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

Contents of /code/trunk/pcre_compile.c

Parent Directory Parent Directory | Revision Log Revision Log


Revision 1064 - (show annotations)
Tue Oct 16 15:54:12 2012 UTC (7 years, 1 month ago) by chpe
File MIME type: text/plain
File size: 272170 byte(s)
Error occurred while calculating annotation data.
pcre32: compile: Use uint32 to store characters in compile_branch

Do this to preserve any 32-bit data character in 32-bit non-UTF-32 mode.
1 /*************************************************
2 * Perl-Compatible Regular Expressions *
3 *************************************************/
4
5 /* PCRE is a library of functions to support regular expressions whose syntax
6 and semantics are as close as possible to those of the Perl 5 language.
7
8 Written by Philip Hazel
9 Copyright (c) 1997-2012 University of Cambridge
10
11 -----------------------------------------------------------------------------
12 Redistribution and use in source and binary forms, with or without
13 modification, are permitted provided that the following conditions are met:
14
15 * Redistributions of source code must retain the above copyright notice,
16 this list of conditions and the following disclaimer.
17
18 * Redistributions in binary form must reproduce the above copyright
19 notice, this list of conditions and the following disclaimer in the
20 documentation and/or other materials provided with the distribution.
21
22 * Neither the name of the University of Cambridge nor the names of its
23 contributors may be used to endorse or promote products derived from
24 this software without specific prior written permission.
25
26 THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
27 AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28 IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29 ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE
30 LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
31 CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
32 SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
33 INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
34 CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
35 ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
36 POSSIBILITY OF SUCH DAMAGE.
37 -----------------------------------------------------------------------------
38 */
39
40
41 /* This module contains the external function pcre_compile(), along with
42 supporting internal functions that are not used by other modules. */
43
44
45 #ifdef HAVE_CONFIG_H
46 #include "config.h"
47 #endif
48
49 #define NLBLOCK cd /* Block containing newline information */
50 #define PSSTART start_pattern /* Field containing processed string start */
51 #define PSEND end_pattern /* Field containing processed string end */
52
53 #include "pcre_internal.h"
54
55
56 /* When PCRE_DEBUG is defined, we need the pcre(16|32)_printint() function, which
57 is also used by pcretest. PCRE_DEBUG is not defined when building a production
58 library. We do not need to select pcre16_printint.c specially, because the
59 COMPILE_PCREx macro will already be appropriately set. */
60
61 #ifdef PCRE_DEBUG
62 /* pcre_printint.c should not include any headers */
63 #define PCRE_INCLUDED
64 #include "pcre_printint.c"
65 #undef PCRE_INCLUDED
66 #endif
67
68
69 /* Macro for setting individual bits in class bitmaps. */
70
71 #define SETBIT(a,b) a[(b)/8] |= (1 << ((b)&7))
72
73 /* Maximum length value to check against when making sure that the integer that
74 holds the compiled pattern length does not overflow. We make it a bit less than
75 INT_MAX to allow for adding in group terminating bytes, so that we don't have
76 to check them every time. */
77
78 #define OFLOW_MAX (INT_MAX - 20)
79
80 /* Definitions to allow mutual recursion */
81
82 static int
83 add_list_to_class(pcre_uint8 *, pcre_uchar **, int, compile_data *,
84 const pcre_uint32 *, unsigned int);
85
86 static BOOL
87 compile_regex(int, pcre_uchar **, const pcre_uchar **, int *, BOOL, BOOL,
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 #define REQ_MASK (REQ_CASELESS | REQ_VARY)
127
128 /* Repeated character flags. */
129
130 #define UTF_LENGTH 0x10000000l /* The char contains its length. */
131
132 /* Table for handling escaped characters in the range '0'-'z'. Positive returns
133 are simple data values; negative values are for special things like \d and so
134 on. Zero means further processing is needed (for things like \x), or the escape
135 is invalid. */
136
137 #ifndef EBCDIC
138
139 /* This is the "normal" table for ASCII systems or for EBCDIC systems running
140 in UTF-8 mode. */
141
142 static const short int escapes[] = {
143 0, 0,
144 0, 0,
145 0, 0,
146 0, 0,
147 0, 0,
148 CHAR_COLON, CHAR_SEMICOLON,
149 CHAR_LESS_THAN_SIGN, CHAR_EQUALS_SIGN,
150 CHAR_GREATER_THAN_SIGN, CHAR_QUESTION_MARK,
151 CHAR_COMMERCIAL_AT, -ESC_A,
152 -ESC_B, -ESC_C,
153 -ESC_D, -ESC_E,
154 0, -ESC_G,
155 -ESC_H, 0,
156 0, -ESC_K,
157 0, 0,
158 -ESC_N, 0,
159 -ESC_P, -ESC_Q,
160 -ESC_R, -ESC_S,
161 0, 0,
162 -ESC_V, -ESC_W,
163 -ESC_X, 0,
164 -ESC_Z, CHAR_LEFT_SQUARE_BRACKET,
165 CHAR_BACKSLASH, CHAR_RIGHT_SQUARE_BRACKET,
166 CHAR_CIRCUMFLEX_ACCENT, CHAR_UNDERSCORE,
167 CHAR_GRAVE_ACCENT, 7,
168 -ESC_b, 0,
169 -ESC_d, ESC_e,
170 ESC_f, 0,
171 -ESC_h, 0,
172 0, -ESC_k,
173 0, 0,
174 ESC_n, 0,
175 -ESC_p, 0,
176 ESC_r, -ESC_s,
177 ESC_tee, 0,
178 -ESC_v, -ESC_w,
179 0, 0,
180 -ESC_z
181 };
182
183 #else
184
185 /* This is the "abnormal" table for EBCDIC systems without UTF-8 support. */
186
187 static const short int escapes[] = {
188 /* 48 */ 0, 0, 0, '.', '<', '(', '+', '|',
189 /* 50 */ '&', 0, 0, 0, 0, 0, 0, 0,
190 /* 58 */ 0, 0, '!', '$', '*', ')', ';', '~',
191 /* 60 */ '-', '/', 0, 0, 0, 0, 0, 0,
192 /* 68 */ 0, 0, '|', ',', '%', '_', '>', '?',
193 /* 70 */ 0, 0, 0, 0, 0, 0, 0, 0,
194 /* 78 */ 0, '`', ':', '#', '@', '\'', '=', '"',
195 /* 80 */ 0, 7, -ESC_b, 0, -ESC_d, ESC_e, ESC_f, 0,
196 /* 88 */-ESC_h, 0, 0, '{', 0, 0, 0, 0,
197 /* 90 */ 0, 0, -ESC_k, 'l', 0, ESC_n, 0, -ESC_p,
198 /* 98 */ 0, ESC_r, 0, '}', 0, 0, 0, 0,
199 /* A0 */ 0, '~', -ESC_s, ESC_tee, 0,-ESC_v, -ESC_w, 0,
200 /* A8 */ 0,-ESC_z, 0, 0, 0, '[', 0, 0,
201 /* B0 */ 0, 0, 0, 0, 0, 0, 0, 0,
202 /* B8 */ 0, 0, 0, 0, 0, ']', '=', '-',
203 /* C0 */ '{',-ESC_A, -ESC_B, -ESC_C, -ESC_D,-ESC_E, 0, -ESC_G,
204 /* C8 */-ESC_H, 0, 0, 0, 0, 0, 0, 0,
205 /* D0 */ '}', 0, -ESC_K, 0, 0,-ESC_N, 0, -ESC_P,
206 /* D8 */-ESC_Q,-ESC_R, 0, 0, 0, 0, 0, 0,
207 /* E0 */ '\\', 0, -ESC_S, 0, 0,-ESC_V, -ESC_W, -ESC_X,
208 /* E8 */ 0,-ESC_Z, 0, 0, 0, 0, 0, 0,
209 /* F0 */ 0, 0, 0, 0, 0, 0, 0, 0,
210 /* F8 */ 0, 0, 0, 0, 0, 0, 0, 0
211 };
212 #endif
213
214
215 /* Table of special "verbs" like (*PRUNE). This is a short table, so it is
216 searched linearly. Put all the names into a single string, in order to reduce
217 the number of relocations when a shared library is dynamically linked. The
218 string is built from string macros so that it works in UTF-8 mode on EBCDIC
219 platforms. */
220
221 typedef struct verbitem {
222 int len; /* Length of verb name */
223 int op; /* Op when no arg, or -1 if arg mandatory */
224 int op_arg; /* Op when arg present, or -1 if not allowed */
225 } verbitem;
226
227 static const char verbnames[] =
228 "\0" /* Empty name is a shorthand for MARK */
229 STRING_MARK0
230 STRING_ACCEPT0
231 STRING_COMMIT0
232 STRING_F0
233 STRING_FAIL0
234 STRING_PRUNE0
235 STRING_SKIP0
236 STRING_THEN;
237
238 static const verbitem verbs[] = {
239 { 0, -1, OP_MARK },
240 { 4, -1, OP_MARK },
241 { 6, OP_ACCEPT, -1 },
242 { 6, OP_COMMIT, -1 },
243 { 1, OP_FAIL, -1 },
244 { 4, OP_FAIL, -1 },
245 { 5, OP_PRUNE, OP_PRUNE_ARG },
246 { 4, OP_SKIP, OP_SKIP_ARG },
247 { 4, OP_THEN, OP_THEN_ARG }
248 };
249
250 static const int verbcount = sizeof(verbs)/sizeof(verbitem);
251
252
253 /* Tables of names of POSIX character classes and their lengths. The names are
254 now all in a single string, to reduce the number of relocations when a shared
255 library is dynamically loaded. The list of lengths is terminated by a zero
256 length entry. The first three must be alpha, lower, upper, as this is assumed
257 for handling case independence. */
258
259 static const char posix_names[] =
260 STRING_alpha0 STRING_lower0 STRING_upper0 STRING_alnum0
261 STRING_ascii0 STRING_blank0 STRING_cntrl0 STRING_digit0
262 STRING_graph0 STRING_print0 STRING_punct0 STRING_space0
263 STRING_word0 STRING_xdigit;
264
265 static const pcre_uint8 posix_name_lengths[] = {
266 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 4, 6, 0 };
267
268 /* Table of class bit maps for each POSIX class. Each class is formed from a
269 base map, with an optional addition or removal of another map. Then, for some
270 classes, there is some additional tweaking: for [:blank:] the vertical space
271 characters are removed, and for [:alpha:] and [:alnum:] the underscore
272 character is removed. The triples in the table consist of the base map offset,
273 second map offset or -1 if no second map, and a non-negative value for map
274 addition or a negative value for map subtraction (if there are two maps). The
275 absolute value of the third field has these meanings: 0 => no tweaking, 1 =>
276 remove vertical space characters, 2 => remove underscore. */
277
278 static const int posix_class_maps[] = {
279 cbit_word, cbit_digit, -2, /* alpha */
280 cbit_lower, -1, 0, /* lower */
281 cbit_upper, -1, 0, /* upper */
282 cbit_word, -1, 2, /* alnum - word without underscore */
283 cbit_print, cbit_cntrl, 0, /* ascii */
284 cbit_space, -1, 1, /* blank - a GNU extension */
285 cbit_cntrl, -1, 0, /* cntrl */
286 cbit_digit, -1, 0, /* digit */
287 cbit_graph, -1, 0, /* graph */
288 cbit_print, -1, 0, /* print */
289 cbit_punct, -1, 0, /* punct */
290 cbit_space, -1, 0, /* space */
291 cbit_word, -1, 0, /* word - a Perl extension */
292 cbit_xdigit,-1, 0 /* xdigit */
293 };
294
295 /* Table of substitutes for \d etc when PCRE_UCP is set. The POSIX class
296 substitutes must be in the order of the names, defined above, and there are
297 both positive and negative cases. NULL means no substitute. */
298
299 #ifdef SUPPORT_UCP
300 static const pcre_uchar string_PNd[] = {
301 CHAR_BACKSLASH, CHAR_P, CHAR_LEFT_CURLY_BRACKET,
302 CHAR_N, CHAR_d, CHAR_RIGHT_CURLY_BRACKET, '\0' };
303 static const pcre_uchar string_pNd[] = {
304 CHAR_BACKSLASH, CHAR_p, CHAR_LEFT_CURLY_BRACKET,
305 CHAR_N, CHAR_d, CHAR_RIGHT_CURLY_BRACKET, '\0' };
306 static const pcre_uchar string_PXsp[] = {
307 CHAR_BACKSLASH, CHAR_P, CHAR_LEFT_CURLY_BRACKET,
308 CHAR_X, CHAR_s, CHAR_p, CHAR_RIGHT_CURLY_BRACKET, '\0' };
309 static const pcre_uchar string_pXsp[] = {
310 CHAR_BACKSLASH, CHAR_p, CHAR_LEFT_CURLY_BRACKET,
311 CHAR_X, CHAR_s, CHAR_p, CHAR_RIGHT_CURLY_BRACKET, '\0' };
312 static const pcre_uchar string_PXwd[] = {
313 CHAR_BACKSLASH, CHAR_P, CHAR_LEFT_CURLY_BRACKET,
314 CHAR_X, CHAR_w, CHAR_d, CHAR_RIGHT_CURLY_BRACKET, '\0' };
315 static const pcre_uchar string_pXwd[] = {
316 CHAR_BACKSLASH, CHAR_p, CHAR_LEFT_CURLY_BRACKET,
317 CHAR_X, CHAR_w, CHAR_d, CHAR_RIGHT_CURLY_BRACKET, '\0' };
318
319 static const pcre_uchar *substitutes[] = {
320 string_PNd, /* \D */
321 string_pNd, /* \d */
322 string_PXsp, /* \S */ /* NOTE: Xsp is Perl space */
323 string_pXsp, /* \s */
324 string_PXwd, /* \W */
325 string_pXwd /* \w */
326 };
327
328 static const pcre_uchar string_pL[] = {
329 CHAR_BACKSLASH, CHAR_p, CHAR_LEFT_CURLY_BRACKET,
330 CHAR_L, CHAR_RIGHT_CURLY_BRACKET, '\0' };
331 static const pcre_uchar string_pLl[] = {
332 CHAR_BACKSLASH, CHAR_p, CHAR_LEFT_CURLY_BRACKET,
333 CHAR_L, CHAR_l, CHAR_RIGHT_CURLY_BRACKET, '\0' };
334 static const pcre_uchar string_pLu[] = {
335 CHAR_BACKSLASH, CHAR_p, CHAR_LEFT_CURLY_BRACKET,
336 CHAR_L, CHAR_u, CHAR_RIGHT_CURLY_BRACKET, '\0' };
337 static const pcre_uchar string_pXan[] = {
338 CHAR_BACKSLASH, CHAR_p, CHAR_LEFT_CURLY_BRACKET,
339 CHAR_X, CHAR_a, CHAR_n, CHAR_RIGHT_CURLY_BRACKET, '\0' };
340 static const pcre_uchar string_h[] = {
341 CHAR_BACKSLASH, CHAR_h, '\0' };
342 static const pcre_uchar string_pXps[] = {
343 CHAR_BACKSLASH, CHAR_p, CHAR_LEFT_CURLY_BRACKET,
344 CHAR_X, CHAR_p, CHAR_s, CHAR_RIGHT_CURLY_BRACKET, '\0' };
345 static const pcre_uchar string_PL[] = {
346 CHAR_BACKSLASH, CHAR_P, CHAR_LEFT_CURLY_BRACKET,
347 CHAR_L, CHAR_RIGHT_CURLY_BRACKET, '\0' };
348 static const pcre_uchar string_PLl[] = {
349 CHAR_BACKSLASH, CHAR_P, CHAR_LEFT_CURLY_BRACKET,
350 CHAR_L, CHAR_l, CHAR_RIGHT_CURLY_BRACKET, '\0' };
351 static const pcre_uchar string_PLu[] = {
352 CHAR_BACKSLASH, CHAR_P, CHAR_LEFT_CURLY_BRACKET,
353 CHAR_L, CHAR_u, CHAR_RIGHT_CURLY_BRACKET, '\0' };
354 static const pcre_uchar string_PXan[] = {
355 CHAR_BACKSLASH, CHAR_P, CHAR_LEFT_CURLY_BRACKET,
356 CHAR_X, CHAR_a, CHAR_n, CHAR_RIGHT_CURLY_BRACKET, '\0' };
357 static const pcre_uchar string_H[] = {
358 CHAR_BACKSLASH, CHAR_H, '\0' };
359 static const pcre_uchar string_PXps[] = {
360 CHAR_BACKSLASH, CHAR_P, CHAR_LEFT_CURLY_BRACKET,
361 CHAR_X, CHAR_p, CHAR_s, CHAR_RIGHT_CURLY_BRACKET, '\0' };
362
363 static const pcre_uchar *posix_substitutes[] = {
364 string_pL, /* alpha */
365 string_pLl, /* lower */
366 string_pLu, /* upper */
367 string_pXan, /* alnum */
368 NULL, /* ascii */
369 string_h, /* blank */
370 NULL, /* cntrl */
371 string_pNd, /* digit */
372 NULL, /* graph */
373 NULL, /* print */
374 NULL, /* punct */
375 string_pXps, /* space */ /* NOTE: Xps is POSIX space */
376 string_pXwd, /* word */
377 NULL, /* xdigit */
378 /* Negated cases */
379 string_PL, /* ^alpha */
380 string_PLl, /* ^lower */
381 string_PLu, /* ^upper */
382 string_PXan, /* ^alnum */
383 NULL, /* ^ascii */
384 string_H, /* ^blank */
385 NULL, /* ^cntrl */
386 string_PNd, /* ^digit */
387 NULL, /* ^graph */
388 NULL, /* ^print */
389 NULL, /* ^punct */
390 string_PXps, /* ^space */ /* NOTE: Xps is POSIX space */
391 string_PXwd, /* ^word */
392 NULL /* ^xdigit */
393 };
394 #define POSIX_SUBSIZE (sizeof(posix_substitutes) / sizeof(pcre_uchar *))
395 #endif
396
397 #define STRING(a) # a
398 #define XSTRING(s) STRING(s)
399
400 /* The texts of compile-time error messages. These are "char *" because they
401 are passed to the outside world. Do not ever re-use any error number, because
402 they are documented. Always add a new error instead. Messages marked DEAD below
403 are no longer used. This used to be a table of strings, but in order to reduce
404 the number of relocations needed when a shared library is loaded dynamically,
405 it is now one long string. We cannot use a table of offsets, because the
406 lengths of inserts such as XSTRING(MAX_NAME_SIZE) are not known. Instead, we
407 simply count through to the one we want - this isn't a performance issue
408 because these strings are used only when there is a compilation error.
409
410 Each substring ends with \0 to insert a null character. This includes the final
411 substring, so that the whole string ends with \0\0, which can be detected when
412 counting through. */
413
414 static const char error_texts[] =
415 "no error\0"
416 "\\ at end of pattern\0"
417 "\\c at end of pattern\0"
418 "unrecognized character follows \\\0"
419 "numbers out of order in {} quantifier\0"
420 /* 5 */
421 "number too big in {} quantifier\0"
422 "missing terminating ] for character class\0"
423 "invalid escape sequence in character class\0"
424 "range out of order in character class\0"
425 "nothing to repeat\0"
426 /* 10 */
427 "operand of unlimited repeat could match the empty string\0" /** DEAD **/
428 "internal error: unexpected repeat\0"
429 "unrecognized character after (? or (?-\0"
430 "POSIX named classes are supported only within a class\0"
431 "missing )\0"
432 /* 15 */
433 "reference to non-existent subpattern\0"
434 "erroffset passed as NULL\0"
435 "unknown option bit(s) set\0"
436 "missing ) after comment\0"
437 "parentheses nested too deeply\0" /** DEAD **/
438 /* 20 */
439 "regular expression is too large\0"
440 "failed to get memory\0"
441 "unmatched parentheses\0"
442 "internal error: code overflow\0"
443 "unrecognized character after (?<\0"
444 /* 25 */
445 "lookbehind assertion is not fixed length\0"
446 "malformed number or name after (?(\0"
447 "conditional group contains more than two branches\0"
448 "assertion expected after (?(\0"
449 "(?R or (?[+-]digits must be followed by )\0"
450 /* 30 */
451 "unknown POSIX class name\0"
452 "POSIX collating elements are not supported\0"
453 "this version of PCRE is compiled without UTF support\0"
454 "spare error\0" /** DEAD **/
455 "character value in \\x{...} sequence is too large\0"
456 /* 35 */
457 "invalid condition (?(0)\0"
458 "\\C not allowed in lookbehind assertion\0"
459 "PCRE does not support \\L, \\l, \\N{name}, \\U, or \\u\0"
460 "number after (?C is > 255\0"
461 "closing ) for (?C expected\0"
462 /* 40 */
463 "recursive call could loop indefinitely\0"
464 "unrecognized character after (?P\0"
465 "syntax error in subpattern name (missing terminator)\0"
466 "two named subpatterns have the same name\0"
467 "invalid UTF-8 string\0"
468 /* 45 */
469 "support for \\P, \\p, and \\X has not been compiled\0"
470 "malformed \\P or \\p sequence\0"
471 "unknown property name after \\P or \\p\0"
472 "subpattern name is too long (maximum " XSTRING(MAX_NAME_SIZE) " characters)\0"
473 "too many named subpatterns (maximum " XSTRING(MAX_NAME_COUNT) ")\0"
474 /* 50 */
475 "repeated subpattern is too long\0" /** DEAD **/
476 "octal value is greater than \\377 in 8-bit non-UTF-8 mode\0"
477 "internal error: overran compiling workspace\0"
478 "internal error: previously-checked referenced subpattern not found\0"
479 "DEFINE group contains more than one branch\0"
480 /* 55 */
481 "repeating a DEFINE group is not allowed\0" /** DEAD **/
482 "inconsistent NEWLINE options\0"
483 "\\g is not followed by a braced, angle-bracketed, or quoted name/number or by a plain number\0"
484 "a numbered reference must not be zero\0"
485 "an argument is not allowed for (*ACCEPT), (*FAIL), or (*COMMIT)\0"
486 /* 60 */
487 "(*VERB) not recognized\0"
488 "number is too big\0"
489 "subpattern name expected\0"
490 "digit expected after (?+\0"
491 "] is an invalid data character in JavaScript compatibility mode\0"
492 /* 65 */
493 "different names for subpatterns of the same number are not allowed\0"
494 "(*MARK) must have an argument\0"
495 "this version of PCRE is not compiled with Unicode property support\0"
496 "\\c must be followed by an ASCII character\0"
497 "\\k is not followed by a braced, angle-bracketed, or quoted name\0"
498 /* 70 */
499 "internal error: unknown opcode in find_fixedlength()\0"
500 "\\N is not supported in a class\0"
501 "too many forward references\0"
502 "disallowed Unicode code point (>= 0xd800 && <= 0xdfff)\0"
503 "invalid UTF-16 string\0"
504 /* 75 */
505 "name is too long in (*MARK), (*PRUNE), (*SKIP), or (*THEN)\0"
506 "character value in \\u.... sequence is too large\0"
507 "invalid UTF-32 string\0"
508 ;
509
510 /* Table to identify digits and hex digits. This is used when compiling
511 patterns. Note that the tables in chartables are dependent on the locale, and
512 may mark arbitrary characters as digits - but the PCRE compiling code expects
513 to handle only 0-9, a-z, and A-Z as digits when compiling. That is why we have
514 a private table here. It costs 256 bytes, but it is a lot faster than doing
515 character value tests (at least in some simple cases I timed), and in some
516 applications one wants PCRE to compile efficiently as well as match
517 efficiently.
518
519 For convenience, we use the same bit definitions as in chartables:
520
521 0x04 decimal digit
522 0x08 hexadecimal digit
523
524 Then we can use ctype_digit and ctype_xdigit in the code. */
525
526 /* Using a simple comparison for decimal numbers rather than a memory read
527 is much faster, and the resulting code is simpler (the compiler turns it
528 into a subtraction and unsigned comparison). */
529
530 #define IS_DIGIT(x) ((x) >= CHAR_0 && (x) <= CHAR_9)
531
532 #ifndef EBCDIC
533
534 /* This is the "normal" case, for ASCII systems, and EBCDIC systems running in
535 UTF-8 mode. */
536
537 static const pcre_uint8 digitab[] =
538 {
539 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 0- 7 */
540 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 8- 15 */
541 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 16- 23 */
542 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 24- 31 */
543 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - ' */
544 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* ( - / */
545 0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c, /* 0 - 7 */
546 0x0c,0x0c,0x00,0x00,0x00,0x00,0x00,0x00, /* 8 - ? */
547 0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* @ - G */
548 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* H - O */
549 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* P - W */
550 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* X - _ */
551 0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* ` - g */
552 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* h - o */
553 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* p - w */
554 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* x -127 */
555 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 128-135 */
556 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 136-143 */
557 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 144-151 */
558 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 152-159 */
559 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 160-167 */
560 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 168-175 */
561 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 176-183 */
562 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 184-191 */
563 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 192-199 */
564 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 200-207 */
565 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 208-215 */
566 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 216-223 */
567 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 224-231 */
568 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 232-239 */
569 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 240-247 */
570 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00};/* 248-255 */
571
572 #else
573
574 /* This is the "abnormal" case, for EBCDIC systems not running in UTF-8 mode. */
575
576 static const pcre_uint8 digitab[] =
577 {
578 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 0- 7 0 */
579 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 8- 15 */
580 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 16- 23 10 */
581 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 24- 31 */
582 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 32- 39 20 */
583 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 40- 47 */
584 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 48- 55 30 */
585 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 56- 63 */
586 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - 71 40 */
587 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 72- | */
588 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* & - 87 50 */
589 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 88- 95 */
590 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - -103 60 */
591 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 104- ? */
592 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 112-119 70 */
593 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 120- " */
594 0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* 128- g 80 */
595 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* h -143 */
596 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 144- p 90 */
597 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* q -159 */
598 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 160- x A0 */
599 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* y -175 */
600 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* ^ -183 B0 */
601 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 184-191 */
602 0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* { - G C0 */
603 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* H -207 */
604 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* } - P D0 */
605 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* Q -223 */
606 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* \ - X E0 */
607 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* Y -239 */
608 0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c, /* 0 - 7 F0 */
609 0x0c,0x0c,0x00,0x00,0x00,0x00,0x00,0x00};/* 8 -255 */
610
611 static const pcre_uint8 ebcdic_chartab[] = { /* chartable partial dup */
612 0x80,0x00,0x00,0x00,0x00,0x01,0x00,0x00, /* 0- 7 */
613 0x00,0x00,0x00,0x00,0x01,0x01,0x00,0x00, /* 8- 15 */
614 0x00,0x00,0x00,0x00,0x00,0x01,0x00,0x00, /* 16- 23 */
615 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 24- 31 */
616 0x00,0x00,0x00,0x00,0x00,0x01,0x00,0x00, /* 32- 39 */
617 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 40- 47 */
618 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 48- 55 */
619 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 56- 63 */
620 0x01,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - 71 */
621 0x00,0x00,0x00,0x80,0x00,0x80,0x80,0x80, /* 72- | */
622 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* & - 87 */
623 0x00,0x00,0x00,0x80,0x80,0x80,0x00,0x00, /* 88- 95 */
624 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - -103 */
625 0x00,0x00,0x00,0x00,0x00,0x10,0x00,0x80, /* 104- ? */
626 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 112-119 */
627 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 120- " */
628 0x00,0x1a,0x1a,0x1a,0x1a,0x1a,0x1a,0x12, /* 128- g */
629 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* h -143 */
630 0x00,0x12,0x12,0x12,0x12,0x12,0x12,0x12, /* 144- p */
631 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* q -159 */
632 0x00,0x00,0x12,0x12,0x12,0x12,0x12,0x12, /* 160- x */
633 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* y -175 */
634 0x80,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* ^ -183 */
635 0x00,0x00,0x80,0x00,0x00,0x00,0x00,0x00, /* 184-191 */
636 0x80,0x1a,0x1a,0x1a,0x1a,0x1a,0x1a,0x12, /* { - G */
637 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* H -207 */
638 0x00,0x12,0x12,0x12,0x12,0x12,0x12,0x12, /* } - P */
639 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* Q -223 */
640 0x00,0x00,0x12,0x12,0x12,0x12,0x12,0x12, /* \ - X */
641 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* Y -239 */
642 0x1c,0x1c,0x1c,0x1c,0x1c,0x1c,0x1c,0x1c, /* 0 - 7 */
643 0x1c,0x1c,0x00,0x00,0x00,0x00,0x00,0x00};/* 8 -255 */
644 #endif
645
646
647
648
649 /*************************************************
650 * Find an error text *
651 *************************************************/
652
653 /* The error texts are now all in one long string, to save on relocations. As
654 some of the text is of unknown length, we can't use a table of offsets.
655 Instead, just count through the strings. This is not a performance issue
656 because it happens only when there has been a compilation error.
657
658 Argument: the error number
659 Returns: pointer to the error string
660 */
661
662 static const char *
663 find_error_text(int n)
664 {
665 const char *s = error_texts;
666 for (; n > 0; n--)
667 {
668 while (*s++ != 0) {};
669 if (*s == 0) return "Error text not found (please report)";
670 }
671 return s;
672 }
673
674
675 /*************************************************
676 * Expand the workspace *
677 *************************************************/
678
679 /* This function is called during the second compiling phase, if the number of
680 forward references fills the existing workspace, which is originally a block on
681 the stack. A larger block is obtained from malloc() unless the ultimate limit
682 has been reached or the increase will be rather small.
683
684 Argument: pointer to the compile data block
685 Returns: 0 if all went well, else an error number
686 */
687
688 static int
689 expand_workspace(compile_data *cd)
690 {
691 pcre_uchar *newspace;
692 int newsize = cd->workspace_size * 2;
693
694 if (newsize > COMPILE_WORK_SIZE_MAX) newsize = COMPILE_WORK_SIZE_MAX;
695 if (cd->workspace_size >= COMPILE_WORK_SIZE_MAX ||
696 newsize - cd->workspace_size < WORK_SIZE_SAFETY_MARGIN)
697 return ERR72;
698
699 newspace = (PUBL(malloc))(IN_UCHARS(newsize));
700 if (newspace == NULL) return ERR21;
701 memcpy(newspace, cd->start_workspace, cd->workspace_size * sizeof(pcre_uchar));
702 cd->hwm = (pcre_uchar *)newspace + (cd->hwm - cd->start_workspace);
703 if (cd->workspace_size > COMPILE_WORK_SIZE)
704 (PUBL(free))((void *)cd->start_workspace);
705 cd->start_workspace = newspace;
706 cd->workspace_size = newsize;
707 return 0;
708 }
709
710
711
712 /*************************************************
713 * Check for counted repeat *
714 *************************************************/
715
716 /* This function is called when a '{' is encountered in a place where it might
717 start a quantifier. It looks ahead to see if it really is a quantifier or not.
718 It is only a quantifier if it is one of the forms {ddd} {ddd,} or {ddd,ddd}
719 where the ddds are digits.
720
721 Arguments:
722 p pointer to the first char after '{'
723
724 Returns: TRUE or FALSE
725 */
726
727 static BOOL
728 is_counted_repeat(const pcre_uchar *p)
729 {
730 if (!IS_DIGIT(*p)) return FALSE;
731 p++;
732 while (IS_DIGIT(*p)) p++;
733 if (*p == CHAR_RIGHT_CURLY_BRACKET) return TRUE;
734
735 if (*p++ != CHAR_COMMA) return FALSE;
736 if (*p == CHAR_RIGHT_CURLY_BRACKET) return TRUE;
737
738 if (!IS_DIGIT(*p)) return FALSE;
739 p++;
740 while (IS_DIGIT(*p)) p++;
741
742 return (*p == CHAR_RIGHT_CURLY_BRACKET);
743 }
744
745
746
747 /*************************************************
748 * Handle escapes *
749 *************************************************/
750
751 /* This function is called when a \ has been encountered. It either returns a
752 positive value for a simple escape such as \n, or 0 for a data character
753 which will be placed in chptr. A backreference to group n is returned as
754 negative n. When UTF-8 is enabled, a positive value greater than 255 may
755 be returned in chptr.
756 On entry,ptr is pointing at the \. On exit, it is on the final character of the
757 escape sequence.
758
759 Arguments:
760 ptrptr points to the pattern position pointer
761 chptr points to the data character
762 errorcodeptr points to the errorcode variable
763 bracount number of previous extracting brackets
764 options the options bits
765 isclass TRUE if inside a character class
766
767 Returns: zero => a data character
768 positive => a special escape sequence
769 negative => a back reference
770 on error, errorcodeptr is set
771 */
772
773 static int
774 check_escape(const pcre_uchar **ptrptr, pcre_uint32 *chptr, int *errorcodeptr,
775 int bracount, int options, BOOL isclass)
776 {
777 /* PCRE_UTF16 has the same value as PCRE_UTF8. */
778 BOOL utf = (options & PCRE_UTF8) != 0;
779 const pcre_uchar *ptr = *ptrptr + 1;
780 pcre_uint32 c;
781 int escape = 0;
782 int i;
783
784 GETCHARINCTEST(c, ptr); /* Get character value, increment pointer */
785 ptr--; /* Set pointer back to the last byte */
786
787 /* If backslash is at the end of the pattern, it's an error. */
788
789 if (c == 0) *errorcodeptr = ERR1;
790
791 /* Non-alphanumerics are literals. For digits or letters, do an initial lookup
792 in a table. A non-zero result is something that can be returned immediately.
793 Otherwise further processing may be required. */
794
795 #ifndef EBCDIC /* ASCII/UTF-8 coding */
796 /* Not alphanumeric */
797 else if (c < CHAR_0 || c > CHAR_z) {}
798 else if ((i = escapes[c - CHAR_0]) != 0) { if (i > 0) c = (pcre_uint32)i; else escape = -i; }
799
800 #else /* EBCDIC coding */
801 /* Not alphanumeric */
802 else if (c < CHAR_a || (!MAX_255(c) || (ebcdic_chartab[c] & 0x0E) == 0)) {}
803 else if ((i = escapes[c - 0x48]) != 0) { if (i > 0) c = (pcre_uint32)i; else escape = -i; }
804 #endif
805
806 /* Escapes that need further processing, or are illegal. */
807
808 else
809 {
810 const pcre_uchar *oldptr;
811 BOOL braced, negated, overflow;
812 int s;
813
814 switch (c)
815 {
816 /* A number of Perl escapes are not handled by PCRE. We give an explicit
817 error. */
818
819 case CHAR_l:
820 case CHAR_L:
821 *errorcodeptr = ERR37;
822 break;
823
824 case CHAR_u:
825 if ((options & PCRE_JAVASCRIPT_COMPAT) != 0)
826 {
827 /* In JavaScript, \u must be followed by four hexadecimal numbers.
828 Otherwise it is a lowercase u letter. */
829 if (MAX_255(ptr[1]) && (digitab[ptr[1]] & ctype_xdigit) != 0
830 && MAX_255(ptr[2]) && (digitab[ptr[2]] & ctype_xdigit) != 0
831 && MAX_255(ptr[3]) && (digitab[ptr[3]] & ctype_xdigit) != 0
832 && MAX_255(ptr[4]) && (digitab[ptr[4]] & ctype_xdigit) != 0)
833 {
834 c = 0;
835 for (i = 0; i < 4; ++i)
836 {
837 register pcre_uint32 cc = *(++ptr);
838 #ifndef EBCDIC /* ASCII/UTF-8 coding */
839 if (cc >= CHAR_a) cc -= 32; /* Convert to upper case */
840 c = (c << 4) + cc - ((cc < CHAR_A)? CHAR_0 : (CHAR_A - 10));
841 #else /* EBCDIC coding */
842 if (cc >= CHAR_a && cc <= CHAR_z) cc += 64; /* Convert to upper case */
843 c = (c << 4) + cc - ((cc >= CHAR_0)? CHAR_0 : (CHAR_A - 10));
844 #endif
845 }
846
847 #if defined COMPILE_PCRE8
848 if (c > (utf ? 0x10ffff : 0xff))
849 #elif defined COMPILE_PCRE16
850 if (c > (utf ? 0x10ffff : 0xffff))
851 #elif defined COMPILE_PCRE32
852 if (utf && c > 0x10ffff)
853 #endif
854 {
855 *errorcodeptr = ERR76;
856 }
857 else if (utf && c >= 0xd800 && c <= 0xdfff) *errorcodeptr = ERR73;
858 }
859 }
860 else
861 *errorcodeptr = ERR37;
862 break;
863
864 case CHAR_U:
865 /* In JavaScript, \U is an uppercase U letter. */
866 if ((options & PCRE_JAVASCRIPT_COMPAT) == 0) *errorcodeptr = ERR37;
867 break;
868
869 /* In a character class, \g is just a literal "g". Outside a character
870 class, \g must be followed by one of a number of specific things:
871
872 (1) A number, either plain or braced. If positive, it is an absolute
873 backreference. If negative, it is a relative backreference. This is a Perl
874 5.10 feature.
875
876 (2) Perl 5.10 also supports \g{name} as a reference to a named group. This
877 is part of Perl's movement towards a unified syntax for back references. As
878 this is synonymous with \k{name}, we fudge it up by pretending it really
879 was \k.
880
881 (3) For Oniguruma compatibility we also support \g followed by a name or a
882 number either in angle brackets or in single quotes. However, these are
883 (possibly recursive) subroutine calls, _not_ backreferences. Just return
884 the ESC_g code (cf \k). */
885
886 case CHAR_g:
887 if (isclass) break;
888 if (ptr[1] == CHAR_LESS_THAN_SIGN || ptr[1] == CHAR_APOSTROPHE)
889 {
890 escape = ESC_g;
891 break;
892 }
893
894 /* Handle the Perl-compatible cases */
895
896 if (ptr[1] == CHAR_LEFT_CURLY_BRACKET)
897 {
898 const pcre_uchar *p;
899 for (p = ptr+2; *p != 0 && *p != CHAR_RIGHT_CURLY_BRACKET; p++)
900 if (*p != CHAR_MINUS && !IS_DIGIT(*p)) break;
901 if (*p != 0 && *p != CHAR_RIGHT_CURLY_BRACKET)
902 {
903 escape = ESC_k;
904 break;
905 }
906 braced = TRUE;
907 ptr++;
908 }
909 else braced = FALSE;
910
911 if (ptr[1] == CHAR_MINUS)
912 {
913 negated = TRUE;
914 ptr++;
915 }
916 else negated = FALSE;
917
918 /* The integer range is limited by the machine's int representation. */
919 s = 0;
920 overflow = FALSE;
921 while (IS_DIGIT(ptr[1]))
922 {
923 if (s > INT_MAX / 10 - 1) /* Integer overflow */
924 {
925 overflow = TRUE;
926 break;
927 }
928 s = s * 10 + (int)(*(++ptr) - CHAR_0);
929 }
930 if (overflow) /* Integer overflow */
931 {
932 while (IS_DIGIT(ptr[1]))
933 ptr++;
934 *errorcodeptr = ERR61;
935 break;
936 }
937
938 if (braced && *(++ptr) != CHAR_RIGHT_CURLY_BRACKET)
939 {
940 *errorcodeptr = ERR57;
941 break;
942 }
943
944 if (s == 0)
945 {
946 *errorcodeptr = ERR58;
947 break;
948 }
949
950 if (negated)
951 {
952 if (s > bracount)
953 {
954 *errorcodeptr = ERR15;
955 break;
956 }
957 s = bracount - (s - 1);
958 }
959
960 escape = -s;
961 break;
962
963 /* The handling of escape sequences consisting of a string of digits
964 starting with one that is not zero is not straightforward. By experiment,
965 the way Perl works seems to be as follows:
966
967 Outside a character class, the digits are read as a decimal number. If the
968 number is less than 10, or if there are that many previous extracting
969 left brackets, then it is a back reference. Otherwise, up to three octal
970 digits are read to form an escaped byte. Thus \123 is likely to be octal
971 123 (cf \0123, which is octal 012 followed by the literal 3). If the octal
972 value is greater than 377, the least significant 8 bits are taken. Inside a
973 character class, \ followed by a digit is always an octal number. */
974
975 case CHAR_1: case CHAR_2: case CHAR_3: case CHAR_4: case CHAR_5:
976 case CHAR_6: case CHAR_7: case CHAR_8: case CHAR_9:
977
978 if (!isclass)
979 {
980 oldptr = ptr;
981 /* The integer range is limited by the machine's int representation. */
982 s = (int)(c -CHAR_0);
983 overflow = FALSE;
984 while (IS_DIGIT(ptr[1]))
985 {
986 if (s > INT_MAX / 10 - 1) /* Integer overflow */
987 {
988 overflow = TRUE;
989 break;
990 }
991 s = s * 10 + (int)(*(++ptr) - CHAR_0);
992 }
993 if (overflow) /* Integer overflow */
994 {
995 while (IS_DIGIT(ptr[1]))
996 ptr++;
997 *errorcodeptr = ERR61;
998 break;
999 }
1000 if (s < 10 || s <= bracount)
1001 {
1002 escape = -s;
1003 break;
1004 }
1005 ptr = oldptr; /* Put the pointer back and fall through */
1006 }
1007
1008 /* Handle an octal number following \. If the first digit is 8 or 9, Perl
1009 generates a binary zero byte and treats the digit as a following literal.
1010 Thus we have to pull back the pointer by one. */
1011
1012 if ((c = *ptr) >= CHAR_8)
1013 {
1014 ptr--;
1015 c = 0;
1016 break;
1017 }
1018
1019 /* \0 always starts an octal number, but we may drop through to here with a
1020 larger first octal digit. The original code used just to take the least
1021 significant 8 bits of octal numbers (I think this is what early Perls used
1022 to do). Nowadays we allow for larger numbers in UTF-8 mode and 16-bit mode,
1023 but no more than 3 octal digits. */
1024
1025 case CHAR_0:
1026 c -= CHAR_0;
1027 while(i++ < 2 && ptr[1] >= CHAR_0 && ptr[1] <= CHAR_7)
1028 c = c * 8 + *(++ptr) - CHAR_0;
1029 #ifdef COMPILE_PCRE8
1030 if (!utf && c > 0xff) *errorcodeptr = ERR51;
1031 #endif
1032 break;
1033
1034 /* \x is complicated. \x{ddd} is a character number which can be greater
1035 than 0xff in utf or non-8bit mode, but only if the ddd are hex digits.
1036 If not, { is treated as a data character. */
1037
1038 case CHAR_x:
1039 if ((options & PCRE_JAVASCRIPT_COMPAT) != 0)
1040 {
1041 /* In JavaScript, \x must be followed by two hexadecimal numbers.
1042 Otherwise it is a lowercase x letter. */
1043 if (MAX_255(ptr[1]) && (digitab[ptr[1]] & ctype_xdigit) != 0
1044 && MAX_255(ptr[2]) && (digitab[ptr[2]] & ctype_xdigit) != 0)
1045 {
1046 c = 0;
1047 for (i = 0; i < 2; ++i)
1048 {
1049 register pcre_uint32 cc = *(++ptr);
1050 #ifndef EBCDIC /* ASCII/UTF-8 coding */
1051 if (cc >= CHAR_a) cc -= 32; /* Convert to upper case */
1052 c = (c << 4) + cc - ((cc < CHAR_A)? CHAR_0 : (CHAR_A - 10));
1053 #else /* EBCDIC coding */
1054 if (cc >= CHAR_a && cc <= CHAR_z) cc += 64; /* Convert to upper case */
1055 c = (c << 4) + cc - ((cc >= CHAR_0)? CHAR_0 : (CHAR_A - 10));
1056 #endif
1057 }
1058 }
1059 break;
1060 }
1061
1062 if (ptr[1] == CHAR_LEFT_CURLY_BRACKET)
1063 {
1064 const pcre_uchar *pt = ptr + 2;
1065
1066 c = 0;
1067 overflow = FALSE;
1068 while (MAX_255(*pt) && (digitab[*pt] & ctype_xdigit) != 0)
1069 {
1070 register pcre_uint32 cc = *pt++;
1071 if (c == 0 && cc == CHAR_0) continue; /* Leading zeroes */
1072
1073 #ifdef COMPILE_PCRE32
1074 if (c >= 0x10000000l) { overflow = TRUE; break; }
1075 #endif
1076
1077 #ifndef EBCDIC /* ASCII/UTF-8 coding */
1078 if (cc >= CHAR_a) cc -= 32; /* Convert to upper case */
1079 c = (c << 4) + cc - ((cc < CHAR_A)? CHAR_0 : (CHAR_A - 10));
1080 #else /* EBCDIC coding */
1081 if (cc >= CHAR_a && cc <= CHAR_z) cc += 64; /* Convert to upper case */
1082 c = (c << 4) + cc - ((cc >= CHAR_0)? CHAR_0 : (CHAR_A - 10));
1083 #endif
1084
1085 #if defined COMPILE_PCRE8
1086 if (c > (utf ? 0x10ffff : 0xff)) { overflow = TRUE; break; }
1087 #elif defined COMPILE_PCRE16
1088 if (c > (utf ? 0x10ffff : 0xffff)) { overflow = TRUE; break; }
1089 #elif defined COMPILE_PCRE32
1090 if (utf && c > 0x10ffff) { overflow = TRUE; break; }
1091 #endif
1092 }
1093
1094 if (overflow)
1095 {
1096 while (MAX_255(*pt) && (digitab[*pt] & ctype_xdigit) != 0) pt++;
1097 *errorcodeptr = ERR34;
1098 }
1099
1100 if (*pt == CHAR_RIGHT_CURLY_BRACKET)
1101 {
1102 if (utf && c >= 0xd800 && c <= 0xdfff) *errorcodeptr = ERR73;
1103 ptr = pt;
1104 break;
1105 }
1106
1107 /* If the sequence of hex digits does not end with '}', then we don't
1108 recognize this construct; fall through to the normal \x handling. */
1109 }
1110
1111 /* Read just a single-byte hex-defined char */
1112
1113 c = 0;
1114 while (i++ < 2 && MAX_255(ptr[1]) && (digitab[ptr[1]] & ctype_xdigit) != 0)
1115 {
1116 pcre_uint32 cc; /* Some compilers don't like */
1117 cc = *(++ptr); /* ++ in initializers */
1118 #ifndef EBCDIC /* ASCII/UTF-8 coding */
1119 if (cc >= CHAR_a) cc -= 32; /* Convert to upper case */
1120 c = c * 16 + cc - ((cc < CHAR_A)? CHAR_0 : (CHAR_A - 10));
1121 #else /* EBCDIC coding */
1122 if (cc <= CHAR_z) cc += 64; /* Convert to upper case */
1123 c = c * 16 + cc - ((cc >= CHAR_0)? CHAR_0 : (CHAR_A - 10));
1124 #endif
1125 }
1126 break;
1127
1128 /* For \c, a following letter is upper-cased; then the 0x40 bit is flipped.
1129 An error is given if the byte following \c is not an ASCII character. This
1130 coding is ASCII-specific, but then the whole concept of \cx is
1131 ASCII-specific. (However, an EBCDIC equivalent has now been added.) */
1132
1133 case CHAR_c:
1134 c = *(++ptr);
1135 if (c == 0)
1136 {
1137 *errorcodeptr = ERR2;
1138 break;
1139 }
1140 #ifndef EBCDIC /* ASCII/UTF-8 coding */
1141 if (c > 127) /* Excludes all non-ASCII in either mode */
1142 {
1143 *errorcodeptr = ERR68;
1144 break;
1145 }
1146 if (c >= CHAR_a && c <= CHAR_z) c -= 32;
1147 c ^= 0x40;
1148 #else /* EBCDIC coding */
1149 if (c >= CHAR_a && c <= CHAR_z) c += 64;
1150 c ^= 0xC0;
1151 #endif
1152 break;
1153
1154 /* PCRE_EXTRA enables extensions to Perl in the matter of escapes. Any
1155 other alphanumeric following \ is an error if PCRE_EXTRA was set;
1156 otherwise, for Perl compatibility, it is a literal. This code looks a bit
1157 odd, but there used to be some cases other than the default, and there may
1158 be again in future, so I haven't "optimized" it. */
1159
1160 default:
1161 if ((options & PCRE_EXTRA) != 0) switch(c)
1162 {
1163 default:
1164 *errorcodeptr = ERR3;
1165 break;
1166 }
1167 break;
1168 }
1169 }
1170
1171 /* Perl supports \N{name} for character names, as well as plain \N for "not
1172 newline". PCRE does not support \N{name}. However, it does support
1173 quantification such as \N{2,3}. */
1174
1175 if (escape == ESC_N && ptr[1] == CHAR_LEFT_CURLY_BRACKET &&
1176 !is_counted_repeat(ptr+2))
1177 *errorcodeptr = ERR37;
1178
1179 /* If PCRE_UCP is set, we change the values for \d etc. */
1180
1181 if ((options & PCRE_UCP) != 0 && escape >= ESC_D && escape <= ESC_w)
1182 escape += (ESC_DU - ESC_D);
1183
1184 /* Set the pointer to the final character before returning. */
1185
1186 *ptrptr = ptr;
1187 *chptr = c;
1188 return escape;
1189 }
1190
1191 #ifdef SUPPORT_UCP
1192 /*************************************************
1193 * Handle \P and \p *
1194 *************************************************/
1195
1196 /* This function is called after \P or \p has been encountered, provided that
1197 PCRE is compiled with support for Unicode properties. On entry, ptrptr is
1198 pointing at the P or p. On exit, it is pointing at the final character of the
1199 escape sequence.
1200
1201 Argument:
1202 ptrptr points to the pattern position pointer
1203 negptr points to a boolean that is set TRUE for negation else FALSE
1204 dptr points to an int that is set to the detailed property value
1205 errorcodeptr points to the error code variable
1206
1207 Returns: type value from ucp_type_table, or -1 for an invalid type
1208 */
1209
1210 static int
1211 get_ucp(const pcre_uchar **ptrptr, BOOL *negptr, int *dptr, int *errorcodeptr)
1212 {
1213 int c, i, bot, top;
1214 const pcre_uchar *ptr = *ptrptr;
1215 pcre_uchar name[32];
1216
1217 c = *(++ptr);
1218 if (c == 0) goto ERROR_RETURN;
1219
1220 *negptr = FALSE;
1221
1222 /* \P or \p can be followed by a name in {}, optionally preceded by ^ for
1223 negation. */
1224
1225 if (c == CHAR_LEFT_CURLY_BRACKET)
1226 {
1227 if (ptr[1] == CHAR_CIRCUMFLEX_ACCENT)
1228 {
1229 *negptr = TRUE;
1230 ptr++;
1231 }
1232 for (i = 0; i < (int)(sizeof(name) / sizeof(pcre_uchar)) - 1; i++)
1233 {
1234 c = *(++ptr);
1235 if (c == 0) goto ERROR_RETURN;
1236 if (c == CHAR_RIGHT_CURLY_BRACKET) break;
1237 name[i] = c;
1238 }
1239 if (c != CHAR_RIGHT_CURLY_BRACKET) goto ERROR_RETURN;
1240 name[i] = 0;
1241 }
1242
1243 /* Otherwise there is just one following character */
1244
1245 else
1246 {
1247 name[0] = c;
1248 name[1] = 0;
1249 }
1250
1251 *ptrptr = ptr;
1252
1253 /* Search for a recognized property name using binary chop */
1254
1255 bot = 0;
1256 top = PRIV(utt_size);
1257
1258 while (bot < top)
1259 {
1260 i = (bot + top) >> 1;
1261 c = STRCMP_UC_C8(name, PRIV(utt_names) + PRIV(utt)[i].name_offset);
1262 if (c == 0)
1263 {
1264 *dptr = PRIV(utt)[i].value;
1265 return PRIV(utt)[i].type;
1266 }
1267 if (c > 0) bot = i + 1; else top = i;
1268 }
1269
1270 *errorcodeptr = ERR47;
1271 *ptrptr = ptr;
1272 return -1;
1273
1274 ERROR_RETURN:
1275 *errorcodeptr = ERR46;
1276 *ptrptr = ptr;
1277 return -1;
1278 }
1279 #endif
1280
1281
1282
1283
1284 /*************************************************
1285 * Read repeat counts *
1286 *************************************************/
1287
1288 /* Read an item of the form {n,m} and return the values. This is called only
1289 after is_counted_repeat() has confirmed that a repeat-count quantifier exists,
1290 so the syntax is guaranteed to be correct, but we need to check the values.
1291
1292 Arguments:
1293 p pointer to first char after '{'
1294 minp pointer to int for min
1295 maxp pointer to int for max
1296 returned as -1 if no max
1297 errorcodeptr points to error code variable
1298
1299 Returns: pointer to '}' on success;
1300 current ptr on error, with errorcodeptr set non-zero
1301 */
1302
1303 static const pcre_uchar *
1304 read_repeat_counts(const pcre_uchar *p, int *minp, int *maxp, int *errorcodeptr)
1305 {
1306 int min = 0;
1307 int max = -1;
1308
1309 /* Read the minimum value and do a paranoid check: a negative value indicates
1310 an integer overflow. */
1311
1312 while (IS_DIGIT(*p)) min = min * 10 + *p++ - CHAR_0;
1313 if (min < 0 || min > 65535)
1314 {
1315 *errorcodeptr = ERR5;
1316 return p;
1317 }
1318
1319 /* Read the maximum value if there is one, and again do a paranoid on its size.
1320 Also, max must not be less than min. */
1321
1322 if (*p == CHAR_RIGHT_CURLY_BRACKET) max = min; else
1323 {
1324 if (*(++p) != CHAR_RIGHT_CURLY_BRACKET)
1325 {
1326 max = 0;
1327 while(IS_DIGIT(*p)) max = max * 10 + *p++ - CHAR_0;
1328 if (max < 0 || max > 65535)
1329 {
1330 *errorcodeptr = ERR5;
1331 return p;
1332 }
1333 if (max < min)
1334 {
1335 *errorcodeptr = ERR4;
1336 return p;
1337 }
1338 }
1339 }
1340
1341 /* Fill in the required variables, and pass back the pointer to the terminating
1342 '}'. */
1343
1344 *minp = min;
1345 *maxp = max;
1346 return p;
1347 }
1348
1349
1350
1351 /*************************************************
1352 * Subroutine for finding forward reference *
1353 *************************************************/
1354
1355 /* This recursive function is called only from find_parens() below. The
1356 top-level call starts at the beginning of the pattern. All other calls must
1357 start at a parenthesis. It scans along a pattern's text looking for capturing
1358 subpatterns, and counting them. If it finds a named pattern that matches the
1359 name it is given, it returns its number. Alternatively, if the name is NULL, it
1360 returns when it reaches a given numbered subpattern. Recursion is used to keep
1361 track of subpatterns that reset the capturing group numbers - the (?| feature.
1362
1363 This function was originally called only from the second pass, in which we know
1364 that if (?< or (?' or (?P< is encountered, the name will be correctly
1365 terminated because that is checked in the first pass. There is now one call to
1366 this function in the first pass, to check for a recursive back reference by
1367 name (so that we can make the whole group atomic). In this case, we need check
1368 only up to the current position in the pattern, and that is still OK because
1369 and previous occurrences will have been checked. To make this work, the test
1370 for "end of pattern" is a check against cd->end_pattern in the main loop,
1371 instead of looking for a binary zero. This means that the special first-pass
1372 call can adjust cd->end_pattern temporarily. (Checks for binary zero while
1373 processing items within the loop are OK, because afterwards the main loop will
1374 terminate.)
1375
1376 Arguments:
1377 ptrptr address of the current character pointer (updated)
1378 cd compile background data
1379 name name to seek, or NULL if seeking a numbered subpattern
1380 lorn name length, or subpattern number if name is NULL
1381 xmode TRUE if we are in /x mode
1382 utf TRUE if we are in UTF-8 / UTF-16 / UTF-32 mode
1383 count pointer to the current capturing subpattern number (updated)
1384
1385 Returns: the number of the named subpattern, or -1 if not found
1386 */
1387
1388 static int
1389 find_parens_sub(pcre_uchar **ptrptr, compile_data *cd, const pcre_uchar *name, int lorn,
1390 BOOL xmode, BOOL utf, int *count)
1391 {
1392 pcre_uchar *ptr = *ptrptr;
1393 int start_count = *count;
1394 int hwm_count = start_count;
1395 BOOL dup_parens = FALSE;
1396
1397 /* If the first character is a parenthesis, check on the type of group we are
1398 dealing with. The very first call may not start with a parenthesis. */
1399
1400 if (ptr[0] == CHAR_LEFT_PARENTHESIS)
1401 {
1402 /* Handle specials such as (*SKIP) or (*UTF8) etc. */
1403
1404 if (ptr[1] == CHAR_ASTERISK) ptr += 2;
1405
1406 /* Handle a normal, unnamed capturing parenthesis. */
1407
1408 else if (ptr[1] != CHAR_QUESTION_MARK)
1409 {
1410 *count += 1;
1411 if (name == NULL && *count == lorn) return *count;
1412 ptr++;
1413 }
1414
1415 /* All cases now have (? at the start. Remember when we are in a group
1416 where the parenthesis numbers are duplicated. */
1417
1418 else if (ptr[2] == CHAR_VERTICAL_LINE)
1419 {
1420 ptr += 3;
1421 dup_parens = TRUE;
1422 }
1423
1424 /* Handle comments; all characters are allowed until a ket is reached. */
1425
1426 else if (ptr[2] == CHAR_NUMBER_SIGN)
1427 {
1428 for (ptr += 3; *ptr != 0; ptr++) if (*ptr == CHAR_RIGHT_PARENTHESIS) break;
1429 goto FAIL_EXIT;
1430 }
1431
1432 /* Handle a condition. If it is an assertion, just carry on so that it
1433 is processed as normal. If not, skip to the closing parenthesis of the
1434 condition (there can't be any nested parens). */
1435
1436 else if (ptr[2] == CHAR_LEFT_PARENTHESIS)
1437 {
1438 ptr += 2;
1439 if (ptr[1] != CHAR_QUESTION_MARK)
1440 {
1441 while (*ptr != 0 && *ptr != CHAR_RIGHT_PARENTHESIS) ptr++;
1442 if (*ptr != 0) ptr++;
1443 }
1444 }
1445
1446 /* Start with (? but not a condition. */
1447
1448 else
1449 {
1450 ptr += 2;
1451 if (*ptr == CHAR_P) ptr++; /* Allow optional P */
1452
1453 /* We have to disambiguate (?<! and (?<= from (?<name> for named groups */
1454
1455 if ((*ptr == CHAR_LESS_THAN_SIGN && ptr[1] != CHAR_EXCLAMATION_MARK &&
1456 ptr[1] != CHAR_EQUALS_SIGN) || *ptr == CHAR_APOSTROPHE)
1457 {
1458 int term;
1459 const pcre_uchar *thisname;
1460 *count += 1;
1461 if (name == NULL && *count == lorn) return *count;
1462 term = *ptr++;
1463 if (term == CHAR_LESS_THAN_SIGN) term = CHAR_GREATER_THAN_SIGN;
1464 thisname = ptr;
1465 while (*ptr != term) ptr++;
1466 if (name != NULL && lorn == ptr - thisname &&
1467 STRNCMP_UC_UC(name, thisname, lorn) == 0)
1468 return *count;
1469 term++;
1470 }
1471 }
1472 }
1473
1474 /* Past any initial parenthesis handling, scan for parentheses or vertical
1475 bars. Stop if we get to cd->end_pattern. Note that this is important for the
1476 first-pass call when this value is temporarily adjusted to stop at the current
1477 position. So DO NOT change this to a test for binary zero. */
1478
1479 for (; ptr < cd->end_pattern; ptr++)
1480 {
1481 /* Skip over backslashed characters and also entire \Q...\E */
1482
1483 if (*ptr == CHAR_BACKSLASH)
1484 {
1485 if (*(++ptr) == 0) goto FAIL_EXIT;
1486 if (*ptr == CHAR_Q) for (;;)
1487 {
1488 while (*(++ptr) != 0 && *ptr != CHAR_BACKSLASH) {};
1489 if (*ptr == 0) goto FAIL_EXIT;
1490 if (*(++ptr) == CHAR_E) break;
1491 }
1492 continue;
1493 }
1494
1495 /* Skip over character classes; this logic must be similar to the way they
1496 are handled for real. If the first character is '^', skip it. Also, if the
1497 first few characters (either before or after ^) are \Q\E or \E we skip them
1498 too. This makes for compatibility with Perl. Note the use of STR macros to
1499 encode "Q\\E" so that it works in UTF-8 on EBCDIC platforms. */
1500
1501 if (*ptr == CHAR_LEFT_SQUARE_BRACKET)
1502 {
1503 BOOL negate_class = FALSE;
1504 for (;;)
1505 {
1506 if (ptr[1] == CHAR_BACKSLASH)
1507 {
1508 if (ptr[2] == CHAR_E)
1509 ptr+= 2;
1510 else if (STRNCMP_UC_C8(ptr + 2,
1511 STR_Q STR_BACKSLASH STR_E, 3) == 0)
1512 ptr += 4;
1513 else
1514 break;
1515 }
1516 else if (!negate_class && ptr[1] == CHAR_CIRCUMFLEX_ACCENT)
1517 {
1518 negate_class = TRUE;
1519 ptr++;
1520 }
1521 else break;
1522 }
1523
1524 /* If the next character is ']', it is a data character that must be
1525 skipped, except in JavaScript compatibility mode. */
1526
1527 if (ptr[1] == CHAR_RIGHT_SQUARE_BRACKET &&
1528 (cd->external_options & PCRE_JAVASCRIPT_COMPAT) == 0)
1529 ptr++;
1530
1531 while (*(++ptr) != CHAR_RIGHT_SQUARE_BRACKET)
1532 {
1533 if (*ptr == 0) return -1;
1534 if (*ptr == CHAR_BACKSLASH)
1535 {
1536 if (*(++ptr) == 0) goto FAIL_EXIT;
1537 if (*ptr == CHAR_Q) for (;;)
1538 {
1539 while (*(++ptr) != 0 && *ptr != CHAR_BACKSLASH) {};
1540 if (*ptr == 0) goto FAIL_EXIT;
1541 if (*(++ptr) == CHAR_E) break;
1542 }
1543 continue;
1544 }
1545 }
1546 continue;
1547 }
1548
1549 /* Skip comments in /x mode */
1550
1551 if (xmode && *ptr == CHAR_NUMBER_SIGN)
1552 {
1553 ptr++;
1554 while (*ptr != 0)
1555 {
1556 if (IS_NEWLINE(ptr)) { ptr += cd->nllen - 1; break; }
1557 ptr++;
1558 #ifdef SUPPORT_UTF
1559 if (utf) FORWARDCHAR(ptr);
1560 #endif
1561 }
1562 if (*ptr == 0) goto FAIL_EXIT;
1563 continue;
1564 }
1565
1566 /* Check for the special metacharacters */
1567
1568 if (*ptr == CHAR_LEFT_PARENTHESIS)
1569 {
1570 int rc = find_parens_sub(&ptr, cd, name, lorn, xmode, utf, count);
1571 if (rc > 0) return rc;
1572 if (*ptr == 0) goto FAIL_EXIT;
1573 }
1574
1575 else if (*ptr == CHAR_RIGHT_PARENTHESIS)
1576 {
1577 if (dup_parens && *count < hwm_count) *count = hwm_count;
1578 goto FAIL_EXIT;
1579 }
1580
1581 else if (*ptr == CHAR_VERTICAL_LINE && dup_parens)
1582 {
1583 if (*count > hwm_count) hwm_count = *count;
1584 *count = start_count;
1585 }
1586 }
1587
1588 FAIL_EXIT:
1589 *ptrptr = ptr;
1590 return -1;
1591 }
1592
1593
1594
1595
1596 /*************************************************
1597 * Find forward referenced subpattern *
1598 *************************************************/
1599
1600 /* This function scans along a pattern's text looking for capturing
1601 subpatterns, and counting them. If it finds a named pattern that matches the
1602 name it is given, it returns its number. Alternatively, if the name is NULL, it
1603 returns when it reaches a given numbered subpattern. This is used for forward
1604 references to subpatterns. We used to be able to start this scan from the
1605 current compiling point, using the current count value from cd->bracount, and
1606 do it all in a single loop, but the addition of the possibility of duplicate
1607 subpattern numbers means that we have to scan from the very start, in order to
1608 take account of such duplicates, and to use a recursive function to keep track
1609 of the different types of group.
1610
1611 Arguments:
1612 cd compile background data
1613 name name to seek, or NULL if seeking a numbered subpattern
1614 lorn name length, or subpattern number if name is NULL
1615 xmode TRUE if we are in /x mode
1616 utf TRUE if we are in UTF-8 / UTF-16 / UTF-32 mode
1617
1618 Returns: the number of the found subpattern, or -1 if not found
1619 */
1620
1621 static int
1622 find_parens(compile_data *cd, const pcre_uchar *name, int lorn, BOOL xmode,
1623 BOOL utf)
1624 {
1625 pcre_uchar *ptr = (pcre_uchar *)cd->start_pattern;
1626 int count = 0;
1627 int rc;
1628
1629 /* If the pattern does not start with an opening parenthesis, the first call
1630 to find_parens_sub() will scan right to the end (if necessary). However, if it
1631 does start with a parenthesis, find_parens_sub() will return when it hits the
1632 matching closing parens. That is why we have to have a loop. */
1633
1634 for (;;)
1635 {
1636 rc = find_parens_sub(&ptr, cd, name, lorn, xmode, utf, &count);
1637 if (rc > 0 || *ptr++ == 0) break;
1638 }
1639
1640 return rc;
1641 }
1642
1643
1644
1645
1646 /*************************************************
1647 * Find first significant op code *
1648 *************************************************/
1649
1650 /* This is called by several functions that scan a compiled expression looking
1651 for a fixed first character, or an anchoring op code etc. It skips over things
1652 that do not influence this. For some calls, it makes sense to skip negative
1653 forward and all backward assertions, and also the \b assertion; for others it
1654 does not.
1655
1656 Arguments:
1657 code pointer to the start of the group
1658 skipassert TRUE if certain assertions are to be skipped
1659
1660 Returns: pointer to the first significant opcode
1661 */
1662
1663 static const pcre_uchar*
1664 first_significant_code(const pcre_uchar *code, BOOL skipassert)
1665 {
1666 for (;;)
1667 {
1668 switch ((int)*code)
1669 {
1670 case OP_ASSERT_NOT:
1671 case OP_ASSERTBACK:
1672 case OP_ASSERTBACK_NOT:
1673 if (!skipassert) return code;
1674 do code += GET(code, 1); while (*code == OP_ALT);
1675 code += PRIV(OP_lengths)[*code];
1676 break;
1677
1678 case OP_WORD_BOUNDARY:
1679 case OP_NOT_WORD_BOUNDARY:
1680 if (!skipassert) return code;
1681 /* Fall through */
1682
1683 case OP_CALLOUT:
1684 case OP_CREF:
1685 case OP_NCREF:
1686 case OP_RREF:
1687 case OP_NRREF:
1688 case OP_DEF:
1689 code += PRIV(OP_lengths)[*code];
1690 break;
1691
1692 default:
1693 return code;
1694 }
1695 }
1696 /* Control never reaches here */
1697 }
1698
1699
1700
1701
1702 /*************************************************
1703 * Find the fixed length of a branch *
1704 *************************************************/
1705
1706 /* Scan a branch and compute the fixed length of subject that will match it,
1707 if the length is fixed. This is needed for dealing with backward assertions.
1708 In UTF8 mode, the result is in characters rather than bytes. The branch is
1709 temporarily terminated with OP_END when this function is called.
1710
1711 This function is called when a backward assertion is encountered, so that if it
1712 fails, the error message can point to the correct place in the pattern.
1713 However, we cannot do this when the assertion contains subroutine calls,
1714 because they can be forward references. We solve this by remembering this case
1715 and doing the check at the end; a flag specifies which mode we are running in.
1716
1717 Arguments:
1718 code points to the start of the pattern (the bracket)
1719 utf TRUE in UTF-8 / UTF-16 / UTF-32 mode
1720 atend TRUE if called when the pattern is complete
1721 cd the "compile data" structure
1722
1723 Returns: the fixed length,
1724 or -1 if there is no fixed length,
1725 or -2 if \C was encountered (in UTF-8 mode only)
1726 or -3 if an OP_RECURSE item was encountered and atend is FALSE
1727 or -4 if an unknown opcode was encountered (internal error)
1728 */
1729
1730 static int
1731 find_fixedlength(pcre_uchar *code, BOOL utf, BOOL atend, compile_data *cd)
1732 {
1733 int length = -1;
1734
1735 register int branchlength = 0;
1736 register pcre_uchar *cc = code + 1 + LINK_SIZE;
1737
1738 /* Scan along the opcodes for this branch. If we get to the end of the
1739 branch, check the length against that of the other branches. */
1740
1741 for (;;)
1742 {
1743 int d;
1744 pcre_uchar *ce, *cs;
1745 register int op = *cc;
1746
1747 switch (op)
1748 {
1749 /* We only need to continue for OP_CBRA (normal capturing bracket) and
1750 OP_BRA (normal non-capturing bracket) because the other variants of these
1751 opcodes are all concerned with unlimited repeated groups, which of course
1752 are not of fixed length. */
1753
1754 case OP_CBRA:
1755 case OP_BRA:
1756 case OP_ONCE:
1757 case OP_ONCE_NC:
1758 case OP_COND:
1759 d = find_fixedlength(cc + ((op == OP_CBRA)? IMM2_SIZE : 0), utf, atend, cd);
1760 if (d < 0) return d;
1761 branchlength += d;
1762 do cc += GET(cc, 1); while (*cc == OP_ALT);
1763 cc += 1 + LINK_SIZE;
1764 break;
1765
1766 /* Reached end of a branch; if it's a ket it is the end of a nested call.
1767 If it's ALT it is an alternation in a nested call. An ACCEPT is effectively
1768 an ALT. If it is END it's the end of the outer call. All can be handled by
1769 the same code. Note that we must not include the OP_KETRxxx opcodes here,
1770 because they all imply an unlimited repeat. */
1771
1772 case OP_ALT:
1773 case OP_KET:
1774 case OP_END:
1775 case OP_ACCEPT:
1776 case OP_ASSERT_ACCEPT:
1777 if (length < 0) length = branchlength;
1778 else if (length != branchlength) return -1;
1779 if (*cc != OP_ALT) return length;
1780 cc += 1 + LINK_SIZE;
1781 branchlength = 0;
1782 break;
1783
1784 /* A true recursion implies not fixed length, but a subroutine call may
1785 be OK. If the subroutine is a forward reference, we can't deal with
1786 it until the end of the pattern, so return -3. */
1787
1788 case OP_RECURSE:
1789 if (!atend) return -3;
1790 cs = ce = (pcre_uchar *)cd->start_code + GET(cc, 1); /* Start subpattern */
1791 do ce += GET(ce, 1); while (*ce == OP_ALT); /* End subpattern */
1792 if (cc > cs && cc < ce) return -1; /* Recursion */
1793 d = find_fixedlength(cs + IMM2_SIZE, utf, atend, cd);
1794 if (d < 0) return d;
1795 branchlength += d;
1796 cc += 1 + LINK_SIZE;
1797 break;
1798
1799 /* Skip over assertive subpatterns */
1800
1801 case OP_ASSERT:
1802 case OP_ASSERT_NOT:
1803 case OP_ASSERTBACK:
1804 case OP_ASSERTBACK_NOT:
1805 do cc += GET(cc, 1); while (*cc == OP_ALT);
1806 cc += PRIV(OP_lengths)[*cc];
1807 break;
1808
1809 /* Skip over things that don't match chars */
1810
1811 case OP_MARK:
1812 case OP_PRUNE_ARG:
1813 case OP_SKIP_ARG:
1814 case OP_THEN_ARG:
1815 cc += cc[1] + PRIV(OP_lengths)[*cc];
1816 break;
1817
1818 case OP_CALLOUT:
1819 case OP_CIRC:
1820 case OP_CIRCM:
1821 case OP_CLOSE:
1822 case OP_COMMIT:
1823 case OP_CREF:
1824 case OP_DEF:
1825 case OP_DOLL:
1826 case OP_DOLLM:
1827 case OP_EOD:
1828 case OP_EODN:
1829 case OP_FAIL:
1830 case OP_NCREF:
1831 case OP_NRREF:
1832 case OP_NOT_WORD_BOUNDARY:
1833 case OP_PRUNE:
1834 case OP_REVERSE:
1835 case OP_RREF:
1836 case OP_SET_SOM:
1837 case OP_SKIP:
1838 case OP_SOD:
1839 case OP_SOM:
1840 case OP_THEN:
1841 case OP_WORD_BOUNDARY:
1842 cc += PRIV(OP_lengths)[*cc];
1843 break;
1844
1845 /* Handle literal characters */
1846
1847 case OP_CHAR:
1848 case OP_CHARI:
1849 case OP_NOT:
1850 case OP_NOTI:
1851 branchlength++;
1852 cc += 2;
1853 #if defined SUPPORT_UTF && !defined COMPILE_PCRE32
1854 if (utf && HAS_EXTRALEN(cc[-1])) cc += GET_EXTRALEN(cc[-1]);
1855 #endif
1856 break;
1857
1858 /* Handle exact repetitions. The count is already in characters, but we
1859 need to skip over a multibyte character in UTF8 mode. */
1860
1861 case OP_EXACT:
1862 case OP_EXACTI:
1863 case OP_NOTEXACT:
1864 case OP_NOTEXACTI:
1865 branchlength += GET2(cc,1);
1866 cc += 2 + IMM2_SIZE;
1867 #if defined SUPPORT_UTF && !defined COMPILE_PCRE32
1868 if (utf && HAS_EXTRALEN(cc[-1])) cc += GET_EXTRALEN(cc[-1]);
1869 #endif
1870 break;
1871
1872 case OP_TYPEEXACT:
1873 branchlength += GET2(cc,1);
1874 if (cc[1 + IMM2_SIZE] == OP_PROP || cc[1 + IMM2_SIZE] == OP_NOTPROP)
1875 cc += 2;
1876 cc += 1 + IMM2_SIZE + 1;
1877 break;
1878
1879 /* Handle single-char matchers */
1880
1881 case OP_PROP:
1882 case OP_NOTPROP:
1883 cc += 2;
1884 /* Fall through */
1885
1886 case OP_HSPACE:
1887 case OP_VSPACE:
1888 case OP_NOT_HSPACE:
1889 case OP_NOT_VSPACE:
1890 case OP_NOT_DIGIT:
1891 case OP_DIGIT:
1892 case OP_NOT_WHITESPACE:
1893 case OP_WHITESPACE:
1894 case OP_NOT_WORDCHAR:
1895 case OP_WORDCHAR:
1896 case OP_ANY:
1897 case OP_ALLANY:
1898 branchlength++;
1899 cc++;
1900 break;
1901
1902 /* The single-byte matcher isn't allowed. This only happens in UTF-8 mode;
1903 otherwise \C is coded as OP_ALLANY. */
1904
1905 case OP_ANYBYTE:
1906 return -2;
1907
1908 /* Check a class for variable quantification */
1909
1910 #if defined SUPPORT_UTF || defined COMPILE_PCRE16 || defined COMPILE_PCRE32
1911 case OP_XCLASS:
1912 cc += GET(cc, 1) - PRIV(OP_lengths)[OP_CLASS];
1913 /* Fall through */
1914 #endif
1915
1916 case OP_CLASS:
1917 case OP_NCLASS:
1918 cc += PRIV(OP_lengths)[OP_CLASS];
1919
1920 switch (*cc)
1921 {
1922 case OP_CRPLUS:
1923 case OP_CRMINPLUS:
1924 case OP_CRSTAR:
1925 case OP_CRMINSTAR:
1926 case OP_CRQUERY:
1927 case OP_CRMINQUERY:
1928 return -1;
1929
1930 case OP_CRRANGE:
1931 case OP_CRMINRANGE:
1932 if (GET2(cc,1) != GET2(cc,1+IMM2_SIZE)) return -1;
1933 branchlength += GET2(cc,1);
1934 cc += 1 + 2 * IMM2_SIZE;
1935 break;
1936
1937 default:
1938 branchlength++;
1939 }
1940 break;
1941
1942 /* Anything else is variable length */
1943
1944 case OP_ANYNL:
1945 case OP_BRAMINZERO:
1946 case OP_BRAPOS:
1947 case OP_BRAPOSZERO:
1948 case OP_BRAZERO:
1949 case OP_CBRAPOS:
1950 case OP_EXTUNI:
1951 case OP_KETRMAX:
1952 case OP_KETRMIN:
1953 case OP_KETRPOS:
1954 case OP_MINPLUS:
1955 case OP_MINPLUSI:
1956 case OP_MINQUERY:
1957 case OP_MINQUERYI:
1958 case OP_MINSTAR:
1959 case OP_MINSTARI:
1960 case OP_MINUPTO:
1961 case OP_MINUPTOI:
1962 case OP_NOTMINPLUS:
1963 case OP_NOTMINPLUSI:
1964 case OP_NOTMINQUERY:
1965 case OP_NOTMINQUERYI:
1966 case OP_NOTMINSTAR:
1967 case OP_NOTMINSTARI:
1968 case OP_NOTMINUPTO:
1969 case OP_NOTMINUPTOI:
1970 case OP_NOTPLUS:
1971 case OP_NOTPLUSI:
1972 case OP_NOTPOSPLUS:
1973 case OP_NOTPOSPLUSI:
1974 case OP_NOTPOSQUERY:
1975 case OP_NOTPOSQUERYI:
1976 case OP_NOTPOSSTAR:
1977 case OP_NOTPOSSTARI:
1978 case OP_NOTPOSUPTO:
1979 case OP_NOTPOSUPTOI:
1980 case OP_NOTQUERY:
1981 case OP_NOTQUERYI:
1982 case OP_NOTSTAR:
1983 case OP_NOTSTARI:
1984 case OP_NOTUPTO:
1985 case OP_NOTUPTOI:
1986 case OP_PLUS:
1987 case OP_PLUSI:
1988 case OP_POSPLUS:
1989 case OP_POSPLUSI:
1990 case OP_POSQUERY:
1991 case OP_POSQUERYI:
1992 case OP_POSSTAR:
1993 case OP_POSSTARI:
1994 case OP_POSUPTO:
1995 case OP_POSUPTOI:
1996 case OP_QUERY:
1997 case OP_QUERYI:
1998 case OP_REF:
1999 case OP_REFI:
2000 case OP_SBRA:
2001 case OP_SBRAPOS:
2002 case OP_SCBRA:
2003 case OP_SCBRAPOS:
2004 case OP_SCOND:
2005 case OP_SKIPZERO:
2006 case OP_STAR:
2007 case OP_STARI:
2008 case OP_TYPEMINPLUS:
2009 case OP_TYPEMINQUERY:
2010 case OP_TYPEMINSTAR:
2011 case OP_TYPEMINUPTO:
2012 case OP_TYPEPLUS:
2013 case OP_TYPEPOSPLUS:
2014 case OP_TYPEPOSQUERY:
2015 case OP_TYPEPOSSTAR:
2016 case OP_TYPEPOSUPTO:
2017 case OP_TYPEQUERY:
2018 case OP_TYPESTAR:
2019 case OP_TYPEUPTO:
2020 case OP_UPTO:
2021 case OP_UPTOI:
2022 return -1;
2023
2024 /* Catch unrecognized opcodes so that when new ones are added they
2025 are not forgotten, as has happened in the past. */
2026
2027 default:
2028 return -4;
2029 }
2030 }
2031 /* Control never gets here */
2032 }
2033
2034
2035
2036
2037 /*************************************************
2038 * Scan compiled regex for specific bracket *
2039 *************************************************/
2040
2041 /* This little function scans through a compiled pattern until it finds a
2042 capturing bracket with the given number, or, if the number is negative, an
2043 instance of OP_REVERSE for a lookbehind. The function is global in the C sense
2044 so that it can be called from pcre_study() when finding the minimum matching
2045 length.
2046
2047 Arguments:
2048 code points to start of expression
2049 utf TRUE in UTF-8 / UTF-16 / UTF-32 mode
2050 number the required bracket number or negative to find a lookbehind
2051
2052 Returns: pointer to the opcode for the bracket, or NULL if not found
2053 */
2054
2055 const pcre_uchar *
2056 PRIV(find_bracket)(const pcre_uchar *code, BOOL utf, int number)
2057 {
2058 for (;;)
2059 {
2060 register int c = *code;
2061
2062 if (c == OP_END) return NULL;
2063
2064 /* XCLASS is used for classes that cannot be represented just by a bit
2065 map. This includes negated single high-valued characters. The length in
2066 the table is zero; the actual length is stored in the compiled code. */
2067
2068 if (c == OP_XCLASS) code += GET(code, 1);
2069
2070 /* Handle recursion */
2071
2072 else if (c == OP_REVERSE)
2073 {
2074 if (number < 0) return (pcre_uchar *)code;
2075 code += PRIV(OP_lengths)[c];
2076 }
2077
2078 /* Handle capturing bracket */
2079
2080 else if (c == OP_CBRA || c == OP_SCBRA ||
2081 c == OP_CBRAPOS || c == OP_SCBRAPOS)
2082 {
2083 int n = GET2(code, 1+LINK_SIZE);
2084 if (n == number) return (pcre_uchar *)code;
2085 code += PRIV(OP_lengths)[c];
2086 }
2087
2088 /* Otherwise, we can get the item's length from the table, except that for
2089 repeated character types, we have to test for \p and \P, which have an extra
2090 two bytes of parameters, and for MARK/PRUNE/SKIP/THEN with an argument, we
2091 must add in its length. */
2092
2093 else
2094 {
2095 switch(c)
2096 {
2097 case OP_TYPESTAR:
2098 case OP_TYPEMINSTAR:
2099 case OP_TYPEPLUS:
2100 case OP_TYPEMINPLUS:
2101 case OP_TYPEQUERY:
2102 case OP_TYPEMINQUERY:
2103 case OP_TYPEPOSSTAR:
2104 case OP_TYPEPOSPLUS:
2105 case OP_TYPEPOSQUERY:
2106 if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
2107 break;
2108
2109 case OP_TYPEUPTO:
2110 case OP_TYPEMINUPTO:
2111 case OP_TYPEEXACT:
2112 case OP_TYPEPOSUPTO:
2113 if (code[1 + IMM2_SIZE] == OP_PROP || code[1 + IMM2_SIZE] == OP_NOTPROP)
2114 code += 2;
2115 break;
2116
2117 case OP_MARK:
2118 case OP_PRUNE_ARG:
2119 case OP_SKIP_ARG:
2120 code += code[1];
2121 break;
2122
2123 case OP_THEN_ARG:
2124 code += code[1];
2125 break;
2126 }
2127
2128 /* Add in the fixed length from the table */
2129
2130 code += PRIV(OP_lengths)[c];
2131
2132 /* In UTF-8 mode, opcodes that are followed by a character may be followed by
2133 a multi-byte character. The length in the table is a minimum, so we have to
2134 arrange to skip the extra bytes. */
2135
2136 #if defined SUPPORT_UTF && !defined COMPILE_PCRE32
2137 if (utf) switch(c)
2138 {
2139 case OP_CHAR:
2140 case OP_CHARI:
2141 case OP_EXACT:
2142 case OP_EXACTI:
2143 case OP_UPTO:
2144 case OP_UPTOI:
2145 case OP_MINUPTO:
2146 case OP_MINUPTOI:
2147 case OP_POSUPTO:
2148 case OP_POSUPTOI:
2149 case OP_STAR:
2150 case OP_STARI:
2151 case OP_MINSTAR:
2152 case OP_MINSTARI:
2153 case OP_POSSTAR:
2154 case OP_POSSTARI:
2155 case OP_PLUS:
2156 case OP_PLUSI:
2157 case OP_MINPLUS:
2158 case OP_MINPLUSI:
2159 case OP_POSPLUS:
2160 case OP_POSPLUSI:
2161 case OP_QUERY:
2162 case OP_QUERYI:
2163 case OP_MINQUERY:
2164 case OP_MINQUERYI:
2165 case OP_POSQUERY:
2166 case OP_POSQUERYI:
2167 if (HAS_EXTRALEN(code[-1])) code += GET_EXTRALEN(code[-1]);
2168 break;
2169 }
2170 #else
2171 (void)(utf); /* Keep compiler happy by referencing function argument */
2172 #endif
2173 }
2174 }
2175 }
2176
2177
2178
2179 /*************************************************
2180 * Scan compiled regex for recursion reference *
2181 *************************************************/
2182
2183 /* This little function scans through a compiled pattern until it finds an
2184 instance of OP_RECURSE.
2185
2186 Arguments:
2187 code points to start of expression
2188 utf TRUE in UTF-8 / UTF-16 / UTF-32 mode
2189
2190 Returns: pointer to the opcode for OP_RECURSE, or NULL if not found
2191 */
2192
2193 static const pcre_uchar *
2194 find_recurse(const pcre_uchar *code, BOOL utf)
2195 {
2196 for (;;)
2197 {
2198 register int c = *code;
2199 if (c == OP_END) return NULL;
2200 if (c == OP_RECURSE) return code;
2201
2202 /* XCLASS is used for classes that cannot be represented just by a bit
2203 map. This includes negated single high-valued characters. The length in
2204 the table is zero; the actual length is stored in the compiled code. */
2205
2206 if (c == OP_XCLASS) code += GET(code, 1);
2207
2208 /* Otherwise, we can get the item's length from the table, except that for
2209 repeated character types, we have to test for \p and \P, which have an extra
2210 two bytes of parameters, and for MARK/PRUNE/SKIP/THEN with an argument, we
2211 must add in its length. */
2212
2213 else
2214 {
2215 switch(c)
2216 {
2217 case OP_TYPESTAR:
2218 case OP_TYPEMINSTAR:
2219 case OP_TYPEPLUS:
2220 case OP_TYPEMINPLUS:
2221 case OP_TYPEQUERY:
2222 case OP_TYPEMINQUERY:
2223 case OP_TYPEPOSSTAR:
2224 case OP_TYPEPOSPLUS:
2225 case OP_TYPEPOSQUERY:
2226 if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
2227 break;
2228
2229 case OP_TYPEPOSUPTO:
2230 case OP_TYPEUPTO:
2231 case OP_TYPEMINUPTO:
2232 case OP_TYPEEXACT:
2233 if (code[1 + IMM2_SIZE] == OP_PROP || code[1 + IMM2_SIZE] == OP_NOTPROP)
2234 code += 2;
2235 break;
2236
2237 case OP_MARK:
2238 case OP_PRUNE_ARG:
2239 case OP_SKIP_ARG:
2240 code += code[1];
2241 break;
2242
2243 case OP_THEN_ARG:
2244 code += code[1];
2245 break;
2246 }
2247
2248 /* Add in the fixed length from the table */
2249
2250 code += PRIV(OP_lengths)[c];
2251
2252 /* In UTF-8 mode, opcodes that are followed by a character may be followed
2253 by a multi-byte character. The length in the table is a minimum, so we have
2254 to arrange to skip the extra bytes. */
2255
2256 #if defined SUPPORT_UTF && !defined COMPILE_PCRE32
2257 if (utf) switch(c)
2258 {
2259 case OP_CHAR:
2260 case OP_CHARI:
2261 case OP_NOT:
2262 case OP_NOTI:
2263 case OP_EXACT:
2264 case OP_EXACTI:
2265 case OP_NOTEXACT:
2266 case OP_NOTEXACTI:
2267 case OP_UPTO:
2268 case OP_UPTOI:
2269 case OP_NOTUPTO:
2270 case OP_NOTUPTOI:
2271 case OP_MINUPTO:
2272 case OP_MINUPTOI:
2273 case OP_NOTMINUPTO:
2274 case OP_NOTMINUPTOI:
2275 case OP_POSUPTO:
2276 case OP_POSUPTOI:
2277 case OP_NOTPOSUPTO:
2278 case OP_NOTPOSUPTOI:
2279 case OP_STAR:
2280 case OP_STARI:
2281 case OP_NOTSTAR:
2282 case OP_NOTSTARI:
2283 case OP_MINSTAR:
2284 case OP_MINSTARI:
2285 case OP_NOTMINSTAR:
2286 case OP_NOTMINSTARI:
2287 case OP_POSSTAR:
2288 case OP_POSSTARI:
2289 case OP_NOTPOSSTAR:
2290 case OP_NOTPOSSTARI:
2291 case OP_PLUS:
2292 case OP_PLUSI:
2293 case OP_NOTPLUS:
2294 case OP_NOTPLUSI:
2295 case OP_MINPLUS:
2296 case OP_MINPLUSI:
2297 case OP_NOTMINPLUS:
2298 case OP_NOTMINPLUSI:
2299 case OP_POSPLUS:
2300 case OP_POSPLUSI:
2301 case OP_NOTPOSPLUS:
2302 case OP_NOTPOSPLUSI:
2303 case OP_QUERY:
2304 case OP_QUERYI:
2305 case OP_NOTQUERY:
2306 case OP_NOTQUERYI:
2307 case OP_MINQUERY:
2308 case OP_MINQUERYI:
2309 case OP_NOTMINQUERY:
2310 case OP_NOTMINQUERYI:
2311 case OP_POSQUERY:
2312 case OP_POSQUERYI:
2313 case OP_NOTPOSQUERY:
2314 case OP_NOTPOSQUERYI:
2315 if (HAS_EXTRALEN(code[-1])) code += GET_EXTRALEN(code[-1]);
2316 break;
2317 }
2318 #else
2319 (void)(utf); /* Keep compiler happy by referencing function argument */
2320 #endif
2321 }
2322 }
2323 }
2324
2325
2326
2327 /*************************************************
2328 * Scan compiled branch for non-emptiness *
2329 *************************************************/
2330
2331 /* This function scans through a branch of a compiled pattern to see whether it
2332 can match the empty string or not. It is called from could_be_empty()
2333 below and from compile_branch() when checking for an unlimited repeat of a
2334 group that can match nothing. Note that first_significant_code() skips over
2335 backward and negative forward assertions when its final argument is TRUE. If we
2336 hit an unclosed bracket, we return "empty" - this means we've struck an inner
2337 bracket whose current branch will already have been scanned.
2338
2339 Arguments:
2340 code points to start of search
2341 endcode points to where to stop
2342 utf TRUE if in UTF-8 / UTF-16 / UTF-32 mode
2343 cd contains pointers to tables etc.
2344
2345 Returns: TRUE if what is matched could be empty
2346 */
2347
2348 static BOOL
2349 could_be_empty_branch(const pcre_uchar *code, const pcre_uchar *endcode,
2350 BOOL utf, compile_data *cd)
2351 {
2352 register int c;
2353 for (code = first_significant_code(code + PRIV(OP_lengths)[*code], TRUE);
2354 code < endcode;
2355 code = first_significant_code(code + PRIV(OP_lengths)[c], TRUE))
2356 {
2357 const pcre_uchar *ccode;
2358
2359 c = *code;
2360
2361 /* Skip over forward assertions; the other assertions are skipped by
2362 first_significant_code() with a TRUE final argument. */
2363
2364 if (c == OP_ASSERT)
2365 {
2366 do code += GET(code, 1); while (*code == OP_ALT);
2367 c = *code;
2368 continue;
2369 }
2370
2371 /* For a recursion/subroutine call, if its end has been reached, which
2372 implies a backward reference subroutine call, we can scan it. If it's a
2373 forward reference subroutine call, we can't. To detect forward reference
2374 we have to scan up the list that is kept in the workspace. This function is
2375 called only when doing the real compile, not during the pre-compile that
2376 measures the size of the compiled pattern. */
2377
2378 if (c == OP_RECURSE)
2379 {
2380 const pcre_uchar *scode;
2381 BOOL empty_branch;
2382
2383 /* Test for forward reference */
2384
2385 for (scode = cd->start_workspace; scode < cd->hwm; scode += LINK_SIZE)
2386 if (GET(scode, 0) == code + 1 - cd->start_code) return TRUE;
2387
2388 /* Not a forward reference, test for completed backward reference */
2389
2390 empty_branch = FALSE;
2391 scode = cd->start_code + GET(code, 1);
2392 if (GET(scode, 1) == 0) return TRUE; /* Unclosed */
2393
2394 /* Completed backwards reference */
2395
2396 do
2397 {
2398 if (could_be_empty_branch(scode, endcode, utf, cd))
2399 {
2400 empty_branch = TRUE;
2401 break;
2402 }
2403 scode += GET(scode, 1);
2404 }
2405 while (*scode == OP_ALT);
2406
2407 if (!empty_branch) return FALSE; /* All branches are non-empty */
2408 continue;
2409 }
2410
2411 /* Groups with zero repeats can of course be empty; skip them. */
2412
2413 if (c == OP_BRAZERO || c == OP_BRAMINZERO || c == OP_SKIPZERO ||
2414 c == OP_BRAPOSZERO)
2415 {
2416 code += PRIV(OP_lengths)[c];
2417 do code += GET(code, 1); while (*code == OP_ALT);
2418 c = *code;
2419 continue;
2420 }
2421
2422 /* A nested group that is already marked as "could be empty" can just be
2423 skipped. */
2424
2425 if (c == OP_SBRA || c == OP_SBRAPOS ||
2426 c == OP_SCBRA || c == OP_SCBRAPOS)
2427 {
2428 do code += GET(code, 1); while (*code == OP_ALT);
2429 c = *code;
2430 continue;
2431 }
2432
2433 /* For other groups, scan the branches. */
2434
2435 if (c == OP_BRA || c == OP_BRAPOS ||
2436 c == OP_CBRA || c == OP_CBRAPOS ||
2437 c == OP_ONCE || c == OP_ONCE_NC ||
2438 c == OP_COND)
2439 {
2440 BOOL empty_branch;
2441 if (GET(code, 1) == 0) return TRUE; /* Hit unclosed bracket */
2442
2443 /* If a conditional group has only one branch, there is a second, implied,
2444 empty branch, so just skip over the conditional, because it could be empty.
2445 Otherwise, scan the individual branches of the group. */
2446
2447 if (c == OP_COND && code[GET(code, 1)] != OP_ALT)
2448 code += GET(code, 1);
2449 else
2450 {
2451 empty_branch = FALSE;
2452 do
2453 {
2454 if (!empty_branch && could_be_empty_branch(code, endcode, utf, cd))
2455 empty_branch = TRUE;
2456 code += GET(code, 1);
2457 }
2458 while (*code == OP_ALT);
2459 if (!empty_branch) return FALSE; /* All branches are non-empty */
2460 }
2461
2462 c = *code;
2463 continue;
2464 }
2465
2466 /* Handle the other opcodes */
2467
2468 switch (c)
2469 {
2470 /* Check for quantifiers after a class. XCLASS is used for classes that
2471 cannot be represented just by a bit map. This includes negated single
2472 high-valued characters. The length in PRIV(OP_lengths)[] is zero; the
2473 actual length is stored in the compiled code, so we must update "code"
2474 here. */
2475
2476 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
2477 case OP_XCLASS:
2478 ccode = code += GET(code, 1);
2479 goto CHECK_CLASS_REPEAT;
2480 #endif
2481
2482 case OP_CLASS:
2483 case OP_NCLASS:
2484 ccode = code + PRIV(OP_lengths)[OP_CLASS];
2485
2486 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
2487 CHECK_CLASS_REPEAT:
2488 #endif
2489
2490 switch (*ccode)
2491 {
2492 case OP_CRSTAR: /* These could be empty; continue */
2493 case OP_CRMINSTAR:
2494 case OP_CRQUERY:
2495 case OP_CRMINQUERY:
2496 break;
2497
2498 default: /* Non-repeat => class must match */
2499 case OP_CRPLUS: /* These repeats aren't empty */
2500 case OP_CRMINPLUS:
2501 return FALSE;
2502
2503 case OP_CRRANGE:
2504 case OP_CRMINRANGE:
2505 if (GET2(ccode, 1) > 0) return FALSE; /* Minimum > 0 */
2506 break;
2507 }
2508 break;
2509
2510 /* Opcodes that must match a character */
2511
2512 case OP_PROP:
2513 case OP_NOTPROP:
2514 case OP_EXTUNI:
2515 case OP_NOT_DIGIT:
2516 case OP_DIGIT:
2517 case OP_NOT_WHITESPACE:
2518 case OP_WHITESPACE:
2519 case OP_NOT_WORDCHAR:
2520 case OP_WORDCHAR:
2521 case OP_ANY:
2522 case OP_ALLANY:
2523 case OP_ANYBYTE:
2524 case OP_CHAR:
2525 case OP_CHARI:
2526 case OP_NOT:
2527 case OP_NOTI:
2528 case OP_PLUS:
2529 case OP_MINPLUS:
2530 case OP_POSPLUS:
2531 case OP_EXACT:
2532 case OP_NOTPLUS:
2533 case OP_NOTMINPLUS:
2534 case OP_NOTPOSPLUS:
2535 case OP_NOTEXACT:
2536 case OP_TYPEPLUS:
2537 case OP_TYPEMINPLUS:
2538 case OP_TYPEPOSPLUS:
2539 case OP_TYPEEXACT:
2540 return FALSE;
2541
2542 /* These are going to continue, as they may be empty, but we have to
2543 fudge the length for the \p and \P cases. */
2544
2545 case OP_TYPESTAR:
2546 case OP_TYPEMINSTAR:
2547 case OP_TYPEPOSSTAR:
2548 case OP_TYPEQUERY:
2549 case OP_TYPEMINQUERY:
2550 case OP_TYPEPOSQUERY:
2551 if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
2552 break;
2553
2554 /* Same for these */
2555
2556 case OP_TYPEUPTO:
2557 case OP_TYPEMINUPTO:
2558 case OP_TYPEPOSUPTO:
2559 if (code[1 + IMM2_SIZE] == OP_PROP || code[1 + IMM2_SIZE] == OP_NOTPROP)
2560 code += 2;
2561 break;
2562
2563 /* End of branch */
2564
2565 case OP_KET:
2566 case OP_KETRMAX:
2567 case OP_KETRMIN:
2568 case OP_KETRPOS:
2569 case OP_ALT:
2570 return TRUE;
2571
2572 /* In UTF-8 mode, STAR, MINSTAR, POSSTAR, QUERY, MINQUERY, POSQUERY, UPTO,
2573 MINUPTO, and POSUPTO may be followed by a multibyte character */
2574
2575 #if defined SUPPORT_UTF && !defined COMPILE_PCRE32
2576 case OP_STAR:
2577 case OP_STARI:
2578 case OP_MINSTAR:
2579 case OP_MINSTARI:
2580 case OP_POSSTAR:
2581 case OP_POSSTARI:
2582 case OP_QUERY:
2583 case OP_QUERYI:
2584 case OP_MINQUERY:
2585 case OP_MINQUERYI:
2586 case OP_POSQUERY:
2587 case OP_POSQUERYI:
2588 if (utf && HAS_EXTRALEN(code[1])) code += GET_EXTRALEN(code[1]);
2589 break;
2590
2591 case OP_UPTO:
2592 case OP_UPTOI:
2593 case OP_MINUPTO:
2594 case OP_MINUPTOI:
2595 case OP_POSUPTO:
2596 case OP_POSUPTOI:
2597 if (utf && HAS_EXTRALEN(code[1 + IMM2_SIZE])) code += GET_EXTRALEN(code[1 + IMM2_SIZE]);
2598 break;
2599 #endif
2600
2601 /* MARK, and PRUNE/SKIP/THEN with an argument must skip over the argument
2602 string. */
2603
2604 case OP_MARK:
2605 case OP_PRUNE_ARG:
2606 case OP_SKIP_ARG:
2607 code += code[1];
2608 break;
2609
2610 case OP_THEN_ARG:
2611 code += code[1];
2612 break;
2613
2614 /* None of the remaining opcodes are required to match a character. */
2615
2616 default:
2617 break;
2618 }
2619 }
2620
2621 return TRUE;
2622 }
2623
2624
2625
2626 /*************************************************
2627 * Scan compiled regex for non-emptiness *
2628 *************************************************/
2629
2630 /* This function is called to check for left recursive calls. We want to check
2631 the current branch of the current pattern to see if it could match the empty
2632 string. If it could, we must look outwards for branches at other levels,
2633 stopping when we pass beyond the bracket which is the subject of the recursion.
2634 This function is called only during the real compile, not during the
2635 pre-compile.
2636
2637 Arguments:
2638 code points to start of the recursion
2639 endcode points to where to stop (current RECURSE item)
2640 bcptr points to the chain of current (unclosed) branch starts
2641 utf TRUE if in UTF-8 / UTF-16 / UTF-32 mode
2642 cd pointers to tables etc
2643
2644 Returns: TRUE if what is matched could be empty
2645 */
2646
2647 static BOOL
2648 could_be_empty(const pcre_uchar *code, const pcre_uchar *endcode,
2649 branch_chain *bcptr, BOOL utf, compile_data *cd)
2650 {
2651 while (bcptr != NULL && bcptr->current_branch >= code)
2652 {
2653 if (!could_be_empty_branch(bcptr->current_branch, endcode, utf, cd))
2654 return FALSE;
2655 bcptr = bcptr->outer;
2656 }
2657 return TRUE;
2658 }
2659
2660
2661
2662 /*************************************************
2663 * Check for POSIX class syntax *
2664 *************************************************/
2665
2666 /* This function is called when the sequence "[:" or "[." or "[=" is
2667 encountered in a character class. It checks whether this is followed by a
2668 sequence of characters terminated by a matching ":]" or ".]" or "=]". If we
2669 reach an unescaped ']' without the special preceding character, return FALSE.
2670
2671 Originally, this function only recognized a sequence of letters between the
2672 terminators, but it seems that Perl recognizes any sequence of characters,
2673 though of course unknown POSIX names are subsequently rejected. Perl gives an
2674 "Unknown POSIX class" error for [:f\oo:] for example, where previously PCRE
2675 didn't consider this to be a POSIX class. Likewise for [:1234:].
2676
2677 The problem in trying to be exactly like Perl is in the handling of escapes. We
2678 have to be sure that [abc[:x\]pqr] is *not* treated as containing a POSIX
2679 class, but [abc[:x\]pqr:]] is (so that an error can be generated). The code
2680 below handles the special case of \], but does not try to do any other escape
2681 processing. This makes it different from Perl for cases such as [:l\ower:]
2682 where Perl recognizes it as the POSIX class "lower" but PCRE does not recognize
2683 "l\ower". This is a lesser evil that not diagnosing bad classes when Perl does,
2684 I think.
2685
2686 A user pointed out that PCRE was rejecting [:a[:digit:]] whereas Perl was not.
2687 It seems that the appearance of a nested POSIX class supersedes an apparent
2688 external class. For example, [:a[:digit:]b:] matches "a", "b", ":", or
2689 a digit.
2690
2691 In Perl, unescaped square brackets may also appear as part of class names. For
2692 example, [:a[:abc]b:] gives unknown POSIX class "[:abc]b:]". However, for
2693 [:a[:abc]b][b:] it gives unknown POSIX class "[:abc]b][b:]", which does not
2694 seem right at all. PCRE does not allow closing square brackets in POSIX class
2695 names.
2696
2697 Arguments:
2698 ptr pointer to the initial [
2699 endptr where to return the end pointer
2700
2701 Returns: TRUE or FALSE
2702 */
2703
2704 static BOOL
2705 check_posix_syntax(const pcre_uchar *ptr, const pcre_uchar **endptr)
2706 {
2707 int terminator; /* Don't combine these lines; the Solaris cc */
2708 terminator = *(++ptr); /* compiler warns about "non-constant" initializer. */
2709 for (++ptr; *ptr != 0; ptr++)
2710 {
2711 if (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET)
2712 ptr++;
2713 else if (*ptr == CHAR_RIGHT_SQUARE_BRACKET) return FALSE;
2714 else
2715 {
2716 if (*ptr == terminator && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET)
2717 {
2718 *endptr = ptr;
2719 return TRUE;
2720 }
2721 if (*ptr == CHAR_LEFT_SQUARE_BRACKET &&
2722 (ptr[1] == CHAR_COLON || ptr[1] == CHAR_DOT ||
2723 ptr[1] == CHAR_EQUALS_SIGN) &&
2724 check_posix_syntax(ptr, endptr))
2725 return FALSE;
2726 }
2727 }
2728 return FALSE;
2729 }
2730
2731
2732
2733
2734 /*************************************************
2735 * Check POSIX class name *
2736 *************************************************/
2737
2738 /* This function is called to check the name given in a POSIX-style class entry
2739 such as [:alnum:].
2740
2741 Arguments:
2742 ptr points to the first letter
2743 len the length of the name
2744
2745 Returns: a value representing the name, or -1 if unknown
2746 */
2747
2748 static int
2749 check_posix_name(const pcre_uchar *ptr, int len)
2750 {
2751 const char *pn = posix_names;
2752 register int yield = 0;
2753 while (posix_name_lengths[yield] != 0)
2754 {
2755 if (len == posix_name_lengths[yield] &&
2756 STRNCMP_UC_C8(ptr, pn, len) == 0) return yield;
2757 pn += posix_name_lengths[yield] + 1;
2758 yield++;
2759 }
2760 return -1;
2761 }
2762
2763
2764 /*************************************************
2765 * Adjust OP_RECURSE items in repeated group *
2766 *************************************************/
2767
2768 /* OP_RECURSE items contain an offset from the start of the regex to the group
2769 that is referenced. This means that groups can be replicated for fixed
2770 repetition simply by copying (because the recursion is allowed to refer to
2771 earlier groups that are outside the current group). However, when a group is
2772 optional (i.e. the minimum quantifier is zero), OP_BRAZERO or OP_SKIPZERO is
2773 inserted before it, after it has been compiled. This means that any OP_RECURSE
2774 items within it that refer to the group itself or any contained groups have to
2775 have their offsets adjusted. That one of the jobs of this function. Before it
2776 is called, the partially compiled regex must be temporarily terminated with
2777 OP_END.
2778
2779 This function has been extended with the possibility of forward references for
2780 recursions and subroutine calls. It must also check the list of such references
2781 for the group we are dealing with. If it finds that one of the recursions in
2782 the current group is on this list, it adjusts the offset in the list, not the
2783 value in the reference (which is a group number).
2784
2785 Arguments:
2786 group points to the start of the group
2787 adjust the amount by which the group is to be moved
2788 utf TRUE in UTF-8 / UTF-16 / UTF-32 mode
2789 cd contains pointers to tables etc.
2790 save_hwm the hwm forward reference pointer at the start of the group
2791
2792 Returns: nothing
2793 */
2794
2795 static void
2796 adjust_recurse(pcre_uchar *group, int adjust, BOOL utf, compile_data *cd,
2797 pcre_uchar *save_hwm)
2798 {
2799 pcre_uchar *ptr = group;
2800
2801 while ((ptr = (pcre_uchar *)find_recurse(ptr, utf)) != NULL)
2802 {
2803 int offset;
2804 pcre_uchar *hc;
2805
2806 /* See if this recursion is on the forward reference list. If so, adjust the
2807 reference. */
2808
2809 for (hc = save_hwm; hc < cd->hwm; hc += LINK_SIZE)
2810 {
2811 offset = GET(hc, 0);
2812 if (cd->start_code + offset == ptr + 1)
2813 {
2814 PUT(hc, 0, offset + adjust);
2815 break;
2816 }
2817 }
2818
2819 /* Otherwise, adjust the recursion offset if it's after the start of this
2820 group. */
2821
2822 if (hc >= cd->hwm)
2823 {
2824 offset = GET(ptr, 1);
2825 if (cd->start_code + offset >= group) PUT(ptr, 1, offset + adjust);
2826 }
2827
2828 ptr += 1 + LINK_SIZE;
2829 }
2830 }
2831
2832
2833
2834 /*************************************************
2835 * Insert an automatic callout point *
2836 *************************************************/
2837
2838 /* This function is called when the PCRE_AUTO_CALLOUT option is set, to insert
2839 callout points before each pattern item.
2840
2841 Arguments:
2842 code current code pointer
2843 ptr current pattern pointer
2844 cd pointers to tables etc
2845
2846 Returns: new code pointer
2847 */
2848
2849 static pcre_uchar *
2850 auto_callout(pcre_uchar *code, const pcre_uchar *ptr, compile_data *cd)
2851 {
2852 *code++ = OP_CALLOUT;
2853 *code++ = 255;
2854 PUT(code, 0, (int)(ptr - cd->start_pattern)); /* Pattern offset */
2855 PUT(code, LINK_SIZE, 0); /* Default length */
2856 return code + 2 * LINK_SIZE;
2857 }
2858
2859
2860
2861 /*************************************************
2862 * Complete a callout item *
2863 *************************************************/
2864
2865 /* A callout item contains the length of the next item in the pattern, which
2866 we can't fill in till after we have reached the relevant point. This is used
2867 for both automatic and manual callouts.
2868
2869 Arguments:
2870 previous_callout points to previous callout item
2871 ptr current pattern pointer
2872 cd pointers to tables etc
2873
2874 Returns: nothing
2875 */
2876
2877 static void
2878 complete_callout(pcre_uchar *previous_callout, const pcre_uchar *ptr, compile_data *cd)
2879 {
2880 int length = (int)(ptr - cd->start_pattern - GET(previous_callout, 2));
2881 PUT(previous_callout, 2 + LINK_SIZE, length);
2882 }
2883
2884
2885
2886 #ifdef SUPPORT_UCP
2887 /*************************************************
2888 * Get othercase range *
2889 *************************************************/
2890
2891 /* This function is passed the start and end of a class range, in UTF-8 mode
2892 with UCP support. It searches up the characters, looking for ranges of
2893 characters in the "other" case. Each call returns the next one, updating the
2894 start address. A character with multiple other cases is returned on its own
2895 with a special return value.
2896
2897 Arguments:
2898 cptr points to starting character value; updated
2899 d end value
2900 ocptr where to put start of othercase range
2901 odptr where to put end of othercase range
2902
2903 Yield: -1 when no more
2904 0 when a range is returned
2905 >0 the CASESET offset for char with multiple other cases
2906 in this case, ocptr contains the original
2907 */
2908
2909 static int
2910 get_othercase_range(pcre_uint32 *cptr, pcre_uint32 d, pcre_uint32 *ocptr,
2911 pcre_uint32 *odptr)
2912 {
2913 pcre_uint32 c, othercase, next;
2914 int co;
2915
2916 /* Find the first character that has an other case. If it has multiple other
2917 cases, return its case offset value. */
2918
2919 for (c = *cptr; c <= d; c++)
2920 {
2921 if ((co = UCD_CASESET(c)) != 0)
2922 {
2923 *ocptr = c++; /* Character that has the set */
2924 *cptr = c; /* Rest of input range */
2925 return co;
2926 }
2927 if ((othercase = UCD_OTHERCASE(c)) != c) break;
2928 }
2929
2930 if (c > d) return -1; /* Reached end of range */
2931
2932 *ocptr = othercase;
2933 next = othercase + 1;
2934
2935 for (++c; c <= d; c++)
2936 {
2937 if (UCD_OTHERCASE(c) != next) break;
2938 next++;
2939 }
2940
2941 *odptr = next - 1; /* End of othercase range */
2942 *cptr = c; /* Rest of input range */
2943 return 0;
2944 }
2945
2946
2947
2948 /*************************************************
2949 * Check a character and a property *
2950 *************************************************/
2951
2952 /* This function is called by check_auto_possessive() when a property item
2953 is adjacent to a fixed character.
2954
2955 Arguments:
2956 c the character
2957 ptype the property type
2958 pdata the data for the type
2959 negated TRUE if it's a negated property (\P or \p{^)
2960
2961 Returns: TRUE if auto-possessifying is OK
2962 */
2963
2964 static BOOL
2965 check_char_prop(pcre_uint32 c, int ptype, int pdata, BOOL negated)
2966 {
2967 #ifdef SUPPORT_UCP
2968 const pcre_uint32 *p;
2969 #endif
2970
2971 const ucd_record *prop = GET_UCD(c);
2972
2973 switch(ptype)
2974 {
2975 case PT_LAMP:
2976 return (prop->chartype == ucp_Lu ||
2977 prop->chartype == ucp_Ll ||
2978 prop->chartype == ucp_Lt) == negated;
2979
2980 case PT_GC:
2981 return (pdata == PRIV(ucp_gentype)[prop->chartype]) == negated;
2982
2983 case PT_PC:
2984 return (pdata == prop->chartype) == negated;
2985
2986 case PT_SC:
2987 return (pdata == prop->script) == negated;
2988
2989 /* These are specials */
2990
2991 case PT_ALNUM:
2992 return (PRIV(ucp_gentype)[prop->chartype] == ucp_L ||
2993 PRIV(ucp_gentype)[prop->chartype] == ucp_N) == negated;
2994
2995 case PT_SPACE: /* Perl space */
2996 return (PRIV(ucp_gentype)[prop->chartype] == ucp_Z ||
2997 c == CHAR_HT || c == CHAR_NL || c == CHAR_FF || c == CHAR_CR)
2998 == negated;
2999
3000 case PT_PXSPACE: /* POSIX space */
3001 return (PRIV(ucp_gentype)[prop->chartype] == ucp_Z ||
3002 c == CHAR_HT || c == CHAR_NL || c == CHAR_VT ||
3003 c == CHAR_FF || c == CHAR_CR)
3004 == negated;
3005
3006 case PT_WORD:
3007 return (PRIV(ucp_gentype)[prop->chartype] == ucp_L ||
3008 PRIV(ucp_gentype)[prop->chartype] == ucp_N ||
3009 c == CHAR_UNDERSCORE) == negated;
3010
3011 #ifdef SUPPORT_UCP
3012 case PT_CLIST:
3013 p = PRIV(ucd_caseless_sets) + prop->caseset;
3014 for (;;)
3015 {
3016 if ((unsigned int)c < *p) return !negated;
3017 if ((unsigned int)c == *p++) return negated;
3018 }
3019 break; /* Control never reaches here */
3020 #endif
3021 }
3022
3023 return FALSE;
3024 }
3025 #endif /* SUPPORT_UCP */
3026
3027
3028
3029 /*************************************************
3030 * Check if auto-possessifying is possible *
3031 *************************************************/
3032
3033 /* This function is called for unlimited repeats of certain items, to see
3034 whether the next thing could possibly match the repeated item. If not, it makes
3035 sense to automatically possessify the repeated item.
3036
3037 Arguments:
3038 previous pointer to the repeated opcode
3039 utf TRUE in UTF-8 / UTF-16 / UTF-32 mode
3040 ptr next character in pattern
3041 options options bits
3042 cd contains pointers to tables etc.
3043
3044 Returns: TRUE if possessifying is wanted
3045 */
3046
3047 static BOOL
3048 check_auto_possessive(const pcre_uchar *previous, BOOL utf,
3049 const pcre_uchar *ptr, int options, compile_data *cd)
3050 {
3051 pcre_uint32 c = NOTACHAR;
3052 pcre_uint32 next;
3053 int escape;
3054 int op_code = *previous++;
3055
3056 /* Skip whitespace and comments in extended mode */
3057
3058 if ((options & PCRE_EXTENDED) != 0)
3059 {
3060 for (;;)
3061 {
3062 while (MAX_255(*ptr) && (cd->ctypes[*ptr] & ctype_space) != 0) ptr++;
3063 if (*ptr == CHAR_NUMBER_SIGN)
3064 {
3065 ptr++;
3066 while (*ptr != 0)
3067 {
3068 if (IS_NEWLINE(ptr)) { ptr += cd->nllen; break; }
3069 ptr++;
3070 #ifdef SUPPORT_UTF
3071 if (utf) FORWARDCHAR(ptr);
3072 #endif
3073 }
3074 }
3075 else break;
3076 }
3077 }
3078
3079 /* If the next item is one that we can handle, get its value. A non-negative
3080 value is a character, a negative value is an escape value. */
3081
3082 if (*ptr == CHAR_BACKSLASH)
3083 {
3084 int temperrorcode = 0;
3085 escape = check_escape(&ptr, &next, &temperrorcode, cd->bracount, options, FALSE);
3086 if (temperrorcode != 0) return FALSE;
3087 ptr++; /* Point after the escape sequence */
3088 }
3089 else if (!MAX_255(*ptr) || (cd->ctypes[*ptr] & ctype_meta) == 0)
3090 {
3091 escape = 0;
3092 #ifdef SUPPORT_UTF
3093 if (utf) { GETCHARINC(next, ptr); } else
3094 #endif
3095 next = *ptr++;
3096 }
3097 else return FALSE;
3098
3099 /* Skip whitespace and comments in extended mode */
3100
3101 if ((options & PCRE_EXTENDED) != 0)
3102 {
3103 for (;;)
3104 {
3105 while (MAX_255(*ptr) && (cd->ctypes[*ptr] & ctype_space) != 0) ptr++;
3106 if (*ptr == CHAR_NUMBER_SIGN)
3107 {
3108 ptr++;
3109 while (*ptr != 0)
3110 {
3111 if (IS_NEWLINE(ptr)) { ptr += cd->nllen; break; }
3112 ptr++;
3113 #ifdef SUPPORT_UTF
3114 if (utf) FORWARDCHAR(ptr);
3115 #endif
3116 }
3117 }
3118 else break;
3119 }
3120 }
3121
3122 /* If the next thing is itself optional, we have to give up. */
3123
3124 if (*ptr == CHAR_ASTERISK || *ptr == CHAR_QUESTION_MARK ||
3125 STRNCMP_UC_C8(ptr, STR_LEFT_CURLY_BRACKET STR_0 STR_COMMA, 3) == 0)
3126 return FALSE;
3127
3128 /* If the previous item is a character, get its value. */
3129
3130 if (op_code == OP_CHAR || op_code == OP_CHARI ||
3131 op_code == OP_NOT || op_code == OP_NOTI)
3132 //if (escape == 0) switch(op_code)
3133 {
3134 #ifdef SUPPORT_UTF
3135 GETCHARTEST(c, previous);
3136 #else
3137 c = *previous;
3138 #endif
3139 }
3140
3141 /* Now compare the next item with the previous opcode. First, handle cases when
3142 the next item is a character. */
3143
3144 if (escape == 0)
3145 {
3146 /* For a caseless UTF match, the next character may have more than one other
3147 case, which maps to the special PT_CLIST property. Check this first. */
3148
3149 #ifdef SUPPORT_UCP
3150 if (utf && c != NOTACHAR && (options & PCRE_CASELESS) != 0)
3151 {
3152 int ocs = UCD_CASESET(next);
3153 if (ocs > 0) return check_char_prop(c, PT_CLIST, ocs, op_code >= OP_NOT);
3154 }
3155 #endif
3156
3157 switch(op_code)
3158 {
3159 case OP_CHAR:
3160 return c != next;
3161
3162 /* For CHARI (caseless character) we must check the other case. If we have
3163 Unicode property support, we can use it to test the other case of
3164 high-valued characters. We know that next can have only one other case,
3165 because multi-other-case characters are dealt with above. */
3166
3167 case OP_CHARI:
3168 if (c == next) return FALSE;
3169 #ifdef SUPPORT_UTF
3170 if (utf)
3171 {
3172 pcre_uint32 othercase;
3173 if (next < 128) othercase = cd->fcc[next]; else
3174 #ifdef SUPPORT_UCP
3175 othercase = UCD_OTHERCASE(next);
3176 #else
3177 othercase = NOTACHAR;
3178 #endif
3179 return c != othercase;
3180 }
3181 else
3182 #endif /* SUPPORT_UTF */
3183 return (c != TABLE_GET(next, cd->fcc, next)); /* Not UTF */
3184
3185 case OP_NOT:
3186 return c == next;
3187
3188 case OP_NOTI:
3189 if (c == next) return TRUE;
3190 #ifdef SUPPORT_UTF
3191 if (utf)
3192 {
3193 pcre_uint32 othercase;
3194 if (next < 128) othercase = cd->fcc[next]; else
3195 #ifdef SUPPORT_UCP
3196 othercase = UCD_OTHERCASE(next);
3197 #else
3198 othercase = NOTACHAR;
3199 #endif
3200 return c == othercase;
3201 }
3202 else
3203 #endif /* SUPPORT_UTF */
3204 return (c == TABLE_GET(next, cd->fcc, next)); /* Not UTF */
3205
3206 /* Note that OP_DIGIT etc. are generated only when PCRE_UCP is *not* set.
3207 When it is set, \d etc. are converted into OP_(NOT_)PROP codes. */
3208
3209 case OP_DIGIT:
3210 return next > 255 || (cd->ctypes[next] & ctype_digit) == 0;
3211
3212 case OP_NOT_DIGIT:
3213 return next <= 255 && (cd->ctypes[next] & ctype_digit) != 0;
3214
3215 case OP_WHITESPACE:
3216 return next > 255 || (cd->ctypes[next] & ctype_space) == 0;
3217
3218 case OP_NOT_WHITESPACE:
3219 return next <= 255 && (cd->ctypes[next] & ctype_space) != 0;
3220
3221 case OP_WORDCHAR:
3222 return next > 255 || (cd->ctypes[next] & ctype_word) == 0;
3223
3224 case OP_NOT_WORDCHAR:
3225 return next <= 255 && (cd->ctypes[next] & ctype_word) != 0;
3226
3227 case OP_HSPACE:
3228 case OP_NOT_HSPACE:
3229 switch(next)
3230 {
3231 HSPACE_CASES:
3232 return op_code == OP_NOT_HSPACE;
3233
3234 default:
3235 return op_code != OP_NOT_HSPACE;
3236 }
3237
3238 case OP_ANYNL:
3239 case OP_VSPACE:
3240 case OP_NOT_VSPACE:
3241 switch(next)
3242 {
3243 VSPACE_CASES:
3244 return op_code == OP_NOT_VSPACE;
3245
3246 default:
3247 return op_code != OP_NOT_VSPACE;
3248 }
3249
3250 #ifdef SUPPORT_UCP
3251 case OP_PROP:
3252 return check_char_prop(next, previous[0], previous[1], FALSE);
3253
3254 case OP_NOTPROP:
3255 return check_char_prop(next, previous[0], previous[1], TRUE);
3256 #endif
3257
3258 default:
3259 return FALSE;
3260 }
3261 }
3262
3263 /* Handle the case when the next item is \d, \s, etc. Note that when PCRE_UCP
3264 is set, \d turns into ESC_du rather than ESC_d, etc., so ESC_d etc. are
3265 generated only when PCRE_UCP is *not* set, that is, when only ASCII
3266 characteristics are recognized. Similarly, the opcodes OP_DIGIT etc. are
3267 replaced by OP_PROP codes when PCRE_UCP is set. */
3268
3269 switch(op_code)
3270 {
3271 case OP_CHAR:
3272 case OP_CHARI:
3273 switch(escape)
3274 {
3275 case ESC_d:
3276 return c > 255 || (cd->ctypes[c] & ctype_digit) == 0;
3277
3278 case ESC_D:
3279 return c <= 255 && (cd->ctypes[c] & ctype_digit) != 0;
3280
3281 case ESC_s:
3282 return c > 255 || (cd->ctypes[c] & ctype_space) == 0;
3283
3284 case ESC_S:
3285 return c <= 255 && (cd->ctypes[c] & ctype_space) != 0;
3286
3287 case ESC_w:
3288 return c > 255 || (cd->ctypes[c] & ctype_word) == 0;
3289
3290 case ESC_W:
3291 return c <= 255 && (cd->ctypes[c] & ctype_word) != 0;
3292
3293 case ESC_h:
3294 case ESC_H:
3295 switch(c)
3296 {
3297 HSPACE_CASES:
3298 return escape != ESC_h;
3299
3300 default:
3301 return escape == ESC_h;
3302 }
3303
3304 case ESC_v:
3305 case ESC_V:
3306 switch(c)
3307 {
3308 VSPACE_CASES:
3309 return escape != ESC_v;
3310
3311 default:
3312 return escape == ESC_v;
3313 }
3314
3315 /* When PCRE_UCP is set, these values get generated for \d etc. Find
3316 their substitutions and process them. The result will always be either
3317 ESC_p or ESC_P. Then fall through to process those values. */
3318
3319 #ifdef SUPPORT_UCP
3320 case ESC_du:
3321 case ESC_DU:
3322 case ESC_wu:
3323 case ESC_WU:
3324 case ESC_su:
3325 case ESC_SU:
3326 {
3327 int temperrorcode = 0;
3328 ptr = substitutes[escape - ESC_DU];
3329 escape = check_escape(&ptr, &next, &temperrorcode, 0, options, FALSE);
3330 if (temperrorcode != 0) return FALSE;
3331 ptr++; /* For compatibility */
3332 }
3333 /* Fall through */
3334
3335 case ESC_p:
3336 case ESC_P:
3337 {
3338 int ptype, pdata, errorcodeptr;
3339 BOOL negated;
3340
3341 ptr--; /* Make ptr point at the p or P */
3342 ptype = get_ucp(&ptr, &negated, &pdata, &errorcodeptr);
3343 if (ptype < 0) return FALSE;
3344 ptr++; /* Point past the final curly ket */
3345
3346 /* If the property item is optional, we have to give up. (When generated
3347 from \d etc by PCRE_UCP, this test will have been applied much earlier,
3348 to the original \d etc. At this point, ptr will point to a zero byte. */
3349
3350 if (*ptr == CHAR_ASTERISK || *ptr == CHAR_QUESTION_MARK ||
3351 STRNCMP_UC_C8(ptr, STR_LEFT_CURLY_BRACKET STR_0 STR_COMMA, 3) == 0)
3352 return FALSE;
3353
3354 /* Do the property check. */
3355
3356 return check_char_prop(c, ptype, pdata, (escape == ESC_P) != negated);
3357 }
3358 #endif
3359
3360 default:
3361 return FALSE;
3362 }
3363
3364 /* In principle, support for Unicode properties should be integrated here as
3365 well. It means re-organizing the above code so as to get hold of the property
3366 values before switching on the op-code. However, I wonder how many patterns
3367 combine ASCII \d etc with Unicode properties? (Note that if PCRE_UCP is set,
3368 these op-codes are never generated.) */
3369
3370 case OP_DIGIT:
3371 return escape == ESC_D || escape == ESC_s || escape == ESC_W ||
3372 escape == ESC_h || escape == ESC_v || escape == ESC_R;
3373
3374 case OP_NOT_DIGIT:
3375 return escape == ESC_d;
3376
3377 case OP_WHITESPACE:
3378 return escape == ESC_S || escape == ESC_d || escape == ESC_w;
3379
3380 case OP_NOT_WHITESPACE:
3381 return escape == ESC_s || escape == ESC_h || escape == ESC_v || escape == ESC_R;
3382
3383 case OP_HSPACE:
3384 return escape == ESC_S || escape == ESC_H || escape == ESC_d ||
3385 escape == ESC_w || escape == ESC_v || escape == ESC_R;
3386
3387 case OP_NOT_HSPACE:
3388 return escape == ESC_h;
3389
3390 /* Can't have \S in here because VT matches \S (Perl anomaly) */
3391 case OP_ANYNL:
3392 case OP_VSPACE:
3393 return escape == ESC_V || escape == ESC_d || escape == ESC_w;
3394
3395 case OP_NOT_VSPACE:
3396 return escape == ESC_v || escape == ESC_R;
3397
3398 case OP_WORDCHAR:
3399 return escape == ESC_W || escape == ESC_s || escape == ESC_h ||
3400 escape == ESC_v || escape == ESC_R;
3401
3402 case OP_NOT_WORDCHAR:
3403 return escape == ESC_w || escape == ESC_d;
3404
3405 default:
3406 return FALSE;
3407 }
3408
3409 /* Control does not reach here */
3410 }
3411
3412
3413
3414 /*************************************************
3415 * Add a character or range to a class *
3416 *************************************************/
3417
3418 /* This function packages up the logic of adding a character or range of
3419 characters to a class. The character values in the arguments will be within the
3420 valid values for the current mode (8-bit, 16-bit, UTF, etc). This function is
3421 mutually recursive with the function immediately below.
3422
3423 Arguments:
3424 classbits the bit map for characters < 256
3425 uchardptr points to the pointer for extra data
3426 options the options word
3427 cd contains pointers to tables etc.
3428 start start of range character
3429 end end of range character
3430
3431 Returns: the number of < 256 characters added
3432 the pointer to extra data is updated
3433 */
3434
3435 static int
3436 add_to_class(pcre_uint8 *classbits, pcre_uchar **uchardptr, int options,
3437 compile_data *cd, pcre_uint32 start, pcre_uint32 end)
3438 {
3439 pcre_uint32 c;
3440 int n8 = 0;
3441
3442 /* If caseless matching is required, scan the range and process alternate
3443 cases. In Unicode, there are 8-bit characters that have alternate cases that
3444 are greater than 255 and vice-versa. Sometimes we can just extend the original
3445 range. */
3446
3447 if ((options & PCRE_CASELESS) != 0)
3448 {
3449 #ifdef SUPPORT_UCP
3450 if ((options & PCRE_UTF8) != 0)
3451 {
3452 int rc;
3453 pcre_uint32 oc, od;
3454
3455 options &= ~PCRE_CASELESS; /* Remove for recursive calls */
3456 c = start;
3457
3458 while ((rc = get_othercase_range(&c, end, &oc, &od)) >= 0)
3459 {
3460 /* Handle a single character that has more than one other case. */
3461
3462 if (rc > 0) n8 += add_list_to_class(classbits, uchardptr, options, cd,
3463 PRIV(ucd_caseless_sets) + rc, oc);
3464
3465 /* Do nothing if the other case range is within the original range. */
3466
3467 else if (oc >= start && od <= end) continue;
3468
3469 /* Extend the original range if there is overlap, noting that if oc < c, we
3470 can't have od > end because a subrange is always shorter than the basic
3471 range. Otherwise, use a recursive call to add the additional range. */
3472
3473 else if (oc < start && od >= start - 1) start = oc; /* Extend downwards */
3474 else if (od > end && oc <= end + 1) end = od; /* Extend upwards */
3475 else n8 += add_to_class(classbits, uchardptr, options, cd, oc, od);
3476 }
3477 }
3478 else
3479 #endif /* SUPPORT_UCP */
3480
3481 /* Not UTF-mode, or no UCP */
3482
3483 for (c = start; c <= end && c < 256; c++)
3484 {
3485 SETBIT(classbits, cd->fcc[c]);
3486 n8++;
3487 }
3488 }
3489
3490 /* Now handle the original range. Adjust the final value according to the bit
3491 length - this means that the same lists of (e.g.) horizontal spaces can be used
3492 in all cases. */
3493
3494 #if defined COMPILE_PCRE8
3495 #ifdef SUPPORT_UTF
3496 if ((options & PCRE_UTF8) == 0)
3497 #endif
3498 if (end > 0xff) end = 0xff;
3499
3500 #elif defined COMPILE_PCRE16
3501 #ifdef SUPPORT_UTF
3502 if ((options & PCRE_UTF16) == 0)
3503 #endif
3504 if (end > 0xffff) end = 0xffff;
3505
3506 #endif /* COMPILE_PCRE[8|16] */
3507
3508 /* If all characters are less than 256, use the bit map. Otherwise use extra
3509 data. */
3510
3511 if (end < 0x100)
3512 {
3513 for (c = start; c <= end; c++)
3514 {
3515 n8++;
3516 SETBIT(classbits, c);
3517 }
3518 }
3519
3520 else
3521 {
3522 pcre_uchar *uchardata = *uchardptr;
3523
3524 #ifdef SUPPORT_UTF
3525 if ((options & PCRE_UTF8) != 0) /* All UTFs use the same flag bit */
3526 {
3527 if (start < end)
3528 {
3529 *uchardata++ = XCL_RANGE;
3530 uchardata += PRIV(ord2utf)(start, uchardata);
3531 uchardata += PRIV(ord2utf)(end, uchardata);
3532 }
3533 else if (start == end)
3534 {
3535 *uchardata++ = XCL_SINGLE;
3536 uchardata += PRIV(ord2utf)(start, uchardata);
3537 }
3538 }
3539 else
3540 #endif /* SUPPORT_UTF */
3541
3542 /* Without UTF support, character values are constrained by the bit length,
3543 and can only be > 256 for 16-bit and 32-bit libraries. */
3544
3545 #ifdef COMPILE_PCRE8
3546 {}
3547 #else
3548 if (start < end)
3549 {
3550 *uchardata++ = XCL_RANGE;
3551 *uchardata++ = start;
3552 *uchardata++ = end;
3553 }
3554 else if (start == end)
3555 {
3556 *uchardata++ = XCL_SINGLE;
3557 *uchardata++ = start;
3558 }
3559 #endif
3560
3561 *uchardptr = uchardata; /* Updata extra data pointer */
3562 }
3563
3564 return n8; /* Number of 8-bit characters */
3565 }
3566
3567
3568
3569
3570 /*************************************************
3571 * Add a list of characters to a class *
3572 *************************************************/
3573
3574 /* This function is used for adding a list of case-equivalent characters to a
3575 class, and also for adding a list of horizontal or vertical whitespace. If the
3576 list is in order (which it should be), ranges of characters are detected and
3577 handled appropriately. This function is mutually recursive with the function
3578 above.
3579
3580 Arguments:
3581 classbits the bit map for characters < 256
3582 uchardptr points to the pointer for extra data
3583 options the options word
3584 cd contains pointers to tables etc.
3585 p points to row of 32-bit values, terminated by NOTACHAR
3586 except character to omit; this is used when adding lists of
3587 case-equivalent characters to avoid including the one we
3588 already know about
3589
3590 Returns: the number of < 256 characters added
3591 the pointer to extra data is updated
3592 */
3593
3594 static int
3595 add_list_to_class(pcre_uint8 *classbits, pcre_uchar **uchardptr, int options,
3596 compile_data *cd, const pcre_uint32 *p, unsigned int except)
3597 {
3598 int n8 = 0;
3599 while (p[0] < NOTACHAR)
3600 {
3601 int n = 0;
3602 if (p[0] != except)
3603 {
3604 while(p[n+1] == p[0] + n + 1) n++;
3605 n8 += add_to_class(classbits, uchardptr, options, cd, p[0], p[n]);
3606 }
3607 p += n + 1;
3608 }
3609 return n8;
3610 }
3611
3612
3613
3614 /*************************************************
3615 * Add characters not in a list to a class *
3616 *************************************************/
3617
3618 /* This function is used for adding the complement of a list of horizontal or
3619 vertical whitespace to a class. The list must be in order.
3620
3621 Arguments:
3622 classbits the bit map for characters < 256
3623 uchardptr points to the pointer for extra data
3624 options the options word
3625 cd contains pointers to tables etc.
3626 p points to row of 32-bit values, terminated by NOTACHAR
3627
3628 Returns: the number of < 256 characters added
3629 the pointer to extra data is updated
3630 */
3631
3632 static int
3633 add_not_list_to_class(pcre_uint8 *classbits, pcre_uchar **uchardptr,
3634 int options, compile_data *cd, const pcre_uint32 *p)
3635 {
3636 BOOL utf = (options & PCRE_UTF8) != 0;
3637 int n8 = 0;
3638 if (p[0] > 0)
3639 n8 += add_to_class(classbits, uchardptr, options, cd, 0, p[0] - 1);
3640 while (p[0] < NOTACHAR)
3641 {
3642 while (p[1] == p[0] + 1) p++;
3643 n8 += add_to_class(classbits, uchardptr, options, cd, p[0] + 1,
3644 (p[1] == NOTACHAR) ? (utf ? 0x10ffffu : 0xffffffffu) : p[1] - 1);
3645 p++;
3646 }
3647 return n8;
3648 }
3649
3650
3651
3652 /*************************************************
3653 * Compile one branch *
3654 *************************************************/
3655
3656 /* Scan the pattern, compiling it into the a vector. If the options are
3657 changed during the branch, the pointer is used to change the external options
3658 bits. This function is used during the pre-compile phase when we are trying
3659 to find out the amount of memory needed, as well as during the real compile
3660 phase. The value of lengthptr distinguishes the two phases.
3661
3662 Arguments:
3663 optionsptr pointer to the option bits
3664 codeptr points to the pointer to the current code point
3665 ptrptr points to the current pattern pointer
3666 errorcodeptr points to error code variable
3667 firstcharptr set to initial literal character, or < 0 (REQ_UNSET, REQ_NONE)
3668 reqcharptr set to the last literal character required, else < 0
3669 bcptr points to current branch chain
3670 cond_depth conditional nesting depth
3671 cd contains pointers to tables etc.
3672 lengthptr NULL during the real compile phase
3673 points to length accumulator during pre-compile phase
3674
3675 Returns: TRUE on success
3676 FALSE, with *errorcodeptr set non-zero on error
3677 */
3678
3679 static BOOL
3680 compile_branch(int *optionsptr, pcre_uchar **codeptr,
3681 const pcre_uchar **ptrptr, int *errorcodeptr, pcre_int32 *firstcharptr,
3682 pcre_int32 *reqcharptr, branch_chain *bcptr, int cond_depth,
3683 compile_data *cd, int *lengthptr)
3684 {
3685 int repeat_type, op_type;
3686 int repeat_min = 0, repeat_max = 0; /* To please picky compilers */
3687 int bravalue = 0;
3688 int greedy_default, greedy_non_default;
3689 pcre_int32 firstchar, reqchar;
3690 pcre_int32 zeroreqchar, zerofirstchar;
3691 pcre_int32 req_caseopt, reqvary, tempreqvary;
3692 int options = *optionsptr; /* May change dynamically */
3693 int after_manual_callout = 0;
3694 int length_prevgroup = 0;
3695 register pcre_uint32 c;
3696 int escape;
3697 register pcre_uchar *code = *codeptr;
3698 pcre_uchar *last_code = code;
3699 pcre_uchar *orig_code = code;
3700 pcre_uchar *tempcode;
3701 BOOL inescq = FALSE;
3702 BOOL groupsetfirstchar = FALSE;
3703 const pcre_uchar *ptr = *ptrptr;
3704 const pcre_uchar *tempptr;
3705 const pcre_uchar *nestptr = NULL;
3706 pcre_uchar *previous = NULL;
3707 pcre_uchar *previous_callout = NULL;
3708 pcre_uchar *save_hwm = NULL;
3709 pcre_uint8 classbits[32];
3710
3711 /* We can fish out the UTF-8 setting once and for all into a BOOL, but we
3712 must not do this for other options (e.g. PCRE_EXTENDED) because they may change
3713 dynamically as we process the pattern. */
3714
3715 #ifdef SUPPORT_UTF
3716 /* PCRE_UTF[16|32] have the same value as PCRE_UTF8. */
3717 BOOL utf = (options & PCRE_UTF8) != 0;
3718 pcre_uchar utf_chars[6];
3719 #else
3720 BOOL utf = FALSE;
3721 #endif
3722
3723 /* Helper variables for OP_XCLASS opcode (for characters > 255). We define
3724 class_uchardata always so that it can be passed to add_to_class() always,
3725 though it will not be used in non-UTF 8-bit cases. This avoids having to supply
3726 alternative calls for the different cases. */
3727
3728 pcre_uchar *class_uchardata;
3729 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
3730 BOOL xclass;
3731 pcre_uchar *class_uchardata_base;
3732 #endif
3733
3734 #ifdef PCRE_DEBUG
3735 if (lengthptr != NULL) DPRINTF((">> start branch\n"));
3736 #endif
3737
3738 /* Set up the default and non-default settings for greediness */
3739
3740 greedy_default = ((options & PCRE_UNGREEDY) != 0);
3741 greedy_non_default = greedy_default ^ 1;
3742
3743 /* Initialize no first byte, no required byte. REQ_UNSET means "no char
3744 matching encountered yet". It gets changed to REQ_NONE if we hit something that
3745 matches a non-fixed char first char; reqchar just remains unset if we never
3746 find one.
3747
3748 When we hit a repeat whose minimum is zero, we may have to adjust these values
3749 to take the zero repeat into account. This is implemented by setting them to
3750 zerofirstbyte and zeroreqchar when such a repeat is encountered. The individual
3751 item types that can be repeated set these backoff variables appropriately. */
3752
3753 firstchar = reqchar = zerofirstchar = zeroreqchar = REQ_UNSET;
3754
3755 /* The variable req_caseopt contains either the REQ_CASELESS value
3756 or zero, according to the current setting of the caseless flag. The
3757 REQ_CASELESS leaves the lower 28 bit empty. It is added into the
3758 firstchar or reqchar variables to record the case status of the
3759 value. This is used only for ASCII characters. */
3760
3761 req_caseopt = ((options & PCRE_CASELESS) != 0)? REQ_CASELESS:0;
3762
3763 /* Switch on next character until the end of the branch */
3764
3765 for (;; ptr++)
3766 {
3767 BOOL negate_class;
3768 BOOL should_flip_negation;
3769 BOOL possessive_quantifier;
3770 BOOL is_quantifier;
3771 BOOL is_recurse;
3772 BOOL reset_bracount;
3773 int class_has_8bitchar;
3774 int class_one_char;
3775 int newoptions;
3776 int recno;
3777 int refsign;
3778 int skipbytes;
3779 int subreqchar;
3780 int subfirstchar;
3781 int terminator;
3782 int mclength;
3783 int tempbracount;
3784 int ec; // FIXMEchpe pcre_uint32
3785 pcre_uchar mcbuffer[8];
3786
3787 /* Get next character in the pattern */
3788
3789 c = *ptr;
3790
3791 /* If we are at the end of a nested substitution, revert to the outer level
3792 string. Nesting only happens one level deep. */
3793
3794 if (c == 0 && nestptr != NULL)
3795 {
3796 ptr = nestptr;
3797 nestptr = NULL;
3798 c = *ptr;
3799 }
3800
3801 /* If we are in the pre-compile phase, accumulate the length used for the
3802 previous cycle of this loop. */
3803
3804 if (lengthptr != NULL)
3805 {
3806 #ifdef PCRE_DEBUG
3807 if (code > cd->hwm) cd->hwm = code; /* High water info */
3808 #endif
3809 if (code > cd->start_workspace + cd->workspace_size -
3810 WORK_SIZE_SAFETY_MARGIN) /* Check for overrun */
3811 {
3812 *errorcodeptr = ERR52;
3813 goto FAILED;
3814 }
3815
3816 /* There is at least one situation where code goes backwards: this is the
3817 case of a zero quantifier after a class (e.g. [ab]{0}). At compile time,
3818 the class is simply eliminated. However, it is created first, so we have to
3819 allow memory for it. Therefore, don't ever reduce the length at this point.
3820 */
3821
3822 if (code < last_code) code = last_code;
3823
3824 /* Paranoid check for integer overflow */
3825
3826 if (OFLOW_MAX - *lengthptr < code - last_code)
3827 {
3828 *errorcodeptr = ERR20;
3829 goto FAILED;
3830 }
3831
3832 *lengthptr += (int)(code - last_code);
3833 DPRINTF(("length=%d added %d c=%c (0x%x)\n", *lengthptr,
3834 (int)(code - last_code), c, c));
3835
3836 /* If "previous" is set and it is not at the start of the work space, move
3837 it back to there, in order to avoid filling up the work space. Otherwise,
3838 if "previous" is NULL, reset the current code pointer to the start. */
3839
3840 if (previous != NULL)
3841 {
3842 if (previous > orig_code)
3843 {
3844 memmove(orig_code, previous, IN_UCHARS(code - previous));
3845 code -= previous - orig_code;
3846 previous = orig_code;
3847 }
3848 }
3849 else code = orig_code;
3850
3851 /* Remember where this code item starts so we can pick up the length
3852 next time round. */
3853
3854 last_code = code;
3855 }
3856
3857 /* In the real compile phase, just check the workspace used by the forward
3858 reference list. */
3859
3860 else if (cd->hwm > cd->start_workspace + cd->workspace_size -
3861 WORK_SIZE_SAFETY_MARGIN)
3862 {
3863 *errorcodeptr = ERR52;
3864 goto FAILED;
3865 }
3866
3867 /* If in \Q...\E, check for the end; if not, we have a literal */
3868
3869 if (inescq && c != 0)
3870 {
3871 if (c == CHAR_BACKSLASH && ptr[1] == CHAR_E)
3872 {
3873 inescq = FALSE;
3874 ptr++;
3875 continue;
3876 }
3877 else
3878 {
3879 if (previous_callout != NULL)
3880 {
3881 if (lengthptr == NULL) /* Don't attempt in pre-compile phase */
3882 complete_callout(previous_callout, ptr, cd);
3883 previous_callout = NULL;
3884 }
3885 if ((options & PCRE_AUTO_CALLOUT) != 0)
3886 {
3887 previous_callout = code;
3888 code = auto_callout(code, ptr, cd);
3889 }
3890 goto NORMAL_CHAR;
3891 }
3892 }
3893
3894 /* Fill in length of a previous callout, except when the next thing is
3895 a quantifier. */
3896
3897 is_quantifier =
3898 c == CHAR_ASTERISK || c == CHAR_PLUS || c == CHAR_QUESTION_MARK ||
3899 (c == CHAR_LEFT_CURLY_BRACKET && is_counted_repeat(ptr+1));
3900
3901 if (!is_quantifier && previous_callout != NULL &&
3902 after_manual_callout-- <= 0)
3903 {
3904 if (lengthptr == NULL) /* Don't attempt in pre-compile phase */
3905 complete_callout(previous_callout, ptr, cd);
3906 previous_callout = NULL;
3907 }
3908
3909 /* In extended mode, skip white space and comments. */
3910
3911 if ((options & PCRE_EXTENDED) != 0)
3912 {
3913 if (MAX_255(*ptr) && (cd->ctypes[c] & ctype_space) != 0) continue;
3914 if (c == CHAR_NUMBER_SIGN)
3915 {
3916 ptr++;
3917 while (*ptr != 0)
3918 {
3919 if (IS_NEWLINE(ptr)) { ptr += cd->nllen - 1; break; }
3920 ptr++;
3921 #ifdef SUPPORT_UTF
3922 if (utf) FORWARDCHAR(ptr);
3923 #endif
3924 }
3925 if (*ptr != 0) continue;
3926
3927 /* Else fall through to handle end of string */
3928 c = 0;
3929 }
3930 }
3931
3932 /* No auto callout for quantifiers. */
3933
3934 if ((options & PCRE_AUTO_CALLOUT) != 0 && !is_quantifier)
3935 {
3936 previous_callout = code;
3937 code = auto_callout(code, ptr, cd);
3938 }
3939
3940 switch(c)
3941 {
3942 /* ===================================================================*/
3943 case 0: /* The branch terminates at string end */
3944 case CHAR_VERTICAL_LINE: /* or | or ) */
3945 case CHAR_RIGHT_PARENTHESIS:
3946 *firstcharptr = firstchar;
3947 *reqcharptr = reqchar;
3948 *codeptr = code;
3949 *ptrptr = ptr;
3950 if (lengthptr != NULL)
3951 {
3952 if (OFLOW_MAX - *lengthptr < code - last_code)
3953 {
3954 *errorcodeptr = ERR20;
3955 goto FAILED;
3956 }
3957 *lengthptr += (int)(code - last_code); /* To include callout length */
3958 DPRINTF((">> end branch\n"));
3959 }
3960 return TRUE;
3961
3962
3963 /* ===================================================================*/
3964 /* Handle single-character metacharacters. In multiline mode, ^ disables
3965 the setting of any following char as a first character. */
3966
3967 case CHAR_CIRCUMFLEX_ACCENT:
3968 previous = NULL;
3969 if ((options & PCRE_MULTILINE) != 0)
3970 {
3971 if (firstchar == REQ_UNSET) firstchar = REQ_NONE;
3972 *code++ = OP_CIRCM;
3973 }
3974 else *code++ = OP_CIRC;
3975 break;
3976
3977 case CHAR_DOLLAR_SIGN:
3978 previous = NULL;
3979 *code++ = ((options & PCRE_MULTILINE) != 0)? OP_DOLLM : OP_DOLL;
3980 break;
3981
3982 /* There can never be a first char if '.' is first, whatever happens about
3983 repeats. The value of reqchar doesn't change either. */
3984
3985 case CHAR_DOT:
3986 if (firstchar == REQ_UNSET) firstchar = REQ_NONE;
3987 zerofirstchar = firstchar;
3988 zeroreqchar = reqchar;
3989 previous = code;
3990 *code++ = ((options & PCRE_DOTALL) != 0)? OP_ALLANY: OP_ANY;
3991 break;
3992
3993
3994 /* ===================================================================*/
3995 /* Character classes. If the included characters are all < 256, we build a
3996 32-byte bitmap of the permitted characters, except in the special case
3997 where there is only one such character. For negated classes, we build the
3998 map as usual, then invert it at the end. However, we use a different opcode
3999 so that data characters > 255 can be handled correctly.
4000
4001 If the class contains characters outside the 0-255 range, a different
4002 opcode is compiled. It may optionally have a bit map for characters < 256,
4003 but those above are are explicitly listed afterwards. A flag byte tells
4004 whether the bitmap is present, and whether this is a negated class or not.
4005
4006 In JavaScript compatibility mode, an isolated ']' causes an error. In
4007 default (Perl) mode, it is treated as a data character. */
4008
4009 case CHAR_RIGHT_SQUARE_BRACKET:
4010 if ((cd->external_options & PCRE_JAVASCRIPT_COMPAT) != 0)
4011 {
4012 *errorcodeptr = ERR64;
4013 goto FAILED;
4014 }
4015 goto NORMAL_CHAR;
4016
4017 case CHAR_LEFT_SQUARE_BRACKET:
4018 previous = code;
4019
4020 /* PCRE supports POSIX class stuff inside a class. Perl gives an error if
4021 they are encountered at the top level, so we'll do that too. */
4022
4023 if ((ptr[1] == CHAR_COLON || ptr[1] == CHAR_DOT ||
4024 ptr[1] == CHAR_EQUALS_SIGN) &&
4025 check_posix_syntax(ptr, &tempptr))
4026 {
4027 *errorcodeptr = (ptr[1] == CHAR_COLON)? ERR13 : ERR31;
4028 goto FAILED;
4029 }
4030
4031 /* If the first character is '^', set the negation flag and skip it. Also,
4032 if the first few characters (either before or after ^) are \Q\E or \E we
4033 skip them too. This makes for compatibility with Perl. */
4034
4035 negate_class = FALSE;
4036 for (;;)
4037 {
4038 c = *(++ptr);
4039 if (c == CHAR_BACKSLASH)
4040 {
4041 if (ptr[1] == CHAR_E)
4042 ptr++;
4043 else if (STRNCMP_UC_C8(ptr + 1, STR_Q STR_BACKSLASH STR_E, 3) == 0)
4044 ptr += 3;
4045 else
4046 break;
4047 }
4048 else if (!negate_class && c == CHAR_CIRCUMFLEX_ACCENT)
4049 negate_class = TRUE;
4050 else break;
4051 }
4052
4053 /* Empty classes are allowed in JavaScript compatibility mode. Otherwise,
4054 an initial ']' is taken as a data character -- the code below handles
4055 that. In JS mode, [] must always fail, so generate OP_FAIL, whereas
4056 [^] must match any character, so generate OP_ALLANY. */
4057
4058 if (c == CHAR_RIGHT_SQUARE_BRACKET &&
4059 (cd->external_options & PCRE_JAVASCRIPT_COMPAT) != 0)
4060 {
4061 *code++ = negate_class? OP_ALLANY : OP_FAIL;
4062 if (firstchar == REQ_UNSET) firstchar = REQ_NONE;
4063 zerofirstchar = firstchar;
4064 break;
4065 }
4066
4067 /* If a class contains a negative special such as \S, we need to flip the
4068 negation flag at the end, so that support for characters > 255 works
4069 correctly (they are all included in the class). */
4070
4071 should_flip_negation = FALSE;
4072
4073 /* For optimization purposes, we track some properties of the class:
4074 class_has_8bitchar will be non-zero if the class contains at least one <
4075 256 character; class_one_char will be 1 if the class contains just one
4076 character. */
4077
4078 class_has_8bitchar = 0;
4079 class_one_char = 0;
4080
4081 /* Initialize the 32-char bit map to all zeros. We build the map in a
4082 temporary bit of memory, in case the class contains fewer than two
4083 8-bit characters because in that case the compiled code doesn't use the bit
4084 map. */
4085
4086 memset(classbits, 0, 32 * sizeof(pcre_uint8));
4087
4088 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
4089 xclass = FALSE;
4090 class_uchardata = code + LINK_SIZE + 2; /* For XCLASS items */
4091 class_uchardata_base = class_uchardata; /* Save the start */
4092 #endif
4093
4094 /* Process characters until ] is reached. By writing this as a "do" it
4095 means that an initial ] is taken as a data character. At the start of the
4096 loop, c contains the first byte of the character. */
4097
4098 if (c != 0) do
4099 {
4100 const pcre_uchar *oldptr;
4101
4102 #if defined SUPPORT_UTF && !defined COMPILE_PCRE32
4103 if (utf && HAS_EXTRALEN(c))
4104 { /* Braces are required because the */
4105 GETCHARLEN(c, ptr, ptr); /* macro generates multiple statements */
4106 }
4107 #endif
4108
4109 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
4110 /* In the pre-compile phase, accumulate the length of any extra
4111 data and reset the pointer. This is so that very large classes that
4112 contain a zillion > 255 characters no longer overwrite the work space
4113 (which is on the stack). We have to remember that there was XCLASS data,
4114 however. */
4115
4116 if (lengthptr != NULL && class_uchardata > class_uchardata_base)
4117 {
4118 xclass = TRUE;
4119 *lengthptr += class_uchardata - class_uchardata_base;
4120 class_uchardata = class_uchardata_base;
4121 }
4122 #endif
4123
4124 /* Inside \Q...\E everything is literal except \E */
4125
4126 if (inescq)
4127 {
4128 if (c == CHAR_BACKSLASH && ptr[1] == CHAR_E) /* If we are at \E */
4129 {
4130 inescq = FALSE; /* Reset literal state */
4131 ptr++; /* Skip the 'E' */
4132 continue; /* Carry on with next */
4133 }
4134 goto CHECK_RANGE; /* Could be range if \E follows */
4135 }
4136
4137 /* Handle POSIX class names. Perl allows a negation extension of the
4138 form [:^name:]. A square bracket that doesn't match the syntax is
4139 treated as a literal. We also recognize the POSIX constructions
4140 [.ch.] and [=ch=] ("collating elements") and fault them, as Perl
4141 5.6 and 5.8 do. */
4142
4143 if (c == CHAR_LEFT_SQUARE_BRACKET &&
4144 (ptr[1] == CHAR_COLON || ptr[1] == CHAR_DOT ||
4145 ptr[1] == CHAR_EQUALS_SIGN) && check_posix_syntax(ptr, &tempptr))
4146 {
4147 BOOL local_negate = FALSE;
4148 int posix_class, taboffset, tabopt;
4149 register const pcre_uint8 *cbits = cd->cbits;
4150 pcre_uint8 pbits[32];
4151
4152 if (ptr[1] != CHAR_COLON)
4153 {
4154 *errorcodeptr = ERR31;
4155 goto FAILED;
4156 }
4157
4158 ptr += 2;
4159 if (*ptr == CHAR_CIRCUMFLEX_ACCENT)
4160 {
4161 local_negate = TRUE;
4162 should_flip_negation = TRUE; /* Note negative special */
4163 ptr++;
4164 }
4165
4166 posix_class = check_posix_name(ptr, (int)(tempptr - ptr));
4167 if (posix_class < 0)
4168 {
4169 *errorcodeptr = ERR30;
4170 goto FAILED;
4171 }
4172
4173 /* If matching is caseless, upper and lower are converted to
4174 alpha. This relies on the fact that the class table starts with
4175 alpha, lower, upper as the first 3 entries. */
4176
4177 if ((options & PCRE_CASELESS) != 0 && posix_class <= 2)
4178 posix_class = 0;
4179
4180 /* When PCRE_UCP is set, some of the POSIX classes are converted to
4181 different escape sequences that use Unicode properties. */
4182
4183 #ifdef SUPPORT_UCP
4184 if ((options & PCRE_UCP) != 0)
4185 {
4186 int pc = posix_class + ((local_negate)? POSIX_SUBSIZE/2 : 0);
4187 if (posix_substitutes[pc] != NULL)
4188 {
4189 nestptr = tempptr + 1;
4190 ptr = posix_substitutes[pc] - 1;
4191 continue;
4192 }
4193 }
4194 #endif
4195 /* In the non-UCP case, we build the bit map for the POSIX class in a
4196 chunk of local store because we may be adding and subtracting from it,
4197 and we don't want to subtract bits that may be in the main map already.
4198 At the end we or the result into the bit map that is being built. */
4199
4200 posix_class *= 3;
4201
4202 /* Copy in the first table (always present) */
4203
4204 memcpy(pbits, cbits + posix_class_maps[posix_class],
4205 32 * sizeof(pcre_uint8));
4206
4207 /* If there is a second table, add or remove it as required. */
4208
4209 taboffset = posix_class_maps[posix_class + 1];
4210 tabopt = posix_class_maps[posix_class + 2];
4211
4212 if (taboffset >= 0)
4213 {
4214 if (tabopt >= 0)
4215 for (c = 0; c < 32; c++) pbits[c] |= cbits[c + taboffset];
4216 else
4217 for (c = 0; c < 32; c++) pbits[c] &= ~cbits[c + taboffset];
4218 }
4219
4220 /* Now see if we need to remove any special characters. An option
4221 value of 1 removes vertical space and 2 removes underscore. */
4222
4223 if (tabopt < 0) tabopt = -tabopt;
4224 if (tabopt == 1) pbits[1] &= ~0x3c;
4225 else if (tabopt == 2) pbits[11] &= 0x7f;
4226
4227 /* Add the POSIX table or its complement into the main table that is
4228 being built and we are done. */
4229
4230 if (local_negate)
4231 for (c = 0; c < 32; c++) classbits[c] |= ~pbits[c];
4232 else
4233 for (c = 0; c < 32; c++) classbits[c] |= pbits[c];
4234
4235 ptr = tempptr + 1;
4236 /* Every class contains at least one < 256 character. */
4237 class_has_8bitchar = 1;
4238 /* Every class contains at least two characters. */
4239 class_one_char = 2;
4240 continue; /* End of POSIX syntax handling */
4241 }
4242
4243 /* Backslash may introduce a single character, or it may introduce one
4244 of the specials, which just set a flag. The sequence \b is a special
4245 case. Inside a class (and only there) it is treated as backspace. We
4246 assume that other escapes have more than one character in them, so
4247 speculatively set both class_has_8bitchar and class_one_char bigger
4248 than one. Unrecognized escapes fall through and are either treated
4249 as literal characters (by default), or are faulted if
4250 PCRE_EXTRA is set. */
4251
4252 if (c == CHAR_BACKSLASH)
4253 {
4254 escape = check_escape(&ptr, &ec, errorcodeptr, cd->bracount, options, TRUE);
4255
4256 if (*errorcodeptr != 0) goto FAILED;
4257
4258 if (escape == 0)
4259 c = ec;
4260 else if (escape == ESC_b) c = CHAR_BS; /* \b is backspace in a class */
4261 else if (escape == ESC_N) /* \N is not supported in a class */
4262 {
4263 *errorcodeptr = ERR71;
4264 goto FAILED;
4265 }
4266 else if (escape == ESC_Q) /* Handle start of quoted string */
4267 {
4268 if (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
4269 {
4270 ptr += 2; /* avoid empty string */
4271 }
4272 else inescq = TRUE;
4273 continue;
4274 }
4275 else if (escape == ESC_E) continue; /* Ignore orphan \E */
4276
4277 else
4278 {
4279 register const pcre_uint8 *cbits = cd->cbits;
4280 /* Every class contains at least two < 256 characters. */
4281 class_has_8bitchar++;
4282 /* Every class contains at least two characters. */
4283 class_one_char += 2;
4284
4285 switch (escape)
4286 {
4287 #ifdef SUPPORT_UCP
4288 case ESC_du: /* These are the values given for \d etc */
4289 case ESC_DU: /* when PCRE_UCP is set. We replace the */
4290 case ESC_wu: /* escape sequence with an appropriate \p */
4291 case ESC_WU: /* or \P to test Unicode properties instead */
4292 case ESC_su: /* of the default ASCII testing. */
4293 case ESC_SU:
4294 nestptr = ptr;
4295 ptr = substitutes[escape - ESC_DU] - 1; /* Just before substitute */
4296 class_has_8bitchar--; /* Undo! */
4297 continue;
4298 #endif
4299 case ESC_d:
4300 for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_digit];
4301 continue;
4302
4303 case ESC_D:
4304 should_flip_negation = TRUE;
4305 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_digit];
4306 continue;
4307
4308 case ESC_w:
4309 for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_word];
4310 continue;
4311
4312 case ESC_W:
4313 should_flip_negation = TRUE;
4314 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_word];
4315 continue;
4316
4317 /* Perl 5.004 onwards omits VT from \s, but we must preserve it
4318 if it was previously set by something earlier in the character
4319 class. Luckily, the value of CHAR_VT is 0x0b in both ASCII and
4320 EBCDIC, so we lazily just adjust the appropriate bit. */
4321
4322 case ESC_s:
4323 classbits[0] |= cbits[cbit_space];
4324 classbits[1] |= cbits[cbit_space+1] & ~0x08;
4325 for (c = 2; c < 32; c++) classbits[c] |= cbits[c+cbit_space];
4326 continue;
4327
4328 case ESC_S:
4329 should_flip_negation = TRUE;
4330 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_space];
4331 classbits[1] |= 0x08; /* Perl 5.004 onwards omits VT from \s */
4332 continue;
4333
4334 /* The rest apply in both UCP and non-UCP cases. */
4335
4336 case ESC_h:
4337 (void)add_list_to_class(classbits, &class_uchardata, options, cd,
4338 PRIV(hspace_list), NOTACHAR);
4339 continue;
4340
4341 case ESC_H:
4342 (void)add_not_list_to_class(classbits, &class_uchardata, options,
4343 cd, PRIV(hspace_list));
4344 continue;
4345
4346 case ESC_v:
4347 (void)add_list_to_class(classbits, &class_uchardata, options, cd,
4348 PRIV(vspace_list), NOTACHAR);
4349 continue;
4350
4351 case ESC_V:
4352 (void)add_not_list_to_class(classbits, &class_uchardata, options,
4353 cd, PRIV(vspace_list));
4354 continue;
4355
4356 #ifdef SUPPORT_UCP
4357 case ESC_p:
4358 case ESC_P:
4359 {
4360 BOOL negated;
4361 int pdata;
4362 int ptype = get_ucp(&ptr, &negated, &pdata, errorcodeptr);
4363 if (ptype < 0) goto FAILED;
4364 *class_uchardata++ = ((escape == ESC_p) != negated)?
4365 XCL_PROP : XCL_NOTPROP;
4366 *class_uchardata++ = ptype;
4367 *class_uchardata++ = pdata;
4368 class_has_8bitchar--; /* Undo! */
4369 continue;
4370 }
4371 #endif
4372 /* Unrecognized escapes are faulted if PCRE is running in its
4373 strict mode. By default, for compatibility with Perl, they are
4374 treated as literals. */
4375
4376 default:
4377 if ((options & PCRE_EXTRA) != 0)
4378 {
4379 *errorcodeptr = ERR7;
4380 goto FAILED;
4381 }
4382 class_has_8bitchar--; /* Undo the speculative increase. */
4383 class_one_char -= 2; /* Undo the speculative increase. */
4384 c = *ptr; /* Get the final character and fall through */
4385 break;
4386 }
4387 }
4388
4389 /* Fall through if the escape just defined a single character (c >= 0).
4390 This may be greater than 256. */
4391
4392 escape = 0;
4393
4394 } /* End of backslash handling */
4395
4396 /* A character may be followed by '-' to form a range. However, Perl does
4397 not permit ']' to be the end of the range. A '-' character at the end is
4398 treated as a literal. Perl ignores orphaned \E sequences entirely. The
4399 code for handling \Q and \E is messy. */
4400
4401 CHECK_RANGE:
4402 while (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
4403 {
4404 inescq = FALSE;
4405 ptr += 2;
4406 }
4407 oldptr = ptr;
4408
4409 /* Remember if \r or \n were explicitly used */
4410
4411 if (c == CHAR_CR || c == CHAR_NL) cd->external_flags |= PCRE_HASCRORLF;
4412
4413 /* Check for range */
4414
4415 if (!inescq && ptr[1] == CHAR_MINUS)
4416 {
4417 int d;
4418 ptr += 2;
4419 while (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_E) ptr += 2;
4420
4421 /* If we hit \Q (not followed by \E) at this point, go into escaped
4422 mode. */
4423
4424 while (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_Q)
4425 {
4426 ptr += 2;
4427 if (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_E)
4428 { ptr += 2; continue; }
4429 inescq = TRUE;
4430 break;
4431 }
4432
4433 /* Minus (hyphen) at the end of a class is treated as a literal, so put
4434 back the pointer and jump to handle the character that preceded it. */
4435
4436 if (*ptr == 0 || (!inescq && *ptr == CHAR_RIGHT_SQUARE_BRACKET))
4437 {
4438 ptr = oldptr;
4439 goto CLASS_SINGLE_CHARACTER;
4440 }
4441
4442 /* Otherwise, we have a potential range; pick up the next character */
4443
4444 #ifdef SUPPORT_UTF
4445 if (utf)
4446 { /* Braces are required because the */
4447 GETCHARLEN(d, ptr, ptr); /* macro generates multiple statements */
4448 }
4449 else
4450 #endif
4451 d = *ptr; /* Not UTF-8 mode */
4452
4453 /* The second part of a range can be a single-character escape, but
4454 not any of the other escapes. Perl 5.6 treats a hyphen as a literal
4455 in such circumstances. */
4456
4457 if (!inescq && d == CHAR_BACKSLASH)
4458 {
4459 int descape;
4460 descape = check_escape(&ptr, &d, errorcodeptr, cd->bracount, options, TRUE);
4461 if (*errorcodeptr != 0) goto FAILED;
4462
4463 /* \b is backspace; any other special means the '-' was literal. */
4464
4465 if (descape != 0)
4466 {
4467 if (descape == ESC_b) d = CHAR_BS; else
4468 {
4469 ptr = oldptr;
4470 goto CLASS_SINGLE_CHARACTER; /* A few lines below */
4471 }
4472 }
4473 }
4474
4475 /* Check that the two values are in the correct order. Optimize
4476 one-character ranges. */
4477
4478 if (d < c)
4479 {
4480 *errorcodeptr = ERR8;
4481 goto FAILED;
4482 }
4483 if (d == c) goto CLASS_SINGLE_CHARACTER; /* A few lines below */
4484
4485 /* We have found a character range, so single character optimizations
4486 cannot be done anymore. Any value greater than 1 indicates that there
4487 is more than one character. */
4488
4489 class_one_char = 2;
4490
4491 /* Remember an explicit \r or \n, and add the range to the class. */
4492
4493 if (d == CHAR_CR || d == CHAR_NL) cd->external_flags |= PCRE_HASCRORLF;
4494
4495 class_has_8bitchar +=
4496 add_to_class(classbits, &class_uchardata, options, cd, c, d);
4497
4498 continue; /* Go get the next char in the class */
4499 }
4500
4501 /* Handle a single character - we can get here for a normal non-escape
4502 char, or after \ that introduces a single character or for an apparent
4503 range that isn't. Only the value 1 matters for class_one_char, so don't
4504 increase it if it is already 2 or more ... just in case there's a class
4505 with a zillion characters in it. */
4506
4507 CLASS_SINGLE_CHARACTER:
4508 if (class_one_char < 2) class_one_char++;
4509
4510 /* If class_one_char is 1, we have the first single character in the
4511 class, and there have been no prior ranges, or XCLASS items generated by
4512 escapes. If this is the final character in the class, we can optimize by
4513 turning the item into a 1-character OP_CHAR[I] if it's positive, or
4514 OP_NOT[I] if it's negative. In the positive case, it can cause firstchar
4515 to be set. Otherwise, there can be no first char if this item is first,
4516 whatever repeat count may follow. In the case of reqchar, save the
4517 previous value for reinstating. */
4518
4519 if (class_one_char == 1 && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET)
4520 {
4521 ptr++;
4522 zeroreqchar = reqchar;
4523
4524 if (negate_class)
4525 {
4526 #ifdef SUPPORT_UCP
4527 // FIXMEchpe pcreuint32?
4528 int d;
4529 #endif
4530 if (firstchar == REQ_UNSET) firstchar = REQ_NONE;
4531 zerofirstchar = firstchar;
4532
4533 /* For caseless UTF-8 mode when UCP support is available, check
4534 whether this character has more than one other case. If so, generate
4535 a special OP_NOTPROP item instead of OP_NOTI. */
4536
4537 #ifdef SUPPORT_UCP
4538 if (utf && (options & PCRE_CASELESS) != 0 &&
4539 (d = UCD_CASESET(c)) != 0)
4540 {
4541 *code++ = OP_NOTPROP;
4542 *code++ = PT_CLIST;
4543 *code++ = d;
4544 }
4545 else
4546 #endif
4547 /* Char has only one other case, or UCP not available */
4548
4549 {
4550 *code++ = ((options & PCRE_CASELESS) != 0)? OP_NOTI: OP_NOT;
4551 #if defined SUPPORT_UTF && !defined COMPILE_PCRE32
4552 if (utf && c > MAX_VALUE_FOR_SINGLE_CHAR)
4553 code += PRIV(ord2utf)(c, code);
4554 else
4555 #endif
4556 *code++ = c;
4557 }
4558
4559 /* We are finished with this character class */
4560
4561 goto END_CLASS;
4562 }
4563
4564 /* For a single, positive character, get the value into mcbuffer, and
4565 then we can handle this with the normal one-character code. */
4566
4567 #if defined SUPPORT_UTF && !defined COMPILE_PCRE32
4568 if (utf && c > MAX_VALUE_FOR_SINGLE_CHAR)
4569 mclength = PRIV(ord2utf)(c, mcbuffer);
4570 else
4571 #endif
4572 {
4573 mcbuffer[0] = c;
4574 mclength = 1;
4575 }
4576 goto ONE_CHAR;
4577 } /* End of 1-char optimization */
4578
4579 /* There is more than one character in the class, or an XCLASS item
4580 has been generated. Add this character to the class. */
4581
4582 class_has_8bitchar +=
4583 add_to_class(classbits, &class_uchardata, options, cd, c, c);
4584 }
4585
4586 /* Loop until ']' reached. This "while" is the end of the "do" far above.
4587 If we are at the end of an internal nested string, revert to the outer
4588 string. */
4589
4590 while (((c = *(++ptr)) != 0 ||
4591 (nestptr != NULL &&
4592 (ptr = nestptr, nestptr = NULL, c = *(++ptr)) != 0)) &&
4593 (c != CHAR_RIGHT_SQUARE_BRACKET || inescq));
4594
4595 /* Check for missing terminating ']' */
4596
4597 if (c == 0)
4598 {
4599 *errorcodeptr = ERR6;
4600 goto FAILED;
4601 }
4602
4603 /* We will need an XCLASS if data has been placed in class_uchardata. In
4604 the second phase this is a sufficient test. However, in the pre-compile
4605 phase, class_uchardata gets emptied to prevent workspace overflow, so it
4606 only if the very last character in the class needs XCLASS will it contain
4607 anything at this point. For this reason, xclass gets set TRUE above when
4608 uchar_classdata is emptied, and that's why this code is the way it is here
4609 instead of just doing a test on class_uchardata below. */
4610
4611 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
4612 if (class_uchardata > class_uchardata_base) xclass = TRUE;
4613 #endif
4614
4615 /* If this is the first thing in the branch, there can be no first char
4616 setting, whatever the repeat count. Any reqchar setting must remain
4617 unchanged after any kind of repeat. */
4618
4619 if (firstchar == REQ_UNSET) firstchar = REQ_NONE;
4620 zerofirstchar = firstchar;
4621 zeroreqchar = reqchar;
4622
4623 /* If there are characters with values > 255, we have to compile an
4624 extended class, with its own opcode, unless there was a negated special
4625 such as \S in the class, and PCRE_UCP is not set, because in that case all
4626 characters > 255 are in the class, so any that were explicitly given as
4627 well can be ignored. If (when there are explicit characters > 255 that must
4628 be listed) there are no characters < 256, we can omit the bitmap in the
4629 actual compiled code. */
4630
4631 #ifdef SUPPORT_UTF
4632 if (xclass && (!should_flip_negation || (options & PCRE_UCP) != 0))
4633 #elif !defined COMPILE_PCRE8
4634 if (xclass && !should_flip_negation)
4635 #endif
4636 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
4637 {
4638 *class_uchardata++ = XCL_END; /* Marks the end of extra data */
4639 *code++ = OP_XCLASS;
4640 code += LINK_SIZE;
4641 *code = negate_class? XCL_NOT:0;
4642
4643 /* If the map is required, move up the extra data to make room for it;
4644 otherwise just move the code pointer to the end of the extra data. */
4645
4646 if (class_has_8bitchar > 0)
4647 {
4648 *code++ |= XCL_MAP;
4649 memmove(code + (32 / sizeof(pcre_uchar)), code,
4650 IN_UCHARS(class_uchardata - code));
4651 memcpy(code, classbits, 32);
4652 code = class_uchardata + (32 / sizeof(pcre_uchar));
4653 }
4654 else code = class_uchardata;
4655
4656 /* Now fill in the complete length of the item */
4657
4658 PUT(previous, 1, (int)(code - previous));
4659 break; /* End of class handling */
4660 }
4661 #endif
4662
4663 /* If there are no characters > 255, or they are all to be included or
4664 excluded, set the opcode to OP_CLASS or OP_NCLASS, depending on whether the
4665 whole class was negated and whether there were negative specials such as \S
4666 (non-UCP) in the class. Then copy the 32-byte map into the code vector,
4667 negating it if necessary. */
4668
4669 *code++ = (negate_class == should_flip_negation) ? OP_CLASS : OP_NCLASS;
4670 if (lengthptr == NULL) /* Save time in the pre-compile phase */
4671 {
4672 if (negate_class)
4673 for (c = 0; c < 32; c++) classbits[c] = ~classbits[c];
4674 memcpy(code, classbits, 32);
4675 }
4676 code += 32 / sizeof(pcre_uchar);
4677
4678 END_CLASS:
4679 break;
4680
4681
4682 /* ===================================================================*/
4683 /* Various kinds of repeat; '{' is not necessarily a quantifier, but this
4684 has been tested above. */
4685
4686 case CHAR_LEFT_CURLY_BRACKET:
4687 if (!is_quantifier) goto NORMAL_CHAR;
4688 ptr = read_repeat_counts(ptr+1, &repeat_min, &repeat_max, errorcodeptr);
4689 if (*errorcodeptr != 0) goto FAILED;
4690 goto REPEAT;
4691
4692 case CHAR_ASTERISK:
4693 repeat_min = 0;
4694 repeat_max = -1;
4695 goto REPEAT;
4696
4697 case CHAR_PLUS:
4698 repeat_min = 1;
4699 repeat_max = -1;
4700 goto REPEAT;
4701
4702 case CHAR_QUESTION_MARK:
4703 repeat_min = 0;
4704 repeat_max = 1;
4705
4706 REPEAT:
4707 if (previous == NULL)
4708 {
4709 *errorcodeptr = ERR9;
4710 goto FAILED;
4711 }
4712
4713 if (repeat_min == 0)
4714 {
4715 firstchar = zerofirstchar; /* Adjust for zero repeat */
4716 reqchar = zeroreqchar; /* Ditto */
4717 }
4718
4719 /* Remember whether this is a variable length repeat */
4720
4721 reqvary = (repeat_min == repeat_max)? 0 : REQ_VARY;
4722
4723 op_type = 0; /* Default single-char op codes */
4724 possessive_quantifier = FALSE; /* Default not possessive quantifier */
4725
4726 /* Save start of previous item, in case we have to move it up in order to
4727 insert something before it. */
4728
4729 tempcode = previous;
4730
4731 /* If the next character is '+', we have a possessive quantifier. This
4732 implies greediness, whatever the setting of the PCRE_UNGREEDY option.
4733 If the next character is '?' this is a minimizing repeat, by default,
4734 but if PCRE_UNGREEDY is set, it works the other way round. We change the
4735 repeat type to the non-default. */
4736
4737 if (ptr[1] == CHAR_PLUS)
4738 {
4739 repeat_type = 0; /* Force greedy */
4740 possessive_quantifier = TRUE;
4741 ptr++;
4742 }
4743 else if (ptr[1] == CHAR_QUESTION_MARK)
4744 {
4745 repeat_type = greedy_non_default;
4746 ptr++;
4747 }
4748 else repeat_type = greedy_default;
4749
4750 /* If previous was a recursion call, wrap it in atomic brackets so that
4751 previous becomes the atomic group. All recursions were so wrapped in the
4752 past, but it no longer happens for non-repeated recursions. In fact, the
4753 repeated ones could be re-implemented independently so as not to need this,
4754 but for the moment we rely on the code for repeating groups. */
4755
4756 if (*previous == OP_RECURSE)
4757 {
4758 memmove(previous + 1 + LINK_SIZE, previous, IN_UCHARS(1 + LINK_SIZE));
4759 *previous = OP_ONCE;
4760 PUT(previous, 1, 2 + 2*LINK_SIZE);
4761 previous[2 + 2*LINK_SIZE] = OP_KET;
4762 PUT(previous, 3 + 2*LINK_SIZE, 2 + 2*LINK_SIZE);
4763 code += 2 + 2 * LINK_SIZE;
4764 length_prevgroup = 3 + 3*LINK_SIZE;
4765
4766 /* When actually compiling, we need to check whether this was a forward
4767 reference, and if so, adjust the offset. */
4768
4769 if (lengthptr == NULL && cd->hwm >= cd->start_workspace + LINK_SIZE)
4770 {
4771 int offset = GET(cd->hwm, -LINK_SIZE);
4772 if (offset == previous + 1 - cd->start_code)
4773 PUT(cd->hwm, -LINK_SIZE, offset + 1 + LINK_SIZE);
4774 }
4775 }
4776
4777 /* Now handle repetition for the different types of item. */
4778
4779 /* If previous was a character or negated character match, abolish the item
4780 and generate a repeat item instead. If a char item has a minimum of more
4781 than one, ensure that it is set in reqchar - it might not be if a sequence
4782 such as x{3} is the first thing in a branch because the x will have gone
4783 into firstchar instead. */
4784
4785 if (*previous == OP_CHAR || *previous == OP_CHARI
4786 || *previous == OP_NOT || *previous == OP_NOTI)
4787 {
4788 switch (*previous)
4789 {
4790 default: /* Make compiler happy. */
4791 case OP_CHAR: op_type = OP_STAR - OP_STAR; break;
4792 case OP_CHARI: op_type = OP_STARI - OP_STAR; break;
4793 case OP_NOT: op_type = OP_NOTSTAR - OP_STAR; break;
4794 case OP_NOTI: op_type = OP_NOTSTARI - OP_STAR; break;
4795 }
4796
4797 /* Deal with UTF characters that take up more than one character. It's
4798 easier to write this out separately than try to macrify it. Use c to
4799 hold the length of the character in bytes, plus UTF_LENGTH to flag that
4800 it's a length rather than a small character. */
4801
4802 #if defined SUPPORT_UTF && !defined COMPILE_PCRE32
4803 if (utf && NOT_FIRSTCHAR(code[-1]))
4804 {
4805 pcre_uchar *lastchar = code - 1;
4806 BACKCHAR(lastchar);
4807 c = (int)(code - lastchar); /* Length of UTF-8 character */
4808 memcpy(utf_chars, lastchar, IN_UCHARS(c)); /* Save the char */
4809 c |= UTF_LENGTH; /* Flag c as a length */
4810 }
4811 else
4812 #endif /* SUPPORT_UTF */
4813
4814 /* Handle the case of a single charater - either with no UTF support, or
4815 with UTF disabled, or for a single character UTF character. */
4816 {
4817 c = code[-1];
4818 if (*previous <= OP_CHARI && repeat_min > 1)
4819 reqchar = c | req_caseopt | cd->req_varyopt;
4820 }
4821
4822 /* If the repetition is unlimited, it pays to see if the next thing on
4823 the line is something that cannot possibly match this character. If so,
4824 automatically possessifying this item gains some performance in the case
4825 where the match fails. */
4826
4827 if (!possessive_quantifier &&
4828 repeat_max < 0 &&
4829 check_auto_possessive(previous, utf, ptr + 1, options, cd))
4830 {
4831 repeat_type = 0; /* Force greedy */
4832 possessive_quantifier = TRUE;
4833 }
4834
4835 goto OUTPUT_SINGLE_REPEAT; /* Code shared with single character types */
4836 }
4837
4838 /* If previous was a character type match (\d or similar), abolish it and
4839 create a suitable repeat item. The code is shared with single-character
4840 repeats by setting op_type to add a suitable offset into repeat_type. Note
4841 the the Unicode property types will be present only when SUPPORT_UCP is
4842 defined, but we don't wrap the little bits of code here because it just
4843 makes it horribly messy. */
4844
4845 else if (*previous < OP_EODN)
4846 {
4847 pcre_uchar *oldcode;
4848 int prop_type, prop_value;
4849 op_type = OP_TYPESTAR - OP_STAR; /* Use type opcodes */
4850 c = *previous;
4851
4852 if (!possessive_quantifier &&
4853 repeat_max < 0 &&
4854 check_auto_possessive(previous, utf, ptr + 1, options, cd))
4855 {
4856 repeat_type = 0; /* Force greedy */
4857 possessive_quantifier = TRUE;
4858 }
4859
4860 OUTPUT_SINGLE_REPEAT:
4861 if (*previous == OP_PROP || *previous == OP_NOTPROP)
4862 {
4863 prop_type = previous[1];
4864 prop_value = previous[2];
4865 }
4866 else prop_type = prop_value = -1;
4867
4868 oldcode = code;
4869 code = previous; /* Usually overwrite previous item */
4870
4871 /* If the maximum is zero then the minimum must also be zero; Perl allows
4872 this case, so we do too - by simply omitting the item altogether. */
4873
4874 if (repeat_max == 0) goto END_REPEAT;
4875
4876 /*--------------------------------------------------------------------*/
4877 /* This code is obsolete from release 8.00; the restriction was finally
4878 removed: */
4879
4880 /* All real repeats make it impossible to handle partial matching (maybe
4881 one day we will be able to remove this restriction). */
4882
4883 /* if (repeat_max != 1) cd->external_flags |= PCRE_NOPARTIAL; */
4884 /*--------------------------------------------------------------------*/
4885
4886 /* Combine the op_type with the repeat_type */
4887
4888 repeat_type += op_type;
4889
4890 /* A minimum of zero is handled either as the special case * or ?, or as
4891 an UPTO, with the maximum given. */
4892
4893 if (repeat_min == 0)
4894 {
4895 if (repeat_max == -1) *code++ = OP_STAR + repeat_type;
4896 else if (repeat_max == 1) *code++ = OP_QUERY + repeat_type;
4897 else
4898 {
4899 *code++ = OP_UPTO + repeat_type;
4900 PUT2INC(code, 0, repeat_max);
4901 }
4902 }
4903
4904 /* A repeat minimum of 1 is optimized into some special cases. If the
4905 maximum is unlimited, we use OP_PLUS. Otherwise, the original item is
4906 left in place and, if the maximum is greater than 1, we use OP_UPTO with
4907 one less than the maximum. */
4908
4909 else if (repeat_min == 1)
4910 {
4911 if (repeat_max == -1)
4912 *code++ = OP_PLUS + repeat_type;
4913 else
4914 {
4915 code = oldcode; /* leave previous item in place */
4916 if (repeat_max == 1) goto END_REPEAT;
4917 *code++ = OP_UPTO + repeat_type;
4918 PUT2INC(code, 0, repeat_max - 1);
4919 }
4920 }
4921
4922 /* The case {n,n} is just an EXACT, while the general case {n,m} is
4923 handled as an EXACT followed by an UPTO. */
4924
4925 else
4926 {
4927 *code++ = OP_EXACT + op_type; /* NB EXACT doesn't have repeat_type */
4928 PUT2INC(code, 0, repeat_min);
4929
4930 /* If the maximum is unlimited, insert an OP_STAR. Before doing so,
4931 we have to insert the character for the previous code. For a repeated
4932 Unicode property match, there are two extra bytes that define the
4933 required property. In UTF-8 mode, long characters have their length in
4934 c, with the UTF_LENGTH bit as a flag. */
4935
4936 if (repeat_max < 0)
4937 {
4938 #if defined SUPPORT_UTF && !defined COMPILE_PCRE32
4939 if (utf && (c & UTF_LENGTH) != 0)
4940 {
4941 memcpy(code, utf_chars, IN_UCHARS(c & 7));
4942 code += c & 7;
4943 }
4944 else
4945 #endif
4946 {
4947 *code++ = c;
4948 if (prop_type >= 0)
4949 {
4950 *code++ = prop_type;
4951 *code++ = prop_value;
4952 }
4953 }
4954 *code++ = OP_STAR + repeat_type;
4955 }
4956
4957 /* Else insert an UPTO if the max is greater than the min, again
4958 preceded by the character, for the previously inserted code. If the
4959 UPTO is just for 1 instance, we can use QUERY instead. */
4960
4961 else if (repeat_max != repeat_min)
4962 {
4963 #if defined SUPPORT_UTF && !defined COMPILE_PCRE32
4964 if (utf && (c & UTF_LENGTH) != 0)
4965 {
4966 memcpy(code, utf_chars, IN_UCHARS(c & 7));
4967 code += c & 7;
4968 }
4969 else
4970 #endif
4971 *code++ = c;
4972 if (prop_type >= 0)
4973 {
4974 *code++ = prop_type;
4975 *code++ = prop_value;
4976 }
4977 repeat_max -= repeat_min;
4978
4979 if (repeat_max == 1)
4980 {
4981 *code++ = OP_QUERY + repeat_type;
4982 }
4983 else
4984 {
4985 *code++ = OP_UPTO + repeat_type;
4986 PUT2INC(code, 0, repeat_max);
4987 }
4988 }
4989 }
4990
4991 /* The character or character type itself comes last in all cases. */
4992
4993 #if defined SUPPORT_UTF && !defined COMPILE_PCRE32
4994 if (utf && (c & UTF_LENGTH) != 0)
4995 {
4996 memcpy(code, utf_chars, IN_UCHARS(c & 7));
4997 code += c & 7;
4998 }
4999 else
5000 #endif
5001 *code++ = c;
5002
5003 /* For a repeated Unicode property match, there are two extra bytes that
5004 define the required property. */
5005
5006 #ifdef SUPPORT_UCP
5007 if (prop_type >= 0)
5008 {
5009 *code++ = prop_type;
5010 *code++ = prop_value;
5011 }
5012 #endif
5013 }
5014
5015 /* If previous was a character class or a back reference, we put the repeat
5016 stuff after it, but just skip the item if the repeat was {0,0}. */
5017
5018 else if (*previous == OP_CLASS ||
5019 *previous == OP_NCLASS ||
5020 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
5021 *previous == OP_XCLASS ||
5022 #endif
5023 *previous == OP_REF ||
5024 *previous == OP_REFI)
5025 {
5026 if (repeat_max == 0)
5027 {
5028 code = previous;
5029 goto END_REPEAT;
5030 }
5031
5032 /*--------------------------------------------------------------------*/
5033 /* This code is obsolete from release 8.00; the restriction was finally
5034 removed: */
5035
5036 /* All real repeats make it impossible to handle partial matching (maybe
5037 one day we will be able to remove this restriction). */
5038
5039 /* if (repeat_max != 1) cd->external_flags |= PCRE_NOPARTIAL; */
5040 /*--------------------------------------------------------------------*/
5041
5042 if (repeat_min == 0 && repeat_max == -1)
5043 *code++ = OP_CRSTAR + repeat_type;
5044 else if (repeat_min == 1 && repeat_max == -1)
5045 *code++ = OP_CRPLUS + repeat_type;
5046 else if (repeat_min == 0 && repeat_max == 1)
5047 *code++ = OP_CRQUERY + repeat_type;
5048 else
5049 {
5050 *code++ = OP_CRRANGE + repeat_type;
5051 PUT2INC(code, 0, repeat_min);
5052 if (repeat_max == -1) repeat_max = 0; /* 2-byte encoding for max */
5053 PUT2INC(code, 0, repeat_max);
5054 }
5055 }
5056
5057 /* If previous was a bracket group, we may have to replicate it in certain
5058 cases. Note that at this point we can encounter only the "basic" bracket
5059 opcodes such as BRA and CBRA, as this is the place where they get converted
5060 into the more special varieties such as BRAPOS and SBRA. A test for >=
5061 OP_ASSERT and <= OP_COND includes ASSERT, ASSERT_NOT, ASSERTBACK,
5062 ASSERTBACK_NOT, ONCE, BRA, CBRA, and COND. Originally, PCRE did not allow
5063 repetition of assertions, but now it does, for Perl compatibility. */
5064
5065 else if (*previous >= OP_ASSERT && *previous <= OP_COND)
5066 {
5067 register int i;
5068 int len = (int)(code - previous);
5069 pcre_uchar *bralink = NULL;
5070 pcre_uchar *brazeroptr = NULL;
5071
5072 /* Repeating a DEFINE group is pointless, but Perl allows the syntax, so
5073 we just ignore the repeat. */
5074
5075 if (*previous == OP_COND && previous[LINK_SIZE+1] == OP_DEF)
5076 goto END_REPEAT;
5077
5078 /* There is no sense in actually repeating assertions. The only potential
5079 use of repetition is in cases when the assertion is optional. Therefore,
5080 if the minimum is greater than zero, just ignore the repeat. If the
5081 maximum is not not zero or one, set it to 1. */
5082
5083 if (*previous < OP_ONCE) /* Assertion */
5084 {
5085 if (repeat_min > 0) goto END_REPEAT;
5086 if (repeat_max < 0 || repeat_max > 1) repeat_max = 1;
5087 }
5088
5089 /* The case of a zero minimum is special because of the need to stick
5090 OP_BRAZERO in front of it, and because the group appears once in the
5091 data, whereas in other cases it appears the minimum number of times. For
5092 this reason, it is simplest to treat this case separately, as otherwise
5093 the code gets far too messy. There are several special subcases when the
5094 minimum is zero. */
5095
5096 if (repeat_min == 0)
5097 {
5098 /* If the maximum is also zero, we used to just omit the group from the
5099 output altogether, like this:
5100
5101 ** if (repeat_max == 0)
5102 ** {
5103 ** code = previous;
5104 ** goto END_REPEAT;
5105 ** }
5106
5107 However, that fails when a group or a subgroup within it is referenced
5108 as a subroutine from elsewhere in the pattern, so now we stick in
5109 OP_SKIPZERO in front of it so that it is skipped on execution. As we
5110 don't have a list of which groups are referenced, we cannot do this
5111 selectively.
5112
5113 If the maximum is 1 or unlimited, we just have to stick in the BRAZERO
5114 and do no more at this point. However, we do need to adjust any
5115 OP_RECURSE calls inside the group that refer to the group itself or any
5116 internal or forward referenced group, because the offset is from the
5117 start of the whole regex. Temporarily terminate the pattern while doing
5118 this. */
5119
5120 if (repeat_max <= 1) /* Covers 0, 1, and unlimited */
5121 {
5122 *code = OP_END;
5123 adjust_recurse(previous, 1, utf, cd, save_hwm);
5124 memmove(previous + 1, previous, IN_UCHARS(len));
5125 code++;
5126 if (repeat_max == 0)
5127 {
5128 *previous++ = OP_SKIPZERO;
5129 goto END_REPEAT;
5130 }
5131 brazeroptr = previous; /* Save for possessive optimizing */
5132 *previous++ = OP_BRAZERO + repeat_type;
5133 }
5134
5135 /* If the maximum is greater than 1 and limited, we have to replicate
5136 in a nested fashion, sticking OP_BRAZERO before each set of brackets.
5137 The first one has to be handled carefully because it's the original
5138 copy, which has to be moved up. The remainder can be handled by code
5139 that is common with the non-zero minimum case below. We have to
5140 adjust the value or repeat_max, since one less copy is required. Once
5141 again, we may have to adjust any OP_RECURSE calls inside the group. */
5142
5143 else
5144 {
5145 int offset;
5146 *code = OP_END;
5147 adjust_recurse(previous, 2 + LINK_SIZE, utf, cd, save_hwm);
5148 memmove(previous + 2 + LINK_SIZE, previous, IN_UCHARS(len));
5149 code += 2 + LINK_SIZE;
5150 *previous++ = OP_BRAZERO + repeat_type;
5151 *previous++ = OP_BRA;
5152
5153 /* We chain together the bracket offset fields that have to be
5154 filled in later when the ends of the brackets are reached. */
5155
5156 offset = (bralink == NULL)? 0 : (int)(previous - bralink);
5157 bralink = previous;
5158 PUTINC(previous, 0, offset);
5159 }
5160
5161 repeat_max--;
5162 }
5163
5164 /* If the minimum is greater than zero, replicate the group as many
5165 times as necessary, and adjust the maximum to the number of subsequent
5166 copies that we need. If we set a first char from the group, and didn't
5167 set a required char, copy the latter from the former. If there are any
5168 forward reference subroutine calls in the group, there will be entries on
5169 the workspace list; replicate these with an appropriate increment. */
5170
5171 else
5172 {
5173 if (repeat_min > 1)
5174 {
5175 /* In the pre-compile phase, we don't actually do the replication. We
5176 just adjust the length as if we had. Do some paranoid checks for
5177 potential integer overflow. The INT64_OR_DOUBLE type is a 64-bit
5178 integer type when available, otherwise double. */
5179
5180 if (lengthptr != NULL)
5181 {
5182 int delta = (repeat_min - 1)*length_prevgroup;
5183 if ((INT64_OR_DOUBLE)(repeat_min - 1)*
5184 (INT64_OR_DOUBLE)length_prevgroup >
5185 (INT64_OR_DOUBLE)INT_MAX ||
5186 OFLOW_MAX - *lengthptr < delta)
5187 {
5188 *errorcodeptr = ERR20;
5189 goto FAILED;
5190 }
5191 *lengthptr += delta;
5192 }
5193
5194 /* This is compiling for real. If there is a set first byte for
5195 the group, and we have not yet set a "required byte", set it. Make
5196 sure there is enough workspace for copying forward references before
5197 doing the copy. */
5198
5199 else
5200 {
5201 if (groupsetfirstchar && reqchar < 0) reqchar = firstchar;
5202
5203 for (i = 1; i < repeat_min; i++)
5204 {
5205 pcre_uchar *hc;
5206 pcre_uchar *this_hwm = cd->hwm;
5207 memcpy(code, previous, IN_UCHARS(len));
5208
5209 while (cd->hwm > cd->start_workspace + cd->workspace_size -
5210 WORK_SIZE_SAFETY_MARGIN - (this_hwm - save_hwm))
5211 {
5212 int save_offset = save_hwm - cd->start_workspace;
5213 int this_offset = this_hwm - cd->start_workspace;
5214 *errorcodeptr = expand_workspace(cd);
5215 if (*errorcodeptr != 0) goto FAILED;
5216 save_hwm = (pcre_uchar *)cd->start_workspace + save_offset;
5217 this_hwm = (pcre_uchar *)cd->start_workspace + this_offset;
5218 }
5219
5220 for (hc = save_hwm; hc < this_hwm; hc += LINK_SIZE)
5221 {
5222 PUT(cd->hwm, 0, GET(hc, 0) + len);
5223 cd->hwm += LINK_SIZE;
5224 }
5225 save_hwm = this_hwm;
5226 code += len;
5227 }
5228 }
5229 }
5230
5231 if (repeat_max > 0) repeat_max -= repeat_min;
5232 }
5233
5234 /* This code is common to both the zero and non-zero minimum cases. If
5235 the maximum is limited, it replicates the group in a nested fashion,
5236 remembering the bracket starts on a stack. In the case of a zero minimum,
5237 the first one was set up above. In all cases the repeat_max now specifies
5238 the number of additional copies needed. Again, we must remember to
5239 replicate entries on the forward reference list. */
5240
5241 if (repeat_max >= 0)
5242 {
5243 /* In the pre-compile phase, we don't actually do the replication. We
5244 just adjust the length as if we had. For each repetition we must add 1
5245 to the length for BRAZERO and for all but the last repetition we must
5246 add 2 + 2*LINKSIZE to allow for the nesting that occurs. Do some
5247 paranoid checks to avoid integer overflow. The INT64_OR_DOUBLE type is
5248 a 64-bit integer type when available, otherwise double. */
5249
5250 if (lengthptr != NULL && repeat_max > 0)
5251 {
5252 int delta = repeat_max * (length_prevgroup + 1 + 2 + 2*LINK_SIZE) -
5253 2 - 2*LINK_SIZE; /* Last one doesn't nest */
5254 if ((INT64_OR_DOUBLE)repeat_max *
5255 (INT64_OR_DOUBLE)(length_prevgroup + 1 + 2 + 2*LINK_SIZE)
5256 > (INT64_OR_DOUBLE)INT_MAX ||
5257 OFLOW_MAX - *lengthptr < delta)
5258 {
5259 *errorcodeptr = ERR20;
5260 goto FAILED;
5261 }
5262 *lengthptr += delta;
5263 }
5264
5265 /* This is compiling for real */
5266
5267 else for (i = repeat_max - 1; i >= 0; i--)
5268 {
5269 pcre_uchar *hc;
5270 pcre_uchar *this_hwm = cd->hwm;
5271
5272 *code++ = OP_BRAZERO + repeat_type;
5273
5274 /* All but the final copy start a new nesting, maintaining the
5275 chain of brackets outstanding. */
5276
5277 if (i != 0)
5278 {
5279 int offset;
5280 *code++ = OP_BRA;
5281 offset = (bralink == NULL)? 0 : (int)(code - bralink);
5282 bralink = code;
5283 PUTINC(code, 0, offset);
5284 }
5285
5286 memcpy(code, previous, IN_UCHARS(len));
5287
5288 /* Ensure there is enough workspace for forward references before
5289 copying them. */
5290
5291 while (cd->hwm > cd->start_workspace + cd->workspace_size -
5292 WORK_SIZE_SAFETY_MARGIN - (this_hwm - save_hwm))
5293 {
5294 int save_offset = save_hwm - cd->start_workspace;
5295 int this_offset = this_hwm - cd->start_workspace;
5296 *errorcodeptr = expand_workspace(cd);
5297 if (*errorcodeptr != 0) goto FAILED;
5298 save_hwm = (pcre_uchar *)cd->start_workspace + save_offset;
5299 this_hwm = (pcre_uchar *)cd->start_workspace + this_offset;
5300 }
5301
5302 for (hc = save_hwm; hc < this_hwm; hc += LINK_SIZE)
5303 {
5304 PUT(cd->hwm, 0, GET(hc, 0) + len + ((i != 0)? 2+LINK_SIZE : 1));
5305 cd->hwm += LINK_SIZE;
5306 }
5307 save_hwm = this_hwm;
5308 code += len;
5309 }
5310
5311 /* Now chain through the pending brackets, and fill in their length
5312 fields (which are holding the chain links pro tem). */
5313
5314 while (bralink != NULL)
5315 {
5316 int oldlinkoffset;
5317 int offset = (int)(code - bralink + 1);
5318 pcre_uchar *bra = code - offset;
5319 oldlinkoffset = GET(bra, 1);
5320 bralink = (oldlinkoffset == 0)? NULL : bralink - oldlinkoffset;
5321 *code++ = OP_KET;
5322 PUTINC(code, 0, offset);
5323 PUT(bra, 1, offset);
5324 }
5325 }
5326
5327 /* If the maximum is unlimited, set a repeater in the final copy. For
5328 ONCE brackets, that's all we need to do. However, possessively repeated
5329 ONCE brackets can be converted into non-capturing brackets, as the
5330 behaviour of (?:xx)++ is the same as (?>xx)++ and this saves having to
5331 deal with possessive ONCEs specially.
5332
5333 Otherwise, when we are doing the actual compile phase, check to see
5334 whether this group is one that could match an empty string. If so,
5335 convert the initial operator to the S form (e.g. OP_BRA -> OP_SBRA) so
5336 that runtime checking can be done. [This check is also applied to ONCE
5337 groups at runtime, but in a different way.]
5338
5339 Then, if the quantifier was possessive and the bracket is not a
5340 conditional, we convert the BRA code to the POS form, and the KET code to
5341 KETRPOS. (It turns out to be convenient at runtime to detect this kind of
5342 subpattern at both the start and at the end.) The use of special opcodes
5343 makes it possible to reduce greatly the stack usage in pcre_exec(). If
5344 the group is preceded by OP_BRAZERO, convert this to OP_BRAPOSZERO.
5345
5346 Then, if the minimum number of matches is 1 or 0, cancel the possessive
5347 flag so that the default action below, of wrapping everything inside
5348 atomic brackets, does not happen. When the minimum is greater than 1,
5349 there will be earlier copies of the group, and so we still have to wrap
5350 the whole thing. */
5351
5352 else
5353 {
5354 pcre_uchar *ketcode = code - 1 - LINK_SIZE;
5355 pcre_uchar *bracode = ketcode - GET(ketcode, 1);
5356
5357 /* Convert possessive ONCE brackets to non-capturing */
5358
5359 if ((*bracode == OP_ONCE || *bracode == OP_ONCE_NC) &&
5360 possessive_quantifier) *bracode = OP_BRA;
5361
5362 /* For non-possessive ONCE brackets, all we need to do is to
5363 set the KET. */
5364
5365 if (*bracode == OP_ONCE || *bracode == OP_ONCE_NC)
5366 *ketcode = OP_KETRMAX + repeat_type;
5367
5368 /* Handle non-ONCE brackets and possessive ONCEs (which have been
5369 converted to non-capturing above). */
5370
5371 else
5372 {
5373 /* In the compile phase, check for empty string matching. */
5374
5375 if (lengthptr == NULL)
5376 {
5377 pcre_uchar *scode = bracode;
5378 do
5379 {
5380 if (could_be_empty_branch(scode, ketcode, utf, cd))
5381 {
5382 *bracode += OP_SBRA - OP_BRA;
5383 break;
5384 }
5385 scode += GET(scode, 1);
5386 }
5387 while (*scode == OP_ALT);
5388 }
5389
5390 /* Handle possessive quantifiers. */
5391
5392 if (possessive_quantifier)
5393 {
5394 /* For COND brackets, we wrap the whole thing in a possessively
5395 repeated non-capturing bracket, because we have not invented POS
5396 versions of the COND opcodes. Because we are moving code along, we
5397 must ensure that any pending recursive references are updated. */
5398
5399 if (*bracode == OP_COND || *bracode == OP_SCOND)
5400 {
5401 int nlen = (int)(code - bracode);
5402 *code = OP_END;
5403 adjust_recurse(bracode, 1 + LINK_SIZE, utf, cd, save_hwm);
5404 memmove(bracode + 1 + LINK_SIZE, bracode, IN_UCHARS(nlen));
5405 code += 1 + LINK_SIZE;
5406 nlen += 1 + LINK_SIZE;
5407 *bracode = OP_BRAPOS;
5408 *code++ = OP_KETRPOS;
5409 PUTINC(code, 0, nlen);
5410 PUT(bracode, 1, nlen);
5411 }
5412
5413 /* For non-COND brackets, we modify the BRA code and use KETRPOS. */
5414
5415 else
5416 {
5417 *bracode += 1; /* Switch to xxxPOS opcodes */
5418 *ketcode = OP_KETRPOS;
5419 }
5420
5421 /* If the minimum is zero, mark it as possessive, then unset the
5422 possessive flag when the minimum is 0 or 1. */
5423
5424 if (brazeroptr != NULL) *brazeroptr = OP_BRAPOSZERO;
5425 if (repeat_min < 2) possessive_quantifier = FALSE;
5426 }
5427
5428 /* Non-possessive quantifier */
5429
5430 else *ketcode = OP_KETRMAX + repeat_type;
5431 }
5432 }
5433 }
5434
5435 /* If previous is OP_FAIL, it was generated by an empty class [] in
5436 JavaScript mode. The other ways in which OP_FAIL can be generated, that is
5437 by (*FAIL) or (?!) set previous to NULL, which gives a "nothing to repeat"
5438 error above. We can just ignore the repeat in JS case. */
5439
5440 else if (*previous == OP_FAIL) goto END_REPEAT;
5441
5442 /* Else there's some kind of shambles */
5443
5444 else
5445 {
5446 *errorcodeptr = ERR11;
5447 goto FAILED;
5448 }
5449
5450 /* If the character following a repeat is '+', or if certain optimization
5451 tests above succeeded, possessive_quantifier is TRUE. For some opcodes,
5452 there are special alternative opcodes for this case. For anything else, we
5453 wrap the entire repeated item inside OP_ONCE brackets. Logically, the '+'
5454 notation is just syntactic sugar, taken from Sun's Java package, but the
5455 special opcodes can optimize it.
5456
5457 Some (but not all) possessively repeated subpatterns have already been
5458 completely handled in the code just above. For them, possessive_quantifier
5459 is always FALSE at this stage.
5460
5461 Note that the repeated item starts at tempcode, not at previous, which
5462 might be the first part of a string whose (former) last char we repeated.
5463
5464 Possessifying an 'exact' quantifier has no effect, so we can ignore it. But
5465 an 'upto' may follow. We skip over an 'exact' item, and then test the
5466 length of what remains before proceeding. */
5467
5468 if (possessive_quantifier)
5469 {
5470 int len;
5471
5472 if (*tempcode == OP_TYPEEXACT)
5473 tempcode += PRIV(OP_lengths)[*tempcode] +
5474 ((tempcode[1 + IMM2_SIZE] == OP_PROP
5475 || tempcode[1 + IMM2_SIZE] == OP_NOTPROP)? 2 : 0);
5476
5477 else if (*tempcode == OP_EXACT || *tempcode == OP_NOTEXACT)
5478 {
5479 tempcode += PRIV(OP_lengths)[*tempcode];
5480 #if defined SUPPORT_UTF && !defined COMPILE_PCRE32
5481 if (utf && HAS_EXTRALEN(tempcode[-1]))
5482 tempcode += GET_EXTRALEN(tempcode[-1]);
5483 #endif
5484 }
5485
5486 len = (int)(code - tempcode);
5487 if (len > 0) switch (*tempcode)
5488 {
5489 case OP_STAR: *tempcode = OP_POSSTAR; break;
5490 case OP_PLUS: *tempcode = OP_POSPLUS; break;
5491 case OP_QUERY: *tempcode = OP_POSQUERY; break;
5492 case OP_UPTO: *tempcode = OP_POSUPTO; break;
5493
5494 case OP_STARI: *tempcode = OP_POSSTARI; break;
5495 case OP_PLUSI: *tempcode = OP_POSPLUSI; break;
5496 case OP_QUERYI: *tempcode = OP_POSQUERYI; break;
5497 case OP_UPTOI: *tempcode = OP_POSUPTOI; break;
5498
5499 case OP_NOTSTAR: *tempcode = OP_NOTPOSSTAR; break;
5500 case OP_NOTPLUS: *tempcode = OP_NOTPOSPLUS; break;
5501 case OP_NOTQUERY: *tempcode = OP_NOTPOSQUERY; break;
5502 case OP_NOTUPTO: *tempcode = OP_NOTPOSUPTO; break;
5503
5504 case OP_NOTSTARI: *tempcode = OP_NOTPOSSTARI; break;
5505 case OP_NOTPLUSI: *tempcode = OP_NOTPOSPLUSI; break;
5506 case OP_NOTQUERYI: *tempcode = OP_NOTPOSQUERYI; break;
5507 case OP_NOTUPTOI: *tempcode = OP_NOTPOSUPTOI; break;
5508
5509 case OP_TYPESTAR: *tempcode = OP_TYPEPOSSTAR; break;
5510 case OP_TYPEPLUS: *tempcode = OP_TYPEPOSPLUS; break;
5511 case OP_TYPEQUERY: *tempcode = OP_TYPEPOSQUERY; break;
5512 case OP_TYPEUPTO: *tempcode = OP_TYPEPOSUPTO; break;
5513
5514 /* Because we are moving code along, we must ensure that any
5515 pending recursive references are updated. */
5516
5517 default:
5518 *code = OP_END;
5519 adjust_recurse(tempcode, 1 + LINK_SIZE, utf, cd, save_hwm);
5520 memmove(tempcode + 1 + LINK_SIZE, tempcode, IN_UCHARS(len));
5521 code += 1 + LINK_SIZE;
5522 len += 1 + LINK_SIZE;
5523 tempcode[0] = OP_ONCE;
5524 *code++ = OP_KET;
5525 PUTINC(code, 0, len);
5526 PUT(tempcode, 1, len);
5527 break;
5528 }
5529 }
5530
5531 /* In all case we no longer have a previous item. We also set the
5532 "follows varying string" flag for subsequently encountered reqchars if
5533 it isn't already set and we have just passed a varying length item. */
5534
5535 END_REPEAT:
5536 previous = NULL;
5537 cd->req_varyopt |= reqvary;
5538 break;
5539
5540
5541 /* ===================================================================*/
5542 /* Start of nested parenthesized sub-expression, or comment or lookahead or
5543 lookbehind or option setting or condition or all the other extended
5544 parenthesis forms. */
5545
5546 case CHAR_LEFT_PARENTHESIS:
5547 newoptions = options;
5548 skipbytes = 0;
5549 bravalue = OP_CBRA;
5550 save_hwm = cd->hwm;
5551 reset_bracount = FALSE;
5552
5553 /* First deal with various "verbs" that can be introduced by '*'. */
5554
5555 ptr++;
5556 if (ptr[0] == CHAR_ASTERISK && (ptr[1] == ':'
5557 || (MAX_255(ptr[1]) && ((cd->ctypes[ptr[1]] & ctype_letter) != 0))))
5558 {
5559 int i, namelen;
5560 int arglen = 0;
5561 const char *vn = verbnames;
5562 const pcre_uchar *name = ptr + 1;
5563 const pcre_uchar *arg = NULL;
5564 previous = NULL;
5565 ptr++;
5566 while (MAX_255(*ptr) && (cd->ctypes[*ptr] & ctype_letter) != 0) ptr++;
5567 namelen = (int)(ptr - name);
5568
5569 /* It appears that Perl allows any characters whatsoever, other than
5570 a closing parenthesis, to appear in arguments, so we no longer insist on
5571 letters, digits, and underscores. */
5572
5573 if (*ptr == CHAR_COLON)
5574 {
5575 arg = ++ptr;
5576 while (*ptr != 0 && *ptr != CHAR_RIGHT_PARENTHESIS) ptr++;
5577 arglen = (int)(ptr - arg);
5578 if ((unsigned int)arglen > MAX_MARK)
5579 {
5580 *errorcodeptr = ERR75;
5581 goto FAILED;
5582 }
5583 }
5584
5585 if (*ptr != CHAR_RIGHT_PARENTHESIS)
5586 {
5587 *errorcodeptr = ERR60;
5588 goto FAILED;
5589 }
5590
5591 /* Scan the table of verb names */
5592
5593 for (i = 0; i < verbcount; i++)
5594 {
5595 if (namelen == verbs[i].len &&
5596 STRNCMP_UC_C8(name, vn, namelen) == 0)
5597 {
5598 int setverb;
5599
5600 /* Check for open captures before ACCEPT and convert it to
5601 ASSERT_ACCEPT if in an assertion. */
5602
5603 if (verbs[i].op == OP_ACCEPT)
5604 {
5605 open_capitem *oc;
5606 if (arglen != 0)
5607 {
5608 *errorcodeptr = ERR59;
5609 goto FAILED;
5610 }
5611 cd->had_accept = TRUE;
5612 for (oc = cd->open_caps; oc != NULL; oc = oc->next)
5613 {
5614 *code++ = OP_CLOSE;
5615 PUT2INC(code, 0, oc->number);
5616 }
5617 setverb = *code++ =
5618 (cd->assert_depth > 0)? OP_ASSERT_ACCEPT : OP_ACCEPT;
5619
5620 /* Do not set firstchar after *ACCEPT */
5621 if (firstchar == REQ_UNSET) firstchar = REQ_NONE;
5622 }
5623
5624 /* Handle other cases with/without an argument */
5625
5626 else if (arglen == 0)
5627 {
5628 if (verbs[i].op < 0) /* Argument is mandatory */
5629 {
5630 *errorcodeptr = ERR66;
5631 goto FAILED;
5632 }
5633 setverb = *code++ = verbs[i].op;
5634 }
5635
5636 else
5637 {
5638 if (verbs[i].op_arg < 0) /* Argument is forbidden */
5639 {
5640 *errorcodeptr = ERR59;
5641 goto FAILED;
5642 }
5643 setverb = *code++ = verbs[i].op_arg;
5644 *code++ = arglen;
5645 memcpy(code, arg, IN_UCHARS(arglen));
5646 code += arglen;
5647 *code++ = 0;
5648 }
5649
5650 switch (setverb)
5651 {
5652 case OP_THEN:
5653 case OP_THEN_ARG:
5654 cd->external_flags |= PCRE_HASTHEN;
5655 break;
5656
5657 case OP_PRUNE:
5658 case OP_PRUNE_ARG:
5659 case OP_SKIP:
5660 case OP_SKIP_ARG:
5661 cd->had_pruneorskip = TRUE;
5662 break;
5663 }
5664
5665 break; /* Found verb, exit loop */
5666 }
5667
5668 vn += verbs[i].len + 1;
5669 }
5670
5671 if (i < verbcount) continue; /* Successfully handled a verb */
5672 *errorcodeptr = ERR60; /* Verb not recognized */
5673 goto FAILED;
5674 }
5675
5676 /* Deal with the extended parentheses; all are introduced by '?', and the
5677 appearance of any of them means that this is not a capturing group. */
5678
5679 else if (*ptr == CHAR_QUESTION_MARK)
5680 {
5681 int i, set, unset, namelen;
5682 int *optset;
5683 const pcre_uchar *name;
5684 pcre_uchar *slot;
5685
5686 switch (*(++ptr))
5687 {
5688 case CHAR_NUMBER_SIGN: /* Comment; skip to ket */
5689 ptr++;
5690 while (*ptr != 0 && *ptr != CHAR_RIGHT_PARENTHESIS) ptr++;
5691 if (*ptr == 0)
5692 {
5693 *errorcodeptr = ERR18;
5694 goto FAILED;
5695 }
5696 continue;
5697
5698
5699 /* ------------------------------------------------------------ */
5700 case CHAR_VERTICAL_LINE: /* Reset capture count for each branch */
5701 reset_bracount = TRUE;
5702 /* Fall through */
5703
5704 /* ------------------------------------------------------------ */
5705 case CHAR_COLON: /* Non-capturing bracket */
5706 bravalue = OP_BRA;
5707 ptr++;
5708 break;
5709
5710
5711 /* ------------------------------------------------------------ */
5712 case CHAR_LEFT_PARENTHESIS:
5713 bravalue = OP_COND; /* Conditional group */
5714
5715 /* A condition can be an assertion, a number (referring to a numbered
5716 group), a name (referring to a named group), or 'R', referring to
5717 recursion. R<digits> and R&name are also permitted for recursion tests.
5718
5719 There are several syntaxes for testing a named group: (?(name)) is used
5720 by Python; Perl 5.10 onwards uses (?(<name>) or (?('name')).
5721
5722 There are two unfortunate ambiguities, caused by history. (a) 'R' can
5723 be the recursive thing or the name 'R' (and similarly for 'R' followed
5724 by digits), and (b) a number could be a name that consists of digits.
5725 In both cases, we look for a name first; if not found, we try the other
5726 cases. */
5727
5728 /* For conditions that are assertions, check the syntax, and then exit
5729 the switch. This will take control down to where bracketed groups,
5730 including assertions, are processed. */
5731
5732 if (ptr[1] == CHAR_QUESTION_MARK && (ptr[2] == CHAR_EQUALS_SIGN ||
5733 ptr[2] == CHAR_EXCLAMATION_MARK || ptr[2] == CHAR_LESS_THAN_SIGN))
5734 break;
5735
5736 /* Most other conditions use OP_CREF (a couple change to OP_RREF
5737 below), and all need to skip 1+IMM2_SIZE bytes at the start of the group. */
5738
5739 code[1+LINK_SIZE] = OP_CREF;
5740 skipbytes = 1+IMM2_SIZE;
5741 refsign = -1;
5742
5743 /* Check for a test for recursion in a named group. */
5744
5745 if (ptr[1] == CHAR_R && ptr[2] == CHAR_AMPERSAND)
5746 {
5747 terminator = -1;
5748 ptr += 2;
5749 code[1+LINK_SIZE] = OP_RREF; /* Change the type of test */
5750 }
5751
5752 /* Check for a test for a named group's having been set, using the Perl
5753 syntax (?(<name>) or (?('name') */
5754
5755 else if (ptr[1] == CHAR_LESS_THAN_SIGN)
5756 {
5757 terminator = CHAR_GREATER_THAN_SIGN;
5758 ptr++;
5759 }
5760 else if (ptr[1] == CHAR_APOSTROPHE)
5761 {
5762 terminator = CHAR_APOSTROPHE;
5763 ptr++;
5764 }
5765 else
5766 {
5767 terminator = 0;
5768 if (ptr[1] == CHAR_MINUS || ptr[1] == CHAR_PLUS) refsign = *(++ptr);
5769 }
5770
5771 /* We now expect to read a name; any thing else is an error */
5772
5773 if (!MAX_255(ptr[1]) || (cd->ctypes[ptr[1]] & ctype_word) == 0)
5774 {
5775 ptr += 1; /* To get the right offset */
5776 *errorcodeptr = ERR28;
5777 goto FAILED;
5778 }
5779
5780 /* Read the name, but also get it as a number if it's all digits */
5781
5782 recno = 0;
5783 name = ++ptr;
5784 while (MAX_255(*ptr) && (cd->ctypes[*ptr] & ctype_word) != 0)
5785 {
5786 if (recno >= 0)
5787 recno = (IS_DIGIT(*ptr))? recno * 10 + *ptr - CHAR_0 : -1;
5788 ptr++;
5789 }
5790 namelen = (int)(ptr - name);
5791
5792 if ((terminator > 0 && *ptr++ != terminator) ||
5793 *ptr++ != CHAR_RIGHT_PARENTHESIS)
5794 {
5795 ptr--; /* Error offset */
5796 *errorcodeptr = ERR26;
5797 goto FAILED;
5798 }
5799
5800 /* Do no further checking in the pre-compile phase. */
5801
5802 if (lengthptr != NULL) break;
5803
5804 /* In the real compile we do the work of looking for the actual
5805 reference. If the string started with "+" or "-" we require the rest to
5806 be digits, in which case recno will be set. */
5807
5808 if (refsign > 0)
5809 {
5810 if (recno <= 0)
5811 {
5812 *errorcodeptr = ERR58;
5813 goto FAILED;
5814 }
5815 recno = (refsign == CHAR_MINUS)?
5816 cd->bracount - recno + 1 : recno +cd->bracount;
5817 if (recno <= 0 || recno > cd->final_bracount)
5818 {
5819 *errorcodeptr = ERR15;
5820 goto FAILED;
5821 }
5822 PUT2(code, 2+LINK_SIZE, recno);
5823 break;
5824 }
5825
5826 /* Otherwise (did not start with "+" or "-"), start by looking for the
5827 name. If we find a name, add one to the opcode to change OP_CREF or
5828 OP_RREF into OP_NCREF or OP_NRREF. These behave exactly the same,
5829 except they record that the reference was originally to a name. The
5830 information is used to check duplicate names. */
5831
5832 slot = cd->name_table;
5833 for (i = 0; i < cd->names_found; i++)
5834 {
5835 if (STRNCMP_UC_UC(name, slot+IMM2_SIZE, namelen) == 0) break;
5836 slot += cd->name_entry_size;
5837 }
5838
5839 /* Found a previous named subpattern */
5840
5841 if (i < cd->names_found)
5842 {
5843 recno = GET2(slot, 0);
5844 PUT2(code, 2+LINK_SIZE, recno);
5845 code[1+LINK_SIZE]++;
5846 }
5847
5848 /* Search the pattern for a forward reference */
5849
5850 else if ((i = find_parens(cd, name, namelen,
5851 (options & PCRE_EXTENDED) != 0, utf)) > 0)
5852 {
5853 PUT2(code, 2+LINK_SIZE, i);
5854 code[1+LINK_SIZE]++;
5855 }
5856
5857 /* If terminator == 0 it means that the name followed directly after
5858 the opening parenthesis [e.g. (?(abc)...] and in this case there are
5859 some further alternatives to try. For the cases where terminator != 0
5860 [things like (?(<name>... or (?('name')... or (?(R&name)... ] we have
5861 now checked all the possibilities, so give an error. */
5862
5863 else if (terminator != 0)
5864 {
5865 *errorcodeptr = ERR15;
5866 goto FAILED;
5867 }
5868
5869 /* Check for (?(R) for recursion. Allow digits after R to specify a
5870 specific group number. */
5871
5872 else if (*name == CHAR_R)
5873 {
5874 recno = 0;
5875 for (i = 1; i < namelen; i++)
5876 {
5877 if (!IS_DIGIT(name[i]))
5878 {
5879 *errorcodeptr = ERR15;
5880 goto FAILED;
5881 }
5882 recno = recno * 10 + name[i] - CHAR_0;
5883 }
5884 if (recno == 0) recno = RREF_ANY;
5885 code[1+LINK_SIZE] = OP_RREF; /* Change test type */
5886 PUT2(code, 2+LINK_SIZE, recno);
5887 }
5888
5889 /* Similarly, check for the (?(DEFINE) "condition", which is always
5890 false. */
5891
5892 else if (namelen == 6 && STRNCMP_UC_C8(name, STRING_DEFINE, 6) == 0)
5893 {
5894 code[1+LINK_SIZE] = OP_DEF;
5895 skipbytes = 1;
5896 }
5897
5898 /* Check for the "name" actually being a subpattern number. We are
5899 in the second pass here, so final_bracount is set. */
5900
5901 else if (recno > 0 && recno <= cd->final_bracount)
5902 {
5903 PUT2(code, 2+LINK_SIZE, recno);
5904 }
5905
5906 /* Either an unidentified subpattern, or a reference to (?(0) */
5907
5908 else
5909 {
5910 *errorcodeptr = (recno == 0)? ERR35: ERR15;
5911 goto FAILED;
5912 }
5913 break;
5914
5915
5916 /* ------------------------------------------------------------ */
5917 case CHAR_EQUALS_SIGN: /* Positive lookahead */
5918 bravalue = OP_ASSERT;
5919 cd->assert_depth += 1;
5920 ptr++;
5921 break;
5922
5923
5924 /* ------------------------------------------------------------ */
5925 case CHAR_EXCLAMATION_MARK: /* Negative lookahead */
5926 ptr++;
5927 if (*ptr == CHAR_RIGHT_PARENTHESIS) /* Optimize (?!) */
5928 {
5929 *code++ = OP_FAIL;
5930 previous = NULL;
5931 continue;
5932 }
5933 bravalue = OP_ASSERT_NOT;
5934 cd->assert_depth += 1;
5935 break;
5936
5937
5938 /* ------------------------------------------------------------ */
5939 case CHAR_LESS_THAN_SIGN: /* Lookbehind or named define */
5940 switch (ptr[1])
5941 {
5942 case CHAR_EQUALS_SIGN: /* Positive lookbehind */
5943 bravalue = OP_ASSERTBACK;
5944 cd->assert_depth += 1;
5945 ptr += 2;
5946 break;
5947
5948 case CHAR_EXCLAMATION_MARK: /* Negative lookbehind */
5949 bravalue = OP_ASSERTBACK_NOT;
5950 cd->assert_depth += 1;
5951 ptr += 2;
5952 break;
5953
5954 default: /* Could be name define, else bad */
5955 if (MAX_255(ptr[1]) && (cd->ctypes[ptr[1]] & ctype_word) != 0)
5956 goto DEFINE_NAME;
5957 ptr++; /* Correct offset for error */
5958 *errorcodeptr = ERR24;
5959 goto FAILED;
5960 }
5961 break;
5962
5963
5964 /* ------------------------------------------------------------ */
5965 case CHAR_GREATER_THAN_SIGN: /* One-time brackets */
5966 bravalue = OP_ONCE;
5967 ptr++;
5968 break;
5969
5970
5971 /* ------------------------------------------------------------ */
5972 case CHAR_C: /* Callout - may be followed by digits; */
5973 previous_callout = code; /* Save for later completion */
5974 after_manual_callout = 1; /* Skip one item before completing */
5975 *code++ = OP_CALLOUT;
5976 {
5977 int n = 0;
5978 ptr++;
5979 while(IS_DIGIT(*ptr))
5980 n = n * 10 + *ptr++ - CHAR_0;
5981 if (*ptr != CHAR_RIGHT_PARENTHESIS)
5982 {
5983 *errorcodeptr = ERR39;
5984 goto FAILED;
5985 }
5986 if (n > 255)
5987 {
5988 *errorcodeptr = ERR38;
5989 goto FAILED;
5990 }
5991 *code++ = n;
5992 PUT(code, 0, (int)(ptr - cd->start_pattern + 1)); /* Pattern offset */
5993 PUT(code, LINK_SIZE, 0); /* Default length */
5994 code += 2 * LINK_SIZE;
5995 }
5996 previous = NULL;
5997 continue;
5998
5999
6000 /* ------------------------------------------------------------ */
6001 case CHAR_P: /* Python-style named subpattern handling */
6002 if (*(++ptr) == CHAR_EQUALS_SIGN ||
6003 *ptr == CHAR_GREATER_THAN_SIGN) /* Reference or recursion */
6004 {
6005 is_recurse = *ptr == CHAR_GREATER_THAN_SIGN;
6006 terminator = CHAR_RIGHT_PARENTHESIS;
6007 goto NAMED_REF_OR_RECURSE;
6008 }
6009 else if (*ptr != CHAR_LESS_THAN_SIGN) /* Test for Python-style defn */
6010 {
6011 *errorcodeptr = ERR41;
6012 goto FAILED;
6013 }
6014 /* Fall through to handle (?P< as (?< is handled */
6015
6016
6017 /* ------------------------------------------------------------ */
6018 DEFINE_NAME: /* Come here from (?< handling */
6019 case CHAR_APOSTROPHE:
6020 {
6021 terminator = (*ptr == CHAR_LESS_THAN_SIGN)?
6022 CHAR_GREATER_THAN_SIGN : CHAR_APOSTROPHE;
6023 name = ++ptr;
6024
6025 while (MAX_255(*ptr) && (cd->ctypes[*ptr] & ctype_word) != 0) ptr++;
6026 namelen = (int)(ptr - name);
6027
6028 /* In the pre-compile phase, just do a syntax check. */
6029
6030 if (lengthptr != NULL)
6031 {
6032 if (*ptr != terminator)
6033 {
6034 *errorcodeptr = ERR42;
6035 goto FAILED;
6036 }
6037 if (cd->names_found >= MAX_NAME_COUNT)
6038 {
6039 *errorcodeptr = ERR49;
6040 goto FAILED;
6041 }
6042 if (namelen + IMM2_SIZE + 1 > cd->name_entry_size)
6043 {
6044 cd->name_entry_size = namelen + IMM2_SIZE + 1;
6045 if (namelen > MAX_NAME_SIZE)
6046 {
6047 *errorcodeptr = ERR48;
6048 goto FAILED;
6049 }
6050 }
6051 }
6052
6053 /* In the real compile, create the entry in the table, maintaining
6054 alphabetical order. Duplicate names for different numbers are
6055 permitted only if PCRE_DUPNAMES is set. Duplicate names for the same
6056 number are always OK. (An existing number can be re-used if (?|
6057 appears in the pattern.) In either event, a duplicate name results in
6058 a duplicate entry in the table, even if the number is the same. This
6059 is because the number of names, and hence the table size, is computed
6060 in the pre-compile, and it affects various numbers and pointers which
6061 would all have to be modified, and the compiled code moved down, if
6062 duplicates with the same number were omitted from the table. This
6063 doesn't seem worth the hassle. However, *different* names for the
6064 same number are not permitted. */
6065
6066 else
6067 {
6068 BOOL dupname = FALSE;
6069 slot = cd->name_table;
6070
6071 for (i = 0; i < cd->names_found; i++)
6072 {
6073 int crc = memcmp(name, slot+IMM2_SIZE, IN_UCHARS(namelen));
6074 if (crc == 0)
6075 {
6076 if (slot[IMM2_SIZE+namelen] == 0)
6077 {
6078 if (GET2(slot, 0) != cd->bracount + 1 &&
6079 (options & PCRE_DUPNAMES) == 0)
6080 {
6081 *errorcodeptr = ERR43;
6082 goto FAILED;
6083 }
6084 else dupname = TRUE;
6085 }
6086 else crc = -1; /* Current name is a substring */
6087 }
6088
6089 /* Make space in the table and break the loop for an earlier
6090 name. For a duplicate or later name, carry on. We do this for
6091 duplicates so that in the simple case (when ?(| is not used) they
6092 are in order of their numbers. */
6093
6094 if (crc < 0)
6095 {
6096 memmove(slot + cd->name_entry_size, slot,
6097 IN_UCHARS((cd->names_found - i) * cd->name_entry_size));
6098 break;
6099 }
6100
6101 /* Continue the loop for a later or duplicate name */
6102
6103 slot += cd->name_entry_size;
6104 }
6105
6106 /* For non-duplicate names, check for a duplicate number before
6107 adding the new name. */
6108
6109 if (!dupname)
6110 {
6111 pcre_uchar *cslot = cd->name_table;
6112 for (i = 0; i < cd->names_found; i++)
6113 {
6114 if (cslot != slot)
6115 {
6116 if (GET2(cslot, 0) == cd->bracount + 1)
6117 {
6118 *errorcodeptr = ERR65;
6119 goto FAILED;
6120 }
6121 }
6122 else i--;
6123 cslot += cd->name_entry_size;
6124 }
6125 }
6126
6127 PUT2(slot, 0, cd->bracount + 1);
6128 memcpy(slot + IMM2_SIZE, name, IN_UCHARS(namelen));
6129 slot[IMM2_SIZE + namelen] = 0;
6130 }
6131 }
6132
6133 /* In both pre-compile and compile, count the number of names we've
6134 encountered. */
6135
6136 cd->names_found++;
6137 ptr++; /* Move past > or ' */
6138 goto NUMBERED_GROUP;
6139
6140
6141 /* ------------------------------------------------------------ */
6142 case CHAR_AMPERSAND: /* Perl recursion/subroutine syntax */
6143 terminator = CHAR_RIGHT_PARENTHESIS;
6144 is_recurse = TRUE;
6145 /* Fall through */
6146
6147 /* We come here from the Python syntax above that handles both
6148 references (?P=name) and recursion (?P>name), as well as falling
6149 through from the Perl recursion syntax (?&name). We also come here from
6150 the Perl \k<name> or \k'name' back reference syntax and the \k{name}
6151 .NET syntax, and the Oniguruma \g<...> and \g'...' subroutine syntax. */
6152
6153 NAMED_REF_OR_RECURSE:
6154 name = ++ptr;
6155 while (MAX_255(*ptr) && (cd->ctypes[*ptr] & ctype_word) != 0) ptr++;
6156 namelen = (int)(ptr - name);
6157
6158 /* In the pre-compile phase, do a syntax check. We used to just set
6159 a dummy reference number, because it was not used in the first pass.
6160 However, with the change of recursive back references to be atomic,
6161 we have to look for the number so that this state can be identified, as
6162 otherwise the incorrect length is computed. If it's not a backwards
6163 reference, the dummy number will do. */
6164
6165 if (lengthptr != NULL)
6166 {
6167 const pcre_uchar *temp;
6168
6169 if (namelen == 0)
6170 {
6171 *errorcodeptr = ERR62;
6172 goto FAILED;
6173 }
6174 if (*ptr != terminator)
6175 {
6176 *errorcodeptr = ERR42;
6177 goto FAILED;
6178 }
6179 if (namelen > MAX_NAME_SIZE)
6180 {
6181 *errorcodeptr = ERR48;
6182 goto FAILED;
6183 }
6184
6185 /* The name table does not exist in the first pass, so we cannot
6186 do a simple search as in the code below. Instead, we have to scan the
6187 pattern to find the number. It is important that we scan it only as
6188 far as we have got because the syntax of named subpatterns has not
6189 been checked for the rest of the pattern, and find_parens() assumes
6190 correct syntax. In any case, it's a waste of resources to scan
6191 further. We stop the scan at the current point by temporarily
6192 adjusting the value of cd->endpattern. */
6193
6194 temp = cd->end_pattern;
6195 cd->end_pattern = ptr;
6196 recno = find_parens(cd, name, namelen,
6197 (options & PCRE_EXTENDED) != 0, utf);
6198 cd->end_pattern = temp;
6199 if (recno < 0) recno = 0; /* Forward ref; set dummy number */
6200 }
6201
6202 /* In the real compile, seek the name in the table. We check the name
6203 first, and then check that we have reached the end of the name in the
6204 table. That way, if the name that is longer than any in the table,
6205 the comparison will fail without reading beyond the table entry. */
6206
6207 else
6208 {
6209 slot = cd->name_table;
6210 for (i = 0; i < cd->names_found; i++)
6211 {
6212 if (STRNCMP_UC_UC(name, slot+IMM2_SIZE, namelen) == 0 &&
6213 slot[IMM2_SIZE+namelen] == 0)
6214 break;
6215 slot += cd->name_entry_size;
6216 }
6217
6218 if (i < cd->names_found) /* Back reference */
6219 {
6220 recno = GET2(slot, 0);
6221 }
6222 else if ((recno = /* Forward back reference */
6223 find_parens(cd, name, namelen,
6224 (options & PCRE_EXTENDED) != 0, utf)) <= 0)
6225 {
6226 *errorcodeptr = ERR15;
6227 goto FAILED;
6228 }
6229 }
6230
6231 /* In both phases, we can now go to the code than handles numerical
6232 recursion or backreferences. */
6233
6234 if (is_recurse) goto HANDLE_RECURSION;
6235 else goto HANDLE_REFERENCE;
6236
6237
6238 /* ------------------------------------------------------------ */
6239 case CHAR_R: /* Recursion */
6240 ptr++; /* Same as (?0) */
6241 /* Fall through */
6242
6243
6244 /* ------------------------------------------------------------ */
6245 case CHAR_MINUS: case CHAR_PLUS: /* Recursion or subroutine */
6246 case CHAR_0: case CHAR_1: case CHAR_2: case CHAR_3: case CHAR_4:
6247 case CHAR_5: case CHAR_6: case CHAR_7: case CHAR_8: case CHAR_9:
6248 {
6249 const pcre_uchar *called;
6250 terminator = CHAR_RIGHT_PARENTHESIS;
6251
6252 /* Come here from the \g<...> and \g'...' code (Oniguruma
6253 compatibility). However, the syntax has been checked to ensure that
6254 the ... are a (signed) number, so that neither ERR63 nor ERR29 will
6255 be called on this path, nor with the jump to OTHER_CHAR_AFTER_QUERY
6256 ever be taken. */
6257
6258 HANDLE_NUMERICAL_RECURSION:
6259
6260 if ((refsign = *ptr) == CHAR_PLUS)
6261 {
6262 ptr++;
6263 if (!IS_DIGIT(*ptr))
6264 {
6265 *errorcodeptr = ERR63;
6266 goto FAILED;
6267 }
6268 }
6269 else if (refsign == CHAR_MINUS)
6270 {
6271 if (!IS_DIGIT(ptr[1]))
6272 goto OTHER_CHAR_AFTER_QUERY;
6273 ptr++;
6274 }
6275
6276 recno = 0;
6277 while(IS_DIGIT(*ptr))
6278 recno = recno * 10 + *ptr++ - CHAR_0;
6279
6280 if (*ptr != terminator)
6281 {
6282 *errorcodeptr = ERR29;
6283 goto FAILED;
6284 }
6285
6286 if (refsign == CHAR_MINUS)
6287 {
6288 if (recno == 0)
6289 {
6290 *errorcodeptr = ERR58;
6291 goto FAILED;
6292 }
6293 recno = cd->bracount - recno + 1;
6294 if (recno <= 0)
6295 {
6296 *errorcodeptr = ERR15;
6297 goto FAILED;
6298 }
6299 }
6300 else if (refsign == CHAR_PLUS)
6301 {
6302 if (recno == 0)
6303 {
6304 *errorcodeptr = ERR58;
6305 goto FAILED;
6306 }
6307 recno += cd->bracount;
6308 }
6309
6310 /* Come here from code above that handles a named recursion */
6311
6312 HANDLE_RECURSION:
6313
6314 previous = code;
6315 called = cd->start_code;
6316
6317 /* When we are actually compiling, find the bracket that is being
6318 referenced. Temporarily end the regex in case it doesn't exist before
6319 this point. If we end up with a forward reference, first check that
6320 the bracket does occur later so we can give the error (and position)
6321 now. Then remember this forward reference in the workspace so it can
6322 be filled in at the end. */
6323
6324 if (lengthptr == NULL)
6325 {
6326 *code = OP_END;
6327 if (recno != 0)
6328 called = PRIV(find_bracket)(cd->start_code, utf, recno);
6329
6330 /* Forward reference */
6331
6332 if (called == NULL)
6333 {
6334 if (find_parens(cd, NULL, recno,
6335 (options & PCRE_EXTENDED) != 0, utf) < 0)
6336 {
6337 *errorcodeptr = ERR15;
6338 goto FAILED;
6339 }
6340
6341 /* Fudge the value of "called" so that when it is inserted as an
6342 offset below, what it actually inserted is the reference number
6343 of the group. Then remember the forward reference. */
6344
6345 called = cd->start_code + recno;
6346 if (cd->hwm >= cd->start_workspace + cd->workspace_size -
6347 WORK_SIZE_SAFETY_MARGIN)
6348 {
6349 *errorcodeptr = expand_workspace(cd);
6350 if (*errorcodeptr != 0) goto FAILED;
6351 }
6352 PUTINC(cd->hwm, 0, (int)(code + 1 - cd->start_code));
6353 }
6354
6355 /* If not a forward reference, and the subpattern is still open,
6356 this is a recursive call. We check to see if this is a left
6357 recursion that could loop for ever, and diagnose that case. We
6358 must not, however, do this check if we are in a conditional
6359 subpattern because the condition might be testing for recursion in
6360 a pattern such as /(?(R)a+|(?R)b)/, which is perfectly valid.
6361 Forever loops are also detected at runtime, so those that occur in
6362 conditional subpatterns will be picked up then. */
6363
6364 else if (GET(called, 1) == 0 && cond_depth <= 0 &&
6365 could_be_empty(called, code, bcptr, utf, cd))
6366 {
6367 *errorcodeptr = ERR40;
6368 goto FAILED;
6369 }
6370 }
6371
6372 /* Insert the recursion/subroutine item. It does not have a set first
6373 character (relevant if it is repeated, because it will then be
6374 wrapped with ONCE brackets). */
6375
6376 *code = OP_RECURSE;
6377 PUT(code, 1, (int)(called - cd->start_code));
6378 code += 1 + LINK_SIZE;
6379 groupsetfirstchar = FALSE;
6380 }
6381
6382 /* Can't determine a first byte now */
6383
6384 if (firstchar == REQ_UNSET) firstchar = REQ_NONE;
6385 continue;
6386
6387
6388 /* ------------------------------------------------------------ */
6389 default: /* Other characters: check option setting */
6390 OTHER_CHAR_AFTER_QUERY:
6391 set = unset = 0;
6392 optset = &set;
6393
6394 while (*ptr != CHAR_RIGHT_PARENTHESIS && *ptr != CHAR_COLON)
6395 {
6396 switch (*ptr++)
6397 {
6398 case CHAR_MINUS: optset = &unset; break;
6399
6400 case CHAR_J: /* Record that it changed in the external options */
6401 *optset |= PCRE_DUPNAMES;
6402 cd->external_flags |= PCRE_JCHANGED;
6403 break;
6404
6405 case CHAR_i: *optset |= PCRE_CASELESS; break;
6406 case CHAR_m: *optset |= PCRE_MULTILINE; break;
6407 case CHAR_s: *optset |= PCRE_DOTALL; break;
6408 case CHAR_x: *optset |= PCRE_EXTENDED; break;
6409 case CHAR_U: *optset |= PCRE_UNGREEDY; break;
6410 case CHAR_X: *optset |= PCRE_EXTRA; break;
6411
6412 default: *errorcodeptr = ERR12;
6413 ptr--; /* Correct the offset */
6414 goto FAILED;
6415 }
6416 }
6417
6418 /* Set up the changed option bits, but don't change anything yet. */
6419
6420 newoptions = (options | set) & (~unset);
6421
6422 /* If the options ended with ')' this is not the start of a nested
6423 group with option changes, so the options change at this level. If this
6424 item is right at the start of the pattern, the options can be
6425 abstracted and made external in the pre-compile phase, and ignored in
6426 the compile phase. This can be helpful when matching -- for instance in
6427 caseless checking of required bytes.
6428
6429 If the code pointer is not (cd->start_code + 1 + LINK_SIZE), we are
6430 definitely *not* at the start of the pattern because something has been
6431 compiled. In the pre-compile phase, however, the code pointer can have
6432 that value after the start, because it gets reset as code is discarded
6433 during the pre-compile. However, this can happen only at top level - if
6434 we are within parentheses, the starting BRA will still be present. At
6435 any parenthesis level, the length value can be used to test if anything
6436 has been compiled at that level. Thus, a test for both these conditions
6437 is necessary to ensure we correctly detect the start of the pattern in
6438 both phases.
6439
6440 If we are not at the pattern start, reset the greedy defaults and the
6441 case value for firstchar and reqchar. */
6442
6443 if (*ptr == CHAR_RIGHT_PARENTHESIS)
6444 {
6445 if (code == cd->start_code + 1 + LINK_SIZE &&
6446 (lengthptr == NULL || *lengthptr == 2 + 2*LINK_SIZE))
6447 {
6448 cd->external_options = newoptions;
6449 }
6450 else
6451 {
6452 greedy_default = ((newoptions & PCRE_UNGREEDY) != 0);
6453 greedy_non_default = greedy_default ^ 1;
6454 req_caseopt = ((newoptions & PCRE_CASELESS) != 0)? REQ_CASELESS:0;
6455 }
6456
6457 /* Change options at this level, and pass them back for use
6458 in subsequent branches. */
6459
6460 *optionsptr = options = newoptions;
6461 previous = NULL; /* This item can't be repeated */
6462 continue; /* It is complete */
6463 }
6464
6465 /* If the options ended with ':' we are heading into a nested group
6466 with possible change of options. Such groups are non-capturing and are
6467 not assertions of any kind. All we need to do is skip over the ':';
6468 the newoptions value is handled below. */
6469
6470 bravalue = OP_BRA;
6471 ptr++;
6472 } /* End of switch for character following (? */
6473 } /* End of (? handling */
6474
6475 /* Opening parenthesis not followed by '*' or '?'. If PCRE_NO_AUTO_CAPTURE
6476 is set, all unadorned brackets become non-capturing and behave like (?:...)
6477 brackets. */
6478
6479 else if ((options & PCRE_NO_AUTO_CAPTURE) != 0)
6480 {
6481 bravalue = OP_BRA;
6482 }
6483
6484 /* Else we have a capturing group. */
6485
6486 else
6487 {
6488 NUMBERED_GROUP:
6489 cd->bracount += 1;
6490 PUT2(code, 1+LINK_SIZE, cd->bracount);
6491 skipbytes = IMM2_SIZE;
6492 }
6493
6494 /* Process nested bracketed regex. Assertions used not to be repeatable,
6495 but this was changed for Perl compatibility, so all kinds can now be
6496 repeated. We copy code into a non-register variable (tempcode) in order to
6497 be able to pass its address because some compilers complain otherwise. */
6498
6499 previous = code; /* For handling repetition */
6500 *code = bravalue;
6501 tempcode = code;
6502 tempreqvary = cd->req_varyopt; /* Save value before bracket */
6503 tempbracount = cd->bracount; /* Save value before bracket */
6504 length_prevgroup = 0; /* Initialize for pre-compile phase */
6505
6506 if (!compile_regex(
6507 newoptions, /* The complete new option state */
6508 &tempcode, /* Where to put code (updated) */
6509 &ptr, /* Input pointer (updated) */
6510 errorcodeptr, /* Where to put an error message */
6511 (bravalue == OP_ASSERTBACK ||
6512 bravalue == OP_ASSERTBACK_NOT), /* TRUE if back assert */
6513 reset_bracount, /* True if (?| group */
6514 skipbytes, /* Skip over bracket number */
6515 cond_depth +
6516 ((bravalue == OP_COND)?1:0), /* Depth of condition subpatterns */
6517 &subfirstchar, /* For possible first char */
6518 &subreqchar, /* For possible last char */
6519 bcptr, /* Current branch chain */
6520 cd, /* Tables block */
6521 (lengthptr == NULL)? NULL : /* Actual compile phase */
6522 &length_prevgroup /* Pre-compile phase */
6523 ))
6524 goto FAILED;
6525
6526 /* If this was an atomic group and there are no capturing groups within it,
6527 generate OP_ONCE_NC instead of OP_ONCE. */
6528
6529 if (bravalue == OP_ONCE && cd->bracount <= tempbracount)
6530 *code = OP_ONCE_NC;
6531
6532 if (bravalue >= OP_ASSERT && bravalue <= OP_ASSERTBACK_NOT)
6533 cd->assert_depth -= 1;
6534
6535 /* At the end of compiling, code is still pointing to the start of the
6536 group, while tempcode has been updated to point past the end of the group.
6537 The pattern pointer (ptr) is on the bracket.
6538
6539 If this is a conditional bracket, check that there are no more than
6540 two branches in the group, or just one if it's a DEFINE group. We do this
6541 in the real compile phase, not in the pre-pass, where the whole group may
6542 not be available. */
6543
6544 if (bravalue == OP_COND && lengthptr == NULL)
6545 {
6546 pcre_uchar *tc = code;
6547 int condcount = 0;
6548
6549 do {
6550 condcount++;
6551 tc += GET(tc,1);
6552 }
6553 while (*tc != OP_KET);
6554
6555 /* A DEFINE group is never obeyed inline (the "condition" is always
6556 false). It must have only one branch. */
6557
6558 if (code[LINK_SIZE+1] == OP_DEF)
6559 {
6560 if (condcount > 1)
6561 {
6562 *errorcodeptr = ERR54;
6563 goto FAILED;
6564 }
6565 bravalue = OP_DEF; /* Just a flag to suppress char handling below */
6566 }
6567
6568 /* A "normal" conditional group. If there is just one branch, we must not
6569 make use of its firstchar or reqchar, because this is equivalent to an
6570 empty second branch. */
6571
6572 else
6573 {
6574 if (condcount > 2)
6575 {
6576 *errorcodeptr = ERR27;
6577 goto FAILED;
6578 }
6579 if (condcount == 1) subfirstchar = subreqchar = REQ_NONE;
6580 }
6581 }
6582
6583 /* Error if hit end of pattern */
6584
6585 if (*ptr != CHAR_RIGHT_PARENTHESIS)
6586 {
6587 *errorcodeptr = ERR14;
6588 goto FAILED;
6589 }
6590
6591 /* In the pre-compile phase, update the length by the length of the group,
6592 less the brackets at either end. Then reduce the compiled code to just a
6593 set of non-capturing brackets so that it doesn't use much memory if it is
6594 duplicated by a quantifier.*/
6595
6596 if (lengthptr != NULL)
6597 {
6598 if (OFLOW_MAX - *lengthptr < length_prevgroup - 2 - 2*LINK_SIZE)
6599 {
6600 *errorcodeptr = ERR20;
6601 goto FAILED;
6602 }
6603 *lengthptr += length_prevgroup - 2 - 2*LINK_SIZE;
6604 code++; /* This already contains bravalue */
6605 PUTINC(code, 0, 1 + LINK_SIZE);
6606 *code++ = OP_KET;
6607 PUTINC(code, 0, 1 + LINK_SIZE);
6608 break; /* No need to waste time with special character handling */
6609 }
6610
6611 /* Otherwise update the main code pointer to the end of the group. */
6612
6613 code = tempcode;
6614
6615 /* For a DEFINE group, required and first character settings are not
6616 relevant. */
6617
6618 if (bravalue == OP_DEF) break;
6619
6620 /* Handle updating of the required and first characters for other types of
6621 group. Update for normal brackets of all kinds, and conditions with two
6622 branches (see code above). If the bracket is followed by a quantifier with
6623 zero repeat, we have to back off. Hence the definition of zeroreqchar and
6624 zerofirstchar outside the main loop so that they can be accessed for the
6625 back off. */
6626
6627 zeroreqchar = reqchar;
6628 zerofirstchar = firstchar;
6629 groupsetfirstchar = FALSE;
6630
6631 if (bravalue >= OP_ONCE)
6632 {
6633 /* If we have not yet set a firstchar in this branch, take it from the
6634 subpattern, remembering that it was set here so that a repeat of more
6635 than one can replicate it as reqchar if necessary. If the subpattern has
6636 no firstchar, set "none" for the whole branch. In both cases, a zero
6637 repeat forces firstchar to "none". */
6638
6639 if (firstchar == REQ_UNSET)
6640 {
6641