/[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 759 - (show annotations) (download)
Mon Nov 21 18:57:34 2011 UTC (3 years ago) by zherczeg
File MIME type: text/plain
File size: 254383 byte(s)
remove double pcre_uint8 declaration and fix some string literals
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 char 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 = _pcre_utt_size;
1117
1118 while (bot < top)
1119 {
1120 i = (bot + top) >> 1;
1121 c = strcmp(name, _pcre_utt_names + _pcre_utt[i].name_offset);
1122 if (c == 0)
1123 {
1124 *dptr = _pcre_utt[i].value;
1125 return _pcre_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((const char *)name, (const char *)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((const char *)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 += _pcre_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 += _pcre_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)? 2:0), 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] + _pcre_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 += _pcre_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 += _pcre_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 += 4;
1725 #ifdef SUPPORT_UTF8
1726 if (utf8 && cc[-1] >= 0xc0) cc += _pcre_utf8_table4[cc[-1] & 0x3f];
1727 #endif
1728 break;
1729
1730 case OP_TYPEEXACT:
1731 branchlength += GET2(cc,1);
1732 if (cc[3] == OP_PROP || cc[3] == OP_NOTPROP) cc += 2;
1733 cc += 4;
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,3)) return -1;
1790 branchlength += GET2(cc,1);
1791 cc += 5;
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 _pcre_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 += _pcre_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 += _pcre_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[3] == OP_PROP || code[3] == OP_NOTPROP) code += 2;
1971 break;
1972
1973 case OP_MARK:
1974 case OP_PRUNE_ARG:
1975 case OP_SKIP_ARG:
1976 code += code[1];
1977 break;
1978
1979 case OP_THEN_ARG:
1980 code += code[1];
1981 break;
1982 }
1983
1984 /* Add in the fixed length from the table */
1985
1986 code += _pcre_OP_lengths[c];
1987
1988 /* In UTF-8 mode, opcodes that are followed by a character may be followed by
1989 a multi-byte character. The length in the table is a minimum, so we have to
1990 arrange to skip the extra bytes. */
1991
1992 #ifdef SUPPORT_UTF8
1993 if (utf8) switch(c)
1994 {
1995 case OP_CHAR:
1996 case OP_CHARI:
1997 case OP_EXACT:
1998 case OP_EXACTI:
1999 case OP_UPTO:
2000 case OP_UPTOI:
2001 case OP_MINUPTO:
2002 case OP_MINUPTOI:
2003 case OP_POSUPTO:
2004 case OP_POSUPTOI:
2005 case OP_STAR:
2006 case OP_STARI:
2007 case OP_MINSTAR:
2008 case OP_MINSTARI:
2009 case OP_POSSTAR:
2010 case OP_POSSTARI:
2011 case OP_PLUS:
2012 case OP_PLUSI:
2013 case OP_MINPLUS:
2014 case OP_MINPLUSI:
2015 case OP_POSPLUS:
2016 case OP_POSPLUSI:
2017 case OP_QUERY:
2018 case OP_QUERYI:
2019 case OP_MINQUERY:
2020 case OP_MINQUERYI:
2021 case OP_POSQUERY:
2022 case OP_POSQUERYI:
2023 if (code[-1] >= 0xc0) code += _pcre_utf8_table4[code[-1] & 0x3f];
2024 break;
2025 }
2026 #else
2027 (void)(utf8); /* Keep compiler happy by referencing function argument */
2028 #endif
2029 }
2030 }
2031 }
2032
2033
2034
2035 /*************************************************
2036 * Scan compiled regex for recursion reference *
2037 *************************************************/
2038
2039 /* This little function scans through a compiled pattern until it finds an
2040 instance of OP_RECURSE.
2041
2042 Arguments:
2043 code points to start of expression
2044 utf8 TRUE in UTF-8 mode
2045
2046 Returns: pointer to the opcode for OP_RECURSE, or NULL if not found
2047 */
2048
2049 static const pcre_uchar *
2050 find_recurse(const pcre_uchar *code, BOOL utf8)
2051 {
2052 for (;;)
2053 {
2054 register int c = *code;
2055 if (c == OP_END) return NULL;
2056 if (c == OP_RECURSE) return code;
2057
2058 /* XCLASS is used for classes that cannot be represented just by a bit
2059 map. This includes negated single high-valued characters. The length in
2060 the table is zero; the actual length is stored in the compiled code. */
2061
2062 if (c == OP_XCLASS) code += GET(code, 1);
2063
2064 /* Otherwise, we can get the item's length from the table, except that for
2065 repeated character types, we have to test for \p and \P, which have an extra
2066 two bytes of parameters, and for MARK/PRUNE/SKIP/THEN with an argument, we
2067 must add in its length. */
2068
2069 else
2070 {
2071 switch(c)
2072 {
2073 case OP_TYPESTAR:
2074 case OP_TYPEMINSTAR:
2075 case OP_TYPEPLUS:
2076 case OP_TYPEMINPLUS:
2077 case OP_TYPEQUERY:
2078 case OP_TYPEMINQUERY:
2079 case OP_TYPEPOSSTAR:
2080 case OP_TYPEPOSPLUS:
2081 case OP_TYPEPOSQUERY:
2082 if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
2083 break;
2084
2085 case OP_TYPEPOSUPTO:
2086 case OP_TYPEUPTO:
2087 case OP_TYPEMINUPTO:
2088 case OP_TYPEEXACT:
2089 if (code[3] == OP_PROP || code[3] == OP_NOTPROP) code += 2;
2090 break;
2091
2092 case OP_MARK:
2093 case OP_PRUNE_ARG:
2094 case OP_SKIP_ARG:
2095 code += code[1];
2096 break;
2097
2098 case OP_THEN_ARG:
2099 code += code[1];
2100 break;
2101 }
2102
2103 /* Add in the fixed length from the table */
2104
2105 code += _pcre_OP_lengths[c];
2106
2107 /* In UTF-8 mode, opcodes that are followed by a character may be followed
2108 by a multi-byte character. The length in the table is a minimum, so we have
2109 to arrange to skip the extra bytes. */
2110
2111 #ifdef SUPPORT_UTF8
2112 if (utf8) switch(c)
2113 {
2114 case OP_CHAR:
2115 case OP_CHARI:
2116 case OP_EXACT:
2117 case OP_EXACTI:
2118 case OP_UPTO:
2119 case OP_UPTOI:
2120 case OP_MINUPTO:
2121 case OP_MINUPTOI:
2122 case OP_POSUPTO:
2123 case OP_POSUPTOI:
2124 case OP_STAR:
2125 case OP_STARI:
2126 case OP_MINSTAR:
2127 case OP_MINSTARI:
2128 case OP_POSSTAR:
2129 case OP_POSSTARI:
2130 case OP_PLUS:
2131 case OP_PLUSI:
2132 case OP_MINPLUS:
2133 case OP_MINPLUSI:
2134 case OP_POSPLUS:
2135 case OP_POSPLUSI:
2136 case OP_QUERY:
2137 case OP_QUERYI:
2138 case OP_MINQUERY:
2139 case OP_MINQUERYI:
2140 case OP_POSQUERY:
2141 case OP_POSQUERYI:
2142 if (code[-1] >= 0xc0) code += _pcre_utf8_table4[code[-1] & 0x3f];
2143 break;
2144 }
2145 #else
2146 (void)(utf8); /* Keep compiler happy by referencing function argument */
2147 #endif
2148 }
2149 }
2150 }
2151
2152
2153
2154 /*************************************************
2155 * Scan compiled branch for non-emptiness *
2156 *************************************************/
2157
2158 /* This function scans through a branch of a compiled pattern to see whether it
2159 can match the empty string or not. It is called from could_be_empty()
2160 below and from compile_branch() when checking for an unlimited repeat of a
2161 group that can match nothing. Note that first_significant_code() skips over
2162 backward and negative forward assertions when its final argument is TRUE. If we
2163 hit an unclosed bracket, we return "empty" - this means we've struck an inner
2164 bracket whose current branch will already have been scanned.
2165
2166 Arguments:
2167 code points to start of search
2168 endcode points to where to stop
2169 utf8 TRUE if in UTF8 mode
2170 cd contains pointers to tables etc.
2171
2172 Returns: TRUE if what is matched could be empty
2173 */
2174
2175 static BOOL
2176 could_be_empty_branch(const pcre_uchar *code, const pcre_uchar *endcode,
2177 BOOL utf8, compile_data *cd)
2178 {
2179 register int c;
2180 for (code = first_significant_code(code + _pcre_OP_lengths[*code], TRUE);
2181 code < endcode;
2182 code = first_significant_code(code + _pcre_OP_lengths[c], TRUE))
2183 {
2184 const pcre_uchar *ccode;
2185
2186 c = *code;
2187
2188 /* Skip over forward assertions; the other assertions are skipped by
2189 first_significant_code() with a TRUE final argument. */
2190
2191 if (c == OP_ASSERT)
2192 {
2193 do code += GET(code, 1); while (*code == OP_ALT);
2194 c = *code;
2195 continue;
2196 }
2197
2198 /* For a recursion/subroutine call, if its end has been reached, which
2199 implies a backward reference subroutine call, we can scan it. If it's a
2200 forward reference subroutine call, we can't. To detect forward reference
2201 we have to scan up the list that is kept in the workspace. This function is
2202 called only when doing the real compile, not during the pre-compile that
2203 measures the size of the compiled pattern. */
2204
2205 if (c == OP_RECURSE)
2206 {
2207 const pcre_uchar *scode;
2208 BOOL empty_branch;
2209
2210 /* Test for forward reference */
2211
2212 for (scode = cd->start_workspace; scode < cd->hwm; scode += LINK_SIZE)
2213 if (GET(scode, 0) == code + 1 - cd->start_code) return TRUE;
2214
2215 /* Not a forward reference, test for completed backward reference */
2216
2217 empty_branch = FALSE;
2218 scode = cd->start_code + GET(code, 1);
2219 if (GET(scode, 1) == 0) return TRUE; /* Unclosed */
2220
2221 /* Completed backwards reference */
2222
2223 do
2224 {
2225 if (could_be_empty_branch(scode, endcode, utf8, cd))
2226 {
2227 empty_branch = TRUE;
2228 break;
2229 }
2230 scode += GET(scode, 1);
2231 }
2232 while (*scode == OP_ALT);
2233
2234 if (!empty_branch) return FALSE; /* All branches are non-empty */
2235 continue;
2236 }
2237
2238 /* Groups with zero repeats can of course be empty; skip them. */
2239
2240 if (c == OP_BRAZERO || c == OP_BRAMINZERO || c == OP_SKIPZERO ||
2241 c == OP_BRAPOSZERO)
2242 {
2243 code += _pcre_OP_lengths[c];
2244 do code += GET(code, 1); while (*code == OP_ALT);
2245 c = *code;
2246 continue;
2247 }
2248
2249 /* A nested group that is already marked as "could be empty" can just be
2250 skipped. */
2251
2252 if (c == OP_SBRA || c == OP_SBRAPOS ||
2253 c == OP_SCBRA || c == OP_SCBRAPOS)
2254 {
2255 do code += GET(code, 1); while (*code == OP_ALT);
2256 c = *code;
2257 continue;
2258 }
2259
2260 /* For other groups, scan the branches. */
2261
2262 if (c == OP_BRA || c == OP_BRAPOS ||
2263 c == OP_CBRA || c == OP_CBRAPOS ||
2264 c == OP_ONCE || c == OP_ONCE_NC ||
2265 c == OP_COND)
2266 {
2267 BOOL empty_branch;
2268 if (GET(code, 1) == 0) return TRUE; /* Hit unclosed bracket */
2269
2270 /* If a conditional group has only one branch, there is a second, implied,
2271 empty branch, so just skip over the conditional, because it could be empty.
2272 Otherwise, scan the individual branches of the group. */
2273
2274 if (c == OP_COND && code[GET(code, 1)] != OP_ALT)
2275 code += GET(code, 1);
2276 else
2277 {
2278 empty_branch = FALSE;
2279 do
2280 {
2281 if (!empty_branch && could_be_empty_branch(code, endcode, utf8, cd))
2282 empty_branch = TRUE;
2283 code += GET(code, 1);
2284 }
2285 while (*code == OP_ALT);
2286 if (!empty_branch) return FALSE; /* All branches are non-empty */
2287 }
2288
2289 c = *code;
2290 continue;
2291 }
2292
2293 /* Handle the other opcodes */
2294
2295 switch (c)
2296 {
2297 /* Check for quantifiers after a class. XCLASS is used for classes that
2298 cannot be represented just by a bit map. This includes negated single
2299 high-valued characters. The length in _pcre_OP_lengths[] is zero; the
2300 actual length is stored in the compiled code, so we must update "code"
2301 here. */
2302
2303 #ifdef SUPPORT_UTF8
2304 case OP_XCLASS:
2305 ccode = code += GET(code, 1);
2306 goto CHECK_CLASS_REPEAT;
2307 #endif
2308
2309 case OP_CLASS:
2310 case OP_NCLASS:
2311 ccode = code + 33;
2312
2313 #ifdef SUPPORT_UTF8
2314 CHECK_CLASS_REPEAT:
2315 #endif
2316
2317 switch (*ccode)
2318 {
2319 case OP_CRSTAR: /* These could be empty; continue */
2320 case OP_CRMINSTAR:
2321 case OP_CRQUERY:
2322 case OP_CRMINQUERY:
2323 break;
2324
2325 default: /* Non-repeat => class must match */
2326 case OP_CRPLUS: /* These repeats aren't empty */
2327 case OP_CRMINPLUS:
2328 return FALSE;
2329
2330 case OP_CRRANGE:
2331 case OP_CRMINRANGE:
2332 if (GET2(ccode, 1) > 0) return FALSE; /* Minimum > 0 */
2333 break;
2334 }
2335 break;
2336
2337 /* Opcodes that must match a character */
2338
2339 case OP_PROP:
2340 case OP_NOTPROP:
2341 case OP_EXTUNI:
2342 case OP_NOT_DIGIT:
2343 case OP_DIGIT:
2344 case OP_NOT_WHITESPACE:
2345 case OP_WHITESPACE:
2346 case OP_NOT_WORDCHAR:
2347 case OP_WORDCHAR:
2348 case OP_ANY:
2349 case OP_ALLANY:
2350 case OP_ANYBYTE:
2351 case OP_CHAR:
2352 case OP_CHARI:
2353 case OP_NOT:
2354 case OP_NOTI:
2355 case OP_PLUS:
2356 case OP_MINPLUS:
2357 case OP_POSPLUS:
2358 case OP_EXACT:
2359 case OP_NOTPLUS:
2360 case OP_NOTMINPLUS:
2361 case OP_NOTPOSPLUS:
2362 case OP_NOTEXACT:
2363 case OP_TYPEPLUS:
2364 case OP_TYPEMINPLUS:
2365 case OP_TYPEPOSPLUS:
2366 case OP_TYPEEXACT:
2367 return FALSE;
2368
2369 /* These are going to continue, as they may be empty, but we have to
2370 fudge the length for the \p and \P cases. */
2371
2372 case OP_TYPESTAR:
2373 case OP_TYPEMINSTAR:
2374 case OP_TYPEPOSSTAR:
2375 case OP_TYPEQUERY:
2376 case OP_TYPEMINQUERY:
2377 case OP_TYPEPOSQUERY:
2378 if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
2379 break;
2380
2381 /* Same for these */
2382
2383 case OP_TYPEUPTO:
2384 case OP_TYPEMINUPTO:
2385 case OP_TYPEPOSUPTO:
2386 if (code[3] == OP_PROP || code[3] == OP_NOTPROP) code += 2;
2387 break;
2388
2389 /* End of branch */
2390
2391 case OP_KET:
2392 case OP_KETRMAX:
2393 case OP_KETRMIN:
2394 case OP_KETRPOS:
2395 case OP_ALT:
2396 return TRUE;
2397
2398 /* In UTF-8 mode, STAR, MINSTAR, POSSTAR, QUERY, MINQUERY, POSQUERY, UPTO,
2399 MINUPTO, and POSUPTO may be followed by a multibyte character */
2400
2401 #ifdef SUPPORT_UTF8
2402 case OP_STAR:
2403 case OP_STARI:
2404 case OP_MINSTAR:
2405 case OP_MINSTARI:
2406 case OP_POSSTAR:
2407 case OP_POSSTARI:
2408 case OP_QUERY:
2409 case OP_QUERYI:
2410 case OP_MINQUERY:
2411 case OP_MINQUERYI:
2412 case OP_POSQUERY:
2413 case OP_POSQUERYI:
2414 if (utf8 && code[1] >= 0xc0) code += _pcre_utf8_table4[code[1] & 0x3f];
2415 break;
2416
2417 case OP_UPTO:
2418 case OP_UPTOI:
2419 case OP_MINUPTO:
2420 case OP_MINUPTOI:
2421 case OP_POSUPTO:
2422 case OP_POSUPTOI:
2423 if (utf8 && code[3] >= 0xc0) code += _pcre_utf8_table4[code[3] & 0x3f];
2424 break;
2425 #endif
2426
2427 /* MARK, and PRUNE/SKIP/THEN with an argument must skip over the argument
2428 string. */
2429
2430 case OP_MARK:
2431 case OP_PRUNE_ARG:
2432 case OP_SKIP_ARG:
2433 code += code[1];
2434 break;
2435
2436 case OP_THEN_ARG:
2437 code += code[1];
2438 break;
2439
2440 /* None of the remaining opcodes are required to match a character. */
2441
2442 default:
2443 break;
2444 }
2445 }
2446
2447 return TRUE;
2448 }
2449
2450
2451
2452 /*************************************************
2453 * Scan compiled regex for non-emptiness *
2454 *************************************************/
2455
2456 /* This function is called to check for left recursive calls. We want to check
2457 the current branch of the current pattern to see if it could match the empty
2458 string. If it could, we must look outwards for branches at other levels,
2459 stopping when we pass beyond the bracket which is the subject of the recursion.
2460 This function is called only during the real compile, not during the
2461 pre-compile.
2462
2463 Arguments:
2464 code points to start of the recursion
2465 endcode points to where to stop (current RECURSE item)
2466 bcptr points to the chain of current (unclosed) branch starts
2467 utf8 TRUE if in UTF-8 mode
2468 cd pointers to tables etc
2469
2470 Returns: TRUE if what is matched could be empty
2471 */
2472
2473 static BOOL
2474 could_be_empty(const pcre_uchar *code, const pcre_uchar *endcode,
2475 branch_chain *bcptr, BOOL utf8, compile_data *cd)
2476 {
2477 while (bcptr != NULL && bcptr->current_branch >= code)
2478 {
2479 if (!could_be_empty_branch(bcptr->current_branch, endcode, utf8, cd))
2480 return FALSE;
2481 bcptr = bcptr->outer;
2482 }
2483 return TRUE;
2484 }
2485
2486
2487
2488 /*************************************************
2489 * Check for POSIX class syntax *
2490 *************************************************/
2491
2492 /* This function is called when the sequence "[:" or "[." or "[=" is
2493 encountered in a character class. It checks whether this is followed by a
2494 sequence of characters terminated by a matching ":]" or ".]" or "=]". If we
2495 reach an unescaped ']' without the special preceding character, return FALSE.
2496
2497 Originally, this function only recognized a sequence of letters between the
2498 terminators, but it seems that Perl recognizes any sequence of characters,
2499 though of course unknown POSIX names are subsequently rejected. Perl gives an
2500 "Unknown POSIX class" error for [:f\oo:] for example, where previously PCRE
2501 didn't consider this to be a POSIX class. Likewise for [:1234:].
2502
2503 The problem in trying to be exactly like Perl is in the handling of escapes. We
2504 have to be sure that [abc[:x\]pqr] is *not* treated as containing a POSIX
2505 class, but [abc[:x\]pqr:]] is (so that an error can be generated). The code
2506 below handles the special case of \], but does not try to do any other escape
2507 processing. This makes it different from Perl for cases such as [:l\ower:]
2508 where Perl recognizes it as the POSIX class "lower" but PCRE does not recognize
2509 "l\ower". This is a lesser evil that not diagnosing bad classes when Perl does,
2510 I think.
2511
2512 A user pointed out that PCRE was rejecting [:a[:digit:]] whereas Perl was not.
2513 It seems that the appearance of a nested POSIX class supersedes an apparent
2514 external class. For example, [:a[:digit:]b:] matches "a", "b", ":", or
2515 a digit.
2516
2517 In Perl, unescaped square brackets may also appear as part of class names. For
2518 example, [:a[:abc]b:] gives unknown POSIX class "[:abc]b:]". However, for
2519 [:a[:abc]b][b:] it gives unknown POSIX class "[:abc]b][b:]", which does not
2520 seem right at all. PCRE does not allow closing square brackets in POSIX class
2521 names.
2522
2523 Arguments:
2524 ptr pointer to the initial [
2525 endptr where to return the end pointer
2526
2527 Returns: TRUE or FALSE
2528 */
2529
2530 static BOOL
2531 check_posix_syntax(const pcre_uchar *ptr, const pcre_uchar **endptr)
2532 {
2533 int terminator; /* Don't combine these lines; the Solaris cc */
2534 terminator = *(++ptr); /* compiler warns about "non-constant" initializer. */
2535 for (++ptr; *ptr != 0; ptr++)
2536 {
2537 if (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET)
2538 ptr++;
2539 else if (*ptr == CHAR_RIGHT_SQUARE_BRACKET) return FALSE;
2540 else
2541 {
2542 if (*ptr == terminator && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET)
2543 {
2544 *endptr = ptr;
2545 return TRUE;
2546 }
2547 if (*ptr == CHAR_LEFT_SQUARE_BRACKET &&
2548 (ptr[1] == CHAR_COLON || ptr[1] == CHAR_DOT ||
2549 ptr[1] == CHAR_EQUALS_SIGN) &&
2550 check_posix_syntax(ptr, endptr))
2551 return FALSE;
2552 }
2553 }
2554 return FALSE;
2555 }
2556
2557
2558
2559
2560 /*************************************************
2561 * Check POSIX class name *
2562 *************************************************/
2563
2564 /* This function is called to check the name given in a POSIX-style class entry
2565 such as [:alnum:].
2566
2567 Arguments:
2568 ptr points to the first letter
2569 len the length of the name
2570
2571 Returns: a value representing the name, or -1 if unknown
2572 */
2573
2574 static int
2575 check_posix_name(const pcre_uchar *ptr, int len)
2576 {
2577 const char *pn = posix_names;
2578 register int yield = 0;
2579 while (posix_name_lengths[yield] != 0)
2580 {
2581 if (len == posix_name_lengths[yield] &&
2582 strncmp((const char *)ptr, pn, len) == 0) return yield;
2583 pn += posix_name_lengths[yield] + 1;
2584 yield++;
2585 }
2586 return -1;
2587 }
2588
2589
2590 /*************************************************
2591 * Adjust OP_RECURSE items in repeated group *
2592 *************************************************/
2593
2594 /* OP_RECURSE items contain an offset from the start of the regex to the group
2595 that is referenced. This means that groups can be replicated for fixed
2596 repetition simply by copying (because the recursion is allowed to refer to
2597 earlier groups that are outside the current group). However, when a group is
2598 optional (i.e. the minimum quantifier is zero), OP_BRAZERO or OP_SKIPZERO is
2599 inserted before it, after it has been compiled. This means that any OP_RECURSE
2600 items within it that refer to the group itself or any contained groups have to
2601 have their offsets adjusted. That one of the jobs of this function. Before it
2602 is called, the partially compiled regex must be temporarily terminated with
2603 OP_END.
2604
2605 This function has been extended with the possibility of forward references for
2606 recursions and subroutine calls. It must also check the list of such references
2607 for the group we are dealing with. If it finds that one of the recursions in
2608 the current group is on this list, it adjusts the offset in the list, not the
2609 value in the reference (which is a group number).
2610
2611 Arguments:
2612 group points to the start of the group
2613 adjust the amount by which the group is to be moved
2614 utf8 TRUE in UTF-8 mode
2615 cd contains pointers to tables etc.
2616 save_hwm the hwm forward reference pointer at the start of the group
2617
2618 Returns: nothing
2619 */
2620
2621 static void
2622 adjust_recurse(pcre_uchar *group, int adjust, BOOL utf8, compile_data *cd,
2623 pcre_uchar *save_hwm)
2624 {
2625 pcre_uchar *ptr = group;
2626
2627 while ((ptr = (pcre_uchar *)find_recurse(ptr, utf8)) != NULL)
2628 {
2629 int offset;
2630 pcre_uchar *hc;
2631
2632 /* See if this recursion is on the forward reference list. If so, adjust the
2633 reference. */
2634
2635 for (hc = save_hwm; hc < cd->hwm; hc += LINK_SIZE)
2636 {
2637 offset = GET(hc, 0);
2638 if (cd->start_code + offset == ptr + 1)
2639 {
2640 PUT(hc, 0, offset + adjust);
2641 break;
2642 }
2643 }
2644
2645 /* Otherwise, adjust the recursion offset if it's after the start of this
2646 group. */
2647
2648 if (hc >= cd->hwm)
2649 {
2650 offset = GET(ptr, 1);
2651 if (cd->start_code + offset >= group) PUT(ptr, 1, offset + adjust);
2652 }
2653
2654 ptr += 1 + LINK_SIZE;
2655 }
2656 }
2657
2658
2659
2660 /*************************************************
2661 * Insert an automatic callout point *
2662 *************************************************/
2663
2664 /* This function is called when the PCRE_AUTO_CALLOUT option is set, to insert
2665 callout points before each pattern item.
2666
2667 Arguments:
2668 code current code pointer
2669 ptr current pattern pointer
2670 cd pointers to tables etc
2671
2672 Returns: new code pointer
2673 */
2674
2675 static pcre_uchar *
2676 auto_callout(pcre_uchar *code, const pcre_uchar *ptr, compile_data *cd)
2677 {
2678 *code++ = OP_CALLOUT;
2679 *code++ = 255;
2680 PUT(code, 0, (int)(ptr - cd->start_pattern)); /* Pattern offset */
2681 PUT(code, LINK_SIZE, 0); /* Default length */
2682 return code + 2 * LINK_SIZE;
2683 }
2684
2685
2686
2687 /*************************************************
2688 * Complete a callout item *
2689 *************************************************/
2690
2691 /* A callout item contains the length of the next item in the pattern, which
2692 we can't fill in till after we have reached the relevant point. This is used
2693 for both automatic and manual callouts.
2694
2695 Arguments:
2696 previous_callout points to previous callout item
2697 ptr current pattern pointer
2698 cd pointers to tables etc
2699
2700 Returns: nothing
2701 */
2702
2703 static void
2704 complete_callout(pcre_uchar *previous_callout, const pcre_uchar *ptr, compile_data *cd)
2705 {
2706 int length = (int)(ptr - cd->start_pattern - GET(previous_callout, 2));
2707 PUT(previous_callout, 2 + LINK_SIZE, length);
2708 }
2709
2710
2711
2712 #ifdef SUPPORT_UCP
2713 /*************************************************
2714 * Get othercase range *
2715 *************************************************/
2716
2717 /* This function is passed the start and end of a class range, in UTF-8 mode
2718 with UCP support. It searches up the characters, looking for internal ranges of
2719 characters in the "other" case. Each call returns the next one, updating the
2720 start address.
2721
2722 Arguments:
2723 cptr points to starting character value; updated
2724 d end value
2725 ocptr where to put start of othercase range
2726 odptr where to put end of othercase range
2727
2728 Yield: TRUE when range returned; FALSE when no more
2729 */
2730
2731 static BOOL
2732 get_othercase_range(unsigned int *cptr, unsigned int d, unsigned int *ocptr,
2733 unsigned int *odptr)
2734 {
2735 unsigned int c, othercase, next;
2736
2737 for (c = *cptr; c <= d; c++)
2738 { if ((othercase = UCD_OTHERCASE(c)) != c) break; }
2739
2740 if (c > d) return FALSE;
2741
2742 *ocptr = othercase;
2743 next = othercase + 1;
2744
2745 for (++c; c <= d; c++)
2746 {
2747 if (UCD_OTHERCASE(c) != next) break;
2748 next++;
2749 }
2750
2751 *odptr = next - 1;
2752 *cptr = c;
2753
2754 return TRUE;
2755 }
2756
2757
2758
2759 /*************************************************
2760 * Check a character and a property *
2761 *************************************************/
2762
2763 /* This function is called by check_auto_possessive() when a property item
2764 is adjacent to a fixed character.
2765
2766 Arguments:
2767 c the character
2768 ptype the property type
2769 pdata the data for the type
2770 negated TRUE if it's a negated property (\P or \p{^)
2771
2772 Returns: TRUE if auto-possessifying is OK
2773 */
2774
2775 static BOOL
2776 check_char_prop(int c, int ptype, int pdata, BOOL negated)
2777 {
2778 const ucd_record *prop = GET_UCD(c);
2779 switch(ptype)
2780 {
2781 case PT_LAMP:
2782 return (prop->chartype == ucp_Lu ||
2783 prop->chartype == ucp_Ll ||
2784 prop->chartype == ucp_Lt) == negated;
2785
2786 case PT_GC:
2787 return (pdata == _pcre_ucp_gentype[prop->chartype]) == negated;
2788
2789 case PT_PC:
2790 return (pdata == prop->chartype) == negated;
2791
2792 case PT_SC:
2793 return (pdata == prop->script) == negated;
2794
2795 /* These are specials */
2796
2797 case PT_ALNUM:
2798 return (_pcre_ucp_gentype[prop->chartype] == ucp_L ||
2799 _pcre_ucp_gentype[prop->chartype] == ucp_N) == negated;
2800
2801 case PT_SPACE: /* Perl space */
2802 return (_pcre_ucp_gentype[prop->chartype] == ucp_Z ||
2803 c == CHAR_HT || c == CHAR_NL || c == CHAR_FF || c == CHAR_CR)
2804 == negated;
2805
2806 case PT_PXSPACE: /* POSIX space */
2807 return (_pcre_ucp_gentype[prop->chartype] == ucp_Z ||
2808 c == CHAR_HT || c == CHAR_NL || c == CHAR_VT ||
2809 c == CHAR_FF || c == CHAR_CR)
2810 == negated;
2811
2812 case PT_WORD:
2813 return (_pcre_ucp_gentype[prop->chartype] == ucp_L ||
2814 _pcre_ucp_gentype[prop->chartype] == ucp_N ||
2815 c == CHAR_UNDERSCORE) == negated;
2816 }
2817 return FALSE;
2818 }
2819 #endif /* SUPPORT_UCP */
2820
2821
2822
2823 /*************************************************
2824 * Check if auto-possessifying is possible *
2825 *************************************************/
2826
2827 /* This function is called for unlimited repeats of certain items, to see
2828 whether the next thing could possibly match the repeated item. If not, it makes
2829 sense to automatically possessify the repeated item.
2830
2831 Arguments:
2832 previous pointer to the repeated opcode
2833 utf8 TRUE in UTF-8 mode
2834 ptr next character in pattern
2835 options options bits
2836 cd contains pointers to tables etc.
2837
2838 Returns: TRUE if possessifying is wanted
2839 */
2840
2841 static BOOL
2842 check_auto_possessive(const pcre_uchar *previous, BOOL utf8,
2843 const pcre_uchar *ptr, int options, compile_data *cd)
2844 {
2845 int c, next;
2846 int op_code = *previous++;
2847
2848 /* Skip whitespace and comments in extended mode */
2849
2850 if ((options & PCRE_EXTENDED) != 0)
2851 {
2852 for (;;)
2853 {
2854 while ((cd->ctypes[*ptr] & ctype_space) != 0) ptr++;
2855 if (*ptr == CHAR_NUMBER_SIGN)
2856 {
2857 ptr++;
2858 while (*ptr != 0)
2859 {
2860 if (IS_NEWLINE(ptr)) { ptr += cd->nllen; break; }
2861 ptr++;
2862 #ifdef SUPPORT_UTF8
2863 if (utf8) while ((*ptr & 0xc0) == 0x80) ptr++;
2864 #endif
2865 }
2866 }
2867 else break;
2868 }
2869 }
2870
2871 /* If the next item is one that we can handle, get its value. A non-negative
2872 value is a character, a negative value is an escape value. */
2873
2874 if (*ptr == CHAR_BACKSLASH)
2875 {
2876 int temperrorcode = 0;
2877 next = check_escape(&ptr, &temperrorcode, cd->bracount, options, FALSE);
2878 if (temperrorcode != 0) return FALSE;
2879 ptr++; /* Point after the escape sequence */
2880 }
2881
2882 else if ((cd->ctypes[*ptr] & ctype_meta) == 0)
2883 {
2884 #ifdef SUPPORT_UTF8
2885 if (utf8) { GETCHARINC(next, ptr); } else
2886 #endif
2887 next = *ptr++;
2888 }
2889
2890 else return FALSE;
2891
2892 /* Skip whitespace and comments in extended mode */
2893
2894 if ((options & PCRE_EXTENDED) != 0)
2895 {
2896 for (;;)
2897 {
2898 while ((cd->ctypes[*ptr] & ctype_space) != 0) ptr++;
2899 if (*ptr == CHAR_NUMBER_SIGN)
2900 {
2901 ptr++;
2902 while (*ptr != 0)
2903 {
2904 if (IS_NEWLINE(ptr)) { ptr += cd->nllen; break; }
2905 ptr++;
2906 #ifdef SUPPORT_UTF8
2907 if (utf8) while ((*ptr & 0xc0) == 0x80) ptr++;
2908 #endif
2909 }
2910 }
2911 else break;
2912 }
2913 }
2914
2915 /* If the next thing is itself optional, we have to give up. */
2916
2917 if (*ptr == CHAR_ASTERISK || *ptr == CHAR_QUESTION_MARK ||
2918 strncmp((char *)ptr, STR_LEFT_CURLY_BRACKET STR_0 STR_COMMA, 3) == 0)
2919 return FALSE;
2920
2921 /* Now compare the next item with the previous opcode. First, handle cases when
2922 the next item is a character. */
2923
2924 if (next >= 0) switch(op_code)
2925 {
2926 case OP_CHAR:
2927 #ifdef SUPPORT_UTF8
2928 GETCHARTEST(c, previous);
2929 #else
2930 c = *previous;
2931 #endif
2932 return c != next;
2933
2934 /* For CHARI (caseless character) we must check the other case. If we have
2935 Unicode property support, we can use it to test the other case of
2936 high-valued characters. */
2937
2938 case OP_CHARI:
2939 #ifdef SUPPORT_UTF8
2940 GETCHARTEST(c, previous);
2941 #else
2942 c = *previous;
2943 #endif
2944 if (c == next) return FALSE;
2945 #ifdef SUPPORT_UTF8
2946 if (utf8)
2947 {
2948 unsigned int othercase;
2949 if (next < 128) othercase = cd->fcc[next]; else
2950 #ifdef SUPPORT_UCP
2951 othercase = UCD_OTHERCASE((unsigned int)next);
2952 #else
2953 othercase = NOTACHAR;
2954 #endif
2955 return (unsigned int)c != othercase;
2956 }
2957 else
2958 #endif /* SUPPORT_UTF8 */
2959 return (c != cd->fcc[next]); /* Non-UTF-8 mode */
2960
2961 /* For OP_NOT and OP_NOTI, the data is always a single-byte character. These
2962 opcodes are not used for multi-byte characters, because they are coded using
2963 an XCLASS instead. */
2964
2965 case OP_NOT:
2966 return (c = *previous) == next;
2967
2968 case OP_NOTI:
2969 if ((c = *previous) == next) return TRUE;
2970 #ifdef SUPPORT_UTF8
2971 if (utf8)
2972 {
2973 unsigned int othercase;
2974 if (next < 128) othercase = cd->fcc[next]; else
2975 #ifdef SUPPORT_UCP
2976 othercase = UCD_OTHERCASE(next);
2977 #else
2978 othercase = NOTACHAR;
2979 #endif
2980 return (unsigned int)c == othercase;
2981 }
2982 else
2983 #endif /* SUPPORT_UTF8 */
2984 return (c == cd->fcc[next]); /* Non-UTF-8 mode */
2985
2986 /* Note that OP_DIGIT etc. are generated only when PCRE_UCP is *not* set.
2987 When it is set, \d etc. are converted into OP_(NOT_)PROP codes. */
2988
2989 case OP_DIGIT:
2990 return next > 127 || (cd->ctypes[next] & ctype_digit) == 0;
2991
2992 case OP_NOT_DIGIT:
2993 return next <= 127 && (cd->ctypes[next] & ctype_digit) != 0;
2994
2995 case OP_WHITESPACE:
2996 return next > 127 || (cd->ctypes[next] & ctype_space) == 0;
2997
2998 case OP_NOT_WHITESPACE:
2999 return next <= 127 && (cd->ctypes[next] & ctype_space) != 0;
3000
3001 case OP_WORDCHAR:
3002 return next > 127 || (cd->ctypes[next] & ctype_word) == 0;
3003
3004 case OP_NOT_WORDCHAR:
3005 return next <= 127 && (cd->ctypes[next] & ctype_word) != 0;
3006
3007 case OP_HSPACE:
3008 case OP_NOT_HSPACE:
3009 switch(next)
3010 {
3011 case 0x09:
3012 case 0x20:
3013 case 0xa0:
3014 case 0x1680:
3015 case 0x180e:
3016 case 0x2000:
3017 case 0x2001:
3018 case 0x2002:
3019 case 0x2003:
3020 case 0x2004:
3021 case 0x2005:
3022 case 0x2006:
3023 case 0x2007:
3024 case 0x2008:
3025 case 0x2009:
3026 case 0x200A:
3027 case 0x202f:
3028 case 0x205f:
3029 case 0x3000:
3030 return op_code == OP_NOT_HSPACE;
3031 default:
3032 return op_code != OP_NOT_HSPACE;
3033 }
3034
3035 case OP_ANYNL:
3036 case OP_VSPACE:
3037 case OP_NOT_VSPACE:
3038 switch(next)
3039 {
3040 case 0x0a:
3041 case 0x0b:
3042 case 0x0c:
3043 case 0x0d:
3044 case 0x85:
3045 case 0x2028:
3046 case 0x2029:
3047 return op_code == OP_NOT_VSPACE;
3048 default:
3049 return op_code != OP_NOT_VSPACE;
3050 }
3051
3052 #ifdef SUPPORT_UCP
3053 case OP_PROP:
3054 return check_char_prop(next, previous[0], previous[1], FALSE);
3055
3056 case OP_NOTPROP:
3057 return check_char_prop(next, previous[0], previous[1], TRUE);
3058 #endif
3059
3060 default:
3061 return FALSE;
3062 }
3063
3064
3065 /* Handle the case when the next item is \d, \s, etc. Note that when PCRE_UCP
3066 is set, \d turns into ESC_du rather than ESC_d, etc., so ESC_d etc. are
3067 generated only when PCRE_UCP is *not* set, that is, when only ASCII
3068 characteristics are recognized. Similarly, the opcodes OP_DIGIT etc. are
3069 replaced by OP_PROP codes when PCRE_UCP is set. */
3070
3071 switch(op_code)
3072 {
3073 case OP_CHAR:
3074 case OP_CHARI:
3075 #ifdef SUPPORT_UTF8
3076 GETCHARTEST(c, previous);
3077 #else
3078 c = *previous;
3079 #endif
3080 switch(-next)
3081 {
3082 case ESC_d:
3083 return c > 127 || (cd->ctypes[c] & ctype_digit) == 0;
3084
3085 case ESC_D:
3086 return c <= 127 && (cd->ctypes[c] & ctype_digit) != 0;
3087
3088 case ESC_s:
3089 return c > 127 || (cd->ctypes[c] & ctype_space) == 0;
3090
3091 case ESC_S:
3092 return c <= 127 && (cd->ctypes[c] & ctype_space) != 0;
3093
3094 case ESC_w:
3095 return c > 127 || (cd->ctypes[c] & ctype_word) == 0;
3096
3097 case ESC_W:
3098 return c <= 127 && (cd->ctypes[c] & ctype_word) != 0;
3099
3100 case ESC_h:
3101 case ESC_H:
3102 switch(c)
3103 {
3104 case 0x09:
3105 case 0x20:
3106 case 0xa0:
3107 case 0x1680:
3108 case 0x180e:
3109 case 0x2000:
3110 case 0x2001:
3111 case 0x2002:
3112 case 0x2003:
3113 case 0x2004:
3114 case 0x2005:
3115 case 0x2006:
3116 case 0x2007:
3117 case 0x2008:
3118 case 0x2009:
3119 case 0x200A:
3120 case 0x202f:
3121 case 0x205f:
3122 case 0x3000:
3123 return -next != ESC_h;
3124 default:
3125 return -next == ESC_h;
3126 }
3127
3128 case ESC_v:
3129 case ESC_V:
3130 switch(c)
3131 {
3132 case 0x0a:
3133 case 0x0b:
3134 case 0x0c:
3135 case 0x0d:
3136 case 0x85:
3137 case 0x2028:
3138 case 0x2029:
3139 return -next != ESC_v;
3140 default:
3141 return -next == ESC_v;
3142 }
3143
3144 /* When PCRE_UCP is set, these values get generated for \d etc. Find
3145 their substitutions and process them. The result will always be either
3146 -ESC_p or -ESC_P. Then fall through to process those values. */
3147
3148 #ifdef SUPPORT_UCP
3149 case ESC_du:
3150 case ESC_DU:
3151 case ESC_wu:
3152 case ESC_WU:
3153 case ESC_su:
3154 case ESC_SU:
3155 {
3156 int temperrorcode = 0;
3157 ptr = substitutes[-next - ESC_DU];
3158 next = check_escape(&ptr, &temperrorcode, 0, options, FALSE);
3159 if (temperrorcode != 0) return FALSE;
3160 ptr++; /* For compatibility */
3161 }
3162 /* Fall through */
3163
3164 case ESC_p:
3165 case ESC_P:
3166 {
3167 int ptype, pdata, errorcodeptr;
3168 BOOL negated;
3169
3170 ptr--; /* Make ptr point at the p or P */
3171 ptype = get_ucp(&ptr, &negated, &pdata, &errorcodeptr);
3172 if (ptype < 0) return FALSE;
3173 ptr++; /* Point past the final curly ket */
3174
3175 /* If the property item is optional, we have to give up. (When generated
3176 from \d etc by PCRE_UCP, this test will have been applied much earlier,
3177 to the original \d etc. At this point, ptr will point to a zero byte. */
3178
3179 if (*ptr == CHAR_ASTERISK || *ptr == CHAR_QUESTION_MARK ||
3180 strncmp((char *)ptr, STR_LEFT_CURLY_BRACKET STR_0 STR_COMMA, 3) == 0)
3181 return FALSE;
3182
3183 /* Do the property check. */
3184
3185 return check_char_prop(c, ptype, pdata, (next == -ESC_P) != negated);
3186 }
3187 #endif
3188
3189 default:
3190 return FALSE;
3191 }
3192
3193 /* In principle, support for Unicode properties should be integrated here as
3194 well. It means re-organizing the above code so as to get hold of the property
3195 values before switching on the op-code. However, I wonder how many patterns
3196 combine ASCII \d etc with Unicode properties? (Note that if PCRE_UCP is set,
3197 these op-codes are never generated.) */
3198
3199 case OP_DIGIT:
3200 return next == -ESC_D || next == -ESC_s || next == -ESC_W ||
3201 next == -ESC_h || next == -ESC_v || next == -ESC_R;
3202
3203 case OP_NOT_DIGIT:
3204 return next == -ESC_d;
3205
3206 case OP_WHITESPACE:
3207 return next == -ESC_S || next == -ESC_d || next == -ESC_w || next == -ESC_R;
3208
3209 case OP_NOT_WHITESPACE:
3210 return next == -ESC_s || next == -ESC_h || next == -ESC_v;
3211
3212 case OP_HSPACE:
3213 return next == -ESC_S || next == -ESC_H || next == -ESC_d ||
3214 next == -ESC_w || next == -ESC_v || next == -ESC_R;
3215
3216 case OP_NOT_HSPACE:
3217 return next == -ESC_h;
3218
3219 /* Can't have \S in here because VT matches \S (Perl anomaly) */
3220 case OP_ANYNL:
3221 case OP_VSPACE:
3222 return next == -ESC_V || next == -ESC_d || next == -ESC_w;
3223
3224 case OP_NOT_VSPACE:
3225 return next == -ESC_v || next == -ESC_R;
3226
3227 case OP_WORDCHAR:
3228 return next == -ESC_W || next == -ESC_s || next == -ESC_h ||
3229 next == -ESC_v || next == -ESC_R;
3230
3231 case OP_NOT_WORDCHAR:
3232 return next == -ESC_w || next == -ESC_d;
3233
3234 default:
3235 return FALSE;
3236 }
3237
3238 /* Control does not reach here */
3239 }
3240
3241
3242
3243 /*************************************************
3244 * Compile one branch *
3245 *************************************************/
3246
3247 /* Scan the pattern, compiling it into the a vector. If the options are
3248 changed during the branch, the pointer is used to change the external options
3249 bits. This function is used during the pre-compile phase when we are trying
3250 to find out the amount of memory needed, as well as during the real compile
3251 phase. The value of lengthptr distinguishes the two phases.
3252
3253 Arguments:
3254 optionsptr pointer to the option bits
3255 codeptr points to the pointer to the current code point
3256 ptrptr points to the current pattern pointer
3257 errorcodeptr points to error code variable
3258 firstbyteptr set to initial literal character, or < 0 (REQ_UNSET, REQ_NONE)
3259 reqbyteptr set to the last literal character required, else < 0
3260 bcptr points to current branch chain
3261 cond_depth conditional nesting depth
3262 cd contains pointers to tables etc.
3263 lengthptr NULL during the real compile phase
3264 points to length accumulator during pre-compile phase
3265
3266 Returns: TRUE on success
3267 FALSE, with *errorcodeptr set non-zero on error
3268 */
3269
3270 static BOOL
3271 compile_branch(int *optionsptr, pcre_uchar **codeptr,
3272 const pcre_uchar **ptrptr, int *errorcodeptr, int *firstbyteptr,
3273 int *reqbyteptr, branch_chain *bcptr, int cond_depth, compile_data *cd,
3274 int *lengthptr)
3275 {
3276 int repeat_type, op_type;
3277 int repeat_min = 0, repeat_max = 0; /* To please picky compilers */
3278 int bravalue = 0;
3279 int greedy_default, greedy_non_default;
3280 int firstbyte, reqbyte;
3281 int zeroreqbyte, zerofirstbyte;
3282 int req_caseopt, reqvary, tempreqvary;
3283 int options = *optionsptr; /* May change dynamically */
3284 int after_manual_callout = 0;
3285 int length_prevgroup = 0;
3286 register int c;
3287 register pcre_uchar *code = *codeptr;
3288 pcre_uchar *last_code = code;
3289 pcre_uchar *orig_code = code;
3290 pcre_uchar *tempcode;
3291 BOOL inescq = FALSE;
3292 BOOL groupsetfirstbyte = FALSE;
3293 const pcre_uchar *ptr = *ptrptr;
3294 const pcre_uchar *tempptr;
3295 const pcre_uchar *nestptr = NULL;
3296 pcre_uchar *previous = NULL;
3297 pcre_uchar *previous_callout = NULL;
3298 pcre_uchar *save_hwm = NULL;
3299 pcre_uchar classbits[32];
3300
3301 /* We can fish out the UTF-8 setting once and for all into a BOOL, but we
3302 must not do this for other options (e.g. PCRE_EXTENDED) because they may change
3303 dynamically as we process the pattern. */
3304
3305 #ifdef SUPPORT_UTF8
3306 BOOL class_utf8;
3307 BOOL utf8 = (options & PCRE_UTF8) != 0;
3308 pcre_uint8 *class_utf8data;
3309 pcre_uint8 *class_utf8data_base;
3310 pcre_uint8 utf8_char[6];
3311 #else
3312 BOOL utf8 = FALSE;
3313 #endif
3314
3315 #ifdef PCRE_DEBUG
3316 if (lengthptr != NULL) DPRINTF((">> start branch\n"));
3317 #endif
3318
3319 /* Set up the default and non-default settings for greediness */
3320
3321 greedy_default = ((options & PCRE_UNGREEDY) != 0);
3322 greedy_non_default = greedy_default ^ 1;
3323
3324 /* Initialize no first byte, no required byte. REQ_UNSET means "no char
3325 matching encountered yet". It gets changed to REQ_NONE if we hit something that
3326 matches a non-fixed char first char; reqbyte just remains unset if we never
3327 find one.
3328
3329 When we hit a repeat whose minimum is zero, we may have to adjust these values
3330 to take the zero repeat into account. This is implemented by setting them to
3331 zerofirstbyte and zeroreqbyte when such a repeat is encountered. The individual
3332 item types that can be repeated set these backoff variables appropriately. */
3333
3334 firstbyte = reqbyte = zerofirstbyte = zeroreqbyte = REQ_UNSET;
3335
3336 /* The variable req_caseopt contains either the REQ_CASELESS value or zero,
3337 according to the current setting of the caseless flag. REQ_CASELESS is a bit
3338 value > 255. It is added into the firstbyte or reqbyte variables to record the
3339 case status of the value. This is used only for ASCII characters. */
3340
3341 req_caseopt = ((options & PCRE_CASELESS) != 0)? REQ_CASELESS : 0;
3342
3343 /* Switch on next character until the end of the branch */
3344
3345 for (;; ptr++)
3346 {
3347 BOOL negate_class;
3348 BOOL should_flip_negation;
3349 BOOL possessive_quantifier;
3350 BOOL is_quantifier;
3351 BOOL is_recurse;
3352 BOOL reset_bracount;
3353 int class_charcount;
3354 int class_lastchar;
3355 int newoptions;
3356 int recno;
3357 int refsign;
3358 int skipbytes;
3359 int subreqbyte;
3360 int subfirstbyte;
3361 int terminator;
3362 int mclength;
3363 int tempbracount;
3364 pcre_uchar mcbuffer[8];
3365
3366 /* Get next byte in the pattern */
3367
3368 c = *ptr;
3369
3370 /* If we are at the end of a nested substitution, revert to the outer level
3371 string. Nesting only happens one level deep. */
3372
3373 if (c == 0 && nestptr != NULL)
3374 {
3375 ptr = nestptr;
3376 nestptr = NULL;
3377 c = *ptr;
3378 }
3379
3380 /* If we are in the pre-compile phase, accumulate the length used for the
3381 previous cycle of this loop. */
3382
3383 if (lengthptr != NULL)
3384 {
3385 #ifdef PCRE_DEBUG
3386 if (code > cd->hwm) cd->hwm = code; /* High water info */
3387 #endif
3388 if (code > cd->start_workspace + WORK_SIZE_CHECK) /* Check for overrun */
3389 {
3390 *errorcodeptr = ERR52;
3391 goto FAILED;
3392 }
3393
3394 /* There is at least one situation where code goes backwards: this is the
3395 case of a zero quantifier after a class (e.g. [ab]{0}). At compile time,
3396 the class is simply eliminated. However, it is created first, so we have to
3397 allow memory for it. Therefore, don't ever reduce the length at this point.
3398 */
3399
3400 if (code < last_code) code = last_code;
3401
3402 /* Paranoid check for integer overflow */
3403
3404 if (OFLOW_MAX - *lengthptr < code - last_code)
3405 {
3406 *errorcodeptr = ERR20;
3407 goto FAILED;
3408 }
3409
3410 *lengthptr += (int)(code - last_code);
3411 DPRINTF(("length=%d added %d c=%c\n", *lengthptr, (int)(code - last_code),
3412 c));
3413
3414 /* If "previous" is set and it is not at the start of the work space, move
3415 it back to there, in order to avoid filling up the work space. Otherwise,
3416 if "previous" is NULL, reset the current code pointer to the start. */
3417
3418 if (previous != NULL)
3419 {
3420 if (previous > orig_code)
3421 {
3422 memmove(orig_code, previous, code - previous);
3423 code -= previous - orig_code;
3424 previous = orig_code;
3425 }
3426 }
3427 else code = orig_code;
3428
3429 /* Remember where this code item starts so we can pick up the length
3430 next time round. */
3431
3432 last_code = code;
3433 }
3434
3435 /* In the real compile phase, just check the workspace used by the forward
3436 reference list. */
3437
3438 else if (cd->hwm > cd->start_workspace + WORK_SIZE_CHECK)
3439 {
3440 *errorcodeptr = ERR52;
3441 goto FAILED;
3442 }
3443
3444 /* If in \Q...\E, check for the end; if not, we have a literal */
3445
3446 if (inescq && c != 0)
3447 {
3448 if (c == CHAR_BACKSLASH && ptr[1] == CHAR_E)
3449 {
3450 inescq = FALSE;
3451 ptr++;
3452 continue;
3453 }
3454 else
3455 {
3456 if (previous_callout != NULL)
3457 {
3458 if (lengthptr == NULL) /* Don't attempt in pre-compile phase */
3459 complete_callout(previous_callout, ptr, cd);
3460 previous_callout = NULL;
3461 }
3462 if ((options & PCRE_AUTO_CALLOUT) != 0)
3463 {
3464 previous_callout = code;
3465 code = auto_callout(code, ptr, cd);
3466 }
3467 goto NORMAL_CHAR;
3468 }
3469 }
3470
3471 /* Fill in length of a previous callout, except when the next thing is
3472 a quantifier. */
3473
3474 is_quantifier =
3475 c == CHAR_ASTERISK || c == CHAR_PLUS || c == CHAR_QUESTION_MARK ||
3476 (c == CHAR_LEFT_CURLY_BRACKET && is_counted_repeat(ptr+1));
3477
3478 if (!is_quantifier && previous_callout != NULL &&
3479 after_manual_callout-- <= 0)
3480 {
3481 if (lengthptr == NULL) /* Don't attempt in pre-compile phase */
3482 complete_callout(previous_callout, ptr, cd);
3483 previous_callout = NULL;
3484 }
3485
3486 /* In extended mode, skip white space and comments. */
3487
3488 if ((options & PCRE_EXTENDED) != 0)
3489 {
3490 if ((cd->ctypes[c] & ctype_space) != 0) continue;
3491 if (c == CHAR_NUMBER_SIGN)
3492 {
3493 ptr++;
3494 while (*ptr != 0)
3495 {
3496 if (IS_NEWLINE(ptr)) { ptr += cd->nllen - 1; break; }
3497 ptr++;
3498 #ifdef SUPPORT_UTF8
3499 if (utf8) while ((*ptr & 0xc0) == 0x80) ptr++;
3500 #endif
3501 }
3502 if (*ptr != 0) continue;
3503
3504 /* Else fall through to handle end of string */
3505 c = 0;
3506 }
3507 }
3508
3509 /* No auto callout for quantifiers. */
3510
3511 if ((options & PCRE_AUTO_CALLOUT) != 0 && !is_quantifier)
3512 {
3513 previous_callout = code;
3514 code = auto_callout(code, ptr, cd);
3515 }
3516
3517 switch(c)
3518 {
3519 /* ===================================================================*/
3520 case 0: /* The branch terminates at string end */
3521 case CHAR_VERTICAL_LINE: /* or | or ) */
3522 case CHAR_RIGHT_PARENTHESIS:
3523 *firstbyteptr = firstbyte;
3524 *reqbyteptr = reqbyte;
3525 *codeptr = code;
3526 *ptrptr = ptr;
3527 if (lengthptr != NULL)
3528 {
3529 if (OFLOW_MAX - *lengthptr < code - last_code)
3530 {
3531 *errorcodeptr = ERR20;
3532 goto FAILED;
3533 }
3534 *lengthptr += (int)(code - last_code); /* To include callout length */
3535 DPRINTF((">> end branch\n"));
3536 }
3537 return TRUE;
3538
3539
3540 /* ===================================================================*/
3541 /* Handle single-character metacharacters. In multiline mode, ^ disables
3542 the setting of any following char as a first character. */
3543
3544 case CHAR_CIRCUMFLEX_ACCENT:
3545 previous = NULL;
3546 if ((options & PCRE_MULTILINE) != 0)
3547 {
3548 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
3549 *code++ = OP_CIRCM;
3550 }
3551 else *code++ = OP_CIRC;
3552 break;
3553
3554 case CHAR_DOLLAR_SIGN:
3555 previous = NULL;
3556 *code++ = ((options & PCRE_MULTILINE) != 0)? OP_DOLLM : OP_DOLL;
3557 break;
3558
3559 /* There can never be a first char if '.' is first, whatever happens about
3560 repeats. The value of reqbyte doesn't change either. */
3561
3562 case CHAR_DOT:
3563 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
3564 zerofirstbyte = firstbyte;
3565 zeroreqbyte = reqbyte;
3566 previous = code;
3567 *code++ = ((options & PCRE_DOTALL) != 0)? OP_ALLANY: OP_ANY;
3568 break;
3569
3570
3571 /* ===================================================================*/
3572 /* Character classes. If the included characters are all < 256, we build a
3573 32-byte bitmap of the permitted characters, except in the special case
3574 where there is only one such character. For negated classes, we build the
3575 map as usual, then invert it at the end. However, we use a different opcode
3576 so that data characters > 255 can be handled correctly.
3577
3578 If the class contains characters outside the 0-255 range, a different
3579 opcode is compiled. It may optionally have a bit map for characters < 256,
3580 but those above are are explicitly listed afterwards. A flag byte tells
3581 whether the bitmap is present, and whether this is a negated class or not.
3582
3583 In JavaScript compatibility mode, an isolated ']' causes an error. In
3584 default (Perl) mode, it is treated as a data character. */
3585
3586 case CHAR_RIGHT_SQUARE_BRACKET:
3587 if ((cd->external_options & PCRE_JAVASCRIPT_COMPAT) != 0)
3588 {
3589 *errorcodeptr = ERR64;
3590 goto FAILED;
3591 }
3592 goto NORMAL_CHAR;
3593
3594 case CHAR_LEFT_SQUARE_BRACKET:
3595 previous = code;
3596
3597 /* PCRE supports POSIX class stuff inside a class. Perl gives an error if
3598 they are encountered at the top level, so we'll do that too. */
3599
3600 if ((ptr[1] == CHAR_COLON || ptr[1] == CHAR_DOT ||
3601 ptr[1] == CHAR_EQUALS_SIGN) &&
3602 check_posix_syntax(ptr, &tempptr))
3603 {
3604 *errorcodeptr = (ptr[1] == CHAR_COLON)? ERR13 : ERR31;
3605 goto FAILED;
3606 }
3607
3608 /* If the first character is '^', set the negation flag and skip it. Also,
3609 if the first few characters (either before or after ^) are \Q\E or \E we
3610 skip them too. This makes for compatibility with Perl. */
3611
3612 negate_class = FALSE;
3613 for (;;)
3614 {
3615 c = *(++ptr);
3616 if (c == CHAR_BACKSLASH)
3617 {
3618 if (ptr[1] == CHAR_E)
3619 ptr++;
3620 else if (strncmp((const char *)ptr+1,
3621 STR_Q STR_BACKSLASH STR_E, 3) == 0)
3622 ptr += 3;
3623 else
3624 break;
3625 }
3626 else if (!negate_class && c == CHAR_CIRCUMFLEX_ACCENT)
3627 negate_class = TRUE;
3628 else break;
3629 }
3630
3631 /* Empty classes are allowed in JavaScript compatibility mode. Otherwise,
3632 an initial ']' is taken as a data character -- the code below handles
3633 that. In JS mode, [] must always fail, so generate OP_FAIL, whereas
3634 [^] must match any character, so generate OP_ALLANY. */
3635
3636 if (c == CHAR_RIGHT_SQUARE_BRACKET &&
3637 (cd->external_options & PCRE_JAVASCRIPT_COMPAT) != 0)
3638 {
3639 *code++ = negate_class? OP_ALLANY : OP_FAIL;
3640 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
3641 zerofirstbyte = firstbyte;
3642 break;
3643 }
3644
3645 /* If a class contains a negative special such as \S, we need to flip the
3646 negation flag at the end, so that support for characters > 255 works
3647 correctly (they are all included in the class). */
3648
3649 should_flip_negation = FALSE;
3650
3651 /* Keep a count of chars with values < 256 so that we can optimize the case
3652 of just a single character (as long as it's < 256). However, For higher
3653 valued UTF-8 characters, we don't yet do any optimization. */
3654
3655 class_charcount = 0;
3656 class_lastchar = -1;
3657
3658 /* Initialize the 32-char bit map to all zeros. We build the map in a
3659 temporary bit of memory, in case the class contains only 1 character (less
3660 than 256), because in that case the compiled code doesn't use the bit map.
3661 */
3662
3663 memset(classbits, 0, 32 * sizeof(pcre_uint8));
3664
3665 #ifdef SUPPORT_UTF8
3666 class_utf8 = FALSE; /* No chars >= 256 */
3667 class_utf8data = code + LINK_SIZE + 2; /* For UTF-8 items */
3668 class_utf8data_base = class_utf8data; /* For resetting in pass 1 */
3669 #endif
3670
3671 /* Process characters until ] is reached. By writing this as a "do" it
3672 means that an initial ] is taken as a data character. At the start of the
3673 loop, c contains the first byte of the character. */
3674
3675 if (c != 0) do
3676 {
3677 const pcre_uchar *oldptr;
3678
3679 #ifdef SUPPORT_UTF8
3680 if (utf8 && c > 127)
3681 { /* Braces are required because the */
3682 GETCHARLEN(c, ptr, ptr); /* macro generates multiple statements */
3683 }
3684
3685 /* In the pre-compile phase, accumulate the length of any UTF-8 extra
3686 data and reset the pointer. This is so that very large classes that
3687 contain a zillion UTF-8 characters no longer overwrite the work space
3688 (which is on the stack). */
3689
3690 if (lengthptr != NULL)
3691 {
3692 *lengthptr += class_utf8data - class_utf8data_base;
3693 class_utf8data = class_utf8data_base;
3694 }
3695
3696 #endif
3697
3698 /* Inside \Q...\E everything is literal except \E */
3699
3700 if (inescq)
3701 {
3702 if (c == CHAR_BACKSLASH && ptr[1] == CHAR_E) /* If we are at \E */
3703 {
3704 inescq = FALSE; /* Reset literal state */
3705 ptr++; /* Skip the 'E' */
3706 continue; /* Carry on with next */
3707 }
3708 goto CHECK_RANGE; /* Could be range if \E follows */
3709 }
3710
3711 /* Handle POSIX class names. Perl allows a negation extension of the
3712 form [:^name:]. A square bracket that doesn't match the syntax is
3713 treated as a literal. We also recognize the POSIX constructions
3714 [.ch.] and [=ch=] ("collating elements") and fault them, as Perl
3715 5.6 and 5.8 do. */
3716
3717 if (c == CHAR_LEFT_SQUARE_BRACKET &&
3718 (ptr[1] == CHAR_COLON || ptr[1] == CHAR_DOT ||
3719 ptr[1] == CHAR_EQUALS_SIGN) && check_posix_syntax(ptr, &tempptr))
3720 {
3721 BOOL local_negate = FALSE;
3722 int posix_class, taboffset, tabopt;
3723 register const pcre_uint8 *cbits = cd->cbits;
3724 pcre_uint8 pbits[32];
3725
3726 if (ptr[1] != CHAR_COLON)
3727 {
3728 *errorcodeptr = ERR31;
3729 goto FAILED;
3730 }
3731
3732 ptr += 2;
3733 if (*ptr == CHAR_CIRCUMFLEX_ACCENT)
3734 {
3735 local_negate = TRUE;
3736 should_flip_negation = TRUE; /* Note negative special */
3737 ptr++;
3738 }
3739
3740 posix_class = check_posix_name(ptr, (int)(tempptr - ptr));
3741 if (posix_class < 0)
3742 {
3743 *errorcodeptr = ERR30;
3744 goto FAILED;
3745 }
3746
3747 /* If matching is caseless, upper and lower are converted to
3748 alpha. This relies on the fact that the class table starts with
3749 alpha, lower, upper as the first 3 entries. */
3750
3751 if ((options & PCRE_CASELESS) != 0 && posix_class <= 2)
3752 posix_class = 0;
3753
3754 /* When PCRE_UCP is set, some of the POSIX classes are converted to
3755 different escape sequences that use Unicode properties. */
3756
3757 #ifdef SUPPORT_UCP
3758 if ((options & PCRE_UCP) != 0)
3759 {
3760 int pc = posix_class + ((local_negate)? POSIX_SUBSIZE/2 : 0);
3761 if (posix_substitutes[pc] != NULL)
3762 {
3763 nestptr = tempptr + 1;
3764 ptr = posix_substitutes[pc] - 1;
3765 continue;
3766 }
3767 }
3768 #endif
3769 /* In the non-UCP case, we build the bit map for the POSIX class in a
3770 chunk of local store because we may be adding and subtracting from it,
3771 and we don't want to subtract bits that may be in the main map already.
3772 At the end we or the result into the bit map that is being built. */
3773
3774 posix_class *= 3;
3775
3776 /* Copy in the first table (always present) */
3777
3778 memcpy(pbits, cbits + posix_class_maps[posix_class],
3779 32 * sizeof(pcre_uint8));
3780
3781 /* If there is a second table, add or remove it as required. */
3782
3783 taboffset = posix_class_maps[posix_class + 1];
3784 tabopt = posix_class_maps[posix_class + 2];
3785
3786 if (taboffset >= 0)
3787 {
3788 if (tabopt >= 0)
3789 for (c = 0; c < 32; c++) pbits[c] |= cbits[c + taboffset];
3790 else
3791 for (c = 0; c < 32; c++) pbits[c] &= ~cbits[c + taboffset];
3792 }
3793
3794 /* Not see if we need to remove any special characters. An option
3795 value of 1 removes vertical space and 2 removes underscore. */
3796
3797 if (tabopt < 0) tabopt = -tabopt;
3798 if (tabopt == 1) pbits[1] &= ~0x3c;
3799 else if (tabopt == 2) pbits[11] &= 0x7f;
3800
3801 /* Add the POSIX table or its complement into the main table that is
3802 being built and we are done. */
3803
3804 if (local_negate)
3805 for (c = 0; c < 32; c++) classbits[c] |= ~pbits[c];
3806 else
3807 for (c = 0; c < 32; c++) classbits[c] |= pbits[c];
3808
3809 ptr = tempptr + 1;
3810 class_charcount = 10; /* Set > 1; assumes more than 1 per class */
3811 continue; /* End of POSIX syntax handling */
3812 }
3813
3814 /* Backslash may introduce a single character, or it may introduce one
3815 of the specials, which just set a flag. The sequence \b is a special
3816 case. Inside a class (and only there) it is treated as backspace. We
3817 assume that other escapes have more than one character in them, so set
3818 class_charcount bigger than one. Unrecognized escapes fall through and
3819 are either treated as literal characters (by default), or are faulted if
3820 PCRE_EXTRA is set. */
3821
3822 if (c == CHAR_BACKSLASH)
3823 {
3824 c = check_escape(&ptr, errorcodeptr, cd->bracount, options, TRUE);
3825 if (*errorcodeptr != 0) goto FAILED;
3826
3827 if (-c == ESC_b) c = CHAR_BS; /* \b is backspace in a class */
3828 else if (-c == ESC_Q) /* Handle start of quoted string */
3829 {
3830 if (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
3831 {
3832 ptr += 2; /* avoid empty string */
3833 }
3834 else inescq = TRUE;
3835 continue;
3836 }
3837 else if (-c == ESC_E) continue; /* Ignore orphan \E */
3838
3839 if (c < 0)
3840 {
3841 register const pcre_uint8 *cbits = cd->cbits;
3842 class_charcount += 2; /* Greater than 1 is what matters */
3843
3844 switch (-c)
3845 {
3846 #ifdef SUPPORT_UCP
3847 case ESC_du: /* These are the values given for \d etc */
3848 case ESC_DU: /* when PCRE_UCP is set. We replace the */
3849 case ESC_wu: /* escape sequence with an appropriate \p */
3850 case ESC_WU: /* or \P to test Unicode properties instead */
3851 case ESC_su: /* of the default ASCII testing. */
3852 case ESC_SU:
3853 nestptr = ptr;
3854 ptr = substitutes[-c - ESC_DU] - 1; /* Just before substitute */
3855 class_charcount -= 2; /* Undo! */
3856 continue;
3857 #endif
3858 case ESC_d:
3859 for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_digit];
3860 continue;
3861
3862 case ESC_D:
3863 should_flip_negation = TRUE;
3864 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_digit];
3865 continue;
3866
3867 case ESC_w:
3868 for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_word];
3869 continue;
3870
3871 case ESC_W:
3872 should_flip_negation = TRUE;
3873 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_word];
3874 continue;
3875
3876 /* Perl 5.004 onwards omits VT from \s, but we must preserve it
3877 if it was previously set by something earlier in the character
3878 class. */
3879
3880 case ESC_s:
3881 classbits[0] |= cbits[cbit_space];
3882 classbits[1] |= cbits[cbit_space+1] & ~0x08;
3883 for (c = 2; c < 32; c++) classbits[c] |= cbits[c+cbit_space];
3884 continue;
3885
3886 case ESC_S:
3887 should_flip_negation = TRUE;
3888 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_space];
3889 classbits[1] |= 0x08; /* Perl 5.004 onwards omits VT from \s */
3890 continue;
3891
3892 case ESC_h:
3893 SETBIT(classbits, 0x09); /* VT */
3894 SETBIT(classbits, 0x20); /* SPACE */
3895 SETBIT(classbits, 0xa0); /* NSBP */
3896 #ifdef SUPPORT_UTF8
3897 if (utf8)
3898 {
3899 class_utf8 = TRUE;
3900 *class_utf8data++ = XCL_SINGLE;
3901 class_utf8data += _pcre_ord2utf8(0x1680, class_utf8data);
3902 *class_utf8data++ = XCL_SINGLE;
3903 class_utf8data += _pcre_ord2utf8(0x180e, class_utf8data);
3904 *class_utf8data++ = XCL_RANGE;
3905 class_utf8data += _pcre_ord2utf8(0x2000, class_utf8data);
3906 class_utf8data += _pcre_ord2utf8(0x200A, class_utf8data);
3907 *class_utf8data++ = XCL_SINGLE;
3908 class_utf8data += _pcre_ord2utf8(0x202f, class_utf8data);
3909 *class_utf8data++ = XCL_SINGLE;
3910 class_utf8data += _pcre_ord2utf8(0x205f, class_utf8data);
3911 *class_utf8data++ = XCL_SINGLE;
3912 class_utf8data += _pcre_ord2utf8(0x3000, class_utf8data);
3913 }
3914 #endif
3915 continue;
3916
3917 case ESC_H:
3918 for (c = 0; c < 32; c++)
3919 {
3920 int x = 0xff;
3921 switch (c)
3922 {
3923 case 0x09/8: x ^= 1 << (0x09%8); break;
3924 case 0x20/8: x ^= 1 << (0x20%8); break;
3925 case 0xa0/8: x ^= 1 << (0xa0%8); break;
3926 default: break;
3927 }
3928 classbits[c] |= x;
3929 }
3930
3931 #ifdef SUPPORT_UTF8
3932 if (utf8)
3933 {
3934 class_utf8 = TRUE;
3935 *class_utf8data++ = XCL_RANGE;
3936 class_utf8data += _pcre_ord2utf8(0x0100, class_utf8data);
3937 class_utf8data += _pcre_ord2utf8(0x167f, class_utf8data);
3938 *class_utf8data++ = XCL_RANGE;
3939 class_utf8data += _pcre_ord2utf8(0x1681, class_utf8data);
3940 class_utf8data += _pcre_ord2utf8(0x180d, class_utf8data);
3941 *class_utf8data++ = XCL_RANGE;
3942 class_utf8data += _pcre_ord2utf8(0x180f, class_utf8data);
3943 class_utf8data += _pcre_ord2utf8(0x1fff, class_utf8data);
3944 *class_utf8data++ = XCL_RANGE;
3945 class_utf8data += _pcre_ord2utf8(0x200B, class_utf8data);
3946 class_utf8data += _pcre_ord2utf8(0x202e, class_utf8data);
3947 *class_utf8data++ = XCL_RANGE;
3948 class_utf8data += _pcre_ord2utf8(0x2030, class_utf8data);
3949 class_utf8data += _pcre_ord2utf8(0x205e, class_utf8data);
3950 *class_utf8data++ = XCL_RANGE;
3951 class_utf8data += _pcre_ord2utf8(0x2060, class_utf8data);
3952 class_utf8data += _pcre_ord2utf8(0x2fff, class_utf8data);
3953 *class_utf8data++ = XCL_RANGE;
3954 class_utf8data += _pcre_ord2utf8(0x3001, class_utf8data);
3955 class_utf8data += _pcre_ord2utf8(0x7fffffff, class_utf8data);
3956 }
3957 #endif
3958 continue;
3959
3960 case ESC_v:
3961 SETBIT(classbits, 0x0a); /* LF */
3962 SETBIT(classbits, 0x0b); /* VT */
3963 SETBIT(classbits, 0x0c); /* FF */
3964 SETBIT(classbits, 0x0d); /* CR */
3965 SETBIT(classbits, 0x85); /* NEL */
3966 #ifdef SUPPORT_UTF8
3967 if (utf8)
3968 {
3969 class_utf8 = TRUE;
3970 *class_utf8data++ = XCL_RANGE;
3971 class_utf8data += _pcre_ord2utf8(0x2028, class_utf8data);
3972 class_utf8data += _pcre_ord2utf8(0x2029, class_utf8data);
3973 }
3974 #endif
3975 continue;
3976
3977 case ESC_V:
3978 for (c = 0; c < 32; c++)
3979 {
3980 int x = 0xff;
3981 switch (c)
3982 {
3983 case 0x0a/8: x ^= 1 << (0x0a%8);
3984 x ^= 1 << (0x0b%8);
3985 x ^= 1 << (0x0c%8);
3986 x ^= 1 << (0x0d%8);
3987 break;
3988 case 0x85/8: x ^= 1 << (0x85%8); break;
3989 default: break;
3990 }
3991 classbits[c] |= x;
3992 }
3993
3994 #ifdef SUPPORT_UTF8
3995 if (utf8)
3996 {
3997 class_utf8 = TRUE;
3998 *class_utf8data++ = XCL_RANGE;
3999 class_utf8data += _pcre_ord2utf8(0x0100, class_utf8data);
4000 class_utf8data += _pcre_ord2utf8(0x2027, class_utf8data);
4001 *class_utf8data++ = XCL_RANGE;
4002 class_utf8data += _pcre_ord2utf8(0x2029, class_utf8data);
4003 class_utf8data += _pcre_ord2utf8(0x7fffffff, class_utf8data);
4004 }
4005 #endif
4006 continue;
4007
4008 #ifdef SUPPORT_UCP
4009 case ESC_p:
4010 case ESC_P:
4011 {
4012 BOOL negated;
4013 int pdata;
4014 int ptype = get_ucp(&ptr, &negated, &pdata, errorcodeptr);
4015 if (ptype < 0) goto FAILED;
4016 class_utf8 = TRUE;
4017 *class_utf8data++ = ((-c == ESC_p) != negated)?
4018 XCL_PROP : XCL_NOTPROP;
4019 *class_utf8data++ = ptype;
4020 *class_utf8data++ = pdata;
4021 class_charcount -= 2; /* Not a < 256 character */
4022 continue;
4023 }
4024 #endif
4025 /* Unrecognized escapes are faulted if PCRE is running in its
4026 strict mode. By default, for compatibility with Perl, they are
4027 treated as literals. */
4028
4029 default:
4030 if ((options & PCRE_EXTRA) != 0)
4031 {
4032 *errorcodeptr = ERR7;
4033 goto FAILED;
4034 }
4035 class_charcount -= 2; /* Undo the default count from above */
4036 c = *ptr; /* Get the final character and fall through */
4037 break;
4038 }
4039 }
4040
4041 /* Fall through if we have a single character (c >= 0). This may be
4042 greater than 256 in UTF-8 mode. */
4043
4044 } /* End of backslash handling */
4045
4046 /* A single character may be followed by '-' to form a range. However,
4047 Perl does not permit ']' to be the end of the range. A '-' character
4048 at the end is treated as a literal. Perl ignores orphaned \E sequences
4049 entirely. The code for handling \Q and \E is messy. */
4050
4051 CHECK_RANGE:
4052 while (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
4053 {
4054 inescq = FALSE;
4055 ptr += 2;
4056 }
4057
4058 oldptr = ptr;
4059
4060 /* Remember \r or \n */
4061
4062 if (c == CHAR_CR || c == CHAR_NL) cd->external_flags |= PCRE_HASCRORLF;
4063
4064 /* Check for range */
4065
4066 if (!inescq && ptr[1] == CHAR_MINUS)
4067 {
4068 int d;
4069 ptr += 2;
4070 while (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_E) ptr += 2;
4071
4072 /* If we hit \Q (not followed by \E) at this point, go into escaped
4073 mode. */
4074
4075 while (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_Q)
4076 {
4077 ptr += 2;
4078 if (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_E)
4079 { ptr += 2; continue; }
4080 inescq = TRUE;
4081 break;
4082 }
4083
4084 if (*ptr == 0 || (!inescq && *ptr == CHAR_RIGHT_SQUARE_BRACKET))
4085 {
4086 ptr = oldptr;
4087 goto LONE_SINGLE_CHARACTER;
4088 }
4089
4090 #ifdef SUPPORT_UTF8
4091 if (utf8)
4092 { /* Braces are required because the */
4093 GETCHARLEN(d, ptr, ptr); /* macro generates multiple statements */
4094 }
4095 else
4096 #endif
4097 d = *ptr; /* Not UTF-8 mode */
4098
4099 /* The second part of a range can be a single-character escape, but
4100 not any of the other escapes. Perl 5.6 treats a hyphen as a literal
4101 in such circumstances. */
4102
4103 if (!inescq && d == CHAR_BACKSLASH)
4104 {
4105 d = check_escape(&ptr, errorcodeptr, cd->bracount, options, TRUE);
4106 if (*errorcodeptr != 0) goto FAILED;
4107
4108 /* \b is backspace; any other special means the '-' was literal */
4109
4110 if (d < 0)
4111 {
4112 if (d == -ESC_b) d = CHAR_BS; else
4113 {
4114 ptr = oldptr;
4115 goto LONE_SINGLE_CHARACTER; /* A few lines below */
4116 }
4117 }
4118 }
4119
4120 /* Check that the two values are in the correct order. Optimize
4121 one-character ranges */
4122
4123 if (d < c)
4124 {
4125 *errorcodeptr = ERR8;
4126 goto FAILED;
4127 }
4128
4129 if (d == c) goto LONE_SINGLE_CHARACTER; /* A few lines below */
4130
4131 /* Remember \r or \n */
4132
4133 if (d == CHAR_CR || d == CHAR_NL) cd->external_flags |= PCRE_HASCRORLF;
4134
4135 /* In UTF-8 mode, if the upper limit is > 255, or > 127 for caseless
4136 matching, we have to use an XCLASS with extra data items. Caseless
4137 matching for characters > 127 is available only if UCP support is
4138 available. */
4139
4140 #ifdef SUPPORT_UTF8
4141 if (utf8 && (d > 255 || ((options & PCRE_CASELESS) != 0 && d > 127)))
4142 {
4143 class_utf8 = TRUE;
4144
4145 /* With UCP support, we can find the other case equivalents of
4146 the relevant characters. There may be several ranges. Optimize how
4147 they fit with the basic range. */
4148
4149 #ifdef SUPPORT_UCP
4150 if ((options & PCRE_CASELESS) != 0)
4151 {
4152 unsigned int occ, ocd;
4153 unsigned int cc = c;
4154 unsigned int origd = d;
4155 while (get_othercase_range(&cc, origd, &occ, &ocd))
4156 {
4157 if (occ >= (unsigned int)c &&
4158 ocd <= (unsigned int)d)
4159 continue; /* Skip embedded ranges */
4160
4161 if (occ < (unsigned int)c &&
4162 ocd >= (unsigned int)c - 1) /* Extend the basic range */
4163 { /* if there is overlap, */
4164 c = occ; /* noting that if occ < c */
4165 continue; /* we can't have ocd > d */
4166 } /* because a subrange is */
4167 if (ocd > (unsigned int)d &&
4168 occ <= (unsigned int)d + 1) /* always shorter than */
4169 { /* the basic range. */
4170 d = ocd;
4171 continue;
4172 }
4173
4174 if (occ == ocd)
4175 {
4176 *class_utf8data++ = XCL_SINGLE;
4177 }
4178 else
4179 {
4180 *class_utf8data++ = XCL_RANGE;
4181 class_utf8data += _pcre_ord2utf8(occ, class_utf8data);
4182 }
4183 class_utf8data += _pcre_ord2utf8(ocd, class_utf8data);
4184 }
4185 }
4186 #endif /* SUPPORT_UCP */
4187
4188 /* Now record the original range, possibly modified for UCP caseless
4189 overlapping ranges. */
4190
4191 *class_utf8data++ = XCL_RANGE;
4192 class_utf8data += _pcre_ord2utf8(c, class_utf8data);
4193 class_utf8data += _pcre_ord2utf8(d, class_utf8data);
4194
4195 /* With UCP support, we are done. Without UCP support, there is no
4196 caseless matching for UTF-8 characters > 127; we can use the bit map
4197 for the smaller ones. */
4198
4199 #ifdef SUPPORT_UCP
4200 continue; /* With next character in the class */
4201 #else
4202 if ((options & PCRE_CASELESS) == 0 || c > 127) continue;
4203
4204 /* Adjust upper limit and fall through to set up the map */
4205
4206 d = 127;
4207
4208 #endif /* SUPPORT_UCP */
4209 }
4210 #endif /* SUPPORT_UTF8 */
4211
4212 /* We use the bit map for all cases when not in UTF-8 mode; else
4213 ranges that lie entirely within 0-127 when there is UCP support; else
4214 for partial ranges without UCP support. */
4215
4216 class_charcount += d - c + 1;
4217 class_lastchar = d;
4218
4219 /* We can save a bit of time by skipping this in the pre-compile. */
4220
4221 if (lengthptr == NULL) for (; c <= d; c++)
4222 {
4223 classbits[c/8] |= (1 << (c&7));
4224 if ((options & PCRE_CASELESS) != 0)
4225 {
4226 int uc = cd->fcc[c]; /* flip case */
4227 classbits[uc/8] |= (1 << (uc&7));
4228 }
4229 }
4230
4231 continue; /* Go get the next char in the class */
4232 }
4233
4234 /* Handle a lone single character - we can get here for a normal
4235 non-escape char, or after \ that introduces a single character or for an
4236 apparent range that isn't. */
4237
4238 LONE_SINGLE_CHARACTER:
4239
4240 /* Handle a character that cannot go in the bit map */
4241
4242 #ifdef SUPPORT_UTF8
4243 if (utf8 && (c > 255 || ((options & PCRE_CASELESS) != 0 && c > 127)))
4244 {
4245 class_utf8 = TRUE;
4246 *class_utf8data++ = XCL_SINGLE;
4247 class_utf8data += _pcre_ord2utf8(c, class_utf8data);
4248
4249 #ifdef SUPPORT_UCP
4250 if ((options & PCRE_CASELESS) != 0)
4251 {
4252 unsigned int othercase;
4253 if ((othercase = UCD_OTHERCASE(c)) != c)
4254 {
4255 *class_utf8data++ = XCL_SINGLE;
4256 class_utf8data += _pcre_ord2utf8(othercase, class_utf8data);
4257 }
4258 }
4259 #endif /* SUPPORT_UCP */
4260
4261 }
4262 else
4263 #endif /* SUPPORT_UTF8 */
4264
4265 /* Handle a single-byte character */
4266 {
4267 classbits[c/8] |= (1 << (c&7));
4268 if ((options & PCRE_CASELESS) != 0)
4269 {
4270 c = cd->fcc[c]; /* flip case */
4271 classbits[c/8] |= (1 << (c&7));
4272 }
4273 class_charcount++;
4274 class_lastchar = c;
4275 }
4276 }
4277
4278 /* Loop until ']' reached. This "while" is the end of the "do" far above.
4279 If we are at the end of an internal nested string, revert to the outer
4280 string. */
4281
4282 while (((c = *(++ptr)) != 0 ||
4283 (nestptr != NULL &&
4284 (ptr = nestptr, nestptr = NULL, c = *(++ptr)) != 0)) &&
4285 (c != CHAR_RIGHT_SQUARE_BRACKET || inescq));
4286
4287 /* Check for missing terminating ']' */
4288
4289 if (c == 0)
4290 {
4291 *errorcodeptr = ERR6;
4292 goto FAILED;
4293 }
4294
4295 /* If class_charcount is 1, we saw precisely one character whose value is
4296 less than 256. As long as there were no characters >= 128 and there was no
4297 use of \p or \P, in other words, no use of any XCLASS features, we can
4298 optimize.
4299
4300 In UTF-8 mode, we can optimize the negative case only if there were no
4301 characters >= 128 because OP_NOT and the related opcodes like OP_NOTSTAR
4302 operate on single-bytes characters only. This is an historical hangover.
4303 Maybe one day we can tidy these opcodes to handle multi-byte characters.
4304
4305 The optimization throws away the bit map. We turn the item into a
4306 1-character OP_CHAR[I] if it's positive, or OP_NOT[I] if it's negative.
4307 Note that OP_NOT[I] does not support multibyte characters. In the positive
4308 case, it can cause firstbyte to be set. Otherwise, there can be no first
4309 char if this item is first, whatever repeat count may follow. In the case
4310 of reqbyte, save the previous value for reinstating. */
4311
4312 #ifdef SUPPORT_UTF8
4313 if (class_charcount == 1 && !class_utf8 &&
4314 (!utf8 || !negate_class || class_lastchar < 128))
4315 #else
4316 if (class_charcount == 1)
4317 #endif
4318 {
4319 zeroreqbyte = reqbyte;
4320
4321 /* The OP_NOT[I] opcodes work on one-byte characters only. */
4322
4323 if (negate_class)
4324 {
4325 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
4326 zerofirstbyte = firstbyte;
4327 *code++ = ((options & PCRE_CASELESS) != 0)? OP_NOTI: OP_NOT;
4328 *code++ = class_lastchar;
4329 break;
4330 }
4331
4332 /* For a single, positive character, get the value into mcbuffer, and
4333 then we can handle this with the normal one-character code. */
4334
4335 #ifdef SUPPORT_UTF8
4336 if (utf8 && class_lastchar > 127)
4337 mclength = _pcre_ord2utf8(class_lastchar, mcbuffer);
4338 else
4339 #endif
4340 {
4341 mcbuffer[0] = class_lastchar;
4342 mclength = 1;
4343 }
4344 goto ONE_CHAR;
4345 } /* End of 1-char optimization */
4346
4347 /* The general case - not the one-char optimization. If this is the first
4348 thing in the branch, there can be no first char setting, whatever the
4349 repeat count. Any reqbyte setting must remain unchanged after any kind of
4350 repeat. */
4351
4352 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
4353 zerofirstbyte = firstbyte;
4354 zeroreqbyte = reqbyte;
4355
4356 /* If there are characters with values > 255, we have to compile an
4357 extended class, with its own opcode, unless there was a negated special
4358 such as \S in the class, and PCRE_UCP is not set, because in that case all
4359 characters > 255 are in the class, so any that were explicitly given as
4360 well can be ignored. If (when there are explicit characters > 255 that must
4361 be listed) there are no characters < 256, we can omit the bitmap in the
4362 actual compiled code. */
4363
4364 #ifdef SUPPORT_UTF8
4365 if (class_utf8 && (!should_flip_negation || (options & PCRE_UCP) != 0))
4366 {
4367 *class_utf8data++ = XCL_END; /* Marks the end of extra data */
4368 *code++ = OP_XCLASS;
4369 code += LINK_SIZE;
4370 *code = negate_class? XCL_NOT : 0;
4371
4372 /* If the map is required, move up the extra data to make room for it;
4373 otherwise just move the code pointer to the end of the extra data. */
4374
4375 if (class_charcount > 0)
4376 {
4377 *code++ |= XCL_MAP;
4378 memmove(code + 32, code, class_utf8data - code);
4379 memcpy(code, classbits, 32);
4380 code = class_utf8data + 32;
4381 }
4382 else code = class_utf8data;
4383
4384 /* Now fill in the complete length of the item */
4385
4386 PUT(previous, 1, code - previous);
4387 break; /* End of class handling */
4388 }
4389 #endif
4390
4391 /* If there are no characters > 255, or they are all to be included or
4392 excluded, set the opcode to OP_CLASS or OP_NCLASS, depending on whether the
4393 whole class was negated and whether there were negative specials such as \S
4394 (non-UCP) in the class. Then copy the 32-byte map into the code vector,
4395 negating it if necessary. */
4396
4397 *code++ = (negate_class == should_flip_negation) ? OP_CLASS : OP_NCLASS;
4398 if (negate_class)
4399 {
4400 if (lengthptr == NULL) /* Save time in the pre-compile phase */
4401 for (c = 0; c < 32; c++) code[c] = ~classbits[c];
4402 }
4403 else
4404 {
4405 memcpy(code, classbits, 32);
4406 }
4407 code += 32;
4408 break;
4409
4410
4411 /* ===================================================================*/
4412 /* Various kinds of repeat; '{' is not necessarily a quantifier, but this
4413 has been tested above. */
4414
4415 case CHAR_LEFT_CURLY_BRACKET:
4416 if (!is_quantifier) goto NORMAL_CHAR;
4417 ptr = read_repeat_counts(ptr+1, &repeat_min, &repeat_max, errorcodeptr);
4418 if (*errorcodeptr != 0) goto FAILED;
4419 goto REPEAT;
4420
4421 case CHAR_ASTERISK:
4422 repeat_min = 0;
4423 repeat_max = -1;
4424 goto REPEAT;
4425
4426 case CHAR_PLUS:
4427 repeat_min = 1;
4428 repeat_max = -1;
4429 goto REPEAT;
4430
4431 case CHAR_QUESTION_MARK:
4432 repeat_min = 0;
4433 repeat_max = 1;
4434
4435 REPEAT:
4436 if (previous == NULL)
4437 {
4438 *errorcodeptr = ERR9;
4439 goto FAILED;
4440 }
4441
4442 if (repeat_min == 0)
4443 {
4444 firstbyte = zerofirstbyte; /* Adjust for zero repeat */
4445 reqbyte = zeroreqbyte; /* Ditto */
4446 }
4447
4448 /* Remember whether this is a variable length repeat */
4449
4450 reqvary = (repeat_min == repeat_max)? 0 : REQ_VARY;
4451
4452 op_type = 0; /* Default single-char op codes */
4453 possessive_quantifier = FALSE; /* Default not possessive quantifier */
4454
4455 /* Save start of previous item, in case we have to move it up in order to
4456 insert something before it. */
4457
4458 tempcode = previous;
4459
4460 /* If the next character is '+', we have a possessive quantifier. This
4461 implies greediness, whatever the setting of the PCRE_UNGREEDY option.
4462 If the next character is '?' this is a minimizing repeat, by default,
4463 but if PCRE_UNGREEDY is set, it works the other way round. We change the
4464 repeat type to the non-default. */
4465
4466 if (ptr[1] == CHAR_PLUS)
4467 {
4468 repeat_type = 0; /* Force greedy */
4469 possessive_quantifier = TRUE;
4470 ptr++;
4471 }
4472 else if (ptr[1] == CHAR_QUESTION_MARK)
4473 {
4474 repeat_type = greedy_non_default;
4475 ptr++;
4476 }
4477 else repeat_type = greedy_default;
4478
4479 /* If previous was a recursion call, wrap it in atomic brackets so that
4480 previous becomes the atomic group. All recursions were so wrapped in the
4481 past, but it no longer happens for non-repeated recursions. In fact, the
4482 repeated ones could be re-implemented independently so as not to need this,
4483 but for the moment we rely on the code for repeating groups. */
4484
4485 if (*previous == OP_RECURSE)
4486 {
4487 memmove(previous + 1 + LINK_SIZE, previous, 1 + LINK_SIZE);
4488 *previous = OP_ONCE;
4489 PUT(previous, 1, 2 + 2*LINK_SIZE);
4490 previous[2 + 2*LINK_SIZE] = OP_KET;
4491 PUT(previous, 3 + 2*LINK_SIZE, 2 + 2*LINK_SIZE);
4492 code += 2 + 2 * LINK_SIZE;
4493 length_prevgroup = 3 + 3*LINK_SIZE;
4494
4495 /* When actually compiling, we need to check whether this was a forward
4496 reference, and if so, adjust the offset. */
4497
4498 if (lengthptr == NULL && cd->hwm >= cd->start_workspace + LINK_SIZE)
4499 {
4500 int offset = GET(cd->hwm, -LINK_SIZE);
4501 if (offset == previous + 1 - cd->start_code)
4502 PUT(cd->hwm, -LINK_SIZE, offset + 1 + LINK_SIZE);
4503 }
4504 }
4505
4506 /* Now handle repetition for the different types of item. */
4507
4508 /* If previous was a character match, abolish the item and generate a
4509 repeat item instead. If a char item has a minumum of more than one, ensure
4510 that it is set in reqbyte - it might not be if a sequence such as x{3} is
4511 the first thing in a branch because the x will have gone into firstbyte
4512 instead. */
4513
4514 if (*previous == OP_CHAR || *previous == OP_CHARI)
4515 {
4516 op_type = (*previous == OP_CHAR)? 0 : OP_STARI - OP_STAR;
4517
4518 /* Deal with UTF-8 characters that take up more than one byte. It's
4519 easier to write this out separately than try to macrify it. Use c to
4520 hold the length of the character in bytes, plus 0x80 to flag that it's a
4521 length rather than a small character. */
4522
4523 #ifdef SUPPORT_UTF8
4524 if (utf8 && (code[-1] & 0x80) != 0)
4525 {
4526 pcre_uchar *lastchar = code - 1;
4527 while((*lastchar & 0xc0) == 0x80) lastchar--;
4528 c = code - lastchar; /* Length of UTF-8 character */
4529 memcpy(utf8_char, lastchar, c); /* Save the char */
4530 c |= 0x80; /* Flag c as a length */
4531 }
4532 else
4533 #endif
4534
4535 /* Handle the case of a single byte - either with no UTF8 support, or
4536 with UTF-8 disabled, or for a UTF-8 character < 128. */
4537
4538 {
4539 c = code[-1];
4540 if (repeat_min > 1) reqbyte = c | req_caseopt | cd->req_varyopt;
4541 }
4542
4543 /* If the repetition is unlimited, it pays to see if the next thing on
4544 the line is something that cannot possibly match this character. If so,
4545 automatically possessifying this item gains some performance in the case
4546 where the match fails. */
4547
4548 if (!possessive_quantifier &&
4549 repeat_max < 0 &&
4550 check_auto_possessive(previous, utf8, ptr + 1, options, cd))
4551 {
4552 repeat_type = 0; /* Force greedy */
4553 possessive_quantifier = TRUE;
4554 }
4555
4556 goto OUTPUT_SINGLE_REPEAT; /* Code shared with single character types */
4557 }
4558
4559 /* If previous was a single negated character ([^a] or similar), we use
4560 one of the special opcodes, replacing it. The code is shared with single-
4561 character repeats by setting opt_type to add a suitable offset into
4562 repeat_type. We can also test for auto-possessification. OP_NOT and OP_NOTI
4563 are currently used only for single-byte chars. */
4564
4565 else if (*previous == OP_NOT || *previous == OP_NOTI)
4566 {
4567 op_type = ((*previous == OP_NOT)? OP_NOTSTAR : OP_NOTSTARI) - OP_STAR;
4568 c = previous[1];
4569 if (!possessive_quantifier &&
4570 repeat_max < 0 &&
4571 check_auto_possessive(previous, utf8, ptr + 1, options, cd))
4572 {
4573 repeat_type = 0; /* Force greedy */
4574 possessive_quantifier = TRUE;
4575 }
4576 goto OUTPUT_SINGLE_REPEAT;
4577 }
4578
4579 /* If previous was a character type match (\d or similar), abolish it and
4580 create a suitable repeat item. The code is shared with single-character
4581 repeats by setting op_type to add a suitable offset into repeat_type. Note
4582 the the Unicode property types will be present only when SUPPORT_UCP is
4583 defined, but we don't wrap the little bits of code here because it just
4584 makes it horribly messy. */
4585
4586 else if (*previous < OP_EODN)
4587 {
4588 pcre_uchar *oldcode;
4589 int prop_type, prop_value;
4590 op_type = OP_TYPESTAR - OP_STAR; /* Use type opcodes */
4591 c = *previous;
4592
4593 if (!possessive_quantifier &&
4594 repeat_max < 0 &&
4595 check_auto_possessive(previous, utf8, ptr + 1, options, cd))
4596 {
4597 repeat_type = 0; /* Force greedy */
4598 possessive_quantifier = TRUE;
4599 }
4600
4601 OUTPUT_SINGLE_REPEAT:
4602 if (*previous == OP_PROP || *previous == OP_NOTPROP)
4603 {
4604 prop_type = previous[1];
4605 prop_value = previous[2];
4606 }
4607 else prop_type = prop_value = -1;
4608
4609 oldcode = code;
4610 code = previous; /* Usually overwrite previous item */
4611
4612 /* If the maximum is zero then the minimum must also be zero; Perl allows
4613 this case, so we do too - by simply omitting the item altogether. */
4614
4615 if (repeat_max == 0) goto END_REPEAT;
4616
4617 /*--------------------------------------------------------------------*/
4618 /* This code is obsolete from release 8.00; the restriction was finally
4619 removed: */
4620
4621 /* All real repeats make it impossible to handle partial matching (maybe
4622 one day we will be able to remove this restriction). */
4623
4624 /* if (repeat_max != 1) cd->external_flags |= PCRE_NOPARTIAL; */
4625 /*--------------------------------------------------------------------*/
4626
4627 /* Combine the op_type with the repeat_type */
4628
4629 repeat_type += op_type;
4630
4631 /* A minimum of zero is handled either as the special case * or ?, or as
4632 an UPTO, with the maximum given. */
4633
4634 if (repeat_min == 0)
4635 {
4636 if (repeat_max == -1) *code++ = OP_STAR + repeat_type;
4637 else if (repeat_max == 1) *code++ = OP_QUERY + repeat_type;
4638 else
4639 {
4640 *code++ = OP_UPTO + repeat_type;
4641 PUT2INC(code, 0, repeat_max);
4642 }
4643 }
4644
4645 /* A repeat minimum of 1 is optimized into some special cases. If the
4646 maximum is unlimited, we use OP_PLUS. Otherwise, the original item is
4647 left in place and, if the maximum is greater than 1, we use OP_UPTO with
4648 one less than the maximum. */
4649
4650 else if (repeat_min == 1)
4651 {
4652 if (repeat_max == -1)
4653 *code++ = OP_PLUS + repeat_type;
4654 else
4655 {
4656 code = oldcode; /* leave previous item in place */
4657 if (repeat_max == 1) goto END_REPEAT;
4658 *code++ = OP_UPTO + repeat_type;
4659 PUT2INC(code, 0, repeat_max - 1);
4660 }
4661 }
4662
4663 /* The case {n,n} is just an EXACT, while the general case {n,m} is
4664 handled as an EXACT followed by an UPTO. */
4665
4666 else
4667 {
4668 *code++ = OP_EXACT + op_type; /* NB EXACT doesn't have repeat_type */
4669 PUT2INC(code, 0, repeat_min);
4670
4671 /* If the maximum is unlimited, insert an OP_STAR. Before doing so,
4672 we have to insert the character for the previous code. For a repeated
4673 Unicode property match, there are two extra bytes that define the
4674 required property. In UTF-8 mode, long characters have their length in
4675 c, with the 0x80 bit as a flag. */
4676
4677 if (repeat_max < 0)
4678 {
4679 #ifdef SUPPORT_UTF8
4680 if (utf8 && c >= 128)
4681 {
4682 memcpy(code, utf8_char, c & 7);
4683 code += c & 7;
4684 }
4685 else
4686 #endif
4687 {
4688 *code++ = c;
4689 if (prop_type >= 0)
4690 {
4691 *code++ = prop_type;
4692 *code++ = prop_value;
4693 }
4694 }
4695 *code++ = OP_STAR + repeat_type;
4696 }
4697
4698 /* Else insert an UPTO if the max is greater than the min, again
4699 preceded by the character, for the previously inserted code. If the
4700 UPTO is just for 1 instance, we can use QUERY instead. */
4701
4702 else if (repeat_max != repeat_min)
4703 {
4704 #ifdef SUPPORT_UTF8
4705 if (utf8 && c >= 128)
4706 {
4707 memcpy(code, utf8_char, c & 7);
4708 code += c & 7;
4709 }
4710 else
4711 #endif
4712 *code++ = c;
4713 if (prop_type >= 0)
4714 {
4715 *code++ = prop_type;
4716 *code++ = prop_value;
4717 }
4718 repeat_max -= repeat_min;
4719
4720 if (repeat_max == 1)
4721 {
4722 *code++ = OP_QUERY + repeat_type;
4723 }
4724 else
4725 {
4726 *code++ = OP_UPTO + repeat_type;
4727 PUT2INC(code, 0, repeat_max);
4728 }
4729 }
4730 }
4731
4732 /* The character or character type itself comes last in all cases. */
4733
4734 #ifdef SUPPORT_UTF8
4735 if (utf8 && c >= 128)
4736 {
4737 memcpy(code, utf8_char, c & 7);
4738 code += c & 7;
4739 }
4740 else
4741 #endif
4742 *code++ = c;
4743
4744 /* For a repeated Unicode property match, there are two extra bytes that
4745 define the required property. */
4746
4747 #ifdef SUPPORT_UCP
4748 if (prop_type >= 0)
4749 {
4750 *code++ = prop_type;
4751 *code++ = prop_value;
4752 }
4753 #endif
4754 }
4755
4756 /* If previous was a character class or a back reference, we put the repeat
4757 stuff after it, but just skip the item if the repeat was {0,0}. */
4758
4759 else if (*previous == OP_CLASS ||
4760 *previous == OP_NCLASS ||
4761 #ifdef SUPPORT_UTF8
4762 *previous == OP_XCLASS ||
4763 #endif
4764 *previous == OP_REF ||
4765 *previous == OP_REFI)
4766 {
4767 if (repeat_max == 0)
4768 {
4769 code = previous;
4770 goto END_REPEAT;
4771 }
4772
4773 /*--------------------------------------------------------------------*/
4774 /* This code is obsolete from release 8.00; the restriction was finally
4775 removed: */
4776
4777 /* All real repeats make it impossible to handle partial matching (maybe
4778 one day we will be able to remove this restriction). */
4779
4780 /* if (repeat_max != 1) cd->external_flags |= PCRE_NOPARTIAL; */
4781 /*--------------------------------------------------------------------*/
4782
4783 if (repeat_min == 0 && repeat_max == -1)
4784 *code++ = OP_CRSTAR + repeat_type;
4785 else if (repeat_min == 1 && repeat_max == -1)
4786 *code++ = OP_CRPLUS + repeat_type;
4787 else if (repeat_min == 0 && repeat_max == 1)
4788 *code++ = OP_CRQUERY + repeat_type;
4789 else
4790 {
4791 *code++ = OP_CRRANGE + repeat_type;
4792 PUT2INC(code, 0, repeat_min);
4793 if (repeat_max == -1) repeat_max = 0; /* 2-byte encoding for max */
4794 PUT2INC(code, 0, repeat_max);
4795 }
4796 }
4797
4798 /* If previous was a bracket group, we may have to replicate it in certain
4799 cases. Note that at this point we can encounter only the "basic" bracket
4800 opcodes such as BRA and CBRA, as this is the place where they get converted
4801 into the more special varieties such as BRAPOS and SBRA. A test for >=
4802 OP_ASSERT and <= OP_COND includes ASSERT, ASSERT_NOT, ASSERTBACK,
4803 ASSERTBACK_NOT, ONCE, BRA, CBRA, and COND. Originally, PCRE did not allow
4804 repetition of assertions, but now it does, for Perl compatibility. */
4805
4806 else if (*previous >= OP_ASSERT && *previous <= OP_COND)
4807 {
4808 register int i;
4809 int len = (int)(code - previous);
4810 pcre_uchar *bralink = NULL;
4811 pcre_uchar *brazeroptr = NULL;
4812
4813 /* Repeating a DEFINE group is pointless, but Perl allows the syntax, so
4814 we just ignore the repeat. */
4815
4816 if (*previous == OP_COND && previous[LINK_SIZE+1] == OP_DEF)
4817 goto END_REPEAT;
4818
4819 /* There is no sense in actually repeating assertions. The only potential
4820 use of repetition is in cases when the assertion is optional. Therefore,
4821 if the minimum is greater than zero, just ignore the repeat. If the
4822 maximum is not not zero or one, set it to 1. */
4823
4824 if (*previous < OP_ONCE) /* Assertion */
4825 {
4826 if (repeat_min > 0) goto END_REPEAT;
4827 if (repeat_max < 0 || repeat_max > 1) repeat_max = 1;
4828 }
4829
4830 /* The case of a zero minimum is special because of the need to stick
4831 OP_BRAZERO in front of it, and because the group appears once in the
4832 data, whereas in other cases it appears the minimum number of times. For
4833 this reason, it is simplest to treat this case separately, as otherwise
4834 the code gets far too messy. There are several special subcases when the
4835 minimum is zero. */
4836
4837 if (repeat_min == 0)
4838 {
4839 /* If the maximum is also zero, we used to just omit the group from the
4840 output altogether, like this:
4841
4842 ** if (repeat_max == 0)
4843 ** {
4844 ** code = previous;
4845 ** goto END_REPEAT;
4846 ** }
4847
4848 However, that fails when a group or a subgroup within it is referenced
4849 as a subroutine from elsewhere in the pattern, so now we stick in
4850 OP_SKIPZERO in front of it so that it is skipped on execution. As we
4851 don't have a list of which groups are referenced, we cannot do this
4852 selectively.
4853
4854 If the maximum is 1 or unlimited, we just have to stick in the BRAZERO
4855 and do no more at this point. However, we do need to adjust any
4856 OP_RECURSE calls inside the group that refer to the group itself or any
4857 internal or forward referenced group, because the offset is from the
4858 start of the whole regex. Temporarily terminate the pattern while doing
4859 this. */
4860
4861 if (repeat_max <= 1) /* Covers 0, 1, and unlimited */
4862 {
4863 *code = OP_END;
4864 adjust_recurse(previous, 1, utf8, cd, save_hwm);
4865 memmove(previous+1, previous, len);
4866 code++;
4867 if (repeat_max == 0)
4868 {
4869 *previous++ = OP_SKIPZERO;
4870 goto END_REPEAT;
4871 }
4872 brazeroptr = previous; /* Save for possessive optimizing */
4873 *previous++ = OP_BRAZERO + repeat_type;
4874 }
4875
4876 /* If the maximum is greater than 1 and limited, we have to replicate
4877 in a nested fashion, sticking OP_BRAZERO before each set of brackets.
4878 The first one has to be handled carefully because it's the original
4879 copy, which has to be moved up. The remainder can be handled by code
4880 that is common with the non-zero minimum case below. We have to
4881 adjust the value or repeat_max, since one less copy is required. Once
4882 again, we may have to adjust any OP_RECURSE calls inside the group. */
4883
4884 else
4885 {
4886 int offset;
4887 *code = OP_END;
4888 adjust_recurse(previous, 2 + LINK_SIZE, utf8, cd, save_hwm);
4889 memmove(previous + 2 + LINK_SIZE, previous, len);
4890 code += 2 + LINK_SIZE;
4891 *previous++ = OP_BRAZERO + repeat_type;
4892 *previous++ = OP_BRA;
4893
4894 /* We chain together the bracket offset fields that have to be
4895 filled in later when the ends of the brackets are reached. */
4896
4897 offset = (bralink == NULL)? 0 : (int)(previous - bralink);
4898 bralink = previous;
4899 PUTINC(previous, 0, offset);
4900 }
4901
4902 repeat_max--;
4903 }
4904
4905 /* If the minimum is greater than zero, replicate the group as many
4906 times as necessary, and adjust the maximum to the number of subsequent
4907 copies that we need. If we set a first char from the group, and didn't
4908 set a required char, copy the latter from the former. If there are any
4909 forward reference subroutine calls in the group, there will be entries on
4910 the workspace list; replicate these with an appropriate increment. */
4911
4912 else
4913 {
4914 if (repeat_min > 1)
4915 {
4916 /* In the pre-compile phase, we don't actually do the replication. We
4917 just adjust the length as if we had. Do some paranoid checks for
4918 potential integer overflow. The INT64_OR_DOUBLE type is a 64-bit
4919 integer type when available, otherwise double. */
4920
4921 if (lengthptr != NULL)
4922 {
4923 int delta = (repeat_min - 1)*length_prevgroup;
4924 if ((INT64_OR_DOUBLE)(repeat_min - 1)*
4925 (INT64_OR_DOUBLE)length_prevgroup >
4926 (INT64_OR_DOUBLE)INT_MAX ||
4927 OFLOW_MAX - *lengthptr < delta)
4928 {
4929 *errorcodeptr = ERR20;
4930 goto FAILED;
4931 }
4932 *lengthptr += delta;
4933 }
4934
4935 /* This is compiling for real */
4936
4937 else
4938 {
4939 if (groupsetfirstbyte && reqbyte < 0) reqbyte = firstbyte;
4940 for (i = 1; i < repeat_min; i++)
4941 {
4942 pcre_uchar *hc;
4943 pcre_uchar *this_hwm = cd->hwm;
4944 memcpy(code, previous, len);
4945 for (hc = save_hwm; hc < this_hwm; hc += LINK_SIZE)
4946 {
4947 PUT(cd->hwm, 0, GET(hc, 0) + len);
4948 cd->hwm += LINK_SIZE;
4949 }
4950 save_hwm = this_hwm;
4951 code += len;
4952 }
4953 }
4954 }
4955
4956 if (repeat_max > 0) repeat_max -= repeat_min;
4957 }
4958
4959 /* This code is common to both the zero and non-zero minimum cases. If
4960 the maximum is limited, it replicates the group in a nested fashion,
4961 remembering the bracket starts on a stack. In the case of a zero minimum,
4962 the first one was set up above. In all cases the repeat_max now specifies
4963 the number of additional copies needed. Again, we must remember to
4964 replicate entries on the forward reference list. */
4965
4966 if (repeat_max >= 0)
4967 {
4968 /* In the pre-compile phase, we don't actually do the replication. We
4969 just adjust the length as if we had. For each repetition we must add 1
4970 to the length for BRAZERO and for all but the last repetition we must
4971 add 2 + 2*LINKSIZE to allow for the nesting that occurs. Do some
4972 paranoid checks to avoid integer overflow. The INT64_OR_DOUBLE type is
4973 a 64-bit integer type when available, otherwise double. */
4974
4975 if (lengthptr != NULL && repeat_max > 0)
4976 {
4977 int delta = repeat_max * (length_prevgroup + 1 + 2 + 2*LINK_SIZE) -
4978 2 - 2*LINK_SIZE; /* Last one doesn't nest */
4979 if ((INT64_OR_DOUBLE)repeat_max *
4980 (INT64_OR_DOUBLE)(length_prevgroup + 1 + 2 + 2*LINK_SIZE)
4981 > (INT64_OR_DOUBLE)INT_MAX ||
4982 OFLOW_MAX - *lengthptr < delta)
4983 {
4984 *errorcodeptr = ERR20;
4985 goto FAILED;
4986 }
4987 *lengthptr += delta;
4988 }
4989
4990 /* This is compiling for real */
4991
4992 else for (i = repeat_max - 1; i >= 0; i--)
4993 {
4994 pcre_uchar *hc;
4995 pcre_uchar *this_hwm = cd->hwm;
4996
4997 *code++ = OP_BRAZERO + repeat_type;
4998
4999 /* All but the final copy start a new nesting, maintaining the
5000 chain of brackets outstanding. */
5001
5002 if (i != 0)
5003 {
5004 int offset;
5005 *code++ = OP_BRA;
5006 offset = (bralink == NULL)? 0 : (int)(code - bralink);
5007 bralink = code;
5008 PUTINC(code, 0, offset);
5009 }
5010
5011 memcpy(code, previous, len);
5012 for (hc = save_hwm; hc < this_hwm; hc += LINK_SIZE)
5013 {
5014 PUT(cd->hwm, 0, GET(hc, 0) + len + ((i != 0)? 2+LINK_SIZE : 1));
5015 cd->hwm += LINK_SIZE;
5016 }
5017 save_hwm = this_hwm;
5018 code += len;
5019 }
5020
5021 /* Now chain through the pending brackets, and fill in their length
5022 fields (which are holding the chain links pro tem). */
5023
5024 while (bralink != NULL)
5025 {
5026 int oldlinkoffset;
5027 int offset = (int)(code - bralink + 1);
5028 pcre_uchar *bra = code - offset;
5029 oldlinkoffset = GET(bra, 1);
5030 bralink = (oldlinkoffset == 0)? NULL : bralink - oldlinkoffset;
5031 *code++ = OP_KET;
5032 PUTINC(code, 0, offset);
5033 PUT(bra, 1, offset);
5034 }
5035 }
5036
5037 /* If the maximum is unlimited, set a repeater in the final copy. For
5038 ONCE brackets, that's all we need to do. However, possessively repeated
5039 ONCE brackets can be converted into non-capturing brackets, as the
5040 behaviour of (?:xx)++ is the same as (?>xx)++ and this saves having to
5041 deal with possessive ONCEs specially.
5042
5043 Otherwise, when we are doing the actual compile phase, check to see
5044 whether this group is one that could match an empty string. If so,
5045 convert the initial operator to the S form (e.g. OP_BRA -> OP_SBRA) so
5046 that runtime checking can be done. [This check is also applied to ONCE
5047 groups at runtime, but in a different way.]
5048
5049 Then, if the quantifier was possessive and the bracket is not a
5050 conditional, we convert the BRA code to the POS form, and the KET code to
5051 KETRPOS. (It turns out to be convenient at runtime to detect this kind of
5052 subpattern at both the start and at the end.) The use of special opcodes
5053 makes it possible to reduce greatly the stack usage in pcre_exec(). If
5054 the group is preceded by OP_BRAZERO, convert this to OP_BRAPOSZERO.
5055
5056 Then, if the minimum number of matches is 1 or 0, cancel the possessive
5057 flag so that the default action below, of wrapping everything inside
5058 atomic brackets, does not happen. When the minimum is greater than 1,
5059 there will be earlier copies of the group, and so we still have to wrap
5060 the whole thing. */
5061
5062 else
5063 {
5064 pcre_uchar *ketcode = code - 1 - LINK_SIZE;
5065 pcre_uchar *bracode = ketcode - GET(ketcode, 1);
5066
5067 /* Convert possessive ONCE brackets to non-capturing */
5068
5069 if ((*bracode == OP_ONCE || *bracode == OP_ONCE_NC) &&
5070 possessive_quantifier) *bracode = OP_BRA;
5071
5072 /* For non-possessive ONCE brackets, all we need to do is to
5073 set the KET. */
5074
5075 if (*bracode == OP_ONCE || *bracode == OP_ONCE_NC)
5076 *ketcode = OP_KETRMAX + repeat_type;
5077
5078 /* Handle non-ONCE brackets and possessive ONCEs (which have been
5079 converted to non-capturing above). */
5080
5081 else
5082 {
5083 /* In the compile phase, check for empty string matching. */
5084
5085 if (lengthptr == NULL)
5086 {
5087 pcre_uchar *scode = bracode;
5088 do
5089 {
5090 if (could_be_empty_branch(scode, ketcode, utf8, cd))
5091 {
5092 *bracode += OP_SBRA - OP_BRA;
5093 break;
5094 }
5095 scode += GET(scode, 1);
5096 }
5097 while (*scode == OP_ALT);
5098 }
5099
5100 /* Handle possessive quantifiers. */
5101
5102 if (possessive_quantifier)
5103 {
5104 /* For COND brackets, we wrap the whole thing in a possessively
5105 repeated non-capturing bracket, because we have not invented POS
5106 versions of the COND opcodes. Because we are moving code along, we
5107 must ensure that any pending recursive references are updated. */
5108
5109 if (*bracode == OP_COND || *bracode == OP_SCOND)
5110 {
5111 int nlen = (int)(code - bracode);
5112 *code = OP_END;
5113 adjust_recurse(bracode, 1 + LINK_SIZE, utf8, cd, save_hwm);
5114 memmove(bracode + 1+LINK_SIZE, bracode, nlen);
5115 code += 1 + LINK_SIZE;
5116 nlen += 1 + LINK_SIZE;
5117 *bracode = OP_BRAPOS;
5118 *code++ = OP_KETRPOS;
5119 PUTINC(code, 0, nlen);
5120 PUT(bracode, 1, nlen);
5121 }
5122
5123 /* For non-COND brackets, we modify the BRA code and use KETRPOS. */
5124
5125 else
5126 {
5127 *bracode += 1; /* Switch to xxxPOS opcodes */
5128 *ketcode = OP_KETRPOS;
5129 }
5130
5131 /* If the minimum is zero, mark it as possessive, then unset the
5132 possessive flag when the minimum is 0 or 1. */
5133
5134 if (brazeroptr != NULL) *brazeroptr = OP_BRAPOSZERO;
5135 if (repeat_min < 2) possessive_quantifier = FALSE;
5136 }
5137
5138 /* Non-possessive quantifier */
5139
5140 else *ketcode = OP_KETRMAX + repeat_type;
5141 }
5142 }
5143 }
5144
5145 /* If previous is OP_FAIL, it was generated by an empty class [] in
5146 JavaScript mode. The other ways in which OP_FAIL can be generated, that is
5147 by (*FAIL) or (?!) set previous to NULL, which gives a "nothing to repeat"
5148 error above. We can just ignore the repeat in JS case. */
5149
5150 else if (*previous == OP_FAIL) goto END_REPEAT;
5151
5152 /* Else there's some kind of shambles */
5153
5154 else
5155 {
5156 *errorcodeptr = ERR11;
5157 goto FAILED;
5158 }
5159
5160 /* If the character following a repeat is '+', or if certain optimization
5161 tests above succeeded, possessive_quantifier is TRUE. For some opcodes,
5162 there are special alternative opcodes for this case. For anything else, we
5163 wrap the entire repeated item inside OP_ONCE brackets. Logically, the '+'
5164 notation is just syntactic sugar, taken from Sun's Java package, but the
5165 special opcodes can optimize it.
5166
5167 Some (but not all) possessively repeated subpatterns have already been
5168 completely handled in the code just above. For them, possessive_quantifier
5169 is always FALSE at this stage.
5170
5171 Note that the repeated item starts at tempcode, not at previous, which
5172 might be the first part of a string whose (former) last char we repeated.
5173
5174 Possessifying an 'exact' quantifier has no effect, so we can ignore it. But
5175 an 'upto' may follow. We skip over an 'exact' item, and then test the
5176 length of what remains before proceeding. */
5177
5178 if (possessive_quantifier)
5179 {
5180 int len;
5181
5182 if (*tempcode == OP_TYPEEXACT)
5183 tempcode += _pcre_OP_lengths[*tempcode] +
5184 ((tempcode[3] == OP_PROP || tempcode[3] == OP_NOTPROP)? 2 : 0);
5185
5186 else if (*tempcode == OP_EXACT || *tempcode == OP_NOTEXACT)
5187 {
5188 tempcode += _pcre_OP_lengths[*tempcode];
5189 #ifdef SUPPORT_UTF8
5190 if (utf8 && tempcode[-1] >= 0xc0)
5191 tempcode += _pcre_utf8_table4[tempcode[-1] & 0x3f];
5192 #endif
5193 }
5194
5195 len = (int)(code - tempcode);
5196 if (len > 0) switch (*tempcode)
5197 {
5198 case OP_STAR: *tempcode = OP_POSSTAR; break;
5199 case OP_PLUS: *tempcode = OP_POSPLUS; break;
5200 case OP_QUERY: *tempcode = OP_POSQUERY; break;
5201 case OP_UPTO: *tempcode = OP_POSUPTO; break;
5202
5203 case OP_STARI: *tempcode = OP_POSSTARI; break;
5204 case OP_PLUSI: *tempcode = OP_POSPLUSI; break;
5205 case OP_QUERYI: *tempcode = OP_POSQUERYI; break;
5206 case OP_UPTOI: *tempcode = OP_POSUPTOI; break;
5207
5208 case OP_NOTSTAR: *tempcode = OP_NOTPOSSTAR; break;
5209 case OP_NOTPLUS: *tempcode = OP_NOTPOSPLUS; break;
5210 case OP_NOTQUERY: *tempcode = OP_NOTPOSQUERY; break;
5211 case OP_NOTUPTO: *tempcode = OP_NOTPOSUPTO; break;
5212
5213 case OP_NOTSTARI: *tempcode = OP_NOTPOSSTARI; break;
5214 case OP_NOTPLUSI: *tempcode = OP_NOTPOSPLUSI; break;
5215 case OP_NOTQUERYI: *tempcode = OP_NOTPOSQUERYI; break;
5216 case OP_NOTUPTOI: *tempcode = OP_NOTPOSUPTOI; break;
5217
5218 case OP_TYPESTAR: *tempcode = OP_TYPEPOSSTAR; break;
5219 case OP_TYPEPLUS: *tempcode = OP_TYPEPOSPLUS; break;
5220 case OP_TYPEQUERY: *tempcode = OP_TYPEPOSQUERY; break;
5221 case OP_TYPEUPTO: *tempcode = OP_TYPEPOSUPTO; break;
5222
5223 /* Because we are moving code along, we must ensure that any
5224 pending recursive references are updated. */
5225
5226 default:
5227 *code = OP_END;
5228 adjust_recurse(tempcode, 1 + LINK_SIZE, utf8, cd, save_hwm);
5229 memmove(tempcode + 1+LINK_SIZE, tempcode, len);
5230 code += 1 + LINK_SIZE;
5231 len += 1 + LINK_SIZE;
5232 tempcode[0] = OP_ONCE;
5233 *code++ = OP_KET;
5234 PUTINC(code, 0, len);
5235 PUT(tempcode, 1, len);
5236 break;
5237 }
5238 }
5239
5240 /* In all case we no longer have a previous item. We also set the
5241 "follows varying string" flag for subsequently encountered reqbytes if
5242 it isn't already set and we have just passed a varying length item. */
5243
5244 END_REPEAT:
5245 previous = NULL;
5246 cd->req_varyopt |= reqvary;
5247 break;
5248
5249
5250 /* ===================================================================*/
5251 /* Start of nested parenthesized sub-expression, or comment or lookahead or
5252 lookbehind or option setting or condition or all the other extended
5253 parenthesis forms. */
5254
5255 case CHAR_LEFT_PARENTHESIS:
5256 newoptions = options;
5257 skipbytes = 0;
5258 bravalue = OP_CBRA;
5259 save_hwm = cd->hwm;
5260 reset_bracount = FALSE;
5261
5262 /* First deal with various "verbs" that can be introduced by '*'. */
5263
5264 if (*(++ptr) == CHAR_ASTERISK &&
5265 ((cd->ctypes[ptr[1]] & ctype_letter) != 0 || ptr[1] == ':'))
5266 {
5267 int i, namelen;
5268 int arglen = 0;
5269 const char *vn = verbnames;
5270 const pcre_uchar *name = ptr + 1;
5271 const pcre_uchar *arg = NULL;
5272 previous = NULL;
5273 while ((cd->ctypes[*++ptr] & ctype_letter) != 0) {};
5274 namelen = (int)(ptr - name);
5275
5276 /* It appears that Perl allows any characters whatsoever, other than
5277 a closing parenthesis, to appear in arguments, so we no longer insist on
5278 letters, digits, and underscores. */
5279
5280 if (*ptr == CHAR_COLON)
5281 {
5282 arg = ++ptr;
5283 while (*ptr != 0 && *ptr != CHAR_RIGHT_PARENTHESIS) ptr++;
5284 arglen = (int)(ptr - arg);
5285 }
5286
5287 if (*ptr != CHAR_RIGHT_PARENTHESIS)
5288 {
5289 *errorcodeptr = ERR60;
5290 goto FAILED;
5291 }
5292
5293 /* Scan the table of verb names */
5294
5295 for (i = 0; i < verbcount; i++)
5296 {
5297 if (namelen == verbs[i].len &&
5298 strncmp((char *)name, vn, namelen) == 0)
5299 {
5300 /* Check for open captures before ACCEPT and convert it to
5301 ASSERT_ACCEPT if in an assertion. */
5302
5303 if (verbs[i].op == OP_ACCEPT)
5304 {
5305 open_capitem *oc;
5306 if (arglen != 0)
5307 {
5308 *errorcodeptr = ERR59;
5309 goto FAILED;
5310 }
5311 cd->had_accept = TRUE;
5312 for (oc = cd->open_caps; oc != NULL; oc = oc->next)
5313 {
5314 *code++ = OP_CLOSE;
5315 PUT2INC(code, 0, oc->number);
5316 }
5317 *code++ = (cd->assert_depth > 0)? OP_ASSERT_ACCEPT : OP_ACCEPT;
5318
5319 /* Do not set firstbyte after *ACCEPT */
5320 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
5321 }
5322
5323 /* Handle other cases with/without an argument */
5324
5325 else if (arglen == 0)
5326 {
5327 if (verbs[i].op < 0) /* Argument is mandatory */
5328 {
5329 *errorcodeptr = ERR66;
5330 goto FAILED;
5331 }
5332 *code = verbs[i].op;
5333 if (*code++ == OP_THEN) cd->external_flags |= PCRE_HASTHEN;
5334 }
5335
5336 else
5337 {
5338 if (verbs[i].op_arg < 0) /* Argument is forbidden */
5339 {
5340 *errorcodeptr = ERR59;
5341 goto FAILED;
5342 }
5343 *code = verbs[i].op_arg;
5344 if (*code++ == OP_THEN_ARG) cd->external_flags |= PCRE_HASTHEN;
5345 *code++ = arglen;
5346 memcpy(code, arg, arglen);
5347 code += arglen;
5348 *code++ = 0;
5349 }
5350
5351 break; /* Found verb, exit loop */
5352 }
5353
5354 vn += verbs[i].len + 1;
5355 }
5356
5357 if (i < verbcount) continue; /* Successfully handled a verb */
5358 *errorcodeptr = ERR60; /* Verb not recognized */
5359 goto FAILED;
5360 }
5361
5362 /* Deal with the extended parentheses; all are introduced by '?', and the
5363 appearance of any of them means that this is not a capturing group. */
5364
5365 else if (*ptr == CHAR_QUESTION_MARK)
5366 {
5367 int i, set, unset, namelen;
5368 int *optset;
5369 const pcre_uchar *name;
5370 pcre_uchar *slot;
5371
5372 switch (*(++ptr))
5373 {
5374 case CHAR_NUMBER_SIGN: /* Comment; skip to ket */
5375 ptr++;
5376 while (*ptr != 0 && *ptr != CHAR_RIGHT_PARENTHESIS) ptr++;
5377 if (*ptr == 0)
5378 {
5379 *errorcodeptr = ERR18;
5380 goto FAILED;
5381 }
5382 continue;
5383
5384
5385 /* ------------------------------------------------------------ */
5386 case CHAR_VERTICAL_LINE: /* Reset capture count for each branch */
5387 reset_bracount = TRUE;
5388 /* Fall through */
5389
5390 /* ------------------------------------------------------------ */
5391 case CHAR_COLON: /* Non-capturing bracket */
5392 bravalue = OP_BRA;
5393 ptr++;
5394 break;
5395
5396
5397 /* ------------------------------------------------------------ */
5398 case CHAR_LEFT_PARENTHESIS:
5399 bravalue = OP_COND; /* Conditional group */
5400
5401 /* A condition can be an assertion, a number (referring to a numbered
5402 group), a name (referring to a named group), or 'R', referring to
5403 recursion. R<digits> and R&name are also permitted for recursion tests.
5404
5405 There are several syntaxes for testing a named group: (?(name)) is used
5406 by Python; Perl 5.10 onwards uses (?(<name>) or (?('name')).
5407
5408 There are two unfortunate ambiguities, caused by history. (a) 'R' can
5409 be the recursive thing or the name 'R' (and similarly for 'R' followed
5410 by digits), and (b) a number could be a name that consists of digits.
5411 In both cases, we look for a name first; if not found, we try the other
5412 cases. */
5413
5414 /* For conditions that are assertions, check the syntax, and then exit
5415 the switch. This will take control down to where bracketed groups,
5416 including assertions, are processed. */
5417
5418 if (ptr[1] == CHAR_QUESTION_MARK && (ptr[2] == CHAR_EQUALS_SIGN ||
5419 ptr[2] == CHAR_EXCLAMATION_MARK || ptr[2] == CHAR_LESS_THAN_SIGN))
5420 break;
5421
5422 /* Most other conditions use OP_CREF (a couple change to OP_RREF
5423 below), and all need to skip 3 bytes at the start of the group. */
5424
5425 code[1+LINK_SIZE] = OP_CREF;
5426 skipbytes = 3;
5427 refsign = -1;
5428
5429 /* Check for a test for recursion in a named group. */
5430
5431 if (ptr[1] == CHAR_R && ptr[2] == CHAR_AMPERSAND)
5432 {
5433 terminator = -1;
5434 ptr += 2;
5435 code[1+LINK_SIZE] = OP_RREF; /* Change the type of test */
5436 }
5437
5438 /* Check for a test for a named group's having been set, using the Perl
5439 syntax (?(<name>) or (?('name') */
5440
5441 else if (ptr[1] == CHAR_LESS_THAN_SIGN)
5442 {
5443 terminator = CHAR_GREATER_THAN_SIGN;
5444 ptr++;
5445 }
5446 else if (ptr[1] == CHAR_APOSTROPHE)
5447 {
5448 terminator = CHAR_APOSTROPHE;
5449 ptr++;
5450 }
5451 else
5452 {
5453 terminator = 0;
5454 if (ptr[1] == CHAR_MINUS || ptr[1] == CHAR_PLUS) refsign = *(++ptr);
5455 }
5456
5457 /* We now expect to read a name; any thing else is an error */
5458
5459 if ((cd->ctypes[ptr[1]] & ctype_word) == 0)
5460 {
5461 ptr += 1; /* To get the right offset */
5462 *errorcodeptr = ERR28;
5463 goto FAILED;
5464 }
5465
5466 /* Read the name, but also get it as a number if it's all digits */
5467
5468 recno = 0;
5469 name = ++ptr;
5470 while ((cd->ctypes[*ptr] & ctype_word) != 0)
5471 {
5472 if (recno >= 0)
5473 recno = ((digitab[*ptr] & ctype_digit) != 0)?
5474 recno * 10 + *ptr - CHAR_0 : -1;
5475 ptr++;
5476 }
5477 namelen = (int)(ptr - name);
5478
5479 if ((terminator > 0 && *ptr++ != terminator) ||
5480 *ptr++ != CHAR_RIGHT_PARENTHESIS)
5481 {
5482 ptr--; /* Error offset */
5483 *errorcodeptr = ERR26;
5484 goto FAILED;
5485 }
5486
5487 /* Do no further checking in the pre-compile phase. */
5488
5489 if (lengthptr != NULL) break;
5490
5491 /* In the real compile we do the work of looking for the actual
5492 reference. If the string started with "+" or "-" we require the rest to
5493 be digits, in which case recno will be set. */
5494
5495 if (refsign > 0)
5496 {
5497 if (recno <= 0)
5498 {
5499 *errorcodeptr = ERR58;
5500 goto FAILED;
5501 }
5502 recno = (refsign == CHAR_MINUS)?
5503 cd->bracount - recno + 1 : recno +cd->bracount;
5504 if (recno <= 0 || recno > cd->final_bracount)
5505 {
5506 *errorcodeptr = ERR15;
5507 goto FAILED;
5508 }
5509 PUT2(code, 2+LINK_SIZE, recno);
5510 break;
5511 }
5512
5513 /* Otherwise (did not start with "+" or "-"), start by looking for the
5514 name. If we find a name, add one to the opcode to change OP_CREF or
5515 OP_RREF into OP_NCREF or OP_NRREF. These behave exactly the same,
5516 except they record that the reference was originally to a name. The
5517 information is used to check duplicate names. */
5518
5519 slot = cd->name_table;
5520 for (i = 0; i < cd->names_found; i++)
5521 {
5522 if (strncmp((char *)name, (char *)slot+2, namelen) == 0) break;
5523 slot += cd->name_entry_size;
5524 }
5525
5526 /* Found a previous named subpattern */
5527
5528 if (i < cd->names_found)
5529 {
5530 recno = GET2(slot, 0);
5531 PUT2(code, 2+LINK_SIZE, recno);
5532 code[1+LINK_SIZE]++;
5533 }
5534
5535 /* Search the pattern for a forward reference */
5536
5537 else if ((i = find_parens(cd, name, namelen,
5538 (options & PCRE_EXTENDED) != 0, utf8)) > 0)
5539 {
5540 PUT2(code, 2+LINK_SIZE, i);
5541 code[1+LINK_SIZE]++;
5542 }
5543
5544 /* If terminator == 0 it means that the name followed directly after
5545 the opening parenthesis [e.g. (?(abc)...] and in this case there are
5546 some further alternatives to try. For the cases where terminator != 0
5547 [things like (?(<name>... or (?('name')... or (?(R&name)... ] we have
5548 now checked all the possibilities, so give an error. */
5549
5550 else if (terminator != 0)
5551 {
5552 *errorcodeptr = ERR15;
5553 goto FAILED;
5554 }
5555
5556 /* Check for (?(R) for recursion. Allow digits after R to specify a
5557 specific group number. */
5558
5559 else if (*name == CHAR_R)
5560 {
5561 recno = 0;
5562 for (i = 1; i < namelen; i++)
5563 {
5564 if ((digitab[name[i]] & ctype_digit) == 0)
5565 {
5566 *errorcodeptr = ERR15;
5567 goto FAILED;
5568 }
5569 recno = recno * 10 + name[i] - CHAR_0;
5570 }
5571 if (recno == 0) recno = RREF_ANY;
5572 code[1+LINK_SIZE] = OP_RREF; /* Change test type */
5573 PUT2(code, 2+LINK_SIZE, recno);
5574 }
5575
5576 /* Similarly, check for the (?(DEFINE) "condition", which is always
5577 false. */
5578
5579 else if (namelen == 6 && strncmp((char *)name, STRING_DEFINE, 6) == 0)
5580 {
5581 code[1+LINK_SIZE] = OP_DEF;
5582 skipbytes = 1;
5583 }
5584
5585 /* Check for the "name" actually being a subpattern number. We are
5586 in the second pass here, so final_bracount is set. */
5587
5588 else if (recno > 0 && recno <= cd->final_bracount)
5589 {
5590 PUT2(code, 2+LINK_SIZE, recno);
5591 }
5592
5593 /* Either an unidentified subpattern, or a reference to (?(0) */
5594
5595 else
5596 {
5597 *errorcodeptr = (recno == 0)? ERR35: ERR15;
5598 goto FAILED;
5599 }
5600 break;
5601
5602
5603 /* ------------------------------------------------------------ */
5604 case CHAR_EQUALS_SIGN: /* Positive lookahead */
5605 bravalue = OP_ASSERT;
5606 cd->assert_depth += 1;
5607 ptr++;
5608 break;
5609
5610
5611 /* ------------------------------------------------------------ */
5612 case CHAR_EXCLAMATION_MARK: /* Negative lookahead */
5613 ptr++;
5614 if (*ptr == CHAR_RIGHT_PARENTHESIS) /* Optimize (?!) */
5615 {
5616 *code++ = OP_FAIL;
5617 previous = NULL;
5618 continue;
5619 }
5620 bravalue = OP_ASSERT_NOT;
5621 cd->assert_depth += 1;
5622 break;
5623
5624
5625 /* ------------------------------------------------------------ */
5626 case CHAR_LESS_THAN_SIGN: /* Lookbehind or named define */
5627 switch (ptr[1])
5628 {
5629 case CHAR_EQUALS_SIGN: /* Positive lookbehind */
5630 bravalue = OP_ASSERTBACK;
5631 cd->assert_depth += 1;
5632 ptr += 2;
5633 break;
5634
5635 case CHAR_EXCLAMATION_MARK: /* Negative lookbehind */
5636 bravalue = OP_ASSERTBACK_NOT;
5637 cd->assert_depth += 1;
5638 ptr += 2;
5639 break;
5640
5641 default: /* Could be name define, else bad */
5642 if ((cd->ctypes[ptr[1]] & ctype_word) != 0) goto DEFINE_NAME;
5643 ptr++; /* Correct offset for error */
5644 *errorcodeptr = ERR24;
5645 goto FAILED;
5646 }
5647 break;
5648
5649
5650 /* ------------------------------------------------------------ */
5651 case CHAR_GREATER_THAN_SIGN: /* One-time brackets */
5652 bravalue = OP_ONCE;
5653 ptr++;
5654 break;
5655
5656
5657 /* ------------------------------------------------------------ */
5658 case CHAR_C: /* Callout - may be followed by digits; */
5659 previous_callout = code; /* Save for later completion */
5660 after_manual_callout = 1; /* Skip one item before completing */
5661 *code++ = OP_CALLOUT;
5662 {
5663 int n = 0;
5664 while ((digitab[*(++ptr)] & ctype_digit) != 0)
5665 n = n * 10 + *ptr - CHAR_0;
5666 if (*ptr != CHAR_RIGHT_PARENTHESIS)
5667 {
5668 *errorcodeptr = ERR39;
5669 goto FAILED;
5670 }
5671 if (n > 255)
5672 {
5673 *errorcodeptr = ERR38;
5674 goto FAILED;
5675 }
5676 *code++ = n;
5677 PUT(code, 0, (int)(ptr - cd->start_pattern + 1)); /* Pattern offset */
5678 PUT(code, LINK_SIZE, 0); /* Default length */
5679 code += 2 * LINK_SIZE;
5680 }
5681 previous = NULL;
5682 continue;
5683
5684
5685 /* ------------------------------------------------------------ */
5686 case CHAR_P: /* Python-style named subpattern handling */
5687 if (*(++ptr) == CHAR_EQUALS_SIGN ||
5688 *ptr == CHAR_GREATER_THAN_SIGN) /* Reference or recursion */
5689 {
5690 is_recurse = *ptr == CHAR_GREATER_THAN_SIGN;
5691 terminator = CHAR_RIGHT_PARENTHESIS;
5692 goto NAMED_REF_OR_RECURSE;
5693 }
5694 else if (*ptr != CHAR_LESS_THAN_SIGN) /* Test for Python-style defn */
5695 {
5696 *errorcodeptr = ERR41;
5697 goto FAILED;
5698 }
5699 /* Fall through to handle (?P< as (?< is handled */
5700
5701
5702 /* ------------------------------------------------------------ */
5703 DEFINE_NAME: /* Come here from (?< handling */
5704 case CHAR_APOSTROPHE:
5705 {
5706 terminator = (*ptr == CHAR_LESS_THAN_SIGN)?
5707 CHAR_GREATER_THAN_SIGN : CHAR_APOSTROPHE;
5708 name = ++ptr;
5709
5710 while ((cd->ctypes[*ptr] & ctype_word) != 0) ptr++;
5711 namelen = (int)(ptr - name);
5712
5713 /* In the pre-compile phase, just do a syntax check. */
5714
5715 if (lengthptr != NULL)
5716 {
5717 if (*ptr != terminator)
5718 {
5719 *errorcodeptr = ERR42;
5720 goto FAILED;
5721 }
5722 if (cd->names_found >= MAX_NAME_COUNT)
5723 {
5724 *errorcodeptr = ERR49;
5725 goto FAILED;
5726 }
5727 if (namelen + 3 > cd->name_entry_size)
5728 {
5729 cd->name_entry_size = namelen + 3;
5730 if (namelen > MAX_NAME_SIZE)
5731 {
5732 *errorcodeptr = ERR48;
5733 goto FAILED;
5734 }
5735 }
5736 }
5737
5738 /* In the real compile, create the entry in the table, maintaining
5739 alphabetical order. Duplicate names for different numbers are
5740 permitted only if PCRE_DUPNAMES is set. Duplicate names for the same
5741 number are always OK. (An existing number can be re-used if (?|
5742 appears in the pattern.) In either event, a duplicate name results in
5743 a duplicate entry in the table, even if the number is the same. This
5744 is because the number of names, and hence the table size, is computed
5745 in the pre-compile, and it affects various numbers and pointers which
5746 would all have to be modified, and the compiled code moved down, if
5747 duplicates with the same number were omitted from the table. This
5748 doesn't seem worth the hassle. However, *different* names for the
5749 same number are not permitted. */
5750
5751 else
5752 {
5753 BOOL dupname = FALSE;
5754 slot = cd->name_table;
5755
5756 for (i = 0; i < cd->names_found; i++)
5757 {
5758 int crc = memcmp(name, slot+2, namelen);
5759 if (crc == 0)
5760 {
5761 if (slot[2+namelen] == 0)
5762 {
5763 if (GET2(slot, 0) != cd->bracount + 1 &&
5764 (options & PCRE_DUPNAMES) == 0)
5765 {
5766 *errorcodeptr = ERR43;
5767 goto FAILED;
5768 }
5769 else dupname = TRUE;
5770 }
5771 else crc = -1; /* Current name is a substring */
5772 }
5773
5774 /* Make space in the table and break the loop for an earlier
5775 name. For a duplicate or later name, carry on. We do this for
5776 duplicates so that in the simple case (when ?(| is not used) they
5777 are in order of their numbers. */
5778
5779 if (crc < 0)
5780 {
5781 memmove(slot + cd->name_entry_size, slot,
5782 (cd->names_found - i) * cd->name_entry_size);
5783 break;
5784 }
5785
5786 /* Continue the loop for a later or duplicate name */
5787
5788 slot += cd->name_entry_size;
5789 }
5790
5791 /* For non-duplicate names, check for a duplicate number before
5792 adding the new name. */
5793
5794 if (!dupname)
5795 {
5796 pcre_uchar *cslot = cd->name_table;
5797 for (i = 0; i < cd->names_found; i++)
5798 {
5799 if (cslot != slot)
5800 {
5801 if (GET2(cslot, 0) == cd->bracount + 1)
5802 {
5803 *errorcodeptr = ERR65;
5804 goto FAILED;
5805 }
5806 }
5807 else i--;
5808 cslot += cd->name_entry_size;
5809 }
5810 }
5811
5812 PUT2(slot, 0, cd->bracount + 1);
5813 memcpy(slot + 2, name, namelen);
5814 slot[2+namelen] = 0;
5815 }
5816 }
5817
5818 /* In both pre-compile and compile, count the number of names we've
5819 encountered. */
5820
5821 cd->names_found++;
5822 ptr++; /* Move past > or ' */
5823 goto NUMBERED_GROUP;
5824
5825
5826 /* ------------------------------------------------------------ */
5827 case CHAR_AMPERSAND: /* Perl recursion/subroutine syntax */
5828 terminator = CHAR_RIGHT_PARENTHESIS;
5829 is_recurse = TRUE;
5830 /* Fall through */
5831
5832 /* We come here from the Python syntax above that handles both
5833 references (?P=name) and recursion (?P>name), as well as falling
5834 through from the Perl recursion syntax (?&name). We also come here from
5835 the Perl \k<name> or \k'name' back reference syntax and the \k{name}
5836 .NET syntax, and the Oniguruma \g<...> and \g'...' subroutine syntax. */
5837
5838 NAMED_REF_OR_RECURSE:
5839 name = ++ptr;
5840 while ((cd->ctypes[*ptr] & ctype_word) != 0) ptr++;
5841 namelen = (int)(ptr - name);
5842
5843 /* In the pre-compile phase, do a syntax check. We used to just set
5844 a dummy reference number, because it was not used in the first pass.
5845 However, with the change of recursive back references to be atomic,
5846 we have to look for the number so that this state can be identified, as
5847 otherwise the incorrect length is computed. If it's not a backwards
5848 reference, the dummy number will do. */
5849
5850 if (lengthptr != NULL)
5851 {
5852 const pcre_uchar *temp;
5853
5854 if (namelen == 0)
5855 {
5856 *errorcodeptr = ERR62;
5857 goto FAILED;
5858 }
5859 if (*ptr != terminator)
5860 {
5861 *errorcodeptr = ERR42;
5862 goto FAILED;
5863 }
5864 if (namelen > MAX_NAME_SIZE)
5865 {
5866 *errorcodeptr = ERR48;
5867 goto FAILED;
5868 }
5869
5870 /* The name table does not exist in the first pass, so we cannot
5871 do a simple search as in the code below. Instead, we have to scan the
5872 pattern to find the number. It is important that we scan it only as
5873 far as we have got because the syntax of named subpatterns has not
5874 been checked for the rest of the pattern, and find_parens() assumes
5875 correct syntax. In any case, it's a waste of resources to scan
5876 further. We stop the scan at the current point by temporarily
5877 adjusting the value of cd->endpattern. */
5878
5879 temp = cd->end_pattern;
5880 cd->end_pattern = ptr;
5881 recno = find_parens(cd, name, namelen,
5882 (options & PCRE_EXTENDED) != 0, utf8);
5883 cd->end_pattern = temp;
5884 if (recno < 0) recno = 0; /* Forward ref; set dummy number */
5885 }
5886
5887 /* In the real compile, seek the name in the table. We check the name
5888 first, and then check that we have reached the end of the name in the
5889 table. That way, if the name that is longer than any in the table,
5890 the comparison will fail without reading beyond the table entry. */
5891
5892 else
5893 {
5894 slot = cd->name_table;
5895 for (i = 0; i < cd->names_found; i++)
5896 {
5897 if (strncmp((char *)name, (char *)slot+2, namelen) == 0 &&
5898 slot[2+namelen] == 0)
5899 break;
5900 slot += cd->name_entry_size;
5901 }
5902
5903 if (i < cd->names_found) /* Back reference */
5904 {
5905 recno = GET2(slot, 0);
5906 }
5907 else if ((recno = /* Forward back reference */
5908 find_parens(cd, name, namelen,
5909 (options & PCRE_EXTENDED) != 0, utf8)) <= 0)
5910 {
5911 *errorcodeptr = ERR15;
5912 goto FAILED;
5913 }
5914 }
5915
5916 /* In both phases, we can now go to the code than handles numerical
5917 recursion or backreferences. */
5918
5919 if (is_recurse) goto HANDLE_RECURSION;
5920 else goto HANDLE_REFERENCE;
5921
5922
5923 /* ------------------------------------------------------------ */
5924 case CHAR_R: /* Recursion */
5925 ptr++; /* Same as (?0) */
5926 /* Fall through */
5927
5928
5929 /* ------------------------------------------------------------ */
5930 case CHAR_MINUS: case CHAR_PLUS: /* Recursion or subroutine */
5931 case CHAR_0: case CHAR_1: case CHAR_2: case CHAR_3: case CHAR_4:
5932 case CHAR_5: case CHAR_6: case CHAR_7: case CHAR_8: case CHAR_9:
5933 {
5934 const pcre_uchar *called;
5935 terminator = CHAR_RIGHT_PARENTHESIS;
5936
5937 /* Come here from the \g<...> and \g'...' code (Oniguruma
5938 compatibility). However, the syntax has been checked to ensure that
5939 the ... are a (signed) number, so that neither ERR63 nor ERR29 will
5940 be called on this path, nor with the jump to OTHER_CHAR_AFTER_QUERY
5941 ever be taken. */
5942
5943 HANDLE_NUMERICAL_RECURSION:
5944
5945 if ((refsign = *ptr) == CHAR_PLUS)
5946 {
5947 ptr++;
5948 if ((digitab[*ptr] & ctype_digit) == 0)
5949 {
5950 *errorcodeptr = ERR63;
5951 goto FAILED;
5952 }
5953 }
5954 else if (refsign == CHAR_MINUS)
5955 {
5956 if ((digitab[ptr[1]] & ctype_digit) == 0)
5957 goto OTHER_CHAR_AFTER_QUERY;
5958 ptr++;
5959 }
5960
5961 recno = 0;
5962 while((digitab[*ptr] & ctype_digit) != 0)
5963 recno = recno * 10 + *ptr++ - CHAR_0;
5964
5965 if (*ptr != terminator)
5966 {
5967 *errorcodeptr = ERR29;
5968 goto FAILED;
5969 }
5970
5971 if (refsign == CHAR_MINUS)
5972 {
5973 if (recno == 0)
5974 {
5975 *errorcodeptr = ERR58;
5976 goto FAILED;
5977 }
5978 recno = cd->bracount - recno + 1;
5979 if (recno <= 0)
5980 {
5981 *errorcodeptr = ERR15;
5982 goto FAILED;
5983 }
5984 }
5985 else if (refsign == CHAR_PLUS)
5986 {
5987 if (recno == 0)
5988 {
5989 *errorcodeptr = ERR58;
5990 goto FAILED;
5991 }
5992 recno += cd->bracount;
5993 }
5994
5995 /* Come here from code above that handles a named recursion */
5996
5997 HANDLE_RECURSION:
5998
5999 previous = code;
6000 called = cd->start_code;
6001
6002 /* When we are actually compiling, find the bracket that is being
6003 referenced. Temporarily end the regex in case it doesn't exist before
6004 this point. If we end up with a forward reference, first check that
6005 the bracket does occur later so we can give the error (and position)
6006 now. Then remember this forward reference in the workspace so it can
6007 be filled in at the end. */
6008
6009 if (lengthptr == NULL)
6010 {
6011 *code = OP_END;
6012 if (recno != 0)
6013 called = _pcre_find_bracket(cd->start_code, utf8, recno);
6014
6015 /* Forward reference */
6016
6017 if (called == NULL)
6018 {
6019 if (find_parens(cd, NULL, recno,
6020 (options & PCRE_EXTENDED) != 0, utf8) < 0)
6021 {
6022 *errorcodeptr = ERR15;
6023 goto FAILED;
6024 }
6025
6026 /* Fudge the value of "called" so that when it is inserted as an
6027 offset below, what it actually inserted is the reference number
6028 of the group. Then remember the forward reference. */
6029
6030 called = cd->start_code + recno;
6031 PUTINC(cd->hwm, 0, (int)(code + 1 - cd->start_code));
6032 }
6033
6034 /* If not a forward reference, and the subpattern is still open,
6035 this is a recursive call. We check to see if this is a left
6036 recursion that could loop for ever, and diagnose that case. We
6037 must not, however, do this check if we are in a conditional
6038 subpattern because the condition might be testing for recursion in
6039 a pattern such as /(?(R)a+|(?R)b)/, which is perfectly valid.
6040 Forever loops are also detected at runtime, so those that occur in
6041 conditional subpatterns will be picked up then. */
6042
6043 else if (GET(called, 1) == 0 && cond_depth <= 0 &&
6044 could_be_empty(called, code, bcptr, utf8, cd))
6045 {
6046 *errorcodeptr = ERR40;
6047 goto FAILED;
6048 }
6049 }
6050
6051 /* Insert the recursion/subroutine item. */
6052
6053 *code = OP_RECURSE;
6054 PUT(code, 1, (int)(called - cd->start_code));
6055 code += 1 + LINK_SIZE;
6056 }
6057
6058 /* Can't determine a first byte now */
6059
6060 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
6061 continue;
6062
6063
6064 /* ------------------------------------------------------------ */
6065 default: /* Other characters: check option setting */
6066 OTHER_CHAR_AFTER_QUERY:
6067 set = unset = 0;
6068 optset = &set;
6069
6070 while (*ptr != CHAR_RIGHT_PARENTHESIS && *ptr != CHAR_COLON)
6071 {
6072 switch (*ptr++)
6073 {
6074 case CHAR_MINUS: optset = &unset; break;
6075
6076 case CHAR_J: /* Record that it changed in the external options */
6077 *optset |= PCRE_DUPNAMES;
6078 cd->external_flags |= PCRE_JCHANGED;
6079 break;
6080
6081 case CHAR_i: *optset |= PCRE_CASELESS; break;
6082 case CHAR_m: *optset |= PCRE_MULTILINE; break;
6083 case CHAR_s: *optset |= PCRE_DOTALL; break;
6084 case CHAR_x: *optset |= PCRE_EXTENDED; break;
6085 case CHAR_U: *optset |= PCRE_UNGREEDY; break;
6086 case CHAR_X: *optset |= PCRE_EXTRA; break;
6087
6088 default: *errorcodeptr = ERR12;
6089 ptr--; /* Correct the offset */
6090 goto FAILED;
6091 }
6092 }
6093
6094 /* Set up the changed option bits, but don't change anything yet. */
6095
6096 newoptions = (options | set) & (~unset);
6097
6098 /* If the options ended with ')' this is not the start of a nested
6099 group with option changes, so the options change at this level. If this
6100 item is right at the start of the pattern, the options can be
6101 abstracted and made external in the pre-compile phase, and ignored in
6102 the compile phase. This can be helpful when matching -- for instance in
6103 caseless checking of required bytes.
6104
6105 If the code pointer is not (cd->start_code + 1 + LINK_SIZE), we are
6106 definitely *not* at the start of the pattern because something has been
6107 compiled. In the pre-compile phase, however, the code pointer can have
6108 that value after the start, because it gets reset as code is discarded
6109 during the pre-compile. However, this can happen only at top level - if
6110 we are within parentheses, the starting BRA will still be present. At
6111 any parenthesis level, the length value can be used to test if anything
6112 has been compiled at that level. Thus, a test for both these conditions
6113 is necessary to ensure we correctly detect the start of the pattern in
6114 both phases.
6115
6116 If we are not at the pattern start, reset the greedy defaults and the
6117 case value for firstbyte and reqbyte. */
6118
6119 if (*ptr == CHAR_RIGHT_PARENTHESIS)
6120 {
6121 if (code == cd->start_code + 1 + LINK_SIZE &&
6122 (lengthptr == NULL || *lengthptr == 2 + 2*LINK_SIZE))
6123 {
6124 cd->external_options = newoptions;
6125 }
6126 else
6127 {
6128 greedy_default = ((newoptions & PCRE_UNGREEDY) != 0);
6129 greedy_non_default = greedy_default ^ 1;
6130 req_caseopt = ((newoptions & PCRE_CASELESS) != 0)? REQ_CASELESS : 0;
6131 }
6132
6133 /* Change options at this level, and pass them back for use
6134 in subsequent branches. */
6135
6136 *optionsptr = options = newoptions;
6137 previous = NULL; /* This item can't be repeated */
6138 continue; /* It is complete */
6139 }
6140
6141 /* If the options ended with ':' we are heading into a nested group
6142 with possible change of options. Such groups are non-capturing and are
6143 not assertions of any kind. All we need to do is skip over the ':';
6144 the newoptions value is handled below. */
6145
6146 bravalue = OP_BRA;
6147 ptr++;
6148 } /* End of switch for character following (? */
6149 } /* End of (? handling */
6150
6151 /* Opening parenthesis not followed by '*' or '?'. If PCRE_NO_AUTO_CAPTURE
6152 is set, all unadorned brackets become non-capturing and behave like (?:...)
6153 brackets. */
6154
6155 else if ((options & PCRE_NO_AUTO_CAPTURE) != 0)
6156 {
6157 bravalue = OP_BRA;
6158 }
6159
6160 /* Else we have a capturing group. */
6161
6162 else
6163 {
6164 NUMBERED_GROUP:
6165 cd->bracount += 1;
6166 PUT2(code, 1+LINK_SIZE, cd->bracount);
6167 skipbytes = 2;
6168 }
6169
6170 /* Process nested bracketed regex. Assertions used not to be repeatable,
6171 but this was changed for Perl compatibility, so all kinds can now be
6172 repeated. We copy code into a non-register variable (tempcode) in order to
6173 be able to pass its address because some compilers complain otherwise. */
6174
6175 previous = code; /* For handling repetition */
6176 *code = bravalue;
6177 tempcode = code;
6178 tempreqvary = cd->req_varyopt; /* Save value before bracket */
6179 tempbracount = cd->bracount; /* Save value before bracket */
6180 length_prevgroup = 0; /* Initialize for pre-compile phase */
6181
6182 if (!compile_regex(
6183 newoptions, /* The complete new option state */
6184 &tempcode, /* Where to put code (updated) */
6185 &ptr, /* Input pointer (updated) */
6186 errorcodeptr, /* Where to put an error message */
6187 (bravalue == OP_ASSERTBACK ||
6188 bravalue == OP_ASSERTBACK_NOT), /* TRUE if back assert */
6189 reset_bracount, /* True if (?| group */
6190 skipbytes, /* Skip over bracket number */
6191 cond_depth +
6192 ((bravalue == OP_COND)?1:0), /* Depth of condition subpatterns */
6193 &subfirstbyte, /* For possible first char */
6194 &subreqbyte, /* For possible last char */
6195 bcptr, /* Current branch chain */
6196 cd, /* Tables block */
6197 (lengthptr == NULL)? NULL : /* Actual compile phase */
6198 &length_prevgroup /* Pre-compile phase */
6199 ))
6200 goto FAILED;
6201
6202 /* If this was an atomic group and there are no capturing groups within it,
6203 generate OP_ONCE_NC instead of OP_ONCE. */
6204
6205 if (bravalue == OP_ONCE && cd->bracount <= tempbracount)
6206 *code = OP_ONCE_NC;
6207
6208 if (bravalue >= OP_ASSERT && bravalue <= OP_ASSERTBACK_NOT)
6209 cd->assert_depth -= 1;
6210
6211 /* At the end of compiling, code is still pointing to the start of the
6212 group, while tempcode has been updated to point past the end of the group.
6213 The pattern pointer (ptr) is on the bracket.
6214
6215 If this is a conditional bracket, check that there are no more than
6216 two branches in the group, or just one if it's a DEFINE group. We do this
6217 in the real compile phase, not in the pre-pass, where the whole group may
6218 not be available. */
6219
6220 if (bravalue == OP_COND && lengthptr == NULL)
6221 {
6222 pcre_uchar *tc = code;
6223 int condcount = 0;
6224
6225 do {
6226 condcount++;
6227 tc += GET(tc,1);
6228 }
6229 while (*tc != OP_KET);
6230
6231 /* A DEFINE group is never obeyed inline (the "condition" is always
6232 false). It must have only one branch. */
6233
6234 if (code[LINK_SIZE+1] == OP_DEF)
6235 {
6236 if (condcount > 1)
6237 {
6238 *errorcodeptr = ERR54;
6239 goto FAILED;
6240 }
6241 bravalue = OP_DEF; /* Just a flag to suppress char handling below */
6242 }
6243
6244 /* A "normal" conditional group. If there is just one branch, we must not
6245 make use of its firstbyte or reqbyte, because this is equivalent to an
6246 empty second branch. */
6247
6248 else
6249 {
6250 if (condcount > 2)
6251 {
6252 *errorcodeptr = ERR27;
6253 goto FAILED;
6254 }
6255 if (condcount == 1) subfirstbyte = subreqbyte = REQ_NONE;
6256 }
6257 }
6258
6259 /* Error if hit end of pattern */
6260
6261 if (*ptr != CHAR_RIGHT_PARENTHESIS)
6262 {
6263 *errorcodeptr = ERR14;
6264 goto FAILED;
6265 }
6266
6267 /* In the pre-compile phase, update the length by the length of the group,
6268 less the brackets at either end. Then reduce the compiled code to just a
6269 set of non-capturing brackets so that it doesn't use much memory if it is
6270 duplicated by a quantifier.*/
6271
6272 if (lengthptr != NULL)
6273 {
6274 if (OFLOW_MAX - *lengthptr < length_prevgroup - 2 - 2*LINK_SIZE)
6275 {
6276 *errorcodeptr = ERR20;
6277 goto FAILED;
6278 }
6279 *lengthptr += length_prevgroup - 2 - 2*LINK_SIZE;
6280 code++; /* This already contains bravalue */
6281 PUTINC(code, 0, 1 + LINK_SIZE);
6282 *code++ = OP_KET;
6283 PUTINC(code, 0, 1 + LINK_SIZE);
6284 break; /* No need to waste time with special character handling */
6285 }
6286
6287 /* Otherwise update the main code pointer to the end of the group. */
6288
6289 code = tempcode;
6290
6291 /* For a DEFINE group, required and first character settings are not
6292 relevant. */
6293
6294 if (bravalue == OP_DEF) break;
6295
6296 /* Handle updating of the required and first characters for other types of
6297 group. Update for normal brackets of all kinds, and conditions with two
6298 branches (see code above). If the bracket is followed by a quantifier with
6299 zero repeat, we have to back off. Hence the definition of zeroreqbyte and
6300 zerofirstbyte outside the main loop so that they can be accessed for the
6301 back off. */
6302
6303 zeroreqbyte = reqbyte;
6304 zerofirstbyte = firstbyte;
6305 groupsetfirstbyte = FALSE;
6306
6307 if (bravalue >= OP_ONCE)
6308 {
6309 /* If we have not yet set a firstbyte in this branch, take it from the
6310 subpattern, remembering that it was set here so that a repeat of more
6311 than one can replicate it as reqbyte if necessary. If the subpattern has
6312 no firstbyte, set "none" for the whole branch. In both cases, a zero
6313 repeat forces firstbyte to "none". */
6314
6315 if (firstbyte == REQ_UNSET)
6316 {
6317 if (subfirstbyte >= 0)
6318 {
6319 firstbyte = subfirstbyte;
6320 groupsetfirstbyte = TRUE;
6321 }
6322 else firstbyte = REQ_NONE;
6323 zerofirstbyte = REQ_NONE;
6324 }
6325
6326 /* If firstbyte was previously set, convert the subpattern's firstbyte
6327 into reqbyte if there wasn't one, using the vary flag that was in
6328 existence beforehand. */
6329
6330 else if (subfirstbyte >= 0 && subreqbyte < 0)
6331 subreqbyte = subfirstbyte | tempreqvary;
6332
6333 /* If the subpattern set a required byte (or set a first byte that isn't
6334 really the first byte - see above), set it. */
6335
6336 if (subreqbyte >= 0) reqbyte = subreqbyte;
6337 }
6338
6339 /* For a forward assertion, we take the reqbyte, if set. This can be
6340 helpful if the pattern that follows the assertion doesn't set a different
6341 char. For example, it's useful for /(?=abcde).+/. We can't set firstbyte
6342 for an assertion, however because it leads to incorrect effect for patterns
6343 such as /(?=a)a.+/ when the "real" "a" would then become a reqbyte instead
6344 of a firstbyte. This is overcome by a scan at the end if there's no
6345 firstbyte, looking for an asserted first char. */
6346
6347 else if (bravalue == OP_ASSERT && subreqbyte >= 0) reqbyte = subreqbyte;
6348 break; /* End of processing '(' */
6349
6350
6351 /* ===================================================================*/
6352 /* Handle metasequences introduced by \. For ones like \d, the ESC_ values
6353 are arranged to be the negation of the corresponding OP_values in the
6354 default case when PCRE_UCP is not set. For the back references, the values
6355 are ESC_REF plus the reference number. Only back references and those types
6356 that consume a character may be repeated. We can test for values between
6357 ESC_b and ESC_Z for the latter; this may have to change if any new ones are
6358 ever created. */
6359
6360 case CHAR_BACKSLASH:
6361 tempptr = ptr;
6362 c = check_escape(&ptr, errorcodeptr, cd->bracount, options, FALSE);
6363 if (*errorcodeptr != 0) goto FAILED;
6364
6365 if (c < 0)
6366 {
6367 if (-c == ESC_Q) /* Handle start of quoted string */
6368 {
6369 if (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
6370 ptr += 2; /* avoid empty string */
6371 else inescq = TRUE;
6372 continue;
6373 }
6374
6375 if (-c == ESC_E) continue; /* Perl ignores an orphan \E */
6376
6377 /* For metasequences that actually match a character, we disable the
6378 setting of a first character if it hasn't already been set. */
6379
6380 if (firstbyte == REQ_UNSET && -c > ESC_b && -c < ESC_Z)
6381 firstbyte = REQ_NONE;
6382
6383 /* Set values to reset to if this is followed by a zero repeat. */
6384
6385 zerofirstbyte = firstbyte;
6386 zeroreqbyte = reqbyte;
6387
6388 /* \g<name> or \g'name' is a subroutine call by name and \g<n> or \g'n'
6389 is a subroutine call by number (Oniguruma syntax). In fact, the value
6390 -ESC_g is returned only for these cases. So we don't need to check for <
6391 or ' if the value is -ESC_g. For the Perl syntax \g{n} the value is
6392 -ESC_REF+n, and for the Perl syntax \g{name} the result is -ESC_k (as
6393 that is a synonym for a named back reference). */
6394
6395 if (-c == ESC_g)
6396 {
6397 const pcre_uchar *p;
6398 save_hwm = cd->hwm; /* Normally this is set when '(' is read */
6399 terminator = (*(++ptr) == CHAR_LESS_THAN_SIGN)?
6400 CHAR_GREATER_THAN_SIGN : CHAR_APOSTROPHE;
6401
6402 /* These two statements stop the compiler for warning about possibly
6403 unset variables caused by the jump to HANDLE_NUMERICAL_RECURSION. In
6404 fact, because we actually check for a number below, the paths that
6405 would actually be in error are never taken. */
6406
6407 skipbytes = 0;
6408 reset_bracount = FALSE;
6409
6410 /* Test for a name */
6411
6412 if (ptr[1] != CHAR_PLUS && ptr[1] != CHAR_MINUS)
6413 {
6414 BOOL isnumber = TRUE;
6415 for (p = ptr + 1; *p != 0 && *p != terminator; p++)
6416 {
6417 if ((cd->ctypes[*p] & ctype_digit) == 0) isnumber = FALSE;
6418 if ((cd->ctypes[*p] & ctype_word) == 0) break;
6419 }
6420 if (*p != terminator)
6421 {
6422 *errorcodeptr = ERR57;
6423 break;
6424 }
6425 if (isnumber)
6426 {
6427 ptr++;
6428 goto HANDLE_NUMERICAL_RECURSION;
6429 }
6430 is_recurse = TRUE;
6431 goto NAMED_REF_OR_RECURSE;
6432 }
6433
6434 /* Test a signed number in angle brackets or quotes. */
6435
6436 p = ptr + 2;
6437 while ((digitab[*p] & ctype_digit) != 0) p++;
6438 if (*p != terminator)
6439 {
6440 *errorcodeptr = ERR57;
6441 break;
6442 }
6443 ptr++;
6444 goto HANDLE_NUMERICAL_RECURSION;
6445 }
6446
6447 /* \k<name> or \k'name' is a back reference by name (Perl syntax).
6448 We also support \k{name} (.NET syntax). */
6449
6450 if (-c == ESC_k)
6451 {
6452 if ((ptr[1] != CHAR_LESS_THAN_SIGN &&
6453 ptr[1] != CHAR_APOSTROPHE && ptr[1] != CHAR_LEFT_CURLY_BRACKET))
6454 {
6455 *errorcodeptr = ERR69;
6456 break;
6457 }
6458 is_recurse = FALSE;
6459 terminator = (*(++ptr) == CHAR_LESS_THAN_SIGN)?
6460 CHAR_GREATER_THAN_SIGN : (*ptr == CHAR_APOSTROPHE)?
6461 CHAR_APOSTROPHE : CHAR_RIGHT_CURLY_BRACKET;
6462 goto NAMED_REF_OR_RECURSE;
6463 }
6464
6465 /* Back references are handled specially; must disable firstbyte if
6466 not set to cope with cases like (?=(\w+))\1: which would otherwise set
6467 ':' later. */
6468
6469 if (-c >= ESC_REF)
6470 {
6471 open_capitem *oc;
6472 recno = -c - ESC_REF;
6473
6474 HANDLE_REFERENCE: /* Come here from named backref handling */
6475 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
6476 previous = code;
6477 *code++ = ((options & PCRE_CASELESS) != 0)? OP_REFI : OP_REF;
6478 PUT2INC(code, 0, recno);
6479 cd->backref_map |= (recno < 32)? (1 << recno) : 1;
6480 if (recno > cd->top_backref) cd->top_backref = recno;
6481
6482 /* Check to see if this back reference is recursive, that it, it
6483 is inside the group that it references. A flag is set so that the
6484 group can be made atomic. */
6485
6486 for (oc = cd->open_caps; oc != NULL; oc = oc->next)
6487 {
6488 if (oc->number == recno)
6489 {
6490 oc->flag = TRUE;
6491 break;
6492 }
6493 }
6494 }
6495
6496 /* So are Unicode property matches, if supported. */
6497
6498 #ifdef SUPPORT_UCP
6499 else if (-c == ESC_P || -c == ESC_p)
6500 {
6501 BOOL negated;
6502 int pdata;
6503 int ptype = get_ucp(&ptr, &negated, &pdata, errorcodeptr);
6504 if (ptype < 0) goto FAILED;
6505 previous = code;
6506 *code++ = ((-c == ESC_p) != negated)? OP_PROP : OP_NOTPROP;
6507 *code++ = ptype;
6508 *code++ = pdata;
6509 }
6510 #else
6511
6512 /* If Unicode properties are not supported, \X, \P, and \p are not
6513 allowed. */
6514
6515 else if (-c == ESC_X || -c == ESC_P || -c == ESC_p)
6516 {
6517 *errorcodeptr = ERR45;
6518 goto FAILED;
6519 }
6520 #endif
6521
6522 /* For the rest (including \X when Unicode properties are supported), we
6523 can obtain the OP value by negating the escape value in the default
6524 situation when PCRE_UCP is not set. When it *is* set, we substitute
6525 Unicode property tests. */
6526
6527 else
6528 {
6529 #ifdef SUPPORT_UCP
6530 if (-c >= ESC_DU && -c <= ESC_wu)
6531 {
6532 nestptr = ptr + 1; /* Where to resume */
6533 ptr = substitutes[-c - ESC_DU] - 1; /* Just before substitute */
6534 }
6535 else
6536 #endif
6537 /* In non-UTF-8 mode, we turn \C into OP_ALLANY instead of OP_ANYBYTE
6538 so that it works in DFA mode and in lookbehinds. */
6539
6540 {
6541 previous = (-c > ESC_b && -c < ESC_Z)? code : NULL;
6542 *code++ = (!utf8 && c == -ESC_C)? OP_ALLANY : -c;
6543 }
6544 }
6545 continue;
6546 }
6547
6548 /* We have a data character whose value is in c. In UTF-8 mode it may have
6549 a value > 127. We set its representation in the length/buffer, and then
6550 handle it as a data character. */
6551
6552 #ifdef SUPPORT_UTF8
6553 if (utf8 && c > 127)
6554 mclength = _pcre_ord2utf8(c, mcbuffer);
6555 else
6556 #endif
6557
6558 {
6559 mcbuffer[0] = c;
6560 mclength = 1;
6561 }
6562 goto ONE_CHAR;
6563
6564
6565 /* ===================================================================*/
6566 /* Handle a literal character. It is guaranteed not to be whitespace or #
6567 when the extended flag is set. If we are in UTF-8 mode, it may be a
6568 multi-byte literal character. */
6569
6570 default:
6571 NORMAL_CHAR:
6572 mclength = 1;
6573 mcbuffer[0] = c;
6574
6575 #ifdef SUPPORT_UTF8
6576 if (utf8 && c >= 0xc0)
6577 {
6578 while ((ptr[1] & 0xc0) == 0x80)
6579 mcbuffer[mclength++] = *(++ptr);
6580 }
6581 #endif
6582
6583 /* At this point we have the character's bytes in mcbuffer, and the length
6584 in mclength. When not in UTF-8 mode, the length is always 1. */
6585
6586 ONE_CHAR:
6587 previous = code;
6588 *code++ = ((options & PCRE_CASELESS) != 0)? OP_CHARI : OP_CHAR;
6589 for (c = 0; c < mclength; c++) *code++ = mcbuffer[c];
6590
6591 /* Remember if \r or \n were seen */
6592
6593 if (mcbuffer[0] == CHAR_CR || mcbuffer[0] == CHAR_NL)
6594 cd->external_flags |= PCRE_HASCRORLF;
6595
6596 /* Set the first and required bytes appropriately. If no previous first
6597 byte, set it from this character, but revert to none on a zero repeat.
6598 Otherwise, leave the firstbyte value alone, and don't change it on a zero
6599 repeat. */
6600
6601 if (firstbyte == REQ_UNSET)
6602 {
6603 zerofirstbyte = REQ_NONE;
6604 zeroreqbyte = reqbyte;
6605
6606 /* If the character is more than one byte long, we can set firstbyte
6607 only if it is not to be matched caselessly. */
6608
6609 if (mclength == 1 || req_caseopt == 0)
6610