/[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 806 - (show annotations)
Thu Dec 15 11:57:39 2011 UTC (7 years, 9 months ago) by zherczeg
File MIME type: text/plain
File size: 266427 byte(s)
Error occurred while calculating annotation data.
lcc and inline printint.c 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) - 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 switch (op)
1713 {
1714 /* We only need to continue for OP_CBRA (normal capturing bracket) and
1715 OP_BRA (normal non-capturing bracket) because the other variants of these
1716 opcodes are all concerned with unlimited repeated groups, which of course
1717 are not of fixed length. */
1718
1719 case OP_CBRA:
1720 case OP_BRA:
1721 case OP_ONCE:
1722 case OP_ONCE_NC:
1723 case OP_COND:
1724 d = find_fixedlength(cc + ((op == OP_CBRA)? IMM2_SIZE : 0), utf, atend, cd);
1725 if (d < 0) return d;
1726 branchlength += d;
1727 do cc += GET(cc, 1); while (*cc == OP_ALT);
1728 cc += 1 + LINK_SIZE;
1729 break;
1730
1731 /* Reached end of a branch; if it's a ket it is the end of a nested call.
1732 If it's ALT it is an alternation in a nested call. An ACCEPT is effectively
1733 an ALT. If it is END it's the end of the outer call. All can be handled by
1734 the same code. Note that we must not include the OP_KETRxxx opcodes here,
1735 because they all imply an unlimited repeat. */
1736
1737 case OP_ALT:
1738 case OP_KET:
1739 case OP_END:
1740 case OP_ACCEPT:
1741 case OP_ASSERT_ACCEPT:
1742 if (length < 0) length = branchlength;
1743 else if (length != branchlength) return -1;
1744 if (*cc != OP_ALT) return length;
1745 cc += 1 + LINK_SIZE;
1746 branchlength = 0;
1747 break;
1748
1749 /* A true recursion implies not fixed length, but a subroutine call may
1750 be OK. If the subroutine is a forward reference, we can't deal with
1751 it until the end of the pattern, so return -3. */
1752
1753 case OP_RECURSE:
1754 if (!atend) return -3;
1755 cs = ce = (pcre_uchar *)cd->start_code + GET(cc, 1); /* Start subpattern */
1756 do ce += GET(ce, 1); while (*ce == OP_ALT); /* End subpattern */
1757 if (cc > cs && cc < ce) return -1; /* Recursion */
1758 d = find_fixedlength(cs + 2, utf, atend, cd);
1759 if (d < 0) return d;
1760 branchlength += d;
1761 cc += 1 + LINK_SIZE;
1762 break;
1763
1764 /* Skip over assertive subpatterns */
1765
1766 case OP_ASSERT:
1767 case OP_ASSERT_NOT:
1768 case OP_ASSERTBACK:
1769 case OP_ASSERTBACK_NOT:
1770 do cc += GET(cc, 1); while (*cc == OP_ALT);
1771 /* Fall through */
1772
1773 /* Skip over things that don't match chars */
1774
1775 case OP_MARK:
1776 case OP_PRUNE_ARG:
1777 case OP_SKIP_ARG:
1778 case OP_THEN_ARG:
1779 cc += cc[1] + PRIV(OP_lengths)[*cc];
1780 break;
1781
1782 case OP_CALLOUT:
1783 case OP_CIRC:
1784 case OP_CIRCM:
1785 case OP_CLOSE:
1786 case OP_COMMIT:
1787 case OP_CREF:
1788 case OP_DEF:
1789 case OP_DOLL:
1790 case OP_DOLLM:
1791 case OP_EOD:
1792 case OP_EODN:
1793 case OP_FAIL:
1794 case OP_NCREF:
1795 case OP_NRREF:
1796 case OP_NOT_WORD_BOUNDARY:
1797 case OP_PRUNE:
1798 case OP_REVERSE:
1799 case OP_RREF:
1800 case OP_SET_SOM:
1801 case OP_SKIP:
1802 case OP_SOD:
1803 case OP_SOM:
1804 case OP_THEN:
1805 case OP_WORD_BOUNDARY:
1806 cc += PRIV(OP_lengths)[*cc];
1807 break;
1808
1809 /* Handle literal characters */
1810
1811 case OP_CHAR:
1812 case OP_CHARI:
1813 case OP_NOT:
1814 case OP_NOTI:
1815 branchlength++;
1816 cc += 2;
1817 #ifdef SUPPORT_UTF
1818 if (utf && HAS_EXTRALEN(cc[-1])) cc += GET_EXTRALEN(cc[-1]);
1819 #endif
1820 break;
1821
1822 /* Handle exact repetitions. The count is already in characters, but we
1823 need to skip over a multibyte character in UTF8 mode. */
1824
1825 case OP_EXACT:
1826 case OP_EXACTI:
1827 case OP_NOTEXACT:
1828 case OP_NOTEXACTI:
1829 branchlength += GET2(cc,1);
1830 cc += 2 + IMM2_SIZE;
1831 #ifdef SUPPORT_UTF
1832 if (utf && HAS_EXTRALEN(cc[-1])) cc += GET_EXTRALEN(cc[-1]);
1833 #endif
1834 break;
1835
1836 case OP_TYPEEXACT:
1837 branchlength += GET2(cc,1);
1838 if (cc[1 + IMM2_SIZE] == OP_PROP || cc[1 + IMM2_SIZE] == OP_NOTPROP) cc += 2;
1839 cc += 1 + IMM2_SIZE + 1;
1840 break;
1841
1842 /* Handle single-char matchers */
1843
1844 case OP_PROP:
1845 case OP_NOTPROP:
1846 cc += 2;
1847 /* Fall through */
1848
1849 case OP_HSPACE:
1850 case OP_VSPACE:
1851 case OP_NOT_HSPACE:
1852 case OP_NOT_VSPACE:
1853 case OP_NOT_DIGIT:
1854 case OP_DIGIT:
1855 case OP_NOT_WHITESPACE:
1856 case OP_WHITESPACE:
1857 case OP_NOT_WORDCHAR:
1858 case OP_WORDCHAR:
1859 case OP_ANY:
1860 case OP_ALLANY:
1861 branchlength++;
1862 cc++;
1863 break;
1864
1865 /* The single-byte matcher isn't allowed. This only happens in UTF-8 mode;
1866 otherwise \C is coded as OP_ALLANY. */
1867
1868 case OP_ANYBYTE:
1869 return -2;
1870
1871 /* Check a class for variable quantification */
1872
1873 #if defined SUPPORT_UTF || defined COMPILE_PCRE16
1874 case OP_XCLASS:
1875 cc += GET(cc, 1) - PRIV(OP_lengths)[OP_CLASS];
1876 /* Fall through */
1877 #endif
1878
1879 case OP_CLASS:
1880 case OP_NCLASS:
1881 cc += PRIV(OP_lengths)[OP_CLASS];
1882
1883 switch (*cc)
1884 {
1885 case OP_CRPLUS:
1886 case OP_CRMINPLUS:
1887 case OP_CRSTAR:
1888 case OP_CRMINSTAR:
1889 case OP_CRQUERY:
1890 case OP_CRMINQUERY:
1891 return -1;
1892
1893 case OP_CRRANGE:
1894 case OP_CRMINRANGE:
1895 if (GET2(cc,1) != GET2(cc,1+IMM2_SIZE)) return -1;
1896 branchlength += GET2(cc,1);
1897 cc += 1 + 2 * IMM2_SIZE;
1898 break;
1899
1900 default:
1901 branchlength++;
1902 }
1903 break;
1904
1905 /* Anything else is variable length */
1906
1907 case OP_ANYNL:
1908 case OP_BRAMINZERO:
1909 case OP_BRAPOS:
1910 case OP_BRAPOSZERO:
1911 case OP_BRAZERO:
1912 case OP_CBRAPOS:
1913 case OP_EXTUNI:
1914 case OP_KETRMAX:
1915 case OP_KETRMIN:
1916 case OP_KETRPOS:
1917 case OP_MINPLUS:
1918 case OP_MINPLUSI:
1919 case OP_MINQUERY:
1920 case OP_MINQUERYI:
1921 case OP_MINSTAR:
1922 case OP_MINSTARI:
1923 case OP_MINUPTO:
1924 case OP_MINUPTOI:
1925 case OP_NOTMINPLUS:
1926 case OP_NOTMINPLUSI:
1927 case OP_NOTMINQUERY:
1928 case OP_NOTMINQUERYI:
1929 case OP_NOTMINSTAR:
1930 case OP_NOTMINSTARI:
1931 case OP_NOTMINUPTO:
1932 case OP_NOTMINUPTOI:
1933 case OP_NOTPLUS:
1934 case OP_NOTPLUSI:
1935 case OP_NOTPOSPLUS:
1936 case OP_NOTPOSPLUSI:
1937 case OP_NOTPOSQUERY:
1938 case OP_NOTPOSQUERYI:
1939 case OP_NOTPOSSTAR:
1940 case OP_NOTPOSSTARI:
1941 case OP_NOTPOSUPTO:
1942 case OP_NOTPOSUPTOI:
1943 case OP_NOTQUERY:
1944 case OP_NOTQUERYI:
1945 case OP_NOTSTAR:
1946 case OP_NOTSTARI:
1947 case OP_NOTUPTO:
1948 case OP_NOTUPTOI:
1949 case OP_PLUS:
1950 case OP_PLUSI:
1951 case OP_POSPLUS:
1952 case OP_POSPLUSI:
1953 case OP_POSQUERY:
1954 case OP_POSQUERYI:
1955 case OP_POSSTAR:
1956 case OP_POSSTARI:
1957 case OP_POSUPTO:
1958 case OP_POSUPTOI:
1959 case OP_QUERY:
1960 case OP_QUERYI:
1961 case OP_REF:
1962 case OP_REFI:
1963 case OP_SBRA:
1964 case OP_SBRAPOS:
1965 case OP_SCBRA:
1966 case OP_SCBRAPOS:
1967 case OP_SCOND:
1968 case OP_SKIPZERO:
1969 case OP_STAR:
1970 case OP_STARI:
1971 case OP_TYPEMINPLUS:
1972 case OP_TYPEMINQUERY:
1973 case OP_TYPEMINSTAR:
1974 case OP_TYPEMINUPTO:
1975 case OP_TYPEPLUS:
1976 case OP_TYPEPOSPLUS:
1977 case OP_TYPEPOSQUERY:
1978 case OP_TYPEPOSSTAR:
1979 case OP_TYPEPOSUPTO:
1980 case OP_TYPEQUERY:
1981 case OP_TYPESTAR:
1982 case OP_TYPEUPTO:
1983 case OP_UPTO:
1984 case OP_UPTOI:
1985 return -1;
1986
1987 /* Catch unrecognized opcodes so that when new ones are added they
1988 are not forgotten, as has happened in the past. */
1989
1990 default:
1991 return -4;
1992 }
1993 }
1994 /* Control never gets here */
1995 }
1996
1997
1998
1999
2000 /*************************************************
2001 * Scan compiled regex for specific bracket *
2002 *************************************************/
2003
2004 /* This little function scans through a compiled pattern until it finds a
2005 capturing bracket with the given number, or, if the number is negative, an
2006 instance of OP_REVERSE for a lookbehind. The function is global in the C sense
2007 so that it can be called from pcre_study() when finding the minimum matching
2008 length.
2009
2010 Arguments:
2011 code points to start of expression
2012 utf TRUE in UTF-8 / UTF-16 mode
2013 number the required bracket number or negative to find a lookbehind
2014
2015 Returns: pointer to the opcode for the bracket, or NULL if not found
2016 */
2017
2018 const pcre_uchar *
2019 PRIV(find_bracket)(const pcre_uchar *code, BOOL utf, int number)
2020 {
2021 for (;;)
2022 {
2023 register int c = *code;
2024
2025 if (c == OP_END) return NULL;
2026
2027 /* XCLASS is used for classes that cannot be represented just by a bit
2028 map. This includes negated single high-valued characters. The length in
2029 the table is zero; the actual length is stored in the compiled code. */
2030
2031 if (c == OP_XCLASS) code += GET(code, 1);
2032
2033 /* Handle recursion */
2034
2035 else if (c == OP_REVERSE)
2036 {
2037 if (number < 0) return (pcre_uchar *)code;
2038 code += PRIV(OP_lengths)[c];
2039 }
2040
2041 /* Handle capturing bracket */
2042
2043 else if (c == OP_CBRA || c == OP_SCBRA ||
2044 c == OP_CBRAPOS || c == OP_SCBRAPOS)
2045 {
2046 int n = GET2(code, 1+LINK_SIZE);
2047 if (n == number) return (pcre_uchar *)code;
2048 code += PRIV(OP_lengths)[c];
2049 }
2050
2051 /* Otherwise, we can get the item's length from the table, except that for
2052 repeated character types, we have to test for \p and \P, which have an extra
2053 two bytes of parameters, and for MARK/PRUNE/SKIP/THEN with an argument, we
2054 must add in its length. */
2055
2056 else
2057 {
2058 switch(c)
2059 {
2060 case OP_TYPESTAR:
2061 case OP_TYPEMINSTAR:
2062 case OP_TYPEPLUS:
2063 case OP_TYPEMINPLUS:
2064 case OP_TYPEQUERY:
2065 case OP_TYPEMINQUERY:
2066 case OP_TYPEPOSSTAR:
2067 case OP_TYPEPOSPLUS:
2068 case OP_TYPEPOSQUERY:
2069 if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
2070 break;
2071
2072 case OP_TYPEUPTO:
2073 case OP_TYPEMINUPTO:
2074 case OP_TYPEEXACT:
2075 case OP_TYPEPOSUPTO:
2076 if (code[1 + IMM2_SIZE] == OP_PROP
2077 || code[1 + IMM2_SIZE] == OP_NOTPROP) code += 2;
2078 break;
2079
2080 case OP_MARK:
2081 case OP_PRUNE_ARG:
2082 case OP_SKIP_ARG:
2083 code += code[1];
2084 break;
2085
2086 case OP_THEN_ARG:
2087 code += code[1];
2088 break;
2089 }
2090
2091 /* Add in the fixed length from the table */
2092
2093 code += PRIV(OP_lengths)[c];
2094
2095 /* In UTF-8 mode, opcodes that are followed by a character may be followed by
2096 a multi-byte character. The length in the table is a minimum, so we have to
2097 arrange to skip the extra bytes. */
2098
2099 #ifdef SUPPORT_UTF
2100 if (utf) switch(c)
2101 {
2102 case OP_CHAR:
2103 case OP_CHARI:
2104 case OP_EXACT:
2105 case OP_EXACTI:
2106 case OP_UPTO:
2107 case OP_UPTOI:
2108 case OP_MINUPTO:
2109 case OP_MINUPTOI:
2110 case OP_POSUPTO:
2111 case OP_POSUPTOI:
2112 case OP_STAR:
2113 case OP_STARI:
2114 case OP_MINSTAR:
2115 case OP_MINSTARI:
2116 case OP_POSSTAR:
2117 case OP_POSSTARI:
2118 case OP_PLUS:
2119 case OP_PLUSI:
2120 case OP_MINPLUS:
2121 case OP_MINPLUSI:
2122 case OP_POSPLUS:
2123 case OP_POSPLUSI:
2124 case OP_QUERY:
2125 case OP_QUERYI:
2126 case OP_MINQUERY:
2127 case OP_MINQUERYI:
2128 case OP_POSQUERY:
2129 case OP_POSQUERYI:
2130 if (HAS_EXTRALEN(code[-1])) code += GET_EXTRALEN(code[-1]);
2131 break;
2132 }
2133 #else
2134 (void)(utf); /* Keep compiler happy by referencing function argument */
2135 #endif
2136 }
2137 }
2138 }
2139
2140
2141
2142 /*************************************************
2143 * Scan compiled regex for recursion reference *
2144 *************************************************/
2145
2146 /* This little function scans through a compiled pattern until it finds an
2147 instance of OP_RECURSE.
2148
2149 Arguments:
2150 code points to start of expression
2151 utf TRUE in UTF-8 / UTF-16 mode
2152
2153 Returns: pointer to the opcode for OP_RECURSE, or NULL if not found
2154 */
2155
2156 static const pcre_uchar *
2157 find_recurse(const pcre_uchar *code, BOOL utf)
2158 {
2159 for (;;)
2160 {
2161 register int c = *code;
2162 if (c == OP_END) return NULL;
2163 if (c == OP_RECURSE) return code;
2164
2165 /* XCLASS is used for classes that cannot be represented just by a bit
2166 map. This includes negated single high-valued characters. The length in
2167 the table is zero; the actual length is stored in the compiled code. */
2168
2169 if (c == OP_XCLASS) code += GET(code, 1);
2170
2171 /* Otherwise, we can get the item's length from the table, except that for
2172 repeated character types, we have to test for \p and \P, which have an extra
2173 two bytes of parameters, and for MARK/PRUNE/SKIP/THEN with an argument, we
2174 must add in its length. */
2175
2176 else
2177 {
2178 switch(c)
2179 {
2180 case OP_TYPESTAR:
2181 case OP_TYPEMINSTAR:
2182 case OP_TYPEPLUS:
2183 case OP_TYPEMINPLUS:
2184 case OP_TYPEQUERY:
2185 case OP_TYPEMINQUERY:
2186 case OP_TYPEPOSSTAR:
2187 case OP_TYPEPOSPLUS:
2188 case OP_TYPEPOSQUERY:
2189 if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
2190 break;
2191
2192 case OP_TYPEPOSUPTO:
2193 case OP_TYPEUPTO:
2194 case OP_TYPEMINUPTO:
2195 case OP_TYPEEXACT:
2196 if (code[1 + IMM2_SIZE] == OP_PROP
2197 || code[1 + IMM2_SIZE] == OP_NOTPROP) code += 2;
2198 break;
2199
2200 case OP_MARK:
2201 case OP_PRUNE_ARG:
2202 case OP_SKIP_ARG:
2203 code += code[1];
2204 break;
2205
2206 case OP_THEN_ARG:
2207 code += code[1];
2208 break;
2209 }
2210
2211 /* Add in the fixed length from the table */
2212
2213 code += PRIV(OP_lengths)[c];
2214
2215 /* In UTF-8 mode, opcodes that are followed by a character may be followed
2216 by a multi-byte character. The length in the table is a minimum, so we have
2217 to arrange to skip the extra bytes. */
2218
2219 #ifdef SUPPORT_UTF
2220 if (utf) switch(c)
2221 {
2222 case OP_CHAR:
2223 case OP_CHARI:
2224 case OP_EXACT:
2225 case OP_EXACTI:
2226 case OP_UPTO:
2227 case OP_UPTOI:
2228 case OP_MINUPTO:
2229 case OP_MINUPTOI:
2230 case OP_POSUPTO:
2231 case OP_POSUPTOI:
2232 case OP_STAR:
2233 case OP_STARI:
2234 case OP_MINSTAR:
2235 case OP_MINSTARI:
2236 case OP_POSSTAR:
2237 case OP_POSSTARI:
2238 case OP_PLUS:
2239 case OP_PLUSI:
2240 case OP_MINPLUS:
2241 case OP_MINPLUSI:
2242 case OP_POSPLUS:
2243 case OP_POSPLUSI:
2244 case OP_QUERY:
2245 case OP_QUERYI:
2246 case OP_MINQUERY:
2247 case OP_MINQUERYI:
2248 case OP_POSQUERY:
2249 case OP_POSQUERYI:
2250 if (HAS_EXTRALEN(code[-1])) code += GET_EXTRALEN(code[-1]);
2251 break;
2252 }
2253 #else
2254 (void)(utf); /* Keep compiler happy by referencing function argument */
2255 #endif
2256 }
2257 }
2258 }
2259
2260
2261
2262 /*************************************************
2263 * Scan compiled branch for non-emptiness *
2264 *************************************************/
2265
2266 /* This function scans through a branch of a compiled pattern to see whether it
2267 can match the empty string or not. It is called from could_be_empty()
2268 below and from compile_branch() when checking for an unlimited repeat of a
2269 group that can match nothing. Note that first_significant_code() skips over
2270 backward and negative forward assertions when its final argument is TRUE. If we
2271 hit an unclosed bracket, we return "empty" - this means we've struck an inner
2272 bracket whose current branch will already have been scanned.
2273
2274 Arguments:
2275 code points to start of search
2276 endcode points to where to stop
2277 utf TRUE if in UTF-8 / UTF-16 mode
2278 cd contains pointers to tables etc.
2279
2280 Returns: TRUE if what is matched could be empty
2281 */
2282
2283 static BOOL
2284 could_be_empty_branch(const pcre_uchar *code, const pcre_uchar *endcode,
2285 BOOL utf, compile_data *cd)
2286 {
2287 register int c;
2288 for (code = first_significant_code(code + PRIV(OP_lengths)[*code], TRUE);
2289 code < endcode;
2290 code = first_significant_code(code + PRIV(OP_lengths)[c], TRUE))
2291 {
2292 const pcre_uchar *ccode;
2293
2294 c = *code;
2295
2296 /* Skip over forward assertions; the other assertions are skipped by
2297 first_significant_code() with a TRUE final argument. */
2298
2299 if (c == OP_ASSERT)
2300 {
2301 do code += GET(code, 1); while (*code == OP_ALT);
2302 c = *code;
2303 continue;
2304 }
2305
2306 /* For a recursion/subroutine call, if its end has been reached, which
2307 implies a backward reference subroutine call, we can scan it. If it's a
2308 forward reference subroutine call, we can't. To detect forward reference
2309 we have to scan up the list that is kept in the workspace. This function is
2310 called only when doing the real compile, not during the pre-compile that
2311 measures the size of the compiled pattern. */
2312
2313 if (c == OP_RECURSE)
2314 {
2315 const pcre_uchar *scode;
2316 BOOL empty_branch;
2317
2318 /* Test for forward reference */
2319
2320 for (scode = cd->start_workspace; scode < cd->hwm; scode += LINK_SIZE)
2321 if (GET(scode, 0) == code + 1 - cd->start_code) return TRUE;
2322
2323 /* Not a forward reference, test for completed backward reference */
2324
2325 empty_branch = FALSE;
2326 scode = cd->start_code + GET(code, 1);
2327 if (GET(scode, 1) == 0) return TRUE; /* Unclosed */
2328
2329 /* Completed backwards reference */
2330
2331 do
2332 {
2333 if (could_be_empty_branch(scode, endcode, utf, cd))
2334 {
2335 empty_branch = TRUE;
2336 break;
2337 }
2338 scode += GET(scode, 1);
2339 }
2340 while (*scode == OP_ALT);
2341
2342 if (!empty_branch) return FALSE; /* All branches are non-empty */
2343 continue;
2344 }
2345
2346 /* Groups with zero repeats can of course be empty; skip them. */
2347
2348 if (c == OP_BRAZERO || c == OP_BRAMINZERO || c == OP_SKIPZERO ||
2349 c == OP_BRAPOSZERO)
2350 {
2351 code += PRIV(OP_lengths)[c];
2352 do code += GET(code, 1); while (*code == OP_ALT);
2353 c = *code;
2354 continue;
2355 }
2356
2357 /* A nested group that is already marked as "could be empty" can just be
2358 skipped. */
2359
2360 if (c == OP_SBRA || c == OP_SBRAPOS ||
2361 c == OP_SCBRA || c == OP_SCBRAPOS)
2362 {
2363 do code += GET(code, 1); while (*code == OP_ALT);
2364 c = *code;
2365 continue;
2366 }
2367
2368 /* For other groups, scan the branches. */
2369
2370 if (c == OP_BRA || c == OP_BRAPOS ||
2371 c == OP_CBRA || c == OP_CBRAPOS ||
2372 c == OP_ONCE || c == OP_ONCE_NC ||
2373 c == OP_COND)
2374 {
2375 BOOL empty_branch;
2376 if (GET(code, 1) == 0) return TRUE; /* Hit unclosed bracket */
2377
2378 /* If a conditional group has only one branch, there is a second, implied,
2379 empty branch, so just skip over the conditional, because it could be empty.
2380 Otherwise, scan the individual branches of the group. */
2381
2382 if (c == OP_COND && code[GET(code, 1)] != OP_ALT)
2383 code += GET(code, 1);
2384 else
2385 {
2386 empty_branch = FALSE;
2387 do
2388 {
2389 if (!empty_branch && could_be_empty_branch(code, endcode, utf, cd))
2390 empty_branch = TRUE;
2391 code += GET(code, 1);
2392 }
2393 while (*code == OP_ALT);
2394 if (!empty_branch) return FALSE; /* All branches are non-empty */
2395 }
2396
2397 c = *code;
2398 continue;
2399 }
2400
2401 /* Handle the other opcodes */
2402
2403 switch (c)
2404 {
2405 /* Check for quantifiers after a class. XCLASS is used for classes that
2406 cannot be represented just by a bit map. This includes negated single
2407 high-valued characters. The length in PRIV(OP_lengths)[] is zero; the
2408 actual length is stored in the compiled code, so we must update "code"
2409 here. */
2410
2411 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
2412 case OP_XCLASS:
2413 ccode = code += GET(code, 1);
2414 goto CHECK_CLASS_REPEAT;
2415 #endif
2416
2417 case OP_CLASS:
2418 case OP_NCLASS:
2419 ccode = code + PRIV(OP_lengths)[OP_CLASS];
2420
2421 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
2422 CHECK_CLASS_REPEAT:
2423 #endif
2424
2425 switch (*ccode)
2426 {
2427 case OP_CRSTAR: /* These could be empty; continue */
2428 case OP_CRMINSTAR:
2429 case OP_CRQUERY:
2430 case OP_CRMINQUERY:
2431 break;
2432
2433 default: /* Non-repeat => class must match */
2434 case OP_CRPLUS: /* These repeats aren't empty */
2435 case OP_CRMINPLUS:
2436 return FALSE;
2437
2438 case OP_CRRANGE:
2439 case OP_CRMINRANGE:
2440 if (GET2(ccode, 1) > 0) return FALSE; /* Minimum > 0 */
2441 break;
2442 }
2443 break;
2444
2445 /* Opcodes that must match a character */
2446
2447 case OP_PROP:
2448 case OP_NOTPROP:
2449 case OP_EXTUNI:
2450 case OP_NOT_DIGIT:
2451 case OP_DIGIT:
2452 case OP_NOT_WHITESPACE:
2453 case OP_WHITESPACE:
2454 case OP_NOT_WORDCHAR:
2455 case OP_WORDCHAR:
2456 case OP_ANY:
2457 case OP_ALLANY:
2458 case OP_ANYBYTE:
2459 case OP_CHAR:
2460 case OP_CHARI:
2461 case OP_NOT:
2462 case OP_NOTI:
2463 case OP_PLUS:
2464 case OP_MINPLUS:
2465 case OP_POSPLUS:
2466 case OP_EXACT:
2467 case OP_NOTPLUS:
2468 case OP_NOTMINPLUS:
2469 case OP_NOTPOSPLUS:
2470 case OP_NOTEXACT:
2471 case OP_TYPEPLUS:
2472 case OP_TYPEMINPLUS:
2473 case OP_TYPEPOSPLUS:
2474 case OP_TYPEEXACT:
2475 return FALSE;
2476
2477 /* These are going to continue, as they may be empty, but we have to
2478 fudge the length for the \p and \P cases. */
2479
2480 case OP_TYPESTAR:
2481 case OP_TYPEMINSTAR:
2482 case OP_TYPEPOSSTAR:
2483 case OP_TYPEQUERY:
2484 case OP_TYPEMINQUERY:
2485 case OP_TYPEPOSQUERY:
2486 if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
2487 break;
2488
2489 /* Same for these */
2490
2491 case OP_TYPEUPTO:
2492 case OP_TYPEMINUPTO:
2493 case OP_TYPEPOSUPTO:
2494 if (code[1 + IMM2_SIZE] == OP_PROP
2495 || code[1 + IMM2_SIZE] == OP_NOTPROP) code += 2;
2496 break;
2497
2498 /* End of branch */
2499
2500 case OP_KET:
2501 case OP_KETRMAX:
2502 case OP_KETRMIN:
2503 case OP_KETRPOS:
2504 case OP_ALT:
2505 return TRUE;
2506
2507 /* In UTF-8 mode, STAR, MINSTAR, POSSTAR, QUERY, MINQUERY, POSQUERY, UPTO,
2508 MINUPTO, and POSUPTO may be followed by a multibyte character */
2509
2510 #ifdef SUPPORT_UTF
2511 case OP_STAR:
2512 case OP_STARI:
2513 case OP_MINSTAR:
2514 case OP_MINSTARI:
2515 case OP_POSSTAR:
2516 case OP_POSSTARI:
2517 case OP_QUERY:
2518 case OP_QUERYI:
2519 case OP_MINQUERY:
2520 case OP_MINQUERYI:
2521 case OP_POSQUERY:
2522 case OP_POSQUERYI:
2523 if (utf && HAS_EXTRALEN(code[1])) code += GET_EXTRALEN(code[1]);
2524 break;
2525
2526 case OP_UPTO:
2527 case OP_UPTOI:
2528 case OP_MINUPTO:
2529 case OP_MINUPTOI:
2530 case OP_POSUPTO:
2531 case OP_POSUPTOI:
2532 if (utf && HAS_EXTRALEN(code[1 + IMM2_SIZE])) code += GET_EXTRALEN(code[1 + IMM2_SIZE]);
2533 break;
2534 #endif
2535
2536 /* MARK, and PRUNE/SKIP/THEN with an argument must skip over the argument
2537 string. */
2538
2539 case OP_MARK:
2540 case OP_PRUNE_ARG:
2541 case OP_SKIP_ARG:
2542 code += code[1];
2543 break;
2544
2545 case OP_THEN_ARG:
2546 code += code[1];
2547 break;
2548
2549 /* None of the remaining opcodes are required to match a character. */
2550
2551 default:
2552 break;
2553 }
2554 }
2555
2556 return TRUE;
2557 }
2558
2559
2560
2561 /*************************************************
2562 * Scan compiled regex for non-emptiness *
2563 *************************************************/
2564
2565 /* This function is called to check for left recursive calls. We want to check
2566 the current branch of the current pattern to see if it could match the empty
2567 string. If it could, we must look outwards for branches at other levels,
2568 stopping when we pass beyond the bracket which is the subject of the recursion.
2569 This function is called only during the real compile, not during the
2570 pre-compile.
2571
2572 Arguments:
2573 code points to start of the recursion
2574 endcode points to where to stop (current RECURSE item)
2575 bcptr points to the chain of current (unclosed) branch starts
2576 utf TRUE if in UTF-8 / UTF-16 mode
2577 cd pointers to tables etc
2578
2579 Returns: TRUE if what is matched could be empty
2580 */
2581
2582 static BOOL
2583 could_be_empty(const pcre_uchar *code, const pcre_uchar *endcode,
2584 branch_chain *bcptr, BOOL utf, compile_data *cd)
2585 {
2586 while (bcptr != NULL && bcptr->current_branch >= code)
2587 {
2588 if (!could_be_empty_branch(bcptr->current_branch, endcode, utf, cd))
2589 return FALSE;
2590 bcptr = bcptr->outer;
2591 }
2592 return TRUE;
2593 }
2594
2595
2596
2597 /*************************************************
2598 * Check for POSIX class syntax *
2599 *************************************************/
2600
2601 /* This function is called when the sequence "[:" or "[." or "[=" is
2602 encountered in a character class. It checks whether this is followed by a
2603 sequence of characters terminated by a matching ":]" or ".]" or "=]". If we
2604 reach an unescaped ']' without the special preceding character, return FALSE.
2605
2606 Originally, this function only recognized a sequence of letters between the
2607 terminators, but it seems that Perl recognizes any sequence of characters,
2608 though of course unknown POSIX names are subsequently rejected. Perl gives an
2609 "Unknown POSIX class" error for [:f\oo:] for example, where previously PCRE
2610 didn't consider this to be a POSIX class. Likewise for [:1234:].
2611
2612 The problem in trying to be exactly like Perl is in the handling of escapes. We
2613 have to be sure that [abc[:x\]pqr] is *not* treated as containing a POSIX
2614 class, but [abc[:x\]pqr:]] is (so that an error can be generated). The code
2615 below handles the special case of \], but does not try to do any other escape
2616 processing. This makes it different from Perl for cases such as [:l\ower:]
2617 where Perl recognizes it as the POSIX class "lower" but PCRE does not recognize
2618 "l\ower". This is a lesser evil that not diagnosing bad classes when Perl does,
2619 I think.
2620
2621 A user pointed out that PCRE was rejecting [:a[:digit:]] whereas Perl was not.
2622 It seems that the appearance of a nested POSIX class supersedes an apparent
2623 external class. For example, [:a[:digit:]b:] matches "a", "b", ":", or
2624 a digit.
2625
2626 In Perl, unescaped square brackets may also appear as part of class names. For
2627 example, [:a[:abc]b:] gives unknown POSIX class "[:abc]b:]". However, for
2628 [:a[:abc]b][b:] it gives unknown POSIX class "[:abc]b][b:]", which does not
2629 seem right at all. PCRE does not allow closing square brackets in POSIX class
2630 names.
2631
2632 Arguments:
2633 ptr pointer to the initial [
2634 endptr where to return the end pointer
2635
2636 Returns: TRUE or FALSE
2637 */
2638
2639 static BOOL
2640 check_posix_syntax(const pcre_uchar *ptr, const pcre_uchar **endptr)
2641 {
2642 int terminator; /* Don't combine these lines; the Solaris cc */
2643 terminator = *(++ptr); /* compiler warns about "non-constant" initializer. */
2644 for (++ptr; *ptr != 0; ptr++)
2645 {
2646 if (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET)
2647 ptr++;
2648 else if (*ptr == CHAR_RIGHT_SQUARE_BRACKET) return FALSE;
2649 else
2650 {
2651 if (*ptr == terminator && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET)
2652 {
2653 *endptr = ptr;
2654 return TRUE;
2655 }
2656 if (*ptr == CHAR_LEFT_SQUARE_BRACKET &&
2657 (ptr[1] == CHAR_COLON || ptr[1] == CHAR_DOT ||
2658 ptr[1] == CHAR_EQUALS_SIGN) &&
2659 check_posix_syntax(ptr, endptr))
2660 return FALSE;
2661 }
2662 }
2663 return FALSE;
2664 }
2665
2666
2667
2668
2669 /*************************************************
2670 * Check POSIX class name *
2671 *************************************************/
2672
2673 /* This function is called to check the name given in a POSIX-style class entry
2674 such as [:alnum:].
2675
2676 Arguments:
2677 ptr points to the first letter
2678 len the length of the name
2679
2680 Returns: a value representing the name, or -1 if unknown
2681 */
2682
2683 static int
2684 check_posix_name(const pcre_uchar *ptr, int len)
2685 {
2686 const char *pn = posix_names;
2687 register int yield = 0;
2688 while (posix_name_lengths[yield] != 0)
2689 {
2690 if (len == posix_name_lengths[yield] &&
2691 STRNCMP_UC_C8(ptr, pn, len) == 0) return yield;
2692 pn += posix_name_lengths[yield] + 1;
2693 yield++;
2694 }
2695 return -1;
2696 }
2697
2698
2699 /*************************************************
2700 * Adjust OP_RECURSE items in repeated group *
2701 *************************************************/
2702
2703 /* OP_RECURSE items contain an offset from the start of the regex to the group
2704 that is referenced. This means that groups can be replicated for fixed
2705 repetition simply by copying (because the recursion is allowed to refer to
2706 earlier groups that are outside the current group). However, when a group is
2707 optional (i.e. the minimum quantifier is zero), OP_BRAZERO or OP_SKIPZERO is
2708 inserted before it, after it has been compiled. This means that any OP_RECURSE
2709 items within it that refer to the group itself or any contained groups have to
2710 have their offsets adjusted. That one of the jobs of this function. Before it
2711 is called, the partially compiled regex must be temporarily terminated with
2712 OP_END.
2713
2714 This function has been extended with the possibility of forward references for
2715 recursions and subroutine calls. It must also check the list of such references
2716 for the group we are dealing with. If it finds that one of the recursions in
2717 the current group is on this list, it adjusts the offset in the list, not the
2718 value in the reference (which is a group number).
2719
2720 Arguments:
2721 group points to the start of the group
2722 adjust the amount by which the group is to be moved
2723 utf TRUE in UTF-8 / UTF-16 mode
2724 cd contains pointers to tables etc.
2725 save_hwm the hwm forward reference pointer at the start of the group
2726
2727 Returns: nothing
2728 */
2729
2730 static void
2731 adjust_recurse(pcre_uchar *group, int adjust, BOOL utf, compile_data *cd,
2732 pcre_uchar *save_hwm)
2733 {
2734 pcre_uchar *ptr = group;
2735
2736 while ((ptr = (pcre_uchar *)find_recurse(ptr, utf)) != NULL)
2737 {
2738 int offset;
2739 pcre_uchar *hc;
2740
2741 /* See if this recursion is on the forward reference list. If so, adjust the
2742 reference. */
2743
2744 for (hc = save_hwm; hc < cd->hwm; hc += LINK_SIZE)
2745 {
2746 offset = GET(hc, 0);
2747 if (cd->start_code + offset == ptr + 1)
2748 {
2749 PUT(hc, 0, offset + adjust);
2750 break;
2751 }
2752 }
2753
2754 /* Otherwise, adjust the recursion offset if it's after the start of this
2755 group. */
2756
2757 if (hc >= cd->hwm)
2758 {
2759 offset = GET(ptr, 1);
2760 if (cd->start_code + offset >= group) PUT(ptr, 1, offset + adjust);
2761 }
2762
2763 ptr += 1 + LINK_SIZE;
2764 }
2765 }
2766
2767
2768
2769 /*************************************************
2770 * Insert an automatic callout point *
2771 *************************************************/
2772
2773 /* This function is called when the PCRE_AUTO_CALLOUT option is set, to insert
2774 callout points before each pattern item.
2775
2776 Arguments:
2777 code current code pointer
2778 ptr current pattern pointer
2779 cd pointers to tables etc
2780
2781 Returns: new code pointer
2782 */
2783
2784 static pcre_uchar *
2785 auto_callout(pcre_uchar *code, const pcre_uchar *ptr, compile_data *cd)
2786 {
2787 *code++ = OP_CALLOUT;
2788 *code++ = 255;
2789 PUT(code, 0, (int)(ptr - cd->start_pattern)); /* Pattern offset */
2790 PUT(code, LINK_SIZE, 0); /* Default length */
2791 return code + 2 * LINK_SIZE;
2792 }
2793
2794
2795
2796 /*************************************************
2797 * Complete a callout item *
2798 *************************************************/
2799
2800 /* A callout item contains the length of the next item in the pattern, which
2801 we can't fill in till after we have reached the relevant point. This is used
2802 for both automatic and manual callouts.
2803
2804 Arguments:
2805 previous_callout points to previous callout item
2806 ptr current pattern pointer
2807 cd pointers to tables etc
2808
2809 Returns: nothing
2810 */
2811
2812 static void
2813 complete_callout(pcre_uchar *previous_callout, const pcre_uchar *ptr, compile_data *cd)
2814 {
2815 int length = (int)(ptr - cd->start_pattern - GET(previous_callout, 2));
2816 PUT(previous_callout, 2 + LINK_SIZE, length);
2817 }
2818
2819
2820
2821 #ifdef SUPPORT_UCP
2822 /*************************************************
2823 * Get othercase range *
2824 *************************************************/
2825
2826 /* This function is passed the start and end of a class range, in UTF-8 mode
2827 with UCP support. It searches up the characters, looking for internal ranges of
2828 characters in the "other" case. Each call returns the next one, updating the
2829 start address.
2830
2831 Arguments:
2832 cptr points to starting character value; updated
2833 d end value
2834 ocptr where to put start of othercase range
2835 odptr where to put end of othercase range
2836
2837 Yield: TRUE when range returned; FALSE when no more
2838 */
2839
2840 static BOOL
2841 get_othercase_range(unsigned int *cptr, unsigned int d, unsigned int *ocptr,
2842 unsigned int *odptr)
2843 {
2844 unsigned int c, othercase, next;
2845
2846 for (c = *cptr; c <= d; c++)
2847 { if ((othercase = UCD_OTHERCASE(c)) != c) break; }
2848
2849 if (c > d) return FALSE;
2850
2851 *ocptr = othercase;
2852 next = othercase + 1;
2853
2854 for (++c; c <= d; c++)
2855 {
2856 if (UCD_OTHERCASE(c) != next) break;
2857 next++;
2858 }
2859
2860 *odptr = next - 1;
2861 *cptr = c;
2862
2863 return TRUE;
2864 }
2865
2866
2867
2868 /*************************************************
2869 * Check a character and a property *
2870 *************************************************/
2871
2872 /* This function is called by check_auto_possessive() when a property item
2873 is adjacent to a fixed character.
2874
2875 Arguments:
2876 c the character
2877 ptype the property type
2878 pdata the data for the type
2879 negated TRUE if it's a negated property (\P or \p{^)
2880
2881 Returns: TRUE if auto-possessifying is OK
2882 */
2883
2884 static BOOL
2885 check_char_prop(int c, int ptype, int pdata, BOOL negated)
2886 {
2887 const ucd_record *prop = GET_UCD(c);
2888 switch(ptype)
2889 {
2890 case PT_LAMP:
2891 return (prop->chartype == ucp_Lu ||
2892 prop->chartype == ucp_Ll ||
2893 prop->chartype == ucp_Lt) == negated;
2894
2895 case PT_GC:
2896 return (pdata == PRIV(ucp_gentype)[prop->chartype]) == negated;
2897
2898 case PT_PC:
2899 return (pdata == prop->chartype) == negated;
2900
2901 case PT_SC:
2902 return (pdata == prop->script) == negated;
2903
2904 /* These are specials */
2905
2906 case PT_ALNUM:
2907 return (PRIV(ucp_gentype)[prop->chartype] == ucp_L ||
2908 PRIV(ucp_gentype)[prop->chartype] == ucp_N) == negated;
2909
2910 case PT_SPACE: /* Perl space */
2911 return (PRIV(ucp_gentype)[prop->chartype] == ucp_Z ||
2912 c == CHAR_HT || c == CHAR_NL || c == CHAR_FF || c == CHAR_CR)
2913 == negated;
2914
2915 case PT_PXSPACE: /* POSIX space */
2916 return (PRIV(ucp_gentype)[prop->chartype] == ucp_Z ||
2917 c == CHAR_HT || c == CHAR_NL || c == CHAR_VT ||
2918 c == CHAR_FF || c == CHAR_CR)
2919 == negated;
2920
2921 case PT_WORD:
2922 return (PRIV(ucp_gentype)[prop->chartype] == ucp_L ||
2923 PRIV(ucp_gentype)[prop->chartype] == ucp_N ||
2924 c == CHAR_UNDERSCORE) == negated;
2925 }
2926 return FALSE;
2927 }
2928 #endif /* SUPPORT_UCP */
2929
2930
2931
2932 /*************************************************
2933 * Check if auto-possessifying is possible *
2934 *************************************************/
2935
2936 /* This function is called for unlimited repeats of certain items, to see
2937 whether the next thing could possibly match the repeated item. If not, it makes
2938 sense to automatically possessify the repeated item.
2939
2940 Arguments:
2941 previous pointer to the repeated opcode
2942 utf TRUE in UTF-8 / UTF-16 mode
2943 ptr next character in pattern
2944 options options bits
2945 cd contains pointers to tables etc.
2946
2947 Returns: TRUE if possessifying is wanted
2948 */
2949
2950 static BOOL
2951 check_auto_possessive(const pcre_uchar *previous, BOOL utf,
2952 const pcre_uchar *ptr, int options, compile_data *cd)
2953 {
2954 pcre_int32 c, next;
2955 int op_code = *previous++;
2956
2957 /* Skip whitespace and comments in extended mode */
2958
2959 if ((options & PCRE_EXTENDED) != 0)
2960 {
2961 for (;;)
2962 {
2963 while (MAX_255(*ptr) && (cd->ctypes[*ptr] & ctype_space) != 0) ptr++;
2964 if (*ptr == CHAR_NUMBER_SIGN)
2965 {
2966 ptr++;
2967 while (*ptr != 0)
2968 {
2969 if (IS_NEWLINE(ptr)) { ptr += cd->nllen; break; }
2970 ptr++;
2971 #ifdef SUPPORT_UTF
2972 if (utf) FORWARDCHAR(ptr);
2973 #endif
2974 }
2975 }
2976 else break;
2977 }
2978 }
2979
2980 /* If the next item is one that we can handle, get its value. A non-negative
2981 value is a character, a negative value is an escape value. */
2982
2983 if (*ptr == CHAR_BACKSLASH)
2984 {
2985 int temperrorcode = 0;
2986 next = check_escape(&ptr, &temperrorcode, cd->bracount, options, FALSE);
2987 if (temperrorcode != 0) return FALSE;
2988 ptr++; /* Point after the escape sequence */
2989 }
2990 else if (!MAX_255(*ptr) || (cd->ctypes[*ptr] & ctype_meta) == 0)
2991 {
2992 #ifdef SUPPORT_UTF
2993 if (utf) { GETCHARINC(next, ptr); } else
2994 #endif
2995 next = *ptr++;
2996 }
2997 else return FALSE;
2998
2999 /* Skip whitespace and comments in extended mode */
3000
3001 if ((options & PCRE_EXTENDED) != 0)
3002 {
3003 for (;;)
3004 {
3005 while (MAX_255(*ptr) && (cd->ctypes[*ptr] & ctype_space) != 0) ptr++;
3006 if (*ptr == CHAR_NUMBER_SIGN)
3007 {
3008 ptr++;
3009 while (*ptr != 0)
3010 {
3011 if (IS_NEWLINE(ptr)) { ptr += cd->nllen; break; }
3012 ptr++;
3013 #ifdef SUPPORT_UTF
3014 if (utf) FORWARDCHAR(ptr);
3015 #endif
3016 }
3017 }
3018 else break;
3019 }
3020 }
3021
3022 /* If the next thing is itself optional, we have to give up. */
3023
3024 if (*ptr == CHAR_ASTERISK || *ptr == CHAR_QUESTION_MARK ||
3025 STRNCMP_UC_C8(ptr, STR_LEFT_CURLY_BRACKET STR_0 STR_COMMA, 3) == 0)
3026 return FALSE;
3027
3028 /* Now compare the next item with the previous opcode. First, handle cases when
3029 the next item is a character. */
3030
3031 if (next >= 0) switch(op_code)
3032 {
3033 case OP_CHAR:
3034 #ifdef SUPPORT_UTF
3035 GETCHARTEST(c, previous);
3036 #else
3037 c = *previous;
3038 #endif
3039 return c != next;
3040
3041 /* For CHARI (caseless character) we must check the other case. If we have
3042 Unicode property support, we can use it to test the other case of
3043 high-valued characters. */
3044
3045 case OP_CHARI:
3046 #ifdef SUPPORT_UTF
3047 GETCHARTEST(c, previous);
3048 #else
3049 c = *previous;
3050 #endif
3051 if (c == next) return FALSE;
3052 #ifdef SUPPORT_UTF
3053 if (utf)
3054 {
3055 unsigned int othercase;
3056 if (next < 128) othercase = cd->fcc[next]; else
3057 #ifdef SUPPORT_UCP
3058 othercase = UCD_OTHERCASE((unsigned int)next);
3059 #else
3060 othercase = NOTACHAR;
3061 #endif
3062 return (unsigned int)c != othercase;
3063 }
3064 else
3065 #endif /* SUPPORT_UTF */
3066 return (c != TABLE_GET(next, cd->fcc, next)); /* Non-UTF-8 mode */
3067
3068 /* For OP_NOT and OP_NOTI, the data is always a single-byte character. These
3069 opcodes are not used for multi-byte characters, because they are coded using
3070 an XCLASS instead. */
3071
3072 case OP_NOT:
3073 return (c = *previous) == next;
3074
3075 case OP_NOTI:
3076 if ((c = *previous) == next) return TRUE;
3077 #ifdef SUPPORT_UTF
3078 if (utf)
3079 {
3080 unsigned int othercase;
3081 if (next < 128) othercase = cd->fcc[next]; else
3082 #ifdef SUPPORT_UCP
3083 othercase = UCD_OTHERCASE(next);
3084 #else
3085 othercase = NOTACHAR;
3086 #endif
3087 return (unsigned int)c == othercase;
3088 }
3089 else
3090 #endif /* SUPPORT_UTF */
3091 return (c == TABLE_GET(next, cd->fcc, next)); /* Non-UTF-8 mode */
3092
3093 /* Note that OP_DIGIT etc. are generated only when PCRE_UCP is *not* set.
3094 When it is set, \d etc. are converted into OP_(NOT_)PROP codes. */
3095
3096 case OP_DIGIT:
3097 return next > 127 || (cd->ctypes[next] & ctype_digit) == 0;
3098
3099 case OP_NOT_DIGIT:
3100 return next <= 127 && (cd->ctypes[next] & ctype_digit) != 0;
3101
3102 case OP_WHITESPACE:
3103 return next > 127 || (cd->ctypes[next] & ctype_space) == 0;
3104
3105 case OP_NOT_WHITESPACE:
3106 return next <= 127 && (cd->ctypes[next] & ctype_space) != 0;
3107
3108 case OP_WORDCHAR:
3109 return next > 127 || (cd->ctypes[next] & ctype_word) == 0;
3110
3111 case OP_NOT_WORDCHAR:
3112 return next <= 127 && (cd->ctypes[next] & ctype_word) != 0;
3113
3114 case OP_HSPACE:
3115 case OP_NOT_HSPACE:
3116 switch(next)
3117 {
3118 case 0x09:
3119 case 0x20:
3120 case 0xa0:
3121 case 0x1680:
3122 case 0x180e:
3123 case 0x2000:
3124 case 0x2001:
3125 case 0x2002:
3126 case 0x2003:
3127 case 0x2004:
3128 case 0x2005:
3129 case 0x2006:
3130 case 0x2007:
3131 case 0x2008:
3132 case 0x2009:
3133 case 0x200A:
3134 case 0x202f:
3135 case 0x205f:
3136 case 0x3000:
3137 return op_code == OP_NOT_HSPACE;
3138 default:
3139 return op_code != OP_NOT_HSPACE;
3140 }
3141
3142 case OP_ANYNL:
3143 case OP_VSPACE:
3144 case OP_NOT_VSPACE:
3145 switch(next)
3146 {
3147 case 0x0a:
3148 case 0x0b:
3149 case 0x0c:
3150 case 0x0d:
3151 case 0x85:
3152 case 0x2028:
3153 case 0x2029:
3154 return op_code == OP_NOT_VSPACE;
3155 default:
3156 return op_code != OP_NOT_VSPACE;
3157 }
3158
3159 #ifdef SUPPORT_UCP
3160 case OP_PROP:
3161 return check_char_prop(next, previous[0], previous[1], FALSE);
3162
3163 case OP_NOTPROP:
3164 return check_char_prop(next, previous[0], previous[1], TRUE);
3165 #endif
3166
3167 default:
3168 return FALSE;
3169 }
3170
3171
3172 /* Handle the case when the next item is \d, \s, etc. Note that when PCRE_UCP
3173 is set, \d turns into ESC_du rather than ESC_d, etc., so ESC_d etc. are
3174 generated only when PCRE_UCP is *not* set, that is, when only ASCII
3175 characteristics are recognized. Similarly, the opcodes OP_DIGIT etc. are
3176 replaced by OP_PROP codes when PCRE_UCP is set. */
3177
3178 switch(op_code)
3179 {
3180 case OP_CHAR:
3181 case OP_CHARI:
3182 #ifdef SUPPORT_UTF
3183 GETCHARTEST(c, previous);
3184 #else
3185 c = *previous;
3186 #endif
3187 switch(-next)
3188 {
3189 case ESC_d:
3190 return c > 127 || (cd->ctypes[c] & ctype_digit) == 0;
3191
3192 case ESC_D:
3193 return c <= 127 && (cd->ctypes[c] & ctype_digit) != 0;
3194
3195 case ESC_s:
3196 return c > 127 || (cd->ctypes[c] & ctype_space) == 0;
3197
3198 case ESC_S:
3199 return c <= 127 && (cd->ctypes[c] & ctype_space) != 0;
3200
3201 case ESC_w:
3202 return c > 127 || (cd->ctypes[c] & ctype_word) == 0;
3203
3204 case ESC_W:
3205 return c <= 127 && (cd->ctypes[c] & ctype_word) != 0;
3206
3207 case ESC_h:
3208 case ESC_H:
3209 switch(c)
3210 {
3211 case 0x09:
3212 case 0x20:
3213 case 0xa0:
3214 case 0x1680:
3215 case 0x180e:
3216 case 0x2000:
3217 case 0x2001:
3218 case 0x2002:
3219 case 0x2003:
3220 case 0x2004:
3221 case 0x2005:
3222 case 0x2006:
3223 case 0x2007:
3224 case 0x2008:
3225 case 0x2009:
3226 case 0x200A:
3227 case 0x202f:
3228 case 0x205f:
3229 case 0x3000:
3230 return -next != ESC_h;
3231 default:
3232 return -next == ESC_h;
3233 }
3234
3235 case ESC_v:
3236 case ESC_V:
3237 switch(c)
3238 {
3239 case 0x0a:
3240 case 0x0b:
3241 case 0x0c:
3242 case 0x0d:
3243 case 0x85:
3244 case 0x2028:
3245 case 0x2029:
3246 return -next != ESC_v;
3247 default:
3248 return -next == ESC_v;
3249 }
3250
3251 /* When PCRE_UCP is set, these values get generated for \d etc. Find
3252 their substitutions and process them. The result will always be either
3253 -ESC_p or -ESC_P. Then fall through to process those values. */
3254
3255 #ifdef SUPPORT_UCP
3256 case ESC_du:
3257 case ESC_DU:
3258 case ESC_wu:
3259 case ESC_WU:
3260 case ESC_su:
3261 case ESC_SU:
3262 {
3263 int temperrorcode = 0;
3264 ptr = substitutes[-next - ESC_DU];
3265 next = check_escape(&ptr, &temperrorcode, 0, options, FALSE);
3266 if (temperrorcode != 0) return FALSE;
3267 ptr++; /* For compatibility */
3268 }
3269 /* Fall through */
3270
3271 case ESC_p:
3272 case ESC_P:
3273 {
3274 int ptype, pdata, errorcodeptr;
3275 BOOL negated;
3276
3277 ptr--; /* Make ptr point at the p or P */
3278 ptype = get_ucp(&ptr, &negated, &pdata, &errorcodeptr);
3279 if (ptype < 0) return FALSE;
3280 ptr++; /* Point past the final curly ket */
3281
3282 /* If the property item is optional, we have to give up. (When generated
3283 from \d etc by PCRE_UCP, this test will have been applied much earlier,
3284 to the original \d etc. At this point, ptr will point to a zero byte. */
3285
3286 if (*ptr == CHAR_ASTERISK || *ptr == CHAR_QUESTION_MARK ||
3287 STRNCMP_UC_C8(ptr, STR_LEFT_CURLY_BRACKET STR_0 STR_COMMA, 3) == 0)
3288 return FALSE;
3289
3290 /* Do the property check. */
3291
3292 return check_char_prop(c, ptype, pdata, (next == -ESC_P) != negated);
3293 }
3294 #endif
3295
3296 default:
3297 return FALSE;
3298 }
3299
3300 /* In principle, support for Unicode properties should be integrated here as
3301 well. It means re-organizing the above code so as to get hold of the property
3302 values before switching on the op-code. However, I wonder how many patterns
3303 combine ASCII \d etc with Unicode properties? (Note that if PCRE_UCP is set,
3304 these op-codes are never generated.) */
3305
3306 case OP_DIGIT:
3307 return next == -ESC_D || next == -ESC_s || next == -ESC_W ||
3308 next == -ESC_h || next == -ESC_v || next == -ESC_R;
3309
3310 case OP_NOT_DIGIT:
3311 return next == -ESC_d;
3312
3313 case OP_WHITESPACE:
3314 return next == -ESC_S || next == -ESC_d || next == -ESC_w || next == -ESC_R;
3315
3316 case OP_NOT_WHITESPACE:
3317 return next == -ESC_s || next == -ESC_h || next == -ESC_v;
3318
3319 case OP_HSPACE:
3320 return next == -ESC_S || next == -ESC_H || next == -ESC_d ||
3321 next == -ESC_w || next == -ESC_v || next == -ESC_R;
3322
3323 case OP_NOT_HSPACE:
3324 return next == -ESC_h;
3325
3326 /* Can't have \S in here because VT matches \S (Perl anomaly) */
3327 case OP_ANYNL:
3328 case OP_VSPACE:
3329 return next == -ESC_V || next == -ESC_d || next == -ESC_w;
3330
3331 case OP_NOT_VSPACE:
3332 return next == -ESC_v || next == -ESC_R;
3333
3334 case OP_WORDCHAR:
3335 return next == -ESC_W || next == -ESC_s || next == -ESC_h ||
3336 next == -ESC_v || next == -ESC_R;
3337
3338 case OP_NOT_WORDCHAR:
3339 return next == -ESC_w || next == -ESC_d;
3340
3341 default:
3342 return FALSE;
3343 }
3344
3345 /* Control does not reach here */
3346 }
3347
3348
3349
3350 /*************************************************
3351 * Compile one branch *
3352 *************************************************/
3353
3354 /* Scan the pattern, compiling it into the a vector. If the options are
3355 changed during the branch, the pointer is used to change the external options
3356 bits. This function is used during the pre-compile phase when we are trying
3357 to find out the amount of memory needed, as well as during the real compile
3358 phase. The value of lengthptr distinguishes the two phases.
3359
3360 Arguments:
3361 optionsptr pointer to the option bits
3362 codeptr points to the pointer to the current code point
3363 ptrptr points to the current pattern pointer
3364 errorcodeptr points to error code variable
3365 firstcharptr set to initial literal character, or < 0 (REQ_UNSET, REQ_NONE)
3366 reqcharptr set to the last literal character required, else < 0
3367 bcptr points to current branch chain
3368 cond_depth conditional nesting depth
3369 cd contains pointers to tables etc.
3370 lengthptr NULL during the real compile phase
3371 points to length accumulator during pre-compile phase
3372
3373 Returns: TRUE on success
3374 FALSE, with *errorcodeptr set non-zero on error
3375 */
3376
3377 static BOOL
3378 compile_branch(int *optionsptr, pcre_uchar **codeptr,
3379 const pcre_uchar **ptrptr, int *errorcodeptr, pcre_int32 *firstcharptr,
3380 pcre_int32 *reqcharptr, branch_chain *bcptr, int cond_depth,
3381 compile_data *cd, int *lengthptr)
3382 {
3383 int repeat_type, op_type;
3384 int repeat_min = 0, repeat_max = 0; /* To please picky compilers */
3385 int bravalue = 0;
3386 int greedy_default, greedy_non_default;
3387 pcre_int32 firstchar, reqchar;
3388 pcre_int32 zeroreqchar, zerofirstchar;
3389 pcre_int32 req_caseopt, reqvary, tempreqvary;
3390 int options = *optionsptr; /* May change dynamically */
3391 int after_manual_callout = 0;
3392 int length_prevgroup = 0;
3393 register int c;
3394 register pcre_uchar *code = *codeptr;
3395 pcre_uchar *last_code = code;
3396 pcre_uchar *orig_code = code;
3397 pcre_uchar *tempcode;
3398 BOOL inescq = FALSE;
3399 BOOL groupsetfirstchar = FALSE;
3400 const pcre_uchar *ptr = *ptrptr;
3401 const pcre_uchar *tempptr;
3402 const pcre_uchar *nestptr = NULL;
3403 pcre_uchar *previous = NULL;
3404 pcre_uchar *previous_callout = NULL;
3405 pcre_uchar *save_hwm = NULL;
3406 pcre_uint8 classbits[32];
3407
3408 /* We can fish out the UTF-8 setting once and for all into a BOOL, but we
3409 must not do this for other options (e.g. PCRE_EXTENDED) because they may change
3410 dynamically as we process the pattern. */
3411
3412 #ifdef SUPPORT_UTF
3413 /* PCRE_UTF16 has the same value as PCRE_UTF8. */
3414 BOOL utf = (options & PCRE_UTF8) != 0;
3415 pcre_uchar utf_chars[6];
3416 #else
3417 BOOL utf = FALSE;
3418 #endif
3419
3420 /* Helper variables for OP_XCLASS opcode (for characters > 255). */
3421
3422 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
3423 BOOL xclass;
3424 pcre_uchar *class_uchardata;
3425 pcre_uchar *class_uchardata_base;
3426 #endif
3427
3428 #ifdef PCRE_DEBUG
3429 if (lengthptr != NULL) DPRINTF((">> start branch\n"));
3430 #endif
3431
3432 /* Set up the default and non-default settings for greediness */
3433
3434 greedy_default = ((options & PCRE_UNGREEDY) != 0);
3435 greedy_non_default = greedy_default ^ 1;
3436
3437 /* Initialize no first byte, no required byte. REQ_UNSET means "no char
3438 matching encountered yet". It gets changed to REQ_NONE if we hit something that
3439 matches a non-fixed char first char; reqchar just remains unset if we never
3440 find one.
3441
3442 When we hit a repeat whose minimum is zero, we may have to adjust these values
3443 to take the zero repeat into account. This is implemented by setting them to
3444 zerofirstbyte and zeroreqchar when such a repeat is encountered. The individual
3445 item types that can be repeated set these backoff variables appropriately. */
3446
3447 firstchar = reqchar = zerofirstchar = zeroreqchar = REQ_UNSET;
3448
3449 /* The variable req_caseopt contains either the REQ_CASELESS value
3450 or zero, according to the current setting of the caseless flag. The
3451 REQ_CASELESS leaves the lower 28 bit empty. It is added into the
3452 firstchar or reqchar variables to record the case status of the
3453 value. This is used only for ASCII characters. */
3454
3455 req_caseopt = ((options & PCRE_CASELESS) != 0)? REQ_CASELESS:0;
3456
3457 /* Switch on next character until the end of the branch */
3458
3459 for (;; ptr++)
3460 {
3461 BOOL negate_class;
3462 BOOL should_flip_negation;
3463 BOOL possessive_quantifier;
3464 BOOL is_quantifier;
3465 BOOL is_recurse;
3466 BOOL reset_bracount;
3467 int class_has_8bitchar;
3468 int class_single_char;
3469 int newoptions;
3470 int recno;
3471 int refsign;
3472 int skipbytes;
3473 int subreqchar;
3474 int subfirstchar;
3475 int terminator;
3476 int mclength;
3477 int tempbracount;
3478 pcre_uchar mcbuffer[8];
3479
3480 /* Get next character in the pattern */
3481
3482 c = *ptr;
3483
3484 /* If we are at the end of a nested substitution, revert to the outer level
3485 string. Nesting only happens one level deep. */
3486
3487 if (c == 0 && nestptr != NULL)
3488 {
3489 ptr = nestptr;
3490 nestptr = NULL;
3491 c = *ptr;
3492 }
3493
3494 /* If we are in the pre-compile phase, accumulate the length used for the
3495 previous cycle of this loop. */
3496
3497 if (lengthptr != NULL)
3498 {
3499 #ifdef PCRE_DEBUG
3500 if (code > cd->hwm) cd->hwm = code; /* High water info */
3501 #endif
3502 if (code > cd->start_workspace + cd->workspace_size -
3503 WORK_SIZE_SAFETY_MARGIN) /* Check for overrun */
3504 {
3505 *errorcodeptr = ERR52;
3506 goto FAILED;
3507 }
3508
3509 /* There is at least one situation where code goes backwards: this is the
3510 case of a zero quantifier after a class (e.g. [ab]{0}). At compile time,
3511 the class is simply eliminated. However, it is created first, so we have to
3512 allow memory for it. Therefore, don't ever reduce the length at this point.
3513 */
3514
3515 if (code < last_code) code = last_code;
3516
3517 /* Paranoid check for integer overflow */
3518
3519 if (OFLOW_MAX - *lengthptr < code - last_code)
3520 {
3521 *errorcodeptr = ERR20;
3522 goto FAILED;
3523 }
3524
3525 *lengthptr += (int)(code - last_code);
3526 DPRINTF(("length=%d added %d c=%c (0x%x)\n", *lengthptr,
3527 (int)(code - last_code), c, c));
3528
3529 /* If "previous" is set and it is not at the start of the work space, move
3530 it back to there, in order to avoid filling up the work space. Otherwise,
3531 if "previous" is NULL, reset the current code pointer to the start. */
3532
3533 if (previous != NULL)
3534 {
3535 if (previous > orig_code)
3536 {
3537 memmove(orig_code, previous, IN_UCHARS(code - previous));
3538 code -= previous - orig_code;
3539 previous = orig_code;
3540 }
3541 }
3542 else code = orig_code;
3543
3544 /* Remember where this code item starts so we can pick up the length
3545 next time round. */
3546
3547 last_code = code;
3548 }
3549
3550 /* In the real compile phase, just check the workspace used by the forward
3551 reference list. */
3552
3553 else if (cd->hwm > cd->start_workspace + cd->workspace_size -
3554 WORK_SIZE_SAFETY_MARGIN)
3555 {
3556 *errorcodeptr = ERR52;
3557 goto FAILED;
3558 }
3559
3560 /* If in \Q...\E, check for the end; if not, we have a literal */
3561
3562 if (inescq && c != 0)
3563 {
3564 if (c == CHAR_BACKSLASH && ptr[1] == CHAR_E)
3565 {
3566 inescq = FALSE;
3567 ptr++;
3568 continue;
3569 }
3570 else
3571 {
3572 if (previous_callout != NULL)
3573 {
3574 if (lengthptr == NULL) /* Don't attempt in pre-compile phase */
3575 complete_callout(previous_callout, ptr, cd);
3576 previous_callout = NULL;
3577 }
3578 if ((options & PCRE_AUTO_CALLOUT) != 0)
3579 {
3580 previous_callout = code;
3581 code = auto_callout(code, ptr, cd);
3582 }
3583 goto NORMAL_CHAR;
3584 }
3585 }
3586
3587 /* Fill in length of a previous callout, except when the next thing is
3588 a quantifier. */
3589
3590 is_quantifier =
3591 c == CHAR_ASTERISK || c == CHAR_PLUS || c == CHAR_QUESTION_MARK ||
3592 (c == CHAR_LEFT_CURLY_BRACKET && is_counted_repeat(ptr+1));
3593
3594 if (!is_quantifier && previous_callout != NULL &&
3595 after_manual_callout-- <= 0)
3596 {
3597 if (lengthptr == NULL) /* Don't attempt in pre-compile phase */
3598 complete_callout(previous_callout, ptr, cd);
3599 previous_callout = NULL;
3600 }
3601
3602 /* In extended mode, skip white space and comments. */
3603
3604 if ((options & PCRE_EXTENDED) != 0)
3605 {
3606 if (MAX_255(*ptr) && (cd->ctypes[c] & ctype_space) != 0) continue;
3607 if (c == CHAR_NUMBER_SIGN)
3608 {
3609 ptr++;
3610 while (*ptr != 0)
3611 {
3612 if (IS_NEWLINE(ptr)) { ptr += cd->nllen - 1; break; }
3613 ptr++;
3614 #ifdef SUPPORT_UTF
3615 if (utf) FORWARDCHAR(ptr);
3616 #endif
3617 }
3618 if (*ptr != 0) continue;
3619
3620 /* Else fall through to handle end of string */
3621 c = 0;
3622 }
3623 }
3624
3625 /* No auto callout for quantifiers. */
3626
3627 if ((options & PCRE_AUTO_CALLOUT) != 0 && !is_quantifier)
3628 {
3629 previous_callout = code;
3630 code = auto_callout(code, ptr, cd);
3631 }
3632
3633 switch(c)
3634 {
3635 /* ===================================================================*/
3636 case 0: /* The branch terminates at string end */
3637 case CHAR_VERTICAL_LINE: /* or | or ) */
3638 case CHAR_RIGHT_PARENTHESIS:
3639 *firstcharptr = firstchar;
3640 *reqcharptr = reqchar;
3641 *codeptr = code;
3642 *ptrptr = ptr;
3643 if (lengthptr != NULL)
3644 {
3645 if (OFLOW_MAX - *lengthptr < code - last_code)
3646 {
3647 *errorcodeptr = ERR20;
3648 goto FAILED;
3649 }
3650 *lengthptr += (int)(code - last_code); /* To include callout length */
3651 DPRINTF((">> end branch\n"));
3652 }
3653 return TRUE;
3654
3655
3656 /* ===================================================================*/
3657 /* Handle single-character metacharacters. In multiline mode, ^ disables
3658 the setting of any following char as a first character. */
3659
3660 case CHAR_CIRCUMFLEX_ACCENT:
3661 previous = NULL;
3662 if ((options & PCRE_MULTILINE) != 0)
3663 {
3664 if (firstchar == REQ_UNSET) firstchar = REQ_NONE;
3665 *code++ = OP_CIRCM;
3666 }
3667 else *code++ = OP_CIRC;
3668 break;
3669
3670 case CHAR_DOLLAR_SIGN:
3671 previous = NULL;
3672 *code++ = ((options & PCRE_MULTILINE) != 0)? OP_DOLLM : OP_DOLL;
3673 break;
3674
3675 /* There can never be a first char if '.' is first, whatever happens about
3676 repeats. The value of reqchar doesn't change either. */
3677
3678 case CHAR_DOT:
3679 if (firstchar == REQ_UNSET) firstchar = REQ_NONE;
3680 zerofirstchar = firstchar;
3681 zeroreqchar = reqchar;
3682 previous = code;
3683 *code++ = ((options & PCRE_DOTALL) != 0)? OP_ALLANY: OP_ANY;
3684 break;
3685
3686
3687 /* ===================================================================*/
3688 /* Character classes. If the included characters are all < 256, we build a
3689 32-byte bitmap of the permitted characters, except in the special case
3690 where there is only one such character. For negated classes, we build the
3691 map as usual, then invert it at the end. However, we use a different opcode
3692 so that data characters > 255 can be handled correctly.
3693
3694 If the class contains characters outside the 0-255 range, a different
3695 opcode is compiled. It may optionally have a bit map for characters < 256,
3696 but those above are are explicitly listed afterwards. A flag byte tells
3697 whether the bitmap is present, and whether this is a negated class or not.
3698
3699 In JavaScript compatibility mode, an isolated ']' causes an error. In
3700 default (Perl) mode, it is treated as a data character. */
3701
3702 case CHAR_RIGHT_SQUARE_BRACKET:
3703 if ((cd->external_options & PCRE_JAVASCRIPT_COMPAT) != 0)
3704 {
3705 *errorcodeptr = ERR64;
3706 goto FAILED;
3707 }
3708 goto NORMAL_CHAR;
3709
3710 case CHAR_LEFT_SQUARE_BRACKET:
3711 previous = code;
3712
3713 /* PCRE supports POSIX class stuff inside a class. Perl gives an error if
3714 they are encountered at the top level, so we'll do that too. */
3715
3716 if ((ptr[1] == CHAR_COLON || ptr[1] == CHAR_DOT ||
3717 ptr[1] == CHAR_EQUALS_SIGN) &&
3718 check_posix_syntax(ptr, &tempptr))
3719 {
3720 *errorcodeptr = (ptr[1] == CHAR_COLON)? ERR13 : ERR31;
3721 goto FAILED;
3722 }
3723
3724 /* If the first character is '^', set the negation flag and skip it. Also,
3725 if the first few characters (either before or after ^) are \Q\E or \E we
3726 skip them too. This makes for compatibility with Perl. */
3727
3728 negate_class = FALSE;
3729 for (;;)
3730 {
3731 c = *(++ptr);
3732 if (c == CHAR_BACKSLASH)
3733 {
3734 if (ptr[1] == CHAR_E)
3735 ptr++;
3736 else if (STRNCMP_UC_C8(ptr + 1, STR_Q STR_BACKSLASH STR_E, 3) == 0)
3737 ptr += 3;
3738 else
3739 break;
3740 }
3741 else if (!negate_class && c == CHAR_CIRCUMFLEX_ACCENT)
3742 negate_class = TRUE;
3743 else break;
3744 }
3745
3746 /* Empty classes are allowed in JavaScript compatibility mode. Otherwise,
3747 an initial ']' is taken as a data character -- the code below handles
3748 that. In JS mode, [] must always fail, so generate OP_FAIL, whereas
3749 [^] must match any character, so generate OP_ALLANY. */
3750
3751 if (c == CHAR_RIGHT_SQUARE_BRACKET &&
3752 (cd->external_options & PCRE_JAVASCRIPT_COMPAT) != 0)
3753 {
3754 *code++ = negate_class? OP_ALLANY : OP_FAIL;
3755 if (firstchar == REQ_UNSET) firstchar = REQ_NONE;
3756 zerofirstchar = firstchar;
3757 break;
3758 }
3759
3760 /* If a class contains a negative special such as \S, we need to flip the
3761 negation flag at the end, so that support for characters > 255 works
3762 correctly (they are all included in the class). */
3763
3764 should_flip_negation = FALSE;
3765
3766 /* For optimization purposes, we track some properties of the class.
3767 class_has_8bitchar will be non-zero, if the class contains at least one
3768 < 256 character. class_single_char will be 1 if the class contains only
3769 a single character. */
3770
3771 class_has_8bitchar = 0;
3772 class_single_char = 0;
3773
3774 /* Initialize the 32-char bit map to all zeros. We build the map in a
3775 temporary bit of memory, in case the class contains only 1 character (less
3776 than 256), because in that case the compiled code doesn't use the bit map.
3777 */
3778
3779 memset(classbits, 0, 32 * sizeof(pcre_uint8));
3780
3781 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
3782 xclass = FALSE; /* No chars >= 256 */
3783 class_uchardata = code + LINK_SIZE + 2; /* For UTF-8 items */
3784 class_uchardata_base = class_uchardata; /* For resetting in pass 1 */
3785 #endif
3786
3787 /* Process characters until ] is reached. By writing this as a "do" it
3788 means that an initial ] is taken as a data character. At the start of the
3789 loop, c contains the first byte of the character. */
3790
3791 if (c != 0) do
3792 {
3793 const pcre_uchar *oldptr;
3794
3795 #ifdef SUPPORT_UTF
3796 if (utf && HAS_EXTRALEN(c))
3797 { /* Braces are required because the */
3798 GETCHARLEN(c, ptr, ptr); /* macro generates multiple statements */
3799 }
3800 #endif
3801
3802 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
3803 /* In the pre-compile phase, accumulate the length of any extra
3804 data and reset the pointer. This is so that very large classes that
3805 contain a zillion > 255 characters no longer overwrite the work space
3806 (which is on the stack). */
3807
3808 if (lengthptr != NULL)
3809 {
3810 *lengthptr += class_uchardata - class_uchardata_base;
3811 class_uchardata = class_uchardata_base;
3812 }
3813 #endif
3814
3815 /* Inside \Q...\E everything is literal except \E */
3816
3817 if (inescq)
3818 {
3819 if (c == CHAR_BACKSLASH && ptr[1] == CHAR_E) /* If we are at \E */
3820 {
3821 inescq = FALSE; /* Reset literal state */
3822 ptr++; /* Skip the 'E' */
3823 continue; /* Carry on with next */
3824 }
3825 goto CHECK_RANGE; /* Could be range if \E follows */
3826 }
3827
3828 /* Handle POSIX class names. Perl allows a negation extension of the
3829 form [:^name:]. A square bracket that doesn't match the syntax is
3830 treated as a literal. We also recognize the POSIX constructions
3831 [.ch.] and [=ch=] ("collating elements") and fault them, as Perl
3832 5.6 and 5.8 do. */
3833
3834 if (c == CHAR_LEFT_SQUARE_BRACKET &&
3835 (ptr[1] == CHAR_COLON || ptr[1] == CHAR_DOT ||
3836 ptr[1] == CHAR_EQUALS_SIGN) && check_posix_syntax(ptr, &tempptr))
3837 {
3838 BOOL local_negate = FALSE;
3839 int posix_class, taboffset, tabopt;
3840 register const pcre_uint8 *cbits = cd->cbits;
3841 pcre_uint8 pbits[32];
3842
3843 if (ptr[1] != CHAR_COLON)
3844 {
3845 *errorcodeptr = ERR31;
3846 goto FAILED;
3847 }
3848
3849 ptr += 2;
3850 if (*ptr == CHAR_CIRCUMFLEX_ACCENT)
3851 {
3852 local_negate = TRUE;
3853 should_flip_negation = TRUE; /* Note negative special */
3854 ptr++;
3855 }
3856
3857 posix_class = check_posix_name(ptr, (int)(tempptr - ptr));
3858 if (posix_class < 0)
3859 {
3860 *errorcodeptr = ERR30;
3861 goto FAILED;
3862 }
3863
3864 /* If matching is caseless, upper and lower are converted to
3865 alpha. This relies on the fact that the class table starts with
3866 alpha, lower, upper as the first 3 entries. */
3867
3868 if ((options & PCRE_CASELESS) != 0 && posix_class <= 2)
3869 posix_class = 0;
3870
3871 /* When PCRE_UCP is set, some of the POSIX classes are converted to
3872 different escape sequences that use Unicode properties. */
3873
3874 #ifdef SUPPORT_UCP
3875 if ((options & PCRE_UCP) != 0)
3876 {
3877 int pc = posix_class + ((local_negate)? POSIX_SUBSIZE/2 : 0);
3878 if (posix_substitutes[pc] != NULL)
3879 {
3880 nestptr = tempptr + 1;
3881 ptr = posix_substitutes[pc] - 1;
3882 continue;
3883 }
3884 }
3885 #endif
3886 /* In the non-UCP case, we build the bit map for the POSIX class in a
3887 chunk of local store because we may be adding and subtracting from it,
3888 and we don't want to subtract bits that may be in the main map already.
3889 At the end we or the result into the bit map that is being built. */
3890
3891 posix_class *= 3;
3892
3893 /* Copy in the first table (always present) */
3894
3895 memcpy(pbits, cbits + posix_class_maps[posix_class],
3896 32 * sizeof(pcre_uint8));
3897
3898 /* If there is a second table, add or remove it as required. */
3899
3900 taboffset = posix_class_maps[posix_class + 1];
3901 tabopt = posix_class_maps[posix_class + 2];
3902
3903 if (taboffset >= 0)
3904 {
3905 if (tabopt >= 0)
3906 for (c = 0; c < 32; c++) pbits[c] |= cbits[c + taboffset];
3907 else
3908 for (c = 0; c < 32; c++) pbits[c] &= ~cbits[c + taboffset];
3909 }
3910
3911 /* Not see if we need to remove any special characters. An option
3912 value of 1 removes vertical space and 2 removes underscore. */
3913
3914 if (tabopt < 0) tabopt = -tabopt;
3915 if (tabopt == 1) pbits[1] &= ~0x3c;
3916 else if (tabopt == 2) pbits[11] &= 0x7f;
3917
3918 /* Add the POSIX table or its complement into the main table that is
3919 being built and we are done. */
3920
3921 if (local_negate)
3922 for (c = 0; c < 32; c++) classbits[c] |= ~pbits[c];
3923 else
3924 for (c = 0; c < 32; c++) classbits[c] |= pbits[c];
3925
3926 ptr = tempptr + 1;
3927 /* Every class contains at least one < 256 characters. */
3928 class_has_8bitchar = 1;
3929 /* Every class contains at least two characters. */
3930 class_single_char = 2;
3931 continue; /* End of POSIX syntax handling */
3932 }
3933
3934 /* Backslash may introduce a single character, or it may introduce one
3935 of the specials, which just set a flag. The sequence \b is a special
3936 case. Inside a class (and only there) it is treated as backspace. We
3937 assume that other escapes have more than one character in them, so
3938 speculatively set both class_has_8bitchar and class_single_char bigger
3939 than one. Unrecognized escapes fall through and are either treated
3940 as literal characters (by default), or are faulted if
3941 PCRE_EXTRA is set. */
3942
3943 if (c == CHAR_BACKSLASH)
3944 {
3945 c = check_escape(&ptr, errorcodeptr, cd->bracount, options, TRUE);
3946 if (*errorcodeptr != 0) goto FAILED;
3947
3948 if (-c == ESC_b) c = CHAR_BS; /* \b is backspace in a class */
3949 else if (-c == ESC_N) /* \N is not supported in a class */
3950 {
3951 *errorcodeptr = ERR71;
3952 goto FAILED;
3953 }
3954 else if (-c == ESC_Q) /* Handle start of quoted string */
3955 {
3956 if (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
3957 {
3958 ptr += 2; /* avoid empty string */
3959 }
3960 else inescq = TRUE;
3961 continue;
3962 }
3963 else if (-c == ESC_E) continue; /* Ignore orphan \E */
3964
3965 if (c < 0)
3966 {
3967 register const pcre_uint8 *cbits = cd->cbits;
3968 /* Every class contains at least two < 256 characters. */
3969 class_has_8bitchar++;
3970 /* Every class contains at least two characters. */
3971 class_single_char += 2;
3972
3973 switch (-c)
3974 {
3975 #ifdef SUPPORT_UCP
3976 case ESC_du: /* These are the values given for \d etc */
3977 case ESC_DU: /* when PCRE_UCP is set. We replace the */
3978 case ESC_wu: /* escape sequence with an appropriate \p */
3979 case ESC_WU: /* or \P to test Unicode properties instead */
3980 case ESC_su: /* of the default ASCII testing. */
3981 case ESC_SU:
3982 nestptr = ptr;
3983 ptr = substitutes[-c - ESC_DU] - 1; /* Just before substitute */
3984 class_has_8bitchar--; /* Undo! */
3985 continue;
3986 #endif
3987 case ESC_d:
3988 for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_digit];
3989 continue;
3990
3991 case ESC_D:
3992 should_flip_negation = TRUE;
3993 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_digit];
3994 continue;
3995
3996 case ESC_w:
3997 for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_word];
3998 continue;
3999
4000 case ESC_W:
4001 should_flip_negation = TRUE;
4002 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_word];
4003 continue;
4004
4005 /* Perl 5.004 onwards omits VT from \s, but we must preserve it
4006 if it was previously set by something earlier in the character
4007 class. */
4008
4009 case ESC_s:
4010 classbits[0] |= cbits[cbit_space];
4011 classbits[1] |= cbits[cbit_space+1] & ~0x08;
4012 for (c = 2; c < 32; c++) classbits[c] |= cbits[c+cbit_space];
4013 continue;
4014
4015 case ESC_S:
4016 should_flip_negation = TRUE;
4017 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_space];
4018 classbits[1] |= 0x08; /* Perl 5.004 onwards omits VT from \s */
4019 continue;
4020
4021 case ESC_h:
4022 SETBIT(classbits, 0x09); /* VT */
4023 SETBIT(classbits, 0x20); /* SPACE */
4024 SETBIT(classbits, 0xa0); /* NSBP */
4025 #ifdef SUPPORT_UTF
4026 if (utf)
4027 {
4028 xclass = TRUE;
4029 *class_uchardata++ = XCL_SINGLE;
4030 class_uchardata += PRIV(ord2utf)(0x1680, class_uchardata);
4031 *class_uchardata++ = XCL_SINGLE;
4032 class_uchardata += PRIV(ord2utf)(0x180e, class_uchardata);
4033 *class_uchardata++ = XCL_RANGE;
4034 class_uchardata += PRIV(ord2utf)(0x2000, class_uchardata);
4035 class_uchardata += PRIV(ord2utf)(0x200A, class_uchardata);
4036 *class_uchardata++ = XCL_SINGLE;
4037 class_uchardata += PRIV(ord2utf)(0x202f, class_uchardata);
4038 *class_uchardata++ = XCL_SINGLE;
4039 class_uchardata += PRIV(ord2utf)(0x205f, class_uchardata);
4040 *class_uchardata++ = XCL_SINGLE;
4041 class_uchardata += PRIV(ord2utf)(0x3000, class_uchardata);
4042 }
4043 #endif
4044 continue;
4045
4046 case ESC_H:
4047 for (c = 0; c < 32; c++)
4048 {
4049 int x = 0xff;
4050 switch (c)
4051 {
4052 case 0x09/8: x ^= 1 << (0x09%8); break;
4053 case 0x20/8: x ^= 1 << (0x20%8); break;
4054 case 0xa0/8: x ^= 1 << (0xa0%8); break;
4055 default: break;
4056 }
4057 classbits[c] |= x;
4058 }
4059
4060 #ifdef SUPPORT_UTF
4061 if (utf)
4062 {
4063 xclass = TRUE;
4064 *class_uchardata++ = XCL_RANGE;
4065 class_uchardata += PRIV(ord2utf)(0x0100, class_uchardata);
4066 class_uchardata += PRIV(ord2utf)(0x167f, class_uchardata);
4067 *class_uchardata++ = XCL_RANGE;
4068 class_uchardata += PRIV(ord2utf)(0x1681, class_uchardata);
4069 class_uchardata += PRIV(ord2utf)(0x180d, class_uchardata);
4070 *class_uchardata++ = XCL_RANGE;
4071 class_uchardata += PRIV(ord2utf)(0x180f, class_uchardata);
4072 class_uchardata += PRIV(ord2utf)(0x1fff, class_uchardata);
4073 *class_uchardata++ = XCL_RANGE;
4074 class_uchardata += PRIV(ord2utf)(0x200B, class_uchardata);
4075 class_uchardata += PRIV(ord2utf)(0x202e, class_uchardata);
4076 *class_uchardata++ = XCL_RANGE;
4077 class_uchardata += PRIV(ord2utf)(0x2030, class_uchardata);
4078 class_uchardata += PRIV(ord2utf)(0x205e, class_uchardata);
4079 *class_uchardata++ = XCL_RANGE;
4080 class_uchardata += PRIV(ord2utf)(0x2060, class_uchardata);
4081 class_uchardata += PRIV(ord2utf)(0x2fff, class_uchardata);
4082 *class_uchardata++ = XCL_RANGE;
4083 class_uchardata += PRIV(ord2utf)(0x3001, class_uchardata);
4084 class_uchardata += PRIV(ord2utf)(0x10ffff, class_uchardata);
4085 }
4086 #endif
4087 continue;
4088
4089 case ESC_v:
4090 SETBIT(classbits, 0x0a); /* LF */
4091 SETBIT(classbits, 0x0b); /* VT */
4092 SETBIT(classbits, 0x0c); /* FF */
4093 SETBIT(classbits, 0x0d); /* CR */
4094 SETBIT(classbits, 0x85); /* NEL */
4095 #ifdef SUPPORT_UTF
4096 if (utf)
4097 {
4098 xclass = TRUE;
4099 *class_uchardata++ = XCL_RANGE;
4100 class_uchardata += PRIV(ord2utf)(0x2028, class_uchardata);
4101 class_uchardata += PRIV(ord2utf)(0x2029, class_uchardata);
4102 }
4103 #endif
4104 continue;
4105
4106 case ESC_V:
4107 for (c = 0; c < 32; c++)
4108 {
4109 int x = 0xff;
4110 switch (c)
4111 {
4112 case 0x0a/8: x ^= 1 << (0x0a%8);
4113 x ^= 1 << (0x0b%8);
4114 x ^= 1 << (0x0c%8);
4115 x ^= 1 << (0x0d%8);
4116 break;
4117 case 0x85/8: x ^= 1 << (0x85%8); break;
4118 default: break;
4119 }
4120 classbits[c] |= x;
4121 }
4122
4123 #ifdef SUPPORT_UTF
4124 if (utf)
4125 {
4126 xclass = TRUE;
4127 *class_uchardata++ = XCL_RANGE;
4128 class_uchardata += PRIV(ord2utf)(0x0100, class_uchardata);
4129 class_uchardata += PRIV(ord2utf)(0x2027, class_uchardata);
4130 *class_uchardata++ = XCL_RANGE;
4131 class_uchardata += PRIV(ord2utf)(0x2029, class_uchardata);
4132 class_uchardata += PRIV(ord2utf)(0x10ffff, class_uchardata);
4133 }
4134 #endif
4135 continue;
4136
4137 #ifdef SUPPORT_UCP
4138 case ESC_p:
4139 case ESC_P:
4140 {
4141 BOOL negated;
4142 int pdata;
4143 int ptype = get_ucp(&ptr, &negated, &pdata, errorcodeptr);
4144 if (ptype < 0) goto FAILED;
4145 xclass = TRUE;
4146 *class_uchardata++ = ((-c == ESC_p) != negated)?
4147 XCL_PROP : XCL_NOTPROP;
4148 *class_uchardata++ = ptype;
4149 *class_uchardata++ = pdata;
4150 class_has_8bitchar--; /* Undo! */
4151 continue;
4152 }
4153 #endif
4154 /* Unrecognized escapes are faulted if PCRE is running in its
4155 strict mode. By default, for compatibility with Perl, they are
4156 treated as literals. */
4157
4158 default:
4159 if ((options & PCRE_EXTRA) != 0)
4160 {
4161 *errorcodeptr = ERR7;
4162 goto FAILED;
4163 }
4164 class_has_8bitchar--; /* Undo the speculative increase. */
4165 class_single_char -= 2; /* Undo the speculative increase. */
4166 c = *ptr; /* Get the final character and fall through */
4167 break;
4168 }
4169 }
4170
4171 /* Fall through if we have a single character (c >= 0). This may be
4172 greater than 256. */
4173
4174 } /* End of backslash handling */
4175
4176 /* A single character may be followed by '-' to form a range. However,
4177 Perl does not permit ']' to be the end of the range. A '-' character
4178 at the end is treated as a literal. Perl ignores orphaned \E sequences
4179 entirely. The code for handling \Q and \E is messy. */
4180
4181 CHECK_RANGE:
4182 while (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
4183 {
4184 inescq = FALSE;
4185 ptr += 2;
4186 }
4187
4188 oldptr = ptr;
4189
4190 /* Remember \r or \n */
4191
4192 if (c == CHAR_CR || c == CHAR_NL) cd->external_flags |= PCRE_HASCRORLF;
4193
4194 /* Check for range */
4195
4196 if (!inescq && ptr[1] == CHAR_MINUS)
4197 {
4198 int d;
4199 ptr += 2;
4200 while (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_E) ptr += 2;
4201
4202 /* If we hit \Q (not followed by \E) at this point, go into escaped
4203 mode. */
4204
4205 while (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_Q)
4206 {
4207 ptr += 2;
4208 if (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_E)
4209 { ptr += 2; continue; }
4210 inescq = TRUE;
4211 break;
4212 }
4213
4214 if (*ptr == 0 || (!inescq && *ptr == CHAR_RIGHT_SQUARE_BRACKET))
4215 {
4216 ptr = oldptr;
4217 goto LONE_SINGLE_CHARACTER;
4218 }
4219
4220 #ifdef SUPPORT_UTF
4221 if (utf)
4222 { /* Braces are required because the */
4223 GETCHARLEN(d, ptr, ptr); /* macro generates multiple statements */
4224 }
4225 else
4226 #endif
4227 d = *ptr; /* Not UTF-8 mode */
4228
4229 /* The second part of a range can be a single-character escape, but
4230 not any of the other escapes. Perl 5.6 treats a hyphen as a literal
4231 in such circumstances. */
4232
4233 if (!inescq && d == CHAR_BACKSLASH)
4234 {
4235 d = check_escape(&ptr, errorcodeptr, cd->bracount, options, TRUE);
4236 if (*errorcodeptr != 0) goto FAILED;
4237
4238 /* \b is backspace; any other special means the '-' was literal */
4239
4240 if (d < 0)
4241 {
4242 if (d == -ESC_b) d = CHAR_BS; else
4243 {
4244 ptr = oldptr;
4245 goto LONE_SINGLE_CHARACTER; /* A few lines below */
4246 }
4247 }
4248 }
4249
4250 /* Check that the two values are in the correct order. Optimize
4251 one-character ranges */
4252
4253 if (d < c)
4254 {
4255 *errorcodeptr = ERR8;
4256 goto FAILED;
4257 }
4258
4259 if (d == c) goto LONE_SINGLE_CHARACTER; /* A few lines below */
4260
4261 /* Remember \r or \n */
4262
4263 if (d == CHAR_CR || d == CHAR_NL) cd->external_flags |= PCRE_HASCRORLF;
4264
4265 /* Since we found a character range, single character optimizations
4266 cannot be done anymore. */
4267 class_single_char = 2;
4268
4269 /* In UTF-8 mode, if the upper limit is > 255, or > 127 for caseless
4270 matching, we have to use an XCLASS with extra data items. Caseless
4271 matching for characters > 127 is available only if UCP support is
4272 available. */
4273
4274 #if defined SUPPORT_UTF && !(defined COMPILE_PCRE8)
4275 if ((d > 255) || (utf && ((options & PCRE_CASELESS) != 0 && d > 127)))
4276 #elif defined SUPPORT_UTF
4277 if (utf && (d > 255 || ((options & PCRE_CASELESS) != 0 && d > 127)))
4278 #elif !(defined COMPILE_PCRE8)
4279 if (d > 255)
4280 #endif
4281 #if defined SUPPORT_UTF || !(defined COMPILE_PCRE8)
4282 {
4283 xclass = TRUE;
4284
4285 /* With UCP support, we can find the other case equivalents of
4286 the relevant characters. There may be several ranges. Optimize how
4287 they fit with the basic range. */
4288
4289 #ifdef SUPPORT_UCP
4290 #ifndef COMPILE_PCRE8
4291 if (utf && (options & PCRE_CASELESS) != 0)
4292 #else
4293 if ((options & PCRE_CASELESS) != 0)
4294 #endif
4295 {
4296 unsigned int occ, ocd;
4297 unsigned int cc = c;
4298 unsigned int origd = d;
4299 while (get_othercase_range(&cc, origd, &occ, &ocd))
4300 {
4301 if (occ >= (unsigned int)c &&
4302 ocd <= (unsigned int)d)
4303 continue; /* Skip embedded ranges */
4304
4305 if (occ < (unsigned int)c &&
4306 ocd >= (unsigned int)c - 1) /* Extend the basic range */
4307 { /* if there is overlap, */
4308 c = occ; /* noting that if occ < c */
4309 continue; /* we can't have ocd > d */
4310 } /* because a subrange is */
4311 if (ocd > (unsigned int)d &&
4312 occ <= (unsigned int)d + 1) /* always shorter than */
4313 { /* the basic range. */
4314 d = ocd;
4315 continue;
4316 }
4317
4318 if (occ == ocd)
4319 {
4320 *class_uchardata++ = XCL_SINGLE;
4321 }
4322 else
4323 {
4324 *class_uchardata++ = XCL_RANGE;
4325 class_uchardata += PRIV(ord2utf)(occ, class_uchardata);
4326 }
4327 class_uchardata += PRIV(ord2utf)(ocd, class_uchardata);
4328 }
4329 }
4330 #endif /* SUPPORT_UCP */
4331
4332 /* Now record the original range, possibly modified for UCP caseless
4333 overlapping ranges. */
4334
4335 *class_uchardata++ = XCL_RANGE;
4336 #ifdef SUPPORT_UTF
4337 #ifndef COMPILE_PCRE8
4338 if (utf)
4339 {
4340 class_uchardata += PRIV(ord2utf)(c, class_uchardata);
4341 class_uchardata += PRIV(ord2utf)(d, class_uchardata);
4342 }
4343 else
4344 {
4345 *class_uchardata++ = c;
4346 *class_uchardata++ = d;
4347 }
4348 #else
4349 class_uchardata += PRIV(ord2utf)(c, class_uchardata);
4350 class_uchardata += PRIV(ord2utf)(d, class_uchardata);
4351 #endif
4352 #else /* SUPPORT_UTF */
4353 *class_uchardata++ = c;
4354 *class_uchardata++ = d;
4355 #endif /* SUPPORT_UTF */
4356
4357 /* With UCP support, we are done. Without UCP support, there is no
4358 caseless matching for UTF characters > 127; we can use the bit map
4359 for the smaller ones. As for 16 bit characters without UTF, we
4360 can still use */
4361
4362 #ifdef SUPPORT_UCP
4363 #ifndef COMPILE_PCRE8
4364 if (utf)
4365 #endif
4366 continue; /* With next character in the class */
4367 #endif /* SUPPORT_UCP */
4368
4369 #if defined SUPPORT_UTF && !defined(SUPPORT_UCP) && !(defined COMPILE_PCRE8)
4370 if (utf)
4371 {
4372 if ((options & PCRE_CASELESS) == 0 || c > 127) continue;
4373 /* Adjust upper limit and fall through to set up the map */
4374 d = 127;
4375 }
4376 else
4377 {
4378 if (c > 255) continue;
4379 /* Adjust upper limit and fall through to set up the map */
4380 d = 255;
4381 }
4382 #elif defined SUPPORT_UTF && !defined(SUPPORT_UCP)
4383 if ((options & PCRE_CASELESS) == 0 || c > 127) continue;
4384 /* Adjust upper limit and fall through to set up the map */
4385 d = 127;
4386 #else
4387 if (c > 255) continue;
4388 /* Adjust upper limit and fall through to set up the map */
4389 d = 255;
4390 #endif /* SUPPORT_UTF && !SUPPORT_UCP && !COMPILE_PCRE8 */
4391 }
4392 #endif /* SUPPORT_UTF || !COMPILE_PCRE8 */
4393
4394 /* We use the bit map for 8 bit mode, or when the characters fall
4395 partially or entirely to [0-255] ([0-127] for UCP) ranges. */
4396
4397 class_has_8bitchar = 1;
4398
4399 /* We can save a bit of time by skipping this in the pre-compile. */
4400
4401 if (lengthptr == NULL) for (; c <= d; c++)
4402 {
4403 classbits[c/8] |= (1 << (c&7));
4404 if ((options & PCRE_CASELESS) != 0)
4405 {
4406 int uc = cd->fcc[c]; /* flip case */
4407 classbits[uc/8] |= (1 << (uc&7));
4408 }
4409 }
4410
4411 continue; /* Go get the next char in the class */
4412 }
4413
4414 /* Handle a lone single character - we can get here for a normal
4415 non-escape char, or after \ that introduces a single character or for an
4416 apparent range that isn't. */
4417
4418 LONE_SINGLE_CHARACTER:
4419
4420 /* Only the value of 1 matters for class_single_char. */
4421 if (class_single_char < 2) class_single_char++;
4422
4423 /* If class_charcount is 1, we saw precisely one character. As long as
4424 there were no negated characters >= 128 and there was no use of \p or \P,
4425 in other words, no use of any XCLASS features, we can optimize.
4426
4427 In UTF-8 mode, we can optimize the negative case only if there were no
4428 characters >= 128 because OP_NOT and the related opcodes like OP_NOTSTAR
4429 operate on single-bytes characters only. This is an historical hangover.
4430 Maybe one day we can tidy these opcodes to handle multi-byte characters.
4431
4432 The optimization throws away the bit map. We turn the item into a
4433 1-character OP_CHAR[I] if it's positive, or OP_NOT[I] if it's negative.
4434 Note that OP_NOT[I] does not support multibyte characters. In the positive
4435 case, it can cause firstchar to be set. Otherwise, there can be no first
4436 char if this item is first, whatever repeat count may follow. In the case
4437 of reqchar, save the previous value for reinstating. */
4438
4439 #ifdef SUPPORT_UTF
4440 if (class_single_char == 1 && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET
4441 && (!utf || !negate_class || c < (MAX_VALUE_FOR_SINGLE_CHAR + 1)))
4442 #else
4443 if (class_single_char == 1 && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET)
4444 #endif
4445 {
4446 ptr++;
4447 zeroreqchar = reqchar;
4448
4449 /* The OP_NOT[I] opcodes work on single characters only. */
4450
4451 if (negate_class)
4452 {
4453 if (firstchar == REQ_UNSET) firstchar = REQ_NONE;
4454 zerofirstchar = firstchar;
4455 *code++ = ((options & PCRE_CASELESS) != 0)? OP_NOTI: OP_NOT;
4456 *code++ = c;
4457 goto NOT_CHAR;
4458 }
4459
4460 /* For a single, positive character, get the value into mcbuffer, and
4461 then we can handle this with the normal one-character code. */
4462
4463 #ifdef SUPPORT_UTF
4464 if (utf && c > MAX_VALUE_FOR_SINGLE_CHAR)
4465 mclength = PRIV(ord2utf)(c, mcbuffer);
4466 else
4467 #endif
4468 {
4469 mcbuffer[0] = c;
4470 mclength = 1;
4471 }
4472 goto ONE_CHAR;
4473 } /* End of 1-char optimization */
4474
4475 /* Handle a character that cannot go in the bit map. */
4476
4477 #if defined SUPPORT_UTF && !(defined COMPILE_PCRE8)
4478 if ((c > 255) || (utf && ((options & PCRE_CASELESS) != 0 && c > 127)))
4479 #elif defined SUPPORT_UTF
4480 if (utf && (c > 255 || ((options & PCRE_CASELESS) != 0 && c > 127)))
4481 #elif !(defined COMPILE_PCRE8)
4482 if (c > 255)
4483 #endif
4484
4485 #if defined SUPPORT_UTF || !(defined COMPILE_PCRE8)
4486 {
4487 xclass = TRUE;
4488 *class_uchardata++ = XCL_SINGLE;
4489 #ifdef SUPPORT_UTF
4490 #ifndef COMPILE_PCRE8
4491 /* In non 8 bit mode, we can get here even if we are not in UTF mode. */
4492 if (!utf)
4493 *class_uchardata++ = c;
4494 else
4495 #endif
4496 class_uchardata += PRIV(ord2utf)(c, class_uchardata);
4497 #else /* SUPPORT_UTF */
4498 *class_uchardata++ = c;
4499 #endif /* SUPPORT_UTF */
4500
4501 #ifdef SUPPORT_UCP
4502 #ifdef COMPILE_PCRE8
4503 if ((options & PCRE_CASELESS) != 0)
4504 #else
4505 /* In non 8 bit mode, we can get here even if we are not in UTF mode. */
4506 if (utf && (options & PCRE_CASELESS) != 0)
4507 #endif
4508 {
4509 unsigned int othercase;
4510 if ((othercase = UCD_OTHERCASE(c)) != c)
4511 {
4512 *class_uchardata++ = XCL_SINGLE;
4513 class_uchardata += PRIV(ord2utf)(othercase, class_uchardata);
4514 }
4515 }
4516 #endif /* SUPPORT_UCP */
4517
4518 }
4519 else
4520 #endif /* SUPPORT_UTF || COMPILE_PCRE16 */
4521
4522 /* Handle a single-byte character */
4523 {
4524 class_has_8bitchar = 1;
4525 classbits[c/8] |= (1 << (c&7));
4526 if ((options & PCRE_CASELESS) != 0)
4527 {
4528 c = cd->fcc[c]; /* flip case */
4529 classbits[c/8] |= (1 << (c&7));
4530 }
4531 }
4532 }
4533
4534 /* Loop until ']' reached. This "while" is the end of the "do" far above.
4535 If we are at the end of an internal nested string, revert to the outer
4536 string. */
4537
4538 while (((c = *(++ptr)) != 0 ||
4539 (nestptr != NULL &&
4540 (ptr = nestptr, nestptr = NULL, c = *(++ptr)) != 0)) &&
4541 (c != CHAR_RIGHT_SQUARE_BRACKET || inescq));
4542
4543 /* Check for missing terminating ']' */
4544
4545 if (c == 0)
4546 {
4547 *errorcodeptr = ERR6;
4548 goto FAILED;
4549 }
4550
4551 /* If this is the first thing in the branch, there can be no first char
4552 setting, whatever the repeat count. Any reqchar setting must remain
4553 unchanged after any kind of repeat. */
4554
4555 if (firstchar == REQ_UNSET) firstchar = REQ_NONE;
4556 zerofirstchar = firstchar;
4557 zeroreqchar = reqchar;
4558
4559 /* If there are characters with values > 255, we have to compile an
4560 extended class, with its own opcode, unless there was a negated special
4561 such as \S in the class, and PCRE_UCP is not set, because in that case all
4562 characters > 255 are in the class, so any that were explicitly given as
4563 well can be ignored. If (when there are explicit characters > 255 that must
4564 be listed) there are no characters < 256, we can omit the bitmap in the
4565 actual compiled code. */
4566
4567 #ifdef SUPPORT_UTF
4568 if (xclass && (!should_flip_negation || (options & PCRE_UCP) != 0))
4569 #endif
4570 #ifndef COMPILE_PCRE8
4571 if (xclass && !should_flip_negation)
4572 #endif
4573 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
4574 {
4575 *class_uchardata++ = XCL_END; /* Marks the end of extra data */
4576 *code++ = OP_XCLASS;
4577 code += LINK_SIZE;
4578 *code = negate_class? XCL_NOT:0;
4579
4580 /* If the map is required, move up the extra data to make room for it;
4581 otherwise just move the code pointer to the end of the extra data. */
4582
4583 if (class_has_8bitchar > 0)
4584 {
4585 *code++ |= XCL_MAP;
4586 memmove(code + (32 / sizeof(pcre_uchar)), code,
4587 IN_UCHARS(class_uchardata - code));
4588 memcpy(code, classbits, 32);
4589 code = class_uchardata + (32 / sizeof(pcre_uchar));
4590 }
4591 else code = class_uchardata;
4592
4593 /* Now fill in the complete length of the item */
4594
4595 PUT(previous, 1, (int)(code - previous));
4596 break; /* End of class handling */
4597 }
4598 #endif
4599
4600 /* If there are no characters > 255, or they are all to be included or
4601 excluded, set the opcode to OP_CLASS or OP_NCLASS, depending on whether the
4602 whole class was negated and whether there were negative specials such as \S
4603 (non-UCP) in the class. Then copy the 32-byte map into the code vector,
4604 negating it if necessary. */
4605
4606 *code++ = (negate_class == should_flip_negation) ? OP_CLASS : OP_NCLASS;
4607 if (lengthptr == NULL) /* Save time in the pre-compile phase */
4608 {
4609 if (negate_class)
4610 for (c = 0; c < 32; c++) classbits[c] = ~classbits[c];
4611 memcpy(code, classbits, 32);
4612 }
4613 code += 32 / sizeof(pcre_uchar);
4614 NOT_CHAR:
4615 break;
4616
4617
4618 /* ===================================================================*/
4619 /* Various kinds of repeat; '{' is not necessarily a quantifier, but this
4620 has been tested above. */
4621
4622 case CHAR_LEFT_CURLY_BRACKET:
4623 if (!is_quantifier) goto NORMAL_CHAR;
4624 ptr = read_repeat_counts(ptr+1, &repeat_min, &repeat_max, errorcodeptr);
4625 if (*errorcodeptr != 0) goto FAILED;
4626 goto REPEAT;
4627
4628 case CHAR_ASTERISK:
4629 repeat_min = 0;
4630 repeat_max = -1;
4631 goto REPEAT;
4632
4633 case CHAR_PLUS:
4634 repeat_min = 1;
4635 repeat_max = -1;
4636 goto REPEAT;
4637
4638 case CHAR_QUESTION_MARK:
4639 repeat_min = 0;
4640 repeat_max = 1;
4641
4642 REPEAT:
4643 if (previous == NULL)
4644 {
4645 *errorcodeptr = ERR9;
4646 goto FAILED;
4647 }
4648
4649 if (repeat_min == 0)
4650 {
4651 firstchar = zerofirstchar; /* Adjust for zero repeat */
4652 reqchar = zeroreqchar; /* Ditto */
4653 }
4654
4655 /* Remember whether this is a variable length repeat */
4656
4657 reqvary = (repeat_min == repeat_max)? 0 : REQ_VARY;
4658
4659 op_type = 0; /* Default single-char op codes */
4660 possessive_quantifier = FALSE; /* Default not possessive quantifier */
4661
4662 /* Save start of previous item, in case we have to move it up in order to
4663 insert something before it. */
4664
4665 tempcode = previous;
4666
4667 /* If the next character is '+', we have a possessive quantifier. This
4668 implies greediness, whatever the setting of the PCRE_UNGREEDY option.
4669 If the next character is '?' this is a minimizing repeat, by default,
4670 but if PCRE_UNGREEDY is set, it works the other way round. We change the
4671 repeat type to the non-default. */
4672
4673 if (ptr[1] == CHAR_PLUS)
4674 {
4675 repeat_type = 0; /* Force greedy */
4676 possessive_quantifier = TRUE;
4677 ptr++;
4678 }
4679 else if (ptr[1] == CHAR_QUESTION_MARK)
4680 {
4681 repeat_type = greedy_non_default;
4682 ptr++;
4683 }
4684 else repeat_type = greedy_default;
4685
4686 /* If previous was a recursion call, wrap it in atomic brackets so that
4687 previous becomes the atomic group. All recursions were so wrapped in the
4688 past, but it no longer happens for non-repeated recursions. In fact, the
4689 repeated ones could be re-implemented independently so as not to need this,
4690 but for the moment we rely on the code for repeating groups. */
4691
4692 if (*previous == OP_RECURSE)
4693 {
4694 memmove(previous + 1 + LINK_SIZE, previous, IN_UCHARS(1 + LINK_SIZE));
4695 *previous = OP_ONCE;
4696 PUT(previous, 1, 2 + 2*LINK_SIZE);
4697 previous[2 + 2*LINK_SIZE] = OP_KET;
4698 PUT(previous, 3 + 2*LINK_SIZE, 2 + 2*LINK_SIZE);
4699 code += 2 + 2 * LINK_SIZE;
4700 length_prevgroup = 3 + 3*LINK_SIZE;
4701
4702 /* When actually compiling, we need to check whether this was a forward
4703 reference, and if so, adjust the offset. */
4704
4705 if (lengthptr == NULL && cd->hwm >= cd->start_workspace + LINK_SIZE)
4706 {
4707 int offset = GET(cd->hwm, -LINK_SIZE);
4708 if (offset == previous + 1 - cd->start_code)
4709 PUT(cd->hwm, -LINK_SIZE, offset + 1 + LINK_SIZE);
4710 }
4711 }
4712
4713 /* Now handle repetition for the different types of item. */
4714
4715 /* If previous was a character match, abolish the item and generate a
4716 repeat item instead. If a char item has a minumum of more than one, ensure
4717 that it is set in reqchar - it might not be if a sequence such as x{3} is
4718 the first thing in a branch because the x will have gone into firstchar
4719 instead. */
4720
4721 if (*previous == OP_CHAR || *previous == OP_CHARI)
4722 {
4723 op_type = (*previous == OP_CHAR)? 0 : OP_STARI - OP_STAR;
4724
4725 /* Deal with UTF characters that take up more than one character. It's
4726 easier to write this out separately than try to macrify it. Use c to
4727 hold the length of the character in bytes, plus UTF_LENGTH to flag that
4728 it's a length rather than a small character. */
4729
4730 #ifdef SUPPORT_UTF
4731 if (utf && NOT_FIRSTCHAR(code[-1]))
4732 {
4733 pcre_uchar *lastchar = code - 1;
4734 BACKCHAR(lastchar);
4735 c = (int)(code - lastchar); /* Length of UTF-8 character */
4736 memcpy(utf_chars, lastchar, IN_UCHARS(c)); /* Save the char */
4737 c |= UTF_LENGTH; /* Flag c as a length */
4738 }
4739 else
4740 #endif /* SUPPORT_UTF */
4741
4742 /* Handle the case of a single charater - either with no UTF support, or
4743 with UTF disabled, or for a single character UTF character. */
4744 {
4745 c = code[-1];
4746 if (repeat_min > 1) reqchar = c | req_caseopt | cd->req_varyopt;
4747 }
4748
4749 /* If the repetition is unlimited, it pays to see if the next thing on
4750 the line is something that cannot possibly match this character. If so,
4751 automatically possessifying this item gains some performance in the case
4752 where the match fails. */
4753
4754 if (!possessive_quantifier &&
4755 repeat_max < 0 &&
4756 check_auto_possessive(previous, utf, ptr + 1, options, cd))
4757 {
4758 repeat_type = 0; /* Force greedy */
4759 possessive_quantifier = TRUE;
4760 }
4761
4762 goto OUTPUT_SINGLE_REPEAT; /* Code shared with single character types */
4763 }
4764
4765 /* If previous was a single negated character ([^a] or similar), we use
4766 one of the special opcodes, replacing it. The code is shared with single-
4767 character repeats by setting opt_type to add a suitable offset into
4768 repeat_type. We can also test for auto-possessification. OP_NOT and OP_NOTI
4769 are currently used only for single-byte chars. */
4770
4771 else if (*previous == OP_NOT || *previous == OP_NOTI)
4772 {
4773 op_type = ((*previous == OP_NOT)? OP_NOTSTAR : OP_NOTSTARI) - OP_STAR;
4774 c = previous[1];
4775 if (!possessive_quantifier &&
4776 repeat_max < 0 &&
4777 check_auto_possessive(previous, utf, ptr + 1, options, cd))
4778 {
4779 repeat_type = 0; /* Force greedy */
4780 possessive_quantifier = TRUE;
4781 }
4782 goto OUTPUT_SINGLE_REPEAT;
4783 }
4784
4785 /* If previous was a character type match (\d or similar), abolish it and
4786 create a suitable repeat item. The code is shared with single-character
4787 repeats by setting op_type to add a suitable offset into repeat_type. Note
4788 the the Unicode property types will be present only when SUPPORT_UCP is
4789 defined, but we don't wrap the little bits of code here because it just
4790 makes it horribly messy. */
4791
4792 else if (*previous < OP_EODN)
4793 {
4794 pcre_uchar *oldcode;
4795 int prop_type, prop_value;
4796 op_type = OP_TYPESTAR - OP_STAR; /* Use type opcodes */
4797 c = *previous;
4798
4799 if (!possessive_quantifier &&
4800 repeat_max < 0 &&
4801 check_auto_possessive(previous, utf, ptr + 1, options, cd))
4802 {
4803 repeat_type = 0; /* Force greedy */
4804 possessive_quantifier = TRUE;
4805 }
4806
4807 OUTPUT_SINGLE_REPEAT:
4808 if (*previous == OP_PROP || *previous == OP_NOTPROP)
4809 {
4810 prop_type = previous[1];
4811 prop_value = previous[2];
4812 }
4813 else prop_type = prop_value = -1;
4814
4815 oldcode = code;
4816 code = previous; /* Usually overwrite previous item */
4817
4818 /* If the maximum is zero then the minimum must also be zero; Perl allows
4819 this case, so we do too - by simply omitting the item altogether. */
4820
4821 if (repeat_max == 0) goto END_REPEAT;
4822
4823 /*--------------------------------------------------------------------*/
4824 /* This code is obsolete from release 8.00; the restriction was finally
4825 removed: */
4826
4827 /* All real repeats make it impossible to handle partial matching (maybe
4828 one day we will be able to remove this restriction). */
4829
4830 /* if (repeat_max != 1) cd->external_flags |= PCRE_NOPARTIAL; */
4831 /*--------------------------------------------------------------------*/
4832
4833 /* Combine the op_type with the repeat_type */
4834
4835 repeat_type += op_type;
4836
4837 /* A minimum of zero is handled either as the special case * or ?, or as
4838 an UPTO, with the maximum given. */
4839
4840 if (repeat_min == 0)
4841 {
4842 if (repeat_max == -1) *code++ = OP_STAR + repeat_type;
4843 else if (repeat_max == 1) *code++ = OP_QUERY + repeat_type;
4844 else
4845 {
4846 *code++ = OP_UPTO + repeat_type;
4847 PUT2INC(code, 0, repeat_max);
4848 }
4849 }
4850
4851 /* A repeat minimum of 1 is optimized into some special cases. If the
4852 maximum is unlimited, we use OP_PLUS. Otherwise, the original item is
4853 left in place and, if the maximum is greater than 1, we use OP_UPTO with
4854 one less than the maximum. */
4855
4856 else if (repeat_min == 1)
4857 {
4858 if (repeat_max == -1)
4859 *code++ = OP_PLUS + repeat_type;
4860 else
4861 {
4862 code = oldcode; /* leave previous item in place */
4863 if (repeat_max == 1) goto END_REPEAT;
4864 *code++ = OP_UPTO + repeat_type;
4865 PUT2INC(code, 0, repeat_max - 1);
4866 }
4867 }
4868
4869 /* The case {n,n} is just an EXACT, while the general case {n,m} is
4870 handled as an EXACT followed by an UPTO. */
4871
4872 else
4873 {
4874 *code++ = OP_EXACT + op_type; /* NB EXACT doesn't have repeat_type */
4875 PUT2INC(code, 0, repeat_min);
4876
4877 /* If the maximum is unlimited, insert an OP_STAR. Before doing so,
4878 we have to insert the character for the previous code. For a repeated
4879 Unicode property match, there are two extra bytes that define the
4880 required property. In UTF-8 mode, long characters have their length in
4881 c, with the UTF_LENGTH bit as a flag. */
4882
4883 if (repeat_max < 0)
4884 {
4885 #ifdef SUPPORT_UTF
4886 if (utf && (c & UTF_LENGTH) != 0)
4887 {
4888 memcpy(code, utf_chars, IN_UCHARS(c & 7));
4889 code += c & 7;
4890 }
4891 else
4892 #endif
4893 {
4894 *code++ = c;
4895 if (prop_type >= 0)
4896 {
4897 *code++ = prop_type;
4898 *code++ = prop_value;
4899 }
4900 }
4901 *code++ = OP_STAR + repeat_type;
4902 }
4903
4904 /* Else insert an UPTO if the max is greater than the min, again
4905 preceded by the character, for the previously inserted code. If the
4906 UPTO is just for 1 instance, we can use QUERY instead. */
4907
4908 else if (repeat_max != repeat_min)
4909 {
4910 #ifdef SUPPORT_UTF
4911 if (utf && (c & UTF_LENGTH) != 0)
4912 {
4913 memcpy(code, utf_chars, IN_UCHARS(c & 7));
4914 code += c & 7;
4915 }
4916 else
4917 #endif
4918 *code++ = c;
4919 if (prop_type >= 0)
4920 {
4921 *code++ = prop_type;
4922 *code++ = prop_value;
4923 }
4924 repeat_max -= repeat_min;
4925
4926 if (repeat_max == 1)
4927 {
4928 *code++ = OP_QUERY + repeat_type;
4929 }
4930 else
4931 {
4932 *code++ = OP_UPTO + repeat_type;
4933 PUT2INC(code, 0, repeat_max);
4934 }
4935 }
4936 }
4937
4938 /* The character or character type itself comes last in all cases. */
4939
4940 #ifdef SUPPORT_UTF
4941 if (utf && (c & UTF_LENGTH) != 0)
4942 {
4943 memcpy(code, utf_chars, IN_UCHARS(c & 7));
4944 code += c & 7;
4945 }
4946 else
4947 #endif
4948 *code++ = c;
4949
4950 /* For a repeated Unicode property match, there are two extra bytes that
4951 define the required property. */
4952
4953 #ifdef SUPPORT_UCP
4954 if (prop_type >= 0)
4955 {
4956 *code++ = prop_type;
4957 *code++ = prop_value;
4958 }
4959 #endif
4960 }
4961
4962 /* If previous was a character class or a back reference, we put the repeat
4963 stuff after it, but just skip the item if the repeat was {0,0}. */
4964
4965 else if (*previous == OP_CLASS ||
4966 *previous == OP_NCLASS ||
4967 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
4968 *previous == OP_XCLASS ||
4969 #endif
4970 *previous == OP_REF ||
4971 *previous == OP_REFI)
4972 {
4973 if (repeat_max == 0)
4974 {
4975 code = previous;
4976 goto END_REPEAT;
4977 }
4978
4979 /*--------------------------------------------------------------------*/
4980 /* This code is obsolete from release 8.00; the restriction was finally
4981 removed: */
4982
4983 /* All real repeats make it impossible to handle partial matching (maybe
4984 one day we will be able to remove this restriction). */
4985
4986 /* if (repeat_max != 1) cd->external_flags |= PCRE_NOPARTIAL; */
4987 /*--------------------------------------------------------------------*/
4988
4989 if (repeat_min == 0 && repeat_max == -1)
4990 *code++ = OP_CRSTAR + repeat_type;
4991 else if (repeat_min == 1 && repeat_max == -1)
4992 *code++ = OP_CRPLUS + repeat_type;
4993 else if (repeat_min == 0 && repeat_max == 1)
4994 *code++ = OP_CRQUERY + repeat_type;
4995 else
4996 {
4997 *code++ = OP_CRRANGE + repeat_type;
4998 PUT2INC(code, 0, repeat_min);
4999 if (repeat_max == -1) repeat_max = 0; /* 2-byte encoding for max */
5000 PUT2INC(code, 0, repeat_max);
5001 }
5002 }
5003
5004 /* If previous was a bracket group, we may have to replicate it in certain
5005 cases. Note that at this point we can encounter only the "basic" bracket
5006 opcodes such as BRA and CBRA, as this is the place where they get converted
5007 into the more special varieties such as BRAPOS and SBRA. A test for >=
5008 OP_ASSERT and <= OP_COND includes ASSERT, ASSERT_NOT, ASSERTBACK,
5009 ASSERTBACK_NOT, ONCE, BRA, CBRA, and COND. Originally, PCRE did not allow
5010 repetition of assertions, but now it does, for Perl compatibility. */
5011
5012 else if (*previous >= OP_ASSERT && *previous <= OP_COND)
5013 {
5014 register int i;
5015 int len = (int)(code - previous);
5016 pcre_uchar *bralink = NULL;
5017 pcre_uchar *brazeroptr = NULL;
5018
5019 /* Repeating a DEFINE group is pointless, but Perl allows the syntax, so
5020 we just ignore the repeat. */
5021
5022 if (*previous == OP_COND && previous[LINK_SIZE+1] == OP_DEF)
5023 goto END_REPEAT;
5024
5025 /* There is no sense in actually repeating assertions. The only potential
5026 use of repetition is in cases when the assertion is optional. Therefore,
5027 if the minimum is greater than zero, just ignore the repeat. If the
5028 maximum is not not zero or one, set it to 1. */
5029
5030 if (*previous < OP_ONCE) /* Assertion */
5031 {
5032 if (repeat_min > 0) goto END_REPEAT;
5033 if (repeat_max < 0 || repeat_max > 1) repeat_max = 1;
5034 }
5035
5036 /* The case of a zero minimum is special because of the need to stick
5037 OP_BRAZERO in front of it, and because the group appears once in the
5038 data, whereas in other cases it appears the minimum number of times. For
5039 this reason, it is simplest to treat this case separately, as otherwise
5040 the code gets far too messy. There are several special subcases when the
5041 minimum is zero. */
5042
5043 if (repeat_min == 0)
5044 {
5045 /* If the maximum is also zero, we used to just omit the group from the
5046 output altogether, like this:
5047
5048 ** if (repeat_max == 0)
5049 ** {
5050 ** code = previous;
5051 ** goto END_REPEAT;
5052 ** }
5053
5054 However, that fails when a group or a subgroup within it is referenced
5055 as a subroutine from elsewhere in the pattern, so now we stick in
5056 OP_SKIPZERO in front of it so that it is skipped on execution. As we
5057 don't have a list of which groups are referenced, we cannot do this
5058 selectively.
5059
5060 If the maximum is 1 or unlimited, we just have to stick in the BRAZERO
5061 and do no more at this point. However, we do need to adjust any
5062 OP_RECURSE calls inside the group that refer to the group itself or any
5063 internal or forward referenced group, because the offset is from the
5064 start of the whole regex. Temporarily terminate the pattern while doing
5065 this. */
5066
5067 if (repeat_max <= 1) /* Covers 0, 1, and unlimited */
5068 {
5069 *code = OP_END;
5070 adjust_recurse(previous, 1, utf, cd, save_hwm);
5071 memmove(previous + 1, previous, IN_UCHARS(len));
5072 code++;
5073 if (repeat_max == 0)
5074 {
5075 *previous++ = OP_SKIPZERO;
5076 goto END_REPEAT;
5077 }
5078 brazeroptr = previous; /* Save for possessive optimizing */
5079 *previous++ = OP_BRAZERO + repeat_type;
5080 }
5081
5082 /* If the maximum is greater than 1 and limited, we have to replicate
5083 in a nested fashion, sticking OP_BRAZERO before each set of brackets.
5084 The first one has to be handled carefully because it's the original
5085 copy, which has to be moved up. The remainder can be handled by code
5086 that is common with the non-zero minimum case below. We have to
5087 adjust the value or repeat_max, since one less copy is required. Once
5088 again, we may have to adjust any OP_RECURSE calls inside the group. */
5089
5090 else
5091 {
5092 int offset;
5093 *code = OP_END;
5094 adjust_recurse(previous, 2 + LINK_SIZE, utf, cd, save_hwm);
5095 memmove(previous + 2 + LINK_SIZE, previous, IN_UCHARS(len));
5096 code += 2 + LINK_SIZE;
5097 *previous++ = OP_BRAZERO + repeat_type;
5098 *previous++ = OP_BRA;
5099
5100 /* We chain together the bracket offset fields that have to be
5101 filled in later when the ends of the brackets are reached. */
5102
5103 offset = (bralink == NULL)? 0 : (int)(previous - bralink);
5104 bralink = previous;
5105 PUTINC(previous, 0, offset);
5106 }
5107
5108 repeat_max--;
5109 }
5110
5111 /* If the minimum is greater than zero, replicate the group as many
5112 times as necessary, and adjust the maximum to the number of subsequent
5113 copies that we need. If we set a first char from the group, and didn't
5114 set a required char, copy the latter from the former. If there are any
5115 forward reference subroutine calls in the group, there will be entries on
5116 the workspace list; replicate these with an appropriate increment. */
5117
5118 else
5119 {
5120 if (repeat_min > 1)
5121 {
5122 /* In the pre-compile phase, we don't actually do the replication. We
5123 just adjust the length as if we had. Do some paranoid checks for
5124 potential integer overflow. The INT64_OR_DOUBLE type is a 64-bit
5125 integer type when available, otherwise double. */
5126
5127 if (lengthptr != NULL)
5128 {
5129 int delta = (repeat_min - 1)*length_prevgroup;
5130 if ((INT64_OR_DOUBLE)(repeat_min - 1)*
5131 (INT64_OR_DOUBLE)length_prevgroup >
5132 (INT64_OR_DOUBLE)INT_MAX ||
5133 OFLOW_MAX - *lengthptr < delta)
5134 {
5135 *errorcodeptr = ERR20;
5136 goto FAILED;
5137 }
5138 *lengthptr += delta;
5139 }
5140
5141 /* This is compiling for real. If there is a set first byte for
5142 the group, and we have not yet set a "required byte", set it. Make
5143 sure there is enough workspace for copying forward references before
5144 doing the copy. */
5145
5146 else
5147 {
5148 if (groupsetfirstchar && reqchar < 0) reqchar = firstchar;
5149
5150 for (i = 1; i < repeat_min; i++)
5151 {
5152 pcre_uchar *hc;
5153 pcre_uchar *this_hwm = cd->hwm;
5154 memcpy(code, previous, IN_UCHARS(len));
5155
5156 while (cd->hwm > cd->start_workspace + cd->workspace_size -
5157 WORK_SIZE_SAFETY_MARGIN - (this_hwm - save_hwm))
5158 {
5159 int save_offset = save_hwm - cd->start_workspace;
5160 int this_offset = this_hwm - cd->start_workspace;
5161 *errorcodeptr = expand_workspace(cd);
5162 if (*errorcodeptr != 0) goto FAILED;
5163 save_hwm = (pcre_uchar *)cd->start_workspace + save_offset;
5164 this_hwm = (pcre_uchar *)cd->start_workspace + this_offset;
5165 }
5166
5167 for (hc = save_hwm; hc < this_hwm; hc += LINK_SIZE)
5168 {
5169 PUT(cd->hwm, 0, GET(hc, 0) + len);
5170 cd->hwm += LINK_SIZE;
5171 }
5172 save_hwm = this_hwm;
5173 code += len;
5174 }
5175 }
5176 }
5177
5178 if (repeat_max > 0) repeat_max -= repeat_min;
5179 }
5180
5181 /* This code is common to both the zero and non-zero minimum cases. If
5182 the maximum is limited, it replicates the group in a nested fashion,
5183 remembering the bracket starts on a stack. In the case of a zero minimum,
5184 the first one was set up above. In all cases the repeat_max now specifies
5185 the number of additional copies needed. Again, we must remember to
5186 replicate entries on the forward reference list. */
5187
5188 if (repeat_max >= 0)
5189 {
5190 /* In the pre-compile phase, we don't actually do the replication. We
5191 just adjust the length as if we had. For each repetition we must add 1
5192 to the length for BRAZERO and for all but the last repetition we must
5193 add 2 + 2*LINKSIZE to allow for the nesting that occurs. Do some
5194 paranoid checks to avoid integer overflow. The INT64_OR_DOUBLE type is
5195 a 64-bit integer type when available, otherwise double. */
5196
5197 if (lengthptr != NULL && repeat_max > 0)
5198 {
5199 int delta = repeat_max * (length_prevgroup + 1 + 2 + 2*LINK_SIZE) -
5200 2 - 2*LINK_SIZE; /* Last one doesn't nest */
5201 if ((INT64_OR_DOUBLE)repeat_max *
5202 (INT64_OR_DOUBLE)(length_prevgroup + 1 + 2 + 2*LINK_SIZE)
5203 > (INT64_OR_DOUBLE)INT_MAX ||
5204 OFLOW_MAX - *lengthptr < delta)
5205 {
5206 *errorcodeptr = ERR20;
5207 goto FAILED;
5208 }
5209 *lengthptr += delta;
5210 }
5211
5212 /* This is compiling for real */
5213
5214 else for (i = repeat_max - 1; i >= 0; i--)
5215 {
5216 pcre_uchar *hc;
5217 pcre_uchar *this_hwm = cd->hwm;
5218
5219 *code++ = OP_BRAZERO + repeat_type;
5220
5221 /* All but the final copy start a new nesting, maintaining the
5222 chain of brackets outstanding. */
5223
5224 if (i != 0)
5225 {
5226 int offset;
5227 *code++ = OP_BRA;
5228 offset = (bralink == NULL)? 0 : (int)(code - bralink);
5229 bralink = code;
5230 PUTINC(code, 0, offset);
5231 }
5232
5233 memcpy(code, previous, IN_UCHARS(len));
5234
5235 /* Ensure there is enough workspace for forward references before
5236 copying them. */
5237
5238 while (cd->hwm > cd->start_workspace + cd->workspace_size -
5239 WORK_SIZE_SAFETY_MARGIN - (this_hwm - save_hwm))
5240 {
5241 int save_offset = save_hwm - cd->start_workspace;
5242 int this_offset = this_hwm - cd->start_workspace;
5243 *errorcodeptr = expand_workspace(cd);
5244 if (*errorcodeptr != 0) goto FAILED;
5245 save_hwm = (pcre_uchar *)cd->start_workspace + save_offset;
5246 this_hwm = (pcre_uchar *)cd->start_workspace + this_offset;
5247 }
5248
5249 for (hc = save_hwm; hc < this_hwm; hc += LINK_SIZE)
5250 {
5251 PUT(cd->hwm, 0, GET(hc, 0) + len + ((i != 0)? 2+LINK_SIZE : 1));
5252 cd->hwm += LINK_SIZE;
5253 }
5254 save_hwm = this_hwm;
5255 code += len;
5256 }
5257
5258 /* Now chain through the pending brackets, and fill in their length
5259 fields (which are holding the chain links pro tem). */
5260
5261 while (bralink != NULL)
5262 {
5263 int oldlinkoffset;
5264 int offset = (int)(code - bralink + 1);
5265 pcre_uchar *bra = code - offset;
5266 oldlinkoffset = GET(bra, 1);
5267 bralink = (oldlinkoffset == 0)? NULL : bralink - oldlinkoffset;
5268 *code++ = OP_KET;
5269 PUTINC(code, 0, offset);
5270 PUT(bra, 1, offset);
5271 }
5272 }
5273
5274 /* If the maximum is unlimited, set a repeater in the final copy. For
5275 ONCE brackets, that's all we need to do. However, possessively repeated
5276 ONCE brackets can be converted into non-capturing brackets, as the
5277 behaviour of (?:xx)++ is the same as (?>xx)++ and this saves having to
5278 deal with possessive ONCEs specially.
5279
5280 Otherwise, when we are doing the actual compile phase, check to see
5281 whether this group is one that could match an empty string. If so,
5282 convert the initial operator to the S form (e.g. OP_BRA -> OP_SBRA) so
5283 that runtime checking can be done. [This check is also applied to ONCE
5284 groups at runtime, but in a different way.]
5285
5286 Then, if the quantifier was possessive and the bracket is not a
5287 conditional, we convert the BRA code to the POS form, and the KET code to
5288 KETRPOS. (It turns out to be convenient at runtime to detect this kind of
5289 subpattern at both the start and at the end.) The use of special opcodes
5290 makes it possible to reduce greatly the stack usage in pcre_exec(). If
5291 the group is preceded by OP_BRAZERO, convert this to OP_BRAPOSZERO.
5292
5293 Then, if the minimum number of matches is 1 or 0, cancel the possessive
5294 flag so that the default action below, of wrapping everything inside
5295 atomic brackets, does not happen. When the minimum is greater than 1,
5296 there will be earlier copies of the group, and so we still have to wrap
5297 the whole thing. */
5298
5299 else
5300 {
5301 pcre_uchar *ketcode = code - 1 - LINK_SIZE;
5302 pcre_uchar *bracode = ketcode - GET(ketcode, 1);
5303
5304 /* Convert possessive ONCE brackets to non-capturing */
5305
5306 if ((*bracode == OP_ONCE || *bracode == OP_ONCE_NC) &&
5307 possessive_quantifier) *bracode = OP_BRA;
5308
5309 /* For non-possessive ONCE brackets, all we need to do is to
5310 set the KET. */
5311
5312 if (*bracode == OP_ONCE || *bracode == OP_ONCE_NC)
5313 *ketcode = OP_KETRMAX + repeat_type;
5314
5315 /* Handle non-ONCE brackets and possessive ONCEs (which have been
5316 converted to non-capturing above). */
5317
5318 else
5319 {
5320 /* In the compile phase, check for empty string matching. */
5321
5322 if (lengthptr == NULL)
5323 {
5324 pcre_uchar *scode = bracode;
5325 do
5326 {
5327 if (could_be_empty_branch(scode, ketcode, utf, cd))
5328 {
5329 *bracode += OP_SBRA - OP_BRA;
5330 break;
5331 }
5332 scode += GET(scode, 1);
5333 }
5334 while (*scode == OP_ALT);
5335 }
5336
5337 /* Handle possessive quantifiers. */
5338
5339 if (possessive_quantifier)
5340 {
5341 /* For COND brackets, we wrap the whole thing in a possessively
5342 repeated non-capturing bracket, because we have not invented POS
5343 versions of the COND opcodes. Because we are moving code along, we
5344 must ensure that any pending recursive references are updated. */
5345
5346 if (*bracode == OP_COND || *bracode == OP_SCOND)
5347 {
5348 int nlen = (int)(code - bracode);
5349 *code = OP_END;
5350 adjust_recurse(bracode, 1 + LINK_SIZE, utf, cd, save_hwm);
5351 memmove(bracode + 1 + LINK_SIZE, bracode, IN_UCHARS(nlen));
5352 code += 1 + LINK_SIZE;
5353 nlen += 1 + LINK_SIZE;
5354 *bracode = OP_BRAPOS;
5355 *code++ = OP_KETRPOS;
5356 PUTINC(code, 0, nlen);
5357 PUT(bracode, 1, nlen);
5358 }
5359
5360 /* For non-COND brackets, we modify the BRA code and use KETRPOS. */
5361
5362 else
5363 {
5364 *bracode += 1; /* Switch to xxxPOS opcodes */
5365 *ketcode = OP_KETRPOS;
5366 }
5367
5368 /* If the minimum is zero, mark it as possessive, then unset the
5369 possessive flag when the minimum is 0 or 1. */
5370
5371 if (brazeroptr != NULL) *brazeroptr = OP_BRAPOSZERO;
5372 if (repeat_min < 2) possessive_quantifier = FALSE;
5373 }
5374
5375 /* Non-possessive quantifier */
5376
5377 else *ketcode = OP_KETRMAX + repeat_type;
5378 }
5379 }
5380 }
5381
5382 /* If previous is OP_FAIL, it was generated by an empty class [] in
5383 JavaScript mode. The other ways in which OP_FAIL can be generated, that is
5384 by (*FAIL) or (?!) set previous to NULL, which gives a "nothing to repeat"
5385 error above. We can just ignore the repeat in JS case. */
5386
5387 else if (*previous == OP_FAIL) goto END_REPEAT;
5388
5389 /* Else there's some kind of shambles */
5390
5391 else
5392 {
5393 *errorcodeptr = ERR11;
5394 goto FAILED;
5395 }
5396
5397 /* If the character following a repeat is '+', or if certain optimization
5398 tests above succeeded, possessive_quantifier is TRUE. For some opcodes,
5399 there are special alternative opcodes for this case. For anything else, we
5400 wrap the entire repeated item inside OP_ONCE brackets. Logically, the '+'
5401 notation is just syntactic sugar, taken from Sun's Java package, but the
5402 special opcodes can optimize it.
5403
5404 Some (but not all) possessively repeated subpatterns have already been
5405 completely handled in the code just above. For them, possessive_quantifier
5406 is always FALSE at this stage.
5407
5408 Note that the repeated item starts at tempcode, not at previous, which
5409 might be the first part of a string whose (former) last char we repeated.
5410
5411 Possessifying an 'exact' quantifier has no effect, so we can ignore it. But
5412 an 'upto' may follow. We skip over an 'exact' item, and then test the
5413 length of what remains before proceeding. */
5414
5415 if (possessive_quantifier)
5416 {
5417 int len;
5418
5419 if (*tempcode == OP_TYPEEXACT)
5420 tempcode += PRIV(OP_lengths)[*tempcode] +
5421 ((tempcode[1 + IMM2_SIZE] == OP_PROP
5422 || tempcode[1 + IMM2_SIZE] == OP_NOTPROP)? 2 : 0);
5423
5424 else if (*tempcode == OP_EXACT || *tempcode == OP_NOTEXACT)
5425 {
5426 tempcode += PRIV(OP_lengths)[*tempcode];
5427 #ifdef SUPPORT_UTF
5428 if (utf && HAS_EXTRALEN(tempcode[-1]))
5429 tempcode += GET_EXTRALEN(tempcode[-1]);
5430 #endif
5431 }
5432
5433 len = (int)(code - tempcode);
5434 if (len > 0) switch (*tempcode)
5435 {
5436 case OP_STAR: *tempcode = OP_POSSTAR; break;
5437 case OP_PLUS: *tempcode = OP_POSPLUS; break;
5438 case OP_QUERY: *tempcode = OP_POSQUERY; break;
5439 case OP_UPTO: *tempcode = OP_POSUPTO; break;
5440
5441 case OP_STARI: *tempcode = OP_POSSTARI; break;
5442 case OP_PLUSI: *tempcode = OP_POSPLUSI; break;
5443 case OP_QUERYI: *tempcode = OP_POSQUERYI; break;
5444 case OP_UPTOI: *tempcode = OP_POSUPTOI; break;
5445
5446 case OP_NOTSTAR: *tempcode = OP_NOTPOSSTAR; break;
5447 case OP_NOTPLUS: *tempcode = OP_NOTPOSPLUS; break;
5448 case OP_NOTQUERY: *tempcode = OP_NOTPOSQUERY; break;
5449 case OP_NOTUPTO: *tempcode = OP_NOTPOSUPTO; break;
5450
5451 case OP_NOTSTARI: *tempcode = OP_NOTPOSSTARI; break;
5452 case OP_NOTPLUSI: *tempcode = OP_NOTPOSPLUSI; break;
5453 case OP_NOTQUERYI: *tempcode = OP_NOTPOSQUERYI; break;
5454 case OP_NOTUPTOI: *tempcode = OP_NOTPOSUPTOI; break;
5455
5456 case OP_TYPESTAR: *tempcode = OP_TYPEPOSSTAR; break;
5457 case OP_TYPEPLUS: *tempcode = OP_TYPEPOSPLUS; break;
5458 case OP_TYPEQUERY: *tempcode = OP_TYPEPOSQUERY; break;
5459 case OP_TYPEUPTO: *tempcode = OP_TYPEPOSUPTO; break;
5460
5461 /* Because we are moving code along, we must ensure that any
5462 pending recursive references are updated. */
5463
5464 default:
5465 *code = OP_END;
5466 adjust_recurse(tempcode, 1 + LINK_SIZE, utf, cd, save_hwm);
5467 memmove(tempcode + 1 + LINK_SIZE, tempcode, IN_UCHARS(len));
5468 code += 1 + LINK_SIZE;
5469 len += 1 + LINK_SIZE;
5470 tempcode[0] = OP_ONCE;
5471 *code++ = OP_KET;
5472 PUTINC(code, 0, len);
5473 PUT(tempcode, 1, len);
5474 break;
5475 }
5476 }
5477
5478 /* In all case we no longer have a previous item. We also set the
5479 "follows varying string" flag for subsequently encountered reqchars if
5480 it isn't already set and we have just passed a varying length item. */
5481
5482 END_REPEAT:
5483 previous = NULL;
5484 cd->req_varyopt |= reqvary;
5485 break;
5486
5487
5488 /* ===================================================================*/
5489 /* Start of nested parenthesized sub-expression, or comment or lookahead or
5490 lookbehind or option setting or condition or all the other extended
5491 parenthesis forms. */
5492
5493 case CHAR_LEFT_PARENTHESIS:
5494 newoptions = options;
5495 skipbytes = 0;
5496 bravalue = OP_CBRA;
5497 save_hwm = cd->hwm;
5498 reset_bracount = FALSE;
5499
5500 /* First deal with various "verbs" that can be introduced by '*'. */
5501
5502 ptr++;
5503 if (ptr[0] == CHAR_ASTERISK && (ptr[1] == ':'
5504 || (MAX_255(ptr[1]) && ((cd->ctypes[ptr[1]] & ctype_letter) != 0))))
5505 {
5506 int i, namelen;
5507 int arglen = 0;
5508 const char *vn = verbnames;
5509 const pcre_uchar *name = ptr + 1;
5510 const pcre_uchar *arg = NULL;
5511 previous = NULL;
5512 ptr++;
5513 while (MAX_255(*ptr) && (cd->ctypes[*ptr] & ctype_letter) != 0) ptr++;
5514 namelen = (int)(ptr - name);
5515
5516 /* It appears that Perl allows any characters whatsoever, other than
5517 a closing parenthesis, to appear in arguments, so we no longer insist on
5518 letters, digits, and underscores. */
5519
5520 if (*ptr == CHAR_COLON)
5521 {
5522 arg = ++ptr;
5523 while (*ptr != 0 && *ptr != CHAR_RIGHT_PARENTHESIS) ptr++;
5524 arglen = (int)(ptr - arg);
5525 }
5526
5527 if (*ptr != CHAR_RIGHT_PARENTHESIS)
5528 {
5529 *errorcodeptr = ERR60;
5530 goto FAILED;
5531 }
5532
5533 /* Scan the table of verb names */
5534
5535 for (i = 0; i < verbcount; i++)
5536 {
5537 if (namelen == verbs[i].len &&
5538 STRNCMP_UC_C8(name, vn, namelen) == 0)
5539 {
5540 /* Check for open captures before ACCEPT and convert it to
5541 ASSERT_ACCEPT if in an assertion. */
5542
5543 if (verbs[i].op == OP_ACCEPT)
5544 {
5545 open_capitem *oc;
5546 if (arglen != 0)
5547 {
5548 *errorcodeptr = ERR59;
5549 goto FAILED;
5550 }
5551 cd->had_accept = TRUE;
5552 for (oc = cd->open_caps; oc != NULL; oc = oc->next)
5553 {
5554 *code++ = OP_CLOSE;
5555 PUT2INC(code, 0, oc->number);
5556 }
5557 *code++ = (cd->assert_depth > 0)? OP_ASSERT_ACCEPT : OP_ACCEPT;
5558
5559 /* Do not set firstchar after *ACCEPT */
5560 if (firstchar == REQ_UNSET) firstchar = REQ_NONE;
5561 }
5562
5563 /* Handle other cases with/without an argument */
5564
5565 else if (arglen == 0)
5566 {
5567 if (verbs[i].op < 0) /* Argument is mandatory */
5568 {
5569 *errorcodeptr = ERR66;
5570 goto FAILED;
5571 }
5572 *code = verbs[i].op;
5573 if (*code++ == OP_THEN) cd->external_flags |= PCRE_HASTHEN;
5574 }
5575
5576 else
5577 {
5578 if (verbs[i].op_arg < 0) /* Argument is forbidden */
5579 {
5580 *errorcodeptr = ERR59;
5581 goto FAILED;
5582 }
5583 *code = verbs[i].op_arg;
5584 if (*code++ == OP_THEN_ARG) cd->external_flags |= PCRE_HASTHEN;
5585 *code++ = arglen;
5586 memcpy(code, arg, IN_UCHARS(arglen));
5587 code += arglen;
5588 *code++ = 0;
5589 }
5590
5591 break; /* Found verb, exit loop */
5592 }
5593
5594 vn += verbs[i].len + 1;
5595 }
5596
5597 if (i < verbcount) continue; /* Successfully handled a verb */
5598 *errorcodeptr = ERR60; /* Verb not recognized */
5599 goto FAILED;
5600 }
5601
5602 /* Deal with the extended parentheses; all are introduced by '?', and the
5603 appearance of any of them means that this is not a capturing group. */
5604
5605 else if (*ptr == CHAR_QUESTION_MARK)
5606 {
5607 int i, set, unset, namelen;
5608 int *optset;
5609 const pcre_uchar *name;
5610 pcre_uchar *slot;
5611
5612 switch (*(++ptr))
5613 {
5614 case CHAR_NUMBER_SIGN: /* Comment; skip to ket */
5615 ptr++;
5616 while (*ptr != 0 && *ptr != CHAR_RIGHT_PARENTHESIS) ptr++;
5617 if (*ptr == 0)
5618 {
5619 *errorcodeptr = ERR18;
5620 goto FAILED;
5621 }
5622 continue;
5623
5624
5625 /* ------------------------------------------------------------ */
5626 case CHAR_VERTICAL_LINE: /* Reset capture count for each branch */
5627 reset_bracount = TRUE;
5628 /* Fall through */
5629
5630 /* ------------------------------------------------------------ */
5631 case CHAR_COLON: /* Non-capturing bracket */
5632 bravalue = OP_BRA;
5633 ptr++;
5634 break;
5635
5636
5637 /* ------------------------------------------------------------ */
5638 case CHAR_LEFT_PARENTHESIS:
5639 bravalue = OP_COND; /* Conditional group */
5640
5641 /* A condition can be an assertion, a number (referring to a numbered
5642 group), a name (referring to a named group), or 'R', referring to
5643 recursion. R<digits> and R&name are also permitted for recursion tests.
5644
5645 There are several syntaxes for testing a named group: (?(name)) is used
5646 by Python; Perl 5.10 onwards uses (?(<name>) or (?('name')).
5647
5648 There are two unfortunate ambiguities, caused by history. (a) 'R' can
5649 be the recursive thing or the name 'R' (and similarly for 'R' followed
5650 by digits), and (b) a number could be a name that consists of digits.
5651 In both cases, we look for a name first; if not found, we try the other
5652 cases. */
5653
5654 /* For conditions that are assertions, check the syntax, and then exit
5655 the switch. This will take control down to where bracketed groups,
5656 including assertions, are processed. */
5657
5658 if (ptr[1] == CHAR_QUESTION_MARK && (ptr[2] == CHAR_EQUALS_SIGN ||
5659 ptr[2] == CHAR_EXCLAMATION_MARK || ptr[2] == CHAR_LESS_THAN_SIGN))
5660 break;
5661
5662 /* Most other conditions use OP_CREF (a couple change to OP_RREF
5663 below), and all need to skip 1+IMM2_SIZE bytes at the start of the group. */
5664
5665 code[1+LINK_SIZE] = OP_CREF;
5666 skipbytes = 1+IMM2_SIZE;
5667 refsign = -1;
5668
5669 /* Check for a test for recursion in a named group. */
5670
5671 if (ptr[1] == CHAR_R && ptr[2] == CHAR_AMPERSAND)
5672 {
5673 terminator = -1;
5674 ptr += 2;
5675 code[1+LINK_SIZE] = OP_RREF; /* Change the type of test */
5676 }
5677
5678 /* Check for a test for a named group's having been set, using the Perl
5679 syntax (?(<name>) or (?('name') */
5680
5681 else if (ptr[1] == CHAR_LESS_THAN_SIGN)
5682 {
5683 terminator = CHAR_GREATER_THAN_SIGN;
5684 ptr++;
5685 }
5686 else if (ptr[1] == CHAR_APOSTROPHE)
5687 {
5688 terminator = CHAR_APOSTROPHE;
5689 ptr++;
5690 }
5691 else
5692 {
5693 terminator = 0;
5694 if (ptr[1] == CHAR_MINUS || ptr[1] == CHAR_PLUS) refsign = *(++ptr);
5695 }
5696
5697 /* We now expect to read a name; any thing else is an error */
5698
5699 if (!MAX_255(ptr[1]) || (cd->ctypes[ptr[1]] & ctype_word) == 0)
5700 {
5701 ptr += 1; /* To get the right offset */
5702 *errorcodeptr = ERR28;
5703 goto FAILED;
5704 }
5705
5706 /* Read the name, but also get it as a number if it's all digits */
5707
5708 recno = 0;
5709 name = ++ptr;
5710 while (MAX_255(*ptr) && (cd->ctypes[*ptr] & ctype_word) != 0)
5711 {
5712 if (recno >= 0)
5713 recno = (IS_DIGIT(*ptr))? recno * 10 + *ptr - CHAR_0 : -1;
5714 ptr++;
5715 }
5716 namelen = (int)(ptr - name);
5717
5718 if ((terminator > 0 && *ptr++ != terminator) ||
5719 *ptr++ != CHAR_RIGHT_PARENTHESIS)
5720 {
5721 ptr--; /* Error offset */
5722 *errorcodeptr = ERR26;
5723 goto FAILED;
5724 }
5725
5726 /* Do no further checking in the pre-compile phase. */
5727
5728 if (lengthptr != NULL) break;
5729
5730 /* In the real compile we do the work of looking for the actual
5731 reference. If the string started with "+" or "-" we require the rest to
5732 be digits, in which case recno will be set. */
5733
5734 if (refsign > 0)
5735 {
5736 if (recno <= 0)
5737 {
5738 *errorcodeptr = ERR58;
5739 goto FAILED;
5740 }
5741 recno = (refsign == CHAR_MINUS)?
5742 cd->bracount - recno + 1 : recno +cd->bracount;
5743 if (recno <= 0 || recno > cd->final_bracount)
5744 {
5745 *errorcodeptr = ERR15;
5746 goto FAILED;
5747 }
5748 PUT2(code, 2+LINK_SIZE, recno);
5749 break;
5750 }
5751
5752 /* Otherwise (did not start with "+" or "-"), start by looking for the
5753 name. If we find a name, add one to the opcode to change OP_CREF or
5754 OP_RREF into OP_NCREF or OP_NRREF. These behave exactly the same,
5755 except they record that the reference was originally to a name. The
5756 information is used to check duplicate names. */
5757
5758 slot = cd->name_table;
5759 for (i = 0; i < cd->names_found; i++)
5760 {
5761 if (STRNCMP_UC_UC(name, slot+IMM2_SIZE, namelen) == 0) break;
5762 slot += cd->name_entry_size;
5763 }
5764
5765 /* Found a previous named subpattern */
5766
5767 if (i < cd->names_found)
5768 {
5769 recno = GET2(slot, 0);
5770 PUT2(code, 2+LINK_SIZE, recno);
5771 code[1+LINK_SIZE]++;
5772 }
5773
5774 /* Search the pattern for a forward reference */
5775
5776 else if ((i = find_parens(cd, name, namelen,
5777 (options & PCRE_EXTENDED) != 0, utf)) > 0)
5778 {
5779 PUT2(code, 2+LINK_SIZE, i);
5780 code[1+LINK_SIZE]++;
5781 }
5782
5783 /* If terminator == 0 it means that the name followed directly after
5784 the opening parenthesis [e.g. (?(abc)...] and in this case there are
5785 some further alternatives to try. For the cases where terminator != 0
5786 [things like (?(<name>... or (?('name')... or (?(R&name)... ] we have
5787 now checked all the possibilities, so give an error. */
5788
5789 else if (terminator != 0)
5790 {
5791 *errorcodeptr = ERR15;
5792 goto FAILED;
5793 }
5794
5795 /* Check for (?(R) for recursion. Allow digits after R to specify a
5796 specific group number. */
5797
5798 else if (*name == CHAR_R)
5799 {
5800 recno = 0;
5801 for (i = 1; i < namelen; i++)
5802 {
5803 if (!IS_DIGIT(name[i]))
5804 {
5805 *errorcodeptr = ERR15;
5806 goto FAILED;
5807 }
5808 recno = recno * 10 + name[i] - CHAR_0;
5809 }
5810 if (recno == 0) recno = RREF_ANY;
5811 code[1+LINK_SIZE] = OP_RREF; /* Change test type */
5812 PUT2(code, 2+LINK_SIZE, recno);
5813 }
5814
5815 /* Similarly, check for the (?(DEFINE) "condition", which is always
5816 false. */
5817
5818 else if (namelen == 6 && STRNCMP_UC_C8(name, STRING_DEFINE, 6) == 0)
5819 {
5820 code[1+LINK_SIZE] = OP_DEF;
5821 skipbytes = 1;
5822 }
5823
5824 /* Check for the "name" actually being a subpattern number. We are
5825 in the second pass here, so final_bracount is set. */
5826
5827 else if (recno > 0 && recno <= cd->final_bracount)
5828 {
5829 PUT2(code, 2+LINK_SIZE, recno);
5830 }
5831
5832 /* Either an unidentified subpattern, or a reference to (?(0) */
5833
5834 else
5835 {
5836 *errorcodeptr = (recno == 0)? ERR35: ERR15;
5837 goto FAILED;
5838 }
5839 break;
5840
5841
5842 /* ------------------------------------------------------------ */
5843 case CHAR_EQUALS_SIGN: /* Positive lookahead */
5844 bravalue = OP_ASSERT;
5845 cd->assert_depth += 1;
5846 ptr++;
5847 break;
5848
5849
5850 /* ------------------------------------------------------------ */
5851 case CHAR_EXCLAMATION_MARK: /* Negative lookahead */
5852 ptr++;
5853 if (*ptr == CHAR_RIGHT_PARENTHESIS) /* Optimize (?!) */
5854 {
5855 *code++ = OP_FAIL;
5856 previous = NULL;
5857 continue;
5858 }
5859 bravalue = OP_ASSERT_NOT;
5860 cd->assert_depth += 1;
5861 break;
5862
5863
5864 /* ------------------------------------------------------------ */
5865 case CHAR_LESS_THAN_SIGN: /* Lookbehind or named define */
5866 switch (ptr[1])
5867 {
5868 case CHAR_EQUALS_SIGN: /* Positive lookbehind */
5869 bravalue = OP_ASSERTBACK;
5870 cd->assert_depth += 1;
5871 ptr += 2;
5872 break;
5873
5874 case CHAR_EXCLAMATION_MARK: /* Negative lookbehind */
5875 bravalue = OP_ASSERTBACK_NOT;
5876 cd->assert_depth += 1;
5877 ptr += 2;
5878 break;
5879
5880 default: /* Could be name define, else bad */
5881 if (MAX_255(ptr[1]) && (cd->ctypes[ptr[1]] & ctype_word) != 0)
5882 goto DEFINE_NAME;
5883 ptr++; /* Correct offset for error */
5884 *errorcodeptr = ERR24;
5885 goto FAILED;
5886 }
5887 break;
5888
5889
5890 /* ------------------------------------------------------------ */
5891 case CHAR_GREATER_THAN_SIGN: /* One-time brackets */
5892 bravalue = OP_ONCE;
5893 ptr++;
5894 break;
5895
5896
5897 /* ------------------------------------------------------------ */
5898 case CHAR_C: /* Callout - may be followed by digits; */
5899 previous_callout = code; /* Save for later completion */
5900 after_manual_callout = 1; /* Skip one item before completing */
5901 *code++ = OP_CALLOUT;
5902 {
5903 int n = 0;
5904 ptr++;
5905 while(IS_DIGIT(*ptr))
5906 n = n * 10 + *ptr++ - CHAR_0;
5907 if (*ptr != CHAR_RIGHT_PARENTHESIS)
5908 {
5909 *errorcodeptr = ERR39;
5910 goto FAILED;
5911 }
5912 if (n > 255)
5913 {
5914 *errorcodeptr = ERR38;
5915 goto FAILED;
5916 }
5917 *code++ = n;
5918 PUT(code, 0, (int)(ptr - cd->start_pattern + 1)); /* Pattern offset */
5919 PUT(code, LINK_SIZE, 0); /* Default length */
5920 code += 2 * LINK_SIZE;
5921 }
5922 previous = NULL;
5923 continue;
5924
5925
5926 /* ------------------------------------------------------------ */
5927 case CHAR_P: /* Python-style named subpattern handling */
5928 if (*(++ptr) == CHAR_EQUALS_SIGN ||
5929 *ptr == CHAR_GREATER_THAN_SIGN) /* Reference or recursion */
5930 {
5931 is_recurse = *ptr == CHAR_GREATER_THAN_SIGN;
5932 terminator = CHAR_RIGHT_PARENTHESIS;
5933 goto NAMED_REF_OR_RECURSE;
5934 }
5935 else if (*ptr != CHAR_LESS_THAN_SIGN) /* Test for Python-style defn */
5936 {
5937 *errorcodeptr = ERR41;
5938 goto FAILED;
5939 }
5940 /* Fall through to handle (?P< as (?< is handled */
5941
5942
5943 /* ------------------------------------------------------------ */
5944 DEFINE_NAME: /* Come here from (?< handling */
5945 case CHAR_APOSTROPHE:
5946 {
5947 terminator = (*ptr == CHAR_LESS_THAN_SIGN)?
5948 CHAR_GREATER_THAN_SIGN : CHAR_APOSTROPHE;
5949 name = ++ptr;
5950
5951 while (MAX_255(*ptr) && (cd->ctypes[*ptr] & ctype_word) != 0) ptr++;
5952 namelen = (int)(ptr - name);
5953
5954 /* In the pre-compile phase, just do a syntax check. */
5955
5956 if (lengthptr != NULL)
5957 {
5958 if (*ptr != terminator)
5959 {
5960 *errorcodeptr = ERR42;
5961 goto FAILED;
5962 }
5963 if (cd->names_found >= MAX_NAME_COUNT)
5964 {
5965 *errorcodeptr = ERR49;
5966 goto FAILED;
5967 }
5968 if (namelen + IMM2_SIZE + 1 > cd->name_entry_size)
5969 {
5970 cd->name_entry_size = namelen + IMM2_SIZE + 1;
5971 if (namelen > MAX_NAME_SIZE)
5972 {
5973 *errorcodeptr = ERR48;
5974 goto FAILED;
5975 }
5976 }
5977 }
5978
5979 /* In the real compile, create the entry in the table, maintaining
5980 alphabetical order. Duplicate names for different numbers are
5981 permitted only if PCRE_DUPNAMES is set. Duplicate names for the same
5982 number are always OK. (An existing number can be re-used if (?|
5983 appears in the pattern.) In either event, a duplicate name results in
5984 a duplicate entry in the table, even if the number is the same. This
5985 is because the number of names, and hence the table size, is computed
5986 in the pre-compile, and it affects various numbers and pointers which
5987 would all have to be modified, and the compiled code moved down, if
5988 duplicates with the same number were omitted from the table. This
5989 doesn't seem worth the hassle. However, *different* names for the
5990 same number are not permitted. */
5991
5992 else
5993 {
5994 BOOL dupname = FALSE;
5995 slot = cd->name_table;
5996
5997 for (i = 0; i < cd->names_found; i++)
5998 {
5999 int crc = memcmp(name, slot+IMM2_SIZE, IN_UCHARS(namelen));
6000 if (crc == 0)
6001 {
6002 if (slot[IMM2_SIZE+namelen] == 0)
6003 {
6004 if (GET2(slot, 0) != cd->bracount + 1 &&
6005 (options & PCRE_DUPNAMES) == 0)
6006 {
6007 *errorcodeptr = ERR43;
6008 goto FAILED;
6009 }
6010 else dupname = TRUE;
6011 }
6012 else crc = -1; /* Current name is a substring */
6013 }
6014
6015 /* Make space in the table and break the loop for an earlier
6016 name. For a duplicate or later name, carry on. We do this for
6017 duplicates so that in the simple case (when ?(| is not used) they
6018 are in order of their numbers. */
6019
6020 if (crc < 0)
6021 {
6022 memmove(slot + cd->name_entry_size, slot,
6023 IN_UCHARS((cd->names_found - i) * cd->name_entry_size));
6024 break;
6025 }
6026
6027 /* Continue the loop for a later or duplicate name */
6028
6029 slot += cd->name_entry_size;
6030 }
6031
6032 /* For non-duplicate names, check for a duplicate number before
6033 adding the new name. */
6034
6035 if (!dupname)
6036 {
6037 pcre_uchar *cslot = cd->name_table;
6038 for (i = 0; i < cd->names_found; i++)
6039 {
6040 if (cslot != slot)
6041 {
6042 if (GET2(cslot, 0) == cd->bracount + 1)
6043 {
6044 *errorcodeptr = ERR65;
6045 goto FAILED;
6046 }
6047 }
6048 else i--;
6049 cslot += cd->name_entry_size;
6050 }
6051 }
6052
6053 PUT2(slot, 0, cd->bracount + 1);
6054 memcpy(slot + IMM2_SIZE, name, IN_UCHARS(namelen));
6055 slot[IMM2_SIZE + namelen] = 0;
6056 }
6057 }
6058
6059 /* In both pre-compile and compile, count the number of names we've
6060 encountered. */
6061
6062 cd->names_found++;
6063 ptr++; /* Move past > or ' */
6064 goto NUMBERED_GROUP;
6065
6066
6067 /* ------------------------------------------------------------ */
6068 case CHAR_AMPERSAND: /* Perl recursion/subroutine syntax */
6069 terminator = CHAR_RIGHT_PARENTHESIS;
6070 is_recurse = TRUE;
6071 /* Fall through */
6072
6073 /* We come here from the Python syntax above that handles both
6074 references (?P=name) and recursion (?P>name), as well as falling
6075 through from the Perl recursion syntax (?&name). We also come here from
6076 the Perl \k<name> or \k'name' back reference syntax and the \k{name}
6077 .NET syntax, and the Oniguruma \g<...> and \g'...' subroutine syntax. */
6078
6079 NAMED_REF_OR_RECURSE:
6080 name = ++ptr;
6081 while (MAX_255(*ptr) && (cd->ctypes[*ptr] & ctype_word) != 0) ptr++;
6082 namelen = (int)(ptr - name);
6083
6084 /* In the pre-compile phase, do a syntax check. We used to just set
6085 a dummy reference number, because it was not used in the first pass.
6086 However, with the change of recursive back references to be atomic,
6087 we have to look for the number so that this state can be identified, as
6088 otherwise the incorrect length is computed. If it's not a backwards
6089 reference, the dummy number will do. */
6090
6091 if (lengthptr != NULL)
6092 {
6093 const pcre_uchar *temp;
6094
6095 if (namelen == 0)
6096 {
6097 *errorcodeptr = ERR62;
6098 goto FAILED;
6099 }
6100 if (*ptr != terminator)
6101 {
6102 *errorcodeptr = ERR42;
6103 goto FAILED;
6104 }
6105 if (namelen > MAX_NAME_SIZE)
6106 {
6107 *errorcodeptr = ERR48;
6108 goto FAILED;
6109 }
6110
6111 /* The name table does not exist in the first pass, so we cannot
6112 do a simple search as in the code below. Instead, we have to scan the
6113 pattern to find the number. It is important that we scan it only as
6114 far as we have got because the syntax of named subpatterns has not
6115 been checked for the rest of the pattern, and find_parens() assumes
6116 correct syntax. In any case, it's a waste of resources to scan
6117 further. We stop the scan at the current point by temporarily
6118 adjusting the value of cd->endpattern. */
6119
6120 temp = cd->end_pattern;
6121 cd->end_pattern = ptr;
6122 recno = find_parens(cd, name, namelen,
6123 (options & PCRE_EXTENDED) != 0, utf);
6124 cd->end_pattern = temp;
6125 if (recno < 0) recno = 0; /* Forward ref; set dummy number */
6126 }
6127
6128 /* In the real compile, seek the name in the table. We check the name
6129 first, and then check that we have reached the end of the name in the
6130 table. That way, if the name that is longer than any in the table,
6131 the comparison will fail without reading beyond the table entry. */
6132
6133 else
6134 {
6135 slot = cd->name_table;
6136 for (i = 0; i < cd->names_found; i++)
6137 {
6138 if (STRNCMP_UC_UC(name, slot+IMM2_SIZE, namelen) == 0 &&
6139 slot[IMM2_SIZE+namelen] == 0)
6140 break;
6141 slot += cd->name_entry_size;
6142 }
6143
6144 if (i < cd->names_found) /* Back reference */
6145 {
6146 recno = GET2(slot, 0);
6147 }
6148 else if ((recno = /* Forward back reference */
6149 find_parens(cd, name, namelen,
6150 (options & PCRE_EXTENDED) != 0, utf)) <= 0)
6151 {
6152 *errorcodeptr = ERR15;
6153 goto FAILED;
6154 }
6155 }
6156
6157 /* In both phases, we can now go to the code than handles numerical
6158 recursion or backreferences. */
6159
6160 if (is_recurse) goto HANDLE_RECURSION;
6161 else goto HANDLE_REFERENCE;
6162
6163
6164 /* ------------------------------------------------------------ */
6165 case CHAR_R: /* Recursion */
6166 ptr++; /* Same as (?0) */
6167 /* Fall through */
6168
6169
6170 /* ------------------------------------------------------------ */
6171 case CHAR_MINUS: case CHAR_PLUS: /* Recursion or subroutine */
6172 case CHAR_0: case CHAR_1: case CHAR_2: case CHAR_3: case CHAR_4:
6173 case CHAR_5: case CHAR_6: case CHAR_7: case CHAR_8: case CHAR_9:
6174 {
6175 const pcre_uchar *called;
6176 terminator = CHAR_RIGHT_PARENTHESIS;
6177
6178 /* Come here from the \g<...> and \g'...' code (Oniguruma
6179 compatibility). However, the syntax has been checked to ensure that
6180 the ... are a (signed) number, so that neither ERR63 nor ERR29 will
6181 be called on this path, nor with the jump to OTHER_CHAR_AFTER_QUERY
6182 ever be taken. */
6183
6184 HANDLE_NUMERICAL_RECURSION:
6185
6186 if ((refsign = *ptr) == CHAR_PLUS)
6187 {
6188 ptr++;
6189 if (!IS_DIGIT(*ptr))
6190 {
6191 *errorcodeptr = ERR63;
6192 goto FAILED;
6193 }
6194 }
6195 else if (refsign == CHAR_MINUS)
6196 {
6197 if (!IS_DIGIT(ptr[1]))
6198 goto OTHER_CHAR_AFTER_QUERY;
6199 ptr++;
6200 }
6201
6202 recno = 0;
6203 while(IS_DIGIT(*ptr))
6204 recno = recno * 10 + *ptr++ - CHAR_0;
6205
6206 if (*ptr != terminator)
6207 {
6208 *errorcodeptr = ERR29;
6209 goto FAILED;
6210 }
6211
6212 if (refsign == CHAR_MINUS)
6213 {
6214 if (recno == 0)
6215 {
6216 *errorcodeptr = ERR58;
6217 goto FAILED;
6218 }
6219 recno = cd->bracount - recno + 1;
6220 if (recno <= 0)
6221 {
6222 *errorcodeptr = ERR15;
6223 goto FAILED;
6224 }
6225 }
6226 else if (refsign == CHAR_PLUS)
6227 {
6228 if (recno == 0)
6229 {
6230 *errorcodeptr = ERR58;
6231 goto FAILED;
6232 }
6233 recno += cd->bracount;
6234 }
6235
6236 /* Come here from code above that handles a named recursion */
6237
6238 HANDLE_RECURSION:
6239
6240 previous = code;
6241 called = cd->start_code;
6242
6243 /* When we are actually compiling, find the bracket that is being
6244 referenced. Temporarily end the regex in case it doesn't exist before
6245 this point. If we end up with a forward reference, first check that
6246 the bracket does occur later so we can give the error (and position)
6247 now. Then remember this forward reference in the workspace so it can
6248 be filled in at the end. */
6249
6250 if (lengthptr == NULL)
6251 {
6252 *code = OP_END;
6253 if (recno != 0)
6254 called = PRIV(find_bracket)(cd->start_code, utf, recno);
6255
6256 /* Forward reference */
6257
6258 if (called == NULL)
6259 {
6260 if (find_parens(cd, NULL, recno,
6261 (options & PCRE_EXTENDED) != 0, utf) < 0)
6262 {
6263 *errorcodeptr = ERR15;
6264 goto FAILED;
6265 }
6266
6267 /* Fudge the value of "called" so that when it is inserted as an
6268 offset below, what it actually inserted is the reference number
6269 of the group. Then remember the forward reference. */
6270
6271 called = cd->start_code + recno;
6272 if (cd->hwm >= cd->start_workspace + cd->workspace_size -
6273 WORK_SIZE_SAFETY_MARGIN)
6274 {
6275 *errorcodeptr = expand_workspace(cd);
6276 if (*errorcodeptr != 0) goto FAILED;
6277 }
6278 PUTINC(cd->hwm, 0, (int)(code + 1 - cd->start_code));
6279 }
6280
6281 /* If not a forward reference, and the subpattern is still open,
6282 this is a recursive call. We check to see if this is a left
6283 recursion that could loop for ever, and diagnose that case. We
6284 must not, however, do this check if we are in a conditional
6285 subpattern because the condition might be testing for recursion in
6286 a pattern such as /(?(R)a+|(?R)b)/, which is perfectly valid.
6287 Forever loops are also detected at runtime, so those that occur in
6288 conditional subpatterns will be picked up then. */
6289
6290 else if (GET(called, 1) == 0 && cond_depth <= 0 &&
6291 could_be_empty(called, code, bcptr, utf, cd))
6292 {
6293 *errorcodeptr = ERR40;
6294 goto FAILED;
6295 }
6296 }
6297
6298 /* Insert the recursion/subroutine item. It does not have a set first
6299 character (relevant if it is repeated, because it will then be
6300 wrapped with ONCE brackets). */
6301
6302 *code = OP_RECURSE;
6303 PUT(code, 1, (int)(called - cd->start_code));
6304 code += 1 + LINK_SIZE;
6305 groupsetfirstchar = FALSE;
6306 }
6307
6308 /* Can't determine a first byte now */
6309
6310 if (firstchar == REQ_UNSET) firstchar = REQ_NONE;
6311 continue;
6312
6313
6314 /* ------------------------------------------------------------ */
6315 default: /* Other characters: check option setting */
6316 OTHER_CHAR_AFTER_QUERY:
6317 set = unset = 0;
6318 optset = &set;
6319
6320 while (*ptr != CHAR_RIGHT_PARENTHESIS && *ptr != CHAR_COLON)
6321 {
6322 switch (*ptr++)
6323 {
6324 case CHAR_MINUS: optset = &unset; break;
6325
6326 case CHAR_J: /* Record that it changed in the external options */
6327 *optset |= PCRE_DUPNAMES;
6328 cd->external_flags |= PCRE_JCHANGED;
6329 break;
6330
6331 case CHAR_i: *optset |= PCRE_CASELESS; break;
6332 case CHAR_m: *optset |= PCRE_MULTILINE; break;
6333 case CHAR_s: *optset |= PCRE_DOTALL; break;
6334 case CHAR_x: *optset |= PCRE_EXTENDED; break;
6335 case CHAR_U: *optset |= PCRE_UNGREEDY; break;
6336 case CHAR_X: *optset |= PCRE_EXTRA; break;
6337
6338 default: *errorcodeptr = ERR12;
6339 ptr--; /* Correct the offset */
6340 goto FAILED;
6341 }
6342 }
6343
6344 /* Set up the changed option bits, but don't change anything yet. */
6345
6346 newoptions = (options | set) & (~unset);
6347
6348 /* If the options ended with ')' this is not the start of a nested
6349 group with option changes, so the options change at this level. If this
6350 item is right at the start of the pattern, the options can be
6351 abstracted and made external in the pre-compile phase, and ignored in
6352 the compile phase. This can be helpful when matching -- for instance in
6353 caseless checking of required bytes.
6354
6355 If the code pointer is not (cd->start_code + 1 + LINK_SIZE), we are
6356 definitely *not* at the start of the pattern because something has been
6357 compiled. In the pre-compile phase, however, the code pointer can have
6358 that value after the start, because it gets reset as code is discarded
6359 during the pre-compile. However, this can happen only at top level - if
6360 we are within parentheses, the starting BRA will still be present. At
6361 any parenthesis level, the length value can be used to test if anything
6362 has been compiled at that level. Thus, a test for both these conditions
6363 is necessary to ensure we correctly detect the start of the pattern in
6364 both phases.
6365
6366 If we are not at the pattern start, reset the greedy defaults and the
6367 case value for firstchar and reqchar. */
6368
6369 if (*ptr == CHAR_RIGHT_PARENTHESIS)
6370 {
6371 if (code == cd->start_code + 1 + LINK_SIZE &&
6372 (lengthptr == NULL || *lengthptr == 2 + 2*LINK_SIZE))
6373 {
6374 cd->external_options = newoptions;
6375 }
6376 else
6377 {
6378 greedy_default = ((newoptions & PCRE_UNGREEDY) != 0);
6379 greedy_non_default = greedy_default ^ 1;
6380 req_caseopt = ((newoptions & PCRE_CASELESS) != 0)? REQ_CASELESS:0;
6381 }
6382
6383 /* Change options at this level, and pass them back for use
6384 in subsequent branches. */
6385
6386 *optionsptr = options = newoptions;
6387 previous = NULL; /* This item can't be repeated */
6388 continue; /* It is complete */
6389 }
6390
6391 /* If the options ended with ':' we are heading into a nested group
6392 with possible change of options. Such groups are non-capturing and are
6393 not assertions of any kind. All we need to do is skip over the ':';
6394 the newoptions value is handled below. */
6395
6396 bravalue = OP_BRA;
6397 ptr++;
6398 } /* End of switch for character following (? */
6399 } /* End of (? handling */
6400
6401 /* Opening parenthesis not followed by '*' or '?'. If PCRE_NO_AUTO_CAPTURE
6402 is set, all unadorned brackets become non-capturing and behave like (?:...)
6403 brackets. */
6404
6405 else if ((options & PCRE_NO_AUTO_CAPTURE) != 0)
6406 {
6407 bravalue = OP_BRA;
6408 }
6409
6410 /* Else we have a capturing group. */
6411
6412 else
6413 {
6414 NUMBERED_GROUP:
6415 cd->bracount += 1;
6416 PUT2(code, 1+LINK_SIZE, cd->bracount);
6417 skipbytes = IMM2_SIZE;
6418 }
6419
6420 /* Process nested bracketed regex. Assertions used not to be repeatable,
6421 but this was changed for Perl compatibility, so all kinds can now be
6422 repeated. We copy code into a non-register variable (tempcode) in order to
6423 be able to pass its address because some compilers complain otherwise. */
6424
6425 previous = code; /* For handling repetition */
6426 *code = bravalue;
6427 tempcode = code;
6428 tempreqvary = cd->req_varyopt; /* Save value before bracket */
6429 tempbracount = cd->bracount; /* Save value before bracket */
6430 length_prevgroup = 0; /* Initialize for pre-compile phase */
6431
6432 if (!compile_regex(
6433 newoptions, /* The complete new option state */
6434 &tempcode, /* Where to put code (updated) */
6435 &ptr, /* Input pointer (updated) */
6436 errorcodeptr, /* Where to put an error message */
6437 (bravalue == OP_ASSERTBACK ||
6438 bravalue == OP_ASSERTBACK_NOT), /* TRUE if back assert */
6439 reset_bracount, /* True if (?| group */
6440 skipbytes, /* Skip over bracket number */
6441 cond_depth +
6442 ((bravalue == OP_COND)?1:0), /* Depth of condition subpatterns */
6443 &subfirstchar, /* For possible first char */
6444 &subreqchar, /* For possible last char */
6445 bcptr, /* Current branch chain */
6446 cd, /* Tables block */
6447 (lengthptr == NULL)? NULL : /* Actual compile phase */
6448 &length_prevgroup /* Pre-compile phase */
6449 ))
6450 goto FAILED;
6451
6452 /* If this was an atomic group and there are no capturing groups within it,
6453 generate OP_ONCE_NC instead of OP_ONCE. */
6454
6455 if (bravalue == OP_ONCE && cd->bracount <= tempbracount)
6456 *code = OP_ONCE_NC;
6457
6458 if (bravalue >= OP_ASSERT && bravalue <= OP_ASSERTBACK_NOT)
6459 cd->assert_depth -= 1;
6460
6461 /* At the end of compiling, code is still pointing to the start of the
6462 group, while tempcode has been updated to point past the end of the group.
6463 The pattern pointer (ptr) is on the bracket.
6464
6465 If this is a conditional bracket, check that there are no more than
6466 two branches in the group, or just one if it's a DEFINE group. We do this
6467 in the real compile phase, not in the pre-pass, where the whole group may
6468 not be available. */
6469
6470 if (bravalue == OP_COND && lengthptr == NULL)
6471 {
6472 pcre_uchar *tc = code;
6473 int condcount = 0;
6474
6475 do {
6476 condcount++;
6477 tc += GET(tc,1);
6478 }
6479 while (*tc != OP_KET);
6480
6481 /* A DEFINE group is never obeyed inline (the "condition" is always
6482 false). It must have only one branch. */
6483
6484 if (code[LINK_SIZE+1] == OP_DEF)
6485 {
6486 if (condcount > 1)
6487 {
6488 *errorcodeptr = ERR54;
6489 goto FAILED;
6490 }
6491 bravalue = OP_DEF; /* Just a flag to suppress char handling below */
6492 }
6493
6494 /* A "normal" conditional group. If there is just one branch, we must not
6495 make use of its firstchar or reqchar, because this is equivalent to an
6496 empty second branch. */
6497
6498 else
6499 {
6500 if (condcount > 2)
6501 {
6502 *errorcodeptr = ERR27;
6503 goto FAILED;
6504 }
6505 if (condcount == 1) subfirstchar = subreqchar = REQ_NONE;
6506 }
6507 }
6508
6509 /* Error if hit end of pattern */
6510
6511 if (*ptr != CHAR_RIGHT_PARENTHESIS)
6512 {
6513 *errorcodeptr = ERR14;
6514 goto FAILED;
6515 }
6516
6517 /* In the pre-compile phase, update the length by the length of the group,
6518 less the brackets at either end. Then reduce the compiled code to just a
6519 set of non-capturing brackets so that it doesn't use much memory if it is
6520 duplicated by a quantifier.*/
6521
6522 if (lengthptr != NULL)
6523 {
6524 if (OFLOW_MAX - *lengthptr < length_prevgroup - 2 - 2*LINK_SIZE)
6525 {
6526 *errorcodeptr = ERR20;
6527 goto FAILED;
6528 }
6529 *lengthptr += length_prevgroup - 2 - 2*LINK_SIZE;
6530 code++; /* This already contains bravalue */
6531 PUTINC(code, 0, 1 + LINK_SIZE);
6532 *code++ = OP_KET;
6533 PUTINC(code, 0, 1 + LINK_SIZE);
6534 break; /* No need to waste time with special character handling */
6535 }
6536
6537 /* Otherwise update the main code pointer to the end of the group. */
6538
6539 code = tempcode;
6540
6541 /* For a DEFINE group, required and first character settings are not
6542 relevant. */
6543
6544 if (bravalue == OP_DEF) break;
6545
6546 /* Handle updating of the required and first characters for other types of
6547 group. Update for normal brackets of all kinds, and conditions with two
6548 branches (see code above). If the bracket is followed by a quantifier with
6549 zero repeat, we have to back off. Hence the definition of zeroreqchar and
6550 zerofirstchar outside the main loop so that they can be accessed for the
6551 back off. */
6552
6553 zeroreqchar = reqchar;
6554 zerofirstchar = firstchar;
6555 groupsetfirstchar = FALSE;
6556
6557 if (bravalue >= OP_ONCE)
6558 {
6559 /* If we have not yet set a firstchar in this branch, take it from the
6560 subpattern, remembering that it was set here so that a repeat of more
6561 than one can replicate it as reqchar if necessary. If the subpattern has
6562 no firstchar, set "none" for the whole branch. In both cases, a zero
6563 repeat forces firstchar to "none". */
6564
6565 if (firstchar == REQ_UNSET)
6566 {
6567 if (subfirstchar >= 0)
6568 {
6569 firstchar = subfirstchar;
6570 groupsetfirstchar = TRUE;
6571 }
6572 else firstchar = REQ_NONE;
6573 zerofirstchar = REQ_NONE;
6574 }
6575
6576 /* If firstchar was previously set, convert the subpattern's firstchar
6577 into reqchar if there wasn't one, using the vary flag that was in
6578 existence beforehand. */
6579
6580 else if (subfirstchar >= 0 && subreqchar < 0)
6581 subreqchar = subfirstchar | tempreqvary;
6582
6583 /* If the subpattern set a required byte (or set a first byte that isn't
6584 really the first byte - see above), set it. */
6585
6586 if (subreqchar >= 0) reqchar = subreqchar;
6587 }
6588
6589 /* For a forward assertion, we take the reqchar, if set. This can be
6590 helpful if the pattern that follows the assertion doesn't set a different
6591 char. For example, it's useful for /(?=abcde).+/. We can't set firstchar
6592 for an assertion, however because it leads to incorrect effect for patterns
6593 such as /(?=a)a.+/ when the "real" "a" would then become a reqchar instead
6594 of a firstchar. This is overcome by a scan at the end if there's no
6595 firstchar, looking for an asserted first char. */
6596
6597 else if (bravalue == OP_ASSERT && subreqchar >= 0) reqchar = subreqchar;
6598 break; /* End of processing '(' */
6599
6600
6601 /* ===================================================================*/
6602 /* Handle metasequences introduced by \. For ones like \d, the ESC_ values
6603 are arranged to be the negation of the corresponding OP_values in the
6604 default case when PCRE_UCP is not set. For the back references, the values
6605 are ESC_REF plus the reference number. Only back references and those types
6606 that consume a character may be repeated. We can test for values between
6607 ESC_b and ESC_Z for the latter; this may have to change if any new ones are
6608 ever created. */
6609
6610 case CHAR_BACKSLASH:
6611 tempptr = ptr;
6612 c = check_escape(&ptr, errorcodeptr, cd->bracount, options, FALSE);
6613 if (*errorcodeptr != 0) goto FAILED;
6614
6615 if (c < 0)
6616 {
6617 if (-c == ESC_Q) /* Handle start of quoted string */
6618 {
6619 if (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
6620 ptr += 2; /* avoid empty string */
6621 else inescq = TRUE;
6622 continue;
6623 }
6624
6625 if (-c == ESC_E) continue; /* Perl ignores an orphan \E */
6626
6627 /* For metasequences that actually match a character, we disable the
6628 setting of a first character if it hasn't already been set. */
6629
6630 if (firstchar == REQ_UNSET && -c > ESC_b && -c < ESC_Z)
6631 firstchar = REQ_NONE;