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

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

Parent Directory Parent Directory | Revision Log Revision Log


Revision 774 - (show annotations)
Thu Dec 1 06:08:45 2011 UTC (7 years, 9 months ago) by zherczeg
File MIME type: text/plain
File size: 257449 byte(s)
better digit parsing, first_byte, req_byte are renamed to first_char req_char respectively
1 /*************************************************
2 * Perl-Compatible Regular Expressions *
3 *************************************************/
4
5 /* PCRE is a library of functions to support regular expressions whose syntax
6 and semantics are as close as possible to those of the Perl 5 language.
7
8 Written by Philip Hazel
9 Copyright (c) 1997-2011 University of Cambridge
10
11 -----------------------------------------------------------------------------
12 Redistribution and use in source and binary forms, with or without
13 modification, are permitted provided that the following conditions are met:
14
15 * Redistributions of source code must retain the above copyright notice,
16 this list of conditions and the following disclaimer.
17
18 * Redistributions in binary form must reproduce the above copyright
19 notice, this list of conditions and the following disclaimer in the
20 documentation and/or other materials provided with the distribution.
21
22 * Neither the name of the University of Cambridge nor the names of its
23 contributors may be used to endorse or promote products derived from
24 this software without specific prior written permission.
25
26 THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
27 AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28 IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29 ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE
30 LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
31 CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
32 SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
33 INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
34 CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
35 ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
36 POSSIBILITY OF SUCH DAMAGE.
37 -----------------------------------------------------------------------------
38 */
39
40
41 /* This module contains the external function pcre_compile(), along with
42 supporting internal functions that are not used by other modules. */
43
44
45 #ifdef HAVE_CONFIG_H
46 #include "config.h"
47 #endif
48
49 #define NLBLOCK cd /* Block containing newline information */
50 #define PSSTART start_pattern /* Field containing processed string start */
51 #define PSEND end_pattern /* Field containing processed string end */
52
53 #include "pcre_internal.h"
54
55
56 /* When PCRE_DEBUG is defined, we need the pcre_printint() function, which is
57 also used by pcretest. PCRE_DEBUG is not defined when building a production
58 library. */
59
60 #ifdef PCRE_DEBUG
61 #include "pcre_printint.src"
62 #endif
63
64
65 /* Macro for setting individual bits in class bitmaps. */
66
67 #define SETBIT(a,b) a[b/8] |= (1 << (b%8))
68
69 /* Maximum length value to check against when making sure that the integer that
70 holds the compiled pattern length does not overflow. We make it a bit less than
71 INT_MAX to allow for adding in group terminating bytes, so that we don't have
72 to check them every time. */
73
74 #define OFLOW_MAX (INT_MAX - 20)
75
76
77 /*************************************************
78 * Code parameters and static tables *
79 *************************************************/
80
81 /* This value specifies the size of stack workspace that is used during the
82 first pre-compile phase that determines how much memory is required. The regex
83 is partly compiled into this space, but the compiled parts are discarded as
84 soon as they can be, so that hopefully there will never be an overrun. The code
85 does, however, check for an overrun. The largest amount I've seen used is 218,
86 so this number is very generous.
87
88 The same workspace is used during the second, actual compile phase for
89 remembering forward references to groups so that they can be filled in at the
90 end. Each entry in this list occupies LINK_SIZE bytes, so even when LINK_SIZE
91 is 4 there is plenty of room. */
92
93 #define COMPILE_WORK_SIZE (4096)
94
95 /* The overrun tests check for a slightly smaller size so that they detect the
96 overrun before it actually does run off the end of the data block. */
97
98 #define WORK_SIZE_CHECK (COMPILE_WORK_SIZE - 100)
99
100 /* Private flags added to firstchar and reqchar. */
101
102 #define REQ_CASELESS 0x10000000l /* Indicates caselessness */
103 #define REQ_VARY 0x20000000l /* Reqchar followed non-literal item */
104
105 /* Table for handling escaped characters in the range '0'-'z'. Positive returns
106 are simple data values; negative values are for special things like \d and so
107 on. Zero means further processing is needed (for things like \x), or the escape
108 is invalid. */
109
110 #ifndef EBCDIC
111
112 /* This is the "normal" table for ASCII systems or for EBCDIC systems running
113 in UTF-8 mode. */
114
115 static const short int escapes[] = {
116 0, 0,
117 0, 0,
118 0, 0,
119 0, 0,
120 0, 0,
121 CHAR_COLON, CHAR_SEMICOLON,
122 CHAR_LESS_THAN_SIGN, CHAR_EQUALS_SIGN,
123 CHAR_GREATER_THAN_SIGN, CHAR_QUESTION_MARK,
124 CHAR_COMMERCIAL_AT, -ESC_A,
125 -ESC_B, -ESC_C,
126 -ESC_D, -ESC_E,
127 0, -ESC_G,
128 -ESC_H, 0,
129 0, -ESC_K,
130 0, 0,
131 -ESC_N, 0,
132 -ESC_P, -ESC_Q,
133 -ESC_R, -ESC_S,
134 0, 0,
135 -ESC_V, -ESC_W,
136 -ESC_X, 0,
137 -ESC_Z, CHAR_LEFT_SQUARE_BRACKET,
138 CHAR_BACKSLASH, CHAR_RIGHT_SQUARE_BRACKET,
139 CHAR_CIRCUMFLEX_ACCENT, CHAR_UNDERSCORE,
140 CHAR_GRAVE_ACCENT, 7,
141 -ESC_b, 0,
142 -ESC_d, ESC_e,
143 ESC_f, 0,
144 -ESC_h, 0,
145 0, -ESC_k,
146 0, 0,
147 ESC_n, 0,
148 -ESC_p, 0,
149 ESC_r, -ESC_s,
150 ESC_tee, 0,
151 -ESC_v, -ESC_w,
152 0, 0,
153 -ESC_z
154 };
155
156 #else
157
158 /* This is the "abnormal" table for EBCDIC systems without UTF-8 support. */
159
160 static const short int escapes[] = {
161 /* 48 */ 0, 0, 0, '.', '<', '(', '+', '|',
162 /* 50 */ '&', 0, 0, 0, 0, 0, 0, 0,
163 /* 58 */ 0, 0, '!', '$', '*', ')', ';', '~',
164 /* 60 */ '-', '/', 0, 0, 0, 0, 0, 0,
165 /* 68 */ 0, 0, '|', ',', '%', '_', '>', '?',
166 /* 70 */ 0, 0, 0, 0, 0, 0, 0, 0,
167 /* 78 */ 0, '`', ':', '#', '@', '\'', '=', '"',
168 /* 80 */ 0, 7, -ESC_b, 0, -ESC_d, ESC_e, ESC_f, 0,
169 /* 88 */-ESC_h, 0, 0, '{', 0, 0, 0, 0,
170 /* 90 */ 0, 0, -ESC_k, 'l', 0, ESC_n, 0, -ESC_p,
171 /* 98 */ 0, ESC_r, 0, '}', 0, 0, 0, 0,
172 /* A0 */ 0, '~', -ESC_s, ESC_tee, 0,-ESC_v, -ESC_w, 0,
173 /* A8 */ 0,-ESC_z, 0, 0, 0, '[', 0, 0,
174 /* B0 */ 0, 0, 0, 0, 0, 0, 0, 0,
175 /* B8 */ 0, 0, 0, 0, 0, ']', '=', '-',
176 /* C0 */ '{',-ESC_A, -ESC_B, -ESC_C, -ESC_D,-ESC_E, 0, -ESC_G,
177 /* C8 */-ESC_H, 0, 0, 0, 0, 0, 0, 0,
178 /* D0 */ '}', 0, -ESC_K, 0, 0,-ESC_N, 0, -ESC_P,
179 /* D8 */-ESC_Q,-ESC_R, 0, 0, 0, 0, 0, 0,
180 /* E0 */ '\\', 0, -ESC_S, 0, 0,-ESC_V, -ESC_W, -ESC_X,
181 /* E8 */ 0,-ESC_Z, 0, 0, 0, 0, 0, 0,
182 /* F0 */ 0, 0, 0, 0, 0, 0, 0, 0,
183 /* F8 */ 0, 0, 0, 0, 0, 0, 0, 0
184 };
185 #endif
186
187
188 /* Table of special "verbs" like (*PRUNE). This is a short table, so it is
189 searched linearly. Put all the names into a single string, in order to reduce
190 the number of relocations when a shared library is dynamically linked. The
191 string is built from string macros so that it works in UTF-8 mode on EBCDIC
192 platforms. */
193
194 typedef struct verbitem {
195 int len; /* Length of verb name */
196 int op; /* Op when no arg, or -1 if arg mandatory */
197 int op_arg; /* Op when arg present, or -1 if not allowed */
198 } verbitem;
199
200 static const char verbnames[] =
201 "\0" /* Empty name is a shorthand for MARK */
202 STRING_MARK0
203 STRING_ACCEPT0
204 STRING_COMMIT0
205 STRING_F0
206 STRING_FAIL0
207 STRING_PRUNE0
208 STRING_SKIP0
209 STRING_THEN;
210
211 static const verbitem verbs[] = {
212 { 0, -1, OP_MARK },
213 { 4, -1, OP_MARK },
214 { 6, OP_ACCEPT, -1 },
215 { 6, OP_COMMIT, -1 },
216 { 1, OP_FAIL, -1 },
217 { 4, OP_FAIL, -1 },
218 { 5, OP_PRUNE, OP_PRUNE_ARG },
219 { 4, OP_SKIP, OP_SKIP_ARG },
220 { 4, OP_THEN, OP_THEN_ARG }
221 };
222
223 static const int verbcount = sizeof(verbs)/sizeof(verbitem);
224
225
226 /* Tables of names of POSIX character classes and their lengths. The names are
227 now all in a single string, to reduce the number of relocations when a shared
228 library is dynamically loaded. The list of lengths is terminated by a zero
229 length entry. The first three must be alpha, lower, upper, as this is assumed
230 for handling case independence. */
231
232 static const char posix_names[] =
233 STRING_alpha0 STRING_lower0 STRING_upper0 STRING_alnum0
234 STRING_ascii0 STRING_blank0 STRING_cntrl0 STRING_digit0
235 STRING_graph0 STRING_print0 STRING_punct0 STRING_space0
236 STRING_word0 STRING_xdigit;
237
238 static const pcre_uint8 posix_name_lengths[] = {
239 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 4, 6, 0 };
240
241 /* Table of class bit maps for each POSIX class. Each class is formed from a
242 base map, with an optional addition or removal of another map. Then, for some
243 classes, there is some additional tweaking: for [:blank:] the vertical space
244 characters are removed, and for [:alpha:] and [:alnum:] the underscore
245 character is removed. The triples in the table consist of the base map offset,
246 second map offset or -1 if no second map, and a non-negative value for map
247 addition or a negative value for map subtraction (if there are two maps). The
248 absolute value of the third field has these meanings: 0 => no tweaking, 1 =>
249 remove vertical space characters, 2 => remove underscore. */
250
251 static const int posix_class_maps[] = {
252 cbit_word, cbit_digit, -2, /* alpha */
253 cbit_lower, -1, 0, /* lower */
254 cbit_upper, -1, 0, /* upper */
255 cbit_word, -1, 2, /* alnum - word without underscore */
256 cbit_print, cbit_cntrl, 0, /* ascii */
257 cbit_space, -1, 1, /* blank - a GNU extension */
258 cbit_cntrl, -1, 0, /* cntrl */
259 cbit_digit, -1, 0, /* digit */
260 cbit_graph, -1, 0, /* graph */
261 cbit_print, -1, 0, /* print */
262 cbit_punct, -1, 0, /* punct */
263 cbit_space, -1, 0, /* space */
264 cbit_word, -1, 0, /* word - a Perl extension */
265 cbit_xdigit,-1, 0 /* xdigit */
266 };
267
268 /* Table of substitutes for \d etc when PCRE_UCP is set. The POSIX class
269 substitutes must be in the order of the names, defined above, and there are
270 both positive and negative cases. NULL means no substitute. */
271
272 #ifdef SUPPORT_UCP
273 static const pcre_uchar string_PNd[] = {
274 CHAR_BACKSLASH, CHAR_P, CHAR_LEFT_CURLY_BRACKET,
275 CHAR_N, CHAR_d, CHAR_RIGHT_CURLY_BRACKET, '\0' };
276 static const pcre_uchar string_pNd[] = {
277 CHAR_BACKSLASH, CHAR_p, CHAR_LEFT_CURLY_BRACKET,
278 CHAR_N, CHAR_d, CHAR_RIGHT_CURLY_BRACKET, '\0' };
279 static const pcre_uchar string_PXsp[] = {
280 CHAR_BACKSLASH, CHAR_P, CHAR_LEFT_CURLY_BRACKET,
281 CHAR_X, CHAR_s, CHAR_p, CHAR_RIGHT_CURLY_BRACKET, '\0' };
282 static const pcre_uchar string_pXsp[] = {
283 CHAR_BACKSLASH, CHAR_p, CHAR_LEFT_CURLY_BRACKET,
284 CHAR_X, CHAR_s, CHAR_p, CHAR_RIGHT_CURLY_BRACKET, '\0' };
285 static const pcre_uchar string_PXwd[] = {
286 CHAR_BACKSLASH, CHAR_P, CHAR_LEFT_CURLY_BRACKET,
287 CHAR_X, CHAR_w, CHAR_d, CHAR_RIGHT_CURLY_BRACKET, '\0' };
288 static const pcre_uchar string_pXwd[] = {
289 CHAR_BACKSLASH, CHAR_p, CHAR_LEFT_CURLY_BRACKET,
290 CHAR_X, CHAR_w, CHAR_d, CHAR_RIGHT_CURLY_BRACKET, '\0' };
291
292 static const pcre_uchar *substitutes[] = {
293 string_PNd, /* \D */
294 string_pNd, /* \d */
295 string_PXsp, /* \S */ /* NOTE: Xsp is Perl space */
296 string_pXsp, /* \s */
297 string_PXwd, /* \W */
298 string_pXwd /* \w */
299 };
300
301 static const pcre_uchar string_pL[] = {
302 CHAR_BACKSLASH, CHAR_p, CHAR_LEFT_CURLY_BRACKET,
303 CHAR_L, CHAR_RIGHT_CURLY_BRACKET, '\0' };
304 static const pcre_uchar string_pLl[] = {
305 CHAR_BACKSLASH, CHAR_p, CHAR_LEFT_CURLY_BRACKET,
306 CHAR_L, CHAR_l, CHAR_RIGHT_CURLY_BRACKET, '\0' };
307 static const pcre_uchar string_pLu[] = {
308 CHAR_BACKSLASH, CHAR_p, CHAR_LEFT_CURLY_BRACKET,
309 CHAR_L, CHAR_u, CHAR_RIGHT_CURLY_BRACKET, '\0' };
310 static const pcre_uchar string_pXan[] = {
311 CHAR_BACKSLASH, CHAR_p, CHAR_LEFT_CURLY_BRACKET,
312 CHAR_X, CHAR_a, CHAR_n, CHAR_RIGHT_CURLY_BRACKET, '\0' };
313 static const pcre_uchar string_h[] = {
314 CHAR_BACKSLASH, CHAR_h, '\0' };
315 static const pcre_uchar string_pXps[] = {
316 CHAR_BACKSLASH, CHAR_p, CHAR_LEFT_CURLY_BRACKET,
317 CHAR_X, CHAR_p, CHAR_s, CHAR_RIGHT_CURLY_BRACKET, '\0' };
318 static const pcre_uchar string_PL[] = {
319 CHAR_BACKSLASH, CHAR_P, CHAR_LEFT_CURLY_BRACKET,
320 CHAR_L, CHAR_RIGHT_CURLY_BRACKET, '\0' };
321 static const pcre_uchar string_PLl[] = {
322 CHAR_BACKSLASH, CHAR_P, CHAR_LEFT_CURLY_BRACKET,
323 CHAR_L, CHAR_l, CHAR_RIGHT_CURLY_BRACKET, '\0' };
324 static const pcre_uchar string_PLu[] = {
325 CHAR_BACKSLASH, CHAR_P, CHAR_LEFT_CURLY_BRACKET,
326 CHAR_L, CHAR_u, CHAR_RIGHT_CURLY_BRACKET, '\0' };
327 static const pcre_uchar string_PXan[] = {
328 CHAR_BACKSLASH, CHAR_P, CHAR_LEFT_CURLY_BRACKET,
329 CHAR_X, CHAR_a, CHAR_n, CHAR_RIGHT_CURLY_BRACKET, '\0' };
330 static const pcre_uchar string_H[] = {
331 CHAR_BACKSLASH, CHAR_H, '\0' };
332 static const pcre_uchar string_PXps[] = {
333 CHAR_BACKSLASH, CHAR_P, CHAR_LEFT_CURLY_BRACKET,
334 CHAR_X, CHAR_p, CHAR_s, CHAR_RIGHT_CURLY_BRACKET, '\0' };
335
336 static const pcre_uchar *posix_substitutes[] = {
337 string_pL, /* alpha */
338 string_pLl, /* lower */
339 string_pLu, /* upper */
340 string_pXan, /* alnum */
341 NULL, /* ascii */
342 string_h, /* blank */
343 NULL, /* cntrl */
344 string_pNd, /* digit */
345 NULL, /* graph */
346 NULL, /* print */
347 NULL, /* punct */
348 string_pXps, /* space */ /* NOTE: Xps is POSIX space */
349 string_pXwd, /* word */
350 NULL, /* xdigit */
351 /* Negated cases */
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 };
367 #define POSIX_SUBSIZE (sizeof(posix_substitutes) / sizeof(pcre_uchar *))
368 #endif
369
370 #define STRING(a) # a
371 #define XSTRING(s) STRING(s)
372
373 /* The texts of compile-time error messages. These are "char *" because they
374 are passed to the outside world. Do not ever re-use any error number, because
375 they are documented. Always add a new error instead. Messages marked DEAD below
376 are no longer used. This used to be a table of strings, but in order to reduce
377 the number of relocations needed when a shared library is loaded dynamically,
378 it is now one long string. We cannot use a table of offsets, because the
379 lengths of inserts such as XSTRING(MAX_NAME_SIZE) are not known. Instead, we
380 simply count through to the one we want - this isn't a performance issue
381 because these strings are used only when there is a compilation error.
382
383 Each substring ends with \0 to insert a null character. This includes the final
384 substring, so that the whole string ends with \0\0, which can be detected when
385 counting through. */
386
387 static const char error_texts[] =
388 "no error\0"
389 "\\ at end of pattern\0"
390 "\\c at end of pattern\0"
391 "unrecognized character follows \\\0"
392 "numbers out of order in {} quantifier\0"
393 /* 5 */
394 "number too big in {} quantifier\0"
395 "missing terminating ] for character class\0"
396 "invalid escape sequence in character class\0"
397 "range out of order in character class\0"
398 "nothing to repeat\0"
399 /* 10 */
400 "operand of unlimited repeat could match the empty string\0" /** DEAD **/
401 "internal error: unexpected repeat\0"
402 "unrecognized character after (? or (?-\0"
403 "POSIX named classes are supported only within a class\0"
404 "missing )\0"
405 /* 15 */
406 "reference to non-existent subpattern\0"
407 "erroffset passed as NULL\0"
408 "unknown option bit(s) set\0"
409 "missing ) after comment\0"
410 "parentheses nested too deeply\0" /** DEAD **/
411 /* 20 */
412 "regular expression is too large\0"
413 "failed to get memory\0"
414 "unmatched parentheses\0"
415 "internal error: code overflow\0"
416 "unrecognized character after (?<\0"
417 /* 25 */
418 "lookbehind assertion is not fixed length\0"
419 "malformed number or name after (?(\0"
420 "conditional group contains more than two branches\0"
421 "assertion expected after (?(\0"
422 "(?R or (?[+-]digits must be followed by )\0"
423 /* 30 */
424 "unknown POSIX class name\0"
425 "POSIX collating elements are not supported\0"
426 "this version of PCRE is not compiled with PCRE_UTF8 support\0"
427 "spare error\0" /** DEAD **/
428 "character value in \\x{...} sequence is too large\0"
429 /* 35 */
430 "invalid condition (?(0)\0"
431 "\\C not allowed in lookbehind assertion\0"
432 "PCRE does not support \\L, \\l, \\N{name}, \\U, or \\u\0"
433 "number after (?C is > 255\0"
434 "closing ) for (?C expected\0"
435 /* 40 */
436 "recursive call could loop indefinitely\0"
437 "unrecognized character after (?P\0"
438 "syntax error in subpattern name (missing terminator)\0"
439 "two named subpatterns have the same name\0"
440 "invalid UTF-8 string\0"
441 /* 45 */
442 "support for \\P, \\p, and \\X has not been compiled\0"
443 "malformed \\P or \\p sequence\0"
444 "unknown property name after \\P or \\p\0"
445 "subpattern name is too long (maximum " XSTRING(MAX_NAME_SIZE) " characters)\0"
446 "too many named subpatterns (maximum " XSTRING(MAX_NAME_COUNT) ")\0"
447 /* 50 */
448 "repeated subpattern is too long\0" /** DEAD **/
449 "octal value is greater than \\377 (not in UTF-8 mode)\0"
450 "internal error: overran compiling workspace\0"
451 "internal error: previously-checked referenced subpattern not found\0"
452 "DEFINE group contains more than one branch\0"
453 /* 55 */
454 "repeating a DEFINE group is not allowed\0" /** DEAD **/
455 "inconsistent NEWLINE options\0"
456 "\\g is not followed by a braced, angle-bracketed, or quoted name/number or by a plain number\0"
457 "a numbered reference must not be zero\0"
458 "an argument is not allowed for (*ACCEPT), (*FAIL), or (*COMMIT)\0"
459 /* 60 */
460 "(*VERB) not recognized\0"
461 "number is too big\0"
462 "subpattern name expected\0"
463 "digit expected after (?+\0"
464 "] is an invalid data character in JavaScript compatibility mode\0"
465 /* 65 */
466 "different names for subpatterns of the same number are not allowed\0"
467 "(*MARK) must have an argument\0"
468 "this version of PCRE is not compiled with PCRE_UCP support\0"
469 "\\c must be followed by an ASCII character\0"
470 "\\k is not followed by a braced, angle-bracketed, or quoted name\0"
471 /* 70 */
472 "internal error: unknown opcode in find_fixedlength()\0"
473 ;
474
475 /* Table to identify digits and hex digits. This is used when compiling
476 patterns. Note that the tables in chartables are dependent on the locale, and
477 may mark arbitrary characters as digits - but the PCRE compiling code expects
478 to handle only 0-9, a-z, and A-Z as digits when compiling. That is why we have
479 a private table here. It costs 256 bytes, but it is a lot faster than doing
480 character value tests (at least in some simple cases I timed), and in some
481 applications one wants PCRE to compile efficiently as well as match
482 efficiently.
483
484 For convenience, we use the same bit definitions as in chartables:
485
486 0x04 decimal digit
487 0x08 hexadecimal digit
488
489 Then we can use ctype_digit and ctype_xdigit in the code. */
490
491 /* Using a simple comparison for decimal numbers rather than a memory read
492 is much faster, and the resulting code is simpler (the compiler turns it
493 into a subtraction and unsigned comparison). */
494
495 #define IS_DIGIT(x) ((x) >= CHAR_0 && (x) <= CHAR_9)
496
497 #ifndef EBCDIC
498
499 /* This is the "normal" case, for ASCII systems, and EBCDIC systems running in
500 UTF-8 mode. */
501
502 static const pcre_uint8 digitab[] =
503 {
504 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 0- 7 */
505 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 8- 15 */
506 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 16- 23 */
507 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 24- 31 */
508 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - ' */
509 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* ( - / */
510 0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c, /* 0 - 7 */
511 0x0c,0x0c,0x00,0x00,0x00,0x00,0x00,0x00, /* 8 - ? */
512 0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* @ - G */
513 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* H - O */
514 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* P - W */
515 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* X - _ */
516 0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* ` - g */
517 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* h - o */
518 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* p - w */
519 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* x -127 */
520 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 128-135 */
521 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 136-143 */
522 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 144-151 */
523 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 152-159 */
524 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 160-167 */
525 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 168-175 */
526 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 176-183 */
527 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 184-191 */
528 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 192-199 */
529 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 200-207 */
530 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 208-215 */
531 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 216-223 */
532 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 224-231 */
533 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 232-239 */
534 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 240-247 */
535 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00};/* 248-255 */
536
537 #else
538
539 /* This is the "abnormal" case, for EBCDIC systems not running in UTF-8 mode. */
540
541 static const pcre_unit8 digitab[] =
542 {
543 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 0- 7 0 */
544 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 8- 15 */
545 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 16- 23 10 */
546 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 24- 31 */
547 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 32- 39 20 */
548 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 40- 47 */
549 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 48- 55 30 */
550 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 56- 63 */
551 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - 71 40 */
552 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 72- | */
553 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* & - 87 50 */
554 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 88- 95 */
555 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - -103 60 */
556 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 104- ? */
557 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 112-119 70 */
558 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 120- " */
559 0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* 128- g 80 */
560 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* h -143 */
561 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 144- p 90 */
562 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* q -159 */
563 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 160- x A0 */
564 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* y -175 */
565 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* ^ -183 B0 */
566 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 184-191 */
567 0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* { - G C0 */
568 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* H -207 */
569 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* } - P D0 */
570 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* Q -223 */
571 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* \ - X E0 */
572 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* Y -239 */
573 0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c, /* 0 - 7 F0 */
574 0x0c,0x0c,0x00,0x00,0x00,0x00,0x00,0x00};/* 8 -255 */
575
576 static const pcre_uint8 ebcdic_chartab[] = { /* chartable partial dup */
577 0x80,0x00,0x00,0x00,0x00,0x01,0x00,0x00, /* 0- 7 */
578 0x00,0x00,0x00,0x00,0x01,0x01,0x00,0x00, /* 8- 15 */
579 0x00,0x00,0x00,0x00,0x00,0x01,0x00,0x00, /* 16- 23 */
580 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 24- 31 */
581 0x00,0x00,0x00,0x00,0x00,0x01,0x00,0x00, /* 32- 39 */
582 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 40- 47 */
583 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 48- 55 */
584 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 56- 63 */
585 0x01,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - 71 */
586 0x00,0x00,0x00,0x80,0x00,0x80,0x80,0x80, /* 72- | */
587 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* & - 87 */
588 0x00,0x00,0x00,0x80,0x80,0x80,0x00,0x00, /* 88- 95 */
589 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - -103 */
590 0x00,0x00,0x00,0x00,0x00,0x10,0x00,0x80, /* 104- ? */
591 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 112-119 */
592 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 120- " */
593 0x00,0x1a,0x1a,0x1a,0x1a,0x1a,0x1a,0x12, /* 128- g */
594 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* h -143 */
595 0x00,0x12,0x12,0x12,0x12,0x12,0x12,0x12, /* 144- p */
596 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* q -159 */
597 0x00,0x00,0x12,0x12,0x12,0x12,0x12,0x12, /* 160- x */
598 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* y -175 */
599 0x80,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* ^ -183 */
600 0x00,0x00,0x80,0x00,0x00,0x00,0x00,0x00, /* 184-191 */
601 0x80,0x1a,0x1a,0x1a,0x1a,0x1a,0x1a,0x12, /* { - G */
602 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* H -207 */
603 0x00,0x12,0x12,0x12,0x12,0x12,0x12,0x12, /* } - P */
604 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* Q -223 */
605 0x00,0x00,0x12,0x12,0x12,0x12,0x12,0x12, /* \ - X */
606 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* Y -239 */
607 0x1c,0x1c,0x1c,0x1c,0x1c,0x1c,0x1c,0x1c, /* 0 - 7 */
608 0x1c,0x1c,0x00,0x00,0x00,0x00,0x00,0x00};/* 8 -255 */
609 #endif
610
611
612 /* Definition to allow mutual recursion */
613
614 static BOOL
615 compile_regex(int, pcre_uchar **, const pcre_uchar **, int *, BOOL, BOOL, int, int,
616 int *, int *, branch_chain *, compile_data *, int *);
617
618
619
620 /*************************************************
621 * Find an error text *
622 *************************************************/
623
624 /* The error texts are now all in one long string, to save on relocations. As
625 some of the text is of unknown length, we can't use a table of offsets.
626 Instead, just count through the strings. This is not a performance issue
627 because it happens only when there has been a compilation error.
628
629 Argument: the error number
630 Returns: pointer to the error string
631 */
632
633 static const char *
634 find_error_text(int n)
635 {
636 const char *s = error_texts;
637 for (; n > 0; n--)
638 {
639 while (*s++ != 0) {};
640 if (*s == 0) return "Error text not found (please report)";
641 }
642 return s;
643 }
644
645
646 /*************************************************
647 * Check for counted repeat *
648 *************************************************/
649
650 /* This function is called when a '{' is encountered in a place where it might
651 start a quantifier. It looks ahead to see if it really is a quantifier or not.
652 It is only a quantifier if it is one of the forms {ddd} {ddd,} or {ddd,ddd}
653 where the ddds are digits.
654
655 Arguments:
656 p pointer to the first char after '{'
657
658 Returns: TRUE or FALSE
659 */
660
661 static BOOL
662 is_counted_repeat(const pcre_uchar *p)
663 {
664 if (!IS_DIGIT(*p)) return FALSE;
665 p++;
666 while (IS_DIGIT(*p)) p++;
667 if (*p == CHAR_RIGHT_CURLY_BRACKET) return TRUE;
668
669 if (*p++ != CHAR_COMMA) return FALSE;
670 if (*p == CHAR_RIGHT_CURLY_BRACKET) return TRUE;
671
672 if (!IS_DIGIT(*p)) return FALSE;
673 p++;
674 while (IS_DIGIT(*p)) p++;
675
676 return (*p == CHAR_RIGHT_CURLY_BRACKET);
677 }
678
679
680
681 /*************************************************
682 * Handle escapes *
683 *************************************************/
684
685 /* This function is called when a \ has been encountered. It either returns a
686 positive value for a simple escape such as \n, or a negative value which
687 encodes one of the more complicated things such as \d. A backreference to group
688 n is returned as -(ESC_REF + n); ESC_REF is the highest ESC_xxx macro. When
689 UTF-8 is enabled, a positive value greater than 255 may be returned. On entry,
690 ptr is pointing at the \. On exit, it is on the final character of the escape
691 sequence.
692
693 Arguments:
694 ptrptr points to the pattern position pointer
695 errorcodeptr points to the errorcode variable
696 bracount number of previous extracting brackets
697 options the options bits
698 isclass TRUE if inside a character class
699
700 Returns: zero or positive => a data character
701 negative => a special escape sequence
702 on error, errorcodeptr is set
703 */
704
705 static int
706 check_escape(const pcre_uchar **ptrptr, int *errorcodeptr, int bracount,
707 int options, BOOL isclass)
708 {
709 BOOL utf8 = (options & PCRE_UTF8) != 0;
710 const pcre_uchar *ptr = *ptrptr + 1;
711 int c, i;
712
713 GETCHARINCTEST(c, ptr); /* Get character value, increment pointer */
714 ptr--; /* Set pointer back to the last byte */
715
716 /* If backslash is at the end of the pattern, it's an error. */
717
718 if (c == 0) *errorcodeptr = ERR1;
719
720 /* Non-alphanumerics are literals. For digits or letters, do an initial lookup
721 in a table. A non-zero result is something that can be returned immediately.
722 Otherwise further processing may be required. */
723
724 #ifndef EBCDIC /* ASCII/UTF-8 coding */
725 /* Not alphanumeric */
726 else if (c < CHAR_0 || c > CHAR_z) {}
727 else if ((i = escapes[c - CHAR_0]) != 0) c = i;
728
729 #else /* EBCDIC coding */
730 /* Not alphanumeric */
731 else if (c < 'a' || (!MAX_255(c) || (ebcdic_chartab[c] & 0x0E) == 0)) {}
732 else if ((i = escapes[c - 0x48]) != 0) c = i;
733 #endif
734
735 /* Escapes that need further processing, or are illegal. */
736
737 else
738 {
739 const pcre_uchar *oldptr;
740 BOOL braced, negated;
741
742 switch (c)
743 {
744 /* A number of Perl escapes are not handled by PCRE. We give an explicit
745 error. */
746
747 case CHAR_l:
748 case CHAR_L:
749 *errorcodeptr = ERR37;
750 break;
751
752 case CHAR_u:
753 if ((options & PCRE_JAVASCRIPT_COMPAT) != 0)
754 {
755 /* In JavaScript, \u must be followed by four hexadecimal numbers.
756 Otherwise it is a lowercase u letter. */
757 if (MAX_255(ptr[1]) && (digitab[ptr[1]] & ctype_xdigit) != 0
758 && MAX_255(ptr[2]) && (digitab[ptr[2]] & ctype_xdigit) != 0
759 && MAX_255(ptr[3]) && (digitab[ptr[3]] & ctype_xdigit) != 0
760 && MAX_255(ptr[4]) && (digitab[ptr[4]] & ctype_xdigit) != 0)
761 {
762 c = 0;
763 for (i = 0; i < 4; ++i)
764 {
765 register int cc = *(++ptr);
766 #ifndef EBCDIC /* ASCII/UTF-8 coding */
767 if (cc >= CHAR_a) cc -= 32; /* Convert to upper case */
768 c = (c << 4) + cc - ((cc < CHAR_A)? CHAR_0 : (CHAR_A - 10));
769 #else /* EBCDIC coding */
770 if (cc >= CHAR_a && cc <= CHAR_z) cc += 64; /* Convert to upper case */
771 c = (c << 4) + cc - ((cc >= CHAR_0)? CHAR_0 : (CHAR_A - 10));
772 #endif
773 }
774 }
775 }
776 else
777 *errorcodeptr = ERR37;
778 break;
779
780 case CHAR_U:
781 /* In JavaScript, \U is an uppercase U letter. */
782 if ((options & PCRE_JAVASCRIPT_COMPAT) == 0) *errorcodeptr = ERR37;
783 break;
784
785 /* In a character class, \g is just a literal "g". Outside a character
786 class, \g must be followed by one of a number of specific things:
787
788 (1) A number, either plain or braced. If positive, it is an absolute
789 backreference. If negative, it is a relative backreference. This is a Perl
790 5.10 feature.
791
792 (2) Perl 5.10 also supports \g{name} as a reference to a named group. This
793 is part of Perl's movement towards a unified syntax for back references. As
794 this is synonymous with \k{name}, we fudge it up by pretending it really
795 was \k.
796
797 (3) For Oniguruma compatibility we also support \g followed by a name or a
798 number either in angle brackets or in single quotes. However, these are
799 (possibly recursive) subroutine calls, _not_ backreferences. Just return
800 the -ESC_g code (cf \k). */
801
802 case CHAR_g:
803 if (isclass) break;
804 if (ptr[1] == CHAR_LESS_THAN_SIGN || ptr[1] == CHAR_APOSTROPHE)
805 {
806 c = -ESC_g;
807 break;
808 }
809
810 /* Handle the Perl-compatible cases */
811
812 if (ptr[1] == CHAR_LEFT_CURLY_BRACKET)
813 {
814 const pcre_uchar *p;
815 for (p = ptr+2; *p != 0 && *p != CHAR_RIGHT_CURLY_BRACKET; p++)
816 if (*p != CHAR_MINUS && !IS_DIGIT(*p)) break;
817 if (*p != 0 && *p != CHAR_RIGHT_CURLY_BRACKET)
818 {
819 c = -ESC_k;
820 break;
821 }
822 braced = TRUE;
823 ptr++;
824 }
825 else braced = FALSE;
826
827 if (ptr[1] == CHAR_MINUS)
828 {
829 negated = TRUE;
830 ptr++;
831 }
832 else negated = FALSE;
833
834 /* The integer range is limited by the machine's int representation. */
835 c = 0;
836 while (IS_DIGIT(ptr[1]))
837 {
838 if (((unsigned int)c) > INT_MAX / 10) /* Integer overflow */
839 {
840 c = -1;
841 break;
842 }
843 c = c * 10 + *(++ptr) - CHAR_0;
844 }
845 if (((unsigned int)c) > INT_MAX) /* Integer overflow */
846 {
847 while (IS_DIGIT(ptr[1]))
848 ptr++;
849 *errorcodeptr = ERR61;
850 break;
851 }
852
853 if (braced && *(++ptr) != CHAR_RIGHT_CURLY_BRACKET)
854 {
855 *errorcodeptr = ERR57;
856 break;
857 }
858
859 if (c == 0)
860 {
861 *errorcodeptr = ERR58;
862 break;
863 }
864
865 if (negated)
866 {
867 if (c > bracount)
868 {
869 *errorcodeptr = ERR15;
870 break;
871 }
872 c = bracount - (c - 1);
873 }
874
875 c = -(ESC_REF + c);
876 break;
877
878 /* The handling of escape sequences consisting of a string of digits
879 starting with one that is not zero is not straightforward. By experiment,
880 the way Perl works seems to be as follows:
881
882 Outside a character class, the digits are read as a decimal number. If the
883 number is less than 10, or if there are that many previous extracting
884 left brackets, then it is a back reference. Otherwise, up to three octal
885 digits are read to form an escaped byte. Thus \123 is likely to be octal
886 123 (cf \0123, which is octal 012 followed by the literal 3). If the octal
887 value is greater than 377, the least significant 8 bits are taken. Inside a
888 character class, \ followed by a digit is always an octal number. */
889
890 case CHAR_1: case CHAR_2: case CHAR_3: case CHAR_4: case CHAR_5:
891 case CHAR_6: case CHAR_7: case CHAR_8: case CHAR_9:
892
893 if (!isclass)
894 {
895 oldptr = ptr;
896 /* The integer range is limited by the machine's int representation. */
897 c -= CHAR_0;
898 while (IS_DIGIT(ptr[1]))
899 {
900 if (((unsigned int)c) > INT_MAX / 10) /* Integer overflow */
901 {
902 c = -1;
903 break;
904 }
905 c = c * 10 + *(++ptr) - CHAR_0;
906 }
907 if (((unsigned int)c) > INT_MAX) /* Integer overflow */
908 {
909 while (IS_DIGIT(ptr[1]))
910 ptr++;
911 *errorcodeptr = ERR61;
912 break;
913 }
914 if (c < 10 || c <= bracount)
915 {
916 c = -(ESC_REF + c);
917 break;
918 }
919 ptr = oldptr; /* Put the pointer back and fall through */
920 }
921
922 /* Handle an octal number following \. If the first digit is 8 or 9, Perl
923 generates a binary zero byte and treats the digit as a following literal.
924 Thus we have to pull back the pointer by one. */
925
926 if ((c = *ptr) >= CHAR_8)
927 {
928 ptr--;
929 c = 0;
930 break;
931 }
932
933 /* \0 always starts an octal number, but we may drop through to here with a
934 larger first octal digit. The original code used just to take the least
935 significant 8 bits of octal numbers (I think this is what early Perls used
936 to do). Nowadays we allow for larger numbers in UTF-8 mode, but no more
937 than 3 octal digits. */
938
939 case CHAR_0:
940 c -= CHAR_0;
941 while(i++ < 2 && ptr[1] >= CHAR_0 && ptr[1] <= CHAR_7)
942 c = c * 8 + *(++ptr) - CHAR_0;
943 if (!utf8 && c > 0xff) *errorcodeptr = ERR51;
944 break;
945
946 /* \x is complicated. \x{ddd} is a character number which can be greater
947 than 0xff in utf8 mode, but only if the ddd are hex digits. If not, { is
948 treated as a data character. */
949
950 case CHAR_x:
951 if ((options & PCRE_JAVASCRIPT_COMPAT) != 0)
952 {
953 /* In JavaScript, \x must be followed by two hexadecimal numbers.
954 Otherwise it is a lowercase x letter. */
955 if (MAX_255(ptr[1]) && (digitab[ptr[1]] & ctype_xdigit) != 0
956 && MAX_255(ptr[2]) && (digitab[ptr[2]] & ctype_xdigit) != 0)
957 {
958 c = 0;
959 for (i = 0; i < 2; ++i)
960 {
961 register int cc = *(++ptr);
962 #ifndef EBCDIC /* ASCII/UTF-8 coding */
963 if (cc >= CHAR_a) cc -= 32; /* Convert to upper case */
964 c = (c << 4) + cc - ((cc < CHAR_A)? CHAR_0 : (CHAR_A - 10));
965 #else /* EBCDIC coding */
966 if (cc >= CHAR_a && cc <= CHAR_z) cc += 64; /* Convert to upper case */
967 c = (c << 4) + cc - ((cc >= CHAR_0)? CHAR_0 : (CHAR_A - 10));
968 #endif
969 }
970 }
971 break;
972 }
973
974 if (ptr[1] == CHAR_LEFT_CURLY_BRACKET)
975 {
976 const pcre_uchar *pt = ptr + 2;
977 int count = 0;
978
979 c = 0;
980 while (MAX_255(*pt) && (digitab[*pt] & ctype_xdigit) != 0)
981 {
982 register int cc = *pt++;
983 if (c == 0 && cc == CHAR_0) continue; /* Leading zeroes */
984 count++;
985
986 #ifndef EBCDIC /* ASCII/UTF-8 coding */
987 if (cc >= CHAR_a) cc -= 32; /* Convert to upper case */
988 c = (c << 4) + cc - ((cc < CHAR_A)? CHAR_0 : (CHAR_A - 10));
989 #else /* EBCDIC coding */
990 if (cc >= CHAR_a && cc <= CHAR_z) cc += 64; /* Convert to upper case */
991 c = (c << 4) + cc - ((cc >= CHAR_0)? CHAR_0 : (CHAR_A - 10));
992 #endif
993 }
994
995 if (*pt == CHAR_RIGHT_CURLY_BRACKET)
996 {
997 #ifdef COMPILE_PCRE8
998 if (c < 0 || count > (utf8? 8:2)) *errorcodeptr = ERR34;
999 #else
1000 #ifdef COMPILE_PCRE16
1001 if (c < 0 || count > (utf8? 8:4)) *errorcodeptr = ERR34;
1002 #endif
1003 #endif
1004 ptr = pt;
1005 break;
1006 }
1007
1008 /* If the sequence of hex digits does not end with '}', then we don't
1009 recognize this construct; fall through to the normal \x handling. */
1010 }
1011
1012 /* Read just a single-byte hex-defined char */
1013
1014 c = 0;
1015 while (i++ < 2 && MAX_255(ptr[1]) && (digitab[ptr[1]] & ctype_xdigit) != 0)
1016 {
1017 int cc; /* Some compilers don't like */
1018 cc = *(++ptr); /* ++ in initializers */
1019 #ifndef EBCDIC /* ASCII/UTF-8 coding */
1020 if (cc >= CHAR_a) cc -= 32; /* Convert to upper case */
1021 c = c * 16 + cc - ((cc < CHAR_A)? CHAR_0 : (CHAR_A - 10));
1022 #else /* EBCDIC coding */
1023 if (cc <= CHAR_z) cc += 64; /* Convert to upper case */
1024 c = c * 16 + cc - ((cc >= CHAR_0)? CHAR_0 : (CHAR_A - 10));
1025 #endif
1026 }
1027 break;
1028
1029 /* For \c, a following letter is upper-cased; then the 0x40 bit is flipped.
1030 An error is given if the byte following \c is not an ASCII character. This
1031 coding is ASCII-specific, but then the whole concept of \cx is
1032 ASCII-specific. (However, an EBCDIC equivalent has now been added.) */
1033
1034 case CHAR_c:
1035 c = *(++ptr);
1036 if (c == 0)
1037 {
1038 *errorcodeptr = ERR2;
1039 break;
1040 }
1041 #ifndef EBCDIC /* ASCII/UTF-8 coding */
1042 if (c > 127) /* Excludes all non-ASCII in either mode */
1043 {
1044 *errorcodeptr = ERR68;
1045 break;
1046 }
1047 if (c >= CHAR_a && c <= CHAR_z) c -= 32;
1048 c ^= 0x40;
1049 #else /* EBCDIC coding */
1050 if (c >= CHAR_a && c <= CHAR_z) c += 64;
1051 c ^= 0xC0;
1052 #endif
1053 break;
1054
1055 /* PCRE_EXTRA enables extensions to Perl in the matter of escapes. Any
1056 other alphanumeric following \ is an error if PCRE_EXTRA was set;
1057 otherwise, for Perl compatibility, it is a literal. This code looks a bit
1058 odd, but there used to be some cases other than the default, and there may
1059 be again in future, so I haven't "optimized" it. */
1060
1061 default:
1062 if ((options & PCRE_EXTRA) != 0) switch(c)
1063 {
1064 default:
1065 *errorcodeptr = ERR3;
1066 break;
1067 }
1068 break;
1069 }
1070 }
1071
1072 /* Perl supports \N{name} for character names, as well as plain \N for "not
1073 newline". PCRE does not support \N{name}. However, it does support
1074 quantification such as \N{2,3}. */
1075
1076 if (c == -ESC_N && ptr[1] == CHAR_LEFT_CURLY_BRACKET &&
1077 !is_counted_repeat(ptr+2))
1078 *errorcodeptr = ERR37;
1079
1080 /* If PCRE_UCP is set, we change the values for \d etc. */
1081
1082 if ((options & PCRE_UCP) != 0 && c <= -ESC_D && c >= -ESC_w)
1083 c -= (ESC_DU - ESC_D);
1084
1085 /* Set the pointer to the final character before returning. */
1086
1087 *ptrptr = ptr;
1088 return c;
1089 }
1090
1091
1092
1093 #ifdef SUPPORT_UCP
1094 /*************************************************
1095 * Handle \P and \p *
1096 *************************************************/
1097
1098 /* This function is called after \P or \p has been encountered, provided that
1099 PCRE is compiled with support for Unicode properties. On entry, ptrptr is
1100 pointing at the P or p. On exit, it is pointing at the final character of the
1101 escape sequence.
1102
1103 Argument:
1104 ptrptr points to the pattern position pointer
1105 negptr points to a boolean that is set TRUE for negation else FALSE
1106 dptr points to an int that is set to the detailed property value
1107 errorcodeptr points to the error code variable
1108
1109 Returns: type value from ucp_type_table, or -1 for an invalid type
1110 */
1111
1112 static int
1113 get_ucp(const pcre_uchar **ptrptr, BOOL *negptr, int *dptr, int *errorcodeptr)
1114 {
1115 int c, i, bot, top;
1116 const pcre_uchar *ptr = *ptrptr;
1117 pcre_uchar name[32];
1118
1119 c = *(++ptr);
1120 if (c == 0) goto ERROR_RETURN;
1121
1122 *negptr = FALSE;
1123
1124 /* \P or \p can be followed by a name in {}, optionally preceded by ^ for
1125 negation. */
1126
1127 if (c == CHAR_LEFT_CURLY_BRACKET)
1128 {
1129 if (ptr[1] == CHAR_CIRCUMFLEX_ACCENT)
1130 {
1131 *negptr = TRUE;
1132 ptr++;
1133 }
1134 for (i = 0; i < (int)sizeof(name) - 1; i++)
1135 {
1136 c = *(++ptr);
1137 if (c == 0) goto ERROR_RETURN;
1138 if (c == CHAR_RIGHT_CURLY_BRACKET) break;
1139 name[i] = c;
1140 }
1141 if (c != CHAR_RIGHT_CURLY_BRACKET) goto ERROR_RETURN;
1142 name[i] = 0;
1143 }
1144
1145 /* Otherwise there is just one following character */
1146
1147 else
1148 {
1149 name[0] = c;
1150 name[1] = 0;
1151 }
1152
1153 *ptrptr = ptr;
1154
1155 /* Search for a recognized property name using binary chop */
1156
1157 bot = 0;
1158 top = PRIV(utt_size);
1159
1160 while (bot < top)
1161 {
1162 i = (bot + top) >> 1;
1163 c = STRCMP_UC_C8(name, PRIV(utt_names) + PRIV(utt)[i].name_offset);
1164 if (c == 0)
1165 {
1166 *dptr = PRIV(utt)[i].value;
1167 return PRIV(utt)[i].type;
1168 }
1169 if (c > 0) bot = i + 1; else top = i;
1170 }
1171
1172 *errorcodeptr = ERR47;
1173 *ptrptr = ptr;
1174 return -1;
1175
1176 ERROR_RETURN:
1177 *errorcodeptr = ERR46;
1178 *ptrptr = ptr;
1179 return -1;
1180 }
1181 #endif
1182
1183
1184
1185
1186 /*************************************************
1187 * Read repeat counts *
1188 *************************************************/
1189
1190 /* Read an item of the form {n,m} and return the values. This is called only
1191 after is_counted_repeat() has confirmed that a repeat-count quantifier exists,
1192 so the syntax is guaranteed to be correct, but we need to check the values.
1193
1194 Arguments:
1195 p pointer to first char after '{'
1196 minp pointer to int for min
1197 maxp pointer to int for max
1198 returned as -1 if no max
1199 errorcodeptr points to error code variable
1200
1201 Returns: pointer to '}' on success;
1202 current ptr on error, with errorcodeptr set non-zero
1203 */
1204
1205 static const pcre_uchar *
1206 read_repeat_counts(const pcre_uchar *p, int *minp, int *maxp, int *errorcodeptr)
1207 {
1208 int min = 0;
1209 int max = -1;
1210
1211 /* Read the minimum value and do a paranoid check: a negative value indicates
1212 an integer overflow. */
1213
1214 while (IS_DIGIT(*p)) min = min * 10 + *p++ - CHAR_0;
1215 if (min < 0 || min > 65535)
1216 {
1217 *errorcodeptr = ERR5;
1218 return p;
1219 }
1220
1221 /* Read the maximum value if there is one, and again do a paranoid on its size.
1222 Also, max must not be less than min. */
1223
1224 if (*p == CHAR_RIGHT_CURLY_BRACKET) max = min; else
1225 {
1226 if (*(++p) != CHAR_RIGHT_CURLY_BRACKET)
1227 {
1228 max = 0;
1229 while(IS_DIGIT(*p)) max = max * 10 + *p++ - CHAR_0;
1230 if (max < 0 || max > 65535)
1231 {
1232 *errorcodeptr = ERR5;
1233 return p;
1234 }
1235 if (max < min)
1236 {
1237 *errorcodeptr = ERR4;
1238 return p;
1239 }
1240 }
1241 }
1242
1243 /* Fill in the required variables, and pass back the pointer to the terminating
1244 '}'. */
1245
1246 *minp = min;
1247 *maxp = max;
1248 return p;
1249 }
1250
1251
1252
1253 /*************************************************
1254 * Subroutine for finding forward reference *
1255 *************************************************/
1256
1257 /* This recursive function is called only from find_parens() below. The
1258 top-level call starts at the beginning of the pattern. All other calls must
1259 start at a parenthesis. It scans along a pattern's text looking for capturing
1260 subpatterns, and counting them. If it finds a named pattern that matches the
1261 name it is given, it returns its number. Alternatively, if the name is NULL, it
1262 returns when it reaches a given numbered subpattern. Recursion is used to keep
1263 track of subpatterns that reset the capturing group numbers - the (?| feature.
1264
1265 This function was originally called only from the second pass, in which we know
1266 that if (?< or (?' or (?P< is encountered, the name will be correctly
1267 terminated because that is checked in the first pass. There is now one call to
1268 this function in the first pass, to check for a recursive back reference by
1269 name (so that we can make the whole group atomic). In this case, we need check
1270 only up to the current position in the pattern, and that is still OK because
1271 and previous occurrences will have been checked. To make this work, the test
1272 for "end of pattern" is a check against cd->end_pattern in the main loop,
1273 instead of looking for a binary zero. This means that the special first-pass
1274 call can adjust cd->end_pattern temporarily. (Checks for binary zero while
1275 processing items within the loop are OK, because afterwards the main loop will
1276 terminate.)
1277
1278 Arguments:
1279 ptrptr address of the current character pointer (updated)
1280 cd compile background data
1281 name name to seek, or NULL if seeking a numbered subpattern
1282 lorn name length, or subpattern number if name is NULL
1283 xmode TRUE if we are in /x mode
1284 utf8 TRUE if we are in UTF-8 mode
1285 count pointer to the current capturing subpattern number (updated)
1286
1287 Returns: the number of the named subpattern, or -1 if not found
1288 */
1289
1290 static int
1291 find_parens_sub(pcre_uchar **ptrptr, compile_data *cd, const pcre_uchar *name, int lorn,
1292 BOOL xmode, BOOL utf8, int *count)
1293 {
1294 pcre_uchar *ptr = *ptrptr;
1295 int start_count = *count;
1296 int hwm_count = start_count;
1297 BOOL dup_parens = FALSE;
1298
1299 /* If the first character is a parenthesis, check on the type of group we are
1300 dealing with. The very first call may not start with a parenthesis. */
1301
1302 if (ptr[0] == CHAR_LEFT_PARENTHESIS)
1303 {
1304 /* Handle specials such as (*SKIP) or (*UTF8) etc. */
1305
1306 if (ptr[1] == CHAR_ASTERISK) ptr += 2;
1307
1308 /* Handle a normal, unnamed capturing parenthesis. */
1309
1310 else if (ptr[1] != CHAR_QUESTION_MARK)
1311 {
1312 *count += 1;
1313 if (name == NULL && *count == lorn) return *count;
1314 ptr++;
1315 }
1316
1317 /* All cases now have (? at the start. Remember when we are in a group
1318 where the parenthesis numbers are duplicated. */
1319
1320 else if (ptr[2] == CHAR_VERTICAL_LINE)
1321 {
1322 ptr += 3;
1323 dup_parens = TRUE;
1324 }
1325
1326 /* Handle comments; all characters are allowed until a ket is reached. */
1327
1328 else if (ptr[2] == CHAR_NUMBER_SIGN)
1329 {
1330 for (ptr += 3; *ptr != 0; ptr++) if (*ptr == CHAR_RIGHT_PARENTHESIS) break;
1331 goto FAIL_EXIT;
1332 }
1333
1334 /* Handle a condition. If it is an assertion, just carry on so that it
1335 is processed as normal. If not, skip to the closing parenthesis of the
1336 condition (there can't be any nested parens). */
1337
1338 else if (ptr[2] == CHAR_LEFT_PARENTHESIS)
1339 {
1340 ptr += 2;
1341 if (ptr[1] != CHAR_QUESTION_MARK)
1342 {
1343 while (*ptr != 0 && *ptr != CHAR_RIGHT_PARENTHESIS) ptr++;
1344 if (*ptr != 0) ptr++;
1345 }
1346 }
1347
1348 /* Start with (? but not a condition. */
1349
1350 else
1351 {
1352 ptr += 2;
1353 if (*ptr == CHAR_P) ptr++; /* Allow optional P */
1354
1355 /* We have to disambiguate (?<! and (?<= from (?<name> for named groups */
1356
1357 if ((*ptr == CHAR_LESS_THAN_SIGN && ptr[1] != CHAR_EXCLAMATION_MARK &&
1358 ptr[1] != CHAR_EQUALS_SIGN) || *ptr == CHAR_APOSTROPHE)
1359 {
1360 int term;
1361 const pcre_uchar *thisname;
1362 *count += 1;
1363 if (name == NULL && *count == lorn) return *count;
1364 term = *ptr++;
1365 if (term == CHAR_LESS_THAN_SIGN) term = CHAR_GREATER_THAN_SIGN;
1366 thisname = ptr;
1367 while (*ptr != term) ptr++;
1368 if (name != NULL && lorn == ptr - thisname &&
1369 STRNCMP_UC_UC(name, thisname, lorn) == 0)
1370 return *count;
1371 term++;
1372 }
1373 }
1374 }
1375
1376 /* Past any initial parenthesis handling, scan for parentheses or vertical
1377 bars. Stop if we get to cd->end_pattern. Note that this is important for the
1378 first-pass call when this value is temporarily adjusted to stop at the current
1379 position. So DO NOT change this to a test for binary zero. */
1380
1381 for (; ptr < cd->end_pattern; ptr++)
1382 {
1383 /* Skip over backslashed characters and also entire \Q...\E */
1384
1385 if (*ptr == CHAR_BACKSLASH)
1386 {
1387 if (*(++ptr) == 0) goto FAIL_EXIT;
1388 if (*ptr == CHAR_Q) for (;;)
1389 {
1390 while (*(++ptr) != 0 && *ptr != CHAR_BACKSLASH) {};
1391 if (*ptr == 0) goto FAIL_EXIT;
1392 if (*(++ptr) == CHAR_E) break;
1393 }
1394 continue;
1395 }
1396
1397 /* Skip over character classes; this logic must be similar to the way they
1398 are handled for real. If the first character is '^', skip it. Also, if the
1399 first few characters (either before or after ^) are \Q\E or \E we skip them
1400 too. This makes for compatibility with Perl. Note the use of STR macros to
1401 encode "Q\\E" so that it works in UTF-8 on EBCDIC platforms. */
1402
1403 if (*ptr == CHAR_LEFT_SQUARE_BRACKET)
1404 {
1405 BOOL negate_class = FALSE;
1406 for (;;)
1407 {
1408 if (ptr[1] == CHAR_BACKSLASH)
1409 {
1410 if (ptr[2] == CHAR_E)
1411 ptr+= 2;
1412 else if (STRNCMP_UC_C8(ptr + 2,
1413 STR_Q STR_BACKSLASH STR_E, 3) == 0)
1414 ptr += 4;
1415 else
1416 break;
1417 }
1418 else if (!negate_class && ptr[1] == CHAR_CIRCUMFLEX_ACCENT)
1419 {
1420 negate_class = TRUE;
1421 ptr++;
1422 }
1423 else break;
1424 }
1425
1426 /* If the next character is ']', it is a data character that must be
1427 skipped, except in JavaScript compatibility mode. */
1428
1429 if (ptr[1] == CHAR_RIGHT_SQUARE_BRACKET &&
1430 (cd->external_options & PCRE_JAVASCRIPT_COMPAT) == 0)
1431 ptr++;
1432
1433 while (*(++ptr) != CHAR_RIGHT_SQUARE_BRACKET)
1434 {
1435 if (*ptr == 0) return -1;
1436 if (*ptr == CHAR_BACKSLASH)
1437 {
1438 if (*(++ptr) == 0) goto FAIL_EXIT;
1439 if (*ptr == CHAR_Q) for (;;)
1440 {
1441 while (*(++ptr) != 0 && *ptr != CHAR_BACKSLASH) {};
1442 if (*ptr == 0) goto FAIL_EXIT;
1443 if (*(++ptr) == CHAR_E) break;
1444 }
1445 continue;
1446 }
1447 }
1448 continue;
1449 }
1450
1451 /* Skip comments in /x mode */
1452
1453 if (xmode && *ptr == CHAR_NUMBER_SIGN)
1454 {
1455 ptr++;
1456 while (*ptr != 0)
1457 {
1458 if (IS_NEWLINE(ptr)) { ptr += cd->nllen - 1; break; }
1459 ptr++;
1460 #ifdef SUPPORT_UTF8
1461 if (utf8) while ((*ptr & 0xc0) == 0x80) ptr++;
1462 #endif
1463 }
1464 if (*ptr == 0) goto FAIL_EXIT;
1465 continue;
1466 }
1467
1468 /* Check for the special metacharacters */
1469
1470 if (*ptr == CHAR_LEFT_PARENTHESIS)
1471 {
1472 int rc = find_parens_sub(&ptr, cd, name, lorn, xmode, utf8, count);
1473 if (rc > 0) return rc;
1474 if (*ptr == 0) goto FAIL_EXIT;
1475 }
1476
1477 else if (*ptr == CHAR_RIGHT_PARENTHESIS)
1478 {
1479 if (dup_parens && *count < hwm_count) *count = hwm_count;
1480 goto FAIL_EXIT;
1481 }
1482
1483 else if (*ptr == CHAR_VERTICAL_LINE && dup_parens)
1484 {
1485 if (*count > hwm_count) hwm_count = *count;
1486 *count = start_count;
1487 }
1488 }
1489
1490 FAIL_EXIT:
1491 *ptrptr = ptr;
1492 return -1;
1493 }
1494
1495
1496
1497
1498 /*************************************************
1499 * Find forward referenced subpattern *
1500 *************************************************/
1501
1502 /* This function scans along a pattern's text looking for capturing
1503 subpatterns, and counting them. If it finds a named pattern that matches the
1504 name it is given, it returns its number. Alternatively, if the name is NULL, it
1505 returns when it reaches a given numbered subpattern. This is used for forward
1506 references to subpatterns. We used to be able to start this scan from the
1507 current compiling point, using the current count value from cd->bracount, and
1508 do it all in a single loop, but the addition of the possibility of duplicate
1509 subpattern numbers means that we have to scan from the very start, in order to
1510 take account of such duplicates, and to use a recursive function to keep track
1511 of the different types of group.
1512
1513 Arguments:
1514 cd compile background data
1515 name name to seek, or NULL if seeking a numbered subpattern
1516 lorn name length, or subpattern number if name is NULL
1517 xmode TRUE if we are in /x mode
1518 utf8 TRUE if we are in UTF-8 mode
1519
1520 Returns: the number of the found subpattern, or -1 if not found
1521 */
1522
1523 static int
1524 find_parens(compile_data *cd, const pcre_uchar *name, int lorn, BOOL xmode,
1525 BOOL utf8)
1526 {
1527 pcre_uchar *ptr = (pcre_uchar *)cd->start_pattern;
1528 int count = 0;
1529 int rc;
1530
1531 /* If the pattern does not start with an opening parenthesis, the first call
1532 to find_parens_sub() will scan right to the end (if necessary). However, if it
1533 does start with a parenthesis, find_parens_sub() will return when it hits the
1534 matching closing parens. That is why we have to have a loop. */
1535
1536 for (;;)
1537 {
1538 rc = find_parens_sub(&ptr, cd, name, lorn, xmode, utf8, &count);
1539 if (rc > 0 || *ptr++ == 0) break;
1540 }
1541
1542 return rc;
1543 }
1544
1545
1546
1547
1548 /*************************************************
1549 * Find first significant op code *
1550 *************************************************/
1551
1552 /* This is called by several functions that scan a compiled expression looking
1553 for a fixed first character, or an anchoring op code etc. It skips over things
1554 that do not influence this. For some calls, it makes sense to skip negative
1555 forward and all backward assertions, and also the \b assertion; for others it
1556 does not.
1557
1558 Arguments:
1559 code pointer to the start of the group
1560 skipassert TRUE if certain assertions are to be skipped
1561
1562 Returns: pointer to the first significant opcode
1563 */
1564
1565 static const pcre_uchar*
1566 first_significant_code(const pcre_uchar *code, BOOL skipassert)
1567 {
1568 for (;;)
1569 {
1570 switch ((int)*code)
1571 {
1572 case OP_ASSERT_NOT:
1573 case OP_ASSERTBACK:
1574 case OP_ASSERTBACK_NOT:
1575 if (!skipassert) return code;
1576 do code += GET(code, 1); while (*code == OP_ALT);
1577 code += PRIV(OP_lengths)[*code];
1578 break;
1579
1580 case OP_WORD_BOUNDARY:
1581 case OP_NOT_WORD_BOUNDARY:
1582 if (!skipassert) return code;
1583 /* Fall through */
1584
1585 case OP_CALLOUT:
1586 case OP_CREF:
1587 case OP_NCREF:
1588 case OP_RREF:
1589 case OP_NRREF:
1590 case OP_DEF:
1591 code += PRIV(OP_lengths)[*code];
1592 break;
1593
1594 default:
1595 return code;
1596 }
1597 }
1598 /* Control never reaches here */
1599 }
1600
1601
1602
1603
1604 /*************************************************
1605 * Find the fixed length of a branch *
1606 *************************************************/
1607
1608 /* Scan a branch and compute the fixed length of subject that will match it,
1609 if the length is fixed. This is needed for dealing with backward assertions.
1610 In UTF8 mode, the result is in characters rather than bytes. The branch is
1611 temporarily terminated with OP_END when this function is called.
1612
1613 This function is called when a backward assertion is encountered, so that if it
1614 fails, the error message can point to the correct place in the pattern.
1615 However, we cannot do this when the assertion contains subroutine calls,
1616 because they can be forward references. We solve this by remembering this case
1617 and doing the check at the end; a flag specifies which mode we are running in.
1618
1619 Arguments:
1620 code points to the start of the pattern (the bracket)
1621 utf8 TRUE in UTF-8 mode
1622 atend TRUE if called when the pattern is complete
1623 cd the "compile data" structure
1624
1625 Returns: the fixed length,
1626 or -1 if there is no fixed length,
1627 or -2 if \C was encountered (in UTF-8 mode only)
1628 or -3 if an OP_RECURSE item was encountered and atend is FALSE
1629 or -4 if an unknown opcode was encountered (internal error)
1630 */
1631
1632 static int
1633 find_fixedlength(pcre_uchar *code, BOOL utf8, BOOL atend, compile_data *cd)
1634 {
1635 int length = -1;
1636
1637 register int branchlength = 0;
1638 register pcre_uchar *cc = code + 1 + LINK_SIZE;
1639
1640 /* Scan along the opcodes for this branch. If we get to the end of the
1641 branch, check the length against that of the other branches. */
1642
1643 for (;;)
1644 {
1645 int d;
1646 pcre_uchar *ce, *cs;
1647 register int op = *cc;
1648 switch (op)
1649 {
1650 /* We only need to continue for OP_CBRA (normal capturing bracket) and
1651 OP_BRA (normal non-capturing bracket) because the other variants of these
1652 opcodes are all concerned with unlimited repeated groups, which of course
1653 are not of fixed length. */
1654
1655 case OP_CBRA:
1656 case OP_BRA:
1657 case OP_ONCE:
1658 case OP_ONCE_NC:
1659 case OP_COND:
1660 d = find_fixedlength(cc + ((op == OP_CBRA)? IMM2_SIZE : 0), utf8, atend, cd);
1661 if (d < 0) return d;
1662 branchlength += d;
1663 do cc += GET(cc, 1); while (*cc == OP_ALT);
1664 cc += 1 + LINK_SIZE;
1665 break;
1666
1667 /* Reached end of a branch; if it's a ket it is the end of a nested call.
1668 If it's ALT it is an alternation in a nested call. An ACCEPT is effectively
1669 an ALT. If it is END it's the end of the outer call. All can be handled by
1670 the same code. Note that we must not include the OP_KETRxxx opcodes here,
1671 because they all imply an unlimited repeat. */
1672
1673 case OP_ALT:
1674 case OP_KET:
1675 case OP_END:
1676 case OP_ACCEPT:
1677 case OP_ASSERT_ACCEPT:
1678 if (length < 0) length = branchlength;
1679 else if (length != branchlength) return -1;
1680 if (*cc != OP_ALT) return length;
1681 cc += 1 + LINK_SIZE;
1682 branchlength = 0;
1683 break;
1684
1685 /* A true recursion implies not fixed length, but a subroutine call may
1686 be OK. If the subroutine is a forward reference, we can't deal with
1687 it until the end of the pattern, so return -3. */
1688
1689 case OP_RECURSE:
1690 if (!atend) return -3;
1691 cs = ce = (pcre_uchar *)cd->start_code + GET(cc, 1); /* Start subpattern */
1692 do ce += GET(ce, 1); while (*ce == OP_ALT); /* End subpattern */
1693 if (cc > cs && cc < ce) return -1; /* Recursion */
1694 d = find_fixedlength(cs + 2, utf8, atend, cd);
1695 if (d < 0) return d;
1696 branchlength += d;
1697 cc += 1 + LINK_SIZE;
1698 break;
1699
1700 /* Skip over assertive subpatterns */
1701
1702 case OP_ASSERT:
1703 case OP_ASSERT_NOT:
1704 case OP_ASSERTBACK:
1705 case OP_ASSERTBACK_NOT:
1706 do cc += GET(cc, 1); while (*cc == OP_ALT);
1707 /* Fall through */
1708
1709 /* Skip over things that don't match chars */
1710
1711 case OP_MARK:
1712 case OP_PRUNE_ARG:
1713 case OP_SKIP_ARG:
1714 case OP_THEN_ARG:
1715 cc += cc[1] + PRIV(OP_lengths)[*cc];
1716 break;
1717
1718 case OP_CALLOUT:
1719 case OP_CIRC:
1720 case OP_CIRCM:
1721 case OP_CLOSE:
1722 case OP_COMMIT:
1723 case OP_CREF:
1724 case OP_DEF:
1725 case OP_DOLL:
1726 case OP_DOLLM:
1727 case OP_EOD:
1728 case OP_EODN:
1729 case OP_FAIL:
1730 case OP_NCREF:
1731 case OP_NRREF:
1732 case OP_NOT_WORD_BOUNDARY:
1733 case OP_PRUNE:
1734 case OP_REVERSE:
1735 case OP_RREF:
1736 case OP_SET_SOM:
1737 case OP_SKIP:
1738 case OP_SOD:
1739 case OP_SOM:
1740 case OP_THEN:
1741 case OP_WORD_BOUNDARY:
1742 cc += PRIV(OP_lengths)[*cc];
1743 break;
1744
1745 /* Handle literal characters */
1746
1747 case OP_CHAR:
1748 case OP_CHARI:
1749 case OP_NOT:
1750 case OP_NOTI:
1751 branchlength++;
1752 cc += 2;
1753 #ifdef SUPPORT_UTF8
1754 if (utf8 && cc[-1] >= 0xc0) cc += PRIV(utf8_table4)[cc[-1] & 0x3f];
1755 #endif
1756 break;
1757
1758 /* Handle exact repetitions. The count is already in characters, but we
1759 need to skip over a multibyte character in UTF8 mode. */
1760
1761 case OP_EXACT:
1762 case OP_EXACTI:
1763 case OP_NOTEXACT:
1764 case OP_NOTEXACTI:
1765 branchlength += GET2(cc,1);
1766 cc += 2 + IMM2_SIZE;
1767 #ifdef SUPPORT_UTF8
1768 if (utf8 && cc[-1] >= 0xc0) cc += PRIV(utf8_table4)[cc[-1] & 0x3f];
1769 #endif
1770 break;
1771
1772 case OP_TYPEEXACT:
1773 branchlength += GET2(cc,1);
1774 if (cc[1 + IMM2_SIZE] == OP_PROP || cc[1 + IMM2_SIZE] == OP_NOTPROP) cc += 2;
1775 cc += 1 + IMM2_SIZE + 1;
1776 break;
1777
1778 /* Handle single-char matchers */
1779
1780 case OP_PROP:
1781 case OP_NOTPROP:
1782 cc += 2;
1783 /* Fall through */
1784
1785 case OP_HSPACE:
1786 case OP_VSPACE:
1787 case OP_NOT_HSPACE:
1788 case OP_NOT_VSPACE:
1789 case OP_NOT_DIGIT:
1790 case OP_DIGIT:
1791 case OP_NOT_WHITESPACE:
1792 case OP_WHITESPACE:
1793 case OP_NOT_WORDCHAR:
1794 case OP_WORDCHAR:
1795 case OP_ANY:
1796 case OP_ALLANY:
1797 branchlength++;
1798 cc++;
1799 break;
1800
1801 /* The single-byte matcher isn't allowed. This only happens in UTF-8 mode;
1802 otherwise \C is coded as OP_ALLANY. */
1803
1804 case OP_ANYBYTE:
1805 return -2;
1806
1807 /* Check a class for variable quantification */
1808
1809 #if defined SUPPORT_UTF || defined COMPILE_PCRE16
1810 case OP_XCLASS:
1811 cc += GET(cc, 1) - PRIV(OP_lengths)[OP_CLASS];
1812 /* Fall through */
1813 #endif
1814
1815 case OP_CLASS:
1816 case OP_NCLASS:
1817 cc += PRIV(OP_lengths)[OP_CLASS];
1818
1819 switch (*cc)
1820 {
1821 case OP_CRPLUS:
1822 case OP_CRMINPLUS:
1823 case OP_CRSTAR:
1824 case OP_CRMINSTAR:
1825 case OP_CRQUERY:
1826 case OP_CRMINQUERY:
1827 return -1;
1828
1829 case OP_CRRANGE:
1830 case OP_CRMINRANGE:
1831 if (GET2(cc,1) != GET2(cc,1+IMM2_SIZE)) return -1;
1832 branchlength += GET2(cc,1);
1833 cc += 1 + 2 * IMM2_SIZE;
1834 break;
1835
1836 default:
1837 branchlength++;
1838 }
1839 break;
1840
1841 /* Anything else is variable length */
1842
1843 case OP_ANYNL:
1844 case OP_BRAMINZERO:
1845 case OP_BRAPOS:
1846 case OP_BRAPOSZERO:
1847 case OP_BRAZERO:
1848 case OP_CBRAPOS:
1849 case OP_EXTUNI:
1850 case OP_KETRMAX:
1851 case OP_KETRMIN:
1852 case OP_KETRPOS:
1853 case OP_MINPLUS:
1854 case OP_MINPLUSI:
1855 case OP_MINQUERY:
1856 case OP_MINQUERYI:
1857 case OP_MINSTAR:
1858 case OP_MINSTARI:
1859 case OP_MINUPTO:
1860 case OP_MINUPTOI:
1861 case OP_NOTMINPLUS:
1862 case OP_NOTMINPLUSI:
1863 case OP_NOTMINQUERY:
1864 case OP_NOTMINQUERYI:
1865 case OP_NOTMINSTAR:
1866 case OP_NOTMINSTARI:
1867 case OP_NOTMINUPTO:
1868 case OP_NOTMINUPTOI:
1869 case OP_NOTPLUS:
1870 case OP_NOTPLUSI:
1871 case OP_NOTPOSPLUS:
1872 case OP_NOTPOSPLUSI:
1873 case OP_NOTPOSQUERY:
1874 case OP_NOTPOSQUERYI:
1875 case OP_NOTPOSSTAR:
1876 case OP_NOTPOSSTARI:
1877 case OP_NOTPOSUPTO:
1878 case OP_NOTPOSUPTOI:
1879 case OP_NOTQUERY:
1880 case OP_NOTQUERYI:
1881 case OP_NOTSTAR:
1882 case OP_NOTSTARI:
1883 case OP_NOTUPTO:
1884 case OP_NOTUPTOI:
1885 case OP_PLUS:
1886 case OP_PLUSI:
1887 case OP_POSPLUS:
1888 case OP_POSPLUSI:
1889 case OP_POSQUERY:
1890 case OP_POSQUERYI:
1891 case OP_POSSTAR:
1892 case OP_POSSTARI:
1893 case OP_POSUPTO:
1894 case OP_POSUPTOI:
1895 case OP_QUERY:
1896 case OP_QUERYI:
1897 case OP_REF:
1898 case OP_REFI:
1899 case OP_SBRA:
1900 case OP_SBRAPOS:
1901 case OP_SCBRA:
1902 case OP_SCBRAPOS:
1903 case OP_SCOND:
1904 case OP_SKIPZERO:
1905 case OP_STAR:
1906 case OP_STARI:
1907 case OP_TYPEMINPLUS:
1908 case OP_TYPEMINQUERY:
1909 case OP_TYPEMINSTAR:
1910 case OP_TYPEMINUPTO:
1911 case OP_TYPEPLUS:
1912 case OP_TYPEPOSPLUS:
1913 case OP_TYPEPOSQUERY:
1914 case OP_TYPEPOSSTAR:
1915 case OP_TYPEPOSUPTO:
1916 case OP_TYPEQUERY:
1917 case OP_TYPESTAR:
1918 case OP_TYPEUPTO:
1919 case OP_UPTO:
1920 case OP_UPTOI:
1921 return -1;
1922
1923 /* Catch unrecognized opcodes so that when new ones are added they
1924 are not forgotten, as has happened in the past. */
1925
1926 default:
1927 return -4;
1928 }
1929 }
1930 /* Control never gets here */
1931 }
1932
1933
1934
1935
1936 /*************************************************
1937 * Scan compiled regex for specific bracket *
1938 *************************************************/
1939
1940 /* This little function scans through a compiled pattern until it finds a
1941 capturing bracket with the given number, or, if the number is negative, an
1942 instance of OP_REVERSE for a lookbehind. The function is global in the C sense
1943 so that it can be called from pcre_study() when finding the minimum matching
1944 length.
1945
1946 Arguments:
1947 code points to start of expression
1948 utf8 TRUE in UTF-8 mode
1949 number the required bracket number or negative to find a lookbehind
1950
1951 Returns: pointer to the opcode for the bracket, or NULL if not found
1952 */
1953
1954 const pcre_uchar *
1955 PRIV(find_bracket)(const pcre_uchar *code, BOOL utf8, int number)
1956 {
1957 for (;;)
1958 {
1959 register int c = *code;
1960
1961 if (c == OP_END) return NULL;
1962
1963 /* XCLASS is used for classes that cannot be represented just by a bit
1964 map. This includes negated single high-valued characters. The length in
1965 the table is zero; the actual length is stored in the compiled code. */
1966
1967 if (c == OP_XCLASS) code += GET(code, 1);
1968
1969 /* Handle recursion */
1970
1971 else if (c == OP_REVERSE)
1972 {
1973 if (number < 0) return (pcre_uchar *)code;
1974 code += PRIV(OP_lengths)[c];
1975 }
1976
1977 /* Handle capturing bracket */
1978
1979 else if (c == OP_CBRA || c == OP_SCBRA ||
1980 c == OP_CBRAPOS || c == OP_SCBRAPOS)
1981 {
1982 int n = GET2(code, 1+LINK_SIZE);
1983 if (n == number) return (pcre_uchar *)code;
1984 code += PRIV(OP_lengths)[c];
1985 }
1986
1987 /* Otherwise, we can get the item's length from the table, except that for
1988 repeated character types, we have to test for \p and \P, which have an extra
1989 two bytes of parameters, and for MARK/PRUNE/SKIP/THEN with an argument, we
1990 must add in its length. */
1991
1992 else
1993 {
1994 switch(c)
1995 {
1996 case OP_TYPESTAR:
1997 case OP_TYPEMINSTAR:
1998 case OP_TYPEPLUS:
1999 case OP_TYPEMINPLUS:
2000 case OP_TYPEQUERY:
2001 case OP_TYPEMINQUERY:
2002 case OP_TYPEPOSSTAR:
2003 case OP_TYPEPOSPLUS:
2004 case OP_TYPEPOSQUERY:
2005 if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
2006 break;
2007
2008 case OP_TYPEUPTO:
2009 case OP_TYPEMINUPTO:
2010 case OP_TYPEEXACT:
2011 case OP_TYPEPOSUPTO:
2012 if (code[1 + IMM2_SIZE] == OP_PROP
2013 || code[1 + IMM2_SIZE] == OP_NOTPROP) code += 2;
2014 break;
2015
2016 case OP_MARK:
2017 case OP_PRUNE_ARG:
2018 case OP_SKIP_ARG:
2019 code += code[1];
2020 break;
2021
2022 case OP_THEN_ARG:
2023 code += code[1];
2024 break;
2025 }
2026
2027 /* Add in the fixed length from the table */
2028
2029 code += PRIV(OP_lengths)[c];
2030
2031 /* In UTF-8 mode, opcodes that are followed by a character may be followed by
2032 a multi-byte character. The length in the table is a minimum, so we have to
2033 arrange to skip the extra bytes. */
2034
2035 #ifdef SUPPORT_UTF8
2036 if (utf8) switch(c)
2037 {
2038 case OP_CHAR:
2039 case OP_CHARI:
2040 case OP_EXACT:
2041 case OP_EXACTI:
2042 case OP_UPTO:
2043 case OP_UPTOI:
2044 case OP_MINUPTO:
2045 case OP_MINUPTOI:
2046 case OP_POSUPTO:
2047 case OP_POSUPTOI:
2048 case OP_STAR:
2049 case OP_STARI:
2050 case OP_MINSTAR:
2051 case OP_MINSTARI:
2052 case OP_POSSTAR:
2053 case OP_POSSTARI:
2054 case OP_PLUS:
2055 case OP_PLUSI:
2056 case OP_MINPLUS:
2057 case OP_MINPLUSI:
2058 case OP_POSPLUS:
2059 case OP_POSPLUSI:
2060 case OP_QUERY:
2061 case OP_QUERYI:
2062 case OP_MINQUERY:
2063 case OP_MINQUERYI:
2064 case OP_POSQUERY:
2065 case OP_POSQUERYI:
2066 if (code[-1] >= 0xc0) code += PRIV(utf8_table4)[code[-1] & 0x3f];
2067 break;
2068 }
2069 #else
2070 (void)(utf8); /* Keep compiler happy by referencing function argument */
2071 #endif
2072 }
2073 }
2074 }
2075
2076
2077
2078 /*************************************************
2079 * Scan compiled regex for recursion reference *
2080 *************************************************/
2081
2082 /* This little function scans through a compiled pattern until it finds an
2083 instance of OP_RECURSE.
2084
2085 Arguments:
2086 code points to start of expression
2087 utf8 TRUE in UTF-8 mode
2088
2089 Returns: pointer to the opcode for OP_RECURSE, or NULL if not found
2090 */
2091
2092 static const pcre_uchar *
2093 find_recurse(const pcre_uchar *code, BOOL utf8)
2094 {
2095 for (;;)
2096 {
2097 register int c = *code;
2098 if (c == OP_END) return NULL;
2099 if (c == OP_RECURSE) return code;
2100
2101 /* XCLASS is used for classes that cannot be represented just by a bit
2102 map. This includes negated single high-valued characters. The length in
2103 the table is zero; the actual length is stored in the compiled code. */
2104
2105 if (c == OP_XCLASS) code += GET(code, 1);
2106
2107 /* Otherwise, we can get the item's length from the table, except that for
2108 repeated character types, we have to test for \p and \P, which have an extra
2109 two bytes of parameters, and for MARK/PRUNE/SKIP/THEN with an argument, we
2110 must add in its length. */
2111
2112 else
2113 {
2114 switch(c)
2115 {
2116 case OP_TYPESTAR:
2117 case OP_TYPEMINSTAR:
2118 case OP_TYPEPLUS:
2119 case OP_TYPEMINPLUS:
2120 case OP_TYPEQUERY:
2121 case OP_TYPEMINQUERY:
2122 case OP_TYPEPOSSTAR:
2123 case OP_TYPEPOSPLUS:
2124 case OP_TYPEPOSQUERY:
2125 if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
2126 break;
2127
2128 case OP_TYPEPOSUPTO:
2129 case OP_TYPEUPTO:
2130 case OP_TYPEMINUPTO:
2131 case OP_TYPEEXACT:
2132 if (code[1 + IMM2_SIZE] == OP_PROP
2133 || code[1 + IMM2_SIZE] == OP_NOTPROP) code += 2;
2134 break;
2135
2136 case OP_MARK:
2137 case OP_PRUNE_ARG:
2138 case OP_SKIP_ARG:
2139 code += code[1];
2140 break;
2141
2142 case OP_THEN_ARG:
2143 code += code[1];
2144 break;
2145 }
2146
2147 /* Add in the fixed length from the table */
2148
2149 code += PRIV(OP_lengths)[c];
2150
2151 /* In UTF-8 mode, opcodes that are followed by a character may be followed
2152 by a multi-byte character. The length in the table is a minimum, so we have
2153 to arrange to skip the extra bytes. */
2154
2155 #ifdef SUPPORT_UTF8
2156 if (utf8) switch(c)
2157 {
2158 case OP_CHAR:
2159 case OP_CHARI:
2160 case OP_EXACT:
2161 case OP_EXACTI:
2162 case OP_UPTO:
2163 case OP_UPTOI:
2164 case OP_MINUPTO:
2165 case OP_MINUPTOI:
2166 case OP_POSUPTO:
2167 case OP_POSUPTOI:
2168 case OP_STAR:
2169 case OP_STARI:
2170 case OP_MINSTAR:
2171 case OP_MINSTARI:
2172 case OP_POSSTAR:
2173 case OP_POSSTARI:
2174 case OP_PLUS:
2175 case OP_PLUSI:
2176 case OP_MINPLUS:
2177 case OP_MINPLUSI:
2178 case OP_POSPLUS:
2179 case OP_POSPLUSI:
2180 case OP_QUERY:
2181 case OP_QUERYI:
2182 case OP_MINQUERY:
2183 case OP_MINQUERYI:
2184 case OP_POSQUERY:
2185 case OP_POSQUERYI:
2186 if (code[-1] >= 0xc0) code += PRIV(utf8_table4)[code[-1] & 0x3f];
2187 break;
2188 }
2189 #else
2190 (void)(utf8); /* Keep compiler happy by referencing function argument */
2191 #endif
2192 }
2193 }
2194 }
2195
2196
2197
2198 /*************************************************
2199 * Scan compiled branch for non-emptiness *
2200 *************************************************/
2201
2202 /* This function scans through a branch of a compiled pattern to see whether it
2203 can match the empty string or not. It is called from could_be_empty()
2204 below and from compile_branch() when checking for an unlimited repeat of a
2205 group that can match nothing. Note that first_significant_code() skips over
2206 backward and negative forward assertions when its final argument is TRUE. If we
2207 hit an unclosed bracket, we return "empty" - this means we've struck an inner
2208 bracket whose current branch will already have been scanned.
2209
2210 Arguments:
2211 code points to start of search
2212 endcode points to where to stop
2213 utf8 TRUE if in UTF8 mode
2214 cd contains pointers to tables etc.
2215
2216 Returns: TRUE if what is matched could be empty
2217 */
2218
2219 static BOOL
2220 could_be_empty_branch(const pcre_uchar *code, const pcre_uchar *endcode,
2221 BOOL utf8, compile_data *cd)
2222 {
2223 register int c;
2224 for (code = first_significant_code(code + PRIV(OP_lengths)[*code], TRUE);
2225 code < endcode;
2226 code = first_significant_code(code + PRIV(OP_lengths)[c], TRUE))
2227 {
2228 const pcre_uchar *ccode;
2229
2230 c = *code;
2231
2232 /* Skip over forward assertions; the other assertions are skipped by
2233 first_significant_code() with a TRUE final argument. */
2234
2235 if (c == OP_ASSERT)
2236 {
2237 do code += GET(code, 1); while (*code == OP_ALT);
2238 c = *code;
2239 continue;
2240 }
2241
2242 /* For a recursion/subroutine call, if its end has been reached, which
2243 implies a backward reference subroutine call, we can scan it. If it's a
2244 forward reference subroutine call, we can't. To detect forward reference
2245 we have to scan up the list that is kept in the workspace. This function is
2246 called only when doing the real compile, not during the pre-compile that
2247 measures the size of the compiled pattern. */
2248
2249 if (c == OP_RECURSE)
2250 {
2251 const pcre_uchar *scode;
2252 BOOL empty_branch;
2253
2254 /* Test for forward reference */
2255
2256 for (scode = cd->start_workspace; scode < cd->hwm; scode += LINK_SIZE)
2257 if (GET(scode, 0) == code + 1 - cd->start_code) return TRUE;
2258
2259 /* Not a forward reference, test for completed backward reference */
2260
2261 empty_branch = FALSE;
2262 scode = cd->start_code + GET(code, 1);
2263 if (GET(scode, 1) == 0) return TRUE; /* Unclosed */
2264
2265 /* Completed backwards reference */
2266
2267 do
2268 {
2269 if (could_be_empty_branch(scode, endcode, utf8, cd))
2270 {
2271 empty_branch = TRUE;
2272 break;
2273 }
2274 scode += GET(scode, 1);
2275 }
2276 while (*scode == OP_ALT);
2277
2278 if (!empty_branch) return FALSE; /* All branches are non-empty */
2279 continue;
2280 }
2281
2282 /* Groups with zero repeats can of course be empty; skip them. */
2283
2284 if (c == OP_BRAZERO || c == OP_BRAMINZERO || c == OP_SKIPZERO ||
2285 c == OP_BRAPOSZERO)
2286 {
2287 code += PRIV(OP_lengths)[c];
2288 do code += GET(code, 1); while (*code == OP_ALT);
2289 c = *code;
2290 continue;
2291 }
2292
2293 /* A nested group that is already marked as "could be empty" can just be
2294 skipped. */
2295
2296 if (c == OP_SBRA || c == OP_SBRAPOS ||
2297 c == OP_SCBRA || c == OP_SCBRAPOS)
2298 {
2299 do code += GET(code, 1); while (*code == OP_ALT);
2300 c = *code;
2301 continue;
2302 }
2303
2304 /* For other groups, scan the branches. */
2305
2306 if (c == OP_BRA || c == OP_BRAPOS ||
2307 c == OP_CBRA || c == OP_CBRAPOS ||
2308 c == OP_ONCE || c == OP_ONCE_NC ||
2309 c == OP_COND)
2310 {
2311 BOOL empty_branch;
2312 if (GET(code, 1) == 0) return TRUE; /* Hit unclosed bracket */
2313
2314 /* If a conditional group has only one branch, there is a second, implied,
2315 empty branch, so just skip over the conditional, because it could be empty.
2316 Otherwise, scan the individual branches of the group. */
2317
2318 if (c == OP_COND && code[GET(code, 1)] != OP_ALT)
2319 code += GET(code, 1);
2320 else
2321 {
2322 empty_branch = FALSE;
2323 do
2324 {
2325 if (!empty_branch && could_be_empty_branch(code, endcode, utf8, cd))
2326 empty_branch = TRUE;
2327 code += GET(code, 1);
2328 }
2329 while (*code == OP_ALT);
2330 if (!empty_branch) return FALSE; /* All branches are non-empty */
2331 }
2332
2333 c = *code;
2334 continue;
2335 }
2336
2337 /* Handle the other opcodes */
2338
2339 switch (c)
2340 {
2341 /* Check for quantifiers after a class. XCLASS is used for classes that
2342 cannot be represented just by a bit map. This includes negated single
2343 high-valued characters. The length in PRIV(OP_lengths)[] is zero; the
2344 actual length is stored in the compiled code, so we must update "code"
2345 here. */
2346
2347 #ifdef SUPPORT_UTF8
2348 case OP_XCLASS:
2349 ccode = code += GET(code, 1);
2350 goto CHECK_CLASS_REPEAT;
2351 #endif
2352
2353 case OP_CLASS:
2354 case OP_NCLASS:
2355 ccode = code + PRIV(OP_lengths)[OP_CLASS];
2356
2357 #ifdef SUPPORT_UTF8
2358 CHECK_CLASS_REPEAT:
2359 #endif
2360
2361 switch (*ccode)
2362 {
2363 case OP_CRSTAR: /* These could be empty; continue */
2364 case OP_CRMINSTAR:
2365 case OP_CRQUERY:
2366 case OP_CRMINQUERY:
2367 break;
2368
2369 default: /* Non-repeat => class must match */
2370 case OP_CRPLUS: /* These repeats aren't empty */
2371 case OP_CRMINPLUS:
2372 return FALSE;
2373
2374 case OP_CRRANGE:
2375 case OP_CRMINRANGE:
2376 if (GET2(ccode, 1) > 0) return FALSE; /* Minimum > 0 */
2377 break;
2378 }
2379 break;
2380
2381 /* Opcodes that must match a character */
2382
2383 case OP_PROP:
2384 case OP_NOTPROP:
2385 case OP_EXTUNI:
2386 case OP_NOT_DIGIT:
2387 case OP_DIGIT:
2388 case OP_NOT_WHITESPACE:
2389 case OP_WHITESPACE:
2390 case OP_NOT_WORDCHAR:
2391 case OP_WORDCHAR:
2392 case OP_ANY:
2393 case OP_ALLANY:
2394 case OP_ANYBYTE:
2395 case OP_CHAR:
2396 case OP_CHARI:
2397 case OP_NOT:
2398 case OP_NOTI:
2399 case OP_PLUS:
2400 case OP_MINPLUS:
2401 case OP_POSPLUS:
2402 case OP_EXACT:
2403 case OP_NOTPLUS:
2404 case OP_NOTMINPLUS:
2405 case OP_NOTPOSPLUS:
2406 case OP_NOTEXACT:
2407 case OP_TYPEPLUS:
2408 case OP_TYPEMINPLUS:
2409 case OP_TYPEPOSPLUS:
2410 case OP_TYPEEXACT:
2411 return FALSE;
2412
2413 /* These are going to continue, as they may be empty, but we have to
2414 fudge the length for the \p and \P cases. */
2415
2416 case OP_TYPESTAR:
2417 case OP_TYPEMINSTAR:
2418 case OP_TYPEPOSSTAR:
2419 case OP_TYPEQUERY:
2420 case OP_TYPEMINQUERY:
2421 case OP_TYPEPOSQUERY:
2422 if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
2423 break;
2424
2425 /* Same for these */
2426
2427 case OP_TYPEUPTO:
2428 case OP_TYPEMINUPTO:
2429 case OP_TYPEPOSUPTO:
2430 if (code[1 + IMM2_SIZE] == OP_PROP
2431 || code[1 + IMM2_SIZE] == OP_NOTPROP) code += 2;
2432 break;
2433
2434 /* End of branch */
2435
2436 case OP_KET:
2437 case OP_KETRMAX:
2438 case OP_KETRMIN:
2439 case OP_KETRPOS:
2440 case OP_ALT:
2441 return TRUE;
2442
2443 /* In UTF-8 mode, STAR, MINSTAR, POSSTAR, QUERY, MINQUERY, POSQUERY, UPTO,
2444 MINUPTO, and POSUPTO may be followed by a multibyte character */
2445
2446 #ifdef SUPPORT_UTF8
2447 case OP_STAR:
2448 case OP_STARI:
2449 case OP_MINSTAR:
2450 case OP_MINSTARI:
2451 case OP_POSSTAR:
2452 case OP_POSSTARI:
2453 case OP_QUERY:
2454 case OP_QUERYI:
2455 case OP_MINQUERY:
2456 case OP_MINQUERYI:
2457 case OP_POSQUERY:
2458 case OP_POSQUERYI:
2459 if (utf8 && code[1] >= 0xc0) code += PRIV(utf8_table4)[code[1] & 0x3f];
2460 break;
2461
2462 case OP_UPTO:
2463 case OP_UPTOI:
2464 case OP_MINUPTO:
2465 case OP_MINUPTOI:
2466 case OP_POSUPTO:
2467 case OP_POSUPTOI:
2468 if (utf8 && code[1 + IMM2_SIZE] >= 0xc0) code += PRIV(utf8_table4)[code[1 + IMM2_SIZE] & 0x3f];
2469 break;
2470 #endif
2471
2472 /* MARK, and PRUNE/SKIP/THEN with an argument must skip over the argument
2473 string. */
2474
2475 case OP_MARK:
2476 case OP_PRUNE_ARG:
2477 case OP_SKIP_ARG:
2478 code += code[1];
2479 break;
2480
2481 case OP_THEN_ARG:
2482 code += code[1];
2483 break;
2484
2485 /* None of the remaining opcodes are required to match a character. */
2486
2487 default:
2488 break;
2489 }
2490 }
2491
2492 return TRUE;
2493 }
2494
2495
2496
2497 /*************************************************
2498 * Scan compiled regex for non-emptiness *
2499 *************************************************/
2500
2501 /* This function is called to check for left recursive calls. We want to check
2502 the current branch of the current pattern to see if it could match the empty
2503 string. If it could, we must look outwards for branches at other levels,
2504 stopping when we pass beyond the bracket which is the subject of the recursion.
2505 This function is called only during the real compile, not during the
2506 pre-compile.
2507
2508 Arguments:
2509 code points to start of the recursion
2510 endcode points to where to stop (current RECURSE item)
2511 bcptr points to the chain of current (unclosed) branch starts
2512 utf8 TRUE if in UTF-8 mode
2513 cd pointers to tables etc
2514
2515 Returns: TRUE if what is matched could be empty
2516 */
2517
2518 static BOOL
2519 could_be_empty(const pcre_uchar *code, const pcre_uchar *endcode,
2520 branch_chain *bcptr, BOOL utf8, compile_data *cd)
2521 {
2522 while (bcptr != NULL && bcptr->current_branch >= code)
2523 {
2524 if (!could_be_empty_branch(bcptr->current_branch, endcode, utf8, cd))
2525 return FALSE;
2526 bcptr = bcptr->outer;
2527 }
2528 return TRUE;
2529 }
2530
2531
2532
2533 /*************************************************
2534 * Check for POSIX class syntax *
2535 *************************************************/
2536
2537 /* This function is called when the sequence "[:" or "[." or "[=" is
2538 encountered in a character class. It checks whether this is followed by a
2539 sequence of characters terminated by a matching ":]" or ".]" or "=]". If we
2540 reach an unescaped ']' without the special preceding character, return FALSE.
2541
2542 Originally, this function only recognized a sequence of letters between the
2543 terminators, but it seems that Perl recognizes any sequence of characters,
2544 though of course unknown POSIX names are subsequently rejected. Perl gives an
2545 "Unknown POSIX class" error for [:f\oo:] for example, where previously PCRE
2546 didn't consider this to be a POSIX class. Likewise for [:1234:].
2547
2548 The problem in trying to be exactly like Perl is in the handling of escapes. We
2549 have to be sure that [abc[:x\]pqr] is *not* treated as containing a POSIX
2550 class, but [abc[:x\]pqr:]] is (so that an error can be generated). The code
2551 below handles the special case of \], but does not try to do any other escape
2552 processing. This makes it different from Perl for cases such as [:l\ower:]
2553 where Perl recognizes it as the POSIX class "lower" but PCRE does not recognize
2554 "l\ower". This is a lesser evil that not diagnosing bad classes when Perl does,
2555 I think.
2556
2557 A user pointed out that PCRE was rejecting [:a[:digit:]] whereas Perl was not.
2558 It seems that the appearance of a nested POSIX class supersedes an apparent
2559 external class. For example, [:a[:digit:]b:] matches "a", "b", ":", or
2560 a digit.
2561
2562 In Perl, unescaped square brackets may also appear as part of class names. For
2563 example, [:a[:abc]b:] gives unknown POSIX class "[:abc]b:]". However, for
2564 [:a[:abc]b][b:] it gives unknown POSIX class "[:abc]b][b:]", which does not
2565 seem right at all. PCRE does not allow closing square brackets in POSIX class
2566 names.
2567
2568 Arguments:
2569 ptr pointer to the initial [
2570 endptr where to return the end pointer
2571
2572 Returns: TRUE or FALSE
2573 */
2574
2575 static BOOL
2576 check_posix_syntax(const pcre_uchar *ptr, const pcre_uchar **endptr)
2577 {
2578 int terminator; /* Don't combine these lines; the Solaris cc */
2579 terminator = *(++ptr); /* compiler warns about "non-constant" initializer. */
2580 for (++ptr; *ptr != 0; ptr++)
2581 {
2582 if (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET)
2583 ptr++;
2584 else if (*ptr == CHAR_RIGHT_SQUARE_BRACKET) return FALSE;
2585 else
2586 {
2587 if (*ptr == terminator && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET)
2588 {
2589 *endptr = ptr;
2590 return TRUE;
2591 }
2592 if (*ptr == CHAR_LEFT_SQUARE_BRACKET &&
2593 (ptr[1] == CHAR_COLON || ptr[1] == CHAR_DOT ||
2594 ptr[1] == CHAR_EQUALS_SIGN) &&
2595 check_posix_syntax(ptr, endptr))
2596 return FALSE;
2597 }
2598 }
2599 return FALSE;
2600 }
2601
2602
2603
2604
2605 /*************************************************
2606 * Check POSIX class name *
2607 *************************************************/
2608
2609 /* This function is called to check the name given in a POSIX-style class entry
2610 such as [:alnum:].
2611
2612 Arguments:
2613 ptr points to the first letter
2614 len the length of the name
2615
2616 Returns: a value representing the name, or -1 if unknown
2617 */
2618
2619 static int
2620 check_posix_name(const pcre_uchar *ptr, int len)
2621 {
2622 const char *pn = posix_names;
2623 register int yield = 0;
2624 while (posix_name_lengths[yield] != 0)
2625 {
2626 if (len == posix_name_lengths[yield] &&
2627 STRNCMP_UC_C8(ptr, pn, len) == 0) return yield;
2628 pn += posix_name_lengths[yield] + 1;
2629 yield++;
2630 }
2631 return -1;
2632 }
2633
2634
2635 /*************************************************
2636 * Adjust OP_RECURSE items in repeated group *
2637 *************************************************/
2638
2639 /* OP_RECURSE items contain an offset from the start of the regex to the group
2640 that is referenced. This means that groups can be replicated for fixed
2641 repetition simply by copying (because the recursion is allowed to refer to
2642 earlier groups that are outside the current group). However, when a group is
2643 optional (i.e. the minimum quantifier is zero), OP_BRAZERO or OP_SKIPZERO is
2644 inserted before it, after it has been compiled. This means that any OP_RECURSE
2645 items within it that refer to the group itself or any contained groups have to
2646 have their offsets adjusted. That one of the jobs of this function. Before it
2647 is called, the partially compiled regex must be temporarily terminated with
2648 OP_END.
2649
2650 This function has been extended with the possibility of forward references for
2651 recursions and subroutine calls. It must also check the list of such references
2652 for the group we are dealing with. If it finds that one of the recursions in
2653 the current group is on this list, it adjusts the offset in the list, not the
2654 value in the reference (which is a group number).
2655
2656 Arguments:
2657 group points to the start of the group
2658 adjust the amount by which the group is to be moved
2659 utf8 TRUE in UTF-8 mode
2660 cd contains pointers to tables etc.
2661 save_hwm the hwm forward reference pointer at the start of the group
2662
2663 Returns: nothing
2664 */
2665
2666 static void
2667 adjust_recurse(pcre_uchar *group, int adjust, BOOL utf8, compile_data *cd,
2668 pcre_uchar *save_hwm)
2669 {
2670 pcre_uchar *ptr = group;
2671
2672 while ((ptr = (pcre_uchar *)find_recurse(ptr, utf8)) != NULL)
2673 {
2674 int offset;
2675 pcre_uchar *hc;
2676
2677 /* See if this recursion is on the forward reference list. If so, adjust the
2678 reference. */
2679
2680 for (hc = save_hwm; hc < cd->hwm; hc += LINK_SIZE)
2681 {
2682 offset = GET(hc, 0);
2683 if (cd->start_code + offset == ptr + 1)
2684 {
2685 PUT(hc, 0, offset + adjust);
2686 break;
2687 }
2688 }
2689
2690 /* Otherwise, adjust the recursion offset if it's after the start of this
2691 group. */
2692
2693 if (hc >= cd->hwm)
2694 {
2695 offset = GET(ptr, 1);
2696 if (cd->start_code + offset >= group) PUT(ptr, 1, offset + adjust);
2697 }
2698
2699 ptr += 1 + LINK_SIZE;
2700 }
2701 }
2702
2703
2704
2705 /*************************************************
2706 * Insert an automatic callout point *
2707 *************************************************/
2708
2709 /* This function is called when the PCRE_AUTO_CALLOUT option is set, to insert
2710 callout points before each pattern item.
2711
2712 Arguments:
2713 code current code pointer
2714 ptr current pattern pointer
2715 cd pointers to tables etc
2716
2717 Returns: new code pointer
2718 */
2719
2720 static pcre_uchar *
2721 auto_callout(pcre_uchar *code, const pcre_uchar *ptr, compile_data *cd)
2722 {
2723 *code++ = OP_CALLOUT;
2724 *code++ = 255;
2725 PUT(code, 0, (int)(ptr - cd->start_pattern)); /* Pattern offset */
2726 PUT(code, LINK_SIZE, 0); /* Default length */
2727 return code + 2 * LINK_SIZE;
2728 }
2729
2730
2731
2732 /*************************************************
2733 * Complete a callout item *
2734 *************************************************/
2735
2736 /* A callout item contains the length of the next item in the pattern, which
2737 we can't fill in till after we have reached the relevant point. This is used
2738 for both automatic and manual callouts.
2739
2740 Arguments:
2741 previous_callout points to previous callout item
2742 ptr current pattern pointer
2743 cd pointers to tables etc
2744
2745 Returns: nothing
2746 */
2747
2748 static void
2749 complete_callout(pcre_uchar *previous_callout, const pcre_uchar *ptr, compile_data *cd)
2750 {
2751 int length = (int)(ptr - cd->start_pattern - GET(previous_callout, 2));
2752 PUT(previous_callout, 2 + LINK_SIZE, length);
2753 }
2754
2755
2756
2757 #ifdef SUPPORT_UCP
2758 /*************************************************
2759 * Get othercase range *
2760 *************************************************/
2761
2762 /* This function is passed the start and end of a class range, in UTF-8 mode
2763 with UCP support. It searches up the characters, looking for internal ranges of
2764 characters in the "other" case. Each call returns the next one, updating the
2765 start address.
2766
2767 Arguments:
2768 cptr points to starting character value; updated
2769 d end value
2770 ocptr where to put start of othercase range
2771 odptr where to put end of othercase range
2772
2773 Yield: TRUE when range returned; FALSE when no more
2774 */
2775
2776 static BOOL
2777 get_othercase_range(unsigned int *cptr, unsigned int d, unsigned int *ocptr,
2778 unsigned int *odptr)
2779 {
2780 unsigned int c, othercase, next;
2781
2782 for (c = *cptr; c <= d; c++)
2783 { if ((othercase = UCD_OTHERCASE(c)) != c) break; }
2784
2785 if (c > d) return FALSE;
2786
2787 *ocptr = othercase;
2788 next = othercase + 1;
2789
2790 for (++c; c <= d; c++)
2791 {
2792 if (UCD_OTHERCASE(c) != next) break;
2793 next++;
2794 }
2795
2796 *odptr = next - 1;
2797 *cptr = c;
2798
2799 return TRUE;
2800 }
2801
2802
2803
2804 /*************************************************
2805 * Check a character and a property *
2806 *************************************************/
2807
2808 /* This function is called by check_auto_possessive() when a property item
2809 is adjacent to a fixed character.
2810
2811 Arguments:
2812 c the character
2813 ptype the property type
2814 pdata the data for the type
2815 negated TRUE if it's a negated property (\P or \p{^)
2816
2817 Returns: TRUE if auto-possessifying is OK
2818 */
2819
2820 static BOOL
2821 check_char_prop(int c, int ptype, int pdata, BOOL negated)
2822 {
2823 const ucd_record *prop = GET_UCD(c);
2824 switch(ptype)
2825 {
2826 case PT_LAMP:
2827 return (prop->chartype == ucp_Lu ||
2828 prop->chartype == ucp_Ll ||
2829 prop->chartype == ucp_Lt) == negated;
2830
2831 case PT_GC:
2832 return (pdata == PRIV(ucp_gentype)[prop->chartype]) == negated;
2833
2834 case PT_PC:
2835 return (pdata == prop->chartype) == negated;
2836
2837 case PT_SC:
2838 return (pdata == prop->script) == negated;
2839
2840 /* These are specials */
2841
2842 case PT_ALNUM:
2843 return (PRIV(ucp_gentype)[prop->chartype] == ucp_L ||
2844 PRIV(ucp_gentype)[prop->chartype] == ucp_N) == negated;
2845
2846 case PT_SPACE: /* Perl space */
2847 return (PRIV(ucp_gentype)[prop->chartype] == ucp_Z ||
2848 c == CHAR_HT || c == CHAR_NL || c == CHAR_FF || c == CHAR_CR)
2849 == negated;
2850
2851 case PT_PXSPACE: /* POSIX space */
2852 return (PRIV(ucp_gentype)[prop->chartype] == ucp_Z ||
2853 c == CHAR_HT || c == CHAR_NL || c == CHAR_VT ||
2854 c == CHAR_FF || c == CHAR_CR)
2855 == negated;
2856
2857 case PT_WORD:
2858 return (PRIV(ucp_gentype)[prop->chartype] == ucp_L ||
2859 PRIV(ucp_gentype)[prop->chartype] == ucp_N ||
2860 c == CHAR_UNDERSCORE) == negated;
2861 }
2862 return FALSE;
2863 }
2864 #endif /* SUPPORT_UCP */
2865
2866
2867
2868 /*************************************************
2869 * Check if auto-possessifying is possible *
2870 *************************************************/
2871
2872 /* This function is called for unlimited repeats of certain items, to see
2873 whether the next thing could possibly match the repeated item. If not, it makes
2874 sense to automatically possessify the repeated item.
2875
2876 Arguments:
2877 previous pointer to the repeated opcode
2878 utf8 TRUE in UTF-8 mode
2879 ptr next character in pattern
2880 options options bits
2881 cd contains pointers to tables etc.
2882
2883 Returns: TRUE if possessifying is wanted
2884 */
2885
2886 static BOOL
2887 check_auto_possessive(const pcre_uchar *previous, BOOL utf8,
2888 const pcre_uchar *ptr, int options, compile_data *cd)
2889 {
2890 int c, next;
2891 int op_code = *previous++;
2892
2893 /* Skip whitespace and comments in extended mode */
2894
2895 if ((options & PCRE_EXTENDED) != 0)
2896 {
2897 for (;;)
2898 {
2899 while ((cd->ctypes[*ptr] & ctype_space) != 0) ptr++;
2900 if (*ptr == CHAR_NUMBER_SIGN)
2901 {
2902 ptr++;
2903 while (*ptr != 0)
2904 {
2905 if (IS_NEWLINE(ptr)) { ptr += cd->nllen; break; }
2906 ptr++;
2907 #ifdef SUPPORT_UTF8
2908 if (utf8) while ((*ptr & 0xc0) == 0x80) ptr++;
2909 #endif
2910 }
2911 }
2912 else break;
2913 }
2914 }
2915
2916 /* If the next item is one that we can handle, get its value. A non-negative
2917 value is a character, a negative value is an escape value. */
2918
2919 if (*ptr == CHAR_BACKSLASH)
2920 {
2921 int temperrorcode = 0;
2922 next = check_escape(&ptr, &temperrorcode, cd->bracount, options, FALSE);
2923 if (temperrorcode != 0) return FALSE;
2924 ptr++; /* Point after the escape sequence */
2925 }
2926
2927 else if ((cd->ctypes[*ptr] & ctype_meta) == 0)
2928 {
2929 #ifdef SUPPORT_UTF8
2930 if (utf8) { GETCHARINC(next, ptr); } else
2931 #endif
2932 next = *ptr++;
2933 }
2934
2935 else return FALSE;
2936
2937 /* Skip whitespace and comments in extended mode */
2938
2939 if ((options & PCRE_EXTENDED) != 0)
2940 {
2941 for (;;)
2942 {
2943 while ((cd->ctypes[*ptr] & ctype_space) != 0) ptr++;
2944 if (*ptr == CHAR_NUMBER_SIGN)
2945 {
2946 ptr++;
2947 while (*ptr != 0)
2948 {
2949 if (IS_NEWLINE(ptr)) { ptr += cd->nllen; break; }
2950 ptr++;
2951 #ifdef SUPPORT_UTF8
2952 if (utf8) while ((*ptr & 0xc0) == 0x80) ptr++;
2953 #endif
2954 }
2955 }
2956 else break;
2957 }
2958 }
2959
2960 /* If the next thing is itself optional, we have to give up. */
2961
2962 if (*ptr == CHAR_ASTERISK || *ptr == CHAR_QUESTION_MARK ||
2963 STRNCMP_UC_C8(ptr, STR_LEFT_CURLY_BRACKET STR_0 STR_COMMA, 3) == 0)
2964 return FALSE;
2965
2966 /* Now compare the next item with the previous opcode. First, handle cases when
2967 the next item is a character. */
2968
2969 if (next >= 0) switch(op_code)
2970 {
2971 case OP_CHAR:
2972 #ifdef SUPPORT_UTF8
2973 GETCHARTEST(c, previous);
2974 #else
2975 c = *previous;
2976 #endif
2977 return c != next;
2978
2979 /* For CHARI (caseless character) we must check the other case. If we have
2980 Unicode property support, we can use it to test the other case of
2981 high-valued characters. */
2982
2983 case OP_CHARI:
2984 #ifdef SUPPORT_UTF8
2985 GETCHARTEST(c, previous);
2986 #else
2987 c = *previous;
2988 #endif
2989 if (c == next) return FALSE;
2990 #ifdef SUPPORT_UTF8
2991 if (utf8)
2992 {
2993 unsigned int othercase;
2994 if (next < 128) othercase = cd->fcc[next]; else
2995 #ifdef SUPPORT_UCP
2996 othercase = UCD_OTHERCASE((unsigned int)next);
2997 #else
2998 othercase = NOTACHAR;
2999 #endif
3000 return (unsigned int)c != othercase;
3001 }
3002 else
3003 #endif /* SUPPORT_UTF8 */
3004 return (c != cd->fcc[next]); /* Non-UTF-8 mode */
3005
3006 /* For OP_NOT and OP_NOTI, the data is always a single-byte character. These
3007 opcodes are not used for multi-byte characters, because they are coded using
3008 an XCLASS instead. */
3009
3010 case OP_NOT:
3011 return (c = *previous) == next;
3012
3013 case OP_NOTI:
3014 if ((c = *previous) == next) return TRUE;
3015 #ifdef SUPPORT_UTF8
3016 if (utf8)
3017 {
3018 unsigned int othercase;
3019 if (next < 128) othercase = cd->fcc[next]; else
3020 #ifdef SUPPORT_UCP
3021 othercase = UCD_OTHERCASE(next);
3022 #else
3023 othercase = NOTACHAR;
3024 #endif
3025 return (unsigned int)c == othercase;
3026 }
3027 else
3028 #endif /* SUPPORT_UTF8 */
3029 return (c == cd->fcc[next]); /* Non-UTF-8 mode */
3030
3031 /* Note that OP_DIGIT etc. are generated only when PCRE_UCP is *not* set.
3032 When it is set, \d etc. are converted into OP_(NOT_)PROP codes. */
3033
3034 case OP_DIGIT:
3035 return next > 127 || (cd->ctypes[next] & ctype_digit) == 0;
3036
3037 case OP_NOT_DIGIT:
3038 return next <= 127 && (cd->ctypes[next] & ctype_digit) != 0;
3039
3040 case OP_WHITESPACE:
3041 return next > 127 || (cd->ctypes[next] & ctype_space) == 0;
3042
3043 case OP_NOT_WHITESPACE:
3044 return next <= 127 && (cd->ctypes[next] & ctype_space) != 0;
3045
3046 case OP_WORDCHAR:
3047 return next > 127 || (cd->ctypes[next] & ctype_word) == 0;
3048
3049 case OP_NOT_WORDCHAR:
3050 return next <= 127 && (cd->ctypes[next] & ctype_word) != 0;
3051
3052 case OP_HSPACE:
3053 case OP_NOT_HSPACE:
3054 switch(next)
3055 {
3056 case 0x09:
3057 case 0x20:
3058 case 0xa0:
3059 case 0x1680:
3060 case 0x180e:
3061 case 0x2000:
3062 case 0x2001:
3063 case 0x2002:
3064 case 0x2003:
3065 case 0x2004:
3066 case 0x2005:
3067 case 0x2006:
3068 case 0x2007:
3069 case 0x2008:
3070 case 0x2009:
3071 case 0x200A:
3072 case 0x202f:
3073 case 0x205f:
3074 case 0x3000:
3075 return op_code == OP_NOT_HSPACE;
3076 default:
3077 return op_code != OP_NOT_HSPACE;
3078 }
3079
3080 case OP_ANYNL:
3081 case OP_VSPACE:
3082 case OP_NOT_VSPACE:
3083 switch(next)
3084 {
3085 case 0x0a:
3086 case 0x0b:
3087 case 0x0c:
3088 case 0x0d:
3089 case 0x85:
3090 case 0x2028:
3091 case 0x2029:
3092 return op_code == OP_NOT_VSPACE;
3093 default:
3094 return op_code != OP_NOT_VSPACE;
3095 }
3096
3097 #ifdef SUPPORT_UCP
3098 case OP_PROP:
3099 return check_char_prop(next, previous[0], previous[1], FALSE);
3100
3101 case OP_NOTPROP:
3102 return check_char_prop(next, previous[0], previous[1], TRUE);
3103 #endif
3104
3105 default:
3106 return FALSE;
3107 }
3108
3109
3110 /* Handle the case when the next item is \d, \s, etc. Note that when PCRE_UCP
3111 is set, \d turns into ESC_du rather than ESC_d, etc., so ESC_d etc. are
3112 generated only when PCRE_UCP is *not* set, that is, when only ASCII
3113 characteristics are recognized. Similarly, the opcodes OP_DIGIT etc. are
3114 replaced by OP_PROP codes when PCRE_UCP is set. */
3115
3116 switch(op_code)
3117 {
3118 case OP_CHAR:
3119 case OP_CHARI:
3120 #ifdef SUPPORT_UTF8
3121 GETCHARTEST(c, previous);
3122 #else
3123 c = *previous;
3124 #endif
3125 switch(-next)
3126 {
3127 case ESC_d:
3128 return c > 127 || (cd->ctypes[c] & ctype_digit) == 0;
3129
3130 case ESC_D:
3131 return c <= 127 && (cd->ctypes[c] & ctype_digit) != 0;
3132
3133 case ESC_s:
3134 return c > 127 || (cd->ctypes[c] & ctype_space) == 0;
3135
3136 case ESC_S:
3137 return c <= 127 && (cd->ctypes[c] & ctype_space) != 0;
3138
3139 case ESC_w:
3140 return c > 127 || (cd->ctypes[c] & ctype_word) == 0;
3141
3142 case ESC_W:
3143 return c <= 127 && (cd->ctypes[c] & ctype_word) != 0;
3144
3145 case ESC_h:
3146 case ESC_H:
3147 switch(c)
3148 {
3149 case 0x09:
3150 case 0x20:
3151 case 0xa0:
3152 case 0x1680:
3153 case 0x180e:
3154 case 0x2000:
3155 case 0x2001:
3156 case 0x2002:
3157 case 0x2003:
3158 case 0x2004:
3159 case 0x2005:
3160 case 0x2006:
3161 case 0x2007:
3162 case 0x2008:
3163 case 0x2009:
3164 case 0x200A:
3165 case 0x202f:
3166 case 0x205f:
3167 case 0x3000:
3168 return -next != ESC_h;
3169 default:
3170 return -next == ESC_h;
3171 }
3172
3173 case ESC_v:
3174 case ESC_V:
3175 switch(c)
3176 {
3177 case 0x0a:
3178 case 0x0b:
3179 case 0x0c:
3180 case 0x0d:
3181 case 0x85:
3182 case 0x2028:
3183 case 0x2029:
3184 return -next != ESC_v;
3185 default:
3186 return -next == ESC_v;
3187 }
3188
3189 /* When PCRE_UCP is set, these values get generated for \d etc. Find
3190 their substitutions and process them. The result will always be either
3191 -ESC_p or -ESC_P. Then fall through to process those values. */
3192
3193 #ifdef SUPPORT_UCP
3194 case ESC_du:
3195 case ESC_DU:
3196 case ESC_wu:
3197 case ESC_WU:
3198 case ESC_su:
3199 case ESC_SU:
3200 {
3201 int temperrorcode = 0;
3202 ptr = substitutes[-next - ESC_DU];
3203 next = check_escape(&ptr, &temperrorcode, 0, options, FALSE);
3204 if (temperrorcode != 0) return FALSE;
3205 ptr++; /* For compatibility */
3206 }
3207 /* Fall through */
3208
3209 case ESC_p:
3210 case ESC_P:
3211 {
3212 int ptype, pdata, errorcodeptr;
3213 BOOL negated;
3214
3215 ptr--; /* Make ptr point at the p or P */
3216 ptype = get_ucp(&ptr, &negated, &pdata, &errorcodeptr);
3217 if (ptype < 0) return FALSE;
3218 ptr++; /* Point past the final curly ket */
3219
3220 /* If the property item is optional, we have to give up. (When generated
3221 from \d etc by PCRE_UCP, this test will have been applied much earlier,
3222 to the original \d etc. At this point, ptr will point to a zero byte. */
3223
3224 if (*ptr == CHAR_ASTERISK || *ptr == CHAR_QUESTION_MARK ||
3225 STRNCMP_UC_C8(ptr, STR_LEFT_CURLY_BRACKET STR_0 STR_COMMA, 3) == 0)
3226 return FALSE;
3227
3228 /* Do the property check. */
3229
3230 return check_char_prop(c, ptype, pdata, (next == -ESC_P) != negated);
3231 }
3232 #endif
3233
3234 default:
3235 return FALSE;
3236 }
3237
3238 /* In principle, support for Unicode properties should be integrated here as
3239 well. It means re-organizing the above code so as to get hold of the property
3240 values before switching on the op-code. However, I wonder how many patterns
3241 combine ASCII \d etc with Unicode properties? (Note that if PCRE_UCP is set,
3242 these op-codes are never generated.) */
3243
3244 case OP_DIGIT:
3245 return next == -ESC_D || next == -ESC_s || next == -ESC_W ||
3246 next == -ESC_h || next == -ESC_v || next == -ESC_R;
3247
3248 case OP_NOT_DIGIT:
3249 return next == -ESC_d;
3250
3251 case OP_WHITESPACE:
3252 return next == -ESC_S || next == -ESC_d || next == -ESC_w || next == -ESC_R;
3253
3254 case OP_NOT_WHITESPACE:
3255 return next == -ESC_s || next == -ESC_h || next == -ESC_v;
3256
3257 case OP_HSPACE:
3258 return next == -ESC_S || next == -ESC_H || next == -ESC_d ||
3259 next == -ESC_w || next == -ESC_v || next == -ESC_R;
3260
3261 case OP_NOT_HSPACE:
3262 return next == -ESC_h;
3263
3264 /* Can't have \S in here because VT matches \S (Perl anomaly) */
3265 case OP_ANYNL:
3266 case OP_VSPACE:
3267 return next == -ESC_V || next == -ESC_d || next == -ESC_w;
3268
3269 case OP_NOT_VSPACE:
3270 return next == -ESC_v || next == -ESC_R;
3271
3272 case OP_WORDCHAR:
3273 return next == -ESC_W || next == -ESC_s || next == -ESC_h ||
3274 next == -ESC_v || next == -ESC_R;
3275
3276 case OP_NOT_WORDCHAR:
3277 return next == -ESC_w || next == -ESC_d;
3278
3279 default:
3280 return FALSE;
3281 }
3282
3283 /* Control does not reach here */
3284 }
3285
3286
3287
3288 /*************************************************
3289 * Compile one branch *
3290 *************************************************/
3291
3292 /* Scan the pattern, compiling it into the a vector. If the options are
3293 changed during the branch, the pointer is used to change the external options
3294 bits. This function is used during the pre-compile phase when we are trying
3295 to find out the amount of memory needed, as well as during the real compile
3296 phase. The value of lengthptr distinguishes the two phases.
3297
3298 Arguments:
3299 optionsptr pointer to the option bits
3300 codeptr points to the pointer to the current code point
3301 ptrptr points to the current pattern pointer
3302 errorcodeptr points to error code variable
3303 firstcharptr set to initial literal character, or < 0 (REQ_UNSET, REQ_NONE)
3304 reqcharptr set to the last literal character required, else < 0
3305 bcptr points to current branch chain
3306 cond_depth conditional nesting depth
3307 cd contains pointers to tables etc.
3308 lengthptr NULL during the real compile phase
3309 points to length accumulator during pre-compile phase
3310
3311 Returns: TRUE on success
3312 FALSE, with *errorcodeptr set non-zero on error
3313 */
3314
3315 static BOOL
3316 compile_branch(int *optionsptr, pcre_uchar **codeptr,
3317 const pcre_uchar **ptrptr, int *errorcodeptr, pcre_int32 *firstcharptr,
3318 pcre_int32 *reqcharptr, branch_chain *bcptr, int cond_depth,
3319 compile_data *cd, int *lengthptr)
3320 {
3321 int repeat_type, op_type;
3322 int repeat_min = 0, repeat_max = 0; /* To please picky compilers */
3323 int bravalue = 0;
3324 int greedy_default, greedy_non_default;
3325 pcre_int32 firstchar, reqchar;
3326 pcre_int32 zeroreqchar, zerofirstchar;
3327 pcre_int32 req_caseopt, reqvary, tempreqvary;
3328 int options = *optionsptr; /* May change dynamically */
3329 int after_manual_callout = 0;
3330 int length_prevgroup = 0;
3331 register int c;
3332 register pcre_uchar *code = *codeptr;
3333 pcre_uchar *last_code = code;
3334 pcre_uchar *orig_code = code;
3335 pcre_uchar *tempcode;
3336 BOOL inescq = FALSE;
3337 BOOL groupsetfirstchar = FALSE;
3338 const pcre_uchar *ptr = *ptrptr;
3339 const pcre_uchar *tempptr;
3340 const pcre_uchar *nestptr = NULL;
3341 pcre_uchar *previous = NULL;
3342 pcre_uchar *previous_callout = NULL;
3343 pcre_uchar *save_hwm = NULL;
3344 pcre_uint8 classbits[32];
3345
3346 /* We can fish out the UTF-8 setting once and for all into a BOOL, but we
3347 must not do this for other options (e.g. PCRE_EXTENDED) because they may change
3348 dynamically as we process the pattern. */
3349
3350 #ifdef SUPPORT_UTF8
3351 BOOL utf8 = (options & PCRE_UTF8) != 0;
3352 pcre_uint8 utf8_char[6];
3353 #else
3354 BOOL utf8 = FALSE;
3355 #endif
3356
3357 /* Helper variables for OP_XCLASS opcode (for characters > 255). */
3358
3359 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
3360 BOOL xclass;
3361 pcre_uchar *class_uchardata;
3362 pcre_uchar *class_uchardata_base;
3363 #endif
3364
3365 #ifdef PCRE_DEBUG
3366 if (lengthptr != NULL) DPRINTF((">> start branch\n"));
3367 #endif
3368
3369 /* Set up the default and non-default settings for greediness */
3370
3371 greedy_default = ((options & PCRE_UNGREEDY) != 0);
3372 greedy_non_default = greedy_default ^ 1;
3373
3374 /* Initialize no first byte, no required byte. REQ_UNSET means "no char
3375 matching encountered yet". It gets changed to REQ_NONE if we hit something that
3376 matches a non-fixed char first char; reqchar just remains unset if we never
3377 find one.
3378
3379 When we hit a repeat whose minimum is zero, we may have to adjust these values
3380 to take the zero repeat into account. This is implemented by setting them to
3381 zerofirstbyte and zeroreqchar when such a repeat is encountered. The individual
3382 item types that can be repeated set these backoff variables appropriately. */
3383
3384 firstchar = reqchar = zerofirstchar = zeroreqchar = REQ_UNSET;
3385
3386 /* The variable req_caseopt contains either the REQ_CASELESS value
3387 or zero, according to the current setting of the caseless flag. The
3388 REQ_CASELESS leaves the lower 28 bit empty. It is added into the
3389 firstchar or reqchar variables to record the case status of the
3390 value. This is used only for ASCII characters. */
3391
3392 req_caseopt = ((options & PCRE_CASELESS) != 0)? REQ_CASELESS:0;
3393
3394 /* Switch on next character until the end of the branch */
3395
3396 for (;; ptr++)
3397 {
3398 BOOL negate_class;
3399 BOOL should_flip_negation;
3400 BOOL possessive_quantifier;
3401 BOOL is_quantifier;
3402 BOOL is_recurse;
3403 BOOL reset_bracount;
3404 int class_charcount;
3405 int class_lastchar;
3406 int newoptions;
3407 int recno;
3408 int refsign;
3409 int skipbytes;
3410 int subreqchar;
3411 int subfirstchar;
3412 int terminator;
3413 int mclength;
3414 int tempbracount;
3415 pcre_uchar mcbuffer[8];
3416
3417 /* Get next byte in the pattern */
3418
3419 c = *ptr;
3420
3421 /* If we are at the end of a nested substitution, revert to the outer level
3422 string. Nesting only happens one level deep. */
3423
3424 if (c == 0 && nestptr != NULL)
3425 {
3426 ptr = nestptr;
3427 nestptr = NULL;
3428 c = *ptr;
3429 }
3430
3431 /* If we are in the pre-compile phase, accumulate the length used for the
3432 previous cycle of this loop. */
3433
3434 if (lengthptr != NULL)
3435 {
3436 #ifdef PCRE_DEBUG
3437 if (code > cd->hwm) cd->hwm = code; /* High water info */
3438 #endif
3439 if (code > cd->start_workspace + WORK_SIZE_CHECK) /* Check for overrun */
3440 {
3441 *errorcodeptr = ERR52;
3442 goto FAILED;
3443 }
3444
3445 /* There is at least one situation where code goes backwards: this is the
3446 case of a zero quantifier after a class (e.g. [ab]{0}). At compile time,
3447 the class is simply eliminated. However, it is created first, so we have to
3448 allow memory for it. Therefore, don't ever reduce the length at this point.
3449 */
3450
3451 if (code < last_code) code = last_code;
3452
3453 /* Paranoid check for integer overflow */
3454
3455 if (OFLOW_MAX - *lengthptr < code - last_code)
3456 {
3457 *errorcodeptr = ERR20;
3458 goto FAILED;
3459 }
3460
3461 *lengthptr += (int)(code - last_code);
3462 DPRINTF(("length=%d added %d c=%c\n", *lengthptr, (int)(code - last_code),
3463 c));
3464
3465 /* If "previous" is set and it is not at the start of the work space, move
3466 it back to there, in order to avoid filling up the work space. Otherwise,
3467 if "previous" is NULL, reset the current code pointer to the start. */
3468
3469 if (previous != NULL)
3470 {
3471 if (previous > orig_code)
3472 {
3473 memmove(orig_code, previous, IN_UCHARS(code - previous));
3474 code -= previous - orig_code;
3475 previous = orig_code;
3476 }
3477 }
3478 else code = orig_code;
3479
3480 /* Remember where this code item starts so we can pick up the length
3481 next time round. */
3482
3483 last_code = code;
3484 }
3485
3486 /* In the real compile phase, just check the workspace used by the forward
3487 reference list. */
3488
3489 else if (cd->hwm > cd->start_workspace + WORK_SIZE_CHECK)
3490 {
3491 *errorcodeptr = ERR52;
3492 goto FAILED;
3493 }
3494
3495 /* If in \Q...\E, check for the end; if not, we have a literal */
3496
3497 if (inescq && c != 0)
3498 {
3499 if (c == CHAR_BACKSLASH && ptr[1] == CHAR_E)
3500 {
3501 inescq = FALSE;
3502 ptr++;
3503 continue;
3504 }
3505 else
3506 {
3507 if (previous_callout != NULL)
3508 {
3509 if (lengthptr == NULL) /* Don't attempt in pre-compile phase */
3510 complete_callout(previous_callout, ptr, cd);
3511 previous_callout = NULL;
3512 }
3513 if ((options & PCRE_AUTO_CALLOUT) != 0)
3514 {
3515 previous_callout = code;
3516 code = auto_callout(code, ptr, cd);
3517 }
3518 goto NORMAL_CHAR;
3519 }
3520 }
3521
3522 /* Fill in length of a previous callout, except when the next thing is
3523 a quantifier. */
3524
3525 is_quantifier =
3526 c == CHAR_ASTERISK || c == CHAR_PLUS || c == CHAR_QUESTION_MARK ||
3527 (c == CHAR_LEFT_CURLY_BRACKET && is_counted_repeat(ptr+1));
3528
3529 if (!is_quantifier && previous_callout != NULL &&
3530 after_manual_callout-- <= 0)
3531 {
3532 if (lengthptr == NULL) /* Don't attempt in pre-compile phase */
3533 complete_callout(previous_callout, ptr, cd);
3534 previous_callout = NULL;
3535 }
3536
3537 /* In extended mode, skip white space and comments. */
3538
3539 if ((options & PCRE_EXTENDED) != 0)
3540 {
3541 if ((cd->ctypes[c] & ctype_space) != 0) continue;
3542 if (c == CHAR_NUMBER_SIGN)
3543 {
3544 ptr++;
3545 while (*ptr != 0)
3546 {
3547 if (IS_NEWLINE(ptr)) { ptr += cd->nllen - 1; break; }
3548 ptr++;
3549 #ifdef SUPPORT_UTF8
3550 if (utf8) while ((*ptr & 0xc0) == 0x80) ptr++;
3551 #endif
3552 }
3553 if (*ptr != 0) continue;
3554
3555 /* Else fall through to handle end of string */
3556 c = 0;
3557 }
3558 }
3559
3560 /* No auto callout for quantifiers. */
3561
3562 if ((options & PCRE_AUTO_CALLOUT) != 0 && !is_quantifier)
3563 {
3564 previous_callout = code;
3565 code = auto_callout(code, ptr, cd);
3566 }
3567
3568 switch(c)
3569 {
3570 /* ===================================================================*/
3571 case 0: /* The branch terminates at string end */
3572 case CHAR_VERTICAL_LINE: /* or | or ) */
3573 case CHAR_RIGHT_PARENTHESIS:
3574 *firstcharptr = firstchar;
3575 *reqcharptr = reqchar;
3576 *codeptr = code;
3577 *ptrptr = ptr;
3578 if (lengthptr != NULL)
3579 {
3580 if (OFLOW_MAX - *lengthptr < code - last_code)
3581 {
3582 *errorcodeptr = ERR20;
3583 goto FAILED;
3584 }
3585 *lengthptr += (int)(code - last_code); /* To include callout length */
3586 DPRINTF((">> end branch\n"));
3587 }
3588 return TRUE;
3589
3590
3591 /* ===================================================================*/
3592 /* Handle single-character metacharacters. In multiline mode, ^ disables
3593 the setting of any following char as a first character. */
3594
3595 case CHAR_CIRCUMFLEX_ACCENT:
3596 previous = NULL;
3597 if ((options & PCRE_MULTILINE) != 0)
3598 {
3599 if (firstchar == REQ_UNSET) firstchar = REQ_NONE;
3600 *code++ = OP_CIRCM;
3601 }
3602 else *code++ = OP_CIRC;
3603 break;
3604
3605 case CHAR_DOLLAR_SIGN:
3606 previous = NULL;
3607 *code++ = ((options & PCRE_MULTILINE) != 0)? OP_DOLLM : OP_DOLL;
3608 break;
3609
3610 /* There can never be a first char if '.' is first, whatever happens about
3611 repeats. The value of reqchar doesn't change either. */
3612
3613 case CHAR_DOT:
3614 if (firstchar == REQ_UNSET) firstchar = REQ_NONE;
3615 zerofirstchar = firstchar;
3616 zeroreqchar = reqchar;
3617 previous = code;
3618 *code++ = ((options & PCRE_DOTALL) != 0)? OP_ALLANY: OP_ANY;
3619 break;
3620
3621
3622 /* ===================================================================*/
3623 /* Character classes. If the included characters are all < 256, we build a
3624 32-byte bitmap of the permitted characters, except in the special case
3625 where there is only one such character. For negated classes, we build the
3626 map as usual, then invert it at the end. However, we use a different opcode
3627 so that data characters > 255 can be handled correctly.
3628
3629 If the class contains characters outside the 0-255 range, a different
3630 opcode is compiled. It may optionally have a bit map for characters < 256,
3631 but those above are are explicitly listed afterwards. A flag byte tells
3632 whether the bitmap is present, and whether this is a negated class or not.
3633
3634 In JavaScript compatibility mode, an isolated ']' causes an error. In
3635 default (Perl) mode, it is treated as a data character. */
3636
3637 case CHAR_RIGHT_SQUARE_BRACKET:
3638 if ((cd->external_options & PCRE_JAVASCRIPT_COMPAT) != 0)
3639 {
3640 *errorcodeptr = ERR64;
3641 goto FAILED;
3642 }
3643 goto NORMAL_CHAR;
3644
3645 case CHAR_LEFT_SQUARE_BRACKET:
3646 previous = code;
3647
3648 /* PCRE supports POSIX class stuff inside a class. Perl gives an error if
3649 they are encountered at the top level, so we'll do that too. */
3650
3651 if ((ptr[1] == CHAR_COLON || ptr[1] == CHAR_DOT ||
3652 ptr[1] == CHAR_EQUALS_SIGN) &&
3653 check_posix_syntax(ptr, &tempptr))
3654 {
3655 *errorcodeptr = (ptr[1] == CHAR_COLON)? ERR13 : ERR31;
3656 goto FAILED;
3657 }
3658
3659 /* If the first character is '^', set the negation flag and skip it. Also,
3660 if the first few characters (either before or after ^) are \Q\E or \E we
3661 skip them too. This makes for compatibility with Perl. */
3662
3663 negate_class = FALSE;
3664 for (;;)
3665 {
3666 c = *(++ptr);
3667 if (c == CHAR_BACKSLASH)
3668 {
3669 if (ptr[1] == CHAR_E)
3670 ptr++;
3671 else if (STRNCMP_UC_C8(ptr + 1, STR_Q STR_BACKSLASH STR_E, 3) == 0)
3672 ptr += 3;
3673 else
3674 break;
3675 }
3676 else if (!negate_class && c == CHAR_CIRCUMFLEX_ACCENT)
3677 negate_class = TRUE;
3678 else break;
3679 }
3680
3681 /* Empty classes are allowed in JavaScript compatibility mode. Otherwise,
3682 an initial ']' is taken as a data character -- the code below handles
3683 that. In JS mode, [] must always fail, so generate OP_FAIL, whereas
3684 [^] must match any character, so generate OP_ALLANY. */
3685
3686 if (c == CHAR_RIGHT_SQUARE_BRACKET &&
3687 (cd->external_options & PCRE_JAVASCRIPT_COMPAT) != 0)
3688 {
3689 *code++ = negate_class? OP_ALLANY : OP_FAIL;
3690 if (firstchar == REQ_UNSET) firstchar = REQ_NONE;
3691 zerofirstchar = firstchar;
3692 break;
3693 }
3694
3695 /* If a class contains a negative special such as \S, we need to flip the
3696 negation flag at the end, so that support for characters > 255 works
3697 correctly (they are all included in the class). */
3698
3699 should_flip_negation = FALSE;
3700
3701 /* Keep a count of chars with values < 256 so that we can optimize the case
3702 of just a single character (as long as it's < 256). However, For higher
3703 valued UTF-8 characters, we don't yet do any optimization. */
3704
3705 class_charcount = 0;
3706 class_lastchar = -1;
3707
3708 /* Initialize the 32-char bit map to all zeros. We build the map in a
3709 temporary bit of memory, in case the class contains only 1 character (less
3710 than 256), because in that case the compiled code doesn't use the bit map.
3711 */
3712
3713 memset(classbits, 0, 32 * sizeof(pcre_uint8));
3714
3715 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
3716 xclass = FALSE; /* No chars >= 256 */
3717 class_uchardata = code + LINK_SIZE + 2; /* For UTF-8 items */
3718 class_uchardata_base = class_uchardata; /* For resetting in pass 1 */
3719 #endif
3720
3721 /* Process characters until ] is reached. By writing this as a "do" it
3722 means that an initial ] is taken as a data character. At the start of the
3723 loop, c contains the first byte of the character. */
3724
3725 if (c != 0) do
3726 {
3727 const pcre_uchar *oldptr;
3728
3729 #ifdef SUPPORT_UTF8
3730 if (utf8 && c > 127)
3731 { /* Braces are required because the */
3732 GETCHARLEN(c, ptr, ptr); /* macro generates multiple statements */
3733 }
3734 #endif
3735
3736 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
3737 /* In the pre-compile phase, accumulate the length of any extra
3738 data and reset the pointer. This is so that very large classes that
3739 contain a zillion > 255 characters no longer overwrite the work space
3740 (which is on the stack). */
3741
3742 if (lengthptr != NULL)
3743 {
3744 *lengthptr += class_uchardata - class_uchardata_base;
3745 class_uchardata = class_uchardata_base;
3746 }
3747 #endif
3748
3749 /* Inside \Q...\E everything is literal except \E */
3750
3751 if (inescq)
3752 {
3753 if (c == CHAR_BACKSLASH && ptr[1] == CHAR_E) /* If we are at \E */
3754 {
3755 inescq = FALSE; /* Reset literal state */
3756 ptr++; /* Skip the 'E' */
3757 continue; /* Carry on with next */
3758 }
3759 goto CHECK_RANGE; /* Could be range if \E follows */
3760 }
3761
3762 /* Handle POSIX class names. Perl allows a negation extension of the
3763 form [:^name:]. A square bracket that doesn't match the syntax is
3764 treated as a literal. We also recognize the POSIX constructions
3765 [.ch.] and [=ch=] ("collating elements") and fault them, as Perl
3766 5.6 and 5.8 do. */
3767
3768 if (c == CHAR_LEFT_SQUARE_BRACKET &&
3769 (ptr[1] == CHAR_COLON || ptr[1] == CHAR_DOT ||
3770 ptr[1] == CHAR_EQUALS_SIGN) && check_posix_syntax(ptr, &tempptr))
3771 {
3772 BOOL local_negate = FALSE;
3773 int posix_class, taboffset, tabopt;
3774 register const pcre_uint8 *cbits = cd->cbits;
3775 pcre_uint8 pbits[32];
3776
3777 if (ptr[1] != CHAR_COLON)
3778 {
3779 *errorcodeptr = ERR31;
3780 goto FAILED;
3781 }
3782
3783 ptr += 2;
3784 if (*ptr == CHAR_CIRCUMFLEX_ACCENT)
3785 {
3786 local_negate = TRUE;
3787 should_flip_negation = TRUE; /* Note negative special */
3788 ptr++;
3789 }
3790
3791 posix_class = check_posix_name(ptr, (int)(tempptr - ptr));
3792 if (posix_class < 0)
3793 {
3794 *errorcodeptr = ERR30;
3795 goto FAILED;
3796 }
3797
3798 /* If matching is caseless, upper and lower are converted to
3799 alpha. This relies on the fact that the class table starts with
3800 alpha, lower, upper as the first 3 entries. */
3801
3802 if ((options & PCRE_CASELESS) != 0 && posix_class <= 2)
3803 posix_class = 0;
3804
3805 /* When PCRE_UCP is set, some of the POSIX classes are converted to
3806 different escape sequences that use Unicode properties. */
3807
3808 #ifdef SUPPORT_UCP
3809 if ((options & PCRE_UCP) != 0)
3810 {
3811 int pc = posix_class + ((local_negate)? POSIX_SUBSIZE/2 : 0);
3812 if (posix_substitutes[pc] != NULL)
3813 {
3814 nestptr = tempptr + 1;
3815 ptr = posix_substitutes[pc] - 1;
3816 continue;
3817 }
3818 }
3819 #endif
3820 /* In the non-UCP case, we build the bit map for the POSIX class in a
3821 chunk of local store because we may be adding and subtracting from it,
3822 and we don't want to subtract bits that may be in the main map already.
3823 At the end we or the result into the bit map that is being built. */
3824
3825 posix_class *= 3;
3826
3827 /* Copy in the first table (always present) */
3828
3829 memcpy(pbits, cbits + posix_class_maps[posix_class],
3830 32 * sizeof(pcre_uint8));
3831
3832 /* If there is a second table, add or remove it as required. */
3833
3834 taboffset = posix_class_maps[posix_class + 1];
3835 tabopt = posix_class_maps[posix_class + 2];
3836
3837 if (taboffset >= 0)
3838 {
3839 if (tabopt >= 0)
3840 for (c = 0; c < 32; c++) pbits[c] |= cbits[c + taboffset];
3841 else
3842 for (c = 0; c < 32; c++) pbits[c] &= ~cbits[c + taboffset];
3843 }
3844
3845 /* Not see if we need to remove any special characters. An option
3846 value of 1 removes vertical space and 2 removes underscore. */
3847
3848 if (tabopt < 0) tabopt = -tabopt;
3849 if (tabopt == 1) pbits[1] &= ~0x3c;
3850 else if (tabopt == 2) pbits[11] &= 0x7f;
3851
3852 /* Add the POSIX table or its complement into the main table that is
3853 being built and we are done. */
3854
3855 if (local_negate)
3856 for (c = 0; c < 32; c++) classbits[c] |= ~pbits[c];
3857 else
3858 for (c = 0; c < 32; c++) classbits[c] |= pbits[c];
3859
3860 ptr = tempptr + 1;
3861 class_charcount = 10; /* Set > 1; assumes more than 1 per class */
3862 continue; /* End of POSIX syntax handling */
3863 }
3864
3865 /* Backslash may introduce a single character, or it may introduce one
3866 of the specials, which just set a flag. The sequence \b is a special
3867 case. Inside a class (and only there) it is treated as backspace. We
3868 assume that other escapes have more than one character in them, so set
3869 class_charcount bigger than one. Unrecognized escapes fall through and
3870 are either treated as literal characters (by default), or are faulted if
3871 PCRE_EXTRA is set. */
3872
3873 if (c == CHAR_BACKSLASH)
3874 {
3875 c = check_escape(&ptr, errorcodeptr, cd->bracount, options, TRUE);
3876 if (*errorcodeptr != 0) goto FAILED;
3877
3878 if (-c == ESC_b) c = CHAR_BS; /* \b is backspace in a class */
3879 else if (-c == ESC_Q) /* Handle start of quoted string */
3880 {
3881 if (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
3882 {
3883 ptr += 2; /* avoid empty string */
3884 }
3885 else inescq = TRUE;
3886 continue;
3887 }
3888 else if (-c == ESC_E) continue; /* Ignore orphan \E */
3889
3890 if (c < 0)
3891 {
3892 register const pcre_uint8 *cbits = cd->cbits;
3893 class_charcount += 2; /* Greater than 1 is what matters */
3894
3895 switch (-c)
3896 {
3897 #ifdef SUPPORT_UCP
3898 case ESC_du: /* These are the values given for \d etc */
3899 case ESC_DU: /* when PCRE_UCP is set. We replace the */
3900 case ESC_wu: /* escape sequence with an appropriate \p */
3901 case ESC_WU: /* or \P to test Unicode properties instead */
3902 case ESC_su: /* of the default ASCII testing. */
3903 case ESC_SU:
3904 nestptr = ptr;
3905 ptr = substitutes[-c - ESC_DU] - 1; /* Just before substitute */
3906 class_charcount -= 2; /* Undo! */
3907 continue;
3908 #endif
3909 case ESC_d:
3910 for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_digit];
3911 continue;
3912
3913 case ESC_D:
3914 should_flip_negation = TRUE;
3915 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_digit];
3916 continue;
3917
3918 case ESC_w:
3919 for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_word];
3920 continue;
3921
3922 case ESC_W:
3923 should_flip_negation = TRUE;
3924 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_word];
3925 continue;
3926
3927 /* Perl 5.004 onwards omits VT from \s, but we must preserve it
3928 if it was previously set by something earlier in the character
3929 class. */
3930
3931 case ESC_s:
3932 classbits[0] |= cbits[cbit_space];
3933 classbits[1] |= cbits[cbit_space+1] & ~0x08;
3934 for (c = 2; c < 32; c++) classbits[c] |= cbits[c+cbit_space];
3935 continue;
3936
3937 case ESC_S:
3938 should_flip_negation = TRUE;
3939 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_space];
3940 classbits[1] |= 0x08; /* Perl 5.004 onwards omits VT from \s */
3941 continue;
3942
3943 case ESC_h:
3944 SETBIT(classbits, 0x09); /* VT */
3945 SETBIT(classbits, 0x20); /* SPACE */
3946 SETBIT(classbits, 0xa0); /* NSBP */
3947 #ifdef SUPPORT_UTF
3948 if (utf8)
3949 {
3950 xclass = TRUE;
3951 *class_uchardata++ = XCL_SINGLE;
3952 class_uchardata += PRIV(ord2utf8)(0x1680, class_uchardata);
3953 *class_uchardata++ = XCL_SINGLE;
3954 class_uchardata += PRIV(ord2utf8)(0x180e, class_uchardata);
3955 *class_uchardata++ = XCL_RANGE;
3956 class_uchardata += PRIV(ord2utf8)(0x2000, class_uchardata);
3957 class_uchardata += PRIV(ord2utf8)(0x200A, class_uchardata);
3958 *class_uchardata++ = XCL_SINGLE;
3959 class_uchardata += PRIV(ord2utf8)(0x202f, class_uchardata);
3960 *class_uchardata++ = XCL_SINGLE;
3961 class_uchardata += PRIV(ord2utf8)(0x205f, class_uchardata);
3962 *class_uchardata++ = XCL_SINGLE;
3963 class_uchardata += PRIV(ord2utf8)(0x3000, class_uchardata);
3964 }
3965 #endif
3966 continue;
3967
3968 case ESC_H:
3969 for (c = 0; c < 32; c++)
3970 {
3971 int x = 0xff;
3972 switch (c)
3973 {
3974 case 0x09/8: x ^= 1 << (0x09%8); break;
3975 case 0x20/8: x ^= 1 << (0x20%8); break;
3976 case 0xa0/8: x ^= 1 << (0xa0%8); break;
3977 default: break;
3978 }
3979 classbits[c] |= x;
3980 }
3981
3982 #ifdef SUPPORT_UTF
3983 if (utf8)
3984 {
3985 xclass = TRUE;
3986 *class_uchardata++ = XCL_RANGE;
3987 class_uchardata += PRIV(ord2utf8)(0x0100, class_uchardata);
3988 class_uchardata += PRIV(ord2utf8)(0x167f, class_uchardata);
3989 *class_uchardata++ = XCL_RANGE;
3990 class_uchardata += PRIV(ord2utf8)(0x1681, class_uchardata);
3991 class_uchardata += PRIV(ord2utf8)(0x180d, class_uchardata);
3992 *class_uchardata++ = XCL_RANGE;
3993 class_uchardata += PRIV(ord2utf8)(0x180f, class_uchardata);
3994 class_uchardata += PRIV(ord2utf8)(0x1fff, class_uchardata);
3995 *class_uchardata++ = XCL_RANGE;
3996 class_uchardata += PRIV(ord2utf8)(0x200B, class_uchardata);
3997 class_uchardata += PRIV(ord2utf8)(0x202e, class_uchardata);
3998 *class_uchardata++ = XCL_RANGE;
3999 class_uchardata += PRIV(ord2utf8)(0x2030, class_uchardata);
4000 class_uchardata += PRIV(ord2utf8)(0x205e, class_uchardata);
4001 *class_uchardata++ = XCL_RANGE;
4002 class_uchardata += PRIV(ord2utf8)(0x2060, class_uchardata);
4003 class_uchardata += PRIV(ord2utf8)(0x2fff, class_uchardata);
4004 *class_uchardata++ = XCL_RANGE;
4005 class_uchardata += PRIV(ord2utf8)(0x3001, class_uchardata);
4006 class_uchardata += PRIV(ord2utf8)(0x7fffffff, class_uchardata);
4007 }
4008 #endif
4009 continue;
4010
4011 case ESC_v:
4012 SETBIT(classbits, 0x0a); /* LF */
4013 SETBIT(classbits, 0x0b); /* VT */
4014 SETBIT(classbits, 0x0c); /* FF */
4015 SETBIT(classbits, 0x0d); /* CR */
4016 SETBIT(classbits, 0x85); /* NEL */
4017 #ifdef SUPPORT_UTF
4018 if (utf8)
4019 {
4020 xclass = TRUE;
4021 *class_uchardata++ = XCL_RANGE;
4022 class_uchardata += PRIV(ord2utf8)(0x2028, class_uchardata);
4023 class_uchardata += PRIV(ord2utf8)(0x2029, class_uchardata);
4024 }
4025 #endif
4026 continue;
4027
4028 case ESC_V:
4029 for (c = 0; c < 32; c++)
4030 {
4031 int x = 0xff;
4032 switch (c)
4033 {
4034 case 0x0a/8: x ^= 1 << (0x0a%8);
4035 x ^= 1 << (0x0b%8);
4036 x ^= 1 << (0x0c%8);
4037 x ^= 1 << (0x0d%8);
4038 break;
4039 case 0x85/8: x ^= 1 << (0x85%8); break;
4040 default: break;
4041 }
4042 classbits[c] |= x;
4043 }
4044
4045 #ifdef SUPPORT_UTF
4046 if (utf8)
4047 {
4048 xclass = TRUE;
4049 *class_uchardata++ = XCL_RANGE;
4050 class_uchardata += PRIV(ord2utf8)(0x0100, class_uchardata);
4051 class_uchardata += PRIV(ord2utf8)(0x2027, class_uchardata);
4052 *class_uchardata++ = XCL_RANGE;
4053 class_uchardata += PRIV(ord2utf8)(0x2029, class_uchardata);
4054 class_uchardata += PRIV(ord2utf8)(0x7fffffff, class_uchardata);
4055 }
4056 #endif
4057 continue;
4058
4059 #ifdef SUPPORT_UCP
4060 case ESC_p:
4061 case ESC_P:
4062 {
4063 BOOL negated;
4064 int pdata;
4065 int ptype = get_ucp(&ptr, &negated, &pdata, errorcodeptr);
4066 if (ptype < 0) goto FAILED;
4067 xclass = TRUE;
4068 *class_uchardata++ = ((-c == ESC_p) != negated)?
4069 XCL_PROP : XCL_NOTPROP;
4070 *class_uchardata++ = ptype;
4071 *class_uchardata++ = pdata;
4072 class_charcount -= 2; /* Not a < 256 character */
4073 continue;
4074 }
4075 #endif
4076 /* Unrecognized escapes are faulted if PCRE is running in its
4077 strict mode. By default, for compatibility with Perl, they are
4078 treated as literals. */
4079
4080 default:
4081 if ((options & PCRE_EXTRA) != 0)
4082 {
4083 *errorcodeptr = ERR7;
4084 goto FAILED;
4085 }
4086 class_charcount -= 2; /* Undo the default count from above */
4087 c = *ptr; /* Get the final character and fall through */
4088 break;
4089 }
4090 }
4091
4092 /* Fall through if we have a single character (c >= 0). This may be
4093 greater than 256 mode. */
4094
4095 } /* End of backslash handling */
4096
4097 /* A single character may be followed by '-' to form a range. However,
4098 Perl does not permit ']' to be the end of the range. A '-' character
4099 at the end is treated as a literal. Perl ignores orphaned \E sequences
4100 entirely. The code for handling \Q and \E is messy. */
4101
4102 CHECK_RANGE:
4103 while (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
4104 {
4105 inescq = FALSE;
4106 ptr += 2;
4107 }
4108
4109 oldptr = ptr;
4110
4111 /* Remember \r or \n */
4112
4113 if (c == CHAR_CR || c == CHAR_NL) cd->external_flags |= PCRE_HASCRORLF;
4114
4115 /* Check for range */
4116
4117 if (!inescq && ptr[1] == CHAR_MINUS)
4118 {
4119 int d;
4120 ptr += 2;
4121 while (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_E) ptr += 2;
4122
4123 /* If we hit \Q (not followed by \E) at this point, go into escaped
4124 mode. */
4125
4126 while (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_Q)
4127 {
4128 ptr += 2;
4129 if (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_E)
4130 { ptr += 2; continue; }
4131 inescq = TRUE;
4132 break;
4133 }
4134
4135 if (*ptr == 0 || (!inescq && *ptr == CHAR_RIGHT_SQUARE_BRACKET))
4136 {
4137 ptr = oldptr;
4138 goto LONE_SINGLE_CHARACTER;
4139 }
4140
4141 #ifdef SUPPORT_UTF8
4142 if (utf8)
4143 { /* Braces are required because the */
4144 GETCHARLEN(d, ptr, ptr); /* macro generates multiple statements */
4145 }
4146 else
4147 #endif
4148 d = *ptr; /* Not UTF-8 mode */
4149
4150 /* The second part of a range can be a single-character escape, but
4151 not any of the other escapes. Perl 5.6 treats a hyphen as a literal
4152 in such circumstances. */
4153
4154 if (!inescq && d == CHAR_BACKSLASH)
4155 {
4156 d = check_escape(&ptr, errorcodeptr, cd->bracount, options, TRUE);
4157 if (*errorcodeptr != 0) goto FAILED;
4158
4159 /* \b is backspace; any other special means the '-' was literal */
4160
4161 if (d < 0)
4162 {
4163 if (d == -ESC_b) d = CHAR_BS; else
4164 {
4165 ptr = oldptr;
4166 goto LONE_SINGLE_CHARACTER; /* A few lines below */
4167 }
4168 }
4169 }
4170
4171 /* Check that the two values are in the correct order. Optimize
4172 one-character ranges */
4173
4174 if (d < c)
4175 {
4176 *errorcodeptr = ERR8;
4177 goto FAILED;
4178 }
4179
4180 if (d == c) goto LONE_SINGLE_CHARACTER; /* A few lines below */
4181
4182 /* Remember \r or \n */
4183
4184 if (d == CHAR_CR || d == CHAR_NL) cd->external_flags |= PCRE_HASCRORLF;
4185
4186 /* In UTF-8 mode, if the upper limit is > 255, or > 127 for caseless
4187 matching, we have to use an XCLASS with extra data items. Caseless
4188 matching for characters > 127 is available only if UCP support is
4189 available. */
4190
4191 #ifdef SUPPORT_UTF
4192 if (utf8 && (d > 255 || ((options & PCRE_CASELESS) != 0 && d > 127)))
4193 #endif
4194 #ifndef COMPILE_PCRE8
4195 if (d > 255)
4196 #endif
4197 #if defined SUPPORT_UTF || defined COMPILE_PCRE16
4198 {
4199 xclass = TRUE;
4200
4201 /* With UCP support, we can find the other case equivalents of
4202 the relevant characters. There may be several ranges. Optimize how
4203 they fit with the basic range. */
4204
4205 #ifdef SUPPORT_UCP
4206 if ((options & PCRE_CASELESS) != 0)
4207 {
4208 unsigned int occ, ocd;
4209 unsigned int cc = c;
4210 unsigned int origd = d;
4211 while (get_othercase_range(&cc, origd, &occ, &ocd))
4212 {
4213 if (occ >= (unsigned int)c &&
4214 ocd <= (unsigned int)d)
4215 continue; /* Skip embedded ranges */
4216
4217 if (occ < (unsigned int)c &&
4218 ocd >= (unsigned int)c - 1) /* Extend the basic range */
4219 { /* if there is overlap, */
4220 c = occ; /* noting that if occ < c */
4221 continue; /* we can't have ocd > d */
4222 } /* because a subrange is */
4223 if (ocd > (unsigned int)d &&
4224 occ <= (unsigned int)d + 1) /* always shorter than */
4225 { /* the basic range. */
4226 d = ocd;
4227 continue;
4228 }
4229
4230 if (occ == ocd)
4231 {
4232 *class_uchardata++ = XCL_SINGLE;
4233 }
4234 else
4235 {
4236 *class_uchardata++ = XCL_RANGE;
4237 class_uchardata += PRIV(ord2utf8)(occ, class_uchardata);
4238 }
4239 class_uchardata += PRIV(ord2utf8)(ocd, class_uchardata);
4240 }
4241 }
4242 #endif /* SUPPORT_UCP */
4243
4244 /* Now record the original range, possibly modified for UCP caseless
4245 overlapping ranges. */
4246
4247 *class_uchardata++ = XCL_RANGE;
4248 #ifdef SUPPORT_UTF
4249 class_uchardata += PRIV(ord2utf8)(c, class_uchardata);
4250 class_uchardata += PRIV(ord2utf8)(d, class_uchardata);
4251 #else
4252 *class_uchardata++ = c;
4253 *class_uchardata++ = d;
4254 #endif
4255
4256 /* With UCP support, we are done. Without UCP support, there is no
4257 caseless matching for UTF characters > 127; we can use the bit map
4258 for the smaller ones. As for 16 bit characters without UTF, we
4259 can still use */
4260
4261 #ifdef SUPPORT_UCP
4262 continue; /* With next character in the class */
4263 #else
4264 #ifdef SUPPORT_UTF
4265 if ((options & PCRE_CASELESS) == 0 || c > 127) continue;
4266 /* Adjust upper limit and fall through to set up the map */
4267 d = 127;
4268 #else
4269 if (c > 255) continue;
4270 /* Adjust upper limit and fall through to set up the map */
4271 d = 255;
4272 #endif /* SUPPORT_UTF */
4273 #endif /* SUPPORT_UCP */
4274 }
4275 #endif /* SUPPORT_UTF8 || COMPILE_PCRE16 */
4276
4277 /* We use the bit map for 8 bit mode, or when the characters fall
4278 partially or entirely to [0-255] ([0-127] for UCP) ranges. */
4279
4280 class_charcount += d - c + 1;
4281 class_lastchar = d;
4282
4283 /* We can save a bit of time by skipping this in the pre-compile. */
4284
4285 if (lengthptr == NULL) for (; c <= d; c++)
4286 {
4287 classbits[c/8] |= (1 << (c&7));
4288 if ((options & PCRE_CASELESS) != 0)
4289 {
4290 int uc = cd->fcc[c]; /* flip case */
4291 classbits[uc/8] |= (1 << (uc&7));
4292 }
4293 }
4294
4295 continue; /* Go get the next char in the class */
4296 }
4297
4298 /* Handle a lone single character - we can get here for a normal
4299 non-escape char, or after \ that introduces a single character or for an
4300 apparent range that isn't. */
4301
4302 LONE_SINGLE_CHARACTER:
4303
4304 /* Handle a character that cannot go in the bit map */
4305
4306 #ifdef SUPPORT_UTF
4307 if (utf8 && (c > 255 || ((options & PCRE_CASELESS) != 0 && c > 127)))
4308 #endif
4309 #ifndef COMPILE_PCRE8
4310 if (c > 255)
4311 #endif
4312 #if defined SUPPORT_UTF || defined COMPILE_PCRE16
4313 {
4314 xclass = TRUE;
4315 *class_uchardata++ = XCL_SINGLE;
4316 #ifdef SUPPORT_UTF
4317 class_uchardata += PRIV(ord2utf8)(c, class_uchardata);
4318 #else
4319 *class_uchardata++ = c;
4320 #endif
4321
4322 #ifdef SUPPORT_UCP
4323 if ((options & PCRE_CASELESS) != 0)
4324 {
4325 unsigned int othercase;
4326 if ((othercase = UCD_OTHERCASE(c)) != c)
4327 {
4328 *class_uchardata++ = XCL_SINGLE;
4329 class_uchardata += PRIV(ord2utf8)(othercase, class_uchardata);
4330 }
4331 }
4332 #endif /* SUPPORT_UCP */
4333
4334 }
4335 else
4336 #endif /* SUPPORT_UTF8 */
4337
4338 /* Handle a single-byte character */
4339 {
4340 classbits[c/8] |= (1 << (c&7));
4341 if ((options & PCRE_CASELESS) != 0)
4342 {
4343 c = cd->fcc[c]; /* flip case */
4344 classbits[c/8] |= (1 << (c&7));
4345 }
4346 class_charcount++;
4347 class_lastchar = c;
4348 }
4349 }
4350
4351 /* Loop until ']' reached. This "while" is the end of the "do" far above.
4352 If we are at the end of an internal nested string, revert to the outer
4353 string. */
4354
4355 while (((c = *(++ptr)) != 0 ||
4356 (nestptr != NULL &&
4357 (ptr = nestptr, nestptr = NULL, c = *(++ptr)) != 0)) &&
4358 (c != CHAR_RIGHT_SQUARE_BRACKET || inescq));
4359
4360 /* Check for missing terminating ']' */
4361
4362 if (c == 0)
4363 {
4364 *errorcodeptr = ERR6;
4365 goto FAILED;
4366 }
4367
4368 /* If class_charcount is 1, we saw precisely one character whose value is
4369 less than 256. As long as there were no characters >= 128 and there was no
4370 use of \p or \P, in other words, no use of any XCLASS features, we can
4371 optimize.
4372
4373 In UTF-8 mode, we can optimize the negative case only if there were no
4374 characters >= 128 because OP_NOT and the related opcodes like OP_NOTSTAR
4375 operate on single-bytes characters only. This is an historical hangover.
4376 Maybe one day we can tidy these opcodes to handle multi-byte characters.
4377
4378 The optimization throws away the bit map. We turn the item into a
4379 1-character OP_CHAR[I] if it's positive, or OP_NOT[I] if it's negative.
4380 Note that OP_NOT[I] does not support multibyte characters. In the positive
4381 case, it can cause firstchar to be set. Otherwise, there can be no first
4382 char if this item is first, whatever repeat count may follow. In the case
4383 of reqchar, save the previous value for reinstating. */
4384
4385 #ifdef SUPPORT_UTF
4386 if (class_charcount == 1 && !xclass &&
4387 (!utf8 || !negate_class || class_lastchar < 128))
4388 #elif defined COMPILE_PCRE8
4389 if (class_charcount == 1)
4390 #else
4391 if (class_charcount == 1 && !xclass)
4392 #endif
4393 {
4394 zeroreqchar = reqchar;
4395
4396 /* The OP_NOT[I] opcodes work on one-byte characters only. */
4397
4398 if (negate_class)
4399 {
4400 if (firstchar == REQ_UNSET) firstchar = REQ_NONE;
4401 zerofirstchar = firstchar;
4402 *code++ = ((options & PCRE_CASELESS) != 0)? OP_NOTI: OP_NOT;
4403 *code++ = class_lastchar;
4404 break;
4405 }
4406
4407 /* For a single, positive character, get the value into mcbuffer, and
4408 then we can handle this with the normal one-character code. */
4409
4410 #ifdef SUPPORT_UTF8
4411 if (utf8 && class_lastchar > 127)
4412 mclength = PRIV(ord2utf8)(class_lastchar, mcbuffer);
4413 else
4414 #endif
4415 {
4416 mcbuffer[0] = class_lastchar;
4417 mclength = 1;
4418 }
4419 goto ONE_CHAR;
4420 } /* End of 1-char optimization */
4421
4422 /* The general case - not the one-char optimization. If this is the first
4423 thing in the branch, there can be no first char setting, whatever the
4424 repeat count. Any reqchar setting must remain unchanged after any kind of
4425 repeat. */
4426
4427 if (firstchar == REQ_UNSET) firstchar = REQ_NONE;
4428 zerofirstchar = firstchar;
4429 zeroreqchar = reqchar;
4430
4431 /* If there are characters with values > 255, we have to compile an
4432 extended class, with its own opcode, unless there was a negated special
4433 such as \S in the class, and PCRE_UCP is not set, because in that case all
4434 characters > 255 are in the class, so any that were explicitly given as
4435 well can be ignored. If (when there are explicit characters > 255 that must
4436 be listed) there are no characters < 256, we can omit the bitmap in the
4437 actual compiled code. */
4438
4439 #ifdef SUPPORT_UTF
4440 if (xclass && (!should_flip_negation || (options & PCRE_UCP) != 0))
4441 #endif
4442 #ifndef COMPILE_PCRE8
4443 if (xclass && !should_flip_negation)
4444 #endif
4445 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
4446 {
4447 *class_uchardata++ = XCL_END; /* Marks the end of extra data */
4448 *code++ = OP_XCLASS;
4449 code += LINK_SIZE;
4450 *code = negate_class? XCL_NOT:0;
4451
4452 /* If the map is required, move up the extra data to make room for it;
4453 otherwise just move the code pointer to the end of the extra data. */
4454
4455 if (class_charcount > 0)
4456 {
4457 *code++ |= XCL_MAP;
4458 memmove(code + (32 / sizeof(pcre_uchar)), code,
4459 IN_UCHARS(class_uchardata - code));
4460 memcpy(code, classbits, 32);
4461 code = class_uchardata + (32 / sizeof(pcre_uchar));
4462 }
4463 else code = class_uchardata;
4464
4465 /* Now fill in the complete length of the item */
4466
4467 PUT(previous, 1, code - previous);
4468 break; /* End of class handling */
4469 }
4470 #endif
4471
4472 /* If there are no characters > 255, or they are all to be included or
4473 excluded, set the opcode to OP_CLASS or OP_NCLASS, depending on whether the
4474 whole class was negated and whether there were negative specials such as \S
4475 (non-UCP) in the class. Then copy the 32-byte map into the code vector,
4476 negating it if necessary. */
4477
4478 *code++ = (negate_class == should_flip_negation) ? OP_CLASS : OP_NCLASS;
4479 if (lengthptr == NULL) /* Save time in the pre-compile phase */
4480 {
4481 if (negate_class)
4482 for (c = 0; c < 32; c++) classbits[c] = ~classbits[c];
4483 memcpy(code, classbits, 32);
4484 }
4485 code += 32 / sizeof(pcre_uchar);
4486 break;
4487
4488
4489 /* ===================================================================*/
4490 /* Various kinds of repeat; '{' is not necessarily a quantifier, but this
4491 has been tested above. */
4492
4493 case CHAR_LEFT_CURLY_BRACKET:
4494 if (!is_quantifier) goto NORMAL_CHAR;
4495 ptr = read_repeat_counts(ptr+1, &repeat_min, &repeat_max, errorcodeptr);
4496 if (*errorcodeptr != 0) goto FAILED;
4497 goto REPEAT;
4498
4499 case CHAR_ASTERISK:
4500 repeat_min = 0;
4501 repeat_max = -1;
4502 goto REPEAT;
4503
4504 case CHAR_PLUS:
4505 repeat_min = 1;
4506 repeat_max = -1;
4507 goto REPEAT;
4508
4509 case CHAR_QUESTION_MARK:
4510 repeat_min = 0;
4511 repeat_max = 1;
4512
4513 REPEAT:
4514 if (previous == NULL)
4515 {
4516 *errorcodeptr = ERR9;
4517 goto FAILED;
4518 }
4519
4520 if (repeat_min == 0)
4521 {
4522 firstchar = zerofirstchar; /* Adjust for zero repeat */
4523 reqchar = zeroreqchar; /* Ditto */
4524 }
4525
4526 /* Remember whether this is a variable length repeat */
4527
4528 reqvary = (repeat_min == repeat_max)? 0 : REQ_VARY;
4529
4530 op_type = 0; /* Default single-char op codes */
4531 possessive_quantifier = FALSE; /* Default not possessive quantifier */
4532
4533 /* Save start of previous item, in case we have to move it up in order to
4534 insert something before it. */
4535
4536 tempcode = previous;
4537
4538 /* If the next character is '+', we have a possessive quantifier. This
4539 implies greediness, whatever the setting of the PCRE_UNGREEDY option.
4540 If the next character is '?' this is a minimizing repeat, by default,
4541 but if PCRE_UNGREEDY is set, it works the other way round. We change the
4542 repeat type to the non-default. */
4543
4544 if (ptr[1] == CHAR_PLUS)
4545 {
4546 repeat_type = 0; /* Force greedy */
4547 possessive_quantifier = TRUE;
4548 ptr++;
4549 }
4550 else if (ptr[1] == CHAR_QUESTION_MARK)
4551 {
4552 repeat_type = greedy_non_default;
4553 ptr++;
4554 }
4555 else repeat_type = greedy_default;
4556
4557 /* If previous was a recursion call, wrap it in atomic brackets so that
4558 previous becomes the atomic group. All recursions were so wrapped in the
4559 past, but it no longer happens for non-repeated recursions. In fact, the
4560 repeated ones could be re-implemented independently so as not to need this,
4561 but for the moment we rely on the code for repeating groups. */
4562
4563 if (*previous == OP_RECURSE)
4564 {
4565 memmove(previous + 1 + LINK_SIZE, previous, IN_UCHARS(1 + LINK_SIZE));
4566 *previous = OP_ONCE;
4567 PUT(previous, 1, 2 + 2*LINK_SIZE);
4568 previous[2 + 2*LINK_SIZE] = OP_KET;
4569 PUT(previous, 3 + 2*LINK_SIZE, 2 + 2*LINK_SIZE);
4570 code += 2 + 2 * LINK_SIZE;
4571 length_prevgroup = 3 + 3*LINK_SIZE;
4572
4573 /* When actually compiling, we need to check whether this was a forward
4574 reference, and if so, adjust the offset. */
4575
4576 if (lengthptr == NULL && cd->hwm >= cd->start_workspace + LINK_SIZE)
4577 {
4578 int offset = GET(cd->hwm, -LINK_SIZE);
4579 if (offset == previous + 1 - cd->start_code)
4580 PUT(cd->hwm, -LINK_SIZE, offset + 1 + LINK_SIZE);
4581 }
4582 }
4583
4584 /* Now handle repetition for the different types of item. */
4585
4586 /* If previous was a character match, abolish the item and generate a
4587 repeat item instead. If a char item has a minumum of more than one, ensure
4588 that it is set in reqchar - it might not be if a sequence such as x{3} is
4589 the first thing in a branch because the x will have gone into firstchar
4590 instead. */
4591
4592 if (*previous == OP_CHAR || *previous == OP_CHARI)
4593 {
4594 op_type = (*previous == OP_CHAR)? 0 : OP_STARI - OP_STAR;
4595
4596 /* Deal with UTF-8 characters that take up more than one byte. It's
4597 easier to write this out separately than try to macrify it. Use c to
4598 hold the length of the character in bytes, plus 0x80 to flag that it's a
4599 length rather than a small character. */
4600
4601 #ifdef SUPPORT_UTF8
4602 if (utf8 && (code[-1] & 0x80) != 0)
4603 {
4604 pcre_uchar *lastchar = code - 1;
4605 while((*lastchar & 0xc0) == 0x80) lastchar--;
4606 c = code - lastchar; /* Length of UTF-8 character */
4607 memcpy(utf8_char, lastchar, c); /* Save the char */
4608 c |= 0x80; /* Flag c as a length */
4609 }
4610 else
4611 #endif
4612
4613 /* Handle the case of a single byte - either with no UTF8 support, or
4614 with UTF-8 disabled, or for a UTF-8 character < 128. */
4615
4616 {
4617 c = code[-1];
4618 if (repeat_min > 1) reqchar = c | req_caseopt | cd->req_varyopt;
4619 }
4620
4621 /* If the repetition is unlimited, it pays to see if the next thing on
4622 the line is something that cannot possibly match this character. If so,
4623 automatically possessifying this item gains some performance in the case
4624 where the match fails. */
4625
4626 if (!possessive_quantifier &&
4627 repeat_max < 0 &&
4628 check_auto_possessive(previous, utf8, ptr + 1, options, cd))
4629 {
4630 repeat_type = 0; /* Force greedy */
4631 possessive_quantifier = TRUE;
4632 }
4633
4634 goto OUTPUT_SINGLE_REPEAT; /* Code shared with single character types */
4635 }
4636
4637 /* If previous was a single negated character ([^a] or similar), we use
4638 one of the special opcodes, replacing it. The code is shared with single-
4639 character repeats by setting opt_type to add a suitable offset into
4640 repeat_type. We can also test for auto-possessification. OP_NOT and OP_NOTI
4641 are currently used only for single-byte chars. */
4642
4643 else if (*previous == OP_NOT || *previous == OP_NOTI)
4644 {
4645 op_type = ((*previous == OP_NOT)? OP_NOTSTAR : OP_NOTSTARI) - OP_STAR;
4646 c = previous[1];
4647 if (!possessive_quantifier &&
4648 repeat_max < 0 &&
4649 check_auto_possessive(previous, utf8, ptr + 1, options, cd))
4650 {
4651 repeat_type = 0; /* Force greedy */
4652 possessive_quantifier = TRUE;
4653 }
4654 goto OUTPUT_SINGLE_REPEAT;
4655 }
4656
4657 /* If previous was a character type match (\d or similar), abolish it and
4658 create a suitable repeat item. The code is shared with single-character
4659 repeats by setting op_type to add a suitable offset into repeat_type. Note
4660 the the Unicode property types will be present only when SUPPORT_UCP is
4661 defined, but we don't wrap the little bits of code here because it just
4662 makes it horribly messy. */
4663
4664 else if (*previous < OP_EODN)
4665 {
4666 pcre_uchar *oldcode;
4667 int prop_type, prop_value;
4668 op_type = OP_TYPESTAR - OP_STAR; /* Use type opcodes */
4669 c = *previous;
4670
4671 if (!possessive_quantifier &&
4672 repeat_max < 0 &&
4673 check_auto_possessive(previous, utf8, ptr + 1, options, cd))
4674 {
4675 repeat_type = 0; /* Force greedy */
4676 possessive_quantifier = TRUE;
4677 }
4678
4679 OUTPUT_SINGLE_REPEAT:
4680 if (*previous == OP_PROP || *previous == OP_NOTPROP)
4681 {
4682 prop_type = previous[1];
4683 prop_value = previous[2];
4684 }
4685 else prop_type = prop_value = -1;
4686
4687 oldcode = code;
4688 code = previous; /* Usually overwrite previous item */
4689
4690 /* If the maximum is zero then the minimum must also be zero; Perl allows
4691 this case, so we do too - by simply omitting the item altogether. */
4692
4693 if (repeat_max == 0) goto END_REPEAT;
4694
4695 /*--------------------------------------------------------------------*/
4696 /* This code is obsolete from release 8.00; the restriction was finally
4697 removed: */
4698
4699 /* All real repeats make it impossible to handle partial matching (maybe
4700 one day we will be able to remove this restriction). */
4701
4702 /* if (repeat_max != 1) cd->external_flags |= PCRE_NOPARTIAL; */
4703 /*--------------------------------------------------------------------*/
4704
4705 /* Combine the op_type with the repeat_type */
4706
4707 repeat_type += op_type;
4708
4709 /* A minimum of zero is handled either as the special case * or ?, or as
4710 an UPTO, with the maximum given. */
4711
4712 if (repeat_min == 0)
4713 {
4714 if (repeat_max == -1) *code++ = OP_STAR + repeat_type;
4715 else if (repeat_max == 1) *code++ = OP_QUERY + repeat_type;
4716 else
4717 {
4718 *code++ = OP_UPTO + repeat_type;
4719 PUT2INC(code, 0, repeat_max);
4720 }
4721 }
4722
4723 /* A repeat minimum of 1 is optimized into some special cases. If the
4724 maximum is unlimited, we use OP_PLUS. Otherwise, the original item is
4725 left in place and, if the maximum is greater than 1, we use OP_UPTO with
4726 one less than the maximum. */
4727
4728 else if (repeat_min == 1)
4729 {
4730 if (repeat_max == -1)
4731 *code++ = OP_PLUS + repeat_type;
4732 else
4733 {
4734 code = oldcode; /* leave previous item in place */
4735 if (repeat_max == 1) goto END_REPEAT;
4736 *code++ = OP_UPTO + repeat_type;
4737 PUT2INC(code, 0, repeat_max - 1);
4738 }
4739 }
4740
4741 /* The case {n,n} is just an EXACT, while the general case {n,m} is
4742 handled as an EXACT followed by an UPTO. */
4743
4744 else
4745 {
4746 *code++ = OP_EXACT + op_type; /* NB EXACT doesn't have repeat_type */
4747 PUT2INC(code, 0, repeat_min);
4748
4749 /* If the maximum is unlimited, insert an OP_STAR. Before doing so,
4750 we have to insert the character for the previous code. For a repeated
4751 Unicode property match, there are two extra bytes that define the
4752 required property. In UTF-8 mode, long characters have their length in
4753 c, with the 0x80 bit as a flag. */
4754
4755 if (repeat_max < 0)
4756 {
4757 #ifdef SUPPORT_UTF8
4758 if (utf8 && c >= 128)
4759 {
4760 memcpy(code, utf8_char, c & 7);
4761 code += c & 7;
4762 }
4763 else
4764 #endif
4765 {
4766 *code++ = c;
4767 if (prop_type >= 0)
4768 {
4769 *code++ = prop_type;
4770 *code++ = prop_value;
4771 }
4772 }
4773 *code++ = OP_STAR + repeat_type;
4774 }
4775
4776 /* Else insert an UPTO if the max is greater than the min, again
4777 preceded by the character, for the previously inserted code. If the
4778 UPTO is just for 1 instance, we can use QUERY instead. */
4779
4780 else if (repeat_max != repeat_min)
4781 {
4782 #ifdef SUPPORT_UTF8
4783 if (utf8 && c >= 128)
4784 {
4785 memcpy(code, utf8_char, c & 7);
4786 code += c & 7;
4787 }
4788 else
4789 #endif
4790 *code++ = c;
4791 if (prop_type >= 0)
4792 {
4793 *code++ = prop_type;
4794 *code++ = prop_value;
4795 }
4796 repeat_max -= repeat_min;
4797
4798 if (repeat_max == 1)
4799 {
4800 *code++ = OP_QUERY + repeat_type;
4801 }
4802 else
4803 {
4804 *code++ = OP_UPTO + repeat_type;
4805 PUT2INC(code, 0, repeat_max);
4806 }
4807 }
4808 }
4809
4810 /* The character or character type itself comes last in all cases. */
4811
4812 #ifdef SUPPORT_UTF8
4813 if (utf8 && c >= 128)
4814 {
4815 memcpy(code, utf8_char, c & 7);
4816 code += c & 7;
4817 }
4818 else
4819 #endif
4820 *code++ = c;
4821
4822 /* For a repeated Unicode property match, there are two extra bytes that
4823 define the required property. */
4824
4825 #ifdef SUPPORT_UCP
4826 if (prop_type >= 0)
4827 {
4828 *code++ = prop_type;
4829 *code++ = prop_value;
4830 }
4831 #endif
4832 }
4833
4834 /* If previous was a character class or a back reference, we put the repeat
4835 stuff after it, but just skip the item if the repeat was {0,0}. */
4836
4837 else if (*previous == OP_CLASS ||
4838 *previous == OP_NCLASS ||
4839 #if defined SUPPORT_UTF8 || !defined COMPILE_PCRE8
4840 *previous == OP_XCLASS ||
4841 #endif
4842 *previous == OP_REF ||
4843 *previous == OP_REFI)
4844 {
4845 if (repeat_max == 0)
4846 {
4847 code = previous;
4848 goto END_REPEAT;
4849 }
4850
4851 /*--------------------------------------------------------------------*/
4852 /* This code is obsolete from release 8.00; the restriction was finally
4853 removed: */
4854
4855 /* All real repeats make it impossible to handle partial matching (maybe
4856 one day we will be able to remove this restriction). */
4857
4858 /* if (repeat_max != 1) cd->external_flags |= PCRE_NOPARTIAL; */
4859 /*--------------------------------------------------------------------*/
4860
4861 if (repeat_min == 0 && repeat_max == -1)
4862 *code++ = OP_CRSTAR + repeat_type;
4863 else if (repeat_min == 1 && repeat_max == -1)
4864 *code++ = OP_CRPLUS + repeat_type;
4865 else if (repeat_min == 0 && repeat_max == 1)
4866 *code++ = OP_CRQUERY + repeat_type;
4867 else
4868 {
4869 *code++ = OP_CRRANGE + repeat_type;
4870 PUT2INC(code, 0, repeat_min);
4871 if (repeat_max == -1) repeat_max = 0; /* 2-byte encoding for max */
4872 PUT2INC(code, 0, repeat_max);
4873 }
4874 }
4875
4876 /* If previous was a bracket group, we may have to replicate it in certain
4877 cases. Note that at this point we can encounter only the "basic" bracket
4878 opcodes such as BRA and CBRA, as this is the place where they get converted
4879 into the more special varieties such as BRAPOS and SBRA. A test for >=
4880 OP_ASSERT and <= OP_COND includes ASSERT, ASSERT_NOT, ASSERTBACK,
4881 ASSERTBACK_NOT, ONCE, BRA, CBRA, and COND. Originally, PCRE did not allow
4882 repetition of assertions, but now it does, for Perl compatibility. */
4883
4884 else if (*previous >= OP_ASSERT && *previous <= OP_COND)
4885 {
4886 register int i;
4887 int len = (int)(code - previous);
4888 pcre_uchar *bralink = NULL;
4889 pcre_uchar *brazeroptr = NULL;
4890
4891 /* Repeating a DEFINE group is pointless, but Perl allows the syntax, so
4892 we just ignore the repeat. */
4893
4894 if (*previous == OP_COND && previous[LINK_SIZE+1] == OP_DEF)
4895 goto END_REPEAT;
4896
4897 /* There is no sense in actually repeating assertions. The only potential
4898 use of repetition is in cases when the assertion is optional. Therefore,
4899 if the minimum is greater than zero, just ignore the repeat. If the
4900 maximum is not not zero or one, set it to 1. */
4901
4902 if (*previous < OP_ONCE) /* Assertion */
4903 {
4904 if (repeat_min > 0) goto END_REPEAT;
4905 if (repeat_max < 0 || repeat_max > 1) repeat_max = 1;
4906 }
4907
4908 /* The case of a zero minimum is special because of the need to stick
4909 OP_BRAZERO in front of it, and because the group appears once in the
4910 data, whereas in other cases it appears the minimum number of times. For
4911 this reason, it is simplest to treat this case separately, as otherwise
4912 the code gets far too messy. There are several special subcases when the
4913 minimum is zero. */
4914
4915 if (repeat_min == 0)
4916 {
4917 /* If the maximum is also zero, we used to just omit the group from the
4918 output altogether, like this:
4919
4920 ** if (repeat_max == 0)
4921 ** {
4922 ** code = previous;
4923 ** goto END_REPEAT;
4924 ** }
4925
4926 However, that fails when a group or a subgroup within it is referenced
4927 as a subroutine from elsewhere in the pattern, so now we stick in
4928 OP_SKIPZERO in front of it so that it is skipped on execution. As we
4929 don't have a list of which groups are referenced, we cannot do this
4930 selectively.
4931
4932 If the maximum is 1 or unlimited, we just have to stick in the BRAZERO
4933 and do no more at this point. However, we do need to adjust any
4934 OP_RECURSE calls inside the group that refer to the group itself or any
4935 internal or forward referenced group, because the offset is from the
4936 start of the whole regex. Temporarily terminate the pattern while doing
4937 this. */
4938
4939 if (repeat_max <= 1) /* Covers 0, 1, and unlimited */
4940 {
4941 *code = OP_END;
4942 adjust_recurse(previous, 1, utf8, cd, save_hwm);
4943 memmove(previous + 1, previous, IN_UCHARS(len));
4944 code++;
4945 if (repeat_max == 0)
4946 {
4947 *previous++ = OP_SKIPZERO;
4948 goto END_REPEAT;
4949 }
4950 brazeroptr = previous; /* Save for possessive optimizing */
4951 *previous++ = OP_BRAZERO + repeat_type;
4952 }
4953
4954 /* If the maximum is greater than 1 and limited, we have to replicate
4955 in a nested fashion, sticking OP_BRAZERO before each set of brackets.
4956 The first one has to be handled carefully because it's the original
4957 copy, which has to be moved up. The remainder can be handled by code
4958 that is common with the non-zero minimum case below. We have to
4959 adjust the value or repeat_max, since one less copy is required. Once
4960 again, we may have to adjust any OP_RECURSE calls inside the group. */
4961
4962 else
4963 {
4964 int offset;
4965 *code = OP_END;
4966 adjust_recurse(previous, 2 + LINK_SIZE, utf8, cd, save_hwm);
4967 memmove(previous + 2 + LINK_SIZE, previous, IN_UCHARS(len));
4968 code += 2 + LINK_SIZE;
4969 *previous++ = OP_BRAZERO + repeat_type;
4970 *previous++ = OP_BRA;
4971
4972 /* We chain together the bracket offset fields that have to be
4973 filled in later when the ends of the brackets are reached. */
4974
4975 offset = (bralink == NULL)? 0 : (int)(previous - bralink);
4976 bralink = previous;
4977 PUTINC(previous, 0, offset);
4978 }
4979
4980 repeat_max--;
4981 }
4982
4983 /* If the minimum is greater than zero, replicate the group as many
4984 times as necessary, and adjust the maximum to the number of subsequent
4985 copies that we need. If we set a first char from the group, and didn't
4986 set a required char, copy the latter from the former. If there are any
4987 forward reference subroutine calls in the group, there will be entries on
4988 the workspace list; replicate these with an appropriate increment. */
4989
4990 else
4991 {
4992 if (repeat_min > 1)
4993 {
4994 /* In the pre-compile phase, we don't actually do the replication. We
4995 just adjust the length as if we had. Do some paranoid checks for
4996 potential integer overflow. The INT64_OR_DOUBLE type is a 64-bit
4997 integer type when available, otherwise double. */
4998
4999 if (lengthptr != NULL)
5000 {
5001 int delta = (repeat_min - 1)*length_prevgroup;
5002 if ((INT64_OR_DOUBLE)(repeat_min - 1)*
5003 (INT64_OR_DOUBLE)length_prevgroup >
5004 (INT64_OR_DOUBLE)INT_MAX ||
5005 OFLOW_MAX - *lengthptr < delta)
5006 {
5007 *errorcodeptr = ERR20;
5008 goto FAILED;
5009 }
5010 *lengthptr += delta;
5011 }
5012
5013 /* This is compiling for real */
5014
5015 else
5016 {
5017 if (groupsetfirstchar && reqchar < 0) reqchar = firstchar;
5018 for (i = 1; i < repeat_min; i++)
5019 {
5020 pcre_uchar *hc;
5021 pcre_uchar *this_hwm = cd->hwm;
5022 memcpy(code, previous, IN_UCHARS(len));
5023 for (hc = save_hwm; hc < this_hwm; hc += LINK_SIZE)
5024 {
5025 PUT(cd->hwm, 0, GET(hc, 0) + len);
5026 cd->hwm += LINK_SIZE;
5027 }
5028 save_hwm = this_hwm;
5029 code += len;
5030 }
5031 }
5032 }
5033
5034 if (repeat_max > 0) repeat_max -= repeat_min;
5035 }
5036
5037 /* This code is common to both the zero and non-zero minimum cases. If
5038 the maximum is limited, it replicates the group in a nested fashion,
5039 remembering the bracket starts on a stack. In the case of a zero minimum,
5040 the first one was set up above. In all cases the repeat_max now specifies
5041 the number of additional copies needed. Again, we must remember to
5042 replicate entries on the forward reference list. */
5043
5044 if (repeat_max >= 0)
5045 {
5046 /* In the pre-compile phase, we don't actually do the replication. We
5047 just adjust the length as if we had. For each repetition we must add 1
5048 to the length for BRAZERO and for all but the last repetition we must
5049 add 2 + 2*LINKSIZE to allow for the nesting that occurs. Do some
5050 paranoid checks to avoid integer overflow. The INT64_OR_DOUBLE type is
5051 a 64-bit integer type when available, otherwise double. */
5052
5053 if (lengthptr != NULL && repeat_max > 0)
5054 {
5055 int delta = repeat_max * (length_prevgroup + 1 + 2 + 2*LINK_SIZE) -
5056 2 - 2*LINK_SIZE; /* Last one doesn't nest */
5057 if ((INT64_OR_DOUBLE)repeat_max *
5058 (INT64_OR_DOUBLE)(length_prevgroup + 1 + 2 + 2*LINK_SIZE)
5059 > (INT64_OR_DOUBLE)INT_MAX ||
5060 OFLOW_MAX - *lengthptr < delta)
5061 {
5062 *errorcodeptr = ERR20;
5063 goto FAILED;
5064 }
5065 *lengthptr += delta;
5066 }
5067
5068 /* This is compiling for real */
5069
5070 else for (i = repeat_max - 1; i >= 0; i--)
5071 {
5072 pcre_uchar *hc;
5073 pcre_uchar *this_hwm = cd->hwm;
5074
5075 *code++ = OP_BRAZERO + repeat_type;
5076
5077 /* All but the final copy start a new nesting, maintaining the
5078 chain of brackets outstanding. */
5079
5080 if (i != 0)
5081 {
5082 int offset;
5083 *code++ = OP_BRA;
5084 offset = (bralink == NULL)? 0 : (int)(code - bralink);
5085 bralink = code;
5086 PUTINC(code, 0, offset);
5087 }
5088
5089 memcpy(code, previous, IN_UCHARS(len));
5090 for (hc = save_hwm; hc < this_hwm; hc += LINK_SIZE)
5091 {
5092 PUT(cd->hwm, 0, GET(hc, 0) + len + ((i != 0)? 2+LINK_SIZE : 1));
5093 cd->hwm += LINK_SIZE;
5094 }
5095 save_hwm = this_hwm;
5096 code += len;
5097 }
5098
5099 /* Now chain through the pending brackets, and fill in their length
5100 fields (which are holding the chain links pro tem). */
5101
5102 while (bralink != NULL)
5103 {
5104 int oldlinkoffset;
5105 int offset = (int)(code - bralink + 1);
5106 pcre_uchar *bra = code - offset;
5107 oldlinkoffset = GET(bra, 1);
5108 bralink = (oldlinkoffset == 0)? NULL : bralink - oldlinkoffset;
5109 *code++ = OP_KET;
5110 PUTINC(code, 0, offset);
5111 PUT(bra, 1, offset);
5112 }
5113 }
5114
5115 /* If the maximum is unlimited, set a repeater in the final copy. For
5116 ONCE brackets, that's all we need to do. However, possessively repeated
5117 ONCE brackets can be converted into non-capturing brackets, as the
5118 behaviour of (?:xx)++ is the same as (?>xx)++ and this saves having to
5119 deal with possessive ONCEs specially.
5120
5121 Otherwise, when we are doing the actual compile phase, check to see
5122 whether this group is one that could match an empty string. If so,
5123 convert the initial operator to the S form (e.g. OP_BRA -> OP_SBRA) so
5124 that runtime checking can be done. [This check is also applied to ONCE
5125 groups at runtime, but in a different way.]
5126
5127 Then, if the quantifier was possessive and the bracket is not a
5128 conditional, we convert the BRA code to the POS form, and the KET code to
5129 KETRPOS. (It turns out to be convenient at runtime to detect this kind of
5130 subpattern at both the start and at the end.) The use of special opcodes
5131 makes it possible to reduce greatly the stack usage in pcre_exec(). If
5132 the group is preceded by OP_BRAZERO, convert this to OP_BRAPOSZERO.
5133
5134 Then, if the minimum number of matches is 1 or 0, cancel the possessive
5135 flag so that the default action below, of wrapping everything inside
5136 atomic brackets, does not happen. When the minimum is greater than 1,
5137 there will be earlier copies of the group, and so we still have to wrap
5138 the whole thing. */
5139
5140 else
5141 {
5142 pcre_uchar *ketcode = code - 1 - LINK_SIZE;
5143 pcre_uchar *bracode = ketcode - GET(ketcode, 1);
5144
5145 /* Convert possessive ONCE brackets to non-capturing */
5146
5147 if ((*bracode == OP_ONCE || *bracode == OP_ONCE_NC) &&
5148 possessive_quantifier) *bracode = OP_BRA;
5149
5150 /* For non-possessive ONCE brackets, all we need to do is to
5151 set the KET. */
5152
5153 if (*bracode == OP_ONCE || *bracode == OP_ONCE_NC)
5154 *ketcode = OP_KETRMAX + repeat_type;
5155
5156 /* Handle non-ONCE brackets and possessive ONCEs (which have been
5157 converted to non-capturing above). */
5158
5159 else
5160 {
5161 /* In the compile phase, check for empty string matching. */
5162
5163 if (lengthptr == NULL)
5164 {
5165 pcre_uchar *scode = bracode;
5166 do
5167 {
5168 if (could_be_empty_branch(scode, ketcode, utf8, cd))
5169 {
5170 *bracode += OP_SBRA - OP_BRA;
5171 break;
5172 }
5173 scode += GET(scode, 1);
5174 }
5175 while (*scode == OP_ALT);
5176 }
5177
5178 /* Handle possessive quantifiers. */
5179
5180 if (possessive_quantifier)
5181 {
5182 /* For COND brackets, we wrap the whole thing in a possessively
5183 repeated non-capturing bracket, because we have not invented POS
5184 versions of the COND opcodes. Because we are moving code along, we
5185 must ensure that any pending recursive references are updated. */
5186
5187 if (*bracode == OP_COND || *bracode == OP_SCOND)
5188 {
5189 int nlen = (int)(code - bracode);
5190 *code = OP_END;
5191 adjust_recurse(bracode, 1 + LINK_SIZE, utf8, cd, save_hwm);
5192 memmove(bracode + 1 + LINK_SIZE, bracode, IN_UCHARS(nlen));
5193 code += 1 + LINK_SIZE;
5194 nlen += 1 + LINK_SIZE;
5195 *bracode = OP_BRAPOS;
5196 *code++ = OP_KETRPOS;
5197 PUTINC(code, 0, nlen);
5198 PUT(bracode, 1, nlen);
5199 }
5200
5201 /* For non-COND brackets, we modify the BRA code and use KETRPOS. */
5202
5203 else
5204 {
5205 *bracode += 1; /* Switch to xxxPOS opcodes */
5206 *ketcode = OP_KETRPOS;
5207 }
5208
5209 /* If the minimum is zero, mark it as possessive, then unset the
5210 possessive flag when the minimum is 0 or 1. */
5211
5212 if (brazeroptr != NULL) *brazeroptr = OP_BRAPOSZERO;
5213 if (repeat_min < 2) possessive_quantifier = FALSE;
5214 }
5215
5216 /* Non-possessive quantifier */
5217
5218 else *ketcode = OP_KETRMAX + repeat_type;
5219 }
5220 }
5221 }
5222
5223 /* If previous is OP_FAIL, it was generated by an empty class [] in
5224 JavaScript mode. The other ways in which OP_FAIL can be generated, that is
5225 by (*FAIL) or (?!) set previous to NULL, which gives a "nothing to repeat"
5226 error above. We can just ignore the repeat in JS case. */
5227
5228 else if (*previous == OP_FAIL) goto END_REPEAT;
5229
5230 /* Else there's some kind of shambles */
5231
5232 else
5233 {
5234 *errorcodeptr = ERR11;
5235 goto FAILED;
5236 }
5237
5238 /* If the character following a repeat is '+', or if certain optimization
5239 tests above succeeded, possessive_quantifier is TRUE. For some opcodes,
5240 there are special alternative opcodes for this case. For anything else, we
5241 wrap the entire repeated item inside OP_ONCE brackets. Logically, the '+'
5242 notation is just syntactic sugar, taken from Sun's Java package, but the
5243 special opcodes can optimize it.
5244
5245 Some (but not all) possessively repeated subpatterns have already been
5246 completely handled in the code just above. For them, possessive_quantifier
5247 is always FALSE at this stage.
5248
5249 Note that the repeated item starts at tempcode, not at previous, which
5250 might be the first part of a string whose (former) last char we repeated.
5251
5252 Possessifying an 'exact' quantifier has no effect, so we can ignore it. But
5253 an 'upto' may follow. We skip over an 'exact' item, and then test the
5254 length of what remains before proceeding. */
5255
5256 if (possessive_quantifier)
5257 {
5258 int len;
5259
5260 if (*tempcode == OP_TYPEEXACT)
5261 tempcode += PRIV(OP_lengths)[*tempcode] +
5262 ((tempcode[1 + IMM2_SIZE] == OP_PROP
5263 || tempcode[1 + IMM2_SIZE] == OP_NOTPROP)? 2 : 0);
5264
5265 else if (*tempcode == OP_EXACT || *tempcode == OP_NOTEXACT)
5266 {
5267 tempcode += PRIV(OP_lengths)[*tempcode];
5268 #ifdef SUPPORT_UTF8
5269 if (utf8 && tempcode[-1] >= 0xc0)
5270 tempcode += PRIV(utf8_table4)[tempcode[-1] & 0x3f];
5271 #endif
5272 }
5273
5274 len = (int)(code - tempcode);
5275 if (len > 0) switch (*tempcode)
5276 {
5277 case OP_STAR: *tempcode = OP_POSSTAR; break;
5278 case OP_PLUS: *tempcode = OP_POSPLUS; break;
5279 case OP_QUERY: *tempcode = OP_POSQUERY; break;
5280 case OP_UPTO: *tempcode = OP_POSUPTO; break;
5281
5282 case OP_STARI: *tempcode = OP_POSSTARI; break;
5283 case OP_PLUSI: *tempcode = OP_POSPLUSI; break;
5284 case OP_QUERYI: *tempcode = OP_POSQUERYI; break;
5285 case OP_UPTOI: *tempcode = OP_POSUPTOI; break;
5286
5287 case OP_NOTSTAR: *tempcode = OP_NOTPOSSTAR; break;
5288 case OP_NOTPLUS: *tempcode = OP_NOTPOSPLUS; break;
5289 case OP_NOTQUERY: *tempcode = OP_NOTPOSQUERY; break;
5290 case OP_NOTUPTO: *tempcode = OP_NOTPOSUPTO; break;
5291
5292 case OP_NOTSTARI: *tempcode = OP_NOTPOSSTARI; break;
5293 case OP_NOTPLUSI: *tempcode = OP_NOTPOSPLUSI; break;
5294 case OP_NOTQUERYI: *tempcode = OP_NOTPOSQUERYI; break;
5295 case OP_NOTUPTOI: *tempcode = OP_NOTPOSUPTOI; break;
5296
5297 case OP_TYPESTAR: *tempcode = OP_TYPEPOSSTAR; break;
5298 case OP_TYPEPLUS: *tempcode = OP_TYPEPOSPLUS; break;
5299 case OP_TYPEQUERY: *tempcode = OP_TYPEPOSQUERY; break;
5300 case OP_TYPEUPTO: *tempcode = OP_TYPEPOSUPTO; break;
5301
5302 /* Because we are moving code along, we must ensure that any
5303 pending recursive references are updated. */
5304
5305 default:
5306 *code = OP_END;
5307 adjust_recurse(tempcode, 1 + LINK_SIZE, utf8, cd, save_hwm);
5308 memmove(tempcode + 1 + LINK_SIZE, tempcode, IN_UCHARS(len));
5309 code += 1 + LINK_SIZE;
5310 len += 1 + LINK_SIZE;
5311 tempcode[0] = OP_ONCE;
5312 *code++ = OP_KET;
5313 PUTINC(code, 0, len);
5314 PUT(tempcode, 1, len);
5315 break;
5316 }
5317 }
5318
5319 /* In all case we no longer have a previous item. We also set the
5320 "follows varying string" flag for subsequently encountered reqchars if
5321 it isn't already set and we have just passed a varying length item. */
5322
5323 END_REPEAT:
5324 previous = NULL;
5325 cd->req_varyopt |= reqvary;
5326 break;
5327
5328
5329 /* ===================================================================*/
5330 /* Start of nested parenthesized sub-expression, or comment or lookahead or
5331 lookbehind or option setting or condition or all the other extended
5332 parenthesis forms. */
5333
5334 case CHAR_LEFT_PARENTHESIS:
5335 newoptions = options;
5336 skipbytes = 0;
5337 bravalue = OP_CBRA;
5338 save_hwm = cd->hwm;
5339 reset_bracount = FALSE;
5340
5341 /* First deal with various "verbs" that can be introduced by '*'. */
5342
5343 if (*(++ptr) == CHAR_ASTERISK &&
5344 ((cd->ctypes[ptr[1]] & ctype_letter) != 0 || ptr[1] == ':'))
5345 {
5346 int i, namelen;
5347 int arglen = 0;
5348 const char *vn = verbnames;
5349 const pcre_uchar *name = ptr + 1;
5350 const pcre_uchar *arg = NULL;
5351 previous = NULL;
5352 while ((cd->ctypes[*++ptr] & ctype_letter) != 0) {};
5353 namelen = (int)(ptr - name);
5354
5355 /* It appears that Perl allows any characters whatsoever, other than
5356 a closing parenthesis, to appear in arguments, so we no longer insist on
5357 letters, digits, and underscores. */
5358
5359 if (*ptr == CHAR_COLON)
5360 {
5361 arg = ++ptr;
5362 while (*ptr != 0 && *ptr != CHAR_RIGHT_PARENTHESIS) ptr++;
5363 arglen = (int)(ptr - arg);
5364 }
5365
5366 if (*ptr != CHAR_RIGHT_PARENTHESIS)
5367 {
5368 *errorcodeptr = ERR60;
5369 goto FAILED;
5370 }
5371
5372 /* Scan the table of verb names */
5373
5374 for (i = 0; i < verbcount; i++)
5375 {
5376 if (namelen == verbs[i].len &&
5377 STRNCMP_UC_C8(name, vn, namelen) == 0)
5378 {
5379 /* Check for open captures before ACCEPT and convert it to
5380 ASSERT_ACCEPT if in an assertion. */
5381
5382 if (verbs[i].op == OP_ACCEPT)
5383 {
5384 open_capitem *oc;
5385 if (arglen != 0)
5386 {
5387 *errorcodeptr = ERR59;
5388 goto FAILED;
5389 }
5390 cd->had_accept = TRUE;
5391 for (oc = cd->open_caps; oc != NULL; oc = oc->next)
5392 {
5393 *code++ = OP_CLOSE;
5394 PUT2INC(code, 0, oc->number);
5395 }
5396 *code++ = (cd->assert_depth > 0)? OP_ASSERT_ACCEPT : OP_ACCEPT;
5397
5398 /* Do not set firstchar after *ACCEPT */
5399 if (firstchar == REQ_UNSET) firstchar = REQ_NONE;
5400 }
5401
5402 /* Handle other cases with/without an argument */
5403
5404 else if (arglen == 0)
5405 {
5406 if (verbs[i].op < 0) /* Argument is mandatory */
5407 {
5408 *errorcodeptr = ERR66;
5409 goto FAILED;
5410 }
5411 *code = verbs[i].op;
5412 if (*code++ == OP_THEN) cd->external_flags |= PCRE_HASTHEN;
5413 }
5414
5415 else
5416 {
5417 if (verbs[i].op_arg < 0) /* Argument is forbidden */
5418 {
5419 *errorcodeptr = ERR59;
5420 goto FAILED;
5421 }
5422 *code = verbs[i].op_arg;
5423 if (*code++ == OP_THEN_ARG) cd->external_flags |= PCRE_HASTHEN;
5424 *code++ = arglen;
5425 memcpy(code, arg, IN_UCHARS(arglen));
5426 code += arglen;
5427 *code++ = 0;
5428 }
5429
5430 break; /* Found verb, exit loop */
5431 }
5432
5433 vn += verbs[i].len + 1;
5434 }
5435
5436 if (i < verbcount) continue; /* Successfully handled a verb */
5437 *errorcodeptr = ERR60; /* Verb not recognized */
5438 goto FAILED;
5439 }
5440
5441 /* Deal with the extended parentheses; all are introduced by '?', and the
5442 appearance of any of them means that this is not a capturing group. */
5443
5444 else if (*ptr == CHAR_QUESTION_MARK)
5445 {
5446 int i, set, unset, namelen;
5447 int *optset;
5448 const pcre_uchar *name;
5449 pcre_uchar *slot;
5450
5451 switch (*(++ptr))
5452 {
5453 case CHAR_NUMBER_SIGN: /* Comment; skip to ket */
5454 ptr++;
5455 while (*ptr != 0 && *ptr != CHAR_RIGHT_PARENTHESIS) ptr++;
5456 if (*ptr == 0)
5457 {
5458 *errorcodeptr = ERR18;
5459 goto FAILED;
5460 }
5461 continue;
5462
5463
5464 /* ------------------------------------------------------------ */
5465 case CHAR_VERTICAL_LINE: /* Reset capture count for each branch */
5466 reset_bracount = TRUE;
5467 /* Fall through */
5468
5469 /* ------------------------------------------------------------ */
5470 case CHAR_COLON: /* Non-capturing bracket */
5471 bravalue = OP_BRA;
5472 ptr++;
5473 break;
5474
5475
5476 /* ------------------------------------------------------------ */
5477 case CHAR_LEFT_PARENTHESIS:
5478 bravalue = OP_COND; /* Conditional group */
5479
5480 /* A condition can be an assertion, a number (referring to a numbered
5481 group), a name (referring to a named group), or 'R', referring to
5482 recursion. R<digits> and R&name are also permitted for recursion tests.
5483
5484 There are several syntaxes for testing a named group: (?(name)) is used
5485 by Python; Perl 5.10 onwards uses (?(<name>) or (?('name')).
5486
5487 There are two unfortunate ambiguities, caused by history. (a) 'R' can
5488 be the recursive thing or the name 'R' (and similarly for 'R' followed
5489 by digits), and (b) a number could be a name that consists of digits.
5490 In both cases, we look for a name first; if not found, we try the other
5491 cases. */
5492
5493 /* For conditions that are assertions, check the syntax, and then exit
5494 the switch. This will take control down to where bracketed groups,
5495 including assertions, are processed. */
5496
5497 if (ptr[1] == CHAR_QUESTION_MARK && (ptr[2] == CHAR_EQUALS_SIGN ||
5498 ptr[2] == CHAR_EXCLAMATION_MARK || ptr[2] == CHAR_LESS_THAN_SIGN))
5499 break;
5500
5501 /* Most other conditions use OP_CREF (a couple change to OP_RREF
5502 below), and all need to skip 1+IMM2_SIZE bytes at the start of the group. */
5503
5504 code[1+LINK_SIZE] = OP_CREF;
5505 skipbytes = 1+IMM2_SIZE;
5506 refsign = -1;
5507
5508 /* Check for a test for recursion in a named group. */
5509
5510 if (ptr[1] == CHAR_R && ptr[2] == CHAR_AMPERSAND)
5511 {
5512 terminator = -1;
5513 ptr += 2;
5514 code[1+LINK_SIZE] = OP_RREF; /* Change the type of test */
5515 }
5516
5517 /* Check for a test for a named group's having been set, using the Perl
5518 syntax (?(<name>) or (?('name') */
5519
5520 else if (ptr[1] == CHAR_LESS_THAN_SIGN)
5521 {
5522 terminator = CHAR_GREATER_THAN_SIGN;
5523 ptr++;
5524 }
5525 else if (ptr[1] == CHAR_APOSTROPHE)
5526 {
5527 terminator = CHAR_APOSTROPHE;
5528 ptr++;
5529 }
5530 else
5531 {
5532 terminator = 0;
5533 if (ptr[1] == CHAR_MINUS || ptr[1] == CHAR_PLUS) refsign = *(++ptr);
5534 }
5535
5536 /* We now expect to read a name; any thing else is an error */
5537
5538 if ((cd->ctypes[ptr[1]] & ctype_word) == 0)
5539 {
5540 ptr += 1; /* To get the right offset */
5541 *errorcodeptr = ERR28;
5542 goto FAILED;
5543 }
5544
5545 /* Read the name, but also get it as a number if it's all digits */
5546
5547 recno = 0;
5548 name = ++ptr;
5549 while ((cd->ctypes[*ptr] & ctype_word) != 0)
5550 {
5551 if (recno >= 0)
5552 recno = (IS_DIGIT(*ptr))? recno * 10 + *ptr - CHAR_0 : -1;
5553 ptr++;
5554 }
5555 namelen = (int)(ptr - name);
5556
5557 if ((terminator > 0 && *ptr++ != terminator) ||
5558 *ptr++ != CHAR_RIGHT_PARENTHESIS)
5559 {
5560 ptr--; /* Error offset */
5561 *errorcodeptr = ERR26;
5562 goto FAILED;
5563 }
5564
5565 /* Do no further checking in the pre-compile phase. */
5566
5567 if (lengthptr != NULL) break;
5568
5569 /* In the real compile we do the work of looking for the actual
5570 reference. If the string started with "+" or "-" we require the rest to
5571 be digits, in which case recno will be set. */
5572
5573 if (refsign > 0)
5574 {
5575 if (recno <= 0)
5576 {
5577 *errorcodeptr = ERR58;
5578 goto FAILED;
5579 }
5580 recno = (refsign == CHAR_MINUS)?
5581 cd->bracount - recno + 1 : recno +cd->bracount;
5582 if (recno <= 0 || recno > cd->final_bracount)
5583 {
5584 *errorcodeptr = ERR15;
5585 goto FAILED;
5586 }
5587 PUT2(code, 2+LINK_SIZE, recno);
5588 break;
5589 }
5590
5591 /* Otherwise (did not start with "+" or "-"), start by looking for the
5592 name. If we find a name, add one to the opcode to change OP_CREF or
5593 OP_RREF into OP_NCREF or OP_NRREF. These behave exactly the same,
5594 except they record that the reference was originally to a name. The
5595 information is used to check duplicate names. */
5596
5597 slot = cd->name_table;
5598 for (i = 0; i < cd->names_found; i++)
5599 {
5600 if (STRNCMP_UC_UC(name, slot+IMM2_SIZE, namelen) == 0) break;
5601 slot += cd->name_entry_size;
5602 }
5603
5604 /* Found a previous named subpattern */
5605
5606 if (i < cd->names_found)
5607 {
5608 recno = GET2(slot, 0);
5609 PUT2(code, 2+LINK_SIZE, recno);
5610 code[1+LINK_SIZE]++;
5611 }
5612
5613 /* Search the pattern for a forward reference */
5614
5615 else if ((i = find_parens(cd, name, namelen,
5616 (options & PCRE_EXTENDED) != 0, utf8)) > 0)
5617 {
5618 PUT2(code, 2+LINK_SIZE, i);
5619 code[1+LINK_SIZE]++;
5620 }
5621
5622 /* If terminator == 0 it means that the name followed directly after
5623 the opening parenthesis [e.g. (?(abc)...] and in this case there are
5624 some further alternatives to try. For the cases where terminator != 0
5625 [things like (?(<name>... or (?('name')... or (?(R&name)... ] we have
5626 now checked all the possibilities, so give an error. */
5627
5628 else if (terminator != 0)
5629 {
5630 *errorcodeptr = ERR15;
5631 goto FAILED;
5632 }
5633
5634 /* Check for (?(R) for recursion. Allow digits after R to specify a
5635 specific group number. */
5636
5637 else if (*name == CHAR_R)
5638 {
5639 recno = 0;
5640 for (i = 1; i < namelen; i++)
5641 {
5642 if (!IS_DIGIT(name[i]))
5643 {
5644 *errorcodeptr = ERR15;
5645 goto FAILED;
5646 }
5647 recno = recno * 10 + name[i] - CHAR_0;
5648 }
5649 if (recno == 0) recno = RREF_ANY;
5650 code[1+LINK_SIZE] = OP_RREF; /* Change test type */
5651 PUT2(code, 2+LINK_SIZE, recno);
5652 }
5653
5654 /* Similarly, check for the (?(DEFINE) "condition", which is always
5655 false. */
5656
5657 else if (namelen == 6 && STRNCMP_UC_C8(name, STRING_DEFINE, 6) == 0)
5658 {
5659 code[1+LINK_SIZE] = OP_DEF;
5660 skipbytes = 1;
5661 }
5662
5663 /* Check for the "name" actually being a subpattern number. We are
5664 in the second pass here, so final_bracount is set. */
5665
5666 else if (recno > 0 && recno <= cd->final_bracount)
5667 {
5668 PUT2(code, 2+LINK_SIZE, recno);
5669 }
5670
5671 /* Either an unidentified subpattern, or a reference to (?(0) */
5672
5673 else
5674 {
5675 *errorcodeptr = (recno == 0)? ERR35: ERR15;
5676 goto FAILED;
5677 }
5678 break;
5679
5680
5681 /* ------------------------------------------------------------ */
5682 case CHAR_EQUALS_SIGN: /* Positive lookahead */
5683 bravalue = OP_ASSERT;
5684 cd->assert_depth += 1;
5685 ptr++;
5686 break;
5687
5688
5689 /* ------------------------------------------------------------ */
5690 case CHAR_EXCLAMATION_MARK: /* Negative lookahead */
5691 ptr++;
5692 if (*ptr == CHAR_RIGHT_PARENTHESIS) /* Optimize (?!) */
5693 {
5694 *code++ = OP_FAIL;
5695 previous = NULL;
5696 continue;
5697 }
5698 bravalue = OP_ASSERT_NOT;
5699 cd->assert_depth += 1;
5700 break;
5701
5702
5703 /* ------------------------------------------------------------ */
5704 case CHAR_LESS_THAN_SIGN: /* Lookbehind or named define */
5705 switch (ptr[1])
5706 {
5707 case CHAR_EQUALS_SIGN: /* Positive lookbehind */
5708 bravalue = OP_ASSERTBACK;
5709 cd->assert_depth += 1;
5710 ptr += 2;
5711 break;
5712
5713 case CHAR_EXCLAMATION_MARK: /* Negative lookbehind */
5714 bravalue = OP_ASSERTBACK_NOT;
5715 cd->assert_depth += 1;
5716 ptr += 2;
5717 break;
5718
5719 default: /* Could be name define, else bad */
5720 if ((cd->ctypes[ptr[1]] & ctype_word) != 0) goto DEFINE_NAME;
5721 ptr++; /* Correct offset for error */
5722 *errorcodeptr = ERR24;
5723 goto FAILED;
5724 }
5725 break;
5726
5727
5728 /* ------------------------------------------------------------ */
5729 case CHAR_GREATER_THAN_SIGN: /* One-time brackets */
5730 bravalue = OP_ONCE;
5731 ptr++;
5732 break;
5733
5734
5735 /* ------------------------------------------------------------ */
5736 case CHAR_C: /* Callout - may be followed by digits; */
5737 previous_callout = code; /* Save for later completion */
5738 after_manual_callout = 1; /* Skip one item before completing */
5739 *code++ = OP_CALLOUT;
5740 {
5741 int n = 0;
5742 ptr++;
5743 while(IS_DIGIT(*ptr))
5744 n = n * 10 + *ptr++ - CHAR_0;
5745 if (*ptr != CHAR_RIGHT_PARENTHESIS)
5746 {
5747 *errorcodeptr = ERR39;
5748 goto FAILED;
5749 }
5750 if (n > 255)
5751 {
5752 *errorcodeptr = ERR38;
5753 goto FAILED;
5754 }
5755 *code++ = n;
5756 PUT(code, 0, (int)(ptr - cd->start_pattern + 1)); /* Pattern offset */
5757 PUT(code, LINK_SIZE, 0); /* Default length */
5758 code += 2 * LINK_SIZE;
5759 }
5760 previous = NULL;
5761 continue;
5762
5763
5764 /* ------------------------------------------------------------ */
5765 case CHAR_P: /* Python-style named subpattern handling */
5766 if (*(++ptr) == CHAR_EQUALS_SIGN ||
5767 *ptr == CHAR_GREATER_THAN_SIGN) /* Reference or recursion */
5768 {
5769 is_recurse = *ptr == CHAR_GREATER_THAN_SIGN;
5770 terminator = CHAR_RIGHT_PARENTHESIS;
5771 goto NAMED_REF_OR_RECURSE;
5772 }
5773 else if (*ptr != CHAR_LESS_THAN_SIGN) /* Test for Python-style defn */
5774 {
5775 *errorcodeptr = ERR41;
5776 goto FAILED;
5777 }
5778 /* Fall through to handle (?P< as (?< is handled */
5779
5780
5781 /* ------------------------------------------------------------ */
5782 DEFINE_NAME: /* Come here from (?< handling */
5783 case CHAR_APOSTROPHE:
5784 {
5785 terminator = (*ptr == CHAR_LESS_THAN_SIGN)?
5786 CHAR_GREATER_THAN_SIGN : CHAR_APOSTROPHE;
5787 name = ++ptr;
5788
5789 while ((cd->ctypes[*ptr] & ctype_word) != 0) ptr++;
5790 namelen = (int)(ptr - name);
5791
5792 /* In the pre-compile phase, just do a syntax check. */
5793
5794 if (lengthptr != NULL)
5795 {
5796 if (*ptr != terminator)
5797 {
5798 *errorcodeptr = ERR42;
5799 goto FAILED;
5800 }
5801 if (cd->names_found >= MAX_NAME_COUNT)
5802 {
5803 *errorcodeptr = ERR49;
5804 goto FAILED;
5805 }
5806 if (namelen + 3 > cd->name_entry_size)
5807 {
5808 cd->name_entry_size = namelen + 3;
5809 if (namelen > MAX_NAME_SIZE)
5810 {
5811 *errorcodeptr = ERR48;
5812 goto FAILED;
5813 }
5814 }
5815 }
5816
5817 /* In the real compile, create the entry in the table, maintaining
5818 alphabetical order. Duplicate names for different numbers are
5819 permitted only if PCRE_DUPNAMES is set. Duplicate names for the same
5820 number are always OK. (An existing number can be re-used if (?|
5821 appears in the pattern.) In either event, a duplicate name results in
5822 a duplicate entry in the table, even if the number is the same. This
5823 is because the number of names, and hence the table size, is computed
5824 in the pre-compile, and it affects various numbers and pointers which
5825 would all have to be modified, and the compiled code moved down, if
5826 duplicates with the same number were omitted from the table. This
5827 doesn't seem worth the hassle. However, *different* names for the
5828 same number are not permitted. */
5829
5830 else
5831 {
5832 BOOL dupname = FALSE;
5833 slot = cd->name_table;
5834
5835 for (i = 0; i < cd->names_found; i++)
5836 {
5837 int crc = memcmp(name, slot+2, namelen);
5838 if (crc == 0)
5839 {
5840 if (slot[2+namelen] == 0)
5841 {
5842 if (GET2(slot, 0) != cd->bracount + 1 &&
5843 (options & PCRE_DUPNAMES) == 0)
5844 {
5845 *errorcodeptr = ERR43;
5846 goto FAILED;
5847 }
5848 else dupname = TRUE;
5849 }
5850 else crc = -1; /* Current name is a substring */
5851 }
5852
5853 /* Make space in the table and break the loop for an earlier
5854 name. For a duplicate or later name, carry on. We do this for
5855 duplicates so that in the simple case (when ?(| is not used) they
5856 are in order of their numbers. */
5857
5858 if (crc < 0)
5859 {
5860 memmove(slot + cd->name_entry_size, slot,
5861 IN_UCHARS((cd->names_found - i) * cd->name_entry_size));
5862 break;
5863 }
5864
5865 /* Continue the loop for a later or duplicate name */
5866
5867 slot += cd->name_entry_size;
5868 }
5869
5870 /* For non-duplicate names, check for a duplicate number before
5871 adding the new name. */
5872
5873 if (!dupname)
5874 {
5875 pcre_uchar *cslot = cd->name_table;
5876 for (i = 0; i < cd->names_found; i++)
5877 {
5878 if (cslot != slot)
5879 {
5880 if (GET2(cslot, 0) == cd->bracount + 1)
5881 {
5882 *errorcodeptr = ERR65;
5883 goto FAILED;
5884 }
5885 }
5886 else i--;
5887 cslot += cd->name_entry_size;
5888 }
5889 }
5890
5891 PUT2(slot, 0, cd->bracount + 1);
5892 memcpy(slot + 2, name, IN_UCHARS(namelen));
5893 slot[2 + namelen] = 0;
5894 }
5895 }
5896
5897 /* In both pre-compile and compile, count the number of names we've
5898 encountered. */
5899
5900 cd->names_found++;
5901 ptr++; /* Move past > or ' */
5902 goto NUMBERED_GROUP;
5903
5904
5905 /* ------------------------------------------------------------ */
5906 case CHAR_AMPERSAND: /* Perl recursion/subroutine syntax */
5907 terminator = CHAR_RIGHT_PARENTHESIS;
5908 is_recurse = TRUE;
5909 /* Fall through */
5910
5911 /* We come here from the Python syntax above that handles both
5912 references (?P=name) and recursion (?P>name), as well as falling
5913 through from the Perl recursion syntax (?&name). We also come here from
5914 the Perl \k<name> or \k'name' back reference syntax and the \k{name}
5915 .NET syntax, and the Oniguruma \g<...> and \g'...' subroutine syntax. */
5916
5917 NAMED_REF_OR_RECURSE:
5918 name = ++ptr;
5919 while ((cd->ctypes[*ptr] & ctype_word) != 0) ptr++;
5920 namelen = (int)(ptr - name);
5921
5922 /* In the pre-compile phase, do a syntax check. We used to just set
5923 a dummy reference number, because it was not used in the first pass.
5924 However, with the change of recursive back references to be atomic,
5925 we have to look for the number so that this state can be identified, as
5926 otherwise the incorrect length is computed. If it's not a backwards
5927 reference, the dummy number will do. */
5928
5929 if (lengthptr != NULL)
5930 {
5931 const pcre_uchar *temp;
5932
5933 if (namelen == 0)
5934 {
5935 *errorcodeptr = ERR62;
5936 goto FAILED;
5937 }
5938 if (*ptr != terminator)
5939 {
5940 *errorcodeptr = ERR42;
5941 goto FAILED;
5942 }
5943 if (namelen > MAX_NAME_SIZE)
5944 {
5945 *errorcodeptr = ERR48;
5946 goto FAILED;
5947 }
5948
5949 /* The name table does not exist in the first pass, so we cannot
5950 do a simple search as in the code below. Instead, we have to scan the
5951 pattern to find the number. It is important that we scan it only as
5952 far as we have got because the syntax of named subpatterns has not
5953 been checked for the rest of the pattern, and find_parens() assumes
5954 correct syntax. In any case, it's a waste of resources to scan
5955 further. We stop the scan at the current point by temporarily
5956 adjusting the value of cd->endpattern. */
5957
5958 temp = cd->end_pattern;
5959 cd->end_pattern = ptr;
5960 recno = find_parens(cd, name, namelen,
5961 (options & PCRE_EXTENDED) != 0, utf8);
5962 cd->end_pattern = temp;
5963 if (recno < 0) recno = 0; /* Forward ref; set dummy number */
5964 }
5965
5966 /* In the real compile, seek the name in the table. We check the name
5967 first, and then check that we have reached the end of the name in the
5968 table. That way, if the name that is longer than any in the table,
5969 the comparison will fail without reading beyond the table entry. */
5970
5971 else
5972 {
5973 slot = cd->name_table;
5974 for (i = 0; i < cd->names_found; i++)
5975 {
5976 if (STRNCMP_UC_UC(name, slot+IMM2_SIZE, namelen) == 0 &&
5977 slot[2+namelen] == 0)
5978 break;
5979 slot += cd->name_entry_size;
5980 }
5981
5982 if (i < cd->names_found) /* Back reference */
5983 {
5984 recno = GET2(slot, 0);
5985 }
5986 else if ((recno = /* Forward back reference */
5987 find_parens(cd, name, namelen,
5988 (options & PCRE_EXTENDED) != 0, utf8)) <= 0)
5989 {
5990 *errorcodeptr = ERR15;
5991 goto FAILED;
5992 }
5993 }
5994
5995 /* In both phases, we can now go to the code than handles numerical
5996 recursion or backreferences. */
5997
5998 if (is_recurse) goto HANDLE_RECURSION;
5999 else goto HANDLE_REFERENCE;
6000
6001
6002 /* ------------------------------------------------------------ */
6003 case CHAR_R: /* Recursion */
6004 ptr++; /* Same as (?0) */
6005 /* Fall through */
6006
6007
6008 /* ------------------------------------------------------------ */
6009 case CHAR_MINUS: case CHAR_PLUS: /* Recursion or subroutine */
6010 case CHAR_0: case CHAR_1: case CHAR_2: case CHAR_3: case CHAR_4:
6011 case CHAR_5: case CHAR_6: case CHAR_7: case CHAR_8: case CHAR_9:
6012 {
6013 const pcre_uchar *called;
6014 terminator = CHAR_RIGHT_PARENTHESIS;
6015
6016 /* Come here from the \g<...> and \g'...' code (Oniguruma
6017 compatibility). However, the syntax has been checked to ensure that
6018 the ... are a (signed) number, so that neither ERR63 nor ERR29 will
6019 be called on this path, nor with the jump to OTHER_CHAR_AFTER_QUERY
6020 ever be taken. */
6021
6022 HANDLE_NUMERICAL_RECURSION:
6023
6024 if ((refsign = *ptr) == CHAR_PLUS)
6025 {
6026 ptr++;
6027 if (!IS_DIGIT(*ptr))
6028 {
6029 *errorcodeptr = ERR63;
6030 goto FAILED;
6031 }
6032 }
6033 else if (refsign == CHAR_MINUS)
6034 {
6035 if (!IS_DIGIT(ptr[1]))
6036 goto OTHER_CHAR_AFTER_QUERY;
6037 ptr++;
6038 }
6039
6040 recno = 0;
6041 while(IS_DIGIT(*ptr))
6042 recno = recno * 10 + *ptr++ - CHAR_0;
6043
6044 if (*ptr != terminator)
6045 {
6046 *errorcodeptr = ERR29;
6047 goto FAILED;
6048 }
6049
6050 if (refsign == CHAR_MINUS)
6051 {
6052 if (recno == 0)
6053 {
6054 *errorcodeptr = ERR58;
6055 goto FAILED;
6056 }
6057 recno = cd->bracount - recno + 1;
6058 if (recno <= 0)
6059 {
6060 *errorcodeptr = ERR15;
6061 goto FAILED;
6062 }
6063 }
6064 else if (refsign == CHAR_PLUS)
6065 {
6066 if (recno == 0)
6067 {
6068 *errorcodeptr = ERR58;
6069 goto FAILED;
6070 }
6071 recno += cd->bracount;
6072 }
6073
6074 /* Come here from code above that handles a named recursion */
6075
6076 HANDLE_RECURSION:
6077
6078 previous = code;
6079 called = cd->start_code;
6080
6081 /* When we are actually compiling, find the bracket that is being
6082 referenced. Temporarily end the regex in case it doesn't exist before
6083 this point. If we end up with a forward reference, first check that
6084 the bracket does occur later so we can give the error (and position)
6085 now. Then remember this forward reference in the workspace so it can
6086 be filled in at the end. */
6087
6088 if (lengthptr == NULL)
6089 {
6090 *code = OP_END;
6091 if (recno != 0)
6092 called = PRIV(find_bracket)(cd->start_code, utf8, recno);
6093
6094 /* Forward reference */
6095
6096 if (called == NULL)
6097 {
6098 if (find_parens(cd, NULL, recno,
6099 (options & PCRE_EXTENDED) != 0, utf8) < 0)
6100 {
6101 *errorcodeptr = ERR15;
6102 goto FAILED;
6103 }
6104
6105 /* Fudge the value of "called" so that when it is inserted as an
6106 offset below, what it actually inserted is the reference number
6107 of the group. Then remember the forward reference. */
6108
6109 called = cd->start_code + recno;
6110 PUTINC(cd->hwm, 0, (int)(code + 1 - cd->start_code));
6111 }
6112
6113 /* If not a forward reference, and the subpattern is still open,
6114 this is a recursive call. We check to see if this is a left
6115 recursion that could loop for ever, and diagnose that case. We
6116 must not, however, do this check if we are in a conditional
6117 subpattern because the condition might be testing for recursion in
6118 a pattern such as /(?(R)a+|(?R)b)/, which is perfectly valid.
6119 Forever loops are also detected at runtime, so those that occur in
6120 conditional subpatterns will be picked up then. */
6121
6122 else if (GET(called, 1) == 0 && cond_depth <= 0 &&
6123 could_be_empty(called, code, bcptr, utf8, cd))
6124 {
6125 *errorcodeptr = ERR40;
6126 goto FAILED;
6127 }
6128 }
6129
6130 /* Insert the recursion/subroutine item. */
6131
6132 *code = OP_RECURSE;
6133 PUT(code, 1, (int)(called - cd->start_code));
6134 code += 1 + LINK_SIZE;
6135 }
6136
6137 /* Can't determine a first byte now */
6138
6139 if (firstchar == REQ_UNSET) firstchar = REQ_NONE;
6140 continue;
6141
6142
6143 /* ------------------------------------------------------------ */
6144 default: /* Other characters: check option setting */
6145 OTHER_CHAR_AFTER_QUERY:
6146 set = unset = 0;
6147 optset = &set;
6148
6149 while (*ptr != CHAR_RIGHT_PARENTHESIS && *ptr != CHAR_COLON)
6150 {
6151 switch (*ptr++)
6152 {
6153 case CHAR_MINUS: optset = &unset; break;
6154
6155 case CHAR_J: /* Record that it changed in the external options */
6156 *optset |= PCRE_DUPNAMES;
6157 cd->external_flags |= PCRE_JCHANGED;
6158 break;
6159
6160 case CHAR_i: *optset |= PCRE_CASELESS; break;
6161 case CHAR_m: *optset |= PCRE_MULTILINE; break;
6162 case CHAR_s: *optset |= PCRE_DOTALL; break;
6163 case CHAR_x: *optset |= PCRE_EXTENDED; break;
6164 case CHAR_U: *optset |= PCRE_UNGREEDY; break;
6165 case CHAR_X: *optset |= PCRE_EXTRA; break;
6166
6167 default: *errorcodeptr = ERR12;
6168 ptr--; /* Correct the offset */
6169 goto FAILED;
6170 }
6171 }
6172
6173 /* Set up the changed option bits, but don't change anything yet. */
6174
6175 newoptions = (options | set) & (~unset);
6176
6177 /* If the options ended with ')' this is not the start of a nested
6178 group with option changes, so the options change at this level. If this
6179 item is right at the start of the pattern, the options can be
6180 abstracted and made external in the pre-compile phase, and ignored in
6181 the compile phase. This can be helpful when matching -- for instance in
6182 caseless checking of required bytes.
6183
6184 If the code pointer is not (cd->start_code + 1 + LINK_SIZE), we are
6185 definitely *not* at the start of the pattern because something has been
6186 compiled. In the pre-compile phase, however, the code pointer can have
6187 that value after the start, because it gets reset as code is discarded
6188 during the pre-compile. However, this can happen only at top level - if
6189 we are within parentheses, the starting BRA will still be present. At
6190 any parenthesis level, the length value can be used to test if anything
6191 has been compiled at that level. Thus, a test for both these conditions
6192 is necessary to ensure we correctly detect the start of the pattern in
6193 both phases.
6194
6195 If we are not at the pattern start, reset the greedy defaults and the
6196 case value for firstchar and reqchar. */
6197
6198 if (*ptr == CHAR_RIGHT_PARENTHESIS)
6199 {
6200 if (code == cd->start_code + 1 + LINK_SIZE &&
6201 (lengthptr == NULL || *lengthptr == 2 + 2*LINK_SIZE))
6202 {
6203 cd->external_options = newoptions;
6204 }
6205 else
6206 {
6207 greedy_default = ((newoptions & PCRE_UNGREEDY) != 0);
6208 greedy_non_default = greedy_default ^ 1;
6209 req_caseopt = ((newoptions & PCRE_CASELESS) != 0)? REQ_CASELESS:0;
6210 }
6211
6212 /* Change options at this level, and pass them back for use
6213 in subsequent branches. */
6214
6215 *optionsptr = options = newoptions;
6216 previous = NULL; /* This item can't be repeated */
6217 continue; /* It is complete */
6218 }
6219
6220 /* If the options ended with ':' we are heading into a nested group
6221 with possible change of options. Such groups are non-capturing and are
6222 not assertions of any kind. All we need to do is skip over the ':';
6223 the newoptions value is handled below. */
6224
6225 bravalue = OP_BRA;
6226 ptr++;
6227 } /* End of switch for character following (? */
6228 } /* End of (? handling */
6229
6230 /* Opening parenthesis not followed by '*' or '?'. If PCRE_NO_AUTO_CAPTURE
6231 is set, all unadorned brackets become non-capturing and behave like (?:...)
6232 brackets. */
6233
6234 else if ((options & PCRE_NO_AUTO_CAPTURE) != 0)
6235 {
6236 bravalue = OP_BRA;
6237 }
6238
6239 /* Else we have a capturing group. */
6240
6241 else
6242 {
6243 NUMBERED_GROUP:
6244 cd->bracount += 1;
6245 PUT2(code, 1+LINK_SIZE, cd->bracount);
6246 skipbytes = IMM2_SIZE;
6247 }
6248
6249 /* Process nested bracketed regex. Assertions used not to be repeatable,
6250 but this was changed for Perl compatibility, so all kinds can now be
6251 repeated. We copy code into a non-register variable (tempcode) in order to
6252 be able to pass its address because some compilers complain otherwise. */
6253
6254 previous = code; /* For handling repetition */
6255 *code = bravalue;
6256 tempcode = code;
6257 tempreqvary = cd->req_varyopt; /* Save value before bracket */
6258 tempbracount = cd->bracount; /* Save value before bracket */
6259 length_prevgroup = 0; /* Initialize for pre-compile phase */
6260
6261 if (!compile_regex(
6262 newoptions, /* The complete new option state */
6263 &tempcode, /* Where to put code (updated) */
6264 &ptr, /* Input pointer (updated) */
6265 errorcodeptr, /* Where to put an error message */
6266 (bravalue == OP_ASSERTBACK ||
6267 bravalue == OP_ASSERTBACK_NOT), /* TRUE if back assert */
6268 reset_bracount, /* True if (?| group */
6269 skipbytes, /* Skip over bracket number */
6270 cond_depth +
6271 ((bravalue == OP_COND)?1:0), /* Depth of condition subpatterns */
6272 &subfirstchar, /* For possible first char */
6273 &subreqchar, /* For possible last char */
6274 bcptr, /* Current branch chain */
6275 cd, /* Tables block */
6276 (lengthptr == NULL)? NULL : /* Actual compile phase */
6277 &length_prevgroup /* Pre-compile phase */
6278 ))
6279 goto FAILED;
6280
6281 /* If this was an atomic group and there are no capturing groups within it,
6282 generate OP_ONCE_NC instead of OP_ONCE. */
6283
6284 if (bravalue == OP_ONCE && cd->bracount <= tempbracount)
6285 *code = OP_ONCE_NC;
6286
6287 if (bravalue >= OP_ASSERT && bravalue <= OP_ASSERTBACK_NOT)
6288 cd->assert_depth -= 1;
6289
6290 /* At the end of compiling, code is still pointing to the start of the
6291 group, while tempcode has been updated to point past the end of the group.
6292 The pattern pointer (ptr) is on the bracket.
6293
6294 If this is a conditional bracket, check that there are no more than
6295 two branches in the group, or just one if it's a DEFINE group. We do this
6296 in the real compile phase, not in the pre-pass, where the whole group may
6297 not be available. */
6298
6299 if (bravalue == OP_COND && lengthptr == NULL)
6300 {
6301 pcre_uchar *tc = code;
6302 int condcount = 0;
6303
6304 do {
6305 condcount++;
6306 tc += GET(tc,1);
6307 }
6308 while (*tc != OP_KET);
6309
6310 /* A DEFINE group is never obeyed inline (the "condition" is always
6311 false). It must have only one branch. */
6312
6313 if (code[LINK_SIZE+1] == OP_DEF)
6314 {
6315 if (condcount > 1)
6316 {
6317 *errorcodeptr = ERR54;
6318 goto FAILED;
6319 }
6320 bravalue = OP_DEF; /* Just a flag to suppress char handling below */
6321 }
6322
6323 /* A "normal" conditional group. If there is just one branch, we must not
6324 make use of its firstchar or reqchar, because this is equivalent to an
6325 empty second branch. */
6326
6327 else
6328 {
6329 if (condcount > 2)
6330 {
6331 *errorcodeptr = ERR27;
6332 goto FAILED;
6333 }
6334 if (condcount == 1) subfirstchar = subreqchar = REQ_NONE;
6335 }
6336 }
6337
6338 /* Error if hit end of pattern */
6339
6340 if (*ptr != CHAR_RIGHT_PARENTHESIS)
6341 {
6342 *errorcodeptr = ERR14;
6343 goto FAILED;
6344 }
6345
6346 /* In the pre-compile phase, update the length by the length of the group,
6347 less the brackets at either end. Then reduce the compiled code to just a
6348 set of non-capturing brackets so that it doesn't use much memory if it is
6349 duplicated by a quantifier.*/
6350
6351 if (lengthptr != NULL)
6352 {
6353 if (OFLOW_MAX - *lengthptr < length_prevgroup - 2 - 2*LINK_SIZE)
6354 {
6355 *errorcodeptr = ERR20;
6356 goto FAILED;
6357 }
6358 *lengthptr += length_prevgroup - 2 - 2*LINK_SIZE;
6359 code++; /* This already contains bravalue */
6360 PUTINC(code, 0, 1 + LINK_SIZE);
6361 *code++ = OP_KET;
6362 PUTINC(code, 0, 1 + LINK_SIZE);
6363 break; /* No need to waste time with special character handling */
6364 }
6365
6366 /* Otherwise update the main code pointer to the end of the group. */
6367
6368 code = tempcode;
6369
6370 /* For a DEFINE group, required and first character settings are not
6371 relevant. */
6372
6373 if (bravalue == OP_DEF) break;
6374
6375 /* Handle updating of the required and first characters for other types of
6376 group. Update for normal brackets of all kinds, and conditions with two
6377 branches (see code above). If the bracket is followed by a quantifier with
6378 zero repeat, we have to back off. Hence the definition of zeroreqchar and
6379 zerofirstchar outside the main loop so that they can be accessed for the
6380 back off. */
6381
6382 zeroreqchar = reqchar;
6383 zerofirstchar = firstchar;
6384 groupsetfirstchar = FALSE;
6385
6386 if (bravalue >= OP_ONCE)
6387 {
6388 /* If we have not yet set a firstchar in this branch, take it from the
6389 subpattern, remembering that it was set here so that a repeat of more
6390 than one can replicate it as reqchar if necessary. If the subpattern has
6391 no firstchar, set "none" for the whole branch. In both cases, a zero
6392 repeat forces firstchar to "none". */
6393
6394 if (firstchar == REQ_UNSET)
6395 {
6396 if (subfirstchar >= 0)
6397 {
6398 firstchar = subfirstchar;
6399 groupsetfirstchar = TRUE;
6400 }
6401 else firstchar = REQ_NONE;
6402 zerofirstchar = REQ_NONE;
6403 }
6404
6405 /* If firstchar was previously set, convert the subpattern's firstchar
6406 into reqchar if there wasn't one, using the vary flag that was in
6407 existence beforehand. */
6408
6409 else if (subfirstchar >= 0 && subreqchar < 0)
6410 subreqchar = subfirstchar | tempreqvary;
6411
6412 /* If the subpattern set a required byte (or set a first byte that isn't
6413 really the first byte - see above), set it. */
6414
6415 if (subreqchar >= 0) reqchar = subreqchar;
6416 }
6417
6418 /* For a forward assertion, we take the reqchar, if set. This can be
6419 helpful if the pattern that follows the assertion doesn't set a different
6420 char. For example, it's useful for /(?=abcde).+/. We can't set firstchar
6421 for an assertion, however because it leads to incorrect effect for patterns
6422 such as /(?=a)a.+/ when the "real" "a" would then become a reqchar instead
6423 of a firstchar. This is overcome by a scan at the end if there's no
6424 firstchar, looking for an asserted first char. */
6425
6426 else if (bravalue == OP_ASSERT && subreqchar >= 0) reqchar = subreqchar;
6427 break; /* End of processing '(' */
6428
6429
6430 /* ===================================================================*/
6431 /* Handle metasequences introduced by \. For ones like \d, the ESC_ values
6432 are arranged to be the negation of the corresponding OP_values in the
6433 default case when PCRE_UCP is not set. For the back references, the values
6434 are ESC_REF plus the reference number. Only back references and those types
6435 that consume a character may be repeated. We can test for values between
6436 ESC_b and ESC_Z for the latter; this may have to change if any new ones are
6437 ever created. */
6438
6439 case CHAR_BACKSLASH:
6440 tempptr = ptr;
6441 c = check_escape(&ptr, errorcodeptr, cd->bracount, options, FALSE);
6442 if (*errorcodeptr != 0) goto FAILED;
6443
6444 if (c < 0)
6445 {
6446 if (-c == ESC_Q) /* Handle start of quoted string */
6447 {
6448 if (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
6449 ptr += 2; /* avoid empty string */
6450 else inescq = TRUE;
6451 continue;
6452 }
6453
6454 if (-c == ESC_E) continue; /* Perl ignores an orphan \E */
6455
6456 /* For metasequences that actually match a character, we disable the
6457 setting of a first character if it hasn't already been set. */
6458
6459 if (firstchar == REQ_UNSET && -c > ESC_b && -c < ESC_Z)
6460 firstchar = REQ_NONE;
6461
6462 /* Set values to reset to if this is followed by a zero repeat. */
6463
6464 zerofirstchar = firstchar;
6465 zeroreqchar = reqchar;
6466
6467 /* \g<name> or \g'name' is a subroutine call by name and \g<n> or \g'n'
6468 is a subroutine call by number (Oniguruma syntax). In fact, the value
6469 -ESC_g is returned only for these cases. So we don't need to check for <
6470 or ' if the value is -ESC_g. For the Perl syntax \g{n} the value is
6471 -ESC_REF+n, and for the Perl syntax \g{name} the result is -ESC_k (as
6472 that is a synonym for a named back reference). */
6473
6474 if (-c == ESC_g)
6475 {
6476 const pcre_uchar *p;
6477 save_hwm = cd->hwm; /* Normally this is set when '(' is read */
6478 terminator = (*(++ptr) == CHAR_LESS_THAN_SIGN)?
6479 CHAR_GREATER_THAN_SIGN : CHAR_APOSTROPHE;
6480
6481 /* These two statements stop the compiler for warning about possibly
6482 unset variables caused by the jump to HANDLE_NUMERICAL_RECURSION. In
6483 fact, because we actually check for a number below, the paths that
6484 would actually be in error are never taken. */
6485
6486 skipbytes = 0;
6487 reset_bracount = FALSE;
6488
6489 /* Test for a name */
6490
6491 if (ptr[1] != CHAR_PLUS && ptr[1] != CHAR_MINUS)
6492 {
6493 BOOL isnumber = TRUE;
6494 for (p = ptr + 1; *p != 0 && *p != terminator; p++)
6495 {
6496 if ((cd->ctypes[*p] & ctype_digit) == 0) isnumber = FALSE;
6497 if ((cd->ctypes[*p] & ctype_word) == 0) break;
6498 }
6499 if (*p != terminator)
6500 {
6501 *errorcodeptr = ERR57;
6502 break;
6503 }
6504 if (isnumber)
6505 {
6506 ptr++;
6507 goto HANDLE_NUMERICAL_RECURSION;
6508 }
6509 is_recurse = TRUE;
6510 goto NAMED_REF_OR_RECURSE;
6511 }
6512
6513 /* Test a signed number in angle brackets or quotes. */
6514
6515 p = ptr + 2;
6516 while (IS_DIGIT(*p)) p++;
6517 if (*p != terminator)
6518 {
6519 *errorcodeptr = ERR57;
6520 break;
6521 }
6522 ptr++;
6523 goto HANDLE_NUMERICAL_RECURSION;
6524 }
6525
6526 /* \k<name> or \k'name' is a back reference by name (Perl syntax).
6527 We also support \k{name} (.NET syntax). */
6528
6529 if (-c == ESC_k)
6530 {
6531 if ((ptr[1] != CHAR_LESS_THAN_SIGN &&
6532 ptr[1] != CHAR_APOSTROPHE && ptr[1] != CHAR_LEFT_CURLY_BRACKET))
6533 {
6534 *errorcodeptr = ERR69;
6535 break;
6536 }
6537 is_recurse = FALSE;
6538 terminator = (*(++ptr) == CHAR_LESS_THAN_SIGN)?
6539 CHAR_GREATER_THAN_SIGN : (*ptr == CHAR_APOSTROPHE)?
6540 CHAR_APOSTROPHE : CHAR_RIGHT_CURLY_BRACKET;
6541 goto NAMED_REF_OR_RECURSE;
6542 }
6543
6544 /* Back references are handled specially; must disable firstchar if
6545 not set to cope with cases like (?=(\w+))\1: which would otherwise set
6546 ':' later. */
6547
6548 if (-c >= ESC_REF)
6549 {
6550 open_capitem *oc;
6551 recno = -c - ESC_REF;
6552
6553 HANDLE_REFERENCE: /* Come here from named backref handling */
6554 if (firstchar == REQ_UNSET) firstchar = REQ_NONE;
6555 previous = code;
6556 *code++ = ((options & PCRE_CASELESS) != 0)? OP_REFI : OP_REF;
6557 PUT2INC(code, 0, recno);
6558 cd->backref_map |= (recno < 32)? (1 << recno) : 1;
6559 if (recno > cd->top_backref) cd->top_backref = recno;
6560
6561 /* Check to see if this back reference is recursive, that it, it
6562 is inside the group that it references. A flag is set so that the
6563 group can be made atomic. */
6564
6565 for (oc = cd->open_caps; oc != NULL; oc = oc->next)
6566 {
6567 if (oc->number == recno)
6568 {
6569 oc->flag = TRUE;
6570 break;
6571 }
6572 }
6573 }
6574
6575 /* So are Unicode property matches, if supported. */
6576
6577 #ifdef SUPPORT_UCP
6578 else if (-c == ESC_P || -c == ESC_p)
6579 {
6580 BOOL negated;
6581 int pdata;
6582 int ptype = get_ucp(&ptr, &negated, &pdata, errorcodeptr);
6583 if (ptype < 0) goto FAILED;
6584 previous = code;
6585 *code++ = ((-c == ESC_p) != negated)? OP_PROP : OP_NOTPROP;
6586 *code++ = ptype;
6587 *code++ = pdata;
6588 }
6589 #else
6590
6591 /* If Unicode properties are not supported, \X, \P, and \p are not
6592 allowed. */
6593
6594 else if (-c == ESC_X || -c == ESC_P || -c == ESC_p)
6595 {
6596 *errorcodeptr = ERR45;
6597 goto FAILED;
6598 }
6599 #endif
6600
6601 /* For the rest (including \X when Unicode properties are supported), we
6602 can obtain the OP value by negating the escape value in the default
6603 situation when PCRE_UCP is not set. When it *is* set, we substitute
6604 Unicode property tests. */
6605
6606 else
6607 {
6608 #ifdef SUPPORT_UCP
6609 if (-c >= ESC_DU && -c <= ESC_wu)
6610 {
6611 nestptr = ptr + 1; /* Where to resume */
6612 ptr = substitutes[-c - ESC_DU] - 1; /* Just before substitute */
6613 }
6614 else
6615 #endif
6616 /* In non-UTF-8 mode, we turn \C into OP_ALLANY instead of OP_ANYBYTE
6617 so that it works in DFA mode and in lookbehinds. */
6618
6619 {
6620 previous = (-c > ESC_b && -c < ESC_Z)? code : NULL;
6621 *code++ = (!utf8 && c == -ESC_C)? OP_ALLANY : -c;
6622 }
6623 }
6624 continue;
6625 }
6626
6627 /* We have a data character whose value is in c. In UTF-8 mode it may have
6628 a value > 127. We set its representation in the length/buffer, and then
6629 handle it as a data character. */
6630
6631 #ifdef SUPPORT_UTF8
6632 if (utf8 && c > 127)
6633 mclength = PRIV(ord2utf8)(c, mcbuffer);
6634 else
6635 #endif
6636
6637 {
6638 mcbuffer[0] = c;
6639 mclength = 1;
6640 }
6641 goto ONE_CHAR;
6642
6643
6644 /* ===================================================================*/