/[pcre]/code/branches/pcre16/pcre_compile.c
ViewVC logotype

Contents of /code/branches/pcre16/pcre_compile.c

Parent Directory Parent Directory | Revision Log Revision Log


Revision 812 - (show annotations)
Mon Dec 19 21:23:42 2011 UTC (7 years, 7 months ago) by zherczeg
File MIME type: text/plain
File size: 266487 byte(s)
minor fixes
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) 1997-2011 University of Cambridge
10
11 -----------------------------------------------------------------------------
12 Redistribution and use in source and binary forms, with or without
13 modification, are permitted provided that the following conditions are met:
14
15 * Redistributions of source code must retain the above copyright notice,
16 this list of conditions and the following disclaimer.
17
18 * Redistributions in binary form must reproduce the above copyright
19 notice, this list of conditions and the following disclaimer in the
20 documentation and/or other materials provided with the distribution.
21
22 * Neither the name of the University of Cambridge nor the names of its
23 contributors may be used to endorse or promote products derived from
24 this software without specific prior written permission.
25
26 THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
27 AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28 IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29 ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE
30 LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
31 CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
32 SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
33 INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
34 CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
35 ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
36 POSSIBILITY OF SUCH DAMAGE.
37 -----------------------------------------------------------------------------
38 */
39
40
41 /* This module contains the external function pcre_compile(), along with
42 supporting internal functions that are not used by other modules. */
43
44
45 #ifdef HAVE_CONFIG_H
46 #include "config.h"
47 #endif
48
49 #define NLBLOCK cd /* Block containing newline information */
50 #define PSSTART start_pattern /* Field containing processed string start */
51 #define PSEND end_pattern /* Field containing processed string end */
52
53 #include "pcre_internal.h"
54
55
56 /* When PCRE_DEBUG is defined, we need the pcre(16)_printint() function, which
57 is also used by pcretest. PCRE_DEBUG is not defined when building a production
58 library. We do not need to select pcre16_printint.c specially, because the
59 COMPILE_PCREx macro will already be appropriately set. */
60
61 #ifdef PCRE_DEBUG
62 /* pcre_printint.c should not include any headers */
63 #define PCRE_INCLUDED
64 #include "pcre_printint.c"
65 #undef PCRE_INCLUDED
66 #endif
67
68
69 /* Macro for setting individual bits in class bitmaps. */
70
71 #define SETBIT(a,b) a[b/8] |= (1 << (b%8))
72
73 /* Maximum length value to check against when making sure that the integer that
74 holds the compiled pattern length does not overflow. We make it a bit less than
75 INT_MAX to allow for adding in group terminating bytes, so that we don't have
76 to check them every time. */
77
78 #define OFLOW_MAX (INT_MAX - 20)
79
80
81 /*************************************************
82 * Code parameters and static tables *
83 *************************************************/
84
85 /* This value specifies the size of stack workspace that is used during the
86 first pre-compile phase that determines how much memory is required. The regex
87 is partly compiled into this space, but the compiled parts are discarded as
88 soon as they can be, so that hopefully there will never be an overrun. The code
89 does, however, check for an overrun. The largest amount I've seen used is 218,
90 so this number is very generous.
91
92 The same workspace is used during the second, actual compile phase for
93 remembering forward references to groups so that they can be filled in at the
94 end. Each entry in this list occupies LINK_SIZE bytes, so even when LINK_SIZE
95 is 4 there is plenty of room for most patterns. However, the memory can get
96 filled up by repetitions of forward references, for example patterns like
97 /(?1){0,1999}(b)/, and one user did hit the limit. The code has been changed so
98 that the workspace is expanded using malloc() in this situation. The value
99 below is therefore a minimum, and we put a maximum on it for safety. The
100 minimum is now also defined in terms of LINK_SIZE so that the use of malloc()
101 kicks in at the same number of forward references in all cases. */
102
103 #define COMPILE_WORK_SIZE (2048*LINK_SIZE)
104 #define COMPILE_WORK_SIZE_MAX (100*COMPILE_WORK_SIZE)
105
106 /* The overrun tests check for a slightly smaller size so that they detect the
107 overrun before it actually does run off the end of the data block. */
108
109 #define WORK_SIZE_SAFETY_MARGIN (100)
110
111 /* Private flags added to firstchar and reqchar. */
112
113 #define REQ_CASELESS 0x10000000l /* Indicates caselessness */
114 #define REQ_VARY 0x20000000l /* Reqchar followed non-literal item */
115
116 /* Repeated character flags. */
117
118 #define UTF_LENGTH 0x10000000l /* The char contains its length. */
119
120 /* Table for handling escaped characters in the range '0'-'z'. Positive returns
121 are simple data values; negative values are for special things like \d and so
122 on. Zero means further processing is needed (for things like \x), or the escape
123 is invalid. */
124
125 #ifndef EBCDIC
126
127 /* This is the "normal" table for ASCII systems or for EBCDIC systems running
128 in UTF-8 mode. */
129
130 static const short int escapes[] = {
131 0, 0,
132 0, 0,
133 0, 0,
134 0, 0,
135 0, 0,
136 CHAR_COLON, CHAR_SEMICOLON,
137 CHAR_LESS_THAN_SIGN, CHAR_EQUALS_SIGN,
138 CHAR_GREATER_THAN_SIGN, CHAR_QUESTION_MARK,
139 CHAR_COMMERCIAL_AT, -ESC_A,
140 -ESC_B, -ESC_C,
141 -ESC_D, -ESC_E,
142 0, -ESC_G,
143 -ESC_H, 0,
144 0, -ESC_K,
145 0, 0,
146 -ESC_N, 0,
147 -ESC_P, -ESC_Q,
148 -ESC_R, -ESC_S,
149 0, 0,
150 -ESC_V, -ESC_W,
151 -ESC_X, 0,
152 -ESC_Z, CHAR_LEFT_SQUARE_BRACKET,
153 CHAR_BACKSLASH, CHAR_RIGHT_SQUARE_BRACKET,
154 CHAR_CIRCUMFLEX_ACCENT, CHAR_UNDERSCORE,
155 CHAR_GRAVE_ACCENT, 7,
156 -ESC_b, 0,
157 -ESC_d, ESC_e,
158 ESC_f, 0,
159 -ESC_h, 0,
160 0, -ESC_k,
161 0, 0,
162 ESC_n, 0,
163 -ESC_p, 0,
164 ESC_r, -ESC_s,
165 ESC_tee, 0,
166 -ESC_v, -ESC_w,
167 0, 0,
168 -ESC_z
169 };
170
171 #else
172
173 /* This is the "abnormal" table for EBCDIC systems without UTF-8 support. */
174
175 static const short int escapes[] = {
176 /* 48 */ 0, 0, 0, '.', '<', '(', '+', '|',
177 /* 50 */ '&', 0, 0, 0, 0, 0, 0, 0,
178 /* 58 */ 0, 0, '!', '$', '*', ')', ';', '~',
179 /* 60 */ '-', '/', 0, 0, 0, 0, 0, 0,
180 /* 68 */ 0, 0, '|', ',', '%', '_', '>', '?',
181 /* 70 */ 0, 0, 0, 0, 0, 0, 0, 0,
182 /* 78 */ 0, '`', ':', '#', '@', '\'', '=', '"',
183 /* 80 */ 0, 7, -ESC_b, 0, -ESC_d, ESC_e, ESC_f, 0,
184 /* 88 */-ESC_h, 0, 0, '{', 0, 0, 0, 0,
185 /* 90 */ 0, 0, -ESC_k, 'l', 0, ESC_n, 0, -ESC_p,
186 /* 98 */ 0, ESC_r, 0, '}', 0, 0, 0, 0,
187 /* A0 */ 0, '~', -ESC_s, ESC_tee, 0,-ESC_v, -ESC_w, 0,
188 /* A8 */ 0,-ESC_z, 0, 0, 0, '[', 0, 0,
189 /* B0 */ 0, 0, 0, 0, 0, 0, 0, 0,
190 /* B8 */ 0, 0, 0, 0, 0, ']', '=', '-',
191 /* C0 */ '{',-ESC_A, -ESC_B, -ESC_C, -ESC_D,-ESC_E, 0, -ESC_G,
192 /* C8 */-ESC_H, 0, 0, 0, 0, 0, 0, 0,
193 /* D0 */ '}', 0, -ESC_K, 0, 0,-ESC_N, 0, -ESC_P,
194 /* D8 */-ESC_Q,-ESC_R, 0, 0, 0, 0, 0, 0,
195 /* E0 */ '\\', 0, -ESC_S, 0, 0,-ESC_V, -ESC_W, -ESC_X,
196 /* E8 */ 0,-ESC_Z, 0, 0, 0, 0, 0, 0,
197 /* F0 */ 0, 0, 0, 0, 0, 0, 0, 0,
198 /* F8 */ 0, 0, 0, 0, 0, 0, 0, 0
199 };
200 #endif
201
202
203 /* Table of special "verbs" like (*PRUNE). This is a short table, so it is
204 searched linearly. Put all the names into a single string, in order to reduce
205 the number of relocations when a shared library is dynamically linked. The
206 string is built from string macros so that it works in UTF-8 mode on EBCDIC
207 platforms. */
208
209 typedef struct verbitem {
210 int len; /* Length of verb name */
211 int op; /* Op when no arg, or -1 if arg mandatory */
212 int op_arg; /* Op when arg present, or -1 if not allowed */
213 } verbitem;
214
215 static const char verbnames[] =
216 "\0" /* Empty name is a shorthand for MARK */
217 STRING_MARK0
218 STRING_ACCEPT0
219 STRING_COMMIT0
220 STRING_F0
221 STRING_FAIL0
222 STRING_PRUNE0
223 STRING_SKIP0
224 STRING_THEN;
225
226 static const verbitem verbs[] = {
227 { 0, -1, OP_MARK },
228 { 4, -1, OP_MARK },
229 { 6, OP_ACCEPT, -1 },
230 { 6, OP_COMMIT, -1 },
231 { 1, OP_FAIL, -1 },
232 { 4, OP_FAIL, -1 },
233 { 5, OP_PRUNE, OP_PRUNE_ARG },
234 { 4, OP_SKIP, OP_SKIP_ARG },
235 { 4, OP_THEN, OP_THEN_ARG }
236 };
237
238 static const int verbcount = sizeof(verbs)/sizeof(verbitem);
239
240
241 /* Tables of names of POSIX character classes and their lengths. The names are
242 now all in a single string, to reduce the number of relocations when a shared
243 library is dynamically loaded. The list of lengths is terminated by a zero
244 length entry. The first three must be alpha, lower, upper, as this is assumed
245 for handling case independence. */
246
247 static const char posix_names[] =
248 STRING_alpha0 STRING_lower0 STRING_upper0 STRING_alnum0
249 STRING_ascii0 STRING_blank0 STRING_cntrl0 STRING_digit0
250 STRING_graph0 STRING_print0 STRING_punct0 STRING_space0
251 STRING_word0 STRING_xdigit;
252
253 static const pcre_uint8 posix_name_lengths[] = {
254 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 4, 6, 0 };
255
256 /* Table of class bit maps for each POSIX class. Each class is formed from a
257 base map, with an optional addition or removal of another map. Then, for some
258 classes, there is some additional tweaking: for [:blank:] the vertical space
259 characters are removed, and for [:alpha:] and [:alnum:] the underscore
260 character is removed. The triples in the table consist of the base map offset,
261 second map offset or -1 if no second map, and a non-negative value for map
262 addition or a negative value for map subtraction (if there are two maps). The
263 absolute value of the third field has these meanings: 0 => no tweaking, 1 =>
264 remove vertical space characters, 2 => remove underscore. */
265
266 static const int posix_class_maps[] = {
267 cbit_word, cbit_digit, -2, /* alpha */
268 cbit_lower, -1, 0, /* lower */
269 cbit_upper, -1, 0, /* upper */
270 cbit_word, -1, 2, /* alnum - word without underscore */
271 cbit_print, cbit_cntrl, 0, /* ascii */
272 cbit_space, -1, 1, /* blank - a GNU extension */
273 cbit_cntrl, -1, 0, /* cntrl */
274 cbit_digit, -1, 0, /* digit */
275 cbit_graph, -1, 0, /* graph */
276 cbit_print, -1, 0, /* print */
277 cbit_punct, -1, 0, /* punct */
278 cbit_space, -1, 0, /* space */
279 cbit_word, -1, 0, /* word - a Perl extension */
280 cbit_xdigit,-1, 0 /* xdigit */
281 };
282
283 /* Table of substitutes for \d etc when PCRE_UCP is set. The POSIX class
284 substitutes must be in the order of the names, defined above, and there are
285 both positive and negative cases. NULL means no substitute. */
286
287 #ifdef SUPPORT_UCP
288 static const pcre_uchar string_PNd[] = {
289 CHAR_BACKSLASH, CHAR_P, CHAR_LEFT_CURLY_BRACKET,
290 CHAR_N, CHAR_d, CHAR_RIGHT_CURLY_BRACKET, '\0' };
291 static const pcre_uchar string_pNd[] = {
292 CHAR_BACKSLASH, CHAR_p, CHAR_LEFT_CURLY_BRACKET,
293 CHAR_N, CHAR_d, CHAR_RIGHT_CURLY_BRACKET, '\0' };
294 static const pcre_uchar string_PXsp[] = {
295 CHAR_BACKSLASH, CHAR_P, CHAR_LEFT_CURLY_BRACKET,
296 CHAR_X, CHAR_s, CHAR_p, CHAR_RIGHT_CURLY_BRACKET, '\0' };
297 static const pcre_uchar string_pXsp[] = {
298 CHAR_BACKSLASH, CHAR_p, CHAR_LEFT_CURLY_BRACKET,
299 CHAR_X, CHAR_s, CHAR_p, CHAR_RIGHT_CURLY_BRACKET, '\0' };
300 static const pcre_uchar string_PXwd[] = {
301 CHAR_BACKSLASH, CHAR_P, CHAR_LEFT_CURLY_BRACKET,
302 CHAR_X, CHAR_w, CHAR_d, CHAR_RIGHT_CURLY_BRACKET, '\0' };
303 static const pcre_uchar string_pXwd[] = {
304 CHAR_BACKSLASH, CHAR_p, CHAR_LEFT_CURLY_BRACKET,
305 CHAR_X, CHAR_w, CHAR_d, CHAR_RIGHT_CURLY_BRACKET, '\0' };
306
307 static const pcre_uchar *substitutes[] = {
308 string_PNd, /* \D */
309 string_pNd, /* \d */
310 string_PXsp, /* \S */ /* NOTE: Xsp is Perl space */
311 string_pXsp, /* \s */
312 string_PXwd, /* \W */
313 string_pXwd /* \w */
314 };
315
316 static const pcre_uchar string_pL[] = {
317 CHAR_BACKSLASH, CHAR_p, CHAR_LEFT_CURLY_BRACKET,
318 CHAR_L, CHAR_RIGHT_CURLY_BRACKET, '\0' };
319 static const pcre_uchar string_pLl[] = {
320 CHAR_BACKSLASH, CHAR_p, CHAR_LEFT_CURLY_BRACKET,
321 CHAR_L, CHAR_l, CHAR_RIGHT_CURLY_BRACKET, '\0' };
322 static const pcre_uchar string_pLu[] = {
323 CHAR_BACKSLASH, CHAR_p, CHAR_LEFT_CURLY_BRACKET,
324 CHAR_L, CHAR_u, CHAR_RIGHT_CURLY_BRACKET, '\0' };
325 static const pcre_uchar string_pXan[] = {
326 CHAR_BACKSLASH, CHAR_p, CHAR_LEFT_CURLY_BRACKET,
327 CHAR_X, CHAR_a, CHAR_n, CHAR_RIGHT_CURLY_BRACKET, '\0' };
328 static const pcre_uchar string_h[] = {
329 CHAR_BACKSLASH, CHAR_h, '\0' };
330 static const pcre_uchar string_pXps[] = {
331 CHAR_BACKSLASH, CHAR_p, CHAR_LEFT_CURLY_BRACKET,
332 CHAR_X, CHAR_p, CHAR_s, CHAR_RIGHT_CURLY_BRACKET, '\0' };
333 static const pcre_uchar string_PL[] = {
334 CHAR_BACKSLASH, CHAR_P, CHAR_LEFT_CURLY_BRACKET,
335 CHAR_L, CHAR_RIGHT_CURLY_BRACKET, '\0' };
336 static const pcre_uchar string_PLl[] = {
337 CHAR_BACKSLASH, CHAR_P, CHAR_LEFT_CURLY_BRACKET,
338 CHAR_L, CHAR_l, CHAR_RIGHT_CURLY_BRACKET, '\0' };
339 static const pcre_uchar string_PLu[] = {
340 CHAR_BACKSLASH, CHAR_P, CHAR_LEFT_CURLY_BRACKET,
341 CHAR_L, CHAR_u, CHAR_RIGHT_CURLY_BRACKET, '\0' };
342 static const pcre_uchar string_PXan[] = {
343 CHAR_BACKSLASH, CHAR_P, CHAR_LEFT_CURLY_BRACKET,
344 CHAR_X, CHAR_a, CHAR_n, CHAR_RIGHT_CURLY_BRACKET, '\0' };
345 static const pcre_uchar string_H[] = {
346 CHAR_BACKSLASH, CHAR_H, '\0' };
347 static const pcre_uchar string_PXps[] = {
348 CHAR_BACKSLASH, CHAR_P, CHAR_LEFT_CURLY_BRACKET,
349 CHAR_X, CHAR_p, CHAR_s, CHAR_RIGHT_CURLY_BRACKET, '\0' };
350
351 static const pcre_uchar *posix_substitutes[] = {
352 string_pL, /* alpha */
353 string_pLl, /* lower */
354 string_pLu, /* upper */
355 string_pXan, /* alnum */
356 NULL, /* ascii */
357 string_h, /* blank */
358 NULL, /* cntrl */
359 string_pNd, /* digit */
360 NULL, /* graph */
361 NULL, /* print */
362 NULL, /* punct */
363 string_pXps, /* space */ /* NOTE: Xps is POSIX space */
364 string_pXwd, /* word */
365 NULL, /* xdigit */
366 /* Negated cases */
367 string_PL, /* ^alpha */
368 string_PLl, /* ^lower */
369 string_PLu, /* ^upper */
370 string_PXan, /* ^alnum */
371 NULL, /* ^ascii */
372 string_H, /* ^blank */
373 NULL, /* ^cntrl */
374 string_PNd, /* ^digit */
375 NULL, /* ^graph */
376 NULL, /* ^print */
377 NULL, /* ^punct */
378 string_PXps, /* ^space */ /* NOTE: Xps is POSIX space */
379 string_PXwd, /* ^word */
380 NULL /* ^xdigit */
381 };
382 #define POSIX_SUBSIZE (sizeof(posix_substitutes) / sizeof(pcre_uchar *))
383 #endif
384
385 #define STRING(a) # a
386 #define XSTRING(s) STRING(s)
387
388 /* The texts of compile-time error messages. These are "char *" because they
389 are passed to the outside world. Do not ever re-use any error number, because
390 they are documented. Always add a new error instead. Messages marked DEAD below
391 are no longer used. This used to be a table of strings, but in order to reduce
392 the number of relocations needed when a shared library is loaded dynamically,
393 it is now one long string. We cannot use a table of offsets, because the
394 lengths of inserts such as XSTRING(MAX_NAME_SIZE) are not known. Instead, we
395 simply count through to the one we want - this isn't a performance issue
396 because these strings are used only when there is a compilation error.
397
398 Each substring ends with \0 to insert a null character. This includes the final
399 substring, so that the whole string ends with \0\0, which can be detected when
400 counting through. */
401
402 static const char error_texts[] =
403 "no error\0"
404 "\\ at end of pattern\0"
405 "\\c at end of pattern\0"
406 "unrecognized character follows \\\0"
407 "numbers out of order in {} quantifier\0"
408 /* 5 */
409 "number too big in {} quantifier\0"
410 "missing terminating ] for character class\0"
411 "invalid escape sequence in character class\0"
412 "range out of order in character class\0"
413 "nothing to repeat\0"
414 /* 10 */
415 "operand of unlimited repeat could match the empty string\0" /** DEAD **/
416 "internal error: unexpected repeat\0"
417 "unrecognized character after (? or (?-\0"
418 "POSIX named classes are supported only within a class\0"
419 "missing )\0"
420 /* 15 */
421 "reference to non-existent subpattern\0"
422 "erroffset passed as NULL\0"
423 "unknown option bit(s) set\0"
424 "missing ) after comment\0"
425 "parentheses nested too deeply\0" /** DEAD **/
426 /* 20 */
427 "regular expression is too large\0"
428 "failed to get memory\0"
429 "unmatched parentheses\0"
430 "internal error: code overflow\0"
431 "unrecognized character after (?<\0"
432 /* 25 */
433 "lookbehind assertion is not fixed length\0"
434 "malformed number or name after (?(\0"
435 "conditional group contains more than two branches\0"
436 "assertion expected after (?(\0"
437 "(?R or (?[+-]digits must be followed by )\0"
438 /* 30 */
439 "unknown POSIX class name\0"
440 "POSIX collating elements are not supported\0"
441 "this version of PCRE is not compiled with PCRE_UTF8 support\0"
442 "spare error\0" /** DEAD **/
443 "character value in \\x{...} sequence is too large\0"
444 /* 35 */
445 "invalid condition (?(0)\0"
446 "\\C not allowed in lookbehind assertion\0"
447 "PCRE does not support \\L, \\l, \\N{name}, \\U, or \\u\0"
448 "number after (?C is > 255\0"
449 "closing ) for (?C expected\0"
450 /* 40 */
451 "recursive call could loop indefinitely\0"
452 "unrecognized character after (?P\0"
453 "syntax error in subpattern name (missing terminator)\0"
454 "two named subpatterns have the same name\0"
455 "invalid UTF-8 string\0"
456 /* 45 */
457 "support for \\P, \\p, and \\X has not been compiled\0"
458 "malformed \\P or \\p sequence\0"
459 "unknown property name after \\P or \\p\0"
460 "subpattern name is too long (maximum " XSTRING(MAX_NAME_SIZE) " characters)\0"
461 "too many named subpatterns (maximum " XSTRING(MAX_NAME_COUNT) ")\0"
462 /* 50 */
463 "repeated subpattern is too long\0" /** DEAD **/
464 "octal value is greater than \\377 (not in UTF-8 mode)\0"
465 "internal error: overran compiling workspace\0"
466 "internal error: previously-checked referenced subpattern not found\0"
467 "DEFINE group contains more than one branch\0"
468 /* 55 */
469 "repeating a DEFINE group is not allowed\0" /** DEAD **/
470 "inconsistent NEWLINE options\0"
471 "\\g is not followed by a braced, angle-bracketed, or quoted name/number or by a plain number\0"
472 "a numbered reference must not be zero\0"
473 "an argument is not allowed for (*ACCEPT), (*FAIL), or (*COMMIT)\0"
474 /* 60 */
475 "(*VERB) not recognized\0"
476 "number is too big\0"
477 "subpattern name expected\0"
478 "digit expected after (?+\0"
479 "] is an invalid data character in JavaScript compatibility mode\0"
480 /* 65 */
481 "different names for subpatterns of the same number are not allowed\0"
482 "(*MARK) must have an argument\0"
483 "this version of PCRE is not compiled with PCRE_UCP support\0"
484 "\\c must be followed by an ASCII character\0"
485 "\\k is not followed by a braced, angle-bracketed, or quoted name\0"
486 /* 70 */
487 "internal error: unknown opcode in find_fixedlength()\0"
488 "\\N is not supported in a class\0"
489 "too many forward references\0"
490 "disallowed UTF-8/16 code point (>= 0xd800 && <= 0xdfff)\0"
491 ;
492
493 /* Table to identify digits and hex digits. This is used when compiling
494 patterns. Note that the tables in chartables are dependent on the locale, and
495 may mark arbitrary characters as digits - but the PCRE compiling code expects
496 to handle only 0-9, a-z, and A-Z as digits when compiling. That is why we have
497 a private table here. It costs 256 bytes, but it is a lot faster than doing
498 character value tests (at least in some simple cases I timed), and in some
499 applications one wants PCRE to compile efficiently as well as match
500 efficiently.
501
502 For convenience, we use the same bit definitions as in chartables:
503
504 0x04 decimal digit
505 0x08 hexadecimal digit
506
507 Then we can use ctype_digit and ctype_xdigit in the code. */
508
509 /* Using a simple comparison for decimal numbers rather than a memory read
510 is much faster, and the resulting code is simpler (the compiler turns it
511 into a subtraction and unsigned comparison). */
512
513 #define IS_DIGIT(x) ((x) >= CHAR_0 && (x) <= CHAR_9)
514
515 #ifndef EBCDIC
516
517 /* This is the "normal" case, for ASCII systems, and EBCDIC systems running in
518 UTF-8 mode. */
519
520 static const pcre_uint8 digitab[] =
521 {
522 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 0- 7 */
523 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 8- 15 */
524 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 16- 23 */
525 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 24- 31 */
526 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - ' */
527 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* ( - / */
528 0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c, /* 0 - 7 */
529 0x0c,0x0c,0x00,0x00,0x00,0x00,0x00,0x00, /* 8 - ? */
530 0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* @ - G */
531 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* H - O */
532 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* P - W */
533 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* X - _ */
534 0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* ` - g */
535 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* h - o */
536 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* p - w */
537 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* x -127 */
538 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 128-135 */
539 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 136-143 */
540 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 144-151 */
541 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 152-159 */
542 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 160-167 */
543 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 168-175 */
544 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 176-183 */
545 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 184-191 */
546 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 192-199 */
547 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 200-207 */
548 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 208-215 */
549 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 216-223 */
550 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 224-231 */
551 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 232-239 */
552 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 240-247 */
553 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00};/* 248-255 */
554
555 #else
556
557 /* This is the "abnormal" case, for EBCDIC systems not running in UTF-8 mode. */
558
559 static const pcre_uint8 digitab[] =
560 {
561 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 0- 7 0 */
562 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 8- 15 */
563 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 16- 23 10 */
564 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 24- 31 */
565 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 32- 39 20 */
566 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 40- 47 */
567 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 48- 55 30 */
568 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 56- 63 */
569 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - 71 40 */
570 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 72- | */
571 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* & - 87 50 */
572 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 88- 95 */
573 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - -103 60 */
574 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 104- ? */
575 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 112-119 70 */
576 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 120- " */
577 0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* 128- g 80 */
578 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* h -143 */
579 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 144- p 90 */
580 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* q -159 */
581 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 160- x A0 */
582 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* y -175 */
583 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* ^ -183 B0 */
584 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 184-191 */
585 0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* { - G C0 */
586 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* H -207 */
587 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* } - P D0 */
588 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* Q -223 */
589 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* \ - X E0 */
590 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* Y -239 */
591 0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c, /* 0 - 7 F0 */
592 0x0c,0x0c,0x00,0x00,0x00,0x00,0x00,0x00};/* 8 -255 */
593
594 static const pcre_uint8 ebcdic_chartab[] = { /* chartable partial dup */
595 0x80,0x00,0x00,0x00,0x00,0x01,0x00,0x00, /* 0- 7 */
596 0x00,0x00,0x00,0x00,0x01,0x01,0x00,0x00, /* 8- 15 */
597 0x00,0x00,0x00,0x00,0x00,0x01,0x00,0x00, /* 16- 23 */
598 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 24- 31 */
599 0x00,0x00,0x00,0x00,0x00,0x01,0x00,0x00, /* 32- 39 */
600 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 40- 47 */
601 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 48- 55 */
602 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 56- 63 */
603 0x01,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - 71 */
604 0x00,0x00,0x00,0x80,0x00,0x80,0x80,0x80, /* 72- | */
605 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* & - 87 */
606 0x00,0x00,0x00,0x80,0x80,0x80,0x00,0x00, /* 88- 95 */
607 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - -103 */
608 0x00,0x00,0x00,0x00,0x00,0x10,0x00,0x80, /* 104- ? */
609 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 112-119 */
610 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 120- " */
611 0x00,0x1a,0x1a,0x1a,0x1a,0x1a,0x1a,0x12, /* 128- g */
612 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* h -143 */
613 0x00,0x12,0x12,0x12,0x12,0x12,0x12,0x12, /* 144- p */
614 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* q -159 */
615 0x00,0x00,0x12,0x12,0x12,0x12,0x12,0x12, /* 160- x */
616 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* y -175 */
617 0x80,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* ^ -183 */
618 0x00,0x00,0x80,0x00,0x00,0x00,0x00,0x00, /* 184-191 */
619 0x80,0x1a,0x1a,0x1a,0x1a,0x1a,0x1a,0x12, /* { - G */
620 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* H -207 */
621 0x00,0x12,0x12,0x12,0x12,0x12,0x12,0x12, /* } - P */
622 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* Q -223 */
623 0x00,0x00,0x12,0x12,0x12,0x12,0x12,0x12, /* \ - X */
624 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* Y -239 */
625 0x1c,0x1c,0x1c,0x1c,0x1c,0x1c,0x1c,0x1c, /* 0 - 7 */
626 0x1c,0x1c,0x00,0x00,0x00,0x00,0x00,0x00};/* 8 -255 */
627 #endif
628
629
630 /* Definition to allow mutual recursion */
631
632 static BOOL
633 compile_regex(int, pcre_uchar **, const pcre_uchar **, int *, BOOL, BOOL, int, int,
634 int *, int *, branch_chain *, compile_data *, int *);
635
636
637
638 /*************************************************
639 * Find an error text *
640 *************************************************/
641
642 /* The error texts are now all in one long string, to save on relocations. As
643 some of the text is of unknown length, we can't use a table of offsets.
644 Instead, just count through the strings. This is not a performance issue
645 because it happens only when there has been a compilation error.
646
647 Argument: the error number
648 Returns: pointer to the error string
649 */
650
651 static const char *
652 find_error_text(int n)
653 {
654 const char *s = error_texts;
655 for (; n > 0; n--)
656 {
657 while (*s++ != 0) {};
658 if (*s == 0) return "Error text not found (please report)";
659 }
660 return s;
661 }
662
663
664 /*************************************************
665 * Expand the workspace *
666 *************************************************/
667
668 /* This function is called during the second compiling phase, if the number of
669 forward references fills the existing workspace, which is originally a block on
670 the stack. A larger block is obtained from malloc() unless the ultimate limit
671 has been reached or the increase will be rather small.
672
673 Argument: pointer to the compile data block
674 Returns: 0 if all went well, else an error number
675 */
676
677 static int
678 expand_workspace(compile_data *cd)
679 {
680 pcre_uchar *newspace;
681 int newsize = cd->workspace_size * 2;
682
683 if (newsize > COMPILE_WORK_SIZE_MAX) newsize = COMPILE_WORK_SIZE_MAX;
684 if (cd->workspace_size >= COMPILE_WORK_SIZE_MAX ||
685 newsize - cd->workspace_size < WORK_SIZE_SAFETY_MARGIN)
686 return ERR72;
687
688 newspace = (PUBL(malloc))(newsize);
689 if (newspace == NULL) return ERR21;
690
691 memcpy(newspace, cd->start_workspace, cd->workspace_size);
692 cd->hwm = (pcre_uchar *)newspace + (cd->hwm - cd->start_workspace);
693 if (cd->workspace_size > COMPILE_WORK_SIZE)
694 (PUBL(free))((void *)cd->start_workspace);
695 cd->start_workspace = newspace;
696 cd->workspace_size = newsize;
697 return 0;
698 }
699
700
701
702 /*************************************************
703 * Check for counted repeat *
704 *************************************************/
705
706 /* This function is called when a '{' is encountered in a place where it might
707 start a quantifier. It looks ahead to see if it really is a quantifier or not.
708 It is only a quantifier if it is one of the forms {ddd} {ddd,} or {ddd,ddd}
709 where the ddds are digits.
710
711 Arguments:
712 p pointer to the first char after '{'
713
714 Returns: TRUE or FALSE
715 */
716
717 static BOOL
718 is_counted_repeat(const pcre_uchar *p)
719 {
720 if (!IS_DIGIT(*p)) return FALSE;
721 p++;
722 while (IS_DIGIT(*p)) p++;
723 if (*p == CHAR_RIGHT_CURLY_BRACKET) return TRUE;
724
725 if (*p++ != CHAR_COMMA) return FALSE;
726 if (*p == CHAR_RIGHT_CURLY_BRACKET) return TRUE;
727
728 if (!IS_DIGIT(*p)) return FALSE;
729 p++;
730 while (IS_DIGIT(*p)) p++;
731
732 return (*p == CHAR_RIGHT_CURLY_BRACKET);
733 }
734
735
736
737 /*************************************************
738 * Handle escapes *
739 *************************************************/
740
741 /* This function is called when a \ has been encountered. It either returns a
742 positive value for a simple escape such as \n, or a negative value which
743 encodes one of the more complicated things such as \d. A backreference to group
744 n is returned as -(ESC_REF + n); ESC_REF is the highest ESC_xxx macro. When
745 UTF-8 is enabled, a positive value greater than 255 may be returned. On entry,
746 ptr is pointing at the \. On exit, it is on the final character of the escape
747 sequence.
748
749 Arguments:
750 ptrptr points to the pattern position pointer
751 errorcodeptr points to the errorcode variable
752 bracount number of previous extracting brackets
753 options the options bits
754 isclass TRUE if inside a character class
755
756 Returns: zero or positive => a data character
757 negative => a special escape sequence
758 on error, errorcodeptr is set
759 */
760
761 static int
762 check_escape(const pcre_uchar **ptrptr, int *errorcodeptr, int bracount,
763 int options, BOOL isclass)
764 {
765 /* PCRE_UTF16 has the same value as PCRE_UTF8. */
766 BOOL utf = (options & PCRE_UTF8) != 0;
767 const pcre_uchar *ptr = *ptrptr + 1;
768 pcre_int32 c;
769 int i;
770
771 GETCHARINCTEST(c, ptr); /* Get character value, increment pointer */
772 ptr--; /* Set pointer back to the last byte */
773
774 /* If backslash is at the end of the pattern, it's an error. */
775
776 if (c == 0) *errorcodeptr = ERR1;
777
778 /* Non-alphanumerics are literals. For digits or letters, do an initial lookup
779 in a table. A non-zero result is something that can be returned immediately.
780 Otherwise further processing may be required. */
781
782 #ifndef EBCDIC /* ASCII/UTF-8 coding */
783 /* Not alphanumeric */
784 else if (c < CHAR_0 || c > CHAR_z) {}
785 else if ((i = escapes[c - CHAR_0]) != 0) c = i;
786
787 #else /* EBCDIC coding */
788 /* Not alphanumeric */
789 else if (c < 'a' || (!MAX_255(c) || (ebcdic_chartab[c] & 0x0E) == 0)) {}
790 else if ((i = escapes[c - 0x48]) != 0) c = i;
791 #endif
792
793 /* Escapes that need further processing, or are illegal. */
794
795 else
796 {
797 const pcre_uchar *oldptr;
798 BOOL braced, negated;
799
800 switch (c)
801 {
802 /* A number of Perl escapes are not handled by PCRE. We give an explicit
803 error. */
804
805 case CHAR_l:
806 case CHAR_L:
807 *errorcodeptr = ERR37;
808 break;
809
810 case CHAR_u:
811 if ((options & PCRE_JAVASCRIPT_COMPAT) != 0)
812 {
813 /* In JavaScript, \u must be followed by four hexadecimal numbers.
814 Otherwise it is a lowercase u letter. */
815 if (MAX_255(ptr[1]) && (digitab[ptr[1]] & ctype_xdigit) != 0
816 && MAX_255(ptr[2]) && (digitab[ptr[2]] & ctype_xdigit) != 0
817 && MAX_255(ptr[3]) && (digitab[ptr[3]] & ctype_xdigit) != 0
818 && MAX_255(ptr[4]) && (digitab[ptr[4]] & ctype_xdigit) != 0)
819 {
820 c = 0;
821 for (i = 0; i < 4; ++i)
822 {
823 register int cc = *(++ptr);
824 #ifndef EBCDIC /* ASCII/UTF-8 coding */
825 if (cc >= CHAR_a) cc -= 32; /* Convert to upper case */
826 c = (c << 4) + cc - ((cc < CHAR_A)? CHAR_0 : (CHAR_A - 10));
827 #else /* EBCDIC coding */
828 if (cc >= CHAR_a && cc <= CHAR_z) cc += 64; /* Convert to upper case */
829 c = (c << 4) + cc - ((cc >= CHAR_0)? CHAR_0 : (CHAR_A - 10));
830 #endif
831 }
832 }
833 }
834 else
835 *errorcodeptr = ERR37;
836 break;
837
838 case CHAR_U:
839 /* In JavaScript, \U is an uppercase U letter. */
840 if ((options & PCRE_JAVASCRIPT_COMPAT) == 0) *errorcodeptr = ERR37;
841 break;
842
843 /* In a character class, \g is just a literal "g". Outside a character
844 class, \g must be followed by one of a number of specific things:
845
846 (1) A number, either plain or braced. If positive, it is an absolute
847 backreference. If negative, it is a relative backreference. This is a Perl
848 5.10 feature.
849
850 (2) Perl 5.10 also supports \g{name} as a reference to a named group. This
851 is part of Perl's movement towards a unified syntax for back references. As
852 this is synonymous with \k{name}, we fudge it up by pretending it really
853 was \k.
854
855 (3) For Oniguruma compatibility we also support \g followed by a name or a
856 number either in angle brackets or in single quotes. However, these are
857 (possibly recursive) subroutine calls, _not_ backreferences. Just return
858 the -ESC_g code (cf \k). */
859
860 case CHAR_g:
861 if (isclass) break;
862 if (ptr[1] == CHAR_LESS_THAN_SIGN || ptr[1] == CHAR_APOSTROPHE)
863 {
864 c = -ESC_g;
865 break;
866 }
867
868 /* Handle the Perl-compatible cases */
869
870 if (ptr[1] == CHAR_LEFT_CURLY_BRACKET)
871 {
872 const pcre_uchar *p;
873 for (p = ptr+2; *p != 0 && *p != CHAR_RIGHT_CURLY_BRACKET; p++)
874 if (*p != CHAR_MINUS && !IS_DIGIT(*p)) break;
875 if (*p != 0 && *p != CHAR_RIGHT_CURLY_BRACKET)
876 {
877 c = -ESC_k;
878 break;
879 }
880 braced = TRUE;
881 ptr++;
882 }
883 else braced = FALSE;
884
885 if (ptr[1] == CHAR_MINUS)
886 {
887 negated = TRUE;
888 ptr++;
889 }
890 else negated = FALSE;
891
892 /* The integer range is limited by the machine's int representation. */
893 c = 0;
894 while (IS_DIGIT(ptr[1]))
895 {
896 if (((unsigned int)c) > INT_MAX / 10) /* Integer overflow */
897 {
898 c = -1;
899 break;
900 }
901 c = c * 10 + *(++ptr) - CHAR_0;
902 }
903 if (((unsigned int)c) > INT_MAX) /* Integer overflow */
904 {
905 while (IS_DIGIT(ptr[1]))
906 ptr++;
907 *errorcodeptr = ERR61;
908 break;
909 }
910
911 if (braced && *(++ptr) != CHAR_RIGHT_CURLY_BRACKET)
912 {
913 *errorcodeptr = ERR57;
914 break;
915 }
916
917 if (c == 0)
918 {
919 *errorcodeptr = ERR58;
920 break;
921 }
922
923 if (negated)
924 {
925 if (c > bracount)
926 {
927 *errorcodeptr = ERR15;
928 break;
929 }
930 c = bracount - (c - 1);
931 }
932
933 c = -(ESC_REF + c);
934 break;
935
936 /* The handling of escape sequences consisting of a string of digits
937 starting with one that is not zero is not straightforward. By experiment,
938 the way Perl works seems to be as follows:
939
940 Outside a character class, the digits are read as a decimal number. If the
941 number is less than 10, or if there are that many previous extracting
942 left brackets, then it is a back reference. Otherwise, up to three octal
943 digits are read to form an escaped byte. Thus \123 is likely to be octal
944 123 (cf \0123, which is octal 012 followed by the literal 3). If the octal
945 value is greater than 377, the least significant 8 bits are taken. Inside a
946 character class, \ followed by a digit is always an octal number. */
947
948 case CHAR_1: case CHAR_2: case CHAR_3: case CHAR_4: case CHAR_5:
949 case CHAR_6: case CHAR_7: case CHAR_8: case CHAR_9:
950
951 if (!isclass)
952 {
953 oldptr = ptr;
954 /* The integer range is limited by the machine's int representation. */
955 c -= CHAR_0;
956 while (IS_DIGIT(ptr[1]))
957 {
958 if (((unsigned int)c) > INT_MAX / 10) /* Integer overflow */
959 {
960 c = -1;
961 break;
962 }
963 c = c * 10 + *(++ptr) - CHAR_0;
964 }
965 if (((unsigned int)c) > INT_MAX) /* Integer overflow */
966 {
967 while (IS_DIGIT(ptr[1]))
968 ptr++;
969 *errorcodeptr = ERR61;
970 break;
971 }
972 if (c < 10 || c <= bracount)
973 {
974 c = -(ESC_REF + c);
975 break;
976 }
977 ptr = oldptr; /* Put the pointer back and fall through */
978 }
979
980 /* Handle an octal number following \. If the first digit is 8 or 9, Perl
981 generates a binary zero byte and treats the digit as a following literal.
982 Thus we have to pull back the pointer by one. */
983
984 if ((c = *ptr) >= CHAR_8)
985 {
986 ptr--;
987 c = 0;
988 break;
989 }
990
991 /* \0 always starts an octal number, but we may drop through to here with a
992 larger first octal digit. The original code used just to take the least
993 significant 8 bits of octal numbers (I think this is what early Perls used
994 to do). Nowadays we allow for larger numbers in UTF-8 mode, but no more
995 than 3 octal digits. */
996
997 case CHAR_0:
998 c -= CHAR_0;
999 while(i++ < 2 && ptr[1] >= CHAR_0 && ptr[1] <= CHAR_7)
1000 c = c * 8 + *(++ptr) - CHAR_0;
1001 if (!utf && c > 0xff) *errorcodeptr = ERR51;
1002 break;
1003
1004 /* \x is complicated. \x{ddd} is a character number which can be greater
1005 than 0xff in utf or non-8bit mode, but only if the ddd are hex digits.
1006 If not, { is treated as a data character. */
1007
1008 case CHAR_x:
1009 if ((options & PCRE_JAVASCRIPT_COMPAT) != 0)
1010 {
1011 /* In JavaScript, \x must be followed by two hexadecimal numbers.
1012 Otherwise it is a lowercase x letter. */
1013 if (MAX_255(ptr[1]) && (digitab[ptr[1]] & ctype_xdigit) != 0
1014 && MAX_255(ptr[2]) && (digitab[ptr[2]] & ctype_xdigit) != 0)
1015 {
1016 c = 0;
1017 for (i = 0; i < 2; ++i)
1018 {
1019 register int cc = *(++ptr);
1020 #ifndef EBCDIC /* ASCII/UTF-8 coding */
1021 if (cc >= CHAR_a) cc -= 32; /* Convert to upper case */
1022 c = (c << 4) + cc - ((cc < CHAR_A)? CHAR_0 : (CHAR_A - 10));
1023 #else /* EBCDIC coding */
1024 if (cc >= CHAR_a && cc <= CHAR_z) cc += 64; /* Convert to upper case */
1025 c = (c << 4) + cc - ((cc >= CHAR_0)? CHAR_0 : (CHAR_A - 10));
1026 #endif
1027 }
1028 }
1029 break;
1030 }
1031
1032 if (ptr[1] == CHAR_LEFT_CURLY_BRACKET)
1033 {
1034 const pcre_uchar *pt = ptr + 2;
1035
1036 c = 0;
1037 while (MAX_255(*pt) && (digitab[*pt] & ctype_xdigit) != 0)
1038 {
1039 register int cc = *pt++;
1040 if (c == 0 && cc == CHAR_0) continue; /* Leading zeroes */
1041
1042 #ifndef EBCDIC /* ASCII/UTF-8 coding */
1043 if (cc >= CHAR_a) cc -= 32; /* Convert to upper case */
1044 c = (c << 4) + cc - ((cc < CHAR_A)? CHAR_0 : (CHAR_A - 10));
1045 #else /* EBCDIC coding */
1046 if (cc >= CHAR_a && cc <= CHAR_z) cc += 64; /* Convert to upper case */
1047 c = (c << 4) + cc - ((cc >= CHAR_0)? CHAR_0 : (CHAR_A - 10));
1048 #endif
1049
1050 #ifdef COMPILE_PCRE8
1051 if (c > (utf ? 0x10ffff : 0xff)) { c = -1; break; }
1052 #else
1053 #ifdef COMPILE_PCRE16
1054 if (c > (utf ? 0x10ffff : 0xffff)) { c = -1; break; }
1055 #endif
1056 #endif
1057 }
1058
1059 if (c < 0)
1060 {
1061 while (MAX_255(*pt) && (digitab[*pt] & ctype_xdigit) != 0) pt++;
1062 *errorcodeptr = ERR34;
1063 }
1064
1065 if (*pt == CHAR_RIGHT_CURLY_BRACKET)
1066 {
1067 if (utf && c >= 0xd800 && c <= 0xdfff) *errorcodeptr = ERR73;
1068 ptr = pt;
1069 break;
1070 }
1071
1072 /* If the sequence of hex digits does not end with '}', then we don't
1073 recognize this construct; fall through to the normal \x handling. */
1074 }
1075
1076 /* Read just a single-byte hex-defined char */
1077
1078 c = 0;
1079 while (i++ < 2 && MAX_255(ptr[1]) && (digitab[ptr[1]] & ctype_xdigit) != 0)
1080 {
1081 int cc; /* Some compilers don't like */
1082 cc = *(++ptr); /* ++ in initializers */
1083 #ifndef EBCDIC /* ASCII/UTF-8 coding */
1084 if (cc >= CHAR_a) cc -= 32; /* Convert to upper case */
1085 c = c * 16 + cc - ((cc < CHAR_A)? CHAR_0 : (CHAR_A - 10));
1086 #else /* EBCDIC coding */
1087 if (cc <= CHAR_z) cc += 64; /* Convert to upper case */
1088 c = c * 16 + cc - ((cc >= CHAR_0)? CHAR_0 : (CHAR_A - 10));
1089 #endif
1090 }
1091 break;
1092
1093 /* For \c, a following letter is upper-cased; then the 0x40 bit is flipped.
1094 An error is given if the byte following \c is not an ASCII character. This
1095 coding is ASCII-specific, but then the whole concept of \cx is
1096 ASCII-specific. (However, an EBCDIC equivalent has now been added.) */
1097
1098 case CHAR_c:
1099 c = *(++ptr);
1100 if (c == 0)
1101 {
1102 *errorcodeptr = ERR2;
1103 break;
1104 }
1105 #ifndef EBCDIC /* ASCII/UTF-8 coding */
1106 if (c > 127) /* Excludes all non-ASCII in either mode */
1107 {
1108 *errorcodeptr = ERR68;
1109 break;
1110 }
1111 if (c >= CHAR_a && c <= CHAR_z) c -= 32;
1112 c ^= 0x40;
1113 #else /* EBCDIC coding */
1114 if (c >= CHAR_a && c <= CHAR_z) c += 64;
1115 c ^= 0xC0;
1116 #endif
1117 break;
1118
1119 /* PCRE_EXTRA enables extensions to Perl in the matter of escapes. Any
1120 other alphanumeric following \ is an error if PCRE_EXTRA was set;
1121 otherwise, for Perl compatibility, it is a literal. This code looks a bit
1122 odd, but there used to be some cases other than the default, and there may
1123 be again in future, so I haven't "optimized" it. */
1124
1125 default:
1126 if ((options & PCRE_EXTRA) != 0) switch(c)
1127 {
1128 default:
1129 *errorcodeptr = ERR3;
1130 break;
1131 }
1132 break;
1133 }
1134 }
1135
1136 /* Perl supports \N{name} for character names, as well as plain \N for "not
1137 newline". PCRE does not support \N{name}. However, it does support
1138 quantification such as \N{2,3}. */
1139
1140 if (c == -ESC_N && ptr[1] == CHAR_LEFT_CURLY_BRACKET &&
1141 !is_counted_repeat(ptr+2))
1142 *errorcodeptr = ERR37;
1143
1144 /* If PCRE_UCP is set, we change the values for \d etc. */
1145
1146 if ((options & PCRE_UCP) != 0 && c <= -ESC_D && c >= -ESC_w)
1147 c -= (ESC_DU - ESC_D);
1148
1149 /* Set the pointer to the final character before returning. */
1150
1151 *ptrptr = ptr;
1152 return c;
1153 }
1154
1155
1156
1157 #ifdef SUPPORT_UCP
1158 /*************************************************
1159 * Handle \P and \p *
1160 *************************************************/
1161
1162 /* This function is called after \P or \p has been encountered, provided that
1163 PCRE is compiled with support for Unicode properties. On entry, ptrptr is
1164 pointing at the P or p. On exit, it is pointing at the final character of the
1165 escape sequence.
1166
1167 Argument:
1168 ptrptr points to the pattern position pointer
1169 negptr points to a boolean that is set TRUE for negation else FALSE
1170 dptr points to an int that is set to the detailed property value
1171 errorcodeptr points to the error code variable
1172
1173 Returns: type value from ucp_type_table, or -1 for an invalid type
1174 */
1175
1176 static int
1177 get_ucp(const pcre_uchar **ptrptr, BOOL *negptr, int *dptr, int *errorcodeptr)
1178 {
1179 int c, i, bot, top;
1180 const pcre_uchar *ptr = *ptrptr;
1181 pcre_uchar name[32];
1182
1183 c = *(++ptr);
1184 if (c == 0) goto ERROR_RETURN;
1185
1186 *negptr = FALSE;
1187
1188 /* \P or \p can be followed by a name in {}, optionally preceded by ^ for
1189 negation. */
1190
1191 if (c == CHAR_LEFT_CURLY_BRACKET)
1192 {
1193 if (ptr[1] == CHAR_CIRCUMFLEX_ACCENT)
1194 {
1195 *negptr = TRUE;
1196 ptr++;
1197 }
1198 for (i = 0; i < (int)(sizeof(name) / sizeof(pcre_uchar)) - 1; i++)
1199 {
1200 c = *(++ptr);
1201 if (c == 0) goto ERROR_RETURN;
1202 if (c == CHAR_RIGHT_CURLY_BRACKET) break;
1203 name[i] = c;
1204 }
1205 if (c != CHAR_RIGHT_CURLY_BRACKET) goto ERROR_RETURN;
1206 name[i] = 0;
1207 }
1208
1209 /* Otherwise there is just one following character */
1210
1211 else
1212 {
1213 name[0] = c;
1214 name[1] = 0;
1215 }
1216
1217 *ptrptr = ptr;
1218
1219 /* Search for a recognized property name using binary chop */
1220
1221 bot = 0;
1222 top = PRIV(utt_size);
1223
1224 while (bot < top)
1225 {
1226 i = (bot + top) >> 1;
1227 c = STRCMP_UC_C8(name, PRIV(utt_names) + PRIV(utt)[i].name_offset);
1228 if (c == 0)
1229 {
1230 *dptr = PRIV(utt)[i].value;
1231 return PRIV(utt)[i].type;
1232 }
1233 if (c > 0) bot = i + 1; else top = i;
1234 }
1235
1236 *errorcodeptr = ERR47;
1237 *ptrptr = ptr;
1238 return -1;
1239
1240 ERROR_RETURN:
1241 *errorcodeptr = ERR46;
1242 *ptrptr = ptr;
1243 return -1;
1244 }
1245 #endif
1246
1247
1248
1249
1250 /*************************************************
1251 * Read repeat counts *
1252 *************************************************/
1253
1254 /* Read an item of the form {n,m} and return the values. This is called only
1255 after is_counted_repeat() has confirmed that a repeat-count quantifier exists,
1256 so the syntax is guaranteed to be correct, but we need to check the values.
1257
1258 Arguments:
1259 p pointer to first char after '{'
1260 minp pointer to int for min
1261 maxp pointer to int for max
1262 returned as -1 if no max
1263 errorcodeptr points to error code variable
1264
1265 Returns: pointer to '}' on success;
1266 current ptr on error, with errorcodeptr set non-zero
1267 */
1268
1269 static const pcre_uchar *
1270 read_repeat_counts(const pcre_uchar *p, int *minp, int *maxp, int *errorcodeptr)
1271 {
1272 int min = 0;
1273 int max = -1;
1274
1275 /* Read the minimum value and do a paranoid check: a negative value indicates
1276 an integer overflow. */
1277
1278 while (IS_DIGIT(*p)) min = min * 10 + *p++ - CHAR_0;
1279 if (min < 0 || min > 65535)
1280 {
1281 *errorcodeptr = ERR5;
1282 return p;
1283 }
1284
1285 /* Read the maximum value if there is one, and again do a paranoid on its size.
1286 Also, max must not be less than min. */
1287
1288 if (*p == CHAR_RIGHT_CURLY_BRACKET) max = min; else
1289 {
1290 if (*(++p) != CHAR_RIGHT_CURLY_BRACKET)
1291 {
1292 max = 0;
1293 while(IS_DIGIT(*p)) max = max * 10 + *p++ - CHAR_0;
1294 if (max < 0 || max > 65535)
1295 {
1296 *errorcodeptr = ERR5;
1297 return p;
1298 }
1299 if (max < min)
1300 {
1301 *errorcodeptr = ERR4;
1302 return p;
1303 }
1304 }
1305 }
1306
1307 /* Fill in the required variables, and pass back the pointer to the terminating
1308 '}'. */
1309
1310 *minp = min;
1311 *maxp = max;
1312 return p;
1313 }
1314
1315
1316
1317 /*************************************************
1318 * Subroutine for finding forward reference *
1319 *************************************************/
1320
1321 /* This recursive function is called only from find_parens() below. The
1322 top-level call starts at the beginning of the pattern. All other calls must
1323 start at a parenthesis. It scans along a pattern's text looking for capturing
1324 subpatterns, and counting them. If it finds a named pattern that matches the
1325 name it is given, it returns its number. Alternatively, if the name is NULL, it
1326 returns when it reaches a given numbered subpattern. Recursion is used to keep
1327 track of subpatterns that reset the capturing group numbers - the (?| feature.
1328
1329 This function was originally called only from the second pass, in which we know
1330 that if (?< or (?' or (?P< is encountered, the name will be correctly
1331 terminated because that is checked in the first pass. There is now one call to
1332 this function in the first pass, to check for a recursive back reference by
1333 name (so that we can make the whole group atomic). In this case, we need check
1334 only up to the current position in the pattern, and that is still OK because
1335 and previous occurrences will have been checked. To make this work, the test
1336 for "end of pattern" is a check against cd->end_pattern in the main loop,
1337 instead of looking for a binary zero. This means that the special first-pass
1338 call can adjust cd->end_pattern temporarily. (Checks for binary zero while
1339 processing items within the loop are OK, because afterwards the main loop will
1340 terminate.)
1341
1342 Arguments:
1343 ptrptr address of the current character pointer (updated)
1344 cd compile background data
1345 name name to seek, or NULL if seeking a numbered subpattern
1346 lorn name length, or subpattern number if name is NULL
1347 xmode TRUE if we are in /x mode
1348 utf TRUE if we are in UTF-8 / UTF-16 mode
1349 count pointer to the current capturing subpattern number (updated)
1350
1351 Returns: the number of the named subpattern, or -1 if not found
1352 */
1353
1354 static int
1355 find_parens_sub(pcre_uchar **ptrptr, compile_data *cd, const pcre_uchar *name, int lorn,
1356 BOOL xmode, BOOL utf, int *count)
1357 {
1358 pcre_uchar *ptr = *ptrptr;
1359 int start_count = *count;
1360 int hwm_count = start_count;
1361 BOOL dup_parens = FALSE;
1362
1363 /* If the first character is a parenthesis, check on the type of group we are
1364 dealing with. The very first call may not start with a parenthesis. */
1365
1366 if (ptr[0] == CHAR_LEFT_PARENTHESIS)
1367 {
1368 /* Handle specials such as (*SKIP) or (*UTF8) etc. */
1369
1370 if (ptr[1] == CHAR_ASTERISK) ptr += 2;
1371
1372 /* Handle a normal, unnamed capturing parenthesis. */
1373
1374 else if (ptr[1] != CHAR_QUESTION_MARK)
1375 {
1376 *count += 1;
1377 if (name == NULL && *count == lorn) return *count;
1378 ptr++;
1379 }
1380
1381 /* All cases now have (? at the start. Remember when we are in a group
1382 where the parenthesis numbers are duplicated. */
1383
1384 else if (ptr[2] == CHAR_VERTICAL_LINE)
1385 {
1386 ptr += 3;
1387 dup_parens = TRUE;
1388 }
1389
1390 /* Handle comments; all characters are allowed until a ket is reached. */
1391
1392 else if (ptr[2] == CHAR_NUMBER_SIGN)
1393 {
1394 for (ptr += 3; *ptr != 0; ptr++) if (*ptr == CHAR_RIGHT_PARENTHESIS) break;
1395 goto FAIL_EXIT;
1396 }
1397
1398 /* Handle a condition. If it is an assertion, just carry on so that it
1399 is processed as normal. If not, skip to the closing parenthesis of the
1400 condition (there can't be any nested parens). */
1401
1402 else if (ptr[2] == CHAR_LEFT_PARENTHESIS)
1403 {
1404 ptr += 2;
1405 if (ptr[1] != CHAR_QUESTION_MARK)
1406 {
1407 while (*ptr != 0 && *ptr != CHAR_RIGHT_PARENTHESIS) ptr++;
1408 if (*ptr != 0) ptr++;
1409 }
1410 }
1411
1412 /* Start with (? but not a condition. */
1413
1414 else
1415 {
1416 ptr += 2;
1417 if (*ptr == CHAR_P) ptr++; /* Allow optional P */
1418
1419 /* We have to disambiguate (?<! and (?<= from (?<name> for named groups */
1420
1421 if ((*ptr == CHAR_LESS_THAN_SIGN && ptr[1] != CHAR_EXCLAMATION_MARK &&
1422 ptr[1] != CHAR_EQUALS_SIGN) || *ptr == CHAR_APOSTROPHE)
1423 {
1424 int term;
1425 const pcre_uchar *thisname;
1426 *count += 1;
1427 if (name == NULL && *count == lorn) return *count;
1428 term = *ptr++;
1429 if (term == CHAR_LESS_THAN_SIGN) term = CHAR_GREATER_THAN_SIGN;
1430 thisname = ptr;
1431 while (*ptr != term) ptr++;
1432 if (name != NULL && lorn == ptr - thisname &&
1433 STRNCMP_UC_UC(name, thisname, lorn) == 0)
1434 return *count;
1435 term++;
1436 }
1437 }
1438 }
1439
1440 /* Past any initial parenthesis handling, scan for parentheses or vertical
1441 bars. Stop if we get to cd->end_pattern. Note that this is important for the
1442 first-pass call when this value is temporarily adjusted to stop at the current
1443 position. So DO NOT change this to a test for binary zero. */
1444
1445 for (; ptr < cd->end_pattern; ptr++)
1446 {
1447 /* Skip over backslashed characters and also entire \Q...\E */
1448
1449 if (*ptr == CHAR_BACKSLASH)
1450 {
1451 if (*(++ptr) == 0) goto FAIL_EXIT;
1452 if (*ptr == CHAR_Q) for (;;)
1453 {
1454 while (*(++ptr) != 0 && *ptr != CHAR_BACKSLASH) {};
1455 if (*ptr == 0) goto FAIL_EXIT;
1456 if (*(++ptr) == CHAR_E) break;
1457 }
1458 continue;
1459 }
1460
1461 /* Skip over character classes; this logic must be similar to the way they
1462 are handled for real. If the first character is '^', skip it. Also, if the
1463 first few characters (either before or after ^) are \Q\E or \E we skip them
1464 too. This makes for compatibility with Perl. Note the use of STR macros to
1465 encode "Q\\E" so that it works in UTF-8 on EBCDIC platforms. */
1466
1467 if (*ptr == CHAR_LEFT_SQUARE_BRACKET)
1468 {
1469 BOOL negate_class = FALSE;
1470 for (;;)
1471 {
1472 if (ptr[1] == CHAR_BACKSLASH)
1473 {
1474 if (ptr[2] == CHAR_E)
1475 ptr+= 2;
1476 else if (STRNCMP_UC_C8(ptr + 2,
1477 STR_Q STR_BACKSLASH STR_E, 3) == 0)
1478 ptr += 4;
1479 else
1480 break;
1481 }
1482 else if (!negate_class && ptr[1] == CHAR_CIRCUMFLEX_ACCENT)
1483 {
1484 negate_class = TRUE;
1485 ptr++;
1486 }
1487 else break;
1488 }
1489
1490 /* If the next character is ']', it is a data character that must be
1491 skipped, except in JavaScript compatibility mode. */
1492
1493 if (ptr[1] == CHAR_RIGHT_SQUARE_BRACKET &&
1494 (cd->external_options & PCRE_JAVASCRIPT_COMPAT) == 0)
1495 ptr++;
1496
1497 while (*(++ptr) != CHAR_RIGHT_SQUARE_BRACKET)
1498 {
1499 if (*ptr == 0) return -1;
1500 if (*ptr == CHAR_BACKSLASH)
1501 {
1502 if (*(++ptr) == 0) goto FAIL_EXIT;
1503 if (*ptr == CHAR_Q) for (;;)
1504 {
1505 while (*(++ptr) != 0 && *ptr != CHAR_BACKSLASH) {};
1506 if (*ptr == 0) goto FAIL_EXIT;
1507 if (*(++ptr) == CHAR_E) break;
1508 }
1509 continue;
1510 }
1511 }
1512 continue;
1513 }
1514
1515 /* Skip comments in /x mode */
1516
1517 if (xmode && *ptr == CHAR_NUMBER_SIGN)
1518 {
1519 ptr++;
1520 while (*ptr != 0)
1521 {
1522 if (IS_NEWLINE(ptr)) { ptr += cd->nllen - 1; break; }
1523 ptr++;
1524 #ifdef SUPPORT_UTF
1525 if (utf) FORWARDCHAR(ptr);
1526 #endif
1527 }
1528 if (*ptr == 0) goto FAIL_EXIT;
1529 continue;
1530 }
1531
1532 /* Check for the special metacharacters */
1533
1534 if (*ptr == CHAR_LEFT_PARENTHESIS)
1535 {
1536 int rc = find_parens_sub(&ptr, cd, name, lorn, xmode, utf, count);
1537 if (rc > 0) return rc;
1538 if (*ptr == 0) goto FAIL_EXIT;
1539 }
1540
1541 else if (*ptr == CHAR_RIGHT_PARENTHESIS)
1542 {
1543 if (dup_parens && *count < hwm_count) *count = hwm_count;
1544 goto FAIL_EXIT;
1545 }
1546
1547 else if (*ptr == CHAR_VERTICAL_LINE && dup_parens)
1548 {
1549 if (*count > hwm_count) hwm_count = *count;
1550 *count = start_count;
1551 }
1552 }
1553
1554 FAIL_EXIT:
1555 *ptrptr = ptr;
1556 return -1;
1557 }
1558
1559
1560
1561
1562 /*************************************************
1563 * Find forward referenced subpattern *
1564 *************************************************/
1565
1566 /* This function scans along a pattern's text looking for capturing
1567 subpatterns, and counting them. If it finds a named pattern that matches the
1568 name it is given, it returns its number. Alternatively, if the name is NULL, it
1569 returns when it reaches a given numbered subpattern. This is used for forward
1570 references to subpatterns. We used to be able to start this scan from the
1571 current compiling point, using the current count value from cd->bracount, and
1572 do it all in a single loop, but the addition of the possibility of duplicate
1573 subpattern numbers means that we have to scan from the very start, in order to
1574 take account of such duplicates, and to use a recursive function to keep track
1575 of the different types of group.
1576
1577 Arguments:
1578 cd compile background data
1579 name name to seek, or NULL if seeking a numbered subpattern
1580 lorn name length, or subpattern number if name is NULL
1581 xmode TRUE if we are in /x mode
1582 utf TRUE if we are in UTF-8 / UTF-16 mode
1583
1584 Returns: the number of the found subpattern, or -1 if not found
1585 */
1586
1587 static int
1588 find_parens(compile_data *cd, const pcre_uchar *name, int lorn, BOOL xmode,
1589 BOOL utf)
1590 {
1591 pcre_uchar *ptr = (pcre_uchar *)cd->start_pattern;
1592 int count = 0;
1593 int rc;
1594
1595 /* If the pattern does not start with an opening parenthesis, the first call
1596 to find_parens_sub() will scan right to the end (if necessary). However, if it
1597 does start with a parenthesis, find_parens_sub() will return when it hits the
1598 matching closing parens. That is why we have to have a loop. */
1599
1600 for (;;)
1601 {
1602 rc = find_parens_sub(&ptr, cd, name, lorn, xmode, utf, &count);
1603 if (rc > 0 || *ptr++ == 0) break;
1604 }
1605
1606 return rc;
1607 }
1608
1609
1610
1611
1612 /*************************************************
1613 * Find first significant op code *
1614 *************************************************/
1615
1616 /* This is called by several functions that scan a compiled expression looking
1617 for a fixed first character, or an anchoring op code etc. It skips over things
1618 that do not influence this. For some calls, it makes sense to skip negative
1619 forward and all backward assertions, and also the \b assertion; for others it
1620 does not.
1621
1622 Arguments:
1623 code pointer to the start of the group
1624 skipassert TRUE if certain assertions are to be skipped
1625
1626 Returns: pointer to the first significant opcode
1627 */
1628
1629 static const pcre_uchar*
1630 first_significant_code(const pcre_uchar *code, BOOL skipassert)
1631 {
1632 for (;;)
1633 {
1634 switch ((int)*code)
1635 {
1636 case OP_ASSERT_NOT:
1637 case OP_ASSERTBACK:
1638 case OP_ASSERTBACK_NOT:
1639 if (!skipassert) return code;
1640 do code += GET(code, 1); while (*code == OP_ALT);
1641 code += PRIV(OP_lengths)[*code];
1642 break;
1643
1644 case OP_WORD_BOUNDARY:
1645 case OP_NOT_WORD_BOUNDARY:
1646 if (!skipassert) return code;
1647 /* Fall through */
1648
1649 case OP_CALLOUT:
1650 case OP_CREF:
1651 case OP_NCREF:
1652 case OP_RREF:
1653 case OP_NRREF:
1654 case OP_DEF:
1655 code += PRIV(OP_lengths)[*code];
1656 break;
1657
1658 default:
1659 return code;
1660 }
1661 }
1662 /* Control never reaches here */
1663 }
1664
1665
1666
1667
1668 /*************************************************
1669 * Find the fixed length of a branch *
1670 *************************************************/
1671
1672 /* Scan a branch and compute the fixed length of subject that will match it,
1673 if the length is fixed. This is needed for dealing with backward assertions.
1674 In UTF8 mode, the result is in characters rather than bytes. The branch is
1675 temporarily terminated with OP_END when this function is called.
1676
1677 This function is called when a backward assertion is encountered, so that if it
1678 fails, the error message can point to the correct place in the pattern.
1679 However, we cannot do this when the assertion contains subroutine calls,
1680 because they can be forward references. We solve this by remembering this case
1681 and doing the check at the end; a flag specifies which mode we are running in.
1682
1683 Arguments:
1684 code points to the start of the pattern (the bracket)
1685 utf TRUE in UTF-8 / UTF-16 mode
1686 atend TRUE if called when the pattern is complete
1687 cd the "compile data" structure
1688
1689 Returns: the fixed length,
1690 or -1 if there is no fixed length,
1691 or -2 if \C was encountered (in UTF-8 mode only)
1692 or -3 if an OP_RECURSE item was encountered and atend is FALSE
1693 or -4 if an unknown opcode was encountered (internal error)
1694 */
1695
1696 static int
1697 find_fixedlength(pcre_uchar *code, BOOL utf, BOOL atend, compile_data *cd)
1698 {
1699 int length = -1;
1700
1701 register int branchlength = 0;
1702 register pcre_uchar *cc = code + 1 + LINK_SIZE;
1703
1704 /* Scan along the opcodes for this branch. If we get to the end of the
1705 branch, check the length against that of the other branches. */
1706
1707 for (;;)
1708 {
1709 int d;
1710 pcre_uchar *ce, *cs;
1711 register int op = *cc;
1712
1713 switch (op)
1714 {
1715 /* We only need to continue for OP_CBRA (normal capturing bracket) and
1716 OP_BRA (normal non-capturing bracket) because the other variants of these
1717 opcodes are all concerned with unlimited repeated groups, which of course
1718 are not of fixed length. */
1719
1720 case OP_CBRA:
1721 case OP_BRA:
1722 case OP_ONCE:
1723 case OP_ONCE_NC:
1724 case OP_COND:
1725 d = find_fixedlength(cc + ((op == OP_CBRA)? IMM2_SIZE : 0), utf, atend, cd);
1726 if (d < 0) return d;
1727 branchlength += d;
1728 do cc += GET(cc, 1); while (*cc == OP_ALT);
1729 cc += 1 + LINK_SIZE;
1730 break;
1731
1732 /* Reached end of a branch; if it's a ket it is the end of a nested call.
1733 If it's ALT it is an alternation in a nested call. An ACCEPT is effectively
1734 an ALT. If it is END it's the end of the outer call. All can be handled by
1735 the same code. Note that we must not include the OP_KETRxxx opcodes here,
1736 because they all imply an unlimited repeat. */
1737
1738 case OP_ALT:
1739 case OP_KET:
1740 case OP_END:
1741 case OP_ACCEPT:
1742 case OP_ASSERT_ACCEPT:
1743 if (length < 0) length = branchlength;
1744 else if (length != branchlength) return -1;
1745 if (*cc != OP_ALT) return length;
1746 cc += 1 + LINK_SIZE;
1747 branchlength = 0;
1748 break;
1749
1750 /* A true recursion implies not fixed length, but a subroutine call may
1751 be OK. If the subroutine is a forward reference, we can't deal with
1752 it until the end of the pattern, so return -3. */
1753
1754 case OP_RECURSE:
1755 if (!atend) return -3;
1756 cs = ce = (pcre_uchar *)cd->start_code + GET(cc, 1); /* Start subpattern */
1757 do ce += GET(ce, 1); while (*ce == OP_ALT); /* End subpattern */
1758 if (cc > cs && cc < ce) return -1; /* Recursion */
1759 d = find_fixedlength(cs + 2, utf, atend, cd);
1760 if (d < 0) return d;
1761 branchlength += d;
1762 cc += 1 + LINK_SIZE;
1763 break;
1764
1765 /* Skip over assertive subpatterns */
1766
1767 case OP_ASSERT:
1768 case OP_ASSERT_NOT:
1769 case OP_ASSERTBACK:
1770 case OP_ASSERTBACK_NOT:
1771 do cc += GET(cc, 1); while (*cc == OP_ALT);
1772 cc += PRIV(OP_lengths)[*cc];
1773 break;
1774
1775 /* Skip over things that don't match chars */
1776
1777 case OP_MARK:
1778 case OP_PRUNE_ARG:
1779 case OP_SKIP_ARG:
1780 case OP_THEN_ARG:
1781 cc += cc[1] + PRIV(OP_lengths)[*cc];
1782 break;
1783
1784 case OP_CALLOUT:
1785 case OP_CIRC:
1786 case OP_CIRCM:
1787 case OP_CLOSE:
1788 case OP_COMMIT:
1789 case OP_CREF:
1790 case OP_DEF:
1791 case OP_DOLL:
1792 case OP_DOLLM:
1793 case OP_EOD:
1794 case OP_EODN:
1795 case OP_FAIL:
1796 case OP_NCREF:
1797 case OP_NRREF:
1798 case OP_NOT_WORD_BOUNDARY:
1799 case OP_PRUNE:
1800 case OP_REVERSE:
1801 case OP_RREF:
1802 case OP_SET_SOM:
1803 case OP_SKIP:
1804 case OP_SOD:
1805 case OP_SOM:
1806 case OP_THEN:
1807 case OP_WORD_BOUNDARY:
1808 cc += PRIV(OP_lengths)[*cc];
1809 break;
1810
1811 /* Handle literal characters */
1812
1813 case OP_CHAR:
1814 case OP_CHARI:
1815 case OP_NOT:
1816 case OP_NOTI:
1817 branchlength++;
1818 cc += 2;
1819 #ifdef SUPPORT_UTF
1820 if (utf && HAS_EXTRALEN(cc[-1])) cc += GET_EXTRALEN(cc[-1]);
1821 #endif
1822 break;
1823
1824 /* Handle exact repetitions. The count is already in characters, but we
1825 need to skip over a multibyte character in UTF8 mode. */
1826
1827 case OP_EXACT:
1828 case OP_EXACTI:
1829 case OP_NOTEXACT:
1830 case OP_NOTEXACTI:
1831 branchlength += GET2(cc,1);
1832 cc += 2 + IMM2_SIZE;
1833 #ifdef SUPPORT_UTF
1834 if (utf && HAS_EXTRALEN(cc[-1])) cc += GET_EXTRALEN(cc[-1]);
1835 #endif
1836 break;
1837
1838 case OP_TYPEEXACT:
1839 branchlength += GET2(cc,1);
1840 if (cc[1 + IMM2_SIZE] == OP_PROP || cc[1 + IMM2_SIZE] == OP_NOTPROP) cc += 2;
1841 cc += 1 + IMM2_SIZE + 1;
1842 break;
1843
1844 /* Handle single-char matchers */
1845
1846 case OP_PROP:
1847 case OP_NOTPROP:
1848 cc += 2;
1849 /* Fall through */
1850
1851 case OP_HSPACE:
1852 case OP_VSPACE:
1853 case OP_NOT_HSPACE:
1854 case OP_NOT_VSPACE:
1855 case OP_NOT_DIGIT:
1856 case OP_DIGIT:
1857 case OP_NOT_WHITESPACE:
1858 case OP_WHITESPACE:
1859 case OP_NOT_WORDCHAR:
1860 case OP_WORDCHAR:
1861 case OP_ANY:
1862 case OP_ALLANY:
1863 branchlength++;
1864 cc++;
1865 break;
1866
1867 /* The single-byte matcher isn't allowed. This only happens in UTF-8 mode;
1868 otherwise \C is coded as OP_ALLANY. */
1869
1870 case OP_ANYBYTE:
1871 return -2;
1872
1873 /* Check a class for variable quantification */
1874
1875 #if defined SUPPORT_UTF || defined COMPILE_PCRE16
1876 case OP_XCLASS:
1877 cc += GET(cc, 1) - PRIV(OP_lengths)[OP_CLASS];
1878 /* Fall through */
1879 #endif
1880
1881 case OP_CLASS:
1882 case OP_NCLASS:
1883 cc += PRIV(OP_lengths)[OP_CLASS];
1884
1885 switch (*cc)
1886 {
1887 case OP_CRPLUS:
1888 case OP_CRMINPLUS:
1889 case OP_CRSTAR:
1890 case OP_CRMINSTAR:
1891 case OP_CRQUERY:
1892 case OP_CRMINQUERY:
1893 return -1;
1894
1895 case OP_CRRANGE:
1896 case OP_CRMINRANGE:
1897 if (GET2(cc,1) != GET2(cc,1+IMM2_SIZE)) return -1;
1898 branchlength += GET2(cc,1);
1899 cc += 1 + 2 * IMM2_SIZE;
1900 break;
1901
1902 default:
1903 branchlength++;
1904 }
1905 break;
1906
1907 /* Anything else is variable length */
1908
1909 case OP_ANYNL:
1910 case OP_BRAMINZERO:
1911 case OP_BRAPOS:
1912 case OP_BRAPOSZERO:
1913 case OP_BRAZERO:
1914 case OP_CBRAPOS:
1915 case OP_EXTUNI:
1916 case OP_KETRMAX:
1917 case OP_KETRMIN:
1918 case OP_KETRPOS:
1919 case OP_MINPLUS:
1920 case OP_MINPLUSI:
1921 case OP_MINQUERY:
1922 case OP_MINQUERYI:
1923 case OP_MINSTAR:
1924 case OP_MINSTARI:
1925 case OP_MINUPTO:
1926 case OP_MINUPTOI:
1927 case OP_NOTMINPLUS:
1928 case OP_NOTMINPLUSI:
1929 case OP_NOTMINQUERY:
1930 case OP_NOTMINQUERYI:
1931 case OP_NOTMINSTAR:
1932 case OP_NOTMINSTARI:
1933 case OP_NOTMINUPTO:
1934 case OP_NOTMINUPTOI:
1935 case OP_NOTPLUS:
1936 case OP_NOTPLUSI:
1937 case OP_NOTPOSPLUS:
1938 case OP_NOTPOSPLUSI:
1939 case OP_NOTPOSQUERY:
1940 case OP_NOTPOSQUERYI:
1941 case OP_NOTPOSSTAR:
1942 case OP_NOTPOSSTARI:
1943 case OP_NOTPOSUPTO:
1944 case OP_NOTPOSUPTOI:
1945 case OP_NOTQUERY:
1946 case OP_NOTQUERYI:
1947 case OP_NOTSTAR:
1948 case OP_NOTSTARI:
1949 case OP_NOTUPTO:
1950 case OP_NOTUPTOI:
1951 case OP_PLUS:
1952 case OP_PLUSI:
1953 case OP_POSPLUS:
1954 case OP_POSPLUSI:
1955 case OP_POSQUERY:
1956 case OP_POSQUERYI:
1957 case OP_POSSTAR:
1958 case OP_POSSTARI:
1959 case OP_POSUPTO:
1960 case OP_POSUPTOI:
1961 case OP_QUERY:
1962 case OP_QUERYI:
1963 case OP_REF:
1964 case OP_REFI:
1965 case OP_SBRA:
1966 case OP_SBRAPOS:
1967 case OP_SCBRA:
1968 case OP_SCBRAPOS:
1969 case OP_SCOND:
1970 case OP_SKIPZERO:
1971 case OP_STAR:
1972 case OP_STARI:
1973 case OP_TYPEMINPLUS:
1974 case OP_TYPEMINQUERY:
1975 case OP_TYPEMINSTAR:
1976 case OP_TYPEMINUPTO:
1977 case OP_TYPEPLUS:
1978 case OP_TYPEPOSPLUS:
1979 case OP_TYPEPOSQUERY:
1980 case OP_TYPEPOSSTAR:
1981 case OP_TYPEPOSUPTO:
1982 case OP_TYPEQUERY:
1983 case OP_TYPESTAR:
1984 case OP_TYPEUPTO:
1985 case OP_UPTO:
1986 case OP_UPTOI:
1987 return -1;
1988
1989 /* Catch unrecognized opcodes so that when new ones are added they
1990 are not forgotten, as has happened in the past. */
1991
1992 default:
1993 return -4;
1994 }
1995 }
1996 /* Control never gets here */
1997 }
1998
1999
2000
2001
2002 /*************************************************
2003 * Scan compiled regex for specific bracket *
2004 *************************************************/
2005
2006 /* This little function scans through a compiled pattern until it finds a
2007 capturing bracket with the given number, or, if the number is negative, an
2008 instance of OP_REVERSE for a lookbehind. The function is global in the C sense
2009 so that it can be called from pcre_study() when finding the minimum matching
2010 length.
2011
2012 Arguments:
2013 code points to start of expression
2014 utf TRUE in UTF-8 / UTF-16 mode
2015 number the required bracket number or negative to find a lookbehind
2016
2017 Returns: pointer to the opcode for the bracket, or NULL if not found
2018 */
2019
2020 const pcre_uchar *
2021 PRIV(find_bracket)(const pcre_uchar *code, BOOL utf, int number)
2022 {
2023 for (;;)
2024 {
2025 register int c = *code;
2026
2027 if (c == OP_END) return NULL;
2028
2029 /* XCLASS is used for classes that cannot be represented just by a bit
2030 map. This includes negated single high-valued characters. The length in
2031 the table is zero; the actual length is stored in the compiled code. */
2032
2033 if (c == OP_XCLASS) code += GET(code, 1);
2034
2035 /* Handle recursion */
2036
2037 else if (c == OP_REVERSE)
2038 {
2039 if (number < 0) return (pcre_uchar *)code;
2040 code += PRIV(OP_lengths)[c];
2041 }
2042
2043 /* Handle capturing bracket */
2044
2045 else if (c == OP_CBRA || c == OP_SCBRA ||
2046 c == OP_CBRAPOS || c == OP_SCBRAPOS)
2047 {
2048 int n = GET2(code, 1+LINK_SIZE);
2049 if (n == number) return (pcre_uchar *)code;
2050 code += PRIV(OP_lengths)[c];
2051 }
2052
2053 /* Otherwise, we can get the item's length from the table, except that for
2054 repeated character types, we have to test for \p and \P, which have an extra
2055 two bytes of parameters, and for MARK/PRUNE/SKIP/THEN with an argument, we
2056 must add in its length. */
2057
2058 else
2059 {
2060 switch(c)
2061 {
2062 case OP_TYPESTAR:
2063 case OP_TYPEMINSTAR:
2064 case OP_TYPEPLUS:
2065 case OP_TYPEMINPLUS:
2066 case OP_TYPEQUERY:
2067 case OP_TYPEMINQUERY:
2068 case OP_TYPEPOSSTAR:
2069 case OP_TYPEPOSPLUS:
2070 case OP_TYPEPOSQUERY:
2071 if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
2072 break;
2073
2074 case OP_TYPEUPTO:
2075 case OP_TYPEMINUPTO:
2076 case OP_TYPEEXACT:
2077 case OP_TYPEPOSUPTO:
2078 if (code[1 + IMM2_SIZE] == OP_PROP
2079 || code[1 + IMM2_SIZE] == OP_NOTPROP) code += 2;
2080 break;
2081
2082 case OP_MARK:
2083 case OP_PRUNE_ARG:
2084 case OP_SKIP_ARG:
2085 code += code[1];
2086 break;
2087
2088 case OP_THEN_ARG:
2089 code += code[1];
2090 break;
2091 }
2092
2093 /* Add in the fixed length from the table */
2094
2095 code += PRIV(OP_lengths)[c];
2096
2097 /* In UTF-8 mode, opcodes that are followed by a character may be followed by
2098 a multi-byte character. The length in the table is a minimum, so we have to
2099 arrange to skip the extra bytes. */
2100
2101 #ifdef SUPPORT_UTF
2102 if (utf) switch(c)
2103 {
2104 case OP_CHAR:
2105 case OP_CHARI:
2106 case OP_EXACT:
2107 case OP_EXACTI:
2108 case OP_UPTO:
2109 case OP_UPTOI:
2110 case OP_MINUPTO:
2111 case OP_MINUPTOI:
2112 case OP_POSUPTO:
2113 case OP_POSUPTOI:
2114 case OP_STAR:
2115 case OP_STARI:
2116 case OP_MINSTAR:
2117 case OP_MINSTARI:
2118 case OP_POSSTAR:
2119 case OP_POSSTARI:
2120 case OP_PLUS:
2121 case OP_PLUSI:
2122 case OP_MINPLUS:
2123 case OP_MINPLUSI:
2124 case OP_POSPLUS:
2125 case OP_POSPLUSI:
2126 case OP_QUERY:
2127 case OP_QUERYI:
2128 case OP_MINQUERY:
2129 case OP_MINQUERYI:
2130 case OP_POSQUERY:
2131 case OP_POSQUERYI:
2132 if (HAS_EXTRALEN(code[-1])) code += GET_EXTRALEN(code[-1]);
2133 break;
2134 }
2135 #else
2136 (void)(utf); /* Keep compiler happy by referencing function argument */
2137 #endif
2138 }
2139 }
2140 }
2141
2142
2143
2144 /*************************************************
2145 * Scan compiled regex for recursion reference *
2146 *************************************************/
2147
2148 /* This little function scans through a compiled pattern until it finds an
2149 instance of OP_RECURSE.
2150
2151 Arguments:
2152 code points to start of expression
2153 utf TRUE in UTF-8 / UTF-16 mode
2154
2155 Returns: pointer to the opcode for OP_RECURSE, or NULL if not found
2156 */
2157
2158 static const pcre_uchar *
2159 find_recurse(const pcre_uchar *code, BOOL utf)
2160 {
2161 for (;;)
2162 {
2163 register int c = *code;
2164 if (c == OP_END) return NULL;
2165 if (c == OP_RECURSE) return code;
2166
2167 /* XCLASS is used for classes that cannot be represented just by a bit
2168 map. This includes negated single high-valued characters. The length in
2169 the table is zero; the actual length is stored in the compiled code. */
2170
2171 if (c == OP_XCLASS) code += GET(code, 1);
2172
2173 /* Otherwise, we can get the item's length from the table, except that for
2174 repeated character types, we have to test for \p and \P, which have an extra
2175 two bytes of parameters, and for MARK/PRUNE/SKIP/THEN with an argument, we
2176 must add in its length. */
2177
2178 else
2179 {
2180 switch(c)
2181 {
2182 case OP_TYPESTAR:
2183 case OP_TYPEMINSTAR:
2184 case OP_TYPEPLUS:
2185 case OP_TYPEMINPLUS:
2186 case OP_TYPEQUERY:
2187 case OP_TYPEMINQUERY:
2188 case OP_TYPEPOSSTAR:
2189 case OP_TYPEPOSPLUS:
2190 case OP_TYPEPOSQUERY:
2191 if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
2192 break;
2193
2194 case OP_TYPEPOSUPTO:
2195 case OP_TYPEUPTO:
2196 case OP_TYPEMINUPTO:
2197 case OP_TYPEEXACT:
2198 if (code[1 + IMM2_SIZE] == OP_PROP
2199 || code[1 + IMM2_SIZE] == OP_NOTPROP) code += 2;
2200 break;
2201
2202 case OP_MARK:
2203 case OP_PRUNE_ARG:
2204 case OP_SKIP_ARG:
2205 code += code[1];
2206 break;
2207
2208 case OP_THEN_ARG:
2209 code += code[1];
2210 break;
2211 }
2212
2213 /* Add in the fixed length from the table */
2214
2215 code += PRIV(OP_lengths)[c];
2216
2217 /* In UTF-8 mode, opcodes that are followed by a character may be followed
2218 by a multi-byte character. The length in the table is a minimum, so we have
2219 to arrange to skip the extra bytes. */
2220
2221 #ifdef SUPPORT_UTF
2222 if (utf) switch(c)
2223 {
2224 case OP_CHAR:
2225 case OP_CHARI:
2226 case OP_EXACT:
2227 case OP_EXACTI:
2228 case OP_UPTO:
2229 case OP_UPTOI:
2230 case OP_MINUPTO:
2231 case OP_MINUPTOI:
2232 case OP_POSUPTO:
2233 case OP_POSUPTOI:
2234 case OP_STAR:
2235 case OP_STARI:
2236 case OP_MINSTAR:
2237 case OP_MINSTARI:
2238 case OP_POSSTAR:
2239 case OP_POSSTARI:
2240 case OP_PLUS:
2241 case OP_PLUSI:
2242 case OP_MINPLUS:
2243 case OP_MINPLUSI:
2244 case OP_POSPLUS:
2245 case OP_POSPLUSI:
2246 case OP_QUERY:
2247 case OP_QUERYI:
2248 case OP_MINQUERY:
2249 case OP_MINQUERYI:
2250 case OP_POSQUERY:
2251 case OP_POSQUERYI:
2252 if (HAS_EXTRALEN(code[-1])) code += GET_EXTRALEN(code[-1]);
2253 break;
2254 }
2255 #else
2256 (void)(utf); /* Keep compiler happy by referencing function argument */
2257 #endif
2258 }
2259 }
2260 }
2261
2262
2263
2264 /*************************************************
2265 * Scan compiled branch for non-emptiness *
2266 *************************************************/
2267
2268 /* This function scans through a branch of a compiled pattern to see whether it
2269 can match the empty string or not. It is called from could_be_empty()
2270 below and from compile_branch() when checking for an unlimited repeat of a
2271 group that can match nothing. Note that first_significant_code() skips over
2272 backward and negative forward assertions when its final argument is TRUE. If we
2273 hit an unclosed bracket, we return "empty" - this means we've struck an inner
2274 bracket whose current branch will already have been scanned.
2275
2276 Arguments:
2277 code points to start of search
2278 endcode points to where to stop
2279 utf TRUE if in UTF-8 / UTF-16 mode
2280 cd contains pointers to tables etc.
2281
2282 Returns: TRUE if what is matched could be empty
2283 */
2284
2285 static BOOL
2286 could_be_empty_branch(const pcre_uchar *code, const pcre_uchar *endcode,
2287 BOOL utf, compile_data *cd)
2288 {
2289 register int c;
2290 for (code = first_significant_code(code + PRIV(OP_lengths)[*code], TRUE);
2291 code < endcode;
2292 code = first_significant_code(code + PRIV(OP_lengths)[c], TRUE))
2293 {
2294 const pcre_uchar *ccode;
2295
2296 c = *code;
2297
2298 /* Skip over forward assertions; the other assertions are skipped by
2299 first_significant_code() with a TRUE final argument. */
2300
2301 if (c == OP_ASSERT)
2302 {
2303 do code += GET(code, 1); while (*code == OP_ALT);
2304 c = *code;
2305 continue;
2306 }
2307
2308 /* For a recursion/subroutine call, if its end has been reached, which
2309 implies a backward reference subroutine call, we can scan it. If it's a
2310 forward reference subroutine call, we can't. To detect forward reference
2311 we have to scan up the list that is kept in the workspace. This function is
2312 called only when doing the real compile, not during the pre-compile that
2313 measures the size of the compiled pattern. */
2314
2315 if (c == OP_RECURSE)
2316 {
2317 const pcre_uchar *scode;
2318 BOOL empty_branch;
2319
2320 /* Test for forward reference */
2321
2322 for (scode = cd->start_workspace; scode < cd->hwm; scode += LINK_SIZE)
2323 if (GET(scode, 0) == code + 1 - cd->start_code) return TRUE;
2324
2325 /* Not a forward reference, test for completed backward reference */
2326
2327 empty_branch = FALSE;
2328 scode = cd->start_code + GET(code, 1);
2329 if (GET(scode, 1) == 0) return TRUE; /* Unclosed */
2330
2331 /* Completed backwards reference */
2332
2333 do
2334 {
2335 if (could_be_empty_branch(scode, endcode, utf, cd))
2336 {
2337 empty_branch = TRUE;
2338 break;
2339 }
2340 scode += GET(scode, 1);
2341 }
2342 while (*scode == OP_ALT);
2343
2344 if (!empty_branch) return FALSE; /* All branches are non-empty */
2345 continue;
2346 }
2347
2348 /* Groups with zero repeats can of course be empty; skip them. */
2349
2350 if (c == OP_BRAZERO || c == OP_BRAMINZERO || c == OP_SKIPZERO ||
2351 c == OP_BRAPOSZERO)
2352 {
2353 code += PRIV(OP_lengths)[c];
2354 do code += GET(code, 1); while (*code == OP_ALT);
2355 c = *code;
2356 continue;
2357 }
2358
2359 /* A nested group that is already marked as "could be empty" can just be
2360 skipped. */
2361
2362 if (c == OP_SBRA || c == OP_SBRAPOS ||
2363 c == OP_SCBRA || c == OP_SCBRAPOS)
2364 {
2365 do code += GET(code, 1); while (*code == OP_ALT);
2366 c = *code;
2367 continue;
2368 }
2369
2370 /* For other groups, scan the branches. */
2371
2372 if (c == OP_BRA || c == OP_BRAPOS ||
2373 c == OP_CBRA || c == OP_CBRAPOS ||
2374 c == OP_ONCE || c == OP_ONCE_NC ||
2375 c == OP_COND)
2376 {
2377 BOOL empty_branch;
2378 if (GET(code, 1) == 0) return TRUE; /* Hit unclosed bracket */
2379
2380 /* If a conditional group has only one branch, there is a second, implied,
2381 empty branch, so just skip over the conditional, because it could be empty.
2382 Otherwise, scan the individual branches of the group. */
2383
2384 if (c == OP_COND && code[GET(code, 1)] != OP_ALT)
2385 code += GET(code, 1);
2386 else
2387 {
2388 empty_branch = FALSE;
2389 do
2390 {
2391 if (!empty_branch && could_be_empty_branch(code, endcode, utf, cd))
2392 empty_branch = TRUE;
2393 code += GET(code, 1);
2394 }
2395 while (*code == OP_ALT);
2396 if (!empty_branch) return FALSE; /* All branches are non-empty */
2397 }
2398
2399 c = *code;
2400 continue;
2401 }
2402
2403 /* Handle the other opcodes */
2404
2405 switch (c)
2406 {
2407 /* Check for quantifiers after a class. XCLASS is used for classes that
2408 cannot be represented just by a bit map. This includes negated single
2409 high-valued characters. The length in PRIV(OP_lengths)[] is zero; the
2410 actual length is stored in the compiled code, so we must update "code"
2411 here. */
2412
2413 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
2414 case OP_XCLASS:
2415 ccode = code += GET(code, 1);
2416 goto CHECK_CLASS_REPEAT;
2417 #endif
2418
2419 case OP_CLASS:
2420 case OP_NCLASS:
2421 ccode = code + PRIV(OP_lengths)[OP_CLASS];
2422
2423 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
2424 CHECK_CLASS_REPEAT:
2425 #endif
2426
2427 switch (*ccode)
2428 {
2429 case OP_CRSTAR: /* These could be empty; continue */
2430 case OP_CRMINSTAR:
2431 case OP_CRQUERY:
2432 case OP_CRMINQUERY:
2433 break;
2434
2435 default: /* Non-repeat => class must match */
2436 case OP_CRPLUS: /* These repeats aren't empty */
2437 case OP_CRMINPLUS:
2438 return FALSE;
2439
2440 case OP_CRRANGE:
2441 case OP_CRMINRANGE:
2442 if (GET2(ccode, 1) > 0) return FALSE; /* Minimum > 0 */
2443 break;
2444 }
2445 break;
2446
2447 /* Opcodes that must match a character */
2448
2449 case OP_PROP:
2450 case OP_NOTPROP:
2451 case OP_EXTUNI:
2452 case OP_NOT_DIGIT:
2453 case OP_DIGIT:
2454 case OP_NOT_WHITESPACE:
2455 case OP_WHITESPACE:
2456 case OP_NOT_WORDCHAR:
2457 case OP_WORDCHAR:
2458 case OP_ANY:
2459 case OP_ALLANY:
2460 case OP_ANYBYTE:
2461 case OP_CHAR:
2462 case OP_CHARI:
2463 case OP_NOT:
2464 case OP_NOTI:
2465 case OP_PLUS:
2466 case OP_MINPLUS:
2467 case OP_POSPLUS:
2468 case OP_EXACT:
2469 case OP_NOTPLUS:
2470 case OP_NOTMINPLUS:
2471 case OP_NOTPOSPLUS:
2472 case OP_NOTEXACT:
2473 case OP_TYPEPLUS:
2474 case OP_TYPEMINPLUS:
2475 case OP_TYPEPOSPLUS:
2476 case OP_TYPEEXACT:
2477 return FALSE;
2478
2479 /* These are going to continue, as they may be empty, but we have to
2480 fudge the length for the \p and \P cases. */
2481
2482 case OP_TYPESTAR:
2483 case OP_TYPEMINSTAR:
2484 case OP_TYPEPOSSTAR:
2485 case OP_TYPEQUERY:
2486 case OP_TYPEMINQUERY:
2487 case OP_TYPEPOSQUERY:
2488 if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
2489 break;
2490
2491 /* Same for these */
2492
2493 case OP_TYPEUPTO:
2494 case OP_TYPEMINUPTO:
2495 case OP_TYPEPOSUPTO:
2496 if (code[1 + IMM2_SIZE] == OP_PROP
2497 || code[1 + IMM2_SIZE] == OP_NOTPROP) code += 2;
2498 break;
2499
2500 /* End of branch */
2501
2502 case OP_KET:
2503 case OP_KETRMAX:
2504 case OP_KETRMIN:
2505 case OP_KETRPOS:
2506 case OP_ALT:
2507 return TRUE;
2508
2509 /* In UTF-8 mode, STAR, MINSTAR, POSSTAR, QUERY, MINQUERY, POSQUERY, UPTO,
2510 MINUPTO, and POSUPTO may be followed by a multibyte character */
2511
2512 #ifdef SUPPORT_UTF
2513 case OP_STAR:
2514 case OP_STARI:
2515 case OP_MINSTAR:
2516 case OP_MINSTARI:
2517 case OP_POSSTAR:
2518 case OP_POSSTARI:
2519 case OP_QUERY:
2520 case OP_QUERYI:
2521 case OP_MINQUERY:
2522 case OP_MINQUERYI:
2523 case OP_POSQUERY:
2524 case OP_POSQUERYI:
2525 if (utf && HAS_EXTRALEN(code[1])) code += GET_EXTRALEN(code[1]);
2526 break;
2527
2528 case OP_UPTO:
2529 case OP_UPTOI:
2530 case OP_MINUPTO:
2531 case OP_MINUPTOI:
2532 case OP_POSUPTO:
2533 case OP_POSUPTOI:
2534 if (utf && HAS_EXTRALEN(code[1 + IMM2_SIZE])) code += GET_EXTRALEN(code[1 + IMM2_SIZE]);
2535 break;
2536 #endif
2537
2538 /* MARK, and PRUNE/SKIP/THEN with an argument must skip over the argument
2539 string. */
2540
2541 case OP_MARK:
2542 case OP_PRUNE_ARG:
2543 case OP_SKIP_ARG:
2544 code += code[1];
2545 break;
2546
2547 case OP_THEN_ARG:
2548 code += code[1];
2549 break;
2550
2551 /* None of the remaining opcodes are required to match a character. */
2552
2553 default:
2554 break;
2555 }
2556 }
2557
2558 return TRUE;
2559 }
2560
2561
2562
2563 /*************************************************
2564 * Scan compiled regex for non-emptiness *
2565 *************************************************/
2566
2567 /* This function is called to check for left recursive calls. We want to check
2568 the current branch of the current pattern to see if it could match the empty
2569 string. If it could, we must look outwards for branches at other levels,
2570 stopping when we pass beyond the bracket which is the subject of the recursion.
2571 This function is called only during the real compile, not during the
2572 pre-compile.
2573
2574 Arguments:
2575 code points to start of the recursion
2576 endcode points to where to stop (current RECURSE item)
2577 bcptr points to the chain of current (unclosed) branch starts
2578 utf TRUE if in UTF-8 / UTF-16 mode
2579 cd pointers to tables etc
2580
2581 Returns: TRUE if what is matched could be empty
2582 */
2583
2584 static BOOL
2585 could_be_empty(const pcre_uchar *code, const pcre_uchar *endcode,
2586 branch_chain *bcptr, BOOL utf, compile_data *cd)
2587 {
2588 while (bcptr != NULL && bcptr->current_branch >= code)
2589 {
2590 if (!could_be_empty_branch(bcptr->current_branch, endcode, utf, cd))
2591 return FALSE;
2592 bcptr = bcptr->outer;
2593 }
2594 return TRUE;
2595 }
2596
2597
2598
2599 /*************************************************
2600 * Check for POSIX class syntax *
2601 *************************************************/
2602
2603 /* This function is called when the sequence "[:" or "[." or "[=" is
2604 encountered in a character class. It checks whether this is followed by a
2605 sequence of characters terminated by a matching ":]" or ".]" or "=]". If we
2606 reach an unescaped ']' without the special preceding character, return FALSE.
2607
2608 Originally, this function only recognized a sequence of letters between the
2609 terminators, but it seems that Perl recognizes any sequence of characters,
2610 though of course unknown POSIX names are subsequently rejected. Perl gives an
2611 "Unknown POSIX class" error for [:f\oo:] for example, where previously PCRE
2612 didn't consider this to be a POSIX class. Likewise for [:1234:].
2613
2614 The problem in trying to be exactly like Perl is in the handling of escapes. We
2615 have to be sure that [abc[:x\]pqr] is *not* treated as containing a POSIX
2616 class, but [abc[:x\]pqr:]] is (so that an error can be generated). The code
2617 below handles the special case of \], but does not try to do any other escape
2618 processing. This makes it different from Perl for cases such as [:l\ower:]
2619 where Perl recognizes it as the POSIX class "lower" but PCRE does not recognize
2620 "l\ower". This is a lesser evil that not diagnosing bad classes when Perl does,
2621 I think.
2622
2623 A user pointed out that PCRE was rejecting [:a[:digit:]] whereas Perl was not.
2624 It seems that the appearance of a nested POSIX class supersedes an apparent
2625 external class. For example, [:a[:digit:]b:] matches "a", "b", ":", or
2626 a digit.
2627
2628 In Perl, unescaped square brackets may also appear as part of class names. For
2629 example, [:a[:abc]b:] gives unknown POSIX class "[:abc]b:]". However, for
2630 [:a[:abc]b][b:] it gives unknown POSIX class "[:abc]b][b:]", which does not
2631 seem right at all. PCRE does not allow closing square brackets in POSIX class
2632 names.
2633
2634 Arguments:
2635 ptr pointer to the initial [
2636 endptr where to return the end pointer
2637
2638 Returns: TRUE or FALSE
2639 */
2640
2641 static BOOL
2642 check_posix_syntax(const pcre_uchar *ptr, const pcre_uchar **endptr)
2643 {
2644 int terminator; /* Don't combine these lines; the Solaris cc */
2645 terminator = *(++ptr); /* compiler warns about "non-constant" initializer. */
2646 for (++ptr; *ptr != 0; ptr++)
2647 {
2648 if (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET)
2649 ptr++;
2650 else if (*ptr == CHAR_RIGHT_SQUARE_BRACKET) return FALSE;
2651 else
2652 {
2653 if (*ptr == terminator && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET)
2654 {
2655 *endptr = ptr;
2656 return TRUE;
2657 }
2658 if (*ptr == CHAR_LEFT_SQUARE_BRACKET &&
2659 (ptr[1] == CHAR_COLON || ptr[1] == CHAR_DOT ||
2660 ptr[1] == CHAR_EQUALS_SIGN) &&
2661 check_posix_syntax(ptr, endptr))
2662 return FALSE;
2663 }
2664 }
2665 return FALSE;
2666 }
2667
2668
2669
2670
2671 /*************************************************
2672 * Check POSIX class name *
2673 *************************************************/
2674
2675 /* This function is called to check the name given in a POSIX-style class entry
2676 such as [:alnum:].
2677
2678 Arguments:
2679 ptr points to the first letter
2680 len the length of the name
2681
2682 Returns: a value representing the name, or -1 if unknown
2683 */
2684
2685 static int
2686 check_posix_name(const pcre_uchar *ptr, int len)
2687 {
2688 const char *pn = posix_names;
2689 register int yield = 0;
2690 while (posix_name_lengths[yield] != 0)
2691 {
2692 if (len == posix_name_lengths[yield] &&
2693 STRNCMP_UC_C8(ptr, pn, len) == 0) return yield;
2694 pn += posix_name_lengths[yield] + 1;
2695 yield++;
2696 }
2697 return -1;
2698 }
2699
2700
2701 /*************************************************
2702 * Adjust OP_RECURSE items in repeated group *
2703 *************************************************/
2704
2705 /* OP_RECURSE items contain an offset from the start of the regex to the group
2706 that is referenced. This means that groups can be replicated for fixed
2707 repetition simply by copying (because the recursion is allowed to refer to
2708 earlier groups that are outside the current group). However, when a group is
2709 optional (i.e. the minimum quantifier is zero), OP_BRAZERO or OP_SKIPZERO is
2710 inserted before it, after it has been compiled. This means that any OP_RECURSE
2711 items within it that refer to the group itself or any contained groups have to
2712 have their offsets adjusted. That one of the jobs of this function. Before it
2713 is called, the partially compiled regex must be temporarily terminated with
2714 OP_END.
2715
2716 This function has been extended with the possibility of forward references for
2717 recursions and subroutine calls. It must also check the list of such references
2718 for the group we are dealing with. If it finds that one of the recursions in
2719 the current group is on this list, it adjusts the offset in the list, not the
2720 value in the reference (which is a group number).
2721
2722 Arguments:
2723 group points to the start of the group
2724 adjust the amount by which the group is to be moved
2725 utf TRUE in UTF-8 / UTF-16 mode
2726 cd contains pointers to tables etc.
2727 save_hwm the hwm forward reference pointer at the start of the group
2728
2729 Returns: nothing
2730 */
2731
2732 static void
2733 adjust_recurse(pcre_uchar *group, int adjust, BOOL utf, compile_data *cd,
2734 pcre_uchar *save_hwm)
2735 {
2736 pcre_uchar *ptr = group;
2737
2738 while ((ptr = (pcre_uchar *)find_recurse(ptr, utf)) != NULL)
2739 {
2740 int offset;
2741 pcre_uchar *hc;
2742
2743 /* See if this recursion is on the forward reference list. If so, adjust the
2744 reference. */
2745
2746 for (hc = save_hwm; hc < cd->hwm; hc += LINK_SIZE)
2747 {
2748 offset = GET(hc, 0);
2749 if (cd->start_code + offset == ptr + 1)
2750 {
2751 PUT(hc, 0, offset + adjust);
2752 break;
2753 }
2754 }
2755
2756 /* Otherwise, adjust the recursion offset if it's after the start of this
2757 group. */
2758
2759 if (hc >= cd->hwm)
2760 {
2761 offset = GET(ptr, 1);
2762 if (cd->start_code + offset >= group) PUT(ptr, 1, offset + adjust);
2763 }
2764
2765 ptr += 1 + LINK_SIZE;
2766 }
2767 }
2768
2769
2770
2771 /*************************************************
2772 * Insert an automatic callout point *
2773 *************************************************/
2774
2775 /* This function is called when the PCRE_AUTO_CALLOUT option is set, to insert
2776 callout points before each pattern item.
2777
2778 Arguments:
2779 code current code pointer
2780 ptr current pattern pointer
2781 cd pointers to tables etc
2782
2783 Returns: new code pointer
2784 */
2785
2786 static pcre_uchar *
2787 auto_callout(pcre_uchar *code, const pcre_uchar *ptr, compile_data *cd)
2788 {
2789 *code++ = OP_CALLOUT;
2790 *code++ = 255;
2791 PUT(code, 0, (int)(ptr - cd->start_pattern)); /* Pattern offset */
2792 PUT(code, LINK_SIZE, 0); /* Default length */
2793 return code + 2 * LINK_SIZE;
2794 }
2795
2796
2797
2798 /*************************************************
2799 * Complete a callout item *
2800 *************************************************/
2801
2802 /* A callout item contains the length of the next item in the pattern, which
2803 we can't fill in till after we have reached the relevant point. This is used
2804 for both automatic and manual callouts.
2805
2806 Arguments:
2807 previous_callout points to previous callout item
2808 ptr current pattern pointer
2809 cd pointers to tables etc
2810
2811 Returns: nothing
2812 */
2813
2814 static void
2815 complete_callout(pcre_uchar *previous_callout, const pcre_uchar *ptr, compile_data *cd)
2816 {
2817 int length = (int)(ptr - cd->start_pattern - GET(previous_callout, 2));
2818 PUT(previous_callout, 2 + LINK_SIZE, length);
2819 }
2820
2821
2822
2823 #ifdef SUPPORT_UCP
2824 /*************************************************
2825 * Get othercase range *
2826 *************************************************/
2827
2828 /* This function is passed the start and end of a class range, in UTF-8 mode
2829 with UCP support. It searches up the characters, looking for internal ranges of
2830 characters in the "other" case. Each call returns the next one, updating the
2831 start address.
2832
2833 Arguments:
2834 cptr points to starting character value; updated
2835 d end value
2836 ocptr where to put start of othercase range
2837 odptr where to put end of othercase range
2838
2839 Yield: TRUE when range returned; FALSE when no more
2840 */
2841
2842 static BOOL
2843 get_othercase_range(unsigned int *cptr, unsigned int d, unsigned int *ocptr,
2844 unsigned int *odptr)
2845 {
2846 unsigned int c, othercase, next;
2847
2848 for (c = *cptr; c <= d; c++)
2849 { if ((othercase = UCD_OTHERCASE(c)) != c) break; }
2850
2851 if (c > d) return FALSE;
2852
2853 *ocptr = othercase;
2854 next = othercase + 1;
2855
2856 for (++c; c <= d; c++)
2857 {
2858 if (UCD_OTHERCASE(c) != next) break;
2859 next++;
2860 }
2861
2862 *odptr = next - 1;
2863 *cptr = c;
2864
2865 return TRUE;
2866 }
2867
2868
2869
2870 /*************************************************
2871 * Check a character and a property *
2872 *************************************************/
2873
2874 /* This function is called by check_auto_possessive() when a property item
2875 is adjacent to a fixed character.
2876
2877 Arguments:
2878 c the character
2879 ptype the property type
2880 pdata the data for the type
2881 negated TRUE if it's a negated property (\P or \p{^)
2882
2883 Returns: TRUE if auto-possessifying is OK
2884 */
2885
2886 static BOOL
2887 check_char_prop(int c, int ptype, int pdata, BOOL negated)
2888 {
2889 const ucd_record *prop = GET_UCD(c);
2890 switch(ptype)
2891 {
2892 case PT_LAMP:
2893 return (prop->chartype == ucp_Lu ||
2894 prop->chartype == ucp_Ll ||
2895 prop->chartype == ucp_Lt) == negated;
2896
2897 case PT_GC:
2898 return (pdata == PRIV(ucp_gentype)[prop->chartype]) == negated;
2899
2900 case PT_PC:
2901 return (pdata == prop->chartype) == negated;
2902
2903 case PT_SC:
2904 return (pdata == prop->script) == negated;
2905
2906 /* These are specials */
2907
2908 case PT_ALNUM:
2909 return (PRIV(ucp_gentype)[prop->chartype] == ucp_L ||
2910 PRIV(ucp_gentype)[prop->chartype] == ucp_N) == negated;
2911
2912 case PT_SPACE: /* Perl space */
2913 return (PRIV(ucp_gentype)[prop->chartype] == ucp_Z ||
2914 c == CHAR_HT || c == CHAR_NL || c == CHAR_FF || c == CHAR_CR)
2915 == negated;
2916
2917 case PT_PXSPACE: /* POSIX space */
2918 return (PRIV(ucp_gentype)[prop->chartype] == ucp_Z ||
2919 c == CHAR_HT || c == CHAR_NL || c == CHAR_VT ||
2920 c == CHAR_FF || c == CHAR_CR)
2921 == negated;
2922
2923 case PT_WORD:
2924 return (PRIV(ucp_gentype)[prop->chartype] == ucp_L ||
2925 PRIV(ucp_gentype)[prop->chartype] == ucp_N ||
2926 c == CHAR_UNDERSCORE) == negated;
2927 }
2928 return FALSE;
2929 }
2930 #endif /* SUPPORT_UCP */
2931
2932
2933
2934 /*************************************************
2935 * Check if auto-possessifying is possible *
2936 *************************************************/
2937
2938 /* This function is called for unlimited repeats of certain items, to see
2939 whether the next thing could possibly match the repeated item. If not, it makes
2940 sense to automatically possessify the repeated item.
2941
2942 Arguments:
2943 previous pointer to the repeated opcode
2944 utf TRUE in UTF-8 / UTF-16 mode
2945 ptr next character in pattern
2946 options options bits
2947 cd contains pointers to tables etc.
2948
2949 Returns: TRUE if possessifying is wanted
2950 */
2951
2952 static BOOL
2953 check_auto_possessive(const pcre_uchar *previous, BOOL utf,
2954 const pcre_uchar *ptr, int options, compile_data *cd)
2955 {
2956 pcre_int32 c, next;
2957 int op_code = *previous++;
2958
2959 /* Skip whitespace and comments in extended mode */
2960
2961 if ((options & PCRE_EXTENDED) != 0)
2962 {
2963 for (;;)
2964 {
2965 while (MAX_255(*ptr) && (cd->ctypes[*ptr] & ctype_space) != 0) ptr++;
2966 if (*ptr == CHAR_NUMBER_SIGN)
2967 {
2968 ptr++;
2969 while (*ptr != 0)
2970 {
2971 if (IS_NEWLINE(ptr)) { ptr += cd->nllen; break; }
2972 ptr++;
2973 #ifdef SUPPORT_UTF
2974 if (utf) FORWARDCHAR(ptr);
2975 #endif
2976 }
2977 }
2978 else break;
2979 }
2980 }
2981
2982 /* If the next item is one that we can handle, get its value. A non-negative
2983 value is a character, a negative value is an escape value. */
2984
2985 if (*ptr == CHAR_BACKSLASH)
2986 {
2987 int temperrorcode = 0;
2988 next = check_escape(&ptr, &temperrorcode, cd->bracount, options, FALSE);
2989 if (temperrorcode != 0) return FALSE;
2990 ptr++; /* Point after the escape sequence */
2991 }
2992 else if (!MAX_255(*ptr) || (cd->ctypes[*ptr] & ctype_meta) == 0)
2993 {
2994 #ifdef SUPPORT_UTF
2995 if (utf) { GETCHARINC(next, ptr); } else
2996 #endif
2997 next = *ptr++;
2998 }
2999 else return FALSE;
3000
3001 /* Skip whitespace and comments in extended mode */
3002
3003 if ((options & PCRE_EXTENDED) != 0)
3004 {
3005 for (;;)
3006 {
3007 while (MAX_255(*ptr) && (cd->ctypes[*ptr] & ctype_space) != 0) ptr++;
3008 if (*ptr == CHAR_NUMBER_SIGN)
3009 {
3010 ptr++;
3011 while (*ptr != 0)
3012 {
3013 if (IS_NEWLINE(ptr)) { ptr += cd->nllen; break; }
3014 ptr++;
3015 #ifdef SUPPORT_UTF
3016 if (utf) FORWARDCHAR(ptr);
3017 #endif
3018 }
3019 }
3020 else break;
3021 }
3022 }
3023
3024 /* If the next thing is itself optional, we have to give up. */
3025
3026 if (*ptr == CHAR_ASTERISK || *ptr == CHAR_QUESTION_MARK ||
3027 STRNCMP_UC_C8(ptr, STR_LEFT_CURLY_BRACKET STR_0 STR_COMMA, 3) == 0)
3028 return FALSE;
3029
3030 /* Now compare the next item with the previous opcode. First, handle cases when
3031 the next item is a character. */
3032
3033 if (next >= 0) switch(op_code)
3034 {
3035 case OP_CHAR:
3036 #ifdef SUPPORT_UTF
3037 GETCHARTEST(c, previous);
3038 #else
3039 c = *previous;
3040 #endif
3041 return c != next;
3042
3043 /* For CHARI (caseless character) we must check the other case. If we have
3044 Unicode property support, we can use it to test the other case of
3045 high-valued characters. */
3046
3047 case OP_CHARI:
3048 #ifdef SUPPORT_UTF
3049 GETCHARTEST(c, previous);
3050 #else
3051 c = *previous;
3052 #endif
3053 if (c == next) return FALSE;
3054 #ifdef SUPPORT_UTF
3055 if (utf)
3056 {
3057 unsigned int othercase;
3058 if (next < 128) othercase = cd->fcc[next]; else
3059 #ifdef SUPPORT_UCP
3060 othercase = UCD_OTHERCASE((unsigned int)next);
3061 #else
3062 othercase = NOTACHAR;
3063 #endif
3064 return (unsigned int)c != othercase;
3065 }
3066 else
3067 #endif /* SUPPORT_UTF */
3068 return (c != TABLE_GET(next, cd->fcc, next)); /* Non-UTF-8 mode */
3069
3070 /* For OP_NOT and OP_NOTI, the data is always a single-byte character. These
3071 opcodes are not used for multi-byte characters, because they are coded using
3072 an XCLASS instead. */
3073
3074 case OP_NOT:
3075 return (c = *previous) == next;
3076
3077 case OP_NOTI:
3078 if ((c = *previous) == next) return TRUE;
3079 #ifdef SUPPORT_UTF
3080 if (utf)
3081 {
3082 unsigned int othercase;
3083 if (next < 128) othercase = cd->fcc[next]; else
3084 #ifdef SUPPORT_UCP
3085 othercase = UCD_OTHERCASE(next);
3086 #else
3087 othercase = NOTACHAR;
3088 #endif
3089 return (unsigned int)c == othercase;
3090 }
3091 else
3092 #endif /* SUPPORT_UTF */
3093 return (c == TABLE_GET(next, cd->fcc, next)); /* Non-UTF-8 mode */
3094
3095 /* Note that OP_DIGIT etc. are generated only when PCRE_UCP is *not* set.
3096 When it is set, \d etc. are converted into OP_(NOT_)PROP codes. */
3097
3098 case OP_DIGIT:
3099 return next > 127 || (cd->ctypes[next] & ctype_digit) == 0;
3100
3101 case OP_NOT_DIGIT:
3102 return next <= 127 && (cd->ctypes[next] & ctype_digit) != 0;
3103
3104 case OP_WHITESPACE:
3105 return next > 127 || (cd->ctypes[next] & ctype_space) == 0;
3106
3107 case OP_NOT_WHITESPACE:
3108 return next <= 127 && (cd->ctypes[next] & ctype_space) != 0;
3109
3110 case OP_WORDCHAR:
3111 return next > 127 || (cd->ctypes[next] & ctype_word) == 0;
3112
3113 case OP_NOT_WORDCHAR:
3114 return next <= 127 && (cd->ctypes[next] & ctype_word) != 0;
3115
3116 case OP_HSPACE:
3117 case OP_NOT_HSPACE:
3118 switch(next)
3119 {
3120 case 0x09:
3121 case 0x20:
3122 case 0xa0:
3123 case 0x1680:
3124 case 0x180e:
3125 case 0x2000:
3126 case 0x2001:
3127 case 0x2002:
3128 case 0x2003:
3129 case 0x2004:
3130 case 0x2005:
3131 case 0x2006:
3132 case 0x2007:
3133 case 0x2008:
3134 case 0x2009:
3135 case 0x200A:
3136 case 0x202f:
3137 case 0x205f:
3138 case 0x3000:
3139 return op_code == OP_NOT_HSPACE;
3140 default:
3141 return op_code != OP_NOT_HSPACE;
3142 }
3143
3144 case OP_ANYNL:
3145 case OP_VSPACE:
3146 case OP_NOT_VSPACE:
3147 switch(next)
3148 {
3149 case 0x0a:
3150 case 0x0b:
3151 case 0x0c:
3152 case 0x0d:
3153 case 0x85:
3154 case 0x2028:
3155 case 0x2029:
3156 return op_code == OP_NOT_VSPACE;
3157 default:
3158 return op_code != OP_NOT_VSPACE;
3159 }
3160
3161 #ifdef SUPPORT_UCP
3162 case OP_PROP:
3163 return check_char_prop(next, previous[0], previous[1], FALSE);
3164
3165 case OP_NOTPROP:
3166 return check_char_prop(next, previous[0], previous[1], TRUE);
3167 #endif
3168
3169 default:
3170 return FALSE;
3171 }
3172
3173
3174 /* Handle the case when the next item is \d, \s, etc. Note that when PCRE_UCP
3175 is set, \d turns into ESC_du rather than ESC_d, etc., so ESC_d etc. are
3176 generated only when PCRE_UCP is *not* set, that is, when only ASCII
3177 characteristics are recognized. Similarly, the opcodes OP_DIGIT etc. are
3178 replaced by OP_PROP codes when PCRE_UCP is set. */
3179
3180 switch(op_code)
3181 {
3182 case OP_CHAR:
3183 case OP_CHARI:
3184 #ifdef SUPPORT_UTF
3185 GETCHARTEST(c, previous);
3186 #else
3187 c = *previous;
3188 #endif
3189 switch(-next)
3190 {
3191 case ESC_d:
3192 return c > 127 || (cd->ctypes[c] & ctype_digit) == 0;
3193
3194 case ESC_D:
3195 return c <= 127 && (cd->ctypes[c] & ctype_digit) != 0;
3196
3197 case ESC_s:
3198 return c > 127 || (cd->ctypes[c] & ctype_space) == 0;
3199
3200 case ESC_S:
3201 return c <= 127 && (cd->ctypes[c] & ctype_space) != 0;
3202
3203 case ESC_w:
3204 return c > 127 || (cd->ctypes[c] & ctype_word) == 0;
3205
3206 case ESC_W:
3207 return c <= 127 && (cd->ctypes[c] & ctype_word) != 0;
3208
3209 case ESC_h:
3210 case ESC_H:
3211 switch(c)
3212 {
3213 case 0x09:
3214 case 0x20:
3215 case 0xa0:
3216 case 0x1680:
3217 case 0x180e:
3218 case 0x2000:
3219 case 0x2001:
3220 case 0x2002:
3221 case 0x2003:
3222 case 0x2004:
3223 case 0x2005:
3224 case 0x2006:
3225 case 0x2007:
3226 case 0x2008:
3227 case 0x2009:
3228 case 0x200A:
3229 case 0x202f:
3230 case 0x205f:
3231 case 0x3000:
3232 return -next != ESC_h;
3233 default:
3234 return -next == ESC_h;
3235 }
3236
3237 case ESC_v:
3238 case ESC_V:
3239 switch(c)
3240 {
3241 case 0x0a:
3242 case 0x0b:
3243 case 0x0c:
3244 case 0x0d:
3245 case 0x85:
3246 case 0x2028:
3247 case 0x2029:
3248 return -next != ESC_v;
3249 default:
3250 return -next == ESC_v;
3251 }
3252
3253 /* When PCRE_UCP is set, these values get generated for \d etc. Find
3254 their substitutions and process them. The result will always be either
3255 -ESC_p or -ESC_P. Then fall through to process those values. */
3256
3257 #ifdef SUPPORT_UCP
3258 case ESC_du:
3259 case ESC_DU:
3260 case ESC_wu:
3261 case ESC_WU:
3262 case ESC_su:
3263 case ESC_SU:
3264 {
3265 int temperrorcode = 0;
3266 ptr = substitutes[-next - ESC_DU];
3267 next = check_escape(&ptr, &temperrorcode, 0, options, FALSE);
3268 if (temperrorcode != 0) return FALSE;
3269 ptr++; /* For compatibility */
3270 }
3271 /* Fall through */
3272
3273 case ESC_p:
3274 case ESC_P:
3275 {
3276 int ptype, pdata, errorcodeptr;
3277 BOOL negated;
3278
3279 ptr--; /* Make ptr point at the p or P */
3280 ptype = get_ucp(&ptr, &negated, &pdata, &errorcodeptr);
3281 if (ptype < 0) return FALSE;
3282 ptr++; /* Point past the final curly ket */
3283
3284 /* If the property item is optional, we have to give up. (When generated
3285 from \d etc by PCRE_UCP, this test will have been applied much earlier,
3286 to the original \d etc. At this point, ptr will point to a zero byte. */
3287
3288 if (*ptr == CHAR_ASTERISK || *ptr == CHAR_QUESTION_MARK ||
3289 STRNCMP_UC_C8(ptr, STR_LEFT_CURLY_BRACKET STR_0 STR_COMMA, 3) == 0)
3290 return FALSE;
3291
3292 /* Do the property check. */
3293
3294 return check_char_prop(c, ptype, pdata, (next == -ESC_P) != negated);
3295 }
3296 #endif
3297
3298 default:
3299 return FALSE;
3300 }
3301
3302 /* In principle, support for Unicode properties should be integrated here as
3303 well. It means re-organizing the above code so as to get hold of the property
3304 values before switching on the op-code. However, I wonder how many patterns
3305 combine ASCII \d etc with Unicode properties? (Note that if PCRE_UCP is set,
3306 these op-codes are never generated.) */
3307
3308 case OP_DIGIT:
3309 return next == -ESC_D || next == -ESC_s || next == -ESC_W ||
3310 next == -ESC_h || next == -ESC_v || next == -ESC_R;
3311
3312 case OP_NOT_DIGIT:
3313 return next == -ESC_d;
3314
3315 case OP_WHITESPACE:
3316 return next == -ESC_S || next == -ESC_d || next == -ESC_w || next == -ESC_R;
3317
3318 case OP_NOT_WHITESPACE:
3319 return next == -ESC_s || next == -ESC_h || next == -ESC_v;
3320
3321 case OP_HSPACE:
3322 return next == -ESC_S || next == -ESC_H || next == -ESC_d ||
3323 next == -ESC_w || next == -ESC_v || next == -ESC_R;
3324
3325 case OP_NOT_HSPACE:
3326 return next == -ESC_h;
3327
3328 /* Can't have \S in here because VT matches \S (Perl anomaly) */
3329 case OP_ANYNL:
3330 case OP_VSPACE:
3331 return next == -ESC_V || next == -ESC_d || next == -ESC_w;
3332
3333 case OP_NOT_VSPACE:
3334 return next == -ESC_v || next == -ESC_R;
3335
3336 case OP_WORDCHAR:
3337 return next == -ESC_W || next == -ESC_s || next == -ESC_h ||
3338 next == -ESC_v || next == -ESC_R;
3339
3340 case OP_NOT_WORDCHAR:
3341 return next == -ESC_w || next == -ESC_d;
3342
3343 default:
3344 return FALSE;
3345 }
3346
3347 /* Control does not reach here */
3348 }
3349
3350
3351
3352 /*************************************************
3353 * Compile one branch *
3354 *************************************************/
3355
3356 /* Scan the pattern, compiling it into the a vector. If the options are
3357 changed during the branch, the pointer is used to change the external options
3358 bits. This function is used during the pre-compile phase when we are trying
3359 to find out the amount of memory needed, as well as during the real compile
3360 phase. The value of lengthptr distinguishes the two phases.
3361
3362 Arguments:
3363 optionsptr pointer to the option bits
3364 codeptr points to the pointer to the current code point
3365 ptrptr points to the current pattern pointer
3366 errorcodeptr points to error code variable
3367 firstcharptr set to initial literal character, or < 0 (REQ_UNSET, REQ_NONE)
3368 reqcharptr set to the last literal character required, else < 0
3369 bcptr points to current branch chain
3370 cond_depth conditional nesting depth
3371 cd contains pointers to tables etc.
3372 lengthptr NULL during the real compile phase
3373 points to length accumulator during pre-compile phase
3374
3375 Returns: TRUE on success
3376 FALSE, with *errorcodeptr set non-zero on error
3377 */
3378
3379 static BOOL
3380 compile_branch(int *optionsptr, pcre_uchar **codeptr,
3381 const pcre_uchar **ptrptr, int *errorcodeptr, pcre_int32 *firstcharptr,
3382 pcre_int32 *reqcharptr, branch_chain *bcptr, int cond_depth,
3383 compile_data *cd, int *lengthptr)
3384 {
3385 int repeat_type, op_type;
3386 int repeat_min = 0, repeat_max = 0; /* To please picky compilers */
3387 int bravalue = 0;
3388 int greedy_default, greedy_non_default;
3389 pcre_int32 firstchar, reqchar;
3390 pcre_int32 zeroreqchar, zerofirstchar;
3391 pcre_int32 req_caseopt, reqvary, tempreqvary;
3392 int options = *optionsptr; /* May change dynamically */
3393 int after_manual_callout = 0;
3394 int length_prevgroup = 0;
3395 register int c;
3396 register pcre_uchar *code = *codeptr;
3397 pcre_uchar *last_code = code;
3398 pcre_uchar *orig_code = code;
3399 pcre_uchar *tempcode;
3400 BOOL inescq = FALSE;
3401 BOOL groupsetfirstchar = FALSE;
3402 const pcre_uchar *ptr = *ptrptr;
3403 const pcre_uchar *tempptr;
3404 const pcre_uchar *nestptr = NULL;
3405 pcre_uchar *previous = NULL;
3406 pcre_uchar *previous_callout = NULL;
3407 pcre_uchar *save_hwm = NULL;
3408 pcre_uint8 classbits[32];
3409
3410 /* We can fish out the UTF-8 setting once and for all into a BOOL, but we
3411 must not do this for other options (e.g. PCRE_EXTENDED) because they may change
3412 dynamically as we process the pattern. */
3413
3414 #ifdef SUPPORT_UTF
3415 /* PCRE_UTF16 has the same value as PCRE_UTF8. */
3416 BOOL utf = (options & PCRE_UTF8) != 0;
3417 pcre_uchar utf_chars[6];
3418 #else
3419 BOOL utf = FALSE;
3420 #endif
3421
3422 /* Helper variables for OP_XCLASS opcode (for characters > 255). */
3423
3424 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
3425 BOOL xclass;
3426 pcre_uchar *class_uchardata;
3427 pcre_uchar *class_uchardata_base;
3428 #endif
3429
3430 #ifdef PCRE_DEBUG
3431 if (lengthptr != NULL) DPRINTF((">> start branch\n"));
3432 #endif
3433
3434 /* Set up the default and non-default settings for greediness */
3435
3436 greedy_default = ((options & PCRE_UNGREEDY) != 0);
3437 greedy_non_default = greedy_default ^ 1;
3438
3439 /* Initialize no first byte, no required byte. REQ_UNSET means "no char
3440 matching encountered yet". It gets changed to REQ_NONE if we hit something that
3441 matches a non-fixed char first char; reqchar just remains unset if we never
3442 find one.
3443
3444 When we hit a repeat whose minimum is zero, we may have to adjust these values
3445 to take the zero repeat into account. This is implemented by setting them to
3446 zerofirstbyte and zeroreqchar when such a repeat is encountered. The individual
3447 item types that can be repeated set these backoff variables appropriately. */
3448
3449 firstchar = reqchar = zerofirstchar = zeroreqchar = REQ_UNSET;
3450
3451 /* The variable req_caseopt contains either the REQ_CASELESS value
3452 or zero, according to the current setting of the caseless flag. The
3453 REQ_CASELESS leaves the lower 28 bit empty. It is added into the
3454 firstchar or reqchar variables to record the case status of the
3455 value. This is used only for ASCII characters. */
3456
3457 req_caseopt = ((options & PCRE_CASELESS) != 0)? REQ_CASELESS:0;
3458
3459 /* Switch on next character until the end of the branch */
3460
3461 for (;; ptr++)
3462 {
3463 BOOL negate_class;
3464 BOOL should_flip_negation;
3465 BOOL possessive_quantifier;
3466 BOOL is_quantifier;
3467 BOOL is_recurse;
3468 BOOL reset_bracount;
3469 int class_has_8bitchar;
3470 int class_single_char;
3471 int newoptions;
3472 int recno;
3473 int refsign;
3474 int skipbytes;
3475 int subreqchar;
3476 int subfirstchar;
3477 int terminator;
3478 int mclength;
3479 int tempbracount;
3480 pcre_uchar mcbuffer[8];
3481
3482 /* Get next character in the pattern */
3483
3484 c = *ptr;
3485
3486 /* If we are at the end of a nested substitution, revert to the outer level
3487 string. Nesting only happens one level deep. */
3488
3489 if (c == 0 && nestptr != NULL)
3490 {
3491 ptr = nestptr;
3492 nestptr = NULL;
3493 c = *ptr;
3494 }
3495
3496 /* If we are in the pre-compile phase, accumulate the length used for the
3497 previous cycle of this loop. */
3498
3499 if (lengthptr != NULL)
3500 {
3501 #ifdef PCRE_DEBUG
3502 if (code > cd->hwm) cd->hwm = code; /* High water info */
3503 #endif
3504 if (code > cd->start_workspace + cd->workspace_size -
3505 WORK_SIZE_SAFETY_MARGIN) /* Check for overrun */
3506 {
3507 *errorcodeptr = ERR52;
3508 goto FAILED;
3509 }
3510
3511 /* There is at least one situation where code goes backwards: this is the
3512 case of a zero quantifier after a class (e.g. [ab]{0}). At compile time,
3513 the class is simply eliminated. However, it is created first, so we have to
3514 allow memory for it. Therefore, don't ever reduce the length at this point.
3515 */
3516
3517 if (code < last_code) code = last_code;
3518
3519 /* Paranoid check for integer overflow */
3520
3521 if (OFLOW_MAX - *lengthptr < code - last_code)
3522 {
3523 *errorcodeptr = ERR20;
3524 goto FAILED;
3525 }
3526
3527 *lengthptr += (int)(code - last_code);
3528 DPRINTF(("length=%d added %d c=%c (0x%x)\n", *lengthptr,
3529 (int)(code - last_code), c, c));
3530
3531 /* If "previous" is set and it is not at the start of the work space, move
3532 it back to there, in order to avoid filling up the work space. Otherwise,
3533 if "previous" is NULL, reset the current code pointer to the start. */
3534
3535 if (previous != NULL)
3536 {
3537 if (previous > orig_code)
3538 {
3539 memmove(orig_code, previous, IN_UCHARS(code - previous));
3540 code -= previous - orig_code;
3541 previous = orig_code;
3542 }
3543 }
3544 else code = orig_code;
3545
3546 /* Remember where this code item starts so we can pick up the length
3547 next time round. */
3548
3549 last_code = code;
3550 }
3551
3552 /* In the real compile phase, just check the workspace used by the forward
3553 reference list. */
3554
3555 else if (cd->hwm > cd->start_workspace + cd->workspace_size -
3556 WORK_SIZE_SAFETY_MARGIN)
3557 {
3558 *errorcodeptr = ERR52;
3559 goto FAILED;
3560 }
3561
3562 /* If in \Q...\E, check for the end; if not, we have a literal */
3563
3564 if (inescq && c != 0)
3565 {
3566 if (c == CHAR_BACKSLASH && ptr[1] == CHAR_E)
3567 {
3568 inescq = FALSE;
3569 ptr++;
3570 continue;
3571 }
3572 else
3573 {
3574 if (previous_callout != NULL)
3575 {
3576 if (lengthptr == NULL) /* Don't attempt in pre-compile phase */
3577 complete_callout(previous_callout, ptr, cd);
3578 previous_callout = NULL;
3579 }
3580 if ((options & PCRE_AUTO_CALLOUT) != 0)
3581 {
3582 previous_callout = code;
3583 code = auto_callout(code, ptr, cd);
3584 }
3585 goto NORMAL_CHAR;
3586 }
3587 }
3588
3589 /* Fill in length of a previous callout, except when the next thing is
3590 a quantifier. */
3591
3592 is_quantifier =
3593 c == CHAR_ASTERISK || c == CHAR_PLUS || c == CHAR_QUESTION_MARK ||
3594 (c == CHAR_LEFT_CURLY_BRACKET && is_counted_repeat(ptr+1));
3595
3596 if (!is_quantifier && previous_callout != NULL &&
3597 after_manual_callout-- <= 0)
3598 {
3599 if (lengthptr == NULL) /* Don't attempt in pre-compile phase */
3600 complete_callout(previous_callout, ptr, cd);
3601 previous_callout = NULL;
3602 }
3603
3604 /* In extended mode, skip white space and comments. */
3605
3606 if ((options & PCRE_EXTENDED) != 0)
3607 {
3608 if (MAX_255(*ptr) && (cd->ctypes[c] & ctype_space) != 0) continue;
3609 if (c == CHAR_NUMBER_SIGN)
3610 {
3611 ptr++;
3612 while (*ptr != 0)
3613 {
3614 if (IS_NEWLINE(ptr)) { ptr += cd->nllen - 1; break; }
3615 ptr++;
3616 #ifdef SUPPORT_UTF
3617 if (utf) FORWARDCHAR(ptr);
3618 #endif
3619 }
3620 if (*ptr != 0) continue;
3621
3622 /* Else fall through to handle end of string */
3623 c = 0;
3624 }
3625 }
3626
3627 /* No auto callout for quantifiers. */
3628
3629 if ((options & PCRE_AUTO_CALLOUT) != 0 && !is_quantifier)
3630 {
3631 previous_callout = code;
3632 code = auto_callout(code, ptr, cd);
3633 }
3634
3635 switch(c)
3636 {
3637 /* ===================================================================*/
3638 case 0: /* The branch terminates at string end */
3639 case CHAR_VERTICAL_LINE: /* or | or ) */
3640 case CHAR_RIGHT_PARENTHESIS:
3641 *firstcharptr = firstchar;
3642 *reqcharptr = reqchar;
3643 *codeptr = code;
3644 *ptrptr = ptr;
3645 if (lengthptr != NULL)
3646 {
3647 if (OFLOW_MAX - *lengthptr < code - last_code)
3648 {
3649 *errorcodeptr = ERR20;
3650 goto FAILED;
3651 }
3652 *lengthptr += (int)(code - last_code); /* To include callout length */
3653 DPRINTF((">> end branch\n"));
3654 }
3655 return TRUE;
3656
3657
3658 /* ===================================================================*/
3659 /* Handle single-character metacharacters. In multiline mode, ^ disables
3660 the setting of any following char as a first character. */
3661
3662 case CHAR_CIRCUMFLEX_ACCENT:
3663 previous = NULL;
3664 if ((options & PCRE_MULTILINE) != 0)
3665 {
3666 if (firstchar == REQ_UNSET) firstchar = REQ_NONE;
3667 *code++ = OP_CIRCM;
3668 }
3669 else *code++ = OP_CIRC;
3670 break;
3671
3672 case CHAR_DOLLAR_SIGN:
3673 previous = NULL;
3674 *code++ = ((options & PCRE_MULTILINE) != 0)? OP_DOLLM : OP_DOLL;
3675 break;
3676
3677 /* There can never be a first char if '.' is first, whatever happens about
3678 repeats. The value of reqchar doesn't change either. */
3679
3680 case CHAR_DOT:
3681 if (firstchar == REQ_UNSET) firstchar = REQ_NONE;
3682 zerofirstchar = firstchar;
3683 zeroreqchar = reqchar;
3684 previous = code;
3685 *code++ = ((options & PCRE_DOTALL) != 0)? OP_ALLANY: OP_ANY;
3686 break;
3687
3688
3689 /* ===================================================================*/
3690 /* Character classes. If the included characters are all < 256, we build a
3691 32-byte bitmap of the permitted characters, except in the special case
3692 where there is only one such character. For negated classes, we build the
3693 map as usual, then invert it at the end. However, we use a different opcode
3694 so that data characters > 255 can be handled correctly.
3695
3696 If the class contains characters outside the 0-255 range, a different
3697 opcode is compiled. It may optionally have a bit map for characters < 256,
3698 but those above are are explicitly listed afterwards. A flag byte tells
3699 whether the bitmap is present, and whether this is a negated class or not.
3700
3701 In JavaScript compatibility mode, an isolated ']' causes an error. In
3702 default (Perl) mode, it is treated as a data character. */
3703
3704 case CHAR_RIGHT_SQUARE_BRACKET:
3705 if ((cd->external_options & PCRE_JAVASCRIPT_COMPAT) != 0)
3706 {
3707 *errorcodeptr = ERR64;
3708 goto FAILED;
3709 }
3710 goto NORMAL_CHAR;
3711
3712 case CHAR_LEFT_SQUARE_BRACKET:
3713 previous = code;
3714
3715 /* PCRE supports POSIX class stuff inside a class. Perl gives an error if
3716 they are encountered at the top level, so we'll do that too. */
3717
3718 if ((ptr[1] == CHAR_COLON || ptr[1] == CHAR_DOT ||
3719 ptr[1] == CHAR_EQUALS_SIGN) &&
3720 check_posix_syntax(ptr, &tempptr))
3721 {
3722 *errorcodeptr = (ptr[1] == CHAR_COLON)? ERR13 : ERR31;
3723 goto FAILED;
3724 }
3725
3726 /* If the first character is '^', set the negation flag and skip it. Also,
3727 if the first few characters (either before or after ^) are \Q\E or \E we
3728 skip them too. This makes for compatibility with Perl. */
3729
3730 negate_class = FALSE;
3731 for (;;)
3732 {
3733 c = *(++ptr);
3734 if (c == CHAR_BACKSLASH)
3735 {
3736 if (ptr[1] == CHAR_E)
3737 ptr++;
3738 else if (STRNCMP_UC_C8(ptr + 1, STR_Q STR_BACKSLASH STR_E, 3) == 0)
3739 ptr += 3;
3740 else
3741 break;
3742 }
3743 else if (!negate_class && c == CHAR_CIRCUMFLEX_ACCENT)
3744 negate_class = TRUE;
3745 else break;
3746 }
3747
3748 /* Empty classes are allowed in JavaScript compatibility mode. Otherwise,
3749 an initial ']' is taken as a data character -- the code below handles
3750 that. In JS mode, [] must always fail, so generate OP_FAIL, whereas
3751 [^] must match any character, so generate OP_ALLANY. */
3752
3753 if (c == CHAR_RIGHT_SQUARE_BRACKET &&
3754 (cd->external_options & PCRE_JAVASCRIPT_COMPAT) != 0)
3755 {
3756 *code++ = negate_class? OP_ALLANY : OP_FAIL;
3757 if (firstchar == REQ_UNSET) firstchar = REQ_NONE;
3758 zerofirstchar = firstchar;
3759 break;
3760 }
3761
3762 /* If a class contains a negative special such as \S, we need to flip the
3763 negation flag at the end, so that support for characters > 255 works
3764 correctly (they are all included in the class). */
3765
3766 should_flip_negation = FALSE;
3767
3768 /* For optimization purposes, we track some properties of the class.
3769 class_has_8bitchar will be non-zero, if the class contains at least one
3770 < 256 character. class_single_char will be 1 if the class contains only
3771 a single character. */
3772
3773 class_has_8bitchar = 0;
3774 class_single_char = 0;
3775
3776 /* Initialize the 32-char bit map to all zeros. We build the map in a
3777 temporary bit of memory, in case the class contains only 1 character (less
3778 than 256), because in that case the compiled code doesn't use the bit map.
3779 */
3780
3781 memset(classbits, 0, 32 * sizeof(pcre_uint8));
3782
3783 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
3784 xclass = FALSE; /* No chars >= 256 */
3785 class_uchardata = code + LINK_SIZE + 2; /* For UTF-8 items */
3786 class_uchardata_base = class_uchardata; /* For resetting in pass 1 */
3787 #endif
3788
3789 /* Process characters until ] is reached. By writing this as a "do" it
3790 means that an initial ] is taken as a data character. At the start of the
3791 loop, c contains the first byte of the character. */
3792
3793 if (c != 0) do
3794 {
3795 const pcre_uchar *oldptr;
3796
3797 #ifdef SUPPORT_UTF
3798 if (utf && HAS_EXTRALEN(c))
3799 { /* Braces are required because the */
3800 GETCHARLEN(c, ptr, ptr); /* macro generates multiple statements */
3801 }
3802 #endif
3803
3804 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
3805 /* In the pre-compile phase, accumulate the length of any extra
3806 data and reset the pointer. This is so that very large classes that
3807 contain a zillion > 255 characters no longer overwrite the work space
3808 (which is on the stack). */
3809
3810 if (lengthptr != NULL)
3811 {
3812 *lengthptr += class_uchardata - class_uchardata_base;
3813 class_uchardata = class_uchardata_base;
3814 }
3815 #endif
3816
3817 /* Inside \Q...\E everything is literal except \E */
3818
3819 if (inescq)
3820 {
3821 if (c == CHAR_BACKSLASH && ptr[1] == CHAR_E) /* If we are at \E */
3822 {
3823 inescq = FALSE; /* Reset literal state */
3824 ptr++; /* Skip the 'E' */
3825 continue; /* Carry on with next */
3826 }
3827 goto CHECK_RANGE; /* Could be range if \E follows */
3828 }
3829
3830 /* Handle POSIX class names. Perl allows a negation extension of the
3831 form [:^name:]. A square bracket that doesn't match the syntax is
3832 treated as a literal. We also recognize the POSIX constructions
3833 [.ch.] and [=ch=] ("collating elements") and fault them, as Perl
3834 5.6 and 5.8 do. */
3835
3836 if (c == CHAR_LEFT_SQUARE_BRACKET &&
3837 (ptr[1] == CHAR_COLON || ptr[1] == CHAR_DOT ||
3838 ptr[1] == CHAR_EQUALS_SIGN) && check_posix_syntax(ptr, &tempptr))
3839 {
3840 BOOL local_negate = FALSE;
3841 int posix_class, taboffset, tabopt;
3842 register const pcre_uint8 *cbits = cd->cbits;
3843 pcre_uint8 pbits[32];
3844
3845 if (ptr[1] != CHAR_COLON)
3846 {
3847 *errorcodeptr = ERR31;
3848 goto FAILED;
3849 }
3850
3851 ptr += 2;
3852 if (*ptr == CHAR_CIRCUMFLEX_ACCENT)
3853 {
3854 local_negate = TRUE;
3855 should_flip_negation = TRUE; /* Note negative special */
3856 ptr++;
3857 }
3858
3859 posix_class = check_posix_name(ptr, (int)(tempptr - ptr));
3860 if (posix_class < 0)
3861 {
3862 *errorcodeptr = ERR30;
3863 goto FAILED;
3864 }
3865
3866 /* If matching is caseless, upper and lower are converted to
3867 alpha. This relies on the fact that the class table starts with
3868 alpha, lower, upper as the first 3 entries. */
3869
3870 if ((options & PCRE_CASELESS) != 0 && posix_class <= 2)
3871 posix_class = 0;
3872
3873 /* When PCRE_UCP is set, some of the POSIX classes are converted to
3874 different escape sequences that use Unicode properties. */
3875
3876 #ifdef SUPPORT_UCP
3877 if ((options & PCRE_UCP) != 0)
3878 {
3879 int pc = posix_class + ((local_negate)? POSIX_SUBSIZE/2 : 0);
3880 if (posix_substitutes[pc] != NULL)
3881 {
3882 nestptr = tempptr + 1;
3883 ptr = posix_substitutes[pc] - 1;
3884 continue;
3885 }
3886 }
3887 #endif
3888 /* In the non-UCP case, we build the bit map for the POSIX class in a
3889 chunk of local store because we may be adding and subtracting from it,
3890 and we don't want to subtract bits that may be in the main map already.
3891 At the end we or the result into the bit map that is being built. */
3892
3893 posix_class *= 3;
3894
3895 /* Copy in the first table (always present) */
3896
3897 memcpy(pbits, cbits + posix_class_maps[posix_class],
3898 32 * sizeof(pcre_uint8));
3899
3900 /* If there is a second table, add or remove it as required. */
3901
3902 taboffset = posix_class_maps[posix_class + 1];
3903 tabopt = posix_class_maps[posix_class + 2];
3904
3905 if (taboffset >= 0)
3906 {
3907 if (tabopt >= 0)
3908 for (c = 0; c < 32; c++) pbits[c] |= cbits[c + taboffset];
3909 else
3910 for (c = 0; c < 32; c++) pbits[c] &= ~cbits[c + taboffset];
3911 }
3912
3913 /* Not see if we need to remove any special characters. An option
3914 value of 1 removes vertical space and 2 removes underscore. */
3915
3916 if (tabopt < 0) tabopt = -tabopt;
3917 if (tabopt == 1) pbits[1] &= ~0x3c;
3918 else if (tabopt == 2) pbits[11] &= 0x7f;
3919
3920 /* Add the POSIX table or its complement into the main table that is
3921 being built and we are done. */
3922
3923 if (local_negate)
3924 for (c = 0; c < 32; c++) classbits[c] |= ~pbits[c];
3925 else
3926 for (c = 0; c < 32; c++) classbits[c] |= pbits[c];
3927
3928 ptr = tempptr + 1;
3929 /* Every class contains at least one < 256 characters. */
3930 class_has_8bitchar = 1;
3931 /* Every class contains at least two characters. */
3932 class_single_char = 2;
3933 continue; /* End of POSIX syntax handling */
3934 }
3935
3936 /* Backslash may introduce a single character, or it may introduce one
3937 of the specials, which just set a flag. The sequence \b is a special
3938 case. Inside a class (and only there) it is treated as backspace. We
3939 assume that other escapes have more than one character in them, so
3940 speculatively set both class_has_8bitchar and class_single_char bigger
3941 than one. Unrecognized escapes fall through and are either treated
3942 as literal characters (by default), or are faulted if
3943 PCRE_EXTRA is set. */
3944
3945 if (c == CHAR_BACKSLASH)
3946 {
3947 c = check_escape(&ptr, errorcodeptr, cd->bracount, options, TRUE);
3948 if (*errorcodeptr != 0) goto FAILED;
3949
3950 if (-c == ESC_b) c = CHAR_BS; /* \b is backspace in a class */
3951 else if (-c == ESC_N) /* \N is not supported in a class */
3952 {
3953 *errorcodeptr = ERR71;
3954 goto FAILED;
3955 }
3956 else if (-c == ESC_Q) /* Handle start of quoted string */
3957 {
3958 if (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
3959 {
3960 ptr += 2; /* avoid empty string */
3961 }
3962 else inescq = TRUE;
3963 continue;
3964 }
3965 else if (-c == ESC_E) continue; /* Ignore orphan \E */
3966
3967 if (c < 0)
3968 {
3969 register const pcre_uint8 *cbits = cd->cbits;
3970 /* Every class contains at least two < 256 characters. */
3971 class_has_8bitchar++;
3972 /* Every class contains at least two characters. */
3973 class_single_char += 2;
3974
3975 switch (-c)
3976 {
3977 #ifdef SUPPORT_UCP
3978 case ESC_du: /* These are the values given for \d etc */
3979 case ESC_DU: /* when PCRE_UCP is set. We replace the */
3980 case ESC_wu: /* escape sequence with an appropriate \p */
3981 case ESC_WU: /* or \P to test Unicode properties instead */
3982 case ESC_su: /* of the default ASCII testing. */
3983 case ESC_SU:
3984 nestptr = ptr;
3985 ptr = substitutes[-c - ESC_DU] - 1; /* Just before substitute */
3986 class_has_8bitchar--; /* Undo! */
3987 continue;
3988 #endif
3989 case ESC_d:
3990 for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_digit];
3991 continue;
3992
3993 case ESC_D:
3994 should_flip_negation = TRUE;
3995 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_digit];
3996 continue;
3997
3998 case ESC_w:
3999 for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_word];
4000 continue;
4001
4002 case ESC_W:
4003 should_flip_negation = TRUE;
4004 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_word];
4005 continue;
4006
4007 /* Perl 5.004 onwards omits VT from \s, but we must preserve it
4008 if it was previously set by something earlier in the character
4009 class. */
4010
4011 case ESC_s:
4012 classbits[0] |= cbits[cbit_space];
4013 classbits[1] |= cbits[cbit_space+1] & ~0x08;
4014 for (c = 2; c < 32; c++) classbits[c] |= cbits[c+cbit_space];
4015 continue;
4016
4017 case ESC_S:
4018 should_flip_negation = TRUE;
4019 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_space];
4020 classbits[1] |= 0x08; /* Perl 5.004 onwards omits VT from \s */
4021 continue;
4022
4023 case ESC_h:
4024 SETBIT(classbits, 0x09); /* VT */
4025 SETBIT(classbits, 0x20); /* SPACE */
4026 SETBIT(classbits, 0xa0); /* NSBP */
4027 #ifdef SUPPORT_UTF
4028 if (utf)
4029 {
4030 xclass = TRUE;
4031 *class_uchardata++ = XCL_SINGLE;
4032 class_uchardata += PRIV(ord2utf)(0x1680, class_uchardata);
4033 *class_uchardata++ = XCL_SINGLE;
4034 class_uchardata += PRIV(ord2utf)(0x180e, class_uchardata);
4035 *class_uchardata++ = XCL_RANGE;
4036 class_uchardata += PRIV(ord2utf)(0x2000, class_uchardata);
4037 class_uchardata += PRIV(ord2utf)(0x200A, class_uchardata);
4038 *class_uchardata++ = XCL_SINGLE;
4039 class_uchardata += PRIV(ord2utf)(0x202f, class_uchardata);
4040 *class_uchardata++ = XCL_SINGLE;
4041 class_uchardata += PRIV(ord2utf)(0x205f, class_uchardata);
4042 *class_uchardata++ = XCL_SINGLE;
4043 class_uchardata += PRIV(ord2utf)(0x3000, class_uchardata);
4044 }
4045 #endif
4046 continue;
4047
4048 case ESC_H:
4049 for (c = 0; c < 32; c++)
4050 {
4051 int x = 0xff;
4052 switch (c)
4053 {
4054 case 0x09/8: x ^= 1 << (0x09%8); break;
4055 case 0x20/8: x ^= 1 << (0x20%8); break;
4056 case 0xa0/8: x ^= 1 << (0xa0%8); break;
4057 default: break;
4058 }
4059 classbits[c] |= x;
4060 }
4061
4062 #ifdef SUPPORT_UTF
4063 if (utf)
4064 {
4065 xclass = TRUE;
4066 *class_uchardata++ = XCL_RANGE;
4067 class_uchardata += PRIV(ord2utf)(0x0100, class_uchardata);
4068 class_uchardata += PRIV(ord2utf)(0x167f, class_uchardata);
4069 *class_uchardata++ = XCL_RANGE;
4070 class_uchardata += PRIV(ord2utf)(0x1681, class_uchardata);
4071 class_uchardata += PRIV(ord2utf)(0x180d, class_uchardata);
4072 *class_uchardata++ = XCL_RANGE;
4073 class_uchardata += PRIV(ord2utf)(0x180f, class_uchardata);
4074 class_uchardata += PRIV(ord2utf)(0x1fff, class_uchardata);
4075 *class_uchardata++ = XCL_RANGE;
4076 class_uchardata += PRIV(ord2utf)(0x200B, class_uchardata);
4077 class_uchardata += PRIV(ord2utf)(0x202e, class_uchardata);
4078 *class_uchardata++ = XCL_RANGE;
4079 class_uchardata += PRIV(ord2utf)(0x2030, class_uchardata);
4080 class_uchardata += PRIV(ord2utf)(0x205e, class_uchardata);
4081 *class_uchardata++ = XCL_RANGE;
4082 class_uchardata += PRIV(ord2utf)(0x2060, class_uchardata);
4083 class_uchardata += PRIV(ord2utf)(0x2fff, class_uchardata);
4084 *class_uchardata++ = XCL_RANGE;
4085 class_uchardata += PRIV(ord2utf)(0x3001, class_uchardata);
4086 class_uchardata += PRIV(ord2utf)(0x10ffff, class_uchardata);
4087 }
4088 #endif
4089 continue;
4090
4091 case ESC_v:
4092 SETBIT(classbits, 0x0a); /* LF */
4093 SETBIT(classbits, 0x0b); /* VT */
4094 SETBIT(classbits, 0x0c); /* FF */
4095 SETBIT(classbits, 0x0d); /* CR */
4096 SETBIT(classbits, 0x85); /* NEL */
4097 #ifdef SUPPORT_UTF
4098 if (utf)
4099 {
4100 xclass = TRUE;
4101 *class_uchardata++ = XCL_RANGE;
4102 class_uchardata += PRIV(ord2utf)(0x2028, class_uchardata);
4103 class_uchardata += PRIV(ord2utf)(0x2029, class_uchardata);
4104 }
4105 #endif
4106 continue;
4107
4108 case ESC_V:
4109 for (c = 0; c < 32; c++)
4110 {
4111 int x = 0xff;
4112 switch (c)
4113 {
4114 case 0x0a/8: x ^= 1 << (0x0a%8);
4115 x ^= 1 << (0x0b%8);
4116 x ^= 1 << (0x0c%8);
4117 x ^= 1 << (0x0d%8);
4118 break;
4119 case 0x85/8: x ^= 1 << (0x85%8); break;
4120 default: break;
4121 }
4122 classbits[c] |= x;
4123 }
4124
4125 #ifdef SUPPORT_UTF
4126 if (utf)
4127 {
4128 xclass = TRUE;
4129 *class_uchardata++ = XCL_RANGE;
4130 class_uchardata += PRIV(ord2utf)(0x0100, class_uchardata);
4131 class_uchardata += PRIV(ord2utf)(0x2027, class_uchardata);
4132 *class_uchardata++ = XCL_RANGE;
4133 class_uchardata += PRIV(ord2utf)(0x2029, class_uchardata);
4134 class_uchardata += PRIV(ord2utf)(0x10ffff, class_uchardata);
4135 }
4136 #endif
4137 continue;
4138
4139 #ifdef SUPPORT_UCP
4140 case ESC_p:
4141 case ESC_P:
4142 {
4143 BOOL negated;
4144 int pdata;
4145 int ptype = get_ucp(&ptr, &negated, &pdata, errorcodeptr);
4146 if (ptype < 0) goto FAILED;
4147 xclass = TRUE;
4148 *class_uchardata++ = ((-c == ESC_p) != negated)?
4149 XCL_PROP : XCL_NOTPROP;
4150 *class_uchardata++ = ptype;
4151 *class_uchardata++ = pdata;
4152 class_has_8bitchar--; /* Undo! */
4153 continue;
4154 }
4155 #endif
4156 /* Unrecognized escapes are faulted if PCRE is running in its
4157 strict mode. By default, for compatibility with Perl, they are
4158 treated as literals. */
4159
4160 default:
4161 if ((options & PCRE_EXTRA) != 0)
4162 {
4163 *errorcodeptr = ERR7;
4164 goto FAILED;
4165 }
4166 class_has_8bitchar--; /* Undo the speculative increase. */
4167 class_single_char -= 2; /* Undo the speculative increase. */
4168 c = *ptr; /* Get the final character and fall through */
4169 break;
4170 }
4171 }
4172
4173 /* Fall through if we have a single character (c >= 0). This may be
4174 greater than 256. */
4175
4176 } /* End of backslash handling */
4177
4178 /* A single character may be followed by '-' to form a range. However,
4179 Perl does not permit ']' to be the end of the range. A '-' character
4180 at the end is treated as a literal. Perl ignores orphaned \E sequences
4181 entirely. The code for handling \Q and \E is messy. */
4182
4183 CHECK_RANGE:
4184 while (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
4185 {
4186 inescq = FALSE;
4187 ptr += 2;
4188 }
4189
4190 oldptr = ptr;
4191
4192 /* Remember \r or \n */
4193
4194 if (c == CHAR_CR || c == CHAR_NL) cd->external_flags |= PCRE_HASCRORLF;
4195
4196 /* Check for range */
4197
4198 if (!inescq && ptr[1] == CHAR_MINUS)
4199 {
4200 int d;
4201 ptr += 2;
4202 while (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_E) ptr += 2;
4203
4204 /* If we hit \Q (not followed by \E) at this point, go into escaped
4205 mode. */
4206
4207 while (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_Q)
4208 {
4209 ptr += 2;
4210 if (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_E)
4211 { ptr += 2; continue; }
4212 inescq = TRUE;
4213 break;
4214 }
4215
4216 if (*ptr == 0 || (!inescq && *ptr == CHAR_RIGHT_SQUARE_BRACKET))
4217 {
4218 ptr = oldptr;
4219 goto LONE_SINGLE_CHARACTER;
4220 }
4221
4222 #ifdef SUPPORT_UTF
4223 if (utf)
4224 { /* Braces are required because the */
4225 GETCHARLEN(d, ptr, ptr); /* macro generates multiple statements */
4226 }
4227 else
4228 #endif
4229 d = *ptr; /* Not UTF-8 mode */
4230
4231 /* The second part of a range can be a single-character escape, but
4232 not any of the other escapes. Perl 5.6 treats a hyphen as a literal
4233 in such circumstances. */
4234
4235 if (!inescq && d == CHAR_BACKSLASH)
4236 {
4237 d = check_escape(&ptr, errorcodeptr, cd->bracount, options, TRUE);
4238 if (*errorcodeptr != 0) goto FAILED;
4239
4240 /* \b is backspace; any other special means the '-' was literal */
4241
4242 if (d < 0)
4243 {
4244 if (d == -ESC_b) d = CHAR_BS; else
4245 {
4246 ptr = oldptr;
4247 goto LONE_SINGLE_CHARACTER; /* A few lines below */
4248 }
4249 }
4250 }
4251
4252 /* Check that the two values are in the correct order. Optimize
4253 one-character ranges */
4254
4255 if (d < c)
4256 {
4257 *errorcodeptr = ERR8;
4258 goto FAILED;
4259 }
4260
4261 if (d == c) goto LONE_SINGLE_CHARACTER; /* A few lines below */
4262
4263 /* Remember \r or \n */
4264
4265 if (d == CHAR_CR || d == CHAR_NL) cd->external_flags |= PCRE_HASCRORLF;
4266
4267 /* Since we found a character range, single character optimizations
4268 cannot be done anymore. */
4269 class_single_char = 2;
4270
4271 /* In UTF-8 mode, if the upper limit is > 255, or > 127 for caseless
4272 matching, we have to use an XCLASS with extra data items. Caseless
4273 matching for characters > 127 is available only if UCP support is
4274 available. */
4275
4276 #if defined SUPPORT_UTF && !(defined COMPILE_PCRE8)
4277 if ((d > 255) || (utf && ((options & PCRE_CASELESS) != 0 && d > 127)))
4278 #elif defined SUPPORT_UTF
4279 if (utf && (d > 255 || ((options & PCRE_CASELESS) != 0 && d > 127)))
4280 #elif !(defined COMPILE_PCRE8)
4281 if (d > 255)
4282 #endif
4283 #if defined SUPPORT_UTF || !(defined COMPILE_PCRE8)
4284 {
4285 xclass = TRUE;
4286
4287 /* With UCP support, we can find the other case equivalents of
4288 the relevant characters. There may be several ranges. Optimize how
4289 they fit with the basic range. */
4290
4291 #ifdef SUPPORT_UCP
4292 #ifndef COMPILE_PCRE8
4293 if (utf && (options & PCRE_CASELESS) != 0)
4294 #else
4295 if ((options & PCRE_CASELESS) != 0)
4296 #endif
4297 {
4298 unsigned int occ, ocd;
4299 unsigned int cc = c;
4300 unsigned int origd = d;
4301 while (get_othercase_range(&cc, origd, &occ, &ocd))
4302 {
4303 if (occ >= (unsigned int)c &&
4304 ocd <= (unsigned int)d)
4305 continue; /* Skip embedded ranges */
4306
4307 if (occ < (unsigned int)c &&
4308 ocd >= (unsigned int)c - 1) /* Extend the basic range */
4309 { /* if there is overlap, */
4310 c = occ; /* noting that if occ < c */
4311 continue; /* we can't have ocd > d */
4312 } /* because a subrange is */
4313 if (ocd > (unsigned int)d &&
4314 occ <= (unsigned int)d + 1) /* always shorter than */
4315 { /* the basic range. */
4316 d = ocd;
4317 continue;
4318 }
4319
4320 if (occ == ocd)
4321 {
4322 *class_uchardata++ = XCL_SINGLE;
4323 }
4324 else
4325 {
4326 *class_uchardata++ = XCL_RANGE;
4327 class_uchardata += PRIV(ord2utf)(occ, class_uchardata);
4328 }
4329 class_uchardata += PRIV(ord2utf)(ocd, class_uchardata);
4330 }
4331 }
4332 #endif /* SUPPORT_UCP */
4333
4334 /* Now record the original range, possibly modified for UCP caseless
4335 overlapping ranges. */
4336
4337 *class_uchardata++ = XCL_RANGE;
4338 #ifdef SUPPORT_UTF
4339 #ifndef COMPILE_PCRE8
4340 if (utf)
4341 {
4342 class_uchardata += PRIV(ord2utf)(c, class_uchardata);
4343 class_uchardata += PRIV(ord2utf)(d, class_uchardata);
4344 }
4345 else
4346 {
4347 *class_uchardata++ = c;
4348 *class_uchardata++ = d;
4349 }
4350 #else
4351 class_uchardata += PRIV(ord2utf)(c, class_uchardata);
4352 class_uchardata += PRIV(ord2utf)(d, class_uchardata);
4353 #endif
4354 #else /* SUPPORT_UTF */
4355 *class_uchardata++ = c;
4356 *class_uchardata++ = d;
4357 #endif /* SUPPORT_UTF */
4358
4359 /* With UCP support, we are done. Without UCP support, there is no
4360 caseless matching for UTF characters > 127; we can use the bit map
4361 for the smaller ones. As for 16 bit characters without UTF, we
4362 can still use */
4363
4364 #ifdef SUPPORT_UCP
4365 #ifndef COMPILE_PCRE8
4366 if (utf)
4367 #endif
4368 continue; /* With next character in the class */
4369 #endif /* SUPPORT_UCP */
4370
4371 #if defined SUPPORT_UTF && !defined(SUPPORT_UCP) && !(defined COMPILE_PCRE8)
4372 if (utf)
4373 {
4374 if ((options & PCRE_CASELESS) == 0 || c > 127) continue;
4375 /* Adjust upper limit and fall through to set up the map */
4376 d = 127;
4377 }
4378 else
4379 {
4380 if (c > 255) continue;
4381 /* Adjust upper limit and fall through to set up the map */
4382 d = 255;
4383 }
4384 #elif defined SUPPORT_UTF && !defined(SUPPORT_UCP)
4385 if ((options & PCRE_CASELESS) == 0 || c > 127) continue;
4386 /* Adjust upper limit and fall through to set up the map */
4387 d = 127;
4388 #else
4389 if (c > 255) continue;
4390 /* Adjust upper limit and fall through to set up the map */
4391 d = 255;
4392 #endif /* SUPPORT_UTF && !SUPPORT_UCP && !COMPILE_PCRE8 */
4393 }
4394 #endif /* SUPPORT_UTF || !COMPILE_PCRE8 */
4395
4396 /* We use the bit map for 8 bit mode, or when the characters fall
4397 partially or entirely to [0-255] ([0-127] for UCP) ranges. */
4398
4399 class_has_8bitchar = 1;
4400
4401 /* We can save a bit of time by skipping this in the pre-compile. */
4402
4403 if (lengthptr == NULL) for (; c <= d; c++)
4404 {
4405 classbits[c/8] |= (1 << (c&7));
4406 if ((options & PCRE_CASELESS) != 0)
4407 {
4408 int uc = cd->fcc[c]; /* flip case */
4409 classbits[uc/8] |= (1 << (uc&7));
4410 }
4411 }
4412
4413 continue; /* Go get the next char in the class */
4414 }
4415
4416 /* Handle a lone single character - we can get here for a normal
4417 non-escape char, or after \ that introduces a single character or for an
4418 apparent range that isn't. */
4419
4420 LONE_SINGLE_CHARACTER:
4421
4422 /* Only the value of 1 matters for class_single_char. */
4423 if (class_single_char < 2) class_single_char++;
4424
4425 /* If class_charcount is 1, we saw precisely one character. As long as
4426 there were no negated characters >= 128 and there was no use of \p or \P,
4427 in other words, no use of any XCLASS features, we can optimize.
4428
4429 In UTF-8 mode, we can optimize the negative case only if there were no
4430 characters >= 128 because OP_NOT and the related opcodes like OP_NOTSTAR
4431 operate on single-bytes characters only. This is an historical hangover.
4432 Maybe one day we can tidy these opcodes to handle multi-byte characters.
4433
4434 The optimization throws away the bit map. We turn the item into a
4435 1-character OP_CHAR[I] if it's positive, or OP_NOT[I] if it's negative.
4436 Note that OP_NOT[I] does not support multibyte characters. In the positive
4437 case, it can cause firstchar to be set. Otherwise, there can be no first
4438 char if this item is first, whatever repeat count may follow. In the case
4439 of reqchar, save the previous value for reinstating. */
4440
4441 #ifdef SUPPORT_UTF
4442 if (class_single_char == 1 && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET
4443 && (!utf || !negate_class || c < (MAX_VALUE_FOR_SINGLE_CHAR + 1)))
4444 #else
4445 if (class_single_char == 1 && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET)
4446 #endif
4447 {
4448 ptr++;
4449 zeroreqchar = reqchar;
4450
4451 /* The OP_NOT[I] opcodes work on single characters only. */
4452
4453 if (negate_class)
4454 {
4455 if (firstchar == REQ_UNSET) firstchar = REQ_NONE;
4456 zerofirstchar = firstchar;
4457 *code++ = ((options & PCRE_CASELESS) != 0)? OP_NOTI: OP_NOT;
4458 *code++ = c;
4459 goto NOT_CHAR;
4460 }
4461
4462 /* For a single, positive character, get the value into mcbuffer, and
4463 then we can handle this with the normal one-character code. */
4464
4465 #ifdef SUPPORT_UTF
4466 if (utf && c > MAX_VALUE_FOR_SINGLE_CHAR)
4467 mclength = PRIV(ord2utf)(c, mcbuffer);
4468 else
4469 #endif
4470 {
4471 mcbuffer[0] = c;
4472 mclength = 1;
4473 }
4474 goto ONE_CHAR;
4475 } /* End of 1-char optimization */
4476
4477 /* Handle a character that cannot go in the bit map. */
4478
4479 #if defined SUPPORT_UTF && !(defined COMPILE_PCRE8)
4480 if ((c > 255) || (utf && ((options & PCRE_CASELESS) != 0 && c > 127)))
4481 #elif defined SUPPORT_UTF
4482 if (utf && (c > 255 || ((options & PCRE_CASELESS) != 0 && c > 127)))
4483 #elif !(defined COMPILE_PCRE8)
4484 if (c > 255)
4485 #endif
4486
4487 #if defined SUPPORT_UTF || !(defined COMPILE_PCRE8)
4488 {
4489 xclass = TRUE;
4490 *class_uchardata++ = XCL_SINGLE;
4491 #ifdef SUPPORT_UTF
4492 #ifndef COMPILE_PCRE8
4493 /* In non 8 bit mode, we can get here even if we are not in UTF mode. */
4494 if (!utf)
4495 *class_uchardata++ = c;
4496 else
4497 #endif
4498 class_uchardata += PRIV(ord2utf)(c, class_uchardata);
4499 #else /* SUPPORT_UTF */
4500 *class_uchardata++ = c;
4501 #endif /* SUPPORT_UTF */
4502
4503 #ifdef SUPPORT_UCP
4504 #ifdef COMPILE_PCRE8
4505 if ((options & PCRE_CASELESS) != 0)
4506 #else
4507 /* In non 8 bit mode, we can get here even if we are not in UTF mode. */
4508 if (utf && (options & PCRE_CASELESS) != 0)
4509 #endif
4510 {
4511 unsigned int othercase;
4512 if ((othercase = UCD_OTHERCASE(c)) != c)
4513 {
4514 *class_uchardata++ = XCL_SINGLE;
4515 class_uchardata += PRIV(ord2utf)(othercase, class_uchardata);
4516 }
4517 }
4518 #endif /* SUPPORT_UCP */
4519
4520 }
4521 else
4522 #endif /* SUPPORT_UTF || COMPILE_PCRE16 */
4523
4524 /* Handle a single-byte character */
4525 {
4526 class_has_8bitchar = 1;
4527 classbits[c/8] |= (1 << (c&7));
4528 if ((options & PCRE_CASELESS) != 0)
4529 {
4530 c = cd->fcc[c]; /* flip case */
4531 classbits[c/8] |= (1 << (c&7));
4532 }
4533 }
4534 }
4535
4536 /* Loop until ']' reached. This "while" is the end of the "do" far above.
4537 If we are at the end of an internal nested string, revert to the outer
4538 string. */
4539
4540 while (((c = *(++ptr)) != 0 ||
4541 (nestptr != NULL &&
4542 (ptr = nestptr, nestptr = NULL, c = *(++ptr)) != 0)) &&
4543 (c != CHAR_RIGHT_SQUARE_BRACKET || inescq));
4544
4545 /* Check for missing terminating ']' */
4546
4547 if (c == 0)
4548 {
4549 *errorcodeptr = ERR6;
4550 goto FAILED;
4551 }
4552
4553 /* If this is the first thing in the branch, there can be no first char
4554 setting, whatever the repeat count. Any reqchar setting must remain
4555 unchanged after any kind of repeat. */
4556
4557 if (firstchar == REQ_UNSET) firstchar = REQ_NONE;
4558 zerofirstchar = firstchar;
4559 zeroreqchar = reqchar;
4560
4561 /* If there are characters with values > 255, we have to compile an
4562 extended class, with its own opcode, unless there was a negated special
4563 such as \S in the class, and PCRE_UCP is not set, because in that case all
4564 characters > 255 are in the class, so any that were explicitly given as
4565 well can be ignored. If (when there are explicit characters > 255 that must
4566 be listed) there are no characters < 256, we can omit the bitmap in the
4567 actual compiled code. */
4568
4569 #ifdef SUPPORT_UTF
4570 if (xclass && (!should_flip_negation || (options & PCRE_UCP) != 0))
4571 #elif !defined COMPILE_PCRE8
4572 if (xclass && !should_flip_negation)
4573 #endif
4574 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
4575 {
4576 *class_uchardata++ = XCL_END; /* Marks the end of extra data */
4577 *code++ = OP_XCLASS;
4578 code += LINK_SIZE;
4579 *code = negate_class? XCL_NOT:0;
4580
4581 /* If the map is required, move up the extra data to make room for it;
4582 otherwise just move the code pointer to the end of the extra data. */
4583
4584 if (class_has_8bitchar > 0)
4585 {
4586 *code++ |= XCL_MAP;
4587 memmove(code + (32 / sizeof(pcre_uchar)), code,
4588 IN_UCHARS(class_uchardata - code));
4589 memcpy(code, classbits, 32);
4590 code = class_uchardata + (32 / sizeof(pcre_uchar));
4591 }
4592 else code = class_uchardata;
4593
4594 /* Now fill in the complete length of the item */
4595
4596 PUT(previous, 1, (int)(code - previous));
4597 break; /* End of class handling */
4598 }
4599 #endif
4600
4601 /* If there are no characters > 255, or they are all to be included or
4602 excluded, set the opcode to OP_CLASS or OP_NCLASS, depending on whether the
4603 whole class was negated and whether there were negative specials such as \S
4604 (non-UCP) in the class. Then copy the 32-byte map into the code vector,
4605 negating it if necessary. */
4606
4607 *code++ = (negate_class == should_flip_negation) ? OP_CLASS : OP_NCLASS;
4608 if (lengthptr == NULL) /* Save time in the pre-compile phase */
4609 {
4610 if (negate_class)
4611 for (c = 0; c < 32; c++) classbits[c] = ~classbits[c];
4612 memcpy(code, classbits, 32);
4613 }
4614 code += 32 / sizeof(pcre_uchar);
4615 NOT_CHAR:
4616 break;
4617
4618
4619 /* ===================================================================*/
4620 /* Various kinds of repeat; '{' is not necessarily a quantifier, but this
4621 has been tested above. */
4622
4623 case CHAR_LEFT_CURLY_BRACKET:
4624 if (!is_quantifier) goto NORMAL_CHAR;
4625 ptr = read_repeat_counts(ptr+1, &repeat_min, &repeat_max, errorcodeptr);
4626 if (*errorcodeptr != 0) goto FAILED;
4627 goto REPEAT;
4628
4629 case CHAR_ASTERISK:
4630 repeat_min = 0;
4631 repeat_max = -1;
4632 goto REPEAT;
4633
4634 case CHAR_PLUS:
4635 repeat_min = 1;
4636 repeat_max = -1;
4637 goto REPEAT;
4638
4639 case CHAR_QUESTION_MARK:
4640 repeat_min = 0;
4641 repeat_max = 1;
4642
4643 REPEAT:
4644 if (previous == NULL)
4645 {
4646 *errorcodeptr = ERR9;
4647 goto FAILED;
4648 }
4649
4650 if (repeat_min == 0)
4651 {
4652 firstchar = zerofirstchar; /* Adjust for zero repeat */
4653 reqchar = zeroreqchar; /* Ditto */
4654 }
4655
4656 /* Remember whether this is a variable length repeat */
4657
4658 reqvary = (repeat_min == repeat_max)? 0 : REQ_VARY;
4659
4660 op_type = 0; /* Default single-char op codes */
4661 possessive_quantifier = FALSE; /* Default not possessive quantifier */
4662
4663 /* Save start of previous item, in case we have to move it up in order to
4664 insert something before it. */
4665
4666 tempcode = previous;
4667
4668 /* If the next character is '+', we have a possessive quantifier. This
4669 implies greediness, whatever the setting of the PCRE_UNGREEDY option.
4670 If the next character is '?' this is a minimizing repeat, by default,
4671 but if PCRE_UNGREEDY is set, it works the other way round. We change the
4672 repeat type to the non-default. */
4673
4674 if (ptr[1] == CHAR_PLUS)
4675 {
4676 repeat_type = 0; /* Force greedy */
4677 possessive_quantifier = TRUE;
4678 ptr++;
4679 }
4680 else if (ptr[1] == CHAR_QUESTION_MARK)
4681 {
4682 repeat_type = greedy_non_default;
4683 ptr++;
4684 }
4685 else repeat_type = greedy_default;
4686
4687 /* If previous was a recursion call, wrap it in atomic brackets so that
4688 previous becomes the atomic group. All recursions were so wrapped in the
4689 past, but it no longer happens for non-repeated recursions. In fact, the
4690 repeated ones could be re-implemented independently so as not to need this,
4691 but for the moment we rely on the code for repeating groups. */
4692
4693 if (*previous == OP_RECURSE)
4694 {
4695 memmove(previous + 1 + LINK_SIZE, previous, IN_UCHARS(1 + LINK_SIZE));
4696 *previous = OP_ONCE;
4697 PUT(previous, 1, 2 + 2*LINK_SIZE);
4698 previous[2 + 2*LINK_SIZE] = OP_KET;
4699 PUT(previous, 3 + 2*LINK_SIZE, 2 + 2*LINK_SIZE);
4700 code += 2 + 2 * LINK_SIZE;
4701 length_prevgroup = 3 + 3*LINK_SIZE;
4702
4703 /* When actually compiling, we need to check whether this was a forward
4704 reference, and if so, adjust the offset. */
4705
4706 if (lengthptr == NULL && cd->hwm >= cd->start_workspace + LINK_SIZE)
4707 {
4708 int offset = GET(cd->hwm, -LINK_SIZE);
4709 if (offset == previous + 1 - cd->start_code)
4710 PUT(cd->hwm, -LINK_SIZE, offset + 1 + LINK_SIZE);
4711 }
4712 }
4713
4714 /* Now handle repetition for the different types of item. */
4715
4716 /* If previous was a character match, abolish the item and generate a
4717 repeat item instead. If a char item has a minumum of more than one, ensure
4718 that it is set in reqchar - it might not be if a sequence such as x{3} is
4719 the first thing in a branch because the x will have gone into firstchar
4720 instead. */
4721
4722 if (*previous == OP_CHAR || *previous == OP_CHARI)
4723 {
4724 op_type = (*previous == OP_CHAR)? 0 : OP_STARI - OP_STAR;
4725
4726 /* Deal with UTF characters that take up more than one character. It's
4727 easier to write this out separately than try to macrify it. Use c to
4728 hold the length of the character in bytes, plus UTF_LENGTH to flag that
4729 it's a length rather than a small character. */
4730
4731 #ifdef SUPPORT_UTF
4732 if (utf && NOT_FIRSTCHAR(code[-1]))
4733 {
4734 pcre_uchar *lastchar = code - 1;
4735 BACKCHAR(lastchar);
4736 c = (int)(code - lastchar); /* Length of UTF-8 character */
4737 memcpy(utf_chars, lastchar, IN_UCHARS(c)); /* Save the char */
4738 c |= UTF_LENGTH; /* Flag c as a length */
4739 }
4740 else
4741 #endif /* SUPPORT_UTF */
4742
4743 /* Handle the case of a single charater - either with no UTF support, or
4744 with UTF disabled, or for a single character UTF character. */
4745 {
4746 c = code[-1];
4747 if (repeat_min > 1) reqchar = c | req_caseopt | cd->req_varyopt;
4748 }
4749
4750 /* If the repetition is unlimited, it pays to see if the next thing on
4751 the line is something that cannot possibly match this character. If so,
4752 automatically possessifying this item gains some performance in the case
4753 where the match fails. */
4754
4755 if (!possessive_quantifier &&
4756 repeat_max < 0 &&
4757 check_auto_possessive(previous, utf, ptr + 1, options, cd))
4758 {
4759 repeat_type = 0; /* Force greedy */
4760 possessive_quantifier = TRUE;
4761 }
4762
4763 goto OUTPUT_SINGLE_REPEAT; /* Code shared with single character types */
4764 }
4765
4766 /* If previous was a single negated character ([^a] or similar), we use
4767 one of the special opcodes, replacing it. The code is shared with single-
4768 character repeats by setting opt_type to add a suitable offset into
4769 repeat_type. We can also test for auto-possessification. OP_NOT and OP_NOTI
4770 are currently used only for single-byte chars. */
4771
4772 else if (*previous == OP_NOT || *previous == OP_NOTI)
4773 {
4774 op_type = ((*previous == OP_NOT)? OP_NOTSTAR : OP_NOTSTARI) - OP_STAR;
4775 c = previous[1];
4776 if (!possessive_quantifier &&
4777 repeat_max < 0 &&
4778 check_auto_possessive(previous, utf, ptr + 1, options, cd))
4779 {
4780 repeat_type = 0; /* Force greedy */
4781 possessive_quantifier = TRUE;
4782 }
4783 goto OUTPUT_SINGLE_REPEAT;
4784 }
4785
4786 /* If previous was a character type match (\d or similar), abolish it and
4787 create a suitable repeat item. The code is shared with single-character
4788 repeats by setting op_type to add a suitable offset into repeat_type. Note
4789 the the Unicode property types will be present only when SUPPORT_UCP is
4790 defined, but we don't wrap the little bits of code here because it just
4791 makes it horribly messy. */
4792
4793 else if (*previous < OP_EODN)
4794 {
4795 pcre_uchar *oldcode;
4796 int prop_type, prop_value;
4797 op_type = OP_TYPESTAR - OP_STAR; /* Use type opcodes */
4798 c = *previous;
4799
4800 if (!possessive_quantifier &&
4801 repeat_max < 0 &&
4802 check_auto_possessive(previous, utf, ptr + 1, options, cd))
4803 {
4804 repeat_type = 0; /* Force greedy */
4805 possessive_quantifier = TRUE;
4806 }
4807
4808 OUTPUT_SINGLE_REPEAT:
4809 if (*previous == OP_PROP || *previous == OP_NOTPROP)
4810 {
4811 prop_type = previous[1];
4812 prop_value = previous[2];
4813 }
4814 else prop_type = prop_value = -1;
4815
4816 oldcode = code;
4817 code = previous; /* Usually overwrite previous item */
4818
4819 /* If the maximum is zero then the minimum must also be zero; Perl allows
4820 this case, so we do too - by simply omitting the item altogether. */
4821
4822 if (repeat_max == 0) goto END_REPEAT;
4823
4824 /*--------------------------------------------------------------------*/
4825 /* This code is obsolete from release 8.00; the restriction was finally
4826 removed: */
4827
4828 /* All real repeats make it impossible to handle partial matching (maybe
4829 one day we will be able to remove this restriction). */
4830
4831 /* if (repeat_max != 1) cd->external_flags |= PCRE_NOPARTIAL; */
4832 /*--------------------------------------------------------------------*/
4833
4834 /* Combine the op_type with the repeat_type */
4835
4836 repeat_type += op_type;
4837
4838 /* A minimum of zero is handled either as the special case * or ?, or as
4839 an UPTO, with the maximum given. */
4840
4841 if (repeat_min == 0)
4842 {
4843 if (repeat_max == -1) *code++ = OP_STAR + repeat_type;
4844 else if (repeat_max == 1) *code++ = OP_QUERY + repeat_type;
4845 else
4846 {
4847 *code++ = OP_UPTO + repeat_type;
4848 PUT2INC(code, 0, repeat_max);
4849 }
4850 }
4851
4852 /* A repeat minimum of 1 is optimized into some special cases. If the
4853 maximum is unlimited, we use OP_PLUS. Otherwise, the original item is
4854 left in place and, if the maximum is greater than 1, we use OP_UPTO with
4855 one less than the maximum. */
4856
4857 else if (repeat_min == 1)
4858 {
4859 if (repeat_max == -1)
4860 *code++ = OP_PLUS + repeat_type;
4861 else
4862 {
4863 code = oldcode; /* leave previous item in place */
4864 if (repeat_max == 1) goto END_REPEAT;
4865 *code++ = OP_UPTO + repeat_type;
4866 PUT2INC(code, 0, repeat_max - 1);
4867 }
4868 }
4869
4870 /* The case {n,n} is just an EXACT, while the general case {n,m} is
4871 handled as an EXACT followed by an UPTO. */
4872
4873 else
4874 {
4875 *code++ = OP_EXACT + op_type; /* NB EXACT doesn't have repeat_type */
4876 PUT2INC(code, 0, repeat_min);
4877
4878 /* If the maximum is unlimited, insert an OP_STAR. Before doing so,
4879 we have to insert the character for the previous code. For a repeated
4880 Unicode property match, there are two extra bytes that define the
4881 required property. In UTF-8 mode, long characters have their length in
4882 c, with the UTF_LENGTH bit as a flag. */
4883
4884 if (repeat_max < 0)
4885 {
4886 #ifdef SUPPORT_UTF
4887 if (utf && (c & UTF_LENGTH) != 0)
4888 {
4889 memcpy(code, utf_chars, IN_UCHARS(c & 7));
4890 code += c & 7;
4891 }
4892 else
4893 #endif
4894 {
4895 *code++ = c;
4896 if (prop_type >= 0)
4897 {
4898 *code++ = prop_type;
4899 *code++ = prop_value;
4900 }
4901 }
4902 *code++ = OP_STAR + repeat_type;
4903 }
4904
4905 /* Else insert an UPTO if the max is greater than the min, again
4906 preceded by the character, for the previously inserted code. If the
4907 UPTO is just for 1 instance, we can use QUERY instead. */
4908
4909 else if (repeat_max != repeat_min)
4910 {
4911 #ifdef SUPPORT_UTF
4912 if (utf && (c & UTF_LENGTH) != 0)
4913 {
4914 memcpy(code, utf_chars, IN_UCHARS(c & 7));
4915 code += c & 7;
4916 }
4917 else
4918 #endif
4919 *code++ = c;
4920 if (prop_type >= 0)
4921 {
4922 *code++ = prop_type;
4923 *code++ = prop_value;
4924 }
4925 repeat_max -= repeat_min;
4926
4927 if (repeat_max == 1)
4928 {
4929 *code++ = OP_QUERY + repeat_type;
4930 }
4931 else
4932 {
4933 *code++ = OP_UPTO + repeat_type;
4934 PUT2INC(code, 0, repeat_max);
4935 }
4936 }
4937 }
4938
4939 /* The character or character type itself comes last in all cases. */
4940
4941 #ifdef SUPPORT_UTF
4942 if (utf && (c & UTF_LENGTH) != 0)
4943 {
4944 memcpy(code, utf_chars, IN_UCHARS(c & 7));
4945 code += c & 7;
4946 }
4947 else
4948 #endif
4949 *code++ = c;
4950
4951 /* For a repeated Unicode property match, there are two extra bytes that
4952 define the required property. */
4953
4954 #ifdef SUPPORT_UCP
4955 if (prop_type >= 0)
4956 {
4957 *code++ = prop_type;
4958 *code++ = prop_value;
4959 }
4960 #endif
4961 }
4962
4963 /* If previous was a character class or a back reference, we put the repeat
4964 stuff after it, but just skip the item if the repeat was {0,0}. */
4965
4966 else if (*previous == OP_CLASS ||
4967 *previous == OP_NCLASS ||
4968 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
4969 *previous == OP_XCLASS ||
4970 #endif
4971 *previous == OP_REF ||
4972 *previous == OP_REFI)
4973 {
4974 if (repeat_max == 0)
4975 {
4976 code = previous;
4977 goto END_REPEAT;
4978 }
4979
4980 /*--------------------------------------------------------------------*/
4981 /* This code is obsolete from release 8.00; the restriction was finally
4982 removed: */
4983
4984 /* All real repeats make it impossible to handle partial matching (maybe
4985 one day we will be able to remove this restriction). */
4986
4987 /* if (repeat_max != 1) cd->external_flags |= PCRE_NOPARTIAL; */
4988 /*--------------------------------------------------------------------*/
4989
4990 if (repeat_min == 0 && repeat_max == -1)
4991 *code++ = OP_CRSTAR + repeat_type;
4992 else if (repeat_min == 1 && repeat_max == -1)
4993 *code++ = OP_CRPLUS + repeat_type;
4994 else if (repeat_min == 0 && repeat_max == 1)
4995 *code++ = OP_CRQUERY + repeat_type;
4996 else
4997 {
4998 *code++ = OP_CRRANGE + repeat_type;
4999 PUT2INC(code, 0, repeat_min);
5000 if (repeat_max == -1) repeat_max = 0; /* 2-byte encoding for max */
5001 PUT2INC(code, 0, repeat_max);
5002 }
5003 }
5004
5005 /* If previous was a bracket group, we may have to replicate it in certain
5006 cases. Note that at this point we can encounter only the "basic" bracket
5007 opcodes such as BRA and CBRA, as this is the place where they get converted
5008 into the more special varieties such as BRAPOS and SBRA. A test for >=
5009 OP_ASSERT and <= OP_COND includes ASSERT, ASSERT_NOT, ASSERTBACK,
5010 ASSERTBACK_NOT, ONCE, BRA, CBRA, and COND. Originally, PCRE did not allow
5011 repetition of assertions, but now it does, for Perl compatibility. */
5012
5013 else if (*previous >= OP_ASSERT && *previous <= OP_COND)
5014 {
5015 register int i;
5016 int len = (int)(code - previous);
5017 pcre_uchar *bralink = NULL;
5018 pcre_uchar *brazeroptr = NULL;
5019
5020 /* Repeating a DEFINE group is pointless, but Perl allows the syntax, so
5021 we just ignore the repeat. */
5022
5023 if (*previous == OP_COND && previous[LINK_SIZE+1] == OP_DEF)
5024 goto END_REPEAT;
5025
5026 /* There is no sense in actually repeating assertions. The only potential
5027 use of repetition is in cases when the assertion is optional. Therefore,
5028 if the minimum is greater than zero, just ignore the repeat. If the
5029 maximum is not not zero or one, set it to 1. */
5030
5031 if (*previous < OP_ONCE) /* Assertion */
5032 {
5033 if (repeat_min > 0) goto END_REPEAT;
5034 if (repeat_max < 0 || repeat_max > 1) repeat_max = 1;
5035 }
5036
5037 /* The case of a zero minimum is special because of the need to stick
5038 OP_BRAZERO in front of it, and because the group appears once in the
5039 data, whereas in other cases it appears the minimum number of times. For
5040 this reason, it is simplest to treat this case separately, as otherwise
5041 the code gets far too messy. There are several special subcases when the
5042 minimum is zero. */
5043
5044 if (repeat_min == 0)
5045 {
5046 /* If the maximum is also zero, we used to just omit the group from the
5047 output altogether, like this:
5048
5049 ** if (repeat_max == 0)
5050 ** {
5051 ** code = previous;
5052 ** goto END_REPEAT;
5053 ** }
5054
5055 However, that fails when a group or a subgroup within it is referenced
5056 as a subroutine from elsewhere in the pattern, so now we stick in
5057 OP_SKIPZERO in front of it so that it is skipped on execution. As we
5058 don't have a list of which groups are referenced, we cannot do this
5059 selectively.
5060
5061 If the maximum is 1 or unlimited, we just have to stick in the BRAZERO
5062 and do no more at this point. However, we do need to adjust any
5063 OP_RECURSE calls inside the group that refer to the group itself or any
5064 internal or forward referenced group, because the offset is from the
5065 start of the whole regex. Temporarily terminate the pattern while doing
5066 this. */
5067
5068 if (repeat_max <= 1) /* Covers 0, 1, and unlimited */
5069 {
5070 *code = OP_END;
5071 adjust_recurse(previous, 1, utf, cd, save_hwm);
5072 memmove(previous + 1, previous, IN_UCHARS(len));
5073 code++;
5074 if (repeat_max == 0)
5075 {
5076 *previous++ = OP_SKIPZERO;
5077 goto END_REPEAT;
5078 }
5079 brazeroptr = previous; /* Save for possessive optimizing */
5080 *previous++ = OP_BRAZERO + repeat_type;
5081 }
5082
5083 /* If the maximum is greater than 1 and limited, we have to replicate
5084 in a nested fashion, sticking OP_BRAZERO before each set of brackets.
5085 The first one has to be handled carefully because it's the original
5086 copy, which has to be moved up. The remainder can be handled by code
5087 that is common with the non-zero minimum case below. We have to
5088 adjust the value or repeat_max, since one less copy is required. Once
5089 again, we may have to adjust any OP_RECURSE calls inside the group. */
5090
5091 else
5092 {
5093 int offset;
5094 *code = OP_END;
5095 adjust_recurse(previous, 2 + LINK_SIZE, utf, cd, save_hwm);
5096 memmove(previous + 2 + LINK_SIZE, previous, IN_UCHARS(len));
5097 code += 2 + LINK_SIZE;
5098 *previous++ = OP_BRAZERO + repeat_type;
5099 *previous++ = OP_BRA;
5100
5101 /* We chain together the bracket offset fields that have to be
5102 filled in later when the ends of the brackets are reached. */
5103
5104 offset = (bralink == NULL)? 0 : (int)(previous - bralink);
5105 bralink = previous;
5106 PUTINC(previous, 0, offset);
5107 }
5108
5109 repeat_max--;
5110 }
5111
5112 /* If the minimum is greater than zero, replicate the group as many
5113 times as necessary, and adjust the maximum to the number of subsequent
5114 copies that we need. If we set a first char from the group, and didn't
5115 set a required char, copy the latter from the former. If there are any
5116 forward reference subroutine calls in the group, there will be entries on
5117 the workspace list; replicate these with an appropriate increment. */
5118
5119 else
5120 {
5121 if (repeat_min > 1)
5122 {
5123 /* In the pre-compile phase, we don't actually do the replication. We
5124 just adjust the length as if we had. Do some paranoid checks for
5125 potential integer overflow. The INT64_OR_DOUBLE type is a 64-bit
5126 integer type when available, otherwise double. */
5127
5128 if (lengthptr != NULL)
5129 {
5130 int delta = (repeat_min - 1)*length_prevgroup;
5131 if ((INT64_OR_DOUBLE)(repeat_min - 1)*
5132 (INT64_OR_DOUBLE)length_prevgroup >
5133 (INT64_OR_DOUBLE)INT_MAX ||
5134 OFLOW_MAX - *lengthptr < delta)
5135 {
5136 *errorcodeptr = ERR20;
5137 goto FAILED;
5138 }
5139 *lengthptr += delta;
5140 }
5141
5142 /* This is compiling for real. If there is a set first byte for
5143 the group, and we have not yet set a "required byte", set it. Make
5144 sure there is enough workspace for copying forward references before
5145 doing the copy. */
5146
5147 else
5148 {
5149 if (groupsetfirstchar && reqchar < 0) reqchar = firstchar;
5150
5151 for (i = 1; i < repeat_min; i++)
5152 {
5153 pcre_uchar *hc;
5154 pcre_uchar *this_hwm = cd->hwm;
5155 memcpy(code, previous, IN_UCHARS(len));
5156
5157 while (cd->hwm > cd->start_workspace + cd->workspace_size -
5158 WORK_SIZE_SAFETY_MARGIN - (this_hwm - save_hwm))
5159 {
5160 int save_offset = save_hwm - cd->start_workspace;
5161 int this_offset = this_hwm - cd->start_workspace;
5162 *errorcodeptr = expand_workspace(cd);
5163 if (*errorcodeptr != 0) goto FAILED;
5164 save_hwm = (pcre_uchar *)cd->start_workspace + save_offset;
5165 this_hwm = (pcre_uchar *)cd->start_workspace + this_offset;
5166 }
5167
5168 for (hc = save_hwm; hc < this_hwm; hc += LINK_SIZE)
5169 {
5170 PUT(cd->hwm, 0, GET(hc, 0) + len);
5171 cd->hwm += LINK_SIZE;
5172 }
5173 save_hwm = this_hwm;
5174 code += len;
5175 }
5176 }
5177 }
5178
5179 if (repeat_max > 0) repeat_max -= repeat_min;
5180 }
5181
5182 /* This code is common to both the zero and non-zero minimum cases. If
5183 the maximum is limited, it replicates the group in a nested fashion,
5184 remembering the bracket starts on a stack. In the case of a zero minimum,
5185 the first one was set up above. In all cases the repeat_max now specifies
5186 the number of additional copies needed. Again, we must remember to
5187 replicate entries on the forward reference list. */
5188
5189 if (repeat_max >= 0)
5190 {
5191 /* In the pre-compile phase, we don't actually do the replication. We
5192 just adjust the length as if we had. For each repetition we must add 1
5193 to the length for BRAZERO and for all but the last repetition we must
5194 add 2 + 2*LINKSIZE to allow for the nesting that occurs. Do some
5195 paranoid checks to avoid integer overflow. The INT64_OR_DOUBLE type is
5196 a 64-bit integer type when available, otherwise double. */
5197
5198 if (lengthptr != NULL && repeat_max > 0)
5199 {
5200 int delta = repeat_max * (length_prevgroup + 1 + 2 + 2*LINK_SIZE) -
5201 2 - 2*LINK_SIZE; /* Last one doesn't nest */
5202 if ((INT64_OR_DOUBLE)repeat_max *
5203 (INT64_OR_DOUBLE)(length_prevgroup + 1 + 2 + 2*LINK_SIZE)
5204 > (INT64_OR_DOUBLE)INT_MAX ||
5205 OFLOW_MAX - *lengthptr < delta)
5206 {
5207 *errorcodeptr = ERR20;
5208 goto FAILED;
5209 }
5210 *lengthptr += delta;
5211 }
5212
5213 /* This is compiling for real */
5214
5215 else for (i = repeat_max - 1; i >= 0; i--)
5216 {
5217 pcre_uchar *hc;
5218 pcre_uchar *this_hwm = cd->hwm;
5219
5220 *code++ = OP_BRAZERO + repeat_type;
5221
5222 /* All but the final copy start a new nesting, maintaining the
5223 chain of brackets outstanding. */
5224
5225 if (i != 0)
5226 {
5227 int offset;
5228 *code++ = OP_BRA;
5229 offset = (bralink == NULL)? 0 : (int)(code - bralink);
5230 bralink = code;
5231 PUTINC(code, 0, offset);
5232 }
5233
5234 memcpy(code, previous, IN_UCHARS(len));
5235
5236 /* Ensure there is enough workspace for forward references before
5237 copying them. */
5238
5239 while (cd->hwm > cd->start_workspace + cd->workspace_size -
5240 WORK_SIZE_SAFETY_MARGIN - (this_hwm - save_hwm))
5241 {
5242 int save_offset = save_hwm - cd->start_workspace;
5243 int this_offset = this_hwm - cd->start_workspace;
5244 *errorcodeptr = expand_workspace(cd);
5245 if (*errorcodeptr != 0) goto FAILED;
5246 save_hwm = (pcre_uchar *)cd->start_workspace + save_offset;
5247 this_hwm = (pcre_uchar *)cd->start_workspace + this_offset;
5248 }
5249
5250 for (hc = save_hwm; hc < this_hwm; hc += LINK_SIZE)
5251 {
5252 PUT(cd->hwm, 0, GET(hc, 0) + len + ((i != 0)? 2+LINK_SIZE : 1));
5253 cd->hwm += LINK_SIZE;
5254 }
5255 save_hwm = this_hwm;
5256 code += len;
5257 }
5258
5259 /* Now chain through the pending brackets, and fill in their length
5260 fields (which are holding the chain links pro tem). */
5261
5262 while (bralink != NULL)
5263 {
5264 int oldlinkoffset;
5265 int offset = (int)(code - bralink + 1);
5266 pcre_uchar *bra = code - offset;
5267 oldlinkoffset = GET(bra, 1);
5268 bralink = (oldlinkoffset == 0)? NULL : bralink - oldlinkoffset;
5269 *code++ = OP_KET;
5270 PUTINC(code, 0, offset);
5271 PUT(bra, 1, offset);
5272 }
5273 }
5274
5275 /* If the maximum is unlimited, set a repeater in the final copy. For
5276 ONCE brackets, that's all we need to do. However, possessively repeated
5277 ONCE brackets can be converted into non-capturing brackets, as the
5278 behaviour of (?:xx)++ is the same as (?>xx)++ and this saves having to
5279 deal with possessive ONCEs specially.
5280
5281 Otherwise, when we are doing the actual compile phase, check to see
5282 whether this group is one that could match an empty string. If so,
5283 convert the initial operator to the S form (e.g. OP_BRA -> OP_SBRA) so
5284 that runtime checking can be done. [This check is also applied to ONCE
5285 groups at runtime, but in a different way.]
5286
5287 Then, if the quantifier was possessive and the bracket is not a
5288 conditional, we convert the BRA code to the POS form, and the KET code to
5289 KETRPOS. (It turns out to be convenient at runtime to detect this kind of
5290 subpattern at both the start and at the end.) The use of special opcodes
5291 makes it possible to reduce greatly the stack usage in pcre_exec(). If
5292 the group is preceded by OP_BRAZERO, convert this to OP_BRAPOSZERO.
5293
5294 Then, if the minimum number of matches is 1 or 0, cancel the possessive
5295 flag so that the default action below, of wrapping everything inside
5296 atomic brackets, does not happen. When the minimum is greater than 1,
5297 there will be earlier copies of the group, and so we still have to wrap
5298 the whole thing. */
5299
5300 else
5301 {
5302 pcre_uchar *ketcode = code - 1 - LINK_SIZE;
5303 pcre_uchar *bracode = ketcode - GET(ketcode, 1);
5304
5305 /* Convert possessive ONCE brackets to non-capturing */
5306
5307 if ((*bracode == OP_ONCE || *bracode == OP_ONCE_NC) &&
5308 possessive_quantifier) *bracode = OP_BRA;
5309
5310 /* For non-possessive ONCE brackets, all we need to do is to
5311 set the KET. */
5312
5313 if (*bracode == OP_ONCE || *bracode == OP_ONCE_NC)
5314 *ketcode = OP_KETRMAX + repeat_type;
5315
5316 /* Handle non-ONCE brackets and possessive ONCEs (which have been
5317 converted to non-capturing above). */
5318
5319 else
5320 {
5321 /* In the compile phase, check for empty string matching. */
5322
5323 if (lengthptr == NULL)
5324 {
5325 pcre_uchar *scode = bracode;
5326 do
5327 {
5328 if (could_be_empty_branch(scode, ketcode, utf, cd))
5329 {
5330 *bracode += OP_SBRA - OP_BRA;
5331 break;
5332 }
5333 scode += GET(scode, 1);
5334 }
5335 while (*scode == OP_ALT);
5336 }
5337
5338 /* Handle possessive quantifiers. */
5339
5340 if (possessive_quantifier)
5341 {
5342 /* For COND brackets, we wrap the whole thing in a possessively
5343 repeated non-capturing bracket, because we have not invented POS
5344 versions of the COND opcodes. Because we are moving code along, we
5345 must ensure that any pending recursive references are updated. */
5346
5347 if (*bracode == OP_COND || *bracode == OP_SCOND)
5348 {
5349 int nlen = (int)(code - bracode);
5350 *code = OP_END;
5351 adjust_recurse(bracode, 1 + LINK_SIZE, utf, cd, save_hwm);
5352 memmove(bracode + 1 + LINK_SIZE, bracode, IN_UCHARS(nlen));
5353 code += 1 + LINK_SIZE;
5354 nlen += 1 + LINK_SIZE;
5355 *bracode = OP_BRAPOS;
5356 *code++ = OP_KETRPOS;
5357 PUTINC(code, 0, nlen);
5358 PUT(bracode, 1, nlen);
5359 }
5360
5361 /* For non-COND brackets, we modify the BRA code and use KETRPOS. */
5362
5363 else
5364 {
5365 *bracode += 1; /* Switch to xxxPOS opcodes */
5366 *ketcode = OP_KETRPOS;
5367 }
5368
5369 /* If the minimum is zero, mark it as possessive, then unset the
5370 possessive flag when the minimum is 0 or 1. */
5371
5372 if (brazeroptr != NULL) *brazeroptr = OP_BRAPOSZERO;
5373 if (repeat_min < 2) possessive_quantifier = FALSE;
5374 }
5375
5376 /* Non-possessive quantifier */
5377
5378 else *ketcode = OP_KETRMAX + repeat_type;
5379 }
5380 }
5381 }
5382
5383 /* If previous is OP_FAIL, it was generated by an empty class [] in
5384 JavaScript mode. The other ways in which OP_FAIL can be generated, that is
5385 by (*FAIL) or (?!) set previous to NULL, which gives a "nothing to repeat"
5386 error above. We can just ignore the repeat in JS case. */
5387
5388 else if (*previous == OP_FAIL) goto END_REPEAT;
5389
5390 /* Else there's some kind of shambles */
5391
5392 else
5393 {
5394 *errorcodeptr = ERR11;
5395 goto FAILED;
5396 }
5397
5398 /* If the character following a repeat is '+', or if certain optimization
5399 tests above succeeded, possessive_quantifier is TRUE. For some opcodes,
5400 there are special alternative opcodes for this case. For anything else, we
5401 wrap the entire repeated item inside OP_ONCE brackets. Logically, the '+'
5402 notation is just syntactic sugar, taken from Sun's Java package, but the
5403 special opcodes can optimize it.
5404
5405 Some (but not all) possessively repeated subpatterns have already been
5406 completely handled in the code just above. For them, possessive_quantifier
5407 is always FALSE at this stage.
5408
5409 Note that the repeated item starts at tempcode, not at previous, which
5410 might be the first part of a string whose (former) last char we repeated.
5411
5412 Possessifying an 'exact' quantifier has no effect, so we can ignore it. But
5413 an 'upto' may follow. We skip over an 'exact' item, and then test the
5414 length of what remains before proceeding. */
5415
5416 if (possessive_quantifier)
5417 {
5418 int len;
5419
5420 if (*tempcode == OP_TYPEEXACT)
5421 tempcode += PRIV(OP_lengths)[*tempcode] +
5422 ((tempcode[1 + IMM2_SIZE] == OP_PROP
5423 || tempcode[1 + IMM2_SIZE] == OP_NOTPROP)? 2 : 0);
5424
5425 else if (*tempcode == OP_EXACT || *tempcode == OP_NOTEXACT)
5426 {
5427 tempcode += PRIV(OP_lengths)[*tempcode];
5428 #ifdef SUPPORT_UTF
5429 if (utf && HAS_EXTRALEN(tempcode[-1]))
5430 tempcode += GET_EXTRALEN(tempcode[-1]);
5431 #endif
5432 }
5433
5434 len = (int)(code - tempcode);
5435 if (len > 0) switch (*tempcode)
5436 {
5437 case OP_STAR: *tempcode = OP_POSSTAR; break;
5438 case OP_PLUS: *tempcode = OP_POSPLUS; break;
5439 case OP_QUERY: *tempcode = OP_POSQUERY; break;
5440 case OP_UPTO: *tempcode = OP_POSUPTO; break;
5441
5442 case OP_STARI: *tempcode = OP_POSSTARI; break;
5443 case OP_PLUSI: *tempcode = OP_POSPLUSI; break;
5444 case OP_QUERYI: *tempcode = OP_POSQUERYI; break;
5445 case OP_UPTOI: *tempcode = OP_POSUPTOI; break;
5446
5447 case OP_NOTSTAR: *tempcode = OP_NOTPOSSTAR; break;
5448 case OP_NOTPLUS: *tempcode = OP_NOTPOSPLUS; break;
5449 case OP_NOTQUERY: *tempcode = OP_NOTPOSQUERY; break;
5450 case OP_NOTUPTO: *tempcode = OP_NOTPOSUPTO; break;
5451
5452 case OP_NOTSTARI: *tempcode = OP_NOTPOSSTARI; break;
5453 case OP_NOTPLUSI: *tempcode = OP_NOTPOSPLUSI; break;
5454 case OP_NOTQUERYI: *tempcode = OP_NOTPOSQUERYI; break;
5455 case OP_NOTUPTOI: *tempcode = OP_NOTPOSUPTOI; break;
5456
5457 case OP_TYPESTAR: *tempcode = OP_TYPEPOSSTAR; break;
5458 case OP_TYPEPLUS: *tempcode = OP_TYPEPOSPLUS; break;
5459 case OP_TYPEQUERY: *tempcode = OP_TYPEPOSQUERY; break;
5460 case OP_TYPEUPTO: *tempcode = OP_TYPEPOSUPTO; break;
5461
5462 /* Because we are moving code along, we must ensure that any
5463 pending recursive references are updated. */
5464
5465 default:
5466 *code = OP_END;
5467 adjust_recurse(tempcode, 1 + LINK_SIZE, utf, cd, save_hwm);
5468 memmove(tempcode + 1 + LINK_SIZE, tempcode, IN_UCHARS(len));
5469 code += 1 + LINK_SIZE;
5470 len += 1 + LINK_SIZE;
5471 tempcode[0] = OP_ONCE;
5472 *code++ = OP_KET;
5473 PUTINC(code, 0, len);
5474 PUT(tempcode, 1, len);
5475 break;
5476 }
5477 }
5478
5479 /* In all case we no longer have a previous item. We also set the
5480 "follows varying string" flag for subsequently encountered reqchars if
5481 it isn't already set and we have just passed a varying length item. */
5482
5483 END_REPEAT:
5484 previous = NULL;
5485 cd->req_varyopt |= reqvary;
5486 break;
5487
5488
5489 /* ===================================================================*/
5490 /* Start of nested parenthesized sub-expression, or comment or lookahead or
5491 lookbehind or option setting or condition or all the other extended
5492 parenthesis forms. */
5493
5494 case CHAR_LEFT_PARENTHESIS:
5495 newoptions = options;
5496 skipbytes = 0;
5497 bravalue = OP_CBRA;
5498 save_hwm = cd->hwm;
5499 reset_bracount = FALSE;
5500
5501 /* First deal with various "verbs" that can be introduced by '*'. */
5502
5503 ptr++;
5504 if (ptr[0] == CHAR_ASTERISK && (ptr[1] == ':'
5505 || (MAX_255(ptr[1]) && ((cd->ctypes[ptr[1]] & ctype_letter) != 0))))
5506 {
5507 int i, namelen;
5508 int arglen = 0;
5509 const char *vn = verbnames;
5510 const pcre_uchar *name = ptr + 1;
5511 const pcre_uchar *arg = NULL;
5512 previous = NULL;
5513 ptr++;
5514 while (MAX_255(*ptr) && (cd->ctypes[*ptr] & ctype_letter) != 0) ptr++;
5515 namelen = (int)(ptr - name);
5516
5517 /* It appears that Perl allows any characters whatsoever, other than
5518 a closing parenthesis, to appear in arguments, so we no longer insist on
5519 letters, digits, and underscores. */
5520
5521 if (*ptr == CHAR_COLON)
5522 {
5523 arg = ++ptr;
5524 while (*ptr != 0 && *ptr != CHAR_RIGHT_PARENTHESIS) ptr++;
5525 arglen = (int)(ptr - arg);
5526 }
5527
5528 if (*ptr != CHAR_RIGHT_PARENTHESIS)
5529 {
5530 *errorcodeptr = ERR60;
5531 goto FAILED;
5532 }
5533
5534 /* Scan the table of verb names */
5535
5536 for (i = 0; i < verbcount; i++)
5537 {
5538 if (namelen == verbs[i].len &&
5539 STRNCMP_UC_C8(name, vn, namelen) == 0)
5540 {
5541 /* Check for open captures before ACCEPT and convert it to
5542 ASSERT_ACCEPT if in an assertion. */
5543
5544 if (verbs[i].op == OP_ACCEPT)
5545 {
5546 open_capitem *oc;
5547 if (arglen != 0)
5548 {
5549 *errorcodeptr = ERR59;
5550 goto FAILED;
5551 }
5552 cd->had_accept = TRUE;
5553 for (oc = cd->open_caps; oc != NULL; oc = oc->next)
5554 {
5555 *code++ = OP_CLOSE;
5556 PUT2INC(code, 0, oc->number);
5557 }
5558 *code++ = (cd->assert_depth > 0)? OP_ASSERT_ACCEPT : OP_ACCEPT;
5559
5560 /* Do not set firstchar after *ACCEPT */
5561 if (firstchar == REQ_UNSET) firstchar = REQ_NONE;
5562 }
5563
5564 /* Handle other cases with/without an argument */
5565
5566 else if (arglen == 0)
5567 {
5568 if (verbs[i].op < 0) /* Argument is mandatory */
5569 {
5570 *errorcodeptr = ERR66;
5571 goto FAILED;
5572 }
5573 *code = verbs[i].op;
5574 if (*code++ == OP_THEN) cd->external_flags |= PCRE_HASTHEN;
5575 }
5576
5577 else
5578 {
5579 if (verbs[i].op_arg < 0) /* Argument is forbidden */
5580 {
5581 *errorcodeptr = ERR59;
5582 goto FAILED;
5583 }
5584 *code = verbs[i].op_arg;
5585 if (*code++ == OP_THEN_ARG) cd->external_flags |= PCRE_HASTHEN;
5586 *code++ = arglen;
5587 memcpy(code, arg, IN_UCHARS(arglen));
5588 code += arglen;
5589 *code++ = 0;
5590 }
5591
5592 break; /* Found verb, exit loop */
5593 }
5594
5595 vn += verbs[i].len + 1;
5596 }
5597
5598 if (i < verbcount) continue; /* Successfully handled a verb */
5599 *errorcodeptr = ERR60; /* Verb not recognized */
5600 goto FAILED;
5601 }
5602
5603 /* Deal with the extended parentheses; all are introduced by '?', and the
5604 appearance of any of them means that this is not a capturing group. */
5605
5606 else if (*ptr == CHAR_QUESTION_MARK)
5607 {
5608 int i, set, unset, namelen;
5609 int *optset;
5610 const pcre_uchar *name;
5611 pcre_uchar *slot;
5612
5613 switch (*(++ptr))
5614 {
5615 case CHAR_NUMBER_SIGN: /* Comment; skip to ket */
5616 ptr++;
5617 while (*ptr != 0 && *ptr != CHAR_RIGHT_PARENTHESIS) ptr++;
5618 if (*ptr == 0)
5619 {
5620 *errorcodeptr = ERR18;
5621 goto FAILED;
5622 }
5623 continue;
5624
5625
5626 /* ------------------------------------------------------------ */
5627 case CHAR_VERTICAL_LINE: /* Reset capture count for each branch */
5628 reset_bracount = TRUE;
5629 /* Fall through */
5630
5631 /* ------------------------------------------------------------ */
5632 case CHAR_COLON: /* Non-capturing bracket */
5633 bravalue = OP_BRA;
5634 ptr++;
5635 break;
5636
5637
5638 /* ------------------------------------------------------------ */
5639 case CHAR_LEFT_PARENTHESIS:
5640 bravalue = OP_COND; /* Conditional group */
5641
5642 /* A condition can be an assertion, a number (referring to a numbered
5643 group), a name (referring to a named group), or 'R', referring to
5644 recursion. R<digits> and R&name are also permitted for recursion tests.
5645
5646 There are several syntaxes for testing a named group: (?(name)) is used
5647 by Python; Perl 5.10 onwards uses (?(<name>) or (?('name')).
5648
5649 There are two unfortunate ambiguities, caused by history. (a) 'R' can
5650 be the recursive thing or the name 'R' (and similarly for 'R' followed
5651 by digits), and (b) a number could be a name that consists of digits.
5652 In both cases, we look for a name first; if not found, we try the other
5653 cases. */
5654
5655 /* For conditions that are assertions, check the syntax, and then exit
5656 the switch. This will take control down to where bracketed groups,
5657 including assertions, are processed. */
5658
5659 if (ptr[1] == CHAR_QUESTION_MARK && (ptr[2] == CHAR_EQUALS_SIGN ||
5660 ptr[2] == CHAR_EXCLAMATION_MARK || ptr[2] == CHAR_LESS_THAN_SIGN))
5661 break;
5662
5663 /* Most other conditions use OP_CREF (a couple change to OP_RREF
5664 below), and all need to skip 1+IMM2_SIZE bytes at the start of the group. */
5665
5666 code[1+LINK_SIZE] = OP_CREF;
5667 skipbytes = 1+IMM2_SIZE;
5668 refsign = -1;
5669
5670 /* Check for a test for recursion in a named group. */
5671
5672 if (ptr[1] == CHAR_R && ptr[2] == CHAR_AMPERSAND)
5673 {
5674 terminator = -1;
5675 ptr += 2;
5676 code[1+LINK_SIZE] = OP_RREF; /* Change the type of test */
5677 }
5678
5679 /* Check for a test for a named group's having been set, using the Perl
5680 syntax (?(<name>) or (?('name') */
5681
5682 else if (ptr[1] == CHAR_LESS_THAN_SIGN)
5683 {
5684 terminator = CHAR_GREATER_THAN_SIGN;
5685 ptr++;
5686 }
5687 else if (ptr[1] == CHAR_APOSTROPHE)
5688 {
5689 terminator = CHAR_APOSTROPHE;
5690 ptr++;
5691 }
5692 else
5693 {
5694 terminator = 0;
5695 if (ptr[1] == CHAR_MINUS || ptr[1] == CHAR_PLUS) refsign = *(++ptr);
5696 }
5697
5698 /* We now expect to read a name; any thing else is an error */
5699
5700 if (!MAX_255(ptr[1]) || (cd->ctypes[ptr[1]] & ctype_word) == 0)
5701 {
5702 ptr += 1; /* To get the right offset */
5703 *errorcodeptr = ERR28;
5704 goto FAILED;
5705 }
5706
5707 /* Read the name, but also get it as a number if it's all digits */
5708
5709 recno = 0;
5710 name = ++ptr;
5711 while (MAX_255(*ptr) && (cd->ctypes[*ptr] & ctype_word) != 0)
5712 {
5713 if (recno >= 0)
5714 recno = (IS_DIGIT(*ptr))? recno * 10 + *ptr - CHAR_0 : -1;
5715 ptr++;
5716 }
5717 namelen = (int)(ptr - name);
5718
5719 if ((terminator > 0 && *ptr++ != terminator) ||
5720 *ptr++ != CHAR_RIGHT_PARENTHESIS)
5721 {
5722 ptr--; /* Error offset */
5723 *errorcodeptr = ERR26;
5724 goto FAILED;
5725 }
5726
5727 /* Do no further checking in the pre-compile phase. */
5728
5729 if (lengthptr != NULL) break;
5730
5731 /* In the real compile we do the work of looking for the actual
5732 reference. If the string started with "+" or "-" we require the rest to
5733 be digits, in which case recno will be set. */
5734
5735 if (refsign > 0)
5736 {
5737 if (recno <= 0)
5738 {
5739 *errorcodeptr = ERR58;
5740 goto FAILED;
5741 }
5742 recno = (refsign == CHAR_MINUS)?
5743 cd->bracount - recno + 1 : recno +cd->bracount;
5744 if (recno <= 0 || recno > cd->final_bracount)
5745 {
5746 *errorcodeptr = ERR15;
5747 goto FAILED;
5748 }
5749 PUT2(code, 2+LINK_SIZE, recno);
5750 break;
5751 }
5752
5753 /* Otherwise (did not start with "+" or "-"), start by looking for the
5754 name. If we find a name, add one to the opcode to change OP_CREF or
5755 OP_RREF into OP_NCREF or OP_NRREF. These behave exactly the same,
5756 except they record that the reference was originally to a name. The
5757 information is used to check duplicate names. */
5758
5759 slot = cd->name_table;
5760 for (i = 0; i < cd->names_found; i++)
5761 {
5762 if (STRNCMP_UC_UC(name, slot+IMM2_SIZE, namelen) == 0) break;
5763 slot += cd->name_entry_size;
5764 }
5765
5766 /* Found a previous named subpattern */
5767
5768 if (i < cd->names_found)
5769 {
5770 recno = GET2(slot, 0);
5771 PUT2(code, 2+LINK_SIZE, recno);
5772 code[1+LINK_SIZE]++;
5773 }
5774
5775 /* Search the pattern for a forward reference */
5776
5777 else if ((i = find_parens(cd, name, namelen,
5778 (options & PCRE_EXTENDED) != 0, utf)) > 0)
5779 {
5780 PUT2(code, 2+LINK_SIZE, i);
5781 code[1+LINK_SIZE]++;
5782 }
5783
5784 /* If terminator == 0 it means that the name followed directly after
5785 the opening parenthesis [e.g. (?(abc)...] and in this case there are
5786 some further alternatives to try. For the cases where terminator != 0
5787 [things like (?(<name>... or (?('name')... or (?(R&name)... ] we have
5788 now checked all the possibilities, so give an error. */
5789
5790 else if (terminator != 0)
5791 {
5792 *errorcodeptr = ERR15;
5793 goto FAILED;
5794 }
5795
5796 /* Check for (?(R) for recursion. Allow digits after R to specify a
5797 specific group number. */
5798
5799 else if (*name == CHAR_R)
5800 {
5801 recno = 0;
5802 for (i = 1; i < namelen; i++)
5803 {
5804 if (!IS_DIGIT(name[i]))
5805 {
5806 *errorcodeptr = ERR15;
5807 goto FAILED;
5808 }
5809 recno = recno * 10 + name[i] - CHAR_0;
5810 }
5811 if (recno == 0) recno = RREF_ANY;
5812 code[1+LINK_SIZE] = OP_RREF; /* Change test type */
5813 PUT2(code, 2+LINK_SIZE, recno);
5814 }
5815
5816 /* Similarly, check for the (?(DEFINE) "condition", which is always
5817 false. */
5818
5819 else if (namelen == 6 && STRNCMP_UC_C8(name, STRING_DEFINE, 6) == 0)
5820 {
5821 code[1+LINK_SIZE] = OP_DEF;
5822 skipbytes = 1;
5823 }
5824
5825 /* Check for the "name" actually being a subpattern number. We are
5826 in the second pass here, so final_bracount is set. */
5827
5828 else if (recno > 0 && recno <= cd->final_bracount)
5829 {
5830 PUT2(code, 2+LINK_SIZE, recno);
5831 }
5832
5833 /* Either an unidentified subpattern, or a reference to (?(0) */
5834
5835 else
5836 {
5837 *errorcodeptr = (recno == 0)? ERR35: ERR15;
5838 goto FAILED;
5839 }
5840 break;
5841
5842
5843 /* ------------------------------------------------------------ */
5844 case CHAR_EQUALS_SIGN: /* Positive lookahead */
5845 bravalue = OP_ASSERT;
5846 cd->assert_depth += 1;
5847 ptr++;
5848 break;
5849
5850
5851 /* ------------------------------------------------------------ */
5852 case CHAR_EXCLAMATION_MARK: /* Negative lookahead */
5853 ptr++;
5854 if (*ptr == CHAR_RIGHT_PARENTHESIS) /* Optimize (?!) */
5855 {
5856 *code++ = OP_FAIL;
5857 previous = NULL;
5858 continue;
5859 }
5860 bravalue = OP_ASSERT_NOT;
5861 cd->assert_depth += 1;
5862 break;
5863
5864
5865 /* ------------------------------------------------------------ */
5866 case CHAR_LESS_THAN_SIGN: /* Lookbehind or named define */
5867 switch (ptr[1])
5868 {
5869 case CHAR_EQUALS_SIGN: /* Positive lookbehind */
5870 bravalue = OP_ASSERTBACK;
5871 cd->assert_depth += 1;
5872 ptr += 2;
5873 break;
5874
5875 case CHAR_EXCLAMATION_MARK: /* Negative lookbehind */
5876 bravalue = OP_ASSERTBACK_NOT;
5877 cd->assert_depth += 1;
5878 ptr += 2;
5879 break;
5880
5881 default: /* Could be name define, else bad */
5882 if (MAX_255(ptr[1]) && (cd->ctypes[ptr[1]] & ctype_word) != 0)
5883 goto DEFINE_NAME;
5884 ptr++; /* Correct offset for error */
5885 *errorcodeptr = ERR24;
5886 goto FAILED;
5887 }
5888 break;
5889
5890
5891 /* ------------------------------------------------------------ */
5892 case CHAR_GREATER_THAN_SIGN: /* One-time brackets */
5893 bravalue = OP_ONCE;
5894 ptr++;
5895 break;
5896
5897
5898 /* ------------------------------------------------------------ */
5899 case CHAR_C: /* Callout - may be followed by digits; */
5900 previous_callout = code; /* Save for later completion */
5901 after_manual_callout = 1; /* Skip one item before completing */
5902 *code++ = OP_CALLOUT;
5903 {
5904 int n = 0;
5905 ptr++;
5906 while(IS_DIGIT(*ptr))
5907 n = n * 10 + *ptr++ - CHAR_0;
5908 if (*ptr != CHAR_RIGHT_PARENTHESIS)
5909 {
5910 *errorcodeptr = ERR39;
5911 goto FAILED;
5912 }
5913 if (n > 255)
5914 {
5915 *errorcodeptr = ERR38;
5916 goto FAILED;
5917 }
5918 *code++ = n;
5919 PUT(code, 0, (int)(ptr - cd->start_pattern + 1)); /* Pattern offset */
5920 PUT(code, LINK_SIZE, 0); /* Default length */
5921 code += 2 * LINK_SIZE;
5922 }
5923 previous = NULL;
5924 continue;
5925
5926
5927 /* ------------------------------------------------------------ */
5928 case CHAR_P: /* Python-style named subpattern handling */
5929 if (*(++ptr) == CHAR_EQUALS_SIGN ||
5930 *ptr == CHAR_GREATER_THAN_SIGN) /* Reference or recursion */
5931 {
5932 is_recurse = *ptr == CHAR_GREATER_THAN_SIGN;
5933 terminator = CHAR_RIGHT_PARENTHESIS;
5934 goto NAMED_REF_OR_RECURSE;
5935 }
5936 else if (*ptr != CHAR_LESS_THAN_SIGN) /* Test for Python-style defn */
5937 {
5938 *errorcodeptr = ERR41;
5939 goto FAILED;
5940 }
5941 /* Fall through to handle (?P< as (?< is handled */
5942
5943
5944 /* ------------------------------------------------------------ */
5945 DEFINE_NAME: /* Come here from (?< handling */
5946 case CHAR_APOSTROPHE:
5947 {
5948 terminator = (*ptr == CHAR_LESS_THAN_SIGN)?
5949 CHAR_GREATER_THAN_SIGN : CHAR_APOSTROPHE;
5950 name = ++ptr;
5951
5952 while (MAX_255(*ptr) && (cd->ctypes[*ptr] & ctype_word) != 0) ptr++;
5953 namelen = (int)(ptr - name);
5954
5955 /* In the pre-compile phase, just do a syntax check. */
5956
5957 if (lengthptr != NULL)
5958 {
5959 if (*ptr != terminator)
5960 {
5961 *errorcodeptr = ERR42;
5962 goto FAILED;
5963 }
5964 if (cd->names_found >= MAX_NAME_COUNT)
5965 {
5966 *errorcodeptr = ERR49;
5967 goto FAILED;
5968 }
5969 if (namelen + IMM2_SIZE + 1 > cd->name_entry_size)
5970 {
5971 cd->name_entry_size = namelen + IMM2_SIZE + 1;
5972 if (namelen > MAX_NAME_SIZE)
5973 {
5974 *errorcodeptr = ERR48;
5975 goto FAILED;
5976 }
5977 }
5978 }
5979
5980 /* In the real compile, create the entry in the table, maintaining
5981 alphabetical order. Duplicate names for different numbers are
5982 permitted only if PCRE_DUPNAMES is set. Duplicate names for the same
5983 number are always OK. (An existing number can be re-used if (?|
5984 appears in the pattern.) In either event, a duplicate name results in
5985 a duplicate entry in the table, even if the number is the same. This
5986 is because the number of names, and hence the table size, is computed
5987 in the pre-compile, and it affects various numbers and pointers which
5988 would all have to be modified, and the compiled code moved down, if
5989 duplicates with the same number were omitted from the table. This
5990 doesn't seem worth the hassle. However, *different* names for the
5991 same number are not permitted. */
5992
5993 else
5994 {
5995 BOOL dupname = FALSE;
5996 slot = cd->name_table;
5997
5998 for (i = 0; i < cd->names_found; i++)
5999 {
6000 int crc = memcmp(name, slot+IMM2_SIZE, IN_UCHARS(namelen));
6001 if (crc == 0)
6002 {
6003 if (slot[IMM2_SIZE+namelen] == 0)
6004 {
6005 if (GET2(slot, 0) != cd->bracount + 1 &&
6006 (options & PCRE_DUPNAMES) == 0)
6007 {
6008 *errorcodeptr = ERR43;
6009 goto FAILED;
6010 }
6011 else dupname = TRUE;
6012 }
6013 else crc = -1; /* Current name is a substring */
6014 }
6015
6016 /* Make space in the table and break the loop for an earlier
6017 name. For a duplicate or later name, carry on. We do this for
6018 duplicates so that in the simple case (when ?(| is not used) they
6019 are in order of their numbers. */
6020
6021 if (crc < 0)
6022 {
6023 memmove(slot + cd->name_entry_size, slot,
6024 IN_UCHARS((cd->names_found - i) * cd->name_entry_size));
6025 break;
6026 }
6027
6028 /* Continue the loop for a later or duplicate name */
6029
6030 slot += cd->name_entry_size;
6031 }
6032
6033 /* For non-duplicate names, check for a duplicate number before
6034 adding the new name. */
6035
6036 if (!dupname)
6037 {
6038 pcre_uchar *cslot = cd->name_table;
6039 for (i = 0; i < cd->names_found; i++)
6040 {
6041 if (cslot != slot)
6042 {
6043 if (GET2(cslot, 0) == cd->bracount + 1)
6044 {
6045 *errorcodeptr = ERR65;
6046 goto FAILED;
6047 }
6048 }
6049 else i--;
6050 cslot += cd->name_entry_size;
6051 }
6052 }
6053
6054 PUT2(slot, 0, cd->bracount + 1);
6055 memcpy(slot + IMM2_SIZE, name, IN_UCHARS(namelen));
6056 slot[IMM2_SIZE + namelen] = 0;
6057 }
6058 }
6059
6060 /* In both pre-compile and compile, count the number of names we've
6061 encountered. */
6062
6063 cd->names_found++;
6064 ptr++; /* Move past > or ' */
6065 goto NUMBERED_GROUP;
6066
6067
6068 /* ------------------------------------------------------------ */
6069 case CHAR_AMPERSAND: /* Perl recursion/subroutine syntax */
6070 terminator = CHAR_RIGHT_PARENTHESIS;
6071 is_recurse = TRUE;
6072 /* Fall through */
6073
6074 /* We come here from the Python syntax above that handles both
6075 references (?P=name) and recursion (?P>name), as well as falling
6076 through from the Perl recursion syntax (?&name). We also come here from
6077 the Perl \k<name> or \k'name' back reference syntax and the \k{name}
6078 .NET syntax, and the Oniguruma \g<...> and \g'...' subroutine syntax. */
6079
6080 NAMED_REF_OR_RECURSE:
6081 name = ++ptr;
6082 while (MAX_255(*ptr) && (cd->ctypes[*ptr] & ctype_word) != 0) ptr++;
6083 namelen = (int)(ptr - name);
6084
6085 /* In the pre-compile phase, do a syntax check. We used to just set
6086 a dummy reference number, because it was not used in the first pass.
6087 However, with the change of recursive back references to be atomic,
6088 we have to look for the number so that this state can be identified, as
6089 otherwise the incorrect length is computed. If it's not a backwards
6090 reference, the dummy number will do. */
6091
6092 if (lengthptr != NULL)
6093 {
6094 const pcre_uchar *temp;
6095
6096 if (namelen == 0)
6097 {
6098 *errorcodeptr = ERR62;
6099 goto FAILED;
6100 }
6101 if (*ptr != terminator)
6102 {
6103 *errorcodeptr = ERR42;
6104 goto FAILED;
6105 }
6106 if (namelen > MAX_NAME_SIZE)
6107 {
6108 *errorcodeptr = ERR48;
6109 goto FAILED;
6110 }
6111
6112 /* The name table does not exist in the first pass, so we cannot
6113 do a simple search as in the code below. Instead, we have to scan the
6114 pattern to find the number. It is important that we scan it only as
6115 far as we have got because the syntax of named subpatterns has not
6116 been checked for the rest of the pattern, and find_parens() assumes
6117 correct syntax. In any case, it's a waste of resources to scan
6118 further. We stop the scan at the current point by temporarily
6119 adjusting the value of cd->endpattern. */
6120
6121 temp = cd->end_pattern;
6122 cd->end_pattern = ptr;
6123 recno = find_parens(cd, name, namelen,
6124 (options & PCRE_EXTENDED) != 0, utf);
6125 cd->end_pattern = temp;
6126 if (recno < 0) recno = 0; /* Forward ref; set dummy number */
6127 }
6128
6129 /* In the real compile, seek the name in the table. We check the name
6130 first, and then check that we have reached the end of the name in the
6131 table. That way, if the name that is longer than any in the table,
6132 the comparison will fail without reading beyond the table entry. */
6133
6134 else
6135 {
6136 slot = cd->name_table;
6137 for (i = 0; i < cd->names_found; i++)
6138 {
6139 if (STRNCMP_UC_UC(name, slot+IMM2_SIZE, namelen) == 0 &&
6140 slot[IMM2_SIZE+namelen] == 0)
6141 break;
6142 slot += cd->name_entry_size;
6143 }
6144
6145 if (i < cd->names_found) /* Back reference */
6146 {
6147 recno = GET2(slot, 0);
6148 }
6149 else if ((recno = /* Forward back reference */
6150 find_parens(cd, name, namelen,
6151 (options & PCRE_EXTENDED) != 0, utf)) <= 0)
6152 {
6153 *errorcodeptr = ERR15;
6154 goto FAILED;
6155 }
6156 }
6157
6158 /* In both phases, we can now go to the code than handles numerical
6159 recursion or backreferences. */
6160
6161 if (is_recurse) goto HANDLE_RECURSION;
6162 else goto HANDLE_REFERENCE;
6163
6164
6165 /* ------------------------------------------------------------ */
6166 case CHAR_R: /* Recursion */
6167 ptr++; /* Same as (?0) */
6168 /* Fall through */
6169
6170
6171 /* ------------------------------------------------------------ */
6172 case CHAR_MINUS: case CHAR_PLUS: /* Recursion or subroutine */
6173 case CHAR_0: case CHAR_1: case CHAR_2: case CHAR_3: case CHAR_4:
6174 case CHAR_5: case CHAR_6: case CHAR_7: case CHAR_8: case CHAR_9:
6175 {
6176 const pcre_uchar *called;
6177 terminator = CHAR_RIGHT_PARENTHESIS;
6178
6179 /* Come here from the \g<...> and \g'...' code (Oniguruma
6180 compatibility). However, the syntax has been checked to ensure that
6181 the ... are a (signed) number, so that neither ERR63 nor ERR29 will
6182 be called on this path, nor with the jump to OTHER_CHAR_AFTER_QUERY
6183 ever be taken. */
6184
6185 HANDLE_NUMERICAL_RECURSION:
6186
6187 if ((refsign = *ptr) == CHAR_PLUS)
6188 {
6189 ptr++;
6190 if (!IS_DIGIT(*ptr))
6191 {
6192 *errorcodeptr = ERR63;
6193 goto FAILED;
6194 }
6195 }
6196 else if (refsign == CHAR_MINUS)
6197 {
6198 if (!IS_DIGIT(ptr[1]))
6199 goto OTHER_CHAR_AFTER_QUERY;
6200 ptr++;
6201 }
6202
6203 recno = 0;
6204 while(IS_DIGIT(*ptr))
6205 recno = recno * 10 + *ptr++ - CHAR_0;
6206
6207 if (*ptr != terminator)
6208 {
6209 *errorcodeptr = ERR29;
6210 goto FAILED;
6211 }
6212
6213 if (refsign == CHAR_MINUS)
6214 {
6215 if (recno == 0)
6216 {
6217 *errorcodeptr = ERR58;
6218 goto FAILED;
6219 }
6220 recno = cd->bracount - recno + 1;
6221 if (recno <= 0)
6222 {
6223 *errorcodeptr = ERR15;
6224 goto FAILED;
6225 }
6226 }
6227 else if (refsign == CHAR_PLUS)
6228 {
6229 if (recno == 0)
6230 {
6231 *errorcodeptr = ERR58;
6232 goto FAILED;
6233 }
6234 recno += cd->bracount;
6235 }
6236
6237 /* Come here from code above that handles a named recursion */
6238
6239 HANDLE_RECURSION:
6240
6241 previous = code;
6242 called = cd->start_code;
6243
6244 /* When we are actually compiling, find the bracket that is being
6245 referenced. Temporarily end the regex in case it doesn't exist before
6246 this point. If we end up with a forward reference, first check that
6247 the bracket does occur later so we can give the error (and position)
6248 now. Then remember this forward reference in the workspace so it can
6249 be filled in at the end. */
6250
6251 if (lengthptr == NULL)
6252 {
6253 *code = OP_END;
6254 if (recno != 0)
6255 called = PRIV(find_bracket)(cd->start_code, utf, recno);
6256
6257 /* Forward reference */
6258
6259 if (called == NULL)
6260 {
6261 if (find_parens(cd, NULL, recno,
6262 (options & PCRE_EXTENDED) != 0, utf) < 0)
6263 {
6264 *errorcodeptr = ERR15;
6265 goto FAILED;
6266 }
6267
6268 /* Fudge the value of "called" so that when it is inserted as an
6269 offset below, what it actually inserted is the reference number
6270 of the group. Then remember the forward reference. */
6271
6272 called = cd->start_code + recno;
6273 if (cd->hwm >= cd->start_workspace + cd->workspace_size -
6274 WORK_SIZE_SAFETY_MARGIN)
6275 {
6276 *errorcodeptr = expand_workspace(cd);
6277 if (*errorcodeptr != 0) goto FAILED;
6278 }
6279 PUTINC(cd->hwm, 0, (int)(code + 1 - cd->start_code));
6280 }
6281
6282 /* If not a forward reference, and the subpattern is still open,
6283 this is a recursive call. We check to see if this is a left
6284 recursion that could loop for ever, and diagnose that case. We
6285 must not, however, do this check if we are in a conditional
6286 subpattern because the condition might be testing for recursion in
6287 a pattern such as /(?(R)a+|(?R)b)/, which is perfectly valid.
6288 Forever loops are also detected at runtime, so those that occur in
6289 conditional subpatterns will be picked up then. */
6290
6291 else if (GET(called, 1) == 0 && cond_depth <= 0 &&
6292 could_be_empty(called, code, bcptr, utf, cd))
6293 {
6294 *errorcodeptr = ERR40;
6295 goto FAILED;
6296 }
6297 }
6298
6299 /* Insert the recursion/subroutine item. It does not have a set first
6300 character (relevant if it is repeated, because it will then be
6301 wrapped with ONCE brackets). */
6302
6303 *code = OP_RECURSE;
6304 PUT(code, 1, (int)(called - cd->start_code));
6305 code += 1 + LINK_SIZE;
6306 groupsetfirstchar = FALSE;
6307 }
6308
6309 /* Can't determine a first byte now */
6310
6311 if (firstchar == REQ_UNSET) firstchar = REQ_NONE;
6312 continue;
6313
6314
6315 /* ------------------------------------------------------------ */
6316 default: /* Other characters: check option setting */
6317 OTHER_CHAR_AFTER_QUERY:
6318 set = unset = 0;
6319 optset = &set;
6320
6321 while (*ptr != CHAR_RIGHT_PARENTHESIS && *ptr != CHAR_COLON)
6322 {
6323 switch (*ptr++)
6324 {
6325 case CHAR_MINUS: optset = &unset; break;
6326
6327 case CHAR_J: /* Record that it changed in the external options */
6328 *optset |= PCRE_DUPNAMES;
6329 cd->external_flags |= PCRE_JCHANGED;
6330 break;
6331
6332 case CHAR_i: *optset |= PCRE_CASELESS; break;
6333 case CHAR_m: *optset |= PCRE_MULTILINE; break;
6334 case CHAR_s: *optset |= PCRE_DOTALL; break;
6335 case CHAR_x: *optset |= PCRE_EXTENDED; break;
6336 case CHAR_U: *optset |= PCRE_UNGREEDY; break;
6337 case CHAR_X: *optset |= PCRE_EXTRA; break;
6338
6339 default: *errorcodeptr = ERR12;
6340 ptr--; /* Correct the offset */
6341 goto FAILED;
6342 }
6343 }
6344
6345 /* Set up the changed option bits, but don't change anything yet. */
6346
6347 newoptions = (options | set) & (~unset);
6348
6349 /* If the options ended with ')' this is not the start of a nested
6350 group with option changes, so the options change at this level. If this
6351 item is right at the start of the pattern, the options can be
6352 abstracted and made external in the pre-compile phase, and ignored in
6353 the compile phase. This can be helpful when matching -- for instance in
6354 caseless checking of required bytes.
6355
6356 If the code pointer is not (cd->start_code + 1 + LINK_SIZE), we are
6357 definitely *not* at the start of the pattern because something has been
6358 compiled. In the pre-compile phase, however, the code pointer can have
6359 that value after the start, because it gets reset as code is discarded
6360 during the pre-compile. However, this can happen only at top level - if
6361 we are within parentheses, the starting BRA will still be present. At
6362 any parenthesis level, the length value can be used to test if anything
6363 has been compiled at that level. Thus, a test for both these conditions
6364 is necessary to ensure we correctly detect the start of the pattern in
6365 both phases.
6366
6367 If we are not at the pattern start, reset the greedy defaults and the
6368 case value for firstchar and reqchar. */
6369
6370 if (*ptr == CHAR_RIGHT_PARENTHESIS)
6371 {
6372 if (code == cd->start_code + 1 + LINK_SIZE &&
6373 (lengthptr == NULL || *lengthptr == 2 + 2*LINK_SIZE))
6374 {
6375 cd->external_options = newoptions;
6376 }
6377 else
6378 {
6379 greedy_default = ((newoptions & PCRE_UNGREEDY) != 0);
6380 greedy_non_default = greedy_default ^ 1;
6381 req_caseopt = ((newoptions & PCRE_CASELESS) != 0)? REQ_CASELESS:0;
6382 }
6383
6384 /* Change options at this level, and pass them back for use
6385 in subsequent branches. */
6386
6387 *optionsptr = options = newoptions;
6388 previous = NULL; /* This item can't be repeated */
6389 continue; /* It is complete */
6390 }
6391
6392 /* If the options ended with ':' we are heading into a nested group
6393 with possible change of options. Such groups are non-capturing and are
6394 not assertions of any kind. All we need to do is skip over the ':';
6395 the newoptions value is handled below. */
6396
6397 bravalue = OP_BRA;
6398 ptr++;
6399 } /* End of switch for character following (? */
6400 } /* End of (? handling */
6401
6402 /* Opening parenthesis not followed by '*' or '?'. If PCRE_NO_AUTO_CAPTURE
6403 is set, all unadorned brackets become non-capturing and behave like (?:...)
6404 brackets. */
6405
6406 else if ((options & PCRE_NO_AUTO_CAPTURE) != 0)
6407 {
6408 bravalue = OP_BRA;
6409 }
6410
6411 /* Else we have a capturing group. */
6412
6413 else
6414 {
6415 NUMBERED_GROUP:
6416 cd->bracount += 1;
6417 PUT2(code, 1+LINK_SIZE, cd->bracount);
6418 skipbytes = IMM2_SIZE;
6419 }
6420
6421 /* Process nested bracketed regex. Assertions used not to be repeatable,
6422 but this was changed for Perl compatibility, so all kinds can now be
6423 repeated. We copy code into a non-register variable (tempcode) in order to
6424 be able to pass its address because some compilers complain otherwise. */
6425
6426 previous = code; /* For handling repetition */
6427 *code = bravalue;
6428 tempcode = code;
6429 tempreqvary = cd->req_varyopt; /* Save value before bracket */
6430 tempbracount = cd->bracount; /* Save value before bracket */
6431 length_prevgroup = 0; /* Initialize for pre-compile phase */
6432
6433 if (!compile_regex(
6434 newoptions, /* The complete new option state */
6435 &tempcode, /* Where to put code (updated) */
6436 &ptr, /* Input pointer (updated) */
6437 errorcodeptr, /* Where to put an error message */
6438 (bravalue == OP_ASSERTBACK ||
6439 bravalue == OP_ASSERTBACK_NOT), /* TRUE if back assert */
6440 reset_bracount, /* True if (?| group */
6441 skipbytes, /* Skip over bracket number */
6442 cond_depth +
6443 ((bravalue == OP_COND)?1:0), /* Depth of condition subpatterns */
6444 &subfirstchar, /* For possible first char */
6445 &subreqchar, /* For possible last char */
6446 bcptr, /* Current branch chain */
6447 cd, /* Tables block */
6448 (lengthptr == NULL)? NULL : /* Actual compile phase */
6449 &length_prevgroup /* Pre-compile phase */
6450 ))
6451 goto FAILED;
6452
6453 /* If this was an atomic group and there are no capturing groups within it,
6454 generate OP_ONCE_NC instead of OP_ONCE. */
6455
6456 if (bravalue == OP_ONCE && cd->bracount <= tempbracount)
6457 *code = OP_ONCE_NC;
6458
6459 if (bravalue >= OP_ASSERT && bravalue <= OP_ASSERTBACK_NOT)
6460 cd->assert_depth -= 1;
6461
6462 /* At the end of compiling, code is still pointing to the start of the
6463 group, while tempcode has been updated to point past the end of the group.
6464 The pattern pointer (ptr) is on the bracket.
6465
6466 If this is a conditional bracket, check that there are no more than
6467 two branches in the group, or just one if it's a DEFINE group. We do this
6468 in the real compile phase, not in the pre-pass, where the whole group may
6469 not be available. */
6470
6471 if (bravalue == OP_COND && lengthptr == NULL)
6472 {
6473 pcre_uchar *tc = code;
6474 int condcount = 0;
6475
6476 do {
6477 condcount++;
6478 tc += GET(tc,1);
6479 }
6480 while (*tc != OP_KET);
6481
6482 /* A DEFINE group is never obeyed inline (the "condition" is always
6483 false). It must have only one branch. */
6484
6485 if (code[LINK_SIZE+1] == OP_DEF)
6486 {
6487 if (condcount > 1)
6488 {
6489 *errorcodeptr = ERR54;
6490 goto FAILED;
6491 }
6492 bravalue = OP_DEF; /* Just a flag to suppress char handling below */
6493 }
6494
6495 /* A "normal" conditional group. If there is just one branch, we must not
6496 make use of its firstchar or reqchar, because this is equivalent to an
6497 empty second branch. */
6498
6499 else
6500 {
6501 if (condcount > 2)
6502 {
6503 *errorcodeptr = ERR27;
6504 goto FAILED;
6505 }
6506 if (condcount == 1) subfirstchar = subreqchar = REQ_NONE;
6507 }
6508 }
6509
6510 /* Error if hit end of pattern */
6511
6512 if (*ptr != CHAR_RIGHT_PARENTHESIS)
6513 {
6514 *errorcodeptr = ERR14;
6515 goto FAILED;
6516 }
6517
6518 /* In the pre-compile phase, update the length by the length of the group,
6519 less the brackets at either end. Then reduce the compiled code to just a
6520 set of non-capturing brackets so that it doesn't use much memory if it is
6521 duplicated by a quantifier.*/
6522
6523 if (lengthptr != NULL)
6524 {
6525 if (OFLOW_MAX - *lengthptr < length_prevgroup - 2 - 2*LINK_SIZE)
6526 {
6527 *errorcodeptr = ERR20;
6528 goto FAILED;
6529 }
6530 *lengthptr += length_prevgroup - 2 - 2*LINK_SIZE;
6531 code++; /* This already contains bravalue */
6532 PUTINC(code, 0, 1 + LINK_SIZE);
6533 *code++ = OP_KET;
6534 PUTINC(code, 0, 1 + LINK_SIZE);
6535 break; /* No need to waste time with special character handling */
6536 }
6537
6538 /* Otherwise update the main code pointer to the end of the group. */
6539
6540 code = tempcode;
6541
6542 /* For a DEFINE group, required and first character settings are not
6543 relevant. */
6544
6545 if (bravalue == OP_DEF) break;
6546
6547 /* Handle updating of the required and first characters for other types of
6548 group. Update for normal brackets of all kinds, and conditions with two
6549 branches (see code above). If the bracket is followed by a quantifier with
6550 zero repeat, we have to back off. Hence the definition of zeroreqchar and
6551 zerofirstchar outside the main loop so that they can be accessed for the
6552 back off. */
6553
6554 zeroreqchar = reqchar;
6555 zerofirstchar = firstchar;
6556 groupsetfirstchar = FALSE;
6557
6558 if (bravalue >= OP_ONCE)
6559 {
6560 /* If we have not yet set a firstchar in this branch, take it from the
6561 subpattern, remembering that it was set here so that a repeat of more
6562 than one can replicate it as reqchar if necessary. If the subpattern has
6563 no firstchar, set "none" for the whole branch. In both cases, a zero
6564 repeat forces firstchar to "none". */
6565
6566 if (firstchar == REQ_UNSET)
6567 {
6568 if (subfirstchar >= 0)
6569 {
6570 firstchar = subfirstchar;
6571 groupsetfirstchar = TRUE;
6572 }
6573 else firstchar = REQ_NONE;
6574 zerofirstchar = REQ_NONE;
6575 }
6576
6577 /* If firstchar was previously set, convert the subpattern's firstchar
6578 into reqchar if there wasn't one, using the vary flag that was in
6579 existence beforehand. */
6580
6581 else if (subfirstchar >= 0 && subreqchar < 0)
6582 subreqchar = subfirstchar | tempreqvary;
6583
6584 /* If the subpattern set a required byte (or set a first byte that isn't
6585 really the first byte - see above), set it. */
6586
6587 if (subreqchar >= 0) reqchar = subreqchar;
6588 }
6589
6590 /* For a forward assertion, we take the reqchar, if set. This can be
6591 helpful if the pattern that follows the assertion doesn't set a different
6592 char. For example, it's useful for /(?=abcde).+/. We can't set firstchar
6593 for an assertion, however because it leads to incorrect effect for patterns
6594 such as /(?=a)a.+/ when the "real" "a" would then become a reqchar instead
6595 of a firstchar. This is overcome by a scan at the end if there's no
6596 firstchar, looking for an asserted first char. */
6597
6598 else if (bravalue == OP_ASSERT && subreqchar >= 0) reqchar = subreqchar;
6599 break; /* End of processing '(' */
6600
6601
6602 /* ===================================================================*/
6603 /* Handle metasequences introduced by \. For ones like \d, the ESC_ values
6604 are arranged to be the negation of the corresponding OP_values in the
6605 default case when PCRE_UCP is not set. For the back references, the values
6606 are ESC_REF plus the reference number. Only back references and those types
6607 that consume a character may be repeated. We can test for values between
6608 ESC_b and ESC_Z for the latter; this may have to change if any new ones are
6609 ever created. */
6610
6611 case CHAR_BACKSLASH:
6612 tempptr = ptr;
6613 c = check_escape(&ptr, errorcodeptr, cd->bracount, options, FALSE);
6614 if (*errorcodeptr != 0) goto FAILED;
6615
6616 if (c < 0)
6617 {
6618 if (-c == ESC_Q) /* Handle start of quoted string */
6619 {
6620 if (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
6621 ptr += 2; /* avoid empty string */
6622 else inescq = TRUE;
6623 continue;
6624 }
6625
6626 if (-c == ESC_E) continue; /* Perl ignores an orphan \E */
6627
6628 /* For metasequences that actually match a character, we disable the
6629 setting of a first character if it hasn't already been set. */
6630
6631 if (firstchar == REQ_UNSET && -c > ESC_b && -c < ESC_Z)
6632