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

Contents of /code/trunk/pcre_compile.c

Parent Directory Parent Directory | Revision Log Revision Log


Revision 1033 - (show annotations)
Mon Sep 10 11:02:48 2012 UTC (7 years, 1 month ago) by ph10
File MIME type: text/plain
File size: 272041 byte(s)
General spring-clean of EBCDIC-related issues in the code, which had decayed 
over time. Also the documentation. Added one test that can be run in an ASCII
world to do a little testing of EBCDIC-related things. 
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 < CHAR_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 CHAR_HT:
3172 case CHAR_SPACE:
3173 #ifndef EBCDIC
3174 case 0xa0:
3175 case 0x1680:
3176 case 0x180e:
3177 case 0x2000:
3178 case 0x2001:
3179 case 0x2002:
3180 case 0x2003:
3181 case 0x2004:
3182 case 0x2005:
3183 case 0x2006:
3184 case 0x2007:
3185 case 0x2008:
3186 case 0x2009:
3187 case 0x200A:
3188 case 0x202f:
3189 case 0x205f:
3190 case 0x3000:
3191 #endif /* Not EBCDIC */
3192 return op_code == OP_NOT_HSPACE;
3193 default:
3194 return op_code != OP_NOT_HSPACE;
3195 }
3196
3197 case OP_ANYNL:
3198 case OP_VSPACE:
3199 case OP_NOT_VSPACE:
3200 switch(next)
3201 {
3202 case CHAR_LF:
3203 case CHAR_VT:
3204 case CHAR_FF:
3205 case CHAR_CR:
3206 case CHAR_NEL:
3207 #ifndef EBCDIC
3208 case 0x2028:
3209 case 0x2029:
3210 #endif
3211 return op_code == OP_NOT_VSPACE;
3212 default:
3213 return op_code != OP_NOT_VSPACE;
3214 }
3215
3216 #ifdef SUPPORT_UCP
3217 case OP_PROP:
3218 return check_char_prop(next, previous[0], previous[1], FALSE);
3219
3220 case OP_NOTPROP:
3221 return check_char_prop(next, previous[0], previous[1], TRUE);
3222 #endif
3223
3224 default:
3225 return FALSE;
3226 }
3227
3228
3229 /* Handle the case when the next item is \d, \s, etc. Note that when PCRE_UCP
3230 is set, \d turns into ESC_du rather than ESC_d, etc., so ESC_d etc. are
3231 generated only when PCRE_UCP is *not* set, that is, when only ASCII
3232 characteristics are recognized. Similarly, the opcodes OP_DIGIT etc. are
3233 replaced by OP_PROP codes when PCRE_UCP is set. */
3234
3235 switch(op_code)
3236 {
3237 case OP_CHAR:
3238 case OP_CHARI:
3239 #ifdef SUPPORT_UTF
3240 GETCHARTEST(c, previous);
3241 #else
3242 c = *previous;
3243 #endif
3244 switch(-next)
3245 {
3246 case ESC_d:
3247 return c > 255 || (cd->ctypes[c] & ctype_digit) == 0;
3248
3249 case ESC_D:
3250 return c <= 255 && (cd->ctypes[c] & ctype_digit) != 0;
3251
3252 case ESC_s:
3253 return c > 255 || (cd->ctypes[c] & ctype_space) == 0;
3254
3255 case ESC_S:
3256 return c <= 255 && (cd->ctypes[c] & ctype_space) != 0;
3257
3258 case ESC_w:
3259 return c > 255 || (cd->ctypes[c] & ctype_word) == 0;
3260
3261 case ESC_W:
3262 return c <= 255 && (cd->ctypes[c] & ctype_word) != 0;
3263
3264 case ESC_h:
3265 case ESC_H:
3266 switch(c)
3267 {
3268 case CHAR_HT:
3269 case CHAR_SPACE:
3270 #ifndef EBCDIC
3271 case 0xa0:
3272 case 0x1680:
3273 case 0x180e:
3274 case 0x2000:
3275 case 0x2001:
3276 case 0x2002:
3277 case 0x2003:
3278 case 0x2004:
3279 case 0x2005:
3280 case 0x2006:
3281 case 0x2007:
3282 case 0x2008:
3283 case 0x2009:
3284 case 0x200A:
3285 case 0x202f:
3286 case 0x205f:
3287 case 0x3000:
3288 #endif /* Not EBCDIC */
3289 return -next != ESC_h;
3290 default:
3291 return -next == ESC_h;
3292 }
3293
3294 case ESC_v:
3295 case ESC_V:
3296 switch(c)
3297 {
3298 case CHAR_LF:
3299 case CHAR_VT:
3300 case CHAR_FF:
3301 case CHAR_CR:
3302 case CHAR_NEL:
3303 #ifndef EBCDIC
3304 case 0x2028:
3305 case 0x2029:
3306 #endif /* Not EBCDIC */
3307 return -next != ESC_v;
3308 default:
3309 return -next == ESC_v;
3310 }
3311
3312 /* When PCRE_UCP is set, these values get generated for \d etc. Find
3313 their substitutions and process them. The result will always be either
3314 -ESC_p or -ESC_P. Then fall through to process those values. */
3315
3316 #ifdef SUPPORT_UCP
3317 case ESC_du:
3318 case ESC_DU:
3319 case ESC_wu:
3320 case ESC_WU:
3321 case ESC_su:
3322 case ESC_SU:
3323 {
3324 int temperrorcode = 0;
3325 ptr = substitutes[-next - ESC_DU];
3326 next = check_escape(&ptr, &temperrorcode, 0, options, FALSE);
3327 if (temperrorcode != 0) return FALSE;
3328 ptr++; /* For compatibility */
3329 }
3330 /* Fall through */
3331
3332 case ESC_p:
3333 case ESC_P:
3334 {
3335 int ptype, pdata, errorcodeptr;
3336 BOOL negated;
3337
3338 ptr--; /* Make ptr point at the p or P */
3339 ptype = get_ucp(&ptr, &negated, &pdata, &errorcodeptr);
3340 if (ptype < 0) return FALSE;
3341 ptr++; /* Point past the final curly ket */
3342
3343 /* If the property item is optional, we have to give up. (When generated
3344 from \d etc by PCRE_UCP, this test will have been applied much earlier,
3345 to the original \d etc. At this point, ptr will point to a zero byte. */
3346
3347 if (*ptr == CHAR_ASTERISK || *ptr == CHAR_QUESTION_MARK ||
3348 STRNCMP_UC_C8(ptr, STR_LEFT_CURLY_BRACKET STR_0 STR_COMMA, 3) == 0)
3349 return FALSE;
3350
3351 /* Do the property check. */
3352
3353 return check_char_prop(c, ptype, pdata, (next == -ESC_P) != negated);
3354 }
3355 #endif
3356
3357 default:
3358 return FALSE;
3359 }
3360
3361 /* In principle, support for Unicode properties should be integrated here as
3362 well. It means re-organizing the above code so as to get hold of the property
3363 values before switching on the op-code. However, I wonder how many patterns
3364 combine ASCII \d etc with Unicode properties? (Note that if PCRE_UCP is set,
3365 these op-codes are never generated.) */
3366
3367 case OP_DIGIT:
3368 return next == -ESC_D || next == -ESC_s || next == -ESC_W ||
3369 next == -ESC_h || next == -ESC_v || next == -ESC_R;
3370
3371 case OP_NOT_DIGIT:
3372 return next == -ESC_d;
3373
3374 case OP_WHITESPACE:
3375 return next == -ESC_S || next == -ESC_d || next == -ESC_w;
3376
3377 case OP_NOT_WHITESPACE:
3378 return next == -ESC_s || next == -ESC_h || next == -ESC_v || next == -ESC_R;
3379
3380 case OP_HSPACE:
3381 return next == -ESC_S || next == -ESC_H || next == -ESC_d ||
3382 next == -ESC_w || next == -ESC_v || next == -ESC_R;
3383
3384 case OP_NOT_HSPACE:
3385 return next == -ESC_h;
3386
3387 /* Can't have \S in here because VT matches \S (Perl anomaly) */
3388 case OP_ANYNL:
3389 case OP_VSPACE:
3390 return next == -ESC_V || next == -ESC_d || next == -ESC_w;
3391
3392 case OP_NOT_VSPACE:
3393 return next == -ESC_v || next == -ESC_R;
3394
3395 case OP_WORDCHAR:
3396 return next == -ESC_W || next == -ESC_s || next == -ESC_h ||
3397 next == -ESC_v || next == -ESC_R;
3398
3399 case OP_NOT_WORDCHAR:
3400 return next == -ESC_w || next == -ESC_d;
3401
3402 default:
3403 return FALSE;
3404 }
3405
3406 /* Control does not reach here */
3407 }
3408
3409
3410
3411 /*************************************************
3412 * Compile one branch *
3413 *************************************************/
3414
3415 /* Scan the pattern, compiling it into the a vector. If the options are
3416 changed during the branch, the pointer is used to change the external options
3417 bits. This function is used during the pre-compile phase when we are trying
3418 to find out the amount of memory needed, as well as during the real compile
3419 phase. The value of lengthptr distinguishes the two phases.
3420
3421 Arguments:
3422 optionsptr pointer to the option bits
3423 codeptr points to the pointer to the current code point
3424 ptrptr points to the current pattern pointer
3425 errorcodeptr points to error code variable
3426 firstcharptr set to initial literal character, or < 0 (REQ_UNSET, REQ_NONE)
3427 reqcharptr set to the last literal character required, else < 0
3428 bcptr points to current branch chain
3429 cond_depth conditional nesting depth
3430 cd contains pointers to tables etc.
3431 lengthptr NULL during the real compile phase
3432 points to length accumulator during pre-compile phase
3433
3434 Returns: TRUE on success
3435 FALSE, with *errorcodeptr set non-zero on error
3436 */
3437
3438 static BOOL
3439 compile_branch(int *optionsptr, pcre_uchar **codeptr,
3440 const pcre_uchar **ptrptr, int *errorcodeptr, pcre_int32 *firstcharptr,
3441 pcre_int32 *reqcharptr, branch_chain *bcptr, int cond_depth,
3442 compile_data *cd, int *lengthptr)
3443 {
3444 int repeat_type, op_type;
3445 int repeat_min = 0, repeat_max = 0; /* To please picky compilers */
3446 int bravalue = 0;
3447 int greedy_default, greedy_non_default;
3448 pcre_int32 firstchar, reqchar;
3449 pcre_int32 zeroreqchar, zerofirstchar;
3450 pcre_int32 req_caseopt, reqvary, tempreqvary;
3451 int options = *optionsptr; /* May change dynamically */
3452 int after_manual_callout = 0;
3453 int length_prevgroup = 0;
3454 register int c;
3455 register pcre_uchar *code = *codeptr;
3456 pcre_uchar *last_code = code;
3457 pcre_uchar *orig_code = code;
3458 pcre_uchar *tempcode;
3459 BOOL inescq = FALSE;
3460 BOOL groupsetfirstchar = FALSE;
3461 const pcre_uchar *ptr = *ptrptr;
3462 const pcre_uchar *tempptr;
3463 const pcre_uchar *nestptr = NULL;
3464 pcre_uchar *previous = NULL;
3465 pcre_uchar *previous_callout = NULL;
3466 pcre_uchar *save_hwm = NULL;
3467 pcre_uint8 classbits[32];
3468
3469 /* We can fish out the UTF-8 setting once and for all into a BOOL, but we
3470 must not do this for other options (e.g. PCRE_EXTENDED) because they may change
3471 dynamically as we process the pattern. */
3472
3473 #ifdef SUPPORT_UTF
3474 /* PCRE_UTF16 has the same value as PCRE_UTF8. */
3475 BOOL utf = (options & PCRE_UTF8) != 0;
3476 pcre_uchar utf_chars[6];
3477 #else
3478 BOOL utf = FALSE;
3479 #endif
3480
3481 /* Helper variables for OP_XCLASS opcode (for characters > 255). */
3482
3483 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
3484 BOOL xclass;
3485 pcre_uchar *class_uchardata;
3486 pcre_uchar *class_uchardata_base;
3487 #endif
3488
3489 #ifdef PCRE_DEBUG
3490 if (lengthptr != NULL) DPRINTF((">> start branch\n"));
3491 #endif
3492
3493 /* Set up the default and non-default settings for greediness */
3494
3495 greedy_default = ((options & PCRE_UNGREEDY) != 0);
3496 greedy_non_default = greedy_default ^ 1;
3497
3498 /* Initialize no first byte, no required byte. REQ_UNSET means "no char
3499 matching encountered yet". It gets changed to REQ_NONE if we hit something that
3500 matches a non-fixed char first char; reqchar just remains unset if we never
3501 find one.
3502
3503 When we hit a repeat whose minimum is zero, we may have to adjust these values
3504 to take the zero repeat into account. This is implemented by setting them to
3505 zerofirstbyte and zeroreqchar when such a repeat is encountered. The individual
3506 item types that can be repeated set these backoff variables appropriately. */
3507
3508 firstchar = reqchar = zerofirstchar = zeroreqchar = REQ_UNSET;
3509
3510 /* The variable req_caseopt contains either the REQ_CASELESS value
3511 or zero, according to the current setting of the caseless flag. The
3512 REQ_CASELESS leaves the lower 28 bit empty. It is added into the
3513 firstchar or reqchar variables to record the case status of the
3514 value. This is used only for ASCII characters. */
3515
3516 req_caseopt = ((options & PCRE_CASELESS) != 0)? REQ_CASELESS:0;
3517
3518 /* Switch on next character until the end of the branch */
3519
3520 for (;; ptr++)
3521 {
3522 BOOL negate_class;
3523 BOOL should_flip_negation;
3524 BOOL possessive_quantifier;
3525 BOOL is_quantifier;
3526 BOOL is_recurse;
3527 BOOL reset_bracount;
3528 int class_has_8bitchar;
3529 int class_single_char;
3530 int newoptions;
3531 int recno;
3532 int refsign;
3533 int skipbytes;
3534 int subreqchar;
3535 int subfirstchar;
3536 int terminator;
3537 int mclength;
3538 int tempbracount;
3539 pcre_uchar mcbuffer[8];
3540
3541 /* Get next character in the pattern */
3542
3543 c = *ptr;
3544
3545 /* If we are at the end of a nested substitution, revert to the outer level
3546 string. Nesting only happens one level deep. */
3547
3548 if (c == 0 && nestptr != NULL)
3549 {
3550 ptr = nestptr;
3551 nestptr = NULL;
3552 c = *ptr;
3553 }
3554
3555 /* If we are in the pre-compile phase, accumulate the length used for the
3556 previous cycle of this loop. */
3557
3558 if (lengthptr != NULL)
3559 {
3560 #ifdef PCRE_DEBUG
3561 if (code > cd->hwm) cd->hwm = code; /* High water info */
3562 #endif
3563 if (code > cd->start_workspace + cd->workspace_size -
3564 WORK_SIZE_SAFETY_MARGIN) /* Check for overrun */
3565 {
3566 *errorcodeptr = ERR52;
3567 goto FAILED;
3568 }
3569
3570 /* There is at least one situation where code goes backwards: this is the
3571 case of a zero quantifier after a class (e.g. [ab]{0}). At compile time,
3572 the class is simply eliminated. However, it is created first, so we have to
3573 allow memory for it. Therefore, don't ever reduce the length at this point.
3574 */
3575
3576 if (code < last_code) code = last_code;
3577
3578 /* Paranoid check for integer overflow */
3579
3580 if (OFLOW_MAX - *lengthptr < code - last_code)
3581 {
3582 *errorcodeptr = ERR20;
3583 goto FAILED;
3584 }
3585
3586 *lengthptr += (int)(code - last_code);
3587 DPRINTF(("length=%d added %d c=%c (0x%x)\n", *lengthptr,
3588 (int)(code - last_code), c, c));
3589
3590 /* If "previous" is set and it is not at the start of the work space, move
3591 it back to there, in order to avoid filling up the work space. Otherwise,
3592 if "previous" is NULL, reset the current code pointer to the start. */
3593
3594 if (previous != NULL)
3595 {
3596 if (previous > orig_code)
3597 {
3598 memmove(orig_code, previous, IN_UCHARS(code - previous));
3599 code -= previous - orig_code;
3600 previous = orig_code;
3601 }
3602 }
3603 else code = orig_code;
3604
3605 /* Remember where this code item starts so we can pick up the length
3606 next time round. */
3607
3608 last_code = code;
3609 }
3610
3611 /* In the real compile phase, just check the workspace used by the forward
3612 reference list. */
3613
3614 else if (cd->hwm > cd->start_workspace + cd->workspace_size -
3615 WORK_SIZE_SAFETY_MARGIN)
3616 {
3617 *errorcodeptr = ERR52;
3618 goto FAILED;
3619 }
3620
3621 /* If in \Q...\E, check for the end; if not, we have a literal */
3622
3623 if (inescq && c != 0)
3624 {
3625 if (c == CHAR_BACKSLASH && ptr[1] == CHAR_E)
3626 {
3627 inescq = FALSE;
3628 ptr++;
3629 continue;
3630 }
3631 else
3632 {
3633 if (previous_callout != NULL)
3634 {
3635 if (lengthptr == NULL) /* Don't attempt in pre-compile phase */
3636 complete_callout(previous_callout, ptr, cd);
3637 previous_callout = NULL;
3638 }
3639 if ((options & PCRE_AUTO_CALLOUT) != 0)
3640 {
3641 previous_callout = code;
3642 code = auto_callout(code, ptr, cd);
3643 }
3644 goto NORMAL_CHAR;
3645 }
3646 }
3647
3648 /* Fill in length of a previous callout, except when the next thing is
3649 a quantifier. */
3650
3651 is_quantifier =
3652 c == CHAR_ASTERISK || c == CHAR_PLUS || c == CHAR_QUESTION_MARK ||
3653 (c == CHAR_LEFT_CURLY_BRACKET && is_counted_repeat(ptr+1));
3654
3655 if (!is_quantifier && previous_callout != NULL &&
3656 after_manual_callout-- <= 0)
3657 {
3658 if (lengthptr == NULL) /* Don't attempt in pre-compile phase */
3659 complete_callout(previous_callout, ptr, cd);
3660 previous_callout = NULL;
3661 }
3662
3663 /* In extended mode, skip white space and comments. */
3664
3665 if ((options & PCRE_EXTENDED) != 0)
3666 {
3667 if (MAX_255(*ptr) && (cd->ctypes[c] & ctype_space) != 0) continue;
3668 if (c == CHAR_NUMBER_SIGN)
3669 {
3670 ptr++;
3671 while (*ptr != 0)
3672 {
3673 if (IS_NEWLINE(ptr)) { ptr += cd->nllen - 1; break; }
3674 ptr++;
3675 #ifdef SUPPORT_UTF
3676 if (utf) FORWARDCHAR(ptr);
3677 #endif
3678 }
3679 if (*ptr != 0) continue;
3680
3681 /* Else fall through to handle end of string */
3682 c = 0;
3683 }
3684 }
3685
3686 /* No auto callout for quantifiers. */
3687
3688 if ((options & PCRE_AUTO_CALLOUT) != 0 && !is_quantifier)
3689 {
3690 previous_callout = code;
3691 code = auto_callout(code, ptr, cd);
3692 }
3693
3694 switch(c)
3695 {
3696 /* ===================================================================*/
3697 case 0: /* The branch terminates at string end */
3698 case CHAR_VERTICAL_LINE: /* or | or ) */
3699 case CHAR_RIGHT_PARENTHESIS:
3700 *firstcharptr = firstchar;
3701 *reqcharptr = reqchar;
3702 *codeptr = code;
3703 *ptrptr = ptr;
3704 if (lengthptr != NULL)
3705 {
3706 if (OFLOW_MAX - *lengthptr < code - last_code)
3707 {
3708 *errorcodeptr = ERR20;
3709 goto FAILED;
3710 }
3711 *lengthptr += (int)(code - last_code); /* To include callout length */
3712 DPRINTF((">> end branch\n"));
3713 }
3714 return TRUE;
3715
3716
3717 /* ===================================================================*/
3718 /* Handle single-character metacharacters. In multiline mode, ^ disables
3719 the setting of any following char as a first character. */
3720
3721 case CHAR_CIRCUMFLEX_ACCENT:
3722 previous = NULL;
3723 if ((options & PCRE_MULTILINE) != 0)
3724 {
3725 if (firstchar == REQ_UNSET) firstchar = REQ_NONE;
3726 *code++ = OP_CIRCM;
3727 }
3728 else *code++ = OP_CIRC;
3729 break;
3730
3731 case CHAR_DOLLAR_SIGN:
3732 previous = NULL;
3733 *code++ = ((options & PCRE_MULTILINE) != 0)? OP_DOLLM : OP_DOLL;
3734 break;
3735
3736 /* There can never be a first char if '.' is first, whatever happens about
3737 repeats. The value of reqchar doesn't change either. */
3738
3739 case CHAR_DOT:
3740 if (firstchar == REQ_UNSET) firstchar = REQ_NONE;
3741 zerofirstchar = firstchar;
3742 zeroreqchar = reqchar;
3743 previous = code;
3744 *code++ = ((options & PCRE_DOTALL) != 0)? OP_ALLANY: OP_ANY;
3745 break;
3746
3747
3748 /* ===================================================================*/
3749 /* Character classes. If the included characters are all < 256, we build a
3750 32-byte bitmap of the permitted characters, except in the special case
3751 where there is only one such character. For negated classes, we build the
3752 map as usual, then invert it at the end. However, we use a different opcode
3753 so that data characters > 255 can be handled correctly.
3754
3755 If the class contains characters outside the 0-255 range, a different
3756 opcode is compiled. It may optionally have a bit map for characters < 256,
3757 but those above are are explicitly listed afterwards. A flag byte tells
3758 whether the bitmap is present, and whether this is a negated class or not.
3759
3760 In JavaScript compatibility mode, an isolated ']' causes an error. In
3761 default (Perl) mode, it is treated as a data character. */
3762
3763 case CHAR_RIGHT_SQUARE_BRACKET:
3764 if ((cd->external_options & PCRE_JAVASCRIPT_COMPAT) != 0)
3765 {
3766 *errorcodeptr = ERR64;
3767 goto FAILED;
3768 }
3769 goto NORMAL_CHAR;
3770
3771 case CHAR_LEFT_SQUARE_BRACKET:
3772 previous = code;
3773
3774 /* PCRE supports POSIX class stuff inside a class. Perl gives an error if
3775 they are encountered at the top level, so we'll do that too. */
3776
3777 if ((ptr[1] == CHAR_COLON || ptr[1] == CHAR_DOT ||
3778 ptr[1] == CHAR_EQUALS_SIGN) &&
3779 check_posix_syntax(ptr, &tempptr))
3780 {
3781 *errorcodeptr = (ptr[1] == CHAR_COLON)? ERR13 : ERR31;
3782 goto FAILED;
3783 }
3784
3785 /* If the first character is '^', set the negation flag and skip it. Also,
3786 if the first few characters (either before or after ^) are \Q\E or \E we
3787 skip them too. This makes for compatibility with Perl. */
3788
3789 negate_class = FALSE;
3790 for (;;)
3791 {
3792 c = *(++ptr);
3793 if (c == CHAR_BACKSLASH)
3794 {
3795 if (ptr[1] == CHAR_E)
3796 ptr++;
3797 else if (STRNCMP_UC_C8(ptr + 1, STR_Q STR_BACKSLASH STR_E, 3) == 0)
3798 ptr += 3;
3799 else
3800 break;
3801 }
3802 else if (!negate_class && c == CHAR_CIRCUMFLEX_ACCENT)
3803 negate_class = TRUE;
3804 else break;
3805 }
3806
3807 /* Empty classes are allowed in JavaScript compatibility mode. Otherwise,
3808 an initial ']' is taken as a data character -- the code below handles
3809 that. In JS mode, [] must always fail, so generate OP_FAIL, whereas
3810 [^] must match any character, so generate OP_ALLANY. */
3811
3812 if (c == CHAR_RIGHT_SQUARE_BRACKET &&
3813 (cd->external_options & PCRE_JAVASCRIPT_COMPAT) != 0)
3814 {
3815 *code++ = negate_class? OP_ALLANY : OP_FAIL;
3816 if (firstchar == REQ_UNSET) firstchar = REQ_NONE;
3817 zerofirstchar = firstchar;
3818 break;
3819 }
3820
3821 /* If a class contains a negative special such as \S, we need to flip the
3822 negation flag at the end, so that support for characters > 255 works
3823 correctly (they are all included in the class). */
3824
3825 should_flip_negation = FALSE;
3826
3827 /* For optimization purposes, we track some properties of the class.
3828 class_has_8bitchar will be non-zero, if the class contains at least one
3829 < 256 character. class_single_char will be 1 if the class contains only
3830 a single character. */
3831
3832 class_has_8bitchar = 0;
3833 class_single_char = 0;
3834
3835 /* Initialize the 32-char bit map to all zeros. We build the map in a
3836 temporary bit of memory, in case the class contains only 1 character (less
3837 than 256), because in that case the compiled code doesn't use the bit map.
3838 */
3839
3840 memset(classbits, 0, 32 * sizeof(pcre_uint8));
3841
3842 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
3843 xclass = FALSE; /* No chars >= 256 */
3844 class_uchardata = code + LINK_SIZE + 2; /* For UTF-8 items */
3845 class_uchardata_base = class_uchardata; /* For resetting in pass 1 */
3846 #endif
3847
3848 /* Process characters until ] is reached. By writing this as a "do" it
3849 means that an initial ] is taken as a data character. At the start of the
3850 loop, c contains the first byte of the character. */
3851
3852 if (c != 0) do
3853 {
3854 const pcre_uchar *oldptr;
3855
3856 #ifdef SUPPORT_UTF
3857 if (utf && HAS_EXTRALEN(c))
3858 { /* Braces are required because the */
3859 GETCHARLEN(c, ptr, ptr); /* macro generates multiple statements */
3860 }
3861 #endif
3862
3863 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
3864 /* In the pre-compile phase, accumulate the length of any extra
3865 data and reset the pointer. This is so that very large classes that
3866 contain a zillion > 255 characters no longer overwrite the work space
3867 (which is on the stack). */
3868
3869 if (lengthptr != NULL)
3870 {
3871 *lengthptr += class_uchardata - class_uchardata_base;
3872 class_uchardata = class_uchardata_base;
3873 }
3874 #endif
3875
3876 /* Inside \Q...\E everything is literal except \E */
3877
3878 if (inescq)
3879 {
3880 if (c == CHAR_BACKSLASH && ptr[1] == CHAR_E) /* If we are at \E */
3881 {
3882 inescq = FALSE; /* Reset literal state */
3883 ptr++; /* Skip the 'E' */
3884 continue; /* Carry on with next */
3885 }
3886 goto CHECK_RANGE; /* Could be range if \E follows */
3887 }
3888
3889 /* Handle POSIX class names. Perl allows a negation extension of the
3890 form [:^name:]. A square bracket that doesn't match the syntax is
3891 treated as a literal. We also recognize the POSIX constructions
3892 [.ch.] and [=ch=] ("collating elements") and fault them, as Perl
3893 5.6 and 5.8 do. */
3894
3895 if (c == CHAR_LEFT_SQUARE_BRACKET &&
3896 (ptr[1] == CHAR_COLON || ptr[1] == CHAR_DOT ||
3897 ptr[1] == CHAR_EQUALS_SIGN) && check_posix_syntax(ptr, &tempptr))
3898 {
3899 BOOL local_negate = FALSE;
3900 int posix_class, taboffset, tabopt;
3901 register const pcre_uint8 *cbits = cd->cbits;
3902 pcre_uint8 pbits[32];
3903
3904 if (ptr[1] != CHAR_COLON)
3905 {
3906 *errorcodeptr = ERR31;
3907 goto FAILED;
3908 }
3909
3910 ptr += 2;
3911 if (*ptr == CHAR_CIRCUMFLEX_ACCENT)
3912 {
3913 local_negate = TRUE;
3914 should_flip_negation = TRUE; /* Note negative special */
3915 ptr++;
3916 }
3917
3918 posix_class = check_posix_name(ptr, (int)(tempptr - ptr));
3919 if (posix_class < 0)
3920 {
3921 *errorcodeptr = ERR30;
3922 goto FAILED;
3923 }
3924
3925 /* If matching is caseless, upper and lower are converted to
3926 alpha. This relies on the fact that the class table starts with
3927 alpha, lower, upper as the first 3 entries. */
3928
3929 if ((options & PCRE_CASELESS) != 0 && posix_class <= 2)
3930 posix_class = 0;
3931
3932 /* When PCRE_UCP is set, some of the POSIX classes are converted to
3933 different escape sequences that use Unicode properties. */
3934
3935 #ifdef SUPPORT_UCP
3936 if ((options & PCRE_UCP) != 0)
3937 {
3938 int pc = posix_class + ((local_negate)? POSIX_SUBSIZE/2 : 0);
3939 if (posix_substitutes[pc] != NULL)
3940 {
3941 nestptr = tempptr + 1;
3942 ptr = posix_substitutes[pc] - 1;
3943 continue;
3944 }
3945 }
3946 #endif
3947 /* In the non-UCP case, we build the bit map for the POSIX class in a
3948 chunk of local store because we may be adding and subtracting from it,
3949 and we don't want to subtract bits that may be in the main map already.
3950 At the end we or the result into the bit map that is being built. */
3951
3952 posix_class *= 3;
3953
3954 /* Copy in the first table (always present) */
3955
3956 memcpy(pbits, cbits + posix_class_maps[posix_class],
3957 32 * sizeof(pcre_uint8));
3958
3959 /* If there is a second table, add or remove it as required. */
3960
3961 taboffset = posix_class_maps[posix_class + 1];
3962 tabopt = posix_class_maps[posix_class + 2];
3963
3964 if (taboffset >= 0)
3965 {
3966 if (tabopt >= 0)
3967 for (c = 0; c < 32; c++) pbits[c] |= cbits[c + taboffset];
3968 else
3969 for (c = 0; c < 32; c++) pbits[c] &= ~cbits[c + taboffset];
3970 }
3971
3972 /* Not see if we need to remove any special characters. An option
3973 value of 1 removes vertical space and 2 removes underscore. */
3974
3975 if (tabopt < 0) tabopt = -tabopt;
3976 if (tabopt == 1) pbits[1] &= ~0x3c;
3977 else if (tabopt == 2) pbits[11] &= 0x7f;
3978
3979 /* Add the POSIX table or its complement into the main table that is
3980 being built and we are done. */
3981
3982 if (local_negate)
3983 for (c = 0; c < 32; c++) classbits[c] |= ~pbits[c];
3984 else
3985 for (c = 0; c < 32; c++) classbits[c] |= pbits[c];
3986
3987 ptr = tempptr + 1;
3988 /* Every class contains at least one < 256 characters. */
3989 class_has_8bitchar = 1;
3990 /* Every class contains at least two characters. */
3991 class_single_char = 2;
3992 continue; /* End of POSIX syntax handling */
3993 }
3994
3995 /* Backslash may introduce a single character, or it may introduce one
3996 of the specials, which just set a flag. The sequence \b is a special
3997 case. Inside a class (and only there) it is treated as backspace. We
3998 assume that other escapes have more than one character in them, so
3999 speculatively set both class_has_8bitchar and class_single_char bigger
4000 than one. Unrecognized escapes fall through and are either treated
4001 as literal characters (by default), or are faulted if
4002 PCRE_EXTRA is set. */
4003
4004 if (c == CHAR_BACKSLASH)
4005 {
4006 c = check_escape(&ptr, errorcodeptr, cd->bracount, options, TRUE);
4007 if (*errorcodeptr != 0) goto FAILED;
4008
4009 if (-c == ESC_b) c = CHAR_BS; /* \b is backspace in a class */
4010 else if (-c == ESC_N) /* \N is not supported in a class */
4011 {
4012 *errorcodeptr = ERR71;
4013 goto FAILED;
4014 }
4015 else if (-c == ESC_Q) /* Handle start of quoted string */
4016 {
4017 if (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
4018 {
4019 ptr += 2; /* avoid empty string */
4020 }
4021 else inescq = TRUE;
4022 continue;
4023 }
4024 else if (-c == ESC_E) continue; /* Ignore orphan \E */
4025
4026 if (c < 0)
4027 {
4028 register const pcre_uint8 *cbits = cd->cbits;
4029 /* Every class contains at least two < 256 characters. */
4030 class_has_8bitchar++;
4031 /* Every class contains at least two characters. */
4032 class_single_char += 2;
4033
4034 switch (-c)
4035 {
4036 #ifdef SUPPORT_UCP
4037 case ESC_du: /* These are the values given for \d etc */
4038 case ESC_DU: /* when PCRE_UCP is set. We replace the */
4039 case ESC_wu: /* escape sequence with an appropriate \p */
4040 case ESC_WU: /* or \P to test Unicode properties instead */
4041 case ESC_su: /* of the default ASCII testing. */
4042 case ESC_SU:
4043 nestptr = ptr;
4044 ptr = substitutes[-c - ESC_DU] - 1; /* Just before substitute */
4045 class_has_8bitchar--; /* Undo! */
4046 continue;
4047 #endif
4048 case ESC_d:
4049 for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_digit];
4050 continue;
4051
4052 case ESC_D:
4053 should_flip_negation = TRUE;
4054 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_digit];
4055 continue;
4056
4057 case ESC_w:
4058 for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_word];
4059 continue;
4060
4061 case ESC_W:
4062 should_flip_negation = TRUE;
4063 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_word];
4064 continue;
4065
4066 /* Perl 5.004 onwards omits VT from \s, but we must preserve it
4067 if it was previously set by something earlier in the character
4068 class. Luckily, the value of CHAR_VT is 0x0b in both ASCII and
4069 EBCDIC, so we lazily just adjust the appropriate bit. */
4070
4071 case ESC_s:
4072 classbits[0] |= cbits[cbit_space];
4073 classbits[1] |= cbits[cbit_space+1] & ~0x08;
4074 for (c = 2; c < 32; c++) classbits[c] |= cbits[c+cbit_space];
4075 continue;
4076
4077 case ESC_S:
4078 should_flip_negation = TRUE;
4079 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_space];
4080 classbits[1] |= 0x08; /* Perl 5.004 onwards omits VT from \s */
4081 continue;
4082
4083 case ESC_h:
4084 SETBIT(classbits, CHAR_HT);
4085 SETBIT(classbits, CHAR_SPACE);
4086 #ifndef EBCDIC
4087 SETBIT(classbits, 0xa0); /* NSBP */
4088 #ifndef COMPILE_PCRE8
4089 xclass = TRUE;
4090 *class_uchardata++ = XCL_SINGLE;
4091 *class_uchardata++ = 0x1680;
4092 *class_uchardata++ = XCL_SINGLE;
4093 *class_uchardata++ = 0x180e;
4094 *class_uchardata++ = XCL_RANGE;
4095 *class_uchardata++ = 0x2000;
4096 *class_uchardata++ = 0x200a;
4097 *class_uchardata++ = XCL_SINGLE;
4098 *class_uchardata++ = 0x202f;
4099 *class_uchardata++ = XCL_SINGLE;
4100 *class_uchardata++ = 0x205f;
4101 *class_uchardata++ = XCL_SINGLE;
4102 *class_uchardata++ = 0x3000;
4103 #elif defined SUPPORT_UTF
4104 if (utf)
4105 {
4106 xclass = TRUE;
4107 *class_uchardata++ = XCL_SINGLE;
4108 class_uchardata += PRIV(ord2utf)(0x1680, class_uchardata);
4109 *class_uchardata++ = XCL_SINGLE;
4110 class_uchardata += PRIV(ord2utf)(0x180e, class_uchardata);
4111 *class_uchardata++ = XCL_RANGE;
4112 class_uchardata += PRIV(ord2utf)(0x2000, class_uchardata);
4113 class_uchardata += PRIV(ord2utf)(0x200a, class_uchardata);
4114 *class_uchardata++ = XCL_SINGLE;
4115 class_uchardata += PRIV(ord2utf)(0x202f, class_uchardata);
4116 *class_uchardata++ = XCL_SINGLE;
4117 class_uchardata += PRIV(ord2utf)(0x205f, class_uchardata);
4118 *class_uchardata++ = XCL_SINGLE;
4119 class_uchardata += PRIV(ord2utf)(0x3000, class_uchardata);
4120 }
4121 #endif
4122 #endif /* Not EBCDIC */
4123 continue;
4124
4125 case ESC_H:
4126 for (c = 0; c < 32; c++)
4127 {
4128 int x = 0xff;
4129 switch (c)
4130 {
4131 case CHAR_HT/8: x ^= 1 << (CHAR_HT%8); break;
4132 case CHAR_SPACE/8: x ^= 1 << (CHAR_SPACE%8); break;
4133 #ifndef EBCDIC
4134 case 0xa0/8: x ^= 1 << (0xa0%8); break; /* NSBSP */
4135 #endif
4136 default: break;
4137 }
4138 classbits[c] |= x;
4139 }
4140 #ifndef EBCDIC
4141 #ifndef COMPILE_PCRE8
4142 xclass = TRUE;
4143 *class_uchardata++ = XCL_RANGE;
4144 *class_uchardata++ = 0x0100;
4145 *class_uchardata++ = 0x167f;
4146 *class_uchardata++ = XCL_RANGE;
4147 *class_uchardata++ = 0x1681;
4148 *class_uchardata++ = 0x180d;
4149 *class_uchardata++ = XCL_RANGE;
4150 *class_uchardata++ = 0x180f;
4151 *class_uchardata++ = 0x1fff;
4152 *class_uchardata++ = XCL_RANGE;
4153 *class_uchardata++ = 0x200b;
4154 *class_uchardata++ = 0x202e;
4155 *class_uchardata++ = XCL_RANGE;
4156 *class_uchardata++ = 0x2030;
4157 *class_uchardata++ = 0x205e;
4158 *class_uchardata++ = XCL_RANGE;
4159 *class_uchardata++ = 0x2060;
4160 *class_uchardata++ = 0x2fff;
4161 *class_uchardata++ = XCL_RANGE;
4162 *class_uchardata++ = 0x3001;
4163 #ifdef SUPPORT_UTF
4164 if (utf)
4165 class_uchardata += PRIV(ord2utf)(0x10ffff, class_uchardata);
4166 else
4167 #endif /* SUPPORT_UTF */
4168 *class_uchardata++ = 0xffff;
4169 #elif defined SUPPORT_UTF
4170 if (utf)
4171 {
4172 xclass = TRUE;
4173 *class_uchardata++ = XCL_RANGE;
4174 class_uchardata += PRIV(ord2utf)(0x0100, class_uchardata);
4175 class_uchardata += PRIV(ord2utf)(0x167f, class_uchardata);
4176 *class_uchardata++ = XCL_RANGE;
4177 class_uchardata += PRIV(ord2utf)(0x1681, class_uchardata);
4178 class_uchardata += PRIV(ord2utf)(0x180d, class_uchardata);
4179 *class_uchardata++ = XCL_RANGE;
4180 class_uchardata += PRIV(ord2utf)(0x180f, class_uchardata);
4181 class_uchardata += PRIV(ord2utf)(0x1fff, class_uchardata);
4182 *class_uchardata++ = XCL_RANGE;
4183 class_uchardata += PRIV(ord2utf)(0x200b, class_uchardata);
4184 class_uchardata += PRIV(ord2utf)(0x202e, class_uchardata);
4185 *class_uchardata++ = XCL_RANGE;
4186 class_uchardata += PRIV(ord2utf)(0x2030, class_uchardata);
4187 class_uchardata += PRIV(ord2utf)(0x205e, class_uchardata);
4188 *class_uchardata++ = XCL_RANGE;
4189 class_uchardata += PRIV(ord2utf)(0x2060, class_uchardata);
4190 class_uchardata += PRIV(ord2utf)(0x2fff, class_uchardata);
4191 *class_uchardata++ = XCL_RANGE;
4192 class_uchardata += PRIV(ord2utf)(0x3001, class_uchardata);
4193 class_uchardata += PRIV(ord2utf)(0x10ffff, class_uchardata);
4194 }
4195 #endif
4196 #endif /* Not EBCDIC */
4197 continue;
4198
4199 case ESC_v:
4200 SETBIT(classbits, CHAR_LF);
4201 SETBIT(classbits, CHAR_VT);
4202 SETBIT(classbits, CHAR_FF);
4203 SETBIT(classbits, CHAR_CR);
4204 SETBIT(classbits, CHAR_NEL);
4205 #ifndef EBCDIC
4206 #ifndef COMPILE_PCRE8
4207 xclass = TRUE;
4208 *class_uchardata++ = XCL_RANGE;
4209 *class_uchardata++ = 0x2028;
4210 *class_uchardata++ = 0x2029;
4211 #elif defined SUPPORT_UTF
4212 if (utf)
4213 {
4214 xclass = TRUE;
4215 *class_uchardata++ = XCL_RANGE;
4216 class_uchardata += PRIV(ord2utf)(0x2028, class_uchardata);
4217 class_uchardata += PRIV(ord2utf)(0x2029, class_uchardata);
4218 }
4219 #endif
4220 #endif /* Not EBCDIC */
4221 continue;
4222
4223 case ESC_V:
4224 for (c = 0; c < 32; c++)
4225 {
4226 int x = 0xff;
4227 switch (c)
4228 {
4229 case CHAR_LF/8: x ^= 1 << (CHAR_LF%8);
4230 x ^= 1 << (CHAR_VT%8);
4231 x ^= 1 << (CHAR_FF%8);
4232 x ^= 1 << (CHAR_CR%8);
4233 break;
4234 case CHAR_NEL/8: x ^= 1 << (CHAR_NEL%8); break;
4235 default: break;
4236 }
4237 classbits[c] |= x;
4238 }
4239
4240 #ifndef EBCDIC
4241 #ifndef COMPILE_PCRE8
4242 xclass = TRUE;
4243 *class_uchardata++ = XCL_RANGE;
4244 *class_uchardata++ = 0x0100;
4245 *class_uchardata++ = 0x2027;
4246 *class_uchardata++ = XCL_RANGE;
4247 *class_uchardata++ = 0x202a;
4248 #ifdef SUPPORT_UTF
4249 if (utf)
4250 class_uchardata += PRIV(ord2utf)(0x10ffff, class_uchardata);
4251 else
4252 #endif
4253 *class_uchardata++ = 0xffff;
4254 #elif defined SUPPORT_UTF
4255 if (utf)
4256 {
4257 xclass = TRUE;
4258 *class_uchardata++ = XCL_RANGE;
4259 class_uchardata += PRIV(ord2utf)(0x0100, class_uchardata);
4260 class_uchardata += PRIV(ord2utf)(0x2027, class_uchardata);
4261 *class_uchardata++ = XCL_RANGE;
4262 class_uchardata += PRIV(ord2utf)(0x202a, class_uchardata);
4263 class_uchardata += PRIV(ord2utf)(0x10ffff, class_uchardata);
4264 }
4265 #endif
4266 #endif /* Not EBCDIC */
4267 continue;
4268
4269 #ifdef SUPPORT_UCP
4270 case ESC_p:
4271 case ESC_P:
4272 {
4273 BOOL negated;
4274 int pdata;
4275 int ptype = get_ucp(&ptr, &negated, &pdata, errorcodeptr);
4276 if (ptype < 0) goto FAILED;
4277 xclass = TRUE;
4278 *class_uchardata++ = ((-c == ESC_p) != negated)?
4279 XCL_PROP : XCL_NOTPROP;
4280 *class_uchardata++ = ptype;
4281 *class_uchardata++ = pdata;
4282 class_has_8bitchar--; /* Undo! */
4283 continue;
4284 }
4285 #endif
4286 /* Unrecognized escapes are faulted if PCRE is running in its
4287 strict mode. By default, for compatibility with Perl, they are
4288 treated as literals. */
4289
4290 default:
4291 if ((options & PCRE_EXTRA) != 0)
4292 {
4293 *errorcodeptr = ERR7;
4294 goto FAILED;
4295 }
4296 class_has_8bitchar--; /* Undo the speculative increase. */
4297 class_single_char -= 2; /* Undo the speculative increase. */
4298 c = *ptr; /* Get the final character and fall through */
4299 break;
4300 }
4301 }
4302
4303 /* Fall through if we have a single character (c >= 0). This may be
4304 greater than 256. */
4305
4306 } /* End of backslash handling */
4307
4308 /* A single character may be followed by '-' to form a range. However,
4309 Perl does not permit ']' to be the end of the range. A '-' character
4310 at the end is treated as a literal. Perl ignores orphaned \E sequences
4311 entirely. The code for handling \Q and \E is messy. */
4312
4313 CHECK_RANGE:
4314 while (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
4315 {
4316 inescq = FALSE;
4317 ptr += 2;
4318 }
4319
4320 oldptr = ptr;
4321
4322 /* Remember \r or \n */
4323
4324 if (c == CHAR_CR || c == CHAR_NL) cd->external_flags |= PCRE_HASCRORLF;
4325
4326 /* Check for range */
4327
4328 if (!inescq && ptr[1] == CHAR_MINUS)
4329 {
4330 int d;
4331 ptr += 2;
4332 while (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_E) ptr += 2;
4333
4334 /* If we hit \Q (not followed by \E) at this point, go into escaped
4335 mode. */
4336
4337 while (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_Q)
4338 {
4339 ptr += 2;
4340 if (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_E)
4341 { ptr += 2; continue; }
4342 inescq = TRUE;
4343 break;
4344 }
4345
4346 if (*ptr == 0 || (!inescq && *ptr == CHAR_RIGHT_SQUARE_BRACKET))
4347 {
4348 ptr = oldptr;
4349 goto LONE_SINGLE_CHARACTER;
4350 }
4351
4352 #ifdef SUPPORT_UTF
4353 if (utf)
4354 { /* Braces are required because the */
4355 GETCHARLEN(d, ptr, ptr); /* macro generates multiple statements */
4356 }
4357 else
4358 #endif
4359 d = *ptr; /* Not UTF-8 mode */
4360
4361 /* The second part of a range can be a single-character escape, but
4362 not any of the other escapes. Perl 5.6 treats a hyphen as a literal
4363 in such circumstances. */
4364
4365 if (!inescq && d == CHAR_BACKSLASH)
4366 {
4367 d = check_escape(&ptr, errorcodeptr, cd->bracount, options, TRUE);
4368 if (*errorcodeptr != 0) goto FAILED;
4369
4370 /* \b is backspace; any other special means the '-' was literal */
4371
4372 if (d < 0)
4373 {
4374 if (d == -ESC_b) d = CHAR_BS; else
4375 {
4376 ptr = oldptr;
4377 goto LONE_SINGLE_CHARACTER; /* A few lines below */
4378 }
4379 }
4380 }
4381
4382 /* Check that the two values are in the correct order. Optimize
4383 one-character ranges */
4384
4385 if (d < c)
4386 {
4387 *errorcodeptr = ERR8;
4388 goto FAILED;
4389 }
4390
4391 if (d == c) goto LONE_SINGLE_CHARACTER; /* A few lines below */
4392
4393 /* Remember \r or \n */
4394
4395 if (d == CHAR_CR || d == CHAR_NL) cd->external_flags |= PCRE_HASCRORLF;
4396
4397 /* Since we found a character range, single character optimizations
4398 cannot be done anymore. */
4399 class_single_char = 2;
4400
4401 /* In UTF-8 mode, if the upper limit is > 255, or > 127 for caseless
4402 matching, we have to use an XCLASS with extra data items. Caseless
4403 matching for characters > 127 is available only if UCP support is
4404 available. */
4405
4406 #if defined SUPPORT_UTF && !(defined COMPILE_PCRE8)
4407 if ((d > 255) || (utf && ((options & PCRE_CASELESS) != 0 && d > 127)))
4408 #elif defined SUPPORT_UTF
4409 if (utf && (d > 255 || ((options & PCRE_CASELESS) != 0 && d > 127)))
4410 #elif !(defined COMPILE_PCRE8)
4411 if (d > 255)
4412 #endif
4413 #if defined SUPPORT_UTF || !(defined COMPILE_PCRE8)
4414 {
4415 xclass = TRUE;
4416
4417 /* With UCP support, we can find the other case equivalents of
4418 the relevant characters. There may be several ranges. Optimize how
4419 they fit with the basic range. */
4420
4421 #ifdef SUPPORT_UCP
4422 #ifndef COMPILE_PCRE8
4423 if (utf && (options & PCRE_CASELESS) != 0)
4424 #else
4425 if ((options & PCRE_CASELESS) != 0)
4426 #endif
4427 {
4428 unsigned int occ, ocd;
4429 unsigned int cc = c;
4430 unsigned int origd = d;
4431 while (get_othercase_range(&cc, origd, &occ, &ocd))
4432 {
4433 if (occ >= (unsigned int)c &&
4434 ocd <= (unsigned int)d)
4435 continue; /* Skip embedded ranges */
4436
4437 if (occ < (unsigned int)c &&
4438 ocd >= (unsigned int)c - 1) /* Extend the basic range */
4439 { /* if there is overlap, */
4440 c = occ; /* noting that if occ < c */
4441 continue; /* we can't have ocd > d */
4442 } /* because a subrange is */
4443 if (ocd > (unsigned int)d &&
4444 occ <= (unsigned int)d + 1) /* always shorter than */
4445 { /* the basic range. */
4446 d = ocd;
4447 continue;
4448 }
4449
4450 if (occ == ocd)
4451 {
4452 *class_uchardata++ = XCL_SINGLE;
4453 }
4454 else
4455 {
4456 *class_uchardata++ = XCL_RANGE;
4457 class_uchardata += PRIV(ord2utf)(occ, class_uchardata);
4458 }
4459 class_uchardata += PRIV(ord2utf)(ocd, class_uchardata);
4460 }
4461 }
4462 #endif /* SUPPORT_UCP */
4463
4464 /* Now record the original range, possibly modified for UCP caseless
4465 overlapping ranges. */
4466
4467 *class_uchardata++ = XCL_RANGE;
4468 #ifdef SUPPORT_UTF
4469 #ifndef COMPILE_PCRE8
4470 if (utf)
4471 {
4472 class_uchardata += PRIV(ord2utf)(c, class_uchardata);
4473 class_uchardata += PRIV(ord2utf)(d, class_uchardata);
4474 }
4475 else
4476 {
4477 *class_uchardata++ = c;
4478 *class_uchardata++ = d;
4479 }
4480 #else
4481 class_uchardata += PRIV(ord2utf)(c, class_uchardata);
4482 class_uchardata += PRIV(ord2utf)(d, class_uchardata);
4483 #endif
4484 #else /* SUPPORT_UTF */
4485 *class_uchardata++ = c;
4486 *class_uchardata++ = d;
4487 #endif /* SUPPORT_UTF */
4488
4489 /* With UCP support, we are done. Without UCP support, there is no
4490 caseless matching for UTF characters > 127; we can use the bit map
4491 for the smaller ones. As for 16 bit characters without UTF, we
4492 can still use */
4493
4494 #ifdef SUPPORT_UCP
4495 #ifndef COMPILE_PCRE8
4496 if (utf)
4497 #endif
4498 continue; /* With next character in the class */
4499 #endif /* SUPPORT_UCP */
4500
4501 #if defined SUPPORT_UTF && !defined(SUPPORT_UCP) && !(defined COMPILE_PCRE8)
4502 if (utf)
4503 {
4504 if ((options & PCRE_CASELESS) == 0 || c > 127) continue;
4505 /* Adjust upper limit and fall through to set up the map */
4506 d = 127;
4507 }
4508 else
4509 {
4510 if (c > 255) continue;
4511 /* Adjust upper limit and fall through to set up the map */
4512 d = 255;
4513 }
4514 #elif defined SUPPORT_UTF && !defined(SUPPORT_UCP)
4515 if ((options & PCRE_CASELESS) == 0 || c > 127) continue;
4516 /* Adjust upper limit and fall through to set up the map */
4517 d = 127;
4518 #else
4519 if (c > 255) continue;
4520 /* Adjust upper limit and fall through to set up the map */
4521 d = 255;
4522 #endif /* SUPPORT_UTF && !SUPPORT_UCP && !COMPILE_PCRE8 */
4523 }
4524 #endif /* SUPPORT_UTF || !COMPILE_PCRE8 */
4525
4526 /* We use the bit map for 8 bit mode, or when the characters fall
4527 partially or entirely to [0-255] ([0-127] for UCP) ranges. */
4528
4529 class_has_8bitchar = 1;
4530
4531 /* We can save a bit of time by skipping this in the pre-compile. */
4532
4533 if (lengthptr == NULL) for (; c <= d; c++)
4534 {
4535 classbits[c/8] |= (1 << (c&7));
4536 if ((options & PCRE_CASELESS) != 0)
4537 {
4538 int uc = cd->fcc[c]; /* flip case */
4539 classbits[uc/8] |= (1 << (uc&7));
4540 }
4541 }
4542
4543 continue; /* Go get the next char in the class */
4544 }
4545
4546 /* Handle a lone single character - we can get here for a normal
4547 non-escape char, or after \ that introduces a single character or for an
4548 apparent range that isn't. */
4549
4550 LONE_SINGLE_CHARACTER:
4551
4552 /* Only the value of 1 matters for class_single_char. */
4553
4554 if (class_single_char < 2) class_single_char++;
4555
4556 /* If class_charcount is 1, we saw precisely one character. As long as
4557 there was no use of \p or \P, in other words, no use of any XCLASS
4558 features, we can optimize.
4559
4560 The optimization throws away the bit map. We turn the item into a
4561 1-character OP_CHAR[I] if it's positive, or OP_NOT[I] if it's negative.
4562 In the positive case, it can cause firstchar to be set. Otherwise, there
4563 can be no first char if this item is first, whatever repeat count may
4564 follow. In the case of reqchar, save the previous value for reinstating. */
4565
4566 if (class_single_char == 1 && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET)
4567 {
4568 ptr++;
4569 zeroreqchar = reqchar;
4570
4571 if (negate_class)
4572 {
4573 if (firstchar == REQ_UNSET) firstchar = REQ_NONE;
4574 zerofirstchar = firstchar;
4575 *code++ = ((options & PCRE_CASELESS) != 0)? OP_NOTI: OP_NOT;
4576 #ifdef SUPPORT_UTF
4577 if (utf && c > MAX_VALUE_FOR_SINGLE_CHAR)
4578 code += PRIV(ord2utf)(c, code);
4579 else
4580 #endif
4581 *code++ = c;
4582 goto NOT_CHAR;
4583 }
4584
4585 /* For a single, positive character, get the value into mcbuffer, and
4586 then we can handle this with the normal one-character code. */
4587
4588 #ifdef SUPPORT_UTF
4589 if (utf && c > MAX_VALUE_FOR_SINGLE_CHAR)
4590 mclength = PRIV(ord2utf)(c, mcbuffer);
4591 else
4592 #endif
4593 {
4594 mcbuffer[0] = c;
4595 mclength = 1;
4596 }
4597 goto ONE_CHAR;
4598 } /* End of 1-char optimization */
4599
4600 /* Handle a character that cannot go in the bit map. */
4601
4602 #if defined SUPPORT_UTF && !(defined COMPILE_PCRE8)
4603 if ((c > 255) || (utf && ((options & PCRE_CASELESS) != 0 && c > 127)))
4604 #elif defined SUPPORT_UTF
4605 if (utf && (c > 255 || ((options & PCRE_CASELESS) != 0 && c > 127)))
4606 #elif !(defined COMPILE_PCRE8)
4607 if (c > 255)
4608 #endif
4609
4610 #if defined SUPPORT_UTF || !(defined COMPILE_PCRE8)
4611 {
4612 xclass = TRUE;
4613 *class_uchardata++ = XCL_SINGLE;
4614 #ifdef SUPPORT_UTF
4615 #ifndef COMPILE_PCRE8
4616 /* In non 8 bit mode, we can get here even if we are not in UTF mode. */
4617 if (!utf)
4618 *class_uchardata++ = c;
4619 else
4620 #endif
4621 class_uchardata += PRIV(ord2utf)(c, class_uchardata);
4622 #else /* SUPPORT_UTF */
4623 *class_uchardata++ = c;
4624 #endif /* SUPPORT_UTF */
4625
4626 #ifdef SUPPORT_UCP
4627 #ifdef COMPILE_PCRE8
4628 if ((options & PCRE_CASELESS) != 0)
4629 #else
4630 /* In non 8 bit mode, we can get here even if we are not in UTF mode. */
4631 if (utf && (options & PCRE_CASELESS) != 0)
4632 #endif
4633 {
4634 unsigned int othercase;
4635 if ((int)(othercase = UCD_OTHERCASE(c)) != c)
4636 {
4637 *class_uchardata++ = XCL_SINGLE;
4638 class_uchardata += PRIV(ord2utf)(othercase, class_uchardata);
4639 }
4640 }
4641 #endif /* SUPPORT_UCP */
4642
4643 }
4644 else
4645 #endif /* SUPPORT_UTF || COMPILE_PCRE16 */
4646
4647 /* Handle a single-byte character */
4648 {
4649 class_has_8bitchar = 1;
4650 classbits[c/8] |= (1 << (c&7));
4651 if ((options & PCRE_CASELESS) != 0)
4652 {
4653 c = cd->fcc[c]; /* flip case */
4654 classbits[c/8] |= (1 << (c&7));
4655 }
4656 }
4657 }
4658
4659 /* Loop until ']' reached. This "while" is the end of the "do" far above.
4660 If we are at the end of an internal nested string, revert to the outer
4661 string. */
4662
4663 while (((c = *(++ptr)) != 0 ||
4664 (nestptr != NULL &&
4665 (ptr = nestptr, nestptr = NULL, c = *(++ptr)) != 0)) &&
4666 (c != CHAR_RIGHT_SQUARE_BRACKET || inescq));
4667
4668 /* Check for missing terminating ']' */
4669
4670 if (c == 0)
4671 {
4672 *errorcodeptr = ERR6;
4673 goto FAILED;
4674 }
4675
4676 /* If this is the first thing in the branch, there can be no first char
4677 setting, whatever the repeat count. Any reqchar setting must remain
4678 unchanged after any kind of repeat. */
4679
4680 if (firstchar == REQ_UNSET) firstchar = REQ_NONE;
4681 zerofirstchar = firstchar;
4682 zeroreqchar = reqchar;
4683
4684 /* If there are characters with values > 255, we have to compile an
4685 extended class, with its own opcode, unless there was a negated special
4686 such as \S in the class, and PCRE_UCP is not set, because in that case all
4687 characters > 255 are in the class, so any that were explicitly given as
4688 well can be ignored. If (when there are explicit characters > 255 that must
4689 be listed) there are no characters < 256, we can omit the bitmap in the
4690 actual compiled code. */
4691
4692 #ifdef SUPPORT_UTF
4693 if (xclass && (!should_flip_negation || (options & PCRE_UCP) != 0))
4694 #elif !defined COMPILE_PCRE8
4695 if (xclass && !should_flip_negation)
4696 #endif
4697 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
4698 {
4699 *class_uchardata++ = XCL_END; /* Marks the end of extra data */
4700 *code++ = OP_XCLASS;
4701 code += LINK_SIZE;
4702 *code = negate_class? XCL_NOT:0;
4703
4704 /* If the map is required, move up the extra data to make room for it;
4705 otherwise just move the code pointer to the end of the extra data. */
4706
4707 if (class_has_8bitchar > 0)
4708 {
4709 *code++ |= XCL_MAP;
4710 memmove(code + (32 / sizeof(pcre_uchar)), code,
4711 IN_UCHARS(class_uchardata - code));
4712 memcpy(code, classbits, 32);
4713 code = class_uchardata + (32 / sizeof(pcre_uchar));
4714 }
4715 else code = class_uchardata;
4716
4717 /* Now fill in the complete length of the item */
4718
4719 PUT(previous, 1, (int)(code - previous));
4720 break; /* End of class handling */
4721 }
4722 #endif
4723
4724 /* If there are no characters > 255, or they are all to be included or
4725 excluded, set the opcode to OP_CLASS or OP_NCLASS, depending on whether the
4726 whole class was negated and whether there were negative specials such as \S
4727 (non-UCP) in the class. Then copy the 32-byte map into the code vector,
4728 negating it if necessary. */
4729
4730 *code++ = (negate_class == should_flip_negation) ? OP_CLASS : OP_NCLASS;
4731 if (lengthptr == NULL) /* Save time in the pre-compile phase */
4732 {
4733 if (negate_class)
4734 for (c = 0; c < 32; c++) classbits[c] = ~classbits[c];
4735 memcpy(code, classbits, 32);
4736 }
4737 code += 32 / sizeof(pcre_uchar);
4738 NOT_CHAR:
4739 break;
4740
4741
4742 /* ===================================================================*/
4743 /* Various kinds of repeat; '{' is not necessarily a quantifier, but this
4744 has been tested above. */
4745
4746 case CHAR_LEFT_CURLY_BRACKET:
4747 if (!is_quantifier) goto NORMAL_CHAR;
4748 ptr = read_repeat_counts(ptr+1, &repeat_min, &repeat_max, errorcodeptr);
4749 if (*errorcodeptr != 0) goto FAILED;
4750 goto REPEAT;
4751
4752 case CHAR_ASTERISK:
4753 repeat_min = 0;
4754 repeat_max = -1;
4755 goto REPEAT;
4756
4757 case CHAR_PLUS:
4758 repeat_min = 1;
4759 repeat_max = -1;
4760 goto REPEAT;
4761
4762 case CHAR_QUESTION_MARK:
4763 repeat_min = 0;
4764 repeat_max = 1;
4765
4766 REPEAT:
4767 if (previous == NULL)
4768 {
4769 *errorcodeptr = ERR9;
4770 goto FAILED;
4771 }
4772
4773 if (repeat_min == 0)
4774 {
4775 firstchar = zerofirstchar; /* Adjust for zero repeat */
4776 reqchar = zeroreqchar; /* Ditto */
4777 }
4778
4779 /* Remember whether this is a variable length repeat */
4780
4781 reqvary = (repeat_min == repeat_max)? 0 : REQ_VARY;
4782
4783 op_type = 0; /* Default single-char op codes */
4784 possessive_quantifier = FALSE; /* Default not possessive quantifier */
4785
4786 /* Save start of previous item, in case we have to move it up in order to
4787 insert something before it. */
4788
4789 tempcode = previous;
4790
4791 /* If the next character is '+', we have a possessive quantifier. This
4792 implies greediness, whatever the setting of the PCRE_UNGREEDY option.
4793 If the next character is '?' this is a minimizing repeat, by default,
4794 but if PCRE_UNGREEDY is set, it works the other way round. We change the
4795 repeat type to the non-default. */
4796
4797 if (ptr[1] == CHAR_PLUS)
4798 {
4799 repeat_type = 0; /* Force greedy */
4800 possessive_quantifier = TRUE;
4801 ptr++;
4802 }
4803 else if (ptr[1] == CHAR_QUESTION_MARK)
4804 {
4805 repeat_type = greedy_non_default;
4806 ptr++;
4807 }
4808 else repeat_type = greedy_default;
4809
4810 /* If previous was a recursion call, wrap it in atomic brackets so that
4811 previous becomes the atomic group. All recursions were so wrapped in the
4812 past, but it no longer happens for non-repeated recursions. In fact, the
4813 repeated ones could be re-implemented independently so as not to need this,
4814 but for the moment we rely on the code for repeating groups. */
4815
4816 if (*previous == OP_RECURSE)
4817 {
4818 memmove(previous + 1 + LINK_SIZE, previous, IN_UCHARS(1 + LINK_SIZE));
4819 *previous = OP_ONCE;
4820 PUT(previous, 1, 2 + 2*LINK_SIZE);
4821 previous[2 + 2*LINK_SIZE] = OP_KET;
4822 PUT(previous, 3 + 2*LINK_SIZE, 2 + 2*LINK_SIZE);
4823 code += 2 + 2 * LINK_SIZE;
4824 length_prevgroup = 3 + 3*LINK_SIZE;
4825
4826 /* When actually compiling, we need to check whether this was a forward
4827 reference, and if so, adjust the offset. */
4828
4829 if (lengthptr == NULL && cd->hwm >= cd->start_workspace + LINK_SIZE)
4830 {
4831 int offset = GET(cd->hwm, -LINK_SIZE);
4832 if (offset == previous + 1 - cd->start_code)
4833 PUT(cd->hwm, -LINK_SIZE, offset + 1 + LINK_SIZE);
4834 }
4835 }
4836
4837 /* Now handle repetition for the different types of item. */
4838
4839 /* If previous was a character or negated character match, abolish the item
4840 and generate a repeat item instead. If a char item has a minimum of more
4841 than one, ensure that it is set in reqchar - it might not be if a sequence
4842 such as x{3} is the first thing in a branch because the x will have gone
4843 into firstchar instead. */
4844
4845 if (*previous == OP_CHAR || *previous == OP_CHARI
4846 || *previous == OP_NOT || *previous == OP_NOTI)
4847 {
4848 switch (*previous)
4849 {
4850 default: /* Make compiler happy. */
4851 case OP_CHAR: op_type = OP_STAR - OP_STAR; break;
4852 case OP_CHARI: op_type = OP_STARI - OP_STAR; break;
4853 case OP_NOT: op_type = OP_NOTSTAR - OP_STAR; break;
4854 case OP_NOTI: op_type = OP_NOTSTARI - OP_STAR; break;
4855 }
4856
4857 /* Deal with UTF characters that take up more than one character. It's
4858 easier to write this out separately than try to macrify it. Use c to
4859 hold the length of the character in bytes, plus UTF_LENGTH to flag that
4860 it's a length rather than a small character. */
4861
4862 #ifdef SUPPORT_UTF
4863 if (utf && NOT_FIRSTCHAR(code[-1]))
4864 {
4865 pcre_uchar *lastchar = code - 1;
4866 BACKCHAR(lastchar);
4867 c = (int)(code - lastchar); /* Length of UTF-8 character */
4868 memcpy(utf_chars, lastchar, IN_UCHARS(c)); /* Save the char */
4869 c |= UTF_LENGTH; /* Flag c as a length */
4870 }
4871 else
4872 #endif /* SUPPORT_UTF */
4873
4874 /* Handle the case of a single charater - either with no UTF support, or
4875 with UTF disabled, or for a single character UTF character. */
4876 {
4877 c = code[-1];
4878 if (*previous <= OP_CHARI && repeat_min > 1)
4879 reqchar = c | req_caseopt | cd->req_varyopt;
4880 }
4881
4882 /* If the repetition is unlimited, it pays to see if the next thing on
4883 the line is something that cannot possibly match this character. If so,
4884 automatically possessifying this item gains some performance in the case
4885 where the match fails. */
4886
4887 if (!possessive_quantifier &&
4888 repeat_max < 0 &&
4889 check_auto_possessive(previous, utf, ptr + 1, options, cd))
4890 {
4891 repeat_type = 0; /* Force greedy */
4892 possessive_quantifier = TRUE;
4893 }
4894
4895 goto OUTPUT_SINGLE_REPEAT; /* Code shared with single character types */
4896 }
4897
4898 /* If previous was a character type match (\d or similar), abolish it and
4899 create a suitable repeat item. The code is shared with single-character
4900 repeats by setting op_type to add a suitable offset into repeat_type. Note
4901 the the Unicode property types will be present only when SUPPORT_UCP is
4902 defined, but we don't wrap the little bits of code here because it just
4903 makes it horribly messy. */
4904
4905 else if (*previous < OP_EODN)
4906 {
4907 pcre_uchar *oldcode;
4908 int prop_type, prop_value;
4909 op_type = OP_TYPESTAR - OP_STAR; /* Use type opcodes */
4910 c = *previous;
4911
4912 if (!possessive_quantifier &&
4913 repeat_max < 0 &&
4914 check_auto_possessive(previous, utf, ptr + 1, options, cd))
4915 {
4916 repeat_type = 0; /* Force greedy */
4917 possessive_quantifier = TRUE;
4918 }
4919
4920 OUTPUT_SINGLE_REPEAT:
4921 if (*previous == OP_PROP || *previous == OP_NOTPROP)
4922 {
4923 prop_type = previous[1];
4924 prop_value = previous[2];
4925 }
4926 else prop_type = prop_value = -1;
4927
4928 oldcode = code;
4929 code = previous; /* Usually overwrite previous item */
4930
4931 /* If the maximum is zero then the minimum must also be zero; Perl allows
4932 this case, so we do too - by simply omitting the item altogether. */
4933
4934 if (repeat_max == 0) goto END_REPEAT;
4935
4936 /*--------------------------------------------------------------------*/
4937 /* This code is obsolete from release 8.00; the restriction was finally
4938 removed: */
4939
4940 /* All real repeats make it impossible to handle partial matching (maybe
4941 one day we will be able to remove this restriction). */
4942
4943 /* if (repeat_max != 1) cd->external_flags |= PCRE_NOPARTIAL; */
4944 /*--------------------------------------------------------------------*/
4945
4946 /* Combine the op_type with the repeat_type */
4947
4948 repeat_type += op_type;
4949
4950 /* A minimum of zero is handled either as the special case * or ?, or as
4951 an UPTO, with the maximum given. */
4952
4953 if (repeat_min == 0)
4954 {
4955 if (repeat_max == -1) *code++ = OP_STAR + repeat_type;
4956 else if (repeat_max == 1) *code++ = OP_QUERY + repeat_type;
4957 else
4958 {
4959 *code++ = OP_UPTO + repeat_type;
4960 PUT2INC(code, 0, repeat_max);
4961 }
4962 }
4963
4964 /* A repeat minimum of 1 is optimized into some special cases. If the
4965 maximum is unlimited, we use OP_PLUS. Otherwise, the original item is
4966 left in place and, if the maximum is greater than 1, we use OP_UPTO with
4967 one less than the maximum. */
4968
4969 else if (repeat_min == 1)
4970 {
4971 if (repeat_max == -1)
4972 *code++ = OP_PLUS + repeat_type;
4973 else
4974 {
4975 code = oldcode; /* leave previous item in place */
4976 if (repeat_max == 1) goto END_REPEAT;
4977 *code++ = OP_UPTO + repeat_type;
4978 PUT2INC(code, 0, repeat_max - 1);
4979 }
4980 }
4981
4982 /* The case {n,n} is just an EXACT, while the general case {n,m} is
4983 handled as an EXACT followed by an UPTO. */
4984
4985 else
4986 {
4987 *code++ = OP_EXACT + op_type; /* NB EXACT doesn't have repeat_type */
4988 PUT2INC(code, 0, repeat_min);
4989
4990 /* If the maximum is unlimited, insert an OP_STAR. Before doing so,
4991 we have to insert the character for the previous code. For a repeated
4992 Unicode property match, there are two extra bytes that define the
4993 required property. In UTF-8 mode, long characters have their length in
4994 c, with the UTF_LENGTH bit as a flag. */
4995
4996 if (repeat_max < 0)
4997 {
4998 #ifdef SUPPORT_UTF
4999 if (utf && (c & UTF_LENGTH) != 0)
5000 {
5001 memcpy(code, utf_chars, IN_UCHARS(c & 7));
5002 code += c & 7;
5003 }
5004 else
5005 #endif
5006 {
5007 *code++ = c;
5008 if (prop_type >= 0)
5009 {
5010 *code++ = prop_type;
5011 *code++ = prop_value;
5012 }
5013 }
5014 *code++ = OP_STAR + repeat_type;
5015 }
5016
5017 /* Else insert an UPTO if the max is greater than the min, again
5018 preceded by the character, for the previously inserted code. If the
5019 UPTO is just for 1 instance, we can use QUERY instead. */
5020
5021 else if (repeat_max != repeat_min)
5022 {
5023 #ifdef SUPPORT_UTF
5024 if (utf && (c & UTF_LENGTH) != 0)
5025 {
5026 memcpy(code, utf_chars, IN_UCHARS(c & 7));
5027 code += c & 7;
5028 }
5029 else
5030 #endif
5031 *code++ = c;
5032 if (prop_type >= 0)
5033 {
5034 *code++ = prop_type;
5035 *code++ = prop_value;
5036 }
5037 repeat_max -= repeat_min;
5038
5039 if (repeat_max == 1)
5040 {
5041 *code++ = OP_QUERY + repeat_type;
5042 }
5043 else
5044 {
5045 *code++ = OP_UPTO + repeat_type;
5046 PUT2INC(code, 0, repeat_max);
5047 }
5048 }
5049 }
5050
5051 /* The character or character type itself comes last in all cases. */
5052
5053 #ifdef SUPPORT_UTF
5054 if (utf && (c & UTF_LENGTH) != 0)
5055 {
5056 memcpy(code, utf_chars, IN_UCHARS(c & 7));
5057 code += c & 7;
5058 }
5059 else
5060 #endif
5061 *code++ = c;
5062
5063 /* For a repeated Unicode property match, there are two extra bytes that
5064 define the required property. */
5065
5066 #ifdef SUPPORT_UCP
5067 if (prop_type >= 0)
5068 {
5069 *code++ = prop_type;
5070 *code++ = prop_value;
5071 }
5072 #endif
5073 }
5074
5075 /* If previous was a character class or a back reference, we put the repeat
5076 stuff after it, but just skip the item if the repeat was {0,0}. */
5077
5078 else if (*previous == OP_CLASS ||
5079 *previous == OP_NCLASS ||
5080 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
5081 *previous == OP_XCLASS ||
5082 #endif
5083 *previous == OP_REF ||
5084 *previous == OP_REFI)
5085 {
5086 if (repeat_max == 0)
5087 {
5088 code = previous;
5089 goto END_REPEAT;
5090 }
5091
5092 /*--------------------------------------------------------------------*/
5093 /* This code is obsolete from release 8.00; the restriction was finally
5094 removed: */
5095
5096 /* All real repeats make it impossible to handle partial matching (maybe
5097 one day we will be able to remove this restriction). */
5098
5099 /* if (repeat_max != 1) cd->external_flags |= PCRE_NOPARTIAL; */
5100 /*--------------------------------------------------------------------*/
5101
5102 if (repeat_min == 0 && repeat_max == -1)
5103 *code++ = OP_CRSTAR + repeat_type;
5104 else if (repeat_min == 1 && repeat_max == -1)
5105 *code++ = OP_CRPLUS + repeat_type;
5106 else if (repeat_min == 0 && repeat_max == 1)
5107 *code++ = OP_CRQUERY + repeat_type;
5108 else
5109 {
5110 *code++ = OP_CRRANGE + repeat_type;
5111 PUT2INC(code, 0, repeat_min);
5112 if (repeat_max == -1) repeat_max = 0; /* 2-byte encoding for max */
5113 PUT2INC(code, 0, repeat_max);
5114 }
5115 }
5116
5117 /* If previous was a bracket group, we may have to replicate it in certain
5118 cases. Note that at this point we can encounter only the "basic" bracket
5119 opcodes such as BRA and CBRA, as this is the place where they get converted
5120 into the more special varieties such as BRAPOS and SBRA. A test for >=
5121 OP_ASSERT and <= OP_COND includes ASSERT, ASSERT_NOT, ASSERTBACK,
5122 ASSERTBACK_NOT, ONCE, BRA, CBRA, and COND. Originally, PCRE did not allow
5123 repetition of assertions, but now it does, for Perl compatibility. */
5124
5125 else if (*previous >= OP_ASSERT && *previous <= OP_COND)
5126 {
5127 register int i;
5128 int len = (int)(code - previous);
5129 pcre_uchar *bralink = NULL;
5130 pcre_uchar *brazeroptr = NULL;
5131
5132 /* Repeating a DEFINE group is pointless, but Perl allows the syntax, so
5133 we just ignore the repeat. */
5134
5135 if (*previous == OP_COND && previous[LINK_SIZE+1] == OP_DEF)
5136 goto END_REPEAT;
5137
5138 /* There is no sense in actually repeating assertions. The only potential
5139 use of repetition is in cases when the assertion is optional. Therefore,
5140 if the minimum is greater than zero, just ignore the repeat. If the
5141 maximum is not not zero or one, set it to 1. */
5142
5143 if (*previous < OP_ONCE) /* Assertion */
5144 {
5145 if (repeat_min > 0) goto END_REPEAT;
5146 if (repeat_max < 0 || repeat_max > 1) repeat_max = 1;
5147 }
5148
5149 /* The case of a zero minimum is special because of the need to stick
5150 OP_BRAZERO in front of it, and because the group appears once in the
5151 data, whereas in other cases it appears the minimum number of times. For
5152 this reason, it is simplest to treat this case separately, as otherwise
5153 the code gets far too messy. There are several special subcases when the
5154 minimum is zero. */
5155
5156 if (repeat_min == 0)
5157 {
5158 /* If the maximum is also zero, we used to just omit the group from the
5159 output altogether, like this:
5160
5161 ** if (repeat_max == 0)
5162 ** {
5163 ** code = previous;
5164 ** goto END_REPEAT;
5165 ** }
5166
5167 However, that fails when a group or a subgroup within it is referenced
5168 as a subroutine from elsewhere in the pattern, so now we stick in
5169 OP_SKIPZERO in front of it so that it is skipped on execution. As we
5170 don't have a list of which groups are referenced, we cannot do this
5171 selectively.
5172
5173 If the maximum is 1 or unlimited, we just have to stick in the BRAZERO
5174 and do no more at this point. However, we do need to adjust any
5175 OP_RECURSE calls inside the group that refer to the group itself or any
5176 internal or forward referenced group, because the offset is from the
5177 start of the whole regex. Temporarily terminate the pattern while doing
5178 this. */
5179
5180 if (repeat_max <= 1) /* Covers 0, 1, and unlimited */
5181 {
5182 *code = OP_END;
5183 adjust_recurse(previous, 1, utf, cd, save_hwm);
5184 memmove(previous + 1, previous, IN_UCHARS(len));
5185 code++;
5186 if (repeat_max == 0)
5187 {
5188 *previous++ = OP_SKIPZERO;
5189 goto END_REPEAT;
5190 }
5191 brazeroptr = previous; /* Save for possessive optimizing */
5192 *previous++ = OP_BRAZERO + repeat_type;
5193 }
5194
5195 /* If the maximum is greater than 1 and limited, we have to replicate
5196 in a nested fashion, sticking OP_BRAZERO before each set of brackets.
5197 The first one has to be handled carefully because it's the original
5198 copy, which has to be moved up. The remainder can be handled by code
5199 that is common with the non-zero minimum case below. We have to
5200 adjust the value or repeat_max, since one less copy is required. Once
5201 again, we may have to adjust any OP_RECURSE calls inside the group. */
5202
5203 else
5204 {
5205 int offset;
5206 *code = OP_END;
5207 adjust_recurse(previous, 2 + LINK_SIZE, utf, cd, save_hwm);
5208 memmove(previous + 2 + LINK_SIZE, previous, IN_UCHARS(len));
5209 code += 2 + LINK_SIZE;
5210 *previous++ = OP_BRAZERO + repeat_type;
5211 *previous++ = OP_BRA;
5212
5213 /* We chain together the bracket offset fields that have to be
5214 filled in later when the ends of the brackets are reached. */
5215
5216 offset = (bralink == NULL)? 0 : (int)(previous - bralink);
5217 bralink = previous;
5218 PUTINC(previous, 0, offset);
5219 }
5220
5221 repeat_max--;
5222 }
5223
5224 /* If the minimum is greater than zero, replicate the group as many
5225 times as necessary, and adjust the maximum to the number of subsequent
5226 copies that we need. If we set a first char from the group, and didn't
5227 set a required char, copy the latter from the former. If there are any
5228 forward reference subroutine calls in the group, there will be entries on
5229 the workspace list; replicate these with an appropriate increment. */
5230
5231 else
5232 {
5233 if (repeat_min > 1)
5234 {
5235 /* In the pre-compile phase, we don't actually do the replication. We
5236 just adjust the length as if we had. Do some paranoid checks for
5237 potential integer overflow. The INT64_OR_DOUBLE type is a 64-bit
5238 integer type when available, otherwise double. */
5239
5240 if (lengthptr != NULL)
5241 {
5242 int delta = (repeat_min - 1)*length_prevgroup;
5243 if ((INT64_OR_DOUBLE)(repeat_min - 1)*
5244 (INT64_OR_DOUBLE)length_prevgroup >
5245 (INT64_OR_DOUBLE)INT_MAX ||
5246 OFLOW_MAX - *lengthptr < delta)
5247 {
5248 *errorcodeptr = ERR20;
5249 goto FAILED;
5250 }
5251 *lengthptr += delta;
5252 }
5253
5254 /* This is compiling for real. If there is a set first byte for
5255 the group, and we have not yet set a "required byte", set it. Make
5256 sure there is enough workspace for copying forward references before
5257 doing the copy. */
5258
5259 else
5260 {
5261 if (groupsetfirstchar && reqchar < 0) reqchar = firstchar;
5262
5263 for (i = 1; i < repeat_min; i++)
5264 {
5265 pcre_uchar *hc;
5266 pcre_uchar *this_hwm = cd->hwm;
5267 memcpy(code, previous, IN_UCHARS(len));
5268
5269 while (cd->hwm > cd->start_workspace + cd->workspace_size -
5270 WORK_SIZE_SAFETY_MARGIN - (this_hwm - save_hwm))
5271 {
5272 int save_offset = save_hwm - cd->start_workspace;
5273 int this_offset = this_hwm - cd->start_workspace;
5274 *errorcodeptr = expand_workspace(cd);
5275 if (*errorcodeptr != 0) goto FAILED;
5276 save_hwm = (pcre_uchar *)cd->start_workspace + save_offset;
5277 this_hwm = (pcre_uchar *)cd->start_workspace + this_offset;
5278 }
5279
5280 for (hc = save_hwm; hc < this_hwm; hc += LINK_SIZE)
5281 {
5282 PUT(cd->hwm, 0, GET(hc, 0) + len);
5283 cd->hwm += LINK_SIZE;
5284 }
5285 save_hwm = this_hwm;
5286 code += len;
5287 }
5288 }
5289 }
5290
5291 if (repeat_max > 0) repeat_max -= repeat_min;
5292 }
5293
5294 /* This code is common to both the zero and non-zero minimum cases. If
5295 the maximum is limited, it replicates the group in a nested fashion,
5296 remembering the bracket starts on a stack. In the case of a zero minimum,
5297 the first one was set up above. In all cases the repeat_max now specifies
5298 the number of additional copies needed. Again, we must remember to
5299 replicate entries on the forward reference list. */
5300
5301 if (repeat_max >= 0)
5302 {
5303 /* In the pre-compile phase, we don't actually do the replication. We
5304 just adjust the length as if we had. For each repetition we must add 1
5305 to the length for BRAZERO and for all but the last repetition we must
5306 add 2 + 2*LINKSIZE to allow for the nesting that occurs. Do some
5307 paranoid checks to avoid integer overflow. The INT64_OR_DOUBLE type is
5308 a 64-bit integer type when available, otherwise double. */
5309
5310 if (lengthptr != NULL && repeat_max > 0)
5311 {
5312 int delta = repeat_max * (length_prevgroup + 1 + 2 + 2*LINK_SIZE) -
5313 2 - 2*LINK_SIZE; /* Last one doesn't nest */
5314 if ((INT64_OR_DOUBLE)repeat_max *
5315 (INT64_OR_DOUBLE)(length_prevgroup + 1 + 2 + 2*LINK_SIZE)
5316 > (INT64_OR_DOUBLE)INT_MAX ||
5317 OFLOW_MAX - *lengthptr < delta)
5318 {
5319 *errorcodeptr = ERR20;
5320 goto FAILED;
5321 }
5322 *lengthptr += delta;
5323 }
5324
5325 /* This is compiling for real */
5326
5327 else for (i = repeat_max - 1; i >= 0; i--)
5328 {
5329 pcre_uchar *hc;
5330 pcre_uchar *this_hwm = cd->hwm;
5331
5332 *code++ = OP_BRAZERO + repeat_type;
5333
5334 /* All but the final copy start a new nesting, maintaining the
5335 chain of brackets outstanding. */
5336
5337 if (i != 0)
5338 {
5339 int offset;
5340 *code++ = OP_BRA;
5341 offset = (bralink == NULL)? 0 : (int)(code - bralink);
5342 bralink = code;
5343 PUTINC(code, 0, offset);
5344 }
5345
5346 memcpy(code, previous, IN_UCHARS(len));
5347
5348 /* Ensure there is enough workspace for forward references before
5349 copying them. */
5350
5351 while (cd->hwm > cd->start_workspace + cd->workspace_size -
5352 WORK_SIZE_SAFETY_MARGIN - (this_hwm - save_hwm))
5353 {
5354 int save_offset = save_hwm - cd->start_workspace;
5355 int this_offset = this_hwm - cd->start_workspace;
5356 *errorcodeptr = expand_workspace(cd);
5357 if (*errorcodeptr != 0) goto FAILED;
5358 save_hwm = (pcre_uchar *)cd->start_workspace + save_offset;
5359 this_hwm = (pcre_uchar *)cd->start_workspace + this_offset;
5360 }
5361
5362 for (hc = save_hwm; hc < this_hwm; hc += LINK_SIZE)
5363 {
5364 PUT(cd->hwm, 0, GET(hc, 0) + len + ((i != 0)? 2+LINK_SIZE : 1));
5365 cd->hwm += LINK_SIZE;
5366 }
5367 save_hwm = this_hwm;
5368 code += len;
5369 }
5370
5371 /* Now chain through the pending brackets, and fill in their length
5372 fields (which are holding the chain links pro tem). */
5373
5374 while (bralink != NULL)
5375 {
5376 int oldlinkoffset;
5377 int offset = (int)(code - bralink + 1);
5378 pcre_uchar *bra = code - offset;
5379 oldlinkoffset = GET(bra, 1);
5380 bralink = (oldlinkoffset == 0)? NULL : bralink - oldlinkoffset;
5381 *code++ = OP_KET;
5382 PUTINC(code, 0, offset);
5383 PUT(bra, 1, offset);
5384 }
5385 }
5386
5387 /* If the maximum is unlimited, set a repeater in the final copy. For
5388 ONCE brackets, that's all we need to do. However, possessively repeated
5389 ONCE brackets can be converted into non-capturing brackets, as the
5390 behaviour of (?:xx)++ is the same as (?>xx)++ and this saves having to
5391 deal with possessive ONCEs specially.
5392
5393 Otherwise, when we are doing the actual compile phase, check to see
5394 whether this group is one that could match an empty string. If so,
5395 convert the initial operator to the S form (e.g. OP_BRA -> OP_SBRA) so
5396 that runtime checking can be done. [This check is also applied to ONCE
5397 groups at runtime, but in a different way.]
5398
5399 Then, if the quantifier was possessive and the bracket is not a
5400 conditional, we convert the BRA code to the POS form, and the KET code to
5401 KETRPOS. (It turns out to be convenient at runtime to detect this kind of
5402 subpattern at both the start and at the end.) The use of special opcodes
5403 makes it possible to reduce greatly the stack usage in pcre_exec(). If
5404 the group is preceded by OP_BRAZERO, convert this to OP_BRAPOSZERO.
5405
5406 Then, if the minimum number of matches is 1 or 0, cancel the possessive
5407 flag so that the default action below, of wrapping everything inside
5408 atomic brackets, does not happen. When the minimum is greater than 1,
5409 there will be earlier copies of the group, and so we still have to wrap
5410 the whole thing. */
5411
5412 else
5413 {
5414 pcre_uchar *ketcode = code - 1 - LINK_SIZE;
5415 pcre_uchar *bracode = ketcode - GET(ketcode, 1);
5416
5417 /* Convert possessive ONCE brackets to non-capturing */
5418
5419 if ((*bracode == OP_ONCE || *bracode == OP_ONCE_NC) &&
5420 possessive_quantifier) *bracode = OP_BRA;
5421
5422 /* For non-possessive ONCE brackets, all we need to do is to
5423 set the KET. */
5424
5425 if (*bracode == OP_ONCE || *bracode == OP_ONCE_NC)
5426 *ketcode = OP_KETRMAX + repeat_type;
5427
5428 /* Handle non-ONCE brackets and possessive ONCEs (which have been
5429 converted to non-capturing above). */
5430
5431 else
5432 {
5433 /* In the compile phase, check for empty string matching. */
5434
5435 if (lengthptr == NULL)
5436 {
5437 pcre_uchar *scode = bracode;
5438 do
5439 {
5440 if (could_be_empty_branch(scode, ketcode, utf, cd))
5441 {
5442 *bracode += OP_SBRA - OP_BRA;
5443 break;
5444 }
5445 scode += GET(scode, 1);
5446 }
5447 while (*scode == OP_ALT);
5448 }
5449
5450 /* Handle possessive quantifiers. */
5451
5452 if (possessive_quantifier)
5453 {
5454 /* For COND brackets, we wrap the whole thing in a possessively
5455 repeated non-capturing bracket, because we have not invented POS
5456 versions of the COND opcodes. Because we are moving code along, we
5457 must ensure that any pending recursive references are updated. */
5458
5459 if (*bracode == OP_COND || *bracode == OP_SCOND)
5460 {
5461 int nlen = (int)(code - bracode);
5462 *code = OP_END;
5463 adjust_recurse(bracode, 1 + LINK_SIZE, utf, cd, save_hwm);
5464 memmove(bracode + 1 + LINK_SIZE, bracode, IN_UCHARS(nlen));
5465 code += 1 + LINK_SIZE;
5466 nlen += 1 + LINK_SIZE;
5467 *bracode = OP_BRAPOS;
5468 *code++ = OP_KETRPOS;
5469 PUTINC(code, 0, nlen);
5470 PUT(bracode, 1, nlen);
5471 }
5472
5473 /* For non-COND brackets, we modify the BRA code and use KETRPOS. */
5474
5475 else
5476 {
5477 *bracode += 1; /* Switch to xxxPOS opcodes */
5478 *ketcode = OP_KETRPOS;
5479 }
5480
5481 /* If the minimum is zero, mark it as possessive, then unset the
5482 possessive flag when the minimum is 0 or 1. */
5483
5484 if (brazeroptr != NULL) *brazeroptr = OP_BRAPOSZERO;
5485 if (repeat_min < 2) possessive_quantifier = FALSE;
5486 }
5487
5488 /* Non-possessive quantifier */
5489
5490 else *ketcode = OP_KETRMAX + repeat_type;
5491 }
5492 }
5493 }
5494
5495 /* If previous is OP_FAIL, it was generated by an empty class [] in
5496 JavaScript mode. The other ways in which OP_FAIL can be generated, that is
5497 by (*FAIL) or (?!) set previous to NULL, which gives a "nothing to repeat"
5498 error above. We can just ignore the repeat in JS case. */
5499
5500 else if (*previous == OP_FAIL) goto END_REPEAT;
5501
5502 /* Else there's some kind of shambles */
5503
5504 else
5505 {
5506 *errorcodeptr = ERR11;
5507 goto FAILED;
5508 }
5509
5510 /* If the character following a repeat is '+', or if certain optimization
5511 tests above succeeded, possessive_quantifier is TRUE. For some opcodes,
5512 there are special alternative opcodes for this case. For anything else, we
5513 wrap the entire repeated item inside OP_ONCE brackets. Logically, the '+'
5514 notation is just syntactic sugar, taken from Sun's Java package, but the
5515 special opcodes can optimize it.
5516
5517 Some (but not all) possessively repeated subpatterns have already been
5518 completely handled in the code just above. For them, possessive_quantifier
5519 is always FALSE at this stage.
5520
5521 Note that the repeated item starts at tempcode, not at previous, which
5522 might be the first part of a string whose (former) last char we repeated.
5523
5524 Possessifying an 'exact' quantifier has no effect, so we can ignore it. But
5525 an 'upto' may follow. We skip over an 'exact' item, and then test the
5526 length of what remains before proceeding. */
5527
5528 if (possessive_quantifier)
5529 {
5530 int len;
5531
5532 if (*tempcode == OP_TYPEEXACT)
5533 tempcode += PRIV(OP_lengths)[*tempcode] +
5534 ((tempcode[1 + IMM2_SIZE] == OP_PROP
5535 || tempcode[1 + IMM2_SIZE] == OP_NOTPROP)? 2 : 0);
5536
5537 else if (*tempcode == OP_EXACT || *tempcode == OP_NOTEXACT)
5538 {
5539 tempcode += PRIV(OP_lengths)[*tempcode];
5540 #ifdef SUPPORT_UTF
5541 if (utf && HAS_EXTRALEN(tempcode[-1]))
5542 tempcode += GET_EXTRALEN(tempcode[-1]);
5543 #endif
5544 }
5545
5546 len = (int)(code - tempcode);
5547 if (len > 0) switch (*tempcode)
5548 {
5549 case OP_STAR: *tempcode = OP_POSSTAR; break;
5550 case OP_PLUS: *tempcode = OP_POSPLUS; break;
5551 case OP_QUERY: *tempcode = OP_POSQUERY; break;
5552 case OP_UPTO: *tempcode = OP_POSUPTO; break;
5553
5554 case OP_STARI: *tempcode = OP_POSSTARI; break;
5555 case OP_PLUSI: *tempcode = OP_POSPLUSI; break;
5556 case OP_QUERYI: *tempcode = OP_POSQUERYI; break;
5557 case OP_UPTOI: *tempcode = OP_POSUPTOI; break;
5558
5559 case OP_NOTSTAR: *tempcode = OP_NOTPOSSTAR; break;
5560 case OP_NOTPLUS: *tempcode = OP_NOTPOSPLUS; break;
5561 case OP_NOTQUERY: *tempcode = OP_NOTPOSQUERY; break;
5562 case OP_NOTUPTO: *tempcode = OP_NOTPOSUPTO; break;
5563
5564 case OP_NOTSTARI: *tempcode = OP_NOTPOSSTARI; break;
5565 case OP_NOTPLUSI: *tempcode = OP_NOTPOSPLUSI; break;
5566 case OP_NOTQUERYI: *tempcode = OP_NOTPOSQUERYI; break;
5567 case OP_NOTUPTOI: *tempcode = OP_NOTPOSUPTOI; break;
5568
5569 case OP_TYPESTAR: *tempcode = OP_TYPEPOSSTAR; break;
5570 case OP_TYPEPLUS: *tempcode = OP_TYPEPOSPLUS; break;
5571 case OP_TYPEQUERY: *tempcode = OP_TYPEPOSQUERY; break;
5572 case OP_TYPEUPTO: *tempcode = OP_TYPEPOSUPTO; break;
5573
5574 /* Because we are moving code along, we must ensure that any
5575 pending recursive references are updated. */
5576
5577 default:
5578 *code = OP_END;
5579 adjust_recurse(tempcode, 1 + LINK_SIZE, utf, cd, save_hwm);
5580 memmove(tempcode + 1 + LINK_SIZE, tempcode, IN_UCHARS(len));
5581 code += 1 + LINK_SIZE;
5582 len += 1 + LINK_SIZE;
5583 tempcode[0] = OP_ONCE;
5584 *code++ = OP_KET;
5585 PUTINC(code, 0, len);
5586 PUT(tempcode, 1, len);
5587 break;
5588 }
5589 }
5590
5591 /* In all case we no longer have a previous item. We also set the
5592 "follows varying string" flag for subsequently encountered reqchars if
5593 it isn't already set and we have just passed a varying length item. */
5594
5595 END_REPEAT:
5596 previous = NULL;
5597 cd->req_varyopt |= reqvary;
5598 break;
5599
5600
5601 /* ===================================================================*/
5602 /* Start of nested parenthesized sub-expression, or comment or lookahead or
5603 lookbehind or option setting or condition or all the other extended
5604 parenthesis forms. */
5605
5606 case CHAR_LEFT_PARENTHESIS:
5607 newoptions = options;
5608 skipbytes = 0;
5609 bravalue = OP_CBRA;
5610 save_hwm = cd->hwm;
5611 reset_bracount = FALSE;
5612
5613 /* First deal with various "verbs" that can be introduced by '*'. */
5614
5615 ptr++;
5616 if (ptr[0] == CHAR_ASTERISK && (ptr[1] == ':'
5617 || (MAX_255(ptr[1]) && ((cd->ctypes[ptr[1]] & ctype_letter) != 0))))
5618 {
5619 int i, namelen;
5620 int arglen = 0;
5621 const char *vn = verbnames;
5622 const pcre_uchar *name = ptr + 1;
5623 const pcre_uchar *arg = NULL;
5624 previous = NULL;
5625 ptr++;
5626 while (MAX_255(*ptr) && (cd->ctypes[*ptr] & ctype_letter) != 0) ptr++;
5627 namelen = (int)(ptr - name);
5628
5629 /* It appears that Perl allows any characters whatsoever, other than
5630 a closing parenthesis, to appear in arguments, so we no longer insist on
5631 letters, digits, and underscores. */
5632
5633 if (*ptr == CHAR_COLON)
5634 {
5635 arg = ++ptr;
5636 while (*ptr != 0 && *ptr != CHAR_RIGHT_PARENTHESIS) ptr++;
5637 arglen = (int)(ptr - arg);
5638 if (arglen > (int)MAX_MARK)
5639 {
5640 *errorcodeptr = ERR75;
5641 goto FAILED;
5642 }
5643 }
5644
5645 if (*ptr != CHAR_RIGHT_PARENTHESIS)
5646 {
5647 *errorcodeptr = ERR60;
5648 goto FAILED;
5649 }
5650
5651 /* Scan the table of verb names */
5652
5653 for (i = 0; i < verbcount; i++)
5654 {
5655 if (namelen == verbs[i].len &&
5656 STRNCMP_UC_C8(name, vn, namelen) == 0)
5657 {
5658 int setverb;
5659
5660 /* Check for open captures before ACCEPT and convert it to
5661 ASSERT_ACCEPT if in an assertion. */
5662
5663 if (verbs[i].op == OP_ACCEPT)
5664 {
5665 open_capitem *oc;
5666 if (arglen != 0)
5667 {
5668 *errorcodeptr = ERR59;
5669 goto FAILED;
5670 }
5671 cd->had_accept = TRUE;
5672 for (oc = cd->open_caps; oc != NULL; oc = oc->next)
5673 {
5674 *code++ = OP_CLOSE;
5675 PUT2INC(code, 0, oc->number);
5676 }
5677 setverb = *code++ =
5678 (cd->assert_depth > 0)? OP_ASSERT_ACCEPT : OP_ACCEPT;
5679
5680 /* Do not set firstchar after *ACCEPT */
5681 if (firstchar == REQ_UNSET) firstchar = REQ_NONE;
5682 }
5683
5684 /* Handle other cases with/without an argument */
5685
5686 else if (arglen == 0)
5687 {
5688 if (verbs[i].op < 0) /* Argument is mandatory */
5689 {
5690 *errorcodeptr = ERR66;
5691 goto FAILED;
5692 }
5693 setverb = *code++ = verbs[i].op;
5694 }
5695
5696 else
5697 {
5698 if (verbs[i].op_arg < 0) /* Argument is forbidden */
5699 {
5700 *errorcodeptr = ERR59;
5701 goto FAILED;
5702 }
5703 setverb = *code++ = verbs[i].op_arg;
5704 *code++ = arglen;
5705 memcpy(code, arg, IN_UCHARS(arglen));
5706 code += arglen;
5707 *code++ = 0;
5708 }
5709
5710 switch (setverb)
5711 {
5712 case OP_THEN:
5713 case OP_THEN_ARG:
5714 cd->external_flags |= PCRE_HASTHEN;
5715 break;
5716
5717 case OP_PRUNE:
5718 case OP_PRUNE_ARG:
5719 case OP_SKIP:
5720 case OP_SKIP_ARG:
5721 cd->had_pruneorskip = TRUE;
5722 break;
5723 }
5724
5725 break; /* Found verb, exit loop */
5726 }
5727
5728 vn += verbs[i].len + 1;
5729 }
5730
5731 if (i < verbcount) continue; /* Successfully handled a verb */
5732 *errorcodeptr = ERR60; /* Verb not recognized */
5733 goto FAILED;
5734 }
5735
5736 /* Deal with the extended parentheses; all are introduced by '?', and the
5737 appearance of any of them means that this is not a capturing group. */
5738
5739 else if (*ptr == CHAR_QUESTION_MARK)
5740 {
5741 int i, set, unset, namelen;
5742 int *optset;
5743 const pcre_uchar *name;
5744 pcre_uchar *slot;
5745
5746 switch (*(++ptr))
5747 {
5748 case CHAR_NUMBER_SIGN: /* Comment; skip to ket */
5749 ptr++;
5750 while (*ptr != 0 && *ptr != CHAR_RIGHT_PARENTHESIS) ptr++;
5751 if (*ptr == 0)
5752 {
5753 *errorcodeptr = ERR18;
5754 goto FAILED;
5755 }
5756 continue;
5757
5758
5759 /* ------------------------------------------------------------ */
5760 case CHAR_VERTICAL_LINE: /* Reset capture count for each branch */
5761 reset_bracount = TRUE;
5762 /* Fall through */
5763
5764 /* ------------------------------------------------------------ */
5765 case CHAR_COLON: /* Non-capturing bracket */
5766 bravalue = OP_BRA;
5767 ptr++;
5768 break;
5769
5770
5771 /* ------------------------------------------------------------ */
5772 case CHAR_LEFT_PARENTHESIS:
5773 bravalue = OP_COND; /* Conditional group */
5774
5775 /* A condition can be an assertion, a number (referring to a numbered
5776 group), a name (referring to a named group), or 'R', referring to
5777 recursion. R<digits> and R&name are also permitted for recursion tests.
5778
5779 There are several syntaxes for testing a named group: (?(name)) is used
5780 by Python; Perl 5.10 onwards uses (?(<name>) or (?('name')).
5781
5782 There are two unfortunate ambiguities, caused by history. (a) 'R' can
5783 be the recursive thing or the name 'R' (and similarly for 'R' followed
5784 by digits), and (b) a number could be a name that consists of digits.
5785 In both cases, we look for a name first; if not found, we try the other
5786 cases. */
5787
5788 /* For conditions that are assertions, check the syntax, and then exit
5789 the switch. This will take control down to where bracketed groups,
5790 including assertions, are processed. */
5791
5792 if (ptr[1] == CHAR_QUESTION_MARK && (ptr[2] == CHAR_EQUALS_SIGN ||
5793 ptr[2] == CHAR_EXCLAMATION_MARK || ptr[2] == CHAR_LESS_THAN_SIGN))
5794 break;
5795
5796 /* Most other conditions use OP_CREF (a couple change to OP_RREF
5797 below), and all need to skip 1+IMM2_SIZE bytes at the start of the group. */
5798
5799 code[1+LINK_SIZE] = OP_CREF;
5800 skipbytes = 1+IMM2_SIZE;
5801 refsign = -1;
5802
5803 /* Check for a test for recursion in a named group. */
5804
5805 if (ptr[1] == CHAR_R && ptr[2] == CHAR_AMPERSAND)
5806 {
5807 terminator = -1;
5808 ptr += 2;
5809 code[1+LINK_SIZE] = OP_RREF; /* Change the type of test */
5810 }
5811
5812 /* Check for a test for a named group's having been set, using the Perl
5813 syntax (?(<name>) or (?('name') */
5814
5815 else if (ptr[1] == CHAR_LESS_THAN_SIGN)
5816 {
5817 terminator = CHAR_GREATER_THAN_SIGN;
5818 ptr++;
5819 }
5820 else if (ptr[1] == CHAR_APOSTROPHE)
5821 {
5822 terminator = CHAR_APOSTROPHE;
5823 ptr++;
5824 }
5825 else
5826 {
5827 terminator = 0;
5828 if (ptr[1] == CHAR_MINUS || ptr[1] == CHAR_PLUS) refsign = *(++ptr);
5829 }
5830
5831 /* We now expect to read a name; any thing else is an error */
5832
5833 if (!MAX_255(ptr[1]) || (cd->ctypes[ptr[1]] & ctype_word) == 0)
5834 {
5835 ptr += 1; /* To get the right offset */
5836 *errorcodeptr = ERR28;
5837 goto FAILED;
5838 }
5839
5840 /* Read the name, but also get it as a number if it's all digits */
5841
5842 recno = 0;
5843 name = ++ptr;
5844 while (MAX_255(*ptr) && (cd->ctypes[*ptr] & ctype_word) != 0)
5845 {
5846 if (recno >= 0)
5847 recno = (IS_DIGIT(*ptr))? recno * 10 + *ptr - CHAR_0 : -1;
5848 ptr++;
5849 }
5850 namelen = (int)(ptr - name);
5851
5852 if ((terminator > 0 && *ptr++ != terminator) ||
5853 *ptr++ != CHAR_RIGHT_PARENTHESIS)
5854 {
5855 ptr--; /* Error offset */
5856 *errorcodeptr = ERR26;
5857 goto FAILED;
5858 }
5859
5860 /* Do no further checking in the pre-compile phase. */
5861
5862 if (lengthptr != NULL) break;
5863
5864 /* In the real compile we do the work of looking for the actual
5865 reference. If the string started with "+" or "-" we require the rest to
5866 be digits, in which case recno will be set. */
5867
5868 if (refsign > 0)
5869 {
5870 if (recno <= 0)
5871 {
5872 *errorcodeptr = ERR58;
5873 goto FAILED;
5874 }
5875 recno = (refsign == CHAR_MINUS)?
5876 cd->bracount - recno + 1 : recno +cd->bracount;
5877 if (recno <= 0 || recno > cd->final_bracount)
5878 {
5879 *errorcodeptr = ERR15;
5880 goto FAILED;
5881 }
5882 PUT2(code, 2+LINK_SIZE, recno);
5883 break;
5884 }
5885
5886 /* Otherwise (did not start with "+" or "-"), start by looking for the
5887 name. If we find a name, add one to the opcode to change OP_CREF or
5888 OP_RREF into OP_NCREF or OP_NRREF. These behave exactly the same,
5889 except they record that the reference was originally to a name. The
5890 information is used to check duplicate names. */
5891
5892 slot = cd->name_table;
5893 for (i = 0; i < cd->names_found; i++)
5894 {
5895 if (STRNCMP_UC_UC(name, slot+IMM2_SIZE, namelen) == 0) break;
5896 slot += cd->name_entry_size;
5897 }
5898
5899 /* Found a previous named subpattern */
5900
5901 if (i < cd->names_found)
5902 {
5903 recno = GET2(slot, 0);
5904 PUT2(code, 2+LINK_SIZE, recno);
5905 code[1+LINK_SIZE]++;
5906 }
5907
5908 /* Search the pattern for a forward reference */
5909
5910 else if ((i = find_parens(cd, name, namelen,
5911 (options & PCRE_EXTENDED) != 0, utf)) > 0)
5912 {
5913 PUT2(code, 2+LINK_SIZE, i);
5914 code[1+LINK_SIZE]++;
5915 }
5916
5917 /* If terminator == 0 it means that the name followed directly after
5918 the opening parenthesis [e.g. (?(abc)...] and in this case there are
5919 some further alternatives to try. For the cases where terminator != 0
5920 [things like (?(<name>... or (?('name')... or (?(R&name)... ] we have
5921 now checked all the possibilities, so give an error. */
5922
5923 else if (terminator != 0)
5924 {
5925 *errorcodeptr = ERR15;
5926 goto FAILED;
5927 }
5928
5929 /* Check for (?(R) for recursion. Allow digits after R to specify a
5930 specific group number. */
5931
5932 else if (*name == CHAR_R)
5933 {
5934 recno = 0;
5935 for (i = 1; i < namelen; i++)
5936 {
5937 if (!IS_DIGIT(name[i]))
5938 {
5939 *errorcodeptr = ERR15;
5940 goto FAILED;
5941 }
5942 recno = recno * 10 + name[i] - CHAR_0;
5943 }
5944 if (recno == 0) recno = RREF_ANY;
5945 code[1+LINK_SIZE] = OP_RREF; /* Change test type */
5946 PUT2(code, 2+LINK_SIZE, recno);
5947 }
5948
5949 /* Similarly, check for the (?(DEFINE) "condition", which is always
5950 false. */
5951
5952 else if (namelen == 6 && STRNCMP_UC_C8(name, STRING_DEFINE, 6) == 0)
5953 {
5954 code[1+LINK_SIZE] = OP_DEF;
5955 skipbytes = 1;
5956 }
5957
5958 /* Check for the "name" actually being a subpattern number. We are
5959 in the second pass here, so final_bracount is set. */
5960
5961 else if (recno > 0 && recno <= cd->final_bracount)
5962 {
5963 PUT2(code, 2+LINK_SIZE, recno);
5964 }
5965
5966 /* Either an unidentified subpattern, or a reference to (?(0) */
5967
5968 else
5969 {
5970 *errorcodeptr = (recno == 0)? ERR35: ERR15;
5971 goto FAILED;
5972 }
5973 break;
5974
5975
5976 /* ------------------------------------------------------------ */
5977 case CHAR_EQUALS_SIGN: /* Positive lookahead */
5978 bravalue = OP_ASSERT;
5979 cd->assert_depth += 1;
5980 ptr++;
5981 break;
5982
5983
5984 /* ------------------------------------------------------------ */
5985 case CHAR_EXCLAMATION_MARK: /* Negative lookahead */
5986 ptr++;
5987 if (*ptr == CHAR_RIGHT_PARENTHESIS) /* Optimize (?!) */
5988 {
5989 *code++ = OP_FAIL;
5990 previous = NULL;
5991 continue;
5992 }
5993 bravalue = OP_ASSERT_NOT;
5994 cd->assert_depth += 1;
5995 break;
5996
5997
5998 /* ------------------------------------------------------------ */
5999 case CHAR_LESS_THAN_SIGN: /* Lookbehind or named define */
6000 switch (ptr[1])
6001 {
6002 case CHAR_EQUALS_SIGN: /* Positive lookbehind */
6003 bravalue = OP_ASSERTBACK;
6004 cd->assert_depth += 1;
6005 ptr += 2;
6006 break;
6007
6008 case CHAR_EXCLAMATION_MARK: /* Negative lookbehind */
6009 bravalue = OP_ASSERTBACK_NOT;
6010 cd->assert_depth += 1;
6011 ptr += 2;
6012 break;
6013
6014 default: /* Could be name define, else bad */
6015 if (MAX_255(ptr[1]) && (cd->ctypes[ptr[1]] & ctype_word) != 0)
6016 goto DEFINE_NAME;
6017 ptr++; /* Correct offset for error */
6018 *errorcodeptr = ERR24;
6019 goto FAILED;
6020 }
6021 break;
6022
6023
6024 /* ------------------------------------------------------------ */
6025 case CHAR_GREATER_THAN_SIGN: /* One-time brackets */
6026 bravalue = OP_ONCE;
6027 ptr++;
6028 break;
6029
6030
6031 /* ------------------------------------------------------------ */
6032 case CHAR_C: /* Callout - may be followed by digits; */
6033 previous_callout = code; /* Save for later completion */
6034 after_manual_callout = 1; /* Skip one item before completing */
6035 *code++ = OP_CALLOUT;
6036 {
6037 int n = 0;
6038 ptr++;
6039 while(IS_DIGIT(*ptr))
6040 n = n * 10 + *ptr++ - CHAR_0;
6041 if (*ptr != CHAR_RIGHT_PARENTHESIS)
6042 {
6043 *errorcodeptr = ERR39;
6044 goto FAILED;
6045 }
6046 if (n > 255)
6047 {
6048 *errorcodeptr = ERR38;
6049 goto FAILED;
6050 }
6051 *code++ = n;
6052 PUT(code, 0, (int)(ptr - cd->start_pattern + 1)); /* Pattern offset */
6053 PUT(code, LINK_SIZE, 0); /* Default length */
6054 code += 2 * LINK_SIZE;
6055 }
6056 previous = NULL;
6057 continue;
6058
6059
6060 /* ------------------------------------------------------------ */
6061 case CHAR_P: /* Python-style named subpattern handling */
6062 if (*(++ptr) == CHAR_EQUALS_SIGN ||
6063 *ptr == CHAR_GREATER_THAN_SIGN) /* Reference or recursion */
6064 {
6065 is_recurse = *ptr == CHAR_GREATER_THAN_SIGN;
6066 terminator = CHAR_RIGHT_PARENTHESIS;
6067 goto NAMED_REF_OR_RECURSE;
6068 }
6069 else if (*ptr != CHAR_LESS_THAN_SIGN) /* Test for Python-style defn */
6070 {
6071 *errorcodeptr = ERR41;
6072 goto FAILED;
6073 }
6074 /* Fall through to handle (?P< as (?< is handled */
6075
6076
6077 /* ------------------------------------------------------------ */
6078 DEFINE_NAME: /* Come here from (?< handling */
6079 case CHAR_APOSTROPHE:
6080 {
6081 terminator = (*ptr == CHAR_LESS_THAN_SIGN)?
6082 CHAR_GREATER_THAN_SIGN : CHAR_APOSTROPHE;
6083 name = ++ptr;
6084
6085 while (MAX_255(*ptr) && (cd->ctypes[*ptr] & ctype_word) != 0) ptr++;
6086 namelen = (int)(ptr - name);
6087
6088 /* In the pre-compile phase, just do a syntax check. */
6089
6090 if (lengthptr != NULL)
6091 {
6092 if (*ptr != terminator)
6093 {
6094 *errorcodeptr = ERR42;
6095 goto FAILED;
6096 }
6097 if (cd->names_found >= MAX_NAME_COUNT)
6098 {
6099 *errorcodeptr = ERR49;
6100 goto FAILED;
6101 }
6102 if (namelen + IMM2_SIZE + 1 > cd->name_entry_size)
6103 {
6104 cd->name_entry_size = namelen + IMM2_SIZE + 1;
6105 if (namelen > MAX_NAME_SIZE)
6106 {
6107 *errorcodeptr = ERR48;
6108 goto FAILED;
6109 }
6110 }
6111 }
6112
6113 /* In the real compile, create the entry in the table, maintaining
6114 alphabetical order. Duplicate names for different numbers are
6115 permitted only if PCRE_DUPNAMES is set. Duplicate names for the same
6116 number are always OK. (An existing number can be re-used if (?|
6117 appears in the pattern.) In either event, a duplicate name results in
6118 a duplicate entry in the table, even if the number is the same. This
6119 is because the number of names, and hence the table size, is computed
6120 in the pre-compile, and it affects various numbers and pointers which
6121 would all have to be modified, and the compiled code moved down, if
6122 duplicates with the same number were omitted from the table. This
6123 doesn't seem worth the hassle. However, *different* names for the
6124 same number are not permitted. */
6125
6126 else
6127 {
6128 BOOL dupname = FALSE;
6129 slot = cd->name_table;
6130
6131 for (i = 0; i < cd->names_found; i++)
6132 {
6133 int crc = memcmp(name, slot+IMM2_SIZE, IN_UCHARS(namelen));
6134 if (crc == 0)
6135 {
6136 if (slot[IMM2_SIZE+namelen] == 0)
6137 {
6138 if (GET2(slot, 0) != cd->bracount + 1 &&
6139 (options & PCRE_DUPNAMES) == 0)
6140 {
6141 *errorcodeptr = ERR43;
6142 goto FAILED;
6143 }
6144 else dupname = TRUE;
6145 }
6146 else crc = -1; /* Current name is a substring */
6147 }
6148
6149 /* Make space in the table and break the loop for an earlier
6150 name. For a duplicate or later name, carry on. We do this for
6151 duplicates so that in the simple case (when ?(| is not used) they
6152 are in order of their numbers. */
6153
6154 if (crc < 0)
6155 {
6156 memmove(slot + cd->name_entry_size, slot,
6157 IN_UCHARS((cd->names_found - i) * cd->name_entry_size));
6158 break;
6159 }
6160
6161 /* Continue the loop for a later or duplicate name */
6162
6163 slot += cd->name_entry_size;
6164 }
6165
6166 /* For non-duplicate names, check for a duplicate number before
6167 adding the new name. */
6168
6169 if (!dupname)
6170 {
6171 pcre_uchar *cslot = cd->name_table;
6172 for (i = 0; i < cd->names_found; i++)
6173 {
6174 if (cslot != slot)
6175 {
6176 if (GET2(cslot, 0) == cd->bracount + 1)
6177 {
6178 *errorcodeptr = ERR65;
6179 goto FAILED;
6180 }
6181 }
6182 else i--;
6183 cslot += cd->name_entry_size;
6184 }
6185 }
6186
6187 PUT2(slot, 0, cd->bracount + 1);
6188 memcpy(slot + IMM2_SIZE, name, IN_UCHARS(namelen));
6189 slot[IMM2_SIZE + namelen] = 0;
6190 }
6191 }
6192
6193 /* In both pre-compile and compile, count the number of names we've
6194 encountered. */
6195
6196 cd->names_found++;
6197 ptr++; /* Move past > or ' */
6198 goto NUMBERED_GROUP;
6199
6200
6201 /* ------------------------------------------------------------ */
6202 case CHAR_AMPERSAND: /* Perl recursion/subroutine syntax */
6203 terminator = CHAR_RIGHT_PARENTHESIS;
6204 is_recurse = TRUE;
6205 /* Fall through */
6206
6207 /* We come here from the Python syntax above that handles both
6208 references (?P=name) and recursion (?P>name), as well as falling
6209 through from the Perl recursion syntax (?&name). We also come here from
6210 the Perl \k<name> or \k'name' back reference syntax and the \k{name}
6211 .NET syntax, and the Oniguruma \g<...> and \g'...' subroutine syntax. */
6212
6213 NAMED_REF_OR_RECURSE:
6214 name = ++ptr;
6215 while (MAX_255(*ptr) && (cd->ctypes[*ptr] & ctype_word) != 0) ptr++;
6216 namelen = (int)(ptr - name);
6217
6218 /* In the pre-compile phase, do a syntax check. We used to just set
6219 a dummy reference number, because it was not used in the first pass.
6220 However, with the change of recursive back references to be atomic,
6221 we have to look for the number so that this state can be identified, as
6222 otherwise the incorrect length is computed. If it's not a backwards
6223 reference, the dummy number will do. */
6224
6225 if (lengthptr != NULL)
6226 {
6227 const pcre_uchar *temp;
6228
6229 if (namelen == 0)
6230 {
6231 *errorcodeptr = ERR62;
6232 goto FAILED;
6233 }
6234 if (*ptr != terminator)
6235 {
6236 *errorcodeptr = ERR42;
6237 goto FAILED;
6238 }
6239 if (namelen > MAX_NAME_SIZE)
6240 {
6241 *errorcodeptr = ERR48;
6242 goto FAILED;
6243 }
6244
6245 /* The name table does not exist in the first pass, so we cannot
6246 do a simple search as in the code below. Instead, we have to scan the
6247 pattern to find the number. It is important that we scan it only as
6248 far as we have got because the syntax of named subpatterns has not
6249 been checked for the rest of the pattern, and find_parens() assumes
6250 correct syntax. In any case, it's a waste of resources to scan
6251 further. We stop the scan at the current point by temporarily
6252 adjusting the value of cd->endpattern. */
6253
6254 temp = cd->end_pattern;
6255 cd->end_pattern = ptr;
6256 recno = find_parens(cd, name, namelen,
6257 (options & PCRE_EXTENDED) != 0, utf);
6258 cd->end_pattern = temp;
6259 if (recno < 0) recno = 0; /* Forward ref; set dummy number */
6260 }
6261
6262 /* In the real compile, seek the name in the table. We check the name
6263 first, and then check that we have reached the end of the name in the
6264 table. That way, if the name that is longer than any in the table,
6265 the comparison will fail without reading beyond the table entry. */
6266
6267 else
6268 {
6269 slot = cd->name_table;
6270 for (i = 0; i < cd->names_found; i++)
6271 {
6272 if (STRNCMP_UC_UC(name, slot+IMM2_SIZE, namelen) == 0 &&
6273 slot[IMM2_SIZE+namelen] == 0)
6274 break;
6275 slot += cd->name_entry_size;
6276 }
6277
6278 if (i < cd->names_found) /* Back reference */
6279 {
6280 recno = GET2(slot, 0);
6281 }
6282 else if ((recno = /* Forward back reference */
6283 find_parens(cd, name, namelen,
6284 (options & PCRE_EXTENDED) != 0, utf)) <= 0)
6285 {
6286 *errorcodeptr = ERR15;
6287 goto FAILED;
6288 }
6289 }
6290
6291 /* In both phases, we can now go to the code than handles numerical
6292 recursion or backreferences. */
6293
6294 if (is_recurse) goto HANDLE_RECURSION;
6295 else goto HANDLE_REFERENCE;
6296
6297
6298 /* ------------------------------------------------------------ */
6299 case CHAR_R: /* Recursion */
6300 ptr++; /* Same as (?0) */
6301 /* Fall through */
6302
6303
6304 /* ------------------------------------------------------------ */
6305 case CHAR_MINUS: case CHAR_PLUS: /* Recursion or subroutine */
6306 case CHAR_0: case CHAR_1: case CHAR_2: case CHAR_3: case CHAR_4:
6307 case CHAR_5: case CHAR_6: case CHAR_7: case CHAR_8: case CHAR_9:
6308 {
6309 const pcre_uchar *called;
6310 terminator = CHAR_RIGHT_PARENTHESIS;
6311
6312 /* Come here from the \g<...> and \g'...' code (Oniguruma
6313 compatibility). However, the syntax has been checked to ensure that
6314 the ... are a (signed) number, so that neither ERR63 nor ERR29 will
6315 be called on this path, nor with the jump to OTHER_CHAR_AFTER_QUERY
6316 ever be taken. */
6317
6318 HANDLE_NUMERICAL_RECURSION:
6319
6320 if ((refsign = *ptr) == CHAR_PLUS)
6321 {
6322 ptr++;
6323 if (!IS_DIGIT(*ptr))
6324 {
6325 *errorcodeptr = ERR63;
6326 goto FAILED;
6327 }
6328 }
6329 else if (refsign == CHAR_MINUS)
6330 {
6331 if (!IS_DIGIT(ptr[1]))
6332 goto OTHER_CHAR_AFTER_QUERY;
6333 ptr++;
6334 }
6335
6336 recno = 0;
6337 while(IS_DIGIT(*ptr))
6338 recno = recno * 10 + *ptr++ - CHAR_0;
6339
6340 if (*ptr != terminator)
6341 {
6342 *errorcodeptr = ERR29;
6343 goto FAILED;
6344 }
6345
6346 if (refsign == CHAR_MINUS)
6347 {
6348 if (recno == 0)
6349 {
6350 *errorcodeptr = ERR58;
6351 goto FAILED;
6352 }
6353 recno = cd->bracount - recno + 1;
6354 if (recno <= 0)
6355 {
6356 *errorcodeptr = ERR15;
6357 goto FAILED;
6358 }
6359 }
6360 else if (refsign == CHAR_PLUS)
6361 {
6362 if (recno == 0)
6363 {
6364 *errorcodeptr = ERR58;
6365 goto FAILED;
6366 }
6367 recno += cd->bracount;
6368 }
6369
6370 /* Come here from code above that handles a named recursion */
6371
6372 HANDLE_RECURSION:
6373
6374 previous = code;
6375 called = cd->start_code;
6376
6377 /* When we are actually compiling, find the bracket that is being
6378 referenced. Temporarily end the regex in case it doesn't exist before
6379 this point. If we end up with a forward reference, first check that
6380 the bracket does occur later so we can give the error (and position)
6381 now. Then remember this forward reference in the workspace so it can
6382 be filled in at the end. */
6383
6384 if (lengthptr == NULL)
6385 {
6386 *code = OP_END;
6387 if (recno != 0)
6388 called = PRIV(find_bracket)(cd->start_code, utf, recno);
6389
6390 /* Forward reference */
6391
6392 if (called == NULL)
6393 {
6394 if (find_parens(cd, NULL, recno,
6395 (options & PCRE_EXTENDED) != 0, utf) < 0)
6396 {
6397 *errorcodeptr = ERR15;
6398 goto FAILED;
6399 }
6400
6401 /* Fudge the value of "called" so that when it is inserted as an
6402 offset below, what it actually inserted is the reference number
6403 of the group. Then remember the forward reference. */
6404
6405 called = cd->start_code + recno;
6406 if (cd->hwm >= cd->start_workspace + cd->workspace_size -
6407 WORK_SIZE_SAFETY_MARGIN)
6408 {
6409 *errorcodeptr = expand_workspace(cd);
6410 if (*errorcodeptr != 0) goto FAILED;
6411 }
6412 PUTINC(cd->hwm, 0, (int)(code + 1 - cd->start_code));
6413 }
6414
6415 /* If not a forward reference, and the subpattern is still open,
6416 this is a recursive call. We check to see if this is a left
6417 recursion that could loop for ever, and diagnose that case. We
6418 must not, however, do this check if we are in a conditional
6419 subpattern because the condition might be testing for recursion in
6420 a pattern such as /(?(R)a+|(?R)b)/, which is perfectly valid.
6421 Forever loops are also detected at runtime, so those that occur in
6422 conditional subpatterns will be picked up then. */
6423
6424 else if (GET(called, 1) == 0 && cond_depth <= 0 &&
6425 could_be_empty(called, code, bcptr, utf, cd))
6426 {
6427 *errorcodeptr = ERR40;
6428 goto FAILED;
6429 }
6430 }
6431
6432 /* Insert the recursion/subroutine item. It does not have a set first
6433 character (relevant if it is repeated, because it will then be
6434 wrapped with ONCE brackets). */
6435
6436 *code = OP_RECURSE;
6437 PUT(code, 1, (int)(called - cd->start_code));
6438 code += 1 + LINK_SIZE;
6439 groupsetfirstchar = FALSE;
6440 }
6441
6442 /* Can't determine a first byte now */
6443
6444 if (firstchar == REQ_UNSET) firstchar = REQ_NONE;
6445 continue;
6446
6447
6448 /* ------------------------------------------------------------ */
6449 default: /* Other characters: check option setting */
6450 OTHER_CHAR_AFTER_QUERY:
6451 set = unset = 0;
6452 optset = &set;
6453
6454 while (*ptr != CHAR_RIGHT_PARENTHESIS && *ptr != CHAR_COLON)
6455 {
6456 switch (*ptr++)
6457 {
6458 case CHAR_MINUS: optset = &unset; break;
6459
6460 case CHAR_J: /* Record that it changed in the external options */
6461 *optset |= PCRE_DUPNAMES;
6462 cd->external_flags |= PCRE_JCHANGED;
6463 break;
6464
6465 case CHAR_i: *optset |= PCRE_CASELESS; break;
6466 case CHAR_m: *optset |= PCRE_MULTILINE; break;
6467 case CHAR_s: *optset |= PCRE_DOTALL; break;
6468 case CHAR_x: *optset |= PCRE_EXTENDED; break;
6469 case CHAR_U: *optset |= PCRE_UNGREEDY; break;
6470 case CHAR_X: *optset |= PCRE_EXTRA; break;
6471
6472 default: *errorcodeptr = ERR12;
6473 ptr--; /* Correct the offset */
6474 goto FAILED;
6475 }
6476 }
6477
6478 /* Set up the changed option bits, but don't change anything yet. */
6479
6480 newoptions = (options | set) & (~unset);
6481
6482 /* If the options ended with ')' this is not the start of a nested
6483 group with option changes, so the options change at this level. If this
6484 item is right at the start of the pattern, the options can be
6485 abstracted and made external in the pre-compile phase, and ignored in
6486 the compile phase. This can be helpful when matching -- for instance in
6487 caseless checking of required bytes.
6488
6489 If the code pointer is not (cd->start_code + 1 + LINK_SIZE), we are
6490 definitely *not* at the start of the pattern because something has been
6491 compiled. In the pre-compile phase, however, the code pointer can have
6492 that value after the start, because it gets reset as code is discarded
6493 during the pre-compile. However, this can happen only at top level - if
6494 we are within parentheses, the starting BRA will still be present. At
6495 any parenthesis level, the length value can be used to test if anything
6496 has been compiled at that level. Thus, a test for both these conditions
6497 is necessary to ensure we correctly detect the start of the pattern in
6498 both phases.
6499
6500 If we are not at the pattern start, reset the greedy defaults and the
6501 case value for firstchar and reqchar. */
6502
6503 if (*ptr == CHAR_RIGHT_PARENTHESIS)
6504 {
6505 if (code == cd->start_code + 1 + LINK_SIZE &&
6506 (lengthptr == NULL || *lengthptr == 2 + 2*LINK_SIZE))
6507 {
6508 cd->external_options = newoptions;
6509 }
6510 else
6511 {
6512 greedy_default = ((newoptions & PCRE_UNGREEDY) != 0);
6513 greedy_non_default = greedy_default ^ 1;
6514 req_caseopt = ((newoptions & PCRE_CASELESS) != 0)? REQ_CASELESS:0;
6515 }
6516
6517 /* Change options at this level, and pass them back for use
6518 in subsequent branches. */
6519
6520 *optionsptr = options = newoptions;
6521 previous = NULL; /* This item can't be repeated */
6522 continue; /* It is complete */
6523 }
6524
6525 /* If the options ended with ':' we are heading into a nested group
6526 with possible change of options. Such groups are non-capturing and are
6527 not assertions of any kind. All we need to do is skip over the ':';
6528 the newoptions value is handled below. */
6529
6530 bravalue = OP_BRA;
6531 ptr++;
6532 } /* End of switch for character following (? */
6533 } /* End of (? handling */
6534
6535 /* Opening parenthesis not followed by '*' or '?'. If PCRE_NO_AUTO_CAPTURE
6536 is set, all unadorned brackets become non-capturing and behave like (?:...)
6537 brackets. */
6538
6539 else if ((options & PCRE_NO_AUTO_CAPTURE) != 0)
6540 {
6541 bravalue = OP_BRA;
6542 }
6543
6544 /* Else we have a capturing group. */
6545
6546 else
6547 {
6548 NUMBERED_GROUP:
6549 cd->bracount += 1;
6550 PUT2(code, 1+LINK_SIZE, cd->bracount);
6551 skipbytes = IMM2_SIZE;
6552 }
6553
6554 /* Process nested bracketed regex. Assertions used not to be repeatable,
6555 but this was changed for Perl compatibility, so all kinds can now be
6556 repeated. We copy code into a non-register variable (tempcode) in order to
6557 be able to pass its address because some compilers complain otherwise. */
6558
6559 previous = code; /* For handling repetition */
6560 *code = bravalue;
6561 tempcode = code;
6562 tempreqvary = cd->req_varyopt; /* Save value before bracket */
6563 tempbracount = cd->bracount; /* Save value before bracket */
6564 length_prevgroup = 0; /* Initialize for pre-compile phase */
6565
6566 if (!compile_regex(
6567 newoptions, /* The complete new option state */
6568 &tempcode, /* Where to put code (updated) */
6569 &ptr, /* Input pointer (updated) */
6570 errorcodeptr, /* Where to put an error message */
6571 (bravalue == OP_ASSERTBACK ||
6572 bravalue == OP_ASSERTBACK_NOT), /* TRUE if back assert */
6573 reset_bracount, /* True if (?| group */
6574 skipbytes, /* Skip over bracket number */
6575 cond_depth +
6576 ((bravalue == OP_COND)?1:0), /* Depth of condition subpatterns */
6577 &subfirstchar, /* For possible first char */
6578 &subreqchar, /* For possible last char */
6579 bcptr, /* Current branch chain */
6580 cd, /* Tables block */
6581 (lengthptr == NULL)? NULL : /* Actual compile phase */
6582 &length_prevgroup /* Pre-compile phase */
6583 ))
6584 goto FAILED;
6585
6586 /* If this was an atomic group and there are no capturing groups within it,
6587 generate OP_ONCE_NC instead of OP_ONCE. */
6588
6589 if (bravalue == OP_ONCE && cd->bracount <= tempbracount)
6590 *code = OP_ONCE_NC;
6591
6592 if (bravalue >= OP_ASSERT && bravalue <= OP_ASSERTBACK_NOT)
6593 cd->assert_depth -= 1;
6594
6595 /* At the end of compiling, code is still pointing to the start of the
6596 group, while tempcode has been updated to point past the end of the group.
6597 The pattern pointer (ptr) is on the bracket.
6598
6599 If this is a conditional bracket, check that there are no more than
6600 two branches in the group, or just one if it's a DEFINE group. We do this
6601 in the real compile phase, not in the pre-pass, where the whole group may
6602 not be available. */
6603
6604 if (bravalue == OP_COND && lengthptr == NULL)
6605 {
6606 pcre_uchar *tc = code;
6607 int condcount = 0;
6608
6609 do {
6610 condcount++;
6611 tc += GET(tc,1);
6612 }
6613 while (*tc != OP_KET);
6614
6615 /* A DEFINE group is never obeyed inline (the "condition" is always
6616 false). It must have only one branch. */
6617
6618 if (code[LINK_SIZE+1] == OP_DEF)
6619 {
6620 if (condcount > 1)
6621 {
6622 *errorcodeptr = ERR54;
6623 goto FAILED;
6624 }
6625 bravalue = OP_DEF; /* Just a flag to suppress char handling below */
6626 }
6627
6628 /* A "normal" conditional group. If there is just one branch, we must not
6629 make use of its firstchar or reqchar, because this is equivalent to an
6630 empty second branch. */
6631
6632 else
6633 {
6634 if (condcount > 2)
6635 {
6636 *errorcodeptr = ERR27;
6637 goto FAILED;
6638 }
6639 if (condcount == 1) subfirstchar = subreqchar = REQ_NONE;
6640 }
6641 }
6642
66