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

Contents of /code/trunk/pcre_compile.c

Parent Directory Parent Directory | Revision Log Revision Log


Revision 982 - (show annotations)
Wed Jun 20 15:15:27 2012 UTC (7 years, 3 months ago) by ph10
File MIME type: text/plain
File size: 269756 byte(s)
Source tidies for 8.31-RC2.
1 /*************************************************
2 * Perl-Compatible Regular Expressions *
3 *************************************************/
4
5 /* PCRE is a library of functions to support regular expressions whose syntax
6 and semantics are as close as possible to those of the Perl 5 language.
7
8 Written by Philip Hazel
9 Copyright (c) 1997-2012 University of Cambridge
10
11 -----------------------------------------------------------------------------
12 Redistribution and use in source and binary forms, with or without
13 modification, are permitted provided that the following conditions are met:
14
15 * Redistributions of source code must retain the above copyright notice,
16 this list of conditions and the following disclaimer.
17
18 * Redistributions in binary form must reproduce the above copyright
19 notice, this list of conditions and the following disclaimer in the
20 documentation and/or other materials provided with the distribution.
21
22 * Neither the name of the University of Cambridge nor the names of its
23 contributors may be used to endorse or promote products derived from
24 this software without specific prior written permission.
25
26 THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
27 AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28 IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29 ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE
30 LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
31 CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
32 SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
33 INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
34 CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
35 ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
36 POSSIBILITY OF SUCH DAMAGE.
37 -----------------------------------------------------------------------------
38 */
39
40
41 /* This module contains the external function pcre_compile(), along with
42 supporting internal functions that are not used by other modules. */
43
44
45 #ifdef HAVE_CONFIG_H
46 #include "config.h"
47 #endif
48
49 #define NLBLOCK cd /* Block containing newline information */
50 #define PSSTART start_pattern /* Field containing processed string start */
51 #define PSEND end_pattern /* Field containing processed string end */
52
53 #include "pcre_internal.h"
54
55
56 /* When PCRE_DEBUG is defined, we need the pcre(16)_printint() function, which
57 is also used by pcretest. PCRE_DEBUG is not defined when building a production
58 library. We do not need to select pcre16_printint.c specially, because the
59 COMPILE_PCREx macro will already be appropriately set. */
60
61 #ifdef PCRE_DEBUG
62 /* pcre_printint.c should not include any headers */
63 #define PCRE_INCLUDED
64 #include "pcre_printint.c"
65 #undef PCRE_INCLUDED
66 #endif
67
68
69 /* Macro for setting individual bits in class bitmaps. */
70
71 #define SETBIT(a,b) a[b/8] |= (1 << (b%8))
72
73 /* Maximum length value to check against when making sure that the integer that
74 holds the compiled pattern length does not overflow. We make it a bit less than
75 INT_MAX to allow for adding in group terminating bytes, so that we don't have
76 to check them every time. */
77
78 #define OFLOW_MAX (INT_MAX - 20)
79
80
81 /*************************************************
82 * Code parameters and static tables *
83 *************************************************/
84
85 /* This value specifies the size of stack workspace that is used during the
86 first pre-compile phase that determines how much memory is required. The regex
87 is partly compiled into this space, but the compiled parts are discarded as
88 soon as they can be, so that hopefully there will never be an overrun. The code
89 does, however, check for an overrun. The largest amount I've seen used is 218,
90 so this number is very generous.
91
92 The same workspace is used during the second, actual compile phase for
93 remembering forward references to groups so that they can be filled in at the
94 end. Each entry in this list occupies LINK_SIZE bytes, so even when LINK_SIZE
95 is 4 there is plenty of room for most patterns. However, the memory can get
96 filled up by repetitions of forward references, for example patterns like
97 /(?1){0,1999}(b)/, and one user did hit the limit. The code has been changed so
98 that the workspace is expanded using malloc() in this situation. The value
99 below is therefore a minimum, and we put a maximum on it for safety. The
100 minimum is now also defined in terms of LINK_SIZE so that the use of malloc()
101 kicks in at the same number of forward references in all cases. */
102
103 #define COMPILE_WORK_SIZE (2048*LINK_SIZE)
104 #define COMPILE_WORK_SIZE_MAX (100*COMPILE_WORK_SIZE)
105
106 /* The overrun tests check for a slightly smaller size so that they detect the
107 overrun before it actually does run off the end of the data block. */
108
109 #define WORK_SIZE_SAFETY_MARGIN (100)
110
111 /* Private flags added to firstchar and reqchar. */
112
113 #define REQ_CASELESS 0x10000000l /* Indicates caselessness */
114 #define REQ_VARY 0x20000000l /* Reqchar followed non-literal item */
115
116 /* Repeated character flags. */
117
118 #define UTF_LENGTH 0x10000000l /* The char contains its length. */
119
120 /* Table for handling escaped characters in the range '0'-'z'. Positive returns
121 are simple data values; negative values are for special things like \d and so
122 on. Zero means further processing is needed (for things like \x), or the escape
123 is invalid. */
124
125 #ifndef EBCDIC
126
127 /* This is the "normal" table for ASCII systems or for EBCDIC systems running
128 in UTF-8 mode. */
129
130 static const short int escapes[] = {
131 0, 0,
132 0, 0,
133 0, 0,
134 0, 0,
135 0, 0,
136 CHAR_COLON, CHAR_SEMICOLON,
137 CHAR_LESS_THAN_SIGN, CHAR_EQUALS_SIGN,
138 CHAR_GREATER_THAN_SIGN, CHAR_QUESTION_MARK,
139 CHAR_COMMERCIAL_AT, -ESC_A,
140 -ESC_B, -ESC_C,
141 -ESC_D, -ESC_E,
142 0, -ESC_G,
143 -ESC_H, 0,
144 0, -ESC_K,
145 0, 0,
146 -ESC_N, 0,
147 -ESC_P, -ESC_Q,
148 -ESC_R, -ESC_S,
149 0, 0,
150 -ESC_V, -ESC_W,
151 -ESC_X, 0,
152 -ESC_Z, CHAR_LEFT_SQUARE_BRACKET,
153 CHAR_BACKSLASH, CHAR_RIGHT_SQUARE_BRACKET,
154 CHAR_CIRCUMFLEX_ACCENT, CHAR_UNDERSCORE,
155 CHAR_GRAVE_ACCENT, 7,
156 -ESC_b, 0,
157 -ESC_d, ESC_e,
158 ESC_f, 0,
159 -ESC_h, 0,
160 0, -ESC_k,
161 0, 0,
162 ESC_n, 0,
163 -ESC_p, 0,
164 ESC_r, -ESC_s,
165 ESC_tee, 0,
166 -ESC_v, -ESC_w,
167 0, 0,
168 -ESC_z
169 };
170
171 #else
172
173 /* This is the "abnormal" table for EBCDIC systems without UTF-8 support. */
174
175 static const short int escapes[] = {
176 /* 48 */ 0, 0, 0, '.', '<', '(', '+', '|',
177 /* 50 */ '&', 0, 0, 0, 0, 0, 0, 0,
178 /* 58 */ 0, 0, '!', '$', '*', ')', ';', '~',
179 /* 60 */ '-', '/', 0, 0, 0, 0, 0, 0,
180 /* 68 */ 0, 0, '|', ',', '%', '_', '>', '?',
181 /* 70 */ 0, 0, 0, 0, 0, 0, 0, 0,
182 /* 78 */ 0, '`', ':', '#', '@', '\'', '=', '"',
183 /* 80 */ 0, 7, -ESC_b, 0, -ESC_d, ESC_e, ESC_f, 0,
184 /* 88 */-ESC_h, 0, 0, '{', 0, 0, 0, 0,
185 /* 90 */ 0, 0, -ESC_k, 'l', 0, ESC_n, 0, -ESC_p,
186 /* 98 */ 0, ESC_r, 0, '}', 0, 0, 0, 0,
187 /* A0 */ 0, '~', -ESC_s, ESC_tee, 0,-ESC_v, -ESC_w, 0,
188 /* A8 */ 0,-ESC_z, 0, 0, 0, '[', 0, 0,
189 /* B0 */ 0, 0, 0, 0, 0, 0, 0, 0,
190 /* B8 */ 0, 0, 0, 0, 0, ']', '=', '-',
191 /* C0 */ '{',-ESC_A, -ESC_B, -ESC_C, -ESC_D,-ESC_E, 0, -ESC_G,
192 /* C8 */-ESC_H, 0, 0, 0, 0, 0, 0, 0,
193 /* D0 */ '}', 0, -ESC_K, 0, 0,-ESC_N, 0, -ESC_P,
194 /* D8 */-ESC_Q,-ESC_R, 0, 0, 0, 0, 0, 0,
195 /* E0 */ '\\', 0, -ESC_S, 0, 0,-ESC_V, -ESC_W, -ESC_X,
196 /* E8 */ 0,-ESC_Z, 0, 0, 0, 0, 0, 0,
197 /* F0 */ 0, 0, 0, 0, 0, 0, 0, 0,
198 /* F8 */ 0, 0, 0, 0, 0, 0, 0, 0
199 };
200 #endif
201
202
203 /* Table of special "verbs" like (*PRUNE). This is a short table, so it is
204 searched linearly. Put all the names into a single string, in order to reduce
205 the number of relocations when a shared library is dynamically linked. The
206 string is built from string macros so that it works in UTF-8 mode on EBCDIC
207 platforms. */
208
209 typedef struct verbitem {
210 int len; /* Length of verb name */
211 int op; /* Op when no arg, or -1 if arg mandatory */
212 int op_arg; /* Op when arg present, or -1 if not allowed */
213 } verbitem;
214
215 static const char verbnames[] =
216 "\0" /* Empty name is a shorthand for MARK */
217 STRING_MARK0
218 STRING_ACCEPT0
219 STRING_COMMIT0
220 STRING_F0
221 STRING_FAIL0
222 STRING_PRUNE0
223 STRING_SKIP0
224 STRING_THEN;
225
226 static const verbitem verbs[] = {
227 { 0, -1, OP_MARK },
228 { 4, -1, OP_MARK },
229 { 6, OP_ACCEPT, -1 },
230 { 6, OP_COMMIT, -1 },
231 { 1, OP_FAIL, -1 },
232 { 4, OP_FAIL, -1 },
233 { 5, OP_PRUNE, OP_PRUNE_ARG },
234 { 4, OP_SKIP, OP_SKIP_ARG },
235 { 4, OP_THEN, OP_THEN_ARG }
236 };
237
238 static const int verbcount = sizeof(verbs)/sizeof(verbitem);
239
240
241 /* Tables of names of POSIX character classes and their lengths. The names are
242 now all in a single string, to reduce the number of relocations when a shared
243 library is dynamically loaded. The list of lengths is terminated by a zero
244 length entry. The first three must be alpha, lower, upper, as this is assumed
245 for handling case independence. */
246
247 static const char posix_names[] =
248 STRING_alpha0 STRING_lower0 STRING_upper0 STRING_alnum0
249 STRING_ascii0 STRING_blank0 STRING_cntrl0 STRING_digit0
250 STRING_graph0 STRING_print0 STRING_punct0 STRING_space0
251 STRING_word0 STRING_xdigit;
252
253 static const pcre_uint8 posix_name_lengths[] = {
254 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 4, 6, 0 };
255
256 /* Table of class bit maps for each POSIX class. Each class is formed from a
257 base map, with an optional addition or removal of another map. Then, for some
258 classes, there is some additional tweaking: for [:blank:] the vertical space
259 characters are removed, and for [:alpha:] and [:alnum:] the underscore
260 character is removed. The triples in the table consist of the base map offset,
261 second map offset or -1 if no second map, and a non-negative value for map
262 addition or a negative value for map subtraction (if there are two maps). The
263 absolute value of the third field has these meanings: 0 => no tweaking, 1 =>
264 remove vertical space characters, 2 => remove underscore. */
265
266 static const int posix_class_maps[] = {
267 cbit_word, cbit_digit, -2, /* alpha */
268 cbit_lower, -1, 0, /* lower */
269 cbit_upper, -1, 0, /* upper */
270 cbit_word, -1, 2, /* alnum - word without underscore */
271 cbit_print, cbit_cntrl, 0, /* ascii */
272 cbit_space, -1, 1, /* blank - a GNU extension */
273 cbit_cntrl, -1, 0, /* cntrl */
274 cbit_digit, -1, 0, /* digit */
275 cbit_graph, -1, 0, /* graph */
276 cbit_print, -1, 0, /* print */
277 cbit_punct, -1, 0, /* punct */
278 cbit_space, -1, 0, /* space */
279 cbit_word, -1, 0, /* word - a Perl extension */
280 cbit_xdigit,-1, 0 /* xdigit */
281 };
282
283 /* Table of substitutes for \d etc when PCRE_UCP is set. The POSIX class
284 substitutes must be in the order of the names, defined above, and there are
285 both positive and negative cases. NULL means no substitute. */
286
287 #ifdef SUPPORT_UCP
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_pNd[] = {
292 CHAR_BACKSLASH, CHAR_p, CHAR_LEFT_CURLY_BRACKET,
293 CHAR_N, CHAR_d, 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_pXsp[] = {
298 CHAR_BACKSLASH, CHAR_p, CHAR_LEFT_CURLY_BRACKET,
299 CHAR_X, CHAR_s, CHAR_p, 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 static const pcre_uchar string_pXwd[] = {
304 CHAR_BACKSLASH, CHAR_p, CHAR_LEFT_CURLY_BRACKET,
305 CHAR_X, CHAR_w, CHAR_d, CHAR_RIGHT_CURLY_BRACKET, '\0' };
306
307 static const pcre_uchar *substitutes[] = {
308 string_PNd, /* \D */
309 string_pNd, /* \d */
310 string_PXsp, /* \S */ /* NOTE: Xsp is Perl space */
311 string_pXsp, /* \s */
312 string_PXwd, /* \W */
313 string_pXwd /* \w */
314 };
315
316 static const pcre_uchar string_pL[] = {
317 CHAR_BACKSLASH, CHAR_p, CHAR_LEFT_CURLY_BRACKET,
318 CHAR_L, CHAR_RIGHT_CURLY_BRACKET, '\0' };
319 static const pcre_uchar string_pLl[] = {
320 CHAR_BACKSLASH, CHAR_p, CHAR_LEFT_CURLY_BRACKET,
321 CHAR_L, CHAR_l, CHAR_RIGHT_CURLY_BRACKET, '\0' };
322 static const pcre_uchar string_pLu[] = {
323 CHAR_BACKSLASH, CHAR_p, CHAR_LEFT_CURLY_BRACKET,
324 CHAR_L, CHAR_u, CHAR_RIGHT_CURLY_BRACKET, '\0' };
325 static const pcre_uchar string_pXan[] = {
326 CHAR_BACKSLASH, CHAR_p, CHAR_LEFT_CURLY_BRACKET,
327 CHAR_X, CHAR_a, CHAR_n, CHAR_RIGHT_CURLY_BRACKET, '\0' };
328 static const pcre_uchar string_h[] = {
329 CHAR_BACKSLASH, CHAR_h, '\0' };
330 static const pcre_uchar string_pXps[] = {
331 CHAR_BACKSLASH, CHAR_p, CHAR_LEFT_CURLY_BRACKET,
332 CHAR_X, CHAR_p, CHAR_s, CHAR_RIGHT_CURLY_BRACKET, '\0' };
333 static const pcre_uchar string_PL[] = {
334 CHAR_BACKSLASH, CHAR_P, CHAR_LEFT_CURLY_BRACKET,
335 CHAR_L, CHAR_RIGHT_CURLY_BRACKET, '\0' };
336 static const pcre_uchar string_PLl[] = {
337 CHAR_BACKSLASH, CHAR_P, CHAR_LEFT_CURLY_BRACKET,
338 CHAR_L, CHAR_l, CHAR_RIGHT_CURLY_BRACKET, '\0' };
339 static const pcre_uchar string_PLu[] = {
340 CHAR_BACKSLASH, CHAR_P, CHAR_LEFT_CURLY_BRACKET,
341 CHAR_L, CHAR_u, CHAR_RIGHT_CURLY_BRACKET, '\0' };
342 static const pcre_uchar string_PXan[] = {
343 CHAR_BACKSLASH, CHAR_P, CHAR_LEFT_CURLY_BRACKET,
344 CHAR_X, CHAR_a, CHAR_n, CHAR_RIGHT_CURLY_BRACKET, '\0' };
345 static const pcre_uchar string_H[] = {
346 CHAR_BACKSLASH, CHAR_H, '\0' };
347 static const pcre_uchar string_PXps[] = {
348 CHAR_BACKSLASH, CHAR_P, CHAR_LEFT_CURLY_BRACKET,
349 CHAR_X, CHAR_p, CHAR_s, CHAR_RIGHT_CURLY_BRACKET, '\0' };
350
351 static const pcre_uchar *posix_substitutes[] = {
352 string_pL, /* alpha */
353 string_pLl, /* lower */
354 string_pLu, /* upper */
355 string_pXan, /* alnum */
356 NULL, /* ascii */
357 string_h, /* blank */
358 NULL, /* cntrl */
359 string_pNd, /* digit */
360 NULL, /* graph */
361 NULL, /* print */
362 NULL, /* punct */
363 string_pXps, /* space */ /* NOTE: Xps is POSIX space */
364 string_pXwd, /* word */
365 NULL, /* xdigit */
366 /* Negated cases */
367 string_PL, /* ^alpha */
368 string_PLl, /* ^lower */
369 string_PLu, /* ^upper */
370 string_PXan, /* ^alnum */
371 NULL, /* ^ascii */
372 string_H, /* ^blank */
373 NULL, /* ^cntrl */
374 string_PNd, /* ^digit */
375 NULL, /* ^graph */
376 NULL, /* ^print */
377 NULL, /* ^punct */
378 string_PXps, /* ^space */ /* NOTE: Xps is POSIX space */
379 string_PXwd, /* ^word */
380 NULL /* ^xdigit */
381 };
382 #define POSIX_SUBSIZE (sizeof(posix_substitutes) / sizeof(pcre_uchar *))
383 #endif
384
385 #define STRING(a) # a
386 #define XSTRING(s) STRING(s)
387
388 /* The texts of compile-time error messages. These are "char *" because they
389 are passed to the outside world. Do not ever re-use any error number, because
390 they are documented. Always add a new error instead. Messages marked DEAD below
391 are no longer used. This used to be a table of strings, but in order to reduce
392 the number of relocations needed when a shared library is loaded dynamically,
393 it is now one long string. We cannot use a table of offsets, because the
394 lengths of inserts such as XSTRING(MAX_NAME_SIZE) are not known. Instead, we
395 simply count through to the one we want - this isn't a performance issue
396 because these strings are used only when there is a compilation error.
397
398 Each substring ends with \0 to insert a null character. This includes the final
399 substring, so that the whole string ends with \0\0, which can be detected when
400 counting through. */
401
402 static const char error_texts[] =
403 "no error\0"
404 "\\ at end of pattern\0"
405 "\\c at end of pattern\0"
406 "unrecognized character follows \\\0"
407 "numbers out of order in {} quantifier\0"
408 /* 5 */
409 "number too big in {} quantifier\0"
410 "missing terminating ] for character class\0"
411 "invalid escape sequence in character class\0"
412 "range out of order in character class\0"
413 "nothing to repeat\0"
414 /* 10 */
415 "operand of unlimited repeat could match the empty string\0" /** DEAD **/
416 "internal error: unexpected repeat\0"
417 "unrecognized character after (? or (?-\0"
418 "POSIX named classes are supported only within a class\0"
419 "missing )\0"
420 /* 15 */
421 "reference to non-existent subpattern\0"
422 "erroffset passed as NULL\0"
423 "unknown option bit(s) set\0"
424 "missing ) after comment\0"
425 "parentheses nested too deeply\0" /** DEAD **/
426 /* 20 */
427 "regular expression is too large\0"
428 "failed to get memory\0"
429 "unmatched parentheses\0"
430 "internal error: code overflow\0"
431 "unrecognized character after (?<\0"
432 /* 25 */
433 "lookbehind assertion is not fixed length\0"
434 "malformed number or name after (?(\0"
435 "conditional group contains more than two branches\0"
436 "assertion expected after (?(\0"
437 "(?R or (?[+-]digits must be followed by )\0"
438 /* 30 */
439 "unknown POSIX class name\0"
440 "POSIX collating elements are not supported\0"
441 "this version of PCRE is compiled without UTF support\0"
442 "spare error\0" /** DEAD **/
443 "character value in \\x{...} sequence is too large\0"
444 /* 35 */
445 "invalid condition (?(0)\0"
446 "\\C not allowed in lookbehind assertion\0"
447 "PCRE does not support \\L, \\l, \\N{name}, \\U, or \\u\0"
448 "number after (?C is > 255\0"
449 "closing ) for (?C expected\0"
450 /* 40 */
451 "recursive call could loop indefinitely\0"
452 "unrecognized character after (?P\0"
453 "syntax error in subpattern name (missing terminator)\0"
454 "two named subpatterns have the same name\0"
455 "invalid UTF-8 string\0"
456 /* 45 */
457 "support for \\P, \\p, and \\X has not been compiled\0"
458 "malformed \\P or \\p sequence\0"
459 "unknown property name after \\P or \\p\0"
460 "subpattern name is too long (maximum " XSTRING(MAX_NAME_SIZE) " characters)\0"
461 "too many named subpatterns (maximum " XSTRING(MAX_NAME_COUNT) ")\0"
462 /* 50 */
463 "repeated subpattern is too long\0" /** DEAD **/
464 "octal value is greater than \\377 in 8-bit non-UTF-8 mode\0"
465 "internal error: overran compiling workspace\0"
466 "internal error: previously-checked referenced subpattern not found\0"
467 "DEFINE group contains more than one branch\0"
468 /* 55 */
469 "repeating a DEFINE group is not allowed\0" /** DEAD **/
470 "inconsistent NEWLINE options\0"
471 "\\g is not followed by a braced, angle-bracketed, or quoted name/number or by a plain number\0"
472 "a numbered reference must not be zero\0"
473 "an argument is not allowed for (*ACCEPT), (*FAIL), or (*COMMIT)\0"
474 /* 60 */
475 "(*VERB) not recognized\0"
476 "number is too big\0"
477 "subpattern name expected\0"
478 "digit expected after (?+\0"
479 "] is an invalid data character in JavaScript compatibility mode\0"
480 /* 65 */
481 "different names for subpatterns of the same number are not allowed\0"
482 "(*MARK) must have an argument\0"
483 "this version of PCRE is not compiled with Unicode property support\0"
484 "\\c must be followed by an ASCII character\0"
485 "\\k is not followed by a braced, angle-bracketed, or quoted name\0"
486 /* 70 */
487 "internal error: unknown opcode in find_fixedlength()\0"
488 "\\N is not supported in a class\0"
489 "too many forward references\0"
490 "disallowed Unicode code point (>= 0xd800 && <= 0xdfff)\0"
491 "invalid UTF-16 string\0"
492 /* 75 */
493 "name is too long in (*MARK), (*PRUNE), (*SKIP), or (*THEN)\0"
494 "character value in \\u.... sequence is too large\0"
495 ;
496
497 /* Table to identify digits and hex digits. This is used when compiling
498 patterns. Note that the tables in chartables are dependent on the locale, and
499 may mark arbitrary characters as digits - but the PCRE compiling code expects
500 to handle only 0-9, a-z, and A-Z as digits when compiling. That is why we have
501 a private table here. It costs 256 bytes, but it is a lot faster than doing
502 character value tests (at least in some simple cases I timed), and in some
503 applications one wants PCRE to compile efficiently as well as match
504 efficiently.
505
506 For convenience, we use the same bit definitions as in chartables:
507
508 0x04 decimal digit
509 0x08 hexadecimal digit
510
511 Then we can use ctype_digit and ctype_xdigit in the code. */
512
513 /* Using a simple comparison for decimal numbers rather than a memory read
514 is much faster, and the resulting code is simpler (the compiler turns it
515 into a subtraction and unsigned comparison). */
516
517 #define IS_DIGIT(x) ((x) >= CHAR_0 && (x) <= CHAR_9)
518
519 #ifndef EBCDIC
520
521 /* This is the "normal" case, for ASCII systems, and EBCDIC systems running in
522 UTF-8 mode. */
523
524 static const pcre_uint8 digitab[] =
525 {
526 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 0- 7 */
527 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 8- 15 */
528 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 16- 23 */
529 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 24- 31 */
530 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - ' */
531 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* ( - / */
532 0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c, /* 0 - 7 */
533 0x0c,0x0c,0x00,0x00,0x00,0x00,0x00,0x00, /* 8 - ? */
534 0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* @ - G */
535 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* H - O */
536 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* P - W */
537 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* X - _ */
538 0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* ` - g */
539 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* h - o */
540 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* p - w */
541 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* x -127 */
542 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 128-135 */
543 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 136-143 */
544 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 144-151 */
545 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 152-159 */
546 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 160-167 */
547 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 168-175 */
548 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 176-183 */
549 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 184-191 */
550 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 192-199 */
551 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 200-207 */
552 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 208-215 */
553 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 216-223 */
554 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 224-231 */
555 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 232-239 */
556 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 240-247 */
557 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00};/* 248-255 */
558
559 #else
560
561 /* This is the "abnormal" case, for EBCDIC systems not running in UTF-8 mode. */
562
563 static const pcre_uint8 digitab[] =
564 {
565 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 0- 7 0 */
566 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 8- 15 */
567 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 16- 23 10 */
568 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 24- 31 */
569 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 32- 39 20 */
570 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 40- 47 */
571 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 48- 55 30 */
572 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 56- 63 */
573 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - 71 40 */
574 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 72- | */
575 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* & - 87 50 */
576 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 88- 95 */
577 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - -103 60 */
578 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 104- ? */
579 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 112-119 70 */
580 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 120- " */
581 0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* 128- g 80 */
582 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* h -143 */
583 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 144- p 90 */
584 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* q -159 */
585 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 160- x A0 */
586 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* y -175 */
587 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* ^ -183 B0 */
588 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 184-191 */
589 0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* { - G C0 */
590 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* H -207 */
591 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* } - P D0 */
592 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* Q -223 */
593 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* \ - X E0 */
594 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* Y -239 */
595 0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c, /* 0 - 7 F0 */
596 0x0c,0x0c,0x00,0x00,0x00,0x00,0x00,0x00};/* 8 -255 */
597
598 static const pcre_uint8 ebcdic_chartab[] = { /* chartable partial dup */
599 0x80,0x00,0x00,0x00,0x00,0x01,0x00,0x00, /* 0- 7 */
600 0x00,0x00,0x00,0x00,0x01,0x01,0x00,0x00, /* 8- 15 */
601 0x00,0x00,0x00,0x00,0x00,0x01,0x00,0x00, /* 16- 23 */
602 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 24- 31 */
603 0x00,0x00,0x00,0x00,0x00,0x01,0x00,0x00, /* 32- 39 */
604 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 40- 47 */
605 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 48- 55 */
606 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 56- 63 */
607 0x01,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - 71 */
608 0x00,0x00,0x00,0x80,0x00,0x80,0x80,0x80, /* 72- | */
609 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* & - 87 */
610 0x00,0x00,0x00,0x80,0x80,0x80,0x00,0x00, /* 88- 95 */
611 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - -103 */
612 0x00,0x00,0x00,0x00,0x00,0x10,0x00,0x80, /* 104- ? */
613 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 112-119 */
614 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 120- " */
615 0x00,0x1a,0x1a,0x1a,0x1a,0x1a,0x1a,0x12, /* 128- g */
616 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* h -143 */
617 0x00,0x12,0x12,0x12,0x12,0x12,0x12,0x12, /* 144- p */
618 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* q -159 */
619 0x00,0x00,0x12,0x12,0x12,0x12,0x12,0x12, /* 160- x */
620 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* y -175 */
621 0x80,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* ^ -183 */
622 0x00,0x00,0x80,0x00,0x00,0x00,0x00,0x00, /* 184-191 */
623 0x80,0x1a,0x1a,0x1a,0x1a,0x1a,0x1a,0x12, /* { - G */
624 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* H -207 */
625 0x00,0x12,0x12,0x12,0x12,0x12,0x12,0x12, /* } - P */
626 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* Q -223 */
627 0x00,0x00,0x12,0x12,0x12,0x12,0x12,0x12, /* \ - X */
628 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* Y -239 */
629 0x1c,0x1c,0x1c,0x1c,0x1c,0x1c,0x1c,0x1c, /* 0 - 7 */
630 0x1c,0x1c,0x00,0x00,0x00,0x00,0x00,0x00};/* 8 -255 */
631 #endif
632
633
634 /* Definition to allow mutual recursion */
635
636 static BOOL
637 compile_regex(int, pcre_uchar **, const pcre_uchar **, int *, BOOL, BOOL, int, int,
638 int *, int *, branch_chain *, compile_data *, int *);
639
640
641
642 /*************************************************
643 * Find an error text *
644 *************************************************/
645
646 /* The error texts are now all in one long string, to save on relocations. As
647 some of the text is of unknown length, we can't use a table of offsets.
648 Instead, just count through the strings. This is not a performance issue
649 because it happens only when there has been a compilation error.
650
651 Argument: the error number
652 Returns: pointer to the error string
653 */
654
655 static const char *
656 find_error_text(int n)
657 {
658 const char *s = error_texts;
659 for (; n > 0; n--)
660 {
661 while (*s++ != 0) {};
662 if (*s == 0) return "Error text not found (please report)";
663 }
664 return s;
665 }
666
667
668 /*************************************************
669 * Expand the workspace *
670 *************************************************/
671
672 /* This function is called during the second compiling phase, if the number of
673 forward references fills the existing workspace, which is originally a block on
674 the stack. A larger block is obtained from malloc() unless the ultimate limit
675 has been reached or the increase will be rather small.
676
677 Argument: pointer to the compile data block
678 Returns: 0 if all went well, else an error number
679 */
680
681 static int
682 expand_workspace(compile_data *cd)
683 {
684 pcre_uchar *newspace;
685 int newsize = cd->workspace_size * 2;
686
687 if (newsize > COMPILE_WORK_SIZE_MAX) newsize = COMPILE_WORK_SIZE_MAX;
688 if (cd->workspace_size >= COMPILE_WORK_SIZE_MAX ||
689 newsize - cd->workspace_size < WORK_SIZE_SAFETY_MARGIN)
690 return ERR72;
691
692 newspace = (PUBL(malloc))(IN_UCHARS(newsize));
693 if (newspace == NULL) return ERR21;
694 memcpy(newspace, cd->start_workspace, cd->workspace_size * sizeof(pcre_uchar));
695 cd->hwm = (pcre_uchar *)newspace + (cd->hwm - cd->start_workspace);
696 if (cd->workspace_size > COMPILE_WORK_SIZE)
697 (PUBL(free))((void *)cd->start_workspace);
698 cd->start_workspace = newspace;
699 cd->workspace_size = newsize;
700 return 0;
701 }
702
703
704
705 /*************************************************
706 * Check for counted repeat *
707 *************************************************/
708
709 /* This function is called when a '{' is encountered in a place where it might
710 start a quantifier. It looks ahead to see if it really is a quantifier or not.
711 It is only a quantifier if it is one of the forms {ddd} {ddd,} or {ddd,ddd}
712 where the ddds are digits.
713
714 Arguments:
715 p pointer to the first char after '{'
716
717 Returns: TRUE or FALSE
718 */
719
720 static BOOL
721 is_counted_repeat(const pcre_uchar *p)
722 {
723 if (!IS_DIGIT(*p)) return FALSE;
724 p++;
725 while (IS_DIGIT(*p)) p++;
726 if (*p == CHAR_RIGHT_CURLY_BRACKET) return TRUE;
727
728 if (*p++ != CHAR_COMMA) return FALSE;
729 if (*p == CHAR_RIGHT_CURLY_BRACKET) return TRUE;
730
731 if (!IS_DIGIT(*p)) return FALSE;
732 p++;
733 while (IS_DIGIT(*p)) p++;
734
735 return (*p == CHAR_RIGHT_CURLY_BRACKET);
736 }
737
738
739
740 /*************************************************
741 * Handle escapes *
742 *************************************************/
743
744 /* This function is called when a \ has been encountered. It either returns a
745 positive value for a simple escape such as \n, or a negative value which
746 encodes one of the more complicated things such as \d. A backreference to group
747 n is returned as -(ESC_REF + n); ESC_REF is the highest ESC_xxx macro. When
748 UTF-8 is enabled, a positive value greater than 255 may be returned. On entry,
749 ptr is pointing at the \. On exit, it is on the final character of the escape
750 sequence.
751
752 Arguments:
753 ptrptr points to the pattern position pointer
754 errorcodeptr points to the errorcode variable
755 bracount number of previous extracting brackets
756 options the options bits
757 isclass TRUE if inside a character class
758
759 Returns: zero or positive => a data character
760 negative => a special escape sequence
761 on error, errorcodeptr is set
762 */
763
764 static int
765 check_escape(const pcre_uchar **ptrptr, int *errorcodeptr, int bracount,
766 int options, BOOL isclass)
767 {
768 /* PCRE_UTF16 has the same value as PCRE_UTF8. */
769 BOOL utf = (options & PCRE_UTF8) != 0;
770 const pcre_uchar *ptr = *ptrptr + 1;
771 pcre_int32 c;
772 int i;
773
774 GETCHARINCTEST(c, ptr); /* Get character value, increment pointer */
775 ptr--; /* Set pointer back to the last byte */
776
777 /* If backslash is at the end of the pattern, it's an error. */
778
779 if (c == 0) *errorcodeptr = ERR1;
780
781 /* Non-alphanumerics are literals. For digits or letters, do an initial lookup
782 in a table. A non-zero result is something that can be returned immediately.
783 Otherwise further processing may be required. */
784
785 #ifndef EBCDIC /* ASCII/UTF-8 coding */
786 /* Not alphanumeric */
787 else if (c < CHAR_0 || c > CHAR_z) {}
788 else if ((i = escapes[c - CHAR_0]) != 0) c = i;
789
790 #else /* EBCDIC coding */
791 /* Not alphanumeric */
792 else if (c < 'a' || (!MAX_255(c) || (ebcdic_chartab[c] & 0x0E) == 0)) {}
793 else if ((i = escapes[c - 0x48]) != 0) c = i;
794 #endif
795
796 /* Escapes that need further processing, or are illegal. */
797
798 else
799 {
800 const pcre_uchar *oldptr;
801 BOOL braced, negated;
802
803 switch (c)
804 {
805 /* A number of Perl escapes are not handled by PCRE. We give an explicit
806 error. */
807
808 case CHAR_l:
809 case CHAR_L:
810 *errorcodeptr = ERR37;
811 break;
812
813 case CHAR_u:
814 if ((options & PCRE_JAVASCRIPT_COMPAT) != 0)
815 {
816 /* In JavaScript, \u must be followed by four hexadecimal numbers.
817 Otherwise it is a lowercase u letter. */
818 if (MAX_255(ptr[1]) && (digitab[ptr[1]] & ctype_xdigit) != 0
819 && MAX_255(ptr[2]) && (digitab[ptr[2]] & ctype_xdigit) != 0
820 && MAX_255(ptr[3]) && (digitab[ptr[3]] & ctype_xdigit) != 0
821 && MAX_255(ptr[4]) && (digitab[ptr[4]] & ctype_xdigit) != 0)
822 {
823 c = 0;
824 for (i = 0; i < 4; ++i)
825 {
826 register int cc = *(++ptr);
827 #ifndef EBCDIC /* ASCII/UTF-8 coding */
828 if (cc >= CHAR_a) cc -= 32; /* Convert to upper case */
829 c = (c << 4) + cc - ((cc < CHAR_A)? CHAR_0 : (CHAR_A - 10));
830 #else /* EBCDIC coding */
831 if (cc >= CHAR_a && cc <= CHAR_z) cc += 64; /* Convert to upper case */
832 c = (c << 4) + cc - ((cc >= CHAR_0)? CHAR_0 : (CHAR_A - 10));
833 #endif
834 }
835
836 #ifdef COMPILE_PCRE8
837 if (c > (utf ? 0x10ffff : 0xff))
838 #else
839 #ifdef COMPILE_PCRE16
840 if (c > (utf ? 0x10ffff : 0xffff))
841 #endif
842 #endif
843 {
844 *errorcodeptr = ERR76;
845 }
846 else if (utf && c >= 0xd800 && c <= 0xdfff) *errorcodeptr = ERR73;
847 }
848 }
849 else
850 *errorcodeptr = ERR37;
851 break;
852
853 case CHAR_U:
854 /* In JavaScript, \U is an uppercase U letter. */
855 if ((options & PCRE_JAVASCRIPT_COMPAT) == 0) *errorcodeptr = ERR37;
856 break;
857
858 /* In a character class, \g is just a literal "g". Outside a character
859 class, \g must be followed by one of a number of specific things:
860
861 (1) A number, either plain or braced. If positive, it is an absolute
862 backreference. If negative, it is a relative backreference. This is a Perl
863 5.10 feature.
864
865 (2) Perl 5.10 also supports \g{name} as a reference to a named group. This
866 is part of Perl's movement towards a unified syntax for back references. As
867 this is synonymous with \k{name}, we fudge it up by pretending it really
868 was \k.
869
870 (3) For Oniguruma compatibility we also support \g followed by a name or a
871 number either in angle brackets or in single quotes. However, these are
872 (possibly recursive) subroutine calls, _not_ backreferences. Just return
873 the -ESC_g code (cf \k). */
874
875 case CHAR_g:
876 if (isclass) break;
877 if (ptr[1] == CHAR_LESS_THAN_SIGN || ptr[1] == CHAR_APOSTROPHE)
878 {
879 c = -ESC_g;
880 break;
881 }
882
883 /* Handle the Perl-compatible cases */
884
885 if (ptr[1] == CHAR_LEFT_CURLY_BRACKET)
886 {
887 const pcre_uchar *p;
888 for (p = ptr+2; *p != 0 && *p != CHAR_RIGHT_CURLY_BRACKET; p++)
889 if (*p != CHAR_MINUS && !IS_DIGIT(*p)) break;
890 if (*p != 0 && *p != CHAR_RIGHT_CURLY_BRACKET)
891 {
892 c = -ESC_k;
893 break;
894 }
895 braced = TRUE;
896 ptr++;
897 }
898 else braced = FALSE;
899
900 if (ptr[1] == CHAR_MINUS)
901 {
902 negated = TRUE;
903 ptr++;
904 }
905 else negated = FALSE;
906
907 /* The integer range is limited by the machine's int representation. */
908 c = 0;
909 while (IS_DIGIT(ptr[1]))
910 {
911 if (((unsigned int)c) > INT_MAX / 10) /* Integer overflow */
912 {
913 c = -1;
914 break;
915 }
916 c = c * 10 + *(++ptr) - CHAR_0;
917 }
918 if (((unsigned int)c) > INT_MAX) /* Integer overflow */
919 {
920 while (IS_DIGIT(ptr[1]))
921 ptr++;
922 *errorcodeptr = ERR61;
923 break;
924 }
925
926 if (braced && *(++ptr) != CHAR_RIGHT_CURLY_BRACKET)
927 {
928 *errorcodeptr = ERR57;
929 break;
930 }
931
932 if (c == 0)
933 {
934 *errorcodeptr = ERR58;
935 break;
936 }
937
938 if (negated)
939 {
940 if (c > bracount)
941 {
942 *errorcodeptr = ERR15;
943 break;
944 }
945 c = bracount - (c - 1);
946 }
947
948 c = -(ESC_REF + c);
949 break;
950
951 /* The handling of escape sequences consisting of a string of digits
952 starting with one that is not zero is not straightforward. By experiment,
953 the way Perl works seems to be as follows:
954
955 Outside a character class, the digits are read as a decimal number. If the
956 number is less than 10, or if there are that many previous extracting
957 left brackets, then it is a back reference. Otherwise, up to three octal
958 digits are read to form an escaped byte. Thus \123 is likely to be octal
959 123 (cf \0123, which is octal 012 followed by the literal 3). If the octal
960 value is greater than 377, the least significant 8 bits are taken. Inside a
961 character class, \ followed by a digit is always an octal number. */
962
963 case CHAR_1: case CHAR_2: case CHAR_3: case CHAR_4: case CHAR_5:
964 case CHAR_6: case CHAR_7: case CHAR_8: case CHAR_9:
965
966 if (!isclass)
967 {
968 oldptr = ptr;
969 /* The integer range is limited by the machine's int representation. */
970 c -= CHAR_0;
971 while (IS_DIGIT(ptr[1]))
972 {
973 if (((unsigned int)c) > INT_MAX / 10) /* Integer overflow */
974 {
975 c = -1;
976 break;
977 }
978 c = c * 10 + *(++ptr) - CHAR_0;
979 }
980 if (((unsigned int)c) > INT_MAX) /* Integer overflow */
981 {
982 while (IS_DIGIT(ptr[1]))
983 ptr++;
984 *errorcodeptr = ERR61;
985 break;
986 }
987 if (c < 10 || c <= bracount)
988 {
989 c = -(ESC_REF + c);
990 break;
991 }
992 ptr = oldptr; /* Put the pointer back and fall through */
993 }
994
995 /* Handle an octal number following \. If the first digit is 8 or 9, Perl
996 generates a binary zero byte and treats the digit as a following literal.
997 Thus we have to pull back the pointer by one. */
998
999 if ((c = *ptr) >= CHAR_8)
1000 {
1001 ptr--;
1002 c = 0;
1003 break;
1004 }
1005
1006 /* \0 always starts an octal number, but we may drop through to here with a
1007 larger first octal digit. The original code used just to take the least
1008 significant 8 bits of octal numbers (I think this is what early Perls used
1009 to do). Nowadays we allow for larger numbers in UTF-8 mode and 16-bit mode,
1010 but no more than 3 octal digits. */
1011
1012 case CHAR_0:
1013 c -= CHAR_0;
1014 while(i++ < 2 && ptr[1] >= CHAR_0 && ptr[1] <= CHAR_7)
1015 c = c * 8 + *(++ptr) - CHAR_0;
1016 #ifdef COMPILE_PCRE8
1017 if (!utf && c > 0xff) *errorcodeptr = ERR51;
1018 #endif
1019 break;
1020
1021 /* \x is complicated. \x{ddd} is a character number which can be greater
1022 than 0xff in utf or non-8bit mode, but only if the ddd are hex digits.
1023 If not, { is treated as a data character. */
1024
1025 case CHAR_x:
1026 if ((options & PCRE_JAVASCRIPT_COMPAT) != 0)
1027 {
1028 /* In JavaScript, \x must be followed by two hexadecimal numbers.
1029 Otherwise it is a lowercase x letter. */
1030 if (MAX_255(ptr[1]) && (digitab[ptr[1]] & ctype_xdigit) != 0
1031 && MAX_255(ptr[2]) && (digitab[ptr[2]] & ctype_xdigit) != 0)
1032 {
1033 c = 0;
1034 for (i = 0; i < 2; ++i)
1035 {
1036 register int cc = *(++ptr);
1037 #ifndef EBCDIC /* ASCII/UTF-8 coding */
1038 if (cc >= CHAR_a) cc -= 32; /* Convert to upper case */
1039 c = (c << 4) + cc - ((cc < CHAR_A)? CHAR_0 : (CHAR_A - 10));
1040 #else /* EBCDIC coding */
1041 if (cc >= CHAR_a && cc <= CHAR_z) cc += 64; /* Convert to upper case */
1042 c = (c << 4) + cc - ((cc >= CHAR_0)? CHAR_0 : (CHAR_A - 10));
1043 #endif
1044 }
1045 }
1046 break;
1047 }
1048
1049 if (ptr[1] == CHAR_LEFT_CURLY_BRACKET)
1050 {
1051 const pcre_uchar *pt = ptr + 2;
1052
1053 c = 0;
1054 while (MAX_255(*pt) && (digitab[*pt] & ctype_xdigit) != 0)
1055 {
1056 register int cc = *pt++;
1057 if (c == 0 && cc == CHAR_0) continue; /* Leading zeroes */
1058
1059 #ifndef EBCDIC /* ASCII/UTF-8 coding */
1060 if (cc >= CHAR_a) cc -= 32; /* Convert to upper case */
1061 c = (c << 4) + cc - ((cc < CHAR_A)? CHAR_0 : (CHAR_A - 10));
1062 #else /* EBCDIC coding */
1063 if (cc >= CHAR_a && cc <= CHAR_z) cc += 64; /* Convert to upper case */
1064 c = (c << 4) + cc - ((cc >= CHAR_0)? CHAR_0 : (CHAR_A - 10));
1065 #endif
1066
1067 #ifdef COMPILE_PCRE8
1068 if (c > (utf ? 0x10ffff : 0xff)) { c = -1; break; }
1069 #else
1070 #ifdef COMPILE_PCRE16
1071 if (c > (utf ? 0x10ffff : 0xffff)) { c = -1; break; }
1072 #endif
1073 #endif
1074 }
1075
1076 if (c < 0)
1077 {
1078 while (MAX_255(*pt) && (digitab[*pt] & ctype_xdigit) != 0) pt++;
1079 *errorcodeptr = ERR34;
1080 }
1081
1082 if (*pt == CHAR_RIGHT_CURLY_BRACKET)
1083 {
1084 if (utf && c >= 0xd800 && c <= 0xdfff) *errorcodeptr = ERR73;
1085 ptr = pt;
1086 break;
1087 }
1088
1089 /* If the sequence of hex digits does not end with '}', then we don't
1090 recognize this construct; fall through to the normal \x handling. */
1091 }
1092
1093 /* Read just a single-byte hex-defined char */
1094
1095 c = 0;
1096 while (i++ < 2 && MAX_255(ptr[1]) && (digitab[ptr[1]] & ctype_xdigit) != 0)
1097 {
1098 int cc; /* Some compilers don't like */
1099 cc = *(++ptr); /* ++ in initializers */
1100 #ifndef EBCDIC /* ASCII/UTF-8 coding */
1101 if (cc >= CHAR_a) cc -= 32; /* Convert to upper case */
1102 c = c * 16 + cc - ((cc < CHAR_A)? CHAR_0 : (CHAR_A - 10));
1103 #else /* EBCDIC coding */
1104 if (cc <= CHAR_z) cc += 64; /* Convert to upper case */
1105 c = c * 16 + cc - ((cc >= CHAR_0)? CHAR_0 : (CHAR_A - 10));
1106 #endif
1107 }
1108 break;
1109
1110 /* For \c, a following letter is upper-cased; then the 0x40 bit is flipped.
1111 An error is given if the byte following \c is not an ASCII character. This
1112 coding is ASCII-specific, but then the whole concept of \cx is
1113 ASCII-specific. (However, an EBCDIC equivalent has now been added.) */
1114
1115 case CHAR_c:
1116 c = *(++ptr);
1117 if (c == 0)
1118 {
1119 *errorcodeptr = ERR2;
1120 break;
1121 }
1122 #ifndef EBCDIC /* ASCII/UTF-8 coding */
1123 if (c > 127) /* Excludes all non-ASCII in either mode */
1124 {
1125 *errorcodeptr = ERR68;
1126 break;
1127 }
1128 if (c >= CHAR_a && c <= CHAR_z) c -= 32;
1129 c ^= 0x40;
1130 #else /* EBCDIC coding */
1131 if (c >= CHAR_a && c <= CHAR_z) c += 64;
1132 c ^= 0xC0;
1133 #endif
1134 break;
1135
1136 /* PCRE_EXTRA enables extensions to Perl in the matter of escapes. Any
1137 other alphanumeric following \ is an error if PCRE_EXTRA was set;
1138 otherwise, for Perl compatibility, it is a literal. This code looks a bit
1139 odd, but there used to be some cases other than the default, and there may
1140 be again in future, so I haven't "optimized" it. */
1141
1142 default:
1143 if ((options & PCRE_EXTRA) != 0) switch(c)
1144 {
1145 default:
1146 *errorcodeptr = ERR3;
1147 break;
1148 }
1149 break;
1150 }
1151 }
1152
1153 /* Perl supports \N{name} for character names, as well as plain \N for "not
1154 newline". PCRE does not support \N{name}. However, it does support
1155 quantification such as \N{2,3}. */
1156
1157 if (c == -ESC_N && ptr[1] == CHAR_LEFT_CURLY_BRACKET &&
1158 !is_counted_repeat(ptr+2))
1159 *errorcodeptr = ERR37;
1160
1161 /* If PCRE_UCP is set, we change the values for \d etc. */
1162
1163 if ((options & PCRE_UCP) != 0 && c <= -ESC_D && c >= -ESC_w)
1164 c -= (ESC_DU - ESC_D);
1165
1166 /* Set the pointer to the final character before returning. */
1167
1168 *ptrptr = ptr;
1169 return c;
1170 }
1171
1172
1173
1174 #ifdef SUPPORT_UCP
1175 /*************************************************
1176 * Handle \P and \p *
1177 *************************************************/
1178
1179 /* This function is called after \P or \p has been encountered, provided that
1180 PCRE is compiled with support for Unicode properties. On entry, ptrptr is
1181 pointing at the P or p. On exit, it is pointing at the final character of the
1182 escape sequence.
1183
1184 Argument:
1185 ptrptr points to the pattern position pointer
1186 negptr points to a boolean that is set TRUE for negation else FALSE
1187 dptr points to an int that is set to the detailed property value
1188 errorcodeptr points to the error code variable
1189
1190 Returns: type value from ucp_type_table, or -1 for an invalid type
1191 */
1192
1193 static int
1194 get_ucp(const pcre_uchar **ptrptr, BOOL *negptr, int *dptr, int *errorcodeptr)
1195 {
1196 int c, i, bot, top;
1197 const pcre_uchar *ptr = *ptrptr;
1198 pcre_uchar name[32];
1199
1200 c = *(++ptr);
1201 if (c == 0) goto ERROR_RETURN;
1202
1203 *negptr = FALSE;
1204
1205 /* \P or \p can be followed by a name in {}, optionally preceded by ^ for
1206 negation. */
1207
1208 if (c == CHAR_LEFT_CURLY_BRACKET)
1209 {
1210 if (ptr[1] == CHAR_CIRCUMFLEX_ACCENT)
1211 {
1212 *negptr = TRUE;
1213 ptr++;
1214 }
1215 for (i = 0; i < (int)(sizeof(name) / sizeof(pcre_uchar)) - 1; i++)
1216 {
1217 c = *(++ptr);
1218 if (c == 0) goto ERROR_RETURN;
1219 if (c == CHAR_RIGHT_CURLY_BRACKET) break;
1220 name[i] = c;
1221 }
1222 if (c != CHAR_RIGHT_CURLY_BRACKET) goto ERROR_RETURN;
1223 name[i] = 0;
1224 }
1225
1226 /* Otherwise there is just one following character */
1227
1228 else
1229 {
1230 name[0] = c;
1231 name[1] = 0;
1232 }
1233
1234 *ptrptr = ptr;
1235
1236 /* Search for a recognized property name using binary chop */
1237
1238 bot = 0;
1239 top = PRIV(utt_size);
1240
1241 while (bot < top)
1242 {
1243 i = (bot + top) >> 1;
1244 c = STRCMP_UC_C8(name, PRIV(utt_names) + PRIV(utt)[i].name_offset);
1245 if (c == 0)
1246 {
1247 *dptr = PRIV(utt)[i].value;
1248 return PRIV(utt)[i].type;
1249 }
1250 if (c > 0) bot = i + 1; else top = i;
1251 }
1252
1253 *errorcodeptr = ERR47;
1254 *ptrptr = ptr;
1255 return -1;
1256
1257 ERROR_RETURN:
1258 *errorcodeptr = ERR46;
1259 *ptrptr = ptr;
1260 return -1;
1261 }
1262 #endif
1263
1264
1265
1266
1267 /*************************************************
1268 * Read repeat counts *
1269 *************************************************/
1270
1271 /* Read an item of the form {n,m} and return the values. This is called only
1272 after is_counted_repeat() has confirmed that a repeat-count quantifier exists,
1273 so the syntax is guaranteed to be correct, but we need to check the values.
1274
1275 Arguments:
1276 p pointer to first char after '{'
1277 minp pointer to int for min
1278 maxp pointer to int for max
1279 returned as -1 if no max
1280 errorcodeptr points to error code variable
1281
1282 Returns: pointer to '}' on success;
1283 current ptr on error, with errorcodeptr set non-zero
1284 */
1285
1286 static const pcre_uchar *
1287 read_repeat_counts(const pcre_uchar *p, int *minp, int *maxp, int *errorcodeptr)
1288 {
1289 int min = 0;
1290 int max = -1;
1291
1292 /* Read the minimum value and do a paranoid check: a negative value indicates
1293 an integer overflow. */
1294
1295 while (IS_DIGIT(*p)) min = min * 10 + *p++ - CHAR_0;
1296 if (min < 0 || min > 65535)
1297 {
1298 *errorcodeptr = ERR5;
1299 return p;
1300 }
1301
1302 /* Read the maximum value if there is one, and again do a paranoid on its size.
1303 Also, max must not be less than min. */
1304
1305 if (*p == CHAR_RIGHT_CURLY_BRACKET) max = min; else
1306 {
1307 if (*(++p) != CHAR_RIGHT_CURLY_BRACKET)
1308 {
1309 max = 0;
1310 while(IS_DIGIT(*p)) max = max * 10 + *p++ - CHAR_0;
1311 if (max < 0 || max > 65535)
1312 {
1313 *errorcodeptr = ERR5;
1314 return p;
1315 }
1316 if (max < min)
1317 {
1318 *errorcodeptr = ERR4;
1319 return p;
1320 }
1321 }
1322 }
1323
1324 /* Fill in the required variables, and pass back the pointer to the terminating
1325 '}'. */
1326
1327 *minp = min;
1328 *maxp = max;
1329 return p;
1330 }
1331
1332
1333
1334 /*************************************************
1335 * Subroutine for finding forward reference *
1336 *************************************************/
1337
1338 /* This recursive function is called only from find_parens() below. The
1339 top-level call starts at the beginning of the pattern. All other calls must
1340 start at a parenthesis. It scans along a pattern's text looking for capturing
1341 subpatterns, and counting them. If it finds a named pattern that matches the
1342 name it is given, it returns its number. Alternatively, if the name is NULL, it
1343 returns when it reaches a given numbered subpattern. Recursion is used to keep
1344 track of subpatterns that reset the capturing group numbers - the (?| feature.
1345
1346 This function was originally called only from the second pass, in which we know
1347 that if (?< or (?' or (?P< is encountered, the name will be correctly
1348 terminated because that is checked in the first pass. There is now one call to
1349 this function in the first pass, to check for a recursive back reference by
1350 name (so that we can make the whole group atomic). In this case, we need check
1351 only up to the current position in the pattern, and that is still OK because
1352 and previous occurrences will have been checked. To make this work, the test
1353 for "end of pattern" is a check against cd->end_pattern in the main loop,
1354 instead of looking for a binary zero. This means that the special first-pass
1355 call can adjust cd->end_pattern temporarily. (Checks for binary zero while
1356 processing items within the loop are OK, because afterwards the main loop will
1357 terminate.)
1358
1359 Arguments:
1360 ptrptr address of the current character pointer (updated)
1361 cd compile background data
1362 name name to seek, or NULL if seeking a numbered subpattern
1363 lorn name length, or subpattern number if name is NULL
1364 xmode TRUE if we are in /x mode
1365 utf TRUE if we are in UTF-8 / UTF-16 mode
1366 count pointer to the current capturing subpattern number (updated)
1367
1368 Returns: the number of the named subpattern, or -1 if not found
1369 */
1370
1371 static int
1372 find_parens_sub(pcre_uchar **ptrptr, compile_data *cd, const pcre_uchar *name, int lorn,
1373 BOOL xmode, BOOL utf, int *count)
1374 {
1375 pcre_uchar *ptr = *ptrptr;
1376 int start_count = *count;
1377 int hwm_count = start_count;
1378 BOOL dup_parens = FALSE;
1379
1380 /* If the first character is a parenthesis, check on the type of group we are
1381 dealing with. The very first call may not start with a parenthesis. */
1382
1383 if (ptr[0] == CHAR_LEFT_PARENTHESIS)
1384 {
1385 /* Handle specials such as (*SKIP) or (*UTF8) etc. */
1386
1387 if (ptr[1] == CHAR_ASTERISK) ptr += 2;
1388
1389 /* Handle a normal, unnamed capturing parenthesis. */
1390
1391 else if (ptr[1] != CHAR_QUESTION_MARK)
1392 {
1393 *count += 1;
1394 if (name == NULL && *count == lorn) return *count;
1395 ptr++;
1396 }
1397
1398 /* All cases now have (? at the start. Remember when we are in a group
1399 where the parenthesis numbers are duplicated. */
1400
1401 else if (ptr[2] == CHAR_VERTICAL_LINE)
1402 {
1403 ptr += 3;
1404 dup_parens = TRUE;
1405 }
1406
1407 /* Handle comments; all characters are allowed until a ket is reached. */
1408
1409 else if (ptr[2] == CHAR_NUMBER_SIGN)
1410 {
1411 for (ptr += 3; *ptr != 0; ptr++) if (*ptr == CHAR_RIGHT_PARENTHESIS) break;
1412 goto FAIL_EXIT;
1413 }
1414
1415 /* Handle a condition. If it is an assertion, just carry on so that it
1416 is processed as normal. If not, skip to the closing parenthesis of the
1417 condition (there can't be any nested parens). */
1418
1419 else if (ptr[2] == CHAR_LEFT_PARENTHESIS)
1420 {
1421 ptr += 2;
1422 if (ptr[1] != CHAR_QUESTION_MARK)
1423 {
1424 while (*ptr != 0 && *ptr != CHAR_RIGHT_PARENTHESIS) ptr++;
1425 if (*ptr != 0) ptr++;
1426 }
1427 }
1428
1429 /* Start with (? but not a condition. */
1430
1431 else
1432 {
1433 ptr += 2;
1434 if (*ptr == CHAR_P) ptr++; /* Allow optional P */
1435
1436 /* We have to disambiguate (?<! and (?<= from (?<name> for named groups */
1437
1438 if ((*ptr == CHAR_LESS_THAN_SIGN && ptr[1] != CHAR_EXCLAMATION_MARK &&
1439 ptr[1] != CHAR_EQUALS_SIGN) || *ptr == CHAR_APOSTROPHE)
1440 {
1441 int term;
1442 const pcre_uchar *thisname;
1443 *count += 1;
1444 if (name == NULL && *count == lorn) return *count;
1445 term = *ptr++;
1446 if (term == CHAR_LESS_THAN_SIGN) term = CHAR_GREATER_THAN_SIGN;
1447 thisname = ptr;
1448 while (*ptr != term) ptr++;
1449 if (name != NULL && lorn == ptr - thisname &&
1450 STRNCMP_UC_UC(name, thisname, lorn) == 0)
1451 return *count;
1452 term++;
1453 }
1454 }
1455 }
1456
1457 /* Past any initial parenthesis handling, scan for parentheses or vertical
1458 bars. Stop if we get to cd->end_pattern. Note that this is important for the
1459 first-pass call when this value is temporarily adjusted to stop at the current
1460 position. So DO NOT change this to a test for binary zero. */
1461
1462 for (; ptr < cd->end_pattern; ptr++)
1463 {
1464 /* Skip over backslashed characters and also entire \Q...\E */
1465
1466 if (*ptr == CHAR_BACKSLASH)
1467 {
1468 if (*(++ptr) == 0) goto FAIL_EXIT;
1469 if (*ptr == CHAR_Q) for (;;)
1470 {
1471 while (*(++ptr) != 0 && *ptr != CHAR_BACKSLASH) {};
1472 if (*ptr == 0) goto FAIL_EXIT;
1473 if (*(++ptr) == CHAR_E) break;
1474 }
1475 continue;
1476 }
1477
1478 /* Skip over character classes; this logic must be similar to the way they
1479 are handled for real. If the first character is '^', skip it. Also, if the
1480 first few characters (either before or after ^) are \Q\E or \E we skip them
1481 too. This makes for compatibility with Perl. Note the use of STR macros to
1482 encode "Q\\E" so that it works in UTF-8 on EBCDIC platforms. */
1483
1484 if (*ptr == CHAR_LEFT_SQUARE_BRACKET)
1485 {
1486 BOOL negate_class = FALSE;
1487 for (;;)
1488 {
1489 if (ptr[1] == CHAR_BACKSLASH)
1490 {
1491 if (ptr[2] == CHAR_E)
1492 ptr+= 2;
1493 else if (STRNCMP_UC_C8(ptr + 2,
1494 STR_Q STR_BACKSLASH STR_E, 3) == 0)
1495 ptr += 4;
1496 else
1497 break;
1498 }
1499 else if (!negate_class && ptr[1] == CHAR_CIRCUMFLEX_ACCENT)
1500 {
1501 negate_class = TRUE;
1502 ptr++;
1503 }
1504 else break;
1505 }
1506
1507 /* If the next character is ']', it is a data character that must be
1508 skipped, except in JavaScript compatibility mode. */
1509
1510 if (ptr[1] == CHAR_RIGHT_SQUARE_BRACKET &&
1511 (cd->external_options & PCRE_JAVASCRIPT_COMPAT) == 0)
1512 ptr++;
1513
1514 while (*(++ptr) != CHAR_RIGHT_SQUARE_BRACKET)
1515 {
1516 if (*ptr == 0) return -1;
1517 if (*ptr == CHAR_BACKSLASH)
1518 {
1519 if (*(++ptr) == 0) goto FAIL_EXIT;
1520 if (*ptr == CHAR_Q) for (;;)
1521 {
1522 while (*(++ptr) != 0 && *ptr != CHAR_BACKSLASH) {};
1523 if (*ptr == 0) goto FAIL_EXIT;
1524 if (*(++ptr) == CHAR_E) break;
1525 }
1526 continue;
1527 }
1528 }
1529 continue;
1530 }
1531
1532 /* Skip comments in /x mode */
1533
1534 if (xmode && *ptr == CHAR_NUMBER_SIGN)
1535 {
1536 ptr++;
1537 while (*ptr != 0)
1538 {
1539 if (IS_NEWLINE(ptr)) { ptr += cd->nllen - 1; break; }
1540 ptr++;
1541 #ifdef SUPPORT_UTF
1542 if (utf) FORWARDCHAR(ptr);
1543 #endif
1544 }
1545 if (*ptr == 0) goto FAIL_EXIT;
1546 continue;
1547 }
1548
1549 /* Check for the special metacharacters */
1550
1551 if (*ptr == CHAR_LEFT_PARENTHESIS)
1552 {
1553 int rc = find_parens_sub(&ptr, cd, name, lorn, xmode, utf, count);
1554 if (rc > 0) return rc;
1555 if (*ptr == 0) goto FAIL_EXIT;
1556 }
1557
1558 else if (*ptr == CHAR_RIGHT_PARENTHESIS)
1559 {
1560 if (dup_parens && *count < hwm_count) *count = hwm_count;
1561 goto FAIL_EXIT;
1562 }
1563
1564 else if (*ptr == CHAR_VERTICAL_LINE && dup_parens)
1565 {
1566 if (*count > hwm_count) hwm_count = *count;
1567 *count = start_count;
1568 }
1569 }
1570
1571 FAIL_EXIT:
1572 *ptrptr = ptr;
1573 return -1;
1574 }
1575
1576
1577
1578
1579 /*************************************************
1580 * Find forward referenced subpattern *
1581 *************************************************/
1582
1583 /* This function scans along a pattern's text looking for capturing
1584 subpatterns, and counting them. If it finds a named pattern that matches the
1585 name it is given, it returns its number. Alternatively, if the name is NULL, it
1586 returns when it reaches a given numbered subpattern. This is used for forward
1587 references to subpatterns. We used to be able to start this scan from the
1588 current compiling point, using the current count value from cd->bracount, and
1589 do it all in a single loop, but the addition of the possibility of duplicate
1590 subpattern numbers means that we have to scan from the very start, in order to
1591 take account of such duplicates, and to use a recursive function to keep track
1592 of the different types of group.
1593
1594 Arguments:
1595 cd compile background data
1596 name name to seek, or NULL if seeking a numbered subpattern
1597 lorn name length, or subpattern number if name is NULL
1598 xmode TRUE if we are in /x mode
1599 utf TRUE if we are in UTF-8 / UTF-16 mode
1600
1601 Returns: the number of the found subpattern, or -1 if not found
1602 */
1603
1604 static int
1605 find_parens(compile_data *cd, const pcre_uchar *name, int lorn, BOOL xmode,
1606 BOOL utf)
1607 {
1608 pcre_uchar *ptr = (pcre_uchar *)cd->start_pattern;
1609 int count = 0;
1610 int rc;
1611
1612 /* If the pattern does not start with an opening parenthesis, the first call
1613 to find_parens_sub() will scan right to the end (if necessary). However, if it
1614 does start with a parenthesis, find_parens_sub() will return when it hits the
1615 matching closing parens. That is why we have to have a loop. */
1616
1617 for (;;)
1618 {
1619 rc = find_parens_sub(&ptr, cd, name, lorn, xmode, utf, &count);
1620 if (rc > 0 || *ptr++ == 0) break;
1621 }
1622
1623 return rc;
1624 }
1625
1626
1627
1628
1629 /*************************************************
1630 * Find first significant op code *
1631 *************************************************/
1632
1633 /* This is called by several functions that scan a compiled expression looking
1634 for a fixed first character, or an anchoring op code etc. It skips over things
1635 that do not influence this. For some calls, it makes sense to skip negative
1636 forward and all backward assertions, and also the \b assertion; for others it
1637 does not.
1638
1639 Arguments:
1640 code pointer to the start of the group
1641 skipassert TRUE if certain assertions are to be skipped
1642
1643 Returns: pointer to the first significant opcode
1644 */
1645
1646 static const pcre_uchar*
1647 first_significant_code(const pcre_uchar *code, BOOL skipassert)
1648 {
1649 for (;;)
1650 {
1651 switch ((int)*code)
1652 {
1653 case OP_ASSERT_NOT:
1654 case OP_ASSERTBACK:
1655 case OP_ASSERTBACK_NOT:
1656 if (!skipassert) return code;
1657 do code += GET(code, 1); while (*code == OP_ALT);
1658 code += PRIV(OP_lengths)[*code];
1659 break;
1660
1661 case OP_WORD_BOUNDARY:
1662 case OP_NOT_WORD_BOUNDARY:
1663 if (!skipassert) return code;
1664 /* Fall through */
1665
1666 case OP_CALLOUT:
1667 case OP_CREF:
1668 case OP_NCREF:
1669 case OP_RREF:
1670 case OP_NRREF:
1671 case OP_DEF:
1672 code += PRIV(OP_lengths)[*code];
1673 break;
1674
1675 default:
1676 return code;
1677 }
1678 }
1679 /* Control never reaches here */
1680 }
1681
1682
1683
1684
1685 /*************************************************
1686 * Find the fixed length of a branch *
1687 *************************************************/
1688
1689 /* Scan a branch and compute the fixed length of subject that will match it,
1690 if the length is fixed. This is needed for dealing with backward assertions.
1691 In UTF8 mode, the result is in characters rather than bytes. The branch is
1692 temporarily terminated with OP_END when this function is called.
1693
1694 This function is called when a backward assertion is encountered, so that if it
1695 fails, the error message can point to the correct place in the pattern.
1696 However, we cannot do this when the assertion contains subroutine calls,
1697 because they can be forward references. We solve this by remembering this case
1698 and doing the check at the end; a flag specifies which mode we are running in.
1699
1700 Arguments:
1701 code points to the start of the pattern (the bracket)
1702 utf TRUE in UTF-8 / UTF-16 mode
1703 atend TRUE if called when the pattern is complete
1704 cd the "compile data" structure
1705
1706 Returns: the fixed length,
1707 or -1 if there is no fixed length,
1708 or -2 if \C was encountered (in UTF-8 mode only)
1709 or -3 if an OP_RECURSE item was encountered and atend is FALSE
1710 or -4 if an unknown opcode was encountered (internal error)
1711 */
1712
1713 static int
1714 find_fixedlength(pcre_uchar *code, BOOL utf, BOOL atend, compile_data *cd)
1715 {
1716 int length = -1;
1717
1718 register int branchlength = 0;
1719 register pcre_uchar *cc = code + 1 + LINK_SIZE;
1720
1721 /* Scan along the opcodes for this branch. If we get to the end of the
1722 branch, check the length against that of the other branches. */
1723
1724 for (;;)
1725 {
1726 int d;
1727 pcre_uchar *ce, *cs;
1728 register int op = *cc;
1729
1730 switch (op)
1731 {
1732 /* We only need to continue for OP_CBRA (normal capturing bracket) and
1733 OP_BRA (normal non-capturing bracket) because the other variants of these
1734 opcodes are all concerned with unlimited repeated groups, which of course
1735 are not of fixed length. */
1736
1737 case OP_CBRA:
1738 case OP_BRA:
1739 case OP_ONCE:
1740 case OP_ONCE_NC:
1741 case OP_COND:
1742 d = find_fixedlength(cc + ((op == OP_CBRA)? IMM2_SIZE : 0), utf, atend, cd);
1743 if (d < 0) return d;
1744 branchlength += d;
1745 do cc += GET(cc, 1); while (*cc == OP_ALT);
1746 cc += 1 + LINK_SIZE;
1747 break;
1748
1749 /* Reached end of a branch; if it's a ket it is the end of a nested call.
1750 If it's ALT it is an alternation in a nested call. An ACCEPT is effectively
1751 an ALT. If it is END it's the end of the outer call. All can be handled by
1752 the same code. Note that we must not include the OP_KETRxxx opcodes here,
1753 because they all imply an unlimited repeat. */
1754
1755 case OP_ALT:
1756 case OP_KET:
1757 case OP_END:
1758 case OP_ACCEPT:
1759 case OP_ASSERT_ACCEPT:
1760 if (length < 0) length = branchlength;
1761 else if (length != branchlength) return -1;
1762 if (*cc != OP_ALT) return length;
1763 cc += 1 + LINK_SIZE;
1764 branchlength = 0;
1765 break;
1766
1767 /* A true recursion implies not fixed length, but a subroutine call may
1768 be OK. If the subroutine is a forward reference, we can't deal with
1769 it until the end of the pattern, so return -3. */
1770
1771 case OP_RECURSE:
1772 if (!atend) return -3;
1773 cs = ce = (pcre_uchar *)cd->start_code + GET(cc, 1); /* Start subpattern */
1774 do ce += GET(ce, 1); while (*ce == OP_ALT); /* End subpattern */
1775 if (cc > cs && cc < ce) return -1; /* Recursion */
1776 d = find_fixedlength(cs + IMM2_SIZE, utf, atend, cd);
1777 if (d < 0) return d;
1778 branchlength += d;
1779 cc += 1 + LINK_SIZE;
1780 break;
1781
1782 /* Skip over assertive subpatterns */
1783
1784 case OP_ASSERT:
1785 case OP_ASSERT_NOT:
1786 case OP_ASSERTBACK:
1787 case OP_ASSERTBACK_NOT:
1788 do cc += GET(cc, 1); while (*cc == OP_ALT);
1789 cc += PRIV(OP_lengths)[*cc];
1790 break;
1791
1792 /* Skip over things that don't match chars */
1793
1794 case OP_MARK:
1795 case OP_PRUNE_ARG:
1796 case OP_SKIP_ARG:
1797 case OP_THEN_ARG:
1798 cc += cc[1] + PRIV(OP_lengths)[*cc];
1799 break;
1800
1801 case OP_CALLOUT:
1802 case OP_CIRC:
1803 case OP_CIRCM:
1804 case OP_CLOSE:
1805 case OP_COMMIT:
1806 case OP_CREF:
1807 case OP_DEF:
1808 case OP_DOLL:
1809 case OP_DOLLM:
1810 case OP_EOD:
1811 case OP_EODN:
1812 case OP_FAIL:
1813 case OP_NCREF:
1814 case OP_NRREF:
1815 case OP_NOT_WORD_BOUNDARY:
1816 case OP_PRUNE:
1817 case OP_REVERSE:
1818 case OP_RREF:
1819 case OP_SET_SOM:
1820 case OP_SKIP:
1821 case OP_SOD:
1822 case OP_SOM:
1823 case OP_THEN:
1824 case OP_WORD_BOUNDARY:
1825 cc += PRIV(OP_lengths)[*cc];
1826 break;
1827
1828 /* Handle literal characters */
1829
1830 case OP_CHAR:
1831 case OP_CHARI:
1832 case OP_NOT:
1833 case OP_NOTI:
1834 branchlength++;
1835 cc += 2;
1836 #ifdef SUPPORT_UTF
1837 if (utf && HAS_EXTRALEN(cc[-1])) cc += GET_EXTRALEN(cc[-1]);
1838 #endif
1839 break;
1840
1841 /* Handle exact repetitions. The count is already in characters, but we
1842 need to skip over a multibyte character in UTF8 mode. */
1843
1844 case OP_EXACT:
1845 case OP_EXACTI:
1846 case OP_NOTEXACT:
1847 case OP_NOTEXACTI:
1848 branchlength += GET2(cc,1);
1849 cc += 2 + IMM2_SIZE;
1850 #ifdef SUPPORT_UTF
1851 if (utf && HAS_EXTRALEN(cc[-1])) cc += GET_EXTRALEN(cc[-1]);
1852 #endif
1853 break;
1854
1855 case OP_TYPEEXACT:
1856 branchlength += GET2(cc,1);
1857 if (cc[1 + IMM2_SIZE] == OP_PROP || cc[1 + IMM2_SIZE] == OP_NOTPROP) cc += 2;
1858 cc += 1 + IMM2_SIZE + 1;
1859 break;
1860
1861 /* Handle single-char matchers */
1862
1863 case OP_PROP:
1864 case OP_NOTPROP:
1865 cc += 2;
1866 /* Fall through */
1867
1868 case OP_HSPACE:
1869 case OP_VSPACE:
1870 case OP_NOT_HSPACE:
1871 case OP_NOT_VSPACE:
1872 case OP_NOT_DIGIT:
1873 case OP_DIGIT:
1874 case OP_NOT_WHITESPACE:
1875 case OP_WHITESPACE:
1876 case OP_NOT_WORDCHAR:
1877 case OP_WORDCHAR:
1878 case OP_ANY:
1879 case OP_ALLANY:
1880 branchlength++;
1881 cc++;
1882 break;
1883
1884 /* The single-byte matcher isn't allowed. This only happens in UTF-8 mode;
1885 otherwise \C is coded as OP_ALLANY. */
1886
1887 case OP_ANYBYTE:
1888 return -2;
1889
1890 /* Check a class for variable quantification */
1891
1892 #if defined SUPPORT_UTF || defined COMPILE_PCRE16
1893 case OP_XCLASS:
1894 cc += GET(cc, 1) - PRIV(OP_lengths)[OP_CLASS];
1895 /* Fall through */
1896 #endif
1897
1898 case OP_CLASS:
1899 case OP_NCLASS:
1900 cc += PRIV(OP_lengths)[OP_CLASS];
1901
1902 switch (*cc)
1903 {
1904 case OP_CRPLUS:
1905 case OP_CRMINPLUS:
1906 case OP_CRSTAR:
1907 case OP_CRMINSTAR:
1908 case OP_CRQUERY:
1909 case OP_CRMINQUERY:
1910 return -1;
1911
1912 case OP_CRRANGE:
1913 case OP_CRMINRANGE:
1914 if (GET2(cc,1) != GET2(cc,1+IMM2_SIZE)) return -1;
1915 branchlength += GET2(cc,1);
1916 cc += 1 + 2 * IMM2_SIZE;
1917 break;
1918
1919 default:
1920 branchlength++;
1921 }
1922 break;
1923
1924 /* Anything else is variable length */
1925
1926 case OP_ANYNL:
1927 case OP_BRAMINZERO:
1928 case OP_BRAPOS:
1929 case OP_BRAPOSZERO:
1930 case OP_BRAZERO:
1931 case OP_CBRAPOS:
1932 case OP_EXTUNI:
1933 case OP_KETRMAX:
1934 case OP_KETRMIN:
1935 case OP_KETRPOS:
1936 case OP_MINPLUS:
1937 case OP_MINPLUSI:
1938 case OP_MINQUERY:
1939 case OP_MINQUERYI:
1940 case OP_MINSTAR:
1941 case OP_MINSTARI:
1942 case OP_MINUPTO:
1943 case OP_MINUPTOI:
1944 case OP_NOTMINPLUS:
1945 case OP_NOTMINPLUSI:
1946 case OP_NOTMINQUERY:
1947 case OP_NOTMINQUERYI:
1948 case OP_NOTMINSTAR:
1949 case OP_NOTMINSTARI:
1950 case OP_NOTMINUPTO:
1951 case OP_NOTMINUPTOI:
1952 case OP_NOTPLUS:
1953 case OP_NOTPLUSI:
1954 case OP_NOTPOSPLUS:
1955 case OP_NOTPOSPLUSI:
1956 case OP_NOTPOSQUERY:
1957 case OP_NOTPOSQUERYI:
1958 case OP_NOTPOSSTAR:
1959 case OP_NOTPOSSTARI:
1960 case OP_NOTPOSUPTO:
1961 case OP_NOTPOSUPTOI:
1962 case OP_NOTQUERY:
1963 case OP_NOTQUERYI:
1964 case OP_NOTSTAR:
1965 case OP_NOTSTARI:
1966 case OP_NOTUPTO:
1967 case OP_NOTUPTOI:
1968 case OP_PLUS:
1969 case OP_PLUSI:
1970 case OP_POSPLUS:
1971 case OP_POSPLUSI:
1972 case OP_POSQUERY:
1973 case OP_POSQUERYI:
1974 case OP_POSSTAR:
1975 case OP_POSSTARI:
1976 case OP_POSUPTO:
1977 case OP_POSUPTOI:
1978 case OP_QUERY:
1979 case OP_QUERYI:
1980 case OP_REF:
1981 case OP_REFI:
1982 case OP_SBRA:
1983 case OP_SBRAPOS:
1984 case OP_SCBRA:
1985 case OP_SCBRAPOS:
1986 case OP_SCOND:
1987 case OP_SKIPZERO:
1988 case OP_STAR:
1989 case OP_STARI:
1990 case OP_TYPEMINPLUS:
1991 case OP_TYPEMINQUERY:
1992 case OP_TYPEMINSTAR:
1993 case OP_TYPEMINUPTO:
1994 case OP_TYPEPLUS:
1995 case OP_TYPEPOSPLUS:
1996 case OP_TYPEPOSQUERY:
1997 case OP_TYPEPOSSTAR:
1998 case OP_TYPEPOSUPTO:
1999 case OP_TYPEQUERY:
2000 case OP_TYPESTAR:
2001 case OP_TYPEUPTO:
2002 case OP_UPTO:
2003 case OP_UPTOI:
2004 return -1;
2005
2006 /* Catch unrecognized opcodes so that when new ones are added they
2007 are not forgotten, as has happened in the past. */
2008
2009 default:
2010 return -4;
2011 }
2012 }
2013 /* Control never gets here */
2014 }
2015
2016
2017
2018
2019 /*************************************************
2020 * Scan compiled regex for specific bracket *
2021 *************************************************/
2022
2023 /* This little function scans through a compiled pattern until it finds a
2024 capturing bracket with the given number, or, if the number is negative, an
2025 instance of OP_REVERSE for a lookbehind. The function is global in the C sense
2026 so that it can be called from pcre_study() when finding the minimum matching
2027 length.
2028
2029 Arguments:
2030 code points to start of expression
2031 utf TRUE in UTF-8 / UTF-16 mode
2032 number the required bracket number or negative to find a lookbehind
2033
2034 Returns: pointer to the opcode for the bracket, or NULL if not found
2035 */
2036
2037 const pcre_uchar *
2038 PRIV(find_bracket)(const pcre_uchar *code, BOOL utf, int number)
2039 {
2040 for (;;)
2041 {
2042 register int c = *code;
2043
2044 if (c == OP_END) return NULL;
2045
2046 /* XCLASS is used for classes that cannot be represented just by a bit
2047 map. This includes negated single high-valued characters. The length in
2048 the table is zero; the actual length is stored in the compiled code. */
2049
2050 if (c == OP_XCLASS) code += GET(code, 1);
2051
2052 /* Handle recursion */
2053
2054 else if (c == OP_REVERSE)
2055 {
2056 if (number < 0) return (pcre_uchar *)code;
2057 code += PRIV(OP_lengths)[c];
2058 }
2059
2060 /* Handle capturing bracket */
2061
2062 else if (c == OP_CBRA || c == OP_SCBRA ||
2063 c == OP_CBRAPOS || c == OP_SCBRAPOS)
2064 {
2065 int n = GET2(code, 1+LINK_SIZE);
2066 if (n == number) return (pcre_uchar *)code;
2067 code += PRIV(OP_lengths)[c];
2068 }
2069
2070 /* Otherwise, we can get the item's length from the table, except that for
2071 repeated character types, we have to test for \p and \P, which have an extra
2072 two bytes of parameters, and for MARK/PRUNE/SKIP/THEN with an argument, we
2073 must add in its length. */
2074
2075 else
2076 {
2077 switch(c)
2078 {
2079 case OP_TYPESTAR:
2080 case OP_TYPEMINSTAR:
2081 case OP_TYPEPLUS:
2082 case OP_TYPEMINPLUS:
2083 case OP_TYPEQUERY:
2084 case OP_TYPEMINQUERY:
2085 case OP_TYPEPOSSTAR:
2086 case OP_TYPEPOSPLUS:
2087 case OP_TYPEPOSQUERY:
2088 if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
2089 break;
2090
2091 case OP_TYPEUPTO:
2092 case OP_TYPEMINUPTO:
2093 case OP_TYPEEXACT:
2094 case OP_TYPEPOSUPTO:
2095 if (code[1 + IMM2_SIZE] == OP_PROP
2096 || code[1 + IMM2_SIZE] == OP_NOTPROP) code += 2;
2097 break;
2098
2099 case OP_MARK:
2100 case OP_PRUNE_ARG:
2101 case OP_SKIP_ARG:
2102 code += code[1];
2103 break;
2104
2105 case OP_THEN_ARG:
2106 code += code[1];
2107 break;
2108 }
2109
2110 /* Add in the fixed length from the table */
2111
2112 code += PRIV(OP_lengths)[c];
2113
2114 /* In UTF-8 mode, opcodes that are followed by a character may be followed by
2115 a multi-byte character. The length in the table is a minimum, so we have to
2116 arrange to skip the extra bytes. */
2117
2118 #ifdef SUPPORT_UTF
2119 if (utf) switch(c)
2120 {
2121 case OP_CHAR:
2122 case OP_CHARI:
2123 case OP_EXACT:
2124 case OP_EXACTI:
2125 case OP_UPTO:
2126 case OP_UPTOI:
2127 case OP_MINUPTO:
2128 case OP_MINUPTOI:
2129 case OP_POSUPTO:
2130 case OP_POSUPTOI:
2131 case OP_STAR:
2132 case OP_STARI:
2133 case OP_MINSTAR:
2134 case OP_MINSTARI:
2135 case OP_POSSTAR:
2136 case OP_POSSTARI:
2137 case OP_PLUS:
2138 case OP_PLUSI:
2139 case OP_MINPLUS:
2140 case OP_MINPLUSI:
2141 case OP_POSPLUS:
2142 case OP_POSPLUSI:
2143 case OP_QUERY:
2144 case OP_QUERYI:
2145 case OP_MINQUERY:
2146 case OP_MINQUERYI:
2147 case OP_POSQUERY:
2148 case OP_POSQUERYI:
2149 if (HAS_EXTRALEN(code[-1])) code += GET_EXTRALEN(code[-1]);
2150 break;
2151 }
2152 #else
2153 (void)(utf); /* Keep compiler happy by referencing function argument */
2154 #endif
2155 }
2156 }
2157 }
2158
2159
2160
2161 /*************************************************
2162 * Scan compiled regex for recursion reference *
2163 *************************************************/
2164
2165 /* This little function scans through a compiled pattern until it finds an
2166 instance of OP_RECURSE.
2167
2168 Arguments:
2169 code points to start of expression
2170 utf TRUE in UTF-8 / UTF-16 mode
2171
2172 Returns: pointer to the opcode for OP_RECURSE, or NULL if not found
2173 */
2174
2175 static const pcre_uchar *
2176 find_recurse(const pcre_uchar *code, BOOL utf)
2177 {
2178 for (;;)
2179 {
2180 register int c = *code;
2181 if (c == OP_END) return NULL;
2182 if (c == OP_RECURSE) return code;
2183
2184 /* XCLASS is used for classes that cannot be represented just by a bit
2185 map. This includes negated single high-valued characters. The length in
2186 the table is zero; the actual length is stored in the compiled code. */
2187
2188 if (c == OP_XCLASS) code += GET(code, 1);
2189
2190 /* Otherwise, we can get the item's length from the table, except that for
2191 repeated character types, we have to test for \p and \P, which have an extra
2192 two bytes of parameters, and for MARK/PRUNE/SKIP/THEN with an argument, we
2193 must add in its length. */
2194
2195 else
2196 {
2197 switch(c)
2198 {
2199 case OP_TYPESTAR:
2200 case OP_TYPEMINSTAR:
2201 case OP_TYPEPLUS:
2202 case OP_TYPEMINPLUS:
2203 case OP_TYPEQUERY:
2204 case OP_TYPEMINQUERY:
2205 case OP_TYPEPOSSTAR:
2206 case OP_TYPEPOSPLUS:
2207 case OP_TYPEPOSQUERY:
2208 if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
2209 break;
2210
2211 case OP_TYPEPOSUPTO:
2212 case OP_TYPEUPTO:
2213 case OP_TYPEMINUPTO:
2214 case OP_TYPEEXACT:
2215 if (code[1 + IMM2_SIZE] == OP_PROP
2216 || code[1 + IMM2_SIZE] == OP_NOTPROP) code += 2;
2217 break;
2218
2219 case OP_MARK:
2220 case OP_PRUNE_ARG:
2221 case OP_SKIP_ARG:
2222 code += code[1];
2223 break;
2224
2225 case OP_THEN_ARG:
2226 code += code[1];
2227 break;
2228 }
2229
2230 /* Add in the fixed length from the table */
2231
2232 code += PRIV(OP_lengths)[c];
2233
2234 /* In UTF-8 mode, opcodes that are followed by a character may be followed
2235 by a multi-byte character. The length in the table is a minimum, so we have
2236 to arrange to skip the extra bytes. */
2237
2238 #ifdef SUPPORT_UTF
2239 if (utf) switch(c)
2240 {
2241 case OP_CHAR:
2242 case OP_CHARI:
2243 case OP_NOT:
2244 case OP_NOTI:
2245 case OP_EXACT:
2246 case OP_EXACTI:
2247 case OP_NOTEXACT:
2248 case OP_NOTEXACTI:
2249 case OP_UPTO:
2250 case OP_UPTOI:
2251 case OP_NOTUPTO:
2252 case OP_NOTUPTOI:
2253 case OP_MINUPTO:
2254 case OP_MINUPTOI:
2255 case OP_NOTMINUPTO:
2256 case OP_NOTMINUPTOI:
2257 case OP_POSUPTO:
2258 case OP_POSUPTOI:
2259 case OP_NOTPOSUPTO:
2260 case OP_NOTPOSUPTOI:
2261 case OP_STAR:
2262 case OP_STARI:
2263 case OP_NOTSTAR:
2264 case OP_NOTSTARI:
2265 case OP_MINSTAR:
2266 case OP_MINSTARI:
2267 case OP_NOTMINSTAR:
2268 case OP_NOTMINSTARI:
2269 case OP_POSSTAR:
2270 case OP_POSSTARI:
2271 case OP_NOTPOSSTAR:
2272 case OP_NOTPOSSTARI:
2273 case OP_PLUS:
2274 case OP_PLUSI:
2275 case OP_NOTPLUS:
2276 case OP_NOTPLUSI:
2277 case OP_MINPLUS:
2278 case OP_MINPLUSI:
2279 case OP_NOTMINPLUS:
2280 case OP_NOTMINPLUSI:
2281 case OP_POSPLUS:
2282 case OP_POSPLUSI:
2283 case OP_NOTPOSPLUS:
2284 case OP_NOTPOSPLUSI:
2285 case OP_QUERY:
2286 case OP_QUERYI:
2287 case OP_NOTQUERY:
2288 case OP_NOTQUERYI:
2289 case OP_MINQUERY:
2290 case OP_MINQUERYI:
2291 case OP_NOTMINQUERY:
2292 case OP_NOTMINQUERYI:
2293 case OP_POSQUERY:
2294 case OP_POSQUERYI:
2295 case OP_NOTPOSQUERY:
2296 case OP_NOTPOSQUERYI:
2297 if (HAS_EXTRALEN(code[-1])) code += GET_EXTRALEN(code[-1]);
2298 break;
2299 }
2300 #else
2301 (void)(utf); /* Keep compiler happy by referencing function argument */
2302 #endif
2303 }
2304 }
2305 }
2306
2307
2308
2309 /*************************************************
2310 * Scan compiled branch for non-emptiness *
2311 *************************************************/
2312
2313 /* This function scans through a branch of a compiled pattern to see whether it
2314 can match the empty string or not. It is called from could_be_empty()
2315 below and from compile_branch() when checking for an unlimited repeat of a
2316 group that can match nothing. Note that first_significant_code() skips over
2317 backward and negative forward assertions when its final argument is TRUE. If we
2318 hit an unclosed bracket, we return "empty" - this means we've struck an inner
2319 bracket whose current branch will already have been scanned.
2320
2321 Arguments:
2322 code points to start of search
2323 endcode points to where to stop
2324 utf TRUE if in UTF-8 / UTF-16 mode
2325 cd contains pointers to tables etc.
2326
2327 Returns: TRUE if what is matched could be empty
2328 */
2329
2330 static BOOL
2331 could_be_empty_branch(const pcre_uchar *code, const pcre_uchar *endcode,
2332 BOOL utf, compile_data *cd)
2333 {
2334 register int c;
2335 for (code = first_significant_code(code + PRIV(OP_lengths)[*code], TRUE);
2336 code < endcode;
2337 code = first_significant_code(code + PRIV(OP_lengths)[c], TRUE))
2338 {
2339 const pcre_uchar *ccode;
2340
2341 c = *code;
2342
2343 /* Skip over forward assertions; the other assertions are skipped by
2344 first_significant_code() with a TRUE final argument. */
2345
2346 if (c == OP_ASSERT)
2347 {
2348 do code += GET(code, 1); while (*code == OP_ALT);
2349 c = *code;
2350 continue;
2351 }
2352
2353 /* For a recursion/subroutine call, if its end has been reached, which
2354 implies a backward reference subroutine call, we can scan it. If it's a
2355 forward reference subroutine call, we can't. To detect forward reference
2356 we have to scan up the list that is kept in the workspace. This function is
2357 called only when doing the real compile, not during the pre-compile that
2358 measures the size of the compiled pattern. */
2359
2360 if (c == OP_RECURSE)
2361 {
2362 const pcre_uchar *scode;
2363 BOOL empty_branch;
2364
2365 /* Test for forward reference */
2366
2367 for (scode = cd->start_workspace; scode < cd->hwm; scode += LINK_SIZE)
2368 if (GET(scode, 0) == code + 1 - cd->start_code) return TRUE;
2369
2370 /* Not a forward reference, test for completed backward reference */
2371
2372 empty_branch = FALSE;
2373 scode = cd->start_code + GET(code, 1);
2374 if (GET(scode, 1) == 0) return TRUE; /* Unclosed */
2375
2376 /* Completed backwards reference */
2377
2378 do
2379 {
2380 if (could_be_empty_branch(scode, endcode, utf, cd))
2381 {
2382 empty_branch = TRUE;
2383 break;
2384 }
2385 scode += GET(scode, 1);
2386 }
2387 while (*scode == OP_ALT);
2388
2389 if (!empty_branch) return FALSE; /* All branches are non-empty */
2390 continue;
2391 }
2392
2393 /* Groups with zero repeats can of course be empty; skip them. */
2394
2395 if (c == OP_BRAZERO || c == OP_BRAMINZERO || c == OP_SKIPZERO ||
2396 c == OP_BRAPOSZERO)
2397 {
2398 code += PRIV(OP_lengths)[c];
2399 do code += GET(code, 1); while (*code == OP_ALT);
2400 c = *code;
2401 continue;
2402 }
2403
2404 /* A nested group that is already marked as "could be empty" can just be
2405 skipped. */
2406
2407 if (c == OP_SBRA || c == OP_SBRAPOS ||
2408 c == OP_SCBRA || c == OP_SCBRAPOS)
2409 {
2410 do code += GET(code, 1); while (*code == OP_ALT);
2411 c = *code;
2412 continue;
2413 }
2414
2415 /* For other groups, scan the branches. */
2416
2417 if (c == OP_BRA || c == OP_BRAPOS ||
2418 c == OP_CBRA || c == OP_CBRAPOS ||
2419 c == OP_ONCE || c == OP_ONCE_NC ||
2420 c == OP_COND)
2421 {
2422 BOOL empty_branch;
2423 if (GET(code, 1) == 0) return TRUE; /* Hit unclosed bracket */
2424
2425 /* If a conditional group has only one branch, there is a second, implied,
2426 empty branch, so just skip over the conditional, because it could be empty.
2427 Otherwise, scan the individual branches of the group. */
2428
2429 if (c == OP_COND && code[GET(code, 1)] != OP_ALT)
2430 code += GET(code, 1);
2431 else
2432 {
2433 empty_branch = FALSE;
2434 do
2435 {
2436 if (!empty_branch && could_be_empty_branch(code, endcode, utf, cd))
2437 empty_branch = TRUE;
2438 code += GET(code, 1);
2439 }
2440 while (*code == OP_ALT);
2441 if (!empty_branch) return FALSE; /* All branches are non-empty */
2442 }
2443
2444 c = *code;
2445 continue;
2446 }
2447
2448 /* Handle the other opcodes */
2449
2450 switch (c)
2451 {
2452 /* Check for quantifiers after a class. XCLASS is used for classes that
2453 cannot be represented just by a bit map. This includes negated single
2454 high-valued characters. The length in PRIV(OP_lengths)[] is zero; the
2455 actual length is stored in the compiled code, so we must update "code"
2456 here. */
2457
2458 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
2459 case OP_XCLASS:
2460 ccode = code += GET(code, 1);
2461 goto CHECK_CLASS_REPEAT;
2462 #endif
2463
2464 case OP_CLASS:
2465 case OP_NCLASS:
2466 ccode = code + PRIV(OP_lengths)[OP_CLASS];
2467
2468 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
2469 CHECK_CLASS_REPEAT:
2470 #endif
2471
2472 switch (*ccode)
2473 {
2474 case OP_CRSTAR: /* These could be empty; continue */
2475 case OP_CRMINSTAR:
2476 case OP_CRQUERY:
2477 case OP_CRMINQUERY:
2478 break;
2479
2480 default: /* Non-repeat => class must match */
2481 case OP_CRPLUS: /* These repeats aren't empty */
2482 case OP_CRMINPLUS:
2483 return FALSE;
2484
2485 case OP_CRRANGE:
2486 case OP_CRMINRANGE:
2487 if (GET2(ccode, 1) > 0) return FALSE; /* Minimum > 0 */
2488 break;
2489 }
2490 break;
2491
2492 /* Opcodes that must match a character */
2493
2494 case OP_PROP:
2495 case OP_NOTPROP:
2496 case OP_EXTUNI:
2497 case OP_NOT_DIGIT:
2498 case OP_DIGIT:
2499 case OP_NOT_WHITESPACE:
2500 case OP_WHITESPACE:
2501 case OP_NOT_WORDCHAR:
2502 case OP_WORDCHAR:
2503 case OP_ANY:
2504 case OP_ALLANY:
2505 case OP_ANYBYTE:
2506 case OP_CHAR:
2507 case OP_CHARI:
2508 case OP_NOT:
2509 case OP_NOTI:
2510 case OP_PLUS:
2511 case OP_MINPLUS:
2512 case OP_POSPLUS:
2513 case OP_EXACT:
2514 case OP_NOTPLUS:
2515 case OP_NOTMINPLUS:
2516 case OP_NOTPOSPLUS:
2517 case OP_NOTEXACT:
2518 case OP_TYPEPLUS:
2519 case OP_TYPEMINPLUS:
2520 case OP_TYPEPOSPLUS:
2521 case OP_TYPEEXACT:
2522 return FALSE;
2523
2524 /* These are going to continue, as they may be empty, but we have to
2525 fudge the length for the \p and \P cases. */
2526
2527 case OP_TYPESTAR:
2528 case OP_TYPEMINSTAR:
2529 case OP_TYPEPOSSTAR:
2530 case OP_TYPEQUERY:
2531 case OP_TYPEMINQUERY:
2532 case OP_TYPEPOSQUERY:
2533 if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
2534 break;
2535
2536 /* Same for these */
2537
2538 case OP_TYPEUPTO:
2539 case OP_TYPEMINUPTO:
2540 case OP_TYPEPOSUPTO:
2541 if (code[1 + IMM2_SIZE] == OP_PROP
2542 || code[1 + IMM2_SIZE] == OP_NOTPROP) code += 2;
2543 break;
2544
2545 /* End of branch */
2546
2547 case OP_KET:
2548 case OP_KETRMAX:
2549 case OP_KETRMIN:
2550 case OP_KETRPOS:
2551 case OP_ALT:
2552 return TRUE;
2553
2554 /* In UTF-8 mode, STAR, MINSTAR, POSSTAR, QUERY, MINQUERY, POSQUERY, UPTO,
2555 MINUPTO, and POSUPTO may be followed by a multibyte character */
2556
2557 #ifdef SUPPORT_UTF
2558 case OP_STAR:
2559 case OP_STARI:
2560 case OP_MINSTAR:
2561 case OP_MINSTARI:
2562 case OP_POSSTAR:
2563 case OP_POSSTARI:
2564 case OP_QUERY:
2565 case OP_QUERYI:
2566 case OP_MINQUERY:
2567 case OP_MINQUERYI:
2568 case OP_POSQUERY:
2569 case OP_POSQUERYI:
2570 if (utf && HAS_EXTRALEN(code[1])) code += GET_EXTRALEN(code[1]);
2571 break;
2572
2573 case OP_UPTO:
2574 case OP_UPTOI:
2575 case OP_MINUPTO:
2576 case OP_MINUPTOI:
2577 case OP_POSUPTO:
2578 case OP_POSUPTOI:
2579 if (utf && HAS_EXTRALEN(code[1 + IMM2_SIZE])) code += GET_EXTRALEN(code[1 + IMM2_SIZE]);
2580 break;
2581 #endif
2582
2583 /* MARK, and PRUNE/SKIP/THEN with an argument must skip over the argument
2584 string. */
2585
2586 case OP_MARK:
2587 case OP_PRUNE_ARG:
2588 case OP_SKIP_ARG:
2589 code += code[1];
2590 break;
2591
2592 case OP_THEN_ARG:
2593 code += code[1];
2594 break;
2595
2596 /* None of the remaining opcodes are required to match a character. */
2597
2598 default:
2599 break;
2600 }
2601 }
2602
2603 return TRUE;
2604 }
2605
2606
2607
2608 /*************************************************
2609 * Scan compiled regex for non-emptiness *
2610 *************************************************/
2611
2612 /* This function is called to check for left recursive calls. We want to check
2613 the current branch of the current pattern to see if it could match the empty
2614 string. If it could, we must look outwards for branches at other levels,
2615 stopping when we pass beyond the bracket which is the subject of the recursion.
2616 This function is called only during the real compile, not during the
2617 pre-compile.
2618
2619 Arguments:
2620 code points to start of the recursion
2621 endcode points to where to stop (current RECURSE item)
2622 bcptr points to the chain of current (unclosed) branch starts
2623 utf TRUE if in UTF-8 / UTF-16 mode
2624 cd pointers to tables etc
2625
2626 Returns: TRUE if what is matched could be empty
2627 */
2628
2629 static BOOL
2630 could_be_empty(const pcre_uchar *code, const pcre_uchar *endcode,
2631 branch_chain *bcptr, BOOL utf, compile_data *cd)
2632 {
2633 while (bcptr != NULL && bcptr->current_branch >= code)
2634 {
2635 if (!could_be_empty_branch(bcptr->current_branch, endcode, utf, cd))
2636 return FALSE;
2637 bcptr = bcptr->outer;
2638 }
2639 return TRUE;
2640 }
2641
2642
2643
2644 /*************************************************
2645 * Check for POSIX class syntax *
2646 *************************************************/
2647
2648 /* This function is called when the sequence "[:" or "[." or "[=" is
2649 encountered in a character class. It checks whether this is followed by a
2650 sequence of characters terminated by a matching ":]" or ".]" or "=]". If we
2651 reach an unescaped ']' without the special preceding character, return FALSE.
2652
2653 Originally, this function only recognized a sequence of letters between the
2654 terminators, but it seems that Perl recognizes any sequence of characters,
2655 though of course unknown POSIX names are subsequently rejected. Perl gives an
2656 "Unknown POSIX class" error for [:f\oo:] for example, where previously PCRE
2657 didn't consider this to be a POSIX class. Likewise for [:1234:].
2658
2659 The problem in trying to be exactly like Perl is in the handling of escapes. We
2660 have to be sure that [abc[:x\]pqr] is *not* treated as containing a POSIX
2661 class, but [abc[:x\]pqr:]] is (so that an error can be generated). The code
2662 below handles the special case of \], but does not try to do any other escape
2663 processing. This makes it different from Perl for cases such as [:l\ower:]
2664 where Perl recognizes it as the POSIX class "lower" but PCRE does not recognize
2665 "l\ower". This is a lesser evil that not diagnosing bad classes when Perl does,
2666 I think.
2667
2668 A user pointed out that PCRE was rejecting [:a[:digit:]] whereas Perl was not.
2669 It seems that the appearance of a nested POSIX class supersedes an apparent
2670 external class. For example, [:a[:digit:]b:] matches "a", "b", ":", or
2671 a digit.
2672
2673 In Perl, unescaped square brackets may also appear as part of class names. For
2674 example, [:a[:abc]b:] gives unknown POSIX class "[:abc]b:]". However, for
2675 [:a[:abc]b][b:] it gives unknown POSIX class "[:abc]b][b:]", which does not
2676 seem right at all. PCRE does not allow closing square brackets in POSIX class
2677 names.
2678
2679 Arguments:
2680 ptr pointer to the initial [
2681 endptr where to return the end pointer
2682
2683 Returns: TRUE or FALSE
2684 */
2685
2686 static BOOL
2687 check_posix_syntax(const pcre_uchar *ptr, const pcre_uchar **endptr)
2688 {
2689 int terminator; /* Don't combine these lines; the Solaris cc */
2690 terminator = *(++ptr); /* compiler warns about "non-constant" initializer. */
2691 for (++ptr; *ptr != 0; ptr++)
2692 {
2693 if (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET)
2694 ptr++;
2695 else if (*ptr == CHAR_RIGHT_SQUARE_BRACKET) return FALSE;
2696 else
2697 {
2698 if (*ptr == terminator && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET)
2699 {
2700 *endptr = ptr;
2701 return TRUE;
2702 }
2703 if (*ptr == CHAR_LEFT_SQUARE_BRACKET &&
2704 (ptr[1] == CHAR_COLON || ptr[1] == CHAR_DOT ||
2705 ptr[1] == CHAR_EQUALS_SIGN) &&
2706 check_posix_syntax(ptr, endptr))
2707 return FALSE;
2708 }
2709 }
2710 return FALSE;
2711 }
2712
2713
2714
2715
2716 /*************************************************
2717 * Check POSIX class name *
2718 *************************************************/
2719
2720 /* This function is called to check the name given in a POSIX-style class entry
2721 such as [:alnum:].
2722
2723 Arguments:
2724 ptr points to the first letter
2725 len the length of the name
2726
2727 Returns: a value representing the name, or -1 if unknown
2728 */
2729
2730 static int
2731 check_posix_name(const pcre_uchar *ptr, int len)
2732 {
2733 const char *pn = posix_names;
2734 register int yield = 0;
2735 while (posix_name_lengths[yield] != 0)
2736 {
2737 if (len == posix_name_lengths[yield] &&
2738 STRNCMP_UC_C8(ptr, pn, len) == 0) return yield;
2739 pn += posix_name_lengths[yield] + 1;
2740 yield++;
2741 }
2742 return -1;
2743 }
2744
2745
2746 /*************************************************
2747 * Adjust OP_RECURSE items in repeated group *
2748 *************************************************/
2749
2750 /* OP_RECURSE items contain an offset from the start of the regex to the group
2751 that is referenced. This means that groups can be replicated for fixed
2752 repetition simply by copying (because the recursion is allowed to refer to
2753 earlier groups that are outside the current group). However, when a group is
2754 optional (i.e. the minimum quantifier is zero), OP_BRAZERO or OP_SKIPZERO is
2755 inserted before it, after it has been compiled. This means that any OP_RECURSE
2756 items within it that refer to the group itself or any contained groups have to
2757 have their offsets adjusted. That one of the jobs of this function. Before it
2758 is called, the partially compiled regex must be temporarily terminated with
2759 OP_END.
2760
2761 This function has been extended with the possibility of forward references for
2762 recursions and subroutine calls. It must also check the list of such references
2763 for the group we are dealing with. If it finds that one of the recursions in
2764 the current group is on this list, it adjusts the offset in the list, not the
2765 value in the reference (which is a group number).
2766
2767 Arguments:
2768 group points to the start of the group
2769 adjust the amount by which the group is to be moved
2770 utf TRUE in UTF-8 / UTF-16 mode
2771 cd contains pointers to tables etc.
2772 save_hwm the hwm forward reference pointer at the start of the group
2773
2774 Returns: nothing
2775 */
2776
2777 static void
2778 adjust_recurse(pcre_uchar *group, int adjust, BOOL utf, compile_data *cd,
2779 pcre_uchar *save_hwm)
2780 {
2781 pcre_uchar *ptr = group;
2782
2783 while ((ptr = (pcre_uchar *)find_recurse(ptr, utf)) != NULL)
2784 {
2785 int offset;
2786 pcre_uchar *hc;
2787
2788 /* See if this recursion is on the forward reference list. If so, adjust the
2789 reference. */
2790
2791 for (hc = save_hwm; hc < cd->hwm; hc += LINK_SIZE)
2792 {
2793 offset = GET(hc, 0);
2794 if (cd->start_code + offset == ptr + 1)
2795 {
2796 PUT(hc, 0, offset + adjust);
2797 break;
2798 }
2799 }
2800
2801 /* Otherwise, adjust the recursion offset if it's after the start of this
2802 group. */
2803
2804 if (hc >= cd->hwm)
2805 {
2806 offset = GET(ptr, 1);
2807 if (cd->start_code + offset >= group) PUT(ptr, 1, offset + adjust);
2808 }
2809
2810 ptr += 1 + LINK_SIZE;
2811 }
2812 }
2813
2814
2815
2816 /*************************************************
2817 * Insert an automatic callout point *
2818 *************************************************/
2819
2820 /* This function is called when the PCRE_AUTO_CALLOUT option is set, to insert
2821 callout points before each pattern item.
2822
2823 Arguments:
2824 code current code pointer
2825 ptr current pattern pointer
2826 cd pointers to tables etc
2827
2828 Returns: new code pointer
2829 */
2830
2831 static pcre_uchar *
2832 auto_callout(pcre_uchar *code, const pcre_uchar *ptr, compile_data *cd)
2833 {
2834 *code++ = OP_CALLOUT;
2835 *code++ = 255;
2836 PUT(code, 0, (int)(ptr - cd->start_pattern)); /* Pattern offset */
2837 PUT(code, LINK_SIZE, 0); /* Default length */
2838 return code + 2 * LINK_SIZE;
2839 }
2840
2841
2842
2843 /*************************************************
2844 * Complete a callout item *
2845 *************************************************/
2846
2847 /* A callout item contains the length of the next item in the pattern, which
2848 we can't fill in till after we have reached the relevant point. This is used
2849 for both automatic and manual callouts.
2850
2851 Arguments:
2852 previous_callout points to previous callout item
2853 ptr current pattern pointer
2854 cd pointers to tables etc
2855
2856 Returns: nothing
2857 */
2858
2859 static void
2860 complete_callout(pcre_uchar *previous_callout, const pcre_uchar *ptr, compile_data *cd)
2861 {
2862 int length = (int)(ptr - cd->start_pattern - GET(previous_callout, 2));
2863 PUT(previous_callout, 2 + LINK_SIZE, length);
2864 }
2865
2866
2867
2868 #ifdef SUPPORT_UCP
2869 /*************************************************
2870 * Get othercase range *
2871 *************************************************/
2872
2873 /* This function is passed the start and end of a class range, in UTF-8 mode
2874 with UCP support. It searches up the characters, looking for internal ranges of
2875 characters in the "other" case. Each call returns the next one, updating the
2876 start address.
2877
2878 Arguments:
2879 cptr points to starting character value; updated
2880 d end value
2881 ocptr where to put start of othercase range
2882 odptr where to put end of othercase range
2883
2884 Yield: TRUE when range returned; FALSE when no more
2885 */
2886
2887 static BOOL
2888 get_othercase_range(unsigned int *cptr, unsigned int d, unsigned int *ocptr,
2889 unsigned int *odptr)
2890 {
2891 unsigned int c, othercase, next;
2892
2893 for (c = *cptr; c <= d; c++)
2894 { if ((othercase = UCD_OTHERCASE(c)) != c) break; }
2895
2896 if (c > d) return FALSE;
2897
2898 *ocptr = othercase;
2899 next = othercase + 1;
2900
2901 for (++c; c <= d; c++)
2902 {
2903 if (UCD_OTHERCASE(c) != next) break;
2904 next++;
2905 }
2906
2907 *odptr = next - 1;
2908 *cptr = c;
2909
2910 return TRUE;
2911 }
2912
2913
2914
2915 /*************************************************
2916 * Check a character and a property *
2917 *************************************************/
2918
2919 /* This function is called by check_auto_possessive() when a property item
2920 is adjacent to a fixed character.
2921
2922 Arguments:
2923 c the character
2924 ptype the property type
2925 pdata the data for the type
2926 negated TRUE if it's a negated property (\P or \p{^)
2927
2928 Returns: TRUE if auto-possessifying is OK
2929 */
2930
2931 static BOOL
2932 check_char_prop(int c, int ptype, int pdata, BOOL negated)
2933 {
2934 const ucd_record *prop = GET_UCD(c);
2935 switch(ptype)
2936 {
2937 case PT_LAMP:
2938 return (prop->chartype == ucp_Lu ||
2939 prop->chartype == ucp_Ll ||
2940 prop->chartype == ucp_Lt) == negated;
2941
2942 case PT_GC:
2943 return (pdata == PRIV(ucp_gentype)[prop->chartype]) == negated;
2944
2945 case PT_PC:
2946 return (pdata == prop->chartype) == negated;
2947
2948 case PT_SC:
2949 return (pdata == prop->script) == negated;
2950
2951 /* These are specials */
2952
2953 case PT_ALNUM:
2954 return (PRIV(ucp_gentype)[prop->chartype] == ucp_L ||
2955 PRIV(ucp_gentype)[prop->chartype] == ucp_N) == negated;
2956
2957 case PT_SPACE: /* Perl space */
2958 return (PRIV(ucp_gentype)[prop->chartype] == ucp_Z ||
2959 c == CHAR_HT || c == CHAR_NL || c == CHAR_FF || c == CHAR_CR)
2960 == negated;
2961
2962 case PT_PXSPACE: /* POSIX space */
2963 return (PRIV(ucp_gentype)[prop->chartype] == ucp_Z ||
2964 c == CHAR_HT || c == CHAR_NL || c == CHAR_VT ||
2965 c == CHAR_FF || c == CHAR_CR)
2966 == negated;
2967
2968 case PT_WORD:
2969 return (PRIV(ucp_gentype)[prop->chartype] == ucp_L ||
2970 PRIV(ucp_gentype)[prop->chartype] == ucp_N ||
2971 c == CHAR_UNDERSCORE) == negated;
2972 }
2973 return FALSE;
2974 }
2975 #endif /* SUPPORT_UCP */
2976
2977
2978
2979 /*************************************************
2980 * Check if auto-possessifying is possible *
2981 *************************************************/
2982
2983 /* This function is called for unlimited repeats of certain items, to see
2984 whether the next thing could possibly match the repeated item. If not, it makes
2985 sense to automatically possessify the repeated item.
2986
2987 Arguments:
2988 previous pointer to the repeated opcode
2989 utf TRUE in UTF-8 / UTF-16 mode
2990 ptr next character in pattern
2991 options options bits
2992 cd contains pointers to tables etc.
2993
2994 Returns: TRUE if possessifying is wanted
2995 */
2996
2997 static BOOL
2998 check_auto_possessive(const pcre_uchar *previous, BOOL utf,
2999 const pcre_uchar *ptr, int options, compile_data *cd)
3000 {
3001 pcre_int32 c, next;
3002 int op_code = *previous++;
3003
3004 /* Skip whitespace and comments in extended mode */
3005
3006 if ((options & PCRE_EXTENDED) != 0)
3007 {
3008 for (;;)
3009 {
3010 while (MAX_255(*ptr) && (cd->ctypes[*ptr] & ctype_space) != 0) ptr++;
3011 if (*ptr == CHAR_NUMBER_SIGN)
3012 {
3013 ptr++;
3014 while (*ptr != 0)
3015 {
3016 if (IS_NEWLINE(ptr)) { ptr += cd->nllen; break; }
3017 ptr++;
3018 #ifdef SUPPORT_UTF
3019 if (utf) FORWARDCHAR(ptr);
3020 #endif
3021 }
3022 }
3023 else break;
3024 }
3025 }
3026
3027 /* If the next item is one that we can handle, get its value. A non-negative
3028 value is a character, a negative value is an escape value. */
3029
3030 if (*ptr == CHAR_BACKSLASH)
3031 {
3032 int temperrorcode = 0;
3033 next = check_escape(&ptr, &temperrorcode, cd->bracount, options, FALSE);
3034 if (temperrorcode != 0) return FALSE;
3035 ptr++; /* Point after the escape sequence */
3036 }
3037 else if (!MAX_255(*ptr) || (cd->ctypes[*ptr] & ctype_meta) == 0)
3038 {
3039 #ifdef SUPPORT_UTF
3040 if (utf) { GETCHARINC(next, ptr); } else
3041 #endif
3042 next = *ptr++;
3043 }
3044 else return FALSE;
3045
3046 /* Skip whitespace and comments in extended mode */
3047
3048 if ((options & PCRE_EXTENDED) != 0)
3049 {
3050 for (;;)
3051 {
3052 while (MAX_255(*ptr) && (cd->ctypes[*ptr] & ctype_space) != 0) ptr++;
3053 if (*ptr == CHAR_NUMBER_SIGN)
3054 {
3055 ptr++;
3056 while (*ptr != 0)
3057 {
3058 if (IS_NEWLINE(ptr)) { ptr += cd->nllen; break; }
3059 ptr++;
3060 #ifdef SUPPORT_UTF
3061 if (utf) FORWARDCHAR(ptr);
3062 #endif
3063 }
3064 }
3065 else break;
3066 }
3067 }
3068
3069 /* If the next thing is itself optional, we have to give up. */
3070
3071 if (*ptr == CHAR_ASTERISK || *ptr == CHAR_QUESTION_MARK ||
3072 STRNCMP_UC_C8(ptr, STR_LEFT_CURLY_BRACKET STR_0 STR_COMMA, 3) == 0)
3073 return FALSE;
3074
3075 /* Now compare the next item with the previous opcode. First, handle cases when
3076 the next item is a character. */
3077
3078 if (next >= 0) switch(op_code)
3079 {
3080 case OP_CHAR:
3081 #ifdef SUPPORT_UTF
3082 GETCHARTEST(c, previous);
3083 #else
3084 c = *previous;
3085 #endif
3086 return c != next;
3087
3088 /* For CHARI (caseless character) we must check the other case. If we have
3089 Unicode property support, we can use it to test the other case of
3090 high-valued characters. */
3091
3092 case OP_CHARI:
3093 #ifdef SUPPORT_UTF
3094 GETCHARTEST(c, previous);
3095 #else
3096 c = *previous;
3097 #endif
3098 if (c == next) return FALSE;
3099 #ifdef SUPPORT_UTF
3100 if (utf)
3101 {
3102 unsigned int othercase;
3103 if (next < 128) othercase = cd->fcc[next]; else
3104 #ifdef SUPPORT_UCP
3105 othercase = UCD_OTHERCASE((unsigned int)next);
3106 #else
3107 othercase = NOTACHAR;
3108 #endif
3109 return (unsigned int)c != othercase;
3110 }
3111 else
3112 #endif /* SUPPORT_UTF */
3113 return (c != TABLE_GET((unsigned int)next, cd->fcc, next)); /* Non-UTF-8 mode */
3114
3115 case OP_NOT:
3116 #ifdef SUPPORT_UTF
3117 GETCHARTEST(c, previous);
3118 #else
3119 c = *previous;
3120 #endif
3121 return c == next;
3122
3123 case OP_NOTI:
3124 #ifdef SUPPORT_UTF
3125 GETCHARTEST(c, previous);
3126 #else
3127 c = *previous;
3128 #endif
3129 if (c == next) return TRUE;
3130 #ifdef SUPPORT_UTF
3131 if (utf)
3132 {
3133 unsigned int othercase;
3134 if (next < 128) othercase = cd->fcc[next]; else
3135 #ifdef SUPPORT_UCP
3136 othercase = UCD_OTHERCASE((unsigned int)next);
3137 #else
3138 othercase = NOTACHAR;
3139 #endif
3140 return (unsigned int)c == othercase;
3141 }
3142 else
3143 #endif /* SUPPORT_UTF */
3144 return (c == TABLE_GET((unsigned int)next, cd->fcc, next)); /* Non-UTF-8 mode */
3145
3146 /* Note that OP_DIGIT etc. are generated only when PCRE_UCP is *not* set.
3147 When it is set, \d etc. are converted into OP_(NOT_)PROP codes. */
3148
3149 case OP_DIGIT:
3150 return next > 255 || (cd->ctypes[next] & ctype_digit) == 0;
3151
3152 case OP_NOT_DIGIT:
3153 return next <= 255 && (cd->ctypes[next] & ctype_digit) != 0;
3154
3155 case OP_WHITESPACE:
3156 return next > 255 || (cd->ctypes[next] & ctype_space) == 0;
3157
3158 case OP_NOT_WHITESPACE:
3159 return next <= 255 && (cd->ctypes[next] & ctype_space) != 0;
3160
3161 case OP_WORDCHAR:
3162 return next > 255 || (cd->ctypes[next] & ctype_word) == 0;
3163
3164 case OP_NOT_WORDCHAR:
3165 return next <= 255 && (cd->ctypes[next] & ctype_word) != 0;
3166
3167 case OP_HSPACE:
3168 case OP_NOT_HSPACE:
3169 switch(next)
3170 {
3171 case 0x09:
3172 case 0x20:
3173 case 0xa0:
3174 case 0x1680:
3175 case 0x180e:
3176 case 0x2000:
3177 case 0x2001:
3178 case 0x2002:
3179 case 0x2003:
3180 case 0x2004:
3181 case 0x2005:
3182 case 0x2006:
3183 case 0x2007:
3184 case 0x2008:
3185 case 0x2009:
3186 case 0x200A:
3187 case 0x202f:
3188 case 0x205f:
3189 case 0x3000:
3190 return op_code == OP_NOT_HSPACE;
3191 default:
3192 return op_code != OP_NOT_HSPACE;
3193 }
3194
3195 case OP_ANYNL:
3196 case OP_VSPACE:
3197 case OP_NOT_VSPACE:
3198 switch(next)
3199 {
3200 case 0x0a:
3201 case 0x0b:
3202 case 0x0c:
3203 case 0x0d:
3204 case 0x85:
3205 case 0x2028:
3206 case 0x2029:
3207 return op_code == OP_NOT_VSPACE;
3208 default:
3209 return op_code != OP_NOT_VSPACE;
3210 }
3211
3212 #ifdef SUPPORT_UCP
3213 case OP_PROP:
3214 return check_char_prop(next, previous[0], previous[1], FALSE);
3215
3216 case OP_NOTPROP:
3217 return check_char_prop(next, previous[0], previous[1], TRUE);
3218 #endif
3219
3220 default:
3221 return FALSE;
3222 }
3223
3224
3225 /* Handle the case when the next item is \d, \s, etc. Note that when PCRE_UCP
3226 is set, \d turns into ESC_du rather than ESC_d, etc., so ESC_d etc. are
3227 generated only when PCRE_UCP is *not* set, that is, when only ASCII
3228 characteristics are recognized. Similarly, the opcodes OP_DIGIT etc. are
3229 replaced by OP_PROP codes when PCRE_UCP is set. */
3230
3231 switch(op_code)
3232 {
3233 case OP_CHAR:
3234 case OP_CHARI:
3235 #ifdef SUPPORT_UTF
3236 GETCHARTEST(c, previous);
3237 #else
3238 c = *previous;
3239 #endif
3240 switch(-next)
3241 {
3242 case ESC_d:
3243 return c > 255 || (cd->ctypes[c] & ctype_digit) == 0;
3244
3245 case ESC_D:
3246 return c <= 255 && (cd->ctypes[c] & ctype_digit) != 0;
3247
3248 case ESC_s:
3249 return c > 255 || (cd->ctypes[c] & ctype_space) == 0;
3250
3251 case ESC_S:
3252 return c <= 255 && (cd->ctypes[c] & ctype_space) != 0;
3253
3254 case ESC_w:
3255 return c > 255 || (cd->ctypes[c] & ctype_word) == 0;
3256
3257 case ESC_W:
3258 return c <= 255 && (cd->ctypes[c] & ctype_word) != 0;
3259
3260 case ESC_h:
3261 case ESC_H:
3262 switch(c)
3263 {
3264 case 0x09:
3265 case 0x20:
3266 case 0xa0:
3267 case 0x1680:
3268 case 0x180e:
3269 case 0x2000:
3270 case 0x2001:
3271 case 0x2002:
3272 case 0x2003:
3273 case 0x2004:
3274 case 0x2005:
3275 case 0x2006:
3276 case 0x2007:
3277 case 0x2008:
3278 case 0x2009:
3279 case 0x200A:
3280 case 0x202f:
3281 case 0x205f:
3282 case 0x3000:
3283 return -next != ESC_h;
3284 default:
3285 return -next == ESC_h;
3286 }
3287
3288 case ESC_v:
3289 case ESC_V:
3290 switch(c)
3291 {
3292 case 0x0a:
3293 case 0x0b:
3294 case 0x0c:
3295 case 0x0d:
3296 case 0x85:
3297 case 0x2028:
3298 case 0x2029:
3299 return -next != ESC_v;
3300 default:
3301 return -next == ESC_v;
3302 }
3303
3304 /* When PCRE_UCP is set, these values get generated for \d etc. Find
3305 their substitutions and process them. The result will always be either
3306 -ESC_p or -ESC_P. Then fall through to process those values. */
3307
3308 #ifdef SUPPORT_UCP
3309 case ESC_du:
3310 case ESC_DU:
3311 case ESC_wu:
3312 case ESC_WU:
3313 case ESC_su:
3314 case ESC_SU:
3315 {
3316 int temperrorcode = 0;
3317 ptr = substitutes[-next - ESC_DU];
3318 next = check_escape(&ptr, &temperrorcode, 0, options, FALSE);
3319 if (temperrorcode != 0) return FALSE;
3320 ptr++; /* For compatibility */
3321 }
3322 /* Fall through */
3323
3324 case ESC_p:
3325 case ESC_P:
3326 {
3327 int ptype, pdata, errorcodeptr;
3328 BOOL negated;
3329
3330 ptr--; /* Make ptr point at the p or P */
3331 ptype = get_ucp(&ptr, &negated, &pdata, &errorcodeptr);
3332 if (ptype < 0) return FALSE;
3333 ptr++; /* Point past the final curly ket */
3334
3335 /* If the property item is optional, we have to give up. (When generated
3336 from \d etc by PCRE_UCP, this test will have been applied much earlier,
3337 to the original \d etc. At this point, ptr will point to a zero byte. */
3338
3339 if (*ptr == CHAR_ASTERISK || *ptr == CHAR_QUESTION_MARK ||
3340 STRNCMP_UC_C8(ptr, STR_LEFT_CURLY_BRACKET STR_0 STR_COMMA, 3) == 0)
3341 return FALSE;
3342
3343 /* Do the property check. */
3344
3345 return check_char_prop(c, ptype, pdata, (next == -ESC_P) != negated);
3346 }
3347 #endif
3348
3349 default:
3350 return FALSE;
3351 }
3352
3353 /* In principle, support for Unicode properties should be integrated here as
3354 well. It means re-organizing the above code so as to get hold of the property
3355 values before switching on the op-code. However, I wonder how many patterns
3356 combine ASCII \d etc with Unicode properties? (Note that if PCRE_UCP is set,
3357 these op-codes are never generated.) */
3358
3359 case OP_DIGIT:
3360 return next == -ESC_D || next == -ESC_s || next == -ESC_W ||
3361 next == -ESC_h || next == -ESC_v || next == -ESC_R;
3362
3363 case OP_NOT_DIGIT:
3364 return next == -ESC_d;
3365
3366 case OP_WHITESPACE:
3367 return next == -ESC_S || next == -ESC_d || next == -ESC_w;
3368
3369 case OP_NOT_WHITESPACE:
3370 return next == -ESC_s || next == -ESC_h || next == -ESC_v || next == -ESC_R;
3371
3372 case OP_HSPACE:
3373 return next == -ESC_S || next == -ESC_H || next == -ESC_d ||
3374 next == -ESC_w || next == -ESC_v || next == -ESC_R;
3375
3376 case OP_NOT_HSPACE:
3377 return next == -ESC_h;
3378
3379 /* Can't have \S in here because VT matches \S (Perl anomaly) */
3380 case OP_ANYNL:
3381 case OP_VSPACE:
3382 return next == -ESC_V || next == -ESC_d || next == -ESC_w;
3383
3384 case OP_NOT_VSPACE:
3385 return next == -ESC_v || next == -ESC_R;
3386
3387 case OP_WORDCHAR:
3388 return next == -ESC_W || next == -ESC_s || next == -ESC_h ||
3389 next == -ESC_v || next == -ESC_R;
3390
3391 case OP_NOT_WORDCHAR:
3392 return next == -ESC_w || next == -ESC_d;
3393
3394 default:
3395 return FALSE;
3396 }
3397
3398 /* Control does not reach here */
3399 }
3400
3401
3402
3403 /*************************************************
3404 * Compile one branch *
3405 *************************************************/
3406
3407 /* Scan the pattern, compiling it into the a vector. If the options are
3408 changed during the branch, the pointer is used to change the external options
3409 bits. This function is used during the pre-compile phase when we are trying
3410 to find out the amount of memory needed, as well as during the real compile
3411 phase. The value of lengthptr distinguishes the two phases.
3412
3413 Arguments:
3414 optionsptr pointer to the option bits
3415 codeptr points to the pointer to the current code point
3416 ptrptr points to the current pattern pointer
3417 errorcodeptr points to error code variable
3418 firstcharptr set to initial literal character, or < 0 (REQ_UNSET, REQ_NONE)
3419 reqcharptr set to the last literal character required, else < 0
3420 bcptr points to current branch chain
3421 cond_depth conditional nesting depth
3422 cd contains pointers to tables etc.
3423 lengthptr NULL during the real compile phase
3424 points to length accumulator during pre-compile phase
3425
3426 Returns: TRUE on success
3427 FALSE, with *errorcodeptr set non-zero on error
3428 */
3429
3430 static BOOL
3431 compile_branch(int *optionsptr, pcre_uchar **codeptr,
3432 const pcre_uchar **ptrptr, int *errorcodeptr, pcre_int32 *firstcharptr,
3433 pcre_int32 *reqcharptr, branch_chain *bcptr, int cond_depth,
3434 compile_data *cd, int *lengthptr)
3435 {
3436 int repeat_type, op_type;
3437 int repeat_min = 0, repeat_max = 0; /* To please picky compilers */
3438 int bravalue = 0;
3439 int greedy_default, greedy_non_default;
3440 pcre_int32 firstchar, reqchar;
3441 pcre_int32 zeroreqchar, zerofirstchar;
3442 pcre_int32 req_caseopt, reqvary, tempreqvary;
3443 int options = *optionsptr; /* May change dynamically */
3444 int after_manual_callout = 0;
3445 int length_prevgroup = 0;
3446 register int c;
3447 register pcre_uchar *code = *codeptr;
3448 pcre_uchar *last_code = code;
3449 pcre_uchar *orig_code = code;
3450 pcre_uchar *tempcode;
3451 BOOL inescq = FALSE;
3452 BOOL groupsetfirstchar = FALSE;
3453 const pcre_uchar *ptr = *ptrptr;
3454 const pcre_uchar *tempptr;
3455 const pcre_uchar *nestptr = NULL;
3456 pcre_uchar *previous = NULL;
3457 pcre_uchar *previous_callout = NULL;
3458 pcre_uchar *save_hwm = NULL;
3459 pcre_uint8 classbits[32];
3460
3461 /* We can fish out the UTF-8 setting once and for all into a BOOL, but we
3462 must not do this for other options (e.g. PCRE_EXTENDED) because they may change
3463 dynamically as we process the pattern. */
3464
3465 #ifdef SUPPORT_UTF
3466 /* PCRE_UTF16 has the same value as PCRE_UTF8. */
3467 BOOL utf = (options & PCRE_UTF8) != 0;
3468 pcre_uchar utf_chars[6];
3469 #else
3470 BOOL utf = FALSE;
3471 #endif
3472
3473 /* Helper variables for OP_XCLASS opcode (for characters > 255). */
3474
3475 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
3476 BOOL xclass;
3477 pcre_uchar *class_uchardata;
3478 pcre_uchar *class_uchardata_base;
3479 #endif
3480
3481 #ifdef PCRE_DEBUG
3482 if (lengthptr != NULL) DPRINTF((">> start branch\n"));
3483 #endif
3484
3485 /* Set up the default and non-default settings for greediness */
3486
3487 greedy_default = ((options & PCRE_UNGREEDY) != 0);
3488 greedy_non_default = greedy_default ^ 1;
3489
3490 /* Initialize no first byte, no required byte. REQ_UNSET means "no char
3491 matching encountered yet". It gets changed to REQ_NONE if we hit something that
3492 matches a non-fixed char first char; reqchar just remains unset if we never
3493 find one.
3494
3495 When we hit a repeat whose minimum is zero, we may have to adjust these values
3496 to take the zero repeat into account. This is implemented by setting them to
3497 zerofirstbyte and zeroreqchar when such a repeat is encountered. The individual
3498 item types that can be repeated set these backoff variables appropriately. */
3499
3500 firstchar = reqchar = zerofirstchar = zeroreqchar = REQ_UNSET;
3501
3502 /* The variable req_caseopt contains either the REQ_CASELESS value
3503 or zero, according to the current setting of the caseless flag. The
3504 REQ_CASELESS leaves the lower 28 bit empty. It is added into the
3505 firstchar or reqchar variables to record the case status of the
3506 value. This is used only for ASCII characters. */
3507
3508 req_caseopt = ((options & PCRE_CASELESS) != 0)? REQ_CASELESS:0;
3509
3510 /* Switch on next character until the end of the branch */
3511
3512 for (;; ptr++)
3513 {
3514 BOOL negate_class;
3515 BOOL should_flip_negation;
3516 BOOL possessive_quantifier;
3517 BOOL is_quantifier;
3518 BOOL is_recurse;
3519 BOOL reset_bracount;
3520 int class_has_8bitchar;
3521 int class_single_char;
3522 int newoptions;
3523 int recno;
3524 int refsign;
3525 int skipbytes;
3526 int subreqchar;
3527 int subfirstchar;
3528 int terminator;
3529 int mclength;
3530 int tempbracount;
3531 pcre_uchar mcbuffer[8];
3532
3533 /* Get next character in the pattern */
3534
3535 c = *ptr;
3536
3537 /* If we are at the end of a nested substitution, revert to the outer level
3538 string. Nesting only happens one level deep. */
3539
3540 if (c == 0 && nestptr != NULL)
3541 {
3542 ptr = nestptr;
3543 nestptr = NULL;
3544 c = *ptr;
3545 }
3546
3547 /* If we are in the pre-compile phase, accumulate the length used for the
3548 previous cycle of this loop. */
3549
3550 if (lengthptr != NULL)
3551 {
3552 #ifdef PCRE_DEBUG
3553 if (code > cd->hwm) cd->hwm = code; /* High water info */
3554 #endif
3555 if (code > cd->start_workspace + cd->workspace_size -
3556 WORK_SIZE_SAFETY_MARGIN) /* Check for overrun */
3557 {
3558 *errorcodeptr = ERR52;
3559 goto FAILED;
3560 }
3561
3562 /* There is at least one situation where code goes backwards: this is the
3563 case of a zero quantifier after a class (e.g. [ab]{0}). At compile time,
3564 the class is simply eliminated. However, it is created first, so we have to
3565 allow memory for it. Therefore, don't ever reduce the length at this point.
3566 */
3567
3568 if (code < last_code) code = last_code;
3569
3570 /* Paranoid check for integer overflow */
3571
3572 if (OFLOW_MAX - *lengthptr < code - last_code)
3573 {
3574 *errorcodeptr = ERR20;
3575 goto FAILED;
3576 }
3577
3578 *lengthptr += (int)(code - last_code);
3579 DPRINTF(("length=%d added %d c=%c (0x%x)\n", *lengthptr,
3580 (int)(code - last_code), c, c));
3581
3582 /* If "previous" is set and it is not at the start of the work space, move
3583 it back to there, in order to avoid filling up the work space. Otherwise,
3584 if "previous" is NULL, reset the current code pointer to the start. */
3585
3586 if (previous != NULL)
3587 {
3588 if (previous > orig_code)
3589 {
3590 memmove(orig_code, previous, IN_UCHARS(code - previous));
3591 code -= previous - orig_code;
3592 previous = orig_code;
3593 }
3594 }
3595 else code = orig_code;
3596
3597 /* Remember where this code item starts so we can pick up the length
3598 next time round. */
3599
3600 last_code = code;
3601 }
3602
3603 /* In the real compile phase, just check the workspace used by the forward
3604 reference list. */
3605
3606 else if (cd->hwm > cd->start_workspace + cd->workspace_size -
3607 WORK_SIZE_SAFETY_MARGIN)
3608 {
3609 *errorcodeptr = ERR52;
3610 goto FAILED;
3611 }
3612
3613 /* If in \Q...\E, check for the end; if not, we have a literal */
3614
3615 if (inescq && c != 0)
3616 {
3617 if (c == CHAR_BACKSLASH && ptr[1] == CHAR_E)
3618 {
3619 inescq = FALSE;
3620 ptr++;
3621 continue;
3622 }
3623 else
3624 {
3625 if (previous_callout != NULL)
3626 {
3627 if (lengthptr == NULL) /* Don't attempt in pre-compile phase */
3628 complete_callout(previous_callout, ptr, cd);
3629 previous_callout = NULL;
3630 }
3631 if ((options & PCRE_AUTO_CALLOUT) != 0)
3632 {
3633 previous_callout = code;
3634 code = auto_callout(code, ptr, cd);
3635 }
3636 goto NORMAL_CHAR;
3637 }
3638 }
3639
3640 /* Fill in length of a previous callout, except when the next thing is
3641 a quantifier. */
3642
3643 is_quantifier =
3644 c == CHAR_ASTERISK || c == CHAR_PLUS || c == CHAR_QUESTION_MARK ||
3645 (c == CHAR_LEFT_CURLY_BRACKET && is_counted_repeat(ptr+1));
3646
3647 if (!is_quantifier && previous_callout != NULL &&
3648 after_manual_callout-- <= 0)
3649 {
3650 if (lengthptr == NULL) /* Don't attempt in pre-compile phase */
3651 complete_callout(previous_callout, ptr, cd);
3652 previous_callout = NULL;
3653 }
3654
3655 /* In extended mode, skip white space and comments. */
3656
3657 if ((options & PCRE_EXTENDED) != 0)
3658 {
3659 if (MAX_255(*ptr) && (cd->ctypes[c] & ctype_space) != 0) continue;
3660 if (c == CHAR_NUMBER_SIGN)
3661 {
3662 ptr++;
3663 while (*ptr != 0)
3664 {
3665 if (IS_NEWLINE(ptr)) { ptr += cd->nllen - 1; break; }
3666 ptr++;
3667 #ifdef SUPPORT_UTF
3668 if (utf) FORWARDCHAR(ptr);
3669 #endif
3670 }
3671 if (*ptr != 0) continue;
3672
3673 /* Else fall through to handle end of string */
3674 c = 0;
3675 }
3676 }
3677
3678 /* No auto callout for quantifiers. */
3679
3680 if ((options & PCRE_AUTO_CALLOUT) != 0 && !is_quantifier)
3681 {
3682 previous_callout = code;
3683 code = auto_callout(code, ptr, cd);
3684 }
3685
3686 switch(c)
3687 {
3688 /* ===================================================================*/
3689 case 0: /* The branch terminates at string end */
3690 case CHAR_VERTICAL_LINE: /* or | or ) */
3691 case CHAR_RIGHT_PARENTHESIS:
3692 *firstcharptr = firstchar;
3693 *reqcharptr = reqchar;
3694 *codeptr = code;
3695 *ptrptr = ptr;
3696 if (lengthptr != NULL)
3697 {
3698 if (OFLOW_MAX - *lengthptr < code - last_code)
3699 {
3700 *errorcodeptr = ERR20;
3701 goto FAILED;
3702 }
3703 *lengthptr += (int)(code - last_code); /* To include callout length */
3704 DPRINTF((">> end branch\n"));
3705 }
3706 return TRUE;
3707
3708
3709 /* ===================================================================*/
3710 /* Handle single-character metacharacters. In multiline mode, ^ disables
3711 the setting of any following char as a first character. */
3712
3713 case CHAR_CIRCUMFLEX_ACCENT:
3714 previous = NULL;
3715 if ((options & PCRE_MULTILINE) != 0)
3716 {
3717 if (firstchar == REQ_UNSET) firstchar = REQ_NONE;
3718 *code++ = OP_CIRCM;
3719 }
3720 else *code++ = OP_CIRC;
3721 break;
3722
3723 case CHAR_DOLLAR_SIGN:
3724 previous = NULL;
3725 *code++ = ((options & PCRE_MULTILINE) != 0)? OP_DOLLM : OP_DOLL;
3726 break;
3727
3728 /* There can never be a first char if '.' is first, whatever happens about
3729 repeats. The value of reqchar doesn't change either. */
3730
3731 case CHAR_DOT:
3732 if (firstchar == REQ_UNSET) firstchar = REQ_NONE;
3733 zerofirstchar = firstchar;
3734 zeroreqchar = reqchar;
3735 previous = code;
3736 *code++ = ((options & PCRE_DOTALL) != 0)? OP_ALLANY: OP_ANY;
3737 break;
3738
3739
3740 /* ===================================================================*/
3741 /* Character classes. If the included characters are all < 256, we build a
3742 32-byte bitmap of the permitted characters, except in the special case
3743 where there is only one such character. For negated classes, we build the
3744 map as usual, then invert it at the end. However, we use a different opcode
3745 so that data characters > 255 can be handled correctly.
3746
3747 If the class contains characters outside the 0-255 range, a different
3748 opcode is compiled. It may optionally have a bit map for characters < 256,
3749 but those above are are explicitly listed afterwards. A flag byte tells
3750 whether the bitmap is present, and whether this is a negated class or not.
3751
3752 In JavaScript compatibility mode, an isolated ']' causes an error. In
3753 default (Perl) mode, it is treated as a data character. */
3754
3755 case CHAR_RIGHT_SQUARE_BRACKET:
3756 if ((cd->external_options & PCRE_JAVASCRIPT_COMPAT) != 0)
3757 {
3758 *errorcodeptr = ERR64;
3759 goto FAILED;
3760 }
3761 goto NORMAL_CHAR;
3762
3763 case CHAR_LEFT_SQUARE_BRACKET:
3764 previous = code;
3765
3766 /* PCRE supports POSIX class stuff inside a class. Perl gives an error if
3767 they are encountered at the top level, so we'll do that too. */
3768
3769 if ((ptr[1] == CHAR_COLON || ptr[1] == CHAR_DOT ||
3770 ptr[1] == CHAR_EQUALS_SIGN) &&
3771 check_posix_syntax(ptr, &tempptr))
3772 {
3773 *errorcodeptr = (ptr[1] == CHAR_COLON)? ERR13 : ERR31;
3774 goto FAILED;
3775 }
3776
3777 /* If the first character is '^', set the negation flag and skip it. Also,
3778 if the first few characters (either before or after ^) are \Q\E or \E we
3779 skip them too. This makes for compatibility with Perl. */
3780
3781 negate_class = FALSE;
3782 for (;;)
3783 {
3784 c = *(++ptr);
3785 if (c == CHAR_BACKSLASH)
3786 {
3787 if (ptr[1] == CHAR_E)
3788 ptr++;
3789 else if (STRNCMP_UC_C8(ptr + 1, STR_Q STR_BACKSLASH STR_E, 3) == 0)
3790 ptr += 3;
3791 else
3792 break;
3793 }
3794 else if (!negate_class && c == CHAR_CIRCUMFLEX_ACCENT)
3795 negate_class = TRUE;
3796 else break;
3797 }
3798
3799 /* Empty classes are allowed in JavaScript compatibility mode. Otherwise,
3800 an initial ']' is taken as a data character -- the code below handles
3801 that. In JS mode, [] must always fail, so generate OP_FAIL, whereas
3802 [^] must match any character, so generate OP_ALLANY. */
3803
3804 if (c == CHAR_RIGHT_SQUARE_BRACKET &&
3805 (cd->external_options & PCRE_JAVASCRIPT_COMPAT) != 0)
3806 {
3807 *code++ = negate_class? OP_ALLANY : OP_FAIL;
3808 if (firstchar == REQ_UNSET) firstchar = REQ_NONE;
3809 zerofirstchar = firstchar;
3810 break;
3811 }
3812
3813 /* If a class contains a negative special such as \S, we need to flip the
3814 negation flag at the end, so that support for characters > 255 works
3815 correctly (they are all included in the class). */
3816
3817 should_flip_negation = FALSE;
3818
3819 /* For optimization purposes, we track some properties of the class.
3820 class_has_8bitchar will be non-zero, if the class contains at least one
3821 < 256 character. class_single_char will be 1 if the class contains only
3822 a single character. */
3823
3824 class_has_8bitchar = 0;
3825 class_single_char = 0;
3826
3827 /* Initialize the 32-char bit map to all zeros. We build the map in a
3828 temporary bit of memory, in case the class contains only 1 character (less
3829 than 256), because in that case the compiled code doesn't use the bit map.
3830 */
3831
3832 memset(classbits, 0, 32 * sizeof(pcre_uint8));
3833
3834 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
3835 xclass = FALSE; /* No chars >= 256 */
3836 class_uchardata = code + LINK_SIZE + 2; /* For UTF-8 items */
3837 class_uchardata_base = class_uchardata; /* For resetting in pass 1 */
3838 #endif
3839
3840 /* Process characters until ] is reached. By writing this as a "do" it
3841 means that an initial ] is taken as a data character. At the start of the
3842 loop, c contains the first byte of the character. */
3843
3844 if (c != 0) do
3845 {
3846 const pcre_uchar *oldptr;
3847
3848 #ifdef SUPPORT_UTF
3849 if (utf && HAS_EXTRALEN(c))
3850 { /* Braces are required because the */
3851 GETCHARLEN(c, ptr, ptr); /* macro generates multiple statements */
3852 }
3853 #endif
3854
3855 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
3856 /* In the pre-compile phase, accumulate the length of any extra
3857 data and reset the pointer. This is so that very large classes that
3858 contain a zillion > 255 characters no longer overwrite the work space
3859 (which is on the stack). */
3860
3861 if (lengthptr != NULL)
3862 {
3863 *lengthptr += class_uchardata - class_uchardata_base;
3864 class_uchardata = class_uchardata_base;
3865 }
3866 #endif
3867
3868 /* Inside \Q...\E everything is literal except \E */
3869
3870 if (inescq)
3871 {
3872 if (c == CHAR_BACKSLASH && ptr[1] == CHAR_E) /* If we are at \E */
3873 {
3874 inescq = FALSE; /* Reset literal state */
3875 ptr++; /* Skip the 'E' */
3876 continue; /* Carry on with next */
3877 }
3878 goto CHECK_RANGE; /* Could be range if \E follows */
3879 }
3880
3881 /* Handle POSIX class names. Perl allows a negation extension of the
3882 form [:^name:]. A square bracket that doesn't match the syntax is
3883 treated as a literal. We also recognize the POSIX constructions
3884 [.ch.] and [=ch=] ("collating elements") and fault them, as Perl
3885 5.6 and 5.8 do. */
3886
3887 if (c == CHAR_LEFT_SQUARE_BRACKET &&
3888 (ptr[1] == CHAR_COLON || ptr[1] == CHAR_DOT ||
3889 ptr[1] == CHAR_EQUALS_SIGN) && check_posix_syntax(ptr, &tempptr))
3890 {
3891 BOOL local_negate = FALSE;
3892 int posix_class, taboffset, tabopt;
3893 register const pcre_uint8 *cbits = cd->cbits;
3894 pcre_uint8 pbits[32];
3895
3896 if (ptr[1] != CHAR_COLON)
3897 {
3898 *errorcodeptr = ERR31;
3899 goto FAILED;
3900 }
3901
3902 ptr += 2;
3903 if (*ptr == CHAR_CIRCUMFLEX_ACCENT)
3904 {
3905 local_negate = TRUE;
3906 should_flip_negation = TRUE; /* Note negative special */
3907 ptr++;
3908 }
3909
3910 posix_class = check_posix_name(ptr, (int)(tempptr - ptr));
3911 if (posix_class < 0)
3912 {
3913 *errorcodeptr = ERR30;
3914 goto FAILED;
3915 }
3916
3917 /* If matching is caseless, upper and lower are converted to
3918 alpha. This relies on the fact that the class table starts with
3919 alpha, lower, upper as the first 3 entries. */
3920
3921 if ((options & PCRE_CASELESS) != 0 && posix_class <= 2)
3922 posix_class = 0;
3923
3924 /* When PCRE_UCP is set, some of the POSIX classes are converted to
3925 different escape sequences that use Unicode properties. */
3926
3927 #ifdef SUPPORT_UCP
3928 if ((options & PCRE_UCP) != 0)
3929 {
3930 int pc = posix_class + ((local_negate)? POSIX_SUBSIZE/2 : 0);
3931 if (posix_substitutes[pc] != NULL)
3932 {
3933 nestptr = tempptr + 1;
3934 ptr = posix_substitutes[pc] - 1;
3935 continue;
3936 }
3937 }
3938 #endif
3939 /* In the non-UCP case, we build the bit map for the POSIX class in a
3940 chunk of local store because we may be adding and subtracting from it,
3941 and we don't want to subtract bits that may be in the main map already.
3942 At the end we or the result into the bit map that is being built. */
3943
3944 posix_class *= 3;
3945
3946 /* Copy in the first table (always present) */
3947
3948 memcpy(pbits, cbits + posix_class_maps[posix_class],
3949 32 * sizeof(pcre_uint8));
3950
3951 /* If there is a second table, add or remove it as required. */
3952
3953 taboffset = posix_class_maps[posix_class + 1];
3954 tabopt = posix_class_maps[posix_class + 2];
3955
3956 if (taboffset >= 0)
3957 {
3958 if (tabopt >= 0)
3959 for (c = 0; c < 32; c++) pbits[c] |= cbits[c + taboffset];
3960 else
3961 for (c = 0; c < 32; c++) pbits[c] &= ~cbits[c + taboffset];
3962 }
3963
3964 /* Not see if we need to remove any special characters. An option
3965 value of 1 removes vertical space and 2 removes underscore. */
3966
3967 if (tabopt < 0) tabopt = -tabopt;
3968 if (tabopt == 1) pbits[1] &= ~0x3c;
3969 else if (tabopt == 2) pbits[11] &= 0x7f;
3970
3971 /* Add the POSIX table or its complement into the main table that is
3972 being built and we are done. */
3973
3974 if (local_negate)
3975 for (c = 0; c < 32; c++) classbits[c] |= ~pbits[c];
3976 else
3977 for (c = 0; c < 32; c++) classbits[c] |= pbits[c];
3978
3979 ptr = tempptr + 1;
3980 /* Every class contains at least one < 256 characters. */
3981 class_has_8bitchar = 1;
3982 /* Every class contains at least two characters. */
3983 class_single_char = 2;
3984 continue; /* End of POSIX syntax handling */
3985 }
3986
3987 /* Backslash may introduce a single character, or it may introduce one
3988 of the specials, which just set a flag. The sequence \b is a special
3989 case. Inside a class (and only there) it is treated as backspace. We
3990 assume that other escapes have more than one character in them, so
3991 speculatively set both class_has_8bitchar and class_single_char bigger
3992 than one. Unrecognized escapes fall through and are either treated
3993 as literal characters (by default), or are faulted if
3994 PCRE_EXTRA is set. */
3995
3996 if (c == CHAR_BACKSLASH)
3997 {
3998 c = check_escape(&ptr, errorcodeptr, cd->bracount, options, TRUE);
3999 if (*errorcodeptr != 0) goto FAILED;
4000
4001 if (-c == ESC_b) c = CHAR_BS; /* \b is backspace in a class */
4002 else if (-c == ESC_N) /* \N is not supported in a class */
4003 {
4004 *errorcodeptr = ERR71;
4005 goto FAILED;
4006 }
4007 else if (-c == ESC_Q) /* Handle start of quoted string */
4008 {
4009 if (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
4010 {
4011 ptr += 2; /* avoid empty string */
4012 }
4013 else inescq = TRUE;
4014 continue;
4015 }
4016 else if (-c == ESC_E) continue; /* Ignore orphan \E */
4017
4018 if (c < 0)
4019 {
4020 register const pcre_uint8 *cbits = cd->cbits;
4021 /* Every class contains at least two < 256 characters. */
4022 class_has_8bitchar++;
4023 /* Every class contains at least two characters. */
4024 class_single_char += 2;
4025
4026 switch (-c)
4027 {
4028 #ifdef SUPPORT_UCP
4029 case ESC_du: /* These are the values given for \d etc */
4030 case ESC_DU: /* when PCRE_UCP is set. We replace the */
4031 case ESC_wu: /* escape sequence with an appropriate \p */
4032 case ESC_WU: /* or \P to test Unicode properties instead */
4033 case ESC_su: /* of the default ASCII testing. */
4034 case ESC_SU:
4035 nestptr = ptr;
4036 ptr = substitutes[-c - ESC_DU] - 1; /* Just before substitute */
4037 class_has_8bitchar--; /* Undo! */
4038 continue;
4039 #endif
4040 case ESC_d:
4041 for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_digit];
4042 continue;
4043
4044 case ESC_D:
4045 should_flip_negation = TRUE;
4046 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_digit];
4047 continue;
4048
4049 case ESC_w:
4050 for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_word];
4051 continue;
4052
4053 case ESC_W:
4054 should_flip_negation = TRUE;
4055 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_word];
4056 continue;
4057
4058 /* Perl 5.004 onwards omits VT from \s, but we must preserve it
4059 if it was previously set by something earlier in the character
4060 class. */
4061
4062 case ESC_s:
4063 classbits[0] |= cbits[cbit_space];
4064 classbits[1] |= cbits[cbit_space+1] & ~0x08;
4065 for (c = 2; c < 32; c++) classbits[c] |= cbits[c+cbit_space];
4066 continue;
4067
4068 case ESC_S:
4069 should_flip_negation = TRUE;
4070 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_space];
4071 classbits[1] |= 0x08; /* Perl 5.004 onwards omits VT from \s */
4072 continue;
4073
4074 case ESC_h:
4075 SETBIT(classbits, 0x09); /* VT */
4076 SETBIT(classbits, 0x20); /* SPACE */
4077 SETBIT(classbits, 0xa0); /* NSBP */
4078 #ifndef COMPILE_PCRE8
4079 xclass = TRUE;
4080 *class_uchardata++ = XCL_SINGLE;
4081 *class_uchardata++ = 0x1680;
4082 *class_uchardata++ = XCL_SINGLE;
4083 *class_uchardata++ = 0x180e;
4084 *class_uchardata++ = XCL_RANGE;
4085 *class_uchardata++ = 0x2000;
4086 *class_uchardata++ = 0x200a;
4087 *class_uchardata++ = XCL_SINGLE;
4088 *class_uchardata++ = 0x202f;
4089 *class_uchardata++ = XCL_SINGLE;
4090 *class_uchardata++ = 0x205f;
4091 *class_uchardata++ = XCL_SINGLE;
4092 *class_uchardata++ = 0x3000;
4093 #elif defined SUPPORT_UTF
4094 if (utf)
4095 {
4096 xclass = TRUE;
4097 *class_uchardata++ = XCL_SINGLE;
4098 class_uchardata += PRIV(ord2utf)(0x1680, class_uchardata);
4099 *class_uchardata++ = XCL_SINGLE;
4100 class_uchardata += PRIV(ord2utf)(0x180e, class_uchardata);
4101 *class_uchardata++ = XCL_RANGE;
4102 class_uchardata += PRIV(ord2utf)(0x2000, class_uchardata);
4103 class_uchardata += PRIV(ord2utf)(0x200a, class_uchardata);
4104 *class_uchardata++ = XCL_SINGLE;
4105 class_uchardata += PRIV(ord2utf)(0x202f, class_uchardata);
4106 *class_uchardata++ = XCL_SINGLE;
4107 class_uchardata += PRIV(ord2utf)(0x205f, class_uchardata);
4108 *class_uchardata++ = XCL_SINGLE;
4109 class_uchardata += PRIV(ord2utf)(0x3000, class_uchardata);
4110 }
4111 #endif
4112 continue;
4113
4114 case ESC_H:
4115 for (c = 0; c < 32; c++)
4116 {
4117 int x = 0xff;
4118 switch (c)
4119 {
4120 case 0x09/8: x ^= 1 << (0x09%8); break;
4121 case 0x20/8: x ^= 1 << (0x20%8); break;
4122 case 0xa0/8: x ^= 1 << (0xa0%8); break;
4123 default: break;
4124 }
4125 classbits[c] |= x;
4126 }
4127 #ifndef COMPILE_PCRE8
4128 xclass = TRUE;
4129 *class_uchardata++ = XCL_RANGE;
4130 *class_uchardata++ = 0x0100;
4131 *class_uchardata++ = 0x167f;
4132 *class_uchardata++ = XCL_RANGE;
4133 *class_uchardata++ = 0x1681;
4134 *class_uchardata++ = 0x180d;
4135 *class_uchardata++ = XCL_RANGE;
4136 *class_uchardata++ = 0x180f;
4137 *class_uchardata++ = 0x1fff;
4138 *class_uchardata++ = XCL_RANGE;
4139 *class_uchardata++ = 0x200b;
4140 *class_uchardata++ = 0x202e;
4141 *class_uchardata++ = XCL_RANGE;
4142 *class_uchardata++ = 0x2030;
4143 *class_uchardata++ = 0x205e;
4144 *class_uchardata++ = XCL_RANGE;
4145 *class_uchardata++ = 0x2060;
4146 *class_uchardata++ = 0x2fff;
4147 *class_uchardata++ = XCL_RANGE;
4148 *class_uchardata++ = 0x3001;
4149 #ifdef SUPPORT_UTF
4150 if (utf)
4151 class_uchardata += PRIV(ord2utf)(0x10ffff, class_uchardata);
4152 else
4153 #endif
4154 *class_uchardata++ = 0xffff;
4155 #elif defined SUPPORT_UTF
4156 if (utf)
4157 {
4158 xclass = TRUE;
4159 *class_uchardata++ = XCL_RANGE;
4160 class_uchardata += PRIV(ord2utf)(0x0100, class_uchardata);
4161 class_uchardata += PRIV(ord2utf)(0x167f, class_uchardata);
4162 *class_uchardata++ = XCL_RANGE;
4163 class_uchardata += PRIV(ord2utf)(0x1681, class_uchardata);
4164 class_uchardata += PRIV(ord2utf)(0x180d, class_uchardata);
4165 *class_uchardata++ = XCL_RANGE;
4166 class_uchardata += PRIV(ord2utf)(0x180f, class_uchardata);
4167 class_uchardata += PRIV(ord2utf)(0x1fff, class_uchardata);
4168 *class_uchardata++ = XCL_RANGE;
4169 class_uchardata += PRIV(ord2utf)(0x200b, class_uchardata);
4170 class_uchardata += PRIV(ord2utf)(0x202e, class_uchardata);
4171 *class_uchardata++ = XCL_RANGE;
4172 class_uchardata += PRIV(ord2utf)(0x2030, class_uchardata);
4173 class_uchardata += PRIV(ord2utf)(0x205e, class_uchardata);
4174 *class_uchardata++ = XCL_RANGE;
4175 class_uchardata += PRIV(ord2utf)(0x2060, class_uchardata);
4176 class_uchardata += PRIV(ord2utf)(0x2fff, class_uchardata);
4177 *class_uchardata++ = XCL_RANGE;
4178 class_uchardata += PRIV(ord2utf)(0x3001, class_uchardata);
4179 class_uchardata += PRIV(ord2utf)(0x10ffff, class_uchardata);
4180 }
4181 #endif
4182 continue;
4183
4184 case ESC_v:
4185 SETBIT(classbits, 0x0a); /* LF */
4186 SETBIT(classbits, 0x0b); /* VT */
4187 SETBIT(classbits, 0x0c); /* FF */
4188 SETBIT(classbits, 0x0d); /* CR */
4189 SETBIT(classbits, 0x85); /* NEL */
4190 #ifndef COMPILE_PCRE8
4191 xclass = TRUE;
4192 *class_uchardata++ = XCL_RANGE;
4193 *class_uchardata++ = 0x2028;
4194 *class_uchardata++ = 0x2029;
4195 #elif defined SUPPORT_UTF
4196 if (utf)
4197 {
4198 xclass = TRUE;
4199 *class_uchardata++ = XCL_RANGE;
4200 class_uchardata += PRIV(ord2utf)(0x2028, class_uchardata);
4201 class_uchardata += PRIV(ord2utf)(0x2029, class_uchardata);
4202 }
4203 #endif
4204 continue;
4205
4206 case ESC_V:
4207 for (c = 0; c < 32; c++)
4208 {
4209 int x = 0xff;
4210 switch (c)
4211 {
4212 case 0x0a/8: x ^= 1 << (0x0a%8);
4213 x ^= 1 << (0x0b%8);
4214 x ^= 1 << (0x0c%8);
4215 x ^= 1 << (0x0d%8);
4216 break;
4217 case 0x85/8: x ^= 1 << (0x85%8); break;
4218 default: break;
4219 }
4220 classbits[c] |= x;
4221 }
4222
4223 #ifndef COMPILE_PCRE8
4224 xclass = TRUE;
4225 *class_uchardata++ = XCL_RANGE;
4226 *class_uchardata++ = 0x0100;
4227 *class_uchardata++ = 0x2027;
4228 *class_uchardata++ = XCL_RANGE;
4229 *class_uchardata++ = 0x202a;
4230 #ifdef SUPPORT_UTF
4231 if (utf)
4232 class_uchardata += PRIV(ord2utf)(0x10ffff, class_uchardata);
4233 else
4234 #endif
4235 *class_uchardata++ = 0xffff;
4236 #elif defined SUPPORT_UTF
4237 if (utf)
4238 {
4239 xclass = TRUE;
4240 *class_uchardata++ = XCL_RANGE;
4241 class_uchardata += PRIV(ord2utf)(0x0100, class_uchardata);
4242 class_uchardata += PRIV(ord2utf)(0x2027, class_uchardata);
4243 *class_uchardata++ = XCL_RANGE;
4244 class_uchardata += PRIV(ord2utf)(0x202a, class_uchardata);
4245 class_uchardata += PRIV(ord2utf)(0x10ffff, class_uchardata);
4246 }
4247 #endif
4248 continue;
4249
4250 #ifdef SUPPORT_UCP
4251 case ESC_p:
4252 case ESC_P:
4253 {
4254 BOOL negated;
4255 int pdata;
4256 int ptype = get_ucp(&ptr, &negated, &pdata, errorcodeptr);
4257 if (ptype < 0) goto FAILED;
4258 xclass = TRUE;
4259 *class_uchardata++ = ((-c == ESC_p) != negated)?
4260 XCL_PROP : XCL_NOTPROP;
4261 *class_uchardata++ = ptype;
4262 *class_uchardata++ = pdata;
4263 class_has_8bitchar--; /* Undo! */
4264 continue;
4265 }
4266 #endif
4267 /* Unrecognized escapes are faulted if PCRE is running in its
4268 strict mode. By default, for compatibility with Perl, they are
4269 treated as literals. */
4270
4271 default:
4272 if ((options & PCRE_EXTRA) != 0)
4273 {
4274 *errorcodeptr = ERR7;
4275 goto FAILED;
4276 }
4277 class_has_8bitchar--; /* Undo the speculative increase. */
4278 class_single_char -= 2; /* Undo the speculative increase. */
4279 c = *ptr; /* Get the final character and fall through */
4280 break;
4281 }
4282 }
4283
4284 /* Fall through if we have a single character (c >= 0). This may be
4285 greater than 256. */
4286
4287 } /* End of backslash handling */
4288
4289 /* A single character may be followed by '-' to form a range. However,
4290 Perl does not permit ']' to be the end of the range. A '-' character
4291 at the end is treated as a literal. Perl ignores orphaned \E sequences
4292 entirely. The code for handling \Q and \E is messy. */
4293
4294 CHECK_RANGE:
4295 while (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
4296 {
4297 inescq = FALSE;
4298 ptr += 2;
4299 }
4300
4301 oldptr = ptr;
4302
4303 /* Remember \r or \n */
4304
4305 if (c == CHAR_CR || c == CHAR_NL) cd->external_flags |= PCRE_HASCRORLF;
4306
4307 /* Check for range */
4308
4309 if (!inescq && ptr[1] == CHAR_MINUS)
4310 {
4311 int d;
4312 ptr += 2;
4313 while (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_E) ptr += 2;
4314
4315 /* If we hit \Q (not followed by \E) at this point, go into escaped
4316 mode. */
4317
4318 while (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_Q)
4319 {
4320 ptr += 2;
4321 if (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_E)
4322 { ptr += 2; continue; }
4323 inescq = TRUE;
4324 break;
4325 }
4326
4327 if (*ptr == 0 || (!inescq && *ptr == CHAR_RIGHT_SQUARE_BRACKET))
4328 {
4329 ptr = oldptr;
4330 goto LONE_SINGLE_CHARACTER;
4331 }
4332
4333 #ifdef SUPPORT_UTF
4334 if (utf)
4335 { /* Braces are required because the */
4336 GETCHARLEN(d, ptr, ptr); /* macro generates multiple statements */
4337 }
4338 else
4339 #endif
4340 d = *ptr; /* Not UTF-8 mode */
4341
4342 /* The second part of a range can be a single-character escape, but
4343 not any of the other escapes. Perl 5.6 treats a hyphen as a literal
4344 in such circumstances. */
4345
4346 if (!inescq && d == CHAR_BACKSLASH)
4347 {
4348 d = check_escape(&ptr, errorcodeptr, cd->bracount, options, TRUE);
4349 if (*errorcodeptr != 0) goto FAILED;
4350
4351 /* \b is backspace; any other special means the '-' was literal */
4352
4353 if (d < 0)
4354 {
4355 if (d == -ESC_b) d = CHAR_BS; else
4356 {
4357 ptr = oldptr;
4358 goto LONE_SINGLE_CHARACTER; /* A few lines below */
4359 }
4360 }
4361 }
4362
4363 /* Check that the two values are in the correct order. Optimize
4364 one-character ranges */
4365
4366 if (d < c)
4367 {
4368 *errorcodeptr = ERR8;
4369 goto FAILED;
4370 }
4371
4372 if (d == c) goto LONE_SINGLE_CHARACTER; /* A few lines below */
4373
4374 /* Remember \r or \n */
4375
4376 if (d == CHAR_CR || d == CHAR_NL) cd->external_flags |= PCRE_HASCRORLF;
4377
4378 /* Since we found a character range, single character optimizations
4379 cannot be done anymore. */
4380 class_single_char = 2;
4381
4382 /* In UTF-8 mode, if the upper limit is > 255, or > 127 for caseless
4383 matching, we have to use an XCLASS with extra data items. Caseless
4384 matching for characters > 127 is available only if UCP support is
4385 available. */
4386
4387 #if defined SUPPORT_UTF && !(defined COMPILE_PCRE8)
4388 if ((d > 255) || (utf && ((options & PCRE_CASELESS) != 0 && d > 127)))
4389 #elif defined SUPPORT_UTF
4390 if (utf && (d > 255 || ((options & PCRE_CASELESS) != 0 && d > 127)))
4391 #elif !(defined COMPILE_PCRE8)
4392 if (d > 255)
4393 #endif
4394 #if defined SUPPORT_UTF || !(defined COMPILE_PCRE8)
4395 {
4396 xclass = TRUE;
4397
4398 /* With UCP support, we can find the other case equivalents of
4399 the relevant characters. There may be several ranges. Optimize how
4400 they fit with the basic range. */
4401
4402 #ifdef SUPPORT_UCP
4403 #ifndef COMPILE_PCRE8
4404 if (utf && (options & PCRE_CASELESS) != 0)
4405 #else
4406 if ((options & PCRE_CASELESS) != 0)
4407 #endif
4408 {
4409 unsigned int occ, ocd;
4410 unsigned int cc = c;
4411 unsigned int origd = d;
4412 while (get_othercase_range(&cc, origd, &occ, &ocd))
4413 {
4414 if (occ >= (unsigned int)c &&
4415 ocd <= (unsigned int)d)
4416 continue; /* Skip embedded ranges */
4417
4418 if (occ < (unsigned int)c &&
4419 ocd >= (unsigned int)c - 1) /* Extend the basic range */
4420 { /* if there is overlap, */
4421 c = occ; /* noting that if occ < c */
4422 continue; /* we can't have ocd > d */
4423 } /* because a subrange is */
4424 if (ocd > (unsigned int)d &&
4425 occ <= (unsigned int)d + 1) /* always shorter than */
4426 { /* the basic range. */
4427 d = ocd;
4428 continue;
4429 }
4430
4431 if (occ == ocd)
4432 {
4433 *class_uchardata++ = XCL_SINGLE;
4434 }
4435 else
4436 {
4437 *class_uchardata++ = XCL_RANGE;
4438 class_uchardata += PRIV(ord2utf)(occ, class_uchardata);
4439 }
4440 class_uchardata += PRIV(ord2utf)(ocd, class_uchardata);
4441 }
4442 }
4443 #endif /* SUPPORT_UCP */
4444
4445 /* Now record the original range, possibly modified for UCP caseless
4446 overlapping ranges. */
4447
4448 *class_uchardata++ = XCL_RANGE;
4449 #ifdef SUPPORT_UTF
4450 #ifndef COMPILE_PCRE8
4451 if (utf)
4452 {
4453 class_uchardata += PRIV(ord2utf)(c, class_uchardata);
4454 class_uchardata += PRIV(ord2utf)(d, class_uchardata);
4455 }
4456 else
4457 {
4458 *class_uchardata++ = c;
4459 *class_uchardata++ = d;
4460 }
4461 #else
4462 class_uchardata += PRIV(ord2utf)(c, class_uchardata);
4463 class_uchardata += PRIV(ord2utf)(d, class_uchardata);
4464 #endif
4465 #else /* SUPPORT_UTF */
4466 *class_uchardata++ = c;
4467 *class_uchardata++ = d;
4468 #endif /* SUPPORT_UTF */
4469
4470 /* With UCP support, we are done. Without UCP support, there is no
4471 caseless matching for UTF characters > 127; we can use the bit map
4472 for the smaller ones. As for 16 bit characters without UTF, we
4473 can still use */
4474
4475 #ifdef SUPPORT_UCP
4476 #ifndef COMPILE_PCRE8
4477 if (utf)
4478 #endif
4479 continue; /* With next character in the class */
4480 #endif /* SUPPORT_UCP */
4481
4482 #if defined SUPPORT_UTF && !defined(SUPPORT_UCP) && !(defined COMPILE_PCRE8)
4483 if (utf)
4484 {
4485 if ((options & PCRE_CASELESS) == 0 || c > 127) continue;
4486 /* Adjust upper limit and fall through to set up the map */
4487 d = 127;
4488 }
4489 else
4490 {
4491 if (c > 255) continue;
4492 /* Adjust upper limit and fall through to set up the map */
4493 d = 255;
4494 }
4495 #elif defined SUPPORT_UTF && !defined(SUPPORT_UCP)
4496 if ((options & PCRE_CASELESS) == 0 || c > 127) continue;
4497 /* Adjust upper limit and fall through to set up the map */
4498 d = 127;
4499 #else
4500 if (c > 255) continue;
4501 /* Adjust upper limit and fall through to set up the map */
4502 d = 255;
4503 #endif /* SUPPORT_UTF && !SUPPORT_UCP && !COMPILE_PCRE8 */
4504 }
4505 #endif /* SUPPORT_UTF || !COMPILE_PCRE8 */
4506
4507 /* We use the bit map for 8 bit mode, or when the characters fall
4508 partially or entirely to [0-255] ([0-127] for UCP) ranges. */
4509
4510 class_has_8bitchar = 1;
4511
4512 /* We can save a bit of time by skipping this in the pre-compile. */
4513
4514 if (lengthptr == NULL) for (; c <= d; c++)
4515 {
4516 classbits[c/8] |= (1 << (c&7));
4517 if ((options & PCRE_CASELESS) != 0)
4518 {
4519 int uc = cd->fcc[c]; /* flip case */
4520 classbits[uc/8] |= (1 << (uc&7));
4521 }
4522 }
4523
4524 continue; /* Go get the next char in the class */
4525 }
4526
4527 /* Handle a lone single character - we can get here for a normal
4528 non-escape char, or after \ that introduces a single character or for an
4529 apparent range that isn't. */
4530
4531 LONE_SINGLE_CHARACTER:
4532
4533 /* Only the value of 1 matters for class_single_char. */
4534
4535 if (class_single_char < 2) class_single_char++;
4536
4537 /* If class_charcount is 1, we saw precisely one character. As long as
4538 there was no use of \p or \P, in other words, no use of any XCLASS
4539 features, we can optimize.
4540
4541 The optimization throws away the bit map. We turn the item into a
4542 1-character OP_CHAR[I] if it's positive, or OP_NOT[I] if it's negative.
4543 In the positive case, it can cause firstchar to be set. Otherwise, there
4544 can be no first char if this item is first, whatever repeat count may
4545 follow. In the case of reqchar, save the previous value for reinstating. */
4546
4547 if (class_single_char == 1 && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET)
4548 {
4549 ptr++;
4550 zeroreqchar = reqchar;
4551
4552 if (negate_class)
4553 {
4554 if (firstchar == REQ_UNSET) firstchar = REQ_NONE;
4555 zerofirstchar = firstchar;
4556 *code++ = ((options & PCRE_CASELESS) != 0)? OP_NOTI: OP_NOT;
4557 #ifdef SUPPORT_UTF
4558 if (utf && c > MAX_VALUE_FOR_SINGLE_CHAR)
4559 code += PRIV(ord2utf)(c, code);
4560 else
4561 #endif
4562 *code++ = c;
4563 goto NOT_CHAR;
4564 }
4565
4566 /* For a single, positive character, get the value into mcbuffer, and
4567 then we can handle this with the normal one-character code. */
4568
4569 #ifdef SUPPORT_UTF
4570 if (utf && c > MAX_VALUE_FOR_SINGLE_CHAR)
4571 mclength = PRIV(ord2utf)(c, mcbuffer);
4572 else
4573 #endif
4574 {
4575 mcbuffer[0] = c;
4576 mclength = 1;
4577 }
4578 goto ONE_CHAR;
4579 } /* End of 1-char optimization */
4580
4581 /* Handle a character that cannot go in the bit map. */
4582
4583 #if defined SUPPORT_UTF && !(defined COMPILE_PCRE8)
4584 if ((c > 255) || (utf && ((options & PCRE_CASELESS) != 0 && c > 127)))
4585 #elif defined SUPPORT_UTF
4586 if (utf && (c > 255 || ((options & PCRE_CASELESS) != 0 && c > 127)))
4587 #elif !(defined COMPILE_PCRE8)
4588 if (c > 255)
4589 #endif
4590
4591 #if defined SUPPORT_UTF || !(defined COMPILE_PCRE8)
4592 {
4593 xclass = TRUE;
4594 *class_uchardata++ = XCL_SINGLE;
4595 #ifdef SUPPORT_UTF
4596 #ifndef COMPILE_PCRE8
4597 /* In non 8 bit mode, we can get here even if we are not in UTF mode. */
4598 if (!utf)
4599 *class_uchardata++ = c;
4600 else
4601 #endif
4602 class_uchardata += PRIV(ord2utf)(c, class_uchardata);
4603 #else /* SUPPORT_UTF */
4604 *class_uchardata++ = c;
4605 #endif /* SUPPORT_UTF */
4606
4607 #ifdef SUPPORT_UCP
4608 #ifdef COMPILE_PCRE8
4609 if ((options & PCRE_CASELESS) != 0)
4610 #else
4611 /* In non 8 bit mode, we can get here even if we are not in UTF mode. */
4612 if (utf && (options & PCRE_CASELESS) != 0)
4613 #endif
4614 {
4615 unsigned int othercase;
4616 if ((int)(othercase = UCD_OTHERCASE(c)) != c)
4617 {
4618 *class_uchardata++ = XCL_SINGLE;
4619 class_uchardata += PRIV(ord2utf)(othercase, class_uchardata);
4620 }
4621 }
4622 #endif /* SUPPORT_UCP */
4623
4624 }
4625 else
4626 #endif /* SUPPORT_UTF || COMPILE_PCRE16 */
4627
4628 /* Handle a single-byte character */
4629 {
4630 class_has_8bitchar = 1;
4631 classbits[c/8] |= (1 << (c&7));
4632 if ((options & PCRE_CASELESS) != 0)
4633 {
4634 c = cd->fcc[c]; /* flip case */
4635 classbits[c/8] |= (1 << (c&7));
4636 }
4637 }
4638 }
4639
4640 /* Loop until ']' reached. This "while" is the end of the "do" far above.
4641 If we are at the end of an internal nested string, revert to the outer
4642 string. */
4643
4644 while (((c = *(++ptr)) != 0 ||
4645 (nestptr != NULL &&
4646 (ptr = nestptr, nestptr = NULL, c = *(++ptr)) != 0)) &&
4647 (c != CHAR_RIGHT_SQUARE_BRACKET || inescq));
4648
4649 /* Check for missing terminating ']' */
4650
4651 if (c == 0)
4652 {
4653 *errorcodeptr = ERR6;
4654 goto FAILED;
4655 }
4656
4657 /* If this is the first thing in the branch, there can be no first char
4658 setting, whatever the repeat count. Any reqchar setting must remain
4659 unchanged after any kind of repeat. */
4660
4661 if (firstchar == REQ_UNSET) firstchar = REQ_NONE;
4662 zerofirstchar = firstchar;
4663 zeroreqchar = reqchar;
4664
4665 /* If there are characters with values > 255, we have to compile an
4666 extended class, with its own opcode, unless there was a negated special
4667 such as \S in the class, and PCRE_UCP is not set, because in that case all
4668 characters > 255 are in the class, so any that were explicitly given as
4669 well can be ignored. If (when there are explicit characters > 255 that must
4670 be listed) there are no characters < 256, we can omit the bitmap in the
4671 actual compiled code. */
4672
4673 #ifdef SUPPORT_UTF
4674 if (xclass && (!should_flip_negation || (options & PCRE_UCP) != 0))
4675 #elif !defined COMPILE_PCRE8
4676 if (xclass && !should_flip_negation)
4677 #endif
4678 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
4679 {
4680 *class_uchardata++ = XCL_END; /* Marks the end of extra data */
4681 *code++ = OP_XCLASS;
4682 code += LINK_SIZE;
4683 *code = negate_class? XCL_NOT:0;
4684
4685 /* If the map is required, move up the extra data to make room for it;
4686 otherwise just move the code pointer to the end of the extra data. */
4687
4688 if (class_has_8bitchar > 0)
4689 {
4690 *code++ |= XCL_MAP;
4691 memmove(code + (32 / sizeof(pcre_uchar)), code,
4692 IN_UCHARS(class_uchardata - code));
4693 memcpy(code, classbits, 32);
4694 code = class_uchardata + (32 / sizeof(pcre_uchar));
4695 }
4696 else code = class_uchardata;
4697
4698 /* Now fill in the complete length of the item */
4699
4700 PUT(previous, 1, (int)(code - previous));
4701 break; /* End of class handling */
4702 }
4703 #endif
4704
4705 /* If there are no characters > 255, or they are all to be included or
4706 excluded, set the opcode to OP_CLASS or OP_NCLASS, depending on whether the
4707 whole class was negated and whether there were negative specials such as \S
4708 (non-UCP) in the class. Then copy the 32-byte map into the code vector,
4709 negating it if necessary. */
4710
4711 *code++ = (negate_class == should_flip_negation) ? OP_CLASS : OP_NCLASS;
4712 if (lengthptr == NULL) /* Save time in the pre-compile phase */
4713 {
4714 if (negate_class)
4715 for (c = 0; c < 32; c++) classbits[c] = ~classbits[c];
4716 memcpy(code, classbits, 32);
4717 }
4718 code += 32 / sizeof(pcre_uchar);
4719 NOT_CHAR:
4720 break;
4721
4722
4723 /* ===================================================================*/
4724 /* Various kinds of repeat; '{' is not necessarily a quantifier, but this
4725 has been tested above. */
4726
4727 case CHAR_LEFT_CURLY_BRACKET:
4728 if (!is_quantifier) goto NORMAL_CHAR;
4729 ptr = read_repeat_counts(ptr+1, &repeat_min, &repeat_max, errorcodeptr);
4730 if (*errorcodeptr != 0) goto FAILED;
4731 goto REPEAT;
4732
4733 case CHAR_ASTERISK:
4734 repeat_min = 0;
4735 repeat_max = -1;
4736 goto REPEAT;
4737
4738 case CHAR_PLUS:
4739 repeat_min = 1;
4740 repeat_max = -1;
4741 goto REPEAT;
4742
4743 case CHAR_QUESTION_MARK:
4744 repeat_min = 0;
4745 repeat_max = 1;
4746
4747 REPEAT:
4748 if (previous == NULL)
4749 {
4750 *errorcodeptr = ERR9;
4751 goto FAILED;
4752 }
4753
4754 if (repeat_min == 0)
4755 {
4756 firstchar = zerofirstchar; /* Adjust for zero repeat */
4757 reqchar = zeroreqchar; /* Ditto */
4758 }
4759
4760 /* Remember whether this is a variable length repeat */
4761
4762 reqvary = (repeat_min == repeat_max)? 0 : REQ_VARY;
4763
4764 op_type = 0; /* Default single-char op codes */
4765 possessive_quantifier = FALSE; /* Default not possessive quantifier */
4766
4767 /* Save start of previous item, in case we have to move it up in order to
4768 insert something before it. */
4769
4770 tempcode = previous;
4771
4772 /* If the next character is '+', we have a possessive quantifier. This
4773 implies greediness, whatever the setting of the PCRE_UNGREEDY option.
4774 If the next character is '?' this is a minimizing repeat, by default,
4775 but if PCRE_UNGREEDY is set, it works the other way round. We change the
4776 repeat type to the non-default. */
4777
4778 if (ptr[1] == CHAR_PLUS)
4779 {
4780 repeat_type = 0; /* Force greedy */
4781 possessive_quantifier = TRUE;
4782 ptr++;
4783 }
4784 else if (ptr[1] == CHAR_QUESTION_MARK)
4785 {
4786 repeat_type = greedy_non_default;
4787 ptr++;
4788 }
4789 else repeat_type = greedy_default;
4790
4791 /* If previous was a recursion call, wrap it in atomic brackets so that
4792 previous becomes the atomic group. All recursions were so wrapped in the
4793 past, but it no longer happens for non-repeated recursions. In fact, the
4794 repeated ones could be re-implemented independently so as not to need this,
4795 but for the moment we rely on the code for repeating groups. */
4796
4797 if (*previous == OP_RECURSE)
4798 {
4799 memmove(previous + 1 + LINK_SIZE, previous, IN_UCHARS(1 + LINK_SIZE));
4800 *previous = OP_ONCE;
4801 PUT(previous, 1, 2 + 2*LINK_SIZE);
4802 previous[2 + 2*LINK_SIZE] = OP_KET;
4803 PUT(previous, 3 + 2*LINK_SIZE, 2 + 2*LINK_SIZE);
4804 code += 2 + 2 * LINK_SIZE;
4805 length_prevgroup = 3 + 3*LINK_SIZE;
4806
4807 /* When actually compiling, we need to check whether this was a forward
4808 reference, and if so, adjust the offset. */
4809
4810 if (lengthptr == NULL && cd->hwm >= cd->start_workspace + LINK_SIZE)
4811 {
4812 int offset = GET(cd->hwm, -LINK_SIZE);
4813 if (offset == previous + 1 - cd->start_code)
4814 PUT(cd->hwm, -LINK_SIZE, offset + 1 + LINK_SIZE);
4815 }
4816 }
4817
4818 /* Now handle repetition for the different types of item. */
4819
4820 /* If previous was a character or negated character match, abolish the item
4821 and generate a repeat item instead. If a char item has a minimum of more
4822 than one, ensure that it is set in reqchar - it might not be if a sequence
4823 such as x{3} is the first thing in a branch because the x will have gone
4824 into firstchar instead. */
4825
4826 if (*previous == OP_CHAR || *previous == OP_CHARI
4827 || *previous == OP_NOT || *previous == OP_NOTI)
4828 {
4829 switch (*previous)
4830 {
4831 default: /* Make compiler happy. */
4832 case OP_CHAR: op_type = OP_STAR - OP_STAR; break;
4833 case OP_CHARI: op_type = OP_STARI - OP_STAR; break;
4834 case OP_NOT: op_type = OP_NOTSTAR - OP_STAR; break;
4835 case OP_NOTI: op_type = OP_NOTSTARI - OP_STAR; break;
4836 }
4837
4838 /* Deal with UTF characters that take up more than one character. It's
4839 easier to write this out separately than try to macrify it. Use c to
4840 hold the length of the character in bytes, plus UTF_LENGTH to flag that
4841 it's a length rather than a small character. */
4842
4843 #ifdef SUPPORT_UTF
4844 if (utf && NOT_FIRSTCHAR(code[-1]))
4845 {
4846 pcre_uchar *lastchar = code - 1;
4847 BACKCHAR(lastchar);
4848 c = (int)(code - lastchar); /* Length of UTF-8 character */
4849 memcpy(utf_chars, lastchar, IN_UCHARS(c)); /* Save the char */
4850 c |= UTF_LENGTH; /* Flag c as a length */
4851 }
4852 else
4853 #endif /* SUPPORT_UTF */
4854
4855 /* Handle the case of a single charater - either with no UTF support, or
4856 with UTF disabled, or for a single character UTF character. */
4857 {
4858 c = code[-1];
4859 if (*previous <= OP_CHARI && repeat_min > 1)
4860 reqchar = c | req_caseopt | cd->req_varyopt;
4861 }
4862
4863 /* If the repetition is unlimited, it pays to see if the next thing on
4864 the line is something that cannot possibly match this character. If so,
4865 automatically possessifying this item gains some performance in the case
4866 where the match fails. */
4867
4868 if (!possessive_quantifier &&
4869 repeat_max < 0 &&
4870 check_auto_possessive(previous, utf, ptr + 1, options, cd))
4871 {
4872 repeat_type = 0; /* Force greedy */
4873 possessive_quantifier = TRUE;
4874 }
4875
4876 goto OUTPUT_SINGLE_REPEAT; /* Code shared with single character types */
4877 }
4878
4879 /* If previous was a character type match (\d or similar), abolish it and
4880 create a suitable repeat item. The code is shared with single-character
4881 repeats by setting op_type to add a suitable offset into repeat_type. Note
4882 the the Unicode property types will be present only when SUPPORT_UCP is
4883 defined, but we don't wrap the little bits of code here because it just
4884 makes it horribly messy. */
4885
4886 else if (*previous < OP_EODN)
4887 {
4888 pcre_uchar *oldcode;
4889 int prop_type, prop_value;
4890 op_type = OP_TYPESTAR - OP_STAR; /* Use type opcodes */
4891 c = *previous;
4892
4893 if (!possessive_quantifier &&
4894 repeat_max < 0 &&
4895 check_auto_possessive(previous, utf, ptr + 1, options, cd))
4896 {
4897 repeat_type = 0; /* Force greedy */
4898 possessive_quantifier = TRUE;
4899 }
4900
4901 OUTPUT_SINGLE_REPEAT:
4902 if (*previous == OP_PROP || *previous == OP_NOTPROP)
4903 {
4904 prop_type = previous[1];
4905 prop_value = previous[2];
4906 }
4907 else prop_type = prop_value = -1;
4908
4909 oldcode = code;
4910 code = previous; /* Usually overwrite previous item */
4911
4912 /* If the maximum is zero then the minimum must also be zero; Perl allows
4913 this case, so we do too - by simply omitting the item altogether. */
4914
4915 if (repeat_max == 0) goto END_REPEAT;
4916
4917 /*--------------------------------------------------------------------*/
4918 /* This code is obsolete from release 8.00; the restriction was finally
4919 removed: */
4920
4921 /* All real repeats make it impossible to handle partial matching (maybe
4922 one day we will be able to remove this restriction). */
4923
4924 /* if (repeat_max != 1) cd->external_flags |= PCRE_NOPARTIAL; */
4925 /*--------------------------------------------------------------------*/
4926
4927 /* Combine the op_type with the repeat_type */
4928
4929 repeat_type += op_type;
4930
4931 /* A minimum of zero is handled either as the special case * or ?, or as
4932 an UPTO, with the maximum given. */
4933
4934 if (repeat_min == 0)
4935 {
4936 if (repeat_max == -1) *code++ = OP_STAR + repeat_type;
4937 else if (repeat_max == 1) *code++ = OP_QUERY + repeat_type;
4938 else
4939 {
4940 *code++ = OP_UPTO + repeat_type;
4941 PUT2INC(code, 0, repeat_max);
4942 }
4943 }
4944
4945 /* A repeat minimum of 1 is optimized into some special cases. If the
4946 maximum is unlimited, we use OP_PLUS. Otherwise, the original item is
4947 left in place and, if the maximum is greater than 1, we use OP_UPTO with
4948 one less than the maximum. */
4949
4950 else if (repeat_min == 1)
4951 {
4952 if (repeat_max == -1)
4953 *code++ = OP_PLUS + repeat_type;
4954 else
4955 {
4956 code = oldcode; /* leave previous item in place */
4957 if (repeat_max == 1) goto END_REPEAT;
4958 *code++ = OP_UPTO + repeat_type;
4959 PUT2INC(code, 0, repeat_max - 1);
4960 }
4961 }
4962
4963 /* The case {n,n} is just an EXACT, while the general case {n,m} is
4964 handled as an EXACT followed by an UPTO. */
4965
4966 else
4967 {
4968 *code++ = OP_EXACT + op_type; /* NB EXACT doesn't have repeat_type */
4969 PUT2INC(code, 0, repeat_min);
4970
4971 /* If the maximum is unlimited, insert an OP_STAR. Before doing so,
4972 we have to insert the character for the previous code. For a repeated
4973 Unicode property match, there are two extra bytes that define the
4974 required property. In UTF-8 mode, long characters have their length in
4975 c, with the UTF_LENGTH bit as a flag. */
4976
4977 if (repeat_max < 0)
4978 {
4979 #ifdef SUPPORT_UTF
4980 if (utf && (c & UTF_LENGTH) != 0)
4981 {
4982 memcpy(code, utf_chars, IN_UCHARS(c & 7));
4983 code += c & 7;
4984 }
4985 else
4986 #endif
4987 {
4988 *code++ = c;
4989 if (prop_type >= 0)
4990 {
4991 *code++ = prop_type;
4992 *code++ = prop_value;
4993 }
4994 }
4995 *code++ = OP_STAR + repeat_type;
4996 }
4997
4998 /* Else insert an UPTO if the max is greater than the min, again
4999 preceded by the character, for the previously inserted code. If the
5000 UPTO is just for 1 instance, we can use QUERY instead. */
5001
5002 else if (repeat_max != repeat_min)
5003 {
5004 #ifdef SUPPORT_UTF
5005 if (utf && (c & UTF_LENGTH) != 0)
5006 {
5007 memcpy(code, utf_chars, IN_UCHARS(c & 7));
5008 code += c & 7;
5009 }
5010 else
5011 #endif
5012 *code++ = c;
5013 if (prop_type >= 0)
5014 {
5015 *code++ = prop_type;
5016 *code++ = prop_value;
5017 }
5018 repeat_max -= repeat_min;
5019
5020 if (repeat_max == 1)
5021 {
5022 *code++ = OP_QUERY + repeat_type;
5023 }
5024 else
5025 {
5026 *code++ = OP_UPTO + repeat_type;
5027 PUT2INC(code, 0, repeat_max);
5028 }
5029 }
5030 }
5031
5032 /* The character or character type itself comes last in all cases. */
5033
5034 #ifdef SUPPORT_UTF
5035 if (utf && (c & UTF_LENGTH) != 0)
5036 {
5037 memcpy(code, utf_chars, IN_UCHARS(c & 7));
5038 code += c & 7;
5039 }
5040 else
5041 #endif
5042 *code++ = c;
5043
5044 /* For a repeated Unicode property match, there are two extra bytes that
5045 define the required property. */
5046
5047 #ifdef SUPPORT_UCP
5048 if (prop_type >= 0)
5049 {
5050 *code++ = prop_type;
5051 *code++ = prop_value;
5052 }
5053 #endif
5054 }
5055
5056 /* If previous was a character class or a back reference, we put the repeat
5057 stuff after it, but just skip the item if the repeat was {0,0}. */
5058
5059 else if (*previous == OP_CLASS ||
5060 *previous == OP_NCLASS ||
5061 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
5062 *previous == OP_XCLASS ||
5063 #endif
5064 *previous == OP_REF ||
5065 *previous == OP_REFI)
5066 {
5067 if (repeat_max == 0)
5068 {
5069 code = previous;
5070 goto END_REPEAT;
5071 }
5072
5073 /*--------------------------------------------------------------------*/
5074 /* This code is obsolete from release 8.00; the restriction was finally
5075 removed: */
5076
5077 /* All real repeats make it impossible to handle partial matching (maybe
5078 one day we will be able to remove this restriction). */
5079
5080 /* if (repeat_max != 1) cd->external_flags |= PCRE_NOPARTIAL; */
5081 /*--------------------------------------------------------------------*/
5082
5083 if (repeat_min == 0 && repeat_max == -1)
5084 *code++ = OP_CRSTAR + repeat_type;
5085 else if (repeat_min == 1 && repeat_max == -1)
5086 *code++ = OP_CRPLUS + repeat_type;
5087 else if (repeat_min == 0 && repeat_max == 1)
5088 *code++ = OP_CRQUERY + repeat_type;
5089 else
5090 {
5091 *code++ = OP_CRRANGE + repeat_type;
5092 PUT2INC(code, 0, repeat_min);
5093 if (repeat_max == -1) repeat_max = 0; /* 2-byte encoding for max */
5094 PUT2INC(code, 0, repeat_max);
5095 }
5096 }
5097
5098 /* If previous was a bracket group, we may have to replicate it in certain
5099 cases. Note that at this point we can encounter only the "basic" bracket
5100 opcodes such as BRA and CBRA, as this is the place where they get converted
5101 into the more special varieties such as BRAPOS and SBRA. A test for >=
5102 OP_ASSERT and <= OP_COND includes ASSERT, ASSERT_NOT, ASSERTBACK,
5103 ASSERTBACK_NOT, ONCE, BRA, CBRA, and COND. Originally, PCRE did not allow
5104 repetition of assertions, but now it does, for Perl compatibility. */
5105
5106 else if (*previous >= OP_ASSERT && *previous <= OP_COND)
5107 {
5108 register int i;
5109 int len = (int)(code - previous);
5110 pcre_uchar *bralink = NULL;
5111 pcre_uchar *brazeroptr = NULL;
5112
5113 /* Repeating a DEFINE group is pointless, but Perl allows the syntax, so
5114 we just ignore the repeat. */
5115
5116 if (*previous == OP_COND && previous[LINK_SIZE+1] == OP_DEF)
5117 goto END_REPEAT;
5118
5119 /* There is no sense in actually repeating assertions. The only potential
5120 use of repetition is in cases when the assertion is optional. Therefore,
5121 if the minimum is greater than zero, just ignore the repeat. If the
5122 maximum is not not zero or one, set it to 1. */
5123
5124 if (*previous < OP_ONCE) /* Assertion */
5125 {
5126 if (repeat_min > 0) goto END_REPEAT;
5127 if (repeat_max < 0 || repeat_max > 1) repeat_max = 1;
5128 }
5129
5130 /* The case of a zero minimum is special because of the need to stick
5131 OP_BRAZERO in front of it, and because the group appears once in the
5132 data, whereas in other cases it appears the minimum number of times. For
5133 this reason, it is simplest to treat this case separately, as otherwise
5134 the code gets far too messy. There are several special subcases when the
5135 minimum is zero. */
5136
5137 if (repeat_min == 0)
5138 {
5139 /* If the maximum is also zero, we used to just omit the group from the
5140 output altogether, like this:
5141
5142 ** if (repeat_max == 0)
5143 ** {
5144 ** code = previous;
5145 ** goto END_REPEAT;
5146 ** }
5147
5148 However, that fails when a group or a subgroup within it is referenced
5149 as a subroutine from elsewhere in the pattern, so now we stick in
5150 OP_SKIPZERO in front of it so that it is skipped on execution. As we
5151 don't have a list of which groups are referenced, we cannot do this
5152 selectively.
5153
5154 If the maximum is 1 or unlimited, we just have to stick in the BRAZERO
5155 and do no more at this point. However, we do need to adjust any
5156 OP_RECURSE calls inside the group that refer to the group itself or any
5157 internal or forward referenced group, because the offset is from the
5158 start of the whole regex. Temporarily terminate the pattern while doing
5159 this. */
5160
5161 if (repeat_max <= 1) /* Covers 0, 1, and unlimited */
5162 {
5163 *code = OP_END;
5164 adjust_recurse(previous, 1, utf, cd, save_hwm);
5165 memmove(previous + 1, previous, IN_UCHARS(len));
5166 code++;
5167 if (repeat_max == 0)
5168 {
5169 *previous++ = OP_SKIPZERO;
5170 goto END_REPEAT;
5171 }
5172 brazeroptr = previous; /* Save for possessive optimizing */
5173 *previous++ = OP_BRAZERO + repeat_type;
5174 }
5175
5176 /* If the maximum is greater than 1 and limited, we have to replicate
5177 in a nested fashion, sticking OP_BRAZERO before each set of brackets.
5178 The first one has to be handled carefully because it's the original
5179 copy, which has to be moved up. The remainder can be handled by code
5180 that is common with the non-zero minimum case below. We have to
5181 adjust the value or repeat_max, since one less copy is required. Once
5182 again, we may have to adjust any OP_RECURSE calls inside the group. */
5183
5184 else
5185 {
5186 int offset;
5187 *code = OP_END;
5188 adjust_recurse(previous, 2 + LINK_SIZE, utf, cd, save_hwm);
5189 memmove(previous + 2 + LINK_SIZE, previous, IN_UCHARS(len));
5190 code += 2 + LINK_SIZE;
5191 *previous++ = OP_BRAZERO + repeat_type;
5192 *previous++ = OP_BRA;
5193
5194 /* We chain together the bracket offset fields that have to be
5195 filled in later when the ends of the brackets are reached. */
5196
5197 offset = (bralink == NULL)? 0 : (int)(previous - bralink);
5198 bralink = previous;
5199 PUTINC(previous, 0, offset);
5200 }
5201
5202 repeat_max--;
5203 }
5204
5205 /* If the minimum is greater than zero, replicate the group as many
5206 times as necessary, and adjust the maximum to the number of subsequent
5207 copies that we need. If we set a first char from the group, and didn't
5208 set a required char, copy the latter from the former. If there are any
5209 forward reference subroutine calls in the group, there will be entries on
5210 the workspace list; replicate these with an appropriate increment. */
5211
5212 else
5213 {
5214 if (repeat_min > 1)
5215 {
5216 /* In the pre-compile phase, we don't actually do the replication. We
5217 just adjust the length as if we had. Do some paranoid checks for
5218 potential integer overflow. The INT64_OR_DOUBLE type is a 64-bit
5219 integer type when available, otherwise double. */
5220
5221 if (lengthptr != NULL)
5222 {
5223 int delta = (repeat_min - 1)*length_prevgroup;
5224 if ((INT64_OR_DOUBLE)(repeat_min - 1)*
5225 (INT64_OR_DOUBLE)length_prevgroup >
5226 (INT64_OR_DOUBLE)INT_MAX ||
5227 OFLOW_MAX - *lengthptr < delta)
5228 {
5229 *errorcodeptr = ERR20;
5230 goto FAILED;
5231 }
5232 *lengthptr += delta;
5233 }
5234
5235 /* This is compiling for real. If there is a set first byte for
5236 the group, and we have not yet set a "required byte", set it. Make
5237 sure there is enough workspace for copying forward references before
5238 doing the copy. */
5239
5240 else
5241 {
5242 if (groupsetfirstchar && reqchar < 0) reqchar = firstchar;
5243
5244 for (i = 1; i < repeat_min; i++)
5245 {
5246 pcre_uchar *hc;
5247 pcre_uchar *this_hwm = cd->hwm;
5248 memcpy(code, previous, IN_UCHARS(len));
5249
5250 while (cd->hwm > cd->start_workspace + cd->workspace_size -
5251 WORK_SIZE_SAFETY_MARGIN - (this_hwm - save_hwm))
5252 {
5253 int save_offset = save_hwm - cd->start_workspace;
5254 int this_offset = this_hwm - cd->start_workspace;
5255 *errorcodeptr = expand_workspace(cd);
5256 if (*errorcodeptr != 0) goto FAILED;
5257 save_hwm = (pcre_uchar *)cd->start_workspace + save_offset;
5258 this_hwm = (pcre_uchar *)cd->start_workspace + this_offset;
5259 }
5260
5261 for (hc = save_hwm; hc < this_hwm; hc += LINK_SIZE)
5262 {
5263 PUT(cd->hwm, 0, GET(hc, 0) + len);
5264 cd->hwm += LINK_SIZE;
5265 }
5266 save_hwm = this_hwm;
5267 code += len;
5268 }
5269 }
5270 }
5271
5272 if (repeat_max > 0) repeat_max -= repeat_min;
5273 }
5274
5275 /* This code is common to both the zero and non-zero minimum cases. If
5276 the maximum is limited, it replicates the group in a nested fashion,
5277 remembering the bracket starts on a stack. In the case of a zero minimum,
5278 the first one was set up above. In all cases the repeat_max now specifies
5279 the number of additional copies needed. Again, we must remember to
5280 replicate entries on the forward reference list. */
5281
5282 if (repeat_max >= 0)
5283 {
5284 /* In the pre-compile phase, we don't actually do the replication. We
5285 just adjust the length as if we had. For each repetition we must add 1
5286 to the length for BRAZERO and for all but the last repetition we must
5287 add 2 + 2*LINKSIZE to allow for the nesting that occurs. Do some
5288 paranoid checks to avoid integer overflow. The INT64_OR_DOUBLE type is
5289 a 64-bit integer type when available, otherwise double. */
5290
5291 if (lengthptr != NULL && repeat_max > 0)
5292 {
5293 int delta = repeat_max * (length_prevgroup + 1 + 2 + 2*LINK_SIZE) -
5294 2 - 2*LINK_SIZE; /* Last one doesn't nest */
5295 if ((INT64_OR_DOUBLE)repeat_max *
5296 (INT64_OR_DOUBLE)(length_prevgroup + 1 + 2 + 2*LINK_SIZE)
5297 > (INT64_OR_DOUBLE)INT_MAX ||
5298 OFLOW_MAX - *lengthptr < delta)
5299 {
5300 *errorcodeptr = ERR20;
5301 goto FAILED;
5302 }
5303 *lengthptr += delta;
5304 }
5305
5306 /* This is compiling for real */
5307
5308 else for (i = repeat_max - 1; i >= 0; i--)
5309 {
5310 pcre_uchar *hc;
5311 pcre_uchar *this_hwm = cd->hwm;
5312
5313 *code++ = OP_BRAZERO + repeat_type;
5314
5315 /* All but the final copy start a new nesting, maintaining the
5316 chain of brackets outstanding. */
5317
5318 if (i != 0)
5319 {
5320 int offset;
5321 *code++ = OP_BRA;
5322 offset = (bralink == NULL)? 0 : (int)(code - bralink);
5323 bralink = code;
5324 PUTINC(code, 0, offset);
5325 }
5326
5327 memcpy(code, previous, IN_UCHARS(len));
5328
5329 /* Ensure there is enough workspace for forward references before
5330 copying them. */
5331
5332 while (cd->hwm > cd->start_workspace + cd->workspace_size -
5333 WORK_SIZE_SAFETY_MARGIN - (this_hwm - save_hwm))
5334 {
5335 int save_offset = save_hwm - cd->start_workspace;
5336 int this_offset = this_hwm - cd->start_workspace;
5337 *errorcodeptr = expand_workspace(cd);
5338 if (*errorcodeptr != 0) goto FAILED;
5339 save_hwm = (pcre_uchar *)cd->start_workspace + save_offset;
5340 this_hwm = (pcre_uchar *)cd->start_workspace + this_offset;
5341 }
5342
5343 for (hc = save_hwm; hc < this_hwm; hc += LINK_SIZE)
5344 {
5345 PUT(cd->hwm, 0, GET(hc, 0) + len + ((i != 0)? 2+LINK_SIZE : 1));
5346 cd->hwm += LINK_SIZE;
5347 }
5348 save_hwm = this_hwm;
5349 code += len;
5350 }
5351
5352 /* Now chain through the pending brackets, and fill in their length
5353 fields (which are holding the chain links pro tem). */
5354
5355 while (bralink != NULL)
5356 {
5357 int oldlinkoffset;
5358 int offset = (int)(code - bralink + 1);
5359 pcre_uchar *bra = code - offset;
5360 oldlinkoffset = GET(bra, 1);
5361 bralink = (oldlinkoffset == 0)? NULL : bralink - oldlinkoffset;
5362 *code++ = OP_KET;
5363 PUTINC(code, 0, offset);
5364 PUT(bra, 1, offset);
5365 }
5366 }
5367
5368 /* If the maximum is unlimited, set a repeater in the final copy. For
5369 ONCE brackets, that's all we need to do. However, possessively repeated
5370 ONCE brackets can be converted into non-capturing brackets, as the
5371 behaviour of (?:xx)++ is the same as (?>xx)++ and this saves having to
5372 deal with possessive ONCEs specially.
5373
5374 Otherwise, when we are doing the actual compile phase, check to see
5375 whether this group is one that could match an empty string. If so,
5376 convert the initial operator to the S form (e.g. OP_BRA -> OP_SBRA) so
5377 that runtime checking can be done. [This check is also applied to ONCE
5378 groups at runtime, but in a different way.]
5379
5380 Then, if the quantifier was possessive and the bracket is not a
5381 conditional, we convert the BRA code to the POS form, and the KET code to
5382 KETRPOS. (It turns out to be convenient at runtime to detect this kind of
5383 subpattern at both the start and at the end.) The use of special opcodes
5384 makes it possible to reduce greatly the stack usage in pcre_exec(). If
5385 the group is preceded by OP_BRAZERO, convert this to OP_BRAPOSZERO.
5386
5387 Then, if the minimum number of matches is 1 or 0, cancel the possessive
5388 flag so that the default action below, of wrapping everything inside
5389 atomic brackets, does not happen. When the minimum is greater than 1,
5390 there will be earlier copies of the group, and so we still have to wrap
5391 the whole thing. */
5392
5393 else
5394 {
5395 pcre_uchar *ketcode = code - 1 - LINK_SIZE;
5396 pcre_uchar *bracode = ketcode - GET(ketcode, 1);
5397
5398 /* Convert possessive ONCE brackets to non-capturing */
5399
5400 if ((*bracode == OP_ONCE || *bracode == OP_ONCE_NC) &&
5401 possessive_quantifier) *bracode = OP_BRA;
5402
5403 /* For non-possessive ONCE brackets, all we need to do is to
5404 set the KET. */
5405
5406 if (*bracode == OP_ONCE || *bracode == OP_ONCE_NC)
5407 *ketcode = OP_KETRMAX + repeat_type;
5408
5409 /* Handle non-ONCE brackets and possessive ONCEs (which have been
5410 converted to non-capturing above). */
5411
5412 else
5413 {
5414 /* In the compile phase, check for empty string matching. */
5415
5416 if (lengthptr == NULL)
5417 {
5418 pcre_uchar *scode = bracode;
5419 do
5420 {
5421 if (could_be_empty_branch(scode, ketcode, utf, cd))
5422 {
5423 *bracode += OP_SBRA - OP_BRA;
5424 break;
5425 }
5426 scode += GET(scode, 1);
5427 }
5428 while (*scode == OP_ALT);
5429 }
5430
5431 /* Handle possessive quantifiers. */
5432
5433 if (possessive_quantifier)
5434 {
5435 /* For COND brackets, we wrap the whole thing in a possessively
5436 repeated non-capturing bracket, because we have not invented POS
5437 versions of the COND opcodes. Because we are moving code along, we
5438 must ensure that any pending recursive references are updated. */
5439
5440 if (*bracode == OP_COND || *bracode == OP_SCOND)
5441 {
5442 int nlen = (int)(code - bracode);
5443 *code = OP_END;
5444 adjust_recurse(bracode, 1 + LINK_SIZE, utf, cd, save_hwm);
5445 memmove(bracode + 1 + LINK_SIZE, bracode, IN_UCHARS(nlen));
5446 code += 1 + LINK_SIZE;
5447 nlen += 1 + LINK_SIZE;
5448 *bracode = OP_BRAPOS;
5449 *code++ = OP_KETRPOS;
5450 PUTINC(code, 0, nlen);
5451 PUT(bracode, 1, nlen);
5452 }
5453
5454 /* For non-COND brackets, we modify the BRA code and use KETRPOS. */
5455
5456 else
5457 {
5458 *bracode += 1; /* Switch to xxxPOS opcodes */
5459 *ketcode = OP_KETRPOS;
5460 }
5461
5462 /* If the minimum is zero, mark it as possessive, then unset the
5463 possessive flag when the minimum is 0 or 1. */
5464
5465 if (brazeroptr != NULL) *brazeroptr = OP_BRAPOSZERO;
5466 if (repeat_min < 2) possessive_quantifier = FALSE;
5467 }
5468
5469 /* Non-possessive quantifier */
5470
5471 else *ketcode = OP_KETRMAX + repeat_type;
5472 }
5473 }
5474 }
5475
5476 /* If previous is OP_FAIL, it was generated by an empty class [] in
5477 JavaScript mode. The other ways in which OP_FAIL can be generated, that is
5478 by (*FAIL) or (?!) set previous to NULL, which gives a "nothing to repeat"
5479 error above. We can just ignore the repeat in JS case. */
5480
5481 else if (*previous == OP_FAIL) goto END_REPEAT;
5482
5483 /* Else there's some kind of shambles */
5484
5485 else
5486 {
5487 *errorcodeptr = ERR11;
5488 goto FAILED;
5489 }
5490
5491 /* If the character following a repeat is '+', or if certain optimization
5492 tests above succeeded, possessive_quantifier is TRUE. For some opcodes,
5493 there are special alternative opcodes for this case. For anything else, we
5494 wrap the entire repeated item inside OP_ONCE brackets. Logically, the '+'
5495 notation is just syntactic sugar, taken from Sun's Java package, but the
5496 special opcodes can optimize it.
5497
5498 Some (but not all) possessively repeated subpatterns have already been
5499 completely handled in the code just above. For them, possessive_quantifier
5500 is always FALSE at this stage.
5501
5502 Note that the repeated item starts at tempcode, not at previous, which
5503 might be the first part of a string whose (former) last char we repeated.
5504
5505 Possessifying an 'exact' quantifier has no effect, so we can ignore it. But
5506 an 'upto' may follow. We skip over an 'exact' item, and then test the
5507 length of what remains before proceeding. */
5508
5509 if (possessive_quantifier)
5510 {
5511 int len;
5512
5513 if (*tempcode == OP_TYPEEXACT)
5514 tempcode += PRIV(OP_lengths)[*tempcode] +
5515 ((tempcode[1 + IMM2_SIZE] == OP_PROP
5516 || tempcode[1 + IMM2_SIZE] == OP_NOTPROP)? 2 : 0);
5517
5518 else if (*tempcode == OP_EXACT || *tempcode == OP_NOTEXACT)
5519 {
5520 tempcode += PRIV(OP_lengths)[*tempcode];
5521 #ifdef SUPPORT_UTF
5522 if (utf && HAS_EXTRALEN(tempcode[-1]))
5523 tempcode += GET_EXTRALEN(tempcode[-1]);
5524 #endif
5525 }
5526
5527 len = (int)(code - tempcode);
5528 if (len > 0) switch (*tempcode)
5529 {
5530 case OP_STAR: *tempcode = OP_POSSTAR; break;
5531 case OP_PLUS: *tempcode = OP_POSPLUS; break;
5532 case OP_QUERY: *tempcode = OP_POSQUERY; break;
5533 case OP_UPTO: *tempcode = OP_POSUPTO; break;
5534
5535 case OP_STARI: *tempcode = OP_POSSTARI; break;
5536 case OP_PLUSI: *tempcode = OP_POSPLUSI; break;
5537 case OP_QUERYI: *tempcode = OP_POSQUERYI; break;
5538 case OP_UPTOI: *tempcode = OP_POSUPTOI; break;
5539
5540 case OP_NOTSTAR: *tempcode = OP_NOTPOSSTAR; break;
5541 case OP_NOTPLUS: *tempcode = OP_NOTPOSPLUS; break;
5542 case OP_NOTQUERY: *tempcode = OP_NOTPOSQUERY; break;
5543 case OP_NOTUPTO: *tempcode = OP_NOTPOSUPTO; break;
5544
5545 case OP_NOTSTARI: *tempcode = OP_NOTPOSSTARI; break;
5546 case OP_NOTPLUSI: *tempcode = OP_NOTPOSPLUSI; break;
5547 case OP_NOTQUERYI: *tempcode = OP_NOTPOSQUERYI; break;
5548 case OP_NOTUPTOI: *tempcode = OP_NOTPOSUPTOI; break;
5549
5550 case OP_TYPESTAR: *tempcode = OP_TYPEPOSSTAR; break;
5551 case OP_TYPEPLUS: *tempcode = OP_TYPEPOSPLUS; break;
5552 case OP_TYPEQUERY: *tempcode = OP_TYPEPOSQUERY; break;
5553 case OP_TYPEUPTO: *tempcode = OP_TYPEPOSUPTO; break;
5554
5555 /* Because we are moving code along, we must ensure that any
5556 pending recursive references are updated. */
5557
5558 default:
5559 *code = OP_END;
5560 adjust_recurse(tempcode, 1 + LINK_SIZE, utf, cd, save_hwm);
5561 memmove(tempcode + 1 + LINK_SIZE, tempcode, IN_UCHARS(len));
5562 code += 1 + LINK_SIZE;
5563 len += 1 + LINK_SIZE;
5564 tempcode[0] = OP_ONCE;
5565 *code++ = OP_KET;
5566 PUTINC(code, 0, len);
5567 PUT(tempcode, 1, len);
5568 break;
5569 }
5570 }
5571
5572 /* In all case we no longer have a previous item. We also set the
5573 "follows varying string" flag for subsequently encountered reqchars if
5574 it isn't already set and we have just passed a varying length item. */
5575
5576 END_REPEAT:
5577 previous = NULL;
5578 cd->req_varyopt |= reqvary;
5579 break;
5580
5581
5582 /* ===================================================================*/
5583 /* Start of nested parenthesized sub-expression, or comment or lookahead or
5584 lookbehind or option setting or condition or all the other extended
5585 parenthesis forms. */
5586
5587 case CHAR_LEFT_PARENTHESIS:
5588 newoptions = options;
5589 skipbytes = 0;
5590 bravalue = OP_CBRA;
5591 save_hwm = cd->hwm;
5592 reset_bracount = FALSE;
5593
5594 /* First deal with various "verbs" that can be introduced by '*'. */
5595
5596 ptr++;
5597 if (ptr[0] == CHAR_ASTERISK && (ptr[1] == ':'
5598 || (MAX_255(ptr[1]) && ((cd->ctypes[ptr[1]] & ctype_letter) != 0))))
5599 {
5600 int i, namelen;
5601 int arglen = 0;
5602 const char *vn = verbnames;
5603 const pcre_uchar *name = ptr + 1;
5604 const pcre_uchar *arg = NULL;
5605 previous = NULL;
5606 ptr++;
5607 while (MAX_255(*ptr) && (cd->ctypes[*ptr] & ctype_letter) != 0) ptr++;
5608 namelen = (int)(ptr - name);
5609
5610 /* It appears that Perl allows any characters whatsoever, other than
5611 a closing parenthesis, to appear in arguments, so we no longer insist on
5612 letters, digits, and underscores. */
5613
5614 if (*ptr == CHAR_COLON)
5615 {
5616 arg = ++ptr;
5617 while (*ptr != 0 && *ptr != CHAR_RIGHT_PARENTHESIS) ptr++;
5618 arglen = (int)(ptr - arg);
5619 if (arglen > (int)MAX_MARK)
5620 {
5621 *errorcodeptr = ERR75;
5622 goto FAILED;
5623 }
5624 }
5625
5626 if (*ptr != CHAR_RIGHT_PARENTHESIS)
5627 {
5628 *errorcodeptr = ERR60;
5629 goto FAILED;
5630 }
5631
5632 /* Scan the table of verb names */
5633
5634 for (i = 0; i < verbcount; i++)
5635 {
5636 if (namelen == verbs[i].len &&
5637 STRNCMP_UC_C8(name, vn, namelen) == 0)
5638 {
5639 /* Check for open captures before ACCEPT and convert it to
5640 ASSERT_ACCEPT if in an assertion. */
5641
5642 if (verbs[i].op == OP_ACCEPT)
5643 {
5644 open_capitem *oc;
5645 if (arglen != 0)
5646 {
5647 *errorcodeptr = ERR59;
5648 goto FAILED;
5649 }
5650 cd->had_accept = TRUE;
5651 for (oc = cd->open_caps; oc != NULL; oc = oc->next)
5652 {
5653 *code++ = OP_CLOSE;
5654 PUT2INC(code, 0, oc->number);
5655 }
5656 *code++ = (cd->assert_depth > 0)? OP_ASSERT_ACCEPT : OP_ACCEPT;
5657
5658 /* Do not set firstchar after *ACCEPT */
5659 if (firstchar == REQ_UNSET) firstchar = REQ_NONE;
5660 }
5661
5662 /* Handle other cases with/without an argument */
5663
5664 else if (arglen == 0)
5665 {
5666 if (verbs[i].op < 0) /* Argument is mandatory */
5667 {
5668 *errorcodeptr = ERR66;
5669 goto FAILED;
5670 }
5671 *code = verbs[i].op;
5672 if (*code++ == OP_THEN) cd->external_flags |= PCRE_HASTHEN;
5673 }
5674
5675 else
5676 {
5677 if (verbs[i].op_arg < 0) /* Argument is forbidden */
5678 {
5679 *errorcodeptr = ERR59;
5680 goto FAILED;
5681 }
5682 *code = verbs[i].op_arg;
5683 if (*code++ == OP_THEN_ARG) cd->external_flags |= PCRE_HASTHEN;
5684 *code++ = arglen;
5685 memcpy(code, arg, IN_UCHARS(arglen));
5686 code += arglen;
5687 *code++ = 0;
5688 }
5689
5690 break; /* Found verb, exit loop */
5691 }
5692
5693 vn += verbs[i].len + 1;
5694 }
5695
5696 if (i < verbcount) continue; /* Successfully handled a verb */
5697 *errorcodeptr = ERR60; /* Verb not recognized */
5698 goto FAILED;
5699 }
5700
5701 /* Deal with the extended parentheses; all are introduced by '?', and the
5702 appearance of any of them means that this is not a capturing group. */
5703
5704 else if (*ptr == CHAR_QUESTION_MARK)
5705 {
5706 int i, set, unset, namelen;
5707 int *optset;
5708 const pcre_uchar *name;
5709 pcre_uchar *slot;
5710
5711 switch (*(++ptr))
5712 {
5713 case CHAR_NUMBER_SIGN: /* Comment; skip to ket */
5714 ptr++;
5715 while (*ptr != 0 && *ptr != CHAR_RIGHT_PARENTHESIS) ptr++;
5716 if (*ptr == 0)
5717 {
5718 *errorcodeptr = ERR18;
5719 goto FAILED;
5720 }
5721 continue;
5722
5723
5724 /* ------------------------------------------------------------ */
5725 case CHAR_VERTICAL_LINE: /* Reset capture count for each branch */
5726 reset_bracount = TRUE;
5727 /* Fall through */
5728
5729 /* ------------------------------------------------------------ */
5730 case CHAR_COLON: /* Non-capturing bracket */
5731 bravalue = OP_BRA;
5732 ptr++;
5733 break;
5734
5735
5736 /* ------------------------------------------------------------ */
5737 case CHAR_LEFT_PARENTHESIS:
5738 bravalue = OP_COND; /* Conditional group */
5739
5740 /* A condition can be an assertion, a number (referring to a numbered
5741 group), a name (referring to a named group), or 'R', referring to
5742 recursion. R<digits> and R&name are also permitted for recursion tests.
5743
5744 There are several syntaxes for testing a named group: (?(name)) is used
5745 by Python; Perl 5.10 onwards uses (?(<name>) or (?('name')).
5746
5747 There are two unfortunate ambiguities, caused by history. (a) 'R' can
5748 be the recursive thing or the name 'R' (and similarly for 'R' followed
5749 by digits), and (b) a number could be a name that consists of digits.
5750 In both cases, we look for a name first; if not found, we try the other
5751 cases. */
5752
5753 /* For conditions that are assertions, check the syntax, and then exit
5754 the switch. This will take control down to where bracketed groups,
5755 including assertions, are processed. */
5756
5757 if (ptr[1] == CHAR_QUESTION_MARK && (ptr[2] == CHAR_EQUALS_SIGN ||
5758 ptr[2] == CHAR_EXCLAMATION_MARK || ptr[2] == CHAR_LESS_THAN_SIGN))
5759 break;
5760
5761 /* Most other conditions use OP_CREF (a couple change to OP_RREF
5762 below), and all need to skip 1+IMM2_SIZE bytes at the start of the group. */
5763
5764 code[1+LINK_SIZE] = OP_CREF;
5765 skipbytes = 1+IMM2_SIZE;
5766 refsign = -1;
5767
5768 /* Check for a test for recursion in a named group. */
5769
5770 if (ptr[1] == CHAR_R && ptr[2] == CHAR_AMPERSAND)
5771 {
5772 terminator = -1;
5773 ptr += 2;
5774 code[1+LINK_SIZE] = OP_RREF; /* Change the type of test */
5775 }
5776
5777 /* Check for a test for a named group's having been set, using the Perl
5778 syntax (?(<name>) or (?('name') */
5779
5780 else if (ptr[1] == CHAR_LESS_THAN_SIGN)
5781 {
5782 terminator = CHAR_GREATER_THAN_SIGN;
5783 ptr++;
5784 }
5785 else if (ptr[1] == CHAR_APOSTROPHE)
5786 {
5787 terminator = CHAR_APOSTROPHE;
5788 ptr++;
5789 }
5790 else
5791 {
5792 terminator = 0;
5793 if (ptr[1] == CHAR_MINUS || ptr[1] == CHAR_PLUS) refsign = *(++ptr);
5794 }
5795
5796 /* We now expect to read a name; any thing else is an error */
5797
5798 if (!MAX_255(ptr[1]) || (cd->ctypes[ptr[1]] & ctype_word) == 0)
5799 {
5800 ptr += 1; /* To get the right offset */
5801 *errorcodeptr = ERR28;
5802 goto FAILED;
5803 }
5804
5805 /* Read the name, but also get it as a number if it's all digits */
5806
5807 recno = 0;
5808 name = ++ptr;
5809 while (MAX_255(*ptr) && (cd->ctypes[*ptr] & ctype_word) != 0)
5810 {
5811 if (recno >= 0)
5812 recno = (IS_DIGIT(*ptr))? recno * 10 + *ptr - CHAR_0 : -1;
5813 ptr++;
5814 }
5815 namelen = (int)(ptr - name);
5816
5817 if ((terminator > 0 && *ptr++ != terminator) ||
5818 *ptr++ != CHAR_RIGHT_PARENTHESIS)
5819 {
5820 ptr--; /* Error offset */
5821 *errorcodeptr = ERR26;
5822 goto FAILED;
5823 }
5824
5825 /* Do no further checking in the pre-compile phase. */
5826
5827 if (lengthptr != NULL) break;
5828
5829 /* In the real compile we do the work of looking for the actual
5830 reference. If the string started with "+" or "-" we require the rest to
5831 be digits, in which case recno will be set. */
5832
5833 if (refsign > 0)
5834 {
5835 if (recno <= 0)
5836 {
5837 *errorcodeptr = ERR58;
5838 goto FAILED;
5839 }
5840 recno = (refsign == CHAR_MINUS)?
5841 cd->bracount - recno + 1 : recno +cd->bracount;
5842 if (recno <= 0 || recno > cd->final_bracount)
5843 {
5844 *errorcodeptr = ERR15;
5845 goto FAILED;
5846 }
5847 PUT2(code, 2+LINK_SIZE, recno);
5848 break;
5849 }
5850
5851 /* Otherwise (did not start with "+" or "-"), start by looking for the
5852 name. If we find a name, add one to the opcode to change OP_CREF or
5853 OP_RREF into OP_NCREF or OP_NRREF. These behave exactly the same,
5854 except they record that the reference was originally to a name. The
5855 information is used to check duplicate names. */
5856
5857 slot = cd->name_table;
5858 for (i = 0; i < cd->names_found; i++)
5859 {
5860 if (STRNCMP_UC_UC(name, slot+IMM2_SIZE, namelen) == 0) break;
5861 slot += cd->name_entry_size;
5862 }
5863
5864 /* Found a previous named subpattern */
5865
5866 if (i < cd->names_found)
5867 {
5868 recno = GET2(slot, 0);
5869 PUT2(code, 2+LINK_SIZE, recno);
5870 code[1+LINK_SIZE]++;
5871 }
5872
5873 /* Search the pattern for a forward reference */
5874
5875 else if ((i = find_parens(cd, name, namelen,
5876 (options & PCRE_EXTENDED) != 0, utf)) > 0)
5877 {
5878 PUT2(code, 2+LINK_SIZE, i);
5879 code[1+LINK_SIZE]++;
5880 }
5881
5882 /* If terminator == 0 it means that the name followed directly after
5883 the opening parenthesis [e.g. (?(abc)...] and in this case there are
5884 some further alternatives to try. For the cases where terminator != 0
5885 [things like (?(<name>... or (?('name')... or (?(R&name)... ] we have
5886 now checked all the possibilities, so give an error. */
5887
5888 else if (terminator != 0)
5889 {
5890 *errorcodeptr = ERR15;
5891 goto FAILED;
5892 }
5893
5894 /* Check for (?(R) for recursion. Allow digits after R to specify a
5895 specific group number. */
5896
5897 else if (*name == CHAR_R)
5898 {
5899 recno = 0;
5900 for (i = 1; i < namelen; i++)
5901 {
5902 if (!IS_DIGIT(name[i]))
5903 {
5904 *errorcodeptr = ERR15;
5905 goto FAILED;
5906 }
5907 recno = recno * 10 + name[i] - CHAR_0;
5908 }
5909 if (recno == 0) recno = RREF_ANY;
5910 code[1+LINK_SIZE] = OP_RREF; /* Change test type */
5911 PUT2(code, 2+LINK_SIZE, recno);
5912 }
5913
5914 /* Similarly, check for the (?(DEFINE) "condition", which is always
5915 false. */
5916
5917 else if (namelen == 6 && STRNCMP_UC_C8(name, STRING_DEFINE, 6) == 0)
5918 {
5919 code[1+LINK_SIZE] = OP_DEF;
5920 skipbytes = 1;
5921 }
5922
5923 /* Check for the "name" actually being a subpattern number. We are
5924 in the second pass here, so final_bracount is set. */
5925
5926 else if (recno > 0 && recno <= cd->final_bracount)
5927 {
5928 PUT2(code, 2+LINK_SIZE, recno);
5929 }
5930
5931 /* Either an unidentified subpattern, or a reference to (?(0) */
5932
5933 else
5934 {
5935 *errorcodeptr = (recno == 0)? ERR35: ERR15;
5936 goto FAILED;
5937 }
5938 break;
5939
5940
5941 /* ------------------------------------------------------------ */
5942 case CHAR_EQUALS_SIGN: /* Positive lookahead */
5943 bravalue = OP_ASSERT;
5944 cd->assert_depth += 1;
5945 ptr++;
5946 break;
5947
5948
5949 /* ------------------------------------------------------------ */
5950 case CHAR_EXCLAMATION_MARK: /* Negative lookahead */
5951 ptr++;
5952 if (*ptr == CHAR_RIGHT_PARENTHESIS) /* Optimize (?!) */
5953 {
5954 *code++ = OP_FAIL;
5955 previous = NULL;
5956 continue;
5957 }
5958 bravalue = OP_ASSERT_NOT;
5959 cd->assert_depth += 1;
5960 break;
5961
5962
5963 /* ------------------------------------------------------------ */
5964 case CHAR_LESS_THAN_SIGN: /* Lookbehind or named define */
5965 switch (ptr[1])
5966 {
5967 case CHAR_EQUALS_SIGN: /* Positive lookbehind */
5968 bravalue = OP_ASSERTBACK;
5969 cd->assert_depth += 1;
5970 ptr += 2;
5971 break;
5972
5973 case CHAR_EXCLAMATION_MARK: /* Negative lookbehind */
5974 bravalue = OP_ASSERTBACK_NOT;
5975 cd->assert_depth += 1;
5976 ptr += 2;
5977 break;
5978
5979 default: /* Could be name define, else bad */
5980 if (MAX_255(ptr[1]) && (cd->ctypes[ptr[1]] & ctype_word) != 0)
5981 goto DEFINE_NAME;
5982 ptr++; /* Correct offset for error */
5983 *errorcodeptr = ERR24;
5984 goto FAILED;
5985 }
5986 break;
5987
5988
5989 /* ------------------------------------------------------------ */
5990 case CHAR_GREATER_THAN_SIGN: /* One-time brackets */
5991 bravalue = OP_ONCE;
5992 ptr++;
5993 break;
5994
5995
5996 /* ------------------------------------------------------------ */
5997 case CHAR_C: /* Callout - may be followed by digits; */
5998 previous_callout = code; /* Save for later completion */
5999 after_manual_callout = 1; /* Skip one item before completing */
6000 *code++ = OP_CALLOUT;
6001 {
6002 int n = 0;
6003 ptr++;
6004 while(IS_DIGIT(*ptr))
6005 n = n * 10 + *ptr++ - CHAR_0;
6006 if (*ptr != CHAR_RIGHT_PARENTHESIS)
6007 {
6008 *errorcodeptr = ERR39;
6009 goto FAILED;
6010 }
6011 if (n > 255)
6012 {
6013 *errorcodeptr = ERR38;
6014 goto FAILED;
6015 }
6016 *code++ = n;
6017 PUT(code, 0, (int)(ptr - cd->start_pattern + 1)); /* Pattern offset */
6018 PUT(code, LINK_SIZE, 0); /* Default length */
6019 code += 2 * LINK_SIZE;
6020 }
6021 previous = NULL;
6022 continue;
6023
6024
6025 /* ------------------------------------------------------------ */
6026 case CHAR_P: /* Python-style named subpattern handling */
6027 if (*(++ptr) == CHAR_EQUALS_SIGN ||
6028 *ptr == CHAR_GREATER_THAN_SIGN) /* Reference or recursion */
6029 {
6030 is_recurse = *ptr == CHAR_GREATER_THAN_SIGN;
6031 terminator = CHAR_RIGHT_PARENTHESIS;
6032 goto NAMED_REF_OR_RECURSE;
6033 }
6034 else if (*ptr != CHAR_LESS_THAN_SIGN) /* Test for Python-style defn */
6035 {
6036 *errorcodeptr = ERR41;
6037 goto FAILED;
6038 }
6039 /* Fall through to handle (?P< as (?< is handled */
6040
6041
6042 /* ------------------------------------------------------------ */
6043 DEFINE_NAME: /* Come here from (?< handling */
6044 case CHAR_APOSTROPHE:
6045 {
6046 terminator = (*ptr == CHAR_LESS_THAN_SIGN)?
6047 CHAR_GREATER_THAN_SIGN : CHAR_APOSTROPHE;
6048 name = ++ptr;
6049
6050 while (MAX_255(*ptr) && (cd->ctypes[*ptr] & ctype_word) != 0) ptr++;
6051 namelen = (int)(ptr - name);
6052
6053 /* In the pre-compile phase, just do a syntax check. */
6054
6055 if (lengthptr != NULL)
6056 {
6057 if (*ptr != terminator)
6058 {
6059 *errorcodeptr = ERR42;
6060 goto FAILED;
6061 }
6062 if (cd->names_found >= MAX_NAME_COUNT)
6063 {
6064 *errorcodeptr = ERR49;
6065 goto FAILED;
6066 }
6067 if (namelen + IMM2_SIZE + 1 > cd->name_entry_size)
6068 {
6069 cd->name_entry_size = namelen + IMM2_SIZE + 1;
6070 if (namelen > MAX_NAME_SIZE)
6071 {
6072 *errorcodeptr = ERR48;
6073 goto FAILED;
6074 }
6075 }
6076 }
6077
6078 /* In the real compile, create the entry in the table, maintaining
6079 alphabetical order. Duplicate names for different numbers are
6080 permitted only if PCRE_DUPNAMES is set. Duplicate names for the same
6081 number are always OK. (An existing number can be re-used if (?|
6082 appears in the pattern.) In either event, a duplicate name results in
6083 a duplicate entry in the table, even if the number is the same. This
6084 is because the number of names, and hence the table size, is computed
6085 in the pre-compile, and it affects various numbers and pointers which
6086 would all have to be modified, and the compiled code moved down, if
6087 duplicates with the same number were omitted from the table. This
6088 doesn't seem worth the hassle. However, *different* names for the
6089 same number are not permitted. */
6090
6091 else
6092 {
6093 BOOL dupname = FALSE;
6094 slot = cd->name_table;
6095
6096 for (i = 0; i < cd->names_found; i++)
6097 {
6098 int crc = memcmp(name, slot+IMM2_SIZE, IN_UCHARS(namelen));
6099 if (crc == 0)
6100 {
6101 if (slot[IMM2_SIZE+namelen] == 0)
6102 {
6103 if (GET2(slot, 0) != cd->bracount + 1 &&
6104 (options & PCRE_DUPNAMES) == 0)
6105 {
6106 *errorcodeptr = ERR43;
6107 goto FAILED;
6108 }
6109 else dupname = TRUE;
6110 }
6111 else crc = -1; /* Current name is a substring */
6112 }
6113
6114 /* Make space in the table and break the loop for an earlier
6115 name. For a duplicate or later name, carry on. We do this for
6116 duplicates so that in the simple case (when ?(| is not used) they
6117 are in order of their numbers. */
6118
6119 if (crc < 0)
6120 {
6121 memmove(slot + cd->name_entry_size, slot,
6122 IN_UCHARS((cd->names_found - i) * cd->name_entry_size));
6123 break;
6124 }
6125
6126 /* Continue the loop for a later or duplicate name */
6127
6128 slot += cd->name_entry_size;
6129 }
6130
6131 /* For non-duplicate names, check for a duplicate number before
6132 adding the new name. */
6133
6134 if (!dupname)
6135 {
6136 pcre_uchar *cslot = cd->name_table;
6137 for (i = 0; i < cd->names_found; i++)
6138 {
6139 if (cslot != slot)
6140 {
6141 if (GET2(cslot, 0) == cd->bracount + 1)
6142 {
6143 *errorcodeptr = ERR65;
6144 goto FAILED;
6145 }
6146 }
6147 else i--;
6148 cslot += cd->name_entry_size;
6149 }
6150 }
6151
6152 PUT2(slot, 0, cd->bracount + 1);
6153 memcpy(slot + IMM2_SIZE, name, IN_UCHARS(namelen));
6154 slot[IMM2_SIZE + namelen] = 0;
6155 }
6156 }
6157
6158 /* In both pre-compile and compile, count the number of names we've
6159 encountered. */
6160
6161 cd->names_found++;
6162 ptr++; /* Move past > or ' */
6163 goto NUMBERED_GROUP;
6164
6165
6166 /* ------------------------------------------------------------ */
6167 case CHAR_AMPERSAND: /* Perl recursion/subroutine syntax */
6168 terminator = CHAR_RIGHT_PARENTHESIS;
6169 is_recurse = TRUE;
6170 /* Fall through */
6171
6172 /* We come here from the Python syntax above that handles both
6173 references (?P=name) and recursion (?P>name), as well as falling
6174 through from the Perl recursion syntax (?&name). We also come here from
6175 the Perl \k<name> or \k'name' back reference syntax and the \k{name}
6176 .NET syntax, and the Oniguruma \g<...> and \g'...' subroutine syntax. */
6177
6178 NAMED_REF_OR_RECURSE:
6179 name = ++ptr;
6180 while (MAX_255(*ptr) && (cd->ctypes[*ptr] & ctype_word) != 0) ptr++;
6181 namelen = (int)(ptr - name);
6182
6183 /* In the pre-compile phase, do a syntax check. We used to just set
6184 a dummy reference number, because it was not used in the first pass.
6185 However, with the change of recursive back references to be atomic,
6186 we have to look for the number so that this state can be identified, as
6187 otherwise the incorrect length is computed. If it's not a backwards
6188 reference, the dummy number will do. */
6189
6190 if (lengthptr != NULL)
6191 {
6192 const pcre_uchar *temp;
6193
6194 if (namelen == 0)
6195 {
6196 *errorcodeptr = ERR62;
6197 goto FAILED;
6198 }
6199 if (*ptr != terminator)
6200 {
6201 *errorcodeptr = ERR42;
6202 goto FAILED;
6203 }
6204 if (namelen > MAX_NAME_SIZE)
6205 {
6206 *errorcodeptr = ERR48;
6207 goto FAILED;
6208 }
6209
6210 /* The name table does not exist in the first pass, so we cannot
6211 do a simple search as in the code below. Instead, we have to scan the
6212 pattern to find the number. It is important that we scan it only as
6213 far as we have got because the syntax of named subpatterns has not
6214 been checked for the rest of the pattern, and find_parens() assumes
6215 correct syntax. In any case, it's a waste of resources to scan
6216 further. We stop the scan at the current point by temporarily
6217 adjusting the value of cd->endpattern. */
6218
6219 temp = cd->end_pattern;
6220 cd->end_pattern = ptr;
6221 recno = find_parens(cd, name, namelen,
6222 (options & PCRE_EXTENDED) != 0, utf);
6223 cd->end_pattern = temp;
6224 if (recno < 0) recno = 0; /* Forward ref; set dummy number */
6225 }
6226
6227 /* In the real compile, seek the name in the table. We check the name
6228 first, and then check that we have reached the end of the name in the
6229 table. That way, if the name that is longer than any in the table,
6230 the comparison will fail without reading beyond the table entry. */
6231
6232 else
6233 {
6234 slot = cd->name_table;
6235 for (i = 0; i < cd->names_found; i++)
6236 {
6237 if (STRNCMP_UC_UC(name, slot+IMM2_SIZE, namelen) == 0 &&
6238 slot[IMM2_SIZE+namelen] == 0)
6239 break;
6240 slot += cd->name_entry_size;
6241 }
6242
6243 if (i < cd->names_found) /* Back reference */
6244 {
6245 recno = GET2(slot, 0);
6246 }
6247 else if ((recno = /* Forward back reference */
6248 find_parens(cd, name, namelen,
6249 (options & PCRE_EXTENDED) != 0, utf)) <= 0)
6250 {
6251 *errorcodeptr = ERR15;
6252 goto FAILED;
6253 }
6254 }
6255
6256 /* In both phases, we can now go to the code than handles numerical
6257 recursion or backreferences. */
6258
6259 if (is_recurse) goto HANDLE_RECURSION;
6260 else goto HANDLE_REFERENCE;
6261
6262
6263 /* ------------------------------------------------------------ */
6264 case CHAR_R: /* Recursion */
6265 ptr++; /* Same as (?0) */
6266 /* Fall through */
6267
6268
6269 /* ------------------------------------------------------------ */
6270 case CHAR_MINUS: case CHAR_PLUS: /* Recursion or subroutine */
6271 case CHAR_0: case CHAR_1: case CHAR_2: case CHAR_3: case CHAR_4:
6272 case CHAR_5: case CHAR_6: case CHAR_7: case CHAR_8: case CHAR_9:
6273 {
6274 const pcre_uchar *called;
6275 terminator = CHAR_RIGHT_PARENTHESIS;
6276
6277 /* Come here from the \g<...> and \g'...' code (Oniguruma
6278 compatibility). However, the syntax has been checked to ensure that
6279 the ... are a (signed) number, so that neither ERR63 nor ERR29 will
6280 be called on this path, nor with the jump to OTHER_CHAR_AFTER_QUERY
6281 ever be taken. */
6282
6283 HANDLE_NUMERICAL_RECURSION:
6284
6285 if ((refsign = *ptr) == CHAR_PLUS)
6286 {
6287 ptr++;
6288 if (!IS_DIGIT(*ptr))
6289 {
6290 *errorcodeptr = ERR63;
6291 goto FAILED;
6292 }
6293 }
6294 else if (refsign == CHAR_MINUS)
6295 {
6296 if (!IS_DIGIT(ptr[1]))
6297 goto OTHER_CHAR_AFTER_QUERY;
6298 ptr++;
6299 }
6300
6301 recno = 0;
6302 while(IS_DIGIT(*ptr))
6303 recno = recno * 10 + *ptr++ - CHAR_0;
6304
6305 if (*ptr != terminator)
6306 {
6307 *errorcodeptr = ERR29;
6308 goto FAILED;
6309 }
6310
6311 if (refsign == CHAR_MINUS)
6312 {
6313 if (recno == 0)
6314 {
6315 *errorcodeptr = ERR58;
6316 goto FAILED;
6317 }
6318 recno = cd->bracount - recno + 1;
6319 if (recno <= 0)
6320 {
6321 *errorcodeptr = ERR15;
6322 goto FAILED;
6323 }
6324 }
6325 else if (refsign == CHAR_PLUS)
6326 {
6327 if (recno == 0)
6328 {
6329 *errorcodeptr = ERR58;
6330 goto FAILED;
6331 }
6332 recno += cd->bracount;
6333 }
6334
6335 /* Come here from code above that handles a named recursion */
6336
6337 HANDLE_RECURSION:
6338
6339 previous = code;
6340 called = cd->start_code;
6341
6342 /* When we are actually compiling, find the bracket that is being
6343 referenced. Temporarily end the regex in case it doesn't exist before
6344 this point. If we end up with a forward reference, first check that
6345 the bracket does occur later so we can give the error (and position)
6346 now. Then remember this forward reference in the workspace so it can
6347 be filled in at the end. */
6348
6349 if (lengthptr == NULL)
6350 {
6351 *code = OP_END;
6352 if (recno != 0)
6353 called = PRIV(find_bracket)(cd->start_code, utf, recno);
6354
6355 /* Forward reference */
6356
6357 if (called == NULL)
6358 {
6359 if (find_parens(cd, NULL, recno,
6360 (options & PCRE_EXTENDED) != 0, utf) < 0)
6361 {
6362 *errorcodeptr = ERR15;
6363 goto FAILED;
6364 }
6365
6366 /* Fudge the value of "called" so that when it is inserted as an
6367 offset below, what it actually inserted is the reference number
6368 of the group. Then remember the forward reference. */
6369
6370 called = cd->start_code + recno;
6371 if (cd->hwm >= cd->start_workspace + cd->workspace_size -
6372 WORK_SIZE_SAFETY_MARGIN)
6373 {
6374 *errorcodeptr = expand_workspace(cd);
6375 if (*errorcodeptr != 0) goto FAILED;
6376 }
6377 PUTINC(cd->hwm, 0, (int)(code + 1 - cd->start_code));
6378 }
6379
6380 /* If not a forward reference, and the subpattern is still open,
6381 this is a recursive call. We check to see if this is a left
6382 recursion that could loop for ever, and diagnose that case. We
6383 must not, however, do this check if we are in a conditional
6384 subpattern because the condition might be testing for recursion in
6385 a pattern such as /(?(R)a+|(?R)b)/, which is perfectly valid.
6386 Forever loops are also detected at runtime, so those that occur in
6387 conditional subpatterns will be picked up then. */
6388
6389 else if (GET(called, 1) == 0 && cond_depth <= 0 &&
6390 could_be_empty(called, code, bcptr, utf, cd))
6391 {
6392 *errorcodeptr = ERR40;
6393 goto FAILED;
6394 }
6395 }
6396
6397 /* Insert the recursion/subroutine item. It does not have a set first
6398 character (relevant if it is repeated, because it will then be
6399 wrapped with ONCE brackets). */
6400
6401 *code = OP_RECURSE;
6402 PUT(code, 1, (int)(called - cd->start_code));
6403 code += 1 + LINK_SIZE;
6404 groupsetfirstchar = FALSE;
6405 }
6406
6407 /* Can't determine a first byte now */
6408
6409 if (firstchar == REQ_UNSET) firstchar = REQ_NONE;
6410 continue;
6411
6412
6413 /* ------------------------------------------------------------ */
6414 default: /* Other characters: check option setting */
6415 OTHER_CHAR_AFTER_QUERY:
6416 set = unset = 0;
6417 optset = &set;
6418
6419 while (*ptr != CHAR_RIGHT_PARENTHESIS && *ptr != CHAR_COLON)
6420 {
6421 switch (*ptr++)
6422 {
6423 case CHAR_MINUS: optset = &unset; break;
6424
6425 case CHAR_J: /* Record that it changed in the external options */
6426 *optset |= PCRE_DUPNAMES;
6427 cd->external_flags |= PCRE_JCHANGED;
6428 break;
6429
6430 case CHAR_i: *optset |= PCRE_CASELESS; break;
6431 case CHAR_m: *optset |= PCRE_MULTILINE; break;
6432 case CHAR_s: *optset |= PCRE_DOTALL; break;
6433 case CHAR_x: *optset |= PCRE_EXTENDED; break;
6434 case CHAR_U: *optset |= PCRE_UNGREEDY; break;
6435 case CHAR_X: *optset |= PCRE_EXTRA; break;
6436
6437 default: *errorcodeptr = ERR12;
6438 ptr--; /* Correct the offset */
6439 goto FAILED;
6440 }
6441 }
6442
6443 /* Set up the changed option bits, but don't change anything yet. */
6444
6445 newoptions = (options | set) & (~unset);
6446
6447 /* If the options ended with ')' this is not the start of a nested
6448 group with option changes, so the options change at this level. If this
6449 item is right at the start of the pattern, the options can be
6450 abstracted and made external in the pre-compile phase, and ignored in
6451 the compile phase. This can be helpful when matching -- for instance in
6452 caseless checking of required bytes.
6453
6454 If the code pointer is not (cd->start_code + 1 + LINK_SIZE), we are
6455 definitely *not* at the start of the pattern because something has been
6456 compiled. In the pre-compile phase, however, the code pointer can have
6457 that value after the start, because it gets reset as code is discarded
6458 during the pre-compile. However, this can happen only at top level - if
6459 we are within parentheses, the starting BRA will still be present. At
6460 any parenthesis level, the length value can be used to test if anything
6461 has been compiled at that level. Thus, a test for both these conditions
6462 is necessary to ensure we correctly detect the start of the pattern in
6463 both phases.
6464
6465 If we are not at the pattern start, reset the greedy defaults and the
6466 case value for firstchar and reqchar. */
6467
6468 if (*ptr == CHAR_RIGHT_PARENTHESIS)
6469 {
6470 if (code == cd->start_code + 1 + LINK_SIZE &&
6471 (lengthptr == NULL || *lengthptr == 2 + 2*LINK_SIZE))
6472 {
6473 cd->external_options = newoptions;
6474 }
6475 else
6476 {
6477 greedy_default = ((newoptions & PCRE_UNGREEDY) != 0);
6478 greedy_non_default = greedy_default ^ 1;
6479 req_caseopt = ((newoptions & PCRE_CASELESS) != 0)? REQ_CASELESS:0;
6480 }
6481
6482 /* Change options at this level, and pass them back for use
6483 in subsequent branches. */
6484
6485 *optionsptr = options = newoptions;
6486 previous = NULL; /* This item can't be repeated */
6487 continue; /* It is complete */
6488 }
6489
6490 /* If the options ended with ':' we are heading into a nested group
6491 with possible change of options. Such groups are non-capturing and are
6492 not assertions of any kind. All we need to do is skip over the ':';
6493 the newoptions value is handled below. */
6494
6495 bravalue = OP_BRA;
6496 ptr++;
6497 } /* End of switch for character following (? */
6498 } /* End of (? handling */
6499
6500 /* Opening parenthesis not followed by '*' or '?'. If PCRE_NO_AUTO_CAPTURE
6501 is set, all unadorned brackets become non-capturing and behave like (?:...)
6502 brackets. */
6503
6504 else if ((options & PCRE_NO_AUTO_CAPTURE) != 0)
6505 {
6506 bravalue = OP_BRA;
6507 }
6508
6509 /* Else we have a capturing group. */
6510
6511 else
6512 {
6513 NUMBERED_GROUP:
6514 cd->bracount += 1;
6515 PUT2(code, 1+LINK_SIZE, cd->bracount);
6516 skipbytes = IMM2_SIZE;
6517 }
6518
6519 /* Process nested bracketed regex. Assertions used not to be repeatable,
6520 but this was changed for Perl compatibility, so all kinds can now be
6521 repeated. We copy code into a non-register variable (tempcode) in order to
6522 be able to pass its address because some compilers complain otherwise. */
6523
6524 previous = code; /* For handling repetition */
6525 *code = bravalue;
6526 tempcode = code;
6527 tempreqvary = cd->req_varyopt; /* Save value before bracket */
6528 tempbracount = cd->bracount; /* Save value before bracket */
6529 length_prevgroup = 0; /* Initialize for pre-compile phase */
6530
6531 if (!compile_regex(
6532 newoptions, /* The complete new option state */
6533 &tempcode, /* Where to put code (updated) */
6534 &ptr, /* Input pointer (updated) */
6535 errorcodeptr, /* Where to put an error message */
6536 (bravalue == OP_ASSERTBACK ||
6537 bravalue == OP_ASSERTBACK_NOT), /* TRUE if back assert */
6538 reset_bracount, /* True if (?| group */
6539 skipbytes, /* Skip over bracket number */
6540 cond_depth +
6541 ((bravalue == OP_COND)?1:0), /* Depth of condition subpatterns */
6542 &subfirstchar, /* For possible first char */
6543 &subreqchar, /* For possible last char */
6544 bcptr, /* Current branch chain */
6545 cd, /* Tables block */
6546 (lengthptr == NULL)? NULL : /* Actual compile phase */
6547 &length_prevgroup /* Pre-compile phase */
6548 ))
6549 goto FAILED;
6550
6551 /* If this was an atomic group and there are no capturing groups within it,
6552 generate OP_ONCE_NC instead of OP_ONCE. */
6553
6554 if (bravalue == OP_ONCE && cd->bracount <= tempbracount)
6555 *code = OP_ONCE_NC;
6556
6557 if (bravalue >= OP_ASSERT && bravalue <= OP_ASSERTBACK_NOT)
6558 cd->assert_depth -= 1;
6559
6560 /* At the end of compiling, code is still pointing to the start of the
6561 group, while tempcode has been updated to point past the end of the group.
6562 The pattern pointer (ptr) is on the bracket.
6563
6564 If this is a conditional bracket, check that there are no more than
6565 two branches in the group, or just one if it's a DEFINE group. We do this
6566 in the real compile phase, not in the pre-pass, where the whole group may
6567 not be available. */
6568
6569 if (bravalue == OP_COND && lengthptr == NULL)
6570 {
6571 pcre_uchar *tc = code;
6572 int condcount = 0;
6573
6574 do {
6575 condcount++;
6576 tc += GET(tc,1);
6577 }
6578 while (*tc != OP_KET);
6579
6580 /* A DEFINE group is never obeyed inline (the "condition" is always
6581 false). It must have only one branch. */
6582
6583 if (code[LINK_SIZE+1] == OP_DEF)
6584 {
6585 if (condcount > 1)
6586 {
6587 *errorcodeptr = ERR54;
6588 goto FAILED;
6589 }
6590 bravalue = OP_DEF; /* Just a flag to suppress char handling below */
6591 }
6592
6593 /* A "normal" conditional group. If there is just one branch, we must not
6594 make use of its firstchar or reqchar, because this is equivalent to an
6595 empty second branch. */
6596
6597 else
6598 {
6599 if (condcount > 2)
6600 {
6601 *errorcodeptr = ERR27;
6602 goto FAILED;
6603 }
6604 if (condcount == 1) subfirstchar = subreqchar = REQ_NONE;
6605 }
6606 }
6607
6608 /* Error if hit end of pattern */
6609
6610 if (*ptr != CHAR_RIGHT_PARENTHESIS)
6611 {
6612 *errorcodeptr = ERR14;
6613 goto FAILED;
6614 }
6615
6616 /* In the pre-compile phase, update the length by the length of the group,
6617 less the brackets at either end. Then reduce the compiled code to just a
6618 set of non-capturing brackets so that it doesn't use much memory if it is
6619 duplicated by a quantifier.*/
6620
6621 if (lengthptr != NULL)
6622 {
6623 if (OFLOW_MAX - *lengthptr < length_prevgroup - 2 - 2*LINK_SIZE)
6624 {
6625 *errorcodeptr = ERR20;
6626 goto FAILED;
6627 }
6628 *lengthptr += length_prevgroup - 2 - 2*LINK_SIZE;
6629 code++; /* This already contains bravalue */
6630 PUTINC(code, 0, 1 + LINK_SIZE);
6631 *code++ = OP_KET;
6632 PUTINC(code, 0, 1 + LINK_SIZE);
6633 break; /* No need to waste time with special character handling */
6634 }
6635
6636 /* Otherwise update the main code pointer to the end of the group. */
6637
6638 code = tempcode;
6639
6640 /* For a DEFINE group, required and first character settings are not
6641 relevant. */
6642