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

Contents of /code/trunk/pcre_compile.c

Parent Directory Parent Directory | Revision Log Revision Log


Revision 1048 - (show annotations)
Fri Sep 28 16:03:14 2012 UTC (7 years ago) by ph10
File MIME type: text/plain
File size: 269284 byte(s)
Fix bugs in new caseless support code in the auto-possessifying function.
1 /*************************************************
2 * Perl-Compatible Regular Expressions *
3 *************************************************/
4
5 /* PCRE is a library of functions to support regular expressions whose syntax
6 and semantics are as close as possible to those of the Perl 5 language.
7
8 Written by Philip Hazel
9 Copyright (c) 1997-2012 University of Cambridge
10
11 -----------------------------------------------------------------------------
12 Redistribution and use in source and binary forms, with or without
13 modification, are permitted provided that the following conditions are met:
14
15 * Redistributions of source code must retain the above copyright notice,
16 this list of conditions and the following disclaimer.
17
18 * Redistributions in binary form must reproduce the above copyright
19 notice, this list of conditions and the following disclaimer in the
20 documentation and/or other materials provided with the distribution.
21
22 * Neither the name of the University of Cambridge nor the names of its
23 contributors may be used to endorse or promote products derived from
24 this software without specific prior written permission.
25
26 THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
27 AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28 IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29 ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE
30 LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
31 CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
32 SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
33 INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
34 CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
35 ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
36 POSSIBILITY OF SUCH DAMAGE.
37 -----------------------------------------------------------------------------
38 */
39
40
41 /* This module contains the external function pcre_compile(), along with
42 supporting internal functions that are not used by other modules. */
43
44
45 #ifdef HAVE_CONFIG_H
46 #include "config.h"
47 #endif
48
49 #define NLBLOCK cd /* Block containing newline information */
50 #define PSSTART start_pattern /* Field containing processed string start */
51 #define PSEND end_pattern /* Field containing processed string end */
52
53 #include "pcre_internal.h"
54
55
56 /* When PCRE_DEBUG is defined, we need the pcre(16)_printint() function, which
57 is also used by pcretest. PCRE_DEBUG is not defined when building a production
58 library. We do not need to select pcre16_printint.c specially, because the
59 COMPILE_PCREx macro will already be appropriately set. */
60
61 #ifdef PCRE_DEBUG
62 /* pcre_printint.c should not include any headers */
63 #define PCRE_INCLUDED
64 #include "pcre_printint.c"
65 #undef PCRE_INCLUDED
66 #endif
67
68
69 /* Macro for setting individual bits in class bitmaps. */
70
71 #define SETBIT(a,b) a[(b)/8] |= (1 << ((b)&7))
72
73 /* Maximum length value to check against when making sure that the integer that
74 holds the compiled pattern length does not overflow. We make it a bit less than
75 INT_MAX to allow for adding in group terminating bytes, so that we don't have
76 to check them every time. */
77
78 #define OFLOW_MAX (INT_MAX - 20)
79
80 /* Definitions to allow mutual recursion */
81
82 static int
83 add_list_to_class(pcre_uint8 *, pcre_uchar **, int, compile_data *,
84 const pcre_uint32 *, unsigned int);
85
86 static BOOL
87 compile_regex(int, pcre_uchar **, const pcre_uchar **, int *, BOOL, BOOL,
88 int, int, int *, int *, branch_chain *, compile_data *, int *);
89
90
91
92 /*************************************************
93 * Code parameters and static tables *
94 *************************************************/
95
96 /* This value specifies the size of stack workspace that is used during the
97 first pre-compile phase that determines how much memory is required. The regex
98 is partly compiled into this space, but the compiled parts are discarded as
99 soon as they can be, so that hopefully there will never be an overrun. The code
100 does, however, check for an overrun. The largest amount I've seen used is 218,
101 so this number is very generous.
102
103 The same workspace is used during the second, actual compile phase for
104 remembering forward references to groups so that they can be filled in at the
105 end. Each entry in this list occupies LINK_SIZE bytes, so even when LINK_SIZE
106 is 4 there is plenty of room for most patterns. However, the memory can get
107 filled up by repetitions of forward references, for example patterns like
108 /(?1){0,1999}(b)/, and one user did hit the limit. The code has been changed so
109 that the workspace is expanded using malloc() in this situation. The value
110 below is therefore a minimum, and we put a maximum on it for safety. The
111 minimum is now also defined in terms of LINK_SIZE so that the use of malloc()
112 kicks in at the same number of forward references in all cases. */
113
114 #define COMPILE_WORK_SIZE (2048*LINK_SIZE)
115 #define COMPILE_WORK_SIZE_MAX (100*COMPILE_WORK_SIZE)
116
117 /* The overrun tests check for a slightly smaller size so that they detect the
118 overrun before it actually does run off the end of the data block. */
119
120 #define WORK_SIZE_SAFETY_MARGIN (100)
121
122 /* Private flags added to firstchar and reqchar. */
123
124 #define REQ_CASELESS 0x10000000l /* Indicates caselessness */
125 #define REQ_VARY 0x20000000l /* Reqchar followed non-literal item */
126
127 /* Repeated character flags. */
128
129 #define UTF_LENGTH 0x10000000l /* The char contains its length. */
130
131 /* Table for handling escaped characters in the range '0'-'z'. Positive returns
132 are simple data values; negative values are for special things like \d and so
133 on. Zero means further processing is needed (for things like \x), or the escape
134 is invalid. */
135
136 #ifndef EBCDIC
137
138 /* This is the "normal" table for ASCII systems or for EBCDIC systems running
139 in UTF-8 mode. */
140
141 static const short int escapes[] = {
142 0, 0,
143 0, 0,
144 0, 0,
145 0, 0,
146 0, 0,
147 CHAR_COLON, CHAR_SEMICOLON,
148 CHAR_LESS_THAN_SIGN, CHAR_EQUALS_SIGN,
149 CHAR_GREATER_THAN_SIGN, CHAR_QUESTION_MARK,
150 CHAR_COMMERCIAL_AT, -ESC_A,
151 -ESC_B, -ESC_C,
152 -ESC_D, -ESC_E,
153 0, -ESC_G,
154 -ESC_H, 0,
155 0, -ESC_K,
156 0, 0,
157 -ESC_N, 0,
158 -ESC_P, -ESC_Q,
159 -ESC_R, -ESC_S,
160 0, 0,
161 -ESC_V, -ESC_W,
162 -ESC_X, 0,
163 -ESC_Z, CHAR_LEFT_SQUARE_BRACKET,
164 CHAR_BACKSLASH, CHAR_RIGHT_SQUARE_BRACKET,
165 CHAR_CIRCUMFLEX_ACCENT, CHAR_UNDERSCORE,
166 CHAR_GRAVE_ACCENT, 7,
167 -ESC_b, 0,
168 -ESC_d, ESC_e,
169 ESC_f, 0,
170 -ESC_h, 0,
171 0, -ESC_k,
172 0, 0,
173 ESC_n, 0,
174 -ESC_p, 0,
175 ESC_r, -ESC_s,
176 ESC_tee, 0,
177 -ESC_v, -ESC_w,
178 0, 0,
179 -ESC_z
180 };
181
182 #else
183
184 /* This is the "abnormal" table for EBCDIC systems without UTF-8 support. */
185
186 static const short int escapes[] = {
187 /* 48 */ 0, 0, 0, '.', '<', '(', '+', '|',
188 /* 50 */ '&', 0, 0, 0, 0, 0, 0, 0,
189 /* 58 */ 0, 0, '!', '$', '*', ')', ';', '~',
190 /* 60 */ '-', '/', 0, 0, 0, 0, 0, 0,
191 /* 68 */ 0, 0, '|', ',', '%', '_', '>', '?',
192 /* 70 */ 0, 0, 0, 0, 0, 0, 0, 0,
193 /* 78 */ 0, '`', ':', '#', '@', '\'', '=', '"',
194 /* 80 */ 0, 7, -ESC_b, 0, -ESC_d, ESC_e, ESC_f, 0,
195 /* 88 */-ESC_h, 0, 0, '{', 0, 0, 0, 0,
196 /* 90 */ 0, 0, -ESC_k, 'l', 0, ESC_n, 0, -ESC_p,
197 /* 98 */ 0, ESC_r, 0, '}', 0, 0, 0, 0,
198 /* A0 */ 0, '~', -ESC_s, ESC_tee, 0,-ESC_v, -ESC_w, 0,
199 /* A8 */ 0,-ESC_z, 0, 0, 0, '[', 0, 0,
200 /* B0 */ 0, 0, 0, 0, 0, 0, 0, 0,
201 /* B8 */ 0, 0, 0, 0, 0, ']', '=', '-',
202 /* C0 */ '{',-ESC_A, -ESC_B, -ESC_C, -ESC_D,-ESC_E, 0, -ESC_G,
203 /* C8 */-ESC_H, 0, 0, 0, 0, 0, 0, 0,
204 /* D0 */ '}', 0, -ESC_K, 0, 0,-ESC_N, 0, -ESC_P,
205 /* D8 */-ESC_Q,-ESC_R, 0, 0, 0, 0, 0, 0,
206 /* E0 */ '\\', 0, -ESC_S, 0, 0,-ESC_V, -ESC_W, -ESC_X,
207 /* E8 */ 0,-ESC_Z, 0, 0, 0, 0, 0, 0,
208 /* F0 */ 0, 0, 0, 0, 0, 0, 0, 0,
209 /* F8 */ 0, 0, 0, 0, 0, 0, 0, 0
210 };
211 #endif
212
213
214 /* Table of special "verbs" like (*PRUNE). This is a short table, so it is
215 searched linearly. Put all the names into a single string, in order to reduce
216 the number of relocations when a shared library is dynamically linked. The
217 string is built from string macros so that it works in UTF-8 mode on EBCDIC
218 platforms. */
219
220 typedef struct verbitem {
221 int len; /* Length of verb name */
222 int op; /* Op when no arg, or -1 if arg mandatory */
223 int op_arg; /* Op when arg present, or -1 if not allowed */
224 } verbitem;
225
226 static const char verbnames[] =
227 "\0" /* Empty name is a shorthand for MARK */
228 STRING_MARK0
229 STRING_ACCEPT0
230 STRING_COMMIT0
231 STRING_F0
232 STRING_FAIL0
233 STRING_PRUNE0
234 STRING_SKIP0
235 STRING_THEN;
236
237 static const verbitem verbs[] = {
238 { 0, -1, OP_MARK },
239 { 4, -1, OP_MARK },
240 { 6, OP_ACCEPT, -1 },
241 { 6, OP_COMMIT, -1 },
242 { 1, OP_FAIL, -1 },
243 { 4, OP_FAIL, -1 },
244 { 5, OP_PRUNE, OP_PRUNE_ARG },
245 { 4, OP_SKIP, OP_SKIP_ARG },
246 { 4, OP_THEN, OP_THEN_ARG }
247 };
248
249 static const int verbcount = sizeof(verbs)/sizeof(verbitem);
250
251
252 /* Tables of names of POSIX character classes and their lengths. The names are
253 now all in a single string, to reduce the number of relocations when a shared
254 library is dynamically loaded. The list of lengths is terminated by a zero
255 length entry. The first three must be alpha, lower, upper, as this is assumed
256 for handling case independence. */
257
258 static const char posix_names[] =
259 STRING_alpha0 STRING_lower0 STRING_upper0 STRING_alnum0
260 STRING_ascii0 STRING_blank0 STRING_cntrl0 STRING_digit0
261 STRING_graph0 STRING_print0 STRING_punct0 STRING_space0
262 STRING_word0 STRING_xdigit;
263
264 static const pcre_uint8 posix_name_lengths[] = {
265 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 4, 6, 0 };
266
267 /* Table of class bit maps for each POSIX class. Each class is formed from a
268 base map, with an optional addition or removal of another map. Then, for some
269 classes, there is some additional tweaking: for [:blank:] the vertical space
270 characters are removed, and for [:alpha:] and [:alnum:] the underscore
271 character is removed. The triples in the table consist of the base map offset,
272 second map offset or -1 if no second map, and a non-negative value for map
273 addition or a negative value for map subtraction (if there are two maps). The
274 absolute value of the third field has these meanings: 0 => no tweaking, 1 =>
275 remove vertical space characters, 2 => remove underscore. */
276
277 static const int posix_class_maps[] = {
278 cbit_word, cbit_digit, -2, /* alpha */
279 cbit_lower, -1, 0, /* lower */
280 cbit_upper, -1, 0, /* upper */
281 cbit_word, -1, 2, /* alnum - word without underscore */
282 cbit_print, cbit_cntrl, 0, /* ascii */
283 cbit_space, -1, 1, /* blank - a GNU extension */
284 cbit_cntrl, -1, 0, /* cntrl */
285 cbit_digit, -1, 0, /* digit */
286 cbit_graph, -1, 0, /* graph */
287 cbit_print, -1, 0, /* print */
288 cbit_punct, -1, 0, /* punct */
289 cbit_space, -1, 0, /* space */
290 cbit_word, -1, 0, /* word - a Perl extension */
291 cbit_xdigit,-1, 0 /* xdigit */
292 };
293
294 /* Table of substitutes for \d etc when PCRE_UCP is set. The POSIX class
295 substitutes must be in the order of the names, defined above, and there are
296 both positive and negative cases. NULL means no substitute. */
297
298 #ifdef SUPPORT_UCP
299 static const pcre_uchar string_PNd[] = {
300 CHAR_BACKSLASH, CHAR_P, CHAR_LEFT_CURLY_BRACKET,
301 CHAR_N, CHAR_d, CHAR_RIGHT_CURLY_BRACKET, '\0' };
302 static const pcre_uchar string_pNd[] = {
303 CHAR_BACKSLASH, CHAR_p, CHAR_LEFT_CURLY_BRACKET,
304 CHAR_N, CHAR_d, CHAR_RIGHT_CURLY_BRACKET, '\0' };
305 static const pcre_uchar string_PXsp[] = {
306 CHAR_BACKSLASH, CHAR_P, CHAR_LEFT_CURLY_BRACKET,
307 CHAR_X, CHAR_s, CHAR_p, CHAR_RIGHT_CURLY_BRACKET, '\0' };
308 static const pcre_uchar string_pXsp[] = {
309 CHAR_BACKSLASH, CHAR_p, CHAR_LEFT_CURLY_BRACKET,
310 CHAR_X, CHAR_s, CHAR_p, CHAR_RIGHT_CURLY_BRACKET, '\0' };
311 static const pcre_uchar string_PXwd[] = {
312 CHAR_BACKSLASH, CHAR_P, CHAR_LEFT_CURLY_BRACKET,
313 CHAR_X, CHAR_w, CHAR_d, CHAR_RIGHT_CURLY_BRACKET, '\0' };
314 static const pcre_uchar string_pXwd[] = {
315 CHAR_BACKSLASH, CHAR_p, CHAR_LEFT_CURLY_BRACKET,
316 CHAR_X, CHAR_w, CHAR_d, CHAR_RIGHT_CURLY_BRACKET, '\0' };
317
318 static const pcre_uchar *substitutes[] = {
319 string_PNd, /* \D */
320 string_pNd, /* \d */
321 string_PXsp, /* \S */ /* NOTE: Xsp is Perl space */
322 string_pXsp, /* \s */
323 string_PXwd, /* \W */
324 string_pXwd /* \w */
325 };
326
327 static const pcre_uchar string_pL[] = {
328 CHAR_BACKSLASH, CHAR_p, CHAR_LEFT_CURLY_BRACKET,
329 CHAR_L, CHAR_RIGHT_CURLY_BRACKET, '\0' };
330 static const pcre_uchar string_pLl[] = {
331 CHAR_BACKSLASH, CHAR_p, CHAR_LEFT_CURLY_BRACKET,
332 CHAR_L, CHAR_l, CHAR_RIGHT_CURLY_BRACKET, '\0' };
333 static const pcre_uchar string_pLu[] = {
334 CHAR_BACKSLASH, CHAR_p, CHAR_LEFT_CURLY_BRACKET,
335 CHAR_L, CHAR_u, CHAR_RIGHT_CURLY_BRACKET, '\0' };
336 static const pcre_uchar string_pXan[] = {
337 CHAR_BACKSLASH, CHAR_p, CHAR_LEFT_CURLY_BRACKET,
338 CHAR_X, CHAR_a, CHAR_n, CHAR_RIGHT_CURLY_BRACKET, '\0' };
339 static const pcre_uchar string_h[] = {
340 CHAR_BACKSLASH, CHAR_h, '\0' };
341 static const pcre_uchar string_pXps[] = {
342 CHAR_BACKSLASH, CHAR_p, CHAR_LEFT_CURLY_BRACKET,
343 CHAR_X, CHAR_p, CHAR_s, CHAR_RIGHT_CURLY_BRACKET, '\0' };
344 static const pcre_uchar string_PL[] = {
345 CHAR_BACKSLASH, CHAR_P, CHAR_LEFT_CURLY_BRACKET,
346 CHAR_L, CHAR_RIGHT_CURLY_BRACKET, '\0' };
347 static const pcre_uchar string_PLl[] = {
348 CHAR_BACKSLASH, CHAR_P, CHAR_LEFT_CURLY_BRACKET,
349 CHAR_L, CHAR_l, CHAR_RIGHT_CURLY_BRACKET, '\0' };
350 static const pcre_uchar string_PLu[] = {
351 CHAR_BACKSLASH, CHAR_P, CHAR_LEFT_CURLY_BRACKET,
352 CHAR_L, CHAR_u, CHAR_RIGHT_CURLY_BRACKET, '\0' };
353 static const pcre_uchar string_PXan[] = {
354 CHAR_BACKSLASH, CHAR_P, CHAR_LEFT_CURLY_BRACKET,
355 CHAR_X, CHAR_a, CHAR_n, CHAR_RIGHT_CURLY_BRACKET, '\0' };
356 static const pcre_uchar string_H[] = {
357 CHAR_BACKSLASH, CHAR_H, '\0' };
358 static const pcre_uchar string_PXps[] = {
359 CHAR_BACKSLASH, CHAR_P, CHAR_LEFT_CURLY_BRACKET,
360 CHAR_X, CHAR_p, CHAR_s, CHAR_RIGHT_CURLY_BRACKET, '\0' };
361
362 static const pcre_uchar *posix_substitutes[] = {
363 string_pL, /* alpha */
364 string_pLl, /* lower */
365 string_pLu, /* upper */
366 string_pXan, /* alnum */
367 NULL, /* ascii */
368 string_h, /* blank */
369 NULL, /* cntrl */
370 string_pNd, /* digit */
371 NULL, /* graph */
372 NULL, /* print */
373 NULL, /* punct */
374 string_pXps, /* space */ /* NOTE: Xps is POSIX space */
375 string_pXwd, /* word */
376 NULL, /* xdigit */
377 /* Negated cases */
378 string_PL, /* ^alpha */
379 string_PLl, /* ^lower */
380 string_PLu, /* ^upper */
381 string_PXan, /* ^alnum */
382 NULL, /* ^ascii */
383 string_H, /* ^blank */
384 NULL, /* ^cntrl */
385 string_PNd, /* ^digit */
386 NULL, /* ^graph */
387 NULL, /* ^print */
388 NULL, /* ^punct */
389 string_PXps, /* ^space */ /* NOTE: Xps is POSIX space */
390 string_PXwd, /* ^word */
391 NULL /* ^xdigit */
392 };
393 #define POSIX_SUBSIZE (sizeof(posix_substitutes) / sizeof(pcre_uchar *))
394 #endif
395
396 #define STRING(a) # a
397 #define XSTRING(s) STRING(s)
398
399 /* The texts of compile-time error messages. These are "char *" because they
400 are passed to the outside world. Do not ever re-use any error number, because
401 they are documented. Always add a new error instead. Messages marked DEAD below
402 are no longer used. This used to be a table of strings, but in order to reduce
403 the number of relocations needed when a shared library is loaded dynamically,
404 it is now one long string. We cannot use a table of offsets, because the
405 lengths of inserts such as XSTRING(MAX_NAME_SIZE) are not known. Instead, we
406 simply count through to the one we want - this isn't a performance issue
407 because these strings are used only when there is a compilation error.
408
409 Each substring ends with \0 to insert a null character. This includes the final
410 substring, so that the whole string ends with \0\0, which can be detected when
411 counting through. */
412
413 static const char error_texts[] =
414 "no error\0"
415 "\\ at end of pattern\0"
416 "\\c at end of pattern\0"
417 "unrecognized character follows \\\0"
418 "numbers out of order in {} quantifier\0"
419 /* 5 */
420 "number too big in {} quantifier\0"
421 "missing terminating ] for character class\0"
422 "invalid escape sequence in character class\0"
423 "range out of order in character class\0"
424 "nothing to repeat\0"
425 /* 10 */
426 "operand of unlimited repeat could match the empty string\0" /** DEAD **/
427 "internal error: unexpected repeat\0"
428 "unrecognized character after (? or (?-\0"
429 "POSIX named classes are supported only within a class\0"
430 "missing )\0"
431 /* 15 */
432 "reference to non-existent subpattern\0"
433 "erroffset passed as NULL\0"
434 "unknown option bit(s) set\0"
435 "missing ) after comment\0"
436 "parentheses nested too deeply\0" /** DEAD **/
437 /* 20 */
438 "regular expression is too large\0"
439 "failed to get memory\0"
440 "unmatched parentheses\0"
441 "internal error: code overflow\0"
442 "unrecognized character after (?<\0"
443 /* 25 */
444 "lookbehind assertion is not fixed length\0"
445 "malformed number or name after (?(\0"
446 "conditional group contains more than two branches\0"
447 "assertion expected after (?(\0"
448 "(?R or (?[+-]digits must be followed by )\0"
449 /* 30 */
450 "unknown POSIX class name\0"
451 "POSIX collating elements are not supported\0"
452 "this version of PCRE is compiled without UTF support\0"
453 "spare error\0" /** DEAD **/
454 "character value in \\x{...} sequence is too large\0"
455 /* 35 */
456 "invalid condition (?(0)\0"
457 "\\C not allowed in lookbehind assertion\0"
458 "PCRE does not support \\L, \\l, \\N{name}, \\U, or \\u\0"
459 "number after (?C is > 255\0"
460 "closing ) for (?C expected\0"
461 /* 40 */
462 "recursive call could loop indefinitely\0"
463 "unrecognized character after (?P\0"
464 "syntax error in subpattern name (missing terminator)\0"
465 "two named subpatterns have the same name\0"
466 "invalid UTF-8 string\0"
467 /* 45 */
468 "support for \\P, \\p, and \\X has not been compiled\0"
469 "malformed \\P or \\p sequence\0"
470 "unknown property name after \\P or \\p\0"
471 "subpattern name is too long (maximum " XSTRING(MAX_NAME_SIZE) " characters)\0"
472 "too many named subpatterns (maximum " XSTRING(MAX_NAME_COUNT) ")\0"
473 /* 50 */
474 "repeated subpattern is too long\0" /** DEAD **/
475 "octal value is greater than \\377 in 8-bit non-UTF-8 mode\0"
476 "internal error: overran compiling workspace\0"
477 "internal error: previously-checked referenced subpattern not found\0"
478 "DEFINE group contains more than one branch\0"
479 /* 55 */
480 "repeating a DEFINE group is not allowed\0" /** DEAD **/
481 "inconsistent NEWLINE options\0"
482 "\\g is not followed by a braced, angle-bracketed, or quoted name/number or by a plain number\0"
483 "a numbered reference must not be zero\0"
484 "an argument is not allowed for (*ACCEPT), (*FAIL), or (*COMMIT)\0"
485 /* 60 */
486 "(*VERB) not recognized\0"
487 "number is too big\0"
488 "subpattern name expected\0"
489 "digit expected after (?+\0"
490 "] is an invalid data character in JavaScript compatibility mode\0"
491 /* 65 */
492 "different names for subpatterns of the same number are not allowed\0"
493 "(*MARK) must have an argument\0"
494 "this version of PCRE is not compiled with Unicode property support\0"
495 "\\c must be followed by an ASCII character\0"
496 "\\k is not followed by a braced, angle-bracketed, or quoted name\0"
497 /* 70 */
498 "internal error: unknown opcode in find_fixedlength()\0"
499 "\\N is not supported in a class\0"
500 "too many forward references\0"
501 "disallowed Unicode code point (>= 0xd800 && <= 0xdfff)\0"
502 "invalid UTF-16 string\0"
503 /* 75 */
504 "name is too long in (*MARK), (*PRUNE), (*SKIP), or (*THEN)\0"
505 "character value in \\u.... sequence is too large\0"
506 ;
507
508 /* Table to identify digits and hex digits. This is used when compiling
509 patterns. Note that the tables in chartables are dependent on the locale, and
510 may mark arbitrary characters as digits - but the PCRE compiling code expects
511 to handle only 0-9, a-z, and A-Z as digits when compiling. That is why we have
512 a private table here. It costs 256 bytes, but it is a lot faster than doing
513 character value tests (at least in some simple cases I timed), and in some
514 applications one wants PCRE to compile efficiently as well as match
515 efficiently.
516
517 For convenience, we use the same bit definitions as in chartables:
518
519 0x04 decimal digit
520 0x08 hexadecimal digit
521
522 Then we can use ctype_digit and ctype_xdigit in the code. */
523
524 /* Using a simple comparison for decimal numbers rather than a memory read
525 is much faster, and the resulting code is simpler (the compiler turns it
526 into a subtraction and unsigned comparison). */
527
528 #define IS_DIGIT(x) ((x) >= CHAR_0 && (x) <= CHAR_9)
529
530 #ifndef EBCDIC
531
532 /* This is the "normal" case, for ASCII systems, and EBCDIC systems running in
533 UTF-8 mode. */
534
535 static const pcre_uint8 digitab[] =
536 {
537 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 0- 7 */
538 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 8- 15 */
539 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 16- 23 */
540 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 24- 31 */
541 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - ' */
542 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* ( - / */
543 0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c, /* 0 - 7 */
544 0x0c,0x0c,0x00,0x00,0x00,0x00,0x00,0x00, /* 8 - ? */
545 0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* @ - G */
546 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* H - O */
547 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* P - W */
548 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* X - _ */
549 0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* ` - g */
550 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* h - o */
551 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* p - w */
552 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* x -127 */
553 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 128-135 */
554 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 136-143 */
555 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 144-151 */
556 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 152-159 */
557 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 160-167 */
558 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 168-175 */
559 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 176-183 */
560 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 184-191 */
561 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 192-199 */
562 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 200-207 */
563 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 208-215 */
564 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 216-223 */
565 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 224-231 */
566 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 232-239 */
567 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 240-247 */
568 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00};/* 248-255 */
569
570 #else
571
572 /* This is the "abnormal" case, for EBCDIC systems not running in UTF-8 mode. */
573
574 static const pcre_uint8 digitab[] =
575 {
576 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 0- 7 0 */
577 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 8- 15 */
578 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 16- 23 10 */
579 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 24- 31 */
580 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 32- 39 20 */
581 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 40- 47 */
582 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 48- 55 30 */
583 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 56- 63 */
584 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - 71 40 */
585 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 72- | */
586 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* & - 87 50 */
587 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 88- 95 */
588 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - -103 60 */
589 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 104- ? */
590 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 112-119 70 */
591 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 120- " */
592 0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* 128- g 80 */
593 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* h -143 */
594 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 144- p 90 */
595 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* q -159 */
596 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 160- x A0 */
597 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* y -175 */
598 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* ^ -183 B0 */
599 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 184-191 */
600 0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* { - G C0 */
601 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* H -207 */
602 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* } - P D0 */
603 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* Q -223 */
604 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* \ - X E0 */
605 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* Y -239 */
606 0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c, /* 0 - 7 F0 */
607 0x0c,0x0c,0x00,0x00,0x00,0x00,0x00,0x00};/* 8 -255 */
608
609 static const pcre_uint8 ebcdic_chartab[] = { /* chartable partial dup */
610 0x80,0x00,0x00,0x00,0x00,0x01,0x00,0x00, /* 0- 7 */
611 0x00,0x00,0x00,0x00,0x01,0x01,0x00,0x00, /* 8- 15 */
612 0x00,0x00,0x00,0x00,0x00,0x01,0x00,0x00, /* 16- 23 */
613 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 24- 31 */
614 0x00,0x00,0x00,0x00,0x00,0x01,0x00,0x00, /* 32- 39 */
615 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 40- 47 */
616 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 48- 55 */
617 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 56- 63 */
618 0x01,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - 71 */
619 0x00,0x00,0x00,0x80,0x00,0x80,0x80,0x80, /* 72- | */
620 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* & - 87 */
621 0x00,0x00,0x00,0x80,0x80,0x80,0x00,0x00, /* 88- 95 */
622 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - -103 */
623 0x00,0x00,0x00,0x00,0x00,0x10,0x00,0x80, /* 104- ? */
624 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 112-119 */
625 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 120- " */
626 0x00,0x1a,0x1a,0x1a,0x1a,0x1a,0x1a,0x12, /* 128- g */
627 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* h -143 */
628 0x00,0x12,0x12,0x12,0x12,0x12,0x12,0x12, /* 144- p */
629 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* q -159 */
630 0x00,0x00,0x12,0x12,0x12,0x12,0x12,0x12, /* 160- x */
631 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* y -175 */
632 0x80,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* ^ -183 */
633 0x00,0x00,0x80,0x00,0x00,0x00,0x00,0x00, /* 184-191 */
634 0x80,0x1a,0x1a,0x1a,0x1a,0x1a,0x1a,0x12, /* { - G */
635 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* H -207 */
636 0x00,0x12,0x12,0x12,0x12,0x12,0x12,0x12, /* } - P */
637 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* Q -223 */
638 0x00,0x00,0x12,0x12,0x12,0x12,0x12,0x12, /* \ - X */
639 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* Y -239 */
640 0x1c,0x1c,0x1c,0x1c,0x1c,0x1c,0x1c,0x1c, /* 0 - 7 */
641 0x1c,0x1c,0x00,0x00,0x00,0x00,0x00,0x00};/* 8 -255 */
642 #endif
643
644
645
646
647 /*************************************************
648 * Find an error text *
649 *************************************************/
650
651 /* The error texts are now all in one long string, to save on relocations. As
652 some of the text is of unknown length, we can't use a table of offsets.
653 Instead, just count through the strings. This is not a performance issue
654 because it happens only when there has been a compilation error.
655
656 Argument: the error number
657 Returns: pointer to the error string
658 */
659
660 static const char *
661 find_error_text(int n)
662 {
663 const char *s = error_texts;
664 for (; n > 0; n--)
665 {
666 while (*s++ != 0) {};
667 if (*s == 0) return "Error text not found (please report)";
668 }
669 return s;
670 }
671
672
673 /*************************************************
674 * Expand the workspace *
675 *************************************************/
676
677 /* This function is called during the second compiling phase, if the number of
678 forward references fills the existing workspace, which is originally a block on
679 the stack. A larger block is obtained from malloc() unless the ultimate limit
680 has been reached or the increase will be rather small.
681
682 Argument: pointer to the compile data block
683 Returns: 0 if all went well, else an error number
684 */
685
686 static int
687 expand_workspace(compile_data *cd)
688 {
689 pcre_uchar *newspace;
690 int newsize = cd->workspace_size * 2;
691
692 if (newsize > COMPILE_WORK_SIZE_MAX) newsize = COMPILE_WORK_SIZE_MAX;
693 if (cd->workspace_size >= COMPILE_WORK_SIZE_MAX ||
694 newsize - cd->workspace_size < WORK_SIZE_SAFETY_MARGIN)
695 return ERR72;
696
697 newspace = (PUBL(malloc))(IN_UCHARS(newsize));
698 if (newspace == NULL) return ERR21;
699 memcpy(newspace, cd->start_workspace, cd->workspace_size * sizeof(pcre_uchar));
700 cd->hwm = (pcre_uchar *)newspace + (cd->hwm - cd->start_workspace);
701 if (cd->workspace_size > COMPILE_WORK_SIZE)
702 (PUBL(free))((void *)cd->start_workspace);
703 cd->start_workspace = newspace;
704 cd->workspace_size = newsize;
705 return 0;
706 }
707
708
709
710 /*************************************************
711 * Check for counted repeat *
712 *************************************************/
713
714 /* This function is called when a '{' is encountered in a place where it might
715 start a quantifier. It looks ahead to see if it really is a quantifier or not.
716 It is only a quantifier if it is one of the forms {ddd} {ddd,} or {ddd,ddd}
717 where the ddds are digits.
718
719 Arguments:
720 p pointer to the first char after '{'
721
722 Returns: TRUE or FALSE
723 */
724
725 static BOOL
726 is_counted_repeat(const pcre_uchar *p)
727 {
728 if (!IS_DIGIT(*p)) return FALSE;
729 p++;
730 while (IS_DIGIT(*p)) p++;
731 if (*p == CHAR_RIGHT_CURLY_BRACKET) return TRUE;
732
733 if (*p++ != CHAR_COMMA) return FALSE;
734 if (*p == CHAR_RIGHT_CURLY_BRACKET) return TRUE;
735
736 if (!IS_DIGIT(*p)) return FALSE;
737 p++;
738 while (IS_DIGIT(*p)) p++;
739
740 return (*p == CHAR_RIGHT_CURLY_BRACKET);
741 }
742
743
744
745 /*************************************************
746 * Handle escapes *
747 *************************************************/
748
749 /* This function is called when a \ has been encountered. It either returns a
750 positive value for a simple escape such as \n, or a negative value which
751 encodes one of the more complicated things such as \d. A backreference to group
752 n is returned as -(ESC_REF + n); ESC_REF is the highest ESC_xxx macro. When
753 UTF-8 is enabled, a positive value greater than 255 may be returned. On entry,
754 ptr is pointing at the \. On exit, it is on the final character of the escape
755 sequence.
756
757 Arguments:
758 ptrptr points to the pattern position pointer
759 errorcodeptr points to the errorcode variable
760 bracount number of previous extracting brackets
761 options the options bits
762 isclass TRUE if inside a character class
763
764 Returns: zero or positive => a data character
765 negative => a special escape sequence
766 on error, errorcodeptr is set
767 */
768
769 static int
770 check_escape(const pcre_uchar **ptrptr, int *errorcodeptr, int bracount,
771 int options, BOOL isclass)
772 {
773 /* PCRE_UTF16 has the same value as PCRE_UTF8. */
774 BOOL utf = (options & PCRE_UTF8) != 0;
775 const pcre_uchar *ptr = *ptrptr + 1;
776 pcre_int32 c;
777 int i;
778
779 GETCHARINCTEST(c, ptr); /* Get character value, increment pointer */
780 ptr--; /* Set pointer back to the last byte */
781
782 /* If backslash is at the end of the pattern, it's an error. */
783
784 if (c == 0) *errorcodeptr = ERR1;
785
786 /* Non-alphanumerics are literals. For digits or letters, do an initial lookup
787 in a table. A non-zero result is something that can be returned immediately.
788 Otherwise further processing may be required. */
789
790 #ifndef EBCDIC /* ASCII/UTF-8 coding */
791 /* Not alphanumeric */
792 else if (c < CHAR_0 || c > CHAR_z) {}
793 else if ((i = escapes[c - CHAR_0]) != 0) c = i;
794
795 #else /* EBCDIC coding */
796 /* Not alphanumeric */
797 else if (c < CHAR_a || (!MAX_255(c) || (ebcdic_chartab[c] & 0x0E) == 0)) {}
798 else if ((i = escapes[c - 0x48]) != 0) c = i;
799 #endif
800
801 /* Escapes that need further processing, or are illegal. */
802
803 else
804 {
805 const pcre_uchar *oldptr;
806 BOOL braced, negated;
807
808 switch (c)
809 {
810 /* A number of Perl escapes are not handled by PCRE. We give an explicit
811 error. */
812
813 case CHAR_l:
814 case CHAR_L:
815 *errorcodeptr = ERR37;
816 break;
817
818 case CHAR_u:
819 if ((options & PCRE_JAVASCRIPT_COMPAT) != 0)
820 {
821 /* In JavaScript, \u must be followed by four hexadecimal numbers.
822 Otherwise it is a lowercase u letter. */
823 if (MAX_255(ptr[1]) && (digitab[ptr[1]] & ctype_xdigit) != 0
824 && MAX_255(ptr[2]) && (digitab[ptr[2]] & ctype_xdigit) != 0
825 && MAX_255(ptr[3]) && (digitab[ptr[3]] & ctype_xdigit) != 0
826 && MAX_255(ptr[4]) && (digitab[ptr[4]] & ctype_xdigit) != 0)
827 {
828 c = 0;
829 for (i = 0; i < 4; ++i)
830 {
831 register int cc = *(++ptr);
832 #ifndef EBCDIC /* ASCII/UTF-8 coding */
833 if (cc >= CHAR_a) cc -= 32; /* Convert to upper case */
834 c = (c << 4) + cc - ((cc < CHAR_A)? CHAR_0 : (CHAR_A - 10));
835 #else /* EBCDIC coding */
836 if (cc >= CHAR_a && cc <= CHAR_z) cc += 64; /* Convert to upper case */
837 c = (c << 4) + cc - ((cc >= CHAR_0)? CHAR_0 : (CHAR_A - 10));
838 #endif
839 }
840
841 #ifdef COMPILE_PCRE8
842 if (c > (utf ? 0x10ffff : 0xff))
843 #else
844 #ifdef COMPILE_PCRE16
845 if (c > (utf ? 0x10ffff : 0xffff))
846 #endif
847 #endif
848 {
849 *errorcodeptr = ERR76;
850 }
851 else if (utf && c >= 0xd800 && c <= 0xdfff) *errorcodeptr = ERR73;
852 }
853 }
854 else
855 *errorcodeptr = ERR37;
856 break;
857
858 case CHAR_U:
859 /* In JavaScript, \U is an uppercase U letter. */
860 if ((options & PCRE_JAVASCRIPT_COMPAT) == 0) *errorcodeptr = ERR37;
861 break;
862
863 /* In a character class, \g is just a literal "g". Outside a character
864 class, \g must be followed by one of a number of specific things:
865
866 (1) A number, either plain or braced. If positive, it is an absolute
867 backreference. If negative, it is a relative backreference. This is a Perl
868 5.10 feature.
869
870 (2) Perl 5.10 also supports \g{name} as a reference to a named group. This
871 is part of Perl's movement towards a unified syntax for back references. As
872 this is synonymous with \k{name}, we fudge it up by pretending it really
873 was \k.
874
875 (3) For Oniguruma compatibility we also support \g followed by a name or a
876 number either in angle brackets or in single quotes. However, these are
877 (possibly recursive) subroutine calls, _not_ backreferences. Just return
878 the -ESC_g code (cf \k). */
879
880 case CHAR_g:
881 if (isclass) break;
882 if (ptr[1] == CHAR_LESS_THAN_SIGN || ptr[1] == CHAR_APOSTROPHE)
883 {
884 c = -ESC_g;
885 break;
886 }
887
888 /* Handle the Perl-compatible cases */
889
890 if (ptr[1] == CHAR_LEFT_CURLY_BRACKET)
891 {
892 const pcre_uchar *p;
893 for (p = ptr+2; *p != 0 && *p != CHAR_RIGHT_CURLY_BRACKET; p++)
894 if (*p != CHAR_MINUS && !IS_DIGIT(*p)) break;
895 if (*p != 0 && *p != CHAR_RIGHT_CURLY_BRACKET)
896 {
897 c = -ESC_k;
898 break;
899 }
900 braced = TRUE;
901 ptr++;
902 }
903 else braced = FALSE;
904
905 if (ptr[1] == CHAR_MINUS)
906 {
907 negated = TRUE;
908 ptr++;
909 }
910 else negated = FALSE;
911
912 /* The integer range is limited by the machine's int representation. */
913 c = 0;
914 while (IS_DIGIT(ptr[1]))
915 {
916 if (((unsigned int)c) > INT_MAX / 10) /* Integer overflow */
917 {
918 c = -1;
919 break;
920 }
921 c = c * 10 + *(++ptr) - CHAR_0;
922 }
923 if (((unsigned int)c) > INT_MAX) /* Integer overflow */
924 {
925 while (IS_DIGIT(ptr[1]))
926 ptr++;
927 *errorcodeptr = ERR61;
928 break;
929 }
930
931 if (braced && *(++ptr) != CHAR_RIGHT_CURLY_BRACKET)
932 {
933 *errorcodeptr = ERR57;
934 break;
935 }
936
937 if (c == 0)
938 {
939 *errorcodeptr = ERR58;
940 break;
941 }
942
943 if (negated)
944 {
945 if (c > bracount)
946 {
947 *errorcodeptr = ERR15;
948 break;
949 }
950 c = bracount - (c - 1);
951 }
952
953 c = -(ESC_REF + c);
954 break;
955
956 /* The handling of escape sequences consisting of a string of digits
957 starting with one that is not zero is not straightforward. By experiment,
958 the way Perl works seems to be as follows:
959
960 Outside a character class, the digits are read as a decimal number. If the
961 number is less than 10, or if there are that many previous extracting
962 left brackets, then it is a back reference. Otherwise, up to three octal
963 digits are read to form an escaped byte. Thus \123 is likely to be octal
964 123 (cf \0123, which is octal 012 followed by the literal 3). If the octal
965 value is greater than 377, the least significant 8 bits are taken. Inside a
966 character class, \ followed by a digit is always an octal number. */
967
968 case CHAR_1: case CHAR_2: case CHAR_3: case CHAR_4: case CHAR_5:
969 case CHAR_6: case CHAR_7: case CHAR_8: case CHAR_9:
970
971 if (!isclass)
972 {
973 oldptr = ptr;
974 /* The integer range is limited by the machine's int representation. */
975 c -= CHAR_0;
976 while (IS_DIGIT(ptr[1]))
977 {
978 if (((unsigned int)c) > INT_MAX / 10) /* Integer overflow */
979 {
980 c = -1;
981 break;
982 }
983 c = c * 10 + *(++ptr) - CHAR_0;
984 }
985 if (((unsigned int)c) > INT_MAX) /* Integer overflow */
986 {
987 while (IS_DIGIT(ptr[1]))
988 ptr++;
989 *errorcodeptr = ERR61;
990 break;
991 }
992 if (c < 10 || c <= bracount)
993 {
994 c = -(ESC_REF + c);
995 break;
996 }
997 ptr = oldptr; /* Put the pointer back and fall through */
998 }
999
1000 /* Handle an octal number following \. If the first digit is 8 or 9, Perl
1001 generates a binary zero byte and treats the digit as a following literal.
1002 Thus we have to pull back the pointer by one. */
1003
1004 if ((c = *ptr) >= CHAR_8)
1005 {
1006 ptr--;
1007 c = 0;
1008 break;
1009 }
1010
1011 /* \0 always starts an octal number, but we may drop through to here with a
1012 larger first octal digit. The original code used just to take the least
1013 significant 8 bits of octal numbers (I think this is what early Perls used
1014 to do). Nowadays we allow for larger numbers in UTF-8 mode and 16-bit mode,
1015 but no more than 3 octal digits. */
1016
1017 case CHAR_0:
1018 c -= CHAR_0;
1019 while(i++ < 2 && ptr[1] >= CHAR_0 && ptr[1] <= CHAR_7)
1020 c = c * 8 + *(++ptr) - CHAR_0;
1021 #ifdef COMPILE_PCRE8
1022 if (!utf && c > 0xff) *errorcodeptr = ERR51;
1023 #endif
1024 break;
1025
1026 /* \x is complicated. \x{ddd} is a character number which can be greater
1027 than 0xff in utf or non-8bit mode, but only if the ddd are hex digits.
1028 If not, { is treated as a data character. */
1029
1030 case CHAR_x:
1031 if ((options & PCRE_JAVASCRIPT_COMPAT) != 0)
1032 {
1033 /* In JavaScript, \x must be followed by two hexadecimal numbers.
1034 Otherwise it is a lowercase x letter. */
1035 if (MAX_255(ptr[1]) && (digitab[ptr[1]] & ctype_xdigit) != 0
1036 && MAX_255(ptr[2]) && (digitab[ptr[2]] & ctype_xdigit) != 0)
1037 {
1038 c = 0;
1039 for (i = 0; i < 2; ++i)
1040 {
1041 register int cc = *(++ptr);
1042 #ifndef EBCDIC /* ASCII/UTF-8 coding */
1043 if (cc >= CHAR_a) cc -= 32; /* Convert to upper case */
1044 c = (c << 4) + cc - ((cc < CHAR_A)? CHAR_0 : (CHAR_A - 10));
1045 #else /* EBCDIC coding */
1046 if (cc >= CHAR_a && cc <= CHAR_z) cc += 64; /* Convert to upper case */
1047 c = (c << 4) + cc - ((cc >= CHAR_0)? CHAR_0 : (CHAR_A - 10));
1048 #endif
1049 }
1050 }
1051 break;
1052 }
1053
1054 if (ptr[1] == CHAR_LEFT_CURLY_BRACKET)
1055 {
1056 const pcre_uchar *pt = ptr + 2;
1057
1058 c = 0;
1059 while (MAX_255(*pt) && (digitab[*pt] & ctype_xdigit) != 0)
1060 {
1061 register int cc = *pt++;
1062 if (c == 0 && cc == CHAR_0) continue; /* Leading zeroes */
1063
1064 #ifndef EBCDIC /* ASCII/UTF-8 coding */
1065 if (cc >= CHAR_a) cc -= 32; /* Convert to upper case */
1066 c = (c << 4) + cc - ((cc < CHAR_A)? CHAR_0 : (CHAR_A - 10));
1067 #else /* EBCDIC coding */
1068 if (cc >= CHAR_a && cc <= CHAR_z) cc += 64; /* Convert to upper case */
1069 c = (c << 4) + cc - ((cc >= CHAR_0)? CHAR_0 : (CHAR_A - 10));
1070 #endif
1071
1072 #ifdef COMPILE_PCRE8
1073 if (c > (utf ? 0x10ffff : 0xff)) { c = -1; break; }
1074 #else
1075 #ifdef COMPILE_PCRE16
1076 if (c > (utf ? 0x10ffff : 0xffff)) { c = -1; break; }
1077 #endif
1078 #endif
1079 }
1080
1081 if (c < 0)
1082 {
1083 while (MAX_255(*pt) && (digitab[*pt] & ctype_xdigit) != 0) pt++;
1084 *errorcodeptr = ERR34;
1085 }
1086
1087 if (*pt == CHAR_RIGHT_CURLY_BRACKET)
1088 {
1089 if (utf && c >= 0xd800 && c <= 0xdfff) *errorcodeptr = ERR73;
1090 ptr = pt;
1091 break;
1092 }
1093
1094 /* If the sequence of hex digits does not end with '}', then we don't
1095 recognize this construct; fall through to the normal \x handling. */
1096 }
1097
1098 /* Read just a single-byte hex-defined char */
1099
1100 c = 0;
1101 while (i++ < 2 && MAX_255(ptr[1]) && (digitab[ptr[1]] & ctype_xdigit) != 0)
1102 {
1103 int cc; /* Some compilers don't like */
1104 cc = *(++ptr); /* ++ in initializers */
1105 #ifndef EBCDIC /* ASCII/UTF-8 coding */
1106 if (cc >= CHAR_a) cc -= 32; /* Convert to upper case */
1107 c = c * 16 + cc - ((cc < CHAR_A)? CHAR_0 : (CHAR_A - 10));
1108 #else /* EBCDIC coding */
1109 if (cc <= CHAR_z) cc += 64; /* Convert to upper case */
1110 c = c * 16 + cc - ((cc >= CHAR_0)? CHAR_0 : (CHAR_A - 10));
1111 #endif
1112 }
1113 break;
1114
1115 /* For \c, a following letter is upper-cased; then the 0x40 bit is flipped.
1116 An error is given if the byte following \c is not an ASCII character. This
1117 coding is ASCII-specific, but then the whole concept of \cx is
1118 ASCII-specific. (However, an EBCDIC equivalent has now been added.) */
1119
1120 case CHAR_c:
1121 c = *(++ptr);
1122 if (c == 0)
1123 {
1124 *errorcodeptr = ERR2;
1125 break;
1126 }
1127 #ifndef EBCDIC /* ASCII/UTF-8 coding */
1128 if (c > 127) /* Excludes all non-ASCII in either mode */
1129 {
1130 *errorcodeptr = ERR68;
1131 break;
1132 }
1133 if (c >= CHAR_a && c <= CHAR_z) c -= 32;
1134 c ^= 0x40;
1135 #else /* EBCDIC coding */
1136 if (c >= CHAR_a && c <= CHAR_z) c += 64;
1137 c ^= 0xC0;
1138 #endif
1139 break;
1140
1141 /* PCRE_EXTRA enables extensions to Perl in the matter of escapes. Any
1142 other alphanumeric following \ is an error if PCRE_EXTRA was set;
1143 otherwise, for Perl compatibility, it is a literal. This code looks a bit
1144 odd, but there used to be some cases other than the default, and there may
1145 be again in future, so I haven't "optimized" it. */
1146
1147 default:
1148 if ((options & PCRE_EXTRA) != 0) switch(c)
1149 {
1150 default:
1151 *errorcodeptr = ERR3;
1152 break;
1153 }
1154 break;
1155 }
1156 }
1157
1158 /* Perl supports \N{name} for character names, as well as plain \N for "not
1159 newline". PCRE does not support \N{name}. However, it does support
1160 quantification such as \N{2,3}. */
1161
1162 if (c == -ESC_N && ptr[1] == CHAR_LEFT_CURLY_BRACKET &&
1163 !is_counted_repeat(ptr+2))
1164 *errorcodeptr = ERR37;
1165
1166 /* If PCRE_UCP is set, we change the values for \d etc. */
1167
1168 if ((options & PCRE_UCP) != 0 && c <= -ESC_D && c >= -ESC_w)
1169 c -= (ESC_DU - ESC_D);
1170
1171 /* Set the pointer to the final character before returning. */
1172
1173 *ptrptr = ptr;
1174 return c;
1175 }
1176
1177
1178
1179 #ifdef SUPPORT_UCP
1180 /*************************************************
1181 * Handle \P and \p *
1182 *************************************************/
1183
1184 /* This function is called after \P or \p has been encountered, provided that
1185 PCRE is compiled with support for Unicode properties. On entry, ptrptr is
1186 pointing at the P or p. On exit, it is pointing at the final character of the
1187 escape sequence.
1188
1189 Argument:
1190 ptrptr points to the pattern position pointer
1191 negptr points to a boolean that is set TRUE for negation else FALSE
1192 dptr points to an int that is set to the detailed property value
1193 errorcodeptr points to the error code variable
1194
1195 Returns: type value from ucp_type_table, or -1 for an invalid type
1196 */
1197
1198 static int
1199 get_ucp(const pcre_uchar **ptrptr, BOOL *negptr, int *dptr, int *errorcodeptr)
1200 {
1201 int c, i, bot, top;
1202 const pcre_uchar *ptr = *ptrptr;
1203 pcre_uchar name[32];
1204
1205 c = *(++ptr);
1206 if (c == 0) goto ERROR_RETURN;
1207
1208 *negptr = FALSE;
1209
1210 /* \P or \p can be followed by a name in {}, optionally preceded by ^ for
1211 negation. */
1212
1213 if (c == CHAR_LEFT_CURLY_BRACKET)
1214 {
1215 if (ptr[1] == CHAR_CIRCUMFLEX_ACCENT)
1216 {
1217 *negptr = TRUE;
1218 ptr++;
1219 }
1220 for (i = 0; i < (int)(sizeof(name) / sizeof(pcre_uchar)) - 1; i++)
1221 {
1222 c = *(++ptr);
1223 if (c == 0) goto ERROR_RETURN;
1224 if (c == CHAR_RIGHT_CURLY_BRACKET) break;
1225 name[i] = c;
1226 }
1227 if (c != CHAR_RIGHT_CURLY_BRACKET) goto ERROR_RETURN;
1228 name[i] = 0;
1229 }
1230
1231 /* Otherwise there is just one following character */
1232
1233 else
1234 {
1235 name[0] = c;
1236 name[1] = 0;
1237 }
1238
1239 *ptrptr = ptr;
1240
1241 /* Search for a recognized property name using binary chop */
1242
1243 bot = 0;
1244 top = PRIV(utt_size);
1245
1246 while (bot < top)
1247 {
1248 i = (bot + top) >> 1;
1249 c = STRCMP_UC_C8(name, PRIV(utt_names) + PRIV(utt)[i].name_offset);
1250 if (c == 0)
1251 {
1252 *dptr = PRIV(utt)[i].value;
1253 return PRIV(utt)[i].type;
1254 }
1255 if (c > 0) bot = i + 1; else top = i;
1256 }
1257
1258 *errorcodeptr = ERR47;
1259 *ptrptr = ptr;
1260 return -1;
1261
1262 ERROR_RETURN:
1263 *errorcodeptr = ERR46;
1264 *ptrptr = ptr;
1265 return -1;
1266 }
1267 #endif
1268
1269
1270
1271
1272 /*************************************************
1273 * Read repeat counts *
1274 *************************************************/
1275
1276 /* Read an item of the form {n,m} and return the values. This is called only
1277 after is_counted_repeat() has confirmed that a repeat-count quantifier exists,
1278 so the syntax is guaranteed to be correct, but we need to check the values.
1279
1280 Arguments:
1281 p pointer to first char after '{'
1282 minp pointer to int for min
1283 maxp pointer to int for max
1284 returned as -1 if no max
1285 errorcodeptr points to error code variable
1286
1287 Returns: pointer to '}' on success;
1288 current ptr on error, with errorcodeptr set non-zero
1289 */
1290
1291 static const pcre_uchar *
1292 read_repeat_counts(const pcre_uchar *p, int *minp, int *maxp, int *errorcodeptr)
1293 {
1294 int min = 0;
1295 int max = -1;
1296
1297 /* Read the minimum value and do a paranoid check: a negative value indicates
1298 an integer overflow. */
1299
1300 while (IS_DIGIT(*p)) min = min * 10 + *p++ - CHAR_0;
1301 if (min < 0 || min > 65535)
1302 {
1303 *errorcodeptr = ERR5;
1304 return p;
1305 }
1306
1307 /* Read the maximum value if there is one, and again do a paranoid on its size.
1308 Also, max must not be less than min. */
1309
1310 if (*p == CHAR_RIGHT_CURLY_BRACKET) max = min; else
1311 {
1312 if (*(++p) != CHAR_RIGHT_CURLY_BRACKET)
1313 {
1314 max = 0;
1315 while(IS_DIGIT(*p)) max = max * 10 + *p++ - CHAR_0;
1316 if (max < 0 || max > 65535)
1317 {
1318 *errorcodeptr = ERR5;
1319 return p;
1320 }
1321 if (max < min)
1322 {
1323 *errorcodeptr = ERR4;
1324 return p;
1325 }
1326 }
1327 }
1328
1329 /* Fill in the required variables, and pass back the pointer to the terminating
1330 '}'. */
1331
1332 *minp = min;
1333 *maxp = max;
1334 return p;
1335 }
1336
1337
1338
1339 /*************************************************
1340 * Subroutine for finding forward reference *
1341 *************************************************/
1342
1343 /* This recursive function is called only from find_parens() below. The
1344 top-level call starts at the beginning of the pattern. All other calls must
1345 start at a parenthesis. It scans along a pattern's text looking for capturing
1346 subpatterns, and counting them. If it finds a named pattern that matches the
1347 name it is given, it returns its number. Alternatively, if the name is NULL, it
1348 returns when it reaches a given numbered subpattern. Recursion is used to keep
1349 track of subpatterns that reset the capturing group numbers - the (?| feature.
1350
1351 This function was originally called only from the second pass, in which we know
1352 that if (?< or (?' or (?P< is encountered, the name will be correctly
1353 terminated because that is checked in the first pass. There is now one call to
1354 this function in the first pass, to check for a recursive back reference by
1355 name (so that we can make the whole group atomic). In this case, we need check
1356 only up to the current position in the pattern, and that is still OK because
1357 and previous occurrences will have been checked. To make this work, the test
1358 for "end of pattern" is a check against cd->end_pattern in the main loop,
1359 instead of looking for a binary zero. This means that the special first-pass
1360 call can adjust cd->end_pattern temporarily. (Checks for binary zero while
1361 processing items within the loop are OK, because afterwards the main loop will
1362 terminate.)
1363
1364 Arguments:
1365 ptrptr address of the current character pointer (updated)
1366 cd compile background data
1367 name name to seek, or NULL if seeking a numbered subpattern
1368 lorn name length, or subpattern number if name is NULL
1369 xmode TRUE if we are in /x mode
1370 utf TRUE if we are in UTF-8 / UTF-16 mode
1371 count pointer to the current capturing subpattern number (updated)
1372
1373 Returns: the number of the named subpattern, or -1 if not found
1374 */
1375
1376 static int
1377 find_parens_sub(pcre_uchar **ptrptr, compile_data *cd, const pcre_uchar *name, int lorn,
1378 BOOL xmode, BOOL utf, int *count)
1379 {
1380 pcre_uchar *ptr = *ptrptr;
1381 int start_count = *count;
1382 int hwm_count = start_count;
1383 BOOL dup_parens = FALSE;
1384
1385 /* If the first character is a parenthesis, check on the type of group we are
1386 dealing with. The very first call may not start with a parenthesis. */
1387
1388 if (ptr[0] == CHAR_LEFT_PARENTHESIS)
1389 {
1390 /* Handle specials such as (*SKIP) or (*UTF8) etc. */
1391
1392 if (ptr[1] == CHAR_ASTERISK) ptr += 2;
1393
1394 /* Handle a normal, unnamed capturing parenthesis. */
1395
1396 else if (ptr[1] != CHAR_QUESTION_MARK)
1397 {
1398 *count += 1;
1399 if (name == NULL && *count == lorn) return *count;
1400 ptr++;
1401 }
1402
1403 /* All cases now have (? at the start. Remember when we are in a group
1404 where the parenthesis numbers are duplicated. */
1405
1406 else if (ptr[2] == CHAR_VERTICAL_LINE)
1407 {
1408 ptr += 3;
1409 dup_parens = TRUE;
1410 }
1411
1412 /* Handle comments; all characters are allowed until a ket is reached. */
1413
1414 else if (ptr[2] == CHAR_NUMBER_SIGN)
1415 {
1416 for (ptr += 3; *ptr != 0; ptr++) if (*ptr == CHAR_RIGHT_PARENTHESIS) break;
1417 goto FAIL_EXIT;
1418 }
1419
1420 /* Handle a condition. If it is an assertion, just carry on so that it
1421 is processed as normal. If not, skip to the closing parenthesis of the
1422 condition (there can't be any nested parens). */
1423
1424 else if (ptr[2] == CHAR_LEFT_PARENTHESIS)
1425 {
1426 ptr += 2;
1427 if (ptr[1] != CHAR_QUESTION_MARK)
1428 {
1429 while (*ptr != 0 && *ptr != CHAR_RIGHT_PARENTHESIS) ptr++;
1430 if (*ptr != 0) ptr++;
1431 }
1432 }
1433
1434 /* Start with (? but not a condition. */
1435
1436 else
1437 {
1438 ptr += 2;
1439 if (*ptr == CHAR_P) ptr++; /* Allow optional P */
1440
1441 /* We have to disambiguate (?<! and (?<= from (?<name> for named groups */
1442
1443 if ((*ptr == CHAR_LESS_THAN_SIGN && ptr[1] != CHAR_EXCLAMATION_MARK &&
1444 ptr[1] != CHAR_EQUALS_SIGN) || *ptr == CHAR_APOSTROPHE)
1445 {
1446 int term;
1447 const pcre_uchar *thisname;
1448 *count += 1;
1449 if (name == NULL && *count == lorn) return *count;
1450 term = *ptr++;
1451 if (term == CHAR_LESS_THAN_SIGN) term = CHAR_GREATER_THAN_SIGN;
1452 thisname = ptr;
1453 while (*ptr != term) ptr++;
1454 if (name != NULL && lorn == ptr - thisname &&
1455 STRNCMP_UC_UC(name, thisname, lorn) == 0)
1456 return *count;
1457 term++;
1458 }
1459 }
1460 }
1461
1462 /* Past any initial parenthesis handling, scan for parentheses or vertical
1463 bars. Stop if we get to cd->end_pattern. Note that this is important for the
1464 first-pass call when this value is temporarily adjusted to stop at the current
1465 position. So DO NOT change this to a test for binary zero. */
1466
1467 for (; ptr < cd->end_pattern; ptr++)
1468 {
1469 /* Skip over backslashed characters and also entire \Q...\E */
1470
1471 if (*ptr == CHAR_BACKSLASH)
1472 {
1473 if (*(++ptr) == 0) goto FAIL_EXIT;
1474 if (*ptr == CHAR_Q) for (;;)
1475 {
1476 while (*(++ptr) != 0 && *ptr != CHAR_BACKSLASH) {};
1477 if (*ptr == 0) goto FAIL_EXIT;
1478 if (*(++ptr) == CHAR_E) break;
1479 }
1480 continue;
1481 }
1482
1483 /* Skip over character classes; this logic must be similar to the way they
1484 are handled for real. If the first character is '^', skip it. Also, if the
1485 first few characters (either before or after ^) are \Q\E or \E we skip them
1486 too. This makes for compatibility with Perl. Note the use of STR macros to
1487 encode "Q\\E" so that it works in UTF-8 on EBCDIC platforms. */
1488
1489 if (*ptr == CHAR_LEFT_SQUARE_BRACKET)
1490 {
1491 BOOL negate_class = FALSE;
1492 for (;;)
1493 {
1494 if (ptr[1] == CHAR_BACKSLASH)
1495 {
1496 if (ptr[2] == CHAR_E)
1497 ptr+= 2;
1498 else if (STRNCMP_UC_C8(ptr + 2,
1499 STR_Q STR_BACKSLASH STR_E, 3) == 0)
1500 ptr += 4;
1501 else
1502 break;
1503 }
1504 else if (!negate_class && ptr[1] == CHAR_CIRCUMFLEX_ACCENT)
1505 {
1506 negate_class = TRUE;
1507 ptr++;
1508 }
1509 else break;
1510 }
1511
1512 /* If the next character is ']', it is a data character that must be
1513 skipped, except in JavaScript compatibility mode. */
1514
1515 if (ptr[1] == CHAR_RIGHT_SQUARE_BRACKET &&
1516 (cd->external_options & PCRE_JAVASCRIPT_COMPAT) == 0)
1517 ptr++;
1518
1519 while (*(++ptr) != CHAR_RIGHT_SQUARE_BRACKET)
1520 {
1521 if (*ptr == 0) return -1;
1522 if (*ptr == CHAR_BACKSLASH)
1523 {
1524 if (*(++ptr) == 0) goto FAIL_EXIT;
1525 if (*ptr == CHAR_Q) for (;;)
1526 {
1527 while (*(++ptr) != 0 && *ptr != CHAR_BACKSLASH) {};
1528 if (*ptr == 0) goto FAIL_EXIT;
1529 if (*(++ptr) == CHAR_E) break;
1530 }
1531 continue;
1532 }
1533 }
1534 continue;
1535 }
1536
1537 /* Skip comments in /x mode */
1538
1539 if (xmode && *ptr == CHAR_NUMBER_SIGN)
1540 {
1541 ptr++;
1542 while (*ptr != 0)
1543 {
1544 if (IS_NEWLINE(ptr)) { ptr += cd->nllen - 1; break; }
1545 ptr++;
1546 #ifdef SUPPORT_UTF
1547 if (utf) FORWARDCHAR(ptr);
1548 #endif
1549 }
1550 if (*ptr == 0) goto FAIL_EXIT;
1551 continue;
1552 }
1553
1554 /* Check for the special metacharacters */
1555
1556 if (*ptr == CHAR_LEFT_PARENTHESIS)
1557 {
1558 int rc = find_parens_sub(&ptr, cd, name, lorn, xmode, utf, count);
1559 if (rc > 0) return rc;
1560 if (*ptr == 0) goto FAIL_EXIT;
1561 }
1562
1563 else if (*ptr == CHAR_RIGHT_PARENTHESIS)
1564 {
1565 if (dup_parens && *count < hwm_count) *count = hwm_count;
1566 goto FAIL_EXIT;
1567 }
1568
1569 else if (*ptr == CHAR_VERTICAL_LINE && dup_parens)
1570 {
1571 if (*count > hwm_count) hwm_count = *count;
1572 *count = start_count;
1573 }
1574 }
1575
1576 FAIL_EXIT:
1577 *ptrptr = ptr;
1578 return -1;
1579 }
1580
1581
1582
1583
1584 /*************************************************
1585 * Find forward referenced subpattern *
1586 *************************************************/
1587
1588 /* This function scans along a pattern's text looking for capturing
1589 subpatterns, and counting them. If it finds a named pattern that matches the
1590 name it is given, it returns its number. Alternatively, if the name is NULL, it
1591 returns when it reaches a given numbered subpattern. This is used for forward
1592 references to subpatterns. We used to be able to start this scan from the
1593 current compiling point, using the current count value from cd->bracount, and
1594 do it all in a single loop, but the addition of the possibility of duplicate
1595 subpattern numbers means that we have to scan from the very start, in order to
1596 take account of such duplicates, and to use a recursive function to keep track
1597 of the different types of group.
1598
1599 Arguments:
1600 cd compile background data
1601 name name to seek, or NULL if seeking a numbered subpattern
1602 lorn name length, or subpattern number if name is NULL
1603 xmode TRUE if we are in /x mode
1604 utf TRUE if we are in UTF-8 / UTF-16 mode
1605
1606 Returns: the number of the found subpattern, or -1 if not found
1607 */
1608
1609 static int
1610 find_parens(compile_data *cd, const pcre_uchar *name, int lorn, BOOL xmode,
1611 BOOL utf)
1612 {
1613 pcre_uchar *ptr = (pcre_uchar *)cd->start_pattern;
1614 int count = 0;
1615 int rc;
1616
1617 /* If the pattern does not start with an opening parenthesis, the first call
1618 to find_parens_sub() will scan right to the end (if necessary). However, if it
1619 does start with a parenthesis, find_parens_sub() will return when it hits the
1620 matching closing parens. That is why we have to have a loop. */
1621
1622 for (;;)
1623 {
1624 rc = find_parens_sub(&ptr, cd, name, lorn, xmode, utf, &count);
1625 if (rc > 0 || *ptr++ == 0) break;
1626 }
1627
1628 return rc;
1629 }
1630
1631
1632
1633
1634 /*************************************************
1635 * Find first significant op code *
1636 *************************************************/
1637
1638 /* This is called by several functions that scan a compiled expression looking
1639 for a fixed first character, or an anchoring op code etc. It skips over things
1640 that do not influence this. For some calls, it makes sense to skip negative
1641 forward and all backward assertions, and also the \b assertion; for others it
1642 does not.
1643
1644 Arguments:
1645 code pointer to the start of the group
1646 skipassert TRUE if certain assertions are to be skipped
1647
1648 Returns: pointer to the first significant opcode
1649 */
1650
1651 static const pcre_uchar*
1652 first_significant_code(const pcre_uchar *code, BOOL skipassert)
1653 {
1654 for (;;)
1655 {
1656 switch ((int)*code)
1657 {
1658 case OP_ASSERT_NOT:
1659 case OP_ASSERTBACK:
1660 case OP_ASSERTBACK_NOT:
1661 if (!skipassert) return code;
1662 do code += GET(code, 1); while (*code == OP_ALT);
1663 code += PRIV(OP_lengths)[*code];
1664 break;
1665
1666 case OP_WORD_BOUNDARY:
1667 case OP_NOT_WORD_BOUNDARY:
1668 if (!skipassert) return code;
1669 /* Fall through */
1670
1671 case OP_CALLOUT:
1672 case OP_CREF:
1673 case OP_NCREF:
1674 case OP_RREF:
1675 case OP_NRREF:
1676 case OP_DEF:
1677 code += PRIV(OP_lengths)[*code];
1678 break;
1679
1680 default:
1681 return code;
1682 }
1683 }
1684 /* Control never reaches here */
1685 }
1686
1687
1688
1689
1690 /*************************************************
1691 * Find the fixed length of a branch *
1692 *************************************************/
1693
1694 /* Scan a branch and compute the fixed length of subject that will match it,
1695 if the length is fixed. This is needed for dealing with backward assertions.
1696 In UTF8 mode, the result is in characters rather than bytes. The branch is
1697 temporarily terminated with OP_END when this function is called.
1698
1699 This function is called when a backward assertion is encountered, so that if it
1700 fails, the error message can point to the correct place in the pattern.
1701 However, we cannot do this when the assertion contains subroutine calls,
1702 because they can be forward references. We solve this by remembering this case
1703 and doing the check at the end; a flag specifies which mode we are running in.
1704
1705 Arguments:
1706 code points to the start of the pattern (the bracket)
1707 utf TRUE in UTF-8 / UTF-16 mode
1708 atend TRUE if called when the pattern is complete
1709 cd the "compile data" structure
1710
1711 Returns: the fixed length,
1712 or -1 if there is no fixed length,
1713 or -2 if \C was encountered (in UTF-8 mode only)
1714 or -3 if an OP_RECURSE item was encountered and atend is FALSE
1715 or -4 if an unknown opcode was encountered (internal error)
1716 */
1717
1718 static int
1719 find_fixedlength(pcre_uchar *code, BOOL utf, BOOL atend, compile_data *cd)
1720 {
1721 int length = -1;
1722
1723 register int branchlength = 0;
1724 register pcre_uchar *cc = code + 1 + LINK_SIZE;
1725
1726 /* Scan along the opcodes for this branch. If we get to the end of the
1727 branch, check the length against that of the other branches. */
1728
1729 for (;;)
1730 {
1731 int d;
1732 pcre_uchar *ce, *cs;
1733 register int op = *cc;
1734
1735 switch (op)
1736 {
1737 /* We only need to continue for OP_CBRA (normal capturing bracket) and
1738 OP_BRA (normal non-capturing bracket) because the other variants of these
1739 opcodes are all concerned with unlimited repeated groups, which of course
1740 are not of fixed length. */
1741
1742 case OP_CBRA:
1743 case OP_BRA:
1744 case OP_ONCE:
1745 case OP_ONCE_NC:
1746 case OP_COND:
1747 d = find_fixedlength(cc + ((op == OP_CBRA)? IMM2_SIZE : 0), utf, atend, cd);
1748 if (d < 0) return d;
1749 branchlength += d;
1750 do cc += GET(cc, 1); while (*cc == OP_ALT);
1751 cc += 1 + LINK_SIZE;
1752 break;
1753
1754 /* Reached end of a branch; if it's a ket it is the end of a nested call.
1755 If it's ALT it is an alternation in a nested call. An ACCEPT is effectively
1756 an ALT. If it is END it's the end of the outer call. All can be handled by
1757 the same code. Note that we must not include the OP_KETRxxx opcodes here,
1758 because they all imply an unlimited repeat. */
1759
1760 case OP_ALT:
1761 case OP_KET:
1762 case OP_END:
1763 case OP_ACCEPT:
1764 case OP_ASSERT_ACCEPT:
1765 if (length < 0) length = branchlength;
1766 else if (length != branchlength) return -1;
1767 if (*cc != OP_ALT) return length;
1768 cc += 1 + LINK_SIZE;
1769 branchlength = 0;
1770 break;
1771
1772 /* A true recursion implies not fixed length, but a subroutine call may
1773 be OK. If the subroutine is a forward reference, we can't deal with
1774 it until the end of the pattern, so return -3. */
1775
1776 case OP_RECURSE:
1777 if (!atend) return -3;
1778 cs = ce = (pcre_uchar *)cd->start_code + GET(cc, 1); /* Start subpattern */
1779 do ce += GET(ce, 1); while (*ce == OP_ALT); /* End subpattern */
1780 if (cc > cs && cc < ce) return -1; /* Recursion */
1781 d = find_fixedlength(cs + IMM2_SIZE, utf, atend, cd);
1782 if (d < 0) return d;
1783 branchlength += d;
1784 cc += 1 + LINK_SIZE;
1785 break;
1786
1787 /* Skip over assertive subpatterns */
1788
1789 case OP_ASSERT:
1790 case OP_ASSERT_NOT:
1791 case OP_ASSERTBACK:
1792 case OP_ASSERTBACK_NOT:
1793 do cc += GET(cc, 1); while (*cc == OP_ALT);
1794 cc += PRIV(OP_lengths)[*cc];
1795 break;
1796
1797 /* Skip over things that don't match chars */
1798
1799 case OP_MARK:
1800 case OP_PRUNE_ARG:
1801 case OP_SKIP_ARG:
1802 case OP_THEN_ARG:
1803 cc += cc[1] + PRIV(OP_lengths)[*cc];
1804 break;
1805
1806 case OP_CALLOUT:
1807 case OP_CIRC:
1808 case OP_CIRCM:
1809 case OP_CLOSE:
1810 case OP_COMMIT:
1811 case OP_CREF:
1812 case OP_DEF:
1813 case OP_DOLL:
1814 case OP_DOLLM:
1815 case OP_EOD:
1816 case OP_EODN:
1817 case OP_FAIL:
1818 case OP_NCREF:
1819 case OP_NRREF:
1820 case OP_NOT_WORD_BOUNDARY:
1821 case OP_PRUNE:
1822 case OP_REVERSE:
1823 case OP_RREF:
1824 case OP_SET_SOM:
1825 case OP_SKIP:
1826 case OP_SOD:
1827 case OP_SOM:
1828 case OP_THEN:
1829 case OP_WORD_BOUNDARY:
1830 cc += PRIV(OP_lengths)[*cc];
1831 break;
1832
1833 /* Handle literal characters */
1834
1835 case OP_CHAR:
1836 case OP_CHARI:
1837 case OP_NOT:
1838 case OP_NOTI:
1839 branchlength++;
1840 cc += 2;
1841 #ifdef SUPPORT_UTF
1842 if (utf && HAS_EXTRALEN(cc[-1])) cc += GET_EXTRALEN(cc[-1]);
1843 #endif
1844 break;
1845
1846 /* Handle exact repetitions. The count is already in characters, but we
1847 need to skip over a multibyte character in UTF8 mode. */
1848
1849 case OP_EXACT:
1850 case OP_EXACTI:
1851 case OP_NOTEXACT:
1852 case OP_NOTEXACTI:
1853 branchlength += GET2(cc,1);
1854 cc += 2 + IMM2_SIZE;
1855 #ifdef SUPPORT_UTF
1856 if (utf && HAS_EXTRALEN(cc[-1])) cc += GET_EXTRALEN(cc[-1]);
1857 #endif
1858 break;
1859
1860 case OP_TYPEEXACT:
1861 branchlength += GET2(cc,1);
1862 if (cc[1 + IMM2_SIZE] == OP_PROP || cc[1 + IMM2_SIZE] == OP_NOTPROP)
1863 cc += 2;
1864 cc += 1 + IMM2_SIZE + 1;
1865 break;
1866
1867 /* Handle single-char matchers */
1868
1869 case OP_PROP:
1870 case OP_NOTPROP:
1871 cc += 2;
1872 /* Fall through */
1873
1874 case OP_HSPACE:
1875 case OP_VSPACE:
1876 case OP_NOT_HSPACE:
1877 case OP_NOT_VSPACE:
1878 case OP_NOT_DIGIT:
1879 case OP_DIGIT:
1880 case OP_NOT_WHITESPACE:
1881 case OP_WHITESPACE:
1882 case OP_NOT_WORDCHAR:
1883 case OP_WORDCHAR:
1884 case OP_ANY:
1885 case OP_ALLANY:
1886 branchlength++;
1887 cc++;
1888 break;
1889
1890 /* The single-byte matcher isn't allowed. This only happens in UTF-8 mode;
1891 otherwise \C is coded as OP_ALLANY. */
1892
1893 case OP_ANYBYTE:
1894 return -2;
1895
1896 /* Check a class for variable quantification */
1897
1898 #if defined SUPPORT_UTF || defined COMPILE_PCRE16
1899 case OP_XCLASS:
1900 cc += GET(cc, 1) - PRIV(OP_lengths)[OP_CLASS];
1901 /* Fall through */
1902 #endif
1903
1904 case OP_CLASS:
1905 case OP_NCLASS:
1906 cc += PRIV(OP_lengths)[OP_CLASS];
1907
1908 switch (*cc)
1909 {
1910 case OP_CRPLUS:
1911 case OP_CRMINPLUS:
1912 case OP_CRSTAR:
1913 case OP_CRMINSTAR:
1914 case OP_CRQUERY:
1915 case OP_CRMINQUERY:
1916 return -1;
1917
1918 case OP_CRRANGE:
1919 case OP_CRMINRANGE:
1920 if (GET2(cc,1) != GET2(cc,1+IMM2_SIZE)) return -1;
1921 branchlength += GET2(cc,1);
1922 cc += 1 + 2 * IMM2_SIZE;
1923 break;
1924
1925 default:
1926 branchlength++;
1927 }
1928 break;
1929
1930 /* Anything else is variable length */
1931
1932 case OP_ANYNL:
1933 case OP_BRAMINZERO:
1934 case OP_BRAPOS:
1935 case OP_BRAPOSZERO:
1936 case OP_BRAZERO:
1937 case OP_CBRAPOS:
1938 case OP_EXTUNI:
1939 case OP_KETRMAX:
1940 case OP_KETRMIN:
1941 case OP_KETRPOS:
1942 case OP_MINPLUS:
1943 case OP_MINPLUSI:
1944 case OP_MINQUERY:
1945 case OP_MINQUERYI:
1946 case OP_MINSTAR:
1947 case OP_MINSTARI:
1948 case OP_MINUPTO:
1949 case OP_MINUPTOI:
1950 case OP_NOTMINPLUS:
1951 case OP_NOTMINPLUSI:
1952 case OP_NOTMINQUERY:
1953 case OP_NOTMINQUERYI:
1954 case OP_NOTMINSTAR:
1955 case OP_NOTMINSTARI:
1956 case OP_NOTMINUPTO:
1957 case OP_NOTMINUPTOI:
1958 case OP_NOTPLUS:
1959 case OP_NOTPLUSI:
1960 case OP_NOTPOSPLUS:
1961 case OP_NOTPOSPLUSI:
1962 case OP_NOTPOSQUERY:
1963 case OP_NOTPOSQUERYI:
1964 case OP_NOTPOSSTAR:
1965 case OP_NOTPOSSTARI:
1966 case OP_NOTPOSUPTO:
1967 case OP_NOTPOSUPTOI:
1968 case OP_NOTQUERY:
1969 case OP_NOTQUERYI:
1970 case OP_NOTSTAR:
1971 case OP_NOTSTARI:
1972 case OP_NOTUPTO:
1973 case OP_NOTUPTOI:
1974 case OP_PLUS:
1975 case OP_PLUSI:
1976 case OP_POSPLUS:
1977 case OP_POSPLUSI:
1978 case OP_POSQUERY:
1979 case OP_POSQUERYI:
1980 case OP_POSSTAR:
1981 case OP_POSSTARI:
1982 case OP_POSUPTO:
1983 case OP_POSUPTOI:
1984 case OP_QUERY:
1985 case OP_QUERYI:
1986 case OP_REF:
1987 case OP_REFI:
1988 case OP_SBRA:
1989 case OP_SBRAPOS:
1990 case OP_SCBRA:
1991 case OP_SCBRAPOS:
1992 case OP_SCOND:
1993 case OP_SKIPZERO:
1994 case OP_STAR:
1995 case OP_STARI:
1996 case OP_TYPEMINPLUS:
1997 case OP_TYPEMINQUERY:
1998 case OP_TYPEMINSTAR:
1999 case OP_TYPEMINUPTO:
2000 case OP_TYPEPLUS:
2001 case OP_TYPEPOSPLUS:
2002 case OP_TYPEPOSQUERY:
2003 case OP_TYPEPOSSTAR:
2004 case OP_TYPEPOSUPTO:
2005 case OP_TYPEQUERY:
2006 case OP_TYPESTAR:
2007 case OP_TYPEUPTO:
2008 case OP_UPTO:
2009 case OP_UPTOI:
2010 return -1;
2011
2012 /* Catch unrecognized opcodes so that when new ones are added they
2013 are not forgotten, as has happened in the past. */
2014
2015 default:
2016 return -4;
2017 }
2018 }
2019 /* Control never gets here */
2020 }
2021
2022
2023
2024
2025 /*************************************************
2026 * Scan compiled regex for specific bracket *
2027 *************************************************/
2028
2029 /* This little function scans through a compiled pattern until it finds a
2030 capturing bracket with the given number, or, if the number is negative, an
2031 instance of OP_REVERSE for a lookbehind. The function is global in the C sense
2032 so that it can be called from pcre_study() when finding the minimum matching
2033 length.
2034
2035 Arguments:
2036 code points to start of expression
2037 utf TRUE in UTF-8 / UTF-16 mode
2038 number the required bracket number or negative to find a lookbehind
2039
2040 Returns: pointer to the opcode for the bracket, or NULL if not found
2041 */
2042
2043 const pcre_uchar *
2044 PRIV(find_bracket)(const pcre_uchar *code, BOOL utf, int number)
2045 {
2046 for (;;)
2047 {
2048 register int c = *code;
2049
2050 if (c == OP_END) return NULL;
2051
2052 /* XCLASS is used for classes that cannot be represented just by a bit
2053 map. This includes negated single high-valued characters. The length in
2054 the table is zero; the actual length is stored in the compiled code. */
2055
2056 if (c == OP_XCLASS) code += GET(code, 1);
2057
2058 /* Handle recursion */
2059
2060 else if (c == OP_REVERSE)
2061 {
2062 if (number < 0) return (pcre_uchar *)code;
2063 code += PRIV(OP_lengths)[c];
2064 }
2065
2066 /* Handle capturing bracket */
2067
2068 else if (c == OP_CBRA || c == OP_SCBRA ||
2069 c == OP_CBRAPOS || c == OP_SCBRAPOS)
2070 {
2071 int n = GET2(code, 1+LINK_SIZE);
2072 if (n == number) return (pcre_uchar *)code;
2073 code += PRIV(OP_lengths)[c];
2074 }
2075
2076 /* Otherwise, we can get the item's length from the table, except that for
2077 repeated character types, we have to test for \p and \P, which have an extra
2078 two bytes of parameters, and for MARK/PRUNE/SKIP/THEN with an argument, we
2079 must add in its length. */
2080
2081 else
2082 {
2083 switch(c)
2084 {
2085 case OP_TYPESTAR:
2086 case OP_TYPEMINSTAR:
2087 case OP_TYPEPLUS:
2088 case OP_TYPEMINPLUS:
2089 case OP_TYPEQUERY:
2090 case OP_TYPEMINQUERY:
2091 case OP_TYPEPOSSTAR:
2092 case OP_TYPEPOSPLUS:
2093 case OP_TYPEPOSQUERY:
2094 if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
2095 break;
2096
2097 case OP_TYPEUPTO:
2098 case OP_TYPEMINUPTO:
2099 case OP_TYPEEXACT:
2100 case OP_TYPEPOSUPTO:
2101 if (code[1 + IMM2_SIZE] == OP_PROP || code[1 + IMM2_SIZE] == OP_NOTPROP)
2102 code += 2;
2103 break;
2104
2105 case OP_MARK:
2106 case OP_PRUNE_ARG:
2107 case OP_SKIP_ARG:
2108 code += code[1];
2109 break;
2110
2111 case OP_THEN_ARG:
2112 code += code[1];
2113 break;
2114 }
2115
2116 /* Add in the fixed length from the table */
2117
2118 code += PRIV(OP_lengths)[c];
2119
2120 /* In UTF-8 mode, opcodes that are followed by a character may be followed by
2121 a multi-byte character. The length in the table is a minimum, so we have to
2122 arrange to skip the extra bytes. */
2123
2124 #ifdef SUPPORT_UTF
2125 if (utf) switch(c)
2126 {
2127 case OP_CHAR:
2128 case OP_CHARI:
2129 case OP_EXACT:
2130 case OP_EXACTI:
2131 case OP_UPTO:
2132 case OP_UPTOI:
2133 case OP_MINUPTO:
2134 case OP_MINUPTOI:
2135 case OP_POSUPTO:
2136 case OP_POSUPTOI:
2137 case OP_STAR:
2138 case OP_STARI:
2139 case OP_MINSTAR:
2140 case OP_MINSTARI:
2141 case OP_POSSTAR:
2142 case OP_POSSTARI:
2143 case OP_PLUS:
2144 case OP_PLUSI:
2145 case OP_MINPLUS:
2146 case OP_MINPLUSI:
2147 case OP_POSPLUS:
2148 case OP_POSPLUSI:
2149 case OP_QUERY:
2150 case OP_QUERYI:
2151 case OP_MINQUERY:
2152 case OP_MINQUERYI:
2153 case OP_POSQUERY:
2154 case OP_POSQUERYI:
2155 if (HAS_EXTRALEN(code[-1])) code += GET_EXTRALEN(code[-1]);
2156 break;
2157 }
2158 #else
2159 (void)(utf); /* Keep compiler happy by referencing function argument */
2160 #endif
2161 }
2162 }
2163 }
2164
2165
2166
2167 /*************************************************
2168 * Scan compiled regex for recursion reference *
2169 *************************************************/
2170
2171 /* This little function scans through a compiled pattern until it finds an
2172 instance of OP_RECURSE.
2173
2174 Arguments:
2175 code points to start of expression
2176 utf TRUE in UTF-8 / UTF-16 mode
2177
2178 Returns: pointer to the opcode for OP_RECURSE, or NULL if not found
2179 */
2180
2181 static const pcre_uchar *
2182 find_recurse(const pcre_uchar *code, BOOL utf)
2183 {
2184 for (;;)
2185 {
2186 register int c = *code;
2187 if (c == OP_END) return NULL;
2188 if (c == OP_RECURSE) return code;
2189
2190 /* XCLASS is used for classes that cannot be represented just by a bit
2191 map. This includes negated single high-valued characters. The length in
2192 the table is zero; the actual length is stored in the compiled code. */
2193
2194 if (c == OP_XCLASS) code += GET(code, 1);
2195
2196 /* Otherwise, we can get the item's length from the table, except that for
2197 repeated character types, we have to test for \p and \P, which have an extra
2198 two bytes of parameters, and for MARK/PRUNE/SKIP/THEN with an argument, we
2199 must add in its length. */
2200
2201 else
2202 {
2203 switch(c)
2204 {
2205 case OP_TYPESTAR:
2206 case OP_TYPEMINSTAR:
2207 case OP_TYPEPLUS:
2208 case OP_TYPEMINPLUS:
2209 case OP_TYPEQUERY:
2210 case OP_TYPEMINQUERY:
2211 case OP_TYPEPOSSTAR:
2212 case OP_TYPEPOSPLUS:
2213 case OP_TYPEPOSQUERY:
2214 if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
2215 break;
2216
2217 case OP_TYPEPOSUPTO:
2218 case OP_TYPEUPTO:
2219 case OP_TYPEMINUPTO:
2220 case OP_TYPEEXACT:
2221 if (code[1 + IMM2_SIZE] == OP_PROP || code[1 + IMM2_SIZE] == OP_NOTPROP)
2222 code += 2;
2223 break;
2224
2225 case OP_MARK:
2226 case OP_PRUNE_ARG:
2227 case OP_SKIP_ARG:
2228 code += code[1];
2229 break;
2230
2231 case OP_THEN_ARG:
2232 code += code[1];
2233 break;
2234 }
2235
2236 /* Add in the fixed length from the table */
2237
2238 code += PRIV(OP_lengths)[c];
2239
2240 /* In UTF-8 mode, opcodes that are followed by a character may be followed
2241 by a multi-byte character. The length in the table is a minimum, so we have
2242 to arrange to skip the extra bytes. */
2243
2244 #ifdef SUPPORT_UTF
2245 if (utf) switch(c)
2246 {
2247 case OP_CHAR:
2248 case OP_CHARI:
2249 case OP_NOT:
2250 case OP_NOTI:
2251 case OP_EXACT:
2252 case OP_EXACTI:
2253 case OP_NOTEXACT:
2254 case OP_NOTEXACTI:
2255 case OP_UPTO:
2256 case OP_UPTOI:
2257 case OP_NOTUPTO:
2258 case OP_NOTUPTOI:
2259 case OP_MINUPTO:
2260 case OP_MINUPTOI:
2261 case OP_NOTMINUPTO:
2262 case OP_NOTMINUPTOI:
2263 case OP_POSUPTO:
2264 case OP_POSUPTOI:
2265 case OP_NOTPOSUPTO:
2266 case OP_NOTPOSUPTOI:
2267 case OP_STAR:
2268 case OP_STARI:
2269 case OP_NOTSTAR:
2270 case OP_NOTSTARI:
2271 case OP_MINSTAR:
2272 case OP_MINSTARI:
2273 case OP_NOTMINSTAR:
2274 case OP_NOTMINSTARI:
2275 case OP_POSSTAR:
2276 case OP_POSSTARI:
2277 case OP_NOTPOSSTAR:
2278 case OP_NOTPOSSTARI:
2279 case OP_PLUS:
2280 case OP_PLUSI:
2281 case OP_NOTPLUS:
2282 case OP_NOTPLUSI:
2283 case OP_MINPLUS:
2284 case OP_MINPLUSI:
2285 case OP_NOTMINPLUS:
2286 case OP_NOTMINPLUSI:
2287 case OP_POSPLUS:
2288 case OP_POSPLUSI:
2289 case OP_NOTPOSPLUS:
2290 case OP_NOTPOSPLUSI:
2291 case OP_QUERY:
2292 case OP_QUERYI:
2293 case OP_NOTQUERY:
2294 case OP_NOTQUERYI:
2295 case OP_MINQUERY:
2296 case OP_MINQUERYI:
2297 case OP_NOTMINQUERY:
2298 case OP_NOTMINQUERYI:
2299 case OP_POSQUERY:
2300 case OP_POSQUERYI:
2301 case OP_NOTPOSQUERY:
2302 case OP_NOTPOSQUERYI:
2303 if (HAS_EXTRALEN(code[-1])) code += GET_EXTRALEN(code[-1]);
2304 break;
2305 }
2306 #else
2307 (void)(utf); /* Keep compiler happy by referencing function argument */
2308 #endif
2309 }
2310 }
2311 }
2312
2313
2314
2315 /*************************************************
2316 * Scan compiled branch for non-emptiness *
2317 *************************************************/
2318
2319 /* This function scans through a branch of a compiled pattern to see whether it
2320 can match the empty string or not. It is called from could_be_empty()
2321 below and from compile_branch() when checking for an unlimited repeat of a
2322 group that can match nothing. Note that first_significant_code() skips over
2323 backward and negative forward assertions when its final argument is TRUE. If we
2324 hit an unclosed bracket, we return "empty" - this means we've struck an inner
2325 bracket whose current branch will already have been scanned.
2326
2327 Arguments:
2328 code points to start of search
2329 endcode points to where to stop
2330 utf TRUE if in UTF-8 / UTF-16 mode
2331 cd contains pointers to tables etc.
2332
2333 Returns: TRUE if what is matched could be empty
2334 */
2335
2336 static BOOL
2337 could_be_empty_branch(const pcre_uchar *code, const pcre_uchar *endcode,
2338 BOOL utf, compile_data *cd)
2339 {
2340 register int c;
2341 for (code = first_significant_code(code + PRIV(OP_lengths)[*code], TRUE);
2342 code < endcode;
2343 code = first_significant_code(code + PRIV(OP_lengths)[c], TRUE))
2344 {
2345 const pcre_uchar *ccode;
2346
2347 c = *code;
2348
2349 /* Skip over forward assertions; the other assertions are skipped by
2350 first_significant_code() with a TRUE final argument. */
2351
2352 if (c == OP_ASSERT)
2353 {
2354 do code += GET(code, 1); while (*code == OP_ALT);
2355 c = *code;
2356 continue;
2357 }
2358
2359 /* For a recursion/subroutine call, if its end has been reached, which
2360 implies a backward reference subroutine call, we can scan it. If it's a
2361 forward reference subroutine call, we can't. To detect forward reference
2362 we have to scan up the list that is kept in the workspace. This function is
2363 called only when doing the real compile, not during the pre-compile that
2364 measures the size of the compiled pattern. */
2365
2366 if (c == OP_RECURSE)
2367 {
2368 const pcre_uchar *scode;
2369 BOOL empty_branch;
2370
2371 /* Test for forward reference */
2372
2373 for (scode = cd->start_workspace; scode < cd->hwm; scode += LINK_SIZE)
2374 if (GET(scode, 0) == code + 1 - cd->start_code) return TRUE;
2375
2376 /* Not a forward reference, test for completed backward reference */
2377
2378 empty_branch = FALSE;
2379 scode = cd->start_code + GET(code, 1);
2380 if (GET(scode, 1) == 0) return TRUE; /* Unclosed */
2381
2382 /* Completed backwards reference */
2383
2384 do
2385 {
2386 if (could_be_empty_branch(scode, endcode, utf, cd))
2387 {
2388 empty_branch = TRUE;
2389 break;
2390 }
2391 scode += GET(scode, 1);
2392 }
2393 while (*scode == OP_ALT);
2394
2395 if (!empty_branch) return FALSE; /* All branches are non-empty */
2396 continue;
2397 }
2398
2399 /* Groups with zero repeats can of course be empty; skip them. */
2400
2401 if (c == OP_BRAZERO || c == OP_BRAMINZERO || c == OP_SKIPZERO ||
2402 c == OP_BRAPOSZERO)
2403 {
2404 code += PRIV(OP_lengths)[c];
2405 do code += GET(code, 1); while (*code == OP_ALT);
2406 c = *code;
2407 continue;
2408 }
2409
2410 /* A nested group that is already marked as "could be empty" can just be
2411 skipped. */
2412
2413 if (c == OP_SBRA || c == OP_SBRAPOS ||
2414 c == OP_SCBRA || c == OP_SCBRAPOS)
2415 {
2416 do code += GET(code, 1); while (*code == OP_ALT);
2417 c = *code;
2418 continue;
2419 }
2420
2421 /* For other groups, scan the branches. */
2422
2423 if (c == OP_BRA || c == OP_BRAPOS ||
2424 c == OP_CBRA || c == OP_CBRAPOS ||
2425 c == OP_ONCE || c == OP_ONCE_NC ||
2426 c == OP_COND)
2427 {
2428 BOOL empty_branch;
2429 if (GET(code, 1) == 0) return TRUE; /* Hit unclosed bracket */
2430
2431 /* If a conditional group has only one branch, there is a second, implied,
2432 empty branch, so just skip over the conditional, because it could be empty.
2433 Otherwise, scan the individual branches of the group. */
2434
2435 if (c == OP_COND && code[GET(code, 1)] != OP_ALT)
2436 code += GET(code, 1);
2437 else
2438 {
2439 empty_branch = FALSE;
2440 do
2441 {
2442 if (!empty_branch && could_be_empty_branch(code, endcode, utf, cd))
2443 empty_branch = TRUE;
2444 code += GET(code, 1);
2445 }
2446 while (*code == OP_ALT);
2447 if (!empty_branch) return FALSE; /* All branches are non-empty */
2448 }
2449
2450 c = *code;
2451 continue;
2452 }
2453
2454 /* Handle the other opcodes */
2455
2456 switch (c)
2457 {
2458 /* Check for quantifiers after a class. XCLASS is used for classes that
2459 cannot be represented just by a bit map. This includes negated single
2460 high-valued characters. The length in PRIV(OP_lengths)[] is zero; the
2461 actual length is stored in the compiled code, so we must update "code"
2462 here. */
2463
2464 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
2465 case OP_XCLASS:
2466 ccode = code += GET(code, 1);
2467 goto CHECK_CLASS_REPEAT;
2468 #endif
2469
2470 case OP_CLASS:
2471 case OP_NCLASS:
2472 ccode = code + PRIV(OP_lengths)[OP_CLASS];
2473
2474 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
2475 CHECK_CLASS_REPEAT:
2476 #endif
2477
2478 switch (*ccode)
2479 {
2480 case OP_CRSTAR: /* These could be empty; continue */
2481 case OP_CRMINSTAR:
2482 case OP_CRQUERY:
2483 case OP_CRMINQUERY:
2484 break;
2485
2486 default: /* Non-repeat => class must match */
2487 case OP_CRPLUS: /* These repeats aren't empty */
2488 case OP_CRMINPLUS:
2489 return FALSE;
2490
2491 case OP_CRRANGE:
2492 case OP_CRMINRANGE:
2493 if (GET2(ccode, 1) > 0) return FALSE; /* Minimum > 0 */
2494 break;
2495 }
2496 break;
2497
2498 /* Opcodes that must match a character */
2499
2500 case OP_PROP:
2501 case OP_NOTPROP:
2502 case OP_EXTUNI:
2503 case OP_NOT_DIGIT:
2504 case OP_DIGIT:
2505 case OP_NOT_WHITESPACE:
2506 case OP_WHITESPACE:
2507 case OP_NOT_WORDCHAR:
2508 case OP_WORDCHAR:
2509 case OP_ANY:
2510 case OP_ALLANY:
2511 case OP_ANYBYTE:
2512 case OP_CHAR:
2513 case OP_CHARI:
2514 case OP_NOT:
2515 case OP_NOTI:
2516 case OP_PLUS:
2517 case OP_MINPLUS:
2518 case OP_POSPLUS:
2519 case OP_EXACT:
2520 case OP_NOTPLUS:
2521 case OP_NOTMINPLUS:
2522 case OP_NOTPOSPLUS:
2523 case OP_NOTEXACT:
2524 case OP_TYPEPLUS:
2525 case OP_TYPEMINPLUS:
2526 case OP_TYPEPOSPLUS:
2527 case OP_TYPEEXACT:
2528 return FALSE;
2529
2530 /* These are going to continue, as they may be empty, but we have to
2531 fudge the length for the \p and \P cases. */
2532
2533 case OP_TYPESTAR:
2534 case OP_TYPEMINSTAR:
2535 case OP_TYPEPOSSTAR:
2536 case OP_TYPEQUERY:
2537 case OP_TYPEMINQUERY:
2538 case OP_TYPEPOSQUERY:
2539 if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
2540 break;
2541
2542 /* Same for these */
2543
2544 case OP_TYPEUPTO:
2545 case OP_TYPEMINUPTO:
2546 case OP_TYPEPOSUPTO:
2547 if (code[1 + IMM2_SIZE] == OP_PROP || code[1 + IMM2_SIZE] == OP_NOTPROP)
2548 code += 2;
2549 break;
2550
2551 /* End of branch */
2552
2553 case OP_KET:
2554 case OP_KETRMAX:
2555 case OP_KETRMIN:
2556 case OP_KETRPOS:
2557 case OP_ALT:
2558 return TRUE;
2559
2560 /* In UTF-8 mode, STAR, MINSTAR, POSSTAR, QUERY, MINQUERY, POSQUERY, UPTO,
2561 MINUPTO, and POSUPTO may be followed by a multibyte character */
2562
2563 #ifdef SUPPORT_UTF
2564 case OP_STAR:
2565 case OP_STARI:
2566 case OP_MINSTAR:
2567 case OP_MINSTARI:
2568 case OP_POSSTAR:
2569 case OP_POSSTARI:
2570 case OP_QUERY:
2571 case OP_QUERYI:
2572 case OP_MINQUERY:
2573 case OP_MINQUERYI:
2574 case OP_POSQUERY:
2575 case OP_POSQUERYI:
2576 if (utf && HAS_EXTRALEN(code[1])) code += GET_EXTRALEN(code[1]);
2577 break;
2578
2579 case OP_UPTO:
2580 case OP_UPTOI:
2581 case OP_MINUPTO:
2582 case OP_MINUPTOI:
2583 case OP_POSUPTO:
2584 case OP_POSUPTOI:
2585 if (utf && HAS_EXTRALEN(code[1 + IMM2_SIZE])) code += GET_EXTRALEN(code[1 + IMM2_SIZE]);
2586 break;
2587 #endif
2588
2589 /* MARK, and PRUNE/SKIP/THEN with an argument must skip over the argument
2590 string. */
2591
2592 case OP_MARK:
2593 case OP_PRUNE_ARG:
2594 case OP_SKIP_ARG:
2595 code += code[1];
2596 break;
2597
2598 case OP_THEN_ARG:
2599 code += code[1];
2600 break;
2601
2602 /* None of the remaining opcodes are required to match a character. */
2603
2604 default:
2605 break;
2606 }
2607 }
2608
2609 return TRUE;
2610 }
2611
2612
2613
2614 /*************************************************
2615 * Scan compiled regex for non-emptiness *
2616 *************************************************/
2617
2618 /* This function is called to check for left recursive calls. We want to check
2619 the current branch of the current pattern to see if it could match the empty
2620 string. If it could, we must look outwards for branches at other levels,
2621 stopping when we pass beyond the bracket which is the subject of the recursion.
2622 This function is called only during the real compile, not during the
2623 pre-compile.
2624
2625 Arguments:
2626 code points to start of the recursion
2627 endcode points to where to stop (current RECURSE item)
2628 bcptr points to the chain of current (unclosed) branch starts
2629 utf TRUE if in UTF-8 / UTF-16 mode
2630 cd pointers to tables etc
2631
2632 Returns: TRUE if what is matched could be empty
2633 */
2634
2635 static BOOL
2636 could_be_empty(const pcre_uchar *code, const pcre_uchar *endcode,
2637 branch_chain *bcptr, BOOL utf, compile_data *cd)
2638 {
2639 while (bcptr != NULL && bcptr->current_branch >= code)
2640 {
2641 if (!could_be_empty_branch(bcptr->current_branch, endcode, utf, cd))
2642 return FALSE;
2643 bcptr = bcptr->outer;
2644 }
2645 return TRUE;
2646 }
2647
2648
2649
2650 /*************************************************
2651 * Check for POSIX class syntax *
2652 *************************************************/
2653
2654 /* This function is called when the sequence "[:" or "[." or "[=" is
2655 encountered in a character class. It checks whether this is followed by a
2656 sequence of characters terminated by a matching ":]" or ".]" or "=]". If we
2657 reach an unescaped ']' without the special preceding character, return FALSE.
2658
2659 Originally, this function only recognized a sequence of letters between the
2660 terminators, but it seems that Perl recognizes any sequence of characters,
2661 though of course unknown POSIX names are subsequently rejected. Perl gives an
2662 "Unknown POSIX class" error for [:f\oo:] for example, where previously PCRE
2663 didn't consider this to be a POSIX class. Likewise for [:1234:].
2664
2665 The problem in trying to be exactly like Perl is in the handling of escapes. We
2666 have to be sure that [abc[:x\]pqr] is *not* treated as containing a POSIX
2667 class, but [abc[:x\]pqr:]] is (so that an error can be generated). The code
2668 below handles the special case of \], but does not try to do any other escape
2669 processing. This makes it different from Perl for cases such as [:l\ower:]
2670 where Perl recognizes it as the POSIX class "lower" but PCRE does not recognize
2671 "l\ower". This is a lesser evil that not diagnosing bad classes when Perl does,
2672 I think.
2673
2674 A user pointed out that PCRE was rejecting [:a[:digit:]] whereas Perl was not.
2675 It seems that the appearance of a nested POSIX class supersedes an apparent
2676 external class. For example, [:a[:digit:]b:] matches "a", "b", ":", or
2677 a digit.
2678
2679 In Perl, unescaped square brackets may also appear as part of class names. For
2680 example, [:a[:abc]b:] gives unknown POSIX class "[:abc]b:]". However, for
2681 [:a[:abc]b][b:] it gives unknown POSIX class "[:abc]b][b:]", which does not
2682 seem right at all. PCRE does not allow closing square brackets in POSIX class
2683 names.
2684
2685 Arguments:
2686 ptr pointer to the initial [
2687 endptr where to return the end pointer
2688
2689 Returns: TRUE or FALSE
2690 */
2691
2692 static BOOL
2693 check_posix_syntax(const pcre_uchar *ptr, const pcre_uchar **endptr)
2694 {
2695 int terminator; /* Don't combine these lines; the Solaris cc */
2696 terminator = *(++ptr); /* compiler warns about "non-constant" initializer. */
2697 for (++ptr; *ptr != 0; ptr++)
2698 {
2699 if (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET)
2700 ptr++;
2701 else if (*ptr == CHAR_RIGHT_SQUARE_BRACKET) return FALSE;
2702 else
2703 {
2704 if (*ptr == terminator && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET)
2705 {
2706 *endptr = ptr;
2707 return TRUE;
2708 }
2709 if (*ptr == CHAR_LEFT_SQUARE_BRACKET &&
2710 (ptr[1] == CHAR_COLON || ptr[1] == CHAR_DOT ||
2711 ptr[1] == CHAR_EQUALS_SIGN) &&
2712 check_posix_syntax(ptr, endptr))
2713 return FALSE;
2714 }
2715 }
2716 return FALSE;
2717 }
2718
2719
2720
2721
2722 /*************************************************
2723 * Check POSIX class name *
2724 *************************************************/
2725
2726 /* This function is called to check the name given in a POSIX-style class entry
2727 such as [:alnum:].
2728
2729 Arguments:
2730 ptr points to the first letter
2731 len the length of the name
2732
2733 Returns: a value representing the name, or -1 if unknown
2734 */
2735
2736 static int
2737 check_posix_name(const pcre_uchar *ptr, int len)
2738 {
2739 const char *pn = posix_names;
2740 register int yield = 0;
2741 while (posix_name_lengths[yield] != 0)
2742 {
2743 if (len == posix_name_lengths[yield] &&
2744 STRNCMP_UC_C8(ptr, pn, len) == 0) return yield;
2745 pn += posix_name_lengths[yield] + 1;
2746 yield++;
2747 }
2748 return -1;
2749 }
2750
2751
2752 /*************************************************
2753 * Adjust OP_RECURSE items in repeated group *
2754 *************************************************/
2755
2756 /* OP_RECURSE items contain an offset from the start of the regex to the group
2757 that is referenced. This means that groups can be replicated for fixed
2758 repetition simply by copying (because the recursion is allowed to refer to
2759 earlier groups that are outside the current group). However, when a group is
2760 optional (i.e. the minimum quantifier is zero), OP_BRAZERO or OP_SKIPZERO is
2761 inserted before it, after it has been compiled. This means that any OP_RECURSE
2762 items within it that refer to the group itself or any contained groups have to
2763 have their offsets adjusted. That one of the jobs of this function. Before it
2764 is called, the partially compiled regex must be temporarily terminated with
2765 OP_END.
2766
2767 This function has been extended with the possibility of forward references for
2768 recursions and subroutine calls. It must also check the list of such references
2769 for the group we are dealing with. If it finds that one of the recursions in
2770 the current group is on this list, it adjusts the offset in the list, not the
2771 value in the reference (which is a group number).
2772
2773 Arguments:
2774 group points to the start of the group
2775 adjust the amount by which the group is to be moved
2776 utf TRUE in UTF-8 / UTF-16 mode
2777 cd contains pointers to tables etc.
2778 save_hwm the hwm forward reference pointer at the start of the group
2779
2780 Returns: nothing
2781 */
2782
2783 static void
2784 adjust_recurse(pcre_uchar *group, int adjust, BOOL utf, compile_data *cd,
2785 pcre_uchar *save_hwm)
2786 {
2787 pcre_uchar *ptr = group;
2788
2789 while ((ptr = (pcre_uchar *)find_recurse(ptr, utf)) != NULL)
2790 {
2791 int offset;
2792 pcre_uchar *hc;
2793
2794 /* See if this recursion is on the forward reference list. If so, adjust the
2795 reference. */
2796
2797 for (hc = save_hwm; hc < cd->hwm; hc += LINK_SIZE)
2798 {
2799 offset = GET(hc, 0);
2800 if (cd->start_code + offset == ptr + 1)
2801 {
2802 PUT(hc, 0, offset + adjust);
2803 break;
2804 }
2805 }
2806
2807 /* Otherwise, adjust the recursion offset if it's after the start of this
2808 group. */
2809
2810 if (hc >= cd->hwm)
2811 {
2812 offset = GET(ptr, 1);
2813 if (cd->start_code + offset >= group) PUT(ptr, 1, offset + adjust);
2814 }
2815
2816 ptr += 1 + LINK_SIZE;
2817 }
2818 }
2819
2820
2821
2822 /*************************************************
2823 * Insert an automatic callout point *
2824 *************************************************/
2825
2826 /* This function is called when the PCRE_AUTO_CALLOUT option is set, to insert
2827 callout points before each pattern item.
2828
2829 Arguments:
2830 code current code pointer
2831 ptr current pattern pointer
2832 cd pointers to tables etc
2833
2834 Returns: new code pointer
2835 */
2836
2837 static pcre_uchar *
2838 auto_callout(pcre_uchar *code, const pcre_uchar *ptr, compile_data *cd)
2839 {
2840 *code++ = OP_CALLOUT;
2841 *code++ = 255;
2842 PUT(code, 0, (int)(ptr - cd->start_pattern)); /* Pattern offset */
2843 PUT(code, LINK_SIZE, 0); /* Default length */
2844 return code + 2 * LINK_SIZE;
2845 }
2846
2847
2848
2849 /*************************************************
2850 * Complete a callout item *
2851 *************************************************/
2852
2853 /* A callout item contains the length of the next item in the pattern, which
2854 we can't fill in till after we have reached the relevant point. This is used
2855 for both automatic and manual callouts.
2856
2857 Arguments:
2858 previous_callout points to previous callout item
2859 ptr current pattern pointer
2860 cd pointers to tables etc
2861
2862 Returns: nothing
2863 */
2864
2865 static void
2866 complete_callout(pcre_uchar *previous_callout, const pcre_uchar *ptr, compile_data *cd)
2867 {
2868 int length = (int)(ptr - cd->start_pattern - GET(previous_callout, 2));
2869 PUT(previous_callout, 2 + LINK_SIZE, length);
2870 }
2871
2872
2873
2874 #ifdef SUPPORT_UCP
2875 /*************************************************
2876 * Get othercase range *
2877 *************************************************/
2878
2879 /* This function is passed the start and end of a class range, in UTF-8 mode
2880 with UCP support. It searches up the characters, looking for ranges of
2881 characters in the "other" case. Each call returns the next one, updating the
2882 start address. A character with multiple other cases is returned on its own
2883 with a special return value.
2884
2885 Arguments:
2886 cptr points to starting character value; updated
2887 d end value
2888 ocptr where to put start of othercase range
2889 odptr where to put end of othercase range
2890
2891 Yield: -1 when no more
2892 0 when a range is returned
2893 >0 the CASESET offset for char with multiple other cases
2894 in this case, ocptr contains the original
2895 */
2896
2897 static int
2898 get_othercase_range(unsigned int *cptr, unsigned int d, unsigned int *ocptr,
2899 unsigned int *odptr)
2900 {
2901 unsigned int c, othercase, next;
2902 int co;
2903
2904 /* Find the first character that has an other case. If it has multiple other
2905 cases, return its case offset value. */
2906
2907 for (c = *cptr; c <= d; c++)
2908 {
2909 if ((co = UCD_CASESET(c)) != 0)
2910 {
2911 *ocptr = c++; /* Character that has the set */
2912 *cptr = c; /* Rest of input range */
2913 return co;
2914 }
2915 if ((othercase = UCD_OTHERCASE(c)) != c) break;
2916 }
2917
2918 if (c > d) return -1; /* Reached end of range */
2919
2920 *ocptr = othercase;
2921 next = othercase + 1;
2922
2923 for (++c; c <= d; c++)
2924 {
2925 if (UCD_OTHERCASE(c) != next) break;
2926 next++;
2927 }
2928
2929 *odptr = next - 1; /* End of othercase range */
2930 *cptr = c; /* Rest of input range */
2931 return 0;
2932 }
2933
2934
2935
2936 /*************************************************
2937 * Check a character and a property *
2938 *************************************************/
2939
2940 /* This function is called by check_auto_possessive() when a property item
2941 is adjacent to a fixed character.
2942
2943 Arguments:
2944 c the character
2945 ptype the property type
2946 pdata the data for the type
2947 negated TRUE if it's a negated property (\P or \p{^)
2948
2949 Returns: TRUE if auto-possessifying is OK
2950 */
2951
2952 static BOOL
2953 check_char_prop(int c, int ptype, int pdata, BOOL negated)
2954 {
2955 #ifdef SUPPORT_UCP
2956 const pcre_uint32 *p;
2957 #endif
2958
2959 const ucd_record *prop = GET_UCD(c);
2960
2961 switch(ptype)
2962 {
2963 case PT_LAMP:
2964 return (prop->chartype == ucp_Lu ||
2965 prop->chartype == ucp_Ll ||
2966 prop->chartype == ucp_Lt) == negated;
2967
2968 case PT_GC:
2969 return (pdata == PRIV(ucp_gentype)[prop->chartype]) == negated;
2970
2971 case PT_PC:
2972 return (pdata == prop->chartype) == negated;
2973
2974 case PT_SC:
2975 return (pdata == prop->script) == negated;
2976
2977 /* These are specials */
2978
2979 case PT_ALNUM:
2980 return (PRIV(ucp_gentype)[prop->chartype] == ucp_L ||
2981 PRIV(ucp_gentype)[prop->chartype] == ucp_N) == negated;
2982
2983 case PT_SPACE: /* Perl space */
2984 return (PRIV(ucp_gentype)[prop->chartype] == ucp_Z ||
2985 c == CHAR_HT || c == CHAR_NL || c == CHAR_FF || c == CHAR_CR)
2986 == negated;
2987
2988 case PT_PXSPACE: /* POSIX space */
2989 return (PRIV(ucp_gentype)[prop->chartype] == ucp_Z ||
2990 c == CHAR_HT || c == CHAR_NL || c == CHAR_VT ||
2991 c == CHAR_FF || c == CHAR_CR)
2992 == negated;
2993
2994 case PT_WORD:
2995 return (PRIV(ucp_gentype)[prop->chartype] == ucp_L ||
2996 PRIV(ucp_gentype)[prop->chartype] == ucp_N ||
2997 c == CHAR_UNDERSCORE) == negated;
2998
2999 #ifdef SUPPORT_UCP
3000 case PT_CLIST:
3001 p = PRIV(ucd_caseless_sets) + prop->caseset;
3002 for (;;)
3003 {
3004 if ((unsigned int)c < *p) return !negated;
3005 if ((unsigned int)c == *p++) return negated;
3006 }
3007 break; /* Control never reaches here */
3008 #endif
3009 }
3010
3011 return FALSE;
3012 }
3013 #endif /* SUPPORT_UCP */
3014
3015
3016
3017 /*************************************************
3018 * Check if auto-possessifying is possible *
3019 *************************************************/
3020
3021 /* This function is called for unlimited repeats of certain items, to see
3022 whether the next thing could possibly match the repeated item. If not, it makes
3023 sense to automatically possessify the repeated item.
3024
3025 Arguments:
3026 previous pointer to the repeated opcode
3027 utf TRUE in UTF-8 / UTF-16 mode
3028 ptr next character in pattern
3029 options options bits
3030 cd contains pointers to tables etc.
3031
3032 Returns: TRUE if possessifying is wanted
3033 */
3034
3035 static BOOL
3036 check_auto_possessive(const pcre_uchar *previous, BOOL utf,
3037 const pcre_uchar *ptr, int options, compile_data *cd)
3038 {
3039 pcre_int32 c = NOTACHAR;
3040 pcre_int32 next;
3041 int op_code = *previous++;
3042
3043 /* Skip whitespace and comments in extended mode */
3044
3045 if ((options & PCRE_EXTENDED) != 0)
3046 {
3047 for (;;)
3048 {
3049 while (MAX_255(*ptr) && (cd->ctypes[*ptr] & ctype_space) != 0) ptr++;
3050 if (*ptr == CHAR_NUMBER_SIGN)
3051 {
3052 ptr++;
3053 while (*ptr != 0)
3054 {
3055 if (IS_NEWLINE(ptr)) { ptr += cd->nllen; break; }
3056 ptr++;
3057 #ifdef SUPPORT_UTF
3058 if (utf) FORWARDCHAR(ptr);
3059 #endif
3060 }
3061 }
3062 else break;
3063 }
3064 }
3065
3066 /* If the next item is one that we can handle, get its value. A non-negative
3067 value is a character, a negative value is an escape value. */
3068
3069 if (*ptr == CHAR_BACKSLASH)
3070 {
3071 int temperrorcode = 0;
3072 next = check_escape(&ptr, &temperrorcode, cd->bracount, options, FALSE);
3073 if (temperrorcode != 0) return FALSE;
3074 ptr++; /* Point after the escape sequence */
3075 }
3076 else if (!MAX_255(*ptr) || (cd->ctypes[*ptr] & ctype_meta) == 0)
3077 {
3078 #ifdef SUPPORT_UTF
3079 if (utf) { GETCHARINC(next, ptr); } else
3080 #endif
3081 next = *ptr++;
3082 }
3083 else return FALSE;
3084
3085 /* Skip whitespace and comments in extended mode */
3086
3087 if ((options & PCRE_EXTENDED) != 0)
3088 {
3089 for (;;)
3090 {
3091 while (MAX_255(*ptr) && (cd->ctypes[*ptr] & ctype_space) != 0) ptr++;
3092 if (*ptr == CHAR_NUMBER_SIGN)
3093 {
3094 ptr++;
3095 while (*ptr != 0)
3096 {
3097 if (IS_NEWLINE(ptr)) { ptr += cd->nllen; break; }
3098 ptr++;
3099 #ifdef SUPPORT_UTF
3100 if (utf) FORWARDCHAR(ptr);
3101 #endif
3102 }
3103 }
3104 else break;
3105 }
3106 }
3107
3108 /* If the next thing is itself optional, we have to give up. */
3109
3110 if (*ptr == CHAR_ASTERISK || *ptr == CHAR_QUESTION_MARK ||
3111 STRNCMP_UC_C8(ptr, STR_LEFT_CURLY_BRACKET STR_0 STR_COMMA, 3) == 0)
3112 return FALSE;
3113
3114 /* If the previous item is a character, get its value. */
3115
3116 if (op_code == OP_CHAR || op_code == OP_CHARI ||
3117 op_code == OP_NOT || op_code == OP_NOTI)
3118 {
3119 #ifdef SUPPORT_UTF
3120 GETCHARTEST(c, previous);
3121 #else
3122 c = *previous;
3123 #endif
3124 }
3125
3126 /* Now compare the next item with the previous opcode. First, handle cases when
3127 the next item is a character. */
3128
3129 if (next >= 0)
3130 {
3131 /* For a caseless UTF match, the next character may have more than one other
3132 case, which maps to the special PT_CLIST property. Check this first. */
3133
3134 #ifdef SUPPORT_UCP
3135 if (utf && (unsigned int)c != NOTACHAR && (options & PCRE_CASELESS) != 0)
3136 {
3137 int ocs = UCD_CASESET(next);
3138 if (ocs > 0) return check_char_prop(c, PT_CLIST, ocs, op_code >= OP_NOT);
3139 }
3140 #endif
3141
3142 switch(op_code)
3143 {
3144 case OP_CHAR:
3145 return c != next;
3146
3147 /* For CHARI (caseless character) we must check the other case. If we have
3148 Unicode property support, we can use it to test the other case of
3149 high-valued characters. We know that next can have only one other case,
3150 because multi-other-case characters are dealt with above. */
3151
3152 case OP_CHARI:
3153 if (c == next) return FALSE;
3154 #ifdef SUPPORT_UTF
3155 if (utf)
3156 {
3157 unsigned int othercase;
3158 if (next < 128) othercase = cd->fcc[next]; else
3159 #ifdef SUPPORT_UCP
3160 othercase = UCD_OTHERCASE((unsigned int)next);
3161 #else
3162 othercase = NOTACHAR;
3163 #endif
3164 return (unsigned int)c != othercase;
3165 }
3166 else
3167 #endif /* SUPPORT_UTF */
3168 return (c != TABLE_GET((unsigned int)next, cd->fcc, next)); /* Not UTF */
3169
3170 case OP_NOT:
3171 return c == next;
3172
3173 case OP_NOTI:
3174 if (c == next) return TRUE;
3175 #ifdef SUPPORT_UTF
3176 if (utf)
3177 {
3178 unsigned int othercase;
3179 if (next < 128) othercase = cd->fcc[next]; else
3180 #ifdef SUPPORT_UCP
3181 othercase = UCD_OTHERCASE((unsigned int)next);
3182 #else
3183 othercase = NOTACHAR;
3184 #endif
3185 return (unsigned int)c == othercase;
3186 }
3187 else
3188 #endif /* SUPPORT_UTF */
3189 return (c == TABLE_GET((unsigned int)next, cd->fcc, next)); /* Not UTF */
3190
3191 /* Note that OP_DIGIT etc. are generated only when PCRE_UCP is *not* set.
3192 When it is set, \d etc. are converted into OP_(NOT_)PROP codes. */
3193
3194 case OP_DIGIT:
3195 return next > 255 || (cd->ctypes[next] & ctype_digit) == 0;
3196
3197 case OP_NOT_DIGIT:
3198 return next <= 255 && (cd->ctypes[next] & ctype_digit) != 0;
3199
3200 case OP_WHITESPACE:
3201 return next > 255 || (cd->ctypes[next] & ctype_space) == 0;
3202
3203 case OP_NOT_WHITESPACE:
3204 return next <= 255 && (cd->ctypes[next] & ctype_space) != 0;
3205
3206 case OP_WORDCHAR:
3207 return next > 255 || (cd->ctypes[next] & ctype_word) == 0;
3208
3209 case OP_NOT_WORDCHAR:
3210 return next <= 255 && (cd->ctypes[next] & ctype_word) != 0;
3211
3212 case OP_HSPACE:
3213 case OP_NOT_HSPACE:
3214 switch(next)
3215 {
3216 HSPACE_CASES:
3217 return op_code == OP_NOT_HSPACE;
3218
3219 default:
3220 return op_code != OP_NOT_HSPACE;
3221 }
3222
3223 case OP_ANYNL:
3224 case OP_VSPACE:
3225 case OP_NOT_VSPACE:
3226 switch(next)
3227 {
3228 VSPACE_CASES:
3229 return op_code == OP_NOT_VSPACE;
3230
3231 default:
3232 return op_code != OP_NOT_VSPACE;
3233 }
3234
3235 #ifdef SUPPORT_UCP
3236 case OP_PROP:
3237 return check_char_prop(next, previous[0], previous[1], FALSE);
3238
3239 case OP_NOTPROP:
3240 return check_char_prop(next, previous[0], previous[1], TRUE);
3241 #endif
3242
3243 default:
3244 return FALSE;
3245 }
3246 }
3247
3248 /* Handle the case when the next item is \d, \s, etc. Note that when PCRE_UCP
3249 is set, \d turns into ESC_du rather than ESC_d, etc., so ESC_d etc. are
3250 generated only when PCRE_UCP is *not* set, that is, when only ASCII
3251 characteristics are recognized. Similarly, the opcodes OP_DIGIT etc. are
3252 replaced by OP_PROP codes when PCRE_UCP is set. */
3253
3254 switch(op_code)
3255 {
3256 case OP_CHAR:
3257 case OP_CHARI:
3258 switch(-next)
3259 {
3260 case ESC_d:
3261 return c > 255 || (cd->ctypes[c] & ctype_digit) == 0;
3262
3263 case ESC_D:
3264 return c <= 255 && (cd->ctypes[c] & ctype_digit) != 0;
3265
3266 case ESC_s:
3267 return c > 255 || (cd->ctypes[c] & ctype_space) == 0;
3268
3269 case ESC_S:
3270 return c <= 255 && (cd->ctypes[c] & ctype_space) != 0;
3271
3272 case ESC_w:
3273 return c > 255 || (cd->ctypes[c] & ctype_word) == 0;
3274
3275 case ESC_W:
3276 return c <= 255 && (cd->ctypes[c] & ctype_word) != 0;
3277
3278 case ESC_h:
3279 case ESC_H:
3280 switch(c)
3281 {
3282 HSPACE_CASES:
3283 return -next != ESC_h;
3284
3285 default:
3286 return -next == ESC_h;
3287 }
3288
3289 case ESC_v:
3290 case ESC_V:
3291 switch(c)
3292 {
3293 VSPACE_CASES:
3294 return -next != ESC_v;
3295
3296 default:
3297 return -next == ESC_v;
3298 }
3299
3300 /* When PCRE_UCP is set, these values get generated for \d etc. Find
3301 their substitutions and process them. The result will always be either
3302 -ESC_p or -ESC_P. Then fall through to process those values. */
3303
3304 #ifdef SUPPORT_UCP
3305 case ESC_du:
3306 case ESC_DU:
3307 case ESC_wu:
3308 case ESC_WU:
3309 case ESC_su:
3310 case ESC_SU:
3311 {
3312 int temperrorcode = 0;
3313 ptr = substitutes[-next - ESC_DU];
3314 next = check_escape(&ptr, &temperrorcode, 0, options, FALSE);
3315 if (temperrorcode != 0) return FALSE;
3316 ptr++; /* For compatibility */
3317 }
3318 /* Fall through */
3319
3320 case ESC_p:
3321 case ESC_P:
3322 {
3323 int ptype, pdata, errorcodeptr;
3324 BOOL negated;
3325
3326 ptr--; /* Make ptr point at the p or P */
3327 ptype = get_ucp(&ptr, &negated, &pdata, &errorcodeptr);
3328 if (ptype < 0) return FALSE;
3329 ptr++; /* Point past the final curly ket */
3330
3331 /* If the property item is optional, we have to give up. (When generated
3332 from \d etc by PCRE_UCP, this test will have been applied much earlier,
3333 to the original \d etc. At this point, ptr will point to a zero byte. */
3334
3335 if (*ptr == CHAR_ASTERISK || *ptr == CHAR_QUESTION_MARK ||
3336 STRNCMP_UC_C8(ptr, STR_LEFT_CURLY_BRACKET STR_0 STR_COMMA, 3) == 0)
3337 return FALSE;
3338
3339 /* Do the property check. */
3340
3341 return check_char_prop(c, ptype, pdata, (next == -ESC_P) != negated);
3342 }
3343 #endif
3344
3345 default:
3346 return FALSE;
3347 }
3348
3349 /* In principle, support for Unicode properties should be integrated here as
3350 well. It means re-organizing the above code so as to get hold of the property
3351 values before switching on the op-code. However, I wonder how many patterns
3352 combine ASCII \d etc with Unicode properties? (Note that if PCRE_UCP is set,
3353 these op-codes are never generated.) */
3354
3355 case OP_DIGIT:
3356 return next == -ESC_D || next == -ESC_s || next == -ESC_W ||
3357 next == -ESC_h || next == -ESC_v || next == -ESC_R;
3358
3359 case OP_NOT_DIGIT:
3360 return next == -ESC_d;
3361
3362 case OP_WHITESPACE:
3363 return next == -ESC_S || next == -ESC_d || next == -ESC_w;
3364
3365 case OP_NOT_WHITESPACE:
3366 return next == -ESC_s || next == -ESC_h || next == -ESC_v || next == -ESC_R;
3367
3368 case OP_HSPACE:
3369 return next == -ESC_S || next == -ESC_H || next == -ESC_d ||
3370 next == -ESC_w || next == -ESC_v || next == -ESC_R;
3371
3372 case OP_NOT_HSPACE:
3373 return next == -ESC_h;
3374
3375 /* Can't have \S in here because VT matches \S (Perl anomaly) */
3376 case OP_ANYNL:
3377 case OP_VSPACE:
3378 return next == -ESC_V || next == -ESC_d || next == -ESC_w;
3379
3380 case OP_NOT_VSPACE:
3381 return next == -ESC_v || next == -ESC_R;
3382
3383 case OP_WORDCHAR:
3384 return next == -ESC_W || next == -ESC_s || next == -ESC_h ||
3385 next == -ESC_v || next == -ESC_R;
3386
3387 case OP_NOT_WORDCHAR:
3388 return next == -ESC_w || next == -ESC_d;
3389
3390 default:
3391 return FALSE;
3392 }
3393
3394 /* Control does not reach here */
3395 }
3396
3397
3398
3399 /*************************************************
3400 * Add a character or range to a class *
3401 *************************************************/
3402
3403 /* This function packages up the logic of adding a character or range of
3404 characters to a class. The character values in the arguments will be within the
3405 valid values for the current mode (8-bit, 16-bit, UTF, etc). This function is
3406 mutually recursive with the function immediately below.
3407
3408 Arguments:
3409 classbits the bit map for characters < 256
3410 uchardptr points to the pointer for extra data
3411 options the options word
3412 cd contains pointers to tables etc.
3413 start start of range character
3414 end end of range character
3415
3416 Returns: the number of < 256 characters added
3417 the pointer to extra data is updated
3418 */
3419
3420 static int
3421 add_to_class(pcre_uint8 *classbits, pcre_uchar **uchardptr, int options,
3422 compile_data *cd, unsigned int start, unsigned int end)
3423 {
3424 unsigned int c;
3425 int n8 = 0;
3426
3427 /* If caseless matching is required, scan the range and process alternate
3428 cases. In Unicode, there are 8-bit characters that have alternate cases that
3429 are greater than 255 and vice-versa. Sometimes we can just extend the original
3430 range. */
3431
3432 if ((options & PCRE_CASELESS) != 0)
3433 {
3434 #ifdef SUPPORT_UCP
3435 if ((options & PCRE_UTF8) != 0)
3436 {
3437 int rc;
3438 unsigned int oc, od;
3439
3440 options &= ~PCRE_CASELESS; /* Remove for recursive calls */
3441 c = start;
3442
3443 while ((rc = get_othercase_range(&c, end, &oc, &od)) >= 0)
3444 {
3445 /* Handle a single character that has more than one other case. */
3446
3447 if (rc > 0) n8 += add_list_to_class(classbits, uchardptr, options, cd,
3448 PRIV(ucd_caseless_sets) + rc, oc);
3449
3450 /* Do nothing if the other case range is within the original range. */
3451
3452 else if (oc >= start && od <= end) continue;
3453
3454 /* Extend the original range if there is overlap, noting that if oc < c, we
3455 can't have od > end because a subrange is always shorter than the basic
3456 range. Otherwise, use a recursive call to add the additional range. */
3457
3458 else if (oc < start && od >= start - 1) start = oc; /* Extend downwards */
3459 else if (od > end && oc <= end + 1) end = od; /* Extend upwards */
3460 else n8 += add_to_class(classbits, uchardptr, options, cd, oc, od);
3461 }
3462 }
3463 else
3464 #endif /* SUPPORT_UCP */
3465
3466 /* Not UTF-mode, or no UCP */
3467
3468 for (c = start; c <= end && c < 256; c++)
3469 {
3470 SETBIT(classbits, cd->fcc[c]);
3471 n8++;
3472 }
3473 }
3474
3475 /* Now handle the original range. Adjust the final value according to the bit
3476 length - this means that the same lists of (e.g.) horizontal spaces can be used
3477 in all cases. */
3478
3479 #ifdef COMPILE_PCRE8
3480 #ifdef SUPPORT_UTF
3481 if ((options & PCRE_UTF8) == 0)
3482 #endif
3483 if (end > 0xff) end = 0xff;
3484 #endif
3485
3486 #ifdef COMPILE_PCRE16
3487 #ifdef SUPPORT_UTF
3488 if ((options & PCRE_UTF16) == 0)
3489 #endif
3490 if (end > 0xffff) end = 0xffff;
3491 #endif
3492
3493 /* If all characters are less than 256, use the bit map. Otherwise use extra
3494 data. */
3495
3496 if (end < 0x100)
3497 {
3498 for (c = start; c <= end; c++)
3499 {
3500 n8++;
3501 SETBIT(classbits, c);
3502 }
3503 }
3504
3505 else
3506 {
3507 pcre_uchar *uchardata = *uchardptr;
3508
3509 #ifdef SUPPORT_UTF
3510 if ((options & PCRE_UTF8) != 0) /* All UTFs use the same flag bit */
3511 {
3512 if (start < end)
3513 {
3514 *uchardata++ = XCL_RANGE;
3515 uchardata += PRIV(ord2utf)(start, uchardata);
3516 uchardata += PRIV(ord2utf)(end, uchardata);
3517 }
3518 else if (start == end)
3519 {
3520 *uchardata++ = XCL_SINGLE;
3521 uchardata += PRIV(ord2utf)(start, uchardata);
3522 }
3523 }
3524 else
3525 #endif /* SUPPORT_UTF */
3526
3527 /* Without UTF support, character values are constrained by the bit length,
3528 and can only be > 256 for 16-bit and 32-bit libraries. */
3529
3530 #ifdef COMPILE_PCRE8
3531 {}
3532 #else
3533 if (start < end)
3534 {
3535 *uchardata++ = XCL_RANGE;
3536 *uchardata++ = start;
3537 *uchardata++ = end;
3538 }
3539 else if (start == end)
3540 {
3541 *uchardata++ = XCL_SINGLE;
3542 *uchardata++ = start;
3543 }
3544 #endif
3545
3546 *uchardptr = uchardata; /* Updata extra data pointer */
3547 }
3548
3549 return n8; /* Number of 8-bit characters */
3550 }
3551
3552
3553
3554
3555 /*************************************************
3556 * Add a list of characters to a class *
3557 *************************************************/
3558
3559 /* This function is used for adding a list of case-equivalent characters to a
3560 class, and also for adding a list of horizontal or vertical whitespace. If the
3561 list is in order (which it should be), ranges of characters are detected and
3562 handled appropriately. This function is mutually recursive with the function
3563 above.
3564
3565 Arguments:
3566 classbits the bit map for characters < 256
3567 uchardptr points to the pointer for extra data
3568 options the options word
3569 cd contains pointers to tables etc.
3570 p points to row of 32-bit values, terminated by NOTACHAR
3571 except character to omit; this is used when adding lists of
3572 case-equivalent characters to avoid including the one we
3573 already know about
3574
3575 Returns: the number of < 256 characters added
3576 the pointer to extra data is updated
3577 */
3578
3579 static int
3580 add_list_to_class(pcre_uint8 *classbits, pcre_uchar **uchardptr, int options,
3581 compile_data *cd, const pcre_uint32 *p, unsigned int except)
3582 {
3583 int n8 = 0;
3584 while (p[0] < NOTACHAR)
3585 {
3586 int n = 0;
3587 if (p[0] != except)
3588 {
3589 while(p[n+1] == p[0] + n + 1) n++;
3590 n8 += add_to_class(classbits, uchardptr, options, cd, p[0], p[n]);
3591 }
3592 p += n + 1;
3593 }
3594 return n8;
3595 }
3596
3597
3598
3599 /*************************************************
3600 * Add characters not in a list to a class *
3601 *************************************************/
3602
3603 /* This function is used for adding the complement of a list of horizontal or
3604 vertical whitespace to a class. The list must be in order.
3605
3606 Arguments:
3607 classbits the bit map for characters < 256
3608 uchardptr points to the pointer for extra data
3609 options the options word
3610 cd contains pointers to tables etc.
3611 p points to row of 32-bit values, terminated by NOTACHAR
3612
3613 Returns: the number of < 256 characters added
3614 the pointer to extra data is updated
3615 */
3616
3617 static int
3618 add_not_list_to_class(pcre_uint8 *classbits, pcre_uchar **uchardptr,
3619 int options, compile_data *cd, const pcre_uint32 *p)
3620 {
3621 int n8 = 0;
3622 if (p[0] > 0)
3623 n8 += add_to_class(classbits, uchardptr, options, cd, 0, p[0] - 1);
3624 while (p[0] < NOTACHAR)
3625 {
3626 while (p[1] == p[0] + 1) p++;
3627 n8 += add_to_class(classbits, uchardptr, options, cd, p[0] + 1,
3628 (p[1] == NOTACHAR)? 0x10ffff : p[1] - 1);
3629 p++;
3630 }
3631 return n8;
3632 }
3633
3634
3635
3636 /*************************************************
3637 * Compile one branch *
3638 *************************************************/
3639
3640 /* Scan the pattern, compiling it into the a vector. If the options are
3641 changed during the branch, the pointer is used to change the external options
3642 bits. This function is used during the pre-compile phase when we are trying
3643 to find out the amount of memory needed, as well as during the real compile
3644 phase. The value of lengthptr distinguishes the two phases.
3645
3646 Arguments:
3647 optionsptr pointer to the option bits
3648 codeptr points to the pointer to the current code point
3649 ptrptr points to the current pattern pointer
3650 errorcodeptr points to error code variable
3651 firstcharptr set to initial literal character, or < 0 (REQ_UNSET, REQ_NONE)
3652 reqcharptr set to the last literal character required, else < 0
3653 bcptr points to current branch chain
3654 cond_depth conditional nesting depth
3655 cd contains pointers to tables etc.
3656 lengthptr NULL during the real compile phase
3657 points to length accumulator during pre-compile phase
3658
3659 Returns: TRUE on success
3660 FALSE, with *errorcodeptr set non-zero on error
3661 */
3662
3663 static BOOL
3664 compile_branch(int *optionsptr, pcre_uchar **codeptr,
3665 const pcre_uchar **ptrptr, int *errorcodeptr, pcre_int32 *firstcharptr,
3666 pcre_int32 *reqcharptr, branch_chain *bcptr, int cond_depth,
3667 compile_data *cd, int *lengthptr)
3668 {
3669 int repeat_type, op_type;
3670 int repeat_min = 0, repeat_max = 0; /* To please picky compilers */
3671 int bravalue = 0;
3672 int greedy_default, greedy_non_default;
3673 pcre_int32 firstchar, reqchar;
3674 pcre_int32 zeroreqchar, zerofirstchar;
3675 pcre_int32 req_caseopt, reqvary, tempreqvary;
3676 int options = *optionsptr; /* May change dynamically */
3677 int after_manual_callout = 0;
3678 int length_prevgroup = 0;
3679 register int c;
3680 register pcre_uchar *code = *codeptr;
3681 pcre_uchar *last_code = code;
3682 pcre_uchar *orig_code = code;
3683 pcre_uchar *tempcode;
3684 BOOL inescq = FALSE;
3685 BOOL groupsetfirstchar = FALSE;
3686 const pcre_uchar *ptr = *ptrptr;
3687 const pcre_uchar *tempptr;
3688 const pcre_uchar *nestptr = NULL;
3689 pcre_uchar *previous = NULL;
3690 pcre_uchar *previous_callout = NULL;
3691 pcre_uchar *save_hwm = NULL;
3692 pcre_uint8 classbits[32];
3693
3694 /* We can fish out the UTF-8 setting once and for all into a BOOL, but we
3695 must not do this for other options (e.g. PCRE_EXTENDED) because they may change
3696 dynamically as we process the pattern. */
3697
3698 #ifdef SUPPORT_UTF
3699 /* PCRE_UTF16 has the same value as PCRE_UTF8. */
3700 BOOL utf = (options & PCRE_UTF8) != 0;
3701 pcre_uchar utf_chars[6];
3702 #else
3703 BOOL utf = FALSE;
3704 #endif
3705
3706 /* Helper variables for OP_XCLASS opcode (for characters > 255). We define
3707 class_uchardata always so that it can be passed to add_to_class() always,
3708 though it will not be used in non-UTF 8-bit cases. This avoids having to supply
3709 alternative calls for the different cases. */
3710
3711 pcre_uchar *class_uchardata;
3712 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
3713 BOOL xclass;
3714 pcre_uchar *class_uchardata_base;
3715 #endif
3716
3717 #ifdef PCRE_DEBUG
3718 if (lengthptr != NULL) DPRINTF((">> start branch\n"));
3719 #endif
3720
3721 /* Set up the default and non-default settings for greediness */
3722
3723 greedy_default = ((options & PCRE_UNGREEDY) != 0);
3724 greedy_non_default = greedy_default ^ 1;
3725
3726 /* Initialize no first byte, no required byte. REQ_UNSET means "no char
3727 matching encountered yet". It gets changed to REQ_NONE if we hit something that
3728 matches a non-fixed char first char; reqchar just remains unset if we never
3729 find one.
3730
3731 When we hit a repeat whose minimum is zero, we may have to adjust these values
3732 to take the zero repeat into account. This is implemented by setting them to
3733 zerofirstbyte and zeroreqchar when such a repeat is encountered. The individual
3734 item types that can be repeated set these backoff variables appropriately. */
3735
3736 firstchar = reqchar = zerofirstchar = zeroreqchar = REQ_UNSET;
3737
3738 /* The variable req_caseopt contains either the REQ_CASELESS value
3739 or zero, according to the current setting of the caseless flag. The
3740 REQ_CASELESS leaves the lower 28 bit empty. It is added into the
3741 firstchar or reqchar variables to record the case status of the
3742 value. This is used only for ASCII characters. */
3743
3744 req_caseopt = ((options & PCRE_CASELESS) != 0)? REQ_CASELESS:0;
3745
3746 /* Switch on next character until the end of the branch */
3747
3748 for (;; ptr++)
3749 {
3750 BOOL negate_class;
3751 BOOL should_flip_negation;
3752 BOOL possessive_quantifier;
3753 BOOL is_quantifier;
3754 BOOL is_recurse;
3755 BOOL reset_bracount;
3756 int class_has_8bitchar;
3757 int class_one_char;
3758 int newoptions;
3759 int recno;
3760 int refsign;
3761 int skipbytes;
3762 int subreqchar;
3763 int subfirstchar;
3764 int terminator;
3765 int mclength;
3766 int tempbracount;
3767 pcre_uchar mcbuffer[8];
3768
3769 /* Get next character in the pattern */
3770
3771 c = *ptr;
3772
3773 /* If we are at the end of a nested substitution, revert to the outer level
3774 string. Nesting only happens one level deep. */
3775
3776 if (c == 0 && nestptr != NULL)
3777 {
3778 ptr = nestptr;
3779 nestptr = NULL;
3780 c = *ptr;
3781 }
3782
3783 /* If we are in the pre-compile phase, accumulate the length used for the
3784 previous cycle of this loop. */
3785
3786 if (lengthptr != NULL)
3787 {
3788 #ifdef PCRE_DEBUG
3789 if (code > cd->hwm) cd->hwm = code; /* High water info */
3790 #endif
3791 if (code > cd->start_workspace + cd->workspace_size -
3792 WORK_SIZE_SAFETY_MARGIN) /* Check for overrun */
3793 {
3794 *errorcodeptr = ERR52;
3795 goto FAILED;
3796 }
3797
3798 /* There is at least one situation where code goes backwards: this is the
3799 case of a zero quantifier after a class (e.g. [ab]{0}). At compile time,
3800 the class is simply eliminated. However, it is created first, so we have to
3801 allow memory for it. Therefore, don't ever reduce the length at this point.
3802 */
3803
3804 if (code < last_code) code = last_code;
3805
3806 /* Paranoid check for integer overflow */
3807
3808 if (OFLOW_MAX - *lengthptr < code - last_code)
3809 {
3810 *errorcodeptr = ERR20;
3811 goto FAILED;
3812 }
3813
3814 *lengthptr += (int)(code - last_code);
3815 DPRINTF(("length=%d added %d c=%c (0x%x)\n", *lengthptr,
3816 (int)(code - last_code), c, c));
3817
3818 /* If "previous" is set and it is not at the start of the work space, move
3819 it back to there, in order to avoid filling up the work space. Otherwise,
3820 if "previous" is NULL, reset the current code pointer to the start. */
3821
3822 if (previous != NULL)
3823 {
3824 if (previous > orig_code)
3825 {
3826 memmove(orig_code, previous, IN_UCHARS(code - previous));
3827 code -= previous - orig_code;
3828 previous = orig_code;
3829 }
3830 }
3831 else code = orig_code;
3832
3833 /* Remember where this code item starts so we can pick up the length
3834 next time round. */
3835
3836 last_code = code;
3837 }
3838
3839 /* In the real compile phase, just check the workspace used by the forward
3840 reference list. */
3841
3842 else if (cd->hwm > cd->start_workspace + cd->workspace_size -
3843 WORK_SIZE_SAFETY_MARGIN)
3844 {
3845 *errorcodeptr = ERR52;
3846 goto FAILED;
3847 }
3848
3849 /* If in \Q...\E, check for the end; if not, we have a literal */
3850
3851 if (inescq && c != 0)
3852 {
3853 if (c == CHAR_BACKSLASH && ptr[1] == CHAR_E)
3854 {
3855 inescq = FALSE;
3856 ptr++;
3857 continue;
3858 }
3859 else
3860 {
3861 if (previous_callout != NULL)
3862 {
3863 if (lengthptr == NULL) /* Don't attempt in pre-compile phase */
3864 complete_callout(previous_callout, ptr, cd);
3865 previous_callout = NULL;
3866 }
3867 if ((options & PCRE_AUTO_CALLOUT) != 0)
3868 {
3869 previous_callout = code;
3870 code = auto_callout(code, ptr, cd);
3871 }
3872 goto NORMAL_CHAR;
3873 }
3874 }
3875
3876 /* Fill in length of a previous callout, except when the next thing is
3877 a quantifier. */
3878
3879 is_quantifier =
3880 c == CHAR_ASTERISK || c == CHAR_PLUS || c == CHAR_QUESTION_MARK ||
3881 (c == CHAR_LEFT_CURLY_BRACKET && is_counted_repeat(ptr+1));
3882
3883 if (!is_quantifier && previous_callout != NULL &&
3884 after_manual_callout-- <= 0)
3885 {
3886 if (lengthptr == NULL) /* Don't attempt in pre-compile phase */
3887 complete_callout(previous_callout, ptr, cd);
3888 previous_callout = NULL;
3889 }
3890
3891 /* In extended mode, skip white space and comments. */
3892
3893 if ((options & PCRE_EXTENDED) != 0)
3894 {
3895 if (MAX_255(*ptr) && (cd->ctypes[c] & ctype_space) != 0) continue;
3896 if (c == CHAR_NUMBER_SIGN)
3897 {
3898 ptr++;
3899 while (*ptr != 0)
3900 {
3901 if (IS_NEWLINE(ptr)) { ptr += cd->nllen - 1; break; }
3902 ptr++;
3903 #ifdef SUPPORT_UTF
3904 if (utf) FORWARDCHAR(ptr);
3905 #endif
3906 }
3907 if (*ptr != 0) continue;
3908
3909 /* Else fall through to handle end of string */
3910 c = 0;
3911 }
3912 }
3913
3914 /* No auto callout for quantifiers. */
3915
3916 if ((options & PCRE_AUTO_CALLOUT) != 0 && !is_quantifier)
3917 {
3918 previous_callout = code;
3919 code = auto_callout(code, ptr, cd);
3920 }
3921
3922 switch(c)
3923 {
3924 /* ===================================================================*/
3925 case 0: /* The branch terminates at string end */
3926 case CHAR_VERTICAL_LINE: /* or | or ) */
3927 case CHAR_RIGHT_PARENTHESIS:
3928 *firstcharptr = firstchar;
3929 *reqcharptr = reqchar;
3930 *codeptr = code;
3931 *ptrptr = ptr;
3932 if (lengthptr != NULL)
3933 {
3934 if (OFLOW_MAX - *lengthptr < code - last_code)
3935 {
3936 *errorcodeptr = ERR20;
3937 goto FAILED;
3938 }
3939 *lengthptr += (int)(code - last_code); /* To include callout length */
3940 DPRINTF((">> end branch\n"));
3941 }
3942 return TRUE;
3943
3944
3945 /* ===================================================================*/
3946 /* Handle single-character metacharacters. In multiline mode, ^ disables
3947 the setting of any following char as a first character. */
3948
3949 case CHAR_CIRCUMFLEX_ACCENT:
3950 previous = NULL;
3951 if ((options & PCRE_MULTILINE) != 0)
3952 {
3953 if (firstchar == REQ_UNSET) firstchar = REQ_NONE;
3954 *code++ = OP_CIRCM;
3955 }
3956 else *code++ = OP_CIRC;
3957 break;
3958
3959 case CHAR_DOLLAR_SIGN:
3960 previous = NULL;
3961 *code++ = ((options & PCRE_MULTILINE) != 0)? OP_DOLLM : OP_DOLL;
3962 break;
3963
3964 /* There can never be a first char if '.' is first, whatever happens about
3965 repeats. The value of reqchar doesn't change either. */
3966
3967 case CHAR_DOT:
3968 if (firstchar == REQ_UNSET) firstchar = REQ_NONE;
3969 zerofirstchar = firstchar;
3970 zeroreqchar = reqchar;
3971 previous = code;
3972 *code++ = ((options & PCRE_DOTALL) != 0)? OP_ALLANY: OP_ANY;
3973 break;
3974
3975
3976 /* ===================================================================*/
3977 /* Character classes. If the included characters are all < 256, we build a
3978 32-byte bitmap of the permitted characters, except in the special case
3979 where there is only one such character. For negated classes, we build the
3980 map as usual, then invert it at the end. However, we use a different opcode
3981 so that data characters > 255 can be handled correctly.
3982
3983 If the class contains characters outside the 0-255 range, a different
3984 opcode is compiled. It may optionally have a bit map for characters < 256,
3985 but those above are are explicitly listed afterwards. A flag byte tells
3986 whether the bitmap is present, and whether this is a negated class or not.
3987
3988 In JavaScript compatibility mode, an isolated ']' causes an error. In
3989 default (Perl) mode, it is treated as a data character. */
3990
3991 case CHAR_RIGHT_SQUARE_BRACKET:
3992 if ((cd->external_options & PCRE_JAVASCRIPT_COMPAT) != 0)
3993 {
3994 *errorcodeptr = ERR64;
3995 goto FAILED;
3996 }
3997 goto NORMAL_CHAR;
3998
3999 case CHAR_LEFT_SQUARE_BRACKET:
4000 previous = code;
4001
4002 /* PCRE supports POSIX class stuff inside a class. Perl gives an error if
4003 they are encountered at the top level, so we'll do that too. */
4004
4005 if ((ptr[1] == CHAR_COLON || ptr[1] == CHAR_DOT ||
4006 ptr[1] == CHAR_EQUALS_SIGN) &&
4007 check_posix_syntax(ptr, &tempptr))
4008 {
4009 *errorcodeptr = (ptr[1] == CHAR_COLON)? ERR13 : ERR31;
4010 goto FAILED;
4011 }
4012
4013 /* If the first character is '^', set the negation flag and skip it. Also,
4014 if the first few characters (either before or after ^) are \Q\E or \E we
4015 skip them too. This makes for compatibility with Perl. */
4016
4017 negate_class = FALSE;
4018 for (;;)
4019 {
4020 c = *(++ptr);
4021 if (c == CHAR_BACKSLASH)
4022 {
4023 if (ptr[1] == CHAR_E)
4024 ptr++;
4025 else if (STRNCMP_UC_C8(ptr + 1, STR_Q STR_BACKSLASH STR_E, 3) == 0)
4026 ptr += 3;
4027 else
4028 break;
4029 }
4030 else if (!negate_class && c == CHAR_CIRCUMFLEX_ACCENT)
4031 negate_class = TRUE;
4032 else break;
4033 }
4034
4035 /* Empty classes are allowed in JavaScript compatibility mode. Otherwise,
4036 an initial ']' is taken as a data character -- the code below handles
4037 that. In JS mode, [] must always fail, so generate OP_FAIL, whereas
4038 [^] must match any character, so generate OP_ALLANY. */
4039
4040 if (c == CHAR_RIGHT_SQUARE_BRACKET &&
4041 (cd->external_options & PCRE_JAVASCRIPT_COMPAT) != 0)
4042 {
4043 *code++ = negate_class? OP_ALLANY : OP_FAIL;
4044 if (firstchar == REQ_UNSET) firstchar = REQ_NONE;
4045 zerofirstchar = firstchar;
4046 break;
4047 }
4048
4049 /* If a class contains a negative special such as \S, we need to flip the
4050 negation flag at the end, so that support for characters > 255 works
4051 correctly (they are all included in the class). */
4052
4053 should_flip_negation = FALSE;
4054
4055 /* For optimization purposes, we track some properties of the class:
4056 class_has_8bitchar will be non-zero if the class contains at least one <
4057 256 character; class_one_char will be 1 if the class contains just one
4058 character. */
4059
4060 class_has_8bitchar = 0;
4061 class_one_char = 0;
4062
4063 /* Initialize the 32-char bit map to all zeros. We build the map in a
4064 temporary bit of memory, in case the class contains fewer than two
4065 8-bit characters because in that case the compiled code doesn't use the bit
4066 map. */
4067
4068 memset(classbits, 0, 32 * sizeof(pcre_uint8));
4069
4070 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
4071 xclass = FALSE;
4072 class_uchardata = code + LINK_SIZE + 2; /* For XCLASS items */
4073 class_uchardata_base = class_uchardata; /* Save the start */
4074 #endif
4075
4076 /* Process characters until ] is reached. By writing this as a "do" it
4077 means that an initial ] is taken as a data character. At the start of the
4078 loop, c contains the first byte of the character. */
4079
4080 if (c != 0) do
4081 {
4082 const pcre_uchar *oldptr;
4083
4084 #ifdef SUPPORT_UTF
4085 if (utf && HAS_EXTRALEN(c))
4086 { /* Braces are required because the */
4087 GETCHARLEN(c, ptr, ptr); /* macro generates multiple statements */
4088 }
4089 #endif
4090
4091 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
4092 /* In the pre-compile phase, accumulate the length of any extra
4093 data and reset the pointer. This is so that very large classes that
4094 contain a zillion > 255 characters no longer overwrite the work space
4095 (which is on the stack). We have to remember that there was XCLASS data,
4096 however. */
4097
4098 if (lengthptr != NULL && class_uchardata > class_uchardata_base)
4099 {
4100 xclass = TRUE;
4101 *lengthptr += class_uchardata - class_uchardata_base;
4102 class_uchardata = class_uchardata_base;
4103 }
4104 #endif
4105
4106 /* Inside \Q...\E everything is literal except \E */
4107
4108 if (inescq)
4109 {
4110 if (c == CHAR_BACKSLASH && ptr[1] == CHAR_E) /* If we are at \E */
4111 {
4112 inescq = FALSE; /* Reset literal state */
4113 ptr++; /* Skip the 'E' */
4114 continue; /* Carry on with next */
4115 }
4116 goto CHECK_RANGE; /* Could be range if \E follows */
4117 }
4118
4119 /* Handle POSIX class names. Perl allows a negation extension of the
4120 form [:^name:]. A square bracket that doesn't match the syntax is
4121 treated as a literal. We also recognize the POSIX constructions
4122 [.ch.] and [=ch=] ("collating elements") and fault them, as Perl
4123 5.6 and 5.8 do. */
4124
4125 if (c == CHAR_LEFT_SQUARE_BRACKET &&
4126 (ptr[1] == CHAR_COLON || ptr[1] == CHAR_DOT ||
4127 ptr[1] == CHAR_EQUALS_SIGN) && check_posix_syntax(ptr, &tempptr))
4128 {
4129 BOOL local_negate = FALSE;
4130 int posix_class, taboffset, tabopt;
4131 register const pcre_uint8 *cbits = cd->cbits;
4132 pcre_uint8 pbits[32];
4133
4134 if (ptr[1] != CHAR_COLON)
4135 {
4136 *errorcodeptr = ERR31;
4137 goto FAILED;
4138 }
4139
4140 ptr += 2;
4141 if (*ptr == CHAR_CIRCUMFLEX_ACCENT)
4142 {
4143 local_negate = TRUE;
4144 should_flip_negation = TRUE; /* Note negative special */
4145 ptr++;
4146 }
4147
4148 posix_class = check_posix_name(ptr, (int)(tempptr - ptr));
4149 if (posix_class < 0)
4150 {
4151 *errorcodeptr = ERR30;
4152 goto FAILED;
4153 }
4154
4155 /* If matching is caseless, upper and lower are converted to
4156 alpha. This relies on the fact that the class table starts with
4157 alpha, lower, upper as the first 3 entries. */
4158
4159 if ((options & PCRE_CASELESS) != 0 && posix_class <= 2)
4160 posix_class = 0;
4161
4162 /* When PCRE_UCP is set, some of the POSIX classes are converted to
4163 different escape sequences that use Unicode properties. */
4164
4165 #ifdef SUPPORT_UCP
4166 if ((options & PCRE_UCP) != 0)
4167 {
4168 int pc = posix_class + ((local_negate)? POSIX_SUBSIZE/2 : 0);
4169 if (posix_substitutes[pc] != NULL)
4170 {
4171 nestptr = tempptr + 1;
4172 ptr = posix_substitutes[pc] - 1;
4173 continue;
4174 }
4175 }
4176 #endif
4177 /* In the non-UCP case, we build the bit map for the POSIX class in a
4178 chunk of local store because we may be adding and subtracting from it,
4179 and we don't want to subtract bits that may be in the main map already.
4180 At the end we or the result into the bit map that is being built. */
4181
4182 posix_class *= 3;
4183
4184 /* Copy in the first table (always present) */
4185
4186 memcpy(pbits, cbits + posix_class_maps[posix_class],
4187 32 * sizeof(pcre_uint8));
4188
4189 /* If there is a second table, add or remove it as required. */
4190
4191 taboffset = posix_class_maps[posix_class + 1];
4192 tabopt = posix_class_maps[posix_class + 2];
4193
4194 if (taboffset >= 0)
4195 {
4196 if (tabopt >= 0)
4197 for (c = 0; c < 32; c++) pbits[c] |= cbits[c + taboffset];
4198 else
4199 for (c = 0; c < 32; c++) pbits[c] &= ~cbits[c + taboffset];
4200 }
4201
4202 /* Now see if we need to remove any special characters. An option
4203 value of 1 removes vertical space and 2 removes underscore. */
4204
4205 if (tabopt < 0) tabopt = -tabopt;
4206 if (tabopt == 1) pbits[1] &= ~0x3c;
4207 else if (tabopt == 2) pbits[11] &= 0x7f;
4208
4209 /* Add the POSIX table or its complement into the main table that is
4210 being built and we are done. */
4211
4212 if (local_negate)
4213 for (c = 0; c < 32; c++) classbits[c] |= ~pbits[c];
4214 else
4215 for (c = 0; c < 32; c++) classbits[c] |= pbits[c];
4216
4217 ptr = tempptr + 1;
4218 /* Every class contains at least one < 256 character. */
4219 class_has_8bitchar = 1;
4220 /* Every class contains at least two characters. */
4221 class_one_char = 2;
4222 continue; /* End of POSIX syntax handling */
4223 }
4224
4225 /* Backslash may introduce a single character, or it may introduce one
4226 of the specials, which just set a flag. The sequence \b is a special
4227 case. Inside a class (and only there) it is treated as backspace. We
4228 assume that other escapes have more than one character in them, so
4229 speculatively set both class_has_8bitchar and class_one_char bigger
4230 than one. Unrecognized escapes fall through and are either treated
4231 as literal characters (by default), or are faulted if
4232 PCRE_EXTRA is set. */
4233
4234 if (c == CHAR_BACKSLASH)
4235 {
4236 c = check_escape(&ptr, errorcodeptr, cd->bracount, options, TRUE);
4237 if (*errorcodeptr != 0) goto FAILED;
4238
4239 if (-c == ESC_b) c = CHAR_BS; /* \b is backspace in a class */
4240 else if (-c == ESC_N) /* \N is not supported in a class */
4241 {
4242 *errorcodeptr = ERR71;
4243 goto FAILED;
4244 }
4245 else if (-c == ESC_Q) /* Handle start of quoted string */
4246 {
4247 if (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
4248 {
4249 ptr += 2; /* avoid empty string */
4250 }
4251 else inescq = TRUE;
4252 continue;
4253 }
4254 else if (-c == ESC_E) continue; /* Ignore orphan \E */
4255
4256 if (c < 0)
4257 {
4258 register const pcre_uint8 *cbits = cd->cbits;
4259 /* Every class contains at least two < 256 characters. */
4260 class_has_8bitchar++;
4261 /* Every class contains at least two characters. */
4262 class_one_char += 2;
4263
4264 switch (-c)
4265 {
4266 #ifdef SUPPORT_UCP
4267 case ESC_du: /* These are the values given for \d etc */
4268 case ESC_DU: /* when PCRE_UCP is set. We replace the */
4269 case ESC_wu: /* escape sequence with an appropriate \p */
4270 case ESC_WU: /* or \P to test Unicode properties instead */
4271 case ESC_su: /* of the default ASCII testing. */
4272 case ESC_SU:
4273 nestptr = ptr;
4274 ptr = substitutes[-c - ESC_DU] - 1; /* Just before substitute */
4275 class_has_8bitchar--; /* Undo! */
4276 continue;
4277 #endif
4278 case ESC_d:
4279 for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_digit];
4280 continue;
4281
4282 case ESC_D:
4283 should_flip_negation = TRUE;
4284 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_digit];
4285 continue;
4286
4287 case ESC_w:
4288 for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_word];
4289 continue;
4290
4291 case ESC_W:
4292 should_flip_negation = TRUE;
4293 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_word];
4294 continue;
4295
4296 /* Perl 5.004 onwards omits VT from \s, but we must preserve it
4297 if it was previously set by something earlier in the character
4298 class. Luckily, the value of CHAR_VT is 0x0b in both ASCII and
4299 EBCDIC, so we lazily just adjust the appropriate bit. */
4300
4301 case ESC_s:
4302 classbits[0] |= cbits[cbit_space];
4303 classbits[1] |= cbits[cbit_space+1] & ~0x08;
4304 for (c = 2; c < 32; c++) classbits[c] |= cbits[c+cbit_space];
4305 continue;
4306
4307 case ESC_S:
4308 should_flip_negation = TRUE;
4309 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_space];
4310 classbits[1] |= 0x08; /* Perl 5.004 onwards omits VT from \s */
4311 continue;
4312
4313 /* The rest apply in both UCP and non-UCP cases. */
4314
4315 case ESC_h:
4316 (void)add_list_to_class(classbits, &class_uchardata, options, cd,
4317 PRIV(hspace_list), NOTACHAR);
4318 continue;
4319
4320 case ESC_H:
4321 (void)add_not_list_to_class(classbits, &class_uchardata, options,
4322 cd, PRIV(hspace_list));
4323 continue;
4324
4325 case ESC_v:
4326 (void)add_list_to_class(classbits, &class_uchardata, options, cd,
4327 PRIV(vspace_list), NOTACHAR);
4328 continue;
4329
4330 case ESC_V:
4331 (void)add_not_list_to_class(classbits, &class_uchardata, options,
4332 cd, PRIV(vspace_list));
4333 continue;
4334
4335 #ifdef SUPPORT_UCP
4336 case ESC_p:
4337 case ESC_P:
4338 {
4339 BOOL negated;
4340 int pdata;
4341 int ptype = get_ucp(&ptr, &negated, &pdata, errorcodeptr);
4342 if (ptype < 0) goto FAILED;
4343 *class_uchardata++ = ((-c == ESC_p) != negated)?
4344 XCL_PROP : XCL_NOTPROP;
4345 *class_uchardata++ = ptype;
4346 *class_uchardata++ = pdata;
4347 class_has_8bitchar--; /* Undo! */
4348 continue;
4349 }
4350 #endif
4351 /* Unrecognized escapes are faulted if PCRE is running in its
4352 strict mode. By default, for compatibility with Perl, they are
4353 treated as literals. */
4354
4355 default:
4356 if ((options & PCRE_EXTRA) != 0)
4357 {
4358 *errorcodeptr = ERR7;
4359 goto FAILED;
4360 }
4361 class_has_8bitchar--; /* Undo the speculative increase. */
4362 class_one_char -= 2; /* Undo the speculative increase. */
4363 c = *ptr; /* Get the final character and fall through */
4364 break;
4365 }
4366 }
4367
4368 /* Fall through if the escape just defined a single character (c >= 0).
4369 This may be greater than 256. */
4370
4371 } /* End of backslash handling */
4372
4373 /* A character may be followed by '-' to form a range. However, Perl does
4374 not permit ']' to be the end of the range. A '-' character at the end is
4375 treated as a literal. Perl ignores orphaned \E sequences entirely. The
4376 code for handling \Q and \E is messy. */
4377
4378 CHECK_RANGE:
4379 while (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
4380 {
4381 inescq = FALSE;
4382 ptr += 2;
4383 }
4384 oldptr = ptr;
4385
4386 /* Remember if \r or \n were explicitly used */
4387
4388 if (c == CHAR_CR || c == CHAR_NL) cd->external_flags |= PCRE_HASCRORLF;
4389
4390 /* Check for range */
4391
4392 if (!inescq && ptr[1] == CHAR_MINUS)
4393 {
4394 int d;
4395 ptr += 2;
4396 while (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_E) ptr += 2;
4397
4398 /* If we hit \Q (not followed by \E) at this point, go into escaped
4399 mode. */
4400
4401 while (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_Q)
4402 {
4403 ptr += 2;
4404 if (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_E)
4405 { ptr += 2; continue; }
4406 inescq = TRUE;
4407 break;
4408 }
4409
4410 /* Minus (hyphen) at the end of a class is treated as a literal, so put
4411 back the pointer and jump to handle the character that preceded it. */
4412
4413 if (*ptr == 0 || (!inescq && *ptr == CHAR_RIGHT_SQUARE_BRACKET))
4414 {
4415 ptr = oldptr;
4416 goto CLASS_SINGLE_CHARACTER;
4417 }
4418
4419 /* Otherwise, we have a potential range; pick up the next character */
4420
4421 #ifdef SUPPORT_UTF
4422 if (utf)
4423 { /* Braces are required because the */
4424 GETCHARLEN(d, ptr, ptr); /* macro generates multiple statements */
4425 }
4426 else
4427 #endif
4428 d = *ptr; /* Not UTF-8 mode */
4429
4430 /* The second part of a range can be a single-character escape, but
4431 not any of the other escapes. Perl 5.6 treats a hyphen as a literal
4432 in such circumstances. */
4433
4434 if (!inescq && d == CHAR_BACKSLASH)
4435 {
4436 d = check_escape(&ptr, errorcodeptr, cd->bracount, options, TRUE);
4437 if (*errorcodeptr != 0) goto FAILED;
4438
4439 /* \b is backspace; any other special means the '-' was literal. */
4440
4441 if (d < 0)
4442 {
4443 if (d == -ESC_b) d = CHAR_BS; else
4444 {
4445 ptr = oldptr;
4446 goto CLASS_SINGLE_CHARACTER; /* A few lines below */
4447 }
4448 }
4449 }
4450
4451 /* Check that the two values are in the correct order. Optimize
4452 one-character ranges. */
4453
4454 if (d < c)
4455 {
4456 *errorcodeptr = ERR8;
4457 goto FAILED;
4458 }
4459 if (d == c) goto CLASS_SINGLE_CHARACTER; /* A few lines below */
4460
4461 /* We have found a character range, so single character optimizations
4462 cannot be done anymore. Any value greater than 1 indicates that there
4463 is more than one character. */
4464
4465 class_one_char = 2;
4466
4467 /* Remember an explicit \r or \n, and add the range to the class. */
4468
4469 if (d == CHAR_CR || d == CHAR_NL) cd->external_flags |= PCRE_HASCRORLF;
4470
4471 class_has_8bitchar +=
4472 add_to_class(classbits, &class_uchardata, options, cd, c, d);
4473
4474 continue; /* Go get the next char in the class */
4475 }
4476
4477 /* Handle a single character - we can get here for a normal non-escape
4478 char, or after \ that introduces a single character or for an apparent
4479 range that isn't. Only the value 1 matters for class_one_char, so don't
4480 increase it if it is already 2 or more ... just in case there's a class
4481 with a zillion characters in it. */
4482
4483 CLASS_SINGLE_CHARACTER:
4484 if (class_one_char < 2) class_one_char++;
4485
4486 /* If class_one_char is 1, we have the first single character in the
4487 class, and there have been no prior ranges, or XCLASS items generated by
4488 escapes. If this is the final character in the class, we can optimize by
4489 turning the item into a 1-character OP_CHAR[I] if it's positive, or
4490 OP_NOT[I] if it's negative. In the positive case, it can cause firstchar
4491 to be set. Otherwise, there can be no first char if this item is first,
4492 whatever repeat count may follow. In the case of reqchar, save the
4493 previous value for reinstating. */
4494
4495 if (class_one_char == 1 && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET)
4496 {
4497 ptr++;
4498 zeroreqchar = reqchar;
4499
4500 if (negate_class)
4501 {
4502 #ifdef SUPPORT_UCP
4503 int d;
4504 #endif
4505 if (firstchar == REQ_UNSET) firstchar = REQ_NONE;
4506 zerofirstchar = firstchar;
4507
4508 /* For caseless UTF-8 mode when UCP support is available, check
4509 whether this character has more than one other case. If so, generate
4510 a special OP_NOTPROP item instead of OP_NOTI. */
4511
4512 #ifdef SUPPORT_UCP
4513 if (utf && (options & PCRE_CASELESS) != 0 &&
4514 (d = UCD_CASESET(c)) != 0)
4515 {
4516 *code++ = OP_NOTPROP;
4517 *code++ = PT_CLIST;
4518 *code++ = d;
4519 }
4520 else
4521 #endif
4522 /* Char has only one other case, or UCP not available */
4523
4524 {
4525 *code++ = ((options & PCRE_CASELESS) != 0)? OP_NOTI: OP_NOT;
4526 #ifdef SUPPORT_UTF
4527 if (utf && c > MAX_VALUE_FOR_SINGLE_CHAR)
4528 code += PRIV(ord2utf)(c, code);
4529 else
4530 #endif
4531 *code++ = c;
4532 }
4533
4534 /* We are finished with this character class */
4535
4536 goto END_CLASS;
4537 }
4538
4539 /* For a single, positive character, get the value into mcbuffer, and
4540 then we can handle this with the normal one-character code. */
4541
4542 #ifdef SUPPORT_UTF
4543 if (utf && c > MAX_VALUE_FOR_SINGLE_CHAR)
4544 mclength = PRIV(ord2utf)(c, mcbuffer);
4545 else
4546 #endif
4547 {
4548 mcbuffer[0] = c;
4549 mclength = 1;
4550 }
4551 goto ONE_CHAR;
4552 } /* End of 1-char optimization */
4553
4554 /* There is more than one character in the class, or an XCLASS item
4555 has been generated. Add this character to the class. */
4556
4557 class_has_8bitchar +=
4558 add_to_class(classbits, &class_uchardata, options, cd, c, c);
4559 }
4560
4561 /* Loop until ']' reached. This "while" is the end of the "do" far above.
4562 If we are at the end of an internal nested string, revert to the outer
4563 string. */
4564
4565 while (((c = *(++ptr)) != 0 ||
4566 (nestptr != NULL &&
4567 (ptr = nestptr, nestptr = NULL, c = *(++ptr)) != 0)) &&
4568 (c != CHAR_RIGHT_SQUARE_BRACKET || inescq));
4569
4570 /* Check for missing terminating ']' */
4571
4572 if (c == 0)
4573 {
4574 *errorcodeptr = ERR6;
4575 goto FAILED;
4576 }
4577
4578 /* We will need an XCLASS if data has been placed in class_uchardata. In
4579 the second phase this is a sufficient test. However, in the pre-compile
4580 phase, class_uchardata gets emptied to prevent workspace overflow, so it
4581 only if the very last character in the class needs XCLASS will it contain
4582 anything at this point. For this reason, xclass gets set TRUE above when
4583 uchar_classdata is emptied, and that's why this code is the way it is here
4584 instead of just doing a test on class_uchardata below. */
4585
4586 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
4587 if (class_uchardata > class_uchardata_base) xclass = TRUE;
4588 #endif
4589
4590 /* If this is the first thing in the branch, there can be no first char
4591 setting, whatever the repeat count. Any reqchar setting must remain
4592 unchanged after any kind of repeat. */
4593
4594 if (firstchar == REQ_UNSET) firstchar = REQ_NONE;
4595 zerofirstchar = firstchar;
4596 zeroreqchar = reqchar;
4597
4598 /* If there are characters with values > 255, we have to compile an
4599 extended class, with its own opcode, unless there was a negated special
4600 such as \S in the class, and PCRE_UCP is not set, because in that case all
4601 characters > 255 are in the class, so any that were explicitly given as
4602 well can be ignored. If (when there are explicit characters > 255 that must
4603 be listed) there are no characters < 256, we can omit the bitmap in the
4604 actual compiled code. */
4605
4606 #ifdef SUPPORT_UTF
4607 if (xclass && (!should_flip_negation || (options & PCRE_UCP) != 0))
4608 #elif !defined COMPILE_PCRE8
4609 if (xclass && !should_flip_negation)
4610 #endif
4611 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
4612 {
4613 *class_uchardata++ = XCL_END; /* Marks the end of extra data */
4614 *code++ = OP_XCLASS;
4615 code += LINK_SIZE;
4616 *code = negate_class? XCL_NOT:0;
4617
4618 /* If the map is required, move up the extra data to make room for it;
4619 otherwise just move the code pointer to the end of the extra data. */
4620
4621 if (class_has_8bitchar > 0)
4622 {
4623 *code++ |= XCL_MAP;
4624 memmove(code + (32 / sizeof(pcre_uchar)), code,
4625 IN_UCHARS(class_uchardata - code));
4626 memcpy(code, classbits, 32);
4627 code = class_uchardata + (32 / sizeof(pcre_uchar));
4628 }
4629 else code = class_uchardata;
4630
4631 /* Now fill in the complete length of the item */
4632
4633 PUT(previous, 1, (int)(code - previous));
4634 break; /* End of class handling */
4635 }
4636 #endif
4637
4638 /* If there are no characters > 255, or they are all to be included or
4639 excluded, set the opcode to OP_CLASS or OP_NCLASS, depending on whether the
4640 whole class was negated and whether there were negative specials such as \S
4641 (non-UCP) in the class. Then copy the 32-byte map into the code vector,
4642 negating it if necessary. */
4643
4644 *code++ = (negate_class == should_flip_negation) ? OP_CLASS : OP_NCLASS;
4645 if (lengthptr == NULL) /* Save time in the pre-compile phase */
4646 {
4647 if (negate_class)
4648 for (c = 0; c < 32; c++) classbits[c] = ~classbits[c];
4649 memcpy(code, classbits, 32);
4650 }
4651 code += 32 / sizeof(pcre_uchar);
4652
4653 END_CLASS:
4654 break;
4655
4656
4657 /* ===================================================================*/
4658 /* Various kinds of repeat; '{' is not necessarily a quantifier, but this
4659 has been tested above. */
4660
4661 case CHAR_LEFT_CURLY_BRACKET:
4662 if (!is_quantifier) goto NORMAL_CHAR;
4663 ptr = read_repeat_counts(ptr+1, &repeat_min, &repeat_max, errorcodeptr);
4664 if (*errorcodeptr != 0) goto FAILED;
4665 goto REPEAT;
4666
4667 case CHAR_ASTERISK:
4668 repeat_min = 0;
4669 repeat_max = -1;
4670 goto REPEAT;
4671
4672 case CHAR_PLUS:
4673 repeat_min = 1;
4674 repeat_max = -1;
4675 goto REPEAT;
4676
4677 case CHAR_QUESTION_MARK:
4678 repeat_min = 0;
4679 repeat_max = 1;
4680
4681 REPEAT:
4682 if (previous == NULL)
4683 {
4684 *errorcodeptr = ERR9;
4685 goto FAILED;
4686 }
4687
4688 if (repeat_min == 0)
4689 {
4690 firstchar = zerofirstchar; /* Adjust for zero repeat */
4691 reqchar = zeroreqchar; /* Ditto */
4692 }
4693
4694 /* Remember whether this is a variable length repeat */
4695
4696 reqvary = (repeat_min == repeat_max)? 0 : REQ_VARY;
4697
4698 op_type = 0; /* Default single-char op codes */
4699 possessive_quantifier = FALSE; /* Default not possessive quantifier */
4700
4701 /* Save start of previous item, in case we have to move it up in order to
4702 insert something before it. */
4703
4704 tempcode = previous;
4705
4706 /* If the next character is '+', we have a possessive quantifier. This
4707 implies greediness, whatever the setting of the PCRE_UNGREEDY option.
4708 If the next character is '?' this is a minimizing repeat, by default,
4709 but if PCRE_UNGREEDY is set, it works the other way round. We change the
4710 repeat type to the non-default. */
4711
4712 if (ptr[1] == CHAR_PLUS)
4713 {
4714 repeat_type = 0; /* Force greedy */
4715 possessive_quantifier = TRUE;
4716 ptr++;
4717 }
4718 else if (ptr[1] == CHAR_QUESTION_MARK)
4719 {
4720 repeat_type = greedy_non_default;
4721 ptr++;
4722 }
4723 else repeat_type = greedy_default;
4724
4725 /* If previous was a recursion call, wrap it in atomic brackets so that
4726 previous becomes the atomic group. All recursions were so wrapped in the
4727 past, but it no longer happens for non-repeated recursions. In fact, the
4728 repeated ones could be re-implemented independently so as not to need this,
4729 but for the moment we rely on the code for repeating groups. */
4730
4731 if (*previous == OP_RECURSE)
4732 {
4733 memmove(previous + 1 + LINK_SIZE, previous, IN_UCHARS(1 + LINK_SIZE));
4734 *previous = OP_ONCE;
4735 PUT(previous, 1, 2 + 2*LINK_SIZE);
4736 previous[2 + 2*LINK_SIZE] = OP_KET;
4737 PUT(previous, 3 + 2*LINK_SIZE, 2 + 2*LINK_SIZE);
4738 code += 2 + 2 * LINK_SIZE;
4739 length_prevgroup = 3 + 3*LINK_SIZE;
4740
4741 /* When actually compiling, we need to check whether this was a forward
4742 reference, and if so, adjust the offset. */
4743
4744 if (lengthptr == NULL && cd->hwm >= cd->start_workspace + LINK_SIZE)
4745 {
4746 int offset = GET(cd->hwm, -LINK_SIZE);
4747 if (offset == previous + 1 - cd->start_code)
4748 PUT(cd->hwm, -LINK_SIZE, offset + 1 + LINK_SIZE);
4749 }
4750 }
4751
4752 /* Now handle repetition for the different types of item. */
4753
4754 /* If previous was a character or negated character match, abolish the item
4755 and generate a repeat item instead. If a char item has a minimum of more
4756 than one, ensure that it is set in reqchar - it might not be if a sequence
4757 such as x{3} is the first thing in a branch because the x will have gone
4758 into firstchar instead. */
4759
4760 if (*previous == OP_CHAR || *previous == OP_CHARI
4761 || *previous == OP_NOT || *previous == OP_NOTI)
4762 {
4763 switch (*previous)
4764 {
4765 default: /* Make compiler happy. */
4766 case OP_CHAR: op_type = OP_STAR - OP_STAR; break;
4767 case OP_CHARI: op_type = OP_STARI - OP_STAR; break;
4768 case OP_NOT: op_type = OP_NOTSTAR - OP_STAR; break;
4769 case OP_NOTI: op_type = OP_NOTSTARI - OP_STAR; break;
4770 }
4771
4772 /* Deal with UTF characters that take up more than one character. It's
4773 easier to write this out separately than try to macrify it. Use c to
4774 hold the length of the character in bytes, plus UTF_LENGTH to flag that
4775 it's a length rather than a small character. */
4776
4777 #ifdef SUPPORT_UTF
4778 if (utf && NOT_FIRSTCHAR(code[-1]))
4779 {
4780 pcre_uchar *lastchar = code - 1;
4781 BACKCHAR(lastchar);
4782 c = (int)(code - lastchar); /* Length of UTF-8 character */
4783 memcpy(utf_chars, lastchar, IN_UCHARS(c)); /* Save the char */
4784 c |= UTF_LENGTH; /* Flag c as a length */
4785 }
4786 else
4787 #endif /* SUPPORT_UTF */
4788
4789 /* Handle the case of a single charater - either with no UTF support, or
4790 with UTF disabled, or for a single character UTF character. */
4791 {
4792 c = code[-1];
4793 if (*previous <= OP_CHARI && repeat_min > 1)
4794 reqchar = c | req_caseopt | cd->req_varyopt;
4795 }
4796
4797 /* If the repetition is unlimited, it pays to see if the next thing on
4798 the line is something that cannot possibly match this character. If so,
4799 automatically possessifying this item gains some performance in the case
4800 where the match fails. */
4801
4802 if (!possessive_quantifier &&
4803 repeat_max < 0 &&
4804 check_auto_possessive(previous, utf, ptr + 1, options, cd))
4805 {
4806 repeat_type = 0; /* Force greedy */
4807 possessive_quantifier = TRUE;
4808 }
4809
4810 goto OUTPUT_SINGLE_REPEAT; /* Code shared with single character types */
4811 }
4812
4813 /* If previous was a character type match (\d or similar), abolish it and
4814 create a suitable repeat item. The code is shared with single-character
4815 repeats by setting op_type to add a suitable offset into repeat_type. Note
4816 the the Unicode property types will be present only when SUPPORT_UCP is
4817 defined, but we don't wrap the little bits of code here because it just
4818 makes it horribly messy. */
4819
4820 else if (*previous < OP_EODN)
4821 {
4822 pcre_uchar *oldcode;
4823 int prop_type, prop_value;
4824 op_type = OP_TYPESTAR - OP_STAR; /* Use type opcodes */
4825 c = *previous;
4826
4827 if (!possessive_quantifier &&
4828 repeat_max < 0 &&
4829 check_auto_possessive(previous, utf, ptr + 1, options, cd))
4830 {
4831 repeat_type = 0; /* Force greedy */
4832 possessive_quantifier = TRUE;
4833 }
4834
4835 OUTPUT_SINGLE_REPEAT:
4836 if (*previous == OP_PROP || *previous == OP_NOTPROP)
4837 {
4838 prop_type = previous[1];
4839 prop_value = previous[2];
4840 }
4841 else prop_type = prop_value = -1;
4842
4843 oldcode = code;
4844 code = previous; /* Usually overwrite previous item */
4845
4846 /* If the maximum is zero then the minimum must also be zero; Perl allows
4847 this case, so we do too - by simply omitting the item altogether. */
4848
4849 if (repeat_max == 0) goto END_REPEAT;
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 /* Combine the op_type with the repeat_type */
4862
4863 repeat_type += op_type;
4864
4865 /* A minimum of zero is handled either as the special case * or ?, or as
4866 an UPTO, with the maximum given. */
4867
4868 if (repeat_min == 0)
4869 {
4870 if (repeat_max == -1) *code++ = OP_STAR + repeat_type;
4871 else if (repeat_max == 1) *code++ = OP_QUERY + repeat_type;
4872 else
4873 {
4874 *code++ = OP_UPTO + repeat_type;
4875 PUT2INC(code, 0, repeat_max);
4876 }
4877 }
4878
4879 /* A repeat minimum of 1 is optimized into some special cases. If the
4880 maximum is unlimited, we use OP_PLUS. Otherwise, the original item is
4881 left in place and, if the maximum is greater than 1, we use OP_UPTO with
4882 one less than the maximum. */
4883
4884 else if (repeat_min == 1)
4885 {
4886 if (repeat_max == -1)
4887 *code++ = OP_PLUS + repeat_type;
4888 else
4889 {
4890 code = oldcode; /* leave previous item in place */
4891 if (repeat_max == 1) goto END_REPEAT;
4892 *code++ = OP_UPTO + repeat_type;
4893 PUT2INC(code, 0, repeat_max - 1);
4894 }
4895 }
4896
4897 /* The case {n,n} is just an EXACT, while the general case {n,m} is
4898 handled as an EXACT followed by an UPTO. */
4899
4900 else
4901 {
4902 *code++ = OP_EXACT + op_type; /* NB EXACT doesn't have repeat_type */
4903 PUT2INC(code, 0, repeat_min);
4904
4905 /* If the maximum is unlimited, insert an OP_STAR. Before doing so,
4906 we have to insert the character for the previous code. For a repeated
4907 Unicode property match, there are two extra bytes that define the
4908 required property. In UTF-8 mode, long characters have their length in
4909 c, with the UTF_LENGTH bit as a flag. */
4910
4911 if (repeat_max < 0)
4912 {
4913 #ifdef SUPPORT_UTF
4914 if (utf && (c & UTF_LENGTH) != 0)
4915 {
4916 memcpy(code, utf_chars, IN_UCHARS(c & 7));
4917 code += c & 7;
4918 }
4919 else
4920 #endif
4921 {
4922 *code++ = c;
4923 if (prop_type >= 0)
4924 {
4925 *code++ = prop_type;
4926 *code++ = prop_value;
4927 }
4928 }
4929 *code++ = OP_STAR + repeat_type;
4930 }
4931
4932 /* Else insert an UPTO if the max is greater than the min, again
4933 preceded by the character, for the previously inserted code. If the
4934 UPTO is just for 1 instance, we can use QUERY instead. */
4935
4936 else if (repeat_max != repeat_min)
4937 {
4938 #ifdef SUPPORT_UTF
4939 if (utf && (c & UTF_LENGTH) != 0)
4940 {
4941 memcpy(code, utf_chars, IN_UCHARS(c & 7));
4942 code += c & 7;
4943 }
4944 else
4945 #endif
4946 *code++ = c;
4947 if (prop_type >= 0)
4948 {
4949 *code++ = prop_type;
4950 *code++ = prop_value;
4951 }
4952 repeat_max -= repeat_min;
4953
4954 if (repeat_max == 1)
4955 {
4956 *code++ = OP_QUERY + repeat_type;
4957 }
4958 else
4959 {
4960 *code++ = OP_UPTO + repeat_type;
4961 PUT2INC(code, 0, repeat_max);
4962 }
4963 }
4964 }
4965
4966 /* The character or character type itself comes last in all cases. */
4967
4968 #ifdef SUPPORT_UTF
4969 if (utf && (c & UTF_LENGTH) != 0)
4970 {
4971 memcpy(code, utf_chars, IN_UCHARS(c & 7));
4972 code += c & 7;
4973 }
4974 else
4975 #endif
4976 *code++ = c;
4977
4978 /* For a repeated Unicode property match, there are two extra bytes that
4979 define the required property. */
4980
4981 #ifdef SUPPORT_UCP
4982 if (prop_type >= 0)
4983 {
4984 *code++ = prop_type;
4985 *code++ = prop_value;
4986 }
4987 #endif
4988 }
4989
4990 /* If previous was a character class or a back reference, we put the repeat
4991 stuff after it, but just skip the item if the repeat was {0,0}. */
4992
4993 else if (*previous == OP_CLASS ||
4994 *previous == OP_NCLASS ||
4995 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
4996 *previous == OP_XCLASS ||
4997 #endif
4998 *previous == OP_REF ||
4999 *previous == OP_REFI)
5000 {
5001 if (repeat_max == 0)
5002 {
5003 code = previous;
5004 goto END_REPEAT;
5005 }
5006
5007 /*--------------------------------------------------------------------*/
5008 /* This code is obsolete from release 8.00; the restriction was finally
5009 removed: */
5010
5011 /* All real repeats make it impossible to handle partial matching (maybe
5012 one day we will be able to remove this restriction). */
5013
5014 /* if (repeat_max != 1) cd->external_flags |= PCRE_NOPARTIAL; */
5015 /*--------------------------------------------------------------------*/
5016
5017 if (repeat_min == 0 && repeat_max == -1)
5018 *code++ = OP_CRSTAR + repeat_type;
5019 else if (repeat_min == 1 && repeat_max == -1)
5020 *code++ = OP_CRPLUS + repeat_type;
5021 else if (repeat_min == 0 && repeat_max == 1)
5022 *code++ = OP_CRQUERY + repeat_type;
5023 else
5024 {
5025 *code++ = OP_CRRANGE + repeat_type;
5026 PUT2INC(code, 0, repeat_min);
5027 if (repeat_max == -1) repeat_max = 0; /* 2-byte encoding for max */
5028 PUT2INC(code, 0, repeat_max);
5029 }
5030 }
5031
5032 /* If previous was a bracket group, we may have to replicate it in certain
5033 cases. Note that at this point we can encounter only the "basic" bracket
5034 opcodes such as BRA and CBRA, as this is the place where they get converted
5035 into the more special varieties such as BRAPOS and SBRA. A test for >=
5036 OP_ASSERT and <= OP_COND includes ASSERT, ASSERT_NOT, ASSERTBACK,
5037 ASSERTBACK_NOT, ONCE, BRA, CBRA, and COND. Originally, PCRE did not allow
5038 repetition of assertions, but now it does, for Perl compatibility. */
5039
5040 else if (*previous >= OP_ASSERT && *previous <= OP_COND)
5041 {
5042 register int i;
5043 int len = (int)(code - previous);
5044 pcre_uchar *bralink = NULL;
5045 pcre_uchar *brazeroptr = NULL;
5046
5047 /* Repeating a DEFINE group is pointless, but Perl allows the syntax, so
5048 we just ignore the repeat. */
5049
5050 if (*previous == OP_COND && previous[LINK_SIZE+1] == OP_DEF)
5051 goto END_REPEAT;
5052
5053 /* There is no sense in actually repeating assertions. The only potential
5054 use of repetition is in cases when the assertion is optional. Therefore,
5055 if the minimum is greater than zero, just ignore the repeat. If the
5056 maximum is not not zero or one, set it to 1. */
5057
5058 if (*previous < OP_ONCE) /* Assertion */
5059 {
5060 if (repeat_min > 0) goto END_REPEAT;
5061 if (repeat_max < 0 || repeat_max > 1) repeat_max = 1;
5062 }
5063
5064 /* The case of a zero minimum is special because of the need to stick
5065 OP_BRAZERO in front of it, and because the group appears once in the
5066 data, whereas in other cases it appears the minimum number of times. For
5067 this reason, it is simplest to treat this case separately, as otherwise
5068 the code gets far too messy. There are several special subcases when the
5069 minimum is zero. */
5070
5071 if (repeat_min == 0)
5072 {
5073 /* If the maximum is also zero, we used to just omit the group from the
5074 output altogether, like this:
5075
5076 ** if (repeat_max == 0)
5077 ** {
5078 ** code = previous;
5079 ** goto END_REPEAT;
5080 ** }
5081
5082 However, that fails when a group or a subgroup within it is referenced
5083 as a subroutine from elsewhere in the pattern, so now we stick in
5084 OP_SKIPZERO in front of it so that it is skipped on execution. As we
5085 don't have a list of which groups are referenced, we cannot do this
5086 selectively.
5087
5088 If the maximum is 1 or unlimited, we just have to stick in the BRAZERO
5089 and do no more at this point. However, we do need to adjust any
5090 OP_RECURSE calls inside the group that refer to the group itself or any
5091 internal or forward referenced group, because the offset is from the
5092 start of the whole regex. Temporarily terminate the pattern while doing
5093 this. */
5094
5095 if (repeat_max <= 1) /* Covers 0, 1, and unlimited */
5096 {
5097 *code = OP_END;
5098 adjust_recurse(previous, 1, utf, cd, save_hwm);
5099 memmove(previous + 1, previous, IN_UCHARS(len));
5100 code++;
5101 if (repeat_max == 0)
5102 {
5103 *previous++ = OP_SKIPZERO;
5104 goto END_REPEAT;
5105 }
5106 brazeroptr = previous; /* Save for possessive optimizing */
5107 *previous++ = OP_BRAZERO + repeat_type;
5108 }
5109
5110 /* If the maximum is greater than 1 and limited, we have to replicate
5111 in a nested fashion, sticking OP_BRAZERO before each set of brackets.
5112 The first one has to be handled carefully because it's the original
5113 copy, which has to be moved up. The remainder can be handled by code
5114 that is common with the non-zero minimum case below. We have to
5115 adjust the value or repeat_max, since one less copy is required. Once
5116 again, we may have to adjust any OP_RECURSE calls inside the group. */
5117
5118 else
5119 {
5120 int offset;
5121 *code = OP_END;
5122 adjust_recurse(previous, 2 + LINK_SIZE, utf, cd, save_hwm);
5123 memmove(previous + 2 + LINK_SIZE, previous, IN_UCHARS(len));
5124 code += 2 + LINK_SIZE;
5125 *previous++ = OP_BRAZERO + repeat_type;
5126 *previous++ = OP_BRA;
5127
5128 /* We chain together the bracket offset fields that have to be
5129 filled in later when the ends of the brackets are reached. */
5130
5131 offset = (bralink == NULL)? 0 : (int)(previous - bralink);
5132 bralink = previous;
5133 PUTINC(previous, 0, offset);
5134 }
5135
5136 repeat_max--;
5137 }
5138
5139 /* If the minimum is greater than zero, replicate the group as many
5140 times as necessary, and adjust the maximum to the number of subsequent
5141 copies that we need. If we set a first char from the group, and didn't
5142 set a required char, copy the latter from the former. If there are any
5143 forward reference subroutine calls in the group, there will be entries on
5144 the workspace list; replicate these with an appropriate increment. */
5145
5146 else
5147 {
5148 if (repeat_min > 1)
5149 {
5150 /* In the pre-compile phase, we don't actually do the replication. We
5151 just adjust the length as if we had. Do some paranoid checks for
5152 potential integer overflow. The INT64_OR_DOUBLE type is a 64-bit
5153 integer type when available, otherwise double. */
5154
5155 if (lengthptr != NULL)
5156 {
5157 int delta = (repeat_min - 1)*length_prevgroup;
5158 if ((INT64_OR_DOUBLE)(repeat_min - 1)*
5159 (INT64_OR_DOUBLE)length_prevgroup >
5160 (INT64_OR_DOUBLE)INT_MAX ||
5161 OFLOW_MAX - *lengthptr < delta)
5162 {
5163 *errorcodeptr = ERR20;
5164 goto FAILED;
5165 }
5166 *lengthptr += delta;
5167 }
5168
5169 /* This is compiling for real. If there is a set first byte for
5170 the group, and we have not yet set a "required byte", set it. Make
5171 sure there is enough workspace for copying forward references before
5172 doing the copy. */
5173
5174 else
5175 {
5176 if (groupsetfirstchar && reqchar < 0) reqchar = firstchar;
5177
5178 for (i = 1; i < repeat_min; i++)
5179 {
5180 pcre_uchar *hc;
5181 pcre_uchar *this_hwm = cd->hwm;
5182 memcpy(code, previous, IN_UCHARS(len));
5183
5184 while (cd->hwm > cd->start_workspace + cd->workspace_size -
5185 WORK_SIZE_SAFETY_MARGIN - (this_hwm - save_hwm))
5186 {
5187 int save_offset = save_hwm - cd->start_workspace;
5188 int this_offset = this_hwm - cd->start_workspace;
5189 *errorcodeptr = expand_workspace(cd);
5190 if (*errorcodeptr != 0) goto FAILED;
5191 save_hwm = (pcre_uchar *)cd->start_workspace + save_offset;
5192 this_hwm = (pcre_uchar *)cd->start_workspace + this_offset;
5193 }
5194
5195 for (hc = save_hwm; hc < this_hwm; hc += LINK_SIZE)
5196 {
5197 PUT(cd->hwm, 0, GET(hc, 0) + len);
5198 cd->hwm += LINK_SIZE;
5199 }
5200 save_hwm = this_hwm;
5201 code += len;
5202 }
5203 }
5204 }
5205
5206 if (repeat_max > 0) repeat_max -= repeat_min;
5207 }
5208
5209 /* This code is common to both the zero and non-zero minimum cases. If
5210 the maximum is limited, it replicates the group in a nested fashion,
5211 remembering the bracket starts on a stack. In the case of a zero minimum,
5212 the first one was set up above. In all cases the repeat_max now specifies
5213 the number of additional copies needed. Again, we must remember to
5214 replicate entries on the forward reference list. */
5215
5216 if (repeat_max >= 0)
5217 {
5218 /* In the pre-compile phase, we don't actually do the replication. We
5219 just adjust the length as if we had. For each repetition we must add 1
5220 to the length for BRAZERO and for all but the last repetition we must
5221 add 2 + 2*LINKSIZE to allow for the nesting that occurs. Do some
5222 paranoid checks to avoid integer overflow. The INT64_OR_DOUBLE type is
5223 a 64-bit integer type when available, otherwise double. */
5224
5225 if (lengthptr != NULL && repeat_max > 0)
5226 {
5227 int delta = repeat_max * (length_prevgroup + 1 + 2 + 2*LINK_SIZE) -
5228 2 - 2*LINK_SIZE; /* Last one doesn't nest */
5229 if ((INT64_OR_DOUBLE)repeat_max *
5230 (INT64_OR_DOUBLE)(length_prevgroup + 1 + 2 + 2*LINK_SIZE)
5231 > (INT64_OR_DOUBLE)INT_MAX ||
5232 OFLOW_MAX - *lengthptr < delta)
5233 {
5234 *errorcodeptr = ERR20;
5235 goto FAILED;
5236 }
5237 *lengthptr += delta;
5238 }
5239
5240 /* This is compiling for real */
5241
5242 else for (i = repeat_max - 1; i >= 0; i--)
5243 {
5244 pcre_uchar *hc;
5245 pcre_uchar *this_hwm = cd->hwm;
5246
5247 *code++ = OP_BRAZERO + repeat_type;
5248
5249 /* All but the final copy start a new nesting, maintaining the
5250 chain of brackets outstanding. */
5251
5252 if (i != 0)
5253 {
5254 int offset;
5255 *code++ = OP_BRA;
5256 offset = (bralink == NULL)? 0 : (int)(code - bralink);
5257 bralink = code;
5258 PUTINC(code, 0, offset);
5259 }
5260
5261 memcpy(code, previous, IN_UCHARS(len));
5262
5263 /* Ensure there is enough workspace for forward references before
5264 copying them. */
5265
5266 while (cd->hwm > cd->start_workspace + cd->workspace_size -
5267 WORK_SIZE_SAFETY_MARGIN - (this_hwm - save_hwm))
5268 {
5269 int save_offset = save_hwm - cd->start_workspace;
5270 int this_offset = this_hwm - cd->start_workspace;
5271 *errorcodeptr = expand_workspace(cd);
5272 if (*errorcodeptr != 0) goto FAILED;
5273 save_hwm = (pcre_uchar *)cd->start_workspace + save_offset;
5274 this_hwm = (pcre_uchar *)cd->start_workspace + this_offset;
5275 }
5276
5277 for (hc = save_hwm; hc < this_hwm; hc += LINK_SIZE)
5278 {
5279 PUT(cd->hwm, 0, GET(hc, 0) + len + ((i != 0)? 2+LINK_SIZE : 1));
5280 cd->hwm += LINK_SIZE;
5281 }
5282 save_hwm = this_hwm;
5283 code += len;
5284 }
5285
5286 /* Now chain through the pending brackets, and fill in their length
5287 fields (which are holding the chain links pro tem). */
5288
5289 while (bralink != NULL)
5290 {
5291 int oldlinkoffset;
5292 int offset = (int)(code - bralink + 1);
5293 pcre_uchar *bra = code - offset;
5294 oldlinkoffset = GET(bra, 1);
5295 bralink = (oldlinkoffset == 0)? NULL : bralink - oldlinkoffset;
5296 *code++ = OP_KET;
5297 PUTINC(code, 0, offset);
5298 PUT(bra, 1, offset);
5299 }
5300 }
5301
5302 /* If the maximum is unlimited, set a repeater in the final copy. For
5303 ONCE brackets, that's all we need to do. However, possessively repeated
5304 ONCE brackets can be converted into non-capturing brackets, as the
5305 behaviour of (?:xx)++ is the same as (?>xx)++ and this saves having to
5306 deal with possessive ONCEs specially.
5307
5308 Otherwise, when we are doing the actual compile phase, check to see
5309 whether this group is one that could match an empty string. If so,
5310 convert the initial operator to the S form (e.g. OP_BRA -> OP_SBRA) so
5311 that runtime checking can be done. [This check is also applied to ONCE
5312 groups at runtime, but in a different way.]
5313
5314 Then, if the quantifier was possessive and the bracket is not a
5315 conditional, we convert the BRA code to the POS form, and the KET code to
5316 KETRPOS. (It turns out to be convenient at runtime to detect this kind of
5317 subpattern at both the start and at the end.) The use of special opcodes
5318 makes it possible to reduce greatly the stack usage in pcre_exec(). If
5319 the group is preceded by OP_BRAZERO, convert this to OP_BRAPOSZERO.
5320
5321 Then, if the minimum number of matches is 1 or 0, cancel the possessive
5322 flag so that the default action below, of wrapping everything inside
5323 atomic brackets, does not happen. When the minimum is greater than 1,
5324 there will be earlier copies of the group, and so we still have to wrap
5325 the whole thing. */
5326
5327 else
5328 {
5329 pcre_uchar *ketcode = code - 1 - LINK_SIZE;
5330 pcre_uchar *bracode = ketcode - GET(ketcode, 1);
5331
5332 /* Convert possessive ONCE brackets to non-capturing */
5333
5334 if ((*bracode == OP_ONCE || *bracode == OP_ONCE_NC) &&
5335 possessive_quantifier) *bracode = OP_BRA;
5336
5337 /* For non-possessive ONCE brackets, all we need to do is to
5338 set the KET. */
5339
5340 if (*bracode == OP_ONCE || *bracode == OP_ONCE_NC)
5341 *ketcode = OP_KETRMAX + repeat_type;
5342
5343 /* Handle non-ONCE brackets and possessive ONCEs (which have been
5344 converted to non-capturing above). */
5345
5346 else
5347 {
5348 /* In the compile phase, check for empty string matching. */
5349
5350 if (lengthptr == NULL)
5351 {
5352 pcre_uchar *scode = bracode;
5353 do
5354 {
5355 if (could_be_empty_branch(scode, ketcode, utf, cd))
5356 {
5357 *bracode += OP_SBRA - OP_BRA;
5358 break;
5359 }
5360 scode += GET(scode, 1);
5361 }
5362 while (*scode == OP_ALT);
5363 }
5364
5365 /* Handle possessive quantifiers. */
5366
5367 if (possessive_quantifier)
5368 {
5369 /* For COND brackets, we wrap the whole thing in a possessively
5370 repeated non-capturing bracket, because we have not invented POS
5371 versions of the COND opcodes. Because we are moving code along, we
5372 must ensure that any pending recursive references are updated. */
5373
5374 if (*bracode == OP_COND || *bracode == OP_SCOND)
5375 {
5376 int nlen = (int)(code - bracode);
5377 *code = OP_END;
5378 adjust_recurse(bracode, 1 + LINK_SIZE, utf, cd, save_hwm);
5379 memmove(bracode + 1 + LINK_SIZE, bracode, IN_UCHARS(nlen));
5380 code += 1 + LINK_SIZE;
5381 nlen += 1 + LINK_SIZE;
5382 *bracode = OP_BRAPOS;
5383 *code++ = OP_KETRPOS;
5384 PUTINC(code, 0, nlen);
5385 PUT(bracode, 1, nlen);
5386 }
5387
5388 /* For non-COND brackets, we modify the BRA code and use KETRPOS. */
5389
5390 else
5391 {
5392 *bracode += 1; /* Switch to xxxPOS opcodes */
5393 *ketcode = OP_KETRPOS;
5394 }
5395
5396 /* If the minimum is zero, mark it as possessive, then unset the
5397 possessive flag when the minimum is 0 or 1. */
5398
5399 if (brazeroptr != NULL) *brazeroptr = OP_BRAPOSZERO;
5400 if (repeat_min < 2) possessive_quantifier = FALSE;
5401 }
5402
5403 /* Non-possessive quantifier */
5404
5405 else *ketcode = OP_KETRMAX + repeat_type;
5406 }
5407 }
5408 }
5409
5410 /* If previous is OP_FAIL, it was generated by an empty class [] in
5411 JavaScript mode. The other ways in which OP_FAIL can be generated, that is
5412 by (*FAIL) or (?!) set previous to NULL, which gives a "nothing to repeat"
5413 error above. We can just ignore the repeat in JS case. */
5414
5415 else if (*previous == OP_FAIL) goto END_REPEAT;
5416
5417 /* Else there's some kind of shambles */
5418
5419 else
5420 {
5421 *errorcodeptr = ERR11;
5422 goto FAILED;
5423 }
5424
5425 /* If the character following a repeat is '+', or if certain optimization
5426 tests above succeeded, possessive_quantifier is TRUE. For some opcodes,
5427 there are special alternative opcodes for this case. For anything else, we
5428 wrap the entire repeated item inside OP_ONCE brackets. Logically, the '+'
5429 notation is just syntactic sugar, taken from Sun's Java package, but the
5430 special opcodes can optimize it.
5431
5432 Some (but not all) possessively repeated subpatterns have already been
5433 completely handled in the code just above. For them, possessive_quantifier
5434 is always FALSE at this stage.
5435
5436 Note that the repeated item starts at tempcode, not at previous, which
5437 might be the first part of a string whose (former) last char we repeated.
5438
5439 Possessifying an 'exact' quantifier has no effect, so we can ignore it. But
5440 an 'upto' may follow. We skip over an 'exact' item, and then test the
5441 length of what remains before proceeding. */
5442
5443 if (possessive_quantifier)
5444 {
5445 int len;
5446
5447 if (*tempcode == OP_TYPEEXACT)
5448 tempcode += PRIV(OP_lengths)[*tempcode] +
5449 ((tempcode[1 + IMM2_SIZE] == OP_PROP
5450 || tempcode[1 + IMM2_SIZE] == OP_NOTPROP)? 2 : 0);
5451
5452 else if (*tempcode == OP_EXACT || *tempcode == OP_NOTEXACT)
5453 {
5454 tempcode += PRIV(OP_lengths)[*tempcode];
5455 #ifdef SUPPORT_UTF
5456 if (utf && HAS_EXTRALEN(tempcode[-1]))
5457 tempcode += GET_EXTRALEN(tempcode[-1]);
5458 #endif
5459 }
5460
5461 len = (int)(code - tempcode);
5462 if (len > 0) switch (*tempcode)
5463 {
5464 case OP_STAR: *tempcode = OP_POSSTAR; break;
5465 case OP_PLUS: *tempcode = OP_POSPLUS; break;
5466 case OP_QUERY: *tempcode = OP_POSQUERY; break;
5467 case OP_UPTO: *tempcode = OP_POSUPTO; break;
5468
5469 case OP_STARI: *tempcode = OP_POSSTARI; break;
5470 case OP_PLUSI: *tempcode = OP_POSPLUSI; break;
5471 case OP_QUERYI: *tempcode = OP_POSQUERYI; break;
5472 case OP_UPTOI: *tempcode = OP_POSUPTOI; break;
5473
5474 case OP_NOTSTAR: *tempcode = OP_NOTPOSSTAR; break;
5475 case OP_NOTPLUS: *tempcode = OP_NOTPOSPLUS; break;
5476 case OP_NOTQUERY: *tempcode = OP_NOTPOSQUERY; break;
5477 case OP_NOTUPTO: *tempcode = OP_NOTPOSUPTO; break;
5478
5479 case OP_NOTSTARI: *tempcode = OP_NOTPOSSTARI; break;
5480 case OP_NOTPLUSI: *tempcode = OP_NOTPOSPLUSI; break;
5481 case OP_NOTQUERYI: *tempcode = OP_NOTPOSQUERYI; break;
5482 case OP_NOTUPTOI: *tempcode = OP_NOTPOSUPTOI; break;
5483
5484 case OP_TYPESTAR: *tempcode = OP_TYPEPOSSTAR; break;
5485 case OP_TYPEPLUS: *tempcode = OP_TYPEPOSPLUS; break;
5486 case OP_TYPEQUERY: *tempcode = OP_TYPEPOSQUERY; break;
5487 case OP_TYPEUPTO: *tempcode = OP_TYPEPOSUPTO; break;
5488
5489 /* Because we are moving code along, we must ensure that any
5490 pending recursive references are updated. */
5491
5492 default:
5493 *code = OP_END;
5494 adjust_recurse(tempcode, 1 + LINK_SIZE, utf, cd, save_hwm);
5495 memmove(tempcode + 1 + LINK_SIZE, tempcode, IN_UCHARS(len));
5496 code += 1 + LINK_SIZE;
5497 len += 1 + LINK_SIZE;
5498 tempcode[0] = OP_ONCE;
5499 *code++ = OP_KET;
5500 PUTINC(code, 0, len);
5501 PUT(tempcode, 1, len);
5502 break;
5503 }
5504 }
5505
5506 /* In all case we no longer have a previous item. We also set the
5507 "follows varying string" flag for subsequently encountered reqchars if
5508 it isn't already set and we have just passed a varying length item. */
5509
5510 END_REPEAT:
5511 previous = NULL;
5512 cd->req_varyopt |= reqvary;
5513 break;
5514
5515
5516 /* ===================================================================*/
5517 /* Start of nested parenthesized sub-expression, or comment or lookahead or
5518 lookbehind or option setting or condition or all the other extended
5519 parenthesis forms. */
5520
5521 case CHAR_LEFT_PARENTHESIS:
5522 newoptions = options;
5523 skipbytes = 0;
5524 bravalue = OP_CBRA;
5525 save_hwm = cd->hwm;
5526 reset_bracount = FALSE;
5527
5528 /* First deal with various "verbs" that can be introduced by '*'. */
5529
5530 ptr++;
5531 if (ptr[0] == CHAR_ASTERISK && (ptr[1] == ':'
5532 || (MAX_255(ptr[1]) && ((cd->ctypes[ptr[1]] & ctype_letter) != 0))))
5533 {
5534 int i, namelen;
5535 int arglen = 0;
5536 const char *vn = verbnames;
5537 const pcre_uchar *name = ptr + 1;
5538 const pcre_uchar *arg = NULL;
5539 previous = NULL;
5540 ptr++;
5541 while (MAX_255(*ptr) && (cd->ctypes[*ptr] & ctype_letter) != 0) ptr++;
5542 namelen = (int)(ptr - name);
5543
5544 /* It appears that Perl allows any characters whatsoever, other than
5545 a closing parenthesis, to appear in arguments, so we no longer insist on
5546 letters, digits, and underscores. */
5547
5548 if (*ptr == CHAR_COLON)
5549 {
5550 arg = ++ptr;
5551 while (*ptr != 0 && *ptr != CHAR_RIGHT_PARENTHESIS) ptr++;
5552 arglen = (int)(ptr - arg);
5553 if (arglen > (int)MAX_MARK)
5554 {
5555 *errorcodeptr = ERR75;
5556 goto FAILED;
5557 }
5558 }
5559
5560 if (*ptr != CHAR_RIGHT_PARENTHESIS)
5561 {
5562 *errorcodeptr = ERR60;
5563 goto FAILED;
5564 }
5565
5566 /* Scan the table of verb names */
5567
5568 for (i = 0; i < verbcount; i++)
5569 {
5570 if (namelen == verbs[i].len &&
5571 STRNCMP_UC_C8(name, vn, namelen) == 0)
5572 {
5573 int setverb;
5574
5575 /* Check for open captures before ACCEPT and convert it to
5576 ASSERT_ACCEPT if in an assertion. */
5577
5578 if (verbs[i].op == OP_ACCEPT)
5579 {
5580 open_capitem *oc;
5581 if (arglen != 0)
5582 {
5583 *errorcodeptr = ERR59;
5584 goto FAILED;
5585 }
5586 cd->had_accept = TRUE;
5587 for (oc = cd->open_caps; oc != NULL; oc = oc->next)
5588 {
5589 *code++ = OP_CLOSE;
5590 PUT2INC(code, 0, oc->number);
5591 }
5592 setverb = *code++ =
5593 (cd->assert_depth > 0)? OP_ASSERT_ACCEPT : OP_ACCEPT;
5594
5595 /* Do not set firstchar after *ACCEPT */
5596 if (firstchar == REQ_UNSET) firstchar = REQ_NONE;
5597 }
5598
5599 /* Handle other cases with/without an argument */
5600
5601 else if (arglen == 0)
5602 {
5603 if (verbs[i].op < 0) /* Argument is mandatory */
5604 {
5605 *errorcodeptr = ERR66;
5606 goto FAILED;
5607 }
5608 setverb = *code++ = verbs[i].op;
5609 }
5610
5611 else
5612 {
5613 if (verbs[i].op_arg < 0) /* Argument is forbidden */
5614 {
5615 *errorcodeptr = ERR59;
5616 goto FAILED;
5617 }
5618 setverb = *code++ = verbs[i].op_arg;
5619 *code++ = arglen;
5620 memcpy(code, arg, IN_UCHARS(arglen));
5621 code += arglen;
5622 *code++ = 0;
5623 }
5624
5625 switch (setverb)
5626 {
5627 case OP_THEN:
5628 case OP_THEN_ARG:
5629 cd->external_flags |= PCRE_HASTHEN;
5630 break;
5631
5632 case OP_PRUNE:
5633 case OP_PRUNE_ARG:
5634 case OP_SKIP:
5635 case OP_SKIP_ARG:
5636 cd->had_pruneorskip = TRUE;
5637 break;
5638 }
5639
5640 break; /* Found verb, exit loop */
5641 }
5642
5643 vn += verbs[i].len + 1;
5644 }
5645
5646 if (i < verbcount) continue; /* Successfully handled a verb */
5647 *errorcodeptr = ERR60; /* Verb not recognized */
5648 goto FAILED;
5649 }
5650
5651 /* Deal with the extended parentheses; all are introduced by '?', and the
5652 appearance of any of them means that this is not a capturing group. */
5653
5654 else if (*ptr == CHAR_QUESTION_MARK)
5655 {
5656 int i, set, unset, namelen;
5657 int *optset;
5658 const pcre_uchar *name;
5659 pcre_uchar *slot;
5660
5661 switch (*(++ptr))
5662 {
5663 case CHAR_NUMBER_SIGN: /* Comment; skip to ket */
5664 ptr++;
5665 while (*ptr != 0 && *ptr != CHAR_RIGHT_PARENTHESIS) ptr++;
5666 if (*ptr == 0)
5667 {
5668 *errorcodeptr = ERR18;
5669 goto FAILED;
5670 }
5671 continue;
5672
5673
5674 /* ------------------------------------------------------------ */
5675 case CHAR_VERTICAL_LINE: /* Reset capture count for each branch */
5676 reset_bracount = TRUE;
5677 /* Fall through */
5678
5679 /* ------------------------------------------------------------ */
5680 case CHAR_COLON: /* Non-capturing bracket */
5681 bravalue = OP_BRA;
5682 ptr++;
5683 break;
5684
5685
5686 /* ------------------------------------------------------------ */
5687 case CHAR_LEFT_PARENTHESIS:
5688 bravalue = OP_COND; /* Conditional group */
5689
5690 /* A condition can be an assertion, a number (referring to a numbered
5691 group), a name (referring to a named group), or 'R', referring to
5692 recursion. R<digits> and R&name are also permitted for recursion tests.
5693
5694 There are several syntaxes for testing a named group: (?(name)) is used
5695 by Python; Perl 5.10 onwards uses (?(<name>) or (?('name')).
5696
5697 There are two unfortunate ambiguities, caused by history. (a) 'R' can
5698 be the recursive thing or the name 'R' (and similarly for 'R' followed
5699 by digits), and (b) a number could be a name that consists of digits.
5700 In both cases, we look for a name first; if not found, we try the other
5701 cases. */
5702
5703 /* For conditions that are assertions, check the syntax, and then exit
5704 the switch. This will take control down to where bracketed groups,
5705 including assertions, are processed. */
5706
5707 if (ptr[1] == CHAR_QUESTION_MARK && (ptr[2] == CHAR_EQUALS_SIGN ||
5708 ptr[2] == CHAR_EXCLAMATION_MARK || ptr[2] == CHAR_LESS_THAN_SIGN))
5709 break;
5710
5711 /* Most other conditions use OP_CREF (a couple change to OP_RREF
5712 below), and all need to skip 1+IMM2_SIZE bytes at the start of the group. */
5713
5714 code[1+LINK_SIZE] = OP_CREF;
5715 skipbytes = 1+IMM2_SIZE;
5716 refsign = -1;
5717
5718 /* Check for a test for recursion in a named group. */
5719
5720 if (ptr[1] == CHAR_R && ptr[2] == CHAR_AMPERSAND)
5721 {
5722 terminator = -1;
5723 ptr += 2;
5724 code[1+LINK_SIZE] = OP_RREF; /* Change the type of test */
5725 }
5726
5727 /* Check for a test for a named group's having been set, using the Perl
5728 syntax (?(<name>) or (?('name') */
5729
5730 else if (ptr[1] == CHAR_LESS_THAN_SIGN)
5731 {
5732 terminator = CHAR_GREATER_THAN_SIGN;
5733 ptr++;
5734 }
5735 else if (ptr[1] == CHAR_APOSTROPHE)
5736 {
5737 terminator = CHAR_APOSTROPHE;
5738 ptr++;
5739 }
5740 else
5741 {
5742 terminator = 0;
5743 if (ptr[1] == CHAR_MINUS || ptr[1] == CHAR_PLUS) refsign = *(++ptr);
5744 }
5745
5746 /* We now expect to read a name; any thing else is an error */
5747
5748 if (!MAX_255(ptr[1]) || (cd->ctypes[ptr[1]] & ctype_word) == 0)
5749 {
5750 ptr += 1; /* To get the right offset */
5751 *errorcodeptr = ERR28;
5752 goto FAILED;
5753 }
5754
5755 /* Read the name, but also get it as a number if it's all digits */
5756
5757 recno = 0;
5758 name = ++ptr;
5759 while (MAX_255(*ptr) && (cd->ctypes[*ptr] & ctype_word) != 0)
5760 {
5761 if (recno >= 0)
5762 recno = (IS_DIGIT(*ptr))? recno * 10 + *ptr - CHAR_0 : -1;
5763 ptr++;
5764 }
5765 namelen = (int)(ptr - name);
5766
5767 if ((terminator > 0 && *ptr++ != terminator) ||
5768 *ptr++ != CHAR_RIGHT_PARENTHESIS)
5769 {
5770 ptr--; /* Error offset */
5771 *errorcodeptr = ERR26;
5772 goto FAILED;
5773 }
5774
5775 /* Do no further checking in the pre-compile phase. */
5776
5777 if (lengthptr != NULL) break;
5778
5779 /* In the real compile we do the work of looking for the actual
5780 reference. If the string started with "+" or "-" we require the rest to
5781 be digits, in which case recno will be set. */
5782
5783 if (refsign > 0)
5784 {
5785 if (recno <= 0)
5786 {
5787 *errorcodeptr = ERR58;
5788 goto FAILED;
5789 }
5790 recno = (refsign == CHAR_MINUS)?
5791 cd->bracount - recno + 1 : recno +cd->bracount;
5792 if (recno <= 0 || recno > cd->final_bracount)
5793 {
5794 *errorcodeptr = ERR15;
5795 goto FAILED;
5796 }
5797 PUT2(code, 2+LINK_SIZE, recno);
5798 break;
5799 }
5800
5801 /* Otherwise (did not start with "+" or "-"), start by looking for the
5802 name. If we find a name, add one to the opcode to change OP_CREF or
5803 OP_RREF into OP_NCREF or OP_NRREF. These behave exactly the same,
5804 except they record that the reference was originally to a name. The
5805 information is used to check duplicate names. */
5806
5807 slot = cd->name_table;
5808 for (i = 0; i < cd->names_found; i++)
5809 {
5810 if (STRNCMP_UC_UC(name, slot+IMM2_SIZE, namelen) == 0) break;
5811 slot += cd->name_entry_size;
5812 }
5813
5814 /* Found a previous named subpattern */
5815
5816 if (i < cd->names_found)
5817 {
5818 recno = GET2(slot, 0);
5819 PUT2(code, 2+LINK_SIZE, recno);
5820 code[1+LINK_SIZE]++;
5821 }
5822
5823 /* Search the pattern for a forward reference */
5824
5825 else if ((i = find_parens(cd, name, namelen,
5826 (options & PCRE_EXTENDED) != 0, utf)) > 0)
5827 {
5828 PUT2(code, 2+LINK_SIZE, i);
5829 code[1+LINK_SIZE]++;
5830 }
5831
5832 /* If terminator == 0 it means that the name followed directly after
5833 the opening parenthesis [e.g. (?(abc)...] and in this case there are
5834 some further alternatives to try. For the cases where terminator != 0
5835 [things like (?(<name>... or (?('name')... or (?(R&name)... ] we have
5836 now checked all the possibilities, so give an error. */
5837
5838 else if (terminator != 0)
5839 {
5840 *errorcodeptr = ERR15;
5841 goto FAILED;
5842 }
5843
5844 /* Check for (?(R) for recursion. Allow digits after R to specify a
5845 specific group number. */
5846
5847 else if (*name == CHAR_R)
5848 {
5849 recno = 0;
5850 for (i = 1; i < namelen; i++)
5851 {
5852 if (!IS_DIGIT(name[i]))
5853 {
5854 *errorcodeptr = ERR15;
5855 goto FAILED;
5856 }
5857 recno = recno * 10 + name[i] - CHAR_0;
5858 }
5859 if (recno == 0) recno = RREF_ANY;
5860 code[1+LINK_SIZE] = OP_RREF; /* Change test type */
5861 PUT2(code, 2+LINK_SIZE, recno);
5862 }
5863
5864 /* Similarly, check for the (?(DEFINE) "condition", which is always
5865 false. */
5866
5867 else if (namelen == 6 && STRNCMP_UC_C8(name, STRING_DEFINE, 6) == 0)
5868 {
5869 code[1+LINK_SIZE] = OP_DEF;
5870 skipbytes = 1;
5871 }
5872
5873 /* Check for the "name" actually being a subpattern number. We are
5874 in the second pass here, so final_bracount is set. */
5875
5876 else if (recno > 0 && recno <= cd->final_bracount)
5877 {
5878 PUT2(code, 2+LINK_SIZE, recno);
5879 }
5880
5881 /* Either an unidentified subpattern, or a reference to (?(0) */
5882
5883 else
5884 {
5885 *errorcodeptr = (recno == 0)? ERR35: ERR15;
5886 goto FAILED;
5887 }
5888 break;
5889
5890
5891 /* ------------------------------------------------------------ */
5892 case CHAR_EQUALS_SIGN: /* Positive lookahead */
5893 bravalue = OP_ASSERT;
5894 cd->assert_depth += 1;
5895 ptr++;
5896 break;
5897
5898
5899 /* ------------------------------------------------------------ */
5900 case CHAR_EXCLAMATION_MARK: /* Negative lookahead */
5901 ptr++;
5902 if (*ptr == CHAR_RIGHT_PARENTHESIS) /* Optimize (?!) */
5903 {
5904 *code++ = OP_FAIL;
5905 previous = NULL;
5906 continue;
5907 }
5908 bravalue = OP_ASSERT_NOT;
5909 cd->assert_depth += 1;
5910 break;
5911
5912
5913 /* ------------------------------------------------------------ */
5914 case CHAR_LESS_THAN_SIGN: /* Lookbehind or named define */
5915 switch (ptr[1])
5916 {
5917 case CHAR_EQUALS_SIGN: /* Positive lookbehind */
5918 bravalue = OP_ASSERTBACK;
5919 cd->assert_depth += 1;
5920 ptr += 2;
5921 break;
5922
5923 case CHAR_EXCLAMATION_MARK: /* Negative lookbehind */
5924 bravalue = OP_ASSERTBACK_NOT;
5925 cd->assert_depth += 1;
5926 ptr += 2;
5927 break;
5928
5929 default: /* Could be name define, else bad */
5930 if (MAX_255(ptr[1]) && (cd->ctypes[ptr[1]] & ctype_word) != 0)
5931 goto DEFINE_NAME;
5932 ptr++; /* Correct offset for error */
5933 *errorcodeptr = ERR24;
5934 goto FAILED;
5935 }
5936 break;
5937
5938
5939 /* ------------------------------------------------------------ */
5940 case CHAR_GREATER_THAN_SIGN: /* One-time brackets */
5941 bravalue = OP_ONCE;
5942 ptr++;
5943 break;
5944
5945
5946 /* ------------------------------------------------------------ */
5947 case CHAR_C: /* Callout - may be followed by digits; */
5948 previous_callout = code; /* Save for later completion */
5949 after_manual_callout = 1; /* Skip one item before completing */
5950 *code++ = OP_CALLOUT;
5951 {
5952 int n = 0;
5953 ptr++;
5954 while(IS_DIGIT(*ptr))
5955 n = n * 10 + *ptr++ - CHAR_0;
5956 if (*ptr != CHAR_RIGHT_PARENTHESIS)
5957 {
5958 *errorcodeptr = ERR39;
5959 goto FAILED;
5960 }
5961 if (n > 255)
5962 {
5963 *errorcodeptr = ERR38;
5964 goto FAILED;
5965 }
5966 *code++ = n;
5967 PUT(code, 0, (int)(ptr - cd->start_pattern + 1)); /* Pattern offset */
5968 PUT(code, LINK_SIZE, 0); /* Default length */
5969 code += 2 * LINK_SIZE;
5970 }
5971 previous = NULL;
5972 continue;
5973
5974
5975 /* ------------------------------------------------------------ */
5976 case CHAR_P: /* Python-style named subpattern handling */
5977 if (*(++ptr) == CHAR_EQUALS_SIGN ||
5978 *ptr == CHAR_GREATER_THAN_SIGN) /* Reference or recursion */
5979 {
5980 is_recurse = *ptr == CHAR_GREATER_THAN_SIGN;
5981 terminator = CHAR_RIGHT_PARENTHESIS;
5982 goto NAMED_REF_OR_RECURSE;
5983 }
5984 else if (*ptr != CHAR_LESS_THAN_SIGN) /* Test for Python-style defn */
5985 {
5986 *errorcodeptr = ERR41;
5987 goto FAILED;
5988 }
5989 /* Fall through to handle (?P< as (?< is handled */
5990
5991
5992 /* ------------------------------------------------------------ */
5993 DEFINE_NAME: /* Come here from (?< handling */
5994 case CHAR_APOSTROPHE:
5995 {
5996 terminator = (*ptr == CHAR_LESS_THAN_SIGN)?
5997 CHAR_GREATER_THAN_SIGN : CHAR_APOSTROPHE;
5998 name = ++ptr;
5999
6000 while (MAX_255(*ptr) && (cd->ctypes[*ptr] & ctype_word) != 0) ptr++;
6001 namelen = (int)(ptr - name);
6002
6003 /* In the pre-compile phase, just do a syntax check. */
6004
6005 if (lengthptr != NULL)
6006 {
6007 if (*ptr != terminator)
6008 {
6009 *errorcodeptr = ERR42;
6010 goto FAILED;
6011 }
6012 if (cd->names_found >= MAX_NAME_COUNT)
6013 {
6014 *errorcodeptr = ERR49;
6015 goto FAILED;
6016 }
6017 if (namelen + IMM2_SIZE + 1 > cd->name_entry_size)
6018 {
6019 cd->name_entry_size = namelen + IMM2_SIZE + 1;
6020 if (namelen > MAX_NAME_SIZE)
6021 {
6022 *errorcodeptr = ERR48;
6023 goto FAILED;
6024 }
6025 }
6026 }
6027
6028 /* In the real compile, create the entry in the table, maintaining
6029 alphabetical order. Duplicate names for different numbers are
6030 permitted only if PCRE_DUPNAMES is set. Duplicate names for the same
6031 number are always OK. (An existing number can be re-used if (?|
6032 appears in the pattern.) In either event, a duplicate name results in
6033 a duplicate entry in the table, even if the number is the same. This
6034 is because the number of names, and hence the table size, is computed
6035 in the pre-compile, and it affects various numbers and pointers which
6036 would all have to be modified, and the compiled code moved down, if
6037 duplicates with the same number were omitted from the table. This
6038 doesn't seem worth the hassle. However, *different* names for the
6039 same number are not permitted. */
6040
6041 else
6042 {
6043 BOOL dupname = FALSE;
6044 slot = cd->name_table;
6045
6046 for (i = 0; i < cd->names_found; i++)
6047 {
6048 int crc = memcmp(name, slot+IMM2_SIZE, IN_UCHARS(namelen));
6049 if (crc == 0)
6050 {
6051 if (slot[IMM2_SIZE+namelen] == 0)
6052 {
6053 if (GET2(slot, 0) != cd->bracount + 1 &&
6054 (options & PCRE_DUPNAMES) == 0)
6055 {
6056 *errorcodeptr = ERR43;
6057 goto FAILED;
6058 }
6059 else dupname = TRUE;
6060 }
6061 else crc = -1; /* Current name is a substring */
6062 }
6063
6064 /* Make space in the table and break the loop for an earlier
6065 name. For a duplicate or later name, carry on. We do this for
6066 duplicates so that in the simple case (when ?(| is not used) they
6067 are in order of their numbers. */
6068
6069 if (crc < 0)
6070 {
6071 memmove(slot + cd->name_entry_size, slot,
6072 IN_UCHARS((cd->names_found - i) * cd->name_entry_size));
6073 break;
6074 }
6075
6076 /* Continue the loop for a later or duplicate name */
6077
6078 slot += cd->name_entry_size;
6079 }
6080
6081 /* For non-duplicate names, check for a duplicate number before
6082 adding the new name. */
6083
6084 if (!dupname)
6085 {
6086 pcre_uchar *cslot = cd->name_table;
6087 for (i = 0; i < cd->names_found; i++)
6088 {
6089 if (cslot != slot)
6090 {
6091 if (GET2(cslot, 0) == cd->bracount + 1)
6092 {
6093 *errorcodeptr = ERR65;
6094 goto FAILED;
6095 }
6096 }
6097 else i--;
6098 cslot += cd->name_entry_size;
6099 }
6100 }
6101
6102 PUT2(slot, 0, cd->bracount + 1);
6103 memcpy(slot + IMM2_SIZE, name, IN_UCHARS(namelen));
6104 slot[IMM2_SIZE + namelen] = 0;
6105 }
6106 }
6107
6108 /* In both pre-compile and compile, count the number of names we've
6109 encountered. */
6110
6111 cd->names_found++;
6112 ptr++; /* Move past > or ' */
6113 goto NUMBERED_GROUP;
6114
6115
6116 /* ------------------------------------------------------------ */
6117 case CHAR_AMPERSAND: /* Perl recursion/subroutine syntax */
6118 terminator = CHAR_RIGHT_PARENTHESIS;
6119 is_recurse = TRUE;
6120 /* Fall through */
6121
6122 /* We come here from the Python syntax above that handles both
6123 references (?P=name) and recursion (?P>name), as well as falling
6124 through from the Perl recursion syntax (?&name). We also come here from
6125 the Perl \k<name> or \k'name' back reference syntax and the \k{name}
6126 .NET syntax, and the Oniguruma \g<...> and \g'...' subroutine syntax. */
6127
6128 NAMED_REF_OR_RECURSE:
6129 name = ++ptr;
6130 while (MAX_255(*ptr) && (cd->ctypes[*ptr] & ctype_word) != 0) ptr++;
6131 namelen = (int)(ptr - name);
6132
6133 /* In the pre-compile phase, do a syntax check. We used to just set
6134 a dummy reference number, because it was not used in the first pass.
6135 However, with the change of recursive back references to be atomic,
6136 we have to look for the number so that this state can be identified, as
6137 otherwise the incorrect length is computed. If it's not a backwards
6138 reference, the dummy number will do. */
6139
6140 if (lengthptr != NULL)
6141 {
6142 const pcre_uchar *temp;
6143
6144 if (namelen == 0)
6145 {
6146 *errorcodeptr = ERR62;
6147 goto FAILED;
6148 }
6149 if (*ptr != terminator)
6150 {
6151 *errorcodeptr = ERR42;
6152 goto FAILED;
6153 }
6154 if (namelen > MAX_NAME_SIZE)
6155 {
6156 *errorcodeptr = ERR48;
6157 goto FAILED;
6158 }
6159
6160 /* The name table does not exist in the first pass, so we cannot
6161 do a simple search as in the code below. Instead, we have to scan the
6162 pattern to find the number. It is important that we scan it only as
6163 far as we have got because the syntax of named subpatterns has not
6164 been checked for the rest of the pattern, and find_parens() assumes
6165 correct syntax. In any case, it's a waste of resources to scan
6166 further. We stop the scan at the current point by temporarily
6167 adjusting the value of cd->endpattern. */
6168
6169 temp = cd->end_pattern;
6170 cd->end_pattern = ptr;
6171 recno = find_parens(cd, name, namelen,
6172 (options & PCRE_EXTENDED) != 0, utf);
6173 cd->end_pattern = temp;
6174 if (recno < 0) recno = 0; /* Forward ref; set dummy number */
6175 }
6176
6177 /* In the real compile, seek the name in the table. We check the name
6178 first, and then check that we have reached the end of the name in the
6179 table. That way, if the name that is longer than any in the table,
6180 the comparison will fail without reading beyond the table entry. */
6181
6182 else
6183 {
6184 slot = cd->name_table;
6185 for (i = 0; i < cd->names_found; i++)
6186 {
6187 if (STRNCMP_UC_UC(name, slot+IMM2_SIZE, namelen) == 0 &&
6188 slot[IMM2_SIZE+namelen] == 0)
6189 break;
6190 slot += cd->name_entry_size;
6191 }
6192
6193 if (i < cd->names_found) /* Back reference */
6194 {
6195 recno = GET2(slot, 0);
6196 }
6197 else if ((recno = /* Forward back reference */
6198 find_parens(cd, name, namelen,
6199 (options & PCRE_EXTENDED) != 0, utf)) <= 0)
6200 {
6201 *errorcodeptr = ERR15;
6202 goto FAILED;
6203 }
6204 }
6205
6206 /* In both phases, we can now go to the code than handles numerical
6207 recursion or backreferences. */
6208
6209 if (is_recurse) goto HANDLE_RECURSION;
6210 else goto HANDLE_REFERENCE;
6211
6212
6213 /* ------------------------------------------------------------ */
6214 case CHAR_R: /* Recursion */
6215 ptr++; /* Same as (?0) */
6216 /* Fall through */
6217
6218
6219 /* ------------------------------------------------------------ */
6220 case CHAR_MINUS: case CHAR_PLUS: /* Recursion or subroutine */
6221 case CHAR_0: case CHAR_1: case CHAR_2: case CHAR_3: case CHAR_4:
6222 case CHAR_5: case CHAR_6: case CHAR_7: case CHAR_8: case CHAR_9:
6223 {
6224 const pcre_uchar *called;
6225 terminator = CHAR_RIGHT_PARENTHESIS;
6226
6227 /* Come here from the \g<...> and \g'...' code (Oniguruma
6228 compatibility). However, the syntax has been checked to ensure that
6229 the ... are a (signed) number, so that neither ERR63 nor ERR29 will
6230 be called on this path, nor with the jump to OTHER_CHAR_AFTER_QUERY
6231 ever be taken. */
6232
6233 HANDLE_NUMERICAL_RECURSION:
6234
6235 if ((refsign = *ptr) == CHAR_PLUS)
6236 {
6237 ptr++;
6238 if (!IS_DIGIT(*ptr))
6239 {
6240 *errorcodeptr = ERR63;
6241 goto FAILED;
6242 }
6243 }
6244 else if (refsign == CHAR_MINUS)
6245 {
6246 if (!IS_DIGIT(ptr[1]))
6247 goto OTHER_CHAR_AFTER_QUERY;
6248 ptr++;
6249 }
6250
6251 recno = 0;
6252 while(IS_DIGIT(*ptr))
6253 recno = recno * 10 + *ptr++ - CHAR_0;
6254
6255 if (*ptr != terminator)
6256 {
6257 *errorcodeptr = ERR29;
6258 goto FAILED;
6259 }
6260
6261 if (refsign == CHAR_MINUS)
6262 {
6263 if (recno == 0)
6264 {
6265 *errorcodeptr = ERR58;
6266 goto FAILED;
6267 }
6268 recno = cd->bracount - recno + 1;
6269 if (recno <= 0)
6270 {
6271 *errorcodeptr = ERR15;
6272 goto FAILED;
6273 }
6274 }
6275 else if (refsign == CHAR_PLUS)
6276 {
6277 if (recno == 0)
6278 {
6279 *errorcodeptr = ERR58;
6280 goto FAILED;
6281 }
6282 recno += cd->bracount;
6283 }
6284
6285 /* Come here from code above that handles a named recursion */
6286
6287 HANDLE_RECURSION:
6288
6289 previous = code;
6290 called = cd->start_code;
6291
6292 /* When we are actually compiling, find the bracket that is being
6293 referenced. Temporarily end the regex in case it doesn't exist before
6294 this point. If we end up with a forward reference, first check that
6295 the bracket does occur later so we can give the error (and position)
6296 now. Then remember this forward reference in the workspace so it can
6297 be filled in at the end. */
6298
6299 if (lengthptr == NULL)
6300 {
6301 *code = OP_END;
6302 if (recno != 0)
6303 called = PRIV(find_bracket)(cd->start_code, utf, recno);
6304
6305 /* Forward reference */
6306
6307 if (called == NULL)
6308 {
6309 if (find_parens(cd, NULL, recno,
6310 (options & PCRE_EXTENDED) != 0, utf) < 0)
6311 {
6312 *errorcodeptr = ERR15;
6313 goto FAILED;
6314 }
6315
6316 /* Fudge the value of "called" so that when it is inserted as an
6317 offset below, what it actually inserted is the reference number
6318 of the group. Then remember the forward reference. */
6319
6320 called = cd->start_code + recno;
6321 if (cd->hwm >= cd->start_workspace + cd->workspace_size -
6322 WORK_SIZE_SAFETY_MARGIN)
6323 {
6324 *errorcodeptr = expand_workspace(cd);
6325 if (*errorcodeptr != 0) goto FAILED;
6326 }
6327 PUTINC(cd->hwm, 0, (int)(code + 1 - cd->start_code));
6328 }
6329
6330 /* If not a forward reference, and the subpattern is still open,
6331 this is a recursive call. We check to see if this is a left
6332 recursion that could loop for ever, and diagnose that case. We
6333 must not, however, do this check if we are in a conditional
6334 subpattern because the condition might be testing for recursion in
6335 a pattern such as /(?(R)a+|(?R)b)/, which is perfectly valid.
6336 Forever loops are also detected at runtime, so those that occur in
6337 conditional subpatterns will be picked up then. */
6338
6339 else if (GET(called, 1) == 0 && cond_depth <= 0 &&
6340 could_be_empty(called, code, bcptr, utf, cd))
6341 {
6342 *errorcodeptr = ERR40;
6343 goto FAILED;
6344 }
6345 }
6346
6347 /* Insert the recursion/subroutine item. It does not have a set first
6348 character (relevant if it is repeated, because it will then be
6349 wrapped with ONCE brackets). */
6350
6351 *code = OP_RECURSE;
6352 PUT(code, 1, (int)(called - cd->start_code));
6353 code += 1 + LINK_SIZE;
6354 groupsetfirstchar = FALSE;
6355 }
6356
6357 /* Can't determine a first byte now */
6358
6359 if (firstchar == REQ_UNSET) firstchar = REQ_NONE;
6360 continue;
6361
6362
6363 /* ------------------------------------------------------------ */
6364 default: /* Other characters: check option setting */
6365 OTHER_CHAR_AFTER_QUERY:
6366 set = unset = 0;
6367 optset = &set;
6368
6369 while (*ptr != CHAR_RIGHT_PARENTHESIS && *ptr != CHAR_COLON)
6370 {
6371 switch (*ptr++)
6372 {
6373 case CHAR_MINUS: optset = &unset; break;
6374
6375 case CHAR_J: /* Record that it changed in the external options */
6376 *optset |= PCRE_DUPNAMES;
6377 cd->external_flags |= PCRE_JCHANGED;
6378 break;
6379
6380 case CHAR_i: *optset |= PCRE_CASELESS; break;
6381 case CHAR_m: *optset |= PCRE_MULTILINE; break;
6382 case CHAR_s: *optset |= PCRE_DOTALL; break;
6383 case CHAR_x: *optset |= PCRE_EXTENDED; break;
6384 case CHAR_U: *optset |= PCRE_UNGREEDY; break;
6385 case CHAR_X: *optset |= PCRE_EXTRA; break;
6386
6387 default: *errorcodeptr = ERR12;
6388 ptr--; /* Correct the offset */
6389 goto FAILED;
6390 }
6391 }
6392
6393 /* Set up the changed option bits, but don't change anything yet. */
6394
6395 newoptions = (options | set) & (~unset);
6396
6397 /* If the options ended with ')' this is not the start of a nested
6398 group with option changes, so the options change at this level. If this
6399 item is right at the start of the pattern, the options can be
6400 abstracted and made external in the pre-compile phase, and ignored in
6401 the compile phase. This can be helpful when matching -- for instance in
6402 caseless checking of required bytes.
6403
6404 If the code pointer is not (cd->start_code + 1 + LINK_SIZE), we are
6405 definitely *not* at the start of the pattern because something has been
6406 compiled. In the pre-compile phase, however, the code pointer can have
6407 that value after the start, because it gets reset as code is discarded
6408 during the pre-compile. However, this can happen only at top level - if
6409 we are within parentheses, the starting BRA will still be present. At
6410 any parenthesis level, the length value can be used to test if anything
6411 has been compiled at that level. Thus, a test for both these conditions
6412 is necessary to ensure we correctly detect the start of the pattern in
6413 both phases.
6414
6415 If we are not at the pattern start, reset the greedy defaults and the
6416 case value for firstchar and reqchar. */
6417
6418 if (*ptr == CHAR_RIGHT_PARENTHESIS)
6419 {
6420 if (code == cd->start_code + 1 + LINK_SIZE &&
6421 (lengthptr == NULL || *lengthptr == 2 + 2*LINK_SIZE))
6422 {
6423 cd->external_options = newoptions;
6424 }
6425 else
6426 {
6427 greedy_default = ((newoptions & PCRE_UNGREEDY) != 0);
6428 greedy_non_default = greedy_default ^ 1;
6429 req_caseopt = ((newoptions & PCRE_CASELESS) != 0)? REQ_CASELESS:0;
6430 }
6431
6432 /* Change options at this level, and pass them back for use
6433 in subsequent branches. */
6434
6435 *optionsptr = options = newoptions;
6436 previous = NULL; /* This item can't be repeated */
6437 continue; /* It is complete */
6438 }
6439
6440 /* If the options ended with ':' we are heading into a nested group
6441 with possible change of options. Such groups are non-capturing and are
6442 not assertions of any kind. All we need to do is skip over the ':';
6443 the newoptions value is handled below. */
6444
6445 bravalue = OP_BRA;
6446 ptr++;
6447 } /* End of switch for character following (? */
6448 } /* End of (? handling */
6449
6450 /* Opening parenthesis not followed by '*' or '?'. If PCRE_NO_AUTO_CAPTURE
6451 is set, all unadorned brackets become non-capturing and behave like (?:...)
6452 brackets. */
6453
6454 else if ((options & PCRE_NO_AUTO_CAPTURE) != 0)
6455 {
6456 bravalue = OP_BRA;
6457 }
6458
6459 /* Else we have a capturing group. */
6460
6461 else
6462 {
6463 NUMBERED_GROUP:
6464 cd->bracount += 1;
6465 PUT2(code, 1+LINK_SIZE, cd->bracount);
6466 skipbytes = IMM2_SIZE;
6467 }
6468
6469 /* Process nested bracketed regex. Assertions used not to be repeatable,
6470 but this was changed for Perl compatibility, so all kinds can now be
6471 repeated. We copy code into a non-register variable (tempcode) in order to
6472 be able to pass its address because some compilers complain otherwise. */
6473
6474 previous = code; /* For handling repetition */
6475 *code = bravalue;
6476 tempcode = code;
6477 tempreqvary = cd->req_varyopt; /* Save value before bracket */
6478 tempbracount = cd->bracount; /* Save value before bracket */
6479 length_prevgroup = 0; /* Initialize for pre-compile phase */
6480
6481 if (!compile_regex(
6482 newoptions, /* The complete new option state */
6483 &tempcode, /* Where to put code (updated) */
6484 &ptr, /* Input pointer (updated) */
6485 errorcodeptr, /* Where to put an error message */
6486 (bravalue == OP_ASSERTBACK ||
6487 bravalue == OP_ASSERTBACK_NOT), /* TRUE if back assert */
6488 reset_bracount, /* True if (?| group */
6489 skipbytes, /* Skip over bracket number */
6490 cond_depth +
6491 ((bravalue == OP_COND)?1:0), /* Depth of condition subpatterns */
6492 &subfirstchar, /* For possible first char */
6493 &subreqchar, /* For possible last char */
6494 bcptr, /* Current branch chain */
6495 cd, /* Tables block */
6496 (lengthptr == NULL)? NULL : /* Actual compile phase */
6497 &length_prevgroup /* Pre-compile phase */
6498 ))
6499 goto FAILED;
6500
6501 /* If this was an atomic group and there are no capturing groups within it,
6502 generate OP_ONCE_NC instead of OP_ONCE. */
6503
6504 if (bravalue == OP_ONCE && cd->bracount <= tempbracount)
6505 *code = OP_ONCE_NC;
6506
6507 if (bravalue >= OP_ASSERT && bravalue <= OP_ASSERTBACK_NOT)
6508 cd->assert_depth -= 1;
6509
6510 /* At the end of compiling, code is still pointing to the start of the
6511 group, while tempcode has been updated to point past the end of the group.
6512 The pattern pointer (ptr) is on the bracket.
6513
6514 If this is a conditional bracket, check that there are no more than
6515 two branches in the group, or just one if it's a DEFINE group. We do this
6516 in the real compile phase, not in the pre-pass, where the whole group may
6517 not be available. */
6518
6519 if (bravalue == OP_COND && lengthptr == NULL)
6520 {
6521 pcre_uchar *tc = code;
6522 int condcount = 0;
6523
6524 do {
6525 condcount++;
6526 tc += GET(tc,1);
6527 }
6528 while (*tc != OP_KET);
6529
6530 /* A DEFINE group is never obeyed inline (the "condition" is always
6531 false). It must have only one branch. */
6532
6533 if (code[LINK_SIZE+1] == OP_DEF)
6534 {
6535 if (condcount > 1)
6536 {
6537 *errorcodeptr = ERR54;
6538 goto FAILED;
6539 }
6540 bravalue = OP_DEF; /* Just a flag to suppress char handling below */
6541 }
6542
6543 /* A "normal" conditional group. If there is just one branch, we must not
6544 make use of its firstchar or reqchar, because this is equivalent to an
6545 empty second branch. */
6546
6547 else
6548 {
6549 if (condcount > 2)
6550 {
6551 *errorcodeptr = ERR27;
6552 goto FAILED;
6553 }
6554 if (condcount == 1) subfirstchar = subreqchar = REQ_NONE;
6555 }
6556 }
6557
6558 /* Error if hit end of pattern */
6559
6560 if (*ptr != CHAR_RIGHT_PARENTHESIS)
6561 {
6562 *errorcodeptr = ERR14;
6563 goto FAILED;
6564 }
6565
6566 /* In the pre-compile phase, update the length by the length of the group,
6567 less the brackets at either end. Then reduce the compiled code to just a
6568 set of non-capturing brackets so that it doesn't use much memory if it is
6569 duplicated by a quantifier.*/
6570
6571 if (lengthptr != NULL)
6572 {
6573 if (OFLOW_MAX - *lengthptr < length_prevgroup - 2 - 2*LINK_SIZE)
6574 {
6575 *errorcodeptr = ERR20;
6576 goto FAILED;
6577 }
6578 *lengthptr += length_prevgroup - 2 - 2*LINK_SIZE;
6579 code++; /* This already contains bravalue */
6580 PUTINC(code, 0, 1 + LINK_SIZE);
6581 *code++ = OP_KET;
6582 PUTINC(code, 0, 1 + LINK_SIZE);
6583 break; /* No need to waste time with special character handling */
6584 }
6585
6586 /* Otherwise update the main code pointer to the end of the group. */
6587
6588 code = tempcode;
6589
6590 /* For a DEFINE group, required and first character settings are not
6591 relevant. */
6592
6593 if (bravalue == OP_DEF) break;
6594
6595 /* Handle updating of the required and first characters for other types of
6596 group. Update for normal brackets of all kinds, and conditions with two
6597 branches (see code above). If the bracket is followed by a quantifier with
6598 zero repeat, we have to back off. Hence the definition of zeroreqchar and
6599 zerofirstchar outside the main loop so that they can be accessed for the
6600 back off. */
6601
6602 zeroreqchar = reqchar;
6603 zerofirstchar = firstchar;
6604 groupsetfirstchar = FALSE;
6605
6606 if (bravalue >= OP_ONCE)
6607 {
6608 /* If we have not yet set a firstchar in this branch, take it from the
6609 subpattern, remembering that it was set here so that a repeat of more
6610 than one can replicate it as reqchar if necessary. If the subpattern has
6611 no firstchar, set "none" for the whole branch. In both cases, a zero
6612 repeat forces firstchar to "none". */
6613
6614 if (firstchar == REQ_UNSET)
6615 {
6616 if (subfirstchar >= 0)
6617 {
6618 firstchar = subfirstchar;
6619 groupsetfirstchar = TRUE;
6620 }
6621 else firstchar = REQ_NONE;
6622 zerofirstchar = REQ_NONE;
6623 }
6624
6625 /* If firstchar was previously set, convert the subpattern's firstchar
6626 into reqchar if there wasn't one, using the vary flag that was in
6627 existence beforehand. */
6628
6629 else if (subfirstchar >= 0 && subreqchar < 0)
6630 subreqchar = subfirstchar | tempreqvary;
6631
6632 /* If the subpattern set a required byte (or set a first byte that isn't
6633 really the first byte - see above), set it. */
6634
6635 if (subreqchar >= 0) reqchar = subreqchar;
6636 }
6637
6638 /* For a forward assertion, we take the reqchar, if set. This can be
6639 helpful if the pattern that follows the assertion doesn't set a different
6640 char. For example, it's useful for /(?=abcde).+/. We can't set firstchar
6641 for an assertion, however because it leads to incorrect effect for patterns
6642 such as /(?=a)a.+/ when the "real" "a" would then become a reqchar instead
6643 of a firstchar. This is overcome by a scan at the end if there's no
6644 firstchar, looking for an asserted first char. */