/[pcre]/code/tags/pcre-8.33/pcre_compile.c
ViewVC logotype

Contents of /code/tags/pcre-8.33/pcre_compile.c

Parent Directory Parent Directory | Revision Log Revision Log


Revision 975 - (show annotations)
Sat Jun 2 11:03:06 2012 UTC (7 years, 5 months ago) by ph10
Original Path: code/trunk/pcre_compile.c
File MIME type: text/plain
File size: 269421 byte(s)
Document update for 8.31-RC1 test release.
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) 1997-2012 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 compiled without UTF 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 in 8-bit non-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 Unicode property 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 Unicode code point (>= 0xd800 && <= 0xdfff)\0"
491 "invalid UTF-16 string\0"
492 /* 75 */
493 "name is too long in (*MARK), (*PRUNE), (*SKIP), or (*THEN)\0"
494 ;
495
496 /* Table to identify digits and hex digits. This is used when compiling
497 patterns. Note that the tables in chartables are dependent on the locale, and
498 may mark arbitrary characters as digits - but the PCRE compiling code expects
499 to handle only 0-9, a-z, and A-Z as digits when compiling. That is why we have
500 a private table here. It costs 256 bytes, but it is a lot faster than doing
501 character value tests (at least in some simple cases I timed), and in some
502 applications one wants PCRE to compile efficiently as well as match
503 efficiently.
504
505 For convenience, we use the same bit definitions as in chartables:
506
507 0x04 decimal digit
508 0x08 hexadecimal digit
509
510 Then we can use ctype_digit and ctype_xdigit in the code. */
511
512 /* Using a simple comparison for decimal numbers rather than a memory read
513 is much faster, and the resulting code is simpler (the compiler turns it
514 into a subtraction and unsigned comparison). */
515
516 #define IS_DIGIT(x) ((x) >= CHAR_0 && (x) <= CHAR_9)
517
518 #ifndef EBCDIC
519
520 /* This is the "normal" case, for ASCII systems, and EBCDIC systems running in
521 UTF-8 mode. */
522
523 static const pcre_uint8 digitab[] =
524 {
525 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 0- 7 */
526 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 8- 15 */
527 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 16- 23 */
528 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 24- 31 */
529 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - ' */
530 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* ( - / */
531 0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c, /* 0 - 7 */
532 0x0c,0x0c,0x00,0x00,0x00,0x00,0x00,0x00, /* 8 - ? */
533 0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* @ - G */
534 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* H - O */
535 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* P - W */
536 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* X - _ */
537 0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* ` - g */
538 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* h - o */
539 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* p - w */
540 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* x -127 */
541 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 128-135 */
542 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 136-143 */
543 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 144-151 */
544 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 152-159 */
545 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 160-167 */
546 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 168-175 */
547 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 176-183 */
548 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 184-191 */
549 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 192-199 */
550 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 200-207 */
551 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 208-215 */
552 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 216-223 */
553 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 224-231 */
554 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 232-239 */
555 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 240-247 */
556 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00};/* 248-255 */
557
558 #else
559
560 /* This is the "abnormal" case, for EBCDIC systems not running in UTF-8 mode. */
561
562 static const pcre_uint8 digitab[] =
563 {
564 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 0- 7 0 */
565 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 8- 15 */
566 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 16- 23 10 */
567 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 24- 31 */
568 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 32- 39 20 */
569 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 40- 47 */
570 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 48- 55 30 */
571 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 56- 63 */
572 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - 71 40 */
573 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 72- | */
574 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* & - 87 50 */
575 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 88- 95 */
576 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - -103 60 */
577 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 104- ? */
578 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 112-119 70 */
579 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 120- " */
580 0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* 128- g 80 */
581 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* h -143 */
582 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 144- p 90 */
583 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* q -159 */
584 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 160- x A0 */
585 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* y -175 */
586 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* ^ -183 B0 */
587 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 184-191 */
588 0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* { - G C0 */
589 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* H -207 */
590 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* } - P D0 */
591 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* Q -223 */
592 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* \ - X E0 */
593 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* Y -239 */
594 0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c, /* 0 - 7 F0 */
595 0x0c,0x0c,0x00,0x00,0x00,0x00,0x00,0x00};/* 8 -255 */
596
597 static const pcre_uint8 ebcdic_chartab[] = { /* chartable partial dup */
598 0x80,0x00,0x00,0x00,0x00,0x01,0x00,0x00, /* 0- 7 */
599 0x00,0x00,0x00,0x00,0x01,0x01,0x00,0x00, /* 8- 15 */
600 0x00,0x00,0x00,0x00,0x00,0x01,0x00,0x00, /* 16- 23 */
601 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 24- 31 */
602 0x00,0x00,0x00,0x00,0x00,0x01,0x00,0x00, /* 32- 39 */
603 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 40- 47 */
604 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 48- 55 */
605 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 56- 63 */
606 0x01,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - 71 */
607 0x00,0x00,0x00,0x80,0x00,0x80,0x80,0x80, /* 72- | */
608 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* & - 87 */
609 0x00,0x00,0x00,0x80,0x80,0x80,0x00,0x00, /* 88- 95 */
610 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - -103 */
611 0x00,0x00,0x00,0x00,0x00,0x10,0x00,0x80, /* 104- ? */
612 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 112-119 */
613 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 120- " */
614 0x00,0x1a,0x1a,0x1a,0x1a,0x1a,0x1a,0x12, /* 128- g */
615 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* h -143 */
616 0x00,0x12,0x12,0x12,0x12,0x12,0x12,0x12, /* 144- p */
617 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* q -159 */
618 0x00,0x00,0x12,0x12,0x12,0x12,0x12,0x12, /* 160- x */
619 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* y -175 */
620 0x80,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* ^ -183 */
621 0x00,0x00,0x80,0x00,0x00,0x00,0x00,0x00, /* 184-191 */
622 0x80,0x1a,0x1a,0x1a,0x1a,0x1a,0x1a,0x12, /* { - G */
623 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* H -207 */
624 0x00,0x12,0x12,0x12,0x12,0x12,0x12,0x12, /* } - P */
625 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* Q -223 */
626 0x00,0x00,0x12,0x12,0x12,0x12,0x12,0x12, /* \ - X */
627 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* Y -239 */
628 0x1c,0x1c,0x1c,0x1c,0x1c,0x1c,0x1c,0x1c, /* 0 - 7 */
629 0x1c,0x1c,0x00,0x00,0x00,0x00,0x00,0x00};/* 8 -255 */
630 #endif
631
632
633 /* Definition to allow mutual recursion */
634
635 static BOOL
636 compile_regex(int, pcre_uchar **, const pcre_uchar **, int *, BOOL, BOOL, int, int,
637 int *, int *, branch_chain *, compile_data *, int *);
638
639
640
641 /*************************************************
642 * Find an error text *
643 *************************************************/
644
645 /* The error texts are now all in one long string, to save on relocations. As
646 some of the text is of unknown length, we can't use a table of offsets.
647 Instead, just count through the strings. This is not a performance issue
648 because it happens only when there has been a compilation error.
649
650 Argument: the error number
651 Returns: pointer to the error string
652 */
653
654 static const char *
655 find_error_text(int n)
656 {
657 const char *s = error_texts;
658 for (; n > 0; n--)
659 {
660 while (*s++ != 0) {};
661 if (*s == 0) return "Error text not found (please report)";
662 }
663 return s;
664 }
665
666
667 /*************************************************
668 * Expand the workspace *
669 *************************************************/
670
671 /* This function is called during the second compiling phase, if the number of
672 forward references fills the existing workspace, which is originally a block on
673 the stack. A larger block is obtained from malloc() unless the ultimate limit
674 has been reached or the increase will be rather small.
675
676 Argument: pointer to the compile data block
677 Returns: 0 if all went well, else an error number
678 */
679
680 static int
681 expand_workspace(compile_data *cd)
682 {
683 pcre_uchar *newspace;
684 int newsize = cd->workspace_size * 2;
685
686 if (newsize > COMPILE_WORK_SIZE_MAX) newsize = COMPILE_WORK_SIZE_MAX;
687 if (cd->workspace_size >= COMPILE_WORK_SIZE_MAX ||
688 newsize - cd->workspace_size < WORK_SIZE_SAFETY_MARGIN)
689 return ERR72;
690
691 newspace = (PUBL(malloc))(IN_UCHARS(newsize));
692 if (newspace == NULL) return ERR21;
693 memcpy(newspace, cd->start_workspace, cd->workspace_size * sizeof(pcre_uchar));
694 cd->hwm = (pcre_uchar *)newspace + (cd->hwm - cd->start_workspace);
695 if (cd->workspace_size > COMPILE_WORK_SIZE)
696 (PUBL(free))((void *)cd->start_workspace);
697 cd->start_workspace = newspace;
698 cd->workspace_size = newsize;
699 return 0;
700 }
701
702
703
704 /*************************************************
705 * Check for counted repeat *
706 *************************************************/
707
708 /* This function is called when a '{' is encountered in a place where it might
709 start a quantifier. It looks ahead to see if it really is a quantifier or not.
710 It is only a quantifier if it is one of the forms {ddd} {ddd,} or {ddd,ddd}
711 where the ddds are digits.
712
713 Arguments:
714 p pointer to the first char after '{'
715
716 Returns: TRUE or FALSE
717 */
718
719 static BOOL
720 is_counted_repeat(const pcre_uchar *p)
721 {
722 if (!IS_DIGIT(*p)) return FALSE;
723 p++;
724 while (IS_DIGIT(*p)) p++;
725 if (*p == CHAR_RIGHT_CURLY_BRACKET) return TRUE;
726
727 if (*p++ != CHAR_COMMA) return FALSE;
728 if (*p == CHAR_RIGHT_CURLY_BRACKET) return TRUE;
729
730 if (!IS_DIGIT(*p)) return FALSE;
731 p++;
732 while (IS_DIGIT(*p)) p++;
733
734 return (*p == CHAR_RIGHT_CURLY_BRACKET);
735 }
736
737
738
739 /*************************************************
740 * Handle escapes *
741 *************************************************/
742
743 /* This function is called when a \ has been encountered. It either returns a
744 positive value for a simple escape such as \n, or a negative value which
745 encodes one of the more complicated things such as \d. A backreference to group
746 n is returned as -(ESC_REF + n); ESC_REF is the highest ESC_xxx macro. When
747 UTF-8 is enabled, a positive value greater than 255 may be returned. On entry,
748 ptr is pointing at the \. On exit, it is on the final character of the escape
749 sequence.
750
751 Arguments:
752 ptrptr points to the pattern position pointer
753 errorcodeptr points to the errorcode variable
754 bracount number of previous extracting brackets
755 options the options bits
756 isclass TRUE if inside a character class
757
758 Returns: zero or positive => a data character
759 negative => a special escape sequence
760 on error, errorcodeptr is set
761 */
762
763 static int
764 check_escape(const pcre_uchar **ptrptr, int *errorcodeptr, int bracount,
765 int options, BOOL isclass)
766 {
767 /* PCRE_UTF16 has the same value as PCRE_UTF8. */
768 BOOL utf = (options & PCRE_UTF8) != 0;
769 const pcre_uchar *ptr = *ptrptr + 1;
770 pcre_int32 c;
771 int i;
772
773 GETCHARINCTEST(c, ptr); /* Get character value, increment pointer */
774 ptr--; /* Set pointer back to the last byte */
775
776 /* If backslash is at the end of the pattern, it's an error. */
777
778 if (c == 0) *errorcodeptr = ERR1;
779
780 /* Non-alphanumerics are literals. For digits or letters, do an initial lookup
781 in a table. A non-zero result is something that can be returned immediately.
782 Otherwise further processing may be required. */
783
784 #ifndef EBCDIC /* ASCII/UTF-8 coding */
785 /* Not alphanumeric */
786 else if (c < CHAR_0 || c > CHAR_z) {}
787 else if ((i = escapes[c - CHAR_0]) != 0) c = i;
788
789 #else /* EBCDIC coding */
790 /* Not alphanumeric */
791 else if (c < 'a' || (!MAX_255(c) || (ebcdic_chartab[c] & 0x0E) == 0)) {}
792 else if ((i = escapes[c - 0x48]) != 0) c = i;
793 #endif
794
795 /* Escapes that need further processing, or are illegal. */
796
797 else
798 {
799 const pcre_uchar *oldptr;
800 BOOL braced, negated;
801
802 switch (c)
803 {
804 /* A number of Perl escapes are not handled by PCRE. We give an explicit
805 error. */
806
807 case CHAR_l:
808 case CHAR_L:
809 *errorcodeptr = ERR37;
810 break;
811
812 case CHAR_u:
813 if ((options & PCRE_JAVASCRIPT_COMPAT) != 0)
814 {
815 /* In JavaScript, \u must be followed by four hexadecimal numbers.
816 Otherwise it is a lowercase u letter. */
817 if (MAX_255(ptr[1]) && (digitab[ptr[1]] & ctype_xdigit) != 0
818 && MAX_255(ptr[2]) && (digitab[ptr[2]] & ctype_xdigit) != 0
819 && MAX_255(ptr[3]) && (digitab[ptr[3]] & ctype_xdigit) != 0
820 && MAX_255(ptr[4]) && (digitab[ptr[4]] & ctype_xdigit) != 0)
821 {
822 c = 0;
823 for (i = 0; i < 4; ++i)
824 {
825 register int cc = *(++ptr);
826 #ifndef EBCDIC /* ASCII/UTF-8 coding */
827 if (cc >= CHAR_a) cc -= 32; /* Convert to upper case */
828 c = (c << 4) + cc - ((cc < CHAR_A)? CHAR_0 : (CHAR_A - 10));
829 #else /* EBCDIC coding */
830 if (cc >= CHAR_a && cc <= CHAR_z) cc += 64; /* Convert to upper case */
831 c = (c << 4) + cc - ((cc >= CHAR_0)? CHAR_0 : (CHAR_A - 10));
832 #endif
833 }
834 }
835 }
836 else
837 *errorcodeptr = ERR37;
838 break;
839
840 case CHAR_U:
841 /* In JavaScript, \U is an uppercase U letter. */
842 if ((options & PCRE_JAVASCRIPT_COMPAT) == 0) *errorcodeptr = ERR37;
843 break;
844
845 /* In a character class, \g is just a literal "g". Outside a character
846 class, \g must be followed by one of a number of specific things:
847
848 (1) A number, either plain or braced. If positive, it is an absolute
849 backreference. If negative, it is a relative backreference. This is a Perl
850 5.10 feature.
851
852 (2) Perl 5.10 also supports \g{name} as a reference to a named group. This
853 is part of Perl's movement towards a unified syntax for back references. As
854 this is synonymous with \k{name}, we fudge it up by pretending it really
855 was \k.
856
857 (3) For Oniguruma compatibility we also support \g followed by a name or a
858 number either in angle brackets or in single quotes. However, these are
859 (possibly recursive) subroutine calls, _not_ backreferences. Just return
860 the -ESC_g code (cf \k). */
861
862 case CHAR_g:
863 if (isclass) break;
864 if (ptr[1] == CHAR_LESS_THAN_SIGN || ptr[1] == CHAR_APOSTROPHE)
865 {
866 c = -ESC_g;
867 break;
868 }
869
870 /* Handle the Perl-compatible cases */
871
872 if (ptr[1] == CHAR_LEFT_CURLY_BRACKET)
873 {
874 const pcre_uchar *p;
875 for (p = ptr+2; *p != 0 && *p != CHAR_RIGHT_CURLY_BRACKET; p++)
876 if (*p != CHAR_MINUS && !IS_DIGIT(*p)) break;
877 if (*p != 0 && *p != CHAR_RIGHT_CURLY_BRACKET)
878 {
879 c = -ESC_k;
880 break;
881 }
882 braced = TRUE;
883 ptr++;
884 }
885 else braced = FALSE;
886
887 if (ptr[1] == CHAR_MINUS)
888 {
889 negated = TRUE;
890 ptr++;
891 }
892 else negated = FALSE;
893
894 /* The integer range is limited by the machine's int representation. */
895 c = 0;
896 while (IS_DIGIT(ptr[1]))
897 {
898 if (((unsigned int)c) > INT_MAX / 10) /* Integer overflow */
899 {
900 c = -1;
901 break;
902 }
903 c = c * 10 + *(++ptr) - CHAR_0;
904 }
905 if (((unsigned int)c) > INT_MAX) /* Integer overflow */
906 {
907 while (IS_DIGIT(ptr[1]))
908 ptr++;
909 *errorcodeptr = ERR61;
910 break;
911 }
912
913 if (braced && *(++ptr) != CHAR_RIGHT_CURLY_BRACKET)
914 {
915 *errorcodeptr = ERR57;
916 break;
917 }
918
919 if (c == 0)
920 {
921 *errorcodeptr = ERR58;
922 break;
923 }
924
925 if (negated)
926 {
927 if (c > bracount)
928 {
929 *errorcodeptr = ERR15;
930 break;
931 }
932 c = bracount - (c - 1);
933 }
934
935 c = -(ESC_REF + c);
936 break;
937
938 /* The handling of escape sequences consisting of a string of digits
939 starting with one that is not zero is not straightforward. By experiment,
940 the way Perl works seems to be as follows:
941
942 Outside a character class, the digits are read as a decimal number. If the
943 number is less than 10, or if there are that many previous extracting
944 left brackets, then it is a back reference. Otherwise, up to three octal
945 digits are read to form an escaped byte. Thus \123 is likely to be octal
946 123 (cf \0123, which is octal 012 followed by the literal 3). If the octal
947 value is greater than 377, the least significant 8 bits are taken. Inside a
948 character class, \ followed by a digit is always an octal number. */
949
950 case CHAR_1: case CHAR_2: case CHAR_3: case CHAR_4: case CHAR_5:
951 case CHAR_6: case CHAR_7: case CHAR_8: case CHAR_9:
952
953 if (!isclass)
954 {
955 oldptr = ptr;
956 /* The integer range is limited by the machine's int representation. */
957 c -= CHAR_0;
958 while (IS_DIGIT(ptr[1]))
959 {
960 if (((unsigned int)c) > INT_MAX / 10) /* Integer overflow */
961 {
962 c = -1;
963 break;
964 }
965 c = c * 10 + *(++ptr) - CHAR_0;
966 }
967 if (((unsigned int)c) > INT_MAX) /* Integer overflow */
968 {
969 while (IS_DIGIT(ptr[1]))
970 ptr++;
971 *errorcodeptr = ERR61;
972 break;
973 }
974 if (c < 10 || c <= bracount)
975 {
976 c = -(ESC_REF + c);
977 break;
978 }
979 ptr = oldptr; /* Put the pointer back and fall through */
980 }
981
982 /* Handle an octal number following \. If the first digit is 8 or 9, Perl
983 generates a binary zero byte and treats the digit as a following literal.
984 Thus we have to pull back the pointer by one. */
985
986 if ((c = *ptr) >= CHAR_8)
987 {
988 ptr--;
989 c = 0;
990 break;
991 }
992
993 /* \0 always starts an octal number, but we may drop through to here with a
994 larger first octal digit. The original code used just to take the least
995 significant 8 bits of octal numbers (I think this is what early Perls used
996 to do). Nowadays we allow for larger numbers in UTF-8 mode and 16-bit mode,
997 but no more than 3 octal digits. */
998
999 case CHAR_0:
1000 c -= CHAR_0;
1001 while(i++ < 2 && ptr[1] >= CHAR_0 && ptr[1] <= CHAR_7)
1002 c = c * 8 + *(++ptr) - CHAR_0;
1003 #ifdef COMPILE_PCRE8
1004 if (!utf && c > 0xff) *errorcodeptr = ERR51;
1005 #endif
1006 break;
1007
1008 /* \x is complicated. \x{ddd} is a character number which can be greater
1009 than 0xff in utf or non-8bit mode, but only if the ddd are hex digits.
1010 If not, { is treated as a data character. */
1011
1012 case CHAR_x:
1013 if ((options & PCRE_JAVASCRIPT_COMPAT) != 0)
1014 {
1015 /* In JavaScript, \x must be followed by two hexadecimal numbers.
1016 Otherwise it is a lowercase x letter. */
1017 if (MAX_255(ptr[1]) && (digitab[ptr[1]] & ctype_xdigit) != 0
1018 && MAX_255(ptr[2]) && (digitab[ptr[2]] & ctype_xdigit) != 0)
1019 {
1020 c = 0;
1021 for (i = 0; i < 2; ++i)
1022 {
1023 register int cc = *(++ptr);
1024 #ifndef EBCDIC /* ASCII/UTF-8 coding */
1025 if (cc >= CHAR_a) cc -= 32; /* Convert to upper case */
1026 c = (c << 4) + cc - ((cc < CHAR_A)? CHAR_0 : (CHAR_A - 10));
1027 #else /* EBCDIC coding */
1028 if (cc >= CHAR_a && cc <= CHAR_z) cc += 64; /* Convert to upper case */
1029 c = (c << 4) + cc - ((cc >= CHAR_0)? CHAR_0 : (CHAR_A - 10));
1030 #endif
1031 }
1032 }
1033 break;
1034 }
1035
1036 if (ptr[1] == CHAR_LEFT_CURLY_BRACKET)
1037 {
1038 const pcre_uchar *pt = ptr + 2;
1039
1040 c = 0;
1041 while (MAX_255(*pt) && (digitab[*pt] & ctype_xdigit) != 0)
1042 {
1043 register int cc = *pt++;
1044 if (c == 0 && cc == CHAR_0) continue; /* Leading zeroes */
1045
1046 #ifndef EBCDIC /* ASCII/UTF-8 coding */
1047 if (cc >= CHAR_a) cc -= 32; /* Convert to upper case */
1048 c = (c << 4) + cc - ((cc < CHAR_A)? CHAR_0 : (CHAR_A - 10));
1049 #else /* EBCDIC coding */
1050 if (cc >= CHAR_a && cc <= CHAR_z) cc += 64; /* Convert to upper case */
1051 c = (c << 4) + cc - ((cc >= CHAR_0)? CHAR_0 : (CHAR_A - 10));
1052 #endif
1053
1054 #ifdef COMPILE_PCRE8
1055 if (c > (utf ? 0x10ffff : 0xff)) { c = -1; break; }
1056 #else
1057 #ifdef COMPILE_PCRE16
1058 if (c > (utf ? 0x10ffff : 0xffff)) { c = -1; break; }
1059 #endif
1060 #endif
1061 }
1062
1063 if (c < 0)
1064 {
1065 while (MAX_255(*pt) && (digitab[*pt] & ctype_xdigit) != 0) pt++;
1066 *errorcodeptr = ERR34;
1067 }
1068
1069 if (*pt == CHAR_RIGHT_CURLY_BRACKET)
1070 {
1071 if (utf && c >= 0xd800 && c <= 0xdfff) *errorcodeptr = ERR73;
1072 ptr = pt;
1073 break;
1074 }
1075
1076 /* If the sequence of hex digits does not end with '}', then we don't
1077 recognize this construct; fall through to the normal \x handling. */
1078 }
1079
1080 /* Read just a single-byte hex-defined char */
1081
1082 c = 0;
1083 while (i++ < 2 && MAX_255(ptr[1]) && (digitab[ptr[1]] & ctype_xdigit) != 0)
1084 {
1085 int cc; /* Some compilers don't like */
1086 cc = *(++ptr); /* ++ in initializers */
1087 #ifndef EBCDIC /* ASCII/UTF-8 coding */
1088 if (cc >= CHAR_a) cc -= 32; /* Convert to upper case */
1089 c = c * 16 + cc - ((cc < CHAR_A)? CHAR_0 : (CHAR_A - 10));
1090 #else /* EBCDIC coding */
1091 if (cc <= CHAR_z) cc += 64; /* Convert to upper case */
1092 c = c * 16 + cc - ((cc >= CHAR_0)? CHAR_0 : (CHAR_A - 10));
1093 #endif
1094 }
1095 break;
1096
1097 /* For \c, a following letter is upper-cased; then the 0x40 bit is flipped.
1098 An error is given if the byte following \c is not an ASCII character. This
1099 coding is ASCII-specific, but then the whole concept of \cx is
1100 ASCII-specific. (However, an EBCDIC equivalent has now been added.) */
1101
1102 case CHAR_c:
1103 c = *(++ptr);
1104 if (c == 0)
1105 {
1106 *errorcodeptr = ERR2;
1107 break;
1108 }
1109 #ifndef EBCDIC /* ASCII/UTF-8 coding */
1110 if (c > 127) /* Excludes all non-ASCII in either mode */
1111 {
1112 *errorcodeptr = ERR68;
1113 break;
1114 }
1115 if (c >= CHAR_a && c <= CHAR_z) c -= 32;
1116 c ^= 0x40;
1117 #else /* EBCDIC coding */
1118 if (c >= CHAR_a && c <= CHAR_z) c += 64;
1119 c ^= 0xC0;
1120 #endif
1121 break;
1122
1123 /* PCRE_EXTRA enables extensions to Perl in the matter of escapes. Any
1124 other alphanumeric following \ is an error if PCRE_EXTRA was set;
1125 otherwise, for Perl compatibility, it is a literal. This code looks a bit
1126 odd, but there used to be some cases other than the default, and there may
1127 be again in future, so I haven't "optimized" it. */
1128
1129 default:
1130 if ((options & PCRE_EXTRA) != 0) switch(c)
1131 {
1132 default:
1133 *errorcodeptr = ERR3;
1134 break;
1135 }
1136 break;
1137 }
1138 }
1139
1140 /* Perl supports \N{name} for character names, as well as plain \N for "not
1141 newline". PCRE does not support \N{name}. However, it does support
1142 quantification such as \N{2,3}. */
1143
1144 if (c == -ESC_N && ptr[1] == CHAR_LEFT_CURLY_BRACKET &&
1145 !is_counted_repeat(ptr+2))
1146 *errorcodeptr = ERR37;
1147
1148 /* If PCRE_UCP is set, we change the values for \d etc. */
1149
1150 if ((options & PCRE_UCP) != 0 && c <= -ESC_D && c >= -ESC_w)
1151 c -= (ESC_DU - ESC_D);
1152
1153 /* Set the pointer to the final character before returning. */
1154
1155 *ptrptr = ptr;
1156 return c;
1157 }
1158
1159
1160
1161 #ifdef SUPPORT_UCP
1162 /*************************************************
1163 * Handle \P and \p *
1164 *************************************************/
1165
1166 /* This function is called after \P or \p has been encountered, provided that
1167 PCRE is compiled with support for Unicode properties. On entry, ptrptr is
1168 pointing at the P or p. On exit, it is pointing at the final character of the
1169 escape sequence.
1170
1171 Argument:
1172 ptrptr points to the pattern position pointer
1173 negptr points to a boolean that is set TRUE for negation else FALSE
1174 dptr points to an int that is set to the detailed property value
1175 errorcodeptr points to the error code variable
1176
1177 Returns: type value from ucp_type_table, or -1 for an invalid type
1178 */
1179
1180 static int
1181 get_ucp(const pcre_uchar **ptrptr, BOOL *negptr, int *dptr, int *errorcodeptr)
1182 {
1183 int c, i, bot, top;
1184 const pcre_uchar *ptr = *ptrptr;
1185 pcre_uchar name[32];
1186
1187 c = *(++ptr);
1188 if (c == 0) goto ERROR_RETURN;
1189
1190 *negptr = FALSE;
1191
1192 /* \P or \p can be followed by a name in {}, optionally preceded by ^ for
1193 negation. */
1194
1195 if (c == CHAR_LEFT_CURLY_BRACKET)
1196 {
1197 if (ptr[1] == CHAR_CIRCUMFLEX_ACCENT)
1198 {
1199 *negptr = TRUE;
1200 ptr++;
1201 }
1202 for (i = 0; i < (int)(sizeof(name) / sizeof(pcre_uchar)) - 1; i++)
1203 {
1204 c = *(++ptr);
1205 if (c == 0) goto ERROR_RETURN;
1206 if (c == CHAR_RIGHT_CURLY_BRACKET) break;
1207 name[i] = c;
1208 }
1209 if (c != CHAR_RIGHT_CURLY_BRACKET) goto ERROR_RETURN;
1210 name[i] = 0;
1211 }
1212
1213 /* Otherwise there is just one following character */
1214
1215 else
1216 {
1217 name[0] = c;
1218 name[1] = 0;
1219 }
1220
1221 *ptrptr = ptr;
1222
1223 /* Search for a recognized property name using binary chop */
1224
1225 bot = 0;
1226 top = PRIV(utt_size);
1227
1228 while (bot < top)
1229 {
1230 i = (bot + top) >> 1;
1231 c = STRCMP_UC_C8(name, PRIV(utt_names) + PRIV(utt)[i].name_offset);
1232 if (c == 0)
1233 {
1234 *dptr = PRIV(utt)[i].value;
1235 return PRIV(utt)[i].type;
1236 }
1237 if (c > 0) bot = i + 1; else top = i;
1238 }
1239
1240 *errorcodeptr = ERR47;
1241 *ptrptr = ptr;
1242 return -1;
1243
1244 ERROR_RETURN:
1245 *errorcodeptr = ERR46;
1246 *ptrptr = ptr;
1247 return -1;
1248 }
1249 #endif
1250
1251
1252
1253
1254 /*************************************************
1255 * Read repeat counts *
1256 *************************************************/
1257
1258 /* Read an item of the form {n,m} and return the values. This is called only
1259 after is_counted_repeat() has confirmed that a repeat-count quantifier exists,
1260 so the syntax is guaranteed to be correct, but we need to check the values.
1261
1262 Arguments:
1263 p pointer to first char after '{'
1264 minp pointer to int for min
1265 maxp pointer to int for max
1266 returned as -1 if no max
1267 errorcodeptr points to error code variable
1268
1269 Returns: pointer to '}' on success;
1270 current ptr on error, with errorcodeptr set non-zero
1271 */
1272
1273 static const pcre_uchar *
1274 read_repeat_counts(const pcre_uchar *p, int *minp, int *maxp, int *errorcodeptr)
1275 {
1276 int min = 0;
1277 int max = -1;
1278
1279 /* Read the minimum value and do a paranoid check: a negative value indicates
1280 an integer overflow. */
1281
1282 while (IS_DIGIT(*p)) min = min * 10 + *p++ - CHAR_0;
1283 if (min < 0 || min > 65535)
1284 {
1285 *errorcodeptr = ERR5;
1286 return p;
1287 }
1288
1289 /* Read the maximum value if there is one, and again do a paranoid on its size.
1290 Also, max must not be less than min. */
1291
1292 if (*p == CHAR_RIGHT_CURLY_BRACKET) max = min; else
1293 {
1294 if (*(++p) != CHAR_RIGHT_CURLY_BRACKET)
1295 {
1296 max = 0;
1297 while(IS_DIGIT(*p)) max = max * 10 + *p++ - CHAR_0;
1298 if (max < 0 || max > 65535)
1299 {
1300 *errorcodeptr = ERR5;
1301 return p;
1302 }
1303 if (max < min)
1304 {
1305 *errorcodeptr = ERR4;
1306 return p;
1307 }
1308 }
1309 }
1310
1311 /* Fill in the required variables, and pass back the pointer to the terminating
1312 '}'. */
1313
1314 *minp = min;
1315 *maxp = max;
1316 return p;
1317 }
1318
1319
1320
1321 /*************************************************
1322 * Subroutine for finding forward reference *
1323 *************************************************/
1324
1325 /* This recursive function is called only from find_parens() below. The
1326 top-level call starts at the beginning of the pattern. All other calls must
1327 start at a parenthesis. It scans along a pattern's text looking for capturing
1328 subpatterns, and counting them. If it finds a named pattern that matches the
1329 name it is given, it returns its number. Alternatively, if the name is NULL, it
1330 returns when it reaches a given numbered subpattern. Recursion is used to keep
1331 track of subpatterns that reset the capturing group numbers - the (?| feature.
1332
1333 This function was originally called only from the second pass, in which we know
1334 that if (?< or (?' or (?P< is encountered, the name will be correctly
1335 terminated because that is checked in the first pass. There is now one call to
1336 this function in the first pass, to check for a recursive back reference by
1337 name (so that we can make the whole group atomic). In this case, we need check
1338 only up to the current position in the pattern, and that is still OK because
1339 and previous occurrences will have been checked. To make this work, the test
1340 for "end of pattern" is a check against cd->end_pattern in the main loop,
1341 instead of looking for a binary zero. This means that the special first-pass
1342 call can adjust cd->end_pattern temporarily. (Checks for binary zero while
1343 processing items within the loop are OK, because afterwards the main loop will
1344 terminate.)
1345
1346 Arguments:
1347 ptrptr address of the current character pointer (updated)
1348 cd compile background data
1349 name name to seek, or NULL if seeking a numbered subpattern
1350 lorn name length, or subpattern number if name is NULL
1351 xmode TRUE if we are in /x mode
1352 utf TRUE if we are in UTF-8 / UTF-16 mode
1353 count pointer to the current capturing subpattern number (updated)
1354
1355 Returns: the number of the named subpattern, or -1 if not found
1356 */
1357
1358 static int
1359 find_parens_sub(pcre_uchar **ptrptr, compile_data *cd, const pcre_uchar *name, int lorn,
1360 BOOL xmode, BOOL utf, int *count)
1361 {
1362 pcre_uchar *ptr = *ptrptr;
1363 int start_count = *count;
1364 int hwm_count = start_count;
1365 BOOL dup_parens = FALSE;
1366
1367 /* If the first character is a parenthesis, check on the type of group we are
1368 dealing with. The very first call may not start with a parenthesis. */
1369
1370 if (ptr[0] == CHAR_LEFT_PARENTHESIS)
1371 {
1372 /* Handle specials such as (*SKIP) or (*UTF8) etc. */
1373
1374 if (ptr[1] == CHAR_ASTERISK) ptr += 2;
1375
1376 /* Handle a normal, unnamed capturing parenthesis. */
1377
1378 else if (ptr[1] != CHAR_QUESTION_MARK)
1379 {
1380 *count += 1;
1381 if (name == NULL && *count == lorn) return *count;
1382 ptr++;
1383 }
1384
1385 /* All cases now have (? at the start. Remember when we are in a group
1386 where the parenthesis numbers are duplicated. */
1387
1388 else if (ptr[2] == CHAR_VERTICAL_LINE)
1389 {
1390 ptr += 3;
1391 dup_parens = TRUE;
1392 }
1393
1394 /* Handle comments; all characters are allowed until a ket is reached. */
1395
1396 else if (ptr[2] == CHAR_NUMBER_SIGN)
1397 {
1398 for (ptr += 3; *ptr != 0; ptr++) if (*ptr == CHAR_RIGHT_PARENTHESIS) break;
1399 goto FAIL_EXIT;
1400 }
1401
1402 /* Handle a condition. If it is an assertion, just carry on so that it
1403 is processed as normal. If not, skip to the closing parenthesis of the
1404 condition (there can't be any nested parens). */
1405
1406 else if (ptr[2] == CHAR_LEFT_PARENTHESIS)
1407 {
1408 ptr += 2;
1409 if (ptr[1] != CHAR_QUESTION_MARK)
1410 {
1411 while (*ptr != 0 && *ptr != CHAR_RIGHT_PARENTHESIS) ptr++;
1412 if (*ptr != 0) ptr++;
1413 }
1414 }
1415
1416 /* Start with (? but not a condition. */
1417
1418 else
1419 {
1420 ptr += 2;
1421 if (*ptr == CHAR_P) ptr++; /* Allow optional P */
1422
1423 /* We have to disambiguate (?<! and (?<= from (?<name> for named groups */
1424
1425 if ((*ptr == CHAR_LESS_THAN_SIGN && ptr[1] != CHAR_EXCLAMATION_MARK &&
1426 ptr[1] != CHAR_EQUALS_SIGN) || *ptr == CHAR_APOSTROPHE)
1427 {
1428 int term;
1429 const pcre_uchar *thisname;
1430 *count += 1;
1431 if (name == NULL && *count == lorn) return *count;
1432 term = *ptr++;
1433 if (term == CHAR_LESS_THAN_SIGN) term = CHAR_GREATER_THAN_SIGN;
1434 thisname = ptr;
1435 while (*ptr != term) ptr++;
1436 if (name != NULL && lorn == ptr - thisname &&
1437 STRNCMP_UC_UC(name, thisname, lorn) == 0)
1438 return *count;
1439 term++;
1440 }
1441 }
1442 }
1443
1444 /* Past any initial parenthesis handling, scan for parentheses or vertical
1445 bars. Stop if we get to cd->end_pattern. Note that this is important for the
1446 first-pass call when this value is temporarily adjusted to stop at the current
1447 position. So DO NOT change this to a test for binary zero. */
1448
1449 for (; ptr < cd->end_pattern; ptr++)
1450 {
1451 /* Skip over backslashed characters and also entire \Q...\E */
1452
1453 if (*ptr == CHAR_BACKSLASH)
1454 {
1455 if (*(++ptr) == 0) goto FAIL_EXIT;
1456 if (*ptr == CHAR_Q) for (;;)
1457 {
1458 while (*(++ptr) != 0 && *ptr != CHAR_BACKSLASH) {};
1459 if (*ptr == 0) goto FAIL_EXIT;
1460 if (*(++ptr) == CHAR_E) break;
1461 }
1462 continue;
1463 }
1464
1465 /* Skip over character classes; this logic must be similar to the way they
1466 are handled for real. If the first character is '^', skip it. Also, if the
1467 first few characters (either before or after ^) are \Q\E or \E we skip them
1468 too. This makes for compatibility with Perl. Note the use of STR macros to
1469 encode "Q\\E" so that it works in UTF-8 on EBCDIC platforms. */
1470
1471 if (*ptr == CHAR_LEFT_SQUARE_BRACKET)
1472 {
1473 BOOL negate_class = FALSE;
1474 for (;;)
1475 {
1476 if (ptr[1] == CHAR_BACKSLASH)
1477 {
1478 if (ptr[2] == CHAR_E)
1479 ptr+= 2;
1480 else if (STRNCMP_UC_C8(ptr + 2,
1481 STR_Q STR_BACKSLASH STR_E, 3) == 0)
1482 ptr += 4;
1483 else
1484 break;
1485 }
1486 else if (!negate_class && ptr[1] == CHAR_CIRCUMFLEX_ACCENT)
1487 {
1488 negate_class = TRUE;
1489 ptr++;
1490 }
1491 else break;
1492 }
1493
1494 /* If the next character is ']', it is a data character that must be
1495 skipped, except in JavaScript compatibility mode. */
1496
1497 if (ptr[1] == CHAR_RIGHT_SQUARE_BRACKET &&
1498 (cd->external_options & PCRE_JAVASCRIPT_COMPAT) == 0)
1499 ptr++;
1500
1501 while (*(++ptr) != CHAR_RIGHT_SQUARE_BRACKET)
1502 {
1503 if (*ptr == 0) return -1;
1504 if (*ptr == CHAR_BACKSLASH)
1505 {
1506 if (*(++ptr) == 0) goto FAIL_EXIT;
1507 if (*ptr == CHAR_Q) for (;;)
1508 {
1509 while (*(++ptr) != 0 && *ptr != CHAR_BACKSLASH) {};
1510 if (*ptr == 0) goto FAIL_EXIT;
1511 if (*(++ptr) == CHAR_E) break;
1512 }
1513 continue;
1514 }
1515 }
1516 continue;
1517 }
1518
1519 /* Skip comments in /x mode */
1520
1521 if (xmode && *ptr == CHAR_NUMBER_SIGN)
1522 {
1523 ptr++;
1524 while (*ptr != 0)
1525 {
1526 if (IS_NEWLINE(ptr)) { ptr += cd->nllen - 1; break; }
1527 ptr++;
1528 #ifdef SUPPORT_UTF
1529 if (utf) FORWARDCHAR(ptr);
1530 #endif
1531 }
1532 if (*ptr == 0) goto FAIL_EXIT;
1533 continue;
1534 }
1535
1536 /* Check for the special metacharacters */
1537
1538 if (*ptr == CHAR_LEFT_PARENTHESIS)
1539 {
1540 int rc = find_parens_sub(&ptr, cd, name, lorn, xmode, utf, count);
1541 if (rc > 0) return rc;
1542 if (*ptr == 0) goto FAIL_EXIT;
1543 }
1544
1545 else if (*ptr == CHAR_RIGHT_PARENTHESIS)
1546 {
1547 if (dup_parens && *count < hwm_count) *count = hwm_count;
1548 goto FAIL_EXIT;
1549 }
1550
1551 else if (*ptr == CHAR_VERTICAL_LINE && dup_parens)
1552 {
1553 if (*count > hwm_count) hwm_count = *count;
1554 *count = start_count;
1555 }
1556 }
1557
1558 FAIL_EXIT:
1559 *ptrptr = ptr;
1560 return -1;
1561 }
1562
1563
1564
1565
1566 /*************************************************
1567 * Find forward referenced subpattern *
1568 *************************************************/
1569
1570 /* This function scans along a pattern's text looking for capturing
1571 subpatterns, and counting them. If it finds a named pattern that matches the
1572 name it is given, it returns its number. Alternatively, if the name is NULL, it
1573 returns when it reaches a given numbered subpattern. This is used for forward
1574 references to subpatterns. We used to be able to start this scan from the
1575 current compiling point, using the current count value from cd->bracount, and
1576 do it all in a single loop, but the addition of the possibility of duplicate
1577 subpattern numbers means that we have to scan from the very start, in order to
1578 take account of such duplicates, and to use a recursive function to keep track
1579 of the different types of group.
1580
1581 Arguments:
1582 cd compile background data
1583 name name to seek, or NULL if seeking a numbered subpattern
1584 lorn name length, or subpattern number if name is NULL
1585 xmode TRUE if we are in /x mode
1586 utf TRUE if we are in UTF-8 / UTF-16 mode
1587
1588 Returns: the number of the found subpattern, or -1 if not found
1589 */
1590
1591 static int
1592 find_parens(compile_data *cd, const pcre_uchar *name, int lorn, BOOL xmode,
1593 BOOL utf)
1594 {
1595 pcre_uchar *ptr = (pcre_uchar *)cd->start_pattern;
1596 int count = 0;
1597 int rc;
1598
1599 /* If the pattern does not start with an opening parenthesis, the first call
1600 to find_parens_sub() will scan right to the end (if necessary). However, if it
1601 does start with a parenthesis, find_parens_sub() will return when it hits the
1602 matching closing parens. That is why we have to have a loop. */
1603
1604 for (;;)
1605 {
1606 rc = find_parens_sub(&ptr, cd, name, lorn, xmode, utf, &count);
1607 if (rc > 0 || *ptr++ == 0) break;
1608 }
1609
1610 return rc;
1611 }
1612
1613
1614
1615
1616 /*************************************************
1617 * Find first significant op code *
1618 *************************************************/
1619
1620 /* This is called by several functions that scan a compiled expression looking
1621 for a fixed first character, or an anchoring op code etc. It skips over things
1622 that do not influence this. For some calls, it makes sense to skip negative
1623 forward and all backward assertions, and also the \b assertion; for others it
1624 does not.
1625
1626 Arguments:
1627 code pointer to the start of the group
1628 skipassert TRUE if certain assertions are to be skipped
1629
1630 Returns: pointer to the first significant opcode
1631 */
1632
1633 static const pcre_uchar*
1634 first_significant_code(const pcre_uchar *code, BOOL skipassert)
1635 {
1636 for (;;)
1637 {
1638 switch ((int)*code)
1639 {
1640 case OP_ASSERT_NOT:
1641 case OP_ASSERTBACK:
1642 case OP_ASSERTBACK_NOT:
1643 if (!skipassert) return code;
1644 do code += GET(code, 1); while (*code == OP_ALT);
1645 code += PRIV(OP_lengths)[*code];
1646 break;
1647
1648 case OP_WORD_BOUNDARY:
1649 case OP_NOT_WORD_BOUNDARY:
1650 if (!skipassert) return code;
1651 /* Fall through */
1652
1653 case OP_CALLOUT:
1654 case OP_CREF:
1655 case OP_NCREF:
1656 case OP_RREF:
1657 case OP_NRREF:
1658 case OP_DEF:
1659 code += PRIV(OP_lengths)[*code];
1660 break;
1661
1662 default:
1663 return code;
1664 }
1665 }
1666 /* Control never reaches here */
1667 }
1668
1669
1670
1671
1672 /*************************************************
1673 * Find the fixed length of a branch *
1674 *************************************************/
1675
1676 /* Scan a branch and compute the fixed length of subject that will match it,
1677 if the length is fixed. This is needed for dealing with backward assertions.
1678 In UTF8 mode, the result is in characters rather than bytes. The branch is
1679 temporarily terminated with OP_END when this function is called.
1680
1681 This function is called when a backward assertion is encountered, so that if it
1682 fails, the error message can point to the correct place in the pattern.
1683 However, we cannot do this when the assertion contains subroutine calls,
1684 because they can be forward references. We solve this by remembering this case
1685 and doing the check at the end; a flag specifies which mode we are running in.
1686
1687 Arguments:
1688 code points to the start of the pattern (the bracket)
1689 utf TRUE in UTF-8 / UTF-16 mode
1690 atend TRUE if called when the pattern is complete
1691 cd the "compile data" structure
1692
1693 Returns: the fixed length,
1694 or -1 if there is no fixed length,
1695 or -2 if \C was encountered (in UTF-8 mode only)
1696 or -3 if an OP_RECURSE item was encountered and atend is FALSE
1697 or -4 if an unknown opcode was encountered (internal error)
1698 */
1699
1700 static int
1701 find_fixedlength(pcre_uchar *code, BOOL utf, BOOL atend, compile_data *cd)
1702 {
1703 int length = -1;
1704
1705 register int branchlength = 0;
1706 register pcre_uchar *cc = code + 1 + LINK_SIZE;
1707
1708 /* Scan along the opcodes for this branch. If we get to the end of the
1709 branch, check the length against that of the other branches. */
1710
1711 for (;;)
1712 {
1713 int d;
1714 pcre_uchar *ce, *cs;
1715 register int op = *cc;
1716
1717 switch (op)
1718 {
1719 /* We only need to continue for OP_CBRA (normal capturing bracket) and
1720 OP_BRA (normal non-capturing bracket) because the other variants of these
1721 opcodes are all concerned with unlimited repeated groups, which of course
1722 are not of fixed length. */
1723
1724 case OP_CBRA:
1725 case OP_BRA:
1726 case OP_ONCE:
1727 case OP_ONCE_NC:
1728 case OP_COND:
1729 d = find_fixedlength(cc + ((op == OP_CBRA)? IMM2_SIZE : 0), utf, atend, cd);
1730 if (d < 0) return d;
1731 branchlength += d;
1732 do cc += GET(cc, 1); while (*cc == OP_ALT);
1733 cc += 1 + LINK_SIZE;
1734 break;
1735
1736 /* Reached end of a branch; if it's a ket it is the end of a nested call.
1737 If it's ALT it is an alternation in a nested call. An ACCEPT is effectively
1738 an ALT. If it is END it's the end of the outer call. All can be handled by
1739 the same code. Note that we must not include the OP_KETRxxx opcodes here,
1740 because they all imply an unlimited repeat. */
1741
1742 case OP_ALT:
1743 case OP_KET:
1744 case OP_END:
1745 case OP_ACCEPT:
1746 case OP_ASSERT_ACCEPT:
1747 if (length < 0) length = branchlength;
1748 else if (length != branchlength) return -1;
1749 if (*cc != OP_ALT) return length;
1750 cc += 1 + LINK_SIZE;
1751 branchlength = 0;
1752 break;
1753
1754 /* A true recursion implies not fixed length, but a subroutine call may
1755 be OK. If the subroutine is a forward reference, we can't deal with
1756 it until the end of the pattern, so return -3. */
1757
1758 case OP_RECURSE:
1759 if (!atend) return -3;
1760 cs = ce = (pcre_uchar *)cd->start_code + GET(cc, 1); /* Start subpattern */
1761 do ce += GET(ce, 1); while (*ce == OP_ALT); /* End subpattern */
1762 if (cc > cs && cc < ce) return -1; /* Recursion */
1763 d = find_fixedlength(cs + IMM2_SIZE, utf, atend, cd);
1764 if (d < 0) return d;
1765 branchlength += d;
1766 cc += 1 + LINK_SIZE;
1767 break;
1768
1769 /* Skip over assertive subpatterns */
1770
1771 case OP_ASSERT:
1772 case OP_ASSERT_NOT:
1773 case OP_ASSERTBACK:
1774 case OP_ASSERTBACK_NOT:
1775 do cc += GET(cc, 1); while (*cc == OP_ALT);
1776 cc += PRIV(OP_lengths)[*cc];
1777 break;
1778
1779 /* Skip over things that don't match chars */
1780
1781 case OP_MARK:
1782 case OP_PRUNE_ARG:
1783 case OP_SKIP_ARG:
1784 case OP_THEN_ARG:
1785 cc += cc[1] + PRIV(OP_lengths)[*cc];
1786 break;
1787
1788 case OP_CALLOUT:
1789 case OP_CIRC:
1790 case OP_CIRCM:
1791 case OP_CLOSE:
1792 case OP_COMMIT:
1793 case OP_CREF:
1794 case OP_DEF:
1795 case OP_DOLL:
1796 case OP_DOLLM:
1797 case OP_EOD:
1798 case OP_EODN:
1799 case OP_FAIL:
1800 case OP_NCREF:
1801 case OP_NRREF:
1802 case OP_NOT_WORD_BOUNDARY:
1803 case OP_PRUNE:
1804 case OP_REVERSE:
1805 case OP_RREF:
1806 case OP_SET_SOM:
1807 case OP_SKIP:
1808 case OP_SOD:
1809 case OP_SOM:
1810 case OP_THEN:
1811 case OP_WORD_BOUNDARY:
1812 cc += PRIV(OP_lengths)[*cc];
1813 break;
1814
1815 /* Handle literal characters */
1816
1817 case OP_CHAR:
1818 case OP_CHARI:
1819 case OP_NOT:
1820 case OP_NOTI:
1821 branchlength++;
1822 cc += 2;
1823 #ifdef SUPPORT_UTF
1824 if (utf && HAS_EXTRALEN(cc[-1])) cc += GET_EXTRALEN(cc[-1]);
1825 #endif
1826 break;
1827
1828 /* Handle exact repetitions. The count is already in characters, but we
1829 need to skip over a multibyte character in UTF8 mode. */
1830
1831 case OP_EXACT:
1832 case OP_EXACTI:
1833 case OP_NOTEXACT:
1834 case OP_NOTEXACTI:
1835 branchlength += GET2(cc,1);
1836 cc += 2 + IMM2_SIZE;
1837 #ifdef SUPPORT_UTF
1838 if (utf && HAS_EXTRALEN(cc[-1])) cc += GET_EXTRALEN(cc[-1]);
1839 #endif
1840 break;
1841
1842 case OP_TYPEEXACT:
1843 branchlength += GET2(cc,1);
1844 if (cc[1 + IMM2_SIZE] == OP_PROP || cc[1 + IMM2_SIZE] == OP_NOTPROP) cc += 2;
1845 cc += 1 + IMM2_SIZE + 1;
1846 break;
1847
1848 /* Handle single-char matchers */
1849
1850 case OP_PROP:
1851 case OP_NOTPROP:
1852 cc += 2;
1853 /* Fall through */
1854
1855 case OP_HSPACE:
1856 case OP_VSPACE:
1857 case OP_NOT_HSPACE:
1858 case OP_NOT_VSPACE:
1859 case OP_NOT_DIGIT:
1860 case OP_DIGIT:
1861 case OP_NOT_WHITESPACE:
1862 case OP_WHITESPACE:
1863 case OP_NOT_WORDCHAR:
1864 case OP_WORDCHAR:
1865 case OP_ANY:
1866 case OP_ALLANY:
1867 branchlength++;
1868 cc++;
1869 break;
1870
1871 /* The single-byte matcher isn't allowed. This only happens in UTF-8 mode;
1872 otherwise \C is coded as OP_ALLANY. */
1873
1874 case OP_ANYBYTE:
1875 return -2;
1876
1877 /* Check a class for variable quantification */
1878
1879 #if defined SUPPORT_UTF || defined COMPILE_PCRE16
1880 case OP_XCLASS:
1881 cc += GET(cc, 1) - PRIV(OP_lengths)[OP_CLASS];
1882 /* Fall through */
1883 #endif
1884
1885 case OP_CLASS:
1886 case OP_NCLASS:
1887 cc += PRIV(OP_lengths)[OP_CLASS];
1888
1889 switch (*cc)
1890 {
1891 case OP_CRPLUS:
1892 case OP_CRMINPLUS:
1893 case OP_CRSTAR:
1894 case OP_CRMINSTAR:
1895 case OP_CRQUERY:
1896 case OP_CRMINQUERY:
1897 return -1;
1898
1899 case OP_CRRANGE:
1900 case OP_CRMINRANGE:
1901 if (GET2(cc,1) != GET2(cc,1+IMM2_SIZE)) return -1;
1902 branchlength += GET2(cc,1);
1903 cc += 1 + 2 * IMM2_SIZE;
1904 break;
1905
1906 default:
1907 branchlength++;
1908 }
1909 break;
1910
1911 /* Anything else is variable length */
1912
1913 case OP_ANYNL:
1914 case OP_BRAMINZERO:
1915 case OP_BRAPOS:
1916 case OP_BRAPOSZERO:
1917 case OP_BRAZERO:
1918 case OP_CBRAPOS:
1919 case OP_EXTUNI:
1920 case OP_KETRMAX:
1921 case OP_KETRMIN:
1922 case OP_KETRPOS:
1923 case OP_MINPLUS:
1924 case OP_MINPLUSI:
1925 case OP_MINQUERY:
1926 case OP_MINQUERYI:
1927 case OP_MINSTAR:
1928 case OP_MINSTARI:
1929 case OP_MINUPTO:
1930 case OP_MINUPTOI:
1931 case OP_NOTMINPLUS:
1932 case OP_NOTMINPLUSI:
1933 case OP_NOTMINQUERY:
1934 case OP_NOTMINQUERYI:
1935 case OP_NOTMINSTAR:
1936 case OP_NOTMINSTARI:
1937 case OP_NOTMINUPTO:
1938 case OP_NOTMINUPTOI:
1939 case OP_NOTPLUS:
1940 case OP_NOTPLUSI:
1941 case OP_NOTPOSPLUS:
1942 case OP_NOTPOSPLUSI:
1943 case OP_NOTPOSQUERY:
1944 case OP_NOTPOSQUERYI:
1945 case OP_NOTPOSSTAR:
1946 case OP_NOTPOSSTARI:
1947 case OP_NOTPOSUPTO:
1948 case OP_NOTPOSUPTOI:
1949 case OP_NOTQUERY:
1950 case OP_NOTQUERYI:
1951 case OP_NOTSTAR:
1952 case OP_NOTSTARI:
1953 case OP_NOTUPTO:
1954 case OP_NOTUPTOI:
1955 case OP_PLUS:
1956 case OP_PLUSI:
1957 case OP_POSPLUS:
1958 case OP_POSPLUSI:
1959 case OP_POSQUERY:
1960 case OP_POSQUERYI:
1961 case OP_POSSTAR:
1962 case OP_POSSTARI:
1963 case OP_POSUPTO:
1964 case OP_POSUPTOI:
1965 case OP_QUERY:
1966 case OP_QUERYI:
1967 case OP_REF:
1968 case OP_REFI:
1969 case OP_SBRA:
1970 case OP_SBRAPOS:
1971 case OP_SCBRA:
1972 case OP_SCBRAPOS:
1973 case OP_SCOND:
1974 case OP_SKIPZERO:
1975 case OP_STAR:
1976 case OP_STARI:
1977 case OP_TYPEMINPLUS:
1978 case OP_TYPEMINQUERY:
1979 case OP_TYPEMINSTAR:
1980 case OP_TYPEMINUPTO:
1981 case OP_TYPEPLUS:
1982 case OP_TYPEPOSPLUS:
1983 case OP_TYPEPOSQUERY:
1984 case OP_TYPEPOSSTAR:
1985 case OP_TYPEPOSUPTO:
1986 case OP_TYPEQUERY:
1987 case OP_TYPESTAR:
1988 case OP_TYPEUPTO:
1989 case OP_UPTO:
1990 case OP_UPTOI:
1991 return -1;
1992
1993 /* Catch unrecognized opcodes so that when new ones are added they
1994 are not forgotten, as has happened in the past. */
1995
1996 default:
1997 return -4;
1998 }
1999 }
2000 /* Control never gets here */
2001 }
2002
2003
2004
2005
2006 /*************************************************
2007 * Scan compiled regex for specific bracket *
2008 *************************************************/
2009
2010 /* This little function scans through a compiled pattern until it finds a
2011 capturing bracket with the given number, or, if the number is negative, an
2012 instance of OP_REVERSE for a lookbehind. The function is global in the C sense
2013 so that it can be called from pcre_study() when finding the minimum matching
2014 length.
2015
2016 Arguments:
2017 code points to start of expression
2018 utf TRUE in UTF-8 / UTF-16 mode
2019 number the required bracket number or negative to find a lookbehind
2020
2021 Returns: pointer to the opcode for the bracket, or NULL if not found
2022 */
2023
2024 const pcre_uchar *
2025 PRIV(find_bracket)(const pcre_uchar *code, BOOL utf, int number)
2026 {
2027 for (;;)
2028 {
2029 register int c = *code;
2030
2031 if (c == OP_END) return NULL;
2032
2033 /* XCLASS is used for classes that cannot be represented just by a bit
2034 map. This includes negated single high-valued characters. The length in
2035 the table is zero; the actual length is stored in the compiled code. */
2036
2037 if (c == OP_XCLASS) code += GET(code, 1);
2038
2039 /* Handle recursion */
2040
2041 else if (c == OP_REVERSE)
2042 {
2043 if (number < 0) return (pcre_uchar *)code;
2044 code += PRIV(OP_lengths)[c];
2045 }
2046
2047 /* Handle capturing bracket */
2048
2049 else if (c == OP_CBRA || c == OP_SCBRA ||
2050 c == OP_CBRAPOS || c == OP_SCBRAPOS)
2051 {
2052 int n = GET2(code, 1+LINK_SIZE);
2053 if (n == number) return (pcre_uchar *)code;
2054 code += PRIV(OP_lengths)[c];
2055 }
2056
2057 /* Otherwise, we can get the item's length from the table, except that for
2058 repeated character types, we have to test for \p and \P, which have an extra
2059 two bytes of parameters, and for MARK/PRUNE/SKIP/THEN with an argument, we
2060 must add in its length. */
2061
2062 else
2063 {
2064 switch(c)
2065 {
2066 case OP_TYPESTAR:
2067 case OP_TYPEMINSTAR:
2068 case OP_TYPEPLUS:
2069 case OP_TYPEMINPLUS:
2070 case OP_TYPEQUERY:
2071 case OP_TYPEMINQUERY:
2072 case OP_TYPEPOSSTAR:
2073 case OP_TYPEPOSPLUS:
2074 case OP_TYPEPOSQUERY:
2075 if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
2076 break;
2077
2078 case OP_TYPEUPTO:
2079 case OP_TYPEMINUPTO:
2080 case OP_TYPEEXACT:
2081 case OP_TYPEPOSUPTO:
2082 if (code[1 + IMM2_SIZE] == OP_PROP
2083 || code[1 + IMM2_SIZE] == OP_NOTPROP) code += 2;
2084 break;
2085
2086 case OP_MARK:
2087 case OP_PRUNE_ARG:
2088 case OP_SKIP_ARG:
2089 code += code[1];
2090 break;
2091
2092 case OP_THEN_ARG:
2093 code += code[1];
2094 break;
2095 }
2096
2097 /* Add in the fixed length from the table */
2098
2099 code += PRIV(OP_lengths)[c];
2100
2101 /* In UTF-8 mode, opcodes that are followed by a character may be followed by
2102 a multi-byte character. The length in the table is a minimum, so we have to
2103 arrange to skip the extra bytes. */
2104
2105 #ifdef SUPPORT_UTF
2106 if (utf) switch(c)
2107 {
2108 case OP_CHAR:
2109 case OP_CHARI:
2110 case OP_EXACT:
2111 case OP_EXACTI:
2112 case OP_UPTO:
2113 case OP_UPTOI:
2114 case OP_MINUPTO:
2115 case OP_MINUPTOI:
2116 case OP_POSUPTO:
2117 case OP_POSUPTOI:
2118 case OP_STAR:
2119 case OP_STARI:
2120 case OP_MINSTAR:
2121 case OP_MINSTARI:
2122 case OP_POSSTAR:
2123 case OP_POSSTARI:
2124 case OP_PLUS:
2125 case OP_PLUSI:
2126 case OP_MINPLUS:
2127 case OP_MINPLUSI:
2128 case OP_POSPLUS:
2129 case OP_POSPLUSI:
2130 case OP_QUERY:
2131 case OP_QUERYI:
2132 case OP_MINQUERY:
2133 case OP_MINQUERYI:
2134 case OP_POSQUERY:
2135 case OP_POSQUERYI:
2136 if (HAS_EXTRALEN(code[-1])) code += GET_EXTRALEN(code[-1]);
2137 break;
2138 }
2139 #else
2140 (void)(utf); /* Keep compiler happy by referencing function argument */
2141 #endif
2142 }
2143 }
2144 }
2145
2146
2147
2148 /*************************************************
2149 * Scan compiled regex for recursion reference *
2150 *************************************************/
2151
2152 /* This little function scans through a compiled pattern until it finds an
2153 instance of OP_RECURSE.
2154
2155 Arguments:
2156 code points to start of expression
2157 utf TRUE in UTF-8 / UTF-16 mode
2158
2159 Returns: pointer to the opcode for OP_RECURSE, or NULL if not found
2160 */
2161
2162 static const pcre_uchar *
2163 find_recurse(const pcre_uchar *code, BOOL utf)
2164 {
2165 for (;;)
2166 {
2167 register int c = *code;
2168 if (c == OP_END) return NULL;
2169 if (c == OP_RECURSE) return code;
2170
2171 /* XCLASS is used for classes that cannot be represented just by a bit
2172 map. This includes negated single high-valued characters. The length in
2173 the table is zero; the actual length is stored in the compiled code. */
2174
2175 if (c == OP_XCLASS) code += GET(code, 1);
2176
2177 /* Otherwise, we can get the item's length from the table, except that for
2178 repeated character types, we have to test for \p and \P, which have an extra
2179 two bytes of parameters, and for MARK/PRUNE/SKIP/THEN with an argument, we
2180 must add in its length. */
2181
2182 else
2183 {
2184 switch(c)
2185 {
2186 case OP_TYPESTAR:
2187 case OP_TYPEMINSTAR:
2188 case OP_TYPEPLUS:
2189 case OP_TYPEMINPLUS:
2190 case OP_TYPEQUERY:
2191 case OP_TYPEMINQUERY:
2192 case OP_TYPEPOSSTAR:
2193 case OP_TYPEPOSPLUS:
2194 case OP_TYPEPOSQUERY:
2195 if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
2196 break;
2197
2198 case OP_TYPEPOSUPTO:
2199 case OP_TYPEUPTO:
2200 case OP_TYPEMINUPTO:
2201 case OP_TYPEEXACT:
2202 if (code[1 + IMM2_SIZE] == OP_PROP
2203 || code[1 + IMM2_SIZE] == OP_NOTPROP) code += 2;
2204 break;
2205
2206 case OP_MARK:
2207 case OP_PRUNE_ARG:
2208 case OP_SKIP_ARG:
2209 code += code[1];
2210 break;
2211
2212 case OP_THEN_ARG:
2213 code += code[1];
2214 break;
2215 }
2216
2217 /* Add in the fixed length from the table */
2218
2219 code += PRIV(OP_lengths)[c];
2220
2221 /* In UTF-8 mode, opcodes that are followed by a character may be followed
2222 by a multi-byte character. The length in the table is a minimum, so we have
2223 to arrange to skip the extra bytes. */
2224
2225 #ifdef SUPPORT_UTF
2226 if (utf) switch(c)
2227 {
2228 case OP_CHAR:
2229 case OP_CHARI:
2230 case OP_NOT:
2231 case OP_NOTI:
2232 case OP_EXACT:
2233 case OP_EXACTI:
2234 case OP_NOTEXACT:
2235 case OP_NOTEXACTI:
2236 case OP_UPTO:
2237 case OP_UPTOI:
2238 case OP_NOTUPTO:
2239 case OP_NOTUPTOI:
2240 case OP_MINUPTO:
2241 case OP_MINUPTOI:
2242 case OP_NOTMINUPTO:
2243 case OP_NOTMINUPTOI:
2244 case OP_POSUPTO:
2245 case OP_POSUPTOI:
2246 case OP_NOTPOSUPTO:
2247 case OP_NOTPOSUPTOI:
2248 case OP_STAR:
2249 case OP_STARI:
2250 case OP_NOTSTAR:
2251 case OP_NOTSTARI:
2252 case OP_MINSTAR:
2253 case OP_MINSTARI:
2254 case OP_NOTMINSTAR:
2255 case OP_NOTMINSTARI:
2256 case OP_POSSTAR:
2257 case OP_POSSTARI:
2258 case OP_NOTPOSSTAR:
2259 case OP_NOTPOSSTARI:
2260 case OP_PLUS:
2261 case OP_PLUSI:
2262 case OP_NOTPLUS:
2263 case OP_NOTPLUSI:
2264 case OP_MINPLUS:
2265 case OP_MINPLUSI:
2266 case OP_NOTMINPLUS:
2267 case OP_NOTMINPLUSI:
2268 case OP_POSPLUS:
2269 case OP_POSPLUSI:
2270 case OP_NOTPOSPLUS:
2271 case OP_NOTPOSPLUSI:
2272 case OP_QUERY:
2273 case OP_QUERYI:
2274 case OP_NOTQUERY:
2275 case OP_NOTQUERYI:
2276 case OP_MINQUERY:
2277 case OP_MINQUERYI:
2278 case OP_NOTMINQUERY:
2279 case OP_NOTMINQUERYI:
2280 case OP_POSQUERY:
2281 case OP_POSQUERYI:
2282 case OP_NOTPOSQUERY:
2283 case OP_NOTPOSQUERYI:
2284 if (HAS_EXTRALEN(code[-1])) code += GET_EXTRALEN(code[-1]);
2285 break;
2286 }
2287 #else
2288 (void)(utf); /* Keep compiler happy by referencing function argument */
2289 #endif
2290 }
2291 }
2292 }
2293
2294
2295
2296 /*************************************************
2297 * Scan compiled branch for non-emptiness *
2298 *************************************************/
2299
2300 /* This function scans through a branch of a compiled pattern to see whether it
2301 can match the empty string or not. It is called from could_be_empty()
2302 below and from compile_branch() when checking for an unlimited repeat of a
2303 group that can match nothing. Note that first_significant_code() skips over
2304 backward and negative forward assertions when its final argument is TRUE. If we
2305 hit an unclosed bracket, we return "empty" - this means we've struck an inner
2306 bracket whose current branch will already have been scanned.
2307
2308 Arguments:
2309 code points to start of search
2310 endcode points to where to stop
2311 utf TRUE if in UTF-8 / UTF-16 mode
2312 cd contains pointers to tables etc.
2313
2314 Returns: TRUE if what is matched could be empty
2315 */
2316
2317 static BOOL
2318 could_be_empty_branch(const pcre_uchar *code, const pcre_uchar *endcode,
2319 BOOL utf, compile_data *cd)
2320 {
2321 register int c;
2322 for (code = first_significant_code(code + PRIV(OP_lengths)[*code], TRUE);
2323 code < endcode;
2324 code = first_significant_code(code + PRIV(OP_lengths)[c], TRUE))
2325 {
2326 const pcre_uchar *ccode;
2327
2328 c = *code;
2329
2330 /* Skip over forward assertions; the other assertions are skipped by
2331 first_significant_code() with a TRUE final argument. */
2332
2333 if (c == OP_ASSERT)
2334 {
2335 do code += GET(code, 1); while (*code == OP_ALT);
2336 c = *code;
2337 continue;
2338 }
2339
2340 /* For a recursion/subroutine call, if its end has been reached, which
2341 implies a backward reference subroutine call, we can scan it. If it's a
2342 forward reference subroutine call, we can't. To detect forward reference
2343 we have to scan up the list that is kept in the workspace. This function is
2344 called only when doing the real compile, not during the pre-compile that
2345 measures the size of the compiled pattern. */
2346
2347 if (c == OP_RECURSE)
2348 {
2349 const pcre_uchar *scode;
2350 BOOL empty_branch;
2351
2352 /* Test for forward reference */
2353
2354 for (scode = cd->start_workspace; scode < cd->hwm; scode += LINK_SIZE)
2355 if (GET(scode, 0) == code + 1 - cd->start_code) return TRUE;
2356
2357 /* Not a forward reference, test for completed backward reference */
2358
2359 empty_branch = FALSE;
2360 scode = cd->start_code + GET(code, 1);
2361 if (GET(scode, 1) == 0) return TRUE; /* Unclosed */
2362
2363 /* Completed backwards reference */
2364
2365 do
2366 {
2367 if (could_be_empty_branch(scode, endcode, utf, cd))
2368 {
2369 empty_branch = TRUE;
2370 break;
2371 }
2372 scode += GET(scode, 1);
2373 }
2374 while (*scode == OP_ALT);
2375
2376 if (!empty_branch) return FALSE; /* All branches are non-empty */
2377 continue;
2378 }
2379
2380 /* Groups with zero repeats can of course be empty; skip them. */
2381
2382 if (c == OP_BRAZERO || c == OP_BRAMINZERO || c == OP_SKIPZERO ||
2383 c == OP_BRAPOSZERO)
2384 {
2385 code += PRIV(OP_lengths)[c];
2386 do code += GET(code, 1); while (*code == OP_ALT);
2387 c = *code;
2388 continue;
2389 }
2390
2391 /* A nested group that is already marked as "could be empty" can just be
2392 skipped. */
2393
2394 if (c == OP_SBRA || c == OP_SBRAPOS ||
2395 c == OP_SCBRA || c == OP_SCBRAPOS)
2396 {
2397 do code += GET(code, 1); while (*code == OP_ALT);
2398 c = *code;
2399 continue;
2400 }
2401
2402 /* For other groups, scan the branches. */
2403
2404 if (c == OP_BRA || c == OP_BRAPOS ||
2405 c == OP_CBRA || c == OP_CBRAPOS ||
2406 c == OP_ONCE || c == OP_ONCE_NC ||
2407 c == OP_COND)
2408 {
2409 BOOL empty_branch;
2410 if (GET(code, 1) == 0) return TRUE; /* Hit unclosed bracket */
2411
2412 /* If a conditional group has only one branch, there is a second, implied,
2413 empty branch, so just skip over the conditional, because it could be empty.
2414 Otherwise, scan the individual branches of the group. */
2415
2416 if (c == OP_COND && code[GET(code, 1)] != OP_ALT)
2417 code += GET(code, 1);
2418 else
2419 {
2420 empty_branch = FALSE;
2421 do
2422 {
2423 if (!empty_branch && could_be_empty_branch(code, endcode, utf, cd))
2424 empty_branch = TRUE;
2425 code += GET(code, 1);
2426 }
2427 while (*code == OP_ALT);
2428 if (!empty_branch) return FALSE; /* All branches are non-empty */
2429 }
2430
2431 c = *code;
2432 continue;
2433 }
2434
2435 /* Handle the other opcodes */
2436
2437 switch (c)
2438 {
2439 /* Check for quantifiers after a class. XCLASS is used for classes that
2440 cannot be represented just by a bit map. This includes negated single
2441 high-valued characters. The length in PRIV(OP_lengths)[] is zero; the
2442 actual length is stored in the compiled code, so we must update "code"
2443 here. */
2444
2445 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
2446 case OP_XCLASS:
2447 ccode = code += GET(code, 1);
2448 goto CHECK_CLASS_REPEAT;
2449 #endif
2450
2451 case OP_CLASS:
2452 case OP_NCLASS:
2453 ccode = code + PRIV(OP_lengths)[OP_CLASS];
2454
2455 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
2456 CHECK_CLASS_REPEAT:
2457 #endif
2458
2459 switch (*ccode)
2460 {
2461 case OP_CRSTAR: /* These could be empty; continue */
2462 case OP_CRMINSTAR:
2463 case OP_CRQUERY:
2464 case OP_CRMINQUERY:
2465 break;
2466
2467 default: /* Non-repeat => class must match */
2468 case OP_CRPLUS: /* These repeats aren't empty */
2469 case OP_CRMINPLUS:
2470 return FALSE;
2471
2472 case OP_CRRANGE:
2473 case OP_CRMINRANGE:
2474 if (GET2(ccode, 1) > 0) return FALSE; /* Minimum > 0 */
2475 break;
2476 }
2477 break;
2478
2479 /* Opcodes that must match a character */
2480
2481 case OP_PROP:
2482 case OP_NOTPROP:
2483 case OP_EXTUNI:
2484 case OP_NOT_DIGIT:
2485 case OP_DIGIT:
2486 case OP_NOT_WHITESPACE:
2487 case OP_WHITESPACE:
2488 case OP_NOT_WORDCHAR:
2489 case OP_WORDCHAR:
2490 case OP_ANY:
2491 case OP_ALLANY:
2492 case OP_ANYBYTE:
2493 case OP_CHAR:
2494 case OP_CHARI:
2495 case OP_NOT:
2496 case OP_NOTI:
2497 case OP_PLUS:
2498 case OP_MINPLUS:
2499 case OP_POSPLUS:
2500 case OP_EXACT:
2501 case OP_NOTPLUS:
2502 case OP_NOTMINPLUS:
2503 case OP_NOTPOSPLUS:
2504 case OP_NOTEXACT:
2505 case OP_TYPEPLUS:
2506 case OP_TYPEMINPLUS:
2507 case OP_TYPEPOSPLUS:
2508 case OP_TYPEEXACT:
2509 return FALSE;
2510
2511 /* These are going to continue, as they may be empty, but we have to
2512 fudge the length for the \p and \P cases. */
2513
2514 case OP_TYPESTAR:
2515 case OP_TYPEMINSTAR:
2516 case OP_TYPEPOSSTAR:
2517 case OP_TYPEQUERY:
2518 case OP_TYPEMINQUERY:
2519 case OP_TYPEPOSQUERY:
2520 if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
2521 break;
2522
2523 /* Same for these */
2524
2525 case OP_TYPEUPTO:
2526 case OP_TYPEMINUPTO:
2527 case OP_TYPEPOSUPTO:
2528 if (code[1 + IMM2_SIZE] == OP_PROP
2529 || code[1 + IMM2_SIZE] == OP_NOTPROP) code += 2;
2530 break;
2531
2532 /* End of branch */
2533
2534 case OP_KET:
2535 case OP_KETRMAX:
2536 case OP_KETRMIN:
2537 case OP_KETRPOS:
2538 case OP_ALT:
2539 return TRUE;
2540
2541 /* In UTF-8 mode, STAR, MINSTAR, POSSTAR, QUERY, MINQUERY, POSQUERY, UPTO,
2542 MINUPTO, and POSUPTO may be followed by a multibyte character */
2543
2544 #ifdef SUPPORT_UTF
2545 case OP_STAR:
2546 case OP_STARI:
2547 case OP_MINSTAR:
2548 case OP_MINSTARI:
2549 case OP_POSSTAR:
2550 case OP_POSSTARI:
2551 case OP_QUERY:
2552 case OP_QUERYI:
2553 case OP_MINQUERY:
2554 case OP_MINQUERYI:
2555 case OP_POSQUERY:
2556 case OP_POSQUERYI:
2557 if (utf && HAS_EXTRALEN(code[1])) code += GET_EXTRALEN(code[1]);
2558 break;
2559
2560 case OP_UPTO:
2561 case OP_UPTOI:
2562 case OP_MINUPTO:
2563 case OP_MINUPTOI:
2564 case OP_POSUPTO:
2565 case OP_POSUPTOI:
2566 if (utf && HAS_EXTRALEN(code[1 + IMM2_SIZE])) code += GET_EXTRALEN(code[1 + IMM2_SIZE]);
2567 break;
2568 #endif
2569
2570 /* MARK, and PRUNE/SKIP/THEN with an argument must skip over the argument
2571 string. */
2572
2573 case OP_MARK:
2574 case OP_PRUNE_ARG:
2575 case OP_SKIP_ARG:
2576 code += code[1];
2577 break;
2578
2579 case OP_THEN_ARG:
2580 code += code[1];
2581 break;
2582
2583 /* None of the remaining opcodes are required to match a character. */
2584
2585 default:
2586 break;
2587 }
2588 }
2589
2590 return TRUE;
2591 }
2592
2593
2594
2595 /*************************************************
2596 * Scan compiled regex for non-emptiness *
2597 *************************************************/
2598
2599 /* This function is called to check for left recursive calls. We want to check
2600 the current branch of the current pattern to see if it could match the empty
2601 string. If it could, we must look outwards for branches at other levels,
2602 stopping when we pass beyond the bracket which is the subject of the recursion.
2603 This function is called only during the real compile, not during the
2604 pre-compile.
2605
2606 Arguments:
2607 code points to start of the recursion
2608 endcode points to where to stop (current RECURSE item)
2609 bcptr points to the chain of current (unclosed) branch starts
2610 utf TRUE if in UTF-8 / UTF-16 mode
2611 cd pointers to tables etc
2612
2613 Returns: TRUE if what is matched could be empty
2614 */
2615
2616 static BOOL
2617 could_be_empty(const pcre_uchar *code, const pcre_uchar *endcode,
2618 branch_chain *bcptr, BOOL utf, compile_data *cd)
2619 {
2620 while (bcptr != NULL && bcptr->current_branch >= code)
2621 {
2622 if (!could_be_empty_branch(bcptr->current_branch, endcode, utf, cd))
2623 return FALSE;
2624 bcptr = bcptr->outer;
2625 }
2626 return TRUE;
2627 }
2628
2629
2630
2631 /*************************************************
2632 * Check for POSIX class syntax *
2633 *************************************************/
2634
2635 /* This function is called when the sequence "[:" or "[." or "[=" is
2636 encountered in a character class. It checks whether this is followed by a
2637 sequence of characters terminated by a matching ":]" or ".]" or "=]". If we
2638 reach an unescaped ']' without the special preceding character, return FALSE.
2639
2640 Originally, this function only recognized a sequence of letters between the
2641 terminators, but it seems that Perl recognizes any sequence of characters,
2642 though of course unknown POSIX names are subsequently rejected. Perl gives an
2643 "Unknown POSIX class" error for [:f\oo:] for example, where previously PCRE
2644 didn't consider this to be a POSIX class. Likewise for [:1234:].
2645
2646 The problem in trying to be exactly like Perl is in the handling of escapes. We
2647 have to be sure that [abc[:x\]pqr] is *not* treated as containing a POSIX
2648 class, but [abc[:x\]pqr:]] is (so that an error can be generated). The code
2649 below handles the special case of \], but does not try to do any other escape
2650 processing. This makes it different from Perl for cases such as [:l\ower:]
2651 where Perl recognizes it as the POSIX class "lower" but PCRE does not recognize
2652 "l\ower". This is a lesser evil that not diagnosing bad classes when Perl does,
2653 I think.
2654
2655 A user pointed out that PCRE was rejecting [:a[:digit:]] whereas Perl was not.
2656 It seems that the appearance of a nested POSIX class supersedes an apparent
2657 external class. For example, [:a[:digit:]b:] matches "a", "b", ":", or
2658 a digit.
2659
2660 In Perl, unescaped square brackets may also appear as part of class names. For
2661 example, [:a[:abc]b:] gives unknown POSIX class "[:abc]b:]". However, for
2662 [:a[:abc]b][b:] it gives unknown POSIX class "[:abc]b][b:]", which does not
2663 seem right at all. PCRE does not allow closing square brackets in POSIX class
2664 names.
2665
2666 Arguments:
2667 ptr pointer to the initial [
2668 endptr where to return the end pointer
2669
2670 Returns: TRUE or FALSE
2671 */
2672
2673 static BOOL
2674 check_posix_syntax(const pcre_uchar *ptr, const pcre_uchar **endptr)
2675 {
2676 int terminator; /* Don't combine these lines; the Solaris cc */
2677 terminator = *(++ptr); /* compiler warns about "non-constant" initializer. */
2678 for (++ptr; *ptr != 0; ptr++)
2679 {
2680 if (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET)
2681 ptr++;
2682 else if (*ptr == CHAR_RIGHT_SQUARE_BRACKET) return FALSE;
2683 else
2684 {
2685 if (*ptr == terminator && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET)
2686 {
2687 *endptr = ptr;
2688 return TRUE;
2689 }
2690 if (*ptr == CHAR_LEFT_SQUARE_BRACKET &&
2691 (ptr[1] == CHAR_COLON || ptr[1] == CHAR_DOT ||
2692 ptr[1] == CHAR_EQUALS_SIGN) &&
2693 check_posix_syntax(ptr, endptr))
2694 return FALSE;
2695 }
2696 }
2697 return FALSE;
2698 }
2699
2700
2701
2702
2703 /*************************************************
2704 * Check POSIX class name *
2705 *************************************************/
2706
2707 /* This function is called to check the name given in a POSIX-style class entry
2708 such as [:alnum:].
2709
2710 Arguments:
2711 ptr points to the first letter
2712 len the length of the name
2713
2714 Returns: a value representing the name, or -1 if unknown
2715 */
2716
2717 static int
2718 check_posix_name(const pcre_uchar *ptr, int len)
2719 {
2720 const char *pn = posix_names;
2721 register int yield = 0;
2722 while (posix_name_lengths[yield] != 0)
2723 {
2724 if (len == posix_name_lengths[yield] &&
2725 STRNCMP_UC_C8(ptr, pn, len) == 0) return yield;
2726 pn += posix_name_lengths[yield] + 1;
2727 yield++;
2728 }
2729 return -1;
2730 }
2731
2732
2733 /*************************************************
2734 * Adjust OP_RECURSE items in repeated group *
2735 *************************************************/
2736
2737 /* OP_RECURSE items contain an offset from the start of the regex to the group
2738 that is referenced. This means that groups can be replicated for fixed
2739 repetition simply by copying (because the recursion is allowed to refer to
2740 earlier groups that are outside the current group). However, when a group is
2741 optional (i.e. the minimum quantifier is zero), OP_BRAZERO or OP_SKIPZERO is
2742 inserted before it, after it has been compiled. This means that any OP_RECURSE
2743 items within it that refer to the group itself or any contained groups have to
2744 have their offsets adjusted. That one of the jobs of this function. Before it
2745 is called, the partially compiled regex must be temporarily terminated with
2746 OP_END.
2747
2748 This function has been extended with the possibility of forward references for
2749 recursions and subroutine calls. It must also check the list of such references
2750 for the group we are dealing with. If it finds that one of the recursions in
2751 the current group is on this list, it adjusts the offset in the list, not the
2752 value in the reference (which is a group number).
2753
2754 Arguments:
2755 group points to the start of the group
2756 adjust the amount by which the group is to be moved
2757 utf TRUE in UTF-8 / UTF-16 mode
2758 cd contains pointers to tables etc.
2759 save_hwm the hwm forward reference pointer at the start of the group
2760
2761 Returns: nothing
2762 */
2763
2764 static void
2765 adjust_recurse(pcre_uchar *group, int adjust, BOOL utf, compile_data *cd,
2766 pcre_uchar *save_hwm)
2767 {
2768 pcre_uchar *ptr = group;
2769
2770 while ((ptr = (pcre_uchar *)find_recurse(ptr, utf)) != NULL)
2771 {
2772 int offset;
2773 pcre_uchar *hc;
2774
2775 /* See if this recursion is on the forward reference list. If so, adjust the
2776 reference. */
2777
2778 for (hc = save_hwm; hc < cd->hwm; hc += LINK_SIZE)
2779 {
2780 offset = GET(hc, 0);
2781 if (cd->start_code + offset == ptr + 1)
2782 {
2783 PUT(hc, 0, offset + adjust);
2784 break;
2785 }
2786 }
2787
2788 /* Otherwise, adjust the recursion offset if it's after the start of this
2789 group. */
2790
2791 if (hc >= cd->hwm)
2792 {
2793 offset = GET(ptr, 1);
2794 if (cd->start_code + offset >= group) PUT(ptr, 1, offset + adjust);
2795 }
2796
2797 ptr += 1 + LINK_SIZE;
2798 }
2799 }
2800
2801
2802
2803 /*************************************************
2804 * Insert an automatic callout point *
2805 *************************************************/
2806
2807 /* This function is called when the PCRE_AUTO_CALLOUT option is set, to insert
2808 callout points before each pattern item.
2809
2810 Arguments:
2811 code current code pointer
2812 ptr current pattern pointer
2813 cd pointers to tables etc
2814
2815 Returns: new code pointer
2816 */
2817
2818 static pcre_uchar *
2819 auto_callout(pcre_uchar *code, const pcre_uchar *ptr, compile_data *cd)
2820 {
2821 *code++ = OP_CALLOUT;
2822 *code++ = 255;
2823 PUT(code, 0, (int)(ptr - cd->start_pattern)); /* Pattern offset */
2824 PUT(code, LINK_SIZE, 0); /* Default length */
2825 return code + 2 * LINK_SIZE;
2826 }
2827
2828
2829
2830 /*************************************************
2831 * Complete a callout item *
2832 *************************************************/
2833
2834 /* A callout item contains the length of the next item in the pattern, which
2835 we can't fill in till after we have reached the relevant point. This is used
2836 for both automatic and manual callouts.
2837
2838 Arguments:
2839 previous_callout points to previous callout item
2840 ptr current pattern pointer
2841 cd pointers to tables etc
2842
2843 Returns: nothing
2844 */
2845
2846 static void
2847 complete_callout(pcre_uchar *previous_callout, const pcre_uchar *ptr, compile_data *cd)
2848 {
2849 int length = (int)(ptr - cd->start_pattern - GET(previous_callout, 2));
2850 PUT(previous_callout, 2 + LINK_SIZE, length);
2851 }
2852
2853
2854
2855 #ifdef SUPPORT_UCP
2856 /*************************************************
2857 * Get othercase range *
2858 *************************************************/
2859
2860 /* This function is passed the start and end of a class range, in UTF-8 mode
2861 with UCP support. It searches up the characters, looking for internal ranges of
2862 characters in the "other" case. Each call returns the next one, updating the
2863 start address.
2864
2865 Arguments:
2866 cptr points to starting character value; updated
2867 d end value
2868 ocptr where to put start of othercase range
2869 odptr where to put end of othercase range
2870
2871 Yield: TRUE when range returned; FALSE when no more
2872 */
2873
2874 static BOOL
2875 get_othercase_range(unsigned int *cptr, unsigned int d, unsigned int *ocptr,
2876 unsigned int *odptr)
2877 {
2878 unsigned int c, othercase, next;
2879
2880 for (c = *cptr; c <= d; c++)
2881 { if ((othercase = UCD_OTHERCASE(c)) != c) break; }
2882
2883 if (c > d) return FALSE;
2884
2885 *ocptr = othercase;
2886 next = othercase + 1;
2887
2888 for (++c; c <= d; c++)
2889 {
2890 if (UCD_OTHERCASE(c) != next) break;
2891 next++;
2892 }
2893
2894 *odptr = next - 1;
2895 *cptr = c;
2896
2897 return TRUE;
2898 }
2899
2900
2901
2902 /*************************************************
2903 * Check a character and a property *
2904 *************************************************/
2905
2906 /* This function is called by check_auto_possessive() when a property item
2907 is adjacent to a fixed character.
2908
2909 Arguments:
2910 c the character
2911 ptype the property type
2912 pdata the data for the type
2913 negated TRUE if it's a negated property (\P or \p{^)
2914
2915 Returns: TRUE if auto-possessifying is OK
2916 */
2917
2918 static BOOL
2919 check_char_prop(int c, int ptype, int pdata, BOOL negated)
2920 {
2921 const ucd_record *prop = GET_UCD(c);
2922 switch(ptype)
2923 {
2924 case PT_LAMP:
2925 return (prop->chartype == ucp_Lu ||
2926 prop->chartype == ucp_Ll ||
2927 prop->chartype == ucp_Lt) == negated;
2928
2929 case PT_GC:
2930 return (pdata == PRIV(ucp_gentype)[prop->chartype]) == negated;
2931
2932 case PT_PC:
2933 return (pdata == prop->chartype) == negated;
2934
2935 case PT_SC:
2936 return (pdata == prop->script) == negated;
2937
2938 /* These are specials */
2939
2940 case PT_ALNUM:
2941 return (PRIV(ucp_gentype)[prop->chartype] == ucp_L ||
2942 PRIV(ucp_gentype)[prop->chartype] == ucp_N) == negated;
2943
2944 case PT_SPACE: /* Perl space */
2945 return (PRIV(ucp_gentype)[prop->chartype] == ucp_Z ||
2946 c == CHAR_HT || c == CHAR_NL || c == CHAR_FF || c == CHAR_CR)
2947 == negated;
2948
2949 case PT_PXSPACE: /* POSIX space */
2950 return (PRIV(ucp_gentype)[prop->chartype] == ucp_Z ||
2951 c == CHAR_HT || c == CHAR_NL || c == CHAR_VT ||
2952 c == CHAR_FF || c == CHAR_CR)
2953 == negated;
2954
2955 case PT_WORD:
2956 return (PRIV(ucp_gentype)[prop->chartype] == ucp_L ||
2957 PRIV(ucp_gentype)[prop->chartype] == ucp_N ||
2958 c == CHAR_UNDERSCORE) == negated;
2959 }
2960 return FALSE;
2961 }
2962 #endif /* SUPPORT_UCP */
2963
2964
2965
2966 /*************************************************
2967 * Check if auto-possessifying is possible *
2968 *************************************************/
2969
2970 /* This function is called for unlimited repeats of certain items, to see
2971 whether the next thing could possibly match the repeated item. If not, it makes
2972 sense to automatically possessify the repeated item.
2973
2974 Arguments:
2975 previous pointer to the repeated opcode
2976 utf TRUE in UTF-8 / UTF-16 mode
2977 ptr next character in pattern
2978 options options bits
2979 cd contains pointers to tables etc.
2980
2981 Returns: TRUE if possessifying is wanted
2982 */
2983
2984 static BOOL
2985 check_auto_possessive(const pcre_uchar *previous, BOOL utf,
2986 const pcre_uchar *ptr, int options, compile_data *cd)
2987 {
2988 pcre_int32 c, next;
2989 int op_code = *previous++;
2990
2991 /* Skip whitespace and comments in extended mode */
2992
2993 if ((options & PCRE_EXTENDED) != 0)
2994 {
2995 for (;;)
2996 {
2997 while (MAX_255(*ptr) && (cd->ctypes[*ptr] & ctype_space) != 0) ptr++;
2998 if (*ptr == CHAR_NUMBER_SIGN)
2999 {
3000 ptr++;
3001 while (*ptr != 0)
3002 {
3003 if (IS_NEWLINE(ptr)) { ptr += cd->nllen; break; }
3004 ptr++;
3005 #ifdef SUPPORT_UTF
3006 if (utf) FORWARDCHAR(ptr);
3007 #endif
3008 }
3009 }
3010 else break;
3011 }
3012 }
3013
3014 /* If the next item is one that we can handle, get its value. A non-negative
3015 value is a character, a negative value is an escape value. */
3016
3017 if (*ptr == CHAR_BACKSLASH)
3018 {
3019 int temperrorcode = 0;
3020 next = check_escape(&ptr, &temperrorcode, cd->bracount, options, FALSE);
3021 if (temperrorcode != 0) return FALSE;
3022 ptr++; /* Point after the escape sequence */
3023 }
3024 else if (!MAX_255(*ptr) || (cd->ctypes[*ptr] & ctype_meta) == 0)
3025 {
3026 #ifdef SUPPORT_UTF
3027 if (utf) { GETCHARINC(next, ptr); } else
3028 #endif
3029 next = *ptr++;
3030 }
3031 else return FALSE;
3032
3033 /* Skip whitespace and comments in extended mode */
3034
3035 if ((options & PCRE_EXTENDED) != 0)
3036 {
3037 for (;;)
3038 {
3039 while (MAX_255(*ptr) && (cd->ctypes[*ptr] & ctype_space) != 0) ptr++;
3040 if (*ptr == CHAR_NUMBER_SIGN)
3041 {
3042 ptr++;
3043 while (*ptr != 0)
3044 {
3045 if (IS_NEWLINE(ptr)) { ptr += cd->nllen; break; }
3046 ptr++;
3047 #ifdef SUPPORT_UTF
3048 if (utf) FORWARDCHAR(ptr);
3049 #endif
3050 }
3051 }
3052 else break;
3053 }
3054 }
3055
3056 /* If the next thing is itself optional, we have to give up. */
3057
3058 if (*ptr == CHAR_ASTERISK || *ptr == CHAR_QUESTION_MARK ||
3059 STRNCMP_UC_C8(ptr, STR_LEFT_CURLY_BRACKET STR_0 STR_COMMA, 3) == 0)
3060 return FALSE;
3061
3062 /* Now compare the next item with the previous opcode. First, handle cases when
3063 the next item is a character. */
3064
3065 if (next >= 0) switch(op_code)
3066 {
3067 case OP_CHAR:
3068 #ifdef SUPPORT_UTF
3069 GETCHARTEST(c, previous);
3070 #else
3071 c = *previous;
3072 #endif
3073 return c != next;
3074
3075 /* For CHARI (caseless character) we must check the other case. If we have
3076 Unicode property support, we can use it to test the other case of
3077 high-valued characters. */
3078
3079 case OP_CHARI:
3080 #ifdef SUPPORT_UTF
3081 GETCHARTEST(c, previous);
3082 #else
3083 c = *previous;
3084 #endif
3085 if (c == next) return FALSE;
3086 #ifdef SUPPORT_UTF
3087 if (utf)
3088 {
3089 unsigned int othercase;
3090 if (next < 128) othercase = cd->fcc[next]; else
3091 #ifdef SUPPORT_UCP
3092 othercase = UCD_OTHERCASE((unsigned int)next);
3093 #else
3094 othercase = NOTACHAR;
3095 #endif
3096 return (unsigned int)c != othercase;
3097 }
3098 else
3099 #endif /* SUPPORT_UTF */
3100 return (c != TABLE_GET((unsigned int)next, cd->fcc, next)); /* Non-UTF-8 mode */
3101
3102 case OP_NOT:
3103 #ifdef SUPPORT_UTF
3104 GETCHARTEST(c, previous);
3105 #else
3106 c = *previous;
3107 #endif
3108 return c == next;
3109
3110 case OP_NOTI:
3111 #ifdef SUPPORT_UTF
3112 GETCHARTEST(c, previous);
3113 #else
3114 c = *previous;
3115 #endif
3116 if (c == next) return TRUE;
3117 #ifdef SUPPORT_UTF
3118 if (utf)
3119 {
3120 unsigned int othercase;
3121 if (next < 128) othercase = cd->fcc[next]; else
3122 #ifdef SUPPORT_UCP
3123 othercase = UCD_OTHERCASE((unsigned int)next);
3124 #else
3125 othercase = NOTACHAR;
3126 #endif
3127 return (unsigned int)c == othercase;
3128 }
3129 else
3130 #endif /* SUPPORT_UTF */
3131 return (c == TABLE_GET((unsigned int)next, cd->fcc, next)); /* Non-UTF-8 mode */
3132
3133 /* Note that OP_DIGIT etc. are generated only when PCRE_UCP is *not* set.
3134 When it is set, \d etc. are converted into OP_(NOT_)PROP codes. */
3135
3136 case OP_DIGIT:
3137 return next > 255 || (cd->ctypes[next] & ctype_digit) == 0;
3138
3139 case OP_NOT_DIGIT:
3140 return next <= 255 && (cd->ctypes[next] & ctype_digit) != 0;
3141
3142 case OP_WHITESPACE:
3143 return next > 255 || (cd->ctypes[next] & ctype_space) == 0;
3144
3145 case OP_NOT_WHITESPACE:
3146 return next <= 255 && (cd->ctypes[next] & ctype_space) != 0;
3147
3148 case OP_WORDCHAR:
3149 return next > 255 || (cd->ctypes[next] & ctype_word) == 0;
3150
3151 case OP_NOT_WORDCHAR:
3152 return next <= 255 && (cd->ctypes[next] & ctype_word) != 0;
3153
3154 case OP_HSPACE:
3155 case OP_NOT_HSPACE:
3156 switch(next)
3157 {
3158 case 0x09:
3159 case 0x20:
3160 case 0xa0:
3161 case 0x1680:
3162 case 0x180e:
3163 case 0x2000:
3164 case 0x2001:
3165 case 0x2002:
3166 case 0x2003:
3167 case 0x2004:
3168 case 0x2005:
3169 case 0x2006:
3170 case 0x2007:
3171 case 0x2008:
3172 case 0x2009:
3173 case 0x200A:
3174 case 0x202f:
3175 case 0x205f:
3176 case 0x3000:
3177 return op_code == OP_NOT_HSPACE;
3178 default:
3179 return op_code != OP_NOT_HSPACE;
3180 }
3181
3182 case OP_ANYNL:
3183 case OP_VSPACE:
3184 case OP_NOT_VSPACE:
3185 switch(next)
3186 {
3187 case 0x0a:
3188 case 0x0b:
3189 case 0x0c:
3190 case 0x0d:
3191 case 0x85:
3192 case 0x2028:
3193 case 0x2029:
3194 return op_code == OP_NOT_VSPACE;
3195 default:
3196 return op_code != OP_NOT_VSPACE;
3197 }
3198
3199 #ifdef SUPPORT_UCP
3200 case OP_PROP:
3201 return check_char_prop(next, previous[0], previous[1], FALSE);
3202
3203 case OP_NOTPROP:
3204 return check_char_prop(next, previous[0], previous[1], TRUE);
3205 #endif
3206
3207 default:
3208 return FALSE;
3209 }
3210
3211
3212 /* Handle the case when the next item is \d, \s, etc. Note that when PCRE_UCP
3213 is set, \d turns into ESC_du rather than ESC_d, etc., so ESC_d etc. are
3214 generated only when PCRE_UCP is *not* set, that is, when only ASCII
3215 characteristics are recognized. Similarly, the opcodes OP_DIGIT etc. are
3216 replaced by OP_PROP codes when PCRE_UCP is set. */
3217
3218 switch(op_code)
3219 {
3220 case OP_CHAR:
3221 case OP_CHARI:
3222 #ifdef SUPPORT_UTF
3223 GETCHARTEST(c, previous);
3224 #else
3225 c = *previous;
3226 #endif
3227 switch(-next)
3228 {
3229 case ESC_d:
3230 return c > 255 || (cd->ctypes[c] & ctype_digit) == 0;
3231
3232 case ESC_D:
3233 return c <= 255 && (cd->ctypes[c] & ctype_digit) != 0;
3234
3235 case ESC_s:
3236 return c > 255 || (cd->ctypes[c] & ctype_space) == 0;
3237
3238 case ESC_S:
3239 return c <= 255 && (cd->ctypes[c] & ctype_space) != 0;
3240
3241 case ESC_w:
3242 return c > 255 || (cd->ctypes[c] & ctype_word) == 0;
3243
3244 case ESC_W:
3245 return c <= 255 && (cd->ctypes[c] & ctype_word) != 0;
3246
3247 case ESC_h:
3248 case ESC_H:
3249 switch(c)
3250 {
3251 case 0x09:
3252 case 0x20:
3253 case 0xa0:
3254 case 0x1680:
3255 case 0x180e:
3256 case 0x2000:
3257 case 0x2001:
3258 case 0x2002:
3259 case 0x2003:
3260 case 0x2004:
3261 case 0x2005:
3262 case 0x2006:
3263 case 0x2007:
3264 case 0x2008:
3265 case 0x2009:
3266 case 0x200A:
3267 case 0x202f:
3268 case 0x205f:
3269 case 0x3000:
3270 return -next != ESC_h;
3271 default:
3272 return -next == ESC_h;
3273 }
3274
3275 case ESC_v:
3276 case ESC_V:
3277 switch(c)
3278 {
3279 case 0x0a:
3280 case 0x0b:
3281 case 0x0c:
3282 case 0x0d:
3283 case 0x85:
3284 case 0x2028:
3285 case 0x2029:
3286 return -next != ESC_v;
3287 default:
3288 return -next == ESC_v;
3289 }
3290
3291 /* When PCRE_UCP is set, these values get generated for \d etc. Find
3292 their substitutions and process them. The result will always be either
3293 -ESC_p or -ESC_P. Then fall through to process those values. */
3294
3295 #ifdef SUPPORT_UCP
3296 case ESC_du:
3297 case ESC_DU:
3298 case ESC_wu:
3299 case ESC_WU:
3300 case ESC_su:
3301 case ESC_SU:
3302 {
3303 int temperrorcode = 0;
3304 ptr = substitutes[-next - ESC_DU];
3305 next = check_escape(&ptr, &temperrorcode, 0, options, FALSE);
3306 if (temperrorcode != 0) return FALSE;
3307 ptr++; /* For compatibility */
3308 }
3309 /* Fall through */
3310
3311 case ESC_p:
3312 case ESC_P:
3313 {
3314 int ptype, pdata, errorcodeptr;
3315 BOOL negated;
3316
3317 ptr--; /* Make ptr point at the p or P */
3318 ptype = get_ucp(&ptr, &negated, &pdata, &errorcodeptr);
3319 if (ptype < 0) return FALSE;
3320 ptr++; /* Point past the final curly ket */
3321
3322 /* If the property item is optional, we have to give up. (When generated
3323 from \d etc by PCRE_UCP, this test will have been applied much earlier,
3324 to the original \d etc. At this point, ptr will point to a zero byte. */
3325
3326 if (*ptr == CHAR_ASTERISK || *ptr == CHAR_QUESTION_MARK ||
3327 STRNCMP_UC_C8(ptr, STR_LEFT_CURLY_BRACKET STR_0 STR_COMMA, 3) == 0)
3328 return FALSE;
3329
3330 /* Do the property check. */
3331
3332 return check_char_prop(c, ptype, pdata, (next == -ESC_P) != negated);
3333 }
3334 #endif
3335
3336 default:
3337 return FALSE;
3338 }
3339
3340 /* In principle, support for Unicode properties should be integrated here as
3341 well. It means re-organizing the above code so as to get hold of the property
3342 values before switching on the op-code. However, I wonder how many patterns
3343 combine ASCII \d etc with Unicode properties? (Note that if PCRE_UCP is set,
3344 these op-codes are never generated.) */
3345
3346 case OP_DIGIT:
3347 return next == -ESC_D || next == -ESC_s || next == -ESC_W ||
3348 next == -ESC_h || next == -ESC_v || next == -ESC_R;
3349
3350 case OP_NOT_DIGIT:
3351 return next == -ESC_d;
3352
3353 case OP_WHITESPACE:
3354 return next == -ESC_S || next == -ESC_d || next == -ESC_w;
3355
3356 case OP_NOT_WHITESPACE:
3357 return next == -ESC_s || next == -ESC_h || next == -ESC_v || next == -ESC_R;
3358
3359 case OP_HSPACE:
3360 return next == -ESC_S || next == -ESC_H || next == -ESC_d ||
3361 next == -ESC_w || next == -ESC_v || next == -ESC_R;
3362
3363 case OP_NOT_HSPACE:
3364 return next == -ESC_h;
3365
3366 /* Can't have \S in here because VT matches \S (Perl anomaly) */
3367 case OP_ANYNL:
3368 case OP_VSPACE:
3369 return next == -ESC_V || next == -ESC_d || next == -ESC_w;
3370
3371 case OP_NOT_VSPACE:
3372 return next == -ESC_v || next == -ESC_R;
3373
3374 case OP_WORDCHAR:
3375 return next == -ESC_W || next == -ESC_s || next == -ESC_h ||
3376 next == -ESC_v || next == -ESC_R;
3377
3378 case OP_NOT_WORDCHAR:
3379 return next == -ESC_w || next == -ESC_d;
3380
3381 default:
3382 return FALSE;
3383 }
3384
3385 /* Control does not reach here */
3386 }
3387
3388
3389
3390 /*************************************************
3391 * Compile one branch *
3392 *************************************************/
3393
3394 /* Scan the pattern, compiling it into the a vector. If the options are
3395 changed during the branch, the pointer is used to change the external options
3396 bits. This function is used during the pre-compile phase when we are trying
3397 to find out the amount of memory needed, as well as during the real compile
3398 phase. The value of lengthptr distinguishes the two phases.
3399
3400 Arguments:
3401 optionsptr pointer to the option bits
3402 codeptr points to the pointer to the current code point
3403 ptrptr points to the current pattern pointer
3404 errorcodeptr points to error code variable
3405 firstcharptr set to initial literal character, or < 0 (REQ_UNSET, REQ_NONE)
3406 reqcharptr set to the last literal character required, else < 0
3407 bcptr points to current branch chain
3408 cond_depth conditional nesting depth
3409 cd contains pointers to tables etc.
3410 lengthptr NULL during the real compile phase
3411 points to length accumulator during pre-compile phase
3412
3413 Returns: TRUE on success
3414 FALSE, with *errorcodeptr set non-zero on error
3415 */
3416
3417 static BOOL
3418 compile_branch(int *optionsptr, pcre_uchar **codeptr,
3419 const pcre_uchar **ptrptr, int *errorcodeptr, pcre_int32 *firstcharptr,
3420 pcre_int32 *reqcharptr, branch_chain *bcptr, int cond_depth,
3421 compile_data *cd, int *lengthptr)
3422 {
3423 int repeat_type, op_type;
3424 int repeat_min = 0, repeat_max = 0; /* To please picky compilers */
3425 int bravalue = 0;
3426 int greedy_default, greedy_non_default;
3427 pcre_int32 firstchar, reqchar;
3428 pcre_int32 zeroreqchar, zerofirstchar;
3429 pcre_int32 req_caseopt, reqvary, tempreqvary;
3430 int options = *optionsptr; /* May change dynamically */
3431 int after_manual_callout = 0;
3432 int length_prevgroup = 0;
3433 register int c;
3434 register pcre_uchar *code = *codeptr;
3435 pcre_uchar *last_code = code;
3436 pcre_uchar *orig_code = code;
3437 pcre_uchar *tempcode;
3438 BOOL inescq = FALSE;
3439 BOOL groupsetfirstchar = FALSE;
3440 const pcre_uchar *ptr = *ptrptr;
3441 const pcre_uchar *tempptr;
3442 const pcre_uchar *nestptr = NULL;
3443 pcre_uchar *previous = NULL;
3444 pcre_uchar *previous_callout = NULL;
3445 pcre_uchar *save_hwm = NULL;
3446 pcre_uint8 classbits[32];
3447
3448 /* We can fish out the UTF-8 setting once and for all into a BOOL, but we
3449 must not do this for other options (e.g. PCRE_EXTENDED) because they may change
3450 dynamically as we process the pattern. */
3451
3452 #ifdef SUPPORT_UTF
3453 /* PCRE_UTF16 has the same value as PCRE_UTF8. */
3454 BOOL utf = (options & PCRE_UTF8) != 0;
3455 pcre_uchar utf_chars[6];
3456 #else
3457 BOOL utf = FALSE;
3458 #endif
3459
3460 /* Helper variables for OP_XCLASS opcode (for characters > 255). */
3461
3462 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
3463 BOOL xclass;
3464 pcre_uchar *class_uchardata;
3465 pcre_uchar *class_uchardata_base;
3466 #endif
3467
3468 #ifdef PCRE_DEBUG
3469 if (lengthptr != NULL) DPRINTF((">> start branch\n"));
3470 #endif
3471
3472 /* Set up the default and non-default settings for greediness */
3473
3474 greedy_default = ((options & PCRE_UNGREEDY) != 0);
3475 greedy_non_default = greedy_default ^ 1;
3476
3477 /* Initialize no first byte, no required byte. REQ_UNSET means "no char
3478 matching encountered yet". It gets changed to REQ_NONE if we hit something that
3479 matches a non-fixed char first char; reqchar just remains unset if we never
3480 find one.
3481
3482 When we hit a repeat whose minimum is zero, we may have to adjust these values
3483 to take the zero repeat into account. This is implemented by setting them to
3484 zerofirstbyte and zeroreqchar when such a repeat is encountered. The individual
3485 item types that can be repeated set these backoff variables appropriately. */
3486
3487 firstchar = reqchar = zerofirstchar = zeroreqchar = REQ_UNSET;
3488
3489 /* The variable req_caseopt contains either the REQ_CASELESS value
3490 or zero, according to the current setting of the caseless flag. The
3491 REQ_CASELESS leaves the lower 28 bit empty. It is added into the
3492 firstchar or reqchar variables to record the case status of the
3493 value. This is used only for ASCII characters. */
3494
3495 req_caseopt = ((options & PCRE_CASELESS) != 0)? REQ_CASELESS:0;
3496
3497 /* Switch on next character until the end of the branch */
3498
3499 for (;; ptr++)
3500 {
3501 BOOL negate_class;
3502 BOOL should_flip_negation;
3503 BOOL possessive_quantifier;
3504 BOOL is_quantifier;
3505 BOOL is_recurse;
3506 BOOL reset_bracount;
3507 int class_has_8bitchar;
3508 int class_single_char;
3509 int newoptions;
3510 int recno;
3511 int refsign;
3512 int skipbytes;
3513 int subreqchar;
3514 int subfirstchar;
3515 int terminator;
3516 int mclength;
3517 int tempbracount;
3518 pcre_uchar mcbuffer[8];
3519
3520 /* Get next character in the pattern */
3521
3522 c = *ptr;
3523
3524 /* If we are at the end of a nested substitution, revert to the outer level
3525 string. Nesting only happens one level deep. */
3526
3527 if (c == 0 && nestptr != NULL)
3528 {
3529 ptr = nestptr;
3530 nestptr = NULL;
3531 c = *ptr;
3532 }
3533
3534 /* If we are in the pre-compile phase, accumulate the length used for the
3535 previous cycle of this loop. */
3536
3537 if (lengthptr != NULL)
3538 {
3539 #ifdef PCRE_DEBUG
3540 if (code > cd->hwm) cd->hwm = code; /* High water info */
3541 #endif
3542 if (code > cd->start_workspace + cd->workspace_size -
3543 WORK_SIZE_SAFETY_MARGIN) /* Check for overrun */
3544 {
3545 *errorcodeptr = ERR52;
3546 goto FAILED;
3547 }
3548
3549 /* There is at least one situation where code goes backwards: this is the
3550 case of a zero quantifier after a class (e.g. [ab]{0}). At compile time,
3551 the class is simply eliminated. However, it is created first, so we have to
3552 allow memory for it. Therefore, don't ever reduce the length at this point.
3553 */
3554
3555 if (code < last_code) code = last_code;
3556
3557 /* Paranoid check for integer overflow */
3558
3559 if (OFLOW_MAX - *lengthptr < code - last_code)
3560 {
3561 *errorcodeptr = ERR20;
3562 goto FAILED;
3563 }
3564
3565 *lengthptr += (int)(code - last_code);
3566 DPRINTF(("length=%d added %d c=%c (0x%x)\n", *lengthptr,
3567 (int)(code - last_code), c, c));
3568
3569 /* If "previous" is set and it is not at the start of the work space, move
3570 it back to there, in order to avoid filling up the work space. Otherwise,
3571 if "previous" is NULL, reset the current code pointer to the start. */
3572
3573 if (previous != NULL)
3574 {
3575 if (previous > orig_code)
3576 {
3577 memmove(orig_code, previous, IN_UCHARS(code - previous));
3578 code -= previous - orig_code;
3579 previous = orig_code;
3580 }
3581 }
3582 else code = orig_code;
3583
3584 /* Remember where this code item starts so we can pick up the length
3585 next time round. */
3586
3587 last_code = code;
3588 }
3589
3590 /* In the real compile phase, just check the workspace used by the forward
3591 reference list. */
3592
3593 else if (cd->hwm > cd->start_workspace + cd->workspace_size -
3594 WORK_SIZE_SAFETY_MARGIN)
3595 {
3596 *errorcodeptr = ERR52;
3597 goto FAILED;
3598 }
3599
3600 /* If in \Q...\E, check for the end; if not, we have a literal */
3601
3602 if (inescq && c != 0)
3603 {
3604 if (c == CHAR_BACKSLASH && ptr[1] == CHAR_E)
3605 {
3606 inescq = FALSE;
3607 ptr++;
3608 continue;
3609 }
3610 else
3611 {
3612 if (previous_callout != NULL)
3613 {
3614 if (lengthptr == NULL) /* Don't attempt in pre-compile phase */
3615 complete_callout(previous_callout, ptr, cd);
3616 previous_callout = NULL;
3617 }
3618 if ((options & PCRE_AUTO_CALLOUT) != 0)
3619 {
3620 previous_callout = code;
3621 code = auto_callout(code, ptr, cd);
3622 }
3623 goto NORMAL_CHAR;
3624 }
3625 }
3626
3627 /* Fill in length of a previous callout, except when the next thing is
3628 a quantifier. */
3629
3630 is_quantifier =
3631 c == CHAR_ASTERISK || c == CHAR_PLUS || c == CHAR_QUESTION_MARK ||
3632 (c == CHAR_LEFT_CURLY_BRACKET && is_counted_repeat(ptr+1));
3633
3634 if (!is_quantifier && previous_callout != NULL &&
3635 after_manual_callout-- <= 0)
3636 {
3637 if (lengthptr == NULL) /* Don't attempt in pre-compile phase */
3638 complete_callout(previous_callout, ptr, cd);
3639 previous_callout = NULL;
3640 }
3641
3642 /* In extended mode, skip white space and comments. */
3643
3644 if ((options & PCRE_EXTENDED) != 0)
3645 {
3646 if (MAX_255(*ptr) && (cd->ctypes[c] & ctype_space) != 0) continue;
3647 if (c == CHAR_NUMBER_SIGN)
3648 {
3649 ptr++;
3650 while (*ptr != 0)
3651 {
3652 if (IS_NEWLINE(ptr)) { ptr += cd->nllen - 1; break; }
3653 ptr++;
3654 #ifdef SUPPORT_UTF
3655 if (utf) FORWARDCHAR(ptr);
3656 #endif
3657 }
3658 if (*ptr != 0) continue;
3659
3660 /* Else fall through to handle end of string */
3661 c = 0;
3662 }
3663 }
3664
3665 /* No auto callout for quantifiers. */
3666
3667 if ((options & PCRE_AUTO_CALLOUT) != 0 && !is_quantifier)
3668 {
3669 previous_callout = code;
3670 code = auto_callout(code, ptr, cd);
3671 }
3672
3673 switch(c)
3674 {
3675 /* ===================================================================*/
3676 case 0: /* The branch terminates at string end */
3677 case CHAR_VERTICAL_LINE: /* or | or ) */
3678 case CHAR_RIGHT_PARENTHESIS:
3679 *firstcharptr = firstchar;
3680 *reqcharptr = reqchar;
3681 *codeptr = code;
3682 *ptrptr = ptr;
3683 if (lengthptr != NULL)
3684 {
3685 if (OFLOW_MAX - *lengthptr < code - last_code)
3686 {
3687 *errorcodeptr = ERR20;
3688 goto FAILED;
3689 }
3690 *lengthptr += (int)(code - last_code); /* To include callout length */
3691 DPRINTF((">> end branch\n"));
3692 }
3693 return TRUE;
3694
3695
3696 /* ===================================================================*/
3697 /* Handle single-character metacharacters. In multiline mode, ^ disables
3698 the setting of any following char as a first character. */
3699
3700 case CHAR_CIRCUMFLEX_ACCENT:
3701 previous = NULL;
3702 if ((options & PCRE_MULTILINE) != 0)
3703 {
3704 if (firstchar == REQ_UNSET) firstchar = REQ_NONE;
3705 *code++ = OP_CIRCM;
3706 }
3707 else *code++ = OP_CIRC;
3708 break;
3709
3710 case CHAR_DOLLAR_SIGN:
3711 previous = NULL;
3712 *code++ = ((options & PCRE_MULTILINE) != 0)? OP_DOLLM : OP_DOLL;
3713 break;
3714
3715 /* There can never be a first char if '.' is first, whatever happens about
3716 repeats. The value of reqchar doesn't change either. */
3717
3718 case CHAR_DOT:
3719 if (firstchar == REQ_UNSET) firstchar = REQ_NONE;
3720 zerofirstchar = firstchar;
3721 zeroreqchar = reqchar;
3722 previous = code;
3723 *code++ = ((options & PCRE_DOTALL) != 0)? OP_ALLANY: OP_ANY;
3724 break;
3725
3726
3727 /* ===================================================================*/
3728 /* Character classes. If the included characters are all < 256, we build a
3729 32-byte bitmap of the permitted characters, except in the special case
3730 where there is only one such character. For negated classes, we build the
3731 map as usual, then invert it at the end. However, we use a different opcode
3732 so that data characters > 255 can be handled correctly.
3733
3734 If the class contains characters outside the 0-255 range, a different
3735 opcode is compiled. It may optionally have a bit map for characters < 256,
3736 but those above are are explicitly listed afterwards. A flag byte tells
3737 whether the bitmap is present, and whether this is a negated class or not.
3738
3739 In JavaScript compatibility mode, an isolated ']' causes an error. In
3740 default (Perl) mode, it is treated as a data character. */
3741
3742 case CHAR_RIGHT_SQUARE_BRACKET:
3743 if ((cd->external_options & PCRE_JAVASCRIPT_COMPAT) != 0)
3744 {
3745 *errorcodeptr = ERR64;
3746 goto FAILED;
3747 }
3748 goto NORMAL_CHAR;
3749
3750 case CHAR_LEFT_SQUARE_BRACKET:
3751 previous = code;
3752
3753 /* PCRE supports POSIX class stuff inside a class. Perl gives an error if
3754 they are encountered at the top level, so we'll do that too. */
3755
3756 if ((ptr[1] == CHAR_COLON || ptr[1] == CHAR_DOT ||
3757 ptr[1] == CHAR_EQUALS_SIGN) &&
3758 check_posix_syntax(ptr, &tempptr))
3759 {
3760 *errorcodeptr = (ptr[1] == CHAR_COLON)? ERR13 : ERR31;
3761 goto FAILED;
3762 }
3763
3764 /* If the first character is '^', set the negation flag and skip it. Also,
3765 if the first few characters (either before or after ^) are \Q\E or \E we
3766 skip them too. This makes for compatibility with Perl. */
3767
3768 negate_class = FALSE;
3769 for (;;)
3770 {
3771 c = *(++ptr);
3772 if (c == CHAR_BACKSLASH)
3773 {
3774 if (ptr[1] == CHAR_E)
3775 ptr++;
3776 else if (STRNCMP_UC_C8(ptr + 1, STR_Q STR_BACKSLASH STR_E, 3) == 0)
3777 ptr += 3;
3778 else
3779 break;
3780 }
3781 else if (!negate_class && c == CHAR_CIRCUMFLEX_ACCENT)
3782 negate_class = TRUE;
3783 else break;
3784 }
3785
3786 /* Empty classes are allowed in JavaScript compatibility mode. Otherwise,
3787 an initial ']' is taken as a data character -- the code below handles
3788 that. In JS mode, [] must always fail, so generate OP_FAIL, whereas
3789 [^] must match any character, so generate OP_ALLANY. */
3790
3791 if (c == CHAR_RIGHT_SQUARE_BRACKET &&
3792 (cd->external_options & PCRE_JAVASCRIPT_COMPAT) != 0)
3793 {
3794 *code++ = negate_class? OP_ALLANY : OP_FAIL;
3795 if (firstchar == REQ_UNSET) firstchar = REQ_NONE;
3796 zerofirstchar = firstchar;
3797 break;
3798 }
3799
3800 /* If a class contains a negative special such as \S, we need to flip the
3801 negation flag at the end, so that support for characters > 255 works
3802 correctly (they are all included in the class). */
3803
3804 should_flip_negation = FALSE;
3805
3806 /* For optimization purposes, we track some properties of the class.
3807 class_has_8bitchar will be non-zero, if the class contains at least one
3808 < 256 character. class_single_char will be 1 if the class contains only
3809 a single character. */
3810
3811 class_has_8bitchar = 0;
3812 class_single_char = 0;
3813
3814 /* Initialize the 32-char bit map to all zeros. We build the map in a
3815 temporary bit of memory, in case the class contains only 1 character (less
3816 than 256), because in that case the compiled code doesn't use the bit map.
3817 */
3818
3819 memset(classbits, 0, 32 * sizeof(pcre_uint8));
3820
3821 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
3822 xclass = FALSE; /* No chars >= 256 */
3823 class_uchardata = code + LINK_SIZE + 2; /* For UTF-8 items */
3824 class_uchardata_base = class_uchardata; /* For resetting in pass 1 */
3825 #endif
3826
3827 /* Process characters until ] is reached. By writing this as a "do" it
3828 means that an initial ] is taken as a data character. At the start of the
3829 loop, c contains the first byte of the character. */
3830
3831 if (c != 0) do
3832 {
3833 const pcre_uchar *oldptr;
3834
3835 #ifdef SUPPORT_UTF
3836 if (utf && HAS_EXTRALEN(c))
3837 { /* Braces are required because the */
3838 GETCHARLEN(c, ptr, ptr); /* macro generates multiple statements */
3839 }
3840 #endif
3841
3842 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
3843 /* In the pre-compile phase, accumulate the length of any extra
3844 data and reset the pointer. This is so that very large classes that
3845 contain a zillion > 255 characters no longer overwrite the work space
3846 (which is on the stack). */
3847
3848 if (lengthptr != NULL)
3849 {
3850 *lengthptr += class_uchardata - class_uchardata_base;
3851 class_uchardata = class_uchardata_base;
3852 }
3853 #endif
3854
3855 /* Inside \Q...\E everything is literal except \E */
3856
3857 if (inescq)
3858 {
3859 if (c == CHAR_BACKSLASH && ptr[1] == CHAR_E) /* If we are at \E */
3860 {
3861 inescq = FALSE; /* Reset literal state */
3862 ptr++; /* Skip the 'E' */
3863 continue; /* Carry on with next */
3864 }
3865 goto CHECK_RANGE; /* Could be range if \E follows */
3866 }
3867
3868 /* Handle POSIX class names. Perl allows a negation extension of the
3869 form [:^name:]. A square bracket that doesn't match the syntax is
3870 treated as a literal. We also recognize the POSIX constructions
3871 [.ch.] and [=ch=] ("collating elements") and fault them, as Perl
3872 5.6 and 5.8 do. */
3873
3874 if (c == CHAR_LEFT_SQUARE_BRACKET &&
3875 (ptr[1] == CHAR_COLON || ptr[1] == CHAR_DOT ||
3876 ptr[1] == CHAR_EQUALS_SIGN) && check_posix_syntax(ptr, &tempptr))
3877 {
3878 BOOL local_negate = FALSE;
3879 int posix_class, taboffset, tabopt;
3880 register const pcre_uint8 *cbits = cd->cbits;
3881 pcre_uint8 pbits[32];
3882
3883 if (ptr[1] != CHAR_COLON)
3884 {
3885 *errorcodeptr = ERR31;
3886 goto FAILED;
3887 }
3888
3889 ptr += 2;
3890 if (*ptr == CHAR_CIRCUMFLEX_ACCENT)
3891 {
3892 local_negate = TRUE;
3893 should_flip_negation = TRUE; /* Note negative special */
3894 ptr++;
3895 }
3896
3897 posix_class = check_posix_name(ptr, (int)(tempptr - ptr));
3898 if (posix_class < 0)
3899 {
3900 *errorcodeptr = ERR30;
3901 goto FAILED;
3902 }
3903
3904 /* If matching is caseless, upper and lower are converted to
3905 alpha. This relies on the fact that the class table starts with
3906 alpha, lower, upper as the first 3 entries. */
3907
3908 if ((options & PCRE_CASELESS) != 0 && posix_class <= 2)
3909 posix_class = 0;
3910
3911 /* When PCRE_UCP is set, some of the POSIX classes are converted to
3912 different escape sequences that use Unicode properties. */
3913
3914 #ifdef SUPPORT_UCP
3915 if ((options & PCRE_UCP) != 0)
3916 {
3917 int pc = posix_class + ((local_negate)? POSIX_SUBSIZE/2 : 0);
3918 if (posix_substitutes[pc] != NULL)
3919 {
3920 nestptr = tempptr + 1;
3921 ptr = posix_substitutes[pc] - 1;
3922 continue;
3923 }
3924 }
3925 #endif
3926 /* In the non-UCP case, we build the bit map for the POSIX class in a
3927 chunk of local store because we may be adding and subtracting from it,
3928 and we don't want to subtract bits that may be in the main map already.
3929 At the end we or the result into the bit map that is being built. */
3930
3931 posix_class *= 3;
3932
3933 /* Copy in the first table (always present) */
3934
3935 memcpy(pbits, cbits + posix_class_maps[posix_class],
3936 32 * sizeof(pcre_uint8));
3937
3938 /* If there is a second table, add or remove it as required. */
3939
3940 taboffset = posix_class_maps[posix_class + 1];
3941 tabopt = posix_class_maps[posix_class + 2];
3942
3943 if (taboffset >= 0)
3944 {
3945 if (tabopt >= 0)
3946 for (c = 0; c < 32; c++) pbits[c] |= cbits[c + taboffset];
3947 else
3948 for (c = 0; c < 32; c++) pbits[c] &= ~cbits[c + taboffset];
3949 }
3950
3951 /* Not see if we need to remove any special characters. An option
3952 value of 1 removes vertical space and 2 removes underscore. */
3953
3954 if (tabopt < 0) tabopt = -tabopt;
3955 if (tabopt == 1) pbits[1] &= ~0x3c;
3956 else if (tabopt == 2) pbits[11] &= 0x7f;
3957
3958 /* Add the POSIX table or its complement into the main table that is
3959 being built and we are done. */
3960
3961 if (local_negate)
3962 for (c = 0; c < 32; c++) classbits[c] |= ~pbits[c];
3963 else
3964 for (c = 0; c < 32; c++) classbits[c] |= pbits[c];
3965
3966 ptr = tempptr + 1;
3967 /* Every class contains at least one < 256 characters. */
3968 class_has_8bitchar = 1;
3969 /* Every class contains at least two characters. */
3970 class_single_char = 2;
3971 continue; /* End of POSIX syntax handling */
3972 }
3973
3974 /* Backslash may introduce a single character, or it may introduce one
3975 of the specials, which just set a flag. The sequence \b is a special
3976 case. Inside a class (and only there) it is treated as backspace. We
3977 assume that other escapes have more than one character in them, so
3978 speculatively set both class_has_8bitchar and class_single_char bigger
3979 than one. Unrecognized escapes fall through and are either treated
3980 as literal characters (by default), or are faulted if
3981 PCRE_EXTRA is set. */
3982
3983 if (c == CHAR_BACKSLASH)
3984 {
3985 c = check_escape(&ptr, errorcodeptr, cd->bracount, options, TRUE);
3986 if (*errorcodeptr != 0) goto FAILED;
3987
3988 if (-c == ESC_b) c = CHAR_BS; /* \b is backspace in a class */
3989 else if (-c == ESC_N) /* \N is not supported in a class */
3990 {
3991 *errorcodeptr = ERR71;
3992 goto FAILED;
3993 }
3994 else if (-c == ESC_Q) /* Handle start of quoted string */
3995 {
3996 if (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
3997 {
3998 ptr += 2; /* avoid empty string */
3999 }
4000 else inescq = TRUE;
4001 continue;
4002 }
4003 else if (-c == ESC_E) continue; /* Ignore orphan \E */
4004
4005 if (c < 0)
4006 {
4007 register const pcre_uint8 *cbits = cd->cbits;
4008 /* Every class contains at least two < 256 characters. */
4009 class_has_8bitchar++;
4010 /* Every class contains at least two characters. */
4011 class_single_char += 2;
4012
4013 switch (-c)
4014 {
4015 #ifdef SUPPORT_UCP
4016 case ESC_du: /* These are the values given for \d etc */
4017 case ESC_DU: /* when PCRE_UCP is set. We replace the */
4018 case ESC_wu: /* escape sequence with an appropriate \p */
4019 case ESC_WU: /* or \P to test Unicode properties instead */
4020 case ESC_su: /* of the default ASCII testing. */
4021 case ESC_SU:
4022 nestptr = ptr;
4023 ptr = substitutes[-c - ESC_DU] - 1; /* Just before substitute */
4024 class_has_8bitchar--; /* Undo! */
4025 continue;
4026 #endif
4027 case ESC_d:
4028 for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_digit];
4029 continue;
4030
4031 case ESC_D:
4032 should_flip_negation = TRUE;
4033 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_digit];
4034 continue;
4035
4036 case ESC_w:
4037 for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_word];
4038 continue;
4039
4040 case ESC_W:
4041 should_flip_negation = TRUE;
4042 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_word];
4043 continue;
4044
4045 /* Perl 5.004 onwards omits VT from \s, but we must preserve it
4046 if it was previously set by something earlier in the character
4047 class. */
4048
4049 case ESC_s:
4050 classbits[0] |= cbits[cbit_space];
4051 classbits[1] |= cbits[cbit_space+1] & ~0x08;
4052 for (c = 2; c < 32; c++) classbits[c] |= cbits[c+cbit_space];
4053 continue;
4054
4055 case ESC_S:
4056 should_flip_negation = TRUE;
4057 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_space];
4058 classbits[1] |= 0x08; /* Perl 5.004 onwards omits VT from \s */
4059 continue;
4060
4061 case ESC_h:
4062 SETBIT(classbits, 0x09); /* VT */
4063 SETBIT(classbits, 0x20); /* SPACE */
4064 SETBIT(classbits, 0xa0); /* NSBP */
4065 #ifndef COMPILE_PCRE8
4066 xclass = TRUE;
4067 *class_uchardata++ = XCL_SINGLE;
4068 *class_uchardata++ = 0x1680;
4069 *class_uchardata++ = XCL_SINGLE;
4070 *class_uchardata++ = 0x180e;
4071 *class_uchardata++ = XCL_RANGE;
4072 *class_uchardata++ = 0x2000;
4073 *class_uchardata++ = 0x200a;
4074 *class_uchardata++ = XCL_SINGLE;
4075 *class_uchardata++ = 0x202f;
4076 *class_uchardata++ = XCL_SINGLE;
4077 *class_uchardata++ = 0x205f;
4078 *class_uchardata++ = XCL_SINGLE;
4079 *class_uchardata++ = 0x3000;
4080 #elif defined SUPPORT_UTF
4081 if (utf)
4082 {
4083 xclass = TRUE;
4084 *class_uchardata++ = XCL_SINGLE;
4085 class_uchardata += PRIV(ord2utf)(0x1680, class_uchardata);
4086 *class_uchardata++ = XCL_SINGLE;
4087 class_uchardata += PRIV(ord2utf)(0x180e, class_uchardata);
4088 *class_uchardata++ = XCL_RANGE;
4089 class_uchardata += PRIV(ord2utf)(0x2000, class_uchardata);
4090 class_uchardata += PRIV(ord2utf)(0x200a, class_uchardata);
4091 *class_uchardata++ = XCL_SINGLE;
4092 class_uchardata += PRIV(ord2utf)(0x202f, class_uchardata);
4093 *class_uchardata++ = XCL_SINGLE;
4094 class_uchardata += PRIV(ord2utf)(0x205f, class_uchardata);
4095 *class_uchardata++ = XCL_SINGLE;
4096 class_uchardata += PRIV(ord2utf)(0x3000, class_uchardata);
4097 }
4098 #endif
4099 continue;
4100
4101 case ESC_H:
4102 for (c = 0; c < 32; c++)
4103 {
4104 int x = 0xff;
4105 switch (c)
4106 {
4107 case 0x09/8: x ^= 1 << (0x09%8); break;
4108 case 0x20/8: x ^= 1 << (0x20%8); break;
4109 case 0xa0/8: x ^= 1 << (0xa0%8); break;
4110 default: break;
4111 }
4112 classbits[c] |= x;
4113 }
4114 #ifndef COMPILE_PCRE8
4115 xclass = TRUE;
4116 *class_uchardata++ = XCL_RANGE;
4117 *class_uchardata++ = 0x0100;
4118 *class_uchardata++ = 0x167f;
4119 *class_uchardata++ = XCL_RANGE;
4120 *class_uchardata++ = 0x1681;
4121 *class_uchardata++ = 0x180d;
4122 *class_uchardata++ = XCL_RANGE;
4123 *class_uchardata++ = 0x180f;
4124 *class_uchardata++ = 0x1fff;
4125 *class_uchardata++ = XCL_RANGE;
4126 *class_uchardata++ = 0x200b;
4127 *class_uchardata++ = 0x202e;
4128 *class_uchardata++ = XCL_RANGE;
4129 *class_uchardata++ = 0x2030;
4130 *class_uchardata++ = 0x205e;
4131 *class_uchardata++ = XCL_RANGE;
4132 *class_uchardata++ = 0x2060;
4133 *class_uchardata++ = 0x2fff;
4134 *class_uchardata++ = XCL_RANGE;
4135 *class_uchardata++ = 0x3001;
4136 #ifdef SUPPORT_UTF
4137 if (utf)
4138 class_uchardata += PRIV(ord2utf)(0x10ffff, class_uchardata);
4139 else
4140 #endif
4141 *class_uchardata++ = 0xffff;
4142 #elif defined SUPPORT_UTF
4143 if (utf)
4144 {
4145 xclass = TRUE;
4146 *class_uchardata++ = XCL_RANGE;
4147 class_uchardata += PRIV(ord2utf)(0x0100, class_uchardata);
4148 class_uchardata += PRIV(ord2utf)(0x167f, class_uchardata);
4149 *class_uchardata++ = XCL_RANGE;
4150 class_uchardata += PRIV(ord2utf)(0x1681, class_uchardata);
4151 class_uchardata += PRIV(ord2utf)(0x180d, class_uchardata);
4152 *class_uchardata++ = XCL_RANGE;
4153 class_uchardata += PRIV(ord2utf)(0x180f, class_uchardata);
4154 class_uchardata += PRIV(ord2utf)(0x1fff, class_uchardata);
4155 *class_uchardata++ = XCL_RANGE;
4156 class_uchardata += PRIV(ord2utf)(0x200b, class_uchardata);
4157 class_uchardata += PRIV(ord2utf)(0x202e, class_uchardata);
4158 *class_uchardata++ = XCL_RANGE;
4159 class_uchardata += PRIV(ord2utf)(0x2030, class_uchardata);
4160 class_uchardata += PRIV(ord2utf)(0x205e, class_uchardata);
4161 *class_uchardata++ = XCL_RANGE;
4162 class_uchardata += PRIV(ord2utf)(0x2060, class_uchardata);
4163 class_uchardata += PRIV(ord2utf)(0x2fff, class_uchardata);
4164 *class_uchardata++ = XCL_RANGE;
4165 class_uchardata += PRIV(ord2utf)(0x3001, class_uchardata);
4166 class_uchardata += PRIV(ord2utf)(0x10ffff, class_uchardata);
4167 }
4168 #endif
4169 continue;
4170
4171 case ESC_v:
4172 SETBIT(classbits, 0x0a); /* LF */
4173 SETBIT(classbits, 0x0b); /* VT */
4174 SETBIT(classbits, 0x0c); /* FF */
4175 SETBIT(classbits, 0x0d); /* CR */
4176 SETBIT(classbits, 0x85); /* NEL */
4177 #ifndef COMPILE_PCRE8
4178 xclass = TRUE;
4179 *class_uchardata++ = XCL_RANGE;
4180 *class_uchardata++ = 0x2028;
4181 *class_uchardata++ = 0x2029;
4182 #elif defined SUPPORT_UTF
4183 if (utf)
4184 {
4185 xclass = TRUE;
4186 *class_uchardata++ = XCL_RANGE;
4187 class_uchardata += PRIV(ord2utf)(0x2028, class_uchardata);
4188 class_uchardata += PRIV(ord2utf)(0x2029, class_uchardata);
4189 }
4190 #endif
4191 continue;
4192
4193 case ESC_V:
4194 for (c = 0; c < 32; c++)
4195 {
4196 int x = 0xff;
4197 switch (c)
4198 {
4199 case 0x0a/8: x ^= 1 << (0x0a%8);
4200 x ^= 1 << (0x0b%8);
4201 x ^= 1 << (0x0c%8);
4202 x ^= 1 << (0x0d%8);
4203 break;
4204 case 0x85/8: x ^= 1 << (0x85%8); break;
4205 default: break;
4206 }
4207 classbits[c] |= x;
4208 }
4209
4210 #ifndef COMPILE_PCRE8
4211 xclass = TRUE;
4212 *class_uchardata++ = XCL_RANGE;
4213 *class_uchardata++ = 0x0100;
4214 *class_uchardata++ = 0x2027;
4215 *class_uchardata++ = XCL_RANGE;
4216 *class_uchardata++ = 0x202a;
4217 #ifdef SUPPORT_UTF
4218 if (utf)
4219 class_uchardata += PRIV(ord2utf)(0x10ffff, class_uchardata);
4220 else
4221 #endif
4222 *class_uchardata++ = 0xffff;
4223 #elif defined SUPPORT_UTF
4224 if (utf)
4225 {
4226 xclass = TRUE;
4227 *class_uchardata++ = XCL_RANGE;
4228 class_uchardata += PRIV(ord2utf)(0x0100, class_uchardata);
4229 class_uchardata += PRIV(ord2utf)(0x2027, class_uchardata);
4230 *class_uchardata++ = XCL_RANGE;
4231 class_uchardata += PRIV(ord2utf)(0x202a, class_uchardata);
4232 class_uchardata += PRIV(ord2utf)(0x10ffff, class_uchardata);
4233 }
4234 #endif
4235 continue;
4236
4237 #ifdef SUPPORT_UCP
4238 case ESC_p:
4239 case ESC_P:
4240 {
4241 BOOL negated;
4242 int pdata;
4243 int ptype = get_ucp(&ptr, &negated, &pdata, errorcodeptr);
4244 if (ptype < 0) goto FAILED;
4245 xclass = TRUE;
4246 *class_uchardata++ = ((-c == ESC_p) != negated)?
4247 XCL_PROP : XCL_NOTPROP;
4248 *class_uchardata++ = ptype;
4249 *class_uchardata++ = pdata;
4250 class_has_8bitchar--; /* Undo! */
4251 continue;
4252 }
4253 #endif
4254 /* Unrecognized escapes are faulted if PCRE is running in its
4255 strict mode. By default, for compatibility with Perl, they are
4256 treated as literals. */
4257
4258 default:
4259 if ((options & PCRE_EXTRA) != 0)
4260 {
4261 *errorcodeptr = ERR7;
4262 goto FAILED;
4263 }
4264 class_has_8bitchar--; /* Undo the speculative increase. */
4265 class_single_char -= 2; /* Undo the speculative increase. */
4266 c = *ptr; /* Get the final character and fall through */
4267 break;
4268 }
4269 }
4270
4271 /* Fall through if we have a single character (c >= 0). This may be
4272 greater than 256. */
4273
4274 } /* End of backslash handling */
4275
4276 /* A single character may be followed by '-' to form a range. However,
4277 Perl does not permit ']' to be the end of the range. A '-' character
4278 at the end is treated as a literal. Perl ignores orphaned \E sequences
4279 entirely. The code for handling \Q and \E is messy. */
4280
4281 CHECK_RANGE:
4282 while (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
4283 {
4284 inescq = FALSE;
4285 ptr += 2;
4286 }
4287
4288 oldptr = ptr;
4289
4290 /* Remember \r or \n */
4291
4292 if (c == CHAR_CR || c == CHAR_NL) cd->external_flags |= PCRE_HASCRORLF;
4293
4294 /* Check for range */
4295
4296 if (!inescq && ptr[1] == CHAR_MINUS)
4297 {
4298 int d;
4299 ptr += 2;
4300 while (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_E) ptr += 2;
4301
4302 /* If we hit \Q (not followed by \E) at this point, go into escaped
4303 mode. */
4304
4305 while (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_Q)
4306 {
4307 ptr += 2;
4308 if (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_E)
4309 { ptr += 2; continue; }
4310 inescq = TRUE;
4311 break;
4312 }
4313
4314 if (*ptr == 0 || (!inescq && *ptr == CHAR_RIGHT_SQUARE_BRACKET))
4315 {
4316 ptr = oldptr;
4317 goto LONE_SINGLE_CHARACTER;
4318 }
4319
4320 #ifdef SUPPORT_UTF
4321 if (utf)
4322 { /* Braces are required because the */
4323 GETCHARLEN(d, ptr, ptr); /* macro generates multiple statements */
4324 }
4325 else
4326 #endif
4327 d = *ptr; /* Not UTF-8 mode */
4328
4329 /* The second part of a range can be a single-character escape, but
4330 not any of the other escapes. Perl 5.6 treats a hyphen as a literal
4331 in such circumstances. */
4332
4333 if (!inescq && d == CHAR_BACKSLASH)
4334 {
4335 d = check_escape(&ptr, errorcodeptr, cd->bracount, options, TRUE);
4336 if (*errorcodeptr != 0) goto FAILED;
4337
4338 /* \b is backspace; any other special means the '-' was literal */
4339
4340 if (d < 0)
4341 {
4342 if (d == -ESC_b) d = CHAR_BS; else
4343 {
4344 ptr = oldptr;
4345 goto LONE_SINGLE_CHARACTER; /* A few lines below */
4346 }
4347 }
4348 }
4349
4350 /* Check that the two values are in the correct order. Optimize
4351 one-character ranges */
4352
4353 if (d < c)
4354 {
4355 *errorcodeptr = ERR8;
4356 goto FAILED;
4357 }
4358
4359 if (d == c) goto LONE_SINGLE_CHARACTER; /* A few lines below */
4360
4361 /* Remember \r or \n */
4362
4363 if (d == CHAR_CR || d == CHAR_NL) cd->external_flags |= PCRE_HASCRORLF;
4364
4365 /* Since we found a character range, single character optimizations
4366 cannot be done anymore. */
4367 class_single_char = 2;
4368
4369 /* In UTF-8 mode, if the upper limit is > 255, or > 127 for caseless
4370 matching, we have to use an XCLASS with extra data items. Caseless
4371 matching for characters > 127 is available only if UCP support is
4372 available. */
4373
4374 #if defined SUPPORT_UTF && !(defined COMPILE_PCRE8)
4375 if ((d > 255) || (utf && ((options & PCRE_CASELESS) != 0 && d > 127)))
4376 #elif defined SUPPORT_UTF
4377 if (utf && (d > 255 || ((options & PCRE_CASELESS) != 0 && d > 127)))
4378 #elif !(defined COMPILE_PCRE8)
4379 if (d > 255)
4380 #endif
4381 #if defined SUPPORT_UTF || !(defined COMPILE_PCRE8)
4382 {
4383 xclass = TRUE;
4384
4385 /* With UCP support, we can find the other case equivalents of
4386 the relevant characters. There may be several ranges. Optimize how
4387 they fit with the basic range. */
4388
4389 #ifdef SUPPORT_UCP
4390 #ifndef COMPILE_PCRE8
4391 if (utf && (options & PCRE_CASELESS) != 0)
4392 #else
4393 if ((options & PCRE_CASELESS) != 0)
4394 #endif
4395 {
4396 unsigned int occ, ocd;
4397 unsigned int cc = c;
4398 unsigned int origd = d;
4399 while (get_othercase_range(&cc, origd, &occ, &ocd))
4400 {
4401 if (occ >= (unsigned int)c &&
4402 ocd <= (unsigned int)d)
4403 continue; /* Skip embedded ranges */
4404
4405 if (occ < (unsigned int)c &&
4406 ocd >= (unsigned int)c - 1) /* Extend the basic range */
4407 { /* if there is overlap, */
4408 c = occ; /* noting that if occ < c */
4409 continue; /* we can't have ocd > d */
4410 } /* because a subrange is */
4411 if (ocd > (unsigned int)d &&
4412 occ <= (unsigned int)d + 1) /* always shorter than */
4413 { /* the basic range. */
4414 d = ocd;
4415 continue;
4416 }
4417
4418 if (occ == ocd)
4419 {
4420 *class_uchardata++ = XCL_SINGLE;
4421 }
4422 else
4423 {
4424 *class_uchardata++ = XCL_RANGE;
4425 class_uchardata += PRIV(ord2utf)(occ, class_uchardata);
4426 }
4427 class_uchardata += PRIV(ord2utf)(ocd, class_uchardata);
4428 }
4429 }
4430 #endif /* SUPPORT_UCP */
4431
4432 /* Now record the original range, possibly modified for UCP caseless
4433 overlapping ranges. */
4434
4435 *class_uchardata++ = XCL_RANGE;
4436 #ifdef SUPPORT_UTF
4437 #ifndef COMPILE_PCRE8
4438 if (utf)
4439 {
4440 class_uchardata += PRIV(ord2utf)(c, class_uchardata);
4441 class_uchardata += PRIV(ord2utf)(d, class_uchardata);
4442 }
4443 else
4444 {
4445 *class_uchardata++ = c;
4446 *class_uchardata++ = d;
4447 }
4448 #else
4449 class_uchardata += PRIV(ord2utf)(c, class_uchardata);
4450 class_uchardata += PRIV(ord2utf)(d, class_uchardata);
4451 #endif
4452 #else /* SUPPORT_UTF */
4453 *class_uchardata++ = c;
4454 *class_uchardata++ = d;
4455 #endif /* SUPPORT_UTF */
4456
4457 /* With UCP support, we are done. Without UCP support, there is no
4458 caseless matching for UTF characters > 127; we can use the bit map
4459 for the smaller ones. As for 16 bit characters without UTF, we
4460 can still use */
4461
4462 #ifdef SUPPORT_UCP
4463 #ifndef COMPILE_PCRE8
4464 if (utf)
4465 #endif
4466 continue; /* With next character in the class */
4467 #endif /* SUPPORT_UCP */
4468
4469 #if defined SUPPORT_UTF && !defined(SUPPORT_UCP) && !(defined COMPILE_PCRE8)
4470 if (utf)
4471 {
4472 if ((options & PCRE_CASELESS) == 0 || c > 127) continue;
4473 /* Adjust upper limit and fall through to set up the map */
4474 d = 127;
4475 }
4476 else
4477 {
4478 if (c > 255) continue;
4479 /* Adjust upper limit and fall through to set up the map */
4480 d = 255;
4481 }
4482 #elif defined SUPPORT_UTF && !defined(SUPPORT_UCP)
4483 if ((options & PCRE_CASELESS) == 0 || c > 127) continue;
4484 /* Adjust upper limit and fall through to set up the map */
4485 d = 127;
4486 #else
4487 if (c > 255) continue;
4488 /* Adjust upper limit and fall through to set up the map */
4489 d = 255;
4490 #endif /* SUPPORT_UTF && !SUPPORT_UCP && !COMPILE_PCRE8 */
4491 }
4492 #endif /* SUPPORT_UTF || !COMPILE_PCRE8 */
4493
4494 /* We use the bit map for 8 bit mode, or when the characters fall
4495 partially or entirely to [0-255] ([0-127] for UCP) ranges. */
4496
4497 class_has_8bitchar = 1;
4498
4499 /* We can save a bit of time by skipping this in the pre-compile. */
4500
4501 if (lengthptr == NULL) for (; c <= d; c++)
4502 {
4503 classbits[c/8] |= (1 << (c&7));
4504 if ((options & PCRE_CASELESS) != 0)
4505 {
4506 int uc = cd->fcc[c]; /* flip case */
4507 classbits[uc/8] |= (1 << (uc&7));
4508 }
4509 }
4510
4511 continue; /* Go get the next char in the class */
4512 }
4513
4514 /* Handle a lone single character - we can get here for a normal
4515 non-escape char, or after \ that introduces a single character or for an
4516 apparent range that isn't. */
4517
4518 LONE_SINGLE_CHARACTER:
4519
4520 /* Only the value of 1 matters for class_single_char. */
4521
4522 if (class_single_char < 2) class_single_char++;
4523
4524 /* If class_charcount is 1, we saw precisely one character. As long as
4525 there was no use of \p or \P, in other words, no use of any XCLASS
4526 features, we can optimize.
4527
4528 The optimization throws away the bit map. We turn the item into a
4529 1-character OP_CHAR[I] if it's positive, or OP_NOT[I] if it's negative.
4530 In the positive case, it can cause firstchar to be set. Otherwise, there
4531 can be no first char if this item is first, whatever repeat count may
4532 follow. In the case of reqchar, save the previous value for reinstating. */
4533
4534 if (class_single_char == 1 && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET)
4535 {
4536 ptr++;
4537 zeroreqchar = reqchar;
4538
4539 if (negate_class)
4540 {
4541 if (firstchar == REQ_UNSET) firstchar = REQ_NONE;
4542 zerofirstchar = firstchar;
4543 *code++ = ((options & PCRE_CASELESS) != 0)? OP_NOTI: OP_NOT;
4544 #ifdef SUPPORT_UTF
4545 if (utf && c > MAX_VALUE_FOR_SINGLE_CHAR)
4546 code += PRIV(ord2utf)(c, code);
4547 else
4548 #endif
4549 *code++ = c;
4550 goto NOT_CHAR;
4551 }
4552
4553 /* For a single, positive character, get the value into mcbuffer, and
4554 then we can handle this with the normal one-character code. */
4555
4556 #ifdef SUPPORT_UTF
4557 if (utf && c > MAX_VALUE_FOR_SINGLE_CHAR)
4558 mclength = PRIV(ord2utf)(c, mcbuffer);
4559 else
4560 #endif
4561 {
4562 mcbuffer[0] = c;
4563 mclength = 1;
4564 }
4565 goto ONE_CHAR;
4566 } /* End of 1-char optimization */
4567
4568 /* Handle a character that cannot go in the bit map. */
4569
4570 #if defined SUPPORT_UTF && !(defined COMPILE_PCRE8)
4571 if ((c > 255) || (utf && ((options & PCRE_CASELESS) != 0 && c > 127)))
4572 #elif defined SUPPORT_UTF
4573 if (utf && (c > 255 || ((options & PCRE_CASELESS) != 0 && c > 127)))
4574 #elif !(defined COMPILE_PCRE8)
4575 if (c > 255)
4576 #endif
4577
4578 #if defined SUPPORT_UTF || !(defined COMPILE_PCRE8)
4579 {
4580 xclass = TRUE;
4581 *class_uchardata++ = XCL_SINGLE;
4582 #ifdef SUPPORT_UTF
4583 #ifndef COMPILE_PCRE8
4584 /* In non 8 bit mode, we can get here even if we are not in UTF mode. */
4585 if (!utf)
4586 *class_uchardata++ = c;
4587 else
4588 #endif
4589 class_uchardata += PRIV(ord2utf)(c, class_uchardata);
4590 #else /* SUPPORT_UTF */
4591 *class_uchardata++ = c;
4592 #endif /* SUPPORT_UTF */
4593
4594 #ifdef SUPPORT_UCP
4595 #ifdef COMPILE_PCRE8
4596 if ((options & PCRE_CASELESS) != 0)
4597 #else
4598 /* In non 8 bit mode, we can get here even if we are not in UTF mode. */
4599 if (utf && (options & PCRE_CASELESS) != 0)
4600 #endif
4601 {
4602 unsigned int othercase;
4603 if ((int)(othercase = UCD_OTHERCASE(c)) != c)
4604 {
4605 *class_uchardata++ = XCL_SINGLE;
4606 class_uchardata += PRIV(ord2utf)(othercase, class_uchardata);
4607 }
4608 }
4609 #endif /* SUPPORT_UCP */
4610
4611 }
4612 else
4613 #endif /* SUPPORT_UTF || COMPILE_PCRE16 */
4614
4615 /* Handle a single-byte character */
4616 {
4617 class_has_8bitchar = 1;
4618 classbits[c/8] |= (1 << (c&7));
4619 if ((options & PCRE_CASELESS) != 0)
4620 {
4621 c = cd->fcc[c]; /* flip case */
4622 classbits[c/8] |= (1 << (c&7));
4623 }
4624 }
4625 }
4626
4627 /* Loop until ']' reached. This "while" is the end of the "do" far above.
4628 If we are at the end of an internal nested string, revert to the outer
4629 string. */
4630
4631 while (((c = *(++ptr)) != 0 ||
4632 (nestptr != NULL &&
4633 (ptr = nestptr, nestptr = NULL, c = *(++ptr)) != 0)) &&
4634 (c != CHAR_RIGHT_SQUARE_BRACKET || inescq));
4635
4636 /* Check for missing terminating ']' */
4637
4638 if (c == 0)
4639 {
4640 *errorcodeptr = ERR6;
4641 goto FAILED;
4642 }
4643
4644 /* If this is the first thing in the branch, there can be no first char
4645 setting, whatever the repeat count. Any reqchar setting must remain
4646 unchanged after any kind of repeat. */
4647
4648 if (firstchar == REQ_UNSET) firstchar = REQ_NONE;
4649 zerofirstchar = firstchar;
4650 zeroreqchar = reqchar;
4651
4652 /* If there are characters with values > 255, we have to compile an
4653 extended class, with its own opcode, unless there was a negated special
4654 such as \S in the class, and PCRE_UCP is not set, because in that case all
4655 characters > 255 are in the class, so any that were explicitly given as
4656 well can be ignored. If (when there are explicit characters > 255 that must
4657 be listed) there are no characters < 256, we can omit the bitmap in the
4658 actual compiled code. */
4659
4660 #ifdef SUPPORT_UTF
4661 if (xclass && (!should_flip_negation || (options & PCRE_UCP) != 0))
4662 #elif !defined COMPILE_PCRE8
4663 if (xclass && !should_flip_negation)
4664 #endif
4665 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
4666 {
4667 *class_uchardata++ = XCL_END; /* Marks the end of extra data */
4668 *code++ = OP_XCLASS;
4669 code += LINK_SIZE;
4670 *code = negate_class? XCL_NOT:0;
4671
4672 /* If the map is required, move up the extra data to make room for it;
4673 otherwise just move the code pointer to the end of the extra data. */
4674
4675 if (class_has_8bitchar > 0)
4676 {
4677 *code++ |= XCL_MAP;
4678 memmove(code + (32 / sizeof(pcre_uchar)), code,
4679 IN_UCHARS(class_uchardata - code));
4680 memcpy(code, classbits, 32);
4681 code = class_uchardata + (32 / sizeof(pcre_uchar));
4682 }
4683 else code = class_uchardata;
4684
4685 /* Now fill in the complete length of the item */
4686
4687 PUT(previous, 1, (int)(code - previous));
4688 break; /* End of class handling */
4689 }
4690 #endif
4691
4692 /* If there are no characters > 255, or they are all to be included or
4693 excluded, set the opcode to OP_CLASS or OP_NCLASS, depending on whether the
4694 whole class was negated and whether there were negative specials such as \S
4695 (non-UCP) in the class. Then copy the 32-byte map into the code vector,
4696 negating it if necessary. */
4697
4698 *code++ = (negate_class == should_flip_negation) ? OP_CLASS : OP_NCLASS;
4699 if (lengthptr == NULL) /* Save time in the pre-compile phase */
4700 {
4701 if (negate_class)
4702 for (c = 0; c < 32; c++) classbits[c] = ~classbits[c];
4703 memcpy(code, classbits, 32);
4704 }
4705 code += 32 / sizeof(pcre_uchar);
4706 NOT_CHAR:
4707 break;
4708
4709
4710 /* ===================================================================*/
4711 /* Various kinds of repeat; '{' is not necessarily a quantifier, but this
4712 has been tested above. */
4713
4714 case CHAR_LEFT_CURLY_BRACKET:
4715 if (!is_quantifier) goto NORMAL_CHAR;
4716 ptr = read_repeat_counts(ptr+1, &repeat_min, &repeat_max, errorcodeptr);
4717 if (*errorcodeptr != 0) goto FAILED;
4718 goto REPEAT;
4719
4720 case CHAR_ASTERISK:
4721 repeat_min = 0;
4722 repeat_max = -1;
4723 goto REPEAT;
4724
4725 case CHAR_PLUS:
4726 repeat_min = 1;
4727 repeat_max = -1;
4728 goto REPEAT;
4729
4730 case CHAR_QUESTION_MARK:
4731 repeat_min = 0;
4732 repeat_max = 1;
4733
4734 REPEAT:
4735 if (previous == NULL)
4736 {
4737 *errorcodeptr = ERR9;
4738 goto FAILED;
4739 }
4740
4741 if (repeat_min == 0)
4742 {
4743 firstchar = zerofirstchar; /* Adjust for zero repeat */
4744 reqchar = zeroreqchar; /* Ditto */
4745 }
4746
4747 /* Remember whether this is a variable length repeat */
4748
4749 reqvary = (repeat_min == repeat_max)? 0 : REQ_VARY;
4750
4751 op_type = 0; /* Default single-char op codes */
4752 possessive_quantifier = FALSE; /* Default not possessive quantifier */
4753
4754 /* Save start of previous item, in case we have to move it up in order to
4755 insert something before it. */
4756
4757 tempcode = previous;
4758
4759 /* If the next character is '+', we have a possessive quantifier. This
4760 implies greediness, whatever the setting of the PCRE_UNGREEDY option.
4761 If the next character is '?' this is a minimizing repeat, by default,
4762 but if PCRE_UNGREEDY is set, it works the other way round. We change the
4763 repeat type to the non-default. */
4764
4765 if (ptr[1] == CHAR_PLUS)
4766 {
4767 repeat_type = 0; /* Force greedy */
4768 possessive_quantifier = TRUE;
4769 ptr++;
4770 }
4771 else if (ptr[1] == CHAR_QUESTION_MARK)
4772 {
4773 repeat_type = greedy_non_default;
4774 ptr++;
4775 }
4776 else repeat_type = greedy_default;
4777
4778 /* If previous was a recursion call, wrap it in atomic brackets so that
4779 previous becomes the atomic group. All recursions were so wrapped in the
4780 past, but it no longer happens for non-repeated recursions. In fact, the
4781 repeated ones could be re-implemented independently so as not to need this,
4782 but for the moment we rely on the code for repeating groups. */
4783
4784 if (*previous == OP_RECURSE)
4785 {
4786 memmove(previous + 1 + LINK_SIZE, previous, IN_UCHARS(1 + LINK_SIZE));
4787 *previous = OP_ONCE;
4788 PUT(previous, 1, 2 + 2*LINK_SIZE);
4789 previous[2 + 2*LINK_SIZE] = OP_KET;
4790 PUT(previous, 3 + 2*LINK_SIZE, 2 + 2*LINK_SIZE);
4791 code += 2 + 2 * LINK_SIZE;
4792 length_prevgroup = 3 + 3*LINK_SIZE;
4793
4794 /* When actually compiling, we need to check whether this was a forward
4795 reference, and if so, adjust the offset. */
4796
4797 if (lengthptr == NULL && cd->hwm >= cd->start_workspace + LINK_SIZE)
4798 {
4799 int offset = GET(cd->hwm, -LINK_SIZE);
4800 if (offset == previous + 1 - cd->start_code)
4801 PUT(cd->hwm, -LINK_SIZE, offset + 1 + LINK_SIZE);
4802 }
4803 }
4804
4805 /* Now handle repetition for the different types of item. */
4806
4807 /* If previous was a character or negated character match, abolish the item
4808 and generate a repeat item instead. If a char item has a minimum of more
4809 than one, ensure that it is set in reqchar - it might not be if a sequence
4810 such as x{3} is the first thing in a branch because the x will have gone
4811 into firstchar instead. */
4812
4813 if (*previous == OP_CHAR || *previous == OP_CHARI
4814 || *previous == OP_NOT || *previous == OP_NOTI)
4815 {
4816 switch (*previous)
4817 {
4818 default: /* Make compiler happy. */
4819 case OP_CHAR: op_type = OP_STAR - OP_STAR; break;
4820 case OP_CHARI: op_type = OP_STARI - OP_STAR; break;
4821 case OP_NOT: op_type = OP_NOTSTAR - OP_STAR; break;
4822 case OP_NOTI: op_type = OP_NOTSTARI - OP_STAR; break;
4823 }
4824
4825 /* Deal with UTF characters that take up more than one character. It's
4826 easier to write this out separately than try to macrify it. Use c to
4827 hold the length of the character in bytes, plus UTF_LENGTH to flag that
4828 it's a length rather than a small character. */
4829
4830 #ifdef SUPPORT_UTF
4831 if (utf && NOT_FIRSTCHAR(code[-1]))
4832 {
4833 pcre_uchar *lastchar = code - 1;
4834 BACKCHAR(lastchar);
4835 c = (int)(code - lastchar); /* Length of UTF-8 character */
4836 memcpy(utf_chars, lastchar, IN_UCHARS(c)); /* Save the char */
4837 c |= UTF_LENGTH; /* Flag c as a length */
4838 }
4839 else
4840 #endif /* SUPPORT_UTF */
4841
4842 /* Handle the case of a single charater - either with no UTF support, or
4843 with UTF disabled, or for a single character UTF character. */
4844 {
4845 c = code[-1];
4846 if (*previous <= OP_CHARI && repeat_min > 1)
4847 reqchar = c | req_caseopt | cd->req_varyopt;
4848 }
4849
4850 /* If the repetition is unlimited, it pays to see if the next thing on
4851 the line is something that cannot possibly match this character. If so,
4852 automatically possessifying this item gains some performance in the case
4853 where the match fails. */
4854
4855 if (!possessive_quantifier &&
4856 repeat_max < 0 &&
4857 check_auto_possessive(previous, utf, ptr + 1, options, cd))
4858 {
4859 repeat_type = 0; /* Force greedy */
4860 possessive_quantifier = TRUE;
4861 }
4862
4863 goto OUTPUT_SINGLE_REPEAT; /* Code shared with single character types */
4864 }
4865
4866 /* If previous was a character type match (\d or similar), abolish it and
4867 create a suitable repeat item. The code is shared with single-character
4868 repeats by setting op_type to add a suitable offset into repeat_type. Note
4869 the the Unicode property types will be present only when SUPPORT_UCP is
4870 defined, but we don't wrap the little bits of code here because it just
4871 makes it horribly messy. */
4872
4873 else if (*previous < OP_EODN)
4874 {
4875 pcre_uchar *oldcode;
4876 int prop_type, prop_value;
4877 op_type = OP_TYPESTAR - OP_STAR; /* Use type opcodes */
4878 c = *previous;
4879
4880 if (!possessive_quantifier &&
4881 repeat_max < 0 &&
4882 check_auto_possessive(previous, utf, ptr + 1, options, cd))
4883 {
4884 repeat_type = 0; /* Force greedy */
4885 possessive_quantifier = TRUE;
4886 }
4887
4888 OUTPUT_SINGLE_REPEAT:
4889 if (*previous == OP_PROP || *previous == OP_NOTPROP)
4890 {
4891 prop_type = previous[1];
4892 prop_value = previous[2];
4893 }
4894 else prop_type = prop_value = -1;
4895
4896 oldcode = code;
4897 code = previous; /* Usually overwrite previous item */
4898
4899 /* If the maximum is zero then the minimum must also be zero; Perl allows
4900 this case, so we do too - by simply omitting the item altogether. */
4901
4902 if (repeat_max == 0) goto END_REPEAT;
4903
4904 /*--------------------------------------------------------------------*/
4905 /* This code is obsolete from release 8.00; the restriction was finally
4906 removed: */
4907
4908 /* All real repeats make it impossible to handle partial matching (maybe
4909 one day we will be able to remove this restriction). */
4910
4911 /* if (repeat_max != 1) cd->external_flags |= PCRE_NOPARTIAL; */
4912 /*--------------------------------------------------------------------*/
4913
4914 /* Combine the op_type with the repeat_type */
4915
4916 repeat_type += op_type;
4917
4918 /* A minimum of zero is handled either as the special case * or ?, or as
4919 an UPTO, with the maximum given. */
4920
4921 if (repeat_min == 0)
4922 {
4923 if (repeat_max == -1) *code++ = OP_STAR + repeat_type;
4924 else if (repeat_max == 1) *code++ = OP_QUERY + repeat_type;
4925 else
4926 {
4927 *code++ = OP_UPTO + repeat_type;
4928 PUT2INC(code, 0, repeat_max);
4929 }
4930 }
4931
4932 /* A repeat minimum of 1 is optimized into some special cases. If the
4933 maximum is unlimited, we use OP_PLUS. Otherwise, the original item is
4934 left in place and, if the maximum is greater than 1, we use OP_UPTO with
4935 one less than the maximum. */
4936
4937 else if (repeat_min == 1)
4938 {
4939 if (repeat_max == -1)
4940 *code++ = OP_PLUS + repeat_type;
4941 else
4942 {
4943 code = oldcode; /* leave previous item in place */
4944 if (repeat_max == 1) goto END_REPEAT;
4945 *code++ = OP_UPTO + repeat_type;
4946 PUT2INC(code, 0, repeat_max - 1);
4947 }
4948 }
4949
4950 /* The case {n,n} is just an EXACT, while the general case {n,m} is
4951 handled as an EXACT followed by an UPTO. */
4952
4953 else
4954 {
4955 *code++ = OP_EXACT + op_type; /* NB EXACT doesn't have repeat_type */
4956 PUT2INC(code, 0, repeat_min);
4957
4958 /* If the maximum is unlimited, insert an OP_STAR. Before doing so,
4959 we have to insert the character for the previous code. For a repeated
4960 Unicode property match, there are two extra bytes that define the
4961 required property. In UTF-8 mode, long characters have their length in
4962 c, with the UTF_LENGTH bit as a flag. */
4963
4964 if (repeat_max < 0)
4965 {
4966 #ifdef SUPPORT_UTF
4967 if (utf && (c & UTF_LENGTH) != 0)
4968 {
4969 memcpy(code, utf_chars, IN_UCHARS(c & 7));
4970 code += c & 7;
4971 }
4972 else
4973 #endif
4974 {
4975 *code++ = c;
4976 if (prop_type >= 0)
4977 {
4978 *code++ = prop_type;
4979 *code++ = prop_value;
4980 }
4981 }
4982 *code++ = OP_STAR + repeat_type;
4983 }
4984
4985 /* Else insert an UPTO if the max is greater than the min, again
4986 preceded by the character, for the previously inserted code. If the
4987 UPTO is just for 1 instance, we can use QUERY instead. */
4988
4989 else if (repeat_max != repeat_min)
4990 {
4991 #ifdef SUPPORT_UTF
4992 if (utf && (c & UTF_LENGTH) != 0)
4993 {
4994 memcpy(code, utf_chars, IN_UCHARS(c & 7));
4995 code += c & 7;
4996 }
4997 else
4998 #endif
4999 *code++ = c;
5000 if (prop_type >= 0)
5001 {
5002 *code++ = prop_type;
5003 *code++ = prop_value;
5004 }
5005 repeat_max -= repeat_min;
5006
5007 if (repeat_max == 1)
5008 {
5009 *code++ = OP_QUERY + repeat_type;
5010 }
5011 else
5012 {
5013 *code++ = OP_UPTO + repeat_type;
5014 PUT2INC(code, 0, repeat_max);
5015 }
5016 }
5017 }
5018
5019 /* The character or character type itself comes last in all cases. */
5020
5021 #ifdef SUPPORT_UTF
5022 if (utf && (c & UTF_LENGTH) != 0)
5023 {
5024 memcpy(code, utf_chars, IN_UCHARS(c & 7));
5025 code += c & 7;
5026 }
5027 else
5028 #endif
5029 *code++ = c;
5030
5031 /* For a repeated Unicode property match, there are two extra bytes that
5032 define the required property. */
5033
5034 #ifdef SUPPORT_UCP
5035 if (prop_type >= 0)
5036 {
5037 *code++ = prop_type;
5038 *code++ = prop_value;
5039 }
5040 #endif
5041 }
5042
5043 /* If previous was a character class or a back reference, we put the repeat
5044 stuff after it, but just skip the item if the repeat was {0,0}. */
5045
5046 else if (*previous == OP_CLASS ||
5047 *previous == OP_NCLASS ||
5048 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
5049 *previous == OP_XCLASS ||
5050 #endif
5051 *previous == OP_REF ||
5052 *previous == OP_REFI)
5053 {
5054 if (repeat_max == 0)
5055 {
5056 code = previous;
5057 goto END_REPEAT;
5058 }
5059
5060 /*--------------------------------------------------------------------*/
5061 /* This code is obsolete from release 8.00; the restriction was finally
5062 removed: */
5063
5064 /* All real repeats make it impossible to handle partial matching (maybe
5065 one day we will be able to remove this restriction). */
5066
5067 /* if (repeat_max != 1) cd->external_flags |= PCRE_NOPARTIAL; */
5068 /*--------------------------------------------------------------------*/
5069
5070 if (repeat_min == 0 && repeat_max == -1)
5071 *code++ = OP_CRSTAR + repeat_type;
5072 else if (repeat_min == 1 && repeat_max == -1)
5073 *code++ = OP_CRPLUS + repeat_type;
5074 else if (repeat_min == 0 && repeat_max == 1)
5075 *code++ = OP_CRQUERY + repeat_type;
5076 else
5077 {
5078 *code++ = OP_CRRANGE + repeat_type;
5079 PUT2INC(code, 0, repeat_min);
5080 if (repeat_max == -1) repeat_max = 0; /* 2-byte encoding for max */
5081 PUT2INC(code, 0, repeat_max);
5082 }
5083 }
5084
5085 /* If previous was a bracket group, we may have to replicate it in certain
5086 cases. Note that at this point we can encounter only the "basic" bracket
5087 opcodes such as BRA and CBRA, as this is the place where they get converted
5088 into the more special varieties such as BRAPOS and SBRA. A test for >=
5089 OP_ASSERT and <= OP_COND includes ASSERT, ASSERT_NOT, ASSERTBACK,
5090 ASSERTBACK_NOT, ONCE, BRA, CBRA, and COND. Originally, PCRE did not allow
5091 repetition of assertions, but now it does, for Perl compatibility. */
5092
5093 else if (*previous >= OP_ASSERT && *previous <= OP_COND)
5094 {
5095 register int i;
5096 int len = (int)(code - previous);
5097 pcre_uchar *bralink = NULL;
5098 pcre_uchar *brazeroptr = NULL;
5099
5100 /* Repeating a DEFINE group is pointless, but Perl allows the syntax, so
5101 we just ignore the repeat. */
5102
5103 if (*previous == OP_COND && previous[LINK_SIZE+1] == OP_DEF)
5104 goto END_REPEAT;
5105
5106 /* There is no sense in actually repeating assertions. The only potential
5107 use of repetition is in cases when the assertion is optional. Therefore,
5108 if the minimum is greater than zero, just ignore the repeat. If the
5109 maximum is not not zero or one, set it to 1. */
5110
5111 if (*previous < OP_ONCE) /* Assertion */
5112 {
5113 if (repeat_min > 0) goto END_REPEAT;
5114 if (repeat_max < 0 || repeat_max > 1) repeat_max = 1;
5115 }
5116
5117 /* The case of a zero minimum is special because of the need to stick
5118 OP_BRAZERO in front of it, and because the group appears once in the
5119 data, whereas in other cases it appears the minimum number of times. For
5120 this reason, it is simplest to treat this case separately, as otherwise
5121 the code gets far too messy. There are several special subcases when the
5122 minimum is zero. */
5123
5124 if (repeat_min == 0)
5125 {
5126 /* If the maximum is also zero, we used to just omit the group from the
5127 output altogether, like this:
5128
5129 ** if (repeat_max == 0)
5130 ** {
5131 ** code = previous;
5132 ** goto END_REPEAT;
5133 ** }
5134
5135 However, that fails when a group or a subgroup within it is referenced
5136 as a subroutine from elsewhere in the pattern, so now we stick in
5137 OP_SKIPZERO in front of it so that it is skipped on execution. As we
5138 don't have a list of which groups are referenced, we cannot do this
5139 selectively.
5140
5141 If the maximum is 1 or unlimited, we just have to stick in the BRAZERO
5142 and do no more at this point. However, we do need to adjust any
5143 OP_RECURSE calls inside the group that refer to the group itself or any
5144 internal or forward referenced group, because the offset is from the
5145 start of the whole regex. Temporarily terminate the pattern while doing
5146 this. */
5147
5148 if (repeat_max <= 1) /* Covers 0, 1, and unlimited */
5149 {
5150 *code = OP_END;
5151 adjust_recurse(previous, 1, utf, cd, save_hwm);
5152 memmove(previous + 1, previous, IN_UCHARS(len));
5153 code++;
5154 if (repeat_max == 0)
5155 {
5156 *previous++ = OP_SKIPZERO;
5157 goto END_REPEAT;
5158 }
5159 brazeroptr = previous; /* Save for possessive optimizing */
5160 *previous++ = OP_BRAZERO + repeat_type;
5161 }
5162
5163 /* If the maximum is greater than 1 and limited, we have to replicate
5164 in a nested fashion, sticking OP_BRAZERO before each set of brackets.
5165 The first one has to be handled carefully because it's the original
5166 copy, which has to be moved up. The remainder can be handled by code
5167 that is common with the non-zero minimum case below. We have to
5168 adjust the value or repeat_max, since one less copy is required. Once
5169 again, we may have to adjust any OP_RECURSE calls inside the group. */
5170
5171 else
5172 {
5173 int offset;
5174 *code = OP_END;
5175 adjust_recurse(previous, 2 + LINK_SIZE, utf, cd, save_hwm);
5176 memmove(previous + 2 + LINK_SIZE, previous, IN_UCHARS(len));
5177 code += 2 + LINK_SIZE;
5178 *previous++ = OP_BRAZERO + repeat_type;
5179 *previous++ = OP_BRA;
5180
5181 /* We chain together the bracket offset fields that have to be
5182 filled in later when the ends of the brackets are reached. */
5183
5184 offset = (bralink == NULL)? 0 : (int)(previous - bralink);
5185 bralink = previous;
5186 PUTINC(previous, 0, offset);
5187 }
5188
5189 repeat_max--;
5190 }
5191
5192 /* If the minimum is greater than zero, replicate the group as many
5193 times as necessary, and adjust the maximum to the number of subsequent
5194 copies that we need. If we set a first char from the group, and didn't
5195 set a required char, copy the latter from the former. If there are any
5196 forward reference subroutine calls in the group, there will be entries on
5197 the workspace list; replicate these with an appropriate increment. */
5198
5199 else
5200 {
5201 if (repeat_min > 1)
5202 {
5203 /* In the pre-compile phase, we don't actually do the replication. We
5204 just adjust the length as if we had. Do some paranoid checks for
5205 potential integer overflow. The INT64_OR_DOUBLE type is a 64-bit
5206 integer type when available, otherwise double. */
5207
5208 if (lengthptr != NULL)
5209 {
5210 int delta = (repeat_min - 1)*length_prevgroup;
5211 if ((INT64_OR_DOUBLE)(repeat_min - 1)*
5212 (INT64_OR_DOUBLE)length_prevgroup >
5213 (INT64_OR_DOUBLE)INT_MAX ||
5214 OFLOW_MAX - *lengthptr < delta)
5215 {
5216 *errorcodeptr = ERR20;
5217 goto FAILED;
5218 }
5219 *lengthptr += delta;
5220 }
5221
5222 /* This is compiling for real. If there is a set first byte for
5223 the group, and we have not yet set a "required byte", set it. Make
5224 sure there is enough workspace for copying forward references before
5225 doing the copy. */
5226
5227 else
5228 {
5229 if (groupsetfirstchar && reqchar < 0) reqchar = firstchar;
5230
5231 for (i = 1; i < repeat_min; i++)
5232 {
5233 pcre_uchar *hc;
5234 pcre_uchar *this_hwm = cd->hwm;
5235 memcpy(code, previous, IN_UCHARS(len));
5236
5237 while (cd->hwm > cd->start_workspace + cd->workspace_size -
5238 WORK_SIZE_SAFETY_MARGIN - (this_hwm - save_hwm))
5239 {
5240 int save_offset = save_hwm - cd->start_workspace;
5241 int this_offset = this_hwm - cd->start_workspace;
5242 *errorcodeptr = expand_workspace(cd);
5243 if (*errorcodeptr != 0) goto FAILED;
5244 save_hwm = (pcre_uchar *)cd->start_workspace + save_offset;
5245 this_hwm = (pcre_uchar *)cd->start_workspace + this_offset;
5246 }
5247
5248 for (hc = save_hwm; hc < this_hwm; hc += LINK_SIZE)
5249 {
5250 PUT(cd->hwm, 0, GET(hc, 0) + len);
5251 cd->hwm += LINK_SIZE;
5252 }
5253 save_hwm = this_hwm;
5254 code += len;
5255 }
5256 }
5257 }
5258
5259 if (repeat_max > 0) repeat_max -= repeat_min;
5260 }
5261
5262 /* This code is common to both the zero and non-zero minimum cases. If
5263 the maximum is limited, it replicates the group in a nested fashion,
5264 remembering the bracket starts on a stack. In the case of a zero minimum,
5265 the first one was set up above. In all cases the repeat_max now specifies
5266 the number of additional copies needed. Again, we must remember to
5267 replicate entries on the forward reference list. */
5268
5269 if (repeat_max >= 0)
5270 {
5271 /* In the pre-compile phase, we don't actually do the replication. We
5272 just adjust the length as if we had. For each repetition we must add 1
5273 to the length for BRAZERO and for all but the last repetition we must
5274 add 2 + 2*LINKSIZE to allow for the nesting that occurs. Do some
5275 paranoid checks to avoid integer overflow. The INT64_OR_DOUBLE type is
5276 a 64-bit integer type when available, otherwise double. */
5277
5278 if (lengthptr != NULL && repeat_max > 0)
5279 {
5280 int delta = repeat_max * (length_prevgroup + 1 + 2 + 2*LINK_SIZE) -
5281 2 - 2*LINK_SIZE; /* Last one doesn't nest */
5282 if ((INT64_OR_DOUBLE)repeat_max *
5283 (INT64_OR_DOUBLE)(length_prevgroup + 1 + 2 + 2*LINK_SIZE)
5284 > (INT64_OR_DOUBLE)INT_MAX ||
5285 OFLOW_MAX - *lengthptr < delta)
5286 {
5287 *errorcodeptr = ERR20;
5288 goto FAILED;
5289 }
5290 *lengthptr += delta;
5291 }
5292
5293 /* This is compiling for real */
5294
5295 else for (i = repeat_max - 1; i >= 0; i--)
5296 {
5297 pcre_uchar *hc;
5298 pcre_uchar *this_hwm = cd->hwm;
5299
5300 *code++ = OP_BRAZERO + repeat_type;
5301
5302 /* All but the final copy start a new nesting, maintaining the
5303 chain of brackets outstanding. */
5304
5305 if (i != 0)
5306 {
5307 int offset;
5308 *code++ = OP_BRA;
5309 offset = (bralink == NULL)? 0 : (int)(code - bralink);
5310 bralink = code;
5311 PUTINC(code, 0, offset);
5312 }
5313
5314 memcpy(code, previous, IN_UCHARS(len));
5315
5316 /* Ensure there is enough workspace for forward references before
5317 copying them. */
5318
5319 while (cd->hwm > cd->start_workspace + cd->workspace_size -
5320 WORK_SIZE_SAFETY_MARGIN - (this_hwm - save_hwm))
5321 {
5322 int save_offset = save_hwm - cd->start_workspace;
5323 int this_offset = this_hwm - cd->start_workspace;
5324 *errorcodeptr = expand_workspace(cd);
5325 if (*errorcodeptr != 0) goto FAILED;
5326 save_hwm = (pcre_uchar *)cd->start_workspace + save_offset;
5327 this_hwm = (pcre_uchar *)cd->start_workspace + this_offset;
5328 }
5329
5330 for (hc = save_hwm; hc < this_hwm; hc += LINK_SIZE)
5331 {
5332 PUT(cd->hwm, 0, GET(hc, 0) + len + ((i != 0)? 2+LINK_SIZE : 1));
5333 cd->hwm += LINK_SIZE;
5334 }
5335 save_hwm = this_hwm;
5336 code += len;
5337 }
5338
5339 /* Now chain through the pending brackets, and fill in their length
5340 fields (which are holding the chain links pro tem). */
5341
5342 while (bralink != NULL)
5343 {
5344 int oldlinkoffset;
5345 int offset = (int)(code - bralink + 1);
5346 pcre_uchar *bra = code - offset;
5347 oldlinkoffset = GET(bra, 1);
5348 bralink = (oldlinkoffset == 0)? NULL : bralink - oldlinkoffset;
5349 *code++ = OP_KET;
5350 PUTINC(code, 0, offset);
5351 PUT(bra, 1, offset);
5352 }
5353 }
5354
5355 /* If the maximum is unlimited, set a repeater in the final copy. For
5356 ONCE brackets, that's all we need to do. However, possessively repeated
5357 ONCE brackets can be converted into non-capturing brackets, as the
5358 behaviour of (?:xx)++ is the same as (?>xx)++ and this saves having to
5359 deal with possessive ONCEs specially.
5360
5361 Otherwise, when we are doing the actual compile phase, check to see
5362 whether this group is one that could match an empty string. If so,
5363 convert the initial operator to the S form (e.g. OP_BRA -> OP_SBRA) so
5364 that runtime checking can be done. [This check is also applied to ONCE
5365 groups at runtime, but in a different way.]
5366
5367 Then, if the quantifier was possessive and the bracket is not a
5368 conditional, we convert the BRA code to the POS form, and the KET code to
5369 KETRPOS. (It turns out to be convenient at runtime to detect this kind of
5370 subpattern at both the start and at the end.) The use of special opcodes
5371 makes it possible to reduce greatly the stack usage in pcre_exec(). If
5372 the group is preceded by OP_BRAZERO, convert this to OP_BRAPOSZERO.
5373
5374 Then, if the minimum number of matches is 1 or 0, cancel the possessive
5375 flag so that the default action below, of wrapping everything inside
5376 atomic brackets, does not happen. When the minimum is greater than 1,
5377 there will be earlier copies of the group, and so we still have to wrap
5378 the whole thing. */
5379
5380 else
5381 {
5382 pcre_uchar *ketcode = code - 1 - LINK_SIZE;
5383 pcre_uchar *bracode = ketcode - GET(ketcode, 1);
5384
5385 /* Convert possessive ONCE brackets to non-capturing */
5386
5387 if ((*bracode == OP_ONCE || *bracode == OP_ONCE_NC) &&
5388 possessive_quantifier) *bracode = OP_BRA;
5389
5390 /* For non-possessive ONCE brackets, all we need to do is to
5391 set the KET. */
5392
5393 if (*bracode == OP_ONCE || *bracode == OP_ONCE_NC)
5394 *ketcode = OP_KETRMAX + repeat_type;
5395
5396 /* Handle non-ONCE brackets and possessive ONCEs (which have been
5397 converted to non-capturing above). */
5398
5399 else
5400 {
5401 /* In the compile phase, check for empty string matching. */
5402
5403 if (lengthptr == NULL)
5404 {
5405 pcre_uchar *scode = bracode;
5406 do
5407 {
5408 if (could_be_empty_branch(scode, ketcode, utf, cd))
5409 {
5410 *bracode += OP_SBRA - OP_BRA;
5411 break;
5412 }
5413 scode += GET(scode, 1);
5414 }
5415 while (*scode == OP_ALT);
5416 }
5417
5418 /* Handle possessive quantifiers. */
5419
5420 if (possessive_quantifier)
5421 {
5422 /* For COND brackets, we wrap the whole thing in a possessively
5423 repeated non-capturing bracket, because we have not invented POS
5424 versions of the COND opcodes. Because we are moving code along, we
5425 must ensure that any pending recursive references are updated. */
5426
5427 if (*bracode == OP_COND || *bracode == OP_SCOND)
5428 {
5429 int nlen = (int)(code - bracode);
5430 *code = OP_END;
5431 adjust_recurse(bracode, 1 + LINK_SIZE, utf, cd, save_hwm);
5432 memmove(bracode + 1 + LINK_SIZE, bracode, IN_UCHARS(nlen));
5433 code += 1 + LINK_SIZE;
5434 nlen += 1 + LINK_SIZE;
5435 *bracode = OP_BRAPOS;
5436 *code++ = OP_KETRPOS;
5437 PUTINC(code, 0, nlen);
5438 PUT(bracode, 1, nlen);
5439 }
5440
5441 /* For non-COND brackets, we modify the BRA code and use KETRPOS. */
5442
5443 else
5444 {
5445 *bracode += 1; /* Switch to xxxPOS opcodes */
5446 *ketcode = OP_KETRPOS;
5447 }
5448
5449 /* If the minimum is zero, mark it as possessive, then unset the
5450 possessive flag when the minimum is 0 or 1. */
5451
5452 if (brazeroptr != NULL) *brazeroptr = OP_BRAPOSZERO;
5453 if (repeat_min < 2) possessive_quantifier = FALSE;
5454 }
5455
5456 /* Non-possessive quantifier */
5457
5458 else *ketcode = OP_KETRMAX + repeat_type;
5459 }
5460 }
5461 }
5462
5463 /* If previous is OP_FAIL, it was generated by an empty class [] in
5464 JavaScript mode. The other ways in which OP_FAIL can be generated, that is
5465 by (*FAIL) or (?!) set previous to NULL, which gives a "nothing to repeat"
5466 error above. We can just ignore the repeat in JS case. */
5467
5468 else if (*previous == OP_FAIL) goto END_REPEAT;
5469
5470 /* Else there's some kind of shambles */
5471
5472 else
5473 {
5474 *errorcodeptr = ERR11;
5475 goto FAILED;
5476 }
5477
5478 /* If the character following a repeat is '+', or if certain optimization
5479 tests above succeeded, possessive_quantifier is TRUE. For some opcodes,
5480 there are special alternative opcodes for this case. For anything else, we
5481 wrap the entire repeated item inside OP_ONCE brackets. Logically, the '+'
5482 notation is just syntactic sugar, taken from Sun's Java package, but the
5483 special opcodes can optimize it.
5484
5485 Some (but not all) possessively repeated subpatterns have already been
5486 completely handled in the code just above. For them, possessive_quantifier
5487 is always FALSE at this stage.
5488
5489 Note that the repeated item starts at tempcode, not at previous, which
5490 might be the first part of a string whose (former) last char we repeated.
5491
5492 Possessifying an 'exact' quantifier has no effect, so we can ignore it. But
5493 an 'upto' may follow. We skip over an 'exact' item, and then test the
5494 length of what remains before proceeding. */
5495
5496 if (possessive_quantifier)
5497 {
5498 int len;
5499
5500 if (*tempcode == OP_TYPEEXACT)
5501 tempcode += PRIV(OP_lengths)[*tempcode] +
5502 ((tempcode[1 + IMM2_SIZE] == OP_PROP
5503 || tempcode[1 + IMM2_SIZE] == OP_NOTPROP)? 2 : 0);
5504
5505 else if (*tempcode == OP_EXACT || *tempcode == OP_NOTEXACT)
5506 {
5507 tempcode += PRIV(OP_lengths)[*tempcode];
5508 #ifdef SUPPORT_UTF
5509 if (utf && HAS_EXTRALEN(tempcode[-1]))
5510 tempcode += GET_EXTRALEN(tempcode[-1]);
5511 #endif
5512 }
5513
5514 len = (int)(code - tempcode);
5515 if (len > 0) switch (*tempcode)
5516 {
5517 case OP_STAR: *tempcode = OP_POSSTAR; break;
5518 case OP_PLUS: *tempcode = OP_POSPLUS; break;
5519 case OP_QUERY: *tempcode = OP_POSQUERY; break;
5520 case OP_UPTO: *tempcode = OP_POSUPTO; break;
5521
5522 case OP_STARI: *tempcode = OP_POSSTARI; break;
5523 case OP_PLUSI: *tempcode = OP_POSPLUSI; break;
5524 case OP_QUERYI: *tempcode = OP_POSQUERYI; break;
5525 case OP_UPTOI: *tempcode = OP_POSUPTOI; break;
5526
5527 case OP_NOTSTAR: *tempcode = OP_NOTPOSSTAR; break;
5528 case OP_NOTPLUS: *tempcode = OP_NOTPOSPLUS; break;
5529 case OP_NOTQUERY: *tempcode = OP_NOTPOSQUERY; break;
5530 case OP_NOTUPTO: *tempcode = OP_NOTPOSUPTO; break;
5531
5532 case OP_NOTSTARI: *tempcode = OP_NOTPOSSTARI; break;
5533 case OP_NOTPLUSI: *tempcode = OP_NOTPOSPLUSI; break;
5534 case OP_NOTQUERYI: *tempcode = OP_NOTPOSQUERYI; break;
5535 case OP_NOTUPTOI: *tempcode = OP_NOTPOSUPTOI; break;
5536
5537 case OP_TYPESTAR: *tempcode = OP_TYPEPOSSTAR; break;
5538 case OP_TYPEPLUS: *tempcode = OP_TYPEPOSPLUS; break;
5539 case OP_TYPEQUERY: *tempcode = OP_TYPEPOSQUERY; break;
5540 case OP_TYPEUPTO: *tempcode = OP_TYPEPOSUPTO; break;
5541
5542 /* Because we are moving code along, we must ensure that any
5543 pending recursive references are updated. */
5544
5545 default:
5546 *code = OP_END;
5547 adjust_recurse(tempcode, 1 + LINK_SIZE, utf, cd, save_hwm);
5548 memmove(tempcode + 1 + LINK_SIZE, tempcode, IN_UCHARS(len));
5549 code += 1 + LINK_SIZE;
5550 len += 1 + LINK_SIZE;
5551 tempcode[0] = OP_ONCE;
5552 *code++ = OP_KET;
5553 PUTINC(code, 0, len);
5554 PUT(tempcode, 1, len);
5555 break;
5556 }
5557 }
5558
5559 /* In all case we no longer have a previous item. We also set the
5560 "follows varying string" flag for subsequently encountered reqchars if
5561 it isn't already set and we have just passed a varying length item. */
5562
5563 END_REPEAT:
5564 previous = NULL;
5565 cd->req_varyopt |= reqvary;
5566 break;
5567
5568
5569 /* ===================================================================*/
5570 /* Start of nested parenthesized sub-expression, or comment or lookahead or
5571 lookbehind or option setting or condition or all the other extended
5572 parenthesis forms. */
5573
5574 case CHAR_LEFT_PARENTHESIS:
5575 newoptions = options;
5576 skipbytes = 0;
5577 bravalue = OP_CBRA;
5578 save_hwm = cd->hwm;
5579 reset_bracount = FALSE;
5580
5581 /* First deal with various "verbs" that can be introduced by '*'. */
5582
5583 ptr++;
5584 if (ptr[0] == CHAR_ASTERISK && (ptr[1] == ':'
5585 || (MAX_255(ptr[1]) && ((cd->ctypes[ptr[1]] & ctype_letter) != 0))))
5586 {
5587 int i, namelen;
5588 int arglen = 0;
5589 const char *vn = verbnames;
5590 const pcre_uchar *name = ptr + 1;
5591 const pcre_uchar *arg = NULL;
5592 previous = NULL;
5593 ptr++;
5594 while (MAX_255(*ptr) && (cd->ctypes[*ptr] & ctype_letter) != 0) ptr++;
5595 namelen = (int)(ptr - name);
5596
5597 /* It appears that Perl allows any characters whatsoever, other than
5598 a closing parenthesis, to appear in arguments, so we no longer insist on
5599 letters, digits, and underscores. */
5600
5601 if (*ptr == CHAR_COLON)
5602 {
5603 arg = ++ptr;
5604 while (*ptr != 0 && *ptr != CHAR_RIGHT_PARENTHESIS) ptr++;
5605 arglen = (int)(ptr - arg);
5606 if (arglen > (int)MAX_MARK)
5607 {
5608 *errorcodeptr = ERR75;
5609 goto FAILED;
5610 }
5611 }
5612
5613 if (*ptr != CHAR_RIGHT_PARENTHESIS)
5614 {
5615 *errorcodeptr = ERR60;
5616 goto FAILED;
5617 }
5618
5619 /* Scan the table of verb names */
5620
5621 for (i = 0; i < verbcount; i++)
5622 {
5623 if (namelen == verbs[i].len &&
5624 STRNCMP_UC_C8(name, vn, namelen) == 0)
5625 {
5626 /* Check for open captures before ACCEPT and convert it to
5627 ASSERT_ACCEPT if in an assertion. */
5628
5629 if (verbs[i].op == OP_ACCEPT)
5630 {
5631 open_capitem *oc;
5632 if (arglen != 0)
5633 {
5634 *errorcodeptr = ERR59;
5635 goto FAILED;
5636 }
5637 cd->had_accept = TRUE;
5638 for (oc = cd->open_caps; oc != NULL; oc = oc->next)
5639 {
5640 *code++ = OP_CLOSE;
5641 PUT2INC(code, 0, oc->number);
5642 }
5643 *code++ = (cd->assert_depth > 0)? OP_ASSERT_ACCEPT : OP_ACCEPT;
5644
5645 /* Do not set firstchar after *ACCEPT */
5646 if (firstchar == REQ_UNSET) firstchar = REQ_NONE;
5647 }
5648
5649 /* Handle other cases with/without an argument */
5650
5651 else if (arglen == 0)
5652 {
5653 if (verbs[i].op < 0) /* Argument is mandatory */
5654 {
5655 *errorcodeptr = ERR66;
5656 goto FAILED;
5657 }
5658 *code = verbs[i].op;
5659 if (*code++ == OP_THEN) cd->external_flags |= PCRE_HASTHEN;
5660 }
5661
5662 else
5663 {
5664 if (verbs[i].op_arg < 0) /* Argument is forbidden */
5665 {
5666 *errorcodeptr = ERR59;
5667 goto FAILED;
5668 }
5669 *code = verbs[i].op_arg;
5670 if (*code++ == OP_THEN_ARG) cd->external_flags |= PCRE_HASTHEN;
5671 *code++ = arglen;
5672 memcpy(code, arg, IN_UCHARS(arglen));
5673 code += arglen;
5674 *code++ = 0;
5675 }
5676
5677 break; /* Found verb, exit loop */
5678 }
5679
5680 vn += verbs[i].len + 1;
5681 }
5682
5683 if (i < verbcount) continue; /* Successfully handled a verb */
5684 *errorcodeptr = ERR60; /* Verb not recognized */
5685 goto FAILED;
5686 }
5687
5688 /* Deal with the extended parentheses; all are introduced by '?', and the
5689 appearance of any of them means that this is not a capturing group. */
5690
5691 else if (*ptr == CHAR_QUESTION_MARK)
5692 {
5693 int i, set, unset, namelen;
5694 int *optset;
5695 const pcre_uchar *name;
5696 pcre_uchar *slot;
5697
5698 switch (*(++ptr))
5699 {
5700 case CHAR_NUMBER_SIGN: /* Comment; skip to ket */
5701 ptr++;
5702 while (*ptr != 0 && *ptr != CHAR_RIGHT_PARENTHESIS) ptr++;
5703 if (*ptr == 0)
5704 {
5705 *errorcodeptr = ERR18;
5706 goto FAILED;
5707 }
5708 continue;
5709
5710
5711 /* ------------------------------------------------------------ */
5712 case CHAR_VERTICAL_LINE: /* Reset capture count for each branch */
5713 reset_bracount = TRUE;
5714 /* Fall through */
5715
5716 /* ------------------------------------------------------------ */
5717 case CHAR_COLON: /* Non-capturing bracket */
5718 bravalue = OP_BRA;
5719 ptr++;
5720 break;
5721
5722
5723 /* ------------------------------------------------------------ */
5724 case CHAR_LEFT_PARENTHESIS:
5725 bravalue = OP_COND; /* Conditional group */
5726
5727 /* A condition can be an assertion, a number (referring to a numbered
5728 group), a name (referring to a named group), or 'R', referring to
5729 recursion. R<digits> and R&name are also permitted for recursion tests.
5730
5731 There are several syntaxes for testing a named group: (?(name)) is used
5732 by Python; Perl 5.10 onwards uses (?(<name>) or (?('name')).
5733
5734 There are two unfortunate ambiguities, caused by history. (a) 'R' can
5735 be the recursive thing or the name 'R' (and similarly for 'R' followed
5736 by digits), and (b) a number could be a name that consists of digits.
5737 In both cases, we look for a name first; if not found, we try the other
5738 cases. */
5739
5740 /* For conditions that are assertions, check the syntax, and then exit
5741 the switch. This will take control down to where bracketed groups,
5742 including assertions, are processed. */
5743
5744 if (ptr[1] == CHAR_QUESTION_MARK && (ptr[2] == CHAR_EQUALS_SIGN ||
5745 ptr[2] == CHAR_EXCLAMATION_MARK || ptr[2] == CHAR_LESS_THAN_SIGN))
5746 break;
5747
5748 /* Most other conditions use OP_CREF (a couple change to OP_RREF
5749 below), and all need to skip 1+IMM2_SIZE bytes at the start of the group. */
5750
5751 code[1+LINK_SIZE] = OP_CREF;
5752 skipbytes = 1+IMM2_SIZE;
5753 refsign = -1;
5754
5755 /* Check for a test for recursion in a named group. */
5756
5757 if (ptr[1] == CHAR_R && ptr[2] == CHAR_AMPERSAND)
5758 {
5759 terminator = -1;
5760 ptr += 2;
5761 code[1+LINK_SIZE] = OP_RREF; /* Change the type of test */
5762 }
5763
5764 /* Check for a test for a named group's having been set, using the Perl
5765 syntax (?(<name>) or (?('name') */
5766
5767 else if (ptr[1] == CHAR_LESS_THAN_SIGN)
5768 {
5769 terminator = CHAR_GREATER_THAN_SIGN;
5770 ptr++;
5771 }
5772 else if (ptr[1] == CHAR_APOSTROPHE)
5773 {
5774 terminator = CHAR_APOSTROPHE;
5775 ptr++;
5776 }
5777 else
5778 {
5779 terminator = 0;
5780 if (ptr[1] == CHAR_MINUS || ptr[1] == CHAR_PLUS) refsign = *(++ptr);
5781 }
5782
5783 /* We now expect to read a name; any thing else is an error */
5784
5785 if (!MAX_255(ptr[1]) || (cd->ctypes[ptr[1]] & ctype_word) == 0)
5786 {
5787 ptr += 1; /* To get the right offset */
5788 *errorcodeptr = ERR28;
5789 goto FAILED;
5790 }
5791
5792 /* Read the name, but also get it as a number if it's all digits */
5793
5794 recno = 0;
5795 name = ++ptr;
5796 while (MAX_255(*ptr) && (cd->ctypes[*ptr] & ctype_word) != 0)
5797 {
5798 if (recno >= 0)
5799 recno = (IS_DIGIT(*ptr))? recno * 10 + *ptr - CHAR_0 : -1;
5800 ptr++;
5801 }
5802 namelen = (int)(ptr - name);
5803
5804 if ((terminator > 0 && *ptr++ != terminator) ||
5805 *ptr++ != CHAR_RIGHT_PARENTHESIS)
5806 {
5807 ptr--; /* Error offset */
5808 *errorcodeptr = ERR26;
5809 goto FAILED;
5810 }
5811
5812 /* Do no further checking in the pre-compile phase. */
5813
5814 if (lengthptr != NULL) break;
5815
5816 /* In the real compile we do the work of looking for the actual
5817 reference. If the string started with "+" or "-" we require the rest to
5818 be digits, in which case recno will be set. */
5819
5820 if (refsign > 0)
5821 {
5822 if (recno <= 0)
5823 {
5824 *errorcodeptr = ERR58;
5825 goto FAILED;
5826 }
5827 recno = (refsign == CHAR_MINUS)?
5828 cd->bracount - recno + 1 : recno +cd->bracount;
5829 if (recno <= 0 || recno > cd->final_bracount)
5830 {
5831 *errorcodeptr = ERR15;
5832 goto FAILED;
5833 }
5834 PUT2(code, 2+LINK_SIZE, recno);
5835 break;
5836 }
5837
5838 /* Otherwise (did not start with "+" or "-"), start by looking for the
5839 name. If we find a name, add one to the opcode to change OP_CREF or
5840 OP_RREF into OP_NCREF or OP_NRREF. These behave exactly the same,
5841 except they record that the reference was originally to a name. The
5842 information is used to check duplicate names. */
5843
5844 slot = cd->name_table;
5845 for (i = 0; i < cd->names_found; i++)
5846 {
5847 if (STRNCMP_UC_UC(name, slot+IMM2_SIZE, namelen) == 0) break;
5848 slot += cd->name_entry_size;
5849 }
5850
5851 /* Found a previous named subpattern */
5852
5853 if (i < cd->names_found)
5854 {
5855 recno = GET2(slot, 0);
5856 PUT2(code, 2+LINK_SIZE, recno);
5857 code[1+LINK_SIZE]++;
5858 }
5859
5860 /* Search the pattern for a forward reference */
5861
5862 else if ((i = find_parens(cd, name, namelen,
5863 (options & PCRE_EXTENDED) != 0, utf)) > 0)
5864 {
5865 PUT2(code, 2+LINK_SIZE, i);
5866 code[1+LINK_SIZE]++;
5867 }
5868
5869 /* If terminator == 0 it means that the name followed directly after
5870 the opening parenthesis [e.g. (?(abc)...] and in this case there are
5871 some further alternatives to try. For the cases where terminator != 0
5872 [things like (?(<name>... or (?('name')... or (?(R&name)... ] we have
5873 now checked all the possibilities, so give an error. */
5874
5875 else if (terminator != 0)
5876 {
5877 *errorcodeptr = ERR15;
5878 goto FAILED;
5879 }
5880
5881 /* Check for (?(R) for recursion. Allow digits after R to specify a
5882 specific group number. */
5883
5884 else if (*name == CHAR_R)
5885 {
5886 recno = 0;
5887 for (i = 1; i < namelen; i++)
5888 {
5889 if (!IS_DIGIT(name[i]))
5890 {
5891 *errorcodeptr = ERR15;
5892 goto FAILED;
5893 }
5894 recno = recno * 10 + name[i] - CHAR_0;
5895 }
5896 if (recno == 0) recno = RREF_ANY;
5897 code[1+LINK_SIZE] = OP_RREF; /* Change test type */
5898 PUT2(code, 2+LINK_SIZE, recno);
5899 }
5900
5901 /* Similarly, check for the (?(DEFINE) "condition", which is always
5902 false. */
5903
5904 else if (namelen == 6 && STRNCMP_UC_C8(name, STRING_DEFINE, 6) == 0)
5905 {
5906 code[1+LINK_SIZE] = OP_DEF;
5907 skipbytes = 1;
5908 }
5909
5910 /* Check for the "name" actually being a subpattern number. We are
5911 in the second pass here, so final_bracount is set. */
5912
5913 else if (recno > 0 && recno <= cd->final_bracount)
5914 {
5915 PUT2(code, 2+LINK_SIZE, recno);
5916 }
5917
5918 /* Either an unidentified subpattern, or a reference to (?(0) */
5919
5920 else
5921 {
5922 *errorcodeptr = (recno == 0)? ERR35: ERR15;
5923 goto FAILED;
5924 }
5925 break;
5926
5927
5928 /* ------------------------------------------------------------ */
5929 case CHAR_EQUALS_SIGN: /* Positive lookahead */
5930 bravalue = OP_ASSERT;
5931 cd->assert_depth += 1;
5932 ptr++;
5933 break;
5934
5935
5936 /* ------------------------------------------------------------ */
5937 case CHAR_EXCLAMATION_MARK: /* Negative lookahead */
5938 ptr++;
5939 if (*ptr == CHAR_RIGHT_PARENTHESIS) /* Optimize (?!) */
5940 {
5941 *code++ = OP_FAIL;
5942 previous = NULL;
5943 continue;
5944 }
5945 bravalue = OP_ASSERT_NOT;
5946 cd->assert_depth += 1;
5947 break;
5948
5949
5950 /* ------------------------------------------------------------ */
5951 case CHAR_LESS_THAN_SIGN: /* Lookbehind or named define */
5952 switch (ptr[1])
5953 {
5954 case CHAR_EQUALS_SIGN: /* Positive lookbehind */
5955 bravalue = OP_ASSERTBACK;
5956 cd->assert_depth += 1;
5957 ptr += 2;
5958 break;
5959
5960 case CHAR_EXCLAMATION_MARK: /* Negative lookbehind */
5961 bravalue = OP_ASSERTBACK_NOT;
5962 cd->assert_depth += 1;
5963 ptr += 2;
5964 break;
5965
5966 default: /* Could be name define, else bad */
5967 if (MAX_255(ptr[1]) && (cd->ctypes[ptr[1]] & ctype_word) != 0)
5968 goto DEFINE_NAME;
5969 ptr++; /* Correct offset for error */
5970 *errorcodeptr = ERR24;
5971 goto FAILED;
5972 }
5973 break;
5974
5975
5976 /* ------------------------------------------------------------ */
5977 case CHAR_GREATER_THAN_SIGN: /* One-time brackets */
5978 bravalue = OP_ONCE;
5979 ptr++;
5980 break;
5981
5982
5983 /* ------------------------------------------------------------ */
5984 case CHAR_C: /* Callout - may be followed by digits; */
5985 previous_callout = code; /* Save for later completion */
5986 after_manual_callout = 1; /* Skip one item before completing */
5987 *code++ = OP_CALLOUT;
5988 {
5989 int n = 0;
5990 ptr++;
5991 while(IS_DIGIT(*ptr))
5992 n = n * 10 + *ptr++ - CHAR_0;
5993 if (*ptr != CHAR_RIGHT_PARENTHESIS)
5994 {
5995 *errorcodeptr = ERR39;
5996 goto FAILED;
5997 }
5998 if (n > 255)
5999 {
6000 *errorcodeptr = ERR38;
6001 goto FAILED;
6002 }
6003 *code++ = n;
6004 PUT(code, 0, (int)(ptr - cd->start_pattern + 1)); /* Pattern offset */
6005 PUT(code, LINK_SIZE, 0); /* Default length */
6006 code += 2 * LINK_SIZE;
6007 }
6008 previous = NULL;
6009 continue;
6010
6011
6012 /* ------------------------------------------------------------ */
6013 case CHAR_P: /* Python-style named subpattern handling */
6014 if (*(++ptr) == CHAR_EQUALS_SIGN ||
6015 *ptr == CHAR_GREATER_THAN_SIGN) /* Reference or recursion */
6016 {
6017 is_recurse = *ptr == CHAR_GREATER_THAN_SIGN;
6018 terminator = CHAR_RIGHT_PARENTHESIS;
6019 goto NAMED_REF_OR_RECURSE;
6020 }
6021 else if (*ptr != CHAR_LESS_THAN_SIGN) /* Test for Python-style defn */
6022 {
6023 *errorcodeptr = ERR41;
6024 goto FAILED;
6025 }
6026 /* Fall through to handle (?P< as (?< is handled */
6027
6028
6029 /* ------------------------------------------------------------ */
6030 DEFINE_NAME: /* Come here from (?< handling */
6031 case CHAR_APOSTROPHE:
6032 {
6033 terminator = (*ptr == CHAR_LESS_THAN_SIGN)?
6034 CHAR_GREATER_THAN_SIGN : CHAR_APOSTROPHE;
6035 name = ++ptr;
6036
6037 while (MAX_255(*ptr) && (cd->ctypes[*ptr] & ctype_word) != 0) ptr++;
6038 namelen = (int)(ptr - name);
6039
6040 /* In the pre-compile phase, just do a syntax check. */
6041
6042 if (lengthptr != NULL)
6043 {
6044 if (*ptr != terminator)
6045 {
6046 *errorcodeptr = ERR42;
6047 goto FAILED;
6048 }
6049 if (cd->names_found >= MAX_NAME_COUNT)
6050 {
6051 *errorcodeptr = ERR49;
6052 goto FAILED;
6053 }
6054 if (namelen + IMM2_SIZE + 1 > cd->name_entry_size)
6055 {
6056 cd->name_entry_size = namelen + IMM2_SIZE + 1;
6057 if (namelen > MAX_NAME_SIZE)
6058 {
6059 *errorcodeptr = ERR48;
6060 goto FAILED;
6061 }
6062 }
6063 }
6064
6065 /* In the real compile, create the entry in the table, maintaining
6066 alphabetical order. Duplicate names for different numbers are
6067 permitted only if PCRE_DUPNAMES is set. Duplicate names for the same
6068 number are always OK. (An existing number can be re-used if (?|
6069 appears in the pattern.) In either event, a duplicate name results in
6070 a duplicate entry in the table, even if the number is the same. This
6071 is because the number of names, and hence the table size, is computed
6072 in the pre-compile, and it affects various numbers and pointers which
6073 would all have to be modified, and the compiled code moved down, if
6074 duplicates with the same number were omitted from the table. This
6075 doesn't seem worth the hassle. However, *different* names for the
6076 same number are not permitted. */
6077
6078 else
6079 {
6080 BOOL dupname = FALSE;
6081 slot = cd->name_table;
6082
6083 for (i = 0; i < cd->names_found; i++)
6084 {
6085 int crc = memcmp(name, slot+IMM2_SIZE, IN_UCHARS(namelen));
6086 if (crc == 0)
6087 {
6088 if (slot[IMM2_SIZE+namelen] == 0)
6089 {
6090 if (GET2(slot, 0) != cd->bracount + 1 &&
6091 (options & PCRE_DUPNAMES) == 0)
6092 {
6093 *errorcodeptr = ERR43;
6094 goto FAILED;
6095 }
6096 else dupname = TRUE;
6097 }
6098 else crc = -1; /* Current name is a substring */
6099 }
6100
6101 /* Make space in the table and break the loop for an earlier
6102 name. For a duplicate or later name, carry on. We do this for
6103 duplicates so that in the simple case (when ?(| is not used) they
6104 are in order of their numbers. */
6105
6106 if (crc < 0)
6107 {
6108 memmove(slot + cd->name_entry_size, slot,
6109 IN_UCHARS((cd->names_found - i) * cd->name_entry_size));
6110 break;
6111 }
6112
6113 /* Continue the loop for a later or duplicate name */
6114
6115 slot += cd->name_entry_size;
6116 }
6117
6118 /* For non-duplicate names, check for a duplicate number before
6119 adding the new name. */
6120
6121 if (!dupname)
6122 {
6123 pcre_uchar *cslot = cd->name_table;
6124 for (i = 0; i < cd->names_found; i++)
6125 {
6126 if (cslot != slot)
6127 {
6128 if (GET2(cslot, 0) == cd->bracount + 1)
6129 {
6130 *errorcodeptr = ERR65;
6131 goto FAILED;
6132 }
6133 }
6134 else i--;
6135 cslot += cd->name_entry_size;
6136 }
6137 }
6138
6139 PUT2(slot, 0, cd->bracount + 1);
6140 memcpy(slot + IMM2_SIZE, name, IN_UCHARS(namelen));
6141 slot[IMM2_SIZE + namelen] = 0;
6142 }
6143 }
6144
6145 /* In both pre-compile and compile, count the number of names we've
6146 encountered. */
6147
6148 cd->names_found++;
6149 ptr++; /* Move past > or ' */
6150 goto NUMBERED_GROUP;
6151
6152
6153 /* ------------------------------------------------------------ */
6154 case CHAR_AMPERSAND: /* Perl recursion/subroutine syntax */
6155 terminator = CHAR_RIGHT_PARENTHESIS;
6156 is_recurse = TRUE;
6157 /* Fall through */
6158
6159 /* We come here from the Python syntax above that handles both
6160 references (?P=name) and recursion (?P>name), as well as falling
6161 through from the Perl recursion syntax (?&name). We also come here from
6162 the Perl \k<name> or \k'name' back reference syntax and the \k{name}
6163 .NET syntax, and the Oniguruma \g<...> and \g'...' subroutine syntax. */
6164
6165 NAMED_REF_OR_RECURSE:
6166 name = ++ptr;
6167 while (MAX_255(*ptr) && (cd->ctypes[*ptr] & ctype_word) != 0) ptr++;
6168 namelen = (int)(ptr - name);
6169
6170 /* In the pre-compile phase, do a syntax check. We used to just set
6171 a dummy reference number, because it was not used in the first pass.
6172 However, with the change of recursive back references to be atomic,
6173 we have to look for the number so that this state can be identified, as
6174 otherwise the incorrect length is computed. If it's not a backwards
6175 reference, the dummy number will do. */
6176
6177 if (lengthptr != NULL)
6178 {
6179 const pcre_uchar *temp;
6180
6181 if (namelen == 0)
6182 {
6183 *errorcodeptr = ERR62;
6184 goto FAILED;
6185 }
6186 if (*ptr != terminator)
6187 {
6188 *errorcodeptr = ERR42;
6189 goto FAILED;
6190 }
6191 if (namelen > MAX_NAME_SIZE)
6192 {
6193 *errorcodeptr = ERR48;
6194 goto FAILED;
6195 }
6196
6197 /* The name table does not exist in the first pass, so we cannot
6198 do a simple search as in the code below. Instead, we have to scan the
6199 pattern to find the number. It is important that we scan it only as
6200 far as we have got because the syntax of named subpatterns has not
6201 been checked for the rest of the pattern, and find_parens() assumes
6202 correct syntax. In any case, it's a waste of resources to scan
6203 further. We stop the scan at the current point by temporarily
6204 adjusting the value of cd->endpattern. */
6205
6206 temp = cd->end_pattern;
6207 cd->end_pattern = ptr;
6208 recno = find_parens(cd, name, namelen,
6209 (options & PCRE_EXTENDED) != 0, utf);
6210 cd->end_pattern = temp;
6211 if (recno < 0) recno = 0; /* Forward ref; set dummy number */
6212 }
6213
6214 /* In the real compile, seek the name in the table. We check the name
6215 first, and then check that we have reached the end of the name in the
6216 table. That way, if the name that is longer than any in the table,
6217 the comparison will fail without reading beyond the table entry. */
6218
6219 else
6220 {
6221 slot = cd->name_table;
6222 for (i = 0; i < cd->names_found; i++)
6223 {
6224 if (STRNCMP_UC_UC(name, slot+IMM2_SIZE, namelen) == 0 &&
6225 slot[IMM2_SIZE+namelen] == 0)
6226 break;
6227 slot += cd->name_entry_size;
6228 }
6229
6230 if (i < cd->names_found) /* Back reference */
6231 {
6232 recno = GET2(slot, 0);
6233 }
6234 else if ((recno = /* Forward back reference */
6235 find_parens(cd, name, namelen,
6236 (options & PCRE_EXTENDED) != 0, utf)) <= 0)
6237 {
6238 *errorcodeptr = ERR15;
6239 goto FAILED;
6240 }
6241 }
6242
6243 /* In both phases, we can now go to the code than handles numerical
6244 recursion or backreferences. */
6245
6246 if (is_recurse) goto HANDLE_RECURSION;
6247 else goto HANDLE_REFERENCE;
6248
6249
6250 /* ------------------------------------------------------------ */
6251 case CHAR_R: /* Recursion */
6252 ptr++; /* Same as (?0) */
6253 /* Fall through */
6254
6255
6256 /* ------------------------------------------------------------ */
6257 case CHAR_MINUS: case CHAR_PLUS: /* Recursion or subroutine */
6258 case CHAR_0: case CHAR_1: case CHAR_2: case CHAR_3: case CHAR_4:
6259 case CHAR_5: case CHAR_6: case CHAR_7: case CHAR_8: case CHAR_9:
6260 {
6261 const pcre_uchar *called;
6262 terminator = CHAR_RIGHT_PARENTHESIS;
6263
6264 /* Come here from the \g<...> and \g'...' code (Oniguruma
6265 compatibility). However, the syntax has been checked to ensure that
6266 the ... are a (signed) number, so that neither ERR63 nor ERR29 will
6267 be called on this path, nor with the jump to OTHER_CHAR_AFTER_QUERY
6268 ever be taken. */
6269
6270 HANDLE_NUMERICAL_RECURSION:
6271
6272 if ((refsign = *ptr) == CHAR_PLUS)
6273 {
6274 ptr++;
6275 if (!IS_DIGIT(*ptr))
6276 {
6277 *errorcodeptr = ERR63;
6278 goto FAILED;
6279 }
6280 }
6281 else if (refsign == CHAR_MINUS)
6282 {
6283 if (!IS_DIGIT(ptr[1]))
6284 goto OTHER_CHAR_AFTER_QUERY;
6285 ptr++;
6286 }
6287
6288 recno = 0;
6289 while(IS_DIGIT(*ptr))
6290 recno = recno * 10 + *ptr++ - CHAR_0;
6291
6292 if (*ptr != terminator)
6293 {
6294 *errorcodeptr = ERR29;
6295 goto FAILED;
6296 }
6297
6298 if (refsign == CHAR_MINUS)
6299 {
6300 if (recno == 0)
6301 {
6302 *errorcodeptr = ERR58;
6303 goto FAILED;
6304 }
6305 recno = cd->bracount - recno + 1;
6306 if (recno <= 0)
6307 {
6308 *errorcodeptr = ERR15;
6309 goto FAILED;
6310 }
6311 }
6312 else if (refsign == CHAR_PLUS)
6313 {
6314 if (recno == 0)
6315 {
6316 *errorcodeptr = ERR58;
6317 goto FAILED;
6318 }
6319 recno += cd->bracount;
6320 }
6321
6322 /* Come here from code above that handles a named recursion */
6323
6324 HANDLE_RECURSION:
6325
6326 previous = code;
6327 called = cd->start_code;
6328
6329 /* When we are actually compiling, find the bracket that is being
6330 referenced. Temporarily end the regex in case it doesn't exist before
6331 this point. If we end up with a forward reference, first check that
6332 the bracket does occur later so we can give the error (and position)
6333 now. Then remember this forward reference in the workspace so it can
6334 be filled in at the end. */
6335
6336 if (lengthptr == NULL)
6337 {
6338 *code = OP_END;
6339 if (recno != 0)
6340 called = PRIV(find_bracket)(cd->start_code, utf, recno);
6341
6342 /* Forward reference */
6343
6344 if (called == NULL)
6345 {
6346 if (find_parens(cd, NULL, recno,
6347 (options & PCRE_EXTENDED) != 0, utf) < 0)
6348 {
6349 *errorcodeptr = ERR15;
6350 goto FAILED;
6351 }
6352
6353 /* Fudge the value of "called" so that when it is inserted as an
6354 offset below, what it actually inserted is the reference number
6355 of the group. Then remember the forward reference. */
6356
6357 called = cd->start_code + recno;
6358 if (cd->hwm >= cd->start_workspace + cd->workspace_size -
6359 WORK_SIZE_SAFETY_MARGIN)
6360 {
6361 *errorcodeptr = expand_workspace(cd);
6362 if (*errorcodeptr != 0) goto FAILED;
6363 }
6364 PUTINC(cd->hwm, 0, (int)(code + 1 - cd->start_code));
6365 }
6366
6367 /* If not a forward reference, and the subpattern is still open,
6368 this is a recursive call. We check to see if this is a left
6369 recursion that could loop for ever, and diagnose that case. We
6370 must not, however, do this check if we are in a conditional
6371 subpattern because the condition might be testing for recursion in
6372 a pattern such as /(?(R)a+|(?R)b)/, which is perfectly valid.
6373 Forever loops are also detected at runtime, so those that occur in
6374 conditional subpatterns will be picked up then. */
6375
6376 else if (GET(called, 1) == 0 && cond_depth <= 0 &&
6377 could_be_empty(called, code, bcptr, utf, cd))
6378 {
6379 *errorcodeptr = ERR40;
6380 goto FAILED;
6381 }
6382 }
6383
6384 /* Insert the recursion/subroutine item. It does not have a set first
6385 character (relevant if it is repeated, because it will then be
6386 wrapped with ONCE brackets). */
6387
6388 *code = OP_RECURSE;
6389 PUT(code, 1, (int)(called - cd->start_code));
6390 code += 1 + LINK_SIZE;
6391 groupsetfirstchar = FALSE;
6392 }
6393
6394 /* Can't determine a first byte now */
6395
6396 if (firstchar == REQ_UNSET) firstchar = REQ_NONE;
6397 continue;
6398
6399
6400 /* ------------------------------------------------------------ */
6401 default: /* Other characters: check option setting */
6402 OTHER_CHAR_AFTER_QUERY:
6403 set = unset = 0;
6404 optset = &set;
6405
6406 while (*ptr != CHAR_RIGHT_PARENTHESIS && *ptr != CHAR_COLON)
6407 {
6408 switch (*ptr++)
6409 {
6410 case CHAR_MINUS: optset = &unset; break;
6411
6412 case CHAR_J: /* Record that it changed in the external options */
6413 *optset |= PCRE_DUPNAMES;
6414 cd->external_flags |= PCRE_JCHANGED;
6415 break;
6416
6417 case CHAR_i: *optset |= PCRE_CASELESS; break;
6418 case CHAR_m: *optset |= PCRE_MULTILINE; break;
6419 case CHAR_s: *optset |= PCRE_DOTALL; break;
6420 case CHAR_x: *optset |= PCRE_EXTENDED; break;
6421 case CHAR_U: *optset |= PCRE_UNGREEDY; break;
6422 case CHAR_X: *optset |= PCRE_EXTRA; break;
6423
6424 default: *errorcodeptr = ERR12;
6425 ptr--; /* Correct the offset */
6426 goto FAILED;
6427 }
6428 }
6429
6430 /* Set up the changed option bits, but don't change anything yet. */
6431
6432 newoptions = (options | set) & (~unset);
6433
6434 /* If the options ended with ')' this is not the start of a nested
6435 group with option changes, so the options change at this level. If this
6436 item is right at the start of the pattern, the options can be
6437 abstracted and made external in the pre-compile phase, and ignored in
6438 the compile phase. This can be helpful when matching -- for instance in
6439 caseless checking of required bytes.
6440
6441 If the code pointer is not (cd->start_code + 1 + LINK_SIZE), we are
6442 definitely *not* at the start of the pattern because something has been
6443 compiled. In the pre-compile phase, however, the code pointer can have
6444 that value after the start, because it gets reset as code is discarded
6445 during the pre-compile. However, this can happen only at top level - if
6446 we are within parentheses, the starting BRA will still be present. At
6447 any parenthesis level, the length value can be used to test if anything
6448 has been compiled at that level. Thus, a test for both these conditions
6449 is necessary to ensure we correctly detect the start of the pattern in
6450 both phases.
6451
6452 If we are not at the pattern start, reset the greedy defaults and the
6453 case value for firstchar and reqchar. */
6454
6455 if (*ptr == CHAR_RIGHT_PARENTHESIS)
6456 {
6457 if (code == cd->start_code + 1 + LINK_SIZE &&
6458 (lengthptr == NULL || *lengthptr == 2 + 2*LINK_SIZE))
6459 {
6460 cd->external_options = newoptions;
6461 }
6462 else
6463 {
6464 greedy_default = ((newoptions & PCRE_UNGREEDY) != 0);
6465 greedy_non_default = greedy_default ^ 1;
6466 req_caseopt = ((newoptions & PCRE_CASELESS) != 0)? REQ_CASELESS:0;
6467 }
6468
6469 /* Change options at this level, and pass them back for use
6470 in subsequent branches. */
6471
6472 *optionsptr = options = newoptions;
6473 previous = NULL; /* This item can't be repeated */
6474 continue; /* It is complete */
6475 }
6476
6477 /* If the options ended with ':' we are heading into a nested group
6478 with possible change of options. Such groups are non-capturing and are
6479 not assertions of any kind. All we need to do is skip over the ':';
6480 the newoptions value is handled below. */
6481
6482 bravalue = OP_BRA;
6483 ptr++;
6484 } /* End of switch for character following (? */
6485 } /* End of (? handling */
6486
6487 /* Opening parenthesis not followed by '*' or '?'. If PCRE_NO_AUTO_CAPTURE
6488 is set, all unadorned brackets become non-capturing and behave like (?:...)
6489 brackets. */
6490
6491 else if ((options & PCRE_NO_AUTO_CAPTURE) != 0)
6492 {
6493 bravalue = OP_BRA;
6494 }
6495
6496 /* Else we have a capturing group. */
6497
6498 else
6499 {
6500 NUMBERED_GROUP:
6501 cd->bracount += 1;
6502 PUT2(code, 1+LINK_SIZE, cd->bracount);
6503 skipbytes = IMM2_SIZE;
6504 }
6505
6506 /* Process nested bracketed regex. Assertions used not to be repeatable,
6507 but this was changed for Perl compatibility, so all kinds can now be
6508 repeated. We copy code into a non-register variable (tempcode) in order to
6509 be able to pass its address because some compilers complain otherwise. */
6510
6511 previous = code; /* For handling repetition */
6512 *code = bravalue;
6513 tempcode = code;
6514 tempreqvary = cd->req_varyopt; /* Save value before bracket */
6515 tempbracount = cd->bracount; /* Save value before bracket */
6516 length_prevgroup = 0; /* Initialize for pre-compile phase */
6517
6518 if (!compile_regex(
6519 newoptions, /* The complete new option state */
6520 &tempcode, /* Where to put code (updated) */
6521 &ptr, /* Input pointer (updated) */
6522 errorcodeptr, /* Where to put an error message */
6523 (bravalue == OP_ASSERTBACK ||
6524 bravalue == OP_ASSERTBACK_NOT), /* TRUE if back assert */
6525 reset_bracount, /* True if (?| group */
6526 skipbytes, /* Skip over bracket number */
6527 cond_depth +
6528 ((bravalue == OP_COND)?1:0), /* Depth of condition subpatterns */
6529 &subfirstchar, /* For possible first char */
6530 &subreqchar, /* For possible last char */
6531 bcptr, /* Current branch chain */
6532 cd, /* Tables block */
6533 (lengthptr == NULL)? NULL : /* Actual compile phase */
6534 &length_prevgroup /* Pre-compile phase */
6535 ))
6536 goto FAILED;
6537
6538 /* If this was an atomic group and there are no capturing groups within it,
6539 generate OP_ONCE_NC instead of OP_ONCE. */
6540
6541 if (bravalue == OP_ONCE && cd->bracount <= tempbracount)
6542 *code = OP_ONCE_NC;
6543
6544 if (bravalue >= OP_ASSERT && bravalue <= OP_ASSERTBACK_NOT)
6545 cd->assert_depth -= 1;
6546
6547 /* At the end of compiling, code is still pointing to the start of the
6548 group, while tempcode has been updated to point past the end of the group.
6549 The pattern pointer (ptr) is on the bracket.
6550
6551 If this is a conditional bracket, check that there are no more than
6552 two branches in the group, or just one if it's a DEFINE group. We do this
6553 in the real compile phase, not in the pre-pass, where the whole group may
6554 not be available. */
6555
6556 if (bravalue == OP_COND && lengthptr == NULL)
6557 {
6558 pcre_uchar *tc = code;
6559 int condcount = 0;
6560
6561 do {
6562 condcount++;
6563 tc += GET(tc,1);
6564 }
6565 while (*tc != OP_KET);
6566
6567 /* A DEFINE group is never obeyed inline (the "condition" is always
6568 false). It must have only one branch. */
6569
6570 if (code[LINK_SIZE+1] == OP_DEF)
6571 {
6572 if (condcount > 1)
6573 {
6574 *errorcodeptr = ERR54;
6575 goto FAILED;
6576 }
6577 bravalue = OP_DEF; /* Just a flag to suppress char handling below */
6578 }
6579
6580 /* A "normal" conditional group. If there is just one branch, we must not
6581 make use of its firstchar or reqchar, because this is equivalent to an
6582 empty second branch. */
6583
6584 else
6585 {
6586 if (condcount > 2)
6587 {
6588 *errorcodeptr = ERR27;
6589 goto FAILED;
6590 }
6591 if (condcount == 1) subfirstchar = subreqchar = REQ_NONE;
6592 }
6593 }
6594
6595 /* Error if hit end of pattern */
6596
6597 if (*ptr != CHAR_RIGHT_PARENTHESIS)
6598 {
6599 *errorcodeptr = ERR14;
6600 goto FAILED;
6601 }
6602
6603 /* In the pre-compile phase, update the length by the length of the group,
6604 less the brackets at either end. Then reduce the compiled code to just a
6605 set of non-capturing brackets so that it doesn't use much memory if it is
6606 duplicated by a quantifier.*/
6607
6608 if (lengthptr != NULL)
6609 {
6610 if (OFLOW_MAX - *lengthptr < length_prevgroup - 2 - 2*LINK_SIZE)
6611 {
6612 *errorcodeptr = ERR20;
6613 goto FAILED;
6614 }
6615 *lengthptr += length_prevgroup - 2 - 2*LINK_SIZE;
6616 code++; /* This already contains bravalue */
6617 PUTINC(code, 0, 1 + LINK_SIZE);
6618 *code++ = OP_KET;
6619 PUTINC(code, 0, 1 + LINK_SIZE);
6620 break; /* No need to waste time with special character handling */
6621 }
6622
6623 /* Otherwise update the main code pointer to the end of the group. */
6624
6625 code = tempcode;
6626
6627 /* For a DEFINE group, required and first character settings are not
6628 relevant. */
6629
6630 if (bravalue == OP_DEF) break;
6631
6632 /* Handle updating of the required and first characters for other types of
6633 group. Update for normal brackets of all kinds, and conditions with two
6634 branches (see code above). If the bracket is followed by a quantifier with
6635 zero repeat, we have to back off. Hence the definition of zeroreqchar and
6636 zerofirstchar outside the main loop so that they can be accessed for the
6637 back off. */
6638