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

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

Parent Directory Parent Directory | Revision Log Revision Log


Revision 805 - (show annotations)
Wed Dec 14 16:49:20 2011 UTC (8 years, 2 months ago) by ph10
File MIME type: text/plain
File size: 266309 byte(s)
Preliminary work on pcretest for 16-bit (not complete by any means).
1 /*************************************************
2 * Perl-Compatible Regular Expressions *
3 *************************************************/
4
5 /* PCRE is a library of functions to support regular expressions whose syntax
6 and semantics are as close as possible to those of the Perl 5 language.
7
8 Written by Philip Hazel
9 Copyright (c) 1997-2011 University of Cambridge
10
11 -----------------------------------------------------------------------------
12 Redistribution and use in source and binary forms, with or without
13 modification, are permitted provided that the following conditions are met:
14
15 * Redistributions of source code must retain the above copyright notice,
16 this list of conditions and the following disclaimer.
17
18 * Redistributions in binary form must reproduce the above copyright
19 notice, this list of conditions and the following disclaimer in the
20 documentation and/or other materials provided with the distribution.
21
22 * Neither the name of the University of Cambridge nor the names of its
23 contributors may be used to endorse or promote products derived from
24 this software without specific prior written permission.
25
26 THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
27 AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28 IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29 ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE
30 LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
31 CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
32 SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
33 INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
34 CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
35 ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
36 POSSIBILITY OF SUCH DAMAGE.
37 -----------------------------------------------------------------------------
38 */
39
40
41 /* This module contains the external function pcre_compile(), along with
42 supporting internal functions that are not used by other modules. */
43
44
45 #ifdef HAVE_CONFIG_H
46 #include "config.h"
47 #endif
48
49 #define NLBLOCK cd /* Block containing newline information */
50 #define PSSTART start_pattern /* Field containing processed string start */
51 #define PSEND end_pattern /* Field containing processed string end */
52
53 #include "pcre_internal.h"
54
55
56 /* When PCRE_DEBUG is defined, we need the pcre(16)_printint() function, which
57 is also used by pcretest. PCRE_DEBUG is not defined when building a production
58 library. We do not need to select pcre16_printint.c specially, because the
59 COMPILE_PCREx macro will already be appropriately set. */
60
61 #ifdef PCRE_DEBUG
62 #include "pcre_printint.c"
63 #endif
64
65
66 /* Macro for setting individual bits in class bitmaps. */
67
68 #define SETBIT(a,b) a[b/8] |= (1 << (b%8))
69
70 /* Maximum length value to check against when making sure that the integer that
71 holds the compiled pattern length does not overflow. We make it a bit less than
72 INT_MAX to allow for adding in group terminating bytes, so that we don't have
73 to check them every time. */
74
75 #define OFLOW_MAX (INT_MAX - 20)
76
77
78 /*************************************************
79 * Code parameters and static tables *
80 *************************************************/
81
82 /* This value specifies the size of stack workspace that is used during the
83 first pre-compile phase that determines how much memory is required. The regex
84 is partly compiled into this space, but the compiled parts are discarded as
85 soon as they can be, so that hopefully there will never be an overrun. The code
86 does, however, check for an overrun. The largest amount I've seen used is 218,
87 so this number is very generous.
88
89 The same workspace is used during the second, actual compile phase for
90 remembering forward references to groups so that they can be filled in at the
91 end. Each entry in this list occupies LINK_SIZE bytes, so even when LINK_SIZE
92 is 4 there is plenty of room for most patterns. However, the memory can get
93 filled up by repetitions of forward references, for example patterns like
94 /(?1){0,1999}(b)/, and one user did hit the limit. The code has been changed so
95 that the workspace is expanded using malloc() in this situation. The value
96 below is therefore a minimum, and we put a maximum on it for safety. The
97 minimum is now also defined in terms of LINK_SIZE so that the use of malloc()
98 kicks in at the same number of forward references in all cases. */
99
100 #define COMPILE_WORK_SIZE (2048*LINK_SIZE)
101 #define COMPILE_WORK_SIZE_MAX (100*COMPILE_WORK_SIZE)
102
103 /* The overrun tests check for a slightly smaller size so that they detect the
104 overrun before it actually does run off the end of the data block. */
105
106 #define WORK_SIZE_SAFETY_MARGIN (100)
107
108 /* Private flags added to firstchar and reqchar. */
109
110 #define REQ_CASELESS 0x10000000l /* Indicates caselessness */
111 #define REQ_VARY 0x20000000l /* Reqchar followed non-literal item */
112
113 /* Repeated character flags. */
114
115 #define UTF_LENGTH 0x10000000l /* The char contains its length. */
116
117 /* Table for handling escaped characters in the range '0'-'z'. Positive returns
118 are simple data values; negative values are for special things like \d and so
119 on. Zero means further processing is needed (for things like \x), or the escape
120 is invalid. */
121
122 #ifndef EBCDIC
123
124 /* This is the "normal" table for ASCII systems or for EBCDIC systems running
125 in UTF-8 mode. */
126
127 static const short int escapes[] = {
128 0, 0,
129 0, 0,
130 0, 0,
131 0, 0,
132 0, 0,
133 CHAR_COLON, CHAR_SEMICOLON,
134 CHAR_LESS_THAN_SIGN, CHAR_EQUALS_SIGN,
135 CHAR_GREATER_THAN_SIGN, CHAR_QUESTION_MARK,
136 CHAR_COMMERCIAL_AT, -ESC_A,
137 -ESC_B, -ESC_C,
138 -ESC_D, -ESC_E,
139 0, -ESC_G,
140 -ESC_H, 0,
141 0, -ESC_K,
142 0, 0,
143 -ESC_N, 0,
144 -ESC_P, -ESC_Q,
145 -ESC_R, -ESC_S,
146 0, 0,
147 -ESC_V, -ESC_W,
148 -ESC_X, 0,
149 -ESC_Z, CHAR_LEFT_SQUARE_BRACKET,
150 CHAR_BACKSLASH, CHAR_RIGHT_SQUARE_BRACKET,
151 CHAR_CIRCUMFLEX_ACCENT, CHAR_UNDERSCORE,
152 CHAR_GRAVE_ACCENT, 7,
153 -ESC_b, 0,
154 -ESC_d, ESC_e,
155 ESC_f, 0,
156 -ESC_h, 0,
157 0, -ESC_k,
158 0, 0,
159 ESC_n, 0,
160 -ESC_p, 0,
161 ESC_r, -ESC_s,
162 ESC_tee, 0,
163 -ESC_v, -ESC_w,
164 0, 0,
165 -ESC_z
166 };
167
168 #else
169
170 /* This is the "abnormal" table for EBCDIC systems without UTF-8 support. */
171
172 static const short int escapes[] = {
173 /* 48 */ 0, 0, 0, '.', '<', '(', '+', '|',
174 /* 50 */ '&', 0, 0, 0, 0, 0, 0, 0,
175 /* 58 */ 0, 0, '!', '$', '*', ')', ';', '~',
176 /* 60 */ '-', '/', 0, 0, 0, 0, 0, 0,
177 /* 68 */ 0, 0, '|', ',', '%', '_', '>', '?',
178 /* 70 */ 0, 0, 0, 0, 0, 0, 0, 0,
179 /* 78 */ 0, '`', ':', '#', '@', '\'', '=', '"',
180 /* 80 */ 0, 7, -ESC_b, 0, -ESC_d, ESC_e, ESC_f, 0,
181 /* 88 */-ESC_h, 0, 0, '{', 0, 0, 0, 0,
182 /* 90 */ 0, 0, -ESC_k, 'l', 0, ESC_n, 0, -ESC_p,
183 /* 98 */ 0, ESC_r, 0, '}', 0, 0, 0, 0,
184 /* A0 */ 0, '~', -ESC_s, ESC_tee, 0,-ESC_v, -ESC_w, 0,
185 /* A8 */ 0,-ESC_z, 0, 0, 0, '[', 0, 0,
186 /* B0 */ 0, 0, 0, 0, 0, 0, 0, 0,
187 /* B8 */ 0, 0, 0, 0, 0, ']', '=', '-',
188 /* C0 */ '{',-ESC_A, -ESC_B, -ESC_C, -ESC_D,-ESC_E, 0, -ESC_G,
189 /* C8 */-ESC_H, 0, 0, 0, 0, 0, 0, 0,
190 /* D0 */ '}', 0, -ESC_K, 0, 0,-ESC_N, 0, -ESC_P,
191 /* D8 */-ESC_Q,-ESC_R, 0, 0, 0, 0, 0, 0,
192 /* E0 */ '\\', 0, -ESC_S, 0, 0,-ESC_V, -ESC_W, -ESC_X,
193 /* E8 */ 0,-ESC_Z, 0, 0, 0, 0, 0, 0,
194 /* F0 */ 0, 0, 0, 0, 0, 0, 0, 0,
195 /* F8 */ 0, 0, 0, 0, 0, 0, 0, 0
196 };
197 #endif
198
199
200 /* Table of special "verbs" like (*PRUNE). This is a short table, so it is
201 searched linearly. Put all the names into a single string, in order to reduce
202 the number of relocations when a shared library is dynamically linked. The
203 string is built from string macros so that it works in UTF-8 mode on EBCDIC
204 platforms. */
205
206 typedef struct verbitem {
207 int len; /* Length of verb name */
208 int op; /* Op when no arg, or -1 if arg mandatory */
209 int op_arg; /* Op when arg present, or -1 if not allowed */
210 } verbitem;
211
212 static const char verbnames[] =
213 "\0" /* Empty name is a shorthand for MARK */
214 STRING_MARK0
215 STRING_ACCEPT0
216 STRING_COMMIT0
217 STRING_F0
218 STRING_FAIL0
219 STRING_PRUNE0
220 STRING_SKIP0
221 STRING_THEN;
222
223 static const verbitem verbs[] = {
224 { 0, -1, OP_MARK },
225 { 4, -1, OP_MARK },
226 { 6, OP_ACCEPT, -1 },
227 { 6, OP_COMMIT, -1 },
228 { 1, OP_FAIL, -1 },
229 { 4, OP_FAIL, -1 },
230 { 5, OP_PRUNE, OP_PRUNE_ARG },
231 { 4, OP_SKIP, OP_SKIP_ARG },
232 { 4, OP_THEN, OP_THEN_ARG }
233 };
234
235 static const int verbcount = sizeof(verbs)/sizeof(verbitem);
236
237
238 /* Tables of names of POSIX character classes and their lengths. The names are
239 now all in a single string, to reduce the number of relocations when a shared
240 library is dynamically loaded. The list of lengths is terminated by a zero
241 length entry. The first three must be alpha, lower, upper, as this is assumed
242 for handling case independence. */
243
244 static const char posix_names[] =
245 STRING_alpha0 STRING_lower0 STRING_upper0 STRING_alnum0
246 STRING_ascii0 STRING_blank0 STRING_cntrl0 STRING_digit0
247 STRING_graph0 STRING_print0 STRING_punct0 STRING_space0
248 STRING_word0 STRING_xdigit;
249
250 static const pcre_uint8 posix_name_lengths[] = {
251 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 4, 6, 0 };
252
253 /* Table of class bit maps for each POSIX class. Each class is formed from a
254 base map, with an optional addition or removal of another map. Then, for some
255 classes, there is some additional tweaking: for [:blank:] the vertical space
256 characters are removed, and for [:alpha:] and [:alnum:] the underscore
257 character is removed. The triples in the table consist of the base map offset,
258 second map offset or -1 if no second map, and a non-negative value for map
259 addition or a negative value for map subtraction (if there are two maps). The
260 absolute value of the third field has these meanings: 0 => no tweaking, 1 =>
261 remove vertical space characters, 2 => remove underscore. */
262
263 static const int posix_class_maps[] = {
264 cbit_word, cbit_digit, -2, /* alpha */
265 cbit_lower, -1, 0, /* lower */
266 cbit_upper, -1, 0, /* upper */
267 cbit_word, -1, 2, /* alnum - word without underscore */
268 cbit_print, cbit_cntrl, 0, /* ascii */
269 cbit_space, -1, 1, /* blank - a GNU extension */
270 cbit_cntrl, -1, 0, /* cntrl */
271 cbit_digit, -1, 0, /* digit */
272 cbit_graph, -1, 0, /* graph */
273 cbit_print, -1, 0, /* print */
274 cbit_punct, -1, 0, /* punct */
275 cbit_space, -1, 0, /* space */
276 cbit_word, -1, 0, /* word - a Perl extension */
277 cbit_xdigit,-1, 0 /* xdigit */
278 };
279
280 /* Table of substitutes for \d etc when PCRE_UCP is set. The POSIX class
281 substitutes must be in the order of the names, defined above, and there are
282 both positive and negative cases. NULL means no substitute. */
283
284 #ifdef SUPPORT_UCP
285 static const pcre_uchar string_PNd[] = {
286 CHAR_BACKSLASH, CHAR_P, CHAR_LEFT_CURLY_BRACKET,
287 CHAR_N, CHAR_d, CHAR_RIGHT_CURLY_BRACKET, '\0' };
288 static const pcre_uchar string_pNd[] = {
289 CHAR_BACKSLASH, CHAR_p, CHAR_LEFT_CURLY_BRACKET,
290 CHAR_N, CHAR_d, CHAR_RIGHT_CURLY_BRACKET, '\0' };
291 static const pcre_uchar string_PXsp[] = {
292 CHAR_BACKSLASH, CHAR_P, CHAR_LEFT_CURLY_BRACKET,
293 CHAR_X, CHAR_s, CHAR_p, CHAR_RIGHT_CURLY_BRACKET, '\0' };
294 static const pcre_uchar string_pXsp[] = {
295 CHAR_BACKSLASH, CHAR_p, CHAR_LEFT_CURLY_BRACKET,
296 CHAR_X, CHAR_s, CHAR_p, CHAR_RIGHT_CURLY_BRACKET, '\0' };
297 static const pcre_uchar string_PXwd[] = {
298 CHAR_BACKSLASH, CHAR_P, CHAR_LEFT_CURLY_BRACKET,
299 CHAR_X, CHAR_w, CHAR_d, CHAR_RIGHT_CURLY_BRACKET, '\0' };
300 static const pcre_uchar string_pXwd[] = {
301 CHAR_BACKSLASH, CHAR_p, CHAR_LEFT_CURLY_BRACKET,
302 CHAR_X, CHAR_w, CHAR_d, CHAR_RIGHT_CURLY_BRACKET, '\0' };
303
304 static const pcre_uchar *substitutes[] = {
305 string_PNd, /* \D */
306 string_pNd, /* \d */
307 string_PXsp, /* \S */ /* NOTE: Xsp is Perl space */
308 string_pXsp, /* \s */
309 string_PXwd, /* \W */
310 string_pXwd /* \w */
311 };
312
313 static const pcre_uchar string_pL[] = {
314 CHAR_BACKSLASH, CHAR_p, CHAR_LEFT_CURLY_BRACKET,
315 CHAR_L, CHAR_RIGHT_CURLY_BRACKET, '\0' };
316 static const pcre_uchar string_pLl[] = {
317 CHAR_BACKSLASH, CHAR_p, CHAR_LEFT_CURLY_BRACKET,
318 CHAR_L, CHAR_l, CHAR_RIGHT_CURLY_BRACKET, '\0' };
319 static const pcre_uchar string_pLu[] = {
320 CHAR_BACKSLASH, CHAR_p, CHAR_LEFT_CURLY_BRACKET,
321 CHAR_L, CHAR_u, CHAR_RIGHT_CURLY_BRACKET, '\0' };
322 static const pcre_uchar string_pXan[] = {
323 CHAR_BACKSLASH, CHAR_p, CHAR_LEFT_CURLY_BRACKET,
324 CHAR_X, CHAR_a, CHAR_n, CHAR_RIGHT_CURLY_BRACKET, '\0' };
325 static const pcre_uchar string_h[] = {
326 CHAR_BACKSLASH, CHAR_h, '\0' };
327 static const pcre_uchar string_pXps[] = {
328 CHAR_BACKSLASH, CHAR_p, CHAR_LEFT_CURLY_BRACKET,
329 CHAR_X, CHAR_p, CHAR_s, CHAR_RIGHT_CURLY_BRACKET, '\0' };
330 static const pcre_uchar string_PL[] = {
331 CHAR_BACKSLASH, CHAR_P, CHAR_LEFT_CURLY_BRACKET,
332 CHAR_L, CHAR_RIGHT_CURLY_BRACKET, '\0' };
333 static const pcre_uchar string_PLl[] = {
334 CHAR_BACKSLASH, CHAR_P, CHAR_LEFT_CURLY_BRACKET,
335 CHAR_L, CHAR_l, CHAR_RIGHT_CURLY_BRACKET, '\0' };
336 static const pcre_uchar string_PLu[] = {
337 CHAR_BACKSLASH, CHAR_P, CHAR_LEFT_CURLY_BRACKET,
338 CHAR_L, CHAR_u, CHAR_RIGHT_CURLY_BRACKET, '\0' };
339 static const pcre_uchar string_PXan[] = {
340 CHAR_BACKSLASH, CHAR_P, CHAR_LEFT_CURLY_BRACKET,
341 CHAR_X, CHAR_a, CHAR_n, CHAR_RIGHT_CURLY_BRACKET, '\0' };
342 static const pcre_uchar string_H[] = {
343 CHAR_BACKSLASH, CHAR_H, '\0' };
344 static const pcre_uchar string_PXps[] = {
345 CHAR_BACKSLASH, CHAR_P, CHAR_LEFT_CURLY_BRACKET,
346 CHAR_X, CHAR_p, CHAR_s, CHAR_RIGHT_CURLY_BRACKET, '\0' };
347
348 static const pcre_uchar *posix_substitutes[] = {
349 string_pL, /* alpha */
350 string_pLl, /* lower */
351 string_pLu, /* upper */
352 string_pXan, /* alnum */
353 NULL, /* ascii */
354 string_h, /* blank */
355 NULL, /* cntrl */
356 string_pNd, /* digit */
357 NULL, /* graph */
358 NULL, /* print */
359 NULL, /* punct */
360 string_pXps, /* space */ /* NOTE: Xps is POSIX space */
361 string_pXwd, /* word */
362 NULL, /* xdigit */
363 /* Negated cases */
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 };
379 #define POSIX_SUBSIZE (sizeof(posix_substitutes) / sizeof(pcre_uchar *))
380 #endif
381
382 #define STRING(a) # a
383 #define XSTRING(s) STRING(s)
384
385 /* The texts of compile-time error messages. These are "char *" because they
386 are passed to the outside world. Do not ever re-use any error number, because
387 they are documented. Always add a new error instead. Messages marked DEAD below
388 are no longer used. This used to be a table of strings, but in order to reduce
389 the number of relocations needed when a shared library is loaded dynamically,
390 it is now one long string. We cannot use a table of offsets, because the
391 lengths of inserts such as XSTRING(MAX_NAME_SIZE) are not known. Instead, we
392 simply count through to the one we want - this isn't a performance issue
393 because these strings are used only when there is a compilation error.
394
395 Each substring ends with \0 to insert a null character. This includes the final
396 substring, so that the whole string ends with \0\0, which can be detected when
397 counting through. */
398
399 static const char error_texts[] =
400 "no error\0"
401 "\\ at end of pattern\0"
402 "\\c at end of pattern\0"
403 "unrecognized character follows \\\0"
404 "numbers out of order in {} quantifier\0"
405 /* 5 */
406 "number too big in {} quantifier\0"
407 "missing terminating ] for character class\0"
408 "invalid escape sequence in character class\0"
409 "range out of order in character class\0"
410 "nothing to repeat\0"
411 /* 10 */
412 "operand of unlimited repeat could match the empty string\0" /** DEAD **/
413 "internal error: unexpected repeat\0"
414 "unrecognized character after (? or (?-\0"
415 "POSIX named classes are supported only within a class\0"
416 "missing )\0"
417 /* 15 */
418 "reference to non-existent subpattern\0"
419 "erroffset passed as NULL\0"
420 "unknown option bit(s) set\0"
421 "missing ) after comment\0"
422 "parentheses nested too deeply\0" /** DEAD **/
423 /* 20 */
424 "regular expression is too large\0"
425 "failed to get memory\0"
426 "unmatched parentheses\0"
427 "internal error: code overflow\0"
428 "unrecognized character after (?<\0"
429 /* 25 */
430 "lookbehind assertion is not fixed length\0"
431 "malformed number or name after (?(\0"
432 "conditional group contains more than two branches\0"
433 "assertion expected after (?(\0"
434 "(?R or (?[+-]digits must be followed by )\0"
435 /* 30 */
436 "unknown POSIX class name\0"
437 "POSIX collating elements are not supported\0"
438 "this version of PCRE is not compiled with PCRE_UTF8 support\0"
439 "spare error\0" /** DEAD **/
440 "character value in \\x{...} sequence is too large\0"
441 /* 35 */
442 "invalid condition (?(0)\0"
443 "\\C not allowed in lookbehind assertion\0"
444 "PCRE does not support \\L, \\l, \\N{name}, \\U, or \\u\0"
445 "number after (?C is > 255\0"
446 "closing ) for (?C expected\0"
447 /* 40 */
448 "recursive call could loop indefinitely\0"
449 "unrecognized character after (?P\0"
450 "syntax error in subpattern name (missing terminator)\0"
451 "two named subpatterns have the same name\0"
452 "invalid UTF-8 string\0"
453 /* 45 */
454 "support for \\P, \\p, and \\X has not been compiled\0"
455 "malformed \\P or \\p sequence\0"
456 "unknown property name after \\P or \\p\0"
457 "subpattern name is too long (maximum " XSTRING(MAX_NAME_SIZE) " characters)\0"
458 "too many named subpatterns (maximum " XSTRING(MAX_NAME_COUNT) ")\0"
459 /* 50 */
460 "repeated subpattern is too long\0" /** DEAD **/
461 "octal value is greater than \\377 (not in UTF-8 mode)\0"
462 "internal error: overran compiling workspace\0"
463 "internal error: previously-checked referenced subpattern not found\0"
464 "DEFINE group contains more than one branch\0"
465 /* 55 */
466 "repeating a DEFINE group is not allowed\0" /** DEAD **/
467 "inconsistent NEWLINE options\0"
468 "\\g is not followed by a braced, angle-bracketed, or quoted name/number or by a plain number\0"
469 "a numbered reference must not be zero\0"
470 "an argument is not allowed for (*ACCEPT), (*FAIL), or (*COMMIT)\0"
471 /* 60 */
472 "(*VERB) not recognized\0"
473 "number is too big\0"
474 "subpattern name expected\0"
475 "digit expected after (?+\0"
476 "] is an invalid data character in JavaScript compatibility mode\0"
477 /* 65 */
478 "different names for subpatterns of the same number are not allowed\0"
479 "(*MARK) must have an argument\0"
480 "this version of PCRE is not compiled with PCRE_UCP support\0"
481 "\\c must be followed by an ASCII character\0"
482 "\\k is not followed by a braced, angle-bracketed, or quoted name\0"
483 /* 70 */
484 "internal error: unknown opcode in find_fixedlength()\0"
485 "\\N is not supported in a class\0"
486 "too many forward references\0"
487 "disallowed UTF-8/16 code point (>= 0xd800 && <= 0xdfff)\0"
488 ;
489
490 /* Table to identify digits and hex digits. This is used when compiling
491 patterns. Note that the tables in chartables are dependent on the locale, and
492 may mark arbitrary characters as digits - but the PCRE compiling code expects
493 to handle only 0-9, a-z, and A-Z as digits when compiling. That is why we have
494 a private table here. It costs 256 bytes, but it is a lot faster than doing
495 character value tests (at least in some simple cases I timed), and in some
496 applications one wants PCRE to compile efficiently as well as match
497 efficiently.
498
499 For convenience, we use the same bit definitions as in chartables:
500
501 0x04 decimal digit
502 0x08 hexadecimal digit
503
504 Then we can use ctype_digit and ctype_xdigit in the code. */
505
506 /* Using a simple comparison for decimal numbers rather than a memory read
507 is much faster, and the resulting code is simpler (the compiler turns it
508 into a subtraction and unsigned comparison). */
509
510 #define IS_DIGIT(x) ((x) >= CHAR_0 && (x) <= CHAR_9)
511
512 #ifndef EBCDIC
513
514 /* This is the "normal" case, for ASCII systems, and EBCDIC systems running in
515 UTF-8 mode. */
516
517 static const pcre_uint8 digitab[] =
518 {
519 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 0- 7 */
520 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 8- 15 */
521 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 16- 23 */
522 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 24- 31 */
523 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - ' */
524 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* ( - / */
525 0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c, /* 0 - 7 */
526 0x0c,0x0c,0x00,0x00,0x00,0x00,0x00,0x00, /* 8 - ? */
527 0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* @ - G */
528 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* H - O */
529 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* P - W */
530 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* X - _ */
531 0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* ` - g */
532 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* h - o */
533 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* p - w */
534 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* x -127 */
535 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 128-135 */
536 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 136-143 */
537 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 144-151 */
538 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 152-159 */
539 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 160-167 */
540 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 168-175 */
541 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 176-183 */
542 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 184-191 */
543 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 192-199 */
544 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 200-207 */
545 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 208-215 */
546 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 216-223 */
547 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 224-231 */
548 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 232-239 */
549 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 240-247 */
550 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00};/* 248-255 */
551
552 #else
553
554 /* This is the "abnormal" case, for EBCDIC systems not running in UTF-8 mode. */
555
556 static const pcre_uint8 digitab[] =
557 {
558 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 0- 7 0 */
559 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 8- 15 */
560 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 16- 23 10 */
561 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 24- 31 */
562 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 32- 39 20 */
563 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 40- 47 */
564 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 48- 55 30 */
565 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 56- 63 */
566 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - 71 40 */
567 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 72- | */
568 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* & - 87 50 */
569 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 88- 95 */
570 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - -103 60 */
571 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 104- ? */
572 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 112-119 70 */
573 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 120- " */
574 0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* 128- g 80 */
575 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* h -143 */
576 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 144- p 90 */
577 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* q -159 */
578 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 160- x A0 */
579 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* y -175 */
580 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* ^ -183 B0 */
581 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 184-191 */
582 0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* { - G C0 */
583 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* H -207 */
584 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* } - P D0 */
585 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* Q -223 */
586 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* \ - X E0 */
587 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* Y -239 */
588 0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c, /* 0 - 7 F0 */
589 0x0c,0x0c,0x00,0x00,0x00,0x00,0x00,0x00};/* 8 -255 */
590
591 static const pcre_uint8 ebcdic_chartab[] = { /* chartable partial dup */
592 0x80,0x00,0x00,0x00,0x00,0x01,0x00,0x00, /* 0- 7 */
593 0x00,0x00,0x00,0x00,0x01,0x01,0x00,0x00, /* 8- 15 */
594 0x00,0x00,0x00,0x00,0x00,0x01,0x00,0x00, /* 16- 23 */
595 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 24- 31 */
596 0x00,0x00,0x00,0x00,0x00,0x01,0x00,0x00, /* 32- 39 */
597 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 40- 47 */
598 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 48- 55 */
599 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 56- 63 */
600 0x01,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - 71 */
601 0x00,0x00,0x00,0x80,0x00,0x80,0x80,0x80, /* 72- | */
602 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* & - 87 */
603 0x00,0x00,0x00,0x80,0x80,0x80,0x00,0x00, /* 88- 95 */
604 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - -103 */
605 0x00,0x00,0x00,0x00,0x00,0x10,0x00,0x80, /* 104- ? */
606 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 112-119 */
607 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 120- " */
608 0x00,0x1a,0x1a,0x1a,0x1a,0x1a,0x1a,0x12, /* 128- g */
609 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* h -143 */
610 0x00,0x12,0x12,0x12,0x12,0x12,0x12,0x12, /* 144- p */
611 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* q -159 */
612 0x00,0x00,0x12,0x12,0x12,0x12,0x12,0x12, /* 160- x */
613 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* y -175 */
614 0x80,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* ^ -183 */
615 0x00,0x00,0x80,0x00,0x00,0x00,0x00,0x00, /* 184-191 */
616 0x80,0x1a,0x1a,0x1a,0x1a,0x1a,0x1a,0x12, /* { - G */
617 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* H -207 */
618 0x00,0x12,0x12,0x12,0x12,0x12,0x12,0x12, /* } - P */
619 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* Q -223 */
620 0x00,0x00,0x12,0x12,0x12,0x12,0x12,0x12, /* \ - X */
621 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* Y -239 */
622 0x1c,0x1c,0x1c,0x1c,0x1c,0x1c,0x1c,0x1c, /* 0 - 7 */
623 0x1c,0x1c,0x00,0x00,0x00,0x00,0x00,0x00};/* 8 -255 */
624 #endif
625
626
627 /* Definition to allow mutual recursion */
628
629 static BOOL
630 compile_regex(int, pcre_uchar **, const pcre_uchar **, int *, BOOL, BOOL, int, int,
631 int *, int *, branch_chain *, compile_data *, int *);
632
633
634
635 /*************************************************
636 * Find an error text *
637 *************************************************/
638
639 /* The error texts are now all in one long string, to save on relocations. As
640 some of the text is of unknown length, we can't use a table of offsets.
641 Instead, just count through the strings. This is not a performance issue
642 because it happens only when there has been a compilation error.
643
644 Argument: the error number
645 Returns: pointer to the error string
646 */
647
648 static const char *
649 find_error_text(int n)
650 {
651 const char *s = error_texts;
652 for (; n > 0; n--)
653 {
654 while (*s++ != 0) {};
655 if (*s == 0) return "Error text not found (please report)";
656 }
657 return s;
658 }
659
660
661 /*************************************************
662 * Expand the workspace *
663 *************************************************/
664
665 /* This function is called during the second compiling phase, if the number of
666 forward references fills the existing workspace, which is originally a block on
667 the stack. A larger block is obtained from malloc() unless the ultimate limit
668 has been reached or the increase will be rather small.
669
670 Argument: pointer to the compile data block
671 Returns: 0 if all went well, else an error number
672 */
673
674 static int
675 expand_workspace(compile_data *cd)
676 {
677 pcre_uchar *newspace;
678 int newsize = cd->workspace_size * 2;
679
680 if (newsize > COMPILE_WORK_SIZE_MAX) newsize = COMPILE_WORK_SIZE_MAX;
681 if (cd->workspace_size >= COMPILE_WORK_SIZE_MAX ||
682 newsize - cd->workspace_size < WORK_SIZE_SAFETY_MARGIN)
683 return ERR72;
684
685 newspace = (PUBL(malloc))(newsize);
686 if (newspace == NULL) return ERR21;
687
688 memcpy(newspace, cd->start_workspace, cd->workspace_size);
689 cd->hwm = (pcre_uchar *)newspace + (cd->hwm - cd->start_workspace);
690 if (cd->workspace_size > COMPILE_WORK_SIZE)
691 (PUBL(free))((void *)cd->start_workspace);
692 cd->start_workspace = newspace;
693 cd->workspace_size = newsize;
694 return 0;
695 }
696
697
698
699 /*************************************************
700 * Check for counted repeat *
701 *************************************************/
702
703 /* This function is called when a '{' is encountered in a place where it might
704 start a quantifier. It looks ahead to see if it really is a quantifier or not.
705 It is only a quantifier if it is one of the forms {ddd} {ddd,} or {ddd,ddd}
706 where the ddds are digits.
707
708 Arguments:
709 p pointer to the first char after '{'
710
711 Returns: TRUE or FALSE
712 */
713
714 static BOOL
715 is_counted_repeat(const pcre_uchar *p)
716 {
717 if (!IS_DIGIT(*p)) return FALSE;
718 p++;
719 while (IS_DIGIT(*p)) p++;
720 if (*p == CHAR_RIGHT_CURLY_BRACKET) return TRUE;
721
722 if (*p++ != CHAR_COMMA) return FALSE;
723 if (*p == CHAR_RIGHT_CURLY_BRACKET) return TRUE;
724
725 if (!IS_DIGIT(*p)) return FALSE;
726 p++;
727 while (IS_DIGIT(*p)) p++;
728
729 return (*p == CHAR_RIGHT_CURLY_BRACKET);
730 }
731
732
733
734 /*************************************************
735 * Handle escapes *
736 *************************************************/
737
738 /* This function is called when a \ has been encountered. It either returns a
739 positive value for a simple escape such as \n, or a negative value which
740 encodes one of the more complicated things such as \d. A backreference to group
741 n is returned as -(ESC_REF + n); ESC_REF is the highest ESC_xxx macro. When
742 UTF-8 is enabled, a positive value greater than 255 may be returned. On entry,
743 ptr is pointing at the \. On exit, it is on the final character of the escape
744 sequence.
745
746 Arguments:
747 ptrptr points to the pattern position pointer
748 errorcodeptr points to the errorcode variable
749 bracount number of previous extracting brackets
750 options the options bits
751 isclass TRUE if inside a character class
752
753 Returns: zero or positive => a data character
754 negative => a special escape sequence
755 on error, errorcodeptr is set
756 */
757
758 static int
759 check_escape(const pcre_uchar **ptrptr, int *errorcodeptr, int bracount,
760 int options, BOOL isclass)
761 {
762 /* PCRE_UTF16 has the same value as PCRE_UTF8. */
763 BOOL utf = (options & PCRE_UTF8) != 0;
764 const pcre_uchar *ptr = *ptrptr + 1;
765 pcre_int32 c;
766 int i;
767
768 GETCHARINCTEST(c, ptr); /* Get character value, increment pointer */
769 ptr--; /* Set pointer back to the last byte */
770
771 /* If backslash is at the end of the pattern, it's an error. */
772
773 if (c == 0) *errorcodeptr = ERR1;
774
775 /* Non-alphanumerics are literals. For digits or letters, do an initial lookup
776 in a table. A non-zero result is something that can be returned immediately.
777 Otherwise further processing may be required. */
778
779 #ifndef EBCDIC /* ASCII/UTF-8 coding */
780 /* Not alphanumeric */
781 else if (c < CHAR_0 || c > CHAR_z) {}
782 else if ((i = escapes[c - CHAR_0]) != 0) c = i;
783
784 #else /* EBCDIC coding */
785 /* Not alphanumeric */
786 else if (c < 'a' || (!MAX_255(c) || (ebcdic_chartab[c] & 0x0E) == 0)) {}
787 else if ((i = escapes[c - 0x48]) != 0) c = i;
788 #endif
789
790 /* Escapes that need further processing, or are illegal. */
791
792 else
793 {
794 const pcre_uchar *oldptr;
795 BOOL braced, negated;
796
797 switch (c)
798 {
799 /* A number of Perl escapes are not handled by PCRE. We give an explicit
800 error. */
801
802 case CHAR_l:
803 case CHAR_L:
804 *errorcodeptr = ERR37;
805 break;
806
807 case CHAR_u:
808 if ((options & PCRE_JAVASCRIPT_COMPAT) != 0)
809 {
810 /* In JavaScript, \u must be followed by four hexadecimal numbers.
811 Otherwise it is a lowercase u letter. */
812 if (MAX_255(ptr[1]) && (digitab[ptr[1]] & ctype_xdigit) != 0
813 && MAX_255(ptr[2]) && (digitab[ptr[2]] & ctype_xdigit) != 0
814 && MAX_255(ptr[3]) && (digitab[ptr[3]] & ctype_xdigit) != 0
815 && MAX_255(ptr[4]) && (digitab[ptr[4]] & ctype_xdigit) != 0)
816 {
817 c = 0;
818 for (i = 0; i < 4; ++i)
819 {
820 register int cc = *(++ptr);
821 #ifndef EBCDIC /* ASCII/UTF-8 coding */
822 if (cc >= CHAR_a) cc -= 32; /* Convert to upper case */
823 c = (c << 4) + cc - ((cc < CHAR_A)? CHAR_0 : (CHAR_A - 10));
824 #else /* EBCDIC coding */
825 if (cc >= CHAR_a && cc <= CHAR_z) cc += 64; /* Convert to upper case */
826 c = (c << 4) + cc - ((cc >= CHAR_0)? CHAR_0 : (CHAR_A - 10));
827 #endif
828 }
829 }
830 }
831 else
832 *errorcodeptr = ERR37;
833 break;
834
835 case CHAR_U:
836 /* In JavaScript, \U is an uppercase U letter. */
837 if ((options & PCRE_JAVASCRIPT_COMPAT) == 0) *errorcodeptr = ERR37;
838 break;
839
840 /* In a character class, \g is just a literal "g". Outside a character
841 class, \g must be followed by one of a number of specific things:
842
843 (1) A number, either plain or braced. If positive, it is an absolute
844 backreference. If negative, it is a relative backreference. This is a Perl
845 5.10 feature.
846
847 (2) Perl 5.10 also supports \g{name} as a reference to a named group. This
848 is part of Perl's movement towards a unified syntax for back references. As
849 this is synonymous with \k{name}, we fudge it up by pretending it really
850 was \k.
851
852 (3) For Oniguruma compatibility we also support \g followed by a name or a
853 number either in angle brackets or in single quotes. However, these are
854 (possibly recursive) subroutine calls, _not_ backreferences. Just return
855 the -ESC_g code (cf \k). */
856
857 case CHAR_g:
858 if (isclass) break;
859 if (ptr[1] == CHAR_LESS_THAN_SIGN || ptr[1] == CHAR_APOSTROPHE)
860 {
861 c = -ESC_g;
862 break;
863 }
864
865 /* Handle the Perl-compatible cases */
866
867 if (ptr[1] == CHAR_LEFT_CURLY_BRACKET)
868 {
869 const pcre_uchar *p;
870 for (p = ptr+2; *p != 0 && *p != CHAR_RIGHT_CURLY_BRACKET; p++)
871 if (*p != CHAR_MINUS && !IS_DIGIT(*p)) break;
872 if (*p != 0 && *p != CHAR_RIGHT_CURLY_BRACKET)
873 {
874 c = -ESC_k;
875 break;
876 }
877 braced = TRUE;
878 ptr++;
879 }
880 else braced = FALSE;
881
882 if (ptr[1] == CHAR_MINUS)
883 {
884 negated = TRUE;
885 ptr++;
886 }
887 else negated = FALSE;
888
889 /* The integer range is limited by the machine's int representation. */
890 c = 0;
891 while (IS_DIGIT(ptr[1]))
892 {
893 if (((unsigned int)c) > INT_MAX / 10) /* Integer overflow */
894 {
895 c = -1;
896 break;
897 }
898 c = c * 10 + *(++ptr) - CHAR_0;
899 }
900 if (((unsigned int)c) > INT_MAX) /* Integer overflow */
901 {
902 while (IS_DIGIT(ptr[1]))
903 ptr++;
904 *errorcodeptr = ERR61;
905 break;
906 }
907
908 if (braced && *(++ptr) != CHAR_RIGHT_CURLY_BRACKET)
909 {
910 *errorcodeptr = ERR57;
911 break;
912 }
913
914 if (c == 0)
915 {
916 *errorcodeptr = ERR58;
917 break;
918 }
919
920 if (negated)
921 {
922 if (c > bracount)
923 {
924 *errorcodeptr = ERR15;
925 break;
926 }
927 c = bracount - (c - 1);
928 }
929
930 c = -(ESC_REF + c);
931 break;
932
933 /* The handling of escape sequences consisting of a string of digits
934 starting with one that is not zero is not straightforward. By experiment,
935 the way Perl works seems to be as follows:
936
937 Outside a character class, the digits are read as a decimal number. If the
938 number is less than 10, or if there are that many previous extracting
939 left brackets, then it is a back reference. Otherwise, up to three octal
940 digits are read to form an escaped byte. Thus \123 is likely to be octal
941 123 (cf \0123, which is octal 012 followed by the literal 3). If the octal
942 value is greater than 377, the least significant 8 bits are taken. Inside a
943 character class, \ followed by a digit is always an octal number. */
944
945 case CHAR_1: case CHAR_2: case CHAR_3: case CHAR_4: case CHAR_5:
946 case CHAR_6: case CHAR_7: case CHAR_8: case CHAR_9:
947
948 if (!isclass)
949 {
950 oldptr = ptr;
951 /* The integer range is limited by the machine's int representation. */
952 c -= CHAR_0;
953 while (IS_DIGIT(ptr[1]))
954 {
955 if (((unsigned int)c) > INT_MAX / 10) /* Integer overflow */
956 {
957 c = -1;
958 break;
959 }
960 c = c * 10 + *(++ptr) - CHAR_0;
961 }
962 if (((unsigned int)c) > INT_MAX) /* Integer overflow */
963 {
964 while (IS_DIGIT(ptr[1]))
965 ptr++;
966 *errorcodeptr = ERR61;
967 break;
968 }
969 if (c < 10 || c <= bracount)
970 {
971 c = -(ESC_REF + c);
972 break;
973 }
974 ptr = oldptr; /* Put the pointer back and fall through */
975 }
976
977 /* Handle an octal number following \. If the first digit is 8 or 9, Perl
978 generates a binary zero byte and treats the digit as a following literal.
979 Thus we have to pull back the pointer by one. */
980
981 if ((c = *ptr) >= CHAR_8)
982 {
983 ptr--;
984 c = 0;
985 break;
986 }
987
988 /* \0 always starts an octal number, but we may drop through to here with a
989 larger first octal digit. The original code used just to take the least
990 significant 8 bits of octal numbers (I think this is what early Perls used
991 to do). Nowadays we allow for larger numbers in UTF-8 mode, but no more
992 than 3 octal digits. */
993
994 case CHAR_0:
995 c -= CHAR_0;
996 while(i++ < 2 && ptr[1] >= CHAR_0 && ptr[1] <= CHAR_7)
997 c = c * 8 + *(++ptr) - CHAR_0;
998 if (!utf && c > 0xff) *errorcodeptr = ERR51;
999 break;
1000
1001 /* \x is complicated. \x{ddd} is a character number which can be greater
1002 than 0xff in utf or non-8bit mode, but only if the ddd are hex digits.
1003 If not, { is treated as a data character. */
1004
1005 case CHAR_x:
1006 if ((options & PCRE_JAVASCRIPT_COMPAT) != 0)
1007 {
1008 /* In JavaScript, \x must be followed by two hexadecimal numbers.
1009 Otherwise it is a lowercase x letter. */
1010 if (MAX_255(ptr[1]) && (digitab[ptr[1]] & ctype_xdigit) != 0
1011 && MAX_255(ptr[2]) && (digitab[ptr[2]] & ctype_xdigit) != 0)
1012 {
1013 c = 0;
1014 for (i = 0; i < 2; ++i)
1015 {
1016 register int cc = *(++ptr);
1017 #ifndef EBCDIC /* ASCII/UTF-8 coding */
1018 if (cc >= CHAR_a) cc -= 32; /* Convert to upper case */
1019 c = (c << 4) + cc - ((cc < CHAR_A)? CHAR_0 : (CHAR_A - 10));
1020 #else /* EBCDIC coding */
1021 if (cc >= CHAR_a && cc <= CHAR_z) cc += 64; /* Convert to upper case */
1022 c = (c << 4) + cc - ((cc >= CHAR_0)? CHAR_0 : (CHAR_A - 10));
1023 #endif
1024 }
1025 }
1026 break;
1027 }
1028
1029 if (ptr[1] == CHAR_LEFT_CURLY_BRACKET)
1030 {
1031 const pcre_uchar *pt = ptr + 2;
1032
1033 c = 0;
1034 while (MAX_255(*pt) && (digitab[*pt] & ctype_xdigit) != 0)
1035 {
1036 register int cc = *pt++;
1037 if (c == 0 && cc == CHAR_0) continue; /* Leading zeroes */
1038
1039 #ifndef EBCDIC /* ASCII/UTF-8 coding */
1040 if (cc >= CHAR_a) cc -= 32; /* Convert to upper case */
1041 c = (c << 4) + cc - ((cc < CHAR_A)? CHAR_0 : (CHAR_A - 10));
1042 #else /* EBCDIC coding */
1043 if (cc >= CHAR_a && cc <= CHAR_z) cc += 64; /* Convert to upper case */
1044 c = (c << 4) + cc - ((cc >= CHAR_0)? CHAR_0 : (CHAR_A - 10));
1045 #endif
1046
1047 #ifdef COMPILE_PCRE8
1048 if (c > (utf ? 0x10ffff : 0xff)) { c = -1; break; }
1049 #else
1050 #ifdef COMPILE_PCRE16
1051 if (c > (utf ? 0x10ffff : 0xffff)) { c = -1; break; }
1052 #endif
1053 #endif
1054 }
1055
1056 if (c < 0)
1057 {
1058 while (MAX_255(*pt) && (digitab[*pt] & ctype_xdigit) != 0) pt++;
1059 *errorcodeptr = ERR34;
1060 }
1061
1062 if (*pt == CHAR_RIGHT_CURLY_BRACKET)
1063 {
1064 if (utf && c >= 0xd800 && c <= 0xdfff) *errorcodeptr = ERR73;
1065 ptr = pt;
1066 break;
1067 }
1068
1069 /* If the sequence of hex digits does not end with '}', then we don't
1070 recognize this construct; fall through to the normal \x handling. */
1071 }
1072
1073 /* Read just a single-byte hex-defined char */
1074
1075 c = 0;
1076 while (i++ < 2 && MAX_255(ptr[1]) && (digitab[ptr[1]] & ctype_xdigit) != 0)
1077 {
1078 int cc; /* Some compilers don't like */
1079 cc = *(++ptr); /* ++ in initializers */
1080 #ifndef EBCDIC /* ASCII/UTF-8 coding */
1081 if (cc >= CHAR_a) cc -= 32; /* Convert to upper case */
1082 c = c * 16 + cc - ((cc < CHAR_A)? CHAR_0 : (CHAR_A - 10));
1083 #else /* EBCDIC coding */
1084 if (cc <= CHAR_z) cc += 64; /* Convert to upper case */
1085 c = c * 16 + cc - ((cc >= CHAR_0)? CHAR_0 : (CHAR_A - 10));
1086 #endif
1087 }
1088 break;
1089
1090 /* For \c, a following letter is upper-cased; then the 0x40 bit is flipped.
1091 An error is given if the byte following \c is not an ASCII character. This
1092 coding is ASCII-specific, but then the whole concept of \cx is
1093 ASCII-specific. (However, an EBCDIC equivalent has now been added.) */
1094
1095 case CHAR_c:
1096 c = *(++ptr);
1097 if (c == 0)
1098 {
1099 *errorcodeptr = ERR2;
1100 break;
1101 }
1102 #ifndef EBCDIC /* ASCII/UTF-8 coding */
1103 if (c > 127) /* Excludes all non-ASCII in either mode */
1104 {
1105 *errorcodeptr = ERR68;
1106 break;
1107 }
1108 if (c >= CHAR_a && c <= CHAR_z) c -= 32;
1109 c ^= 0x40;
1110 #else /* EBCDIC coding */
1111 if (c >= CHAR_a && c <= CHAR_z) c += 64;
1112 c ^= 0xC0;
1113 #endif
1114 break;
1115
1116 /* PCRE_EXTRA enables extensions to Perl in the matter of escapes. Any
1117 other alphanumeric following \ is an error if PCRE_EXTRA was set;
1118 otherwise, for Perl compatibility, it is a literal. This code looks a bit
1119 odd, but there used to be some cases other than the default, and there may
1120 be again in future, so I haven't "optimized" it. */
1121
1122 default:
1123 if ((options & PCRE_EXTRA) != 0) switch(c)
1124 {
1125 default:
1126 *errorcodeptr = ERR3;
1127 break;
1128 }
1129 break;
1130 }
1131 }
1132
1133 /* Perl supports \N{name} for character names, as well as plain \N for "not
1134 newline". PCRE does not support \N{name}. However, it does support
1135 quantification such as \N{2,3}. */
1136
1137 if (c == -ESC_N && ptr[1] == CHAR_LEFT_CURLY_BRACKET &&
1138 !is_counted_repeat(ptr+2))
1139 *errorcodeptr = ERR37;
1140
1141 /* If PCRE_UCP is set, we change the values for \d etc. */
1142
1143 if ((options & PCRE_UCP) != 0 && c <= -ESC_D && c >= -ESC_w)
1144 c -= (ESC_DU - ESC_D);
1145
1146 /* Set the pointer to the final character before returning. */
1147
1148 *ptrptr = ptr;
1149 return c;
1150 }
1151
1152
1153
1154 #ifdef SUPPORT_UCP
1155 /*************************************************
1156 * Handle \P and \p *
1157 *************************************************/
1158
1159 /* This function is called after \P or \p has been encountered, provided that
1160 PCRE is compiled with support for Unicode properties. On entry, ptrptr is
1161 pointing at the P or p. On exit, it is pointing at the final character of the
1162 escape sequence.
1163
1164 Argument:
1165 ptrptr points to the pattern position pointer
1166 negptr points to a boolean that is set TRUE for negation else FALSE
1167 dptr points to an int that is set to the detailed property value
1168 errorcodeptr points to the error code variable
1169
1170 Returns: type value from ucp_type_table, or -1 for an invalid type
1171 */
1172
1173 static int
1174 get_ucp(const pcre_uchar **ptrptr, BOOL *negptr, int *dptr, int *errorcodeptr)
1175 {
1176 int c, i, bot, top;
1177 const pcre_uchar *ptr = *ptrptr;
1178 pcre_uchar name[32];
1179
1180 c = *(++ptr);
1181 if (c == 0) goto ERROR_RETURN;
1182
1183 *negptr = FALSE;
1184
1185 /* \P or \p can be followed by a name in {}, optionally preceded by ^ for
1186 negation. */
1187
1188 if (c == CHAR_LEFT_CURLY_BRACKET)
1189 {
1190 if (ptr[1] == CHAR_CIRCUMFLEX_ACCENT)
1191 {
1192 *negptr = TRUE;
1193 ptr++;
1194 }
1195 for (i = 0; i < (int)sizeof(name) - 1; i++)
1196 {
1197 c = *(++ptr);
1198 if (c == 0) goto ERROR_RETURN;
1199 if (c == CHAR_RIGHT_CURLY_BRACKET) break;
1200 name[i] = c;
1201 }
1202 if (c != CHAR_RIGHT_CURLY_BRACKET) goto ERROR_RETURN;
1203 name[i] = 0;
1204 }
1205
1206 /* Otherwise there is just one following character */
1207
1208 else
1209 {
1210 name[0] = c;
1211 name[1] = 0;
1212 }
1213
1214 *ptrptr = ptr;
1215
1216 /* Search for a recognized property name using binary chop */
1217
1218 bot = 0;
1219 top = PRIV(utt_size);
1220
1221 while (bot < top)
1222 {
1223 i = (bot + top) >> 1;
1224 c = STRCMP_UC_C8(name, PRIV(utt_names) + PRIV(utt)[i].name_offset);
1225 if (c == 0)
1226 {
1227 *dptr = PRIV(utt)[i].value;
1228 return PRIV(utt)[i].type;
1229 }
1230 if (c > 0) bot = i + 1; else top = i;
1231 }
1232
1233 *errorcodeptr = ERR47;
1234 *ptrptr = ptr;
1235 return -1;
1236
1237 ERROR_RETURN:
1238 *errorcodeptr = ERR46;
1239 *ptrptr = ptr;
1240 return -1;
1241 }
1242 #endif
1243
1244
1245
1246
1247 /*************************************************
1248 * Read repeat counts *
1249 *************************************************/
1250
1251 /* Read an item of the form {n,m} and return the values. This is called only
1252 after is_counted_repeat() has confirmed that a repeat-count quantifier exists,
1253 so the syntax is guaranteed to be correct, but we need to check the values.
1254
1255 Arguments:
1256 p pointer to first char after '{'
1257 minp pointer to int for min
1258 maxp pointer to int for max
1259 returned as -1 if no max
1260 errorcodeptr points to error code variable
1261
1262 Returns: pointer to '}' on success;
1263 current ptr on error, with errorcodeptr set non-zero
1264 */
1265
1266 static const pcre_uchar *
1267 read_repeat_counts(const pcre_uchar *p, int *minp, int *maxp, int *errorcodeptr)
1268 {
1269 int min = 0;
1270 int max = -1;
1271
1272 /* Read the minimum value and do a paranoid check: a negative value indicates
1273 an integer overflow. */
1274
1275 while (IS_DIGIT(*p)) min = min * 10 + *p++ - CHAR_0;
1276 if (min < 0 || min > 65535)
1277 {
1278 *errorcodeptr = ERR5;
1279 return p;
1280 }
1281
1282 /* Read the maximum value if there is one, and again do a paranoid on its size.
1283 Also, max must not be less than min. */
1284
1285 if (*p == CHAR_RIGHT_CURLY_BRACKET) max = min; else
1286 {
1287 if (*(++p) != CHAR_RIGHT_CURLY_BRACKET)
1288 {
1289 max = 0;
1290 while(IS_DIGIT(*p)) max = max * 10 + *p++ - CHAR_0;
1291 if (max < 0 || max > 65535)
1292 {
1293 *errorcodeptr = ERR5;
1294 return p;
1295 }
1296 if (max < min)
1297 {
1298 *errorcodeptr = ERR4;
1299 return p;
1300 }
1301 }
1302 }
1303
1304 /* Fill in the required variables, and pass back the pointer to the terminating
1305 '}'. */
1306
1307 *minp = min;
1308 *maxp = max;
1309 return p;
1310 }
1311
1312
1313
1314 /*************************************************
1315 * Subroutine for finding forward reference *
1316 *************************************************/
1317
1318 /* This recursive function is called only from find_parens() below. The
1319 top-level call starts at the beginning of the pattern. All other calls must
1320 start at a parenthesis. It scans along a pattern's text looking for capturing
1321 subpatterns, and counting them. If it finds a named pattern that matches the
1322 name it is given, it returns its number. Alternatively, if the name is NULL, it
1323 returns when it reaches a given numbered subpattern. Recursion is used to keep
1324 track of subpatterns that reset the capturing group numbers - the (?| feature.
1325
1326 This function was originally called only from the second pass, in which we know
1327 that if (?< or (?' or (?P< is encountered, the name will be correctly
1328 terminated because that is checked in the first pass. There is now one call to
1329 this function in the first pass, to check for a recursive back reference by
1330 name (so that we can make the whole group atomic). In this case, we need check
1331 only up to the current position in the pattern, and that is still OK because
1332 and previous occurrences will have been checked. To make this work, the test
1333 for "end of pattern" is a check against cd->end_pattern in the main loop,
1334 instead of looking for a binary zero. This means that the special first-pass
1335 call can adjust cd->end_pattern temporarily. (Checks for binary zero while
1336 processing items within the loop are OK, because afterwards the main loop will
1337 terminate.)
1338
1339 Arguments:
1340 ptrptr address of the current character pointer (updated)
1341 cd compile background data
1342 name name to seek, or NULL if seeking a numbered subpattern
1343 lorn name length, or subpattern number if name is NULL
1344 xmode TRUE if we are in /x mode
1345 utf TRUE if we are in UTF-8 / UTF-16 mode
1346 count pointer to the current capturing subpattern number (updated)
1347
1348 Returns: the number of the named subpattern, or -1 if not found
1349 */
1350
1351 static int
1352 find_parens_sub(pcre_uchar **ptrptr, compile_data *cd, const pcre_uchar *name, int lorn,
1353 BOOL xmode, BOOL utf, int *count)
1354 {
1355 pcre_uchar *ptr = *ptrptr;
1356 int start_count = *count;
1357 int hwm_count = start_count;
1358 BOOL dup_parens = FALSE;
1359
1360 /* If the first character is a parenthesis, check on the type of group we are
1361 dealing with. The very first call may not start with a parenthesis. */
1362
1363 if (ptr[0] == CHAR_LEFT_PARENTHESIS)
1364 {
1365 /* Handle specials such as (*SKIP) or (*UTF8) etc. */
1366
1367 if (ptr[1] == CHAR_ASTERISK) ptr += 2;
1368
1369 /* Handle a normal, unnamed capturing parenthesis. */
1370
1371 else if (ptr[1] != CHAR_QUESTION_MARK)
1372 {
1373 *count += 1;
1374 if (name == NULL && *count == lorn) return *count;
1375 ptr++;
1376 }
1377
1378 /* All cases now have (? at the start. Remember when we are in a group
1379 where the parenthesis numbers are duplicated. */
1380
1381 else if (ptr[2] == CHAR_VERTICAL_LINE)
1382 {
1383 ptr += 3;
1384 dup_parens = TRUE;
1385 }
1386
1387 /* Handle comments; all characters are allowed until a ket is reached. */
1388
1389 else if (ptr[2] == CHAR_NUMBER_SIGN)
1390 {
1391 for (ptr += 3; *ptr != 0; ptr++) if (*ptr == CHAR_RIGHT_PARENTHESIS) break;
1392 goto FAIL_EXIT;
1393 }
1394
1395 /* Handle a condition. If it is an assertion, just carry on so that it
1396 is processed as normal. If not, skip to the closing parenthesis of the
1397 condition (there can't be any nested parens). */
1398
1399 else if (ptr[2] == CHAR_LEFT_PARENTHESIS)
1400 {
1401 ptr += 2;
1402 if (ptr[1] != CHAR_QUESTION_MARK)
1403 {
1404 while (*ptr != 0 && *ptr != CHAR_RIGHT_PARENTHESIS) ptr++;
1405 if (*ptr != 0) ptr++;
1406 }
1407 }
1408
1409 /* Start with (? but not a condition. */
1410
1411 else
1412 {
1413 ptr += 2;
1414 if (*ptr == CHAR_P) ptr++; /* Allow optional P */
1415
1416 /* We have to disambiguate (?<! and (?<= from (?<name> for named groups */
1417
1418 if ((*ptr == CHAR_LESS_THAN_SIGN && ptr[1] != CHAR_EXCLAMATION_MARK &&
1419 ptr[1] != CHAR_EQUALS_SIGN) || *ptr == CHAR_APOSTROPHE)
1420 {
1421 int term;
1422 const pcre_uchar *thisname;
1423 *count += 1;
1424 if (name == NULL && *count == lorn) return *count;
1425 term = *ptr++;
1426 if (term == CHAR_LESS_THAN_SIGN) term = CHAR_GREATER_THAN_SIGN;
1427 thisname = ptr;
1428 while (*ptr != term) ptr++;
1429 if (name != NULL && lorn == ptr - thisname &&
1430 STRNCMP_UC_UC(name, thisname, lorn) == 0)
1431 return *count;
1432 term++;
1433 }
1434 }
1435 }
1436
1437 /* Past any initial parenthesis handling, scan for parentheses or vertical
1438 bars. Stop if we get to cd->end_pattern. Note that this is important for the
1439 first-pass call when this value is temporarily adjusted to stop at the current
1440 position. So DO NOT change this to a test for binary zero. */
1441
1442 for (; ptr < cd->end_pattern; ptr++)
1443 {
1444 /* Skip over backslashed characters and also entire \Q...\E */
1445
1446 if (*ptr == CHAR_BACKSLASH)
1447 {
1448 if (*(++ptr) == 0) goto FAIL_EXIT;
1449 if (*ptr == CHAR_Q) for (;;)
1450 {
1451 while (*(++ptr) != 0 && *ptr != CHAR_BACKSLASH) {};
1452 if (*ptr == 0) goto FAIL_EXIT;
1453 if (*(++ptr) == CHAR_E) break;
1454 }
1455 continue;
1456 }
1457
1458 /* Skip over character classes; this logic must be similar to the way they
1459 are handled for real. If the first character is '^', skip it. Also, if the
1460 first few characters (either before or after ^) are \Q\E or \E we skip them
1461 too. This makes for compatibility with Perl. Note the use of STR macros to
1462 encode "Q\\E" so that it works in UTF-8 on EBCDIC platforms. */
1463
1464 if (*ptr == CHAR_LEFT_SQUARE_BRACKET)
1465 {
1466 BOOL negate_class = FALSE;
1467 for (;;)
1468 {
1469 if (ptr[1] == CHAR_BACKSLASH)
1470 {
1471 if (ptr[2] == CHAR_E)
1472 ptr+= 2;
1473 else if (STRNCMP_UC_C8(ptr + 2,
1474 STR_Q STR_BACKSLASH STR_E, 3) == 0)
1475 ptr += 4;
1476 else
1477 break;
1478 }
1479 else if (!negate_class && ptr[1] == CHAR_CIRCUMFLEX_ACCENT)
1480 {
1481 negate_class = TRUE;
1482 ptr++;
1483 }
1484 else break;
1485 }
1486
1487 /* If the next character is ']', it is a data character that must be
1488 skipped, except in JavaScript compatibility mode. */
1489
1490 if (ptr[1] == CHAR_RIGHT_SQUARE_BRACKET &&
1491 (cd->external_options & PCRE_JAVASCRIPT_COMPAT) == 0)
1492 ptr++;
1493
1494 while (*(++ptr) != CHAR_RIGHT_SQUARE_BRACKET)
1495 {
1496 if (*ptr == 0) return -1;
1497 if (*ptr == CHAR_BACKSLASH)
1498 {
1499 if (*(++ptr) == 0) goto FAIL_EXIT;
1500 if (*ptr == CHAR_Q) for (;;)
1501 {
1502 while (*(++ptr) != 0 && *ptr != CHAR_BACKSLASH) {};
1503 if (*ptr == 0) goto FAIL_EXIT;
1504 if (*(++ptr) == CHAR_E) break;
1505 }
1506 continue;
1507 }
1508 }
1509 continue;
1510 }
1511
1512 /* Skip comments in /x mode */
1513
1514 if (xmode && *ptr == CHAR_NUMBER_SIGN)
1515 {
1516 ptr++;
1517 while (*ptr != 0)
1518 {
1519 if (IS_NEWLINE(ptr)) { ptr += cd->nllen - 1; break; }
1520 ptr++;
1521 #ifdef SUPPORT_UTF
1522 if (utf) FORWARDCHAR(ptr);
1523 #endif
1524 }
1525 if (*ptr == 0) goto FAIL_EXIT;
1526 continue;
1527 }
1528
1529 /* Check for the special metacharacters */
1530
1531 if (*ptr == CHAR_LEFT_PARENTHESIS)
1532 {
1533 int rc = find_parens_sub(&ptr, cd, name, lorn, xmode, utf, count);
1534 if (rc > 0) return rc;
1535 if (*ptr == 0) goto FAIL_EXIT;
1536 }
1537
1538 else if (*ptr == CHAR_RIGHT_PARENTHESIS)
1539 {
1540 if (dup_parens && *count < hwm_count) *count = hwm_count;
1541 goto FAIL_EXIT;
1542 }
1543
1544 else if (*ptr == CHAR_VERTICAL_LINE && dup_parens)
1545 {
1546 if (*count > hwm_count) hwm_count = *count;
1547 *count = start_count;
1548 }
1549 }
1550
1551 FAIL_EXIT:
1552 *ptrptr = ptr;
1553 return -1;
1554 }
1555
1556
1557
1558
1559 /*************************************************
1560 * Find forward referenced subpattern *
1561 *************************************************/
1562
1563 /* This function scans along a pattern's text looking for capturing
1564 subpatterns, and counting them. If it finds a named pattern that matches the
1565 name it is given, it returns its number. Alternatively, if the name is NULL, it
1566 returns when it reaches a given numbered subpattern. This is used for forward
1567 references to subpatterns. We used to be able to start this scan from the
1568 current compiling point, using the current count value from cd->bracount, and
1569 do it all in a single loop, but the addition of the possibility of duplicate
1570 subpattern numbers means that we have to scan from the very start, in order to
1571 take account of such duplicates, and to use a recursive function to keep track
1572 of the different types of group.
1573
1574 Arguments:
1575 cd compile background data
1576 name name to seek, or NULL if seeking a numbered subpattern
1577 lorn name length, or subpattern number if name is NULL
1578 xmode TRUE if we are in /x mode
1579 utf TRUE if we are in UTF-8 / UTF-16 mode
1580
1581 Returns: the number of the found subpattern, or -1 if not found
1582 */
1583
1584 static int
1585 find_parens(compile_data *cd, const pcre_uchar *name, int lorn, BOOL xmode,
1586 BOOL utf)
1587 {
1588 pcre_uchar *ptr = (pcre_uchar *)cd->start_pattern;
1589 int count = 0;
1590 int rc;
1591
1592 /* If the pattern does not start with an opening parenthesis, the first call
1593 to find_parens_sub() will scan right to the end (if necessary). However, if it
1594 does start with a parenthesis, find_parens_sub() will return when it hits the
1595 matching closing parens. That is why we have to have a loop. */
1596
1597 for (;;)
1598 {
1599 rc = find_parens_sub(&ptr, cd, name, lorn, xmode, utf, &count);
1600 if (rc > 0 || *ptr++ == 0) break;
1601 }
1602
1603 return rc;
1604 }
1605
1606
1607
1608
1609 /*************************************************
1610 * Find first significant op code *
1611 *************************************************/
1612
1613 /* This is called by several functions that scan a compiled expression looking
1614 for a fixed first character, or an anchoring op code etc. It skips over things
1615 that do not influence this. For some calls, it makes sense to skip negative
1616 forward and all backward assertions, and also the \b assertion; for others it
1617 does not.
1618
1619 Arguments:
1620 code pointer to the start of the group
1621 skipassert TRUE if certain assertions are to be skipped
1622
1623 Returns: pointer to the first significant opcode
1624 */
1625
1626 static const pcre_uchar*
1627 first_significant_code(const pcre_uchar *code, BOOL skipassert)
1628 {
1629 for (;;)
1630 {
1631 switch ((int)*code)
1632 {
1633 case OP_ASSERT_NOT:
1634 case OP_ASSERTBACK:
1635 case OP_ASSERTBACK_NOT:
1636 if (!skipassert) return code;
1637 do code += GET(code, 1); while (*code == OP_ALT);
1638 code += PRIV(OP_lengths)[*code];
1639 break;
1640
1641 case OP_WORD_BOUNDARY:
1642 case OP_NOT_WORD_BOUNDARY:
1643 if (!skipassert) return code;
1644 /* Fall through */
1645
1646 case OP_CALLOUT:
1647 case OP_CREF:
1648 case OP_NCREF:
1649 case OP_RREF:
1650 case OP_NRREF:
1651 case OP_DEF:
1652 code += PRIV(OP_lengths)[*code];
1653 break;
1654
1655 default:
1656 return code;
1657 }
1658 }
1659 /* Control never reaches here */
1660 }
1661
1662
1663
1664
1665 /*************************************************
1666 * Find the fixed length of a branch *
1667 *************************************************/
1668
1669 /* Scan a branch and compute the fixed length of subject that will match it,
1670 if the length is fixed. This is needed for dealing with backward assertions.
1671 In UTF8 mode, the result is in characters rather than bytes. The branch is
1672 temporarily terminated with OP_END when this function is called.
1673
1674 This function is called when a backward assertion is encountered, so that if it
1675 fails, the error message can point to the correct place in the pattern.
1676 However, we cannot do this when the assertion contains subroutine calls,
1677 because they can be forward references. We solve this by remembering this case
1678 and doing the check at the end; a flag specifies which mode we are running in.
1679
1680 Arguments:
1681 code points to the start of the pattern (the bracket)
1682 utf TRUE in UTF-8 / UTF-16 mode
1683 atend TRUE if called when the pattern is complete
1684 cd the "compile data" structure
1685
1686 Returns: the fixed length,
1687 or -1 if there is no fixed length,
1688 or -2 if \C was encountered (in UTF-8 mode only)
1689 or -3 if an OP_RECURSE item was encountered and atend is FALSE
1690 or -4 if an unknown opcode was encountered (internal error)
1691 */
1692
1693 static int
1694 find_fixedlength(pcre_uchar *code, BOOL utf, BOOL atend, compile_data *cd)
1695 {
1696 int length = -1;
1697
1698 register int branchlength = 0;
1699 register pcre_uchar *cc = code + 1 + LINK_SIZE;
1700
1701 /* Scan along the opcodes for this branch. If we get to the end of the
1702 branch, check the length against that of the other branches. */
1703
1704 for (;;)
1705 {
1706 int d;
1707 pcre_uchar *ce, *cs;
1708 register int op = *cc;
1709 switch (op)
1710 {
1711 /* We only need to continue for OP_CBRA (normal capturing bracket) and
1712 OP_BRA (normal non-capturing bracket) because the other variants of these
1713 opcodes are all concerned with unlimited repeated groups, which of course
1714 are not of fixed length. */
1715
1716 case OP_CBRA:
1717 case OP_BRA:
1718 case OP_ONCE:
1719 case OP_ONCE_NC:
1720 case OP_COND:
1721 d = find_fixedlength(cc + ((op == OP_CBRA)? IMM2_SIZE : 0), utf, atend, cd);
1722 if (d < 0) return d;
1723 branchlength += d;
1724 do cc += GET(cc, 1); while (*cc == OP_ALT);
1725 cc += 1 + LINK_SIZE;
1726 break;
1727
1728 /* Reached end of a branch; if it's a ket it is the end of a nested call.
1729 If it's ALT it is an alternation in a nested call. An ACCEPT is effectively
1730 an ALT. If it is END it's the end of the outer call. All can be handled by
1731 the same code. Note that we must not include the OP_KETRxxx opcodes here,
1732 because they all imply an unlimited repeat. */
1733
1734 case OP_ALT:
1735 case OP_KET:
1736 case OP_END:
1737 case OP_ACCEPT:
1738 case OP_ASSERT_ACCEPT:
1739 if (length < 0) length = branchlength;
1740 else if (length != branchlength) return -1;
1741 if (*cc != OP_ALT) return length;
1742 cc += 1 + LINK_SIZE;
1743 branchlength = 0;
1744 break;
1745
1746 /* A true recursion implies not fixed length, but a subroutine call may
1747 be OK. If the subroutine is a forward reference, we can't deal with
1748 it until the end of the pattern, so return -3. */
1749
1750 case OP_RECURSE:
1751 if (!atend) return -3;
1752 cs = ce = (pcre_uchar *)cd->start_code + GET(cc, 1); /* Start subpattern */
1753 do ce += GET(ce, 1); while (*ce == OP_ALT); /* End subpattern */
1754 if (cc > cs && cc < ce) return -1; /* Recursion */
1755 d = find_fixedlength(cs + 2, utf, atend, cd);
1756 if (d < 0) return d;
1757 branchlength += d;
1758 cc += 1 + LINK_SIZE;
1759 break;
1760
1761 /* Skip over assertive subpatterns */
1762
1763 case OP_ASSERT:
1764 case OP_ASSERT_NOT:
1765 case OP_ASSERTBACK:
1766 case OP_ASSERTBACK_NOT:
1767 do cc += GET(cc, 1); while (*cc == OP_ALT);
1768 /* Fall through */
1769
1770 /* Skip over things that don't match chars */
1771
1772 case OP_MARK:
1773 case OP_PRUNE_ARG:
1774 case OP_SKIP_ARG:
1775 case OP_THEN_ARG:
1776 cc += cc[1] + PRIV(OP_lengths)[*cc];
1777 break;
1778
1779 case OP_CALLOUT:
1780 case OP_CIRC:
1781 case OP_CIRCM:
1782 case OP_CLOSE:
1783 case OP_COMMIT:
1784 case OP_CREF:
1785 case OP_DEF:
1786 case OP_DOLL:
1787 case OP_DOLLM:
1788 case OP_EOD:
1789 case OP_EODN:
1790 case OP_FAIL:
1791 case OP_NCREF:
1792 case OP_NRREF:
1793 case OP_NOT_WORD_BOUNDARY:
1794 case OP_PRUNE:
1795 case OP_REVERSE:
1796 case OP_RREF:
1797 case OP_SET_SOM:
1798 case OP_SKIP:
1799 case OP_SOD:
1800 case OP_SOM:
1801 case OP_THEN:
1802 case OP_WORD_BOUNDARY:
1803 cc += PRIV(OP_lengths)[*cc];
1804 break;
1805
1806 /* Handle literal characters */
1807
1808 case OP_CHAR:
1809 case OP_CHARI:
1810 case OP_NOT:
1811 case OP_NOTI:
1812 branchlength++;
1813 cc += 2;
1814 #ifdef SUPPORT_UTF
1815 if (utf && HAS_EXTRALEN(cc[-1])) cc += GET_EXTRALEN(cc[-1]);
1816 #endif
1817 break;
1818
1819 /* Handle exact repetitions. The count is already in characters, but we
1820 need to skip over a multibyte character in UTF8 mode. */
1821
1822 case OP_EXACT:
1823 case OP_EXACTI:
1824 case OP_NOTEXACT:
1825 case OP_NOTEXACTI:
1826 branchlength += GET2(cc,1);
1827 cc += 2 + IMM2_SIZE;
1828 #ifdef SUPPORT_UTF
1829 if (utf && HAS_EXTRALEN(cc[-1])) cc += GET_EXTRALEN(cc[-1]);
1830 #endif
1831 break;
1832
1833 case OP_TYPEEXACT:
1834 branchlength += GET2(cc,1);
1835 if (cc[1 + IMM2_SIZE] == OP_PROP || cc[1 + IMM2_SIZE] == OP_NOTPROP) cc += 2;
1836 cc += 1 + IMM2_SIZE + 1;
1837 break;
1838
1839 /* Handle single-char matchers */
1840
1841 case OP_PROP:
1842 case OP_NOTPROP:
1843 cc += 2;
1844 /* Fall through */
1845
1846 case OP_HSPACE:
1847 case OP_VSPACE:
1848 case OP_NOT_HSPACE:
1849 case OP_NOT_VSPACE:
1850 case OP_NOT_DIGIT:
1851 case OP_DIGIT:
1852 case OP_NOT_WHITESPACE:
1853 case OP_WHITESPACE:
1854 case OP_NOT_WORDCHAR:
1855 case OP_WORDCHAR:
1856 case OP_ANY:
1857 case OP_ALLANY:
1858 branchlength++;
1859 cc++;
1860 break;
1861
1862 /* The single-byte matcher isn't allowed. This only happens in UTF-8 mode;
1863 otherwise \C is coded as OP_ALLANY. */
1864
1865 case OP_ANYBYTE:
1866 return -2;
1867
1868 /* Check a class for variable quantification */
1869
1870 #if defined SUPPORT_UTF || defined COMPILE_PCRE16
1871 case OP_XCLASS:
1872 cc += GET(cc, 1) - PRIV(OP_lengths)[OP_CLASS];
1873 /* Fall through */
1874 #endif
1875
1876 case OP_CLASS:
1877 case OP_NCLASS:
1878 cc += PRIV(OP_lengths)[OP_CLASS];
1879
1880 switch (*cc)
1881 {
1882 case OP_CRPLUS:
1883 case OP_CRMINPLUS:
1884 case OP_CRSTAR:
1885 case OP_CRMINSTAR:
1886 case OP_CRQUERY:
1887 case OP_CRMINQUERY:
1888 return -1;
1889
1890 case OP_CRRANGE:
1891 case OP_CRMINRANGE:
1892 if (GET2(cc,1) != GET2(cc,1+IMM2_SIZE)) return -1;
1893 branchlength += GET2(cc,1);
1894 cc += 1 + 2 * IMM2_SIZE;
1895 break;
1896
1897 default:
1898 branchlength++;
1899 }
1900 break;
1901
1902 /* Anything else is variable length */
1903
1904 case OP_ANYNL:
1905 case OP_BRAMINZERO:
1906 case OP_BRAPOS:
1907 case OP_BRAPOSZERO:
1908 case OP_BRAZERO:
1909 case OP_CBRAPOS:
1910 case OP_EXTUNI:
1911 case OP_KETRMAX:
1912 case OP_KETRMIN:
1913 case OP_KETRPOS:
1914 case OP_MINPLUS:
1915 case OP_MINPLUSI:
1916 case OP_MINQUERY:
1917 case OP_MINQUERYI:
1918 case OP_MINSTAR:
1919 case OP_MINSTARI:
1920 case OP_MINUPTO:
1921 case OP_MINUPTOI:
1922 case OP_NOTMINPLUS:
1923 case OP_NOTMINPLUSI:
1924 case OP_NOTMINQUERY:
1925 case OP_NOTMINQUERYI:
1926 case OP_NOTMINSTAR:
1927 case OP_NOTMINSTARI:
1928 case OP_NOTMINUPTO:
1929 case OP_NOTMINUPTOI:
1930 case OP_NOTPLUS:
1931 case OP_NOTPLUSI:
1932 case OP_NOTPOSPLUS:
1933 case OP_NOTPOSPLUSI:
1934 case OP_NOTPOSQUERY:
1935 case OP_NOTPOSQUERYI:
1936 case OP_NOTPOSSTAR:
1937 case OP_NOTPOSSTARI:
1938 case OP_NOTPOSUPTO:
1939 case OP_NOTPOSUPTOI:
1940 case OP_NOTQUERY:
1941 case OP_NOTQUERYI:
1942 case OP_NOTSTAR:
1943 case OP_NOTSTARI:
1944 case OP_NOTUPTO:
1945 case OP_NOTUPTOI:
1946 case OP_PLUS:
1947 case OP_PLUSI:
1948 case OP_POSPLUS:
1949 case OP_POSPLUSI:
1950 case OP_POSQUERY:
1951 case OP_POSQUERYI:
1952 case OP_POSSTAR:
1953 case OP_POSSTARI:
1954 case OP_POSUPTO:
1955 case OP_POSUPTOI:
1956 case OP_QUERY:
1957 case OP_QUERYI:
1958 case OP_REF:
1959 case OP_REFI:
1960 case OP_SBRA:
1961 case OP_SBRAPOS:
1962 case OP_SCBRA:
1963 case OP_SCBRAPOS:
1964 case OP_SCOND:
1965 case OP_SKIPZERO:
1966 case OP_STAR:
1967 case OP_STARI:
1968 case OP_TYPEMINPLUS:
1969 case OP_TYPEMINQUERY:
1970 case OP_TYPEMINSTAR:
1971 case OP_TYPEMINUPTO:
1972 case OP_TYPEPLUS:
1973 case OP_TYPEPOSPLUS:
1974 case OP_TYPEPOSQUERY:
1975 case OP_TYPEPOSSTAR:
1976 case OP_TYPEPOSUPTO:
1977 case OP_TYPEQUERY:
1978 case OP_TYPESTAR:
1979 case OP_TYPEUPTO:
1980 case OP_UPTO:
1981 case OP_UPTOI:
1982 return -1;
1983
1984 /* Catch unrecognized opcodes so that when new ones are added they
1985 are not forgotten, as has happened in the past. */
1986
1987 default:
1988 return -4;
1989 }
1990 }
1991 /* Control never gets here */
1992 }
1993
1994
1995
1996
1997 /*************************************************
1998 * Scan compiled regex for specific bracket *
1999 *************************************************/
2000
2001 /* This little function scans through a compiled pattern until it finds a
2002 capturing bracket with the given number, or, if the number is negative, an
2003 instance of OP_REVERSE for a lookbehind. The function is global in the C sense
2004 so that it can be called from pcre_study() when finding the minimum matching
2005 length.
2006
2007 Arguments:
2008 code points to start of expression
2009 utf TRUE in UTF-8 / UTF-16 mode
2010 number the required bracket number or negative to find a lookbehind
2011
2012 Returns: pointer to the opcode for the bracket, or NULL if not found
2013 */
2014
2015 const pcre_uchar *
2016 PRIV(find_bracket)(const pcre_uchar *code, BOOL utf, int number)
2017 {
2018 for (;;)
2019 {
2020 register int c = *code;
2021
2022 if (c == OP_END) return NULL;
2023
2024 /* XCLASS is used for classes that cannot be represented just by a bit
2025 map. This includes negated single high-valued characters. The length in
2026 the table is zero; the actual length is stored in the compiled code. */
2027
2028 if (c == OP_XCLASS) code += GET(code, 1);
2029
2030 /* Handle recursion */
2031
2032 else if (c == OP_REVERSE)
2033 {
2034 if (number < 0) return (pcre_uchar *)code;
2035 code += PRIV(OP_lengths)[c];
2036 }
2037
2038 /* Handle capturing bracket */
2039
2040 else if (c == OP_CBRA || c == OP_SCBRA ||
2041 c == OP_CBRAPOS || c == OP_SCBRAPOS)
2042 {
2043 int n = GET2(code, 1+LINK_SIZE);
2044 if (n == number) return (pcre_uchar *)code;
2045 code += PRIV(OP_lengths)[c];
2046 }
2047
2048 /* Otherwise, we can get the item's length from the table, except that for
2049 repeated character types, we have to test for \p and \P, which have an extra
2050 two bytes of parameters, and for MARK/PRUNE/SKIP/THEN with an argument, we
2051 must add in its length. */
2052
2053 else
2054 {
2055 switch(c)
2056 {
2057 case OP_TYPESTAR:
2058 case OP_TYPEMINSTAR:
2059 case OP_TYPEPLUS:
2060 case OP_TYPEMINPLUS:
2061 case OP_TYPEQUERY:
2062 case OP_TYPEMINQUERY:
2063 case OP_TYPEPOSSTAR:
2064 case OP_TYPEPOSPLUS:
2065 case OP_TYPEPOSQUERY:
2066 if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
2067 break;
2068
2069 case OP_TYPEUPTO:
2070 case OP_TYPEMINUPTO:
2071 case OP_TYPEEXACT:
2072 case OP_TYPEPOSUPTO:
2073 if (code[1 + IMM2_SIZE] == OP_PROP
2074 || code[1 + IMM2_SIZE] == OP_NOTPROP) code += 2;
2075 break;
2076
2077 case OP_MARK:
2078 case OP_PRUNE_ARG:
2079 case OP_SKIP_ARG:
2080 code += code[1];
2081 break;
2082
2083 case OP_THEN_ARG:
2084 code += code[1];
2085 break;
2086 }
2087
2088 /* Add in the fixed length from the table */
2089
2090 code += PRIV(OP_lengths)[c];
2091
2092 /* In UTF-8 mode, opcodes that are followed by a character may be followed by
2093 a multi-byte character. The length in the table is a minimum, so we have to
2094 arrange to skip the extra bytes. */
2095
2096 #ifdef SUPPORT_UTF
2097 if (utf) switch(c)
2098 {
2099 case OP_CHAR:
2100 case OP_CHARI:
2101 case OP_EXACT:
2102 case OP_EXACTI:
2103 case OP_UPTO:
2104 case OP_UPTOI:
2105 case OP_MINUPTO:
2106 case OP_MINUPTOI:
2107 case OP_POSUPTO:
2108 case OP_POSUPTOI:
2109 case OP_STAR:
2110 case OP_STARI:
2111 case OP_MINSTAR:
2112 case OP_MINSTARI:
2113 case OP_POSSTAR:
2114 case OP_POSSTARI:
2115 case OP_PLUS:
2116 case OP_PLUSI:
2117 case OP_MINPLUS:
2118 case OP_MINPLUSI:
2119 case OP_POSPLUS:
2120 case OP_POSPLUSI:
2121 case OP_QUERY:
2122 case OP_QUERYI:
2123 case OP_MINQUERY:
2124 case OP_MINQUERYI:
2125 case OP_POSQUERY:
2126 case OP_POSQUERYI:
2127 if (HAS_EXTRALEN(code[-1])) code += GET_EXTRALEN(code[-1]);
2128 break;
2129 }
2130 #else
2131 (void)(utf); /* Keep compiler happy by referencing function argument */
2132 #endif
2133 }
2134 }
2135 }
2136
2137
2138
2139 /*************************************************
2140 * Scan compiled regex for recursion reference *
2141 *************************************************/
2142
2143 /* This little function scans through a compiled pattern until it finds an
2144 instance of OP_RECURSE.
2145
2146 Arguments:
2147 code points to start of expression
2148 utf TRUE in UTF-8 / UTF-16 mode
2149
2150 Returns: pointer to the opcode for OP_RECURSE, or NULL if not found
2151 */
2152
2153 static const pcre_uchar *
2154 find_recurse(const pcre_uchar *code, BOOL utf)
2155 {
2156 for (;;)
2157 {
2158 register int c = *code;
2159 if (c == OP_END) return NULL;
2160 if (c == OP_RECURSE) return code;
2161
2162 /* XCLASS is used for classes that cannot be represented just by a bit
2163 map. This includes negated single high-valued characters. The length in
2164 the table is zero; the actual length is stored in the compiled code. */
2165
2166 if (c == OP_XCLASS) code += GET(code, 1);
2167
2168 /* Otherwise, we can get the item's length from the table, except that for
2169 repeated character types, we have to test for \p and \P, which have an extra
2170 two bytes of parameters, and for MARK/PRUNE/SKIP/THEN with an argument, we
2171 must add in its length. */
2172
2173 else
2174 {
2175 switch(c)
2176 {
2177 case OP_TYPESTAR:
2178 case OP_TYPEMINSTAR:
2179 case OP_TYPEPLUS:
2180 case OP_TYPEMINPLUS:
2181 case OP_TYPEQUERY:
2182 case OP_TYPEMINQUERY:
2183 case OP_TYPEPOSSTAR:
2184 case OP_TYPEPOSPLUS:
2185 case OP_TYPEPOSQUERY:
2186 if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
2187 break;
2188
2189 case OP_TYPEPOSUPTO:
2190 case OP_TYPEUPTO:
2191 case OP_TYPEMINUPTO:
2192 case OP_TYPEEXACT:
2193 if (code[1 + IMM2_SIZE] == OP_PROP
2194 || code[1 + IMM2_SIZE] == OP_NOTPROP) code += 2;
2195 break;
2196
2197 case OP_MARK:
2198 case OP_PRUNE_ARG:
2199 case OP_SKIP_ARG:
2200 code += code[1];
2201 break;
2202
2203 case OP_THEN_ARG:
2204 code += code[1];
2205 break;
2206 }
2207
2208 /* Add in the fixed length from the table */
2209
2210 code += PRIV(OP_lengths)[c];
2211
2212 /* In UTF-8 mode, opcodes that are followed by a character may be followed
2213 by a multi-byte character. The length in the table is a minimum, so we have
2214 to arrange to skip the extra bytes. */
2215
2216 #ifdef SUPPORT_UTF
2217 if (utf) switch(c)
2218 {
2219 case OP_CHAR:
2220 case OP_CHARI:
2221 case OP_EXACT:
2222 case OP_EXACTI:
2223 case OP_UPTO:
2224 case OP_UPTOI:
2225 case OP_MINUPTO:
2226 case OP_MINUPTOI:
2227 case OP_POSUPTO:
2228 case OP_POSUPTOI:
2229 case OP_STAR:
2230 case OP_STARI:
2231 case OP_MINSTAR:
2232 case OP_MINSTARI:
2233 case OP_POSSTAR:
2234 case OP_POSSTARI:
2235 case OP_PLUS:
2236 case OP_PLUSI:
2237 case OP_MINPLUS:
2238 case OP_MINPLUSI:
2239 case OP_POSPLUS:
2240 case OP_POSPLUSI:
2241 case OP_QUERY:
2242 case OP_QUERYI:
2243 case OP_MINQUERY:
2244 case OP_MINQUERYI:
2245 case OP_POSQUERY:
2246 case OP_POSQUERYI:
2247 if (HAS_EXTRALEN(code[-1])) code += GET_EXTRALEN(code[-1]);
2248 break;
2249 }
2250 #else
2251 (void)(utf); /* Keep compiler happy by referencing function argument */
2252 #endif
2253 }
2254 }
2255 }
2256
2257
2258
2259 /*************************************************
2260 * Scan compiled branch for non-emptiness *
2261 *************************************************/
2262
2263 /* This function scans through a branch of a compiled pattern to see whether it
2264 can match the empty string or not. It is called from could_be_empty()
2265 below and from compile_branch() when checking for an unlimited repeat of a
2266 group that can match nothing. Note that first_significant_code() skips over
2267 backward and negative forward assertions when its final argument is TRUE. If we
2268 hit an unclosed bracket, we return "empty" - this means we've struck an inner
2269 bracket whose current branch will already have been scanned.
2270
2271 Arguments:
2272 code points to start of search
2273 endcode points to where to stop
2274 utf TRUE if in UTF-8 / UTF-16 mode
2275 cd contains pointers to tables etc.
2276
2277 Returns: TRUE if what is matched could be empty
2278 */
2279
2280 static BOOL
2281 could_be_empty_branch(const pcre_uchar *code, const pcre_uchar *endcode,
2282 BOOL utf, compile_data *cd)
2283 {
2284 register int c;
2285 for (code = first_significant_code(code + PRIV(OP_lengths)[*code], TRUE);
2286 code < endcode;
2287 code = first_significant_code(code + PRIV(OP_lengths)[c], TRUE))
2288 {
2289 const pcre_uchar *ccode;
2290
2291 c = *code;
2292
2293 /* Skip over forward assertions; the other assertions are skipped by
2294 first_significant_code() with a TRUE final argument. */
2295
2296 if (c == OP_ASSERT)
2297 {
2298 do code += GET(code, 1); while (*code == OP_ALT);
2299 c = *code;
2300 continue;
2301 }
2302
2303 /* For a recursion/subroutine call, if its end has been reached, which
2304 implies a backward reference subroutine call, we can scan it. If it's a
2305 forward reference subroutine call, we can't. To detect forward reference
2306 we have to scan up the list that is kept in the workspace. This function is
2307 called only when doing the real compile, not during the pre-compile that
2308 measures the size of the compiled pattern. */
2309
2310 if (c == OP_RECURSE)
2311 {
2312 const pcre_uchar *scode;
2313 BOOL empty_branch;
2314
2315 /* Test for forward reference */
2316
2317 for (scode = cd->start_workspace; scode < cd->hwm; scode += LINK_SIZE)
2318 if (GET(scode, 0) == code + 1 - cd->start_code) return TRUE;
2319
2320 /* Not a forward reference, test for completed backward reference */
2321
2322 empty_branch = FALSE;
2323 scode = cd->start_code + GET(code, 1);
2324 if (GET(scode, 1) == 0) return TRUE; /* Unclosed */
2325
2326 /* Completed backwards reference */
2327
2328 do
2329 {
2330 if (could_be_empty_branch(scode, endcode, utf, cd))
2331 {
2332 empty_branch = TRUE;
2333 break;
2334 }
2335 scode += GET(scode, 1);
2336 }
2337 while (*scode == OP_ALT);
2338
2339 if (!empty_branch) return FALSE; /* All branches are non-empty */
2340 continue;
2341 }
2342
2343 /* Groups with zero repeats can of course be empty; skip them. */
2344
2345 if (c == OP_BRAZERO || c == OP_BRAMINZERO || c == OP_SKIPZERO ||
2346 c == OP_BRAPOSZERO)
2347 {
2348 code += PRIV(OP_lengths)[c];
2349 do code += GET(code, 1); while (*code == OP_ALT);
2350 c = *code;
2351 continue;
2352 }
2353
2354 /* A nested group that is already marked as "could be empty" can just be
2355 skipped. */
2356
2357 if (c == OP_SBRA || c == OP_SBRAPOS ||
2358 c == OP_SCBRA || c == OP_SCBRAPOS)
2359 {
2360 do code += GET(code, 1); while (*code == OP_ALT);
2361 c = *code;
2362 continue;
2363 }
2364
2365 /* For other groups, scan the branches. */
2366
2367 if (c == OP_BRA || c == OP_BRAPOS ||
2368 c == OP_CBRA || c == OP_CBRAPOS ||
2369 c == OP_ONCE || c == OP_ONCE_NC ||
2370 c == OP_COND)
2371 {
2372 BOOL empty_branch;
2373 if (GET(code, 1) == 0) return TRUE; /* Hit unclosed bracket */
2374
2375 /* If a conditional group has only one branch, there is a second, implied,
2376 empty branch, so just skip over the conditional, because it could be empty.
2377 Otherwise, scan the individual branches of the group. */
2378
2379 if (c == OP_COND && code[GET(code, 1)] != OP_ALT)
2380 code += GET(code, 1);
2381 else
2382 {
2383 empty_branch = FALSE;
2384 do
2385 {
2386 if (!empty_branch && could_be_empty_branch(code, endcode, utf, cd))
2387 empty_branch = TRUE;
2388 code += GET(code, 1);
2389 }
2390 while (*code == OP_ALT);
2391 if (!empty_branch) return FALSE; /* All branches are non-empty */
2392 }
2393
2394 c = *code;
2395 continue;
2396 }
2397
2398 /* Handle the other opcodes */
2399
2400 switch (c)
2401 {
2402 /* Check for quantifiers after a class. XCLASS is used for classes that
2403 cannot be represented just by a bit map. This includes negated single
2404 high-valued characters. The length in PRIV(OP_lengths)[] is zero; the
2405 actual length is stored in the compiled code, so we must update "code"
2406 here. */
2407
2408 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
2409 case OP_XCLASS:
2410 ccode = code += GET(code, 1);
2411 goto CHECK_CLASS_REPEAT;
2412 #endif
2413
2414 case OP_CLASS:
2415 case OP_NCLASS:
2416 ccode = code + PRIV(OP_lengths)[OP_CLASS];
2417
2418 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
2419 CHECK_CLASS_REPEAT:
2420 #endif
2421
2422 switch (*ccode)
2423 {
2424 case OP_CRSTAR: /* These could be empty; continue */
2425 case OP_CRMINSTAR:
2426 case OP_CRQUERY:
2427 case OP_CRMINQUERY:
2428 break;
2429
2430 default: /* Non-repeat => class must match */
2431 case OP_CRPLUS: /* These repeats aren't empty */
2432 case OP_CRMINPLUS:
2433 return FALSE;
2434
2435 case OP_CRRANGE:
2436 case OP_CRMINRANGE:
2437 if (GET2(ccode, 1) > 0) return FALSE; /* Minimum > 0 */
2438 break;
2439 }
2440 break;
2441
2442 /* Opcodes that must match a character */
2443
2444 case OP_PROP:
2445 case OP_NOTPROP:
2446 case OP_EXTUNI:
2447 case OP_NOT_DIGIT:
2448 case OP_DIGIT:
2449 case OP_NOT_WHITESPACE:
2450 case OP_WHITESPACE:
2451 case OP_NOT_WORDCHAR:
2452 case OP_WORDCHAR:
2453 case OP_ANY:
2454 case OP_ALLANY:
2455 case OP_ANYBYTE:
2456 case OP_CHAR:
2457 case OP_CHARI:
2458 case OP_NOT:
2459 case OP_NOTI:
2460 case OP_PLUS:
2461 case OP_MINPLUS:
2462 case OP_POSPLUS:
2463 case OP_EXACT:
2464 case OP_NOTPLUS:
2465 case OP_NOTMINPLUS:
2466 case OP_NOTPOSPLUS:
2467 case OP_NOTEXACT:
2468 case OP_TYPEPLUS:
2469 case OP_TYPEMINPLUS:
2470 case OP_TYPEPOSPLUS:
2471 case OP_TYPEEXACT:
2472 return FALSE;
2473
2474 /* These are going to continue, as they may be empty, but we have to
2475 fudge the length for the \p and \P cases. */
2476
2477 case OP_TYPESTAR:
2478 case OP_TYPEMINSTAR:
2479 case OP_TYPEPOSSTAR:
2480 case OP_TYPEQUERY:
2481 case OP_TYPEMINQUERY:
2482 case OP_TYPEPOSQUERY:
2483 if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
2484 break;
2485
2486 /* Same for these */
2487
2488 case OP_TYPEUPTO:
2489 case OP_TYPEMINUPTO:
2490 case OP_TYPEPOSUPTO:
2491 if (code[1 + IMM2_SIZE] == OP_PROP
2492 || code[1 + IMM2_SIZE] == OP_NOTPROP) code += 2;
2493 break;
2494
2495 /* End of branch */
2496
2497 case OP_KET:
2498 case OP_KETRMAX:
2499 case OP_KETRMIN:
2500 case OP_KETRPOS:
2501 case OP_ALT:
2502 return TRUE;
2503
2504 /* In UTF-8 mode, STAR, MINSTAR, POSSTAR, QUERY, MINQUERY, POSQUERY, UPTO,
2505 MINUPTO, and POSUPTO may be followed by a multibyte character */
2506
2507 #ifdef SUPPORT_UTF
2508 case OP_STAR:
2509 case OP_STARI:
2510 case OP_MINSTAR:
2511 case OP_MINSTARI:
2512 case OP_POSSTAR:
2513 case OP_POSSTARI:
2514 case OP_QUERY:
2515 case OP_QUERYI:
2516 case OP_MINQUERY:
2517 case OP_MINQUERYI:
2518 case OP_POSQUERY:
2519 case OP_POSQUERYI:
2520 if (utf && HAS_EXTRALEN(code[1])) code += GET_EXTRALEN(code[1]);
2521 break;
2522
2523 case OP_UPTO:
2524 case OP_UPTOI:
2525 case OP_MINUPTO:
2526 case OP_MINUPTOI:
2527 case OP_POSUPTO:
2528 case OP_POSUPTOI:
2529 if (utf && HAS_EXTRALEN(code[1 + IMM2_SIZE])) code += GET_EXTRALEN(code[1 + IMM2_SIZE]);
2530 break;
2531 #endif
2532
2533 /* MARK, and PRUNE/SKIP/THEN with an argument must skip over the argument
2534 string. */
2535
2536 case OP_MARK:
2537 case OP_PRUNE_ARG:
2538 case OP_SKIP_ARG:
2539 code += code[1];
2540 break;
2541
2542 case OP_THEN_ARG:
2543 code += code[1];
2544 break;
2545
2546 /* None of the remaining opcodes are required to match a character. */
2547
2548 default:
2549 break;
2550 }
2551 }
2552
2553 return TRUE;
2554 }
2555
2556
2557
2558 /*************************************************
2559 * Scan compiled regex for non-emptiness *
2560 *************************************************/
2561
2562 /* This function is called to check for left recursive calls. We want to check
2563 the current branch of the current pattern to see if it could match the empty
2564 string. If it could, we must look outwards for branches at other levels,
2565 stopping when we pass beyond the bracket which is the subject of the recursion.
2566 This function is called only during the real compile, not during the
2567 pre-compile.
2568
2569 Arguments:
2570 code points to start of the recursion
2571 endcode points to where to stop (current RECURSE item)
2572 bcptr points to the chain of current (unclosed) branch starts
2573 utf TRUE if in UTF-8 / UTF-16 mode
2574 cd pointers to tables etc
2575
2576 Returns: TRUE if what is matched could be empty
2577 */
2578
2579 static BOOL
2580 could_be_empty(const pcre_uchar *code, const pcre_uchar *endcode,
2581 branch_chain *bcptr, BOOL utf, compile_data *cd)
2582 {
2583 while (bcptr != NULL && bcptr->current_branch >= code)
2584 {
2585 if (!could_be_empty_branch(bcptr->current_branch, endcode, utf, cd))
2586 return FALSE;
2587 bcptr = bcptr->outer;
2588 }
2589 return TRUE;
2590 }
2591
2592
2593
2594 /*************************************************
2595 * Check for POSIX class syntax *
2596 *************************************************/
2597
2598 /* This function is called when the sequence "[:" or "[." or "[=" is
2599 encountered in a character class. It checks whether this is followed by a
2600 sequence of characters terminated by a matching ":]" or ".]" or "=]". If we
2601 reach an unescaped ']' without the special preceding character, return FALSE.
2602
2603 Originally, this function only recognized a sequence of letters between the
2604 terminators, but it seems that Perl recognizes any sequence of characters,
2605 though of course unknown POSIX names are subsequently rejected. Perl gives an
2606 "Unknown POSIX class" error for [:f\oo:] for example, where previously PCRE
2607 didn't consider this to be a POSIX class. Likewise for [:1234:].
2608
2609 The problem in trying to be exactly like Perl is in the handling of escapes. We
2610 have to be sure that [abc[:x\]pqr] is *not* treated as containing a POSIX
2611 class, but [abc[:x\]pqr:]] is (so that an error can be generated). The code
2612 below handles the special case of \], but does not try to do any other escape
2613 processing. This makes it different from Perl for cases such as [:l\ower:]
2614 where Perl recognizes it as the POSIX class "lower" but PCRE does not recognize
2615 "l\ower". This is a lesser evil that not diagnosing bad classes when Perl does,
2616 I think.
2617
2618 A user pointed out that PCRE was rejecting [:a[:digit:]] whereas Perl was not.
2619 It seems that the appearance of a nested POSIX class supersedes an apparent
2620 external class. For example, [:a[:digit:]b:] matches "a", "b", ":", or
2621 a digit.
2622
2623 In Perl, unescaped square brackets may also appear as part of class names. For
2624 example, [:a[:abc]b:] gives unknown POSIX class "[:abc]b:]". However, for
2625 [:a[:abc]b][b:] it gives unknown POSIX class "[:abc]b][b:]", which does not
2626 seem right at all. PCRE does not allow closing square brackets in POSIX class
2627 names.
2628
2629 Arguments:
2630 ptr pointer to the initial [
2631 endptr where to return the end pointer
2632
2633 Returns: TRUE or FALSE
2634 */
2635
2636 static BOOL
2637 check_posix_syntax(const pcre_uchar *ptr, const pcre_uchar **endptr)
2638 {
2639 int terminator; /* Don't combine these lines; the Solaris cc */
2640 terminator = *(++ptr); /* compiler warns about "non-constant" initializer. */
2641 for (++ptr; *ptr != 0; ptr++)
2642 {
2643 if (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET)
2644 ptr++;
2645 else if (*ptr == CHAR_RIGHT_SQUARE_BRACKET) return FALSE;
2646 else
2647 {
2648 if (*ptr == terminator && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET)
2649 {
2650 *endptr = ptr;
2651 return TRUE;
2652 }
2653 if (*ptr == CHAR_LEFT_SQUARE_BRACKET &&
2654 (ptr[1] == CHAR_COLON || ptr[1] == CHAR_DOT ||
2655 ptr[1] == CHAR_EQUALS_SIGN) &&
2656 check_posix_syntax(ptr, endptr))
2657 return FALSE;
2658 }
2659 }
2660 return FALSE;
2661 }
2662
2663
2664
2665
2666 /*************************************************
2667 * Check POSIX class name *
2668 *************************************************/
2669
2670 /* This function is called to check the name given in a POSIX-style class entry
2671 such as [:alnum:].
2672
2673 Arguments:
2674 ptr points to the first letter
2675 len the length of the name
2676
2677 Returns: a value representing the name, or -1 if unknown
2678 */
2679
2680 static int
2681 check_posix_name(const pcre_uchar *ptr, int len)
2682 {
2683 const char *pn = posix_names;
2684 register int yield = 0;
2685 while (posix_name_lengths[yield] != 0)
2686 {
2687 if (len == posix_name_lengths[yield] &&
2688 STRNCMP_UC_C8(ptr, pn, len) == 0) return yield;
2689 pn += posix_name_lengths[yield] + 1;
2690 yield++;
2691 }
2692 return -1;
2693 }
2694
2695
2696 /*************************************************
2697 * Adjust OP_RECURSE items in repeated group *
2698 *************************************************/
2699
2700 /* OP_RECURSE items contain an offset from the start of the regex to the group
2701 that is referenced. This means that groups can be replicated for fixed
2702 repetition simply by copying (because the recursion is allowed to refer to
2703 earlier groups that are outside the current group). However, when a group is
2704 optional (i.e. the minimum quantifier is zero), OP_BRAZERO or OP_SKIPZERO is
2705 inserted before it, after it has been compiled. This means that any OP_RECURSE
2706 items within it that refer to the group itself or any contained groups have to
2707 have their offsets adjusted. That one of the jobs of this function. Before it
2708 is called, the partially compiled regex must be temporarily terminated with
2709 OP_END.
2710
2711 This function has been extended with the possibility of forward references for
2712 recursions and subroutine calls. It must also check the list of such references
2713 for the group we are dealing with. If it finds that one of the recursions in
2714 the current group is on this list, it adjusts the offset in the list, not the
2715 value in the reference (which is a group number).
2716
2717 Arguments:
2718 group points to the start of the group
2719 adjust the amount by which the group is to be moved
2720 utf TRUE in UTF-8 / UTF-16 mode
2721 cd contains pointers to tables etc.
2722 save_hwm the hwm forward reference pointer at the start of the group
2723
2724 Returns: nothing
2725 */
2726
2727 static void
2728 adjust_recurse(pcre_uchar *group, int adjust, BOOL utf, compile_data *cd,
2729 pcre_uchar *save_hwm)
2730 {
2731 pcre_uchar *ptr = group;
2732
2733 while ((ptr = (pcre_uchar *)find_recurse(ptr, utf)) != NULL)
2734 {
2735 int offset;
2736 pcre_uchar *hc;
2737
2738 /* See if this recursion is on the forward reference list. If so, adjust the
2739 reference. */
2740
2741 for (hc = save_hwm; hc < cd->hwm; hc += LINK_SIZE)
2742 {
2743 offset = GET(hc, 0);
2744 if (cd->start_code + offset == ptr + 1)
2745 {
2746 PUT(hc, 0, offset + adjust);
2747 break;
2748 }
2749 }
2750
2751 /* Otherwise, adjust the recursion offset if it's after the start of this
2752 group. */
2753
2754 if (hc >= cd->hwm)
2755 {
2756 offset = GET(ptr, 1);
2757 if (cd->start_code + offset >= group) PUT(ptr, 1, offset + adjust);
2758 }
2759
2760 ptr += 1 + LINK_SIZE;
2761 }
2762 }
2763
2764
2765
2766 /*************************************************
2767 * Insert an automatic callout point *
2768 *************************************************/
2769
2770 /* This function is called when the PCRE_AUTO_CALLOUT option is set, to insert
2771 callout points before each pattern item.
2772
2773 Arguments:
2774 code current code pointer
2775 ptr current pattern pointer
2776 cd pointers to tables etc
2777
2778 Returns: new code pointer
2779 */
2780
2781 static pcre_uchar *
2782 auto_callout(pcre_uchar *code, const pcre_uchar *ptr, compile_data *cd)
2783 {
2784 *code++ = OP_CALLOUT;
2785 *code++ = 255;
2786 PUT(code, 0, (int)(ptr - cd->start_pattern)); /* Pattern offset */
2787 PUT(code, LINK_SIZE, 0); /* Default length */
2788 return code + 2 * LINK_SIZE;
2789 }
2790
2791
2792
2793 /*************************************************
2794 * Complete a callout item *
2795 *************************************************/
2796
2797 /* A callout item contains the length of the next item in the pattern, which
2798 we can't fill in till after we have reached the relevant point. This is used
2799 for both automatic and manual callouts.
2800
2801 Arguments:
2802 previous_callout points to previous callout item
2803 ptr current pattern pointer
2804 cd pointers to tables etc
2805
2806 Returns: nothing
2807 */
2808
2809 static void
2810 complete_callout(pcre_uchar *previous_callout, const pcre_uchar *ptr, compile_data *cd)
2811 {
2812 int length = (int)(ptr - cd->start_pattern - GET(previous_callout, 2));
2813 PUT(previous_callout, 2 + LINK_SIZE, length);
2814 }
2815
2816
2817
2818 #ifdef SUPPORT_UCP
2819 /*************************************************
2820 * Get othercase range *
2821 *************************************************/
2822
2823 /* This function is passed the start and end of a class range, in UTF-8 mode
2824 with UCP support. It searches up the characters, looking for internal ranges of
2825 characters in the "other" case. Each call returns the next one, updating the
2826 start address.
2827
2828 Arguments:
2829 cptr points to starting character value; updated
2830 d end value
2831 ocptr where to put start of othercase range
2832 odptr where to put end of othercase range
2833
2834 Yield: TRUE when range returned; FALSE when no more
2835 */
2836
2837 static BOOL
2838 get_othercase_range(unsigned int *cptr, unsigned int d, unsigned int *ocptr,
2839 unsigned int *odptr)
2840 {
2841 unsigned int c, othercase, next;
2842
2843 for (c = *cptr; c <= d; c++)
2844 { if ((othercase = UCD_OTHERCASE(c)) != c) break; }
2845
2846 if (c > d) return FALSE;
2847
2848 *ocptr = othercase;
2849 next = othercase + 1;
2850
2851 for (++c; c <= d; c++)
2852 {
2853 if (UCD_OTHERCASE(c) != next) break;
2854 next++;
2855 }
2856
2857 *odptr = next - 1;
2858 *cptr = c;
2859
2860 return TRUE;
2861 }
2862
2863
2864
2865 /*************************************************
2866 * Check a character and a property *
2867 *************************************************/
2868
2869 /* This function is called by check_auto_possessive() when a property item
2870 is adjacent to a fixed character.
2871
2872 Arguments:
2873 c the character
2874 ptype the property type
2875 pdata the data for the type
2876 negated TRUE if it's a negated property (\P or \p{^)
2877
2878 Returns: TRUE if auto-possessifying is OK
2879 */
2880
2881 static BOOL
2882 check_char_prop(int c, int ptype, int pdata, BOOL negated)
2883 {
2884 const ucd_record *prop = GET_UCD(c);
2885 switch(ptype)
2886 {
2887 case PT_LAMP:
2888 return (prop->chartype == ucp_Lu ||
2889 prop->chartype == ucp_Ll ||
2890 prop->chartype == ucp_Lt) == negated;
2891
2892 case PT_GC:
2893 return (pdata == PRIV(ucp_gentype)[prop->chartype]) == negated;
2894
2895 case PT_PC:
2896 return (pdata == prop->chartype) == negated;
2897
2898 case PT_SC:
2899 return (pdata == prop->script) == negated;
2900
2901 /* These are specials */
2902
2903 case PT_ALNUM:
2904 return (PRIV(ucp_gentype)[prop->chartype] == ucp_L ||
2905 PRIV(ucp_gentype)[prop->chartype] == ucp_N) == negated;
2906
2907 case PT_SPACE: /* Perl space */
2908 return (PRIV(ucp_gentype)[prop->chartype] == ucp_Z ||
2909 c == CHAR_HT || c == CHAR_NL || c == CHAR_FF || c == CHAR_CR)
2910 == negated;
2911
2912 case PT_PXSPACE: /* POSIX space */
2913 return (PRIV(ucp_gentype)[prop->chartype] == ucp_Z ||
2914 c == CHAR_HT || c == CHAR_NL || c == CHAR_VT ||
2915 c == CHAR_FF || c == CHAR_CR)
2916 == negated;
2917
2918 case PT_WORD:
2919 return (PRIV(ucp_gentype)[prop->chartype] == ucp_L ||
2920 PRIV(ucp_gentype)[prop->chartype] == ucp_N ||
2921 c == CHAR_UNDERSCORE) == negated;
2922 }
2923 return FALSE;
2924 }
2925 #endif /* SUPPORT_UCP */
2926
2927
2928
2929 /*************************************************
2930 * Check if auto-possessifying is possible *
2931 *************************************************/
2932
2933 /* This function is called for unlimited repeats of certain items, to see
2934 whether the next thing could possibly match the repeated item. If not, it makes
2935 sense to automatically possessify the repeated item.
2936
2937 Arguments:
2938 previous pointer to the repeated opcode
2939 utf TRUE in UTF-8 / UTF-16 mode
2940 ptr next character in pattern
2941 options options bits
2942 cd contains pointers to tables etc.
2943
2944 Returns: TRUE if possessifying is wanted
2945 */
2946
2947 static BOOL
2948 check_auto_possessive(const pcre_uchar *previous, BOOL utf,
2949 const pcre_uchar *ptr, int options, compile_data *cd)
2950 {
2951 pcre_int32 c, next;
2952 int op_code = *previous++;
2953
2954 /* Skip whitespace and comments in extended mode */
2955
2956 if ((options & PCRE_EXTENDED) != 0)
2957 {
2958 for (;;)
2959 {
2960 while (MAX_255(*ptr) && (cd->ctypes[*ptr] & ctype_space) != 0) ptr++;
2961 if (*ptr == CHAR_NUMBER_SIGN)
2962 {
2963 ptr++;
2964 while (*ptr != 0)
2965 {
2966 if (IS_NEWLINE(ptr)) { ptr += cd->nllen; break; }
2967 ptr++;
2968 #ifdef SUPPORT_UTF
2969 if (utf) FORWARDCHAR(ptr);
2970 #endif
2971 }
2972 }
2973 else break;
2974 }
2975 }
2976
2977 /* If the next item is one that we can handle, get its value. A non-negative
2978 value is a character, a negative value is an escape value. */
2979
2980 if (*ptr == CHAR_BACKSLASH)
2981 {
2982 int temperrorcode = 0;
2983 next = check_escape(&ptr, &temperrorcode, cd->bracount, options, FALSE);
2984 if (temperrorcode != 0) return FALSE;
2985 ptr++; /* Point after the escape sequence */
2986 }
2987 else if (!MAX_255(*ptr) || (cd->ctypes[*ptr] & ctype_meta) == 0)
2988 {
2989 #ifdef SUPPORT_UTF
2990 if (utf) { GETCHARINC(next, ptr); } else
2991 #endif
2992 next = *ptr++;
2993 }
2994 else return FALSE;
2995
2996 /* Skip whitespace and comments in extended mode */
2997
2998 if ((options & PCRE_EXTENDED) != 0)
2999 {
3000 for (;;)
3001 {
3002 while (MAX_255(*ptr) && (cd->ctypes[*ptr] & ctype_space) != 0) ptr++;
3003 if (*ptr == CHAR_NUMBER_SIGN)
3004 {
3005 ptr++;
3006 while (*ptr != 0)
3007 {
3008 if (IS_NEWLINE(ptr)) { ptr += cd->nllen; break; }
3009 ptr++;
3010 #ifdef SUPPORT_UTF
3011 if (utf) FORWARDCHAR(ptr);
3012 #endif
3013 }
3014 }
3015 else break;
3016 }
3017 }
3018
3019 /* If the next thing is itself optional, we have to give up. */
3020
3021 if (*ptr == CHAR_ASTERISK || *ptr == CHAR_QUESTION_MARK ||
3022 STRNCMP_UC_C8(ptr, STR_LEFT_CURLY_BRACKET STR_0 STR_COMMA, 3) == 0)
3023 return FALSE;
3024
3025 /* Now compare the next item with the previous opcode. First, handle cases when
3026 the next item is a character. */
3027
3028 if (next >= 0) switch(op_code)
3029 {
3030 case OP_CHAR:
3031 #ifdef SUPPORT_UTF
3032 GETCHARTEST(c, previous);
3033 #else
3034 c = *previous;
3035 #endif
3036 return c != next;
3037
3038 /* For CHARI (caseless character) we must check the other case. If we have
3039 Unicode property support, we can use it to test the other case of
3040 high-valued characters. */
3041
3042 case OP_CHARI:
3043 #ifdef SUPPORT_UTF
3044 GETCHARTEST(c, previous);
3045 #else
3046 c = *previous;
3047 #endif
3048 if (c == next) return FALSE;
3049 #ifdef SUPPORT_UTF
3050 if (utf)
3051 {
3052 unsigned int othercase;
3053 if (next < 128) othercase = cd->fcc[next]; else
3054 #ifdef SUPPORT_UCP
3055 othercase = UCD_OTHERCASE((unsigned int)next);
3056 #else
3057 othercase = NOTACHAR;
3058 #endif
3059 return (unsigned int)c != othercase;
3060 }
3061 else
3062 #endif /* SUPPORT_UTF */
3063 return (c != cd->fcc[next]); /* Non-UTF-8 mode */
3064
3065 /* For OP_NOT and OP_NOTI, the data is always a single-byte character. These
3066 opcodes are not used for multi-byte characters, because they are coded using
3067 an XCLASS instead. */
3068
3069 case OP_NOT:
3070 return (c = *previous) == next;
3071
3072 case OP_NOTI:
3073 if ((c = *previous) == next) return TRUE;
3074 #ifdef SUPPORT_UTF
3075 if (utf)
3076 {
3077 unsigned int othercase;
3078 if (next < 128) othercase = cd->fcc[next]; else
3079 #ifdef SUPPORT_UCP
3080 othercase = UCD_OTHERCASE(next);
3081 #else
3082 othercase = NOTACHAR;
3083 #endif
3084 return (unsigned int)c == othercase;
3085 }
3086 else
3087 #endif /* SUPPORT_UTF */
3088 return (c == cd->fcc[next]); /* Non-UTF-8 mode */
3089
3090 /* Note that OP_DIGIT etc. are generated only when PCRE_UCP is *not* set.
3091 When it is set, \d etc. are converted into OP_(NOT_)PROP codes. */
3092
3093 case OP_DIGIT:
3094 return next > 127 || (cd->ctypes[next] & ctype_digit) == 0;
3095
3096 case OP_NOT_DIGIT:
3097 return next <= 127 && (cd->ctypes[next] & ctype_digit) != 0;
3098
3099 case OP_WHITESPACE:
3100 return next > 127 || (cd->ctypes[next] & ctype_space) == 0;
3101
3102 case OP_NOT_WHITESPACE:
3103 return next <= 127 && (cd->ctypes[next] & ctype_space) != 0;
3104
3105 case OP_WORDCHAR:
3106 return next > 127 || (cd->ctypes[next] & ctype_word) == 0;
3107
3108 case OP_NOT_WORDCHAR:
3109 return next <= 127 && (cd->ctypes[next] & ctype_word) != 0;
3110
3111 case OP_HSPACE:
3112 case OP_NOT_HSPACE:
3113 switch(next)
3114 {
3115 case 0x09:
3116 case 0x20:
3117 case 0xa0:
3118 case 0x1680:
3119 case 0x180e:
3120 case 0x2000:
3121 case 0x2001:
3122 case 0x2002:
3123 case 0x2003:
3124 case 0x2004:
3125 case 0x2005:
3126 case 0x2006:
3127 case 0x2007:
3128 case 0x2008:
3129 case 0x2009:
3130 case 0x200A:
3131 case 0x202f:
3132 case 0x205f:
3133 case 0x3000:
3134 return op_code == OP_NOT_HSPACE;
3135 default:
3136 return op_code != OP_NOT_HSPACE;
3137 }
3138
3139 case OP_ANYNL:
3140 case OP_VSPACE:
3141 case OP_NOT_VSPACE:
3142 switch(next)
3143 {
3144 case 0x0a:
3145 case 0x0b:
3146 case 0x0c:
3147 case 0x0d:
3148 case 0x85:
3149 case 0x2028:
3150 case 0x2029:
3151 return op_code == OP_NOT_VSPACE;
3152 default:
3153 return op_code != OP_NOT_VSPACE;
3154 }
3155
3156 #ifdef SUPPORT_UCP
3157 case OP_PROP:
3158 return check_char_prop(next, previous[0], previous[1], FALSE);
3159
3160 case OP_NOTPROP:
3161 return check_char_prop(next, previous[0], previous[1], TRUE);
3162 #endif
3163
3164 default:
3165 return FALSE;
3166 }
3167
3168
3169 /* Handle the case when the next item is \d, \s, etc. Note that when PCRE_UCP
3170 is set, \d turns into ESC_du rather than ESC_d, etc., so ESC_d etc. are
3171 generated only when PCRE_UCP is *not* set, that is, when only ASCII
3172 characteristics are recognized. Similarly, the opcodes OP_DIGIT etc. are
3173 replaced by OP_PROP codes when PCRE_UCP is set. */
3174
3175 switch(op_code)
3176 {
3177 case OP_CHAR:
3178 case OP_CHARI:
3179 #ifdef SUPPORT_UTF
3180 GETCHARTEST(c, previous);
3181 #else
3182 c = *previous;
3183 #endif
3184 switch(-next)
3185 {
3186 case ESC_d:
3187 return c > 127 || (cd->ctypes[c] & ctype_digit) == 0;
3188
3189 case ESC_D:
3190 return c <= 127 && (cd->ctypes[c] & ctype_digit) != 0;
3191
3192 case ESC_s:
3193 return c > 127 || (cd->ctypes[c] & ctype_space) == 0;
3194
3195 case ESC_S:
3196 return c <= 127 && (cd->ctypes[c] & ctype_space) != 0;
3197
3198 case ESC_w:
3199 return c > 127 || (cd->ctypes[c] & ctype_word) == 0;
3200
3201 case ESC_W:
3202 return c <= 127 && (cd->ctypes[c] & ctype_word) != 0;
3203
3204 case ESC_h:
3205 case ESC_H:
3206 switch(c)
3207 {
3208 case 0x09:
3209 case 0x20:
3210 case 0xa0:
3211 case 0x1680:
3212 case 0x180e:
3213 case 0x2000:
3214 case 0x2001:
3215 case 0x2002:
3216 case 0x2003:
3217 case 0x2004:
3218 case 0x2005:
3219 case 0x2006:
3220 case 0x2007:
3221 case 0x2008:
3222 case 0x2009:
3223 case 0x200A:
3224 case 0x202f:
3225 case 0x205f:
3226 case 0x3000:
3227 return -next != ESC_h;
3228 default:
3229 return -next == ESC_h;
3230 }
3231
3232 case ESC_v:
3233 case ESC_V:
3234 switch(c)
3235 {
3236 case 0x0a:
3237 case 0x0b:
3238 case 0x0c:
3239 case 0x0d:
3240 case 0x85:
3241 case 0x2028:
3242 case 0x2029:
3243 return -next != ESC_v;
3244 default:
3245 return -next == ESC_v;
3246 }
3247
3248 /* When PCRE_UCP is set, these values get generated for \d etc. Find
3249 their substitutions and process them. The result will always be either
3250 -ESC_p or -ESC_P. Then fall through to process those values. */
3251
3252 #ifdef SUPPORT_UCP
3253 case ESC_du:
3254 case ESC_DU:
3255 case ESC_wu:
3256 case ESC_WU:
3257 case ESC_su:
3258 case ESC_SU:
3259 {
3260 int temperrorcode = 0;
3261 ptr = substitutes[-next - ESC_DU];
3262 next = check_escape(&ptr, &temperrorcode, 0, options, FALSE);
3263 if (temperrorcode != 0) return FALSE;
3264 ptr++; /* For compatibility */
3265 }
3266 /* Fall through */
3267
3268 case ESC_p:
3269 case ESC_P:
3270 {
3271 int ptype, pdata, errorcodeptr;
3272 BOOL negated;
3273
3274 ptr--; /* Make ptr point at the p or P */
3275 ptype = get_ucp(&ptr, &negated, &pdata, &errorcodeptr);
3276 if (ptype < 0) return FALSE;
3277 ptr++; /* Point past the final curly ket */
3278
3279 /* If the property item is optional, we have to give up. (When generated
3280 from \d etc by PCRE_UCP, this test will have been applied much earlier,
3281 to the original \d etc. At this point, ptr will point to a zero byte. */
3282
3283 if (*ptr == CHAR_ASTERISK || *ptr == CHAR_QUESTION_MARK ||
3284 STRNCMP_UC_C8(ptr, STR_LEFT_CURLY_BRACKET STR_0 STR_COMMA, 3) == 0)
3285 return FALSE;
3286
3287 /* Do the property check. */
3288
3289 return check_char_prop(c, ptype, pdata, (next == -ESC_P) != negated);
3290 }
3291 #endif
3292
3293 default:
3294 return FALSE;
3295 }
3296
3297 /* In principle, support for Unicode properties should be integrated here as
3298 well. It means re-organizing the above code so as to get hold of the property
3299 values before switching on the op-code. However, I wonder how many patterns
3300 combine ASCII \d etc with Unicode properties? (Note that if PCRE_UCP is set,
3301 these op-codes are never generated.) */
3302
3303 case OP_DIGIT:
3304 return next == -ESC_D || next == -ESC_s || next == -ESC_W ||
3305 next == -ESC_h || next == -ESC_v || next == -ESC_R;
3306
3307 case OP_NOT_DIGIT:
3308 return next == -ESC_d;
3309
3310 case OP_WHITESPACE:
3311 return next == -ESC_S || next == -ESC_d || next == -ESC_w || next == -ESC_R;
3312
3313 case OP_NOT_WHITESPACE:
3314 return next == -ESC_s || next == -ESC_h || next == -ESC_v;
3315
3316 case OP_HSPACE:
3317 return next == -ESC_S || next == -ESC_H || next == -ESC_d ||
3318 next == -ESC_w || next == -ESC_v || next == -ESC_R;
3319
3320 case OP_NOT_HSPACE:
3321 return next == -ESC_h;
3322
3323 /* Can't have \S in here because VT matches \S (Perl anomaly) */
3324 case OP_ANYNL:
3325 case OP_VSPACE:
3326 return next == -ESC_V || next == -ESC_d || next == -ESC_w;
3327
3328 case OP_NOT_VSPACE:
3329 return next == -ESC_v || next == -ESC_R;
3330
3331 case OP_WORDCHAR:
3332 return next == -ESC_W || next == -ESC_s || next == -ESC_h ||
3333 next == -ESC_v || next == -ESC_R;
3334
3335 case OP_NOT_WORDCHAR:
3336 return next == -ESC_w || next == -ESC_d;
3337
3338 default:
3339 return FALSE;
3340 }
3341
3342 /* Control does not reach here */
3343 }
3344
3345
3346
3347 /*************************************************
3348 * Compile one branch *
3349 *************************************************/
3350
3351 /* Scan the pattern, compiling it into the a vector. If the options are
3352 changed during the branch, the pointer is used to change the external options
3353 bits. This function is used during the pre-compile phase when we are trying
3354 to find out the amount of memory needed, as well as during the real compile
3355 phase. The value of lengthptr distinguishes the two phases.
3356
3357 Arguments:
3358 optionsptr pointer to the option bits
3359 codeptr points to the pointer to the current code point
3360 ptrptr points to the current pattern pointer
3361 errorcodeptr points to error code variable
3362 firstcharptr set to initial literal character, or < 0 (REQ_UNSET, REQ_NONE)
3363 reqcharptr set to the last literal character required, else < 0
3364 bcptr points to current branch chain
3365 cond_depth conditional nesting depth
3366 cd contains pointers to tables etc.
3367 lengthptr NULL during the real compile phase
3368 points to length accumulator during pre-compile phase
3369
3370 Returns: TRUE on success
3371 FALSE, with *errorcodeptr set non-zero on error
3372 */
3373
3374 static BOOL
3375 compile_branch(int *optionsptr, pcre_uchar **codeptr,
3376 const pcre_uchar **ptrptr, int *errorcodeptr, pcre_int32 *firstcharptr,
3377 pcre_int32 *reqcharptr, branch_chain *bcptr, int cond_depth,
3378 compile_data *cd, int *lengthptr)
3379 {
3380 int repeat_type, op_type;
3381 int repeat_min = 0, repeat_max = 0; /* To please picky compilers */
3382 int bravalue = 0;
3383 int greedy_default, greedy_non_default;
3384 pcre_int32 firstchar, reqchar;
3385 pcre_int32 zeroreqchar, zerofirstchar;
3386 pcre_int32 req_caseopt, reqvary, tempreqvary;
3387 int options = *optionsptr; /* May change dynamically */
3388 int after_manual_callout = 0;
3389 int length_prevgroup = 0;
3390 register int c;
3391 register pcre_uchar *code = *codeptr;
3392 pcre_uchar *last_code = code;
3393 pcre_uchar *orig_code = code;
3394 pcre_uchar *tempcode;
3395 BOOL inescq = FALSE;
3396 BOOL groupsetfirstchar = FALSE;
3397 const pcre_uchar *ptr = *ptrptr;
3398 const pcre_uchar *tempptr;
3399 const pcre_uchar *nestptr = NULL;
3400 pcre_uchar *previous = NULL;
3401 pcre_uchar *previous_callout = NULL;
3402 pcre_uchar *save_hwm = NULL;
3403 pcre_uint8 classbits[32];
3404
3405 /* We can fish out the UTF-8 setting once and for all into a BOOL, but we
3406 must not do this for other options (e.g. PCRE_EXTENDED) because they may change
3407 dynamically as we process the pattern. */
3408
3409 #ifdef SUPPORT_UTF
3410 /* PCRE_UTF16 has the same value as PCRE_UTF8. */
3411 BOOL utf = (options & PCRE_UTF8) != 0;
3412 pcre_uchar utf_chars[6];
3413 #else
3414 BOOL utf = FALSE;
3415 #endif
3416
3417 /* Helper variables for OP_XCLASS opcode (for characters > 255). */
3418
3419 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
3420 BOOL xclass;
3421 pcre_uchar *class_uchardata;
3422 pcre_uchar *class_uchardata_base;
3423 #endif
3424
3425 #ifdef PCRE_DEBUG
3426 if (lengthptr != NULL) DPRINTF((">> start branch\n"));
3427 #endif
3428
3429 /* Set up the default and non-default settings for greediness */
3430
3431 greedy_default = ((options & PCRE_UNGREEDY) != 0);
3432 greedy_non_default = greedy_default ^ 1;
3433
3434 /* Initialize no first byte, no required byte. REQ_UNSET means "no char
3435 matching encountered yet". It gets changed to REQ_NONE if we hit something that
3436 matches a non-fixed char first char; reqchar just remains unset if we never
3437 find one.
3438
3439 When we hit a repeat whose minimum is zero, we may have to adjust these values
3440 to take the zero repeat into account. This is implemented by setting them to
3441 zerofirstbyte and zeroreqchar when such a repeat is encountered. The individual
3442 item types that can be repeated set these backoff variables appropriately. */
3443
3444 firstchar = reqchar = zerofirstchar = zeroreqchar = REQ_UNSET;
3445
3446 /* The variable req_caseopt contains either the REQ_CASELESS value
3447 or zero, according to the current setting of the caseless flag. The
3448 REQ_CASELESS leaves the lower 28 bit empty. It is added into the
3449 firstchar or reqchar variables to record the case status of the
3450 value. This is used only for ASCII characters. */
3451
3452 req_caseopt = ((options & PCRE_CASELESS) != 0)? REQ_CASELESS:0;
3453
3454 /* Switch on next character until the end of the branch */
3455
3456 for (;; ptr++)
3457 {
3458 BOOL negate_class;
3459 BOOL should_flip_negation;
3460 BOOL possessive_quantifier;
3461 BOOL is_quantifier;
3462 BOOL is_recurse;
3463 BOOL reset_bracount;
3464 int class_has_8bitchar;
3465 int class_single_char;
3466 int newoptions;
3467 int recno;
3468 int refsign;
3469 int skipbytes;
3470 int subreqchar;
3471 int subfirstchar;
3472 int terminator;
3473 int mclength;
3474 int tempbracount;
3475 pcre_uchar mcbuffer[8];
3476
3477 /* Get next character in the pattern */
3478
3479 c = *ptr;
3480
3481 /* If we are at the end of a nested substitution, revert to the outer level
3482 string. Nesting only happens one level deep. */
3483
3484 if (c == 0 && nestptr != NULL)
3485 {
3486 ptr = nestptr;
3487 nestptr = NULL;
3488 c = *ptr;
3489 }
3490
3491 /* If we are in the pre-compile phase, accumulate the length used for the
3492 previous cycle of this loop. */
3493
3494 if (lengthptr != NULL)
3495 {
3496 #ifdef PCRE_DEBUG
3497 if (code > cd->hwm) cd->hwm = code; /* High water info */
3498 #endif
3499 if (code > cd->start_workspace + cd->workspace_size -
3500 WORK_SIZE_SAFETY_MARGIN) /* Check for overrun */
3501 {
3502 *errorcodeptr = ERR52;
3503 goto FAILED;
3504 }
3505
3506 /* There is at least one situation where code goes backwards: this is the
3507 case of a zero quantifier after a class (e.g. [ab]{0}). At compile time,
3508 the class is simply eliminated. However, it is created first, so we have to
3509 allow memory for it. Therefore, don't ever reduce the length at this point.
3510 */
3511
3512 if (code < last_code) code = last_code;
3513
3514 /* Paranoid check for integer overflow */
3515
3516 if (OFLOW_MAX - *lengthptr < code - last_code)
3517 {
3518 *errorcodeptr = ERR20;
3519 goto FAILED;
3520 }
3521
3522 *lengthptr += (int)(code - last_code);
3523 DPRINTF(("length=%d added %d c=%c (0x%x)\n", *lengthptr,
3524 (int)(code - last_code), c, c));
3525
3526 /* If "previous" is set and it is not at the start of the work space, move
3527 it back to there, in order to avoid filling up the work space. Otherwise,
3528 if "previous" is NULL, reset the current code pointer to the start. */
3529
3530 if (previous != NULL)
3531 {
3532 if (previous > orig_code)
3533 {
3534 memmove(orig_code, previous, IN_UCHARS(code - previous));
3535 code -= previous - orig_code;
3536 previous = orig_code;
3537 }
3538 }
3539 else code = orig_code;
3540
3541 /* Remember where this code item starts so we can pick up the length
3542 next time round. */
3543
3544 last_code = code;
3545 }
3546
3547 /* In the real compile phase, just check the workspace used by the forward
3548 reference list. */
3549
3550 else if (cd->hwm > cd->start_workspace + cd->workspace_size -
3551 WORK_SIZE_SAFETY_MARGIN)
3552 {
3553 *errorcodeptr = ERR52;
3554 goto FAILED;
3555 }
3556
3557 /* If in \Q...\E, check for the end; if not, we have a literal */
3558
3559 if (inescq && c != 0)
3560 {
3561 if (c == CHAR_BACKSLASH && ptr[1] == CHAR_E)
3562 {
3563 inescq = FALSE;
3564 ptr++;
3565 continue;
3566 }
3567 else
3568 {
3569 if (previous_callout != NULL)
3570 {
3571 if (lengthptr == NULL) /* Don't attempt in pre-compile phase */
3572 complete_callout(previous_callout, ptr, cd);
3573 previous_callout = NULL;
3574 }
3575 if ((options & PCRE_AUTO_CALLOUT) != 0)
3576 {
3577 previous_callout = code;
3578 code = auto_callout(code, ptr, cd);
3579 }
3580 goto NORMAL_CHAR;
3581 }
3582 }
3583
3584 /* Fill in length of a previous callout, except when the next thing is
3585 a quantifier. */
3586
3587 is_quantifier =
3588 c == CHAR_ASTERISK || c == CHAR_PLUS || c == CHAR_QUESTION_MARK ||
3589 (c == CHAR_LEFT_CURLY_BRACKET && is_counted_repeat(ptr+1));
3590
3591 if (!is_quantifier && previous_callout != NULL &&
3592 after_manual_callout-- <= 0)
3593 {
3594 if (lengthptr == NULL) /* Don't attempt in pre-compile phase */
3595 complete_callout(previous_callout, ptr, cd);
3596 previous_callout = NULL;
3597 }
3598
3599 /* In extended mode, skip white space and comments. */
3600
3601 if ((options & PCRE_EXTENDED) != 0)
3602 {
3603 if (MAX_255(*ptr) && (cd->ctypes[c] & ctype_space) != 0) continue;
3604 if (c == CHAR_NUMBER_SIGN)
3605 {
3606 ptr++;
3607 while (*ptr != 0)
3608 {
3609 if (IS_NEWLINE(ptr)) { ptr += cd->nllen - 1; break; }
3610 ptr++;
3611 #ifdef SUPPORT_UTF
3612 if (utf) FORWARDCHAR(ptr);
3613 #endif
3614 }
3615 if (*ptr != 0) continue;
3616
3617 /* Else fall through to handle end of string */
3618 c = 0;
3619 }
3620 }
3621
3622 /* No auto callout for quantifiers. */
3623
3624 if ((options & PCRE_AUTO_CALLOUT) != 0 && !is_quantifier)
3625 {
3626 previous_callout = code;
3627 code = auto_callout(code, ptr, cd);
3628 }
3629
3630 switch(c)
3631 {
3632 /* ===================================================================*/
3633 case 0: /* The branch terminates at string end */
3634 case CHAR_VERTICAL_LINE: /* or | or ) */
3635 case CHAR_RIGHT_PARENTHESIS:
3636 *firstcharptr = firstchar;
3637 *reqcharptr = reqchar;
3638 *codeptr = code;
3639 *ptrptr = ptr;
3640 if (lengthptr != NULL)
3641 {
3642 if (OFLOW_MAX - *lengthptr < code - last_code)
3643 {
3644 *errorcodeptr = ERR20;
3645 goto FAILED;
3646 }
3647 *lengthptr += (int)(code - last_code); /* To include callout length */
3648 DPRINTF((">> end branch\n"));
3649 }
3650 return TRUE;
3651
3652
3653 /* ===================================================================*/
3654 /* Handle single-character metacharacters. In multiline mode, ^ disables
3655 the setting of any following char as a first character. */
3656
3657 case CHAR_CIRCUMFLEX_ACCENT:
3658 previous = NULL;
3659 if ((options & PCRE_MULTILINE) != 0)
3660 {
3661 if (firstchar == REQ_UNSET) firstchar = REQ_NONE;
3662 *code++ = OP_CIRCM;
3663 }
3664 else *code++ = OP_CIRC;
3665 break;
3666
3667 case CHAR_DOLLAR_SIGN:
3668 previous = NULL;
3669 *code++ = ((options & PCRE_MULTILINE) != 0)? OP_DOLLM : OP_DOLL;
3670 break;
3671
3672 /* There can never be a first char if '.' is first, whatever happens about
3673 repeats. The value of reqchar doesn't change either. */
3674
3675 case CHAR_DOT:
3676 if (firstchar == REQ_UNSET) firstchar = REQ_NONE;
3677 zerofirstchar = firstchar;
3678 zeroreqchar = reqchar;
3679 previous = code;
3680 *code++ = ((options & PCRE_DOTALL) != 0)? OP_ALLANY: OP_ANY;
3681 break;
3682
3683
3684 /* ===================================================================*/
3685 /* Character classes. If the included characters are all < 256, we build a
3686 32-byte bitmap of the permitted characters, except in the special case
3687 where there is only one such character. For negated classes, we build the
3688 map as usual, then invert it at the end. However, we use a different opcode
3689 so that data characters > 255 can be handled correctly.
3690
3691 If the class contains characters outside the 0-255 range, a different
3692 opcode is compiled. It may optionally have a bit map for characters < 256,
3693 but those above are are explicitly listed afterwards. A flag byte tells
3694 whether the bitmap is present, and whether this is a negated class or not.
3695
3696 In JavaScript compatibility mode, an isolated ']' causes an error. In
3697 default (Perl) mode, it is treated as a data character. */
3698
3699 case CHAR_RIGHT_SQUARE_BRACKET:
3700 if ((cd->external_options & PCRE_JAVASCRIPT_COMPAT) != 0)
3701 {
3702 *errorcodeptr = ERR64;
3703 goto FAILED;
3704 }
3705 goto NORMAL_CHAR;
3706
3707 case CHAR_LEFT_SQUARE_BRACKET:
3708 previous = code;
3709
3710 /* PCRE supports POSIX class stuff inside a class. Perl gives an error if
3711 they are encountered at the top level, so we'll do that too. */
3712
3713 if ((ptr[1] == CHAR_COLON || ptr[1] == CHAR_DOT ||
3714 ptr[1] == CHAR_EQUALS_SIGN) &&
3715 check_posix_syntax(ptr, &tempptr))
3716 {
3717 *errorcodeptr = (ptr[1] == CHAR_COLON)? ERR13 : ERR31;
3718 goto FAILED;
3719 }
3720
3721 /* If the first character is '^', set the negation flag and skip it. Also,
3722 if the first few characters (either before or after ^) are \Q\E or \E we
3723 skip them too. This makes for compatibility with Perl. */
3724
3725 negate_class = FALSE;
3726 for (;;)
3727 {
3728 c = *(++ptr);
3729 if (c == CHAR_BACKSLASH)
3730 {
3731 if (ptr[1] == CHAR_E)
3732 ptr++;
3733 else if (STRNCMP_UC_C8(ptr + 1, STR_Q STR_BACKSLASH STR_E, 3) == 0)
3734 ptr += 3;
3735 else
3736 break;
3737 }
3738 else if (!negate_class && c == CHAR_CIRCUMFLEX_ACCENT)
3739 negate_class = TRUE;
3740 else break;
3741 }
3742
3743 /* Empty classes are allowed in JavaScript compatibility mode. Otherwise,
3744 an initial ']' is taken as a data character -- the code below handles
3745 that. In JS mode, [] must always fail, so generate OP_FAIL, whereas
3746 [^] must match any character, so generate OP_ALLANY. */
3747
3748 if (c == CHAR_RIGHT_SQUARE_BRACKET &&
3749 (cd->external_options & PCRE_JAVASCRIPT_COMPAT) != 0)
3750 {
3751 *code++ = negate_class? OP_ALLANY : OP_FAIL;
3752 if (firstchar == REQ_UNSET) firstchar = REQ_NONE;
3753 zerofirstchar = firstchar;
3754 break;
3755 }
3756
3757 /* If a class contains a negative special such as \S, we need to flip the
3758 negation flag at the end, so that support for characters > 255 works
3759 correctly (they are all included in the class). */
3760
3761 should_flip_negation = FALSE;
3762
3763 /* For optimization purposes, we track some properties of the class.
3764 class_has_8bitchar will be non-zero, if the class contains at least one
3765 < 256 character. class_single_char will be 1 if the class contains only
3766 a single character. */
3767
3768 class_has_8bitchar = 0;
3769 class_single_char = 0;
3770
3771 /* Initialize the 32-char bit map to all zeros. We build the map in a
3772 temporary bit of memory, in case the class contains only 1 character (less
3773 than 256), because in that case the compiled code doesn't use the bit map.
3774 */
3775
3776 memset(classbits, 0, 32 * sizeof(pcre_uint8));
3777
3778 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
3779 xclass = FALSE; /* No chars >= 256 */
3780 class_uchardata = code + LINK_SIZE + 2; /* For UTF-8 items */
3781 class_uchardata_base = class_uchardata; /* For resetting in pass 1 */
3782 #endif
3783
3784 /* Process characters until ] is reached. By writing this as a "do" it
3785 means that an initial ] is taken as a data character. At the start of the
3786 loop, c contains the first byte of the character. */
3787
3788 if (c != 0) do
3789 {
3790 const pcre_uchar *oldptr;
3791
3792 #ifdef SUPPORT_UTF
3793 if (utf && HAS_EXTRALEN(c))
3794 { /* Braces are required because the */
3795 GETCHARLEN(c, ptr, ptr); /* macro generates multiple statements */
3796 }
3797 #endif
3798
3799 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
3800 /* In the pre-compile phase, accumulate the length of any extra
3801 data and reset the pointer. This is so that very large classes that
3802 contain a zillion > 255 characters no longer overwrite the work space
3803 (which is on the stack). */
3804
3805 if (lengthptr != NULL)
3806 {
3807 *lengthptr += class_uchardata - class_uchardata_base;
3808 class_uchardata = class_uchardata_base;
3809 }
3810 #endif
3811
3812 /* Inside \Q...\E everything is literal except \E */
3813
3814 if (inescq)
3815 {
3816 if (c == CHAR_BACKSLASH && ptr[1] == CHAR_E) /* If we are at \E */
3817 {
3818 inescq = FALSE; /* Reset literal state */
3819 ptr++; /* Skip the 'E' */
3820 continue; /* Carry on with next */
3821 }
3822 goto CHECK_RANGE; /* Could be range if \E follows */
3823 }
3824
3825 /* Handle POSIX class names. Perl allows a negation extension of the
3826 form [:^name:]. A square bracket that doesn't match the syntax is
3827 treated as a literal. We also recognize the POSIX constructions
3828 [.ch.] and [=ch=] ("collating elements") and fault them, as Perl
3829 5.6 and 5.8 do. */
3830
3831 if (c == CHAR_LEFT_SQUARE_BRACKET &&
3832 (ptr[1] == CHAR_COLON || ptr[1] == CHAR_DOT ||
3833 ptr[1] == CHAR_EQUALS_SIGN) && check_posix_syntax(ptr, &tempptr))
3834 {
3835 BOOL local_negate = FALSE;
3836 int posix_class, taboffset, tabopt;
3837 register const pcre_uint8 *cbits = cd->cbits;
3838 pcre_uint8 pbits[32];
3839
3840 if (ptr[1] != CHAR_COLON)
3841 {
3842 *errorcodeptr = ERR31;
3843 goto FAILED;
3844 }
3845
3846 ptr += 2;
3847 if (*ptr == CHAR_CIRCUMFLEX_ACCENT)
3848 {
3849 local_negate = TRUE;
3850 should_flip_negation = TRUE; /* Note negative special */
3851 ptr++;
3852 }
3853
3854 posix_class = check_posix_name(ptr, (int)(tempptr - ptr));
3855 if (posix_class < 0)
3856 {
3857 *errorcodeptr = ERR30;
3858 goto FAILED;
3859 }
3860
3861 /* If matching is caseless, upper and lower are converted to
3862 alpha. This relies on the fact that the class table starts with
3863 alpha, lower, upper as the first 3 entries. */
3864
3865 if ((options & PCRE_CASELESS) != 0 && posix_class <= 2)
3866 posix_class = 0;
3867
3868 /* When PCRE_UCP is set, some of the POSIX classes are converted to
3869 different escape sequences that use Unicode properties. */
3870
3871 #ifdef SUPPORT_UCP
3872 if ((options & PCRE_UCP) != 0)
3873 {
3874 int pc = posix_class + ((local_negate)? POSIX_SUBSIZE/2 : 0);
3875 if (posix_substitutes[pc] != NULL)
3876 {
3877 nestptr = tempptr + 1;
3878 ptr = posix_substitutes[pc] - 1;
3879 continue;
3880 }
3881 }
3882 #endif
3883 /* In the non-UCP case, we build the bit map for the POSIX class in a
3884 chunk of local store because we may be adding and subtracting from it,
3885 and we don't want to subtract bits that may be in the main map already.
3886 At the end we or the result into the bit map that is being built. */
3887
3888 posix_class *= 3;
3889
3890 /* Copy in the first table (always present) */
3891
3892 memcpy(pbits, cbits + posix_class_maps[posix_class],
3893 32 * sizeof(pcre_uint8));
3894
3895 /* If there is a second table, add or remove it as required. */
3896
3897 taboffset = posix_class_maps[posix_class + 1];
3898 tabopt = posix_class_maps[posix_class + 2];
3899
3900 if (taboffset >= 0)
3901 {
3902 if (tabopt >= 0)
3903 for (c = 0; c < 32; c++) pbits[c] |= cbits[c + taboffset];
3904 else
3905 for (c = 0; c < 32; c++) pbits[c] &= ~cbits[c + taboffset];
3906 }
3907
3908 /* Not see if we need to remove any special characters. An option
3909 value of 1 removes vertical space and 2 removes underscore. */
3910
3911 if (tabopt < 0) tabopt = -tabopt;
3912 if (tabopt == 1) pbits[1] &= ~0x3c;
3913 else if (tabopt == 2) pbits[11] &= 0x7f;
3914
3915 /* Add the POSIX table or its complement into the main table that is
3916 being built and we are done. */
3917
3918 if (local_negate)
3919 for (c = 0; c < 32; c++) classbits[c] |= ~pbits[c];
3920 else
3921 for (c = 0; c < 32; c++) classbits[c] |= pbits[c];
3922
3923 ptr = tempptr + 1;
3924 /* Every class contains at least one < 256 characters. */
3925 class_has_8bitchar = 1;
3926 /* Every class contains at least two characters. */
3927 class_single_char = 2;
3928 continue; /* End of POSIX syntax handling */
3929 }
3930
3931 /* Backslash may introduce a single character, or it may introduce one
3932 of the specials, which just set a flag. The sequence \b is a special
3933 case. Inside a class (and only there) it is treated as backspace. We
3934 assume that other escapes have more than one character in them, so
3935 speculatively set both class_has_8bitchar and class_single_char bigger
3936 than one. Unrecognized escapes fall through and are either treated
3937 as literal characters (by default), or are faulted if
3938 PCRE_EXTRA is set. */
3939
3940 if (c == CHAR_BACKSLASH)
3941 {
3942 c = check_escape(&ptr, errorcodeptr, cd->bracount, options, TRUE);
3943 if (*errorcodeptr != 0) goto FAILED;
3944
3945 if (-c == ESC_b) c = CHAR_BS; /* \b is backspace in a class */
3946 else if (-c == ESC_N) /* \N is not supported in a class */
3947 {
3948 *errorcodeptr = ERR71;
3949 goto FAILED;
3950 }
3951 else if (-c == ESC_Q) /* Handle start of quoted string */
3952 {
3953 if (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
3954 {
3955 ptr += 2; /* avoid empty string */
3956 }
3957 else inescq = TRUE;
3958 continue;
3959 }
3960 else if (-c == ESC_E) continue; /* Ignore orphan \E */
3961
3962 if (c < 0)
3963 {
3964 register const pcre_uint8 *cbits = cd->cbits;
3965 /* Every class contains at least two < 256 characters. */
3966 class_has_8bitchar++;
3967 /* Every class contains at least two characters. */
3968 class_single_char += 2;
3969
3970 switch (-c)
3971 {
3972 #ifdef SUPPORT_UCP
3973 case ESC_du: /* These are the values given for \d etc */
3974 case ESC_DU: /* when PCRE_UCP is set. We replace the */
3975 case ESC_wu: /* escape sequence with an appropriate \p */
3976 case ESC_WU: /* or \P to test Unicode properties instead */
3977 case ESC_su: /* of the default ASCII testing. */
3978 case ESC_SU:
3979 nestptr = ptr;
3980 ptr = substitutes[-c - ESC_DU] - 1; /* Just before substitute */
3981 class_has_8bitchar--; /* Undo! */
3982 continue;
3983 #endif
3984 case ESC_d:
3985 for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_digit];
3986 continue;
3987
3988 case ESC_D:
3989 should_flip_negation = TRUE;
3990 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_digit];
3991 continue;
3992
3993 case ESC_w:
3994 for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_word];
3995 continue;
3996
3997 case ESC_W:
3998 should_flip_negation = TRUE;
3999 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_word];
4000 continue;
4001
4002 /* Perl 5.004 onwards omits VT from \s, but we must preserve it
4003 if it was previously set by something earlier in the character
4004 class. */
4005
4006 case ESC_s:
4007 classbits[0] |= cbits[cbit_space];
4008 classbits[1] |= cbits[cbit_space+1] & ~0x08;
4009 for (c = 2; c < 32; c++) classbits[c] |= cbits[c+cbit_space];
4010 continue;
4011
4012 case ESC_S:
4013 should_flip_negation = TRUE;
4014 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_space];
4015 classbits[1] |= 0x08; /* Perl 5.004 onwards omits VT from \s */
4016 continue;
4017
4018 case ESC_h:
4019 SETBIT(classbits, 0x09); /* VT */
4020 SETBIT(classbits, 0x20); /* SPACE */
4021 SETBIT(classbits, 0xa0); /* NSBP */
4022 #ifdef SUPPORT_UTF
4023 if (utf)
4024 {
4025 xclass = TRUE;
4026 *class_uchardata++ = XCL_SINGLE;
4027 class_uchardata += PRIV(ord2utf)(0x1680, class_uchardata);
4028 *class_uchardata++ = XCL_SINGLE;
4029 class_uchardata += PRIV(ord2utf)(0x180e, class_uchardata);
4030 *class_uchardata++ = XCL_RANGE;
4031 class_uchardata += PRIV(ord2utf)(0x2000, class_uchardata);
4032 class_uchardata += PRIV(ord2utf)(0x200A, class_uchardata);
4033 *class_uchardata++ = XCL_SINGLE;
4034 class_uchardata += PRIV(ord2utf)(0x202f, class_uchardata);
4035 *class_uchardata++ = XCL_SINGLE;
4036 class_uchardata += PRIV(ord2utf)(0x205f, class_uchardata);
4037 *class_uchardata++ = XCL_SINGLE;
4038 class_uchardata += PRIV(ord2utf)(0x3000, class_uchardata);
4039 }
4040 #endif
4041 continue;
4042
4043 case ESC_H:
4044 for (c = 0; c < 32; c++)
4045 {
4046 int x = 0xff;
4047 switch (c)
4048 {
4049 case 0x09/8: x ^= 1 << (0x09%8); break;
4050 case 0x20/8: x ^= 1 << (0x20%8); break;
4051 case 0xa0/8: x ^= 1 << (0xa0%8); break;
4052 default: break;
4053 }
4054 classbits[c] |= x;
4055 }
4056
4057 #ifdef SUPPORT_UTF
4058 if (utf)
4059 {
4060 xclass = TRUE;
4061 *class_uchardata++ = XCL_RANGE;
4062 class_uchardata += PRIV(ord2utf)(0x0100, class_uchardata);
4063 class_uchardata += PRIV(ord2utf)(0x167f, class_uchardata);
4064 *class_uchardata++ = XCL_RANGE;
4065 class_uchardata += PRIV(ord2utf)(0x1681, class_uchardata);
4066 class_uchardata += PRIV(ord2utf)(0x180d, class_uchardata);
4067 *class_uchardata++ = XCL_RANGE;
4068 class_uchardata += PRIV(ord2utf)(0x180f, class_uchardata);
4069 class_uchardata += PRIV(ord2utf)(0x1fff, class_uchardata);
4070 *class_uchardata++ = XCL_RANGE;
4071 class_uchardata += PRIV(ord2utf)(0x200B, class_uchardata);
4072 class_uchardata += PRIV(ord2utf)(0x202e, class_uchardata);
4073 *class_uchardata++ = XCL_RANGE;
4074 class_uchardata += PRIV(ord2utf)(0x2030, class_uchardata);
4075 class_uchardata += PRIV(ord2utf)(0x205e, class_uchardata);
4076 *class_uchardata++ = XCL_RANGE;
4077 class_uchardata += PRIV(ord2utf)(0x2060, class_uchardata);
4078 class_uchardata += PRIV(ord2utf)(0x2fff, class_uchardata);
4079 *class_uchardata++ = XCL_RANGE;
4080 class_uchardata += PRIV(ord2utf)(0x3001, class_uchardata);
4081 class_uchardata += PRIV(ord2utf)(0x10ffff, class_uchardata);
4082 }
4083 #endif
4084 continue;
4085
4086 case ESC_v:
4087 SETBIT(classbits, 0x0a); /* LF */
4088 SETBIT(classbits, 0x0b); /* VT */
4089 SETBIT(classbits, 0x0c); /* FF */
4090 SETBIT(classbits, 0x0d); /* CR */
4091 SETBIT(classbits, 0x85); /* NEL */
4092 #ifdef SUPPORT_UTF
4093 if (utf)
4094 {
4095 xclass = TRUE;
4096 *class_uchardata++ = XCL_RANGE;
4097 class_uchardata += PRIV(ord2utf)(0x2028, class_uchardata);
4098 class_uchardata += PRIV(ord2utf)(0x2029, class_uchardata);
4099 }
4100 #endif
4101 continue;
4102
4103 case ESC_V:
4104 for (c = 0; c < 32; c++)
4105 {
4106 int x = 0xff;
4107 switch (c)
4108 {
4109 case 0x0a/8: x ^= 1 << (0x0a%8);
4110 x ^= 1 << (0x0b%8);
4111 x ^= 1 << (0x0c%8);
4112 x ^= 1 << (0x0d%8);
4113 break;
4114 case 0x85/8: x ^= 1 << (0x85%8); break;
4115 default: break;
4116 }
4117 classbits[c] |= x;
4118 }
4119
4120 #ifdef SUPPORT_UTF
4121 if (utf)
4122 {
4123 xclass = TRUE;
4124 *class_uchardata++ = XCL_RANGE;
4125 class_uchardata += PRIV(ord2utf)(0x0100, class_uchardata);
4126 class_uchardata += PRIV(ord2utf)(0x2027, class_uchardata);
4127 *class_uchardata++ = XCL_RANGE;
4128 class_uchardata += PRIV(ord2utf)(0x2029, class_uchardata);
4129 class_uchardata += PRIV(ord2utf)(0x10ffff, class_uchardata);
4130 }
4131 #endif
4132 continue;
4133
4134 #ifdef SUPPORT_UCP
4135 case ESC_p:
4136 case ESC_P:
4137 {
4138 BOOL negated;
4139 int pdata;
4140 int ptype = get_ucp(&ptr, &negated, &pdata, errorcodeptr);
4141 if (ptype < 0) goto FAILED;
4142 xclass = TRUE;
4143 *class_uchardata++ = ((-c == ESC_p) != negated)?
4144 XCL_PROP : XCL_NOTPROP;
4145 *class_uchardata++ = ptype;
4146 *class_uchardata++ = pdata;
4147 class_has_8bitchar--; /* Undo! */
4148 continue;
4149 }
4150 #endif
4151 /* Unrecognized escapes are faulted if PCRE is running in its
4152 strict mode. By default, for compatibility with Perl, they are
4153 treated as literals. */
4154
4155 default:
4156 if ((options & PCRE_EXTRA) != 0)
4157 {
4158 *errorcodeptr = ERR7;
4159 goto FAILED;
4160 }
4161 class_has_8bitchar--; /* Undo the speculative increase. */
4162 class_single_char -= 2; /* Undo the speculative increase. */
4163 c = *ptr; /* Get the final character and fall through */
4164 break;
4165 }
4166 }
4167
4168 /* Fall through if we have a single character (c >= 0). This may be
4169 greater than 256. */
4170
4171 } /* End of backslash handling */
4172
4173 /* A single character may be followed by '-' to form a range. However,
4174 Perl does not permit ']' to be the end of the range. A '-' character
4175 at the end is treated as a literal. Perl ignores orphaned \E sequences
4176 entirely. The code for handling \Q and \E is messy. */
4177
4178 CHECK_RANGE:
4179 while (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
4180 {
4181 inescq = FALSE;
4182 ptr += 2;
4183 }
4184
4185 oldptr = ptr;
4186
4187 /* Remember \r or \n */
4188
4189 if (c == CHAR_CR || c == CHAR_NL) cd->external_flags |= PCRE_HASCRORLF;
4190
4191 /* Check for range */
4192
4193 if (!inescq && ptr[1] == CHAR_MINUS)
4194 {
4195 int d;
4196 ptr += 2;
4197 while (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_E) ptr += 2;
4198
4199 /* If we hit \Q (not followed by \E) at this point, go into escaped
4200 mode. */
4201
4202 while (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_Q)
4203 {
4204 ptr += 2;
4205 if (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_E)
4206 { ptr += 2; continue; }
4207 inescq = TRUE;
4208 break;
4209 }
4210
4211 if (*ptr == 0 || (!inescq && *ptr == CHAR_RIGHT_SQUARE_BRACKET))
4212 {
4213 ptr = oldptr;
4214 goto LONE_SINGLE_CHARACTER;
4215 }
4216
4217 #ifdef SUPPORT_UTF
4218 if (utf)
4219 { /* Braces are required because the */
4220 GETCHARLEN(d, ptr, ptr); /* macro generates multiple statements */
4221 }
4222 else
4223 #endif
4224 d = *ptr; /* Not UTF-8 mode */
4225
4226 /* The second part of a range can be a single-character escape, but
4227 not any of the other escapes. Perl 5.6 treats a hyphen as a literal
4228 in such circumstances. */
4229
4230 if (!inescq && d == CHAR_BACKSLASH)
4231 {
4232 d = check_escape(&ptr, errorcodeptr, cd->bracount, options, TRUE);
4233 if (*errorcodeptr != 0) goto FAILED;
4234
4235 /* \b is backspace; any other special means the '-' was literal */
4236
4237 if (d < 0)
4238 {
4239 if (d == -ESC_b) d = CHAR_BS; else
4240 {
4241 ptr = oldptr;
4242 goto LONE_SINGLE_CHARACTER; /* A few lines below */
4243 }
4244 }
4245 }
4246
4247 /* Check that the two values are in the correct order. Optimize
4248 one-character ranges */
4249
4250 if (d < c)
4251 {
4252 *errorcodeptr = ERR8;
4253 goto FAILED;
4254 }
4255
4256 if (d == c) goto LONE_SINGLE_CHARACTER; /* A few lines below */
4257
4258 /* Remember \r or \n */
4259
4260 if (d == CHAR_CR || d == CHAR_NL) cd->external_flags |= PCRE_HASCRORLF;
4261
4262 /* Since we found a character range, single character optimizations
4263 cannot be done anymore. */
4264 class_single_char = 2;
4265
4266 /* In UTF-8 mode, if the upper limit is > 255, or > 127 for caseless
4267 matching, we have to use an XCLASS with extra data items. Caseless
4268 matching for characters > 127 is available only if UCP support is
4269 available. */
4270
4271 #if defined SUPPORT_UTF && !(defined COMPILE_PCRE8)
4272 if ((d > 255) || (utf && ((options & PCRE_CASELESS) != 0 && d > 127)))
4273 #elif defined SUPPORT_UTF
4274 if (utf && (d > 255 || ((options & PCRE_CASELESS) != 0 && d > 127)))
4275 #elif !(defined COMPILE_PCRE8)
4276 if (d > 255)
4277 #endif
4278 #if defined SUPPORT_UTF || !(defined COMPILE_PCRE8)
4279 {
4280 xclass = TRUE;
4281
4282 /* With UCP support, we can find the other case equivalents of
4283 the relevant characters. There may be several ranges. Optimize how
4284 they fit with the basic range. */
4285
4286 #ifdef SUPPORT_UCP
4287 #ifndef COMPILE_PCRE8
4288 if (utf && (options & PCRE_CASELESS) != 0)
4289 #else
4290 if ((options & PCRE_CASELESS) != 0)
4291 #endif
4292 {
4293 unsigned int occ, ocd;
4294 unsigned int cc = c;
4295 unsigned int origd = d;
4296 while (get_othercase_range(&cc, origd, &occ, &ocd))
4297 {
4298 if (occ >= (unsigned int)c &&
4299 ocd <= (unsigned int)d)
4300 continue; /* Skip embedded ranges */
4301
4302 if (occ < (unsigned int)c &&
4303 ocd >= (unsigned int)c - 1) /* Extend the basic range */
4304 { /* if there is overlap, */
4305 c = occ; /* noting that if occ < c */
4306 continue; /* we can't have ocd > d */
4307 } /* because a subrange is */
4308 if (ocd > (unsigned int)d &&
4309 occ <= (unsigned int)d + 1) /* always shorter than */
4310 { /* the basic range. */
4311 d = ocd;
4312 continue;
4313 }
4314
4315 if (occ == ocd)
4316 {
4317 *class_uchardata++ = XCL_SINGLE;
4318 }
4319 else
4320 {
4321 *class_uchardata++ = XCL_RANGE;
4322 class_uchardata += PRIV(ord2utf)(occ, class_uchardata);
4323 }
4324 class_uchardata += PRIV(ord2utf)(ocd, class_uchardata);
4325 }
4326 }
4327 #endif /* SUPPORT_UCP */
4328
4329 /* Now record the original range, possibly modified for UCP caseless
4330 overlapping ranges. */
4331
4332 *class_uchardata++ = XCL_RANGE;
4333 #ifdef SUPPORT_UTF
4334 #ifndef COMPILE_PCRE8
4335 if (utf)
4336 {
4337 class_uchardata += PRIV(ord2utf)(c, class_uchardata);
4338 class_uchardata += PRIV(ord2utf)(d, class_uchardata);
4339 }
4340 else
4341 {
4342 *class_uchardata++ = c;
4343 *class_uchardata++ = d;
4344 }
4345 #else
4346 class_uchardata += PRIV(ord2utf)(c, class_uchardata);
4347 class_uchardata += PRIV(ord2utf)(d, class_uchardata);
4348 #endif
4349 #else /* SUPPORT_UTF */
4350 *class_uchardata++ = c;
4351 *class_uchardata++ = d;
4352 #endif /* SUPPORT_UTF */
4353
4354 /* With UCP support, we are done. Without UCP support, there is no
4355 caseless matching for UTF characters > 127; we can use the bit map
4356 for the smaller ones. As for 16 bit characters without UTF, we
4357 can still use */
4358
4359 #ifdef SUPPORT_UCP
4360 #ifndef COMPILE_PCRE8
4361 if (utf)
4362 #endif
4363 continue; /* With next character in the class */
4364 #endif /* SUPPORT_UCP */
4365
4366 #if defined SUPPORT_UTF && !defined(SUPPORT_UCP) && !(defined COMPILE_PCRE8)
4367 if (utf)
4368 {
4369 if ((options & PCRE_CASELESS) == 0 || c > 127) continue;
4370 /* Adjust upper limit and fall through to set up the map */
4371 d = 127;
4372 }
4373 else
4374 {
4375 if (c > 255) continue;
4376 /* Adjust upper limit and fall through to set up the map */
4377 d = 255;
4378 }
4379 #elif defined SUPPORT_UTF && !defined(SUPPORT_UCP)
4380 if ((options & PCRE_CASELESS) == 0 || c > 127) continue;
4381 /* Adjust upper limit and fall through to set up the map */
4382 d = 127;
4383 #else
4384 if (c > 255) continue;
4385 /* Adjust upper limit and fall through to set up the map */
4386 d = 255;
4387 #endif /* SUPPORT_UTF && !SUPPORT_UCP && !COMPILE_PCRE8 */
4388 }
4389 #endif /* SUPPORT_UTF || !COMPILE_PCRE8 */
4390
4391 /* We use the bit map for 8 bit mode, or when the characters fall
4392 partially or entirely to [0-255] ([0-127] for UCP) ranges. */
4393
4394 class_has_8bitchar = 1;
4395
4396 /* We can save a bit of time by skipping this in the pre-compile. */
4397
4398 if (lengthptr == NULL) for (; c <= d; c++)
4399 {
4400 classbits[c/8] |= (1 << (c&7));
4401 if ((options & PCRE_CASELESS) != 0)
4402 {
4403 int uc = cd->fcc[c]; /* flip case */
4404 classbits[uc/8] |= (1 << (uc&7));
4405 }
4406 }
4407
4408 continue; /* Go get the next char in the class */
4409 }
4410
4411 /* Handle a lone single character - we can get here for a normal
4412 non-escape char, or after \ that introduces a single character or for an
4413 apparent range that isn't. */
4414
4415 LONE_SINGLE_CHARACTER:
4416
4417 /* Only the value of 1 matters for class_single_char. */
4418 if (class_single_char < 2) class_single_char++;
4419
4420 /* If class_charcount is 1, we saw precisely one character. As long as
4421 there were no negated characters >= 128 and there was no use of \p or \P,
4422 in other words, no use of any XCLASS features, we can optimize.
4423
4424 In UTF-8 mode, we can optimize the negative case only if there were no
4425 characters >= 128 because OP_NOT and the related opcodes like OP_NOTSTAR
4426 operate on single-bytes characters only. This is an historical hangover.
4427 Maybe one day we can tidy these opcodes to handle multi-byte characters.
4428
4429 The optimization throws away the bit map. We turn the item into a
4430 1-character OP_CHAR[I] if it's positive, or OP_NOT[I] if it's negative.
4431 Note that OP_NOT[I] does not support multibyte characters. In the positive
4432 case, it can cause firstchar to be set. Otherwise, there can be no first
4433 char if this item is first, whatever repeat count may follow. In the case
4434 of reqchar, save the previous value for reinstating. */
4435
4436 #ifdef SUPPORT_UTF
4437 if (class_single_char == 1 && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET
4438 && (!utf || !negate_class || c < (MAX_VALUE_FOR_SINGLE_CHAR + 1)))
4439 #else
4440 if (class_single_char == 1 && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET)
4441 #endif
4442 {
4443 ptr++;
4444 zeroreqchar = reqchar;
4445
4446 /* The OP_NOT[I] opcodes work on single characters only. */
4447
4448 if (negate_class)
4449 {
4450 if (firstchar == REQ_UNSET) firstchar = REQ_NONE;
4451 zerofirstchar = firstchar;
4452 *code++ = ((options & PCRE_CASELESS) != 0)? OP_NOTI: OP_NOT;
4453 *code++ = c;
4454 goto NOT_CHAR;
4455 }
4456
4457 /* For a single, positive character, get the value into mcbuffer, and
4458 then we can handle this with the normal one-character code. */
4459
4460 #ifdef SUPPORT_UTF
4461 if (utf && c > MAX_VALUE_FOR_SINGLE_CHAR)
4462 mclength = PRIV(ord2utf)(c, mcbuffer);
4463 else
4464 #endif
4465 {
4466 mcbuffer[0] = c;
4467 mclength = 1;
4468 }
4469 goto ONE_CHAR;
4470 } /* End of 1-char optimization */
4471
4472 /* Handle a character that cannot go in the bit map. */
4473
4474 #if defined SUPPORT_UTF && !(defined COMPILE_PCRE8)
4475 if ((c > 255) || (utf && ((options & PCRE_CASELESS) != 0 && c > 127)))
4476 #elif defined SUPPORT_UTF
4477 if (utf && (c > 255 || ((options & PCRE_CASELESS) != 0 && c > 127)))
4478 #elif !(defined COMPILE_PCRE8)
4479 if (c > 255)
4480 #endif
4481
4482 #if defined SUPPORT_UTF || !(defined COMPILE_PCRE8)
4483 {
4484 xclass = TRUE;
4485 *class_uchardata++ = XCL_SINGLE;
4486 #ifdef SUPPORT_UTF
4487 #ifndef COMPILE_PCRE8
4488 /* In non 8 bit mode, we can get here even if we are not in UTF mode. */
4489 if (!utf)
4490 *class_uchardata++ = c;
4491 else
4492 #endif
4493 class_uchardata += PRIV(ord2utf)(c, class_uchardata);
4494 #else /* SUPPORT_UTF */
4495 *class_uchardata++ = c;
4496 #endif /* SUPPORT_UTF */
4497
4498 #ifdef SUPPORT_UCP
4499 #ifdef COMPILE_PCRE8
4500 if ((options & PCRE_CASELESS) != 0)
4501 #else
4502 /* In non 8 bit mode, we can get here even if we are not in UTF mode. */
4503 if (utf && (options & PCRE_CASELESS) != 0)
4504 #endif
4505 {
4506 unsigned int othercase;
4507 if ((othercase = UCD_OTHERCASE(c)) != c)
4508 {
4509 *class_uchardata++ = XCL_SINGLE;
4510 class_uchardata += PRIV(ord2utf)(othercase, class_uchardata);
4511 }
4512 }
4513 #endif /* SUPPORT_UCP */
4514
4515 }
4516 else
4517 #endif /* SUPPORT_UTF || COMPILE_PCRE16 */
4518
4519 /* Handle a single-byte character */
4520 {
4521 class_has_8bitchar = 1;
4522 classbits[c/8] |= (1 << (c&7));
4523 if ((options & PCRE_CASELESS) != 0)
4524 {
4525 c = cd->fcc[c]; /* flip case */
4526 classbits[c/8] |= (1 << (c&7));
4527 }
4528 }
4529 }
4530
4531 /* Loop until ']' reached. This "while" is the end of the "do" far above.
4532 If we are at the end of an internal nested string, revert to the outer
4533 string. */
4534
4535 while (((c = *(++ptr)) != 0 ||
4536 (nestptr != NULL &&
4537 (ptr = nestptr, nestptr = NULL, c = *(++ptr)) != 0)) &&
4538 (c != CHAR_RIGHT_SQUARE_BRACKET || inescq));
4539
4540 /* Check for missing terminating ']' */
4541
4542 if (c == 0)
4543 {
4544 *errorcodeptr = ERR6;
4545 goto FAILED;
4546 }
4547
4548 /* If this is the first thing in the branch, there can be no first char
4549 setting, whatever the repeat count. Any reqchar setting must remain
4550 unchanged after any kind of repeat. */
4551
4552 if (firstchar == REQ_UNSET) firstchar = REQ_NONE;
4553 zerofirstchar = firstchar;
4554 zeroreqchar = reqchar;
4555
4556 /* If there are characters with values > 255, we have to compile an
4557 extended class, with its own opcode, unless there was a negated special
4558 such as \S in the class, and PCRE_UCP is not set, because in that case all
4559 characters > 255 are in the class, so any that were explicitly given as
4560 well can be ignored. If (when there are explicit characters > 255 that must
4561 be listed) there are no characters < 256, we can omit the bitmap in the
4562 actual compiled code. */
4563
4564 #ifdef SUPPORT_UTF
4565 if (xclass && (!should_flip_negation || (options & PCRE_UCP) != 0))
4566 #endif
4567 #ifndef COMPILE_PCRE8
4568 if (xclass && !should_flip_negation)
4569 #endif
4570 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
4571 {
4572 *class_uchardata++ = XCL_END; /* Marks the end of extra data */
4573 *code++ = OP_XCLASS;
4574 code += LINK_SIZE;
4575 *code = negate_class? XCL_NOT:0;
4576
4577 /* If the map is required, move up the extra data to make room for it;
4578 otherwise just move the code pointer to the end of the extra data. */
4579
4580 if (class_has_8bitchar > 0)
4581 {
4582 *code++ |= XCL_MAP;
4583 memmove(code + (32 / sizeof(pcre_uchar)), code,
4584 IN_UCHARS(class_uchardata - code));
4585 memcpy(code, classbits, 32);
4586 code = class_uchardata + (32 / sizeof(pcre_uchar));
4587 }
4588 else code = class_uchardata;
4589
4590 /* Now fill in the complete length of the item */
4591
4592 PUT(previous, 1, (int)(code - previous));
4593 break; /* End of class handling */
4594 }
4595 #endif
4596
4597 /* If there are no characters > 255, or they are all to be included or
4598 excluded, set the opcode to OP_CLASS or OP_NCLASS, depending on whether the
4599 whole class was negated and whether there were negative specials such as \S
4600 (non-UCP) in the class. Then copy the 32-byte map into the code vector,
4601 negating it if necessary. */
4602
4603 *code++ = (negate_class == should_flip_negation) ? OP_CLASS : OP_NCLASS;
4604 if (lengthptr == NULL) /* Save time in the pre-compile phase */
4605 {
4606 if (negate_class)
4607 for (c = 0; c < 32; c++) classbits[c] = ~classbits[c];
4608 memcpy(code, classbits, 32);
4609 }
4610 code += 32 / sizeof(pcre_uchar);
4611 NOT_CHAR:
4612 break;
4613
4614
4615 /* ===================================================================*/
4616 /* Various kinds of repeat; '{' is not necessarily a quantifier, but this
4617 has been tested above. */
4618
4619 case CHAR_LEFT_CURLY_BRACKET:
4620 if (!is_quantifier) goto NORMAL_CHAR;
4621 ptr = read_repeat_counts(ptr+1, &repeat_min, &repeat_max, errorcodeptr);
4622 if (*errorcodeptr != 0) goto FAILED;
4623 goto REPEAT;
4624
4625 case CHAR_ASTERISK:
4626 repeat_min = 0;
4627 repeat_max = -1;
4628 goto REPEAT;
4629
4630 case CHAR_PLUS:
4631 repeat_min = 1;
4632 repeat_max = -1;
4633 goto REPEAT;
4634
4635 case CHAR_QUESTION_MARK:
4636 repeat_min = 0;
4637 repeat_max = 1;
4638
4639 REPEAT:
4640 if (previous == NULL)
4641 {
4642 *errorcodeptr = ERR9;
4643 goto FAILED;
4644 }
4645
4646 if (repeat_min == 0)
4647 {
4648 firstchar = zerofirstchar; /* Adjust for zero repeat */
4649 reqchar = zeroreqchar; /* Ditto */
4650 }
4651
4652 /* Remember whether this is a variable length repeat */
4653
4654 reqvary = (repeat_min == repeat_max)? 0 : REQ_VARY;
4655
4656 op_type = 0; /* Default single-char op codes */
4657 possessive_quantifier = FALSE; /* Default not possessive quantifier */
4658
4659 /* Save start of previous item, in case we have to move it up in order to
4660 insert something before it. */
4661
4662 tempcode = previous;
4663
4664 /* If the next character is '+', we have a possessive quantifier. This
4665 implies greediness, whatever the setting of the PCRE_UNGREEDY option.
4666 If the next character is '?' this is a minimizing repeat, by default,
4667 but if PCRE_UNGREEDY is set, it works the other way round. We change the
4668 repeat type to the non-default. */
4669
4670 if (ptr[1] == CHAR_PLUS)
4671 {
4672 repeat_type = 0; /* Force greedy */
4673 possessive_quantifier = TRUE;
4674 ptr++;
4675 }
4676 else if (ptr[1] == CHAR_QUESTION_MARK)
4677 {
4678 repeat_type = greedy_non_default;
4679 ptr++;
4680 }
4681 else repeat_type = greedy_default;
4682
4683 /* If previous was a recursion call, wrap it in atomic brackets so that
4684 previous becomes the atomic group. All recursions were so wrapped in the
4685 past, but it no longer happens for non-repeated recursions. In fact, the
4686 repeated ones could be re-implemented independently so as not to need this,
4687 but for the moment we rely on the code for repeating groups. */
4688
4689 if (*previous == OP_RECURSE)
4690 {
4691 memmove(previous + 1 + LINK_SIZE, previous, IN_UCHARS(1 + LINK_SIZE));
4692 *previous = OP_ONCE;
4693 PUT(previous, 1, 2 + 2*LINK_SIZE);
4694 previous[2 + 2*LINK_SIZE] = OP_KET;
4695 PUT(previous, 3 + 2*LINK_SIZE, 2 + 2*LINK_SIZE);
4696 code += 2 + 2 * LINK_SIZE;
4697 length_prevgroup = 3 + 3*LINK_SIZE;
4698
4699 /* When actually compiling, we need to check whether this was a forward
4700 reference, and if so, adjust the offset. */
4701
4702 if (lengthptr == NULL && cd->hwm >= cd->start_workspace + LINK_SIZE)
4703 {
4704 int offset = GET(cd->hwm, -LINK_SIZE);
4705 if (offset == previous + 1 - cd->start_code)
4706 PUT(cd->hwm, -LINK_SIZE, offset + 1 + LINK_SIZE);
4707 }
4708 }
4709
4710 /* Now handle repetition for the different types of item. */
4711
4712 /* If previous was a character match, abolish the item and generate a
4713 repeat item instead. If a char item has a minumum of more than one, ensure
4714 that it is set in reqchar - it might not be if a sequence such as x{3} is
4715 the first thing in a branch because the x will have gone into firstchar
4716 instead. */
4717
4718 if (*previous == OP_CHAR || *previous == OP_CHARI)
4719 {
4720 op_type = (*previous == OP_CHAR)? 0 : OP_STARI - OP_STAR;
4721
4722 /* Deal with UTF characters that take up more than one character. It's
4723 easier to write this out separately than try to macrify it. Use c to
4724 hold the length of the character in bytes, plus UTF_LENGTH to flag that
4725 it's a length rather than a small character. */
4726
4727 #ifdef SUPPORT_UTF
4728 if (utf && NOT_FIRSTCHAR(code[-1]))
4729 {
4730 pcre_uchar *lastchar = code - 1;
4731 BACKCHAR(lastchar);
4732 c = (int)(code - lastchar); /* Length of UTF-8 character */
4733 memcpy(utf_chars, lastchar, IN_UCHARS(c)); /* Save the char */
4734 c |= UTF_LENGTH; /* Flag c as a length */
4735 }
4736 else
4737 #endif /* SUPPORT_UTF */
4738
4739 /* Handle the case of a single charater - either with no UTF support, or
4740 with UTF disabled, or for a single character UTF character. */
4741 {
4742 c = code[-1];
4743 if (repeat_min > 1) reqchar = c | req_caseopt | cd->req_varyopt;
4744 }
4745
4746 /* If the repetition is unlimited, it pays to see if the next thing on
4747 the line is something that cannot possibly match this character. If so,
4748 automatically possessifying this item gains some performance in the case
4749 where the match fails. */
4750
4751 if (!possessive_quantifier &&
4752 repeat_max < 0 &&
4753 check_auto_possessive(previous, utf, ptr + 1, options, cd))
4754 {
4755 repeat_type = 0; /* Force greedy */
4756 possessive_quantifier = TRUE;
4757 }
4758
4759 goto OUTPUT_SINGLE_REPEAT; /* Code shared with single character types */
4760 }
4761
4762 /* If previous was a single negated character ([^a] or similar), we use
4763 one of the special opcodes, replacing it. The code is shared with single-
4764 character repeats by setting opt_type to add a suitable offset into
4765 repeat_type. We can also test for auto-possessification. OP_NOT and OP_NOTI
4766 are currently used only for single-byte chars. */
4767
4768 else if (*previous == OP_NOT || *previous == OP_NOTI)
4769 {
4770 op_type = ((*previous == OP_NOT)? OP_NOTSTAR : OP_NOTSTARI) - OP_STAR;
4771 c = previous[1];
4772 if (!possessive_quantifier &&
4773 repeat_max < 0 &&
4774 check_auto_possessive(previous, utf, ptr + 1, options, cd))
4775 {
4776 repeat_type = 0; /* Force greedy */
4777 possessive_quantifier = TRUE;
4778 }
4779 goto OUTPUT_SINGLE_REPEAT;
4780 }
4781
4782 /* If previous was a character type match (\d or similar), abolish it and
4783 create a suitable repeat item. The code is shared with single-character
4784 repeats by setting op_type to add a suitable offset into repeat_type. Note
4785 the the Unicode property types will be present only when SUPPORT_UCP is
4786 defined, but we don't wrap the little bits of code here because it just
4787 makes it horribly messy. */
4788
4789 else if (*previous < OP_EODN)
4790 {
4791 pcre_uchar *oldcode;
4792 int prop_type, prop_value;
4793 op_type = OP_TYPESTAR - OP_STAR; /* Use type opcodes */
4794 c = *previous;
4795
4796 if (!possessive_quantifier &&
4797 repeat_max < 0 &&
4798 check_auto_possessive(previous, utf, ptr + 1, options, cd))
4799 {
4800 repeat_type = 0; /* Force greedy */
4801 possessive_quantifier = TRUE;
4802 }
4803
4804 OUTPUT_SINGLE_REPEAT:
4805 if (*previous == OP_PROP || *previous == OP_NOTPROP)
4806 {
4807 prop_type = previous[1];
4808 prop_value = previous[2];
4809 }
4810 else prop_type = prop_value = -1;
4811
4812 oldcode = code;
4813 code = previous; /* Usually overwrite previous item */
4814
4815 /* If the maximum is zero then the minimum must also be zero; Perl allows
4816 this case, so we do too - by simply omitting the item altogether. */
4817
4818 if (repeat_max == 0) goto END_REPEAT;
4819
4820 /*--------------------------------------------------------------------*/
4821 /* This code is obsolete from release 8.00; the restriction was finally
4822 removed: */
4823
4824 /* All real repeats make it impossible to handle partial matching (maybe
4825 one day we will be able to remove this restriction). */
4826
4827 /* if (repeat_max != 1) cd->external_flags |= PCRE_NOPARTIAL; */
4828 /*--------------------------------------------------------------------*/
4829
4830 /* Combine the op_type with the repeat_type */
4831
4832 repeat_type += op_type;
4833
4834 /* A minimum of zero is handled either as the special case * or ?, or as
4835 an UPTO, with the maximum given. */
4836
4837 if (repeat_min == 0)
4838 {
4839 if (repeat_max == -1) *code++ = OP_STAR + repeat_type;
4840 else if (repeat_max == 1) *code++ = OP_QUERY + repeat_type;
4841 else
4842 {
4843 *code++ = OP_UPTO + repeat_type;
4844 PUT2INC(code, 0, repeat_max);
4845 }
4846 }
4847
4848 /* A repeat minimum of 1 is optimized into some special cases. If the
4849 maximum is unlimited, we use OP_PLUS. Otherwise, the original item is
4850 left in place and, if the maximum is greater than 1, we use OP_UPTO with
4851 one less than the maximum. */
4852
4853 else if (repeat_min == 1)
4854 {
4855 if (repeat_max == -1)
4856 *code++ = OP_PLUS + repeat_type;
4857 else
4858 {
4859 code = oldcode; /* leave previous item in place */
4860 if (repeat_max == 1) goto END_REPEAT;
4861 *code++ = OP_UPTO + repeat_type;
4862 PUT2INC(code, 0, repeat_max - 1);
4863 }
4864 }
4865
4866 /* The case {n,n} is just an EXACT, while the general case {n,m} is
4867 handled as an EXACT followed by an UPTO. */
4868
4869 else
4870 {
4871 *code++ = OP_EXACT + op_type; /* NB EXACT doesn't have repeat_type */
4872 PUT2INC(code, 0, repeat_min);
4873
4874 /* If the maximum is unlimited, insert an OP_STAR. Before doing so,
4875 we have to insert the character for the previous code. For a repeated
4876 Unicode property match, there are two extra bytes that define the
4877 required property. In UTF-8 mode, long characters have their length in
4878 c, with the UTF_LENGTH bit as a flag. */
4879
4880 if (repeat_max < 0)
4881 {
4882 #ifdef SUPPORT_UTF
4883 if (utf && (c & UTF_LENGTH) != 0)
4884 {
4885 memcpy(code, utf_chars, IN_UCHARS(c & 7));
4886 code += c & 7;
4887 }
4888 else
4889 #endif
4890 {
4891 *code++ = c;
4892 if (prop_type >= 0)
4893 {
4894 *code++ = prop_type;
4895 *code++ = prop_value;
4896 }
4897 }
4898 *code++ = OP_STAR + repeat_type;
4899 }
4900
4901 /* Else insert an UPTO if the max is greater than the min, again
4902 preceded by the character, for the previously inserted code. If the
4903 UPTO is just for 1 instance, we can use QUERY instead. */
4904
4905 else if (repeat_max != repeat_min)
4906 {
4907 #ifdef SUPPORT_UTF
4908 if (utf && (c & UTF_LENGTH) != 0)
4909 {
4910 memcpy(code, utf_chars, IN_UCHARS(c & 7));
4911 code += c & 7;
4912 }
4913 else
4914 #endif
4915 *code++ = c;
4916 if (prop_type >= 0)
4917 {
4918 *code++ = prop_type;
4919 *code++ = prop_value;
4920 }
4921 repeat_max -= repeat_min;
4922
4923 if (repeat_max == 1)
4924 {
4925 *code++ = OP_QUERY + repeat_type;
4926 }
4927 else
4928 {
4929 *code++ = OP_UPTO + repeat_type;
4930 PUT2INC(code, 0, repeat_max);
4931 }
4932 }
4933 }
4934
4935 /* The character or character type itself comes last in all cases. */
4936
4937 #ifdef SUPPORT_UTF
4938 if (utf && (c & UTF_LENGTH) != 0)
4939 {
4940 memcpy(code, utf_chars, IN_UCHARS(c & 7));
4941 code += c & 7;
4942 }
4943 else
4944 #endif
4945 *code++ = c;
4946
4947 /* For a repeated Unicode property match, there are two extra bytes that
4948 define the required property. */
4949
4950 #ifdef SUPPORT_UCP
4951 if (prop_type >= 0)
4952 {
4953 *code++ = prop_type;
4954 *code++ = prop_value;
4955 }
4956 #endif
4957 }
4958
4959 /* If previous was a character class or a back reference, we put the repeat
4960 stuff after it, but just skip the item if the repeat was {0,0}. */
4961
4962 else if (*previous == OP_CLASS ||
4963 *previous == OP_NCLASS ||
4964 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
4965 *previous == OP_XCLASS ||
4966 #endif
4967 *previous == OP_REF ||
4968 *previous == OP_REFI)
4969 {
4970 if (repeat_max == 0)
4971 {
4972 code = previous;
4973 goto END_REPEAT;
4974 }
4975
4976 /*--------------------------------------------------------------------*/
4977 /* This code is obsolete from release 8.00; the restriction was finally
4978 removed: */
4979
4980 /* All real repeats make it impossible to handle partial matching (maybe
4981 one day we will be able to remove this restriction). */
4982
4983 /* if (repeat_max != 1) cd->external_flags |= PCRE_NOPARTIAL; */
4984 /*--------------------------------------------------------------------*/
4985
4986 if (repeat_min == 0 && repeat_max == -1)
4987 *code++ = OP_CRSTAR + repeat_type;
4988 else if (repeat_min == 1 && repeat_max == -1)
4989 *code++ = OP_CRPLUS + repeat_type;
4990 else if (repeat_min == 0 && repeat_max == 1)
4991 *code++ = OP_CRQUERY + repeat_type;
4992 else
4993 {
4994 *code++ = OP_CRRANGE + repeat_type;
4995 PUT2INC(code, 0, repeat_min);
4996 if (repeat_max == -1) repeat_max = 0; /* 2-byte encoding for max */
4997 PUT2INC(code, 0, repeat_max);
4998 }
4999 }
5000
5001 /* If previous was a bracket group, we may have to replicate it in certain
5002 cases. Note that at this point we can encounter only the "basic" bracket
5003 opcodes such as BRA and CBRA, as this is the place where they get converted
5004 into the more special varieties such as BRAPOS and SBRA. A test for >=
5005 OP_ASSERT and <= OP_COND includes ASSERT, ASSERT_NOT, ASSERTBACK,
5006 ASSERTBACK_NOT, ONCE, BRA, CBRA, and COND. Originally, PCRE did not allow
5007 repetition of assertions, but now it does, for Perl compatibility. */
5008
5009 else if (*previous >= OP_ASSERT && *previous <= OP_COND)
5010 {
5011 register int i;
5012 int len = (int)(code - previous);
5013 pcre_uchar *bralink = NULL;
5014 pcre_uchar *brazeroptr = NULL;
5015
5016 /* Repeating a DEFINE group is pointless, but Perl allows the syntax, so
5017 we just ignore the repeat. */
5018
5019 if (*previous == OP_COND && previous[LINK_SIZE+1] == OP_DEF)
5020 goto END_REPEAT;
5021
5022 /* There is no sense in actually repeating assertions. The only potential
5023 use of repetition is in cases when the assertion is optional. Therefore,
5024 if the minimum is greater than zero, just ignore the repeat. If the
5025 maximum is not not zero or one, set it to 1. */
5026
5027 if (*previous < OP_ONCE) /* Assertion */
5028 {
5029 if (repeat_min > 0) goto END_REPEAT;
5030 if (repeat_max < 0 || repeat_max > 1) repeat_max = 1;
5031 }
5032
5033 /* The case of a zero minimum is special because of the need to stick
5034 OP_BRAZERO in front of it, and because the group appears once in the
5035 data, whereas in other cases it appears the minimum number of times. For
5036 this reason, it is simplest to treat this case separately, as otherwise
5037 the code gets far too messy. There are several special subcases when the
5038 minimum is zero. */
5039
5040 if (repeat_min == 0)
5041 {
5042 /* If the maximum is also zero, we used to just omit the group from the
5043 output altogether, like this:
5044
5045 ** if (repeat_max == 0)
5046 ** {
5047 ** code = previous;
5048 ** goto END_REPEAT;
5049 ** }
5050
5051 However, that fails when a group or a subgroup within it is referenced
5052 as a subroutine from elsewhere in the pattern, so now we stick in
5053 OP_SKIPZERO in front of it so that it is skipped on execution. As we
5054 don't have a list of which groups are referenced, we cannot do this
5055 selectively.
5056
5057 If the maximum is 1 or unlimited, we just have to stick in the BRAZERO
5058 and do no more at this point. However, we do need to adjust any
5059 OP_RECURSE calls inside the group that refer to the group itself or any
5060 internal or forward referenced group, because the offset is from the
5061 start of the whole regex. Temporarily terminate the pattern while doing
5062 this. */
5063
5064 if (repeat_max <= 1) /* Covers 0, 1, and unlimited */
5065 {
5066 *code = OP_END;
5067 adjust_recurse(previous, 1, utf, cd, save_hwm);
5068 memmove(previous + 1, previous, IN_UCHARS(len));
5069 code++;
5070 if (repeat_max == 0)
5071 {
5072 *previous++ = OP_SKIPZERO;
5073 goto END_REPEAT;
5074 }
5075 brazeroptr = previous; /* Save for possessive optimizing */
5076 *previous++ = OP_BRAZERO + repeat_type;
5077 }
5078
5079 /* If the maximum is greater than 1 and limited, we have to replicate
5080 in a nested fashion, sticking OP_BRAZERO before each set of brackets.
5081 The first one has to be handled carefully because it's the original
5082 copy, which has to be moved up. The remainder can be handled by code
5083 that is common with the non-zero minimum case below. We have to
5084 adjust the value or repeat_max, since one less copy is required. Once
5085 again, we may have to adjust any OP_RECURSE calls inside the group. */
5086
5087 else
5088 {
5089 int offset;
5090 *code = OP_END;
5091 adjust_recurse(previous, 2 + LINK_SIZE, utf, cd, save_hwm);
5092 memmove(previous + 2 + LINK_SIZE, previous, IN_UCHARS(len));
5093 code += 2 + LINK_SIZE;
5094 *previous++ = OP_BRAZERO + repeat_type;
5095 *previous++ = OP_BRA;
5096
5097 /* We chain together the bracket offset fields that have to be
5098 filled in later when the ends of the brackets are reached. */
5099
5100 offset = (bralink == NULL)? 0 : (int)(previous - bralink);
5101 bralink = previous;
5102 PUTINC(previous, 0, offset);
5103 }
5104
5105 repeat_max--;
5106 }
5107
5108 /* If the minimum is greater than zero, replicate the group as many
5109 times as necessary, and adjust the maximum to the number of subsequent
5110 copies that we need. If we set a first char from the group, and didn't
5111 set a required char, copy the latter from the former. If there are any
5112 forward reference subroutine calls in the group, there will be entries on
5113 the workspace list; replicate these with an appropriate increment. */
5114
5115 else
5116 {
5117 if (repeat_min > 1)
5118 {
5119 /* In the pre-compile phase, we don't actually do the replication. We
5120 just adjust the length as if we had. Do some paranoid checks for
5121 potential integer overflow. The INT64_OR_DOUBLE type is a 64-bit
5122 integer type when available, otherwise double. */
5123
5124 if (lengthptr != NULL)
5125 {
5126 int delta = (repeat_min - 1)*length_prevgroup;
5127 if ((INT64_OR_DOUBLE)(repeat_min - 1)*
5128 (INT64_OR_DOUBLE)length_prevgroup >
5129 (INT64_OR_DOUBLE)INT_MAX ||
5130 OFLOW_MAX - *lengthptr < delta)
5131 {
5132 *errorcodeptr = ERR20;
5133 goto FAILED;
5134 }
5135 *lengthptr += delta;
5136 }
5137
5138 /* This is compiling for real. If there is a set first byte for
5139 the group, and we have not yet set a "required byte", set it. Make
5140 sure there is enough workspace for copying forward references before
5141 doing the copy. */
5142
5143 else
5144 {
5145 if (groupsetfirstchar && reqchar < 0) reqchar = firstchar;
5146
5147 for (i = 1; i < repeat_min; i++)
5148 {
5149 pcre_uchar *hc;
5150 pcre_uchar *this_hwm = cd->hwm;
5151 memcpy(code, previous, IN_UCHARS(len));
5152
5153 while (cd->hwm > cd->start_workspace + cd->workspace_size -
5154 WORK_SIZE_SAFETY_MARGIN - (this_hwm - save_hwm))
5155 {
5156 int save_offset = save_hwm - cd->start_workspace;
5157 int this_offset = this_hwm - cd->start_workspace;
5158 *errorcodeptr = expand_workspace(cd);
5159 if (*errorcodeptr != 0) goto FAILED;
5160 save_hwm = (pcre_uchar *)cd->start_workspace + save_offset;
5161 this_hwm = (pcre_uchar *)cd->start_workspace + this_offset;
5162 }
5163
5164 for (hc = save_hwm; hc < this_hwm; hc += LINK_SIZE)
5165 {
5166 PUT(cd->hwm, 0, GET(hc, 0) + len);
5167 cd->hwm += LINK_SIZE;
5168 }
5169 save_hwm = this_hwm;
5170 code += len;
5171 }
5172 }
5173 }
5174
5175 if (repeat_max > 0) repeat_max -= repeat_min;
5176 }
5177
5178 /* This code is common to both the zero and non-zero minimum cases. If
5179 the maximum is limited, it replicates the group in a nested fashion,
5180 remembering the bracket starts on a stack. In the case of a zero minimum,
5181 the first one was set up above. In all cases the repeat_max now specifies
5182 the number of additional copies needed. Again, we must remember to
5183 replicate entries on the forward reference list. */
5184
5185 if (repeat_max >= 0)
5186 {
5187 /* In the pre-compile phase, we don't actually do the replication. We
5188 just adjust the length as if we had. For each repetition we must add 1
5189 to the length for BRAZERO and for all but the last repetition we must
5190 add 2 + 2*LINKSIZE to allow for the nesting that occurs. Do some
5191 paranoid checks to avoid integer overflow. The INT64_OR_DOUBLE type is
5192 a 64-bit integer type when available, otherwise double. */
5193
5194 if (lengthptr != NULL && repeat_max > 0)
5195 {
5196 int delta = repeat_max * (length_prevgroup + 1 + 2 + 2*LINK_SIZE) -
5197 2 - 2*LINK_SIZE; /* Last one doesn't nest */
5198 if ((INT64_OR_DOUBLE)repeat_max *
5199 (INT64_OR_DOUBLE)(length_prevgroup + 1 + 2 + 2*LINK_SIZE)
5200 > (INT64_OR_DOUBLE)INT_MAX ||
5201 OFLOW_MAX - *lengthptr < delta)
5202 {
5203 *errorcodeptr = ERR20;
5204 goto FAILED;
5205 }
5206 *lengthptr += delta;
5207 }
5208
5209 /* This is compiling for real */
5210
5211 else for (i = repeat_max - 1; i >= 0; i--)
5212 {
5213 pcre_uchar *hc;
5214 pcre_uchar *this_hwm = cd->hwm;
5215
5216 *code++ = OP_BRAZERO + repeat_type;
5217
5218 /* All but the final copy start a new nesting, maintaining the
5219 chain of brackets outstanding. */
5220
5221 if (i != 0)
5222 {
5223 int offset;
5224 *code++ = OP_BRA;
5225 offset = (bralink == NULL)? 0 : (int)(code - bralink);
5226 bralink = code;
5227 PUTINC(code, 0, offset);
5228 }
5229
5230 memcpy(code, previous, IN_UCHARS(len));
5231
5232 /* Ensure there is enough workspace for forward references before
5233 copying them. */
5234
5235 while (cd->hwm > cd->start_workspace + cd->workspace_size -
5236 WORK_SIZE_SAFETY_MARGIN - (this_hwm - save_hwm))
5237 {
5238 int save_offset = save_hwm - cd->start_workspace;
5239 int this_offset = this_hwm - cd->start_workspace;
5240 *errorcodeptr = expand_workspace(cd);
5241 if (*errorcodeptr != 0) goto FAILED;
5242 save_hwm = (pcre_uchar *)cd->start_workspace + save_offset;
5243 this_hwm = (pcre_uchar *)cd->start_workspace + this_offset;
5244 }
5245
5246 for (hc = save_hwm; hc < this_hwm; hc += LINK_SIZE)
5247 {
5248 PUT(cd->hwm, 0, GET(hc, 0) + len + ((i != 0)? 2+LINK_SIZE : 1));
5249 cd->hwm += LINK_SIZE;
5250 }
5251 save_hwm = this_hwm;
5252 code += len;
5253 }
5254
5255 /* Now chain through the pending brackets, and fill in their length
5256 fields (which are holding the chain links pro tem). */
5257
5258 while (bralink != NULL)
5259 {
5260 int oldlinkoffset;
5261 int offset = (int)(code - bralink + 1);
5262 pcre_uchar *bra = code - offset;
5263 oldlinkoffset = GET(bra, 1);
5264 bralink = (oldlinkoffset == 0)? NULL : bralink - oldlinkoffset;
5265 *code++ = OP_KET;
5266 PUTINC(code, 0, offset);
5267 PUT(bra, 1, offset);
5268 }
5269 }
5270
5271 /* If the maximum is unlimited, set a repeater in the final copy. For
5272 ONCE brackets, that's all we need to do. However, possessively repeated
5273 ONCE brackets can be converted into non-capturing brackets, as the
5274 behaviour of (?:xx)++ is the same as (?>xx)++ and this saves having to
5275 deal with possessive ONCEs specially.
5276
5277 Otherwise, when we are doing the actual compile phase, check to see
5278 whether this group is one that could match an empty string. If so,
5279 convert the initial operator to the S form (e.g. OP_BRA -> OP_SBRA) so
5280 that runtime checking can be done. [This check is also applied to ONCE
5281 groups at runtime, but in a different way.]
5282
5283 Then, if the quantifier was possessive and the bracket is not a
5284 conditional, we convert the BRA code to the POS form, and the KET code to
5285 KETRPOS. (It turns out to be convenient at runtime to detect this kind of
5286 subpattern at both the start and at the end.) The use of special opcodes
5287 makes it possible to reduce greatly the stack usage in pcre_exec(). If
5288 the group is preceded by OP_BRAZERO, convert this to OP_BRAPOSZERO.
5289
5290 Then, if the minimum number of matches is 1 or 0, cancel the possessive
5291 flag so that the default action below, of wrapping everything inside
5292 atomic brackets, does not happen. When the minimum is greater than 1,
5293 there will be earlier copies of the group, and so we still have to wrap
5294 the whole thing. */
5295
5296 else
5297 {
5298 pcre_uchar *ketcode = code - 1 - LINK_SIZE;
5299 pcre_uchar *bracode = ketcode - GET(ketcode, 1);
5300
5301 /* Convert possessive ONCE brackets to non-capturing */
5302
5303 if ((*bracode == OP_ONCE || *bracode == OP_ONCE_NC) &&
5304 possessive_quantifier) *bracode = OP_BRA;
5305
5306 /* For non-possessive ONCE brackets, all we need to do is to
5307 set the KET. */
5308
5309 if (*bracode == OP_ONCE || *bracode == OP_ONCE_NC)
5310 *ketcode = OP_KETRMAX + repeat_type;
5311
5312 /* Handle non-ONCE brackets and possessive ONCEs (which have been
5313 converted to non-capturing above). */
5314
5315 else
5316 {
5317 /* In the compile phase, check for empty string matching. */
5318
5319 if (lengthptr == NULL)
5320 {
5321 pcre_uchar *scode = bracode;
5322 do
5323 {
5324 if (could_be_empty_branch(scode, ketcode, utf, cd))
5325 {
5326 *bracode += OP_SBRA - OP_BRA;
5327 break;
5328 }
5329 scode += GET(scode, 1);
5330 }
5331 while (*scode == OP_ALT);
5332 }
5333
5334 /* Handle possessive quantifiers. */
5335
5336 if (possessive_quantifier)
5337 {
5338 /* For COND brackets, we wrap the whole thing in a possessively
5339 repeated non-capturing bracket, because we have not invented POS
5340 versions of the COND opcodes. Because we are moving code along, we
5341 must ensure that any pending recursive references are updated. */
5342
5343 if (*bracode == OP_COND || *bracode == OP_SCOND)
5344 {
5345 int nlen = (int)(code - bracode);
5346 *code = OP_END;
5347 adjust_recurse(bracode, 1 + LINK_SIZE, utf, cd, save_hwm);
5348 memmove(bracode + 1 + LINK_SIZE, bracode, IN_UCHARS(nlen));
5349 code += 1 + LINK_SIZE;
5350 nlen += 1 + LINK_SIZE;
5351 *bracode = OP_BRAPOS;
5352 *code++ = OP_KETRPOS;
5353 PUTINC(code, 0, nlen);
5354 PUT(bracode, 1, nlen);
5355 }
5356
5357 /* For non-COND brackets, we modify the BRA code and use KETRPOS. */
5358
5359 else
5360 {
5361 *bracode += 1; /* Switch to xxxPOS opcodes */
5362 *ketcode = OP_KETRPOS;
5363 }
5364
5365 /* If the minimum is zero, mark it as possessive, then unset the
5366 possessive flag when the minimum is 0 or 1. */
5367
5368 if (brazeroptr != NULL) *brazeroptr = OP_BRAPOSZERO;
5369 if (repeat_min < 2) possessive_quantifier = FALSE;
5370 }
5371
5372 /* Non-possessive quantifier */
5373
5374 else *ketcode = OP_KETRMAX + repeat_type;
5375 }
5376 }
5377 }
5378
5379 /* If previous is OP_FAIL, it was generated by an empty class [] in
5380 JavaScript mode. The other ways in which OP_FAIL can be generated, that is
5381 by (*FAIL) or (?!) set previous to NULL, which gives a "nothing to repeat"
5382 error above. We can just ignore the repeat in JS case. */
5383
5384 else if (*previous == OP_FAIL) goto END_REPEAT;
5385
5386 /* Else there's some kind of shambles */
5387
5388 else
5389 {
5390 *errorcodeptr = ERR11;
5391 goto FAILED;
5392 }
5393
5394 /* If the character following a repeat is '+', or if certain optimization
5395 tests above succeeded, possessive_quantifier is TRUE. For some opcodes,
5396 there are special alternative opcodes for this case. For anything else, we
5397 wrap the entire repeated item inside OP_ONCE brackets. Logically, the '+'
5398 notation is just syntactic sugar, taken from Sun's Java package, but the
5399 special opcodes can optimize it.
5400
5401 Some (but not all) possessively repeated subpatterns have already been
5402 completely handled in the code just above. For them, possessive_quantifier
5403 is always FALSE at this stage.
5404
5405 Note that the repeated item starts at tempcode, not at previous, which
5406 might be the first part of a string whose (former) last char we repeated.
5407
5408 Possessifying an 'exact' quantifier has no effect, so we can ignore it. But
5409 an 'upto' may follow. We skip over an 'exact' item, and then test the
5410 length of what remains before proceeding. */
5411
5412 if (possessive_quantifier)
5413 {
5414 int len;
5415
5416 if (*tempcode == OP_TYPEEXACT)
5417 tempcode += PRIV(OP_lengths)[*tempcode] +
5418 ((tempcode[1 + IMM2_SIZE] == OP_PROP
5419 || tempcode[1 + IMM2_SIZE] == OP_NOTPROP)? 2 : 0);
5420
5421 else if (*tempcode == OP_EXACT || *tempcode == OP_NOTEXACT)
5422 {
5423 tempcode += PRIV(OP_lengths)[*tempcode];
5424 #ifdef SUPPORT_UTF
5425 if (utf && HAS_EXTRALEN(tempcode[-1]))
5426 tempcode += GET_EXTRALEN(tempcode[-1]);
5427 #endif
5428 }
5429
5430 len = (int)(code - tempcode);
5431 if (len > 0) switch (*tempcode)
5432 {
5433 case OP_STAR: *tempcode = OP_POSSTAR; break;
5434 case OP_PLUS: *tempcode = OP_POSPLUS; break;
5435 case OP_QUERY: *tempcode = OP_POSQUERY; break;
5436 case OP_UPTO: *tempcode = OP_POSUPTO; break;
5437
5438 case OP_STARI: *tempcode = OP_POSSTARI; break;
5439 case OP_PLUSI: *tempcode = OP_POSPLUSI; break;
5440 case OP_QUERYI: *tempcode = OP_POSQUERYI; break;
5441 case OP_UPTOI: *tempcode = OP_POSUPTOI; break;
5442
5443 case OP_NOTSTAR: *tempcode = OP_NOTPOSSTAR; break;
5444 case OP_NOTPLUS: *tempcode = OP_NOTPOSPLUS; break;
5445 case OP_NOTQUERY: *tempcode = OP_NOTPOSQUERY; break;
5446 case OP_NOTUPTO: *tempcode = OP_NOTPOSUPTO; break;
5447
5448 case OP_NOTSTARI: *tempcode = OP_NOTPOSSTARI; break;
5449 case OP_NOTPLUSI: *tempcode = OP_NOTPOSPLUSI; break;
5450 case OP_NOTQUERYI: *tempcode = OP_NOTPOSQUERYI; break;
5451 case OP_NOTUPTOI: *tempcode = OP_NOTPOSUPTOI; break;
5452
5453 case OP_TYPESTAR: *tempcode = OP_TYPEPOSSTAR; break;
5454 case OP_TYPEPLUS: *tempcode = OP_TYPEPOSPLUS; break;
5455 case OP_TYPEQUERY: *tempcode = OP_TYPEPOSQUERY; break;
5456 case OP_TYPEUPTO: *tempcode = OP_TYPEPOSUPTO; break;
5457
5458 /* Because we are moving code along, we must ensure that any
5459 pending recursive references are updated. */
5460
5461 default:
5462 *code = OP_END;
5463 adjust_recurse(tempcode, 1 + LINK_SIZE, utf, cd, save_hwm);
5464 memmove(tempcode + 1 + LINK_SIZE, tempcode, IN_UCHARS(len));
5465 code += 1 + LINK_SIZE;
5466 len += 1 + LINK_SIZE;
5467 tempcode[0] = OP_ONCE;
5468 *code++ = OP_KET;
5469 PUTINC(code, 0, len);
5470 PUT(tempcode, 1, len);
5471 break;
5472 }
5473 }
5474
5475 /* In all case we no longer have a previous item. We also set the
5476 "follows varying string" flag for subsequently encountered reqchars if
5477 it isn't already set and we have just passed a varying length item. */
5478
5479 END_REPEAT:
5480 previous = NULL;
5481 cd->req_varyopt |= reqvary;
5482 break;
5483
5484
5485 /* ===================================================================*/
5486 /* Start of nested parenthesized sub-expression, or comment or lookahead or
5487 lookbehind or option setting or condition or all the other extended
5488 parenthesis forms. */
5489
5490 case CHAR_LEFT_PARENTHESIS:
5491 newoptions = options;
5492 skipbytes = 0;
5493 bravalue = OP_CBRA;
5494 save_hwm = cd->hwm;
5495 reset_bracount = FALSE;
5496
5497 /* First deal with various "verbs" that can be introduced by '*'. */
5498
5499 ptr++;
5500 if (ptr[0] == CHAR_ASTERISK && (ptr[1] == ':'
5501 || (MAX_255(ptr[1]) && ((cd->ctypes[ptr[1]] & ctype_letter) != 0))))
5502 {
5503 int i, namelen;
5504 int arglen = 0;
5505 const char *vn = verbnames;
5506 const pcre_uchar *name = ptr + 1;
5507 const pcre_uchar *arg = NULL;
5508 previous = NULL;
5509 ptr++;
5510 while (MAX_255(*ptr) && (cd->ctypes[*ptr] & ctype_letter) != 0) ptr++;
5511 namelen = (int)(ptr - name);
5512
5513 /* It appears that Perl allows any characters whatsoever, other than
5514 a closing parenthesis, to appear in arguments, so we no longer insist on
5515 letters, digits, and underscores. */
5516
5517 if (*ptr == CHAR_COLON)
5518 {
5519 arg = ++ptr;
5520 while (*ptr != 0 && *ptr != CHAR_RIGHT_PARENTHESIS) ptr++;
5521 arglen = (int)(ptr - arg);
5522 }
5523
5524 if (*ptr != CHAR_RIGHT_PARENTHESIS)
5525 {
5526 *errorcodeptr = ERR60;
5527 goto FAILED;
5528 }
5529
5530 /* Scan the table of verb names */
5531
5532 for (i = 0; i < verbcount; i++)
5533 {
5534 if (namelen == verbs[i].len &&
5535 STRNCMP_UC_C8(name, vn, namelen) == 0)
5536 {
5537 /* Check for open captures before ACCEPT and convert it to
5538 ASSERT_ACCEPT if in an assertion. */
5539
5540 if (verbs[i].op == OP_ACCEPT)
5541 {
5542 open_capitem *oc;
5543 if (arglen != 0)
5544 {
5545 *errorcodeptr = ERR59;
5546 goto FAILED;
5547 }
5548 cd->had_accept = TRUE;
5549 for (oc = cd->open_caps; oc != NULL; oc = oc->next)
5550 {
5551 *code++ = OP_CLOSE;
5552 PUT2INC(code, 0, oc->number);
5553 }
5554 *code++ = (cd->assert_depth > 0)? OP_ASSERT_ACCEPT : OP_ACCEPT;
5555
5556 /* Do not set firstchar after *ACCEPT */
5557 if (firstchar == REQ_UNSET) firstchar = REQ_NONE;
5558 }
5559
5560 /* Handle other cases with/without an argument */
5561
5562 else if (arglen == 0)
5563 {
5564 if (verbs[i].op < 0) /* Argument is mandatory */
5565 {
5566 *errorcodeptr = ERR66;
5567 goto FAILED;
5568 }
5569 *code = verbs[i].op;
5570 if (*code++ == OP_THEN) cd->external_flags |= PCRE_HASTHEN;
5571 }
5572
5573 else
5574 {
5575 if (verbs[i].op_arg < 0) /* Argument is forbidden */
5576 {
5577 *errorcodeptr = ERR59;
5578 goto FAILED;
5579 }
5580 *code = verbs[i].op_arg;
5581 if (*code++ == OP_THEN_ARG) cd->external_flags |= PCRE_HASTHEN;
5582 *code++ = arglen;
5583 memcpy(code, arg, IN_UCHARS(arglen));
5584 code += arglen;
5585 *code++ = 0;
5586 }
5587
5588 break; /* Found verb, exit loop */
5589 }
5590
5591 vn += verbs[i].len + 1;
5592 }
5593
5594 if (i < verbcount) continue; /* Successfully handled a verb */
5595 *errorcodeptr = ERR60; /* Verb not recognized */
5596 goto FAILED;
5597 }
5598
5599 /* Deal with the extended parentheses; all are introduced by '?', and the
5600 appearance of any of them means that this is not a capturing group. */
5601
5602 else if (*ptr == CHAR_QUESTION_MARK)
5603 {
5604 int i, set, unset, namelen;
5605 int *optset;
5606 const pcre_uchar *name;
5607 pcre_uchar *slot;
5608
5609 switch (*(++ptr))
5610 {
5611 case CHAR_NUMBER_SIGN: /* Comment; skip to ket */
5612 ptr++;
5613 while (*ptr != 0 && *ptr != CHAR_RIGHT_PARENTHESIS) ptr++;
5614 if (*ptr == 0)
5615 {
5616 *errorcodeptr = ERR18;
5617 goto FAILED;
5618 }
5619 continue;
5620
5621
5622 /* ------------------------------------------------------------ */
5623 case CHAR_VERTICAL_LINE: /* Reset capture count for each branch */
5624 reset_bracount = TRUE;
5625 /* Fall through */
5626
5627 /* ------------------------------------------------------------ */
5628 case CHAR_COLON: /* Non-capturing bracket */
5629 bravalue = OP_BRA;
5630 ptr++;
5631 break;
5632
5633
5634 /* ------------------------------------------------------------ */
5635 case CHAR_LEFT_PARENTHESIS:
5636 bravalue = OP_COND; /* Conditional group */
5637
5638 /* A condition can be an assertion, a number (referring to a numbered
5639 group), a name (referring to a named group), or 'R', referring to
5640 recursion. R<digits> and R&name are also permitted for recursion tests.
5641
5642 There are several syntaxes for testing a named group: (?(name)) is used
5643 by Python; Perl 5.10 onwards uses (?(<name>) or (?('name')).
5644
5645 There are two unfortunate ambiguities, caused by history. (a) 'R' can
5646 be the recursive thing or the name 'R' (and similarly for 'R' followed
5647 by digits), and (b) a number could be a name that consists of digits.
5648 In both cases, we look for a name first; if not found, we try the other
5649 cases. */
5650
5651 /* For conditions that are assertions, check the syntax, and then exit
5652 the switch. This will take control down to where bracketed groups,
5653 including assertions, are processed. */
5654
5655 if (ptr[1] == CHAR_QUESTION_MARK && (ptr[2] == CHAR_EQUALS_SIGN ||
5656 ptr[2] == CHAR_EXCLAMATION_MARK || ptr[2] == CHAR_LESS_THAN_SIGN))
5657 break;
5658
5659 /* Most other conditions use OP_CREF (a couple change to OP_RREF
5660 below), and all need to skip 1+IMM2_SIZE bytes at the start of the group. */
5661
5662 code[1+LINK_SIZE] = OP_CREF;
5663 skipbytes = 1+IMM2_SIZE;
5664 refsign = -1;
5665
5666 /* Check for a test for recursion in a named group. */
5667
5668 if (ptr[1] == CHAR_R && ptr[2] == CHAR_AMPERSAND)
5669 {
5670 terminator = -1;
5671 ptr += 2;
5672 code[1+LINK_SIZE] = OP_RREF; /* Change the type of test */
5673 }
5674
5675 /* Check for a test for a named group's having been set, using the Perl
5676 syntax (?(<name>) or (?('name') */
5677
5678 else if (ptr[1] == CHAR_LESS_THAN_SIGN)
5679 {
5680 terminator = CHAR_GREATER_THAN_SIGN;
5681 ptr++;
5682 }
5683 else if (ptr[1] == CHAR_APOSTROPHE)
5684 {
5685 terminator = CHAR_APOSTROPHE;
5686 ptr++;
5687 }
5688 else
5689 {
5690 terminator = 0;
5691 if (ptr[1] == CHAR_MINUS || ptr[1] == CHAR_PLUS) refsign = *(++ptr);
5692 }
5693
5694 /* We now expect to read a name; any thing else is an error */
5695
5696 if (!MAX_255(ptr[1]) || (cd->ctypes[ptr[1]] & ctype_word) == 0)
5697 {
5698 ptr += 1; /* To get the right offset */
5699 *errorcodeptr = ERR28;
5700 goto FAILED;
5701 }
5702
5703 /* Read the name, but also get it as a number if it's all digits */
5704
5705 recno = 0;
5706 name = ++ptr;
5707 while (MAX_255(*ptr) && (cd->ctypes[*ptr] & ctype_word) != 0)
5708 {
5709 if (recno >= 0)
5710 recno = (IS_DIGIT(*ptr))? recno * 10 + *ptr - CHAR_0 : -1;
5711 ptr++;
5712 }
5713 namelen = (int)(ptr - name);
5714
5715 if ((terminator > 0 && *ptr++ != terminator) ||
5716 *ptr++ != CHAR_RIGHT_PARENTHESIS)
5717 {
5718 ptr--; /* Error offset */
5719 *errorcodeptr = ERR26;
5720 goto FAILED;
5721 }
5722
5723 /* Do no further checking in the pre-compile phase. */
5724
5725 if (lengthptr != NULL) break;
5726
5727 /* In the real compile we do the work of looking for the actual
5728 reference. If the string started with "+" or "-" we require the rest to
5729 be digits, in which case recno will be set. */
5730
5731 if (refsign > 0)
5732 {
5733 if (recno <= 0)
5734 {
5735 *errorcodeptr = ERR58;
5736 goto FAILED;
5737 }
5738 recno = (refsign == CHAR_MINUS)?
5739 cd->bracount - recno + 1 : recno +cd->bracount;
5740 if (recno <= 0 || recno > cd->final_bracount)
5741 {
5742 *errorcodeptr = ERR15;
5743 goto FAILED;
5744 }
5745 PUT2(code, 2+LINK_SIZE, recno);
5746 break;
5747 }
5748
5749 /* Otherwise (did not start with "+" or "-"), start by looking for the
5750 name. If we find a name, add one to the opcode to change OP_CREF or
5751 OP_RREF into OP_NCREF or OP_NRREF. These behave exactly the same,
5752 except they record that the reference was originally to a name. The
5753 information is used to check duplicate names. */
5754
5755 slot = cd->name_table;
5756 for (i = 0; i < cd->names_found; i++)
5757 {
5758 if (STRNCMP_UC_UC(name, slot+IMM2_SIZE, namelen) == 0) break;
5759 slot += cd->name_entry_size;
5760 }
5761
5762 /* Found a previous named subpattern */
5763
5764 if (i < cd->names_found)
5765 {
5766 recno = GET2(slot, 0);
5767 PUT2(code, 2+LINK_SIZE, recno);
5768 code[1+LINK_SIZE]++;
5769 }
5770
5771 /* Search the pattern for a forward reference */
5772
5773 else if ((i = find_parens(cd, name, namelen,
5774 (options & PCRE_EXTENDED) != 0, utf)) > 0)
5775 {
5776 PUT2(code, 2+LINK_SIZE, i);
5777 code[1+LINK_SIZE]++;
5778 }
5779
5780 /* If terminator == 0 it means that the name followed directly after
5781 the opening parenthesis [e.g. (?(abc)...] and in this case there are
5782 some further alternatives to try. For the cases where terminator != 0
5783 [things like (?(<name>... or (?('name')... or (?(R&name)... ] we have
5784 now checked all the possibilities, so give an error. */
5785
5786 else if (terminator != 0)
5787 {
5788 *errorcodeptr = ERR15;
5789 goto FAILED;
5790 }
5791
5792 /* Check for (?(R) for recursion. Allow digits after R to specify a
5793 specific group number. */
5794
5795 else if (*name == CHAR_R)
5796 {
5797 recno = 0;
5798 for (i = 1; i < namelen; i++)
5799 {
5800 if (!IS_DIGIT(name[i]))
5801 {
5802 *errorcodeptr = ERR15;
5803 goto FAILED;
5804 }
5805 recno = recno * 10 + name[i] - CHAR_0;
5806 }
5807 if (recno == 0) recno = RREF_ANY;
5808 code[1+LINK_SIZE] = OP_RREF; /* Change test type */
5809 PUT2(code, 2+LINK_SIZE, recno);
5810 }
5811
5812 /* Similarly, check for the (?(DEFINE) "condition", which is always
5813 false. */
5814
5815 else if (namelen == 6 && STRNCMP_UC_C8(name, STRING_DEFINE, 6) == 0)
5816 {
5817 code[1+LINK_SIZE] = OP_DEF;
5818 skipbytes = 1;
5819 }
5820
5821 /* Check for the "name" actually being a subpattern number. We are
5822 in the second pass here, so final_bracount is set. */
5823
5824 else if (recno > 0 && recno <= cd->final_bracount)
5825 {
5826 PUT2(code, 2+LINK_SIZE, recno);
5827 }
5828
5829 /* Either an unidentified subpattern, or a reference to (?(0) */
5830
5831 else
5832 {
5833 *errorcodeptr = (recno == 0)? ERR35: ERR15;
5834 goto FAILED;
5835 }
5836 break;
5837
5838
5839 /* ------------------------------------------------------------ */
5840 case CHAR_EQUALS_SIGN: /* Positive lookahead */
5841 bravalue = OP_ASSERT;
5842 cd->assert_depth += 1;
5843 ptr++;
5844 break;
5845
5846
5847 /* ------------------------------------------------------------ */
5848 case CHAR_EXCLAMATION_MARK: /* Negative lookahead */
5849 ptr++;
5850 if (*ptr == CHAR_RIGHT_PARENTHESIS) /* Optimize (?!) */
5851 {
5852 *code++ = OP_FAIL;
5853 previous = NULL;
5854 continue;
5855 }
5856 bravalue = OP_ASSERT_NOT;
5857 cd->assert_depth += 1;
5858 break;
5859
5860
5861 /* ------------------------------------------------------------ */
5862 case CHAR_LESS_THAN_SIGN: /* Lookbehind or named define */
5863 switch (ptr[1])
5864 {
5865 case CHAR_EQUALS_SIGN: /* Positive lookbehind */
5866 bravalue = OP_ASSERTBACK;
5867 cd->assert_depth += 1;
5868 ptr += 2;
5869 break;
5870
5871 case CHAR_EXCLAMATION_MARK: /* Negative lookbehind */
5872 bravalue = OP_ASSERTBACK_NOT;
5873 cd->assert_depth += 1;
5874 ptr += 2;
5875 break;
5876
5877 default: /* Could be name define, else bad */
5878 if (MAX_255(ptr[1]) && (cd->ctypes[ptr[1]] & ctype_word) != 0)
5879 goto DEFINE_NAME;
5880 ptr++; /* Correct offset for error */
5881 *errorcodeptr = ERR24;
5882 goto FAILED;
5883 }
5884 break;
5885
5886
5887 /* ------------------------------------------------------------ */
5888 case CHAR_GREATER_THAN_SIGN: /* One-time brackets */
5889 bravalue = OP_ONCE;
5890 ptr++;
5891 break;
5892
5893
5894 /* ------------------------------------------------------------ */
5895 case CHAR_C: /* Callout - may be followed by digits; */
5896 previous_callout = code; /* Save for later completion */
5897 after_manual_callout = 1; /* Skip one item before completing */
5898 *code++ = OP_CALLOUT;
5899 {
5900 int n = 0;
5901 ptr++;
5902 while(IS_DIGIT(*ptr))
5903 n = n * 10 + *ptr++ - CHAR_0;
5904 if (*ptr != CHAR_RIGHT_PARENTHESIS)
5905 {
5906 *errorcodeptr = ERR39;
5907 goto FAILED;
5908 }
5909 if (n > 255)
5910 {
5911 *errorcodeptr = ERR38;
5912 goto FAILED;
5913 }
5914 *code++ = n;
5915 PUT(code, 0, (int)(ptr - cd->start_pattern + 1)); /* Pattern offset */
5916 PUT(code, LINK_SIZE, 0); /* Default length */
5917 code += 2 * LINK_SIZE;
5918 }
5919 previous = NULL;
5920 continue;
5921
5922
5923 /* ------------------------------------------------------------ */
5924 case CHAR_P: /* Python-style named subpattern handling */
5925 if (*(++ptr) == CHAR_EQUALS_SIGN ||
5926 *ptr == CHAR_GREATER_THAN_SIGN) /* Reference or recursion */
5927 {
5928 is_recurse = *ptr == CHAR_GREATER_THAN_SIGN;
5929 terminator = CHAR_RIGHT_PARENTHESIS;
5930 goto NAMED_REF_OR_RECURSE;
5931 }
5932 else if (*ptr != CHAR_LESS_THAN_SIGN) /* Test for Python-style defn */
5933 {
5934 *errorcodeptr = ERR41;
5935 goto FAILED;
5936 }
5937 /* Fall through to handle (?P< as (?< is handled */
5938
5939
5940 /* ------------------------------------------------------------ */
5941 DEFINE_NAME: /* Come here from (?< handling */
5942 case CHAR_APOSTROPHE:
5943 {
5944 terminator = (*ptr == CHAR_LESS_THAN_SIGN)?
5945 CHAR_GREATER_THAN_SIGN : CHAR_APOSTROPHE;
5946 name = ++ptr;
5947
5948 while (MAX_255(*ptr) && (cd->ctypes[*ptr] & ctype_word) != 0) ptr++;
5949 namelen = (int)(ptr - name);
5950
5951 /* In the pre-compile phase, just do a syntax check. */
5952
5953 if (lengthptr != NULL)
5954 {
5955 if (*ptr != terminator)
5956 {
5957 *errorcodeptr = ERR42;
5958 goto FAILED;
5959 }
5960 if (cd->names_found >= MAX_NAME_COUNT)
5961 {
5962 *errorcodeptr = ERR49;
5963 goto FAILED;
5964 }
5965 if (namelen + IMM2_SIZE + 1 > cd->name_entry_size)
5966 {
5967 cd->name_entry_size = namelen + IMM2_SIZE + 1;
5968 if (namelen > MAX_NAME_SIZE)
5969 {
5970 *errorcodeptr = ERR48;
5971 goto FAILED;
5972 }
5973 }
5974 }
5975
5976 /* In the real compile, create the entry in the table, maintaining
5977 alphabetical order. Duplicate names for different numbers are
5978 permitted only if PCRE_DUPNAMES is set. Duplicate names for the same
5979 number are always OK. (An existing number can be re-used if (?|
5980 appears in the pattern.) In either event, a duplicate name results in
5981 a duplicate entry in the table, even if the number is the same. This
5982 is because the number of names, and hence the table size, is computed
5983 in the pre-compile, and it affects various numbers and pointers which
5984 would all have to be modified, and the compiled code moved down, if
5985 duplicates with the same number were omitted from the table. This
5986 doesn't seem worth the hassle. However, *different* names for the
5987 same number are not permitted. */
5988
5989 else
5990 {
5991 BOOL dupname = FALSE;
5992 slot = cd->name_table;
5993
5994 for (i = 0; i < cd->names_found; i++)
5995 {
5996 int crc = memcmp(name, slot+IMM2_SIZE, IN_UCHARS(namelen));
5997 if (crc == 0)
5998 {
5999 if (slot[IMM2_SIZE+namelen] == 0)
6000 {
6001 if (GET2(slot, 0) != cd->bracount + 1 &&
6002 (options & PCRE_DUPNAMES) == 0)
6003 {
6004 *errorcodeptr = ERR43;
6005 goto FAILED;
6006 }
6007 else dupname = TRUE;
6008 }
6009 else crc = -1; /* Current name is a substring */
6010 }
6011
6012 /* Make space in the table and break the loop for an earlier
6013 name. For a duplicate or later name, carry on. We do this for
6014 duplicates so that in the simple case (when ?(| is not used) they
6015 are in order of their numbers. */
6016
6017 if (crc < 0)
6018 {
6019 memmove(slot + cd->name_entry_size, slot,
6020 IN_UCHARS((cd->names_found - i) * cd->name_entry_size));
6021 break;
6022 }
6023
6024 /* Continue the loop for a later or duplicate name */
6025
6026 slot += cd->name_entry_size;
6027 }
6028
6029 /* For non-duplicate names, check for a duplicate number before
6030 adding the new name. */
6031
6032 if (!dupname)
6033 {
6034 pcre_uchar *cslot = cd->name_table;
6035 for (i = 0; i < cd->names_found; i++)
6036 {
6037 if (cslot != slot)
6038 {
6039 if (GET2(cslot, 0) == cd->bracount + 1)
6040 {
6041 *errorcodeptr = ERR65;
6042 goto FAILED;
6043 }
6044 }
6045 else i--;
6046 cslot += cd->name_entry_size;
6047 }
6048 }
6049
6050 PUT2(slot, 0, cd->bracount + 1);
6051 memcpy(slot + IMM2_SIZE, name, IN_UCHARS(namelen));
6052 slot[IMM2_SIZE + namelen] = 0;
6053 }
6054 }
6055
6056 /* In both pre-compile and compile, count the number of names we've
6057 encountered. */
6058
6059 cd->names_found++;
6060 ptr++; /* Move past > or ' */
6061 goto NUMBERED_GROUP;
6062
6063
6064 /* ------------------------------------------------------------ */
6065 case CHAR_AMPERSAND: /* Perl recursion/subroutine syntax */
6066 terminator = CHAR_RIGHT_PARENTHESIS;
6067 is_recurse = TRUE;
6068 /* Fall through */
6069
6070 /* We come here from the Python syntax above that handles both
6071 references (?P=name) and recursion (?P>name), as well as falling
6072 through from the Perl recursion syntax (?&name). We also come here from
6073 the Perl \k<name> or \k'name' back reference syntax and the \k{name}
6074 .NET syntax, and the Oniguruma \g<...> and \g'...' subroutine syntax. */
6075
6076 NAMED_REF_OR_RECURSE:
6077 name = ++ptr;
6078 while (MAX_255(*ptr) && (cd->ctypes[*ptr] & ctype_word) != 0) ptr++;
6079 namelen = (int)(ptr - name);
6080
6081 /* In the pre-compile phase, do a syntax check. We used to just set
6082 a dummy reference number, because it was not used in the first pass.
6083 However, with the change of recursive back references to be atomic,
6084 we have to look for the number so that this state can be identified, as
6085 otherwise the incorrect length is computed. If it's not a backwards
6086 reference, the dummy number will do. */
6087
6088 if (lengthptr != NULL)
6089 {
6090 const pcre_uchar *temp;
6091
6092 if (namelen == 0)
6093 {
6094 *errorcodeptr = ERR62;
6095 goto FAILED;
6096 }
6097 if (*ptr != terminator)
6098 {
6099 *errorcodeptr = ERR42;
6100 goto FAILED;
6101 }
6102 if (namelen > MAX_NAME_SIZE)
6103 {
6104 *errorcodeptr = ERR48;
6105 goto FAILED;
6106 }
6107
6108 /* The name table does not exist in the first pass, so we cannot
6109 do a simple search as in the code below. Instead, we have to scan the
6110 pattern to find the number. It is important that we scan it only as
6111 far as we have got because the syntax of named subpatterns has not
6112 been checked for the rest of the pattern, and find_parens() assumes
6113 correct syntax. In any case, it's a waste of resources to scan
6114 further. We stop the scan at the current point by temporarily
6115 adjusting the value of cd->endpattern. */
6116
6117 temp = cd->end_pattern;
6118 cd->end_pattern = ptr;
6119 recno = find_parens(cd, name, namelen,
6120 (options & PCRE_EXTENDED) != 0, utf);
6121 cd->end_pattern = temp;
6122 if (recno < 0) recno = 0; /* Forward ref; set dummy number */
6123 }
6124
6125 /* In the real compile, seek the name in the table. We check the name
6126 first, and then check that we have reached the end of the name in the
6127 table. That way, if the name that is longer than any in the table,
6128 the comparison will fail without reading beyond the table entry. */
6129
6130 else
6131 {
6132 slot = cd->name_table;
6133 for (i = 0; i < cd->names_found; i++)
6134 {
6135 if (STRNCMP_UC_UC(name, slot+IMM2_SIZE, namelen) == 0 &&
6136 slot[IMM2_SIZE+namelen] == 0)
6137 break;
6138 slot += cd->name_entry_size;
6139 }
6140
6141 if (i < cd->names_found) /* Back reference */
6142 {
6143 recno = GET2(slot, 0);
6144 }
6145 else if ((recno = /* Forward back reference */
6146 find_parens(cd, name, namelen,
6147 (options & PCRE_EXTENDED) != 0, utf)) <= 0)
6148 {
6149 *errorcodeptr = ERR15;
6150 goto FAILED;
6151 }
6152 }
6153
6154 /* In both phases, we can now go to the code than handles numerical
6155 recursion or backreferences. */
6156
6157 if (is_recurse) goto HANDLE_RECURSION;
6158 else goto HANDLE_REFERENCE;
6159
6160
6161 /* ------------------------------------------------------------ */
6162 case CHAR_R: /* Recursion */
6163 ptr++; /* Same as (?0) */
6164 /* Fall through */
6165
6166
6167 /* ------------------------------------------------------------ */
6168 case CHAR_MINUS: case CHAR_PLUS: /* Recursion or subroutine */
6169 case CHAR_0: case CHAR_1: case CHAR_2: case CHAR_3: case CHAR_4:
6170 case CHAR_5: case CHAR_6: case CHAR_7: case CHAR_8: case CHAR_9:
6171 {
6172 const pcre_uchar *called;
6173 terminator = CHAR_RIGHT_PARENTHESIS;
6174
6175 /* Come here from the \g<...> and \g'...' code (Oniguruma
6176 compatibility). However, the syntax has been checked to ensure that
6177 the ... are a (signed) number, so that neither ERR63 nor ERR29 will
6178 be called on this path, nor with the jump to OTHER_CHAR_AFTER_QUERY
6179 ever be taken. */
6180
6181 HANDLE_NUMERICAL_RECURSION:
6182
6183 if ((refsign = *ptr) == CHAR_PLUS)
6184 {
6185 ptr++;
6186 if (!IS_DIGIT(*ptr))
6187 {
6188 *errorcodeptr = ERR63;
6189 goto FAILED;
6190 }
6191 }
6192 else if (refsign == CHAR_MINUS)
6193 {
6194 if (!IS_DIGIT(ptr[1]))
6195 goto OTHER_CHAR_AFTER_QUERY;
6196 ptr++;
6197 }
6198
6199 recno = 0;
6200 while(IS_DIGIT(*ptr))
6201 recno = recno * 10 + *ptr++ - CHAR_0;
6202
6203 if (*ptr != terminator)
6204 {
6205 *errorcodeptr = ERR29;
6206 goto FAILED;
6207 }
6208
6209 if (refsign == CHAR_MINUS)
6210 {
6211 if (recno == 0)
6212 {
6213 *errorcodeptr = ERR58;
6214 goto FAILED;
6215 }
6216 recno = cd->bracount - recno + 1;
6217 if (recno <= 0)
6218 {
6219 *errorcodeptr = ERR15;
6220 goto FAILED;
6221 }
6222 }
6223 else if (refsign == CHAR_PLUS)
6224 {
6225 if (recno == 0)
6226 {
6227 *errorcodeptr = ERR58;
6228 goto FAILED;
6229 }
6230 recno += cd->bracount;
6231 }
6232
6233 /* Come here from code above that handles a named recursion */
6234
6235 HANDLE_RECURSION:
6236
6237 previous = code;
6238 called = cd->start_code;
6239
6240 /* When we are actually compiling, find the bracket that is being
6241 referenced. Temporarily end the regex in case it doesn't exist before
6242 this point. If we end up with a forward reference, first check that
6243 the bracket does occur later so we can give the error (and position)
6244 now. Then remember this forward reference in the workspace so it can
6245 be filled in at the end. */
6246
6247 if (lengthptr == NULL)
6248 {
6249 *code = OP_END;
6250 if (recno != 0)
6251 called = PRIV(find_bracket)(cd->start_code, utf, recno);
6252
6253 /* Forward reference */
6254
6255 if (called == NULL)
6256 {
6257 if (find_parens(cd, NULL, recno,
6258 (options & PCRE_EXTENDED) != 0, utf) < 0)
6259 {
6260 *errorcodeptr = ERR15;
6261 goto FAILED;
6262 }
6263
6264 /* Fudge the value of "called" so that when it is inserted as an
6265 offset below, what it actually inserted is the reference number
6266 of the group. Then remember the forward reference. */
6267
6268 called = cd->start_code + recno;
6269 if (cd->hwm >= cd->start_workspace + cd->workspace_size -
6270 WORK_SIZE_SAFETY_MARGIN)
6271 {
6272 *errorcodeptr = expand_workspace(cd);
6273 if (*errorcodeptr != 0) goto FAILED;
6274 }
6275 PUTINC(cd->hwm, 0, (int)(code + 1 - cd->start_code));
6276 }
6277
6278 /* If not a forward reference, and the subpattern is still open,
6279 this is a recursive call. We check to see if this is a left
6280 recursion that could loop for ever, and diagnose that case. We
6281 must not, however, do this check if we are in a conditional
6282 subpattern because the condition might be testing for recursion in
6283 a pattern such as /(?(R)a+|(?R)b)/, which is perfectly valid.
6284 Forever loops are also detected at runtime, so those that occur in
6285 conditional subpatterns will be picked up then. */
6286
6287 else if (GET(called, 1) == 0 && cond_depth <= 0 &&
6288 could_be_empty(called, code, bcptr, utf, cd))
6289 {
6290 *errorcodeptr = ERR40;
6291 goto FAILED;
6292 }
6293 }
6294
6295 /* Insert the recursion/subroutine item. It does not have a set first
6296 character (relevant if it is repeated, because it will then be
6297 wrapped with ONCE brackets). */
6298
6299 *code = OP_RECURSE;
6300 PUT(code, 1, (int)(called - cd->start_code));
6301 code += 1 + LINK_SIZE;
6302 groupsetfirstchar = FALSE;
6303 }
6304
6305 /* Can't determine a first byte now */
6306
6307 if (firstchar == REQ_UNSET) firstchar = REQ_NONE;
6308 continue;
6309
6310
6311 /* ------------------------------------------------------------ */
6312 default: /* Other characters: check option setting */
6313 OTHER_CHAR_AFTER_QUERY:
6314 set = unset = 0;
6315 optset = &set;
6316
6317 while (*ptr != CHAR_RIGHT_PARENTHESIS && *ptr != CHAR_COLON)
6318 {
6319 switch (*ptr++)
6320 {
6321 case CHAR_MINUS: optset = &unset; break;
6322
6323 case CHAR_J: /* Record that it changed in the external options */
6324 *optset |= PCRE_DUPNAMES;
6325 cd->external_flags |= PCRE_JCHANGED;
6326 break;
6327
6328 case CHAR_i: *optset |= PCRE_CASELESS; break;
6329 case CHAR_m: *optset |= PCRE_MULTILINE; break;
6330 case CHAR_s: *optset |= PCRE_DOTALL; break;
6331 case CHAR_x: *optset |= PCRE_EXTENDED; break;
6332 case CHAR_U: *optset |= PCRE_UNGREEDY; break;
6333 case CHAR_X: *optset |= PCRE_EXTRA; break;
6334
6335 default: *errorcodeptr = ERR12;
6336 ptr--; /* Correct the offset */
6337 goto FAILED;
6338 }
6339 }
6340
6341 /* Set up the changed option bits, but don't change anything yet. */
6342
6343 newoptions = (options | set) & (~unset);
6344
6345 /* If the options ended with ')' this is not the start of a nested
6346 group with option changes, so the options change at this level. If this
6347 item is right at the start of the pattern, the options can be
6348 abstracted and made external in the pre-compile phase, and ignored in
6349 the compile phase. This can be helpful when matching -- for instance in
6350 caseless checking of required bytes.
6351
6352 If the code pointer is not (cd->start_code + 1 + LINK_SIZE), we are
6353 definitely *not* at the start of the pattern because something has been
6354 compiled. In the pre-compile phase, however, the code pointer can have
6355 that value after the start, because it gets reset as code is discarded
6356 during the pre-compile. However, this can happen only at top level - if
6357 we are within parentheses, the starting BRA will still be present. At
6358 any parenthesis level, the length value can be used to test if anything
6359 has been compiled at that level. Thus, a test for both these conditions
6360 is necessary to ensure we correctly detect the start of the pattern in
6361 both phases.
6362
6363 If we are not at the pattern start, reset the greedy defaults and the
6364 case value for firstchar and reqchar. */
6365
6366 if (*ptr == CHAR_RIGHT_PARENTHESIS)
6367 {
6368 if (code == cd->start_code + 1 + LINK_SIZE &&
6369 (lengthptr == NULL || *lengthptr == 2 + 2*LINK_SIZE))
6370 {
6371 cd->external_options = newoptions;
6372 }
6373 else
6374 {
6375 greedy_default = ((newoptions & PCRE_UNGREEDY) != 0);
6376 greedy_non_default = greedy_default ^ 1;
6377 req_caseopt = ((newoptions & PCRE_CASELESS) != 0)? REQ_CASELESS:0;
6378 }
6379
6380 /* Change options at this level, and pass them back for use
6381 in subsequent branches. */
6382
6383 *optionsptr = options = newoptions;
6384 previous = NULL; /* This item can't be repeated */
6385 continue; /* It is complete */
6386 }
6387
6388 /* If the options ended with ':' we are heading into a nested group
6389 with possible change of options. Such groups are non-capturing and are
6390 not assertions of any kind. All we need to do is skip over the ':';
6391 the newoptions value is handled below. */
6392
6393 bravalue = OP_BRA;
6394 ptr++;
6395 } /* End of switch for character following (? */
6396 } /* End of (? handling */
6397
6398 /* Opening parenthesis not followed by '*' or '?'. If PCRE_NO_AUTO_CAPTURE
6399 is set, all unadorned brackets become non-capturing and behave like (?:...)
6400 brackets. */
6401
6402 else if ((options & PCRE_NO_AUTO_CAPTURE) != 0)
6403 {
6404 bravalue = OP_BRA;
6405 }
6406
6407 /* Else we have a capturing group. */
6408
6409 else
6410 {
6411 NUMBERED_GROUP:
6412 cd->bracount += 1;
6413 PUT2(code, 1+LINK_SIZE, cd->bracount);
6414 skipbytes = IMM2_SIZE;
6415 }
6416
6417 /* Process nested bracketed regex. Assertions used not to be repeatable,
6418 but this was changed for Perl compatibility, so all kinds can now be
6419 repeated. We copy code into a non-register variable (tempcode) in order to
6420 be able to pass its address because some compilers complain otherwise. */
6421
6422 previous = code; /* For handling repetition */
6423 *code = bravalue;
6424 tempcode = code;
6425 tempreqvary = cd->req_varyopt; /* Save value before bracket */
6426 tempbracount = cd->bracount; /* Save value before bracket */
6427 length_prevgroup = 0; /* Initialize for pre-compile phase */
6428
6429 if (!compile_regex(
6430 newoptions, /* The complete new option state */
6431 &tempcode, /* Where to put code (updated) */
6432 &ptr, /* Input pointer (updated) */
6433 errorcodeptr, /* Where to put an error message */
6434 (bravalue == OP_ASSERTBACK ||
6435 bravalue == OP_ASSERTBACK_NOT), /* TRUE if back assert */
6436 reset_bracount, /* True if (?| group */
6437 skipbytes, /* Skip over bracket number */
6438 cond_depth +
6439 ((bravalue == OP_COND)?1:0), /* Depth of condition subpatterns */
6440 &subfirstchar, /* For possible first char */
6441 &subreqchar, /* For possible last char */
6442 bcptr, /* Current branch chain */
6443 cd, /* Tables block */
6444 (lengthptr == NULL)? NULL : /* Actual compile phase */
6445 &length_prevgroup /* Pre-compile phase */
6446 ))
6447 goto FAILED;
6448
6449 /* If this was an atomic group and there are no capturing groups within it,
6450 generate OP_ONCE_NC instead of OP_ONCE. */
6451
6452 if (bravalue == OP_ONCE && cd->bracount <= tempbracount)
6453 *code = OP_ONCE_NC;
6454
6455 if (bravalue >= OP_ASSERT && bravalue <= OP_ASSERTBACK_NOT)
6456 cd->assert_depth -= 1;
6457
6458 /* At the end of compiling, code is still pointing to the start of the
6459 group, while tempcode has been updated to point past the end of the group.
6460 The pattern pointer (ptr) is on the bracket.
6461
6462 If this is a conditional bracket, check that there are no more than
6463 two branches in the group, or just one if it's a DEFINE group. We do this
6464 in the real compile phase, not in the pre-pass, where the whole group may
6465 not be available. */
6466
6467 if (bravalue == OP_COND && lengthptr == NULL)
6468 {
6469 pcre_uchar *tc = code;
6470 int condcount = 0;
6471
6472 do {
6473 condcount++;
6474 tc += GET(tc,1);
6475 }
6476 while (*tc != OP_KET);
6477
6478 /* A DEFINE group is never obeyed inline (the "condition" is always
6479 false). It must have only one branch. */
6480
6481 if (code[LINK_SIZE+1] == OP_DEF)
6482 {
6483 if (condcount > 1)
6484 {
6485 *errorcodeptr = ERR54;
6486 goto FAILED;
6487 }
6488 bravalue = OP_DEF; /* Just a flag to suppress char handling below */
6489 }
6490
6491 /* A "normal" conditional group. If there is just one branch, we must not
6492 make use of its firstchar or reqchar, because this is equivalent to an
6493 empty second branch. */
6494
6495 else
6496 {
6497 if (condcount > 2)
6498 {
6499 *errorcodeptr = ERR27;
6500 goto FAILED;
6501 }
6502 if (condcount == 1) subfirstchar = subreqchar = REQ_NONE;
6503 }
6504 }
6505
6506 /* Error if hit end of pattern */
6507
6508 if (*ptr != CHAR_RIGHT_PARENTHESIS)
6509 {
6510 *errorcodeptr = ERR14;
6511 goto FAILED;
6512 }
6513
6514 /* In the pre-compile phase, update the length by the length of the group,
6515 less the brackets at either end. Then reduce the compiled code to just a
6516 set of non-capturing brackets so that it doesn't use much memory if it is
6517 duplicated by a quantifier.*/
6518
6519 if (lengthptr != NULL)
6520 {
6521 if (OFLOW_MAX - *lengthptr < length_prevgroup - 2 - 2*LINK_SIZE)
6522 {
6523 *errorcodeptr = ERR20;
6524 goto FAILED;
6525 }
6526 *lengthptr += length_prevgroup - 2 - 2*LINK_SIZE;
6527 code++; /* This already contains bravalue */
6528 PUTINC(code, 0, 1 + LINK_SIZE);
6529 *code++ = OP_KET;
6530 PUTINC(code, 0, 1 + LINK_SIZE);
6531 break; /* No need to waste time with special character handling */
6532 }
6533
6534 /* Otherwise update the main code pointer to the end of the group. */
6535
6536 code = tempcode;
6537
6538 /* For a DEFINE group, required and first character settings are not
6539 relevant. */
6540
6541 if (bravalue == OP_DEF) break;
6542
6543 /* Handle updating of the required and first characters for other types of
6544 group. Update for normal brackets of all kinds, and conditions with two
6545 branches (see code above). If the bracket is followed by a quantifier with
6546 zero repeat, we have to back off. Hence the definition of zeroreqchar and
6547 zerofirstchar outside the main loop so that they can be accessed for the
6548 back off. */
6549
6550 zeroreqchar = reqchar;
6551 zerofirstchar = firstchar;
6552 groupsetfirstchar = FALSE;
6553
6554 if (bravalue >= OP_ONCE)
6555 {
6556 /* If we have not yet set a firstchar in this branch, take it from the
6557 subpattern, remembering that it was set here so that a repeat of more
6558 than one can replicate it as reqchar if necessary. If the subpattern has
6559 no firstchar, set "none" for the whole branch. In both cases, a zero
6560 repeat forces firstchar to "none". */
6561
6562 if (firstchar == REQ_UNSET)
6563 {
6564 if (subfirstchar >= 0)
6565 {
6566 firstchar = subfirstchar;
6567 groupsetfirstchar = TRUE;
6568 }
6569 else firstchar = REQ_NONE;
6570 zerofirstchar = REQ_NONE;
6571 }
6572
6573 /* If firstchar was previously set, convert the subpattern's firstchar
6574 into reqchar if there wasn't one, using the vary flag that was in
6575 existence beforehand. */
6576
6577 else if (subfirstchar >= 0 && subreqchar < 0)
6578 subreqchar = subfirstchar | tempreqvary;
6579
6580 /* If the subpattern set a required byte (or set a first byte that isn't
6581 really the first byte - see above), set it. */
6582
6583 if (subreqchar >= 0) reqchar = subreqchar;
6584 }
6585
6586 /* For a forward assertion, we take the reqchar, if set. This can be
6587 helpful if the pattern that follows the assertion doesn't set a different
6588 char. For example, it's useful for /(?=abcde).+/. We can't set firstchar
6589 for an assertion, however because it leads to incorrect effect for patterns
6590 such as /(?=a)a.+/ when the "real" "a" would then become a reqchar instead
6591 of a firstchar. This is overcome by a scan at the end if there's no
6592 firstchar, looking for an asserted first char. */
6593
6594 else if (bravalue == OP_ASSERT && subreqchar >= 0) reqchar = subreqchar;
6595 break; /* End of processing '(' */
6596
6597
6598 /* ===================================================================*/
6599 /* Handle metasequences introduced by \. For ones like \d, the ESC_ values
6600 are arranged to be the negation of the corresponding OP_values in the
6601 default case when PCRE_UCP is not set. For the back references, the values
6602 are ESC_REF plus the reference number. Only back references and those types
6603 that consume a character may be repeated. We can test for values between
6604 ESC_b and ESC_Z for the latter; this may have to change if any new ones are
6605 ever created. */
6606
6607 case CHAR_BACKSLASH:
6608 tempptr = ptr;
6609 c = check_escape(&ptr, errorcodeptr, cd->bracount, options, FALSE);
6610 if (*errorcodeptr != 0) goto FAILED;
6611
6612 if (c < 0)
6613 {
6614 if (-c == ESC_Q) /* Handle start of quoted string */
6615 {
6616 if (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
6617 ptr += 2; /* avoid empty string */
6618 else inescq = TRUE;
6619 continue;
6620 }
6621
6622 if (-c == ESC_E) continue; /* Perl ignores an orphan \E */
6623
6624 /* For metasequences that actually match a character, we disable the
6625 setting of a first character if it hasn't already been set. */
6626
6627 if (firstchar == REQ_UNSET && -c > ESC_b && -c < ESC_Z)
6628 firstchar = REQ_NONE;
6629
6630 /* Set values to reset to if this is followed by a zero repeat. */
6631
6632 zerofirstchar = firstchar;