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

Contents of /code/trunk/pcre_compile.c

Parent Directory Parent Directory | Revision Log Revision Log


Revision 1045 - (show annotations)
Sun Sep 23 16:50:00 2012 UTC (6 years, 11 months ago) by ph10
File MIME type: text/plain
File size: 266900 byte(s)
Error occurred while calculating annotation data.
Update character class handling to use new character case information; rework 
\h, \H, \v, and \V to use the same apparatus with centrally defined lists.
1 /*************************************************
2 * Perl-Compatible Regular Expressions *
3 *************************************************/
4
5 /* PCRE is a library of functions to support regular expressions whose syntax
6 and semantics are as close as possible to those of the Perl 5 language.
7
8 Written by Philip Hazel
9 Copyright (c) 1997-2012 University of Cambridge
10
11 -----------------------------------------------------------------------------
12 Redistribution and use in source and binary forms, with or without
13 modification, are permitted provided that the following conditions are met:
14
15 * Redistributions of source code must retain the above copyright notice,
16 this list of conditions and the following disclaimer.
17
18 * Redistributions in binary form must reproduce the above copyright
19 notice, this list of conditions and the following disclaimer in the
20 documentation and/or other materials provided with the distribution.
21
22 * Neither the name of the University of Cambridge nor the names of its
23 contributors may be used to endorse or promote products derived from
24 this software without specific prior written permission.
25
26 THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
27 AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28 IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29 ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE
30 LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
31 CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
32 SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
33 INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
34 CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
35 ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
36 POSSIBILITY OF SUCH DAMAGE.
37 -----------------------------------------------------------------------------
38 */
39
40
41 /* This module contains the external function pcre_compile(), along with
42 supporting internal functions that are not used by other modules. */
43
44
45 #ifdef HAVE_CONFIG_H
46 #include "config.h"
47 #endif
48
49 #define NLBLOCK cd /* Block containing newline information */
50 #define PSSTART start_pattern /* Field containing processed string start */
51 #define PSEND end_pattern /* Field containing processed string end */
52
53 #include "pcre_internal.h"
54
55
56 /* When PCRE_DEBUG is defined, we need the pcre(16)_printint() function, which
57 is also used by pcretest. PCRE_DEBUG is not defined when building a production
58 library. We do not need to select pcre16_printint.c specially, because the
59 COMPILE_PCREx macro will already be appropriately set. */
60
61 #ifdef PCRE_DEBUG
62 /* pcre_printint.c should not include any headers */
63 #define PCRE_INCLUDED
64 #include "pcre_printint.c"
65 #undef PCRE_INCLUDED
66 #endif
67
68
69 /* Macro for setting individual bits in class bitmaps. */
70
71 #define SETBIT(a,b) a[(b)/8] |= (1 << ((b)&7))
72
73 /* Maximum length value to check against when making sure that the integer that
74 holds the compiled pattern length does not overflow. We make it a bit less than
75 INT_MAX to allow for adding in group terminating bytes, so that we don't have
76 to check them every time. */
77
78 #define OFLOW_MAX (INT_MAX - 20)
79
80 /* Definitions to allow mutual recursion */
81
82 static int
83 add_list_to_class(pcre_uint8 *, pcre_uchar **, int, compile_data *,
84 const pcre_uint32 *, unsigned int);
85
86 static BOOL
87 compile_regex(int, pcre_uchar **, const pcre_uchar **, int *, BOOL, BOOL,
88 int, int, int *, int *, branch_chain *, compile_data *, int *);
89
90
91
92 /*************************************************
93 * Code parameters and static tables *
94 *************************************************/
95
96 /* This value specifies the size of stack workspace that is used during the
97 first pre-compile phase that determines how much memory is required. The regex
98 is partly compiled into this space, but the compiled parts are discarded as
99 soon as they can be, so that hopefully there will never be an overrun. The code
100 does, however, check for an overrun. The largest amount I've seen used is 218,
101 so this number is very generous.
102
103 The same workspace is used during the second, actual compile phase for
104 remembering forward references to groups so that they can be filled in at the
105 end. Each entry in this list occupies LINK_SIZE bytes, so even when LINK_SIZE
106 is 4 there is plenty of room for most patterns. However, the memory can get
107 filled up by repetitions of forward references, for example patterns like
108 /(?1){0,1999}(b)/, and one user did hit the limit. The code has been changed so
109 that the workspace is expanded using malloc() in this situation. The value
110 below is therefore a minimum, and we put a maximum on it for safety. The
111 minimum is now also defined in terms of LINK_SIZE so that the use of malloc()
112 kicks in at the same number of forward references in all cases. */
113
114 #define COMPILE_WORK_SIZE (2048*LINK_SIZE)
115 #define COMPILE_WORK_SIZE_MAX (100*COMPILE_WORK_SIZE)
116
117 /* The overrun tests check for a slightly smaller size so that they detect the
118 overrun before it actually does run off the end of the data block. */
119
120 #define WORK_SIZE_SAFETY_MARGIN (100)
121
122 /* Private flags added to firstchar and reqchar. */
123
124 #define REQ_CASELESS 0x10000000l /* Indicates caselessness */
125 #define REQ_VARY 0x20000000l /* Reqchar followed non-literal item */
126
127 /* Repeated character flags. */
128
129 #define UTF_LENGTH 0x10000000l /* The char contains its length. */
130
131 /* Table for handling escaped characters in the range '0'-'z'. Positive returns
132 are simple data values; negative values are for special things like \d and so
133 on. Zero means further processing is needed (for things like \x), or the escape
134 is invalid. */
135
136 #ifndef EBCDIC
137
138 /* This is the "normal" table for ASCII systems or for EBCDIC systems running
139 in UTF-8 mode. */
140
141 static const short int escapes[] = {
142 0, 0,
143 0, 0,
144 0, 0,
145 0, 0,
146 0, 0,
147 CHAR_COLON, CHAR_SEMICOLON,
148 CHAR_LESS_THAN_SIGN, CHAR_EQUALS_SIGN,
149 CHAR_GREATER_THAN_SIGN, CHAR_QUESTION_MARK,
150 CHAR_COMMERCIAL_AT, -ESC_A,
151 -ESC_B, -ESC_C,
152 -ESC_D, -ESC_E,
153 0, -ESC_G,
154 -ESC_H, 0,
155 0, -ESC_K,
156 0, 0,
157 -ESC_N, 0,
158 -ESC_P, -ESC_Q,
159 -ESC_R, -ESC_S,
160 0, 0,
161 -ESC_V, -ESC_W,
162 -ESC_X, 0,
163 -ESC_Z, CHAR_LEFT_SQUARE_BRACKET,
164 CHAR_BACKSLASH, CHAR_RIGHT_SQUARE_BRACKET,
165 CHAR_CIRCUMFLEX_ACCENT, CHAR_UNDERSCORE,
166 CHAR_GRAVE_ACCENT, 7,
167 -ESC_b, 0,
168 -ESC_d, ESC_e,
169 ESC_f, 0,
170 -ESC_h, 0,
171 0, -ESC_k,
172 0, 0,
173 ESC_n, 0,
174 -ESC_p, 0,
175 ESC_r, -ESC_s,
176 ESC_tee, 0,
177 -ESC_v, -ESC_w,
178 0, 0,
179 -ESC_z
180 };
181
182 #else
183
184 /* This is the "abnormal" table for EBCDIC systems without UTF-8 support. */
185
186 static const short int escapes[] = {
187 /* 48 */ 0, 0, 0, '.', '<', '(', '+', '|',
188 /* 50 */ '&', 0, 0, 0, 0, 0, 0, 0,
189 /* 58 */ 0, 0, '!', '$', '*', ')', ';', '~',
190 /* 60 */ '-', '/', 0, 0, 0, 0, 0, 0,
191 /* 68 */ 0, 0, '|', ',', '%', '_', '>', '?',
192 /* 70 */ 0, 0, 0, 0, 0, 0, 0, 0,
193 /* 78 */ 0, '`', ':', '#', '@', '\'', '=', '"',
194 /* 80 */ 0, 7, -ESC_b, 0, -ESC_d, ESC_e, ESC_f, 0,
195 /* 88 */-ESC_h, 0, 0, '{', 0, 0, 0, 0,
196 /* 90 */ 0, 0, -ESC_k, 'l', 0, ESC_n, 0, -ESC_p,
197 /* 98 */ 0, ESC_r, 0, '}', 0, 0, 0, 0,
198 /* A0 */ 0, '~', -ESC_s, ESC_tee, 0,-ESC_v, -ESC_w, 0,
199 /* A8 */ 0,-ESC_z, 0, 0, 0, '[', 0, 0,
200 /* B0 */ 0, 0, 0, 0, 0, 0, 0, 0,
201 /* B8 */ 0, 0, 0, 0, 0, ']', '=', '-',
202 /* C0 */ '{',-ESC_A, -ESC_B, -ESC_C, -ESC_D,-ESC_E, 0, -ESC_G,
203 /* C8 */-ESC_H, 0, 0, 0, 0, 0, 0, 0,
204 /* D0 */ '}', 0, -ESC_K, 0, 0,-ESC_N, 0, -ESC_P,
205 /* D8 */-ESC_Q,-ESC_R, 0, 0, 0, 0, 0, 0,
206 /* E0 */ '\\', 0, -ESC_S, 0, 0,-ESC_V, -ESC_W, -ESC_X,
207 /* E8 */ 0,-ESC_Z, 0, 0, 0, 0, 0, 0,
208 /* F0 */ 0, 0, 0, 0, 0, 0, 0, 0,
209 /* F8 */ 0, 0, 0, 0, 0, 0, 0, 0
210 };
211 #endif
212
213
214 /* Table of special "verbs" like (*PRUNE). This is a short table, so it is
215 searched linearly. Put all the names into a single string, in order to reduce
216 the number of relocations when a shared library is dynamically linked. The
217 string is built from string macros so that it works in UTF-8 mode on EBCDIC
218 platforms. */
219
220 typedef struct verbitem {
221 int len; /* Length of verb name */
222 int op; /* Op when no arg, or -1 if arg mandatory */
223 int op_arg; /* Op when arg present, or -1 if not allowed */
224 } verbitem;
225
226 static const char verbnames[] =
227 "\0" /* Empty name is a shorthand for MARK */
228 STRING_MARK0
229 STRING_ACCEPT0
230 STRING_COMMIT0
231 STRING_F0
232 STRING_FAIL0
233 STRING_PRUNE0
234 STRING_SKIP0
235 STRING_THEN;
236
237 static const verbitem verbs[] = {
238 { 0, -1, OP_MARK },
239 { 4, -1, OP_MARK },
240 { 6, OP_ACCEPT, -1 },
241 { 6, OP_COMMIT, -1 },
242 { 1, OP_FAIL, -1 },
243 { 4, OP_FAIL, -1 },
244 { 5, OP_PRUNE, OP_PRUNE_ARG },
245 { 4, OP_SKIP, OP_SKIP_ARG },
246 { 4, OP_THEN, OP_THEN_ARG }
247 };
248
249 static const int verbcount = sizeof(verbs)/sizeof(verbitem);
250
251
252 /* Tables of names of POSIX character classes and their lengths. The names are
253 now all in a single string, to reduce the number of relocations when a shared
254 library is dynamically loaded. The list of lengths is terminated by a zero
255 length entry. The first three must be alpha, lower, upper, as this is assumed
256 for handling case independence. */
257
258 static const char posix_names[] =
259 STRING_alpha0 STRING_lower0 STRING_upper0 STRING_alnum0
260 STRING_ascii0 STRING_blank0 STRING_cntrl0 STRING_digit0
261 STRING_graph0 STRING_print0 STRING_punct0 STRING_space0
262 STRING_word0 STRING_xdigit;
263
264 static const pcre_uint8 posix_name_lengths[] = {
265 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 4, 6, 0 };
266
267 /* Table of class bit maps for each POSIX class. Each class is formed from a
268 base map, with an optional addition or removal of another map. Then, for some
269 classes, there is some additional tweaking: for [:blank:] the vertical space
270 characters are removed, and for [:alpha:] and [:alnum:] the underscore
271 character is removed. The triples in the table consist of the base map offset,
272 second map offset or -1 if no second map, and a non-negative value for map
273 addition or a negative value for map subtraction (if there are two maps). The
274 absolute value of the third field has these meanings: 0 => no tweaking, 1 =>
275 remove vertical space characters, 2 => remove underscore. */
276
277 static const int posix_class_maps[] = {
278 cbit_word, cbit_digit, -2, /* alpha */
279 cbit_lower, -1, 0, /* lower */
280 cbit_upper, -1, 0, /* upper */
281 cbit_word, -1, 2, /* alnum - word without underscore */
282 cbit_print, cbit_cntrl, 0, /* ascii */
283 cbit_space, -1, 1, /* blank - a GNU extension */
284 cbit_cntrl, -1, 0, /* cntrl */
285 cbit_digit, -1, 0, /* digit */
286 cbit_graph, -1, 0, /* graph */
287 cbit_print, -1, 0, /* print */
288 cbit_punct, -1, 0, /* punct */
289 cbit_space, -1, 0, /* space */
290 cbit_word, -1, 0, /* word - a Perl extension */
291 cbit_xdigit,-1, 0 /* xdigit */
292 };
293
294 /* Table of substitutes for \d etc when PCRE_UCP is set. The POSIX class
295 substitutes must be in the order of the names, defined above, and there are
296 both positive and negative cases. NULL means no substitute. */
297
298 #ifdef SUPPORT_UCP
299 static const pcre_uchar string_PNd[] = {
300 CHAR_BACKSLASH, CHAR_P, CHAR_LEFT_CURLY_BRACKET,
301 CHAR_N, CHAR_d, CHAR_RIGHT_CURLY_BRACKET, '\0' };
302 static const pcre_uchar string_pNd[] = {
303 CHAR_BACKSLASH, CHAR_p, CHAR_LEFT_CURLY_BRACKET,
304 CHAR_N, CHAR_d, CHAR_RIGHT_CURLY_BRACKET, '\0' };
305 static const pcre_uchar string_PXsp[] = {
306 CHAR_BACKSLASH, CHAR_P, CHAR_LEFT_CURLY_BRACKET,
307 CHAR_X, CHAR_s, CHAR_p, CHAR_RIGHT_CURLY_BRACKET, '\0' };
308 static const pcre_uchar string_pXsp[] = {
309 CHAR_BACKSLASH, CHAR_p, CHAR_LEFT_CURLY_BRACKET,
310 CHAR_X, CHAR_s, CHAR_p, CHAR_RIGHT_CURLY_BRACKET, '\0' };
311 static const pcre_uchar string_PXwd[] = {
312 CHAR_BACKSLASH, CHAR_P, CHAR_LEFT_CURLY_BRACKET,
313 CHAR_X, CHAR_w, CHAR_d, CHAR_RIGHT_CURLY_BRACKET, '\0' };
314 static const pcre_uchar string_pXwd[] = {
315 CHAR_BACKSLASH, CHAR_p, CHAR_LEFT_CURLY_BRACKET,
316 CHAR_X, CHAR_w, CHAR_d, CHAR_RIGHT_CURLY_BRACKET, '\0' };
317
318 static const pcre_uchar *substitutes[] = {
319 string_PNd, /* \D */
320 string_pNd, /* \d */
321 string_PXsp, /* \S */ /* NOTE: Xsp is Perl space */
322 string_pXsp, /* \s */
323 string_PXwd, /* \W */
324 string_pXwd /* \w */
325 };
326
327 static const pcre_uchar string_pL[] = {
328 CHAR_BACKSLASH, CHAR_p, CHAR_LEFT_CURLY_BRACKET,
329 CHAR_L, CHAR_RIGHT_CURLY_BRACKET, '\0' };
330 static const pcre_uchar string_pLl[] = {
331 CHAR_BACKSLASH, CHAR_p, CHAR_LEFT_CURLY_BRACKET,
332 CHAR_L, CHAR_l, CHAR_RIGHT_CURLY_BRACKET, '\0' };
333 static const pcre_uchar string_pLu[] = {
334 CHAR_BACKSLASH, CHAR_p, CHAR_LEFT_CURLY_BRACKET,
335 CHAR_L, CHAR_u, CHAR_RIGHT_CURLY_BRACKET, '\0' };
336 static const pcre_uchar string_pXan[] = {
337 CHAR_BACKSLASH, CHAR_p, CHAR_LEFT_CURLY_BRACKET,
338 CHAR_X, CHAR_a, CHAR_n, CHAR_RIGHT_CURLY_BRACKET, '\0' };
339 static const pcre_uchar string_h[] = {
340 CHAR_BACKSLASH, CHAR_h, '\0' };
341 static const pcre_uchar string_pXps[] = {
342 CHAR_BACKSLASH, CHAR_p, CHAR_LEFT_CURLY_BRACKET,
343 CHAR_X, CHAR_p, CHAR_s, CHAR_RIGHT_CURLY_BRACKET, '\0' };
344 static const pcre_uchar string_PL[] = {
345 CHAR_BACKSLASH, CHAR_P, CHAR_LEFT_CURLY_BRACKET,
346 CHAR_L, CHAR_RIGHT_CURLY_BRACKET, '\0' };
347 static const pcre_uchar string_PLl[] = {
348 CHAR_BACKSLASH, CHAR_P, CHAR_LEFT_CURLY_BRACKET,
349 CHAR_L, CHAR_l, CHAR_RIGHT_CURLY_BRACKET, '\0' };
350 static const pcre_uchar string_PLu[] = {
351 CHAR_BACKSLASH, CHAR_P, CHAR_LEFT_CURLY_BRACKET,
352 CHAR_L, CHAR_u, CHAR_RIGHT_CURLY_BRACKET, '\0' };
353 static const pcre_uchar string_PXan[] = {
354 CHAR_BACKSLASH, CHAR_P, CHAR_LEFT_CURLY_BRACKET,
355 CHAR_X, CHAR_a, CHAR_n, CHAR_RIGHT_CURLY_BRACKET, '\0' };
356 static const pcre_uchar string_H[] = {
357 CHAR_BACKSLASH, CHAR_H, '\0' };
358 static const pcre_uchar string_PXps[] = {
359 CHAR_BACKSLASH, CHAR_P, CHAR_LEFT_CURLY_BRACKET,
360 CHAR_X, CHAR_p, CHAR_s, CHAR_RIGHT_CURLY_BRACKET, '\0' };
361
362 static const pcre_uchar *posix_substitutes[] = {
363 string_pL, /* alpha */
364 string_pLl, /* lower */
365 string_pLu, /* upper */
366 string_pXan, /* alnum */
367 NULL, /* ascii */
368 string_h, /* blank */
369 NULL, /* cntrl */
370 string_pNd, /* digit */
371 NULL, /* graph */
372 NULL, /* print */
373 NULL, /* punct */
374 string_pXps, /* space */ /* NOTE: Xps is POSIX space */
375 string_pXwd, /* word */
376 NULL, /* xdigit */
377 /* Negated cases */
378 string_PL, /* ^alpha */
379 string_PLl, /* ^lower */
380 string_PLu, /* ^upper */
381 string_PXan, /* ^alnum */
382 NULL, /* ^ascii */
383 string_H, /* ^blank */
384 NULL, /* ^cntrl */
385 string_PNd, /* ^digit */
386 NULL, /* ^graph */
387 NULL, /* ^print */
388 NULL, /* ^punct */
389 string_PXps, /* ^space */ /* NOTE: Xps is POSIX space */
390 string_PXwd, /* ^word */
391 NULL /* ^xdigit */
392 };
393 #define POSIX_SUBSIZE (sizeof(posix_substitutes) / sizeof(pcre_uchar *))
394 #endif
395
396 #define STRING(a) # a
397 #define XSTRING(s) STRING(s)
398
399 /* The texts of compile-time error messages. These are "char *" because they
400 are passed to the outside world. Do not ever re-use any error number, because
401 they are documented. Always add a new error instead. Messages marked DEAD below
402 are no longer used. This used to be a table of strings, but in order to reduce
403 the number of relocations needed when a shared library is loaded dynamically,
404 it is now one long string. We cannot use a table of offsets, because the
405 lengths of inserts such as XSTRING(MAX_NAME_SIZE) are not known. Instead, we
406 simply count through to the one we want - this isn't a performance issue
407 because these strings are used only when there is a compilation error.
408
409 Each substring ends with \0 to insert a null character. This includes the final
410 substring, so that the whole string ends with \0\0, which can be detected when
411 counting through. */
412
413 static const char error_texts[] =
414 "no error\0"
415 "\\ at end of pattern\0"
416 "\\c at end of pattern\0"
417 "unrecognized character follows \\\0"
418 "numbers out of order in {} quantifier\0"
419 /* 5 */
420 "number too big in {} quantifier\0"
421 "missing terminating ] for character class\0"
422 "invalid escape sequence in character class\0"
423 "range out of order in character class\0"
424 "nothing to repeat\0"
425 /* 10 */
426 "operand of unlimited repeat could match the empty string\0" /** DEAD **/
427 "internal error: unexpected repeat\0"
428 "unrecognized character after (? or (?-\0"
429 "POSIX named classes are supported only within a class\0"
430 "missing )\0"
431 /* 15 */
432 "reference to non-existent subpattern\0"
433 "erroffset passed as NULL\0"
434 "unknown option bit(s) set\0"
435 "missing ) after comment\0"
436 "parentheses nested too deeply\0" /** DEAD **/
437 /* 20 */
438 "regular expression is too large\0"
439 "failed to get memory\0"
440 "unmatched parentheses\0"
441 "internal error: code overflow\0"
442 "unrecognized character after (?<\0"
443 /* 25 */
444 "lookbehind assertion is not fixed length\0"
445 "malformed number or name after (?(\0"
446 "conditional group contains more than two branches\0"
447 "assertion expected after (?(\0"
448 "(?R or (?[+-]digits must be followed by )\0"
449 /* 30 */
450 "unknown POSIX class name\0"
451 "POSIX collating elements are not supported\0"
452 "this version of PCRE is compiled without UTF support\0"
453 "spare error\0" /** DEAD **/
454 "character value in \\x{...} sequence is too large\0"
455 /* 35 */
456 "invalid condition (?(0)\0"
457 "\\C not allowed in lookbehind assertion\0"
458 "PCRE does not support \\L, \\l, \\N{name}, \\U, or \\u\0"
459 "number after (?C is > 255\0"
460 "closing ) for (?C expected\0"
461 /* 40 */
462 "recursive call could loop indefinitely\0"
463 "unrecognized character after (?P\0"
464 "syntax error in subpattern name (missing terminator)\0"
465 "two named subpatterns have the same name\0"
466 "invalid UTF-8 string\0"
467 /* 45 */
468 "support for \\P, \\p, and \\X has not been compiled\0"
469 "malformed \\P or \\p sequence\0"
470 "unknown property name after \\P or \\p\0"
471 "subpattern name is too long (maximum " XSTRING(MAX_NAME_SIZE) " characters)\0"
472 "too many named subpatterns (maximum " XSTRING(MAX_NAME_COUNT) ")\0"
473 /* 50 */
474 "repeated subpattern is too long\0" /** DEAD **/
475 "octal value is greater than \\377 in 8-bit non-UTF-8 mode\0"
476 "internal error: overran compiling workspace\0"
477 "internal error: previously-checked referenced subpattern not found\0"
478 "DEFINE group contains more than one branch\0"
479 /* 55 */
480 "repeating a DEFINE group is not allowed\0" /** DEAD **/
481 "inconsistent NEWLINE options\0"
482 "\\g is not followed by a braced, angle-bracketed, or quoted name/number or by a plain number\0"
483 "a numbered reference must not be zero\0"
484 "an argument is not allowed for (*ACCEPT), (*FAIL), or (*COMMIT)\0"
485 /* 60 */
486 "(*VERB) not recognized\0"
487 "number is too big\0"
488 "subpattern name expected\0"
489 "digit expected after (?+\0"
490 "] is an invalid data character in JavaScript compatibility mode\0"
491 /* 65 */
492 "different names for subpatterns of the same number are not allowed\0"
493 "(*MARK) must have an argument\0"
494 "this version of PCRE is not compiled with Unicode property support\0"
495 "\\c must be followed by an ASCII character\0"
496 "\\k is not followed by a braced, angle-bracketed, or quoted name\0"
497 /* 70 */
498 "internal error: unknown opcode in find_fixedlength()\0"
499 "\\N is not supported in a class\0"
500 "too many forward references\0"
501 "disallowed Unicode code point (>= 0xd800 && <= 0xdfff)\0"
502 "invalid UTF-16 string\0"
503 /* 75 */
504 "name is too long in (*MARK), (*PRUNE), (*SKIP), or (*THEN)\0"
505 "character value in \\u.... sequence is too large\0"
506 ;
507
508 /* Table to identify digits and hex digits. This is used when compiling
509 patterns. Note that the tables in chartables are dependent on the locale, and
510 may mark arbitrary characters as digits - but the PCRE compiling code expects
511 to handle only 0-9, a-z, and A-Z as digits when compiling. That is why we have
512 a private table here. It costs 256 bytes, but it is a lot faster than doing
513 character value tests (at least in some simple cases I timed), and in some
514 applications one wants PCRE to compile efficiently as well as match
515 efficiently.
516
517 For convenience, we use the same bit definitions as in chartables:
518
519 0x04 decimal digit
520 0x08 hexadecimal digit
521
522 Then we can use ctype_digit and ctype_xdigit in the code. */
523
524 /* Using a simple comparison for decimal numbers rather than a memory read
525 is much faster, and the resulting code is simpler (the compiler turns it
526 into a subtraction and unsigned comparison). */
527
528 #define IS_DIGIT(x) ((x) >= CHAR_0 && (x) <= CHAR_9)
529
530 #ifndef EBCDIC
531
532 /* This is the "normal" case, for ASCII systems, and EBCDIC systems running in
533 UTF-8 mode. */
534
535 static const pcre_uint8 digitab[] =
536 {
537 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 0- 7 */
538 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 8- 15 */
539 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 16- 23 */
540 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 24- 31 */
541 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - ' */
542 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* ( - / */
543 0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c, /* 0 - 7 */
544 0x0c,0x0c,0x00,0x00,0x00,0x00,0x00,0x00, /* 8 - ? */
545 0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* @ - G */
546 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* H - O */
547 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* P - W */
548 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* X - _ */
549 0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* ` - g */
550 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* h - o */
551 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* p - w */
552 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* x -127 */
553 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 128-135 */
554 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 136-143 */
555 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 144-151 */
556 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 152-159 */
557 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 160-167 */
558 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 168-175 */
559 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 176-183 */
560 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 184-191 */
561 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 192-199 */
562 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 200-207 */
563 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 208-215 */
564 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 216-223 */
565 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 224-231 */
566 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 232-239 */
567 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 240-247 */
568 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00};/* 248-255 */
569
570 #else
571
572 /* This is the "abnormal" case, for EBCDIC systems not running in UTF-8 mode. */
573
574 static const pcre_uint8 digitab[] =
575 {
576 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 0- 7 0 */
577 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 8- 15 */
578 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 16- 23 10 */
579 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 24- 31 */
580 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 32- 39 20 */
581 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 40- 47 */
582 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 48- 55 30 */
583 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 56- 63 */
584 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - 71 40 */
585 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 72- | */
586 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* & - 87 50 */
587 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 88- 95 */
588 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - -103 60 */
589 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 104- ? */
590 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 112-119 70 */
591 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 120- " */
592 0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* 128- g 80 */
593 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* h -143 */
594 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 144- p 90 */
595 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* q -159 */
596 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 160- x A0 */
597 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* y -175 */
598 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* ^ -183 B0 */
599 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 184-191 */
600 0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* { - G C0 */
601 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* H -207 */
602 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* } - P D0 */
603 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* Q -223 */
604 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* \ - X E0 */
605 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* Y -239 */
606 0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c, /* 0 - 7 F0 */
607 0x0c,0x0c,0x00,0x00,0x00,0x00,0x00,0x00};/* 8 -255 */
608
609 static const pcre_uint8 ebcdic_chartab[] = { /* chartable partial dup */
610 0x80,0x00,0x00,0x00,0x00,0x01,0x00,0x00, /* 0- 7 */
611 0x00,0x00,0x00,0x00,0x01,0x01,0x00,0x00, /* 8- 15 */
612 0x00,0x00,0x00,0x00,0x00,0x01,0x00,0x00, /* 16- 23 */
613 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 24- 31 */
614 0x00,0x00,0x00,0x00,0x00,0x01,0x00,0x00, /* 32- 39 */
615 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 40- 47 */
616 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 48- 55 */
617 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 56- 63 */
618 0x01,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - 71 */
619 0x00,0x00,0x00,0x80,0x00,0x80,0x80,0x80, /* 72- | */
620 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* & - 87 */
621 0x00,0x00,0x00,0x80,0x80,0x80,0x00,0x00, /* 88- 95 */
622 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - -103 */
623 0x00,0x00,0x00,0x00,0x00,0x10,0x00,0x80, /* 104- ? */
624 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 112-119 */
625 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 120- " */
626 0x00,0x1a,0x1a,0x1a,0x1a,0x1a,0x1a,0x12, /* 128- g */
627 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* h -143 */
628 0x00,0x12,0x12,0x12,0x12,0x12,0x12,0x12, /* 144- p */
629 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* q -159 */
630 0x00,0x00,0x12,0x12,0x12,0x12,0x12,0x12, /* 160- x */
631 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* y -175 */
632 0x80,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* ^ -183 */
633 0x00,0x00,0x80,0x00,0x00,0x00,0x00,0x00, /* 184-191 */
634 0x80,0x1a,0x1a,0x1a,0x1a,0x1a,0x1a,0x12, /* { - G */
635 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* H -207 */
636 0x00,0x12,0x12,0x12,0x12,0x12,0x12,0x12, /* } - P */
637 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* Q -223 */
638 0x00,0x00,0x12,0x12,0x12,0x12,0x12,0x12, /* \ - X */
639 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* Y -239 */
640 0x1c,0x1c,0x1c,0x1c,0x1c,0x1c,0x1c,0x1c, /* 0 - 7 */
641 0x1c,0x1c,0x00,0x00,0x00,0x00,0x00,0x00};/* 8 -255 */
642 #endif
643
644
645
646
647 /*************************************************
648 * Find an error text *
649 *************************************************/
650
651 /* The error texts are now all in one long string, to save on relocations. As
652 some of the text is of unknown length, we can't use a table of offsets.
653 Instead, just count through the strings. This is not a performance issue
654 because it happens only when there has been a compilation error.
655
656 Argument: the error number
657 Returns: pointer to the error string
658 */
659
660 static const char *
661 find_error_text(int n)
662 {
663 const char *s = error_texts;
664 for (; n > 0; n--)
665 {
666 while (*s++ != 0) {};
667 if (*s == 0) return "Error text not found (please report)";
668 }
669 return s;
670 }
671
672
673 /*************************************************
674 * Expand the workspace *
675 *************************************************/
676
677 /* This function is called during the second compiling phase, if the number of
678 forward references fills the existing workspace, which is originally a block on
679 the stack. A larger block is obtained from malloc() unless the ultimate limit
680 has been reached or the increase will be rather small.
681
682 Argument: pointer to the compile data block
683 Returns: 0 if all went well, else an error number
684 */
685
686 static int
687 expand_workspace(compile_data *cd)
688 {
689 pcre_uchar *newspace;
690 int newsize = cd->workspace_size * 2;
691
692 if (newsize > COMPILE_WORK_SIZE_MAX) newsize = COMPILE_WORK_SIZE_MAX;
693 if (cd->workspace_size >= COMPILE_WORK_SIZE_MAX ||
694 newsize - cd->workspace_size < WORK_SIZE_SAFETY_MARGIN)
695 return ERR72;
696
697 newspace = (PUBL(malloc))(IN_UCHARS(newsize));
698 if (newspace == NULL) return ERR21;
699 memcpy(newspace, cd->start_workspace, cd->workspace_size * sizeof(pcre_uchar));
700 cd->hwm = (pcre_uchar *)newspace + (cd->hwm - cd->start_workspace);
701 if (cd->workspace_size > COMPILE_WORK_SIZE)
702 (PUBL(free))((void *)cd->start_workspace);
703 cd->start_workspace = newspace;
704 cd->workspace_size = newsize;
705 return 0;
706 }
707
708
709
710 /*************************************************
711 * Check for counted repeat *
712 *************************************************/
713
714 /* This function is called when a '{' is encountered in a place where it might
715 start a quantifier. It looks ahead to see if it really is a quantifier or not.
716 It is only a quantifier if it is one of the forms {ddd} {ddd,} or {ddd,ddd}
717 where the ddds are digits.
718
719 Arguments:
720 p pointer to the first char after '{'
721
722 Returns: TRUE or FALSE
723 */
724
725 static BOOL
726 is_counted_repeat(const pcre_uchar *p)
727 {
728 if (!IS_DIGIT(*p)) return FALSE;
729 p++;
730 while (IS_DIGIT(*p)) p++;
731 if (*p == CHAR_RIGHT_CURLY_BRACKET) return TRUE;
732
733 if (*p++ != CHAR_COMMA) return FALSE;
734 if (*p == CHAR_RIGHT_CURLY_BRACKET) return TRUE;
735
736 if (!IS_DIGIT(*p)) return FALSE;
737 p++;
738 while (IS_DIGIT(*p)) p++;
739
740 return (*p == CHAR_RIGHT_CURLY_BRACKET);
741 }
742
743
744
745 /*************************************************
746 * Handle escapes *
747 *************************************************/
748
749 /* This function is called when a \ has been encountered. It either returns a
750 positive value for a simple escape such as \n, or a negative value which
751 encodes one of the more complicated things such as \d. A backreference to group
752 n is returned as -(ESC_REF + n); ESC_REF is the highest ESC_xxx macro. When
753 UTF-8 is enabled, a positive value greater than 255 may be returned. On entry,
754 ptr is pointing at the \. On exit, it is on the final character of the escape
755 sequence.
756
757 Arguments:
758 ptrptr points to the pattern position pointer
759 errorcodeptr points to the errorcode variable
760 bracount number of previous extracting brackets
761 options the options bits
762 isclass TRUE if inside a character class
763
764 Returns: zero or positive => a data character
765 negative => a special escape sequence
766 on error, errorcodeptr is set
767 */
768
769 static int
770 check_escape(const pcre_uchar **ptrptr, int *errorcodeptr, int bracount,
771 int options, BOOL isclass)
772 {
773 /* PCRE_UTF16 has the same value as PCRE_UTF8. */
774 BOOL utf = (options & PCRE_UTF8) != 0;
775 const pcre_uchar *ptr = *ptrptr + 1;
776 pcre_int32 c;
777 int i;
778
779 GETCHARINCTEST(c, ptr); /* Get character value, increment pointer */
780 ptr--; /* Set pointer back to the last byte */
781
782 /* If backslash is at the end of the pattern, it's an error. */
783
784 if (c == 0) *errorcodeptr = ERR1;
785
786 /* Non-alphanumerics are literals. For digits or letters, do an initial lookup
787 in a table. A non-zero result is something that can be returned immediately.
788 Otherwise further processing may be required. */
789
790 #ifndef EBCDIC /* ASCII/UTF-8 coding */
791 /* Not alphanumeric */
792 else if (c < CHAR_0 || c > CHAR_z) {}
793 else if ((i = escapes[c - CHAR_0]) != 0) c = i;
794
795 #else /* EBCDIC coding */
796 /* Not alphanumeric */
797 else if (c < CHAR_a || (!MAX_255(c) || (ebcdic_chartab[c] & 0x0E) == 0)) {}
798 else if ((i = escapes[c - 0x48]) != 0) c = i;
799 #endif
800
801 /* Escapes that need further processing, or are illegal. */
802
803 else
804 {
805 const pcre_uchar *oldptr;
806 BOOL braced, negated;
807
808 switch (c)
809 {
810 /* A number of Perl escapes are not handled by PCRE. We give an explicit
811 error. */
812
813 case CHAR_l:
814 case CHAR_L:
815 *errorcodeptr = ERR37;
816 break;
817
818 case CHAR_u:
819 if ((options & PCRE_JAVASCRIPT_COMPAT) != 0)
820 {
821 /* In JavaScript, \u must be followed by four hexadecimal numbers.
822 Otherwise it is a lowercase u letter. */
823 if (MAX_255(ptr[1]) && (digitab[ptr[1]] & ctype_xdigit) != 0
824 && MAX_255(ptr[2]) && (digitab[ptr[2]] & ctype_xdigit) != 0
825 && MAX_255(ptr[3]) && (digitab[ptr[3]] & ctype_xdigit) != 0
826 && MAX_255(ptr[4]) && (digitab[ptr[4]] & ctype_xdigit) != 0)
827 {
828 c = 0;
829 for (i = 0; i < 4; ++i)
830 {
831 register int cc = *(++ptr);
832 #ifndef EBCDIC /* ASCII/UTF-8 coding */
833 if (cc >= CHAR_a) cc -= 32; /* Convert to upper case */
834 c = (c << 4) + cc - ((cc < CHAR_A)? CHAR_0 : (CHAR_A - 10));
835 #else /* EBCDIC coding */
836 if (cc >= CHAR_a && cc <= CHAR_z) cc += 64; /* Convert to upper case */
837 c = (c << 4) + cc - ((cc >= CHAR_0)? CHAR_0 : (CHAR_A - 10));
838 #endif
839 }
840
841 #ifdef COMPILE_PCRE8
842 if (c > (utf ? 0x10ffff : 0xff))
843 #else
844 #ifdef COMPILE_PCRE16
845 if (c > (utf ? 0x10ffff : 0xffff))
846 #endif
847 #endif
848 {
849 *errorcodeptr = ERR76;
850 }
851 else if (utf && c >= 0xd800 && c <= 0xdfff) *errorcodeptr = ERR73;
852 }
853 }
854 else
855 *errorcodeptr = ERR37;
856 break;
857
858 case CHAR_U:
859 /* In JavaScript, \U is an uppercase U letter. */
860 if ((options & PCRE_JAVASCRIPT_COMPAT) == 0) *errorcodeptr = ERR37;
861 break;
862
863 /* In a character class, \g is just a literal "g". Outside a character
864 class, \g must be followed by one of a number of specific things:
865
866 (1) A number, either plain or braced. If positive, it is an absolute
867 backreference. If negative, it is a relative backreference. This is a Perl
868 5.10 feature.
869
870 (2) Perl 5.10 also supports \g{name} as a reference to a named group. This
871 is part of Perl's movement towards a unified syntax for back references. As
872 this is synonymous with \k{name}, we fudge it up by pretending it really
873 was \k.
874
875 (3) For Oniguruma compatibility we also support \g followed by a name or a
876 number either in angle brackets or in single quotes. However, these are
877 (possibly recursive) subroutine calls, _not_ backreferences. Just return
878 the -ESC_g code (cf \k). */
879
880 case CHAR_g:
881 if (isclass) break;
882 if (ptr[1] == CHAR_LESS_THAN_SIGN || ptr[1] == CHAR_APOSTROPHE)
883 {
884 c = -ESC_g;
885 break;
886 }
887
888 /* Handle the Perl-compatible cases */
889
890 if (ptr[1] == CHAR_LEFT_CURLY_BRACKET)
891 {
892 const pcre_uchar *p;
893 for (p = ptr+2; *p != 0 && *p != CHAR_RIGHT_CURLY_BRACKET; p++)
894 if (*p != CHAR_MINUS && !IS_DIGIT(*p)) break;
895 if (*p != 0 && *p != CHAR_RIGHT_CURLY_BRACKET)
896 {
897 c = -ESC_k;
898 break;
899 }
900 braced = TRUE;
901 ptr++;
902 }
903 else braced = FALSE;
904
905 if (ptr[1] == CHAR_MINUS)
906 {
907 negated = TRUE;
908 ptr++;
909 }
910 else negated = FALSE;
911
912 /* The integer range is limited by the machine's int representation. */
913 c = 0;
914 while (IS_DIGIT(ptr[1]))
915 {
916 if (((unsigned int)c) > INT_MAX / 10) /* Integer overflow */
917 {
918 c = -1;
919 break;
920 }
921 c = c * 10 + *(++ptr) - CHAR_0;
922 }
923 if (((unsigned int)c) > INT_MAX) /* Integer overflow */
924 {
925 while (IS_DIGIT(ptr[1]))
926 ptr++;
927 *errorcodeptr = ERR61;
928 break;
929 }
930
931 if (braced && *(++ptr) != CHAR_RIGHT_CURLY_BRACKET)
932 {
933 *errorcodeptr = ERR57;
934 break;
935 }
936
937 if (c == 0)
938 {
939 *errorcodeptr = ERR58;
940 break;
941 }
942
943 if (negated)
944 {
945 if (c > bracount)
946 {
947 *errorcodeptr = ERR15;
948 break;
949 }
950 c = bracount - (c - 1);
951 }
952
953 c = -(ESC_REF + c);
954 break;
955
956 /* The handling of escape sequences consisting of a string of digits
957 starting with one that is not zero is not straightforward. By experiment,
958 the way Perl works seems to be as follows:
959
960 Outside a character class, the digits are read as a decimal number. If the
961 number is less than 10, or if there are that many previous extracting
962 left brackets, then it is a back reference. Otherwise, up to three octal
963 digits are read to form an escaped byte. Thus \123 is likely to be octal
964 123 (cf \0123, which is octal 012 followed by the literal 3). If the octal
965 value is greater than 377, the least significant 8 bits are taken. Inside a
966 character class, \ followed by a digit is always an octal number. */
967
968 case CHAR_1: case CHAR_2: case CHAR_3: case CHAR_4: case CHAR_5:
969 case CHAR_6: case CHAR_7: case CHAR_8: case CHAR_9:
970
971 if (!isclass)
972 {
973 oldptr = ptr;
974 /* The integer range is limited by the machine's int representation. */
975 c -= CHAR_0;
976 while (IS_DIGIT(ptr[1]))
977 {
978 if (((unsigned int)c) > INT_MAX / 10) /* Integer overflow */
979 {
980 c = -1;
981 break;
982 }
983 c = c * 10 + *(++ptr) - CHAR_0;
984 }
985 if (((unsigned int)c) > INT_MAX) /* Integer overflow */
986 {
987 while (IS_DIGIT(ptr[1]))
988 ptr++;
989 *errorcodeptr = ERR61;
990 break;
991 }
992 if (c < 10 || c <= bracount)
993 {
994 c = -(ESC_REF + c);
995 break;
996 }
997 ptr = oldptr; /* Put the pointer back and fall through */
998 }
999
1000 /* Handle an octal number following \. If the first digit is 8 or 9, Perl
1001 generates a binary zero byte and treats the digit as a following literal.
1002 Thus we have to pull back the pointer by one. */
1003
1004 if ((c = *ptr) >= CHAR_8)
1005 {
1006 ptr--;
1007 c = 0;
1008 break;
1009 }
1010
1011 /* \0 always starts an octal number, but we may drop through to here with a
1012 larger first octal digit. The original code used just to take the least
1013 significant 8 bits of octal numbers (I think this is what early Perls used
1014 to do). Nowadays we allow for larger numbers in UTF-8 mode and 16-bit mode,
1015 but no more than 3 octal digits. */
1016
1017 case CHAR_0:
1018 c -= CHAR_0;
1019 while(i++ < 2 && ptr[1] >= CHAR_0 && ptr[1] <= CHAR_7)
1020 c = c * 8 + *(++ptr) - CHAR_0;
1021 #ifdef COMPILE_PCRE8
1022 if (!utf && c > 0xff) *errorcodeptr = ERR51;
1023 #endif
1024 break;
1025
1026 /* \x is complicated. \x{ddd} is a character number which can be greater
1027 than 0xff in utf or non-8bit mode, but only if the ddd are hex digits.
1028 If not, { is treated as a data character. */
1029
1030 case CHAR_x:
1031 if ((options & PCRE_JAVASCRIPT_COMPAT) != 0)
1032 {
1033 /* In JavaScript, \x must be followed by two hexadecimal numbers.
1034 Otherwise it is a lowercase x letter. */
1035 if (MAX_255(ptr[1]) && (digitab[ptr[1]] & ctype_xdigit) != 0
1036 && MAX_255(ptr[2]) && (digitab[ptr[2]] & ctype_xdigit) != 0)
1037 {
1038 c = 0;
1039 for (i = 0; i < 2; ++i)
1040 {
1041 register int cc = *(++ptr);
1042 #ifndef EBCDIC /* ASCII/UTF-8 coding */
1043 if (cc >= CHAR_a) cc -= 32; /* Convert to upper case */
1044 c = (c << 4) + cc - ((cc < CHAR_A)? CHAR_0 : (CHAR_A - 10));
1045 #else /* EBCDIC coding */
1046 if (cc >= CHAR_a && cc <= CHAR_z) cc += 64; /* Convert to upper case */
1047 c = (c << 4) + cc - ((cc >= CHAR_0)? CHAR_0 : (CHAR_A - 10));
1048 #endif
1049 }
1050 }
1051 break;
1052 }
1053
1054 if (ptr[1] == CHAR_LEFT_CURLY_BRACKET)
1055 {
1056 const pcre_uchar *pt = ptr + 2;
1057
1058 c = 0;
1059 while (MAX_255(*pt) && (digitab[*pt] & ctype_xdigit) != 0)
1060 {
1061 register int cc = *pt++;
1062 if (c == 0 && cc == CHAR_0) continue; /* Leading zeroes */
1063
1064 #ifndef EBCDIC /* ASCII/UTF-8 coding */
1065 if (cc >= CHAR_a) cc -= 32; /* Convert to upper case */
1066 c = (c << 4) + cc - ((cc < CHAR_A)? CHAR_0 : (CHAR_A - 10));
1067 #else /* EBCDIC coding */
1068 if (cc >= CHAR_a && cc <= CHAR_z) cc += 64; /* Convert to upper case */
1069 c = (c << 4) + cc - ((cc >= CHAR_0)? CHAR_0 : (CHAR_A - 10));
1070 #endif
1071
1072 #ifdef COMPILE_PCRE8
1073 if (c > (utf ? 0x10ffff : 0xff)) { c = -1; break; }
1074 #else
1075 #ifdef COMPILE_PCRE16
1076 if (c > (utf ? 0x10ffff : 0xffff)) { c = -1; break; }
1077 #endif
1078 #endif
1079 }
1080
1081 if (c < 0)
1082 {
1083 while (MAX_255(*pt) && (digitab[*pt] & ctype_xdigit) != 0) pt++;
1084 *errorcodeptr = ERR34;
1085 }
1086
1087 if (*pt == CHAR_RIGHT_CURLY_BRACKET)
1088 {
1089 if (utf && c >= 0xd800 && c <= 0xdfff) *errorcodeptr = ERR73;
1090 ptr = pt;
1091 break;
1092 }
1093
1094 /* If the sequence of hex digits does not end with '}', then we don't
1095 recognize this construct; fall through to the normal \x handling. */
1096 }
1097
1098 /* Read just a single-byte hex-defined char */
1099
1100 c = 0;
1101 while (i++ < 2 && MAX_255(ptr[1]) && (digitab[ptr[1]] & ctype_xdigit) != 0)
1102 {
1103 int cc; /* Some compilers don't like */
1104 cc = *(++ptr); /* ++ in initializers */
1105 #ifndef EBCDIC /* ASCII/UTF-8 coding */
1106 if (cc >= CHAR_a) cc -= 32; /* Convert to upper case */
1107 c = c * 16 + cc - ((cc < CHAR_A)? CHAR_0 : (CHAR_A - 10));
1108 #else /* EBCDIC coding */
1109 if (cc <= CHAR_z) cc += 64; /* Convert to upper case */
1110 c = c * 16 + cc - ((cc >= CHAR_0)? CHAR_0 : (CHAR_A - 10));
1111 #endif
1112 }
1113 break;
1114
1115 /* For \c, a following letter is upper-cased; then the 0x40 bit is flipped.
1116 An error is given if the byte following \c is not an ASCII character. This
1117 coding is ASCII-specific, but then the whole concept of \cx is
1118 ASCII-specific. (However, an EBCDIC equivalent has now been added.) */
1119
1120 case CHAR_c:
1121 c = *(++ptr);
1122 if (c == 0)
1123 {
1124 *errorcodeptr = ERR2;
1125 break;
1126 }
1127 #ifndef EBCDIC /* ASCII/UTF-8 coding */
1128 if (c > 127) /* Excludes all non-ASCII in either mode */
1129 {
1130 *errorcodeptr = ERR68;
1131 break;
1132 }
1133 if (c >= CHAR_a && c <= CHAR_z) c -= 32;
1134 c ^= 0x40;
1135 #else /* EBCDIC coding */
1136 if (c >= CHAR_a && c <= CHAR_z) c += 64;
1137 c ^= 0xC0;
1138 #endif
1139 break;
1140
1141 /* PCRE_EXTRA enables extensions to Perl in the matter of escapes. Any
1142 other alphanumeric following \ is an error if PCRE_EXTRA was set;
1143 otherwise, for Perl compatibility, it is a literal. This code looks a bit
1144 odd, but there used to be some cases other than the default, and there may
1145 be again in future, so I haven't "optimized" it. */
1146
1147 default:
1148 if ((options & PCRE_EXTRA) != 0) switch(c)
1149 {
1150 default:
1151 *errorcodeptr = ERR3;
1152 break;
1153 }
1154 break;
1155 }
1156 }
1157
1158 /* Perl supports \N{name} for character names, as well as plain \N for "not
1159 newline". PCRE does not support \N{name}. However, it does support
1160 quantification such as \N{2,3}. */
1161
1162 if (c == -ESC_N && ptr[1] == CHAR_LEFT_CURLY_BRACKET &&
1163 !is_counted_repeat(ptr+2))
1164 *errorcodeptr = ERR37;
1165
1166 /* If PCRE_UCP is set, we change the values for \d etc. */
1167
1168 if ((options & PCRE_UCP) != 0 && c <= -ESC_D && c >= -ESC_w)
1169 c -= (ESC_DU - ESC_D);
1170
1171 /* Set the pointer to the final character before returning. */
1172
1173 *ptrptr = ptr;
1174 return c;
1175 }
1176
1177
1178
1179 #ifdef SUPPORT_UCP
1180 /*************************************************
1181 * Handle \P and \p *
1182 *************************************************/
1183
1184 /* This function is called after \P or \p has been encountered, provided that
1185 PCRE is compiled with support for Unicode properties. On entry, ptrptr is
1186 pointing at the P or p. On exit, it is pointing at the final character of the
1187 escape sequence.
1188
1189 Argument:
1190 ptrptr points to the pattern position pointer
1191 negptr points to a boolean that is set TRUE for negation else FALSE
1192 dptr points to an int that is set to the detailed property value
1193 errorcodeptr points to the error code variable
1194
1195 Returns: type value from ucp_type_table, or -1 for an invalid type
1196 */
1197
1198 static int
1199 get_ucp(const pcre_uchar **ptrptr, BOOL *negptr, int *dptr, int *errorcodeptr)
1200 {
1201 int c, i, bot, top;
1202 const pcre_uchar *ptr = *ptrptr;
1203 pcre_uchar name[32];
1204
1205 c = *(++ptr);
1206 if (c == 0) goto ERROR_RETURN;
1207
1208 *negptr = FALSE;
1209
1210 /* \P or \p can be followed by a name in {}, optionally preceded by ^ for
1211 negation. */
1212
1213 if (c == CHAR_LEFT_CURLY_BRACKET)
1214 {
1215 if (ptr[1] == CHAR_CIRCUMFLEX_ACCENT)
1216 {
1217 *negptr = TRUE;
1218 ptr++;
1219 }
1220 for (i = 0; i < (int)(sizeof(name) / sizeof(pcre_uchar)) - 1; i++)
1221 {
1222 c = *(++ptr);
1223 if (c == 0) goto ERROR_RETURN;
1224 if (c == CHAR_RIGHT_CURLY_BRACKET) break;
1225 name[i] = c;
1226 }
1227 if (c != CHAR_RIGHT_CURLY_BRACKET) goto ERROR_RETURN;
1228 name[i] = 0;
1229 }
1230
1231 /* Otherwise there is just one following character */
1232
1233 else
1234 {
1235 name[0] = c;
1236 name[1] = 0;
1237 }
1238
1239 *ptrptr = ptr;
1240
1241 /* Search for a recognized property name using binary chop */
1242
1243 bot = 0;
1244 top = PRIV(utt_size);
1245
1246 while (bot < top)
1247 {
1248 i = (bot + top) >> 1;
1249 c = STRCMP_UC_C8(name, PRIV(utt_names) + PRIV(utt)[i].name_offset);
1250 if (c == 0)
1251 {
1252 *dptr = PRIV(utt)[i].value;
1253 return PRIV(utt)[i].type;
1254 }
1255 if (c > 0) bot = i + 1; else top = i;
1256 }
1257
1258 *errorcodeptr = ERR47;
1259 *ptrptr = ptr;
1260 return -1;
1261
1262 ERROR_RETURN:
1263 *errorcodeptr = ERR46;
1264 *ptrptr = ptr;
1265 return -1;
1266 }
1267 #endif
1268
1269
1270
1271
1272 /*************************************************
1273 * Read repeat counts *
1274 *************************************************/
1275
1276 /* Read an item of the form {n,m} and return the values. This is called only
1277 after is_counted_repeat() has confirmed that a repeat-count quantifier exists,
1278 so the syntax is guaranteed to be correct, but we need to check the values.
1279
1280 Arguments:
1281 p pointer to first char after '{'
1282 minp pointer to int for min
1283 maxp pointer to int for max
1284 returned as -1 if no max
1285 errorcodeptr points to error code variable
1286
1287 Returns: pointer to '}' on success;
1288 current ptr on error, with errorcodeptr set non-zero
1289 */
1290
1291 static const pcre_uchar *
1292 read_repeat_counts(const pcre_uchar *p, int *minp, int *maxp, int *errorcodeptr)
1293 {
1294 int min = 0;
1295 int max = -1;
1296
1297 /* Read the minimum value and do a paranoid check: a negative value indicates
1298 an integer overflow. */
1299
1300 while (IS_DIGIT(*p)) min = min * 10 + *p++ - CHAR_0;
1301 if (min < 0 || min > 65535)
1302 {
1303 *errorcodeptr = ERR5;
1304 return p;
1305 }
1306
1307 /* Read the maximum value if there is one, and again do a paranoid on its size.
1308 Also, max must not be less than min. */
1309
1310 if (*p == CHAR_RIGHT_CURLY_BRACKET) max = min; else
1311 {
1312 if (*(++p) != CHAR_RIGHT_CURLY_BRACKET)
1313 {
1314 max = 0;
1315 while(IS_DIGIT(*p)) max = max * 10 + *p++ - CHAR_0;
1316 if (max < 0 || max > 65535)
1317 {
1318 *errorcodeptr = ERR5;
1319 return p;
1320 }
1321 if (max < min)
1322 {
1323 *errorcodeptr = ERR4;
1324 return p;
1325 }
1326 }
1327 }
1328
1329 /* Fill in the required variables, and pass back the pointer to the terminating
1330 '}'. */
1331
1332 *minp = min;
1333 *maxp = max;
1334 return p;
1335 }
1336
1337
1338
1339 /*************************************************
1340 * Subroutine for finding forward reference *
1341 *************************************************/
1342
1343 /* This recursive function is called only from find_parens() below. The
1344 top-level call starts at the beginning of the pattern. All other calls must
1345 start at a parenthesis. It scans along a pattern's text looking for capturing
1346 subpatterns, and counting them. If it finds a named pattern that matches the
1347 name it is given, it returns its number. Alternatively, if the name is NULL, it
1348 returns when it reaches a given numbered subpattern. Recursion is used to keep
1349 track of subpatterns that reset the capturing group numbers - the (?| feature.
1350
1351 This function was originally called only from the second pass, in which we know
1352 that if (?< or (?' or (?P< is encountered, the name will be correctly
1353 terminated because that is checked in the first pass. There is now one call to
1354 this function in the first pass, to check for a recursive back reference by
1355 name (so that we can make the whole group atomic). In this case, we need check
1356 only up to the current position in the pattern, and that is still OK because
1357 and previous occurrences will have been checked. To make this work, the test
1358 for "end of pattern" is a check against cd->end_pattern in the main loop,
1359 instead of looking for a binary zero. This means that the special first-pass
1360 call can adjust cd->end_pattern temporarily. (Checks for binary zero while
1361 processing items within the loop are OK, because afterwards the main loop will
1362 terminate.)
1363
1364 Arguments:
1365 ptrptr address of the current character pointer (updated)
1366 cd compile background data
1367 name name to seek, or NULL if seeking a numbered subpattern
1368 lorn name length, or subpattern number if name is NULL
1369 xmode TRUE if we are in /x mode
1370 utf TRUE if we are in UTF-8 / UTF-16 mode
1371 count pointer to the current capturing subpattern number (updated)
1372
1373 Returns: the number of the named subpattern, or -1 if not found
1374 */
1375
1376 static int
1377 find_parens_sub(pcre_uchar **ptrptr, compile_data *cd, const pcre_uchar *name, int lorn,
1378 BOOL xmode, BOOL utf, int *count)
1379 {
1380 pcre_uchar *ptr = *ptrptr;
1381 int start_count = *count;
1382 int hwm_count = start_count;
1383 BOOL dup_parens = FALSE;
1384
1385 /* If the first character is a parenthesis, check on the type of group we are
1386 dealing with. The very first call may not start with a parenthesis. */
1387
1388 if (ptr[0] == CHAR_LEFT_PARENTHESIS)
1389 {
1390 /* Handle specials such as (*SKIP) or (*UTF8) etc. */
1391
1392 if (ptr[1] == CHAR_ASTERISK) ptr += 2;
1393
1394 /* Handle a normal, unnamed capturing parenthesis. */
1395
1396 else if (ptr[1] != CHAR_QUESTION_MARK)
1397 {
1398 *count += 1;
1399 if (name == NULL && *count == lorn) return *count;
1400 ptr++;
1401 }
1402
1403 /* All cases now have (? at the start. Remember when we are in a group
1404 where the parenthesis numbers are duplicated. */
1405
1406 else if (ptr[2] == CHAR_VERTICAL_LINE)
1407 {
1408 ptr += 3;
1409 dup_parens = TRUE;
1410 }
1411
1412 /* Handle comments; all characters are allowed until a ket is reached. */
1413
1414 else if (ptr[2] == CHAR_NUMBER_SIGN)
1415 {
1416 for (ptr += 3; *ptr != 0; ptr++) if (*ptr == CHAR_RIGHT_PARENTHESIS) break;
1417 goto FAIL_EXIT;
1418 }
1419
1420 /* Handle a condition. If it is an assertion, just carry on so that it
1421 is processed as normal. If not, skip to the closing parenthesis of the
1422 condition (there can't be any nested parens). */
1423
1424 else if (ptr[2] == CHAR_LEFT_PARENTHESIS)
1425 {
1426 ptr += 2;
1427 if (ptr[1] != CHAR_QUESTION_MARK)
1428 {
1429 while (*ptr != 0 && *ptr != CHAR_RIGHT_PARENTHESIS) ptr++;
1430 if (*ptr != 0) ptr++;
1431 }
1432 }
1433
1434 /* Start with (? but not a condition. */
1435
1436 else
1437 {
1438 ptr += 2;
1439 if (*ptr == CHAR_P) ptr++; /* Allow optional P */
1440
1441 /* We have to disambiguate (?<! and (?<= from (?<name> for named groups */
1442
1443 if ((*ptr == CHAR_LESS_THAN_SIGN && ptr[1] != CHAR_EXCLAMATION_MARK &&
1444 ptr[1] != CHAR_EQUALS_SIGN) || *ptr == CHAR_APOSTROPHE)
1445 {
1446 int term;
1447 const pcre_uchar *thisname;
1448 *count += 1;
1449 if (name == NULL && *count == lorn) return *count;
1450 term = *ptr++;
1451 if (term == CHAR_LESS_THAN_SIGN) term = CHAR_GREATER_THAN_SIGN;
1452 thisname = ptr;
1453 while (*ptr != term) ptr++;
1454 if (name != NULL && lorn == ptr - thisname &&
1455 STRNCMP_UC_UC(name, thisname, lorn) == 0)
1456 return *count;
1457 term++;
1458 }
1459 }
1460 }
1461
1462 /* Past any initial parenthesis handling, scan for parentheses or vertical
1463 bars. Stop if we get to cd->end_pattern. Note that this is important for the
1464 first-pass call when this value is temporarily adjusted to stop at the current
1465 position. So DO NOT change this to a test for binary zero. */
1466
1467 for (; ptr < cd->end_pattern; ptr++)
1468 {
1469 /* Skip over backslashed characters and also entire \Q...\E */
1470
1471 if (*ptr == CHAR_BACKSLASH)
1472 {
1473 if (*(++ptr) == 0) goto FAIL_EXIT;
1474 if (*ptr == CHAR_Q) for (;;)
1475 {
1476 while (*(++ptr) != 0 && *ptr != CHAR_BACKSLASH) {};
1477 if (*ptr == 0) goto FAIL_EXIT;
1478 if (*(++ptr) == CHAR_E) break;
1479 }
1480 continue;
1481 }
1482
1483 /* Skip over character classes; this logic must be similar to the way they
1484 are handled for real. If the first character is '^', skip it. Also, if the
1485 first few characters (either before or after ^) are \Q\E or \E we skip them
1486 too. This makes for compatibility with Perl. Note the use of STR macros to
1487 encode "Q\\E" so that it works in UTF-8 on EBCDIC platforms. */
1488
1489 if (*ptr == CHAR_LEFT_SQUARE_BRACKET)
1490 {
1491 BOOL negate_class = FALSE;
1492 for (;;)
1493 {
1494 if (ptr[1] == CHAR_BACKSLASH)
1495 {
1496 if (ptr[2] == CHAR_E)
1497 ptr+= 2;
1498 else if (STRNCMP_UC_C8(ptr + 2,
1499 STR_Q STR_BACKSLASH STR_E, 3) == 0)
1500 ptr += 4;
1501 else
1502 break;
1503 }
1504 else if (!negate_class && ptr[1] == CHAR_CIRCUMFLEX_ACCENT)
1505 {
1506 negate_class = TRUE;
1507 ptr++;
1508 }
1509 else break;
1510 }
1511
1512 /* If the next character is ']', it is a data character that must be
1513 skipped, except in JavaScript compatibility mode. */
1514
1515 if (ptr[1] == CHAR_RIGHT_SQUARE_BRACKET &&
1516 (cd->external_options & PCRE_JAVASCRIPT_COMPAT) == 0)
1517 ptr++;
1518
1519 while (*(++ptr) != CHAR_RIGHT_SQUARE_BRACKET)
1520 {
1521 if (*ptr == 0) return -1;
1522 if (*ptr == CHAR_BACKSLASH)
1523 {
1524 if (*(++ptr) == 0) goto FAIL_EXIT;
1525 if (*ptr == CHAR_Q) for (;;)
1526 {
1527 while (*(++ptr) != 0 && *ptr != CHAR_BACKSLASH) {};
1528 if (*ptr == 0) goto FAIL_EXIT;
1529 if (*(++ptr) == CHAR_E) break;
1530 }
1531 continue;
1532 }
1533 }
1534 continue;
1535 }
1536
1537 /* Skip comments in /x mode */
1538
1539 if (xmode && *ptr == CHAR_NUMBER_SIGN)
1540 {
1541 ptr++;
1542 while (*ptr != 0)
1543 {
1544 if (IS_NEWLINE(ptr)) { ptr += cd->nllen - 1; break; }
1545 ptr++;
1546 #ifdef SUPPORT_UTF
1547 if (utf) FORWARDCHAR(ptr);
1548 #endif
1549 }
1550 if (*ptr == 0) goto FAIL_EXIT;
1551 continue;
1552 }
1553
1554 /* Check for the special metacharacters */
1555
1556 if (*ptr == CHAR_LEFT_PARENTHESIS)
1557 {
1558 int rc = find_parens_sub(&ptr, cd, name, lorn, xmode, utf, count);
1559 if (rc > 0) return rc;
1560 if (*ptr == 0) goto FAIL_EXIT;
1561 }
1562
1563 else if (*ptr == CHAR_RIGHT_PARENTHESIS)
1564 {
1565 if (dup_parens && *count < hwm_count) *count = hwm_count;
1566 goto FAIL_EXIT;
1567 }
1568
1569 else if (*ptr == CHAR_VERTICAL_LINE && dup_parens)
1570 {
1571 if (*count > hwm_count) hwm_count = *count;
1572 *count = start_count;
1573 }
1574 }
1575
1576 FAIL_EXIT:
1577 *ptrptr = ptr;
1578 return -1;
1579 }
1580
1581
1582
1583
1584 /*************************************************
1585 * Find forward referenced subpattern *
1586 *************************************************/
1587
1588 /* This function scans along a pattern's text looking for capturing
1589 subpatterns, and counting them. If it finds a named pattern that matches the
1590 name it is given, it returns its number. Alternatively, if the name is NULL, it
1591 returns when it reaches a given numbered subpattern. This is used for forward
1592 references to subpatterns. We used to be able to start this scan from the
1593 current compiling point, using the current count value from cd->bracount, and
1594 do it all in a single loop, but the addition of the possibility of duplicate
1595 subpattern numbers means that we have to scan from the very start, in order to
1596 take account of such duplicates, and to use a recursive function to keep track
1597 of the different types of group.
1598
1599 Arguments:
1600 cd compile background data
1601 name name to seek, or NULL if seeking a numbered subpattern
1602 lorn name length, or subpattern number if name is NULL
1603 xmode TRUE if we are in /x mode
1604 utf TRUE if we are in UTF-8 / UTF-16 mode
1605
1606 Returns: the number of the found subpattern, or -1 if not found
1607 */
1608
1609 static int
1610 find_parens(compile_data *cd, const pcre_uchar *name, int lorn, BOOL xmode,
1611 BOOL utf)
1612 {
1613 pcre_uchar *ptr = (pcre_uchar *)cd->start_pattern;
1614 int count = 0;
1615 int rc;
1616
1617 /* If the pattern does not start with an opening parenthesis, the first call
1618 to find_parens_sub() will scan right to the end (if necessary). However, if it
1619 does start with a parenthesis, find_parens_sub() will return when it hits the
1620 matching closing parens. That is why we have to have a loop. */
1621
1622 for (;;)
1623 {
1624 rc = find_parens_sub(&ptr, cd, name, lorn, xmode, utf, &count);
1625 if (rc > 0 || *ptr++ == 0) break;
1626 }
1627
1628 return rc;
1629 }
1630
1631
1632
1633
1634 /*************************************************
1635 * Find first significant op code *
1636 *************************************************/
1637
1638 /* This is called by several functions that scan a compiled expression looking
1639 for a fixed first character, or an anchoring op code etc. It skips over things
1640 that do not influence this. For some calls, it makes sense to skip negative
1641 forward and all backward assertions, and also the \b assertion; for others it
1642 does not.
1643
1644 Arguments:
1645 code pointer to the start of the group
1646 skipassert TRUE if certain assertions are to be skipped
1647
1648 Returns: pointer to the first significant opcode
1649 */
1650
1651 static const pcre_uchar*
1652 first_significant_code(const pcre_uchar *code, BOOL skipassert)
1653 {
1654 for (;;)
1655 {
1656 switch ((int)*code)
1657 {
1658 case OP_ASSERT_NOT:
1659 case OP_ASSERTBACK:
1660 case OP_ASSERTBACK_NOT:
1661 if (!skipassert) return code;
1662 do code += GET(code, 1); while (*code == OP_ALT);
1663 code += PRIV(OP_lengths)[*code];
1664 break;
1665
1666 case OP_WORD_BOUNDARY:
1667 case OP_NOT_WORD_BOUNDARY:
1668 if (!skipassert) return code;
1669 /* Fall through */
1670
1671 case OP_CALLOUT:
1672 case OP_CREF:
1673 case OP_NCREF:
1674 case OP_RREF:
1675 case OP_NRREF:
1676 case OP_DEF:
1677 code += PRIV(OP_lengths)[*code];
1678 break;
1679
1680 default:
1681 return code;
1682 }
1683 }
1684 /* Control never reaches here */
1685 }
1686
1687
1688
1689
1690 /*************************************************
1691 * Find the fixed length of a branch *
1692 *************************************************/
1693
1694 /* Scan a branch and compute the fixed length of subject that will match it,
1695 if the length is fixed. This is needed for dealing with backward assertions.
1696 In UTF8 mode, the result is in characters rather than bytes. The branch is
1697 temporarily terminated with OP_END when this function is called.
1698
1699 This function is called when a backward assertion is encountered, so that if it
1700 fails, the error message can point to the correct place in the pattern.
1701 However, we cannot do this when the assertion contains subroutine calls,
1702 because they can be forward references. We solve this by remembering this case
1703 and doing the check at the end; a flag specifies which mode we are running in.
1704
1705 Arguments:
1706 code points to the start of the pattern (the bracket)
1707 utf TRUE in UTF-8 / UTF-16 mode
1708 atend TRUE if called when the pattern is complete
1709 cd the "compile data" structure
1710
1711 Returns: the fixed length,
1712 or -1 if there is no fixed length,
1713 or -2 if \C was encountered (in UTF-8 mode only)
1714 or -3 if an OP_RECURSE item was encountered and atend is FALSE
1715 or -4 if an unknown opcode was encountered (internal error)
1716 */
1717
1718 static int
1719 find_fixedlength(pcre_uchar *code, BOOL utf, BOOL atend, compile_data *cd)
1720 {
1721 int length = -1;
1722
1723 register int branchlength = 0;
1724 register pcre_uchar *cc = code + 1 + LINK_SIZE;
1725
1726 /* Scan along the opcodes for this branch. If we get to the end of the
1727 branch, check the length against that of the other branches. */
1728
1729 for (;;)
1730 {
1731 int d;
1732 pcre_uchar *ce, *cs;
1733 register int op = *cc;
1734
1735 switch (op)
1736 {
1737 /* We only need to continue for OP_CBRA (normal capturing bracket) and
1738 OP_BRA (normal non-capturing bracket) because the other variants of these
1739 opcodes are all concerned with unlimited repeated groups, which of course
1740 are not of fixed length. */
1741
1742 case OP_CBRA:
1743 case OP_BRA:
1744 case OP_ONCE:
1745 case OP_ONCE_NC:
1746 case OP_COND:
1747 d = find_fixedlength(cc + ((op == OP_CBRA)? IMM2_SIZE : 0), utf, atend, cd);
1748 if (d < 0) return d;
1749 branchlength += d;
1750 do cc += GET(cc, 1); while (*cc == OP_ALT);
1751 cc += 1 + LINK_SIZE;
1752 break;
1753
1754 /* Reached end of a branch; if it's a ket it is the end of a nested call.
1755 If it's ALT it is an alternation in a nested call. An ACCEPT is effectively
1756 an ALT. If it is END it's the end of the outer call. All can be handled by
1757 the same code. Note that we must not include the OP_KETRxxx opcodes here,
1758 because they all imply an unlimited repeat. */
1759
1760 case OP_ALT:
1761 case OP_KET:
1762 case OP_END:
1763 case OP_ACCEPT:
1764 case OP_ASSERT_ACCEPT:
1765 if (length < 0) length = branchlength;
1766 else if (length != branchlength) return -1;
1767 if (*cc != OP_ALT) return length;
1768 cc += 1 + LINK_SIZE;
1769 branchlength = 0;
1770 break;
1771
1772 /* A true recursion implies not fixed length, but a subroutine call may
1773 be OK. If the subroutine is a forward reference, we can't deal with
1774 it until the end of the pattern, so return -3. */
1775
1776 case OP_RECURSE:
1777 if (!atend) return -3;
1778 cs = ce = (pcre_uchar *)cd->start_code + GET(cc, 1); /* Start subpattern */
1779 do ce += GET(ce, 1); while (*ce == OP_ALT); /* End subpattern */
1780 if (cc > cs && cc < ce) return -1; /* Recursion */
1781 d = find_fixedlength(cs + IMM2_SIZE, utf, atend, cd);
1782 if (d < 0) return d;
1783 branchlength += d;
1784 cc += 1 + LINK_SIZE;
1785 break;
1786
1787 /* Skip over assertive subpatterns */
1788
1789 case OP_ASSERT:
1790 case OP_ASSERT_NOT:
1791 case OP_ASSERTBACK:
1792 case OP_ASSERTBACK_NOT:
1793 do cc += GET(cc, 1); while (*cc == OP_ALT);
1794 cc += PRIV(OP_lengths)[*cc];
1795 break;
1796
1797 /* Skip over things that don't match chars */
1798
1799 case OP_MARK:
1800 case OP_PRUNE_ARG:
1801 case OP_SKIP_ARG:
1802 case OP_THEN_ARG:
1803 cc += cc[1] + PRIV(OP_lengths)[*cc];
1804 break;
1805
1806 case OP_CALLOUT:
1807 case OP_CIRC:
1808 case OP_CIRCM:
1809 case OP_CLOSE:
1810 case OP_COMMIT:
1811 case OP_CREF:
1812 case OP_DEF:
1813 case OP_DOLL:
1814 case OP_DOLLM:
1815 case OP_EOD:
1816 case OP_EODN:
1817 case OP_FAIL:
1818 case OP_NCREF:
1819 case OP_NRREF:
1820 case OP_NOT_WORD_BOUNDARY:
1821 case OP_PRUNE:
1822 case OP_REVERSE:
1823 case OP_RREF:
1824 case OP_SET_SOM:
1825 case OP_SKIP:
1826 case OP_SOD:
1827 case OP_SOM:
1828 case OP_THEN:
1829 case OP_WORD_BOUNDARY:
1830 cc += PRIV(OP_lengths)[*cc];
1831 break;
1832
1833 /* Handle literal characters */
1834
1835 case OP_CHAR:
1836 case OP_CHARI:
1837 case OP_NOT:
1838 case OP_NOTI:
1839 branchlength++;
1840 cc += 2;
1841 #ifdef SUPPORT_UTF
1842 if (utf && HAS_EXTRALEN(cc[-1])) cc += GET_EXTRALEN(cc[-1]);
1843 #endif
1844 break;
1845
1846 /* Handle exact repetitions. The count is already in characters, but we
1847 need to skip over a multibyte character in UTF8 mode. */
1848
1849 case OP_EXACT:
1850 case OP_EXACTI:
1851 case OP_NOTEXACT:
1852 case OP_NOTEXACTI:
1853 branchlength += GET2(cc,1);
1854 cc += 2 + IMM2_SIZE;
1855 #ifdef SUPPORT_UTF
1856 if (utf && HAS_EXTRALEN(cc[-1])) cc += GET_EXTRALEN(cc[-1]);
1857 #endif
1858 break;
1859
1860 case OP_TYPEEXACT:
1861 branchlength += GET2(cc,1);
1862 if (cc[1 + IMM2_SIZE] == OP_PROP || cc[1 + IMM2_SIZE] == OP_NOTPROP) cc += 2;
1863 cc += 1 + IMM2_SIZE + 1;
1864 break;
1865
1866 /* Handle single-char matchers */
1867
1868 case OP_PROP:
1869 case OP_NOTPROP:
1870 cc += 2;
1871 /* Fall through */
1872
1873 case OP_HSPACE:
1874 case OP_VSPACE:
1875 case OP_NOT_HSPACE:
1876 case OP_NOT_VSPACE:
1877 case OP_NOT_DIGIT:
1878 case OP_DIGIT:
1879 case OP_NOT_WHITESPACE:
1880 case OP_WHITESPACE:
1881 case OP_NOT_WORDCHAR:
1882 case OP_WORDCHAR:
1883 case OP_ANY:
1884 case OP_ALLANY:
1885 branchlength++;
1886 cc++;
1887 break;
1888
1889 /* The single-byte matcher isn't allowed. This only happens in UTF-8 mode;
1890 otherwise \C is coded as OP_ALLANY. */
1891
1892 case OP_ANYBYTE:
1893 return -2;
1894
1895 /* Check a class for variable quantification */
1896
1897 #if defined SUPPORT_UTF || defined COMPILE_PCRE16
1898 case OP_XCLASS:
1899 cc += GET(cc, 1) - PRIV(OP_lengths)[OP_CLASS];
1900 /* Fall through */
1901 #endif
1902
1903 case OP_CLASS:
1904 case OP_NCLASS:
1905 cc += PRIV(OP_lengths)[OP_CLASS];
1906
1907 switch (*cc)
1908 {
1909 case OP_CRPLUS:
1910 case OP_CRMINPLUS:
1911 case OP_CRSTAR:
1912 case OP_CRMINSTAR:
1913 case OP_CRQUERY:
1914 case OP_CRMINQUERY:
1915 return -1;
1916
1917 case OP_CRRANGE:
1918 case OP_CRMINRANGE:
1919 if (GET2(cc,1) != GET2(cc,1+IMM2_SIZE)) return -1;
1920 branchlength += GET2(cc,1);
1921 cc += 1 + 2 * IMM2_SIZE;
1922 break;
1923
1924 default:
1925 branchlength++;
1926 }
1927 break;
1928
1929 /* Anything else is variable length */
1930
1931 case OP_ANYNL:
1932 case OP_BRAMINZERO:
1933 case OP_BRAPOS:
1934 case OP_BRAPOSZERO:
1935 case OP_BRAZERO:
1936 case OP_CBRAPOS:
1937 case OP_EXTUNI:
1938 case OP_KETRMAX:
1939 case OP_KETRMIN:
1940 case OP_KETRPOS:
1941 case OP_MINPLUS:
1942 case OP_MINPLUSI:
1943 case OP_MINQUERY:
1944 case OP_MINQUERYI:
1945 case OP_MINSTAR:
1946 case OP_MINSTARI:
1947 case OP_MINUPTO:
1948 case OP_MINUPTOI:
1949 case OP_NOTMINPLUS:
1950 case OP_NOTMINPLUSI:
1951 case OP_NOTMINQUERY:
1952 case OP_NOTMINQUERYI:
1953 case OP_NOTMINSTAR:
1954 case OP_NOTMINSTARI:
1955 case OP_NOTMINUPTO:
1956 case OP_NOTMINUPTOI:
1957 case OP_NOTPLUS:
1958 case OP_NOTPLUSI:
1959 case OP_NOTPOSPLUS:
1960 case OP_NOTPOSPLUSI:
1961 case OP_NOTPOSQUERY:
1962 case OP_NOTPOSQUERYI:
1963 case OP_NOTPOSSTAR:
1964 case OP_NOTPOSSTARI:
1965 case OP_NOTPOSUPTO:
1966 case OP_NOTPOSUPTOI:
1967 case OP_NOTQUERY:
1968 case OP_NOTQUERYI:
1969 case OP_NOTSTAR:
1970 case OP_NOTSTARI:
1971 case OP_NOTUPTO:
1972 case OP_NOTUPTOI:
1973 case OP_PLUS:
1974 case OP_PLUSI:
1975 case OP_POSPLUS:
1976 case OP_POSPLUSI:
1977 case OP_POSQUERY:
1978 case OP_POSQUERYI:
1979 case OP_POSSTAR:
1980 case OP_POSSTARI:
1981 case OP_POSUPTO:
1982 case OP_POSUPTOI:
1983 case OP_QUERY:
1984 case OP_QUERYI:
1985 case OP_REF:
1986 case OP_REFI:
1987 case OP_SBRA:
1988 case OP_SBRAPOS:
1989 case OP_SCBRA:
1990 case OP_SCBRAPOS:
1991 case OP_SCOND:
1992 case OP_SKIPZERO:
1993 case OP_STAR:
1994 case OP_STARI:
1995 case OP_TYPEMINPLUS:
1996 case OP_TYPEMINQUERY:
1997 case OP_TYPEMINSTAR:
1998 case OP_TYPEMINUPTO:
1999 case OP_TYPEPLUS:
2000 case OP_TYPEPOSPLUS:
2001 case OP_TYPEPOSQUERY:
2002 case OP_TYPEPOSSTAR:
2003 case OP_TYPEPOSUPTO:
2004 case OP_TYPEQUERY:
2005 case OP_TYPESTAR:
2006 case OP_TYPEUPTO:
2007 case OP_UPTO:
2008 case OP_UPTOI:
2009 return -1;
2010
2011 /* Catch unrecognized opcodes so that when new ones are added they
2012 are not forgotten, as has happened in the past. */
2013
2014 default:
2015 return -4;
2016 }
2017 }
2018 /* Control never gets here */
2019 }
2020
2021
2022
2023
2024 /*************************************************
2025 * Scan compiled regex for specific bracket *
2026 *************************************************/
2027
2028 /* This little function scans through a compiled pattern until it finds a
2029 capturing bracket with the given number, or, if the number is negative, an
2030 instance of OP_REVERSE for a lookbehind. The function is global in the C sense
2031 so that it can be called from pcre_study() when finding the minimum matching
2032 length.
2033
2034 Arguments:
2035 code points to start of expression
2036 utf TRUE in UTF-8 / UTF-16 mode
2037 number the required bracket number or negative to find a lookbehind
2038
2039 Returns: pointer to the opcode for the bracket, or NULL if not found
2040 */
2041
2042 const pcre_uchar *
2043 PRIV(find_bracket)(const pcre_uchar *code, BOOL utf, int number)
2044 {
2045 for (;;)
2046 {
2047 register int c = *code;
2048
2049 if (c == OP_END) return NULL;
2050
2051 /* XCLASS is used for classes that cannot be represented just by a bit
2052 map. This includes negated single high-valued characters. The length in
2053 the table is zero; the actual length is stored in the compiled code. */
2054
2055 if (c == OP_XCLASS) code += GET(code, 1);
2056
2057 /* Handle recursion */
2058
2059 else if (c == OP_REVERSE)
2060 {
2061 if (number < 0) return (pcre_uchar *)code;
2062 code += PRIV(OP_lengths)[c];
2063 }
2064
2065 /* Handle capturing bracket */
2066
2067 else if (c == OP_CBRA || c == OP_SCBRA ||
2068 c == OP_CBRAPOS || c == OP_SCBRAPOS)
2069 {
2070 int n = GET2(code, 1+LINK_SIZE);
2071 if (n == number) return (pcre_uchar *)code;
2072 code += PRIV(OP_lengths)[c];
2073 }
2074
2075 /* Otherwise, we can get the item's length from the table, except that for
2076 repeated character types, we have to test for \p and \P, which have an extra
2077 two bytes of parameters, and for MARK/PRUNE/SKIP/THEN with an argument, we
2078 must add in its length. */
2079
2080 else
2081 {
2082 switch(c)
2083 {
2084 case OP_TYPESTAR:
2085 case OP_TYPEMINSTAR:
2086 case OP_TYPEPLUS:
2087 case OP_TYPEMINPLUS:
2088 case OP_TYPEQUERY:
2089 case OP_TYPEMINQUERY:
2090 case OP_TYPEPOSSTAR:
2091 case OP_TYPEPOSPLUS:
2092 case OP_TYPEPOSQUERY:
2093 if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
2094 break;
2095
2096 case OP_TYPEUPTO:
2097 case OP_TYPEMINUPTO:
2098 case OP_TYPEEXACT:
2099 case OP_TYPEPOSUPTO:
2100 if (code[1 + IMM2_SIZE] == OP_PROP
2101 || code[1 + IMM2_SIZE] == OP_NOTPROP) code += 2;
2102 break;
2103
2104 case OP_MARK:
2105 case OP_PRUNE_ARG:
2106 case OP_SKIP_ARG:
2107 code += code[1];
2108 break;
2109
2110 case OP_THEN_ARG:
2111 code += code[1];
2112 break;
2113 }
2114
2115 /* Add in the fixed length from the table */
2116
2117 code += PRIV(OP_lengths)[c];
2118
2119 /* In UTF-8 mode, opcodes that are followed by a character may be followed by
2120 a multi-byte character. The length in the table is a minimum, so we have to
2121 arrange to skip the extra bytes. */
2122
2123 #ifdef SUPPORT_UTF
2124 if (utf) switch(c)
2125 {
2126 case OP_CHAR:
2127 case OP_CHARI:
2128 case OP_EXACT:
2129 case OP_EXACTI:
2130 case OP_UPTO:
2131 case OP_UPTOI:
2132 case OP_MINUPTO:
2133 case OP_MINUPTOI:
2134 case OP_POSUPTO:
2135 case OP_POSUPTOI:
2136 case OP_STAR:
2137 case OP_STARI:
2138 case OP_MINSTAR:
2139 case OP_MINSTARI:
2140 case OP_POSSTAR:
2141 case OP_POSSTARI:
2142 case OP_PLUS:
2143 case OP_PLUSI:
2144 case OP_MINPLUS:
2145 case OP_MINPLUSI:
2146 case OP_POSPLUS:
2147 case OP_POSPLUSI:
2148 case OP_QUERY:
2149 case OP_QUERYI:
2150 case OP_MINQUERY:
2151 case OP_MINQUERYI:
2152 case OP_POSQUERY:
2153 case OP_POSQUERYI:
2154 if (HAS_EXTRALEN(code[-1])) code += GET_EXTRALEN(code[-1]);
2155 break;
2156 }
2157 #else
2158 (void)(utf); /* Keep compiler happy by referencing function argument */
2159 #endif
2160 }
2161 }
2162 }
2163
2164
2165
2166 /*************************************************
2167 * Scan compiled regex for recursion reference *
2168 *************************************************/
2169
2170 /* This little function scans through a compiled pattern until it finds an
2171 instance of OP_RECURSE.
2172
2173 Arguments:
2174 code points to start of expression
2175 utf TRUE in UTF-8 / UTF-16 mode
2176
2177 Returns: pointer to the opcode for OP_RECURSE, or NULL if not found
2178 */
2179
2180 static const pcre_uchar *
2181 find_recurse(const pcre_uchar *code, BOOL utf)
2182 {
2183 for (;;)
2184 {
2185 register int c = *code;
2186 if (c == OP_END) return NULL;
2187 if (c == OP_RECURSE) return code;
2188
2189 /* XCLASS is used for classes that cannot be represented just by a bit
2190 map. This includes negated single high-valued characters. The length in
2191 the table is zero; the actual length is stored in the compiled code. */
2192
2193 if (c == OP_XCLASS) code += GET(code, 1);
2194
2195 /* Otherwise, we can get the item's length from the table, except that for
2196 repeated character types, we have to test for \p and \P, which have an extra
2197 two bytes of parameters, and for MARK/PRUNE/SKIP/THEN with an argument, we
2198 must add in its length. */
2199
2200 else
2201 {
2202 switch(c)
2203 {
2204 case OP_TYPESTAR:
2205 case OP_TYPEMINSTAR:
2206 case OP_TYPEPLUS:
2207 case OP_TYPEMINPLUS:
2208 case OP_TYPEQUERY:
2209 case OP_TYPEMINQUERY:
2210 case OP_TYPEPOSSTAR:
2211 case OP_TYPEPOSPLUS:
2212 case OP_TYPEPOSQUERY:
2213 if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
2214 break;
2215
2216 case OP_TYPEPOSUPTO:
2217 case OP_TYPEUPTO:
2218 case OP_TYPEMINUPTO:
2219 case OP_TYPEEXACT:
2220 if (code[1 + IMM2_SIZE] == OP_PROP
2221 || code[1 + IMM2_SIZE] == OP_NOTPROP) code += 2;
2222 break;
2223
2224 case OP_MARK:
2225 case OP_PRUNE_ARG:
2226 case OP_SKIP_ARG:
2227 code += code[1];
2228 break;
2229
2230 case OP_THEN_ARG:
2231 code += code[1];
2232 break;
2233 }
2234
2235 /* Add in the fixed length from the table */
2236
2237 code += PRIV(OP_lengths)[c];
2238
2239 /* In UTF-8 mode, opcodes that are followed by a character may be followed
2240 by a multi-byte character. The length in the table is a minimum, so we have
2241 to arrange to skip the extra bytes. */
2242
2243 #ifdef SUPPORT_UTF
2244 if (utf) switch(c)
2245 {
2246 case OP_CHAR:
2247 case OP_CHARI:
2248 case OP_NOT:
2249 case OP_NOTI:
2250 case OP_EXACT:
2251 case OP_EXACTI:
2252 case OP_NOTEXACT:
2253 case OP_NOTEXACTI:
2254 case OP_UPTO:
2255 case OP_UPTOI:
2256 case OP_NOTUPTO:
2257 case OP_NOTUPTOI:
2258 case OP_MINUPTO:
2259 case OP_MINUPTOI:
2260 case OP_NOTMINUPTO:
2261 case OP_NOTMINUPTOI:
2262 case OP_POSUPTO:
2263 case OP_POSUPTOI:
2264 case OP_NOTPOSUPTO:
2265 case OP_NOTPOSUPTOI:
2266 case OP_STAR:
2267 case OP_STARI:
2268 case OP_NOTSTAR:
2269 case OP_NOTSTARI:
2270 case OP_MINSTAR:
2271 case OP_MINSTARI:
2272 case OP_NOTMINSTAR:
2273 case OP_NOTMINSTARI:
2274 case OP_POSSTAR:
2275 case OP_POSSTARI:
2276 case OP_NOTPOSSTAR:
2277 case OP_NOTPOSSTARI:
2278 case OP_PLUS:
2279 case OP_PLUSI:
2280 case OP_NOTPLUS:
2281 case OP_NOTPLUSI:
2282 case OP_MINPLUS:
2283 case OP_MINPLUSI:
2284 case OP_NOTMINPLUS:
2285 case OP_NOTMINPLUSI:
2286 case OP_POSPLUS:
2287 case OP_POSPLUSI:
2288 case OP_NOTPOSPLUS:
2289 case OP_NOTPOSPLUSI:
2290 case OP_QUERY:
2291 case OP_QUERYI:
2292 case OP_NOTQUERY:
2293 case OP_NOTQUERYI:
2294 case OP_MINQUERY:
2295 case OP_MINQUERYI:
2296 case OP_NOTMINQUERY:
2297 case OP_NOTMINQUERYI:
2298 case OP_POSQUERY:
2299 case OP_POSQUERYI:
2300 case OP_NOTPOSQUERY:
2301 case OP_NOTPOSQUERYI:
2302 if (HAS_EXTRALEN(code[-1])) code += GET_EXTRALEN(code[-1]);
2303 break;
2304 }
2305 #else
2306 (void)(utf); /* Keep compiler happy by referencing function argument */
2307 #endif
2308 }
2309 }
2310 }
2311
2312
2313
2314 /*************************************************
2315 * Scan compiled branch for non-emptiness *
2316 *************************************************/
2317
2318 /* This function scans through a branch of a compiled pattern to see whether it
2319 can match the empty string or not. It is called from could_be_empty()
2320 below and from compile_branch() when checking for an unlimited repeat of a
2321 group that can match nothing. Note that first_significant_code() skips over
2322 backward and negative forward assertions when its final argument is TRUE. If we
2323 hit an unclosed bracket, we return "empty" - this means we've struck an inner
2324 bracket whose current branch will already have been scanned.
2325
2326 Arguments:
2327 code points to start of search
2328 endcode points to where to stop
2329 utf TRUE if in UTF-8 / UTF-16 mode
2330 cd contains pointers to tables etc.
2331
2332 Returns: TRUE if what is matched could be empty
2333 */
2334
2335 static BOOL
2336 could_be_empty_branch(const pcre_uchar *code, const pcre_uchar *endcode,
2337 BOOL utf, compile_data *cd)
2338 {
2339 register int c;
2340 for (code = first_significant_code(code + PRIV(OP_lengths)[*code], TRUE);
2341 code < endcode;
2342 code = first_significant_code(code + PRIV(OP_lengths)[c], TRUE))
2343 {
2344 const pcre_uchar *ccode;
2345
2346 c = *code;
2347
2348 /* Skip over forward assertions; the other assertions are skipped by
2349 first_significant_code() with a TRUE final argument. */
2350
2351 if (c == OP_ASSERT)
2352 {
2353 do code += GET(code, 1); while (*code == OP_ALT);
2354 c = *code;
2355 continue;
2356 }
2357
2358 /* For a recursion/subroutine call, if its end has been reached, which
2359 implies a backward reference subroutine call, we can scan it. If it's a
2360 forward reference subroutine call, we can't. To detect forward reference
2361 we have to scan up the list that is kept in the workspace. This function is
2362 called only when doing the real compile, not during the pre-compile that
2363 measures the size of the compiled pattern. */
2364
2365 if (c == OP_RECURSE)
2366 {
2367 const pcre_uchar *scode;
2368 BOOL empty_branch;
2369
2370 /* Test for forward reference */
2371
2372 for (scode = cd->start_workspace; scode < cd->hwm; scode += LINK_SIZE)
2373 if (GET(scode, 0) == code + 1 - cd->start_code) return TRUE;
2374
2375 /* Not a forward reference, test for completed backward reference */
2376
2377 empty_branch = FALSE;
2378 scode = cd->start_code + GET(code, 1);
2379 if (GET(scode, 1) == 0) return TRUE; /* Unclosed */
2380
2381 /* Completed backwards reference */
2382
2383 do
2384 {
2385 if (could_be_empty_branch(scode, endcode, utf, cd))
2386 {
2387 empty_branch = TRUE;
2388 break;
2389 }
2390 scode += GET(scode, 1);
2391 }
2392 while (*scode == OP_ALT);
2393
2394 if (!empty_branch) return FALSE; /* All branches are non-empty */
2395 continue;
2396 }
2397
2398 /* Groups with zero repeats can of course be empty; skip them. */
2399
2400 if (c == OP_BRAZERO || c == OP_BRAMINZERO || c == OP_SKIPZERO ||
2401 c == OP_BRAPOSZERO)
2402 {
2403 code += PRIV(OP_lengths)[c];
2404 do code += GET(code, 1); while (*code == OP_ALT);
2405 c = *code;
2406 continue;
2407 }
2408
2409 /* A nested group that is already marked as "could be empty" can just be
2410 skipped. */
2411
2412 if (c == OP_SBRA || c == OP_SBRAPOS ||
2413 c == OP_SCBRA || c == OP_SCBRAPOS)
2414 {
2415 do code += GET(code, 1); while (*code == OP_ALT);
2416 c = *code;
2417 continue;
2418 }
2419
2420 /* For other groups, scan the branches. */
2421
2422 if (c == OP_BRA || c == OP_BRAPOS ||
2423 c == OP_CBRA || c == OP_CBRAPOS ||
2424 c == OP_ONCE || c == OP_ONCE_NC ||
2425 c == OP_COND)
2426 {
2427 BOOL empty_branch;
2428 if (GET(code, 1) == 0) return TRUE; /* Hit unclosed bracket */
2429
2430 /* If a conditional group has only one branch, there is a second, implied,
2431 empty branch, so just skip over the conditional, because it could be empty.
2432 Otherwise, scan the individual branches of the group. */
2433
2434 if (c == OP_COND && code[GET(code, 1)] != OP_ALT)
2435 code += GET(code, 1);
2436 else
2437 {
2438 empty_branch = FALSE;
2439 do
2440 {
2441 if (!empty_branch && could_be_empty_branch(code, endcode, utf, cd))
2442 empty_branch = TRUE;
2443 code += GET(code, 1);
2444 }
2445 while (*code == OP_ALT);
2446 if (!empty_branch) return FALSE; /* All branches are non-empty */
2447 }
2448
2449 c = *code;
2450 continue;
2451 }
2452
2453 /* Handle the other opcodes */
2454
2455 switch (c)
2456 {
2457 /* Check for quantifiers after a class. XCLASS is used for classes that
2458 cannot be represented just by a bit map. This includes negated single
2459 high-valued characters. The length in PRIV(OP_lengths)[] is zero; the
2460 actual length is stored in the compiled code, so we must update "code"
2461 here. */
2462
2463 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
2464 case OP_XCLASS:
2465 ccode = code += GET(code, 1);
2466 goto CHECK_CLASS_REPEAT;
2467 #endif
2468
2469 case OP_CLASS:
2470 case OP_NCLASS:
2471 ccode = code + PRIV(OP_lengths)[OP_CLASS];
2472
2473 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
2474 CHECK_CLASS_REPEAT:
2475 #endif
2476
2477 switch (*ccode)
2478 {
2479 case OP_CRSTAR: /* These could be empty; continue */
2480 case OP_CRMINSTAR:
2481 case OP_CRQUERY:
2482 case OP_CRMINQUERY:
2483 break;
2484
2485 default: /* Non-repeat => class must match */
2486 case OP_CRPLUS: /* These repeats aren't empty */
2487 case OP_CRMINPLUS:
2488 return FALSE;
2489
2490 case OP_CRRANGE:
2491 case OP_CRMINRANGE:
2492 if (GET2(ccode, 1) > 0) return FALSE; /* Minimum > 0 */
2493 break;
2494 }
2495 break;
2496
2497 /* Opcodes that must match a character */
2498
2499 case OP_PROP:
2500 case OP_NOTPROP:
2501 case OP_EXTUNI:
2502 case OP_NOT_DIGIT:
2503 case OP_DIGIT:
2504 case OP_NOT_WHITESPACE:
2505 case OP_WHITESPACE:
2506 case OP_NOT_WORDCHAR:
2507 case OP_WORDCHAR:
2508 case OP_ANY:
2509 case OP_ALLANY:
2510 case OP_ANYBYTE:
2511 case OP_CHAR:
2512 case OP_CHARI:
2513 case OP_NOT:
2514 case OP_NOTI:
2515 case OP_PLUS:
2516 case OP_MINPLUS:
2517 case OP_POSPLUS:
2518 case OP_EXACT:
2519 case OP_NOTPLUS:
2520 case OP_NOTMINPLUS:
2521 case OP_NOTPOSPLUS:
2522 case OP_NOTEXACT:
2523 case OP_TYPEPLUS:
2524 case OP_TYPEMINPLUS:
2525 case OP_TYPEPOSPLUS:
2526 case OP_TYPEEXACT:
2527 return FALSE;
2528
2529 /* These are going to continue, as they may be empty, but we have to
2530 fudge the length for the \p and \P cases. */
2531
2532 case OP_TYPESTAR:
2533 case OP_TYPEMINSTAR:
2534 case OP_TYPEPOSSTAR:
2535 case OP_TYPEQUERY:
2536 case OP_TYPEMINQUERY:
2537 case OP_TYPEPOSQUERY:
2538 if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
2539 break;
2540
2541 /* Same for these */
2542
2543 case OP_TYPEUPTO:
2544 case OP_TYPEMINUPTO:
2545 case OP_TYPEPOSUPTO:
2546 if (code[1 + IMM2_SIZE] == OP_PROP
2547 || code[1 + IMM2_SIZE] == OP_NOTPROP) code += 2;
2548 break;
2549
2550 /* End of branch */
2551
2552 case OP_KET:
2553 case OP_KETRMAX:
2554 case OP_KETRMIN:
2555 case OP_KETRPOS:
2556 case OP_ALT:
2557 return TRUE;
2558
2559 /* In UTF-8 mode, STAR, MINSTAR, POSSTAR, QUERY, MINQUERY, POSQUERY, UPTO,
2560 MINUPTO, and POSUPTO may be followed by a multibyte character */
2561
2562 #ifdef SUPPORT_UTF
2563 case OP_STAR:
2564 case OP_STARI:
2565 case OP_MINSTAR:
2566 case OP_MINSTARI:
2567 case OP_POSSTAR:
2568 case OP_POSSTARI:
2569 case OP_QUERY:
2570 case OP_QUERYI:
2571 case OP_MINQUERY:
2572 case OP_MINQUERYI:
2573 case OP_POSQUERY:
2574 case OP_POSQUERYI:
2575 if (utf && HAS_EXTRALEN(code[1])) code += GET_EXTRALEN(code[1]);
2576 break;
2577
2578 case OP_UPTO:
2579 case OP_UPTOI:
2580 case OP_MINUPTO:
2581 case OP_MINUPTOI:
2582 case OP_POSUPTO:
2583 case OP_POSUPTOI:
2584 if (utf && HAS_EXTRALEN(code[1 + IMM2_SIZE])) code += GET_EXTRALEN(code[1 + IMM2_SIZE]);
2585 break;
2586 #endif
2587
2588 /* MARK, and PRUNE/SKIP/THEN with an argument must skip over the argument
2589 string. */
2590
2591 case OP_MARK:
2592 case OP_PRUNE_ARG:
2593 case OP_SKIP_ARG:
2594 code += code[1];
2595 break;
2596
2597 case OP_THEN_ARG:
2598 code += code[1];
2599 break;
2600
2601 /* None of the remaining opcodes are required to match a character. */
2602
2603 default:
2604 break;
2605 }
2606 }
2607
2608 return TRUE;
2609 }
2610
2611
2612
2613 /*************************************************
2614 * Scan compiled regex for non-emptiness *
2615 *************************************************/
2616
2617 /* This function is called to check for left recursive calls. We want to check
2618 the current branch of the current pattern to see if it could match the empty
2619 string. If it could, we must look outwards for branches at other levels,
2620 stopping when we pass beyond the bracket which is the subject of the recursion.
2621 This function is called only during the real compile, not during the
2622 pre-compile.
2623
2624 Arguments:
2625 code points to start of the recursion
2626 endcode points to where to stop (current RECURSE item)
2627 bcptr points to the chain of current (unclosed) branch starts
2628 utf TRUE if in UTF-8 / UTF-16 mode
2629 cd pointers to tables etc
2630
2631 Returns: TRUE if what is matched could be empty
2632 */
2633
2634 static BOOL
2635 could_be_empty(const pcre_uchar *code, const pcre_uchar *endcode,
2636 branch_chain *bcptr, BOOL utf, compile_data *cd)
2637 {
2638 while (bcptr != NULL && bcptr->current_branch >= code)
2639 {
2640 if (!could_be_empty_branch(bcptr->current_branch, endcode, utf, cd))
2641 return FALSE;
2642 bcptr = bcptr->outer;
2643 }
2644 return TRUE;
2645 }
2646
2647
2648
2649 /*************************************************
2650 * Check for POSIX class syntax *
2651 *************************************************/
2652
2653 /* This function is called when the sequence "[:" or "[." or "[=" is
2654 encountered in a character class. It checks whether this is followed by a
2655 sequence of characters terminated by a matching ":]" or ".]" or "=]". If we
2656 reach an unescaped ']' without the special preceding character, return FALSE.
2657
2658 Originally, this function only recognized a sequence of letters between the
2659 terminators, but it seems that Perl recognizes any sequence of characters,
2660 though of course unknown POSIX names are subsequently rejected. Perl gives an
2661 "Unknown POSIX class" error for [:f\oo:] for example, where previously PCRE
2662 didn't consider this to be a POSIX class. Likewise for [:1234:].
2663
2664 The problem in trying to be exactly like Perl is in the handling of escapes. We
2665 have to be sure that [abc[:x\]pqr] is *not* treated as containing a POSIX
2666 class, but [abc[:x\]pqr:]] is (so that an error can be generated). The code
2667 below handles the special case of \], but does not try to do any other escape
2668 processing. This makes it different from Perl for cases such as [:l\ower:]
2669 where Perl recognizes it as the POSIX class "lower" but PCRE does not recognize
2670 "l\ower". This is a lesser evil that not diagnosing bad classes when Perl does,
2671 I think.
2672
2673 A user pointed out that PCRE was rejecting [:a[:digit:]] whereas Perl was not.
2674 It seems that the appearance of a nested POSIX class supersedes an apparent
2675 external class. For example, [:a[:digit:]b:] matches "a", "b", ":", or
2676 a digit.
2677
2678 In Perl, unescaped square brackets may also appear as part of class names. For
2679 example, [:a[:abc]b:] gives unknown POSIX class "[:abc]b:]". However, for
2680 [:a[:abc]b][b:] it gives unknown POSIX class "[:abc]b][b:]", which does not
2681 seem right at all. PCRE does not allow closing square brackets in POSIX class
2682 names.
2683
2684 Arguments:
2685 ptr pointer to the initial [
2686 endptr where to return the end pointer
2687
2688 Returns: TRUE or FALSE
2689 */
2690
2691 static BOOL
2692 check_posix_syntax(const pcre_uchar *ptr, const pcre_uchar **endptr)
2693 {
2694 int terminator; /* Don't combine these lines; the Solaris cc */
2695 terminator = *(++ptr); /* compiler warns about "non-constant" initializer. */
2696 for (++ptr; *ptr != 0; ptr++)
2697 {
2698 if (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET)
2699 ptr++;
2700 else if (*ptr == CHAR_RIGHT_SQUARE_BRACKET) return FALSE;
2701 else
2702 {
2703 if (*ptr == terminator && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET)
2704 {
2705 *endptr = ptr;
2706 return TRUE;
2707 }
2708 if (*ptr == CHAR_LEFT_SQUARE_BRACKET &&
2709 (ptr[1] == CHAR_COLON || ptr[1] == CHAR_DOT ||
2710 ptr[1] == CHAR_EQUALS_SIGN) &&
2711 check_posix_syntax(ptr, endptr))
2712 return FALSE;
2713 }
2714 }
2715 return FALSE;
2716 }
2717
2718
2719
2720
2721 /*************************************************
2722 * Check POSIX class name *
2723 *************************************************/
2724
2725 /* This function is called to check the name given in a POSIX-style class entry
2726 such as [:alnum:].
2727
2728 Arguments:
2729 ptr points to the first letter
2730 len the length of the name
2731
2732 Returns: a value representing the name, or -1 if unknown
2733 */
2734
2735 static int
2736 check_posix_name(const pcre_uchar *ptr, int len)
2737 {
2738 const char *pn = posix_names;
2739 register int yield = 0;
2740 while (posix_name_lengths[yield] != 0)
2741 {
2742 if (len == posix_name_lengths[yield] &&
2743 STRNCMP_UC_C8(ptr, pn, len) == 0) return yield;
2744 pn += posix_name_lengths[yield] + 1;
2745 yield++;
2746 }
2747 return -1;
2748 }
2749
2750
2751 /*************************************************
2752 * Adjust OP_RECURSE items in repeated group *
2753 *************************************************/
2754
2755 /* OP_RECURSE items contain an offset from the start of the regex to the group
2756 that is referenced. This means that groups can be replicated for fixed
2757 repetition simply by copying (because the recursion is allowed to refer to
2758 earlier groups that are outside the current group). However, when a group is
2759 optional (i.e. the minimum quantifier is zero), OP_BRAZERO or OP_SKIPZERO is
2760 inserted before it, after it has been compiled. This means that any OP_RECURSE
2761 items within it that refer to the group itself or any contained groups have to
2762 have their offsets adjusted. That one of the jobs of this function. Before it
2763 is called, the partially compiled regex must be temporarily terminated with
2764 OP_END.
2765
2766 This function has been extended with the possibility of forward references for
2767 recursions and subroutine calls. It must also check the list of such references
2768 for the group we are dealing with. If it finds that one of the recursions in
2769 the current group is on this list, it adjusts the offset in the list, not the
2770 value in the reference (which is a group number).
2771
2772 Arguments:
2773 group points to the start of the group
2774 adjust the amount by which the group is to be moved
2775 utf TRUE in UTF-8 / UTF-16 mode
2776 cd contains pointers to tables etc.
2777 save_hwm the hwm forward reference pointer at the start of the group
2778
2779 Returns: nothing
2780 */
2781
2782 static void
2783 adjust_recurse(pcre_uchar *group, int adjust, BOOL utf, compile_data *cd,
2784 pcre_uchar *save_hwm)
2785 {
2786 pcre_uchar *ptr = group;
2787
2788 while ((ptr = (pcre_uchar *)find_recurse(ptr, utf)) != NULL)
2789 {
2790 int offset;
2791 pcre_uchar *hc;
2792
2793 /* See if this recursion is on the forward reference list. If so, adjust the
2794 reference. */
2795
2796 for (hc = save_hwm; hc < cd->hwm; hc += LINK_SIZE)
2797 {
2798 offset = GET(hc, 0);
2799 if (cd->start_code + offset == ptr + 1)
2800 {
2801 PUT(hc, 0, offset + adjust);
2802 break;
2803 }
2804 }
2805
2806 /* Otherwise, adjust the recursion offset if it's after the start of this
2807 group. */
2808
2809 if (hc >= cd->hwm)
2810 {
2811 offset = GET(ptr, 1);
2812 if (cd->start_code + offset >= group) PUT(ptr, 1, offset + adjust);
2813 }
2814
2815 ptr += 1 + LINK_SIZE;
2816 }
2817 }
2818
2819
2820
2821 /*************************************************
2822 * Insert an automatic callout point *
2823 *************************************************/
2824
2825 /* This function is called when the PCRE_AUTO_CALLOUT option is set, to insert
2826 callout points before each pattern item.
2827
2828 Arguments:
2829 code current code pointer
2830 ptr current pattern pointer
2831 cd pointers to tables etc
2832
2833 Returns: new code pointer
2834 */
2835
2836 static pcre_uchar *
2837 auto_callout(pcre_uchar *code, const pcre_uchar *ptr, compile_data *cd)
2838 {
2839 *code++ = OP_CALLOUT;
2840 *code++ = 255;
2841 PUT(code, 0, (int)(ptr - cd->start_pattern)); /* Pattern offset */
2842 PUT(code, LINK_SIZE, 0); /* Default length */
2843 return code + 2 * LINK_SIZE;
2844 }
2845
2846
2847
2848 /*************************************************
2849 * Complete a callout item *
2850 *************************************************/
2851
2852 /* A callout item contains the length of the next item in the pattern, which
2853 we can't fill in till after we have reached the relevant point. This is used
2854 for both automatic and manual callouts.
2855
2856 Arguments:
2857 previous_callout points to previous callout item
2858 ptr current pattern pointer
2859 cd pointers to tables etc
2860
2861 Returns: nothing
2862 */
2863
2864 static void
2865 complete_callout(pcre_uchar *previous_callout, const pcre_uchar *ptr, compile_data *cd)
2866 {
2867 int length = (int)(ptr - cd->start_pattern - GET(previous_callout, 2));
2868 PUT(previous_callout, 2 + LINK_SIZE, length);
2869 }
2870
2871
2872
2873 #ifdef SUPPORT_UCP
2874 /*************************************************
2875 * Get othercase range *
2876 *************************************************/
2877
2878 /* This function is passed the start and end of a class range, in UTF-8 mode
2879 with UCP support. It searches up the characters, looking for ranges of
2880 characters in the "other" case. Each call returns the next one, updating the
2881 start address. A character with multiple other cases is returned on its own
2882 with a special return value.
2883
2884 Arguments:
2885 cptr points to starting character value; updated
2886 d end value
2887 ocptr where to put start of othercase range
2888 odptr where to put end of othercase range
2889
2890 Yield: -1 when no more
2891 0 when a range is returned
2892 >0 the CASESET offset for char with multiple other cases
2893 in this case, ocptr contains the original
2894 */
2895
2896 static int
2897 get_othercase_range(unsigned int *cptr, unsigned int d, unsigned int *ocptr,
2898 unsigned int *odptr)
2899 {
2900 unsigned int c, othercase, next;
2901 int co;
2902
2903 /* Find the first character that has an other case. If it has multiple other
2904 cases, return its case offset value. */
2905
2906 for (c = *cptr; c <= d; c++)
2907 {
2908 if ((co = UCD_CASESET(c)) != 0)
2909 {
2910 *ocptr = c++; /* Character that has the set */
2911 *cptr = c; /* Rest of input range */
2912 return co;
2913 }
2914 if ((othercase = UCD_OTHERCASE(c)) != c) break;
2915 }
2916
2917 if (c > d) return -1; /* Reached end of range */
2918
2919 *ocptr = othercase;
2920 next = othercase + 1;
2921
2922 for (++c; c <= d; c++)
2923 {
2924 if (UCD_OTHERCASE(c) != next) break;
2925 next++;
2926 }
2927
2928 *odptr = next - 1; /* End of othercase range */
2929 *cptr = c; /* Rest of input range */
2930 return 0;
2931 }
2932
2933
2934
2935 /*************************************************
2936 * Check a character and a property *
2937 *************************************************/
2938
2939 /* This function is called by check_auto_possessive() when a property item
2940 is adjacent to a fixed character.
2941
2942 Arguments:
2943 c the character
2944 ptype the property type
2945 pdata the data for the type
2946 negated TRUE if it's a negated property (\P or \p{^)
2947
2948 Returns: TRUE if auto-possessifying is OK
2949 */
2950
2951 static BOOL
2952 check_char_prop(int c, int ptype, int pdata, BOOL negated)
2953 {
2954 const ucd_record *prop = GET_UCD(c);
2955 switch(ptype)
2956 {
2957 case PT_LAMP:
2958 return (prop->chartype == ucp_Lu ||
2959 prop->chartype == ucp_Ll ||
2960 prop->chartype == ucp_Lt) == negated;
2961
2962 case PT_GC:
2963 return (pdata == PRIV(ucp_gentype)[prop->chartype]) == negated;
2964
2965 case PT_PC:
2966 return (pdata == prop->chartype) == negated;
2967
2968 case PT_SC:
2969 return (pdata == prop->script) == negated;
2970
2971 /* These are specials */
2972
2973 case PT_ALNUM:
2974 return (PRIV(ucp_gentype)[prop->chartype] == ucp_L ||
2975 PRIV(ucp_gentype)[prop->chartype] == ucp_N) == negated;
2976
2977 case PT_SPACE: /* Perl space */
2978 return (PRIV(ucp_gentype)[prop->chartype] == ucp_Z ||
2979 c == CHAR_HT || c == CHAR_NL || c == CHAR_FF || c == CHAR_CR)
2980 == negated;
2981
2982 case PT_PXSPACE: /* POSIX space */
2983 return (PRIV(ucp_gentype)[prop->chartype] == ucp_Z ||
2984 c == CHAR_HT || c == CHAR_NL || c == CHAR_VT ||
2985 c == CHAR_FF || c == CHAR_CR)
2986 == negated;
2987
2988 case PT_WORD:
2989 return (PRIV(ucp_gentype)[prop->chartype] == ucp_L ||
2990 PRIV(ucp_gentype)[prop->chartype] == ucp_N ||
2991 c == CHAR_UNDERSCORE) == negated;
2992 }
2993 return FALSE;
2994 }
2995 #endif /* SUPPORT_UCP */
2996
2997
2998
2999 /*************************************************
3000 * Check if auto-possessifying is possible *
3001 *************************************************/
3002
3003 /* This function is called for unlimited repeats of certain items, to see
3004 whether the next thing could possibly match the repeated item. If not, it makes
3005 sense to automatically possessify the repeated item.
3006
3007 Arguments:
3008 previous pointer to the repeated opcode
3009 utf TRUE in UTF-8 / UTF-16 mode
3010 ptr next character in pattern
3011 options options bits
3012 cd contains pointers to tables etc.
3013
3014 Returns: TRUE if possessifying is wanted
3015 */
3016
3017 static BOOL
3018 check_auto_possessive(const pcre_uchar *previous, BOOL utf,
3019 const pcre_uchar *ptr, int options, compile_data *cd)
3020 {
3021 pcre_int32 c, next;
3022 int op_code = *previous++;
3023
3024 /* Skip whitespace and comments in extended mode */
3025
3026 if ((options & PCRE_EXTENDED) != 0)
3027 {
3028 for (;;)
3029 {
3030 while (MAX_255(*ptr) && (cd->ctypes[*ptr] & ctype_space) != 0) ptr++;
3031 if (*ptr == CHAR_NUMBER_SIGN)
3032 {
3033 ptr++;
3034 while (*ptr != 0)
3035 {
3036 if (IS_NEWLINE(ptr)) { ptr += cd->nllen; break; }
3037 ptr++;
3038 #ifdef SUPPORT_UTF
3039 if (utf) FORWARDCHAR(ptr);
3040 #endif
3041 }
3042 }
3043 else break;
3044 }
3045 }
3046
3047 /* If the next item is one that we can handle, get its value. A non-negative
3048 value is a character, a negative value is an escape value. */
3049
3050 if (*ptr == CHAR_BACKSLASH)
3051 {
3052 int temperrorcode = 0;
3053 next = check_escape(&ptr, &temperrorcode, cd->bracount, options, FALSE);
3054 if (temperrorcode != 0) return FALSE;
3055 ptr++; /* Point after the escape sequence */
3056 }
3057 else if (!MAX_255(*ptr) || (cd->ctypes[*ptr] & ctype_meta) == 0)
3058 {
3059 #ifdef SUPPORT_UTF
3060 if (utf) { GETCHARINC(next, ptr); } else
3061 #endif
3062 next = *ptr++;
3063 }
3064 else return FALSE;
3065
3066 /* Skip whitespace and comments in extended mode */
3067
3068 if ((options & PCRE_EXTENDED) != 0)
3069 {
3070 for (;;)
3071 {
3072 while (MAX_255(*ptr) && (cd->ctypes[*ptr] & ctype_space) != 0) ptr++;
3073 if (*ptr == CHAR_NUMBER_SIGN)
3074 {
3075 ptr++;
3076 while (*ptr != 0)
3077 {
3078 if (IS_NEWLINE(ptr)) { ptr += cd->nllen; break; }
3079 ptr++;
3080 #ifdef SUPPORT_UTF
3081 if (utf) FORWARDCHAR(ptr);
3082 #endif
3083 }
3084 }
3085 else break;
3086 }
3087 }
3088
3089 /* If the next thing is itself optional, we have to give up. */
3090
3091 if (*ptr == CHAR_ASTERISK || *ptr == CHAR_QUESTION_MARK ||
3092 STRNCMP_UC_C8(ptr, STR_LEFT_CURLY_BRACKET STR_0 STR_COMMA, 3) == 0)
3093 return FALSE;
3094
3095 /* Now compare the next item with the previous opcode. First, handle cases when
3096 the next item is a character. */
3097
3098 if (next >= 0) switch(op_code)
3099 {
3100 case OP_CHAR:
3101 #ifdef SUPPORT_UTF
3102 GETCHARTEST(c, previous);
3103 #else
3104 c = *previous;
3105 #endif
3106 return c != next;
3107
3108 /* For CHARI (caseless character) we must check the other case. If we have
3109 Unicode property support, we can use it to test the other case of
3110 high-valued characters. */
3111
3112 case OP_CHARI:
3113 #ifdef SUPPORT_UTF
3114 GETCHARTEST(c, previous);
3115 #else
3116 c = *previous;
3117 #endif
3118 if (c == next) return FALSE;
3119 #ifdef SUPPORT_UTF
3120 if (utf)
3121 {
3122 unsigned int othercase;
3123 if (next < 128) othercase = cd->fcc[next]; else
3124 #ifdef SUPPORT_UCP
3125 othercase = UCD_OTHERCASE((unsigned int)next);
3126 #else
3127 othercase = NOTACHAR;
3128 #endif
3129 return (unsigned int)c != othercase;
3130 }
3131 else
3132 #endif /* SUPPORT_UTF */
3133 return (c != TABLE_GET((unsigned int)next, cd->fcc, next)); /* Non-UTF-8 mode */
3134
3135 case OP_NOT:
3136 #ifdef SUPPORT_UTF
3137 GETCHARTEST(c, previous);
3138 #else
3139 c = *previous;
3140 #endif
3141 return c == next;
3142
3143 case OP_NOTI:
3144 #ifdef SUPPORT_UTF
3145 GETCHARTEST(c, previous);
3146 #else
3147 c = *previous;
3148 #endif
3149 if (c == next) return TRUE;
3150 #ifdef SUPPORT_UTF
3151 if (utf)
3152 {
3153 unsigned int othercase;
3154 if (next < 128) othercase = cd->fcc[next]; else
3155 #ifdef SUPPORT_UCP
3156 othercase = UCD_OTHERCASE((unsigned int)next);
3157 #else
3158 othercase = NOTACHAR;
3159 #endif
3160 return (unsigned int)c == othercase;
3161 }
3162 else
3163 #endif /* SUPPORT_UTF */
3164 return (c == TABLE_GET((unsigned int)next, cd->fcc, next)); /* Non-UTF-8 mode */
3165
3166 /* Note that OP_DIGIT etc. are generated only when PCRE_UCP is *not* set.
3167 When it is set, \d etc. are converted into OP_(NOT_)PROP codes. */
3168
3169 case OP_DIGIT:
3170 return next > 255 || (cd->ctypes[next] & ctype_digit) == 0;
3171
3172 case OP_NOT_DIGIT:
3173 return next <= 255 && (cd->ctypes[next] & ctype_digit) != 0;
3174
3175 case OP_WHITESPACE:
3176 return next > 255 || (cd->ctypes[next] & ctype_space) == 0;
3177
3178 case OP_NOT_WHITESPACE:
3179 return next <= 255 && (cd->ctypes[next] & ctype_space) != 0;
3180
3181 case OP_WORDCHAR:
3182 return next > 255 || (cd->ctypes[next] & ctype_word) == 0;
3183
3184 case OP_NOT_WORDCHAR:
3185 return next <= 255 && (cd->ctypes[next] & ctype_word) != 0;
3186
3187 case OP_HSPACE:
3188 case OP_NOT_HSPACE:
3189 switch(next)
3190 {
3191 HSPACE_CASES:
3192 return op_code == OP_NOT_HSPACE;
3193
3194 default:
3195 return op_code != OP_NOT_HSPACE;
3196 }
3197
3198 case OP_ANYNL:
3199 case OP_VSPACE:
3200 case OP_NOT_VSPACE:
3201 switch(next)
3202 {
3203 VSPACE_CASES:
3204 return op_code == OP_NOT_VSPACE;
3205
3206 default:
3207 return op_code != OP_NOT_VSPACE;
3208 }
3209
3210 #ifdef SUPPORT_UCP
3211 case OP_PROP:
3212 return check_char_prop(next, previous[0], previous[1], FALSE);
3213
3214 case OP_NOTPROP:
3215 return check_char_prop(next, previous[0], previous[1], TRUE);
3216 #endif
3217
3218 default:
3219 return FALSE;
3220 }
3221
3222
3223 /* Handle the case when the next item is \d, \s, etc. Note that when PCRE_UCP
3224 is set, \d turns into ESC_du rather than ESC_d, etc., so ESC_d etc. are
3225 generated only when PCRE_UCP is *not* set, that is, when only ASCII
3226 characteristics are recognized. Similarly, the opcodes OP_DIGIT etc. are
3227 replaced by OP_PROP codes when PCRE_UCP is set. */
3228
3229 switch(op_code)
3230 {
3231 case OP_CHAR:
3232 case OP_CHARI:
3233 #ifdef SUPPORT_UTF
3234 GETCHARTEST(c, previous);
3235 #else
3236 c = *previous;
3237 #endif
3238 switch(-next)
3239 {
3240 case ESC_d:
3241 return c > 255 || (cd->ctypes[c] & ctype_digit) == 0;
3242
3243 case ESC_D:
3244 return c <= 255 && (cd->ctypes[c] & ctype_digit) != 0;
3245
3246 case ESC_s:
3247 return c > 255 || (cd->ctypes[c] & ctype_space) == 0;
3248
3249 case ESC_S:
3250 return c <= 255 && (cd->ctypes[c] & ctype_space) != 0;
3251
3252 case ESC_w:
3253 return c > 255 || (cd->ctypes[c] & ctype_word) == 0;
3254
3255 case ESC_W:
3256 return c <= 255 && (cd->ctypes[c] & ctype_word) != 0;
3257
3258 case ESC_h:
3259 case ESC_H:
3260 switch(c)
3261 {
3262 HSPACE_CASES:
3263 return -next != ESC_h;
3264
3265 default:
3266 return -next == ESC_h;
3267 }
3268
3269 case ESC_v:
3270 case ESC_V:
3271 switch(c)
3272 {
3273 VSPACE_CASES:
3274 return -next != ESC_v;
3275
3276 default:
3277 return -next == ESC_v;
3278 }
3279
3280 /* When PCRE_UCP is set, these values get generated for \d etc. Find
3281 their substitutions and process them. The result will always be either
3282 -ESC_p or -ESC_P. Then fall through to process those values. */
3283
3284 #ifdef SUPPORT_UCP
3285 case ESC_du:
3286 case ESC_DU:
3287 case ESC_wu:
3288 case ESC_WU:
3289 case ESC_su:
3290 case ESC_SU:
3291 {
3292 int temperrorcode = 0;
3293 ptr = substitutes[-next - ESC_DU];
3294 next = check_escape(&ptr, &temperrorcode, 0, options, FALSE);
3295 if (temperrorcode != 0) return FALSE;
3296 ptr++; /* For compatibility */
3297 }
3298 /* Fall through */
3299
3300 case ESC_p:
3301 case ESC_P:
3302 {
3303 int ptype, pdata, errorcodeptr;
3304 BOOL negated;
3305
3306 ptr--; /* Make ptr point at the p or P */
3307 ptype = get_ucp(&ptr, &negated, &pdata, &errorcodeptr);
3308 if (ptype < 0) return FALSE;
3309 ptr++; /* Point past the final curly ket */
3310
3311 /* If the property item is optional, we have to give up. (When generated
3312 from \d etc by PCRE_UCP, this test will have been applied much earlier,
3313 to the original \d etc. At this point, ptr will point to a zero byte. */
3314
3315 if (*ptr == CHAR_ASTERISK || *ptr == CHAR_QUESTION_MARK ||
3316 STRNCMP_UC_C8(ptr, STR_LEFT_CURLY_BRACKET STR_0 STR_COMMA, 3) == 0)
3317 return FALSE;
3318
3319 /* Do the property check. */
3320
3321 return check_char_prop(c, ptype, pdata, (next == -ESC_P) != negated);
3322 }
3323 #endif
3324
3325 default:
3326 return FALSE;
3327 }
3328
3329 /* In principle, support for Unicode properties should be integrated here as
3330 well. It means re-organizing the above code so as to get hold of the property
3331 values before switching on the op-code. However, I wonder how many patterns
3332 combine ASCII \d etc with Unicode properties? (Note that if PCRE_UCP is set,
3333 these op-codes are never generated.) */
3334
3335 case OP_DIGIT:
3336 return next == -ESC_D || next == -ESC_s || next == -ESC_W ||
3337 next == -ESC_h || next == -ESC_v || next == -ESC_R;
3338
3339 case OP_NOT_DIGIT:
3340 return next == -ESC_d;
3341
3342 case OP_WHITESPACE:
3343 return next == -ESC_S || next == -ESC_d || next == -ESC_w;
3344
3345 case OP_NOT_WHITESPACE:
3346 return next == -ESC_s || next == -ESC_h || next == -ESC_v || next == -ESC_R;
3347
3348 case OP_HSPACE:
3349 return next == -ESC_S || next == -ESC_H || next == -ESC_d ||
3350 next == -ESC_w || next == -ESC_v || next == -ESC_R;
3351
3352 case OP_NOT_HSPACE:
3353 return next == -ESC_h;
3354
3355 /* Can't have \S in here because VT matches \S (Perl anomaly) */
3356 case OP_ANYNL:
3357 case OP_VSPACE:
3358 return next == -ESC_V || next == -ESC_d || next == -ESC_w;
3359
3360 case OP_NOT_VSPACE:
3361 return next == -ESC_v || next == -ESC_R;
3362
3363 case OP_WORDCHAR:
3364 return next == -ESC_W || next == -ESC_s || next == -ESC_h ||
3365 next == -ESC_v || next == -ESC_R;
3366
3367 case OP_NOT_WORDCHAR:
3368 return next == -ESC_w || next == -ESC_d;
3369
3370 default:
3371 return FALSE;
3372 }
3373
3374 /* Control does not reach here */
3375 }
3376
3377
3378
3379 /*************************************************
3380 * Add a character or range to a class *
3381 *************************************************/
3382
3383 /* This function packages up the logic of adding a character or range of
3384 characters to a class. The character values in the arguments will be within the
3385 valid values for the current mode (8-bit, 16-bit, UTF, etc). This function is
3386 mutually recursive with the function immediately below.
3387
3388 Arguments:
3389 classbits the bit map for characters < 256
3390 uchardptr points to the pointer for extra data
3391 options the options word
3392 cd contains pointers to tables etc.
3393 start start of range character
3394 end end of range character
3395
3396 Returns: the number of < 256 characters added
3397 the pointer to extra data is updated
3398 */
3399
3400 static int
3401 add_to_class(pcre_uint8 *classbits, pcre_uchar **uchardptr, int options,
3402 compile_data *cd, unsigned int start, unsigned int end)
3403 {
3404 unsigned int c;
3405 int n8 = 0;
3406
3407 /* If caseless matching is required, scan the range and process alternate
3408 cases. In Unicode, there are 8-bit characters that have alternate cases that
3409 are greater than 255 and vice-versa. Sometimes we can just extend the original
3410 range. */
3411
3412 if ((options & PCRE_CASELESS) != 0)
3413 {
3414 #ifdef SUPPORT_UCP
3415 if ((options & PCRE_UTF8) != 0)
3416 {
3417 int rc;
3418 unsigned int oc, od;
3419
3420 options &= ~PCRE_CASELESS; /* Remove for recursive calls */
3421 c = start;
3422
3423 while ((rc = get_othercase_range(&c, end, &oc, &od)) >= 0)
3424 {
3425 /* Handle a single character that has more than one other case. */
3426
3427 if (rc > 0) n8 += add_list_to_class(classbits, uchardptr, options, cd,
3428 PRIV(ucd_caseless_sets) + rc, oc);
3429
3430 /* Do nothing if the other case range is within the original range. */
3431
3432 else if (oc >= start && od <= end) continue;
3433
3434 /* Extend the original range if there is overlap, noting that if oc < c, we
3435 can't have od > end because a subrange is always shorter than the basic
3436 range. Otherwise, use a recursive call to add the additional range. */
3437
3438 else if (oc < start && od >= start - 1) start = oc; /* Extend downwards */
3439 else if (od > end && oc <= end + 1) end = od; /* Extend upwards */
3440 else n8 += add_to_class(classbits, uchardptr, options, cd, oc, od);
3441 }
3442 }
3443 else
3444 #endif /* SUPPORT_UCP */
3445
3446 /* Not UTF-mode, or no UCP */
3447
3448 for (c = start; c <= end && c < 256; c++)
3449 {
3450 SETBIT(classbits, cd->fcc[c]);
3451 n8++;
3452 }
3453 }
3454
3455 /* Now handle the original range. Adjust the final value according to the bit
3456 length - this means that the same lists of (e.g.) horizontal spaces can be used
3457 in all cases. */
3458
3459 #ifdef COMPILE_PCRE8
3460 #ifdef SUPPORT_UTF
3461 if ((options & PCRE_UTF8) == 0)
3462 #endif
3463 if (end > 0xff) end = 0xff;
3464 #endif
3465
3466 #ifdef COMPILE_PCRE16
3467 #ifdef SUPPORT_UTF
3468 if ((options & PCRE_UTF16) == 0)
3469 #endif
3470 if (end > 0xffff) end = 0xffff;
3471 #endif
3472
3473 /* If all characters are less than 256, use the bit map. Otherwise use extra
3474 data. */
3475
3476 if (end < 0x100)
3477 {
3478 for (c = start; c <= end; c++)
3479 {
3480 n8++;
3481 SETBIT(classbits, c);
3482 }
3483 }
3484
3485 else
3486 {
3487 pcre_uchar *uchardata = *uchardptr;
3488
3489 #ifdef SUPPORT_UTF
3490 if ((options & PCRE_UTF8) != 0) /* All UTFs use the same flag bit */
3491 {
3492 if (start < end)
3493 {
3494 *uchardata++ = XCL_RANGE;
3495 uchardata += PRIV(ord2utf)(start, uchardata);
3496 uchardata += PRIV(ord2utf)(end, uchardata);
3497 }
3498 else if (start == end)
3499 {
3500 *uchardata++ = XCL_SINGLE;
3501 uchardata += PRIV(ord2utf)(start, uchardata);
3502 }
3503 }
3504 else
3505 #endif /* SUPPORT_UTF */
3506
3507 /* Without UTF support, character values are constrained by the bit length,
3508 and can only be > 256 for 16-bit and 32-bit libraries. */
3509
3510 #ifdef COMPILE_PCRE8
3511 {}
3512 #else
3513 if (start < end)
3514 {
3515 *uchardata++ = XCL_RANGE;
3516 *uchardata++ = start;
3517 *uchardata++ = end;
3518 }
3519 else if (start == end)
3520 {
3521 *uchardata++ = XCL_SINGLE;
3522 *uchardata++ = start;
3523 }
3524 #endif
3525
3526 *uchardptr = uchardata; /* Updata extra data pointer */
3527 }
3528
3529 return n8; /* Number of 8-bit characters */
3530 }
3531
3532
3533
3534
3535 /*************************************************
3536 * Add a list of characters to a class *
3537 *************************************************/
3538
3539 /* This function is used for adding a list of case-equivalent characters to a
3540 class, and also for adding a list of horizontal or vertical whitespace. If the
3541 list is in order (which it should be), ranges of characters are detected and
3542 handled appropriately. This function is mutually recursive with the function
3543 above.
3544
3545 Arguments:
3546 classbits the bit map for characters < 256
3547 uchardptr points to the pointer for extra data
3548 options the options word
3549 cd contains pointers to tables etc.
3550 p points to row of 32-bit values, terminated by NOTACHAR
3551 except character to omit; this is used when adding lists of
3552 case-equivalent characters to avoid including the one we
3553 already know about
3554
3555 Returns: the number of < 256 characters added
3556 the pointer to extra data is updated
3557 */
3558
3559 static int
3560 add_list_to_class(pcre_uint8 *classbits, pcre_uchar **uchardptr, int options,
3561 compile_data *cd, const pcre_uint32 *p, unsigned int except)
3562 {
3563 int n8 = 0;
3564 while (p[0] < NOTACHAR)
3565 {
3566 int n = 0;
3567 if (p[0] != except)
3568 {
3569 while(p[n+1] == p[0] + n + 1) n++;
3570 n8 += add_to_class(classbits, uchardptr, options, cd, p[0], p[n]);
3571 }
3572 p += n + 1;
3573 }
3574 return n8;
3575 }
3576
3577
3578
3579 /*************************************************
3580 * Add characters not in a list to a class *
3581 *************************************************/
3582
3583 /* This function is used for adding the complement of a list of horizontal or
3584 vertical whitespace to a class. The list must be in order.
3585
3586 Arguments:
3587 classbits the bit map for characters < 256
3588 uchardptr points to the pointer for extra data
3589 options the options word
3590 cd contains pointers to tables etc.
3591 p points to row of 32-bit values, terminated by NOTACHAR
3592
3593 Returns: the number of < 256 characters added
3594 the pointer to extra data is updated
3595 */
3596
3597 static int
3598 add_not_list_to_class(pcre_uint8 *classbits, pcre_uchar **uchardptr,
3599 int options, compile_data *cd, const pcre_uint32 *p)
3600 {
3601 int n8 = 0;
3602 if (p[0] > 0)
3603 n8 += add_to_class(classbits, uchardptr, options, cd, 0, p[0] - 1);
3604 while (p[0] < NOTACHAR)
3605 {
3606 while (p[1] == p[0] + 1) p++;
3607 n8 += add_to_class(classbits, uchardptr, options, cd, p[0] + 1,
3608 (p[1] == NOTACHAR)? 0x10ffff : p[1] - 1);
3609 p++;
3610 }
3611 return n8;
3612 }
3613
3614
3615
3616 /*************************************************
3617 * Compile one branch *
3618 *************************************************/
3619
3620 /* Scan the pattern, compiling it into the a vector. If the options are
3621 changed during the branch, the pointer is used to change the external options
3622 bits. This function is used during the pre-compile phase when we are trying
3623 to find out the amount of memory needed, as well as during the real compile
3624 phase. The value of lengthptr distinguishes the two phases.
3625
3626 Arguments:
3627 optionsptr pointer to the option bits
3628 codeptr points to the pointer to the current code point
3629 ptrptr points to the current pattern pointer
3630 errorcodeptr points to error code variable
3631 firstcharptr set to initial literal character, or < 0 (REQ_UNSET, REQ_NONE)
3632 reqcharptr set to the last literal character required, else < 0
3633 bcptr points to current branch chain
3634 cond_depth conditional nesting depth
3635 cd contains pointers to tables etc.
3636 lengthptr NULL during the real compile phase
3637 points to length accumulator during pre-compile phase
3638
3639 Returns: TRUE on success
3640 FALSE, with *errorcodeptr set non-zero on error
3641 */
3642
3643 static BOOL
3644 compile_branch(int *optionsptr, pcre_uchar **codeptr,
3645 const pcre_uchar **ptrptr, int *errorcodeptr, pcre_int32 *firstcharptr,
3646 pcre_int32 *reqcharptr, branch_chain *bcptr, int cond_depth,
3647 compile_data *cd, int *lengthptr)
3648 {
3649 int repeat_type, op_type;
3650 int repeat_min = 0, repeat_max = 0; /* To please picky compilers */
3651 int bravalue = 0;
3652 int greedy_default, greedy_non_default;
3653 pcre_int32 firstchar, reqchar;
3654 pcre_int32 zeroreqchar, zerofirstchar;
3655 pcre_int32 req_caseopt, reqvary, tempreqvary;
3656 int options = *optionsptr; /* May change dynamically */
3657 int after_manual_callout = 0;
3658 int length_prevgroup = 0;
3659 register int c;
3660 register pcre_uchar *code = *codeptr;
3661 pcre_uchar *last_code = code;
3662 pcre_uchar *orig_code = code;
3663 pcre_uchar *tempcode;
3664 BOOL inescq = FALSE;
3665 BOOL groupsetfirstchar = FALSE;
3666 const pcre_uchar *ptr = *ptrptr;
3667 const pcre_uchar *tempptr;
3668 const pcre_uchar *nestptr = NULL;
3669 pcre_uchar *previous = NULL;
3670 pcre_uchar *previous_callout = NULL;
3671 pcre_uchar *save_hwm = NULL;
3672 pcre_uint8 classbits[32];
3673
3674 /* We can fish out the UTF-8 setting once and for all into a BOOL, but we
3675 must not do this for other options (e.g. PCRE_EXTENDED) because they may change
3676 dynamically as we process the pattern. */
3677
3678 #ifdef SUPPORT_UTF
3679 /* PCRE_UTF16 has the same value as PCRE_UTF8. */
3680 BOOL utf = (options & PCRE_UTF8) != 0;
3681 pcre_uchar utf_chars[6];
3682 #else
3683 BOOL utf = FALSE;
3684 #endif
3685
3686 /* Helper variables for OP_XCLASS opcode (for characters > 255). */
3687
3688 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
3689 BOOL xclass;
3690 pcre_uchar *class_uchardata;
3691 pcre_uchar *class_uchardata_base;
3692 #endif
3693
3694 #ifdef PCRE_DEBUG
3695 if (lengthptr != NULL) DPRINTF((">> start branch\n"));
3696 #endif
3697
3698 /* Set up the default and non-default settings for greediness */
3699
3700 greedy_default = ((options & PCRE_UNGREEDY) != 0);
3701 greedy_non_default = greedy_default ^ 1;
3702
3703 /* Initialize no first byte, no required byte. REQ_UNSET means "no char
3704 matching encountered yet". It gets changed to REQ_NONE if we hit something that
3705 matches a non-fixed char first char; reqchar just remains unset if we never
3706 find one.
3707
3708 When we hit a repeat whose minimum is zero, we may have to adjust these values
3709 to take the zero repeat into account. This is implemented by setting them to
3710 zerofirstbyte and zeroreqchar when such a repeat is encountered. The individual
3711 item types that can be repeated set these backoff variables appropriately. */
3712
3713 firstchar = reqchar = zerofirstchar = zeroreqchar = REQ_UNSET;
3714
3715 /* The variable req_caseopt contains either the REQ_CASELESS value
3716 or zero, according to the current setting of the caseless flag. The
3717 REQ_CASELESS leaves the lower 28 bit empty. It is added into the
3718 firstchar or reqchar variables to record the case status of the
3719 value. This is used only for ASCII characters. */
3720
3721 req_caseopt = ((options & PCRE_CASELESS) != 0)? REQ_CASELESS:0;
3722
3723 /* Switch on next character until the end of the branch */
3724
3725 for (;; ptr++)
3726 {
3727 BOOL negate_class;
3728 BOOL should_flip_negation;
3729 BOOL possessive_quantifier;
3730 BOOL is_quantifier;
3731 BOOL is_recurse;
3732 BOOL reset_bracount;
3733 int class_has_8bitchar;
3734 int class_one_char;
3735 int newoptions;
3736 int recno;
3737 int refsign;
3738 int skipbytes;
3739 int subreqchar;
3740 int subfirstchar;
3741 int terminator;
3742 int mclength;
3743 int tempbracount;
3744 pcre_uchar mcbuffer[8];
3745
3746 /* Get next character in the pattern */
3747
3748 c = *ptr;
3749
3750 /* If we are at the end of a nested substitution, revert to the outer level
3751 string. Nesting only happens one level deep. */
3752
3753 if (c == 0 && nestptr != NULL)
3754 {
3755 ptr = nestptr;
3756 nestptr = NULL;
3757 c = *ptr;
3758 }
3759
3760 /* If we are in the pre-compile phase, accumulate the length used for the
3761 previous cycle of this loop. */
3762
3763 if (lengthptr != NULL)
3764 {
3765 #ifdef PCRE_DEBUG
3766 if (code > cd->hwm) cd->hwm = code; /* High water info */
3767 #endif
3768 if (code > cd->start_workspace + cd->workspace_size -
3769 WORK_SIZE_SAFETY_MARGIN) /* Check for overrun */
3770 {
3771 *errorcodeptr = ERR52;
3772 goto FAILED;
3773 }
3774
3775 /* There is at least one situation where code goes backwards: this is the
3776 case of a zero quantifier after a class (e.g. [ab]{0}). At compile time,
3777 the class is simply eliminated. However, it is created first, so we have to
3778 allow memory for it. Therefore, don't ever reduce the length at this point.
3779 */
3780
3781 if (code < last_code) code = last_code;
3782
3783 /* Paranoid check for integer overflow */
3784
3785 if (OFLOW_MAX - *lengthptr < code - last_code)
3786 {
3787 *errorcodeptr = ERR20;
3788 goto FAILED;
3789 }
3790
3791 *lengthptr += (int)(code - last_code);
3792 DPRINTF(("length=%d added %d c=%c (0x%x)\n", *lengthptr,
3793 (int)(code - last_code), c, c));
3794
3795 /* If "previous" is set and it is not at the start of the work space, move
3796 it back to there, in order to avoid filling up the work space. Otherwise,
3797 if "previous" is NULL, reset the current code pointer to the start. */
3798
3799 if (previous != NULL)
3800 {
3801 if (previous > orig_code)
3802 {
3803 memmove(orig_code, previous, IN_UCHARS(code - previous));
3804 code -= previous - orig_code;
3805 previous = orig_code;
3806 }
3807 }
3808 else code = orig_code;
3809
3810 /* Remember where this code item starts so we can pick up the length
3811 next time round. */
3812
3813 last_code = code;
3814 }
3815
3816 /* In the real compile phase, just check the workspace used by the forward
3817 reference list. */
3818
3819 else if (cd->hwm > cd->start_workspace + cd->workspace_size -
3820 WORK_SIZE_SAFETY_MARGIN)
3821 {
3822 *errorcodeptr = ERR52;
3823 goto FAILED;
3824 }
3825
3826 /* If in \Q...\E, check for the end; if not, we have a literal */
3827
3828 if (inescq && c != 0)
3829 {
3830 if (c == CHAR_BACKSLASH && ptr[1] == CHAR_E)
3831 {
3832 inescq = FALSE;
3833 ptr++;
3834 continue;
3835 }
3836 else
3837 {
3838 if (previous_callout != NULL)
3839 {
3840 if (lengthptr == NULL) /* Don't attempt in pre-compile phase */
3841 complete_callout(previous_callout, ptr, cd);
3842 previous_callout = NULL;
3843 }
3844 if ((options & PCRE_AUTO_CALLOUT) != 0)
3845 {
3846 previous_callout = code;
3847 code = auto_callout(code, ptr, cd);
3848 }
3849 goto NORMAL_CHAR;
3850 }
3851 }
3852
3853 /* Fill in length of a previous callout, except when the next thing is
3854 a quantifier. */
3855
3856 is_quantifier =
3857 c == CHAR_ASTERISK || c == CHAR_PLUS || c == CHAR_QUESTION_MARK ||
3858 (c == CHAR_LEFT_CURLY_BRACKET && is_counted_repeat(ptr+1));
3859
3860 if (!is_quantifier && previous_callout != NULL &&
3861 after_manual_callout-- <= 0)
3862 {
3863 if (lengthptr == NULL) /* Don't attempt in pre-compile phase */
3864 complete_callout(previous_callout, ptr, cd);
3865 previous_callout = NULL;
3866 }
3867
3868 /* In extended mode, skip white space and comments. */
3869
3870 if ((options & PCRE_EXTENDED) != 0)
3871 {
3872 if (MAX_255(*ptr) && (cd->ctypes[c] & ctype_space) != 0) continue;
3873 if (c == CHAR_NUMBER_SIGN)
3874 {
3875 ptr++;
3876 while (*ptr != 0)
3877 {
3878 if (IS_NEWLINE(ptr)) { ptr += cd->nllen - 1; break; }
3879 ptr++;
3880 #ifdef SUPPORT_UTF
3881 if (utf) FORWARDCHAR(ptr);
3882 #endif
3883 }
3884 if (*ptr != 0) continue;
3885
3886 /* Else fall through to handle end of string */
3887 c = 0;
3888 }
3889 }
3890
3891 /* No auto callout for quantifiers. */
3892
3893 if ((options & PCRE_AUTO_CALLOUT) != 0 && !is_quantifier)
3894 {
3895 previous_callout = code;
3896 code = auto_callout(code, ptr, cd);
3897 }
3898
3899 switch(c)
3900 {
3901 /* ===================================================================*/
3902 case 0: /* The branch terminates at string end */
3903 case CHAR_VERTICAL_LINE: /* or | or ) */
3904 case CHAR_RIGHT_PARENTHESIS:
3905 *firstcharptr = firstchar;
3906 *reqcharptr = reqchar;
3907 *codeptr = code;
3908 *ptrptr = ptr;
3909 if (lengthptr != NULL)
3910 {
3911 if (OFLOW_MAX - *lengthptr < code - last_code)
3912 {
3913 *errorcodeptr = ERR20;
3914 goto FAILED;
3915 }
3916 *lengthptr += (int)(code - last_code); /* To include callout length */
3917 DPRINTF((">> end branch\n"));
3918 }
3919 return TRUE;
3920
3921
3922 /* ===================================================================*/
3923 /* Handle single-character metacharacters. In multiline mode, ^ disables
3924 the setting of any following char as a first character. */
3925
3926 case CHAR_CIRCUMFLEX_ACCENT:
3927 previous = NULL;
3928 if ((options & PCRE_MULTILINE) != 0)
3929 {
3930 if (firstchar == REQ_UNSET) firstchar = REQ_NONE;
3931 *code++ = OP_CIRCM;
3932 }
3933 else *code++ = OP_CIRC;
3934 break;
3935
3936 case CHAR_DOLLAR_SIGN:
3937 previous = NULL;
3938 *code++ = ((options & PCRE_MULTILINE) != 0)? OP_DOLLM : OP_DOLL;
3939 break;
3940
3941 /* There can never be a first char if '.' is first, whatever happens about
3942 repeats. The value of reqchar doesn't change either. */
3943
3944 case CHAR_DOT:
3945 if (firstchar == REQ_UNSET) firstchar = REQ_NONE;
3946 zerofirstchar = firstchar;
3947 zeroreqchar = reqchar;
3948 previous = code;
3949 *code++ = ((options & PCRE_DOTALL) != 0)? OP_ALLANY: OP_ANY;
3950 break;
3951
3952
3953 /* ===================================================================*/
3954 /* Character classes. If the included characters are all < 256, we build a
3955 32-byte bitmap of the permitted characters, except in the special case
3956 where there is only one such character. For negated classes, we build the
3957 map as usual, then invert it at the end. However, we use a different opcode
3958 so that data characters > 255 can be handled correctly.
3959
3960 If the class contains characters outside the 0-255 range, a different
3961 opcode is compiled. It may optionally have a bit map for characters < 256,
3962 but those above are are explicitly listed afterwards. A flag byte tells
3963 whether the bitmap is present, and whether this is a negated class or not.
3964
3965 In JavaScript compatibility mode, an isolated ']' causes an error. In
3966 default (Perl) mode, it is treated as a data character. */
3967
3968 case CHAR_RIGHT_SQUARE_BRACKET:
3969 if ((cd->external_options & PCRE_JAVASCRIPT_COMPAT) != 0)
3970 {
3971 *errorcodeptr = ERR64;
3972 goto FAILED;
3973 }
3974 goto NORMAL_CHAR;
3975
3976 case CHAR_LEFT_SQUARE_BRACKET:
3977 previous = code;
3978
3979 /* PCRE supports POSIX class stuff inside a class. Perl gives an error if
3980 they are encountered at the top level, so we'll do that too. */
3981
3982 if ((ptr[1] == CHAR_COLON || ptr[1] == CHAR_DOT ||
3983 ptr[1] == CHAR_EQUALS_SIGN) &&
3984 check_posix_syntax(ptr, &tempptr))
3985 {
3986 *errorcodeptr = (ptr[1] == CHAR_COLON)? ERR13 : ERR31;
3987 goto FAILED;
3988 }
3989
3990 /* If the first character is '^', set the negation flag and skip it. Also,
3991 if the first few characters (either before or after ^) are \Q\E or \E we
3992 skip them too. This makes for compatibility with Perl. */
3993
3994 negate_class = FALSE;
3995 for (;;)
3996 {
3997 c = *(++ptr);
3998 if (c == CHAR_BACKSLASH)
3999 {
4000 if (ptr[1] == CHAR_E)
4001 ptr++;
4002 else if (STRNCMP_UC_C8(ptr + 1, STR_Q STR_BACKSLASH STR_E, 3) == 0)
4003 ptr += 3;
4004 else
4005 break;
4006 }
4007 else if (!negate_class && c == CHAR_CIRCUMFLEX_ACCENT)
4008 negate_class = TRUE;
4009 else break;
4010 }
4011
4012 /* Empty classes are allowed in JavaScript compatibility mode. Otherwise,
4013 an initial ']' is taken as a data character -- the code below handles
4014 that. In JS mode, [] must always fail, so generate OP_FAIL, whereas
4015 [^] must match any character, so generate OP_ALLANY. */
4016
4017 if (c == CHAR_RIGHT_SQUARE_BRACKET &&
4018 (cd->external_options & PCRE_JAVASCRIPT_COMPAT) != 0)
4019 {
4020 *code++ = negate_class? OP_ALLANY : OP_FAIL;
4021 if (firstchar == REQ_UNSET) firstchar = REQ_NONE;
4022 zerofirstchar = firstchar;
4023 break;
4024 }
4025
4026 /* If a class contains a negative special such as \S, we need to flip the
4027 negation flag at the end, so that support for characters > 255 works
4028 correctly (they are all included in the class). */
4029
4030 should_flip_negation = FALSE;
4031
4032 /* For optimization purposes, we track some properties of the class:
4033 class_has_8bitchar will be non-zero if the class contains at least one <
4034 256 character; class_one_char will be 1 if the class contains just one
4035 character. */
4036
4037 class_has_8bitchar = 0;
4038 class_one_char = 0;
4039
4040 /* Initialize the 32-char bit map to all zeros. We build the map in a
4041 temporary bit of memory, in case the class contains fewer than two
4042 8-bit characters because in that case the compiled code doesn't use the bit
4043 map. */
4044
4045 memset(classbits, 0, 32 * sizeof(pcre_uint8));
4046
4047 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
4048 xclass = FALSE;
4049 class_uchardata = code + LINK_SIZE + 2; /* For XCLASS items */
4050 class_uchardata_base = class_uchardata; /* Save the start */
4051 #endif
4052
4053 /* Process characters until ] is reached. By writing this as a "do" it
4054 means that an initial ] is taken as a data character. At the start of the
4055 loop, c contains the first byte of the character. */
4056
4057 if (c != 0) do
4058 {
4059 const pcre_uchar *oldptr;
4060
4061 #ifdef SUPPORT_UTF
4062 if (utf && HAS_EXTRALEN(c))
4063 { /* Braces are required because the */
4064 GETCHARLEN(c, ptr, ptr); /* macro generates multiple statements */
4065 }
4066 #endif
4067
4068 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
4069 /* In the pre-compile phase, accumulate the length of any extra
4070 data and reset the pointer. This is so that very large classes that
4071 contain a zillion > 255 characters no longer overwrite the work space
4072 (which is on the stack). We have to remember that there was XCLASS data,
4073 however. */
4074
4075 if (lengthptr != NULL && class_uchardata > class_uchardata_base)
4076 {
4077 xclass = TRUE;
4078 *lengthptr += class_uchardata - class_uchardata_base;
4079 class_uchardata = class_uchardata_base;
4080 }
4081 #endif
4082
4083 /* Inside \Q...\E everything is literal except \E */
4084
4085 if (inescq)
4086 {
4087 if (c == CHAR_BACKSLASH && ptr[1] == CHAR_E) /* If we are at \E */
4088 {
4089 inescq = FALSE; /* Reset literal state */
4090 ptr++; /* Skip the 'E' */
4091 continue; /* Carry on with next */
4092 }
4093 goto CHECK_RANGE; /* Could be range if \E follows */
4094 }
4095
4096 /* Handle POSIX class names. Perl allows a negation extension of the
4097 form [:^name:]. A square bracket that doesn't match the syntax is
4098 treated as a literal. We also recognize the POSIX constructions
4099 [.ch.] and [=ch=] ("collating elements") and fault them, as Perl
4100 5.6 and 5.8 do. */
4101
4102 if (c == CHAR_LEFT_SQUARE_BRACKET &&
4103 (ptr[1] == CHAR_COLON || ptr[1] == CHAR_DOT ||
4104 ptr[1] == CHAR_EQUALS_SIGN) && check_posix_syntax(ptr, &tempptr))
4105 {
4106 BOOL local_negate = FALSE;
4107 int posix_class, taboffset, tabopt;
4108 register const pcre_uint8 *cbits = cd->cbits;
4109 pcre_uint8 pbits[32];
4110
4111 if (ptr[1] != CHAR_COLON)
4112 {
4113 *errorcodeptr = ERR31;
4114 goto FAILED;
4115 }
4116
4117 ptr += 2;
4118 if (*ptr == CHAR_CIRCUMFLEX_ACCENT)
4119 {
4120 local_negate = TRUE;
4121 should_flip_negation = TRUE; /* Note negative special */
4122 ptr++;
4123 }
4124
4125 posix_class = check_posix_name(ptr, (int)(tempptr - ptr));
4126 if (posix_class < 0)
4127 {
4128 *errorcodeptr = ERR30;
4129 goto FAILED;
4130 }
4131
4132 /* If matching is caseless, upper and lower are converted to
4133 alpha. This relies on the fact that the class table starts with
4134 alpha, lower, upper as the first 3 entries. */
4135
4136 if ((options & PCRE_CASELESS) != 0 && posix_class <= 2)
4137 posix_class = 0;
4138
4139 /* When PCRE_UCP is set, some of the POSIX classes are converted to
4140 different escape sequences that use Unicode properties. */
4141
4142 #ifdef SUPPORT_UCP
4143 if ((options & PCRE_UCP) != 0)
4144 {
4145 int pc = posix_class + ((local_negate)? POSIX_SUBSIZE/2 : 0);
4146 if (posix_substitutes[pc] != NULL)
4147 {
4148 nestptr = tempptr + 1;
4149 ptr = posix_substitutes[pc] - 1;
4150 continue;
4151 }
4152 }
4153 #endif
4154 /* In the non-UCP case, we build the bit map for the POSIX class in a
4155 chunk of local store because we may be adding and subtracting from it,
4156 and we don't want to subtract bits that may be in the main map already.
4157 At the end we or the result into the bit map that is being built. */
4158
4159 posix_class *= 3;
4160
4161 /* Copy in the first table (always present) */
4162
4163 memcpy(pbits, cbits + posix_class_maps[posix_class],
4164 32 * sizeof(pcre_uint8));
4165
4166 /* If there is a second table, add or remove it as required. */
4167
4168 taboffset = posix_class_maps[posix_class + 1];
4169 tabopt = posix_class_maps[posix_class + 2];
4170
4171 if (taboffset >= 0)
4172 {
4173 if (tabopt >= 0)
4174 for (c = 0; c < 32; c++) pbits[c] |= cbits[c + taboffset];
4175 else
4176 for (c = 0; c < 32; c++) pbits[c] &= ~cbits[c + taboffset];
4177 }
4178
4179 /* Now see if we need to remove any special characters. An option
4180 value of 1 removes vertical space and 2 removes underscore. */
4181
4182 if (tabopt < 0) tabopt = -tabopt;
4183 if (tabopt == 1) pbits[1] &= ~0x3c;
4184 else if (tabopt == 2) pbits[11] &= 0x7f;
4185
4186 /* Add the POSIX table or its complement into the main table that is
4187 being built and we are done. */
4188
4189 if (local_negate)
4190 for (c = 0; c < 32; c++) classbits[c] |= ~pbits[c];
4191 else
4192 for (c = 0; c < 32; c++) classbits[c] |= pbits[c];
4193
4194 ptr = tempptr + 1;
4195 /* Every class contains at least one < 256 character. */
4196 class_has_8bitchar = 1;
4197 /* Every class contains at least two characters. */
4198 class_one_char = 2;
4199 continue; /* End of POSIX syntax handling */
4200 }
4201
4202 /* Backslash may introduce a single character, or it may introduce one
4203 of the specials, which just set a flag. The sequence \b is a special
4204 case. Inside a class (and only there) it is treated as backspace. We
4205 assume that other escapes have more than one character in them, so
4206 speculatively set both class_has_8bitchar and class_one_char bigger
4207 than one. Unrecognized escapes fall through and are either treated
4208 as literal characters (by default), or are faulted if
4209 PCRE_EXTRA is set. */
4210
4211 if (c == CHAR_BACKSLASH)
4212 {
4213 c = check_escape(&ptr, errorcodeptr, cd->bracount, options, TRUE);
4214 if (*errorcodeptr != 0) goto FAILED;
4215
4216 if (-c == ESC_b) c = CHAR_BS; /* \b is backspace in a class */
4217 else if (-c == ESC_N) /* \N is not supported in a class */
4218 {
4219 *errorcodeptr = ERR71;
4220 goto FAILED;
4221 }
4222 else if (-c == ESC_Q) /* Handle start of quoted string */
4223 {
4224 if (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
4225 {
4226 ptr += 2; /* avoid empty string */
4227 }
4228 else inescq = TRUE;
4229 continue;
4230 }
4231 else if (-c == ESC_E) continue; /* Ignore orphan \E */
4232
4233 if (c < 0)
4234 {
4235 register const pcre_uint8 *cbits = cd->cbits;
4236 /* Every class contains at least two < 256 characters. */
4237 class_has_8bitchar++;
4238 /* Every class contains at least two characters. */
4239 class_one_char += 2;
4240
4241 switch (-c)
4242 {
4243 #ifdef SUPPORT_UCP
4244 case ESC_du: /* These are the values given for \d etc */
4245 case ESC_DU: /* when PCRE_UCP is set. We replace the */
4246 case ESC_wu: /* escape sequence with an appropriate \p */
4247 case ESC_WU: /* or \P to test Unicode properties instead */
4248 case ESC_su: /* of the default ASCII testing. */
4249 case ESC_SU:
4250 nestptr = ptr;
4251 ptr = substitutes[-c - ESC_DU] - 1; /* Just before substitute */
4252 class_has_8bitchar--; /* Undo! */
4253 continue;
4254 #endif
4255 case ESC_d:
4256 for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_digit];
4257 continue;
4258
4259 case ESC_D:
4260 should_flip_negation = TRUE;
4261 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_digit];
4262 continue;
4263
4264 case ESC_w:
4265 for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_word];
4266 continue;
4267
4268 case ESC_W:
4269 should_flip_negation = TRUE;
4270 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_word];
4271 continue;
4272
4273 /* Perl 5.004 onwards omits VT from \s, but we must preserve it
4274 if it was previously set by something earlier in the character
4275 class. Luckily, the value of CHAR_VT is 0x0b in both ASCII and
4276 EBCDIC, so we lazily just adjust the appropriate bit. */
4277
4278 case ESC_s:
4279 classbits[0] |= cbits[cbit_space];
4280 classbits[1] |= cbits[cbit_space+1] & ~0x08;
4281 for (c = 2; c < 32; c++) classbits[c] |= cbits[c+cbit_space];
4282 continue;
4283
4284 case ESC_S:
4285 should_flip_negation = TRUE;
4286 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_space];
4287 classbits[1] |= 0x08; /* Perl 5.004 onwards omits VT from \s */
4288 continue;
4289
4290 /* The rest apply in both UCP and non-UCP cases. */
4291
4292 case ESC_h:
4293 (void)add_list_to_class(classbits, &class_uchardata, options, cd,
4294 PRIV(hspace_list), NOTACHAR);
4295 continue;
4296
4297 case ESC_H:
4298 (void)add_not_list_to_class(classbits, &class_uchardata, options,
4299 cd, PRIV(hspace_list));
4300 continue;
4301
4302 case ESC_v:
4303 (void)add_list_to_class(classbits, &class_uchardata, options, cd,
4304 PRIV(vspace_list), NOTACHAR);
4305 continue;
4306
4307 case ESC_V:
4308 (void)add_not_list_to_class(classbits, &class_uchardata, options,
4309 cd, PRIV(vspace_list));
4310 continue;
4311
4312 #ifdef SUPPORT_UCP
4313 case ESC_p:
4314 case ESC_P:
4315 {
4316 BOOL negated;
4317 int pdata;
4318 int ptype = get_ucp(&ptr, &negated, &pdata, errorcodeptr);
4319 if (ptype < 0) goto FAILED;
4320 *class_uchardata++ = ((-c == ESC_p) != negated)?
4321 XCL_PROP : XCL_NOTPROP;
4322 *class_uchardata++ = ptype;
4323 *class_uchardata++ = pdata;
4324 class_has_8bitchar--; /* Undo! */
4325 continue;
4326 }
4327 #endif
4328 /* Unrecognized escapes are faulted if PCRE is running in its
4329 strict mode. By default, for compatibility with Perl, they are
4330 treated as literals. */
4331
4332 default:
4333 if ((options & PCRE_EXTRA) != 0)
4334 {
4335 *errorcodeptr = ERR7;
4336 goto FAILED;
4337 }
4338 class_has_8bitchar--; /* Undo the speculative increase. */
4339 class_one_char -= 2; /* Undo the speculative increase. */
4340 c = *ptr; /* Get the final character and fall through */
4341 break;
4342 }
4343 }
4344
4345 /* Fall through if the escape just defined a single character (c >= 0).
4346 This may be greater than 256. */
4347
4348 } /* End of backslash handling */
4349
4350 /* A character may be followed by '-' to form a range. However, Perl does
4351 not permit ']' to be the end of the range. A '-' character at the end is
4352 treated as a literal. Perl ignores orphaned \E sequences entirely. The
4353 code for handling \Q and \E is messy. */
4354
4355 CHECK_RANGE:
4356 while (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
4357 {
4358 inescq = FALSE;
4359 ptr += 2;
4360 }
4361 oldptr = ptr;
4362
4363 /* Remember if \r or \n were explicitly used */
4364
4365 if (c == CHAR_CR || c == CHAR_NL) cd->external_flags |= PCRE_HASCRORLF;
4366
4367 /* Check for range */
4368
4369 if (!inescq && ptr[1] == CHAR_MINUS)
4370 {
4371 int d;
4372 ptr += 2;
4373 while (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_E) ptr += 2;
4374
4375 /* If we hit \Q (not followed by \E) at this point, go into escaped
4376 mode. */
4377
4378 while (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_Q)
4379 {
4380 ptr += 2;
4381 if (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_E)
4382 { ptr += 2; continue; }
4383 inescq = TRUE;
4384 break;
4385 }
4386
4387 /* Minus (hyphen) at the end of a class is treated as a literal, so put
4388 back the pointer and jump to handle the character that preceded it. */
4389
4390 if (*ptr == 0 || (!inescq && *ptr == CHAR_RIGHT_SQUARE_BRACKET))
4391 {
4392 ptr = oldptr;
4393 goto CLASS_SINGLE_CHARACTER;
4394 }
4395
4396 /* Otherwise, we have a potential range; pick up the next character */
4397
4398 #ifdef SUPPORT_UTF
4399 if (utf)
4400 { /* Braces are required because the */
4401 GETCHARLEN(d, ptr, ptr); /* macro generates multiple statements */
4402 }
4403 else
4404 #endif
4405 d = *ptr; /* Not UTF-8 mode */
4406
4407 /* The second part of a range can be a single-character escape, but
4408 not any of the other escapes. Perl 5.6 treats a hyphen as a literal
4409 in such circumstances. */
4410
4411 if (!inescq && d == CHAR_BACKSLASH)
4412 {
4413 d = check_escape(&ptr, errorcodeptr, cd->bracount, options, TRUE);
4414 if (*errorcodeptr != 0) goto FAILED;
4415
4416 /* \b is backspace; any other special means the '-' was literal. */
4417
4418 if (d < 0)
4419 {
4420 if (d == -ESC_b) d = CHAR_BS; else
4421 {
4422 ptr = oldptr;
4423 goto CLASS_SINGLE_CHARACTER; /* A few lines below */
4424 }
4425 }
4426 }
4427
4428 /* Check that the two values are in the correct order. Optimize
4429 one-character ranges. */
4430
4431 if (d < c)
4432 {
4433 *errorcodeptr = ERR8;
4434 goto FAILED;
4435 }
4436 if (d == c) goto CLASS_SINGLE_CHARACTER; /* A few lines below */
4437
4438 /* We have found a character range, so single character optimizations
4439 cannot be done anymore. Any value greater than 1 indicates that there
4440 is more than one character. */
4441
4442 class_one_char = 2;
4443
4444 /* Remember an explicit \r or \n, and add the range to the class. */
4445
4446 if (d == CHAR_CR || d == CHAR_NL) cd->external_flags |= PCRE_HASCRORLF;
4447
4448 class_has_8bitchar +=
4449 add_to_class(classbits, &class_uchardata, options, cd, c, d);
4450
4451 continue; /* Go get the next char in the class */
4452 }
4453
4454 /* Handle a single character - we can get here for a normal non-escape
4455 char, or after \ that introduces a single character or for an apparent
4456 range that isn't. Only the value 1 matters for class_one_char, so don't
4457 increase it if it is already 2 or more ... just in case there's a class
4458 with a zillion characters in it. */
4459
4460 CLASS_SINGLE_CHARACTER:
4461 if (class_one_char < 2) class_one_char++;
4462
4463 /* If class_one_char is 1, we have the first single character in the
4464 class, and there have been no prior ranges, or XCLASS items generated by
4465 escapes. If this is the final character in the class, we can optimize by
4466 turning the item into a 1-character OP_CHAR[I] if it's positive, or
4467 OP_NOT[I] if it's negative. In the positive case, it can cause firstchar
4468 to be set. Otherwise, there can be no first char if this item is first,
4469 whatever repeat count may follow. In the case of reqchar, save the
4470 previous value for reinstating. */
4471
4472 if (class_one_char == 1 && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET)
4473 {
4474 ptr++;
4475 zeroreqchar = reqchar;
4476
4477 if (negate_class)
4478 {
4479 if (firstchar == REQ_UNSET) firstchar = REQ_NONE;
4480 zerofirstchar = firstchar;
4481 *code++ = ((options & PCRE_CASELESS) != 0)? OP_NOTI: OP_NOT;
4482 #ifdef SUPPORT_UTF
4483 if (utf && c > MAX_VALUE_FOR_SINGLE_CHAR)
4484 code += PRIV(ord2utf)(c, code);
4485 else
4486 #endif
4487 *code++ = c;
4488 goto NOT_CHAR;
4489 }
4490
4491 /* For a single, positive character, get the value into mcbuffer, and
4492 then we can handle this with the normal one-character code. */
4493
4494 #ifdef SUPPORT_UTF
4495 if (utf && c > MAX_VALUE_FOR_SINGLE_CHAR)
4496 mclength = PRIV(ord2utf)(c, mcbuffer);
4497 else
4498 #endif
4499 {
4500 mcbuffer[0] = c;
4501 mclength = 1;
4502 }
4503 goto ONE_CHAR;
4504 } /* End of 1-char optimization */
4505
4506 /* There is more than one character in the class, or an XCLASS item
4507 has been generated. Add this character to the class. */
4508
4509 class_has_8bitchar +=
4510 add_to_class(classbits, &class_uchardata, options, cd, c, c);
4511 }
4512
4513 /* Loop until ']' reached. This "while" is the end of the "do" far above.
4514 If we are at the end of an internal nested string, revert to the outer
4515 string. */
4516
4517 while (((c = *(++ptr)) != 0 ||
4518 (nestptr != NULL &&
4519 (ptr = nestptr, nestptr = NULL, c = *(++ptr)) != 0)) &&
4520 (c != CHAR_RIGHT_SQUARE_BRACKET || inescq));
4521
4522 /* Check for missing terminating ']' */
4523
4524 if (c == 0)
4525 {
4526 *errorcodeptr = ERR6;
4527 goto FAILED;
4528 }
4529
4530 /* We will need an XCLASS if data has been placed in class_uchardata. In
4531 the second phase this is a sufficient test. However, in the pre-compile
4532 phase, class_uchardata gets emptied to prevent workspace overflow, so it
4533 only if the very last character in the class needs XCLASS will it contain
4534 anything at this point. For this reason, xclass gets set TRUE above when
4535 uchar_classdata is emptied, and that's why this code is the way it is here
4536 instead of just doing a test on class_uchardata below. */
4537
4538 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
4539 if (class_uchardata > class_uchardata_base) xclass = TRUE;
4540 #endif
4541
4542 /* If this is the first thing in the branch, there can be no first char
4543 setting, whatever the repeat count. Any reqchar setting must remain
4544 unchanged after any kind of repeat. */
4545
4546 if (firstchar == REQ_UNSET) firstchar = REQ_NONE;
4547 zerofirstchar = firstchar;
4548 zeroreqchar = reqchar;
4549
4550 /* If there are characters with values > 255, we have to compile an
4551 extended class, with its own opcode, unless there was a negated special
4552 such as \S in the class, and PCRE_UCP is not set, because in that case all
4553 characters > 255 are in the class, so any that were explicitly given as
4554 well can be ignored. If (when there are explicit characters > 255 that must
4555 be listed) there are no characters < 256, we can omit the bitmap in the
4556 actual compiled code. */
4557
4558 #ifdef SUPPORT_UTF
4559 if (xclass && (!should_flip_negation || (options & PCRE_UCP) != 0))
4560 #elif !defined COMPILE_PCRE8
4561 if (xclass && !should_flip_negation)
4562 #endif
4563 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
4564 {
4565 *class_uchardata++ = XCL_END; /* Marks the end of extra data */
4566 *code++ = OP_XCLASS;
4567 code += LINK_SIZE;
4568 *code = negate_class? XCL_NOT:0;
4569
4570 /* If the map is required, move up the extra data to make room for it;
4571 otherwise just move the code pointer to the end of the extra data. */
4572
4573 if (class_has_8bitchar > 0)
4574 {
4575 *code++ |= XCL_MAP;
4576 memmove(code + (32 / sizeof(pcre_uchar)), code,
4577 IN_UCHARS(class_uchardata - code));
4578 memcpy(code, classbits, 32);
4579 code = class_uchardata + (32 / sizeof(pcre_uchar));
4580 }
4581 else code = class_uchardata;
4582
4583 /* Now fill in the complete length of the item */
4584
4585 PUT(previous, 1, (int)(code - previous));
4586 break; /* End of class handling */
4587 }
4588 #endif
4589
4590 /* If there are no characters > 255, or they are all to be included or
4591 excluded, set the opcode to OP_CLASS or OP_NCLASS, depending on whether the
4592 whole class was negated and whether there were negative specials such as \S
4593 (non-UCP) in the class. Then copy the 32-byte map into the code vector,
4594 negating it if necessary. */
4595
4596 *code++ = (negate_class == should_flip_negation) ? OP_CLASS : OP_NCLASS;
4597 if (lengthptr == NULL) /* Save time in the pre-compile phase */
4598 {
4599 if (negate_class)
4600 for (c = 0; c < 32; c++) classbits[c] = ~classbits[c];
4601 memcpy(code, classbits, 32);
4602 }
4603 code += 32 / sizeof(pcre_uchar);
4604 NOT_CHAR:
4605 break;
4606
4607
4608 /* ===================================================================*/
4609 /* Various kinds of repeat; '{' is not necessarily a quantifier, but this
4610 has been tested above. */
4611
4612 case CHAR_LEFT_CURLY_BRACKET:
4613 if (!is_quantifier) goto NORMAL_CHAR;
4614 ptr = read_repeat_counts(ptr+1, &repeat_min, &repeat_max, errorcodeptr);
4615 if (*errorcodeptr != 0) goto FAILED;
4616 goto REPEAT;
4617
4618 case CHAR_ASTERISK:
4619 repeat_min = 0;
4620 repeat_max = -1;
4621 goto REPEAT;
4622
4623 case CHAR_PLUS:
4624 repeat_min = 1;
4625 repeat_max = -1;
4626 goto REPEAT;
4627
4628 case CHAR_QUESTION_MARK:
4629 repeat_min = 0;
4630 repeat_max = 1;
4631
4632 REPEAT:
4633 if (previous == NULL)
4634 {
4635 *errorcodeptr = ERR9;
4636 goto FAILED;
4637 }
4638
4639 if (repeat_min == 0)
4640 {
4641 firstchar = zerofirstchar; /* Adjust for zero repeat */
4642 reqchar = zeroreqchar; /* Ditto */
4643 }
4644
4645 /* Remember whether this is a variable length repeat */
4646
4647 reqvary = (repeat_min == repeat_max)? 0 : REQ_VARY;
4648
4649 op_type = 0; /* Default single-char op codes */
4650 possessive_quantifier = FALSE; /* Default not possessive quantifier */
4651
4652 /* Save start of previous item, in case we have to move it up in order to
4653 insert something before it. */
4654
4655 tempcode = previous;
4656
4657 /* If the next character is '+', we have a possessive quantifier. This
4658 implies greediness, whatever the setting of the PCRE_UNGREEDY option.
4659 If the next character is '?' this is a minimizing repeat, by default,
4660 but if PCRE_UNGREEDY is set, it works the other way round. We change the
4661 repeat type to the non-default. */
4662
4663 if (ptr[1] == CHAR_PLUS)
4664 {
4665 repeat_type = 0; /* Force greedy */
4666 possessive_quantifier = TRUE;
4667 ptr++;
4668 }
4669 else if (ptr[1] == CHAR_QUESTION_MARK)
4670 {
4671 repeat_type = greedy_non_default;
4672 ptr++;
4673 }
4674 else repeat_type = greedy_default;
4675
4676 /* If previous was a recursion call, wrap it in atomic brackets so that
4677 previous becomes the atomic group. All recursions were so wrapped in the
4678 past, but it no longer happens for non-repeated recursions. In fact, the
4679 repeated ones could be re-implemented independently so as not to need this,
4680 but for the moment we rely on the code for repeating groups. */
4681
4682 if (*previous == OP_RECURSE)
4683 {
4684 memmove(previous + 1 + LINK_SIZE, previous, IN_UCHARS(1 + LINK_SIZE));
4685 *previous = OP_ONCE;
4686 PUT(previous, 1, 2 + 2*LINK_SIZE);
4687 previous[2 + 2*LINK_SIZE] = OP_KET;
4688 PUT(previous, 3 + 2*LINK_SIZE, 2 + 2*LINK_SIZE);
4689 code += 2 + 2 * LINK_SIZE;
4690 length_prevgroup = 3 + 3*LINK_SIZE;
4691
4692 /* When actually compiling, we need to check whether this was a forward
4693 reference, and if so, adjust the offset. */
4694
4695 if (lengthptr == NULL && cd->hwm >= cd->start_workspace + LINK_SIZE)
4696 {
4697 int offset = GET(cd->hwm, -LINK_SIZE);
4698 if (offset == previous + 1 - cd->start_code)
4699 PUT(cd->hwm, -LINK_SIZE, offset + 1 + LINK_SIZE);
4700 }
4701 }
4702
4703 /* Now handle repetition for the different types of item. */
4704
4705 /* If previous was a character or negated character match, abolish the item
4706 and generate a repeat item instead. If a char item has a minimum of more
4707 than one, ensure that it is set in reqchar - it might not be if a sequence
4708 such as x{3} is the first thing in a branch because the x will have gone
4709 into firstchar instead. */
4710
4711 if (*previous == OP_CHAR || *previous == OP_CHARI
4712 || *previous == OP_NOT || *previous == OP_NOTI)
4713 {
4714 switch (*previous)
4715 {
4716 default: /* Make compiler happy. */
4717 case OP_CHAR: op_type = OP_STAR - OP_STAR; break;
4718 case OP_CHARI: op_type = OP_STARI - OP_STAR; break;
4719 case OP_NOT: op_type = OP_NOTSTAR - OP_STAR; break;
4720 case OP_NOTI: op_type = OP_NOTSTARI - OP_STAR; break;
4721 }
4722
4723 /* Deal with UTF characters that take up more than one character. It's
4724 easier to write this out separately than try to macrify it. Use c to
4725 hold the length of the character in bytes, plus UTF_LENGTH to flag that
4726 it's a length rather than a small character. */
4727
4728 #ifdef SUPPORT_UTF
4729 if (utf && NOT_FIRSTCHAR(code[-1]))
4730 {
4731 pcre_uchar *lastchar = code - 1;
4732 BACKCHAR(lastchar);
4733 c = (int)(code - lastchar); /* Length of UTF-8 character */
4734 memcpy(utf_chars, lastchar, IN_UCHARS(c)); /* Save the char */
4735 c |= UTF_LENGTH; /* Flag c as a length */
4736 }
4737 else
4738 #endif /* SUPPORT_UTF */
4739
4740 /* Handle the case of a single charater - either with no UTF support, or
4741 with UTF disabled, or for a single character UTF character. */
4742 {
4743 c = code[-1];
4744 if (*previous <= OP_CHARI && repeat_min > 1)
4745 reqchar = c | req_caseopt | cd->req_varyopt;
4746 }
4747
4748 /* If the repetition is unlimited, it pays to see if the next thing on
4749 the line is something that cannot possibly match this character. If so,
4750 automatically possessifying this item gains some performance in the case
4751 where the match fails. */
4752
4753 if (!possessive_quantifier &&
4754 repeat_max < 0 &&
4755 check_auto_possessive(previous, utf, ptr + 1, options, cd))
4756 {
4757 repeat_type = 0; /* Force greedy */
4758 possessive_quantifier = TRUE;
4759 }
4760
4761 goto OUTPUT_SINGLE_REPEAT; /* Code shared with single character types */
4762 }
4763
4764 /* If previous was a character type match (\d or similar), abolish it and
4765 create a suitable repeat item. The code is shared with single-character
4766 repeats by setting op_type to add a suitable offset into repeat_type. Note
4767 the the Unicode property types will be present only when SUPPORT_UCP is
4768 defined, but we don't wrap the little bits of code here because it just
4769 makes it horribly messy. */
4770
4771 else if (*previous < OP_EODN)
4772 {
4773 pcre_uchar *oldcode;
4774 int prop_type, prop_value;
4775 op_type = OP_TYPESTAR - OP_STAR; /* Use type opcodes */
4776 c = *previous;
4777
4778 if (!possessive_quantifier &&
4779 repeat_max < 0 &&
4780 check_auto_possessive(previous, utf, ptr + 1, options, cd))
4781 {
4782 repeat_type = 0; /* Force greedy */
4783 possessive_quantifier = TRUE;
4784 }
4785
4786 OUTPUT_SINGLE_REPEAT:
4787 if (*previous == OP_PROP || *previous == OP_NOTPROP)
4788 {
4789 prop_type = previous[1];
4790 prop_value = previous[2];
4791 }
4792 else prop_type = prop_value = -1;
4793
4794 oldcode = code;
4795 code = previous; /* Usually overwrite previous item */
4796
4797 /* If the maximum is zero then the minimum must also be zero; Perl allows
4798 this case, so we do too - by simply omitting the item altogether. */
4799
4800 if (repeat_max == 0) goto END_REPEAT;
4801
4802 /*--------------------------------------------------------------------*/
4803 /* This code is obsolete from release 8.00; the restriction was finally
4804 removed: */
4805
4806 /* All real repeats make it impossible to handle partial matching (maybe
4807 one day we will be able to remove this restriction). */
4808
4809 /* if (repeat_max != 1) cd->external_flags |= PCRE_NOPARTIAL; */
4810 /*--------------------------------------------------------------------*/
4811
4812 /* Combine the op_type with the repeat_type */
4813
4814 repeat_type += op_type;
4815
4816 /* A minimum of zero is handled either as the special case * or ?, or as
4817 an UPTO, with the maximum given. */
4818
4819 if (repeat_min == 0)
4820 {
4821 if (repeat_max == -1) *code++ = OP_STAR + repeat_type;
4822 else if (repeat_max == 1) *code++ = OP_QUERY + repeat_type;
4823 else
4824 {
4825 *code++ = OP_UPTO + repeat_type;
4826 PUT2INC(code, 0, repeat_max);
4827 }
4828 }
4829
4830 /* A repeat minimum of 1 is optimized into some special cases. If the
4831 maximum is unlimited, we use OP_PLUS. Otherwise, the original item is
4832 left in place and, if the maximum is greater than 1, we use OP_UPTO with
4833 one less than the maximum. */
4834
4835 else if (repeat_min == 1)
4836 {
4837 if (repeat_max == -1)
4838 *code++ = OP_PLUS + repeat_type;
4839 else
4840 {
4841 code = oldcode; /* leave previous item in place */
4842 if (repeat_max == 1) goto END_REPEAT;
4843 *code++ = OP_UPTO + repeat_type;
4844 PUT2INC(code, 0, repeat_max - 1);
4845 }
4846 }
4847
4848 /* The case {n,n} is just an EXACT, while the general case {n,m} is
4849 handled as an EXACT followed by an UPTO. */
4850
4851 else
4852 {
4853 *code++ = OP_EXACT + op_type; /* NB EXACT doesn't have repeat_type */
4854 PUT2INC(code, 0, repeat_min);
4855
4856 /* If the maximum is unlimited, insert an OP_STAR. Before doing so,
4857 we have to insert the character for the previous code. For a repeated
4858 Unicode property match, there are two extra bytes that define the
4859 required property. In UTF-8 mode, long characters have their length in
4860 c, with the UTF_LENGTH bit as a flag. */
4861
4862 if (repeat_max < 0)
4863 {
4864 #ifdef SUPPORT_UTF
4865 if (utf && (c & UTF_LENGTH) != 0)
4866 {
4867 memcpy(code, utf_chars, IN_UCHARS(c & 7));
4868 code += c & 7;
4869 }
4870 else
4871 #endif
4872 {
4873 *code++ = c;
4874 if (prop_type >= 0)
4875 {
4876 *code++ = prop_type;
4877 *code++ = prop_value;
4878 }
4879 }
4880 *code++ = OP_STAR + repeat_type;
4881 }
4882
4883 /* Else insert an UPTO if the max is greater than the min, again
4884 preceded by the character, for the previously inserted code. If the
4885 UPTO is just for 1 instance, we can use QUERY instead. */
4886
4887 else if (repeat_max != repeat_min)
4888 {
4889 #ifdef SUPPORT_UTF
4890 if (utf && (c & UTF_LENGTH) != 0)
4891 {
4892 memcpy(code, utf_chars, IN_UCHARS(c & 7));
4893 code += c & 7;
4894 }
4895 else
4896 #endif
4897 *code++ = c;
4898 if (prop_type >= 0)
4899 {
4900 *code++ = prop_type;
4901 *code++ = prop_value;
4902 }
4903 repeat_max -= repeat_min;
4904
4905 if (repeat_max == 1)
4906 {
4907 *code++ = OP_QUERY + repeat_type;
4908 }
4909 else
4910 {
4911 *code++ = OP_UPTO + repeat_type;
4912 PUT2INC(code, 0, repeat_max);
4913 }
4914 }
4915 }
4916
4917 /* The character or character type itself comes last in all cases. */
4918
4919 #ifdef SUPPORT_UTF
4920 if (utf && (c & UTF_LENGTH) != 0)
4921 {
4922 memcpy(code, utf_chars, IN_UCHARS(c & 7));
4923 code += c & 7;
4924 }
4925 else
4926 #endif
4927 *code++ = c;
4928
4929 /* For a repeated Unicode property match, there are two extra bytes that
4930 define the required property. */
4931
4932 #ifdef SUPPORT_UCP
4933 if (prop_type >= 0)
4934 {
4935 *code++ = prop_type;
4936 *code++ = prop_value;
4937 }
4938 #endif
4939 }
4940
4941 /* If previous was a character class or a back reference, we put the repeat
4942 stuff after it, but just skip the item if the repeat was {0,0}. */
4943
4944 else if (*previous == OP_CLASS ||
4945 *previous == OP_NCLASS ||
4946 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
4947 *previous == OP_XCLASS ||
4948 #endif
4949 *previous == OP_REF ||
4950 *previous == OP_REFI)
4951 {
4952 if (repeat_max == 0)
4953 {
4954 code = previous;
4955 goto END_REPEAT;
4956 }
4957
4958 /*--------------------------------------------------------------------*/
4959 /* This code is obsolete from release 8.00; the restriction was finally
4960 removed: */
4961
4962 /* All real repeats make it impossible to handle partial matching (maybe
4963 one day we will be able to remove this restriction). */
4964
4965 /* if (repeat_max != 1) cd->external_flags |= PCRE_NOPARTIAL; */
4966 /*--------------------------------------------------------------------*/
4967
4968 if (repeat_min == 0 && repeat_max == -1)
4969 *code++ = OP_CRSTAR + repeat_type;
4970 else if (repeat_min == 1 && repeat_max == -1)
4971 *code++ = OP_CRPLUS + repeat_type;
4972 else if (repeat_min == 0 && repeat_max == 1)
4973 *code++ = OP_CRQUERY + repeat_type;
4974 else
4975 {
4976 *code++ = OP_CRRANGE + repeat_type;
4977 PUT2INC(code, 0, repeat_min);
4978 if (repeat_max == -1) repeat_max = 0; /* 2-byte encoding for max */
4979 PUT2INC(code, 0, repeat_max);
4980 }
4981 }
4982
4983 /* If previous was a bracket group, we may have to replicate it in certain
4984 cases. Note that at this point we can encounter only the "basic" bracket
4985 opcodes such as BRA and CBRA, as this is the place where they get converted
4986 into the more special varieties such as BRAPOS and SBRA. A test for >=
4987 OP_ASSERT and <= OP_COND includes ASSERT, ASSERT_NOT, ASSERTBACK,
4988 ASSERTBACK_NOT, ONCE, BRA, CBRA, and COND. Originally, PCRE did not allow
4989 repetition of assertions, but now it does, for Perl compatibility. */
4990
4991 else if (*previous >= OP_ASSERT && *previous <= OP_COND)
4992 {
4993 register int i;
4994 int len = (int)(code - previous);
4995 pcre_uchar *bralink = NULL;
4996 pcre_uchar *brazeroptr = NULL;
4997
4998 /* Repeating a DEFINE group is pointless, but Perl allows the syntax, so
4999 we just ignore the repeat. */
5000
5001 if (*previous == OP_COND && previous[LINK_SIZE+1] == OP_DEF)
5002 goto END_REPEAT;
5003
5004 /* There is no sense in actually repeating assertions. The only potential
5005 use of repetition is in cases when the assertion is optional. Therefore,
5006 if the minimum is greater than zero, just ignore the repeat. If the
5007 maximum is not not zero or one, set it to 1. */
5008
5009 if (*previous < OP_ONCE) /* Assertion */
5010 {
5011 if (repeat_min > 0) goto END_REPEAT;
5012 if (repeat_max < 0 || repeat_max > 1) repeat_max = 1;
5013 }
5014
5015 /* The case of a zero minimum is special because of the need to stick
5016 OP_BRAZERO in front of it, and because the group appears once in the
5017 data, whereas in other cases it appears the minimum number of times. For
5018 this reason, it is simplest to treat this case separately, as otherwise
5019 the code gets far too messy. There are several special subcases when the
5020 minimum is zero. */
5021
5022 if (repeat_min == 0)
5023 {
5024 /* If the maximum is also zero, we used to just omit the group from the
5025 output altogether, like this:
5026
5027 ** if (repeat_max == 0)
5028 ** {
5029 ** code = previous;
5030 ** goto END_REPEAT;
5031 ** }
5032
5033 However, that fails when a group or a subgroup within it is referenced
5034 as a subroutine from elsewhere in the pattern, so now we stick in
5035 OP_SKIPZERO in front of it so that it is skipped on execution. As we
5036 don't have a list of which groups are referenced, we cannot do this
5037 selectively.
5038
5039 If the maximum is 1 or unlimited, we just have to stick in the BRAZERO
5040 and do no more at this point. However, we do need to adjust any
5041 OP_RECURSE calls inside the group that refer to the group itself or any
5042 internal or forward referenced group, because the offset is from the
5043 start of the whole regex. Temporarily terminate the pattern while doing
5044 this. */
5045
5046 if (repeat_max <= 1) /* Covers 0, 1, and unlimited */
5047 {
5048 *code = OP_END;
5049 adjust_recurse(previous, 1, utf, cd, save_hwm);
5050 memmove(previous + 1, previous, IN_UCHARS(len));
5051 code++;
5052 if (repeat_max == 0)
5053 {
5054 *previous++ = OP_SKIPZERO;
5055 goto END_REPEAT;
5056 }
5057 brazeroptr = previous; /* Save for possessive optimizing */
5058 *previous++ = OP_BRAZERO + repeat_type;
5059 }
5060
5061 /* If the maximum is greater than 1 and limited, we have to replicate
5062 in a nested fashion, sticking OP_BRAZERO before each set of brackets.
5063 The first one has to be handled carefully because it's the original
5064 copy, which has to be moved up. The remainder can be handled by code
5065 that is common with the non-zero minimum case below. We have to
5066 adjust the value or repeat_max, since one less copy is required. Once
5067 again, we may have to adjust any OP_RECURSE calls inside the group. */
5068
5069 else
5070 {
5071 int offset;
5072 *code = OP_END;
5073 adjust_recurse(previous, 2 + LINK_SIZE, utf, cd, save_hwm);
5074 memmove(previous + 2 + LINK_SIZE, previous, IN_UCHARS(len));
5075 code += 2 + LINK_SIZE;
5076 *previous++ = OP_BRAZERO + repeat_type;
5077 *previous++ = OP_BRA;
5078
5079 /* We chain together the bracket offset fields that have to be
5080 filled in later when the ends of the brackets are reached. */
5081
5082 offset = (bralink == NULL)? 0 : (int)(previous - bralink);
5083 bralink = previous;
5084 PUTINC(previous, 0, offset);
5085 }
5086
5087 repeat_max--;
5088 }
5089
5090 /* If the minimum is greater than zero, replicate the group as many
5091 times as necessary, and adjust the maximum to the number of subsequent
5092 copies that we need. If we set a first char from the group, and didn't
5093 set a required char, copy the latter from the former. If there are any
5094 forward reference subroutine calls in the group, there will be entries on
5095 the workspace list; replicate these with an appropriate increment. */
5096
5097 else
5098 {
5099 if (repeat_min > 1)
5100 {
5101 /* In the pre-compile phase, we don't actually do the replication. We
5102 just adjust the length as if we had. Do some paranoid checks for
5103 potential integer overflow. The INT64_OR_DOUBLE type is a 64-bit
5104 integer type when available, otherwise double. */
5105
5106 if (lengthptr != NULL)
5107 {
5108 int delta = (repeat_min - 1)*length_prevgroup;
5109 if ((INT64_OR_DOUBLE)(repeat_min - 1)*
5110 (INT64_OR_DOUBLE)length_prevgroup >
5111 (INT64_OR_DOUBLE)INT_MAX ||
5112 OFLOW_MAX - *lengthptr < delta)
5113 {
5114 *errorcodeptr = ERR20;
5115 goto FAILED;
5116 }
5117 *lengthptr += delta;
5118 }
5119
5120 /* This is compiling for real. If there is a set first byte for
5121 the group, and we have not yet set a "required byte", set it. Make
5122 sure there is enough workspace for copying forward references before
5123 doing the copy. */
5124
5125 else
5126 {
5127 if (groupsetfirstchar && reqchar < 0) reqchar = firstchar;
5128
5129 for (i = 1; i < repeat_min; i++)
5130 {
5131 pcre_uchar *hc;
5132 pcre_uchar *this_hwm = cd->hwm;
5133 memcpy(code, previous, IN_UCHARS(len));
5134
5135 while (cd->hwm > cd->start_workspace + cd->workspace_size -
5136 WORK_SIZE_SAFETY_MARGIN - (this_hwm - save_hwm))
5137 {
5138 int save_offset = save_hwm - cd->start_workspace;
5139 int this_offset = this_hwm - cd->start_workspace;
5140 *errorcodeptr = expand_workspace(cd);
5141 if (*errorcodeptr != 0) goto FAILED;
5142 save_hwm = (pcre_uchar *)cd->start_workspace + save_offset;
5143 this_hwm = (pcre_uchar *)cd->start_workspace + this_offset;
5144 }
5145
5146 for (hc = save_hwm; hc < this_hwm; hc += LINK_SIZE)
5147 {
5148 PUT(cd->hwm, 0, GET(hc, 0) + len);
5149 cd->hwm += LINK_SIZE;
5150 }
5151 save_hwm = this_hwm;
5152 code += len;
5153 }
5154 }
5155 }
5156
5157 if (repeat_max > 0) repeat_max -= repeat_min;
5158 }
5159
5160 /* This code is common to both the zero and non-zero minimum cases. If
5161 the maximum is limited, it replicates the group in a nested fashion,
5162 remembering the bracket starts on a stack. In the case of a zero minimum,
5163 the first one was set up above. In all cases the repeat_max now specifies
5164 the number of additional copies needed. Again, we must remember to
5165 replicate entries on the forward reference list. */
5166
5167 if (repeat_max >= 0)
5168 {
5169 /* In the pre-compile phase, we don't actually do the replication. We
5170 just adjust the length as if we had. For each repetition we must add 1
5171 to the length for BRAZERO and for all but the last repetition we must
5172 add 2 + 2*LINKSIZE to allow for the nesting that occurs. Do some
5173 paranoid checks to avoid integer overflow. The INT64_OR_DOUBLE type is
5174 a 64-bit integer type when available, otherwise double. */
5175
5176 if (lengthptr != NULL && repeat_max > 0)
5177 {
5178 int delta = repeat_max * (length_prevgroup + 1 + 2 + 2*LINK_SIZE) -
5179 2 - 2*LINK_SIZE; /* Last one doesn't nest */
5180 if ((INT64_OR_DOUBLE)repeat_max *
5181 (INT64_OR_DOUBLE)(length_prevgroup + 1 + 2 + 2*LINK_SIZE)
5182 > (INT64_OR_DOUBLE)INT_MAX ||
5183 OFLOW_MAX - *lengthptr < delta)
5184 {
5185 *errorcodeptr = ERR20;
5186 goto FAILED;
5187 }
5188 *lengthptr += delta;
5189 }
5190
5191 /* This is compiling for real */
5192
5193 else for (i = repeat_max - 1; i >= 0; i--)
5194 {
5195 pcre_uchar *hc;
5196 pcre_uchar *this_hwm = cd->hwm;
5197
5198 *code++ = OP_BRAZERO + repeat_type;
5199
5200 /* All but the final copy start a new nesting, maintaining the
5201 chain of brackets outstanding. */
5202
5203 if (i != 0)
5204 {
5205 int offset;
5206 *code++ = OP_BRA;
5207 offset = (bralink == NULL)? 0 : (int)(code - bralink);
5208 bralink = code;
5209 PUTINC(code, 0, offset);
5210 }
5211
5212 memcpy(code, previous, IN_UCHARS(len));
5213
5214 /* Ensure there is enough workspace for forward references before
5215 copying them. */
5216
5217 while (cd->hwm > cd->start_workspace + cd->workspace_size -
5218 WORK_SIZE_SAFETY_MARGIN - (this_hwm - save_hwm))
5219 {
5220 int save_offset = save_hwm - cd->start_workspace;
5221 int this_offset = this_hwm - cd->start_workspace;
5222 *errorcodeptr = expand_workspace(cd);
5223 if (*errorcodeptr != 0) goto FAILED;
5224 save_hwm = (pcre_uchar *)cd->start_workspace + save_offset;
5225 this_hwm = (pcre_uchar *)cd->start_workspace + this_offset;
5226 }
5227
5228 for (hc = save_hwm; hc < this_hwm; hc += LINK_SIZE)
5229 {
5230 PUT(cd->hwm, 0, GET(hc, 0) + len + ((i != 0)? 2+LINK_SIZE : 1));
5231 cd->hwm += LINK_SIZE;
5232 }
5233 save_hwm = this_hwm;
5234 code += len;
5235 }
5236
5237 /* Now chain through the pending brackets, and fill in their length
5238 fields (which are holding the chain links pro tem). */
5239
5240 while (bralink != NULL)
5241 {
5242 int oldlinkoffset;
5243 int offset = (int)(code - bralink + 1);
5244 pcre_uchar *bra = code - offset;
5245 oldlinkoffset = GET(bra, 1);
5246 bralink = (oldlinkoffset == 0)? NULL : bralink - oldlinkoffset;
5247 *code++ = OP_KET;
5248 PUTINC(code, 0, offset);
5249 PUT(bra, 1, offset);
5250 }
5251 }
5252
5253 /* If the maximum is unlimited, set a repeater in the final copy. For
5254 ONCE brackets, that's all we need to do. However, possessively repeated
5255 ONCE brackets can be converted into non-capturing brackets, as the
5256 behaviour of (?:xx)++ is the same as (?>xx)++ and this saves having to
5257 deal with possessive ONCEs specially.
5258
5259 Otherwise, when we are doing the actual compile phase, check to see
5260 whether this group is one that could match an empty string. If so,
5261 convert the initial operator to the S form (e.g. OP_BRA -> OP_SBRA) so
5262 that runtime checking can be done. [This check is also applied to ONCE
5263 groups at runtime, but in a different way.]
5264
5265 Then, if the quantifier was possessive and the bracket is not a
5266 conditional, we convert the BRA code to the POS form, and the KET code to
5267 KETRPOS. (It turns out to be convenient at runtime to detect this kind of
5268 subpattern at both the start and at the end.) The use of special opcodes
5269 makes it possible to reduce greatly the stack usage in pcre_exec(). If
5270 the group is preceded by OP_BRAZERO, convert this to OP_BRAPOSZERO.
5271
5272 Then, if the minimum number of matches is 1 or 0, cancel the possessive
5273 flag so that the default action below, of wrapping everything inside
5274 atomic brackets, does not happen. When the minimum is greater than 1,
5275 there will be earlier copies of the group, and so we still have to wrap
5276 the whole thing. */
5277
5278 else
5279 {
5280 pcre_uchar *ketcode = code - 1 - LINK_SIZE;
5281 pcre_uchar *bracode = ketcode - GET(ketcode, 1);
5282
5283 /* Convert possessive ONCE brackets to non-capturing */
5284
5285 if ((*bracode == OP_ONCE || *bracode == OP_ONCE_NC) &&
5286 possessive_quantifier) *bracode = OP_BRA;
5287
5288 /* For non-possessive ONCE brackets, all we need to do is to
5289 set the KET. */
5290
5291 if (*bracode == OP_ONCE || *bracode == OP_ONCE_NC)
5292 *ketcode = OP_KETRMAX + repeat_type;
5293
5294 /* Handle non-ONCE brackets and possessive ONCEs (which have been
5295 converted to non-capturing above). */
5296
5297 else
5298 {
5299 /* In the compile phase, check for empty string matching. */
5300
5301 if (lengthptr == NULL)
5302 {
5303 pcre_uchar *scode = bracode;
5304 do
5305 {
5306 if (could_be_empty_branch(scode, ketcode, utf, cd))
5307 {
5308 *bracode += OP_SBRA - OP_BRA;
5309 break;
5310 }
5311 scode += GET(scode, 1);
5312 }
5313 while (*scode == OP_ALT);
5314 }
5315
5316 /* Handle possessive quantifiers. */
5317
5318 if (possessive_quantifier)
5319 {
5320 /* For COND brackets, we wrap the whole thing in a possessively
5321 repeated non-capturing bracket, because we have not invented POS
5322 versions of the COND opcodes. Because we are moving code along, we
5323 must ensure that any pending recursive references are updated. */
5324
5325 if (*bracode == OP_COND || *bracode == OP_SCOND)
5326 {
5327 int nlen = (int)(code - bracode);
5328 *code = OP_END;
5329 adjust_recurse(bracode, 1 + LINK_SIZE, utf, cd, save_hwm);
5330 memmove(bracode + 1 + LINK_SIZE, bracode, IN_UCHARS(nlen));
5331 code += 1 + LINK_SIZE;
5332 nlen += 1 + LINK_SIZE;
5333 *bracode = OP_BRAPOS;
5334 *code++ = OP_KETRPOS;
5335 PUTINC(code, 0, nlen);
5336 PUT(bracode, 1, nlen);
5337 }
5338
5339 /* For non-COND brackets, we modify the BRA code and use KETRPOS. */
5340
5341 else
5342 {
5343 *bracode += 1; /* Switch to xxxPOS opcodes */
5344 *ketcode = OP_KETRPOS;
5345 }
5346
5347 /* If the minimum is zero, mark it as possessive, then unset the
5348 possessive flag when the minimum is 0 or 1. */
5349
5350 if (brazeroptr != NULL) *brazeroptr = OP_BRAPOSZERO;
5351 if (repeat_min < 2) possessive_quantifier = FALSE;
5352 }
5353
5354 /* Non-possessive quantifier */
5355
5356 else *ketcode = OP_KETRMAX + repeat_type;
5357 }
5358 }
5359 }
5360
5361 /* If previous is OP_FAIL, it was generated by an empty class [] in
5362 JavaScript mode. The other ways in which OP_FAIL can be generated, that is
5363 by (*FAIL) or (?!) set previous to NULL, which gives a "nothing to repeat"
5364 error above. We can just ignore the repeat in JS case. */
5365
5366 else if (*previous == OP_FAIL) goto END_REPEAT;
5367
5368 /* Else there's some kind of shambles */
5369
5370 else
5371 {
5372 *errorcodeptr = ERR11;
5373 goto FAILED;
5374 }
5375
5376 /* If the character following a repeat is '+', or if certain optimization
5377 tests above succeeded, possessive_quantifier is TRUE. For some opcodes,
5378 there are special alternative opcodes for this case. For anything else, we
5379 wrap the entire repeated item inside OP_ONCE brackets. Logically, the '+'
5380 notation is just syntactic sugar, taken from Sun's Java package, but the
5381 special opcodes can optimize it.
5382
5383 Some (but not all) possessively repeated subpatterns have already been
5384 completely handled in the code just above. For them, possessive_quantifier
5385 is always FALSE at this stage.
5386
5387 Note that the repeated item starts at tempcode, not at previous, which
5388 might be the first part of a string whose (former) last char we repeated.
5389
5390 Possessifying an 'exact' quantifier has no effect, so we can ignore it. But
5391 an 'upto' may follow. We skip over an 'exact' item, and then test the
5392 length of what remains before proceeding. */
5393
5394 if (possessive_quantifier)
5395 {
5396 int len;
5397
5398 if (*tempcode == OP_TYPEEXACT)
5399 tempcode += PRIV(OP_lengths)[*tempcode] +
5400 ((tempcode[1 + IMM2_SIZE] == OP_PROP
5401 || tempcode[1 + IMM2_SIZE] == OP_NOTPROP)? 2 : 0);
5402
5403 else if (*tempcode == OP_EXACT || *tempcode == OP_NOTEXACT)
5404 {
5405 tempcode += PRIV(OP_lengths)[*tempcode];
5406 #ifdef SUPPORT_UTF
5407 if (utf && HAS_EXTRALEN(tempcode[-1]))
5408 tempcode += GET_EXTRALEN(tempcode[-1]);
5409 #endif
5410 }
5411
5412 len = (int)(code - tempcode);
5413 if (len > 0) switch (*tempcode)
5414 {
5415 case OP_STAR: *tempcode = OP_POSSTAR; break;
5416 case OP_PLUS: *tempcode = OP_POSPLUS; break;
5417 case OP_QUERY: *tempcode = OP_POSQUERY; break;
5418 case OP_UPTO: *tempcode = OP_POSUPTO; break;
5419
5420 case OP_STARI: *tempcode = OP_POSSTARI; break;
5421 case OP_PLUSI: *tempcode = OP_POSPLUSI; break;
5422 case OP_QUERYI: *tempcode = OP_POSQUERYI; break;
5423 case OP_UPTOI: *tempcode = OP_POSUPTOI; break;
5424
5425 case OP_NOTSTAR: *tempcode = OP_NOTPOSSTAR; break;
5426 case OP_NOTPLUS: *tempcode = OP_NOTPOSPLUS; break;
5427 case OP_NOTQUERY: *tempcode = OP_NOTPOSQUERY; break;
5428 case OP_NOTUPTO: *tempcode = OP_NOTPOSUPTO; break;
5429
5430 case OP_NOTSTARI: *tempcode = OP_NOTPOSSTARI; break;
5431 case OP_NOTPLUSI: *tempcode = OP_NOTPOSPLUSI; break;
5432 case OP_NOTQUERYI: *tempcode = OP_NOTPOSQUERYI; break;
5433 case OP_NOTUPTOI: *tempcode = OP_NOTPOSUPTOI; break;
5434
5435 case OP_TYPESTAR: *tempcode = OP_TYPEPOSSTAR; break;
5436 case OP_TYPEPLUS: *tempcode = OP_TYPEPOSPLUS; break;
5437 case OP_TYPEQUERY: *tempcode = OP_TYPEPOSQUERY; break;
5438 case OP_TYPEUPTO: *tempcode = OP_TYPEPOSUPTO; break;
5439
5440 /* Because we are moving code along, we must ensure that any
5441 pending recursive references are updated. */
5442
5443 default:
5444 *code = OP_END;
5445 adjust_recurse(tempcode, 1 + LINK_SIZE, utf, cd, save_hwm);
5446 memmove(tempcode + 1 + LINK_SIZE, tempcode, IN_UCHARS(len));
5447 code += 1 + LINK_SIZE;
5448 len += 1 + LINK_SIZE;
5449 tempcode[0] = OP_ONCE;
5450 *code++ = OP_KET;
5451 PUTINC(code, 0, len);
5452 PUT(tempcode, 1, len);
5453 break;
5454 }
5455 }
5456
5457 /* In all case we no longer have a previous item. We also set the
5458 "follows varying string" flag for subsequently encountered reqchars if
5459 it isn't already set and we have just passed a varying length item. */
5460
5461 END_REPEAT:
5462 previous = NULL;
5463 cd->req_varyopt |= reqvary;
5464 break;
5465
5466
5467 /* ===================================================================*/
5468 /* Start of nested parenthesized sub-expression, or comment or lookahead or
5469 lookbehind or option setting or condition or all the other extended
5470 parenthesis forms. */
5471
5472 case CHAR_LEFT_PARENTHESIS:
5473 newoptions = options;
5474 skipbytes = 0;
5475 bravalue = OP_CBRA;
5476 save_hwm = cd->hwm;
5477 reset_bracount = FALSE;
5478
5479 /* First deal with various "verbs" that can be introduced by '*'. */
5480
5481 ptr++;
5482 if (ptr[0] == CHAR_ASTERISK && (ptr[1] == ':'
5483 || (MAX_255(ptr[1]) && ((cd->ctypes[ptr[1]] & ctype_letter) != 0))))
5484 {
5485 int i, namelen;
5486 int arglen = 0;
5487 const char *vn = verbnames;
5488 const pcre_uchar *name = ptr + 1;
5489 const pcre_uchar *arg = NULL;
5490 previous = NULL;
5491 ptr++;
5492 while (MAX_255(*ptr) && (cd->ctypes[*ptr] & ctype_letter) != 0) ptr++;
5493 namelen = (int)(ptr - name);
5494
5495 /* It appears that Perl allows any characters whatsoever, other than
5496 a closing parenthesis, to appear in arguments, so we no longer insist on
5497 letters, digits, and underscores. */
5498
5499 if (*ptr == CHAR_COLON)
5500 {
5501 arg = ++ptr;
5502 while (*ptr != 0 && *ptr != CHAR_RIGHT_PARENTHESIS) ptr++;
5503 arglen = (int)(ptr - arg);
5504 if (arglen > (int)MAX_MARK)
5505 {
5506 *errorcodeptr = ERR75;
5507 goto FAILED;
5508 }
5509 }
5510
5511 if (*ptr != CHAR_RIGHT_PARENTHESIS)
5512 {
5513 *errorcodeptr = ERR60;
5514 goto FAILED;
5515 }
5516
5517 /* Scan the table of verb names */
5518
5519 for (i = 0; i < verbcount; i++)
5520 {
5521 if (namelen == verbs[i].len &&
5522 STRNCMP_UC_C8(name, vn, namelen) == 0)
5523 {
5524 int setverb;
5525
5526 /* Check for open captures before ACCEPT and convert it to
5527 ASSERT_ACCEPT if in an assertion. */
5528
5529 if (verbs[i].op == OP_ACCEPT)
5530 {
5531 open_capitem *oc;
5532 if (arglen != 0)
5533 {
5534 *errorcodeptr = ERR59;
5535 goto FAILED;
5536 }
5537 cd->had_accept = TRUE;
5538 for (oc = cd->open_caps; oc != NULL; oc = oc->next)
5539 {
5540 *code++ = OP_CLOSE;
5541 PUT2INC(code, 0, oc->number);
5542 }
5543 setverb = *code++ =
5544 (cd->assert_depth > 0)? OP_ASSERT_ACCEPT : OP_ACCEPT;
5545
5546 /* Do not set firstchar after *ACCEPT */
5547 if (firstchar == REQ_UNSET) firstchar = REQ_NONE;
5548 }
5549
5550 /* Handle other cases with/without an argument */
5551
5552 else if (arglen == 0)
5553 {
5554 if (verbs[i].op < 0) /* Argument is mandatory */
5555 {
5556 *errorcodeptr = ERR66;
5557 goto FAILED;
5558 }
5559 setverb = *code++ = verbs[i].op;
5560 }
5561
5562 else
5563 {
5564 if (verbs[i].op_arg < 0) /* Argument is forbidden */
5565 {
5566 *errorcodeptr = ERR59;
5567 goto FAILED;
5568 }
5569 setverb = *code++ = verbs[i].op_arg;
5570 *code++ = arglen;
5571 memcpy(code, arg, IN_UCHARS(arglen));
5572 code += arglen;
5573 *code++ = 0;
5574 }
5575
5576 switch (setverb)
5577 {
5578 case OP_THEN:
5579 case OP_THEN_ARG:
5580 cd->external_flags |= PCRE_HASTHEN;
5581 break;
5582
5583 case OP_PRUNE:
5584 case OP_PRUNE_ARG:
5585 case OP_SKIP:
5586 case OP_SKIP_ARG:
5587 cd->had_pruneorskip = TRUE;
5588 break;
5589 }
5590
5591 break; /* Found verb, exit loop */
5592 }
5593
5594 vn += verbs[i].len + 1;
5595 }
5596
5597 if (i < verbcount) continue; /* Successfully handled a verb */
5598 *errorcodeptr = ERR60; /* Verb not recognized */
5599 goto FAILED;
5600 }
5601
5602 /* Deal with the extended parentheses; all are introduced by '?', and the
5603 appearance of any of them means that this is not a capturing group. */
5604
5605 else if (*ptr == CHAR_QUESTION_MARK)
5606 {
5607 int i, set, unset, namelen;
5608 int *optset;
5609 const pcre_uchar *name;
5610 pcre_uchar *slot;
5611
5612 switch (*(++ptr))
5613 {
5614 case CHAR_NUMBER_SIGN: /* Comment; skip to ket */
5615 ptr++;
5616 while (*ptr != 0 && *ptr != CHAR_RIGHT_PARENTHESIS) ptr++;
5617 if (*ptr == 0)
5618 {
5619 *errorcodeptr = ERR18;
5620 goto FAILED;
5621 }
5622 continue;
5623
5624
5625 /* ------------------------------------------------------------ */
5626 case CHAR_VERTICAL_LINE: /* Reset capture count for each branch */
5627 reset_bracount = TRUE;
5628 /* Fall through */
5629
5630 /* ------------------------------------------------------------ */
5631 case CHAR_COLON: /* Non-capturing bracket */
5632 bravalue = OP_BRA;
5633 ptr++;
5634 break;
5635
5636
5637 /* ------------------------------------------------------------ */
5638 case CHAR_LEFT_PARENTHESIS:
5639 bravalue = OP_COND; /* Conditional group */
5640
5641 /* A condition can be an assertion, a number (referring to a numbered
5642 group), a name (referring to a named group), or 'R', referring to
5643 recursion. R<digits> and R&name are also permitted for recursion tests.
5644
5645 There are several syntaxes for testing a named group: (?(name)) is used
5646 by Python; Perl 5.10 onwards uses (?(<name>) or (?('name')).
5647
5648 There are two unfortunate ambiguities, caused by history. (a) 'R' can
5649 be the recursive thing or the name 'R' (and similarly for 'R' followed
5650 by digits), and (b) a number could be a name that consists of digits.
5651 In both cases, we look for a name first; if not found, we try the other
5652 cases. */
5653
5654 /* For conditions that are assertions, check the syntax, and then exit
5655 the switch. This will take control down to where bracketed groups,
5656 including assertions, are processed. */
5657
5658 if (ptr[1] == CHAR_QUESTION_MARK && (ptr[2] == CHAR_EQUALS_SIGN ||
5659 ptr[2] == CHAR_EXCLAMATION_MARK || ptr[2] == CHAR_LESS_THAN_SIGN))
5660 break;
5661
5662 /* Most other conditions use OP_CREF (a couple change to OP_RREF
5663 below), and all need to skip 1+IMM2_SIZE bytes at the start of the group. */
5664
5665 code[1+LINK_SIZE] = OP_CREF;
5666 skipbytes = 1+IMM2_SIZE;
5667 refsign = -1;
5668
5669 /* Check for a test for recursion in a named group. */
5670
5671 if (ptr[1] == CHAR_R && ptr[2] == CHAR_AMPERSAND)
5672 {
5673 terminator = -1;
5674 ptr += 2;
5675 code[1+LINK_SIZE] = OP_RREF; /* Change the type of test */
5676 }
5677
5678 /* Check for a test for a named group's having been set, using the Perl
5679 syntax (?(<name>) or (?('name') */
5680
5681 else if (ptr[1] == CHAR_LESS_THAN_SIGN)
5682 {
5683 terminator = CHAR_GREATER_THAN_SIGN;
5684 ptr++;
5685 }
5686 else if (ptr[1] == CHAR_APOSTROPHE)
5687 {
5688 terminator = CHAR_APOSTROPHE;
5689 ptr++;
5690 }
5691 else
5692 {
5693 terminator = 0;
5694 if (ptr[1] == CHAR_MINUS || ptr[1] == CHAR_PLUS) refsign = *(++ptr);
5695 }
5696
5697 /* We now expect to read a name; any thing else is an error */
5698
5699 if (!MAX_255(ptr[1]) || (cd->ctypes[ptr[1]] & ctype_word) == 0)
5700 {
5701 ptr += 1; /* To get the right offset */
5702 *errorcodeptr = ERR28;
5703 goto FAILED;
5704 }
5705
5706 /* Read the name, but also get it as a number if it's all digits */
5707
5708 recno = 0;
5709 name = ++ptr;
5710 while (MAX_255(*ptr) && (cd->ctypes[*ptr] & ctype_word) != 0)
5711 {
5712 if (recno >= 0)
5713 recno = (IS_DIGIT(*ptr))? recno * 10 + *ptr - CHAR_0 : -1;
5714 ptr++;
5715 }
5716 namelen = (int)(ptr - name);
5717
5718 if ((terminator > 0 && *ptr++ != terminator) ||
5719 *ptr++ != CHAR_RIGHT_PARENTHESIS)
5720 {
5721 ptr--; /* Error offset */
5722 *errorcodeptr = ERR26;
5723 goto FAILED;
5724 }
5725
5726 /* Do no further checking in the pre-compile phase. */
5727
5728 if (lengthptr != NULL) break;
5729
5730 /* In the real compile we do the work of looking for the actual
5731 reference. If the string started with "+" or "-" we require the rest to
5732 be digits, in which case recno will be set. */
5733
5734 if (refsign > 0)
5735 {
5736 if (recno <= 0)
5737 {
5738 *errorcodeptr = ERR58;
5739 goto FAILED;
5740 }
5741 recno = (refsign == CHAR_MINUS)?
5742 cd->bracount - recno + 1 : recno +cd->bracount;
5743 if (recno <= 0 || recno > cd->final_bracount)
5744 {
5745 *errorcodeptr = ERR15;
5746 goto FAILED;
5747 }
5748 PUT2(code, 2+LINK_SIZE, recno);
5749 break;
5750 }
5751
5752 /* Otherwise (did not start with "+" or "-"), start by looking for the
5753 name. If we find a name, add one to the opcode to change OP_CREF or
5754 OP_RREF into OP_NCREF or OP_NRREF. These behave exactly the same,
5755 except they record that the reference was originally to a name. The
5756 information is used to check duplicate names. */
5757
5758 slot = cd->name_table;
5759 for (i = 0; i < cd->names_found; i++)
5760 {
5761 if (STRNCMP_UC_UC(name, slot+IMM2_SIZE, namelen) == 0) break;
5762 slot += cd->name_entry_size;
5763 }
5764
5765 /* Found a previous named subpattern */
5766
5767 if (i < cd->names_found)
5768 {
5769 recno = GET2(slot, 0);
5770 PUT2(code, 2+LINK_SIZE, recno);
5771 code[1+LINK_SIZE]++;
5772 }
5773
5774 /* Search the pattern for a forward reference */
5775
5776 else if ((i = find_parens(cd, name, namelen,
5777 (options & PCRE_EXTENDED) != 0, utf)) > 0)
5778 {
5779 PUT2(code, 2+LINK_SIZE, i);
5780 code[1+LINK_SIZE]++;
5781 }
5782
5783 /* If terminator == 0 it means that the name followed directly after
5784 the opening parenthesis [e.g. (?(abc)...] and in this case there are
5785 some further alternatives to try. For the cases where terminator != 0
5786 [things like (?(<name>... or (?('name')... or (?(R&name)... ] we have
5787 now checked all the possibilities, so give an error. */
5788
5789 else if (terminator != 0)
5790 {
5791 *errorcodeptr = ERR15;
5792 goto FAILED;
5793 }
5794
5795 /* Check for (?(R) for recursion. Allow digits after R to specify a
5796 specific group number. */
5797
5798 else if (*name == CHAR_R)
5799 {
5800 recno = 0;
5801 for (i = 1; i < namelen; i++)
5802 {
5803 if (!IS_DIGIT(name[i]))
5804 {
5805 *errorcodeptr = ERR15;
5806 goto FAILED;
5807 }
5808 recno = recno * 10 + name[i] - CHAR_0;
5809 }
5810 if (recno == 0) recno = RREF_ANY;
5811 code[1+LINK_SIZE] = OP_RREF; /* Change test type */
5812 PUT2(code, 2+LINK_SIZE, recno);
5813 }
5814
5815 /* Similarly, check for the (?(DEFINE) "condition", which is always
5816 false. */
5817
5818 else if (namelen == 6 && STRNCMP_UC_C8(name, STRING_DEFINE, 6) == 0)
5819 {
5820 code[1+LINK_SIZE] = OP_DEF;
5821 skipbytes = 1;
5822 }
5823
5824 /* Check for the "name" actually being a subpattern number. We are
5825 in the second pass here, so final_bracount is set. */
5826
5827 else if (recno > 0 && recno <= cd->final_bracount)
5828 {
5829 PUT2(code, 2+LINK_SIZE, recno);
5830 }
5831
5832 /* Either an unidentified subpattern, or a reference to (?(0) */
5833
5834 else
5835 {
5836 *errorcodeptr = (recno == 0)? ERR35: ERR15;
5837 goto FAILED;
5838 }
5839 break;
5840
5841
5842 /* ------------------------------------------------------------ */
5843 case CHAR_EQUALS_SIGN: /* Positive lookahead */
5844 bravalue = OP_ASSERT;
5845 cd->assert_depth += 1;
5846 ptr++;
5847 break;
5848
5849
5850 /* ------------------------------------------------------------ */
5851 case CHAR_EXCLAMATION_MARK: /* Negative lookahead */
5852 ptr++;
5853 if (*ptr == CHAR_RIGHT_PARENTHESIS) /* Optimize (?!) */
5854 {
5855 *code++ = OP_FAIL;
5856 previous = NULL;
5857 continue;
5858 }
5859 bravalue = OP_ASSERT_NOT;
5860 cd->assert_depth += 1;
5861 break;
5862
5863
5864 /* ------------------------------------------------------------ */
5865 case CHAR_LESS_THAN_SIGN: /* Lookbehind or named define */
5866 switch (ptr[1])
5867 {
5868 case CHAR_EQUALS_SIGN: /* Positive lookbehind */
5869 bravalue = OP_ASSERTBACK;
5870 cd->assert_depth += 1;
5871 ptr += 2;
5872 break;
5873
5874 case CHAR_EXCLAMATION_MARK: /* Negative lookbehind */
5875 bravalue = OP_ASSERTBACK_NOT;
5876 cd->assert_depth += 1;
5877 ptr += 2;
5878 break;
5879
5880 default: /* Could be name define, else bad */
5881 if (MAX_255(ptr[1]) && (cd->ctypes[ptr[1]] & ctype_word) != 0)
5882 goto DEFINE_NAME;
5883 ptr++; /* Correct offset for error */
5884 *errorcodeptr = ERR24;
5885 goto FAILED;
5886 }
5887 break;
5888
5889
5890 /* ------------------------------------------------------------ */
5891 case CHAR_GREATER_THAN_SIGN: /* One-time brackets */
5892 bravalue = OP_ONCE;
5893 ptr++;
5894 break;
5895
5896
5897 /* ------------------------------------------------------------ */
5898 case CHAR_C: /* Callout - may be followed by digits; */
5899 previous_callout = code; /* Save for later completion */
5900 after_manual_callout = 1; /* Skip one item before completing */
5901 *code++ = OP_CALLOUT;
5902 {
5903 int n = 0;
5904 ptr++;
5905 while(IS_DIGIT(*ptr))
5906 n = n * 10 + *ptr++ - CHAR_0;
5907 if (*ptr != CHAR_RIGHT_PARENTHESIS)
5908 {
5909 *errorcodeptr = ERR39;
5910 goto FAILED;
5911 }
5912 if (n > 255)
5913 {
5914 *errorcodeptr = ERR38;
5915 goto FAILED;
5916 }
5917 *code++ = n;
5918 PUT(code, 0, (int)(ptr - cd->start_pattern + 1)); /* Pattern offset */
5919 PUT(code, LINK_SIZE, 0); /* Default length */
5920 code += 2 * LINK_SIZE;
5921 }
5922 previous = NULL;
5923 continue;
5924
5925
5926 /* ------------------------------------------------------------ */
5927 case CHAR_P: /* Python-style named subpattern handling */
5928 if (*(++ptr) == CHAR_EQUALS_SIGN ||
5929 *ptr == CHAR_GREATER_THAN_SIGN) /* Reference or recursion */
5930 {
5931 is_recurse = *ptr == CHAR_GREATER_THAN_SIGN;
5932 terminator = CHAR_RIGHT_PARENTHESIS;
5933 goto NAMED_REF_OR_RECURSE;
5934 }
5935 else if (*ptr != CHAR_LESS_THAN_SIGN) /* Test for Python-style defn */
5936 {
5937 *errorcodeptr = ERR41;
5938 goto FAILED;
5939 }
5940 /* Fall through to handle (?P< as (?< is handled */
5941
5942
5943 /* ------------------------------------------------------------ */
5944 DEFINE_NAME: /* Come here from (?< handling */
5945 case CHAR_APOSTROPHE:
5946 {
5947 terminator = (*ptr == CHAR_LESS_THAN_SIGN)?
5948 CHAR_GREATER_THAN_SIGN : CHAR_APOSTROPHE;
5949 name = ++ptr;
5950
5951 while (MAX_255(*ptr) && (cd->ctypes[*ptr] & ctype_word) != 0) ptr++;
5952 namelen = (int)(ptr - name);
5953
5954 /* In the pre-compile phase, just do a syntax check. */
5955
5956 if (lengthptr != NULL)
5957 {
5958 if (*ptr != terminator)
5959 {
5960 *errorcodeptr = ERR42;
5961 goto FAILED;
5962 }
5963 if (cd->names_found >= MAX_NAME_COUNT)
5964 {
5965 *errorcodeptr = ERR49;
5966 goto FAILED;
5967 }
5968 if (namelen + IMM2_SIZE + 1 > cd->name_entry_size)
5969 {
5970 cd->name_entry_size = namelen + IMM2_SIZE + 1;
5971 if (namelen > MAX_NAME_SIZE)
5972 {
5973 *errorcodeptr = ERR48;
5974 goto FAILED;
5975 }
5976 }
5977 }
5978
5979 /* In the real compile, create the entry in the table, maintaining
5980 alphabetical order. Duplicate names for different numbers are
5981 permitted only if PCRE_DUPNAMES is set. Duplicate names for the same
5982 number are always OK. (An existing number can be re-used if (?|
5983 appears in the pattern.) In either event, a duplicate name results in
5984 a duplicate entry in the table, even if the number is the same. This
5985 is because the number of names, and hence the table size, is computed
5986 in the pre-compile, and it affects various numbers and pointers which
5987 would all have to be modified, and the compiled code moved down, if
5988 duplicates with the same number were omitted from the table. This
5989 doesn't seem worth the hassle. However, *different* names for the
5990 same number are not permitted. */
5991
5992 else
5993 {
5994 BOOL dupname = FALSE;
5995 slot = cd->name_table;
5996
5997 for (i = 0; i < cd->names_found; i++)
5998 {
5999 int crc = memcmp(name, slot+IMM2_SIZE, IN_UCHARS(namelen));
6000 if (crc == 0)
6001 {
6002 if (slot[IMM2_SIZE+namelen] == 0)
6003 {
6004 if (GET2(slot, 0) != cd->bracount + 1 &&
6005 (options & PCRE_DUPNAMES) == 0)
6006 {
6007 *errorcodeptr = ERR43;
6008 goto FAILED;
6009 }
6010 else dupname = TRUE;
6011 }
6012 else crc = -1; /* Current name is a substring */
6013 }
6014
6015 /* Make space in the table and break the loop for an earlier
6016 name. For a duplicate or later name, carry on. We do this for
6017 duplicates so that in the simple case (when ?(| is not used) they
6018 are in order of their numbers. */
6019
6020 if (crc < 0)
6021 {
6022 memmove(slot + cd->name_entry_size, slot,
6023 IN_UCHARS((cd->names_found - i) * cd->name_entry_size));
6024 break;
6025 }
6026
6027 /* Continue the loop for a later or duplicate name */
6028
6029 slot += cd->name_entry_size;
6030 }
6031
6032 /* For non-duplicate names, check for a duplicate number before
6033 adding the new name. */
6034
6035 if (!dupname)
6036 {
6037 pcre_uchar *cslot = cd->name_table;
6038 for (i = 0; i < cd->names_found; i++)
6039 {
6040 if (cslot != slot)
6041 {
6042 if (GET2(cslot, 0) == cd->bracount + 1)
6043 {
6044 *errorcodeptr = ERR65;
6045 goto FAILED;
6046 }
6047 }
6048 else i--;
6049 cslot += cd->name_entry_size;
6050 }
6051 }
6052
6053 PUT2(slot, 0, cd->bracount + 1);
6054 memcpy(slot + IMM2_SIZE, name, IN_UCHARS(namelen));
6055 slot[IMM2_SIZE + namelen] = 0;
6056 }
6057 }
6058
6059 /* In both pre-compile and compile, count the number of names we've
6060 encountered. */
6061
6062 cd->names_found++;
6063 ptr++; /* Move past > or ' */
6064 goto NUMBERED_GROUP;
6065
6066
6067 /* ------------------------------------------------------------ */
6068 case CHAR_AMPERSAND: /* Perl recursion/subroutine syntax */
6069 terminator = CHAR_RIGHT_PARENTHESIS;
6070 is_recurse = TRUE;
6071 /* Fall through */
6072
6073 /* We come here from the Python syntax above that handles both
6074 references (?P=name) and recursion (?P>name), as well as falling
6075 through from the Perl recursion syntax (?&name). We also come here from
6076 the Perl \k<name> or \k'name' back reference syntax and the \k{name}
6077 .NET syntax, and the Oniguruma \g<...> and \g'...' subroutine syntax. */
6078
6079 NAMED_REF_OR_RECURSE:
6080 name = ++ptr;
6081 while (MAX_255(*ptr) && (cd->ctypes[*ptr] & ctype_word) != 0) ptr++;
6082 namelen = (int)(ptr - name);
6083
6084 /* In the pre-compile phase, do a syntax check. We used to just set
6085 a dummy reference number, because it was not used in the first pass.
6086 However, with the change of recursive back references to be atomic,
6087 we have to look for the number so that this state can be identified, as
6088 otherwise the incorrect length is computed. If it's not a backwards
6089 reference, the dummy number will do. */
6090
6091 if (lengthptr != NULL)
6092 {
6093 const pcre_uchar *temp;
6094
6095 if (namelen == 0)
6096 {
6097 *errorcodeptr = ERR62;
6098 goto FAILED;
6099 }
6100 if (*ptr != terminator)
6101 {
6102 *errorcodeptr = ERR42;
6103 goto FAILED;
6104 }
6105 if (namelen > MAX_NAME_SIZE)
6106 {
6107 *errorcodeptr = ERR48;
6108 goto FAILED;
6109 }
6110
6111 /* The name table does not exist in the first pass, so we cannot
6112 do a simple search as in the code below. Instead, we have to scan the
6113 pattern to find the number. It is important that we scan it only as
6114 far as we have got because the syntax of named subpatterns has not
6115 been checked for the rest of the pattern, and find_parens() assumes
6116 correct syntax. In any case, it's a waste of resources to scan
6117 further. We stop the scan at the current point by temporarily
6118 adjusting the value of cd->endpattern. */
6119
6120 temp = cd->end_pattern;
6121 cd->end_pattern = ptr;
6122 recno = find_parens(cd, name, namelen,
6123 (options & PCRE_EXTENDED) != 0, utf);
6124 cd->end_pattern = temp;
6125 if (recno < 0) recno = 0; /* Forward ref; set dummy number */
6126 }
6127
6128 /* In the real compile, seek the name in the table. We check the name
6129 first, and then check that we have reached the end of the name in the
6130 table. That way, if the name that is longer than any in the table,
6131 the comparison will fail without reading beyond the table entry. */
6132
6133 else
6134 {
6135 slot = cd->name_table;
6136 for (i = 0; i < cd->names_found; i++)
6137 {
6138 if (STRNCMP_UC_UC(name, slot+IMM2_SIZE, namelen) == 0 &&
6139 slot[IMM2_SIZE+namelen] == 0)
6140 break;
6141 slot += cd->name_entry_size;
6142 }
6143
6144 if (i < cd->names_found) /* Back reference */
6145 {
6146 recno = GET2(slot, 0);
6147 }
6148 else if ((recno = /* Forward back reference */
6149 find_parens(cd, name, namelen,
6150 (options & PCRE_EXTENDED) != 0, utf)) <= 0)
6151 {
6152 *errorcodeptr = ERR15;
6153 goto FAILED;
6154 }
6155 }
6156
6157 /* In both phases, we can now go to the code than handles numerical
6158 recursion or backreferences. */
6159
6160 if (is_recurse) goto HANDLE_RECURSION;
6161 else goto HANDLE_REFERENCE;
6162
6163
6164 /* ------------------------------------------------------------ */
6165 case CHAR_R: /* Recursion */
6166 ptr++; /* Same as (?0) */
6167 /* Fall through */
6168
6169
6170 /* ------------------------------------------------------------ */
6171 case CHAR_MINUS: case CHAR_PLUS: /* Recursion or subroutine */
6172 case CHAR_0: case CHAR_1: case CHAR_2: case CHAR_3: case CHAR_4:
6173 case CHAR_5: case CHAR_6: case CHAR_7: case CHAR_8: case CHAR_9:
6174 {
6175 const pcre_uchar *called;
6176 terminator = CHAR_RIGHT_PARENTHESIS;
6177
6178 /* Come here from the \g<...> and \g'...' code (Oniguruma
6179 compatibility). However, the syntax has been checked to ensure that
6180 the ... are a (signed) number, so that neither ERR63 nor ERR29 will
6181 be called on this path, nor with the jump to OTHER_CHAR_AFTER_QUERY
6182 ever be taken. */
6183
6184 HANDLE_NUMERICAL_RECURSION:
6185
6186 if ((refsign = *ptr) == CHAR_PLUS)
6187 {
6188 ptr++;
6189 if (!IS_DIGIT(*ptr))
6190 {
6191 *errorcodeptr = ERR63;
6192 goto FAILED;
6193 }
6194 }
6195 else if (refsign == CHAR_MINUS)
6196 {
6197 if (!IS_DIGIT(ptr[1]))
6198 goto OTHER_CHAR_AFTER_QUERY;
6199 ptr++;
6200 }
6201
6202 recno = 0;
6203 while(IS_DIGIT(*ptr))
6204 recno = recno * 10 + *ptr++ - CHAR_0;
6205
6206 if (*ptr != terminator)
6207 {
6208 *errorcodeptr = ERR29;
6209 goto FAILED;
6210 }
6211
6212 if (refsign == CHAR_MINUS)
6213 {
6214 if (recno == 0)
6215 {
6216 *errorcodeptr = ERR58;
6217 goto FAILED;
6218 }
6219 recno = cd->bracount - recno + 1;
6220 if (recno <= 0)
6221 {
6222 *errorcodeptr = ERR15;
6223 goto FAILED;
6224 }
6225 }
6226 else if (refsign == CHAR_PLUS)
6227 {
6228 if (recno == 0)
6229 {
6230 *errorcodeptr = ERR58;
6231 goto FAILED;
6232 }
6233 recno += cd->bracount;
6234 }
6235
6236 /* Come here from code above that handles a named recursion */
6237
6238 HANDLE_RECURSION:
6239
6240 previous = code;
6241 called = cd->start_code;
6242
6243 /* When we are actually compiling, find the bracket that is being
6244 referenced. Temporarily end the regex in case it doesn't exist before
6245 this point. If we end up with a forward reference, first check that
6246 the bracket does occur later so we can give the error (and position)
6247 now. Then remember this forward reference in the workspace so it can
6248 be filled in at the end. */
6249
6250 if (lengthptr == NULL)
6251 {
6252 *code = OP_END;
6253 if (recno != 0)
6254 called = PRIV(find_bracket)(cd->start_code, utf, recno);
6255
6256 /* Forward reference */
6257
6258 if (called == NULL)
6259 {
6260 if (find_parens(cd, NULL, recno,
6261 (options & PCRE_EXTENDED) != 0, utf) < 0)
6262 {
6263 *errorcodeptr = ERR15;
6264 goto FAILED;
6265 }
6266
6267 /* Fudge the value of "called" so that when it is inserted as an
6268 offset below, what it actually inserted is the reference number
6269 of the group. Then remember the forward reference. */
6270
6271 called = cd->start_code + recno;
6272 if (cd->hwm >= cd->start_workspace + cd->workspace_size -
6273 WORK_SIZE_SAFETY_MARGIN)
6274 {
6275 *errorcodeptr = expand_workspace(cd);
6276 if (*errorcodeptr != 0) goto FAILED;
6277 }
6278 PUTINC(cd->hwm, 0, (int)(code + 1 - cd->start_code));
6279 }
6280
6281 /* If not a forward reference, and the subpattern is still open,
6282 this is a recursive call. We check to see if this is a left
6283 recursion that could loop for ever, and diagnose that case. We
6284 must not, however, do this check if we are in a conditional
6285 subpattern because the condition might be testing for recursion in
6286 a pattern such as /(?(R)a+|(?R)b)/, which is perfectly valid.
6287 Forever loops are also detected at runtime, so those that occur in
6288 conditional subpatterns will be picked up then. */
6289
6290 else if (GET(called, 1) == 0 && cond_depth <= 0 &&
6291 could_be_empty(called, code, bcptr, utf, cd))
6292 {
6293 *errorcodeptr = ERR40;
6294 goto FAILED;
6295 }
6296 }
6297
6298 /* Insert the recursion/subroutine item. It does not have a set first
6299 character (relevant if it is repeated, because it will then be
6300 wrapped with ONCE brackets). */
6301
6302 *code = OP_RECURSE;
6303 PUT(code, 1, (int)(called - cd->start_code));
6304 code += 1 + LINK_SIZE;
6305 groupsetfirstchar = FALSE;
6306 }
6307
6308 /* Can't determine a first byte now */
6309
6310 if (firstchar == REQ_UNSET) firstchar = REQ_NONE;
6311 continue;
6312
6313
6314 /* ------------------------------------------------------------ */
6315 default: /* Other characters: check option setting */
6316 OTHER_CHAR_AFTER_QUERY:
6317 set = unset = 0;
6318 optset = &set;
6319
6320 while (*ptr != CHAR_RIGHT_PARENTHESIS && *ptr != CHAR_COLON)
6321 {
6322 switch (*ptr++)
6323 {
6324 case CHAR_MINUS: optset = &unset; break;
6325
6326 case CHAR_J: /* Record that it changed in the external options */
6327 *optset |= PCRE_DUPNAMES;
6328 cd->external_flags |= PCRE_JCHANGED;
6329 break;
6330
6331 case CHAR_i: *optset |= PCRE_CASELESS; break;
6332 case CHAR_m: *optset |= PCRE_MULTILINE; break;
6333 case CHAR_s: *optset |= PCRE_DOTALL; break;
6334 case CHAR_x: *optset |= PCRE_EXTENDED; break;
6335 case CHAR_U: *optset |= PCRE_UNGREEDY; break;
6336 case CHAR_X: *optset |= PCRE_EXTRA; break;
6337
6338 default: *errorcodeptr = ERR12;
6339 ptr--; /* Correct the offset */
6340 goto FAILED;
6341 }
6342 }
6343
6344 /* Set up the changed option bits, but don't change anything yet. */
6345
6346 newoptions = (options | set) & (~unset);
6347
6348 /* If the options ended with ')' this is not the start of a nested
6349 group with option changes, so the options change at this level. If this
6350 item is right at the start of the pattern, the options can be
6351 abstracted and made external in the pre-compile phase, and ignored in
6352 the compile phase. This can be helpful when matching -- for instance in
6353 caseless checking of required bytes.
6354
6355 If the code pointer is not (cd->start_code + 1 + LINK_SIZE), we are
6356 definitely *not* at the start of the pattern because something has been
6357 compiled. In the pre-compile phase, however, the code pointer can have
6358 that value after the start, because it gets reset as code is discarded
6359 during the pre-compile. However, this can happen only at top level - if
6360 we are within parentheses, the starting BRA will still be present. At
6361 any parenthesis level, the length value can be used to test if anything
6362 has been compiled at that level. Thus, a test for both these conditions
6363 is necessary to ensure we correctly detect the start of the pattern in
6364 both phases.
6365
6366 If we are not at the pattern start, reset the greedy defaults and the
6367 case value for firstchar and reqchar. */
6368
6369 if (*ptr == CHAR_RIGHT_PARENTHESIS)
6370 {
6371 if (code == cd->start_code + 1 + LINK_SIZE &&
6372 (lengthptr == NULL || *lengthptr == 2 + 2*LINK_SIZE))
6373 {
6374 cd->external_options = newoptions;
6375 }
6376 else
6377 {
6378 greedy_default = ((newoptions & PCRE_UNGREEDY) != 0);
6379 greedy_non_default = greedy_default ^ 1;
6380 req_caseopt = ((newoptions & PCRE_CASELESS) != 0)? REQ_CASELESS:0;
6381 }
6382
6383 /* Change options at this level, and pass them back for use
6384 in subsequent branches. */
6385
6386 *optionsptr = options = newoptions;
6387 previous = NULL; /* This item can't be repeated */
6388 continue; /* It is complete */
6389 }
6390
6391 /* If the options ended with ':' we are heading into a nested group
6392 with possible change of options. Such groups are non-capturing and are
6393 not assertions of any kind. All we need to do is skip over the ':';
6394 the newoptions value is handled below. */
6395
6396 bravalue = OP_BRA;
6397 ptr++;
6398 } /* End of switch for character following (? */
6399 } /* End of (? handling */
6400
6401 /* Opening parenthesis not followed by '*' or '?'. If PCRE_NO_AUTO_CAPTURE
6402 is set, all unadorned brackets become non-capturing and behave like (?:...)
6403 brackets. */
6404
6405 else if ((options & PCRE_NO_AUTO_CAPTURE) != 0)
6406 {
6407 bravalue = OP_BRA;
6408 }
6409
6410 /* Else we have a capturing group. */
6411
6412 else
6413 {
6414 NUMBERED_GROUP:
6415 cd->bracount += 1;
6416 PUT2(code, 1+LINK_SIZE, cd->bracount);
6417 skipbytes = IMM2_SIZE;
6418 }
6419
6420 /* Process nested bracketed regex. Assertions used not to be repeatable,
6421 but this was changed for Perl compatibility, so all kinds can now be
6422 repeated. We copy code into a non-register variable (tempcode) in order to
6423 be able to pass its address because some compilers complain otherwise. */
6424
6425 previous = code; /* For handling repetition */
6426 *code = bravalue;
6427 tempcode = code;
6428 tempreqvary = cd->req_varyopt; /* Save value before bracket */
6429 tempbracount = cd->bracount; /* Save value before bracket */
6430 length_prevgroup = 0; /* Initialize for pre-compile phase */
6431
6432 if (!compile_regex(
6433 newoptions, /* The complete new option state */
6434 &tempcode, /* Where to put code (updated) */
6435 &ptr, /* Input pointer (updated) */
6436 errorcodeptr, /* Where to put an error message */
6437 (bravalue == OP_ASSERTBACK ||
6438 bravalue == OP_ASSERTBACK_NOT), /* TRUE if back assert */
6439 reset_bracount, /* True if (?| group */
6440 skipbytes, /* Skip over bracket number */
6441 cond_depth +
6442 ((bravalue == OP_COND)?1:0), /* Depth of condition subpatterns */
6443 &subfirstchar, /* For possible first char */
6444 &subreqchar, /* For possible last char */
6445 bcptr, /* Current branch chain */
6446 cd, /* Tables block */
6447 (lengthptr == NULL)? NULL : /* Actual compile phase */
6448 &length_prevgroup /* Pre-compile phase */
6449 ))
6450 goto FAILED;
6451
6452 /* If this was an atomic group and there are no capturing groups within it,
6453 generate OP_ONCE_NC instead of OP_ONCE. */
6454
6455 if (bravalue == OP_ONCE && cd->bracount <= tempbracount)
6456 *code = OP_ONCE_NC;
6457
6458 if (bravalue >= OP_ASSERT && bravalue <= OP_ASSERTBACK_NOT)
6459 cd->assert_depth -= 1;
6460
6461 /* At the end of compiling, code is still pointing to the start of the
6462 group, while tempcode has been updated to point past the end of the group.
6463 The pattern pointer (ptr) is on the bracket.
6464
6465 If this is a conditional bracket, check that there are no more than
6466 two branches in the group, or just one if it's a DEFINE group. We do this
6467 in the real compile phase, not in the pre-pass, where the whole group may
6468 not be available. */
6469
6470 if (bravalue == OP_COND && lengthptr == NULL)
6471 {
6472 pcre_uchar *tc = code;
6473 int condcount = 0;
6474
6475 do {
6476 condcount++;
6477 tc += GET(tc,1);
6478 }
6479 while (*tc != OP_KET);
6480
6481 /* A DEFINE group is never obeyed inline (the "condition" is always
6482 false). It must have only one branch. */
6483
6484 if (code[LINK_SIZE+1] == OP_DEF)
6485 {
6486 if (condcount > 1)
6487 {
6488 *errorcodeptr = ERR54;
6489 goto FAILED;
6490 }
6491 bravalue = OP_DEF; /* Just a flag to suppress char handling below */
6492 }
6493
6494 /* A "normal" conditional group. If there is just one branch, we must not
6495 make use of its firstchar or reqchar, because this is equivalent to an
6496 empty second branch. */
6497
6498 else
6499 {
6500 if (condcount > 2)
6501 {
6502 *errorcodeptr = ERR27;
6503 goto FAILED;
6504 }
6505 if (condcount == 1) subfirstchar = subreqchar = REQ_NONE;
6506 }
6507 }
6508
6509 /* Error if hit end of pattern */
6510
6511 if (*ptr != CHAR_RIGHT_PARENTHESIS)
6512 {
6513 *errorcodeptr = ERR14;
6514 goto FAILED;
6515 }
6516
6517 /* In the pre-compile phase, update the length by the length of the group,
6518 less the brackets at either end. Then reduce the compiled code to just a
6519 set of non-capturing brackets so that it doesn't use much memory if it is
6520 duplicated by a quantifier.*/
6521
6522 if (lengthptr != NULL)
6523 {
6524 if (OFLOW_MAX - *lengthptr < length_prevgroup - 2 - 2*LINK_SIZE)
6525 {
6526 *errorcodeptr = ERR20;
6527 goto FAILED;
6528 }
6529 *lengthptr += length_prevgroup - 2 - 2*LINK_SIZE;
6530 code++; /* This already contains bravalue */
6531 PUTINC(code, 0, 1 + LINK_SIZE);
6532 *code++ = OP_KET;
6533 PUTINC(code, 0, 1 + LINK_SIZE);
6534 break; /* No need to waste time with special character handling */
6535 }
6536
6537 /* Otherwise update the main code pointer to the end of the group. */
6538
6539 code = tempcode;
6540
6541 /* For a DEFINE group, required and first character settings are not
6542 relevant. */
6543
6544 if (bravalue == OP_DEF) break;
6545
6546 /* Handle updating of the required and first characters for other types of
6547 group. Update for normal brackets of all kinds, and conditions with two
6548 branches (see code above). If the bracket is followed by a quantifier with
6549 zero repeat, we have to back off. Hence the definition of zeroreqchar and
6550 zerofirstchar outside the main loop so that they can be accessed for the
6551 back off. */
6552
6553 zeroreqchar = reqchar;
6554 zerofirstchar = firstchar;
6555 groupsetfirstchar = FALSE;
6556
6557 if (bravalue >= OP_ONCE)
6558 {
6559 /* If we have not yet set a firstchar in this branch, take it from the
6560 subpattern, remembering that it was set here so that a repeat of more
6561 than one can replicate it as reqchar if necessary. If the subpattern has
6562 no firstchar, set "none" for the whole branch. In both cases, a zero
6563 repeat forces firstchar to "none". */
6564
6565 if (firstchar == REQ_UNSET)
6566 {
6567 if (subfirstchar >= 0)
6568 {
6569 firstchar = subfirstchar;
6570 groupsetfirstchar = TRUE;
6571 }
6572 else firstchar = REQ_NONE;
6573 zerofirstchar = REQ_NONE;
6574 }
6575
6576 /* If firstchar was previously set, convert the subpattern's firstchar
6577 into reqchar if there wasn't one, using the vary flag that was in
6578 existence beforehand. */
6579
6580 else if (subfirstchar >= 0 && subreqchar < 0)
6581 subreqchar = subfirstchar | tempreqvary;
6582
6583 /* If the subpattern set a required byte (or set a first byte that isn't
6584 really the first byte - see above), set it. */
6585
6586 if (subreqchar >= 0) reqchar = subreqchar;
6587 }
6588
6589 /* For a forward assertion, we take the reqchar, if set. This can be
6590 helpful if the pattern that follows the assertion doesn't set a different
6591 char. For example, it's useful for /(?=abcde).+/. We can't set firstchar
6592 for an assertion, however because it leads to incorrect effect for patterns
6593 such as /(?=a)a.+/ when the "real" "a" would then become a reqchar instead
6594 of a firstchar. This is overcome by a scan at the end if there's no
6595 firstchar, looking for an asserted first char. */
6596
6597 else if (bravalue == OP_ASSERT && subreqchar >= 0) reqchar = subreqchar;
6598 break; /* End of processing '(' */
6599
6600
6601 /* ===================================================================*/
6602 /* Handle metasequences introduced by \. For ones like \d, the ESC_ values
6603 are arranged to be the negation of the corresponding OP_values in the
6604 default case when PCRE_UCP is not set. For the back references, the values
6605 are ESC_REF plus the reference number. Only back references and those types
6606 that consume a character may be repeated. We can test for values between
6607 ESC_b and ESC_Z for the latter; this may have to change if any new ones are
6608 ever created. */
6609
6610 case CHAR_BACKSLASH:
6611 tempptr = ptr;
6612 c = check_escape(&ptr, errorcodeptr, cd->bracount, options, FALSE);
6613 if (*errorcodeptr != 0) goto FAILED;
6614
6615 if (c < 0)
6616 {
6617 if (-c == ESC_Q) /* Handle start of quoted string */
6618 {
6619 if (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
6620 ptr += 2; /* avoid empty string */
6621 else inescq = TRUE;
6622 continue;
6623 }
6624
6625 if (-c == ESC_E) continue; /* Perl ignores an orphan \E */
6626
6627 /* For metasequences that actually match a character, we disable the
6628 setting of a first character if it hasn't already been set. */
6629
6630 if (firstchar == REQ_UNSET && -c > ESC_b && -c < ESC_Z)
6631 firstchar = REQ_NONE;
6632
6633 /* Set values to reset to if this is followed by a zero repeat. */
6634
6635 zerofirstchar = firstchar;
6636 zeroreqchar = reqchar;
6637
6638 /* \g<name> or \g'name' is a subroutine call by name and \g<n> or \g'n'
6639 is a subroutine call by number (Oniguruma syntax). In fact, the value
6640 -ESC_g is returned only for these cases. So we don't need to check for <
6641 or ' if the value is -ESC_g. For the Perl syntax \g{n} the value is