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

Contents of /code/trunk/pcre_compile.c

Parent Directory Parent Directory | Revision Log Revision Log


Revision 1305 - (show annotations)
Mon Apr 1 14:50:45 2013 UTC (6 years, 8 months ago) by ph10
File MIME type: text/plain
File size: 276406 byte(s)
Fix open parens in MAKE/SKIP/PRUNE/THEN name bug.
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|32)_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, int, int,
88 pcre_uint32 *, pcre_int32 *, pcre_uint32 *, pcre_int32 *, branch_chain *,
89 compile_data *, int *);
90
91
92
93 /*************************************************
94 * Code parameters and static tables *
95 *************************************************/
96
97 /* This value specifies the size of stack workspace that is used during the
98 first pre-compile phase that determines how much memory is required. The regex
99 is partly compiled into this space, but the compiled parts are discarded as
100 soon as they can be, so that hopefully there will never be an overrun. The code
101 does, however, check for an overrun. The largest amount I've seen used is 218,
102 so this number is very generous.
103
104 The same workspace is used during the second, actual compile phase for
105 remembering forward references to groups so that they can be filled in at the
106 end. Each entry in this list occupies LINK_SIZE bytes, so even when LINK_SIZE
107 is 4 there is plenty of room for most patterns. However, the memory can get
108 filled up by repetitions of forward references, for example patterns like
109 /(?1){0,1999}(b)/, and one user did hit the limit. The code has been changed so
110 that the workspace is expanded using malloc() in this situation. The value
111 below is therefore a minimum, and we put a maximum on it for safety. The
112 minimum is now also defined in terms of LINK_SIZE so that the use of malloc()
113 kicks in at the same number of forward references in all cases. */
114
115 #define COMPILE_WORK_SIZE (2048*LINK_SIZE)
116 #define COMPILE_WORK_SIZE_MAX (100*COMPILE_WORK_SIZE)
117
118 /* The overrun tests check for a slightly smaller size so that they detect the
119 overrun before it actually does run off the end of the data block. */
120
121 #define WORK_SIZE_SAFETY_MARGIN (100)
122
123 /* Private flags added to firstchar and reqchar. */
124
125 #define REQ_CASELESS (1 << 0) /* Indicates caselessness */
126 #define REQ_VARY (1 << 1) /* Reqchar followed non-literal item */
127 /* Negative values for the firstchar and reqchar flags */
128 #define REQ_UNSET (-2)
129 #define REQ_NONE (-1)
130
131 /* Repeated character flags. */
132
133 #define UTF_LENGTH 0x10000000l /* The char contains its length. */
134
135 /* Table for handling escaped characters in the range '0'-'z'. Positive returns
136 are simple data values; negative values are for special things like \d and so
137 on. Zero means further processing is needed (for things like \x), or the escape
138 is invalid. */
139
140 #ifndef EBCDIC
141
142 /* This is the "normal" table for ASCII systems or for EBCDIC systems running
143 in UTF-8 mode. */
144
145 static const short int escapes[] = {
146 0, 0,
147 0, 0,
148 0, 0,
149 0, 0,
150 0, 0,
151 CHAR_COLON, CHAR_SEMICOLON,
152 CHAR_LESS_THAN_SIGN, CHAR_EQUALS_SIGN,
153 CHAR_GREATER_THAN_SIGN, CHAR_QUESTION_MARK,
154 CHAR_COMMERCIAL_AT, -ESC_A,
155 -ESC_B, -ESC_C,
156 -ESC_D, -ESC_E,
157 0, -ESC_G,
158 -ESC_H, 0,
159 0, -ESC_K,
160 0, 0,
161 -ESC_N, 0,
162 -ESC_P, -ESC_Q,
163 -ESC_R, -ESC_S,
164 0, 0,
165 -ESC_V, -ESC_W,
166 -ESC_X, 0,
167 -ESC_Z, CHAR_LEFT_SQUARE_BRACKET,
168 CHAR_BACKSLASH, CHAR_RIGHT_SQUARE_BRACKET,
169 CHAR_CIRCUMFLEX_ACCENT, CHAR_UNDERSCORE,
170 CHAR_GRAVE_ACCENT, 7,
171 -ESC_b, 0,
172 -ESC_d, ESC_e,
173 ESC_f, 0,
174 -ESC_h, 0,
175 0, -ESC_k,
176 0, 0,
177 ESC_n, 0,
178 -ESC_p, 0,
179 ESC_r, -ESC_s,
180 ESC_tee, 0,
181 -ESC_v, -ESC_w,
182 0, 0,
183 -ESC_z
184 };
185
186 #else
187
188 /* This is the "abnormal" table for EBCDIC systems without UTF-8 support. */
189
190 static const short int escapes[] = {
191 /* 48 */ 0, 0, 0, '.', '<', '(', '+', '|',
192 /* 50 */ '&', 0, 0, 0, 0, 0, 0, 0,
193 /* 58 */ 0, 0, '!', '$', '*', ')', ';', '~',
194 /* 60 */ '-', '/', 0, 0, 0, 0, 0, 0,
195 /* 68 */ 0, 0, '|', ',', '%', '_', '>', '?',
196 /* 70 */ 0, 0, 0, 0, 0, 0, 0, 0,
197 /* 78 */ 0, '`', ':', '#', '@', '\'', '=', '"',
198 /* 80 */ 0, 7, -ESC_b, 0, -ESC_d, ESC_e, ESC_f, 0,
199 /* 88 */-ESC_h, 0, 0, '{', 0, 0, 0, 0,
200 /* 90 */ 0, 0, -ESC_k, 'l', 0, ESC_n, 0, -ESC_p,
201 /* 98 */ 0, ESC_r, 0, '}', 0, 0, 0, 0,
202 /* A0 */ 0, '~', -ESC_s, ESC_tee, 0,-ESC_v, -ESC_w, 0,
203 /* A8 */ 0,-ESC_z, 0, 0, 0, '[', 0, 0,
204 /* B0 */ 0, 0, 0, 0, 0, 0, 0, 0,
205 /* B8 */ 0, 0, 0, 0, 0, ']', '=', '-',
206 /* C0 */ '{',-ESC_A, -ESC_B, -ESC_C, -ESC_D,-ESC_E, 0, -ESC_G,
207 /* C8 */-ESC_H, 0, 0, 0, 0, 0, 0, 0,
208 /* D0 */ '}', 0, -ESC_K, 0, 0,-ESC_N, 0, -ESC_P,
209 /* D8 */-ESC_Q,-ESC_R, 0, 0, 0, 0, 0, 0,
210 /* E0 */ '\\', 0, -ESC_S, 0, 0,-ESC_V, -ESC_W, -ESC_X,
211 /* E8 */ 0,-ESC_Z, 0, 0, 0, 0, 0, 0,
212 /* F0 */ 0, 0, 0, 0, 0, 0, 0, 0,
213 /* F8 */ 0, 0, 0, 0, 0, 0, 0, 0
214 };
215 #endif
216
217
218 /* Table of special "verbs" like (*PRUNE). This is a short table, so it is
219 searched linearly. Put all the names into a single string, in order to reduce
220 the number of relocations when a shared library is dynamically linked. The
221 string is built from string macros so that it works in UTF-8 mode on EBCDIC
222 platforms. */
223
224 typedef struct verbitem {
225 int len; /* Length of verb name */
226 int op; /* Op when no arg, or -1 if arg mandatory */
227 int op_arg; /* Op when arg present, or -1 if not allowed */
228 } verbitem;
229
230 static const char verbnames[] =
231 "\0" /* Empty name is a shorthand for MARK */
232 STRING_MARK0
233 STRING_ACCEPT0
234 STRING_COMMIT0
235 STRING_F0
236 STRING_FAIL0
237 STRING_PRUNE0
238 STRING_SKIP0
239 STRING_THEN;
240
241 static const verbitem verbs[] = {
242 { 0, -1, OP_MARK },
243 { 4, -1, OP_MARK },
244 { 6, OP_ACCEPT, -1 },
245 { 6, OP_COMMIT, -1 },
246 { 1, OP_FAIL, -1 },
247 { 4, OP_FAIL, -1 },
248 { 5, OP_PRUNE, OP_PRUNE_ARG },
249 { 4, OP_SKIP, OP_SKIP_ARG },
250 { 4, OP_THEN, OP_THEN_ARG }
251 };
252
253 static const int verbcount = sizeof(verbs)/sizeof(verbitem);
254
255
256 /* Tables of names of POSIX character classes and their lengths. The names are
257 now all in a single string, to reduce the number of relocations when a shared
258 library is dynamically loaded. The list of lengths is terminated by a zero
259 length entry. The first three must be alpha, lower, upper, as this is assumed
260 for handling case independence. */
261
262 static const char posix_names[] =
263 STRING_alpha0 STRING_lower0 STRING_upper0 STRING_alnum0
264 STRING_ascii0 STRING_blank0 STRING_cntrl0 STRING_digit0
265 STRING_graph0 STRING_print0 STRING_punct0 STRING_space0
266 STRING_word0 STRING_xdigit;
267
268 static const pcre_uint8 posix_name_lengths[] = {
269 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 4, 6, 0 };
270
271 /* Table of class bit maps for each POSIX class. Each class is formed from a
272 base map, with an optional addition or removal of another map. Then, for some
273 classes, there is some additional tweaking: for [:blank:] the vertical space
274 characters are removed, and for [:alpha:] and [:alnum:] the underscore
275 character is removed. The triples in the table consist of the base map offset,
276 second map offset or -1 if no second map, and a non-negative value for map
277 addition or a negative value for map subtraction (if there are two maps). The
278 absolute value of the third field has these meanings: 0 => no tweaking, 1 =>
279 remove vertical space characters, 2 => remove underscore. */
280
281 static const int posix_class_maps[] = {
282 cbit_word, cbit_digit, -2, /* alpha */
283 cbit_lower, -1, 0, /* lower */
284 cbit_upper, -1, 0, /* upper */
285 cbit_word, -1, 2, /* alnum - word without underscore */
286 cbit_print, cbit_cntrl, 0, /* ascii */
287 cbit_space, -1, 1, /* blank - a GNU extension */
288 cbit_cntrl, -1, 0, /* cntrl */
289 cbit_digit, -1, 0, /* digit */
290 cbit_graph, -1, 0, /* graph */
291 cbit_print, -1, 0, /* print */
292 cbit_punct, -1, 0, /* punct */
293 cbit_space, -1, 0, /* space */
294 cbit_word, -1, 0, /* word - a Perl extension */
295 cbit_xdigit,-1, 0 /* xdigit */
296 };
297
298 /* Table of substitutes for \d etc when PCRE_UCP is set. The POSIX class
299 substitutes must be in the order of the names, defined above, and there are
300 both positive and negative cases. NULL means no substitute. */
301
302 #ifdef SUPPORT_UCP
303 static const pcre_uchar string_PNd[] = {
304 CHAR_BACKSLASH, CHAR_P, CHAR_LEFT_CURLY_BRACKET,
305 CHAR_N, CHAR_d, CHAR_RIGHT_CURLY_BRACKET, '\0' };
306 static const pcre_uchar string_pNd[] = {
307 CHAR_BACKSLASH, CHAR_p, CHAR_LEFT_CURLY_BRACKET,
308 CHAR_N, CHAR_d, CHAR_RIGHT_CURLY_BRACKET, '\0' };
309 static const pcre_uchar string_PXsp[] = {
310 CHAR_BACKSLASH, CHAR_P, CHAR_LEFT_CURLY_BRACKET,
311 CHAR_X, CHAR_s, CHAR_p, CHAR_RIGHT_CURLY_BRACKET, '\0' };
312 static const pcre_uchar string_pXsp[] = {
313 CHAR_BACKSLASH, CHAR_p, CHAR_LEFT_CURLY_BRACKET,
314 CHAR_X, CHAR_s, CHAR_p, CHAR_RIGHT_CURLY_BRACKET, '\0' };
315 static const pcre_uchar string_PXwd[] = {
316 CHAR_BACKSLASH, CHAR_P, CHAR_LEFT_CURLY_BRACKET,
317 CHAR_X, CHAR_w, CHAR_d, CHAR_RIGHT_CURLY_BRACKET, '\0' };
318 static const pcre_uchar string_pXwd[] = {
319 CHAR_BACKSLASH, CHAR_p, CHAR_LEFT_CURLY_BRACKET,
320 CHAR_X, CHAR_w, CHAR_d, CHAR_RIGHT_CURLY_BRACKET, '\0' };
321
322 static const pcre_uchar *substitutes[] = {
323 string_PNd, /* \D */
324 string_pNd, /* \d */
325 string_PXsp, /* \S */ /* NOTE: Xsp is Perl space */
326 string_pXsp, /* \s */
327 string_PXwd, /* \W */
328 string_pXwd /* \w */
329 };
330
331 static const pcre_uchar string_pL[] = {
332 CHAR_BACKSLASH, CHAR_p, CHAR_LEFT_CURLY_BRACKET,
333 CHAR_L, CHAR_RIGHT_CURLY_BRACKET, '\0' };
334 static const pcre_uchar string_pLl[] = {
335 CHAR_BACKSLASH, CHAR_p, CHAR_LEFT_CURLY_BRACKET,
336 CHAR_L, CHAR_l, CHAR_RIGHT_CURLY_BRACKET, '\0' };
337 static const pcre_uchar string_pLu[] = {
338 CHAR_BACKSLASH, CHAR_p, CHAR_LEFT_CURLY_BRACKET,
339 CHAR_L, CHAR_u, CHAR_RIGHT_CURLY_BRACKET, '\0' };
340 static const pcre_uchar string_pXan[] = {
341 CHAR_BACKSLASH, CHAR_p, CHAR_LEFT_CURLY_BRACKET,
342 CHAR_X, CHAR_a, CHAR_n, CHAR_RIGHT_CURLY_BRACKET, '\0' };
343 static const pcre_uchar string_h[] = {
344 CHAR_BACKSLASH, CHAR_h, '\0' };
345 static const pcre_uchar string_pXps[] = {
346 CHAR_BACKSLASH, CHAR_p, CHAR_LEFT_CURLY_BRACKET,
347 CHAR_X, CHAR_p, CHAR_s, CHAR_RIGHT_CURLY_BRACKET, '\0' };
348 static const pcre_uchar string_PL[] = {
349 CHAR_BACKSLASH, CHAR_P, CHAR_LEFT_CURLY_BRACKET,
350 CHAR_L, CHAR_RIGHT_CURLY_BRACKET, '\0' };
351 static const pcre_uchar string_PLl[] = {
352 CHAR_BACKSLASH, CHAR_P, CHAR_LEFT_CURLY_BRACKET,
353 CHAR_L, CHAR_l, CHAR_RIGHT_CURLY_BRACKET, '\0' };
354 static const pcre_uchar string_PLu[] = {
355 CHAR_BACKSLASH, CHAR_P, CHAR_LEFT_CURLY_BRACKET,
356 CHAR_L, CHAR_u, CHAR_RIGHT_CURLY_BRACKET, '\0' };
357 static const pcre_uchar string_PXan[] = {
358 CHAR_BACKSLASH, CHAR_P, CHAR_LEFT_CURLY_BRACKET,
359 CHAR_X, CHAR_a, CHAR_n, CHAR_RIGHT_CURLY_BRACKET, '\0' };
360 static const pcre_uchar string_H[] = {
361 CHAR_BACKSLASH, CHAR_H, '\0' };
362 static const pcre_uchar string_PXps[] = {
363 CHAR_BACKSLASH, CHAR_P, CHAR_LEFT_CURLY_BRACKET,
364 CHAR_X, CHAR_p, CHAR_s, CHAR_RIGHT_CURLY_BRACKET, '\0' };
365
366 static const pcre_uchar *posix_substitutes[] = {
367 string_pL, /* alpha */
368 string_pLl, /* lower */
369 string_pLu, /* upper */
370 string_pXan, /* alnum */
371 NULL, /* ascii */
372 string_h, /* blank */
373 NULL, /* cntrl */
374 string_pNd, /* digit */
375 NULL, /* graph */
376 NULL, /* print */
377 NULL, /* punct */
378 string_pXps, /* space */ /* NOTE: Xps is POSIX space */
379 string_pXwd, /* word */
380 NULL, /* xdigit */
381 /* Negated cases */
382 string_PL, /* ^alpha */
383 string_PLl, /* ^lower */
384 string_PLu, /* ^upper */
385 string_PXan, /* ^alnum */
386 NULL, /* ^ascii */
387 string_H, /* ^blank */
388 NULL, /* ^cntrl */
389 string_PNd, /* ^digit */
390 NULL, /* ^graph */
391 NULL, /* ^print */
392 NULL, /* ^punct */
393 string_PXps, /* ^space */ /* NOTE: Xps is POSIX space */
394 string_PXwd, /* ^word */
395 NULL /* ^xdigit */
396 };
397 #define POSIX_SUBSIZE (sizeof(posix_substitutes) / sizeof(pcre_uchar *))
398 #endif
399
400 #define STRING(a) # a
401 #define XSTRING(s) STRING(s)
402
403 /* The texts of compile-time error messages. These are "char *" because they
404 are passed to the outside world. Do not ever re-use any error number, because
405 they are documented. Always add a new error instead. Messages marked DEAD below
406 are no longer used. This used to be a table of strings, but in order to reduce
407 the number of relocations needed when a shared library is loaded dynamically,
408 it is now one long string. We cannot use a table of offsets, because the
409 lengths of inserts such as XSTRING(MAX_NAME_SIZE) are not known. Instead, we
410 simply count through to the one we want - this isn't a performance issue
411 because these strings are used only when there is a compilation error.
412
413 Each substring ends with \0 to insert a null character. This includes the final
414 substring, so that the whole string ends with \0\0, which can be detected when
415 counting through. */
416
417 static const char error_texts[] =
418 "no error\0"
419 "\\ at end of pattern\0"
420 "\\c at end of pattern\0"
421 "unrecognized character follows \\\0"
422 "numbers out of order in {} quantifier\0"
423 /* 5 */
424 "number too big in {} quantifier\0"
425 "missing terminating ] for character class\0"
426 "invalid escape sequence in character class\0"
427 "range out of order in character class\0"
428 "nothing to repeat\0"
429 /* 10 */
430 "operand of unlimited repeat could match the empty string\0" /** DEAD **/
431 "internal error: unexpected repeat\0"
432 "unrecognized character after (? or (?-\0"
433 "POSIX named classes are supported only within a class\0"
434 "missing )\0"
435 /* 15 */
436 "reference to non-existent subpattern\0"
437 "erroffset passed as NULL\0"
438 "unknown option bit(s) set\0"
439 "missing ) after comment\0"
440 "parentheses nested too deeply\0" /** DEAD **/
441 /* 20 */
442 "regular expression is too large\0"
443 "failed to get memory\0"
444 "unmatched parentheses\0"
445 "internal error: code overflow\0"
446 "unrecognized character after (?<\0"
447 /* 25 */
448 "lookbehind assertion is not fixed length\0"
449 "malformed number or name after (?(\0"
450 "conditional group contains more than two branches\0"
451 "assertion expected after (?(\0"
452 "(?R or (?[+-]digits must be followed by )\0"
453 /* 30 */
454 "unknown POSIX class name\0"
455 "POSIX collating elements are not supported\0"
456 "this version of PCRE is compiled without UTF support\0"
457 "spare error\0" /** DEAD **/
458 "character value in \\x{...} sequence is too large\0"
459 /* 35 */
460 "invalid condition (?(0)\0"
461 "\\C not allowed in lookbehind assertion\0"
462 "PCRE does not support \\L, \\l, \\N{name}, \\U, or \\u\0"
463 "number after (?C is > 255\0"
464 "closing ) for (?C expected\0"
465 /* 40 */
466 "recursive call could loop indefinitely\0"
467 "unrecognized character after (?P\0"
468 "syntax error in subpattern name (missing terminator)\0"
469 "two named subpatterns have the same name\0"
470 "invalid UTF-8 string\0"
471 /* 45 */
472 "support for \\P, \\p, and \\X has not been compiled\0"
473 "malformed \\P or \\p sequence\0"
474 "unknown property name after \\P or \\p\0"
475 "subpattern name is too long (maximum " XSTRING(MAX_NAME_SIZE) " characters)\0"
476 "too many named subpatterns (maximum " XSTRING(MAX_NAME_COUNT) ")\0"
477 /* 50 */
478 "repeated subpattern is too long\0" /** DEAD **/
479 "octal value is greater than \\377 in 8-bit non-UTF-8 mode\0"
480 "internal error: overran compiling workspace\0"
481 "internal error: previously-checked referenced subpattern not found\0"
482 "DEFINE group contains more than one branch\0"
483 /* 55 */
484 "repeating a DEFINE group is not allowed\0" /** DEAD **/
485 "inconsistent NEWLINE options\0"
486 "\\g is not followed by a braced, angle-bracketed, or quoted name/number or by a plain number\0"
487 "a numbered reference must not be zero\0"
488 "an argument is not allowed for (*ACCEPT), (*FAIL), or (*COMMIT)\0"
489 /* 60 */
490 "(*VERB) not recognized\0"
491 "number is too big\0"
492 "subpattern name expected\0"
493 "digit expected after (?+\0"
494 "] is an invalid data character in JavaScript compatibility mode\0"
495 /* 65 */
496 "different names for subpatterns of the same number are not allowed\0"
497 "(*MARK) must have an argument\0"
498 "this version of PCRE is not compiled with Unicode property support\0"
499 "\\c must be followed by an ASCII character\0"
500 "\\k is not followed by a braced, angle-bracketed, or quoted name\0"
501 /* 70 */
502 "internal error: unknown opcode in find_fixedlength()\0"
503 "\\N is not supported in a class\0"
504 "too many forward references\0"
505 "disallowed Unicode code point (>= 0xd800 && <= 0xdfff)\0"
506 "invalid UTF-16 string\0"
507 /* 75 */
508 "name is too long in (*MARK), (*PRUNE), (*SKIP), or (*THEN)\0"
509 "character value in \\u.... sequence is too large\0"
510 "invalid UTF-32 string\0"
511 ;
512
513 /* Table to identify digits and hex digits. This is used when compiling
514 patterns. Note that the tables in chartables are dependent on the locale, and
515 may mark arbitrary characters as digits - but the PCRE compiling code expects
516 to handle only 0-9, a-z, and A-Z as digits when compiling. That is why we have
517 a private table here. It costs 256 bytes, but it is a lot faster than doing
518 character value tests (at least in some simple cases I timed), and in some
519 applications one wants PCRE to compile efficiently as well as match
520 efficiently.
521
522 For convenience, we use the same bit definitions as in chartables:
523
524 0x04 decimal digit
525 0x08 hexadecimal digit
526
527 Then we can use ctype_digit and ctype_xdigit in the code. */
528
529 /* Using a simple comparison for decimal numbers rather than a memory read
530 is much faster, and the resulting code is simpler (the compiler turns it
531 into a subtraction and unsigned comparison). */
532
533 #define IS_DIGIT(x) ((x) >= CHAR_0 && (x) <= CHAR_9)
534
535 #ifndef EBCDIC
536
537 /* This is the "normal" case, for ASCII systems, and EBCDIC systems running in
538 UTF-8 mode. */
539
540 static const pcre_uint8 digitab[] =
541 {
542 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 0- 7 */
543 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 8- 15 */
544 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 16- 23 */
545 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 24- 31 */
546 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - ' */
547 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* ( - / */
548 0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c, /* 0 - 7 */
549 0x0c,0x0c,0x00,0x00,0x00,0x00,0x00,0x00, /* 8 - ? */
550 0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* @ - G */
551 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* H - O */
552 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* P - W */
553 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* X - _ */
554 0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* ` - g */
555 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* h - o */
556 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* p - w */
557 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* x -127 */
558 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 128-135 */
559 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 136-143 */
560 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 144-151 */
561 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 152-159 */
562 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 160-167 */
563 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 168-175 */
564 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 176-183 */
565 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 184-191 */
566 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 192-199 */
567 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 200-207 */
568 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 208-215 */
569 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 216-223 */
570 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 224-231 */
571 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 232-239 */
572 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 240-247 */
573 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00};/* 248-255 */
574
575 #else
576
577 /* This is the "abnormal" case, for EBCDIC systems not running in UTF-8 mode. */
578
579 static const pcre_uint8 digitab[] =
580 {
581 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 0- 7 0 */
582 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 8- 15 */
583 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 16- 23 10 */
584 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 24- 31 */
585 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 32- 39 20 */
586 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 40- 47 */
587 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 48- 55 30 */
588 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 56- 63 */
589 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - 71 40 */
590 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 72- | */
591 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* & - 87 50 */
592 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 88- 95 */
593 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - -103 60 */
594 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 104- ? */
595 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 112-119 70 */
596 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 120- " */
597 0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* 128- g 80 */
598 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* h -143 */
599 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 144- p 90 */
600 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* q -159 */
601 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 160- x A0 */
602 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* y -175 */
603 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* ^ -183 B0 */
604 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 184-191 */
605 0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* { - G C0 */
606 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* H -207 */
607 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* } - P D0 */
608 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* Q -223 */
609 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* \ - X E0 */
610 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* Y -239 */
611 0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c, /* 0 - 7 F0 */
612 0x0c,0x0c,0x00,0x00,0x00,0x00,0x00,0x00};/* 8 -255 */
613
614 static const pcre_uint8 ebcdic_chartab[] = { /* chartable partial dup */
615 0x80,0x00,0x00,0x00,0x00,0x01,0x00,0x00, /* 0- 7 */
616 0x00,0x00,0x00,0x00,0x01,0x01,0x00,0x00, /* 8- 15 */
617 0x00,0x00,0x00,0x00,0x00,0x01,0x00,0x00, /* 16- 23 */
618 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 24- 31 */
619 0x00,0x00,0x00,0x00,0x00,0x01,0x00,0x00, /* 32- 39 */
620 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 40- 47 */
621 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 48- 55 */
622 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 56- 63 */
623 0x01,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - 71 */
624 0x00,0x00,0x00,0x80,0x00,0x80,0x80,0x80, /* 72- | */
625 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* & - 87 */
626 0x00,0x00,0x00,0x80,0x80,0x80,0x00,0x00, /* 88- 95 */
627 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - -103 */
628 0x00,0x00,0x00,0x00,0x00,0x10,0x00,0x80, /* 104- ? */
629 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 112-119 */
630 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 120- " */
631 0x00,0x1a,0x1a,0x1a,0x1a,0x1a,0x1a,0x12, /* 128- g */
632 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* h -143 */
633 0x00,0x12,0x12,0x12,0x12,0x12,0x12,0x12, /* 144- p */
634 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* q -159 */
635 0x00,0x00,0x12,0x12,0x12,0x12,0x12,0x12, /* 160- x */
636 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* y -175 */
637 0x80,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* ^ -183 */
638 0x00,0x00,0x80,0x00,0x00,0x00,0x00,0x00, /* 184-191 */
639 0x80,0x1a,0x1a,0x1a,0x1a,0x1a,0x1a,0x12, /* { - G */
640 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* H -207 */
641 0x00,0x12,0x12,0x12,0x12,0x12,0x12,0x12, /* } - P */
642 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* Q -223 */
643 0x00,0x00,0x12,0x12,0x12,0x12,0x12,0x12, /* \ - X */
644 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* Y -239 */
645 0x1c,0x1c,0x1c,0x1c,0x1c,0x1c,0x1c,0x1c, /* 0 - 7 */
646 0x1c,0x1c,0x00,0x00,0x00,0x00,0x00,0x00};/* 8 -255 */
647 #endif
648
649
650
651 /*************************************************
652 * Find an error text *
653 *************************************************/
654
655 /* The error texts are now all in one long string, to save on relocations. As
656 some of the text is of unknown length, we can't use a table of offsets.
657 Instead, just count through the strings. This is not a performance issue
658 because it happens only when there has been a compilation error.
659
660 Argument: the error number
661 Returns: pointer to the error string
662 */
663
664 static const char *
665 find_error_text(int n)
666 {
667 const char *s = error_texts;
668 for (; n > 0; n--)
669 {
670 while (*s++ != CHAR_NULL) {};
671 if (*s == CHAR_NULL) return "Error text not found (please report)";
672 }
673 return s;
674 }
675
676
677 /*************************************************
678 * Expand the workspace *
679 *************************************************/
680
681 /* This function is called during the second compiling phase, if the number of
682 forward references fills the existing workspace, which is originally a block on
683 the stack. A larger block is obtained from malloc() unless the ultimate limit
684 has been reached or the increase will be rather small.
685
686 Argument: pointer to the compile data block
687 Returns: 0 if all went well, else an error number
688 */
689
690 static int
691 expand_workspace(compile_data *cd)
692 {
693 pcre_uchar *newspace;
694 int newsize = cd->workspace_size * 2;
695
696 if (newsize > COMPILE_WORK_SIZE_MAX) newsize = COMPILE_WORK_SIZE_MAX;
697 if (cd->workspace_size >= COMPILE_WORK_SIZE_MAX ||
698 newsize - cd->workspace_size < WORK_SIZE_SAFETY_MARGIN)
699 return ERR72;
700
701 newspace = (PUBL(malloc))(IN_UCHARS(newsize));
702 if (newspace == NULL) return ERR21;
703 memcpy(newspace, cd->start_workspace, cd->workspace_size * sizeof(pcre_uchar));
704 cd->hwm = (pcre_uchar *)newspace + (cd->hwm - cd->start_workspace);
705 if (cd->workspace_size > COMPILE_WORK_SIZE)
706 (PUBL(free))((void *)cd->start_workspace);
707 cd->start_workspace = newspace;
708 cd->workspace_size = newsize;
709 return 0;
710 }
711
712
713
714 /*************************************************
715 * Check for counted repeat *
716 *************************************************/
717
718 /* This function is called when a '{' is encountered in a place where it might
719 start a quantifier. It looks ahead to see if it really is a quantifier or not.
720 It is only a quantifier if it is one of the forms {ddd} {ddd,} or {ddd,ddd}
721 where the ddds are digits.
722
723 Arguments:
724 p pointer to the first char after '{'
725
726 Returns: TRUE or FALSE
727 */
728
729 static BOOL
730 is_counted_repeat(const pcre_uchar *p)
731 {
732 if (!IS_DIGIT(*p)) return FALSE;
733 p++;
734 while (IS_DIGIT(*p)) p++;
735 if (*p == CHAR_RIGHT_CURLY_BRACKET) return TRUE;
736
737 if (*p++ != CHAR_COMMA) return FALSE;
738 if (*p == CHAR_RIGHT_CURLY_BRACKET) return TRUE;
739
740 if (!IS_DIGIT(*p)) return FALSE;
741 p++;
742 while (IS_DIGIT(*p)) p++;
743
744 return (*p == CHAR_RIGHT_CURLY_BRACKET);
745 }
746
747
748
749 /*************************************************
750 * Handle escapes *
751 *************************************************/
752
753 /* This function is called when a \ has been encountered. It either returns a
754 positive value for a simple escape such as \n, or 0 for a data character
755 which will be placed in chptr. A backreference to group n is returned as
756 negative n. When UTF-8 is enabled, a positive value greater than 255 may
757 be returned in chptr.
758 On entry,ptr is pointing at the \. On exit, it is on the final character of the
759 escape sequence.
760
761 Arguments:
762 ptrptr points to the pattern position pointer
763 chptr points to the data character
764 errorcodeptr points to the errorcode variable
765 bracount number of previous extracting brackets
766 options the options bits
767 isclass TRUE if inside a character class
768
769 Returns: zero => a data character
770 positive => a special escape sequence
771 negative => a back reference
772 on error, errorcodeptr is set
773 */
774
775 static int
776 check_escape(const pcre_uchar **ptrptr, pcre_uint32 *chptr, int *errorcodeptr,
777 int bracount, int options, BOOL isclass)
778 {
779 /* PCRE_UTF16 has the same value as PCRE_UTF8. */
780 BOOL utf = (options & PCRE_UTF8) != 0;
781 const pcre_uchar *ptr = *ptrptr + 1;
782 pcre_uint32 c;
783 int escape = 0;
784 int i;
785
786 GETCHARINCTEST(c, ptr); /* Get character value, increment pointer */
787 ptr--; /* Set pointer back to the last byte */
788
789 /* If backslash is at the end of the pattern, it's an error. */
790
791 if (c == CHAR_NULL) *errorcodeptr = ERR1;
792
793 /* Non-alphanumerics are literals. For digits or letters, do an initial lookup
794 in a table. A non-zero result is something that can be returned immediately.
795 Otherwise further processing may be required. */
796
797 #ifndef EBCDIC /* ASCII/UTF-8 coding */
798 /* Not alphanumeric */
799 else if (c < CHAR_0 || c > CHAR_z) {}
800 else if ((i = escapes[c - CHAR_0]) != 0)
801 { if (i > 0) c = (pcre_uint32)i; else escape = -i; }
802
803 #else /* EBCDIC coding */
804 /* Not alphanumeric */
805 else if (c < CHAR_a || (!MAX_255(c) || (ebcdic_chartab[c] & 0x0E) == 0)) {}
806 else if ((i = escapes[c - 0x48]) != 0) { if (i > 0) c = (pcre_uint32)i; else escape = -i; }
807 #endif
808
809 /* Escapes that need further processing, or are illegal. */
810
811 else
812 {
813 const pcre_uchar *oldptr;
814 BOOL braced, negated, overflow;
815 int s;
816
817 switch (c)
818 {
819 /* A number of Perl escapes are not handled by PCRE. We give an explicit
820 error. */
821
822 case CHAR_l:
823 case CHAR_L:
824 *errorcodeptr = ERR37;
825 break;
826
827 case CHAR_u:
828 if ((options & PCRE_JAVASCRIPT_COMPAT) != 0)
829 {
830 /* In JavaScript, \u must be followed by four hexadecimal numbers.
831 Otherwise it is a lowercase u letter. */
832 if (MAX_255(ptr[1]) && (digitab[ptr[1]] & ctype_xdigit) != 0
833 && MAX_255(ptr[2]) && (digitab[ptr[2]] & ctype_xdigit) != 0
834 && MAX_255(ptr[3]) && (digitab[ptr[3]] & ctype_xdigit) != 0
835 && MAX_255(ptr[4]) && (digitab[ptr[4]] & ctype_xdigit) != 0)
836 {
837 c = 0;
838 for (i = 0; i < 4; ++i)
839 {
840 register pcre_uint32 cc = *(++ptr);
841 #ifndef EBCDIC /* ASCII/UTF-8 coding */
842 if (cc >= CHAR_a) cc -= 32; /* Convert to upper case */
843 c = (c << 4) + cc - ((cc < CHAR_A)? CHAR_0 : (CHAR_A - 10));
844 #else /* EBCDIC coding */
845 if (cc >= CHAR_a && cc <= CHAR_z) cc += 64; /* Convert to upper case */
846 c = (c << 4) + cc - ((cc >= CHAR_0)? CHAR_0 : (CHAR_A - 10));
847 #endif
848 }
849
850 #if defined COMPILE_PCRE8
851 if (c > (utf ? 0x10ffffU : 0xffU))
852 #elif defined COMPILE_PCRE16
853 if (c > (utf ? 0x10ffffU : 0xffffU))
854 #elif defined COMPILE_PCRE32
855 if (utf && c > 0x10ffffU)
856 #endif
857 {
858 *errorcodeptr = ERR76;
859 }
860 else if (utf && c >= 0xd800 && c <= 0xdfff) *errorcodeptr = ERR73;
861 }
862 }
863 else
864 *errorcodeptr = ERR37;
865 break;
866
867 case CHAR_U:
868 /* In JavaScript, \U is an uppercase U letter. */
869 if ((options & PCRE_JAVASCRIPT_COMPAT) == 0) *errorcodeptr = ERR37;
870 break;
871
872 /* In a character class, \g is just a literal "g". Outside a character
873 class, \g must be followed by one of a number of specific things:
874
875 (1) A number, either plain or braced. If positive, it is an absolute
876 backreference. If negative, it is a relative backreference. This is a Perl
877 5.10 feature.
878
879 (2) Perl 5.10 also supports \g{name} as a reference to a named group. This
880 is part of Perl's movement towards a unified syntax for back references. As
881 this is synonymous with \k{name}, we fudge it up by pretending it really
882 was \k.
883
884 (3) For Oniguruma compatibility we also support \g followed by a name or a
885 number either in angle brackets or in single quotes. However, these are
886 (possibly recursive) subroutine calls, _not_ backreferences. Just return
887 the ESC_g code (cf \k). */
888
889 case CHAR_g:
890 if (isclass) break;
891 if (ptr[1] == CHAR_LESS_THAN_SIGN || ptr[1] == CHAR_APOSTROPHE)
892 {
893 escape = ESC_g;
894 break;
895 }
896
897 /* Handle the Perl-compatible cases */
898
899 if (ptr[1] == CHAR_LEFT_CURLY_BRACKET)
900 {
901 const pcre_uchar *p;
902 for (p = ptr+2; *p != CHAR_NULL && *p != CHAR_RIGHT_CURLY_BRACKET; p++)
903 if (*p != CHAR_MINUS && !IS_DIGIT(*p)) break;
904 if (*p != CHAR_NULL && *p != CHAR_RIGHT_CURLY_BRACKET)
905 {
906 escape = ESC_k;
907 break;
908 }
909 braced = TRUE;
910 ptr++;
911 }
912 else braced = FALSE;
913
914 if (ptr[1] == CHAR_MINUS)
915 {
916 negated = TRUE;
917 ptr++;
918 }
919 else negated = FALSE;
920
921 /* The integer range is limited by the machine's int representation. */
922 s = 0;
923 overflow = FALSE;
924 while (IS_DIGIT(ptr[1]))
925 {
926 if (s > INT_MAX / 10 - 1) /* Integer overflow */
927 {
928 overflow = TRUE;
929 break;
930 }
931 s = s * 10 + (int)(*(++ptr) - CHAR_0);
932 }
933 if (overflow) /* Integer overflow */
934 {
935 while (IS_DIGIT(ptr[1]))
936 ptr++;
937 *errorcodeptr = ERR61;
938 break;
939 }
940
941 if (braced && *(++ptr) != CHAR_RIGHT_CURLY_BRACKET)
942 {
943 *errorcodeptr = ERR57;
944 break;
945 }
946
947 if (s == 0)
948 {
949 *errorcodeptr = ERR58;
950 break;
951 }
952
953 if (negated)
954 {
955 if (s > bracount)
956 {
957 *errorcodeptr = ERR15;
958 break;
959 }
960 s = bracount - (s - 1);
961 }
962
963 escape = -s;
964 break;
965
966 /* The handling of escape sequences consisting of a string of digits
967 starting with one that is not zero is not straightforward. By experiment,
968 the way Perl works seems to be as follows:
969
970 Outside a character class, the digits are read as a decimal number. If the
971 number is less than 10, or if there are that many previous extracting
972 left brackets, then it is a back reference. Otherwise, up to three octal
973 digits are read to form an escaped byte. Thus \123 is likely to be octal
974 123 (cf \0123, which is octal 012 followed by the literal 3). If the octal
975 value is greater than 377, the least significant 8 bits are taken. Inside a
976 character class, \ followed by a digit is always an octal number. */
977
978 case CHAR_1: case CHAR_2: case CHAR_3: case CHAR_4: case CHAR_5:
979 case CHAR_6: case CHAR_7: case CHAR_8: case CHAR_9:
980
981 if (!isclass)
982 {
983 oldptr = ptr;
984 /* The integer range is limited by the machine's int representation. */
985 s = (int)(c -CHAR_0);
986 overflow = FALSE;
987 while (IS_DIGIT(ptr[1]))
988 {
989 if (s > INT_MAX / 10 - 1) /* Integer overflow */
990 {
991 overflow = TRUE;
992 break;
993 }
994 s = s * 10 + (int)(*(++ptr) - CHAR_0);
995 }
996 if (overflow) /* Integer overflow */
997 {
998 while (IS_DIGIT(ptr[1]))
999 ptr++;
1000 *errorcodeptr = ERR61;
1001 break;
1002 }
1003 if (s < 10 || s <= bracount)
1004 {
1005 escape = -s;
1006 break;
1007 }
1008 ptr = oldptr; /* Put the pointer back and fall through */
1009 }
1010
1011 /* Handle an octal number following \. If the first digit is 8 or 9, Perl
1012 generates a binary zero byte and treats the digit as a following literal.
1013 Thus we have to pull back the pointer by one. */
1014
1015 if ((c = *ptr) >= CHAR_8)
1016 {
1017 ptr--;
1018 c = 0;
1019 break;
1020 }
1021
1022 /* \0 always starts an octal number, but we may drop through to here with a
1023 larger first octal digit. The original code used just to take the least
1024 significant 8 bits of octal numbers (I think this is what early Perls used
1025 to do). Nowadays we allow for larger numbers in UTF-8 mode and 16-bit mode,
1026 but no more than 3 octal digits. */
1027
1028 case CHAR_0:
1029 c -= CHAR_0;
1030 while(i++ < 2 && ptr[1] >= CHAR_0 && ptr[1] <= CHAR_7)
1031 c = c * 8 + *(++ptr) - CHAR_0;
1032 #ifdef COMPILE_PCRE8
1033 if (!utf && c > 0xff) *errorcodeptr = ERR51;
1034 #endif
1035 break;
1036
1037 /* \x is complicated. \x{ddd} is a character number which can be greater
1038 than 0xff in utf or non-8bit mode, but only if the ddd are hex digits.
1039 If not, { is treated as a data character. */
1040
1041 case CHAR_x:
1042 if ((options & PCRE_JAVASCRIPT_COMPAT) != 0)
1043 {
1044 /* In JavaScript, \x must be followed by two hexadecimal numbers.
1045 Otherwise it is a lowercase x letter. */
1046 if (MAX_255(ptr[1]) && (digitab[ptr[1]] & ctype_xdigit) != 0
1047 && MAX_255(ptr[2]) && (digitab[ptr[2]] & ctype_xdigit) != 0)
1048 {
1049 c = 0;
1050 for (i = 0; i < 2; ++i)
1051 {
1052 register pcre_uint32 cc = *(++ptr);
1053 #ifndef EBCDIC /* ASCII/UTF-8 coding */
1054 if (cc >= CHAR_a) cc -= 32; /* Convert to upper case */
1055 c = (c << 4) + cc - ((cc < CHAR_A)? CHAR_0 : (CHAR_A - 10));
1056 #else /* EBCDIC coding */
1057 if (cc >= CHAR_a && cc <= CHAR_z) cc += 64; /* Convert to upper case */
1058 c = (c << 4) + cc - ((cc >= CHAR_0)? CHAR_0 : (CHAR_A - 10));
1059 #endif
1060 }
1061 }
1062 break;
1063 }
1064
1065 if (ptr[1] == CHAR_LEFT_CURLY_BRACKET)
1066 {
1067 const pcre_uchar *pt = ptr + 2;
1068
1069 c = 0;
1070 overflow = FALSE;
1071 while (MAX_255(*pt) && (digitab[*pt] & ctype_xdigit) != 0)
1072 {
1073 register pcre_uint32 cc = *pt++;
1074 if (c == 0 && cc == CHAR_0) continue; /* Leading zeroes */
1075
1076 #ifdef COMPILE_PCRE32
1077 if (c >= 0x10000000l) { overflow = TRUE; break; }
1078 #endif
1079
1080 #ifndef EBCDIC /* ASCII/UTF-8 coding */
1081 if (cc >= CHAR_a) cc -= 32; /* Convert to upper case */
1082 c = (c << 4) + cc - ((cc < CHAR_A)? CHAR_0 : (CHAR_A - 10));
1083 #else /* EBCDIC coding */
1084 if (cc >= CHAR_a && cc <= CHAR_z) cc += 64; /* Convert to upper case */
1085 c = (c << 4) + cc - ((cc >= CHAR_0)? CHAR_0 : (CHAR_A - 10));
1086 #endif
1087
1088 #if defined COMPILE_PCRE8
1089 if (c > (utf ? 0x10ffffU : 0xffU)) { overflow = TRUE; break; }
1090 #elif defined COMPILE_PCRE16
1091 if (c > (utf ? 0x10ffffU : 0xffffU)) { overflow = TRUE; break; }
1092 #elif defined COMPILE_PCRE32
1093 if (utf && c > 0x10ffffU) { overflow = TRUE; break; }
1094 #endif
1095 }
1096
1097 if (overflow)
1098 {
1099 while (MAX_255(*pt) && (digitab[*pt] & ctype_xdigit) != 0) pt++;
1100 *errorcodeptr = ERR34;
1101 }
1102
1103 if (*pt == CHAR_RIGHT_CURLY_BRACKET)
1104 {
1105 if (utf && c >= 0xd800 && c <= 0xdfff) *errorcodeptr = ERR73;
1106 ptr = pt;
1107 break;
1108 }
1109
1110 /* If the sequence of hex digits does not end with '}', then we don't
1111 recognize this construct; fall through to the normal \x handling. */
1112 }
1113
1114 /* Read just a single-byte hex-defined char */
1115
1116 c = 0;
1117 while (i++ < 2 && MAX_255(ptr[1]) && (digitab[ptr[1]] & ctype_xdigit) != 0)
1118 {
1119 pcre_uint32 cc; /* Some compilers don't like */
1120 cc = *(++ptr); /* ++ in initializers */
1121 #ifndef EBCDIC /* ASCII/UTF-8 coding */
1122 if (cc >= CHAR_a) cc -= 32; /* Convert to upper case */
1123 c = c * 16 + cc - ((cc < CHAR_A)? CHAR_0 : (CHAR_A - 10));
1124 #else /* EBCDIC coding */
1125 if (cc <= CHAR_z) cc += 64; /* Convert to upper case */
1126 c = c * 16 + cc - ((cc >= CHAR_0)? CHAR_0 : (CHAR_A - 10));
1127 #endif
1128 }
1129 break;
1130
1131 /* For \c, a following letter is upper-cased; then the 0x40 bit is flipped.
1132 An error is given if the byte following \c is not an ASCII character. This
1133 coding is ASCII-specific, but then the whole concept of \cx is
1134 ASCII-specific. (However, an EBCDIC equivalent has now been added.) */
1135
1136 case CHAR_c:
1137 c = *(++ptr);
1138 if (c == CHAR_NULL)
1139 {
1140 *errorcodeptr = ERR2;
1141 break;
1142 }
1143 #ifndef EBCDIC /* ASCII/UTF-8 coding */
1144 if (c > 127) /* Excludes all non-ASCII in either mode */
1145 {
1146 *errorcodeptr = ERR68;
1147 break;
1148 }
1149 if (c >= CHAR_a && c <= CHAR_z) c -= 32;
1150 c ^= 0x40;
1151 #else /* EBCDIC coding */
1152 if (c >= CHAR_a && c <= CHAR_z) c += 64;
1153 c ^= 0xC0;
1154 #endif
1155 break;
1156
1157 /* PCRE_EXTRA enables extensions to Perl in the matter of escapes. Any
1158 other alphanumeric following \ is an error if PCRE_EXTRA was set;
1159 otherwise, for Perl compatibility, it is a literal. This code looks a bit
1160 odd, but there used to be some cases other than the default, and there may
1161 be again in future, so I haven't "optimized" it. */
1162
1163 default:
1164 if ((options & PCRE_EXTRA) != 0) switch(c)
1165 {
1166 default:
1167 *errorcodeptr = ERR3;
1168 break;
1169 }
1170 break;
1171 }
1172 }
1173
1174 /* Perl supports \N{name} for character names, as well as plain \N for "not
1175 newline". PCRE does not support \N{name}. However, it does support
1176 quantification such as \N{2,3}. */
1177
1178 if (escape == ESC_N && ptr[1] == CHAR_LEFT_CURLY_BRACKET &&
1179 !is_counted_repeat(ptr+2))
1180 *errorcodeptr = ERR37;
1181
1182 /* If PCRE_UCP is set, we change the values for \d etc. */
1183
1184 if ((options & PCRE_UCP) != 0 && escape >= ESC_D && escape <= ESC_w)
1185 escape += (ESC_DU - ESC_D);
1186
1187 /* Set the pointer to the final character before returning. */
1188
1189 *ptrptr = ptr;
1190 *chptr = c;
1191 return escape;
1192 }
1193
1194 #ifdef SUPPORT_UCP
1195 /*************************************************
1196 * Handle \P and \p *
1197 *************************************************/
1198
1199 /* This function is called after \P or \p has been encountered, provided that
1200 PCRE is compiled with support for Unicode properties. On entry, ptrptr is
1201 pointing at the P or p. On exit, it is pointing at the final character of the
1202 escape sequence.
1203
1204 Argument:
1205 ptrptr points to the pattern position pointer
1206 negptr points to a boolean that is set TRUE for negation else FALSE
1207 ptypeptr points to an unsigned int that is set to the type value
1208 pdataptr points to an unsigned int that is set to the detailed property value
1209 errorcodeptr points to the error code variable
1210
1211 Returns: TRUE if the type value was found, or FALSE for an invalid type
1212 */
1213
1214 static BOOL
1215 get_ucp(const pcre_uchar **ptrptr, BOOL *negptr, unsigned int *ptypeptr,
1216 unsigned int *pdataptr, int *errorcodeptr)
1217 {
1218 pcre_uchar c;
1219 int i, bot, top;
1220 const pcre_uchar *ptr = *ptrptr;
1221 pcre_uchar name[32];
1222
1223 c = *(++ptr);
1224 if (c == CHAR_NULL) goto ERROR_RETURN;
1225
1226 *negptr = FALSE;
1227
1228 /* \P or \p can be followed by a name in {}, optionally preceded by ^ for
1229 negation. */
1230
1231 if (c == CHAR_LEFT_CURLY_BRACKET)
1232 {
1233 if (ptr[1] == CHAR_CIRCUMFLEX_ACCENT)
1234 {
1235 *negptr = TRUE;
1236 ptr++;
1237 }
1238 for (i = 0; i < (int)(sizeof(name) / sizeof(pcre_uchar)) - 1; i++)
1239 {
1240 c = *(++ptr);
1241 if (c == CHAR_NULL) goto ERROR_RETURN;
1242 if (c == CHAR_RIGHT_CURLY_BRACKET) break;
1243 name[i] = c;
1244 }
1245 if (c != CHAR_RIGHT_CURLY_BRACKET) goto ERROR_RETURN;
1246 name[i] = 0;
1247 }
1248
1249 /* Otherwise there is just one following character */
1250
1251 else
1252 {
1253 name[0] = c;
1254 name[1] = 0;
1255 }
1256
1257 *ptrptr = ptr;
1258
1259 /* Search for a recognized property name using binary chop */
1260
1261 bot = 0;
1262 top = PRIV(utt_size);
1263
1264 while (bot < top)
1265 {
1266 int r;
1267 i = (bot + top) >> 1;
1268 r = STRCMP_UC_C8(name, PRIV(utt_names) + PRIV(utt)[i].name_offset);
1269 if (r == 0)
1270 {
1271 *ptypeptr = PRIV(utt)[i].type;
1272 *pdataptr = PRIV(utt)[i].value;
1273 return TRUE;
1274 }
1275 if (r > 0) bot = i + 1; else top = i;
1276 }
1277
1278 *errorcodeptr = ERR47;
1279 *ptrptr = ptr;
1280 return FALSE;
1281
1282 ERROR_RETURN:
1283 *errorcodeptr = ERR46;
1284 *ptrptr = ptr;
1285 return FALSE;
1286 }
1287 #endif
1288
1289
1290
1291
1292 /*************************************************
1293 * Read repeat counts *
1294 *************************************************/
1295
1296 /* Read an item of the form {n,m} and return the values. This is called only
1297 after is_counted_repeat() has confirmed that a repeat-count quantifier exists,
1298 so the syntax is guaranteed to be correct, but we need to check the values.
1299
1300 Arguments:
1301 p pointer to first char after '{'
1302 minp pointer to int for min
1303 maxp pointer to int for max
1304 returned as -1 if no max
1305 errorcodeptr points to error code variable
1306
1307 Returns: pointer to '}' on success;
1308 current ptr on error, with errorcodeptr set non-zero
1309 */
1310
1311 static const pcre_uchar *
1312 read_repeat_counts(const pcre_uchar *p, int *minp, int *maxp, int *errorcodeptr)
1313 {
1314 int min = 0;
1315 int max = -1;
1316
1317 /* Read the minimum value and do a paranoid check: a negative value indicates
1318 an integer overflow. */
1319
1320 while (IS_DIGIT(*p)) min = min * 10 + (int)(*p++ - CHAR_0);
1321 if (min < 0 || min > 65535)
1322 {
1323 *errorcodeptr = ERR5;
1324 return p;
1325 }
1326
1327 /* Read the maximum value if there is one, and again do a paranoid on its size.
1328 Also, max must not be less than min. */
1329
1330 if (*p == CHAR_RIGHT_CURLY_BRACKET) max = min; else
1331 {
1332 if (*(++p) != CHAR_RIGHT_CURLY_BRACKET)
1333 {
1334 max = 0;
1335 while(IS_DIGIT(*p)) max = max * 10 + (int)(*p++ - CHAR_0);
1336 if (max < 0 || max > 65535)
1337 {
1338 *errorcodeptr = ERR5;
1339 return p;
1340 }
1341 if (max < min)
1342 {
1343 *errorcodeptr = ERR4;
1344 return p;
1345 }
1346 }
1347 }
1348
1349 /* Fill in the required variables, and pass back the pointer to the terminating
1350 '}'. */
1351
1352 *minp = min;
1353 *maxp = max;
1354 return p;
1355 }
1356
1357
1358
1359 /*************************************************
1360 * Subroutine for finding forward reference *
1361 *************************************************/
1362
1363 /* This recursive function is called only from find_parens() below. The
1364 top-level call starts at the beginning of the pattern. All other calls must
1365 start at a parenthesis. It scans along a pattern's text looking for capturing
1366 subpatterns, and counting them. If it finds a named pattern that matches the
1367 name it is given, it returns its number. Alternatively, if the name is NULL, it
1368 returns when it reaches a given numbered subpattern. Recursion is used to keep
1369 track of subpatterns that reset the capturing group numbers - the (?| feature.
1370
1371 This function was originally called only from the second pass, in which we know
1372 that if (?< or (?' or (?P< is encountered, the name will be correctly
1373 terminated because that is checked in the first pass. There is now one call to
1374 this function in the first pass, to check for a recursive back reference by
1375 name (so that we can make the whole group atomic). In this case, we need check
1376 only up to the current position in the pattern, and that is still OK because
1377 and previous occurrences will have been checked. To make this work, the test
1378 for "end of pattern" is a check against cd->end_pattern in the main loop,
1379 instead of looking for a binary zero. This means that the special first-pass
1380 call can adjust cd->end_pattern temporarily. (Checks for binary zero while
1381 processing items within the loop are OK, because afterwards the main loop will
1382 terminate.)
1383
1384 Arguments:
1385 ptrptr address of the current character pointer (updated)
1386 cd compile background data
1387 name name to seek, or NULL if seeking a numbered subpattern
1388 lorn name length, or subpattern number if name is NULL
1389 xmode TRUE if we are in /x mode
1390 utf TRUE if we are in UTF-8 / UTF-16 / UTF-32 mode
1391 count pointer to the current capturing subpattern number (updated)
1392
1393 Returns: the number of the named subpattern, or -1 if not found
1394 */
1395
1396 static int
1397 find_parens_sub(pcre_uchar **ptrptr, compile_data *cd, const pcre_uchar *name, int lorn,
1398 BOOL xmode, BOOL utf, int *count)
1399 {
1400 pcre_uchar *ptr = *ptrptr;
1401 int start_count = *count;
1402 int hwm_count = start_count;
1403 BOOL dup_parens = FALSE;
1404
1405 /* If the first character is a parenthesis, check on the type of group we are
1406 dealing with. The very first call may not start with a parenthesis. */
1407
1408 if (ptr[0] == CHAR_LEFT_PARENTHESIS)
1409 {
1410 /* Handle specials such as (*SKIP) or (*UTF8) etc. */
1411
1412 if (ptr[1] == CHAR_ASTERISK)
1413 {
1414 ptr += 2;
1415 while (ptr < cd->end_pattern && *ptr != CHAR_RIGHT_PARENTHESIS) ptr++;
1416 }
1417
1418 /* Handle a normal, unnamed capturing parenthesis. */
1419
1420 else if (ptr[1] != CHAR_QUESTION_MARK)
1421 {
1422 *count += 1;
1423 if (name == NULL && *count == lorn) return *count;
1424 ptr++;
1425 }
1426
1427 /* All cases now have (? at the start. Remember when we are in a group
1428 where the parenthesis numbers are duplicated. */
1429
1430 else if (ptr[2] == CHAR_VERTICAL_LINE)
1431 {
1432 ptr += 3;
1433 dup_parens = TRUE;
1434 }
1435
1436 /* Handle comments; all characters are allowed until a ket is reached. */
1437
1438 else if (ptr[2] == CHAR_NUMBER_SIGN)
1439 {
1440 for (ptr += 3; *ptr != CHAR_NULL; ptr++)
1441 if (*ptr == CHAR_RIGHT_PARENTHESIS) break;
1442 goto FAIL_EXIT;
1443 }
1444
1445 /* Handle a condition. If it is an assertion, just carry on so that it
1446 is processed as normal. If not, skip to the closing parenthesis of the
1447 condition (there can't be any nested parens). */
1448
1449 else if (ptr[2] == CHAR_LEFT_PARENTHESIS)
1450 {
1451 ptr += 2;
1452 if (ptr[1] != CHAR_QUESTION_MARK)
1453 {
1454 while (*ptr != CHAR_NULL && *ptr != CHAR_RIGHT_PARENTHESIS) ptr++;
1455 if (*ptr != CHAR_NULL) ptr++;
1456 }
1457 }
1458
1459 /* Start with (? but not a condition. */
1460
1461 else
1462 {
1463 ptr += 2;
1464 if (*ptr == CHAR_P) ptr++; /* Allow optional P */
1465
1466 /* We have to disambiguate (?<! and (?<= from (?<name> for named groups */
1467
1468 if ((*ptr == CHAR_LESS_THAN_SIGN && ptr[1] != CHAR_EXCLAMATION_MARK &&
1469 ptr[1] != CHAR_EQUALS_SIGN) || *ptr == CHAR_APOSTROPHE)
1470 {
1471 pcre_uchar term;
1472 const pcre_uchar *thisname;
1473 *count += 1;
1474 if (name == NULL && *count == lorn) return *count;
1475 term = *ptr++;
1476 if (term == CHAR_LESS_THAN_SIGN) term = CHAR_GREATER_THAN_SIGN;
1477 thisname = ptr;
1478 while (*ptr != term) ptr++;
1479 if (name != NULL && lorn == (int)(ptr - thisname) &&
1480 STRNCMP_UC_UC(name, thisname, (unsigned int)lorn) == 0)
1481 return *count;
1482 term++;
1483 }
1484 }
1485 }
1486
1487 /* Past any initial parenthesis handling, scan for parentheses or vertical
1488 bars. Stop if we get to cd->end_pattern. Note that this is important for the
1489 first-pass call when this value is temporarily adjusted to stop at the current
1490 position. So DO NOT change this to a test for binary zero. */
1491
1492 for (; ptr < cd->end_pattern; ptr++)
1493 {
1494 /* Skip over backslashed characters and also entire \Q...\E */
1495
1496 if (*ptr == CHAR_BACKSLASH)
1497 {
1498 if (*(++ptr) == CHAR_NULL) goto FAIL_EXIT;
1499 if (*ptr == CHAR_Q) for (;;)
1500 {
1501 while (*(++ptr) != CHAR_NULL && *ptr != CHAR_BACKSLASH) {};
1502 if (*ptr == CHAR_NULL) goto FAIL_EXIT;
1503 if (*(++ptr) == CHAR_E) break;
1504 }
1505 continue;
1506 }
1507
1508 /* Skip over character classes; this logic must be similar to the way they
1509 are handled for real. If the first character is '^', skip it. Also, if the
1510 first few characters (either before or after ^) are \Q\E or \E we skip them
1511 too. This makes for compatibility with Perl. Note the use of STR macros to
1512 encode "Q\\E" so that it works in UTF-8 on EBCDIC platforms. */
1513
1514 if (*ptr == CHAR_LEFT_SQUARE_BRACKET)
1515 {
1516 BOOL negate_class = FALSE;
1517 for (;;)
1518 {
1519 if (ptr[1] == CHAR_BACKSLASH)
1520 {
1521 if (ptr[2] == CHAR_E)
1522 ptr+= 2;
1523 else if (STRNCMP_UC_C8(ptr + 2,
1524 STR_Q STR_BACKSLASH STR_E, 3) == 0)
1525 ptr += 4;
1526 else
1527 break;
1528 }
1529 else if (!negate_class && ptr[1] == CHAR_CIRCUMFLEX_ACCENT)
1530 {
1531 negate_class = TRUE;
1532 ptr++;
1533 }
1534 else break;
1535 }
1536
1537 /* If the next character is ']', it is a data character that must be
1538 skipped, except in JavaScript compatibility mode. */
1539
1540 if (ptr[1] == CHAR_RIGHT_SQUARE_BRACKET &&
1541 (cd->external_options & PCRE_JAVASCRIPT_COMPAT) == 0)
1542 ptr++;
1543
1544 while (*(++ptr) != CHAR_RIGHT_SQUARE_BRACKET)
1545 {
1546 if (*ptr == CHAR_NULL) return -1;
1547 if (*ptr == CHAR_BACKSLASH)
1548 {
1549 if (*(++ptr) == CHAR_NULL) goto FAIL_EXIT;
1550 if (*ptr == CHAR_Q) for (;;)
1551 {
1552 while (*(++ptr) != CHAR_NULL && *ptr != CHAR_BACKSLASH) {};
1553 if (*ptr == CHAR_NULL) goto FAIL_EXIT;
1554 if (*(++ptr) == CHAR_E) break;
1555 }
1556 continue;
1557 }
1558 }
1559 continue;
1560 }
1561
1562 /* Skip comments in /x mode */
1563
1564 if (xmode && *ptr == CHAR_NUMBER_SIGN)
1565 {
1566 ptr++;
1567 while (*ptr != CHAR_NULL)
1568 {
1569 if (IS_NEWLINE(ptr)) { ptr += cd->nllen - 1; break; }
1570 ptr++;
1571 #ifdef SUPPORT_UTF
1572 if (utf) FORWARDCHAR(ptr);
1573 #endif
1574 }
1575 if (*ptr == CHAR_NULL) goto FAIL_EXIT;
1576 continue;
1577 }
1578
1579 /* Check for the special metacharacters */
1580
1581 if (*ptr == CHAR_LEFT_PARENTHESIS)
1582 {
1583 int rc = find_parens_sub(&ptr, cd, name, lorn, xmode, utf, count);
1584 if (rc > 0) return rc;
1585 if (*ptr == CHAR_NULL) goto FAIL_EXIT;
1586 }
1587
1588 else if (*ptr == CHAR_RIGHT_PARENTHESIS)
1589 {
1590 if (dup_parens && *count < hwm_count) *count = hwm_count;
1591 goto FAIL_EXIT;
1592 }
1593
1594 else if (*ptr == CHAR_VERTICAL_LINE && dup_parens)
1595 {
1596 if (*count > hwm_count) hwm_count = *count;
1597 *count = start_count;
1598 }
1599 }
1600
1601 FAIL_EXIT:
1602 *ptrptr = ptr;
1603 return -1;
1604 }
1605
1606
1607
1608
1609 /*************************************************
1610 * Find forward referenced subpattern *
1611 *************************************************/
1612
1613 /* This function scans along a pattern's text looking for capturing
1614 subpatterns, and counting them. If it finds a named pattern that matches the
1615 name it is given, it returns its number. Alternatively, if the name is NULL, it
1616 returns when it reaches a given numbered subpattern. This is used for forward
1617 references to subpatterns. We used to be able to start this scan from the
1618 current compiling point, using the current count value from cd->bracount, and
1619 do it all in a single loop, but the addition of the possibility of duplicate
1620 subpattern numbers means that we have to scan from the very start, in order to
1621 take account of such duplicates, and to use a recursive function to keep track
1622 of the different types of group.
1623
1624 Arguments:
1625 cd compile background data
1626 name name to seek, or NULL if seeking a numbered subpattern
1627 lorn name length, or subpattern number if name is NULL
1628 xmode TRUE if we are in /x mode
1629 utf TRUE if we are in UTF-8 / UTF-16 / UTF-32 mode
1630
1631 Returns: the number of the found subpattern, or -1 if not found
1632 */
1633
1634 static int
1635 find_parens(compile_data *cd, const pcre_uchar *name, int lorn, BOOL xmode,
1636 BOOL utf)
1637 {
1638 pcre_uchar *ptr = (pcre_uchar *)cd->start_pattern;
1639 int count = 0;
1640 int rc;
1641
1642 /* If the pattern does not start with an opening parenthesis, the first call
1643 to find_parens_sub() will scan right to the end (if necessary). However, if it
1644 does start with a parenthesis, find_parens_sub() will return when it hits the
1645 matching closing parens. That is why we have to have a loop. */
1646
1647 for (;;)
1648 {
1649 rc = find_parens_sub(&ptr, cd, name, lorn, xmode, utf, &count);
1650 if (rc > 0 || *ptr++ == CHAR_NULL) break;
1651 }
1652
1653 return rc;
1654 }
1655
1656
1657
1658
1659 /*************************************************
1660 * Find first significant op code *
1661 *************************************************/
1662
1663 /* This is called by several functions that scan a compiled expression looking
1664 for a fixed first character, or an anchoring op code etc. It skips over things
1665 that do not influence this. For some calls, it makes sense to skip negative
1666 forward and all backward assertions, and also the \b assertion; for others it
1667 does not.
1668
1669 Arguments:
1670 code pointer to the start of the group
1671 skipassert TRUE if certain assertions are to be skipped
1672
1673 Returns: pointer to the first significant opcode
1674 */
1675
1676 static const pcre_uchar*
1677 first_significant_code(const pcre_uchar *code, BOOL skipassert)
1678 {
1679 for (;;)
1680 {
1681 switch ((int)*code)
1682 {
1683 case OP_ASSERT_NOT:
1684 case OP_ASSERTBACK:
1685 case OP_ASSERTBACK_NOT:
1686 if (!skipassert) return code;
1687 do code += GET(code, 1); while (*code == OP_ALT);
1688 code += PRIV(OP_lengths)[*code];
1689 break;
1690
1691 case OP_WORD_BOUNDARY:
1692 case OP_NOT_WORD_BOUNDARY:
1693 if (!skipassert) return code;
1694 /* Fall through */
1695
1696 case OP_CALLOUT:
1697 case OP_CREF:
1698 case OP_NCREF:
1699 case OP_RREF:
1700 case OP_NRREF:
1701 case OP_DEF:
1702 code += PRIV(OP_lengths)[*code];
1703 break;
1704
1705 default:
1706 return code;
1707 }
1708 }
1709 /* Control never reaches here */
1710 }
1711
1712
1713
1714
1715 /*************************************************
1716 * Find the fixed length of a branch *
1717 *************************************************/
1718
1719 /* Scan a branch and compute the fixed length of subject that will match it,
1720 if the length is fixed. This is needed for dealing with backward assertions.
1721 In UTF8 mode, the result is in characters rather than bytes. The branch is
1722 temporarily terminated with OP_END when this function is called.
1723
1724 This function is called when a backward assertion is encountered, so that if it
1725 fails, the error message can point to the correct place in the pattern.
1726 However, we cannot do this when the assertion contains subroutine calls,
1727 because they can be forward references. We solve this by remembering this case
1728 and doing the check at the end; a flag specifies which mode we are running in.
1729
1730 Arguments:
1731 code points to the start of the pattern (the bracket)
1732 utf TRUE in UTF-8 / UTF-16 / UTF-32 mode
1733 atend TRUE if called when the pattern is complete
1734 cd the "compile data" structure
1735
1736 Returns: the fixed length,
1737 or -1 if there is no fixed length,
1738 or -2 if \C was encountered (in UTF-8 mode only)
1739 or -3 if an OP_RECURSE item was encountered and atend is FALSE
1740 or -4 if an unknown opcode was encountered (internal error)
1741 */
1742
1743 static int
1744 find_fixedlength(pcre_uchar *code, BOOL utf, BOOL atend, compile_data *cd)
1745 {
1746 int length = -1;
1747
1748 register int branchlength = 0;
1749 register pcre_uchar *cc = code + 1 + LINK_SIZE;
1750
1751 /* Scan along the opcodes for this branch. If we get to the end of the
1752 branch, check the length against that of the other branches. */
1753
1754 for (;;)
1755 {
1756 int d;
1757 pcre_uchar *ce, *cs;
1758 register pcre_uchar op = *cc;
1759
1760 switch (op)
1761 {
1762 /* We only need to continue for OP_CBRA (normal capturing bracket) and
1763 OP_BRA (normal non-capturing bracket) because the other variants of these
1764 opcodes are all concerned with unlimited repeated groups, which of course
1765 are not of fixed length. */
1766
1767 case OP_CBRA:
1768 case OP_BRA:
1769 case OP_ONCE:
1770 case OP_ONCE_NC:
1771 case OP_COND:
1772 d = find_fixedlength(cc + ((op == OP_CBRA)? IMM2_SIZE : 0), utf, atend, cd);
1773 if (d < 0) return d;
1774 branchlength += d;
1775 do cc += GET(cc, 1); while (*cc == OP_ALT);
1776 cc += 1 + LINK_SIZE;
1777 break;
1778
1779 /* Reached end of a branch; if it's a ket it is the end of a nested call.
1780 If it's ALT it is an alternation in a nested call. An ACCEPT is effectively
1781 an ALT. If it is END it's the end of the outer call. All can be handled by
1782 the same code. Note that we must not include the OP_KETRxxx opcodes here,
1783 because they all imply an unlimited repeat. */
1784
1785 case OP_ALT:
1786 case OP_KET:
1787 case OP_END:
1788 case OP_ACCEPT:
1789 case OP_ASSERT_ACCEPT:
1790 if (length < 0) length = branchlength;
1791 else if (length != branchlength) return -1;
1792 if (*cc != OP_ALT) return length;
1793 cc += 1 + LINK_SIZE;
1794 branchlength = 0;
1795 break;
1796
1797 /* A true recursion implies not fixed length, but a subroutine call may
1798 be OK. If the subroutine is a forward reference, we can't deal with
1799 it until the end of the pattern, so return -3. */
1800
1801 case OP_RECURSE:
1802 if (!atend) return -3;
1803 cs = ce = (pcre_uchar *)cd->start_code + GET(cc, 1); /* Start subpattern */
1804 do ce += GET(ce, 1); while (*ce == OP_ALT); /* End subpattern */
1805 if (cc > cs && cc < ce) return -1; /* Recursion */
1806 d = find_fixedlength(cs + IMM2_SIZE, utf, atend, cd);
1807 if (d < 0) return d;
1808 branchlength += d;
1809 cc += 1 + LINK_SIZE;
1810 break;
1811
1812 /* Skip over assertive subpatterns */
1813
1814 case OP_ASSERT:
1815 case OP_ASSERT_NOT:
1816 case OP_ASSERTBACK:
1817 case OP_ASSERTBACK_NOT:
1818 do cc += GET(cc, 1); while (*cc == OP_ALT);
1819 cc += PRIV(OP_lengths)[*cc];
1820 break;
1821
1822 /* Skip over things that don't match chars */
1823
1824 case OP_MARK:
1825 case OP_PRUNE_ARG:
1826 case OP_SKIP_ARG:
1827 case OP_THEN_ARG:
1828 cc += cc[1] + PRIV(OP_lengths)[*cc];
1829 break;
1830
1831 case OP_CALLOUT:
1832 case OP_CIRC:
1833 case OP_CIRCM:
1834 case OP_CLOSE:
1835 case OP_COMMIT:
1836 case OP_CREF:
1837 case OP_DEF:
1838 case OP_DOLL:
1839 case OP_DOLLM:
1840 case OP_EOD:
1841 case OP_EODN:
1842 case OP_FAIL:
1843 case OP_NCREF:
1844 case OP_NRREF:
1845 case OP_NOT_WORD_BOUNDARY:
1846 case OP_PRUNE:
1847 case OP_REVERSE:
1848 case OP_RREF:
1849 case OP_SET_SOM:
1850 case OP_SKIP:
1851 case OP_SOD:
1852 case OP_SOM:
1853 case OP_THEN:
1854 case OP_WORD_BOUNDARY:
1855 cc += PRIV(OP_lengths)[*cc];
1856 break;
1857
1858 /* Handle literal characters */
1859
1860 case OP_CHAR:
1861 case OP_CHARI:
1862 case OP_NOT:
1863 case OP_NOTI:
1864 branchlength++;
1865 cc += 2;
1866 #ifdef SUPPORT_UTF
1867 if (utf && HAS_EXTRALEN(cc[-1])) cc += GET_EXTRALEN(cc[-1]);
1868 #endif
1869 break;
1870
1871 /* Handle exact repetitions. The count is already in characters, but we
1872 need to skip over a multibyte character in UTF8 mode. */
1873
1874 case OP_EXACT:
1875 case OP_EXACTI:
1876 case OP_NOTEXACT:
1877 case OP_NOTEXACTI:
1878 branchlength += (int)GET2(cc,1);
1879 cc += 2 + IMM2_SIZE;
1880 #ifdef SUPPORT_UTF
1881 if (utf && HAS_EXTRALEN(cc[-1])) cc += GET_EXTRALEN(cc[-1]);
1882 #endif
1883 break;
1884
1885 case OP_TYPEEXACT:
1886 branchlength += GET2(cc,1);
1887 if (cc[1 + IMM2_SIZE] == OP_PROP || cc[1 + IMM2_SIZE] == OP_NOTPROP)
1888 cc += 2;
1889 cc += 1 + IMM2_SIZE + 1;
1890 break;
1891
1892 /* Handle single-char matchers */
1893
1894 case OP_PROP:
1895 case OP_NOTPROP:
1896 cc += 2;
1897 /* Fall through */
1898
1899 case OP_HSPACE:
1900 case OP_VSPACE:
1901 case OP_NOT_HSPACE:
1902 case OP_NOT_VSPACE:
1903 case OP_NOT_DIGIT:
1904 case OP_DIGIT:
1905 case OP_NOT_WHITESPACE:
1906 case OP_WHITESPACE:
1907 case OP_NOT_WORDCHAR:
1908 case OP_WORDCHAR:
1909 case OP_ANY:
1910 case OP_ALLANY:
1911 branchlength++;
1912 cc++;
1913 break;
1914
1915 /* The single-byte matcher isn't allowed. This only happens in UTF-8 mode;
1916 otherwise \C is coded as OP_ALLANY. */
1917
1918 case OP_ANYBYTE:
1919 return -2;
1920
1921 /* Check a class for variable quantification */
1922
1923 case OP_CLASS:
1924 case OP_NCLASS:
1925 #if defined SUPPORT_UTF || defined COMPILE_PCRE16 || defined COMPILE_PCRE32
1926 case OP_XCLASS:
1927 /* The original code caused an unsigned overflow in 64 bit systems,
1928 so now we use a conditional statement. */
1929 if (op == OP_XCLASS)
1930 cc += GET(cc, 1);
1931 else
1932 cc += PRIV(OP_lengths)[OP_CLASS];
1933 #else
1934 cc += PRIV(OP_lengths)[OP_CLASS];
1935 #endif
1936
1937 switch (*cc)
1938 {
1939 case OP_CRPLUS:
1940 case OP_CRMINPLUS:
1941 case OP_CRSTAR:
1942 case OP_CRMINSTAR:
1943 case OP_CRQUERY:
1944 case OP_CRMINQUERY:
1945 return -1;
1946
1947 case OP_CRRANGE:
1948 case OP_CRMINRANGE:
1949 if (GET2(cc,1) != GET2(cc,1+IMM2_SIZE)) return -1;
1950 branchlength += (int)GET2(cc,1);
1951 cc += 1 + 2 * IMM2_SIZE;
1952 break;
1953
1954 default:
1955 branchlength++;
1956 }
1957 break;
1958
1959 /* Anything else is variable length */
1960
1961 case OP_ANYNL:
1962 case OP_BRAMINZERO:
1963 case OP_BRAPOS:
1964 case OP_BRAPOSZERO:
1965 case OP_BRAZERO:
1966 case OP_CBRAPOS:
1967 case OP_EXTUNI:
1968 case OP_KETRMAX:
1969 case OP_KETRMIN:
1970 case OP_KETRPOS:
1971 case OP_MINPLUS:
1972 case OP_MINPLUSI:
1973 case OP_MINQUERY:
1974 case OP_MINQUERYI:
1975 case OP_MINSTAR:
1976 case OP_MINSTARI:
1977 case OP_MINUPTO:
1978 case OP_MINUPTOI:
1979 case OP_NOTMINPLUS:
1980 case OP_NOTMINPLUSI:
1981 case OP_NOTMINQUERY:
1982 case OP_NOTMINQUERYI:
1983 case OP_NOTMINSTAR:
1984 case OP_NOTMINSTARI:
1985 case OP_NOTMINUPTO:
1986 case OP_NOTMINUPTOI:
1987 case OP_NOTPLUS:
1988 case OP_NOTPLUSI:
1989 case OP_NOTPOSPLUS:
1990 case OP_NOTPOSPLUSI:
1991 case OP_NOTPOSQUERY:
1992 case OP_NOTPOSQUERYI:
1993 case OP_NOTPOSSTAR:
1994 case OP_NOTPOSSTARI:
1995 case OP_NOTPOSUPTO:
1996 case OP_NOTPOSUPTOI:
1997 case OP_NOTQUERY:
1998 case OP_NOTQUERYI:
1999 case OP_NOTSTAR:
2000 case OP_NOTSTARI:
2001 case OP_NOTUPTO:
2002 case OP_NOTUPTOI:
2003 case OP_PLUS:
2004 case OP_PLUSI:
2005 case OP_POSPLUS:
2006 case OP_POSPLUSI:
2007 case OP_POSQUERY:
2008 case OP_POSQUERYI:
2009 case OP_POSSTAR:
2010 case OP_POSSTARI:
2011 case OP_POSUPTO:
2012 case OP_POSUPTOI:
2013 case OP_QUERY:
2014 case OP_QUERYI:
2015 case OP_REF:
2016 case OP_REFI:
2017 case OP_SBRA:
2018 case OP_SBRAPOS:
2019 case OP_SCBRA:
2020 case OP_SCBRAPOS:
2021 case OP_SCOND:
2022 case OP_SKIPZERO:
2023 case OP_STAR:
2024 case OP_STARI:
2025 case OP_TYPEMINPLUS:
2026 case OP_TYPEMINQUERY:
2027 case OP_TYPEMINSTAR:
2028 case OP_TYPEMINUPTO:
2029 case OP_TYPEPLUS:
2030 case OP_TYPEPOSPLUS:
2031 case OP_TYPEPOSQUERY:
2032 case OP_TYPEPOSSTAR:
2033 case OP_TYPEPOSUPTO:
2034 case OP_TYPEQUERY:
2035 case OP_TYPESTAR:
2036 case OP_TYPEUPTO:
2037 case OP_UPTO:
2038 case OP_UPTOI:
2039 return -1;
2040
2041 /* Catch unrecognized opcodes so that when new ones are added they
2042 are not forgotten, as has happened in the past. */
2043
2044 default:
2045 return -4;
2046 }
2047 }
2048 /* Control never gets here */
2049 }
2050
2051
2052
2053
2054 /*************************************************
2055 * Scan compiled regex for specific bracket *
2056 *************************************************/
2057
2058 /* This little function scans through a compiled pattern until it finds a
2059 capturing bracket with the given number, or, if the number is negative, an
2060 instance of OP_REVERSE for a lookbehind. The function is global in the C sense
2061 so that it can be called from pcre_study() when finding the minimum matching
2062 length.
2063
2064 Arguments:
2065 code points to start of expression
2066 utf TRUE in UTF-8 / UTF-16 / UTF-32 mode
2067 number the required bracket number or negative to find a lookbehind
2068
2069 Returns: pointer to the opcode for the bracket, or NULL if not found
2070 */
2071
2072 const pcre_uchar *
2073 PRIV(find_bracket)(const pcre_uchar *code, BOOL utf, int number)
2074 {
2075 for (;;)
2076 {
2077 register pcre_uchar c = *code;
2078
2079 if (c == OP_END) return NULL;
2080
2081 /* XCLASS is used for classes that cannot be represented just by a bit
2082 map. This includes negated single high-valued characters. The length in
2083 the table is zero; the actual length is stored in the compiled code. */
2084
2085 if (c == OP_XCLASS) code += GET(code, 1);
2086
2087 /* Handle recursion */
2088
2089 else if (c == OP_REVERSE)
2090 {
2091 if (number < 0) return (pcre_uchar *)code;
2092 code += PRIV(OP_lengths)[c];
2093 }
2094
2095 /* Handle capturing bracket */
2096
2097 else if (c == OP_CBRA || c == OP_SCBRA ||
2098 c == OP_CBRAPOS || c == OP_SCBRAPOS)
2099 {
2100 int n = (int)GET2(code, 1+LINK_SIZE);
2101 if (n == number) return (pcre_uchar *)code;
2102 code += PRIV(OP_lengths)[c];
2103 }
2104
2105 /* Otherwise, we can get the item's length from the table, except that for
2106 repeated character types, we have to test for \p and \P, which have an extra
2107 two bytes of parameters, and for MARK/PRUNE/SKIP/THEN with an argument, we
2108 must add in its length. */
2109
2110 else
2111 {
2112 switch(c)
2113 {
2114 case OP_TYPESTAR:
2115 case OP_TYPEMINSTAR:
2116 case OP_TYPEPLUS:
2117 case OP_TYPEMINPLUS:
2118 case OP_TYPEQUERY:
2119 case OP_TYPEMINQUERY:
2120 case OP_TYPEPOSSTAR:
2121 case OP_TYPEPOSPLUS:
2122 case OP_TYPEPOSQUERY:
2123 if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
2124 break;
2125
2126 case OP_TYPEUPTO:
2127 case OP_TYPEMINUPTO:
2128 case OP_TYPEEXACT:
2129 case OP_TYPEPOSUPTO:
2130 if (code[1 + IMM2_SIZE] == OP_PROP || code[1 + IMM2_SIZE] == OP_NOTPROP)
2131 code += 2;
2132 break;
2133
2134 case OP_MARK:
2135 case OP_PRUNE_ARG:
2136 case OP_SKIP_ARG:
2137 case OP_THEN_ARG:
2138 code += code[1];
2139 break;
2140 }
2141
2142 /* Add in the fixed length from the table */
2143
2144 code += PRIV(OP_lengths)[c];
2145
2146 /* In UTF-8 mode, opcodes that are followed by a character may be followed by
2147 a multi-byte character. The length in the table is a minimum, so we have to
2148 arrange to skip the extra bytes. */
2149
2150 #if defined SUPPORT_UTF && !defined COMPILE_PCRE32
2151 if (utf) switch(c)
2152 {
2153 case OP_CHAR:
2154 case OP_CHARI:
2155 case OP_EXACT:
2156 case OP_EXACTI:
2157 case OP_UPTO:
2158 case OP_UPTOI:
2159 case OP_MINUPTO:
2160 case OP_MINUPTOI:
2161 case OP_POSUPTO:
2162 case OP_POSUPTOI:
2163 case OP_STAR:
2164 case OP_STARI:
2165 case OP_MINSTAR:
2166 case OP_MINSTARI:
2167 case OP_POSSTAR:
2168 case OP_POSSTARI:
2169 case OP_PLUS:
2170 case OP_PLUSI:
2171 case OP_MINPLUS:
2172 case OP_MINPLUSI:
2173 case OP_POSPLUS:
2174 case OP_POSPLUSI:
2175 case OP_QUERY:
2176 case OP_QUERYI:
2177 case OP_MINQUERY:
2178 case OP_MINQUERYI:
2179 case OP_POSQUERY:
2180 case OP_POSQUERYI:
2181 if (HAS_EXTRALEN(code[-1])) code += GET_EXTRALEN(code[-1]);
2182 break;
2183 }
2184 #else
2185 (void)(utf); /* Keep compiler happy by referencing function argument */
2186 #endif
2187 }
2188 }
2189 }
2190
2191
2192
2193 /*************************************************
2194 * Scan compiled regex for recursion reference *
2195 *************************************************/
2196
2197 /* This little function scans through a compiled pattern until it finds an
2198 instance of OP_RECURSE.
2199
2200 Arguments:
2201 code points to start of expression
2202 utf TRUE in UTF-8 / UTF-16 / UTF-32 mode
2203
2204 Returns: pointer to the opcode for OP_RECURSE, or NULL if not found
2205 */
2206
2207 static const pcre_uchar *
2208 find_recurse(const pcre_uchar *code, BOOL utf)
2209 {
2210 for (;;)
2211 {
2212 register pcre_uchar c = *code;
2213 if (c == OP_END) return NULL;
2214 if (c == OP_RECURSE) return code;
2215
2216 /* XCLASS is used for classes that cannot be represented just by a bit
2217 map. This includes negated single high-valued characters. The length in
2218 the table is zero; the actual length is stored in the compiled code. */
2219
2220 if (c == OP_XCLASS) code += GET(code, 1);
2221
2222 /* Otherwise, we can get the item's length from the table, except that for
2223 repeated character types, we have to test for \p and \P, which have an extra
2224 two bytes of parameters, and for MARK/PRUNE/SKIP/THEN with an argument, we
2225 must add in its length. */
2226
2227 else
2228 {
2229 switch(c)
2230 {
2231 case OP_TYPESTAR:
2232 case OP_TYPEMINSTAR:
2233 case OP_TYPEPLUS:
2234 case OP_TYPEMINPLUS:
2235 case OP_TYPEQUERY:
2236 case OP_TYPEMINQUERY:
2237 case OP_TYPEPOSSTAR:
2238 case OP_TYPEPOSPLUS:
2239 case OP_TYPEPOSQUERY:
2240 if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
2241 break;
2242
2243 case OP_TYPEPOSUPTO:
2244 case OP_TYPEUPTO:
2245 case OP_TYPEMINUPTO:
2246 case OP_TYPEEXACT:
2247 if (code[1 + IMM2_SIZE] == OP_PROP || code[1 + IMM2_SIZE] == OP_NOTPROP)
2248 code += 2;
2249 break;
2250
2251 case OP_MARK:
2252 case OP_PRUNE_ARG:
2253 case OP_SKIP_ARG:
2254 case OP_THEN_ARG:
2255 code += code[1];
2256 break;
2257 }
2258
2259 /* Add in the fixed length from the table */
2260
2261 code += PRIV(OP_lengths)[c];
2262
2263 /* In UTF-8 mode, opcodes that are followed by a character may be followed
2264 by a multi-byte character. The length in the table is a minimum, so we have
2265 to arrange to skip the extra bytes. */
2266
2267 #if defined SUPPORT_UTF && !defined COMPILE_PCRE32
2268 if (utf) switch(c)
2269 {
2270 case OP_CHAR:
2271 case OP_CHARI:
2272 case OP_NOT:
2273 case OP_NOTI:
2274 case OP_EXACT:
2275 case OP_EXACTI:
2276 case OP_NOTEXACT:
2277 case OP_NOTEXACTI:
2278 case OP_UPTO:
2279 case OP_UPTOI:
2280 case OP_NOTUPTO:
2281 case OP_NOTUPTOI:
2282 case OP_MINUPTO:
2283 case OP_MINUPTOI:
2284 case OP_NOTMINUPTO:
2285 case OP_NOTMINUPTOI:
2286 case OP_POSUPTO:
2287 case OP_POSUPTOI:
2288 case OP_NOTPOSUPTO:
2289 case OP_NOTPOSUPTOI:
2290 case OP_STAR:
2291 case OP_STARI:
2292 case OP_NOTSTAR:
2293 case OP_NOTSTARI:
2294 case OP_MINSTAR:
2295 case OP_MINSTARI:
2296 case OP_NOTMINSTAR:
2297 case OP_NOTMINSTARI:
2298 case OP_POSSTAR:
2299 case OP_POSSTARI:
2300 case OP_NOTPOSSTAR:
2301 case OP_NOTPOSSTARI:
2302 case OP_PLUS:
2303 case OP_PLUSI:
2304 case OP_NOTPLUS:
2305 case OP_NOTPLUSI:
2306 case OP_MINPLUS:
2307 case OP_MINPLUSI:
2308 case OP_NOTMINPLUS:
2309 case OP_NOTMINPLUSI:
2310 case OP_POSPLUS:
2311 case OP_POSPLUSI:
2312 case OP_NOTPOSPLUS:
2313 case OP_NOTPOSPLUSI:
2314 case OP_QUERY:
2315 case OP_QUERYI:
2316 case OP_NOTQUERY:
2317 case OP_NOTQUERYI:
2318 case OP_MINQUERY:
2319 case OP_MINQUERYI:
2320 case OP_NOTMINQUERY:
2321 case OP_NOTMINQUERYI:
2322 case OP_POSQUERY:
2323 case OP_POSQUERYI:
2324 case OP_NOTPOSQUERY:
2325 case OP_NOTPOSQUERYI:
2326 if (HAS_EXTRALEN(code[-1])) code += GET_EXTRALEN(code[-1]);
2327 break;
2328 }
2329 #else
2330 (void)(utf); /* Keep compiler happy by referencing function argument */
2331 #endif
2332 }
2333 }
2334 }
2335
2336
2337
2338 /*************************************************
2339 * Scan compiled branch for non-emptiness *
2340 *************************************************/
2341
2342 /* This function scans through a branch of a compiled pattern to see whether it
2343 can match the empty string or not. It is called from could_be_empty()
2344 below and from compile_branch() when checking for an unlimited repeat of a
2345 group that can match nothing. Note that first_significant_code() skips over
2346 backward and negative forward assertions when its final argument is TRUE. If we
2347 hit an unclosed bracket, we return "empty" - this means we've struck an inner
2348 bracket whose current branch will already have been scanned.
2349
2350 Arguments:
2351 code points to start of search
2352 endcode points to where to stop
2353 utf TRUE if in UTF-8 / UTF-16 / UTF-32 mode
2354 cd contains pointers to tables etc.
2355
2356 Returns: TRUE if what is matched could be empty
2357 */
2358
2359 static BOOL
2360 could_be_empty_branch(const pcre_uchar *code, const pcre_uchar *endcode,
2361 BOOL utf, compile_data *cd)
2362 {
2363 register pcre_uchar c;
2364 for (code = first_significant_code(code + PRIV(OP_lengths)[*code], TRUE);
2365 code < endcode;
2366 code = first_significant_code(code + PRIV(OP_lengths)[c], TRUE))
2367 {
2368 const pcre_uchar *ccode;
2369
2370 c = *code;
2371
2372 /* Skip over forward assertions; the other assertions are skipped by
2373 first_significant_code() with a TRUE final argument. */
2374
2375 if (c == OP_ASSERT)
2376 {
2377 do code += GET(code, 1); while (*code == OP_ALT);
2378 c = *code;
2379 continue;
2380 }
2381
2382 /* For a recursion/subroutine call, if its end has been reached, which
2383 implies a backward reference subroutine call, we can scan it. If it's a
2384 forward reference subroutine call, we can't. To detect forward reference
2385 we have to scan up the list that is kept in the workspace. This function is
2386 called only when doing the real compile, not during the pre-compile that
2387 measures the size of the compiled pattern. */
2388
2389 if (c == OP_RECURSE)
2390 {
2391 const pcre_uchar *scode;
2392 BOOL empty_branch;
2393
2394 /* Test for forward reference */
2395
2396 for (scode = cd->start_workspace; scode < cd->hwm; scode += LINK_SIZE)
2397 if ((int)GET(scode, 0) == (int)(code + 1 - cd->start_code)) return TRUE;
2398
2399 /* Not a forward reference, test for completed backward reference */
2400
2401 empty_branch = FALSE;
2402 scode = cd->start_code + GET(code, 1);
2403 if (GET(scode, 1) == 0) return TRUE; /* Unclosed */
2404
2405 /* Completed backwards reference */
2406
2407 do
2408 {
2409 if (could_be_empty_branch(scode, endcode, utf, cd))
2410 {
2411 empty_branch = TRUE;
2412 break;
2413 }
2414 scode += GET(scode, 1);
2415 }
2416 while (*scode == OP_ALT);
2417
2418 if (!empty_branch) return FALSE; /* All branches are non-empty */
2419 continue;
2420 }
2421
2422 /* Groups with zero repeats can of course be empty; skip them. */
2423
2424 if (c == OP_BRAZERO || c == OP_BRAMINZERO || c == OP_SKIPZERO ||
2425 c == OP_BRAPOSZERO)
2426 {
2427 code += PRIV(OP_lengths)[c];
2428 do code += GET(code, 1); while (*code == OP_ALT);
2429 c = *code;
2430 continue;
2431 }
2432
2433 /* A nested group that is already marked as "could be empty" can just be
2434 skipped. */
2435
2436 if (c == OP_SBRA || c == OP_SBRAPOS ||
2437 c == OP_SCBRA || c == OP_SCBRAPOS)
2438 {
2439 do code += GET(code, 1); while (*code == OP_ALT);
2440 c = *code;
2441 continue;
2442 }
2443
2444 /* For other groups, scan the branches. */
2445
2446 if (c == OP_BRA || c == OP_BRAPOS ||
2447 c == OP_CBRA || c == OP_CBRAPOS ||
2448 c == OP_ONCE || c == OP_ONCE_NC ||
2449 c == OP_COND)
2450 {
2451 BOOL empty_branch;
2452 if (GET(code, 1) == 0) return TRUE; /* Hit unclosed bracket */
2453
2454 /* If a conditional group has only one branch, there is a second, implied,
2455 empty branch, so just skip over the conditional, because it could be empty.
2456 Otherwise, scan the individual branches of the group. */
2457
2458 if (c == OP_COND && code[GET(code, 1)] != OP_ALT)
2459 code += GET(code, 1);
2460 else
2461 {
2462 empty_branch = FALSE;
2463 do
2464 {
2465 if (!empty_branch && could_be_empty_branch(code, endcode, utf, cd))
2466 empty_branch = TRUE;
2467 code += GET(code, 1);
2468 }
2469 while (*code == OP_ALT);
2470 if (!empty_branch) return FALSE; /* All branches are non-empty */
2471 }
2472
2473 c = *code;
2474 continue;
2475 }
2476
2477 /* Handle the other opcodes */
2478
2479 switch (c)
2480 {
2481 /* Check for quantifiers after a class. XCLASS is used for classes that
2482 cannot be represented just by a bit map. This includes negated single
2483 high-valued characters. The length in PRIV(OP_lengths)[] is zero; the
2484 actual length is stored in the compiled code, so we must update "code"
2485 here. */
2486
2487 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
2488 case OP_XCLASS:
2489 ccode = code += GET(code, 1);
2490 goto CHECK_CLASS_REPEAT;
2491 #endif
2492
2493 case OP_CLASS:
2494 case OP_NCLASS:
2495 ccode = code + PRIV(OP_lengths)[OP_CLASS];
2496
2497 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
2498 CHECK_CLASS_REPEAT:
2499 #endif
2500
2501 switch (*ccode)
2502 {
2503 case OP_CRSTAR: /* These could be empty; continue */
2504 case OP_CRMINSTAR:
2505 case OP_CRQUERY:
2506 case OP_CRMINQUERY:
2507 break;
2508
2509 default: /* Non-repeat => class must match */
2510 case OP_CRPLUS: /* These repeats aren't empty */
2511 case OP_CRMINPLUS:
2512 return FALSE;
2513
2514 case OP_CRRANGE:
2515 case OP_CRMINRANGE:
2516 if (GET2(ccode, 1) > 0) return FALSE; /* Minimum > 0 */
2517 break;
2518 }
2519 break;
2520
2521 /* Opcodes that must match a character */
2522
2523 case OP_PROP:
2524 case OP_NOTPROP:
2525 case OP_EXTUNI:
2526 case OP_NOT_DIGIT:
2527 case OP_DIGIT:
2528 case OP_NOT_WHITESPACE:
2529 case OP_WHITESPACE:
2530 case OP_NOT_WORDCHAR:
2531 case OP_WORDCHAR:
2532 case OP_ANY:
2533 case OP_ALLANY:
2534 case OP_ANYBYTE:
2535 case OP_CHAR:
2536 case OP_CHARI:
2537 case OP_NOT:
2538 case OP_NOTI:
2539 case OP_PLUS:
2540 case OP_MINPLUS:
2541 case OP_POSPLUS:
2542 case OP_EXACT:
2543 case OP_NOTPLUS:
2544 case OP_NOTMINPLUS:
2545 case OP_NOTPOSPLUS:
2546 case OP_NOTEXACT:
2547 case OP_TYPEPLUS:
2548 case OP_TYPEMINPLUS:
2549 case OP_TYPEPOSPLUS:
2550 case OP_TYPEEXACT:
2551 return FALSE;
2552
2553 /* These are going to continue, as they may be empty, but we have to
2554 fudge the length for the \p and \P cases. */
2555
2556 case OP_TYPESTAR:
2557 case OP_TYPEMINSTAR:
2558 case OP_TYPEPOSSTAR:
2559 case OP_TYPEQUERY:
2560 case OP_TYPEMINQUERY:
2561 case OP_TYPEPOSQUERY:
2562 if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
2563 break;
2564
2565 /* Same for these */
2566
2567 case OP_TYPEUPTO:
2568 case OP_TYPEMINUPTO:
2569 case OP_TYPEPOSUPTO:
2570 if (code[1 + IMM2_SIZE] == OP_PROP || code[1 + IMM2_SIZE] == OP_NOTPROP)
2571 code += 2;
2572 break;
2573
2574 /* End of branch */
2575
2576 case OP_KET:
2577 case OP_KETRMAX:
2578 case OP_KETRMIN:
2579 case OP_KETRPOS:
2580 case OP_ALT:
2581 return TRUE;
2582
2583 /* In UTF-8 mode, STAR, MINSTAR, POSSTAR, QUERY, MINQUERY, POSQUERY, UPTO,
2584 MINUPTO, and POSUPTO may be followed by a multibyte character */
2585
2586 #if defined SUPPORT_UTF && !defined COMPILE_PCRE32
2587 case OP_STAR:
2588 case OP_STARI:
2589 case OP_MINSTAR:
2590 case OP_MINSTARI:
2591 case OP_POSSTAR:
2592 case OP_POSSTARI:
2593 case OP_QUERY:
2594 case OP_QUERYI:
2595 case OP_MINQUERY:
2596 case OP_MINQUERYI:
2597 case OP_POSQUERY:
2598 case OP_POSQUERYI:
2599 if (utf && HAS_EXTRALEN(code[1])) code += GET_EXTRALEN(code[1]);
2600 break;
2601
2602 case OP_UPTO:
2603 case OP_UPTOI:
2604 case OP_MINUPTO:
2605 case OP_MINUPTOI:
2606 case OP_POSUPTO:
2607 case OP_POSUPTOI:
2608 if (utf && HAS_EXTRALEN(code[1 + IMM2_SIZE])) code += GET_EXTRALEN(code[1 + IMM2_SIZE]);
2609 break;
2610 #endif
2611
2612 /* MARK, and PRUNE/SKIP/THEN with an argument must skip over the argument
2613 string. */
2614
2615 case OP_MARK:
2616 case OP_PRUNE_ARG:
2617 case OP_SKIP_ARG:
2618 case OP_THEN_ARG:
2619 code += code[1];
2620 break;
2621
2622 /* None of the remaining opcodes are required to match a character. */
2623
2624 default:
2625 break;
2626 }
2627 }
2628
2629 return TRUE;
2630 }
2631
2632
2633
2634 /*************************************************
2635 * Scan compiled regex for non-emptiness *
2636 *************************************************/
2637
2638 /* This function is called to check for left recursive calls. We want to check
2639 the current branch of the current pattern to see if it could match the empty
2640 string. If it could, we must look outwards for branches at other levels,
2641 stopping when we pass beyond the bracket which is the subject of the recursion.
2642 This function is called only during the real compile, not during the
2643 pre-compile.
2644
2645 Arguments:
2646 code points to start of the recursion
2647 endcode points to where to stop (current RECURSE item)
2648 bcptr points to the chain of current (unclosed) branch starts
2649 utf TRUE if in UTF-8 / UTF-16 / UTF-32 mode
2650 cd pointers to tables etc
2651
2652 Returns: TRUE if what is matched could be empty
2653 */
2654
2655 static BOOL
2656 could_be_empty(const pcre_uchar *code, const pcre_uchar *endcode,
2657 branch_chain *bcptr, BOOL utf, compile_data *cd)
2658 {
2659 while (bcptr != NULL && bcptr->current_branch >= code)
2660 {
2661 if (!could_be_empty_branch(bcptr->current_branch, endcode, utf, cd))
2662 return FALSE;
2663 bcptr = bcptr->outer;
2664 }
2665 return TRUE;
2666 }
2667
2668
2669
2670 /*************************************************
2671 * Check for POSIX class syntax *
2672 *************************************************/
2673
2674 /* This function is called when the sequence "[:" or "[." or "[=" is
2675 encountered in a character class. It checks whether this is followed by a
2676 sequence of characters terminated by a matching ":]" or ".]" or "=]". If we
2677 reach an unescaped ']' without the special preceding character, return FALSE.
2678
2679 Originally, this function only recognized a sequence of letters between the
2680 terminators, but it seems that Perl recognizes any sequence of characters,
2681 though of course unknown POSIX names are subsequently rejected. Perl gives an
2682 "Unknown POSIX class" error for [:f\oo:] for example, where previously PCRE
2683 didn't consider this to be a POSIX class. Likewise for [:1234:].
2684
2685 The problem in trying to be exactly like Perl is in the handling of escapes. We
2686 have to be sure that [abc[:x\]pqr] is *not* treated as containing a POSIX
2687 class, but [abc[:x\]pqr:]] is (so that an error can be generated). The code
2688 below handles the special case of \], but does not try to do any other escape
2689 processing. This makes it different from Perl for cases such as [:l\ower:]
2690 where Perl recognizes it as the POSIX class "lower" but PCRE does not recognize
2691 "l\ower". This is a lesser evil that not diagnosing bad classes when Perl does,
2692 I think.
2693
2694 A user pointed out that PCRE was rejecting [:a[:digit:]] whereas Perl was not.
2695 It seems that the appearance of a nested POSIX class supersedes an apparent
2696 external class. For example, [:a[:digit:]b:] matches "a", "b", ":", or
2697 a digit.
2698
2699 In Perl, unescaped square brackets may also appear as part of class names. For
2700 example, [:a[:abc]b:] gives unknown POSIX class "[:abc]b:]". However, for
2701 [:a[:abc]b][b:] it gives unknown POSIX class "[:abc]b][b:]", which does not
2702 seem right at all. PCRE does not allow closing square brackets in POSIX class
2703 names.
2704
2705 Arguments:
2706 ptr pointer to the initial [
2707 endptr where to return the end pointer
2708
2709 Returns: TRUE or FALSE
2710 */
2711
2712 static BOOL
2713 check_posix_syntax(const pcre_uchar *ptr, const pcre_uchar **endptr)
2714 {
2715 pcre_uchar terminator; /* Don't combine these lines; the Solaris cc */
2716 terminator = *(++ptr); /* compiler warns about "non-constant" initializer. */
2717 for (++ptr; *ptr != CHAR_NULL; ptr++)
2718 {
2719 if (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET)
2720 ptr++;
2721 else if (*ptr == CHAR_RIGHT_SQUARE_BRACKET) return FALSE;
2722 else
2723 {
2724 if (*ptr == terminator && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET)
2725 {
2726 *endptr = ptr;
2727 return TRUE;
2728 }
2729 if (*ptr == CHAR_LEFT_SQUARE_BRACKET &&
2730 (ptr[1] == CHAR_COLON || ptr[1] == CHAR_DOT ||
2731 ptr[1] == CHAR_EQUALS_SIGN) &&
2732 check_posix_syntax(ptr, endptr))
2733 return FALSE;
2734 }
2735 }
2736 return FALSE;
2737 }
2738
2739
2740
2741
2742 /*************************************************
2743 * Check POSIX class name *
2744 *************************************************/
2745
2746 /* This function is called to check the name given in a POSIX-style class entry
2747 such as [:alnum:].
2748
2749 Arguments:
2750 ptr points to the first letter
2751 len the length of the name
2752
2753 Returns: a value representing the name, or -1 if unknown
2754 */
2755
2756 static int
2757 check_posix_name(const pcre_uchar *ptr, int len)
2758 {
2759 const char *pn = posix_names;
2760 register int yield = 0;
2761 while (posix_name_lengths[yield] != 0)
2762 {
2763 if (len == posix_name_lengths[yield] &&
2764 STRNCMP_UC_C8(ptr, pn, (unsigned int)len) == 0) return yield;
2765 pn += posix_name_lengths[yield] + 1;
2766 yield++;
2767 }
2768 return -1;
2769 }
2770
2771
2772 /*************************************************
2773 * Adjust OP_RECURSE items in repeated group *
2774 *************************************************/
2775
2776 /* OP_RECURSE items contain an offset from the start of the regex to the group
2777 that is referenced. This means that groups can be replicated for fixed
2778 repetition simply by copying (because the recursion is allowed to refer to
2779 earlier groups that are outside the current group). However, when a group is
2780 optional (i.e. the minimum quantifier is zero), OP_BRAZERO or OP_SKIPZERO is
2781 inserted before it, after it has been compiled. This means that any OP_RECURSE
2782 items within it that refer to the group itself or any contained groups have to
2783 have their offsets adjusted. That one of the jobs of this function. Before it
2784 is called, the partially compiled regex must be temporarily terminated with
2785 OP_END.
2786
2787 This function has been extended with the possibility of forward references for
2788 recursions and subroutine calls. It must also check the list of such references
2789 for the group we are dealing with. If it finds that one of the recursions in
2790 the current group is on this list, it adjusts the offset in the list, not the
2791 value in the reference (which is a group number).
2792
2793 Arguments:
2794 group points to the start of the group
2795 adjust the amount by which the group is to be moved
2796 utf TRUE in UTF-8 / UTF-16 / UTF-32 mode
2797 cd contains pointers to tables etc.
2798 save_hwm the hwm forward reference pointer at the start of the group
2799
2800 Returns: nothing
2801 */
2802
2803 static void
2804 adjust_recurse(pcre_uchar *group, int adjust, BOOL utf, compile_data *cd,
2805 pcre_uchar *save_hwm)
2806 {
2807 pcre_uchar *ptr = group;
2808
2809 while ((ptr = (pcre_uchar *)find_recurse(ptr, utf)) != NULL)
2810 {
2811 int offset;
2812 pcre_uchar *hc;
2813
2814 /* See if this recursion is on the forward reference list. If so, adjust the
2815 reference. */
2816
2817 for (hc = save_hwm; hc < cd->hwm; hc += LINK_SIZE)
2818 {
2819 offset = (int)GET(hc, 0);
2820 if (cd->start_code + offset == ptr + 1)
2821 {
2822 PUT(hc, 0, offset + adjust);
2823 break;
2824 }
2825 }
2826
2827 /* Otherwise, adjust the recursion offset if it's after the start of this
2828 group. */
2829
2830 if (hc >= cd->hwm)
2831 {
2832 offset = (int)GET(ptr, 1);
2833 if (cd->start_code + offset >= group) PUT(ptr, 1, offset + adjust);
2834 }
2835
2836 ptr += 1 + LINK_SIZE;
2837 }
2838 }
2839
2840
2841
2842 /*************************************************
2843 * Insert an automatic callout point *
2844 *************************************************/
2845
2846 /* This function is called when the PCRE_AUTO_CALLOUT option is set, to insert
2847 callout points before each pattern item.
2848
2849 Arguments:
2850 code current code pointer
2851 ptr current pattern pointer
2852 cd pointers to tables etc
2853
2854 Returns: new code pointer
2855 */
2856
2857 static pcre_uchar *
2858 auto_callout(pcre_uchar *code, const pcre_uchar *ptr, compile_data *cd)
2859 {
2860 *code++ = OP_CALLOUT;
2861 *code++ = 255;
2862 PUT(code, 0, (int)(ptr - cd->start_pattern)); /* Pattern offset */
2863 PUT(code, LINK_SIZE, 0); /* Default length */
2864 return code + 2 * LINK_SIZE;
2865 }
2866
2867
2868
2869 /*************************************************
2870 * Complete a callout item *
2871 *************************************************/
2872
2873 /* A callout item contains the length of the next item in the pattern, which
2874 we can't fill in till after we have reached the relevant point. This is used
2875 for both automatic and manual callouts.
2876
2877 Arguments:
2878 previous_callout points to previous callout item
2879 ptr current pattern pointer
2880 cd pointers to tables etc
2881
2882 Returns: nothing
2883 */
2884
2885 static void
2886 complete_callout(pcre_uchar *previous_callout, const pcre_uchar *ptr, compile_data *cd)
2887 {
2888 int length = (int)(ptr - cd->start_pattern - GET(previous_callout, 2));
2889 PUT(previous_callout, 2 + LINK_SIZE, length);
2890 }
2891
2892
2893
2894 #ifdef SUPPORT_UCP
2895 /*************************************************
2896 * Get othercase range *
2897 *************************************************/
2898
2899 /* This function is passed the start and end of a class range, in UTF-8 mode
2900 with UCP support. It searches up the characters, looking for ranges of
2901 characters in the "other" case. Each call returns the next one, updating the
2902 start address. A character with multiple other cases is returned on its own
2903 with a special return value.
2904
2905 Arguments:
2906 cptr points to starting character value; updated
2907 d end value
2908 ocptr where to put start of othercase range
2909 odptr where to put end of othercase range
2910
2911 Yield: -1 when no more
2912 0 when a range is returned
2913 >0 the CASESET offset for char with multiple other cases
2914 in this case, ocptr contains the original
2915 */
2916
2917 static int
2918 get_othercase_range(pcre_uint32 *cptr, pcre_uint32 d, pcre_uint32 *ocptr,
2919 pcre_uint32 *odptr)
2920 {
2921 pcre_uint32 c, othercase, next;
2922 unsigned int co;
2923
2924 /* Find the first character that has an other case. If it has multiple other
2925 cases, return its case offset value. */
2926
2927 for (c = *cptr; c <= d; c++)
2928 {
2929 if ((co = UCD_CASESET(c)) != 0)
2930 {
2931 *ocptr = c++; /* Character that has the set */
2932 *cptr = c; /* Rest of input range */
2933 return (int)co;
2934 }
2935 if ((othercase = UCD_OTHERCASE(c)) != c) break;
2936 }
2937
2938 if (c > d) return -1; /* Reached end of range */
2939
2940 *ocptr = othercase;
2941 next = othercase + 1;
2942
2943 for (++c; c <= d; c++)
2944 {
2945 if (UCD_OTHERCASE(c) != next) break;
2946 next++;
2947 }
2948
2949 *odptr = next - 1; /* End of othercase range */
2950 *cptr = c; /* Rest of input range */
2951 return 0;
2952 }
2953
2954
2955
2956 /*************************************************
2957 * Check a character and a property *
2958 *************************************************/
2959
2960 /* This function is called by check_auto_possessive() when a property item
2961 is adjacent to a fixed character.
2962
2963 Arguments:
2964 c the character
2965 ptype the property type
2966 pdata the data for the type
2967 negated TRUE if it's a negated property (\P or \p{^)
2968
2969 Returns: TRUE if auto-possessifying is OK
2970 */
2971
2972 static BOOL
2973 check_char_prop(pcre_uint32 c, unsigned int ptype, unsigned int pdata, BOOL negated)
2974 {
2975 #ifdef SUPPORT_UCP
2976 const pcre_uint32 *p;
2977 #endif
2978
2979 const ucd_record *prop = GET_UCD(c);
2980
2981 switch(ptype)
2982 {
2983 case PT_LAMP:
2984 return (prop->chartype == ucp_Lu ||
2985 prop->chartype == ucp_Ll ||
2986 prop->chartype == ucp_Lt) == negated;
2987
2988 case PT_GC:
2989 return (pdata == PRIV(ucp_gentype)[prop->chartype]) == negated;
2990
2991 case PT_PC:
2992 return (pdata == prop->chartype) == negated;
2993
2994 case PT_SC:
2995 return (pdata == prop->script) == negated;
2996
2997 /* These are specials */
2998
2999 case PT_ALNUM:
3000 return (PRIV(ucp_gentype)[prop->chartype] == ucp_L ||
3001 PRIV(ucp_gentype)[prop->chartype] == ucp_N) == negated;
3002
3003 case PT_SPACE: /* Perl space */
3004 return (PRIV(ucp_gentype)[prop->chartype] == ucp_Z ||
3005 c == CHAR_HT || c == CHAR_NL || c == CHAR_FF || c == CHAR_CR)
3006 == negated;
3007
3008 case PT_PXSPACE: /* POSIX space */
3009 return (PRIV(ucp_gentype)[prop->chartype] == ucp_Z ||
3010 c == CHAR_HT || c == CHAR_NL || c == CHAR_VT ||
3011 c == CHAR_FF || c == CHAR_CR)
3012 == negated;
3013
3014 case PT_WORD:
3015 return (PRIV(ucp_gentype)[prop->chartype] == ucp_L ||
3016 PRIV(ucp_gentype)[prop->chartype] == ucp_N ||
3017 c == CHAR_UNDERSCORE) == negated;
3018
3019 #ifdef SUPPORT_UCP
3020 case PT_CLIST:
3021 p = PRIV(ucd_caseless_sets) + prop->caseset;
3022 for (;;)
3023 {
3024 if (c < *p) return !negated;
3025 if (c == *p++) return negated;
3026 }
3027 break; /* Control never reaches here */
3028 #endif
3029 }
3030
3031 return FALSE;
3032 }
3033 #endif /* SUPPORT_UCP */
3034
3035
3036
3037 /*************************************************
3038 * Check if auto-possessifying is possible *
3039 *************************************************/
3040
3041 /* This function is called for unlimited repeats of certain items, to see
3042 whether the next thing could possibly match the repeated item. If not, it makes
3043 sense to automatically possessify the repeated item.
3044
3045 Arguments:
3046 previous pointer to the repeated opcode
3047 utf TRUE in UTF-8 / UTF-16 / UTF-32 mode
3048 ptr next character in pattern
3049 options options bits
3050 cd contains pointers to tables etc.
3051
3052 Returns: TRUE if possessifying is wanted
3053 */
3054
3055 static BOOL
3056 check_auto_possessive(const pcre_uchar *previous, BOOL utf,
3057 const pcre_uchar *ptr, int options, compile_data *cd)
3058 {
3059 pcre_uint32 c = NOTACHAR;
3060 pcre_uint32 next;
3061 int escape;
3062 pcre_uchar op_code = *previous++;
3063
3064 /* Skip whitespace and comments in extended mode */
3065
3066 if ((options & PCRE_EXTENDED) != 0)
3067 {
3068 for (;;)
3069 {
3070 while (MAX_255(*ptr) && (cd->ctypes[*ptr] & ctype_space) != 0) ptr++;
3071 if (*ptr == CHAR_NUMBER_SIGN)
3072 {
3073 ptr++;
3074 while (*ptr != CHAR_NULL)
3075 {
3076 if (IS_NEWLINE(ptr)) { ptr += cd->nllen; break; }
3077 ptr++;
3078 #ifdef SUPPORT_UTF
3079 if (utf) FORWARDCHAR(ptr);
3080 #endif
3081 }
3082 }
3083 else break;
3084 }
3085 }
3086
3087 /* If the next item is one that we can handle, get its value. A non-negative
3088 value is a character, a negative value is an escape value. */
3089
3090 if (*ptr == CHAR_BACKSLASH)
3091 {
3092 int temperrorcode = 0;
3093 escape = check_escape(&ptr, &next, &temperrorcode, cd->bracount, options,
3094 FALSE);
3095 if (temperrorcode != 0) return FALSE;
3096 ptr++; /* Point after the escape sequence */
3097 }
3098 else if (!MAX_255(*ptr) || (cd->ctypes[*ptr] & ctype_meta) == 0)
3099 {
3100 escape = 0;
3101 #ifdef SUPPORT_UTF
3102 if (utf) { GETCHARINC(next, ptr); } else
3103 #endif
3104 next = *ptr++;
3105 }
3106 else return FALSE;
3107
3108 /* Skip whitespace and comments in extended mode */
3109
3110 if ((options & PCRE_EXTENDED) != 0)
3111 {
3112 for (;;)
3113 {
3114 while (MAX_255(*ptr) && (cd->ctypes[*ptr] & ctype_space) != 0) ptr++;
3115 if (*ptr == CHAR_NUMBER_SIGN)
3116 {
3117 ptr++;
3118 while (*ptr != CHAR_NULL)
3119 {
3120 if (IS_NEWLINE(ptr)) { ptr += cd->nllen; break; }
3121 ptr++;
3122 #ifdef SUPPORT_UTF
3123 if (utf) FORWARDCHAR(ptr);
3124 #endif
3125 }
3126 }
3127 else break;
3128 }
3129 }
3130
3131 /* If the next thing is itself optional, we have to give up. */
3132
3133 if (*ptr == CHAR_ASTERISK || *ptr == CHAR_QUESTION_MARK ||
3134 STRNCMP_UC_C8(ptr, STR_LEFT_CURLY_BRACKET STR_0 STR_COMMA, 3) == 0)
3135 return FALSE;
3136
3137 /* If the previous item is a character, get its value. */
3138
3139 if (op_code == OP_CHAR || op_code == OP_CHARI ||
3140 op_code == OP_NOT || op_code == OP_NOTI)
3141 {
3142 #ifdef SUPPORT_UTF
3143 GETCHARTEST(c, previous);
3144 #else
3145 c = *previous;
3146 #endif
3147 }
3148
3149 /* Now compare the next item with the previous opcode. First, handle cases when
3150 the next item is a character. */
3151
3152 if (escape == 0)
3153 {
3154 /* For a caseless UTF match, the next character may have more than one other
3155 case, which maps to the special PT_CLIST property. Check this first. */
3156
3157 #ifdef SUPPORT_UCP
3158 if (utf && c != NOTACHAR && (options & PCRE_CASELESS) != 0)
3159 {
3160 unsigned int ocs = UCD_CASESET(next);
3161 if (ocs > 0) return check_char_prop(c, PT_CLIST, ocs, op_code >= OP_NOT);
3162 }
3163 #endif
3164
3165 switch(op_code)
3166 {
3167 case OP_CHAR:
3168 return c != next;
3169
3170 /* For CHARI (caseless character) we must check the other case. If we have
3171 Unicode property support, we can use it to test the other case of
3172 high-valued characters. We know that next can have only one other case,
3173 because multi-other-case characters are dealt with above. */
3174
3175 case OP_CHARI:
3176 if (c == next) return FALSE;
3177 #ifdef SUPPORT_UTF
3178 if (utf)
3179 {
3180 pcre_uint32 othercase;
3181 if (next < 128) othercase = cd->fcc[next]; else
3182 #ifdef SUPPORT_UCP
3183 othercase = UCD_OTHERCASE(next);
3184 #else
3185 othercase = NOTACHAR;
3186 #endif
3187 return c != othercase;
3188 }
3189 else
3190 #endif /* SUPPORT_UTF */
3191 return (c != TABLE_GET(next, cd->fcc, next)); /* Not UTF */
3192
3193 case OP_NOT:
3194 return c == next;
3195
3196 case OP_NOTI:
3197 if (c == next) return TRUE;
3198 #ifdef SUPPORT_UTF
3199 if (utf)
3200 {
3201 pcre_uint32 othercase;
3202 if (next < 128) othercase = cd->fcc[next]; else
3203 #ifdef SUPPORT_UCP
3204 othercase = UCD_OTHERCASE(next);
3205 #else
3206 othercase = NOTACHAR;
3207 #endif
3208 return c == othercase;
3209 }
3210 else
3211 #endif /* SUPPORT_UTF */
3212 return (c == TABLE_GET(next, cd->fcc, next)); /* Not UTF */
3213
3214 /* Note that OP_DIGIT etc. are generated only when PCRE_UCP is *not* set.
3215 When it is set, \d etc. are converted into OP_(NOT_)PROP codes. */
3216
3217 case OP_DIGIT:
3218 return next > 255 || (cd->ctypes[next] & ctype_digit) == 0;
3219
3220 case OP_NOT_DIGIT:
3221 return next <= 255 && (cd->ctypes[next] & ctype_digit) != 0;
3222
3223 case OP_WHITESPACE:
3224 return next > 255 || (cd->ctypes[next] & ctype_space) == 0;
3225
3226 case OP_NOT_WHITESPACE:
3227 return next <= 255 && (cd->ctypes[next] & ctype_space) != 0;
3228
3229 case OP_WORDCHAR:
3230 return next > 255 || (cd->ctypes[next] & ctype_word) == 0;
3231
3232 case OP_NOT_WORDCHAR:
3233 return next <= 255 && (cd->ctypes[next] & ctype_word) != 0;
3234
3235 case OP_HSPACE:
3236 case OP_NOT_HSPACE:
3237 switch(next)
3238 {
3239 HSPACE_CASES:
3240 return op_code == OP_NOT_HSPACE;
3241
3242 default:
3243 return op_code != OP_NOT_HSPACE;
3244 }
3245
3246 case OP_ANYNL:
3247 case OP_VSPACE:
3248 case OP_NOT_VSPACE:
3249 switch(next)
3250 {
3251 VSPACE_CASES:
3252 return op_code == OP_NOT_VSPACE;
3253
3254 default:
3255 return op_code != OP_NOT_VSPACE;
3256 }
3257
3258 #ifdef SUPPORT_UCP
3259 case OP_PROP:
3260 return check_char_prop(next, previous[0], previous[1], FALSE);
3261
3262 case OP_NOTPROP:
3263 return check_char_prop(next, previous[0], previous[1], TRUE);
3264 #endif
3265
3266 default:
3267 return FALSE;
3268 }
3269 }
3270
3271 /* Handle the case when the next item is \d, \s, etc. Note that when PCRE_UCP
3272 is set, \d turns into ESC_du rather than ESC_d, etc., so ESC_d etc. are
3273 generated only when PCRE_UCP is *not* set, that is, when only ASCII
3274 characteristics are recognized. Similarly, the opcodes OP_DIGIT etc. are
3275 replaced by OP_PROP codes when PCRE_UCP is set. */
3276
3277 switch(op_code)
3278 {
3279 case OP_CHAR:
3280 case OP_CHARI:
3281 switch(escape)
3282 {
3283 case ESC_d:
3284 return c > 255 || (cd->ctypes[c] & ctype_digit) == 0;
3285
3286 case ESC_D:
3287 return c <= 255 && (cd->ctypes[c] & ctype_digit) != 0;
3288
3289 case ESC_s:
3290 return c > 255 || (cd->ctypes[c] & ctype_space) == 0;
3291
3292 case ESC_S:
3293 return c <= 255 && (cd->ctypes[c] & ctype_space) != 0;
3294
3295 case ESC_w:
3296 return c > 255 || (cd->ctypes[c] & ctype_word) == 0;
3297
3298 case ESC_W:
3299 return c <= 255 && (cd->ctypes[c] & ctype_word) != 0;
3300
3301 case ESC_h:
3302 case ESC_H:
3303 switch(c)
3304 {
3305 HSPACE_CASES:
3306 return escape != ESC_h;
3307
3308 default:
3309 return escape == ESC_h;
3310 }
3311
3312 case ESC_v:
3313 case ESC_V:
3314 switch(c)
3315 {
3316 VSPACE_CASES:
3317 return escape != ESC_v;
3318
3319 default:
3320 return escape == ESC_v;
3321 }
3322
3323 /* When PCRE_UCP is set, these values get generated for \d etc. Find
3324 their substitutions and process them. The result will always be either
3325 ESC_p or ESC_P. Then fall through to process those values. */
3326
3327 #ifdef SUPPORT_UCP
3328 case ESC_du:
3329 case ESC_DU:
3330 case ESC_wu:
3331 case ESC_WU:
3332 case ESC_su:
3333 case ESC_SU:
3334 {
3335 int temperrorcode = 0;
3336 ptr = substitutes[escape - ESC_DU];
3337 escape = check_escape(&ptr, &next, &temperrorcode, 0, options, FALSE);
3338 if (temperrorcode != 0) return FALSE;
3339 ptr++; /* For compatibility */
3340 }
3341 /* Fall through */
3342
3343 case ESC_p:
3344 case ESC_P:
3345 {
3346 unsigned int ptype = 0, pdata = 0;
3347 int errorcodeptr;
3348 BOOL negated;
3349
3350 ptr--; /* Make ptr point at the p or P */
3351 if (!get_ucp(&ptr, &negated, &ptype, &pdata, &errorcodeptr))
3352 return FALSE;
3353 ptr++; /* Point past the final curly ket */
3354
3355 /* If the property item is optional, we have to give up. (When generated
3356 from \d etc by PCRE_UCP, this test will have been applied much earlier,
3357 to the original \d etc. At this point, ptr will point to a zero byte. */
3358
3359 if (*ptr == CHAR_ASTERISK || *ptr == CHAR_QUESTION_MARK ||
3360 STRNCMP_UC_C8(ptr, STR_LEFT_CURLY_BRACKET STR_0 STR_COMMA, 3) == 0)
3361 return FALSE;
3362
3363 /* Do the property check. */
3364
3365 return check_char_prop(c, ptype, pdata, (escape == ESC_P) != negated);
3366 }
3367 #endif
3368
3369 default:
3370 return FALSE;
3371 }
3372
3373 /* In principle, support for Unicode properties should be integrated here as
3374 well. It means re-organizing the above code so as to get hold of the property
3375 values before switching on the op-code. However, I wonder how many patterns
3376 combine ASCII \d etc with Unicode properties? (Note that if PCRE_UCP is set,
3377 these op-codes are never generated.) */
3378
3379 case OP_DIGIT:
3380 return escape == ESC_D || escape == ESC_s || escape == ESC_W ||
3381 escape == ESC_h || escape == ESC_v || escape == ESC_R;
3382
3383 case OP_NOT_DIGIT:
3384 return escape == ESC_d;
3385
3386 case OP_WHITESPACE:
3387 return escape == ESC_S || escape == ESC_d || escape == ESC_w;
3388
3389 case OP_NOT_WHITESPACE:
3390 return escape == ESC_s || escape == ESC_h || escape == ESC_v || escape == ESC_R;
3391
3392 case OP_HSPACE:
3393 return escape == ESC_S || escape == ESC_H || escape == ESC_d ||
3394 escape == ESC_w || escape == ESC_v || escape == ESC_R;
3395
3396 case OP_NOT_HSPACE:
3397 return escape == ESC_h;
3398
3399 /* Can't have \S in here because VT matches \S (Perl anomaly) */
3400 case OP_ANYNL:
3401 case OP_VSPACE:
3402 return escape == ESC_V || escape == ESC_d || escape == ESC_w;
3403
3404 case OP_NOT_VSPACE:
3405 return escape == ESC_v || escape == ESC_R;
3406
3407 case OP_WORDCHAR:
3408 return escape == ESC_W || escape == ESC_s || escape == ESC_h ||
3409 escape == ESC_v || escape == ESC_R;
3410
3411 case OP_NOT_WORDCHAR:
3412 return escape == ESC_w || escape == ESC_d;
3413
3414 default:
3415 return FALSE;
3416 }
3417
3418 /* Control does not reach here */
3419 }
3420
3421
3422
3423 /*************************************************
3424 * Add a character or range to a class *
3425 *************************************************/
3426
3427 /* This function packages up the logic of adding a character or range of
3428 characters to a class. The character values in the arguments will be within the
3429 valid values for the current mode (8-bit, 16-bit, UTF, etc). This function is
3430 mutually recursive with the function immediately below.
3431
3432 Arguments:
3433 classbits the bit map for characters < 256
3434 uchardptr points to the pointer for extra data
3435 options the options word
3436 cd contains pointers to tables etc.
3437 start start of range character
3438 end end of range character
3439
3440 Returns: the number of < 256 characters added
3441 the pointer to extra data is updated
3442 */
3443
3444 static int
3445 add_to_class(pcre_uint8 *classbits, pcre_uchar **uchardptr, int options,
3446 compile_data *cd, pcre_uint32 start, pcre_uint32 end)
3447 {
3448 pcre_uint32 c;
3449 int n8 = 0;
3450
3451 /* If caseless matching is required, scan the range and process alternate
3452 cases. In Unicode, there are 8-bit characters that have alternate cases that
3453 are greater than 255 and vice-versa. Sometimes we can just extend the original
3454 range. */
3455
3456 if ((options & PCRE_CASELESS) != 0)
3457 {
3458 #ifdef SUPPORT_UCP
3459 if ((options & PCRE_UTF8) != 0)
3460 {
3461 int rc;
3462 pcre_uint32 oc, od;
3463
3464 options &= ~PCRE_CASELESS; /* Remove for recursive calls */
3465 c = start;
3466
3467 while ((rc = get_othercase_range(&c, end, &oc, &od)) >= 0)
3468 {
3469 /* Handle a single character that has more than one other case. */
3470
3471 if (rc > 0) n8 += add_list_to_class(classbits, uchardptr, options, cd,
3472 PRIV(ucd_caseless_sets) + rc, oc);
3473
3474 /* Do nothing if the other case range is within the original range. */
3475
3476 else if (oc >= start && od <= end) continue;
3477
3478 /* Extend the original range if there is overlap, noting that if oc < c, we
3479 can't have od > end because a subrange is always shorter than the basic
3480 range. Otherwise, use a recursive call to add the additional range. */
3481
3482 else if (oc < start && od >= start - 1) start = oc; /* Extend downwards */
3483 else if (od > end && oc <= end + 1) end = od; /* Extend upwards */
3484 else n8 += add_to_class(classbits, uchardptr, options, cd, oc, od);
3485 }
3486 }
3487 else
3488 #endif /* SUPPORT_UCP */
3489
3490 /* Not UTF-mode, or no UCP */
3491
3492 for (c = start; c <= end && c < 256; c++)
3493 {
3494 SETBIT(classbits, cd->fcc[c]);
3495 n8++;
3496 }
3497 }
3498
3499 /* Now handle the original range. Adjust the final value according to the bit
3500 length - this means that the same lists of (e.g.) horizontal spaces can be used
3501 in all cases. */
3502
3503 #if defined COMPILE_PCRE8
3504 #ifdef SUPPORT_UTF
3505 if ((options & PCRE_UTF8) == 0)
3506 #endif
3507 if (end > 0xff) end = 0xff;
3508
3509 #elif defined COMPILE_PCRE16
3510 #ifdef SUPPORT_UTF
3511 if ((options & PCRE_UTF16) == 0)
3512 #endif
3513 if (end > 0xffff) end = 0xffff;
3514
3515 #endif /* COMPILE_PCRE[8|16] */
3516
3517 /* If all characters are less than 256, use the bit map. Otherwise use extra
3518 data. */
3519
3520 if (end < 0x100)
3521 {
3522 for (c = start; c <= end; c++)
3523 {
3524 n8++;
3525 SETBIT(classbits, c);
3526 }
3527 }
3528
3529 else
3530 {
3531 pcre_uchar *uchardata = *uchardptr;
3532
3533 #ifdef SUPPORT_UTF
3534 if ((options & PCRE_UTF8) != 0) /* All UTFs use the same flag bit */
3535 {
3536 if (start < end)
3537 {
3538 *uchardata++ = XCL_RANGE;
3539 uchardata += PRIV(ord2utf)(start, uchardata);
3540 uchardata += PRIV(ord2utf)(end, uchardata);
3541 }
3542 else if (start == end)
3543 {
3544 *uchardata++ = XCL_SINGLE;
3545 uchardata += PRIV(ord2utf)(start, uchardata);
3546 }
3547 }
3548 else
3549 #endif /* SUPPORT_UTF */
3550
3551 /* Without UTF support, character values are constrained by the bit length,
3552 and can only be > 256 for 16-bit and 32-bit libraries. */
3553
3554 #ifdef COMPILE_PCRE8
3555 {}
3556 #else
3557 if (start < end)
3558 {
3559 *uchardata++ = XCL_RANGE;
3560 *uchardata++ = start;
3561 *uchardata++ = end;
3562 }
3563 else if (start == end)
3564 {
3565 *uchardata++ = XCL_SINGLE;
3566 *uchardata++ = start;
3567 }
3568 #endif
3569
3570 *uchardptr = uchardata; /* Updata extra data pointer */
3571 }
3572
3573 return n8; /* Number of 8-bit characters */
3574 }
3575
3576
3577
3578
3579 /*************************************************
3580 * Add a list of characters to a class *
3581 *************************************************/
3582
3583 /* This function is used for adding a list of case-equivalent characters to a
3584 class, and also for adding a list of horizontal or vertical whitespace. If the
3585 list is in order (which it should be), ranges of characters are detected and
3586 handled appropriately. This function is mutually recursive with the function
3587 above.
3588
3589 Arguments:
3590 classbits the bit map for characters < 256
3591 uchardptr points to the pointer for extra data
3592 options the options word
3593 cd contains pointers to tables etc.
3594 p points to row of 32-bit values, terminated by NOTACHAR
3595 except character to omit; this is used when adding lists of
3596 case-equivalent characters to avoid including the one we
3597 already know about
3598
3599 Returns: the number of < 256 characters added
3600 the pointer to extra data is updated
3601 */
3602
3603 static int
3604 add_list_to_class(pcre_uint8 *classbits, pcre_uchar **uchardptr, int options,
3605 compile_data *cd, const pcre_uint32 *p, unsigned int except)
3606 {
3607 int n8 = 0;
3608 while (p[0] < NOTACHAR)
3609 {
3610 int n = 0;
3611 if (p[0] != except)
3612 {
3613 while(p[n+1] == p[0] + n + 1) n++;
3614 n8 += add_to_class(classbits, uchardptr, options, cd, p[0], p[n]);
3615 }
3616 p += n + 1;
3617 }
3618 return n8;
3619 }
3620
3621
3622
3623 /*************************************************
3624 * Add characters not in a list to a class *
3625 *************************************************/
3626
3627 /* This function is used for adding the complement of a list of horizontal or
3628 vertical whitespace to a class. The list must be in order.
3629
3630 Arguments:
3631 classbits the bit map for characters < 256
3632 uchardptr points to the pointer for extra data
3633 options the options word
3634 cd contains pointers to tables etc.
3635 p points to row of 32-bit values, terminated by NOTACHAR
3636
3637 Returns: the number of < 256 characters added
3638 the pointer to extra data is updated
3639 */
3640
3641 static int
3642 add_not_list_to_class(pcre_uint8 *classbits, pcre_uchar **uchardptr,
3643 int options, compile_data *cd, const pcre_uint32 *p)
3644 {
3645 BOOL utf = (options & PCRE_UTF8) != 0;
3646 int n8 = 0;
3647 if (p[0] > 0)
3648 n8 += add_to_class(classbits, uchardptr, options, cd, 0, p[0] - 1);
3649 while (p[0] < NOTACHAR)
3650 {
3651 while (p[1] == p[0] + 1) p++;
3652 n8 += add_to_class(classbits, uchardptr, options, cd, p[0] + 1,
3653 (p[1] == NOTACHAR) ? (utf ? 0x10ffffu : 0xffffffffu) : p[1] - 1);
3654 p++;
3655 }
3656 return n8;
3657 }
3658
3659
3660
3661 /*************************************************
3662 * Compile one branch *
3663 *************************************************/
3664
3665 /* Scan the pattern, compiling it into the a vector. If the options are
3666 changed during the branch, the pointer is used to change the external options
3667 bits. This function is used during the pre-compile phase when we are trying
3668 to find out the amount of memory needed, as well as during the real compile
3669 phase. The value of lengthptr distinguishes the two phases.
3670
3671 Arguments:
3672 optionsptr pointer to the option bits
3673 codeptr points to the pointer to the current code point
3674 ptrptr points to the current pattern pointer
3675 errorcodeptr points to error code variable
3676 firstcharptr place to put the first required character
3677 firstcharflagsptr place to put the first character flags, or a negative number
3678 reqcharptr place to put the last required character
3679 reqcharflagsptr place to put the last required character flags, or a negative number
3680 bcptr points to current branch chain
3681 cond_depth conditional nesting depth
3682 cd contains pointers to tables etc.
3683 lengthptr NULL during the real compile phase
3684 points to length accumulator during pre-compile phase
3685
3686 Returns: TRUE on success
3687 FALSE, with *errorcodeptr set non-zero on error
3688 */
3689
3690 static BOOL
3691 compile_branch(int *optionsptr, pcre_uchar **codeptr,
3692 const pcre_uchar **ptrptr, int *errorcodeptr,
3693 pcre_uint32 *firstcharptr, pcre_int32 *firstcharflagsptr,
3694 pcre_uint32 *reqcharptr, pcre_int32 *reqcharflagsptr,
3695 branch_chain *bcptr, int cond_depth,
3696 compile_data *cd, int *lengthptr)
3697 {
3698 int repeat_type, op_type;
3699 int repeat_min = 0, repeat_max = 0; /* To please picky compilers */
3700 int bravalue = 0;
3701 int greedy_default, greedy_non_default;
3702 pcre_uint32 firstchar, reqchar;
3703 pcre_int32 firstcharflags, reqcharflags;
3704 pcre_uint32 zeroreqchar, zerofirstchar;
3705 pcre_int32 zeroreqcharflags, zerofirstcharflags;
3706 pcre_int32 req_caseopt, reqvary, tempreqvary;
3707 int options = *optionsptr; /* May change dynamically */
3708 int after_manual_callout = 0;
3709 int length_prevgroup = 0;
3710 register pcre_uint32 c;
3711 int escape;
3712 register pcre_uchar *code = *codeptr;
3713 pcre_uchar *last_code = code;
3714 pcre_uchar *orig_code = code;
3715 pcre_uchar *tempcode;
3716 BOOL inescq = FALSE;
3717 BOOL groupsetfirstchar = FALSE;
3718 const pcre_uchar *ptr = *ptrptr;
3719 const pcre_uchar *tempptr;
3720 const pcre_uchar *nestptr = NULL;
3721 pcre_uchar *previous = NULL;
3722 pcre_uchar *previous_callout = NULL;
3723 pcre_uchar *save_hwm = NULL;
3724 pcre_uint8 classbits[32];
3725
3726 /* We can fish out the UTF-8 setting once and for all into a BOOL, but we
3727 must not do this for other options (e.g. PCRE_EXTENDED) because they may change
3728 dynamically as we process the pattern. */
3729
3730 #ifdef SUPPORT_UTF
3731 /* PCRE_UTF[16|32] have the same value as PCRE_UTF8. */
3732 BOOL utf = (options & PCRE_UTF8) != 0;
3733 #ifndef COMPILE_PCRE32
3734 pcre_uchar utf_chars[6];
3735 #endif
3736 #else
3737 BOOL utf = FALSE;
3738 #endif
3739
3740 /* Helper variables for OP_XCLASS opcode (for characters > 255). We define
3741 class_uchardata always so that it can be passed to add_to_class() always,
3742 though it will not be used in non-UTF 8-bit cases. This avoids having to supply
3743 alternative calls for the different cases. */
3744
3745 pcre_uchar *class_uchardata;
3746 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
3747 BOOL xclass;
3748 pcre_uchar *class_uchardata_base;
3749 #endif
3750
3751 #ifdef PCRE_DEBUG
3752 if (lengthptr != NULL) DPRINTF((">> start branch\n"));
3753 #endif
3754
3755 /* Set up the default and non-default settings for greediness */
3756
3757 greedy_default = ((options & PCRE_UNGREEDY) != 0);
3758 greedy_non_default = greedy_default ^ 1;
3759
3760 /* Initialize no first byte, no required byte. REQ_UNSET means "no char
3761 matching encountered yet". It gets changed to REQ_NONE if we hit something that
3762 matches a non-fixed char first char; reqchar just remains unset if we never
3763 find one.
3764
3765 When we hit a repeat whose minimum is zero, we may have to adjust these values
3766 to take the zero repeat into account. This is implemented by setting them to
3767 zerofirstbyte and zeroreqchar when such a repeat is encountered. The individual
3768 item types that can be repeated set these backoff variables appropriately. */
3769
3770 firstchar = reqchar = zerofirstchar = zeroreqchar = 0;
3771 firstcharflags = reqcharflags = zerofirstcharflags = zeroreqcharflags = REQ_UNSET;
3772
3773 /* The variable req_caseopt contains either the REQ_CASELESS value
3774 or zero, according to the current setting of the caseless flag. The
3775 REQ_CASELESS leaves the lower 28 bit empty. It is added into the
3776 firstchar or reqchar variables to record the case status of the
3777 value. This is used only for ASCII characters. */
3778
3779 req_caseopt = ((options & PCRE_CASELESS) != 0)? REQ_CASELESS:0;
3780
3781 /* Switch on next character until the end of the branch */
3782
3783 for (;; ptr++)
3784 {
3785 BOOL negate_class;
3786 BOOL should_flip_negation;
3787 BOOL possessive_quantifier;
3788 BOOL is_quantifier;
3789 BOOL is_recurse;
3790 BOOL reset_bracount;
3791 int class_has_8bitchar;
3792 int class_one_char;
3793 int newoptions;
3794 int recno;
3795 int refsign;
3796 int skipbytes;
3797 pcre_uint32 subreqchar, subfirstchar;
3798 pcre_int32 subreqcharflags, subfirstcharflags;
3799 int terminator;
3800 unsigned int mclength;
3801 unsigned int tempbracount;
3802 pcre_uint32 ec;
3803 pcre_uchar mcbuffer[8];
3804
3805 /* Get next character in the pattern */
3806
3807 c = *ptr;
3808
3809 /* If we are at the end of a nested substitution, revert to the outer level
3810 string. Nesting only happens one level deep. */
3811
3812 if (c == CHAR_NULL && nestptr != NULL)
3813 {
3814 ptr = nestptr;
3815 nestptr = NULL;
3816 c = *ptr;
3817 }
3818
3819 /* If we are in the pre-compile phase, accumulate the length used for the
3820 previous cycle of this loop. */
3821
3822 if (lengthptr != NULL)
3823 {
3824 #ifdef PCRE_DEBUG
3825 if (code > cd->hwm) cd->hwm = code; /* High water info */
3826 #endif
3827 if (code > cd->start_workspace + cd->workspace_size -
3828 WORK_SIZE_SAFETY_MARGIN) /* Check for overrun */
3829 {
3830 *errorcodeptr = ERR52;
3831 goto FAILED;
3832 }
3833
3834 /* There is at least one situation where code goes backwards: this is the
3835 case of a zero quantifier after a class (e.g. [ab]{0}). At compile time,
3836 the class is simply eliminated. However, it is created first, so we have to
3837 allow memory for it. Therefore, don't ever reduce the length at this point.
3838 */
3839
3840 if (code < last_code) code = last_code;
3841
3842 /* Paranoid check for integer overflow */
3843
3844 if (OFLOW_MAX - *lengthptr < code - last_code)
3845 {
3846 *errorcodeptr = ERR20;
3847 goto FAILED;
3848 }
3849
3850 *lengthptr += (int)(code - last_code);
3851 DPRINTF(("length=%d added %d c=%c (0x%x)\n", *lengthptr,
3852 (int)(code - last_code), c, c));
3853
3854 /* If "previous" is set and it is not at the start of the work space, move
3855 it back to there, in order to avoid filling up the work space. Otherwise,
3856 if "previous" is NULL, reset the current code pointer to the start. */
3857
3858 if (previous != NULL)
3859 {
3860 if (previous > orig_code)
3861 {
3862 memmove(orig_code, previous, IN_UCHARS(code - previous));
3863 code -= previous - orig_code;
3864 previous = orig_code;
3865 }
3866 }
3867 else code = orig_code;
3868
3869 /* Remember where this code item starts so we can pick up the length
3870 next time round. */
3871
3872 last_code = code;
3873 }
3874
3875 /* In the real compile phase, just check the workspace used by the forward
3876 reference list. */
3877
3878 else if (cd->hwm > cd->start_workspace + cd->workspace_size -
3879 WORK_SIZE_SAFETY_MARGIN)
3880 {
3881 *errorcodeptr = ERR52;
3882 goto FAILED;
3883 }
3884
3885 /* If in \Q...\E, check for the end; if not, we have a literal */
3886
3887 if (inescq && c != CHAR_NULL)
3888 {
3889 if (c == CHAR_BACKSLASH && ptr[1] == CHAR_E)
3890 {
3891 inescq = FALSE;
3892 ptr++;
3893 continue;
3894 }
3895 else
3896 {
3897 if (previous_callout != NULL)
3898 {
3899 if (lengthptr == NULL) /* Don't attempt in pre-compile phase */
3900 complete_callout(previous_callout, ptr, cd);
3901 previous_callout = NULL;
3902 }
3903 if ((options & PCRE_AUTO_CALLOUT) != 0)
3904 {
3905 previous_callout = code;
3906 code = auto_callout(code, ptr, cd);
3907 }
3908 goto NORMAL_CHAR;
3909 }
3910 }
3911
3912 /* Fill in length of a previous callout, except when the next thing is
3913 a quantifier. */
3914
3915 is_quantifier =
3916 c == CHAR_ASTERISK || c == CHAR_PLUS || c == CHAR_QUESTION_MARK ||
3917 (c == CHAR_LEFT_CURLY_BRACKET && is_counted_repeat(ptr+1));
3918
3919 if (!is_quantifier && previous_callout != NULL &&
3920 after_manual_callout-- <= 0)
3921 {
3922 if (lengthptr == NULL) /* Don't attempt in pre-compile phase */
3923 complete_callout(previous_callout, ptr, cd);
3924 previous_callout = NULL;
3925 }
3926
3927 /* In extended mode, skip white space and comments. */
3928
3929 if ((options & PCRE_EXTENDED) != 0)
3930 {
3931 if (MAX_255(*ptr) && (cd->ctypes[c] & ctype_space) != 0) continue;
3932 if (c == CHAR_NUMBER_SIGN)
3933 {
3934 ptr++;
3935 while (*ptr != CHAR_NULL)
3936 {
3937 if (IS_NEWLINE(ptr)) { ptr += cd->nllen - 1; break; }
3938 ptr++;
3939 #ifdef SUPPORT_UTF
3940 if (utf) FORWARDCHAR(ptr);
3941 #endif
3942 }
3943 if (*ptr != CHAR_NULL) continue;
3944
3945 /* Else fall through to handle end of string */
3946 c = 0;
3947 }
3948 }
3949
3950 /* No auto callout for quantifiers. */
3951
3952 if ((options & PCRE_AUTO_CALLOUT) != 0 && !is_quantifier)
3953 {
3954 previous_callout = code;
3955 code = auto_callout(code, ptr, cd);
3956 }
3957
3958 switch(c)
3959 {
3960 /* ===================================================================*/
3961 case 0: /* The branch terminates at string end */
3962 case CHAR_VERTICAL_LINE: /* or | or ) */
3963 case CHAR_RIGHT_PARENTHESIS:
3964 *firstcharptr = firstchar;
3965 *firstcharflagsptr = firstcharflags;
3966 *reqcharptr = reqchar;
3967 *reqcharflagsptr = reqcharflags;
3968 *codeptr = code;
3969 *ptrptr = ptr;
3970 if (lengthptr != NULL)
3971 {
3972 if (OFLOW_MAX - *lengthptr < code - last_code)
3973 {
3974 *errorcodeptr = ERR20;
3975 goto FAILED;
3976 }
3977 *lengthptr += (int)(code - last_code); /* To include callout length */
3978 DPRINTF((">> end branch\n"));
3979 }
3980 return TRUE;
3981
3982
3983 /* ===================================================================*/
3984 /* Handle single-character metacharacters. In multiline mode, ^ disables
3985 the setting of any following char as a first character. */
3986
3987 case CHAR_CIRCUMFLEX_ACCENT:
3988 previous = NULL;
3989 if ((options & PCRE_MULTILINE) != 0)
3990 {
3991 if (firstcharflags == REQ_UNSET) firstcharflags = REQ_NONE;
3992 *code++ = OP_CIRCM;
3993 }
3994 else *code++ = OP_CIRC;
3995 break;
3996
3997 case CHAR_DOLLAR_SIGN:
3998 previous = NULL;
3999 *code++ = ((options & PCRE_MULTILINE) != 0)? OP_DOLLM : OP_DOLL;
4000 break;
4001
4002 /* There can never be a first char if '.' is first, whatever happens about
4003 repeats. The value of reqchar doesn't change either. */
4004
4005 case CHAR_DOT:
4006 if (firstcharflags == REQ_UNSET) firstcharflags = REQ_NONE;
4007 zerofirstchar = firstchar;
4008 zerofirstcharflags = firstcharflags;
4009 zeroreqchar = reqchar;
4010 zeroreqcharflags = reqcharflags;
4011 previous = code;
4012 *code++ = ((options & PCRE_DOTALL) != 0)? OP_ALLANY: OP_ANY;
4013 break;
4014
4015
4016 /* ===================================================================*/
4017 /* Character classes. If the included characters are all < 256, we build a
4018 32-byte bitmap of the permitted characters, except in the special case
4019 where there is only one such character. For negated classes, we build the
4020 map as usual, then invert it at the end. However, we use a different opcode
4021 so that data characters > 255 can be handled correctly.
4022
4023 If the class contains characters outside the 0-255 range, a different
4024 opcode is compiled. It may optionally have a bit map for characters < 256,
4025 but those above are are explicitly listed afterwards. A flag byte tells
4026 whether the bitmap is present, and whether this is a negated class or not.
4027
4028 In JavaScript compatibility mode, an isolated ']' causes an error. In
4029 default (Perl) mode, it is treated as a data character. */
4030
4031 case CHAR_RIGHT_SQUARE_BRACKET:
4032 if ((cd->external_options & PCRE_JAVASCRIPT_COMPAT) != 0)
4033 {
4034 *errorcodeptr = ERR64;
4035 goto FAILED;
4036 }
4037 goto NORMAL_CHAR;
4038
4039 case CHAR_LEFT_SQUARE_BRACKET:
4040 previous = code;
4041
4042 /* PCRE supports POSIX class stuff inside a class. Perl gives an error if
4043 they are encountered at the top level, so we'll do that too. */
4044
4045 if ((ptr[1] == CHAR_COLON || ptr[1] == CHAR_DOT ||
4046 ptr[1] == CHAR_EQUALS_SIGN) &&
4047 check_posix_syntax(ptr, &tempptr))
4048 {
4049 *errorcodeptr = (ptr[1] == CHAR_COLON)? ERR13 : ERR31;
4050 goto FAILED;
4051 }
4052
4053 /* If the first character is '^', set the negation flag and skip it. Also,
4054 if the first few characters (either before or after ^) are \Q\E or \E we
4055 skip them too. This makes for compatibility with Perl. */
4056
4057 negate_class = FALSE;
4058 for (;;)
4059 {
4060 c = *(++ptr);
4061 if (c == CHAR_BACKSLASH)
4062 {
4063 if (ptr[1] == CHAR_E)
4064 ptr++;
4065 else if (STRNCMP_UC_C8(ptr + 1, STR_Q STR_BACKSLASH STR_E, 3) == 0)
4066 ptr += 3;
4067 else
4068 break;
4069 }
4070 else if (!negate_class && c == CHAR_CIRCUMFLEX_ACCENT)
4071 negate_class = TRUE;
4072 else break;
4073 }
4074
4075 /* Empty classes are allowed in JavaScript compatibility mode. Otherwise,
4076 an initial ']' is taken as a data character -- the code below handles
4077 that. In JS mode, [] must always fail, so generate OP_FAIL, whereas
4078 [^] must match any character, so generate OP_ALLANY. */
4079
4080 if (c == CHAR_RIGHT_SQUARE_BRACKET &&
4081 (cd->external_options & PCRE_JAVASCRIPT_COMPAT) != 0)
4082 {
4083 *code++ = negate_class? OP_ALLANY : OP_FAIL;
4084 if (firstcharflags == REQ_UNSET) firstcharflags = REQ_NONE;
4085 zerofirstchar = firstchar;
4086 zerofirstcharflags = firstcharflags;
4087 break;
4088 }
4089
4090 /* If a class contains a negative special such as \S, we need to flip the
4091 negation flag at the end, so that support for characters > 255 works
4092 correctly (they are all included in the class). */
4093
4094 should_flip_negation = FALSE;
4095
4096 /* For optimization purposes, we track some properties of the class:
4097 class_has_8bitchar will be non-zero if the class contains at least one <
4098 256 character; class_one_char will be 1 if the class contains just one
4099 character. */
4100
4101 class_has_8bitchar = 0;
4102 class_one_char = 0;
4103
4104 /* Initialize the 32-char bit map to all zeros. We build the map in a
4105 temporary bit of memory, in case the class contains fewer than two
4106 8-bit characters because in that case the compiled code doesn't use the bit
4107 map. */
4108
4109 memset(classbits, 0, 32 * sizeof(pcre_uint8));
4110
4111 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
4112 xclass = FALSE;
4113 class_uchardata = code + LINK_SIZE + 2; /* For XCLASS items */
4114 class_uchardata_base = class_uchardata; /* Save the start */
4115 #endif
4116
4117 /* Process characters until ] is reached. By writing this as a "do" it
4118 means that an initial ] is taken as a data character. At the start of the
4119 loop, c contains the first byte of the character. */
4120
4121 if (c != CHAR_NULL) do
4122 {
4123 const pcre_uchar *oldptr;
4124
4125 #ifdef SUPPORT_UTF
4126 if (utf && HAS_EXTRALEN(c))
4127 { /* Braces are required because the */
4128 GETCHARLEN(c, ptr, ptr); /* macro generates multiple statements */
4129 }
4130 #endif
4131
4132 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
4133 /* In the pre-compile phase, accumulate the length of any extra
4134 data and reset the pointer. This is so that very large classes that
4135 contain a zillion > 255 characters no longer overwrite the work space
4136 (which is on the stack). We have to remember that there was XCLASS data,
4137 however. */
4138
4139 if (lengthptr != NULL && class_uchardata > class_uchardata_base)
4140 {
4141 xclass = TRUE;
4142 *lengthptr += class_uchardata - class_uchardata_base;
4143 class_uchardata = class_uchardata_base;
4144 }
4145 #endif
4146
4147 /* Inside \Q...\E everything is literal except \E */
4148
4149 if (inescq)
4150 {
4151 if (c == CHAR_BACKSLASH && ptr[1] == CHAR_E) /* If we are at \E */
4152 {
4153 inescq = FALSE; /* Reset literal state */
4154 ptr++; /* Skip the 'E' */
4155 continue; /* Carry on with next */
4156 }
4157 goto CHECK_RANGE; /* Could be range if \E follows */
4158 }
4159
4160 /* Handle POSIX class names. Perl allows a negation extension of the
4161 form [:^name:]. A square bracket that doesn't match the syntax is
4162 treated as a literal. We also recognize the POSIX constructions
4163 [.ch.] and [=ch=] ("collating elements") and fault them, as Perl
4164 5.6 and 5.8 do. */
4165
4166 if (c == CHAR_LEFT_SQUARE_BRACKET &&
4167 (ptr[1] == CHAR_COLON || ptr[1] == CHAR_DOT ||
4168 ptr[1] == CHAR_EQUALS_SIGN) && check_posix_syntax(ptr, &tempptr))
4169 {
4170 BOOL local_negate = FALSE;
4171 int posix_class, taboffset, tabopt;
4172 register const pcre_uint8 *cbits = cd->cbits;
4173 pcre_uint8 pbits[32];
4174
4175 if (ptr[1] != CHAR_COLON)
4176 {
4177 *errorcodeptr = ERR31;
4178 goto FAILED;
4179 }
4180
4181 ptr += 2;
4182 if (*ptr == CHAR_CIRCUMFLEX_ACCENT)
4183 {
4184 local_negate = TRUE;
4185 should_flip_negation = TRUE; /* Note negative special */
4186 ptr++;
4187 }
4188
4189 posix_class = check_posix_name(ptr, (int)(tempptr - ptr));
4190 if (posix_class < 0)
4191 {
4192 *errorcodeptr = ERR30;
4193 goto FAILED;
4194 }
4195
4196 /* If matching is caseless, upper and lower are converted to
4197 alpha. This relies on the fact that the class table starts with
4198 alpha, lower, upper as the first 3 entries. */
4199
4200 if ((options & PCRE_CASELESS) != 0 && posix_class <= 2)
4201 posix_class = 0;
4202
4203 /* When PCRE_UCP is set, some of the POSIX classes are converted to
4204 different escape sequences that use Unicode properties. */
4205
4206 #ifdef SUPPORT_UCP
4207 if ((options & PCRE_UCP) != 0)
4208 {
4209 int pc = posix_class + ((local_negate)? POSIX_SUBSIZE/2 : 0);
4210 if (posix_substitutes[pc] != NULL)
4211 {
4212 nestptr = tempptr + 1;
4213 ptr = posix_substitutes[pc] - 1;
4214 continue;
4215 }
4216 }
4217 #endif
4218 /* In the non-UCP case, we build the bit map for the POSIX class in a
4219 chunk of local store because we may be adding and subtracting from it,
4220 and we don't want to subtract bits that may be in the main map already.
4221 At the end we or the result into the bit map that is being built. */
4222
4223 posix_class *= 3;
4224
4225 /* Copy in the first table (always present) */
4226
4227 memcpy(pbits, cbits + posix_class_maps[posix_class],
4228 32 * sizeof(pcre_uint8));
4229
4230 /* If there is a second table, add or remove it as required. */
4231
4232 taboffset = posix_class_maps[posix_class + 1];
4233 tabopt = posix_class_maps[posix_class + 2];
4234
4235 if (taboffset >= 0)
4236 {
4237 if (tabopt >= 0)
4238 for (c = 0; c < 32; c++) pbits[c] |= cbits[c + taboffset];
4239 else
4240 for (c = 0; c < 32; c++) pbits[c] &= ~cbits[c + taboffset];
4241 }
4242
4243 /* Now see if we need to remove any special characters. An option
4244 value of 1 removes vertical space and 2 removes underscore. */
4245
4246 if (tabopt < 0) tabopt = -tabopt;
4247 if (tabopt == 1) pbits[1] &= ~0x3c;
4248 else if (tabopt == 2) pbits[11] &= 0x7f;
4249
4250 /* Add the POSIX table or its complement into the main table that is
4251 being built and we are done. */
4252
4253 if (local_negate)
4254 for (c = 0; c < 32; c++) classbits[c] |= ~pbits[c];
4255 else
4256 for (c = 0; c < 32; c++) classbits[c] |= pbits[c];
4257
4258 ptr = tempptr + 1;
4259 /* Every class contains at least one < 256 character. */
4260 class_has_8bitchar = 1;
4261 /* Every class contains at least two characters. */
4262 class_one_char = 2;
4263 continue; /* End of POSIX syntax handling */
4264 }
4265
4266 /* Backslash may introduce a single character, or it may introduce one
4267 of the specials, which just set a flag. The sequence \b is a special
4268 case. Inside a class (and only there) it is treated as backspace. We
4269 assume that other escapes have more than one character in them, so
4270 speculatively set both class_has_8bitchar and class_one_char bigger
4271 than one. Unrecognized escapes fall through and are either treated
4272 as literal characters (by default), or are faulted if
4273 PCRE_EXTRA is set. */
4274
4275 if (c == CHAR_BACKSLASH)
4276 {
4277 escape = check_escape(&ptr, &ec, errorcodeptr, cd->bracount, options,
4278 TRUE);
4279 if (*errorcodeptr != 0) goto FAILED;
4280 if (escape == 0) c = ec;
4281 else if (escape == ESC_b) c = CHAR_BS; /* \b is backspace in a class */
4282 else if (escape == ESC_N) /* \N is not supported in a class */
4283 {
4284 *errorcodeptr = ERR71;
4285 goto FAILED;
4286 }
4287 else if (escape == ESC_Q) /* Handle start of quoted string */
4288 {
4289 if (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
4290 {
4291 ptr += 2; /* avoid empty string */
4292 }
4293 else inescq = TRUE;
4294 continue;
4295 }
4296 else if (escape == ESC_E) continue; /* Ignore orphan \E */
4297
4298 else
4299 {
4300 register const pcre_uint8 *cbits = cd->cbits;
4301 /* Every class contains at least two < 256 characters. */
4302 class_has_8bitchar++;
4303 /* Every class contains at least two characters. */
4304 class_one_char += 2;
4305
4306 switch (escape)
4307 {
4308 #ifdef SUPPORT_UCP
4309 case ESC_du: /* These are the values given for \d etc */
4310 case ESC_DU: /* when PCRE_UCP is set. We replace the */
4311 case ESC_wu: /* escape sequence with an appropriate \p */
4312 case ESC_WU: /* or \P to test Unicode properties instead */
4313 case ESC_su: /* of the default ASCII testing. */
4314 case ESC_SU:
4315 nestptr = ptr;
4316 ptr = substitutes[escape - ESC_DU] - 1; /* Just before substitute */
4317 class_has_8bitchar--; /* Undo! */
4318 continue;
4319 #endif
4320 case ESC_d:
4321 for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_digit];
4322 continue;
4323
4324 case ESC_D:
4325 should_flip_negation = TRUE;
4326 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_digit];
4327 continue;
4328
4329 case ESC_w:
4330 for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_word];
4331 continue;
4332
4333 case ESC_W:
4334 should_flip_negation = TRUE;
4335 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_word];
4336 continue;
4337
4338 /* Perl 5.004 onwards omits VT from \s, but we must preserve it
4339 if it was previously set by something earlier in the character
4340 class. Luckily, the value of CHAR_VT is 0x0b in both ASCII and
4341 EBCDIC, so we lazily just adjust the appropriate bit. */
4342
4343 case ESC_s:
4344 classbits[0] |= cbits[cbit_space];
4345 classbits[1] |= cbits[cbit_space+1] & ~0x08;
4346 for (c = 2; c < 32; c++) classbits[c] |= cbits[c+cbit_space];
4347 continue;
4348
4349 case ESC_S:
4350 should_flip_negation = TRUE;
4351 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_space];
4352 classbits[1] |= 0x08; /* Perl 5.004 onwards omits VT from \s */
4353 continue;
4354
4355 /* The rest apply in both UCP and non-UCP cases. */
4356
4357 case ESC_h:
4358 (void)add_list_to_class(classbits, &class_uchardata, options, cd,
4359 PRIV(hspace_list), NOTACHAR);
4360 continue;
4361
4362 case ESC_H:
4363 (void)add_not_list_to_class(classbits, &class_uchardata, options,
4364 cd, PRIV(hspace_list));
4365 continue;
4366
4367 case ESC_v:
4368 (void)add_list_to_class(classbits, &class_uchardata, options, cd,
4369 PRIV(vspace_list), NOTACHAR);
4370 continue;
4371
4372 case ESC_V:
4373 (void)add_not_list_to_class(classbits, &class_uchardata, options,
4374 cd, PRIV(vspace_list));
4375 continue;
4376
4377 #ifdef SUPPORT_UCP
4378 case ESC_p:
4379 case ESC_P:
4380 {
4381 BOOL negated;
4382 unsigned int ptype = 0, pdata = 0;
4383 if (!get_ucp(&ptr, &negated, &ptype, &pdata, errorcodeptr))
4384 goto FAILED;
4385 *class_uchardata++ = ((escape == ESC_p) != negated)?
4386 XCL_PROP : XCL_NOTPROP;
4387 *class_uchardata++ = ptype;
4388 *class_uchardata++ = pdata;
4389 class_has_8bitchar--; /* Undo! */
4390 continue;
4391 }
4392 #endif
4393 /* Unrecognized escapes are faulted if PCRE is running in its
4394 strict mode. By default, for compatibility with Perl, they are
4395 treated as literals. */
4396
4397 default:
4398 if ((options & PCRE_EXTRA) != 0)
4399 {
4400 *errorcodeptr = ERR7;
4401 goto FAILED;
4402 }
4403 class_has_8bitchar--; /* Undo the speculative increase. */
4404 class_one_char -= 2; /* Undo the speculative increase. */
4405 c = *ptr; /* Get the final character and fall through */
4406 break;
4407 }
4408 }
4409
4410 /* Fall through if the escape just defined a single character (c >= 0).
4411 This may be greater than 256. */
4412
4413 escape = 0;
4414
4415 } /* End of backslash handling */
4416
4417 /* A character may be followed by '-' to form a range. However, Perl does
4418 not permit ']' to be the end of the range. A '-' character at the end is
4419 treated as a literal. Perl ignores orphaned \E sequences entirely. The
4420 code for handling \Q and \E is messy. */
4421
4422 CHECK_RANGE:
4423 while (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
4424 {
4425 inescq = FALSE;
4426 ptr += 2;
4427 }
4428 oldptr = ptr;
4429
4430 /* Remember if \r or \n were explicitly used */
4431
4432 if (c == CHAR_CR || c == CHAR_NL) cd->external_flags |= PCRE_HASCRORLF;
4433
4434 /* Check for range */
4435
4436 if (!inescq && ptr[1] == CHAR_MINUS)
4437 {
4438 pcre_uint32 d;
4439 ptr += 2;
4440 while (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_E) ptr += 2;
4441
4442 /* If we hit \Q (not followed by \E) at this point, go into escaped
4443 mode. */
4444
4445 while (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_Q)
4446 {
4447 ptr += 2;
4448 if (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_E)
4449 { ptr += 2; continue; }
4450 inescq = TRUE;
4451 break;
4452 }
4453
4454 /* Minus (hyphen) at the end of a class is treated as a literal, so put
4455 back the pointer and jump to handle the character that preceded it. */
4456
4457 if (*ptr == CHAR_NULL || (!inescq && *ptr == CHAR_RIGHT_SQUARE_BRACKET))
4458 {
4459 ptr = oldptr;
4460 goto CLASS_SINGLE_CHARACTER;
4461 }
4462
4463 /* Otherwise, we have a potential range; pick up the next character */
4464
4465 #ifdef SUPPORT_UTF
4466 if (utf)
4467 { /* Braces are required because the */
4468 GETCHARLEN(d, ptr, ptr); /* macro generates multiple statements */
4469 }
4470 else
4471 #endif
4472 d = *ptr; /* Not UTF-8 mode */
4473
4474 /* The second part of a range can be a single-character escape, but
4475 not any of the other escapes. Perl 5.6 treats a hyphen as a literal
4476 in such circumstances. */
4477
4478 if (!inescq && d == CHAR_BACKSLASH)
4479 {
4480 int descape;
4481 descape = check_escape(&ptr, &d, errorcodeptr, cd->bracount, options, TRUE);
4482 if (*errorcodeptr != 0) goto FAILED;
4483
4484 /* \b is backspace; any other special means the '-' was literal. */
4485
4486 if (descape != 0)
4487 {
4488 if (descape == ESC_b) d = CHAR_BS; else
4489 {
4490 ptr = oldptr;
4491 goto CLASS_SINGLE_CHARACTER; /* A few lines below */
4492 }
4493 }
4494 }
4495
4496 /* Check that the two values are in the correct order. Optimize
4497 one-character ranges. */
4498
4499 if (d < c)
4500 {
4501 *errorcodeptr = ERR8;
4502 goto FAILED;
4503 }
4504 if (d == c) goto CLASS_SINGLE_CHARACTER; /* A few lines below */
4505
4506 /* We have found a character range, so single character optimizations
4507 cannot be done anymore. Any value greater than 1 indicates that there
4508 is more than one character. */
4509
4510 class_one_char = 2;
4511
4512 /* Remember an explicit \r or \n, and add the range to the class. */
4513
4514 if (d == CHAR_CR || d == CHAR_NL) cd->external_flags |= PCRE_HASCRORLF;
4515
4516 class_has_8bitchar +=
4517 add_to_class(classbits, &class_uchardata, options, cd, c, d);
4518
4519 continue; /* Go get the next char in the class */
4520 }
4521
4522 /* Handle a single character - we can get here for a normal non-escape
4523 char, or after \ that introduces a single character or for an apparent
4524 range that isn't. Only the value 1 matters for class_one_char, so don't
4525 increase it if it is already 2 or more ... just in case there's a class
4526 with a zillion characters in it. */
4527
4528 CLASS_SINGLE_CHARACTER:
4529 if (class_one_char < 2) class_one_char++;
4530
4531 /* If class_one_char is 1, we have the first single character in the
4532 class, and there have been no prior ranges, or XCLASS items generated by
4533 escapes. If this is the final character in the class, we can optimize by
4534 turning the item into a 1-character OP_CHAR[I] if it's positive, or
4535 OP_NOT[I] if it's negative. In the positive case, it can cause firstchar
4536 to be set. Otherwise, there can be no first char if this item is first,
4537 whatever repeat count may follow. In the case of reqchar, save the
4538 previous value for reinstating. */
4539
4540 if (class_one_char == 1 && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET)
4541 {
4542 ptr++;
4543 zeroreqchar = reqchar;
4544 zeroreqcharflags = reqcharflags;
4545
4546 if (negate_class)
4547 {
4548 #ifdef SUPPORT_UCP
4549 int d;
4550 #endif
4551 if (firstcharflags == REQ_UNSET) firstcharflags = REQ_NONE;
4552 zerofirstchar = firstchar;
4553 zerofirstcharflags = firstcharflags;
4554
4555 /* For caseless UTF-8 mode when UCP support is available, check
4556 whether this character has more than one other case. If so, generate
4557 a special OP_NOTPROP item instead of OP_NOTI. */
4558
4559 #ifdef SUPPORT_UCP
4560 if (utf && (options & PCRE_CASELESS) != 0 &&
4561 (d = UCD_CASESET(c)) != 0)
4562 {
4563 *code++ = OP_NOTPROP;
4564 *code++ = PT_CLIST;
4565 *code++ = d;
4566 }
4567 else
4568 #endif
4569 /* Char has only one other case, or UCP not available */
4570
4571 {
4572 *code++ = ((options & PCRE_CASELESS) != 0)? OP_NOTI: OP_NOT;
4573 #if defined SUPPORT_UTF && !defined COMPILE_PCRE32
4574 if (utf && c > MAX_VALUE_FOR_SINGLE_CHAR)
4575 code += PRIV(ord2utf)(c, code);
4576 else
4577 #endif
4578 *code++ = c;
4579 }
4580
4581 /* We are finished with this character class */
4582
4583 goto END_CLASS;
4584 }
4585
4586 /* For a single, positive character, get the value into mcbuffer, and
4587 then we can handle this with the normal one-character code. */
4588
4589 #if defined SUPPORT_UTF && !defined COMPILE_PCRE32
4590 if (utf && c > MAX_VALUE_FOR_SINGLE_CHAR)
4591 mclength = PRIV(ord2utf)(c, mcbuffer);
4592 else
4593 #endif
4594 {
4595 mcbuffer[0] = c;
4596 mclength = 1;
4597 }
4598 goto ONE_CHAR;
4599 } /* End of 1-char optimization */
4600
4601 /* There is more than one character in the class, or an XCLASS item
4602 has been generated. Add this character to the class. */
4603
4604 class_has_8bitchar +=
4605 add_to_class(classbits, &class_uchardata, options, cd, c, c);
4606 }
4607
4608 /* Loop until ']' reached. This "while" is the end of the "do" far above.
4609 If we are at the end of an internal nested string, revert to the outer
4610 string. */
4611
4612 while (((c = *(++ptr)) != CHAR_NULL ||
4613 (nestptr != NULL &&
4614 (ptr = nestptr, nestptr = NULL, c = *(++ptr)) != CHAR_NULL)) &&
4615 (c != CHAR_RIGHT_SQUARE_BRACKET || inescq));
4616
4617 /* Check for missing terminating ']' */
4618
4619 if (c == CHAR_NULL)
4620 {
4621 *errorcodeptr = ERR6;
4622 goto FAILED;
4623 }
4624
4625 /* We will need an XCLASS if data has been placed in class_uchardata. In
4626 the second phase this is a sufficient test. However, in the pre-compile
4627 phase, class_uchardata gets emptied to prevent workspace overflow, so it
4628 only if the very last character in the class needs XCLASS will it contain
4629 anything at this point. For this reason, xclass gets set TRUE above when
4630 uchar_classdata is emptied, and that's why this code is the way it is here
4631 instead of just doing a test on class_uchardata below. */
4632
4633 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
4634 if (class_uchardata > class_uchardata_base) xclass = TRUE;
4635 #endif
4636
4637 /* If this is the first thing in the branch, there can be no first char
4638 setting, whatever the repeat count. Any reqchar setting must remain
4639 unchanged after any kind of repeat. */
4640
4641 if (firstcharflags == REQ_UNSET) firstcharflags = REQ_NONE;
4642 zerofirstchar = firstchar;
4643 zerofirstcharflags = firstcharflags;
4644 zeroreqchar = reqchar;
4645 zeroreqcharflags = reqcharflags;
4646
4647 /* If there are characters with values > 255, we have to compile an
4648 extended class, with its own opcode, unless there was a negated special
4649 such as \S in the class, and PCRE_UCP is not set, because in that case all
4650 characters > 255 are in the class, so any that were explicitly given as
4651 well can be ignored. If (when there are explicit characters > 255 that must
4652 be listed) there are no characters < 256, we can omit the bitmap in the
4653 actual compiled code. */
4654
4655 #ifdef SUPPORT_UTF
4656 if (xclass && (!should_flip_negation || (options & PCRE_UCP) != 0))
4657 #elif !defined COMPILE_PCRE8
4658 if (xclass && !should_flip_negation)
4659 #endif
4660 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
4661 {
4662 *class_uchardata++ = XCL_END; /* Marks the end of extra data */
4663 *code++ = OP_XCLASS;
4664 code += LINK_SIZE;
4665 *code = negate_class? XCL_NOT:0;
4666
4667 /* If the map is required, move up the extra data to make room for it;
4668 otherwise just move the code pointer to the end of the extra data. */
4669
4670 if (class_has_8bitchar > 0)
4671 {
4672 *code++ |= XCL_MAP;
4673 memmove(code + (32 / sizeof(pcre_uchar)), code,
4674 IN_UCHARS(class_uchardata - code));
4675 memcpy(code, classbits, 32);
4676 code = class_uchardata + (32 / sizeof(pcre_uchar));
4677 }
4678 else code = class_uchardata;
4679
4680 /* Now fill in the complete length of the item */
4681
4682 PUT(previous, 1, (int)(code - previous));
4683 break; /* End of class handling */
4684 }
4685 #endif
4686
4687 /* If there are no characters > 255, or they are all to be included or
4688 excluded, set the opcode to OP_CLASS or OP_NCLASS, depending on whether the
4689 whole class was negated and whether there were negative specials such as \S
4690 (non-UCP) in the class. Then copy the 32-byte map into the code vector,
4691 negating it if necessary. */
4692
4693 *code++ = (negate_class == should_flip_negation) ? OP_CLASS : OP_NCLASS;
4694 if (lengthptr == NULL) /* Save time in the pre-compile phase */
4695 {
4696 if (negate_class)
4697 for (c = 0; c < 32; c++) classbits[c] = ~classbits[c];
4698 memcpy(code, classbits, 32);
4699 }
4700 code += 32 / sizeof(pcre_uchar);
4701
4702 END_CLASS:
4703 break;
4704
4705
4706 /* ===================================================================*/
4707 /* Various kinds of repeat; '{' is not necessarily a quantifier, but this
4708 has been tested above. */
4709
4710 case CHAR_LEFT_CURLY_BRACKET:
4711 if (!is_quantifier) goto NORMAL_CHAR;
4712 ptr = read_repeat_counts(ptr+1, &repeat_min, &repeat_max, errorcodeptr);
4713 if (*errorcodeptr != 0) goto FAILED;
4714 goto REPEAT;
4715
4716 case CHAR_ASTERISK:
4717 repeat_min = 0;
4718 repeat_max = -1;
4719 goto REPEAT;
4720
4721 case CHAR_PLUS:
4722 repeat_min = 1;
4723 repeat_max = -1;
4724 goto REPEAT;
4725
4726 case CHAR_QUESTION_MARK:
4727 repeat_min = 0;
4728 repeat_max = 1;
4729
4730 REPEAT:
4731 if (previous == NULL)
4732 {
4733 *errorcodeptr = ERR9;
4734 goto FAILED;
4735 }
4736
4737 if (repeat_min == 0)
4738 {
4739 firstchar = zerofirstchar; /* Adjust for zero repeat */
4740 firstcharflags = zerofirstcharflags;
4741 reqchar = zeroreqchar; /* Ditto */
4742 reqcharflags = zeroreqcharflags;
4743 }
4744
4745 /* Remember whether this is a variable length repeat */
4746
4747 reqvary = (repeat_min == repeat_max)? 0 : REQ_VARY;
4748
4749 op_type = 0; /* Default single-char op codes */
4750 possessive_quantifier = FALSE; /* Default not possessive quantifier */
4751
4752 /* Save start of previous item, in case we have to move it up in order to
4753 insert something before it. */
4754
4755 tempcode = previous;
4756
4757 /* If the next character is '+', we have a possessive quantifier. This
4758 implies greediness, whatever the setting of the PCRE_UNGREEDY option.
4759 If the next character is '?' this is a minimizing repeat, by default,
4760 but if PCRE_UNGREEDY is set, it works the other way round. We change the
4761 repeat type to the non-default. */
4762
4763 if (ptr[1] == CHAR_PLUS)
4764 {
4765 repeat_type = 0; /* Force greedy */
4766 possessive_quantifier = TRUE;
4767 ptr++;
4768 }
4769 else if (ptr[1] == CHAR_QUESTION_MARK)
4770 {
4771 repeat_type = greedy_non_default;
4772 ptr++;
4773 }
4774 else repeat_type = greedy_default;
4775
4776 /* If previous was a recursion call, wrap it in atomic brackets so that
4777 previous becomes the atomic group. All recursions were so wrapped in the
4778 past, but it no longer happens for non-repeated recursions. In fact, the
4779 repeated ones could be re-implemented independently so as not to need this,
4780 but for the moment we rely on the code for repeating groups. */
4781
4782 if (*previous == OP_RECURSE)
4783 {
4784 memmove(previous + 1 + LINK_SIZE, previous, IN_UCHARS(1 + LINK_SIZE));
4785 *previous = OP_ONCE;
4786 PUT(previous, 1, 2 + 2*LINK_SIZE);
4787 previous[2 + 2*LINK_SIZE] = OP_KET;
4788 PUT(previous, 3 + 2*LINK_SIZE, 2 + 2*LINK_SIZE);
4789 code += 2 + 2 * LINK_SIZE;
4790 length_prevgroup = 3 + 3*LINK_SIZE;
4791
4792 /* When actually compiling, we need to check whether this was a forward
4793 reference, and if so, adjust the offset. */
4794
4795 if (lengthptr == NULL && cd->hwm >= cd->start_workspace + LINK_SIZE)
4796 {
4797 int offset = GET(cd->hwm, -LINK_SIZE);
4798 if (offset == previous + 1 - cd->start_code)
4799 PUT(cd->hwm, -LINK_SIZE, offset + 1 + LINK_SIZE);
4800 }
4801 }
4802
4803 /* Now handle repetition for the different types of item. */
4804
4805 /* If previous was a character or negated character match, abolish the item
4806 and generate a repeat item instead. If a char item has a minimum of more
4807 than one, ensure that it is set in reqchar - it might not be if a sequence
4808 such as x{3} is the first thing in a branch because the x will have gone
4809 into firstchar instead. */
4810
4811 if (*previous == OP_CHAR || *previous == OP_CHARI
4812 || *previous == OP_NOT || *previous == OP_NOTI)
4813 {
4814 switch (*previous)
4815 {
4816 default: /* Make compiler happy. */
4817 case OP_CHAR: op_type = OP_STAR - OP_STAR; break;
4818 case OP_CHARI: op_type = OP_STARI - OP_STAR; break;
4819 case OP_NOT: op_type = OP_NOTSTAR - OP_STAR; break;
4820 case OP_NOTI: op_type = OP_NOTSTARI - OP_STAR; break;
4821 }
4822
4823 /* Deal with UTF characters that take up more than one character. It's
4824 easier to write this out separately than try to macrify it. Use c to
4825 hold the length of the character in bytes, plus UTF_LENGTH to flag that
4826 it's a length rather than a small character. */
4827
4828 #if defined SUPPORT_UTF && !defined COMPILE_PCRE32
4829 if (utf && NOT_FIRSTCHAR(code[-1]))
4830 {
4831 pcre_uchar *lastchar = code - 1;
4832 BACKCHAR(lastchar);
4833 c = (int)(code - lastchar); /* Length of UTF-8 character */
4834 memcpy(utf_chars, lastchar, IN_UCHARS(c)); /* Save the char */
4835 c |= UTF_LENGTH; /* Flag c as a length */
4836 }
4837 else
4838 #endif /* SUPPORT_UTF */
4839
4840 /* Handle the case of a single charater - either with no UTF support, or
4841 with UTF disabled, or for a single character UTF character. */
4842 {
4843 c = code[-1];
4844 if (*previous <= OP_CHARI && repeat_min > 1)
4845 {
4846 reqchar = c;
4847 reqcharflags = req_caseopt | cd->req_varyopt;
4848 }
4849 }
4850
4851 /* If the repetition is unlimited, it pays to see if the next thing on
4852 the line is something that cannot possibly match this character. If so,
4853 automatically possessifying this item gains some performance in the case
4854 where the match fails. */
4855
4856 if (!possessive_quantifier &&
4857 repeat_max < 0 &&
4858 check_auto_possessive(previous, utf, ptr + 1, options, cd))
4859 {
4860 repeat_type = 0; /* Force greedy */
4861 possessive_quantifier = TRUE;
4862 }
4863
4864 goto OUTPUT_SINGLE_REPEAT; /* Code shared with single character types */
4865 }
4866
4867 /* If previous was a character type match (\d or similar), abolish it and
4868 create a suitable repeat item. The code is shared with single-character
4869 repeats by setting op_type to add a suitable offset into repeat_type. Note
4870 the the Unicode property types will be present only when SUPPORT_UCP is
4871 defined, but we don't wrap the little bits of code here because it just
4872 makes it horribly messy. */
4873
4874 else if (*previous < OP_EODN)
4875 {
4876 pcre_uchar *oldcode;
4877 int prop_type, prop_value;
4878 op_type = OP_TYPESTAR - OP_STAR; /* Use type opcodes */
4879 c = *previous;
4880
4881 if (!possessive_quantifier &&
4882 repeat_max < 0 &&
4883 check_auto_possessive(previous, utf, ptr + 1, options, cd))
4884 {
4885 repeat_type = 0; /* Force greedy */
4886 possessive_quantifier = TRUE;
4887 }
4888
4889 OUTPUT_SINGLE_REPEAT:
4890 if (*previous == OP_PROP || *previous == OP_NOTPROP)
4891 {
4892 prop_type = previous[1];
4893 prop_value = previous[2];
4894 }
4895 else prop_type = prop_value = -1;
4896
4897 oldcode = code;
4898 code = previous; /* Usually overwrite previous item */
4899
4900 /* If the maximum is zero then the minimum must also be zero; Perl allows
4901 this case, so we do too - by simply omitting the item altogether. */
4902
4903 if (repeat_max == 0) goto END_REPEAT;
4904
4905 /* Combine the op_type with the repeat_type */
4906
4907 repeat_type += op_type;
4908
4909 /* A minimum of zero is handled either as the special case * or ?, or as
4910 an UPTO, with the maximum given. */
4911
4912 if (repeat_min == 0)
4913 {
4914 if (repeat_max == -1) *code++ = OP_STAR + repeat_type;
4915 else if (repeat_max == 1) *code++ = OP_QUERY + repeat_type;
4916 else
4917 {
4918 *code++ = OP_UPTO + repeat_type;
4919 PUT2INC(code, 0, repeat_max);
4920 }
4921 }
4922
4923 /* A repeat minimum of 1 is optimized into some special cases. If the
4924 maximum is unlimited, we use OP_PLUS. Otherwise, the original item is
4925 left in place and, if the maximum is greater than 1, we use OP_UPTO with
4926 one less than the maximum. */
4927
4928 else if (repeat_min == 1)
4929 {
4930 if (repeat_max == -1)
4931 *code++ = OP_PLUS + repeat_type;
4932 else
4933 {
4934 code = oldcode; /* leave previous item in place */
4935 if (repeat_max == 1) goto END_REPEAT;
4936 *code++ = OP_UPTO + repeat_type;
4937 PUT2INC(code, 0, repeat_max - 1);
4938 }
4939 }
4940
4941 /* The case {n,n} is just an EXACT, while the general case {n,m} is
4942 handled as an EXACT followed by an UPTO. */
4943
4944 else
4945 {
4946 *code++ = OP_EXACT + op_type; /* NB EXACT doesn't have repeat_type */
4947 PUT2INC(code, 0, repeat_min);
4948
4949 /* If the maximum is unlimited, insert an OP_STAR. Before doing so,
4950 we have to insert the character for the previous code. For a repeated
4951 Unicode property match, there are two extra bytes that define the
4952 required property. In UTF-8 mode, long characters have their length in
4953 c, with the UTF_LENGTH bit as a flag. */
4954
4955 if (repeat_max < 0)
4956 {
4957 #if defined SUPPORT_UTF && !defined COMPILE_PCRE32
4958 if (utf && (c & UTF_LENGTH) != 0)
4959 {
4960 memcpy(code, utf_chars, IN_UCHARS(c & 7));
4961 code += c & 7;
4962 }
4963 else
4964 #endif
4965 {
4966 *code++ = c;
4967 if (prop_type >= 0)
4968 {
4969 *code++ = prop_type;
4970 *code++ = prop_value;
4971 }
4972 }
4973 *code++ = OP_STAR + repeat_type;
4974 }
4975
4976 /* Else insert an UPTO if the max is greater than the min, again
4977 preceded by the character, for the previously inserted code. If the
4978 UPTO is just for 1 instance, we can use QUERY instead. */
4979
4980 else if (repeat_max != repeat_min)
4981 {
4982 #if defined SUPPORT_UTF && !defined COMPILE_PCRE32
4983 if (utf && (c & UTF_LENGTH) != 0)
4984 {
4985 memcpy(code, utf_chars, IN_UCHARS(c & 7));
4986 code += c & 7;
4987 }
4988 else
4989 #endif
4990 *code++ = c;
4991 if (prop_type >= 0)
4992 {
4993 *code++ = prop_type;
4994 *code++ = prop_value;
4995 }
4996 repeat_max -= repeat_min;
4997
4998 if (repeat_max == 1)
4999 {
5000 *code++ = OP_QUERY + repeat_type;
5001 }
5002 else
5003 {
5004 *code++ = OP_UPTO + repeat_type;
5005 PUT2INC(code, 0, repeat_max);
5006 }
5007 }
5008 }
5009
5010 /* The character or character type itself comes last in all cases. */
5011
5012 #if defined SUPPORT_UTF && !defined COMPILE_PCRE32
5013 if (utf && (c & UTF_LENGTH) != 0)
5014 {
5015 memcpy(code, utf_chars, IN_UCHARS(c & 7));
5016 code += c & 7;
5017 }
5018 else
5019 #endif
5020 *code++ = c;
5021
5022 /* For a repeated Unicode property match, there are two extra bytes that
5023 define the required property. */
5024
5025 #ifdef SUPPORT_UCP
5026 if (prop_type >= 0)
5027 {
5028 *code++ = prop_type;
5029 *code++ = prop_value;
5030 }
5031 #endif
5032 }
5033
5034 /* If previous was a character class or a back reference, we put the repeat
5035 stuff after it, but just skip the item if the repeat was {0,0}. */
5036
5037 else if (*previous == OP_CLASS ||
5038 *previous == OP_NCLASS ||
5039 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
5040 *previous == OP_XCLASS ||
5041 #endif
5042 *previous == OP_REF ||
5043 *previous == OP_REFI)
5044 {
5045 if (repeat_max == 0)
5046 {
5047 code = previous;
5048 goto END_REPEAT;
5049 }
5050
5051 if (repeat_min == 0 && repeat_max == -1)
5052 *code++ = OP_CRSTAR + repeat_type;
5053 else if (repeat_min == 1 && repeat_max == -1)
5054 *code++ = OP_CRPLUS + repeat_type;
5055 else if (repeat_min == 0 && repeat_max == 1)
5056 *code++ = OP_CRQUERY + repeat_type;
5057 else
5058 {
5059 *code++ = OP_CRRANGE + repeat_type;
5060 PUT2INC(code, 0, repeat_min);
5061 if (repeat_max == -1) repeat_max = 0; /* 2-byte encoding for max */
5062 PUT2INC(code, 0, repeat_max);
5063 }
5064 }
5065
5066 /* If previous was a bracket group, we may have to replicate it in certain
5067 cases. Note that at this point we can encounter only the "basic" bracket
5068 opcodes such as BRA and CBRA, as this is the place where they get converted
5069 into the more special varieties such as BRAPOS and SBRA. A test for >=
5070 OP_ASSERT and <= OP_COND includes ASSERT, ASSERT_NOT, ASSERTBACK,
5071 ASSERTBACK_NOT, ONCE, BRA, CBRA, and COND. Originally, PCRE did not allow
5072 repetition of assertions, but now it does, for Perl compatibility. */
5073
5074 else if (*previous >= OP_ASSERT && *previous <= OP_COND)
5075 {
5076 register int i;
5077 int len = (int)(code - previous);
5078 pcre_uchar *bralink = NULL;
5079 pcre_uchar *brazeroptr = NULL;
5080
5081 /* Repeating a DEFINE group is pointless, but Perl allows the syntax, so
5082 we just ignore the repeat. */
5083
5084 if (*previous == OP_COND && previous[LINK_SIZE+1] == OP_DEF)
5085 goto END_REPEAT;
5086
5087 /* There is no sense in actually repeating assertions. The only potential
5088 use of repetition is in cases when the assertion is optional. Therefore,
5089 if the minimum is greater than zero, just ignore the repeat. If the
5090 maximum is not not zero or one, set it to 1. */
5091
5092 if (*previous < OP_ONCE) /* Assertion */
5093 {
5094 if (repeat_min > 0) goto END_REPEAT;
5095 if (repeat_max < 0 || repeat_max > 1) repeat_max = 1;
5096 }
5097
5098 /* The case of a zero minimum is special because of the need to stick
5099 OP_BRAZERO in front of it, and because the group appears once in the
5100 data, whereas in other cases it appears the minimum number of times. For
5101 this reason, it is simplest to treat this case separately, as otherwise
5102 the code gets far too messy. There are several special subcases when the
5103 minimum is zero. */
5104
5105 if (repeat_min == 0)
5106 {
5107 /* If the maximum is also zero, we used to just omit the group from the
5108 output altogether, like this:
5109
5110 ** if (repeat_max == 0)
5111 ** {
5112 ** code = previous;
5113 ** goto END_REPEAT;
5114 ** }
5115
5116 However, that fails when a group or a subgroup within it is referenced
5117 as a subroutine from elsewhere in the pattern, so now we stick in
5118 OP_SKIPZERO in front of it so that it is skipped on execution. As we
5119 don't have a list of which groups are referenced, we cannot do this
5120 selectively.
5121
5122 If the maximum is 1 or unlimited, we just have to stick in the BRAZERO
5123 and do no more at this point. However, we do need to adjust any
5124 OP_RECURSE calls inside the group that refer to the group itself or any
5125 internal or forward referenced group, because the offset is from the
5126 start of the whole regex. Temporarily terminate the pattern while doing
5127 this. */
5128
5129 if (repeat_max <= 1) /* Covers 0, 1, and unlimited */
5130 {
5131 *code = OP_END;
5132 adjust_recurse(previous, 1, utf, cd, save_hwm);
5133 memmove(previous + 1, previous, IN_UCHARS(len));
5134 code++;
5135 if (repeat_max == 0)
5136 {
5137 *previous++ = OP_SKIPZERO;
5138 goto END_REPEAT;
5139 }
5140 brazeroptr = previous; /* Save for possessive optimizing */
5141 *previous++ = OP_BRAZERO + repeat_type;
5142 }
5143
5144 /* If the maximum is greater than 1 and limited, we have to replicate
5145 in a nested fashion, sticking OP_BRAZERO before each set of brackets.
5146 The first one has to be handled carefully because it's the original
5147 copy, which has to be moved up. The remainder can be handled by code
5148 that is common with the non-zero minimum case below. We have to
5149 adjust the value or repeat_max, since one less copy is required. Once
5150 again, we may have to adjust any OP_RECURSE calls inside the group. */
5151
5152 else
5153 {
5154 int offset;
5155 *code = OP_END;
5156 adjust_recurse(previous, 2 + LINK_SIZE, utf, cd, save_hwm);
5157 memmove(previous + 2 + LINK_SIZE, previous, IN_UCHARS(len));
5158 code += 2 + LINK_SIZE;
5159 *previous++ = OP_BRAZERO + repeat_type;
5160 *previous++ = OP_BRA;
5161
5162 /* We chain together the bracket offset fields that have to be
5163 filled in later when the ends of the brackets are reached. */
5164
5165 offset = (bralink == NULL)? 0 : (int)(previous - bralink);
5166 bralink = previous;
5167 PUTINC(previous, 0, offset);
5168 }
5169
5170 repeat_max--;
5171 }
5172
5173 /* If the minimum is greater than zero, replicate the group as many
5174 times as necessary, and adjust the maximum to the number of subsequent
5175 copies that we need. If we set a first char from the group, and didn't
5176 set a required char, copy the latter from the former. If there are any
5177 forward reference subroutine calls in the group, there will be entries on
5178 the workspace list; replicate these with an appropriate increment. */
5179
5180 else
5181 {
5182 if (repeat_min > 1)
5183 {
5184 /* In the pre-compile phase, we don't actually do the replication. We
5185 just adjust the length as if we had. Do some paranoid checks for
5186 potential integer overflow. The INT64_OR_DOUBLE type is a 64-bit
5187 integer type when available, otherwise double. */
5188
5189 if (lengthptr != NULL)
5190 {
5191 int delta = (repeat_min - 1)*length_prevgroup;
5192 if ((INT64_OR_DOUBLE)(repeat_min - 1)*
5193 (INT64_OR_DOUBLE)length_prevgroup >
5194 (INT64_OR_DOUBLE)INT_MAX ||
5195 OFLOW_MAX - *lengthptr < delta)
5196 {
5197 *errorcodeptr = ERR20;
5198 goto FAILED;
5199 }
5200 *lengthptr += delta;
5201 }
5202
5203 /* This is compiling for real. If there is a set first byte for
5204 the group, and we have not yet set a "required byte", set it. Make
5205 sure there is enough workspace for copying forward references before
5206 doing the copy. */
5207
5208 else
5209 {
5210 if (groupsetfirstchar && reqcharflags < 0)
5211 {
5212 reqchar = firstchar;
5213 reqcharflags = firstcharflags;
5214 }
5215
5216 for (i = 1; i < repeat_min; i++)
5217 {
5218 pcre_uchar *hc;
5219 pcre_uchar *this_hwm = cd->hwm;
5220 memcpy(code, previous, IN_UCHARS(len));
5221
5222 while (cd->hwm > cd->start_workspace + cd->workspace_size -
5223 WORK_SIZE_SAFETY_MARGIN - (this_hwm - save_hwm))
5224 {
5225 int save_offset = save_hwm - cd->start_workspace;
5226 int this_offset = this_hwm - cd->start_workspace;
5227 *errorcodeptr = expand_workspace(cd);
5228 if (*errorcodeptr != 0) goto FAILED;
5229 save_hwm = (pcre_uchar *)cd->start_workspace + save_offset;
5230 this_hwm = (pcre_uchar *)cd->start_workspace + this_offset;
5231 }
5232
5233 for (hc = save_hwm; hc < this_hwm; hc += LINK_SIZE)
5234 {
5235 PUT(cd->hwm, 0, GET(hc, 0) + len);
5236 cd->hwm += LINK_SIZE;
5237 }
5238 save_hwm = this_hwm;
5239 code += len;
5240 }
5241 }
5242 }
5243
5244 if (repeat_max > 0) repeat_max -= repeat_min;
5245 }
5246
5247 /* This code is common to both the zero and non-zero minimum cases. If
5248 the maximum is limited, it replicates the group in a nested fashion,
5249 remembering the bracket starts on a stack. In the case of a zero minimum,
5250 the first one was set up above. In all cases the repeat_max now specifies
5251 the number of additional copies needed. Again, we must remember to
5252 replicate entries on the forward reference list. */
5253
5254 if (repeat_max >= 0)
5255 {
5256 /* In the pre-compile phase, we don't actually do the replication. We
5257 just adjust the length as if we had. For each repetition we must add 1
5258 to the length for BRAZERO and for all but the last repetition we must
5259 add 2 + 2*LINKSIZE to allow for the nesting that occurs. Do some
5260 paranoid checks to avoid integer overflow. The INT64_OR_DOUBLE type is
5261 a 64-bit integer type when available, otherwise double. */
5262
5263 if (lengthptr != NULL && repeat_max > 0)
5264 {
5265 int delta = repeat_max * (length_prevgroup + 1 + 2 + 2*LINK_SIZE) -
5266 2 - 2*LINK_SIZE; /* Last one doesn't nest */
5267 if ((INT64_OR_DOUBLE)repeat_max *
5268 (INT64_OR_DOUBLE)(length_prevgroup + 1 + 2 + 2*LINK_SIZE)
5269 > (INT64_OR_DOUBLE)INT_MAX ||
5270 OFLOW_MAX - *lengthptr < delta)
5271 {
5272 *errorcodeptr = ERR20;
5273 goto FAILED;
5274 }
5275 *lengthptr += delta;
5276 }
5277
5278 /* This is compiling for real */
5279
5280 else for (i = repeat_max - 1; i >= 0; i--)
5281 {
5282 pcre_uchar *hc;
5283 pcre_uchar *this_hwm = cd->hwm;
5284
5285 *code++ = OP_BRAZERO + repeat_type;
5286
5287 /* All but the final copy start a new nesting, maintaining the
5288 chain of brackets outstanding. */
5289
5290 if (i != 0)
5291 {
5292 int offset;
5293 *code++ = OP_BRA;
5294 offset = (bralink == NULL)? 0 : (int)(code - bralink);
5295 bralink = code;
5296 PUTINC(code, 0, offset);
5297 }
5298
5299 memcpy(code, previous, IN_UCHARS(len));
5300
5301 /* Ensure there is enough workspace for forward references before
5302 copying them. */
5303
5304 while (cd->hwm > cd->start_workspace + cd->workspace_size -
5305 WORK_SIZE_SAFETY_MARGIN - (this_hwm - save_hwm))
5306 {
5307 int save_offset = save_hwm - cd->start_workspace;
5308 int this_offset = this_hwm - cd->start_workspace;
5309 *errorcodeptr = expand_workspace(cd);
5310 if (*errorcodeptr != 0) goto FAILED;
5311 save_hwm = (pcre_uchar *)cd->start_workspace + save_offset;
5312 this_hwm = (pcre_uchar *)cd->start_workspace + this_offset;
5313 }
5314
5315 for (hc = save_hwm; hc < this_hwm; hc += LINK_SIZE)
5316 {
5317 PUT(cd->hwm, 0, GET(hc, 0) + len + ((i != 0)? 2+LINK_SIZE : 1));
5318 cd->hwm += LINK_SIZE;
5319 }
5320 save_hwm = this_hwm;
5321 code += len;
5322 }
5323
5324 /* Now chain through the pending brackets, and fill in their length
5325 fields (which are holding the chain links pro tem). */
5326
5327 while (bralink != NULL)
5328 {
5329 int oldlinkoffset;
5330 int offset = (int)(code - bralink + 1);
5331 pcre_uchar *bra = code - offset;
5332 oldlinkoffset = GET(bra, 1);
5333 bralink = (oldlinkoffset == 0)? NULL : bralink - oldlinkoffset;
5334 *code++ = OP_KET;
5335 PUTINC(code, 0, offset);
5336 PUT(bra, 1, offset);
5337 }
5338 }
5339
5340 /* If the maximum is unlimited, set a repeater in the final copy. For
5341 ONCE brackets, that's all we need to do. However, possessively repeated
5342 ONCE brackets can be converted into non-capturing brackets, as the
5343 behaviour of (?:xx)++ is the same as (?>xx)++ and this saves having to
5344 deal with possessive ONCEs specially.
5345
5346 Otherwise, when we are doing the actual compile phase, check to see
5347 whether this group is one that could match an empty string. If so,
5348 convert the initial operator to the S form (e.g. OP_BRA -> OP_SBRA) so
5349 that runtime checking can be done. [This check is also applied to ONCE
5350 groups at runtime, but in a different way.]
5351
5352 Then, if the quantifier was possessive and the bracket is not a
5353 conditional, we convert the BRA code to the POS form, and the KET code to
5354 KETRPOS. (It turns out to be convenient at runtime to detect this kind of
5355 subpattern at both the start and at the end.) The use of special opcodes
5356 makes it possible to reduce greatly the stack usage in pcre_exec(). If
5357 the group is preceded by OP_BRAZERO, convert this to OP_BRAPOSZERO.
5358
5359 Then, if the minimum number of matches is 1 or 0, cancel the possessive
5360 flag so that the default action below, of wrapping everything inside
5361 atomic brackets, does not happen. When the minimum is greater than 1,
5362 there will be earlier copies of the group, and so we still have to wrap
5363 the whole thing. */
5364
5365 else
5366 {
5367 pcre_uchar *ketcode = code - 1 - LINK_SIZE;
5368 pcre_uchar *bracode = ketcode - GET(ketcode, 1);
5369
5370 /* Convert possessive ONCE brackets to non-capturing */
5371
5372 if ((*bracode == OP_ONCE || *bracode == OP_ONCE_NC) &&
5373 possessive_quantifier) *bracode = OP_BRA;
5374
5375 /* For non-possessive ONCE brackets, all we need to do is to
5376 set the KET. */
5377
5378 if (*bracode == OP_ONCE || *bracode == OP_ONCE_NC)
5379 *ketcode = OP_KETRMAX + repeat_type;
5380
5381 /* Handle non-ONCE brackets and possessive ONCEs (which have been
5382 converted to non-capturing above). */
5383
5384 else
5385 {
5386 /* In the compile phase, check for empty string matching. */
5387
5388 if (lengthptr == NULL)
5389 {
5390 pcre_uchar *scode = bracode;
5391 do
5392 {
5393 if (could_be_empty_branch(scode, ketcode, utf, cd))
5394 {
5395 *bracode += OP_SBRA - OP_BRA;
5396 break;
5397 }
5398 scode += GET(scode, 1);
5399 }
5400 while (*scode == OP_ALT);
5401 }
5402
5403 /* Handle possessive quantifiers. */
5404
5405 if (possessive_quantifier)
5406 {
5407 /* For COND brackets, we wrap the whole thing in a possessively
5408 repeated non-capturing bracket, because we have not invented POS
5409 versions of the COND opcodes. Because we are moving code along, we
5410 must ensure that any pending recursive references are updated. */
5411
5412 if (*bracode == OP_COND || *bracode == OP_SCOND)
5413 {
5414 int nlen = (int)(code - bracode);
5415 *code = OP_END;
5416 adjust_recurse(bracode, 1 + LINK_SIZE, utf, cd, save_hwm);
5417 memmove(bracode + 1 + LINK_SIZE, bracode, IN_UCHARS(nlen));
5418 code += 1 + LINK_SIZE;
5419 nlen += 1 + LINK_SIZE;
5420 *bracode = OP_BRAPOS;
5421 *code++ = OP_KETRPOS;
5422 PUTINC(code, 0, nlen);
5423 PUT(bracode, 1, nlen);
5424 }
5425
5426 /* For non-COND brackets, we modify the BRA code and use KETRPOS. */
5427
5428 else
5429 {
5430 *bracode += 1; /* Switch to xxxPOS opcodes */
5431 *ketcode = OP_KETRPOS;
5432 }
5433
5434 /* If the minimum is zero, mark it as possessive, then unset the
5435 possessive flag when the minimum is 0 or 1. */
5436
5437 if (brazeroptr != NULL) *brazeroptr = OP_BRAPOSZERO;
5438 if (repeat_min < 2) possessive_quantifier = FALSE;
5439 }
5440
5441 /* Non-possessive quantifier */
5442
5443 else *ketcode = OP_KETRMAX + repeat_type;
5444 }
5445 }
5446 }
5447
5448 /* If previous is OP_FAIL, it was generated by an empty class [] in
5449 JavaScript mode. The other ways in which OP_FAIL can be generated, that is
5450 by (*FAIL) or (?!) set previous to NULL, which gives a "nothing to repeat"
5451 error above. We can just ignore the repeat in JS case. */
5452
5453 else if (*previous == OP_FAIL) goto END_REPEAT;
5454
5455 /* Else there's some kind of shambles */
5456
5457 else
5458 {
5459 *errorcodeptr = ERR11;
5460 goto FAILED;
5461 }
5462
5463 /* If the character following a repeat is '+', or if certain optimization
5464 tests above succeeded, possessive_quantifier is TRUE. For some opcodes,
5465 there are special alternative opcodes for this case. For anything else, we
5466 wrap the entire repeated item inside OP_ONCE brackets. Logically, the '+'
5467 notation is just syntactic sugar, taken from Sun's Java package, but the
5468 special opcodes can optimize it.
5469
5470 Some (but not all) possessively repeated subpatterns have already been
5471 completely handled in the code just above. For them, possessive_quantifier
5472 is always FALSE at this stage.
5473
5474 Note that the repeated item starts at tempcode, not at previous, which
5475 might be the first part of a string whose (former) last char we repeated.
5476
5477 Possessifying an 'exact' quantifier has no effect, so we can ignore it. But
5478 an 'upto' may follow. We skip over an 'exact' item, and then test the
5479 length of what remains before proceeding. */
5480
5481 if (possessive_quantifier)
5482 {
5483 int len;
5484
5485 if (*tempcode == OP_TYPEEXACT)
5486 tempcode += PRIV(OP_lengths)[*tempcode] +
5487 ((tempcode[1 + IMM2_SIZE] == OP_PROP
5488 || tempcode[1 + IMM2_SIZE] == OP_NOTPROP)? 2 : 0);
5489
5490 else if (*tempcode == OP_EXACT || *tempcode == OP_NOTEXACT)
5491 {
5492 tempcode += PRIV(OP_lengths)[*tempcode];
5493 #ifdef SUPPORT_UTF
5494 if (utf && HAS_EXTRALEN(tempcode[-1]))
5495 tempcode += GET_EXTRALEN(tempcode[-1]);
5496 #endif
5497 }
5498
5499 len = (int)(code - tempcode);
5500 if (len > 0) switch (*tempcode)
5501 {
5502 case OP_STAR: *tempcode = OP_POSSTAR; break;
5503 case OP_PLUS: *tempcode = OP_POSPLUS; break;
5504 case OP_QUERY: *tempcode = OP_POSQUERY; break;
5505 case OP_UPTO: *tempcode = OP_POSUPTO; break;
5506
5507 case OP_STARI: *tempcode = OP_POSSTARI; break;
5508 case OP_PLUSI: *tempcode = OP_POSPLUSI; break;
5509 case OP_QUERYI: *tempcode = OP_POSQUERYI; break;
5510 case OP_UPTOI: *tempcode = OP_POSUPTOI; break;
5511
5512 case OP_NOTSTAR: *tempcode = OP_NOTPOSSTAR; break;
5513 case OP_NOTPLUS: *tempcode = OP_NOTPOSPLUS; break;
5514 case OP_NOTQUERY: *tempcode = OP_NOTPOSQUERY; break;
5515 case OP_NOTUPTO: *tempcode = OP_NOTPOSUPTO; break;
5516
5517 case OP_NOTSTARI: *tempcode = OP_NOTPOSSTARI; break;
5518 case OP_NOTPLUSI: *tempcode = OP_NOTPOSPLUSI; break;
5519 case OP_NOTQUERYI: *tempcode = OP_NOTPOSQUERYI; break;
5520 case OP_NOTUPTOI: *tempcode = OP_NOTPOSUPTOI; break;
5521
5522 case OP_TYPESTAR: *tempcode = OP_TYPEPOSSTAR; break;
5523 case OP_TYPEPLUS: *tempcode = OP_TYPEPOSPLUS; break;
5524 case OP_TYPEQUERY: *tempcode = OP_TYPEPOSQUERY; break;
5525 case OP_TYPEUPTO: *tempcode = OP_TYPEPOSUPTO; break;
5526
5527 /* Because we are moving code along, we must ensure that any
5528 pending recursive references are updated. */
5529
5530 default:
5531 *code = OP_END;
5532 adjust_recurse(tempcode, 1 + LINK_SIZE, utf, cd, save_hwm);
5533 memmove(tempcode + 1 + LINK_SIZE, tempcode, IN_UCHARS(len));
5534 code += 1 + LINK_SIZE;
5535 len += 1 + LINK_SIZE;
5536 tempcode[0] = OP_ONCE;
5537 *code++ = OP_KET;
5538 PUTINC(code, 0, len);
5539 PUT(tempcode, 1, len);
5540 break;
5541 }
5542 }
5543
5544 /* In all case we no longer have a previous item. We also set the
5545 "follows varying string" flag for subsequently encountered reqchars if
5546 it isn't already set and we have just passed a varying length item. */
5547
5548 END_REPEAT:
5549 previous = NULL;
5550 cd->req_varyopt |= reqvary;
5551 break;
5552
5553
5554 /* ===================================================================*/
5555 /* Start of nested parenthesized sub-expression, or comment or lookahead or
5556 lookbehind or option setting or condition or all the other extended
5557 parenthesis forms. */
5558
5559 case CHAR_LEFT_PARENTHESIS:
5560 newoptions = options;
5561 skipbytes = 0;
5562 bravalue = OP_CBRA;
5563 save_hwm = cd->hwm;
5564 reset_bracount = FALSE;
5565
5566 /* First deal with various "verbs" that can be introduced by '*'. */
5567
5568 ptr++;
5569 if (ptr[0] == CHAR_ASTERISK && (ptr[1] == ':'
5570 || (MAX_255(ptr[1]) && ((cd->ctypes[ptr[1]] & ctype_letter) != 0))))
5571 {
5572 int i, namelen;
5573 int arglen = 0;
5574 const char *vn = verbnames;
5575 const pcre_uchar *name = ptr + 1;
5576 const pcre_uchar *arg = NULL;
5577 previous = NULL;
5578 ptr++;
5579 while (MAX_255(*ptr) && (cd->ctypes[*ptr] & ctype_letter) != 0) ptr++;
5580 namelen = (int)(ptr - name);
5581
5582 /* It appears that Perl allows any characters whatsoever, other than
5583 a closing parenthesis, to appear in arguments, so we no longer insist on
5584 letters, digits, and underscores. */
5585
5586 if (*ptr == CHAR_COLON)
5587 {
5588 arg = ++ptr;
5589 while (*ptr != CHAR_NULL && *ptr != CHAR_RIGHT_PARENTHESIS) ptr++;
5590 arglen = (int)(ptr - arg);
5591 if ((unsigned int)arglen > MAX_MARK)
5592 {
5593 *errorcodeptr = ERR75;
5594 goto FAILED;
5595 }
5596 }
5597
5598 if (*ptr != CHAR_RIGHT_PARENTHESIS)
5599 {
5600 *errorcodeptr = ERR60;
5601 goto FAILED;
5602 }
5603
5604 /* Scan the table of verb names */
5605
5606 for (i = 0; i < verbcount; i++)
5607 {
5608 if (namelen == verbs[i].len &&
5609 STRNCMP_UC_C8(name, vn, namelen) == 0)
5610 {
5611 int setverb;
5612
5613 /* Check for open captures before ACCEPT and convert it to
5614 ASSERT_ACCEPT if in an assertion. */
5615
5616 if (verbs[i].op == OP_ACCEPT)
5617 {
5618 open_capitem *oc;
5619 if (arglen != 0)
5620 {
5621 *errorcodeptr = ERR59;
5622 goto FAILED;
5623 }
5624 cd->had_accept = TRUE;
5625 for (oc = cd->open_caps; oc != NULL; oc = oc->next)
5626 {
5627 *code++ = OP_CLOSE;
5628 PUT2INC(code, 0, oc->number);
5629 }
5630 setverb = *code++ =
5631 (cd->assert_depth > 0)? OP_ASSERT_ACCEPT : OP_ACCEPT;
5632
5633 /* Do not set firstchar after *ACCEPT */
5634 if (firstcharflags == REQ_UNSET) firstcharflags = REQ_NONE;
5635 }
5636
5637 /* Handle other cases with/without an argument */
5638
5639 else if (arglen == 0)
5640 {
5641 if (verbs[i].op < 0) /* Argument is mandatory */
5642 {
5643 *errorcodeptr = ERR66;
5644 goto FAILED;
5645 }
5646 setverb = *code++ = verbs[i].op;
5647 }
5648
5649 else
5650 {
5651 if (verbs[i].op_arg < 0) /* Argument is forbidden */
5652 {
5653 *errorcodeptr = ERR59;
5654 goto FAILED;
5655 }
5656 setverb = *code++ = verbs[i].op_arg;
5657 *code++ = arglen;
5658 memcpy(code, arg, IN_UCHARS(arglen));
5659 code += arglen;
5660 *code++ = 0;
5661 }
5662
5663 switch (setverb)
5664 {
5665 case OP_THEN:
5666 case OP_THEN_ARG:
5667 cd->external_flags |= PCRE_HASTHEN;
5668 break;
5669
5670 case OP_PRUNE:
5671 case OP_PRUNE_ARG:
5672 case OP_SKIP:
5673 case OP_SKIP_ARG:
5674 cd->had_pruneorskip = TRUE;
5675 break;
5676 }
5677
5678 break; /* Found verb, exit loop */
5679 }
5680
5681 vn += verbs[i].len + 1;
5682 }
5683
5684 if (i < verbcount) continue; /* Successfully handled a verb */
5685 *errorcodeptr = ERR60; /* Verb not recognized */
5686 goto FAILED;
5687 }
5688
5689 /* Deal with the extended parentheses; all are introduced by '?', and the
5690 appearance of any of them means that this is not a capturing group. */
5691
5692 else if (*ptr == CHAR_QUESTION_MARK)
5693 {
5694 int i, set, unset, namelen;
5695 int *optset;
5696 const pcre_uchar *name;
5697 pcre_uchar *slot;
5698
5699 switch (*(++ptr))
5700 {
5701 case CHAR_NUMBER_SIGN: /* Comment; skip to ket */
5702 ptr++;
5703 while (*ptr != CHAR_NULL && *ptr != CHAR_RIGHT_PARENTHESIS) ptr++;
5704 if (*ptr == CHAR_NULL)
5705 {
5706 *errorcodeptr = ERR18;
5707 goto FAILED;
5708 }
5709 continue;
5710
5711
5712 /* ------------------------------------------------------------ */
5713 case CHAR_VERTICAL_LINE: /* Reset capture count for each branch */
5714 reset_bracount = TRUE;
5715 /* Fall through */
5716
5717 /* ------------------------------------------------------------ */
5718 case CHAR_COLON: /* Non-capturing bracket */
5719 bravalue = OP_BRA;
5720 ptr++;
5721 break;
5722
5723
5724 /* ------------------------------------------------------------ */
5725 case CHAR_LEFT_PARENTHESIS:
5726 bravalue = OP_COND; /* Conditional group */
5727 tempptr = ptr;
5728
5729 /* A condition can be an assertion, a number (referring to a numbered
5730 group), a name (referring to a named group), or 'R', referring to
5731 recursion. R<digits> and R&name are also permitted for recursion tests.
5732
5733 There are several syntaxes for testing a named group: (?(name)) is used
5734 by Python; Perl 5.10 onwards uses (?(<name>) or (?('name')).
5735
5736 There are two unfortunate ambiguities, caused by history. (a) 'R' can
5737 be the recursive thing or the name 'R' (and similarly for 'R' followed
5738 by digits), and (b) a number could be a name that consists of digits.
5739 In both cases, we look for a name first; if not found, we try the other
5740 cases.
5741
5742 For compatibility with auto-callouts, we allow a callout to be
5743 specified before a condition that is an assertion. First, check for the
5744 syntax of a callout; if found, adjust the temporary pointer that is
5745 used to check for an assertion condition. That's all that is needed! */
5746
5747 if (ptr[1] == CHAR_QUESTION_MARK && ptr[2] == CHAR_C)
5748 {
5749 for (i = 3;; i++) if (!IS_DIGIT(ptr[i])) break;
5750 if (ptr[i] == CHAR_RIGHT_PARENTHESIS)
5751 tempptr += i + 1;
5752 }
5753
5754 /* For conditions that are assertions, check the syntax, and then exit
5755 the switch. This will take control down to where bracketed groups,
5756 including assertions, are processed. */
5757
5758 if (tempptr[1] == CHAR_QUESTION_MARK &&
5759 (tempptr[2] == CHAR_EQUALS_SIGN ||
5760 tempptr[2] == CHAR_EXCLAMATION_MARK ||
5761 tempptr[2] == CHAR_LESS_THAN_SIGN))
5762 break;
5763
5764 /* Most other conditions use OP_CREF (a couple change to OP_RREF
5765 below), and all need to skip 1+IMM2_SIZE bytes at the start of the group. */
5766
5767 code[1+LINK_SIZE] = OP_CREF;
5768 skipbytes = 1+IMM2_SIZE;
5769 refsign = -1;
5770
5771 /* Check for a test for recursion in a named group. */
5772
5773 if (ptr[1] == CHAR_R && ptr[2] == CHAR_AMPERSAND)
5774 {
5775 terminator = -1;
5776 ptr += 2;
5777 code[1+LINK_SIZE] = OP_RREF; /* Change the type of test */
5778 }
5779
5780 /* Check for a test for a named group's having been set, using the Perl
5781 syntax (?(<name>) or (?('name') */
5782
5783 else if (ptr[1] == CHAR_LESS_THAN_SIGN)
5784 {
5785 terminator = CHAR_GREATER_THAN_SIGN;
5786 ptr++;
5787 }
5788 else if (ptr[1] == CHAR_APOSTROPHE)
5789 {
5790 terminator = CHAR_APOSTROPHE;
5791 ptr++;
5792 }
5793 else
5794 {
5795 terminator = CHAR_NULL;
5796 if (ptr[1] == CHAR_MINUS || ptr[1] == CHAR_PLUS) refsign = *(++ptr);
5797 }
5798
5799 /* We now expect to read a name; any thing else is an error */
5800
5801 if (!MAX_255(ptr[1]) || (cd->ctypes[ptr[1]] & ctype_word) == 0)
5802 {
5803 ptr += 1; /* To get the right offset */
5804 *errorcodeptr = ERR28;
5805 goto FAILED;
5806 }
5807
5808 /* Read the name, but also get it as a number if it's all digits */
5809
5810 recno = 0;
5811 name = ++ptr;
5812 while (MAX_255(*ptr) && (cd->ctypes[*ptr] & ctype_word) != 0)
5813 {
5814 if (recno >= 0)
5815 recno = (IS_DIGIT(*ptr))? recno * 10 + (int)(*ptr - CHAR_0) : -1;
5816 ptr++;
5817 }
5818 namelen = (int)(ptr - name);
5819
5820 if ((terminator > 0 && *ptr++ != (pcre_uchar)terminator) ||
5821 *ptr++ != CHAR_RIGHT_PARENTHESIS)
5822 {
5823 ptr--; /* Error offset */
5824 *errorcodeptr = ERR26;
5825 goto FAILED;
5826 }
5827
5828 /* Do no further checking in the pre-compile phase. */
5829
5830 if (lengthptr != NULL) break;
5831
5832 /* In the real compile we do the work of looking for the actual
5833 reference. If the string started with "+" or "-" we require the rest to
5834 be digits, in which case recno will be set. */
5835
5836 if (refsign > 0)
5837 {
5838 if (recno <= 0)
5839 {
5840 *errorcodeptr = ERR58;
5841 goto FAILED;
5842 }
5843 recno = (refsign == CHAR_MINUS)?
5844 cd->bracount - recno + 1 : recno +cd->bracount;
5845 if (recno <= 0 || recno > cd->final_bracount)
5846 {
5847 *errorcodeptr = ERR15;
5848 goto FAILED;
5849 }
5850 PUT2(code, 2+LINK_SIZE, recno);
5851 break;
5852 }
5853
5854 /* Otherwise (did not start with "+" or "-"), start by looking for the
5855 name. If we find a name, add one to the opcode to change OP_CREF or
5856 OP_RREF into OP_NCREF or OP_NRREF. These behave exactly the same,
5857 except they record that the reference was originally to a name. The
5858 information is used to check duplicate names. */
5859
5860 slot = cd->name_table;
5861 for (i = 0; i < cd->names_found; i++)
5862 {
5863 if (STRNCMP_UC_UC(name, slot+IMM2_SIZE, namelen) == 0) break;
5864 slot += cd->name_entry_size;
5865 }
5866
5867 /* Found a previous named subpattern */
5868
5869 if (i < cd->names_found)
5870 {
5871 recno = GET2(slot, 0);
5872 PUT2(code, 2+LINK_SIZE, recno);
5873 code[1+LINK_SIZE]++;
5874 }
5875
5876 /* Search the pattern for a forward reference */
5877
5878 else if ((i = find_parens(cd, name, namelen,
5879 (options & PCRE_EXTENDED) != 0, utf)) > 0)
5880 {
5881 PUT2(code, 2+LINK_SIZE, i);
5882 code[1+LINK_SIZE]++;
5883 }
5884
5885 /* If terminator == CHAR_NULL it means that the name followed directly
5886 after the opening parenthesis [e.g. (?(abc)...] and in this case there
5887 are some further alternatives to try. For the cases where terminator !=
5888 0 [things like (?(<name>... or (?('name')... or (?(R&name)... ] we have
5889 now checked all the possibilities, so give an error. */
5890
5891 else if (terminator != CHAR_NULL)
5892 {
5893 *errorcodeptr = ERR15;
5894 goto FAILED;
5895 }
5896
5897 /* Check for (?(R) for recursion. Allow digits after R to specify a
5898 specific group number. */
5899
5900 else if (*name == CHAR_R)
5901 {
5902 recno = 0;
5903 for (i = 1; i < namelen; i++)
5904 {
5905 if (!IS_DIGIT(name[i]))
5906 {
5907 *errorcodeptr = ERR15;
5908 goto FAILED;
5909 }
5910 recno = recno * 10 + name[i] - CHAR_0;
5911 }
5912 if (recno == 0) recno = RREF_ANY;
5913 code[1+LINK_SIZE] = OP_RREF; /* Change test type */
5914 PUT2(code, 2+LINK_SIZE, recno);
5915 }
5916
5917 /* Similarly, check for the (?(DEFINE) "condition", which is always
5918 false. */
5919
5920 else if (namelen == 6 && STRNCMP_UC_C8(name, STRING_DEFINE, 6) == 0)
5921 {
5922 code[1+LINK_SIZE] = OP_DEF;
5923 skipbytes = 1;
5924 }
5925
5926 /* Check for the "name" actually being a subpattern number. We are
5927 in the second pass here, so final_bracount is set. */
5928
5929 else if (recno > 0 && recno <= cd->final_bracount)
5930 {
5931 PUT2(code, 2+LINK_SIZE, recno);
5932 }
5933
5934 /* Either an unidentified subpattern, or a reference to (?(0) */
5935
5936 else
5937 {
5938 *errorcodeptr = (recno == 0)? ERR35: ERR15;
5939 goto FAILED;
5940 }
5941 break;
5942
5943
5944 /* ------------------------------------------------------------ */
5945 case CHAR_EQUALS_SIGN: /* Positive lookahead */
5946 bravalue = OP_ASSERT;
5947 cd->assert_depth += 1;
5948 ptr++;
5949 break;
5950
5951
5952 /* ------------------------------------------------------------ */
5953 case CHAR_EXCLAMATION_MARK: /* Negative lookahead */
5954 ptr++;
5955 if (*ptr == CHAR_RIGHT_PARENTHESIS) /* Optimize (?!) */
5956 {
5957 *code++ = OP_FAIL;
5958 previous = NULL;
5959 continue;
5960 }
5961 bravalue = OP_ASSERT_NOT;
5962 cd->assert_depth += 1;
5963 break;
5964
5965
5966 /* ------------------------------------------------------------ */
5967 case CHAR_LESS_THAN_SIGN: /* Lookbehind or named define */
5968 switch (ptr[1])
5969 {
5970 case CHAR_EQUALS_SIGN: /* Positive lookbehind */
5971 bravalue = OP_ASSERTBACK;
5972 cd->assert_depth += 1;
5973 ptr += 2;
5974 break;
5975
5976 case CHAR_EXCLAMATION_MARK: /* Negative lookbehind */
5977 bravalue = OP_ASSERTBACK_NOT;
5978 cd->assert_depth += 1;
5979 ptr += 2;
5980 break;
5981
5982 default: /* Could be name define, else bad */
5983 if (MAX_255(ptr[1]) && (cd->ctypes[ptr[1]] & ctype_word) != 0)
5984 goto DEFINE_NAME;
5985 ptr++; /* Correct offset for error */
5986 *errorcodeptr = ERR24;
5987 goto FAILED;
5988 }
5989 break;
5990
5991
5992 /* ------------------------------------------------------------ */
5993 case CHAR_GREATER_THAN_SIGN: /* One-time brackets */
5994 bravalue = OP_ONCE;
5995 ptr++;
5996 break;
5997
5998
5999 /* ------------------------------------------------------------ */
6000 case CHAR_C: /* Callout - may be followed by digits; */
6001 previous_callout = code; /* Save for later completion */
6002 after_manual_callout = 1; /* Skip one item before completing */
6003 *code++ = OP_CALLOUT;
6004 {
6005 int n = 0;
6006 ptr++;
6007 while(IS_DIGIT(*ptr))
6008 n = n * 10 + *ptr++ - CHAR_0;
6009 if (*ptr != CHAR_RIGHT_PARENTHESIS)
6010 {
6011 *errorcodeptr = ERR39;
6012 goto FAILED;
6013 }
6014 if (n > 255)
6015 {
6016 *errorcodeptr = ERR38;
6017 goto FAILED;
6018 }
6019 *code++ = n;
6020 PUT(code, 0, (int)(ptr - cd->start_pattern + 1)); /* Pattern offset */
6021 PUT(code, LINK_SIZE, 0); /* Default length */
6022 code += 2 * LINK_SIZE;
6023 }
6024 previous = NULL;
6025 continue;
6026
6027
6028 /* ------------------------------------------------------------ */
6029 case CHAR_P: /* Python-style named subpattern handling */
6030 if (*(++ptr) == CHAR_EQUALS_SIGN ||
6031 *ptr == CHAR_GREATER_THAN_SIGN) /* Reference or recursion */
6032 {
6033 is_recurse = *ptr == CHAR_GREATER_THAN_SIGN;
6034 terminator = CHAR_RIGHT_PARENTHESIS;
6035 goto NAMED_REF_OR_RECURSE;
6036 }
6037 else if (*ptr != CHAR_LESS_THAN_SIGN) /* Test for Python-style defn */
6038 {
6039 *errorcodeptr = ERR41;
6040 goto FAILED;
6041 }
6042 /* Fall through to handle (?P< as (?< is handled */
6043
6044
6045 /* ------------------------------------------------------------ */
6046 DEFINE_NAME: /* Come here from (?< handling */
6047 case CHAR_APOSTROPHE:
6048 {
6049 terminator = (*ptr == CHAR_LESS_THAN_SIGN)?
6050 CHAR_GREATER_THAN_SIGN : CHAR_APOSTROPHE;
6051 name = ++ptr;
6052
6053 while (MAX_255(*ptr) && (cd->ctypes[*ptr] & ctype_word) != 0) ptr++;
6054 namelen = (int)(ptr - name);
6055
6056 /* In the pre-compile phase, just do a syntax check. */
6057
6058 if (lengthptr != NULL)
6059 {
6060 if (*ptr != (pcre_uchar)terminator)
6061 {
6062 *errorcodeptr = ERR42;
6063 goto FAILED;
6064 }
6065 if (cd->names_found >= MAX_NAME_COUNT)
6066 {
6067 *errorcodeptr = ERR49;
6068 goto FAILED;
6069 }
6070 if (namelen + IMM2_SIZE + 1 > cd->name_entry_size)
6071 {
6072 cd->name_entry_size = namelen + IMM2_SIZE + 1;
6073 if (namelen > MAX_NAME_SIZE)
6074 {
6075 *errorcodeptr = ERR48;
6076 goto FAILED;
6077 }
6078 }
6079 }
6080
6081 /* In the real compile, create the entry in the table, maintaining
6082 alphabetical order. Duplicate names for different numbers are
6083 permitted only if PCRE_DUPNAMES is set. Duplicate names for the same
6084 number are always OK. (An existing number can be re-used if (?|
6085 appears in the pattern.) In either event, a duplicate name results in
6086 a duplicate entry in the table, even if the number is the same. This
6087 is because the number of names, and hence the table size, is computed
6088 in the pre-compile, and it affects various numbers and pointers which
6089 would all have to be modified, and the compiled code moved down, if
6090 duplicates with the same number were omitted from the table. This
6091 doesn't seem worth the hassle. However, *different* names for the
6092 same number are not permitted. */
6093
6094 else
6095 {
6096 BOOL dupname = FALSE;
6097 slot = cd->name_table;
6098
6099 for (i = 0; i < cd->names_found; i++)
6100 {
6101 int crc = memcmp(name, slot+IMM2_SIZE, IN_UCHARS(namelen));
6102 if (crc == 0)
6103 {
6104 if (slot[IMM2_SIZE+namelen] == 0)
6105 {
6106 if (GET2(slot, 0) != cd->bracount + 1 &&
6107 (options & PCRE_DUPNAMES) == 0)
6108 {
6109 *errorcodeptr = ERR43;
6110 goto FAILED;
6111 }
6112 else dupname = TRUE;
6113 }
6114 else crc = -1; /* Current name is a substring */
6115 }
6116
6117 /* Make space in the table and break the loop for an earlier
6118 name. For a duplicate or later name, carry on. We do this for
6119 duplicates so that in the simple case (when ?(| is not used) they
6120 are in order of their numbers. */
6121
6122 if (crc < 0)
6123 {
6124 memmove(slot + cd->name_entry_size, slot,
6125 IN_UCHARS((cd->names_found - i) * cd->name_entry_size));
6126 break;
6127 }
6128
6129 /* Continue the loop for a later or duplicate name */
6130
6131 slot += cd->name_entry_size;
6132 }
6133
6134 /* For non-duplicate names, check for a duplicate number before
6135 adding the new name. */
6136
6137 if (!dupname)
6138 {
6139 pcre_uchar *cslot = cd->name_table;
6140 for (i = 0; i < cd->names_found; i++)
6141 {
6142 if (cslot != slot)
6143 {
6144 if (GET2(cslot, 0) == cd->bracount + 1)
6145 {
6146 *errorcodeptr = ERR65;
6147 goto FAILED;
6148 }
6149 }
6150 else i--;
6151 cslot += cd->name_entry_size;
6152 }
6153 }
6154
6155 PUT2(slot, 0, cd->bracount + 1);
6156 memcpy(slot + IMM2_SIZE, name, IN_UCHARS(namelen));
6157 slot[IMM2_SIZE + namelen] = 0;
6158 }
6159 }
6160
6161 /* In both pre-compile and compile, count the number of names we've
6162 encountered. */
6163
6164 cd->names_found++;
6165 ptr++; /* Move past > or ' */
6166 goto NUMBERED_GROUP;
6167
6168
6169 /* ------------------------------------------------------------ */
6170 case CHAR_AMPERSAND: /* Perl recursion/subroutine syntax */
6171 terminator = CHAR_RIGHT_PARENTHESIS;
6172 is_recurse = TRUE;
6173 /* Fall through */
6174
6175 /* We come here from the Python syntax above that handles both
6176 references (?P=name) and recursion (?P>name), as well as falling
6177 through from the Perl recursion syntax (?&name). We also come here from
6178 the Perl \k<name> or \k'name' back reference syntax and the \k{name}
6179 .NET syntax, and the Oniguruma \g<...> and \g'...' subroutine syntax. */
6180
6181 NAMED_REF_OR_RECURSE:
6182 name = ++ptr;
6183 while (MAX_255(*ptr) && (cd->ctypes[*ptr] & ctype_word) != 0) ptr++;
6184 namelen = (int)(ptr - name);
6185
6186 /* In the pre-compile phase, do a syntax check. We used to just set
6187 a dummy reference number, because it was not used in the first pass.
6188 However, with the change of recursive back references to be atomic,
6189 we have to look for the number so that this state can be identified, as
6190 otherwise the incorrect length is computed. If it's not a backwards
6191 reference, the dummy number will do. */
6192
6193 if (lengthptr != NULL)
6194 {
6195 const pcre_uchar *temp;
6196
6197 if (namelen == 0)
6198 {
6199 *errorcodeptr = ERR62;
6200 goto FAILED;
6201 }
6202 if (*ptr != (pcre_uchar)terminator)
6203 {
6204 *errorcodeptr = ERR42;
6205 goto FAILED;
6206 }
6207 if (namelen > MAX_NAME_SIZE)
6208 {
6209 *errorcodeptr = ERR48;
6210 goto FAILED;
6211 }
6212
6213 /* The name table does not exist in the first pass, so we cannot
6214 do a simple search as in the code below. Instead, we have to scan the
6215 pattern to find the number. It is important that we scan it only as
6216 far as we have got because the syntax of named subpatterns has not
6217 been checked for the rest of the pattern, and find_parens() assumes
6218 correct syntax. In any case, it's a waste of resources to scan
6219 further. We stop the scan at the current point by temporarily
6220 adjusting the value of cd->endpattern. */
6221
6222 temp = cd->end_pattern;
6223 cd->end_pattern = ptr;
6224 recno = find_parens(cd, name, namelen,
6225 (options & PCRE_EXTENDED) != 0, utf);
6226 cd->end_pattern = temp;
6227 if (recno < 0) recno = 0; /* Forward ref; set dummy number */
6228 }
6229
6230 /* In the real compile, seek the name in the table. We check the name
6231 first, and then check that we have reached the end of the name in the
6232 table. That way, if the name that is longer than any in the table,
6233 the comparison will fail without reading beyond the table entry. */
6234
6235 else
6236 {
6237 slot = cd->name_table;
6238 for (i = 0; i < cd->names_found; i++)
6239 {
6240 if (STRNCMP_UC_UC(name, slot+IMM2_SIZE, namelen) == 0 &&
6241 slot[IMM2_SIZE+namelen] == 0)
6242 break;
6243 slot += cd->name_entry_size;
6244 }
6245
6246 if (i < cd->names_found) /* Back reference */
6247 {
6248 recno = GET2(slot, 0);
6249 }
6250 else if ((recno = /* Forward back reference */
6251 find_parens(cd, name, namelen,
6252 (options & PCRE_EXTENDED) != 0, utf)) <= 0)
6253 {
6254 *errorcodeptr = ERR15;
6255 goto FAILED;
6256 }
6257 }
6258
6259 /* In both phases, we can now go to the code than handles numerical
6260 recursion or backreferences. */
6261
6262 if (is_recurse) goto HANDLE_RECURSION;
6263 else goto HANDLE_REFERENCE;
6264
6265
6266 /* ------------------------------------------------------------ */
6267 case CHAR_R: /* Recursion */
6268 ptr++; /* Same as (?0) */
6269 /* Fall through */
6270
6271
6272 /* ------------------------------------------------------------ */
6273 case CHAR_MINUS: case CHAR_PLUS: /* Recursion or subroutine */
6274 case CHAR_0: case CHAR_1: case CHAR_2: case CHAR_3: case CHAR_4:
6275 case CHAR_5: case CHAR_6: case CHAR_7: case CHAR_8: case CHAR_9:
6276 {
6277 const pcre_uchar *called;
6278 terminator = CHAR_RIGHT_PARENTHESIS;
6279
6280 /* Come here from the \g<...> and \g'...' code (Oniguruma
6281 compatibility). However, the syntax has been checked to ensure that
6282 the ... are a (signed) number, so that neither ERR63 nor ERR29 will
6283 be called on this path, nor with the jump to OTHER_CHAR_AFTER_QUERY
6284 ever be taken. */
6285
6286 HANDLE_NUMERICAL_RECURSION:
6287
6288 if ((refsign = *ptr) == CHAR_PLUS)
6289 {
6290 ptr++;
6291 if (!IS_DIGIT(*ptr))
6292 {
6293 *errorcodeptr = ERR63;
6294 goto FAILED;
6295 }
6296 }
6297 else if (refsign == CHAR_MINUS)
6298 {
6299 if (!IS_DIGIT(ptr[1]))
6300 goto OTHER_CHAR_AFTER_QUERY;
6301 ptr++;
6302 }
6303
6304 recno = 0;
6305 while(IS_DIGIT(*ptr))
6306 recno = recno * 10 + *ptr++ - CHAR_0;
6307
6308 if (*ptr != (pcre_uchar)terminator)
6309 {
6310 *errorcodeptr = ERR29;
6311 goto FAILED;
6312 }
6313
6314 if (refsign == CHAR_MINUS)
6315 {
6316 if (recno == 0)
6317 {
6318 *errorcodeptr = ERR58;
6319 goto FAILED;
6320 }
6321 recno = cd->bracount - recno + 1;
6322 if (recno <= 0)
6323 {
6324 *errorcodeptr = ERR15;
6325 goto FAILED;
6326 }
6327 }
6328 else if (refsign == CHAR_PLUS)
6329 {
6330 if (recno == 0)
6331 {
6332 *errorcodeptr = ERR58;
6333 goto FAILED;
6334 }
6335 recno += cd->bracount;
6336 }
6337
6338 /* Come here from code above that handles a named recursion */
6339
6340 HANDLE_RECURSION:
6341
6342 previous = code;
6343 called = cd->start_code;
6344
6345 /* When we are actually compiling, find the bracket that is being
6346 referenced. Temporarily end the regex in case it doesn't exist before
6347 this point. If we end up with a forward reference, first check that
6348 the bracket does occur later so we can give the error (and position)
6349 now. Then remember this forward reference in the workspace so it can
6350 be filled in at the end. */
6351
6352 if (lengthptr == NULL)
6353 {
6354 *code = OP_END;
6355 if (recno != 0)
6356 called = PRIV(find_bracket)(cd->start_code, utf, recno);
6357
6358 /* Forward reference */
6359
6360 if (called == NULL)
6361 {
6362 if (find_parens(cd, NULL, recno,
6363 (options & PCRE_EXTENDED) != 0, utf) < 0)
6364 {
6365 *errorcodeptr = ERR15;
6366 goto FAILED;
6367 }
6368
6369 /* Fudge the value of "called" so that when it is inserted as an
6370 offset below, what it actually inserted is the reference number
6371 of the group. Then remember the forward reference. */
6372
6373 called = cd->start_code + recno;
6374 if (cd->hwm >= cd->start_workspace + cd->workspace_size -
6375 WORK_SIZE_SAFETY_MARGIN)
6376 {
6377 *errorcodeptr = expand_workspace(cd);
6378 if (*errorcodeptr != 0) goto FAILED;
6379 }
6380 PUTINC(cd->hwm, 0, (int)(code + 1 - cd->start_code));
6381 }
6382
6383 /* If not a forward reference, and the subpattern is still open,
6384 this is a recursive call. We check to see if this is a left
6385 recursion that could loop for ever, and diagnose that case. We
6386 must not, however, do this check if we are in a conditional
6387 subpattern because the condition might be testing for recursion in
6388 a pattern such as /(?(R)a+|(?R)b)/, which is perfectly valid.
6389 Forever loops are also detected at runtime, so those that occur in
6390 conditional subpatterns will be picked up then. */
6391
6392 else if (GET(called, 1) == 0 && cond_depth <= 0 &&
6393 could_be_empty(called, code, bcptr, utf, cd))
6394 {
6395 *errorcodeptr = ERR40;
6396 goto FAILED;
6397 }
6398 }
6399
6400 /* Insert the recursion/subroutine item. It does not have a set first
6401 character (relevant if it is repeated, because it will then be
6402 wrapped with ONCE brackets). */
6403
6404 *code = OP_RECURSE;
6405 PUT(code, 1, (int)(called - cd->start_code));
6406 code += 1 + LINK_SIZE;
6407 groupsetfirstchar = FALSE;
6408 }
6409
6410 /* Can't determine a first byte now */
6411
6412 if (firstcharflags == REQ_UNSET) firstcharflags = REQ_NONE;
6413 continue;
6414
6415
6416 /* ------------------------------------------------------------ */
6417 default: /* Other characters: check option setting */
6418 OTHER_CHAR_AFTER_QUERY:
6419 set = unset = 0;
6420 optset = &set;
6421
6422 while (*ptr != CHAR_RIGHT_PARENTHESIS && *ptr != CHAR_COLON)
6423 {
6424 switch (*ptr++)
6425 {
6426 case CHAR_MINUS: optset = &unset; break;
6427
6428 case CHAR_J: /* Record that it changed in the external options */
6429 *optset |= PCRE_DUPNAMES;
6430 cd->external_flags |= PCRE_JCHANGED;
6431 break;
6432
6433 case CHAR_i: *optset |= PCRE_CASELESS; break;
6434 case CHAR_m: *optset |= PCRE_MULTILINE; break;
6435 case CHAR_s: *optset |= PCRE_DOTALL; break;
6436 case CHAR_x: *optset |= PCRE_EXTENDED; break;
6437 case CHAR_U: *optset |= PCRE_UNGREEDY; break;
6438 case CHAR_X: *optset |= PCRE_EXTRA; break;
6439
6440 default: *errorcodeptr = ERR12;
6441 ptr--; /* Correct the offset */
6442 goto FAILED;
6443 }
6444 }
6445
6446 /* Set up the changed option bits, but don't change anything yet. */
6447
6448 newoptions = (options | set) & (~unset);
6449
6450 /* If the options ended with ')' this is not the start of a nested
6451 group with option changes, so the options change at this level. If this
6452 item is right at the start of the pattern, the options can be
6453 abstracted and made external in the pre-compile phase, and ignored in
6454 the compile phase. This can be helpful when matching -- for instance in
6455 caseless checking of required bytes.
6456
6457 If the code pointer is not (cd->start_code + 1 + LINK_SIZE), we are
6458 definitely *not* at the start of the pattern because something has been
6459 compiled. In the pre-compile phase, however, the code pointer can have
6460 that value after the start, because it gets reset as code is discarded
6461 during the pre-compile. However, this can happen only at top level - if
6462 we are within parentheses, the starting BRA will still be present. At
6463 any parenthesis level, the length value can be used to test if anything
6464 has been compiled at that level. Thus, a test for both these conditions
6465 is necessary to ensure we correctly detect the start of the pattern in
6466 both phases.
6467
6468 If we are not at the pattern start, reset the greedy defaults and the
6469 case value for firstchar and reqchar. */
6470
6471 if (*ptr == CHAR_RIGHT_PARENTHESIS)
6472 {
6473 if (code == cd->start_code + 1 + LINK_SIZE &&
6474 (lengthptr == NULL || *lengthptr == 2 + 2*LINK_SIZE))
6475 {
6476 cd->external_options = newoptions;
6477 }
6478 else
6479 {
6480 greedy_default = ((newoptions & PCRE_UNGREEDY) != 0);
6481 greedy_non_default = greedy_default ^ 1;
6482 req_caseopt = ((newoptions & PCRE_CASELESS) != 0)? REQ_CASELESS:0;
6483 }
6484
6485 /* Change options at this level, and pass them back for use
6486 in subsequent branches. */
6487
6488 *optionsptr = options = newoptions;
6489 previous = NULL; /* This item can't be repeated */
6490 continue; /* It is complete */
6491 }
6492
6493 /* If the options ended with ':' we are heading into a nested group
6494 with possible change of options. Such groups are non-capturing and are
6495 not assertions of any kind. All we need to do is skip over the ':';
6496 the newoptions value is handled below. */
6497
6498 bravalue = OP_BRA;
6499 ptr++;
6500 } /* End of switch for character following (? */
6501 } /* End of (? handling */
6502
6503 /* Opening parenthesis not followed by '*' or '?'. If PCRE_NO_AUTO_CAPTURE
6504 is set, all unadorned brackets become non-capturing and behave like (?:...)
6505 brackets. */
6506
6507 else if ((options & PCRE_NO_AUTO_CAPTURE) != 0)
6508 {
6509 bravalue = OP_BRA;
6510 }
6511
6512 /* Else we have a capturing group. */
6513
6514 else
6515 {
6516 NUMBERED_GROUP:
6517 cd->bracount += 1;
6518 PUT2(code, 1+LINK_SIZE, cd->bracount);
6519 skipbytes = IMM2_SIZE;
6520 }
6521
6522 /* Process nested bracketed regex. Assertions used not to be repeatable,
6523 but this was changed for Perl compatibility, so all kinds can now be
6524 repeated. We copy code into a non-register variable (tempcode) in order to
6525 be able to pass its address because some compilers complain otherwise. */
6526
6527 previous = code; /* For handling repetition */
6528 *code = bravalue;
6529 tempcode = code;
6530 tempreqvary = cd->req_varyopt; /* Save value before bracket */
6531 tempbracount = cd->bracount; /* Save value before bracket */
6532 length_prevgroup = 0; /* Initialize for pre-compile phase */
6533
6534 if (!compile_regex(
6535 newoptions, /* The complete new option state */
6536 &tempcode, /* Where to put code (updated) */
6537 &ptr, /* Input pointer (updated) */
6538 errorcodeptr, /* Where to put an error message */
6539 (bravalue == OP_ASSERTBACK ||
6540 bravalue == OP_ASSERTBACK_NOT), /* TRUE if back assert */
6541 reset_bracount, /* True if (?| group */
6542 skipbytes, /* Skip over bracket number */
6543 cond_depth +
6544 ((bravalue == OP_COND)?1:0), /* Depth of condition subpatterns */
6545 &subfirstchar, /* For possible first char */
6546 &subfirstcharflags,
6547 &subreqchar, /* For possible last char */
6548 &subreqcharflags,
6549 bcptr, /* Current branch chain */
6550 cd, /* Tables block */
6551 (lengthptr == NULL)? NULL : /* Actual compile phase */
6552 &length_prevgroup /* Pre-compile phase */
6553 ))
6554 goto FAILED;
6555
6556 /* If this was an atomic group and there are no capturing groups within it,
6557 generate OP_ONCE_NC instead of OP_ONCE. */
6558
6559 if (bravalue == OP_ONCE && cd->bracount <= tempbracount)
6560 *code = OP_ONCE_NC;
6561
6562 if (bravalue >= OP_ASSERT && bravalue <= OP_ASSERTBACK_NOT)
6563 cd->assert_depth -= 1;
6564
6565 /* At the end of compiling, code is still pointing to the start of the
6566 group, while tempcode has been updated to point past the end of the group.
6567 The pattern pointer (ptr) is on the bracket.
6568
6569 If this is a conditional bracket, check that there are no more than
6570 two branches in the group, or just one if it's a DEFINE group. We do this
6571 in the real compile phase, not in the pre-pass, where the whole group may
6572 not be available. */
6573
6574 if (bravalue == OP_COND && lengthptr == NULL)
6575 {
6576 pcre_uchar *tc = code;
6577 int condcount = 0;
6578
6579 do {
6580 condcount++;
6581 tc += GET(tc,1);
6582 }
6583 while (*tc != OP_KET);
6584
6585 /* A DEFINE group is never obeyed inline (the "condition" is always
6586 false). It must have only one branch. */
6587
6588 if (code[LINK_SIZE+1] == OP_DEF)
6589 {
6590 if (condcount > 1)
6591 {
6592 *errorcodeptr = ERR54;
6593 goto FAILED;
6594 }
6595 bravalue = OP_DEF; /* Just a flag to suppress char handling below */
6596 }
6597
6598 /* A "normal" conditional group. If there is just one branch, we must not
6599 make use of its firstchar or reqchar, because this is equivalent to an
6600 empty second branch. */
6601
6602 else
6603 {
6604 if (condcount > 2)
6605 {
6606 *errorcodeptr = ERR27;
6607 goto FAILED;
6608 }
6609 if (condcount == 1) subfirstcharflags = subreqcharflags = REQ_NONE;
6610 }
6611 }
6612
6613 /* Error if hit end of pattern */
6614
6615 if (*ptr != CHAR_RIGHT_PARENTHESIS)
6616 {
6617 *errorcodeptr = ERR14;
6618 goto FAILED;
6619 }
6620
6621 /* In the pre-compile phase, update the length by the length of the group,
6622 less the brackets at either end. Then reduce the compiled code to just a
6623 set of non-capturing brackets so that it doesn't use much memory if it is
6624 duplicated by a quantifier.*/
6625
6626 if (lengthptr != NULL)
6627 {
6628 if (OFLOW_MAX - *lengthptr < length_prevgroup - 2 - 2*LINK_SIZE)
6629 {
6630 *errorcodeptr = ERR20;
6631 goto FAILED;
6632 }
6633 *lengthptr += length_prevgroup - 2 - 2*LINK_SIZE;
6634 code++; /* This already contains bravalue */
6635 PUTINC(code, 0, 1 + LINK_SIZE);
6636 *code++ = OP_KET;
6637 PUTINC(code, 0, 1 + LINK_SIZE);
6638 break; /* No need to waste time with special character handling */
6639 }
6640