/[pcre]/code/trunk/pcre_compile.c
ViewVC logotype

Contents of /code/trunk/pcre_compile.c

Parent Directory Parent Directory | Revision Log Revision Log


Revision 904 - (show annotations)
Mon Jan 23 17:30:49 2012 UTC (7 years, 7 months ago) by ph10
File MIME type: text/plain
File size: 269088 byte(s)
Error occurred while calculating annotation data.
Additional casts to avoid compiler warnings, originally from a MS compiler, but 
also given by gcc if you turn on enough warnings.
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 ;
493
494 /* Table to identify digits and hex digits. This is used when compiling
495 patterns. Note that the tables in chartables are dependent on the locale, and
496 may mark arbitrary characters as digits - but the PCRE compiling code expects
497 to handle only 0-9, a-z, and A-Z as digits when compiling. That is why we have
498 a private table here. It costs 256 bytes, but it is a lot faster than doing
499 character value tests (at least in some simple cases I timed), and in some
500 applications one wants PCRE to compile efficiently as well as match
501 efficiently.
502
503 For convenience, we use the same bit definitions as in chartables:
504
505 0x04 decimal digit
506 0x08 hexadecimal digit
507
508 Then we can use ctype_digit and ctype_xdigit in the code. */
509
510 /* Using a simple comparison for decimal numbers rather than a memory read
511 is much faster, and the resulting code is simpler (the compiler turns it
512 into a subtraction and unsigned comparison). */
513
514 #define IS_DIGIT(x) ((x) >= CHAR_0 && (x) <= CHAR_9)
515
516 #ifndef EBCDIC
517
518 /* This is the "normal" case, for ASCII systems, and EBCDIC systems running in
519 UTF-8 mode. */
520
521 static const pcre_uint8 digitab[] =
522 {
523 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 0- 7 */
524 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 8- 15 */
525 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 16- 23 */
526 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 24- 31 */
527 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - ' */
528 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* ( - / */
529 0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c, /* 0 - 7 */
530 0x0c,0x0c,0x00,0x00,0x00,0x00,0x00,0x00, /* 8 - ? */
531 0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* @ - G */
532 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* H - O */
533 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* P - W */
534 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* X - _ */
535 0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* ` - g */
536 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* h - o */
537 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* p - w */
538 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* x -127 */
539 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 128-135 */
540 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 136-143 */
541 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 144-151 */
542 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 152-159 */
543 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 160-167 */
544 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 168-175 */
545 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 176-183 */
546 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 184-191 */
547 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 192-199 */
548 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 200-207 */
549 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 208-215 */
550 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 216-223 */
551 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 224-231 */
552 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 232-239 */
553 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 240-247 */
554 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00};/* 248-255 */
555
556 #else
557
558 /* This is the "abnormal" case, for EBCDIC systems not running in UTF-8 mode. */
559
560 static const pcre_uint8 digitab[] =
561 {
562 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 0- 7 0 */
563 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 8- 15 */
564 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 16- 23 10 */
565 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 24- 31 */
566 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 32- 39 20 */
567 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 40- 47 */
568 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 48- 55 30 */
569 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 56- 63 */
570 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - 71 40 */
571 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 72- | */
572 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* & - 87 50 */
573 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 88- 95 */
574 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - -103 60 */
575 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 104- ? */
576 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 112-119 70 */
577 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 120- " */
578 0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* 128- g 80 */
579 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* h -143 */
580 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 144- p 90 */
581 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* q -159 */
582 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 160- x A0 */
583 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* y -175 */
584 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* ^ -183 B0 */
585 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 184-191 */
586 0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* { - G C0 */
587 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* H -207 */
588 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* } - P D0 */
589 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* Q -223 */
590 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* \ - X E0 */
591 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* Y -239 */
592 0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c, /* 0 - 7 F0 */
593 0x0c,0x0c,0x00,0x00,0x00,0x00,0x00,0x00};/* 8 -255 */
594
595 static const pcre_uint8 ebcdic_chartab[] = { /* chartable partial dup */
596 0x80,0x00,0x00,0x00,0x00,0x01,0x00,0x00, /* 0- 7 */
597 0x00,0x00,0x00,0x00,0x01,0x01,0x00,0x00, /* 8- 15 */
598 0x00,0x00,0x00,0x00,0x00,0x01,0x00,0x00, /* 16- 23 */
599 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 24- 31 */
600 0x00,0x00,0x00,0x00,0x00,0x01,0x00,0x00, /* 32- 39 */
601 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 40- 47 */
602 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 48- 55 */
603 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 56- 63 */
604 0x01,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - 71 */
605 0x00,0x00,0x00,0x80,0x00,0x80,0x80,0x80, /* 72- | */
606 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* & - 87 */
607 0x00,0x00,0x00,0x80,0x80,0x80,0x00,0x00, /* 88- 95 */
608 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - -103 */
609 0x00,0x00,0x00,0x00,0x00,0x10,0x00,0x80, /* 104- ? */
610 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 112-119 */
611 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 120- " */
612 0x00,0x1a,0x1a,0x1a,0x1a,0x1a,0x1a,0x12, /* 128- g */
613 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* h -143 */
614 0x00,0x12,0x12,0x12,0x12,0x12,0x12,0x12, /* 144- p */
615 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* q -159 */
616 0x00,0x00,0x12,0x12,0x12,0x12,0x12,0x12, /* 160- x */
617 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* y -175 */
618 0x80,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* ^ -183 */
619 0x00,0x00,0x80,0x00,0x00,0x00,0x00,0x00, /* 184-191 */
620 0x80,0x1a,0x1a,0x1a,0x1a,0x1a,0x1a,0x12, /* { - G */
621 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* H -207 */
622 0x00,0x12,0x12,0x12,0x12,0x12,0x12,0x12, /* } - P */
623 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* Q -223 */
624 0x00,0x00,0x12,0x12,0x12,0x12,0x12,0x12, /* \ - X */
625 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* Y -239 */
626 0x1c,0x1c,0x1c,0x1c,0x1c,0x1c,0x1c,0x1c, /* 0 - 7 */
627 0x1c,0x1c,0x00,0x00,0x00,0x00,0x00,0x00};/* 8 -255 */
628 #endif
629
630
631 /* Definition to allow mutual recursion */
632
633 static BOOL
634 compile_regex(int, pcre_uchar **, const pcre_uchar **, int *, BOOL, BOOL, int, int,
635 int *, int *, branch_chain *, compile_data *, int *);
636
637
638
639 /*************************************************
640 * Find an error text *
641 *************************************************/
642
643 /* The error texts are now all in one long string, to save on relocations. As
644 some of the text is of unknown length, we can't use a table of offsets.
645 Instead, just count through the strings. This is not a performance issue
646 because it happens only when there has been a compilation error.
647
648 Argument: the error number
649 Returns: pointer to the error string
650 */
651
652 static const char *
653 find_error_text(int n)
654 {
655 const char *s = error_texts;
656 for (; n > 0; n--)
657 {
658 while (*s++ != 0) {};
659 if (*s == 0) return "Error text not found (please report)";
660 }
661 return s;
662 }
663
664
665 /*************************************************
666 * Expand the workspace *
667 *************************************************/
668
669 /* This function is called during the second compiling phase, if the number of
670 forward references fills the existing workspace, which is originally a block on
671 the stack. A larger block is obtained from malloc() unless the ultimate limit
672 has been reached or the increase will be rather small.
673
674 Argument: pointer to the compile data block
675 Returns: 0 if all went well, else an error number
676 */
677
678 static int
679 expand_workspace(compile_data *cd)
680 {
681 pcre_uchar *newspace;
682 int newsize = cd->workspace_size * 2;
683
684 if (newsize > COMPILE_WORK_SIZE_MAX) newsize = COMPILE_WORK_SIZE_MAX;
685 if (cd->workspace_size >= COMPILE_WORK_SIZE_MAX ||
686 newsize - cd->workspace_size < WORK_SIZE_SAFETY_MARGIN)
687 return ERR72;
688
689 newspace = (PUBL(malloc))(IN_UCHARS(newsize));
690 if (newspace == NULL) return ERR21;
691 memcpy(newspace, cd->start_workspace, cd->workspace_size * sizeof(pcre_uchar));
692 cd->hwm = (pcre_uchar *)newspace + (cd->hwm - cd->start_workspace);
693 if (cd->workspace_size > COMPILE_WORK_SIZE)
694 (PUBL(free))((void *)cd->start_workspace);
695 cd->start_workspace = newspace;
696 cd->workspace_size = newsize;
697 return 0;
698 }
699
700
701
702 /*************************************************
703 * Check for counted repeat *
704 *************************************************/
705
706 /* This function is called when a '{' is encountered in a place where it might
707 start a quantifier. It looks ahead to see if it really is a quantifier or not.
708 It is only a quantifier if it is one of the forms {ddd} {ddd,} or {ddd,ddd}
709 where the ddds are digits.
710
711 Arguments:
712 p pointer to the first char after '{'
713
714 Returns: TRUE or FALSE
715 */
716
717 static BOOL
718 is_counted_repeat(const pcre_uchar *p)
719 {
720 if (!IS_DIGIT(*p)) return FALSE;
721 p++;
722 while (IS_DIGIT(*p)) p++;
723 if (*p == CHAR_RIGHT_CURLY_BRACKET) return TRUE;
724
725 if (*p++ != CHAR_COMMA) return FALSE;
726 if (*p == CHAR_RIGHT_CURLY_BRACKET) return TRUE;
727
728 if (!IS_DIGIT(*p)) return FALSE;
729 p++;
730 while (IS_DIGIT(*p)) p++;
731
732 return (*p == CHAR_RIGHT_CURLY_BRACKET);
733 }
734
735
736
737 /*************************************************
738 * Handle escapes *
739 *************************************************/
740
741 /* This function is called when a \ has been encountered. It either returns a
742 positive value for a simple escape such as \n, or a negative value which
743 encodes one of the more complicated things such as \d. A backreference to group
744 n is returned as -(ESC_REF + n); ESC_REF is the highest ESC_xxx macro. When
745 UTF-8 is enabled, a positive value greater than 255 may be returned. On entry,
746 ptr is pointing at the \. On exit, it is on the final character of the escape
747 sequence.
748
749 Arguments:
750 ptrptr points to the pattern position pointer
751 errorcodeptr points to the errorcode variable
752 bracount number of previous extracting brackets
753 options the options bits
754 isclass TRUE if inside a character class
755
756 Returns: zero or positive => a data character
757 negative => a special escape sequence
758 on error, errorcodeptr is set
759 */
760
761 static int
762 check_escape(const pcre_uchar **ptrptr, int *errorcodeptr, int bracount,
763 int options, BOOL isclass)
764 {
765 /* PCRE_UTF16 has the same value as PCRE_UTF8. */
766 BOOL utf = (options & PCRE_UTF8) != 0;
767 const pcre_uchar *ptr = *ptrptr + 1;
768 pcre_int32 c;
769 int i;
770
771 GETCHARINCTEST(c, ptr); /* Get character value, increment pointer */
772 ptr--; /* Set pointer back to the last byte */
773
774 /* If backslash is at the end of the pattern, it's an error. */
775
776 if (c == 0) *errorcodeptr = ERR1;
777
778 /* Non-alphanumerics are literals. For digits or letters, do an initial lookup
779 in a table. A non-zero result is something that can be returned immediately.
780 Otherwise further processing may be required. */
781
782 #ifndef EBCDIC /* ASCII/UTF-8 coding */
783 /* Not alphanumeric */
784 else if (c < CHAR_0 || c > CHAR_z) {}
785 else if ((i = escapes[c - CHAR_0]) != 0) c = i;
786
787 #else /* EBCDIC coding */
788 /* Not alphanumeric */
789 else if (c < 'a' || (!MAX_255(c) || (ebcdic_chartab[c] & 0x0E) == 0)) {}
790 else if ((i = escapes[c - 0x48]) != 0) c = i;
791 #endif
792
793 /* Escapes that need further processing, or are illegal. */
794
795 else
796 {
797 const pcre_uchar *oldptr;
798 BOOL braced, negated;
799
800 switch (c)
801 {
802 /* A number of Perl escapes are not handled by PCRE. We give an explicit
803 error. */
804
805 case CHAR_l:
806 case CHAR_L:
807 *errorcodeptr = ERR37;
808 break;
809
810 case CHAR_u:
811 if ((options & PCRE_JAVASCRIPT_COMPAT) != 0)
812 {
813 /* In JavaScript, \u must be followed by four hexadecimal numbers.
814 Otherwise it is a lowercase u letter. */
815 if (MAX_255(ptr[1]) && (digitab[ptr[1]] & ctype_xdigit) != 0
816 && MAX_255(ptr[2]) && (digitab[ptr[2]] & ctype_xdigit) != 0
817 && MAX_255(ptr[3]) && (digitab[ptr[3]] & ctype_xdigit) != 0
818 && MAX_255(ptr[4]) && (digitab[ptr[4]] & ctype_xdigit) != 0)
819 {
820 c = 0;
821 for (i = 0; i < 4; ++i)
822 {
823 register int cc = *(++ptr);
824 #ifndef EBCDIC /* ASCII/UTF-8 coding */
825 if (cc >= CHAR_a) cc -= 32; /* Convert to upper case */
826 c = (c << 4) + cc - ((cc < CHAR_A)? CHAR_0 : (CHAR_A - 10));
827 #else /* EBCDIC coding */
828 if (cc >= CHAR_a && cc <= CHAR_z) cc += 64; /* Convert to upper case */
829 c = (c << 4) + cc - ((cc >= CHAR_0)? CHAR_0 : (CHAR_A - 10));
830 #endif
831 }
832 }
833 }
834 else
835 *errorcodeptr = ERR37;
836 break;
837
838 case CHAR_U:
839 /* In JavaScript, \U is an uppercase U letter. */
840 if ((options & PCRE_JAVASCRIPT_COMPAT) == 0) *errorcodeptr = ERR37;
841 break;
842
843 /* In a character class, \g is just a literal "g". Outside a character
844 class, \g must be followed by one of a number of specific things:
845
846 (1) A number, either plain or braced. If positive, it is an absolute
847 backreference. If negative, it is a relative backreference. This is a Perl
848 5.10 feature.
849
850 (2) Perl 5.10 also supports \g{name} as a reference to a named group. This
851 is part of Perl's movement towards a unified syntax for back references. As
852 this is synonymous with \k{name}, we fudge it up by pretending it really
853 was \k.
854
855 (3) For Oniguruma compatibility we also support \g followed by a name or a
856 number either in angle brackets or in single quotes. However, these are
857 (possibly recursive) subroutine calls, _not_ backreferences. Just return
858 the -ESC_g code (cf \k). */
859
860 case CHAR_g:
861 if (isclass) break;
862 if (ptr[1] == CHAR_LESS_THAN_SIGN || ptr[1] == CHAR_APOSTROPHE)
863 {
864 c = -ESC_g;
865 break;
866 }
867
868 /* Handle the Perl-compatible cases */
869
870 if (ptr[1] == CHAR_LEFT_CURLY_BRACKET)
871 {
872 const pcre_uchar *p;
873 for (p = ptr+2; *p != 0 && *p != CHAR_RIGHT_CURLY_BRACKET; p++)
874 if (*p != CHAR_MINUS && !IS_DIGIT(*p)) break;
875 if (*p != 0 && *p != CHAR_RIGHT_CURLY_BRACKET)
876 {
877 c = -ESC_k;
878 break;
879 }
880 braced = TRUE;
881 ptr++;
882 }
883 else braced = FALSE;
884
885 if (ptr[1] == CHAR_MINUS)
886 {
887 negated = TRUE;
888 ptr++;
889 }
890 else negated = FALSE;
891
892 /* The integer range is limited by the machine's int representation. */
893 c = 0;
894 while (IS_DIGIT(ptr[1]))
895 {
896 if (((unsigned int)c) > INT_MAX / 10) /* Integer overflow */
897 {
898 c = -1;
899 break;
900 }
901 c = c * 10 + *(++ptr) - CHAR_0;
902 }
903 if (((unsigned int)c) > INT_MAX) /* Integer overflow */
904 {
905 while (IS_DIGIT(ptr[1]))
906 ptr++;
907 *errorcodeptr = ERR61;
908 break;
909 }
910
911 if (braced && *(++ptr) != CHAR_RIGHT_CURLY_BRACKET)
912 {
913 *errorcodeptr = ERR57;
914 break;
915 }
916
917 if (c == 0)
918 {
919 *errorcodeptr = ERR58;
920 break;
921 }
922
923 if (negated)
924 {
925 if (c > bracount)
926 {
927 *errorcodeptr = ERR15;
928 break;
929 }
930 c = bracount - (c - 1);
931 }
932
933 c = -(ESC_REF + c);
934 break;
935
936 /* The handling of escape sequences consisting of a string of digits
937 starting with one that is not zero is not straightforward. By experiment,
938 the way Perl works seems to be as follows:
939
940 Outside a character class, the digits are read as a decimal number. If the
941 number is less than 10, or if there are that many previous extracting
942 left brackets, then it is a back reference. Otherwise, up to three octal
943 digits are read to form an escaped byte. Thus \123 is likely to be octal
944 123 (cf \0123, which is octal 012 followed by the literal 3). If the octal
945 value is greater than 377, the least significant 8 bits are taken. Inside a
946 character class, \ followed by a digit is always an octal number. */
947
948 case CHAR_1: case CHAR_2: case CHAR_3: case CHAR_4: case CHAR_5:
949 case CHAR_6: case CHAR_7: case CHAR_8: case CHAR_9:
950
951 if (!isclass)
952 {
953 oldptr = ptr;
954 /* The integer range is limited by the machine's int representation. */
955 c -= CHAR_0;
956 while (IS_DIGIT(ptr[1]))
957 {
958 if (((unsigned int)c) > INT_MAX / 10) /* Integer overflow */
959 {
960 c = -1;
961 break;
962 }
963 c = c * 10 + *(++ptr) - CHAR_0;
964 }
965 if (((unsigned int)c) > INT_MAX) /* Integer overflow */
966 {
967 while (IS_DIGIT(ptr[1]))
968 ptr++;
969 *errorcodeptr = ERR61;
970 break;
971 }
972 if (c < 10 || c <= bracount)
973 {
974 c = -(ESC_REF + c);
975 break;
976 }
977 ptr = oldptr; /* Put the pointer back and fall through */
978 }
979
980 /* Handle an octal number following \. If the first digit is 8 or 9, Perl
981 generates a binary zero byte and treats the digit as a following literal.
982 Thus we have to pull back the pointer by one. */
983
984 if ((c = *ptr) >= CHAR_8)
985 {
986 ptr--;
987 c = 0;
988 break;
989 }
990
991 /* \0 always starts an octal number, but we may drop through to here with a
992 larger first octal digit. The original code used just to take the least
993 significant 8 bits of octal numbers (I think this is what early Perls used
994 to do). Nowadays we allow for larger numbers in UTF-8 mode and 16-bit mode,
995 but no more than 3 octal digits. */
996
997 case CHAR_0:
998 c -= CHAR_0;
999 while(i++ < 2 && ptr[1] >= CHAR_0 && ptr[1] <= CHAR_7)
1000 c = c * 8 + *(++ptr) - CHAR_0;
1001 #ifdef COMPILE_PCRE8
1002 if (!utf && c > 0xff) *errorcodeptr = ERR51;
1003 #endif
1004 break;
1005
1006 /* \x is complicated. \x{ddd} is a character number which can be greater
1007 than 0xff in utf or non-8bit mode, but only if the ddd are hex digits.
1008 If not, { is treated as a data character. */
1009
1010 case CHAR_x:
1011 if ((options & PCRE_JAVASCRIPT_COMPAT) != 0)
1012 {
1013 /* In JavaScript, \x must be followed by two hexadecimal numbers.
1014 Otherwise it is a lowercase x letter. */
1015 if (MAX_255(ptr[1]) && (digitab[ptr[1]] & ctype_xdigit) != 0
1016 && MAX_255(ptr[2]) && (digitab[ptr[2]] & ctype_xdigit) != 0)
1017 {
1018 c = 0;
1019 for (i = 0; i < 2; ++i)
1020 {
1021 register int cc = *(++ptr);
1022 #ifndef EBCDIC /* ASCII/UTF-8 coding */
1023 if (cc >= CHAR_a) cc -= 32; /* Convert to upper case */
1024 c = (c << 4) + cc - ((cc < CHAR_A)? CHAR_0 : (CHAR_A - 10));
1025 #else /* EBCDIC coding */
1026 if (cc >= CHAR_a && cc <= CHAR_z) cc += 64; /* Convert to upper case */
1027 c = (c << 4) + cc - ((cc >= CHAR_0)? CHAR_0 : (CHAR_A - 10));
1028 #endif
1029 }
1030 }
1031 break;
1032 }
1033
1034 if (ptr[1] == CHAR_LEFT_CURLY_BRACKET)
1035 {
1036 const pcre_uchar *pt = ptr + 2;
1037
1038 c = 0;
1039 while (MAX_255(*pt) && (digitab[*pt] & ctype_xdigit) != 0)
1040 {
1041 register int cc = *pt++;
1042 if (c == 0 && cc == CHAR_0) continue; /* Leading zeroes */
1043
1044 #ifndef EBCDIC /* ASCII/UTF-8 coding */
1045 if (cc >= CHAR_a) cc -= 32; /* Convert to upper case */
1046 c = (c << 4) + cc - ((cc < CHAR_A)? CHAR_0 : (CHAR_A - 10));
1047 #else /* EBCDIC coding */
1048 if (cc >= CHAR_a && cc <= CHAR_z) cc += 64; /* Convert to upper case */
1049 c = (c << 4) + cc - ((cc >= CHAR_0)? CHAR_0 : (CHAR_A - 10));
1050 #endif
1051
1052 #ifdef COMPILE_PCRE8
1053 if (c > (utf ? 0x10ffff : 0xff)) { c = -1; break; }
1054 #else
1055 #ifdef COMPILE_PCRE16
1056 if (c > (utf ? 0x10ffff : 0xffff)) { c = -1; break; }
1057 #endif
1058 #endif
1059 }
1060
1061 if (c < 0)
1062 {
1063 while (MAX_255(*pt) && (digitab[*pt] & ctype_xdigit) != 0) pt++;
1064 *errorcodeptr = ERR34;
1065 }
1066
1067 if (*pt == CHAR_RIGHT_CURLY_BRACKET)
1068 {
1069 if (utf && c >= 0xd800 && c <= 0xdfff) *errorcodeptr = ERR73;
1070 ptr = pt;
1071 break;
1072 }
1073
1074 /* If the sequence of hex digits does not end with '}', then we don't
1075 recognize this construct; fall through to the normal \x handling. */
1076 }
1077
1078 /* Read just a single-byte hex-defined char */
1079
1080 c = 0;
1081 while (i++ < 2 && MAX_255(ptr[1]) && (digitab[ptr[1]] & ctype_xdigit) != 0)
1082 {
1083 int cc; /* Some compilers don't like */
1084 cc = *(++ptr); /* ++ in initializers */
1085 #ifndef EBCDIC /* ASCII/UTF-8 coding */
1086 if (cc >= CHAR_a) cc -= 32; /* Convert to upper case */
1087 c = c * 16 + cc - ((cc < CHAR_A)? CHAR_0 : (CHAR_A - 10));
1088 #else /* EBCDIC coding */
1089 if (cc <= CHAR_z) cc += 64; /* Convert to upper case */
1090 c = c * 16 + cc - ((cc >= CHAR_0)? CHAR_0 : (CHAR_A - 10));
1091 #endif
1092 }
1093 break;
1094
1095 /* For \c, a following letter is upper-cased; then the 0x40 bit is flipped.
1096 An error is given if the byte following \c is not an ASCII character. This
1097 coding is ASCII-specific, but then the whole concept of \cx is
1098 ASCII-specific. (However, an EBCDIC equivalent has now been added.) */
1099
1100 case CHAR_c:
1101 c = *(++ptr);
1102 if (c == 0)
1103 {
1104 *errorcodeptr = ERR2;
1105 break;
1106 }
1107 #ifndef EBCDIC /* ASCII/UTF-8 coding */
1108 if (c > 127) /* Excludes all non-ASCII in either mode */
1109 {
1110 *errorcodeptr = ERR68;
1111 break;
1112 }
1113 if (c >= CHAR_a && c <= CHAR_z) c -= 32;
1114 c ^= 0x40;
1115 #else /* EBCDIC coding */
1116 if (c >= CHAR_a && c <= CHAR_z) c += 64;
1117 c ^= 0xC0;
1118 #endif
1119 break;
1120
1121 /* PCRE_EXTRA enables extensions to Perl in the matter of escapes. Any
1122 other alphanumeric following \ is an error if PCRE_EXTRA was set;
1123 otherwise, for Perl compatibility, it is a literal. This code looks a bit
1124 odd, but there used to be some cases other than the default, and there may
1125 be again in future, so I haven't "optimized" it. */
1126
1127 default:
1128 if ((options & PCRE_EXTRA) != 0) switch(c)
1129 {
1130 default:
1131 *errorcodeptr = ERR3;
1132 break;
1133 }
1134 break;
1135 }
1136 }
1137
1138 /* Perl supports \N{name} for character names, as well as plain \N for "not
1139 newline". PCRE does not support \N{name}. However, it does support
1140 quantification such as \N{2,3}. */
1141
1142 if (c == -ESC_N && ptr[1] == CHAR_LEFT_CURLY_BRACKET &&
1143 !is_counted_repeat(ptr+2))
1144 *errorcodeptr = ERR37;
1145
1146 /* If PCRE_UCP is set, we change the values for \d etc. */
1147
1148 if ((options & PCRE_UCP) != 0 && c <= -ESC_D && c >= -ESC_w)
1149 c -= (ESC_DU - ESC_D);
1150
1151 /* Set the pointer to the final character before returning. */
1152
1153 *ptrptr = ptr;
1154 return c;
1155 }
1156
1157
1158
1159 #ifdef SUPPORT_UCP
1160 /*************************************************
1161 * Handle \P and \p *
1162 *************************************************/
1163
1164 /* This function is called after \P or \p has been encountered, provided that
1165 PCRE is compiled with support for Unicode properties. On entry, ptrptr is
1166 pointing at the P or p. On exit, it is pointing at the final character of the
1167 escape sequence.
1168
1169 Argument:
1170 ptrptr points to the pattern position pointer
1171 negptr points to a boolean that is set TRUE for negation else FALSE
1172 dptr points to an int that is set to the detailed property value
1173 errorcodeptr points to the error code variable
1174
1175 Returns: type value from ucp_type_table, or -1 for an invalid type
1176 */
1177
1178 static int
1179 get_ucp(const pcre_uchar **ptrptr, BOOL *negptr, int *dptr, int *errorcodeptr)
1180 {
1181 int c, i, bot, top;
1182 const pcre_uchar *ptr = *ptrptr;
1183 pcre_uchar name[32];
1184
1185 c = *(++ptr);
1186 if (c == 0) goto ERROR_RETURN;
1187
1188 *negptr = FALSE;
1189
1190 /* \P or \p can be followed by a name in {}, optionally preceded by ^ for
1191 negation. */
1192
1193 if (c == CHAR_LEFT_CURLY_BRACKET)
1194 {
1195 if (ptr[1] == CHAR_CIRCUMFLEX_ACCENT)
1196 {
1197 *negptr = TRUE;
1198 ptr++;
1199 }
1200 for (i = 0; i < (int)(sizeof(name) / sizeof(pcre_uchar)) - 1; i++)
1201 {
1202 c = *(++ptr);
1203 if (c == 0) goto ERROR_RETURN;
1204 if (c == CHAR_RIGHT_CURLY_BRACKET) break;
1205 name[i] = c;
1206 }
1207 if (c != CHAR_RIGHT_CURLY_BRACKET) goto ERROR_RETURN;
1208 name[i] = 0;
1209 }
1210
1211 /* Otherwise there is just one following character */
1212
1213 else
1214 {
1215 name[0] = c;
1216 name[1] = 0;
1217 }
1218
1219 *ptrptr = ptr;
1220
1221 /* Search for a recognized property name using binary chop */
1222
1223 bot = 0;
1224 top = PRIV(utt_size);
1225
1226 while (bot < top)
1227 {
1228 i = (bot + top) >> 1;
1229 c = STRCMP_UC_C8(name, PRIV(utt_names) + PRIV(utt)[i].name_offset);
1230 if (c == 0)
1231 {
1232 *dptr = PRIV(utt)[i].value;
1233 return PRIV(utt)[i].type;
1234 }
1235 if (c > 0) bot = i + 1; else top = i;
1236 }
1237
1238 *errorcodeptr = ERR47;
1239 *ptrptr = ptr;
1240 return -1;
1241
1242 ERROR_RETURN:
1243 *errorcodeptr = ERR46;
1244 *ptrptr = ptr;
1245 return -1;
1246 }
1247 #endif
1248
1249
1250
1251
1252 /*************************************************
1253 * Read repeat counts *
1254 *************************************************/
1255
1256 /* Read an item of the form {n,m} and return the values. This is called only
1257 after is_counted_repeat() has confirmed that a repeat-count quantifier exists,
1258 so the syntax is guaranteed to be correct, but we need to check the values.
1259
1260 Arguments:
1261 p pointer to first char after '{'
1262 minp pointer to int for min
1263 maxp pointer to int for max
1264 returned as -1 if no max
1265 errorcodeptr points to error code variable
1266
1267 Returns: pointer to '}' on success;
1268 current ptr on error, with errorcodeptr set non-zero
1269 */
1270
1271 static const pcre_uchar *
1272 read_repeat_counts(const pcre_uchar *p, int *minp, int *maxp, int *errorcodeptr)
1273 {
1274 int min = 0;
1275 int max = -1;
1276
1277 /* Read the minimum value and do a paranoid check: a negative value indicates
1278 an integer overflow. */
1279
1280 while (IS_DIGIT(*p)) min = min * 10 + *p++ - CHAR_0;
1281 if (min < 0 || min > 65535)
1282 {
1283 *errorcodeptr = ERR5;
1284 return p;
1285 }
1286
1287 /* Read the maximum value if there is one, and again do a paranoid on its size.
1288 Also, max must not be less than min. */
1289
1290 if (*p == CHAR_RIGHT_CURLY_BRACKET) max = min; else
1291 {
1292 if (*(++p) != CHAR_RIGHT_CURLY_BRACKET)
1293 {
1294 max = 0;
1295 while(IS_DIGIT(*p)) max = max * 10 + *p++ - CHAR_0;
1296 if (max < 0 || max > 65535)
1297 {
1298 *errorcodeptr = ERR5;
1299 return p;
1300 }
1301 if (max < min)
1302 {
1303 *errorcodeptr = ERR4;
1304 return p;
1305 }
1306 }
1307 }
1308
1309 /* Fill in the required variables, and pass back the pointer to the terminating
1310 '}'. */
1311
1312 *minp = min;
1313 *maxp = max;
1314 return p;
1315 }
1316
1317
1318
1319 /*************************************************
1320 * Subroutine for finding forward reference *
1321 *************************************************/
1322
1323 /* This recursive function is called only from find_parens() below. The
1324 top-level call starts at the beginning of the pattern. All other calls must
1325 start at a parenthesis. It scans along a pattern's text looking for capturing
1326 subpatterns, and counting them. If it finds a named pattern that matches the
1327 name it is given, it returns its number. Alternatively, if the name is NULL, it
1328 returns when it reaches a given numbered subpattern. Recursion is used to keep
1329 track of subpatterns that reset the capturing group numbers - the (?| feature.
1330
1331 This function was originally called only from the second pass, in which we know
1332 that if (?< or (?' or (?P< is encountered, the name will be correctly
1333 terminated because that is checked in the first pass. There is now one call to
1334 this function in the first pass, to check for a recursive back reference by
1335 name (so that we can make the whole group atomic). In this case, we need check
1336 only up to the current position in the pattern, and that is still OK because
1337 and previous occurrences will have been checked. To make this work, the test
1338 for "end of pattern" is a check against cd->end_pattern in the main loop,
1339 instead of looking for a binary zero. This means that the special first-pass
1340 call can adjust cd->end_pattern temporarily. (Checks for binary zero while
1341 processing items within the loop are OK, because afterwards the main loop will
1342 terminate.)
1343
1344 Arguments:
1345 ptrptr address of the current character pointer (updated)
1346 cd compile background data
1347 name name to seek, or NULL if seeking a numbered subpattern
1348 lorn name length, or subpattern number if name is NULL
1349 xmode TRUE if we are in /x mode
1350 utf TRUE if we are in UTF-8 / UTF-16 mode
1351 count pointer to the current capturing subpattern number (updated)
1352
1353 Returns: the number of the named subpattern, or -1 if not found
1354 */
1355
1356 static int
1357 find_parens_sub(pcre_uchar **ptrptr, compile_data *cd, const pcre_uchar *name, int lorn,
1358 BOOL xmode, BOOL utf, int *count)
1359 {
1360 pcre_uchar *ptr = *ptrptr;
1361 int start_count = *count;
1362 int hwm_count = start_count;
1363 BOOL dup_parens = FALSE;
1364
1365 /* If the first character is a parenthesis, check on the type of group we are
1366 dealing with. The very first call may not start with a parenthesis. */
1367
1368 if (ptr[0] == CHAR_LEFT_PARENTHESIS)
1369 {
1370 /* Handle specials such as (*SKIP) or (*UTF8) etc. */
1371
1372 if (ptr[1] == CHAR_ASTERISK) ptr += 2;
1373
1374 /* Handle a normal, unnamed capturing parenthesis. */
1375
1376 else if (ptr[1] != CHAR_QUESTION_MARK)
1377 {
1378 *count += 1;
1379 if (name == NULL && *count == lorn) return *count;
1380 ptr++;
1381 }
1382
1383 /* All cases now have (? at the start. Remember when we are in a group
1384 where the parenthesis numbers are duplicated. */
1385
1386 else if (ptr[2] == CHAR_VERTICAL_LINE)
1387 {
1388 ptr += 3;
1389 dup_parens = TRUE;
1390 }
1391
1392 /* Handle comments; all characters are allowed until a ket is reached. */
1393
1394 else if (ptr[2] == CHAR_NUMBER_SIGN)
1395 {
1396 for (ptr += 3; *ptr != 0; ptr++) if (*ptr == CHAR_RIGHT_PARENTHESIS) break;
1397 goto FAIL_EXIT;
1398 }
1399
1400 /* Handle a condition. If it is an assertion, just carry on so that it
1401 is processed as normal. If not, skip to the closing parenthesis of the
1402 condition (there can't be any nested parens). */
1403
1404 else if (ptr[2] == CHAR_LEFT_PARENTHESIS)
1405 {
1406 ptr += 2;
1407 if (ptr[1] != CHAR_QUESTION_MARK)
1408 {
1409 while (*ptr != 0 && *ptr != CHAR_RIGHT_PARENTHESIS) ptr++;
1410 if (*ptr != 0) ptr++;
1411 }
1412 }
1413
1414 /* Start with (? but not a condition. */
1415
1416 else
1417 {
1418 ptr += 2;
1419 if (*ptr == CHAR_P) ptr++; /* Allow optional P */
1420
1421 /* We have to disambiguate (?<! and (?<= from (?<name> for named groups */
1422
1423 if ((*ptr == CHAR_LESS_THAN_SIGN && ptr[1] != CHAR_EXCLAMATION_MARK &&
1424 ptr[1] != CHAR_EQUALS_SIGN) || *ptr == CHAR_APOSTROPHE)
1425 {
1426 int term;
1427 const pcre_uchar *thisname;
1428 *count += 1;
1429 if (name == NULL && *count == lorn) return *count;
1430 term = *ptr++;
1431 if (term == CHAR_LESS_THAN_SIGN) term = CHAR_GREATER_THAN_SIGN;
1432 thisname = ptr;
1433 while (*ptr != term) ptr++;
1434 if (name != NULL && lorn == ptr - thisname &&
1435 STRNCMP_UC_UC(name, thisname, lorn) == 0)
1436 return *count;
1437 term++;
1438 }
1439 }
1440 }
1441
1442 /* Past any initial parenthesis handling, scan for parentheses or vertical
1443 bars. Stop if we get to cd->end_pattern. Note that this is important for the
1444 first-pass call when this value is temporarily adjusted to stop at the current
1445 position. So DO NOT change this to a test for binary zero. */
1446
1447 for (; ptr < cd->end_pattern; ptr++)
1448 {
1449 /* Skip over backslashed characters and also entire \Q...\E */
1450
1451 if (*ptr == CHAR_BACKSLASH)
1452 {
1453 if (*(++ptr) == 0) goto FAIL_EXIT;
1454 if (*ptr == CHAR_Q) for (;;)
1455 {
1456 while (*(++ptr) != 0 && *ptr != CHAR_BACKSLASH) {};
1457 if (*ptr == 0) goto FAIL_EXIT;
1458 if (*(++ptr) == CHAR_E) break;
1459 }
1460 continue;
1461 }
1462
1463 /* Skip over character classes; this logic must be similar to the way they
1464 are handled for real. If the first character is '^', skip it. Also, if the
1465 first few characters (either before or after ^) are \Q\E or \E we skip them
1466 too. This makes for compatibility with Perl. Note the use of STR macros to
1467 encode "Q\\E" so that it works in UTF-8 on EBCDIC platforms. */
1468
1469 if (*ptr == CHAR_LEFT_SQUARE_BRACKET)
1470 {
1471 BOOL negate_class = FALSE;
1472 for (;;)
1473 {
1474 if (ptr[1] == CHAR_BACKSLASH)
1475 {
1476 if (ptr[2] == CHAR_E)
1477 ptr+= 2;
1478 else if (STRNCMP_UC_C8(ptr + 2,
1479 STR_Q STR_BACKSLASH STR_E, 3) == 0)
1480 ptr += 4;
1481 else
1482 break;
1483 }
1484 else if (!negate_class && ptr[1] == CHAR_CIRCUMFLEX_ACCENT)
1485 {
1486 negate_class = TRUE;
1487 ptr++;
1488 }
1489 else break;
1490 }
1491
1492 /* If the next character is ']', it is a data character that must be
1493 skipped, except in JavaScript compatibility mode. */
1494
1495 if (ptr[1] == CHAR_RIGHT_SQUARE_BRACKET &&
1496 (cd->external_options & PCRE_JAVASCRIPT_COMPAT) == 0)
1497 ptr++;
1498
1499 while (*(++ptr) != CHAR_RIGHT_SQUARE_BRACKET)
1500 {
1501 if (*ptr == 0) return -1;
1502 if (*ptr == CHAR_BACKSLASH)
1503 {
1504 if (*(++ptr) == 0) goto FAIL_EXIT;
1505 if (*ptr == CHAR_Q) for (;;)
1506 {
1507 while (*(++ptr) != 0 && *ptr != CHAR_BACKSLASH) {};
1508 if (*ptr == 0) goto FAIL_EXIT;
1509 if (*(++ptr) == CHAR_E) break;
1510 }
1511 continue;
1512 }
1513 }
1514 continue;
1515 }
1516
1517 /* Skip comments in /x mode */
1518
1519 if (xmode && *ptr == CHAR_NUMBER_SIGN)
1520 {
1521 ptr++;
1522 while (*ptr != 0)
1523 {
1524 if (IS_NEWLINE(ptr)) { ptr += cd->nllen - 1; break; }
1525 ptr++;
1526 #ifdef SUPPORT_UTF
1527 if (utf) FORWARDCHAR(ptr);
1528 #endif
1529 }
1530 if (*ptr == 0) goto FAIL_EXIT;
1531 continue;
1532 }
1533
1534 /* Check for the special metacharacters */
1535
1536 if (*ptr == CHAR_LEFT_PARENTHESIS)
1537 {
1538 int rc = find_parens_sub(&ptr, cd, name, lorn, xmode, utf, count);
1539 if (rc > 0) return rc;
1540 if (*ptr == 0) goto FAIL_EXIT;
1541 }
1542
1543 else if (*ptr == CHAR_RIGHT_PARENTHESIS)
1544 {
1545 if (dup_parens && *count < hwm_count) *count = hwm_count;
1546 goto FAIL_EXIT;
1547 }
1548
1549 else if (*ptr == CHAR_VERTICAL_LINE && dup_parens)
1550 {
1551 if (*count > hwm_count) hwm_count = *count;
1552 *count = start_count;
1553 }
1554 }
1555
1556 FAIL_EXIT:
1557 *ptrptr = ptr;
1558 return -1;
1559 }
1560
1561
1562
1563
1564 /*************************************************
1565 * Find forward referenced subpattern *
1566 *************************************************/
1567
1568 /* This function scans along a pattern's text looking for capturing
1569 subpatterns, and counting them. If it finds a named pattern that matches the
1570 name it is given, it returns its number. Alternatively, if the name is NULL, it
1571 returns when it reaches a given numbered subpattern. This is used for forward
1572 references to subpatterns. We used to be able to start this scan from the
1573 current compiling point, using the current count value from cd->bracount, and
1574 do it all in a single loop, but the addition of the possibility of duplicate
1575 subpattern numbers means that we have to scan from the very start, in order to
1576 take account of such duplicates, and to use a recursive function to keep track
1577 of the different types of group.
1578
1579 Arguments:
1580 cd compile background data
1581 name name to seek, or NULL if seeking a numbered subpattern
1582 lorn name length, or subpattern number if name is NULL
1583 xmode TRUE if we are in /x mode
1584 utf TRUE if we are in UTF-8 / UTF-16 mode
1585
1586 Returns: the number of the found subpattern, or -1 if not found
1587 */
1588
1589 static int
1590 find_parens(compile_data *cd, const pcre_uchar *name, int lorn, BOOL xmode,
1591 BOOL utf)
1592 {
1593 pcre_uchar *ptr = (pcre_uchar *)cd->start_pattern;
1594 int count = 0;
1595 int rc;
1596
1597 /* If the pattern does not start with an opening parenthesis, the first call
1598 to find_parens_sub() will scan right to the end (if necessary). However, if it
1599 does start with a parenthesis, find_parens_sub() will return when it hits the
1600 matching closing parens. That is why we have to have a loop. */
1601
1602 for (;;)
1603 {
1604 rc = find_parens_sub(&ptr, cd, name, lorn, xmode, utf, &count);
1605 if (rc > 0 || *ptr++ == 0) break;
1606 }
1607
1608 return rc;
1609 }
1610
1611
1612
1613
1614 /*************************************************
1615 * Find first significant op code *
1616 *************************************************/
1617
1618 /* This is called by several functions that scan a compiled expression looking
1619 for a fixed first character, or an anchoring op code etc. It skips over things
1620 that do not influence this. For some calls, it makes sense to skip negative
1621 forward and all backward assertions, and also the \b assertion; for others it
1622 does not.
1623
1624 Arguments:
1625 code pointer to the start of the group
1626 skipassert TRUE if certain assertions are to be skipped
1627
1628 Returns: pointer to the first significant opcode
1629 */
1630
1631 static const pcre_uchar*
1632 first_significant_code(const pcre_uchar *code, BOOL skipassert)
1633 {
1634 for (;;)
1635 {
1636 switch ((int)*code)
1637 {
1638 case OP_ASSERT_NOT:
1639 case OP_ASSERTBACK:
1640 case OP_ASSERTBACK_NOT:
1641 if (!skipassert) return code;
1642 do code += GET(code, 1); while (*code == OP_ALT);
1643 code += PRIV(OP_lengths)[*code];
1644 break;
1645
1646 case OP_WORD_BOUNDARY:
1647 case OP_NOT_WORD_BOUNDARY:
1648 if (!skipassert) return code;
1649 /* Fall through */
1650
1651 case OP_CALLOUT:
1652 case OP_CREF:
1653 case OP_NCREF:
1654 case OP_RREF:
1655 case OP_NRREF:
1656 case OP_DEF:
1657 code += PRIV(OP_lengths)[*code];
1658 break;
1659
1660 default:
1661 return code;
1662 }
1663 }
1664 /* Control never reaches here */
1665 }
1666
1667
1668
1669
1670 /*************************************************
1671 * Find the fixed length of a branch *
1672 *************************************************/
1673
1674 /* Scan a branch and compute the fixed length of subject that will match it,
1675 if the length is fixed. This is needed for dealing with backward assertions.
1676 In UTF8 mode, the result is in characters rather than bytes. The branch is
1677 temporarily terminated with OP_END when this function is called.
1678
1679 This function is called when a backward assertion is encountered, so that if it
1680 fails, the error message can point to the correct place in the pattern.
1681 However, we cannot do this when the assertion contains subroutine calls,
1682 because they can be forward references. We solve this by remembering this case
1683 and doing the check at the end; a flag specifies which mode we are running in.
1684
1685 Arguments:
1686 code points to the start of the pattern (the bracket)
1687 utf TRUE in UTF-8 / UTF-16 mode
1688 atend TRUE if called when the pattern is complete
1689 cd the "compile data" structure
1690
1691 Returns: the fixed length,
1692 or -1 if there is no fixed length,
1693 or -2 if \C was encountered (in UTF-8 mode only)
1694 or -3 if an OP_RECURSE item was encountered and atend is FALSE
1695 or -4 if an unknown opcode was encountered (internal error)
1696 */
1697
1698 static int
1699 find_fixedlength(pcre_uchar *code, BOOL utf, BOOL atend, compile_data *cd)
1700 {
1701 int length = -1;
1702
1703 register int branchlength = 0;
1704 register pcre_uchar *cc = code + 1 + LINK_SIZE;
1705
1706 /* Scan along the opcodes for this branch. If we get to the end of the
1707 branch, check the length against that of the other branches. */
1708
1709 for (;;)
1710 {
1711 int d;
1712 pcre_uchar *ce, *cs;
1713 register int op = *cc;
1714
1715 switch (op)
1716 {
1717 /* We only need to continue for OP_CBRA (normal capturing bracket) and
1718 OP_BRA (normal non-capturing bracket) because the other variants of these
1719 opcodes are all concerned with unlimited repeated groups, which of course
1720 are not of fixed length. */
1721
1722 case OP_CBRA:
1723 case OP_BRA:
1724 case OP_ONCE:
1725 case OP_ONCE_NC:
1726 case OP_COND:
1727 d = find_fixedlength(cc + ((op == OP_CBRA)? IMM2_SIZE : 0), utf, atend, cd);
1728 if (d < 0) return d;
1729 branchlength += d;
1730 do cc += GET(cc, 1); while (*cc == OP_ALT);
1731 cc += 1 + LINK_SIZE;
1732 break;
1733
1734 /* Reached end of a branch; if it's a ket it is the end of a nested call.
1735 If it's ALT it is an alternation in a nested call. An ACCEPT is effectively
1736 an ALT. If it is END it's the end of the outer call. All can be handled by
1737 the same code. Note that we must not include the OP_KETRxxx opcodes here,
1738 because they all imply an unlimited repeat. */
1739
1740 case OP_ALT:
1741 case OP_KET:
1742 case OP_END:
1743 case OP_ACCEPT:
1744 case OP_ASSERT_ACCEPT:
1745 if (length < 0) length = branchlength;
1746 else if (length != branchlength) return -1;
1747 if (*cc != OP_ALT) return length;
1748 cc += 1 + LINK_SIZE;
1749 branchlength = 0;
1750 break;
1751
1752 /* A true recursion implies not fixed length, but a subroutine call may
1753 be OK. If the subroutine is a forward reference, we can't deal with
1754 it until the end of the pattern, so return -3. */
1755
1756 case OP_RECURSE:
1757 if (!atend) return -3;
1758 cs = ce = (pcre_uchar *)cd->start_code + GET(cc, 1); /* Start subpattern */
1759 do ce += GET(ce, 1); while (*ce == OP_ALT); /* End subpattern */
1760 if (cc > cs && cc < ce) return -1; /* Recursion */
1761 d = find_fixedlength(cs + IMM2_SIZE, utf, atend, cd);
1762 if (d < 0) return d;
1763 branchlength += d;
1764 cc += 1 + LINK_SIZE;
1765 break;
1766
1767 /* Skip over assertive subpatterns */
1768
1769 case OP_ASSERT:
1770 case OP_ASSERT_NOT:
1771 case OP_ASSERTBACK:
1772 case OP_ASSERTBACK_NOT:
1773 do cc += GET(cc, 1); while (*cc == OP_ALT);
1774 cc += PRIV(OP_lengths)[*cc];
1775 break;
1776
1777 /* Skip over things that don't match chars */
1778
1779 case OP_MARK:
1780 case OP_PRUNE_ARG:
1781 case OP_SKIP_ARG:
1782 case OP_THEN_ARG:
1783 cc += cc[1] + PRIV(OP_lengths)[*cc];
1784 break;
1785
1786 case OP_CALLOUT:
1787 case OP_CIRC:
1788 case OP_CIRCM:
1789 case OP_CLOSE:
1790 case OP_COMMIT:
1791 case OP_CREF:
1792 case OP_DEF:
1793 case OP_DOLL:
1794 case OP_DOLLM:
1795 case OP_EOD:
1796 case OP_EODN:
1797 case OP_FAIL:
1798 case OP_NCREF:
1799 case OP_NRREF:
1800 case OP_NOT_WORD_BOUNDARY:
1801 case OP_PRUNE:
1802 case OP_REVERSE:
1803 case OP_RREF:
1804 case OP_SET_SOM:
1805 case OP_SKIP:
1806 case OP_SOD:
1807 case OP_SOM:
1808 case OP_THEN:
1809 case OP_WORD_BOUNDARY:
1810 cc += PRIV(OP_lengths)[*cc];
1811 break;
1812
1813 /* Handle literal characters */
1814
1815 case OP_CHAR:
1816 case OP_CHARI:
1817 case OP_NOT:
1818 case OP_NOTI:
1819 branchlength++;
1820 cc += 2;
1821 #ifdef SUPPORT_UTF
1822 if (utf && HAS_EXTRALEN(cc[-1])) cc += GET_EXTRALEN(cc[-1]);
1823 #endif
1824 break;
1825
1826 /* Handle exact repetitions. The count is already in characters, but we
1827 need to skip over a multibyte character in UTF8 mode. */
1828
1829 case OP_EXACT:
1830 case OP_EXACTI:
1831 case OP_NOTEXACT:
1832 case OP_NOTEXACTI:
1833 branchlength += GET2(cc,1);
1834 cc += 2 + IMM2_SIZE;
1835 #ifdef SUPPORT_UTF
1836 if (utf && HAS_EXTRALEN(cc[-1])) cc += GET_EXTRALEN(cc[-1]);
1837 #endif
1838 break;
1839
1840 case OP_TYPEEXACT:
1841 branchlength += GET2(cc,1);
1842 if (cc[1 + IMM2_SIZE] == OP_PROP || cc[1 + IMM2_SIZE] == OP_NOTPROP) cc += 2;
1843 cc += 1 + IMM2_SIZE + 1;
1844 break;
1845
1846 /* Handle single-char matchers */
1847
1848 case OP_PROP:
1849 case OP_NOTPROP:
1850 cc += 2;
1851 /* Fall through */
1852
1853 case OP_HSPACE:
1854 case OP_VSPACE:
1855 case OP_NOT_HSPACE:
1856 case OP_NOT_VSPACE:
1857 case OP_NOT_DIGIT:
1858 case OP_DIGIT:
1859 case OP_NOT_WHITESPACE:
1860 case OP_WHITESPACE:
1861 case OP_NOT_WORDCHAR:
1862 case OP_WORDCHAR:
1863 case OP_ANY:
1864 case OP_ALLANY:
1865 branchlength++;
1866 cc++;
1867 break;
1868
1869 /* The single-byte matcher isn't allowed. This only happens in UTF-8 mode;
1870 otherwise \C is coded as OP_ALLANY. */
1871
1872 case OP_ANYBYTE:
1873 return -2;
1874
1875 /* Check a class for variable quantification */
1876
1877 #if defined SUPPORT_UTF || defined COMPILE_PCRE16
1878 case OP_XCLASS:
1879 cc += GET(cc, 1) - PRIV(OP_lengths)[OP_CLASS];
1880 /* Fall through */
1881 #endif
1882
1883 case OP_CLASS:
1884 case OP_NCLASS:
1885 cc += PRIV(OP_lengths)[OP_CLASS];
1886
1887 switch (*cc)
1888 {
1889 case OP_CRPLUS:
1890 case OP_CRMINPLUS:
1891 case OP_CRSTAR:
1892 case OP_CRMINSTAR:
1893 case OP_CRQUERY:
1894 case OP_CRMINQUERY:
1895 return -1;
1896
1897 case OP_CRRANGE:
1898 case OP_CRMINRANGE:
1899 if (GET2(cc,1) != GET2(cc,1+IMM2_SIZE)) return -1;
1900 branchlength += GET2(cc,1);
1901 cc += 1 + 2 * IMM2_SIZE;
1902 break;
1903
1904 default:
1905 branchlength++;
1906 }
1907 break;
1908
1909 /* Anything else is variable length */
1910
1911 case OP_ANYNL:
1912 case OP_BRAMINZERO:
1913 case OP_BRAPOS:
1914 case OP_BRAPOSZERO:
1915 case OP_BRAZERO:
1916 case OP_CBRAPOS:
1917 case OP_EXTUNI:
1918 case OP_KETRMAX:
1919 case OP_KETRMIN:
1920 case OP_KETRPOS:
1921 case OP_MINPLUS:
1922 case OP_MINPLUSI:
1923 case OP_MINQUERY:
1924 case OP_MINQUERYI:
1925 case OP_MINSTAR:
1926 case OP_MINSTARI:
1927 case OP_MINUPTO:
1928 case OP_MINUPTOI:
1929 case OP_NOTMINPLUS:
1930 case OP_NOTMINPLUSI:
1931 case OP_NOTMINQUERY:
1932 case OP_NOTMINQUERYI:
1933 case OP_NOTMINSTAR:
1934 case OP_NOTMINSTARI:
1935 case OP_NOTMINUPTO:
1936 case OP_NOTMINUPTOI:
1937 case OP_NOTPLUS:
1938 case OP_NOTPLUSI:
1939 case OP_NOTPOSPLUS:
1940 case OP_NOTPOSPLUSI:
1941 case OP_NOTPOSQUERY:
1942 case OP_NOTPOSQUERYI:
1943 case OP_NOTPOSSTAR:
1944 case OP_NOTPOSSTARI:
1945 case OP_NOTPOSUPTO:
1946 case OP_NOTPOSUPTOI:
1947 case OP_NOTQUERY:
1948 case OP_NOTQUERYI:
1949 case OP_NOTSTAR:
1950 case OP_NOTSTARI:
1951 case OP_NOTUPTO:
1952 case OP_NOTUPTOI:
1953 case OP_PLUS:
1954 case OP_PLUSI:
1955 case OP_POSPLUS:
1956 case OP_POSPLUSI:
1957 case OP_POSQUERY:
1958 case OP_POSQUERYI:
1959 case OP_POSSTAR:
1960 case OP_POSSTARI:
1961 case OP_POSUPTO:
1962 case OP_POSUPTOI:
1963 case OP_QUERY:
1964 case OP_QUERYI:
1965 case OP_REF:
1966 case OP_REFI:
1967 case OP_SBRA:
1968 case OP_SBRAPOS:
1969 case OP_SCBRA:
1970 case OP_SCBRAPOS:
1971 case OP_SCOND:
1972 case OP_SKIPZERO:
1973 case OP_STAR:
1974 case OP_STARI:
1975 case OP_TYPEMINPLUS:
1976 case OP_TYPEMINQUERY:
1977 case OP_TYPEMINSTAR:
1978 case OP_TYPEMINUPTO:
1979 case OP_TYPEPLUS:
1980 case OP_TYPEPOSPLUS:
1981 case OP_TYPEPOSQUERY:
1982 case OP_TYPEPOSSTAR:
1983 case OP_TYPEPOSUPTO:
1984 case OP_TYPEQUERY:
1985 case OP_TYPESTAR:
1986 case OP_TYPEUPTO:
1987 case OP_UPTO:
1988 case OP_UPTOI:
1989 return -1;
1990
1991 /* Catch unrecognized opcodes so that when new ones are added they
1992 are not forgotten, as has happened in the past. */
1993
1994 default:
1995 return -4;
1996 }
1997 }
1998 /* Control never gets here */
1999 }
2000
2001
2002
2003
2004 /*************************************************
2005 * Scan compiled regex for specific bracket *
2006 *************************************************/
2007
2008 /* This little function scans through a compiled pattern until it finds a
2009 capturing bracket with the given number, or, if the number is negative, an
2010 instance of OP_REVERSE for a lookbehind. The function is global in the C sense
2011 so that it can be called from pcre_study() when finding the minimum matching
2012 length.
2013
2014 Arguments:
2015 code points to start of expression
2016 utf TRUE in UTF-8 / UTF-16 mode
2017 number the required bracket number or negative to find a lookbehind
2018
2019 Returns: pointer to the opcode for the bracket, or NULL if not found
2020 */
2021
2022 const pcre_uchar *
2023 PRIV(find_bracket)(const pcre_uchar *code, BOOL utf, int number)
2024 {
2025 for (;;)
2026 {
2027 register int c = *code;
2028
2029 if (c == OP_END) return NULL;
2030
2031 /* XCLASS is used for classes that cannot be represented just by a bit
2032 map. This includes negated single high-valued characters. The length in
2033 the table is zero; the actual length is stored in the compiled code. */
2034
2035 if (c == OP_XCLASS) code += GET(code, 1);
2036
2037 /* Handle recursion */
2038
2039 else if (c == OP_REVERSE)
2040 {
2041 if (number < 0) return (pcre_uchar *)code;
2042 code += PRIV(OP_lengths)[c];
2043 }
2044
2045 /* Handle capturing bracket */
2046
2047 else if (c == OP_CBRA || c == OP_SCBRA ||
2048 c == OP_CBRAPOS || c == OP_SCBRAPOS)
2049 {
2050 int n = GET2(code, 1+LINK_SIZE);
2051 if (n == number) return (pcre_uchar *)code;
2052 code += PRIV(OP_lengths)[c];
2053 }
2054
2055 /* Otherwise, we can get the item's length from the table, except that for
2056 repeated character types, we have to test for \p and \P, which have an extra
2057 two bytes of parameters, and for MARK/PRUNE/SKIP/THEN with an argument, we
2058 must add in its length. */
2059
2060 else
2061 {
2062 switch(c)
2063 {
2064 case OP_TYPESTAR:
2065 case OP_TYPEMINSTAR:
2066 case OP_TYPEPLUS:
2067 case OP_TYPEMINPLUS:
2068 case OP_TYPEQUERY:
2069 case OP_TYPEMINQUERY:
2070 case OP_TYPEPOSSTAR:
2071 case OP_TYPEPOSPLUS:
2072 case OP_TYPEPOSQUERY:
2073 if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
2074 break;
2075
2076 case OP_TYPEUPTO:
2077 case OP_TYPEMINUPTO:
2078 case OP_TYPEEXACT:
2079 case OP_TYPEPOSUPTO:
2080 if (code[1 + IMM2_SIZE] == OP_PROP
2081 || code[1 + IMM2_SIZE] == OP_NOTPROP) code += 2;
2082 break;
2083
2084 case OP_MARK:
2085 case OP_PRUNE_ARG:
2086 case OP_SKIP_ARG:
2087 code += code[1];
2088 break;
2089
2090 case OP_THEN_ARG:
2091 code += code[1];
2092 break;
2093 }
2094
2095 /* Add in the fixed length from the table */
2096
2097 code += PRIV(OP_lengths)[c];
2098
2099 /* In UTF-8 mode, opcodes that are followed by a character may be followed by
2100 a multi-byte character. The length in the table is a minimum, so we have to
2101 arrange to skip the extra bytes. */
2102
2103 #ifdef SUPPORT_UTF
2104 if (utf) switch(c)
2105 {
2106 case OP_CHAR:
2107 case OP_CHARI:
2108 case OP_EXACT:
2109 case OP_EXACTI:
2110 case OP_UPTO:
2111 case OP_UPTOI:
2112 case OP_MINUPTO:
2113 case OP_MINUPTOI:
2114 case OP_POSUPTO:
2115 case OP_POSUPTOI:
2116 case OP_STAR:
2117 case OP_STARI:
2118 case OP_MINSTAR:
2119 case OP_MINSTARI:
2120 case OP_POSSTAR:
2121 case OP_POSSTARI:
2122 case OP_PLUS:
2123 case OP_PLUSI:
2124 case OP_MINPLUS:
2125 case OP_MINPLUSI:
2126 case OP_POSPLUS:
2127 case OP_POSPLUSI:
2128 case OP_QUERY:
2129 case OP_QUERYI:
2130 case OP_MINQUERY:
2131 case OP_MINQUERYI:
2132 case OP_POSQUERY:
2133 case OP_POSQUERYI:
2134 if (HAS_EXTRALEN(code[-1])) code += GET_EXTRALEN(code[-1]);
2135 break;
2136 }
2137 #else
2138 (void)(utf); /* Keep compiler happy by referencing function argument */
2139 #endif
2140 }
2141 }
2142 }
2143
2144
2145
2146 /*************************************************
2147 * Scan compiled regex for recursion reference *
2148 *************************************************/
2149
2150 /* This little function scans through a compiled pattern until it finds an
2151 instance of OP_RECURSE.
2152
2153 Arguments:
2154 code points to start of expression
2155 utf TRUE in UTF-8 / UTF-16 mode
2156
2157 Returns: pointer to the opcode for OP_RECURSE, or NULL if not found
2158 */
2159
2160 static const pcre_uchar *
2161 find_recurse(const pcre_uchar *code, BOOL utf)
2162 {
2163 for (;;)
2164 {
2165 register int c = *code;
2166 if (c == OP_END) return NULL;
2167 if (c == OP_RECURSE) return code;
2168
2169 /* XCLASS is used for classes that cannot be represented just by a bit
2170 map. This includes negated single high-valued characters. The length in
2171 the table is zero; the actual length is stored in the compiled code. */
2172
2173 if (c == OP_XCLASS) code += GET(code, 1);
2174
2175 /* Otherwise, we can get the item's length from the table, except that for
2176 repeated character types, we have to test for \p and \P, which have an extra
2177 two bytes of parameters, and for MARK/PRUNE/SKIP/THEN with an argument, we
2178 must add in its length. */
2179
2180 else
2181 {
2182 switch(c)
2183 {
2184 case OP_TYPESTAR:
2185 case OP_TYPEMINSTAR:
2186 case OP_TYPEPLUS:
2187 case OP_TYPEMINPLUS:
2188 case OP_TYPEQUERY:
2189 case OP_TYPEMINQUERY:
2190 case OP_TYPEPOSSTAR:
2191 case OP_TYPEPOSPLUS:
2192 case OP_TYPEPOSQUERY:
2193 if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
2194 break;
2195
2196 case OP_TYPEPOSUPTO:
2197 case OP_TYPEUPTO:
2198 case OP_TYPEMINUPTO:
2199 case OP_TYPEEXACT:
2200 if (code[1 + IMM2_SIZE] == OP_PROP
2201 || code[1 + IMM2_SIZE] == OP_NOTPROP) code += 2;
2202 break;
2203
2204 case OP_MARK:
2205 case OP_PRUNE_ARG:
2206 case OP_SKIP_ARG:
2207 code += code[1];
2208 break;
2209
2210 case OP_THEN_ARG:
2211 code += code[1];
2212 break;
2213 }
2214
2215 /* Add in the fixed length from the table */
2216
2217 code += PRIV(OP_lengths)[c];
2218
2219 /* In UTF-8 mode, opcodes that are followed by a character may be followed
2220 by a multi-byte character. The length in the table is a minimum, so we have
2221 to arrange to skip the extra bytes. */
2222
2223 #ifdef SUPPORT_UTF
2224 if (utf) switch(c)
2225 {
2226 case OP_CHAR:
2227 case OP_CHARI:
2228 case OP_EXACT:
2229 case OP_EXACTI:
2230 case OP_UPTO:
2231 case OP_UPTOI:
2232 case OP_MINUPTO:
2233 case OP_MINUPTOI:
2234 case OP_POSUPTO:
2235 case OP_POSUPTOI:
2236 case OP_STAR:
2237 case OP_STARI:
2238 case OP_MINSTAR:
2239 case OP_MINSTARI:
2240 case OP_POSSTAR:
2241 case OP_POSSTARI:
2242 case OP_PLUS:
2243 case OP_PLUSI:
2244 case OP_MINPLUS:
2245 case OP_MINPLUSI:
2246 case OP_POSPLUS:
2247 case OP_POSPLUSI:
2248 case OP_QUERY:
2249 case OP_QUERYI:
2250 case OP_MINQUERY:
2251 case OP_MINQUERYI:
2252 case OP_POSQUERY:
2253 case OP_POSQUERYI:
2254 if (HAS_EXTRALEN(code[-1])) code += GET_EXTRALEN(code[-1]);
2255 break;
2256 }
2257 #else
2258 (void)(utf); /* Keep compiler happy by referencing function argument */
2259 #endif
2260 }
2261 }
2262 }
2263
2264
2265
2266 /*************************************************
2267 * Scan compiled branch for non-emptiness *
2268 *************************************************/
2269
2270 /* This function scans through a branch of a compiled pattern to see whether it
2271 can match the empty string or not. It is called from could_be_empty()
2272 below and from compile_branch() when checking for an unlimited repeat of a
2273 group that can match nothing. Note that first_significant_code() skips over
2274 backward and negative forward assertions when its final argument is TRUE. If we
2275 hit an unclosed bracket, we return "empty" - this means we've struck an inner
2276 bracket whose current branch will already have been scanned.
2277
2278 Arguments:
2279 code points to start of search
2280 endcode points to where to stop
2281 utf TRUE if in UTF-8 / UTF-16 mode
2282 cd contains pointers to tables etc.
2283
2284 Returns: TRUE if what is matched could be empty
2285 */
2286
2287 static BOOL
2288 could_be_empty_branch(const pcre_uchar *code, const pcre_uchar *endcode,
2289 BOOL utf, compile_data *cd)
2290 {
2291 register int c;
2292 for (code = first_significant_code(code + PRIV(OP_lengths)[*code], TRUE);
2293 code < endcode;
2294 code = first_significant_code(code + PRIV(OP_lengths)[c], TRUE))
2295 {
2296 const pcre_uchar *ccode;
2297
2298 c = *code;
2299
2300 /* Skip over forward assertions; the other assertions are skipped by
2301 first_significant_code() with a TRUE final argument. */
2302
2303 if (c == OP_ASSERT)
2304 {
2305 do code += GET(code, 1); while (*code == OP_ALT);
2306 c = *code;
2307 continue;
2308 }
2309
2310 /* For a recursion/subroutine call, if its end has been reached, which
2311 implies a backward reference subroutine call, we can scan it. If it's a
2312 forward reference subroutine call, we can't. To detect forward reference
2313 we have to scan up the list that is kept in the workspace. This function is
2314 called only when doing the real compile, not during the pre-compile that
2315 measures the size of the compiled pattern. */
2316
2317 if (c == OP_RECURSE)
2318 {
2319 const pcre_uchar *scode;
2320 BOOL empty_branch;
2321
2322 /* Test for forward reference */
2323
2324 for (scode = cd->start_workspace; scode < cd->hwm; scode += LINK_SIZE)
2325 if (GET(scode, 0) == code + 1 - cd->start_code) return TRUE;
2326
2327 /* Not a forward reference, test for completed backward reference */
2328
2329 empty_branch = FALSE;
2330 scode = cd->start_code + GET(code, 1);
2331 if (GET(scode, 1) == 0) return TRUE; /* Unclosed */
2332
2333 /* Completed backwards reference */
2334
2335 do
2336 {
2337 if (could_be_empty_branch(scode, endcode, utf, cd))
2338 {
2339 empty_branch = TRUE;
2340 break;
2341 }
2342 scode += GET(scode, 1);
2343 }
2344 while (*scode == OP_ALT);
2345
2346 if (!empty_branch) return FALSE; /* All branches are non-empty */
2347 continue;
2348 }
2349
2350 /* Groups with zero repeats can of course be empty; skip them. */
2351
2352 if (c == OP_BRAZERO || c == OP_BRAMINZERO || c == OP_SKIPZERO ||
2353 c == OP_BRAPOSZERO)
2354 {
2355 code += PRIV(OP_lengths)[c];
2356 do code += GET(code, 1); while (*code == OP_ALT);
2357 c = *code;
2358 continue;
2359 }
2360
2361 /* A nested group that is already marked as "could be empty" can just be
2362 skipped. */
2363
2364 if (c == OP_SBRA || c == OP_SBRAPOS ||
2365 c == OP_SCBRA || c == OP_SCBRAPOS)
2366 {
2367 do code += GET(code, 1); while (*code == OP_ALT);
2368 c = *code;
2369 continue;
2370 }
2371
2372 /* For other groups, scan the branches. */
2373
2374 if (c == OP_BRA || c == OP_BRAPOS ||
2375 c == OP_CBRA || c == OP_CBRAPOS ||
2376 c == OP_ONCE || c == OP_ONCE_NC ||
2377 c == OP_COND)
2378 {
2379 BOOL empty_branch;
2380 if (GET(code, 1) == 0) return TRUE; /* Hit unclosed bracket */
2381
2382 /* If a conditional group has only one branch, there is a second, implied,
2383 empty branch, so just skip over the conditional, because it could be empty.
2384 Otherwise, scan the individual branches of the group. */
2385
2386 if (c == OP_COND && code[GET(code, 1)] != OP_ALT)
2387 code += GET(code, 1);
2388 else
2389 {
2390 empty_branch = FALSE;
2391 do
2392 {
2393 if (!empty_branch && could_be_empty_branch(code, endcode, utf, cd))
2394 empty_branch = TRUE;
2395 code += GET(code, 1);
2396 }
2397 while (*code == OP_ALT);
2398 if (!empty_branch) return FALSE; /* All branches are non-empty */
2399 }
2400
2401 c = *code;
2402 continue;
2403 }
2404
2405 /* Handle the other opcodes */
2406
2407 switch (c)
2408 {
2409 /* Check for quantifiers after a class. XCLASS is used for classes that
2410 cannot be represented just by a bit map. This includes negated single
2411 high-valued characters. The length in PRIV(OP_lengths)[] is zero; the
2412 actual length is stored in the compiled code, so we must update "code"
2413 here. */
2414
2415 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
2416 case OP_XCLASS:
2417 ccode = code += GET(code, 1);
2418 goto CHECK_CLASS_REPEAT;
2419 #endif
2420
2421 case OP_CLASS:
2422 case OP_NCLASS:
2423 ccode = code + PRIV(OP_lengths)[OP_CLASS];
2424
2425 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
2426 CHECK_CLASS_REPEAT:
2427 #endif
2428
2429 switch (*ccode)
2430 {
2431 case OP_CRSTAR: /* These could be empty; continue */
2432 case OP_CRMINSTAR:
2433 case OP_CRQUERY:
2434 case OP_CRMINQUERY:
2435 break;
2436
2437 default: /* Non-repeat => class must match */
2438 case OP_CRPLUS: /* These repeats aren't empty */
2439 case OP_CRMINPLUS:
2440 return FALSE;
2441
2442 case OP_CRRANGE:
2443 case OP_CRMINRANGE:
2444 if (GET2(ccode, 1) > 0) return FALSE; /* Minimum > 0 */
2445 break;
2446 }
2447 break;
2448
2449 /* Opcodes that must match a character */
2450
2451 case OP_PROP:
2452 case OP_NOTPROP:
2453 case OP_EXTUNI:
2454 case OP_NOT_DIGIT:
2455 case OP_DIGIT:
2456 case OP_NOT_WHITESPACE:
2457 case OP_WHITESPACE:
2458 case OP_NOT_WORDCHAR:
2459 case OP_WORDCHAR:
2460 case OP_ANY:
2461 case OP_ALLANY:
2462 case OP_ANYBYTE:
2463 case OP_CHAR:
2464 case OP_CHARI:
2465 case OP_NOT:
2466 case OP_NOTI:
2467 case OP_PLUS:
2468 case OP_MINPLUS:
2469 case OP_POSPLUS:
2470 case OP_EXACT:
2471 case OP_NOTPLUS:
2472 case OP_NOTMINPLUS:
2473 case OP_NOTPOSPLUS:
2474 case OP_NOTEXACT:
2475 case OP_TYPEPLUS:
2476 case OP_TYPEMINPLUS:
2477 case OP_TYPEPOSPLUS:
2478 case OP_TYPEEXACT:
2479 return FALSE;
2480
2481 /* These are going to continue, as they may be empty, but we have to
2482 fudge the length for the \p and \P cases. */
2483
2484 case OP_TYPESTAR:
2485 case OP_TYPEMINSTAR:
2486 case OP_TYPEPOSSTAR:
2487 case OP_TYPEQUERY:
2488 case OP_TYPEMINQUERY:
2489 case OP_TYPEPOSQUERY:
2490 if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
2491 break;
2492
2493 /* Same for these */
2494
2495 case OP_TYPEUPTO:
2496 case OP_TYPEMINUPTO:
2497 case OP_TYPEPOSUPTO:
2498 if (code[1 + IMM2_SIZE] == OP_PROP
2499 || code[1 + IMM2_SIZE] == OP_NOTPROP) code += 2;
2500 break;
2501
2502 /* End of branch */
2503
2504 case OP_KET:
2505 case OP_KETRMAX:
2506 case OP_KETRMIN:
2507 case OP_KETRPOS:
2508 case OP_ALT:
2509 return TRUE;
2510
2511 /* In UTF-8 mode, STAR, MINSTAR, POSSTAR, QUERY, MINQUERY, POSQUERY, UPTO,
2512 MINUPTO, and POSUPTO may be followed by a multibyte character */
2513
2514 #ifdef SUPPORT_UTF
2515 case OP_STAR:
2516 case OP_STARI:
2517 case OP_MINSTAR:
2518 case OP_MINSTARI:
2519 case OP_POSSTAR:
2520 case OP_POSSTARI:
2521 case OP_QUERY:
2522 case OP_QUERYI:
2523 case OP_MINQUERY:
2524 case OP_MINQUERYI:
2525 case OP_POSQUERY:
2526 case OP_POSQUERYI:
2527 if (utf && HAS_EXTRALEN(code[1])) code += GET_EXTRALEN(code[1]);
2528 break;
2529
2530 case OP_UPTO:
2531 case OP_UPTOI:
2532 case OP_MINUPTO:
2533 case OP_MINUPTOI:
2534 case OP_POSUPTO:
2535 case OP_POSUPTOI:
2536 if (utf && HAS_EXTRALEN(code[1 + IMM2_SIZE])) code += GET_EXTRALEN(code[1 + IMM2_SIZE]);
2537 break;
2538 #endif
2539
2540 /* MARK, and PRUNE/SKIP/THEN with an argument must skip over the argument
2541 string. */
2542
2543 case OP_MARK:
2544 case OP_PRUNE_ARG:
2545 case OP_SKIP_ARG:
2546 code += code[1];
2547 break;
2548
2549 case OP_THEN_ARG:
2550 code += code[1];
2551 break;
2552
2553 /* None of the remaining opcodes are required to match a character. */
2554
2555 default:
2556 break;
2557 }
2558 }
2559
2560 return TRUE;
2561 }
2562
2563
2564
2565 /*************************************************
2566 * Scan compiled regex for non-emptiness *
2567 *************************************************/
2568
2569 /* This function is called to check for left recursive calls. We want to check
2570 the current branch of the current pattern to see if it could match the empty
2571 string. If it could, we must look outwards for branches at other levels,
2572 stopping when we pass beyond the bracket which is the subject of the recursion.
2573 This function is called only during the real compile, not during the
2574 pre-compile.
2575
2576 Arguments:
2577 code points to start of the recursion
2578 endcode points to where to stop (current RECURSE item)
2579 bcptr points to the chain of current (unclosed) branch starts
2580 utf TRUE if in UTF-8 / UTF-16 mode
2581 cd pointers to tables etc
2582
2583 Returns: TRUE if what is matched could be empty
2584 */
2585
2586 static BOOL
2587 could_be_empty(const pcre_uchar *code, const pcre_uchar *endcode,
2588 branch_chain *bcptr, BOOL utf, compile_data *cd)
2589 {
2590 while (bcptr != NULL && bcptr->current_branch >= code)
2591 {
2592 if (!could_be_empty_branch(bcptr->current_branch, endcode, utf, cd))
2593 return FALSE;
2594 bcptr = bcptr->outer;
2595 }
2596 return TRUE;
2597 }
2598
2599
2600
2601 /*************************************************
2602 * Check for POSIX class syntax *
2603 *************************************************/
2604
2605 /* This function is called when the sequence "[:" or "[." or "[=" is
2606 encountered in a character class. It checks whether this is followed by a
2607 sequence of characters terminated by a matching ":]" or ".]" or "=]". If we
2608 reach an unescaped ']' without the special preceding character, return FALSE.
2609
2610 Originally, this function only recognized a sequence of letters between the
2611 terminators, but it seems that Perl recognizes any sequence of characters,
2612 though of course unknown POSIX names are subsequently rejected. Perl gives an
2613 "Unknown POSIX class" error for [:f\oo:] for example, where previously PCRE
2614 didn't consider this to be a POSIX class. Likewise for [:1234:].
2615
2616 The problem in trying to be exactly like Perl is in the handling of escapes. We
2617 have to be sure that [abc[:x\]pqr] is *not* treated as containing a POSIX
2618 class, but [abc[:x\]pqr:]] is (so that an error can be generated). The code
2619 below handles the special case of \], but does not try to do any other escape
2620 processing. This makes it different from Perl for cases such as [:l\ower:]
2621 where Perl recognizes it as the POSIX class "lower" but PCRE does not recognize
2622 "l\ower". This is a lesser evil that not diagnosing bad classes when Perl does,
2623 I think.
2624
2625 A user pointed out that PCRE was rejecting [:a[:digit:]] whereas Perl was not.
2626 It seems that the appearance of a nested POSIX class supersedes an apparent
2627 external class. For example, [:a[:digit:]b:] matches "a", "b", ":", or
2628 a digit.
2629
2630 In Perl, unescaped square brackets may also appear as part of class names. For
2631 example, [:a[:abc]b:] gives unknown POSIX class "[:abc]b:]". However, for
2632 [:a[:abc]b][b:] it gives unknown POSIX class "[:abc]b][b:]", which does not
2633 seem right at all. PCRE does not allow closing square brackets in POSIX class
2634 names.
2635
2636 Arguments:
2637 ptr pointer to the initial [
2638 endptr where to return the end pointer
2639
2640 Returns: TRUE or FALSE
2641 */
2642
2643 static BOOL
2644 check_posix_syntax(const pcre_uchar *ptr, const pcre_uchar **endptr)
2645 {
2646 int terminator; /* Don't combine these lines; the Solaris cc */
2647 terminator = *(++ptr); /* compiler warns about "non-constant" initializer. */
2648 for (++ptr; *ptr != 0; ptr++)
2649 {
2650 if (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET)
2651 ptr++;
2652 else if (*ptr == CHAR_RIGHT_SQUARE_BRACKET) return FALSE;
2653 else
2654 {
2655 if (*ptr == terminator && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET)
2656 {
2657 *endptr = ptr;
2658 return TRUE;
2659 }
2660 if (*ptr == CHAR_LEFT_SQUARE_BRACKET &&
2661 (ptr[1] == CHAR_COLON || ptr[1] == CHAR_DOT ||
2662 ptr[1] == CHAR_EQUALS_SIGN) &&
2663 check_posix_syntax(ptr, endptr))
2664 return FALSE;
2665 }
2666 }
2667 return FALSE;
2668 }
2669
2670
2671
2672
2673 /*************************************************
2674 * Check POSIX class name *
2675 *************************************************/
2676
2677 /* This function is called to check the name given in a POSIX-style class entry
2678 such as [:alnum:].
2679
2680 Arguments:
2681 ptr points to the first letter
2682 len the length of the name
2683
2684 Returns: a value representing the name, or -1 if unknown
2685 */
2686
2687 static int
2688 check_posix_name(const pcre_uchar *ptr, int len)
2689 {
2690 const char *pn = posix_names;
2691 register int yield = 0;
2692 while (posix_name_lengths[yield] != 0)
2693 {
2694 if (len == posix_name_lengths[yield] &&
2695 STRNCMP_UC_C8(ptr, pn, len) == 0) return yield;
2696 pn += posix_name_lengths[yield] + 1;
2697 yield++;
2698 }
2699 return -1;
2700 }
2701
2702
2703 /*************************************************
2704 * Adjust OP_RECURSE items in repeated group *
2705 *************************************************/
2706
2707 /* OP_RECURSE items contain an offset from the start of the regex to the group
2708 that is referenced. This means that groups can be replicated for fixed
2709 repetition simply by copying (because the recursion is allowed to refer to
2710 earlier groups that are outside the current group). However, when a group is
2711 optional (i.e. the minimum quantifier is zero), OP_BRAZERO or OP_SKIPZERO is
2712 inserted before it, after it has been compiled. This means that any OP_RECURSE
2713 items within it that refer to the group itself or any contained groups have to
2714 have their offsets adjusted. That one of the jobs of this function. Before it
2715 is called, the partially compiled regex must be temporarily terminated with
2716 OP_END.
2717
2718 This function has been extended with the possibility of forward references for
2719 recursions and subroutine calls. It must also check the list of such references
2720 for the group we are dealing with. If it finds that one of the recursions in
2721 the current group is on this list, it adjusts the offset in the list, not the
2722 value in the reference (which is a group number).
2723
2724 Arguments:
2725 group points to the start of the group
2726 adjust the amount by which the group is to be moved
2727 utf TRUE in UTF-8 / UTF-16 mode
2728 cd contains pointers to tables etc.
2729 save_hwm the hwm forward reference pointer at the start of the group
2730
2731 Returns: nothing
2732 */
2733
2734 static void
2735 adjust_recurse(pcre_uchar *group, int adjust, BOOL utf, compile_data *cd,
2736 pcre_uchar *save_hwm)
2737 {
2738 pcre_uchar *ptr = group;
2739
2740 while ((ptr = (pcre_uchar *)find_recurse(ptr, utf)) != NULL)
2741 {
2742 int offset;
2743 pcre_uchar *hc;
2744
2745 /* See if this recursion is on the forward reference list. If so, adjust the
2746 reference. */
2747
2748 for (hc = save_hwm; hc < cd->hwm; hc += LINK_SIZE)
2749 {
2750 offset = GET(hc, 0);
2751 if (cd->start_code + offset == ptr + 1)
2752 {
2753 PUT(hc, 0, offset + adjust);
2754 break;
2755 }
2756 }
2757
2758 /* Otherwise, adjust the recursion offset if it's after the start of this
2759 group. */
2760
2761 if (hc >= cd->hwm)
2762 {
2763 offset = GET(ptr, 1);
2764 if (cd->start_code + offset >= group) PUT(ptr, 1, offset + adjust);
2765 }
2766
2767 ptr += 1 + LINK_SIZE;
2768 }
2769 }
2770
2771
2772
2773 /*************************************************
2774 * Insert an automatic callout point *
2775 *************************************************/
2776
2777 /* This function is called when the PCRE_AUTO_CALLOUT option is set, to insert
2778 callout points before each pattern item.
2779
2780 Arguments:
2781 code current code pointer
2782 ptr current pattern pointer
2783 cd pointers to tables etc
2784
2785 Returns: new code pointer
2786 */
2787
2788 static pcre_uchar *
2789 auto_callout(pcre_uchar *code, const pcre_uchar *ptr, compile_data *cd)
2790 {
2791 *code++ = OP_CALLOUT;
2792 *code++ = 255;
2793 PUT(code, 0, (int)(ptr - cd->start_pattern)); /* Pattern offset */
2794 PUT(code, LINK_SIZE, 0); /* Default length */
2795 return code + 2 * LINK_SIZE;
2796 }
2797
2798
2799
2800 /*************************************************
2801 * Complete a callout item *
2802 *************************************************/
2803
2804 /* A callout item contains the length of the next item in the pattern, which
2805 we can't fill in till after we have reached the relevant point. This is used
2806 for both automatic and manual callouts.
2807
2808 Arguments:
2809 previous_callout points to previous callout item
2810 ptr current pattern pointer
2811 cd pointers to tables etc
2812
2813 Returns: nothing
2814 */
2815
2816 static void
2817 complete_callout(pcre_uchar *previous_callout, const pcre_uchar *ptr, compile_data *cd)
2818 {
2819 int length = (int)(ptr - cd->start_pattern - GET(previous_callout, 2));
2820 PUT(previous_callout, 2 + LINK_SIZE, length);
2821 }
2822
2823
2824
2825 #ifdef SUPPORT_UCP
2826 /*************************************************
2827 * Get othercase range *
2828 *************************************************/
2829
2830 /* This function is passed the start and end of a class range, in UTF-8 mode
2831 with UCP support. It searches up the characters, looking for internal ranges of
2832 characters in the "other" case. Each call returns the next one, updating the
2833 start address.
2834
2835 Arguments:
2836 cptr points to starting character value; updated
2837 d end value
2838 ocptr where to put start of othercase range
2839 odptr where to put end of othercase range
2840
2841 Yield: TRUE when range returned; FALSE when no more
2842 */
2843
2844 static BOOL
2845 get_othercase_range(unsigned int *cptr, unsigned int d, unsigned int *ocptr,
2846 unsigned int *odptr)
2847 {
2848 unsigned int c, othercase, next;
2849
2850 for (c = *cptr; c <= d; c++)
2851 { if ((othercase = UCD_OTHERCASE(c)) != c) break; }
2852
2853 if (c > d) return FALSE;
2854
2855 *ocptr = othercase;
2856 next = othercase + 1;
2857
2858 for (++c; c <= d; c++)
2859 {
2860 if (UCD_OTHERCASE(c) != next) break;
2861 next++;
2862 }
2863
2864 *odptr = next - 1;
2865 *cptr = c;
2866
2867 return TRUE;
2868 }
2869
2870
2871
2872 /*************************************************
2873 * Check a character and a property *
2874 *************************************************/
2875
2876 /* This function is called by check_auto_possessive() when a property item
2877 is adjacent to a fixed character.
2878
2879 Arguments:
2880 c the character
2881 ptype the property type
2882 pdata the data for the type
2883 negated TRUE if it's a negated property (\P or \p{^)
2884
2885 Returns: TRUE if auto-possessifying is OK
2886 */
2887
2888 static BOOL
2889 check_char_prop(int c, int ptype, int pdata, BOOL negated)
2890 {
2891 const ucd_record *prop = GET_UCD(c);
2892 switch(ptype)
2893 {
2894 case PT_LAMP:
2895 return (prop->chartype == ucp_Lu ||
2896 prop->chartype == ucp_Ll ||
2897 prop->chartype == ucp_Lt) == negated;
2898
2899 case PT_GC:
2900 return (pdata == PRIV(ucp_gentype)[prop->chartype]) == negated;
2901
2902 case PT_PC:
2903 return (pdata == prop->chartype) == negated;
2904
2905 case PT_SC:
2906 return (pdata == prop->script) == negated;
2907
2908 /* These are specials */
2909
2910 case PT_ALNUM:
2911 return (PRIV(ucp_gentype)[prop->chartype] == ucp_L ||
2912 PRIV(ucp_gentype)[prop->chartype] == ucp_N) == negated;
2913
2914 case PT_SPACE: /* Perl space */
2915 return (PRIV(ucp_gentype)[prop->chartype] == ucp_Z ||
2916 c == CHAR_HT || c == CHAR_NL || c == CHAR_FF || c == CHAR_CR)
2917 == negated;
2918
2919 case PT_PXSPACE: /* POSIX space */
2920 return (PRIV(ucp_gentype)[prop->chartype] == ucp_Z ||
2921 c == CHAR_HT || c == CHAR_NL || c == CHAR_VT ||
2922 c == CHAR_FF || c == CHAR_CR)
2923 == negated;
2924
2925 case PT_WORD:
2926 return (PRIV(ucp_gentype)[prop->chartype] == ucp_L ||
2927 PRIV(ucp_gentype)[prop->chartype] == ucp_N ||
2928 c == CHAR_UNDERSCORE) == negated;
2929 }
2930 return FALSE;
2931 }
2932 #endif /* SUPPORT_UCP */
2933
2934
2935
2936 /*************************************************
2937 * Check if auto-possessifying is possible *
2938 *************************************************/
2939
2940 /* This function is called for unlimited repeats of certain items, to see
2941 whether the next thing could possibly match the repeated item. If not, it makes
2942 sense to automatically possessify the repeated item.
2943
2944 Arguments:
2945 previous pointer to the repeated opcode
2946 utf TRUE in UTF-8 / UTF-16 mode
2947 ptr next character in pattern
2948 options options bits
2949 cd contains pointers to tables etc.
2950
2951 Returns: TRUE if possessifying is wanted
2952 */
2953
2954 static BOOL
2955 check_auto_possessive(const pcre_uchar *previous, BOOL utf,
2956 const pcre_uchar *ptr, int options, compile_data *cd)
2957 {
2958 pcre_int32 c, next;
2959 int op_code = *previous++;
2960
2961 /* Skip whitespace and comments in extended mode */
2962
2963 if ((options & PCRE_EXTENDED) != 0)
2964 {
2965 for (;;)
2966 {
2967 while (MAX_255(*ptr) && (cd->ctypes[*ptr] & ctype_space) != 0) ptr++;
2968 if (*ptr == CHAR_NUMBER_SIGN)
2969 {
2970 ptr++;
2971 while (*ptr != 0)
2972 {
2973 if (IS_NEWLINE(ptr)) { ptr += cd->nllen; break; }
2974 ptr++;
2975 #ifdef SUPPORT_UTF
2976 if (utf) FORWARDCHAR(ptr);
2977 #endif
2978 }
2979 }
2980 else break;
2981 }
2982 }
2983
2984 /* If the next item is one that we can handle, get its value. A non-negative
2985 value is a character, a negative value is an escape value. */
2986
2987 if (*ptr == CHAR_BACKSLASH)
2988 {
2989 int temperrorcode = 0;
2990 next = check_escape(&ptr, &temperrorcode, cd->bracount, options, FALSE);
2991 if (temperrorcode != 0) return FALSE;
2992 ptr++; /* Point after the escape sequence */
2993 }
2994 else if (!MAX_255(*ptr) || (cd->ctypes[*ptr] & ctype_meta) == 0)
2995 {
2996 #ifdef SUPPORT_UTF
2997 if (utf) { GETCHARINC(next, ptr); } else
2998 #endif
2999 next = *ptr++;
3000 }
3001 else return FALSE;
3002
3003 /* Skip whitespace and comments in extended mode */
3004
3005 if ((options & PCRE_EXTENDED) != 0)
3006 {
3007 for (;;)
3008 {
3009 while (MAX_255(*ptr) && (cd->ctypes[*ptr] & ctype_space) != 0) ptr++;
3010 if (*ptr == CHAR_NUMBER_SIGN)
3011 {
3012 ptr++;
3013 while (*ptr != 0)
3014 {
3015 if (IS_NEWLINE(ptr)) { ptr += cd->nllen; break; }
3016 ptr++;
3017 #ifdef SUPPORT_UTF
3018 if (utf) FORWARDCHAR(ptr);
3019 #endif
3020 }
3021 }
3022 else break;
3023 }
3024 }
3025
3026 /* If the next thing is itself optional, we have to give up. */
3027
3028 if (*ptr == CHAR_ASTERISK || *ptr == CHAR_QUESTION_MARK ||
3029 STRNCMP_UC_C8(ptr, STR_LEFT_CURLY_BRACKET STR_0 STR_COMMA, 3) == 0)
3030 return FALSE;
3031
3032 /* Now compare the next item with the previous opcode. First, handle cases when
3033 the next item is a character. */
3034
3035 if (next >= 0) switch(op_code)
3036 {
3037 case OP_CHAR:
3038 #ifdef SUPPORT_UTF
3039 GETCHARTEST(c, previous);
3040 #else
3041 c = *previous;
3042 #endif
3043 return c != next;
3044
3045 /* For CHARI (caseless character) we must check the other case. If we have
3046 Unicode property support, we can use it to test the other case of
3047 high-valued characters. */
3048
3049 case OP_CHARI:
3050 #ifdef SUPPORT_UTF
3051 GETCHARTEST(c, previous);
3052 #else
3053 c = *previous;
3054 #endif
3055 if (c == next) return FALSE;
3056 #ifdef SUPPORT_UTF
3057 if (utf)
3058 {
3059 unsigned int othercase;
3060 if (next < 128) othercase = cd->fcc[next]; else
3061 #ifdef SUPPORT_UCP
3062 othercase = UCD_OTHERCASE((unsigned int)next);
3063 #else
3064 othercase = NOTACHAR;
3065 #endif
3066 return (unsigned int)c != othercase;
3067 }
3068 else
3069 #endif /* SUPPORT_UTF */
3070 return (c != TABLE_GET((unsigned int)next, cd->fcc, next)); /* Non-UTF-8 mode */
3071
3072 /* For OP_NOT and OP_NOTI, the data is always a single-byte character. These
3073 opcodes are not used for multi-byte characters, because they are coded using
3074 an XCLASS instead. */
3075
3076 case OP_NOT:
3077 return (c = *previous) == next;
3078
3079 case OP_NOTI:
3080 if ((c = *previous) == next) return TRUE;
3081 #ifdef SUPPORT_UTF
3082 if (utf)
3083 {
3084 unsigned int othercase;
3085 if (next < 128) othercase = cd->fcc[next]; else
3086 #ifdef SUPPORT_UCP
3087 othercase = UCD_OTHERCASE(next);
3088 #else
3089 othercase = NOTACHAR;
3090 #endif
3091 return (unsigned int)c == othercase;
3092 }
3093 else
3094 #endif /* SUPPORT_UTF */
3095 return (c == (int)(TABLE_GET((unsigned int)next, cd->fcc, next))); /* Non-UTF-8 mode */
3096
3097 /* Note that OP_DIGIT etc. are generated only when PCRE_UCP is *not* set.
3098 When it is set, \d etc. are converted into OP_(NOT_)PROP codes. */
3099
3100 case OP_DIGIT:
3101 return next > 127 || (cd->ctypes[next] & ctype_digit) == 0;
3102
3103 case OP_NOT_DIGIT:
3104 return next <= 127 && (cd->ctypes[next] & ctype_digit) != 0;
3105
3106 case OP_WHITESPACE:
3107 return next > 127 || (cd->ctypes[next] & ctype_space) == 0;
3108
3109 case OP_NOT_WHITESPACE:
3110 return next <= 127 && (cd->ctypes[next] & ctype_space) != 0;
3111
3112 case OP_WORDCHAR:
3113 return next > 127 || (cd->ctypes[next] & ctype_word) == 0;
3114
3115 case OP_NOT_WORDCHAR:
3116 return next <= 127 && (cd->ctypes[next] & ctype_word) != 0;
3117
3118 case OP_HSPACE:
3119 case OP_NOT_HSPACE:
3120 switch(next)
3121 {
3122 case 0x09:
3123 case 0x20:
3124 case 0xa0:
3125 case 0x1680:
3126 case 0x180e:
3127 case 0x2000:
3128 case 0x2001:
3129 case 0x2002:
3130 case 0x2003:
3131 case 0x2004:
3132 case 0x2005:
3133 case 0x2006:
3134 case 0x2007:
3135 case 0x2008:
3136 case 0x2009:
3137 case 0x200A:
3138 case 0x202f:
3139 case 0x205f:
3140 case 0x3000:
3141 return op_code == OP_NOT_HSPACE;
3142 default:
3143 return op_code != OP_NOT_HSPACE;
3144 }
3145
3146 case OP_ANYNL:
3147 case OP_VSPACE:
3148 case OP_NOT_VSPACE:
3149 switch(next)
3150 {
3151 case 0x0a:
3152 case 0x0b:
3153 case 0x0c:
3154 case 0x0d:
3155 case 0x85:
3156 case 0x2028:
3157 case 0x2029:
3158 return op_code == OP_NOT_VSPACE;
3159 default:
3160 return op_code != OP_NOT_VSPACE;
3161 }
3162
3163 #ifdef SUPPORT_UCP
3164 case OP_PROP:
3165 return check_char_prop(next, previous[0], previous[1], FALSE);
3166
3167 case OP_NOTPROP:
3168 return check_char_prop(next, previous[0], previous[1], TRUE);
3169 #endif
3170
3171 default:
3172 return FALSE;
3173 }
3174
3175
3176 /* Handle the case when the next item is \d, \s, etc. Note that when PCRE_UCP
3177 is set, \d turns into ESC_du rather than ESC_d, etc., so ESC_d etc. are
3178 generated only when PCRE_UCP is *not* set, that is, when only ASCII
3179 characteristics are recognized. Similarly, the opcodes OP_DIGIT etc. are
3180 replaced by OP_PROP codes when PCRE_UCP is set. */
3181
3182 switch(op_code)
3183 {
3184 case OP_CHAR:
3185 case OP_CHARI:
3186 #ifdef SUPPORT_UTF
3187 GETCHARTEST(c, previous);
3188 #else
3189 c = *previous;
3190 #endif
3191 switch(-next)
3192 {
3193 case ESC_d:
3194 return c > 127 || (cd->ctypes[c] & ctype_digit) == 0;
3195
3196 case ESC_D:
3197 return c <= 127 && (cd->ctypes[c] & ctype_digit) != 0;
3198
3199 case ESC_s:
3200 return c > 127 || (cd->ctypes[c] & ctype_space) == 0;
3201
3202 case ESC_S:
3203 return c <= 127 && (cd->ctypes[c] & ctype_space) != 0;
3204
3205 case ESC_w:
3206 return c > 127 || (cd->ctypes[c] & ctype_word) == 0;
3207
3208 case ESC_W:
3209 return c <= 127 && (cd->ctypes[c] & ctype_word) != 0;
3210
3211 case ESC_h:
3212 case ESC_H:
3213 switch(c)
3214 {
3215 case 0x09:
3216 case 0x20:
3217 case 0xa0:
3218 case 0x1680:
3219 case 0x180e:
3220 case 0x2000:
3221 case 0x2001:
3222 case 0x2002:
3223 case 0x2003:
3224 case 0x2004:
3225 case 0x2005:
3226 case 0x2006:
3227 case 0x2007:
3228 case 0x2008:
3229 case 0x2009:
3230 case 0x200A:
3231 case 0x202f:
3232 case 0x205f:
3233 case 0x3000:
3234 return -next != ESC_h;
3235 default:
3236 return -next == ESC_h;
3237 }
3238
3239 case ESC_v:
3240 case ESC_V:
3241 switch(c)
3242 {
3243 case 0x0a:
3244 case 0x0b:
3245 case 0x0c:
3246 case 0x0d:
3247 case 0x85:
3248 case 0x2028:
3249 case 0x2029:
3250 return -next != ESC_v;
3251 default:
3252 return -next == ESC_v;
3253 }
3254
3255 /* When PCRE_UCP is set, these values get generated for \d etc. Find
3256 their substitutions and process them. The result will always be either
3257 -ESC_p or -ESC_P. Then fall through to process those values. */
3258
3259 #ifdef SUPPORT_UCP
3260 case ESC_du:
3261 case ESC_DU:
3262 case ESC_wu:
3263 case ESC_WU:
3264 case ESC_su:
3265 case ESC_SU:
3266 {
3267 int temperrorcode = 0;
3268 ptr = substitutes[-next - ESC_DU];
3269 next = check_escape(&ptr, &temperrorcode, 0, options, FALSE);
3270 if (temperrorcode != 0) return FALSE;
3271 ptr++; /* For compatibility */
3272 }
3273 /* Fall through */
3274
3275 case ESC_p:
3276 case ESC_P:
3277 {
3278 int ptype, pdata, errorcodeptr;
3279 BOOL negated;
3280
3281 ptr--; /* Make ptr point at the p or P */
3282 ptype = get_ucp(&ptr, &negated, &pdata, &errorcodeptr);
3283 if (ptype < 0) return FALSE;
3284 ptr++; /* Point past the final curly ket */
3285
3286 /* If the property item is optional, we have to give up. (When generated
3287 from \d etc by PCRE_UCP, this test will have been applied much earlier,
3288 to the original \d etc. At this point, ptr will point to a zero byte. */
3289
3290 if (*ptr == CHAR_ASTERISK || *ptr == CHAR_QUESTION_MARK ||
3291 STRNCMP_UC_C8(ptr, STR_LEFT_CURLY_BRACKET STR_0 STR_COMMA, 3) == 0)
3292 return FALSE;
3293
3294 /* Do the property check. */
3295
3296 return check_char_prop(c, ptype, pdata, (next == -ESC_P) != negated);
3297 }
3298 #endif
3299
3300 default:
3301 return FALSE;
3302 }
3303
3304 /* In principle, support for Unicode properties should be integrated here as
3305 well. It means re-organizing the above code so as to get hold of the property
3306 values before switching on the op-code. However, I wonder how many patterns
3307 combine ASCII \d etc with Unicode properties? (Note that if PCRE_UCP is set,
3308 these op-codes are never generated.) */
3309
3310 case OP_DIGIT:
3311 return next == -ESC_D || next == -ESC_s || next == -ESC_W ||
3312 next == -ESC_h || next == -ESC_v || next == -ESC_R;
3313
3314 case OP_NOT_DIGIT:
3315 return next == -ESC_d;
3316
3317 case OP_WHITESPACE:
3318 return next == -ESC_S || next == -ESC_d || next == -ESC_w || next == -ESC_R;
3319
3320 case OP_NOT_WHITESPACE:
3321 return next == -ESC_s || next == -ESC_h || next == -ESC_v;
3322
3323 case OP_HSPACE:
3324 return next == -ESC_S || next == -ESC_H || next == -ESC_d ||
3325 next == -ESC_w || next == -ESC_v || next == -ESC_R;
3326
3327 case OP_NOT_HSPACE:
3328 return next == -ESC_h;
3329
3330 /* Can't have \S in here because VT matches \S (Perl anomaly) */
3331 case OP_ANYNL:
3332 case OP_VSPACE:
3333 return next == -ESC_V || next == -ESC_d || next == -ESC_w;
3334
3335 case OP_NOT_VSPACE:
3336 return next == -ESC_v || next == -ESC_R;
3337
3338 case OP_WORDCHAR:
3339 return next == -ESC_W || next == -ESC_s || next == -ESC_h ||
3340 next == -ESC_v || next == -ESC_R;
3341
3342 case OP_NOT_WORDCHAR:
3343 return next == -ESC_w || next == -ESC_d;
3344
3345 default:
3346 return FALSE;
3347 }
3348
3349 /* Control does not reach here */
3350 }
3351
3352
3353
3354 /*************************************************
3355 * Compile one branch *
3356 *************************************************/
3357
3358 /* Scan the pattern, compiling it into the a vector. If the options are
3359 changed during the branch, the pointer is used to change the external options
3360 bits. This function is used during the pre-compile phase when we are trying
3361 to find out the amount of memory needed, as well as during the real compile
3362 phase. The value of lengthptr distinguishes the two phases.
3363
3364 Arguments:
3365 optionsptr pointer to the option bits
3366 codeptr points to the pointer to the current code point
3367 ptrptr points to the current pattern pointer
3368 errorcodeptr points to error code variable
3369 firstcharptr set to initial literal character, or < 0 (REQ_UNSET, REQ_NONE)
3370 reqcharptr set to the last literal character required, else < 0
3371 bcptr points to current branch chain
3372 cond_depth conditional nesting depth
3373 cd contains pointers to tables etc.
3374 lengthptr NULL during the real compile phase
3375 points to length accumulator during pre-compile phase
3376
3377 Returns: TRUE on success
3378 FALSE, with *errorcodeptr set non-zero on error
3379 */
3380
3381 static BOOL
3382 compile_branch(int *optionsptr, pcre_uchar **codeptr,
3383 const pcre_uchar **ptrptr, int *errorcodeptr, pcre_int32 *firstcharptr,
3384 pcre_int32 *reqcharptr, branch_chain *bcptr, int cond_depth,
3385 compile_data *cd, int *lengthptr)
3386 {
3387 int repeat_type, op_type;
3388 int repeat_min = 0, repeat_max = 0; /* To please picky compilers */
3389 int bravalue = 0;
3390 int greedy_default, greedy_non_default;
3391 pcre_int32 firstchar, reqchar;
3392 pcre_int32 zeroreqchar, zerofirstchar;
3393 pcre_int32 req_caseopt, reqvary, tempreqvary;
3394 int options = *optionsptr; /* May change dynamically */
3395 int after_manual_callout = 0;
3396 int length_prevgroup = 0;
3397 register int c;
3398 register pcre_uchar *code = *codeptr;
3399 pcre_uchar *last_code = code;
3400 pcre_uchar *orig_code = code;
3401 pcre_uchar *tempcode;
3402 BOOL inescq = FALSE;
3403 BOOL groupsetfirstchar = FALSE;
3404 const pcre_uchar *ptr = *ptrptr;
3405 const pcre_uchar *tempptr;
3406 const pcre_uchar *nestptr = NULL;
3407 pcre_uchar *previous = NULL;
3408 pcre_uchar *previous_callout = NULL;
3409 pcre_uchar *save_hwm = NULL;
3410 pcre_uint8 classbits[32];
3411
3412 /* We can fish out the UTF-8 setting once and for all into a BOOL, but we
3413 must not do this for other options (e.g. PCRE_EXTENDED) because they may change
3414 dynamically as we process the pattern. */
3415
3416 #ifdef SUPPORT_UTF
3417 /* PCRE_UTF16 has the same value as PCRE_UTF8. */
3418 BOOL utf = (options & PCRE_UTF8) != 0;
3419 pcre_uchar utf_chars[6];
3420 #else
3421 BOOL utf = FALSE;
3422 #endif
3423
3424 /* Helper variables for OP_XCLASS opcode (for characters > 255). */
3425
3426 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
3427 BOOL xclass;
3428 pcre_uchar *class_uchardata;
3429 pcre_uchar *class_uchardata_base;
3430 #endif
3431
3432 #ifdef PCRE_DEBUG
3433 if (lengthptr != NULL) DPRINTF((">> start branch\n"));
3434 #endif
3435
3436 /* Set up the default and non-default settings for greediness */
3437
3438 greedy_default = ((options & PCRE_UNGREEDY) != 0);
3439 greedy_non_default = greedy_default ^ 1;
3440
3441 /* Initialize no first byte, no required byte. REQ_UNSET means "no char
3442 matching encountered yet". It gets changed to REQ_NONE if we hit something that
3443 matches a non-fixed char first char; reqchar just remains unset if we never
3444 find one.
3445
3446 When we hit a repeat whose minimum is zero, we may have to adjust these values
3447 to take the zero repeat into account. This is implemented by setting them to
3448 zerofirstbyte and zeroreqchar when such a repeat is encountered. The individual
3449 item types that can be repeated set these backoff variables appropriately. */
3450
3451 firstchar = reqchar = zerofirstchar = zeroreqchar = REQ_UNSET;
3452
3453 /* The variable req_caseopt contains either the REQ_CASELESS value
3454 or zero, according to the current setting of the caseless flag. The
3455 REQ_CASELESS leaves the lower 28 bit empty. It is added into the
3456 firstchar or reqchar variables to record the case status of the
3457 value. This is used only for ASCII characters. */
3458
3459 req_caseopt = ((options & PCRE_CASELESS) != 0)? REQ_CASELESS:0;
3460
3461 /* Switch on next character until the end of the branch */
3462
3463 for (;; ptr++)
3464 {
3465 BOOL negate_class;
3466 BOOL should_flip_negation;
3467 BOOL possessive_quantifier;
3468 BOOL is_quantifier;
3469 BOOL is_recurse;
3470 BOOL reset_bracount;
3471 int class_has_8bitchar;
3472 int class_single_char;
3473 int newoptions;
3474 int recno;
3475 int refsign;
3476 int skipbytes;
3477 int subreqchar;
3478 int subfirstchar;
3479 int terminator;
3480 int mclength;
3481 int tempbracount;
3482 pcre_uchar mcbuffer[8];
3483
3484 /* Get next character in the pattern */
3485
3486 c = *ptr;
3487
3488 /* If we are at the end of a nested substitution, revert to the outer level
3489 string. Nesting only happens one level deep. */
3490
3491 if (c == 0 && nestptr != NULL)
3492 {
3493 ptr = nestptr;
3494 nestptr = NULL;
3495 c = *ptr;
3496 }
3497
3498 /* If we are in the pre-compile phase, accumulate the length used for the
3499 previous cycle of this loop. */
3500
3501 if (lengthptr != NULL)
3502 {
3503 #ifdef PCRE_DEBUG
3504 if (code > cd->hwm) cd->hwm = code; /* High water info */
3505 #endif
3506 if (code > cd->start_workspace + cd->workspace_size -
3507 WORK_SIZE_SAFETY_MARGIN) /* Check for overrun */
3508 {
3509 *errorcodeptr = ERR52;
3510 goto FAILED;
3511 }
3512
3513 /* There is at least one situation where code goes backwards: this is the
3514 case of a zero quantifier after a class (e.g. [ab]{0}). At compile time,
3515 the class is simply eliminated. However, it is created first, so we have to
3516 allow memory for it. Therefore, don't ever reduce the length at this point.
3517 */
3518
3519 if (code < last_code) code = last_code;
3520
3521 /* Paranoid check for integer overflow */
3522
3523 if (OFLOW_MAX - *lengthptr < code - last_code)
3524 {
3525 *errorcodeptr = ERR20;
3526 goto FAILED;
3527 }
3528
3529 *lengthptr += (int)(code - last_code);
3530 DPRINTF(("length=%d added %d c=%c (0x%x)\n", *lengthptr,
3531 (int)(code - last_code), c, c));
3532
3533 /* If "previous" is set and it is not at the start of the work space, move
3534 it back to there, in order to avoid filling up the work space. Otherwise,
3535 if "previous" is NULL, reset the current code pointer to the start. */
3536
3537 if (previous != NULL)
3538 {
3539 if (previous > orig_code)
3540 {
3541 memmove(orig_code, previous, IN_UCHARS(code - previous));
3542 code -= previous - orig_code;
3543 previous = orig_code;
3544 }
3545 }
3546 else code = orig_code;
3547
3548 /* Remember where this code item starts so we can pick up the length
3549 next time round. */
3550
3551 last_code = code;
3552 }
3553
3554 /* In the real compile phase, just check the workspace used by the forward
3555 reference list. */
3556
3557 else if (cd->hwm > cd->start_workspace + cd->workspace_size -
3558 WORK_SIZE_SAFETY_MARGIN)
3559 {
3560 *errorcodeptr = ERR52;
3561 goto FAILED;
3562 }
3563
3564 /* If in \Q...\E, check for the end; if not, we have a literal */
3565
3566 if (inescq && c != 0)
3567 {
3568 if (c == CHAR_BACKSLASH && ptr[1] == CHAR_E)
3569 {
3570 inescq = FALSE;
3571 ptr++;
3572 continue;
3573 }
3574 else
3575 {
3576 if (previous_callout != NULL)
3577 {
3578 if (lengthptr == NULL) /* Don't attempt in pre-compile phase */
3579 complete_callout(previous_callout, ptr, cd);
3580 previous_callout = NULL;
3581 }
3582 if ((options & PCRE_AUTO_CALLOUT) != 0)
3583 {
3584 previous_callout = code;
3585 code = auto_callout(code, ptr, cd);
3586 }
3587 goto NORMAL_CHAR;
3588 }
3589 }
3590
3591 /* Fill in length of a previous callout, except when the next thing is
3592 a quantifier. */
3593
3594 is_quantifier =
3595 c == CHAR_ASTERISK || c == CHAR_PLUS || c == CHAR_QUESTION_MARK ||
3596 (c == CHAR_LEFT_CURLY_BRACKET && is_counted_repeat(ptr+1));
3597
3598 if (!is_quantifier && previous_callout != NULL &&
3599 after_manual_callout-- <= 0)
3600 {
3601 if (lengthptr == NULL) /* Don't attempt in pre-compile phase */
3602 complete_callout(previous_callout, ptr, cd);
3603 previous_callout = NULL;
3604 }
3605
3606 /* In extended mode, skip white space and comments. */
3607
3608 if ((options & PCRE_EXTENDED) != 0)
3609 {
3610 if (MAX_255(*ptr) && (cd->ctypes[c] & ctype_space) != 0) continue;
3611 if (c == CHAR_NUMBER_SIGN)
3612 {
3613 ptr++;
3614 while (*ptr != 0)
3615 {
3616 if (IS_NEWLINE(ptr)) { ptr += cd->nllen - 1; break; }
3617 ptr++;
3618 #ifdef SUPPORT_UTF
3619 if (utf) FORWARDCHAR(ptr);
3620 #endif
3621 }
3622 if (*ptr != 0) continue;
3623
3624 /* Else fall through to handle end of string */
3625 c = 0;
3626 }
3627 }
3628
3629 /* No auto callout for quantifiers. */
3630
3631 if ((options & PCRE_AUTO_CALLOUT) != 0 && !is_quantifier)
3632 {
3633 previous_callout = code;
3634 code = auto_callout(code, ptr, cd);
3635 }
3636
3637 switch(c)
3638 {
3639 /* ===================================================================*/
3640 case 0: /* The branch terminates at string end */
3641 case CHAR_VERTICAL_LINE: /* or | or ) */
3642 case CHAR_RIGHT_PARENTHESIS:
3643 *firstcharptr = firstchar;
3644 *reqcharptr = reqchar;
3645 *codeptr = code;
3646 *ptrptr = ptr;
3647 if (lengthptr != NULL)
3648 {
3649 if (OFLOW_MAX - *lengthptr < code - last_code)
3650 {
3651 *errorcodeptr = ERR20;
3652 goto FAILED;
3653 }
3654 *lengthptr += (int)(code - last_code); /* To include callout length */
3655 DPRINTF((">> end branch\n"));
3656 }
3657 return TRUE;
3658
3659
3660 /* ===================================================================*/
3661 /* Handle single-character metacharacters. In multiline mode, ^ disables
3662 the setting of any following char as a first character. */
3663
3664 case CHAR_CIRCUMFLEX_ACCENT:
3665 previous = NULL;
3666 if ((options & PCRE_MULTILINE) != 0)
3667 {
3668 if (firstchar == REQ_UNSET) firstchar = REQ_NONE;
3669 *code++ = OP_CIRCM;
3670 }
3671 else *code++ = OP_CIRC;
3672 break;
3673
3674 case CHAR_DOLLAR_SIGN:
3675 previous = NULL;
3676 *code++ = ((options & PCRE_MULTILINE) != 0)? OP_DOLLM : OP_DOLL;
3677 break;
3678
3679 /* There can never be a first char if '.' is first, whatever happens about
3680 repeats. The value of reqchar doesn't change either. */
3681
3682 case CHAR_DOT:
3683 if (firstchar == REQ_UNSET) firstchar = REQ_NONE;
3684 zerofirstchar = firstchar;
3685 zeroreqchar = reqchar;
3686 previous = code;
3687 *code++ = ((options & PCRE_DOTALL) != 0)? OP_ALLANY: OP_ANY;
3688 break;
3689
3690
3691 /* ===================================================================*/
3692 /* Character classes. If the included characters are all < 256, we build a
3693 32-byte bitmap of the permitted characters, except in the special case
3694 where there is only one such character. For negated classes, we build the
3695 map as usual, then invert it at the end. However, we use a different opcode
3696 so that data characters > 255 can be handled correctly.
3697
3698 If the class contains characters outside the 0-255 range, a different
3699 opcode is compiled. It may optionally have a bit map for characters < 256,
3700 but those above are are explicitly listed afterwards. A flag byte tells
3701 whether the bitmap is present, and whether this is a negated class or not.
3702
3703 In JavaScript compatibility mode, an isolated ']' causes an error. In
3704 default (Perl) mode, it is treated as a data character. */
3705
3706 case CHAR_RIGHT_SQUARE_BRACKET:
3707 if ((cd->external_options & PCRE_JAVASCRIPT_COMPAT) != 0)
3708 {
3709 *errorcodeptr = ERR64;
3710 goto FAILED;
3711 }
3712 goto NORMAL_CHAR;
3713
3714 case CHAR_LEFT_SQUARE_BRACKET:
3715 previous = code;
3716
3717 /* PCRE supports POSIX class stuff inside a class. Perl gives an error if
3718 they are encountered at the top level, so we'll do that too. */
3719
3720 if ((ptr[1] == CHAR_COLON || ptr[1] == CHAR_DOT ||
3721 ptr[1] == CHAR_EQUALS_SIGN) &&
3722 check_posix_syntax(ptr, &tempptr))
3723 {
3724 *errorcodeptr = (ptr[1] == CHAR_COLON)? ERR13 : ERR31;
3725 goto FAILED;
3726 }
3727
3728 /* If the first character is '^', set the negation flag and skip it. Also,
3729 if the first few characters (either before or after ^) are \Q\E or \E we
3730 skip them too. This makes for compatibility with Perl. */
3731
3732 negate_class = FALSE;
3733 for (;;)
3734 {
3735 c = *(++ptr);
3736 if (c == CHAR_BACKSLASH)
3737 {
3738 if (ptr[1] == CHAR_E)
3739 ptr++;
3740 else if (STRNCMP_UC_C8(ptr + 1, STR_Q STR_BACKSLASH STR_E, 3) == 0)
3741 ptr += 3;
3742 else
3743 break;
3744 }
3745 else if (!negate_class && c == CHAR_CIRCUMFLEX_ACCENT)
3746 negate_class = TRUE;
3747 else break;
3748 }
3749
3750 /* Empty classes are allowed in JavaScript compatibility mode. Otherwise,
3751 an initial ']' is taken as a data character -- the code below handles
3752 that. In JS mode, [] must always fail, so generate OP_FAIL, whereas
3753 [^] must match any character, so generate OP_ALLANY. */
3754
3755 if (c == CHAR_RIGHT_SQUARE_BRACKET &&
3756 (cd->external_options & PCRE_JAVASCRIPT_COMPAT) != 0)
3757 {
3758 *code++ = negate_class? OP_ALLANY : OP_FAIL;
3759 if (firstchar == REQ_UNSET) firstchar = REQ_NONE;
3760 zerofirstchar = firstchar;
3761 break;
3762 }
3763
3764 /* If a class contains a negative special such as \S, we need to flip the
3765 negation flag at the end, so that support for characters > 255 works
3766 correctly (they are all included in the class). */
3767
3768 should_flip_negation = FALSE;
3769
3770 /* For optimization purposes, we track some properties of the class.
3771 class_has_8bitchar will be non-zero, if the class contains at least one
3772 < 256 character. class_single_char will be 1 if the class contains only
3773 a single character. */
3774
3775 class_has_8bitchar = 0;
3776 class_single_char = 0;
3777
3778 /* Initialize the 32-char bit map to all zeros. We build the map in a
3779 temporary bit of memory, in case the class contains only 1 character (less
3780 than 256), because in that case the compiled code doesn't use the bit map.
3781 */
3782
3783 memset(classbits, 0, 32 * sizeof(pcre_uint8));
3784
3785 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
3786 xclass = FALSE; /* No chars >= 256 */
3787 class_uchardata = code + LINK_SIZE + 2; /* For UTF-8 items */
3788 class_uchardata_base = class_uchardata; /* For resetting in pass 1 */
3789 #endif
3790
3791 /* Process characters until ] is reached. By writing this as a "do" it
3792 means that an initial ] is taken as a data character. At the start of the
3793 loop, c contains the first byte of the character. */
3794
3795 if (c != 0) do
3796 {
3797 const pcre_uchar *oldptr;
3798
3799 #ifdef SUPPORT_UTF
3800 if (utf && HAS_EXTRALEN(c))
3801 { /* Braces are required because the */
3802 GETCHARLEN(c, ptr, ptr); /* macro generates multiple statements */
3803 }
3804 #endif
3805
3806 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
3807 /* In the pre-compile phase, accumulate the length of any extra
3808 data and reset the pointer. This is so that very large classes that
3809 contain a zillion > 255 characters no longer overwrite the work space
3810 (which is on the stack). */
3811
3812 if (lengthptr != NULL)
3813 {
3814 *lengthptr += class_uchardata - class_uchardata_base;
3815 class_uchardata = class_uchardata_base;
3816 }
3817 #endif
3818
3819 /* Inside \Q...\E everything is literal except \E */
3820
3821 if (inescq)
3822 {
3823 if (c == CHAR_BACKSLASH && ptr[1] == CHAR_E) /* If we are at \E */
3824 {
3825 inescq = FALSE; /* Reset literal state */
3826 ptr++; /* Skip the 'E' */
3827 continue; /* Carry on with next */
3828 }
3829 goto CHECK_RANGE; /* Could be range if \E follows */
3830 }
3831
3832 /* Handle POSIX class names. Perl allows a negation extension of the
3833 form [:^name:]. A square bracket that doesn't match the syntax is
3834 treated as a literal. We also recognize the POSIX constructions
3835 [.ch.] and [=ch=] ("collating elements") and fault them, as Perl
3836 5.6 and 5.8 do. */
3837
3838 if (c == CHAR_LEFT_SQUARE_BRACKET &&
3839 (ptr[1] == CHAR_COLON || ptr[1] == CHAR_DOT ||
3840 ptr[1] == CHAR_EQUALS_SIGN) && check_posix_syntax(ptr, &tempptr))
3841 {
3842 BOOL local_negate = FALSE;
3843 int posix_class, taboffset, tabopt;
3844 register const pcre_uint8 *cbits = cd->cbits;
3845 pcre_uint8 pbits[32];
3846
3847 if (ptr[1] != CHAR_COLON)
3848 {
3849 *errorcodeptr = ERR31;
3850 goto FAILED;
3851 }
3852
3853 ptr += 2;
3854 if (*ptr == CHAR_CIRCUMFLEX_ACCENT)
3855 {
3856 local_negate = TRUE;
3857 should_flip_negation = TRUE; /* Note negative special */
3858 ptr++;
3859 }
3860
3861 posix_class = check_posix_name(ptr, (int)(tempptr - ptr));
3862 if (posix_class < 0)
3863 {
3864 *errorcodeptr = ERR30;
3865 goto FAILED;
3866 }
3867
3868 /* If matching is caseless, upper and lower are converted to
3869 alpha. This relies on the fact that the class table starts with
3870 alpha, lower, upper as the first 3 entries. */
3871
3872 if ((options & PCRE_CASELESS) != 0 && posix_class <= 2)
3873 posix_class = 0;
3874
3875 /* When PCRE_UCP is set, some of the POSIX classes are converted to
3876 different escape sequences that use Unicode properties. */
3877
3878 #ifdef SUPPORT_UCP
3879 if ((options & PCRE_UCP) != 0)
3880 {
3881 int pc = posix_class + ((local_negate)? POSIX_SUBSIZE/2 : 0);
3882 if (posix_substitutes[pc] != NULL)
3883 {
3884 nestptr = tempptr + 1;
3885 ptr = posix_substitutes[pc] - 1;
3886 continue;
3887 }
3888 }
3889 #endif
3890 /* In the non-UCP case, we build the bit map for the POSIX class in a
3891 chunk of local store because we may be adding and subtracting from it,
3892 and we don't want to subtract bits that may be in the main map already.
3893 At the end we or the result into the bit map that is being built. */
3894
3895 posix_class *= 3;
3896
3897 /* Copy in the first table (always present) */
3898
3899 memcpy(pbits, cbits + posix_class_maps[posix_class],
3900 32 * sizeof(pcre_uint8));
3901
3902 /* If there is a second table, add or remove it as required. */
3903
3904 taboffset = posix_class_maps[posix_class + 1];
3905 tabopt = posix_class_maps[posix_class + 2];
3906
3907 if (taboffset >= 0)
3908 {
3909 if (tabopt >= 0)
3910 for (c = 0; c < 32; c++) pbits[c] |= cbits[c + taboffset];
3911 else
3912 for (c = 0; c < 32; c++) pbits[c] &= ~cbits[c + taboffset];
3913 }
3914
3915 /* Not see if we need to remove any special characters. An option
3916 value of 1 removes vertical space and 2 removes underscore. */
3917
3918 if (tabopt < 0) tabopt = -tabopt;
3919 if (tabopt == 1) pbits[1] &= ~0x3c;
3920 else if (tabopt == 2) pbits[11] &= 0x7f;
3921
3922 /* Add the POSIX table or its complement into the main table that is
3923 being built and we are done. */
3924
3925 if (local_negate)
3926 for (c = 0; c < 32; c++) classbits[c] |= ~pbits[c];
3927 else
3928 for (c = 0; c < 32; c++) classbits[c] |= pbits[c];
3929
3930 ptr = tempptr + 1;
3931 /* Every class contains at least one < 256 characters. */
3932 class_has_8bitchar = 1;
3933 /* Every class contains at least two characters. */
3934 class_single_char = 2;
3935 continue; /* End of POSIX syntax handling */
3936 }
3937
3938 /* Backslash may introduce a single character, or it may introduce one
3939 of the specials, which just set a flag. The sequence \b is a special
3940 case. Inside a class (and only there) it is treated as backspace. We
3941 assume that other escapes have more than one character in them, so
3942 speculatively set both class_has_8bitchar and class_single_char bigger
3943 than one. Unrecognized escapes fall through and are either treated
3944 as literal characters (by default), or are faulted if
3945 PCRE_EXTRA is set. */
3946
3947 if (c == CHAR_BACKSLASH)
3948 {
3949 c = check_escape(&ptr, errorcodeptr, cd->bracount, options, TRUE);
3950 if (*errorcodeptr != 0) goto FAILED;
3951
3952 if (-c == ESC_b) c = CHAR_BS; /* \b is backspace in a class */
3953 else if (-c == ESC_N) /* \N is not supported in a class */
3954 {
3955 *errorcodeptr = ERR71;
3956 goto FAILED;
3957 }
3958 else if (-c == ESC_Q) /* Handle start of quoted string */
3959 {
3960 if (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
3961 {
3962 ptr += 2; /* avoid empty string */
3963 }
3964 else inescq = TRUE;
3965 continue;
3966 }
3967 else if (-c == ESC_E) continue; /* Ignore orphan \E */
3968
3969 if (c < 0)
3970 {
3971 register const pcre_uint8 *cbits = cd->cbits;
3972 /* Every class contains at least two < 256 characters. */
3973 class_has_8bitchar++;
3974 /* Every class contains at least two characters. */
3975 class_single_char += 2;
3976
3977 switch (-c)
3978 {
3979 #ifdef SUPPORT_UCP
3980 case ESC_du: /* These are the values given for \d etc */
3981 case ESC_DU: /* when PCRE_UCP is set. We replace the */
3982 case ESC_wu: /* escape sequence with an appropriate \p */
3983 case ESC_WU: /* or \P to test Unicode properties instead */
3984 case ESC_su: /* of the default ASCII testing. */
3985 case ESC_SU:
3986 nestptr = ptr;
3987 ptr = substitutes[-c - ESC_DU] - 1; /* Just before substitute */
3988 class_has_8bitchar--; /* Undo! */
3989 continue;
3990 #endif
3991 case ESC_d:
3992 for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_digit];
3993 continue;
3994
3995 case ESC_D:
3996 should_flip_negation = TRUE;
3997 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_digit];
3998 continue;
3999
4000 case ESC_w:
4001 for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_word];
4002 continue;
4003
4004 case ESC_W:
4005 should_flip_negation = TRUE;
4006 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_word];
4007 continue;
4008
4009 /* Perl 5.004 onwards omits VT from \s, but we must preserve it
4010 if it was previously set by something earlier in the character
4011 class. */
4012
4013 case ESC_s:
4014 classbits[0] |= cbits[cbit_space];
4015 classbits[1] |= cbits[cbit_space+1] & ~0x08;
4016 for (c = 2; c < 32; c++) classbits[c] |= cbits[c+cbit_space];
4017 continue;
4018
4019 case ESC_S:
4020 should_flip_negation = TRUE;
4021 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_space];
4022 classbits[1] |= 0x08; /* Perl 5.004 onwards omits VT from \s */
4023 continue;
4024
4025 case ESC_h:
4026 SETBIT(classbits, 0x09); /* VT */
4027 SETBIT(classbits, 0x20); /* SPACE */
4028 SETBIT(classbits, 0xa0); /* NSBP */
4029 #ifndef COMPILE_PCRE8
4030 xclass = TRUE;
4031 *class_uchardata++ = XCL_SINGLE;
4032 *class_uchardata++ = 0x1680;
4033 *class_uchardata++ = XCL_SINGLE;
4034 *class_uchardata++ = 0x180e;
4035 *class_uchardata++ = XCL_RANGE;
4036 *class_uchardata++ = 0x2000;
4037 *class_uchardata++ = 0x200a;
4038 *class_uchardata++ = XCL_SINGLE;
4039 *class_uchardata++ = 0x202f;
4040 *class_uchardata++ = XCL_SINGLE;
4041 *class_uchardata++ = 0x205f;
4042 *class_uchardata++ = XCL_SINGLE;
4043 *class_uchardata++ = 0x3000;
4044 #elif defined SUPPORT_UTF
4045 if (utf)
4046 {
4047 xclass = TRUE;
4048 *class_uchardata++ = XCL_SINGLE;
4049 class_uchardata += PRIV(ord2utf)(0x1680, class_uchardata);
4050 *class_uchardata++ = XCL_SINGLE;
4051 class_uchardata += PRIV(ord2utf)(0x180e, class_uchardata);
4052 *class_uchardata++ = XCL_RANGE;
4053 class_uchardata += PRIV(ord2utf)(0x2000, class_uchardata);
4054 class_uchardata += PRIV(ord2utf)(0x200a, class_uchardata);
4055 *class_uchardata++ = XCL_SINGLE;
4056 class_uchardata += PRIV(ord2utf)(0x202f, class_uchardata);
4057 *class_uchardata++ = XCL_SINGLE;
4058 class_uchardata += PRIV(ord2utf)(0x205f, class_uchardata);
4059 *class_uchardata++ = XCL_SINGLE;
4060 class_uchardata += PRIV(ord2utf)(0x3000, class_uchardata);
4061 }
4062 #endif
4063 continue;
4064
4065 case ESC_H:
4066 for (c = 0; c < 32; c++)
4067 {
4068 int x = 0xff;
4069 switch (c)
4070 {
4071 case 0x09/8: x ^= 1 << (0x09%8); break;
4072 case 0x20/8: x ^= 1 << (0x20%8); break;
4073 case 0xa0/8: x ^= 1 << (0xa0%8); break;
4074 default: break;
4075 }
4076 classbits[c] |= x;
4077 }
4078 #ifndef COMPILE_PCRE8
4079 xclass = TRUE;
4080 *class_uchardata++ = XCL_RANGE;
4081 *class_uchardata++ = 0x0100;
4082 *class_uchardata++ = 0x167f;
4083 *class_uchardata++ = XCL_RANGE;
4084 *class_uchardata++ = 0x1681;
4085 *class_uchardata++ = 0x180d;
4086 *class_uchardata++ = XCL_RANGE;
4087 *class_uchardata++ = 0x180f;
4088 *class_uchardata++ = 0x1fff;
4089 *class_uchardata++ = XCL_RANGE;
4090 *class_uchardata++ = 0x200b;
4091 *class_uchardata++ = 0x202e;
4092 *class_uchardata++ = XCL_RANGE;
4093 *class_uchardata++ = 0x2030;
4094 *class_uchardata++ = 0x205e;
4095 *class_uchardata++ = XCL_RANGE;
4096 *class_uchardata++ = 0x2060;
4097 *class_uchardata++ = 0x2fff;
4098 *class_uchardata++ = XCL_RANGE;
4099 *class_uchardata++ = 0x3001;
4100 #ifdef SUPPORT_UTF
4101 if (utf)
4102 class_uchardata += PRIV(ord2utf)(0x10ffff, class_uchardata);
4103 else
4104 #endif
4105 *class_uchardata++ = 0xffff;
4106 #elif defined SUPPORT_UTF
4107 if (utf)
4108 {
4109 xclass = TRUE;
4110 *class_uchardata++ = XCL_RANGE;
4111 class_uchardata += PRIV(ord2utf)(0x0100, class_uchardata);
4112 class_uchardata += PRIV(ord2utf)(0x167f, class_uchardata);
4113 *class_uchardata++ = XCL_RANGE;
4114 class_uchardata += PRIV(ord2utf)(0x1681, class_uchardata);
4115 class_uchardata += PRIV(ord2utf)(0x180d, class_uchardata);
4116 *class_uchardata++ = XCL_RANGE;
4117 class_uchardata += PRIV(ord2utf)(0x180f, class_uchardata);
4118 class_uchardata += PRIV(ord2utf)(0x1fff, class_uchardata);
4119 *class_uchardata++ = XCL_RANGE;
4120 class_uchardata += PRIV(ord2utf)(0x200b, class_uchardata);
4121 class_uchardata += PRIV(ord2utf)(0x202e, class_uchardata);
4122 *class_uchardata++ = XCL_RANGE;
4123 class_uchardata += PRIV(ord2utf)(0x2030, class_uchardata);
4124 class_uchardata += PRIV(ord2utf)(0x205e, class_uchardata);
4125 *class_uchardata++ = XCL_RANGE;
4126 class_uchardata += PRIV(ord2utf)(0x2060, class_uchardata);
4127 class_uchardata += PRIV(ord2utf)(0x2fff, class_uchardata);
4128 *class_uchardata++ = XCL_RANGE;
4129 class_uchardata += PRIV(ord2utf)(0x3001, class_uchardata);
4130 class_uchardata += PRIV(ord2utf)(0x10ffff, class_uchardata);
4131 }
4132 #endif
4133 continue;
4134
4135 case ESC_v:
4136 SETBIT(classbits, 0x0a); /* LF */
4137 SETBIT(classbits, 0x0b); /* VT */
4138 SETBIT(classbits, 0x0c); /* FF */
4139 SETBIT(classbits, 0x0d); /* CR */
4140 SETBIT(classbits, 0x85); /* NEL */
4141 #ifndef COMPILE_PCRE8
4142 xclass = TRUE;
4143 *class_uchardata++ = XCL_RANGE;
4144 *class_uchardata++ = 0x2028;
4145 *class_uchardata++ = 0x2029;
4146 #elif defined SUPPORT_UTF
4147 if (utf)
4148 {
4149 xclass = TRUE;
4150 *class_uchardata++ = XCL_RANGE;
4151 class_uchardata += PRIV(ord2utf)(0x2028, class_uchardata);
4152 class_uchardata += PRIV(ord2utf)(0x2029, class_uchardata);
4153 }
4154 #endif
4155 continue;
4156
4157 case ESC_V:
4158 for (c = 0; c < 32; c++)
4159 {
4160 int x = 0xff;
4161 switch (c)
4162 {
4163 case 0x0a/8: x ^= 1 << (0x0a%8);
4164 x ^= 1 << (0x0b%8);
4165 x ^= 1 << (0x0c%8);
4166 x ^= 1 << (0x0d%8);
4167 break;
4168 case 0x85/8: x ^= 1 << (0x85%8); break;
4169 default: break;
4170 }
4171 classbits[c] |= x;
4172 }
4173
4174 #ifndef COMPILE_PCRE8
4175 xclass = TRUE;
4176 *class_uchardata++ = XCL_RANGE;
4177 *class_uchardata++ = 0x0100;
4178 *class_uchardata++ = 0x2027;
4179 *class_uchardata++ = XCL_RANGE;
4180 *class_uchardata++ = 0x202a;
4181 #ifdef SUPPORT_UTF
4182 if (utf)
4183 class_uchardata += PRIV(ord2utf)(0x10ffff, class_uchardata);
4184 else
4185 #endif
4186 *class_uchardata++ = 0xffff;
4187 #elif defined SUPPORT_UTF
4188 if (utf)
4189 {
4190 xclass = TRUE;
4191 *class_uchardata++ = XCL_RANGE;
4192 class_uchardata += PRIV(ord2utf)(0x0100, class_uchardata);
4193 class_uchardata += PRIV(ord2utf)(0x2027, class_uchardata);
4194 *class_uchardata++ = XCL_RANGE;
4195 class_uchardata += PRIV(ord2utf)(0x202a, class_uchardata);
4196 class_uchardata += PRIV(ord2utf)(0x10ffff, class_uchardata);
4197 }
4198 #endif
4199 continue;
4200
4201 #ifdef SUPPORT_UCP
4202 case ESC_p:
4203 case ESC_P:
4204 {
4205 BOOL negated;
4206 int pdata;
4207 int ptype = get_ucp(&ptr, &negated, &pdata, errorcodeptr);
4208 if (ptype < 0) goto FAILED;
4209 xclass = TRUE;
4210 *class_uchardata++ = ((-c == ESC_p) != negated)?
4211 XCL_PROP : XCL_NOTPROP;
4212 *class_uchardata++ = ptype;
4213 *class_uchardata++ = pdata;
4214 class_has_8bitchar--; /* Undo! */
4215 continue;
4216 }
4217 #endif
4218 /* Unrecognized escapes are faulted if PCRE is running in its
4219 strict mode. By default, for compatibility with Perl, they are
4220 treated as literals. */
4221
4222 default:
4223 if ((options & PCRE_EXTRA) != 0)
4224 {
4225 *errorcodeptr = ERR7;
4226 goto FAILED;
4227 }
4228 class_has_8bitchar--; /* Undo the speculative increase. */
4229 class_single_char -= 2; /* Undo the speculative increase. */
4230 c = *ptr; /* Get the final character and fall through */
4231 break;
4232 }
4233 }
4234
4235 /* Fall through if we have a single character (c >= 0). This may be
4236 greater than 256. */
4237
4238 } /* End of backslash handling */
4239
4240 /* A single character may be followed by '-' to form a range. However,
4241 Perl does not permit ']' to be the end of the range. A '-' character
4242 at the end is treated as a literal. Perl ignores orphaned \E sequences
4243 entirely. The code for handling \Q and \E is messy. */
4244
4245 CHECK_RANGE:
4246 while (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
4247 {
4248 inescq = FALSE;
4249 ptr += 2;
4250 }
4251
4252 oldptr = ptr;
4253
4254 /* Remember \r or \n */
4255
4256 if (c == CHAR_CR || c == CHAR_NL) cd->external_flags |= PCRE_HASCRORLF;
4257
4258 /* Check for range */
4259
4260 if (!inescq && ptr[1] == CHAR_MINUS)
4261 {
4262 int d;
4263 ptr += 2;
4264 while (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_E) ptr += 2;
4265
4266 /* If we hit \Q (not followed by \E) at this point, go into escaped
4267 mode. */
4268
4269 while (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_Q)
4270 {
4271 ptr += 2;
4272 if (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_E)
4273 { ptr += 2; continue; }
4274 inescq = TRUE;
4275 break;
4276 }
4277
4278 if (*ptr == 0 || (!inescq && *ptr == CHAR_RIGHT_SQUARE_BRACKET))
4279 {
4280 ptr = oldptr;
4281 goto LONE_SINGLE_CHARACTER;
4282 }
4283
4284 #ifdef SUPPORT_UTF
4285 if (utf)
4286 { /* Braces are required because the */
4287 GETCHARLEN(d, ptr, ptr); /* macro generates multiple statements */
4288 }
4289 else
4290 #endif
4291 d = *ptr; /* Not UTF-8 mode */
4292
4293 /* The second part of a range can be a single-character escape, but
4294 not any of the other escapes. Perl 5.6 treats a hyphen as a literal
4295 in such circumstances. */
4296
4297 if (!inescq && d == CHAR_BACKSLASH)
4298 {
4299 d = check_escape(&ptr, errorcodeptr, cd->bracount, options, TRUE);
4300 if (*errorcodeptr != 0) goto FAILED;
4301
4302 /* \b is backspace; any other special means the '-' was literal */
4303
4304 if (d < 0)
4305 {
4306 if (d == -ESC_b) d = CHAR_BS; else
4307 {
4308 ptr = oldptr;
4309 goto LONE_SINGLE_CHARACTER; /* A few lines below */
4310 }
4311 }
4312 }
4313
4314 /* Check that the two values are in the correct order. Optimize
4315 one-character ranges */
4316
4317 if (d < c)
4318 {
4319 *errorcodeptr = ERR8;
4320 goto FAILED;
4321 }
4322
4323 if (d == c) goto LONE_SINGLE_CHARACTER; /* A few lines below */
4324
4325 /* Remember \r or \n */
4326
4327 if (d == CHAR_CR || d == CHAR_NL) cd->external_flags |= PCRE_HASCRORLF;
4328
4329 /* Since we found a character range, single character optimizations
4330 cannot be done anymore. */
4331 class_single_char = 2;
4332
4333 /* In UTF-8 mode, if the upper limit is > 255, or > 127 for caseless
4334 matching, we have to use an XCLASS with extra data items. Caseless
4335 matching for characters > 127 is available only if UCP support is
4336 available. */
4337
4338 #if defined SUPPORT_UTF && !(defined COMPILE_PCRE8)
4339 if ((d > 255) || (utf && ((options & PCRE_CASELESS) != 0 && d > 127)))
4340 #elif defined SUPPORT_UTF
4341 if (utf && (d > 255 || ((options & PCRE_CASELESS) != 0 && d > 127)))
4342 #elif !(defined COMPILE_PCRE8)
4343 if (d > 255)
4344 #endif
4345 #if defined SUPPORT_UTF || !(defined COMPILE_PCRE8)
4346 {
4347 xclass = TRUE;
4348
4349 /* With UCP support, we can find the other case equivalents of
4350 the relevant characters. There may be several ranges. Optimize how
4351 they fit with the basic range. */
4352
4353 #ifdef SUPPORT_UCP
4354 #ifndef COMPILE_PCRE8
4355 if (utf && (options & PCRE_CASELESS) != 0)
4356 #else
4357 if ((options & PCRE_CASELESS) != 0)
4358 #endif
4359 {
4360 unsigned int occ, ocd;
4361 unsigned int cc = c;
4362 unsigned int origd = d;
4363 while (get_othercase_range(&cc, origd, &occ, &ocd))
4364 {
4365 if (occ >= (unsigned int)c &&
4366 ocd <= (unsigned int)d)
4367 continue; /* Skip embedded ranges */
4368
4369 if (occ < (unsigned int)c &&
4370 ocd >= (unsigned int)c - 1) /* Extend the basic range */
4371 { /* if there is overlap, */
4372 c = occ; /* noting that if occ < c */
4373 continue; /* we can't have ocd > d */
4374 } /* because a subrange is */
4375 if (ocd > (unsigned int)d &&
4376 occ <= (unsigned int)d + 1) /* always shorter than */
4377 { /* the basic range. */
4378 d = ocd;
4379 continue;
4380 }
4381
4382 if (occ == ocd)
4383 {
4384 *class_uchardata++ = XCL_SINGLE;
4385 }
4386 else
4387 {
4388 *class_uchardata++ = XCL_RANGE;
4389 class_uchardata += PRIV(ord2utf)(occ, class_uchardata);
4390 }
4391 class_uchardata += PRIV(ord2utf)(ocd, class_uchardata);
4392 }
4393 }
4394 #endif /* SUPPORT_UCP */
4395
4396 /* Now record the original range, possibly modified for UCP caseless
4397 overlapping ranges. */
4398
4399 *class_uchardata++ = XCL_RANGE;
4400 #ifdef SUPPORT_UTF
4401 #ifndef COMPILE_PCRE8
4402 if (utf)
4403 {
4404 class_uchardata += PRIV(ord2utf)(c, class_uchardata);
4405 class_uchardata += PRIV(ord2utf)(d, class_uchardata);
4406 }
4407 else
4408 {
4409 *class_uchardata++ = c;
4410 *class_uchardata++ = d;
4411 }
4412 #else
4413 class_uchardata += PRIV(ord2utf)(c, class_uchardata);
4414 class_uchardata += PRIV(ord2utf)(d, class_uchardata);
4415 #endif
4416 #else /* SUPPORT_UTF */
4417 *class_uchardata++ = c;
4418 *class_uchardata++ = d;
4419 #endif /* SUPPORT_UTF */
4420
4421 /* With UCP support, we are done. Without UCP support, there is no
4422 caseless matching for UTF characters > 127; we can use the bit map
4423 for the smaller ones. As for 16 bit characters without UTF, we
4424 can still use */
4425
4426 #ifdef SUPPORT_UCP
4427 #ifndef COMPILE_PCRE8
4428 if (utf)
4429 #endif
4430 continue; /* With next character in the class */
4431 #endif /* SUPPORT_UCP */
4432
4433 #if defined SUPPORT_UTF && !defined(SUPPORT_UCP) && !(defined COMPILE_PCRE8)
4434 if (utf)
4435 {
4436 if ((options & PCRE_CASELESS) == 0 || c > 127) continue;
4437 /* Adjust upper limit and fall through to set up the map */
4438 d = 127;
4439 }
4440 else
4441 {
4442 if (c > 255) continue;
4443 /* Adjust upper limit and fall through to set up the map */
4444 d = 255;
4445 }
4446 #elif defined SUPPORT_UTF && !defined(SUPPORT_UCP)
4447 if ((options & PCRE_CASELESS) == 0 || c > 127) continue;
4448 /* Adjust upper limit and fall through to set up the map */
4449 d = 127;
4450 #else
4451 if (c > 255) continue;
4452 /* Adjust upper limit and fall through to set up the map */
4453 d = 255;
4454 #endif /* SUPPORT_UTF && !SUPPORT_UCP && !COMPILE_PCRE8 */
4455 }
4456 #endif /* SUPPORT_UTF || !COMPILE_PCRE8 */
4457
4458 /* We use the bit map for 8 bit mode, or when the characters fall
4459 partially or entirely to [0-255] ([0-127] for UCP) ranges. */
4460
4461 class_has_8bitchar = 1;
4462
4463 /* We can save a bit of time by skipping this in the pre-compile. */
4464
4465 if (lengthptr == NULL) for (; c <= d; c++)
4466 {
4467 classbits[c/8] |= (1 << (c&7));
4468 if ((options & PCRE_CASELESS) != 0)
4469 {
4470 int uc = cd->fcc[c]; /* flip case */
4471 classbits[uc/8] |= (1 << (uc&7));
4472 }
4473 }
4474
4475 continue; /* Go get the next char in the class */
4476 }
4477
4478 /* Handle a lone single character - we can get here for a normal
4479 non-escape char, or after \ that introduces a single character or for an
4480 apparent range that isn't. */
4481
4482 LONE_SINGLE_CHARACTER:
4483
4484 /* Only the value of 1 matters for class_single_char. */
4485 if (class_single_char < 2) class_single_char++;
4486
4487 /* If class_charcount is 1, we saw precisely one character. As long as
4488 there were no negated characters >= 128 and there was no use of \p or \P,
4489 in other words, no use of any XCLASS features, we can optimize.
4490
4491 In UTF-8 mode, we can optimize the negative case only if there were no
4492 characters >= 128 because OP_NOT and the related opcodes like OP_NOTSTAR
4493 operate on single-bytes characters only. This is an historical hangover.
4494 Maybe one day we can tidy these opcodes to handle multi-byte characters.
4495
4496 The optimization throws away the bit map. We turn the item into a
4497 1-character OP_CHAR[I] if it's positive, or OP_NOT[I] if it's negative.
4498 Note that OP_NOT[I] does not support multibyte characters. In the positive
4499 case, it can cause firstchar to be set. Otherwise, there can be no first
4500 char if this item is first, whatever repeat count may follow. In the case
4501 of reqchar, save the previous value for reinstating. */
4502
4503 #ifdef SUPPORT_UTF
4504 if (class_single_char == 1 && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET
4505 && (!utf || !negate_class || c < (MAX_VALUE_FOR_SINGLE_CHAR + 1)))
4506 #else
4507 if (class_single_char == 1 && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET)
4508 #endif
4509 {
4510 ptr++;
4511 zeroreqchar = reqchar;
4512
4513 /* The OP_NOT[I] opcodes work on single characters only. */
4514
4515 if (negate_class)
4516 {
4517 if (firstchar == REQ_UNSET) firstchar = REQ_NONE;
4518 zerofirstchar = firstchar;
4519 *code++ = ((options & PCRE_CASELESS) != 0)? OP_NOTI: OP_NOT;
4520 *code++ = c;
4521 goto NOT_CHAR;
4522 }
4523
4524 /* For a single, positive character, get the value into mcbuffer, and
4525 then we can handle this with the normal one-character code. */
4526
4527 #ifdef SUPPORT_UTF
4528 if (utf && c > MAX_VALUE_FOR_SINGLE_CHAR)
4529 mclength = PRIV(ord2utf)(c, mcbuffer);
4530 else
4531 #endif
4532 {
4533 mcbuffer[0] = c;
4534 mclength = 1;
4535 }
4536 goto ONE_CHAR;
4537 } /* End of 1-char optimization */
4538
4539 /* Handle a character that cannot go in the bit map. */
4540
4541 #if defined SUPPORT_UTF && !(defined COMPILE_PCRE8)
4542 if ((c > 255) || (utf && ((options & PCRE_CASELESS) != 0 && c > 127)))
4543 #elif defined SUPPORT_UTF
4544 if (utf && (c > 255 || ((options & PCRE_CASELESS) != 0 && c > 127)))
4545 #elif !(defined COMPILE_PCRE8)
4546 if (c > 255)
4547 #endif
4548
4549 #if defined SUPPORT_UTF || !(defined COMPILE_PCRE8)
4550 {
4551 xclass = TRUE;
4552 *class_uchardata++ = XCL_SINGLE;
4553 #ifdef SUPPORT_UTF
4554 #ifndef COMPILE_PCRE8
4555 /* In non 8 bit mode, we can get here even if we are not in UTF mode. */
4556 if (!utf)
4557 *class_uchardata++ = c;
4558 else
4559 #endif
4560 class_uchardata += PRIV(ord2utf)(c, class_uchardata);
4561 #else /* SUPPORT_UTF */
4562 *class_uchardata++ = c;
4563 #endif /* SUPPORT_UTF */
4564
4565 #ifdef SUPPORT_UCP
4566 #ifdef COMPILE_PCRE8
4567 if ((options & PCRE_CASELESS) != 0)
4568 #else
4569 /* In non 8 bit mode, we can get here even if we are not in UTF mode. */
4570 if (utf && (options & PCRE_CASELESS) != 0)
4571 #endif
4572 {
4573 unsigned int othercase;
4574 if ((int)(othercase = UCD_OTHERCASE(c)) != c)
4575 {
4576 *class_uchardata++ = XCL_SINGLE;
4577 class_uchardata += PRIV(ord2utf)(othercase, class_uchardata);
4578 }
4579 }
4580 #endif /* SUPPORT_UCP */
4581
4582 }
4583 else
4584 #endif /* SUPPORT_UTF || COMPILE_PCRE16 */
4585
4586 /* Handle a single-byte character */
4587 {
4588 class_has_8bitchar = 1;
4589 classbits[c/8] |= (1 << (c&7));
4590 if ((options & PCRE_CASELESS) != 0)
4591 {
4592 c = cd->fcc[c]; /* flip case */
4593 classbits[c/8] |= (1 << (c&7));
4594 }
4595 }
4596 }
4597
4598 /* Loop until ']' reached. This "while" is the end of the "do" far above.
4599 If we are at the end of an internal nested string, revert to the outer
4600 string. */
4601
4602 while (((c = *(++ptr)) != 0 ||
4603 (nestptr != NULL &&
4604 (ptr = nestptr, nestptr = NULL, c = *(++ptr)) != 0)) &&
4605 (c != CHAR_RIGHT_SQUARE_BRACKET || inescq));
4606
4607 /* Check for missing terminating ']' */
4608
4609 if (c == 0)
4610 {
4611 *errorcodeptr = ERR6;
4612 goto FAILED;
4613 }
4614
4615 /* If this is the first thing in the branch, there can be no first char
4616 setting, whatever the repeat count. Any reqchar setting must remain
4617 unchanged after any kind of repeat. */
4618
4619 if (firstchar == REQ_UNSET) firstchar = REQ_NONE;
4620 zerofirstchar = firstchar;
4621 zeroreqchar = reqchar;
4622
4623 /* If there are characters with values > 255, we have to compile an
4624 extended class, with its own opcode, unless there was a negated special
4625 such as \S in the class, and PCRE_UCP is not set, because in that case all
4626 characters > 255 are in the class, so any that were explicitly given as
4627 well can be ignored. If (when there are explicit characters > 255 that must
4628 be listed) there are no characters < 256, we can omit the bitmap in the
4629 actual compiled code. */
4630
4631 #ifdef SUPPORT_UTF
4632 if (xclass && (!should_flip_negation || (options & PCRE_UCP) != 0))
4633 #elif !defined COMPILE_PCRE8
4634 if (xclass && !should_flip_negation)
4635 #endif
4636 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
4637 {
4638 *class_uchardata++ = XCL_END; /* Marks the end of extra data */
4639 *code++ = OP_XCLASS;
4640 code += LINK_SIZE;
4641 *code = negate_class? XCL_NOT:0;
4642
4643 /* If the map is required, move up the extra data to make room for it;
4644 otherwise just move the code pointer to the end of the extra data. */
4645
4646 if (class_has_8bitchar > 0)
4647 {
4648 *code++ |= XCL_MAP;
4649 memmove(code + (32 / sizeof(pcre_uchar)), code,
4650 IN_UCHARS(class_uchardata - code));
4651 memcpy(code, classbits, 32);
4652 code = class_uchardata + (32 / sizeof(pcre_uchar));
4653 }
4654 else code = class_uchardata;
4655
4656 /* Now fill in the complete length of the item */
4657
4658 PUT(previous, 1, (int)(code - previous));
4659 break; /* End of class handling */
4660 }
4661 #endif
4662
4663 /* If there are no characters > 255, or they are all to be included or
4664 excluded, set the opcode to OP_CLASS or OP_NCLASS, depending on whether the
4665 whole class was negated and whether there were negative specials such as \S
4666 (non-UCP) in the class. Then copy the 32-byte map into the code vector,
4667 negating it if necessary. */
4668
4669 *code++ = (negate_class == should_flip_negation) ? OP_CLASS : OP_NCLASS;
4670 if (lengthptr == NULL) /* Save time in the pre-compile phase */
4671 {
4672 if (negate_class)
4673 for (c = 0; c < 32; c++) classbits[c] = ~classbits[c];
4674 memcpy(code, classbits, 32);
4675 }
4676 code += 32 / sizeof(pcre_uchar);
4677 NOT_CHAR:
4678 break;
4679
4680
4681 /* ===================================================================*/
4682 /* Various kinds of repeat; '{' is not necessarily a quantifier, but this
4683 has been tested above. */
4684
4685 case CHAR_LEFT_CURLY_BRACKET:
4686 if (!is_quantifier) goto NORMAL_CHAR;
4687 ptr = read_repeat_counts(ptr+1, &repeat_min, &repeat_max, errorcodeptr);
4688 if (*errorcodeptr != 0) goto FAILED;
4689 goto REPEAT;
4690
4691 case CHAR_ASTERISK:
4692 repeat_min = 0;
4693 repeat_max = -1;
4694 goto REPEAT;
4695
4696 case CHAR_PLUS:
4697 repeat_min = 1;
4698 repeat_max = -1;
4699 goto REPEAT;
4700
4701 case CHAR_QUESTION_MARK:
4702 repeat_min = 0;
4703 repeat_max = 1;
4704
4705 REPEAT:
4706 if (previous == NULL)
4707 {
4708 *errorcodeptr = ERR9;
4709 goto FAILED;
4710 }
4711
4712 if (repeat_min == 0)
4713 {
4714 firstchar = zerofirstchar; /* Adjust for zero repeat */
4715 reqchar = zeroreqchar; /* Ditto */
4716 }
4717
4718 /* Remember whether this is a variable length repeat */
4719
4720 reqvary = (repeat_min == repeat_max)? 0 : REQ_VARY;
4721
4722 op_type = 0; /* Default single-char op codes */
4723 possessive_quantifier = FALSE; /* Default not possessive quantifier */
4724
4725 /* Save start of previous item, in case we have to move it up in order to
4726 insert something before it. */
4727
4728 tempcode = previous;
4729
4730 /* If the next character is '+', we have a possessive quantifier. This
4731 implies greediness, whatever the setting of the PCRE_UNGREEDY option.
4732 If the next character is '?' this is a minimizing repeat, by default,
4733 but if PCRE_UNGREEDY is set, it works the other way round. We change the
4734 repeat type to the non-default. */
4735
4736 if (ptr[1] == CHAR_PLUS)
4737 {
4738 repeat_type = 0; /* Force greedy */
4739 possessive_quantifier = TRUE;
4740 ptr++;
4741 }
4742 else if (ptr[1] == CHAR_QUESTION_MARK)
4743 {
4744 repeat_type = greedy_non_default;
4745 ptr++;
4746 }
4747 else repeat_type = greedy_default;
4748
4749 /* If previous was a recursion call, wrap it in atomic brackets so that
4750 previous becomes the atomic group. All recursions were so wrapped in the
4751 past, but it no longer happens for non-repeated recursions. In fact, the
4752 repeated ones could be re-implemented independently so as not to need this,
4753 but for the moment we rely on the code for repeating groups. */
4754
4755 if (*previous == OP_RECURSE)
4756 {
4757 memmove(previous + 1 + LINK_SIZE, previous, IN_UCHARS(1 + LINK_SIZE));
4758 *previous = OP_ONCE;
4759 PUT(previous, 1, 2 + 2*LINK_SIZE);
4760 previous[2 + 2*LINK_SIZE] = OP_KET;
4761 PUT(previous, 3 + 2*LINK_SIZE, 2 + 2*LINK_SIZE);
4762 code += 2 + 2 * LINK_SIZE;
4763 length_prevgroup = 3 + 3*LINK_SIZE;
4764
4765 /* When actually compiling, we need to check whether this was a forward
4766 reference, and if so, adjust the offset. */
4767
4768 if (lengthptr == NULL && cd->hwm >= cd->start_workspace + LINK_SIZE)
4769 {
4770 int offset = GET(cd->hwm, -LINK_SIZE);
4771 if (offset == previous + 1 - cd->start_code)
4772 PUT(cd->hwm, -LINK_SIZE, offset + 1 + LINK_SIZE);
4773 }
4774 }
4775
4776 /* Now handle repetition for the different types of item. */
4777
4778 /* If previous was a character match, abolish the item and generate a
4779 repeat item instead. If a char item has a minumum of more than one, ensure
4780 that it is set in reqchar - it might not be if a sequence such as x{3} is
4781 the first thing in a branch because the x will have gone into firstchar
4782 instead. */
4783
4784 if (*previous == OP_CHAR || *previous == OP_CHARI)
4785 {
4786 op_type = (*previous == OP_CHAR)? 0 : OP_STARI - OP_STAR;
4787
4788 /* Deal with UTF characters that take up more than one character. It's
4789 easier to write this out separately than try to macrify it. Use c to
4790 hold the length of the character in bytes, plus UTF_LENGTH to flag that
4791 it's a length rather than a small character. */
4792
4793 #ifdef SUPPORT_UTF
4794 if (utf && NOT_FIRSTCHAR(code[-1]))
4795 {
4796 pcre_uchar *lastchar = code - 1;
4797 BACKCHAR(lastchar);
4798 c = (int)(code - lastchar); /* Length of UTF-8 character */
4799 memcpy(utf_chars, lastchar, IN_UCHARS(c)); /* Save the char */
4800 c |= UTF_LENGTH; /* Flag c as a length */
4801 }
4802 else
4803 #endif /* SUPPORT_UTF */
4804
4805 /* Handle the case of a single charater - either with no UTF support, or
4806 with UTF disabled, or for a single character UTF character. */
4807 {
4808 c = code[-1];
4809 if (repeat_min > 1) reqchar = c | req_caseopt | cd->req_varyopt;
4810 }
4811
4812 /* If the repetition is unlimited, it pays to see if the next thing on
4813 the line is something that cannot possibly match this character. If so,
4814 automatically possessifying this item gains some performance in the case
4815 where the match fails. */
4816
4817 if (!possessive_quantifier &&
4818 repeat_max < 0 &&
4819 check_auto_possessive(previous, utf, ptr + 1, options, cd))
4820 {
4821 repeat_type = 0; /* Force greedy */
4822 possessive_quantifier = TRUE;
4823 }
4824
4825 goto OUTPUT_SINGLE_REPEAT; /* Code shared with single character types */
4826 }
4827
4828 /* If previous was a single negated character ([^a] or similar), we use
4829 one of the special opcodes, replacing it. The code is shared with single-
4830 character repeats by setting opt_type to add a suitable offset into
4831 repeat_type. We can also test for auto-possessification. OP_NOT and OP_NOTI
4832 are currently used only for single-byte chars. */
4833
4834 else if (*previous == OP_NOT || *previous == OP_NOTI)
4835 {
4836 op_type = ((*previous == OP_NOT)? OP_NOTSTAR : OP_NOTSTARI) - OP_STAR;
4837 c = previous[1];
4838 if (!possessive_quantifier &&
4839 repeat_max < 0 &&
4840 check_auto_possessive(previous, utf, ptr + 1, options, cd))
4841 {
4842 repeat_type = 0; /* Force greedy */
4843 possessive_quantifier = TRUE;
4844 }
4845 goto OUTPUT_SINGLE_REPEAT;
4846 }
4847
4848 /* If previous was a character type match (\d or similar), abolish it and
4849 create a suitable repeat item. The code is shared with single-character
4850 repeats by setting op_type to add a suitable offset into repeat_type. Note
4851 the the Unicode property types will be present only when SUPPORT_UCP is
4852 defined, but we don't wrap the little bits of code here because it just
4853 makes it horribly messy. */
4854
4855 else if (*previous < OP_EODN)
4856 {
4857 pcre_uchar *oldcode;
4858 int prop_type, prop_value;
4859 op_type = OP_TYPESTAR - OP_STAR; /* Use type opcodes */
4860 c = *previous;
4861
4862 if (!possessive_quantifier &&
4863 repeat_max < 0 &&
4864 check_auto_possessive(previous, utf, ptr + 1, options, cd))
4865 {
4866 repeat_type = 0; /* Force greedy */
4867 possessive_quantifier = TRUE;
4868 }
4869
4870 OUTPUT_SINGLE_REPEAT:
4871 if (*previous == OP_PROP || *previous == OP_NOTPROP)
4872 {
4873 prop_type = previous[1];
4874 prop_value = previous[2];
4875 }
4876 else prop_type = prop_value = -1;
4877
4878 oldcode = code;
4879 code = previous; /* Usually overwrite previous item */
4880
4881 /* If the maximum is zero then the minimum must also be zero; Perl allows
4882 this case, so we do too - by simply omitting the item altogether. */
4883
4884 if (repeat_max == 0) goto END_REPEAT;
4885
4886 /*--------------------------------------------------------------------*/
4887 /* This code is obsolete from release 8.00; the restriction was finally
4888 removed: */
4889
4890 /* All real repeats make it impossible to handle partial matching (maybe
4891 one day we will be able to remove this restriction). */
4892
4893 /* if (repeat_max != 1) cd->external_flags |= PCRE_NOPARTIAL; */
4894 /*--------------------------------------------------------------------*/
4895
4896 /* Combine the op_type with the repeat_type */
4897
4898 repeat_type += op_type;
4899
4900 /* A minimum of zero is handled either as the special case * or ?, or as
4901 an UPTO, with the maximum given. */
4902
4903 if (repeat_min == 0)
4904 {
4905 if (repeat_max == -1) *code++ = OP_STAR + repeat_type;
4906 else if (repeat_max == 1) *code++ = OP_QUERY + repeat_type;
4907 else
4908 {
4909 *code++ = OP_UPTO + repeat_type;
4910 PUT2INC(code, 0, repeat_max);
4911 }
4912 }
4913
4914 /* A repeat minimum of 1 is optimized into some special cases. If the
4915 maximum is unlimited, we use OP_PLUS. Otherwise, the original item is
4916 left in place and, if the maximum is greater than 1, we use OP_UPTO with
4917 one less than the maximum. */
4918
4919 else if (repeat_min == 1)
4920 {
4921 if (repeat_max == -1)
4922 *code++ = OP_PLUS + repeat_type;
4923 else
4924 {
4925 code = oldcode; /* leave previous item in place */
4926 if (repeat_max == 1) goto END_REPEAT;
4927 *code++ = OP_UPTO + repeat_type;
4928 PUT2INC(code, 0, repeat_max - 1);
4929 }
4930 }
4931
4932 /* The case {n,n} is just an EXACT, while the general case {n,m} is
4933 handled as an EXACT followed by an UPTO. */
4934
4935 else
4936 {
4937 *code++ = OP_EXACT + op_type; /* NB EXACT doesn't have repeat_type */
4938 PUT2INC(code, 0, repeat_min);
4939
4940 /* If the maximum is unlimited, insert an OP_STAR. Before doing so,
4941 we have to insert the character for the previous code. For a repeated
4942 Unicode property match, there are two extra bytes that define the
4943 required property. In UTF-8 mode, long characters have their length in
4944 c, with the UTF_LENGTH bit as a flag. */
4945
4946 if (repeat_max < 0)
4947 {
4948 #ifdef SUPPORT_UTF
4949 if (utf && (c & UTF_LENGTH) != 0)
4950 {
4951 memcpy(code, utf_chars, IN_UCHARS(c & 7));
4952 code += c & 7;
4953 }
4954 else
4955 #endif
4956 {
4957 *code++ = c;
4958 if (prop_type >= 0)
4959 {
4960 *code++ = prop_type;
4961 *code++ = prop_value;
4962 }
4963 }
4964 *code++ = OP_STAR + repeat_type;
4965 }
4966
4967 /* Else insert an UPTO if the max is greater than the min, again
4968 preceded by the character, for the previously inserted code. If the
4969 UPTO is just for 1 instance, we can use QUERY instead. */
4970
4971 else if (repeat_max != repeat_min)
4972 {
4973 #ifdef SUPPORT_UTF
4974 if (utf && (c & UTF_LENGTH) != 0)
4975 {
4976 memcpy(code, utf_chars, IN_UCHARS(c & 7));
4977 code += c & 7;
4978 }
4979 else
4980 #endif
4981 *code++ = c;
4982 if (prop_type >= 0)
4983 {
4984 *code++ = prop_type;
4985 *code++ = prop_value;
4986 }
4987 repeat_max -= repeat_min;
4988
4989 if (repeat_max == 1)
4990 {
4991 *code++ = OP_QUERY + repeat_type;
4992 }
4993 else
4994 {
4995 *code++ = OP_UPTO + repeat_type;
4996 PUT2INC(code, 0, repeat_max);
4997 }
4998 }
4999 }
5000
5001 /* The character or character type itself comes last in all cases. */
5002
5003 #ifdef SUPPORT_UTF
5004 if (utf && (c & UTF_LENGTH) != 0)
5005 {
5006 memcpy(code, utf_chars, IN_UCHARS(c & 7));
5007 code += c & 7;
5008 }
5009 else
5010 #endif
5011 *code++ = c;
5012
5013 /* For a repeated Unicode property match, there are two extra bytes that
5014 define the required property. */
5015
5016 #ifdef SUPPORT_UCP
5017 if (prop_type >= 0)
5018 {
5019 *code++ = prop_type;
5020 *code++ = prop_value;
5021 }
5022 #endif
5023 }
5024
5025 /* If previous was a character class or a back reference, we put the repeat
5026 stuff after it, but just skip the item if the repeat was {0,0}. */
5027
5028 else if (*previous == OP_CLASS ||
5029 *previous == OP_NCLASS ||
5030 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
5031 *previous == OP_XCLASS ||
5032 #endif
5033 *previous == OP_REF ||
5034 *previous == OP_REFI)
5035 {
5036 if (repeat_max == 0)
5037 {
5038 code = previous;
5039 goto END_REPEAT;
5040 }
5041
5042 /*--------------------------------------------------------------------*/
5043 /* This code is obsolete from release 8.00; the restriction was finally
5044 removed: */
5045
5046 /* All real repeats make it impossible to handle partial matching (maybe
5047 one day we will be able to remove this restriction). */
5048
5049 /* if (repeat_max != 1) cd->external_flags |= PCRE_NOPARTIAL; */
5050 /*--------------------------------------------------------------------*/
5051
5052 if (repeat_min == 0 && repeat_max == -1)
5053 *code++ = OP_CRSTAR + repeat_type;
5054 else if (repeat_min == 1 && repeat_max == -1)
5055 *code++ = OP_CRPLUS + repeat_type;
5056 else if (repeat_min == 0 && repeat_max == 1)
5057 *code++ = OP_CRQUERY + repeat_type;
5058 else
5059 {
5060 *code++ = OP_CRRANGE + repeat_type;
5061 PUT2INC(code, 0, repeat_min);
5062 if (repeat_max == -1) repeat_max = 0; /* 2-byte encoding for max */
5063 PUT2INC(code, 0, repeat_max);
5064 }
5065 }
5066
5067 /* If previous was a bracket group, we may have to replicate it in certain
5068 cases. Note that at this point we can encounter only the "basic" bracket
5069 opcodes such as BRA and CBRA, as this is the place where they get converted
5070 into the more special varieties such as BRAPOS and SBRA. A test for >=
5071 OP_ASSERT and <= OP_COND includes ASSERT, ASSERT_NOT, ASSERTBACK,
5072 ASSERTBACK_NOT, ONCE, BRA, CBRA, and COND. Originally, PCRE did not allow
5073 repetition of assertions, but now it does, for Perl compatibility. */
5074
5075 else if (*previous >= OP_ASSERT && *previous <= OP_COND)
5076 {
5077 register int i;
5078 int len = (int)(code - previous);
5079 pcre_uchar *bralink = NULL;
5080 pcre_uchar *brazeroptr = NULL;
5081
5082 /* Repeating a DEFINE group is pointless, but Perl allows the syntax, so
5083 we just ignore the repeat. */
5084
5085 if (*previous == OP_COND && previous[LINK_SIZE+1] == OP_DEF)
5086 goto END_REPEAT;
5087
5088 /* There is no sense in actually repeating assertions. The only potential
5089 use of repetition is in cases when the assertion is optional. Therefore,
5090 if the minimum is greater than zero, just ignore the repeat. If the
5091 maximum is not not zero or one, set it to 1. */
5092
5093 if (*previous < OP_ONCE) /* Assertion */
5094 {
5095 if (repeat_min > 0) goto END_REPEAT;
5096 if (repeat_max < 0 || repeat_max > 1) repeat_max = 1;
5097 }
5098
5099 /* The case of a zero minimum is special because of the need to stick
5100 OP_BRAZERO in front of it, and because the group appears once in the
5101 data, whereas in other cases it appears the minimum number of times. For
5102 this reason, it is simplest to treat this case separately, as otherwise
5103 the code gets far too messy. There are several special subcases when the
5104 minimum is zero. */
5105
5106 if (repeat_min == 0)
5107 {
5108 /* If the maximum is also zero, we used to just omit the group from the
5109 output altogether, like this:
5110
5111 ** if (repeat_max == 0)
5112 ** {
5113 ** code = previous;
5114 ** goto END_REPEAT;
5115 ** }
5116
5117 However, that fails when a group or a subgroup within it is referenced
5118 as a subroutine from elsewhere in the pattern, so now we stick in
5119 OP_SKIPZERO in front of it so that it is skipped on execution. As we
5120 don't have a list of which groups are referenced, we cannot do this
5121 selectively.
5122
5123 If the maximum is 1 or unlimited, we just have to stick in the BRAZERO
5124 and do no more at this point. However, we do need to adjust any
5125 OP_RECURSE calls inside the group that refer to the group itself or any
5126 internal or forward referenced group, because the offset is from the
5127 start of the whole regex. Temporarily terminate the pattern while doing
5128 this. */
5129
5130 if (repeat_max <= 1) /* Covers 0, 1, and unlimited */
5131 {
5132 *code = OP_END;
5133 adjust_recurse(previous, 1, utf, cd, save_hwm);
5134 memmove(previous + 1, previous, IN_UCHARS(len));
5135 code++;
5136 if (repeat_max == 0)
5137 {
5138 *previous++ = OP_SKIPZERO;
5139 goto END_REPEAT;
5140 }
5141 brazeroptr = previous; /* Save for possessive optimizing */
5142 *previous++ = OP_BRAZERO + repeat_type;
5143 }
5144
5145 /* If the maximum is greater than 1 and limited, we have to replicate
5146 in a nested fashion, sticking OP_BRAZERO before each set of brackets.
5147 The first one has to be handled carefully because it's the original
5148 copy, which has to be moved up. The remainder can be handled by code
5149 that is common with the non-zero minimum case below. We have to
5150 adjust the value or repeat_max, since one less copy is required. Once
5151 again, we may have to adjust any OP_RECURSE calls inside the group. */
5152
5153 else
5154 {
5155 int offset;
5156 *code = OP_END;
5157 adjust_recurse(previous, 2 + LINK_SIZE, utf, cd, save_hwm);
5158 memmove(previous + 2 + LINK_SIZE, previous, IN_UCHARS(len));
5159 code += 2 + LINK_SIZE;
5160 *previous++ = OP_BRAZERO + repeat_type;
5161 *previous++ = OP_BRA;
5162
5163 /* We chain together the bracket offset fields that have to be
5164 filled in later when the ends of the brackets are reached. */
5165
5166 offset = (bralink == NULL)? 0 : (int)(previous - bralink);
5167 bralink = previous;
5168 PUTINC(previous, 0, offset);
5169 }
5170
5171 repeat_max--;
5172 }
5173
5174 /* If the minimum is greater than zero, replicate the group as many
5175 times as necessary, and adjust the maximum to the number of subsequent
5176 copies that we need. If we set a first char from the group, and didn't
5177 set a required char, copy the latter from the former. If there are any
5178 forward reference subroutine calls in the group, there will be entries on
5179 the workspace list; replicate these with an appropriate increment. */
5180
5181 else
5182 {
5183 if (repeat_min > 1)
5184 {
5185 /* In the pre-compile phase, we don't actually do the replication. We
5186 just adjust the length as if we had. Do some paranoid checks for
5187 potential integer overflow. The INT64_OR_DOUBLE type is a 64-bit
5188 integer type when available, otherwise double. */
5189
5190 if (lengthptr != NULL)
5191 {
5192 int delta = (repeat_min - 1)*length_prevgroup;
5193 if ((INT64_OR_DOUBLE)(repeat_min - 1)*
5194 (INT64_OR_DOUBLE)length_prevgroup >
5195 (INT64_OR_DOUBLE)INT_MAX ||
5196 OFLOW_MAX - *lengthptr < delta)
5197 {
5198 *errorcodeptr = ERR20;
5199 goto FAILED;
5200 }
5201 *lengthptr += delta;
5202 }
5203
5204 /* This is compiling for real. If there is a set first byte for
5205 the group, and we have not yet set a "required byte", set it. Make
5206 sure there is enough workspace for copying forward references before
5207 doing the copy. */
5208
5209 else
5210 {
5211 if (groupsetfirstchar && reqchar < 0) reqchar = firstchar;
5212
5213 for (i = 1; i < repeat_min; i++)
5214 {
5215 pcre_uchar *hc;
5216 pcre_uchar *this_hwm = cd->hwm;
5217 memcpy(code, previous, IN_UCHARS(len));
5218
5219 while (cd->hwm > cd->start_workspace + cd->workspace_size -
5220 WORK_SIZE_SAFETY_MARGIN - (this_hwm - save_hwm))
5221 {
5222 int save_offset = save_hwm - cd->start_workspace;
5223 int this_offset = this_hwm - cd->start_workspace;
5224 *errorcodeptr = expand_workspace(cd);
5225 if (*errorcodeptr != 0) goto FAILED;
5226 save_hwm = (pcre_uchar *)cd->start_workspace + save_offset;
5227 this_hwm = (pcre_uchar *)cd->start_workspace + this_offset;
5228 }
5229
5230 for (hc = save_hwm; hc < this_hwm; hc += LINK_SIZE)
5231 {
5232 PUT(cd->hwm, 0, GET(hc, 0) + len);
5233 cd->hwm += LINK_SIZE;
5234 }
5235 save_hwm = this_hwm;
5236 code += len;
5237 }
5238 }
5239 }
5240
5241 if (repeat_max > 0) repeat_max -= repeat_min;
5242 }
5243
5244 /* This code is common to both the zero and non-zero minimum cases. If
5245 the maximum is limited, it replicates the group in a nested fashion,
5246 remembering the bracket starts on a stack. In the case of a zero minimum,
5247 the first one was set up above. In all cases the repeat_max now specifies
5248 the number of additional copies needed. Again, we must remember to
5249 replicate entries on the forward reference list. */
5250
5251 if (repeat_max >= 0)
5252 {
5253 /* In the pre-compile phase, we don't actually do the replication. We
5254 just adjust the length as if we had. For each repetition we must add 1
5255 to the length for BRAZERO and for all but the last repetition we must
5256 add 2 + 2*LINKSIZE to allow for the nesting that occurs. Do some
5257 paranoid checks to avoid integer overflow. The INT64_OR_DOUBLE type is
5258 a 64-bit integer type when available, otherwise double. */
5259
5260 if (lengthptr != NULL && repeat_max > 0)
5261 {
5262 int delta = repeat_max * (length_prevgroup + 1 + 2 + 2*LINK_SIZE) -
5263 2 - 2*LINK_SIZE; /* Last one doesn't nest */
5264 if ((INT64_OR_DOUBLE)repeat_max *
5265 (INT64_OR_DOUBLE)(length_prevgroup + 1 + 2 + 2*LINK_SIZE)
5266 > (INT64_OR_DOUBLE)INT_MAX ||
5267 OFLOW_MAX - *lengthptr < delta)
5268 {
5269 *errorcodeptr = ERR20;
5270 goto FAILED;
5271 }
5272 *lengthptr += delta;
5273 }
5274
5275 /* This is compiling for real */
5276
5277 else for (i = repeat_max - 1; i >= 0; i--)
5278 {
5279 pcre_uchar *hc;
5280 pcre_uchar *this_hwm = cd->hwm;
5281
5282 *code++ = OP_BRAZERO + repeat_type;
5283
5284 /* All but the final copy start a new nesting, maintaining the
5285 chain of brackets outstanding. */
5286
5287 if (i != 0)
5288 {
5289 int offset;
5290 *code++ = OP_BRA;
5291 offset = (bralink == NULL)? 0 : (int)(code - bralink);
5292 bralink = code;
5293 PUTINC(code, 0, offset);
5294 }
5295
5296 memcpy(code, previous, IN_UCHARS(len));
5297
5298 /* Ensure there is enough workspace for forward references before
5299 copying them. */
5300
5301 while (cd->hwm > cd->start_workspace + cd->workspace_size -
5302 WORK_SIZE_SAFETY_MARGIN - (this_hwm - save_hwm))
5303 {
5304 int save_offset = save_hwm - cd->start_workspace;
5305 int this_offset = this_hwm - cd->start_workspace;
5306 *errorcodeptr = expand_workspace(cd);
5307 if (*errorcodeptr != 0) goto FAILED;
5308 save_hwm = (pcre_uchar *)cd->start_workspace + save_offset;
5309 this_hwm = (pcre_uchar *)cd->start_workspace + this_offset;
5310 }
5311
5312 for (hc = save_hwm; hc < this_hwm; hc += LINK_SIZE)
5313 {
5314 PUT(cd->hwm, 0, GET(hc, 0) + len + ((i != 0)? 2+LINK_SIZE : 1));
5315 cd->hwm += LINK_SIZE;
5316 }
5317 save_hwm = this_hwm;
5318 code += len;
5319 }
5320
5321 /* Now chain through the pending brackets, and fill in their length
5322 fields (which are holding the chain links pro tem). */
5323
5324 while (bralink != NULL)
5325 {
5326 int oldlinkoffset;
5327 int offset = (int)(code - bralink + 1);
5328 pcre_uchar *bra = code - offset;
5329 oldlinkoffset = GET(bra, 1);
5330 bralink = (oldlinkoffset == 0)? NULL : bralink - oldlinkoffset;
5331 *code++ = OP_KET;
5332 PUTINC(code, 0, offset);
5333 PUT(bra, 1, offset);
5334 }
5335 }
5336
5337 /* If the maximum is unlimited, set a repeater in the final copy. For
5338 ONCE brackets, that's all we need to do. However, possessively repeated
5339 ONCE brackets can be converted into non-capturing brackets, as the
5340 behaviour of (?:xx)++ is the same as (?>xx)++ and this saves having to
5341 deal with possessive ONCEs specially.
5342
5343 Otherwise, when we are doing the actual compile phase, check to see
5344 whether this group is one that could match an empty string. If so,
5345 convert the initial operator to the S form (e.g. OP_BRA -> OP_SBRA) so
5346 that runtime checking can be done. [This check is also applied to ONCE
5347 groups at runtime, but in a different way.]
5348
5349 Then, if the quantifier was possessive and the bracket is not a
5350 conditional, we convert the BRA code to the POS form, and the KET code to
5351 KETRPOS. (It turns out to be convenient at runtime to detect this kind of
5352 subpattern at both the start and at the end.) The use of special opcodes
5353 makes it possible to reduce greatly the stack usage in pcre_exec(). If
5354 the group is preceded by OP_BRAZERO, convert this to OP_BRAPOSZERO.
5355
5356 Then, if the minimum number of matches is 1 or 0, cancel the possessive
5357 flag so that the default action below, of wrapping everything inside
5358 atomic brackets, does not happen. When the minimum is greater than 1,
5359 there will be earlier copies of the group, and so we still have to wrap
5360 the whole thing. */
5361
5362 else
5363 {
5364 pcre_uchar *ketcode = code - 1 - LINK_SIZE;
5365 pcre_uchar *bracode = ketcode - GET(ketcode, 1);
5366
5367 /* Convert possessive ONCE brackets to non-capturing */
5368
5369 if ((*bracode == OP_ONCE || *bracode == OP_ONCE_NC) &&
5370 possessive_quantifier) *bracode = OP_BRA;
5371
5372 /* For non-possessive ONCE brackets, all we need to do is to
5373 set the KET. */
5374
5375 if (*bracode == OP_ONCE || *bracode == OP_ONCE_NC)
5376 *ketcode = OP_KETRMAX + repeat_type;
5377
5378 /* Handle non-ONCE brackets and possessive ONCEs (which have been
5379 converted to non-capturing above). */
5380
5381 else
5382 {
5383 /* In the compile phase, check for empty string matching. */
5384
5385 if (lengthptr == NULL)
5386 {
5387 pcre_uchar *scode = bracode;
5388 do
5389 {
5390 if (could_be_empty_branch(scode, ketcode, utf, cd))
5391 {
5392 *bracode += OP_SBRA - OP_BRA;
5393 break;
5394 }
5395 scode += GET(scode, 1);
5396 }
5397 while (*scode == OP_ALT);
5398 }
5399
5400 /* Handle possessive quantifiers. */
5401
5402 if (possessive_quantifier)
5403 {
5404 /* For COND brackets, we wrap the whole thing in a possessively
5405 repeated non-capturing bracket, because we have not invented POS
5406 versions of the COND opcodes. Because we are moving code along, we
5407 must ensure that any pending recursive references are updated. */
5408
5409 if (*bracode == OP_COND || *bracode == OP_SCOND)
5410 {
5411 int nlen = (int)(code - bracode);
5412 *code = OP_END;
5413 adjust_recurse(bracode, 1 + LINK_SIZE, utf, cd, save_hwm);
5414 memmove(bracode + 1 + LINK_SIZE, bracode, IN_UCHARS(nlen));
5415 code += 1 + LINK_SIZE;
5416 nlen += 1 + LINK_SIZE;
5417 *bracode = OP_BRAPOS;
5418 *code++ = OP_KETRPOS;
5419 PUTINC(code, 0, nlen);
5420 PUT(bracode, 1, nlen);
5421 }
5422
5423 /* For non-COND brackets, we modify the BRA code and use KETRPOS. */
5424
5425 else
5426 {
5427 *bracode += 1; /* Switch to xxxPOS opcodes */
5428 *ketcode = OP_KETRPOS;
5429 }
5430
5431 /* If the minimum is zero, mark it as possessive, then unset the
5432 possessive flag when the minimum is 0 or 1. */
5433
5434 if (brazeroptr != NULL) *brazeroptr = OP_BRAPOSZERO;
5435 if (repeat_min < 2) possessive_quantifier = FALSE;
5436 }
5437
5438 /* Non-possessive quantifier */
5439
5440 else *ketcode = OP_KETRMAX + repeat_type;
5441 }
5442 }
5443 }
5444
5445 /* If previous is OP_FAIL, it was generated by an empty class [] in
5446 JavaScript mode. The other ways in which OP_FAIL can be generated, that is
5447 by (*FAIL) or (?!) set previous to NULL, which gives a "nothing to repeat"
5448 error above. We can just ignore the repeat in JS case. */
5449
5450 else if (*previous == OP_FAIL) goto END_REPEAT;
5451
5452 /* Else there's some kind of shambles */
5453
5454 else
5455 {
5456 *errorcodeptr = ERR11;
5457 goto FAILED;
5458 }
5459
5460 /* If the character following a repeat is '+', or if certain optimization
5461 tests above succeeded, possessive_quantifier is TRUE. For some opcodes,
5462 there are special alternative opcodes for this case. For anything else, we
5463 wrap the entire repeated item inside OP_ONCE brackets. Logically, the '+'
5464 notation is just syntactic sugar, taken from Sun's Java package, but the
5465 special opcodes can optimize it.
5466
5467 Some (but not all) possessively repeated subpatterns have already been
5468 completely handled in the code just above. For them, possessive_quantifier
5469 is always FALSE at this stage.
5470
5471 Note that the repeated item starts at tempcode, not at previous, which
5472 might be the first part of a string whose (former) last char we repeated.
5473
5474 Possessifying an 'exact' quantifier has no effect, so we can ignore it. But
5475 an 'upto' may follow. We skip over an 'exact' item, and then test the
5476 length of what remains before proceeding. */
5477
5478 if (possessive_quantifier)
5479 {
5480 int len;
5481
5482 if (*tempcode == OP_TYPEEXACT)
5483 tempcode += PRIV(OP_lengths)[*tempcode] +
5484 ((tempcode[1 + IMM2_SIZE] == OP_PROP
5485 || tempcode[1 + IMM2_SIZE] == OP_NOTPROP)? 2 : 0);
5486
5487 else if (*tempcode == OP_EXACT || *tempcode == OP_NOTEXACT)
5488 {
5489 tempcode += PRIV(OP_lengths)[*tempcode];
5490 #ifdef SUPPORT_UTF
5491 if (utf && HAS_EXTRALEN(tempcode[-1]))
5492 tempcode += GET_EXTRALEN(tempcode[-1]);
5493 #endif
5494 }
5495
5496 len = (int)(code - tempcode);
5497 if (len > 0) switch (*tempcode)
5498 {
5499 case OP_STAR: *tempcode = OP_POSSTAR; break;
5500 case OP_PLUS: *tempcode = OP_POSPLUS; break;
5501 case OP_QUERY: *tempcode = OP_POSQUERY; break;
5502 case OP_UPTO: *tempcode = OP_POSUPTO; break;
5503
5504 case OP_STARI: *tempcode = OP_POSSTARI; break;
5505 case OP_PLUSI: *tempcode = OP_POSPLUSI; break;
5506 case OP_QUERYI: *tempcode = OP_POSQUERYI; break;
5507 case OP_UPTOI: *tempcode = OP_POSUPTOI; break;
5508
5509 case OP_NOTSTAR: *tempcode = OP_NOTPOSSTAR; break;
5510 case OP_NOTPLUS: *tempcode = OP_NOTPOSPLUS; break;
5511 case OP_NOTQUERY: *tempcode = OP_NOTPOSQUERY; break;
5512 case OP_NOTUPTO: *tempcode = OP_NOTPOSUPTO; break;
5513
5514 case OP_NOTSTARI: *tempcode = OP_NOTPOSSTARI; break;
5515 case OP_NOTPLUSI: *tempcode = OP_NOTPOSPLUSI; break;
5516 case OP_NOTQUERYI: *tempcode = OP_NOTPOSQUERYI; break;
5517 case OP_NOTUPTOI: *tempcode = OP_NOTPOSUPTOI; break;
5518
5519 case OP_TYPESTAR: *tempcode = OP_TYPEPOSSTAR; break;
5520 case OP_TYPEPLUS: *tempcode = OP_TYPEPOSPLUS; break;
5521 case OP_TYPEQUERY: *tempcode = OP_TYPEPOSQUERY; break;
5522 case OP_TYPEUPTO: *tempcode = OP_TYPEPOSUPTO; break;
5523
5524 /* Because we are moving code along, we must ensure that any
5525 pending recursive references are updated. */
5526
5527 default:
5528 *code = OP_END;
5529 adjust_recurse(tempcode, 1 + LINK_SIZE, utf, cd, save_hwm);
5530 memmove(tempcode + 1 + LINK_SIZE, tempcode, IN_UCHARS(len));
5531 code += 1 + LINK_SIZE;
5532 len += 1 + LINK_SIZE;
5533 tempcode[0] = OP_ONCE;
5534 *code++ = OP_KET;
5535 PUTINC(code, 0, len);
5536 PUT(tempcode, 1, len);
5537 break;
5538 }
5539 }
5540
5541 /* In all case we no longer have a previous item. We also set the
5542 "follows varying string" flag for subsequently encountered reqchars if
5543 it isn't already set and we have just passed a varying length item. */
5544
5545 END_REPEAT:
5546 previous = NULL;
5547 cd->req_varyopt |= reqvary;
5548 break;
5549
5550
5551 /* ===================================================================*/
5552 /* Start of nested parenthesized sub-expression, or comment or lookahead or
5553 lookbehind or option setting or condition or all the other extended
5554 parenthesis forms. */
5555
5556 case CHAR_LEFT_PARENTHESIS:
5557 newoptions = options;
5558 skipbytes = 0;
5559 bravalue = OP_CBRA;
5560 save_hwm = cd->hwm;
5561 reset_bracount = FALSE;
5562
5563 /* First deal with various "verbs" that can be introduced by '*'. */
5564
5565 ptr++;
5566 if (ptr[0] == CHAR_ASTERISK && (ptr[1] == ':'
5567 || (MAX_255(ptr[1]) && ((cd->ctypes[ptr[1]] & ctype_letter) != 0))))
5568 {
5569 int i, namelen;
5570 int arglen = 0;
5571 const char *vn = verbnames;
5572 const pcre_uchar *name = ptr + 1;
5573 const pcre_uchar *arg = NULL;
5574 previous = NULL;
5575 ptr++;
5576 while (MAX_255(*ptr) && (cd->ctypes[*ptr] & ctype_letter) != 0) ptr++;
5577 namelen = (int)(ptr - name);
5578
5579 /* It appears that Perl allows any characters whatsoever, other than
5580 a closing parenthesis, to appear in arguments, so we no longer insist on
5581 letters, digits, and underscores. */
5582
5583 if (*ptr == CHAR_COLON)
5584 {
5585 arg = ++ptr;
5586 while (*ptr != 0 && *ptr != CHAR_RIGHT_PARENTHESIS) ptr++;
5587 arglen = (int)(ptr - arg);
5588 }
5589
5590 if (*ptr != CHAR_RIGHT_PARENTHESIS)
5591 {
5592 *errorcodeptr = ERR60;
5593 goto FAILED;
5594 }
5595
5596 /* Scan the table of verb names */
5597
5598 for (i = 0; i < verbcount; i++)
5599 {
5600 if (namelen == verbs[i].len &&
5601 STRNCMP_UC_C8(name, vn, namelen) == 0)
5602 {
5603 /* Check for open captures before ACCEPT and convert it to
5604 ASSERT_ACCEPT if in an assertion. */
5605
5606 if (verbs[i].op == OP_ACCEPT)
5607 {
5608 open_capitem *oc;
5609 if (arglen != 0)
5610 {
5611 *errorcodeptr = ERR59;
5612 goto FAILED;
5613 }
5614 cd->had_accept = TRUE;
5615 for (oc = cd->open_caps; oc != NULL; oc = oc->next)
5616 {
5617 *code++ = OP_CLOSE;
5618 PUT2INC(code, 0, oc->number);
5619 }
5620 *code++ = (cd->assert_depth > 0)? OP_ASSERT_ACCEPT : OP_ACCEPT;
5621
5622 /* Do not set firstchar after *ACCEPT */
5623 if (firstchar == REQ_UNSET) firstchar = REQ_NONE;
5624 }
5625
5626 /* Handle other cases with/without an argument */
5627
5628 else if (arglen == 0)
5629 {
5630 if (verbs[i].op < 0) /* Argument is mandatory */
5631 {
5632 *errorcodeptr = ERR66;
5633 goto FAILED;
5634 }
5635 *code = verbs[i].op;
5636 if (*code++ == OP_THEN) cd->external_flags |= PCRE_HASTHEN;
5637 }
5638
5639 else
5640 {
5641 if (verbs[i].op_arg < 0) /* Argument is forbidden */
5642 {
5643 *errorcodeptr = ERR59;
5644 goto FAILED;
5645 }
5646 *code = verbs[i].op_arg;
5647 if (*code++ == OP_THEN_ARG) cd->external_flags |= PCRE_HASTHEN;
5648 *code++ = arglen;
5649 memcpy(code, arg, IN_UCHARS(arglen));
5650 code += arglen;
5651 *code++ = 0;
5652 }
5653
5654 break; /* Found verb, exit loop */
5655 }
5656
5657 vn += verbs[i].len + 1;
5658 }
5659
5660 if (i < verbcount) continue; /* Successfully handled a verb */
5661 *errorcodeptr = ERR60; /* Verb not recognized */
5662 goto FAILED;
5663 }
5664
5665 /* Deal with the extended parentheses; all are introduced by '?', and the
5666 appearance of any of them means that this is not a capturing group. */
5667
5668 else if (*ptr == CHAR_QUESTION_MARK)
5669 {
5670 int i, set, unset, namelen;
5671 int *optset;
5672 const pcre_uchar *name;
5673 pcre_uchar *slot;
5674
5675 switch (*(++ptr))
5676 {
5677 case CHAR_NUMBER_SIGN: /* Comment; skip to ket */
5678 ptr++;
5679 while (*ptr != 0 && *ptr != CHAR_RIGHT_PARENTHESIS) ptr++;
5680 if (*ptr == 0)
5681 {
5682 *errorcodeptr = ERR18;
5683 goto FAILED;
5684 }
5685 continue;
5686
5687
5688 /* ------------------------------------------------------------ */
5689 case CHAR_VERTICAL_LINE: /* Reset capture count for each branch */
5690 reset_bracount = TRUE;
5691 /* Fall through */
5692
5693 /* ------------------------------------------------------------ */
5694 case CHAR_COLON: /* Non-capturing bracket */
5695 bravalue = OP_BRA;
5696 ptr++;
5697 break;
5698
5699
5700 /* ------------------------------------------------------------ */
5701 case CHAR_LEFT_PARENTHESIS:
5702 bravalue = OP_COND; /* Conditional group */
5703
5704 /* A condition can be an assertion, a number (referring to a numbered
5705 group), a name (referring to a named group), or 'R', referring to
5706 recursion. R<digits> and R&name are also permitted for recursion tests.
5707
5708 There are several syntaxes for testing a named group: (?(name)) is used
5709 by Python; Perl 5.10 onwards uses (?(<name>) or (?('name')).
5710
5711 There are two unfortunate ambiguities, caused by history. (a) 'R' can
5712 be the recursive thing or the name 'R' (and similarly for 'R' followed
5713 by digits), and (b) a number could be a name that consists of digits.
5714 In both cases, we look for a name first; if not found, we try the other
5715 cases. */
5716
5717 /* For conditions that are assertions, check the syntax, and then exit
5718 the switch. This will take control down to where bracketed groups,
5719 including assertions, are processed. */
5720
5721 if (ptr[1] == CHAR_QUESTION_MARK && (ptr[2] == CHAR_EQUALS_SIGN ||
5722 ptr[2] == CHAR_EXCLAMATION_MARK || ptr[2] == CHAR_LESS_THAN_SIGN))
5723 break;
5724
5725 /* Most other conditions use OP_CREF (a couple change to OP_RREF
5726 below), and all need to skip 1+IMM2_SIZE bytes at the start of the group. */
5727
5728 code[1+LINK_SIZE] = OP_CREF;
5729 skipbytes = 1+IMM2_SIZE;
5730 refsign = -1;
5731
5732 /* Check for a test for recursion in a named group. */
5733
5734 if (ptr[1] == CHAR_R && ptr[2] == CHAR_AMPERSAND)
5735 {
5736 terminator = -1;
5737 ptr += 2;
5738 code[1+LINK_SIZE] = OP_RREF; /* Change the type of test */
5739 }
5740
5741 /* Check for a test for a named group's having been set, using the Perl
5742 syntax (?(<name>) or (?('name') */
5743
5744 else if (ptr[1] == CHAR_LESS_THAN_SIGN)
5745 {
5746 terminator = CHAR_GREATER_THAN_SIGN;
5747 ptr++;
5748 }
5749 else if (ptr[1] == CHAR_APOSTROPHE)
5750 {
5751 terminator = CHAR_APOSTROPHE;
5752 ptr++;
5753 }
5754 else
5755 {
5756 terminator = 0;
5757 if (ptr[1] == CHAR_MINUS || ptr[1] == CHAR_PLUS) refsign = *(++ptr);
5758 }
5759
5760 /* We now expect to read a name; any thing else is an error */
5761
5762 if (!MAX_255(ptr[1]) || (cd->ctypes[ptr[1]] & ctype_word) == 0)
5763 {
5764 ptr += 1; /* To get the right offset */
5765 *errorcodeptr = ERR28;
5766 goto FAILED;
5767 }
5768
5769 /* Read the name, but also get it as a number if it's all digits */
5770
5771 recno = 0;
5772 name = ++ptr;
5773 while (MAX_255(*ptr) && (cd->ctypes[*ptr] & ctype_word) != 0)
5774 {
5775 if (recno >= 0)
5776 recno = (IS_DIGIT(*ptr))? recno * 10 + *ptr - CHAR_0 : -1;
5777 ptr++;
5778 }
5779 namelen = (int)(ptr - name);
5780
5781 if ((terminator > 0 && *ptr++ != terminator) ||
5782 *ptr++ != CHAR_RIGHT_PARENTHESIS)
5783 {
5784 ptr--; /* Error offset */
5785 *errorcodeptr = ERR26;
5786 goto FAILED;
5787 }
5788
5789 /* Do no further checking in the pre-compile phase. */
5790
5791 if (lengthptr != NULL) break;
5792
5793 /* In the real compile we do the work of looking for the actual
5794 reference. If the string started with "+" or "-" we require the rest to
5795 be digits, in which case recno will be set. */
5796
5797 if (refsign > 0)
5798 {
5799 if (recno <= 0)
5800 {
5801 *errorcodeptr = ERR58;
5802 goto FAILED;
5803 }
5804 recno = (refsign == CHAR_MINUS)?
5805 cd->bracount - recno + 1 : recno +cd->bracount;
5806 if (recno <= 0 || recno > cd->final_bracount)
5807 {
5808 *errorcodeptr = ERR15;
5809 goto FAILED;
5810 }
5811 PUT2(code, 2+LINK_SIZE, recno);
5812 break;
5813 }
5814
5815 /* Otherwise (did not start with "+" or "-"), start by looking for the
5816 name. If we find a name, add one to the opcode to change OP_CREF or
5817 OP_RREF into OP_NCREF or OP_NRREF. These behave exactly the same,
5818 except they record that the reference was originally to a name. The
5819 information is used to check duplicate names. */
5820
5821 slot = cd->name_table;
5822 for (i = 0; i < cd->names_found; i++)
5823 {
5824 if (STRNCMP_UC_UC(name, slot+IMM2_SIZE, namelen) == 0) break;
5825 slot += cd->name_entry_size;
5826 }
5827
5828 /* Found a previous named subpattern */
5829
5830 if (i < cd->names_found)
5831 {
5832 recno = GET2(slot, 0);
5833 PUT2(code, 2+LINK_SIZE, recno);
5834 code[1+LINK_SIZE]++;
5835 }
5836
5837 /* Search the pattern for a forward reference */
5838
5839 else if ((i = find_parens(cd, name, namelen,
5840 (options & PCRE_EXTENDED) != 0, utf)) > 0)
5841 {
5842 PUT2(code, 2+LINK_SIZE, i);
5843 code[1+LINK_SIZE]++;
5844 }
5845
5846 /* If terminator == 0 it means that the name followed directly after
5847 the opening parenthesis [e.g. (?(abc)...] and in this case there are
5848 some further alternatives to try. For the cases where terminator != 0
5849 [things like (?(<name>... or (?('name')... or (?(R&name)... ] we have
5850 now checked all the possibilities, so give an error. */
5851
5852 else if (terminator != 0)
5853 {
5854 *errorcodeptr = ERR15;
5855 goto FAILED;
5856 }
5857
5858 /* Check for (?(R) for recursion. Allow digits after R to specify a
5859 specific group number. */
5860
5861 else if (*name == CHAR_R)
5862 {
5863 recno = 0;
5864 for (i = 1; i < namelen; i++)
5865 {
5866 if (!IS_DIGIT(name[i]))
5867 {
5868 *errorcodeptr = ERR15;
5869 goto FAILED;
5870 }
5871 recno = recno * 10 + name[i] - CHAR_0;
5872 }
5873 if (recno == 0) recno = RREF_ANY;
5874 code[1+LINK_SIZE] = OP_RREF; /* Change test type */
5875 PUT2(code, 2+LINK_SIZE, recno);
5876 }
5877
5878 /* Similarly, check for the (?(DEFINE) "condition", which is always
5879 false. */
5880
5881 else if (namelen == 6 && STRNCMP_UC_C8(name, STRING_DEFINE, 6) == 0)
5882 {
5883 code[1+LINK_SIZE] = OP_DEF;
5884 skipbytes = 1;
5885 }
5886
5887 /* Check for the "name" actually being a subpattern number. We are
5888 in the second pass here, so final_bracount is set. */
5889
5890 else if (recno > 0 && recno <= cd->final_bracount)
5891 {
5892 PUT2(code, 2+LINK_SIZE, recno);
5893 }
5894
5895 /* Either an unidentified subpattern, or a reference to (?(0) */
5896
5897 else
5898 {
5899 *errorcodeptr = (recno == 0)? ERR35: ERR15;
5900 goto FAILED;
5901 }
5902 break;
5903
5904
5905 /* ------------------------------------------------------------ */
5906 case CHAR_EQUALS_SIGN: /* Positive lookahead */
5907 bravalue = OP_ASSERT;
5908 cd->assert_depth += 1;
5909 ptr++;
5910 break;
5911
5912
5913 /* ------------------------------------------------------------ */
5914 case CHAR_EXCLAMATION_MARK: /* Negative lookahead */
5915 ptr++;
5916 if (*ptr == CHAR_RIGHT_PARENTHESIS) /* Optimize (?!) */
5917 {
5918 *code++ = OP_FAIL;
5919 previous = NULL;
5920 continue;
5921 }
5922 bravalue = OP_ASSERT_NOT;
5923 cd->assert_depth += 1;
5924 break;
5925
5926
5927 /* ------------------------------------------------------------ */
5928 case CHAR_LESS_THAN_SIGN: /* Lookbehind or named define */
5929 switch (ptr[1])
5930 {
5931 case CHAR_EQUALS_SIGN: /* Positive lookbehind */
5932 bravalue = OP_ASSERTBACK;
5933 cd->assert_depth += 1;
5934 ptr += 2;
5935 break;
5936
5937 case CHAR_EXCLAMATION_MARK: /* Negative lookbehind */
5938 bravalue = OP_ASSERTBACK_NOT;
5939 cd->assert_depth += 1;
5940 ptr += 2;
5941 break;
5942
5943 default: /* Could be name define, else bad */
5944 if (MAX_255(ptr[1]) && (cd->ctypes[ptr[1]] & ctype_word) != 0)
5945 goto DEFINE_NAME;
5946 ptr++; /* Correct offset for error */
5947 *errorcodeptr = ERR24;
5948 goto FAILED;
5949 }
5950 break;
5951
5952
5953 /* ------------------------------------------------------------ */
5954 case CHAR_GREATER_THAN_SIGN: /* One-time brackets */
5955 bravalue = OP_ONCE;
5956 ptr++;
5957 break;
5958
5959
5960 /* ------------------------------------------------------------ */
5961 case CHAR_C: /* Callout - may be followed by digits; */
5962 previous_callout = code; /* Save for later completion */
5963 after_manual_callout = 1; /* Skip one item before completing */
5964 *code++ = OP_CALLOUT;
5965 {
5966 int n = 0;
5967 ptr++;
5968 while(IS_DIGIT(*ptr))
5969 n = n * 10 + *ptr++ - CHAR_0;
5970 if (*ptr != CHAR_RIGHT_PARENTHESIS)
5971 {
5972 *errorcodeptr = ERR39;
5973 goto FAILED;
5974 }
5975 if (n > 255)
5976 {
5977 *errorcodeptr = ERR38;
5978 goto FAILED;
5979 }
5980 *code++ = n;
5981 PUT(code, 0, (int)(ptr - cd->start_pattern + 1)); /* Pattern offset */
5982 PUT(code, LINK_SIZE, 0); /* Default length */
5983 code += 2 * LINK_SIZE;
5984 }
5985 previous = NULL;
5986 continue;
5987
5988
5989 /* ------------------------------------------------------------ */
5990 case CHAR_P: /* Python-style named subpattern handling */
5991 if (*(++ptr) == CHAR_EQUALS_SIGN ||
5992 *ptr == CHAR_GREATER_THAN_SIGN) /* Reference or recursion */
5993 {
5994 is_recurse = *ptr == CHAR_GREATER_THAN_SIGN;
5995 terminator = CHAR_RIGHT_PARENTHESIS;
5996 goto NAMED_REF_OR_RECURSE;
5997 }
5998 else if (*ptr != CHAR_LESS_THAN_SIGN) /* Test for Python-style defn */
5999 {
6000 *errorcodeptr = ERR41;
6001 goto FAILED;
6002 }
6003 /* Fall through to handle (?P< as (?< is handled */
6004
6005
6006 /* ------------------------------------------------------------ */
6007 DEFINE_NAME: /* Come here from (?< handling */
6008 case CHAR_APOSTROPHE:
6009 {
6010 terminator = (*ptr == CHAR_LESS_THAN_SIGN)?
6011 CHAR_GREATER_THAN_SIGN : CHAR_APOSTROPHE;
6012 name = ++ptr;
6013
6014 while (MAX_255(*ptr) && (cd->ctypes[*ptr] & ctype_word) != 0) ptr++;
6015 namelen = (int)(ptr - name);
6016
6017 /* In the pre-compile phase, just do a syntax check. */
6018
6019 if (lengthptr != NULL)
6020 {
6021 if (*ptr != terminator)
6022 {
6023 *errorcodeptr = ERR42;
6024 goto FAILED;
6025 }
6026 if (cd->names_found >= MAX_NAME_COUNT)
6027 {
6028 *errorcodeptr = ERR49;
6029 goto FAILED;
6030 }
6031 if (namelen + IMM2_SIZE + 1 > cd->name_entry_size)
6032 {
6033 cd->name_entry_size = namelen + IMM2_SIZE + 1;
6034 if (namelen > MAX_NAME_SIZE)
6035 {
6036 *errorcodeptr = ERR48;
6037 goto FAILED;
6038 }
6039 }
6040 }
6041
6042 /* In the real compile, create the entry in the table, maintaining
6043 alphabetical order. Duplicate names for different numbers are
6044 permitted only if PCRE_DUPNAMES is set. Duplicate names for the same
6045 number are always OK. (An existing number can be re-used if (?|
6046 appears in the pattern.) In either event, a duplicate name results in
6047 a duplicate entry in the table, even if the number is the same. This
6048 is because the number of names, and hence the table size, is computed
6049 in the pre-compile, and it affects various numbers and pointers which
6050 would all have to be modified, and the compiled code moved down, if
6051 duplicates with the same number were omitted from the table. This
6052 doesn't seem worth the hassle. However, *different* names for the
6053 same number are not permitted. */
6054
6055 else
6056 {
6057 BOOL dupname = FALSE;
6058 slot = cd->name_table;
6059
6060 for (i = 0; i < cd->names_found; i++)
6061 {
6062 int crc = memcmp(name, slot+IMM2_SIZE, IN_UCHARS(namelen));
6063 if (crc == 0)
6064 {
6065 if (slot[IMM2_SIZE+namelen] == 0)
6066 {
6067 if (GET2(slot, 0) != cd->bracount + 1 &&
6068 (options & PCRE_DUPNAMES) == 0)
6069 {
6070 *errorcodeptr = ERR43;
6071 goto FAILED;
6072 }
6073 else dupname = TRUE;
6074 }
6075 else crc = -1; /* Current name is a substring */
6076 }
6077
6078 /* Make space in the table and break the loop for an earlier
6079 name. For a duplicate or later name, carry on. We do this for
6080 duplicates so that in the simple case (when ?(| is not used) they
6081 are in order of their numbers. */
6082
6083 if (crc < 0)
6084 {
6085 memmove(slot + cd->name_entry_size, slot,
6086 IN_UCHARS((cd->names_found - i) * cd->name_entry_size));
6087 break;
6088 }
6089
6090 /* Continue the loop for a later or duplicate name */
6091
6092 slot += cd->name_entry_size;
6093 }
6094
6095 /* For non-duplicate names, check for a duplicate number before
6096 adding the new name. */
6097
6098 if (!dupname)
6099 {
6100 pcre_uchar *cslot = cd->name_table;
6101 for (i = 0; i < cd->names_found; i++)
6102 {
6103 if (cslot != slot)
6104 {
6105 if (GET2(cslot, 0) == cd->bracount + 1)
6106 {
6107 *errorcodeptr = ERR65;
6108 goto FAILED;
6109 }
6110 }
6111 else i--;
6112 cslot += cd->name_entry_size;
6113 }
6114 }
6115
6116 PUT2(slot, 0, cd->bracount + 1);
6117 memcpy(slot + IMM2_SIZE, name, IN_UCHARS(namelen));
6118 slot[IMM2_SIZE + namelen] = 0;
6119 }
6120 }
6121
6122 /* In both pre-compile and compile, count the number of names we've
6123 encountered. */
6124
6125 cd->names_found++;
6126 ptr++; /* Move past > or ' */
6127 goto NUMBERED_GROUP;
6128
6129
6130 /* ------------------------------------------------------------ */
6131 case CHAR_AMPERSAND: /* Perl recursion/subroutine syntax */
6132 terminator = CHAR_RIGHT_PARENTHESIS;
6133 is_recurse = TRUE;
6134 /* Fall through */
6135
6136 /* We come here from the Python syntax above that handles both
6137 references (?P=name) and recursion (?P>name), as well as falling
6138 through from the Perl recursion syntax (?&name). We also come here from
6139 the Perl \k<name> or \k'name' back reference syntax and the \k{name}
6140 .NET syntax, and the Oniguruma \g<...> and \g'...' subroutine syntax. */
6141
6142 NAMED_REF_OR_RECURSE:
6143 name = ++ptr;
6144 while (MAX_255(*ptr) && (cd->ctypes[*ptr] & ctype_word) != 0) ptr++;
6145 namelen = (int)(ptr - name);
6146
6147 /* In the pre-compile phase, do a syntax check. We used to just set
6148 a dummy reference number, because it was not used in the first pass.
6149 However, with the change of recursive back references to be atomic,
6150 we have to look for the number so that this state can be identified, as
6151 otherwise the incorrect length is computed. If it's not a backwards
6152 reference, the dummy number will do. */
6153
6154 if (lengthptr != NULL)
6155 {
6156 const pcre_uchar *temp;
6157
6158 if (namelen == 0)
6159 {
6160 *errorcodeptr = ERR62;
6161 goto FAILED;
6162 }
6163 if (*ptr != terminator)
6164 {
6165 *errorcodeptr = ERR42;
6166 goto FAILED;
6167 }
6168 if (namelen > MAX_NAME_SIZE)
6169 {
6170 *errorcodeptr = ERR48;
6171 goto FAILED;
6172 }
6173
6174 /* The name table does not exist in the first pass, so we cannot
6175 do a simple search as in the code below. Instead, we have to scan the
6176 pattern to find the number. It is important that we scan it only as
6177 far as we have got because the syntax of named subpatterns has not
6178 been checked for the rest of the pattern, and find_parens() assumes
6179 correct syntax. In any case, it's a waste of resources to scan
6180 further. We stop the scan at the current point by temporarily
6181 adjusting the value of cd->endpattern. */
6182
6183 temp = cd->end_pattern;
6184 cd->end_pattern = ptr;
6185 recno = find_parens(cd, name, namelen,
6186 (options & PCRE_EXTENDED) != 0, utf);
6187 cd->end_pattern = temp;
6188 if (recno < 0) recno = 0; /* Forward ref; set dummy number */
6189 }
6190
6191 /* In the real compile, seek the name in the table. We check the name
6192 first, and then check that we have reached the end of the name in the
6193 table. That way, if the name that is longer than any in the table,
6194 the comparison will fail without reading beyond the table entry. */
6195
6196 else
6197 {
6198 slot = cd->name_table;
6199 for (i = 0; i < cd->names_found; i++)
6200 {
6201 if (STRNCMP_UC_UC(name, slot+IMM2_SIZE, namelen) == 0 &&
6202 slot[IMM2_SIZE+namelen] == 0)
6203 break;
6204 slot += cd->name_entry_size;
6205 }
6206
6207 if (i < cd->names_found) /* Back reference */
6208 {
6209 recno = GET2(slot, 0);
6210 }
6211 else if ((recno = /* Forward back reference */
6212 find_parens(cd, name, namelen,
6213 (options & PCRE_EXTENDED) != 0, utf)) <= 0)
6214 {
6215 *errorcodeptr = ERR15;
6216 goto FAILED;
6217 }
6218 }
6219
6220 /* In both phases, we can now go to the code than handles numerical
6221 recursion or backreferences. */
6222
6223 if (is_recurse) goto HANDLE_RECURSION;
6224 else goto HANDLE_REFERENCE;
6225
6226
6227 /* ------------------------------------------------------------ */
6228 case CHAR_R: /* Recursion */
6229 ptr++; /* Same as (?0) */
6230 /* Fall through */
6231
6232
6233 /* ------------------------------------------------------------ */
6234 case CHAR_MINUS: case CHAR_PLUS: /* Recursion or subroutine */
6235 case CHAR_0: case CHAR_1: case CHAR_2: case CHAR_3: case CHAR_4:
6236 case CHAR_5: case CHAR_6: case CHAR_7: case CHAR_8: case CHAR_9:
6237 {
6238 const pcre_uchar *called;
6239 terminator = CHAR_RIGHT_PARENTHESIS;
6240
6241 /* Come here from the \g<...> and \g'...' code (Oniguruma
6242 compatibility). However, the syntax has been checked to ensure that
6243 the ... are a (signed) number, so that neither ERR63 nor ERR29 will
6244 be called on this path, nor with the jump to OTHER_CHAR_AFTER_QUERY
6245 ever be taken. */
6246
6247 HANDLE_NUMERICAL_RECURSION:
6248
6249 if ((refsign = *ptr) == CHAR_PLUS)
6250 {
6251 ptr++;
6252 if (!IS_DIGIT(*ptr))
6253 {
6254 *errorcodeptr = ERR63;
6255 goto FAILED;
6256 }
6257 }
6258 else if (refsign == CHAR_MINUS)
6259 {
6260 if (!IS_DIGIT(ptr[1]))
6261 goto OTHER_CHAR_AFTER_QUERY;
6262 ptr++;
6263 }
6264
6265 recno = 0;
6266 while(IS_DIGIT(*ptr))
6267 recno = recno * 10 + *ptr++ - CHAR_0;
6268
6269 if (*ptr != terminator)
6270 {
6271 *errorcodeptr = ERR29;
6272 goto FAILED;
6273 }
6274
6275 if (refsign == CHAR_MINUS)
6276 {
6277 if (recno == 0)
6278 {
6279 *errorcodeptr = ERR58;
6280 goto FAILED;
6281 }
6282 recno = cd->bracount - recno + 1;
6283 if (recno <= 0)
6284 {
6285 *errorcodeptr = ERR15;
6286 goto FAILED;
6287 }
6288 }
6289 else if (refsign == CHAR_PLUS)
6290 {
6291 if (recno == 0)
6292 {
6293 *errorcodeptr = ERR58;
6294 goto FAILED;
6295 }
6296 recno += cd->bracount;
6297 }
6298
6299 /* Come here from code above that handles a named recursion */
6300
6301 HANDLE_RECURSION:
6302
6303 previous = code;
6304 called = cd->start_code;
6305
6306 /* When we are actually compiling, find the bracket that is being
6307 referenced. Temporarily end the regex in case it doesn't exist before
6308 this point. If we end up with a forward reference, first check that
6309 the bracket does occur later so we can give the error (and position)
6310 now. Then remember this forward reference in the workspace so it can
6311 be filled in at the end. */
6312
6313 if (lengthptr == NULL)
6314 {
6315 *code = OP_END;
6316 if (recno != 0)
6317 called = PRIV(find_bracket)(cd->start_code, utf, recno);
6318
6319 /* Forward reference */
6320
6321 if (called == NULL)
6322 {
6323 if (find_parens(cd, NULL, recno,
6324 (options & PCRE_EXTENDED) != 0, utf) < 0)
6325 {
6326 *errorcodeptr = ERR15;
6327 goto FAILED;
6328 }
6329
6330 /* Fudge the value of "called" so that when it is inserted as an
6331 offset below, what it actually inserted is the reference number
6332 of the group. Then remember the forward reference. */
6333
6334 called = cd->start_code + recno;
6335 if (cd->hwm >= cd->start_workspace + cd->workspace_size -
6336 WORK_SIZE_SAFETY_MARGIN)
6337 {
6338 *errorcodeptr = expand_workspace(cd);
6339 if (*errorcodeptr != 0) goto FAILED;
6340 }
6341 PUTINC(cd->hwm, 0, (int)(code + 1 - cd->start_code));
6342 }
6343
6344 /* If not a forward reference, and the subpattern is still open,
6345 this is a recursive call. We check to see if this is a left
6346 recursion that could loop for ever, and diagnose that case. We
6347 must not, however, do this check if we are in a conditional
6348 subpattern because the condition might be testing for recursion in
6349 a pattern such as /(?(R)a+|(?R)b)/, which is perfectly valid.
6350 Forever loops are also detected at runtime, so those that occur in
6351 conditional subpatterns will be picked up then. */
6352
6353 else if (GET(called, 1) == 0 && cond_depth <= 0 &&
6354 could_be_empty(called, code, bcptr, utf, cd))
6355 {
6356 *errorcodeptr = ERR40;
6357 goto FAILED;
6358 }
6359 }
6360
6361 /* Insert the recursion/subroutine item. It does not have a set first
6362 character (relevant if it is repeated, because it will then be
6363 wrapped with ONCE brackets). */
6364
6365 *code = OP_RECURSE;
6366 PUT(code, 1, (int)(called - cd->start_code));
6367 code += 1 + LINK_SIZE;
6368 groupsetfirstchar = FALSE;
6369 }
6370
6371 /* Can't determine a first byte now */
6372
6373 if (firstchar == REQ_UNSET) firstchar = REQ_NONE;
6374 continue;
6375
6376
6377 /* ------------------------------------------------------------ */
6378 default: /* Other characters: check option setting */
6379 OTHER_CHAR_AFTER_QUERY:
6380 set = unset = 0;
6381 optset = &set;
6382
6383 while (*ptr != CHAR_RIGHT_PARENTHESIS && *ptr != CHAR_COLON)
6384 {
6385 switch (*ptr++)
6386 {
6387 case CHAR_MINUS: optset = &unset; break;
6388
6389 case CHAR_J: /* Record that it changed in the external options */
6390 *optset |= PCRE_DUPNAMES;
6391 cd->external_flags |= PCRE_JCHANGED;
6392 break;
6393
6394 case CHAR_i: *optset |= PCRE_CASELESS; break;
6395 case CHAR_m: *optset |= PCRE_MULTILINE; break;
6396 case CHAR_s: *optset |= PCRE_DOTALL; break;
6397 case CHAR_x: *optset |= PCRE_EXTENDED; break;
6398 case CHAR_U: *optset |= PCRE_UNGREEDY; break;
6399 case CHAR_X: *optset |= PCRE_EXTRA; break;
6400
6401 default: *errorcodeptr = ERR12;
6402 ptr--; /* Correct the offset */
6403 goto FAILED;
6404 }
6405 }
6406
6407 /* Set up the changed option bits, but don't change anything yet. */
6408
6409 newoptions = (options | set) & (~unset);
6410
6411 /* If the options ended with ')' this is not the start of a nested
6412 group with option changes, so the options change at this level. If this
6413 item is right at the start of the pattern, the options can be
6414 abstracted and made external in the pre-compile phase, and ignored in
6415 the compile phase. This can be helpful when matching -- for instance in
6416 caseless checking of required bytes.
6417
6418 If the code pointer is not (cd->start_code + 1 + LINK_SIZE), we are
6419 definitely *not* at the start of the pattern because something has been
6420 compiled. In the pre-compile phase, however, the code pointer can have
6421 that value after the start, because it gets reset as code is discarded
6422 during the pre-compile. However, this can happen only at top level - if
6423 we are within parentheses, the starting BRA will still be present. At
6424 any parenthesis level, the length value can be used to test if anything
6425 has been compiled at that level. Thus, a test for both these conditions
6426 is necessary to ensure we correctly detect the start of the pattern in
6427 both phases.
6428
6429 If we are not at the pattern start, reset the greedy defaults and the
6430 case value for firstchar and reqchar. */
6431
6432 if (*ptr == CHAR_RIGHT_PARENTHESIS)
6433 {
6434 if (code == cd->start_code + 1 + LINK_SIZE &&
6435 (lengthptr == NULL || *lengthptr == 2 + 2*LINK_SIZE))
6436 {
6437 cd->external_options = newoptions;
6438 }
6439 else
6440 {
6441 greedy_default = ((newoptions & PCRE_UNGREEDY) != 0);
6442 greedy_non_default = greedy_default ^ 1;
6443 req_caseopt = ((newoptions & PCRE_CASELESS) != 0)? REQ_CASELESS:0;
6444 }
6445
6446 /* Change options at this level, and pass them back for use
6447 in subsequent branches. */
6448
6449 *optionsptr = options = newoptions;
6450 previous = NULL; /* This item can't be repeated */
6451 continue; /* It is complete */
6452 }
6453
6454 /* If the options ended with ':' we are heading into a nested group
6455 with possible change of options. Such groups are non-capturing and are
6456 not assertions of any kind. All we need to do is skip over the ':';
6457 the newoptions value is handled below. */
6458
6459 bravalue = OP_BRA;
6460 ptr++;
6461 } /* End of switch for character following (? */
6462 } /* End of (? handling */
6463
6464 /* Opening parenthesis not followed by '*' or '?'. If PCRE_NO_AUTO_CAPTURE
6465 is set, all unadorned brackets become non-capturing and behave like (?:...)
6466 brackets. */
6467
6468 else if ((options & PCRE_NO_AUTO_CAPTURE) != 0)
6469 {
6470 bravalue = OP_BRA;
6471 }
6472
6473 /* Else we have a capturing group. */
6474
6475 else
6476 {
6477 NUMBERED_GROUP:
6478 cd->bracount += 1;
6479 PUT2(code, 1+LINK_SIZE, cd->bracount);
6480 skipbytes = IMM2_SIZE;
6481 }
6482
6483 /* Process nested bracketed regex. Assertions used not to be repeatable,
6484 but this was changed for Perl compatibility, so all kinds can now be
6485 repeated. We copy code into a non-register variable (tempcode) in order to
6486 be able to pass its address because some compilers complain otherwise. */
6487
6488 previous = code; /* For handling repetition */
6489 *code = bravalue;
6490 tempcode = code;
6491 tempreqvary = cd->req_varyopt; /* Save value before bracket */
6492 tempbracount = cd->bracount; /* Save value before bracket */
6493 length_prevgroup = 0; /* Initialize for pre-compile phase */
6494
6495 if (!compile_regex(
6496 newoptions, /* The complete new option state */
6497 &tempcode, /* Where to put code (updated) */
6498 &ptr, /* Input pointer (updated) */
6499 errorcodeptr, /* Where to put an error message */
6500 (bravalue == OP_ASSERTBACK ||
6501 bravalue == OP_ASSERTBACK_NOT), /* TRUE if back assert */
6502 reset_bracount, /* True if (?| group */
6503 skipbytes, /* Skip over bracket number */
6504 cond_depth +
6505 ((bravalue == OP_COND)?1:0), /* Depth of condition subpatterns */
6506 &subfirstchar, /* For possible first char */
6507 &subreqchar, /* For possible last char */
6508 bcptr, /* Current branch chain */
6509 cd, /* Tables block */
6510 (lengthptr == NULL)? NULL : /* Actual compile phase */
6511 &length_prevgroup /* Pre-compile phase */
6512 ))
6513 goto FAILED;
6514
6515 /* If this was an atomic group and there are no capturing groups within it,
6516 generate OP_ONCE_NC instead of OP_ONCE. */
6517
6518 if (bravalue == OP_ONCE && cd->bracount <= tempbracount)
6519 *code = OP_ONCE_NC;
6520
6521 if (bravalue >= OP_ASSERT && bravalue <= OP_ASSERTBACK_NOT)
6522 cd->assert_depth -= 1;
6523
6524 /* At the end of compiling, code is still pointing to the start of the
6525 group, while tempcode has been updated to point past the end of the group.
6526 The pattern pointer (ptr) is on the bracket.
6527
6528 If this is a conditional bracket, check that there are no more than
6529 two branches in the group, or just one if it's a DEFINE group. We do this
6530 in the real compile phase, not in the pre-pass, where the whole group may
6531 not be available. */
6532
6533 if (bravalue == OP_COND && lengthptr == NULL)
6534 {
6535 pcre_uchar *tc = code;
6536 int condcount = 0;
6537
6538 do {
6539 condcount++;
6540 tc += GET(tc,1);
6541 }
6542 while (*tc != OP_KET);
6543
6544 /* A DEFINE group is never obeyed inline (the "condition" is always
6545 false). It must have only one branch. */
6546
6547 if (code[LINK_SIZE+1] == OP_DEF)
6548 {
6549 if (condcount > 1)
6550 {
6551 *errorcodeptr = ERR54;
6552 goto FAILED;
6553 }
6554 bravalue = OP_DEF; /* Just a flag to suppress char handling below */
6555 }
6556
6557 /* A "normal" conditional group. If there is just one branch, we must not
6558 make use of its firstchar or reqchar, because this is equivalent to an
6559 empty second branch. */
6560
6561 else
6562 {
6563 if (condcount > 2)
6564 {
6565 *errorcodeptr = ERR27;
6566 goto FAILED;
6567 }
6568 if (condcount == 1) subfirstchar = subreqchar = REQ_NONE;
6569 }
6570 }
6571
6572 /* Error if hit end of pattern */
6573
6574 if (*ptr != CHAR_RIGHT_PARENTHESIS)
6575 {
6576 *errorcodeptr = ERR14;
6577 goto FAILED;
6578 }
6579
6580 /* In the pre-compile phase, update the length by the length of the group,
6581 less the brackets at either end. Then reduce the compiled code to just a
6582 set of non-capturing brackets so that it doesn't use much memory if it is
6583 duplicated by a quantifier.*/
6584
6585 if (lengthptr != NULL)
6586 {
6587 if (OFLOW_MAX - *lengthptr < length_prevgroup - 2 - 2*LINK_SIZE)
6588 {
6589 *errorcodeptr = ERR20;
6590 goto FAILED;
6591 }
6592 *lengthptr += length_prevgroup - 2 - 2*LINK_SIZE;
6593 code++; /* This already contains bravalue */
6594 PUTINC(code, 0, 1 + LINK_SIZE);
6595 *code++ = OP_KET;
6596 PUTINC(code, 0, 1 + LINK_SIZE);
6597 break; /* No need to waste time with special character handling */
6598 }
6599
6600 /* Otherwise update the main code pointer to the end of the group. */
6601
6602 code = tempcode;
6603
6604 /* For a DEFINE group, required and first character settings are not
6605 relevant. */
6606
6607 if (bravalue == OP_DEF) break;
6608
6609 /* Handle updating of the required and first characters for other types of
6610 group. Update for normal brackets of all kinds, and conditions with two
6611 branches (see code above). If the bracket is followed by a quantifier with
6612 zero repeat, we have to back off. Hence the definition of zeroreqchar and
6613 zerofirstchar outside the main loop so that they can be accessed for the
6614 back off. */
6615
6616 zeroreqchar = reqchar;
6617 zerofirstchar = firstchar;
6618 groupsetfirstchar = FALSE;
6619
6620 if (bravalue >= OP_ONCE)
6621 {
6622 /* If we have not yet set a firstchar in this branch, take it from the
6623 subpattern, remembering that it was set here so that a repeat of more
6624 than one can replicate it as reqchar if necessary. If the subpattern has
6625 no firstchar, set "none" for the whole branch. In both cases, a zero
6626 repeat forces firstchar to "none". */
6627
6628 if (firstchar == REQ_UNSET)
6629 {
6630 if (subfirstchar >= 0)
6631 {
6632 firstchar = subfirstchar;
6633 groupsetfirstchar =