/[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 756 - (show annotations) (download)
Mon Nov 21 10:48:42 2011 UTC (2 years, 9 months ago) by ph10
File MIME type: text/plain
File size: 252830 byte(s)
Apply Zoltan's big patch.

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 literal_PNd[] = { '\\', 'P', '{', 'N', 'd', '}', '\0' };
270 static const pcre_uchar literal_pNd[] = { '\\', 'p', '{', 'N', 'd', '}', '\0' };
271 static const pcre_uchar literal_PXsp[] = { '\\', 'P', '{', 'X', 's', 'p', '}', '\0' };
272 static const pcre_uchar literal_pXsp[] = { '\\', 'p', '{', 'X', 's', 'p', '}', '\0' };
273 static const pcre_uchar literal_PXwd[] = { '\\', 'P', '{', 'X', 'w', 'd', '}', '\0' };
274 static const pcre_uchar literal_pXwd[] = { '\\', 'p', '{', 'X', 'w', 'd', '}', '\0' };
275
276 static const pcre_uchar *substitutes[] = {
277 literal_PNd, /* \D */
278 literal_pNd, /* \d */
279 literal_PXsp, /* \S */ /* NOTE: Xsp is Perl space */
280 literal_pXsp, /* \s */
281 literal_PXwd, /* \W */
282 literal_pXwd /* \w */
283 };
284
285 static const pcre_uchar literal_pL[] = { '\\', 'p', '{', 'L', '}', '\0' };
286 static const pcre_uchar literal_pLl[] = { '\\', 'p', '{', 'L', 'l', '}', '\0' };
287 static const pcre_uchar literal_pLu[] = { '\\', 'p', '{', 'L', 'u', '}', '\0' };
288 static const pcre_uchar literal_pXan[] = { '\\', 'p', '{', 'X', 'a', 'n', '}', '\0' };
289 static const pcre_uchar literal_h[] = { '\\', 'h', '\0' };
290 static const pcre_uchar literal_pXps[] = { '\\', 'p', '{', 'X', 'p', 's', '}', '\0' };
291 static const pcre_uchar literal_PL[] = { '\\', 'P', '{', 'L', '}', '\0' };
292 static const pcre_uchar literal_PLl[] = { '\\', 'P', '{', 'L', 'l', '}', '\0' };
293 static const pcre_uchar literal_PLu[] = { '\\', 'P', '{', 'L', 'u', '}', '\0' };
294 static const pcre_uchar literal_PXan[] = { '\\', 'P', '{', 'X', 'a', 'n', '}', '\0' };
295 static const pcre_uchar literal_H[] = { '\\', 'H', '\0' };
296 static const pcre_uchar literal_PXps[] = { '\\', 'P', '{', 'X', 'p', 's', '}', '\0' };
297
298 static const pcre_uchar *posix_substitutes[] = {
299 literal_pL, /* alpha */
300 literal_pLl, /* lower */
301 literal_pLu, /* upper */
302 literal_pXan, /* alnum */
303 NULL, /* ascii */
304 literal_h, /* blank */
305 NULL, /* cntrl */
306 literal_pNd, /* digit */
307 NULL, /* graph */
308 NULL, /* print */
309 NULL, /* punct */
310 literal_pXps, /* space */ /* NOTE: Xps is POSIX space */
311 literal_pXwd, /* word */
312 NULL, /* xdigit */
313 /* Negated cases */
314 literal_PL, /* ^alpha */
315 literal_PLl, /* ^lower */
316 literal_PLu, /* ^upper */
317 literal_PXan, /* ^alnum */
318 NULL, /* ^ascii */
319 literal_H, /* ^blank */
320 NULL, /* ^cntrl */
321 literal_PNd, /* ^digit */
322 NULL, /* ^graph */
323 NULL, /* ^print */
324 NULL, /* ^punct */
325 literal_PXps, /* ^space */ /* NOTE: Xps is POSIX space */
326 literal_PXwd, /* ^word */
327 NULL /* ^xdigit */
328 };
329 #define POSIX_SUBSIZE (sizeof(posix_substitutes) / sizeof(pcre_uchar *))
330 #endif
331
332 #define STRING(a) # a
333 #define XSTRING(s) STRING(s)
334
335 /* The texts of compile-time error messages. These are "char *" because they
336 are passed to the outside world. Do not ever re-use any error number, because
337 they are documented. Always add a new error instead. Messages marked DEAD below
338 are no longer used. This used to be a table of strings, but in order to reduce
339 the number of relocations needed when a shared library is loaded dynamically,
340 it is now one long string. We cannot use a table of offsets, because the
341 lengths of inserts such as XSTRING(MAX_NAME_SIZE) are not known. Instead, we
342 simply count through to the one we want - this isn't a performance issue
343 because these strings are used only when there is a compilation error.
344
345 Each substring ends with \0 to insert a null character. This includes the final
346 substring, so that the whole string ends with \0\0, which can be detected when
347 counting through. */
348
349 static const char error_texts[] =
350 "no error\0"
351 "\\ at end of pattern\0"
352 "\\c at end of pattern\0"
353 "unrecognized character follows \\\0"
354 "numbers out of order in {} quantifier\0"
355 /* 5 */
356 "number too big in {} quantifier\0"
357 "missing terminating ] for character class\0"
358 "invalid escape sequence in character class\0"
359 "range out of order in character class\0"
360 "nothing to repeat\0"
361 /* 10 */
362 "operand of unlimited repeat could match the empty string\0" /** DEAD **/
363 "internal error: unexpected repeat\0"
364 "unrecognized character after (? or (?-\0"
365 "POSIX named classes are supported only within a class\0"
366 "missing )\0"
367 /* 15 */
368 "reference to non-existent subpattern\0"
369 "erroffset passed as NULL\0"
370 "unknown option bit(s) set\0"
371 "missing ) after comment\0"
372 "parentheses nested too deeply\0" /** DEAD **/
373 /* 20 */
374 "regular expression is too large\0"
375 "failed to get memory\0"
376 "unmatched parentheses\0"
377 "internal error: code overflow\0"
378 "unrecognized character after (?<\0"
379 /* 25 */
380 "lookbehind assertion is not fixed length\0"
381 "malformed number or name after (?(\0"
382 "conditional group contains more than two branches\0"
383 "assertion expected after (?(\0"
384 "(?R or (?[+-]digits must be followed by )\0"
385 /* 30 */
386 "unknown POSIX class name\0"
387 "POSIX collating elements are not supported\0"
388 "this version of PCRE is not compiled with PCRE_UTF8 support\0"
389 "spare error\0" /** DEAD **/
390 "character value in \\x{...} sequence is too large\0"
391 /* 35 */
392 "invalid condition (?(0)\0"
393 "\\C not allowed in lookbehind assertion\0"
394 "PCRE does not support \\L, \\l, \\N{name}, \\U, or \\u\0"
395 "number after (?C is > 255\0"
396 "closing ) for (?C expected\0"
397 /* 40 */
398 "recursive call could loop indefinitely\0"
399 "unrecognized character after (?P\0"
400 "syntax error in subpattern name (missing terminator)\0"
401 "two named subpatterns have the same name\0"
402 "invalid UTF-8 string\0"
403 /* 45 */
404 "support for \\P, \\p, and \\X has not been compiled\0"
405 "malformed \\P or \\p sequence\0"
406 "unknown property name after \\P or \\p\0"
407 "subpattern name is too long (maximum " XSTRING(MAX_NAME_SIZE) " characters)\0"
408 "too many named subpatterns (maximum " XSTRING(MAX_NAME_COUNT) ")\0"
409 /* 50 */
410 "repeated subpattern is too long\0" /** DEAD **/
411 "octal value is greater than \\377 (not in UTF-8 mode)\0"
412 "internal error: overran compiling workspace\0"
413 "internal error: previously-checked referenced subpattern not found\0"
414 "DEFINE group contains more than one branch\0"
415 /* 55 */
416 "repeating a DEFINE group is not allowed\0" /** DEAD **/
417 "inconsistent NEWLINE options\0"
418 "\\g is not followed by a braced, angle-bracketed, or quoted name/number or by a plain number\0"
419 "a numbered reference must not be zero\0"
420 "an argument is not allowed for (*ACCEPT), (*FAIL), or (*COMMIT)\0"
421 /* 60 */
422 "(*VERB) not recognized\0"
423 "number is too big\0"
424 "subpattern name expected\0"
425 "digit expected after (?+\0"
426 "] is an invalid data character in JavaScript compatibility mode\0"
427 /* 65 */
428 "different names for subpatterns of the same number are not allowed\0"
429 "(*MARK) must have an argument\0"
430 "this version of PCRE is not compiled with PCRE_UCP support\0"
431 "\\c must be followed by an ASCII character\0"
432 "\\k is not followed by a braced, angle-bracketed, or quoted name\0"
433 /* 70 */
434 "internal error: unknown opcode in find_fixedlength()\0"
435 ;
436
437 /* Table to identify digits and hex digits. This is used when compiling
438 patterns. Note that the tables in chartables are dependent on the locale, and
439 may mark arbitrary characters as digits - but the PCRE compiling code expects
440 to handle only 0-9, a-z, and A-Z as digits when compiling. That is why we have
441 a private table here. It costs 256 bytes, but it is a lot faster than doing
442 character value tests (at least in some simple cases I timed), and in some
443 applications one wants PCRE to compile efficiently as well as match
444 efficiently.
445
446 For convenience, we use the same bit definitions as in chartables:
447
448 0x04 decimal digit
449 0x08 hexadecimal digit
450
451 Then we can use ctype_digit and ctype_xdigit in the code. */
452
453 #ifndef EBCDIC
454
455 /* This is the "normal" case, for ASCII systems, and EBCDIC systems running in
456 UTF-8 mode. */
457
458 static const unsigned char digitab[] =
459 {
460 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 0- 7 */
461 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 8- 15 */
462 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 16- 23 */
463 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 24- 31 */
464 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - ' */
465 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* ( - / */
466 0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c, /* 0 - 7 */
467 0x0c,0x0c,0x00,0x00,0x00,0x00,0x00,0x00, /* 8 - ? */
468 0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* @ - G */
469 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* H - O */
470 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* P - W */
471 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* X - _ */
472 0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* ` - g */
473 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* h - o */
474 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* p - w */
475 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* x -127 */
476 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 128-135 */
477 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 136-143 */
478 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 144-151 */
479 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 152-159 */
480 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 160-167 */
481 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 168-175 */
482 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 176-183 */
483 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 184-191 */
484 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 192-199 */
485 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 200-207 */
486 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 208-215 */
487 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 216-223 */
488 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 224-231 */
489 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 232-239 */
490 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 240-247 */
491 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00};/* 248-255 */
492
493 #else
494
495 /* This is the "abnormal" case, for EBCDIC systems not running in UTF-8 mode. */
496
497 static const unsigned char digitab[] =
498 {
499 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 0- 7 0 */
500 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 8- 15 */
501 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 16- 23 10 */
502 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 24- 31 */
503 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 32- 39 20 */
504 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 40- 47 */
505 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 48- 55 30 */
506 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 56- 63 */
507 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - 71 40 */
508 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 72- | */
509 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* & - 87 50 */
510 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 88- 95 */
511 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - -103 60 */
512 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 104- ? */
513 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 112-119 70 */
514 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 120- " */
515 0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* 128- g 80 */
516 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* h -143 */
517 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 144- p 90 */
518 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* q -159 */
519 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 160- x A0 */
520 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* y -175 */
521 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* ^ -183 B0 */
522 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 184-191 */
523 0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* { - G C0 */
524 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* H -207 */
525 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* } - P D0 */
526 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* Q -223 */
527 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* \ - X E0 */
528 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* Y -239 */
529 0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c, /* 0 - 7 F0 */
530 0x0c,0x0c,0x00,0x00,0x00,0x00,0x00,0x00};/* 8 -255 */
531
532 static const unsigned char ebcdic_chartab[] = { /* chartable partial dup */
533 0x80,0x00,0x00,0x00,0x00,0x01,0x00,0x00, /* 0- 7 */
534 0x00,0x00,0x00,0x00,0x01,0x01,0x00,0x00, /* 8- 15 */
535 0x00,0x00,0x00,0x00,0x00,0x01,0x00,0x00, /* 16- 23 */
536 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 24- 31 */
537 0x00,0x00,0x00,0x00,0x00,0x01,0x00,0x00, /* 32- 39 */
538 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 40- 47 */
539 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 48- 55 */
540 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 56- 63 */
541 0x01,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - 71 */
542 0x00,0x00,0x00,0x80,0x00,0x80,0x80,0x80, /* 72- | */
543 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* & - 87 */
544 0x00,0x00,0x00,0x80,0x80,0x80,0x00,0x00, /* 88- 95 */
545 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - -103 */
546 0x00,0x00,0x00,0x00,0x00,0x10,0x00,0x80, /* 104- ? */
547 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 112-119 */
548 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 120- " */
549 0x00,0x1a,0x1a,0x1a,0x1a,0x1a,0x1a,0x12, /* 128- g */
550 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* h -143 */
551 0x00,0x12,0x12,0x12,0x12,0x12,0x12,0x12, /* 144- p */
552 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* q -159 */
553 0x00,0x00,0x12,0x12,0x12,0x12,0x12,0x12, /* 160- x */
554 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* y -175 */
555 0x80,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* ^ -183 */
556 0x00,0x00,0x80,0x00,0x00,0x00,0x00,0x00, /* 184-191 */
557 0x80,0x1a,0x1a,0x1a,0x1a,0x1a,0x1a,0x12, /* { - G */
558 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* H -207 */
559 0x00,0x12,0x12,0x12,0x12,0x12,0x12,0x12, /* } - P */
560 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* Q -223 */
561 0x00,0x00,0x12,0x12,0x12,0x12,0x12,0x12, /* \ - X */
562 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* Y -239 */
563 0x1c,0x1c,0x1c,0x1c,0x1c,0x1c,0x1c,0x1c, /* 0 - 7 */
564 0x1c,0x1c,0x00,0x00,0x00,0x00,0x00,0x00};/* 8 -255 */
565 #endif
566
567
568 /* Definition to allow mutual recursion */
569
570 static BOOL
571 compile_regex(int, pcre_uchar **, const pcre_uchar **, int *, BOOL, BOOL, int, int,
572 int *, int *, branch_chain *, compile_data *, int *);
573
574
575
576 /*************************************************
577 * Find an error text *
578 *************************************************/
579
580 /* The error texts are now all in one long string, to save on relocations. As
581 some of the text is of unknown length, we can't use a table of offsets.
582 Instead, just count through the strings. This is not a performance issue
583 because it happens only when there has been a compilation error.
584
585 Argument: the error number
586 Returns: pointer to the error string
587 */
588
589 static const char *
590 find_error_text(int n)
591 {
592 const char *s = error_texts;
593 for (; n > 0; n--)
594 {
595 while (*s++ != 0) {};
596 if (*s == 0) return "Error text not found (please report)";
597 }
598 return s;
599 }
600
601
602 /*************************************************
603 * Check for counted repeat *
604 *************************************************/
605
606 /* This function is called when a '{' is encountered in a place where it might
607 start a quantifier. It looks ahead to see if it really is a quantifier or not.
608 It is only a quantifier if it is one of the forms {ddd} {ddd,} or {ddd,ddd}
609 where the ddds are digits.
610
611 Arguments:
612 p pointer to the first char after '{'
613
614 Returns: TRUE or FALSE
615 */
616
617 static BOOL
618 is_counted_repeat(const pcre_uchar *p)
619 {
620 if ((digitab[*p++] & ctype_digit) == 0) return FALSE;
621 while ((digitab[*p] & ctype_digit) != 0) p++;
622 if (*p == CHAR_RIGHT_CURLY_BRACKET) return TRUE;
623
624 if (*p++ != CHAR_COMMA) return FALSE;
625 if (*p == CHAR_RIGHT_CURLY_BRACKET) return TRUE;
626
627 if ((digitab[*p++] & ctype_digit) == 0) return FALSE;
628 while ((digitab[*p] & ctype_digit) != 0) p++;
629
630 return (*p == CHAR_RIGHT_CURLY_BRACKET);
631 }
632
633
634
635 /*************************************************
636 * Handle escapes *
637 *************************************************/
638
639 /* This function is called when a \ has been encountered. It either returns a
640 positive value for a simple escape such as \n, or a negative value which
641 encodes one of the more complicated things such as \d. A backreference to group
642 n is returned as -(ESC_REF + n); ESC_REF is the highest ESC_xxx macro. When
643 UTF-8 is enabled, a positive value greater than 255 may be returned. On entry,
644 ptr is pointing at the \. On exit, it is on the final character of the escape
645 sequence.
646
647 Arguments:
648 ptrptr points to the pattern position pointer
649 errorcodeptr points to the errorcode variable
650 bracount number of previous extracting brackets
651 options the options bits
652 isclass TRUE if inside a character class
653
654 Returns: zero or positive => a data character
655 negative => a special escape sequence
656 on error, errorcodeptr is set
657 */
658
659 static int
660 check_escape(const pcre_uchar **ptrptr, int *errorcodeptr, int bracount,
661 int options, BOOL isclass)
662 {
663 BOOL utf8 = (options & PCRE_UTF8) != 0;
664 const pcre_uchar *ptr = *ptrptr + 1;
665 int c, i;
666
667 GETCHARINCTEST(c, ptr); /* Get character value, increment pointer */
668 ptr--; /* Set pointer back to the last byte */
669
670 /* If backslash is at the end of the pattern, it's an error. */
671
672 if (c == 0) *errorcodeptr = ERR1;
673
674 /* Non-alphanumerics are literals. For digits or letters, do an initial lookup
675 in a table. A non-zero result is something that can be returned immediately.
676 Otherwise further processing may be required. */
677
678 #ifndef EBCDIC /* ASCII/UTF-8 coding */
679 else if (c < CHAR_0 || c > CHAR_z) {} /* Not alphanumeric */
680 else if ((i = escapes[c - CHAR_0]) != 0) c = i;
681
682 #else /* EBCDIC coding */
683 else if (c < 'a' || (ebcdic_chartab[c] & 0x0E) == 0) {} /* Not alphanumeric */
684 else if ((i = escapes[c - 0x48]) != 0) c = i;
685 #endif
686
687 /* Escapes that need further processing, or are illegal. */
688
689 else
690 {
691 const pcre_uchar *oldptr;
692 BOOL braced, negated;
693
694 switch (c)
695 {
696 /* A number of Perl escapes are not handled by PCRE. We give an explicit
697 error. */
698
699 case CHAR_l:
700 case CHAR_L:
701 *errorcodeptr = ERR37;
702 break;
703
704 case CHAR_u:
705 if ((options & PCRE_JAVASCRIPT_COMPAT) != 0)
706 {
707 /* In JavaScript, \u must be followed by four hexadecimal numbers.
708 Otherwise it is a lowercase u letter. */
709 if ((digitab[ptr[1]] & ctype_xdigit) != 0 && (digitab[ptr[2]] & ctype_xdigit) != 0
710 && (digitab[ptr[3]] & ctype_xdigit) != 0 && (digitab[ptr[4]] & ctype_xdigit) != 0)
711 {
712 c = 0;
713 for (i = 0; i < 4; ++i)
714 {
715 register int cc = *(++ptr);
716 #ifndef EBCDIC /* ASCII/UTF-8 coding */
717 if (cc >= CHAR_a) cc -= 32; /* Convert to upper case */
718 c = (c << 4) + cc - ((cc < CHAR_A)? CHAR_0 : (CHAR_A - 10));
719 #else /* EBCDIC coding */
720 if (cc >= CHAR_a && cc <= CHAR_z) cc += 64; /* Convert to upper case */
721 c = (c << 4) + cc - ((cc >= CHAR_0)? CHAR_0 : (CHAR_A - 10));
722 #endif
723 }
724 }
725 }
726 else
727 *errorcodeptr = ERR37;
728 break;
729
730 case CHAR_U:
731 /* In JavaScript, \U is an uppercase U letter. */
732 if ((options & PCRE_JAVASCRIPT_COMPAT) == 0) *errorcodeptr = ERR37;
733 break;
734
735 /* In a character class, \g is just a literal "g". Outside a character
736 class, \g must be followed by one of a number of specific things:
737
738 (1) A number, either plain or braced. If positive, it is an absolute
739 backreference. If negative, it is a relative backreference. This is a Perl
740 5.10 feature.
741
742 (2) Perl 5.10 also supports \g{name} as a reference to a named group. This
743 is part of Perl's movement towards a unified syntax for back references. As
744 this is synonymous with \k{name}, we fudge it up by pretending it really
745 was \k.
746
747 (3) For Oniguruma compatibility we also support \g followed by a name or a
748 number either in angle brackets or in single quotes. However, these are
749 (possibly recursive) subroutine calls, _not_ backreferences. Just return
750 the -ESC_g code (cf \k). */
751
752 case CHAR_g:
753 if (isclass) break;
754 if (ptr[1] == CHAR_LESS_THAN_SIGN || ptr[1] == CHAR_APOSTROPHE)
755 {
756 c = -ESC_g;
757 break;
758 }
759
760 /* Handle the Perl-compatible cases */
761
762 if (ptr[1] == CHAR_LEFT_CURLY_BRACKET)
763 {
764 const pcre_uchar *p;
765 for (p = ptr+2; *p != 0 && *p != CHAR_RIGHT_CURLY_BRACKET; p++)
766 if (*p != CHAR_MINUS && (digitab[*p] & ctype_digit) == 0) break;
767 if (*p != 0 && *p != CHAR_RIGHT_CURLY_BRACKET)
768 {
769 c = -ESC_k;
770 break;
771 }
772 braced = TRUE;
773 ptr++;
774 }
775 else braced = FALSE;
776
777 if (ptr[1] == CHAR_MINUS)
778 {
779 negated = TRUE;
780 ptr++;
781 }
782 else negated = FALSE;
783
784 c = 0;
785 while ((digitab[ptr[1]] & ctype_digit) != 0)
786 c = c * 10 + *(++ptr) - CHAR_0;
787
788 if (c < 0) /* Integer overflow */
789 {
790 *errorcodeptr = ERR61;
791 break;
792 }
793
794 if (braced && *(++ptr) != CHAR_RIGHT_CURLY_BRACKET)
795 {
796 *errorcodeptr = ERR57;
797 break;
798 }
799
800 if (c == 0)
801 {
802 *errorcodeptr = ERR58;
803 break;
804 }
805
806 if (negated)
807 {
808 if (c > bracount)
809 {
810 *errorcodeptr = ERR15;
811 break;
812 }
813 c = bracount - (c - 1);
814 }
815
816 c = -(ESC_REF + c);
817 break;
818
819 /* The handling of escape sequences consisting of a string of digits
820 starting with one that is not zero is not straightforward. By experiment,
821 the way Perl works seems to be as follows:
822
823 Outside a character class, the digits are read as a decimal number. If the
824 number is less than 10, or if there are that many previous extracting
825 left brackets, then it is a back reference. Otherwise, up to three octal
826 digits are read to form an escaped byte. Thus \123 is likely to be octal
827 123 (cf \0123, which is octal 012 followed by the literal 3). If the octal
828 value is greater than 377, the least significant 8 bits are taken. Inside a
829 character class, \ followed by a digit is always an octal number. */
830
831 case CHAR_1: case CHAR_2: case CHAR_3: case CHAR_4: case CHAR_5:
832 case CHAR_6: case CHAR_7: case CHAR_8: case CHAR_9:
833
834 if (!isclass)
835 {
836 oldptr = ptr;
837 c -= CHAR_0;
838 while ((digitab[ptr[1]] & ctype_digit) != 0)
839 c = c * 10 + *(++ptr) - CHAR_0;
840 if (c < 0) /* Integer overflow */
841 {
842 *errorcodeptr = ERR61;
843 break;
844 }
845 if (c < 10 || c <= bracount)
846 {
847 c = -(ESC_REF + c);
848 break;
849 }
850 ptr = oldptr; /* Put the pointer back and fall through */
851 }
852
853 /* Handle an octal number following \. If the first digit is 8 or 9, Perl
854 generates a binary zero byte and treats the digit as a following literal.
855 Thus we have to pull back the pointer by one. */
856
857 if ((c = *ptr) >= CHAR_8)
858 {
859 ptr--;
860 c = 0;
861 break;
862 }
863
864 /* \0 always starts an octal number, but we may drop through to here with a
865 larger first octal digit. The original code used just to take the least
866 significant 8 bits of octal numbers (I think this is what early Perls used
867 to do). Nowadays we allow for larger numbers in UTF-8 mode, but no more
868 than 3 octal digits. */
869
870 case CHAR_0:
871 c -= CHAR_0;
872 while(i++ < 2 && ptr[1] >= CHAR_0 && ptr[1] <= CHAR_7)
873 c = c * 8 + *(++ptr) - CHAR_0;
874 if (!utf8 && c > 255) *errorcodeptr = ERR51;
875 break;
876
877 /* \x is complicated. \x{ddd} is a character number which can be greater
878 than 0xff in utf8 mode, but only if the ddd are hex digits. If not, { is
879 treated as a data character. */
880
881 case CHAR_x:
882 if ((options & PCRE_JAVASCRIPT_COMPAT) != 0)
883 {
884 /* In JavaScript, \x must be followed by two hexadecimal numbers.
885 Otherwise it is a lowercase x letter. */
886 if ((digitab[ptr[1]] & ctype_xdigit) != 0 && (digitab[ptr[2]] & ctype_xdigit) != 0)
887 {
888 c = 0;
889 for (i = 0; i < 2; ++i)
890 {
891 register int cc = *(++ptr);
892 #ifndef EBCDIC /* ASCII/UTF-8 coding */
893 if (cc >= CHAR_a) cc -= 32; /* Convert to upper case */
894 c = (c << 4) + cc - ((cc < CHAR_A)? CHAR_0 : (CHAR_A - 10));
895 #else /* EBCDIC coding */
896 if (cc >= CHAR_a && cc <= CHAR_z) cc += 64; /* Convert to upper case */
897 c = (c << 4) + cc - ((cc >= CHAR_0)? CHAR_0 : (CHAR_A - 10));
898 #endif
899 }
900 }
901 break;
902 }
903
904 if (ptr[1] == CHAR_LEFT_CURLY_BRACKET)
905 {
906 const pcre_uchar *pt = ptr + 2;
907 int count = 0;
908
909 c = 0;
910 while ((digitab[*pt] & ctype_xdigit) != 0)
911 {
912 register int cc = *pt++;
913 if (c == 0 && cc == CHAR_0) continue; /* Leading zeroes */
914 count++;
915
916 #ifndef EBCDIC /* ASCII/UTF-8 coding */
917 if (cc >= CHAR_a) cc -= 32; /* Convert to upper case */
918 c = (c << 4) + cc - ((cc < CHAR_A)? CHAR_0 : (CHAR_A - 10));
919 #else /* EBCDIC coding */
920 if (cc >= CHAR_a && cc <= CHAR_z) cc += 64; /* Convert to upper case */
921 c = (c << 4) + cc - ((cc >= CHAR_0)? CHAR_0 : (CHAR_A - 10));
922 #endif
923 }
924
925 if (*pt == CHAR_RIGHT_CURLY_BRACKET)
926 {
927 if (c < 0 || count > (utf8? 8 : 2)) *errorcodeptr = ERR34;
928 ptr = pt;
929 break;
930 }
931
932 /* If the sequence of hex digits does not end with '}', then we don't
933 recognize this construct; fall through to the normal \x handling. */
934 }
935
936 /* Read just a single-byte hex-defined char */
937
938 c = 0;
939 while (i++ < 2 && (digitab[ptr[1]] & ctype_xdigit) != 0)
940 {
941 int cc; /* Some compilers don't like */
942 cc = *(++ptr); /* ++ in initializers */
943 #ifndef EBCDIC /* ASCII/UTF-8 coding */
944 if (cc >= CHAR_a) cc -= 32; /* Convert to upper case */
945 c = c * 16 + cc - ((cc < CHAR_A)? CHAR_0 : (CHAR_A - 10));
946 #else /* EBCDIC coding */
947 if (cc <= CHAR_z) cc += 64; /* Convert to upper case */
948 c = c * 16 + cc - ((cc >= CHAR_0)? CHAR_0 : (CHAR_A - 10));
949 #endif
950 }
951 break;
952
953 /* For \c, a following letter is upper-cased; then the 0x40 bit is flipped.
954 An error is given if the byte following \c is not an ASCII character. This
955 coding is ASCII-specific, but then the whole concept of \cx is
956 ASCII-specific. (However, an EBCDIC equivalent has now been added.) */
957
958 case CHAR_c:
959 c = *(++ptr);
960 if (c == 0)
961 {
962 *errorcodeptr = ERR2;
963 break;
964 }
965 #ifndef EBCDIC /* ASCII/UTF-8 coding */
966 if (c > 127) /* Excludes all non-ASCII in either mode */
967 {
968 *errorcodeptr = ERR68;
969 break;
970 }
971 if (c >= CHAR_a && c <= CHAR_z) c -= 32;
972 c ^= 0x40;
973 #else /* EBCDIC coding */
974 if (c >= CHAR_a && c <= CHAR_z) c += 64;
975 c ^= 0xC0;
976 #endif
977 break;
978
979 /* PCRE_EXTRA enables extensions to Perl in the matter of escapes. Any
980 other alphanumeric following \ is an error if PCRE_EXTRA was set;
981 otherwise, for Perl compatibility, it is a literal. This code looks a bit
982 odd, but there used to be some cases other than the default, and there may
983 be again in future, so I haven't "optimized" it. */
984
985 default:
986 if ((options & PCRE_EXTRA) != 0) switch(c)
987 {
988 default:
989 *errorcodeptr = ERR3;
990 break;
991 }
992 break;
993 }
994 }
995
996 /* Perl supports \N{name} for character names, as well as plain \N for "not
997 newline". PCRE does not support \N{name}. However, it does support
998 quantification such as \N{2,3}. */
999
1000 if (c == -ESC_N && ptr[1] == CHAR_LEFT_CURLY_BRACKET &&
1001 !is_counted_repeat(ptr+2))
1002 *errorcodeptr = ERR37;
1003
1004 /* If PCRE_UCP is set, we change the values for \d etc. */
1005
1006 if ((options & PCRE_UCP) != 0 && c <= -ESC_D && c >= -ESC_w)
1007 c -= (ESC_DU - ESC_D);
1008
1009 /* Set the pointer to the final character before returning. */
1010
1011 *ptrptr = ptr;
1012 return c;
1013 }
1014
1015
1016
1017 #ifdef SUPPORT_UCP
1018 /*************************************************
1019 * Handle \P and \p *
1020 *************************************************/
1021
1022 /* This function is called after \P or \p has been encountered, provided that
1023 PCRE is compiled with support for Unicode properties. On entry, ptrptr is
1024 pointing at the P or p. On exit, it is pointing at the final character of the
1025 escape sequence.
1026
1027 Argument:
1028 ptrptr points to the pattern position pointer
1029 negptr points to a boolean that is set TRUE for negation else FALSE
1030 dptr points to an int that is set to the detailed property value
1031 errorcodeptr points to the error code variable
1032
1033 Returns: type value from ucp_type_table, or -1 for an invalid type
1034 */
1035
1036 static int
1037 get_ucp(const pcre_uchar **ptrptr, BOOL *negptr, int *dptr, int *errorcodeptr)
1038 {
1039 int c, i, bot, top;
1040 const pcre_uchar *ptr = *ptrptr;
1041 char name[32];
1042
1043 c = *(++ptr);
1044 if (c == 0) goto ERROR_RETURN;
1045
1046 *negptr = FALSE;
1047
1048 /* \P or \p can be followed by a name in {}, optionally preceded by ^ for
1049 negation. */
1050
1051 if (c == CHAR_LEFT_CURLY_BRACKET)
1052 {
1053 if (ptr[1] == CHAR_CIRCUMFLEX_ACCENT)
1054 {
1055 *negptr = TRUE;
1056 ptr++;
1057 }
1058 for (i = 0; i < (int)sizeof(name) - 1; i++)
1059 {
1060 c = *(++ptr);
1061 if (c == 0) goto ERROR_RETURN;
1062 if (c == CHAR_RIGHT_CURLY_BRACKET) break;
1063 name[i] = c;
1064 }
1065 if (c != CHAR_RIGHT_CURLY_BRACKET) goto ERROR_RETURN;
1066 name[i] = 0;
1067 }
1068
1069 /* Otherwise there is just one following character */
1070
1071 else
1072 {
1073 name[0] = c;
1074 name[1] = 0;
1075 }
1076
1077 *ptrptr = ptr;
1078
1079 /* Search for a recognized property name using binary chop */
1080
1081 bot = 0;
1082 top = _pcre_utt_size;
1083
1084 while (bot < top)
1085 {
1086 i = (bot + top) >> 1;
1087 c = strcmp(name, _pcre_utt_names + _pcre_utt[i].name_offset);
1088 if (c == 0)
1089 {
1090 *dptr = _pcre_utt[i].value;
1091 return _pcre_utt[i].type;
1092 }
1093 if (c > 0) bot = i + 1; else top = i;
1094 }
1095
1096 *errorcodeptr = ERR47;
1097 *ptrptr = ptr;
1098 return -1;
1099
1100 ERROR_RETURN:
1101 *errorcodeptr = ERR46;
1102 *ptrptr = ptr;
1103 return -1;
1104 }
1105 #endif
1106
1107
1108
1109
1110 /*************************************************
1111 * Read repeat counts *
1112 *************************************************/
1113
1114 /* Read an item of the form {n,m} and return the values. This is called only
1115 after is_counted_repeat() has confirmed that a repeat-count quantifier exists,
1116 so the syntax is guaranteed to be correct, but we need to check the values.
1117
1118 Arguments:
1119 p pointer to first char after '{'
1120 minp pointer to int for min
1121 maxp pointer to int for max
1122 returned as -1 if no max
1123 errorcodeptr points to error code variable
1124
1125 Returns: pointer to '}' on success;
1126 current ptr on error, with errorcodeptr set non-zero
1127 */
1128
1129 static const pcre_uchar *
1130 read_repeat_counts(const pcre_uchar *p, int *minp, int *maxp, int *errorcodeptr)
1131 {
1132 int min = 0;
1133 int max = -1;
1134
1135 /* Read the minimum value and do a paranoid check: a negative value indicates
1136 an integer overflow. */
1137
1138 while ((digitab[*p] & ctype_digit) != 0) min = min * 10 + *p++ - CHAR_0;
1139 if (min < 0 || min > 65535)
1140 {
1141 *errorcodeptr = ERR5;
1142 return p;
1143 }
1144
1145 /* Read the maximum value if there is one, and again do a paranoid on its size.
1146 Also, max must not be less than min. */
1147
1148 if (*p == CHAR_RIGHT_CURLY_BRACKET) max = min; else
1149 {
1150 if (*(++p) != CHAR_RIGHT_CURLY_BRACKET)
1151 {
1152 max = 0;
1153 while((digitab[*p] & ctype_digit) != 0) max = max * 10 + *p++ - CHAR_0;
1154 if (max < 0 || max > 65535)
1155 {
1156 *errorcodeptr = ERR5;
1157 return p;
1158 }
1159 if (max < min)
1160 {
1161 *errorcodeptr = ERR4;
1162 return p;
1163 }
1164 }
1165 }
1166
1167 /* Fill in the required variables, and pass back the pointer to the terminating
1168 '}'. */
1169
1170 *minp = min;
1171 *maxp = max;
1172 return p;
1173 }
1174
1175
1176
1177 /*************************************************
1178 * Subroutine for finding forward reference *
1179 *************************************************/
1180
1181 /* This recursive function is called only from find_parens() below. The
1182 top-level call starts at the beginning of the pattern. All other calls must
1183 start at a parenthesis. It scans along a pattern's text looking for capturing
1184 subpatterns, and counting them. If it finds a named pattern that matches the
1185 name it is given, it returns its number. Alternatively, if the name is NULL, it
1186 returns when it reaches a given numbered subpattern. Recursion is used to keep
1187 track of subpatterns that reset the capturing group numbers - the (?| feature.
1188
1189 This function was originally called only from the second pass, in which we know
1190 that if (?< or (?' or (?P< is encountered, the name will be correctly
1191 terminated because that is checked in the first pass. There is now one call to
1192 this function in the first pass, to check for a recursive back reference by
1193 name (so that we can make the whole group atomic). In this case, we need check
1194 only up to the current position in the pattern, and that is still OK because
1195 and previous occurrences will have been checked. To make this work, the test
1196 for "end of pattern" is a check against cd->end_pattern in the main loop,
1197 instead of looking for a binary zero. This means that the special first-pass
1198 call can adjust cd->end_pattern temporarily. (Checks for binary zero while
1199 processing items within the loop are OK, because afterwards the main loop will
1200 terminate.)
1201
1202 Arguments:
1203 ptrptr address of the current character pointer (updated)
1204 cd compile background data
1205 name name to seek, or NULL if seeking a numbered subpattern
1206 lorn name length, or subpattern number if name is NULL
1207 xmode TRUE if we are in /x mode
1208 utf8 TRUE if we are in UTF-8 mode
1209 count pointer to the current capturing subpattern number (updated)
1210
1211 Returns: the number of the named subpattern, or -1 if not found
1212 */
1213
1214 static int
1215 find_parens_sub(pcre_uchar **ptrptr, compile_data *cd, const pcre_uchar *name, int lorn,
1216 BOOL xmode, BOOL utf8, int *count)
1217 {
1218 pcre_uchar *ptr = *ptrptr;
1219 int start_count = *count;
1220 int hwm_count = start_count;
1221 BOOL dup_parens = FALSE;
1222
1223 /* If the first character is a parenthesis, check on the type of group we are
1224 dealing with. The very first call may not start with a parenthesis. */
1225
1226 if (ptr[0] == CHAR_LEFT_PARENTHESIS)
1227 {
1228 /* Handle specials such as (*SKIP) or (*UTF8) etc. */
1229
1230 if (ptr[1] == CHAR_ASTERISK) ptr += 2;
1231
1232 /* Handle a normal, unnamed capturing parenthesis. */
1233
1234 else if (ptr[1] != CHAR_QUESTION_MARK)
1235 {
1236 *count += 1;
1237 if (name == NULL && *count == lorn) return *count;
1238 ptr++;
1239 }
1240
1241 /* All cases now have (? at the start. Remember when we are in a group
1242 where the parenthesis numbers are duplicated. */
1243
1244 else if (ptr[2] == CHAR_VERTICAL_LINE)
1245 {
1246 ptr += 3;
1247 dup_parens = TRUE;
1248 }
1249
1250 /* Handle comments; all characters are allowed until a ket is reached. */
1251
1252 else if (ptr[2] == CHAR_NUMBER_SIGN)
1253 {
1254 for (ptr += 3; *ptr != 0; ptr++) if (*ptr == CHAR_RIGHT_PARENTHESIS) break;
1255 goto FAIL_EXIT;
1256 }
1257
1258 /* Handle a condition. If it is an assertion, just carry on so that it
1259 is processed as normal. If not, skip to the closing parenthesis of the
1260 condition (there can't be any nested parens). */
1261
1262 else if (ptr[2] == CHAR_LEFT_PARENTHESIS)
1263 {
1264 ptr += 2;
1265 if (ptr[1] != CHAR_QUESTION_MARK)
1266 {
1267 while (*ptr != 0 && *ptr != CHAR_RIGHT_PARENTHESIS) ptr++;
1268 if (*ptr != 0) ptr++;
1269 }
1270 }
1271
1272 /* Start with (? but not a condition. */
1273
1274 else
1275 {
1276 ptr += 2;
1277 if (*ptr == CHAR_P) ptr++; /* Allow optional P */
1278
1279 /* We have to disambiguate (?<! and (?<= from (?<name> for named groups */
1280
1281 if ((*ptr == CHAR_LESS_THAN_SIGN && ptr[1] != CHAR_EXCLAMATION_MARK &&
1282 ptr[1] != CHAR_EQUALS_SIGN) || *ptr == CHAR_APOSTROPHE)
1283 {
1284 int term;
1285 const pcre_uchar *thisname;
1286 *count += 1;
1287 if (name == NULL && *count == lorn) return *count;
1288 term = *ptr++;
1289 if (term == CHAR_LESS_THAN_SIGN) term = CHAR_GREATER_THAN_SIGN;
1290 thisname = ptr;
1291 while (*ptr != term) ptr++;
1292 if (name != NULL && lorn == ptr - thisname &&
1293 strncmp((const char *)name, (const char *)thisname, lorn) == 0)
1294 return *count;
1295 term++;
1296 }
1297 }
1298 }
1299
1300 /* Past any initial parenthesis handling, scan for parentheses or vertical
1301 bars. Stop if we get to cd->end_pattern. Note that this is important for the
1302 first-pass call when this value is temporarily adjusted to stop at the current
1303 position. So DO NOT change this to a test for binary zero. */
1304
1305 for (; ptr < cd->end_pattern; ptr++)
1306 {
1307 /* Skip over backslashed characters and also entire \Q...\E */
1308
1309 if (*ptr == CHAR_BACKSLASH)
1310 {
1311 if (*(++ptr) == 0) goto FAIL_EXIT;
1312 if (*ptr == CHAR_Q) for (;;)
1313 {
1314 while (*(++ptr) != 0 && *ptr != CHAR_BACKSLASH) {};
1315 if (*ptr == 0) goto FAIL_EXIT;
1316 if (*(++ptr) == CHAR_E) break;
1317 }
1318 continue;
1319 }
1320
1321 /* Skip over character classes; this logic must be similar to the way they
1322 are handled for real. If the first character is '^', skip it. Also, if the
1323 first few characters (either before or after ^) are \Q\E or \E we skip them
1324 too. This makes for compatibility with Perl. Note the use of STR macros to
1325 encode "Q\\E" so that it works in UTF-8 on EBCDIC platforms. */
1326
1327 if (*ptr == CHAR_LEFT_SQUARE_BRACKET)
1328 {
1329 BOOL negate_class = FALSE;
1330 for (;;)
1331 {
1332 if (ptr[1] == CHAR_BACKSLASH)
1333 {
1334 if (ptr[2] == CHAR_E)
1335 ptr+= 2;
1336 else if (strncmp((const char *)ptr+2,
1337 STR_Q STR_BACKSLASH STR_E, 3) == 0)
1338 ptr += 4;
1339 else
1340 break;
1341 }
1342 else if (!negate_class && ptr[1] == CHAR_CIRCUMFLEX_ACCENT)
1343 {
1344 negate_class = TRUE;
1345 ptr++;
1346 }
1347 else break;
1348 }
1349
1350 /* If the next character is ']', it is a data character that must be
1351 skipped, except in JavaScript compatibility mode. */
1352
1353 if (ptr[1] == CHAR_RIGHT_SQUARE_BRACKET &&
1354 (cd->external_options & PCRE_JAVASCRIPT_COMPAT) == 0)
1355 ptr++;
1356
1357 while (*(++ptr) != CHAR_RIGHT_SQUARE_BRACKET)
1358 {
1359 if (*ptr == 0) return -1;
1360 if (*ptr == CHAR_BACKSLASH)
1361 {
1362 if (*(++ptr) == 0) goto FAIL_EXIT;
1363 if (*ptr == CHAR_Q) for (;;)
1364 {
1365 while (*(++ptr) != 0 && *ptr != CHAR_BACKSLASH) {};
1366 if (*ptr == 0) goto FAIL_EXIT;
1367 if (*(++ptr) == CHAR_E) break;
1368 }
1369 continue;
1370 }
1371 }
1372 continue;
1373 }
1374
1375 /* Skip comments in /x mode */
1376
1377 if (xmode && *ptr == CHAR_NUMBER_SIGN)
1378 {
1379 ptr++;
1380 while (*ptr != 0)
1381 {
1382 if (IS_NEWLINE(ptr)) { ptr += cd->nllen - 1; break; }
1383 ptr++;
1384 #ifdef SUPPORT_UTF8
1385 if (utf8) while ((*ptr & 0xc0) == 0x80) ptr++;
1386 #endif
1387 }
1388 if (*ptr == 0) goto FAIL_EXIT;
1389 continue;
1390 }
1391
1392 /* Check for the special metacharacters */
1393
1394 if (*ptr == CHAR_LEFT_PARENTHESIS)
1395 {
1396 int rc = find_parens_sub(&ptr, cd, name, lorn, xmode, utf8, count);
1397 if (rc > 0) return rc;
1398 if (*ptr == 0) goto FAIL_EXIT;
1399 }
1400
1401 else if (*ptr == CHAR_RIGHT_PARENTHESIS)
1402 {
1403 if (dup_parens && *count < hwm_count) *count = hwm_count;
1404 goto FAIL_EXIT;
1405 }
1406
1407 else if (*ptr == CHAR_VERTICAL_LINE && dup_parens)
1408 {
1409 if (*count > hwm_count) hwm_count = *count;
1410 *count = start_count;
1411 }
1412 }
1413
1414 FAIL_EXIT:
1415 *ptrptr = ptr;
1416 return -1;
1417 }
1418
1419
1420
1421
1422 /*************************************************
1423 * Find forward referenced subpattern *
1424 *************************************************/
1425
1426 /* This function scans along a pattern's text looking for capturing
1427 subpatterns, and counting them. If it finds a named pattern that matches the
1428 name it is given, it returns its number. Alternatively, if the name is NULL, it
1429 returns when it reaches a given numbered subpattern. This is used for forward
1430 references to subpatterns. We used to be able to start this scan from the
1431 current compiling point, using the current count value from cd->bracount, and
1432 do it all in a single loop, but the addition of the possibility of duplicate
1433 subpattern numbers means that we have to scan from the very start, in order to
1434 take account of such duplicates, and to use a recursive function to keep track
1435 of the different types of group.
1436
1437 Arguments:
1438 cd compile background data
1439 name name to seek, or NULL if seeking a numbered subpattern
1440 lorn name length, or subpattern number if name is NULL
1441 xmode TRUE if we are in /x mode
1442 utf8 TRUE if we are in UTF-8 mode
1443
1444 Returns: the number of the found subpattern, or -1 if not found
1445 */
1446
1447 static int
1448 find_parens(compile_data *cd, const pcre_uchar *name, int lorn, BOOL xmode,
1449 BOOL utf8)
1450 {
1451 pcre_uchar *ptr = (pcre_uchar *)cd->start_pattern;
1452 int count = 0;
1453 int rc;
1454
1455 /* If the pattern does not start with an opening parenthesis, the first call
1456 to find_parens_sub() will scan right to the end (if necessary). However, if it
1457 does start with a parenthesis, find_parens_sub() will return when it hits the
1458 matching closing parens. That is why we have to have a loop. */
1459
1460 for (;;)
1461 {
1462 rc = find_parens_sub(&ptr, cd, name, lorn, xmode, utf8, &count);
1463 if (rc > 0 || *ptr++ == 0) break;
1464 }
1465
1466 return rc;
1467 }
1468
1469
1470
1471
1472 /*************************************************
1473 * Find first significant op code *
1474 *************************************************/
1475
1476 /* This is called by several functions that scan a compiled expression looking
1477 for a fixed first character, or an anchoring op code etc. It skips over things
1478 that do not influence this. For some calls, it makes sense to skip negative
1479 forward and all backward assertions, and also the \b assertion; for others it
1480 does not.
1481
1482 Arguments:
1483 code pointer to the start of the group
1484 skipassert TRUE if certain assertions are to be skipped
1485
1486 Returns: pointer to the first significant opcode
1487 */
1488
1489 static const pcre_uchar*
1490 first_significant_code(const pcre_uchar *code, BOOL skipassert)
1491 {
1492 for (;;)
1493 {
1494 switch ((int)*code)
1495 {
1496 case OP_ASSERT_NOT:
1497 case OP_ASSERTBACK:
1498 case OP_ASSERTBACK_NOT:
1499 if (!skipassert) return code;
1500 do code += GET(code, 1); while (*code == OP_ALT);
1501 code += _pcre_OP_lengths[*code];
1502 break;
1503
1504 case OP_WORD_BOUNDARY:
1505 case OP_NOT_WORD_BOUNDARY:
1506 if (!skipassert) return code;
1507 /* Fall through */
1508
1509 case OP_CALLOUT:
1510 case OP_CREF:
1511 case OP_NCREF:
1512 case OP_RREF:
1513 case OP_NRREF:
1514 case OP_DEF:
1515 code += _pcre_OP_lengths[*code];
1516 break;
1517
1518 default:
1519 return code;
1520 }
1521 }
1522 /* Control never reaches here */
1523 }
1524
1525
1526
1527
1528 /*************************************************
1529 * Find the fixed length of a branch *
1530 *************************************************/
1531
1532 /* Scan a branch and compute the fixed length of subject that will match it,
1533 if the length is fixed. This is needed for dealing with backward assertions.
1534 In UTF8 mode, the result is in characters rather than bytes. The branch is
1535 temporarily terminated with OP_END when this function is called.
1536
1537 This function is called when a backward assertion is encountered, so that if it
1538 fails, the error message can point to the correct place in the pattern.
1539 However, we cannot do this when the assertion contains subroutine calls,
1540 because they can be forward references. We solve this by remembering this case
1541 and doing the check at the end; a flag specifies which mode we are running in.
1542
1543 Arguments:
1544 code points to the start of the pattern (the bracket)
1545 utf8 TRUE in UTF-8 mode
1546 atend TRUE if called when the pattern is complete
1547 cd the "compile data" structure
1548
1549 Returns: the fixed length,
1550 or -1 if there is no fixed length,
1551 or -2 if \C was encountered (in UTF-8 mode only)
1552 or -3 if an OP_RECURSE item was encountered and atend is FALSE
1553 or -4 if an unknown opcode was encountered (internal error)
1554 */
1555
1556 static int
1557 find_fixedlength(pcre_uchar *code, BOOL utf8, BOOL atend, compile_data *cd)
1558 {
1559 int length = -1;
1560
1561 register int branchlength = 0;
1562 register pcre_uchar *cc = code + 1 + LINK_SIZE;
1563
1564 /* Scan along the opcodes for this branch. If we get to the end of the
1565 branch, check the length against that of the other branches. */
1566
1567 for (;;)
1568 {
1569 int d;
1570 pcre_uchar *ce, *cs;
1571 register int op = *cc;
1572 switch (op)
1573 {
1574 /* We only need to continue for OP_CBRA (normal capturing bracket) and
1575 OP_BRA (normal non-capturing bracket) because the other variants of these
1576 opcodes are all concerned with unlimited repeated groups, which of course
1577 are not of fixed length. */
1578
1579 case OP_CBRA:
1580 case OP_BRA:
1581 case OP_ONCE:
1582 case OP_ONCE_NC:
1583 case OP_COND:
1584 d = find_fixedlength(cc + ((op == OP_CBRA)? 2:0), utf8, atend, cd);
1585 if (d < 0) return d;
1586 branchlength += d;
1587 do cc += GET(cc, 1); while (*cc == OP_ALT);
1588 cc += 1 + LINK_SIZE;
1589 break;
1590
1591 /* Reached end of a branch; if it's a ket it is the end of a nested call.
1592 If it's ALT it is an alternation in a nested call. An ACCEPT is effectively
1593 an ALT. If it is END it's the end of the outer call. All can be handled by
1594 the same code. Note that we must not include the OP_KETRxxx opcodes here,
1595 because they all imply an unlimited repeat. */
1596
1597 case OP_ALT:
1598 case OP_KET:
1599 case OP_END:
1600 case OP_ACCEPT:
1601 case OP_ASSERT_ACCEPT:
1602 if (length < 0) length = branchlength;
1603 else if (length != branchlength) return -1;
1604 if (*cc != OP_ALT) return length;
1605 cc += 1 + LINK_SIZE;
1606 branchlength = 0;
1607 break;
1608
1609 /* A true recursion implies not fixed length, but a subroutine call may
1610 be OK. If the subroutine is a forward reference, we can't deal with
1611 it until the end of the pattern, so return -3. */
1612
1613 case OP_RECURSE:
1614 if (!atend) return -3;
1615 cs = ce = (pcre_uchar *)cd->start_code + GET(cc, 1); /* Start subpattern */
1616 do ce += GET(ce, 1); while (*ce == OP_ALT); /* End subpattern */
1617 if (cc > cs && cc < ce) return -1; /* Recursion */
1618 d = find_fixedlength(cs + 2, utf8, atend, cd);
1619 if (d < 0) return d;
1620 branchlength += d;
1621 cc += 1 + LINK_SIZE;
1622 break;
1623
1624 /* Skip over assertive subpatterns */
1625
1626 case OP_ASSERT:
1627 case OP_ASSERT_NOT:
1628 case OP_ASSERTBACK:
1629 case OP_ASSERTBACK_NOT:
1630 do cc += GET(cc, 1); while (*cc == OP_ALT);
1631 /* Fall through */
1632
1633 /* Skip over things that don't match chars */
1634
1635 case OP_MARK:
1636 case OP_PRUNE_ARG:
1637 case OP_SKIP_ARG:
1638 case OP_THEN_ARG:
1639 cc += cc[1] + _pcre_OP_lengths[*cc];
1640 break;
1641
1642 case OP_CALLOUT:
1643 case OP_CIRC:
1644 case OP_CIRCM:
1645 case OP_CLOSE:
1646 case OP_COMMIT:
1647 case OP_CREF:
1648 case OP_DEF:
1649 case OP_DOLL:
1650 case OP_DOLLM:
1651 case OP_EOD:
1652 case OP_EODN:
1653 case OP_FAIL:
1654 case OP_NCREF:
1655 case OP_NRREF:
1656 case OP_NOT_WORD_BOUNDARY:
1657 case OP_PRUNE:
1658 case OP_REVERSE:
1659 case OP_RREF:
1660 case OP_SET_SOM:
1661 case OP_SKIP:
1662 case OP_SOD:
1663 case OP_SOM:
1664 case OP_THEN:
1665 case OP_WORD_BOUNDARY:
1666 cc += _pcre_OP_lengths[*cc];
1667 break;
1668
1669 /* Handle literal characters */
1670
1671 case OP_CHAR:
1672 case OP_CHARI:
1673 case OP_NOT:
1674 case OP_NOTI:
1675 branchlength++;
1676 cc += 2;
1677 #ifdef SUPPORT_UTF8
1678 if (utf8 && cc[-1] >= 0xc0) cc += _pcre_utf8_table4[cc[-1] & 0x3f];
1679 #endif
1680 break;
1681
1682 /* Handle exact repetitions. The count is already in characters, but we
1683 need to skip over a multibyte character in UTF8 mode. */
1684
1685 case OP_EXACT:
1686 case OP_EXACTI:
1687 case OP_NOTEXACT:
1688 case OP_NOTEXACTI:
1689 branchlength += GET2(cc,1);
1690 cc += 4;
1691 #ifdef SUPPORT_UTF8
1692 if (utf8 && cc[-1] >= 0xc0) cc += _pcre_utf8_table4[cc[-1] & 0x3f];
1693 #endif
1694 break;
1695
1696 case OP_TYPEEXACT:
1697 branchlength += GET2(cc,1);
1698 if (cc[3] == OP_PROP || cc[3] == OP_NOTPROP) cc += 2;
1699 cc += 4;
1700 break;
1701
1702 /* Handle single-char matchers */
1703
1704 case OP_PROP:
1705 case OP_NOTPROP:
1706 cc += 2;
1707 /* Fall through */
1708
1709 case OP_HSPACE:
1710 case OP_VSPACE:
1711 case OP_NOT_HSPACE:
1712 case OP_NOT_VSPACE:
1713 case OP_NOT_DIGIT:
1714 case OP_DIGIT:
1715 case OP_NOT_WHITESPACE:
1716 case OP_WHITESPACE:
1717 case OP_NOT_WORDCHAR:
1718 case OP_WORDCHAR:
1719 case OP_ANY:
1720 case OP_ALLANY:
1721 branchlength++;
1722 cc++;
1723 break;
1724
1725 /* The single-byte matcher isn't allowed. This only happens in UTF-8 mode;
1726 otherwise \C is coded as OP_ALLANY. */
1727
1728 case OP_ANYBYTE:
1729 return -2;
1730
1731 /* Check a class for variable quantification */
1732
1733 #ifdef SUPPORT_UTF8
1734 case OP_XCLASS:
1735 cc += GET(cc, 1) - 33;
1736 /* Fall through */
1737 #endif
1738
1739 case OP_CLASS:
1740 case OP_NCLASS:
1741 cc += 33;
1742
1743 switch (*cc)
1744 {
1745 case OP_CRPLUS:
1746 case OP_CRMINPLUS:
1747 case OP_CRSTAR:
1748 case OP_CRMINSTAR:
1749 case OP_CRQUERY:
1750 case OP_CRMINQUERY:
1751 return -1;
1752
1753 case OP_CRRANGE:
1754 case OP_CRMINRANGE:
1755 if (GET2(cc,1) != GET2(cc,3)) return -1;
1756 branchlength += GET2(cc,1);
1757 cc += 5;
1758 break;
1759
1760 default:
1761 branchlength++;
1762 }
1763 break;
1764
1765 /* Anything else is variable length */
1766
1767 case OP_ANYNL:
1768 case OP_BRAMINZERO:
1769 case OP_BRAPOS:
1770 case OP_BRAPOSZERO:
1771 case OP_BRAZERO:
1772 case OP_CBRAPOS:
1773 case OP_EXTUNI:
1774 case OP_KETRMAX:
1775 case OP_KETRMIN:
1776 case OP_KETRPOS:
1777 case OP_MINPLUS:
1778 case OP_MINPLUSI:
1779 case OP_MINQUERY:
1780 case OP_MINQUERYI:
1781 case OP_MINSTAR:
1782 case OP_MINSTARI:
1783 case OP_MINUPTO:
1784 case OP_MINUPTOI:
1785 case OP_NOTMINPLUS:
1786 case OP_NOTMINPLUSI:
1787 case OP_NOTMINQUERY:
1788 case OP_NOTMINQUERYI:
1789 case OP_NOTMINSTAR:
1790 case OP_NOTMINSTARI:
1791 case OP_NOTMINUPTO:
1792 case OP_NOTMINUPTOI:
1793 case OP_NOTPLUS:
1794 case OP_NOTPLUSI:
1795 case OP_NOTPOSPLUS:
1796 case OP_NOTPOSPLUSI:
1797 case OP_NOTPOSQUERY:
1798 case OP_NOTPOSQUERYI:
1799 case OP_NOTPOSSTAR:
1800 case OP_NOTPOSSTARI:
1801 case OP_NOTPOSUPTO:
1802 case OP_NOTPOSUPTOI:
1803 case OP_NOTQUERY:
1804 case OP_NOTQUERYI:
1805 case OP_NOTSTAR:
1806 case OP_NOTSTARI:
1807 case OP_NOTUPTO:
1808 case OP_NOTUPTOI:
1809 case OP_PLUS:
1810 case OP_PLUSI:
1811 case OP_POSPLUS:
1812 case OP_POSPLUSI:
1813 case OP_POSQUERY:
1814 case OP_POSQUERYI:
1815 case OP_POSSTAR:
1816 case OP_POSSTARI:
1817 case OP_POSUPTO:
1818 case OP_POSUPTOI:
1819 case OP_QUERY:
1820 case OP_QUERYI:
1821 case OP_REF:
1822 case OP_REFI:
1823 case OP_SBRA:
1824 case OP_SBRAPOS:
1825 case OP_SCBRA:
1826 case OP_SCBRAPOS:
1827 case OP_SCOND:
1828 case OP_SKIPZERO:
1829 case OP_STAR:
1830 case OP_STARI:
1831 case OP_TYPEMINPLUS:
1832 case OP_TYPEMINQUERY:
1833 case OP_TYPEMINSTAR:
1834 case OP_TYPEMINUPTO:
1835 case OP_TYPEPLUS:
1836 case OP_TYPEPOSPLUS:
1837 case OP_TYPEPOSQUERY:
1838 case OP_TYPEPOSSTAR:
1839 case OP_TYPEPOSUPTO:
1840 case OP_TYPEQUERY:
1841 case OP_TYPESTAR:
1842 case OP_TYPEUPTO:
1843 case OP_UPTO:
1844 case OP_UPTOI:
1845 return -1;
1846
1847 /* Catch unrecognized opcodes so that when new ones are added they
1848 are not forgotten, as has happened in the past. */
1849
1850 default:
1851 return -4;
1852 }
1853 }
1854 /* Control never gets here */
1855 }
1856
1857
1858
1859
1860 /*************************************************
1861 * Scan compiled regex for specific bracket *
1862 *************************************************/
1863
1864 /* This little function scans through a compiled pattern until it finds a
1865 capturing bracket with the given number, or, if the number is negative, an
1866 instance of OP_REVERSE for a lookbehind. The function is global in the C sense
1867 so that it can be called from pcre_study() when finding the minimum matching
1868 length.
1869
1870 Arguments:
1871 code points to start of expression
1872 utf8 TRUE in UTF-8 mode
1873 number the required bracket number or negative to find a lookbehind
1874
1875 Returns: pointer to the opcode for the bracket, or NULL if not found
1876 */
1877
1878 const pcre_uchar *
1879 _pcre_find_bracket(const pcre_uchar *code, BOOL utf8, int number)
1880 {
1881 for (;;)
1882 {
1883 register int c = *code;
1884
1885 if (c == OP_END) return NULL;
1886
1887 /* XCLASS is used for classes that cannot be represented just by a bit
1888 map. This includes negated single high-valued characters. The length in
1889 the table is zero; the actual length is stored in the compiled code. */
1890
1891 if (c == OP_XCLASS) code += GET(code, 1);
1892
1893 /* Handle recursion */
1894
1895 else if (c == OP_REVERSE)
1896 {
1897 if (number < 0) return (pcre_uchar *)code;
1898 code += _pcre_OP_lengths[c];
1899 }
1900
1901 /* Handle capturing bracket */
1902
1903 else if (c == OP_CBRA || c == OP_SCBRA ||
1904 c == OP_CBRAPOS || c == OP_SCBRAPOS)
1905 {
1906 int n = GET2(code, 1+LINK_SIZE);
1907 if (n == number) return (pcre_uchar *)code;
1908 code += _pcre_OP_lengths[c];
1909 }
1910
1911 /* Otherwise, we can get the item's length from the table, except that for
1912 repeated character types, we have to test for \p and \P, which have an extra
1913 two bytes of parameters, and for MARK/PRUNE/SKIP/THEN with an argument, we
1914 must add in its length. */
1915
1916 else
1917 {
1918 switch(c)
1919 {
1920 case OP_TYPESTAR:
1921 case OP_TYPEMINSTAR:
1922 case OP_TYPEPLUS:
1923 case OP_TYPEMINPLUS:
1924 case OP_TYPEQUERY:
1925 case OP_TYPEMINQUERY:
1926 case OP_TYPEPOSSTAR:
1927 case OP_TYPEPOSPLUS:
1928 case OP_TYPEPOSQUERY:
1929 if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
1930 break;
1931
1932 case OP_TYPEUPTO:
1933 case OP_TYPEMINUPTO:
1934 case OP_TYPEEXACT:
1935 case OP_TYPEPOSUPTO:
1936 if (code[3] == OP_PROP || code[3] == OP_NOTPROP) code += 2;
1937 break;
1938
1939 case OP_MARK:
1940 case OP_PRUNE_ARG:
1941 case OP_SKIP_ARG:
1942 code += code[1];
1943 break;
1944
1945 case OP_THEN_ARG:
1946 code += code[1];
1947 break;
1948 }
1949
1950 /* Add in the fixed length from the table */
1951
1952 code += _pcre_OP_lengths[c];
1953
1954 /* In UTF-8 mode, opcodes that are followed by a character may be followed by
1955 a multi-byte character. The length in the table is a minimum, so we have to
1956 arrange to skip the extra bytes. */
1957
1958 #ifdef SUPPORT_UTF8
1959 if (utf8) switch(c)
1960 {
1961 case OP_CHAR:
1962 case OP_CHARI:
1963 case OP_EXACT:
1964 case OP_EXACTI:
1965 case OP_UPTO:
1966 case OP_UPTOI:
1967 case OP_MINUPTO:
1968 case OP_MINUPTOI:
1969 case OP_POSUPTO:
1970 case OP_POSUPTOI:
1971 case OP_STAR:
1972 case OP_STARI:
1973 case OP_MINSTAR:
1974 case OP_MINSTARI:
1975 case OP_POSSTAR:
1976 case OP_POSSTARI:
1977 case OP_PLUS:
1978 case OP_PLUSI:
1979 case OP_MINPLUS:
1980 case OP_MINPLUSI:
1981 case OP_POSPLUS:
1982 case OP_POSPLUSI:
1983 case OP_QUERY:
1984 case OP_QUERYI:
1985 case OP_MINQUERY:
1986 case OP_MINQUERYI:
1987 case OP_POSQUERY:
1988 case OP_POSQUERYI:
1989 if (code[-1] >= 0xc0) code += _pcre_utf8_table4[code[-1] & 0x3f];
1990 break;
1991 }
1992 #else
1993 (void)(utf8); /* Keep compiler happy by referencing function argument */
1994 #endif
1995 }
1996 }
1997 }
1998
1999
2000
2001 /*************************************************
2002 * Scan compiled regex for recursion reference *
2003 *************************************************/
2004
2005 /* This little function scans through a compiled pattern until it finds an
2006 instance of OP_RECURSE.
2007
2008 Arguments:
2009 code points to start of expression
2010 utf8 TRUE in UTF-8 mode
2011
2012 Returns: pointer to the opcode for OP_RECURSE, or NULL if not found
2013 */
2014
2015 static const pcre_uchar *
2016 find_recurse(const pcre_uchar *code, BOOL utf8)
2017 {
2018 for (;;)
2019 {
2020 register int c = *code;
2021 if (c == OP_END) return NULL;
2022 if (c == OP_RECURSE) return code;
2023
2024 /* XCLASS is used for classes that cannot be represented just by a bit
2025 map. This includes negated single high-valued characters. The length in
2026 the table is zero; the actual length is stored in the compiled code. */
2027
2028 if (c == OP_XCLASS) code += GET(code, 1);
2029
2030 /* Otherwise, we can get the item's length from the table, except that for
2031 repeated character types, we have to test for \p and \P, which have an extra
2032 two bytes of parameters, and for MARK/PRUNE/SKIP/THEN with an argument, we
2033 must add in its length. */
2034
2035 else
2036 {
2037 switch(c)
2038 {
2039 case OP_TYPESTAR:
2040 case OP_TYPEMINSTAR:
2041 case OP_TYPEPLUS:
2042 case OP_TYPEMINPLUS:
2043 case OP_TYPEQUERY:
2044 case OP_TYPEMINQUERY:
2045 case OP_TYPEPOSSTAR:
2046 case OP_TYPEPOSPLUS:
2047 case OP_TYPEPOSQUERY:
2048 if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
2049 break;
2050
2051 case OP_TYPEPOSUPTO:
2052 case OP_TYPEUPTO:
2053 case OP_TYPEMINUPTO:
2054 case OP_TYPEEXACT:
2055 if (code[3] == OP_PROP || code[3] == OP_NOTPROP) code += 2;
2056 break;
2057
2058 case OP_MARK:
2059 case OP_PRUNE_ARG:
2060 case OP_SKIP_ARG:
2061 code += code[1];
2062 break;
2063
2064 case OP_THEN_ARG:
2065 code += code[1];
2066 break;
2067 }
2068
2069 /* Add in the fixed length from the table */
2070
2071 code += _pcre_OP_lengths[c];
2072
2073 /* In UTF-8 mode, opcodes that are followed by a character may be followed
2074 by a multi-byte character. The length in the table is a minimum, so we have
2075 to arrange to skip the extra bytes. */
2076
2077 #ifdef SUPPORT_UTF8
2078 if (utf8) switch(c)
2079 {
2080 case OP_CHAR:
2081 case OP_CHARI:
2082 case OP_EXACT:
2083 case OP_EXACTI:
2084 case OP_UPTO:
2085 case OP_UPTOI:
2086 case OP_MINUPTO:
2087 case OP_MINUPTOI:
2088 case OP_POSUPTO:
2089 case OP_POSUPTOI:
2090 case OP_STAR:
2091 case OP_STARI:
2092 case OP_MINSTAR:
2093 case OP_MINSTARI:
2094 case OP_POSSTAR:
2095 case OP_POSSTARI:
2096 case OP_PLUS:
2097 case OP_PLUSI:
2098 case OP_MINPLUS:
2099 case OP_MINPLUSI:
2100 case OP_POSPLUS:
2101 case OP_POSPLUSI:
2102 case OP_QUERY:
2103 case OP_QUERYI:
2104 case OP_MINQUERY:
2105 case OP_MINQUERYI:
2106 case OP_POSQUERY:
2107 case OP_POSQUERYI:
2108 if (code[-1] >= 0xc0) code += _pcre_utf8_table4[code[-1] & 0x3f];
2109 break;
2110 }
2111 #else
2112 (void)(utf8); /* Keep compiler happy by referencing function argument */
2113 #endif
2114 }
2115 }
2116 }
2117
2118
2119
2120 /*************************************************
2121 * Scan compiled branch for non-emptiness *
2122 *************************************************/
2123
2124 /* This function scans through a branch of a compiled pattern to see whether it
2125 can match the empty string or not. It is called from could_be_empty()
2126 below and from compile_branch() when checking for an unlimited repeat of a
2127 group that can match nothing. Note that first_significant_code() skips over
2128 backward and negative forward assertions when its final argument is TRUE. If we
2129 hit an unclosed bracket, we return "empty" - this means we've struck an inner
2130 bracket whose current branch will already have been scanned.
2131
2132 Arguments:
2133 code points to start of search
2134 endcode points to where to stop
2135 utf8 TRUE if in UTF8 mode
2136 cd contains pointers to tables etc.
2137
2138 Returns: TRUE if what is matched could be empty
2139 */
2140
2141 static BOOL
2142 could_be_empty_branch(const pcre_uchar *code, const pcre_uchar *endcode,
2143 BOOL utf8, compile_data *cd)
2144 {
2145 register int c;
2146 for (code = first_significant_code(code + _pcre_OP_lengths[*code], TRUE);
2147 code < endcode;
2148 code = first_significant_code(code + _pcre_OP_lengths[c], TRUE))
2149 {
2150 const pcre_uchar *ccode;
2151
2152 c = *code;
2153
2154 /* Skip over forward assertions; the other assertions are skipped by
2155 first_significant_code() with a TRUE final argument. */
2156
2157 if (c == OP_ASSERT)
2158 {
2159 do code += GET(code, 1); while (*code == OP_ALT);
2160 c = *code;
2161 continue;
2162 }
2163
2164 /* For a recursion/subroutine call, if its end has been reached, which
2165 implies a backward reference subroutine call, we can scan it. If it's a
2166 forward reference subroutine call, we can't. To detect forward reference
2167 we have to scan up the list that is kept in the workspace. This function is
2168 called only when doing the real compile, not during the pre-compile that
2169 measures the size of the compiled pattern. */
2170
2171 if (c == OP_RECURSE)
2172 {
2173 const pcre_uchar *scode;
2174 BOOL empty_branch;
2175
2176 /* Test for forward reference */
2177
2178 for (scode = cd->start_workspace; scode < cd->hwm; scode += LINK_SIZE)
2179 if (GET(scode, 0) == code + 1 - cd->start_code) return TRUE;
2180
2181 /* Not a forward reference, test for completed backward reference */
2182
2183 empty_branch = FALSE;
2184 scode = cd->start_code + GET(code, 1);
2185 if (GET(scode, 1) == 0) return TRUE; /* Unclosed */
2186
2187 /* Completed backwards reference */
2188
2189 do
2190 {
2191 if (could_be_empty_branch(scode, endcode, utf8, cd))
2192 {
2193 empty_branch = TRUE;
2194 break;
2195 }
2196 scode += GET(scode, 1);
2197 }
2198 while (*scode == OP_ALT);
2199
2200 if (!empty_branch) return FALSE; /* All branches are non-empty */
2201 continue;
2202 }
2203
2204 /* Groups with zero repeats can of course be empty; skip them. */
2205
2206 if (c == OP_BRAZERO || c == OP_BRAMINZERO || c == OP_SKIPZERO ||
2207 c == OP_BRAPOSZERO)
2208 {
2209 code += _pcre_OP_lengths[c];
2210 do code += GET(code, 1); while (*code == OP_ALT);
2211 c = *code;
2212 continue;
2213 }
2214
2215 /* A nested group that is already marked as "could be empty" can just be
2216 skipped. */
2217
2218 if (c == OP_SBRA || c == OP_SBRAPOS ||
2219 c == OP_SCBRA || c == OP_SCBRAPOS)
2220 {
2221 do code += GET(code, 1); while (*code == OP_ALT);
2222 c = *code;
2223 continue;
2224 }
2225
2226 /* For other groups, scan the branches. */
2227
2228 if (c == OP_BRA || c == OP_BRAPOS ||
2229 c == OP_CBRA || c == OP_CBRAPOS ||
2230 c == OP_ONCE || c == OP_ONCE_NC ||
2231 c == OP_COND)
2232 {
2233 BOOL empty_branch;
2234 if (GET(code, 1) == 0) return TRUE; /* Hit unclosed bracket */
2235
2236 /* If a conditional group has only one branch, there is a second, implied,
2237 empty branch, so just skip over the conditional, because it could be empty.
2238 Otherwise, scan the individual branches of the group. */
2239
2240 if (c == OP_COND && code[GET(code, 1)] != OP_ALT)
2241 code += GET(code, 1);
2242 else
2243 {
2244 empty_branch = FALSE;
2245 do
2246 {
2247 if (!empty_branch && could_be_empty_branch(code, endcode, utf8, cd))
2248 empty_branch = TRUE;
2249 code += GET(code, 1);
2250 }
2251 while (*code == OP_ALT);
2252 if (!empty_branch) return FALSE; /* All branches are non-empty */
2253 }
2254
2255 c = *code;
2256 continue;
2257 }
2258
2259 /* Handle the other opcodes */
2260
2261 switch (c)
2262 {
2263 /* Check for quantifiers after a class. XCLASS is used for classes that
2264 cannot be represented just by a bit map. This includes negated single
2265 high-valued characters. The length in _pcre_OP_lengths[] is zero; the
2266 actual length is stored in the compiled code, so we must update "code"
2267 here. */
2268
2269 #ifdef SUPPORT_UTF8
2270 case OP_XCLASS:
2271 ccode = code += GET(code, 1);
2272 goto CHECK_CLASS_REPEAT;
2273 #endif
2274
2275 case OP_CLASS:
2276 case OP_NCLASS:
2277 ccode = code + 33;
2278
2279 #ifdef SUPPORT_UTF8
2280 CHECK_CLASS_REPEAT:
2281 #endif
2282
2283 switch (*ccode)
2284 {
2285 case OP_CRSTAR: /* These could be empty; continue */
2286 case OP_CRMINSTAR:
2287 case OP_CRQUERY:
2288 case OP_CRMINQUERY:
2289 break;
2290
2291 default: /* Non-repeat => class must match */
2292 case OP_CRPLUS: /* These repeats aren't empty */
2293 case OP_CRMINPLUS:
2294 return FALSE;
2295
2296 case OP_CRRANGE:
2297 case OP_CRMINRANGE:
2298 if (GET2(ccode, 1) > 0) return FALSE; /* Minimum > 0 */
2299 break;
2300 }
2301 break;
2302
2303 /* Opcodes that must match a character */
2304
2305 case OP_PROP:
2306 case OP_NOTPROP:
2307 case OP_EXTUNI:
2308 case OP_NOT_DIGIT:
2309 case OP_DIGIT:
2310 case OP_NOT_WHITESPACE:
2311 case OP_WHITESPACE:
2312 case OP_NOT_WORDCHAR:
2313 case OP_WORDCHAR:
2314 case OP_ANY:
2315 case OP_ALLANY:
2316 case OP_ANYBYTE:
2317 case OP_CHAR:
2318 case OP_CHARI:
2319 case OP_NOT:
2320 case OP_NOTI:
2321 case OP_PLUS:
2322 case OP_MINPLUS:
2323 case OP_POSPLUS:
2324 case OP_EXACT:
2325 case OP_NOTPLUS:
2326 case OP_NOTMINPLUS:
2327 case OP_NOTPOSPLUS:
2328 case OP_NOTEXACT:
2329 case OP_TYPEPLUS:
2330 case OP_TYPEMINPLUS:
2331 case OP_TYPEPOSPLUS:
2332 case OP_TYPEEXACT:
2333 return FALSE;
2334
2335 /* These are going to continue, as they may be empty, but we have to
2336 fudge the length for the \p and \P cases. */
2337
2338 case OP_TYPESTAR:
2339 case OP_TYPEMINSTAR:
2340 case OP_TYPEPOSSTAR:
2341 case OP_TYPEQUERY:
2342 case OP_TYPEMINQUERY:
2343 case OP_TYPEPOSQUERY:
2344 if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
2345 break;
2346
2347 /* Same for these */
2348
2349 case OP_TYPEUPTO:
2350 case OP_TYPEMINUPTO:
2351 case OP_TYPEPOSUPTO:
2352 if (code[3] == OP_PROP || code[3] == OP_NOTPROP) code += 2;
2353 break;
2354
2355 /* End of branch */
2356
2357 case OP_KET:
2358 case OP_KETRMAX:
2359 case OP_KETRMIN:
2360 case OP_KETRPOS:
2361 case OP_ALT:
2362 return TRUE;
2363
2364 /* In UTF-8 mode, STAR, MINSTAR, POSSTAR, QUERY, MINQUERY, POSQUERY, UPTO,
2365 MINUPTO, and POSUPTO may be followed by a multibyte character */
2366
2367 #ifdef SUPPORT_UTF8
2368 case OP_STAR:
2369 case OP_STARI:
2370 case OP_MINSTAR:
2371 case OP_MINSTARI:
2372 case OP_POSSTAR:
2373 case OP_POSSTARI:
2374 case OP_QUERY:
2375 case OP_QUERYI:
2376 case OP_MINQUERY:
2377 case OP_MINQUERYI:
2378 case OP_POSQUERY:
2379 case OP_POSQUERYI:
2380 if (utf8 && code[1] >= 0xc0) code += _pcre_utf8_table4[code[1] & 0x3f];
2381 break;
2382
2383 case OP_UPTO:
2384 case OP_UPTOI:
2385 case OP_MINUPTO:
2386 case OP_MINUPTOI:
2387 case OP_POSUPTO:
2388 case OP_POSUPTOI:
2389 if (utf8 && code[3] >= 0xc0) code += _pcre_utf8_table4[code[3] & 0x3f];
2390 break;
2391 #endif
2392
2393 /* MARK, and PRUNE/SKIP/THEN with an argument must skip over the argument
2394 string. */
2395
2396 case OP_MARK:
2397 case OP_PRUNE_ARG:
2398 case OP_SKIP_ARG:
2399 code += code[1];
2400 break;
2401
2402 case OP_THEN_ARG:
2403 code += code[1];
2404 break;
2405
2406 /* None of the remaining opcodes are required to match a character. */
2407
2408 default:
2409 break;
2410 }
2411 }
2412
2413 return TRUE;
2414 }
2415
2416
2417
2418 /*************************************************
2419 * Scan compiled regex for non-emptiness *
2420 *************************************************/
2421
2422 /* This function is called to check for left recursive calls. We want to check
2423 the current branch of the current pattern to see if it could match the empty
2424 string. If it could, we must look outwards for branches at other levels,
2425 stopping when we pass beyond the bracket which is the subject of the recursion.
2426 This function is called only during the real compile, not during the
2427 pre-compile.
2428
2429 Arguments:
2430 code points to start of the recursion
2431 endcode points to where to stop (current RECURSE item)
2432 bcptr points to the chain of current (unclosed) branch starts
2433 utf8 TRUE if in UTF-8 mode
2434 cd pointers to tables etc
2435
2436 Returns: TRUE if what is matched could be empty
2437 */
2438
2439 static BOOL
2440 could_be_empty(const pcre_uchar *code, const pcre_uchar *endcode,
2441 branch_chain *bcptr, BOOL utf8, compile_data *cd)
2442 {
2443 while (bcptr != NULL && bcptr->current_branch >= code)
2444 {
2445 if (!could_be_empty_branch(bcptr->current_branch, endcode, utf8, cd))
2446 return FALSE;
2447 bcptr = bcptr->outer;
2448 }
2449 return TRUE;
2450 }
2451
2452
2453
2454 /*************************************************
2455 * Check for POSIX class syntax *
2456 *************************************************/
2457
2458 /* This function is called when the sequence "[:" or "[." or "[=" is
2459 encountered in a character class. It checks whether this is followed by a
2460 sequence of characters terminated by a matching ":]" or ".]" or "=]". If we
2461 reach an unescaped ']' without the special preceding character, return FALSE.
2462
2463 Originally, this function only recognized a sequence of letters between the
2464 terminators, but it seems that Perl recognizes any sequence of characters,
2465 though of course unknown POSIX names are subsequently rejected. Perl gives an
2466 "Unknown POSIX class" error for [:f\oo:] for example, where previously PCRE
2467 didn't consider this to be a POSIX class. Likewise for [:1234:].
2468
2469 The problem in trying to be exactly like Perl is in the handling of escapes. We
2470 have to be sure that [abc[:x\]pqr] is *not* treated as containing a POSIX
2471 class, but [abc[:x\]pqr:]] is (so that an error can be generated). The code
2472 below handles the special case of \], but does not try to do any other escape
2473 processing. This makes it different from Perl for cases such as [:l\ower:]
2474 where Perl recognizes it as the POSIX class "lower" but PCRE does not recognize
2475 "l\ower". This is a lesser evil that not diagnosing bad classes when Perl does,
2476 I think.
2477
2478 A user pointed out that PCRE was rejecting [:a[:digit:]] whereas Perl was not.
2479 It seems that the appearance of a nested POSIX class supersedes an apparent
2480 external class. For example, [:a[:digit:]b:] matches "a", "b", ":", or
2481 a digit.
2482
2483 In Perl, unescaped square brackets may also appear as part of class names. For
2484 example, [:a[:abc]b:] gives unknown POSIX class "[:abc]b:]". However, for
2485 [:a[:abc]b][b:] it gives unknown POSIX class "[:abc]b][b:]", which does not
2486 seem right at all. PCRE does not allow closing square brackets in POSIX class
2487 names.
2488
2489 Arguments:
2490 ptr pointer to the initial [
2491 endptr where to return the end pointer
2492
2493 Returns: TRUE or FALSE
2494 */
2495
2496 static BOOL
2497 check_posix_syntax(const pcre_uchar *ptr, const pcre_uchar **endptr)
2498 {
2499 int terminator; /* Don't combine these lines; the Solaris cc */
2500 terminator = *(++ptr); /* compiler warns about "non-constant" initializer. */
2501 for (++ptr; *ptr != 0; ptr++)
2502 {
2503 if (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET)
2504 ptr++;
2505 else if (*ptr == CHAR_RIGHT_SQUARE_BRACKET) return FALSE;
2506 else
2507 {
2508 if (*ptr == terminator && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET)
2509 {
2510 *endptr = ptr;
2511 return TRUE;
2512 }
2513 if (*ptr == CHAR_LEFT_SQUARE_BRACKET &&
2514 (ptr[1] == CHAR_COLON || ptr[1] == CHAR_DOT ||
2515 ptr[1] == CHAR_EQUALS_SIGN) &&
2516 check_posix_syntax(ptr, endptr))
2517 return FALSE;
2518 }
2519 }
2520 return FALSE;
2521 }
2522
2523
2524
2525
2526 /*************************************************
2527 * Check POSIX class name *
2528 *************************************************/
2529
2530 /* This function is called to check the name given in a POSIX-style class entry
2531 such as [:alnum:].
2532
2533 Arguments:
2534 ptr points to the first letter
2535 len the length of the name
2536
2537 Returns: a value representing the name, or -1 if unknown
2538 */
2539
2540 static int
2541 check_posix_name(const pcre_uchar *ptr, int len)
2542 {
2543 const char *pn = posix_names;
2544 register int yield = 0;
2545 while (posix_name_lengths[yield] != 0)
2546 {
2547 if (len == posix_name_lengths[yield] &&
2548 strncmp((const char *)ptr, pn, len) == 0) return yield;
2549 pn += posix_name_lengths[yield] + 1;
2550 yield++;
2551 }
2552 return -1;
2553 }
2554
2555
2556 /*************************************************
2557 * Adjust OP_RECURSE items in repeated group *
2558 *************************************************/
2559
2560 /* OP_RECURSE items contain an offset from the start of the regex to the group
2561 that is referenced. This means that groups can be replicated for fixed
2562 repetition simply by copying (because the recursion is allowed to refer to
2563 earlier groups that are outside the current group). However, when a group is
2564 optional (i.e. the minimum quantifier is zero), OP_BRAZERO or OP_SKIPZERO is
2565 inserted before it, after it has been compiled. This means that any OP_RECURSE
2566 items within it that refer to the group itself or any contained groups have to
2567 have their offsets adjusted. That one of the jobs of this function. Before it
2568 is called, the partially compiled regex must be temporarily terminated with
2569 OP_END.
2570
2571 This function has been extended with the possibility of forward references for
2572 recursions and subroutine calls. It must also check the list of such references
2573 for the group we are dealing with. If it finds that one of the recursions in
2574 the current group is on this list, it adjusts the offset in the list, not the
2575 value in the reference (which is a group number).
2576
2577 Arguments:
2578 group points to the start of the group
2579 adjust the amount by which the group is to be moved
2580 utf8 TRUE in UTF-8 mode
2581 cd contains pointers to tables etc.
2582 save_hwm the hwm forward reference pointer at the start of the group
2583
2584 Returns: nothing
2585 */
2586
2587 static void
2588 adjust_recurse(pcre_uchar *group, int adjust, BOOL utf8, compile_data *cd,
2589 pcre_uchar *save_hwm)
2590 {
2591 pcre_uchar *ptr = group;
2592
2593 while ((ptr = (pcre_uchar *)find_recurse(ptr, utf8)) != NULL)
2594 {
2595 int offset;
2596 pcre_uchar *hc;
2597
2598 /* See if this recursion is on the forward reference list. If so, adjust the
2599 reference. */
2600
2601 for (hc = save_hwm; hc < cd->hwm; hc += LINK_SIZE)
2602 {
2603 offset = GET(hc, 0);
2604 if (cd->start_code + offset == ptr + 1)
2605 {
2606 PUT(hc, 0, offset + adjust);
2607 break;
2608 }
2609 }
2610
2611 /* Otherwise, adjust the recursion offset if it's after the start of this
2612 group. */
2613
2614 if (hc >= cd->hwm)
2615 {
2616 offset = GET(ptr, 1);
2617 if (cd->start_code + offset >= group) PUT(ptr, 1, offset + adjust);
2618 }
2619
2620 ptr += 1 + LINK_SIZE;
2621 }
2622 }
2623
2624
2625
2626 /*************************************************
2627 * Insert an automatic callout point *
2628 *************************************************/
2629
2630 /* This function is called when the PCRE_AUTO_CALLOUT option is set, to insert
2631 callout points before each pattern item.
2632
2633 Arguments:
2634 code current code pointer
2635 ptr current pattern pointer
2636 cd pointers to tables etc
2637
2638 Returns: new code pointer
2639 */
2640
2641 static pcre_uchar *
2642 auto_callout(pcre_uchar *code, const pcre_uchar *ptr, compile_data *cd)
2643 {
2644 *code++ = OP_CALLOUT;
2645 *code++ = 255;
2646 PUT(code, 0, (int)(ptr - cd->start_pattern)); /* Pattern offset */
2647 PUT(code, LINK_SIZE, 0); /* Default length */
2648 return code + 2 * LINK_SIZE;
2649 }
2650
2651
2652
2653 /*************************************************
2654 * Complete a callout item *
2655 *************************************************/
2656
2657 /* A callout item contains the length of the next item in the pattern, which
2658 we can't fill in till after we have reached the relevant point. This is used
2659 for both automatic and manual callouts.
2660
2661 Arguments:
2662 previous_callout points to previous callout item
2663 ptr current pattern pointer
2664 cd pointers to tables etc
2665
2666 Returns: nothing
2667 */
2668
2669 static void
2670 complete_callout(pcre_uchar *previous_callout, const pcre_uchar *ptr, compile_data *cd)
2671 {
2672 int length = (int)(ptr - cd->start_pattern - GET(previous_callout, 2));
2673 PUT(previous_callout, 2 + LINK_SIZE, length);
2674 }
2675
2676
2677
2678 #ifdef SUPPORT_UCP
2679 /*************************************************
2680 * Get othercase range *
2681 *************************************************/
2682
2683 /* This function is passed the start and end of a class range, in UTF-8 mode
2684 with UCP support. It searches up the characters, looking for internal ranges of
2685 characters in the "other" case. Each call returns the next one, updating the
2686 start address.
2687
2688 Arguments:
2689 cptr points to starting character value; updated
2690 d end value
2691 ocptr where to put start of othercase range
2692 odptr where to put end of othercase range
2693
2694 Yield: TRUE when range returned; FALSE when no more
2695 */
2696
2697 static BOOL
2698 get_othercase_range(unsigned int *cptr, unsigned int d, unsigned int *ocptr,
2699 unsigned int *odptr)
2700 {
2701 unsigned int c, othercase, next;
2702
2703 for (c = *cptr; c <= d; c++)
2704 { if ((othercase = UCD_OTHERCASE(c)) != c) break; }
2705
2706 if (c > d) return FALSE;
2707
2708 *ocptr = othercase;
2709 next = othercase + 1;
2710
2711 for (++c; c <= d; c++)
2712 {
2713 if (UCD_OTHERCASE(c) != next) break;
2714 next++;
2715 }
2716
2717 *odptr = next - 1;
2718 *cptr = c;
2719
2720 return TRUE;
2721 }
2722
2723
2724
2725 /*************************************************
2726 * Check a character and a property *
2727 *************************************************/
2728
2729 /* This function is called by check_auto_possessive() when a property item
2730 is adjacent to a fixed character.
2731
2732 Arguments:
2733 c the character
2734 ptype the property type
2735 pdata the data for the type
2736 negated TRUE if it's a negated property (\P or \p{^)
2737
2738 Returns: TRUE if auto-possessifying is OK
2739 */
2740
2741 static BOOL
2742 check_char_prop(int c, int ptype, int pdata, BOOL negated)
2743 {
2744 const ucd_record *prop = GET_UCD(c);
2745 switch(ptype)
2746 {
2747 case PT_LAMP:
2748 return (prop->chartype == ucp_Lu ||
2749 prop->chartype == ucp_Ll ||
2750 prop->chartype == ucp_Lt) == negated;
2751
2752 case PT_GC:
2753 return (pdata == _pcre_ucp_gentype[prop->chartype]) == negated;
2754
2755 case PT_PC:
2756 return (pdata == prop->chartype) == negated;
2757
2758 case PT_SC:
2759 return (pdata == prop->script) == negated;
2760
2761 /* These are specials */
2762
2763 case PT_ALNUM:
2764 return (_pcre_ucp_gentype[prop->chartype] == ucp_L ||
2765 _pcre_ucp_gentype[prop->chartype] == ucp_N) == negated;
2766
2767 case PT_SPACE: /* Perl space */
2768 return (_pcre_ucp_gentype[prop->chartype] == ucp_Z ||
2769 c == CHAR_HT || c == CHAR_NL || c == CHAR_FF || c == CHAR_CR)
2770 == negated;
2771
2772 case PT_PXSPACE: /* POSIX space */
2773 return (_pcre_ucp_gentype[prop->chartype] == ucp_Z ||
2774 c == CHAR_HT || c == CHAR_NL || c == CHAR_VT ||
2775 c == CHAR_FF || c == CHAR_CR)
2776 == negated;
2777
2778 case PT_WORD:
2779 return (_pcre_ucp_gentype[prop->chartype] == ucp_L ||
2780 _pcre_ucp_gentype[prop->chartype] == ucp_N ||
2781 c == CHAR_UNDERSCORE) == negated;
2782 }
2783 return FALSE;
2784 }
2785 #endif /* SUPPORT_UCP */
2786
2787
2788
2789 /*************************************************
2790 * Check if auto-possessifying is possible *
2791 *************************************************/
2792
2793 /* This function is called for unlimited repeats of certain items, to see
2794 whether the next thing could possibly match the repeated item. If not, it makes
2795 sense to automatically possessify the repeated item.
2796
2797 Arguments:
2798 previous pointer to the repeated opcode
2799 utf8 TRUE in UTF-8 mode
2800 ptr next character in pattern
2801 options options bits
2802 cd contains pointers to tables etc.
2803
2804 Returns: TRUE if possessifying is wanted
2805 */
2806
2807 static BOOL
2808 check_auto_possessive(const pcre_uchar *previous, BOOL utf8,
2809 const pcre_uchar *ptr, int options, compile_data *cd)
2810 {
2811 int c, next;
2812 int op_code = *previous++;
2813
2814 /* Skip whitespace and comments in extended mode */
2815
2816 if ((options & PCRE_EXTENDED) != 0)
2817 {
2818 for (;;)
2819 {
2820 while ((cd->ctypes[*ptr] & ctype_space) != 0) ptr++;
2821 if (*ptr == CHAR_NUMBER_SIGN)
2822 {
2823 ptr++;
2824 while (*ptr != 0)
2825 {
2826 if (IS_NEWLINE(ptr)) { ptr += cd->nllen; break; }
2827 ptr++;
2828 #ifdef SUPPORT_UTF8
2829 if (utf8) while ((*ptr & 0xc0) == 0x80) ptr++;
2830 #endif
2831 }
2832 }
2833 else break;
2834 }
2835 }
2836
2837 /* If the next item is one that we can handle, get its value. A non-negative
2838 value is a character, a negative value is an escape value. */
2839
2840 if (*ptr == CHAR_BACKSLASH)
2841 {
2842 int temperrorcode = 0;
2843 next = check_escape(&ptr, &temperrorcode, cd->bracount, options, FALSE);
2844 if (temperrorcode != 0) return FALSE;
2845 ptr++; /* Point after the escape sequence */
2846 }
2847
2848 else if ((cd->ctypes[*ptr] & ctype_meta) == 0)
2849 {
2850 #ifdef SUPPORT_UTF8
2851 if (utf8) { GETCHARINC(next, ptr); } else
2852 #endif
2853 next = *ptr++;
2854 }
2855
2856 else return FALSE;
2857
2858 /* Skip whitespace and comments in extended mode */
2859
2860 if ((options & PCRE_EXTENDED) != 0)
2861 {
2862 for (;;)
2863 {
2864 while ((cd->ctypes[*ptr] & ctype_space) != 0) ptr++;
2865 if (*ptr == CHAR_NUMBER_SIGN)
2866 {
2867 ptr++;
2868 while (*ptr != 0)
2869 {
2870 if (IS_NEWLINE(ptr)) { ptr += cd->nllen; break; }
2871 ptr++;
2872 #ifdef SUPPORT_UTF8
2873 if (utf8) while ((*ptr & 0xc0) == 0x80) ptr++;
2874 #endif
2875 }
2876 }
2877 else break;
2878 }
2879 }
2880
2881 /* If the next thing is itself optional, we have to give up. */
2882
2883 if (*ptr == CHAR_ASTERISK || *ptr == CHAR_QUESTION_MARK ||
2884 strncmp((char *)ptr, STR_LEFT_CURLY_BRACKET STR_0 STR_COMMA, 3) == 0)
2885 return FALSE;
2886
2887 /* Now compare the next item with the previous opcode. First, handle cases when
2888 the next item is a character. */
2889
2890 if (next >= 0) switch(op_code)
2891 {
2892 case OP_CHAR:
2893 #ifdef SUPPORT_UTF8
2894 GETCHARTEST(c, previous);
2895 #else
2896 c = *previous;
2897 #endif
2898 return c != next;
2899
2900 /* For CHARI (caseless character) we must check the other case. If we have
2901 Unicode property support, we can use it to test the other case of
2902 high-valued characters. */
2903
2904 case OP_CHARI:
2905 #ifdef SUPPORT_UTF8
2906 GETCHARTEST(c, previous);
2907 #else
2908 c = *previous;
2909 #endif
2910 if (c == next) return FALSE;
2911 #ifdef SUPPORT_UTF8
2912 if (utf8)
2913 {
2914 unsigned int othercase;
2915 if (next < 128) othercase = cd->fcc[next]; else
2916 #ifdef SUPPORT_UCP
2917 othercase = UCD_OTHERCASE((unsigned int)next);
2918 #else
2919 othercase = NOTACHAR;
2920 #endif
2921 return (unsigned int)c != othercase;
2922 }
2923 else
2924 #endif /* SUPPORT_UTF8 */
2925 return (c != cd->fcc[next]); /* Non-UTF-8 mode */
2926
2927 /* For OP_NOT and OP_NOTI, the data is always a single-byte character. These
2928 opcodes are not used for multi-byte characters, because they are coded using
2929 an XCLASS instead. */
2930
2931 case OP_NOT:
2932 return (c = *previous) == next;
2933
2934 case OP_NOTI:
2935 if ((c = *previous) == next) return TRUE;
2936 #ifdef SUPPORT_UTF8
2937 if (utf8)
2938 {
2939 unsigned int othercase;
2940 if (next < 128) othercase = cd->fcc[next]; else
2941 #ifdef SUPPORT_UCP
2942 othercase = UCD_OTHERCASE(next);
2943 #else
2944 othercase = NOTACHAR;
2945 #endif
2946 return (unsigned int)c == othercase;
2947 }
2948 else
2949 #endif /* SUPPORT_UTF8 */
2950 return (c == cd->fcc[next]); /* Non-UTF-8 mode */
2951
2952 /* Note that OP_DIGIT etc. are generated only when PCRE_UCP is *not* set.
2953 When it is set, \d etc. are converted into OP_(NOT_)PROP codes. */
2954
2955 case OP_DIGIT:
2956 return next > 127 || (cd->ctypes[next] & ctype_digit) == 0;
2957
2958 case OP_NOT_DIGIT:
2959 return next <= 127 && (cd->ctypes[next] & ctype_digit) != 0;
2960
2961 case OP_WHITESPACE:
2962 return next > 127 || (cd->ctypes[next] & ctype_space) == 0;
2963
2964 case OP_NOT_WHITESPACE:
2965 return next <= 127 && (cd->ctypes[next] & ctype_space) != 0;
2966
2967 case OP_WORDCHAR:
2968 return next > 127 || (cd->ctypes[next] & ctype_word) == 0;
2969
2970 case OP_NOT_WORDCHAR:
2971 return next <= 127 && (cd->ctypes[next] & ctype_word) != 0;
2972
2973 case OP_HSPACE:
2974 case OP_NOT_HSPACE:
2975 switch(next)
2976 {
2977 case 0x09:
2978 case 0x20:
2979 case 0xa0:
2980 case 0x1680:
2981 case 0x180e:
2982 case 0x2000:
2983 case 0x2001:
2984 case 0x2002:
2985 case 0x2003:
2986 case 0x2004:
2987 case 0x2005:
2988 case 0x2006:
2989 case 0x2007:
2990 case 0x2008:
2991 case 0x2009:
2992 case 0x200A:
2993 case 0x202f:
2994 case 0x205f:
2995 case 0x3000:
2996 return op_code == OP_NOT_HSPACE;
2997 default:
2998 return op_code != OP_NOT_HSPACE;
2999 }
3000
3001 case OP_ANYNL:
3002 case OP_VSPACE:
3003 case OP_NOT_VSPACE:
3004 switch(next)
3005 {
3006 case 0x0a:
3007 case 0x0b:
3008 case 0x0c:
3009 case 0x0d:
3010 case 0x85:
3011 case 0x2028:
3012 case 0x2029:
3013 return op_code == OP_NOT_VSPACE;
3014 default:
3015 return op_code != OP_NOT_VSPACE;
3016 }
3017
3018 #ifdef SUPPORT_UCP
3019 case OP_PROP:
3020 return check_char_prop(next, previous[0], previous[1], FALSE);
3021
3022 case OP_NOTPROP:
3023 return check_char_prop(next, previous[0], previous[1], TRUE);
3024 #endif
3025
3026 default:
3027 return FALSE;
3028 }
3029
3030
3031 /* Handle the case when the next item is \d, \s, etc. Note that when PCRE_UCP
3032 is set, \d turns into ESC_du rather than ESC_d, etc., so ESC_d etc. are
3033 generated only when PCRE_UCP is *not* set, that is, when only ASCII
3034 characteristics are recognized. Similarly, the opcodes OP_DIGIT etc. are
3035 replaced by OP_PROP codes when PCRE_UCP is set. */
3036
3037 switch(op_code)
3038 {
3039 case OP_CHAR:
3040 case OP_CHARI:
3041 #ifdef SUPPORT_UTF8
3042 GETCHARTEST(c, previous);
3043 #else
3044 c = *previous;
3045 #endif
3046 switch(-next)
3047 {
3048 case ESC_d:
3049 return c > 127 || (cd->ctypes[c] & ctype_digit) == 0;
3050
3051 case ESC_D:
3052 return c <= 127 && (cd->ctypes[c] & ctype_digit) != 0;
3053
3054 case ESC_s:
3055 return c > 127 || (cd->ctypes[c] & ctype_space) == 0;
3056
3057 case ESC_S:
3058 return c <= 127 && (cd->ctypes[c] & ctype_space) != 0;
3059
3060 case ESC_w:
3061 return c > 127 || (cd->ctypes[c] & ctype_word) == 0;
3062
3063 case ESC_W:
3064 return c <= 127 && (cd->ctypes[c] & ctype_word) != 0;
3065
3066 case ESC_h:
3067 case ESC_H:
3068 switch(c)
3069 {
3070 case 0x09:
3071 case 0x20:
3072 case 0xa0:
3073 case 0x1680:
3074 case 0x180e:
3075 case 0x2000:
3076 case 0x2001:
3077 case 0x2002:
3078 case 0x2003:
3079 case 0x2004:
3080 case 0x2005:
3081 case 0x2006:
3082 case 0x2007:
3083 case 0x2008:
3084 case 0x2009:
3085 case 0x200A:
3086 case 0x202f:
3087 case 0x205f:
3088 case 0x3000:
3089 return -next != ESC_h;
3090 default:
3091 return -next == ESC_h;
3092 }
3093
3094 case ESC_v:
3095 case ESC_V:
3096 switch(c)
3097 {
3098 case 0x0a:
3099 case 0x0b:
3100 case 0x0c:
3101 case 0x0d:
3102 case 0x85:
3103 case 0x2028:
3104 case 0x2029:
3105 return -next != ESC_v;
3106 default:
3107 return -next == ESC_v;
3108 }
3109
3110 /* When PCRE_UCP is set, these values get generated for \d etc. Find
3111 their substitutions and process them. The result will always be either
3112 -ESC_p or -ESC_P. Then fall through to process those values. */
3113
3114 #ifdef SUPPORT_UCP
3115 case ESC_du:
3116 case ESC_DU:
3117 case ESC_wu:
3118 case ESC_WU:
3119 case ESC_su:
3120 case ESC_SU:
3121 {
3122 int temperrorcode = 0;
3123 ptr = substitutes[-next - ESC_DU];
3124 next = check_escape(&ptr, &temperrorcode, 0, options, FALSE);
3125 if (temperrorcode != 0) return FALSE;
3126 ptr++; /* For compatibility */
3127 }
3128 /* Fall through */
3129
3130 case ESC_p:
3131 case ESC_P:
3132 {
3133 int ptype, pdata, errorcodeptr;
3134 BOOL negated;
3135
3136 ptr--; /* Make ptr point at the p or P */
3137 ptype = get_ucp(&ptr, &negated, &pdata, &errorcodeptr);
3138 if (ptype < 0) return FALSE;
3139 ptr++; /* Point past the final curly ket */
3140
3141 /* If the property item is optional, we have to give up. (When generated
3142 from \d etc by PCRE_UCP, this test will have been applied much earlier,
3143 to the original \d etc. At this point, ptr will point to a zero byte. */
3144
3145 if (*ptr == CHAR_ASTERISK || *ptr == CHAR_QUESTION_MARK ||
3146 strncmp((char *)ptr, STR_LEFT_CURLY_BRACKET STR_0 STR_COMMA, 3) == 0)
3147 return FALSE;
3148
3149 /* Do the property check. */
3150
3151 return check_char_prop(c, ptype, pdata, (next == -ESC_P) != negated);
3152 }
3153 #endif
3154
3155 default:
3156 return FALSE;
3157 }
3158
3159 /* In principle, support for Unicode properties should be integrated here as
3160 well. It means re-organizing the above code so as to get hold of the property
3161 values before switching on the op-code. However, I wonder how many patterns
3162 combine ASCII \d etc with Unicode properties? (Note that if PCRE_UCP is set,
3163 these op-codes are never generated.) */
3164
3165 case OP_DIGIT:
3166 return next == -ESC_D || next == -ESC_s || next == -ESC_W ||
3167 next == -ESC_h || next == -ESC_v || next == -ESC_R;
3168
3169 case OP_NOT_DIGIT:
3170 return next == -ESC_d;
3171
3172 case OP_WHITESPACE:
3173 return next == -ESC_S || next == -ESC_d || next == -ESC_w || next == -ESC_R;
3174
3175 case OP_NOT_WHITESPACE:
3176 return next == -ESC_s || next == -ESC_h || next == -ESC_v;
3177
3178 case OP_HSPACE:
3179 return next == -ESC_S || next == -ESC_H || next == -ESC_d ||
3180 next == -ESC_w || next == -ESC_v || next == -ESC_R;
3181
3182 case OP_NOT_HSPACE:
3183 return next == -ESC_h;
3184
3185 /* Can't have \S in here because VT matches \S (Perl anomaly) */
3186 case OP_ANYNL:
3187 case OP_VSPACE:
3188 return next == -ESC_V || next == -ESC_d || next == -ESC_w;
3189
3190 case OP_NOT_VSPACE:
3191 return next == -ESC_v || next == -ESC_R;
3192
3193 case OP_WORDCHAR:
3194 return next == -ESC_W || next == -ESC_s || next == -ESC_h ||
3195 next == -ESC_v || next == -ESC_R;
3196
3197 case OP_NOT_WORDCHAR:
3198 return next == -ESC_w || next == -ESC_d;
3199
3200 default:
3201 return FALSE;
3202 }
3203
3204 /* Control does not reach here */
3205 }
3206
3207
3208
3209 /*************************************************
3210 * Compile one branch *
3211 *************************************************/
3212
3213 /* Scan the pattern, compiling it into the a vector. If the options are
3214 changed during the branch, the pointer is used to change the external options
3215 bits. This function is used during the pre-compile phase when we are trying
3216 to find out the amount of memory needed, as well as during the real compile
3217 phase. The value of lengthptr distinguishes the two phases.
3218
3219 Arguments:
3220 optionsptr pointer to the option bits
3221 codeptr points to the pointer to the current code point
3222 ptrptr points to the current pattern pointer
3223 errorcodeptr points to error code variable
3224 firstbyteptr set to initial literal character, or < 0 (REQ_UNSET, REQ_NONE)
3225 reqbyteptr set to the last literal character required, else < 0
3226 bcptr points to current branch chain
3227 cond_depth conditional nesting depth
3228 cd contains pointers to tables etc.
3229 lengthptr NULL during the real compile phase
3230 points to length accumulator during pre-compile phase
3231
3232 Returns: TRUE on success
3233 FALSE, with *errorcodeptr set non-zero on error
3234 */
3235
3236 static BOOL
3237 compile_branch(int *optionsptr, pcre_uchar **codeptr,
3238 const pcre_uchar **ptrptr, int *errorcodeptr, int *firstbyteptr,
3239 int *reqbyteptr, branch_chain *bcptr, int cond_depth, compile_data *cd,
3240 int *lengthptr)
3241 {
3242 int repeat_type, op_type;
3243 int repeat_min = 0, repeat_max = 0; /* To please picky compilers */
3244 int bravalue = 0;
3245 int greedy_default, greedy_non_default;
3246 int firstbyte, reqbyte;
3247 int zeroreqbyte, zerofirstbyte;
3248 int req_caseopt, reqvary, tempreqvary;
3249 int options = *optionsptr; /* May change dynamically */
3250 int after_manual_callout = 0;
3251 int length_prevgroup = 0;
3252 register int c;
3253 register pcre_uchar *code = *codeptr;
3254 pcre_uchar *last_code = code;
3255 pcre_uchar *orig_code = code;
3256 pcre_uchar *tempcode;
3257 BOOL inescq = FALSE;
3258 BOOL groupsetfirstbyte = FALSE;
3259 const pcre_uchar *ptr = *ptrptr;
3260 const pcre_uchar *tempptr;
3261 const pcre_uchar *nestptr = NULL;
3262 pcre_uchar *previous = NULL;
3263 pcre_uchar *previous_callout = NULL;
3264 pcre_uchar *save_hwm = NULL;
3265 pcre_uchar classbits[32];
3266
3267 /* We can fish out the UTF-8 setting once and for all into a BOOL, but we
3268 must not do this for other options (e.g. PCRE_EXTENDED) because they may change
3269 dynamically as we process the pattern. */
3270
3271 #ifdef SUPPORT_UTF8
3272 BOOL class_utf8;
3273 BOOL utf8 = (options & PCRE_UTF8) != 0;
3274 pcre_uint8 *class_utf8data;
3275 pcre_uint8 *class_utf8data_base;
3276 pcre_uint8 utf8_char[6];
3277 #else
3278 BOOL utf8 = FALSE;
3279 #endif
3280
3281 #ifdef PCRE_DEBUG
3282 if (lengthptr != NULL) DPRINTF((">> start branch\n"));
3283 #endif
3284
3285 /* Set up the default and non-default settings for greediness */
3286
3287 greedy_default = ((options & PCRE_UNGREEDY) != 0);
3288 greedy_non_default = greedy_default ^ 1;
3289
3290 /* Initialize no first byte, no required byte. REQ_UNSET means "no char
3291 matching encountered yet". It gets changed to REQ_NONE if we hit something that
3292 matches a non-fixed char first char; reqbyte just remains unset if we never
3293 find one.
3294
3295 When we hit a repeat whose minimum is zero, we may have to adjust these values
3296 to take the zero repeat into account. This is implemented by setting them to
3297 zerofirstbyte and zeroreqbyte when such a repeat is encountered. The individual
3298 item types that can be repeated set these backoff variables appropriately. */
3299
3300 firstbyte = reqbyte = zerofirstbyte = zeroreqbyte = REQ_UNSET;
3301
3302 /* The variable req_caseopt contains either the REQ_CASELESS value or zero,
3303 according to the current setting of the caseless flag. REQ_CASELESS is a bit
3304 value > 255. It is added into the firstbyte or reqbyte variables to record the
3305 case status of the value. This is used only for ASCII characters. */
3306
3307 req_caseopt = ((options & PCRE_CASELESS) != 0)? REQ_CASELESS : 0;
3308
3309 /* Switch on next character until the end of the branch */
3310
3311 for (;; ptr++)
3312 {
3313 BOOL negate_class;
3314 BOOL should_flip_negation;
3315 BOOL possessive_quantifier;
3316 BOOL is_quantifier;
3317 BOOL is_recurse;
3318 BOOL reset_bracount;
3319 int class_charcount;
3320 int class_lastchar;
3321 int newoptions;
3322 int recno;
3323 int refsign;
3324 int skipbytes;
3325 int subreqbyte;
3326 int subfirstbyte;
3327 int terminator;
3328 int mclength;
3329 int tempbracount;
3330 pcre_uchar mcbuffer[8];
3331
3332 /* Get next byte in the pattern */
3333
3334 c = *ptr;
3335
3336 /* If we are at the end of a nested substitution, revert to the outer level
3337 string. Nesting only happens one level deep. */
3338
3339 if (c == 0 && nestptr != NULL)
3340 {
3341 ptr = nestptr;
3342 nestptr = NULL;
3343 c = *ptr;
3344 }
3345
3346 /* If we are in the pre-compile phase, accumulate the length used for the
3347 previous cycle of this loop. */
3348
3349 if (lengthptr != NULL)
3350 {
3351 #ifdef PCRE_DEBUG
3352 if (code > cd->hwm) cd->hwm = code; /* High water info */
3353 #endif
3354 if (code > cd->start_workspace + WORK_SIZE_CHECK) /* Check for overrun */
3355 {
3356 *errorcodeptr = ERR52;
3357 goto FAILED;
3358 }
3359
3360 /* There is at least one situation where code goes backwards: this is the
3361 case of a zero quantifier after a class (e.g. [ab]{0}). At compile time,
3362 the class is simply eliminated. However, it is created first, so we have to
3363 allow memory for it. Therefore, don't ever reduce the length at this point.
3364 */
3365
3366 if (code < last_code) code = last_code;
3367
3368 /* Paranoid check for integer overflow */
3369
3370 if (OFLOW_MAX - *lengthptr < code - last_code)
3371 {
3372 *errorcodeptr = ERR20;
3373 goto FAILED;
3374 }
3375
3376 *lengthptr += (int)(code - last_code);
3377 DPRINTF(("length=%d added %d c=%c\n", *lengthptr, (int)(code - last_code),
3378 c));
3379
3380 /* If "previous" is set and it is not at the start of the work space, move
3381 it back to there, in order to avoid filling up the work space. Otherwise,
3382 if "previous" is NULL, reset the current code pointer to the start. */
3383
3384 if (previous != NULL)
3385 {
3386 if (previous > orig_code)
3387 {
3388 memmove(orig_code, previous, code - previous);
3389 code -= previous - orig_code;
3390 previous = orig_code;
3391 }
3392 }
3393 else code = orig_code;
3394
3395 /* Remember where this code item starts so we can pick up the length
3396 next time round. */
3397
3398 last_code = code;
3399 }
3400
3401 /* In the real compile phase, just check the workspace used by the forward
3402 reference list. */
3403
3404 else if (cd->hwm > cd->start_workspace + WORK_SIZE_CHECK)
3405 {
3406 *errorcodeptr = ERR52;
3407 goto FAILED;
3408 }
3409
3410 /* If in \Q...\E, check for the end; if not, we have a literal */
3411
3412 if (inescq && c != 0)
3413 {
3414 if (c == CHAR_BACKSLASH && ptr[1] == CHAR_E)
3415 {
3416 inescq = FALSE;
3417 ptr++;
3418 continue;
3419 }
3420 else
3421 {
3422 if (previous_callout != NULL)
3423 {
3424 if (lengthptr == NULL) /* Don't attempt in pre-compile phase */
3425 complete_callout(previous_callout, ptr, cd);
3426 previous_callout = NULL;
3427 }
3428 if ((options & PCRE_AUTO_CALLOUT) != 0)
3429 {
3430 previous_callout = code;
3431 code = auto_callout(code, ptr, cd);
3432 }
3433 goto NORMAL_CHAR;
3434 }
3435 }
3436
3437 /* Fill in length of a previous callout, except when the next thing is
3438 a quantifier. */
3439
3440 is_quantifier =
3441 c == CHAR_ASTERISK || c == CHAR_PLUS || c == CHAR_QUESTION_MARK ||
3442 (c == CHAR_LEFT_CURLY_BRACKET && is_counted_repeat(ptr+1));
3443
3444 if (!is_quantifier && previous_callout != NULL &&
3445 after_manual_callout-- <= 0)
3446 {
3447 if (lengthptr == NULL) /* Don't attempt in pre-compile phase */
3448 complete_callout(previous_callout, ptr, cd);
3449 previous_callout = NULL;
3450 }
3451
3452 /* In extended mode, skip white space and comments. */
3453
3454 if ((options & PCRE_EXTENDED) != 0)
3455 {
3456 if ((cd->ctypes[c] & ctype_space) != 0) continue;
3457 if (c == CHAR_NUMBER_SIGN)
3458 {
3459 ptr++;
3460 while (*ptr != 0)
3461 {
3462 if (IS_NEWLINE(ptr)) { ptr += cd->nllen - 1; break; }
3463 ptr++;
3464 #ifdef SUPPORT_UTF8
3465 if (utf8) while ((*ptr & 0xc0) == 0x80) ptr++;
3466 #endif
3467 }
3468 if (*ptr != 0) continue;
3469
3470 /* Else fall through to handle end of string */
3471 c = 0;
3472 }
3473 }
3474
3475 /* No auto callout for quantifiers. */
3476
3477 if ((options & PCRE_AUTO_CALLOUT) != 0 && !is_quantifier)
3478 {
3479 previous_callout = code;
3480 code = auto_callout(code, ptr, cd);
3481 }
3482
3483 switch(c)
3484 {
3485 /* ===================================================================*/
3486 case 0: /* The branch terminates at string end */
3487 case CHAR_VERTICAL_LINE: /* or | or ) */
3488 case CHAR_RIGHT_PARENTHESIS:
3489 *firstbyteptr = firstbyte;
3490 *reqbyteptr = reqbyte;
3491 *codeptr = code;
3492 *ptrptr = ptr;
3493 if (lengthptr != NULL)
3494 {
3495 if (OFLOW_MAX - *lengthptr < code - last_code)
3496 {
3497 *errorcodeptr = ERR20;
3498 goto FAILED;
3499 }
3500 *lengthptr += (int)(code - last_code); /* To include callout length */
3501 DPRINTF((">> end branch\n"));
3502 }
3503 return TRUE;
3504
3505
3506 /* ===================================================================*/
3507 /* Handle single-character metacharacters. In multiline mode, ^ disables
3508 the setting of any following char as a first character. */
3509
3510 case CHAR_CIRCUMFLEX_ACCENT:
3511 previous = NULL;
3512 if ((options & PCRE_MULTILINE) != 0)
3513 {
3514 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
3515 *code++ = OP_CIRCM;
3516 }
3517 else *code++ = OP_CIRC;
3518 break;
3519
3520 case CHAR_DOLLAR_SIGN:
3521 previous = NULL;
3522 *code++ = ((options & PCRE_MULTILINE) != 0)? OP_DOLLM : OP_DOLL;
3523 break;
3524
3525 /* There can never be a first char if '.' is first, whatever happens about
3526 repeats. The value of reqbyte doesn't change either. */
3527
3528 case CHAR_DOT:
3529 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
3530 zerofirstbyte = firstbyte;
3531 zeroreqbyte = reqbyte;
3532 previous = code;
3533 *code++ = ((options & PCRE_DOTALL) != 0)? OP_ALLANY: OP_ANY;
3534 break;
3535
3536
3537 /* ===================================================================*/
3538 /* Character classes. If the included characters are all < 256, we build a
3539 32-byte bitmap of the permitted characters, except in the special case
3540 where there is only one such character. For negated classes, we build the
3541 map as usual, then invert it at the end. However, we use a different opcode
3542 so that data characters > 255 can be handled correctly.
3543
3544 If the class contains characters outside the 0-255 range, a different
3545 opcode is compiled. It may optionally have a bit map for characters < 256,
3546 but those above are are explicitly listed afterwards. A flag byte tells
3547 whether the bitmap is present, and whether this is a negated class or not.
3548
3549 In JavaScript compatibility mode, an isolated ']' causes an error. In
3550 default (Perl) mode, it is treated as a data character. */
3551
3552 case CHAR_RIGHT_SQUARE_BRACKET:
3553 if ((cd->external_options & PCRE_JAVASCRIPT_COMPAT) != 0)
3554 {
3555 *errorcodeptr = ERR64;
3556 goto FAILED;
3557 }
3558 goto NORMAL_CHAR;
3559
3560 case CHAR_LEFT_SQUARE_BRACKET:
3561 previous = code;
3562
3563 /* PCRE supports POSIX class stuff inside a class. Perl gives an error if
3564 they are encountered at the top level, so we'll do that too. */
3565
3566 if ((ptr[1] == CHAR_COLON || ptr[1] == CHAR_DOT ||
3567 ptr[1] == CHAR_EQUALS_SIGN) &&
3568 check_posix_syntax(ptr, &tempptr))
3569 {
3570 *errorcodeptr = (ptr[1] == CHAR_COLON)? ERR13 : ERR31;
3571 goto FAILED;
3572 }
3573
3574 /* If the first character is '^', set the negation flag and skip it. Also,
3575 if the first few characters (either before or after ^) are \Q\E or \E we
3576 skip them too. This makes for compatibility with Perl. */
3577
3578 negate_class = FALSE;
3579 for (;;)
3580 {
3581 c = *(++ptr);
3582 if (c == CHAR_BACKSLASH)
3583 {
3584 if (ptr[1] == CHAR_E)
3585 ptr++;
3586 else if (strncmp((const char *)ptr+1,
3587 STR_Q STR_BACKSLASH STR_E, 3) == 0)
3588 ptr += 3;
3589 else
3590 break;
3591 }
3592 else if (!negate_class && c == CHAR_CIRCUMFLEX_ACCENT)
3593 negate_class = TRUE;
3594 else break;
3595 }
3596
3597 /* Empty classes are allowed in JavaScript compatibility mode. Otherwise,
3598 an initial ']' is taken as a data character -- the code below handles
3599 that. In JS mode, [] must always fail, so generate OP_FAIL, whereas
3600 [^] must match any character, so generate OP_ALLANY. */
3601
3602 if (c == CHAR_RIGHT_SQUARE_BRACKET &&
3603 (cd->external_options & PCRE_JAVASCRIPT_COMPAT) != 0)
3604 {
3605 *code++ = negate_class? OP_ALLANY : OP_FAIL;
3606 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
3607 zerofirstbyte = firstbyte;
3608 break;
3609 }
3610
3611 /* If a class contains a negative special such as \S, we need to flip the
3612 negation flag at the end, so that support for characters > 255 works
3613 correctly (they are all included in the class). */
3614
3615 should_flip_negation = FALSE;
3616
3617 /* Keep a count of chars with values < 256 so that we can optimize the case
3618 of just a single character (as long as it's < 256). However, For higher
3619 valued UTF-8 characters, we don't yet do any optimization. */
3620
3621 class_charcount = 0;
3622 class_lastchar = -1;
3623
3624 /* Initialize the 32-char bit map to all zeros. We build the map in a
3625 temporary bit of memory, in case the class contains only 1 character (less
3626 than 256), because in that case the compiled code doesn't use the bit map.
3627 */
3628
3629 memset(classbits, 0, 32 * sizeof(pcre_uint8));
3630
3631 #ifdef SUPPORT_UTF8
3632 class_utf8 = FALSE; /* No chars >= 256 */
3633 class_utf8data = code + LINK_SIZE + 2; /* For UTF-8 items */
3634 class_utf8data_base = class_utf8data; /* For resetting in pass 1 */
3635 #endif
3636
3637 /* Process characters until ] is reached. By writing this as a "do" it
3638 means that an initial ] is taken as a data character. At the start of the
3639 loop, c contains the first byte of the character. */
3640
3641 if (c != 0) do
3642 {
3643 const pcre_uchar *oldptr;
3644
3645 #ifdef SUPPORT_UTF8
3646 if (utf8 && c > 127)
3647 { /* Braces are required because the */
3648 GETCHARLEN(c, ptr, ptr); /* macro generates multiple statements */
3649 }
3650
3651 /* In the pre-compile phase, accumulate the length of any UTF-8 extra
3652 data and reset the pointer. This is so that very large classes that
3653 contain a zillion UTF-8 characters no longer overwrite the work space
3654 (which is on the stack). */
3655
3656 if (lengthptr != NULL)
3657 {
3658 *lengthptr += class_utf8data - class_utf8data_base;
3659 class_utf8data = class_utf8data_base;
3660 }
3661
3662 #endif
3663
3664 /* Inside \Q...\E everything is literal except \E */
3665
3666 if (inescq)
3667 {
3668 if (c == CHAR_BACKSLASH && ptr[1] == CHAR_E) /* If we are at \E */
3669 {
3670 inescq = FALSE; /* Reset literal state */
3671 ptr++; /* Skip the 'E' */
3672 continue; /* Carry on with next */
3673 }
3674 goto CHECK_RANGE; /* Could be range if \E follows */
3675 }
3676
3677 /* Handle POSIX class names. Perl allows a negation extension of the
3678 form [:^name:]. A square bracket that doesn't match the syntax is
3679 treated as a literal. We also recognize the POSIX constructions
3680 [.ch.] and [=ch=] ("collating elements") and fault them, as Perl
3681 5.6 and 5.8 do. */
3682
3683 if (c == CHAR_LEFT_SQUARE_BRACKET &&
3684 (ptr[1] == CHAR_COLON || ptr[1] == CHAR_DOT ||
3685 ptr[1] == CHAR_EQUALS_SIGN) && check_posix_syntax(ptr, &tempptr))
3686 {
3687 BOOL local_negate = FALSE;
3688 int posix_class, taboffset, tabopt;
3689 register const pcre_uint8 *cbits = cd->cbits;
3690 pcre_uint8 pbits[32];
3691
3692 if (ptr[1] != CHAR_COLON)
3693 {
3694 *errorcodeptr = ERR31;
3695 goto FAILED;
3696 }
3697
3698 ptr += 2;
3699 if (*ptr == CHAR_CIRCUMFLEX_ACCENT)
3700 {
3701 local_negate = TRUE;
3702 should_flip_negation = TRUE; /* Note negative special */
3703 ptr++;
3704 }
3705
3706 posix_class = check_posix_name(ptr, (int)(tempptr - ptr));
3707 if (posix_class < 0)
3708 {
3709 *errorcodeptr = ERR30;
3710 goto FAILED;
3711 }
3712
3713 /* If matching is caseless, upper and lower are converted to
3714 alpha. This relies on the fact that the class table starts with
3715 alpha, lower, upper as the first 3 entries. */
3716
3717 if ((options & PCRE_CASELESS) != 0 && posix_class <= 2)
3718 posix_class = 0;
3719
3720 /* When PCRE_UCP is set, some of the POSIX classes are converted to
3721 different escape sequences that use Unicode properties. */
3722
3723 #ifdef SUPPORT_UCP
3724 if ((options & PCRE_UCP) != 0)
3725 {
3726 int pc = posix_class + ((local_negate)? POSIX_SUBSIZE/2 : 0);
3727 if (posix_substitutes[pc] != NULL)
3728 {
3729 nestptr = tempptr + 1;
3730 ptr = posix_substitutes[pc] - 1;
3731 continue;
3732 }
3733 }
3734 #endif
3735 /* In the non-UCP case, we build the bit map for the POSIX class in a
3736 chunk of local store because we may be adding and subtracting from it,
3737 and we don't want to subtract bits that may be in the main map already.
3738 At the end we or the result into the bit map that is being built. */
3739
3740 posix_class *= 3;
3741
3742 /* Copy in the first table (always present) */
3743
3744 memcpy(pbits, cbits + posix_class_maps[posix_class],
3745 32 * sizeof(pcre_uint8));
3746
3747 /* If there is a second table, add or remove it as required. */
3748
3749 taboffset = posix_class_maps[posix_class + 1];
3750 tabopt = posix_class_maps[posix_class + 2];
3751
3752 if (taboffset >= 0)
3753 {
3754 if (tabopt >= 0)
3755 for (c = 0; c < 32; c++) pbits[c] |= cbits[c + taboffset];
3756 else
3757 for (c = 0; c < 32; c++) pbits[c] &= ~cbits[c + taboffset];
3758 }
3759
3760 /* Not see if we need to remove any special characters. An option
3761 value of 1 removes vertical space and 2 removes underscore. */
3762
3763 if (tabopt < 0) tabopt = -tabopt;
3764 if (tabopt == 1) pbits[1] &= ~0x3c;
3765 else if (tabopt == 2) pbits[11] &= 0x7f;
3766
3767 /* Add the POSIX table or its complement into the main table that is
3768 being built and we are done. */
3769
3770 if (local_negate)
3771 for (c = 0; c < 32; c++) classbits[c] |= ~pbits[c];
3772 else
3773 for (c = 0; c < 32; c++) classbits[c] |= pbits[c];
3774
3775 ptr = tempptr + 1;
3776 class_charcount = 10; /* Set > 1; assumes more than 1 per class */
3777 continue; /* End of POSIX syntax handling */
3778 }
3779
3780 /* Backslash may introduce a single character, or it may introduce one
3781 of the specials, which just set a flag. The sequence \b is a special
3782 case. Inside a class (and only there) it is treated as backspace. We
3783 assume that other escapes have more than one character in them, so set
3784 class_charcount bigger than one. Unrecognized escapes fall through and
3785 are either treated as literal characters (by default), or are faulted if
3786 PCRE_EXTRA is set. */
3787
3788 if (c == CHAR_BACKSLASH)
3789 {
3790 c = check_escape(&ptr, errorcodeptr, cd->bracount, options, TRUE);
3791 if (*errorcodeptr != 0) goto FAILED;
3792
3793 if (-c == ESC_b) c = CHAR_BS; /* \b is backspace in a class */
3794 else if (-c == ESC_Q) /* Handle start of quoted string */
3795 {
3796 if (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
3797 {
3798 ptr += 2; /* avoid empty string */
3799 }
3800 else inescq = TRUE;
3801 continue;
3802 }
3803 else if (-c == ESC_E) continue; /* Ignore orphan \E */
3804
3805 if (c < 0)
3806 {
3807 register const pcre_uint8 *cbits = cd->cbits;
3808 class_charcount += 2; /* Greater than 1 is what matters */
3809
3810 switch (-c)
3811 {
3812 #ifdef SUPPORT_UCP
3813 case ESC_du: /* These are the values given for \d etc */
3814 case ESC_DU: /* when PCRE_UCP is set. We replace the */
3815 case ESC_wu: /* escape sequence with an appropriate \p */
3816 case ESC_WU: /* or \P to test Unicode properties instead */
3817 case ESC_su: /* of the default ASCII testing. */
3818 case ESC_SU:
3819 nestptr = ptr;
3820 ptr = substitutes[-c - ESC_DU] - 1; /* Just before substitute */
3821 class_charcount -= 2; /* Undo! */
3822 continue;
3823 #endif
3824 case ESC_d:
3825 for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_digit];
3826 continue;
3827
3828 case ESC_D:
3829 should_flip_negation = TRUE;
3830 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_digit];
3831 continue;
3832
3833 case ESC_w:
3834 for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_word];
3835 continue;
3836
3837 case ESC_W:
3838 should_flip_negation = TRUE;
3839 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_word];
3840 continue;
3841
3842 /* Perl 5.004 onwards omits VT from \s, but we must preserve it
3843 if it was previously set by something earlier in the character
3844 class. */
3845
3846 case ESC_s:
3847 classbits[0] |= cbits[cbit_space];
3848 classbits[1] |= cbits[cbit_space+1] & ~0x08;
3849 for (c = 2; c < 32; c++) classbits[c] |= cbits[c+cbit_space];
3850 continue;
3851
3852 case ESC_S:
3853 should_flip_negation = TRUE;
3854 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_space];
3855 classbits[1] |= 0x08; /* Perl 5.004 onwards omits VT from \s */
3856 continue;
3857
3858 case ESC_h:
3859 SETBIT(classbits, 0x09); /* VT */
3860 SETBIT(classbits, 0x20); /* SPACE */
3861 SETBIT(classbits, 0xa0); /* NSBP */
3862 #ifdef SUPPORT_UTF8
3863 if (utf8)
3864 {
3865 class_utf8 = TRUE;
3866 *class_utf8data++ = XCL_SINGLE;
3867 class_utf8data += _pcre_ord2utf8(0x1680, class_utf8data);
3868 *class_utf8data++ = XCL_SINGLE;
3869 class_utf8data += _pcre_ord2utf8(0x180e, class_utf8data);
3870 *class_utf8data++ = XCL_RANGE;
3871 class_utf8data += _pcre_ord2utf8(0x2000, class_utf8data);
3872 class_utf8data += _pcre_ord2utf8(0x200A, class_utf8data);
3873 *class_utf8data++ = XCL_SINGLE;
3874 class_utf8data += _pcre_ord2utf8(0x202f, class_utf8data);
3875 *class_utf8data++ = XCL_SINGLE;
3876 class_utf8data += _pcre_ord2utf8(0x205f, class_utf8data);
3877 *class_utf8data++ = XCL_SINGLE;
3878 class_utf8data += _pcre_ord2utf8(0x3000, class_utf8data);
3879 }
3880 #endif
3881 continue;
3882
3883 case ESC_H:
3884 for (c = 0; c < 32; c++)
3885 {
3886 int x = 0xff;
3887 switch (c)
3888 {
3889 case 0x09/8: x ^= 1 << (0x09%8); break;
3890 case 0x20/8: x ^= 1 << (0x20%8); break;
3891 case 0xa0/8: x ^= 1 << (0xa0%8); break;
3892 default: break;
3893 }
3894 classbits[c] |= x;
3895 }
3896
3897 #ifdef SUPPORT_UTF8
3898 if (utf8)
3899 {
3900 class_utf8 = TRUE;
3901 *class_utf8data++ = XCL_RANGE;
3902 class_utf8data += _pcre_ord2utf8(0x0100, class_utf8data);
3903 class_utf8data += _pcre_ord2utf8(0x167f, class_utf8data);
3904 *class_utf8data++ = XCL_RANGE;
3905 class_utf8data += _pcre_ord2utf8(0x1681, class_utf8data);
3906 class_utf8data += _pcre_ord2utf8(0x180d, class_utf8data);
3907 *class_utf8data++ = XCL_RANGE;
3908 class_utf8data += _pcre_ord2utf8(0x180f, class_utf8data);
3909 class_utf8data += _pcre_ord2utf8(0x1fff, class_utf8data);
3910 *class_utf8data++ = XCL_RANGE;
3911 class_utf8data += _pcre_ord2utf8(0x200B, class_utf8data);
3912 class_utf8data += _pcre_ord2utf8(0x202e, class_utf8data);
3913 *class_utf8data++ = XCL_RANGE;
3914 class_utf8data += _pcre_ord2utf8(0x2030, class_utf8data);
3915 class_utf8data += _pcre_ord2utf8(0x205e, class_utf8data);
3916 *class_utf8data++ = XCL_RANGE;
3917 class_utf8data += _pcre_ord2utf8(0x2060, class_utf8data);
3918 class_utf8data += _pcre_ord2utf8(0x2fff, class_utf8data);
3919 *class_utf8data++ = XCL_RANGE;
3920 class_utf8data += _pcre_ord2utf8(0x3001, class_utf8data);
3921 class_utf8data += _pcre_ord2utf8(0x7fffffff, class_utf8data);
3922 }
3923 #endif
3924 continue;
3925
3926 case ESC_v:
3927 SETBIT(classbits, 0x0a); /* LF */
3928 SETBIT(classbits, 0x0b); /* VT */
3929 SETBIT(classbits, 0x0c); /* FF */
3930 SETBIT(classbits, 0x0d); /* CR */
3931 SETBIT(classbits, 0x85); /* NEL */
3932 #ifdef SUPPORT_UTF8
3933 if (utf8)
3934 {
3935 class_utf8 = TRUE;
3936 *class_utf8data++ = XCL_RANGE;
3937 class_utf8data += _pcre_ord2utf8(0x2028, class_utf8data);
3938 class_utf8data += _pcre_ord2utf8(0x2029, class_utf8data);
3939 }
3940 #endif
3941 continue;
3942
3943 case ESC_V:
3944 for (c = 0; c < 32; c++)
3945 {
3946 int x = 0xff;
3947 switch (c)
3948 {
3949 case 0x0a/8: x ^= 1 << (0x0a%8);
3950 x ^= 1 << (0x0b%8);
3951 x ^= 1 << (0x0c%8);
3952 x ^= 1 << (0x0d%8);
3953 break;
3954 case 0x85/8: x ^= 1 << (0x85%8); break;
3955 default: break;
3956 }
3957 classbits[c] |= x;
3958 }
3959
3960 #ifdef SUPPORT_UTF8
3961 if (utf8)
3962 {
3963 class_utf8 = TRUE;
3964 *class_utf8data++ = XCL_RANGE;
3965 class_utf8data += _pcre_ord2utf8(0x0100, class_utf8data);
3966 class_utf8data += _pcre_ord2utf8(0x2027, class_utf8data);
3967 *class_utf8data++ = XCL_RANGE;
3968 class_utf8data += _pcre_ord2utf8(0x2029, class_utf8data);
3969 class_utf8data += _pcre_ord2utf8(0x7fffffff, class_utf8data);
3970 }
3971 #endif
3972 continue;
3973
3974 #ifdef SUPPORT_UCP
3975 case ESC_p:
3976 case ESC_P:
3977 {
3978 BOOL negated;
3979 int pdata;
3980 int ptype = get_ucp(&ptr, &negated, &pdata, errorcodeptr);
3981 if (ptype < 0) goto FAILED;
3982 class_utf8 = TRUE;
3983 *class_utf8data++ = ((-c == ESC_p) != negated)?
3984 XCL_PROP : XCL_NOTPROP;
3985 *class_utf8data++ = ptype;
3986 *class_utf8data++ = pdata;
3987 class_charcount -= 2; /* Not a < 256 character */
3988 continue;
3989 }
3990 #endif
3991 /* Unrecognized escapes are faulted if PCRE is running in its
3992 strict mode. By default, for compatibility with Perl, they are
3993 treated as literals. */
3994
3995 default:
3996 if ((options & PCRE_EXTRA) != 0)
3997 {
3998 *errorcodeptr = ERR7;
3999 goto FAILED;
4000 }
4001 class_charcount -= 2; /* Undo the default count from above */
4002 c = *ptr; /* Get the final character and fall through */
4003 break;
4004 }
4005 }
4006
4007 /* Fall through if we have a single character (c >= 0). This may be
4008 greater than 256 in UTF-8 mode. */
4009
4010 } /* End of backslash handling */
4011
4012 /* A single character may be followed by '-' to form a range. However,
4013 Perl does not permit ']' to be the end of the range. A '-' character
4014 at the end is treated as a literal. Perl ignores orphaned \E sequences
4015 entirely. The code for handling \Q and \E is messy. */
4016
4017 CHECK_RANGE:
4018 while (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
4019 {
4020 inescq = FALSE;
4021 ptr += 2;
4022 }
4023
4024 oldptr = ptr;
4025
4026 /* Remember \r or \n */
4027
4028 if (c == CHAR_CR || c == CHAR_NL) cd->external_flags |= PCRE_HASCRORLF;
4029
4030 /* Check for range */
4031
4032 if (!inescq && ptr[1] == CHAR_MINUS)
4033 {
4034 int d;
4035 ptr += 2;
4036 while (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_E) ptr += 2;
4037
4038 /* If we hit \Q (not followed by \E) at this point, go into escaped
4039 mode. */
4040
4041 while (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_Q)
4042 {
4043 ptr += 2;
4044 if (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_E)
4045 { ptr += 2; continue; }
4046 inescq = TRUE;
4047 break;
4048 }
4049
4050 if (*ptr == 0 || (!inescq && *ptr == CHAR_RIGHT_SQUARE_BRACKET))
4051 {
4052 ptr = oldptr;
4053 goto LONE_SINGLE_CHARACTER;
4054 }
4055
4056 #ifdef SUPPORT_UTF8
4057 if (utf8)
4058 { /* Braces are required because the */
4059 GETCHARLEN(d, ptr, ptr); /* macro generates multiple statements */
4060 }
4061 else
4062 #endif
4063 d = *ptr; /* Not UTF-8 mode */
4064
4065 /* The second part of a range can be a single-character escape, but
4066 not any of the other escapes. Perl 5.6 treats a hyphen as a literal
4067 in such circumstances. */
4068
4069 if (!inescq && d == CHAR_BACKSLASH)
4070 {
4071 d = check_escape(&ptr, errorcodeptr, cd->bracount, options, TRUE);
4072 if (*errorcodeptr != 0) goto FAILED;
4073
4074 /* \b is backspace; any other special means the '-' was literal */
4075
4076 if (d < 0)
4077 {
4078 if (d == -ESC_b) d = CHAR_BS; else
4079 {
4080 ptr = oldptr;
4081 goto LONE_SINGLE_CHARACTER; /* A few lines below */
4082 }
4083 }
4084 }
4085
4086 /* Check that the two values are in the correct order. Optimize
4087 one-character ranges */
4088
4089 if (d < c)
4090 {
4091 *errorcodeptr = ERR8;
4092 goto FAILED;
4093 }
4094
4095 if (d == c) goto LONE_SINGLE_CHARACTER; /* A few lines below */
4096
4097 /* Remember \r or \n */
4098
4099 if (d == CHAR_CR || d == CHAR_NL) cd->external_flags |= PCRE_HASCRORLF;
4100
4101 /* In UTF-8 mode, if the upper limit is > 255, or > 127 for caseless
4102 matching, we have to use an XCLASS with extra data items. Caseless
4103 matching for characters > 127 is available only if UCP support is
4104 available. */
4105
4106 #ifdef SUPPORT_UTF8
4107 if (utf8 && (d > 255 || ((options & PCRE_CASELESS) != 0 && d > 127)))
4108 {
4109 class_utf8 = TRUE;
4110
4111 /* With UCP support, we can find the other case equivalents of
4112 the relevant characters. There may be several ranges. Optimize how
4113 they fit with the basic range. */
4114
4115 #ifdef SUPPORT_UCP
4116 if ((options & PCRE_CASELESS) != 0)
4117 {
4118 unsigned int occ, ocd;
4119 unsigned int cc = c;
4120 unsigned int origd = d;
4121 while (get_othercase_range(&cc, origd, &occ, &ocd))
4122 {
4123 if (occ >= (unsigned int)c &&
4124 ocd <= (unsigned int)d)
4125 continue; /* Skip embedded ranges */
4126
4127 if (occ < (unsigned int)c &&
4128 ocd >= (unsigned int)c - 1) /* Extend the basic range */
4129 { /* if there is overlap, */
4130 c = occ; /* noting that if occ < c */
4131 continue; /* we can't have ocd > d */
4132 } /* because a subrange is */
4133 if (ocd > (unsigned int)d &&
4134 occ <= (unsigned int)d + 1) /* always shorter than */
4135 { /* the basic range. */
4136 d = ocd;
4137 continue;
4138 }
4139
4140 if (occ == ocd)
4141 {
4142 *class_utf8data++ = XCL_SINGLE;
4143 }
4144 else
4145 {
4146 *class_utf8data++ = XCL_RANGE;
4147 class_utf8data += _pcre_ord2utf8(occ, class_utf8data);
4148 }
4149 class_utf8data += _pcre_ord2utf8(ocd, class_utf8data);
4150 }
4151 }
4152 #endif /* SUPPORT_UCP */
4153
4154 /* Now record the original range, possibly modified for UCP caseless
4155 overlapping ranges. */
4156
4157 *class_utf8data++ = XCL_RANGE;
4158 class_utf8data += _pcre_ord2utf8(c, class_utf8data);
4159 class_utf8data += _pcre_ord2utf8(d, class_utf8data);
4160
4161 /* With UCP support, we are done. Without UCP support, there is no
4162 caseless matching for UTF-8 characters > 127; we can use the bit map
4163 for the smaller ones. */
4164
4165 #ifdef SUPPORT_UCP
4166 continue; /* With next character in the class */
4167 #else
4168 if ((options & PCRE_CASELESS) == 0 || c > 127) continue;
4169
4170 /* Adjust upper limit and fall through to set up the map */
4171
4172 d = 127;
4173
4174 #endif /* SUPPORT_UCP */
4175 }
4176 #endif /* SUPPORT_UTF8 */
4177
4178 /* We use the bit map for all cases when not in UTF-8 mode; else
4179 ranges that lie entirely within 0-127 when there is UCP support; else
4180 for partial ranges without UCP support. */
4181
4182 class_charcount += d - c + 1;
4183 class_lastchar = d;
4184
4185 /* We can save a bit of time by skipping this in the pre-compile. */
4186
4187 if (lengthptr == NULL) for (; c <= d; c++)
4188 {
4189 classbits[c/8] |= (1 << (c&7));
4190 if ((options & PCRE_CASELESS) != 0)
4191 {
4192 int uc = cd->fcc[c]; /* flip case */
4193 classbits[uc/8] |= (1 << (uc&7));
4194 }
4195 }
4196
4197 continue; /* Go get the next char in the class */
4198 }
4199
4200 /* Handle a lone single character - we can get here for a normal
4201 non-escape char, or after \ that introduces a single character or for an
4202 apparent range that isn't. */
4203
4204 LONE_SINGLE_CHARACTER:
4205
4206 /* Handle a character that cannot go in the bit map */
4207
4208 #ifdef SUPPORT_UTF8
4209 if (utf8 && (c > 255 || ((options & PCRE_CASELESS) != 0 && c > 127)))
4210 {
4211 class_utf8 = TRUE;
4212 *class_utf8data++ = XCL_SINGLE;
4213 class_utf8data += _pcre_ord2utf8(c, class_utf8data);
4214
4215 #ifdef SUPPORT_UCP
4216 if ((options & PCRE_CASELESS) != 0)
4217 {
4218 unsigned int othercase;
4219 if ((othercase = UCD_OTHERCASE(c)) != c)
4220 {
4221 *class_utf8data++ = XCL_SINGLE;
4222 class_utf8data += _pcre_ord2utf8(othercase, class_utf8data);
4223 }
4224 }
4225 #endif /* SUPPORT_UCP */
4226
4227 }
4228 else
4229 #endif /* SUPPORT_UTF8 */
4230
4231 /* Handle a single-byte character */
4232 {
4233 classbits[c/8] |= (1 << (c&7));
4234 if ((options & PCRE_CASELESS) != 0)
4235 {
4236 c = cd->fcc[c]; /* flip case */
4237 classbits[c/8] |= (1 << (c&7));
4238 }
4239 class_charcount++;
4240 class_lastchar = c;
4241 }
4242 }
4243
4244 /* Loop until ']' reached. This "while" is the end of the "do" far above.
4245 If we are at the end of an internal nested string, revert to the outer
4246 string. */
4247
4248 while (((c = *(++ptr)) != 0 ||
4249 (nestptr != NULL &&
4250 (ptr = nestptr, nestptr = NULL, c = *(++ptr)) != 0)) &&
4251 (c != CHAR_RIGHT_SQUARE_BRACKET || inescq));
4252
4253 /* Check for missing terminating ']' */
4254
4255 if (c == 0)
4256 {
4257 *errorcodeptr = ERR6;
4258 goto FAILED;
4259 }
4260
4261 /* If class_charcount is 1, we saw precisely one character whose value is
4262 less than 256. As long as there were no characters >= 128 and there was no
4263 use of \p or \P, in other words, no use of any XCLASS features, we can
4264 optimize.
4265
4266 In UTF-8 mode, we can optimize the negative case only if there were no
4267 characters >= 128 because OP_NOT and the related opcodes like OP_NOTSTAR
4268 operate on single-bytes characters only. This is an historical hangover.
4269 Maybe one day we can tidy these opcodes to handle multi-byte characters.
4270
4271 The optimization throws away the bit map. We turn the item into a
4272 1-character OP_CHAR[I] if it's positive, or OP_NOT[I] if it's negative.
4273 Note that OP_NOT[I] does not support multibyte characters. In the positive
4274 case, it can cause firstbyte to be set. Otherwise, there can be no first
4275 char if this item is first, whatever repeat count may follow. In the case
4276 of reqbyte, save the previous value for reinstating. */
4277
4278 #ifdef SUPPORT_UTF8
4279 if (class_charcount == 1 && !class_utf8 &&
4280 (!utf8 || !negate_class || class_lastchar < 128))
4281 #else
4282 if (class_charcount == 1)
4283 #endif
4284 {
4285 zeroreqbyte = reqbyte;
4286
4287 /* The OP_NOT[I] opcodes work on one-byte characters only. */
4288
4289 if (negate_class)
4290 {
4291 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
4292 zerofirstbyte = firstbyte;
4293 *code++ = ((options & PCRE_CASELESS) != 0)? OP_NOTI: OP_NOT;
4294 *code++ = class_lastchar;
4295 break;
4296 }
4297
4298 /* For a single, positive character, get the value into mcbuffer, and
4299 then we can handle this with the normal one-character code. */
4300
4301 #ifdef SUPPORT_UTF8
4302 if (utf8 && class_lastchar > 127)
4303 mclength = _pcre_ord2utf8(class_lastchar, mcbuffer);
4304 else
4305 #endif
4306 {
4307 mcbuffer[0] = class_lastchar;
4308 mclength = 1;
4309 }
4310 goto ONE_CHAR;
4311 } /* End of 1-char optimization */
4312
4313 /* The general case - not the one-char optimization. If this is the first
4314 thing in the branch, there can be no first char setting, whatever the
4315 repeat count. Any reqbyte setting must remain unchanged after any kind of
4316 repeat. */
4317
4318 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
4319 zerofirstbyte = firstbyte;
4320 zeroreqbyte = reqbyte;
4321
4322 /* If there are characters with values > 255, we have to compile an
4323 extended class, with its own opcode, unless there was a negated special
4324 such as \S in the class, and PCRE_UCP is not set, because in that case all
4325 characters > 255 are in the class, so any that were explicitly given as
4326 well can be ignored. If (when there are explicit characters > 255 that must
4327 be listed) there are no characters < 256, we can omit the bitmap in the
4328 actual compiled code. */
4329
4330 #ifdef SUPPORT_UTF8
4331 if (class_utf8 && (!should_flip_negation || (options & PCRE_UCP) != 0))
4332 {
4333 *class_utf8data++ = XCL_END; /* Marks the end of extra data */
4334 *code++ = OP_XCLASS;
4335 code += LINK_SIZE;
4336 *code = negate_class? XCL_NOT : 0;
4337
4338 /* If the map is required, move up the extra data to make room for it;
4339 otherwise just move the code pointer to the end of the extra data. */
4340
4341 if (class_charcount > 0)
4342 {
4343 *code++ |= XCL_MAP;
4344 memmove(code + 32, code, class_utf8data - code);
4345 memcpy(code, classbits, 32);
4346 code = class_utf8data + 32;
4347 }
4348 else code = class_utf8data;
4349
4350 /* Now fill in the complete length of the item */
4351
4352 PUT(previous, 1, code - previous);
4353 break; /* End of class handling */
4354 }
4355 #endif
4356
4357 /* If there are no characters > 255, or they are all to be included or
4358 excluded, set the opcode to OP_CLASS or OP_NCLASS, depending on whether the
4359 whole class was negated and whether there were negative specials such as \S
4360 (non-UCP) in the class. Then copy the 32-byte map into the code vector,
4361 negating it if necessary. */
4362
4363 *code++ = (negate_class == should_flip_negation) ? OP_CLASS : OP_NCLASS;
4364 if (negate_class)
4365 {
4366 if (lengthptr == NULL) /* Save time in the pre-compile phase */
4367 for (c = 0; c < 32; c++) code[c] = ~classbits[c];
4368 }
4369 else
4370 {
4371 memcpy(code, classbits, 32);
4372 }
4373 code += 32;
4374 break;
4375
4376
4377 /* ===================================================================*/
4378 /* Various kinds of repeat; '{' is not necessarily a quantifier, but this
4379 has been tested above. */
4380
4381 case CHAR_LEFT_CURLY_BRACKET:
4382 if (!is_quantifier) goto NORMAL_CHAR;
4383 ptr = read_repeat_counts(ptr+1, &repeat_min, &repeat_max, errorcodeptr);
4384 if (*errorcodeptr != 0) goto FAILED;
4385 goto REPEAT;
4386
4387 case CHAR_ASTERISK:
4388 repeat_min = 0;
4389 repeat_max = -1;
4390 goto REPEAT;
4391
4392 case CHAR_PLUS:
4393 repeat_min = 1;
4394 repeat_max = -1;
4395 goto REPEAT;
4396
4397 case CHAR_QUESTION_MARK:
4398 repeat_min = 0;
4399 repeat_max = 1;
4400
4401 REPEAT:
4402 if (previous == NULL)
4403 {
4404 *errorcodeptr = ERR9;
4405 goto FAILED;
4406 }
4407
4408 if (repeat_min == 0)
4409 {
4410 firstbyte = zerofirstbyte; /* Adjust for zero repeat */
4411 reqbyte = zeroreqbyte; /* Ditto */
4412 }
4413
4414 /* Remember whether this is a variable length repeat */
4415
4416 reqvary = (repeat_min == repeat_max)? 0 : REQ_VARY;
4417
4418 op_type = 0; /* Default single-char op codes */
4419 possessive_quantifier = FALSE; /* Default not possessive quantifier */
4420
4421 /* Save start of previous item, in case we have to move it up in order to
4422 insert something before it. */
4423
4424 tempcode = previous;
4425
4426 /* If the next character is '+', we have a possessive quantifier. This
4427 implies greediness, whatever the setting of the PCRE_UNGREEDY option.
4428 If the next character is '?' this is a minimizing repeat, by default,
4429 but if PCRE_UNGREEDY is set, it works the other way round. We change the
4430 repeat type to the non-default. */
4431
4432 if (ptr[1] == CHAR_PLUS)
4433 {
4434 repeat_type = 0; /* Force greedy */
4435 possessive_quantifier = TRUE;
4436 ptr++;
4437 }
4438 else if (ptr[1] == CHAR_QUESTION_MARK)
4439 {
4440 repeat_type = greedy_non_default;
4441 ptr++;
4442 }
4443 else repeat_type = greedy_default;
4444
4445 /* If previous was a recursion call, wrap it in atomic brackets so that
4446 previous becomes the atomic group. All recursions were so wrapped in the
4447 past, but it no longer happens for non-repeated recursions. In fact, the
4448 repeated ones could be re-implemented independently so as not to need this,
4449 but for the moment we rely on the code for repeating groups. */
4450
4451 if (*previous == OP_RECURSE)
4452 {
4453 memmove(previous + 1 + LINK_SIZE, previous, 1 + LINK_SIZE);
4454 *previous = OP_ONCE;
4455 PUT(previous, 1, 2 + 2*LINK_SIZE);
4456 previous[2 + 2*LINK_SIZE] = OP_KET;
4457 PUT(previous, 3 + 2*LINK_SIZE, 2 + 2*LINK_SIZE);
4458 code += 2 + 2 * LINK_SIZE;
4459 length_prevgroup = 3 + 3*LINK_SIZE;
4460
4461 /* When actually compiling, we need to check whether this was a forward
4462 reference, and if so, adjust the offset. */
4463
4464 if (lengthptr == NULL && cd->hwm >= cd->start_workspace + LINK_SIZE)
4465 {
4466 int offset = GET(cd->hwm, -LINK_SIZE);
4467 if (offset == previous + 1 - cd->start_code)
4468 PUT(cd->hwm, -LINK_SIZE, offset + 1 + LINK_SIZE);
4469 }
4470 }
4471
4472 /* Now handle repetition for the different types of item. */
4473
4474 /* If previous was a character match, abolish the item and generate a
4475 repeat item instead. If a char item has a minumum of more than one, ensure
4476 that it is set in reqbyte - it might not be if a sequence such as x{3} is
4477 the first thing in a branch because the x will have gone into firstbyte
4478 instead. */
4479
4480 if (*previous == OP_CHAR || *previous == OP_CHARI)
4481 {
4482 op_type = (*previous == OP_CHAR)? 0 : OP_STARI - OP_STAR;
4483
4484 /* Deal with UTF-8 characters that take up more than one byte. It's
4485 easier to write this out separately than try to macrify it. Use c to
4486 hold the length of the character in bytes, plus 0x80 to flag that it's a
4487 length rather than a small character. */
4488
4489 #ifdef SUPPORT_UTF8
4490 if (utf8 && (code[-1] & 0x80) != 0)
4491 {
4492 pcre_uchar *lastchar = code - 1;
4493 while((*lastchar & 0xc0) == 0x80) lastchar--;
4494 c = code - lastchar; /* Length of UTF-8 character */
4495 memcpy(utf8_char, lastchar, c); /* Save the char */
4496 c |= 0x80; /* Flag c as a length */
4497 }
4498 else
4499 #endif
4500
4501 /* Handle the case of a single byte - either with no UTF8 support, or
4502 with UTF-8 disabled, or for a UTF-8 character < 128. */
4503
4504 {
4505 c = code[-1];
4506 if (repeat_min > 1) reqbyte = c | req_caseopt | cd->req_varyopt;
4507 }
4508
4509 /* If the repetition is unlimited, it pays to see if the next thing on
4510 the line is something that cannot possibly match this character. If so,
4511 automatically possessifying this item gains some performance in the case
4512 where the match fails. */
4513
4514 if (!possessive_quantifier &&
4515 repeat_max < 0 &&
4516 check_auto_possessive(previous, utf8, ptr + 1, options, cd))
4517 {
4518 repeat_type = 0; /* Force greedy */
4519 possessive_quantifier = TRUE;
4520 }
4521
4522 goto OUTPUT_SINGLE_REPEAT; /* Code shared with single character types */
4523 }
4524
4525 /* If previous was a single negated character ([^a] or similar), we use
4526 one of the special opcodes, replacing it. The code is shared with single-
4527 character repeats by setting opt_type to add a suitable offset into
4528 repeat_type. We can also test for auto-possessification. OP_NOT and OP_NOTI
4529 are currently used only for single-byte chars. */
4530
4531 else if (*previous == OP_NOT || *previous == OP_NOTI)
4532 {
4533 op_type = ((*previous == OP_NOT)? OP_NOTSTAR : OP_NOTSTARI) - OP_STAR;
4534 c = previous[1];
4535 if (!possessive_quantifier &&
4536 repeat_max < 0 &&
4537 check_auto_possessive(previous, utf8, ptr + 1, options, cd))
4538 {
4539 repeat_type = 0; /* Force greedy */
4540 possessive_quantifier = TRUE;
4541 }
4542 goto OUTPUT_SINGLE_REPEAT;
4543 }
4544
4545 /* If previous was a character type match (\d or similar), abolish it and
4546 create a suitable repeat item. The code is shared with single-character
4547 repeats by setting op_type to add a suitable offset into repeat_type. Note
4548 the the Unicode property types will be present only when SUPPORT_UCP is
4549 defined, but we don't wrap the little bits of code here because it just
4550 makes it horribly messy. */
4551
4552 else if (*previous < OP_EODN)
4553 {
4554 pcre_uchar *oldcode;
4555 int prop_type, prop_value;
4556 op_type = OP_TYPESTAR - OP_STAR; /* Use type opcodes */
4557 c = *previous;
4558
4559 if (!possessive_quantifier &&
4560 repeat_max < 0 &&
4561 check_auto_possessive(previous, utf8, ptr + 1, options, cd))
4562 {
4563 repeat_type = 0; /* Force greedy */
4564 possessive_quantifier = TRUE;
4565 }
4566
4567 OUTPUT_SINGLE_REPEAT:
4568 if (*previous == OP_PROP || *previous == OP_NOTPROP)
4569 {
4570 prop_type = previous[1];
4571 prop_value = previous[2];
4572 }
4573 else prop_type = prop_value = -1;
4574
4575 oldcode = code;
4576 code = previous; /* Usually overwrite previous item */
4577
4578 /* If the maximum is zero then the minimum must also be zero; Perl allows
4579 this case, so we do too - by simply omitting the item altogether. */
4580
4581 if (repeat_max == 0) goto END_REPEAT;
4582
4583 /*--------------------------------------------------------------------*/
4584 /* This code is obsolete from release 8.00; the restriction was finally
4585 removed: */
4586
4587 /* All real repeats make it impossible to handle partial matching (maybe
4588 one day we will be able to remove this restriction). */
4589
4590 /* if (repeat_max != 1) cd->external_flags |= PCRE_NOPARTIAL; */
4591 /*--------------------------------------------------------------------*/
4592
4593 /* Combine the op_type with the repeat_type */
4594
4595 repeat_type += op_type;
4596
4597 /* A minimum of zero is handled either as the special case * or ?, or as
4598 an UPTO, with the maximum given. */
4599
4600 if (repeat_min == 0)
4601 {
4602 if (repeat_max == -1) *code++ = OP_STAR + repeat_type;
4603 else if (repeat_max == 1) *code++ = OP_QUERY + repeat_type;
4604 else
4605 {
4606 *code++ = OP_UPTO + repeat_type;
4607 PUT2INC(code, 0, repeat_max);
4608 }
4609 }
4610
4611 /* A repeat minimum of 1 is optimized into some special cases. If the
4612 maximum is unlimited, we use OP_PLUS. Otherwise, the original item is
4613 left in place and, if the maximum is greater than 1, we use OP_UPTO with
4614 one less than the maximum. */
4615
4616 else if (repeat_min == 1)
4617 {
4618 if (repeat_max == -1)
4619 *code++ = OP_PLUS + repeat_type;
4620 else
4621 {
4622 code = oldcode; /* leave previous item in place */
4623 if (repeat_max == 1) goto END_REPEAT;
4624 *code++ = OP_UPTO + repeat_type;
4625 PUT2INC(code, 0, repeat_max - 1);
4626 }
4627 }
4628
4629 /* The case {n,n} is just an EXACT, while the general case {n,m} is
4630 handled as an EXACT followed by an UPTO. */
4631
4632 else
4633 {
4634 *code++ = OP_EXACT + op_type; /* NB EXACT doesn't have repeat_type */
4635 PUT2INC(code, 0, repeat_min);
4636
4637 /* If the maximum is unlimited, insert an OP_STAR. Before doing so,
4638 we have to insert the character for the previous code. For a repeated
4639 Unicode property match, there are two extra bytes that define the
4640 required property. In UTF-8 mode, long characters have their length in
4641 c, with the 0x80 bit as a flag. */
4642
4643 if (repeat_max < 0)
4644 {
4645 #ifdef SUPPORT_UTF8
4646 if (utf8 && c >= 128)
4647 {
4648 memcpy(code, utf8_char, c & 7);
4649 code += c & 7;
4650 }
4651 else
4652 #endif
4653 {
4654 *code++ = c;
4655 if (prop_type >= 0)
4656 {
4657 *code++ = prop_type;
4658 *code++ = prop_value;
4659 }
4660 }
4661 *code++ = OP_STAR + repeat_type;
4662 }
4663
4664 /* Else insert an UPTO if the max is greater than the min, again
4665 preceded by the character, for the previously inserted code. If the
4666 UPTO is just for 1 instance, we can use QUERY instead. */
4667
4668 else if (repeat_max != repeat_min)
4669 {
4670 #ifdef SUPPORT_UTF8
4671 if (utf8 && c >= 128)
4672 {
4673 memcpy(code, utf8_char, c & 7);
4674 code += c & 7;
4675 }
4676 else
4677 #endif
4678 *code++ = c;
4679 if (prop_type >= 0)
4680 {
4681 *code++ = prop_type;
4682 *code++ = prop_value;
4683 }
4684 repeat_max -= repeat_min;
4685
4686 if (repeat_max == 1)
4687 {
4688 *code++ = OP_QUERY + repeat_type;
4689 }
4690 else
4691 {
4692 *code++ = OP_UPTO + repeat_type;
4693 PUT2INC(code, 0, repeat_max);
4694 }
4695 }
4696 }
4697
4698 /* The character or character type itself comes last in all cases. */
4699
4700 #ifdef SUPPORT_UTF8
4701 if (utf8 && c >= 128)
4702 {
4703 memcpy(code, utf8_char, c & 7);
4704 code += c & 7;
4705 }
4706 else
4707 #endif
4708 *code++ = c;
4709
4710 /* For a repeated Unicode property match, there are two extra bytes that
4711 define the required property. */
4712
4713 #ifdef SUPPORT_UCP
4714 if (prop_type >= 0)
4715 {
4716 *code++ = prop_type;
4717 *code++ = prop_value;
4718 }
4719 #endif
4720 }
4721
4722 /* If previous was a character class or a back reference, we put the repeat
4723 stuff after it, but just skip the item if the repeat was {0,0}. */
4724
4725 else if (*previous == OP_CLASS ||
4726 *previous == OP_NCLASS ||
4727 #ifdef SUPPORT_UTF8
4728 *previous == OP_XCLASS ||
4729 #endif
4730 *previous == OP_REF ||
4731 *previous == OP_REFI)
4732 {
4733 if (repeat_max == 0)
4734 {
4735 code = previous;
4736 goto END_REPEAT;
4737 }
4738
4739 /*--------------------------------------------------------------------*/
4740 /* This code is obsolete from release 8.00; the restriction was finally
4741 removed: */
4742
4743 /* All real repeats make it impossible to handle partial matching (maybe
4744 one day we will be able to remove this restriction). */
4745
4746 /* if (repeat_max != 1) cd->external_flags |= PCRE_NOPARTIAL; */
4747 /*--------------------------------------------------------------------*/
4748
4749 if (repeat_min == 0 && repeat_max == -1)
4750 *code++ = OP_CRSTAR + repeat_type;
4751 else if (repeat_min == 1 && repeat_max == -1)
4752 *code++ = OP_CRPLUS + repeat_type;
4753 else if (repeat_min == 0 && repeat_max == 1)
4754 *code++ = OP_CRQUERY + repeat_type;
4755 else
4756 {
4757 *code++ = OP_CRRANGE + repeat_type;
4758 PUT2INC(code, 0, repeat_min);
4759 if (repeat_max == -1) repeat_max = 0; /* 2-byte encoding for max */
4760 PUT2INC(code, 0, repeat_max);
4761 }
4762 }
4763
4764 /* If previous was a bracket group, we may have to replicate it in certain
4765 cases. Note that at this point we can encounter only the "basic" bracket
4766 opcodes such as BRA and CBRA, as this is the place where they get converted
4767 into the more special varieties such as BRAPOS and SBRA. A test for >=
4768 OP_ASSERT and <= OP_COND includes ASSERT, ASSERT_NOT, ASSERTBACK,
4769 ASSERTBACK_NOT, ONCE, BRA, CBRA, and COND. Originally, PCRE did not allow
4770 repetition of assertions, but now it does, for Perl compatibility. */
4771
4772 else if (*previous >= OP_ASSERT && *previous <= OP_COND)
4773 {
4774 register int i;
4775 int len = (int)(code - previous);
4776 pcre_uchar *bralink = NULL;
4777 pcre_uchar *brazeroptr = NULL;
4778
4779 /* Repeating a DEFINE group is pointless, but Perl allows the syntax, so
4780 we just ignore the repeat. */
4781
4782 if (*previous == OP_COND && previous[LINK_SIZE+1] == OP_DEF)
4783 goto END_REPEAT;
4784
4785 /* There is no sense in actually repeating assertions. The only potential
4786 use of repetition is in cases when the assertion is optional. Therefore,
4787 if the minimum is greater than zero, just ignore the repeat. If the
4788 maximum is not not zero or one, set it to 1. */
4789
4790 if (*previous < OP_ONCE) /* Assertion */
4791 {
4792 if (repeat_min > 0) goto END_REPEAT;
4793 if (repeat_max < 0 || repeat_max > 1) repeat_max = 1;
4794 }
4795
4796 /* The case of a zero minimum is special because of the need to stick
4797 OP_BRAZERO in front of it, and because the group appears once in the
4798 data, whereas in other cases it appears the minimum number of times. For
4799 this reason, it is simplest to treat this case separately, as otherwise
4800 the code gets far too messy. There are several special subcases when the
4801 minimum is zero. */
4802
4803 if (repeat_min == 0)
4804 {
4805 /* If the maximum is also zero, we used to just omit the group from the
4806 output altogether, like this:
4807
4808 ** if (repeat_max == 0)
4809 ** {
4810 ** code = previous;
4811 ** goto END_REPEAT;
4812 ** }
4813
4814 However, that fails when a group or a subgroup within it is referenced
4815 as a subroutine from elsewhere in the pattern, so now we stick in
4816 OP_SKIPZERO in front of it so that it is skipped on execution. As we
4817 don't have a list of which groups are referenced, we cannot do this
4818 selectively.
4819
4820 If the maximum is 1 or unlimited, we just have to stick in the BRAZERO
4821 and do no more at this point. However, we do need to adjust any
4822 OP_RECURSE calls inside the group that refer to the group itself or any
4823 internal or forward referenced group, because the offset is from the
4824 start of the whole regex. Temporarily terminate the pattern while doing
4825 this. */
4826
4827 if (repeat_max <= 1) /* Covers 0, 1, and unlimited */
4828 {
4829 *code = OP_END;
4830 adjust_recurse(previous, 1, utf8, cd, save_hwm);
4831 memmove(previous+1, previous, len);
4832 code++;
4833 if (repeat_max == 0)
4834 {
4835 *previous++ = OP_SKIPZERO;
4836 goto END_REPEAT;
4837 }
4838 brazeroptr = previous; /* Save for possessive optimizing */
4839 *previous++ = OP_BRAZERO + repeat_type;
4840 }
4841
4842 /* If the maximum is greater than 1 and limited, we have to replicate
4843 in a nested fashion, sticking OP_BRAZERO before each set of brackets.
4844 The first one has to be handled carefully because it's the original
4845 copy, which has to be moved up. The remainder can be handled by code
4846 that is common with the non-zero minimum case below. We have to
4847 adjust the value or repeat_max, since one less copy is required. Once
4848 again, we may have to adjust any OP_RECURSE calls inside the group. */
4849
4850 else
4851 {
4852 int offset;
4853 *code = OP_END;
4854 adjust_recurse(previous, 2 + LINK_SIZE, utf8, cd, save_hwm);
4855 memmove(previous + 2 + LINK_SIZE, previous, len);
4856 code += 2 + LINK_SIZE;
4857 *previous++ = OP_BRAZERO + repeat_type;
4858 *previous++ = OP_BRA;
4859
4860 /* We chain together the bracket offset fields that have to be
4861 filled in later when the ends of the brackets are reached. */
4862
4863 offset = (bralink == NULL)? 0 : (int)(previous - bralink);
4864 bralink = previous;
4865 PUTINC(previous, 0, offset);
4866 }
4867
4868 repeat_max--;
4869 }
4870
4871 /* If the minimum is greater than zero, replicate the group as many
4872 times as necessary, and adjust the maximum to the number of subsequent
4873 copies that we need. If we set a first char from the group, and didn't
4874 set a required char, copy the latter from the former. If there are any
4875 forward reference subroutine calls in the group, there will be entries on
4876 the workspace list; replicate these with an appropriate increment. */
4877
4878 else
4879 {
4880 if (repeat_min > 1)
4881 {
4882 /* In the pre-compile phase, we don't actually do the replication. We
4883 just adjust the length as if we had. Do some paranoid checks for
4884 potential integer overflow. The INT64_OR_DOUBLE type is a 64-bit
4885 integer type when available, otherwise double. */
4886
4887 if (lengthptr != NULL)
4888 {
4889 int delta = (repeat_min - 1)*length_prevgroup;
4890 if ((INT64_OR_DOUBLE)(repeat_min - 1)*
4891 (INT64_OR_DOUBLE)length_prevgroup >
4892 (INT64_OR_DOUBLE)INT_MAX ||
4893 OFLOW_MAX - *lengthptr < delta)
4894 {
4895 *errorcodeptr = ERR20;
4896 goto FAILED;
4897 }
4898 *lengthptr += delta;
4899 }
4900
4901 /* This is compiling for real */
4902
4903 else
4904 {
4905 if (groupsetfirstbyte && reqbyte < 0) reqbyte = firstbyte;
4906 for (i = 1; i < repeat_min; i++)
4907 {
4908 pcre_uchar *hc;
4909 pcre_uchar *this_hwm = cd->hwm;
4910 memcpy(code, previous, len);
4911 for (hc = save_hwm; hc < this_hwm; hc += LINK_SIZE)
4912 {
4913 PUT(cd->hwm, 0, GET(hc, 0) + len);
4914 cd->hwm += LINK_SIZE;
4915 }
4916 save_hwm = this_hwm;
4917 code += len;
4918 }
4919 }
4920 }
4921
4922 if (repeat_max > 0) repeat_max -= repeat_min;
4923 }
4924
4925 /* This code is common to both the zero and non-zero minimum cases. If
4926 the maximum is limited, it replicates the group in a nested fashion,
4927 remembering the bracket starts on a stack. In the case of a zero minimum,
4928 the first one was set up above. In all cases the repeat_max now specifies
4929 the number of additional copies needed. Again, we must remember to
4930 replicate entries on the forward reference list. */
4931
4932 if (repeat_max >= 0)
4933 {
4934 /* In the pre-compile phase, we don't actually do the replication. We
4935 just adjust the length as if we had. For each repetition we must add 1
4936 to the length for BRAZERO and for all but the last repetition we must
4937 add 2 + 2*LINKSIZE to allow for the nesting that occurs. Do some
4938 paranoid checks to avoid integer overflow. The INT64_OR_DOUBLE type is
4939 a 64-bit integer type when available, otherwise double. */
4940
4941 if (lengthptr != NULL && repeat_max > 0)
4942 {
4943 int delta = repeat_max * (length_prevgroup + 1 + 2 + 2*LINK_SIZE) -
4944 2 - 2*LINK_SIZE; /* Last one doesn't nest */
4945 if ((INT64_OR_DOUBLE)repeat_max *
4946 (INT64_OR_DOUBLE)(length_prevgroup + 1 + 2 + 2*LINK_SIZE)
4947 > (INT64_OR_DOUBLE)INT_MAX ||
4948 OFLOW_MAX - *lengthptr < delta)
4949 {
4950 *errorcodeptr = ERR20;
4951 goto FAILED;
4952 }
4953 *lengthptr += delta;
4954 }
4955
4956 /* This is compiling for real */
4957
4958 else for (i = repeat_max - 1; i >= 0; i--)
4959 {
4960 pcre_uchar *hc;
4961 pcre_uchar *this_hwm = cd->hwm;
4962
4963 *code++ = OP_BRAZERO + repeat_type;
4964
4965 /* All but the final copy start a new nesting, maintaining the
4966 chain of brackets outstanding. */
4967
4968 if (i != 0)
4969 {
4970 int offset;
4971 *code++ = OP_BRA;
4972 offset = (bralink == NULL)? 0 : (int)(code - bralink);
4973 bralink = code;
4974 PUTINC(code, 0, offset);
4975 }
4976
4977 memcpy(code, previous, len);
4978 for (hc = save_hwm; hc < this_hwm; hc += LINK_SIZE)
4979 {
4980 PUT(cd->hwm, 0, GET(hc, 0) + len + ((i != 0)? 2+LINK_SIZE : 1));
4981 cd->hwm += LINK_SIZE;
4982 }
4983 save_hwm = this_hwm;
4984 code += len;
4985 }
4986
4987 /* Now chain through the pending brackets, and fill in their length
4988 fields (which are holding the chain links pro tem). */
4989
4990 while (bralink != NULL)
4991 {
4992 int oldlinkoffset;
4993 int offset = (int)(code - bralink + 1);
4994 pcre_uchar *bra = code - offset;
4995 oldlinkoffset = GET(bra, 1);
4996 bralink = (oldlinkoffset == 0)? NULL : bralink - oldlinkoffset;
4997 *code++ = OP_KET;
4998 PUTINC(code, 0, offset);
4999 PUT(bra, 1, offset);
5000 }
5001 }
5002
5003 /* If the maximum is unlimited, set a repeater in the final copy. For
5004 ONCE brackets, that's all we need to do. However, possessively repeated
5005 ONCE brackets can be converted into non-capturing brackets, as the
5006 behaviour of (?:xx)++ is the same as (?>xx)++ and this saves having to
5007 deal with possessive ONCEs specially.
5008
5009 Otherwise, when we are doing the actual compile phase, check to see
5010 whether this group is one that could match an empty string. If so,
5011 convert the initial operator to the S form (e.g. OP_BRA -> OP_SBRA) so
5012 that runtime checking can be done. [This check is also applied to ONCE
5013 groups at runtime, but in a different way.]
5014
5015 Then, if the quantifier was possessive and the bracket is not a
5016 conditional, we convert the BRA code to the POS form, and the KET code to
5017 KETRPOS. (It turns out to be convenient at runtime to detect this kind of
5018 subpattern at both the start and at the end.) The use of special opcodes
5019 makes it possible to reduce greatly the stack usage in pcre_exec(). If
5020 the group is preceded by OP_BRAZERO, convert this to OP_BRAPOSZERO.
5021
5022 Then, if the minimum number of matches is 1 or 0, cancel the possessive
5023 flag so that the default action below, of wrapping everything inside
5024 atomic brackets, does not happen. When the minimum is greater than 1,
5025 there will be earlier copies of the group, and so we still have to wrap
5026 the whole thing. */
5027
5028 else
5029 {
5030 pcre_uchar *ketcode = code - 1 - LINK_SIZE;
5031 pcre_uchar *bracode = ketcode - GET(ketcode, 1);
5032
5033 /* Convert possessive ONCE brackets to non-capturing */
5034
5035 if ((*bracode == OP_ONCE || *bracode == OP_ONCE_NC) &&
5036 possessive_quantifier) *bracode = OP_BRA;
5037
5038 /* For non-possessive ONCE brackets, all we need to do is to
5039 set the KET. */
5040
5041 if (*bracode == OP_ONCE || *bracode == OP_ONCE_NC)
5042 *ketcode = OP_KETRMAX + repeat_type;
5043
5044 /* Handle non-ONCE brackets and possessive ONCEs (which have been
5045 converted to non-capturing above). */
5046
5047 else
5048 {
5049 /* In the compile phase, check for empty string matching. */
5050
5051 if (lengthptr == NULL)
5052 {
5053 pcre_uchar *scode = bracode;
5054 do
5055 {
5056 if (could_be_empty_branch(scode, ketcode, utf8, cd))
5057 {
5058 *bracode += OP_SBRA - OP_BRA;
5059 break;
5060 }
5061 scode += GET(scode, 1);
5062 }
5063 while (*scode == OP_ALT);
5064 }
5065
5066 /* Handle possessive quantifiers. */
5067
5068 if (possessive_quantifier)
5069 {
5070 /* For COND brackets, we wrap the whole thing in a possessively
5071 repeated non-capturing bracket, because we have not invented POS
5072 versions of the COND opcodes. Because we are moving code along, we
5073 must ensure that any pending recursive references are updated. */
5074
5075 if (*bracode == OP_COND || *bracode == OP_SCOND)
5076 {
5077 int nlen = (int)(code - bracode);
5078 *code = OP_END;
5079 adjust_recurse(bracode, 1 + LINK_SIZE, utf8, cd, save_hwm);
5080 memmove(bracode + 1+LINK_SIZE, bracode, nlen);
5081 code += 1 + LINK_SIZE;
5082 nlen += 1 + LINK_SIZE;
5083 *bracode = OP_BRAPOS;
5084 *code++ = OP_KETRPOS;
5085 PUTINC(code, 0, nlen);
5086 PUT(bracode, 1, nlen);
5087 }
5088
5089 /* For non-COND brackets, we modify the BRA code and use KETRPOS. */
5090
5091 else
5092 {
5093 *bracode += 1; /* Switch to xxxPOS opcodes */
5094 *ketcode = OP_KETRPOS;
5095 }
5096
5097 /* If the minimum is zero, mark it as possessive, then unset the
5098 possessive flag when the minimum is 0 or 1. */
5099
5100 if (brazeroptr != NULL) *brazeroptr = OP_BRAPOSZERO;
5101 if (repeat_min < 2) possessive_quantifier = FALSE;
5102 }
5103
5104 /* Non-possessive quantifier */
5105
5106 else *ketcode = OP_KETRMAX + repeat_type;
5107 }
5108 }
5109 }
5110
5111 /* If previous is OP_FAIL, it was generated by an empty class [] in
5112 JavaScript mode. The other ways in which OP_FAIL can be generated, that is
5113 by (*FAIL) or (?!) set previous to NULL, which gives a "nothing to repeat"
5114 error above. We can just ignore the repeat in JS case. */
5115
5116 else if (*previous == OP_FAIL) goto END_REPEAT;
5117
5118 /* Else there's some kind of shambles */
5119
5120 else
5121 {
5122 *errorcodeptr = ERR11;
5123 goto FAILED;
5124 }
5125
5126 /* If the character following a repeat is '+', or if certain optimization
5127 tests above succeeded, possessive_quantifier is TRUE. For some opcodes,
5128 there are special alternative opcodes for this case. For anything else, we
5129 wrap the entire repeated item inside OP_ONCE brackets. Logically, the '+'
5130 notation is just syntactic sugar, taken from Sun's Java package, but the
5131 special opcodes can optimize it.
5132
5133 Some (but not all) possessively repeated subpatterns have already been
5134 completely handled in the code just above. For them, possessive_quantifier
5135 is always FALSE at this stage.
5136
5137 Note that the repeated item starts at tempcode, not at previous, which
5138 might be the first part of a string whose (former) last char we repeated.
5139
5140 Possessifying an 'exact' quantifier has no effect, so we can ignore it. But
5141 an 'upto' may follow. We skip over an 'exact' item, and then test the
5142 length of what remains before proceeding. */
5143
5144 if (possessive_quantifier)
5145 {
5146 int len;
5147
5148 if (*tempcode == OP_TYPEEXACT)
5149 tempcode += _pcre_OP_lengths[*tempcode] +
5150 ((tempcode[3] == OP_PROP || tempcode[3] == OP_NOTPROP)? 2 : 0);
5151
5152 else if (*tempcode == OP_EXACT || *tempcode == OP_NOTEXACT)
5153 {
5154 tempcode += _pcre_OP_lengths[*tempcode];
5155 #ifdef SUPPORT_UTF8
5156 if (utf8 && tempcode[-1] >= 0xc0)
5157 tempcode += _pcre_utf8_table4[tempcode[-1] & 0x3f];
5158 #endif
5159 }
5160
5161 len = (int)(code - tempcode);
5162 if (len > 0) switch (*tempcode)
5163 {
5164 case OP_STAR: *tempcode = OP_POSSTAR; break;
5165 case OP_PLUS: *tempcode = OP_POSPLUS; break;
5166 case OP_QUERY: *tempcode = OP_POSQUERY; break;
5167 case OP_UPTO: *tempcode = OP_POSUPTO; break;
5168
5169 case OP_STARI: *tempcode = OP_POSSTARI; break;
5170 case OP_PLUSI: *tempcode = OP_POSPLUSI; break;
5171 case OP_QUERYI: *tempcode = OP_POSQUERYI; break;
5172 case OP_UPTOI: *tempcode = OP_POSUPTOI; break;
5173
5174 case OP_NOTSTAR: *tempcode = OP_NOTPOSSTAR; break;
5175 case OP_NOTPLUS: *tempcode = OP_NOTPOSPLUS; break;
5176 case OP_NOTQUERY: *tempcode = OP_NOTPOSQUERY; break;
5177 case OP_NOTUPTO: *tempcode = OP_NOTPOSUPTO; break;
5178
5179 case OP_NOTSTARI: *tempcode = OP_NOTPOSSTARI; break;
5180 case OP_NOTPLUSI: *tempcode = OP_NOTPOSPLUSI; break;
5181 case OP_NOTQUERYI: *tempcode = OP_NOTPOSQUERYI; break;
5182 case OP_NOTUPTOI: *tempcode = OP_NOTPOSUPTOI; break;
5183
5184 case OP_TYPESTAR: *tempcode = OP_TYPEPOSSTAR; break;
5185 case OP_TYPEPLUS: *tempcode = OP_TYPEPOSPLUS; break;
5186 case OP_TYPEQUERY: *tempcode = OP_TYPEPOSQUERY; break;
5187 case OP_TYPEUPTO: *tempcode = OP_TYPEPOSUPTO; break;
5188
5189 /* Because we are moving code along, we must ensure that any
5190 pending recursive references are updated. */
5191
5192 default:
5193 *code = OP_END;
5194 adjust_recurse(tempcode, 1 + LINK_SIZE, utf8, cd, save_hwm);
5195 memmove(tempcode + 1+LINK_SIZE, tempcode, len);
5196 code += 1 + LINK_SIZE;
5197 len += 1 + LINK_SIZE;
5198 tempcode[0] = OP_ONCE;
5199 *code++ = OP_KET;
5200 PUTINC(code, 0, len);
5201 PUT(tempcode, 1, len);
5202 break;
5203 }
5204 }
5205
5206 /* In all case we no longer have a previous item. We also set the
5207 "follows varying string" flag for subsequently encountered reqbytes if
5208 it isn't already set and we have just passed a varying length item. */
5209
5210 END_REPEAT:
5211 previous = NULL;
5212 cd->req_varyopt |= reqvary;
5213 break;
5214
5215
5216 /* ===================================================================*/
5217 /* Start of nested parenthesized sub-expression, or comment or lookahead or
5218 lookbehind or option setting or condition or all the other extended
5219 parenthesis forms. */
5220
5221 case CHAR_LEFT_PARENTHESIS:
5222 newoptions = options;
5223 skipbytes = 0;
5224 bravalue = OP_CBRA;
5225 save_hwm = cd->hwm;
5226 reset_bracount = FALSE;
5227
5228 /* First deal with various "verbs" that can be introduced by '*'. */
5229
5230 if (*(++ptr) == CHAR_ASTERISK &&
5231 ((cd->ctypes[ptr[1]] & ctype_letter) != 0 || ptr[1] == ':'))
5232 {
5233 int i, namelen;
5234 int arglen = 0;
5235 const char *vn = verbnames;
5236 const pcre_uchar *name = ptr + 1;
5237 const pcre_uchar *arg = NULL;
5238 previous = NULL;
5239 while ((cd->ctypes[*++ptr] & ctype_letter) != 0) {};
5240 namelen = (int)(ptr - name);
5241
5242 /* It appears that Perl allows any characters whatsoever, other than
5243 a closing parenthesis, to appear in arguments, so we no longer insist on
5244 letters, digits, and underscores. */
5245
5246 if (*ptr == CHAR_COLON)
5247 {
5248 arg = ++ptr;
5249 while (*ptr != 0 && *ptr != CHAR_RIGHT_PARENTHESIS) ptr++;
5250 arglen = (int)(ptr - arg);
5251 }
5252
5253 if (*ptr != CHAR_RIGHT_PARENTHESIS)
5254 {
5255 *errorcodeptr = ERR60;
5256 goto FAILED;
5257 }
5258
5259 /* Scan the table of verb names */
5260
5261 for (i = 0; i < verbcount; i++)
5262 {
5263 if (namelen == verbs[i].len &&
5264 strncmp((char *)name, vn, namelen) == 0)
5265 {
5266 /* Check for open captures before ACCEPT and convert it to
5267 ASSERT_ACCEPT if in an assertion. */
5268
5269 if (verbs[i].op == OP_ACCEPT)
5270 {
5271 open_capitem *oc;
5272 if (arglen != 0)
5273 {
5274 *errorcodeptr = ERR59;
5275 goto FAILED;
5276 }
5277 cd->had_accept = TRUE;
5278 for (oc = cd->open_caps; oc != NULL; oc = oc->next)
5279 {
5280 *code++ = OP_CLOSE;
5281 PUT2INC(code, 0, oc->number);
5282 }
5283 *code++ = (cd->assert_depth > 0)? OP_ASSERT_ACCEPT : OP_ACCEPT;
5284
5285 /* Do not set firstbyte after *ACCEPT */
5286 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
5287 }
5288
5289 /* Handle other cases with/without an argument */
5290
5291 else if (arglen == 0)
5292 {
5293 if (verbs[i].op < 0) /* Argument is mandatory */
5294 {
5295 *errorcodeptr = ERR66;
5296 goto FAILED;
5297 }
5298 *code = verbs[i].op;
5299 if (*code++ == OP_THEN) cd->external_flags |= PCRE_HASTHEN;
5300 }
5301
5302 else
5303 {
5304 if (verbs[i].op_arg < 0) /* Argument is forbidden */
5305 {
5306 *errorcodeptr = ERR59;
5307 goto FAILED;
5308 }
5309 *code = verbs[i].op_arg;
5310 if (*code++ == OP_THEN_ARG) cd->external_flags |= PCRE_HASTHEN;
5311 *code++ = arglen;
5312 memcpy(code, arg, arglen);
5313 code += arglen;
5314 *code++ = 0;
5315 }
5316
5317 break; /* Found verb, exit loop */
5318 }
5319
5320 vn += verbs[i].len + 1;
5321 }
5322
5323 if (i < verbcount) continue; /* Successfully handled a verb */
5324 *errorcodeptr = ERR60; /* Verb not recognized */
5325 goto FAILED;
5326 }
5327
5328 /* Deal with the extended parentheses; all are introduced by '?', and the
5329 appearance of any of them means that this is not a capturing group. */
5330
5331 else if (*ptr == CHAR_QUESTION_MARK)
5332 {
5333 int i, set, unset, namelen;
5334 int *optset;
5335 const pcre_uchar *name;
5336 pcre_uchar *slot;
5337
5338 switch (*(++ptr))
5339 {
5340 case CHAR_NUMBER_SIGN: /* Comment; skip to ket */
5341 ptr++;
5342 while (*ptr != 0 && *ptr != CHAR_RIGHT_PARENTHESIS) ptr++;
5343 if (*ptr == 0)
5344 {
5345 *errorcodeptr = ERR18;
5346 goto FAILED;
5347 }
5348 continue;
5349
5350
5351 /* ------------------------------------------------------------ */
5352 case CHAR_VERTICAL_LINE: /* Reset capture count for each branch */
5353 reset_bracount = TRUE;
5354 /* Fall through */
5355
5356 /* ------------------------------------------------------------ */
5357 case CHAR_COLON: /* Non-capturing bracket */
5358 bravalue = OP_BRA;
5359 ptr++;
5360 break;
5361
5362
5363 /* ------------------------------------------------------------ */
5364 case CHAR_LEFT_PARENTHESIS:
5365 bravalue = OP_COND; /* Conditional group */
5366
5367 /* A condition can be an assertion, a number (referring to a numbered
5368 group), a name (referring to a named group), or 'R', referring to
5369 recursion. R<digits> and R&name are also permitted for recursion tests.
5370
5371 There are several syntaxes for testing a named group: (?(name)) is used
5372 by Python; Perl 5.10 onwards uses (?(<name>) or (?('name')).
5373
5374 There are two unfortunate ambiguities, caused by history. (a) 'R' can
5375 be the recursive thing or the name 'R' (and similarly for 'R' followed
5376 by digits), and (b) a number could be a name that consists of digits.
5377 In both cases, we look for a name first; if not found, we try the other
5378 cases. */
5379
5380 /* For conditions that are assertions, check the syntax, and then exit
5381 the switch. This will take control down to where bracketed groups,
5382 including assertions, are processed. */
5383
5384 if (ptr[1] == CHAR_QUESTION_MARK && (ptr[2] == CHAR_EQUALS_SIGN ||
5385 ptr[2] == CHAR_EXCLAMATION_MARK || ptr[2] == CHAR_LESS_THAN_SIGN))
5386 break;
5387
5388 /* Most other conditions use OP_CREF (a couple change to OP_RREF
5389 below), and all need to skip 3 bytes at the start of the group. */
5390
5391 code[1+LINK_SIZE] = OP_CREF;
5392 skipbytes = 3;
5393 refsign = -1;
5394
5395 /* Check for a test for recursion in a named group. */
5396
5397 if (ptr[1] == CHAR_R && ptr[2] == CHAR_AMPERSAND)
5398 {
5399 terminator = -1;
5400 ptr += 2;
5401 code[1+LINK_SIZE] = OP_RREF; /* Change the type of test */
5402 }
5403
5404 /* Check for a test for a named group's having been set, using the Perl
5405 syntax (?(<name>) or (?('name') */
5406
5407 else if (ptr[1] == CHAR_LESS_THAN_SIGN)
5408 {
5409 terminator = CHAR_GREATER_THAN_SIGN;
5410 ptr++;
5411 }
5412 else if (ptr[1] == CHAR_APOSTROPHE)
5413 {
5414 terminator = CHAR_APOSTROPHE;
5415 ptr++;
5416 }
5417 else
5418 {
5419 terminator = 0;
5420 if (ptr[1] == CHAR_MINUS || ptr[1] == CHAR_PLUS) refsign = *(++ptr);
5421 }
5422
5423 /* We now expect to read a name; any thing else is an error */
5424
5425 if ((cd->ctypes[ptr[1]] & ctype_word) == 0)
5426 {
5427 ptr += 1; /* To get the right offset */
5428 *errorcodeptr = ERR28;
5429 goto FAILED;
5430 }
5431
5432 /* Read the name, but also get it as a number if it's all digits */
5433
5434 recno = 0;
5435 name = ++ptr;
5436 while ((cd->ctypes[*ptr] & ctype_word) != 0)
5437 {
5438 if (recno >= 0)
5439 recno = ((digitab[*ptr] & ctype_digit) != 0)?
5440 recno * 10 + *ptr - CHAR_0 : -1;
5441 ptr++;
5442 }
5443 namelen = (int)(ptr - name);
5444
5445 if ((terminator > 0 && *ptr++ != terminator) ||
5446 *ptr++ != CHAR_RIGHT_PARENTHESIS)
5447 {
5448 ptr--; /* Error offset */
5449 *errorcodeptr = ERR26;
5450 goto FAILED;
5451 }
5452
5453 /* Do no further checking in the pre-compile phase. */
5454
5455 if (lengthptr != NULL) break;
5456
5457 /* In the real compile we do the work of looking for the actual
5458 reference. If the string started with "+" or "-" we require the rest to
5459 be digits, in which case recno will be set. */
5460
5461 if (refsign > 0)
5462 {
5463 if (recno <= 0)
5464 {
5465 *errorcodeptr = ERR58;
5466 goto FAILED;
5467 }
5468 recno = (refsign == CHAR_MINUS)?
5469 cd->bracount - recno + 1 : recno +cd->bracount;
5470 if (recno <= 0 || recno > cd->final_bracount)
5471 {
5472 *errorcodeptr = ERR15;
5473 goto FAILED;
5474 }
5475 PUT2(code, 2+LINK_SIZE, recno);
5476 break;
5477 }
5478
5479 /* Otherwise (did not start with "+" or "-"), start by looking for the
5480 name. If we find a name, add one to the opcode to change OP_CREF or
5481 OP_RREF into OP_NCREF or OP_NRREF. These behave exactly the same,
5482 except they record that the reference was originally to a name. The
5483 information is used to check duplicate names. */
5484
5485 slot = cd->name_table;
5486 for (i = 0; i < cd->names_found; i++)
5487 {
5488 if (strncmp((char *)name, (char *)slot+2, namelen) == 0) break;
5489 slot += cd->name_entry_size;
5490 }
5491
5492 /* Found a previous named subpattern */
5493
5494 if (i < cd->names_found)
5495 {
5496 recno = GET2(slot, 0);
5497 PUT2(code, 2+LINK_SIZE, recno);
5498 code[1+LINK_SIZE]++;
5499 }
5500
5501 /* Search the pattern for a forward reference */
5502
5503 else if ((i = find_parens(cd, name, namelen,
5504 (options & PCRE_EXTENDED) != 0, utf8)) > 0)
5505 {
5506 PUT2(code, 2+LINK_SIZE, i);
5507 code[1+LINK_SIZE]++;
5508 }
5509
5510 /* If terminator == 0 it means that the name followed directly after
5511 the opening parenthesis [e.g. (?(abc)...] and in this case there are
5512 some further alternatives to try. For the cases where terminator != 0
5513 [things like (?(<name>... or (?('name')... or (?(R&name)... ] we have
5514 now checked all the possibilities, so give an error. */
5515
5516 else if (terminator != 0)
5517 {
5518 *errorcodeptr = ERR15;
5519 goto FAILED;
5520 }
5521
5522 /* Check for (?(R) for recursion. Allow digits after R to specify a
5523 specific group number. */
5524
5525 else if (*name == CHAR_R)
5526 {
5527 recno = 0;
5528 for (i = 1; i < namelen; i++)
5529 {
5530 if ((digitab[name[i]] & ctype_digit) == 0)
5531 {
5532 *errorcodeptr = ERR15;
5533 goto FAILED;
5534 }
5535 recno = recno * 10 + name[i] - CHAR_0;
5536 }
5537 if (recno == 0) recno = RREF_ANY;
5538 code[1+LINK_SIZE] = OP_RREF; /* Change test type */
5539 PUT2(code, 2+LINK_SIZE, recno);
5540 }
5541
5542 /* Similarly, check for the (?(DEFINE) "condition", which is always
5543 false. */
5544
5545 else if (namelen == 6 && strncmp((char *)name, STRING_DEFINE, 6) == 0)
5546 {
5547 code[1+LINK_SIZE] = OP_DEF;
5548 skipbytes = 1;
5549 }
5550
5551 /* Check for the "name" actually being a subpattern number. We are
5552 in the second pass here, so final_bracount is set. */
5553
5554 else if (recno > 0 && recno <= cd->final_bracount)
5555 {
5556 PUT2(code, 2+LINK_SIZE, recno);
5557 }
5558
5559 /* Either an unidentified subpattern, or a reference to (?(0) */
5560
5561 else
5562 {
5563 *errorcodeptr = (recno == 0)? ERR35: ERR15;
5564 goto FAILED;
5565 }
5566 break;
5567
5568
5569 /* ------------------------------------------------------------ */
5570 case CHAR_EQUALS_SIGN: /* Positive lookahead */
5571 bravalue = OP_ASSERT;
5572 cd->assert_depth += 1;
5573 ptr++;
5574 break;
5575
5576
5577 /* ------------------------------------------------------------ */
5578 case CHAR_EXCLAMATION_MARK: /* Negative lookahead */
5579 ptr++;
5580 if (*ptr == CHAR_RIGHT_PARENTHESIS) /* Optimize (?!) */
5581 {
5582 *code++ = OP_FAIL;
5583 previous = NULL;
5584 continue;
5585 }
5586 bravalue = OP_ASSERT_NOT;
5587 cd->assert_depth += 1;
5588 break;
5589
5590
5591 /* ------------------------------------------------------------ */
5592 case CHAR_LESS_THAN_SIGN: /* Lookbehind or named define */
5593 switch (ptr[1])
5594 {
5595 case CHAR_EQUALS_SIGN: /* Positive lookbehind */
5596 bravalue = OP_ASSERTBACK;
5597 cd->assert_depth += 1;
5598 ptr += 2;
5599 break;
5600
5601 case CHAR_EXCLAMATION_MARK: /* Negative lookbehind */
5602 bravalue = OP_ASSERTBACK_NOT;
5603 cd->assert_depth += 1;
5604 ptr += 2;
5605 break;
5606
5607 default: /* Could be name define, else bad */
5608 if ((cd->ctypes[ptr[1]] & ctype_word) != 0) goto DEFINE_NAME;
5609 ptr++; /* Correct offset for error */
5610 *errorcodeptr = ERR24;
5611 goto FAILED;
5612 }
5613 break;
5614
5615
5616 /* ------------------------------------------------------------ */
5617 case CHAR_GREATER_THAN_SIGN: /* One-time brackets */
5618 bravalue = OP_ONCE;
5619 ptr++;
5620 break;
5621
5622
5623 /* ------------------------------------------------------------ */
5624 case CHAR_C: /* Callout - may be followed by digits; */
5625 previous_callout = code; /* Save for later completion */
5626 after_manual_callout = 1; /* Skip one item before completing */
5627 *code++ = OP_CALLOUT;
5628 {
5629 int n = 0;
5630 while ((digitab[*(++ptr)] & ctype_digit) != 0)
5631 n = n * 10 + *ptr - CHAR_0;
5632 if (*ptr != CHAR_RIGHT_PARENTHESIS)
5633 {
5634 *errorcodeptr = ERR39;
5635 goto FAILED;
5636 }
5637 if (n > 255)
5638 {
5639 *errorcodeptr = ERR38;
5640 goto FAILED;
5641 }
5642 *code++ = n;
5643 PUT(code, 0, (int)(ptr - cd->start_pattern + 1)); /* Pattern offset */
5644 PUT(code, LINK_SIZE, 0); /* Default length */
5645 code += 2 * LINK_SIZE;
5646 }
5647 previous = NULL;
5648 continue;
5649
5650
5651 /* ------------------------------------------------------------ */
5652 case CHAR_P: /* Python-style named subpattern handling */
5653 if (*(++ptr) == CHAR_EQUALS_SIGN ||
5654 *ptr == CHAR_GREATER_THAN_SIGN) /* Reference or recursion */
5655 {
5656 is_recurse = *ptr == CHAR_GREATER_THAN_SIGN;
5657 terminator = CHAR_RIGHT_PARENTHESIS;
5658 goto NAMED_REF_OR_RECURSE;
5659 }
5660 else if (*ptr != CHAR_LESS_THAN_SIGN) /* Test for Python-style defn */
5661 {
5662 *errorcodeptr = ERR41;
5663 goto FAILED;
5664 }
5665 /* Fall through to handle (?P< as (?< is handled */
5666
5667
5668 /* ------------------------------------------------------------ */
5669 DEFINE_NAME: /* Come here from (?< handling */
5670 case CHAR_APOSTROPHE:
5671 {
5672 terminator = (*ptr == CHAR_LESS_THAN_SIGN)?
5673 CHAR_GREATER_THAN_SIGN : CHAR_APOSTROPHE;
5674 name = ++ptr;
5675
5676 while ((cd->ctypes[*ptr] & ctype_word) != 0) ptr++;
5677 namelen = (int)(ptr - name);
5678
5679 /* In the pre-compile phase, just do a syntax check. */
5680
5681 if (lengthptr != NULL)
5682 {
5683 if (*ptr != terminator)
5684 {
5685 *errorcodeptr = ERR42;
5686 goto FAILED;
5687 }
5688 if (cd->names_found >= MAX_NAME_COUNT)
5689 {
5690 *errorcodeptr = ERR49;
5691 goto FAILED;
5692 }
5693 if (namelen + 3 > cd->name_entry_size)
5694 {
5695 cd->name_entry_size = namelen + 3;
5696 if (namelen > MAX_NAME_SIZE)
5697 {
5698 *errorcodeptr = ERR48;
5699 goto FAILED;
5700 }
5701 }
5702 }
5703
5704 /* In the real compile, create the entry in the table, maintaining
5705 alphabetical order. Duplicate names for different numbers are
5706 permitted only if PCRE_DUPNAMES is set. Duplicate names for the same
5707 number are always OK. (An existing number can be re-used if (?|
5708 appears in the pattern.) In either event, a duplicate name results in
5709 a duplicate entry in the table, even if the number is the same. This
5710 is because the number of names, and hence the table size, is computed
5711 in the pre-compile, and it affects various numbers and pointers which
5712 would all have to be modified, and the compiled code moved down, if
5713 duplicates with the same number were omitted from the table. This
5714 doesn't seem worth the hassle. However, *different* names for the
5715 same number are not permitted. */
5716
5717 else
5718 {
5719 BOOL dupname = FALSE;
5720 slot = cd->name_table;
5721
5722 for (i = 0; i < cd->names_found; i++)
5723 {
5724 int crc = memcmp(name, slot+2, namelen);
5725 if (crc == 0)
5726 {
5727 if (slot[2+namelen] == 0)
5728 {
5729 if (GET2(slot, 0) != cd->bracount + 1 &&
5730 (options & PCRE_DUPNAMES) == 0)
5731 {
5732 *errorcodeptr = ERR43;
5733 goto FAILED;
5734 }
5735 else dupname = TRUE;
5736 }
5737 else crc = -1; /* Current name is a substring */
5738 }
5739
5740 /* Make space in the table and break the loop for an earlier
5741 name. For a duplicate or later name, carry on. We do this for
5742 duplicates so that in the simple case (when ?(| is not used) they
5743 are in order of their numbers. */
5744
5745 if (crc < 0)
5746 {
5747 memmove(slot + cd->name_entry_size, slot,
5748 (cd->names_found - i) * cd->name_entry_size);
5749 break;
5750 }
5751
5752 /* Continue the loop for a later or duplicate name */
5753
5754 slot += cd->name_entry_size;
5755 }
5756
5757 /* For non-duplicate names, check for a duplicate number before
5758 adding the new name. */
5759
5760 if (!dupname)
5761 {
5762 pcre_uchar *cslot = cd->name_table;
5763 for (i = 0; i < cd->names_found; i++)
5764 {
5765 if (cslot != slot)
5766 {
5767 if (GET2(cslot, 0) == cd->bracount + 1)
5768 {
5769 *errorcodeptr = ERR65;
5770 goto FAILED;
5771 }
5772 }
5773 else i--;
5774 cslot += cd->name_entry_size;
5775 }
5776 }
5777
5778 PUT2(slot, 0, cd->bracount + 1);
5779 memcpy(slot + 2, name, namelen);
5780 slot[2+namelen] = 0;
5781 }
5782 }
5783
5784 /* In both pre-compile and compile, count the number of names we've
5785 encountered. */
5786
5787 cd->names_found++;
5788 ptr++; /* Move past > or ' */
5789 goto NUMBERED_GROUP;
5790
5791
5792 /* ------------------------------------------------------------ */
5793 case CHAR_AMPERSAND: /* Perl recursion/subroutine syntax */
5794 terminator = CHAR_RIGHT_PARENTHESIS;
5795 is_recurse = TRUE;
5796 /* Fall through */
5797
5798 /* We come here from the Python syntax above that handles both
5799 references (?P=name) and recursion (?P>name), as well as falling
5800 through from the Perl recursion syntax (?&name). We also come here from
5801 the Perl \k<name> or \k'name' back reference syntax and the \k{name}
5802 .NET syntax, and the Oniguruma \g<...> and \g'...' subroutine syntax. */
5803
5804 NAMED_REF_OR_RECURSE:
5805 name = ++ptr;
5806 while ((cd->ctypes[*ptr] & ctype_word) != 0) ptr++;
5807 namelen = (int)(ptr - name);
5808
5809 /* In the pre-compile phase, do a syntax check. We used to just set
5810 a dummy reference number, because it was not used in the first pass.
5811 However, with the change of recursive back references to be atomic,
5812 we have to look for the number so that this state can be identified, as
5813 otherwise the incorrect length is computed. If it's not a backwards
5814 reference, the dummy number will do. */
5815
5816 if (lengthptr != NULL)
5817 {
5818 const pcre_uchar *temp;
5819
5820 if (namelen == 0)
5821 {
5822 *errorcodeptr = ERR62;
5823 goto FAILED;
5824 }
5825 if (*ptr != terminator)
5826 {
5827 *errorcodeptr = ERR42;
5828 goto FAILED;
5829 }
5830 if (namelen > MAX_NAME_SIZE)
5831 {
5832 *errorcodeptr = ERR48;
5833 goto FAILED;
5834 }
5835
5836 /* The name table does not exist in the first pass, so we cannot
5837 do a simple search as in the code below. Instead, we have to scan the
5838 pattern to find the number. It is important that we scan it only as
5839 far as we have got because the syntax of named subpatterns has not
5840 been checked for the rest of the pattern, and find_parens() assumes
5841 correct syntax. In any case, it's a waste of resources to scan
5842 further. We stop the scan at the current point by temporarily
5843 adjusting the value of cd->endpattern. */
5844
5845 temp = cd->end_pattern;
5846 cd->end_pattern = ptr;
5847 recno = find_parens(cd, name, namelen,
5848 (options & PCRE_EXTENDED) != 0, utf8);
5849 cd->end_pattern = temp;
5850 if (recno < 0) recno = 0; /* Forward ref; set dummy number */
5851 }
5852
5853 /* In the real compile, seek the name in the table. We check the name
5854 first, and then check that we have reached the end of the name in the
5855 table. That way, if the name that is longer than any in the table,
5856 the comparison will fail without reading beyond the table entry. */
5857
5858 else
5859 {
5860 slot = cd->name_table;
5861 for (i = 0; i < cd->names_found; i++)
5862 {
5863 if (strncmp((char *)name, (char *)slot+2, namelen) == 0 &&
5864 slot[2+namelen] == 0)
5865 break;
5866 slot += cd->name_entry_size;
5867 }
5868
5869 if (i < cd->names_found) /* Back reference */
5870 {
5871 recno = GET2(slot, 0);
5872 }
5873 else if ((recno = /* Forward back reference */
5874 find_parens(cd, name, namelen,
5875 (options & PCRE_EXTENDED) != 0, utf8)) <= 0)
5876 {
5877 *errorcodeptr = ERR15;
5878 goto FAILED;
5879 }
5880 }
5881
5882 /* In both phases, we can now go to the code than handles numerical
5883 recursion or backreferences. */
5884
5885 if (is_recurse) goto HANDLE_RECURSION;
5886 else goto HANDLE_REFERENCE;
5887
5888
5889 /* ------------------------------------------------------------ */
5890 case CHAR_R: /* Recursion */
5891 ptr++; /* Same as (?0) */
5892 /* Fall through */
5893
5894
5895 /* ------------------------------------------------------------ */
5896 case CHAR_MINUS: case CHAR_PLUS: /* Recursion or subroutine */
5897 case CHAR_0: case CHAR_1: case CHAR_2: case CHAR_3: case CHAR_4:
5898 case CHAR_5: case CHAR_6: case CHAR_7: case CHAR_8: case CHAR_9:
5899 {
5900 const pcre_uchar *called;
5901 terminator = CHAR_RIGHT_PARENTHESIS;
5902
5903 /* Come here from the \g<...> and \g'...' code (Oniguruma
5904 compatibility). However, the syntax has been checked to ensure that
5905 the ... are a (signed) number, so that neither ERR63 nor ERR29 will
5906 be called on this path, nor with the jump to OTHER_CHAR_AFTER_QUERY
5907 ever be taken. */
5908
5909 HANDLE_NUMERICAL_RECURSION:
5910
5911 if ((refsign = *ptr) == CHAR_PLUS)
5912 {
5913 ptr++;
5914 if ((digitab[*ptr] & ctype_digit) == 0)
5915 {
5916 *errorcodeptr = ERR63;
5917 goto FAILED;
5918 }
5919 }
5920 else if (refsign == CHAR_MINUS)
5921 {
5922 if ((digitab[ptr[1]] & ctype_digit) == 0)
5923 goto OTHER_CHAR_AFTER_QUERY;
5924 ptr++;
5925 }
5926
5927 recno = 0;
5928 while((digitab[*ptr] & ctype_digit) != 0)
5929 recno = recno * 10 + *ptr++ - CHAR_0;
5930
5931 if (*ptr != terminator)
5932 {
5933 *errorcodeptr = ERR29;
5934 goto FAILED;
5935 }
5936
5937 if (refsign == CHAR_MINUS)
5938 {
5939 if (recno == 0)
5940 {
5941 *errorcodeptr = ERR58;
5942 goto FAILED;
5943 }
5944 recno = cd->bracount - recno + 1;
5945 if (recno <= 0)
5946 {
5947 *errorcodeptr = ERR15;
5948 goto FAILED;
5949 }
5950 }
5951 else if (refsign == CHAR_PLUS)
5952 {
5953 if (recno == 0)
5954 {
5955 *errorcodeptr = ERR58;
5956 goto FAILED;
5957 }
5958 recno += cd->bracount;
5959 }
5960
5961 /* Come here from code above that handles a named recursion */
5962
5963 HANDLE_RECURSION:
5964
5965 previous = code;
5966 called = cd->start_code;
5967
5968 /* When we are actually compiling, find the bracket that is being
5969 referenced. Temporarily end the regex in case it doesn't exist before
5970 this point. If we end up with a forward reference, first check that
5971 the bracket does occur later so we can give the error (and position)
5972 now. Then remember this forward reference in the workspace so it can
5973 be filled in at the end. */
5974
5975 if (lengthptr == NULL)
5976 {
5977 *code = OP_END;
5978 if (recno != 0)
5979 called = _pcre_find_bracket(cd->start_code, utf8, recno);
5980
5981 /* Forward reference */
5982
5983 if (called == NULL)
5984 {
5985 if (find_parens(cd, NULL, recno,
5986 (options & PCRE_EXTENDED) != 0, utf8) < 0)
5987 {
5988 *errorcodeptr = ERR15;
5989 goto FAILED;
5990 }
5991
5992 /* Fudge the value of "called" so that when it is inserted as an
5993 offset below, what it actually inserted is the reference number
5994 of the group. Then remember the forward reference. */
5995
5996 called = cd->start_code + recno;
5997 PUTINC(cd->hwm, 0, (int)(code + 1 - cd->start_code));
5998 }
5999
6000 /* If not a forward reference, and the subpattern is still open,
6001 this is a recursive call. We check to see if this is a left
6002 recursion that could loop for ever, and diagnose that case. We
6003 must not, however, do this check if we are in a conditional
6004 subpattern because the condition might be testing for recursion in
6005 a pattern such as /(?(R)a+|(?R)b)/, which is perfectly valid.
6006 Forever loops are also detected at runtime, so those that occur in
6007 conditional subpatterns will be picked up then. */
6008
6009 else if (GET(called, 1) == 0 && cond_depth <= 0 &&
6010 could_be_empty(called, code, bcptr, utf8, cd))
6011 {
6012 *errorcodeptr = ERR40;
6013 goto FAILED;
6014 }
6015 }
6016
6017 /* Insert the recursion/subroutine item. */
6018
6019 *code = OP_RECURSE;
6020 PUT(code, 1, (int)(called - cd->start_code));
6021 code += 1 + LINK_SIZE;
6022 }
6023
6024 /* Can't determine a first byte now */
6025
6026 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
6027 continue;
6028
6029
6030 /* ------------------------------------------------------------ */
6031 default: /* Other characters: check option setting */
6032 OTHER_CHAR_AFTER_QUERY:
6033 set = unset = 0;
6034 optset = &set;
6035
6036 while (*ptr != CHAR_RIGHT_PARENTHESIS && *ptr != CHAR_COLON)
6037 {
6038 switch (*ptr++)
6039 {
6040 case CHAR_MINUS: optset = &unset; break;
6041
6042 case CHAR_J: /* Record that it changed in the external options */
6043 *optset |= PCRE_DUPNAMES;
6044 cd->external_flags |= PCRE_JCHANGED;
6045 break;
6046
6047 case CHAR_i: *optset |= PCRE_CASELESS; break;
6048 case CHAR_m: *optset |= PCRE_MULTILINE; break;
6049 case CHAR_s: *optset |= PCRE_DOTALL; break;
6050 case CHAR_x: *optset |= PCRE_EXTENDED; break;
6051 case CHAR_U: *optset |= PCRE_UNGREEDY; break;
6052 case CHAR_X: *optset |= PCRE_EXTRA; break;
6053
6054 default: *errorcodeptr = ERR12;
6055 ptr--; /* Correct the offset */
6056 goto FAILED;
6057 }
6058 }
6059
6060 /* Set up the changed option bits, but don't change anything yet. */
6061
6062 newoptions = (options | set) & (~unset);
6063
6064 /* If the options ended with ')' this is not the start of a nested
6065 group with option changes, so the options change at this level. If this
6066 item is right at the start of the pattern, the options can be
6067 abstracted and made external in the pre-compile phase, and ignored in
6068 the compile phase. This can be helpful when matching -- for instance in
6069 caseless checking of required bytes.
6070
6071 If the code pointer is not (cd->start_code + 1 + LINK_SIZE), we are
6072 definitely *not* at the start of the pattern because something has been
6073 compiled. In the pre-compile phase, however, the code pointer can have
6074 that value after the start, because it gets reset as code is discarded
6075 during the pre-compile. However, this can happen only at top level - if
6076 we are within parentheses, the starting BRA will still be present. At
6077 any parenthesis level, the length value can be used to test if anything
6078 has been compiled at that level. Thus, a test for both these conditions
6079 is necessary to ensure we correctly detect the start of the pattern in
6080 both phases.
6081
6082 If we are not at the pattern start, reset the greedy defaults and the
6083 case value for firstbyte and reqbyte. */
6084
6085 if (*ptr == CHAR_RIGHT_PARENTHESIS)
6086 {
6087 if (code == cd->start_code + 1 + LINK_SIZE &&
6088 (lengthptr == NULL || *lengthptr == 2 + 2*LINK_SIZE))
6089 {
6090 cd->external_options = newoptions;
6091 }
6092 else
6093 {
6094 greedy_default = ((newoptions & PCRE_UNGREEDY) != 0);
6095 greedy_non_default = greedy_default ^ 1;
6096 req_caseopt = ((newoptions & PCRE_CASELESS) != 0)? REQ_CASELESS : 0;
6097 }
6098
6099 /* Change options at this level, and pass them back for use
6100 in subsequent branches. */
6101
6102 *optionsptr = options = newoptions;
6103 previous = NULL; /* This item can't be repeated */
6104 continue; /* It is complete */
6105 }
6106
6107 /* If the options ended with ':' we are heading into a nested group
6108 with possible change of options. Such groups are non-capturing and are
6109 not assertions of any kind. All we need to do is skip over the ':';
6110 the newoptions value is handled below. */
6111
6112 bravalue = OP_BRA;
6113 ptr++;
6114 } /* End of switch for character following (? */
6115 } /* End of (? handling */
6116
6117 /* Opening parenthesis not followed by '*' or '?'. If PCRE_NO_AUTO_CAPTURE
6118 is set, all unadorned brackets become non-capturing and behave like (?:...)
6119 brackets. */
6120
6121 else if ((options & PCRE_NO_AUTO_CAPTURE) != 0)
6122 {
6123 bravalue = OP_BRA;
6124 }
6125
6126 /* Else we have a capturing group. */
6127
6128 else
6129 {
6130 NUMBERED_GROUP:
6131 cd->bracount += 1;
6132 PUT2(code, 1+LINK_SIZE, cd->bracount);
6133 skipbytes = 2;
6134 }
6135
6136 /* Process nested bracketed regex. Assertions used not to be repeatable,
6137 but this was changed for Perl compatibility, so all kinds can now be
6138 repeated. We copy code into a non-register variable (tempcode) in order to
6139 be able to pass its address because some compilers complain otherwise. */
6140
6141 previous = code; /* For handling repetition */
6142 *code = bravalue;
6143 tempcode = code;
6144 tempreqvary = cd->req_varyopt; /* Save value before bracket */
6145 tempbracount = cd->bracount; /* Save value before bracket */
6146 length_prevgroup = 0; /* Initialize for pre-compile phase */
6147
6148 if (!compile_regex(
6149 newoptions, /* The complete new option state */
6150 &tempcode, /* Where to put code (updated) */
6151 &ptr, /* Input pointer (updated) */
6152 errorcodeptr, /* Where to put an error message */
6153 (bravalue == OP_ASSERTBACK ||
6154 bravalue == OP_ASSERTBACK_NOT), /* TRUE if back assert */
6155 reset_bracount, /* True if (?| group */
6156 skipbytes, /* Skip over bracket number */
6157 cond_depth +
6158 ((bravalue == OP_COND)?1:0), /* Depth of condition subpatterns */
6159 &subfirstbyte, /* For possible first char */
6160 &subreqbyte, /* For possible last char */
6161 bcptr, /* Current branch chain */
6162 cd, /* Tables block */
6163 (lengthptr == NULL)? NULL : /* Actual compile phase */
6164 &length_prevgroup /* Pre-compile phase */
6165 ))
6166 goto FAILED;
6167
6168 /* If this was an atomic group and there are no capturing groups within it,
6169 generate OP_ONCE_NC instead of OP_ONCE. */
6170
6171 if (bravalue == OP_ONCE && cd->bracount <= tempbracount)
6172 *code = OP_ONCE_NC;
6173
6174 if (bravalue >= OP_ASSERT && bravalue <= OP_ASSERTBACK_NOT)
6175 cd->assert_depth -= 1;
6176
6177 /* At the end of compiling, code is still pointing to the start of the
6178 group, while tempcode has been updated to point past the end of the group.
6179 The pattern pointer (ptr) is on the bracket.
6180
6181 If this is a conditional bracket, check that there are no more than
6182 two branches in the group, or just one if it's a DEFINE group. We do this
6183 in the real compile phase, not in the pre-pass, where the whole group may
6184 not be available. */
6185
6186 if (bravalue == OP_COND && lengthptr == NULL)
6187 {
6188 pcre_uchar *tc = code;
6189 int condcount = 0;
6190
6191 do {
6192 condcount++;
6193 tc += GET(tc,1);
6194 }
6195 while (*tc != OP_KET);
6196
6197 /* A DEFINE group is never obeyed inline (the "condition" is always
6198 false). It must have only one branch. */
6199
6200 if (code[LINK_SIZE+1] == OP_DEF)
6201 {
6202 if (condcount > 1)
6203 {
6204 *errorcodeptr = ERR54;
6205 goto FAILED;
6206 }
6207 bravalue = OP_DEF; /* Just a flag to suppress char handling below */
6208 }
6209
6210 /* A "normal" conditional group. If there is just one branch, we must not
6211 make use of its firstbyte or reqbyte, because this is equivalent to an
6212 empty second branch. */
6213
6214 else
6215 {
6216 if (condcount > 2)
6217 {
6218 *errorcodeptr = ERR27;
6219 goto FAILED;
6220 }
6221 if (condcount == 1) subfirstbyte = subreqbyte = REQ_NONE;
6222 }
6223 }
6224
6225 /* Error if hit end of pattern */
6226
6227 if (*ptr != CHAR_RIGHT_PARENTHESIS)
6228 {
6229 *errorcodeptr = ERR14;
6230 goto FAILED;
6231 }
6232
6233 /* In the pre-compile phase, update the length by the length of the group,
6234 less the brackets at either end. Then reduce the compiled code to just a
6235 set of non-capturing brackets so that it doesn't use much memory if it is
6236 duplicated by a quantifier.*/
6237
6238 if (lengthptr != NULL)
6239 {
6240 if (OFLOW_MAX - *lengthptr < length_prevgroup - 2 - 2*LINK_SIZE)
6241 {
6242 *errorcodeptr = ERR20;
6243 goto FAILED;
6244 }
6245 *lengthptr += length_prevgroup - 2 - 2*LINK_SIZE;
6246 code++; /* This already contains bravalue */
6247 PUTINC(code, 0, 1 + LINK_SIZE);
6248 *code++ = OP_KET;
6249 PUTINC(code, 0, 1 + LINK_SIZE);
6250 break; /* No need to waste time with special character handling */
6251 }
6252
6253 /* Otherwise update the main code pointer to the end of the group. */
6254
6255 code = tempcode;
6256
6257 /* For a DEFINE group, required and first character settings are not
6258 relevant. */
6259
6260 if (bravalue == OP_DEF) break;
6261
6262 /* Handle updating of the required and first characters for other types of
6263 group. Update for normal brackets of all kinds, and conditions with two
6264 branches (see code above). If the bracket is followed by a quantifier with
6265 zero repeat, we have to back off. Hence the definition of zeroreqbyte and
6266 zerofirstbyte outside the main loop so that they can be accessed for the
6267 back off. */
6268
6269 zeroreqbyte = reqbyte;
6270 zerofirstbyte = firstbyte;
6271 groupsetfirstbyte = FALSE;
6272
6273 if (bravalue >= OP_ONCE)
6274 {
6275 /* If we have not yet set a firstbyte in this branch, take it from the
6276 subpattern, remembering that it was set here so that a repeat of more
6277 than one can replicate it as reqbyte if necessary. If the subpattern has
6278 no firstbyte, set "none" for the whole branch. In both cases, a zero
6279 repeat forces firstbyte to "none". */
6280
6281 if (firstbyte == REQ_UNSET)
6282 {
6283 if (subfirstbyte >= 0)
6284 {
6285 firstbyte = subfirstbyte;
6286 groupsetfirstbyte = TRUE;
6287 }
6288 else firstbyte = REQ_NONE;
6289 zerofirstbyte = REQ_NONE;
6290 }
6291
6292 /* If firstbyte was previously set, convert the subpattern's firstbyte
6293 into reqbyte if there wasn't one, using the vary flag that was in
6294 existence beforehand. */
6295
6296 else if (subfirstbyte >= 0 && subreqbyte < 0)
6297 subreqbyte = subfirstbyte | tempreqvary;
6298
6299 /* If the subpattern set a required byte (or set a first byte that isn't
6300 really the first byte - see above), set it. */
6301
6302 if (subreqbyte >= 0) reqbyte = subreqbyte;
6303 }
6304
6305 /* For a forward assertion, we take the reqbyte, if set. This can be
6306 helpful if the pattern that follows the assertion doesn't set a different
6307 char. For example, it's useful for /(?=abcde).+/. We can't set firstbyte
6308 for an assertion, however because it leads to incorrect effect for patterns
6309 such as /(?=a)a.+/ when the "real" "a" would then become a reqbyte instead
6310 of a firstbyte. This is overcome by a scan at the end if there's no
6311 firstbyte, looking for an asserted first char. */
6312
6313 else if (bravalue == OP_ASSERT && subreqbyte >= 0) reqbyte = subreqbyte;
6314 break; /* End of processing '(' */
6315
6316
6317 /* ===================================================================*/
6318 /* Handle metasequences introduced by \. For ones like \d, the ESC_ values
6319 are arranged to be the negation of the corresponding OP_values in the
6320 default case when PCRE_UCP is not set. For the back references, the values
6321 are ESC_REF plus the reference number. Only back references and those types
6322 that consume a character may be repeated. We can test for values between
6323 ESC_b and ESC_Z for the latter; this may have to change if any new ones are
6324 ever created. */
6325
6326 case CHAR_BACKSLASH:
6327 tempptr = ptr;
6328 c = check_escape(&ptr, errorcodeptr, cd->bracount, options, FALSE);
6329 if (*errorcodeptr != 0) goto FAILED;
6330
6331 if (c < 0)
6332 {
6333 if (-c == ESC_Q) /* Handle start of quoted string */
6334 {
6335 if (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
6336 ptr += 2; /* avoid empty string */
6337 else inescq = TRUE;
6338 continue;
6339 }
6340
6341 if (-c == ESC_E) continue; /* Perl ignores an orphan \E */
6342
6343 /* For metasequences that actually match a character, we disable the
6344 setting of a first character if it hasn't already been set. */
6345
6346 if (firstbyte == REQ_UNSET && -c > ESC_b && -c < ESC_Z)
6347 firstbyte = REQ_NONE;
6348
6349 /* Set values to reset to if this is followed by a zero repeat. */
6350
6351 zerofirstbyte = firstbyte;
6352 zeroreqbyte = reqbyte;
6353
6354 /* \g<name> or \g'name' is a subroutine call by name and \g<n> or \g'n'
6355 is a subroutine call by number (Oniguruma syntax). In fact, the value
6356 -ESC_g is returned only for these cases. So we don't need to check for <
6357 or ' if the value is -ESC_g. For the Perl syntax \g{n} the value is
6358 -ESC_REF+n, and for the Perl syntax \g{name} the result is -ESC_k (as
6359 that is a synonym for a named back reference). */
6360
6361 if (-c == ESC_g)
6362 {
6363 const pcre_uchar *p;
6364 save_hwm = cd->hwm; /* Normally this is set when '(' is read */
6365 terminator = (*(++ptr) == CHAR_LESS_THAN_SIGN)?
6366 CHAR_GREATER_THAN_SIGN : CHAR_APOSTROPHE;
6367
6368 /* These two statements stop the compiler for warning about possibly
6369 unset variables caused by the jump to HANDLE_NUMERICAL_RECURSION. In
6370 fact, because we actually check for a number below, the paths that
6371 would actually be in error are never taken. */
6372
6373 skipbytes = 0;
6374 reset_bracount = FALSE;
6375
6376 /* Test for a name */
6377
6378 if (ptr[1] != CHAR_PLUS && ptr[1] != CHAR_MINUS)
6379 {
6380 BOOL isnumber = TRUE;
6381 for (p = ptr + 1; *p != 0 && *p != terminator; p++)
6382 {
6383 if ((cd->ctypes[*p] & ctype_digit) == 0) isnumber = FALSE;
6384 if ((cd->ctypes[*p] & ctype_word) == 0) break;
6385 }
6386 if (*p != terminator)
6387 {
6388 *errorcodeptr = ERR57;
6389 break;
6390 }
6391 if (isnumber)
6392 {
6393 ptr++;
6394 goto HANDLE_NUMERICAL_RECURSION;
6395 }
6396 is_recurse = TRUE;
6397 goto NAMED_REF_OR_RECURSE;
6398 }
6399
6400 /* Test a signed number in angle brackets or quotes. */
6401
6402 p = ptr + 2;
6403 while ((digitab[*p] & ctype_digit) != 0) p++;
6404 if (*p != terminator)
6405 {
6406 *errorcodeptr = ERR57;
6407 break;
6408 }
6409 ptr++;
6410 goto HANDLE_NUMERICAL_RECURSION;
6411 }
6412
6413 /* \k<name> or \k'name' is a back reference by name (Perl syntax).
6414 We also support \k{name} (.NET syntax). */
6415
6416 if (-c == ESC_k)
6417 {
6418 if ((ptr[1] != CHAR_LESS_THAN_SIGN &&
6419 ptr[1] != CHAR_APOSTROPHE && ptr[1] != CHAR_LEFT_CURLY_BRACKET))
6420 {
6421 *errorcodeptr = ERR69;
6422 break;
6423 }
6424 is_recurse = FALSE;
6425 terminator = (*(++ptr) == CHAR_LESS_THAN_SIGN)?
6426 CHAR_GREATER_THAN_SIGN : (*ptr == CHAR_APOSTROPHE)?
6427 CHAR_APOSTROPHE : CHAR_RIGHT_CURLY_BRACKET;
6428 goto NAMED_REF_OR_RECURSE;
6429 }
6430
6431 /* Back references are handled specially; must disable firstbyte if
6432 not set to cope with cases like (?=(\w+))\1: which would otherwise set
6433 ':' later. */
6434
6435 if (-c >= ESC_REF)
6436 {
6437 open_capitem *oc;
6438 recno = -c - ESC_REF;
6439
6440 HANDLE_REFERENCE: /* Come here from named backref handling */
6441 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
6442 previous = code;
6443 *code++ = ((options & PCRE_CASELESS) != 0)? OP_REFI : OP_REF;
6444 PUT2INC(code, 0, recno);
6445 cd->backref_map |= (recno < 32)? (1 << recno) : 1;
6446 if (recno > cd->top_backref) cd->top_backref = recno;
6447
6448 /* Check to see if this back reference is recursive, that it, it
6449 is inside the group that it references. A flag is set so that the
6450 group can be made atomic. */
6451
6452 for (oc = cd->open_caps; oc != NULL; oc = oc->next)
6453 {
6454 if (oc->number == recno)
6455 {
6456 oc->flag = TRUE;
6457 break;
6458 }
6459 }
6460 }
6461
6462 /* So are Unicode property matches, if supported. */
6463
6464 #ifdef SUPPORT_UCP
6465 else if (-c == ESC_P || -c == ESC_p)
6466 {
6467 BOOL negated;
6468 int pdata;
6469 int ptype = get_ucp(&ptr, &negated, &pdata, errorcodeptr);
6470 if (ptype < 0) goto FAILED;
6471 previous = code;
6472 *code++ = ((-c == ESC_p) != negated)? OP_PROP : OP_NOTPROP;
6473 *code++ = ptype;
6474 *code++ = pdata;
6475 }
6476 #else
6477
6478 /* If Unicode properties are not supported, \X, \P, and \p are not
6479 allowed. */
6480
6481 else if (-c == ESC_X || -c == ESC_P || -c == ESC_p)
6482 {
6483 *errorcodeptr = ERR45;
6484 goto FAILED;
6485 }
6486 #endif
6487
6488 /* For the rest (including \X when Unicode properties are supported), we
6489 can obtain the OP value by negating the escape value in the default
6490 situation when PCRE_UCP is not set. When it *is* set, we substitute
6491 Unicode property tests. */
6492
6493 else
6494 {
6495 #ifdef SUPPORT_UCP
6496 if (-c >= ESC_DU && -c <= ESC_wu)
6497 {
6498 nestptr = ptr + 1; /* Where to resume */
6499 ptr = substitutes[-c - ESC_DU] - 1; /* Just before substitute */
6500 }
6501 else
6502 #endif
6503 /* In non-UTF-8 mode, we turn \C into OP_ALLANY instead of OP_ANYBYTE
6504 so that it works in DFA mode and in lookbehinds. */
6505
6506 {
6507 previous = (-c > ESC_b && -c < ESC_Z)? code : NULL;
6508 *code++ = (!utf8 && c == -ESC_C)? OP_ALLANY : -c;
6509 }
6510 }
6511 continue;
6512 }
6513
6514 /* We have a data character whose value is in c. In UTF-8 mode it may have
6515 a value > 127. We set its representation in the length/buffer, and then
6516 handle it as a data character. */
6517
6518 #ifdef SUPPORT_UTF8
6519 if (utf8 && c > 127)
6520 mclength = _pcre_ord2utf8(c, mcbuffer);
6521 else
6522 #endif
6523
6524 {
6525 mcbuffer[0] = c;
6526 mclength = 1;
6527 }
6528 goto ONE_CHAR;
6529
6530
6531 /* ===================================================================*/
6532 /* Handle a literal character. It is guaranteed not to be whitespace or #
6533 when the extended flag is set. If we are in UTF-8 mode, it may be a
6534 multi-byte literal character. */
6535
6536 default:
6537 NORMAL_CHAR:
6538 mclength = 1;
6539 mcbuffer[0] = c;
6540
6541 #ifdef SUPPORT_UTF8
6542 if (utf8 && c >= 0xc0)
6543 {
6544 while ((ptr[1] & 0xc0) == 0x80)
6545 mcbuffer[mclength++] = *(++ptr);
6546 }
6547 #endif
6548
6549 /* At this point we have the character's bytes in mcbuffer, and the length
6550 in mclength. When not in UTF-8 mode, the length is always 1. */
6551
6552 ONE_CHAR:
6553 previous = code;
6554 *code++ = ((options & PCRE_CASELESS) != 0)? OP_CHARI : OP_CHAR;
6555 for (c = 0; c < mclength; c++) *code++ = mcbuffer[c];
6556
6557 /* Remember if \r or \n were seen */
6558
6559 if (mcbuffer[0] == CHAR_CR || mcbuffer[0] == CHAR_NL)
6560 cd->external_flags |= PCRE_HASCRORLF;
6561
6562 /* Set the first and required bytes appropriately. If no previous first
6563 byte, set it from this character, but revert to none on a zero repeat.
6564 Otherwise, leave the firstbyte value alone, and don't change it on a zero
6565 repeat. */
6566
6567 if (firstbyte == REQ_UNSET)
6568 {
6569 zerofirstbyte = REQ_NONE;
6570 zeroreqbyte = reqbyte;
6571
6572 /* If the character is more than one byte long, we can set firstbyte
6573 only if it is not to be matched caselessly. */
6574
6575 if (mclength == 1 || req_caseopt == 0)
6576 {
6577 firstbyte = mcbuffer[0] | req_caseopt;
6578 if (mclength != 1) reqbyte = code[-1] | cd->req_varyopt;
6579 }
6580 else firstbyte = reqbyte = REQ_NONE;
6581 }
6582
6583 /* firstbyte was previously set; we can set reqbyte only if the length is
6584 1 or the matching is caseful. */
6585
6586 else
6587 {
6588 zerofirstbyte = firstbyte;
6589 zeroreqbyte = reqbyte;
6590 if (mclength == 1 || req_caseopt == 0)
6591 reqbyte = code[-1] | req_caseopt | cd->req_varyopt;
6592 }
6593
6594 break; /* End of literal character handling */
6595 }
6596 } /* end of big loop */
6597
6598
6599 /* Control never reaches here by falling through, only by a goto for all the
6600 error states. Pass back the position in the pattern so that it can be displayed
6601 to the user for diagnosing the error. */
6602
6603 FAILED:
6604 *ptrptr = ptr;
6605 return FALSE;
6606 }
6607
6608
6609
6610
6611 /*************************************************
6612 * Compile sequence of alternatives *
6613 *************************************************/
6614
6615 /* On entry, ptr is pointing past the bracket character, but on return it
6616 points to the closing bracket, or vertical bar, or end of string. The code
6617 variable is pointing at the byte into which the BRA operator has been stored.
6618 This function is used during the pre-compile phase when we are trying to find
6619 out the amount of memory needed, as well as during the real compile phase. The
6620 value of lengthptr distinguishes the two phases.
6621
6622 Arguments:
6623 options option bits, including any changes for this subpattern
6624 codeptr -> the address of the current code pointer
6625 ptrptr -> the address of the current pattern pointer
6626 errorcodeptr -> pointer to error code variable
6627 lookbehind TRUE if this is a lookbehind assertion
6628 reset_bracount TRUE to reset the count for each branch
6629 skipbytes skip this many bytes at start (for brackets and OP_COND)
6630 cond_depth depth of nesting for conditional subpatterns
6631 firstbyteptr place to put the first required character, or a negative number
6632 reqbyteptr place to put the last required character, or a negative number
6633 bcptr pointer to the chain of currently open branches
6634 cd points to the data block with tables pointers etc.
6635 lengthptr NULL during the real compile phase
6636 points to length accumulator during pre-compile phase
6637
6638 Returns: TRUE on success
6639 */
6640
6641 static BOOL
6642 compile_regex(int options, pcre_u