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

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

Parent Directory Parent Directory | Revision Log Revision Log


Revision 769 - (show annotations)
Sun Nov 27 17:40:33 2011 UTC (7 years, 9 months ago) by zherczeg
File MIME type: text/plain
File size: 254859 byte(s)
Update those opcode sizes whose depend on IMM2_SIZE
1 /*************************************************
2 * Perl-Compatible Regular Expressions *
3 *************************************************/
4
5 /* PCRE is a library of functions to support regular expressions whose syntax
6 and semantics are as close as possible to those of the Perl 5 language.
7
8 Written by Philip Hazel
9 Copyright (c) 1997-2011 University of Cambridge
10
11 -----------------------------------------------------------------------------
12 Redistribution and use in source and binary forms, with or without
13 modification, are permitted provided that the following conditions are met:
14
15 * Redistributions of source code must retain the above copyright notice,
16 this list of conditions and the following disclaimer.
17
18 * Redistributions in binary form must reproduce the above copyright
19 notice, this list of conditions and the following disclaimer in the
20 documentation and/or other materials provided with the distribution.
21
22 * Neither the name of the University of Cambridge nor the names of its
23 contributors may be used to endorse or promote products derived from
24 this software without specific prior written permission.
25
26 THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
27 AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28 IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29 ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE
30 LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
31 CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
32 SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
33 INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
34 CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
35 ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
36 POSSIBILITY OF SUCH DAMAGE.
37 -----------------------------------------------------------------------------
38 */
39
40
41 /* This module contains the external function pcre_compile(), along with
42 supporting internal functions that are not used by other modules. */
43
44
45 #ifdef HAVE_CONFIG_H
46 #include "config.h"
47 #endif
48
49 #define NLBLOCK cd /* Block containing newline information */
50 #define PSSTART start_pattern /* Field containing processed string start */
51 #define PSEND end_pattern /* Field containing processed string end */
52
53 #include "pcre_internal.h"
54
55
56 /* When PCRE_DEBUG is defined, we need the pcre_printint() function, which is
57 also used by pcretest. PCRE_DEBUG is not defined when building a production
58 library. */
59
60 #ifdef PCRE_DEBUG
61 #include "pcre_printint.src"
62 #endif
63
64
65 /* Macro for setting individual bits in class bitmaps. */
66
67 #define SETBIT(a,b) a[b/8] |= (1 << (b%8))
68
69 /* Maximum length value to check against when making sure that the integer that
70 holds the compiled pattern length does not overflow. We make it a bit less than
71 INT_MAX to allow for adding in group terminating bytes, so that we don't have
72 to check them every time. */
73
74 #define OFLOW_MAX (INT_MAX - 20)
75
76
77 /*************************************************
78 * Code parameters and static tables *
79 *************************************************/
80
81 /* This value specifies the size of stack workspace that is used during the
82 first pre-compile phase that determines how much memory is required. The regex
83 is partly compiled into this space, but the compiled parts are discarded as
84 soon as they can be, so that hopefully there will never be an overrun. The code
85 does, however, check for an overrun. The largest amount I've seen used is 218,
86 so this number is very generous.
87
88 The same workspace is used during the second, actual compile phase for
89 remembering forward references to groups so that they can be filled in at the
90 end. Each entry in this list occupies LINK_SIZE bytes, so even when LINK_SIZE
91 is 4 there is plenty of room. */
92
93 #define COMPILE_WORK_SIZE (4096)
94
95 /* The overrun tests check for a slightly smaller size so that they detect the
96 overrun before it actually does run off the end of the data block. */
97
98 #define WORK_SIZE_CHECK (COMPILE_WORK_SIZE - 100)
99
100
101 /* Table for handling escaped characters in the range '0'-'z'. Positive returns
102 are simple data values; negative values are for special things like \d and so
103 on. Zero means further processing is needed (for things like \x), or the escape
104 is invalid. */
105
106 #ifndef EBCDIC
107
108 /* This is the "normal" table for ASCII systems or for EBCDIC systems running
109 in UTF-8 mode. */
110
111 static const short int escapes[] = {
112 0, 0,
113 0, 0,
114 0, 0,
115 0, 0,
116 0, 0,
117 CHAR_COLON, CHAR_SEMICOLON,
118 CHAR_LESS_THAN_SIGN, CHAR_EQUALS_SIGN,
119 CHAR_GREATER_THAN_SIGN, CHAR_QUESTION_MARK,
120 CHAR_COMMERCIAL_AT, -ESC_A,
121 -ESC_B, -ESC_C,
122 -ESC_D, -ESC_E,
123 0, -ESC_G,
124 -ESC_H, 0,
125 0, -ESC_K,
126 0, 0,
127 -ESC_N, 0,
128 -ESC_P, -ESC_Q,
129 -ESC_R, -ESC_S,
130 0, 0,
131 -ESC_V, -ESC_W,
132 -ESC_X, 0,
133 -ESC_Z, CHAR_LEFT_SQUARE_BRACKET,
134 CHAR_BACKSLASH, CHAR_RIGHT_SQUARE_BRACKET,
135 CHAR_CIRCUMFLEX_ACCENT, CHAR_UNDERSCORE,
136 CHAR_GRAVE_ACCENT, 7,
137 -ESC_b, 0,
138 -ESC_d, ESC_e,
139 ESC_f, 0,
140 -ESC_h, 0,
141 0, -ESC_k,
142 0, 0,
143 ESC_n, 0,
144 -ESC_p, 0,
145 ESC_r, -ESC_s,
146 ESC_tee, 0,
147 -ESC_v, -ESC_w,
148 0, 0,
149 -ESC_z
150 };
151
152 #else
153
154 /* This is the "abnormal" table for EBCDIC systems without UTF-8 support. */
155
156 static const short int escapes[] = {
157 /* 48 */ 0, 0, 0, '.', '<', '(', '+', '|',
158 /* 50 */ '&', 0, 0, 0, 0, 0, 0, 0,
159 /* 58 */ 0, 0, '!', '$', '*', ')', ';', '~',
160 /* 60 */ '-', '/', 0, 0, 0, 0, 0, 0,
161 /* 68 */ 0, 0, '|', ',', '%', '_', '>', '?',
162 /* 70 */ 0, 0, 0, 0, 0, 0, 0, 0,
163 /* 78 */ 0, '`', ':', '#', '@', '\'', '=', '"',
164 /* 80 */ 0, 7, -ESC_b, 0, -ESC_d, ESC_e, ESC_f, 0,
165 /* 88 */-ESC_h, 0, 0, '{', 0, 0, 0, 0,
166 /* 90 */ 0, 0, -ESC_k, 'l', 0, ESC_n, 0, -ESC_p,
167 /* 98 */ 0, ESC_r, 0, '}', 0, 0, 0, 0,
168 /* A0 */ 0, '~', -ESC_s, ESC_tee, 0,-ESC_v, -ESC_w, 0,
169 /* A8 */ 0,-ESC_z, 0, 0, 0, '[', 0, 0,
170 /* B0 */ 0, 0, 0, 0, 0, 0, 0, 0,
171 /* B8 */ 0, 0, 0, 0, 0, ']', '=', '-',
172 /* C0 */ '{',-ESC_A, -ESC_B, -ESC_C, -ESC_D,-ESC_E, 0, -ESC_G,
173 /* C8 */-ESC_H, 0, 0, 0, 0, 0, 0, 0,
174 /* D0 */ '}', 0, -ESC_K, 0, 0,-ESC_N, 0, -ESC_P,
175 /* D8 */-ESC_Q,-ESC_R, 0, 0, 0, 0, 0, 0,
176 /* E0 */ '\\', 0, -ESC_S, 0, 0,-ESC_V, -ESC_W, -ESC_X,
177 /* E8 */ 0,-ESC_Z, 0, 0, 0, 0, 0, 0,
178 /* F0 */ 0, 0, 0, 0, 0, 0, 0, 0,
179 /* F8 */ 0, 0, 0, 0, 0, 0, 0, 0
180 };
181 #endif
182
183
184 /* Table of special "verbs" like (*PRUNE). This is a short table, so it is
185 searched linearly. Put all the names into a single string, in order to reduce
186 the number of relocations when a shared library is dynamically linked. The
187 string is built from string macros so that it works in UTF-8 mode on EBCDIC
188 platforms. */
189
190 typedef struct verbitem {
191 int len; /* Length of verb name */
192 int op; /* Op when no arg, or -1 if arg mandatory */
193 int op_arg; /* Op when arg present, or -1 if not allowed */
194 } verbitem;
195
196 static const char verbnames[] =
197 "\0" /* Empty name is a shorthand for MARK */
198 STRING_MARK0
199 STRING_ACCEPT0
200 STRING_COMMIT0
201 STRING_F0
202 STRING_FAIL0
203 STRING_PRUNE0
204 STRING_SKIP0
205 STRING_THEN;
206
207 static const verbitem verbs[] = {
208 { 0, -1, OP_MARK },
209 { 4, -1, OP_MARK },
210 { 6, OP_ACCEPT, -1 },
211 { 6, OP_COMMIT, -1 },
212 { 1, OP_FAIL, -1 },
213 { 4, OP_FAIL, -1 },
214 { 5, OP_PRUNE, OP_PRUNE_ARG },
215 { 4, OP_SKIP, OP_SKIP_ARG },
216 { 4, OP_THEN, OP_THEN_ARG }
217 };
218
219 static const int verbcount = sizeof(verbs)/sizeof(verbitem);
220
221
222 /* Tables of names of POSIX character classes and their lengths. The names are
223 now all in a single string, to reduce the number of relocations when a shared
224 library is dynamically loaded. The list of lengths is terminated by a zero
225 length entry. The first three must be alpha, lower, upper, as this is assumed
226 for handling case independence. */
227
228 static const char posix_names[] =
229 STRING_alpha0 STRING_lower0 STRING_upper0 STRING_alnum0
230 STRING_ascii0 STRING_blank0 STRING_cntrl0 STRING_digit0
231 STRING_graph0 STRING_print0 STRING_punct0 STRING_space0
232 STRING_word0 STRING_xdigit;
233
234 static const pcre_uint8 posix_name_lengths[] = {
235 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 4, 6, 0 };
236
237 /* Table of class bit maps for each POSIX class. Each class is formed from a
238 base map, with an optional addition or removal of another map. Then, for some
239 classes, there is some additional tweaking: for [:blank:] the vertical space
240 characters are removed, and for [:alpha:] and [:alnum:] the underscore
241 character is removed. The triples in the table consist of the base map offset,
242 second map offset or -1 if no second map, and a non-negative value for map
243 addition or a negative value for map subtraction (if there are two maps). The
244 absolute value of the third field has these meanings: 0 => no tweaking, 1 =>
245 remove vertical space characters, 2 => remove underscore. */
246
247 static const int posix_class_maps[] = {
248 cbit_word, cbit_digit, -2, /* alpha */
249 cbit_lower, -1, 0, /* lower */
250 cbit_upper, -1, 0, /* upper */
251 cbit_word, -1, 2, /* alnum - word without underscore */
252 cbit_print, cbit_cntrl, 0, /* ascii */
253 cbit_space, -1, 1, /* blank - a GNU extension */
254 cbit_cntrl, -1, 0, /* cntrl */
255 cbit_digit, -1, 0, /* digit */
256 cbit_graph, -1, 0, /* graph */
257 cbit_print, -1, 0, /* print */
258 cbit_punct, -1, 0, /* punct */
259 cbit_space, -1, 0, /* space */
260 cbit_word, -1, 0, /* word - a Perl extension */
261 cbit_xdigit,-1, 0 /* xdigit */
262 };
263
264 /* Table of substitutes for \d etc when PCRE_UCP is set. The POSIX class
265 substitutes must be in the order of the names, defined above, and there are
266 both positive and negative cases. NULL means no substitute. */
267
268 #ifdef SUPPORT_UCP
269 static const pcre_uchar string_PNd[] = {
270 CHAR_BACKSLASH, CHAR_P, CHAR_LEFT_CURLY_BRACKET,
271 CHAR_N, CHAR_d, CHAR_RIGHT_CURLY_BRACKET, '\0' };
272 static const pcre_uchar string_pNd[] = {
273 CHAR_BACKSLASH, CHAR_p, CHAR_LEFT_CURLY_BRACKET,
274 CHAR_N, CHAR_d, CHAR_RIGHT_CURLY_BRACKET, '\0' };
275 static const pcre_uchar string_PXsp[] = {
276 CHAR_BACKSLASH, CHAR_P, CHAR_LEFT_CURLY_BRACKET,
277 CHAR_X, CHAR_s, CHAR_p, CHAR_RIGHT_CURLY_BRACKET, '\0' };
278 static const pcre_uchar string_pXsp[] = {
279 CHAR_BACKSLASH, CHAR_p, CHAR_LEFT_CURLY_BRACKET,
280 CHAR_X, CHAR_s, CHAR_p, CHAR_RIGHT_CURLY_BRACKET, '\0' };
281 static const pcre_uchar string_PXwd[] = {
282 CHAR_BACKSLASH, CHAR_P, CHAR_LEFT_CURLY_BRACKET,
283 CHAR_X, CHAR_w, CHAR_d, CHAR_RIGHT_CURLY_BRACKET, '\0' };
284 static const pcre_uchar string_pXwd[] = {
285 CHAR_BACKSLASH, CHAR_p, CHAR_LEFT_CURLY_BRACKET,
286 CHAR_X, CHAR_w, CHAR_d, CHAR_RIGHT_CURLY_BRACKET, '\0' };
287
288 static const pcre_uchar *substitutes[] = {
289 string_PNd, /* \D */
290 string_pNd, /* \d */
291 string_PXsp, /* \S */ /* NOTE: Xsp is Perl space */
292 string_pXsp, /* \s */
293 string_PXwd, /* \W */
294 string_pXwd /* \w */
295 };
296
297 static const pcre_uchar string_pL[] = {
298 CHAR_BACKSLASH, CHAR_p, CHAR_LEFT_CURLY_BRACKET,
299 CHAR_L, CHAR_RIGHT_CURLY_BRACKET, '\0' };
300 static const pcre_uchar string_pLl[] = {
301 CHAR_BACKSLASH, CHAR_p, CHAR_LEFT_CURLY_BRACKET,
302 CHAR_L, CHAR_l, CHAR_RIGHT_CURLY_BRACKET, '\0' };
303 static const pcre_uchar string_pLu[] = {
304 CHAR_BACKSLASH, CHAR_p, CHAR_LEFT_CURLY_BRACKET,
305 CHAR_L, CHAR_u, CHAR_RIGHT_CURLY_BRACKET, '\0' };
306 static const pcre_uchar string_pXan[] = {
307 CHAR_BACKSLASH, CHAR_p, CHAR_LEFT_CURLY_BRACKET,
308 CHAR_X, CHAR_a, CHAR_n, CHAR_RIGHT_CURLY_BRACKET, '\0' };
309 static const pcre_uchar string_h[] = {
310 CHAR_BACKSLASH, CHAR_h, '\0' };
311 static const pcre_uchar string_pXps[] = {
312 CHAR_BACKSLASH, CHAR_p, CHAR_LEFT_CURLY_BRACKET,
313 CHAR_X, CHAR_p, CHAR_s, CHAR_RIGHT_CURLY_BRACKET, '\0' };
314 static const pcre_uchar string_PL[] = {
315 CHAR_BACKSLASH, CHAR_P, CHAR_LEFT_CURLY_BRACKET,
316 CHAR_L, CHAR_RIGHT_CURLY_BRACKET, '\0' };
317 static const pcre_uchar string_PLl[] = {
318 CHAR_BACKSLASH, CHAR_P, CHAR_LEFT_CURLY_BRACKET,
319 CHAR_L, CHAR_l, CHAR_RIGHT_CURLY_BRACKET, '\0' };
320 static const pcre_uchar string_PLu[] = {
321 CHAR_BACKSLASH, CHAR_P, CHAR_LEFT_CURLY_BRACKET,
322 CHAR_L, CHAR_u, CHAR_RIGHT_CURLY_BRACKET, '\0' };
323 static const pcre_uchar string_PXan[] = {
324 CHAR_BACKSLASH, CHAR_P, CHAR_LEFT_CURLY_BRACKET,
325 CHAR_X, CHAR_a, CHAR_n, CHAR_RIGHT_CURLY_BRACKET, '\0' };
326 static const pcre_uchar string_H[] = {
327 CHAR_BACKSLASH, CHAR_H, '\0' };
328 static const pcre_uchar string_PXps[] = {
329 CHAR_BACKSLASH, CHAR_P, CHAR_LEFT_CURLY_BRACKET,
330 CHAR_X, CHAR_p, CHAR_s, CHAR_RIGHT_CURLY_BRACKET, '\0' };
331
332 static const pcre_uchar *posix_substitutes[] = {
333 string_pL, /* alpha */
334 string_pLl, /* lower */
335 string_pLu, /* upper */
336 string_pXan, /* alnum */
337 NULL, /* ascii */
338 string_h, /* blank */
339 NULL, /* cntrl */
340 string_pNd, /* digit */
341 NULL, /* graph */
342 NULL, /* print */
343 NULL, /* punct */
344 string_pXps, /* space */ /* NOTE: Xps is POSIX space */
345 string_pXwd, /* word */
346 NULL, /* xdigit */
347 /* Negated cases */
348 string_PL, /* ^alpha */
349 string_PLl, /* ^lower */
350 string_PLu, /* ^upper */
351 string_PXan, /* ^alnum */
352 NULL, /* ^ascii */
353 string_H, /* ^blank */
354 NULL, /* ^cntrl */
355 string_PNd, /* ^digit */
356 NULL, /* ^graph */
357 NULL, /* ^print */
358 NULL, /* ^punct */
359 string_PXps, /* ^space */ /* NOTE: Xps is POSIX space */
360 string_PXwd, /* ^word */
361 NULL /* ^xdigit */
362 };
363 #define POSIX_SUBSIZE (sizeof(posix_substitutes) / sizeof(pcre_uchar *))
364 #endif
365
366 #define STRING(a) # a
367 #define XSTRING(s) STRING(s)
368
369 /* The texts of compile-time error messages. These are "char *" because they
370 are passed to the outside world. Do not ever re-use any error number, because
371 they are documented. Always add a new error instead. Messages marked DEAD below
372 are no longer used. This used to be a table of strings, but in order to reduce
373 the number of relocations needed when a shared library is loaded dynamically,
374 it is now one long string. We cannot use a table of offsets, because the
375 lengths of inserts such as XSTRING(MAX_NAME_SIZE) are not known. Instead, we
376 simply count through to the one we want - this isn't a performance issue
377 because these strings are used only when there is a compilation error.
378
379 Each substring ends with \0 to insert a null character. This includes the final
380 substring, so that the whole string ends with \0\0, which can be detected when
381 counting through. */
382
383 static const char error_texts[] =
384 "no error\0"
385 "\\ at end of pattern\0"
386 "\\c at end of pattern\0"
387 "unrecognized character follows \\\0"
388 "numbers out of order in {} quantifier\0"
389 /* 5 */
390 "number too big in {} quantifier\0"
391 "missing terminating ] for character class\0"
392 "invalid escape sequence in character class\0"
393 "range out of order in character class\0"
394 "nothing to repeat\0"
395 /* 10 */
396 "operand of unlimited repeat could match the empty string\0" /** DEAD **/
397 "internal error: unexpected repeat\0"
398 "unrecognized character after (? or (?-\0"
399 "POSIX named classes are supported only within a class\0"
400 "missing )\0"
401 /* 15 */
402 "reference to non-existent subpattern\0"
403 "erroffset passed as NULL\0"
404 "unknown option bit(s) set\0"
405 "missing ) after comment\0"
406 "parentheses nested too deeply\0" /** DEAD **/
407 /* 20 */
408 "regular expression is too large\0"
409 "failed to get memory\0"
410 "unmatched parentheses\0"
411 "internal error: code overflow\0"
412 "unrecognized character after (?<\0"
413 /* 25 */
414 "lookbehind assertion is not fixed length\0"
415 "malformed number or name after (?(\0"
416 "conditional group contains more than two branches\0"
417 "assertion expected after (?(\0"
418 "(?R or (?[+-]digits must be followed by )\0"
419 /* 30 */
420 "unknown POSIX class name\0"
421 "POSIX collating elements are not supported\0"
422 "this version of PCRE is not compiled with PCRE_UTF8 support\0"
423 "spare error\0" /** DEAD **/
424 "character value in \\x{...} sequence is too large\0"
425 /* 35 */
426 "invalid condition (?(0)\0"
427 "\\C not allowed in lookbehind assertion\0"
428 "PCRE does not support \\L, \\l, \\N{name}, \\U, or \\u\0"
429 "number after (?C is > 255\0"
430 "closing ) for (?C expected\0"
431 /* 40 */
432 "recursive call could loop indefinitely\0"
433 "unrecognized character after (?P\0"
434 "syntax error in subpattern name (missing terminator)\0"
435 "two named subpatterns have the same name\0"
436 "invalid UTF-8 string\0"
437 /* 45 */
438 "support for \\P, \\p, and \\X has not been compiled\0"
439 "malformed \\P or \\p sequence\0"
440 "unknown property name after \\P or \\p\0"
441 "subpattern name is too long (maximum " XSTRING(MAX_NAME_SIZE) " characters)\0"
442 "too many named subpatterns (maximum " XSTRING(MAX_NAME_COUNT) ")\0"
443 /* 50 */
444 "repeated subpattern is too long\0" /** DEAD **/
445 "octal value is greater than \\377 (not in UTF-8 mode)\0"
446 "internal error: overran compiling workspace\0"
447 "internal error: previously-checked referenced subpattern not found\0"
448 "DEFINE group contains more than one branch\0"
449 /* 55 */
450 "repeating a DEFINE group is not allowed\0" /** DEAD **/
451 "inconsistent NEWLINE options\0"
452 "\\g is not followed by a braced, angle-bracketed, or quoted name/number or by a plain number\0"
453 "a numbered reference must not be zero\0"
454 "an argument is not allowed for (*ACCEPT), (*FAIL), or (*COMMIT)\0"
455 /* 60 */
456 "(*VERB) not recognized\0"
457 "number is too big\0"
458 "subpattern name expected\0"
459 "digit expected after (?+\0"
460 "] is an invalid data character in JavaScript compatibility mode\0"
461 /* 65 */
462 "different names for subpatterns of the same number are not allowed\0"
463 "(*MARK) must have an argument\0"
464 "this version of PCRE is not compiled with PCRE_UCP support\0"
465 "\\c must be followed by an ASCII character\0"
466 "\\k is not followed by a braced, angle-bracketed, or quoted name\0"
467 /* 70 */
468 "internal error: unknown opcode in find_fixedlength()\0"
469 ;
470
471 /* Table to identify digits and hex digits. This is used when compiling
472 patterns. Note that the tables in chartables are dependent on the locale, and
473 may mark arbitrary characters as digits - but the PCRE compiling code expects
474 to handle only 0-9, a-z, and A-Z as digits when compiling. That is why we have
475 a private table here. It costs 256 bytes, but it is a lot faster than doing
476 character value tests (at least in some simple cases I timed), and in some
477 applications one wants PCRE to compile efficiently as well as match
478 efficiently.
479
480 For convenience, we use the same bit definitions as in chartables:
481
482 0x04 decimal digit
483 0x08 hexadecimal digit
484
485 Then we can use ctype_digit and ctype_xdigit in the code. */
486
487 #ifndef EBCDIC
488
489 /* This is the "normal" case, for ASCII systems, and EBCDIC systems running in
490 UTF-8 mode. */
491
492 static const unsigned char digitab[] =
493 {
494 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 0- 7 */
495 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 8- 15 */
496 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 16- 23 */
497 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 24- 31 */
498 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - ' */
499 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* ( - / */
500 0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c, /* 0 - 7 */
501 0x0c,0x0c,0x00,0x00,0x00,0x00,0x00,0x00, /* 8 - ? */
502 0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* @ - G */
503 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* H - O */
504 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* P - W */
505 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* X - _ */
506 0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* ` - g */
507 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* h - o */
508 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* p - w */
509 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* x -127 */
510 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 128-135 */
511 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 136-143 */
512 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 144-151 */
513 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 152-159 */
514 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 160-167 */
515 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 168-175 */
516 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 176-183 */
517 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 184-191 */
518 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 192-199 */
519 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 200-207 */
520 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 208-215 */
521 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 216-223 */
522 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 224-231 */
523 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 232-239 */
524 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 240-247 */
525 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00};/* 248-255 */
526
527 #else
528
529 /* This is the "abnormal" case, for EBCDIC systems not running in UTF-8 mode. */
530
531 static const unsigned char digitab[] =
532 {
533 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 0- 7 0 */
534 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 8- 15 */
535 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 16- 23 10 */
536 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 24- 31 */
537 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 32- 39 20 */
538 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 40- 47 */
539 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 48- 55 30 */
540 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 56- 63 */
541 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - 71 40 */
542 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 72- | */
543 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* & - 87 50 */
544 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 88- 95 */
545 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - -103 60 */
546 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 104- ? */
547 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 112-119 70 */
548 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 120- " */
549 0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* 128- g 80 */
550 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* h -143 */
551 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 144- p 90 */
552 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* q -159 */
553 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 160- x A0 */
554 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* y -175 */
555 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* ^ -183 B0 */
556 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 184-191 */
557 0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* { - G C0 */
558 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* H -207 */
559 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* } - P D0 */
560 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* Q -223 */
561 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* \ - X E0 */
562 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* Y -239 */
563 0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c, /* 0 - 7 F0 */
564 0x0c,0x0c,0x00,0x00,0x00,0x00,0x00,0x00};/* 8 -255 */
565
566 static const unsigned char ebcdic_chartab[] = { /* chartable partial dup */
567 0x80,0x00,0x00,0x00,0x00,0x01,0x00,0x00, /* 0- 7 */
568 0x00,0x00,0x00,0x00,0x01,0x01,0x00,0x00, /* 8- 15 */
569 0x00,0x00,0x00,0x00,0x00,0x01,0x00,0x00, /* 16- 23 */
570 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 24- 31 */
571 0x00,0x00,0x00,0x00,0x00,0x01,0x00,0x00, /* 32- 39 */
572 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 40- 47 */
573 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 48- 55 */
574 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 56- 63 */
575 0x01,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - 71 */
576 0x00,0x00,0x00,0x80,0x00,0x80,0x80,0x80, /* 72- | */
577 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* & - 87 */
578 0x00,0x00,0x00,0x80,0x80,0x80,0x00,0x00, /* 88- 95 */
579 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - -103 */
580 0x00,0x00,0x00,0x00,0x00,0x10,0x00,0x80, /* 104- ? */
581 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 112-119 */
582 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 120- " */
583 0x00,0x1a,0x1a,0x1a,0x1a,0x1a,0x1a,0x12, /* 128- g */
584 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* h -143 */
585 0x00,0x12,0x12,0x12,0x12,0x12,0x12,0x12, /* 144- p */
586 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* q -159 */
587 0x00,0x00,0x12,0x12,0x12,0x12,0x12,0x12, /* 160- x */
588 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* y -175 */
589 0x80,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* ^ -183 */
590 0x00,0x00,0x80,0x00,0x00,0x00,0x00,0x00, /* 184-191 */
591 0x80,0x1a,0x1a,0x1a,0x1a,0x1a,0x1a,0x12, /* { - G */
592 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* H -207 */
593 0x00,0x12,0x12,0x12,0x12,0x12,0x12,0x12, /* } - P */
594 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* Q -223 */
595 0x00,0x00,0x12,0x12,0x12,0x12,0x12,0x12, /* \ - X */
596 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* Y -239 */
597 0x1c,0x1c,0x1c,0x1c,0x1c,0x1c,0x1c,0x1c, /* 0 - 7 */
598 0x1c,0x1c,0x00,0x00,0x00,0x00,0x00,0x00};/* 8 -255 */
599 #endif
600
601
602 /* Definition to allow mutual recursion */
603
604 static BOOL
605 compile_regex(int, pcre_uchar **, const pcre_uchar **, int *, BOOL, BOOL, int, int,
606 int *, int *, branch_chain *, compile_data *, int *);
607
608
609
610 /*************************************************
611 * Find an error text *
612 *************************************************/
613
614 /* The error texts are now all in one long string, to save on relocations. As
615 some of the text is of unknown length, we can't use a table of offsets.
616 Instead, just count through the strings. This is not a performance issue
617 because it happens only when there has been a compilation error.
618
619 Argument: the error number
620 Returns: pointer to the error string
621 */
622
623 static const char *
624 find_error_text(int n)
625 {
626 const char *s = error_texts;
627 for (; n > 0; n--)
628 {
629 while (*s++ != 0) {};
630 if (*s == 0) return "Error text not found (please report)";
631 }
632 return s;
633 }
634
635
636 /*************************************************
637 * Check for counted repeat *
638 *************************************************/
639
640 /* This function is called when a '{' is encountered in a place where it might
641 start a quantifier. It looks ahead to see if it really is a quantifier or not.
642 It is only a quantifier if it is one of the forms {ddd} {ddd,} or {ddd,ddd}
643 where the ddds are digits.
644
645 Arguments:
646 p pointer to the first char after '{'
647
648 Returns: TRUE or FALSE
649 */
650
651 static BOOL
652 is_counted_repeat(const pcre_uchar *p)
653 {
654 if ((digitab[*p++] & ctype_digit) == 0) return FALSE;
655 while ((digitab[*p] & ctype_digit) != 0) p++;
656 if (*p == CHAR_RIGHT_CURLY_BRACKET) return TRUE;
657
658 if (*p++ != CHAR_COMMA) return FALSE;
659 if (*p == CHAR_RIGHT_CURLY_BRACKET) return TRUE;
660
661 if ((digitab[*p++] & ctype_digit) == 0) return FALSE;
662 while ((digitab[*p] & ctype_digit) != 0) p++;
663
664 return (*p == CHAR_RIGHT_CURLY_BRACKET);
665 }
666
667
668
669 /*************************************************
670 * Handle escapes *
671 *************************************************/
672
673 /* This function is called when a \ has been encountered. It either returns a
674 positive value for a simple escape such as \n, or a negative value which
675 encodes one of the more complicated things such as \d. A backreference to group
676 n is returned as -(ESC_REF + n); ESC_REF is the highest ESC_xxx macro. When
677 UTF-8 is enabled, a positive value greater than 255 may be returned. On entry,
678 ptr is pointing at the \. On exit, it is on the final character of the escape
679 sequence.
680
681 Arguments:
682 ptrptr points to the pattern position pointer
683 errorcodeptr points to the errorcode variable
684 bracount number of previous extracting brackets
685 options the options bits
686 isclass TRUE if inside a character class
687
688 Returns: zero or positive => a data character
689 negative => a special escape sequence
690 on error, errorcodeptr is set
691 */
692
693 static int
694 check_escape(const pcre_uchar **ptrptr, int *errorcodeptr, int bracount,
695 int options, BOOL isclass)
696 {
697 BOOL utf8 = (options & PCRE_UTF8) != 0;
698 const pcre_uchar *ptr = *ptrptr + 1;
699 int c, i;
700
701 GETCHARINCTEST(c, ptr); /* Get character value, increment pointer */
702 ptr--; /* Set pointer back to the last byte */
703
704 /* If backslash is at the end of the pattern, it's an error. */
705
706 if (c == 0) *errorcodeptr = ERR1;
707
708 /* Non-alphanumerics are literals. For digits or letters, do an initial lookup
709 in a table. A non-zero result is something that can be returned immediately.
710 Otherwise further processing may be required. */
711
712 #ifndef EBCDIC /* ASCII/UTF-8 coding */
713 else if (c < CHAR_0 || c > CHAR_z) {} /* Not alphanumeric */
714 else if ((i = escapes[c - CHAR_0]) != 0) c = i;
715
716 #else /* EBCDIC coding */
717 else if (c < 'a' || (ebcdic_chartab[c] & 0x0E) == 0) {} /* Not alphanumeric */
718 else if ((i = escapes[c - 0x48]) != 0) c = i;
719 #endif
720
721 /* Escapes that need further processing, or are illegal. */
722
723 else
724 {
725 const pcre_uchar *oldptr;
726 BOOL braced, negated;
727
728 switch (c)
729 {
730 /* A number of Perl escapes are not handled by PCRE. We give an explicit
731 error. */
732
733 case CHAR_l:
734 case CHAR_L:
735 *errorcodeptr = ERR37;
736 break;
737
738 case CHAR_u:
739 if ((options & PCRE_JAVASCRIPT_COMPAT) != 0)
740 {
741 /* In JavaScript, \u must be followed by four hexadecimal numbers.
742 Otherwise it is a lowercase u letter. */
743 if ((digitab[ptr[1]] & ctype_xdigit) != 0 && (digitab[ptr[2]] & ctype_xdigit) != 0
744 && (digitab[ptr[3]] & ctype_xdigit) != 0 && (digitab[ptr[4]] & ctype_xdigit) != 0)
745 {
746 c = 0;
747 for (i = 0; i < 4; ++i)
748 {
749 register int cc = *(++ptr);
750 #ifndef EBCDIC /* ASCII/UTF-8 coding */
751 if (cc >= CHAR_a) cc -= 32; /* Convert to upper case */
752 c = (c << 4) + cc - ((cc < CHAR_A)? CHAR_0 : (CHAR_A - 10));
753 #else /* EBCDIC coding */
754 if (cc >= CHAR_a && cc <= CHAR_z) cc += 64; /* Convert to upper case */
755 c = (c << 4) + cc - ((cc >= CHAR_0)? CHAR_0 : (CHAR_A - 10));
756 #endif
757 }
758 }
759 }
760 else
761 *errorcodeptr = ERR37;
762 break;
763
764 case CHAR_U:
765 /* In JavaScript, \U is an uppercase U letter. */
766 if ((options & PCRE_JAVASCRIPT_COMPAT) == 0) *errorcodeptr = ERR37;
767 break;
768
769 /* In a character class, \g is just a literal "g". Outside a character
770 class, \g must be followed by one of a number of specific things:
771
772 (1) A number, either plain or braced. If positive, it is an absolute
773 backreference. If negative, it is a relative backreference. This is a Perl
774 5.10 feature.
775
776 (2) Perl 5.10 also supports \g{name} as a reference to a named group. This
777 is part of Perl's movement towards a unified syntax for back references. As
778 this is synonymous with \k{name}, we fudge it up by pretending it really
779 was \k.
780
781 (3) For Oniguruma compatibility we also support \g followed by a name or a
782 number either in angle brackets or in single quotes. However, these are
783 (possibly recursive) subroutine calls, _not_ backreferences. Just return
784 the -ESC_g code (cf \k). */
785
786 case CHAR_g:
787 if (isclass) break;
788 if (ptr[1] == CHAR_LESS_THAN_SIGN || ptr[1] == CHAR_APOSTROPHE)
789 {
790 c = -ESC_g;
791 break;
792 }
793
794 /* Handle the Perl-compatible cases */
795
796 if (ptr[1] == CHAR_LEFT_CURLY_BRACKET)
797 {
798 const pcre_uchar *p;
799 for (p = ptr+2; *p != 0 && *p != CHAR_RIGHT_CURLY_BRACKET; p++)
800 if (*p != CHAR_MINUS && (digitab[*p] & ctype_digit) == 0) break;
801 if (*p != 0 && *p != CHAR_RIGHT_CURLY_BRACKET)
802 {
803 c = -ESC_k;
804 break;
805 }
806 braced = TRUE;
807 ptr++;
808 }
809 else braced = FALSE;
810
811 if (ptr[1] == CHAR_MINUS)
812 {
813 negated = TRUE;
814 ptr++;
815 }
816 else negated = FALSE;
817
818 c = 0;
819 while ((digitab[ptr[1]] & ctype_digit) != 0)
820 c = c * 10 + *(++ptr) - CHAR_0;
821
822 if (c < 0) /* Integer overflow */
823 {
824 *errorcodeptr = ERR61;
825 break;
826 }
827
828 if (braced && *(++ptr) != CHAR_RIGHT_CURLY_BRACKET)
829 {
830 *errorcodeptr = ERR57;
831 break;
832 }
833
834 if (c == 0)
835 {
836 *errorcodeptr = ERR58;
837 break;
838 }
839
840 if (negated)
841 {
842 if (c > bracount)
843 {
844 *errorcodeptr = ERR15;
845 break;
846 }
847 c = bracount - (c - 1);
848 }
849
850 c = -(ESC_REF + c);
851 break;
852
853 /* The handling of escape sequences consisting of a string of digits
854 starting with one that is not zero is not straightforward. By experiment,
855 the way Perl works seems to be as follows:
856
857 Outside a character class, the digits are read as a decimal number. If the
858 number is less than 10, or if there are that many previous extracting
859 left brackets, then it is a back reference. Otherwise, up to three octal
860 digits are read to form an escaped byte. Thus \123 is likely to be octal
861 123 (cf \0123, which is octal 012 followed by the literal 3). If the octal
862 value is greater than 377, the least significant 8 bits are taken. Inside a
863 character class, \ followed by a digit is always an octal number. */
864
865 case CHAR_1: case CHAR_2: case CHAR_3: case CHAR_4: case CHAR_5:
866 case CHAR_6: case CHAR_7: case CHAR_8: case CHAR_9:
867
868 if (!isclass)
869 {
870 oldptr = ptr;
871 c -= CHAR_0;
872 while ((digitab[ptr[1]] & ctype_digit) != 0)
873 c = c * 10 + *(++ptr) - CHAR_0;
874 if (c < 0) /* Integer overflow */
875 {
876 *errorcodeptr = ERR61;
877 break;
878 }
879 if (c < 10 || c <= bracount)
880 {
881 c = -(ESC_REF + c);
882 break;
883 }
884 ptr = oldptr; /* Put the pointer back and fall through */
885 }
886
887 /* Handle an octal number following \. If the first digit is 8 or 9, Perl
888 generates a binary zero byte and treats the digit as a following literal.
889 Thus we have to pull back the pointer by one. */
890
891 if ((c = *ptr) >= CHAR_8)
892 {
893 ptr--;
894 c = 0;
895 break;
896 }
897
898 /* \0 always starts an octal number, but we may drop through to here with a
899 larger first octal digit. The original code used just to take the least
900 significant 8 bits of octal numbers (I think this is what early Perls used
901 to do). Nowadays we allow for larger numbers in UTF-8 mode, but no more
902 than 3 octal digits. */
903
904 case CHAR_0:
905 c -= CHAR_0;
906 while(i++ < 2 && ptr[1] >= CHAR_0 && ptr[1] <= CHAR_7)
907 c = c * 8 + *(++ptr) - CHAR_0;
908 if (!utf8 && c > 255) *errorcodeptr = ERR51;
909 break;
910
911 /* \x is complicated. \x{ddd} is a character number which can be greater
912 than 0xff in utf8 mode, but only if the ddd are hex digits. If not, { is
913 treated as a data character. */
914
915 case CHAR_x:
916 if ((options & PCRE_JAVASCRIPT_COMPAT) != 0)
917 {
918 /* In JavaScript, \x must be followed by two hexadecimal numbers.
919 Otherwise it is a lowercase x letter. */
920 if ((digitab[ptr[1]] & ctype_xdigit) != 0 && (digitab[ptr[2]] & ctype_xdigit) != 0)
921 {
922 c = 0;
923 for (i = 0; i < 2; ++i)
924 {
925 register int cc = *(++ptr);
926 #ifndef EBCDIC /* ASCII/UTF-8 coding */
927 if (cc >= CHAR_a) cc -= 32; /* Convert to upper case */
928 c = (c << 4) + cc - ((cc < CHAR_A)? CHAR_0 : (CHAR_A - 10));
929 #else /* EBCDIC coding */
930 if (cc >= CHAR_a && cc <= CHAR_z) cc += 64; /* Convert to upper case */
931 c = (c << 4) + cc - ((cc >= CHAR_0)? CHAR_0 : (CHAR_A - 10));
932 #endif
933 }
934 }
935 break;
936 }
937
938 if (ptr[1] == CHAR_LEFT_CURLY_BRACKET)
939 {
940 const pcre_uchar *pt = ptr + 2;
941 int count = 0;
942
943 c = 0;
944 while ((digitab[*pt] & ctype_xdigit) != 0)
945 {
946 register int cc = *pt++;
947 if (c == 0 && cc == CHAR_0) continue; /* Leading zeroes */
948 count++;
949
950 #ifndef EBCDIC /* ASCII/UTF-8 coding */
951 if (cc >= CHAR_a) cc -= 32; /* Convert to upper case */
952 c = (c << 4) + cc - ((cc < CHAR_A)? CHAR_0 : (CHAR_A - 10));
953 #else /* EBCDIC coding */
954 if (cc >= CHAR_a && cc <= CHAR_z) cc += 64; /* Convert to upper case */
955 c = (c << 4) + cc - ((cc >= CHAR_0)? CHAR_0 : (CHAR_A - 10));
956 #endif
957 }
958
959 if (*pt == CHAR_RIGHT_CURLY_BRACKET)
960 {
961 if (c < 0 || count > (utf8? 8 : 2)) *errorcodeptr = ERR34;
962 ptr = pt;
963 break;
964 }
965
966 /* If the sequence of hex digits does not end with '}', then we don't
967 recognize this construct; fall through to the normal \x handling. */
968 }
969
970 /* Read just a single-byte hex-defined char */
971
972 c = 0;
973 while (i++ < 2 && (digitab[ptr[1]] & ctype_xdigit) != 0)
974 {
975 int cc; /* Some compilers don't like */
976 cc = *(++ptr); /* ++ in initializers */
977 #ifndef EBCDIC /* ASCII/UTF-8 coding */
978 if (cc >= CHAR_a) cc -= 32; /* Convert to upper case */
979 c = c * 16 + cc - ((cc < CHAR_A)? CHAR_0 : (CHAR_A - 10));
980 #else /* EBCDIC coding */
981 if (cc <= CHAR_z) cc += 64; /* Convert to upper case */
982 c = c * 16 + cc - ((cc >= CHAR_0)? CHAR_0 : (CHAR_A - 10));
983 #endif
984 }
985 break;
986
987 /* For \c, a following letter is upper-cased; then the 0x40 bit is flipped.
988 An error is given if the byte following \c is not an ASCII character. This
989 coding is ASCII-specific, but then the whole concept of \cx is
990 ASCII-specific. (However, an EBCDIC equivalent has now been added.) */
991
992 case CHAR_c:
993 c = *(++ptr);
994 if (c == 0)
995 {
996 *errorcodeptr = ERR2;
997 break;
998 }
999 #ifndef EBCDIC /* ASCII/UTF-8 coding */
1000 if (c > 127) /* Excludes all non-ASCII in either mode */
1001 {
1002 *errorcodeptr = ERR68;
1003 break;
1004 }
1005 if (c >= CHAR_a && c <= CHAR_z) c -= 32;
1006 c ^= 0x40;
1007 #else /* EBCDIC coding */
1008 if (c >= CHAR_a && c <= CHAR_z) c += 64;
1009 c ^= 0xC0;
1010 #endif
1011 break;
1012
1013 /* PCRE_EXTRA enables extensions to Perl in the matter of escapes. Any
1014 other alphanumeric following \ is an error if PCRE_EXTRA was set;
1015 otherwise, for Perl compatibility, it is a literal. This code looks a bit
1016 odd, but there used to be some cases other than the default, and there may
1017 be again in future, so I haven't "optimized" it. */
1018
1019 default:
1020 if ((options & PCRE_EXTRA) != 0) switch(c)
1021 {
1022 default:
1023 *errorcodeptr = ERR3;
1024 break;
1025 }
1026 break;
1027 }
1028 }
1029
1030 /* Perl supports \N{name} for character names, as well as plain \N for "not
1031 newline". PCRE does not support \N{name}. However, it does support
1032 quantification such as \N{2,3}. */
1033
1034 if (c == -ESC_N && ptr[1] == CHAR_LEFT_CURLY_BRACKET &&
1035 !is_counted_repeat(ptr+2))
1036 *errorcodeptr = ERR37;
1037
1038 /* If PCRE_UCP is set, we change the values for \d etc. */
1039
1040 if ((options & PCRE_UCP) != 0 && c <= -ESC_D && c >= -ESC_w)
1041 c -= (ESC_DU - ESC_D);
1042
1043 /* Set the pointer to the final character before returning. */
1044
1045 *ptrptr = ptr;
1046 return c;
1047 }
1048
1049
1050
1051 #ifdef SUPPORT_UCP
1052 /*************************************************
1053 * Handle \P and \p *
1054 *************************************************/
1055
1056 /* This function is called after \P or \p has been encountered, provided that
1057 PCRE is compiled with support for Unicode properties. On entry, ptrptr is
1058 pointing at the P or p. On exit, it is pointing at the final character of the
1059 escape sequence.
1060
1061 Argument:
1062 ptrptr points to the pattern position pointer
1063 negptr points to a boolean that is set TRUE for negation else FALSE
1064 dptr points to an int that is set to the detailed property value
1065 errorcodeptr points to the error code variable
1066
1067 Returns: type value from ucp_type_table, or -1 for an invalid type
1068 */
1069
1070 static int
1071 get_ucp(const pcre_uchar **ptrptr, BOOL *negptr, int *dptr, int *errorcodeptr)
1072 {
1073 int c, i, bot, top;
1074 const pcre_uchar *ptr = *ptrptr;
1075 pcre_uchar name[32];
1076
1077 c = *(++ptr);
1078 if (c == 0) goto ERROR_RETURN;
1079
1080 *negptr = FALSE;
1081
1082 /* \P or \p can be followed by a name in {}, optionally preceded by ^ for
1083 negation. */
1084
1085 if (c == CHAR_LEFT_CURLY_BRACKET)
1086 {
1087 if (ptr[1] == CHAR_CIRCUMFLEX_ACCENT)
1088 {
1089 *negptr = TRUE;
1090 ptr++;
1091 }
1092 for (i = 0; i < (int)sizeof(name) - 1; i++)
1093 {
1094 c = *(++ptr);
1095 if (c == 0) goto ERROR_RETURN;
1096 if (c == CHAR_RIGHT_CURLY_BRACKET) break;
1097 name[i] = c;
1098 }
1099 if (c != CHAR_RIGHT_CURLY_BRACKET) goto ERROR_RETURN;
1100 name[i] = 0;
1101 }
1102
1103 /* Otherwise there is just one following character */
1104
1105 else
1106 {
1107 name[0] = c;
1108 name[1] = 0;
1109 }
1110
1111 *ptrptr = ptr;
1112
1113 /* Search for a recognized property name using binary chop */
1114
1115 bot = 0;
1116 top = PRIV(utt_size);
1117
1118 while (bot < top)
1119 {
1120 i = (bot + top) >> 1;
1121 c = STRCMP_UC_C8(name, PRIV(utt_names) + PRIV(utt)[i].name_offset);
1122 if (c == 0)
1123 {
1124 *dptr = PRIV(utt)[i].value;
1125 return PRIV(utt)[i].type;
1126 }
1127 if (c > 0) bot = i + 1; else top = i;
1128 }
1129
1130 *errorcodeptr = ERR47;
1131 *ptrptr = ptr;
1132 return -1;
1133
1134 ERROR_RETURN:
1135 *errorcodeptr = ERR46;
1136 *ptrptr = ptr;
1137 return -1;
1138 }
1139 #endif
1140
1141
1142
1143
1144 /*************************************************
1145 * Read repeat counts *
1146 *************************************************/
1147
1148 /* Read an item of the form {n,m} and return the values. This is called only
1149 after is_counted_repeat() has confirmed that a repeat-count quantifier exists,
1150 so the syntax is guaranteed to be correct, but we need to check the values.
1151
1152 Arguments:
1153 p pointer to first char after '{'
1154 minp pointer to int for min
1155 maxp pointer to int for max
1156 returned as -1 if no max
1157 errorcodeptr points to error code variable
1158
1159 Returns: pointer to '}' on success;
1160 current ptr on error, with errorcodeptr set non-zero
1161 */
1162
1163 static const pcre_uchar *
1164 read_repeat_counts(const pcre_uchar *p, int *minp, int *maxp, int *errorcodeptr)
1165 {
1166 int min = 0;
1167 int max = -1;
1168
1169 /* Read the minimum value and do a paranoid check: a negative value indicates
1170 an integer overflow. */
1171
1172 while ((digitab[*p] & ctype_digit) != 0) min = min * 10 + *p++ - CHAR_0;
1173 if (min < 0 || min > 65535)
1174 {
1175 *errorcodeptr = ERR5;
1176 return p;
1177 }
1178
1179 /* Read the maximum value if there is one, and again do a paranoid on its size.
1180 Also, max must not be less than min. */
1181
1182 if (*p == CHAR_RIGHT_CURLY_BRACKET) max = min; else
1183 {
1184 if (*(++p) != CHAR_RIGHT_CURLY_BRACKET)
1185 {
1186 max = 0;
1187 while((digitab[*p] & ctype_digit) != 0) max = max * 10 + *p++ - CHAR_0;
1188 if (max < 0 || max > 65535)
1189 {
1190 *errorcodeptr = ERR5;
1191 return p;
1192 }
1193 if (max < min)
1194 {
1195 *errorcodeptr = ERR4;
1196 return p;
1197 }
1198 }
1199 }
1200
1201 /* Fill in the required variables, and pass back the pointer to the terminating
1202 '}'. */
1203
1204 *minp = min;
1205 *maxp = max;
1206 return p;
1207 }
1208
1209
1210
1211 /*************************************************
1212 * Subroutine for finding forward reference *
1213 *************************************************/
1214
1215 /* This recursive function is called only from find_parens() below. The
1216 top-level call starts at the beginning of the pattern. All other calls must
1217 start at a parenthesis. It scans along a pattern's text looking for capturing
1218 subpatterns, and counting them. If it finds a named pattern that matches the
1219 name it is given, it returns its number. Alternatively, if the name is NULL, it
1220 returns when it reaches a given numbered subpattern. Recursion is used to keep
1221 track of subpatterns that reset the capturing group numbers - the (?| feature.
1222
1223 This function was originally called only from the second pass, in which we know
1224 that if (?< or (?' or (?P< is encountered, the name will be correctly
1225 terminated because that is checked in the first pass. There is now one call to
1226 this function in the first pass, to check for a recursive back reference by
1227 name (so that we can make the whole group atomic). In this case, we need check
1228 only up to the current position in the pattern, and that is still OK because
1229 and previous occurrences will have been checked. To make this work, the test
1230 for "end of pattern" is a check against cd->end_pattern in the main loop,
1231 instead of looking for a binary zero. This means that the special first-pass
1232 call can adjust cd->end_pattern temporarily. (Checks for binary zero while
1233 processing items within the loop are OK, because afterwards the main loop will
1234 terminate.)
1235
1236 Arguments:
1237 ptrptr address of the current character pointer (updated)
1238 cd compile background data
1239 name name to seek, or NULL if seeking a numbered subpattern
1240 lorn name length, or subpattern number if name is NULL
1241 xmode TRUE if we are in /x mode
1242 utf8 TRUE if we are in UTF-8 mode
1243 count pointer to the current capturing subpattern number (updated)
1244
1245 Returns: the number of the named subpattern, or -1 if not found
1246 */
1247
1248 static int
1249 find_parens_sub(pcre_uchar **ptrptr, compile_data *cd, const pcre_uchar *name, int lorn,
1250 BOOL xmode, BOOL utf8, int *count)
1251 {
1252 pcre_uchar *ptr = *ptrptr;
1253 int start_count = *count;
1254 int hwm_count = start_count;
1255 BOOL dup_parens = FALSE;
1256
1257 /* If the first character is a parenthesis, check on the type of group we are
1258 dealing with. The very first call may not start with a parenthesis. */
1259
1260 if (ptr[0] == CHAR_LEFT_PARENTHESIS)
1261 {
1262 /* Handle specials such as (*SKIP) or (*UTF8) etc. */
1263
1264 if (ptr[1] == CHAR_ASTERISK) ptr += 2;
1265
1266 /* Handle a normal, unnamed capturing parenthesis. */
1267
1268 else if (ptr[1] != CHAR_QUESTION_MARK)
1269 {
1270 *count += 1;
1271 if (name == NULL && *count == lorn) return *count;
1272 ptr++;
1273 }
1274
1275 /* All cases now have (? at the start. Remember when we are in a group
1276 where the parenthesis numbers are duplicated. */
1277
1278 else if (ptr[2] == CHAR_VERTICAL_LINE)
1279 {
1280 ptr += 3;
1281 dup_parens = TRUE;
1282 }
1283
1284 /* Handle comments; all characters are allowed until a ket is reached. */
1285
1286 else if (ptr[2] == CHAR_NUMBER_SIGN)
1287 {
1288 for (ptr += 3; *ptr != 0; ptr++) if (*ptr == CHAR_RIGHT_PARENTHESIS) break;
1289 goto FAIL_EXIT;
1290 }
1291
1292 /* Handle a condition. If it is an assertion, just carry on so that it
1293 is processed as normal. If not, skip to the closing parenthesis of the
1294 condition (there can't be any nested parens). */
1295
1296 else if (ptr[2] == CHAR_LEFT_PARENTHESIS)
1297 {
1298 ptr += 2;
1299 if (ptr[1] != CHAR_QUESTION_MARK)
1300 {
1301 while (*ptr != 0 && *ptr != CHAR_RIGHT_PARENTHESIS) ptr++;
1302 if (*ptr != 0) ptr++;
1303 }
1304 }
1305
1306 /* Start with (? but not a condition. */
1307
1308 else
1309 {
1310 ptr += 2;
1311 if (*ptr == CHAR_P) ptr++; /* Allow optional P */
1312
1313 /* We have to disambiguate (?<! and (?<= from (?<name> for named groups */
1314
1315 if ((*ptr == CHAR_LESS_THAN_SIGN && ptr[1] != CHAR_EXCLAMATION_MARK &&
1316 ptr[1] != CHAR_EQUALS_SIGN) || *ptr == CHAR_APOSTROPHE)
1317 {
1318 int term;
1319 const pcre_uchar *thisname;
1320 *count += 1;
1321 if (name == NULL && *count == lorn) return *count;
1322 term = *ptr++;
1323 if (term == CHAR_LESS_THAN_SIGN) term = CHAR_GREATER_THAN_SIGN;
1324 thisname = ptr;
1325 while (*ptr != term) ptr++;
1326 if (name != NULL && lorn == ptr - thisname &&
1327 STRNCMP_UC_UC(name, thisname, lorn) == 0)
1328 return *count;
1329 term++;
1330 }
1331 }
1332 }
1333
1334 /* Past any initial parenthesis handling, scan for parentheses or vertical
1335 bars. Stop if we get to cd->end_pattern. Note that this is important for the
1336 first-pass call when this value is temporarily adjusted to stop at the current
1337 position. So DO NOT change this to a test for binary zero. */
1338
1339 for (; ptr < cd->end_pattern; ptr++)
1340 {
1341 /* Skip over backslashed characters and also entire \Q...\E */
1342
1343 if (*ptr == CHAR_BACKSLASH)
1344 {
1345 if (*(++ptr) == 0) goto FAIL_EXIT;
1346 if (*ptr == CHAR_Q) for (;;)
1347 {
1348 while (*(++ptr) != 0 && *ptr != CHAR_BACKSLASH) {};
1349 if (*ptr == 0) goto FAIL_EXIT;
1350 if (*(++ptr) == CHAR_E) break;
1351 }
1352 continue;
1353 }
1354
1355 /* Skip over character classes; this logic must be similar to the way they
1356 are handled for real. If the first character is '^', skip it. Also, if the
1357 first few characters (either before or after ^) are \Q\E or \E we skip them
1358 too. This makes for compatibility with Perl. Note the use of STR macros to
1359 encode "Q\\E" so that it works in UTF-8 on EBCDIC platforms. */
1360
1361 if (*ptr == CHAR_LEFT_SQUARE_BRACKET)
1362 {
1363 BOOL negate_class = FALSE;
1364 for (;;)
1365 {
1366 if (ptr[1] == CHAR_BACKSLASH)
1367 {
1368 if (ptr[2] == CHAR_E)
1369 ptr+= 2;
1370 else if (STRNCMP_UC_C8(ptr + 2,
1371 STR_Q STR_BACKSLASH STR_E, 3) == 0)
1372 ptr += 4;
1373 else
1374 break;
1375 }
1376 else if (!negate_class && ptr[1] == CHAR_CIRCUMFLEX_ACCENT)
1377 {
1378 negate_class = TRUE;
1379 ptr++;
1380 }
1381 else break;
1382 }
1383
1384 /* If the next character is ']', it is a data character that must be
1385 skipped, except in JavaScript compatibility mode. */
1386
1387 if (ptr[1] == CHAR_RIGHT_SQUARE_BRACKET &&
1388 (cd->external_options & PCRE_JAVASCRIPT_COMPAT) == 0)
1389 ptr++;
1390
1391 while (*(++ptr) != CHAR_RIGHT_SQUARE_BRACKET)
1392 {
1393 if (*ptr == 0) return -1;
1394 if (*ptr == CHAR_BACKSLASH)
1395 {
1396 if (*(++ptr) == 0) goto FAIL_EXIT;
1397 if (*ptr == CHAR_Q) for (;;)
1398 {
1399 while (*(++ptr) != 0 && *ptr != CHAR_BACKSLASH) {};
1400 if (*ptr == 0) goto FAIL_EXIT;
1401 if (*(++ptr) == CHAR_E) break;
1402 }
1403 continue;
1404 }
1405 }
1406 continue;
1407 }
1408
1409 /* Skip comments in /x mode */
1410
1411 if (xmode && *ptr == CHAR_NUMBER_SIGN)
1412 {
1413 ptr++;
1414 while (*ptr != 0)
1415 {
1416 if (IS_NEWLINE(ptr)) { ptr += cd->nllen - 1; break; }
1417 ptr++;
1418 #ifdef SUPPORT_UTF8
1419 if (utf8) while ((*ptr & 0xc0) == 0x80) ptr++;
1420 #endif
1421 }
1422 if (*ptr == 0) goto FAIL_EXIT;
1423 continue;
1424 }
1425
1426 /* Check for the special metacharacters */
1427
1428 if (*ptr == CHAR_LEFT_PARENTHESIS)
1429 {
1430 int rc = find_parens_sub(&ptr, cd, name, lorn, xmode, utf8, count);
1431 if (rc > 0) return rc;
1432 if (*ptr == 0) goto FAIL_EXIT;
1433 }
1434
1435 else if (*ptr == CHAR_RIGHT_PARENTHESIS)
1436 {
1437 if (dup_parens && *count < hwm_count) *count = hwm_count;
1438 goto FAIL_EXIT;
1439 }
1440
1441 else if (*ptr == CHAR_VERTICAL_LINE && dup_parens)
1442 {
1443 if (*count > hwm_count) hwm_count = *count;
1444 *count = start_count;
1445 }
1446 }
1447
1448 FAIL_EXIT:
1449 *ptrptr = ptr;
1450 return -1;
1451 }
1452
1453
1454
1455
1456 /*************************************************
1457 * Find forward referenced subpattern *
1458 *************************************************/
1459
1460 /* This function scans along a pattern's text looking for capturing
1461 subpatterns, and counting them. If it finds a named pattern that matches the
1462 name it is given, it returns its number. Alternatively, if the name is NULL, it
1463 returns when it reaches a given numbered subpattern. This is used for forward
1464 references to subpatterns. We used to be able to start this scan from the
1465 current compiling point, using the current count value from cd->bracount, and
1466 do it all in a single loop, but the addition of the possibility of duplicate
1467 subpattern numbers means that we have to scan from the very start, in order to
1468 take account of such duplicates, and to use a recursive function to keep track
1469 of the different types of group.
1470
1471 Arguments:
1472 cd compile background data
1473 name name to seek, or NULL if seeking a numbered subpattern
1474 lorn name length, or subpattern number if name is NULL
1475 xmode TRUE if we are in /x mode
1476 utf8 TRUE if we are in UTF-8 mode
1477
1478 Returns: the number of the found subpattern, or -1 if not found
1479 */
1480
1481 static int
1482 find_parens(compile_data *cd, const pcre_uchar *name, int lorn, BOOL xmode,
1483 BOOL utf8)
1484 {
1485 pcre_uchar *ptr = (pcre_uchar *)cd->start_pattern;
1486 int count = 0;
1487 int rc;
1488
1489 /* If the pattern does not start with an opening parenthesis, the first call
1490 to find_parens_sub() will scan right to the end (if necessary). However, if it
1491 does start with a parenthesis, find_parens_sub() will return when it hits the
1492 matching closing parens. That is why we have to have a loop. */
1493
1494 for (;;)
1495 {
1496 rc = find_parens_sub(&ptr, cd, name, lorn, xmode, utf8, &count);
1497 if (rc > 0 || *ptr++ == 0) break;
1498 }
1499
1500 return rc;
1501 }
1502
1503
1504
1505
1506 /*************************************************
1507 * Find first significant op code *
1508 *************************************************/
1509
1510 /* This is called by several functions that scan a compiled expression looking
1511 for a fixed first character, or an anchoring op code etc. It skips over things
1512 that do not influence this. For some calls, it makes sense to skip negative
1513 forward and all backward assertions, and also the \b assertion; for others it
1514 does not.
1515
1516 Arguments:
1517 code pointer to the start of the group
1518 skipassert TRUE if certain assertions are to be skipped
1519
1520 Returns: pointer to the first significant opcode
1521 */
1522
1523 static const pcre_uchar*
1524 first_significant_code(const pcre_uchar *code, BOOL skipassert)
1525 {
1526 for (;;)
1527 {
1528 switch ((int)*code)
1529 {
1530 case OP_ASSERT_NOT:
1531 case OP_ASSERTBACK:
1532 case OP_ASSERTBACK_NOT:
1533 if (!skipassert) return code;
1534 do code += GET(code, 1); while (*code == OP_ALT);
1535 code += PRIV(OP_lengths)[*code];
1536 break;
1537
1538 case OP_WORD_BOUNDARY:
1539 case OP_NOT_WORD_BOUNDARY:
1540 if (!skipassert) return code;
1541 /* Fall through */
1542
1543 case OP_CALLOUT:
1544 case OP_CREF:
1545 case OP_NCREF:
1546 case OP_RREF:
1547 case OP_NRREF:
1548 case OP_DEF:
1549 code += PRIV(OP_lengths)[*code];
1550 break;
1551
1552 default:
1553 return code;
1554 }
1555 }
1556 /* Control never reaches here */
1557 }
1558
1559
1560
1561
1562 /*************************************************
1563 * Find the fixed length of a branch *
1564 *************************************************/
1565
1566 /* Scan a branch and compute the fixed length of subject that will match it,
1567 if the length is fixed. This is needed for dealing with backward assertions.
1568 In UTF8 mode, the result is in characters rather than bytes. The branch is
1569 temporarily terminated with OP_END when this function is called.
1570
1571 This function is called when a backward assertion is encountered, so that if it
1572 fails, the error message can point to the correct place in the pattern.
1573 However, we cannot do this when the assertion contains subroutine calls,
1574 because they can be forward references. We solve this by remembering this case
1575 and doing the check at the end; a flag specifies which mode we are running in.
1576
1577 Arguments:
1578 code points to the start of the pattern (the bracket)
1579 utf8 TRUE in UTF-8 mode
1580 atend TRUE if called when the pattern is complete
1581 cd the "compile data" structure
1582
1583 Returns: the fixed length,
1584 or -1 if there is no fixed length,
1585 or -2 if \C was encountered (in UTF-8 mode only)
1586 or -3 if an OP_RECURSE item was encountered and atend is FALSE
1587 or -4 if an unknown opcode was encountered (internal error)
1588 */
1589
1590 static int
1591 find_fixedlength(pcre_uchar *code, BOOL utf8, BOOL atend, compile_data *cd)
1592 {
1593 int length = -1;
1594
1595 register int branchlength = 0;
1596 register pcre_uchar *cc = code + 1 + LINK_SIZE;
1597
1598 /* Scan along the opcodes for this branch. If we get to the end of the
1599 branch, check the length against that of the other branches. */
1600
1601 for (;;)
1602 {
1603 int d;
1604 pcre_uchar *ce, *cs;
1605 register int op = *cc;
1606 switch (op)
1607 {
1608 /* We only need to continue for OP_CBRA (normal capturing bracket) and
1609 OP_BRA (normal non-capturing bracket) because the other variants of these
1610 opcodes are all concerned with unlimited repeated groups, which of course
1611 are not of fixed length. */
1612
1613 case OP_CBRA:
1614 case OP_BRA:
1615 case OP_ONCE:
1616 case OP_ONCE_NC:
1617 case OP_COND:
1618 d = find_fixedlength(cc + ((op == OP_CBRA)? IMM2_SIZE : 0), utf8, atend, cd);
1619 if (d < 0) return d;
1620 branchlength += d;
1621 do cc += GET(cc, 1); while (*cc == OP_ALT);
1622 cc += 1 + LINK_SIZE;
1623 break;
1624
1625 /* Reached end of a branch; if it's a ket it is the end of a nested call.
1626 If it's ALT it is an alternation in a nested call. An ACCEPT is effectively
1627 an ALT. If it is END it's the end of the outer call. All can be handled by
1628 the same code. Note that we must not include the OP_KETRxxx opcodes here,
1629 because they all imply an unlimited repeat. */
1630
1631 case OP_ALT:
1632 case OP_KET:
1633 case OP_END:
1634 case OP_ACCEPT:
1635 case OP_ASSERT_ACCEPT:
1636 if (length < 0) length = branchlength;
1637 else if (length != branchlength) return -1;
1638 if (*cc != OP_ALT) return length;
1639 cc += 1 + LINK_SIZE;
1640 branchlength = 0;
1641 break;
1642
1643 /* A true recursion implies not fixed length, but a subroutine call may
1644 be OK. If the subroutine is a forward reference, we can't deal with
1645 it until the end of the pattern, so return -3. */
1646
1647 case OP_RECURSE:
1648 if (!atend) return -3;
1649 cs = ce = (pcre_uchar *)cd->start_code + GET(cc, 1); /* Start subpattern */
1650 do ce += GET(ce, 1); while (*ce == OP_ALT); /* End subpattern */
1651 if (cc > cs && cc < ce) return -1; /* Recursion */
1652 d = find_fixedlength(cs + 2, utf8, atend, cd);
1653 if (d < 0) return d;
1654 branchlength += d;
1655 cc += 1 + LINK_SIZE;
1656 break;
1657
1658 /* Skip over assertive subpatterns */
1659
1660 case OP_ASSERT:
1661 case OP_ASSERT_NOT:
1662 case OP_ASSERTBACK:
1663 case OP_ASSERTBACK_NOT:
1664 do cc += GET(cc, 1); while (*cc == OP_ALT);
1665 /* Fall through */
1666
1667 /* Skip over things that don't match chars */
1668
1669 case OP_MARK:
1670 case OP_PRUNE_ARG:
1671 case OP_SKIP_ARG:
1672 case OP_THEN_ARG:
1673 cc += cc[1] + PRIV(OP_lengths)[*cc];
1674 break;
1675
1676 case OP_CALLOUT:
1677 case OP_CIRC:
1678 case OP_CIRCM:
1679 case OP_CLOSE:
1680 case OP_COMMIT:
1681 case OP_CREF:
1682 case OP_DEF:
1683 case OP_DOLL:
1684 case OP_DOLLM:
1685 case OP_EOD:
1686 case OP_EODN:
1687 case OP_FAIL:
1688 case OP_NCREF:
1689 case OP_NRREF:
1690 case OP_NOT_WORD_BOUNDARY:
1691 case OP_PRUNE:
1692 case OP_REVERSE:
1693 case OP_RREF:
1694 case OP_SET_SOM:
1695 case OP_SKIP:
1696 case OP_SOD:
1697 case OP_SOM:
1698 case OP_THEN:
1699 case OP_WORD_BOUNDARY:
1700 cc += PRIV(OP_lengths)[*cc];
1701 break;
1702
1703 /* Handle literal characters */
1704
1705 case OP_CHAR:
1706 case OP_CHARI:
1707 case OP_NOT:
1708 case OP_NOTI:
1709 branchlength++;
1710 cc += 2;
1711 #ifdef SUPPORT_UTF8
1712 if (utf8 && cc[-1] >= 0xc0) cc += PRIV(utf8_table4)[cc[-1] & 0x3f];
1713 #endif
1714 break;
1715
1716 /* Handle exact repetitions. The count is already in characters, but we
1717 need to skip over a multibyte character in UTF8 mode. */
1718
1719 case OP_EXACT:
1720 case OP_EXACTI:
1721 case OP_NOTEXACT:
1722 case OP_NOTEXACTI:
1723 branchlength += GET2(cc,1);
1724 cc += 2 + IMM2_SIZE;
1725 #ifdef SUPPORT_UTF8
1726 if (utf8 && cc[-1] >= 0xc0) cc += PRIV(utf8_table4)[cc[-1] & 0x3f];
1727 #endif
1728 break;
1729
1730 case OP_TYPEEXACT:
1731 branchlength += GET2(cc,1);
1732 if (cc[1 + IMM2_SIZE] == OP_PROP || cc[1 + IMM2_SIZE] == OP_NOTPROP) cc += 2;
1733 cc += 1 + IMM2_SIZE + 1;
1734 break;
1735
1736 /* Handle single-char matchers */
1737
1738 case OP_PROP:
1739 case OP_NOTPROP:
1740 cc += 2;
1741 /* Fall through */
1742
1743 case OP_HSPACE:
1744 case OP_VSPACE:
1745 case OP_NOT_HSPACE:
1746 case OP_NOT_VSPACE:
1747 case OP_NOT_DIGIT:
1748 case OP_DIGIT:
1749 case OP_NOT_WHITESPACE:
1750 case OP_WHITESPACE:
1751 case OP_NOT_WORDCHAR:
1752 case OP_WORDCHAR:
1753 case OP_ANY:
1754 case OP_ALLANY:
1755 branchlength++;
1756 cc++;
1757 break;
1758
1759 /* The single-byte matcher isn't allowed. This only happens in UTF-8 mode;
1760 otherwise \C is coded as OP_ALLANY. */
1761
1762 case OP_ANYBYTE:
1763 return -2;
1764
1765 /* Check a class for variable quantification */
1766
1767 #ifdef SUPPORT_UTF8
1768 case OP_XCLASS:
1769 cc += GET(cc, 1) - 33;
1770 /* Fall through */
1771 #endif
1772
1773 case OP_CLASS:
1774 case OP_NCLASS:
1775 cc += 33;
1776
1777 switch (*cc)
1778 {
1779 case OP_CRPLUS:
1780 case OP_CRMINPLUS:
1781 case OP_CRSTAR:
1782 case OP_CRMINSTAR:
1783 case OP_CRQUERY:
1784 case OP_CRMINQUERY:
1785 return -1;
1786
1787 case OP_CRRANGE:
1788 case OP_CRMINRANGE:
1789 if (GET2(cc,1) != GET2(cc,1+IMM2_SIZE)) return -1;
1790 branchlength += GET2(cc,1);
1791 cc += 1 + 2 * IMM2_SIZE;
1792 break;
1793
1794 default:
1795 branchlength++;
1796 }
1797 break;
1798
1799 /* Anything else is variable length */
1800
1801 case OP_ANYNL:
1802 case OP_BRAMINZERO:
1803 case OP_BRAPOS:
1804 case OP_BRAPOSZERO:
1805 case OP_BRAZERO:
1806 case OP_CBRAPOS:
1807 case OP_EXTUNI:
1808 case OP_KETRMAX:
1809 case OP_KETRMIN:
1810 case OP_KETRPOS:
1811 case OP_MINPLUS:
1812 case OP_MINPLUSI:
1813 case OP_MINQUERY:
1814 case OP_MINQUERYI:
1815 case OP_MINSTAR:
1816 case OP_MINSTARI:
1817 case OP_MINUPTO:
1818 case OP_MINUPTOI:
1819 case OP_NOTMINPLUS:
1820 case OP_NOTMINPLUSI:
1821 case OP_NOTMINQUERY:
1822 case OP_NOTMINQUERYI:
1823 case OP_NOTMINSTAR:
1824 case OP_NOTMINSTARI:
1825 case OP_NOTMINUPTO:
1826 case OP_NOTMINUPTOI:
1827 case OP_NOTPLUS:
1828 case OP_NOTPLUSI:
1829 case OP_NOTPOSPLUS:
1830 case OP_NOTPOSPLUSI:
1831 case OP_NOTPOSQUERY:
1832 case OP_NOTPOSQUERYI:
1833 case OP_NOTPOSSTAR:
1834 case OP_NOTPOSSTARI:
1835 case OP_NOTPOSUPTO:
1836 case OP_NOTPOSUPTOI:
1837 case OP_NOTQUERY:
1838 case OP_NOTQUERYI:
1839 case OP_NOTSTAR:
1840 case OP_NOTSTARI:
1841 case OP_NOTUPTO:
1842 case OP_NOTUPTOI:
1843 case OP_PLUS:
1844 case OP_PLUSI:
1845 case OP_POSPLUS:
1846 case OP_POSPLUSI:
1847 case OP_POSQUERY:
1848 case OP_POSQUERYI:
1849 case OP_POSSTAR:
1850 case OP_POSSTARI:
1851 case OP_POSUPTO:
1852 case OP_POSUPTOI:
1853 case OP_QUERY:
1854 case OP_QUERYI:
1855 case OP_REF:
1856 case OP_REFI:
1857 case OP_SBRA:
1858 case OP_SBRAPOS:
1859 case OP_SCBRA:
1860 case OP_SCBRAPOS:
1861 case OP_SCOND:
1862 case OP_SKIPZERO:
1863 case OP_STAR:
1864 case OP_STARI:
1865 case OP_TYPEMINPLUS:
1866 case OP_TYPEMINQUERY:
1867 case OP_TYPEMINSTAR:
1868 case OP_TYPEMINUPTO:
1869 case OP_TYPEPLUS:
1870 case OP_TYPEPOSPLUS:
1871 case OP_TYPEPOSQUERY:
1872 case OP_TYPEPOSSTAR:
1873 case OP_TYPEPOSUPTO:
1874 case OP_TYPEQUERY:
1875 case OP_TYPESTAR:
1876 case OP_TYPEUPTO:
1877 case OP_UPTO:
1878 case OP_UPTOI:
1879 return -1;
1880
1881 /* Catch unrecognized opcodes so that when new ones are added they
1882 are not forgotten, as has happened in the past. */
1883
1884 default:
1885 return -4;
1886 }
1887 }
1888 /* Control never gets here */
1889 }
1890
1891
1892
1893
1894 /*************************************************
1895 * Scan compiled regex for specific bracket *
1896 *************************************************/
1897
1898 /* This little function scans through a compiled pattern until it finds a
1899 capturing bracket with the given number, or, if the number is negative, an
1900 instance of OP_REVERSE for a lookbehind. The function is global in the C sense
1901 so that it can be called from pcre_study() when finding the minimum matching
1902 length.
1903
1904 Arguments:
1905 code points to start of expression
1906 utf8 TRUE in UTF-8 mode
1907 number the required bracket number or negative to find a lookbehind
1908
1909 Returns: pointer to the opcode for the bracket, or NULL if not found
1910 */
1911
1912 const pcre_uchar *
1913 PRIV(find_bracket)(const pcre_uchar *code, BOOL utf8, int number)
1914 {
1915 for (;;)
1916 {
1917 register int c = *code;
1918
1919 if (c == OP_END) return NULL;
1920
1921 /* XCLASS is used for classes that cannot be represented just by a bit
1922 map. This includes negated single high-valued characters. The length in
1923 the table is zero; the actual length is stored in the compiled code. */
1924
1925 if (c == OP_XCLASS) code += GET(code, 1);
1926
1927 /* Handle recursion */
1928
1929 else if (c == OP_REVERSE)
1930 {
1931 if (number < 0) return (pcre_uchar *)code;
1932 code += PRIV(OP_lengths)[c];
1933 }
1934
1935 /* Handle capturing bracket */
1936
1937 else if (c == OP_CBRA || c == OP_SCBRA ||
1938 c == OP_CBRAPOS || c == OP_SCBRAPOS)
1939 {
1940 int n = GET2(code, 1+LINK_SIZE);
1941 if (n == number) return (pcre_uchar *)code;
1942 code += PRIV(OP_lengths)[c];
1943 }
1944
1945 /* Otherwise, we can get the item's length from the table, except that for
1946 repeated character types, we have to test for \p and \P, which have an extra
1947 two bytes of parameters, and for MARK/PRUNE/SKIP/THEN with an argument, we
1948 must add in its length. */
1949
1950 else
1951 {
1952 switch(c)
1953 {
1954 case OP_TYPESTAR:
1955 case OP_TYPEMINSTAR:
1956 case OP_TYPEPLUS:
1957 case OP_TYPEMINPLUS:
1958 case OP_TYPEQUERY:
1959 case OP_TYPEMINQUERY:
1960 case OP_TYPEPOSSTAR:
1961 case OP_TYPEPOSPLUS:
1962 case OP_TYPEPOSQUERY:
1963 if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
1964 break;
1965
1966 case OP_TYPEUPTO:
1967 case OP_TYPEMINUPTO:
1968 case OP_TYPEEXACT:
1969 case OP_TYPEPOSUPTO:
1970 if (code[1 + IMM2_SIZE] == OP_PROP
1971 || code[1 + IMM2_SIZE] == OP_NOTPROP) code += 2;
1972 break;
1973
1974 case OP_MARK:
1975 case OP_PRUNE_ARG:
1976 case OP_SKIP_ARG:
1977 code += code[1];
1978 break;
1979
1980 case OP_THEN_ARG:
1981 code += code[1];
1982 break;
1983 }
1984
1985 /* Add in the fixed length from the table */
1986
1987 code += PRIV(OP_lengths)[c];
1988
1989 /* In UTF-8 mode, opcodes that are followed by a character may be followed by
1990 a multi-byte character. The length in the table is a minimum, so we have to
1991 arrange to skip the extra bytes. */
1992
1993 #ifdef SUPPORT_UTF8
1994 if (utf8) switch(c)
1995 {
1996 case OP_CHAR:
1997 case OP_CHARI:
1998 case OP_EXACT:
1999 case OP_EXACTI:
2000 case OP_UPTO:
2001 case OP_UPTOI:
2002 case OP_MINUPTO:
2003 case OP_MINUPTOI:
2004 case OP_POSUPTO:
2005 case OP_POSUPTOI:
2006 case OP_STAR:
2007 case OP_STARI:
2008 case OP_MINSTAR:
2009 case OP_MINSTARI:
2010 case OP_POSSTAR:
2011 case OP_POSSTARI:
2012 case OP_PLUS:
2013 case OP_PLUSI:
2014 case OP_MINPLUS:
2015 case OP_MINPLUSI:
2016 case OP_POSPLUS:
2017 case OP_POSPLUSI:
2018 case OP_QUERY:
2019 case OP_QUERYI:
2020 case OP_MINQUERY:
2021 case OP_MINQUERYI:
2022 case OP_POSQUERY:
2023 case OP_POSQUERYI:
2024 if (code[-1] >= 0xc0) code += PRIV(utf8_table4)[code[-1] & 0x3f];
2025 break;
2026 }
2027 #else
2028 (void)(utf8); /* Keep compiler happy by referencing function argument */
2029 #endif
2030 }
2031 }
2032 }
2033
2034
2035
2036 /*************************************************
2037 * Scan compiled regex for recursion reference *
2038 *************************************************/
2039
2040 /* This little function scans through a compiled pattern until it finds an
2041 instance of OP_RECURSE.
2042
2043 Arguments:
2044 code points to start of expression
2045 utf8 TRUE in UTF-8 mode
2046
2047 Returns: pointer to the opcode for OP_RECURSE, or NULL if not found
2048 */
2049
2050 static const pcre_uchar *
2051 find_recurse(const pcre_uchar *code, BOOL utf8)
2052 {
2053 for (;;)
2054 {
2055 register int c = *code;
2056 if (c == OP_END) return NULL;
2057 if (c == OP_RECURSE) return code;
2058
2059 /* XCLASS is used for classes that cannot be represented just by a bit
2060 map. This includes negated single high-valued characters. The length in
2061 the table is zero; the actual length is stored in the compiled code. */
2062
2063 if (c == OP_XCLASS) code += GET(code, 1);
2064
2065 /* Otherwise, we can get the item's length from the table, except that for
2066 repeated character types, we have to test for \p and \P, which have an extra
2067 two bytes of parameters, and for MARK/PRUNE/SKIP/THEN with an argument, we
2068 must add in its length. */
2069
2070 else
2071 {
2072 switch(c)
2073 {
2074 case OP_TYPESTAR:
2075 case OP_TYPEMINSTAR:
2076 case OP_TYPEPLUS:
2077 case OP_TYPEMINPLUS:
2078 case OP_TYPEQUERY:
2079 case OP_TYPEMINQUERY:
2080 case OP_TYPEPOSSTAR:
2081 case OP_TYPEPOSPLUS:
2082 case OP_TYPEPOSQUERY:
2083 if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
2084 break;
2085
2086 case OP_TYPEPOSUPTO:
2087 case OP_TYPEUPTO:
2088 case OP_TYPEMINUPTO:
2089 case OP_TYPEEXACT:
2090 if (code[1 + IMM2_SIZE] == OP_PROP
2091 || code[1 + IMM2_SIZE] == OP_NOTPROP) code += 2;
2092 break;
2093
2094 case OP_MARK:
2095 case OP_PRUNE_ARG:
2096 case OP_SKIP_ARG:
2097 code += code[1];
2098 break;
2099
2100 case OP_THEN_ARG:
2101 code += code[1];
2102 break;
2103 }
2104
2105 /* Add in the fixed length from the table */
2106
2107 code += PRIV(OP_lengths)[c];
2108
2109 /* In UTF-8 mode, opcodes that are followed by a character may be followed
2110 by a multi-byte character. The length in the table is a minimum, so we have
2111 to arrange to skip the extra bytes. */
2112
2113 #ifdef SUPPORT_UTF8
2114 if (utf8) switch(c)
2115 {
2116 case OP_CHAR:
2117 case OP_CHARI:
2118 case OP_EXACT:
2119 case OP_EXACTI:
2120 case OP_UPTO:
2121 case OP_UPTOI:
2122 case OP_MINUPTO:
2123 case OP_MINUPTOI:
2124 case OP_POSUPTO:
2125 case OP_POSUPTOI:
2126 case OP_STAR:
2127 case OP_STARI:
2128 case OP_MINSTAR:
2129 case OP_MINSTARI:
2130 case OP_POSSTAR:
2131 case OP_POSSTARI:
2132 case OP_PLUS:
2133 case OP_PLUSI:
2134 case OP_MINPLUS:
2135 case OP_MINPLUSI:
2136 case OP_POSPLUS:
2137 case OP_POSPLUSI:
2138 case OP_QUERY:
2139 case OP_QUERYI:
2140 case OP_MINQUERY:
2141 case OP_MINQUERYI:
2142 case OP_POSQUERY:
2143 case OP_POSQUERYI:
2144 if (code[-1] >= 0xc0) code += PRIV(utf8_table4)[code[-1] & 0x3f];
2145 break;
2146 }
2147 #else
2148 (void)(utf8); /* Keep compiler happy by referencing function argument */
2149 #endif
2150 }
2151 }
2152 }
2153
2154
2155
2156 /*************************************************
2157 * Scan compiled branch for non-emptiness *
2158 *************************************************/
2159
2160 /* This function scans through a branch of a compiled pattern to see whether it
2161 can match the empty string or not. It is called from could_be_empty()
2162 below and from compile_branch() when checking for an unlimited repeat of a
2163 group that can match nothing. Note that first_significant_code() skips over
2164 backward and negative forward assertions when its final argument is TRUE. If we
2165 hit an unclosed bracket, we return "empty" - this means we've struck an inner
2166 bracket whose current branch will already have been scanned.
2167
2168 Arguments:
2169 code points to start of search
2170 endcode points to where to stop
2171 utf8 TRUE if in UTF8 mode
2172 cd contains pointers to tables etc.
2173
2174 Returns: TRUE if what is matched could be empty
2175 */
2176
2177 static BOOL
2178 could_be_empty_branch(const pcre_uchar *code, const pcre_uchar *endcode,
2179 BOOL utf8, compile_data *cd)
2180 {
2181 register int c;
2182 for (code = first_significant_code(code + PRIV(OP_lengths)[*code], TRUE);
2183 code < endcode;
2184 code = first_significant_code(code + PRIV(OP_lengths)[c], TRUE))
2185 {
2186 const pcre_uchar *ccode;
2187
2188 c = *code;
2189
2190 /* Skip over forward assertions; the other assertions are skipped by
2191 first_significant_code() with a TRUE final argument. */
2192
2193 if (c == OP_ASSERT)
2194 {
2195 do code += GET(code, 1); while (*code == OP_ALT);
2196 c = *code;
2197 continue;
2198 }
2199
2200 /* For a recursion/subroutine call, if its end has been reached, which
2201 implies a backward reference subroutine call, we can scan it. If it's a
2202 forward reference subroutine call, we can't. To detect forward reference
2203 we have to scan up the list that is kept in the workspace. This function is
2204 called only when doing the real compile, not during the pre-compile that
2205 measures the size of the compiled pattern. */
2206
2207 if (c == OP_RECURSE)
2208 {
2209 const pcre_uchar *scode;
2210 BOOL empty_branch;
2211
2212 /* Test for forward reference */
2213
2214 for (scode = cd->start_workspace; scode < cd->hwm; scode += LINK_SIZE)
2215 if (GET(scode, 0) == code + 1 - cd->start_code) return TRUE;
2216
2217 /* Not a forward reference, test for completed backward reference */
2218
2219 empty_branch = FALSE;
2220 scode = cd->start_code + GET(code, 1);
2221 if (GET(scode, 1) == 0) return TRUE; /* Unclosed */
2222
2223 /* Completed backwards reference */
2224
2225 do
2226 {
2227 if (could_be_empty_branch(scode, endcode, utf8, cd))
2228 {
2229 empty_branch = TRUE;
2230 break;
2231 }
2232 scode += GET(scode, 1);
2233 }
2234 while (*scode == OP_ALT);
2235
2236 if (!empty_branch) return FALSE; /* All branches are non-empty */
2237 continue;
2238 }
2239
2240 /* Groups with zero repeats can of course be empty; skip them. */
2241
2242 if (c == OP_BRAZERO || c == OP_BRAMINZERO || c == OP_SKIPZERO ||
2243 c == OP_BRAPOSZERO)
2244 {
2245 code += PRIV(OP_lengths)[c];
2246 do code += GET(code, 1); while (*code == OP_ALT);
2247 c = *code;
2248 continue;
2249 }
2250
2251 /* A nested group that is already marked as "could be empty" can just be
2252 skipped. */
2253
2254 if (c == OP_SBRA || c == OP_SBRAPOS ||
2255 c == OP_SCBRA || c == OP_SCBRAPOS)
2256 {
2257 do code += GET(code, 1); while (*code == OP_ALT);
2258 c = *code;
2259 continue;
2260 }
2261
2262 /* For other groups, scan the branches. */
2263
2264 if (c == OP_BRA || c == OP_BRAPOS ||
2265 c == OP_CBRA || c == OP_CBRAPOS ||
2266 c == OP_ONCE || c == OP_ONCE_NC ||
2267 c == OP_COND)
2268 {
2269 BOOL empty_branch;
2270 if (GET(code, 1) == 0) return TRUE; /* Hit unclosed bracket */
2271
2272 /* If a conditional group has only one branch, there is a second, implied,
2273 empty branch, so just skip over the conditional, because it could be empty.
2274 Otherwise, scan the individual branches of the group. */
2275
2276 if (c == OP_COND && code[GET(code, 1)] != OP_ALT)
2277 code += GET(code, 1);
2278 else
2279 {
2280 empty_branch = FALSE;
2281 do
2282 {
2283 if (!empty_branch && could_be_empty_branch(code, endcode, utf8, cd))
2284 empty_branch = TRUE;
2285 code += GET(code, 1);
2286 }
2287 while (*code == OP_ALT);
2288 if (!empty_branch) return FALSE; /* All branches are non-empty */
2289 }
2290
2291 c = *code;
2292 continue;
2293 }
2294
2295 /* Handle the other opcodes */
2296
2297 switch (c)
2298 {
2299 /* Check for quantifiers after a class. XCLASS is used for classes that
2300 cannot be represented just by a bit map. This includes negated single
2301 high-valued characters. The length in PRIV(OP_lengths)[] is zero; the
2302 actual length is stored in the compiled code, so we must update "code"
2303 here. */
2304
2305 #ifdef SUPPORT_UTF8
2306 case OP_XCLASS:
2307 ccode = code += GET(code, 1);
2308 goto CHECK_CLASS_REPEAT;
2309 #endif
2310
2311 case OP_CLASS:
2312 case OP_NCLASS:
2313 ccode = code + 33;
2314
2315 #ifdef SUPPORT_UTF8
2316 CHECK_CLASS_REPEAT:
2317 #endif
2318
2319 switch (*ccode)
2320 {
2321 case OP_CRSTAR: /* These could be empty; continue */
2322 case OP_CRMINSTAR:
2323 case OP_CRQUERY:
2324 case OP_CRMINQUERY:
2325 break;
2326
2327 default: /* Non-repeat => class must match */
2328 case OP_CRPLUS: /* These repeats aren't empty */
2329 case OP_CRMINPLUS:
2330 return FALSE;
2331
2332 case OP_CRRANGE:
2333 case OP_CRMINRANGE:
2334 if (GET2(ccode, 1) > 0) return FALSE; /* Minimum > 0 */
2335 break;
2336 }
2337 break;
2338
2339 /* Opcodes that must match a character */
2340
2341 case OP_PROP:
2342 case OP_NOTPROP:
2343 case OP_EXTUNI:
2344 case OP_NOT_DIGIT:
2345 case OP_DIGIT:
2346 case OP_NOT_WHITESPACE:
2347 case OP_WHITESPACE:
2348 case OP_NOT_WORDCHAR:
2349 case OP_WORDCHAR:
2350 case OP_ANY:
2351 case OP_ALLANY:
2352 case OP_ANYBYTE:
2353 case OP_CHAR:
2354 case OP_CHARI:
2355 case OP_NOT:
2356 case OP_NOTI:
2357 case OP_PLUS:
2358 case OP_MINPLUS:
2359 case OP_POSPLUS:
2360 case OP_EXACT:
2361 case OP_NOTPLUS:
2362 case OP_NOTMINPLUS:
2363 case OP_NOTPOSPLUS:
2364 case OP_NOTEXACT:
2365 case OP_TYPEPLUS:
2366 case OP_TYPEMINPLUS:
2367 case OP_TYPEPOSPLUS:
2368 case OP_TYPEEXACT:
2369 return FALSE;
2370
2371 /* These are going to continue, as they may be empty, but we have to
2372 fudge the length for the \p and \P cases. */
2373
2374 case OP_TYPESTAR:
2375 case OP_TYPEMINSTAR:
2376 case OP_TYPEPOSSTAR:
2377 case OP_TYPEQUERY:
2378 case OP_TYPEMINQUERY:
2379 case OP_TYPEPOSQUERY:
2380 if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
2381 break;
2382
2383 /* Same for these */
2384
2385 case OP_TYPEUPTO:
2386 case OP_TYPEMINUPTO:
2387 case OP_TYPEPOSUPTO:
2388 if (code[1 + IMM2_SIZE] == OP_PROP
2389 || code[1 + IMM2_SIZE] == OP_NOTPROP) code += 2;
2390 break;
2391
2392 /* End of branch */
2393
2394 case OP_KET:
2395 case OP_KETRMAX:
2396 case OP_KETRMIN:
2397 case OP_KETRPOS:
2398 case OP_ALT:
2399 return TRUE;
2400
2401 /* In UTF-8 mode, STAR, MINSTAR, POSSTAR, QUERY, MINQUERY, POSQUERY, UPTO,
2402 MINUPTO, and POSUPTO may be followed by a multibyte character */
2403
2404 #ifdef SUPPORT_UTF8
2405 case OP_STAR:
2406 case OP_STARI:
2407 case OP_MINSTAR:
2408 case OP_MINSTARI:
2409 case OP_POSSTAR:
2410 case OP_POSSTARI:
2411 case OP_QUERY:
2412 case OP_QUERYI:
2413 case OP_MINQUERY:
2414 case OP_MINQUERYI:
2415 case OP_POSQUERY:
2416 case OP_POSQUERYI:
2417 if (utf8 && code[1] >= 0xc0) code += PRIV(utf8_table4)[code[1] & 0x3f];
2418 break;
2419
2420 case OP_UPTO:
2421 case OP_UPTOI:
2422 case OP_MINUPTO:
2423 case OP_MINUPTOI:
2424 case OP_POSUPTO:
2425 case OP_POSUPTOI:
2426 if (utf8 && code[1 + IMM2_SIZE] >= 0xc0) code += PRIV(utf8_table4)[code[1 + IMM2_SIZE] & 0x3f];
2427 break;
2428 #endif
2429
2430 /* MARK, and PRUNE/SKIP/THEN with an argument must skip over the argument
2431 string. */
2432
2433 case OP_MARK:
2434 case OP_PRUNE_ARG:
2435 case OP_SKIP_ARG:
2436 code += code[1];
2437 break;
2438
2439 case OP_THEN_ARG:
2440 code += code[1];
2441 break;
2442
2443 /* None of the remaining opcodes are required to match a character. */
2444
2445 default:
2446 break;
2447 }
2448 }
2449
2450 return TRUE;
2451 }
2452
2453
2454
2455 /*************************************************
2456 * Scan compiled regex for non-emptiness *
2457 *************************************************/
2458
2459 /* This function is called to check for left recursive calls. We want to check
2460 the current branch of the current pattern to see if it could match the empty
2461 string. If it could, we must look outwards for branches at other levels,
2462 stopping when we pass beyond the bracket which is the subject of the recursion.
2463 This function is called only during the real compile, not during the
2464 pre-compile.
2465
2466 Arguments:
2467 code points to start of the recursion
2468 endcode points to where to stop (current RECURSE item)
2469 bcptr points to the chain of current (unclosed) branch starts
2470 utf8 TRUE if in UTF-8 mode
2471 cd pointers to tables etc
2472
2473 Returns: TRUE if what is matched could be empty
2474 */
2475
2476 static BOOL
2477 could_be_empty(const pcre_uchar *code, const pcre_uchar *endcode,
2478 branch_chain *bcptr, BOOL utf8, compile_data *cd)
2479 {
2480 while (bcptr != NULL && bcptr->current_branch >= code)
2481 {
2482 if (!could_be_empty_branch(bcptr->current_branch, endcode, utf8, cd))
2483 return FALSE;
2484 bcptr = bcptr->outer;
2485 }
2486 return TRUE;
2487 }
2488
2489
2490
2491 /*************************************************
2492 * Check for POSIX class syntax *
2493 *************************************************/
2494
2495 /* This function is called when the sequence "[:" or "[." or "[=" is
2496 encountered in a character class. It checks whether this is followed by a
2497 sequence of characters terminated by a matching ":]" or ".]" or "=]". If we
2498 reach an unescaped ']' without the special preceding character, return FALSE.
2499
2500 Originally, this function only recognized a sequence of letters between the
2501 terminators, but it seems that Perl recognizes any sequence of characters,
2502 though of course unknown POSIX names are subsequently rejected. Perl gives an
2503 "Unknown POSIX class" error for [:f\oo:] for example, where previously PCRE
2504 didn't consider this to be a POSIX class. Likewise for [:1234:].
2505
2506 The problem in trying to be exactly like Perl is in the handling of escapes. We
2507 have to be sure that [abc[:x\]pqr] is *not* treated as containing a POSIX
2508 class, but [abc[:x\]pqr:]] is (so that an error can be generated). The code
2509 below handles the special case of \], but does not try to do any other escape
2510 processing. This makes it different from Perl for cases such as [:l\ower:]
2511 where Perl recognizes it as the POSIX class "lower" but PCRE does not recognize
2512 "l\ower". This is a lesser evil that not diagnosing bad classes when Perl does,
2513 I think.
2514
2515 A user pointed out that PCRE was rejecting [:a[:digit:]] whereas Perl was not.
2516 It seems that the appearance of a nested POSIX class supersedes an apparent
2517 external class. For example, [:a[:digit:]b:] matches "a", "b", ":", or
2518 a digit.
2519
2520 In Perl, unescaped square brackets may also appear as part of class names. For
2521 example, [:a[:abc]b:] gives unknown POSIX class "[:abc]b:]". However, for
2522 [:a[:abc]b][b:] it gives unknown POSIX class "[:abc]b][b:]", which does not
2523 seem right at all. PCRE does not allow closing square brackets in POSIX class
2524 names.
2525
2526 Arguments:
2527 ptr pointer to the initial [
2528 endptr where to return the end pointer
2529
2530 Returns: TRUE or FALSE
2531 */
2532
2533 static BOOL
2534 check_posix_syntax(const pcre_uchar *ptr, const pcre_uchar **endptr)
2535 {
2536 int terminator; /* Don't combine these lines; the Solaris cc */
2537 terminator = *(++ptr); /* compiler warns about "non-constant" initializer. */
2538 for (++ptr; *ptr != 0; ptr++)
2539 {
2540 if (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET)
2541 ptr++;
2542 else if (*ptr == CHAR_RIGHT_SQUARE_BRACKET) return FALSE;
2543 else
2544 {
2545 if (*ptr == terminator && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET)
2546 {
2547 *endptr = ptr;
2548 return TRUE;
2549 }
2550 if (*ptr == CHAR_LEFT_SQUARE_BRACKET &&
2551 (ptr[1] == CHAR_COLON || ptr[1] == CHAR_DOT ||
2552 ptr[1] == CHAR_EQUALS_SIGN) &&
2553 check_posix_syntax(ptr, endptr))
2554 return FALSE;
2555 }
2556 }
2557 return FALSE;
2558 }
2559
2560
2561
2562
2563 /*************************************************
2564 * Check POSIX class name *
2565 *************************************************/
2566
2567 /* This function is called to check the name given in a POSIX-style class entry
2568 such as [:alnum:].
2569
2570 Arguments:
2571 ptr points to the first letter
2572 len the length of the name
2573
2574 Returns: a value representing the name, or -1 if unknown
2575 */
2576
2577 static int
2578 check_posix_name(const pcre_uchar *ptr, int len)
2579 {
2580 const char *pn = posix_names;
2581 register int yield = 0;
2582 while (posix_name_lengths[yield] != 0)
2583 {
2584 if (len == posix_name_lengths[yield] &&
2585 STRNCMP_UC_C8(ptr, pn, len) == 0) return yield;
2586 pn += posix_name_lengths[yield] + 1;
2587 yield++;
2588 }
2589 return -1;
2590 }
2591
2592
2593 /*************************************************
2594 * Adjust OP_RECURSE items in repeated group *
2595 *************************************************/
2596
2597 /* OP_RECURSE items contain an offset from the start of the regex to the group
2598 that is referenced. This means that groups can be replicated for fixed
2599 repetition simply by copying (because the recursion is allowed to refer to
2600 earlier groups that are outside the current group). However, when a group is
2601 optional (i.e. the minimum quantifier is zero), OP_BRAZERO or OP_SKIPZERO is
2602 inserted before it, after it has been compiled. This means that any OP_RECURSE
2603 items within it that refer to the group itself or any contained groups have to
2604 have their offsets adjusted. That one of the jobs of this function. Before it
2605 is called, the partially compiled regex must be temporarily terminated with
2606 OP_END.
2607
2608 This function has been extended with the possibility of forward references for
2609 recursions and subroutine calls. It must also check the list of such references
2610 for the group we are dealing with. If it finds that one of the recursions in
2611 the current group is on this list, it adjusts the offset in the list, not the
2612 value in the reference (which is a group number).
2613
2614 Arguments:
2615 group points to the start of the group
2616 adjust the amount by which the group is to be moved
2617 utf8 TRUE in UTF-8 mode
2618 cd contains pointers to tables etc.
2619 save_hwm the hwm forward reference pointer at the start of the group
2620
2621 Returns: nothing
2622 */
2623
2624 static void
2625 adjust_recurse(pcre_uchar *group, int adjust, BOOL utf8, compile_data *cd,
2626 pcre_uchar *save_hwm)
2627 {
2628 pcre_uchar *ptr = group;
2629
2630 while ((ptr = (pcre_uchar *)find_recurse(ptr, utf8)) != NULL)
2631 {
2632 int offset;
2633 pcre_uchar *hc;
2634
2635 /* See if this recursion is on the forward reference list. If so, adjust the
2636 reference. */
2637
2638 for (hc = save_hwm; hc < cd->hwm; hc += LINK_SIZE)
2639 {
2640 offset = GET(hc, 0);
2641 if (cd->start_code + offset == ptr + 1)
2642 {
2643 PUT(hc, 0, offset + adjust);
2644 break;
2645 }
2646 }
2647
2648 /* Otherwise, adjust the recursion offset if it's after the start of this
2649 group. */
2650
2651 if (hc >= cd->hwm)
2652 {
2653 offset = GET(ptr, 1);
2654 if (cd->start_code + offset >= group) PUT(ptr, 1, offset + adjust);
2655 }
2656
2657 ptr += 1 + LINK_SIZE;
2658 }
2659 }
2660
2661
2662
2663 /*************************************************
2664 * Insert an automatic callout point *
2665 *************************************************/
2666
2667 /* This function is called when the PCRE_AUTO_CALLOUT option is set, to insert
2668 callout points before each pattern item.
2669
2670 Arguments:
2671 code current code pointer
2672 ptr current pattern pointer
2673 cd pointers to tables etc
2674
2675 Returns: new code pointer
2676 */
2677
2678 static pcre_uchar *
2679 auto_callout(pcre_uchar *code, const pcre_uchar *ptr, compile_data *cd)
2680 {
2681 *code++ = OP_CALLOUT;
2682 *code++ = 255;
2683 PUT(code, 0, (int)(ptr - cd->start_pattern)); /* Pattern offset */
2684 PUT(code, LINK_SIZE, 0); /* Default length */
2685 return code + 2 * LINK_SIZE;
2686 }
2687
2688
2689
2690 /*************************************************
2691 * Complete a callout item *
2692 *************************************************/
2693
2694 /* A callout item contains the length of the next item in the pattern, which
2695 we can't fill in till after we have reached the relevant point. This is used
2696 for both automatic and manual callouts.
2697
2698 Arguments:
2699 previous_callout points to previous callout item
2700 ptr current pattern pointer
2701 cd pointers to tables etc
2702
2703 Returns: nothing
2704 */
2705
2706 static void
2707 complete_callout(pcre_uchar *previous_callout, const pcre_uchar *ptr, compile_data *cd)
2708 {
2709 int length = (int)(ptr - cd->start_pattern - GET(previous_callout, 2));
2710 PUT(previous_callout, 2 + LINK_SIZE, length);
2711 }
2712
2713
2714
2715 #ifdef SUPPORT_UCP
2716 /*************************************************
2717 * Get othercase range *
2718 *************************************************/
2719
2720 /* This function is passed the start and end of a class range, in UTF-8 mode
2721 with UCP support. It searches up the characters, looking for internal ranges of
2722 characters in the "other" case. Each call returns the next one, updating the
2723 start address.
2724
2725 Arguments:
2726 cptr points to starting character value; updated
2727 d end value
2728 ocptr where to put start of othercase range
2729 odptr where to put end of othercase range
2730
2731 Yield: TRUE when range returned; FALSE when no more
2732 */
2733
2734 static BOOL
2735 get_othercase_range(unsigned int *cptr, unsigned int d, unsigned int *ocptr,
2736 unsigned int *odptr)
2737 {
2738 unsigned int c, othercase, next;
2739
2740 for (c = *cptr; c <= d; c++)
2741 { if ((othercase = UCD_OTHERCASE(c)) != c) break; }
2742
2743 if (c > d) return FALSE;
2744
2745 *ocptr = othercase;
2746 next = othercase + 1;
2747
2748 for (++c; c <= d; c++)
2749 {
2750 if (UCD_OTHERCASE(c) != next) break;
2751 next++;
2752 }
2753
2754 *odptr = next - 1;
2755 *cptr = c;
2756
2757 return TRUE;
2758 }
2759
2760
2761
2762 /*************************************************
2763 * Check a character and a property *
2764 *************************************************/
2765
2766 /* This function is called by check_auto_possessive() when a property item
2767 is adjacent to a fixed character.
2768
2769 Arguments:
2770 c the character
2771 ptype the property type
2772 pdata the data for the type
2773 negated TRUE if it's a negated property (\P or \p{^)
2774
2775 Returns: TRUE if auto-possessifying is OK
2776 */
2777
2778 static BOOL
2779 check_char_prop(int c, int ptype, int pdata, BOOL negated)
2780 {
2781 const ucd_record *prop = GET_UCD(c);
2782 switch(ptype)
2783 {
2784 case PT_LAMP:
2785 return (prop->chartype == ucp_Lu ||
2786 prop->chartype == ucp_Ll ||
2787 prop->chartype == ucp_Lt) == negated;
2788
2789 case PT_GC:
2790 return (pdata == PRIV(ucp_gentype)[prop->chartype]) == negated;
2791
2792 case PT_PC:
2793 return (pdata == prop->chartype) == negated;
2794
2795 case PT_SC:
2796 return (pdata == prop->script) == negated;
2797
2798 /* These are specials */
2799
2800 case PT_ALNUM:
2801 return (PRIV(ucp_gentype)[prop->chartype] == ucp_L ||
2802 PRIV(ucp_gentype)[prop->chartype] == ucp_N) == negated;
2803
2804 case PT_SPACE: /* Perl space */
2805 return (PRIV(ucp_gentype)[prop->chartype] == ucp_Z ||
2806 c == CHAR_HT || c == CHAR_NL || c == CHAR_FF || c == CHAR_CR)
2807 == negated;
2808
2809 case PT_PXSPACE: /* POSIX space */
2810 return (PRIV(ucp_gentype)[prop->chartype] == ucp_Z ||
2811 c == CHAR_HT || c == CHAR_NL || c == CHAR_VT ||
2812 c == CHAR_FF || c == CHAR_CR)
2813 == negated;
2814
2815 case PT_WORD:
2816 return (PRIV(ucp_gentype)[prop->chartype] == ucp_L ||
2817 PRIV(ucp_gentype)[prop->chartype] == ucp_N ||
2818 c == CHAR_UNDERSCORE) == negated;
2819 }
2820 return FALSE;
2821 }
2822 #endif /* SUPPORT_UCP */
2823
2824
2825
2826 /*************************************************
2827 * Check if auto-possessifying is possible *
2828 *************************************************/
2829
2830 /* This function is called for unlimited repeats of certain items, to see
2831 whether the next thing could possibly match the repeated item. If not, it makes
2832 sense to automatically possessify the repeated item.
2833
2834 Arguments:
2835 previous pointer to the repeated opcode
2836 utf8 TRUE in UTF-8 mode
2837 ptr next character in pattern
2838 options options bits
2839 cd contains pointers to tables etc.
2840
2841 Returns: TRUE if possessifying is wanted
2842 */
2843
2844 static BOOL
2845 check_auto_possessive(const pcre_uchar *previous, BOOL utf8,
2846 const pcre_uchar *ptr, int options, compile_data *cd)
2847 {
2848 int c, next;
2849 int op_code = *previous++;
2850
2851 /* Skip whitespace and comments in extended mode */
2852
2853 if ((options & PCRE_EXTENDED) != 0)
2854 {
2855 for (;;)
2856 {
2857 while ((cd->ctypes[*ptr] & ctype_space) != 0) ptr++;
2858 if (*ptr == CHAR_NUMBER_SIGN)
2859 {
2860 ptr++;
2861 while (*ptr != 0)
2862 {
2863 if (IS_NEWLINE(ptr)) { ptr += cd->nllen; break; }
2864 ptr++;
2865 #ifdef SUPPORT_UTF8
2866 if (utf8) while ((*ptr & 0xc0) == 0x80) ptr++;
2867 #endif
2868 }
2869 }
2870 else break;
2871 }
2872 }
2873
2874 /* If the next item is one that we can handle, get its value. A non-negative
2875 value is a character, a negative value is an escape value. */
2876
2877 if (*ptr == CHAR_BACKSLASH)
2878 {
2879 int temperrorcode = 0;
2880 next = check_escape(&ptr, &temperrorcode, cd->bracount, options, FALSE);
2881 if (temperrorcode != 0) return FALSE;
2882 ptr++; /* Point after the escape sequence */
2883 }
2884
2885 else if ((cd->ctypes[*ptr] & ctype_meta) == 0)
2886 {
2887 #ifdef SUPPORT_UTF8
2888 if (utf8) { GETCHARINC(next, ptr); } else
2889 #endif
2890 next = *ptr++;
2891 }
2892
2893 else return FALSE;
2894
2895 /* Skip whitespace and comments in extended mode */
2896
2897 if ((options & PCRE_EXTENDED) != 0)
2898 {
2899 for (;;)
2900 {
2901 while ((cd->ctypes[*ptr] & ctype_space) != 0) ptr++;
2902 if (*ptr == CHAR_NUMBER_SIGN)
2903 {
2904 ptr++;
2905 while (*ptr != 0)
2906 {
2907 if (IS_NEWLINE(ptr)) { ptr += cd->nllen; break; }
2908 ptr++;
2909 #ifdef SUPPORT_UTF8
2910 if (utf8) while ((*ptr & 0xc0) == 0x80) ptr++;
2911 #endif
2912 }
2913 }
2914 else break;
2915 }
2916 }
2917
2918 /* If the next thing is itself optional, we have to give up. */
2919
2920 if (*ptr == CHAR_ASTERISK || *ptr == CHAR_QUESTION_MARK ||
2921 STRNCMP_UC_C8(ptr, STR_LEFT_CURLY_BRACKET STR_0 STR_COMMA, 3) == 0)
2922 return FALSE;
2923
2924 /* Now compare the next item with the previous opcode. First, handle cases when
2925 the next item is a character. */
2926
2927 if (next >= 0) switch(op_code)
2928 {
2929 case OP_CHAR:
2930 #ifdef SUPPORT_UTF8
2931 GETCHARTEST(c, previous);
2932 #else
2933 c = *previous;
2934 #endif
2935 return c != next;
2936
2937 /* For CHARI (caseless character) we must check the other case. If we have
2938 Unicode property support, we can use it to test the other case of
2939 high-valued characters. */
2940
2941 case OP_CHARI:
2942 #ifdef SUPPORT_UTF8
2943 GETCHARTEST(c, previous);
2944 #else
2945 c = *previous;
2946 #endif
2947 if (c == next) return FALSE;
2948 #ifdef SUPPORT_UTF8
2949 if (utf8)
2950 {
2951 unsigned int othercase;
2952 if (next < 128) othercase = cd->fcc[next]; else
2953 #ifdef SUPPORT_UCP
2954 othercase = UCD_OTHERCASE((unsigned int)next);
2955 #else
2956 othercase = NOTACHAR;
2957 #endif
2958 return (unsigned int)c != othercase;
2959 }
2960 else
2961 #endif /* SUPPORT_UTF8 */
2962 return (c != cd->fcc[next]); /* Non-UTF-8 mode */
2963
2964 /* For OP_NOT and OP_NOTI, the data is always a single-byte character. These
2965 opcodes are not used for multi-byte characters, because they are coded using
2966 an XCLASS instead. */
2967
2968 case OP_NOT:
2969 return (c = *previous) == next;
2970
2971 case OP_NOTI:
2972 if ((c = *previous) == next) return TRUE;
2973 #ifdef SUPPORT_UTF8
2974 if (utf8)
2975 {
2976 unsigned int othercase;
2977 if (next < 128) othercase = cd->fcc[next]; else
2978 #ifdef SUPPORT_UCP
2979 othercase = UCD_OTHERCASE(next);
2980 #else
2981 othercase = NOTACHAR;
2982 #endif
2983 return (unsigned int)c == othercase;
2984 }
2985 else
2986 #endif /* SUPPORT_UTF8 */
2987 return (c == cd->fcc[next]); /* Non-UTF-8 mode */
2988
2989 /* Note that OP_DIGIT etc. are generated only when PCRE_UCP is *not* set.
2990 When it is set, \d etc. are converted into OP_(NOT_)PROP codes. */
2991
2992 case OP_DIGIT:
2993 return next > 127 || (cd->ctypes[next] & ctype_digit) == 0;
2994
2995 case OP_NOT_DIGIT:
2996 return next <= 127 && (cd->ctypes[next] & ctype_digit) != 0;
2997
2998 case OP_WHITESPACE:
2999 return next > 127 || (cd->ctypes[next] & ctype_space) == 0;
3000
3001 case OP_NOT_WHITESPACE:
3002 return next <= 127 && (cd->ctypes[next] & ctype_space) != 0;
3003
3004 case OP_WORDCHAR:
3005 return next > 127 || (cd->ctypes[next] & ctype_word) == 0;
3006
3007 case OP_NOT_WORDCHAR:
3008 return next <= 127 && (cd->ctypes[next] & ctype_word) != 0;
3009
3010 case OP_HSPACE:
3011 case OP_NOT_HSPACE:
3012 switch(next)
3013 {
3014 case 0x09:
3015 case 0x20:
3016 case 0xa0:
3017 case 0x1680:
3018 case 0x180e:
3019 case 0x2000:
3020 case 0x2001:
3021 case 0x2002:
3022 case 0x2003:
3023 case 0x2004:
3024 case 0x2005:
3025 case 0x2006:
3026 case 0x2007:
3027 case 0x2008:
3028 case 0x2009:
3029 case 0x200A:
3030 case 0x202f:
3031 case 0x205f:
3032 case 0x3000:
3033 return op_code == OP_NOT_HSPACE;
3034 default:
3035 return op_code != OP_NOT_HSPACE;
3036 }
3037
3038 case OP_ANYNL:
3039 case OP_VSPACE:
3040 case OP_NOT_VSPACE:
3041 switch(next)
3042 {
3043 case 0x0a:
3044 case 0x0b:
3045 case 0x0c:
3046 case 0x0d:
3047 case 0x85:
3048 case 0x2028:
3049 case 0x2029:
3050 return op_code == OP_NOT_VSPACE;
3051 default:
3052 return op_code != OP_NOT_VSPACE;
3053 }
3054
3055 #ifdef SUPPORT_UCP
3056 case OP_PROP:
3057 return check_char_prop(next, previous[0], previous[1], FALSE);
3058
3059 case OP_NOTPROP:
3060 return check_char_prop(next, previous[0], previous[1], TRUE);
3061 #endif
3062
3063 default:
3064 return FALSE;
3065 }
3066
3067
3068 /* Handle the case when the next item is \d, \s, etc. Note that when PCRE_UCP
3069 is set, \d turns into ESC_du rather than ESC_d, etc., so ESC_d etc. are
3070 generated only when PCRE_UCP is *not* set, that is, when only ASCII
3071 characteristics are recognized. Similarly, the opcodes OP_DIGIT etc. are
3072 replaced by OP_PROP codes when PCRE_UCP is set. */
3073
3074 switch(op_code)
3075 {
3076 case OP_CHAR:
3077 case OP_CHARI:
3078 #ifdef SUPPORT_UTF8
3079 GETCHARTEST(c, previous);
3080 #else
3081 c = *previous;
3082 #endif
3083 switch(-next)
3084 {
3085 case ESC_d:
3086 return c > 127 || (cd->ctypes[c] & ctype_digit) == 0;
3087
3088 case ESC_D:
3089 return c <= 127 && (cd->ctypes[c] & ctype_digit) != 0;
3090
3091 case ESC_s:
3092 return c > 127 || (cd->ctypes[c] & ctype_space) == 0;
3093
3094 case ESC_S:
3095 return c <= 127 && (cd->ctypes[c] & ctype_space) != 0;
3096
3097 case ESC_w:
3098 return c > 127 || (cd->ctypes[c] & ctype_word) == 0;
3099
3100 case ESC_W:
3101 return c <= 127 && (cd->ctypes[c] & ctype_word) != 0;
3102
3103 case ESC_h:
3104 case ESC_H:
3105 switch(c)
3106 {
3107 case 0x09:
3108 case 0x20:
3109 case 0xa0:
3110 case 0x1680:
3111 case 0x180e:
3112 case 0x2000:
3113 case 0x2001:
3114 case 0x2002:
3115 case 0x2003:
3116 case 0x2004:
3117 case 0x2005:
3118 case 0x2006:
3119 case 0x2007:
3120 case 0x2008:
3121 case 0x2009:
3122 case 0x200A:
3123 case 0x202f:
3124 case 0x205f:
3125 case 0x3000:
3126 return -next != ESC_h;
3127 default:
3128 return -next == ESC_h;
3129 }
3130
3131 case ESC_v:
3132 case ESC_V:
3133 switch(c)
3134 {
3135 case 0x0a:
3136 case 0x0b:
3137 case 0x0c:
3138 case 0x0d:
3139 case 0x85:
3140 case 0x2028:
3141 case 0x2029:
3142 return -next != ESC_v;
3143 default:
3144 return -next == ESC_v;
3145 }
3146
3147 /* When PCRE_UCP is set, these values get generated for \d etc. Find
3148 their substitutions and process them. The result will always be either
3149 -ESC_p or -ESC_P. Then fall through to process those values. */
3150
3151 #ifdef SUPPORT_UCP
3152 case ESC_du:
3153 case ESC_DU:
3154 case ESC_wu:
3155 case ESC_WU:
3156 case ESC_su:
3157 case ESC_SU:
3158 {
3159 int temperrorcode = 0;
3160 ptr = substitutes[-next - ESC_DU];
3161 next = check_escape(&ptr, &temperrorcode, 0, options, FALSE);
3162 if (temperrorcode != 0) return FALSE;
3163 ptr++; /* For compatibility */
3164 }
3165 /* Fall through */
3166
3167 case ESC_p:
3168 case ESC_P:
3169 {
3170 int ptype, pdata, errorcodeptr;
3171 BOOL negated;
3172
3173 ptr--; /* Make ptr point at the p or P */
3174 ptype = get_ucp(&ptr, &negated, &pdata, &errorcodeptr);
3175 if (ptype < 0) return FALSE;
3176 ptr++; /* Point past the final curly ket */
3177
3178 /* If the property item is optional, we have to give up. (When generated
3179 from \d etc by PCRE_UCP, this test will have been applied much earlier,
3180 to the original \d etc. At this point, ptr will point to a zero byte. */
3181
3182 if (*ptr == CHAR_ASTERISK || *ptr == CHAR_QUESTION_MARK ||
3183 STRNCMP_UC_C8(ptr, STR_LEFT_CURLY_BRACKET STR_0 STR_COMMA, 3) == 0)
3184 return FALSE;
3185
3186 /* Do the property check. */
3187
3188 return check_char_prop(c, ptype, pdata, (next == -ESC_P) != negated);
3189 }
3190 #endif
3191
3192 default:
3193 return FALSE;
3194 }
3195
3196 /* In principle, support for Unicode properties should be integrated here as
3197 well. It means re-organizing the above code so as to get hold of the property
3198 values before switching on the op-code. However, I wonder how many patterns
3199 combine ASCII \d etc with Unicode properties? (Note that if PCRE_UCP is set,
3200 these op-codes are never generated.) */
3201
3202 case OP_DIGIT:
3203 return next == -ESC_D || next == -ESC_s || next == -ESC_W ||
3204 next == -ESC_h || next == -ESC_v || next == -ESC_R;
3205
3206 case OP_NOT_DIGIT:
3207 return next == -ESC_d;
3208
3209 case OP_WHITESPACE:
3210 return next == -ESC_S || next == -ESC_d || next == -ESC_w || next == -ESC_R;
3211
3212 case OP_NOT_WHITESPACE:
3213 return next == -ESC_s || next == -ESC_h || next == -ESC_v;
3214
3215 case OP_HSPACE:
3216 return next == -ESC_S || next == -ESC_H || next == -ESC_d ||
3217 next == -ESC_w || next == -ESC_v || next == -ESC_R;
3218
3219 case OP_NOT_HSPACE:
3220 return next == -ESC_h;
3221
3222 /* Can't have \S in here because VT matches \S (Perl anomaly) */
3223 case OP_ANYNL:
3224 case OP_VSPACE:
3225 return next == -ESC_V || next == -ESC_d || next == -ESC_w;
3226
3227 case OP_NOT_VSPACE:
3228 return next == -ESC_v || next == -ESC_R;
3229
3230 case OP_WORDCHAR:
3231 return next == -ESC_W || next == -ESC_s || next == -ESC_h ||
3232 next == -ESC_v || next == -ESC_R;
3233
3234 case OP_NOT_WORDCHAR:
3235 return next == -ESC_w || next == -ESC_d;
3236
3237 default:
3238 return FALSE;
3239 }
3240
3241 /* Control does not reach here */
3242 }
3243
3244
3245
3246 /*************************************************
3247 * Compile one branch *
3248 *************************************************/
3249
3250 /* Scan the pattern, compiling it into the a vector. If the options are
3251 changed during the branch, the pointer is used to change the external options
3252 bits. This function is used during the pre-compile phase when we are trying
3253 to find out the amount of memory needed, as well as during the real compile
3254 phase. The value of lengthptr distinguishes the two phases.
3255
3256 Arguments:
3257 optionsptr pointer to the option bits
3258 codeptr points to the pointer to the current code point
3259 ptrptr points to the current pattern pointer
3260 errorcodeptr points to error code variable
3261 firstbyteptr set to initial literal character, or < 0 (REQ_UNSET, REQ_NONE)
3262 reqbyteptr set to the last literal character required, else < 0
3263 bcptr points to current branch chain
3264 cond_depth conditional nesting depth
3265 cd contains pointers to tables etc.
3266 lengthptr NULL during the real compile phase
3267 points to length accumulator during pre-compile phase
3268
3269 Returns: TRUE on success
3270 FALSE, with *errorcodeptr set non-zero on error
3271 */
3272
3273 static BOOL
3274 compile_branch(int *optionsptr, pcre_uchar **codeptr,
3275 const pcre_uchar **ptrptr, int *errorcodeptr, int *firstbyteptr,
3276 int *reqbyteptr, branch_chain *bcptr, int cond_depth, compile_data *cd,
3277 int *lengthptr)
3278 {
3279 int repeat_type, op_type;
3280 int repeat_min = 0, repeat_max = 0; /* To please picky compilers */
3281 int bravalue = 0;
3282 int greedy_default, greedy_non_default;
3283 int firstbyte, reqbyte;
3284 int zeroreqbyte, zerofirstbyte;
3285 int req_caseopt, reqvary, tempreqvary;
3286 int options = *optionsptr; /* May change dynamically */
3287 int after_manual_callout = 0;
3288 int length_prevgroup = 0;
3289 register int c;
3290 register pcre_uchar *code = *codeptr;
3291 pcre_uchar *last_code = code;
3292 pcre_uchar *orig_code = code;
3293 pcre_uchar *tempcode;
3294 BOOL inescq = FALSE;
3295 BOOL groupsetfirstbyte = FALSE;
3296 const pcre_uchar *ptr = *ptrptr;
3297 const pcre_uchar *tempptr;
3298 const pcre_uchar *nestptr = NULL;
3299 pcre_uchar *previous = NULL;
3300 pcre_uchar *previous_callout = NULL;
3301 pcre_uchar *save_hwm = NULL;
3302 pcre_uchar classbits[32];
3303
3304 /* We can fish out the UTF-8 setting once and for all into a BOOL, but we
3305 must not do this for other options (e.g. PCRE_EXTENDED) because they may change
3306 dynamically as we process the pattern. */
3307
3308 #ifdef SUPPORT_UTF8
3309 BOOL class_utf8;
3310 BOOL utf8 = (options & PCRE_UTF8) != 0;
3311 pcre_uint8 *class_utf8data;
3312 pcre_uint8 *class_utf8data_base;
3313 pcre_uint8 utf8_char[6];
3314 #else
3315 BOOL utf8 = FALSE;
3316 #endif
3317
3318 #ifdef PCRE_DEBUG
3319 if (lengthptr != NULL) DPRINTF((">> start branch\n"));
3320 #endif
3321
3322 /* Set up the default and non-default settings for greediness */
3323
3324 greedy_default = ((options & PCRE_UNGREEDY) != 0);
3325 greedy_non_default = greedy_default ^ 1;
3326
3327 /* Initialize no first byte, no required byte. REQ_UNSET means "no char
3328 matching encountered yet". It gets changed to REQ_NONE if we hit something that
3329 matches a non-fixed char first char; reqbyte just remains unset if we never
3330 find one.
3331
3332 When we hit a repeat whose minimum is zero, we may have to adjust these values
3333 to take the zero repeat into account. This is implemented by setting them to
3334 zerofirstbyte and zeroreqbyte when such a repeat is encountered. The individual
3335 item types that can be repeated set these backoff variables appropriately. */
3336
3337 firstbyte = reqbyte = zerofirstbyte = zeroreqbyte = REQ_UNSET;
3338
3339 /* The variable req_caseopt contains either the REQ_CASELESS value or zero,
3340 according to the current setting of the caseless flag. REQ_CASELESS is a bit
3341 value > 255. It is added into the firstbyte or reqbyte variables to record the
3342 case status of the value. This is used only for ASCII characters. */
3343
3344 req_caseopt = ((options & PCRE_CASELESS) != 0)? REQ_CASELESS : 0;
3345
3346 /* Switch on next character until the end of the branch */
3347
3348 for (;; ptr++)
3349 {
3350 BOOL negate_class;
3351 BOOL should_flip_negation;
3352 BOOL possessive_quantifier;
3353 BOOL is_quantifier;
3354 BOOL is_recurse;
3355 BOOL reset_bracount;
3356 int class_charcount;
3357 int class_lastchar;
3358 int newoptions;
3359 int recno;
3360 int refsign;
3361 int skipbytes;
3362 int subreqbyte;
3363 int subfirstbyte;
3364 int terminator;
3365 int mclength;
3366 int tempbracount;
3367 pcre_uchar mcbuffer[8];
3368
3369 /* Get next byte in the pattern */
3370
3371 c = *ptr;
3372
3373 /* If we are at the end of a nested substitution, revert to the outer level
3374 string. Nesting only happens one level deep. */
3375
3376 if (c == 0 && nestptr != NULL)
3377 {
3378 ptr = nestptr;
3379 nestptr = NULL;
3380 c = *ptr;
3381 }
3382
3383 /* If we are in the pre-compile phase, accumulate the length used for the
3384 previous cycle of this loop. */
3385
3386 if (lengthptr != NULL)
3387 {
3388 #ifdef PCRE_DEBUG
3389 if (code > cd->hwm) cd->hwm = code; /* High water info */
3390 #endif
3391 if (code > cd->start_workspace + WORK_SIZE_CHECK) /* Check for overrun */
3392 {
3393 *errorcodeptr = ERR52;
3394 goto FAILED;
3395 }
3396
3397 /* There is at least one situation where code goes backwards: this is the
3398 case of a zero quantifier after a class (e.g. [ab]{0}). At compile time,
3399 the class is simply eliminated. However, it is created first, so we have to
3400 allow memory for it. Therefore, don't ever reduce the length at this point.
3401 */
3402
3403 if (code < last_code) code = last_code;
3404
3405 /* Paranoid check for integer overflow */
3406
3407 if (OFLOW_MAX - *lengthptr < code - last_code)
3408 {
3409 *errorcodeptr = ERR20;
3410 goto FAILED;
3411 }
3412
3413 *lengthptr += (int)(code - last_code);
3414 DPRINTF(("length=%d added %d c=%c\n", *lengthptr, (int)(code - last_code),
3415 c));
3416
3417 /* If "previous" is set and it is not at the start of the work space, move
3418 it back to there, in order to avoid filling up the work space. Otherwise,
3419 if "previous" is NULL, reset the current code pointer to the start. */
3420
3421 if (previous != NULL)
3422 {
3423 if (previous > orig_code)
3424 {
3425 memmove(orig_code, previous, IN_UCHARS(code - previous));
3426 code -= previous - orig_code;
3427 previous = orig_code;
3428 }
3429 }
3430 else code = orig_code;
3431
3432 /* Remember where this code item starts so we can pick up the length
3433 next time round. */
3434
3435 last_code = code;
3436 }
3437
3438 /* In the real compile phase, just check the workspace used by the forward
3439 reference list. */
3440
3441 else if (cd->hwm > cd->start_workspace + WORK_SIZE_CHECK)
3442 {
3443 *errorcodeptr = ERR52;
3444 goto FAILED;
3445 }
3446
3447 /* If in \Q...\E, check for the end; if not, we have a literal */
3448
3449 if (inescq && c != 0)
3450 {
3451 if (c == CHAR_BACKSLASH && ptr[1] == CHAR_E)
3452 {
3453 inescq = FALSE;
3454 ptr++;
3455 continue;
3456 }
3457 else
3458 {
3459 if (previous_callout != NULL)
3460 {
3461 if (lengthptr == NULL) /* Don't attempt in pre-compile phase */
3462 complete_callout(previous_callout, ptr, cd);
3463 previous_callout = NULL;
3464 }
3465 if ((options & PCRE_AUTO_CALLOUT) != 0)
3466 {
3467 previous_callout = code;
3468 code = auto_callout(code, ptr, cd);
3469 }
3470 goto NORMAL_CHAR;
3471 }
3472 }
3473
3474 /* Fill in length of a previous callout, except when the next thing is
3475 a quantifier. */
3476
3477 is_quantifier =
3478 c == CHAR_ASTERISK || c == CHAR_PLUS || c == CHAR_QUESTION_MARK ||
3479 (c == CHAR_LEFT_CURLY_BRACKET && is_counted_repeat(ptr+1));
3480
3481 if (!is_quantifier && previous_callout != NULL &&
3482 after_manual_callout-- <= 0)
3483 {
3484 if (lengthptr == NULL) /* Don't attempt in pre-compile phase */
3485 complete_callout(previous_callout, ptr, cd);
3486 previous_callout = NULL;
3487 }
3488
3489 /* In extended mode, skip white space and comments. */
3490
3491 if ((options & PCRE_EXTENDED) != 0)
3492 {
3493 if ((cd->ctypes[c] & ctype_space) != 0) continue;
3494 if (c == CHAR_NUMBER_SIGN)
3495 {
3496 ptr++;
3497 while (*ptr != 0)
3498 {
3499 if (IS_NEWLINE(ptr)) { ptr += cd->nllen - 1; break; }
3500 ptr++;
3501 #ifdef SUPPORT_UTF8
3502 if (utf8) while ((*ptr & 0xc0) == 0x80) ptr++;
3503 #endif
3504 }
3505 if (*ptr != 0) continue;
3506
3507 /* Else fall through to handle end of string */
3508 c = 0;
3509 }
3510 }
3511
3512 /* No auto callout for quantifiers. */
3513
3514 if ((options & PCRE_AUTO_CALLOUT) != 0 && !is_quantifier)
3515 {
3516 previous_callout = code;
3517 code = auto_callout(code, ptr, cd);
3518 }
3519
3520 switch(c)
3521 {
3522 /* ===================================================================*/
3523 case 0: /* The branch terminates at string end */
3524 case CHAR_VERTICAL_LINE: /* or | or ) */
3525 case CHAR_RIGHT_PARENTHESIS:
3526 *firstbyteptr = firstbyte;
3527 *reqbyteptr = reqbyte;
3528 *codeptr = code;
3529 *ptrptr = ptr;
3530 if (lengthptr != NULL)
3531 {
3532 if (OFLOW_MAX - *lengthptr < code - last_code)
3533 {
3534 *errorcodeptr = ERR20;
3535 goto FAILED;
3536 }
3537 *lengthptr += (int)(code - last_code); /* To include callout length */
3538 DPRINTF((">> end branch\n"));
3539 }
3540 return TRUE;
3541
3542
3543 /* ===================================================================*/
3544 /* Handle single-character metacharacters. In multiline mode, ^ disables
3545 the setting of any following char as a first character. */
3546
3547 case CHAR_CIRCUMFLEX_ACCENT:
3548 previous = NULL;
3549 if ((options & PCRE_MULTILINE) != 0)
3550 {
3551 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
3552 *code++ = OP_CIRCM;
3553 }
3554 else *code++ = OP_CIRC;
3555 break;
3556
3557 case CHAR_DOLLAR_SIGN:
3558 previous = NULL;
3559 *code++ = ((options & PCRE_MULTILINE) != 0)? OP_DOLLM : OP_DOLL;
3560 break;
3561
3562 /* There can never be a first char if '.' is first, whatever happens about
3563 repeats. The value of reqbyte doesn't change either. */
3564
3565 case CHAR_DOT:
3566 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
3567 zerofirstbyte = firstbyte;
3568 zeroreqbyte = reqbyte;
3569 previous = code;
3570 *code++ = ((options & PCRE_DOTALL) != 0)? OP_ALLANY: OP_ANY;
3571 break;
3572
3573
3574 /* ===================================================================*/
3575 /* Character classes. If the included characters are all < 256, we build a
3576 32-byte bitmap of the permitted characters, except in the special case
3577 where there is only one such character. For negated classes, we build the
3578 map as usual, then invert it at the end. However, we use a different opcode
3579 so that data characters > 255 can be handled correctly.
3580
3581 If the class contains characters outside the 0-255 range, a different
3582 opcode is compiled. It may optionally have a bit map for characters < 256,
3583 but those above are are explicitly listed afterwards. A flag byte tells
3584 whether the bitmap is present, and whether this is a negated class or not.
3585
3586 In JavaScript compatibility mode, an isolated ']' causes an error. In
3587 default (Perl) mode, it is treated as a data character. */
3588
3589 case CHAR_RIGHT_SQUARE_BRACKET:
3590 if ((cd->external_options & PCRE_JAVASCRIPT_COMPAT) != 0)
3591 {
3592 *errorcodeptr = ERR64;
3593 goto FAILED;
3594 }
3595 goto NORMAL_CHAR;
3596
3597 case CHAR_LEFT_SQUARE_BRACKET:
3598 previous = code;
3599
3600 /* PCRE supports POSIX class stuff inside a class. Perl gives an error if
3601 they are encountered at the top level, so we'll do that too. */
3602
3603 if ((ptr[1] == CHAR_COLON || ptr[1] == CHAR_DOT ||
3604 ptr[1] == CHAR_EQUALS_SIGN) &&
3605 check_posix_syntax(ptr, &tempptr))
3606 {
3607 *errorcodeptr = (ptr[1] == CHAR_COLON)? ERR13 : ERR31;
3608 goto FAILED;
3609 }
3610
3611 /* If the first character is '^', set the negation flag and skip it. Also,
3612 if the first few characters (either before or after ^) are \Q\E or \E we
3613 skip them too. This makes for compatibility with Perl. */
3614
3615 negate_class = FALSE;
3616 for (;;)
3617 {
3618 c = *(++ptr);
3619 if (c == CHAR_BACKSLASH)
3620 {
3621 if (ptr[1] == CHAR_E)
3622 ptr++;
3623 else if (STRNCMP_UC_C8(ptr + 1,
3624 STR_Q STR_BACKSLASH STR_E, 3) == 0)
3625 ptr += 3;
3626 else
3627 break;
3628 }
3629 else if (!negate_class && c == CHAR_CIRCUMFLEX_ACCENT)
3630 negate_class = TRUE;
3631 else break;
3632 }
3633
3634 /* Empty classes are allowed in JavaScript compatibility mode. Otherwise,
3635 an initial ']' is taken as a data character -- the code below handles
3636 that. In JS mode, [] must always fail, so generate OP_FAIL, whereas
3637 [^] must match any character, so generate OP_ALLANY. */
3638
3639 if (c == CHAR_RIGHT_SQUARE_BRACKET &&
3640 (cd->external_options & PCRE_JAVASCRIPT_COMPAT) != 0)
3641 {
3642 *code++ = negate_class? OP_ALLANY : OP_FAIL;
3643 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
3644 zerofirstbyte = firstbyte;
3645 break;
3646 }
3647
3648 /* If a class contains a negative special such as \S, we need to flip the
3649 negation flag at the end, so that support for characters > 255 works
3650 correctly (they are all included in the class). */
3651
3652 should_flip_negation = FALSE;
3653
3654 /* Keep a count of chars with values < 256 so that we can optimize the case
3655 of just a single character (as long as it's < 256). However, For higher
3656 valued UTF-8 characters, we don't yet do any optimization. */
3657
3658 class_charcount = 0;
3659 class_lastchar = -1;
3660
3661 /* Initialize the 32-char bit map to all zeros. We build the map in a
3662 temporary bit of memory, in case the class contains only 1 character (less
3663 than 256), because in that case the compiled code doesn't use the bit map.
3664 */
3665
3666 memset(classbits, 0, 32 * sizeof(pcre_uint8));
3667
3668 #ifdef SUPPORT_UTF8
3669 class_utf8 = FALSE; /* No chars >= 256 */
3670 class_utf8data = code + LINK_SIZE + 2; /* For UTF-8 items */
3671 class_utf8data_base = class_utf8data; /* For resetting in pass 1 */
3672 #endif
3673
3674 /* Process characters until ] is reached. By writing this as a "do" it
3675 means that an initial ] is taken as a data character. At the start of the
3676 loop, c contains the first byte of the character. */
3677
3678 if (c != 0) do
3679 {
3680 const pcre_uchar *oldptr;
3681
3682 #ifdef SUPPORT_UTF8
3683 if (utf8 && c > 127)
3684 { /* Braces are required because the */
3685 GETCHARLEN(c, ptr, ptr); /* macro generates multiple statements */
3686 }
3687
3688 /* In the pre-compile phase, accumulate the length of any UTF-8 extra
3689 data and reset the pointer. This is so that very large classes that
3690 contain a zillion UTF-8 characters no longer overwrite the work space
3691 (which is on the stack). */
3692
3693 if (lengthptr != NULL)
3694 {
3695 *lengthptr += class_utf8data - class_utf8data_base;
3696 class_utf8data = class_utf8data_base;
3697 }
3698
3699 #endif
3700
3701 /* Inside \Q...\E everything is literal except \E */
3702
3703 if (inescq)
3704 {
3705 if (c == CHAR_BACKSLASH && ptr[1] == CHAR_E) /* If we are at \E */
3706 {
3707 inescq = FALSE; /* Reset literal state */
3708 ptr++; /* Skip the 'E' */
3709 continue; /* Carry on with next */
3710 }
3711 goto CHECK_RANGE; /* Could be range if \E follows */
3712 }
3713
3714 /* Handle POSIX class names. Perl allows a negation extension of the
3715 form [:^name:]. A square bracket that doesn't match the syntax is
3716 treated as a literal. We also recognize the POSIX constructions
3717 [.ch.] and [=ch=] ("collating elements") and fault them, as Perl
3718 5.6 and 5.8 do. */
3719
3720 if (c == CHAR_LEFT_SQUARE_BRACKET &&
3721 (ptr[1] == CHAR_COLON || ptr[1] == CHAR_DOT ||
3722 ptr[1] == CHAR_EQUALS_SIGN) && check_posix_syntax(ptr, &tempptr))
3723 {
3724 BOOL local_negate = FALSE;
3725 int posix_class, taboffset, tabopt;
3726 register const pcre_uint8 *cbits = cd->cbits;
3727 pcre_uint8 pbits[32];
3728
3729 if (ptr[1] != CHAR_COLON)
3730 {
3731 *errorcodeptr = ERR31;
3732 goto FAILED;
3733 }
3734
3735 ptr += 2;
3736 if (*ptr == CHAR_CIRCUMFLEX_ACCENT)
3737 {
3738 local_negate = TRUE;
3739 should_flip_negation = TRUE; /* Note negative special */
3740 ptr++;
3741 }
3742
3743 posix_class = check_posix_name(ptr, (int)(tempptr - ptr));
3744 if (posix_class < 0)
3745 {
3746 *errorcodeptr = ERR30;
3747 goto FAILED;
3748 }
3749
3750 /* If matching is caseless, upper and lower are converted to
3751 alpha. This relies on the fact that the class table starts with
3752 alpha, lower, upper as the first 3 entries. */
3753
3754 if ((options & PCRE_CASELESS) != 0 && posix_class <= 2)
3755 posix_class = 0;
3756
3757 /* When PCRE_UCP is set, some of the POSIX classes are converted to
3758 different escape sequences that use Unicode properties. */
3759
3760 #ifdef SUPPORT_UCP
3761 if ((options & PCRE_UCP) != 0)
3762 {
3763 int pc = posix_class + ((local_negate)? POSIX_SUBSIZE/2 : 0);
3764 if (posix_substitutes[pc] != NULL)
3765 {
3766 nestptr = tempptr + 1;
3767 ptr = posix_substitutes[pc] - 1;
3768 continue;
3769 }
3770 }
3771 #endif
3772 /* In the non-UCP case, we build the bit map for the POSIX class in a
3773 chunk of local store because we may be adding and subtracting from it,
3774 and we don't want to subtract bits that may be in the main map already.
3775 At the end we or the result into the bit map that is being built. */
3776
3777 posix_class *= 3;
3778
3779 /* Copy in the first table (always present) */
3780
3781 memcpy(pbits, cbits + posix_class_maps[posix_class],
3782 32 * sizeof(pcre_uint8));
3783
3784 /* If there is a second table, add or remove it as required. */
3785
3786 taboffset = posix_class_maps[posix_class + 1];
3787 tabopt = posix_class_maps[posix_class + 2];
3788
3789 if (taboffset >= 0)
3790 {
3791 if (tabopt >= 0)
3792 for (c = 0; c < 32; c++) pbits[c] |= cbits[c + taboffset];
3793 else
3794 for (c = 0; c < 32; c++) pbits[c] &= ~cbits[c + taboffset];
3795 }
3796
3797 /* Not see if we need to remove any special characters. An option
3798 value of 1 removes vertical space and 2 removes underscore. */
3799
3800 if (tabopt < 0) tabopt = -tabopt;
3801 if (tabopt == 1) pbits[1] &= ~0x3c;
3802 else if (tabopt == 2) pbits[11] &= 0x7f;
3803
3804 /* Add the POSIX table or its complement into the main table that is
3805 being built and we are done. */
3806
3807 if (local_negate)
3808 for (c = 0; c < 32; c++) classbits[c] |= ~pbits[c];
3809 else
3810 for (c = 0; c < 32; c++) classbits[c] |= pbits[c];
3811
3812 ptr = tempptr + 1;
3813 class_charcount = 10; /* Set > 1; assumes more than 1 per class */
3814 continue; /* End of POSIX syntax handling */
3815 }
3816
3817 /* Backslash may introduce a single character, or it may introduce one
3818 of the specials, which just set a flag. The sequence \b is a special
3819 case. Inside a class (and only there) it is treated as backspace. We
3820 assume that other escapes have more than one character in them, so set
3821 class_charcount bigger than one. Unrecognized escapes fall through and
3822 are either treated as literal characters (by default), or are faulted if
3823 PCRE_EXTRA is set. */
3824
3825 if (c == CHAR_BACKSLASH)
3826 {
3827 c = check_escape(&ptr, errorcodeptr, cd->bracount, options, TRUE);
3828 if (*errorcodeptr != 0) goto FAILED;
3829
3830 if (-c == ESC_b) c = CHAR_BS; /* \b is backspace in a class */
3831 else if (-c == ESC_Q) /* Handle start of quoted string */
3832 {
3833 if (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
3834 {
3835 ptr += 2; /* avoid empty string */
3836 }
3837 else inescq = TRUE;
3838 continue;
3839 }
3840 else if (-c == ESC_E) continue; /* Ignore orphan \E */
3841
3842 if (c < 0)
3843 {
3844 register const pcre_uint8 *cbits = cd->cbits;
3845 class_charcount += 2; /* Greater than 1 is what matters */
3846
3847 switch (-c)
3848 {
3849 #ifdef SUPPORT_UCP
3850 case ESC_du: /* These are the values given for \d etc */
3851 case ESC_DU: /* when PCRE_UCP is set. We replace the */
3852 case ESC_wu: /* escape sequence with an appropriate \p */
3853 case ESC_WU: /* or \P to test Unicode properties instead */
3854 case ESC_su: /* of the default ASCII testing. */
3855 case ESC_SU:
3856 nestptr = ptr;
3857 ptr = substitutes[-c - ESC_DU] - 1; /* Just before substitute */
3858 class_charcount -= 2; /* Undo! */
3859 continue;
3860 #endif
3861 case ESC_d:
3862 for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_digit];
3863 continue;
3864
3865 case ESC_D:
3866 should_flip_negation = TRUE;
3867 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_digit];
3868 continue;
3869
3870 case ESC_w:
3871 for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_word];
3872 continue;
3873
3874 case ESC_W:
3875 should_flip_negation = TRUE;
3876 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_word];
3877 continue;
3878
3879 /* Perl 5.004 onwards omits VT from \s, but we must preserve it
3880 if it was previously set by something earlier in the character
3881 class. */
3882
3883 case ESC_s:
3884 classbits[0] |= cbits[cbit_space];
3885 classbits[1] |= cbits[cbit_space+1] & ~0x08;
3886 for (c = 2; c < 32; c++) classbits[c] |= cbits[c+cbit_space];
3887 continue;
3888
3889 case ESC_S:
3890 should_flip_negation = TRUE;
3891 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_space];
3892 classbits[1] |= 0x08; /* Perl 5.004 onwards omits VT from \s */
3893 continue;
3894
3895 case ESC_h:
3896 SETBIT(classbits, 0x09); /* VT */
3897 SETBIT(classbits, 0x20); /* SPACE */
3898 SETBIT(classbits, 0xa0); /* NSBP */
3899 #ifdef SUPPORT_UTF8
3900 if (utf8)
3901 {
3902 class_utf8 = TRUE;
3903 *class_utf8data++ = XCL_SINGLE;
3904 class_utf8data += PRIV(ord2utf8)(0x1680, class_utf8data);
3905 *class_utf8data++ = XCL_SINGLE;
3906 class_utf8data += PRIV(ord2utf8)(0x180e, class_utf8data);
3907 *class_utf8data++ = XCL_RANGE;
3908 class_utf8data += PRIV(ord2utf8)(0x2000, class_utf8data);
3909 class_utf8data += PRIV(ord2utf8)(0x200A, class_utf8data);
3910 *class_utf8data++ = XCL_SINGLE;
3911 class_utf8data += PRIV(ord2utf8)(0x202f, class_utf8data);
3912 *class_utf8data++ = XCL_SINGLE;
3913 class_utf8data += PRIV(ord2utf8)(0x205f, class_utf8data);
3914 *class_utf8data++ = XCL_SINGLE;
3915 class_utf8data += PRIV(ord2utf8)(0x3000, class_utf8data);
3916 }
3917 #endif
3918 continue;
3919
3920 case ESC_H:
3921 for (c = 0; c < 32; c++)
3922 {
3923 int x = 0xff;
3924 switch (c)
3925 {
3926 case 0x09/8: x ^= 1 << (0x09%8); break;
3927 case 0x20/8: x ^= 1 << (0x20%8); break;
3928 case 0xa0/8: x ^= 1 << (0xa0%8); break;
3929 default: break;
3930 }
3931 classbits[c] |= x;
3932 }
3933
3934 #ifdef SUPPORT_UTF8
3935 if (utf8)
3936 {
3937 class_utf8 = TRUE;
3938 *class_utf8data++ = XCL_RANGE;
3939 class_utf8data += PRIV(ord2utf8)(0x0100, class_utf8data);
3940 class_utf8data += PRIV(ord2utf8)(0x167f, class_utf8data);
3941 *class_utf8data++ = XCL_RANGE;
3942 class_utf8data += PRIV(ord2utf8)(0x1681, class_utf8data);
3943 class_utf8data += PRIV(ord2utf8)(0x180d, class_utf8data);
3944 *class_utf8data++ = XCL_RANGE;
3945 class_utf8data += PRIV(ord2utf8)(0x180f, class_utf8data);
3946 class_utf8data += PRIV(ord2utf8)(0x1fff, class_utf8data);
3947 *class_utf8data++ = XCL_RANGE;
3948 class_utf8data += PRIV(ord2utf8)(0x200B, class_utf8data);
3949 class_utf8data += PRIV(ord2utf8)(0x202e, class_utf8data);
3950 *class_utf8data++ = XCL_RANGE;
3951 class_utf8data += PRIV(ord2utf8)(0x2030, class_utf8data);
3952 class_utf8data += PRIV(ord2utf8)(0x205e, class_utf8data);
3953 *class_utf8data++ = XCL_RANGE;
3954 class_utf8data += PRIV(ord2utf8)(0x2060, class_utf8data);
3955 class_utf8data += PRIV(ord2utf8)(0x2fff, class_utf8data);
3956 *class_utf8data++ = XCL_RANGE;
3957 class_utf8data += PRIV(ord2utf8)(0x3001, class_utf8data);
3958 class_utf8data += PRIV(ord2utf8)(0x7fffffff, class_utf8data);
3959 }
3960 #endif
3961 continue;
3962
3963 case ESC_v:
3964 SETBIT(classbits, 0x0a); /* LF */
3965 SETBIT(classbits, 0x0b); /* VT */
3966 SETBIT(classbits, 0x0c); /* FF */
3967 SETBIT(classbits, 0x0d); /* CR */
3968 SETBIT(classbits, 0x85); /* NEL */
3969 #ifdef SUPPORT_UTF8
3970 if (utf8)
3971 {
3972 class_utf8 = TRUE;
3973 *class_utf8data++ = XCL_RANGE;
3974 class_utf8data += PRIV(ord2utf8)(0x2028, class_utf8data);
3975 class_utf8data += PRIV(ord2utf8)(0x2029, class_utf8data);
3976 }
3977 #endif
3978 continue;
3979
3980 case ESC_V:
3981 for (c = 0; c < 32; c++)
3982 {
3983 int x = 0xff;
3984 switch (c)
3985 {
3986 case 0x0a/8: x ^= 1 << (0x0a%8);
3987 x ^= 1 << (0x0b%8);
3988 x ^= 1 << (0x0c%8);
3989 x ^= 1 << (0x0d%8);
3990 break;
3991 case 0x85/8: x ^= 1 << (0x85%8); break;
3992 default: break;
3993 }
3994 classbits[c] |= x;
3995 }
3996
3997 #ifdef SUPPORT_UTF8
3998 if (utf8)
3999 {
4000 class_utf8 = TRUE;
4001 *class_utf8data++ = XCL_RANGE;
4002 class_utf8data += PRIV(ord2utf8)(0x0100, class_utf8data);
4003 class_utf8data += PRIV(ord2utf8)(0x2027, class_utf8data);
4004 *class_utf8data++ = XCL_RANGE;
4005 class_utf8data += PRIV(ord2utf8)(0x2029, class_utf8data);
4006 class_utf8data += PRIV(ord2utf8)(0x7fffffff, class_utf8data);
4007 }
4008 #endif
4009 continue;
4010
4011 #ifdef SUPPORT_UCP
4012 case ESC_p:
4013 case ESC_P:
4014 {
4015 BOOL negated;
4016 int pdata;
4017 int ptype = get_ucp(&ptr, &negated, &pdata, errorcodeptr);
4018 if (ptype < 0) goto FAILED;
4019 class_utf8 = TRUE;
4020 *class_utf8data++ = ((-c == ESC_p) != negated)?
4021 XCL_PROP : XCL_NOTPROP;
4022 *class_utf8data++ = ptype;
4023 *class_utf8data++ = pdata;
4024 class_charcount -= 2; /* Not a < 256 character */
4025 continue;
4026 }
4027 #endif
4028 /* Unrecognized escapes are faulted if PCRE is running in its
4029 strict mode. By default, for compatibility with Perl, they are
4030 treated as literals. */
4031
4032 default:
4033 if ((options & PCRE_EXTRA) != 0)
4034 {
4035 *errorcodeptr = ERR7;
4036 goto FAILED;
4037 }
4038 class_charcount -= 2; /* Undo the default count from above */
4039 c = *ptr; /* Get the final character and fall through */
4040 break;
4041 }
4042 }
4043
4044 /* Fall through if we have a single character (c >= 0). This may be
4045 greater than 256 in UTF-8 mode. */
4046
4047 } /* End of backslash handling */
4048
4049 /* A single character may be followed by '-' to form a range. However,
4050 Perl does not permit ']' to be the end of the range. A '-' character
4051 at the end is treated as a literal. Perl ignores orphaned \E sequences
4052 entirely. The code for handling \Q and \E is messy. */
4053
4054 CHECK_RANGE:
4055 while (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
4056 {
4057 inescq = FALSE;
4058 ptr += 2;
4059 }
4060
4061 oldptr = ptr;
4062
4063 /* Remember \r or \n */
4064
4065 if (c == CHAR_CR || c == CHAR_NL) cd->external_flags |= PCRE_HASCRORLF;
4066
4067 /* Check for range */
4068
4069 if (!inescq && ptr[1] == CHAR_MINUS)
4070 {
4071 int d;
4072 ptr += 2;
4073 while (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_E) ptr += 2;
4074
4075 /* If we hit \Q (not followed by \E) at this point, go into escaped
4076 mode. */
4077
4078 while (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_Q)
4079 {
4080 ptr += 2;
4081 if (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_E)
4082 { ptr += 2; continue; }
4083 inescq = TRUE;
4084 break;
4085 }
4086
4087 if (*ptr == 0 || (!inescq && *ptr == CHAR_RIGHT_SQUARE_BRACKET))
4088 {
4089 ptr = oldptr;
4090 goto LONE_SINGLE_CHARACTER;
4091 }
4092
4093 #ifdef SUPPORT_UTF8
4094 if (utf8)
4095 { /* Braces are required because the */
4096 GETCHARLEN(d, ptr, ptr); /* macro generates multiple statements */
4097 }
4098 else
4099 #endif
4100 d = *ptr; /* Not UTF-8 mode */
4101
4102 /* The second part of a range can be a single-character escape, but
4103 not any of the other escapes. Perl 5.6 treats a hyphen as a literal
4104 in such circumstances. */
4105
4106 if (!inescq && d == CHAR_BACKSLASH)
4107 {
4108 d = check_escape(&ptr, errorcodeptr, cd->bracount, options, TRUE);
4109 if (*errorcodeptr != 0) goto FAILED;
4110
4111 /* \b is backspace; any other special means the '-' was literal */
4112
4113 if (d < 0)
4114 {
4115 if (d == -ESC_b) d = CHAR_BS; else
4116 {
4117 ptr = oldptr;
4118 goto LONE_SINGLE_CHARACTER; /* A few lines below */
4119 }
4120 }
4121 }
4122
4123 /* Check that the two values are in the correct order. Optimize
4124 one-character ranges */
4125
4126 if (d < c)
4127 {
4128 *errorcodeptr = ERR8;
4129 goto FAILED;
4130 }
4131
4132 if (d == c) goto LONE_SINGLE_CHARACTER; /* A few lines below */
4133
4134 /* Remember \r or \n */
4135
4136 if (d == CHAR_CR || d == CHAR_NL) cd->external_flags |= PCRE_HASCRORLF;
4137
4138 /* In UTF-8 mode, if the upper limit is > 255, or > 127 for caseless
4139 matching, we have to use an XCLASS with extra data items. Caseless
4140 matching for characters > 127 is available only if UCP support is
4141 available. */
4142
4143 #ifdef SUPPORT_UTF8
4144 if (utf8 && (d > 255 || ((options & PCRE_CASELESS) != 0 && d > 127)))
4145 {
4146 class_utf8 = TRUE;
4147
4148 /* With UCP support, we can find the other case equivalents of
4149 the relevant characters. There may be several ranges. Optimize how
4150 they fit with the basic range. */
4151
4152 #ifdef SUPPORT_UCP
4153 if ((options & PCRE_CASELESS) != 0)
4154 {
4155 unsigned int occ, ocd;
4156 unsigned int cc = c;
4157 unsigned int origd = d;
4158 while (get_othercase_range(&cc, origd, &occ, &ocd))
4159 {
4160 if (occ >= (unsigned int)c &&
4161 ocd <= (unsigned int)d)
4162 continue; /* Skip embedded ranges */
4163
4164 if (occ < (unsigned int)c &&
4165 ocd >= (unsigned int)c - 1) /* Extend the basic range */
4166 { /* if there is overlap, */
4167 c = occ; /* noting that if occ < c */
4168 continue; /* we can't have ocd > d */
4169 } /* because a subrange is */
4170 if (ocd > (unsigned int)d &&
4171 occ <= (unsigned int)d + 1) /* always shorter than */
4172 { /* the basic range. */
4173 d = ocd;
4174 continue;
4175 }
4176
4177 if (occ == ocd)
4178 {
4179 *class_utf8data++ = XCL_SINGLE;
4180 }
4181 else
4182 {
4183 *class_utf8data++ = XCL_RANGE;
4184 class_utf8data += PRIV(ord2utf8)(occ, class_utf8data);
4185 }
4186 class_utf8data += PRIV(ord2utf8)(ocd, class_utf8data);
4187 }
4188 }
4189 #endif /* SUPPORT_UCP */
4190
4191 /* Now record the original range, possibly modified for UCP caseless
4192 overlapping ranges. */
4193
4194 *class_utf8data++ = XCL_RANGE;
4195 class_utf8data += PRIV(ord2utf8)(c, class_utf8data);
4196 class_utf8data += PRIV(ord2utf8)(d, class_utf8data);
4197
4198 /* With UCP support, we are done. Without UCP support, there is no
4199 caseless matching for UTF-8 characters > 127; we can use the bit map
4200 for the smaller ones. */
4201
4202 #ifdef SUPPORT_UCP
4203 continue; /* With next character in the class */
4204 #else
4205 if ((options & PCRE_CASELESS) == 0 || c > 127) continue;
4206
4207 /* Adjust upper limit and fall through to set up the map */
4208
4209 d = 127;
4210
4211 #endif /* SUPPORT_UCP */
4212 }
4213 #endif /* SUPPORT_UTF8 */
4214
4215 /* We use the bit map for all cases when not in UTF-8 mode; else
4216 ranges that lie entirely within 0-127 when there is UCP support; else
4217 for partial ranges without UCP support. */
4218
4219 class_charcount += d - c + 1;
4220 class_lastchar = d;
4221
4222 /* We can save a bit of time by skipping this in the pre-compile. */
4223
4224 if (lengthptr == NULL) for (; c <= d; c++)
4225 {
4226 classbits[c/8] |= (1 << (c&7));
4227 if ((options & PCRE_CASELESS) != 0)
4228 {
4229 int uc = cd->fcc[c]; /* flip case */
4230 classbits[uc/8] |= (1 << (uc&7));
4231 }
4232 }
4233
4234 continue; /* Go get the next char in the class */
4235 }
4236
4237 /* Handle a lone single character - we can get here for a normal
4238 non-escape char, or after \ that introduces a single character or for an
4239 apparent range that isn't. */
4240
4241 LONE_SINGLE_CHARACTER:
4242
4243 /* Handle a character that cannot go in the bit map */
4244
4245 #ifdef SUPPORT_UTF8
4246 if (utf8 && (c > 255 || ((options & PCRE_CASELESS) != 0 && c > 127)))
4247 {
4248 class_utf8 = TRUE;
4249 *class_utf8data++ = XCL_SINGLE;
4250 class_utf8data += PRIV(ord2utf8)(c, class_utf8data);
4251
4252 #ifdef SUPPORT_UCP
4253 if ((options & PCRE_CASELESS) != 0)
4254 {
4255 unsigned int othercase;
4256 if ((othercase = UCD_OTHERCASE(c)) != c)
4257 {
4258 *class_utf8data++ = XCL_SINGLE;
4259 class_utf8data += PRIV(ord2utf8)(othercase, class_utf8data);
4260 }
4261 }
4262 #endif /* SUPPORT_UCP */
4263
4264 }
4265 else
4266 #endif /* SUPPORT_UTF8 */
4267
4268 /* Handle a single-byte character */
4269 {
4270 classbits[c/8] |= (1 << (c&7));
4271 if ((options & PCRE_CASELESS) != 0)
4272 {
4273 c = cd->fcc[c]; /* flip case */
4274 classbits[c/8] |= (1 << (c&7));
4275 }
4276 class_charcount++;
4277 class_lastchar = c;
4278 }
4279 }
4280
4281 /* Loop until ']' reached. This "while" is the end of the "do" far above.
4282 If we are at the end of an internal nested string, revert to the outer
4283 string. */
4284
4285 while (((c = *(++ptr)) != 0 ||
4286 (nestptr != NULL &&
4287 (ptr = nestptr, nestptr = NULL, c = *(++ptr)) != 0)) &&
4288 (c != CHAR_RIGHT_SQUARE_BRACKET || inescq));
4289
4290 /* Check for missing terminating ']' */
4291
4292 if (c == 0)
4293 {
4294 *errorcodeptr = ERR6;
4295 goto FAILED;
4296 }
4297
4298 /* If class_charcount is 1, we saw precisely one character whose value is
4299 less than 256. As long as there were no characters >= 128 and there was no
4300 use of \p or \P, in other words, no use of any XCLASS features, we can
4301 optimize.
4302
4303 In UTF-8 mode, we can optimize the negative case only if there were no
4304 characters >= 128 because OP_NOT and the related opcodes like OP_NOTSTAR
4305 operate on single-bytes characters only. This is an historical hangover.
4306 Maybe one day we can tidy these opcodes to handle multi-byte characters.
4307
4308 The optimization throws away the bit map. We turn the item into a
4309 1-character OP_CHAR[I] if it's positive, or OP_NOT[I] if it's negative.
4310 Note that OP_NOT[I] does not support multibyte characters. In the positive
4311 case, it can cause firstbyte to be set. Otherwise, there can be no first
4312 char if this item is first, whatever repeat count may follow. In the case
4313 of reqbyte, save the previous value for reinstating. */
4314
4315 #ifdef SUPPORT_UTF8
4316 if (class_charcount == 1 && !class_utf8 &&
4317 (!utf8 || !negate_class || class_lastchar < 128))
4318 #else
4319 if (class_charcount == 1)
4320 #endif
4321 {
4322 zeroreqbyte = reqbyte;
4323
4324 /* The OP_NOT[I] opcodes work on one-byte characters only. */
4325
4326 if (negate_class)
4327 {
4328 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
4329 zerofirstbyte = firstbyte;
4330 *code++ = ((options & PCRE_CASELESS) != 0)? OP_NOTI: OP_NOT;
4331 *code++ = class_lastchar;
4332 break;
4333 }
4334
4335 /* For a single, positive character, get the value into mcbuffer, and
4336 then we can handle this with the normal one-character code. */
4337
4338 #ifdef SUPPORT_UTF8
4339 if (utf8 && class_lastchar > 127)
4340 mclength = PRIV(ord2utf8)(class_lastchar, mcbuffer);
4341 else
4342 #endif
4343 {
4344 mcbuffer[0] = class_lastchar;
4345 mclength = 1;
4346 }
4347 goto ONE_CHAR;
4348 } /* End of 1-char optimization */
4349
4350 /* The general case - not the one-char optimization. If this is the first
4351 thing in the branch, there can be no first char setting, whatever the
4352 repeat count. Any reqbyte setting must remain unchanged after any kind of
4353 repeat. */
4354
4355 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
4356 zerofirstbyte = firstbyte;
4357 zeroreqbyte = reqbyte;
4358
4359 /* If there are characters with values > 255, we have to compile an
4360 extended class, with its own opcode, unless there was a negated special
4361 such as \S in the class, and PCRE_UCP is not set, because in that case all
4362 characters > 255 are in the class, so any that were explicitly given as
4363 well can be ignored. If (when there are explicit characters > 255 that must
4364 be listed) there are no characters < 256, we can omit the bitmap in the
4365 actual compiled code. */
4366
4367 #ifdef SUPPORT_UTF8
4368 if (class_utf8 && (!should_flip_negation || (options & PCRE_UCP) != 0))
4369 {
4370 *class_utf8data++ = XCL_END; /* Marks the end of extra data */
4371 *code++ = OP_XCLASS;
4372 code += LINK_SIZE;
4373 *code = negate_class? XCL_NOT : 0;
4374
4375 /* If the map is required, move up the extra data to make room for it;
4376 otherwise just move the code pointer to the end of the extra data. */
4377
4378 if (class_charcount > 0)
4379 {
4380 *code++ |= XCL_MAP;
4381 memmove(code + 32, code, class_utf8data - code);
4382 memcpy(code, classbits, 32);
4383 code = class_utf8data + 32;
4384 }
4385 else code = class_utf8data;
4386
4387 /* Now fill in the complete length of the item */
4388
4389 PUT(previous, 1, code - previous);
4390 break; /* End of class handling */
4391 }
4392 #endif
4393
4394 /* If there are no characters > 255, or they are all to be included or
4395 excluded, set the opcode to OP_CLASS or OP_NCLASS, depending on whether the
4396 whole class was negated and whether there were negative specials such as \S
4397 (non-UCP) in the class. Then copy the 32-byte map into the code vector,
4398 negating it if necessary. */
4399
4400 *code++ = (negate_class == should_flip_negation) ? OP_CLASS : OP_NCLASS;
4401 if (negate_class)
4402 {
4403 if (lengthptr == NULL) /* Save time in the pre-compile phase */
4404 for (c = 0; c < 32; c++) code[c] = ~classbits[c];
4405 }
4406 else
4407 {
4408 memcpy(code, classbits, 32);
4409 }
4410 code += 32;
4411 break;
4412
4413
4414 /* ===================================================================*/
4415 /* Various kinds of repeat; '{' is not necessarily a quantifier, but this
4416 has been tested above. */
4417
4418 case CHAR_LEFT_CURLY_BRACKET:
4419 if (!is_quantifier) goto NORMAL_CHAR;
4420 ptr = read_repeat_counts(ptr+1, &repeat_min, &repeat_max, errorcodeptr);
4421 if (*errorcodeptr != 0) goto FAILED;
4422 goto REPEAT;
4423
4424 case CHAR_ASTERISK:
4425 repeat_min = 0;
4426 repeat_max = -1;
4427 goto REPEAT;
4428
4429 case CHAR_PLUS:
4430 repeat_min = 1;
4431 repeat_max = -1;
4432 goto REPEAT;
4433
4434 case CHAR_QUESTION_MARK:
4435 repeat_min = 0;
4436 repeat_max = 1;
4437
4438 REPEAT:
4439 if (previous == NULL)
4440 {
4441 *errorcodeptr = ERR9;
4442 goto FAILED;
4443 }
4444
4445 if (repeat_min == 0)
4446 {
4447 firstbyte = zerofirstbyte; /* Adjust for zero repeat */
4448 reqbyte = zeroreqbyte; /* Ditto */
4449 }
4450
4451 /* Remember whether this is a variable length repeat */
4452
4453 reqvary = (repeat_min == repeat_max)? 0 : REQ_VARY;
4454
4455 op_type = 0; /* Default single-char op codes */
4456 possessive_quantifier = FALSE; /* Default not possessive quantifier */
4457
4458 /* Save start of previous item, in case we have to move it up in order to
4459 insert something before it. */
4460
4461 tempcode = previous;
4462
4463 /* If the next character is '+', we have a possessive quantifier. This
4464 implies greediness, whatever the setting of the PCRE_UNGREEDY option.
4465 If the next character is '?' this is a minimizing repeat, by default,
4466 but if PCRE_UNGREEDY is set, it works the other way round. We change the
4467 repeat type to the non-default. */
4468
4469 if (ptr[1] == CHAR_PLUS)
4470 {
4471 repeat_type = 0; /* Force greedy */
4472 possessive_quantifier = TRUE;
4473 ptr++;
4474 }
4475 else if (ptr[1] == CHAR_QUESTION_MARK)
4476 {
4477 repeat_type = greedy_non_default;
4478 ptr++;
4479 }
4480 else repeat_type = greedy_default;
4481
4482 /* If previous was a recursion call, wrap it in atomic brackets so that
4483 previous becomes the atomic group. All recursions were so wrapped in the
4484 past, but it no longer happens for non-repeated recursions. In fact, the
4485 repeated ones could be re-implemented independently so as not to need this,
4486 but for the moment we rely on the code for repeating groups. */
4487
4488 if (*previous == OP_RECURSE)
4489 {
4490 memmove(previous + 1 + LINK_SIZE, previous, IN_UCHARS(1 + LINK_SIZE));
4491 *previous = OP_ONCE;
4492 PUT(previous, 1, 2 + 2*LINK_SIZE);
4493 previous[2 + 2*LINK_SIZE] = OP_KET;
4494 PUT(previous, 3 + 2*LINK_SIZE, 2 + 2*LINK_SIZE);
4495 code += 2 + 2 * LINK_SIZE;
4496 length_prevgroup = 3 + 3*LINK_SIZE;
4497
4498 /* When actually compiling, we need to check whether this was a forward
4499 reference, and if so, adjust the offset. */
4500
4501 if (lengthptr == NULL && cd->hwm >= cd->start_workspace + LINK_SIZE)
4502 {
4503 int offset = GET(cd->hwm, -LINK_SIZE);
4504 if (offset == previous + 1 - cd->start_code)
4505 PUT(cd->hwm, -LINK_SIZE, offset + 1 + LINK_SIZE);
4506 }
4507 }
4508
4509 /* Now handle repetition for the different types of item. */
4510
4511 /* If previous was a character match, abolish the item and generate a
4512 repeat item instead. If a char item has a minumum of more than one, ensure
4513 that it is set in reqbyte - it might not be if a sequence such as x{3} is
4514 the first thing in a branch because the x will have gone into firstbyte
4515 instead. */
4516
4517 if (*previous == OP_CHAR || *previous == OP_CHARI)
4518 {
4519 op_type = (*previous == OP_CHAR)? 0 : OP_STARI - OP_STAR;
4520
4521 /* Deal with UTF-8 characters that take up more than one byte. It's
4522 easier to write this out separately than try to macrify it. Use c to
4523 hold the length of the character in bytes, plus 0x80 to flag that it's a
4524 length rather than a small character. */
4525
4526 #ifdef SUPPORT_UTF8
4527 if (utf8 && (code[-1] & 0x80) != 0)
4528 {
4529 pcre_uchar *lastchar = code - 1;
4530 while((*lastchar & 0xc0) == 0x80) lastchar--;
4531 c = code - lastchar; /* Length of UTF-8 character */
4532 memcpy(utf8_char, lastchar, c); /* Save the char */
4533 c |= 0x80; /* Flag c as a length */
4534 }
4535 else
4536 #endif
4537
4538 /* Handle the case of a single byte - either with no UTF8 support, or
4539 with UTF-8 disabled, or for a UTF-8 character < 128. */
4540
4541 {
4542 c = code[-1];
4543 if (repeat_min > 1) reqbyte = c | req_caseopt | cd->req_varyopt;
4544 }
4545
4546 /* If the repetition is unlimited, it pays to see if the next thing on
4547 the line is something that cannot possibly match this character. If so,
4548 automatically possessifying this item gains some performance in the case
4549 where the match fails. */
4550
4551 if (!possessive_quantifier &&
4552 repeat_max < 0 &&
4553 check_auto_possessive(previous, utf8, ptr + 1, options, cd))
4554 {
4555 repeat_type = 0; /* Force greedy */
4556 possessive_quantifier = TRUE;
4557 }
4558
4559 goto OUTPUT_SINGLE_REPEAT; /* Code shared with single character types */
4560 }
4561
4562 /* If previous was a single negated character ([^a] or similar), we use
4563 one of the special opcodes, replacing it. The code is shared with single-
4564 character repeats by setting opt_type to add a suitable offset into
4565 repeat_type. We can also test for auto-possessification. OP_NOT and OP_NOTI
4566 are currently used only for single-byte chars. */
4567
4568 else if (*previous == OP_NOT || *previous == OP_NOTI)
4569 {
4570 op_type = ((*previous == OP_NOT)? OP_NOTSTAR : OP_NOTSTARI) - OP_STAR;
4571 c = previous[1];
4572 if (!possessive_quantifier &&
4573 repeat_max < 0 &&
4574 check_auto_possessive(previous, utf8, ptr + 1, options, cd))
4575 {
4576 repeat_type = 0; /* Force greedy */
4577 possessive_quantifier = TRUE;
4578 }
4579 goto OUTPUT_SINGLE_REPEAT;
4580 }
4581
4582 /* If previous was a character type match (\d or similar), abolish it and
4583 create a suitable repeat item. The code is shared with single-character
4584 repeats by setting op_type to add a suitable offset into repeat_type. Note
4585 the the Unicode property types will be present only when SUPPORT_UCP is
4586 defined, but we don't wrap the little bits of code here because it just
4587 makes it horribly messy. */
4588
4589 else if (*previous < OP_EODN)
4590 {
4591 pcre_uchar *oldcode;
4592 int prop_type, prop_value;
4593 op_type = OP_TYPESTAR - OP_STAR; /* Use type opcodes */
4594 c = *previous;
4595
4596 if (!possessive_quantifier &&
4597 repeat_max < 0 &&
4598 check_auto_possessive(previous, utf8, ptr + 1, options, cd))
4599 {
4600 repeat_type = 0; /* Force greedy */
4601 possessive_quantifier = TRUE;
4602 }
4603
4604 OUTPUT_SINGLE_REPEAT:
4605 if (*previous == OP_PROP || *previous == OP_NOTPROP)
4606 {
4607 prop_type = previous[1];
4608 prop_value = previous[2];
4609 }
4610 else prop_type = prop_value = -1;
4611
4612 oldcode = code;
4613 code = previous; /* Usually overwrite previous item */
4614
4615 /* If the maximum is zero then the minimum must also be zero; Perl allows
4616 this case, so we do too - by simply omitting the item altogether. */
4617
4618 if (repeat_max == 0) goto END_REPEAT;
4619
4620 /*--------------------------------------------------------------------*/
4621 /* This code is obsolete from release 8.00; the restriction was finally
4622 removed: */
4623
4624 /* All real repeats make it impossible to handle partial matching (maybe
4625 one day we will be able to remove this restriction). */
4626
4627 /* if (repeat_max != 1) cd->external_flags |= PCRE_NOPARTIAL; */
4628 /*--------------------------------------------------------------------*/
4629
4630 /* Combine the op_type with the repeat_type */
4631
4632 repeat_type += op_type;
4633
4634 /* A minimum of zero is handled either as the special case * or ?, or as
4635 an UPTO, with the maximum given. */
4636
4637 if (repeat_min == 0)
4638 {
4639 if (repeat_max == -1) *code++ = OP_STAR + repeat_type;
4640 else if (repeat_max == 1) *code++ = OP_QUERY + repeat_type;
4641 else
4642 {
4643 *code++ = OP_UPTO + repeat_type;
4644 PUT2INC(code, 0, repeat_max);
4645 }
4646 }
4647
4648 /* A repeat minimum of 1 is optimized into some special cases. If the
4649 maximum is unlimited, we use OP_PLUS. Otherwise, the original item is
4650 left in place and, if the maximum is greater than 1, we use OP_UPTO with
4651 one less than the maximum. */
4652
4653 else if (repeat_min == 1)
4654 {
4655 if (repeat_max == -1)
4656 *code++ = OP_PLUS + repeat_type;
4657 else
4658 {
4659 code = oldcode; /* leave previous item in place */
4660 if (repeat_max == 1) goto END_REPEAT;
4661 *code++ = OP_UPTO + repeat_type;
4662 PUT2INC(code, 0, repeat_max - 1);
4663 }
4664 }
4665
4666 /* The case {n,n} is just an EXACT, while the general case {n,m} is
4667 handled as an EXACT followed by an UPTO. */
4668
4669 else
4670 {
4671 *code++ = OP_EXACT + op_type; /* NB EXACT doesn't have repeat_type */
4672 PUT2INC(code, 0, repeat_min);
4673
4674 /* If the maximum is unlimited, insert an OP_STAR. Before doing so,
4675 we have to insert the character for the previous code. For a repeated
4676 Unicode property match, there are two extra bytes that define the
4677 required property. In UTF-8 mode, long characters have their length in
4678 c, with the 0x80 bit as a flag. */
4679
4680 if (repeat_max < 0)
4681 {
4682 #ifdef SUPPORT_UTF8
4683 if (utf8 && c >= 128)
4684 {
4685 memcpy(code, utf8_char, c & 7);
4686 code += c & 7;
4687 }
4688 else
4689 #endif
4690 {
4691 *code++ = c;
4692 if (prop_type >= 0)
4693 {
4694 *code++ = prop_type;
4695 *code++ = prop_value;
4696 }
4697 }
4698 *code++ = OP_STAR + repeat_type;
4699 }
4700
4701 /* Else insert an UPTO if the max is greater than the min, again
4702 preceded by the character, for the previously inserted code. If the
4703 UPTO is just for 1 instance, we can use QUERY instead. */
4704
4705 else if (repeat_max != repeat_min)
4706 {
4707 #ifdef SUPPORT_UTF8
4708 if (utf8 && c >= 128)
4709 {
4710 memcpy(code, utf8_char, c & 7);
4711 code += c & 7;
4712 }
4713 else
4714 #endif
4715 *code++ = c;
4716 if (prop_type >= 0)
4717 {
4718 *code++ = prop_type;
4719 *code++ = prop_value;
4720 }
4721 repeat_max -= repeat_min;
4722
4723 if (repeat_max == 1)
4724 {
4725 *code++ = OP_QUERY + repeat_type;
4726 }
4727 else
4728 {
4729 *code++ = OP_UPTO + repeat_type;
4730 PUT2INC(code, 0, repeat_max);
4731 }
4732 }
4733 }
4734
4735 /* The character or character type itself comes last in all cases. */
4736
4737 #ifdef SUPPORT_UTF8
4738 if (utf8 && c >= 128)
4739 {
4740 memcpy(code, utf8_char, c & 7);
4741 code += c & 7;
4742 }
4743 else
4744 #endif
4745 *code++ = c;
4746
4747 /* For a repeated Unicode property match, there are two extra bytes that
4748 define the required property. */
4749
4750 #ifdef SUPPORT_UCP
4751 if (prop_type >= 0)
4752 {
4753 *code++ = prop_type;
4754 *code++ = prop_value;
4755 }
4756 #endif
4757 }
4758
4759 /* If previous was a character class or a back reference, we put the repeat
4760 stuff after it, but just skip the item if the repeat was {0,0}. */
4761
4762 else if (*previous == OP_CLASS ||
4763 *previous == OP_NCLASS ||
4764 #ifdef SUPPORT_UTF8
4765 *previous == OP_XCLASS ||
4766 #endif
4767 *previous == OP_REF ||
4768 *previous == OP_REFI)
4769 {
4770 if (repeat_max == 0)
4771 {
4772 code = previous;
4773 goto END_REPEAT;
4774 }
4775
4776 /*--------------------------------------------------------------------*/
4777 /* This code is obsolete from release 8.00; the restriction was finally
4778 removed: */
4779
4780 /* All real repeats make it impossible to handle partial matching (maybe
4781 one day we will be able to remove this restriction). */
4782
4783 /* if (repeat_max != 1) cd->external_flags |= PCRE_NOPARTIAL; */
4784 /*--------------------------------------------------------------------*/
4785
4786 if (repeat_min == 0 && repeat_max == -1)
4787 *code++ = OP_CRSTAR + repeat_type;
4788 else if (repeat_min == 1 && repeat_max == -1)
4789 *code++ = OP_CRPLUS + repeat_type;
4790 else if (repeat_min == 0 && repeat_max == 1)
4791 *code++ = OP_CRQUERY + repeat_type;
4792 else
4793 {
4794 *code++ = OP_CRRANGE + repeat_type;
4795 PUT2INC(code, 0, repeat_min);
4796 if (repeat_max == -1) repeat_max = 0; /* 2-byte encoding for max */
4797 PUT2INC(code, 0, repeat_max);
4798 }
4799 }
4800
4801 /* If previous was a bracket group, we may have to replicate it in certain
4802 cases. Note that at this point we can encounter only the "basic" bracket
4803 opcodes such as BRA and CBRA, as this is the place where they get converted
4804 into the more special varieties such as BRAPOS and SBRA. A test for >=
4805 OP_ASSERT and <= OP_COND includes ASSERT, ASSERT_NOT, ASSERTBACK,
4806 ASSERTBACK_NOT, ONCE, BRA, CBRA, and COND. Originally, PCRE did not allow
4807 repetition of assertions, but now it does, for Perl compatibility. */
4808
4809 else if (*previous >= OP_ASSERT && *previous <= OP_COND)
4810 {
4811 register int i;
4812 int len = (int)(code - previous);
4813 pcre_uchar *bralink = NULL;
4814 pcre_uchar *brazeroptr = NULL;
4815
4816 /* Repeating a DEFINE group is pointless, but Perl allows the syntax, so
4817 we just ignore the repeat. */
4818
4819 if (*previous == OP_COND && previous[LINK_SIZE+1] == OP_DEF)
4820 goto END_REPEAT;
4821
4822 /* There is no sense in actually repeating assertions. The only potential
4823 use of repetition is in cases when the assertion is optional. Therefore,
4824 if the minimum is greater than zero, just ignore the repeat. If the
4825 maximum is not not zero or one, set it to 1. */
4826
4827 if (*previous < OP_ONCE) /* Assertion */
4828 {
4829 if (repeat_min > 0) goto END_REPEAT;
4830 if (repeat_max < 0 || repeat_max > 1) repeat_max = 1;
4831 }
4832
4833 /* The case of a zero minimum is special because of the need to stick
4834 OP_BRAZERO in front of it, and because the group appears once in the
4835 data, whereas in other cases it appears the minimum number of times. For
4836 this reason, it is simplest to treat this case separately, as otherwise
4837 the code gets far too messy. There are several special subcases when the
4838 minimum is zero. */
4839
4840 if (repeat_min == 0)
4841 {
4842 /* If the maximum is also zero, we used to just omit the group from the
4843 output altogether, like this:
4844
4845 ** if (repeat_max == 0)
4846 ** {
4847 ** code = previous;
4848 ** goto END_REPEAT;
4849 ** }
4850
4851 However, that fails when a group or a subgroup within it is referenced
4852 as a subroutine from elsewhere in the pattern, so now we stick in
4853 OP_SKIPZERO in front of it so that it is skipped on execution. As we
4854 don't have a list of which groups are referenced, we cannot do this
4855 selectively.
4856
4857 If the maximum is 1 or unlimited, we just have to stick in the BRAZERO
4858 and do no more at this point. However, we do need to adjust any
4859 OP_RECURSE calls inside the group that refer to the group itself or any
4860 internal or forward referenced group, because the offset is from the
4861 start of the whole regex. Temporarily terminate the pattern while doing
4862 this. */
4863
4864 if (repeat_max <= 1) /* Covers 0, 1, and unlimited */
4865 {
4866 *code = OP_END;
4867 adjust_recurse(previous, 1, utf8, cd, save_hwm);
4868 memmove(previous + 1, previous, IN_UCHARS(len));
4869 code++;
4870 if (repeat_max == 0)
4871 {
4872 *previous++ = OP_SKIPZERO;
4873 goto END_REPEAT;
4874 }
4875 brazeroptr = previous; /* Save for possessive optimizing */
4876 *previous++ = OP_BRAZERO + repeat_type;
4877 }
4878
4879 /* If the maximum is greater than 1 and limited, we have to replicate
4880 in a nested fashion, sticking OP_BRAZERO before each set of brackets.
4881 The first one has to be handled carefully because it's the original
4882 copy, which has to be moved up. The remainder can be handled by code
4883 that is common with the non-zero minimum case below. We have to
4884 adjust the value or repeat_max, since one less copy is required. Once
4885 again, we may have to adjust any OP_RECURSE calls inside the group. */
4886
4887 else
4888 {
4889 int offset;
4890 *code = OP_END;
4891 adjust_recurse(previous, 2 + LINK_SIZE, utf8, cd, save_hwm);
4892 memmove(previous + 2 + LINK_SIZE, previous, IN_UCHARS(len));
4893 code += 2 + LINK_SIZE;
4894 *previous++ = OP_BRAZERO + repeat_type;
4895 *previous++ = OP_BRA;
4896
4897 /* We chain together the bracket offset fields that have to be
4898 filled in later when the ends of the brackets are reached. */
4899
4900 offset = (bralink == NULL)? 0 : (int)(previous - bralink);
4901 bralink = previous;
4902 PUTINC(previous, 0, offset);
4903 }
4904
4905 repeat_max--;
4906 }
4907
4908 /* If the minimum is greater than zero, replicate the group as many
4909 times as necessary, and adjust the maximum to the number of subsequent
4910 copies that we need. If we set a first char from the group, and didn't
4911 set a required char, copy the latter from the former. If there are any
4912 forward reference subroutine calls in the group, there will be entries on
4913 the workspace list; replicate these with an appropriate increment. */
4914
4915 else
4916 {
4917 if (repeat_min > 1)
4918 {
4919 /* In the pre-compile phase, we don't actually do the replication. We
4920 just adjust the length as if we had. Do some paranoid checks for
4921 potential integer overflow. The INT64_OR_DOUBLE type is a 64-bit
4922 integer type when available, otherwise double. */
4923
4924 if (lengthptr != NULL)
4925 {
4926 int delta = (repeat_min - 1)*length_prevgroup;
4927 if ((INT64_OR_DOUBLE)(repeat_min - 1)*
4928 (INT64_OR_DOUBLE)length_prevgroup >
4929 (INT64_OR_DOUBLE)INT_MAX ||
4930 OFLOW_MAX - *lengthptr < delta)
4931 {
4932 *errorcodeptr = ERR20;
4933 goto FAILED;
4934 }
4935 *lengthptr += delta;
4936 }
4937
4938 /* This is compiling for real */
4939
4940 else
4941 {
4942 if (groupsetfirstbyte && reqbyte < 0) reqbyte = firstbyte;
4943 for (i = 1; i < repeat_min; i++)
4944 {
4945 pcre_uchar *hc;
4946 pcre_uchar *this_hwm = cd->hwm;
4947 memcpy(code, previous, IN_UCHARS(len));
4948 for (hc = save_hwm; hc < this_hwm; hc += LINK_SIZE)
4949 {
4950 PUT(cd->hwm, 0, GET(hc, 0) + len);
4951 cd->hwm += LINK_SIZE;
4952 }
4953 save_hwm = this_hwm;
4954 code += len;
4955 }
4956 }
4957 }
4958
4959 if (repeat_max > 0) repeat_max -= repeat_min;
4960 }
4961
4962 /* This code is common to both the zero and non-zero minimum cases. If
4963 the maximum is limited, it replicates the group in a nested fashion,
4964 remembering the bracket starts on a stack. In the case of a zero minimum,
4965 the first one was set up above. In all cases the repeat_max now specifies
4966 the number of additional copies needed. Again, we must remember to
4967 replicate entries on the forward reference list. */
4968
4969 if (repeat_max >= 0)
4970 {
4971 /* In the pre-compile phase, we don't actually do the replication. We
4972 just adjust the length as if we had. For each repetition we must add 1
4973 to the length for BRAZERO and for all but the last repetition we must
4974 add 2 + 2*LINKSIZE to allow for the nesting that occurs. Do some
4975 paranoid checks to avoid integer overflow. The INT64_OR_DOUBLE type is
4976 a 64-bit integer type when available, otherwise double. */
4977
4978 if (lengthptr != NULL && repeat_max > 0)
4979 {
4980 int delta = repeat_max * (length_prevgroup + 1 + 2 + 2*LINK_SIZE) -
4981 2 - 2*LINK_SIZE; /* Last one doesn't nest */
4982 if ((INT64_OR_DOUBLE)repeat_max *
4983 (INT64_OR_DOUBLE)(length_prevgroup + 1 + 2 + 2*LINK_SIZE)
4984 > (INT64_OR_DOUBLE)INT_MAX ||
4985 OFLOW_MAX - *lengthptr < delta)
4986 {
4987 *errorcodeptr = ERR20;
4988 goto FAILED;
4989 }
4990 *lengthptr += delta;
4991 }
4992
4993 /* This is compiling for real */
4994
4995 else for (i = repeat_max - 1; i >= 0; i--)
4996 {
4997 pcre_uchar *hc;
4998 pcre_uchar *this_hwm = cd->hwm;
4999
5000 *code++ = OP_BRAZERO + repeat_type;
5001
5002 /* All but the final copy start a new nesting, maintaining the
5003 chain of brackets outstanding. */
5004
5005 if (i != 0)
5006 {
5007 int offset;
5008 *code++ = OP_BRA;
5009 offset = (bralink == NULL)? 0 : (int)(code - bralink);
5010 bralink = code;
5011 PUTINC(code, 0, offset);
5012 }
5013
5014 memcpy(code, previous, IN_UCHARS(len));
5015 for (hc = save_hwm; hc < this_hwm; hc += LINK_SIZE)
5016 {
5017 PUT(cd->hwm, 0, GET(hc, 0) + len + ((i != 0)? 2+LINK_SIZE : 1));
5018 cd->hwm += LINK_SIZE;
5019 }
5020 save_hwm = this_hwm;
5021 code += len;
5022 }
5023
5024 /* Now chain through the pending brackets, and fill in their length
5025 fields (which are holding the chain links pro tem). */
5026
5027 while (bralink != NULL)
5028 {
5029 int oldlinkoffset;
5030 int offset = (int)(code - bralink + 1);
5031 pcre_uchar *bra = code - offset;
5032 oldlinkoffset = GET(bra, 1);
5033 bralink = (oldlinkoffset == 0)? NULL : bralink - oldlinkoffset;
5034 *code++ = OP_KET;
5035 PUTINC(code, 0, offset);
5036 PUT(bra, 1, offset);
5037 }
5038 }
5039
5040 /* If the maximum is unlimited, set a repeater in the final copy. For
5041 ONCE brackets, that's all we need to do. However, possessively repeated
5042 ONCE brackets can be converted into non-capturing brackets, as the
5043 behaviour of (?:xx)++ is the same as (?>xx)++ and this saves having to
5044 deal with possessive ONCEs specially.
5045
5046 Otherwise, when we are doing the actual compile phase, check to see
5047 whether this group is one that could match an empty string. If so,
5048 convert the initial operator to the S form (e.g. OP_BRA -> OP_SBRA) so
5049 that runtime checking can be done. [This check is also applied to ONCE
5050 groups at runtime, but in a different way.]
5051
5052 Then, if the quantifier was possessive and the bracket is not a
5053 conditional, we convert the BRA code to the POS form, and the KET code to
5054 KETRPOS. (It turns out to be convenient at runtime to detect this kind of
5055 subpattern at both the start and at the end.) The use of special opcodes
5056 makes it possible to reduce greatly the stack usage in pcre_exec(). If
5057 the group is preceded by OP_BRAZERO, convert this to OP_BRAPOSZERO.
5058
5059 Then, if the minimum number of matches is 1 or 0, cancel the possessive
5060 flag so that the default action below, of wrapping everything inside
5061 atomic brackets, does not happen. When the minimum is greater than 1,
5062 there will be earlier copies of the group, and so we still have to wrap
5063 the whole thing. */
5064
5065 else
5066 {
5067 pcre_uchar *ketcode = code - 1 - LINK_SIZE;
5068 pcre_uchar *bracode = ketcode - GET(ketcode, 1);
5069
5070 /* Convert possessive ONCE brackets to non-capturing */
5071
5072 if ((*bracode == OP_ONCE || *bracode == OP_ONCE_NC) &&
5073 possessive_quantifier) *bracode = OP_BRA;
5074
5075 /* For non-possessive ONCE brackets, all we need to do is to
5076 set the KET. */
5077
5078 if (*bracode == OP_ONCE || *bracode == OP_ONCE_NC)
5079 *ketcode = OP_KETRMAX + repeat_type;
5080
5081 /* Handle non-ONCE brackets and possessive ONCEs (which have been
5082 converted to non-capturing above). */
5083
5084 else
5085 {
5086 /* In the compile phase, check for empty string matching. */
5087
5088 if (lengthptr == NULL)
5089 {
5090 pcre_uchar *scode = bracode;
5091 do
5092 {
5093 if (could_be_empty_branch(scode, ketcode, utf8, cd))
5094 {
5095 *bracode += OP_SBRA - OP_BRA;
5096 break;
5097 }
5098 scode += GET(scode, 1);
5099 }
5100 while (*scode == OP_ALT);
5101 }
5102
5103 /* Handle possessive quantifiers. */
5104
5105 if (possessive_quantifier)
5106 {
5107 /* For COND brackets, we wrap the whole thing in a possessively
5108 repeated non-capturing bracket, because we have not invented POS
5109 versions of the COND opcodes. Because we are moving code along, we
5110 must ensure that any pending recursive references are updated. */
5111
5112 if (*bracode == OP_COND || *bracode == OP_SCOND)
5113 {
5114 int nlen = (int)(code - bracode);
5115 *code = OP_END;
5116 adjust_recurse(bracode, 1 + LINK_SIZE, utf8, cd, save_hwm);
5117 memmove(bracode + 1 + LINK_SIZE, bracode, IN_UCHARS(nlen));
5118 code += 1 + LINK_SIZE;
5119 nlen += 1 + LINK_SIZE;
5120 *bracode = OP_BRAPOS;
5121 *code++ = OP_KETRPOS;
5122 PUTINC(code, 0, nlen);
5123 PUT(bracode, 1, nlen);
5124 }
5125
5126 /* For non-COND brackets, we modify the BRA code and use KETRPOS. */
5127
5128 else
5129 {
5130 *bracode += 1; /* Switch to xxxPOS opcodes */
5131 *ketcode = OP_KETRPOS;
5132 }
5133
5134 /* If the minimum is zero, mark it as possessive, then unset the
5135 possessive flag when the minimum is 0 or 1. */
5136
5137 if (brazeroptr != NULL) *brazeroptr = OP_BRAPOSZERO;
5138 if (repeat_min < 2) possessive_quantifier = FALSE;
5139 }
5140
5141 /* Non-possessive quantifier */
5142
5143 else *ketcode = OP_KETRMAX + repeat_type;
5144 }
5145 }
5146 }
5147
5148 /* If previous is OP_FAIL, it was generated by an empty class [] in
5149 JavaScript mode. The other ways in which OP_FAIL can be generated, that is
5150 by (*FAIL) or (?!) set previous to NULL, which gives a "nothing to repeat"
5151 error above. We can just ignore the repeat in JS case. */
5152
5153 else if (*previous == OP_FAIL) goto END_REPEAT;
5154
5155 /* Else there's some kind of shambles */
5156
5157 else
5158 {
5159 *errorcodeptr = ERR11;
5160 goto FAILED;
5161 }
5162
5163 /* If the character following a repeat is '+', or if certain optimization
5164 tests above succeeded, possessive_quantifier is TRUE. For some opcodes,
5165 there are special alternative opcodes for this case. For anything else, we
5166 wrap the entire repeated item inside OP_ONCE brackets. Logically, the '+'
5167 notation is just syntactic sugar, taken from Sun's Java package, but the
5168 special opcodes can optimize it.
5169
5170 Some (but not all) possessively repeated subpatterns have already been
5171 completely handled in the code just above. For them, possessive_quantifier
5172 is always FALSE at this stage.
5173
5174 Note that the repeated item starts at tempcode, not at previous, which
5175 might be the first part of a string whose (former) last char we repeated.
5176
5177 Possessifying an 'exact' quantifier has no effect, so we can ignore it. But
5178 an 'upto' may follow. We skip over an 'exact' item, and then test the
5179 length of what remains before proceeding. */
5180
5181 if (possessive_quantifier)
5182 {
5183 int len;
5184
5185 if (*tempcode == OP_TYPEEXACT)
5186 tempcode += PRIV(OP_lengths)[*tempcode] +
5187 ((tempcode[1 + IMM2_SIZE] == OP_PROP
5188 || tempcode[1 + IMM2_SIZE] == OP_NOTPROP)? 2 : 0);
5189
5190 else if (*tempcode == OP_EXACT || *tempcode == OP_NOTEXACT)
5191 {
5192 tempcode += PRIV(OP_lengths)[*tempcode];
5193 #ifdef SUPPORT_UTF8
5194 if (utf8 && tempcode[-1] >= 0xc0)
5195 tempcode += PRIV(utf8_table4)[tempcode[-1] & 0x3f];
5196 #endif
5197 }
5198
5199 len = (int)(code - tempcode);
5200 if (len > 0) switch (*tempcode)
5201 {
5202 case OP_STAR: *tempcode = OP_POSSTAR; break;
5203 case OP_PLUS: *tempcode = OP_POSPLUS; break;
5204 case OP_QUERY: *tempcode = OP_POSQUERY; break;
5205 case OP_UPTO: *tempcode = OP_POSUPTO; break;
5206
5207 case OP_STARI: *tempcode = OP_POSSTARI; break;
5208 case OP_PLUSI: *tempcode = OP_POSPLUSI; break;
5209 case OP_QUERYI: *tempcode = OP_POSQUERYI; break;
5210 case OP_UPTOI: *tempcode = OP_POSUPTOI; break;
5211
5212 case OP_NOTSTAR: *tempcode = OP_NOTPOSSTAR; break;
5213 case OP_NOTPLUS: *tempcode = OP_NOTPOSPLUS; break;
5214 case OP_NOTQUERY: *tempcode = OP_NOTPOSQUERY; break;
5215 case OP_NOTUPTO: *tempcode = OP_NOTPOSUPTO; break;
5216
5217 case OP_NOTSTARI: *tempcode = OP_NOTPOSSTARI; break;
5218 case OP_NOTPLUSI: *tempcode = OP_NOTPOSPLUSI; break;
5219 case OP_NOTQUERYI: *tempcode = OP_NOTPOSQUERYI; break;
5220 case OP_NOTUPTOI: *tempcode = OP_NOTPOSUPTOI; break;
5221
5222 case OP_TYPESTAR: *tempcode = OP_TYPEPOSSTAR; break;
5223 case OP_TYPEPLUS: *tempcode = OP_TYPEPOSPLUS; break;
5224 case OP_TYPEQUERY: *tempcode = OP_TYPEPOSQUERY; break;
5225 case OP_TYPEUPTO: *tempcode = OP_TYPEPOSUPTO; break;
5226
5227 /* Because we are moving code along, we must ensure that any
5228 pending recursive references are updated. */
5229
5230 default:
5231 *code = OP_END;
5232 adjust_recurse(tempcode, 1 + LINK_SIZE, utf8, cd, save_hwm);
5233 memmove(tempcode + 1 + LINK_SIZE, tempcode, IN_UCHARS(len));
5234 code += 1 + LINK_SIZE;
5235 len += 1 + LINK_SIZE;
5236 tempcode[0] = OP_ONCE;
5237 *code++ = OP_KET;
5238 PUTINC(code, 0, len);
5239 PUT(tempcode, 1, len);
5240 break;
5241 }
5242 }
5243
5244 /* In all case we no longer have a previous item. We also set the
5245 "follows varying string" flag for subsequently encountered reqbytes if
5246 it isn't already set and we have just passed a varying length item. */
5247
5248 END_REPEAT:
5249 previous = NULL;
5250 cd->req_varyopt |= reqvary;
5251 break;
5252
5253
5254 /* ===================================================================*/
5255 /* Start of nested parenthesized sub-expression, or comment or lookahead or
5256 lookbehind or option setting or condition or all the other extended
5257 parenthesis forms. */
5258
5259 case CHAR_LEFT_PARENTHESIS:
5260 newoptions = options;
5261 skipbytes = 0;
5262 bravalue = OP_CBRA;
5263 save_hwm = cd->hwm;
5264 reset_bracount = FALSE;
5265
5266 /* First deal with various "verbs" that can be introduced by '*'. */
5267
5268 if (*(++ptr) == CHAR_ASTERISK &&
5269 ((cd->ctypes[ptr[1]] & ctype_letter) != 0 || ptr[1] == ':'))
5270 {
5271 int i, namelen;
5272 int arglen = 0;
5273 const char *vn = verbnames;
5274 const pcre_uchar *name = ptr + 1;
5275 const pcre_uchar *arg = NULL;
5276 previous = NULL;
5277 while ((cd->ctypes[*++ptr] & ctype_letter) != 0) {};
5278 namelen = (int)(ptr - name);
5279
5280 /* It appears that Perl allows any characters whatsoever, other than
5281 a closing parenthesis, to appear in arguments, so we no longer insist on
5282 letters, digits, and underscores. */
5283
5284 if (*ptr == CHAR_COLON)
5285 {
5286 arg = ++ptr;
5287 while (*ptr != 0 && *ptr != CHAR_RIGHT_PARENTHESIS) ptr++;
5288 arglen = (int)(ptr - arg);
5289 }
5290
5291 if (*ptr != CHAR_RIGHT_PARENTHESIS)
5292 {
5293 *errorcodeptr = ERR60;
5294 goto FAILED;
5295 }
5296
5297 /* Scan the table of verb names */
5298
5299 for (i = 0; i < verbcount; i++)
5300 {
5301 if (namelen == verbs[i].len &&
5302 STRNCMP_UC_C8(name, vn, namelen) == 0)
5303 {
5304 /* Check for open captures before ACCEPT and convert it to
5305 ASSERT_ACCEPT if in an assertion. */
5306
5307 if (verbs[i].op == OP_ACCEPT)
5308 {
5309 open_capitem *oc;
5310 if (arglen != 0)
5311 {
5312 *errorcodeptr = ERR59;
5313 goto FAILED;
5314 }
5315 cd->had_accept = TRUE;
5316 for (oc = cd->open_caps; oc != NULL; oc = oc->next)
5317 {
5318 *code++ = OP_CLOSE;
5319 PUT2INC(code, 0, oc->number);
5320 }
5321 *code++ = (cd->assert_depth > 0)? OP_ASSERT_ACCEPT : OP_ACCEPT;
5322
5323 /* Do not set firstbyte after *ACCEPT */
5324 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
5325 }
5326
5327 /* Handle other cases with/without an argument */
5328
5329 else if (arglen == 0)
5330 {
5331 if (verbs[i].op < 0) /* Argument is mandatory */
5332 {
5333 *errorcodeptr = ERR66;
5334 goto FAILED;
5335 }
5336 *code = verbs[i].op;
5337 if (*code++ == OP_THEN) cd->external_flags |= PCRE_HASTHEN;
5338 }
5339
5340 else
5341 {
5342 if (verbs[i].op_arg < 0) /* Argument is forbidden */
5343 {
5344 *errorcodeptr = ERR59;
5345 goto FAILED;
5346 }
5347 *code = verbs[i].op_arg;
5348 if (*code++ == OP_THEN_ARG) cd->external_flags |= PCRE_HASTHEN;
5349 *code++ = arglen;
5350 memcpy(code, arg, IN_UCHARS(arglen));
5351 code += arglen;
5352 *code++ = 0;
5353 }
5354
5355 break; /* Found verb, exit loop */
5356 }
5357
5358 vn += verbs[i].len + 1;
5359 }
5360
5361 if (i < verbcount) continue; /* Successfully handled a verb */
5362 *errorcodeptr = ERR60; /* Verb not recognized */
5363 goto FAILED;
5364 }
5365
5366 /* Deal with the extended parentheses; all are introduced by '?', and the
5367 appearance of any of them means that this is not a capturing group. */
5368
5369 else if (*ptr == CHAR_QUESTION_MARK)
5370 {
5371 int i, set, unset, namelen;
5372 int *optset;
5373 const pcre_uchar *name;
5374 pcre_uchar *slot;
5375
5376 switch (*(++ptr))
5377 {
5378 case CHAR_NUMBER_SIGN: /* Comment; skip to ket */
5379 ptr++;
5380 while (*ptr != 0 && *ptr != CHAR_RIGHT_PARENTHESIS) ptr++;
5381 if (*ptr == 0)
5382 {
5383 *errorcodeptr = ERR18;
5384 goto FAILED;
5385 }
5386 continue;
5387
5388
5389 /* ------------------------------------------------------------ */
5390 case CHAR_VERTICAL_LINE: /* Reset capture count for each branch */
5391 reset_bracount = TRUE;
5392 /* Fall through */
5393
5394 /* ------------------------------------------------------------ */
5395 case CHAR_COLON: /* Non-capturing bracket */
5396 bravalue = OP_BRA;
5397 ptr++;
5398 break;
5399
5400
5401 /* ------------------------------------------------------------ */
5402 case CHAR_LEFT_PARENTHESIS:
5403 bravalue = OP_COND; /* Conditional group */
5404
5405 /* A condition can be an assertion, a number (referring to a numbered
5406 group), a name (referring to a named group), or 'R', referring to
5407 recursion. R<digits> and R&name are also permitted for recursion tests.
5408
5409 There are several syntaxes for testing a named group: (?(name)) is used
5410 by Python; Perl 5.10 onwards uses (?(<name>) or (?('name')).
5411
5412 There are two unfortunate ambiguities, caused by history. (a) 'R' can
5413 be the recursive thing or the name 'R' (and similarly for 'R' followed
5414 by digits), and (b) a number could be a name that consists of digits.
5415 In both cases, we look for a name first; if not found, we try the other
5416 cases. */
5417
5418 /* For conditions that are assertions, check the syntax, and then exit
5419 the switch. This will take control down to where bracketed groups,
5420 including assertions, are processed. */
5421
5422 if (ptr[1] == CHAR_QUESTION_MARK && (ptr[2] == CHAR_EQUALS_SIGN ||
5423 ptr[2] == CHAR_EXCLAMATION_MARK || ptr[2] == CHAR_LESS_THAN_SIGN))
5424 break;
5425
5426 /* Most other conditions use OP_CREF (a couple change to OP_RREF
5427 below), and all need to skip 1+IMM2_SIZE bytes at the start of the group. */
5428
5429 code[1+LINK_SIZE] = OP_CREF;
5430 skipbytes = 1+IMM2_SIZE;
5431 refsign = -1;
5432
5433 /* Check for a test for recursion in a named group. */
5434
5435 if (ptr[1] == CHAR_R && ptr[2] == CHAR_AMPERSAND)
5436 {
5437 terminator = -1;
5438 ptr += 2;
5439 code[1+LINK_SIZE] = OP_RREF; /* Change the type of test */
5440 }
5441
5442 /* Check for a test for a named group's having been set, using the Perl
5443 syntax (?(<name>) or (?('name') */
5444
5445 else if (ptr[1] == CHAR_LESS_THAN_SIGN)
5446 {
5447 terminator = CHAR_GREATER_THAN_SIGN;
5448 ptr++;
5449 }
5450 else if (ptr[1] == CHAR_APOSTROPHE)
5451 {
5452 terminator = CHAR_APOSTROPHE;
5453 ptr++;
5454 }
5455 else
5456 {
5457 terminator = 0;
5458 if (ptr[1] == CHAR_MINUS || ptr[1] == CHAR_PLUS) refsign = *(++ptr);
5459 }
5460
5461 /* We now expect to read a name; any thing else is an error */
5462
5463 if ((cd->ctypes[ptr[1]] & ctype_word) == 0)
5464 {
5465 ptr += 1; /* To get the right offset */
5466 *errorcodeptr = ERR28;
5467 goto FAILED;
5468 }
5469
5470 /* Read the name, but also get it as a number if it's all digits */
5471
5472 recno = 0;
5473 name = ++ptr;
5474 while ((cd->ctypes[*ptr] & ctype_word) != 0)
5475 {
5476 if (recno >= 0)
5477 recno = ((digitab[*ptr] & ctype_digit) != 0)?
5478 recno * 10 + *ptr - CHAR_0 : -1;
5479 ptr++;
5480 }
5481 namelen = (int)(ptr - name);
5482
5483 if ((terminator > 0 && *ptr++ != terminator) ||
5484 *ptr++ != CHAR_RIGHT_PARENTHESIS)
5485 {
5486 ptr--; /* Error offset */
5487 *errorcodeptr = ERR26;
5488 goto FAILED;
5489 }
5490
5491 /* Do no further checking in the pre-compile phase. */
5492
5493 if (lengthptr != NULL) break;
5494
5495 /* In the real compile we do the work of looking for the actual
5496 reference. If the string started with "+" or "-" we require the rest to
5497 be digits, in which case recno will be set. */
5498
5499 if (refsign > 0)
5500 {
5501 if (recno <= 0)
5502 {
5503 *errorcodeptr = ERR58;
5504 goto FAILED;
5505 }
5506 recno = (refsign == CHAR_MINUS)?
5507 cd->bracount - recno + 1 : recno +cd->bracount;
5508 if (recno <= 0 || recno > cd->final_bracount)
5509 {
5510 *errorcodeptr = ERR15;
5511 goto FAILED;
5512 }
5513 PUT2(code, 2+LINK_SIZE, recno);
5514 break;
5515 }
5516
5517 /* Otherwise (did not start with "+" or "-"), start by looking for the
5518 name. If we find a name, add one to the opcode to change OP_CREF or
5519 OP_RREF into OP_NCREF or OP_NRREF. These behave exactly the same,
5520 except they record that the reference was originally to a name. The
5521 information is used to check duplicate names. */
5522
5523 slot = cd->name_table;
5524 for (i = 0; i < cd->names_found; i++)
5525 {
5526 if (STRNCMP_UC_UC(name, slot+IMM2_SIZE, namelen) == 0) break;
5527 slot += cd->name_entry_size;
5528 }
5529
5530 /* Found a previous named subpattern */
5531
5532 if (i < cd->names_found)
5533 {
5534 recno = GET2(slot, 0);
5535 PUT2(code, 2+LINK_SIZE, recno);
5536 code[1+LINK_SIZE]++;
5537 }
5538
5539 /* Search the pattern for a forward reference */
5540
5541 else if ((i = find_parens(cd, name, namelen,
5542 (options & PCRE_EXTENDED) != 0, utf8)) > 0)
5543 {
5544 PUT2(code, 2+LINK_SIZE, i);
5545 code[1+LINK_SIZE]++;
5546 }
5547
5548 /* If terminator == 0 it means that the name followed directly after
5549 the opening parenthesis [e.g. (?(abc)...] and in this case there are
5550 some further alternatives to try. For the cases where terminator != 0
5551 [things like (?(<name>... or (?('name')... or (?(R&name)... ] we have
5552 now checked all the possibilities, so give an error. */
5553
5554 else if (terminator != 0)
5555 {
5556 *errorcodeptr = ERR15;
5557 goto FAILED;
5558 }
5559
5560 /* Check for (?(R) for recursion. Allow digits after R to specify a
5561 specific group number. */
5562
5563 else if (*name == CHAR_R)
5564 {
5565 recno = 0;
5566 for (i = 1; i < namelen; i++)
5567 {
5568 if ((digitab[name[i]] & ctype_digit) == 0)
5569 {
5570 *errorcodeptr = ERR15;
5571 goto FAILED;
5572 }
5573 recno = recno * 10 + name[i] - CHAR_0;
5574 }
5575 if (recno == 0) recno = RREF_ANY;
5576 code[1+LINK_SIZE] = OP_RREF; /* Change test type */
5577 PUT2(code, 2+LINK_SIZE, recno);
5578 }
5579
5580 /* Similarly, check for the (?(DEFINE) "condition", which is always
5581 false. */
5582
5583 else if (namelen == 6 && STRNCMP_UC_C8(name, STRING_DEFINE, 6) == 0)
5584 {
5585 code[1+LINK_SIZE] = OP_DEF;
5586 skipbytes = 1;
5587 }
5588
5589 /* Check for the "name" actually being a subpattern number. We are
5590 in the second pass here, so final_bracount is set. */
5591
5592 else if (recno > 0 && recno <= cd->final_bracount)
5593 {
5594 PUT2(code, 2+LINK_SIZE, recno);
5595 }
5596
5597 /* Either an unidentified subpattern, or a reference to (?(0) */
5598
5599 else
5600 {
5601 *errorcodeptr = (recno == 0)? ERR35: ERR15;
5602 goto FAILED;
5603 }
5604 break;
5605
5606
5607 /* ------------------------------------------------------------ */
5608 case CHAR_EQUALS_SIGN: /* Positive lookahead */
5609 bravalue = OP_ASSERT;
5610 cd->assert_depth += 1;
5611 ptr++;
5612 break;
5613
5614
5615 /* ------------------------------------------------------------ */
5616 case CHAR_EXCLAMATION_MARK: /* Negative lookahead */
5617 ptr++;
5618 if (*ptr == CHAR_RIGHT_PARENTHESIS) /* Optimize (?!) */
5619 {
5620 *code++ = OP_FAIL;
5621 previous = NULL;
5622 continue;
5623 }
5624 bravalue = OP_ASSERT_NOT;
5625 cd->assert_depth += 1;
5626 break;
5627
5628
5629 /* ------------------------------------------------------------ */
5630 case CHAR_LESS_THAN_SIGN: /* Lookbehind or named define */
5631 switch (ptr[1])
5632 {
5633 case CHAR_EQUALS_SIGN: /* Positive lookbehind */
5634 bravalue = OP_ASSERTBACK;
5635 cd->assert_depth += 1;
5636 ptr += 2;
5637 break;
5638
5639 case CHAR_EXCLAMATION_MARK: /* Negative lookbehind */
5640 bravalue = OP_ASSERTBACK_NOT;
5641 cd->assert_depth += 1;
5642 ptr += 2;
5643 break;
5644
5645 default: /* Could be name define, else bad */
5646 if ((cd->ctypes[ptr[1]] & ctype_word) != 0) goto DEFINE_NAME;
5647 ptr++; /* Correct offset for error */
5648 *errorcodeptr = ERR24;
5649 goto FAILED;
5650 }
5651 break;
5652
5653
5654 /* ------------------------------------------------------------ */
5655 case CHAR_GREATER_THAN_SIGN: /* One-time brackets */
5656 bravalue = OP_ONCE;
5657 ptr++;
5658 break;
5659
5660
5661 /* ------------------------------------------------------------ */
5662 case CHAR_C: /* Callout - may be followed by digits; */
5663 previous_callout = code; /* Save for later completion */
5664 after_manual_callout = 1; /* Skip one item before completing */
5665 *code++ = OP_CALLOUT;
5666 {
5667 int n = 0;
5668 while ((digitab[*(++ptr)] & ctype_digit) != 0)
5669 n = n * 10 + *ptr - CHAR_0;
5670 if (*ptr != CHAR_RIGHT_PARENTHESIS)
5671 {
5672 *errorcodeptr = ERR39;
5673 goto FAILED;
5674 }
5675 if (n > 255)
5676 {
5677 *errorcodeptr = ERR38;
5678 goto FAILED;
5679 }
5680 *code++ = n;
5681 PUT(code, 0, (int)(ptr - cd->start_pattern + 1)); /* Pattern offset */
5682 PUT(code, LINK_SIZE, 0); /* Default length */
5683 code += 2 * LINK_SIZE;
5684 }
5685 previous = NULL;
5686 continue;
5687
5688
5689 /* ------------------------------------------------------------ */
5690 case CHAR_P: /* Python-style named subpattern handling */
5691 if (*(++ptr) == CHAR_EQUALS_SIGN ||
5692 *ptr == CHAR_GREATER_THAN_SIGN) /* Reference or recursion */
5693 {
5694 is_recurse = *ptr == CHAR_GREATER_THAN_SIGN;
5695 terminator = CHAR_RIGHT_PARENTHESIS;
5696 goto NAMED_REF_OR_RECURSE;
5697 }
5698 else if (*ptr != CHAR_LESS_THAN_SIGN) /* Test for Python-style defn */
5699 {
5700 *errorcodeptr = ERR41;
5701 goto FAILED;
5702 }
5703 /* Fall through to handle (?P< as (?< is handled */
5704
5705
5706 /* ------------------------------------------------------------ */
5707 DEFINE_NAME: /* Come here from (?< handling */
5708 case CHAR_APOSTROPHE:
5709 {
5710 terminator = (*ptr == CHAR_LESS_THAN_SIGN)?
5711 CHAR_GREATER_THAN_SIGN : CHAR_APOSTROPHE;
5712 name = ++ptr;
5713
5714 while ((cd->ctypes[*ptr] & ctype_word) != 0) ptr++;
5715 namelen = (int)(ptr - name);
5716
5717 /* In the pre-compile phase, just do a syntax check. */
5718
5719 if (lengthptr != NULL)
5720 {
5721 if (*ptr != terminator)
5722 {
5723 *errorcodeptr = ERR42;
5724 goto FAILED;
5725 }
5726 if (cd->names_found >= MAX_NAME_COUNT)
5727 {
5728 *errorcodeptr = ERR49;
5729 goto FAILED;
5730 }
5731 if (namelen + 3 > cd->name_entry_size)
5732 {
5733 cd->name_entry_size = namelen + 3;
5734 if (namelen > MAX_NAME_SIZE)
5735 {
5736 *errorcodeptr = ERR48;
5737 goto FAILED;
5738 }
5739 }
5740 }
5741
5742 /* In the real compile, create the entry in the table, maintaining
5743 alphabetical order. Duplicate names for different numbers are
5744 permitted only if PCRE_DUPNAMES is set. Duplicate names for the same
5745 number are always OK. (An existing number can be re-used if (?|
5746 appears in the pattern.) In either event, a duplicate name results in
5747 a duplicate entry in the table, even if the number is the same. This
5748 is because the number of names, and hence the table size, is computed
5749 in the pre-compile, and it affects various numbers and pointers which
5750 would all have to be modified, and the compiled code moved down, if
5751 duplicates with the same number were omitted from the table. This
5752 doesn't seem worth the hassle. However, *different* names for the
5753 same number are not permitted. */
5754
5755 else
5756 {
5757 BOOL dupname = FALSE;
5758 slot = cd->name_table;
5759
5760 for (i = 0; i < cd->names_found; i++)
5761 {
5762 int crc = memcmp(name, slot+2, namelen);
5763 if (crc == 0)
5764 {
5765 if (slot[2+namelen] == 0)
5766 {
5767 if (GET2(slot, 0) != cd->bracount + 1 &&
5768 (options & PCRE_DUPNAMES) == 0)
5769 {
5770 *errorcodeptr = ERR43;
5771 goto FAILED;
5772 }
5773 else dupname = TRUE;
5774 }
5775 else crc = -1; /* Current name is a substring */
5776 }
5777
5778 /* Make space in the table and break the loop for an earlier
5779 name. For a duplicate or later name, carry on. We do this for
5780 duplicates so that in the simple case (when ?(| is not used) they
5781 are in order of their numbers. */
5782
5783 if (crc < 0)
5784 {
5785 memmove(slot + cd->name_entry_size, slot,
5786 IN_UCHARS((cd->names_found - i) * cd->name_entry_size));
5787 break;
5788 }
5789
5790 /* Continue the loop for a later or duplicate name */
5791
5792 slot += cd->name_entry_size;
5793 }
5794
5795 /* For non-duplicate names, check for a duplicate number before
5796 adding the new name. */
5797
5798 if (!dupname)
5799 {
5800 pcre_uchar *cslot = cd->name_table;
5801 for (i = 0; i < cd->names_found; i++)
5802 {
5803 if (cslot != slot)
5804 {
5805 if (GET2(cslot, 0) == cd->bracount + 1)
5806 {
5807 *errorcodeptr = ERR65;
5808 goto FAILED;
5809 }
5810 }
5811 else i--;
5812 cslot += cd->name_entry_size;
5813 }
5814 }
5815
5816 PUT2(slot, 0, cd->bracount + 1);
5817 memcpy(slot + 2, name, IN_UCHARS(namelen));
5818 slot[2 + namelen] = 0;
5819 }
5820 }
5821
5822 /* In both pre-compile and compile, count the number of names we've
5823 encountered. */
5824
5825 cd->names_found++;
5826 ptr++; /* Move past > or ' */
5827 goto NUMBERED_GROUP;
5828
5829
5830 /* ------------------------------------------------------------ */
5831 case CHAR_AMPERSAND: /* Perl recursion/subroutine syntax */
5832 terminator = CHAR_RIGHT_PARENTHESIS;
5833 is_recurse = TRUE;
5834 /* Fall through */
5835
5836 /* We come here from the Python syntax above that handles both
5837 references (?P=name) and recursion (?P>name), as well as falling
5838 through from the Perl recursion syntax (?&name). We also come here from
5839 the Perl \k<name> or \k'name' back reference syntax and the \k{name}
5840 .NET syntax, and the Oniguruma \g<...> and \g'...' subroutine syntax. */
5841
5842 NAMED_REF_OR_RECURSE:
5843 name = ++ptr;
5844 while ((cd->ctypes[*ptr] & ctype_word) != 0) ptr++;
5845 namelen = (int)(ptr - name);
5846
5847 /* In the pre-compile phase, do a syntax check. We used to just set
5848 a dummy reference number, because it was not used in the first pass.
5849 However, with the change of recursive back references to be atomic,
5850 we have to look for the number so that this state can be identified, as
5851 otherwise the incorrect length is computed. If it's not a backwards
5852 reference, the dummy number will do. */
5853
5854 if (lengthptr != NULL)
5855 {
5856 const pcre_uchar *temp;
5857
5858 if (namelen == 0)
5859 {
5860 *errorcodeptr = ERR62;
5861 goto FAILED;
5862 }
5863 if (*ptr != terminator)
5864 {
5865 *errorcodeptr = ERR42;
5866 goto FAILED;
5867 }
5868 if (namelen > MAX_NAME_SIZE)
5869 {
5870 *errorcodeptr = ERR48;
5871 goto FAILED;
5872 }
5873
5874 /* The name table does not exist in the first pass, so we cannot
5875 do a simple search as in the code below. Instead, we have to scan the
5876 pattern to find the number. It is important that we scan it only as
5877 far as we have got because the syntax of named subpatterns has not
5878 been checked for the rest of the pattern, and find_parens() assumes
5879 correct syntax. In any case, it's a waste of resources to scan
5880 further. We stop the scan at the current point by temporarily
5881 adjusting the value of cd->endpattern. */
5882
5883 temp = cd->end_pattern;
5884 cd->end_pattern = ptr;
5885 recno = find_parens(cd, name, namelen,
5886 (options & PCRE_EXTENDED) != 0, utf8);
5887 cd->end_pattern = temp;
5888 if (recno < 0) recno = 0; /* Forward ref; set dummy number */
5889 }
5890
5891 /* In the real compile, seek the name in the table. We check the name
5892 first, and then check that we have reached the end of the name in the
5893 table. That way, if the name that is longer than any in the table,
5894 the comparison will fail without reading beyond the table entry. */
5895
5896 else
5897 {
5898 slot = cd->name_table;
5899 for (i = 0; i < cd->names_found; i++)
5900 {
5901 if (STRNCMP_UC_UC(name, slot+IMM2_SIZE, namelen) == 0 &&
5902 slot[2+namelen] == 0)
5903 break;
5904 slot += cd->name_entry_size;
5905 }
5906
5907 if (i < cd->names_found) /* Back reference */
5908 {
5909 recno = GET2(slot, 0);
5910 }
5911 else if ((recno = /* Forward back reference */
5912 find_parens(cd, name, namelen,
5913 (options & PCRE_EXTENDED) != 0, utf8)) <= 0)
5914 {
5915 *errorcodeptr = ERR15;
5916 goto FAILED;
5917 }
5918 }
5919
5920 /* In both phases, we can now go to the code than handles numerical
5921 recursion or backreferences. */
5922
5923 if (is_recurse) goto HANDLE_RECURSION;
5924 else goto HANDLE_REFERENCE;
5925
5926
5927 /* ------------------------------------------------------------ */
5928 case CHAR_R: /* Recursion */
5929 ptr++; /* Same as (?0) */
5930 /* Fall through */
5931
5932
5933 /* ------------------------------------------------------------ */
5934 case CHAR_MINUS: case CHAR_PLUS: /* Recursion or subroutine */
5935 case CHAR_0: case CHAR_1: case CHAR_2: case CHAR_3: case CHAR_4:
5936 case CHAR_5: case CHAR_6: case CHAR_7: case CHAR_8: case CHAR_9:
5937 {
5938 const pcre_uchar *called;
5939 terminator = CHAR_RIGHT_PARENTHESIS;
5940
5941 /* Come here from the \g<...> and \g'...' code (Oniguruma
5942 compatibility). However, the syntax has been checked to ensure that
5943 the ... are a (signed) number, so that neither ERR63 nor ERR29 will
5944 be called on this path, nor with the jump to OTHER_CHAR_AFTER_QUERY
5945 ever be taken. */
5946
5947 HANDLE_NUMERICAL_RECURSION:
5948
5949 if ((refsign = *ptr) == CHAR_PLUS)
5950 {
5951 ptr++;
5952 if ((digitab[*ptr] & ctype_digit) == 0)
5953 {
5954 *errorcodeptr = ERR63;
5955 goto FAILED;
5956 }
5957 }
5958 else if (refsign == CHAR_MINUS)
5959 {
5960 if ((digitab[ptr[1]] & ctype_digit) == 0)
5961 goto OTHER_CHAR_AFTER_QUERY;
5962 ptr++;
5963 }
5964
5965 recno = 0;
5966 while((digitab[*ptr] & ctype_digit) != 0)
5967 recno = recno * 10 + *ptr++ - CHAR_0;
5968
5969 if (*ptr != terminator)
5970 {
5971 *errorcodeptr = ERR29;
5972 goto FAILED;
5973 }
5974
5975 if (refsign == CHAR_MINUS)
5976 {
5977 if (recno == 0)
5978 {
5979 *errorcodeptr = ERR58;
5980 goto FAILED;
5981 }
5982 recno = cd->bracount - recno + 1;
5983 if (recno <= 0)
5984 {
5985 *errorcodeptr = ERR15;
5986 goto FAILED;
5987 }
5988 }
5989 else if (refsign == CHAR_PLUS)
5990 {
5991 if (recno == 0)
5992 {
5993 *errorcodeptr = ERR58;
5994 goto FAILED;
5995 }
5996 recno += cd->bracount;
5997 }
5998
5999 /* Come here from code above that handles a named recursion */
6000
6001 HANDLE_RECURSION:
6002
6003 previous = code;
6004 called = cd->start_code;
6005
6006 /* When we are actually compiling, find the bracket that is being
6007 referenced. Temporarily end the regex in case it doesn't exist before
6008 this point. If we end up with a forward reference, first check that
6009 the bracket does occur later so we can give the error (and position)
6010 now. Then remember this forward reference in the workspace so it can
6011 be filled in at the end. */
6012
6013 if (lengthptr == NULL)
6014 {
6015 *code = OP_END;
6016 if (recno != 0)
6017 called = PRIV(find_bracket)(cd->start_code, utf8, recno);
6018
6019 /* Forward reference */
6020
6021 if (called == NULL)
6022 {
6023 if (find_parens(cd, NULL, recno,
6024 (options & PCRE_EXTENDED) != 0, utf8) < 0)
6025 {
6026 *errorcodeptr = ERR15;
6027 goto FAILED;
6028 }
6029
6030 /* Fudge the value of "called" so that when it is inserted as an
6031 offset below, what it actually inserted is the reference number
6032 of the group. Then remember the forward reference. */
6033
6034 called = cd->start_code + recno;
6035 PUTINC(cd->hwm, 0, (int)(code + 1 - cd->start_code));
6036 }
6037
6038 /* If not a forward reference, and the subpattern is still open,
6039 this is a recursive call. We check to see if this is a left
6040 recursion that could loop for ever, and diagnose that case. We
6041 must not, however, do this check if we are in a conditional
6042 subpattern because the condition might be testing for recursion in
6043 a pattern such as /(?(R)a+|(?R)b)/, which is perfectly valid.
6044 Forever loops are also detected at runtime, so those that occur in
6045 conditional subpatterns will be picked up then. */
6046
6047 else if (GET(called, 1) == 0 && cond_depth <= 0 &&
6048 could_be_empty(called, code, bcptr, utf8, cd))
6049 {
6050 *errorcodeptr = ERR40;
6051 goto FAILED;
6052 }
6053 }
6054
6055 /* Insert the recursion/subroutine item. */
6056
6057 *code = OP_RECURSE;
6058 PUT(code, 1, (int)(called - cd->start_code));
6059 code += 1 + LINK_SIZE;
6060 }
6061
6062 /* Can't determine a first byte now */
6063
6064 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
6065 continue;
6066
6067
6068 /* ------------------------------------------------------------ */
6069 default: /* Other characters: check option setting */
6070 OTHER_CHAR_AFTER_QUERY:
6071 set = unset = 0;
6072 optset = &set;
6073
6074 while (*ptr != CHAR_RIGHT_PARENTHESIS && *ptr != CHAR_COLON)
6075 {
6076 switch (*ptr++)
6077 {
6078 case CHAR_MINUS: optset = &unset; break;
6079
6080 case CHAR_J: /* Record that it changed in the external options */
6081 *optset |= PCRE_DUPNAMES;
6082 cd->external_flags |= PCRE_JCHANGED;
6083 break;
6084
6085 case CHAR_i: *optset |= PCRE_CASELESS; break;
6086 case CHAR_m: *optset |= PCRE_MULTILINE; break;
6087 case CHAR_s: *optset |= PCRE_DOTALL; break;
6088 case CHAR_x: *optset |= PCRE_EXTENDED; break;
6089 case CHAR_U: *optset |= PCRE_UNGREEDY; break;
6090 case CHAR_X: *optset |= PCRE_EXTRA; break;
6091
6092 default: *errorcodeptr = ERR12;
6093 ptr--; /* Correct the offset */
6094 goto FAILED;
6095 }
6096 }
6097
6098 /* Set up the changed option bits, but don't change anything yet. */
6099
6100 newoptions = (options | set) & (~unset);
6101
6102 /* If the options ended with ')' this is not the start of a nested
6103 group with option changes, so the options change at this level. If this
6104 item is right at the start of the pattern, the options can be
6105 abstracted and made external in the pre-compile phase, and ignored in
6106 the compile phase. This can be helpful when matching -- for instance in
6107 caseless checking of required bytes.
6108
6109 If the code pointer is not (cd->start_code + 1 + LINK_SIZE), we are
6110 definitely *not* at the start of the pattern because something has been
6111 compiled. In the pre-compile phase, however, the code pointer can have
6112 that value after the start, because it gets reset as code is discarded
6113 during the pre-compile. However, this can happen only at top level - if
6114 we are within parentheses, the starting BRA will still be present. At
6115 any parenthesis level, the length value can be used to test if anything
6116 has been compiled at that level. Thus, a test for both these conditions
6117 is necessary to ensure we correctly detect the start of the pattern in
6118 both phases.
6119
6120 If we are not at the pattern start, reset the greedy defaults and the
6121 case value for firstbyte and reqbyte. */
6122
6123 if (*ptr == CHAR_RIGHT_PARENTHESIS)
6124 {
6125 if (code == cd->start_code + 1 + LINK_SIZE &&
6126 (lengthptr == NULL || *lengthptr == 2 + 2*LINK_SIZE))
6127 {
6128 cd->external_options = newoptions;
6129 }
6130 else
6131 {
6132 greedy_default = ((newoptions & PCRE_UNGREEDY) != 0);
6133 greedy_non_default = greedy_default ^ 1;
6134 req_caseopt = ((newoptions & PCRE_CASELESS) != 0)? REQ_CASELESS : 0;
6135 }
6136
6137 /* Change options at this level, and pass them back for use
6138 in subsequent branches. */
6139
6140 *optionsptr = options = newoptions;
6141 previous = NULL; /* This item can't be repeated */
6142 continue; /* It is complete */
6143 }
6144
6145 /* If the options ended with ':' we are heading into a nested group
6146 with possible change of options. Such groups are non-capturing and are
6147 not assertions of any kind. All we need to do is skip over the ':';
6148 the newoptions value is handled below. */
6149
6150 bravalue = OP_BRA;
6151 ptr++;
6152 } /* End of switch for character following (? */
6153 } /* End of (? handling */
6154
6155 /* Opening parenthesis not followed by '*' or '?'. If PCRE_NO_AUTO_CAPTURE
6156 is set, all unadorned brackets become non-capturing and behave like (?:...)
6157 brackets. */
6158
6159 else if ((options & PCRE_NO_AUTO_CAPTURE) != 0)
6160 {
6161 bravalue = OP_BRA;
6162 }
6163
6164 /* Else we have a capturing group. */
6165
6166 else
6167 {
6168 NUMBERED_GROUP:
6169 cd->bracount += 1;
6170 PUT2(code, 1+LINK_SIZE, cd->bracount);
6171 skipbytes = IMM2_SIZE;
6172 }
6173
6174 /* Process nested bracketed regex. Assertions used not to be repeatable,
6175 but this was changed for Perl compatibility, so all kinds can now be
6176 repeated. We copy code into a non-register variable (tempcode) in order to
6177 be able to pass its address because some compilers complain otherwise. */
6178
6179 previous = code; /* For handling repetition */
6180 *code = bravalue;
6181 tempcode = code;
6182 tempreqvary = cd->req_varyopt; /* Save value before bracket */
6183 tempbracount = cd->bracount; /* Save value before bracket */
6184 length_prevgroup = 0; /* Initialize for pre-compile phase */
6185
6186 if (!compile_regex(
6187 newoptions, /* The complete new option state */
6188 &tempcode, /* Where to put code (updated) */
6189 &ptr, /* Input pointer (updated) */
6190 errorcodeptr, /* Where to put an error message */
6191 (bravalue == OP_ASSERTBACK ||
6192 bravalue == OP_ASSERTBACK_NOT), /* TRUE if back assert */
6193 reset_bracount, /* True if (?| group */
6194 skipbytes, /* Skip over bracket number */
6195 cond_depth +
6196 ((bravalue == OP_COND)?1:0), /* Depth of condition subpatterns */
6197 &subfirstbyte, /* For possible first char */
6198 &subreqbyte, /* For possible last char */
6199 bcptr, /* Current branch chain */
6200 cd, /* Tables block */
6201 (lengthptr == NULL)? NULL : /* Actual compile phase */
6202 &length_prevgroup /* Pre-compile phase */
6203 ))
6204 goto FAILED;
6205
6206 /* If this was an atomic group and there are no capturing groups within it,
6207 generate OP_ONCE_NC instead of OP_ONCE. */
6208
6209 if (bravalue == OP_ONCE && cd->bracount <= tempbracount)
6210 *code = OP_ONCE_NC;
6211
6212 if (bravalue >= OP_ASSERT && bravalue <= OP_ASSERTBACK_NOT)
6213 cd->assert_depth -= 1;
6214
6215 /* At the end of compiling, code is still pointing to the start of the
6216 group, while tempcode has been updated to point past the end of the group.
6217 The pattern pointer (ptr) is on the bracket.
6218
6219 If this is a conditional bracket, check that there are no more than
6220 two branches in the group, or just one if it's a DEFINE group. We do this
6221 in the real compile phase, not in the pre-pass, where the whole group may
6222 not be available. */
6223
6224 if (bravalue == OP_COND && lengthptr == NULL)
6225 {
6226 pcre_uchar *tc = code;
6227 int condcount = 0;
6228
6229 do {
6230 condcount++;
6231 tc += GET(tc,1);
6232 }
6233 while (*tc != OP_KET);
6234
6235 /* A DEFINE group is never obeyed inline (the "condition" is always
6236 false). It must have only one branch. */
6237
6238 if (code[LINK_SIZE+1] == OP_DEF)
6239 {
6240 if (condcount > 1)
6241 {
6242 *errorcodeptr = ERR54;
6243 goto FAILED;
6244 }
6245 bravalue = OP_DEF; /* Just a flag to suppress char handling below */
6246 }
6247
6248 /* A "normal" conditional group. If there is just one branch, we must not
6249 make use of its firstbyte or reqbyte, because this is equivalent to an
6250 empty second branch. */
6251
6252 else
6253 {
6254 if (condcount > 2)
6255 {
6256 *errorcodeptr = ERR27;
6257 goto FAILED;
6258 }
6259 if (condcount == 1) subfirstbyte = subreqbyte = REQ_NONE;
6260 }
6261 }
6262
6263 /* Error if hit end of pattern */
6264
6265 if (*ptr != CHAR_RIGHT_PARENTHESIS)
6266 {
6267 *errorcodeptr = ERR14;
6268 goto FAILED;
6269 }
6270
6271 /* In the pre-compile phase, update the length by the length of the group,
6272 less the brackets at either end. Then reduce the compiled code to just a
6273 set of non-capturing brackets so that it doesn't use much memory if it is
6274 duplicated by a quantifier.*/
6275
6276 if (lengthptr != NULL)
6277 {
6278 if (OFLOW_MAX - *lengthptr < length_prevgroup - 2 - 2*LINK_SIZE)
6279 {
6280 *errorcodeptr = ERR20;
6281 goto FAILED;
6282 }
6283 *lengthptr += length_prevgroup - 2 - 2*LINK_SIZE;
6284 code++; /* This already contains bravalue */
6285 PUTINC(code, 0, 1 + LINK_SIZE);
6286 *code++ = OP_KET;
6287 PUTINC(code, 0, 1 + LINK_SIZE);
6288 break; /* No need to waste time with special character handling */
6289 }
6290
6291 /* Otherwise update the main code pointer to the end of the group. */
6292
6293 code = tempcode;
6294
6295 /* For a DEFINE group, required and first character settings are not
6296 relevant. */
6297
6298 if (bravalue == OP_DEF) break;
6299
6300 /* Handle updating of the required and first characters for other types of
6301 group. Update for normal brackets of all kinds, and conditions with two
6302 branches (see code above). If the bracket is followed by a quantifier with
6303 zero repeat, we have to back off. Hence the definition of zeroreqbyte and
6304 zerofirstbyte outside the main loop so that they can be accessed for the
6305 back off. */
6306
6307 zeroreqbyte = reqbyte;
6308 zerofirstbyte = firstbyte;
6309 groupsetfirstbyte = FALSE;
6310
6311 if (bravalue >= OP_ONCE)
6312 {
6313 /* If we have not yet set a firstbyte in this branch, take it from the
6314 subpattern, remembering that it was set here so that a repeat of more
6315 than one can replicate it as reqbyte if necessary. If the subpattern has
6316 no firstbyte, set "none" for the whole branch. In both cases, a zero
6317 repeat forces firstbyte to "none". */
6318
6319 if (firstbyte == REQ_UNSET)
6320 {
6321 if (subfirstbyte >= 0)
6322 {
6323 firstbyte = subfirstbyte;
6324 groupsetfirstbyte = TRUE;
6325 }
6326 else firstbyte = REQ_NONE;
6327 zerofirstbyte = REQ_NONE;
6328 }
6329
6330 /* If firstbyte was previously set, convert the subpattern's firstbyte
6331 into reqbyte if there wasn't one, using the vary flag that was in
6332 existence beforehand. */
6333
6334 else if (subfirstbyte >= 0 && subreqbyte < 0)
6335 subreqbyte = subfirstbyte | tempreqvary;
6336
6337 /* If the subpattern set a required byte (or set a first byte that isn't
6338 really the first byte - see above), set it. */
6339
6340 if (subreqbyte >= 0) reqbyte = subreqbyte;
6341 }
6342
6343 /* For a forward assertion, we take the reqbyte, if set. This can be
6344 helpful if the pattern that follows the assertion doesn't set a different
6345 char. For example, it's useful for /(?=abcde).+/. We can't set firstbyte
6346 for an assertion, however because it leads to incorrect effect for patterns
6347 such as /(?=a)a.+/ when the "real" "a" would then become a reqbyte instead
6348 of a firstbyte. This is overcome by a scan at the end if there's no
6349 firstbyte, looking for an asserted first char. */
6350
6351 else if (bravalue == OP_ASSERT && subreqbyte >= 0) reqbyte = subreqbyte;
6352 break; /* End of processing '(' */
6353
6354
6355 /* ===================================================================*/
6356 /* Handle metasequences introduced by \. For ones like \d, the ESC_ values
6357 are arranged to be the negation of the corresponding OP_values in the
6358 default case when PCRE_UCP is not set. For the back references, the values
6359 are ESC_REF plus the reference number. Only back references and those types
6360 that consume a character may be repeated. We can test for values between
6361 ESC_b and ESC_Z for the latter; this may have to change if any new ones are
6362 ever created. */
6363
6364 case CHAR_BACKSLASH:
6365 tempptr = ptr;
6366 c = check_escape(&ptr, errorcodeptr, cd->bracount, options, FALSE);
6367 if (*errorcodeptr != 0) goto FAILED;
6368
6369 if (c < 0)
6370 {
6371 if (-c == ESC_Q) /* Handle start of quoted string */
6372 {
6373 if (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
6374 ptr += 2; /* avoid empty string */
6375 else inescq = TRUE;
6376 continue;
6377 }
6378
6379 if (-c == ESC_E) continue; /* Perl ignores an orphan \E */
6380
6381 /* For metasequences that actually match a character, we disable the
6382 setting of a first character if it hasn't already been set. */
6383
6384 if (firstbyte == REQ_UNSET && -c > ESC_b && -c < ESC_Z)
6385 firstbyte = REQ_NONE;
6386
6387 /* Set values to reset to if this is followed by a zero repeat. */
6388
6389 zerofirstbyte = firstbyte;
6390 zeroreqbyte = reqbyte;
6391
6392 /* \g<name> or \g'name' is a subroutine call by name and \g<n> or \g'n'
6393 is a subroutine call by number (Oniguruma syntax). In fact, the value
6394 -ESC_g is returned only for these cases. So we don't need to check for <
6395 or ' if the value is -ESC_g. For the Perl syntax \g{n} the value is
6396 -ESC_REF+n, and for the Perl syntax \g{name} the result is -ESC_k (as
6397 that is a synonym for a named back reference). */
6398
6399 if (-c == ESC_g)
6400 {
6401 const pcre_uchar *p;
6402 save_hwm = cd->hwm; /* Normally this is set when '(' is read */
6403 terminator = (*(++ptr) == CHAR_LESS_THAN_SIGN)?
6404 CHAR_GREATER_THAN_SIGN : CHAR_APOSTROPHE;
6405
6406 /* These two statements stop the compiler for warning about possibly
6407 unset variables caused by the jump to HANDLE_NUMERICAL_RECURSION. In
6408 fact, because we actually check for a number below, the paths that
6409 would actually be in error are never taken. */
6410
6411 skipbytes = 0;
6412 reset_bracount = FALSE;
6413
6414 /* Test for a name */
6415
6416 if (ptr[1] != CHAR_PLUS && ptr[1] != CHAR_MINUS)
6417 {
6418 BOOL isnumber = TRUE;
6419 for (p = ptr + 1; *p != 0 && *p != terminator; p++)
6420 {
6421 if ((cd->ctypes[*p] & ctype_digit) == 0) isnumber = FALSE;
6422 if ((cd->ctypes[*p] & ctype_word) == 0) break;
6423 }
6424 if (*p != terminator)
6425 {
6426 *errorcodeptr = ERR57;
6427 break;
6428 }
6429 if (isnumber)
6430 {
6431 ptr++;
6432 goto HANDLE_NUMERICAL_RECURSION;
6433 }
6434 is_recurse = TRUE;
6435 goto NAMED_REF_OR_RECURSE;
6436 }
6437
6438 /* Test a signed number in angle brackets or quotes. */
6439
6440 p = ptr + 2;
6441 while ((digitab[*p] & ctype_digit) != 0) p++;
6442 if (*p != terminator)
6443 {
6444 *errorcodeptr = ERR57;
6445 break;
6446 }
6447 ptr++;
6448 goto HANDLE_NUMERICAL_RECURSION;
6449 }
6450
6451 /* \k<name> or \k'name' is a back reference by name (Perl syntax).
6452 We also support \k{name} (.NET syntax). */
6453
6454 if (-c == ESC_k)
6455 {
6456 if ((ptr[1] != CHAR_LESS_THAN_SIGN &&
6457 ptr[1] != CHAR_APOSTROPHE && ptr[1] != CHAR_LEFT_CURLY_BRACKET))
6458 {
6459 *errorcodeptr = ERR69;
6460 break;
6461 }
6462 is_recurse = FALSE;
6463 terminator = (*(++ptr) == CHAR_LESS_THAN_SIGN)?
6464 CHAR_GREATER_THAN_SIGN : (*ptr == CHAR_APOSTROPHE)?
6465 CHAR_APOSTROPHE : CHAR_RIGHT_CURLY_BRACKET;
6466 goto NAMED_REF_OR_RECURSE;
6467 }
6468
6469 /* Back references are handled specially; must disable firstbyte if
6470 not set to cope with cases like (?=(\w+))\1: which would otherwise set
6471 ':' later. */
6472
6473 if (-c >= ESC_REF)
6474 {
6475 open_capitem *oc;
6476 recno = -c - ESC_REF;
6477
6478 HANDLE_REFERENCE: /* Come here from named backref handling */
6479 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
6480 previous = code;
6481 *code++ = ((options & PCRE_CASELESS) != 0)? OP_REFI : OP_REF;
6482 PUT2INC(code, 0, recno);
6483 cd->backref_map |= (recno < 32)? (1 << recno) : 1;
6484 if (recno > cd->top_backref) cd->top_backref = recno;
6485
6486 /* Check to see if this back reference is recursive, that it, it
6487 is inside the group that it references. A flag is set so that the
6488 group can be made atomic. */
6489
6490 for (oc = cd->open_caps; oc != NULL; oc = oc->next)
6491 {
6492 if (oc->number == recno)
6493 {
6494 oc->flag = TRUE;
6495 break;
6496 }
6497 }
6498 }
6499
6500 /* So are Unicode property matches, if supported. */
6501
6502 #ifdef SUPPORT_UCP
6503 else if (-c == ESC_P || -c == ESC_p)
6504 {
6505 BOOL negated;
6506 int pdata;
6507 int ptype = get_ucp(&ptr, &negated, &pdata, errorcodeptr);
6508 if (ptype < 0) goto FAILED;
6509 previous = code;
6510 *code++ = ((-c == ESC_p) != negated)? OP_PROP : OP_NOTPROP;
6511 *code++ = ptype;
6512 *code++ = pdata;
6513 }
6514 #else
6515
6516 /* If Unicode properties are not supported, \X, \P, and \p are not
6517 allowed. */
6518
6519 else if (-c == ESC_X || -c == ESC_P || -c == ESC_p)
6520 {
6521 *errorcodeptr = ERR45;
6522 goto FAILED;
6523 }
6524 #endif
6525
6526 /* For the rest (including \X when Unicode properties are supported), we
6527 can obtain the OP value by negating the escape value in the default
6528 situation when PCRE_UCP is not set. When it *is* set, we substitute
6529 Unicode property tests. */
6530
6531 else
6532 {
6533 #ifdef SUPPORT_UCP
6534 if (-c >= ESC_DU && -c <= ESC_wu)
6535 {
6536 nestptr = ptr + 1; /* Where to resume */
6537 ptr = substitutes[-c - ESC_DU] - 1; /* Just before substitute */
6538 }
6539 else
6540 #endif
6541 /* In non-UTF-8 mode, we turn \C into OP_ALLANY instead of OP_ANYBYTE
6542 so that it works in DFA mode and in lookbehinds. */
6543
6544 {
6545 previous = (-c > ESC_b && -c < ESC_Z)? code : NULL;
6546 *code++ = (!utf8 && c == -ESC_C)? OP_ALLANY : -c;
6547 }
6548 }
6549 continue;
6550 }
6551
6552 /* We have a data character whose value is in c. In UTF-8 mode it may have
6553 a value > 127. We set its representation in the length/buffer, and then
6554 handle it as a data character. */
6555
6556 #ifdef SUPPORT_UTF8
6557 if (utf8 && c > 127)
6558 mclength = PRIV(ord2utf8)(c, mcbuffer);
6559 else
6560 #endif
6561
6562 {
6563 mcbuffer[0] = c;
6564 mclength = 1;
6565 }
6566 goto ONE_CHAR;
6567
6568
6569 /* ===================================================================*/
6570 /* Handle a literal character. It is guaranteed not to be whitespace or #
6571 when the extended flag is set. If we are in UTF-8 mode, it may be a
6572 multi-byte literal character. */
6573
6574 default:
6575 NORMAL_CHAR:
6576 mclength = 1;
6577 mcbuffer[0] = c;
6578
6579 #ifdef SUPPORT_UTF8
6580 if (utf8 && c >= 0xc0)
6581 {
6582 while ((ptr[1] & 0xc0) == 0x80)
6583 mcbuffer[mclength++] = *(++ptr);
6584 }
6585 #endif
6586
6587 /* At this point we have the character's bytes in mcbuffer, and the length
6588 in mclength. When not in UTF-8 mode, the length is always 1. */
6589
6590 ONE_CHAR:
6591 previous = code;
6592 *code++ = ((options & PCRE_CASELESS) != 0)? OP_CHARI : OP_CHAR;
6593 for (c = 0; c < mclength; c++) *code++ = mcbuffer[c];
6594
6595 /* Remember if \r or \n were seen */
6596
6597 if (mcbuffer[0] == CHAR_CR || mcbuffer[0] == CHAR_NL)
6598 cd->external_flags |= PCRE_HASCRORLF;
6599
6600 /* Set the first and required bytes appropriately. If no previous first
6601 byte, set it from this character, but revert to none on a zero repeat.
6602 Otherwise, leave the firstbyte value alone, and don't change it on a zero
6603 repeat. */
6604
6605 if (firstbyte == REQ_UNSET)
6606 {
6607 zerofirstbyte = REQ_NONE;
6608 zeroreqbyte = reqbyte;
6609
6610 /* If the character is more than one byte long, we can set firstbyte
6611 only if it is not to be matched caselessly. */
6612
6613 if (mclength == 1 || req_caseopt == 0)
6614 {
6615 firstbyte = mcbuffer[0] | req_caseopt;
6616 if (mclength != 1) reqbyte = code[-1] | cd->req_varyopt;
6617 }
6618 else firstbyte = reqbyte = REQ_NONE;
6619 }
6620
6621 /* firstbyte was previously set; we can set reqbyte only if the length is
6622 1 or the matching is caseful. */
6623
6624 else
6625 {
6626 zerofirstbyte = firstbyte;
6627 zeroreqbyte = reqbyte;
6628 if (mclength == 1 || req_caseopt == 0)
6629 reqbyte = code[-1] | req_caseopt | cd->req_varyopt;
6630 }
6631
6632 break; /* End of literal character handling */
6633 }
6634 } /* end of big loop */
6635
6636
6637 /* Control never reaches here by falling through, only by a goto for all the
6638 error states. Pass back the position in the pattern so that it can be displayed
6639 to the user for diagnosing the error. */
6640
6641 FAILED:
6642 *ptrptr = ptr;
6643 return FALSE;
6644 }