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

Contents of /code/trunk/pcre_compile.c

Parent Directory Parent Directory | Revision Log Revision Log


Revision 1060 - (show annotations)
Tue Oct 16 15:53:57 2012 UTC (7 years ago) by chpe
File MIME type: text/plain
File size: 272262 byte(s)
Error occurred while calculating annotation data.
pcre32: compile: Use uint32 in check_escape

Use pcre_uint32 for the data character in check_escape(), so that it
correctly parses \x{} with any 32-bit hex value in 32-bit non-UTF-32
mode.
1 /*************************************************
2 * Perl-Compatible Regular Expressions *
3 *************************************************/
4
5 /* PCRE is a library of functions to support regular expressions whose syntax
6 and semantics are as close as possible to those of the Perl 5 language.
7
8 Written by Philip Hazel
9 Copyright (c) 1997-2012 University of Cambridge
10
11 -----------------------------------------------------------------------------
12 Redistribution and use in source and binary forms, with or without
13 modification, are permitted provided that the following conditions are met:
14
15 * Redistributions of source code must retain the above copyright notice,
16 this list of conditions and the following disclaimer.
17
18 * Redistributions in binary form must reproduce the above copyright
19 notice, this list of conditions and the following disclaimer in the
20 documentation and/or other materials provided with the distribution.
21
22 * Neither the name of the University of Cambridge nor the names of its
23 contributors may be used to endorse or promote products derived from
24 this software without specific prior written permission.
25
26 THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
27 AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28 IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29 ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE
30 LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
31 CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
32 SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
33 INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
34 CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
35 ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
36 POSSIBILITY OF SUCH DAMAGE.
37 -----------------------------------------------------------------------------
38 */
39
40
41 /* This module contains the external function pcre_compile(), along with
42 supporting internal functions that are not used by other modules. */
43
44
45 #ifdef HAVE_CONFIG_H
46 #include "config.h"
47 #endif
48
49 #define NLBLOCK cd /* Block containing newline information */
50 #define PSSTART start_pattern /* Field containing processed string start */
51 #define PSEND end_pattern /* Field containing processed string end */
52
53 #include "pcre_internal.h"
54
55
56 /* When PCRE_DEBUG is defined, we need the pcre(16|32)_printint() function, which
57 is also used by pcretest. PCRE_DEBUG is not defined when building a production
58 library. We do not need to select pcre16_printint.c specially, because the
59 COMPILE_PCREx macro will already be appropriately set. */
60
61 #ifdef PCRE_DEBUG
62 /* pcre_printint.c should not include any headers */
63 #define PCRE_INCLUDED
64 #include "pcre_printint.c"
65 #undef PCRE_INCLUDED
66 #endif
67
68
69 /* Macro for setting individual bits in class bitmaps. */
70
71 #define SETBIT(a,b) a[(b)/8] |= (1 << ((b)&7))
72
73 /* Maximum length value to check against when making sure that the integer that
74 holds the compiled pattern length does not overflow. We make it a bit less than
75 INT_MAX to allow for adding in group terminating bytes, so that we don't have
76 to check them every time. */
77
78 #define OFLOW_MAX (INT_MAX - 20)
79
80 /* Definitions to allow mutual recursion */
81
82 static int
83 add_list_to_class(pcre_uint8 *, pcre_uchar **, int, compile_data *,
84 const pcre_uint32 *, unsigned int);
85
86 static BOOL
87 compile_regex(int, pcre_uchar **, const pcre_uchar **, int *, BOOL, BOOL,
88 int, int, int *, int *, branch_chain *, compile_data *, int *);
89
90
91
92 /*************************************************
93 * Code parameters and static tables *
94 *************************************************/
95
96 /* This value specifies the size of stack workspace that is used during the
97 first pre-compile phase that determines how much memory is required. The regex
98 is partly compiled into this space, but the compiled parts are discarded as
99 soon as they can be, so that hopefully there will never be an overrun. The code
100 does, however, check for an overrun. The largest amount I've seen used is 218,
101 so this number is very generous.
102
103 The same workspace is used during the second, actual compile phase for
104 remembering forward references to groups so that they can be filled in at the
105 end. Each entry in this list occupies LINK_SIZE bytes, so even when LINK_SIZE
106 is 4 there is plenty of room for most patterns. However, the memory can get
107 filled up by repetitions of forward references, for example patterns like
108 /(?1){0,1999}(b)/, and one user did hit the limit. The code has been changed so
109 that the workspace is expanded using malloc() in this situation. The value
110 below is therefore a minimum, and we put a maximum on it for safety. The
111 minimum is now also defined in terms of LINK_SIZE so that the use of malloc()
112 kicks in at the same number of forward references in all cases. */
113
114 #define COMPILE_WORK_SIZE (2048*LINK_SIZE)
115 #define COMPILE_WORK_SIZE_MAX (100*COMPILE_WORK_SIZE)
116
117 /* The overrun tests check for a slightly smaller size so that they detect the
118 overrun before it actually does run off the end of the data block. */
119
120 #define WORK_SIZE_SAFETY_MARGIN (100)
121
122 /* Private flags added to firstchar and reqchar. */
123
124 #define REQ_CASELESS 0x10000000l /* Indicates caselessness */
125 #define REQ_VARY 0x20000000l /* Reqchar followed non-literal item */
126 #define REQ_MASK (REQ_CASELESS | REQ_VARY)
127
128 /* Repeated character flags. */
129
130 #define UTF_LENGTH 0x10000000l /* The char contains its length. */
131
132 /* Table for handling escaped characters in the range '0'-'z'. Positive returns
133 are simple data values; negative values are for special things like \d and so
134 on. Zero means further processing is needed (for things like \x), or the escape
135 is invalid. */
136
137 #ifndef EBCDIC
138
139 /* This is the "normal" table for ASCII systems or for EBCDIC systems running
140 in UTF-8 mode. */
141
142 static const short int escapes[] = {
143 0, 0,
144 0, 0,
145 0, 0,
146 0, 0,
147 0, 0,
148 CHAR_COLON, CHAR_SEMICOLON,
149 CHAR_LESS_THAN_SIGN, CHAR_EQUALS_SIGN,
150 CHAR_GREATER_THAN_SIGN, CHAR_QUESTION_MARK,
151 CHAR_COMMERCIAL_AT, -ESC_A,
152 -ESC_B, -ESC_C,
153 -ESC_D, -ESC_E,
154 0, -ESC_G,
155 -ESC_H, 0,
156 0, -ESC_K,
157 0, 0,
158 -ESC_N, 0,
159 -ESC_P, -ESC_Q,
160 -ESC_R, -ESC_S,
161 0, 0,
162 -ESC_V, -ESC_W,
163 -ESC_X, 0,
164 -ESC_Z, CHAR_LEFT_SQUARE_BRACKET,
165 CHAR_BACKSLASH, CHAR_RIGHT_SQUARE_BRACKET,
166 CHAR_CIRCUMFLEX_ACCENT, CHAR_UNDERSCORE,
167 CHAR_GRAVE_ACCENT, 7,
168 -ESC_b, 0,
169 -ESC_d, ESC_e,
170 ESC_f, 0,
171 -ESC_h, 0,
172 0, -ESC_k,
173 0, 0,
174 ESC_n, 0,
175 -ESC_p, 0,
176 ESC_r, -ESC_s,
177 ESC_tee, 0,
178 -ESC_v, -ESC_w,
179 0, 0,
180 -ESC_z
181 };
182
183 #else
184
185 /* This is the "abnormal" table for EBCDIC systems without UTF-8 support. */
186
187 static const short int escapes[] = {
188 /* 48 */ 0, 0, 0, '.', '<', '(', '+', '|',
189 /* 50 */ '&', 0, 0, 0, 0, 0, 0, 0,
190 /* 58 */ 0, 0, '!', '$', '*', ')', ';', '~',
191 /* 60 */ '-', '/', 0, 0, 0, 0, 0, 0,
192 /* 68 */ 0, 0, '|', ',', '%', '_', '>', '?',
193 /* 70 */ 0, 0, 0, 0, 0, 0, 0, 0,
194 /* 78 */ 0, '`', ':', '#', '@', '\'', '=', '"',
195 /* 80 */ 0, 7, -ESC_b, 0, -ESC_d, ESC_e, ESC_f, 0,
196 /* 88 */-ESC_h, 0, 0, '{', 0, 0, 0, 0,
197 /* 90 */ 0, 0, -ESC_k, 'l', 0, ESC_n, 0, -ESC_p,
198 /* 98 */ 0, ESC_r, 0, '}', 0, 0, 0, 0,
199 /* A0 */ 0, '~', -ESC_s, ESC_tee, 0,-ESC_v, -ESC_w, 0,
200 /* A8 */ 0,-ESC_z, 0, 0, 0, '[', 0, 0,
201 /* B0 */ 0, 0, 0, 0, 0, 0, 0, 0,
202 /* B8 */ 0, 0, 0, 0, 0, ']', '=', '-',
203 /* C0 */ '{',-ESC_A, -ESC_B, -ESC_C, -ESC_D,-ESC_E, 0, -ESC_G,
204 /* C8 */-ESC_H, 0, 0, 0, 0, 0, 0, 0,
205 /* D0 */ '}', 0, -ESC_K, 0, 0,-ESC_N, 0, -ESC_P,
206 /* D8 */-ESC_Q,-ESC_R, 0, 0, 0, 0, 0, 0,
207 /* E0 */ '\\', 0, -ESC_S, 0, 0,-ESC_V, -ESC_W, -ESC_X,
208 /* E8 */ 0,-ESC_Z, 0, 0, 0, 0, 0, 0,
209 /* F0 */ 0, 0, 0, 0, 0, 0, 0, 0,
210 /* F8 */ 0, 0, 0, 0, 0, 0, 0, 0
211 };
212 #endif
213
214
215 /* Table of special "verbs" like (*PRUNE). This is a short table, so it is
216 searched linearly. Put all the names into a single string, in order to reduce
217 the number of relocations when a shared library is dynamically linked. The
218 string is built from string macros so that it works in UTF-8 mode on EBCDIC
219 platforms. */
220
221 typedef struct verbitem {
222 int len; /* Length of verb name */
223 int op; /* Op when no arg, or -1 if arg mandatory */
224 int op_arg; /* Op when arg present, or -1 if not allowed */
225 } verbitem;
226
227 static const char verbnames[] =
228 "\0" /* Empty name is a shorthand for MARK */
229 STRING_MARK0
230 STRING_ACCEPT0
231 STRING_COMMIT0
232 STRING_F0
233 STRING_FAIL0
234 STRING_PRUNE0
235 STRING_SKIP0
236 STRING_THEN;
237
238 static const verbitem verbs[] = {
239 { 0, -1, OP_MARK },
240 { 4, -1, OP_MARK },
241 { 6, OP_ACCEPT, -1 },
242 { 6, OP_COMMIT, -1 },
243 { 1, OP_FAIL, -1 },
244 { 4, OP_FAIL, -1 },
245 { 5, OP_PRUNE, OP_PRUNE_ARG },
246 { 4, OP_SKIP, OP_SKIP_ARG },
247 { 4, OP_THEN, OP_THEN_ARG }
248 };
249
250 static const int verbcount = sizeof(verbs)/sizeof(verbitem);
251
252
253 /* Tables of names of POSIX character classes and their lengths. The names are
254 now all in a single string, to reduce the number of relocations when a shared
255 library is dynamically loaded. The list of lengths is terminated by a zero
256 length entry. The first three must be alpha, lower, upper, as this is assumed
257 for handling case independence. */
258
259 static const char posix_names[] =
260 STRING_alpha0 STRING_lower0 STRING_upper0 STRING_alnum0
261 STRING_ascii0 STRING_blank0 STRING_cntrl0 STRING_digit0
262 STRING_graph0 STRING_print0 STRING_punct0 STRING_space0
263 STRING_word0 STRING_xdigit;
264
265 static const pcre_uint8 posix_name_lengths[] = {
266 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 4, 6, 0 };
267
268 /* Table of class bit maps for each POSIX class. Each class is formed from a
269 base map, with an optional addition or removal of another map. Then, for some
270 classes, there is some additional tweaking: for [:blank:] the vertical space
271 characters are removed, and for [:alpha:] and [:alnum:] the underscore
272 character is removed. The triples in the table consist of the base map offset,
273 second map offset or -1 if no second map, and a non-negative value for map
274 addition or a negative value for map subtraction (if there are two maps). The
275 absolute value of the third field has these meanings: 0 => no tweaking, 1 =>
276 remove vertical space characters, 2 => remove underscore. */
277
278 static const int posix_class_maps[] = {
279 cbit_word, cbit_digit, -2, /* alpha */
280 cbit_lower, -1, 0, /* lower */
281 cbit_upper, -1, 0, /* upper */
282 cbit_word, -1, 2, /* alnum - word without underscore */
283 cbit_print, cbit_cntrl, 0, /* ascii */
284 cbit_space, -1, 1, /* blank - a GNU extension */
285 cbit_cntrl, -1, 0, /* cntrl */
286 cbit_digit, -1, 0, /* digit */
287 cbit_graph, -1, 0, /* graph */
288 cbit_print, -1, 0, /* print */
289 cbit_punct, -1, 0, /* punct */
290 cbit_space, -1, 0, /* space */
291 cbit_word, -1, 0, /* word - a Perl extension */
292 cbit_xdigit,-1, 0 /* xdigit */
293 };
294
295 /* Table of substitutes for \d etc when PCRE_UCP is set. The POSIX class
296 substitutes must be in the order of the names, defined above, and there are
297 both positive and negative cases. NULL means no substitute. */
298
299 #ifdef SUPPORT_UCP
300 static const pcre_uchar string_PNd[] = {
301 CHAR_BACKSLASH, CHAR_P, CHAR_LEFT_CURLY_BRACKET,
302 CHAR_N, CHAR_d, CHAR_RIGHT_CURLY_BRACKET, '\0' };
303 static const pcre_uchar string_pNd[] = {
304 CHAR_BACKSLASH, CHAR_p, CHAR_LEFT_CURLY_BRACKET,
305 CHAR_N, CHAR_d, CHAR_RIGHT_CURLY_BRACKET, '\0' };
306 static const pcre_uchar string_PXsp[] = {
307 CHAR_BACKSLASH, CHAR_P, CHAR_LEFT_CURLY_BRACKET,
308 CHAR_X, CHAR_s, CHAR_p, CHAR_RIGHT_CURLY_BRACKET, '\0' };
309 static const pcre_uchar string_pXsp[] = {
310 CHAR_BACKSLASH, CHAR_p, CHAR_LEFT_CURLY_BRACKET,
311 CHAR_X, CHAR_s, CHAR_p, CHAR_RIGHT_CURLY_BRACKET, '\0' };
312 static const pcre_uchar string_PXwd[] = {
313 CHAR_BACKSLASH, CHAR_P, CHAR_LEFT_CURLY_BRACKET,
314 CHAR_X, CHAR_w, CHAR_d, CHAR_RIGHT_CURLY_BRACKET, '\0' };
315 static const pcre_uchar string_pXwd[] = {
316 CHAR_BACKSLASH, CHAR_p, CHAR_LEFT_CURLY_BRACKET,
317 CHAR_X, CHAR_w, CHAR_d, CHAR_RIGHT_CURLY_BRACKET, '\0' };
318
319 static const pcre_uchar *substitutes[] = {
320 string_PNd, /* \D */
321 string_pNd, /* \d */
322 string_PXsp, /* \S */ /* NOTE: Xsp is Perl space */
323 string_pXsp, /* \s */
324 string_PXwd, /* \W */
325 string_pXwd /* \w */
326 };
327
328 static const pcre_uchar string_pL[] = {
329 CHAR_BACKSLASH, CHAR_p, CHAR_LEFT_CURLY_BRACKET,
330 CHAR_L, CHAR_RIGHT_CURLY_BRACKET, '\0' };
331 static const pcre_uchar string_pLl[] = {
332 CHAR_BACKSLASH, CHAR_p, CHAR_LEFT_CURLY_BRACKET,
333 CHAR_L, CHAR_l, CHAR_RIGHT_CURLY_BRACKET, '\0' };
334 static const pcre_uchar string_pLu[] = {
335 CHAR_BACKSLASH, CHAR_p, CHAR_LEFT_CURLY_BRACKET,
336 CHAR_L, CHAR_u, CHAR_RIGHT_CURLY_BRACKET, '\0' };
337 static const pcre_uchar string_pXan[] = {
338 CHAR_BACKSLASH, CHAR_p, CHAR_LEFT_CURLY_BRACKET,
339 CHAR_X, CHAR_a, CHAR_n, CHAR_RIGHT_CURLY_BRACKET, '\0' };
340 static const pcre_uchar string_h[] = {
341 CHAR_BACKSLASH, CHAR_h, '\0' };
342 static const pcre_uchar string_pXps[] = {
343 CHAR_BACKSLASH, CHAR_p, CHAR_LEFT_CURLY_BRACKET,
344 CHAR_X, CHAR_p, CHAR_s, CHAR_RIGHT_CURLY_BRACKET, '\0' };
345 static const pcre_uchar string_PL[] = {
346 CHAR_BACKSLASH, CHAR_P, CHAR_LEFT_CURLY_BRACKET,
347 CHAR_L, CHAR_RIGHT_CURLY_BRACKET, '\0' };
348 static const pcre_uchar string_PLl[] = {
349 CHAR_BACKSLASH, CHAR_P, CHAR_LEFT_CURLY_BRACKET,
350 CHAR_L, CHAR_l, CHAR_RIGHT_CURLY_BRACKET, '\0' };
351 static const pcre_uchar string_PLu[] = {
352 CHAR_BACKSLASH, CHAR_P, CHAR_LEFT_CURLY_BRACKET,
353 CHAR_L, CHAR_u, CHAR_RIGHT_CURLY_BRACKET, '\0' };
354 static const pcre_uchar string_PXan[] = {
355 CHAR_BACKSLASH, CHAR_P, CHAR_LEFT_CURLY_BRACKET,
356 CHAR_X, CHAR_a, CHAR_n, CHAR_RIGHT_CURLY_BRACKET, '\0' };
357 static const pcre_uchar string_H[] = {
358 CHAR_BACKSLASH, CHAR_H, '\0' };
359 static const pcre_uchar string_PXps[] = {
360 CHAR_BACKSLASH, CHAR_P, CHAR_LEFT_CURLY_BRACKET,
361 CHAR_X, CHAR_p, CHAR_s, CHAR_RIGHT_CURLY_BRACKET, '\0' };
362
363 static const pcre_uchar *posix_substitutes[] = {
364 string_pL, /* alpha */
365 string_pLl, /* lower */
366 string_pLu, /* upper */
367 string_pXan, /* alnum */
368 NULL, /* ascii */
369 string_h, /* blank */
370 NULL, /* cntrl */
371 string_pNd, /* digit */
372 NULL, /* graph */
373 NULL, /* print */
374 NULL, /* punct */
375 string_pXps, /* space */ /* NOTE: Xps is POSIX space */
376 string_pXwd, /* word */
377 NULL, /* xdigit */
378 /* Negated cases */
379 string_PL, /* ^alpha */
380 string_PLl, /* ^lower */
381 string_PLu, /* ^upper */
382 string_PXan, /* ^alnum */
383 NULL, /* ^ascii */
384 string_H, /* ^blank */
385 NULL, /* ^cntrl */
386 string_PNd, /* ^digit */
387 NULL, /* ^graph */
388 NULL, /* ^print */
389 NULL, /* ^punct */
390 string_PXps, /* ^space */ /* NOTE: Xps is POSIX space */
391 string_PXwd, /* ^word */
392 NULL /* ^xdigit */
393 };
394 #define POSIX_SUBSIZE (sizeof(posix_substitutes) / sizeof(pcre_uchar *))
395 #endif
396
397 #define STRING(a) # a
398 #define XSTRING(s) STRING(s)
399
400 /* The texts of compile-time error messages. These are "char *" because they
401 are passed to the outside world. Do not ever re-use any error number, because
402 they are documented. Always add a new error instead. Messages marked DEAD below
403 are no longer used. This used to be a table of strings, but in order to reduce
404 the number of relocations needed when a shared library is loaded dynamically,
405 it is now one long string. We cannot use a table of offsets, because the
406 lengths of inserts such as XSTRING(MAX_NAME_SIZE) are not known. Instead, we
407 simply count through to the one we want - this isn't a performance issue
408 because these strings are used only when there is a compilation error.
409
410 Each substring ends with \0 to insert a null character. This includes the final
411 substring, so that the whole string ends with \0\0, which can be detected when
412 counting through. */
413
414 static const char error_texts[] =
415 "no error\0"
416 "\\ at end of pattern\0"
417 "\\c at end of pattern\0"
418 "unrecognized character follows \\\0"
419 "numbers out of order in {} quantifier\0"
420 /* 5 */
421 "number too big in {} quantifier\0"
422 "missing terminating ] for character class\0"
423 "invalid escape sequence in character class\0"
424 "range out of order in character class\0"
425 "nothing to repeat\0"
426 /* 10 */
427 "operand of unlimited repeat could match the empty string\0" /** DEAD **/
428 "internal error: unexpected repeat\0"
429 "unrecognized character after (? or (?-\0"
430 "POSIX named classes are supported only within a class\0"
431 "missing )\0"
432 /* 15 */
433 "reference to non-existent subpattern\0"
434 "erroffset passed as NULL\0"
435 "unknown option bit(s) set\0"
436 "missing ) after comment\0"
437 "parentheses nested too deeply\0" /** DEAD **/
438 /* 20 */
439 "regular expression is too large\0"
440 "failed to get memory\0"
441 "unmatched parentheses\0"
442 "internal error: code overflow\0"
443 "unrecognized character after (?<\0"
444 /* 25 */
445 "lookbehind assertion is not fixed length\0"
446 "malformed number or name after (?(\0"
447 "conditional group contains more than two branches\0"
448 "assertion expected after (?(\0"
449 "(?R or (?[+-]digits must be followed by )\0"
450 /* 30 */
451 "unknown POSIX class name\0"
452 "POSIX collating elements are not supported\0"
453 "this version of PCRE is compiled without UTF support\0"
454 "spare error\0" /** DEAD **/
455 "character value in \\x{...} sequence is too large\0"
456 /* 35 */
457 "invalid condition (?(0)\0"
458 "\\C not allowed in lookbehind assertion\0"
459 "PCRE does not support \\L, \\l, \\N{name}, \\U, or \\u\0"
460 "number after (?C is > 255\0"
461 "closing ) for (?C expected\0"
462 /* 40 */
463 "recursive call could loop indefinitely\0"
464 "unrecognized character after (?P\0"
465 "syntax error in subpattern name (missing terminator)\0"
466 "two named subpatterns have the same name\0"
467 "invalid UTF-8 string\0"
468 /* 45 */
469 "support for \\P, \\p, and \\X has not been compiled\0"
470 "malformed \\P or \\p sequence\0"
471 "unknown property name after \\P or \\p\0"
472 "subpattern name is too long (maximum " XSTRING(MAX_NAME_SIZE) " characters)\0"
473 "too many named subpatterns (maximum " XSTRING(MAX_NAME_COUNT) ")\0"
474 /* 50 */
475 "repeated subpattern is too long\0" /** DEAD **/
476 "octal value is greater than \\377 in 8-bit non-UTF-8 mode\0"
477 "internal error: overran compiling workspace\0"
478 "internal error: previously-checked referenced subpattern not found\0"
479 "DEFINE group contains more than one branch\0"
480 /* 55 */
481 "repeating a DEFINE group is not allowed\0" /** DEAD **/
482 "inconsistent NEWLINE options\0"
483 "\\g is not followed by a braced, angle-bracketed, or quoted name/number or by a plain number\0"
484 "a numbered reference must not be zero\0"
485 "an argument is not allowed for (*ACCEPT), (*FAIL), or (*COMMIT)\0"
486 /* 60 */
487 "(*VERB) not recognized\0"
488 "number is too big\0"
489 "subpattern name expected\0"
490 "digit expected after (?+\0"
491 "] is an invalid data character in JavaScript compatibility mode\0"
492 /* 65 */
493 "different names for subpatterns of the same number are not allowed\0"
494 "(*MARK) must have an argument\0"
495 "this version of PCRE is not compiled with Unicode property support\0"
496 "\\c must be followed by an ASCII character\0"
497 "\\k is not followed by a braced, angle-bracketed, or quoted name\0"
498 /* 70 */
499 "internal error: unknown opcode in find_fixedlength()\0"
500 "\\N is not supported in a class\0"
501 "too many forward references\0"
502 "disallowed Unicode code point (>= 0xd800 && <= 0xdfff)\0"
503 "invalid UTF-16 string\0"
504 /* 75 */
505 "name is too long in (*MARK), (*PRUNE), (*SKIP), or (*THEN)\0"
506 "character value in \\u.... sequence is too large\0"
507 "invalid UTF-32 string\0"
508 ;
509
510 /* Table to identify digits and hex digits. This is used when compiling
511 patterns. Note that the tables in chartables are dependent on the locale, and
512 may mark arbitrary characters as digits - but the PCRE compiling code expects
513 to handle only 0-9, a-z, and A-Z as digits when compiling. That is why we have
514 a private table here. It costs 256 bytes, but it is a lot faster than doing
515 character value tests (at least in some simple cases I timed), and in some
516 applications one wants PCRE to compile efficiently as well as match
517 efficiently.
518
519 For convenience, we use the same bit definitions as in chartables:
520
521 0x04 decimal digit
522 0x08 hexadecimal digit
523
524 Then we can use ctype_digit and ctype_xdigit in the code. */
525
526 /* Using a simple comparison for decimal numbers rather than a memory read
527 is much faster, and the resulting code is simpler (the compiler turns it
528 into a subtraction and unsigned comparison). */
529
530 #define IS_DIGIT(x) ((x) >= CHAR_0 && (x) <= CHAR_9)
531
532 #ifndef EBCDIC
533
534 /* This is the "normal" case, for ASCII systems, and EBCDIC systems running in
535 UTF-8 mode. */
536
537 static const pcre_uint8 digitab[] =
538 {
539 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 0- 7 */
540 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 8- 15 */
541 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 16- 23 */
542 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 24- 31 */
543 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - ' */
544 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* ( - / */
545 0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c, /* 0 - 7 */
546 0x0c,0x0c,0x00,0x00,0x00,0x00,0x00,0x00, /* 8 - ? */
547 0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* @ - G */
548 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* H - O */
549 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* P - W */
550 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* X - _ */
551 0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* ` - g */
552 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* h - o */
553 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* p - w */
554 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* x -127 */
555 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 128-135 */
556 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 136-143 */
557 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 144-151 */
558 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 152-159 */
559 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 160-167 */
560 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 168-175 */
561 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 176-183 */
562 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 184-191 */
563 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 192-199 */
564 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 200-207 */
565 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 208-215 */
566 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 216-223 */
567 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 224-231 */
568 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 232-239 */
569 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 240-247 */
570 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00};/* 248-255 */
571
572 #else
573
574 /* This is the "abnormal" case, for EBCDIC systems not running in UTF-8 mode. */
575
576 static const pcre_uint8 digitab[] =
577 {
578 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 0- 7 0 */
579 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 8- 15 */
580 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 16- 23 10 */
581 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 24- 31 */
582 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 32- 39 20 */
583 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 40- 47 */
584 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 48- 55 30 */
585 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 56- 63 */
586 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - 71 40 */
587 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 72- | */
588 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* & - 87 50 */
589 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 88- 95 */
590 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - -103 60 */
591 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 104- ? */
592 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 112-119 70 */
593 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 120- " */
594 0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* 128- g 80 */
595 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* h -143 */
596 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 144- p 90 */
597 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* q -159 */
598 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 160- x A0 */
599 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* y -175 */
600 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* ^ -183 B0 */
601 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 184-191 */
602 0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* { - G C0 */
603 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* H -207 */
604 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* } - P D0 */
605 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* Q -223 */
606 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* \ - X E0 */
607 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* Y -239 */
608 0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c, /* 0 - 7 F0 */
609 0x0c,0x0c,0x00,0x00,0x00,0x00,0x00,0x00};/* 8 -255 */
610
611 static const pcre_uint8 ebcdic_chartab[] = { /* chartable partial dup */
612 0x80,0x00,0x00,0x00,0x00,0x01,0x00,0x00, /* 0- 7 */
613 0x00,0x00,0x00,0x00,0x01,0x01,0x00,0x00, /* 8- 15 */
614 0x00,0x00,0x00,0x00,0x00,0x01,0x00,0x00, /* 16- 23 */
615 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 24- 31 */
616 0x00,0x00,0x00,0x00,0x00,0x01,0x00,0x00, /* 32- 39 */
617 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 40- 47 */
618 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 48- 55 */
619 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 56- 63 */
620 0x01,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - 71 */
621 0x00,0x00,0x00,0x80,0x00,0x80,0x80,0x80, /* 72- | */
622 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* & - 87 */
623 0x00,0x00,0x00,0x80,0x80,0x80,0x00,0x00, /* 88- 95 */
624 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - -103 */
625 0x00,0x00,0x00,0x00,0x00,0x10,0x00,0x80, /* 104- ? */
626 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 112-119 */
627 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 120- " */
628 0x00,0x1a,0x1a,0x1a,0x1a,0x1a,0x1a,0x12, /* 128- g */
629 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* h -143 */
630 0x00,0x12,0x12,0x12,0x12,0x12,0x12,0x12, /* 144- p */
631 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* q -159 */
632 0x00,0x00,0x12,0x12,0x12,0x12,0x12,0x12, /* 160- x */
633 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* y -175 */
634 0x80,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* ^ -183 */
635 0x00,0x00,0x80,0x00,0x00,0x00,0x00,0x00, /* 184-191 */
636 0x80,0x1a,0x1a,0x1a,0x1a,0x1a,0x1a,0x12, /* { - G */
637 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* H -207 */
638 0x00,0x12,0x12,0x12,0x12,0x12,0x12,0x12, /* } - P */
639 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* Q -223 */
640 0x00,0x00,0x12,0x12,0x12,0x12,0x12,0x12, /* \ - X */
641 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* Y -239 */
642 0x1c,0x1c,0x1c,0x1c,0x1c,0x1c,0x1c,0x1c, /* 0 - 7 */
643 0x1c,0x1c,0x00,0x00,0x00,0x00,0x00,0x00};/* 8 -255 */
644 #endif
645
646
647
648
649 /*************************************************
650 * Find an error text *
651 *************************************************/
652
653 /* The error texts are now all in one long string, to save on relocations. As
654 some of the text is of unknown length, we can't use a table of offsets.
655 Instead, just count through the strings. This is not a performance issue
656 because it happens only when there has been a compilation error.
657
658 Argument: the error number
659 Returns: pointer to the error string
660 */
661
662 static const char *
663 find_error_text(int n)
664 {
665 const char *s = error_texts;
666 for (; n > 0; n--)
667 {
668 while (*s++ != 0) {};
669 if (*s == 0) return "Error text not found (please report)";
670 }
671 return s;
672 }
673
674
675 /*************************************************
676 * Expand the workspace *
677 *************************************************/
678
679 /* This function is called during the second compiling phase, if the number of
680 forward references fills the existing workspace, which is originally a block on
681 the stack. A larger block is obtained from malloc() unless the ultimate limit
682 has been reached or the increase will be rather small.
683
684 Argument: pointer to the compile data block
685 Returns: 0 if all went well, else an error number
686 */
687
688 static int
689 expand_workspace(compile_data *cd)
690 {
691 pcre_uchar *newspace;
692 int newsize = cd->workspace_size * 2;
693
694 if (newsize > COMPILE_WORK_SIZE_MAX) newsize = COMPILE_WORK_SIZE_MAX;
695 if (cd->workspace_size >= COMPILE_WORK_SIZE_MAX ||
696 newsize - cd->workspace_size < WORK_SIZE_SAFETY_MARGIN)
697 return ERR72;
698
699 newspace = (PUBL(malloc))(IN_UCHARS(newsize));
700 if (newspace == NULL) return ERR21;
701 memcpy(newspace, cd->start_workspace, cd->workspace_size * sizeof(pcre_uchar));
702 cd->hwm = (pcre_uchar *)newspace + (cd->hwm - cd->start_workspace);
703 if (cd->workspace_size > COMPILE_WORK_SIZE)
704 (PUBL(free))((void *)cd->start_workspace);
705 cd->start_workspace = newspace;
706 cd->workspace_size = newsize;
707 return 0;
708 }
709
710
711
712 /*************************************************
713 * Check for counted repeat *
714 *************************************************/
715
716 /* This function is called when a '{' is encountered in a place where it might
717 start a quantifier. It looks ahead to see if it really is a quantifier or not.
718 It is only a quantifier if it is one of the forms {ddd} {ddd,} or {ddd,ddd}
719 where the ddds are digits.
720
721 Arguments:
722 p pointer to the first char after '{'
723
724 Returns: TRUE or FALSE
725 */
726
727 static BOOL
728 is_counted_repeat(const pcre_uchar *p)
729 {
730 if (!IS_DIGIT(*p)) return FALSE;
731 p++;
732 while (IS_DIGIT(*p)) p++;
733 if (*p == CHAR_RIGHT_CURLY_BRACKET) return TRUE;
734
735 if (*p++ != CHAR_COMMA) return FALSE;
736 if (*p == CHAR_RIGHT_CURLY_BRACKET) return TRUE;
737
738 if (!IS_DIGIT(*p)) return FALSE;
739 p++;
740 while (IS_DIGIT(*p)) p++;
741
742 return (*p == CHAR_RIGHT_CURLY_BRACKET);
743 }
744
745
746
747 /*************************************************
748 * Handle escapes *
749 *************************************************/
750
751 /* This function is called when a \ has been encountered. It either returns a
752 positive value for a simple escape such as \n, or 0 for a data character
753 which will be placed in chptr. A backreference to group
754 n is returned as ESC_REF + n; ESC_REF is the highest ESC_xxx macro. When
755 UTF-8 is enabled, a positive value greater than 255 may be returned in chptr.
756 On entry,ptr is pointing at the \. On exit, it is on the final character of the
757 escape sequence.
758
759 Arguments:
760 ptrptr points to the pattern position pointer
761 chptr points to the data character
762 errorcodeptr points to the errorcode variable
763 bracount number of previous extracting brackets
764 options the options bits
765 isclass TRUE if inside a character class
766
767 Returns: zero => a data character
768 positive => a special escape sequence
769 on error, errorcodeptr is set
770 */
771
772 static int
773 check_escape(const pcre_uchar **ptrptr, int *chptr, int *errorcodeptr,
774 int bracount, int options, BOOL isclass)
775 {
776 /* PCRE_UTF16 has the same value as PCRE_UTF8. */
777 BOOL utf = (options & PCRE_UTF8) != 0;
778 const pcre_uchar *ptr = *ptrptr + 1;
779 pcre_uint32 c;
780 int escape = 0;
781 int i;
782
783 GETCHARINCTEST(c, ptr); /* Get character value, increment pointer */
784 ptr--; /* Set pointer back to the last byte */
785
786 /* If backslash is at the end of the pattern, it's an error. */
787
788 if (c == 0) *errorcodeptr = ERR1;
789
790 /* Non-alphanumerics are literals. For digits or letters, do an initial lookup
791 in a table. A non-zero result is something that can be returned immediately.
792 Otherwise further processing may be required. */
793
794 #ifndef EBCDIC /* ASCII/UTF-8 coding */
795 /* Not alphanumeric */
796 else if (c < CHAR_0 || c > CHAR_z) {}
797 else if ((i = escapes[c - CHAR_0]) != 0) { if (i > 0) c = i; else escape = -i; }
798
799 #else /* EBCDIC coding */
800 /* Not alphanumeric */
801 else if (c < CHAR_a || (!MAX_255(c) || (ebcdic_chartab[c] & 0x0E) == 0)) {}
802 else if ((i = escapes[c - 0x48]) != 0) { if (i > 0) c = i; else escape = -i; }
803 #endif
804
805 /* Escapes that need further processing, or are illegal. */
806
807 else
808 {
809 const pcre_uchar *oldptr;
810 BOOL braced, negated;
811
812 switch (c)
813 {
814 /* A number of Perl escapes are not handled by PCRE. We give an explicit
815 error. */
816
817 case CHAR_l:
818 case CHAR_L:
819 *errorcodeptr = ERR37;
820 break;
821
822 case CHAR_u:
823 if ((options & PCRE_JAVASCRIPT_COMPAT) != 0)
824 {
825 /* In JavaScript, \u must be followed by four hexadecimal numbers.
826 Otherwise it is a lowercase u letter. */
827 if (MAX_255(ptr[1]) && (digitab[ptr[1]] & ctype_xdigit) != 0
828 && MAX_255(ptr[2]) && (digitab[ptr[2]] & ctype_xdigit) != 0
829 && MAX_255(ptr[3]) && (digitab[ptr[3]] & ctype_xdigit) != 0
830 && MAX_255(ptr[4]) && (digitab[ptr[4]] & ctype_xdigit) != 0)
831 {
832 c = 0;
833 for (i = 0; i < 4; ++i)
834 {
835 register pcre_uint32 cc = *(++ptr);
836 #ifndef EBCDIC /* ASCII/UTF-8 coding */
837 if (cc >= CHAR_a) cc -= 32; /* Convert to upper case */
838 c = (c << 4) + cc - ((cc < CHAR_A)? CHAR_0 : (CHAR_A - 10));
839 #else /* EBCDIC coding */
840 if (cc >= CHAR_a && cc <= CHAR_z) cc += 64; /* Convert to upper case */
841 c = (c << 4) + cc - ((cc >= CHAR_0)? CHAR_0 : (CHAR_A - 10));
842 #endif
843 }
844
845 #if defined COMPILE_PCRE8
846 if (c > (utf ? 0x10ffff : 0xff))
847 #elif defined COMPILE_PCRE16
848 if (c > (utf ? 0x10ffff : 0xffff))
849 #elif defined COMPILE_PCRE32
850 if (utf && c > 0x10ffff)
851 #endif
852 {
853 *errorcodeptr = ERR76;
854 }
855 else if (utf && c >= 0xd800 && c <= 0xdfff) *errorcodeptr = ERR73;
856 }
857 }
858 else
859 *errorcodeptr = ERR37;
860 break;
861
862 case CHAR_U:
863 /* In JavaScript, \U is an uppercase U letter. */
864 if ((options & PCRE_JAVASCRIPT_COMPAT) == 0) *errorcodeptr = ERR37;
865 break;
866
867 /* In a character class, \g is just a literal "g". Outside a character
868 class, \g must be followed by one of a number of specific things:
869
870 (1) A number, either plain or braced. If positive, it is an absolute
871 backreference. If negative, it is a relative backreference. This is a Perl
872 5.10 feature.
873
874 (2) Perl 5.10 also supports \g{name} as a reference to a named group. This
875 is part of Perl's movement towards a unified syntax for back references. As
876 this is synonymous with \k{name}, we fudge it up by pretending it really
877 was \k.
878
879 (3) For Oniguruma compatibility we also support \g followed by a name or a
880 number either in angle brackets or in single quotes. However, these are
881 (possibly recursive) subroutine calls, _not_ backreferences. Just return
882 the ESC_g code (cf \k). */
883
884 case CHAR_g:
885 if (isclass) break;
886 if (ptr[1] == CHAR_LESS_THAN_SIGN || ptr[1] == CHAR_APOSTROPHE)
887 {
888 escape = ESC_g;
889 break;
890 }
891
892 /* Handle the Perl-compatible cases */
893
894 if (ptr[1] == CHAR_LEFT_CURLY_BRACKET)
895 {
896 const pcre_uchar *p;
897 for (p = ptr+2; *p != 0 && *p != CHAR_RIGHT_CURLY_BRACKET; p++)
898 if (*p != CHAR_MINUS && !IS_DIGIT(*p)) break;
899 if (*p != 0 && *p != CHAR_RIGHT_CURLY_BRACKET)
900 {
901 escape = ESC_k;
902 break;
903 }
904 braced = TRUE;
905 ptr++;
906 }
907 else braced = FALSE;
908
909 if (ptr[1] == CHAR_MINUS)
910 {
911 negated = TRUE;
912 ptr++;
913 }
914 else negated = FALSE;
915
916 /* The integer range is limited by the machine's int representation. */
917 c = 0;
918 while (IS_DIGIT(ptr[1]))
919 {
920 if (((unsigned int)c) > INT_MAX / 10) /* Integer overflow */
921 {
922 c = -1;
923 break;
924 }
925 c = c * 10 + *(++ptr) - CHAR_0;
926 }
927 if (((unsigned int)c) > INT_MAX) /* Integer overflow */
928 {
929 while (IS_DIGIT(ptr[1]))
930 ptr++;
931 *errorcodeptr = ERR61;
932 break;
933 }
934
935 if (braced && *(++ptr) != CHAR_RIGHT_CURLY_BRACKET)
936 {
937 *errorcodeptr = ERR57;
938 break;
939 }
940
941 if (c == 0)
942 {
943 *errorcodeptr = ERR58;
944 break;
945 }
946
947 if (negated)
948 {
949 if (c > bracount)
950 {
951 *errorcodeptr = ERR15;
952 break;
953 }
954 c = bracount - (c - 1);
955 }
956
957 escape = ESC_REF + c;
958 break;
959
960 /* The handling of escape sequences consisting of a string of digits
961 starting with one that is not zero is not straightforward. By experiment,
962 the way Perl works seems to be as follows:
963
964 Outside a character class, the digits are read as a decimal number. If the
965 number is less than 10, or if there are that many previous extracting
966 left brackets, then it is a back reference. Otherwise, up to three octal
967 digits are read to form an escaped byte. Thus \123 is likely to be octal
968 123 (cf \0123, which is octal 012 followed by the literal 3). If the octal
969 value is greater than 377, the least significant 8 bits are taken. Inside a
970 character class, \ followed by a digit is always an octal number. */
971
972 case CHAR_1: case CHAR_2: case CHAR_3: case CHAR_4: case CHAR_5:
973 case CHAR_6: case CHAR_7: case CHAR_8: case CHAR_9:
974
975 if (!isclass)
976 {
977 oldptr = ptr;
978 /* The integer range is limited by the machine's int representation. */
979 c -= CHAR_0;
980 while (IS_DIGIT(ptr[1]))
981 {
982 if (((unsigned int)c) > INT_MAX / 10) /* Integer overflow */
983 {
984 c = -1;
985 break;
986 }
987 c = c * 10 + *(++ptr) - CHAR_0;
988 }
989 if (((unsigned int)c) > INT_MAX) /* Integer overflow */
990 {
991 while (IS_DIGIT(ptr[1]))
992 ptr++;
993 *errorcodeptr = ERR61;
994 break;
995 }
996 if (c < 10 || c <= bracount)
997 {
998 escape = ESC_REF + c;
999 break;
1000 }
1001 ptr = oldptr; /* Put the pointer back and fall through */
1002 }
1003
1004 /* Handle an octal number following \. If the first digit is 8 or 9, Perl
1005 generates a binary zero byte and treats the digit as a following literal.
1006 Thus we have to pull back the pointer by one. */
1007
1008 if ((c = *ptr) >= CHAR_8)
1009 {
1010 ptr--;
1011 c = 0;
1012 break;
1013 }
1014
1015 /* \0 always starts an octal number, but we may drop through to here with a
1016 larger first octal digit. The original code used just to take the least
1017 significant 8 bits of octal numbers (I think this is what early Perls used
1018 to do). Nowadays we allow for larger numbers in UTF-8 mode and 16-bit mode,
1019 but no more than 3 octal digits. */
1020
1021 case CHAR_0:
1022 c -= CHAR_0;
1023 while(i++ < 2 && ptr[1] >= CHAR_0 && ptr[1] <= CHAR_7)
1024 c = c * 8 + *(++ptr) - CHAR_0;
1025 #ifdef COMPILE_PCRE8
1026 if (!utf && c > 0xff) *errorcodeptr = ERR51;
1027 #endif
1028 break;
1029
1030 /* \x is complicated. \x{ddd} is a character number which can be greater
1031 than 0xff in utf or non-8bit mode, but only if the ddd are hex digits.
1032 If not, { is treated as a data character. */
1033
1034 case CHAR_x:
1035 if ((options & PCRE_JAVASCRIPT_COMPAT) != 0)
1036 {
1037 /* In JavaScript, \x must be followed by two hexadecimal numbers.
1038 Otherwise it is a lowercase x letter. */
1039 if (MAX_255(ptr[1]) && (digitab[ptr[1]] & ctype_xdigit) != 0
1040 && MAX_255(ptr[2]) && (digitab[ptr[2]] & ctype_xdigit) != 0)
1041 {
1042 c = 0;
1043 for (i = 0; i < 2; ++i)
1044 {
1045 register pcre_uint32 cc = *(++ptr);
1046 #ifndef EBCDIC /* ASCII/UTF-8 coding */
1047 if (cc >= CHAR_a) cc -= 32; /* Convert to upper case */
1048 c = (c << 4) + cc - ((cc < CHAR_A)? CHAR_0 : (CHAR_A - 10));
1049 #else /* EBCDIC coding */
1050 if (cc >= CHAR_a && cc <= CHAR_z) cc += 64; /* Convert to upper case */
1051 c = (c << 4) + cc - ((cc >= CHAR_0)? CHAR_0 : (CHAR_A - 10));
1052 #endif
1053 }
1054 }
1055 break;
1056 }
1057
1058 if (ptr[1] == CHAR_LEFT_CURLY_BRACKET)
1059 {
1060 const pcre_uchar *pt = ptr + 2;
1061 BOOL overflow;
1062
1063 c = 0;
1064 overflow = FALSE;
1065 while (MAX_255(*pt) && (digitab[*pt] & ctype_xdigit) != 0)
1066 {
1067 register pcre_uint32 cc = *pt++;
1068 if (c == 0 && cc == CHAR_0) continue; /* Leading zeroes */
1069
1070 #ifdef COMPILE_PCRE32
1071 if (c >= 0x10000000l) { overflow = TRUE; break; }
1072 #endif
1073
1074 #ifndef EBCDIC /* ASCII/UTF-8 coding */
1075 if (cc >= CHAR_a) cc -= 32; /* Convert to upper case */
1076 c = (c << 4) + cc - ((cc < CHAR_A)? CHAR_0 : (CHAR_A - 10));
1077 #else /* EBCDIC coding */
1078 if (cc >= CHAR_a && cc <= CHAR_z) cc += 64; /* Convert to upper case */
1079 c = (c << 4) + cc - ((cc >= CHAR_0)? CHAR_0 : (CHAR_A - 10));
1080 #endif
1081
1082 #if defined COMPILE_PCRE8
1083 if (c > (utf ? 0x10ffff : 0xff)) { overflow = TRUE; break; }
1084 #elif defined COMPILE_PCRE16
1085 if (c > (utf ? 0x10ffff : 0xffff)) { overflow = TRUE; break; }
1086 #elif defined COMPILE_PCRE32
1087 if (utf && c > 0x10ffff) { overflow = TRUE; break; }
1088 #endif
1089 }
1090
1091 if (overflow)
1092 {
1093 while (MAX_255(*pt) && (digitab[*pt] & ctype_xdigit) != 0) pt++;
1094 *errorcodeptr = ERR34;
1095 }
1096
1097 if (*pt == CHAR_RIGHT_CURLY_BRACKET)
1098 {
1099 if (utf && c >= 0xd800 && c <= 0xdfff) *errorcodeptr = ERR73;
1100 ptr = pt;
1101 break;
1102 }
1103
1104 /* If the sequence of hex digits does not end with '}', then we don't
1105 recognize this construct; fall through to the normal \x handling. */
1106 }
1107
1108 /* Read just a single-byte hex-defined char */
1109
1110 c = 0;
1111 while (i++ < 2 && MAX_255(ptr[1]) && (digitab[ptr[1]] & ctype_xdigit) != 0)
1112 {
1113 pcre_uint32 cc; /* Some compilers don't like */
1114 cc = *(++ptr); /* ++ in initializers */
1115 #ifndef EBCDIC /* ASCII/UTF-8 coding */
1116 if (cc >= CHAR_a) cc -= 32; /* Convert to upper case */
1117 c = c * 16 + cc - ((cc < CHAR_A)? CHAR_0 : (CHAR_A - 10));
1118 #else /* EBCDIC coding */
1119 if (cc <= CHAR_z) cc += 64; /* Convert to upper case */
1120 c = c * 16 + cc - ((cc >= CHAR_0)? CHAR_0 : (CHAR_A - 10));
1121 #endif
1122 }
1123 break;
1124
1125 /* For \c, a following letter is upper-cased; then the 0x40 bit is flipped.
1126 An error is given if the byte following \c is not an ASCII character. This
1127 coding is ASCII-specific, but then the whole concept of \cx is
1128 ASCII-specific. (However, an EBCDIC equivalent has now been added.) */
1129
1130 case CHAR_c:
1131 c = *(++ptr);
1132 if (c == 0)
1133 {
1134 *errorcodeptr = ERR2;
1135 break;
1136 }
1137 #ifndef EBCDIC /* ASCII/UTF-8 coding */
1138 if (c > 127) /* Excludes all non-ASCII in either mode */
1139 {
1140 *errorcodeptr = ERR68;
1141 break;
1142 }
1143 if (c >= CHAR_a && c <= CHAR_z) c -= 32;
1144 c ^= 0x40;
1145 #else /* EBCDIC coding */
1146 if (c >= CHAR_a && c <= CHAR_z) c += 64;
1147 c ^= 0xC0;
1148 #endif
1149 break;
1150
1151 /* PCRE_EXTRA enables extensions to Perl in the matter of escapes. Any
1152 other alphanumeric following \ is an error if PCRE_EXTRA was set;
1153 otherwise, for Perl compatibility, it is a literal. This code looks a bit
1154 odd, but there used to be some cases other than the default, and there may
1155 be again in future, so I haven't "optimized" it. */
1156
1157 default:
1158 if ((options & PCRE_EXTRA) != 0) switch(c)
1159 {
1160 default:
1161 *errorcodeptr = ERR3;
1162 break;
1163 }
1164 break;
1165 }
1166 }
1167
1168 /* Perl supports \N{name} for character names, as well as plain \N for "not
1169 newline". PCRE does not support \N{name}. However, it does support
1170 quantification such as \N{2,3}. */
1171
1172 if (escape == ESC_N && ptr[1] == CHAR_LEFT_CURLY_BRACKET &&
1173 !is_counted_repeat(ptr+2))
1174 *errorcodeptr = ERR37;
1175
1176 /* If PCRE_UCP is set, we change the values for \d etc. */
1177
1178 if ((options & PCRE_UCP) != 0 && escape >= ESC_D && escape <= ESC_w)
1179 escape += (ESC_DU - ESC_D);
1180
1181 /* Set the pointer to the final character before returning. */
1182
1183 *ptrptr = ptr;
1184 *chptr = c;
1185 return escape;
1186 }
1187
1188 #ifdef SUPPORT_UCP
1189 /*************************************************
1190 * Handle \P and \p *
1191 *************************************************/
1192
1193 /* This function is called after \P or \p has been encountered, provided that
1194 PCRE is compiled with support for Unicode properties. On entry, ptrptr is
1195 pointing at the P or p. On exit, it is pointing at the final character of the
1196 escape sequence.
1197
1198 Argument:
1199 ptrptr points to the pattern position pointer
1200 negptr points to a boolean that is set TRUE for negation else FALSE
1201 dptr points to an int that is set to the detailed property value
1202 errorcodeptr points to the error code variable
1203
1204 Returns: type value from ucp_type_table, or -1 for an invalid type
1205 */
1206
1207 static int
1208 get_ucp(const pcre_uchar **ptrptr, BOOL *negptr, int *dptr, int *errorcodeptr)
1209 {
1210 int c, i, bot, top;
1211 const pcre_uchar *ptr = *ptrptr;
1212 pcre_uchar name[32];
1213
1214 c = *(++ptr);
1215 if (c == 0) goto ERROR_RETURN;
1216
1217 *negptr = FALSE;
1218
1219 /* \P or \p can be followed by a name in {}, optionally preceded by ^ for
1220 negation. */
1221
1222 if (c == CHAR_LEFT_CURLY_BRACKET)
1223 {
1224 if (ptr[1] == CHAR_CIRCUMFLEX_ACCENT)
1225 {
1226 *negptr = TRUE;
1227 ptr++;
1228 }
1229 for (i = 0; i < (int)(sizeof(name) / sizeof(pcre_uchar)) - 1; i++)
1230 {
1231 c = *(++ptr);
1232 if (c == 0) goto ERROR_RETURN;
1233 if (c == CHAR_RIGHT_CURLY_BRACKET) break;
1234 name[i] = c;
1235 }
1236 if (c != CHAR_RIGHT_CURLY_BRACKET) goto ERROR_RETURN;
1237 name[i] = 0;
1238 }
1239
1240 /* Otherwise there is just one following character */
1241
1242 else
1243 {
1244 name[0] = c;
1245 name[1] = 0;
1246 }
1247
1248 *ptrptr = ptr;
1249
1250 /* Search for a recognized property name using binary chop */
1251
1252 bot = 0;
1253 top = PRIV(utt_size);
1254
1255 while (bot < top)
1256 {
1257 i = (bot + top) >> 1;
1258 c = STRCMP_UC_C8(name, PRIV(utt_names) + PRIV(utt)[i].name_offset);
1259 if (c == 0)
1260 {
1261 *dptr = PRIV(utt)[i].value;
1262 return PRIV(utt)[i].type;
1263 }
1264 if (c > 0) bot = i + 1; else top = i;
1265 }
1266
1267 *errorcodeptr = ERR47;
1268 *ptrptr = ptr;
1269 return -1;
1270
1271 ERROR_RETURN:
1272 *errorcodeptr = ERR46;
1273 *ptrptr = ptr;
1274 return -1;
1275 }
1276 #endif
1277
1278
1279
1280
1281 /*************************************************
1282 * Read repeat counts *
1283 *************************************************/
1284
1285 /* Read an item of the form {n,m} and return the values. This is called only
1286 after is_counted_repeat() has confirmed that a repeat-count quantifier exists,
1287 so the syntax is guaranteed to be correct, but we need to check the values.
1288
1289 Arguments:
1290 p pointer to first char after '{'
1291 minp pointer to int for min
1292 maxp pointer to int for max
1293 returned as -1 if no max
1294 errorcodeptr points to error code variable
1295
1296 Returns: pointer to '}' on success;
1297 current ptr on error, with errorcodeptr set non-zero
1298 */
1299
1300 static const pcre_uchar *
1301 read_repeat_counts(const pcre_uchar *p, int *minp, int *maxp, int *errorcodeptr)
1302 {
1303 int min = 0;
1304 int max = -1;
1305
1306 /* Read the minimum value and do a paranoid check: a negative value indicates
1307 an integer overflow. */
1308
1309 while (IS_DIGIT(*p)) min = min * 10 + *p++ - CHAR_0;
1310 if (min < 0 || min > 65535)
1311 {
1312 *errorcodeptr = ERR5;
1313 return p;
1314 }
1315
1316 /* Read the maximum value if there is one, and again do a paranoid on its size.
1317 Also, max must not be less than min. */
1318
1319 if (*p == CHAR_RIGHT_CURLY_BRACKET) max = min; else
1320 {
1321 if (*(++p) != CHAR_RIGHT_CURLY_BRACKET)
1322 {
1323 max = 0;
1324 while(IS_DIGIT(*p)) max = max * 10 + *p++ - CHAR_0;
1325 if (max < 0 || max > 65535)
1326 {
1327 *errorcodeptr = ERR5;
1328 return p;
1329 }
1330 if (max < min)
1331 {
1332 *errorcodeptr = ERR4;
1333 return p;
1334 }
1335 }
1336 }
1337
1338 /* Fill in the required variables, and pass back the pointer to the terminating
1339 '}'. */
1340
1341 *minp = min;
1342 *maxp = max;
1343 return p;
1344 }
1345
1346
1347
1348 /*************************************************
1349 * Subroutine for finding forward reference *
1350 *************************************************/
1351
1352 /* This recursive function is called only from find_parens() below. The
1353 top-level call starts at the beginning of the pattern. All other calls must
1354 start at a parenthesis. It scans along a pattern's text looking for capturing
1355 subpatterns, and counting them. If it finds a named pattern that matches the
1356 name it is given, it returns its number. Alternatively, if the name is NULL, it
1357 returns when it reaches a given numbered subpattern. Recursion is used to keep
1358 track of subpatterns that reset the capturing group numbers - the (?| feature.
1359
1360 This function was originally called only from the second pass, in which we know
1361 that if (?< or (?' or (?P< is encountered, the name will be correctly
1362 terminated because that is checked in the first pass. There is now one call to
1363 this function in the first pass, to check for a recursive back reference by
1364 name (so that we can make the whole group atomic). In this case, we need check
1365 only up to the current position in the pattern, and that is still OK because
1366 and previous occurrences will have been checked. To make this work, the test
1367 for "end of pattern" is a check against cd->end_pattern in the main loop,
1368 instead of looking for a binary zero. This means that the special first-pass
1369 call can adjust cd->end_pattern temporarily. (Checks for binary zero while
1370 processing items within the loop are OK, because afterwards the main loop will
1371 terminate.)
1372
1373 Arguments:
1374 ptrptr address of the current character pointer (updated)
1375 cd compile background data
1376 name name to seek, or NULL if seeking a numbered subpattern
1377 lorn name length, or subpattern number if name is NULL
1378 xmode TRUE if we are in /x mode
1379 utf TRUE if we are in UTF-8 / UTF-16 / UTF-32 mode
1380 count pointer to the current capturing subpattern number (updated)
1381
1382 Returns: the number of the named subpattern, or -1 if not found
1383 */
1384
1385 static int
1386 find_parens_sub(pcre_uchar **ptrptr, compile_data *cd, const pcre_uchar *name, int lorn,
1387 BOOL xmode, BOOL utf, int *count)
1388 {
1389 pcre_uchar *ptr = *ptrptr;
1390 int start_count = *count;
1391 int hwm_count = start_count;
1392 BOOL dup_parens = FALSE;
1393
1394 /* If the first character is a parenthesis, check on the type of group we are
1395 dealing with. The very first call may not start with a parenthesis. */
1396
1397 if (ptr[0] == CHAR_LEFT_PARENTHESIS)
1398 {
1399 /* Handle specials such as (*SKIP) or (*UTF8) etc. */
1400
1401 if (ptr[1] == CHAR_ASTERISK) ptr += 2;
1402
1403 /* Handle a normal, unnamed capturing parenthesis. */
1404
1405 else if (ptr[1] != CHAR_QUESTION_MARK)
1406 {
1407 *count += 1;
1408 if (name == NULL && *count == lorn) return *count;
1409 ptr++;
1410 }
1411
1412 /* All cases now have (? at the start. Remember when we are in a group
1413 where the parenthesis numbers are duplicated. */
1414
1415 else if (ptr[2] == CHAR_VERTICAL_LINE)
1416 {
1417 ptr += 3;
1418 dup_parens = TRUE;
1419 }
1420
1421 /* Handle comments; all characters are allowed until a ket is reached. */
1422
1423 else if (ptr[2] == CHAR_NUMBER_SIGN)
1424 {
1425 for (ptr += 3; *ptr != 0; ptr++) if (*ptr == CHAR_RIGHT_PARENTHESIS) break;
1426 goto FAIL_EXIT;
1427 }
1428
1429 /* Handle a condition. If it is an assertion, just carry on so that it
1430 is processed as normal. If not, skip to the closing parenthesis of the
1431 condition (there can't be any nested parens). */
1432
1433 else if (ptr[2] == CHAR_LEFT_PARENTHESIS)
1434 {
1435 ptr += 2;
1436 if (ptr[1] != CHAR_QUESTION_MARK)
1437 {
1438 while (*ptr != 0 && *ptr != CHAR_RIGHT_PARENTHESIS) ptr++;
1439 if (*ptr != 0) ptr++;
1440 }
1441 }
1442
1443 /* Start with (? but not a condition. */
1444
1445 else
1446 {
1447 ptr += 2;
1448 if (*ptr == CHAR_P) ptr++; /* Allow optional P */
1449
1450 /* We have to disambiguate (?<! and (?<= from (?<name> for named groups */
1451
1452 if ((*ptr == CHAR_LESS_THAN_SIGN && ptr[1] != CHAR_EXCLAMATION_MARK &&
1453 ptr[1] != CHAR_EQUALS_SIGN) || *ptr == CHAR_APOSTROPHE)
1454 {
1455 int term;
1456 const pcre_uchar *thisname;
1457 *count += 1;
1458 if (name == NULL && *count == lorn) return *count;
1459 term = *ptr++;
1460 if (term == CHAR_LESS_THAN_SIGN) term = CHAR_GREATER_THAN_SIGN;
1461 thisname = ptr;
1462 while (*ptr != term) ptr++;
1463 if (name != NULL && lorn == ptr - thisname &&
1464 STRNCMP_UC_UC(name, thisname, lorn) == 0)
1465 return *count;
1466 term++;
1467 }
1468 }
1469 }
1470
1471 /* Past any initial parenthesis handling, scan for parentheses or vertical
1472 bars. Stop if we get to cd->end_pattern. Note that this is important for the
1473 first-pass call when this value is temporarily adjusted to stop at the current
1474 position. So DO NOT change this to a test for binary zero. */
1475
1476 for (; ptr < cd->end_pattern; ptr++)
1477 {
1478 /* Skip over backslashed characters and also entire \Q...\E */
1479
1480 if (*ptr == CHAR_BACKSLASH)
1481 {
1482 if (*(++ptr) == 0) goto FAIL_EXIT;
1483 if (*ptr == CHAR_Q) for (;;)
1484 {
1485 while (*(++ptr) != 0 && *ptr != CHAR_BACKSLASH) {};
1486 if (*ptr == 0) goto FAIL_EXIT;
1487 if (*(++ptr) == CHAR_E) break;
1488 }
1489 continue;
1490 }
1491
1492 /* Skip over character classes; this logic must be similar to the way they
1493 are handled for real. If the first character is '^', skip it. Also, if the
1494 first few characters (either before or after ^) are \Q\E or \E we skip them
1495 too. This makes for compatibility with Perl. Note the use of STR macros to
1496 encode "Q\\E" so that it works in UTF-8 on EBCDIC platforms. */
1497
1498 if (*ptr == CHAR_LEFT_SQUARE_BRACKET)
1499 {
1500 BOOL negate_class = FALSE;
1501 for (;;)
1502 {
1503 if (ptr[1] == CHAR_BACKSLASH)
1504 {
1505 if (ptr[2] == CHAR_E)
1506 ptr+= 2;
1507 else if (STRNCMP_UC_C8(ptr + 2,
1508 STR_Q STR_BACKSLASH STR_E, 3) == 0)
1509 ptr += 4;
1510 else
1511 break;
1512 }
1513 else if (!negate_class && ptr[1] == CHAR_CIRCUMFLEX_ACCENT)
1514 {
1515 negate_class = TRUE;
1516 ptr++;
1517 }
1518 else break;
1519 }
1520
1521 /* If the next character is ']', it is a data character that must be
1522 skipped, except in JavaScript compatibility mode. */
1523
1524 if (ptr[1] == CHAR_RIGHT_SQUARE_BRACKET &&
1525 (cd->external_options & PCRE_JAVASCRIPT_COMPAT) == 0)
1526 ptr++;
1527
1528 while (*(++ptr) != CHAR_RIGHT_SQUARE_BRACKET)
1529 {
1530 if (*ptr == 0) return -1;
1531 if (*ptr == CHAR_BACKSLASH)
1532 {
1533 if (*(++ptr) == 0) goto FAIL_EXIT;
1534 if (*ptr == CHAR_Q) for (;;)
1535 {
1536 while (*(++ptr) != 0 && *ptr != CHAR_BACKSLASH) {};
1537 if (*ptr == 0) goto FAIL_EXIT;
1538 if (*(++ptr) == CHAR_E) break;
1539 }
1540 continue;
1541 }
1542 }
1543 continue;
1544 }
1545
1546 /* Skip comments in /x mode */
1547
1548 if (xmode && *ptr == CHAR_NUMBER_SIGN)
1549 {
1550 ptr++;
1551 while (*ptr != 0)
1552 {
1553 if (IS_NEWLINE(ptr)) { ptr += cd->nllen - 1; break; }
1554 ptr++;
1555 #ifdef SUPPORT_UTF
1556 if (utf) FORWARDCHAR(ptr);
1557 #endif
1558 }
1559 if (*ptr == 0) goto FAIL_EXIT;
1560 continue;
1561 }
1562
1563 /* Check for the special metacharacters */
1564
1565 if (*ptr == CHAR_LEFT_PARENTHESIS)
1566 {
1567 int rc = find_parens_sub(&ptr, cd, name, lorn, xmode, utf, count);
1568 if (rc > 0) return rc;
1569 if (*ptr == 0) goto FAIL_EXIT;
1570 }
1571
1572 else if (*ptr == CHAR_RIGHT_PARENTHESIS)
1573 {
1574 if (dup_parens && *count < hwm_count) *count = hwm_count;
1575 goto FAIL_EXIT;
1576 }
1577
1578 else if (*ptr == CHAR_VERTICAL_LINE && dup_parens)
1579 {
1580 if (*count > hwm_count) hwm_count = *count;
1581 *count = start_count;
1582 }
1583 }
1584
1585 FAIL_EXIT:
1586 *ptrptr = ptr;
1587 return -1;
1588 }
1589
1590
1591
1592
1593 /*************************************************
1594 * Find forward referenced subpattern *
1595 *************************************************/
1596
1597 /* This function scans along a pattern's text looking for capturing
1598 subpatterns, and counting them. If it finds a named pattern that matches the
1599 name it is given, it returns its number. Alternatively, if the name is NULL, it
1600 returns when it reaches a given numbered subpattern. This is used for forward
1601 references to subpatterns. We used to be able to start this scan from the
1602 current compiling point, using the current count value from cd->bracount, and
1603 do it all in a single loop, but the addition of the possibility of duplicate
1604 subpattern numbers means that we have to scan from the very start, in order to
1605 take account of such duplicates, and to use a recursive function to keep track
1606 of the different types of group.
1607
1608 Arguments:
1609 cd compile background data
1610 name name to seek, or NULL if seeking a numbered subpattern
1611 lorn name length, or subpattern number if name is NULL
1612 xmode TRUE if we are in /x mode
1613 utf TRUE if we are in UTF-8 / UTF-16 / UTF-32 mode
1614
1615 Returns: the number of the found subpattern, or -1 if not found
1616 */
1617
1618 static int
1619 find_parens(compile_data *cd, const pcre_uchar *name, int lorn, BOOL xmode,
1620 BOOL utf)
1621 {
1622 pcre_uchar *ptr = (pcre_uchar *)cd->start_pattern;
1623 int count = 0;
1624 int rc;
1625
1626 /* If the pattern does not start with an opening parenthesis, the first call
1627 to find_parens_sub() will scan right to the end (if necessary). However, if it
1628 does start with a parenthesis, find_parens_sub() will return when it hits the
1629 matching closing parens. That is why we have to have a loop. */
1630
1631 for (;;)
1632 {
1633 rc = find_parens_sub(&ptr, cd, name, lorn, xmode, utf, &count);
1634 if (rc > 0 || *ptr++ == 0) break;
1635 }
1636
1637 return rc;
1638 }
1639
1640
1641
1642
1643 /*************************************************
1644 * Find first significant op code *
1645 *************************************************/
1646
1647 /* This is called by several functions that scan a compiled expression looking
1648 for a fixed first character, or an anchoring op code etc. It skips over things
1649 that do not influence this. For some calls, it makes sense to skip negative
1650 forward and all backward assertions, and also the \b assertion; for others it
1651 does not.
1652
1653 Arguments:
1654 code pointer to the start of the group
1655 skipassert TRUE if certain assertions are to be skipped
1656
1657 Returns: pointer to the first significant opcode
1658 */
1659
1660 static const pcre_uchar*
1661 first_significant_code(const pcre_uchar *code, BOOL skipassert)
1662 {
1663 for (;;)
1664 {
1665 switch ((int)*code)
1666 {
1667 case OP_ASSERT_NOT:
1668 case OP_ASSERTBACK:
1669 case OP_ASSERTBACK_NOT:
1670 if (!skipassert) return code;
1671 do code += GET(code, 1); while (*code == OP_ALT);
1672 code += PRIV(OP_lengths)[*code];
1673 break;
1674
1675 case OP_WORD_BOUNDARY:
1676 case OP_NOT_WORD_BOUNDARY:
1677 if (!skipassert) return code;
1678 /* Fall through */
1679
1680 case OP_CALLOUT:
1681 case OP_CREF:
1682 case OP_NCREF:
1683 case OP_RREF:
1684 case OP_NRREF:
1685 case OP_DEF:
1686 code += PRIV(OP_lengths)[*code];
1687 break;
1688
1689 default:
1690 return code;
1691 }
1692 }
1693 /* Control never reaches here */
1694 }
1695
1696
1697
1698
1699 /*************************************************
1700 * Find the fixed length of a branch *
1701 *************************************************/
1702
1703 /* Scan a branch and compute the fixed length of subject that will match it,
1704 if the length is fixed. This is needed for dealing with backward assertions.
1705 In UTF8 mode, the result is in characters rather than bytes. The branch is
1706 temporarily terminated with OP_END when this function is called.
1707
1708 This function is called when a backward assertion is encountered, so that if it
1709 fails, the error message can point to the correct place in the pattern.
1710 However, we cannot do this when the assertion contains subroutine calls,
1711 because they can be forward references. We solve this by remembering this case
1712 and doing the check at the end; a flag specifies which mode we are running in.
1713
1714 Arguments:
1715 code points to the start of the pattern (the bracket)
1716 utf TRUE in UTF-8 / UTF-16 / UTF-32 mode
1717 atend TRUE if called when the pattern is complete
1718 cd the "compile data" structure
1719
1720 Returns: the fixed length,
1721 or -1 if there is no fixed length,
1722 or -2 if \C was encountered (in UTF-8 mode only)
1723 or -3 if an OP_RECURSE item was encountered and atend is FALSE
1724 or -4 if an unknown opcode was encountered (internal error)
1725 */
1726
1727 static int
1728 find_fixedlength(pcre_uchar *code, BOOL utf, BOOL atend, compile_data *cd)
1729 {
1730 int length = -1;
1731
1732 register int branchlength = 0;
1733 register pcre_uchar *cc = code + 1 + LINK_SIZE;
1734
1735 /* Scan along the opcodes for this branch. If we get to the end of the
1736 branch, check the length against that of the other branches. */
1737
1738 for (;;)
1739 {
1740 int d;
1741 pcre_uchar *ce, *cs;
1742 register int op = *cc;
1743
1744 switch (op)
1745 {
1746 /* We only need to continue for OP_CBRA (normal capturing bracket) and
1747 OP_BRA (normal non-capturing bracket) because the other variants of these
1748 opcodes are all concerned with unlimited repeated groups, which of course
1749 are not of fixed length. */
1750
1751 case OP_CBRA:
1752 case OP_BRA:
1753 case OP_ONCE:
1754 case OP_ONCE_NC:
1755 case OP_COND:
1756 d = find_fixedlength(cc + ((op == OP_CBRA)? IMM2_SIZE : 0), utf, atend, cd);
1757 if (d < 0) return d;
1758 branchlength += d;
1759 do cc += GET(cc, 1); while (*cc == OP_ALT);
1760 cc += 1 + LINK_SIZE;
1761 break;
1762
1763 /* Reached end of a branch; if it's a ket it is the end of a nested call.
1764 If it's ALT it is an alternation in a nested call. An ACCEPT is effectively
1765 an ALT. If it is END it's the end of the outer call. All can be handled by
1766 the same code. Note that we must not include the OP_KETRxxx opcodes here,
1767 because they all imply an unlimited repeat. */
1768
1769 case OP_ALT:
1770 case OP_KET:
1771 case OP_END:
1772 case OP_ACCEPT:
1773 case OP_ASSERT_ACCEPT:
1774 if (length < 0) length = branchlength;
1775 else if (length != branchlength) return -1;
1776 if (*cc != OP_ALT) return length;
1777 cc += 1 + LINK_SIZE;
1778 branchlength = 0;
1779 break;
1780
1781 /* A true recursion implies not fixed length, but a subroutine call may
1782 be OK. If the subroutine is a forward reference, we can't deal with
1783 it until the end of the pattern, so return -3. */
1784
1785 case OP_RECURSE:
1786 if (!atend) return -3;
1787 cs = ce = (pcre_uchar *)cd->start_code + GET(cc, 1); /* Start subpattern */
1788 do ce += GET(ce, 1); while (*ce == OP_ALT); /* End subpattern */
1789 if (cc > cs && cc < ce) return -1; /* Recursion */
1790 d = find_fixedlength(cs + IMM2_SIZE, utf, atend, cd);
1791 if (d < 0) return d;
1792 branchlength += d;
1793 cc += 1 + LINK_SIZE;
1794 break;
1795
1796 /* Skip over assertive subpatterns */
1797
1798 case OP_ASSERT:
1799 case OP_ASSERT_NOT:
1800 case OP_ASSERTBACK:
1801 case OP_ASSERTBACK_NOT:
1802 do cc += GET(cc, 1); while (*cc == OP_ALT);
1803 cc += PRIV(OP_lengths)[*cc];
1804 break;
1805
1806 /* Skip over things that don't match chars */
1807
1808 case OP_MARK:
1809 case OP_PRUNE_ARG:
1810 case OP_SKIP_ARG:
1811 case OP_THEN_ARG:
1812 cc += cc[1] + PRIV(OP_lengths)[*cc];
1813 break;
1814
1815 case OP_CALLOUT:
1816 case OP_CIRC:
1817 case OP_CIRCM:
1818 case OP_CLOSE:
1819 case OP_COMMIT:
1820 case OP_CREF:
1821 case OP_DEF:
1822 case OP_DOLL:
1823 case OP_DOLLM:
1824 case OP_EOD:
1825 case OP_EODN:
1826 case OP_FAIL:
1827 case OP_NCREF:
1828 case OP_NRREF:
1829 case OP_NOT_WORD_BOUNDARY:
1830 case OP_PRUNE:
1831 case OP_REVERSE:
1832 case OP_RREF:
1833 case OP_SET_SOM:
1834 case OP_SKIP:
1835 case OP_SOD:
1836 case OP_SOM:
1837 case OP_THEN:
1838 case OP_WORD_BOUNDARY:
1839 cc += PRIV(OP_lengths)[*cc];
1840 break;
1841
1842 /* Handle literal characters */
1843
1844 case OP_CHAR:
1845 case OP_CHARI:
1846 case OP_NOT:
1847 case OP_NOTI:
1848 branchlength++;
1849 cc += 2;
1850 #if defined SUPPORT_UTF && !defined COMPILE_PCRE32
1851 if (utf && HAS_EXTRALEN(cc[-1])) cc += GET_EXTRALEN(cc[-1]);
1852 #endif
1853 break;
1854
1855 /* Handle exact repetitions. The count is already in characters, but we
1856 need to skip over a multibyte character in UTF8 mode. */
1857
1858 case OP_EXACT:
1859 case OP_EXACTI:
1860 case OP_NOTEXACT:
1861 case OP_NOTEXACTI:
1862 branchlength += GET2(cc,1);
1863 cc += 2 + IMM2_SIZE;
1864 #if defined SUPPORT_UTF && !defined COMPILE_PCRE32
1865 if (utf && HAS_EXTRALEN(cc[-1])) cc += GET_EXTRALEN(cc[-1]);
1866 #endif
1867 break;
1868
1869 case OP_TYPEEXACT:
1870 branchlength += GET2(cc,1);
1871 if (cc[1 + IMM2_SIZE] == OP_PROP || cc[1 + IMM2_SIZE] == OP_NOTPROP)
1872 cc += 2;
1873 cc += 1 + IMM2_SIZE + 1;
1874 break;
1875
1876 /* Handle single-char matchers */
1877
1878 case OP_PROP:
1879 case OP_NOTPROP:
1880 cc += 2;
1881 /* Fall through */
1882
1883 case OP_HSPACE:
1884 case OP_VSPACE:
1885 case OP_NOT_HSPACE:
1886 case OP_NOT_VSPACE:
1887 case OP_NOT_DIGIT:
1888 case OP_DIGIT:
1889 case OP_NOT_WHITESPACE:
1890 case OP_WHITESPACE:
1891 case OP_NOT_WORDCHAR:
1892 case OP_WORDCHAR:
1893 case OP_ANY:
1894 case OP_ALLANY:
1895 branchlength++;
1896 cc++;
1897 break;
1898
1899 /* The single-byte matcher isn't allowed. This only happens in UTF-8 mode;
1900 otherwise \C is coded as OP_ALLANY. */
1901
1902 case OP_ANYBYTE:
1903 return -2;
1904
1905 /* Check a class for variable quantification */
1906
1907 #if defined SUPPORT_UTF || defined COMPILE_PCRE16 || defined COMPILE_PCRE32
1908 case OP_XCLASS:
1909 cc += GET(cc, 1) - PRIV(OP_lengths)[OP_CLASS];
1910 /* Fall through */
1911 #endif
1912
1913 case OP_CLASS:
1914 case OP_NCLASS:
1915 cc += PRIV(OP_lengths)[OP_CLASS];
1916
1917 switch (*cc)
1918 {
1919 case OP_CRPLUS:
1920 case OP_CRMINPLUS:
1921 case OP_CRSTAR:
1922 case OP_CRMINSTAR:
1923 case OP_CRQUERY:
1924 case OP_CRMINQUERY:
1925 return -1;
1926
1927 case OP_CRRANGE:
1928 case OP_CRMINRANGE:
1929 if (GET2(cc,1) != GET2(cc,1+IMM2_SIZE)) return -1;
1930 branchlength += GET2(cc,1);
1931 cc += 1 + 2 * IMM2_SIZE;
1932 break;
1933
1934 default:
1935 branchlength++;
1936 }
1937 break;
1938
1939 /* Anything else is variable length */
1940
1941 case OP_ANYNL:
1942 case OP_BRAMINZERO:
1943 case OP_BRAPOS:
1944 case OP_BRAPOSZERO:
1945 case OP_BRAZERO:
1946 case OP_CBRAPOS:
1947 case OP_EXTUNI:
1948 case OP_KETRMAX:
1949 case OP_KETRMIN:
1950 case OP_KETRPOS:
1951 case OP_MINPLUS:
1952 case OP_MINPLUSI:
1953 case OP_MINQUERY:
1954 case OP_MINQUERYI:
1955 case OP_MINSTAR:
1956 case OP_MINSTARI:
1957 case OP_MINUPTO:
1958 case OP_MINUPTOI:
1959 case OP_NOTMINPLUS:
1960 case OP_NOTMINPLUSI:
1961 case OP_NOTMINQUERY:
1962 case OP_NOTMINQUERYI:
1963 case OP_NOTMINSTAR:
1964 case OP_NOTMINSTARI:
1965 case OP_NOTMINUPTO:
1966 case OP_NOTMINUPTOI:
1967 case OP_NOTPLUS:
1968 case OP_NOTPLUSI:
1969 case OP_NOTPOSPLUS:
1970 case OP_NOTPOSPLUSI:
1971 case OP_NOTPOSQUERY:
1972 case OP_NOTPOSQUERYI:
1973 case OP_NOTPOSSTAR:
1974 case OP_NOTPOSSTARI:
1975 case OP_NOTPOSUPTO:
1976 case OP_NOTPOSUPTOI:
1977 case OP_NOTQUERY:
1978 case OP_NOTQUERYI:
1979 case OP_NOTSTAR:
1980 case OP_NOTSTARI:
1981 case OP_NOTUPTO:
1982 case OP_NOTUPTOI:
1983 case OP_PLUS:
1984 case OP_PLUSI:
1985 case OP_POSPLUS:
1986 case OP_POSPLUSI:
1987 case OP_POSQUERY:
1988 case OP_POSQUERYI:
1989 case OP_POSSTAR:
1990 case OP_POSSTARI:
1991 case OP_POSUPTO:
1992 case OP_POSUPTOI:
1993 case OP_QUERY:
1994 case OP_QUERYI:
1995 case OP_REF:
1996 case OP_REFI:
1997 case OP_SBRA:
1998 case OP_SBRAPOS:
1999 case OP_SCBRA:
2000 case OP_SCBRAPOS:
2001 case OP_SCOND:
2002 case OP_SKIPZERO:
2003 case OP_STAR:
2004 case OP_STARI:
2005 case OP_TYPEMINPLUS:
2006 case OP_TYPEMINQUERY:
2007 case OP_TYPEMINSTAR:
2008 case OP_TYPEMINUPTO:
2009 case OP_TYPEPLUS:
2010 case OP_TYPEPOSPLUS:
2011 case OP_TYPEPOSQUERY:
2012 case OP_TYPEPOSSTAR:
2013 case OP_TYPEPOSUPTO:
2014 case OP_TYPEQUERY:
2015 case OP_TYPESTAR:
2016 case OP_TYPEUPTO:
2017 case OP_UPTO:
2018 case OP_UPTOI:
2019 return -1;
2020
2021 /* Catch unrecognized opcodes so that when new ones are added they
2022 are not forgotten, as has happened in the past. */
2023
2024 default:
2025 return -4;
2026 }
2027 }
2028 /* Control never gets here */
2029 }
2030
2031
2032
2033
2034 /*************************************************
2035 * Scan compiled regex for specific bracket *
2036 *************************************************/
2037
2038 /* This little function scans through a compiled pattern until it finds a
2039 capturing bracket with the given number, or, if the number is negative, an
2040 instance of OP_REVERSE for a lookbehind. The function is global in the C sense
2041 so that it can be called from pcre_study() when finding the minimum matching
2042 length.
2043
2044 Arguments:
2045 code points to start of expression
2046 utf TRUE in UTF-8 / UTF-16 / UTF-32 mode
2047 number the required bracket number or negative to find a lookbehind
2048
2049 Returns: pointer to the opcode for the bracket, or NULL if not found
2050 */
2051
2052 const pcre_uchar *
2053 PRIV(find_bracket)(const pcre_uchar *code, BOOL utf, int number)
2054 {
2055 for (;;)
2056 {
2057 register int c = *code;
2058
2059 if (c == OP_END) return NULL;
2060
2061 /* XCLASS is used for classes that cannot be represented just by a bit
2062 map. This includes negated single high-valued characters. The length in
2063 the table is zero; the actual length is stored in the compiled code. */
2064
2065 if (c == OP_XCLASS) code += GET(code, 1);
2066
2067 /* Handle recursion */
2068
2069 else if (c == OP_REVERSE)
2070 {
2071 if (number < 0) return (pcre_uchar *)code;
2072 code += PRIV(OP_lengths)[c];
2073 }
2074
2075 /* Handle capturing bracket */
2076
2077 else if (c == OP_CBRA || c == OP_SCBRA ||
2078 c == OP_CBRAPOS || c == OP_SCBRAPOS)
2079 {
2080 int n = GET2(code, 1+LINK_SIZE);
2081 if (n == number) return (pcre_uchar *)code;
2082 code += PRIV(OP_lengths)[c];
2083 }
2084
2085 /* Otherwise, we can get the item's length from the table, except that for
2086 repeated character types, we have to test for \p and \P, which have an extra
2087 two bytes of parameters, and for MARK/PRUNE/SKIP/THEN with an argument, we
2088 must add in its length. */
2089
2090 else
2091 {
2092 switch(c)
2093 {
2094 case OP_TYPESTAR:
2095 case OP_TYPEMINSTAR:
2096 case OP_TYPEPLUS:
2097 case OP_TYPEMINPLUS:
2098 case OP_TYPEQUERY:
2099 case OP_TYPEMINQUERY:
2100 case OP_TYPEPOSSTAR:
2101 case OP_TYPEPOSPLUS:
2102 case OP_TYPEPOSQUERY:
2103 if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
2104 break;
2105
2106 case OP_TYPEUPTO:
2107 case OP_TYPEMINUPTO:
2108 case OP_TYPEEXACT:
2109 case OP_TYPEPOSUPTO:
2110 if (code[1 + IMM2_SIZE] == OP_PROP || code[1 + IMM2_SIZE] == OP_NOTPROP)
2111 code += 2;
2112 break;
2113
2114 case OP_MARK:
2115 case OP_PRUNE_ARG:
2116 case OP_SKIP_ARG:
2117 code += code[1];
2118 break;
2119
2120 case OP_THEN_ARG:
2121 code += code[1];
2122 break;
2123 }
2124
2125 /* Add in the fixed length from the table */
2126
2127 code += PRIV(OP_lengths)[c];
2128
2129 /* In UTF-8 mode, opcodes that are followed by a character may be followed by
2130 a multi-byte character. The length in the table is a minimum, so we have to
2131 arrange to skip the extra bytes. */
2132
2133 #if defined SUPPORT_UTF && !defined COMPILE_PCRE32
2134 if (utf) switch(c)
2135 {
2136 case OP_CHAR:
2137 case OP_CHARI:
2138 case OP_EXACT:
2139 case OP_EXACTI:
2140 case OP_UPTO:
2141 case OP_UPTOI:
2142 case OP_MINUPTO:
2143 case OP_MINUPTOI:
2144 case OP_POSUPTO:
2145 case OP_POSUPTOI:
2146 case OP_STAR:
2147 case OP_STARI:
2148 case OP_MINSTAR:
2149 case OP_MINSTARI:
2150 case OP_POSSTAR:
2151 case OP_POSSTARI:
2152 case OP_PLUS:
2153 case OP_PLUSI:
2154 case OP_MINPLUS:
2155 case OP_MINPLUSI:
2156 case OP_POSPLUS:
2157 case OP_POSPLUSI:
2158 case OP_QUERY:
2159 case OP_QUERYI:
2160 case OP_MINQUERY:
2161 case OP_MINQUERYI:
2162 case OP_POSQUERY:
2163 case OP_POSQUERYI:
2164 if (HAS_EXTRALEN(code[-1])) code += GET_EXTRALEN(code[-1]);
2165 break;
2166 }
2167 #else
2168 (void)(utf); /* Keep compiler happy by referencing function argument */
2169 #endif
2170 }
2171 }
2172 }
2173
2174
2175
2176 /*************************************************
2177 * Scan compiled regex for recursion reference *
2178 *************************************************/
2179
2180 /* This little function scans through a compiled pattern until it finds an
2181 instance of OP_RECURSE.
2182
2183 Arguments:
2184 code points to start of expression
2185 utf TRUE in UTF-8 / UTF-16 / UTF-32 mode
2186
2187 Returns: pointer to the opcode for OP_RECURSE, or NULL if not found
2188 */
2189
2190 static const pcre_uchar *
2191 find_recurse(const pcre_uchar *code, BOOL utf)
2192 {
2193 for (;;)
2194 {
2195 register int c = *code;
2196 if (c == OP_END) return NULL;
2197 if (c == OP_RECURSE) return code;
2198
2199 /* XCLASS is used for classes that cannot be represented just by a bit
2200 map. This includes negated single high-valued characters. The length in
2201 the table is zero; the actual length is stored in the compiled code. */
2202
2203 if (c == OP_XCLASS) code += GET(code, 1);
2204
2205 /* Otherwise, we can get the item's length from the table, except that for
2206 repeated character types, we have to test for \p and \P, which have an extra
2207 two bytes of parameters, and for MARK/PRUNE/SKIP/THEN with an argument, we
2208 must add in its length. */
2209
2210 else
2211 {
2212 switch(c)
2213 {
2214 case OP_TYPESTAR:
2215 case OP_TYPEMINSTAR:
2216 case OP_TYPEPLUS:
2217 case OP_TYPEMINPLUS:
2218 case OP_TYPEQUERY:
2219 case OP_TYPEMINQUERY:
2220 case OP_TYPEPOSSTAR:
2221 case OP_TYPEPOSPLUS:
2222 case OP_TYPEPOSQUERY:
2223 if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
2224 break;
2225
2226 case OP_TYPEPOSUPTO:
2227 case OP_TYPEUPTO:
2228 case OP_TYPEMINUPTO:
2229 case OP_TYPEEXACT:
2230 if (code[1 + IMM2_SIZE] == OP_PROP || code[1 + IMM2_SIZE] == OP_NOTPROP)
2231 code += 2;
2232 break;
2233
2234 case OP_MARK:
2235 case OP_PRUNE_ARG:
2236 case OP_SKIP_ARG:
2237 code += code[1];
2238 break;
2239
2240 case OP_THEN_ARG:
2241 code += code[1];
2242 break;
2243 }
2244
2245 /* Add in the fixed length from the table */
2246
2247 code += PRIV(OP_lengths)[c];
2248
2249 /* In UTF-8 mode, opcodes that are followed by a character may be followed
2250 by a multi-byte character. The length in the table is a minimum, so we have
2251 to arrange to skip the extra bytes. */
2252
2253 #if defined SUPPORT_UTF && !defined COMPILE_PCRE32
2254 if (utf) switch(c)
2255 {
2256 case OP_CHAR:
2257 case OP_CHARI:
2258 case OP_NOT:
2259 case OP_NOTI:
2260 case OP_EXACT:
2261 case OP_EXACTI:
2262 case OP_NOTEXACT:
2263 case OP_NOTEXACTI:
2264 case OP_UPTO:
2265 case OP_UPTOI:
2266 case OP_NOTUPTO:
2267 case OP_NOTUPTOI:
2268 case OP_MINUPTO:
2269 case OP_MINUPTOI:
2270 case OP_NOTMINUPTO:
2271 case OP_NOTMINUPTOI:
2272 case OP_POSUPTO:
2273 case OP_POSUPTOI:
2274 case OP_NOTPOSUPTO:
2275 case OP_NOTPOSUPTOI:
2276 case OP_STAR:
2277 case OP_STARI:
2278 case OP_NOTSTAR:
2279 case OP_NOTSTARI:
2280 case OP_MINSTAR:
2281 case OP_MINSTARI:
2282 case OP_NOTMINSTAR:
2283 case OP_NOTMINSTARI:
2284 case OP_POSSTAR:
2285 case OP_POSSTARI:
2286 case OP_NOTPOSSTAR:
2287 case OP_NOTPOSSTARI:
2288 case OP_PLUS:
2289 case OP_PLUSI:
2290 case OP_NOTPLUS:
2291 case OP_NOTPLUSI:
2292 case OP_MINPLUS:
2293 case OP_MINPLUSI:
2294 case OP_NOTMINPLUS:
2295 case OP_NOTMINPLUSI:
2296 case OP_POSPLUS:
2297 case OP_POSPLUSI:
2298 case OP_NOTPOSPLUS:
2299 case OP_NOTPOSPLUSI:
2300 case OP_QUERY:
2301 case OP_QUERYI:
2302 case OP_NOTQUERY:
2303 case OP_NOTQUERYI:
2304 case OP_MINQUERY:
2305 case OP_MINQUERYI:
2306 case OP_NOTMINQUERY:
2307 case OP_NOTMINQUERYI:
2308 case OP_POSQUERY:
2309 case OP_POSQUERYI:
2310 case OP_NOTPOSQUERY:
2311 case OP_NOTPOSQUERYI:
2312 if (HAS_EXTRALEN(code[-1])) code += GET_EXTRALEN(code[-1]);
2313 break;
2314 }
2315 #else
2316 (void)(utf); /* Keep compiler happy by referencing function argument */
2317 #endif
2318 }
2319 }
2320 }
2321
2322
2323
2324 /*************************************************
2325 * Scan compiled branch for non-emptiness *
2326 *************************************************/
2327
2328 /* This function scans through a branch of a compiled pattern to see whether it
2329 can match the empty string or not. It is called from could_be_empty()
2330 below and from compile_branch() when checking for an unlimited repeat of a
2331 group that can match nothing. Note that first_significant_code() skips over
2332 backward and negative forward assertions when its final argument is TRUE. If we
2333 hit an unclosed bracket, we return "empty" - this means we've struck an inner
2334 bracket whose current branch will already have been scanned.
2335
2336 Arguments:
2337 code points to start of search
2338 endcode points to where to stop
2339 utf TRUE if in UTF-8 / UTF-16 / UTF-32 mode
2340 cd contains pointers to tables etc.
2341
2342 Returns: TRUE if what is matched could be empty
2343 */
2344
2345 static BOOL
2346 could_be_empty_branch(const pcre_uchar *code, const pcre_uchar *endcode,
2347 BOOL utf, compile_data *cd)
2348 {
2349 register int c;
2350 for (code = first_significant_code(code + PRIV(OP_lengths)[*code], TRUE);
2351 code < endcode;
2352 code = first_significant_code(code + PRIV(OP_lengths)[c], TRUE))
2353 {
2354 const pcre_uchar *ccode;
2355
2356 c = *code;
2357
2358 /* Skip over forward assertions; the other assertions are skipped by
2359 first_significant_code() with a TRUE final argument. */
2360
2361 if (c == OP_ASSERT)
2362 {
2363 do code += GET(code, 1); while (*code == OP_ALT);
2364 c = *code;
2365 continue;
2366 }
2367
2368 /* For a recursion/subroutine call, if its end has been reached, which
2369 implies a backward reference subroutine call, we can scan it. If it's a
2370 forward reference subroutine call, we can't. To detect forward reference
2371 we have to scan up the list that is kept in the workspace. This function is
2372 called only when doing the real compile, not during the pre-compile that
2373 measures the size of the compiled pattern. */
2374
2375 if (c == OP_RECURSE)
2376 {
2377 const pcre_uchar *scode;
2378 BOOL empty_branch;
2379
2380 /* Test for forward reference */
2381
2382 for (scode = cd->start_workspace; scode < cd->hwm; scode += LINK_SIZE)
2383 if (GET(scode, 0) == code + 1 - cd->start_code) return TRUE;
2384
2385 /* Not a forward reference, test for completed backward reference */
2386
2387 empty_branch = FALSE;
2388 scode = cd->start_code + GET(code, 1);
2389 if (GET(scode, 1) == 0) return TRUE; /* Unclosed */
2390
2391 /* Completed backwards reference */
2392
2393 do
2394 {
2395 if (could_be_empty_branch(scode, endcode, utf, cd))
2396 {
2397 empty_branch = TRUE;
2398 break;
2399 }
2400 scode += GET(scode, 1);
2401 }
2402 while (*scode == OP_ALT);
2403
2404 if (!empty_branch) return FALSE; /* All branches are non-empty */
2405 continue;
2406 }
2407
2408 /* Groups with zero repeats can of course be empty; skip them. */
2409
2410 if (c == OP_BRAZERO || c == OP_BRAMINZERO || c == OP_SKIPZERO ||
2411 c == OP_BRAPOSZERO)
2412 {
2413 code += PRIV(OP_lengths)[c];
2414 do code += GET(code, 1); while (*code == OP_ALT);
2415 c = *code;
2416 continue;
2417 }
2418
2419 /* A nested group that is already marked as "could be empty" can just be
2420 skipped. */
2421
2422 if (c == OP_SBRA || c == OP_SBRAPOS ||
2423 c == OP_SCBRA || c == OP_SCBRAPOS)
2424 {
2425 do code += GET(code, 1); while (*code == OP_ALT);
2426 c = *code;
2427 continue;
2428 }
2429
2430 /* For other groups, scan the branches. */
2431
2432 if (c == OP_BRA || c == OP_BRAPOS ||
2433 c == OP_CBRA || c == OP_CBRAPOS ||
2434 c == OP_ONCE || c == OP_ONCE_NC ||
2435 c == OP_COND)
2436 {
2437 BOOL empty_branch;
2438 if (GET(code, 1) == 0) return TRUE; /* Hit unclosed bracket */
2439
2440 /* If a conditional group has only one branch, there is a second, implied,
2441 empty branch, so just skip over the conditional, because it could be empty.
2442 Otherwise, scan the individual branches of the group. */
2443
2444 if (c == OP_COND && code[GET(code, 1)] != OP_ALT)
2445 code += GET(code, 1);
2446 else
2447 {
2448 empty_branch = FALSE;
2449 do
2450 {
2451 if (!empty_branch && could_be_empty_branch(code, endcode, utf, cd))
2452 empty_branch = TRUE;
2453 code += GET(code, 1);
2454 }
2455 while (*code == OP_ALT);
2456 if (!empty_branch) return FALSE; /* All branches are non-empty */
2457 }
2458
2459 c = *code;
2460 continue;
2461 }
2462
2463 /* Handle the other opcodes */
2464
2465 switch (c)
2466 {
2467 /* Check for quantifiers after a class. XCLASS is used for classes that
2468 cannot be represented just by a bit map. This includes negated single
2469 high-valued characters. The length in PRIV(OP_lengths)[] is zero; the
2470 actual length is stored in the compiled code, so we must update "code"
2471 here. */
2472
2473 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
2474 case OP_XCLASS:
2475 ccode = code += GET(code, 1);
2476 goto CHECK_CLASS_REPEAT;
2477 #endif
2478
2479 case OP_CLASS:
2480 case OP_NCLASS:
2481 ccode = code + PRIV(OP_lengths)[OP_CLASS];
2482
2483 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
2484 CHECK_CLASS_REPEAT:
2485 #endif
2486
2487 switch (*ccode)
2488 {
2489 case OP_CRSTAR: /* These could be empty; continue */
2490 case OP_CRMINSTAR:
2491 case OP_CRQUERY:
2492 case OP_CRMINQUERY:
2493 break;
2494
2495 default: /* Non-repeat => class must match */
2496 case OP_CRPLUS: /* These repeats aren't empty */
2497 case OP_CRMINPLUS:
2498 return FALSE;
2499
2500 case OP_CRRANGE:
2501 case OP_CRMINRANGE:
2502 if (GET2(ccode, 1) > 0) return FALSE; /* Minimum > 0 */
2503 break;
2504 }
2505 break;
2506
2507 /* Opcodes that must match a character */
2508
2509 case OP_PROP:
2510 case OP_NOTPROP:
2511 case OP_EXTUNI:
2512 case OP_NOT_DIGIT:
2513 case OP_DIGIT:
2514 case OP_NOT_WHITESPACE:
2515 case OP_WHITESPACE:
2516 case OP_NOT_WORDCHAR:
2517 case OP_WORDCHAR:
2518 case OP_ANY:
2519 case OP_ALLANY:
2520 case OP_ANYBYTE:
2521 case OP_CHAR:
2522 case OP_CHARI:
2523 case OP_NOT:
2524 case OP_NOTI:
2525 case OP_PLUS:
2526 case OP_MINPLUS:
2527 case OP_POSPLUS:
2528 case OP_EXACT:
2529 case OP_NOTPLUS:
2530 case OP_NOTMINPLUS:
2531 case OP_NOTPOSPLUS:
2532 case OP_NOTEXACT:
2533 case OP_TYPEPLUS:
2534 case OP_TYPEMINPLUS:
2535 case OP_TYPEPOSPLUS:
2536 case OP_TYPEEXACT:
2537 return FALSE;
2538
2539 /* These are going to continue, as they may be empty, but we have to
2540 fudge the length for the \p and \P cases. */
2541
2542 case OP_TYPESTAR:
2543 case OP_TYPEMINSTAR:
2544 case OP_TYPEPOSSTAR:
2545 case OP_TYPEQUERY:
2546 case OP_TYPEMINQUERY:
2547 case OP_TYPEPOSQUERY:
2548 if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
2549 break;
2550
2551 /* Same for these */
2552
2553 case OP_TYPEUPTO:
2554 case OP_TYPEMINUPTO:
2555 case OP_TYPEPOSUPTO:
2556 if (code[1 + IMM2_SIZE] == OP_PROP || code[1 + IMM2_SIZE] == OP_NOTPROP)
2557 code += 2;
2558 break;
2559
2560 /* End of branch */
2561
2562 case OP_KET:
2563 case OP_KETRMAX:
2564 case OP_KETRMIN:
2565 case OP_KETRPOS:
2566 case OP_ALT:
2567 return TRUE;
2568
2569 /* In UTF-8 mode, STAR, MINSTAR, POSSTAR, QUERY, MINQUERY, POSQUERY, UPTO,
2570 MINUPTO, and POSUPTO may be followed by a multibyte character */
2571
2572 #if defined SUPPORT_UTF && !defined COMPILE_PCRE32
2573 case OP_STAR:
2574 case OP_STARI:
2575 case OP_MINSTAR:
2576 case OP_MINSTARI:
2577 case OP_POSSTAR:
2578 case OP_POSSTARI:
2579 case OP_QUERY:
2580 case OP_QUERYI:
2581 case OP_MINQUERY:
2582 case OP_MINQUERYI:
2583 case OP_POSQUERY:
2584 case OP_POSQUERYI:
2585 if (utf && HAS_EXTRALEN(code[1])) code += GET_EXTRALEN(code[1]);
2586 break;
2587
2588 case OP_UPTO:
2589 case OP_UPTOI:
2590 case OP_MINUPTO:
2591 case OP_MINUPTOI:
2592 case OP_POSUPTO:
2593 case OP_POSUPTOI:
2594 if (utf && HAS_EXTRALEN(code[1 + IMM2_SIZE])) code += GET_EXTRALEN(code[1 + IMM2_SIZE]);
2595 break;
2596 #endif
2597
2598 /* MARK, and PRUNE/SKIP/THEN with an argument must skip over the argument
2599 string. */
2600
2601 case OP_MARK:
2602 case OP_PRUNE_ARG:
2603 case OP_SKIP_ARG:
2604 code += code[1];
2605 break;
2606
2607 case OP_THEN_ARG:
2608 code += code[1];
2609 break;
2610
2611 /* None of the remaining opcodes are required to match a character. */
2612
2613 default:
2614 break;
2615 }
2616 }
2617
2618 return TRUE;
2619 }
2620
2621
2622
2623 /*************************************************
2624 * Scan compiled regex for non-emptiness *
2625 *************************************************/
2626
2627 /* This function is called to check for left recursive calls. We want to check
2628 the current branch of the current pattern to see if it could match the empty
2629 string. If it could, we must look outwards for branches at other levels,
2630 stopping when we pass beyond the bracket which is the subject of the recursion.
2631 This function is called only during the real compile, not during the
2632 pre-compile.
2633
2634 Arguments:
2635 code points to start of the recursion
2636 endcode points to where to stop (current RECURSE item)
2637 bcptr points to the chain of current (unclosed) branch starts
2638 utf TRUE if in UTF-8 / UTF-16 / UTF-32 mode
2639 cd pointers to tables etc
2640
2641 Returns: TRUE if what is matched could be empty
2642 */
2643
2644 static BOOL
2645 could_be_empty(const pcre_uchar *code, const pcre_uchar *endcode,
2646 branch_chain *bcptr, BOOL utf, compile_data *cd)
2647 {
2648 while (bcptr != NULL && bcptr->current_branch >= code)
2649 {
2650 if (!could_be_empty_branch(bcptr->current_branch, endcode, utf, cd))
2651 return FALSE;
2652 bcptr = bcptr->outer;
2653 }
2654 return TRUE;
2655 }
2656
2657
2658
2659 /*************************************************
2660 * Check for POSIX class syntax *
2661 *************************************************/
2662
2663 /* This function is called when the sequence "[:" or "[." or "[=" is
2664 encountered in a character class. It checks whether this is followed by a
2665 sequence of characters terminated by a matching ":]" or ".]" or "=]". If we
2666 reach an unescaped ']' without the special preceding character, return FALSE.
2667
2668 Originally, this function only recognized a sequence of letters between the
2669 terminators, but it seems that Perl recognizes any sequence of characters,
2670 though of course unknown POSIX names are subsequently rejected. Perl gives an
2671 "Unknown POSIX class" error for [:f\oo:] for example, where previously PCRE
2672 didn't consider this to be a POSIX class. Likewise for [:1234:].
2673
2674 The problem in trying to be exactly like Perl is in the handling of escapes. We
2675 have to be sure that [abc[:x\]pqr] is *not* treated as containing a POSIX
2676 class, but [abc[:x\]pqr:]] is (so that an error can be generated). The code
2677 below handles the special case of \], but does not try to do any other escape
2678 processing. This makes it different from Perl for cases such as [:l\ower:]
2679 where Perl recognizes it as the POSIX class "lower" but PCRE does not recognize
2680 "l\ower". This is a lesser evil that not diagnosing bad classes when Perl does,
2681 I think.
2682
2683 A user pointed out that PCRE was rejecting [:a[:digit:]] whereas Perl was not.
2684 It seems that the appearance of a nested POSIX class supersedes an apparent
2685 external class. For example, [:a[:digit:]b:] matches "a", "b", ":", or
2686 a digit.
2687
2688 In Perl, unescaped square brackets may also appear as part of class names. For
2689 example, [:a[:abc]b:] gives unknown POSIX class "[:abc]b:]". However, for
2690 [:a[:abc]b][b:] it gives unknown POSIX class "[:abc]b][b:]", which does not
2691 seem right at all. PCRE does not allow closing square brackets in POSIX class
2692 names.
2693
2694 Arguments:
2695 ptr pointer to the initial [
2696 endptr where to return the end pointer
2697
2698 Returns: TRUE or FALSE
2699 */
2700
2701 static BOOL
2702 check_posix_syntax(const pcre_uchar *ptr, const pcre_uchar **endptr)
2703 {
2704 int terminator; /* Don't combine these lines; the Solaris cc */
2705 terminator = *(++ptr); /* compiler warns about "non-constant" initializer. */
2706 for (++ptr; *ptr != 0; ptr++)
2707 {
2708 if (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET)
2709 ptr++;
2710 else if (*ptr == CHAR_RIGHT_SQUARE_BRACKET) return FALSE;
2711 else
2712 {
2713 if (*ptr == terminator && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET)
2714 {
2715 *endptr = ptr;
2716 return TRUE;
2717 }
2718 if (*ptr == CHAR_LEFT_SQUARE_BRACKET &&
2719 (ptr[1] == CHAR_COLON || ptr[1] == CHAR_DOT ||
2720 ptr[1] == CHAR_EQUALS_SIGN) &&
2721 check_posix_syntax(ptr, endptr))
2722 return FALSE;
2723 }
2724 }
2725 return FALSE;
2726 }
2727
2728
2729
2730
2731 /*************************************************
2732 * Check POSIX class name *
2733 *************************************************/
2734
2735 /* This function is called to check the name given in a POSIX-style class entry
2736 such as [:alnum:].
2737
2738 Arguments:
2739 ptr points to the first letter
2740 len the length of the name
2741
2742 Returns: a value representing the name, or -1 if unknown
2743 */
2744
2745 static int
2746 check_posix_name(const pcre_uchar *ptr, int len)
2747 {
2748 const char *pn = posix_names;
2749 register int yield = 0;
2750 while (posix_name_lengths[yield] != 0)
2751 {
2752 if (len == posix_name_lengths[yield] &&
2753 STRNCMP_UC_C8(ptr, pn, len) == 0) return yield;
2754 pn += posix_name_lengths[yield] + 1;
2755 yield++;
2756 }
2757 return -1;
2758 }
2759
2760
2761 /*************************************************
2762 * Adjust OP_RECURSE items in repeated group *
2763 *************************************************/
2764
2765 /* OP_RECURSE items contain an offset from the start of the regex to the group
2766 that is referenced. This means that groups can be replicated for fixed
2767 repetition simply by copying (because the recursion is allowed to refer to
2768 earlier groups that are outside the current group). However, when a group is
2769 optional (i.e. the minimum quantifier is zero), OP_BRAZERO or OP_SKIPZERO is
2770 inserted before it, after it has been compiled. This means that any OP_RECURSE
2771 items within it that refer to the group itself or any contained groups have to
2772 have their offsets adjusted. That one of the jobs of this function. Before it
2773 is called, the partially compiled regex must be temporarily terminated with
2774 OP_END.
2775
2776 This function has been extended with the possibility of forward references for
2777 recursions and subroutine calls. It must also check the list of such references
2778 for the group we are dealing with. If it finds that one of the recursions in
2779 the current group is on this list, it adjusts the offset in the list, not the
2780 value in the reference (which is a group number).
2781
2782 Arguments:
2783 group points to the start of the group
2784 adjust the amount by which the group is to be moved
2785 utf TRUE in UTF-8 / UTF-16 / UTF-32 mode
2786 cd contains pointers to tables etc.
2787 save_hwm the hwm forward reference pointer at the start of the group
2788
2789 Returns: nothing
2790 */
2791
2792 static void
2793 adjust_recurse(pcre_uchar *group, int adjust, BOOL utf, compile_data *cd,
2794 pcre_uchar *save_hwm)
2795 {
2796 pcre_uchar *ptr = group;
2797
2798 while ((ptr = (pcre_uchar *)find_recurse(ptr, utf)) != NULL)
2799 {
2800 int offset;
2801 pcre_uchar *hc;
2802
2803 /* See if this recursion is on the forward reference list. If so, adjust the
2804 reference. */
2805
2806 for (hc = save_hwm; hc < cd->hwm; hc += LINK_SIZE)
2807 {
2808 offset = GET(hc, 0);
2809 if (cd->start_code + offset == ptr + 1)
2810 {
2811 PUT(hc, 0, offset + adjust);
2812 break;
2813 }
2814 }
2815
2816 /* Otherwise, adjust the recursion offset if it's after the start of this
2817 group. */
2818
2819 if (hc >= cd->hwm)
2820 {
2821 offset = GET(ptr, 1);
2822 if (cd->start_code + offset >= group) PUT(ptr, 1, offset + adjust);
2823 }
2824
2825 ptr += 1 + LINK_SIZE;
2826 }
2827 }
2828
2829
2830
2831 /*************************************************
2832 * Insert an automatic callout point *
2833 *************************************************/
2834
2835 /* This function is called when the PCRE_AUTO_CALLOUT option is set, to insert
2836 callout points before each pattern item.
2837
2838 Arguments:
2839 code current code pointer
2840 ptr current pattern pointer
2841 cd pointers to tables etc
2842
2843 Returns: new code pointer
2844 */
2845
2846 static pcre_uchar *
2847 auto_callout(pcre_uchar *code, const pcre_uchar *ptr, compile_data *cd)
2848 {
2849 *code++ = OP_CALLOUT;
2850 *code++ = 255;
2851 PUT(code, 0, (int)(ptr - cd->start_pattern)); /* Pattern offset */
2852 PUT(code, LINK_SIZE, 0); /* Default length */
2853 return code + 2 * LINK_SIZE;
2854 }
2855
2856
2857
2858 /*************************************************
2859 * Complete a callout item *
2860 *************************************************/
2861
2862 /* A callout item contains the length of the next item in the pattern, which
2863 we can't fill in till after we have reached the relevant point. This is used
2864 for both automatic and manual callouts.
2865
2866 Arguments:
2867 previous_callout points to previous callout item
2868 ptr current pattern pointer
2869 cd pointers to tables etc
2870
2871 Returns: nothing
2872 */
2873
2874 static void
2875 complete_callout(pcre_uchar *previous_callout, const pcre_uchar *ptr, compile_data *cd)
2876 {
2877 int length = (int)(ptr - cd->start_pattern - GET(previous_callout, 2));
2878 PUT(previous_callout, 2 + LINK_SIZE, length);
2879 }
2880
2881
2882
2883 #ifdef SUPPORT_UCP
2884 /*************************************************
2885 * Get othercase range *
2886 *************************************************/
2887
2888 /* This function is passed the start and end of a class range, in UTF-8 mode
2889 with UCP support. It searches up the characters, looking for ranges of
2890 characters in the "other" case. Each call returns the next one, updating the
2891 start address. A character with multiple other cases is returned on its own
2892 with a special return value.
2893
2894 Arguments:
2895 cptr points to starting character value; updated
2896 d end value
2897 ocptr where to put start of othercase range
2898 odptr where to put end of othercase range
2899
2900 Yield: -1 when no more
2901 0 when a range is returned
2902 >0 the CASESET offset for char with multiple other cases
2903 in this case, ocptr contains the original
2904 */
2905
2906 static int
2907 get_othercase_range(unsigned int *cptr, unsigned int d, unsigned int *ocptr,
2908 unsigned int *odptr)
2909 {
2910 unsigned int c, othercase, next;
2911 int co;
2912
2913 /* Find the first character that has an other case. If it has multiple other
2914 cases, return its case offset value. */
2915
2916 for (c = *cptr; c <= d; c++)
2917 {
2918 if ((co = UCD_CASESET(c)) != 0)
2919 {
2920 *ocptr = c++; /* Character that has the set */
2921 *cptr = c; /* Rest of input range */
2922 return co;
2923 }
2924 if ((othercase = UCD_OTHERCASE(c)) != c) break;
2925 }
2926
2927 if (c > d) return -1; /* Reached end of range */
2928
2929 *ocptr = othercase;
2930 next = othercase + 1;
2931
2932 for (++c; c <= d; c++)
2933 {
2934 if (UCD_OTHERCASE(c) != next) break;
2935 next++;
2936 }
2937
2938 *odptr = next - 1; /* End of othercase range */
2939 *cptr = c; /* Rest of input range */
2940 return 0;
2941 }
2942
2943
2944
2945 /*************************************************
2946 * Check a character and a property *
2947 *************************************************/
2948
2949 /* This function is called by check_auto_possessive() when a property item
2950 is adjacent to a fixed character.
2951
2952 Arguments:
2953 c the character
2954 ptype the property type
2955 pdata the data for the type
2956 negated TRUE if it's a negated property (\P or \p{^)
2957
2958 Returns: TRUE if auto-possessifying is OK
2959 */
2960
2961 static BOOL
2962 check_char_prop(int c, int ptype, int pdata, BOOL negated)
2963 {
2964 #ifdef SUPPORT_UCP
2965 const pcre_uint32 *p;
2966 #endif
2967
2968 const ucd_record *prop = GET_UCD(c);
2969
2970 switch(ptype)
2971 {
2972 case PT_LAMP:
2973 return (prop->chartype == ucp_Lu ||
2974 prop->chartype == ucp_Ll ||
2975 prop->chartype == ucp_Lt) == negated;
2976
2977 case PT_GC:
2978 return (pdata == PRIV(ucp_gentype)[prop->chartype]) == negated;
2979
2980 case PT_PC:
2981 return (pdata == prop->chartype) == negated;
2982
2983 case PT_SC:
2984 return (pdata == prop->script) == negated;
2985
2986 /* These are specials */
2987
2988 case PT_ALNUM:
2989 return (PRIV(ucp_gentype)[prop->chartype] == ucp_L ||
2990 PRIV(ucp_gentype)[prop->chartype] == ucp_N) == negated;
2991
2992 case PT_SPACE: /* Perl space */
2993 return (PRIV(ucp_gentype)[prop->chartype] == ucp_Z ||
2994 c == CHAR_HT || c == CHAR_NL || c == CHAR_FF || c == CHAR_CR)
2995 == negated;
2996
2997 case PT_PXSPACE: /* POSIX space */
2998 return (PRIV(ucp_gentype)[prop->chartype] == ucp_Z ||
2999 c == CHAR_HT || c == CHAR_NL || c == CHAR_VT ||
3000 c == CHAR_FF || c == CHAR_CR)
3001 == negated;
3002
3003 case PT_WORD:
3004 return (PRIV(ucp_gentype)[prop->chartype] == ucp_L ||
3005 PRIV(ucp_gentype)[prop->chartype] == ucp_N ||
3006 c == CHAR_UNDERSCORE) == negated;
3007
3008 #ifdef SUPPORT_UCP
3009 case PT_CLIST:
3010 p = PRIV(ucd_caseless_sets) + prop->caseset;
3011 for (;;)
3012 {
3013 if ((unsigned int)c < *p) return !negated;
3014 if ((unsigned int)c == *p++) return negated;
3015 }
3016 break; /* Control never reaches here */
3017 #endif
3018 }
3019
3020 return FALSE;
3021 }
3022 #endif /* SUPPORT_UCP */
3023
3024
3025
3026 /*************************************************
3027 * Check if auto-possessifying is possible *
3028 *************************************************/
3029
3030 /* This function is called for unlimited repeats of certain items, to see
3031 whether the next thing could possibly match the repeated item. If not, it makes
3032 sense to automatically possessify the repeated item.
3033
3034 Arguments:
3035 previous pointer to the repeated opcode
3036 utf TRUE in UTF-8 / UTF-16 / UTF-32 mode
3037 ptr next character in pattern
3038 options options bits
3039 cd contains pointers to tables etc.
3040
3041 Returns: TRUE if possessifying is wanted
3042 */
3043
3044 static BOOL
3045 check_auto_possessive(const pcre_uchar *previous, BOOL utf,
3046 const pcre_uchar *ptr, int options, compile_data *cd)
3047 {
3048 pcre_int32 c = NOTACHAR; // FIXMEchpe pcre_uint32
3049 pcre_int32 next;
3050 int escape;
3051 int op_code = *previous++;
3052
3053 /* Skip whitespace and comments in extended mode */
3054
3055 if ((options & PCRE_EXTENDED) != 0)
3056 {
3057 for (;;)
3058 {
3059 while (MAX_255(*ptr) && (cd->ctypes[*ptr] & ctype_space) != 0) ptr++;
3060 if (*ptr == CHAR_NUMBER_SIGN)
3061 {
3062 ptr++;
3063 while (*ptr != 0)
3064 {
3065 if (IS_NEWLINE(ptr)) { ptr += cd->nllen; break; }
3066 ptr++;
3067 #ifdef SUPPORT_UTF
3068 if (utf) FORWARDCHAR(ptr);
3069 #endif
3070 }
3071 }
3072 else break;
3073 }
3074 }
3075
3076 /* If the next item is one that we can handle, get its value. A non-negative
3077 value is a character, a negative value is an escape value. */
3078
3079 if (*ptr == CHAR_BACKSLASH)
3080 {
3081 int temperrorcode = 0;
3082 escape = check_escape(&ptr, &next, &temperrorcode, cd->bracount, options, FALSE);
3083 if (temperrorcode != 0) return FALSE;
3084 ptr++; /* Point after the escape sequence */
3085 }
3086 else if (!MAX_255(*ptr) || (cd->ctypes[*ptr] & ctype_meta) == 0)
3087 {
3088 escape = 0;
3089 #ifdef SUPPORT_UTF
3090 if (utf) { GETCHARINC(next, ptr); } else
3091 #endif
3092 next = *ptr++;
3093 }
3094 else return FALSE;
3095
3096 /* Skip whitespace and comments in extended mode */
3097
3098 if ((options & PCRE_EXTENDED) != 0)
3099 {
3100 for (;;)
3101 {
3102 while (MAX_255(*ptr) && (cd->ctypes[*ptr] & ctype_space) != 0) ptr++;
3103 if (*ptr == CHAR_NUMBER_SIGN)
3104 {
3105 ptr++;
3106 while (*ptr != 0)
3107 {
3108 if (IS_NEWLINE(ptr)) { ptr += cd->nllen; break; }
3109 ptr++;
3110 #ifdef SUPPORT_UTF
3111 if (utf) FORWARDCHAR(ptr);
3112 #endif
3113 }
3114 }
3115 else break;
3116 }
3117 }
3118
3119 /* If the next thing is itself optional, we have to give up. */
3120
3121 if (*ptr == CHAR_ASTERISK || *ptr == CHAR_QUESTION_MARK ||
3122 STRNCMP_UC_C8(ptr, STR_LEFT_CURLY_BRACKET STR_0 STR_COMMA, 3) == 0)
3123 return FALSE;
3124
3125 /* If the previous item is a character, get its value. */
3126
3127 if (op_code == OP_CHAR || op_code == OP_CHARI ||
3128 op_code == OP_NOT || op_code == OP_NOTI)
3129 //if (escape == 0) switch(op_code)
3130 {
3131 #ifdef SUPPORT_UTF
3132 GETCHARTEST(c, previous);
3133 #else
3134 c = *previous;
3135 #endif
3136 }
3137
3138 /* Now compare the next item with the previous opcode. First, handle cases when
3139 the next item is a character. */
3140
3141 if (escape == 0)
3142 {
3143 /* For a caseless UTF match, the next character may have more than one other
3144 case, which maps to the special PT_CLIST property. Check this first. */
3145
3146 #ifdef SUPPORT_UCP
3147 if (utf && (unsigned int)c != NOTACHAR && (options & PCRE_CASELESS) != 0)
3148 {
3149 int ocs = UCD_CASESET(next);
3150 if (ocs > 0) return check_char_prop(c, PT_CLIST, ocs, op_code >= OP_NOT);
3151 }
3152 #endif
3153
3154 switch(op_code)
3155 {
3156 case OP_CHAR:
3157 return c != next;
3158
3159 /* For CHARI (caseless character) we must check the other case. If we have
3160 Unicode property support, we can use it to test the other case of
3161 high-valued characters. We know that next can have only one other case,
3162 because multi-other-case characters are dealt with above. */
3163
3164 case OP_CHARI:
3165 if (c == next) return FALSE;
3166 #ifdef SUPPORT_UTF
3167 if (utf)
3168 {
3169 unsigned int othercase;
3170 if (next < 128) othercase = cd->fcc[next]; else
3171 #ifdef SUPPORT_UCP
3172 othercase = UCD_OTHERCASE((unsigned int)next);
3173 #else
3174 othercase = NOTACHAR;
3175 #endif
3176 return (unsigned int)c != othercase;
3177 }
3178 else
3179 #endif /* SUPPORT_UTF */
3180 return (c != TABLE_GET((unsigned int)next, cd->fcc, next)); /* Not UTF */
3181
3182 case OP_NOT:
3183 return c == next;
3184
3185 case OP_NOTI:
3186 if (c == next) return TRUE;
3187 #ifdef SUPPORT_UTF
3188 if (utf)
3189 {
3190 unsigned int othercase;
3191 if (next < 128) othercase = cd->fcc[next]; else
3192 #ifdef SUPPORT_UCP
3193 othercase = UCD_OTHERCASE((unsigned int)next);
3194 #else
3195 othercase = NOTACHAR;
3196 #endif
3197 return (unsigned int)c == othercase;
3198 }
3199 else
3200 #endif /* SUPPORT_UTF */
3201 return (c == TABLE_GET((unsigned int)next, cd->fcc, next)); /* Not UTF */
3202
3203 /* Note that OP_DIGIT etc. are generated only when PCRE_UCP is *not* set.
3204 When it is set, \d etc. are converted into OP_(NOT_)PROP codes. */
3205
3206 case OP_DIGIT:
3207 return next > 255 || (cd->ctypes[next] & ctype_digit) == 0;
3208
3209 case OP_NOT_DIGIT:
3210 return next <= 255 && (cd->ctypes[next] & ctype_digit) != 0;
3211
3212 case OP_WHITESPACE:
3213 return next > 255 || (cd->ctypes[next] & ctype_space) == 0;
3214
3215 case OP_NOT_WHITESPACE:
3216 return next <= 255 && (cd->ctypes[next] & ctype_space) != 0;
3217
3218 case OP_WORDCHAR:
3219 return next > 255 || (cd->ctypes[next] & ctype_word) == 0;
3220
3221 case OP_NOT_WORDCHAR:
3222 return next <= 255 && (cd->ctypes[next] & ctype_word) != 0;
3223
3224 case OP_HSPACE:
3225 case OP_NOT_HSPACE:
3226 switch(next)
3227 {
3228 HSPACE_CASES:
3229 return op_code == OP_NOT_HSPACE;
3230
3231 default:
3232 return op_code != OP_NOT_HSPACE;
3233 }
3234
3235 case OP_ANYNL:
3236 case OP_VSPACE:
3237 case OP_NOT_VSPACE:
3238 switch(next)
3239 {
3240 VSPACE_CASES:
3241 return op_code == OP_NOT_VSPACE;
3242
3243 default:
3244 return op_code != OP_NOT_VSPACE;
3245 }
3246
3247 #ifdef SUPPORT_UCP
3248 case OP_PROP:
3249 return check_char_prop(next, previous[0], previous[1], FALSE);
3250
3251 case OP_NOTPROP:
3252 return check_char_prop(next, previous[0], previous[1], TRUE);
3253 #endif
3254
3255 default:
3256 return FALSE;
3257 }
3258 }
3259
3260 /* Handle the case when the next item is \d, \s, etc. Note that when PCRE_UCP
3261 is set, \d turns into ESC_du rather than ESC_d, etc., so ESC_d etc. are
3262 generated only when PCRE_UCP is *not* set, that is, when only ASCII
3263 characteristics are recognized. Similarly, the opcodes OP_DIGIT etc. are
3264 replaced by OP_PROP codes when PCRE_UCP is set. */
3265
3266 switch(op_code)
3267 {
3268 case OP_CHAR:
3269 case OP_CHARI:
3270 switch(escape)
3271 {
3272 case ESC_d:
3273 return c > 255 || (cd->ctypes[c] & ctype_digit) == 0;
3274
3275 case ESC_D:
3276 return c <= 255 && (cd->ctypes[c] & ctype_digit) != 0;
3277
3278 case ESC_s:
3279 return c > 255 || (cd->ctypes[c] & ctype_space) == 0;
3280
3281 case ESC_S:
3282 return c <= 255 && (cd->ctypes[c] & ctype_space) != 0;
3283
3284 case ESC_w:
3285 return c > 255 || (cd->ctypes[c] & ctype_word) == 0;
3286
3287 case ESC_W:
3288 return c <= 255 && (cd->ctypes[c] & ctype_word) != 0;
3289
3290 case ESC_h:
3291 case ESC_H:
3292 switch(c)
3293 {
3294 HSPACE_CASES:
3295 return escape != ESC_h;
3296
3297 default:
3298 return escape == ESC_h;
3299 }
3300
3301 case ESC_v:
3302 case ESC_V:
3303 switch(c)
3304 {
3305 VSPACE_CASES:
3306 return escape != ESC_v;
3307
3308 default:
3309 return escape == ESC_v;
3310 }
3311
3312 /* When PCRE_UCP is set, these values get generated for \d etc. Find
3313 their substitutions and process them. The result will always be either
3314 ESC_p or ESC_P. Then fall through to process those values. */
3315
3316 #ifdef SUPPORT_UCP
3317 case ESC_du:
3318 case ESC_DU:
3319 case ESC_wu:
3320 case ESC_WU:
3321 case ESC_su:
3322 case ESC_SU:
3323 {
3324 int temperrorcode = 0;
3325 ptr = substitutes[escape - ESC_DU];
3326 escape = check_escape(&ptr, &next, &temperrorcode, 0, options, FALSE);
3327 if (temperrorcode != 0) return FALSE;
3328 ptr++; /* For compatibility */
3329 }
3330 /* Fall through */
3331
3332 case ESC_p:
3333 case ESC_P:
3334 {
3335 int ptype, pdata, errorcodeptr;
3336 BOOL negated;
3337
3338 ptr--; /* Make ptr point at the p or P */
3339 ptype = get_ucp(&ptr, &negated, &pdata, &errorcodeptr);
3340 if (ptype < 0) return FALSE;
3341 ptr++; /* Point past the final curly ket */
3342
3343 /* If the property item is optional, we have to give up. (When generated
3344 from \d etc by PCRE_UCP, this test will have been applied much earlier,
3345 to the original \d etc. At this point, ptr will point to a zero byte. */
3346
3347 if (*ptr == CHAR_ASTERISK || *ptr == CHAR_QUESTION_MARK ||
3348 STRNCMP_UC_C8(ptr, STR_LEFT_CURLY_BRACKET STR_0 STR_COMMA, 3) == 0)
3349 return FALSE;
3350
3351 /* Do the property check. */
3352
3353 return check_char_prop(c, ptype, pdata, (escape == ESC_P) != negated);
3354 }
3355 #endif
3356
3357 default:
3358 return FALSE;
3359 }
3360
3361 /* In principle, support for Unicode properties should be integrated here as
3362 well. It means re-organizing the above code so as to get hold of the property
3363 values before switching on the op-code. However, I wonder how many patterns
3364 combine ASCII \d etc with Unicode properties? (Note that if PCRE_UCP is set,
3365 these op-codes are never generated.) */
3366
3367 case OP_DIGIT:
3368 return escape == ESC_D || escape == ESC_s || escape == ESC_W ||
3369 escape == ESC_h || escape == ESC_v || escape == ESC_R;
3370
3371 case OP_NOT_DIGIT:
3372 return escape == ESC_d;
3373
3374 case OP_WHITESPACE:
3375 return escape == ESC_S || escape == ESC_d || escape == ESC_w;
3376
3377 case OP_NOT_WHITESPACE:
3378 return escape == ESC_s || escape == ESC_h || escape == ESC_v || escape == ESC_R;
3379
3380 case OP_HSPACE:
3381 return escape == ESC_S || escape == ESC_H || escape == ESC_d ||
3382 escape == ESC_w || escape == ESC_v || escape == ESC_R;
3383
3384 case OP_NOT_HSPACE:
3385 return escape == ESC_h;
3386
3387 /* Can't have \S in here because VT matches \S (Perl anomaly) */
3388 case OP_ANYNL:
3389 case OP_VSPACE:
3390 return escape == ESC_V || escape == ESC_d || escape == ESC_w;
3391
3392 case OP_NOT_VSPACE:
3393 return escape == ESC_v || escape == ESC_R;
3394
3395 case OP_WORDCHAR:
3396 return escape == ESC_W || escape == ESC_s || escape == ESC_h ||
3397 escape == ESC_v || escape == ESC_R;
3398
3399 case OP_NOT_WORDCHAR:
3400 return escape == ESC_w || escape == ESC_d;
3401
3402 default:
3403 return FALSE;
3404 }
3405
3406 /* Control does not reach here */
3407 }
3408
3409
3410
3411 /*************************************************
3412 * Add a character or range to a class *
3413 *************************************************/
3414
3415 /* This function packages up the logic of adding a character or range of
3416 characters to a class. The character values in the arguments will be within the
3417 valid values for the current mode (8-bit, 16-bit, UTF, etc). This function is
3418 mutually recursive with the function immediately below.
3419
3420 Arguments:
3421 classbits the bit map for characters < 256
3422 uchardptr points to the pointer for extra data
3423 options the options word
3424 cd contains pointers to tables etc.
3425 start start of range character
3426 end end of range character
3427
3428 Returns: the number of < 256 characters added
3429 the pointer to extra data is updated
3430 */
3431
3432 static int
3433 add_to_class(pcre_uint8 *classbits, pcre_uchar **uchardptr, int options,
3434 compile_data *cd, pcre_uint32 start, pcre_uint32 end)
3435 {
3436 pcre_uint32 c;
3437 int n8 = 0;
3438
3439 /* If caseless matching is required, scan the range and process alternate
3440 cases. In Unicode, there are 8-bit characters that have alternate cases that
3441 are greater than 255 and vice-versa. Sometimes we can just extend the original
3442 range. */
3443
3444 if ((options & PCRE_CASELESS) != 0)
3445 {
3446 #ifdef SUPPORT_UCP
3447 if ((options & PCRE_UTF8) != 0)
3448 {
3449 int rc;
3450 pcre_uint32 oc, od;
3451
3452 options &= ~PCRE_CASELESS; /* Remove for recursive calls */
3453 c = start;
3454
3455 while ((rc = get_othercase_range(&c, end, &oc, &od)) >= 0)
3456 {
3457 /* Handle a single character that has more than one other case. */
3458
3459 if (rc > 0) n8 += add_list_to_class(classbits, uchardptr, options, cd,
3460 PRIV(ucd_caseless_sets) + rc, oc);
3461
3462 /* Do nothing if the other case range is within the original range. */
3463
3464 else if (oc >= start && od <= end) continue;
3465
3466 /* Extend the original range if there is overlap, noting that if oc < c, we
3467 can't have od > end because a subrange is always shorter than the basic
3468 range. Otherwise, use a recursive call to add the additional range. */
3469
3470 else if (oc < start && od >= start - 1) start = oc; /* Extend downwards */
3471 else if (od > end && oc <= end + 1) end = od; /* Extend upwards */
3472 else n8 += add_to_class(classbits, uchardptr, options, cd, oc, od);
3473 }
3474 }
3475 else
3476 #endif /* SUPPORT_UCP */
3477
3478 /* Not UTF-mode, or no UCP */
3479
3480 for (c = start; c <= end && c < 256; c++)
3481 {
3482 SETBIT(classbits, cd->fcc[c]);
3483 n8++;
3484 }
3485 }
3486
3487 /* Now handle the original range. Adjust the final value according to the bit
3488 length - this means that the same lists of (e.g.) horizontal spaces can be used
3489 in all cases. */
3490
3491 #if defined COMPILE_PCRE8
3492 #ifdef SUPPORT_UTF
3493 if ((options & PCRE_UTF8) == 0)
3494 #endif
3495 if (end > 0xff) end = 0xff;
3496
3497 #elif defined COMPILE_PCRE16
3498 #ifdef SUPPORT_UTF
3499 if ((options & PCRE_UTF16) == 0)
3500 #endif
3501 if (end > 0xffff) end = 0xffff;
3502
3503 #endif /* COMPILE_PCRE[8|16] */
3504
3505 /* If all characters are less than 256, use the bit map. Otherwise use extra
3506 data. */
3507
3508 if (end < 0x100)
3509 {
3510 for (c = start; c <= end; c++)
3511 {
3512 n8++;
3513 SETBIT(classbits, c);
3514 }
3515 }
3516
3517 else
3518 {
3519 pcre_uchar *uchardata = *uchardptr;
3520
3521 #ifdef SUPPORT_UTF
3522 if ((options & PCRE_UTF8) != 0) /* All UTFs use the same flag bit */
3523 {
3524 if (start < end)
3525 {
3526 *uchardata++ = XCL_RANGE;
3527 uchardata += PRIV(ord2utf)(start, uchardata);
3528 uchardata += PRIV(ord2utf)(end, uchardata);
3529 }
3530 else if (start == end)
3531 {
3532 *uchardata++ = XCL_SINGLE;
3533 uchardata += PRIV(ord2utf)(start, uchardata);
3534 }
3535 }
3536 else
3537 #endif /* SUPPORT_UTF */
3538
3539 /* Without UTF support, character values are constrained by the bit length,
3540 and can only be > 256 for 16-bit and 32-bit libraries. */
3541
3542 #ifdef COMPILE_PCRE8
3543 {}
3544 #else
3545 if (start < end)
3546 {
3547 *uchardata++ = XCL_RANGE;
3548 *uchardata++ = start;
3549 *uchardata++ = end;
3550 }
3551 else if (start == end)
3552 {
3553 *uchardata++ = XCL_SINGLE;
3554 *uchardata++ = start;
3555 }
3556 #endif
3557
3558 *uchardptr = uchardata; /* Updata extra data pointer */
3559 }
3560
3561 return n8; /* Number of 8-bit characters */
3562 }
3563
3564
3565
3566
3567 /*************************************************
3568 * Add a list of characters to a class *
3569 *************************************************/
3570
3571 /* This function is used for adding a list of case-equivalent characters to a
3572 class, and also for adding a list of horizontal or vertical whitespace. If the
3573 list is in order (which it should be), ranges of characters are detected and
3574 handled appropriately. This function is mutually recursive with the function
3575 above.
3576
3577 Arguments:
3578 classbits the bit map for characters < 256
3579 uchardptr points to the pointer for extra data
3580 options the options word
3581 cd contains pointers to tables etc.
3582 p points to row of 32-bit values, terminated by NOTACHAR
3583 except character to omit; this is used when adding lists of
3584 case-equivalent characters to avoid including the one we
3585 already know about
3586
3587 Returns: the number of < 256 characters added
3588 the pointer to extra data is updated
3589 */
3590
3591 static int
3592 add_list_to_class(pcre_uint8 *classbits, pcre_uchar **uchardptr, int options,
3593 compile_data *cd, const pcre_uint32 *p, unsigned int except)
3594 {
3595 int n8 = 0;
3596 while (p[0] < NOTACHAR)
3597 {
3598 int n = 0;
3599 if (p[0] != except)
3600 {
3601 while(p[n+1] == p[0] + n + 1) n++;
3602 n8 += add_to_class(classbits, uchardptr, options, cd, p[0], p[n]);
3603 }
3604 p += n + 1;
3605 }
3606 return n8;
3607 }
3608
3609
3610
3611 /*************************************************
3612 * Add characters not in a list to a class *
3613 *************************************************/
3614
3615 /* This function is used for adding the complement of a list of horizontal or
3616 vertical whitespace to a class. The list must be in order.
3617
3618 Arguments:
3619 classbits the bit map for characters < 256
3620 uchardptr points to the pointer for extra data
3621 options the options word
3622 cd contains pointers to tables etc.
3623 p points to row of 32-bit values, terminated by NOTACHAR
3624
3625 Returns: the number of < 256 characters added
3626 the pointer to extra data is updated
3627 */
3628
3629 static int
3630 add_not_list_to_class(pcre_uint8 *classbits, pcre_uchar **uchardptr,
3631 int options, compile_data *cd, const pcre_uint32 *p)
3632 {
3633 BOOL utf = (options & PCRE_UTF8) != 0;
3634 int n8 = 0;
3635 if (p[0] > 0)
3636 n8 += add_to_class(classbits, uchardptr, options, cd, 0, p[0] - 1);
3637 while (p[0] < NOTACHAR)
3638 {
3639 while (p[1] == p[0] + 1) p++;
3640 n8 += add_to_class(classbits, uchardptr, options, cd, p[0] + 1,
3641 (p[1] == NOTACHAR) ? (utf ? 0x10ffffu : 0xffffffffu) : p[1] - 1);
3642 p++;
3643 }
3644 return n8;
3645 }
3646
3647
3648
3649 /*************************************************
3650 * Compile one branch *
3651 *************************************************/
3652
3653 /* Scan the pattern, compiling it into the a vector. If the options are
3654 changed during the branch, the pointer is used to change the external options
3655 bits. This function is used during the pre-compile phase when we are trying
3656 to find out the amount of memory needed, as well as during the real compile
3657 phase. The value of lengthptr distinguishes the two phases.
3658
3659 Arguments:
3660 optionsptr pointer to the option bits
3661 codeptr points to the pointer to the current code point
3662 ptrptr points to the current pattern pointer
3663 errorcodeptr points to error code variable
3664 firstcharptr set to initial literal character, or < 0 (REQ_UNSET, REQ_NONE)
3665 reqcharptr set to the last literal character required, else < 0
3666 bcptr points to current branch chain
3667 cond_depth conditional nesting depth
3668 cd contains pointers to tables etc.
3669 lengthptr NULL during the real compile phase
3670 points to length accumulator during pre-compile phase
3671
3672 Returns: TRUE on success
3673 FALSE, with *errorcodeptr set non-zero on error
3674 */
3675
3676 static BOOL
3677 compile_branch(int *optionsptr, pcre_uchar **codeptr,
3678 const pcre_uchar **ptrptr, int *errorcodeptr, pcre_int32 *firstcharptr,
3679 pcre_int32 *reqcharptr, branch_chain *bcptr, int cond_depth,
3680 compile_data *cd, int *lengthptr)
3681 {
3682 int repeat_type, op_type;
3683 int repeat_min = 0, repeat_max = 0; /* To please picky compilers */
3684 int bravalue = 0;
3685 int greedy_default, greedy_non_default;
3686 pcre_int32 firstchar, reqchar;
3687 pcre_int32 zeroreqchar, zerofirstchar;
3688 pcre_int32 req_caseopt, reqvary, tempreqvary;
3689 int options = *optionsptr; /* May change dynamically */
3690 int after_manual_callout = 0;
3691 int length_prevgroup = 0;
3692 register int c;
3693 int escape;
3694 register pcre_uchar *code = *codeptr;
3695 pcre_uchar *last_code = code;
3696 pcre_uchar *orig_code = code;
3697 pcre_uchar *tempcode;
3698 BOOL inescq = FALSE;
3699 BOOL groupsetfirstchar = FALSE;
3700 const pcre_uchar *ptr = *ptrptr;
3701 const pcre_uchar *tempptr;
3702 const pcre_uchar *nestptr = NULL;
3703 pcre_uchar *previous = NULL;
3704 pcre_uchar *previous_callout = NULL;
3705 pcre_uchar *save_hwm = NULL;
3706 pcre_uint8 classbits[32];
3707
3708 /* We can fish out the UTF-8 setting once and for all into a BOOL, but we
3709 must not do this for other options (e.g. PCRE_EXTENDED) because they may change
3710 dynamically as we process the pattern. */
3711
3712 #ifdef SUPPORT_UTF
3713 /* PCRE_UTF[16|32] have the same value as PCRE_UTF8. */
3714 BOOL utf = (options & PCRE_UTF8) != 0;
3715 pcre_uchar utf_chars[6];
3716 #else
3717 BOOL utf = FALSE;
3718 #endif
3719
3720 /* Helper variables for OP_XCLASS opcode (for characters > 255). We define
3721 class_uchardata always so that it can be passed to add_to_class() always,
3722 though it will not be used in non-UTF 8-bit cases. This avoids having to supply
3723 alternative calls for the different cases. */
3724
3725 pcre_uchar *class_uchardata;
3726 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
3727 BOOL xclass;
3728 pcre_uchar *class_uchardata_base;
3729 #endif
3730
3731 #ifdef PCRE_DEBUG
3732 if (lengthptr != NULL) DPRINTF((">> start branch\n"));
3733 #endif
3734
3735 /* Set up the default and non-default settings for greediness */
3736
3737 greedy_default = ((options & PCRE_UNGREEDY) != 0);
3738 greedy_non_default = greedy_default ^ 1;
3739
3740 /* Initialize no first byte, no required byte. REQ_UNSET means "no char
3741 matching encountered yet". It gets changed to REQ_NONE if we hit something that
3742 matches a non-fixed char first char; reqchar just remains unset if we never
3743 find one.
3744
3745 When we hit a repeat whose minimum is zero, we may have to adjust these values
3746 to take the zero repeat into account. This is implemented by setting them to
3747 zerofirstbyte and zeroreqchar when such a repeat is encountered. The individual
3748 item types that can be repeated set these backoff variables appropriately. */
3749
3750 firstchar = reqchar = zerofirstchar = zeroreqchar = REQ_UNSET;
3751
3752 /* The variable req_caseopt contains either the REQ_CASELESS value
3753 or zero, according to the current setting of the caseless flag. The
3754 REQ_CASELESS leaves the lower 28 bit empty. It is added into the
3755 firstchar or reqchar variables to record the case status of the
3756 value. This is used only for ASCII characters. */
3757
3758 req_caseopt = ((options & PCRE_CASELESS) != 0)? REQ_CASELESS:0;
3759
3760 /* Switch on next character until the end of the branch */
3761
3762 for (;; ptr++)
3763 {
3764 BOOL negate_class;
3765 BOOL should_flip_negation;
3766 BOOL possessive_quantifier;
3767 BOOL is_quantifier;
3768 BOOL is_recurse;
3769 BOOL reset_bracount;
3770 int class_has_8bitchar;
3771 int class_one_char;
3772 int newoptions;
3773 int recno;
3774 int refsign;
3775 int skipbytes;
3776 int subreqchar;
3777 int subfirstchar;
3778 int terminator;
3779 int mclength;
3780 int tempbracount;
3781 int ec; // FIXMEchpe pcre_uint32
3782 pcre_uchar mcbuffer[8];
3783
3784 /* Get next character in the pattern */
3785
3786 c = *ptr;
3787
3788 /* If we are at the end of a nested substitution, revert to the outer level
3789 string. Nesting only happens one level deep. */
3790
3791 if (c == 0 && nestptr != NULL)
3792 {
3793 ptr = nestptr;
3794 nestptr = NULL;
3795 c = *ptr;
3796 }
3797
3798 /* If we are in the pre-compile phase, accumulate the length used for the
3799 previous cycle of this loop. */
3800
3801 if (lengthptr != NULL)
3802 {
3803 #ifdef PCRE_DEBUG
3804 if (code > cd->hwm) cd->hwm = code; /* High water info */
3805 #endif
3806 if (code > cd->start_workspace + cd->workspace_size -
3807 WORK_SIZE_SAFETY_MARGIN) /* Check for overrun */
3808 {
3809 *errorcodeptr = ERR52;
3810 goto FAILED;
3811 }
3812
3813 /* There is at least one situation where code goes backwards: this is the
3814 case of a zero quantifier after a class (e.g. [ab]{0}). At compile time,
3815 the class is simply eliminated. However, it is created first, so we have to
3816 allow memory for it. Therefore, don't ever reduce the length at this point.
3817 */
3818
3819 if (code < last_code) code = last_code;
3820
3821 /* Paranoid check for integer overflow */
3822
3823 if (OFLOW_MAX - *lengthptr < code - last_code)
3824 {
3825 *errorcodeptr = ERR20;
3826 goto FAILED;
3827 }
3828
3829 *lengthptr += (int)(code - last_code);
3830 DPRINTF(("length=%d added %d c=%c (0x%x)\n", *lengthptr,
3831 (int)(code - last_code), c, c));
3832
3833 /* If "previous" is set and it is not at the start of the work space, move
3834 it back to there, in order to avoid filling up the work space. Otherwise,
3835 if "previous" is NULL, reset the current code pointer to the start. */
3836
3837 if (previous != NULL)
3838 {
3839 if (previous > orig_code)
3840 {
3841 memmove(orig_code, previous, IN_UCHARS(code - previous));
3842 code -= previous - orig_code;
3843 previous = orig_code;
3844 }
3845 }
3846 else code = orig_code;
3847
3848 /* Remember where this code item starts so we can pick up the length
3849 next time round. */
3850
3851 last_code = code;
3852 }
3853
3854 /* In the real compile phase, just check the workspace used by the forward
3855 reference list. */
3856
3857 else if (cd->hwm > cd->start_workspace + cd->workspace_size -
3858 WORK_SIZE_SAFETY_MARGIN)
3859 {
3860 *errorcodeptr = ERR52;
3861 goto FAILED;
3862 }
3863
3864 /* If in \Q...\E, check for the end; if not, we have a literal */
3865
3866 if (inescq && c != 0)
3867 {
3868 if (c == CHAR_BACKSLASH && ptr[1] == CHAR_E)
3869 {
3870 inescq = FALSE;
3871 ptr++;
3872 continue;
3873 }
3874 else
3875 {
3876 if (previous_callout != NULL)
3877 {
3878 if (lengthptr == NULL) /* Don't attempt in pre-compile phase */
3879 complete_callout(previous_callout, ptr, cd);
3880 previous_callout = NULL;
3881 }
3882 if ((options & PCRE_AUTO_CALLOUT) != 0)
3883 {
3884 previous_callout = code;
3885 code = auto_callout(code, ptr, cd);
3886 }
3887 goto NORMAL_CHAR;
3888 }
3889 }
3890
3891 /* Fill in length of a previous callout, except when the next thing is
3892 a quantifier. */
3893
3894 is_quantifier =
3895 c == CHAR_ASTERISK || c == CHAR_PLUS || c == CHAR_QUESTION_MARK ||
3896 (c == CHAR_LEFT_CURLY_BRACKET && is_counted_repeat(ptr+1));
3897
3898 if (!is_quantifier && previous_callout != NULL &&
3899 after_manual_callout-- <= 0)
3900 {
3901 if (lengthptr == NULL) /* Don't attempt in pre-compile phase */
3902 complete_callout(previous_callout, ptr, cd);
3903 previous_callout = NULL;
3904 }
3905
3906 /* In extended mode, skip white space and comments. */
3907
3908 if ((options & PCRE_EXTENDED) != 0)
3909 {
3910 if (MAX_255(*ptr) && (cd->ctypes[c] & ctype_space) != 0) continue;
3911 if (c == CHAR_NUMBER_SIGN)
3912 {
3913 ptr++;
3914 while (*ptr != 0)
3915 {
3916 if (IS_NEWLINE(ptr)) { ptr += cd->nllen - 1; break; }
3917 ptr++;
3918 #ifdef SUPPORT_UTF
3919 if (utf) FORWARDCHAR(ptr);
3920 #endif
3921 }
3922 if (*ptr != 0) continue;
3923
3924 /* Else fall through to handle end of string */
3925 c = 0;
3926 }
3927 }
3928
3929 /* No auto callout for quantifiers. */
3930
3931 if ((options & PCRE_AUTO_CALLOUT) != 0 && !is_quantifier)
3932 {
3933 previous_callout = code;
3934 code = auto_callout(code, ptr, cd);
3935 }
3936
3937 switch(c)
3938 {
3939 /* ===================================================================*/
3940 case 0: /* The branch terminates at string end */
3941 case CHAR_VERTICAL_LINE: /* or | or ) */
3942 case CHAR_RIGHT_PARENTHESIS:
3943 *firstcharptr = firstchar;
3944 *reqcharptr = reqchar;
3945 *codeptr = code;
3946 *ptrptr = ptr;
3947 if (lengthptr != NULL)
3948 {
3949 if (OFLOW_MAX - *lengthptr < code - last_code)
3950 {
3951 *errorcodeptr = ERR20;
3952 goto FAILED;
3953 }
3954 *lengthptr += (int)(code - last_code); /* To include callout length */
3955 DPRINTF((">> end branch\n"));
3956 }
3957 return TRUE;
3958
3959
3960 /* ===================================================================*/
3961 /* Handle single-character metacharacters. In multiline mode, ^ disables
3962 the setting of any following char as a first character. */
3963
3964 case CHAR_CIRCUMFLEX_ACCENT:
3965 previous = NULL;
3966 if ((options & PCRE_MULTILINE) != 0)
3967 {
3968 if (firstchar == REQ_UNSET) firstchar = REQ_NONE;
3969 *code++ = OP_CIRCM;
3970 }
3971 else *code++ = OP_CIRC;
3972 break;
3973
3974 case CHAR_DOLLAR_SIGN:
3975 previous = NULL;
3976 *code++ = ((options & PCRE_MULTILINE) != 0)? OP_DOLLM : OP_DOLL;
3977 break;
3978
3979 /* There can never be a first char if '.' is first, whatever happens about
3980 repeats. The value of reqchar doesn't change either. */
3981
3982 case CHAR_DOT:
3983 if (firstchar == REQ_UNSET) firstchar = REQ_NONE;
3984 zerofirstchar = firstchar;
3985 zeroreqchar = reqchar;
3986 previous = code;
3987 *code++ = ((options & PCRE_DOTALL) != 0)? OP_ALLANY: OP_ANY;
3988 break;
3989
3990
3991 /* ===================================================================*/
3992 /* Character classes. If the included characters are all < 256, we build a
3993 32-byte bitmap of the permitted characters, except in the special case
3994 where there is only one such character. For negated classes, we build the
3995 map as usual, then invert it at the end. However, we use a different opcode
3996 so that data characters > 255 can be handled correctly.
3997
3998 If the class contains characters outside the 0-255 range, a different
3999 opcode is compiled. It may optionally have a bit map for characters < 256,
4000 but those above are are explicitly listed afterwards. A flag byte tells
4001 whether the bitmap is present, and whether this is a negated class or not.
4002
4003 In JavaScript compatibility mode, an isolated ']' causes an error. In
4004 default (Perl) mode, it is treated as a data character. */
4005
4006 case CHAR_RIGHT_SQUARE_BRACKET:
4007 if ((cd->external_options & PCRE_JAVASCRIPT_COMPAT) != 0)
4008 {
4009 *errorcodeptr = ERR64;
4010 goto FAILED;
4011 }
4012 goto NORMAL_CHAR;
4013
4014 case CHAR_LEFT_SQUARE_BRACKET:
4015 previous = code;
4016
4017 /* PCRE supports POSIX class stuff inside a class. Perl gives an error if
4018 they are encountered at the top level, so we'll do that too. */
4019
4020 if ((ptr[1] == CHAR_COLON || ptr[1] == CHAR_DOT ||
4021 ptr[1] == CHAR_EQUALS_SIGN) &&
4022 check_posix_syntax(ptr, &tempptr))
4023 {
4024 *errorcodeptr = (ptr[1] == CHAR_COLON)? ERR13 : ERR31;
4025 goto FAILED;
4026 }
4027
4028 /* If the first character is '^', set the negation flag and skip it. Also,
4029 if the first few characters (either before or after ^) are \Q\E or \E we
4030 skip them too. This makes for compatibility with Perl. */
4031
4032 negate_class = FALSE;
4033 for (;;)
4034 {
4035 c = *(++ptr);
4036 if (c == CHAR_BACKSLASH)
4037 {
4038 if (ptr[1] == CHAR_E)
4039 ptr++;
4040 else if (STRNCMP_UC_C8(ptr + 1, STR_Q STR_BACKSLASH STR_E, 3) == 0)
4041 ptr += 3;
4042 else
4043 break;
4044 }
4045 else if (!negate_class && c == CHAR_CIRCUMFLEX_ACCENT)
4046 negate_class = TRUE;
4047 else break;
4048 }
4049
4050 /* Empty classes are allowed in JavaScript compatibility mode. Otherwise,
4051 an initial ']' is taken as a data character -- the code below handles
4052 that. In JS mode, [] must always fail, so generate OP_FAIL, whereas
4053 [^] must match any character, so generate OP_ALLANY. */
4054
4055 if (c == CHAR_RIGHT_SQUARE_BRACKET &&
4056 (cd->external_options & PCRE_JAVASCRIPT_COMPAT) != 0)
4057 {
4058 *code++ = negate_class? OP_ALLANY : OP_FAIL;
4059 if (firstchar == REQ_UNSET) firstchar = REQ_NONE;
4060 zerofirstchar = firstchar;
4061 break;
4062 }
4063
4064 /* If a class contains a negative special such as \S, we need to flip the
4065 negation flag at the end, so that support for characters > 255 works
4066 correctly (they are all included in the class). */
4067
4068 should_flip_negation = FALSE;
4069
4070 /* For optimization purposes, we track some properties of the class:
4071 class_has_8bitchar will be non-zero if the class contains at least one <
4072 256 character; class_one_char will be 1 if the class contains just one
4073 character. */
4074
4075 class_has_8bitchar = 0;
4076 class_one_char = 0;
4077
4078 /* Initialize the 32-char bit map to all zeros. We build the map in a
4079 temporary bit of memory, in case the class contains fewer than two
4080 8-bit characters because in that case the compiled code doesn't use the bit
4081 map. */
4082
4083 memset(classbits, 0, 32 * sizeof(pcre_uint8));
4084
4085 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
4086 xclass = FALSE;
4087 class_uchardata = code + LINK_SIZE + 2; /* For XCLASS items */
4088 class_uchardata_base = class_uchardata; /* Save the start */
4089 #endif
4090
4091 /* Process characters until ] is reached. By writing this as a "do" it
4092 means that an initial ] is taken as a data character. At the start of the
4093 loop, c contains the first byte of the character. */
4094
4095 if (c != 0) do
4096 {
4097 const pcre_uchar *oldptr;
4098
4099 #if defined SUPPORT_UTF && !defined COMPILE_PCRE32
4100 if (utf && HAS_EXTRALEN(c))
4101 { /* Braces are required because the */
4102 GETCHARLEN(c, ptr, ptr); /* macro generates multiple statements */
4103 }
4104 #endif
4105
4106 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
4107 /* In the pre-compile phase, accumulate the length of any extra
4108 data and reset the pointer. This is so that very large classes that
4109 contain a zillion > 255 characters no longer overwrite the work space
4110 (which is on the stack). We have to remember that there was XCLASS data,
4111 however. */
4112
4113 if (lengthptr != NULL && class_uchardata > class_uchardata_base)
4114 {
4115 xclass = TRUE;
4116 *lengthptr += class_uchardata - class_uchardata_base;
4117 class_uchardata = class_uchardata_base;
4118 }
4119 #endif
4120
4121 /* Inside \Q...\E everything is literal except \E */
4122
4123 if (inescq)
4124 {
4125 if (c == CHAR_BACKSLASH && ptr[1] == CHAR_E) /* If we are at \E */
4126 {
4127 inescq = FALSE; /* Reset literal state */
4128 ptr++; /* Skip the 'E' */
4129 continue; /* Carry on with next */
4130 }
4131 goto CHECK_RANGE; /* Could be range if \E follows */
4132 }
4133
4134 /* Handle POSIX class names. Perl allows a negation extension of the
4135 form [:^name:]. A square bracket that doesn't match the syntax is
4136 treated as a literal. We also recognize the POSIX constructions
4137 [.ch.] and [=ch=] ("collating elements") and fault them, as Perl
4138 5.6 and 5.8 do. */
4139
4140 if (c == CHAR_LEFT_SQUARE_BRACKET &&
4141 (ptr[1] == CHAR_COLON || ptr[1] == CHAR_DOT ||
4142 ptr[1] == CHAR_EQUALS_SIGN) && check_posix_syntax(ptr, &tempptr))
4143 {
4144 BOOL local_negate = FALSE;
4145 int posix_class, taboffset, tabopt;
4146 register const pcre_uint8 *cbits = cd->cbits;
4147 pcre_uint8 pbits[32];
4148
4149 if (ptr[1] != CHAR_COLON)
4150 {
4151 *errorcodeptr = ERR31;
4152 goto FAILED;
4153 }
4154
4155 ptr += 2;
4156 if (*ptr == CHAR_CIRCUMFLEX_ACCENT)
4157 {
4158 local_negate = TRUE;
4159 should_flip_negation = TRUE; /* Note negative special */
4160 ptr++;
4161 }
4162
4163 posix_class = check_posix_name(ptr, (int)(tempptr - ptr));
4164 if (posix_class < 0)
4165 {
4166 *errorcodeptr = ERR30;
4167 goto FAILED;
4168 }
4169
4170 /* If matching is caseless, upper and lower are converted to
4171 alpha. This relies on the fact that the class table starts with
4172 alpha, lower, upper as the first 3 entries. */
4173
4174 if ((options & PCRE_CASELESS) != 0 && posix_class <= 2)
4175 posix_class = 0;
4176
4177 /* When PCRE_UCP is set, some of the POSIX classes are converted to
4178 different escape sequences that use Unicode properties. */
4179
4180 #ifdef SUPPORT_UCP
4181 if ((options & PCRE_UCP) != 0)
4182 {
4183 int pc = posix_class + ((local_negate)? POSIX_SUBSIZE/2 : 0);
4184 if (posix_substitutes[pc] != NULL)
4185 {
4186 nestptr = tempptr + 1;
4187 ptr = posix_substitutes[pc] - 1;
4188 continue;
4189 }
4190 }
4191 #endif
4192 /* In the non-UCP case, we build the bit map for the POSIX class in a
4193 chunk of local store because we may be adding and subtracting from it,
4194 and we don't want to subtract bits that may be in the main map already.
4195 At the end we or the result into the bit map that is being built. */
4196
4197 posix_class *= 3;
4198
4199 /* Copy in the first table (always present) */
4200
4201 memcpy(pbits, cbits + posix_class_maps[posix_class],
4202 32 * sizeof(pcre_uint8));
4203
4204 /* If there is a second table, add or remove it as required. */
4205
4206 taboffset = posix_class_maps[posix_class + 1];
4207 tabopt = posix_class_maps[posix_class + 2];
4208
4209 if (taboffset >= 0)
4210 {
4211 if (tabopt >= 0)
4212 for (c = 0; c < 32; c++) pbits[c] |= cbits[c + taboffset];
4213 else
4214 for (c = 0; c < 32; c++) pbits[c] &= ~cbits[c + taboffset];
4215 }
4216
4217 /* Now see if we need to remove any special characters. An option
4218 value of 1 removes vertical space and 2 removes underscore. */
4219
4220 if (tabopt < 0) tabopt = -tabopt;
4221 if (tabopt == 1) pbits[1] &= ~0x3c;
4222 else if (tabopt == 2) pbits[11] &= 0x7f;
4223
4224 /* Add the POSIX table or its complement into the main table that is
4225 being built and we are done. */
4226
4227 if (local_negate)
4228 for (c = 0; c < 32; c++) classbits[c] |= ~pbits[c];
4229 else
4230 for (c = 0; c < 32; c++) classbits[c] |= pbits[c];
4231
4232 ptr = tempptr + 1;
4233 /* Every class contains at least one < 256 character. */
4234 class_has_8bitchar = 1;
4235 /* Every class contains at least two characters. */
4236 class_one_char = 2;
4237 continue; /* End of POSIX syntax handling */
4238 }
4239
4240 /* Backslash may introduce a single character, or it may introduce one
4241 of the specials, which just set a flag. The sequence \b is a special
4242 case. Inside a class (and only there) it is treated as backspace. We
4243 assume that other escapes have more than one character in them, so
4244 speculatively set both class_has_8bitchar and class_one_char bigger
4245 than one. Unrecognized escapes fall through and are either treated
4246 as literal characters (by default), or are faulted if
4247 PCRE_EXTRA is set. */
4248
4249 if (c == CHAR_BACKSLASH)
4250 {
4251 escape = check_escape(&ptr, &ec, errorcodeptr, cd->bracount, options, TRUE);
4252
4253 if (*errorcodeptr != 0) goto FAILED;
4254
4255 if (escape == 0)
4256 c = ec;
4257 else if (escape == ESC_b) c = CHAR_BS; /* \b is backspace in a class */
4258 else if (escape == ESC_N) /* \N is not supported in a class */
4259 {
4260 *errorcodeptr = ERR71;
4261 goto FAILED;
4262 }
4263 else if (escape == ESC_Q) /* Handle start of quoted string */
4264 {
4265 if (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
4266 {
4267 ptr += 2; /* avoid empty string */
4268 }
4269 else inescq = TRUE;
4270 continue;
4271 }
4272 else if (escape == ESC_E) continue; /* Ignore orphan \E */
4273
4274 else
4275 {
4276 register const pcre_uint8 *cbits = cd->cbits;
4277 /* Every class contains at least two < 256 characters. */
4278 class_has_8bitchar++;
4279 /* Every class contains at least two characters. */
4280 class_one_char += 2;
4281
4282 switch (escape)
4283 {
4284 #ifdef SUPPORT_UCP
4285 case ESC_du: /* These are the values given for \d etc */
4286 case ESC_DU: /* when PCRE_UCP is set. We replace the */
4287 case ESC_wu: /* escape sequence with an appropriate \p */
4288 case ESC_WU: /* or \P to test Unicode properties instead */
4289 case ESC_su: /* of the default ASCII testing. */
4290 case ESC_SU:
4291 nestptr = ptr;
4292 ptr = substitutes[escape - ESC_DU] - 1; /* Just before substitute */
4293 class_has_8bitchar--; /* Undo! */
4294 continue;
4295 #endif
4296 case ESC_d:
4297 for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_digit];
4298 continue;
4299
4300 case ESC_D:
4301 should_flip_negation = TRUE;
4302 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_digit];
4303 continue;
4304
4305 case ESC_w:
4306 for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_word];
4307 continue;
4308
4309 case ESC_W:
4310 should_flip_negation = TRUE;
4311 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_word];
4312 continue;
4313
4314 /* Perl 5.004 onwards omits VT from \s, but we must preserve it
4315 if it was previously set by something earlier in the character
4316 class. Luckily, the value of CHAR_VT is 0x0b in both ASCII and
4317 EBCDIC, so we lazily just adjust the appropriate bit. */
4318
4319 case ESC_s:
4320 classbits[0] |= cbits[cbit_space];
4321 classbits[1] |= cbits[cbit_space+1] & ~0x08;
4322 for (c = 2; c < 32; c++) classbits[c] |= cbits[c+cbit_space];
4323 continue;
4324
4325 case ESC_S:
4326 should_flip_negation = TRUE;
4327 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_space];
4328 classbits[1] |= 0x08; /* Perl 5.004 onwards omits VT from \s */
4329 continue;
4330
4331 /* The rest apply in both UCP and non-UCP cases. */
4332
4333 case ESC_h:
4334 (void)add_list_to_class(classbits, &class_uchardata, options, cd,
4335 PRIV(hspace_list), NOTACHAR);
4336 continue;
4337
4338 case ESC_H:
4339 (void)add_not_list_to_class(classbits, &class_uchardata, options,
4340 cd, PRIV(hspace_list));
4341 continue;
4342
4343 case ESC_v:
4344 (void)add_list_to_class(classbits, &class_uchardata, options, cd,
4345 PRIV(vspace_list), NOTACHAR);
4346 continue;
4347
4348 case ESC_V:
4349 (void)add_not_list_to_class(classbits, &class_uchardata, options,
4350 cd, PRIV(vspace_list));
4351 continue;
4352
4353 #ifdef SUPPORT_UCP
4354 case ESC_p:
4355 case ESC_P:
4356 {
4357 BOOL negated;
4358 int pdata;
4359 int ptype = get_ucp(&ptr, &negated, &pdata, errorcodeptr);
4360 if (ptype < 0) goto FAILED;
4361 *class_uchardata++ = ((escape == ESC_p) != negated)?
4362 XCL_PROP : XCL_NOTPROP;
4363 *class_uchardata++ = ptype;
4364 *class_uchardata++ = pdata;
4365 class_has_8bitchar--; /* Undo! */
4366 continue;
4367 }
4368 #endif
4369 /* Unrecognized escapes are faulted if PCRE is running in its
4370 strict mode. By default, for compatibility with Perl, they are
4371 treated as literals. */
4372
4373 default:
4374 if ((options & PCRE_EXTRA) != 0)
4375 {
4376 *errorcodeptr = ERR7;
4377 goto FAILED;
4378 }
4379 class_has_8bitchar--; /* Undo the speculative increase. */
4380 class_one_char -= 2; /* Undo the speculative increase. */
4381 c = *ptr; /* Get the final character and fall through */
4382 break;
4383 }
4384 }
4385
4386 /* Fall through if the escape just defined a single character (c >= 0).
4387 This may be greater than 256. */
4388
4389 escape = 0;
4390
4391 } /* End of backslash handling */
4392
4393 /* A character may be followed by '-' to form a range. However, Perl does
4394 not permit ']' to be the end of the range. A '-' character at the end is
4395 treated as a literal. Perl ignores orphaned \E sequences entirely. The
4396 code for handling \Q and \E is messy. */
4397
4398 CHECK_RANGE:
4399 while (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
4400 {
4401 inescq = FALSE;
4402 ptr += 2;
4403 }
4404 oldptr = ptr;
4405
4406 /* Remember if \r or \n were explicitly used */
4407
4408 if (c == CHAR_CR || c == CHAR_NL) cd->external_flags |= PCRE_HASCRORLF;
4409
4410 /* Check for range */
4411
4412 if (!inescq && ptr[1] == CHAR_MINUS)
4413 {
4414 int d;
4415 ptr += 2;
4416 while (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_E) ptr += 2;
4417
4418 /* If we hit \Q (not followed by \E) at this point, go into escaped
4419 mode. */
4420
4421 while (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_Q)
4422 {
4423 ptr += 2;
4424 if (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_E)
4425 { ptr += 2; continue; }
4426 inescq = TRUE;
4427 break;
4428 }
4429
4430 /* Minus (hyphen) at the end of a class is treated as a literal, so put
4431 back the pointer and jump to handle the character that preceded it. */
4432
4433 if (*ptr == 0 || (!inescq && *ptr == CHAR_RIGHT_SQUARE_BRACKET))
4434 {
4435 ptr = oldptr;
4436 goto CLASS_SINGLE_CHARACTER;
4437 }
4438
4439 /* Otherwise, we have a potential range; pick up the next character */
4440
4441 #ifdef SUPPORT_UTF
4442 if (utf)
4443 { /* Braces are required because the */
4444 GETCHARLEN(d, ptr, ptr); /* macro generates multiple statements */
4445 }
4446 else
4447 #endif
4448 d = *ptr; /* Not UTF-8 mode */
4449
4450 /* The second part of a range can be a single-character escape, but
4451 not any of the other escapes. Perl 5.6 treats a hyphen as a literal
4452 in such circumstances. */
4453
4454 if (!inescq && d == CHAR_BACKSLASH)
4455 {
4456 int descape;
4457 descape = check_escape(&ptr, &d, errorcodeptr, cd->bracount, options, TRUE);
4458 if (*errorcodeptr != 0) goto FAILED;
4459
4460 /* \b is backspace; any other special means the '-' was literal. */
4461
4462 if (descape > 0)
4463 {
4464 if (descape == ESC_b) d = CHAR_BS; else
4465 {
4466 ptr = oldptr;
4467 goto CLASS_SINGLE_CHARACTER; /* A few lines below */
4468 }
4469 }
4470 }
4471
4472 /* Check that the two values are in the correct order. Optimize
4473 one-character ranges. */
4474
4475 if (d < c)
4476 {
4477 *errorcodeptr = ERR8;
4478 goto FAILED;
4479 }
4480 if (d == c) goto CLASS_SINGLE_CHARACTER; /* A few lines below */
4481
4482 /* We have found a character range, so single character optimizations
4483 cannot be done anymore. Any value greater than 1 indicates that there
4484 is more than one character. */
4485
4486 class_one_char = 2;
4487
4488 /* Remember an explicit \r or \n, and add the range to the class. */
4489
4490 if (d == CHAR_CR || d == CHAR_NL) cd->external_flags |= PCRE_HASCRORLF;
4491
4492 class_has_8bitchar +=
4493 add_to_class(classbits, &class_uchardata, options, cd, c, d);
4494
4495 continue; /* Go get the next char in the class */
4496 }
4497
4498 /* Handle a single character - we can get here for a normal non-escape
4499 char, or after \ that introduces a single character or for an apparent
4500 range that isn't. Only the value 1 matters for class_one_char, so don't
4501 increase it if it is already 2 or more ... just in case there's a class
4502 with a zillion characters in it. */
4503
4504 CLASS_SINGLE_CHARACTER:
4505 if (class_one_char < 2) class_one_char++;
4506
4507 /* If class_one_char is 1, we have the first single character in the
4508 class, and there have been no prior ranges, or XCLASS items generated by
4509 escapes. If this is the final character in the class, we can optimize by
4510 turning the item into a 1-character OP_CHAR[I] if it's positive, or
4511 OP_NOT[I] if it's negative. In the positive case, it can cause firstchar
4512 to be set. Otherwise, there can be no first char if this item is first,
4513 whatever repeat count may follow. In the case of reqchar, save the
4514 previous value for reinstating. */
4515
4516 if (class_one_char == 1 && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET)
4517 {
4518 ptr++;
4519 zeroreqchar = reqchar;
4520
4521 if (negate_class)
4522 {
4523 #ifdef SUPPORT_UCP
4524 // FIXMEchpe pcreuint32?
4525 int d;
4526 #endif
4527 if (firstchar == REQ_UNSET) firstchar = REQ_NONE;
4528 zerofirstchar = firstchar;
4529
4530 /* For caseless UTF-8 mode when UCP support is available, check
4531 whether this character has more than one other case. If so, generate
4532 a special OP_NOTPROP item instead of OP_NOTI. */
4533
4534 #ifdef SUPPORT_UCP
4535 if (utf && (options & PCRE_CASELESS) != 0 &&
4536 (d = UCD_CASESET(c)) != 0)
4537 {
4538 *code++ = OP_NOTPROP;
4539 *code++ = PT_CLIST;
4540 *code++ = d;
4541 }
4542 else
4543 #endif
4544 /* Char has only one other case, or UCP not available */
4545
4546 {
4547 *code++ = ((options & PCRE_CASELESS) != 0)? OP_NOTI: OP_NOT;
4548 #if defined SUPPORT_UTF && !defined COMPILE_PCRE32
4549 if (utf && c > MAX_VALUE_FOR_SINGLE_CHAR)
4550 code += PRIV(ord2utf)(c, code);
4551 else
4552 #endif
4553 *code++ = c;
4554 }
4555
4556 /* We are finished with this character class */
4557
4558 goto END_CLASS;
4559 }
4560
4561 /* For a single, positive character, get the value into mcbuffer, and
4562 then we can handle this with the normal one-character code. */
4563
4564 #if defined SUPPORT_UTF && !defined COMPILE_PCRE32
4565 if (utf && c > MAX_VALUE_FOR_SINGLE_CHAR)
4566 mclength = PRIV(ord2utf)(c, mcbuffer);
4567 else
4568 #endif
4569 {
4570 mcbuffer[0] = c;
4571 mclength = 1;
4572 }
4573 goto ONE_CHAR;
4574 } /* End of 1-char optimization */
4575
4576 /* There is more than one character in the class, or an XCLASS item
4577 has been generated. Add this character to the class. */
4578
4579 class_has_8bitchar +=
4580 add_to_class(classbits, &class_uchardata, options, cd, c, c);
4581 }
4582
4583 /* Loop until ']' reached. This "while" is the end of the "do" far above.
4584 If we are at the end of an internal nested string, revert to the outer
4585 string. */
4586
4587 while (((c = *(++ptr)) != 0 ||
4588 (nestptr != NULL &&
4589 (ptr = nestptr, nestptr = NULL, c = *(++ptr)) != 0)) &&
4590 (c != CHAR_RIGHT_SQUARE_BRACKET || inescq));
4591
4592 /* Check for missing terminating ']' */
4593
4594 if (c == 0)
4595 {
4596 *errorcodeptr = ERR6;
4597 goto FAILED;
4598 }
4599
4600 /* We will need an XCLASS if data has been placed in class_uchardata. In
4601 the second phase this is a sufficient test. However, in the pre-compile
4602 phase, class_uchardata gets emptied to prevent workspace overflow, so it
4603 only if the very last character in the class needs XCLASS will it contain
4604 anything at this point. For this reason, xclass gets set TRUE above when
4605 uchar_classdata is emptied, and that's why this code is the way it is here
4606 instead of just doing a test on class_uchardata below. */
4607
4608 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
4609 if (class_uchardata > class_uchardata_base) xclass = TRUE;
4610 #endif
4611
4612 /* If this is the first thing in the branch, there can be no first char
4613 setting, whatever the repeat count. Any reqchar setting must remain
4614 unchanged after any kind of repeat. */
4615
4616 if (firstchar == REQ_UNSET) firstchar = REQ_NONE;
4617 zerofirstchar = firstchar;
4618 zeroreqchar = reqchar;
4619
4620 /* If there are characters with values > 255, we have to compile an
4621 extended class, with its own opcode, unless there was a negated special
4622 such as \S in the class, and PCRE_UCP is not set, because in that case all
4623 characters > 255 are in the class, so any that were explicitly given as
4624 well can be ignored. If (when there are explicit characters > 255 that must
4625 be listed) there are no characters < 256, we can omit the bitmap in the
4626 actual compiled code. */
4627
4628 #ifdef SUPPORT_UTF
4629 if (xclass && (!should_flip_negation || (options & PCRE_UCP) != 0))
4630 #elif !defined COMPILE_PCRE8
4631 if (xclass && !should_flip_negation)
4632 #endif
4633 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
4634 {
4635 *class_uchardata++ = XCL_END; /* Marks the end of extra data */
4636 *code++ = OP_XCLASS;
4637 code += LINK_SIZE;
4638 *code = negate_class? XCL_NOT:0;
4639
4640 /* If the map is required, move up the extra data to make room for it;
4641 otherwise just move the code pointer to the end of the extra data. */
4642
4643 if (class_has_8bitchar > 0)
4644 {
4645 *code++ |= XCL_MAP;
4646 memmove(code + (32 / sizeof(pcre_uchar)), code,
4647 IN_UCHARS(class_uchardata - code));
4648 memcpy(code, classbits, 32);
4649 code = class_uchardata + (32 / sizeof(pcre_uchar));
4650 }
4651 else code = class_uchardata;
4652
4653 /* Now fill in the complete length of the item */
4654
4655 PUT(previous, 1, (int)(code - previous));
4656 break; /* End of class handling */
4657 }
4658 #endif
4659
4660 /* If there are no characters > 255, or they are all to be included or
4661 excluded, set the opcode to OP_CLASS or OP_NCLASS, depending on whether the
4662 whole class was negated and whether there were negative specials such as \S
4663 (non-UCP) in the class. Then copy the 32-byte map into the code vector,
4664 negating it if necessary. */
4665
4666 *code++ = (negate_class == should_flip_negation) ? OP_CLASS : OP_NCLASS;
4667 if (lengthptr == NULL) /* Save time in the pre-compile phase */
4668 {
4669 if (negate_class)
4670 for (c = 0; c < 32; c++) classbits[c] = ~classbits[c];
4671 memcpy(code, classbits, 32);
4672 }
4673 code += 32 / sizeof(pcre_uchar);
4674
4675 END_CLASS:
4676 break;
4677
4678
4679 /* ===================================================================*/
4680 /* Various kinds of repeat; '{' is not necessarily a quantifier, but this
4681 has been tested above. */
4682
4683 case CHAR_LEFT_CURLY_BRACKET:
4684 if (!is_quantifier) goto NORMAL_CHAR;
4685 ptr = read_repeat_counts(ptr+1, &repeat_min, &repeat_max, errorcodeptr);
4686 if (*errorcodeptr != 0) goto FAILED;
4687 goto REPEAT;
4688
4689 case CHAR_ASTERISK:
4690 repeat_min = 0;
4691 repeat_max = -1;
4692 goto REPEAT;
4693
4694 case CHAR_PLUS:
4695 repeat_min = 1;
4696 repeat_max = -1;
4697 goto REPEAT;
4698
4699 case CHAR_QUESTION_MARK:
4700 repeat_min = 0;
4701 repeat_max = 1;
4702
4703 REPEAT:
4704 if (previous == NULL)
4705 {
4706 *errorcodeptr = ERR9;
4707 goto FAILED;
4708 }
4709
4710 if (repeat_min == 0)
4711 {
4712 firstchar = zerofirstchar; /* Adjust for zero repeat */
4713 reqchar = zeroreqchar; /* Ditto */
4714 }
4715
4716 /* Remember whether this is a variable length repeat */
4717
4718 reqvary = (repeat_min == repeat_max)? 0 : REQ_VARY;
4719
4720 op_type = 0; /* Default single-char op codes */
4721 possessive_quantifier = FALSE; /* Default not possessive quantifier */
4722
4723 /* Save start of previous item, in case we have to move it up in order to
4724 insert something before it. */
4725
4726 tempcode = previous;
4727
4728 /* If the next character is '+', we have a possessive quantifier. This
4729 implies greediness, whatever the setting of the PCRE_UNGREEDY option.
4730 If the next character is '?' this is a minimizing repeat, by default,
4731 but if PCRE_UNGREEDY is set, it works the other way round. We change the
4732 repeat type to the non-default. */
4733
4734 if (ptr[1] == CHAR_PLUS)
4735 {
4736 repeat_type = 0; /* Force greedy */
4737 possessive_quantifier = TRUE;
4738 ptr++;
4739 }
4740 else if (ptr[1] == CHAR_QUESTION_MARK)
4741 {
4742 repeat_type = greedy_non_default;
4743 ptr++;
4744 }
4745 else repeat_type = greedy_default;
4746
4747 /* If previous was a recursion call, wrap it in atomic brackets so that
4748 previous becomes the atomic group. All recursions were so wrapped in the
4749 past, but it no longer happens for non-repeated recursions. In fact, the
4750 repeated ones could be re-implemented independently so as not to need this,
4751 but for the moment we rely on the code for repeating groups. */
4752
4753 if (*previous == OP_RECURSE)
4754 {
4755 memmove(previous + 1 + LINK_SIZE, previous, IN_UCHARS(1 + LINK_SIZE));
4756 *previous = OP_ONCE;
4757 PUT(previous, 1, 2 + 2*LINK_SIZE);
4758 previous[2 + 2*LINK_SIZE] = OP_KET;
4759 PUT(previous, 3 + 2*LINK_SIZE, 2 + 2*LINK_SIZE);
4760 code += 2 + 2 * LINK_SIZE;
4761 length_prevgroup = 3 + 3*LINK_SIZE;
4762
4763 /* When actually compiling, we need to check whether this was a forward
4764 reference, and if so, adjust the offset. */
4765
4766 if (lengthptr == NULL && cd->hwm >= cd->start_workspace + LINK_SIZE)
4767 {
4768 int offset = GET(cd->hwm, -LINK_SIZE);
4769 if (offset == previous + 1 - cd->start_code)
4770 PUT(cd->hwm, -LINK_SIZE, offset + 1 + LINK_SIZE);
4771 }
4772 }
4773
4774 /* Now handle repetition for the different types of item. */
4775
4776 /* If previous was a character or negated character match, abolish the item
4777 and generate a repeat item instead. If a char item has a minimum of more
4778 than one, ensure that it is set in reqchar - it might not be if a sequence
4779 such as x{3} is the first thing in a branch because the x will have gone
4780 into firstchar instead. */
4781
4782 if (*previous == OP_CHAR || *previous == OP_CHARI
4783 || *previous == OP_NOT || *previous == OP_NOTI)
4784 {
4785 switch (*previous)
4786 {
4787 default: /* Make compiler happy. */
4788 case OP_CHAR: op_type = OP_STAR - OP_STAR; break;
4789 case OP_CHARI: op_type = OP_STARI - OP_STAR; break;
4790 case OP_NOT: op_type = OP_NOTSTAR - OP_STAR; break;
4791 case OP_NOTI: op_type = OP_NOTSTARI - OP_STAR; break;
4792 }
4793
4794 /* Deal with UTF characters that take up more than one character. It's
4795 easier to write this out separately than try to macrify it. Use c to
4796 hold the length of the character in bytes, plus UTF_LENGTH to flag that
4797 it's a length rather than a small character. */
4798
4799 #if defined SUPPORT_UTF && !defined COMPILE_PCRE32
4800 if (utf && NOT_FIRSTCHAR(code[-1]))
4801 {
4802 pcre_uchar *lastchar = code - 1;
4803 BACKCHAR(lastchar);
4804 c = (int)(code - lastchar); /* Length of UTF-8 character */
4805 memcpy(utf_chars, lastchar, IN_UCHARS(c)); /* Save the char */
4806 c |= UTF_LENGTH; /* Flag c as a length */
4807 }
4808 else
4809 #endif /* SUPPORT_UTF */
4810
4811 /* Handle the case of a single charater - either with no UTF support, or
4812 with UTF disabled, or for a single character UTF character. */
4813 {
4814 c = code[-1];
4815 if (*previous <= OP_CHARI && repeat_min > 1)
4816 reqchar = c | req_caseopt | cd->req_varyopt;
4817 }
4818
4819 /* If the repetition is unlimited, it pays to see if the next thing on
4820 the line is something that cannot possibly match this character. If so,
4821 automatically possessifying this item gains some performance in the case
4822 where the match fails. */
4823
4824 if (!possessive_quantifier &&
4825 repeat_max < 0 &&
4826 check_auto_possessive(previous, utf, ptr + 1, options, cd))
4827 {
4828 repeat_type = 0; /* Force greedy */
4829 possessive_quantifier = TRUE;
4830 }
4831
4832 goto OUTPUT_SINGLE_REPEAT; /* Code shared with single character types */
4833 }
4834
4835 /* If previous was a character type match (\d or similar), abolish it and
4836 create a suitable repeat item. The code is shared with single-character
4837 repeats by setting op_type to add a suitable offset into repeat_type. Note
4838 the the Unicode property types will be present only when SUPPORT_UCP is
4839 defined, but we don't wrap the little bits of code here because it just
4840 makes it horribly messy. */
4841
4842 else if (*previous < OP_EODN)
4843 {
4844 pcre_uchar *oldcode;
4845 int prop_type, prop_value;
4846 op_type = OP_TYPESTAR - OP_STAR; /* Use type opcodes */
4847 c = *previous;
4848
4849 if (!possessive_quantifier &&
4850 repeat_max < 0 &&
4851 check_auto_possessive(previous, utf, ptr + 1, options, cd))
4852 {
4853 repeat_type = 0; /* Force greedy */
4854 possessive_quantifier = TRUE;
4855 }
4856
4857 OUTPUT_SINGLE_REPEAT:
4858 if (*previous == OP_PROP || *previous == OP_NOTPROP)
4859 {
4860 prop_type = previous[1];
4861 prop_value = previous[2];
4862 }
4863 else prop_type = prop_value = -1;
4864
4865 oldcode = code;
4866 code = previous; /* Usually overwrite previous item */
4867
4868 /* If the maximum is zero then the minimum must also be zero; Perl allows
4869 this case, so we do too - by simply omitting the item altogether. */
4870
4871 if (repeat_max == 0) goto END_REPEAT;
4872
4873 /*--------------------------------------------------------------------*/
4874 /* This code is obsolete from release 8.00; the restriction was finally
4875 removed: */
4876
4877 /* All real repeats make it impossible to handle partial matching (maybe
4878 one day we will be able to remove this restriction). */
4879
4880 /* if (repeat_max != 1) cd->external_flags |= PCRE_NOPARTIAL; */
4881 /*--------------------------------------------------------------------*/
4882
4883 /* Combine the op_type with the repeat_type */
4884
4885 repeat_type += op_type;
4886
4887 /* A minimum of zero is handled either as the special case * or ?, or as
4888 an UPTO, with the maximum given. */
4889
4890 if (repeat_min == 0)
4891 {
4892 if (repeat_max == -1) *code++ = OP_STAR + repeat_type;
4893 else if (repeat_max == 1) *code++ = OP_QUERY + repeat_type;
4894 else
4895 {
4896 *code++ = OP_UPTO + repeat_type;
4897 PUT2INC(code, 0, repeat_max);
4898 }
4899 }
4900
4901 /* A repeat minimum of 1 is optimized into some special cases. If the
4902 maximum is unlimited, we use OP_PLUS. Otherwise, the original item is
4903 left in place and, if the maximum is greater than 1, we use OP_UPTO with
4904 one less than the maximum. */
4905
4906 else if (repeat_min == 1)
4907 {
4908 if (repeat_max == -1)
4909 *code++ = OP_PLUS + repeat_type;
4910 else
4911 {
4912 code = oldcode; /* leave previous item in place */
4913 if (repeat_max == 1) goto END_REPEAT;
4914 *code++ = OP_UPTO + repeat_type;
4915 PUT2INC(code, 0, repeat_max - 1);
4916 }
4917 }
4918
4919 /* The case {n,n} is just an EXACT, while the general case {n,m} is
4920 handled as an EXACT followed by an UPTO. */
4921
4922 else
4923 {
4924 *code++ = OP_EXACT + op_type; /* NB EXACT doesn't have repeat_type */
4925 PUT2INC(code, 0, repeat_min);
4926
4927 /* If the maximum is unlimited, insert an OP_STAR. Before doing so,
4928 we have to insert the character for the previous code. For a repeated
4929 Unicode property match, there are two extra bytes that define the
4930 required property. In UTF-8 mode, long characters have their length in
4931 c, with the UTF_LENGTH bit as a flag. */
4932
4933 if (repeat_max < 0)
4934 {
4935 #if defined SUPPORT_UTF && !defined COMPILE_PCRE32
4936 if (utf && (c & UTF_LENGTH) != 0)
4937 {
4938 memcpy(code, utf_chars, IN_UCHARS(c & 7));
4939 code += c & 7;
4940 }
4941 else
4942 #endif
4943 {
4944 *code++ = c;
4945 if (prop_type >= 0)
4946 {
4947 *code++ = prop_type;
4948 *code++ = prop_value;
4949 }
4950 }
4951 *code++ = OP_STAR + repeat_type;
4952 }
4953
4954 /* Else insert an UPTO if the max is greater than the min, again
4955 preceded by the character, for the previously inserted code. If the
4956 UPTO is just for 1 instance, we can use QUERY instead. */
4957
4958 else if (repeat_max != repeat_min)
4959 {
4960 #if defined SUPPORT_UTF && !defined COMPILE_PCRE32
4961 if (utf && (c & UTF_LENGTH) != 0)
4962 {
4963 memcpy(code, utf_chars, IN_UCHARS(c & 7));
4964 code += c & 7;
4965 }
4966 else
4967 #endif
4968 *code++ = c;
4969 if (prop_type >= 0)
4970 {
4971 *code++ = prop_type;
4972 *code++ = prop_value;
4973 }
4974 repeat_max -= repeat_min;
4975
4976 if (repeat_max == 1)
4977 {
4978 *code++ = OP_QUERY + repeat_type;
4979 }
4980 else
4981 {
4982 *code++ = OP_UPTO + repeat_type;
4983 PUT2INC(code, 0, repeat_max);
4984 }
4985 }
4986 }
4987
4988 /* The character or character type itself comes last in all cases. */
4989
4990 #if defined SUPPORT_UTF && !defined COMPILE_PCRE32
4991 if (utf && (c & UTF_LENGTH) != 0)
4992 {
4993 memcpy(code, utf_chars, IN_UCHARS(c & 7));
4994 code += c & 7;
4995 }
4996 else
4997 #endif
4998 *code++ = c;
4999
5000 /* For a repeated Unicode property match, there are two extra bytes that
5001 define the required property. */
5002
5003 #ifdef SUPPORT_UCP
5004 if (prop_type >= 0)
5005 {
5006 *code++ = prop_type;
5007 *code++ = prop_value;
5008 }
5009 #endif
5010 }
5011
5012 /* If previous was a character class or a back reference, we put the repeat
5013 stuff after it, but just skip the item if the repeat was {0,0}. */
5014
5015 else if (*previous == OP_CLASS ||
5016 *previous == OP_NCLASS ||
5017 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
5018 *previous == OP_XCLASS ||
5019 #endif
5020 *previous == OP_REF ||
5021 *previous == OP_REFI)
5022 {
5023 if (repeat_max == 0)
5024 {
5025 code = previous;
5026 goto END_REPEAT;
5027 }
5028
5029 /*--------------------------------------------------------------------*/
5030 /* This code is obsolete from release 8.00; the restriction was finally
5031 removed: */
5032
5033 /* All real repeats make it impossible to handle partial matching (maybe
5034 one day we will be able to remove this restriction). */
5035
5036 /* if (repeat_max != 1) cd->external_flags |= PCRE_NOPARTIAL; */
5037 /*--------------------------------------------------------------------*/
5038
5039 if (repeat_min == 0 && repeat_max == -1)
5040 *code++ = OP_CRSTAR + repeat_type;
5041 else if (repeat_min == 1 && repeat_max == -1)
5042 *code++ = OP_CRPLUS + repeat_type;
5043 else if (repeat_min == 0 && repeat_max == 1)
5044 *code++ = OP_CRQUERY + repeat_type;
5045 else
5046 {
5047 *code++ = OP_CRRANGE + repeat_type;
5048 PUT2INC(code, 0, repeat_min);
5049 if (repeat_max == -1) repeat_max = 0; /* 2-byte encoding for max */
5050 PUT2INC(code, 0, repeat_max);
5051 }
5052 }
5053
5054 /* If previous was a bracket group, we may have to replicate it in certain
5055 cases. Note that at this point we can encounter only the "basic" bracket
5056 opcodes such as BRA and CBRA, as this is the place where they get converted
5057 into the more special varieties such as BRAPOS and SBRA. A test for >=
5058 OP_ASSERT and <= OP_COND includes ASSERT, ASSERT_NOT, ASSERTBACK,
5059 ASSERTBACK_NOT, ONCE, BRA, CBRA, and COND. Originally, PCRE did not allow
5060 repetition of assertions, but now it does, for Perl compatibility. */
5061
5062 else if (*previous >= OP_ASSERT && *previous <= OP_COND)
5063 {
5064 register int i;
5065 int len = (int)(code - previous);
5066 pcre_uchar *bralink = NULL;
5067 pcre_uchar *brazeroptr = NULL;
5068
5069 /* Repeating a DEFINE group is pointless, but Perl allows the syntax, so
5070 we just ignore the repeat. */
5071
5072 if (*previous == OP_COND && previous[LINK_SIZE+1] == OP_DEF)
5073 goto END_REPEAT;
5074
5075 /* There is no sense in actually repeating assertions. The only potential
5076 use of repetition is in cases when the assertion is optional. Therefore,
5077 if the minimum is greater than zero, just ignore the repeat. If the
5078 maximum is not not zero or one, set it to 1. */
5079
5080 if (*previous < OP_ONCE) /* Assertion */
5081 {
5082 if (repeat_min > 0) goto END_REPEAT;
5083 if (repeat_max < 0 || repeat_max > 1) repeat_max = 1;
5084 }
5085
5086 /* The case of a zero minimum is special because of the need to stick
5087 OP_BRAZERO in front of it, and because the group appears once in the
5088 data, whereas in other cases it appears the minimum number of times. For
5089 this reason, it is simplest to treat this case separately, as otherwise
5090 the code gets far too messy. There are several special subcases when the
5091 minimum is zero. */
5092
5093 if (repeat_min == 0)
5094 {
5095 /* If the maximum is also zero, we used to just omit the group from the
5096 output altogether, like this:
5097
5098 ** if (repeat_max == 0)
5099 ** {
5100 ** code = previous;
5101 ** goto END_REPEAT;
5102 ** }
5103
5104 However, that fails when a group or a subgroup within it is referenced
5105 as a subroutine from elsewhere in the pattern, so now we stick in
5106 OP_SKIPZERO in front of it so that it is skipped on execution. As we
5107 don't have a list of which groups are referenced, we cannot do this
5108 selectively.
5109
5110 If the maximum is 1 or unlimited, we just have to stick in the BRAZERO
5111 and do no more at this point. However, we do need to adjust any
5112 OP_RECURSE calls inside the group that refer to the group itself or any
5113 internal or forward referenced group, because the offset is from the
5114 start of the whole regex. Temporarily terminate the pattern while doing
5115 this. */
5116
5117 if (repeat_max <= 1) /* Covers 0, 1, and unlimited */
5118 {
5119 *code = OP_END;
5120 adjust_recurse(previous, 1, utf, cd, save_hwm);
5121 memmove(previous + 1, previous, IN_UCHARS(len));
5122 code++;
5123 if (repeat_max == 0)
5124 {
5125 *previous++ = OP_SKIPZERO;
5126 goto END_REPEAT;
5127 }
5128 brazeroptr = previous; /* Save for possessive optimizing */
5129 *previous++ = OP_BRAZERO + repeat_type;
5130 }
5131
5132 /* If the maximum is greater than 1 and limited, we have to replicate
5133 in a nested fashion, sticking OP_BRAZERO before each set of brackets.
5134 The first one has to be handled carefully because it's the original
5135 copy, which has to be moved up. The remainder can be handled by code
5136 that is common with the non-zero minimum case below. We have to
5137 adjust the value or repeat_max, since one less copy is required. Once
5138 again, we may have to adjust any OP_RECURSE calls inside the group. */
5139
5140 else
5141 {
5142 int offset;
5143 *code = OP_END;
5144 adjust_recurse(previous, 2 + LINK_SIZE, utf, cd, save_hwm);
5145 memmove(previous + 2 + LINK_SIZE, previous, IN_UCHARS(len));
5146 code += 2 + LINK_SIZE;
5147 *previous++ = OP_BRAZERO + repeat_type;
5148 *previous++ = OP_BRA;
5149
5150 /* We chain together the bracket offset fields that have to be
5151 filled in later when the ends of the brackets are reached. */
5152
5153 offset = (bralink == NULL)? 0 : (int)(previous - bralink);
5154 bralink = previous;
5155 PUTINC(previous, 0, offset);
5156 }
5157
5158 repeat_max--;
5159 }
5160
5161 /* If the minimum is greater than zero, replicate the group as many
5162 times as necessary, and adjust the maximum to the number of subsequent
5163 copies that we need. If we set a first char from the group, and didn't
5164 set a required char, copy the latter from the former. If there are any
5165 forward reference subroutine calls in the group, there will be entries on
5166 the workspace list; replicate these with an appropriate increment. */
5167
5168 else
5169 {
5170 if (repeat_min > 1)
5171 {
5172 /* In the pre-compile phase, we don't actually do the replication. We
5173 just adjust the length as if we had. Do some paranoid checks for
5174 potential integer overflow. The INT64_OR_DOUBLE type is a 64-bit
5175 integer type when available, otherwise double. */
5176
5177 if (lengthptr != NULL)
5178 {
5179 int delta = (repeat_min - 1)*length_prevgroup;
5180 if ((INT64_OR_DOUBLE)(repeat_min - 1)*
5181 (INT64_OR_DOUBLE)length_prevgroup >
5182 (INT64_OR_DOUBLE)INT_MAX ||
5183 OFLOW_MAX - *lengthptr < delta)
5184 {
5185 *errorcodeptr = ERR20;
5186 goto FAILED;
5187 }
5188 *lengthptr += delta;
5189 }
5190
5191 /* This is compiling for real. If there is a set first byte for
5192 the group, and we have not yet set a "required byte", set it. Make
5193 sure there is enough workspace for copying forward references before
5194 doing the copy. */
5195
5196 else
5197 {
5198 if (groupsetfirstchar && reqchar < 0) reqchar = firstchar;
5199
5200 for (i = 1; i < repeat_min; i++)
5201 {
5202 pcre_uchar *hc;
5203 pcre_uchar *this_hwm = cd->hwm;
5204 memcpy(code, previous, IN_UCHARS(len));
5205
5206 while (cd->hwm > cd->start_workspace + cd->workspace_size -
5207 WORK_SIZE_SAFETY_MARGIN - (this_hwm - save_hwm))
5208 {
5209 int save_offset = save_hwm - cd->start_workspace;
5210 int this_offset = this_hwm - cd->start_workspace;
5211 *errorcodeptr = expand_workspace(cd);
5212 if (*errorcodeptr != 0) goto FAILED;
5213 save_hwm = (pcre_uchar *)cd->start_workspace + save_offset;
5214 this_hwm = (pcre_uchar *)cd->start_workspace + this_offset;
5215 }
5216
5217 for (hc = save_hwm; hc < this_hwm; hc += LINK_SIZE)
5218 {
5219 PUT(cd->hwm, 0, GET(hc, 0) + len);
5220 cd->hwm += LINK_SIZE;
5221 }
5222 save_hwm = this_hwm;
5223 code += len;
5224 }
5225 }
5226 }
5227
5228 if (repeat_max > 0) repeat_max -= repeat_min;
5229 }
5230
5231 /* This code is common to both the zero and non-zero minimum cases. If
5232 the maximum is limited, it replicates the group in a nested fashion,
5233 remembering the bracket starts on a stack. In the case of a zero minimum,
5234 the first one was set up above. In all cases the repeat_max now specifies
5235 the number of additional copies needed. Again, we must remember to
5236 replicate entries on the forward reference list. */
5237
5238 if (repeat_max >= 0)
5239 {
5240 /* In the pre-compile phase, we don't actually do the replication. We
5241 just adjust the length as if we had. For each repetition we must add 1
5242 to the length for BRAZERO and for all but the last repetition we must
5243 add 2 + 2*LINKSIZE to allow for the nesting that occurs. Do some
5244 paranoid checks to avoid integer overflow. The INT64_OR_DOUBLE type is
5245 a 64-bit integer type when available, otherwise double. */
5246
5247 if (lengthptr != NULL && repeat_max > 0)
5248 {
5249 int delta = repeat_max * (length_prevgroup + 1 + 2 + 2*LINK_SIZE) -
5250 2 - 2*LINK_SIZE; /* Last one doesn't nest */
5251 if ((INT64_OR_DOUBLE)repeat_max *
5252 (INT64_OR_DOUBLE)(length_prevgroup + 1 + 2 + 2*LINK_SIZE)
5253 > (INT64_OR_DOUBLE)INT_MAX ||
5254 OFLOW_MAX - *lengthptr < delta)
5255 {
5256 *errorcodeptr = ERR20;
5257 goto FAILED;
5258 }
5259 *lengthptr += delta;
5260 }
5261
5262 /* This is compiling for real */
5263
5264 else for (i = repeat_max - 1; i >= 0; i--)
5265 {
5266 pcre_uchar *hc;
5267 pcre_uchar *this_hwm = cd->hwm;
5268
5269 *code++ = OP_BRAZERO + repeat_type;
5270
5271 /* All but the final copy start a new nesting, maintaining the
5272 chain of brackets outstanding. */
5273
5274 if (i != 0)
5275 {
5276 int offset;
5277 *code++ = OP_BRA;
5278 offset = (bralink == NULL)? 0 : (int)(code - bralink);
5279 bralink = code;
5280 PUTINC(code, 0, offset);
5281 }
5282
5283 memcpy(code, previous, IN_UCHARS(len));
5284
5285 /* Ensure there is enough workspace for forward references before
5286 copying them. */
5287
5288 while (cd->hwm > cd->start_workspace + cd->workspace_size -
5289 WORK_SIZE_SAFETY_MARGIN - (this_hwm - save_hwm))
5290 {
5291 int save_offset = save_hwm - cd->start_workspace;
5292 int this_offset = this_hwm - cd->start_workspace;
5293 *errorcodeptr = expand_workspace(cd);
5294 if (*errorcodeptr != 0) goto FAILED;
5295 save_hwm = (pcre_uchar *)cd->start_workspace + save_offset;
5296 this_hwm = (pcre_uchar *)cd->start_workspace + this_offset;
5297 }
5298
5299 for (hc = save_hwm; hc < this_hwm; hc += LINK_SIZE)
5300 {
5301 PUT(cd->hwm, 0, GET(hc, 0) + len + ((i != 0)? 2+LINK_SIZE : 1));
5302 cd->hwm += LINK_SIZE;
5303 }
5304 save_hwm = this_hwm;
5305 code += len;
5306 }
5307
5308 /* Now chain through the pending brackets, and fill in their length
5309 fields (which are holding the chain links pro tem). */
5310
5311 while (bralink != NULL)
5312 {
5313 int oldlinkoffset;
5314 int offset = (int)(code - bralink + 1);
5315 pcre_uchar *bra = code - offset;
5316 oldlinkoffset = GET(bra, 1);
5317 bralink = (oldlinkoffset == 0)? NULL : bralink - oldlinkoffset;
5318 *code++ = OP_KET;
5319 PUTINC(code, 0, offset);
5320 PUT(bra, 1, offset);
5321 }
5322 }
5323
5324 /* If the maximum is unlimited, set a repeater in the final copy. For
5325 ONCE brackets, that's all we need to do. However, possessively repeated
5326 ONCE brackets can be converted into non-capturing brackets, as the
5327 behaviour of (?:xx)++ is the same as (?>xx)++ and this saves having to
5328 deal with possessive ONCEs specially.
5329
5330 Otherwise, when we are doing the actual compile phase, check to see
5331 whether this group is one that could match an empty string. If so,
5332 convert the initial operator to the S form (e.g. OP_BRA -> OP_SBRA) so
5333 that runtime checking can be done. [This check is also applied to ONCE
5334 groups at runtime, but in a different way.]
5335
5336 Then, if the quantifier was possessive and the bracket is not a
5337 conditional, we convert the BRA code to the POS form, and the KET code to
5338 KETRPOS. (It turns out to be convenient at runtime to detect this kind of
5339 subpattern at both the start and at the end.) The use of special opcodes
5340 makes it possible to reduce greatly the stack usage in pcre_exec(). If
5341 the group is preceded by OP_BRAZERO, convert this to OP_BRAPOSZERO.
5342
5343 Then, if the minimum number of matches is 1 or 0, cancel the possessive
5344 flag so that the default action below, of wrapping everything inside
5345 atomic brackets, does not happen. When the minimum is greater than 1,
5346 there will be earlier copies of the group, and so we still have to wrap
5347 the whole thing. */
5348
5349 else
5350 {
5351 pcre_uchar *ketcode = code - 1 - LINK_SIZE;
5352 pcre_uchar *bracode = ketcode - GET(ketcode, 1);
5353
5354 /* Convert possessive ONCE brackets to non-capturing */
5355
5356 if ((*bracode == OP_ONCE || *bracode == OP_ONCE_NC) &&
5357 possessive_quantifier) *bracode = OP_BRA;
5358
5359 /* For non-possessive ONCE brackets, all we need to do is to
5360 set the KET. */
5361
5362 if (*bracode == OP_ONCE || *bracode == OP_ONCE_NC)
5363 *ketcode = OP_KETRMAX + repeat_type;
5364
5365 /* Handle non-ONCE brackets and possessive ONCEs (which have been
5366 converted to non-capturing above). */
5367
5368 else
5369 {
5370 /* In the compile phase, check for empty string matching. */
5371
5372 if (lengthptr == NULL)
5373 {
5374 pcre_uchar *scode = bracode;
5375 do
5376 {
5377 if (could_be_empty_branch(scode, ketcode, utf, cd))
5378 {
5379 *bracode += OP_SBRA - OP_BRA;
5380 break;
5381 }
5382 scode += GET(scode, 1);
5383 }
5384 while (*scode == OP_ALT);
5385 }
5386
5387 /* Handle possessive quantifiers. */
5388
5389 if (possessive_quantifier)
5390 {
5391 /* For COND brackets, we wrap the whole thing in a possessively
5392 repeated non-capturing bracket, because we have not invented POS
5393 versions of the COND opcodes. Because we are moving code along, we
5394 must ensure that any pending recursive references are updated. */
5395
5396 if (*bracode == OP_COND || *bracode == OP_SCOND)
5397 {
5398 int nlen = (int)(code - bracode);
5399 *code = OP_END;
5400 adjust_recurse(bracode, 1 + LINK_SIZE, utf, cd, save_hwm);
5401 memmove(bracode + 1 + LINK_SIZE, bracode, IN_UCHARS(nlen));
5402 code += 1 + LINK_SIZE;
5403 nlen += 1 + LINK_SIZE;
5404 *bracode = OP_BRAPOS;
5405 *code++ = OP_KETRPOS;
5406 PUTINC(code, 0, nlen);
5407 PUT(bracode, 1, nlen);
5408 }
5409
5410 /* For non-COND brackets, we modify the BRA code and use KETRPOS. */
5411
5412 else
5413 {
5414 *bracode += 1; /* Switch to xxxPOS opcodes */
5415 *ketcode = OP_KETRPOS;
5416 }
5417
5418 /* If the minimum is zero, mark it as possessive, then unset the
5419 possessive flag when the minimum is 0 or 1. */
5420
5421 if (brazeroptr != NULL) *brazeroptr = OP_BRAPOSZERO;
5422 if (repeat_min < 2) possessive_quantifier = FALSE;
5423 }
5424
5425 /* Non-possessive quantifier */
5426
5427 else *ketcode = OP_KETRMAX + repeat_type;
5428 }
5429 }
5430 }
5431
5432 /* If previous is OP_FAIL, it was generated by an empty class [] in
5433 JavaScript mode. The other ways in which OP_FAIL can be generated, that is
5434 by (*FAIL) or (?!) set previous to NULL, which gives a "nothing to repeat"
5435 error above. We can just ignore the repeat in JS case. */
5436
5437 else if (*previous == OP_FAIL) goto END_REPEAT;
5438
5439 /* Else there's some kind of shambles */
5440
5441 else
5442 {
5443 *errorcodeptr = ERR11;
5444 goto FAILED;
5445 }
5446
5447 /* If the character following a repeat is '+', or if certain optimization
5448 tests above succeeded, possessive_quantifier is TRUE. For some opcodes,
5449 there are special alternative opcodes for this case. For anything else, we
5450 wrap the entire repeated item inside OP_ONCE brackets. Logically, the '+'
5451 notation is just syntactic sugar, taken from Sun's Java package, but the
5452 special opcodes can optimize it.
5453
5454 Some (but not all) possessively repeated subpatterns have already been
5455 completely handled in the code just above. For them, possessive_quantifier
5456 is always FALSE at this stage.
5457
5458 Note that the repeated item starts at tempcode, not at previous, which
5459 might be the first part of a string whose (former) last char we repeated.
5460
5461 Possessifying an 'exact' quantifier has no effect, so we can ignore it. But
5462 an 'upto' may follow. We skip over an 'exact' item, and then test the
5463 length of what remains before proceeding. */
5464
5465 if (possessive_quantifier)
5466 {
5467 int len;
5468
5469 if (*tempcode == OP_TYPEEXACT)
5470 tempcode += PRIV(OP_lengths)[*tempcode] +
5471 ((tempcode[1 + IMM2_SIZE] == OP_PROP
5472 || tempcode[1 + IMM2_SIZE] == OP_NOTPROP)? 2 : 0);
5473
5474 else if (*tempcode == OP_EXACT || *tempcode == OP_NOTEXACT)
5475 {
5476 tempcode += PRIV(OP_lengths)[*tempcode];
5477 #if defined SUPPORT_UTF && !defined COMPILE_PCRE32
5478 if (utf && HAS_EXTRALEN(tempcode[-1]))
5479 tempcode += GET_EXTRALEN(tempcode[-1]);
5480 #endif
5481 }
5482
5483 len = (int)(code - tempcode);
5484 if (len > 0) switch (*tempcode)
5485 {
5486 case OP_STAR: *tempcode = OP_POSSTAR; break;
5487 case OP_PLUS: *tempcode = OP_POSPLUS; break;
5488 case OP_QUERY: *tempcode = OP_POSQUERY; break;
5489 case OP_UPTO: *tempcode = OP_POSUPTO; break;
5490
5491 case OP_STARI: *tempcode = OP_POSSTARI; break;
5492 case OP_PLUSI: *tempcode = OP_POSPLUSI; break;
5493 case OP_QUERYI: *tempcode = OP_POSQUERYI; break;
5494 case OP_UPTOI: *tempcode = OP_POSUPTOI; break;
5495
5496 case OP_NOTSTAR: *tempcode = OP_NOTPOSSTAR; break;
5497 case OP_NOTPLUS: *tempcode = OP_NOTPOSPLUS; break;
5498 case OP_NOTQUERY: *tempcode = OP_NOTPOSQUERY; break;
5499 case OP_NOTUPTO: *tempcode = OP_NOTPOSUPTO; break;
5500
5501 case OP_NOTSTARI: *tempcode = OP_NOTPOSSTARI; break;
5502 case OP_NOTPLUSI: *tempcode = OP_NOTPOSPLUSI; break;
5503 case OP_NOTQUERYI: *tempcode = OP_NOTPOSQUERYI; break;
5504 case OP_NOTUPTOI: *tempcode = OP_NOTPOSUPTOI; break;
5505
5506 case OP_TYPESTAR: *tempcode = OP_TYPEPOSSTAR; break;
5507 case OP_TYPEPLUS: *tempcode = OP_TYPEPOSPLUS; break;
5508 case OP_TYPEQUERY: *tempcode = OP_TYPEPOSQUERY; break;
5509 case OP_TYPEUPTO: *tempcode = OP_TYPEPOSUPTO; break;
5510
5511 /* Because we are moving code along, we must ensure that any
5512 pending recursive references are updated. */
5513
5514 default:
5515 *code = OP_END;
5516 adjust_recurse(tempcode, 1 + LINK_SIZE, utf, cd, save_hwm);
5517 memmove(tempcode + 1 + LINK_SIZE, tempcode, IN_UCHARS(len));
5518 code += 1 + LINK_SIZE;
5519 len += 1 + LINK_SIZE;
5520 tempcode[0] = OP_ONCE;
5521 *code++ = OP_KET;
5522 PUTINC(code, 0, len);
5523 PUT(tempcode, 1, len);
5524 break;
5525 }
5526 }
5527
5528 /* In all case we no longer have a previous item. We also set the
5529 "follows varying string" flag for subsequently encountered reqchars if
5530 it isn't already set and we have just passed a varying length item. */
5531
5532 END_REPEAT:
5533 previous = NULL;
5534 cd->req_varyopt |= reqvary;
5535 break;
5536
5537
5538 /* ===================================================================*/
5539 /* Start of nested parenthesized sub-expression, or comment or lookahead or
5540 lookbehind or option setting or condition or all the other extended
5541 parenthesis forms. */
5542
5543 case CHAR_LEFT_PARENTHESIS:
5544 newoptions = options;
5545 skipbytes = 0;
5546 bravalue = OP_CBRA;
5547 save_hwm = cd->hwm;
5548 reset_bracount = FALSE;
5549
5550 /* First deal with various "verbs" that can be introduced by '*'. */
5551
5552 ptr++;
5553 if (ptr[0] == CHAR_ASTERISK && (ptr[1] == ':'
5554 || (MAX_255(ptr[1]) && ((cd->ctypes[ptr[1]] & ctype_letter) != 0))))
5555 {
5556 int i, namelen;
5557 int arglen = 0;
5558 const char *vn = verbnames;
5559 const pcre_uchar *name = ptr + 1;
5560 const pcre_uchar *arg = NULL;
5561 previous = NULL;
5562 ptr++;
5563 while (MAX_255(*ptr) && (cd->ctypes[*ptr] & ctype_letter) != 0) ptr++;
5564 namelen = (int)(ptr - name);
5565
5566 /* It appears that Perl allows any characters whatsoever, other than
5567 a closing parenthesis, to appear in arguments, so we no longer insist on
5568 letters, digits, and underscores. */
5569
5570 if (*ptr == CHAR_COLON)
5571 {
5572 arg = ++ptr;
5573 while (*ptr != 0 && *ptr != CHAR_RIGHT_PARENTHESIS) ptr++;
5574 arglen = (int)(ptr - arg);
5575 if ((unsigned int)arglen > MAX_MARK)
5576 {
5577 *errorcodeptr = ERR75;
5578 goto FAILED;
5579 }
5580 }
5581
5582 if (*ptr != CHAR_RIGHT_PARENTHESIS)
5583 {
5584 *errorcodeptr = ERR60;
5585 goto FAILED;
5586 }
5587
5588 /* Scan the table of verb names */
5589
5590 for (i = 0; i < verbcount; i++)
5591 {
5592 if (namelen == verbs[i].len &&
5593 STRNCMP_UC_C8(name, vn, namelen) == 0)
5594 {
5595 int setverb;
5596
5597 /* Check for open captures before ACCEPT and convert it to
5598 ASSERT_ACCEPT if in an assertion. */
5599
5600 if (verbs[i].op == OP_ACCEPT)
5601 {
5602 open_capitem *oc;
5603 if (arglen != 0)
5604 {
5605 *errorcodeptr = ERR59;
5606 goto FAILED;
5607 }
5608 cd->had_accept = TRUE;
5609 for (oc = cd->open_caps; oc != NULL; oc = oc->next)
5610 {
5611 *code++ = OP_CLOSE;
5612 PUT2INC(code, 0, oc->number);
5613 }
5614 setverb = *code++ =
5615 (cd->assert_depth > 0)? OP_ASSERT_ACCEPT : OP_ACCEPT;
5616
5617 /* Do not set firstchar after *ACCEPT */
5618 if (firstchar == REQ_UNSET) firstchar = REQ_NONE;
5619 }
5620
5621 /* Handle other cases with/without an argument */
5622
5623 else if (arglen == 0)
5624 {
5625 if (verbs[i].op < 0) /* Argument is mandatory */
5626 {
5627 *errorcodeptr = ERR66;
5628 goto FAILED;
5629 }
5630 setverb = *code++ = verbs[i].op;
5631 }
5632
5633 else
5634 {
5635 if (verbs[i].op_arg < 0) /* Argument is forbidden */
5636 {
5637 *errorcodeptr = ERR59;
5638 goto FAILED;
5639 }
5640 setverb = *code++ = verbs[i].op_arg;
5641 *code++ = arglen;
5642 memcpy(code, arg, IN_UCHARS(arglen));
5643 code += arglen;
5644 *code++ = 0;
5645 }
5646
5647 switch (setverb)
5648 {
5649 case OP_THEN:
5650 case OP_THEN_ARG:
5651 cd->external_flags |= PCRE_HASTHEN;
5652 break;
5653
5654 case OP_PRUNE:
5655 case OP_PRUNE_ARG:
5656 case OP_SKIP:
5657 case OP_SKIP_ARG:
5658 cd->had_pruneorskip = TRUE;
5659 break;
5660 }
5661
5662 break; /* Found verb, exit loop */
5663 }
5664
5665 vn += verbs[i].len + 1;
5666 }
5667
5668 if (i < verbcount) continue; /* Successfully handled a verb */
5669 *errorcodeptr = ERR60; /* Verb not recognized */
5670 goto FAILED;
5671 }
5672
5673 /* Deal with the extended parentheses; all are introduced by '?', and the
5674 appearance of any of them means that this is not a capturing group. */
5675
5676 else if (*ptr == CHAR_QUESTION_MARK)
5677 {
5678 int i, set, unset, namelen;
5679 int *optset;
5680 const pcre_uchar *name;
5681 pcre_uchar *slot;
5682
5683 switch (*(++ptr))
5684 {
5685 case CHAR_NUMBER_SIGN: /* Comment; skip to ket */
5686 ptr++;
5687 while (*ptr != 0 && *ptr != CHAR_RIGHT_PARENTHESIS) ptr++;
5688 if (*ptr == 0)
5689 {
5690 *errorcodeptr = ERR18;
5691 goto FAILED;
5692 }
5693 continue;
5694
5695
5696 /* ------------------------------------------------------------ */
5697 case CHAR_VERTICAL_LINE: /* Reset capture count for each branch */
5698 reset_bracount = TRUE;
5699 /* Fall through */
5700
5701 /* ------------------------------------------------------------ */
5702 case CHAR_COLON: /* Non-capturing bracket */
5703 bravalue = OP_BRA;
5704 ptr++;
5705 break;
5706
5707
5708 /* ------------------------------------------------------------ */
5709 case CHAR_LEFT_PARENTHESIS:
5710 bravalue = OP_COND; /* Conditional group */
5711
5712 /* A condition can be an assertion, a number (referring to a numbered
5713 group), a name (referring to a named group), or 'R', referring to
5714 recursion. R<digits> and R&name are also permitted for recursion tests.
5715
5716 There are several syntaxes for testing a named group: (?(name)) is used
5717 by Python; Perl 5.10 onwards uses (?(<name>) or (?('name')).
5718
5719 There are two unfortunate ambiguities, caused by history. (a) 'R' can
5720 be the recursive thing or the name 'R' (and similarly for 'R' followed
5721 by digits), and (b) a number could be a name that consists of digits.
5722 In both cases, we look for a name first; if not found, we try the other
5723 cases. */
5724
5725 /* For conditions that are assertions, check the syntax, and then exit
5726 the switch. This will take control down to where bracketed groups,
5727 including assertions, are processed. */
5728
5729 if (ptr[1] == CHAR_QUESTION_MARK && (ptr[2] == CHAR_EQUALS_SIGN ||
5730 ptr[2] == CHAR_EXCLAMATION_MARK || ptr[2] == CHAR_LESS_THAN_SIGN))
5731 break;
5732
5733 /* Most other conditions use OP_CREF (a couple change to OP_RREF
5734 below), and all need to skip 1+IMM2_SIZE bytes at the start of the group. */
5735
5736 code[1+LINK_SIZE] = OP_CREF;
5737 skipbytes = 1+IMM2_SIZE;
5738 refsign = -1;
5739
5740 /* Check for a test for recursion in a named group. */
5741
5742 if (ptr[1] == CHAR_R && ptr[2] == CHAR_AMPERSAND)
5743 {
5744 terminator = -1;
5745 ptr += 2;
5746 code[1+LINK_SIZE] = OP_RREF; /* Change the type of test */
5747 }
5748
5749 /* Check for a test for a named group's having been set, using the Perl
5750 syntax (?(<name>) or (?('name') */
5751
5752 else if (ptr[1] == CHAR_LESS_THAN_SIGN)
5753 {
5754 terminator = CHAR_GREATER_THAN_SIGN;
5755 ptr++;
5756 }
5757 else if (ptr[1] == CHAR_APOSTROPHE)
5758 {
5759 terminator = CHAR_APOSTROPHE;
5760 ptr++;
5761 }
5762 else
5763 {
5764 terminator = 0;
5765 if (ptr[1] == CHAR_MINUS || ptr[1] == CHAR_PLUS) refsign = *(++ptr);
5766 }
5767
5768 /* We now expect to read a name; any thing else is an error */
5769
5770 if (!MAX_255(ptr[1]) || (cd->ctypes[ptr[1]] & ctype_word) == 0)
5771 {
5772 ptr += 1; /* To get the right offset */
5773 *errorcodeptr = ERR28;
5774 goto FAILED;
5775 }
5776
5777 /* Read the name, but also get it as a number if it's all digits */
5778
5779 recno = 0;
5780 name = ++ptr;
5781 while (MAX_255(*ptr) && (cd->ctypes[*ptr] & ctype_word) != 0)
5782 {
5783 if (recno >= 0)
5784 recno = (IS_DIGIT(*ptr))? recno * 10 + *ptr - CHAR_0 : -1;
5785 ptr++;
5786 }
5787 namelen = (int)(ptr - name);
5788
5789 if ((terminator > 0 && *ptr++ != terminator) ||
5790 *ptr++ != CHAR_RIGHT_PARENTHESIS)
5791 {
5792 ptr--; /* Error offset */
5793 *errorcodeptr = ERR26;
5794 goto FAILED;
5795 }
5796
5797 /* Do no further checking in the pre-compile phase. */
5798
5799 if (lengthptr != NULL) break;
5800
5801 /* In the real compile we do the work of looking for the actual
5802 reference. If the string started with "+" or "-" we require the rest to
5803 be digits, in which case recno will be set. */
5804
5805 if (refsign > 0)
5806 {
5807 if (recno <= 0)
5808 {
5809 *errorcodeptr = ERR58;
5810 goto FAILED;
5811 }
5812 recno = (refsign == CHAR_MINUS)?
5813 cd->bracount - recno + 1 : recno +cd->bracount;
5814 if (recno <= 0 || recno > cd->final_bracount)
5815 {
5816 *errorcodeptr = ERR15;
5817 goto FAILED;
5818 }
5819 PUT2(code, 2+LINK_SIZE, recno);
5820 break;
5821 }
5822
5823 /* Otherwise (did not start with "+" or "-"), start by looking for the
5824 name. If we find a name, add one to the opcode to change OP_CREF or
5825 OP_RREF into OP_NCREF or OP_NRREF. These behave exactly the same,
5826 except they record that the reference was originally to a name. The
5827 information is used to check duplicate names. */
5828
5829 slot = cd->name_table;
5830 for (i = 0; i < cd->names_found; i++)
5831 {
5832 if (STRNCMP_UC_UC(name, slot+IMM2_SIZE, namelen) == 0) break;
5833 slot += cd->name_entry_size;
5834 }
5835
5836 /* Found a previous named subpattern */
5837
5838 if (i < cd->names_found)
5839 {
5840 recno = GET2(slot, 0);
5841 PUT2(code, 2+LINK_SIZE, recno);
5842 code[1+LINK_SIZE]++;
5843 }
5844
5845 /* Search the pattern for a forward reference */
5846
5847 else if ((i = find_parens(cd, name, namelen,
5848 (options & PCRE_EXTENDED) != 0, utf)) > 0)
5849 {
5850 PUT2(code, 2+LINK_SIZE, i);
5851 code[1+LINK_SIZE]++;
5852 }
5853
5854 /* If terminator == 0 it means that the name followed directly after
5855 the opening parenthesis [e.g. (?(abc)...] and in this case there are
5856 some further alternatives to try. For the cases where terminator != 0
5857 [things like (?(<name>... or (?('name')... or (?(R&name)... ] we have
5858 now checked all the possibilities, so give an error. */
5859
5860 else if (terminator != 0)
5861 {
5862 *errorcodeptr = ERR15;
5863 goto FAILED;
5864 }
5865
5866 /* Check for (?(R) for recursion. Allow digits after R to specify a
5867 specific group number. */
5868
5869 else if (*name == CHAR_R)
5870 {
5871 recno = 0;
5872 for (i = 1; i < namelen; i++)
5873 {
5874 if (!IS_DIGIT(name[i]))
5875 {
5876 *errorcodeptr = ERR15;
5877 goto FAILED;
5878 }
5879 recno = recno * 10 + name[i] - CHAR_0;
5880 }
5881 if (recno == 0) recno = RREF_ANY;
5882 code[1+LINK_SIZE] = OP_RREF; /* Change test type */
5883 PUT2(code, 2+LINK_SIZE, recno);
5884 }
5885
5886 /* Similarly, check for the (?(DEFINE) "condition", which is always
5887 false. */
5888
5889 else if (namelen == 6 && STRNCMP_UC_C8(name, STRING_DEFINE, 6) == 0)
5890 {
5891 code[1+LINK_SIZE] = OP_DEF;
5892 skipbytes = 1;
5893 }
5894
5895 /* Check for the "name" actually being a subpattern number. We are
5896 in the second pass here, so final_bracount is set. */
5897
5898 else if (recno > 0 && recno <= cd->final_bracount)
5899 {
5900 PUT2(code, 2+LINK_SIZE, recno);
5901 }
5902
5903 /* Either an unidentified subpattern, or a reference to (?(0) */
5904
5905 else
5906 {
5907 *errorcodeptr = (recno == 0)? ERR35: ERR15;
5908 goto FAILED;
5909 }
5910 break;
5911
5912
5913 /* ------------------------------------------------------------ */
5914 case CHAR_EQUALS_SIGN: /* Positive lookahead */
5915 bravalue = OP_ASSERT;
5916 cd->assert_depth += 1;
5917 ptr++;
5918 break;
5919
5920
5921 /* ------------------------------------------------------------ */
5922 case CHAR_EXCLAMATION_MARK: /* Negative lookahead */
5923 ptr++;
5924 if (*ptr == CHAR_RIGHT_PARENTHESIS) /* Optimize (?!) */
5925 {
5926 *code++ = OP_FAIL;
5927 previous = NULL;
5928 continue;
5929 }
5930 bravalue = OP_ASSERT_NOT;
5931 cd->assert_depth += 1;
5932 break;
5933
5934
5935 /* ------------------------------------------------------------ */
5936 case CHAR_LESS_THAN_SIGN: /* Lookbehind or named define */
5937 switch (ptr[1])
5938 {
5939 case CHAR_EQUALS_SIGN: /* Positive lookbehind */
5940 bravalue = OP_ASSERTBACK;
5941 cd->assert_depth += 1;
5942 ptr += 2;
5943 break;
5944
5945 case CHAR_EXCLAMATION_MARK: /* Negative lookbehind */
5946 bravalue = OP_ASSERTBACK_NOT;
5947 cd->assert_depth += 1;
5948 ptr += 2;
5949 break;
5950
5951 default: /* Could be name define, else bad */
5952 if (MAX_255(ptr[1]) && (cd->ctypes[ptr[1]] & ctype_word) != 0)
5953 goto DEFINE_NAME;
5954 ptr++; /* Correct offset for error */
5955 *errorcodeptr = ERR24;
5956 goto FAILED;
5957 }
5958 break;
5959
5960
5961 /* ------------------------------------------------------------ */
5962 case CHAR_GREATER_THAN_SIGN: /* One-time brackets */
5963 bravalue = OP_ONCE;
5964 ptr++;
5965 break;
5966
5967
5968 /* ------------------------------------------------------------ */
5969 case CHAR_C: /* Callout - may be followed by digits; */
5970 previous_callout = code; /* Save for later completion */
5971 after_manual_callout = 1; /* Skip one item before completing */
5972 *code++ = OP_CALLOUT;
5973 {
5974 int n = 0;
5975 ptr++;
5976 while(IS_DIGIT(*ptr))
5977 n = n * 10 + *ptr++ - CHAR_0;
5978 if (*ptr != CHAR_RIGHT_PARENTHESIS)
5979 {
5980 *errorcodeptr = ERR39;
5981 goto FAILED;
5982 }
5983 if (n > 255)
5984 {
5985 *errorcodeptr = ERR38;
5986 goto FAILED;
5987 }
5988 *code++ = n;
5989 PUT(code, 0, (int)(ptr - cd->start_pattern + 1)); /* Pattern offset */
5990 PUT(code, LINK_SIZE, 0); /* Default length */
5991 code += 2 * LINK_SIZE;
5992 }
5993 previous = NULL;
5994 continue;
5995
5996
5997 /* ------------------------------------------------------------ */
5998 case CHAR_P: /* Python-style named subpattern handling */
5999 if (*(++ptr) == CHAR_EQUALS_SIGN ||
6000 *ptr == CHAR_GREATER_THAN_SIGN) /* Reference or recursion */
6001 {
6002 is_recurse = *ptr == CHAR_GREATER_THAN_SIGN;
6003 terminator = CHAR_RIGHT_PARENTHESIS;
6004 goto NAMED_REF_OR_RECURSE;
6005 }
6006 else if (*ptr != CHAR_LESS_THAN_SIGN) /* Test for Python-style defn */
6007 {
6008 *errorcodeptr = ERR41;
6009 goto FAILED;
6010 }
6011 /* Fall through to handle (?P< as (?< is handled */
6012
6013
6014 /* ------------------------------------------------------------ */
6015 DEFINE_NAME: /* Come here from (?< handling */
6016 case CHAR_APOSTROPHE:
6017 {
6018 terminator = (*ptr == CHAR_LESS_THAN_SIGN)?
6019 CHAR_GREATER_THAN_SIGN : CHAR_APOSTROPHE;
6020 name = ++ptr;
6021
6022 while (MAX_255(*ptr) && (cd->ctypes[*ptr] & ctype_word) != 0) ptr++;
6023 namelen = (int)(ptr - name);
6024
6025 /* In the pre-compile phase, just do a syntax check. */
6026
6027 if (lengthptr != NULL)
6028 {
6029 if (*ptr != terminator)
6030 {
6031 *errorcodeptr = ERR42;
6032 goto FAILED;
6033 }
6034 if (cd->names_found >= MAX_NAME_COUNT)
6035 {
6036 *errorcodeptr = ERR49;
6037 goto FAILED;
6038 }
6039 if (namelen + IMM2_SIZE + 1 > cd->name_entry_size)
6040 {
6041 cd->name_entry_size = namelen + IMM2_SIZE + 1;
6042 if (namelen > MAX_NAME_SIZE)
6043 {
6044 *errorcodeptr = ERR48;
6045 goto FAILED;
6046 }
6047 }
6048 }
6049
6050 /* In the real compile, create the entry in the table, maintaining
6051 alphabetical order. Duplicate names for different numbers are
6052 permitted only if PCRE_DUPNAMES is set. Duplicate names for the same
6053 number are always OK. (An existing number can be re-used if (?|
6054 appears in the pattern.) In either event, a duplicate name results in
6055 a duplicate entry in the table, even if the number is the same. This
6056 is because the number of names, and hence the table size, is computed
6057 in the pre-compile, and it affects various numbers and pointers which
6058 would all have to be modified, and the compiled code moved down, if
6059 duplicates with the same number were omitted from the table. This
6060 doesn't seem worth the hassle. However, *different* names for the
6061 same number are not permitted. */
6062
6063 else
6064 {
6065 BOOL dupname = FALSE;
6066 slot = cd->name_table;
6067
6068 for (i = 0; i < cd->names_found; i++)
6069 {
6070 int crc = memcmp(name, slot+IMM2_SIZE, IN_UCHARS(namelen));
6071 if (crc == 0)
6072 {
6073 if (slot[IMM2_SIZE+namelen] == 0)
6074 {
6075 if (GET2(slot, 0) != cd->bracount + 1 &&
6076 (options & PCRE_DUPNAMES) == 0)
6077 {
6078 *errorcodeptr = ERR43;
6079 goto FAILED;
6080 }
6081 else dupname = TRUE;
6082 }
6083 else crc = -1; /* Current name is a substring */
6084 }
6085
6086 /* Make space in the table and break the loop for an earlier
6087 name. For a duplicate or later name, carry on. We do this for
6088 duplicates so that in the simple case (when ?(| is not used) they
6089 are in order of their numbers. */
6090
6091 if (crc < 0)
6092 {
6093 memmove(slot + cd->name_entry_size, slot,
6094 IN_UCHARS((cd->names_found - i) * cd->name_entry_size));
6095 break;
6096 }
6097
6098 /* Continue the loop for a later or duplicate name */
6099
6100 slot += cd->name_entry_size;
6101 }
6102
6103 /* For non-duplicate names, check for a duplicate number before
6104 adding the new name. */
6105
6106 if (!dupname)
6107 {
6108 pcre_uchar *cslot = cd->name_table;
6109 for (i = 0; i < cd->names_found; i++)
6110 {
6111 if (cslot != slot)
6112 {
6113 if (GET2(cslot, 0) == cd->bracount + 1)
6114 {
6115 *errorcodeptr = ERR65;
6116 goto FAILED;
6117 }
6118 }
6119 else i--;
6120 cslot += cd->name_entry_size;
6121 }
6122 }
6123
6124 PUT2(slot, 0, cd->bracount + 1);
6125 memcpy(slot + IMM2_SIZE, name, IN_UCHARS(namelen));
6126 slot[IMM2_SIZE + namelen] = 0;
6127 }
6128 }
6129
6130 /* In both pre-compile and compile, count the number of names we've
6131 encountered. */
6132
6133 cd->names_found++;
6134 ptr++; /* Move past > or ' */
6135 goto NUMBERED_GROUP;
6136
6137
6138 /* ------------------------------------------------------------ */
6139 case CHAR_AMPERSAND: /* Perl recursion/subroutine syntax */
6140 terminator = CHAR_RIGHT_PARENTHESIS;
6141 is_recurse = TRUE;
6142 /* Fall through */
6143
6144 /* We come here from the Python syntax above that handles both
6145 references (?P=name) and recursion (?P>name), as well as falling
6146 through from the Perl recursion syntax (?&name). We also come here from
6147 the Perl \k<name> or \k'name' back reference syntax and the \k{name}
6148 .NET syntax, and the Oniguruma \g<...> and \g'...' subroutine syntax. */
6149
6150 NAMED_REF_OR_RECURSE:
6151 name = ++ptr;
6152 while (MAX_255(*ptr) && (cd->ctypes[*ptr] & ctype_word) != 0) ptr++;
6153 namelen = (int)(ptr - name);
6154
6155 /* In the pre-compile phase, do a syntax check. We used to just set
6156 a dummy reference number, because it was not used in the first pass.
6157 However, with the change of recursive back references to be atomic,
6158 we have to look for the number so that this state can be identified, as
6159 otherwise the incorrect length is computed. If it's not a backwards
6160 reference, the dummy number will do. */
6161
6162 if (lengthptr != NULL)
6163 {
6164 const pcre_uchar *temp;
6165
6166 if (namelen == 0)
6167 {
6168 *errorcodeptr = ERR62;
6169 goto FAILED;
6170 }
6171 if (*ptr != terminator)
6172 {
6173 *errorcodeptr = ERR42;
6174 goto FAILED;
6175 }
6176 if (namelen > MAX_NAME_SIZE)
6177 {
6178 *errorcodeptr = ERR48;
6179 goto FAILED;
6180 }
6181
6182 /* The name table does not exist in the first pass, so we cannot
6183 do a simple search as in the code below. Instead, we have to scan the
6184 pattern to find the number. It is important that we scan it only as
6185 far as we have got because the syntax of named subpatterns has not
6186 been checked for the rest of the pattern, and find_parens() assumes
6187 correct syntax. In any case, it's a waste of resources to scan
6188 further. We stop the scan at the current point by temporarily
6189 adjusting the value of cd->endpattern. */
6190
6191 temp = cd->end_pattern;
6192 cd->end_pattern = ptr;
6193 recno = find_parens(cd, name, namelen,
6194 (options & PCRE_EXTENDED) != 0, utf);
6195 cd->end_pattern = temp;
6196 if (recno < 0) recno = 0; /* Forward ref; set dummy number */
6197 }
6198
6199 /* In the real compile, seek the name in the table. We check the name
6200 first, and then check that we have reached the end of the name in the
6201 table. That way, if the name that is longer than any in the table,
6202 the comparison will fail without reading beyond the table entry. */
6203
6204 else
6205 {
6206 slot = cd->name_table;
6207 for (i = 0; i < cd->names_found; i++)
6208 {
6209 if (STRNCMP_UC_UC(name, slot+IMM2_SIZE, namelen) == 0 &&
6210 slot[IMM2_SIZE+namelen] == 0)
6211 break;
6212 slot += cd->name_entry_size;
6213 }
6214
6215 if (i < cd->names_found) /* Back reference */
6216 {
6217 recno = GET2(slot, 0);
6218 }
6219 else if ((recno = /* Forward back reference */
6220 find_parens(cd, name, namelen,
6221 (options & PCRE_EXTENDED) != 0, utf)) <= 0)
6222 {
6223 *errorcodeptr = ERR15;
6224 goto FAILED;
6225 }
6226 }
6227
6228 /* In both phases, we can now go to the code than handles numerical
6229 recursion or backreferences. */
6230
6231 if (is_recurse) goto HANDLE_RECURSION;
6232 else goto HANDLE_REFERENCE;
6233
6234
6235 /* ------------------------------------------------------------ */
6236 case CHAR_R: /* Recursion */
6237 ptr++; /* Same as (?0) */
6238 /* Fall through */
6239
6240
6241 /* ------------------------------------------------------------ */
6242 case CHAR_MINUS: case CHAR_PLUS: /* Recursion or subroutine */
6243 case CHAR_0: case CHAR_1: case CHAR_2: case CHAR_3: case CHAR_4:
6244 case CHAR_5: case CHAR_6: case CHAR_7: case CHAR_8: case CHAR_9:
6245 {
6246 const pcre_uchar *called;
6247 terminator = CHAR_RIGHT_PARENTHESIS;
6248
6249 /* Come here from the \g<...> and \g'...' code (Oniguruma
6250 compatibility). However, the syntax has been checked to ensure that
6251 the ... are a (signed) number, so that neither ERR63 nor ERR29 will
6252 be called on this path, nor with the jump to OTHER_CHAR_AFTER_QUERY
6253 ever be taken. */
6254
6255 HANDLE_NUMERICAL_RECURSION:
6256
6257 if ((refsign = *ptr) == CHAR_PLUS)
6258 {
6259 ptr++;
6260 if (!IS_DIGIT(*ptr))
6261 {
6262 *errorcodeptr = ERR63;
6263 goto FAILED;
6264 }
6265 }
6266 else if (refsign == CHAR_MINUS)
6267 {
6268 if (!IS_DIGIT(ptr[1]))
6269 goto OTHER_CHAR_AFTER_QUERY;
6270 ptr++;
6271 }
6272
6273 recno = 0;
6274 while(IS_DIGIT(*ptr))
6275 recno = recno * 10 + *ptr++ - CHAR_0;
6276
6277 if (*ptr != terminator)
6278 {
6279 *errorcodeptr = ERR29;
6280 goto FAILED;
6281 }
6282
6283 if (refsign == CHAR_MINUS)
6284 {
6285 if (recno == 0)
6286 {
6287 *errorcodeptr = ERR58;
6288 goto FAILED;
6289 }
6290 recno = cd->bracount - recno + 1;
6291 if (recno <= 0)
6292 {
6293 *errorcodeptr = ERR15;
6294 goto FAILED;
6295 }
6296 }
6297 else if (refsign == CHAR_PLUS)
6298 {
6299 if (recno == 0)
6300 {
6301 *errorcodeptr = ERR58;
6302 goto FAILED;
6303 }
6304 recno += cd->bracount;
6305 }
6306
6307 /* Come here from code above that handles a named recursion */
6308
6309 HANDLE_RECURSION:
6310
6311 previous = code;
6312 called = cd->start_code;
6313
6314 /* When we are actually compiling, find the bracket that is being
6315 referenced. Temporarily end the regex in case it doesn't exist before
6316 this point. If we end up with a forward reference, first check that
6317 the bracket does occur later so we can give the error (and position)
6318 now. Then remember this forward reference in the workspace so it can
6319 be filled in at the end. */
6320
6321 if (lengthptr == NULL)
6322 {
6323 *code = OP_END;
6324 if (recno != 0)
6325 called = PRIV(find_bracket)(cd->start_code, utf, recno);
6326
6327 /* Forward reference */
6328
6329 if (called == NULL)
6330 {
6331 if (find_parens(cd, NULL, recno,
6332 (options & PCRE_EXTENDED) != 0, utf) < 0)
6333 {
6334 *errorcodeptr = ERR15;
6335 goto FAILED;
6336 }
6337
6338 /* Fudge the value of "called" so that when it is inserted as an
6339 offset below, what it actually inserted is the reference number
6340 of the group. Then remember the forward reference. */
6341
6342 called = cd->start_code + recno;
6343 if (cd->hwm >= cd->start_workspace + cd->workspace_size -
6344 WORK_SIZE_SAFETY_MARGIN)
6345 {
6346 *errorcodeptr = expand_workspace(cd);
6347 if (*errorcodeptr != 0) goto FAILED;
6348 }
6349 PUTINC(cd->hwm, 0, (int)(code + 1 - cd->start_code));
6350 }
6351
6352 /* If not a forward reference, and the subpattern is still open,
6353 this is a recursive call. We check to see if this is a left
6354 recursion that could loop for ever, and diagnose that case. We
6355 must not, however, do this check if we are in a conditional
6356 subpattern because the condition might be testing for recursion in
6357 a pattern such as /(?(R)a+|(?R)b)/, which is perfectly valid.
6358 Forever loops are also detected at runtime, so those that occur in
6359 conditional subpatterns will be picked up then. */
6360
6361 else if (GET(called, 1) == 0 && cond_depth <= 0 &&
6362 could_be_empty(called, code, bcptr, utf, cd))
6363 {
6364 *errorcodeptr = ERR40;
6365 goto FAILED;
6366 }
6367 }
6368
6369 /* Insert the recursion/subroutine item. It does not have a set first
6370 character (relevant if it is repeated, because it will then be
6371 wrapped with ONCE brackets). */
6372
6373 *code = OP_RECURSE;
6374 PUT(code, 1, (int)(called - cd->start_code));
6375 code += 1 + LINK_SIZE;
6376 groupsetfirstchar = FALSE;
6377 }
6378
6379 /* Can't determine a first byte now */
6380
6381 if (firstchar == REQ_UNSET) firstchar = REQ_NONE;
6382 continue;
6383
6384
6385 /* ------------------------------------------------------------ */
6386 default: /* Other characters: check option setting */
6387 OTHER_CHAR_AFTER_QUERY:
6388 set = unset = 0;
6389 optset = &set;
6390
6391 while (*ptr != CHAR_RIGHT_PARENTHESIS && *ptr != CHAR_COLON)
6392 {
6393 switch (*ptr++)
6394 {
6395 case CHAR_MINUS: optset = &unset; break;
6396
6397 case CHAR_J: /* Record that it changed in the external options */
6398 *optset |= PCRE_DUPNAMES;
6399 cd->external_flags |= PCRE_JCHANGED;
6400 break;
6401
6402 case CHAR_i: *optset |= PCRE_CASELESS; break;
6403 case CHAR_m: *optset |= PCRE_MULTILINE; break;
6404 case CHAR_s: *optset |= PCRE_DOTALL; break;
6405 case CHAR_x: *optset |= PCRE_EXTENDED; break;
6406 case CHAR_U: *optset |= PCRE_UNGREEDY; break;
6407 case CHAR_X: *optset |= PCRE_EXTRA; break;
6408
6409 default: *errorcodeptr = ERR12;
6410 ptr--; /* Correct the offset */
6411 goto FAILED;
6412 }
6413 }
6414
6415 /* Set up the changed option bits, but don't change anything yet. */
6416
6417 newoptions = (options | set) & (~unset);
6418
6419 /* If the options ended with ')' this is not the start of a nested
6420 group with option changes, so the options change at this level. If this
6421 item is right at the start of the pattern, the options can be
6422 abstracted and made external in the pre-compile phase, and ignored in
6423 the compile phase. This can be helpful when matching -- for instance in
6424 caseless checking of required bytes.
6425
6426 If the code pointer is not (cd->start_code + 1 + LINK_SIZE), we are
6427 definitely *not* at the start of the pattern because something has been
6428 compiled. In the pre-compile phase, however, the code pointer can have
6429 that value after the start, because it gets reset as code is discarded
6430 during the pre-compile. However, this can happen only at top level - if
6431 we are within parentheses, the starting BRA will still be present. At
6432 any parenthesis level, the length value can be used to test if anything
6433 has been compiled at that level. Thus, a test for both these conditions
6434 is necessary to ensure we correctly detect the start of the pattern in
6435 both phases.
6436
6437 If we are not at the pattern start, reset the greedy defaults and the
6438 case value for firstchar and reqchar. */
6439
6440 if (*ptr == CHAR_RIGHT_PARENTHESIS)
6441 {
6442 if (code == cd->start_code + 1 + LINK_SIZE &&
6443 (lengthptr == NULL || *lengthptr == 2 + 2*LINK_SIZE))
6444 {
6445 cd->external_options = newoptions;
6446 }
6447 else
6448 {
6449 greedy_default = ((newoptions & PCRE_UNGREEDY) != 0);
6450 greedy_non_default = greedy_default ^ 1;
6451 req_caseopt = ((newoptions & PCRE_CASELESS) != 0)? REQ_CASELESS:0;
6452 }
6453
6454 /* Change options at this level, and pass them back for use
6455 in subsequent branches. */
6456
6457 *optionsptr = options = newoptions;
6458 previous = NULL; /* This item can't be repeated */
6459 continue; /* It is complete */
6460 }
6461
6462 /* If the options ended with ':' we are heading into a nested group
6463 with possible change of options. Such groups are non-capturing and are
6464 not assertions of any kind. All we need to do is skip over the ':';
6465 the newoptions value is handled below. */
6466
6467 bravalue = OP_BRA;
6468 ptr++;
6469 } /* End of switch for character following (? */
6470 } /* End of (? handling */
6471
6472 /* Opening parenthesis not followed by '*' or '?'. If PCRE_NO_AUTO_CAPTURE
6473 is set, all unadorned brackets become non-capturing and behave like (?:...)
6474 brackets. */
6475
6476 else if ((options & PCRE_NO_AUTO_CAPTURE) != 0)
6477 {
6478 bravalue = OP_BRA;
6479 }
6480
6481 /* Else we have a capturing group. */
6482
6483 else
6484 {
6485 NUMBERED_GROUP:
6486 cd->bracount += 1;
6487 PUT2(code, 1+LINK_SIZE, cd->bracount);
6488 skipbytes = IMM2_SIZE;
6489 }
6490
6491 /* Process nested bracketed regex. Assertions used not to be repeatable,
6492 but this was changed for Perl compatibility, so all kinds can now be
6493 repeated. We copy code into a non-register variable (tempcode) in order to
6494 be able to pass its address because some compilers complain otherwise. */
6495
6496 previous = code; /* For handling repetition */
6497 *code = bravalue;
6498 tempcode = code;
6499 tempreqvary = cd->req_varyopt; /* Save value before bracket */
6500 tempbracount = cd->bracount; /* Save value before bracket */
6501 length_prevgroup = 0; /* Initialize for pre-compile phase */
6502
6503 if (!compile_regex(
6504 newoptions, /* The complete new option state */
6505 &tempcode, /* Where to put code (updated) */
6506 &ptr, /* Input pointer (updated) */
6507 errorcodeptr, /* Where to put an error message */
6508 (bravalue == OP_ASSERTBACK ||
6509 bravalue == OP_ASSERTBACK_NOT), /* TRUE if back assert */
6510 reset_bracount, /* True if (?| group */
6511 skipbytes, /* Skip over bracket number */
6512 cond_depth +
6513 ((bravalue == OP_COND)?1:0), /* Depth of condition subpatterns */
6514 &subfirstchar, /* For possible first char */
6515 &subreqchar, /* For possible last char */
6516 bcptr, /* Current branch chain */
6517 cd, /* Tables block */
6518 (lengthptr == NULL)? NULL : /* Actual compile phase */
6519 &length_prevgroup /* Pre-compile phase */
6520 ))
6521 goto FAILED;
6522
6523 /* If this was an atomic group and there are no capturing groups within it,
6524 generate OP_ONCE_NC instead of OP_ONCE. */
6525
6526 if (bravalue == OP_ONCE && cd->bracount <= tempbracount)
6527 *code = OP_ONCE_NC;
6528
6529 if (bravalue >= OP_ASSERT && bravalue <= OP_ASSERTBACK_NOT)
6530 cd->assert_depth -= 1;
6531
6532 /* At the end of compiling, code is still pointing to the start of the
6533 group, while tempcode has been updated to point past the end of the group.
6534 The pattern pointer (ptr) is on the bracket.
6535
6536 If this is a conditional bracket, check that there are no more than
6537 two branches in the group, or just one if it's a DEFINE group. We do this
6538 in the real compile phase, not in the pre-pass, where the whole group may
6539 not be available. */
6540
6541 if (bravalue == OP_COND && lengthptr == NULL)
6542 {
6543 pcre_uchar *tc = code;
6544 int condcount = 0;
6545
6546 do {
6547 condcount++;
6548 tc += GET(tc,1);
6549 }
6550 while (*tc != OP_KET);
6551
6552 /* A DEFINE group is never obeyed inline (the "condition" is always
6553 false). It must have only one branch. */
6554
6555 if (code[LINK_SIZE+1] == OP_DEF)
6556 {
6557 if (condcount > 1)
6558 {
6559 *errorcodeptr = ERR54;
6560 goto FAILED;
6561 }
6562 bravalue = OP_DEF; /* Just a flag to suppress char handling below */
6563 }
6564
6565 /* A "normal" conditional group. If there is just one branch, we must not
6566 make use of its firstchar or reqchar, because this is equivalent to an
6567 empty second branch. */
6568
6569 else
6570 {
6571 if (condcount > 2)
6572 {
6573 *errorcodeptr = ERR27;
6574 goto FAILED;
6575 }
6576 if (condcount == 1) subfirstchar = subreqchar = REQ_NONE;
6577 }
6578 }
6579
6580 /* Error if hit end of pattern */
6581
6582 if (*ptr != CHAR_RIGHT_PARENTHESIS)
6583 {
6584 *errorcodeptr = ERR14;
6585 goto FAILED;
6586 }
6587
6588 /* In the pre-compile phase, update the length by the length of the group,
6589 less the brackets at either end. Then reduce the compiled code to just a
6590 set of non-capturing brackets so that it doesn't use much memory if it is
6591 duplicated by a quantifier.*/
6592
6593 if (lengthptr != NULL)
6594 {
6595 if (OFLOW_MAX - *lengthptr < length_prevgroup - 2 - 2*LINK_SIZE)
6596 {
6597 *errorcodeptr = ERR20;
6598 goto FAILED;
6599 }
6600 *lengthptr += length_prevgroup - 2 - 2*LINK_SIZE;
6601 code++; /* This already contains bravalue */
6602 PUTINC(code, 0, 1 + LINK_SIZE);
6603 *code++ = OP_KET;
6604 PUTINC(code, 0, 1 + LINK_SIZE);
6605 break; /* No need to waste time with special character handling */
6606 }
6607
6608 /* Otherwise update the main code pointer to the end of the group. */
6609
6610 code = tempcode;
6611
6612 /* For a DEFINE group, required and first character settings are not
6613 relevant. */
6614
6615 if (bravalue == OP_DEF) break;
6616
6617 /* Handle updating of the required and first characters for other types of
6618 group. Update for normal brackets of all kinds, and conditions with two
6619 branches (see code above). If the bracket is followed by a quantifier with
6620 zero repeat, we have to back off. Hence the definition of zeroreqchar and
6621 zerofirstchar outside the main loop so that they can be accessed for the
6622 back off. */
6623
6624 zeroreqchar = reqchar;
6625 zerofirstchar = firstchar;
6626 groupsetfirstchar = FALSE;
6627
6628 if (bravalue >= OP_ONCE)
6629 {
6630 /* If we have not yet set a firstchar in this branch, take it from the
6631 subpattern, remembering that it was set here so that a repeat of more
6632 than one can replicate it as reqchar if necessary. If the subpattern has
6633 no firstchar, set "none" for the whole branch. In both cases, a zero
6634 repeat forces firstchar to "none". */
6635
6636 if (firstchar == REQ_UNSET)
6637 {
6638 if (subfirstchar >= 0)
6639 {
6640