assertion
8294 reset_bracount TRUE to reset the count for each branch
8295 skipbytes skip this many bytes at start (for brackets and OP_COND)
8296 cond_depth depth of nesting for conditional subpatterns
8297 firstcharptr place to put the first required character
8298 firstcharflagsptr place to put the first character flags, or a negative number
8299 reqcharptr place to put the last required character
8300 reqcharflagsptr place to put the last required character flags, or a negative number
8301 bcptr pointer to the chain of currently open branches
8302 cd points to the data block with tables pointers etc.
8303 lengthptr NULL during the real compile phase
8304 points to length accumulator during pre-compile phase
8305
8306 Returns: TRUE on success
8307 */
8308
8309 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)8310 compile_regex(int options, pcre_uchar **codeptr, const pcre_uchar **ptrptr,
8311 int *errorcodeptr, BOOL lookbehind, BOOL reset_bracount, int skipbytes,
8312 int cond_depth,
8313 pcre_uint32 *firstcharptr, pcre_int32 *firstcharflagsptr,
8314 pcre_uint32 *reqcharptr, pcre_int32 *reqcharflagsptr,
8315 branch_chain *bcptr, compile_data *cd, int *lengthptr)
8316 {
8317 const pcre_uchar *ptr = *ptrptr;
8318 pcre_uchar *code = *codeptr;
8319 pcre_uchar *last_branch = code;
8320 pcre_uchar *start_bracket = code;
8321 pcre_uchar *reverse_count = NULL;
8322 open_capitem capitem;
8323 int capnumber = 0;
8324 pcre_uint32 firstchar, reqchar;
8325 pcre_int32 firstcharflags, reqcharflags;
8326 pcre_uint32 branchfirstchar, branchreqchar;
8327 pcre_int32 branchfirstcharflags, branchreqcharflags;
8328 int length;
8329 unsigned int orig_bracount;
8330 unsigned int max_bracount;
8331 branch_chain bc;
8332 size_t save_hwm_offset;
8333
8334 /* If set, call the external function that checks for stack availability. */
8335
8336 if (PUBL(stack_guard) != NULL && PUBL(stack_guard)())
8337 {
8338 *errorcodeptr= ERR85;
8339 return FALSE;
8340 }
8341
8342 /* Miscellaneous initialization */
8343
8344 bc.outer = bcptr;
8345 bc.current_branch = code;
8346
8347 firstchar = reqchar = 0;
8348 firstcharflags = reqcharflags = REQ_UNSET;
8349
8350 save_hwm_offset = cd->hwm - cd->start_workspace;
8351
8352 /* Accumulate the length for use in the pre-compile phase. Start with the
8353 length of the BRA and KET and any extra bytes that are required at the
8354 beginning. We accumulate in a local variable to save frequent testing of
8355 lenthptr for NULL. We cannot do this by looking at the value of code at the
8356 start and end of each alternative, because compiled items are discarded during
8357 the pre-compile phase so that the work space is not exceeded. */
8358
8359 length = 2 + 2*LINK_SIZE + skipbytes;
8360
8361 /* WARNING: If the above line is changed for any reason, you must also change
8362 the code that abstracts option settings at the start of the pattern and makes
8363 them global. It tests the value of length for (2 + 2*LINK_SIZE) in the
8364 pre-compile phase to find out whether anything has yet been compiled or not. */
8365
8366 /* If this is a capturing subpattern, add to the chain of open capturing items
8367 so that we can detect them if (*ACCEPT) is encountered. This is also used to
8368 detect groups that contain recursive back references to themselves. Note that
8369 only OP_CBRA need be tested here; changing this opcode to one of its variants,
8370 e.g. OP_SCBRAPOS, happens later, after the group has been compiled. */
8371
8372 if (*code == OP_CBRA)
8373 {
8374 capnumber = GET2(code, 1 + LINK_SIZE);
8375 capitem.number = capnumber;
8376 capitem.next = cd->open_caps;
8377 capitem.flag = FALSE;
8378 cd->open_caps = &capitem;
8379 }
8380
8381 /* Offset is set zero to mark that this bracket is still open */
8382
8383 PUT(code, 1, 0);
8384 code += 1 + LINK_SIZE + skipbytes;
8385
8386 /* Loop for each alternative branch */
8387
8388 orig_bracount = max_bracount = cd->bracount;
8389 for (;;)
8390 {
8391 /* For a (?| group, reset the capturing bracket count so that each branch
8392 uses the same numbers. */
8393
8394 if (reset_bracount) cd->bracount = orig_bracount;
8395
8396 /* Set up dummy OP_REVERSE if lookbehind assertion */
8397
8398 if (lookbehind)
8399 {
8400 *code++ = OP_REVERSE;
8401 reverse_count = code;
8402 PUTINC(code, 0, 0);
8403 length += 1 + LINK_SIZE;
8404 }
8405
8406 /* Now compile the branch; in the pre-compile phase its length gets added
8407 into the length. */
8408
8409 if (!compile_branch(&options, &code, &ptr, errorcodeptr, &branchfirstchar,
8410 &branchfirstcharflags, &branchreqchar, &branchreqcharflags, &bc,
8411 cond_depth, cd, (lengthptr == NULL)? NULL : &length))
8412 {
8413 *ptrptr = ptr;
8414 return FALSE;
8415 }
8416
8417 /* Keep the highest bracket count in case (?| was used and some branch
8418 has fewer than the rest. */
8419
8420 if (cd->bracount > max_bracount) max_bracount = cd->bracount;
8421
8422 /* In the real compile phase, there is some post-processing to be done. */
8423
8424 if (lengthptr == NULL)
8425 {
8426 /* If this is the first branch, the firstchar and reqchar values for the
8427 branch become the values for the regex. */
8428
8429 if (*last_branch != OP_ALT)
8430 {
8431 firstchar = branchfirstchar;
8432 firstcharflags = branchfirstcharflags;
8433 reqchar = branchreqchar;
8434 reqcharflags = branchreqcharflags;
8435 }
8436
8437 /* If this is not the first branch, the first char and reqchar have to
8438 match the values from all the previous branches, except that if the
8439 previous value for reqchar didn't have REQ_VARY set, it can still match,
8440 and we set REQ_VARY for the regex. */
8441
8442 else
8443 {
8444 /* If we previously had a firstchar, but it doesn't match the new branch,
8445 we have to abandon the firstchar for the regex, but if there was
8446 previously no reqchar, it takes on the value of the old firstchar. */
8447
8448 if (firstcharflags >= 0 &&
8449 (firstcharflags != branchfirstcharflags || firstchar != branchfirstchar))
8450 {
8451 if (reqcharflags < 0)
8452 {
8453 reqchar = firstchar;
8454 reqcharflags = firstcharflags;
8455 }
8456 firstcharflags = REQ_NONE;
8457 }
8458
8459 /* If we (now or from before) have no firstchar, a firstchar from the
8460 branch becomes a reqchar if there isn't a branch reqchar. */
8461
8462 if (firstcharflags < 0 && branchfirstcharflags >= 0 && branchreqcharflags < 0)
8463 {
8464 branchreqchar = branchfirstchar;
8465 branchreqcharflags = branchfirstcharflags;
8466 }
8467
8468 /* Now ensure that the reqchars match */
8469
8470 if (((reqcharflags & ~REQ_VARY) != (branchreqcharflags & ~REQ_VARY)) ||
8471 reqchar != branchreqchar)
8472 reqcharflags = REQ_NONE;
8473 else
8474 {
8475 reqchar = branchreqchar;
8476 reqcharflags |= branchreqcharflags; /* To "or" REQ_VARY */
8477 }
8478 }
8479
8480 /* If lookbehind, check that this branch matches a fixed-length string, and
8481 put the length into the OP_REVERSE item. Temporarily mark the end of the
8482 branch with OP_END. If the branch contains OP_RECURSE, the result is -3
8483 because there may be forward references that we can't check here. Set a
8484 flag to cause another lookbehind check at the end. Why not do it all at the
8485 end? Because common, erroneous checks are picked up here and the offset of
8486 the problem can be shown. */
8487
8488 if (lookbehind)
8489 {
8490 int fixed_length;
8491 *code = OP_END;
8492 fixed_length = find_fixedlength(last_branch, (options & PCRE_UTF8) != 0,
8493 FALSE, cd, NULL);
8494 DPRINTF(("fixed length = %d\n", fixed_length));
8495 if (fixed_length == -3)
8496 {
8497 cd->check_lookbehind = TRUE;
8498 }
8499 else if (fixed_length < 0)
8500 {
8501 *errorcodeptr = (fixed_length == -2)? ERR36 :
8502 (fixed_length == -4)? ERR70: ERR25;
8503 *ptrptr = ptr;
8504 return FALSE;
8505 }
8506 else
8507 {
8508 if (fixed_length > cd->max_lookbehind)
8509 cd->max_lookbehind = fixed_length;
8510 PUT(reverse_count, 0, fixed_length);
8511 }
8512 }
8513 }
8514
8515 /* Reached end of expression, either ')' or end of pattern. In the real
8516 compile phase, go back through the alternative branches and reverse the chain
8517 of offsets, with the field in the BRA item now becoming an offset to the
8518 first alternative. If there are no alternatives, it points to the end of the
8519 group. The length in the terminating ket is always the length of the whole
8520 bracketed item. Return leaving the pointer at the terminating char. */
8521
8522 if (*ptr != CHAR_VERTICAL_LINE)
8523 {
8524 if (lengthptr == NULL)
8525 {
8526 int branch_length = (int)(code - last_branch);
8527 do
8528 {
8529 int prev_length = GET(last_branch, 1);
8530 PUT(last_branch, 1, branch_length);
8531 branch_length = prev_length;
8532 last_branch -= branch_length;
8533 }
8534 while (branch_length > 0);
8535 }
8536
8537 /* Fill in the ket */
8538
8539 *code = OP_KET;
8540 PUT(code, 1, (int)(code - start_bracket));
8541 code += 1 + LINK_SIZE;
8542
8543 /* If it was a capturing subpattern, check to see if it contained any
8544 recursive back references. If so, we must wrap it in atomic brackets.
8545 Because we are moving code along, we must ensure that any pending recursive
8546 references are updated. In any event, remove the block from the chain. */
8547
8548 if (capnumber > 0)
8549 {
8550 if (cd->open_caps->flag)
8551 {
8552 *code = OP_END;
8553 adjust_recurse(start_bracket, 1 + LINK_SIZE,
8554 (options & PCRE_UTF8) != 0, cd, save_hwm_offset);
8555 memmove(start_bracket + 1 + LINK_SIZE, start_bracket,
8556 IN_UCHARS(code - start_bracket));
8557 *start_bracket = OP_ONCE;
8558 code += 1 + LINK_SIZE;
8559 PUT(start_bracket, 1, (int)(code - start_bracket));
8560 *code = OP_KET;
8561 PUT(code, 1, (int)(code - start_bracket));
8562 code += 1 + LINK_SIZE;
8563 length += 2 + 2*LINK_SIZE;
8564 }
8565 cd->open_caps = cd->open_caps->next;
8566 }
8567
8568 /* Retain the highest bracket number, in case resetting was used. */
8569
8570 cd->bracount = max_bracount;
8571
8572 /* Set values to pass back */
8573
8574 *codeptr = code;
8575 *ptrptr = ptr;
8576 *firstcharptr = firstchar;
8577 *firstcharflagsptr = firstcharflags;
8578 *reqcharptr = reqchar;
8579 *reqcharflagsptr = reqcharflags;
8580 if (lengthptr != NULL)
8581 {
8582 if (OFLOW_MAX - *lengthptr < length)
8583 {
8584 *errorcodeptr = ERR20;
8585 return FALSE;
8586 }
8587 *lengthptr += length;
8588 }
8589 return TRUE;
8590 }
8591
8592 /* Another branch follows. In the pre-compile phase, we can move the code
8593 pointer back to where it was for the start of the first branch. (That is,
8594 pretend that each branch is the only one.)
8595
8596 In the real compile phase, insert an ALT node. Its length field points back
8597 to the previous branch while the bracket remains open. At the end the chain
8598 is reversed. It's done like this so that the start of the bracket has a
8599 zero offset until it is closed, making it possible to detect recursion. */
8600
8601 if (lengthptr != NULL)
8602 {
8603 code = *codeptr + 1 + LINK_SIZE + skipbytes;
8604 length += 1 + LINK_SIZE;
8605 }
8606 else
8607 {
8608 *code = OP_ALT;
8609 PUT(code, 1, (int)(code - last_branch));
8610 bc.current_branch = last_branch = code;
8611 code += 1 + LINK_SIZE;
8612 }
8613
8614 ptr++;
8615 }
8616 /* Control never reaches here */
8617 }
8618
8619
8620
8621
8622 /*************************************************
8623 * Check for anchored expression *
8624 *************************************************/
8625
8626 /* Try to find out if this is an anchored regular expression. Consider each
8627 alternative branch. If they all start with OP_SOD or OP_CIRC, or with a bracket
8628 all of whose alternatives start with OP_SOD or OP_CIRC (recurse ad lib), then
8629 it's anchored. However, if this is a multiline pattern, then only OP_SOD will
8630 be found, because ^ generates OP_CIRCM in that mode.
8631
8632 We can also consider a regex to be anchored if OP_SOM starts all its branches.
8633 This is the code for \G, which means "match at start of match position, taking
8634 into account the match offset".
8635
8636 A branch is also implicitly anchored if it starts with .* and DOTALL is set,
8637 because that will try the rest of the pattern at all possible matching points,
8638 so there is no point trying again.... er ....
8639
8640 .... except when the .* appears inside capturing parentheses, and there is a
8641 subsequent back reference to those parentheses. We haven't enough information
8642 to catch that case precisely.
8643
8644 At first, the best we could do was to detect when .* was in capturing brackets
8645 and the highest back reference was greater than or equal to that level.
8646 However, by keeping a bitmap of the first 31 back references, we can catch some
8647 of the more common cases more precisely.
8648
8649 ... A second exception is when the .* appears inside an atomic group, because
8650 this prevents the number of characters it matches from being adjusted.
8651
8652 Arguments:
8653 code points to start of expression (the bracket)
8654 bracket_map a bitmap of which brackets we are inside while testing; this
8655 handles up to substring 31; after that we just have to take
8656 the less precise approach
8657 cd points to the compile data block
8658 atomcount atomic group level
8659
8660 Returns: TRUE or FALSE
8661 */
8662
8663 static BOOL
is_anchored(register const pcre_uchar * code,unsigned int bracket_map,compile_data * cd,int atomcount)8664 is_anchored(register const pcre_uchar *code, unsigned int bracket_map,
8665 compile_data *cd, int atomcount)
8666 {
8667 do {
8668 const pcre_uchar *scode = first_significant_code(
8669 code + PRIV(OP_lengths)[*code], FALSE);
8670 register int op = *scode;
8671
8672 /* Non-capturing brackets */
8673
8674 if (op == OP_BRA || op == OP_BRAPOS ||
8675 op == OP_SBRA || op == OP_SBRAPOS)
8676 {
8677 if (!is_anchored(scode, bracket_map, cd, atomcount)) return FALSE;
8678 }
8679
8680 /* Capturing brackets */
8681
8682 else if (op == OP_CBRA || op == OP_CBRAPOS ||
8683 op == OP_SCBRA || op == OP_SCBRAPOS)
8684 {
8685 int n = GET2(scode, 1+LINK_SIZE);
8686 int new_map = bracket_map | ((n < 32)? (1U << n) : 1);
8687 if (!is_anchored(scode, new_map, cd, atomcount)) return FALSE;
8688 }
8689
8690 /* Positive forward assertion */
8691
8692 else if (op == OP_ASSERT)
8693 {
8694 if (!is_anchored(scode, bracket_map, cd, atomcount)) return FALSE;
8695 }
8696
8697 /* Condition; not anchored if no second branch */
8698
8699 else if (op == OP_COND)
8700 {
8701 if (scode[GET(scode,1)] != OP_ALT) return FALSE;
8702 if (!is_anchored(scode, bracket_map, cd, atomcount)) return FALSE;
8703 }
8704
8705 /* Atomic groups */
8706
8707 else if (op == OP_ONCE || op == OP_ONCE_NC)
8708 {
8709 if (!is_anchored(scode, bracket_map, cd, atomcount + 1))
8710 return FALSE;
8711 }
8712
8713 /* .* is not anchored unless DOTALL is set (which generates OP_ALLANY) and
8714 it isn't in brackets that are or may be referenced or inside an atomic
8715 group. */
8716
8717 else if ((op == OP_TYPESTAR || op == OP_TYPEMINSTAR ||
8718 op == OP_TYPEPOSSTAR))
8719 {
8720 if (scode[1] != OP_ALLANY || (bracket_map & cd->backref_map) != 0 ||
8721 atomcount > 0 || cd->had_pruneorskip)
8722 return FALSE;
8723 }
8724
8725 /* Check for explicit anchoring */
8726
8727 else if (op != OP_SOD && op != OP_SOM && op != OP_CIRC) return FALSE;
8728
8729 code += GET(code, 1);
8730 }
8731 while (*code == OP_ALT); /* Loop for each alternative */
8732 return TRUE;
8733 }
8734
8735
8736
8737 /*************************************************
8738 * Check for starting with ^ or .* *
8739 *************************************************/
8740
8741 /* This is called to find out if every branch starts with ^ or .* so that
8742 "first char" processing can be done to speed things up in multiline
8743 matching and for non-DOTALL patterns that start with .* (which must start at
8744 the beginning or after \n). As in the case of is_anchored() (see above), we
8745 have to take account of back references to capturing brackets that contain .*
8746 because in that case we can't make the assumption. Also, the appearance of .*
8747 inside atomic brackets or in an assertion, or in a pattern that contains *PRUNE
8748 or *SKIP does not count, because once again the assumption no longer holds.
8749
8750 Arguments:
8751 code points to start of expression (the bracket)
8752 bracket_map a bitmap of which brackets we are inside while testing; this
8753 handles up to substring 31; after that we just have to take
8754 the less precise approach
8755 cd points to the compile data
8756 atomcount atomic group level
8757 inassert TRUE if in an assertion
8758
8759 Returns: TRUE or FALSE
8760 */
8761
8762 static BOOL
is_startline(const pcre_uchar * code,unsigned int bracket_map,compile_data * cd,int atomcount,BOOL inassert)8763 is_startline(const pcre_uchar *code, unsigned int bracket_map,
8764 compile_data *cd, int atomcount, BOOL inassert)
8765 {
8766 do {
8767 const pcre_uchar *scode = first_significant_code(
8768 code + PRIV(OP_lengths)[*code], FALSE);
8769 register int op = *scode;
8770
8771 /* If we are at the start of a conditional assertion group, *both* the
8772 conditional assertion *and* what follows the condition must satisfy the test
8773 for start of line. Other kinds of condition fail. Note that there may be an
8774 auto-callout at the start of a condition. */
8775
8776 if (op == OP_COND)
8777 {
8778 scode += 1 + LINK_SIZE;
8779 if (*scode == OP_CALLOUT) scode += PRIV(OP_lengths)[OP_CALLOUT];
8780 switch (*scode)
8781 {
8782 case OP_CREF:
8783 case OP_DNCREF:
8784 case OP_RREF:
8785 case OP_DNRREF:
8786 case OP_DEF:
8787 case OP_FAIL:
8788 return FALSE;
8789
8790 default: /* Assertion */
8791 if (!is_startline(scode, bracket_map, cd, atomcount, TRUE)) return FALSE;
8792 do scode += GET(scode, 1); while (*scode == OP_ALT);
8793 scode += 1 + LINK_SIZE;
8794 break;
8795 }
8796 scode = first_significant_code(scode, FALSE);
8797 op = *scode;
8798 }
8799
8800 /* Non-capturing brackets */
8801
8802 if (op == OP_BRA || op == OP_BRAPOS ||
8803 op == OP_SBRA || op == OP_SBRAPOS)
8804 {
8805 if (!is_startline(scode, bracket_map, cd, atomcount, inassert)) return FALSE;
8806 }
8807
8808 /* Capturing brackets */
8809
8810 else if (op == OP_CBRA || op == OP_CBRAPOS ||
8811 op == OP_SCBRA || op == OP_SCBRAPOS)
8812 {
8813 int n = GET2(scode, 1+LINK_SIZE);
8814 int new_map = bracket_map | ((n < 32)? (1U << n) : 1);
8815 if (!is_startline(scode, new_map, cd, atomcount, inassert)) return FALSE;
8816 }
8817
8818 /* Positive forward assertions */
8819
8820 else if (op == OP_ASSERT)
8821 {
8822 if (!is_startline(scode, bracket_map, cd, atomcount, TRUE)) return FALSE;
8823 }
8824
8825 /* Atomic brackets */
8826
8827 else if (op == OP_ONCE || op == OP_ONCE_NC)
8828 {
8829 if (!is_startline(scode, bracket_map, cd, atomcount + 1, inassert)) return FALSE;
8830 }
8831
8832 /* .* means "start at start or after \n" if it isn't in atomic brackets or
8833 brackets that may be referenced or an assertion, as long as the pattern does
8834 not contain *PRUNE or *SKIP, because these break the feature. Consider, for
8835 example, /.*?a(*PRUNE)b/ with the subject "aab", which matches "ab", i.e.
8836 not at the start of a line. */
8837
8838 else if (op == OP_TYPESTAR || op == OP_TYPEMINSTAR || op == OP_TYPEPOSSTAR)
8839 {
8840 if (scode[1] != OP_ANY || (bracket_map & cd->backref_map) != 0 ||
8841 atomcount > 0 || cd->had_pruneorskip || inassert)
8842 return FALSE;
8843 }
8844
8845 /* Check for explicit circumflex; anything else gives a FALSE result. Note
8846 in particular that this includes atomic brackets OP_ONCE and OP_ONCE_NC
8847 because the number of characters matched by .* cannot be adjusted inside
8848 them. */
8849
8850 else if (op != OP_CIRC && op != OP_CIRCM) return FALSE;
8851
8852 /* Move on to the next alternative */
8853
8854 code += GET(code, 1);
8855 }
8856 while (*code == OP_ALT); /* Loop for each alternative */
8857 return TRUE;
8858 }
8859
8860
8861
8862 /*************************************************
8863 * Check for asserted fixed first char *
8864 *************************************************/
8865
8866 /* During compilation, the "first char" settings from forward assertions are
8867 discarded, because they can cause conflicts with actual literals that follow.
8868 However, if we end up without a first char setting for an unanchored pattern,
8869 it is worth scanning the regex to see if there is an initial asserted first
8870 char. If all branches start with the same asserted char, or with a
8871 non-conditional bracket all of whose alternatives start with the same asserted
8872 char (recurse ad lib), then we return that char, with the flags set to zero or
8873 REQ_CASELESS; otherwise return zero with REQ_NONE in the flags.
8874
8875 Arguments:
8876 code points to start of expression (the bracket)
8877 flags points to the first char flags, or to REQ_NONE
8878 inassert TRUE if in an assertion
8879
8880 Returns: the fixed first char, or 0 with REQ_NONE in flags
8881 */
8882
8883 static pcre_uint32
find_firstassertedchar(const pcre_uchar * code,pcre_int32 * flags,BOOL inassert)8884 find_firstassertedchar(const pcre_uchar *code, pcre_int32 *flags,
8885 BOOL inassert)
8886 {
8887 register pcre_uint32 c = 0;
8888 int cflags = REQ_NONE;
8889
8890 *flags = REQ_NONE;
8891 do {
8892 pcre_uint32 d;
8893 int dflags;
8894 int xl = (*code == OP_CBRA || *code == OP_SCBRA ||
8895 *code == OP_CBRAPOS || *code == OP_SCBRAPOS)? IMM2_SIZE:0;
8896 const pcre_uchar *scode = first_significant_code(code + 1+LINK_SIZE + xl,
8897 TRUE);
8898 register pcre_uchar op = *scode;
8899
8900 switch(op)
8901 {
8902 default:
8903 return 0;
8904
8905 case OP_BRA:
8906 case OP_BRAPOS:
8907 case OP_CBRA:
8908 case OP_SCBRA:
8909 case OP_CBRAPOS:
8910 case OP_SCBRAPOS:
8911 case OP_ASSERT:
8912 case OP_ONCE:
8913 case OP_ONCE_NC:
8914 d = find_firstassertedchar(scode, &dflags, op == OP_ASSERT);
8915 if (dflags < 0)
8916 return 0;
8917 if (cflags < 0) { c = d; cflags = dflags; } else if (c != d || cflags != dflags) return 0;
8918 break;
8919
8920 case OP_EXACT:
8921 scode += IMM2_SIZE;
8922 /* Fall through */
8923
8924 case OP_CHAR:
8925 case OP_PLUS:
8926 case OP_MINPLUS:
8927 case OP_POSPLUS:
8928 if (!inassert) return 0;
8929 if (cflags < 0) { c = scode[1]; cflags = 0; }
8930 else if (c != scode[1]) return 0;
8931 break;
8932
8933 case OP_EXACTI:
8934 scode += IMM2_SIZE;
8935 /* Fall through */
8936
8937 case OP_CHARI:
8938 case OP_PLUSI:
8939 case OP_MINPLUSI:
8940 case OP_POSPLUSI:
8941 if (!inassert) return 0;
8942 if (cflags < 0) { c = scode[1]; cflags = REQ_CASELESS; }
8943 else if (c != scode[1]) return 0;
8944 break;
8945 }
8946
8947 code += GET(code, 1);
8948 }
8949 while (*code == OP_ALT);
8950
8951 *flags = cflags;
8952 return c;
8953 }
8954
8955
8956
8957 /*************************************************
8958 * Add an entry to the name/number table *
8959 *************************************************/
8960
8961 /* This function is called between compiling passes to add an entry to the
8962 name/number table, maintaining alphabetical order. Checking for permitted
8963 and forbidden duplicates has already been done.
8964
8965 Arguments:
8966 cd the compile data block
8967 name the name to add
8968 length the length of the name
8969 groupno the group number
8970
8971 Returns: nothing
8972 */
8973
8974 static void
add_name(compile_data * cd,const pcre_uchar * name,int length,unsigned int groupno)8975 add_name(compile_data *cd, const pcre_uchar *name, int length,
8976 unsigned int groupno)
8977 {
8978 int i;
8979 pcre_uchar *slot = cd->name_table;
8980
8981 for (i = 0; i < cd->names_found; i++)
8982 {
8983 int crc = memcmp(name, slot+IMM2_SIZE, IN_UCHARS(length));
8984 if (crc == 0 && slot[IMM2_SIZE+length] != 0)
8985 crc = -1; /* Current name is a substring */
8986
8987 /* Make space in the table and break the loop for an earlier name. For a
8988 duplicate or later name, carry on. We do this for duplicates so that in the
8989 simple case (when ?(| is not used) they are in order of their numbers. In all
8990 cases they are in the order in which they appear in the pattern. */
8991
8992 if (crc < 0)
8993 {
8994 memmove(slot + cd->name_entry_size, slot,
8995 IN_UCHARS((cd->names_found - i) * cd->name_entry_size));
8996 break;
8997 }
8998
8999 /* Continue the loop for a later or duplicate name */
9000
9001 slot += cd->name_entry_size;
9002 }
9003
9004 PUT2(slot, 0, groupno);
9005 memcpy(slot + IMM2_SIZE, name, IN_UCHARS(length));
9006 slot[IMM2_SIZE + length] = 0;
9007 cd->names_found++;
9008 }
9009
9010
9011
9012 /*************************************************
9013 * Compile a Regular Expression *
9014 *************************************************/
9015
9016 /* This function takes a string and returns a pointer to a block of store
9017 holding a compiled version of the expression. The original API for this
9018 function had no error code return variable; it is retained for backwards
9019 compatibility. The new function is given a new name.
9020
9021 Arguments:
9022 pattern the regular expression
9023 options various option bits
9024 errorcodeptr pointer to error code variable (pcre_compile2() only)
9025 can be NULL if you don't want a code value
9026 errorptr pointer to pointer to error text
9027 erroroffset ptr offset in pattern where error was detected
9028 tables pointer to character tables or NULL
9029
9030 Returns: pointer to compiled data block, or NULL on error,
9031 with errorptr and erroroffset set
9032 */
9033
9034 #if defined COMPILE_PCRE8
9035 #if defined(ERLANG_INTEGRATION)
9036 PCRE_EXP_DEFN pcre * PCRE_CALL_CONVENTION
erts_pcre_compile(const char * pattern,int options,const char ** errorptr,int * erroroffset,const unsigned char * tables)9037 erts_pcre_compile(const char *pattern, int options, const char **errorptr,
9038 int *erroroffset, const unsigned char *tables)
9039 #else
9040 PCRE_EXP_DEFN pcre * PCRE_CALL_CONVENTION
9041 pcre_compile(const char *pattern, int options, const char **errorptr,
9042 int *erroroffset, const unsigned char *tables)
9043 #endif
9044 #elif defined COMPILE_PCRE16
9045 PCRE_EXP_DEFN pcre16 * PCRE_CALL_CONVENTION
9046 pcre16_compile(PCRE_SPTR16 pattern, int options, const char **errorptr,
9047 int *erroroffset, const unsigned char *tables)
9048 #elif defined COMPILE_PCRE32
9049 PCRE_EXP_DEFN pcre32 * PCRE_CALL_CONVENTION
9050 pcre32_compile(PCRE_SPTR32 pattern, int options, const char **errorptr,
9051 int *erroroffset, const unsigned char *tables)
9052 #endif
9053 {
9054 #if defined COMPILE_PCRE8
9055 #if defined(ERLANG_INTEGRATION)
9056 return erts_pcre_compile2(pattern, options, NULL, errorptr,
9057 erroroffset, tables);
9058 #else
9059 return pcre_compile2(pattern, options, NULL, errorptr, erroroffset, tables);
9060 #endif
9061 #elif defined COMPILE_PCRE16
9062 return pcre16_compile2(pattern, options, NULL, errorptr, erroroffset, tables);
9063 #elif defined COMPILE_PCRE32
9064 return pcre32_compile2(pattern, options, NULL, errorptr, erroroffset, tables);
9065 #endif
9066 }
9067
9068
9069 #if defined COMPILE_PCRE8
9070 #if defined(ERLANG_INTEGRATION)
9071 PCRE_EXP_DEFN pcre * PCRE_CALL_CONVENTION
erts_pcre_compile2(const char * pattern,int options,int * errorcodeptr,const char ** errorptr,int * erroroffset,const unsigned char * tables)9072 erts_pcre_compile2(const char *pattern, int options, int *errorcodeptr,
9073 const char **errorptr, int *erroroffset, const unsigned char *tables)
9074 #else
9075 PCRE_EXP_DEFN pcre * PCRE_CALL_CONVENTION
9076 pcre_compile2(const char *pattern, int options, int *errorcodeptr,
9077 const char **errorptr, int *erroroffset, const unsigned char *tables)
9078 #endif
9079 #elif defined COMPILE_PCRE16
9080 PCRE_EXP_DEFN pcre16 * PCRE_CALL_CONVENTION
9081 pcre16_compile2(PCRE_SPTR16 pattern, int options, int *errorcodeptr,
9082 const char **errorptr, int *erroroffset, const unsigned char *tables)
9083 #elif defined COMPILE_PCRE32
9084 PCRE_EXP_DEFN pcre32 * PCRE_CALL_CONVENTION
9085 pcre32_compile2(PCRE_SPTR32 pattern, int options, int *errorcodeptr,
9086 const char **errorptr, int *erroroffset, const unsigned char *tables)
9087 #endif
9088 {
9089 REAL_PCRE *re;
9090 int length = 1; /* For final END opcode */
9091 pcre_int32 firstcharflags, reqcharflags;
9092 pcre_uint32 firstchar, reqchar;
9093 pcre_uint32 limit_match = PCRE_UINT32_MAX;
9094 pcre_uint32 limit_recursion = PCRE_UINT32_MAX;
9095 int newline;
9096 int errorcode = 0;
9097 int skipatstart = 0;
9098 BOOL utf;
9099 BOOL never_utf = FALSE;
9100 size_t size;
9101 pcre_uchar *code;
9102 const pcre_uchar *codestart;
9103 const pcre_uchar *ptr;
9104 compile_data compile_block;
9105 compile_data *cd = &compile_block;
9106
9107 /* This space is used for "compiling" into during the first phase, when we are
9108 computing the amount of memory that is needed. Compiled items are thrown away
9109 as soon as possible, so that a fairly large buffer should be sufficient for
9110 this purpose. The same space is used in the second phase for remembering where
9111 to fill in forward references to subpatterns. That may overflow, in which case
9112 new memory is obtained from malloc(). */
9113
9114 pcre_uchar cworkspace[COMPILE_WORK_SIZE];
9115
9116 /* This vector is used for remembering name groups during the pre-compile. In a
9117 similar way to cworkspace, it can be expanded using malloc() if necessary. */
9118
9119 named_group named_groups[NAMED_GROUP_LIST_SIZE];
9120
9121 /* Set this early so that early errors get offset 0. */
9122
9123 ptr = (const pcre_uchar *)pattern;
9124
9125 /* We can't pass back an error message if errorptr is NULL; I guess the best we
9126 can do is just return NULL, but we can set a code value if there is a code
9127 pointer. */
9128
9129 if (errorptr == NULL)
9130 {
9131 if (errorcodeptr != NULL) *errorcodeptr = 99;
9132 return NULL;
9133 }
9134
9135 *errorptr = NULL;
9136 if (errorcodeptr != NULL) *errorcodeptr = ERR0;
9137
9138 /* However, we can give a message for this error */
9139
9140 if (erroroffset == NULL)
9141 {
9142 errorcode = ERR16;
9143 goto PCRE_EARLY_ERROR_RETURN2;
9144 }
9145
9146 *erroroffset = 0;
9147
9148 /* Set up pointers to the individual character tables */
9149
9150 if (tables == NULL) tables = PRIV(default_tables);
9151 cd->lcc = tables + lcc_offset;
9152 cd->fcc = tables + fcc_offset;
9153 cd->cbits = tables + cbits_offset;
9154 cd->ctypes = tables + ctypes_offset;
9155
9156 /* Check that all undefined public option bits are zero */
9157
9158 if ((options & ~PUBLIC_COMPILE_OPTIONS) != 0)
9159 {
9160 errorcode = ERR17;
9161 goto PCRE_EARLY_ERROR_RETURN;
9162 }
9163
9164 /* If PCRE_NEVER_UTF is set, remember it. */
9165
9166 if ((options & PCRE_NEVER_UTF) != 0) never_utf = TRUE;
9167
9168 /* Check for global one-time settings at the start of the pattern, and remember
9169 the offset for later. */
9170
9171 cd->external_flags = 0; /* Initialize here for LIMIT_MATCH/RECURSION */
9172
9173 while (ptr[skipatstart] == CHAR_LEFT_PARENTHESIS &&
9174 ptr[skipatstart+1] == CHAR_ASTERISK)
9175 {
9176 int newnl = 0;
9177 int newbsr = 0;
9178
9179 /* For completeness and backward compatibility, (*UTFn) is supported in the
9180 relevant libraries, but (*UTF) is generic and always supported. Note that
9181 PCRE_UTF8 == PCRE_UTF16 == PCRE_UTF32. */
9182
9183 #ifdef COMPILE_PCRE8
9184 if (STRNCMP_UC_C8(ptr+skipatstart+2, STRING_UTF8_RIGHTPAR, 5) == 0)
9185 { skipatstart += 7; options |= PCRE_UTF8; continue; }
9186 #endif
9187 #ifdef COMPILE_PCRE16
9188 if (STRNCMP_UC_C8(ptr+skipatstart+2, STRING_UTF16_RIGHTPAR, 6) == 0)
9189 { skipatstart += 8; options |= PCRE_UTF16; continue; }
9190 #endif
9191 #ifdef COMPILE_PCRE32
9192 if (STRNCMP_UC_C8(ptr+skipatstart+2, STRING_UTF32_RIGHTPAR, 6) == 0)
9193 { skipatstart += 8; options |= PCRE_UTF32; continue; }
9194 #endif
9195
9196 else if (STRNCMP_UC_C8(ptr+skipatstart+2, STRING_UTF_RIGHTPAR, 4) == 0)
9197 { skipatstart += 6; options |= PCRE_UTF8; continue; }
9198 else if (STRNCMP_UC_C8(ptr+skipatstart+2, STRING_UCP_RIGHTPAR, 4) == 0)
9199 { skipatstart += 6; options |= PCRE_UCP; continue; }
9200 else if (STRNCMP_UC_C8(ptr+skipatstart+2, STRING_NO_AUTO_POSSESS_RIGHTPAR, 16) == 0)
9201 { skipatstart += 18; options |= PCRE_NO_AUTO_POSSESS; continue; }
9202 else if (STRNCMP_UC_C8(ptr+skipatstart+2, STRING_NO_START_OPT_RIGHTPAR, 13) == 0)
9203 { skipatstart += 15; options |= PCRE_NO_START_OPTIMIZE; continue; }
9204
9205 else if (STRNCMP_UC_C8(ptr+skipatstart+2, STRING_LIMIT_MATCH_EQ, 12) == 0)
9206 {
9207 pcre_uint32 c = 0;
9208 int p = skipatstart + 14;
9209 while (isdigit(ptr[p]))
9210 {
9211 if (c > PCRE_UINT32_MAX / 10 - 1) break; /* Integer overflow */
9212 c = c*10 + ptr[p++] - CHAR_0;
9213 }
9214 if (ptr[p++] != CHAR_RIGHT_PARENTHESIS) break;
9215 if (c < limit_match)
9216 {
9217 limit_match = c;
9218 cd->external_flags |= PCRE_MLSET;
9219 }
9220 skipatstart = p;
9221 continue;
9222 }
9223
9224 else if (STRNCMP_UC_C8(ptr+skipatstart+2, STRING_LIMIT_RECURSION_EQ, 16) == 0)
9225 {
9226 pcre_uint32 c = 0;
9227 int p = skipatstart + 18;
9228 while (isdigit(ptr[p]))
9229 {
9230 if (c > PCRE_UINT32_MAX / 10 - 1) break; /* Integer overflow check */
9231 c = c*10 + ptr[p++] - CHAR_0;
9232 }
9233 if (ptr[p++] != CHAR_RIGHT_PARENTHESIS) break;
9234 if (c < limit_recursion)
9235 {
9236 limit_recursion = c;
9237 cd->external_flags |= PCRE_RLSET;
9238 }
9239 skipatstart = p;
9240 continue;
9241 }
9242
9243 if (STRNCMP_UC_C8(ptr+skipatstart+2, STRING_CR_RIGHTPAR, 3) == 0)
9244 { skipatstart += 5; newnl = PCRE_NEWLINE_CR; }
9245 else if (STRNCMP_UC_C8(ptr+skipatstart+2, STRING_LF_RIGHTPAR, 3) == 0)
9246 { skipatstart += 5; newnl = PCRE_NEWLINE_LF; }
9247 else if (STRNCMP_UC_C8(ptr+skipatstart+2, STRING_CRLF_RIGHTPAR, 5) == 0)
9248 { skipatstart += 7; newnl = PCRE_NEWLINE_CR + PCRE_NEWLINE_LF; }
9249 else if (STRNCMP_UC_C8(ptr+skipatstart+2, STRING_ANY_RIGHTPAR, 4) == 0)
9250 { skipatstart += 6; newnl = PCRE_NEWLINE_ANY; }
9251 else if (STRNCMP_UC_C8(ptr+skipatstart+2, STRING_ANYCRLF_RIGHTPAR, 8) == 0)
9252 { skipatstart += 10; newnl = PCRE_NEWLINE_ANYCRLF; }
9253
9254 else if (STRNCMP_UC_C8(ptr+skipatstart+2, STRING_BSR_ANYCRLF_RIGHTPAR, 12) == 0)
9255 { skipatstart += 14; newbsr = PCRE_BSR_ANYCRLF; }
9256 else if (STRNCMP_UC_C8(ptr+skipatstart+2, STRING_BSR_UNICODE_RIGHTPAR, 12) == 0)
9257 { skipatstart += 14; newbsr = PCRE_BSR_UNICODE; }
9258
9259 if (newnl != 0)
9260 options = (options & ~PCRE_NEWLINE_BITS) | newnl;
9261 else if (newbsr != 0)
9262 options = (options & ~(PCRE_BSR_ANYCRLF|PCRE_BSR_UNICODE)) | newbsr;
9263 else break;
9264 }
9265
9266 /* PCRE_UTF(16|32) have the same value as PCRE_UTF8. */
9267 utf = (options & PCRE_UTF8) != 0;
9268 if (utf && never_utf)
9269 {
9270 errorcode = ERR78;
9271 goto PCRE_EARLY_ERROR_RETURN2;
9272 }
9273
9274 /* Can't support UTF unless PCRE has been compiled to include the code. The
9275 return of an error code from PRIV(valid_utf)() is a new feature, introduced in
9276 release 8.13. It is passed back from pcre_[dfa_]exec(), but at the moment is
9277 not used here. */
9278
9279 #ifdef SUPPORT_UTF
9280 if (utf && (options & PCRE_NO_UTF8_CHECK) == 0 &&
9281 (errorcode = PRIV(valid_utf)((PCRE_PUCHAR)pattern, -1, erroroffset)) != 0)
9282 {
9283 #if defined COMPILE_PCRE8
9284 errorcode = ERR44;
9285 #elif defined COMPILE_PCRE16
9286 errorcode = ERR74;
9287 #elif defined COMPILE_PCRE32
9288 errorcode = ERR77;
9289 #endif
9290 goto PCRE_EARLY_ERROR_RETURN2;
9291 }
9292 #else
9293 if (utf)
9294 {
9295 errorcode = ERR32;
9296 goto PCRE_EARLY_ERROR_RETURN;
9297 }
9298 #endif
9299
9300 /* Can't support UCP unless PCRE has been compiled to include the code. */
9301
9302 #ifndef SUPPORT_UCP
9303 if ((options & PCRE_UCP) != 0)
9304 {
9305 errorcode = ERR67;
9306 goto PCRE_EARLY_ERROR_RETURN;
9307 }
9308 #endif
9309
9310 /* Check validity of \R options. */
9311
9312 if ((options & (PCRE_BSR_ANYCRLF|PCRE_BSR_UNICODE)) ==
9313 (PCRE_BSR_ANYCRLF|PCRE_BSR_UNICODE))
9314 {
9315 errorcode = ERR56;
9316 goto PCRE_EARLY_ERROR_RETURN;
9317 }
9318
9319 /* Handle different types of newline. The three bits give seven cases. The
9320 current code allows for fixed one- or two-byte sequences, plus "any" and
9321 "anycrlf". */
9322
9323 switch (options & PCRE_NEWLINE_BITS)
9324 {
9325 case 0: newline = NEWLINE; break; /* Build-time default */
9326 case PCRE_NEWLINE_CR: newline = CHAR_CR; break;
9327 case PCRE_NEWLINE_LF: newline = CHAR_NL; break;
9328 case PCRE_NEWLINE_CR+
9329 PCRE_NEWLINE_LF: newline = (CHAR_CR << 8) | CHAR_NL; break;
9330 case PCRE_NEWLINE_ANY: newline = -1; break;
9331 case PCRE_NEWLINE_ANYCRLF: newline = -2; break;
9332 default: errorcode = ERR56; goto PCRE_EARLY_ERROR_RETURN;
9333 }
9334
9335 if (newline == -2)
9336 {
9337 cd->nltype = NLTYPE_ANYCRLF;
9338 }
9339 else if (newline < 0)
9340 {
9341 cd->nltype = NLTYPE_ANY;
9342 }
9343 else
9344 {
9345 cd->nltype = NLTYPE_FIXED;
9346 if (newline > 255)
9347 {
9348 cd->nllen = 2;
9349 cd->nl[0] = (newline >> 8) & 255;
9350 cd->nl[1] = newline & 255;
9351 }
9352 else
9353 {
9354 cd->nllen = 1;
9355 cd->nl[0] = newline;
9356 }
9357 }
9358
9359 /* Maximum back reference and backref bitmap. The bitmap records up to 31 back
9360 references to help in deciding whether (.*) can be treated as anchored or not.
9361 */
9362
9363 cd->top_backref = 0;
9364 cd->backref_map = 0;
9365
9366 /* Reflect pattern for debugging output */
9367
9368 DPRINTF(("------------------------------------------------------------------\n"));
9369 #ifdef PCRE_DEBUG
9370 print_puchar(stdout, (PCRE_PUCHAR)pattern);
9371 #endif
9372 DPRINTF(("\n"));
9373
9374 /* Pretend to compile the pattern while actually just accumulating the length
9375 of memory required. This behaviour is triggered by passing a non-NULL final
9376 argument to compile_regex(). We pass a block of workspace (cworkspace) for it
9377 to compile parts of the pattern into; the compiled code is discarded when it is
9378 no longer needed, so hopefully this workspace will never overflow, though there
9379 is a test for its doing so. */
9380
9381 cd->bracount = cd->final_bracount = 0;
9382 cd->names_found = 0;
9383 cd->name_entry_size = 0;
9384 cd->name_table = NULL;
9385 cd->dupnames = FALSE;
9386 cd->dupgroups = FALSE;
9387 cd->namedrefcount = 0;
9388 cd->start_code = cworkspace;
9389 cd->hwm = cworkspace;
9390 cd->iscondassert = FALSE;
9391 cd->start_workspace = cworkspace;
9392 cd->workspace_size = COMPILE_WORK_SIZE;
9393 cd->named_groups = named_groups;
9394 cd->named_group_list_size = NAMED_GROUP_LIST_SIZE;
9395 cd->start_pattern = (const pcre_uchar *)pattern;
9396 cd->end_pattern = (const pcre_uchar *)(pattern + STRLEN_UC((const pcre_uchar *)pattern));
9397 cd->req_varyopt = 0;
9398 cd->parens_depth = 0;
9399 cd->assert_depth = 0;
9400 cd->max_lookbehind = 0;
9401 cd->external_options = options;
9402 cd->open_caps = NULL;
9403
9404 /* Now do the pre-compile. On error, errorcode will be set non-zero, so we
9405 don't need to look at the result of the function here. The initial options have
9406 been put into the cd block so that they can be changed if an option setting is
9407 found within the regex right at the beginning. Bringing initial option settings
9408 outside can help speed up starting point checks. */
9409
9410 ptr += skipatstart;
9411 code = cworkspace;
9412 *code = OP_BRA;
9413
9414 (void)compile_regex(cd->external_options, &code, &ptr, &errorcode, FALSE,
9415 FALSE, 0, 0, &firstchar, &firstcharflags, &reqchar, &reqcharflags, NULL,
9416 cd, &length);
9417 if (errorcode != 0) goto PCRE_EARLY_ERROR_RETURN;
9418
9419 DPRINTF(("end pre-compile: length=%d workspace=%d\n", length,
9420 (int)(cd->hwm - cworkspace)));
9421
9422 if (length > MAX_PATTERN_SIZE)
9423 {
9424 errorcode = ERR20;
9425 goto PCRE_EARLY_ERROR_RETURN;
9426 }
9427
9428 /* Compute the size of the data block for storing the compiled pattern. Integer
9429 overflow should no longer be possible because nowadays we limit the maximum
9430 value of cd->names_found and cd->name_entry_size. */
9431
9432 size = sizeof(REAL_PCRE) +
9433 (length + cd->names_found * cd->name_entry_size) * sizeof(pcre_uchar);
9434
9435 /* Get the memory. */
9436
9437 re = (REAL_PCRE *)(PUBL(malloc))(size);
9438 if (re == NULL)
9439 {
9440 errorcode = ERR21;
9441 goto PCRE_EARLY_ERROR_RETURN;
9442 }
9443
9444 /* Put in the magic number, and save the sizes, initial options, internal
9445 flags, and character table pointer. NULL is used for the default character
9446 tables. The nullpad field is at the end; it's there to help in the case when a
9447 regex compiled on a system with 4-byte pointers is run on another with 8-byte
9448 pointers. */
9449
9450 re->magic_number = MAGIC_NUMBER;
9451 re->size = (int)size;
9452 re->options = cd->external_options;
9453 re->flags = cd->external_flags;
9454 re->limit_match = limit_match;
9455 re->limit_recursion = limit_recursion;
9456 re->first_char = 0;
9457 re->req_char = 0;
9458 re->name_table_offset = sizeof(REAL_PCRE) / sizeof(pcre_uchar);
9459 re->name_entry_size = cd->name_entry_size;
9460 re->name_count = cd->names_found;
9461 re->ref_count = 0;
9462 re->tables = (tables == PRIV(default_tables))? NULL : tables;
9463 re->nullpad = NULL;
9464 #ifdef COMPILE_PCRE32
9465 re->dummy = 0;
9466 #else
9467 re->dummy1 = re->dummy2 = re->dummy3 = 0;
9468 #endif
9469
9470 /* The starting points of the name/number translation table and of the code are
9471 passed around in the compile data block. The start/end pattern and initial
9472 options are already set from the pre-compile phase, as is the name_entry_size
9473 field. Reset the bracket count and the names_found field. Also reset the hwm
9474 field; this time it's used for remembering forward references to subpatterns.
9475 */
9476
9477 cd->final_bracount = cd->bracount; /* Save for checking forward references */
9478 cd->parens_depth = 0;
9479 cd->assert_depth = 0;
9480 cd->bracount = 0;
9481 cd->max_lookbehind = 0;
9482 cd->name_table = (pcre_uchar *)re + re->name_table_offset;
9483 codestart = cd->name_table + re->name_entry_size * re->name_count;
9484 cd->start_code = codestart;
9485 cd->hwm = (pcre_uchar *)(cd->start_workspace);
9486 cd->iscondassert = FALSE;
9487 cd->req_varyopt = 0;
9488 cd->had_accept = FALSE;
9489 cd->had_pruneorskip = FALSE;
9490 cd->check_lookbehind = FALSE;
9491 cd->open_caps = NULL;
9492
9493 /* If any named groups were found, create the name/number table from the list
9494 created in the first pass. */
9495
9496 if (cd->names_found > 0)
9497 {
9498 int i = cd->names_found;
9499 named_group *ng = cd->named_groups;
9500 cd->names_found = 0;
9501 for (; i > 0; i--, ng++)
9502 add_name(cd, ng->name, ng->length, ng->number);
9503 if (cd->named_group_list_size > NAMED_GROUP_LIST_SIZE)
9504 (PUBL(free))((void *)cd->named_groups);
9505 }
9506
9507 /* Set up a starting, non-extracting bracket, then compile the expression. On
9508 error, errorcode will be set non-zero, so we don't need to look at the result
9509 of the function here. */
9510
9511 ptr = (const pcre_uchar *)pattern + skipatstart;
9512 code = (pcre_uchar *)codestart;
9513 *code = OP_BRA;
9514 (void)compile_regex(re->options, &code, &ptr, &errorcode, FALSE, FALSE, 0, 0,
9515 &firstchar, &firstcharflags, &reqchar, &reqcharflags, NULL, cd, NULL);
9516 re->top_bracket = cd->bracount;
9517 re->top_backref = cd->top_backref;
9518 re->max_lookbehind = cd->max_lookbehind;
9519 re->flags = cd->external_flags | PCRE_MODE;
9520
9521 if (cd->had_accept)
9522 {
9523 reqchar = 0; /* Must disable after (*ACCEPT) */
9524 reqcharflags = REQ_NONE;
9525 }
9526
9527 /* If not reached end of pattern on success, there's an excess bracket. */
9528
9529 if (errorcode == 0 && *ptr != CHAR_NULL) errorcode = ERR22;
9530
9531 /* Fill in the terminating state and check for disastrous overflow, but
9532 if debugging, leave the test till after things are printed out. */
9533
9534 *code++ = OP_END;
9535
9536 #ifndef PCRE_DEBUG
9537 if (code - codestart > length) errorcode = ERR23;
9538 #endif
9539
9540 #ifdef SUPPORT_VALGRIND
9541 /* If the estimated length exceeds the really used length, mark the extra
9542 allocated memory as unaddressable, so that any out-of-bound reads can be
9543 detected. */
9544 VALGRIND_MAKE_MEM_NOACCESS(code, (length - (code - codestart)) * sizeof(pcre_uchar));
9545 #endif
9546
9547 /* Fill in any forward references that are required. There may be repeated
9548 references; optimize for them, as searching a large regex takes time. */
9549
9550 if (cd->hwm > cd->start_workspace)
9551 {
9552 int prev_recno = -1;
9553 const pcre_uchar *groupptr = NULL;
9554 while (errorcode == 0 && cd->hwm > cd->start_workspace)
9555 {
9556 int offset, recno;
9557 cd->hwm -= LINK_SIZE;
9558 offset = GET(cd->hwm, 0);
9559
9560 /* Check that the hwm handling hasn't gone wrong. This whole area is
9561 rewritten in PCRE2 because there are some obscure cases. */
9562
9563 if (offset == 0 || codestart[offset-1] != OP_RECURSE)
9564 {
9565 errorcode = ERR10;
9566 break;
9567 }
9568
9569 recno = GET(codestart, offset);
9570 if (recno != prev_recno)
9571 {
9572 groupptr = PRIV(find_bracket)(codestart, utf, recno);
9573 prev_recno = recno;
9574 }
9575 if (groupptr == NULL) errorcode = ERR53;
9576 else PUT(((pcre_uchar *)codestart), offset, (int)(groupptr - codestart));
9577 }
9578 }
9579
9580 /* If the workspace had to be expanded, free the new memory. Set the pointer to
9581 NULL to indicate that forward references have been filled in. */
9582
9583 if (cd->workspace_size > COMPILE_WORK_SIZE)
9584 (PUBL(free))((void *)cd->start_workspace);
9585 cd->start_workspace = NULL;
9586
9587 /* Give an error if there's back reference to a non-existent capturing
9588 subpattern. */
9589
9590 if (errorcode == 0 && re->top_backref > re->top_bracket) errorcode = ERR15;
9591
9592 /* Unless disabled, check whether any single character iterators can be
9593 auto-possessified. The function overwrites the appropriate opcode values, so
9594 the type of the pointer must be cast. NOTE: the intermediate variable "temp" is
9595 used in this code because at least one compiler gives a warning about loss of
9596 "const" attribute if the cast (pcre_uchar *)codestart is used directly in the
9597 function call. */
9598
9599 if (errorcode == 0 && (options & PCRE_NO_AUTO_POSSESS) == 0)
9600 {
9601 pcre_uchar *temp = (pcre_uchar *)codestart;
9602 auto_possessify(temp, utf, cd);
9603 }
9604
9605 /* If there were any lookbehind assertions that contained OP_RECURSE
9606 (recursions or subroutine calls), a flag is set for them to be checked here,
9607 because they may contain forward references. Actual recursions cannot be fixed
9608 length, but subroutine calls can. It is done like this so that those without
9609 OP_RECURSE that are not fixed length get a diagnosic with a useful offset. The
9610 exceptional ones forgo this. We scan the pattern to check that they are fixed
9611 length, and set their lengths. */
9612
9613 if (errorcode == 0 && cd->check_lookbehind)
9614 {
9615 pcre_uchar *cc = (pcre_uchar *)codestart;
9616
9617 /* Loop, searching for OP_REVERSE items, and process those that do not have
9618 their length set. (Actually, it will also re-process any that have a length
9619 of zero, but that is a pathological case, and it does no harm.) When we find
9620 one, we temporarily terminate the branch it is in while we scan it. */
9621
9622 for (cc = (pcre_uchar *)PRIV(find_bracket)(codestart, utf, -1);
9623 cc != NULL;
9624 cc = (pcre_uchar *)PRIV(find_bracket)(cc, utf, -1))
9625 {
9626 if (GET(cc, 1) == 0)
9627 {
9628 int fixed_length;
9629 pcre_uchar *be = cc - 1 - LINK_SIZE + GET(cc, -LINK_SIZE);
9630 int end_op = *be;
9631 *be = OP_END;
9632 fixed_length = find_fixedlength(cc, (re->options & PCRE_UTF8) != 0, TRUE,
9633 cd, NULL);
9634 *be = end_op;
9635 DPRINTF(("fixed length = %d\n", fixed_length));
9636 if (fixed_length < 0)
9637 {
9638 errorcode = (fixed_length == -2)? ERR36 :
9639 (fixed_length == -4)? ERR70 : ERR25;
9640 break;
9641 }
9642 if (fixed_length > cd->max_lookbehind) cd->max_lookbehind = fixed_length;
9643 PUT(cc, 1, fixed_length);
9644 }
9645 cc += 1 + LINK_SIZE;
9646 }
9647 }
9648
9649 /* Failed to compile, or error while post-processing */
9650
9651 if (errorcode != 0)
9652 {
9653 (PUBL(free))(re);
9654 PCRE_EARLY_ERROR_RETURN:
9655 *erroroffset = (int)(ptr - (const pcre_uchar *)pattern);
9656 PCRE_EARLY_ERROR_RETURN2:
9657 *errorptr = find_error_text(errorcode);
9658 if (errorcodeptr != NULL) *errorcodeptr = errorcode;
9659 return NULL;
9660 }
9661
9662 /* If the anchored option was not passed, set the flag if we can determine that
9663 the pattern is anchored by virtue of ^ characters or \A or anything else, such
9664 as starting with non-atomic .* when DOTALL is set and there are no occurrences
9665 of *PRUNE or *SKIP.
9666
9667 Otherwise, if we know what the first byte has to be, save it, because that
9668 speeds up unanchored matches no end. If not, see if we can set the
9669 PCRE_STARTLINE flag. This is helpful for multiline matches when all branches
9670 start with ^. and also when all branches start with non-atomic .* for
9671 non-DOTALL matches when *PRUNE and SKIP are not present. */
9672
9673 if ((re->options & PCRE_ANCHORED) == 0)
9674 {
9675 if (is_anchored(codestart, 0, cd, 0)) re->options |= PCRE_ANCHORED;
9676 else
9677 {
9678 if (firstcharflags < 0)
9679 firstchar = find_firstassertedchar(codestart, &firstcharflags, FALSE);
9680 if (firstcharflags >= 0) /* Remove caseless flag for non-caseable chars */
9681 {
9682 #if defined COMPILE_PCRE8
9683 re->first_char = firstchar & 0xff;
9684 #elif defined COMPILE_PCRE16
9685 re->first_char = firstchar & 0xffff;
9686 #elif defined COMPILE_PCRE32
9687 re->first_char = firstchar;
9688 #endif
9689 if ((firstcharflags & REQ_CASELESS) != 0)
9690 {
9691 #if defined SUPPORT_UCP && !(defined COMPILE_PCRE8)
9692 /* We ignore non-ASCII first chars in 8 bit mode. */
9693 if (utf)
9694 {
9695 if (re->first_char < 128)
9696 {
9697 if (cd->fcc[re->first_char] != re->first_char)
9698 re->flags |= PCRE_FCH_CASELESS;
9699 }
9700 else if (UCD_OTHERCASE(re->first_char) != re->first_char)
9701 re->flags |= PCRE_FCH_CASELESS;
9702 }
9703 else
9704 #endif
9705 if (MAX_255(re->first_char)
9706 && cd->fcc[re->first_char] != re->first_char)
9707 re->flags |= PCRE_FCH_CASELESS;
9708 }
9709
9710 re->flags |= PCRE_FIRSTSET;
9711 }
9712
9713 else if (is_startline(codestart, 0, cd, 0, FALSE)) re->flags |= PCRE_STARTLINE;
9714 }
9715 }
9716
9717 /* For an anchored pattern, we use the "required byte" only if it follows a
9718 variable length item in the regex. Remove the caseless flag for non-caseable
9719 bytes. */
9720
9721 if (reqcharflags >= 0 &&
9722 ((re->options & PCRE_ANCHORED) == 0 || (reqcharflags & REQ_VARY) != 0))
9723 {
9724 #if defined COMPILE_PCRE8
9725 re->req_char = reqchar & 0xff;
9726 #elif defined COMPILE_PCRE16
9727 re->req_char = reqchar & 0xffff;
9728 #elif defined COMPILE_PCRE32
9729 re->req_char = reqchar;
9730 #endif
9731 if ((reqcharflags & REQ_CASELESS) != 0)
9732 {
9733 #if defined SUPPORT_UCP && !(defined COMPILE_PCRE8)
9734 /* We ignore non-ASCII first chars in 8 bit mode. */
9735 if (utf)
9736 {
9737 if (re->req_char < 128)
9738 {
9739 if (cd->fcc[re->req_char] != re->req_char)
9740 re->flags |= PCRE_RCH_CASELESS;
9741 }
9742 else if (UCD_OTHERCASE(re->req_char) != re->req_char)
9743 re->flags |= PCRE_RCH_CASELESS;
9744 }
9745 else
9746 #endif
9747 if (MAX_255(re->req_char) && cd->fcc[re->req_char] != re->req_char)
9748 re->flags |= PCRE_RCH_CASELESS;
9749 }
9750
9751 re->flags |= PCRE_REQCHSET;
9752 }
9753
9754 /* Print out the compiled data if debugging is enabled. This is never the
9755 case when building a production library. */
9756
9757 #ifdef PCRE_DEBUG
9758 printf("Length = %d top_bracket = %d top_backref = %d\n",
9759 length, re->top_bracket, re->top_backref);
9760
9761 printf("Options=%08x\n", re->options);
9762
9763 if ((re->flags & PCRE_FIRSTSET) != 0)
9764 {
9765 pcre_uchar ch = re->first_char;
9766 const char *caseless =
9767 ((re->flags & PCRE_FCH_CASELESS) == 0)? "" : " (caseless)";
9768 if (PRINTABLE(ch)) printf("First char = %c%s\n", ch, caseless);
9769 else printf("First char = \\x%02x%s\n", ch, caseless);
9770 }
9771
9772 if ((re->flags & PCRE_REQCHSET) != 0)
9773 {
9774 pcre_uchar ch = re->req_char;
9775 const char *caseless =
9776 ((re->flags & PCRE_RCH_CASELESS) == 0)? "" : " (caseless)";
9777 if (PRINTABLE(ch)) printf("Req char = %c%s\n", ch, caseless);
9778 else printf("Req char = \\x%02x%s\n", ch, caseless);
9779 }
9780
9781 #if defined COMPILE_PCRE8
9782 pcre_printint((pcre *)re, stdout, TRUE);
9783 #elif defined COMPILE_PCRE16
9784 pcre16_printint((pcre *)re, stdout, TRUE);
9785 #elif defined COMPILE_PCRE32
9786 pcre32_printint((pcre *)re, stdout, TRUE);
9787 #endif
9788
9789 /* This check is done here in the debugging case so that the code that
9790 was compiled can be seen. */
9791
9792 if (code - codestart > length)
9793 {
9794 (PUBL(free))(re);
9795 *errorptr = find_error_text(ERR23);
9796 *erroroffset = ptr - (pcre_uchar *)pattern;
9797 if (errorcodeptr != NULL) *errorcodeptr = ERR23;
9798 return NULL;
9799 }
9800 #endif /* PCRE_DEBUG */
9801
9802 /* Check for a pattern than can match an empty string, so that this information
9803 can be provided to applications. */
9804
9805 do
9806 {
9807 if (could_be_empty_branch(codestart, code, utf, cd, NULL))
9808 {
9809 re->flags |= PCRE_MATCH_EMPTY;
9810 break;
9811 }
9812 codestart += GET(codestart, 1);
9813 }
9814 while (*codestart == OP_ALT);
9815
9816 #if defined COMPILE_PCRE8
9817 return (pcre *)re;
9818 #elif defined COMPILE_PCRE16
9819 return (pcre16 *)re;
9820 #elif defined COMPILE_PCRE32
9821 return (pcre32 *)re;
9822 #endif
9823 }
9824
9825 /* End of pcre_compile.c */
9826