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

Contents of /code/trunk/pcre_compile.c

Parent Directory Parent Directory | Revision Log Revision Log


Revision 532 - (show annotations) (download)
Tue Jun 1 16:21:42 2010 UTC (3 years, 10 months ago) by ph10
File MIME type: text/plain
File size: 232835 byte(s)
Extend auto-possessify to handle some Unicode properties.

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) 1997-2010 University of Cambridge
10
11 -----------------------------------------------------------------------------
12 Redistribution and use in source and binary forms, with or without
13 modification, are permitted provided that the following conditions are met:
14
15 * Redistributions of source code must retain the above copyright notice,
16 this list of conditions and the following disclaimer.
17
18 * Redistributions in binary form must reproduce the above copyright
19 notice, this list of conditions and the following disclaimer in the
20 documentation and/or other materials provided with the distribution.
21
22 * Neither the name of the University of Cambridge nor the names of its
23 contributors may be used to endorse or promote products derived from
24 this software without specific prior written permission.
25
26 THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
27 AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28 IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29 ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE
30 LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
31 CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
32 SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
33 INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
34 CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
35 ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
36 POSSIBILITY OF SUCH DAMAGE.
37 -----------------------------------------------------------------------------
38 */
39
40
41 /* This module contains the external function pcre_compile(), along with
42 supporting internal functions that are not used by other modules. */
43
44
45 #ifdef HAVE_CONFIG_H
46 #include "config.h"
47 #endif
48
49 #define NLBLOCK cd /* Block containing newline information */
50 #define PSSTART start_pattern /* Field containing processed string start */
51 #define PSEND end_pattern /* Field containing processed string end */
52
53 #include "pcre_internal.h"
54
55
56 /* When PCRE_DEBUG is defined, we need the pcre_printint() function, which is
57 also used by pcretest. PCRE_DEBUG is not defined when building a production
58 library. */
59
60 #ifdef PCRE_DEBUG
61 #include "pcre_printint.src"
62 #endif
63
64
65 /* Macro for setting individual bits in class bitmaps. */
66
67 #define SETBIT(a,b) a[b/8] |= (1 << (b%8))
68
69 /* Maximum length value to check against when making sure that the integer that
70 holds the compiled pattern length does not overflow. We make it a bit less than
71 INT_MAX to allow for adding in group terminating bytes, so that we don't have
72 to check them every time. */
73
74 #define OFLOW_MAX (INT_MAX - 20)
75
76
77 /*************************************************
78 * Code parameters and static tables *
79 *************************************************/
80
81 /* This value specifies the size of stack workspace that is used during the
82 first pre-compile phase that determines how much memory is required. The regex
83 is partly compiled into this space, but the compiled parts are discarded as
84 soon as they can be, so that hopefully there will never be an overrun. The code
85 does, however, check for an overrun. The largest amount I've seen used is 218,
86 so this number is very generous.
87
88 The same workspace is used during the second, actual compile phase for
89 remembering forward references to groups so that they can be filled in at the
90 end. Each entry in this list occupies LINK_SIZE bytes, so even when LINK_SIZE
91 is 4 there is plenty of room. */
92
93 #define COMPILE_WORK_SIZE (4096)
94
95 /* The overrun tests check for a slightly smaller size so that they detect the
96 overrun before it actually does run off the end of the data block. */
97
98 #define WORK_SIZE_CHECK (COMPILE_WORK_SIZE - 100)
99
100
101 /* Table for handling escaped characters in the range '0'-'z'. Positive returns
102 are simple data values; negative values are for special things like \d and so
103 on. Zero means further processing is needed (for things like \x), or the escape
104 is invalid. */
105
106 #ifndef EBCDIC
107
108 /* This is the "normal" table for ASCII systems or for EBCDIC systems running
109 in UTF-8 mode. */
110
111 static const short int escapes[] = {
112 0, 0,
113 0, 0,
114 0, 0,
115 0, 0,
116 0, 0,
117 CHAR_COLON, CHAR_SEMICOLON,
118 CHAR_LESS_THAN_SIGN, CHAR_EQUALS_SIGN,
119 CHAR_GREATER_THAN_SIGN, CHAR_QUESTION_MARK,
120 CHAR_COMMERCIAL_AT, -ESC_A,
121 -ESC_B, -ESC_C,
122 -ESC_D, -ESC_E,
123 0, -ESC_G,
124 -ESC_H, 0,
125 0, -ESC_K,
126 0, 0,
127 -ESC_N, 0,
128 -ESC_P, -ESC_Q,
129 -ESC_R, -ESC_S,
130 0, 0,
131 -ESC_V, -ESC_W,
132 -ESC_X, 0,
133 -ESC_Z, CHAR_LEFT_SQUARE_BRACKET,
134 CHAR_BACKSLASH, CHAR_RIGHT_SQUARE_BRACKET,
135 CHAR_CIRCUMFLEX_ACCENT, CHAR_UNDERSCORE,
136 CHAR_GRAVE_ACCENT, 7,
137 -ESC_b, 0,
138 -ESC_d, ESC_e,
139 ESC_f, 0,
140 -ESC_h, 0,
141 0, -ESC_k,
142 0, 0,
143 ESC_n, 0,
144 -ESC_p, 0,
145 ESC_r, -ESC_s,
146 ESC_tee, 0,
147 -ESC_v, -ESC_w,
148 0, 0,
149 -ESC_z
150 };
151
152 #else
153
154 /* This is the "abnormal" table for EBCDIC systems without UTF-8 support. */
155
156 static const short int escapes[] = {
157 /* 48 */ 0, 0, 0, '.', '<', '(', '+', '|',
158 /* 50 */ '&', 0, 0, 0, 0, 0, 0, 0,
159 /* 58 */ 0, 0, '!', '$', '*', ')', ';', '~',
160 /* 60 */ '-', '/', 0, 0, 0, 0, 0, 0,
161 /* 68 */ 0, 0, '|', ',', '%', '_', '>', '?',
162 /* 70 */ 0, 0, 0, 0, 0, 0, 0, 0,
163 /* 78 */ 0, '`', ':', '#', '@', '\'', '=', '"',
164 /* 80 */ 0, 7, -ESC_b, 0, -ESC_d, ESC_e, ESC_f, 0,
165 /* 88 */-ESC_h, 0, 0, '{', 0, 0, 0, 0,
166 /* 90 */ 0, 0, -ESC_k, 'l', 0, ESC_n, 0, -ESC_p,
167 /* 98 */ 0, ESC_r, 0, '}', 0, 0, 0, 0,
168 /* A0 */ 0, '~', -ESC_s, ESC_tee, 0,-ESC_v, -ESC_w, 0,
169 /* A8 */ 0,-ESC_z, 0, 0, 0, '[', 0, 0,
170 /* B0 */ 0, 0, 0, 0, 0, 0, 0, 0,
171 /* B8 */ 0, 0, 0, 0, 0, ']', '=', '-',
172 /* C0 */ '{',-ESC_A, -ESC_B, -ESC_C, -ESC_D,-ESC_E, 0, -ESC_G,
173 /* C8 */-ESC_H, 0, 0, 0, 0, 0, 0, 0,
174 /* D0 */ '}', 0, -ESC_K, 0, 0,-ESC_N, 0, -ESC_P,
175 /* D8 */-ESC_Q,-ESC_R, 0, 0, 0, 0, 0, 0,
176 /* E0 */ '\\', 0, -ESC_S, 0, 0,-ESC_V, -ESC_W, -ESC_X,
177 /* E8 */ 0,-ESC_Z, 0, 0, 0, 0, 0, 0,
178 /* F0 */ 0, 0, 0, 0, 0, 0, 0, 0,
179 /* F8 */ 0, 0, 0, 0, 0, 0, 0, 0
180 };
181 #endif
182
183
184 /* Table of special "verbs" like (*PRUNE). This is a short table, so it is
185 searched linearly. Put all the names into a single string, in order to reduce
186 the number of relocations when a shared library is dynamically linked. The
187 string is built from string macros so that it works in UTF-8 mode on EBCDIC
188 platforms. */
189
190 typedef struct verbitem {
191 int len; /* Length of verb name */
192 int op; /* Op when no arg, or -1 if arg mandatory */
193 int op_arg; /* Op when arg present, or -1 if not allowed */
194 } verbitem;
195
196 static const char verbnames[] =
197 "\0" /* Empty name is a shorthand for MARK */
198 STRING_MARK0
199 STRING_ACCEPT0
200 STRING_COMMIT0
201 STRING_F0
202 STRING_FAIL0
203 STRING_PRUNE0
204 STRING_SKIP0
205 STRING_THEN;
206
207 static const verbitem verbs[] = {
208 { 0, -1, OP_MARK },
209 { 4, -1, OP_MARK },
210 { 6, OP_ACCEPT, -1 },
211 { 6, OP_COMMIT, -1 },
212 { 1, OP_FAIL, -1 },
213 { 4, OP_FAIL, -1 },
214 { 5, OP_PRUNE, OP_PRUNE_ARG },
215 { 4, OP_SKIP, OP_SKIP_ARG },
216 { 4, OP_THEN, OP_THEN_ARG }
217 };
218
219 static const int verbcount = sizeof(verbs)/sizeof(verbitem);
220
221
222 /* Tables of names of POSIX character classes and their lengths. The names are
223 now all in a single string, to reduce the number of relocations when a shared
224 library is dynamically loaded. The list of lengths is terminated by a zero
225 length entry. The first three must be alpha, lower, upper, as this is assumed
226 for handling case independence. */
227
228 static const char posix_names[] =
229 STRING_alpha0 STRING_lower0 STRING_upper0 STRING_alnum0
230 STRING_ascii0 STRING_blank0 STRING_cntrl0 STRING_digit0
231 STRING_graph0 STRING_print0 STRING_punct0 STRING_space0
232 STRING_word0 STRING_xdigit;
233
234 static const uschar posix_name_lengths[] = {
235 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 4, 6, 0 };
236
237 /* Table of class bit maps for each POSIX class. Each class is formed from a
238 base map, with an optional addition or removal of another map. Then, for some
239 classes, there is some additional tweaking: for [:blank:] the vertical space
240 characters are removed, and for [:alpha:] and [:alnum:] the underscore
241 character is removed. The triples in the table consist of the base map offset,
242 second map offset or -1 if no second map, and a non-negative value for map
243 addition or a negative value for map subtraction (if there are two maps). The
244 absolute value of the third field has these meanings: 0 => no tweaking, 1 =>
245 remove vertical space characters, 2 => remove underscore. */
246
247 static const int posix_class_maps[] = {
248 cbit_word, cbit_digit, -2, /* alpha */
249 cbit_lower, -1, 0, /* lower */
250 cbit_upper, -1, 0, /* upper */
251 cbit_word, -1, 2, /* alnum - word without underscore */
252 cbit_print, cbit_cntrl, 0, /* ascii */
253 cbit_space, -1, 1, /* blank - a GNU extension */
254 cbit_cntrl, -1, 0, /* cntrl */
255 cbit_digit, -1, 0, /* digit */
256 cbit_graph, -1, 0, /* graph */
257 cbit_print, -1, 0, /* print */
258 cbit_punct, -1, 0, /* punct */
259 cbit_space, -1, 0, /* space */
260 cbit_word, -1, 0, /* word - a Perl extension */
261 cbit_xdigit,-1, 0 /* xdigit */
262 };
263
264 /* Table of substitutes for \d etc when PCRE_UCP is set. The POSIX class
265 substitutes must be in the order of the names, defined above, and there are
266 both positive and negative cases. NULL means no substitute. */
267
268 #ifdef SUPPORT_UCP
269 static const uschar *substitutes[] = {
270 (uschar *)"\\P{Nd}", /* \D */
271 (uschar *)"\\p{Nd}", /* \d */
272 (uschar *)"\\P{Xsp}", /* \S */ /* NOTE: Xsp is Perl space */
273 (uschar *)"\\p{Xsp}", /* \s */
274 (uschar *)"\\P{Xwd}", /* \W */
275 (uschar *)"\\p{Xwd}" /* \w */
276 };
277
278 static const uschar *posix_substitutes[] = {
279 (uschar *)"\\p{L}", /* alpha */
280 (uschar *)"\\p{Ll}", /* lower */
281 (uschar *)"\\p{Lu}", /* upper */
282 (uschar *)"\\p{Xan}", /* alnum */
283 NULL, /* ascii */
284 (uschar *)"\\h", /* blank */
285 NULL, /* cntrl */
286 (uschar *)"\\p{Nd}", /* digit */
287 NULL, /* graph */
288 NULL, /* print */
289 NULL, /* punct */
290 (uschar *)"\\p{Xps}", /* space */ /* NOTE: Xps is POSIX space */
291 (uschar *)"\\p{Xwd}", /* word */
292 NULL, /* xdigit */
293 /* Negated cases */
294 (uschar *)"\\P{L}", /* ^alpha */
295 (uschar *)"\\P{Ll}", /* ^lower */
296 (uschar *)"\\P{Lu}", /* ^upper */
297 (uschar *)"\\P{Xan}", /* ^alnum */
298 NULL, /* ^ascii */
299 (uschar *)"\\H", /* ^blank */
300 NULL, /* ^cntrl */
301 (uschar *)"\\P{Nd}", /* ^digit */
302 NULL, /* ^graph */
303 NULL, /* ^print */
304 NULL, /* ^punct */
305 (uschar *)"\\P{Xps}", /* ^space */ /* NOTE: Xps is POSIX space */
306 (uschar *)"\\P{Xwd}", /* ^word */
307 NULL /* ^xdigit */
308 };
309 #define POSIX_SUBSIZE (sizeof(posix_substitutes)/sizeof(uschar *))
310 #endif
311
312 #define STRING(a) # a
313 #define XSTRING(s) STRING(s)
314
315 /* The texts of compile-time error messages. These are "char *" because they
316 are passed to the outside world. Do not ever re-use any error number, because
317 they are documented. Always add a new error instead. Messages marked DEAD below
318 are no longer used. This used to be a table of strings, but in order to reduce
319 the number of relocations needed when a shared library is loaded dynamically,
320 it is now one long string. We cannot use a table of offsets, because the
321 lengths of inserts such as XSTRING(MAX_NAME_SIZE) are not known. Instead, we
322 simply count through to the one we want - this isn't a performance issue
323 because these strings are used only when there is a compilation error.
324
325 Each substring ends with \0 to insert a null character. This includes the final
326 substring, so that the whole string ends with \0\0, which can be detected when
327 counting through. */
328
329 static const char error_texts[] =
330 "no error\0"
331 "\\ at end of pattern\0"
332 "\\c at end of pattern\0"
333 "unrecognized character follows \\\0"
334 "numbers out of order in {} quantifier\0"
335 /* 5 */
336 "number too big in {} quantifier\0"
337 "missing terminating ] for character class\0"
338 "invalid escape sequence in character class\0"
339 "range out of order in character class\0"
340 "nothing to repeat\0"
341 /* 10 */
342 "operand of unlimited repeat could match the empty string\0" /** DEAD **/
343 "internal error: unexpected repeat\0"
344 "unrecognized character after (? or (?-\0"
345 "POSIX named classes are supported only within a class\0"
346 "missing )\0"
347 /* 15 */
348 "reference to non-existent subpattern\0"
349 "erroffset passed as NULL\0"
350 "unknown option bit(s) set\0"
351 "missing ) after comment\0"
352 "parentheses nested too deeply\0" /** DEAD **/
353 /* 20 */
354 "regular expression is too large\0"
355 "failed to get memory\0"
356 "unmatched parentheses\0"
357 "internal error: code overflow\0"
358 "unrecognized character after (?<\0"
359 /* 25 */
360 "lookbehind assertion is not fixed length\0"
361 "malformed number or name after (?(\0"
362 "conditional group contains more than two branches\0"
363 "assertion expected after (?(\0"
364 "(?R or (?[+-]digits must be followed by )\0"
365 /* 30 */
366 "unknown POSIX class name\0"
367 "POSIX collating elements are not supported\0"
368 "this version of PCRE is not compiled with PCRE_UTF8 support\0"
369 "spare error\0" /** DEAD **/
370 "character value in \\x{...} sequence is too large\0"
371 /* 35 */
372 "invalid condition (?(0)\0"
373 "\\C not allowed in lookbehind assertion\0"
374 "PCRE does not support \\L, \\l, \\N{name}, \\U, or \\u\0"
375 "number after (?C is > 255\0"
376 "closing ) for (?C expected\0"
377 /* 40 */
378 "recursive call could loop indefinitely\0"
379 "unrecognized character after (?P\0"
380 "syntax error in subpattern name (missing terminator)\0"
381 "two named subpatterns have the same name\0"
382 "invalid UTF-8 string\0"
383 /* 45 */
384 "support for \\P, \\p, and \\X has not been compiled\0"
385 "malformed \\P or \\p sequence\0"
386 "unknown property name after \\P or \\p\0"
387 "subpattern name is too long (maximum " XSTRING(MAX_NAME_SIZE) " characters)\0"
388 "too many named subpatterns (maximum " XSTRING(MAX_NAME_COUNT) ")\0"
389 /* 50 */
390 "repeated subpattern is too long\0" /** DEAD **/
391 "octal value is greater than \\377 (not in UTF-8 mode)\0"
392 "internal error: overran compiling workspace\0"
393 "internal error: previously-checked referenced subpattern not found\0"
394 "DEFINE group contains more than one branch\0"
395 /* 55 */
396 "repeating a DEFINE group is not allowed\0"
397 "inconsistent NEWLINE options\0"
398 "\\g is not followed by a braced, angle-bracketed, or quoted name/number or by a plain number\0"
399 "a numbered reference must not be zero\0"
400 "an argument is not allowed for (*ACCEPT), (*FAIL), or (*COMMIT)\0"
401 /* 60 */
402 "(*VERB) not recognized\0"
403 "number is too big\0"
404 "subpattern name expected\0"
405 "digit expected after (?+\0"
406 "] is an invalid data character in JavaScript compatibility mode\0"
407 /* 65 */
408 "different names for subpatterns of the same number are not allowed\0"
409 "(*MARK) must have an argument\0"
410 "this version of PCRE is not compiled with PCRE_UCP support\0"
411 ;
412
413 /* Table to identify digits and hex digits. This is used when compiling
414 patterns. Note that the tables in chartables are dependent on the locale, and
415 may mark arbitrary characters as digits - but the PCRE compiling code expects
416 to handle only 0-9, a-z, and A-Z as digits when compiling. That is why we have
417 a private table here. It costs 256 bytes, but it is a lot faster than doing
418 character value tests (at least in some simple cases I timed), and in some
419 applications one wants PCRE to compile efficiently as well as match
420 efficiently.
421
422 For convenience, we use the same bit definitions as in chartables:
423
424 0x04 decimal digit
425 0x08 hexadecimal digit
426
427 Then we can use ctype_digit and ctype_xdigit in the code. */
428
429 #ifndef EBCDIC
430
431 /* This is the "normal" case, for ASCII systems, and EBCDIC systems running in
432 UTF-8 mode. */
433
434 static const unsigned char digitab[] =
435 {
436 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 0- 7 */
437 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 8- 15 */
438 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 16- 23 */
439 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 24- 31 */
440 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - ' */
441 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* ( - / */
442 0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c, /* 0 - 7 */
443 0x0c,0x0c,0x00,0x00,0x00,0x00,0x00,0x00, /* 8 - ? */
444 0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* @ - G */
445 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* H - O */
446 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* P - W */
447 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* X - _ */
448 0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* ` - g */
449 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* h - o */
450 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* p - w */
451 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* x -127 */
452 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 128-135 */
453 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 136-143 */
454 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 144-151 */
455 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 152-159 */
456 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 160-167 */
457 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 168-175 */
458 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 176-183 */
459 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 184-191 */
460 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 192-199 */
461 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 200-207 */
462 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 208-215 */
463 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 216-223 */
464 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 224-231 */
465 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 232-239 */
466 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 240-247 */
467 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00};/* 248-255 */
468
469 #else
470
471 /* This is the "abnormal" case, for EBCDIC systems not running in UTF-8 mode. */
472
473 static const unsigned char digitab[] =
474 {
475 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 0- 7 0 */
476 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 8- 15 */
477 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 16- 23 10 */
478 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 24- 31 */
479 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 32- 39 20 */
480 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 40- 47 */
481 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 48- 55 30 */
482 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 56- 63 */
483 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - 71 40 */
484 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 72- | */
485 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* & - 87 50 */
486 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 88- 95 */
487 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - -103 60 */
488 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 104- ? */
489 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 112-119 70 */
490 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 120- " */
491 0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* 128- g 80 */
492 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* h -143 */
493 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 144- p 90 */
494 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* q -159 */
495 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 160- x A0 */
496 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* y -175 */
497 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* ^ -183 B0 */
498 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 184-191 */
499 0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* { - G C0 */
500 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* H -207 */
501 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* } - P D0 */
502 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* Q -223 */
503 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* \ - X E0 */
504 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* Y -239 */
505 0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c, /* 0 - 7 F0 */
506 0x0c,0x0c,0x00,0x00,0x00,0x00,0x00,0x00};/* 8 -255 */
507
508 static const unsigned char ebcdic_chartab[] = { /* chartable partial dup */
509 0x80,0x00,0x00,0x00,0x00,0x01,0x00,0x00, /* 0- 7 */
510 0x00,0x00,0x00,0x00,0x01,0x01,0x00,0x00, /* 8- 15 */
511 0x00,0x00,0x00,0x00,0x00,0x01,0x00,0x00, /* 16- 23 */
512 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 24- 31 */
513 0x00,0x00,0x00,0x00,0x00,0x01,0x00,0x00, /* 32- 39 */
514 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 40- 47 */
515 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 48- 55 */
516 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 56- 63 */
517 0x01,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - 71 */
518 0x00,0x00,0x00,0x80,0x00,0x80,0x80,0x80, /* 72- | */
519 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* & - 87 */
520 0x00,0x00,0x00,0x80,0x80,0x80,0x00,0x00, /* 88- 95 */
521 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - -103 */
522 0x00,0x00,0x00,0x00,0x00,0x10,0x00,0x80, /* 104- ? */
523 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 112-119 */
524 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 120- " */
525 0x00,0x1a,0x1a,0x1a,0x1a,0x1a,0x1a,0x12, /* 128- g */
526 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* h -143 */
527 0x00,0x12,0x12,0x12,0x12,0x12,0x12,0x12, /* 144- p */
528 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* q -159 */
529 0x00,0x00,0x12,0x12,0x12,0x12,0x12,0x12, /* 160- x */
530 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* y -175 */
531 0x80,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* ^ -183 */
532 0x00,0x00,0x80,0x00,0x00,0x00,0x00,0x00, /* 184-191 */
533 0x80,0x1a,0x1a,0x1a,0x1a,0x1a,0x1a,0x12, /* { - G */
534 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* H -207 */
535 0x00,0x12,0x12,0x12,0x12,0x12,0x12,0x12, /* } - P */
536 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* Q -223 */
537 0x00,0x00,0x12,0x12,0x12,0x12,0x12,0x12, /* \ - X */
538 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* Y -239 */
539 0x1c,0x1c,0x1c,0x1c,0x1c,0x1c,0x1c,0x1c, /* 0 - 7 */
540 0x1c,0x1c,0x00,0x00,0x00,0x00,0x00,0x00};/* 8 -255 */
541 #endif
542
543
544 /* Definition to allow mutual recursion */
545
546 static BOOL
547 compile_regex(int, int, uschar **, const uschar **, int *, BOOL, BOOL, int,
548 int *, int *, branch_chain *, compile_data *, int *);
549
550
551
552 /*************************************************
553 * Find an error text *
554 *************************************************/
555
556 /* The error texts are now all in one long string, to save on relocations. As
557 some of the text is of unknown length, we can't use a table of offsets.
558 Instead, just count through the strings. This is not a performance issue
559 because it happens only when there has been a compilation error.
560
561 Argument: the error number
562 Returns: pointer to the error string
563 */
564
565 static const char *
566 find_error_text(int n)
567 {
568 const char *s = error_texts;
569 for (; n > 0; n--)
570 {
571 while (*s++ != 0) {};
572 if (*s == 0) return "Error text not found (please report)";
573 }
574 return s;
575 }
576
577
578 /*************************************************
579 * Handle escapes *
580 *************************************************/
581
582 /* This function is called when a \ has been encountered. It either returns a
583 positive value for a simple escape such as \n, or a negative value which
584 encodes one of the more complicated things such as \d. A backreference to group
585 n is returned as -(ESC_REF + n); ESC_REF is the highest ESC_xxx macro. When
586 UTF-8 is enabled, a positive value greater than 255 may be returned. On entry,
587 ptr is pointing at the \. On exit, it is on the final character of the escape
588 sequence.
589
590 Arguments:
591 ptrptr points to the pattern position pointer
592 errorcodeptr points to the errorcode variable
593 bracount number of previous extracting brackets
594 options the options bits
595 isclass TRUE if inside a character class
596
597 Returns: zero or positive => a data character
598 negative => a special escape sequence
599 on error, errorcodeptr is set
600 */
601
602 static int
603 check_escape(const uschar **ptrptr, int *errorcodeptr, int bracount,
604 int options, BOOL isclass)
605 {
606 BOOL utf8 = (options & PCRE_UTF8) != 0;
607 const uschar *ptr = *ptrptr + 1;
608 int c, i;
609
610 GETCHARINCTEST(c, ptr); /* Get character value, increment pointer */
611 ptr--; /* Set pointer back to the last byte */
612
613 /* If backslash is at the end of the pattern, it's an error. */
614
615 if (c == 0) *errorcodeptr = ERR1;
616
617 /* Non-alphanumerics are literals. For digits or letters, do an initial lookup
618 in a table. A non-zero result is something that can be returned immediately.
619 Otherwise further processing may be required. */
620
621 #ifndef EBCDIC /* ASCII/UTF-8 coding */
622 else if (c < CHAR_0 || c > CHAR_z) {} /* Not alphanumeric */
623 else if ((i = escapes[c - CHAR_0]) != 0) c = i;
624
625 #else /* EBCDIC coding */
626 else if (c < 'a' || (ebcdic_chartab[c] & 0x0E) == 0) {} /* Not alphanumeric */
627 else if ((i = escapes[c - 0x48]) != 0) c = i;
628 #endif
629
630 /* Escapes that need further processing, or are illegal. */
631
632 else
633 {
634 const uschar *oldptr;
635 BOOL braced, negated;
636
637 switch (c)
638 {
639 /* A number of Perl escapes are not handled by PCRE. We give an explicit
640 error. */
641
642 case CHAR_l:
643 case CHAR_L:
644 case CHAR_u:
645 case CHAR_U:
646 *errorcodeptr = ERR37;
647 break;
648
649 /* \g must be followed by one of a number of specific things:
650
651 (1) A number, either plain or braced. If positive, it is an absolute
652 backreference. If negative, it is a relative backreference. This is a Perl
653 5.10 feature.
654
655 (2) Perl 5.10 also supports \g{name} as a reference to a named group. This
656 is part of Perl's movement towards a unified syntax for back references. As
657 this is synonymous with \k{name}, we fudge it up by pretending it really
658 was \k.
659
660 (3) For Oniguruma compatibility we also support \g followed by a name or a
661 number either in angle brackets or in single quotes. However, these are
662 (possibly recursive) subroutine calls, _not_ backreferences. Just return
663 the -ESC_g code (cf \k). */
664
665 case CHAR_g:
666 if (ptr[1] == CHAR_LESS_THAN_SIGN || ptr[1] == CHAR_APOSTROPHE)
667 {
668 c = -ESC_g;
669 break;
670 }
671
672 /* Handle the Perl-compatible cases */
673
674 if (ptr[1] == CHAR_LEFT_CURLY_BRACKET)
675 {
676 const uschar *p;
677 for (p = ptr+2; *p != 0 && *p != CHAR_RIGHT_CURLY_BRACKET; p++)
678 if (*p != CHAR_MINUS && (digitab[*p] & ctype_digit) == 0) break;
679 if (*p != 0 && *p != CHAR_RIGHT_CURLY_BRACKET)
680 {
681 c = -ESC_k;
682 break;
683 }
684 braced = TRUE;
685 ptr++;
686 }
687 else braced = FALSE;
688
689 if (ptr[1] == CHAR_MINUS)
690 {
691 negated = TRUE;
692 ptr++;
693 }
694 else negated = FALSE;
695
696 c = 0;
697 while ((digitab[ptr[1]] & ctype_digit) != 0)
698 c = c * 10 + *(++ptr) - CHAR_0;
699
700 if (c < 0) /* Integer overflow */
701 {
702 *errorcodeptr = ERR61;
703 break;
704 }
705
706 if (braced && *(++ptr) != CHAR_RIGHT_CURLY_BRACKET)
707 {
708 *errorcodeptr = ERR57;
709 break;
710 }
711
712 if (c == 0)
713 {
714 *errorcodeptr = ERR58;
715 break;
716 }
717
718 if (negated)
719 {
720 if (c > bracount)
721 {
722 *errorcodeptr = ERR15;
723 break;
724 }
725 c = bracount - (c - 1);
726 }
727
728 c = -(ESC_REF + c);
729 break;
730
731 /* The handling of escape sequences consisting of a string of digits
732 starting with one that is not zero is not straightforward. By experiment,
733 the way Perl works seems to be as follows:
734
735 Outside a character class, the digits are read as a decimal number. If the
736 number is less than 10, or if there are that many previous extracting
737 left brackets, then it is a back reference. Otherwise, up to three octal
738 digits are read to form an escaped byte. Thus \123 is likely to be octal
739 123 (cf \0123, which is octal 012 followed by the literal 3). If the octal
740 value is greater than 377, the least significant 8 bits are taken. Inside a
741 character class, \ followed by a digit is always an octal number. */
742
743 case CHAR_1: case CHAR_2: case CHAR_3: case CHAR_4: case CHAR_5:
744 case CHAR_6: case CHAR_7: case CHAR_8: case CHAR_9:
745
746 if (!isclass)
747 {
748 oldptr = ptr;
749 c -= CHAR_0;
750 while ((digitab[ptr[1]] & ctype_digit) != 0)
751 c = c * 10 + *(++ptr) - CHAR_0;
752 if (c < 0) /* Integer overflow */
753 {
754 *errorcodeptr = ERR61;
755 break;
756 }
757 if (c < 10 || c <= bracount)
758 {
759 c = -(ESC_REF + c);
760 break;
761 }
762 ptr = oldptr; /* Put the pointer back and fall through */
763 }
764
765 /* Handle an octal number following \. If the first digit is 8 or 9, Perl
766 generates a binary zero byte and treats the digit as a following literal.
767 Thus we have to pull back the pointer by one. */
768
769 if ((c = *ptr) >= CHAR_8)
770 {
771 ptr--;
772 c = 0;
773 break;
774 }
775
776 /* \0 always starts an octal number, but we may drop through to here with a
777 larger first octal digit. The original code used just to take the least
778 significant 8 bits of octal numbers (I think this is what early Perls used
779 to do). Nowadays we allow for larger numbers in UTF-8 mode, but no more
780 than 3 octal digits. */
781
782 case CHAR_0:
783 c -= CHAR_0;
784 while(i++ < 2 && ptr[1] >= CHAR_0 && ptr[1] <= CHAR_7)
785 c = c * 8 + *(++ptr) - CHAR_0;
786 if (!utf8 && c > 255) *errorcodeptr = ERR51;
787 break;
788
789 /* \x is complicated. \x{ddd} is a character number which can be greater
790 than 0xff in utf8 mode, but only if the ddd are hex digits. If not, { is
791 treated as a data character. */
792
793 case CHAR_x:
794 if (ptr[1] == CHAR_LEFT_CURLY_BRACKET)
795 {
796 const uschar *pt = ptr + 2;
797 int count = 0;
798
799 c = 0;
800 while ((digitab[*pt] & ctype_xdigit) != 0)
801 {
802 register int cc = *pt++;
803 if (c == 0 && cc == CHAR_0) continue; /* Leading zeroes */
804 count++;
805
806 #ifndef EBCDIC /* ASCII/UTF-8 coding */
807 if (cc >= CHAR_a) cc -= 32; /* Convert to upper case */
808 c = (c << 4) + cc - ((cc < CHAR_A)? CHAR_0 : (CHAR_A - 10));
809 #else /* EBCDIC coding */
810 if (cc >= CHAR_a && cc <= CHAR_z) cc += 64; /* Convert to upper case */
811 c = (c << 4) + cc - ((cc >= CHAR_0)? CHAR_0 : (CHAR_A - 10));
812 #endif
813 }
814
815 if (*pt == CHAR_RIGHT_CURLY_BRACKET)
816 {
817 if (c < 0 || count > (utf8? 8 : 2)) *errorcodeptr = ERR34;
818 ptr = pt;
819 break;
820 }
821
822 /* If the sequence of hex digits does not end with '}', then we don't
823 recognize this construct; fall through to the normal \x handling. */
824 }
825
826 /* Read just a single-byte hex-defined char */
827
828 c = 0;
829 while (i++ < 2 && (digitab[ptr[1]] & ctype_xdigit) != 0)
830 {
831 int cc; /* Some compilers don't like */
832 cc = *(++ptr); /* ++ in initializers */
833 #ifndef EBCDIC /* ASCII/UTF-8 coding */
834 if (cc >= CHAR_a) cc -= 32; /* Convert to upper case */
835 c = c * 16 + cc - ((cc < CHAR_A)? CHAR_0 : (CHAR_A - 10));
836 #else /* EBCDIC coding */
837 if (cc <= CHAR_z) cc += 64; /* Convert to upper case */
838 c = c * 16 + cc - ((cc >= CHAR_0)? CHAR_0 : (CHAR_A - 10));
839 #endif
840 }
841 break;
842
843 /* For \c, a following letter is upper-cased; then the 0x40 bit is flipped.
844 This coding is ASCII-specific, but then the whole concept of \cx is
845 ASCII-specific. (However, an EBCDIC equivalent has now been added.) */
846
847 case CHAR_c:
848 c = *(++ptr);
849 if (c == 0)
850 {
851 *errorcodeptr = ERR2;
852 break;
853 }
854
855 #ifndef EBCDIC /* ASCII/UTF-8 coding */
856 if (c >= CHAR_a && c <= CHAR_z) c -= 32;
857 c ^= 0x40;
858 #else /* EBCDIC coding */
859 if (c >= CHAR_a && c <= CHAR_z) c += 64;
860 c ^= 0xC0;
861 #endif
862 break;
863
864 /* PCRE_EXTRA enables extensions to Perl in the matter of escapes. Any
865 other alphanumeric following \ is an error if PCRE_EXTRA was set;
866 otherwise, for Perl compatibility, it is a literal. This code looks a bit
867 odd, but there used to be some cases other than the default, and there may
868 be again in future, so I haven't "optimized" it. */
869
870 default:
871 if ((options & PCRE_EXTRA) != 0) switch(c)
872 {
873 default:
874 *errorcodeptr = ERR3;
875 break;
876 }
877 break;
878 }
879 }
880
881 /* Perl supports \N{name} for character names, as well as plain \N for "not
882 newline". PCRE does not support \N{name}. */
883
884 if (c == -ESC_N && ptr[1] == CHAR_LEFT_CURLY_BRACKET)
885 *errorcodeptr = ERR37;
886
887 /* If PCRE_UCP is set, we change the values for \d etc. */
888
889 if ((options & PCRE_UCP) != 0 && c <= -ESC_D && c >= -ESC_w)
890 c -= (ESC_DU - ESC_D);
891
892 /* Set the pointer to the final character before returning. */
893
894 *ptrptr = ptr;
895 return c;
896 }
897
898
899
900 #ifdef SUPPORT_UCP
901 /*************************************************
902 * Handle \P and \p *
903 *************************************************/
904
905 /* This function is called after \P or \p has been encountered, provided that
906 PCRE is compiled with support for Unicode properties. On entry, ptrptr is
907 pointing at the P or p. On exit, it is pointing at the final character of the
908 escape sequence.
909
910 Argument:
911 ptrptr points to the pattern position pointer
912 negptr points to a boolean that is set TRUE for negation else FALSE
913 dptr points to an int that is set to the detailed property value
914 errorcodeptr points to the error code variable
915
916 Returns: type value from ucp_type_table, or -1 for an invalid type
917 */
918
919 static int
920 get_ucp(const uschar **ptrptr, BOOL *negptr, int *dptr, int *errorcodeptr)
921 {
922 int c, i, bot, top;
923 const uschar *ptr = *ptrptr;
924 char name[32];
925
926 c = *(++ptr);
927 if (c == 0) goto ERROR_RETURN;
928
929 *negptr = FALSE;
930
931 /* \P or \p can be followed by a name in {}, optionally preceded by ^ for
932 negation. */
933
934 if (c == CHAR_LEFT_CURLY_BRACKET)
935 {
936 if (ptr[1] == CHAR_CIRCUMFLEX_ACCENT)
937 {
938 *negptr = TRUE;
939 ptr++;
940 }
941 for (i = 0; i < (int)sizeof(name) - 1; i++)
942 {
943 c = *(++ptr);
944 if (c == 0) goto ERROR_RETURN;
945 if (c == CHAR_RIGHT_CURLY_BRACKET) break;
946 name[i] = c;
947 }
948 if (c != CHAR_RIGHT_CURLY_BRACKET) goto ERROR_RETURN;
949 name[i] = 0;
950 }
951
952 /* Otherwise there is just one following character */
953
954 else
955 {
956 name[0] = c;
957 name[1] = 0;
958 }
959
960 *ptrptr = ptr;
961
962 /* Search for a recognized property name using binary chop */
963
964 bot = 0;
965 top = _pcre_utt_size;
966
967 while (bot < top)
968 {
969 i = (bot + top) >> 1;
970 c = strcmp(name, _pcre_utt_names + _pcre_utt[i].name_offset);
971 if (c == 0)
972 {
973 *dptr = _pcre_utt[i].value;
974 return _pcre_utt[i].type;
975 }
976 if (c > 0) bot = i + 1; else top = i;
977 }
978
979 *errorcodeptr = ERR47;
980 *ptrptr = ptr;
981 return -1;
982
983 ERROR_RETURN:
984 *errorcodeptr = ERR46;
985 *ptrptr = ptr;
986 return -1;
987 }
988 #endif
989
990
991
992
993 /*************************************************
994 * Check for counted repeat *
995 *************************************************/
996
997 /* This function is called when a '{' is encountered in a place where it might
998 start a quantifier. It looks ahead to see if it really is a quantifier or not.
999 It is only a quantifier if it is one of the forms {ddd} {ddd,} or {ddd,ddd}
1000 where the ddds are digits.
1001
1002 Arguments:
1003 p pointer to the first char after '{'
1004
1005 Returns: TRUE or FALSE
1006 */
1007
1008 static BOOL
1009 is_counted_repeat(const uschar *p)
1010 {
1011 if ((digitab[*p++] & ctype_digit) == 0) return FALSE;
1012 while ((digitab[*p] & ctype_digit) != 0) p++;
1013 if (*p == CHAR_RIGHT_CURLY_BRACKET) return TRUE;
1014
1015 if (*p++ != CHAR_COMMA) return FALSE;
1016 if (*p == CHAR_RIGHT_CURLY_BRACKET) return TRUE;
1017
1018 if ((digitab[*p++] & ctype_digit) == 0) return FALSE;
1019 while ((digitab[*p] & ctype_digit) != 0) p++;
1020
1021 return (*p == CHAR_RIGHT_CURLY_BRACKET);
1022 }
1023
1024
1025
1026 /*************************************************
1027 * Read repeat counts *
1028 *************************************************/
1029
1030 /* Read an item of the form {n,m} and return the values. This is called only
1031 after is_counted_repeat() has confirmed that a repeat-count quantifier exists,
1032 so the syntax is guaranteed to be correct, but we need to check the values.
1033
1034 Arguments:
1035 p pointer to first char after '{'
1036 minp pointer to int for min
1037 maxp pointer to int for max
1038 returned as -1 if no max
1039 errorcodeptr points to error code variable
1040
1041 Returns: pointer to '}' on success;
1042 current ptr on error, with errorcodeptr set non-zero
1043 */
1044
1045 static const uschar *
1046 read_repeat_counts(const uschar *p, int *minp, int *maxp, int *errorcodeptr)
1047 {
1048 int min = 0;
1049 int max = -1;
1050
1051 /* Read the minimum value and do a paranoid check: a negative value indicates
1052 an integer overflow. */
1053
1054 while ((digitab[*p] & ctype_digit) != 0) min = min * 10 + *p++ - CHAR_0;
1055 if (min < 0 || min > 65535)
1056 {
1057 *errorcodeptr = ERR5;
1058 return p;
1059 }
1060
1061 /* Read the maximum value if there is one, and again do a paranoid on its size.
1062 Also, max must not be less than min. */
1063
1064 if (*p == CHAR_RIGHT_CURLY_BRACKET) max = min; else
1065 {
1066 if (*(++p) != CHAR_RIGHT_CURLY_BRACKET)
1067 {
1068 max = 0;
1069 while((digitab[*p] & ctype_digit) != 0) max = max * 10 + *p++ - CHAR_0;
1070 if (max < 0 || max > 65535)
1071 {
1072 *errorcodeptr = ERR5;
1073 return p;
1074 }
1075 if (max < min)
1076 {
1077 *errorcodeptr = ERR4;
1078 return p;
1079 }
1080 }
1081 }
1082
1083 /* Fill in the required variables, and pass back the pointer to the terminating
1084 '}'. */
1085
1086 *minp = min;
1087 *maxp = max;
1088 return p;
1089 }
1090
1091
1092
1093 /*************************************************
1094 * Subroutine for finding forward reference *
1095 *************************************************/
1096
1097 /* This recursive function is called only from find_parens() below. The
1098 top-level call starts at the beginning of the pattern. All other calls must
1099 start at a parenthesis. It scans along a pattern's text looking for capturing
1100 subpatterns, and counting them. If it finds a named pattern that matches the
1101 name it is given, it returns its number. Alternatively, if the name is NULL, it
1102 returns when it reaches a given numbered subpattern. We know that if (?P< is
1103 encountered, the name will be terminated by '>' because that is checked in the
1104 first pass. Recursion is used to keep track of subpatterns that reset the
1105 capturing group numbers - the (?| feature.
1106
1107 Arguments:
1108 ptrptr address of the current character pointer (updated)
1109 cd compile background data
1110 name name to seek, or NULL if seeking a numbered subpattern
1111 lorn name length, or subpattern number if name is NULL
1112 xmode TRUE if we are in /x mode
1113 count pointer to the current capturing subpattern number (updated)
1114
1115 Returns: the number of the named subpattern, or -1 if not found
1116 */
1117
1118 static int
1119 find_parens_sub(uschar **ptrptr, compile_data *cd, const uschar *name, int lorn,
1120 BOOL xmode, int *count)
1121 {
1122 uschar *ptr = *ptrptr;
1123 int start_count = *count;
1124 int hwm_count = start_count;
1125 BOOL dup_parens = FALSE;
1126
1127 /* If the first character is a parenthesis, check on the type of group we are
1128 dealing with. The very first call may not start with a parenthesis. */
1129
1130 if (ptr[0] == CHAR_LEFT_PARENTHESIS)
1131 {
1132 if (ptr[1] == CHAR_QUESTION_MARK &&
1133 ptr[2] == CHAR_VERTICAL_LINE)
1134 {
1135 ptr += 3;
1136 dup_parens = TRUE;
1137 }
1138
1139 /* Handle a normal, unnamed capturing parenthesis */
1140
1141 else if (ptr[1] != CHAR_QUESTION_MARK && ptr[1] != CHAR_ASTERISK)
1142 {
1143 *count += 1;
1144 if (name == NULL && *count == lorn) return *count;
1145 ptr++;
1146 }
1147
1148 /* Handle a condition. If it is an assertion, just carry on so that it
1149 is processed as normal. If not, skip to the closing parenthesis of the
1150 condition (there can't be any nested parens. */
1151
1152 else if (ptr[2] == CHAR_LEFT_PARENTHESIS)
1153 {
1154 ptr += 2;
1155 if (ptr[1] != CHAR_QUESTION_MARK)
1156 {
1157 while (*ptr != 0 && *ptr != CHAR_RIGHT_PARENTHESIS) ptr++;
1158 if (*ptr != 0) ptr++;
1159 }
1160 }
1161
1162 /* We have either (? or (* and not a condition */
1163
1164 else
1165 {
1166 ptr += 2;
1167 if (*ptr == CHAR_P) ptr++; /* Allow optional P */
1168
1169 /* We have to disambiguate (?<! and (?<= from (?<name> for named groups */
1170
1171 if ((*ptr == CHAR_LESS_THAN_SIGN && ptr[1] != CHAR_EXCLAMATION_MARK &&
1172 ptr[1] != CHAR_EQUALS_SIGN) || *ptr == CHAR_APOSTROPHE)
1173 {
1174 int term;
1175 const uschar *thisname;
1176 *count += 1;
1177 if (name == NULL && *count == lorn) return *count;
1178 term = *ptr++;
1179 if (term == CHAR_LESS_THAN_SIGN) term = CHAR_GREATER_THAN_SIGN;
1180 thisname = ptr;
1181 while (*ptr != term) ptr++;
1182 if (name != NULL && lorn == ptr - thisname &&
1183 strncmp((const char *)name, (const char *)thisname, lorn) == 0)
1184 return *count;
1185 term++;
1186 }
1187 }
1188 }
1189
1190 /* Past any initial parenthesis handling, scan for parentheses or vertical
1191 bars. */
1192
1193 for (; *ptr != 0; ptr++)
1194 {
1195 /* Skip over backslashed characters and also entire \Q...\E */
1196
1197 if (*ptr == CHAR_BACKSLASH)
1198 {
1199 if (*(++ptr) == 0) goto FAIL_EXIT;
1200 if (*ptr == CHAR_Q) for (;;)
1201 {
1202 while (*(++ptr) != 0 && *ptr != CHAR_BACKSLASH) {};
1203 if (*ptr == 0) goto FAIL_EXIT;
1204 if (*(++ptr) == CHAR_E) break;
1205 }
1206 continue;
1207 }
1208
1209 /* Skip over character classes; this logic must be similar to the way they
1210 are handled for real. If the first character is '^', skip it. Also, if the
1211 first few characters (either before or after ^) are \Q\E or \E we skip them
1212 too. This makes for compatibility with Perl. Note the use of STR macros to
1213 encode "Q\\E" so that it works in UTF-8 on EBCDIC platforms. */
1214
1215 if (*ptr == CHAR_LEFT_SQUARE_BRACKET)
1216 {
1217 BOOL negate_class = FALSE;
1218 for (;;)
1219 {
1220 if (ptr[1] == CHAR_BACKSLASH)
1221 {
1222 if (ptr[2] == CHAR_E)
1223 ptr+= 2;
1224 else if (strncmp((const char *)ptr+2,
1225 STR_Q STR_BACKSLASH STR_E, 3) == 0)
1226 ptr += 4;
1227 else
1228 break;
1229 }
1230 else if (!negate_class && ptr[1] == CHAR_CIRCUMFLEX_ACCENT)
1231 {
1232 negate_class = TRUE;
1233 ptr++;
1234 }
1235 else break;
1236 }
1237
1238 /* If the next character is ']', it is a data character that must be
1239 skipped, except in JavaScript compatibility mode. */
1240
1241 if (ptr[1] == CHAR_RIGHT_SQUARE_BRACKET &&
1242 (cd->external_options & PCRE_JAVASCRIPT_COMPAT) == 0)
1243 ptr++;
1244
1245 while (*(++ptr) != CHAR_RIGHT_SQUARE_BRACKET)
1246 {
1247 if (*ptr == 0) return -1;
1248 if (*ptr == CHAR_BACKSLASH)
1249 {
1250 if (*(++ptr) == 0) goto FAIL_EXIT;
1251 if (*ptr == CHAR_Q) for (;;)
1252 {
1253 while (*(++ptr) != 0 && *ptr != CHAR_BACKSLASH) {};
1254 if (*ptr == 0) goto FAIL_EXIT;
1255 if (*(++ptr) == CHAR_E) break;
1256 }
1257 continue;
1258 }
1259 }
1260 continue;
1261 }
1262
1263 /* Skip comments in /x mode */
1264
1265 if (xmode && *ptr == CHAR_NUMBER_SIGN)
1266 {
1267 while (*(++ptr) != 0 && *ptr != CHAR_NL) {};
1268 if (*ptr == 0) goto FAIL_EXIT;
1269 continue;
1270 }
1271
1272 /* Check for the special metacharacters */
1273
1274 if (*ptr == CHAR_LEFT_PARENTHESIS)
1275 {
1276 int rc = find_parens_sub(&ptr, cd, name, lorn, xmode, count);
1277 if (rc > 0) return rc;
1278 if (*ptr == 0) goto FAIL_EXIT;
1279 }
1280
1281 else if (*ptr == CHAR_RIGHT_PARENTHESIS)
1282 {
1283 if (dup_parens && *count < hwm_count) *count = hwm_count;
1284 *ptrptr = ptr;
1285 return -1;
1286 }
1287
1288 else if (*ptr == CHAR_VERTICAL_LINE && dup_parens)
1289 {
1290 if (*count > hwm_count) hwm_count = *count;
1291 *count = start_count;
1292 }
1293 }
1294
1295 FAIL_EXIT:
1296 *ptrptr = ptr;
1297 return -1;
1298 }
1299
1300
1301
1302
1303 /*************************************************
1304 * Find forward referenced subpattern *
1305 *************************************************/
1306
1307 /* This function scans along a pattern's text looking for capturing
1308 subpatterns, and counting them. If it finds a named pattern that matches the
1309 name it is given, it returns its number. Alternatively, if the name is NULL, it
1310 returns when it reaches a given numbered subpattern. This is used for forward
1311 references to subpatterns. We used to be able to start this scan from the
1312 current compiling point, using the current count value from cd->bracount, and
1313 do it all in a single loop, but the addition of the possibility of duplicate
1314 subpattern numbers means that we have to scan from the very start, in order to
1315 take account of such duplicates, and to use a recursive function to keep track
1316 of the different types of group.
1317
1318 Arguments:
1319 cd compile background data
1320 name name to seek, or NULL if seeking a numbered subpattern
1321 lorn name length, or subpattern number if name is NULL
1322 xmode TRUE if we are in /x mode
1323
1324 Returns: the number of the found subpattern, or -1 if not found
1325 */
1326
1327 static int
1328 find_parens(compile_data *cd, const uschar *name, int lorn, BOOL xmode)
1329 {
1330 uschar *ptr = (uschar *)cd->start_pattern;
1331 int count = 0;
1332 int rc;
1333
1334 /* If the pattern does not start with an opening parenthesis, the first call
1335 to find_parens_sub() will scan right to the end (if necessary). However, if it
1336 does start with a parenthesis, find_parens_sub() will return when it hits the
1337 matching closing parens. That is why we have to have a loop. */
1338
1339 for (;;)
1340 {
1341 rc = find_parens_sub(&ptr, cd, name, lorn, xmode, &count);
1342 if (rc > 0 || *ptr++ == 0) break;
1343 }
1344
1345 return rc;
1346 }
1347
1348
1349
1350
1351 /*************************************************
1352 * Find first significant op code *
1353 *************************************************/
1354
1355 /* This is called by several functions that scan a compiled expression looking
1356 for a fixed first character, or an anchoring op code etc. It skips over things
1357 that do not influence this. For some calls, a change of option is important.
1358 For some calls, it makes sense to skip negative forward and all backward
1359 assertions, and also the \b assertion; for others it does not.
1360
1361 Arguments:
1362 code pointer to the start of the group
1363 options pointer to external options
1364 optbit the option bit whose changing is significant, or
1365 zero if none are
1366 skipassert TRUE if certain assertions are to be skipped
1367
1368 Returns: pointer to the first significant opcode
1369 */
1370
1371 static const uschar*
1372 first_significant_code(const uschar *code, int *options, int optbit,
1373 BOOL skipassert)
1374 {
1375 for (;;)
1376 {
1377 switch ((int)*code)
1378 {
1379 case OP_OPT:
1380 if (optbit > 0 && ((int)code[1] & optbit) != (*options & optbit))
1381 *options = (int)code[1];
1382 code += 2;
1383 break;
1384
1385 case OP_ASSERT_NOT:
1386 case OP_ASSERTBACK:
1387 case OP_ASSERTBACK_NOT:
1388 if (!skipassert) return code;
1389 do code += GET(code, 1); while (*code == OP_ALT);
1390 code += _pcre_OP_lengths[*code];
1391 break;
1392
1393 case OP_WORD_BOUNDARY:
1394 case OP_NOT_WORD_BOUNDARY:
1395 if (!skipassert) return code;
1396 /* Fall through */
1397
1398 case OP_CALLOUT:
1399 case OP_CREF:
1400 case OP_NCREF:
1401 case OP_RREF:
1402 case OP_NRREF:
1403 case OP_DEF:
1404 code += _pcre_OP_lengths[*code];
1405 break;
1406
1407 default:
1408 return code;
1409 }
1410 }
1411 /* Control never reaches here */
1412 }
1413
1414
1415
1416
1417 /*************************************************
1418 * Find the fixed length of a branch *
1419 *************************************************/
1420
1421 /* Scan a branch and compute the fixed length of subject that will match it,
1422 if the length is fixed. This is needed for dealing with backward assertions.
1423 In UTF8 mode, the result is in characters rather than bytes. The branch is
1424 temporarily terminated with OP_END when this function is called.
1425
1426 This function is called when a backward assertion is encountered, so that if it
1427 fails, the error message can point to the correct place in the pattern.
1428 However, we cannot do this when the assertion contains subroutine calls,
1429 because they can be forward references. We solve this by remembering this case
1430 and doing the check at the end; a flag specifies which mode we are running in.
1431
1432 Arguments:
1433 code points to the start of the pattern (the bracket)
1434 options the compiling options
1435 atend TRUE if called when the pattern is complete
1436 cd the "compile data" structure
1437
1438 Returns: the fixed length,
1439 or -1 if there is no fixed length,
1440 or -2 if \C was encountered
1441 or -3 if an OP_RECURSE item was encountered and atend is FALSE
1442 */
1443
1444 static int
1445 find_fixedlength(uschar *code, int options, BOOL atend, compile_data *cd)
1446 {
1447 int length = -1;
1448
1449 register int branchlength = 0;
1450 register uschar *cc = code + 1 + LINK_SIZE;
1451
1452 /* Scan along the opcodes for this branch. If we get to the end of the
1453 branch, check the length against that of the other branches. */
1454
1455 for (;;)
1456 {
1457 int d;
1458 uschar *ce, *cs;
1459 register int op = *cc;
1460 switch (op)
1461 {
1462 case OP_CBRA:
1463 case OP_BRA:
1464 case OP_ONCE:
1465 case OP_COND:
1466 d = find_fixedlength(cc + ((op == OP_CBRA)? 2:0), options, atend, cd);
1467 if (d < 0) return d;
1468 branchlength += d;
1469 do cc += GET(cc, 1); while (*cc == OP_ALT);
1470 cc += 1 + LINK_SIZE;
1471 break;
1472
1473 /* Reached end of a branch; if it's a ket it is the end of a nested
1474 call. If it's ALT it is an alternation in a nested call. If it is
1475 END it's the end of the outer call. All can be handled by the same code. */
1476
1477 case OP_ALT:
1478 case OP_KET:
1479 case OP_KETRMAX:
1480 case OP_KETRMIN:
1481 case OP_END:
1482 if (length < 0) length = branchlength;
1483 else if (length != branchlength) return -1;
1484 if (*cc != OP_ALT) return length;
1485 cc += 1 + LINK_SIZE;
1486 branchlength = 0;
1487 break;
1488
1489 /* A true recursion implies not fixed length, but a subroutine call may
1490 be OK. If the subroutine is a forward reference, we can't deal with
1491 it until the end of the pattern, so return -3. */
1492
1493 case OP_RECURSE:
1494 if (!atend) return -3;
1495 cs = ce = (uschar *)cd->start_code + GET(cc, 1); /* Start subpattern */
1496 do ce += GET(ce, 1); while (*ce == OP_ALT); /* End subpattern */
1497 if (cc > cs && cc < ce) return -1; /* Recursion */
1498 d = find_fixedlength(cs + 2, options, atend, cd);
1499 if (d < 0) return d;
1500 branchlength += d;
1501 cc += 1 + LINK_SIZE;
1502 break;
1503
1504 /* Skip over assertive subpatterns */
1505
1506 case OP_ASSERT:
1507 case OP_ASSERT_NOT:
1508 case OP_ASSERTBACK:
1509 case OP_ASSERTBACK_NOT:
1510 do cc += GET(cc, 1); while (*cc == OP_ALT);
1511 /* Fall through */
1512
1513 /* Skip over things that don't match chars */
1514
1515 case OP_REVERSE:
1516 case OP_CREF:
1517 case OP_NCREF:
1518 case OP_RREF:
1519 case OP_NRREF:
1520 case OP_DEF:
1521 case OP_OPT:
1522 case OP_CALLOUT:
1523 case OP_SOD:
1524 case OP_SOM:
1525 case OP_SET_SOM:
1526 case OP_EOD:
1527 case OP_EODN:
1528 case OP_CIRC:
1529 case OP_DOLL:
1530 case OP_NOT_WORD_BOUNDARY:
1531 case OP_WORD_BOUNDARY:
1532 cc += _pcre_OP_lengths[*cc];
1533 break;
1534
1535 /* Handle literal characters */
1536
1537 case OP_CHAR:
1538 case OP_CHARNC:
1539 case OP_NOT:
1540 branchlength++;
1541 cc += 2;
1542 #ifdef SUPPORT_UTF8
1543 if ((options & PCRE_UTF8) != 0 && cc[-1] >= 0xc0)
1544 cc += _pcre_utf8_table4[cc[-1] & 0x3f];
1545 #endif
1546 break;
1547
1548 /* Handle exact repetitions. The count is already in characters, but we
1549 need to skip over a multibyte character in UTF8 mode. */
1550
1551 case OP_EXACT:
1552 branchlength += GET2(cc,1);
1553 cc += 4;
1554 #ifdef SUPPORT_UTF8
1555 if ((options & PCRE_UTF8) != 0 && cc[-1] >= 0xc0)
1556 cc += _pcre_utf8_table4[cc[-1] & 0x3f];
1557 #endif
1558 break;
1559
1560 case OP_TYPEEXACT:
1561 branchlength += GET2(cc,1);
1562 if (cc[3] == OP_PROP || cc[3] == OP_NOTPROP) cc += 2;
1563 cc += 4;
1564 break;
1565
1566 /* Handle single-char matchers */
1567
1568 case OP_PROP:
1569 case OP_NOTPROP:
1570 cc += 2;
1571 /* Fall through */
1572
1573 case OP_NOT_DIGIT:
1574 case OP_DIGIT:
1575 case OP_NOT_WHITESPACE:
1576 case OP_WHITESPACE:
1577 case OP_NOT_WORDCHAR:
1578 case OP_WORDCHAR:
1579 case OP_ANY:
1580 case OP_ALLANY:
1581 branchlength++;
1582 cc++;
1583 break;
1584
1585 /* The single-byte matcher isn't allowed */
1586
1587 case OP_ANYBYTE:
1588 return -2;
1589
1590 /* Check a class for variable quantification */
1591
1592 #ifdef SUPPORT_UTF8
1593 case OP_XCLASS:
1594 cc += GET(cc, 1) - 33;
1595 /* Fall through */
1596 #endif
1597
1598 case OP_CLASS:
1599 case OP_NCLASS:
1600 cc += 33;
1601
1602 switch (*cc)
1603 {
1604 case OP_CRSTAR:
1605 case OP_CRMINSTAR:
1606 case OP_CRQUERY:
1607 case OP_CRMINQUERY:
1608 return -1;
1609
1610 case OP_CRRANGE:
1611 case OP_CRMINRANGE:
1612 if (GET2(cc,1) != GET2(cc,3)) return -1;
1613 branchlength += GET2(cc,1);
1614 cc += 5;
1615 break;
1616
1617 default:
1618 branchlength++;
1619 }
1620 break;
1621
1622 /* Anything else is variable length */
1623
1624 default:
1625 return -1;
1626 }
1627 }
1628 /* Control never gets here */
1629 }
1630
1631
1632
1633
1634 /*************************************************
1635 * Scan compiled regex for specific bracket *
1636 *************************************************/
1637
1638 /* This little function scans through a compiled pattern until it finds a
1639 capturing bracket with the given number, or, if the number is negative, an
1640 instance of OP_REVERSE for a lookbehind. The function is global in the C sense
1641 so that it can be called from pcre_study() when finding the minimum matching
1642 length.
1643
1644 Arguments:
1645 code points to start of expression
1646 utf8 TRUE in UTF-8 mode
1647 number the required bracket number or negative to find a lookbehind
1648
1649 Returns: pointer to the opcode for the bracket, or NULL if not found
1650 */
1651
1652 const uschar *
1653 _pcre_find_bracket(const uschar *code, BOOL utf8, int number)
1654 {
1655 for (;;)
1656 {
1657 register int c = *code;
1658 if (c == OP_END) return NULL;
1659
1660 /* XCLASS is used for classes that cannot be represented just by a bit
1661 map. This includes negated single high-valued characters. The length in
1662 the table is zero; the actual length is stored in the compiled code. */
1663
1664 if (c == OP_XCLASS) code += GET(code, 1);
1665
1666 /* Handle recursion */
1667
1668 else if (c == OP_REVERSE)
1669 {
1670 if (number < 0) return (uschar *)code;
1671 code += _pcre_OP_lengths[c];
1672 }
1673
1674 /* Handle capturing bracket */
1675
1676 else if (c == OP_CBRA)
1677 {
1678 int n = GET2(code, 1+LINK_SIZE);
1679 if (n == number) return (uschar *)code;
1680 code += _pcre_OP_lengths[c];
1681 }
1682
1683 /* Otherwise, we can get the item's length from the table, except that for
1684 repeated character types, we have to test for \p and \P, which have an extra
1685 two bytes of parameters, and for MARK/PRUNE/SKIP/THEN with an argument, we
1686 must add in its length. */
1687
1688 else
1689 {
1690 switch(c)
1691 {
1692 case OP_TYPESTAR:
1693 case OP_TYPEMINSTAR:
1694 case OP_TYPEPLUS:
1695 case OP_TYPEMINPLUS:
1696 case OP_TYPEQUERY:
1697 case OP_TYPEMINQUERY:
1698 case OP_TYPEPOSSTAR:
1699 case OP_TYPEPOSPLUS:
1700 case OP_TYPEPOSQUERY:
1701 if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
1702 break;
1703
1704 case OP_TYPEUPTO:
1705 case OP_TYPEMINUPTO:
1706 case OP_TYPEEXACT:
1707 case OP_TYPEPOSUPTO:
1708 if (code[3] == OP_PROP || code[3] == OP_NOTPROP) code += 2;
1709 break;
1710
1711 case OP_MARK:
1712 case OP_PRUNE_ARG:
1713 case OP_SKIP_ARG:
1714 case OP_THEN_ARG:
1715 code += code[1];
1716 break;
1717 }
1718
1719 /* Add in the fixed length from the table */
1720
1721 code += _pcre_OP_lengths[c];
1722
1723 /* In UTF-8 mode, opcodes that are followed by a character may be followed by
1724 a multi-byte character. The length in the table is a minimum, so we have to
1725 arrange to skip the extra bytes. */
1726
1727 #ifdef SUPPORT_UTF8
1728 if (utf8) switch(c)
1729 {
1730 case OP_CHAR:
1731 case OP_CHARNC:
1732 case OP_EXACT:
1733 case OP_UPTO:
1734 case OP_MINUPTO:
1735 case OP_POSUPTO:
1736 case OP_STAR:
1737 case OP_MINSTAR:
1738 case OP_POSSTAR:
1739 case OP_PLUS:
1740 case OP_MINPLUS:
1741 case OP_POSPLUS:
1742 case OP_QUERY:
1743 case OP_MINQUERY:
1744 case OP_POSQUERY:
1745 if (code[-1] >= 0xc0) code += _pcre_utf8_table4[code[-1] & 0x3f];
1746 break;
1747 }
1748 #else
1749 (void)(utf8); /* Keep compiler happy by referencing function argument */
1750 #endif
1751 }
1752 }
1753 }
1754
1755
1756
1757 /*************************************************
1758 * Scan compiled regex for recursion reference *
1759 *************************************************/
1760
1761 /* This little function scans through a compiled pattern until it finds an
1762 instance of OP_RECURSE.
1763
1764 Arguments:
1765 code points to start of expression
1766 utf8 TRUE in UTF-8 mode
1767
1768 Returns: pointer to the opcode for OP_RECURSE, or NULL if not found
1769 */
1770
1771 static const uschar *
1772 find_recurse(const uschar *code, BOOL utf8)
1773 {
1774 for (;;)
1775 {
1776 register int c = *code;
1777 if (c == OP_END) return NULL;
1778 if (c == OP_RECURSE) return code;
1779
1780 /* XCLASS is used for classes that cannot be represented just by a bit
1781 map. This includes negated single high-valued characters. The length in
1782 the table is zero; the actual length is stored in the compiled code. */
1783
1784 if (c == OP_XCLASS) code += GET(code, 1);
1785
1786 /* Otherwise, we can get the item's length from the table, except that for
1787 repeated character types, we have to test for \p and \P, which have an extra
1788 two bytes of parameters, and for MARK/PRUNE/SKIP/THEN with an argument, we
1789 must add in its length. */
1790
1791 else
1792 {
1793 switch(c)
1794 {
1795 case OP_TYPESTAR:
1796 case OP_TYPEMINSTAR:
1797 case OP_TYPEPLUS:
1798 case OP_TYPEMINPLUS:
1799 case OP_TYPEQUERY:
1800 case OP_TYPEMINQUERY:
1801 case OP_TYPEPOSSTAR:
1802 case OP_TYPEPOSPLUS:
1803 case OP_TYPEPOSQUERY:
1804 if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
1805 break;
1806
1807 case OP_TYPEPOSUPTO:
1808 case OP_TYPEUPTO:
1809 case OP_TYPEMINUPTO:
1810 case OP_TYPEEXACT:
1811 if (code[3] == OP_PROP || code[3] == OP_NOTPROP) code += 2;
1812 break;
1813
1814 case OP_MARK:
1815 case OP_PRUNE_ARG:
1816 case OP_SKIP_ARG:
1817 case OP_THEN_ARG:
1818 code += code[1];
1819 break;
1820 }
1821
1822 /* Add in the fixed length from the table */
1823
1824 code += _pcre_OP_lengths[c];
1825
1826 /* In UTF-8 mode, opcodes that are followed by a character may be followed
1827 by a multi-byte character. The length in the table is a minimum, so we have
1828 to arrange to skip the extra bytes. */
1829
1830 #ifdef SUPPORT_UTF8
1831 if (utf8) switch(c)
1832 {
1833 case OP_CHAR:
1834 case OP_CHARNC:
1835 case OP_EXACT:
1836 case OP_UPTO:
1837 case OP_MINUPTO:
1838 case OP_POSUPTO:
1839 case OP_STAR:
1840 case OP_MINSTAR:
1841 case OP_POSSTAR:
1842 case OP_PLUS:
1843 case OP_MINPLUS:
1844 case OP_POSPLUS:
1845 case OP_QUERY:
1846 case OP_MINQUERY:
1847 case OP_POSQUERY:
1848 if (code[-1] >= 0xc0) code += _pcre_utf8_table4[code[-1] & 0x3f];
1849 break;
1850 }
1851 #else
1852 (void)(utf8); /* Keep compiler happy by referencing function argument */
1853 #endif
1854 }
1855 }
1856 }
1857
1858
1859
1860 /*************************************************
1861 * Scan compiled branch for non-emptiness *
1862 *************************************************/
1863
1864 /* This function scans through a branch of a compiled pattern to see whether it
1865 can match the empty string or not. It is called from could_be_empty()
1866 below and from compile_branch() when checking for an unlimited repeat of a
1867 group that can match nothing. Note that first_significant_code() skips over
1868 backward and negative forward assertions when its final argument is TRUE. If we
1869 hit an unclosed bracket, we return "empty" - this means we've struck an inner
1870 bracket whose current branch will already have been scanned.
1871
1872 Arguments:
1873 code points to start of search
1874 endcode points to where to stop
1875 utf8 TRUE if in UTF8 mode
1876 cd contains pointers to tables etc.
1877
1878 Returns: TRUE if what is matched could be empty
1879 */
1880
1881 static BOOL
1882 could_be_empty_branch(const uschar *code, const uschar *endcode, BOOL utf8,
1883 compile_data *cd)
1884 {
1885 register int c;
1886 for (code = first_significant_code(code + _pcre_OP_lengths[*code], NULL, 0, TRUE);
1887 code < endcode;
1888 code = first_significant_code(code + _pcre_OP_lengths[c], NULL, 0, TRUE))
1889 {
1890 const uschar *ccode;
1891
1892 c = *code;
1893
1894 /* Skip over forward assertions; the other assertions are skipped by
1895 first_significant_code() with a TRUE final argument. */
1896
1897 if (c == OP_ASSERT)
1898 {
1899 do code += GET(code, 1); while (*code == OP_ALT);
1900 c = *code;
1901 continue;
1902 }
1903
1904 /* Groups with zero repeats can of course be empty; skip them. */
1905
1906 if (c == OP_BRAZERO || c == OP_BRAMINZERO || c == OP_SKIPZERO)
1907 {
1908 code += _pcre_OP_lengths[c];
1909 do code += GET(code, 1); while (*code == OP_ALT);
1910 c = *code;
1911 continue;
1912 }
1913
1914 /* For a recursion/subroutine call, if its end has been reached, which
1915 implies a subroutine call, we can scan it. */
1916
1917 if (c == OP_RECURSE)
1918 {
1919 BOOL empty_branch = FALSE;
1920 const uschar *scode = cd->start_code + GET(code, 1);
1921 if (GET(scode, 1) == 0) return TRUE; /* Unclosed */
1922 do
1923 {
1924 if (could_be_empty_branch(scode, endcode, utf8, cd))
1925 {
1926 empty_branch = TRUE;
1927 break;
1928 }
1929 scode += GET(scode, 1);
1930 }
1931 while (*scode == OP_ALT);
1932 if (!empty_branch) return FALSE; /* All branches are non-empty */
1933 continue;
1934 }
1935
1936 /* For other groups, scan the branches. */
1937
1938 if (c == OP_BRA || c == OP_CBRA || c == OP_ONCE || c == OP_COND)
1939 {
1940 BOOL empty_branch;
1941 if (GET(code, 1) == 0) return TRUE; /* Hit unclosed bracket */
1942
1943 /* If a conditional group has only one branch, there is a second, implied,
1944 empty branch, so just skip over the conditional, because it could be empty.
1945 Otherwise, scan the individual branches of the group. */
1946
1947 if (c == OP_COND && code[GET(code, 1)] != OP_ALT)
1948 code += GET(code, 1);
1949 else
1950 {
1951 empty_branch = FALSE;
1952 do
1953 {
1954 if (!empty_branch && could_be_empty_branch(code, endcode, utf8, cd))
1955 empty_branch = TRUE;
1956 code += GET(code, 1);
1957 }
1958 while (*code == OP_ALT);
1959 if (!empty_branch) return FALSE; /* All branches are non-empty */
1960 }
1961
1962 c = *code;
1963 continue;
1964 }
1965
1966 /* Handle the other opcodes */
1967
1968 switch (c)
1969 {
1970 /* Check for quantifiers after a class. XCLASS is used for classes that
1971 cannot be represented just by a bit map. This includes negated single
1972 high-valued characters. The length in _pcre_OP_lengths[] is zero; the
1973 actual length is stored in the compiled code, so we must update "code"
1974 here. */
1975
1976 #ifdef SUPPORT_UTF8
1977 case OP_XCLASS:
1978 ccode = code += GET(code, 1);
1979 goto CHECK_CLASS_REPEAT;
1980 #endif
1981
1982 case OP_CLASS:
1983 case OP_NCLASS:
1984 ccode = code + 33;
1985
1986 #ifdef SUPPORT_UTF8
1987 CHECK_CLASS_REPEAT:
1988 #endif
1989
1990 switch (*ccode)
1991 {
1992 case OP_CRSTAR: /* These could be empty; continue */
1993 case OP_CRMINSTAR:
1994 case OP_CRQUERY:
1995 case OP_CRMINQUERY:
1996 break;
1997
1998 default: /* Non-repeat => class must match */
1999 case OP_CRPLUS: /* These repeats aren't empty */
2000 case OP_CRMINPLUS:
2001 return FALSE;
2002
2003 case OP_CRRANGE:
2004 case OP_CRMINRANGE:
2005 if (GET2(ccode, 1) > 0) return FALSE; /* Minimum > 0 */
2006 break;
2007 }
2008 break;
2009
2010 /* Opcodes that must match a character */
2011
2012 case OP_PROP:
2013 case OP_NOTPROP:
2014 case OP_EXTUNI:
2015 case OP_NOT_DIGIT:
2016 case OP_DIGIT:
2017 case OP_NOT_WHITESPACE:
2018 case OP_WHITESPACE:
2019 case OP_NOT_WORDCHAR:
2020 case OP_WORDCHAR:
2021 case OP_ANY:
2022 case OP_ALLANY:
2023 case OP_ANYBYTE:
2024 case OP_CHAR:
2025 case OP_CHARNC:
2026 case OP_NOT:
2027 case OP_PLUS:
2028 case OP_MINPLUS:
2029 case OP_POSPLUS:
2030 case OP_EXACT:
2031 case OP_NOTPLUS:
2032 case OP_NOTMINPLUS:
2033 case OP_NOTPOSPLUS:
2034 case OP_NOTEXACT:
2035 case OP_TYPEPLUS:
2036 case OP_TYPEMINPLUS:
2037 case OP_TYPEPOSPLUS:
2038 case OP_TYPEEXACT:
2039 return FALSE;
2040
2041 /* These are going to continue, as they may be empty, but we have to
2042 fudge the length for the \p and \P cases. */
2043
2044 case OP_TYPESTAR:
2045 case OP_TYPEMINSTAR:
2046 case OP_TYPEPOSSTAR:
2047 case OP_TYPEQUERY:
2048 case OP_TYPEMINQUERY:
2049 case OP_TYPEPOSQUERY:
2050 if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
2051 break;
2052
2053 /* Same for these */
2054
2055 case OP_TYPEUPTO:
2056 case OP_TYPEMINUPTO:
2057 case OP_TYPEPOSUPTO:
2058 if (code[3] == OP_PROP || code[3] == OP_NOTPROP) code += 2;
2059 break;
2060
2061 /* End of branch */
2062
2063 case OP_KET:
2064 case OP_KETRMAX:
2065 case OP_KETRMIN:
2066 case OP_ALT:
2067 return TRUE;
2068
2069 /* In UTF-8 mode, STAR, MINSTAR, POSSTAR, QUERY, MINQUERY, POSQUERY, UPTO,
2070 MINUPTO, and POSUPTO may be followed by a multibyte character */
2071
2072 #ifdef SUPPORT_UTF8
2073 case OP_STAR:
2074 case OP_MINSTAR:
2075 case OP_POSSTAR:
2076 case OP_QUERY:
2077 case OP_MINQUERY:
2078 case OP_POSQUERY:
2079 if (utf8 && code[1] >= 0xc0) code += _pcre_utf8_table4[code[1] & 0x3f];
2080 break;
2081
2082 case OP_UPTO:
2083 case OP_MINUPTO:
2084 case OP_POSUPTO:
2085 if (utf8 && code[3] >= 0xc0) code += _pcre_utf8_table4[code[3] & 0x3f];
2086 break;
2087 #endif
2088
2089 /* MARK, and PRUNE/SKIP/THEN with an argument must skip over the argument
2090 string. */
2091
2092 case OP_MARK:
2093 case OP_PRUNE_ARG:
2094 case OP_SKIP_ARG:
2095 case OP_THEN_ARG:
2096 code += code[1];
2097 break;
2098
2099 /* None of the remaining opcodes are required to match a character. */
2100
2101 default:
2102 break;
2103 }
2104 }
2105
2106 return TRUE;
2107 }
2108
2109
2110
2111 /*************************************************
2112 * Scan compiled regex for non-emptiness *
2113 *************************************************/
2114
2115 /* This function is called to check for left recursive calls. We want to check
2116 the current branch of the current pattern to see if it could match the empty
2117 string. If it could, we must look outwards for branches at other levels,
2118 stopping when we pass beyond the bracket which is the subject of the recursion.
2119
2120 Arguments:
2121 code points to start of the recursion
2122 endcode points to where to stop (current RECURSE item)
2123 bcptr points to the chain of current (unclosed) branch starts
2124 utf8 TRUE if in UTF-8 mode
2125 cd pointers to tables etc
2126
2127 Returns: TRUE if what is matched could be empty
2128 */
2129
2130 static BOOL
2131 could_be_empty(const uschar *code, const uschar *endcode, branch_chain *bcptr,
2132 BOOL utf8, compile_data *cd)
2133 {
2134 while (bcptr != NULL && bcptr->current_branch >= code)
2135 {
2136 if (!could_be_empty_branch(bcptr->current_branch, endcode, utf8, cd))
2137 return FALSE;
2138 bcptr = bcptr->outer;
2139 }
2140 return TRUE;
2141 }
2142
2143
2144
2145 /*************************************************
2146 * Check for POSIX class syntax *
2147 *************************************************/
2148
2149 /* This function is called when the sequence "[:" or "[." or "[=" is
2150 encountered in a character class. It checks whether this is followed by a
2151 sequence of characters terminated by a matching ":]" or ".]" or "=]". If we
2152 reach an unescaped ']' without the special preceding character, return FALSE.
2153
2154 Originally, this function only recognized a sequence of letters between the
2155 terminators, but it seems that Perl recognizes any sequence of characters,
2156 though of course unknown POSIX names are subsequently rejected. Perl gives an
2157 "Unknown POSIX class" error for [:f\oo:] for example, where previously PCRE
2158 didn't consider this to be a POSIX class. Likewise for [:1234:].
2159
2160 The problem in trying to be exactly like Perl is in the handling of escapes. We
2161 have to be sure that [abc[:x\]pqr] is *not* treated as containing a POSIX
2162 class, but [abc[:x\]pqr:]] is (so that an error can be generated). The code
2163 below handles the special case of \], but does not try to do any other escape
2164 processing. This makes it different from Perl for cases such as [:l\ower:]
2165 where Perl recognizes it as the POSIX class "lower" but PCRE does not recognize
2166 "l\ower". This is a lesser evil that not diagnosing bad classes when Perl does,
2167 I think.
2168
2169 Arguments:
2170 ptr pointer to the initial [
2171 endptr where to return the end pointer
2172
2173 Returns: TRUE or FALSE
2174 */
2175
2176 static BOOL
2177 check_posix_syntax(const uschar *ptr, const uschar **endptr)
2178 {
2179 int terminator; /* Don't combine these lines; the Solaris cc */
2180 terminator = *(++ptr); /* compiler warns about "non-constant" initializer. */
2181 for (++ptr; *ptr != 0; ptr++)
2182 {
2183 if (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET) ptr++; else
2184 {
2185 if (*ptr == CHAR_RIGHT_SQUARE_BRACKET) return FALSE;
2186 if (*ptr == terminator && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET)
2187 {
2188 *endptr = ptr;
2189 return TRUE;
2190 }
2191 }
2192 }
2193 return FALSE;
2194 }
2195
2196
2197
2198
2199 /*************************************************
2200 * Check POSIX class name *
2201 *************************************************/
2202
2203 /* This function is called to check the name given in a POSIX-style class entry
2204 such as [:alnum:].
2205
2206 Arguments:
2207 ptr points to the first letter
2208 len the length of the name
2209
2210 Returns: a value representing the name, or -1 if unknown
2211 */
2212
2213 static int
2214 check_posix_name(const uschar *ptr, int len)
2215 {
2216 const char *pn = posix_names;
2217 register int yield = 0;
2218 while (posix_name_lengths[yield] != 0)
2219 {
2220 if (len == posix_name_lengths[yield] &&
2221 strncmp((const char *)ptr, pn, len) == 0) return yield;
2222 pn += posix_name_lengths[yield] + 1;
2223 yield++;
2224 }
2225 return -1;
2226 }
2227
2228
2229 /*************************************************
2230 * Adjust OP_RECURSE items in repeated group *
2231 *************************************************/
2232
2233 /* OP_RECURSE items contain an offset from the start of the regex to the group
2234 that is referenced. This means that groups can be replicated for fixed
2235 repetition simply by copying (because the recursion is allowed to refer to
2236 earlier groups that are outside the current group). However, when a group is
2237 optional (i.e. the minimum quantifier is zero), OP_BRAZERO or OP_SKIPZERO is
2238 inserted before it, after it has been compiled. This means that any OP_RECURSE
2239 items within it that refer to the group itself or any contained groups have to
2240 have their offsets adjusted. That one of the jobs of this function. Before it
2241 is called, the partially compiled regex must be temporarily terminated with
2242 OP_END.
2243
2244 This function has been extended with the possibility of forward references for
2245 recursions and subroutine calls. It must also check the list of such references
2246 for the group we are dealing with. If it finds that one of the recursions in
2247 the current group is on this list, it adjusts the offset in the list, not the
2248 value in the reference (which is a group number).
2249
2250 Arguments:
2251 group points to the start of the group
2252 adjust the amount by which the group is to be moved
2253 utf8 TRUE in UTF-8 mode
2254 cd contains pointers to tables etc.
2255 save_hwm the hwm forward reference pointer at the start of the group
2256
2257 Returns: nothing
2258 */
2259
2260 static void
2261 adjust_recurse(uschar *group, int adjust, BOOL utf8, compile_data *cd,
2262 uschar *save_hwm)
2263 {
2264 uschar *ptr = group;
2265
2266 while ((ptr = (uschar *)find_recurse(ptr, utf8)) != NULL)
2267 {
2268 int offset;
2269 uschar *hc;
2270
2271 /* See if this recursion is on the forward reference list. If so, adjust the
2272 reference. */
2273
2274 for (hc = save_hwm; hc < cd->hwm; hc += LINK_SIZE)
2275 {
2276 offset = GET(hc, 0);
2277 if (cd->start_code + offset == ptr + 1)
2278 {
2279 PUT(hc, 0, offset + adjust);
2280 break;
2281 }
2282 }
2283
2284 /* Otherwise, adjust the recursion offset if it's after the start of this
2285 group. */
2286
2287 if (hc >= cd->hwm)
2288 {
2289 offset = GET(ptr, 1);
2290 if (cd->start_code + offset >= group) PUT(ptr, 1, offset + adjust);
2291 }
2292
2293 ptr += 1 + LINK_SIZE;
2294 }
2295 }
2296
2297
2298
2299 /*************************************************
2300 * Insert an automatic callout point *
2301 *************************************************/
2302
2303 /* This function is called when the PCRE_AUTO_CALLOUT option is set, to insert
2304 callout points before each pattern item.
2305
2306 Arguments:
2307 code current code pointer
2308 ptr current pattern pointer
2309 cd pointers to tables etc
2310
2311 Returns: new code pointer
2312 */
2313
2314 static uschar *
2315 auto_callout(uschar *code, const uschar *ptr, compile_data *cd)
2316 {
2317 *code++ = OP_CALLOUT;
2318 *code++ = 255;
2319 PUT(code, 0, (int)(ptr - cd->start_pattern)); /* Pattern offset */
2320 PUT(code, LINK_SIZE, 0); /* Default length */
2321 return code + 2*LINK_SIZE;
2322 }
2323
2324
2325
2326 /*************************************************
2327 * Complete a callout item *
2328 *************************************************/
2329
2330 /* A callout item contains the length of the next item in the pattern, which
2331 we can't fill in till after we have reached the relevant point. This is used
2332 for both automatic and manual callouts.
2333
2334 Arguments:
2335 previous_callout points to previous callout item
2336 ptr current pattern pointer
2337 cd pointers to tables etc
2338
2339 Returns: nothing
2340 */
2341
2342 static void
2343 complete_callout(uschar *previous_callout, const uschar *ptr, compile_data *cd)
2344 {
2345 int length = (int)(ptr - cd->start_pattern - GET(previous_callout, 2));
2346 PUT(previous_callout, 2 + LINK_SIZE, length);
2347 }
2348
2349
2350
2351 #ifdef SUPPORT_UCP
2352 /*************************************************
2353 * Get othercase range *
2354 *************************************************/
2355
2356 /* This function is passed the start and end of a class range, in UTF-8 mode
2357 with UCP support. It searches up the characters, looking for internal ranges of
2358 characters in the "other" case. Each call returns the next one, updating the
2359 start address.
2360
2361 Arguments:
2362 cptr points to starting character value; updated
2363 d end value
2364 ocptr where to put start of othercase range
2365 odptr where to put end of othercase range
2366
2367 Yield: TRUE when range returned; FALSE when no more
2368 */
2369
2370 static BOOL
2371 get_othercase_range(unsigned int *cptr, unsigned int d, unsigned int *ocptr,
2372 unsigned int *odptr)
2373 {
2374 unsigned int c, othercase, next;
2375
2376 for (c = *cptr; c <= d; c++)
2377 { if ((othercase = UCD_OTHERCASE(c)) != c) break; }
2378
2379 if (c > d) return FALSE;
2380
2381 *ocptr = othercase;
2382 next = othercase + 1;
2383
2384 for (++c; c <= d; c++)
2385 {
2386 if (UCD_OTHERCASE(c) != next) break;
2387 next++;
2388 }
2389
2390 *odptr = next - 1;
2391 *cptr = c;
2392
2393 return TRUE;
2394 }
2395
2396
2397
2398 /*************************************************
2399 * Check a character and a property *
2400 *************************************************/
2401
2402 /* This function is called by check_auto_possessive() when a property item
2403 is adjacent to a fixed character.
2404
2405 Arguments:
2406 c the character
2407 ptype the property type
2408 pdata the data for the type
2409 negated TRUE if it's a negated property (\P or \p{^)
2410
2411 Returns: TRUE if auto-possessifying is OK
2412 */
2413
2414 static BOOL
2415 check_char_prop(int c, int ptype, int pdata, BOOL negated)
2416 {
2417 const ucd_record *prop = GET_UCD(c);
2418 switch(ptype)
2419 {
2420 case PT_LAMP:
2421 return (prop->chartype == ucp_Lu ||
2422 prop->chartype == ucp_Ll ||
2423 prop->chartype == ucp_Lt) == negated;
2424
2425 case PT_GC:
2426 return (pdata == _pcre_ucp_gentype[prop->chartype]) == negated;
2427
2428 case PT_PC:
2429 return (pdata == prop->chartype) == negated;
2430
2431 case PT_SC:
2432 return (pdata == prop->script) == negated;
2433
2434 /* These are specials */
2435
2436 case PT_ALNUM:
2437 return (_pcre_ucp_gentype[prop->chartype] == ucp_L ||
2438 _pcre_ucp_gentype[prop->chartype] == ucp_N) == negated;
2439
2440 case PT_SPACE: /* Perl space */
2441 return (_pcre_ucp_gentype[prop->chartype] == ucp_Z ||
2442 c == CHAR_HT || c == CHAR_NL || c == CHAR_FF || c == CHAR_CR)
2443 == negated;
2444
2445 case PT_PXSPACE: /* POSIX space */
2446 return (_pcre_ucp_gentype[prop->chartype] == ucp_Z ||
2447 c == CHAR_HT || c == CHAR_NL || c == CHAR_VT ||
2448 c == CHAR_FF || c == CHAR_CR)
2449 == negated;
2450
2451 case PT_WORD:
2452 return (_pcre_ucp_gentype[prop->chartype] == ucp_L ||
2453 _pcre_ucp_gentype[prop->chartype] == ucp_N ||
2454 c == CHAR_UNDERSCORE) == negated;
2455 }
2456 return FALSE;
2457 }
2458 #endif /* SUPPORT_UCP */
2459
2460
2461
2462 /*************************************************
2463 * Check if auto-possessifying is possible *
2464 *************************************************/
2465
2466 /* This function is called for unlimited repeats of certain items, to see
2467 whether the next thing could possibly match the repeated item. If not, it makes
2468 sense to automatically possessify the repeated item.
2469
2470 Arguments:
2471 previous pointer to the repeated opcode
2472 utf8 TRUE in UTF-8 mode
2473 ptr next character in pattern
2474 options options bits
2475 cd contains pointers to tables etc.
2476
2477 Returns: TRUE if possessifying is wanted
2478 */
2479
2480 static BOOL
2481 check_auto_possessive(const uschar *previous, BOOL utf8, const uschar *ptr,
2482 int options, compile_data *cd)
2483 {
2484 int c, next;
2485 int op_code = *previous++;
2486
2487 /* Skip whitespace and comments in extended mode */
2488
2489 if ((options & PCRE_EXTENDED) != 0)
2490 {
2491 for (;;)
2492 {
2493 while ((cd->ctypes[*ptr] & ctype_space) != 0) ptr++;
2494 if (*ptr == CHAR_NUMBER_SIGN)
2495 {
2496 while (*(++ptr) != 0)
2497 if (IS_NEWLINE(ptr)) { ptr += cd->nllen; break; }
2498 }
2499 else break;
2500 }
2501 }
2502
2503 /* If the next item is one that we can handle, get its value. A non-negative
2504 value is a character, a negative value is an escape value. */
2505
2506 if (*ptr == CHAR_BACKSLASH)
2507 {
2508 int temperrorcode = 0;
2509 next = check_escape(&ptr, &temperrorcode, cd->bracount, options, FALSE);
2510 if (temperrorcode != 0) return FALSE;
2511 ptr++; /* Point after the escape sequence */
2512 }
2513
2514 else if ((cd->ctypes[*ptr] & ctype_meta) == 0)
2515 {
2516 #ifdef SUPPORT_UTF8
2517 if (utf8) { GETCHARINC(next, ptr); } else
2518 #endif
2519 next = *ptr++;
2520 }
2521
2522 else return FALSE;
2523
2524 /* Skip whitespace and comments in extended mode */
2525
2526 if ((options & PCRE_EXTENDED) != 0)
2527 {
2528 for (;;)
2529 {
2530 while ((cd->ctypes[*ptr] & ctype_space) != 0) ptr++;
2531 if (*ptr == CHAR_NUMBER_SIGN)
2532 {
2533 while (*(++ptr) != 0)
2534 if (IS_NEWLINE(ptr)) { ptr += cd->nllen; break; }
2535 }
2536 else break;
2537 }
2538 }
2539
2540 /* If the next thing is itself optional, we have to give up. */
2541
2542 if (*ptr == CHAR_ASTERISK || *ptr == CHAR_QUESTION_MARK ||
2543 strncmp((char *)ptr, STR_LEFT_CURLY_BRACKET STR_0 STR_COMMA, 3) == 0)
2544 return FALSE;
2545
2546 /* Now compare the next item with the previous opcode. First, handle cases when
2547 the next item is a character. */
2548
2549 if (next >= 0) switch(op_code)
2550 {
2551 case OP_CHAR:
2552 #ifdef SUPPORT_UTF8
2553 GETCHARTEST(c, previous);
2554 #else
2555 c = *previous;
2556 #endif
2557 return c != next;
2558
2559 /* For CHARNC (caseless character) we must check the other case. If we have
2560 Unicode property support, we can use it to test the other case of
2561 high-valued characters. */
2562
2563 case OP_CHARNC:
2564 #ifdef SUPPORT_UTF8
2565 GETCHARTEST(c, previous);
2566 #else
2567 c = *previous;
2568 #endif
2569 if (c == next) return FALSE;
2570 #ifdef SUPPORT_UTF8
2571 if (utf8)
2572 {
2573 unsigned int othercase;
2574 if (next < 128) othercase = cd->fcc[next]; else
2575 #ifdef SUPPORT_UCP
2576 othercase = UCD_OTHERCASE((unsigned int)next);
2577 #else
2578 othercase = NOTACHAR;
2579 #endif
2580 return (unsigned int)c != othercase;
2581 }
2582 else
2583 #endif /* SUPPORT_UTF8 */
2584 return (c != cd->fcc[next]); /* Non-UTF-8 mode */
2585
2586 /* For OP_NOT, its data is always a single-byte character. */
2587
2588 case OP_NOT:
2589 if ((c = *previous) == next) return TRUE;
2590 if ((options & PCRE_CASELESS) == 0) return FALSE;
2591 #ifdef SUPPORT_UTF8
2592 if (utf8)
2593 {
2594 unsigned int othercase;
2595 if (next < 128) othercase = cd->fcc[next]; else
2596 #ifdef SUPPORT_UCP
2597 othercase = UCD_OTHERCASE(next);
2598 #else
2599 othercase = NOTACHAR;
2600 #endif
2601 return (unsigned int)c == othercase;
2602 }
2603 else
2604 #endif /* SUPPORT_UTF8 */
2605 return (c == cd->fcc[next]); /* Non-UTF-8 mode */
2606
2607 /* Note that OP_DIGIT etc. are generated only when PCRE_UCP is *not* set.
2608 When it is set, \d etc. are converted into OP_(NOT_)PROP codes. */
2609
2610 case OP_DIGIT:
2611 return next > 127 || (cd->ctypes[next] & ctype_digit) == 0;
2612
2613 case OP_NOT_DIGIT:
2614 return next <= 127 && (cd->ctypes[next] & ctype_digit) != 0;
2615
2616 case OP_WHITESPACE:
2617 return next > 127 || (cd->ctypes[next] & ctype_space) == 0;
2618
2619 case OP_NOT_WHITESPACE:
2620 return next <= 127 && (cd->ctypes[next] & ctype_space) != 0;
2621
2622 case OP_WORDCHAR:
2623 return next > 127 || (cd->ctypes[next] & ctype_word) == 0;
2624
2625 case OP_NOT_WORDCHAR:
2626 return next <= 127 && (cd->ctypes[next] & ctype_word) != 0;
2627
2628 case OP_HSPACE:
2629 case OP_NOT_HSPACE:
2630 switch(next)
2631 {
2632 case 0x09:
2633 case 0x20:
2634 case 0xa0:
2635 case 0x1680:
2636 case 0x180e:
2637 case 0x2000:
2638 case 0x2001:
2639 case 0x2002:
2640 case 0x2003:
2641 case 0x2004:
2642 case 0x2005:
2643 case 0x2006:
2644 case 0x2007:
2645 case 0x2008:
2646 case 0x2009:
2647 case 0x200A:
2648 case 0x202f:
2649 case 0x205f:
2650 case 0x3000:
2651 return op_code == OP_NOT_HSPACE;
2652 default:
2653 return op_code != OP_NOT_HSPACE;
2654 }
2655
2656 case OP_ANYNL:
2657 case OP_VSPACE:
2658 case OP_NOT_VSPACE:
2659 switch(next)
2660 {
2661 case 0x0a:
2662 case 0x0b:
2663 case 0x0c:
2664 case 0x0d:
2665 case 0x85:
2666 case 0x2028:
2667 case 0x2029:
2668 return op_code == OP_NOT_VSPACE;
2669 default:
2670 return op_code != OP_NOT_VSPACE;
2671 }
2672
2673 #ifdef SUPPORT_UCP
2674 case OP_PROP:
2675 return check_char_prop(next, previous[0], previous[1], FALSE);
2676
2677 case OP_NOTPROP:
2678 return check_char_prop(next, previous[0], previous[1], TRUE);
2679 #endif
2680
2681 default:
2682 return FALSE;
2683 }
2684
2685
2686 /* Handle the case when the next item is \d, \s, etc. Note that when PCRE_UCP
2687 is set, \d turns into ESC_du rather than ESC_d, etc., so ESC_d etc. are
2688 generated only when PCRE_UCP is *not* set, that is, when only ASCII
2689 characteristics are recognized. Similarly, the opcodes OP_DIGIT etc. are
2690 replaced by OP_PROP codes when PCRE_UCP is set. */
2691
2692 switch(op_code)
2693 {
2694 case OP_CHAR:
2695 case OP_CHARNC:
2696 #ifdef SUPPORT_UTF8
2697 GETCHARTEST(c, previous);
2698 #else
2699 c = *previous;
2700 #endif
2701 switch(-next)
2702 {
2703 case ESC_d:
2704 return c > 127 || (cd->ctypes[c] & ctype_digit) == 0;
2705
2706 case ESC_D:
2707 return c <= 127 && (cd->ctypes[c] & ctype_digit) != 0;
2708
2709 case ESC_s:
2710 return c > 127 || (cd->ctypes[c] & ctype_space) == 0;
2711
2712 case ESC_S:
2713 return c <= 127 && (cd->ctypes[c] & ctype_space) != 0;
2714
2715 case ESC_w:
2716 return c > 127 || (cd->ctypes[c] & ctype_word) == 0;
2717
2718 case ESC_W:
2719 return c <= 127 && (cd->ctypes[c] & ctype_word) != 0;
2720
2721 case ESC_h:
2722 case ESC_H:
2723 switch(c)
2724 {
2725 case 0x09:
2726 case 0x20:
2727 case 0xa0:
2728 case 0x1680:
2729 case 0x180e:
2730 case 0x2000:
2731 case 0x2001:
2732 case 0x2002:
2733 case 0x2003:
2734 case 0x2004:
2735 case 0x2005:
2736 case 0x2006:
2737 case 0x2007:
2738 case 0x2008:
2739 case 0x2009:
2740 case 0x200A:
2741 case 0x202f:
2742 case 0x205f:
2743 case 0x3000:
2744 return -next != ESC_h;
2745 default:
2746 return -next == ESC_h;
2747 }
2748
2749 case ESC_v:
2750 case ESC_V:
2751 switch(c)
2752 {
2753 case 0x0a:
2754 case 0x0b:
2755 case 0x0c:
2756 case 0x0d:
2757 case 0x85:
2758 case 0x2028:
2759 case 0x2029:
2760 return -next != ESC_v;
2761 default:
2762 return -next == ESC_v;
2763 }
2764
2765 /* When PCRE_UCP is set, these values get generated for \d etc. Find
2766 their substitutions and process them. The result will always be either
2767 -ESC_p or -ESC_P. Then fall through to process those values. */
2768
2769 #ifdef SUPPORT_UCP
2770 case ESC_du:
2771 case ESC_DU:
2772 case ESC_wu:
2773 case ESC_WU:
2774 case ESC_su:
2775 case ESC_SU:
2776 {
2777 int temperrorcode = 0;
2778 ptr = substitutes[-next - ESC_DU];
2779 next = check_escape(&ptr, &temperrorcode, 0, options, FALSE);
2780 if (temperrorcode != 0) return FALSE;
2781 ptr++; /* For compatibility */
2782 }
2783 /* Fall through */
2784
2785 case ESC_p:
2786 case ESC_P:
2787 {
2788 int ptype, pdata, errorcodeptr;
2789 BOOL negated;
2790
2791 ptr--; /* Make ptr point at the p or P */
2792 ptype = get_ucp(&ptr, &negated, &pdata, &errorcodeptr);
2793 if (ptype < 0) return FALSE;
2794 ptr++; /* Point past the final curly ket */
2795
2796 /* If the property item is optional, we have to give up. (When generated
2797 from \d etc by PCRE_UCP, this test will have been applied much earlier,
2798 to the original \d etc. At this point, ptr will point to a zero byte. */
2799
2800 if (*ptr == CHAR_ASTERISK || *ptr == CHAR_QUESTION_MARK ||
2801 strncmp((char *)ptr, STR_LEFT_CURLY_BRACKET STR_0 STR_COMMA, 3) == 0)
2802 return FALSE;
2803
2804 /* Do the property check. */
2805
2806 return check_char_prop(c, ptype, pdata, (next == -ESC_P) != negated);
2807 }
2808 #endif
2809
2810 default:
2811 return FALSE;
2812 }
2813
2814 /* In principle, support for Unicode properties should be integrated here as
2815 well. It means re-organizing the above code so as to get hold of the property
2816 values before switching on the op-code. However, I wonder how many patterns
2817 combine ASCII \d etc with Unicode properties? (Note that if PCRE_UCP is set,
2818 these op-codes are never generated.) */
2819
2820 case OP_DIGIT:
2821 return next == -ESC_D || next == -ESC_s || next == -ESC_W ||
2822 next == -ESC_h || next == -ESC_v || next == -ESC_R;
2823
2824 case OP_NOT_DIGIT:
2825 return next == -ESC_d;
2826
2827 case OP_WHITESPACE:
2828 return next == -ESC_S || next == -ESC_d || next == -ESC_w || next == -ESC_R;
2829
2830 case OP_NOT_WHITESPACE:
2831 return next == -ESC_s || next == -ESC_h || next == -ESC_v;
2832
2833 case OP_HSPACE:
2834 return next == -ESC_S || next == -ESC_H || next == -ESC_d ||
2835 next == -ESC_w || next == -ESC_v || next == -ESC_R;
2836
2837 case OP_NOT_HSPACE:
2838 return next == -ESC_h;
2839
2840 /* Can't have \S in here because VT matches \S (Perl anomaly) */
2841 case OP_ANYNL:
2842 case OP_VSPACE:
2843 return next == -ESC_V || next == -ESC_d || next == -ESC_w;
2844
2845 case OP_NOT_VSPACE:
2846 return next == -ESC_v || next == -ESC_R;
2847
2848 case OP_WORDCHAR:
2849 return next == -ESC_W || next == -ESC_s || next == -ESC_h ||
2850 next == -ESC_v || next == -ESC_R;
2851
2852 case OP_NOT_WORDCHAR:
2853 return next == -ESC_w || next == -ESC_d;
2854
2855 default:
2856 return FALSE;
2857 }
2858
2859 /* Control does not reach here */
2860 }
2861
2862
2863
2864 /*************************************************
2865 * Compile one branch *
2866 *************************************************/
2867
2868 /* Scan the pattern, compiling it into the a vector. If the options are
2869 changed during the branch, the pointer is used to change the external options
2870 bits. This function is used during the pre-compile phase when we are trying
2871 to find out the amount of memory needed, as well as during the real compile
2872 phase. The value of lengthptr distinguishes the two phases.
2873
2874 Arguments:
2875 optionsptr pointer to the option bits
2876 codeptr points to the pointer to the current code point
2877 ptrptr points to the current pattern pointer
2878 errorcodeptr points to error code variable
2879 firstbyteptr set to initial literal character, or < 0 (REQ_UNSET, REQ_NONE)
2880 reqbyteptr set to the last literal character required, else < 0
2881 bcptr points to current branch chain
2882 cd contains pointers to tables etc.
2883 lengthptr NULL during the real compile phase
2884 points to length accumulator during pre-compile phase
2885
2886 Returns: TRUE on success
2887 FALSE, with *errorcodeptr set non-zero on error
2888 */
2889
2890 static BOOL
2891 compile_branch(int *optionsptr, uschar **codeptr, const uschar **ptrptr,
2892 int *errorcodeptr, int *firstbyteptr, int *reqbyteptr, branch_chain *bcptr,
2893 compile_data *cd, int *lengthptr)
2894 {
2895 int repeat_type, op_type;
2896 int repeat_min = 0, repeat_max = 0; /* To please picky compilers */
2897 int bravalue = 0;
2898 int greedy_default, greedy_non_default;
2899 int firstbyte, reqbyte;
2900 int zeroreqbyte, zerofirstbyte;
2901 int req_caseopt, reqvary, tempreqvary;
2902 int options = *optionsptr;
2903 int after_manual_callout = 0;
2904 int length_prevgroup = 0;
2905 register int c;
2906 register uschar *code = *codeptr;
2907 uschar *last_code = code;
2908 uschar *orig_code = code;
2909 uschar *tempcode;
2910 BOOL inescq = FALSE;
2911 BOOL groupsetfirstbyte = FALSE;
2912 const uschar *ptr = *ptrptr;
2913 const uschar *tempptr;
2914 const uschar *nestptr = NULL;
2915 uschar *previous = NULL;
2916 uschar *previous_callout = NULL;
2917 uschar *save_hwm = NULL;
2918 uschar classbits[32];
2919
2920 #ifdef SUPPORT_UTF8
2921 BOOL class_utf8;
2922 BOOL utf8 = (options & PCRE_UTF8) != 0;
2923 uschar *class_utf8data;
2924 uschar *class_utf8data_base;
2925 uschar utf8_char[6];
2926 #else
2927 BOOL utf8 = FALSE;
2928 uschar *utf8_char = NULL;
2929 #endif
2930
2931 #ifdef PCRE_DEBUG
2932 if (lengthptr != NULL) DPRINTF((">> start branch\n"));
2933 #endif
2934
2935 /* Set up the default and non-default settings for greediness */
2936
2937 greedy_default = ((options & PCRE_UNGREEDY) != 0);
2938 greedy_non_default = greedy_default ^ 1;
2939
2940 /* Initialize no first byte, no required byte. REQ_UNSET means "no char
2941 matching encountered yet". It gets changed to REQ_NONE if we hit something that
2942 matches a non-fixed char first char; reqbyte just remains unset if we never
2943 find one.
2944
2945 When we hit a repeat whose minimum is zero, we may have to adjust these values
2946 to take the zero repeat into account. This is implemented by setting them to
2947 zerofirstbyte and zeroreqbyte when such a repeat is encountered. The individual
2948 item types that can be repeated set these backoff variables appropriately. */
2949
2950 firstbyte = reqbyte = zerofirstbyte = zeroreqbyte = REQ_UNSET;
2951
2952 /* The variable req_caseopt contains either the REQ_CASELESS value or zero,
2953 according to the current setting of the caseless flag. REQ_CASELESS is a bit
2954 value > 255. It is added into the firstbyte or reqbyte variables to record the
2955 case status of the value. This is used only for ASCII characters. */
2956
2957 req_caseopt = ((options & PCRE_CASELESS) != 0)? REQ_CASELESS : 0;
2958
2959 /* Switch on next character until the end of the branch */
2960
2961 for (;; ptr++)
2962 {
2963 BOOL negate_class;
2964 BOOL should_flip_negation;
2965 BOOL possessive_quantifier;
2966 BOOL is_quantifier;
2967 BOOL is_recurse;
2968 BOOL reset_bracount;
2969 int class_charcount;
2970 int class_lastchar;
2971 int newoptions;
2972 int recno;
2973 int refsign;
2974 int skipbytes;
2975 int subreqbyte;
2976 int subfirstbyte;
2977 int terminator;
2978 int mclength;
2979 uschar mcbuffer[8];
2980
2981 /* Get next byte in the pattern */
2982
2983 c = *ptr;
2984
2985 /* If we are at the end of a nested substitution, revert to the outer level
2986 string. Nesting only happens one level deep. */
2987
2988 if (c == 0 && nestptr != NULL)
2989 {
2990 ptr = nestptr;
2991 nestptr = NULL;
2992 c = *ptr;
2993 }
2994
2995 /* If we are in the pre-compile phase, accumulate the length used for the
2996 previous cycle of this loop. */
2997
2998 if (lengthptr != NULL)
2999 {
3000 #ifdef PCRE_DEBUG
3001 if (code > cd->hwm) cd->hwm = code; /* High water info */
3002 #endif
3003 if (code > cd->start_workspace + WORK_SIZE_CHECK) /* Check for overrun */
3004 {
3005 *errorcodeptr = ERR52;
3006 goto FAILED;
3007 }
3008
3009 /* There is at least one situation where code goes backwards: this is the
3010 case of a zero quantifier after a class (e.g. [ab]{0}). At compile time,
3011 the class is simply eliminated. However, it is created first, so we have to
3012 allow memory for it. Therefore, don't ever reduce the length at this point.
3013 */
3014
3015 if (code < last_code) code = last_code;
3016
3017 /* Paranoid check for integer overflow */
3018
3019 if (OFLOW_MAX - *lengthptr < code - last_code)
3020 {
3021 *errorcodeptr = ERR20;
3022 goto FAILED;
3023 }
3024
3025 *lengthptr += (int)(code - last_code);
3026 DPRINTF(("length=%d added %d c=%c\n", *lengthptr, code - last_code, c));
3027
3028 /* If "previous" is set and it is not at the start of the work space, move
3029 it back to there, in order to avoid filling up the work space. Otherwise,
3030 if "previous" is NULL, reset the current code pointer to the start. */
3031
3032 if (previous != NULL)
3033 {
3034 if (previous > orig_code)
3035 {
3036 memmove(orig_code, previous, code - previous);
3037 code -= previous - orig_code;
3038 previous = orig_code;
3039 }
3040 }
3041 else code = orig_code;
3042
3043 /* Remember where this code item starts so we can pick up the length
3044 next time round. */
3045
3046 last_code = code;
3047 }
3048
3049 /* In the real compile phase, just check the workspace used by the forward
3050 reference list. */
3051
3052 else if (cd->hwm > cd->start_workspace + WORK_SIZE_CHECK)
3053 {
3054 *errorcodeptr = ERR52;
3055 goto FAILED;
3056 }
3057
3058 /* If in \Q...\E, check for the end; if not, we have a literal */
3059
3060 if (inescq && c != 0)
3061 {
3062 if (c == CHAR_BACKSLASH && ptr[1] == CHAR_E)
3063 {
3064 inescq = FALSE;
3065 ptr++;
3066 continue;
3067 }
3068 else
3069 {
3070 if (previous_callout != NULL)
3071 {
3072 if (lengthptr == NULL) /* Don't attempt in pre-compile phase */
3073 complete_callout(previous_callout, ptr, cd);
3074 previous_callout = NULL;
3075 }
3076 if ((options & PCRE_AUTO_CALLOUT) != 0)
3077 {
3078 previous_callout = code;
3079 code = auto_callout(code, ptr, cd);
3080 }
3081 goto NORMAL_CHAR;
3082 }
3083 }
3084
3085 /* Fill in length of a previous callout, except when the next thing is
3086 a quantifier. */
3087
3088 is_quantifier =
3089 c == CHAR_ASTERISK || c == CHAR_PLUS || c == CHAR_QUESTION_MARK ||
3090 (c == CHAR_LEFT_CURLY_BRACKET && is_counted_repeat(ptr+1));
3091
3092 if (!is_quantifier && previous_callout != NULL &&
3093 after_manual_callout-- <= 0)
3094 {
3095 if (lengthptr == NULL) /* Don't attempt in pre-compile phase */
3096 complete_callout(previous_callout, ptr, cd);
3097 previous_callout = NULL;
3098 }
3099
3100 /* In extended mode, skip white space and comments */
3101
3102 if ((options & PCRE_EXTENDED) != 0)
3103 {
3104 if ((cd->ctypes[c] & ctype_space) != 0) continue;
3105 if (c == CHAR_NUMBER_SIGN)
3106 {
3107 while (*(++ptr) != 0)
3108 {
3109 if (IS_NEWLINE(ptr)) { ptr += cd->nllen - 1; break; }
3110 }
3111 if (*ptr != 0) continue;
3112
3113 /* Else fall through to handle end of string */
3114 c = 0;
3115 }
3116 }
3117
3118 /* No auto callout for quantifiers. */
3119
3120 if ((options & PCRE_AUTO_CALLOUT) != 0 && !is_quantifier)
3121 {
3122 previous_callout = code;
3123 code = auto_callout(code, ptr, cd);
3124 }
3125
3126 switch(c)
3127 {
3128 /* ===================================================================*/
3129 case 0: /* The branch terminates at string end */
3130 case CHAR_VERTICAL_LINE: /* or | or ) */
3131 case CHAR_RIGHT_PARENTHESIS:
3132 *firstbyteptr = firstbyte;
3133 *reqbyteptr = reqbyte;
3134 *codeptr = code;
3135 *ptrptr = ptr;
3136 if (lengthptr != NULL)
3137 {
3138 if (OFLOW_MAX - *lengthptr < code - last_code)
3139 {
3140 *errorcodeptr = ERR20;
3141 goto FAILED;
3142 }
3143 *lengthptr += (int)(code - last_code); /* To include callout length */
3144 DPRINTF((">> end branch\n"));
3145 }
3146 return TRUE;
3147
3148
3149 /* ===================================================================*/
3150 /* Handle single-character metacharacters. In multiline mode, ^ disables
3151 the setting of any following char as a first character. */
3152
3153 case CHAR_CIRCUMFLEX_ACCENT:
3154 if ((options & PCRE_MULTILINE) != 0)
3155 {
3156 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
3157 }
3158 previous = NULL;
3159 *code++ = OP_CIRC;
3160 break;
3161
3162 case CHAR_DOLLAR_SIGN:
3163 previous = NULL;
3164 *code++ = OP_DOLL;
3165 break;
3166
3167 /* There can never be a first char if '.' is first, whatever happens about
3168 repeats. The value of reqbyte doesn't change either. */
3169
3170 case CHAR_DOT:
3171 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
3172 zerofirstbyte = firstbyte;
3173 zeroreqbyte = reqbyte;
3174 previous = code;
3175 *code++ = ((options & PCRE_DOTALL) != 0)? OP_ALLANY: OP_ANY;
3176 break;
3177
3178
3179 /* ===================================================================*/
3180 /* Character classes. If the included characters are all < 256, we build a
3181 32-byte bitmap of the permitted characters, except in the special case
3182 where there is only one such character. For negated classes, we build the
3183 map as usual, then invert it at the end. However, we use a different opcode
3184 so that data characters > 255 can be handled correctly.
3185
3186 If the class contains characters outside the 0-255 range, a different
3187 opcode is compiled. It may optionally have a bit map for characters < 256,
3188 but those above are are explicitly listed afterwards. A flag byte tells
3189 whether the bitmap is present, and whether this is a negated class or not.
3190
3191 In JavaScript compatibility mode, an isolated ']' causes an error. In
3192 default (Perl) mode, it is treated as a data character. */
3193
3194 case CHAR_RIGHT_SQUARE_BRACKET:
3195 if ((cd->external_options & PCRE_JAVASCRIPT_COMPAT) != 0)
3196 {
3197 *errorcodeptr = ERR64;
3198 goto FAILED;
3199 }
3200 goto NORMAL_CHAR;
3201
3202 case CHAR_LEFT_SQUARE_BRACKET:
3203 previous = code;
3204
3205 /* PCRE supports POSIX class stuff inside a class. Perl gives an error if
3206 they are encountered at the top level, so we'll do that too. */
3207
3208 if ((ptr[1] == CHAR_COLON || ptr[1] == CHAR_DOT ||
3209 ptr[1] == CHAR_EQUALS_SIGN) &&
3210 check_posix_syntax(ptr, &tempptr))
3211 {
3212 *errorcodeptr = (ptr[1] == CHAR_COLON)? ERR13 : ERR31;
3213 goto FAILED;
3214 }
3215
3216 /* If the first character is '^', set the negation flag and skip it. Also,
3217 if the first few characters (either before or after ^) are \Q\E or \E we
3218 skip them too. This makes for compatibility with Perl. */
3219
3220 negate_class = FALSE;
3221 for (;;)
3222 {
3223 c = *(++ptr);
3224 if (c == CHAR_BACKSLASH)
3225 {
3226 if (ptr[1] == CHAR_E)
3227 ptr++;
3228 else if (strncmp((const char *)ptr+1,
3229 STR_Q STR_BACKSLASH STR_E, 3) == 0)
3230 ptr += 3;
3231 else
3232 break;
3233 }
3234 else if (!negate_class && c == CHAR_CIRCUMFLEX_ACCENT)
3235 negate_class = TRUE;
3236 else break;
3237 }
3238
3239 /* Empty classes are allowed in JavaScript compatibility mode. Otherwise,
3240 an initial ']' is taken as a data character -- the code below handles
3241 that. In JS mode, [] must always fail, so generate OP_FAIL, whereas
3242 [^] must match any character, so generate OP_ALLANY. */
3243
3244 if (c == CHAR_RIGHT_SQUARE_BRACKET &&
3245 (cd->external_options & PCRE_JAVASCRIPT_COMPAT) != 0)
3246 {
3247 *code++ = negate_class? OP_ALLANY : OP_FAIL;
3248 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
3249 zerofirstbyte = firstbyte;
3250 break;
3251 }
3252
3253 /* If a class contains a negative special such as \S, we need to flip the
3254 negation flag at the end, so that support for characters > 255 works
3255 correctly (they are all included in the class). */
3256
3257 should_flip_negation = FALSE;
3258
3259 /* Keep a count of chars with values < 256 so that we can optimize the case
3260 of just a single character (as long as it's < 256). However, For higher
3261 valued UTF-8 characters, we don't yet do any optimization. */
3262
3263 class_charcount = 0;
3264 class_lastchar = -1;
3265
3266 /* Initialize the 32-char bit map to all zeros. We build the map in a
3267 temporary bit of memory, in case the class contains only 1 character (less
3268 than 256), because in that case the compiled code doesn't use the bit map.
3269 */
3270
3271 memset(classbits, 0, 32 * sizeof(uschar));
3272
3273 #ifdef SUPPORT_UTF8
3274 class_utf8 = FALSE; /* No chars >= 256 */
3275 class_utf8data = code + LINK_SIZE + 2; /* For UTF-8 items */
3276 class_utf8data_base = class_utf8data; /* For resetting in pass 1 */
3277 #endif
3278
3279 /* Process characters until ] is reached. By writing this as a "do" it
3280 means that an initial ] is taken as a data character. At the start of the
3281 loop, c contains the first byte of the character. */
3282
3283 if (c != 0) do
3284 {
3285 const uschar *oldptr;
3286
3287 #ifdef SUPPORT_UTF8
3288 if (utf8 && c > 127)
3289 { /* Braces are required because the */
3290 GETCHARLEN(c, ptr, ptr); /* macro generates multiple statements */
3291 }
3292
3293 /* In the pre-compile phase, accumulate the length of any UTF-8 extra
3294 data and reset the pointer. This is so that very large classes that
3295 contain a zillion UTF-8 characters no longer overwrite the work space
3296 (which is on the stack). */
3297
3298 if (lengthptr != NULL)
3299 {
3300 *lengthptr += class_utf8data - class_utf8data_base;
3301 class_utf8data = class_utf8data_base;
3302 }
3303
3304 #endif
3305
3306 /* Inside \Q...\E everything is literal except \E */
3307
3308 if (inescq)
3309 {
3310 if (c == CHAR_BACKSLASH && ptr[1] == CHAR_E) /* If we are at \E */
3311 {
3312 inescq = FALSE; /* Reset literal state */
3313 ptr++; /* Skip the 'E' */
3314 continue; /* Carry on with next */
3315 }
3316 goto CHECK_RANGE; /* Could be range if \E follows */
3317 }
3318
3319 /* Handle POSIX class names. Perl allows a negation extension of the
3320 form [:^name:]. A square bracket that doesn't match the syntax is
3321 treated as a literal. We also recognize the POSIX constructions
3322 [.ch.] and [=ch=] ("collating elements") and fault them, as Perl
3323 5.6 and 5.8 do. */
3324
3325 if (c == CHAR_LEFT_SQUARE_BRACKET &&
3326 (ptr[1] == CHAR_COLON || ptr[1] == CHAR_DOT ||
3327 ptr[1] == CHAR_EQUALS_SIGN) && check_posix_syntax(ptr, &tempptr))
3328 {
3329 BOOL local_negate = FALSE;
3330 int posix_class, taboffset, tabopt;
3331 register const uschar *cbits = cd->cbits;
3332 uschar pbits[32];
3333
3334 if (ptr[1] != CHAR_COLON)
3335 {
3336 *errorcodeptr = ERR31;
3337 goto FAILED;
3338 }
3339
3340 ptr += 2;
3341 if (*ptr == CHAR_CIRCUMFLEX_ACCENT)
3342 {
3343 local_negate = TRUE;
3344 should_flip_negation = TRUE; /* Note negative special */
3345 ptr++;
3346 }
3347
3348 posix_class = check_posix_name(ptr, (int)(tempptr - ptr));
3349 if (posix_class < 0)
3350 {
3351 *errorcodeptr = ERR30;
3352 goto FAILED;
3353 }
3354
3355 /* If matching is caseless, upper and lower are converted to
3356 alpha. This relies on the fact that the class table starts with
3357 alpha, lower, upper as the first 3 entries. */
3358
3359 if ((options & PCRE_CASELESS) != 0 && posix_class <= 2)
3360 posix_class = 0;
3361
3362 /* When PCRE_UCP is set, some of the POSIX classes are converted to
3363 different escape sequences that use Unicode properties. */
3364
3365 #ifdef SUPPORT_UCP
3366 if ((options & PCRE_UCP) != 0)
3367 {
3368 int pc = posix_class + ((local_negate)? POSIX_SUBSIZE/2 : 0);
3369 if (posix_substitutes[pc] != NULL)
3370 {
3371 nestptr = tempptr + 1;
3372 ptr = posix_substitutes[pc] - 1;
3373 continue;
3374 }
3375 }
3376 #endif
3377 /* In the non-UCP case, we build the bit map for the POSIX class in a
3378 chunk of local store because we may be adding and subtracting from it,
3379 and we don't want to subtract bits that may be in the main map already.
3380 At the end we or the result into the bit map that is being built. */
3381
3382 posix_class *= 3;
3383
3384 /* Copy in the first table (always present) */
3385
3386 memcpy(pbits, cbits + posix_class_maps[posix_class],
3387 32 * sizeof(uschar));
3388
3389 /* If there is a second table, add or remove it as required. */
3390
3391 taboffset = posix_class_maps[posix_class + 1];
3392 tabopt = posix_class_maps[posix_class + 2];
3393
3394 if (taboffset >= 0)
3395 {
3396 if (tabopt >= 0)
3397 for (c = 0; c < 32; c++) pbits[c] |= cbits[c + taboffset];
3398 else
3399 for (c = 0; c < 32; c++) pbits[c] &= ~cbits[c + taboffset];
3400 }
3401
3402 /* Not see if we need to remove any special characters. An option
3403 value of 1 removes vertical space and 2 removes underscore. */
3404
3405 if (tabopt < 0) tabopt = -tabopt;
3406 if (tabopt == 1) pbits[1] &= ~0x3c;
3407 else if (tabopt == 2) pbits[11] &= 0x7f;
3408
3409 /* Add the POSIX table or its complement into the main table that is
3410 being built and we are done. */
3411
3412 if (local_negate)
3413 for (c = 0; c < 32; c++) classbits[c] |= ~pbits[c];
3414 else
3415 for (c = 0; c < 32; c++) classbits[c] |= pbits[c];
3416
3417 ptr = tempptr + 1;
3418 class_charcount = 10; /* Set > 1; assumes more than 1 per class */
3419 continue; /* End of POSIX syntax handling */
3420 }
3421
3422 /* Backslash may introduce a single character, or it may introduce one
3423 of the specials, which just set a flag. The sequence \b is a special
3424 case. Inside a class (and only there) it is treated as backspace. We
3425 assume that other escapes have more than one character in them, so set
3426 class_charcount bigger than one. Unrecognized escapes fall through and
3427 are either treated as literal characters (by default), or are faulted if
3428 PCRE_EXTRA is set. */
3429
3430 if (c == CHAR_BACKSLASH)
3431 {
3432 c = check_escape(&ptr, errorcodeptr, cd->bracount, options, TRUE);
3433 if (*errorcodeptr != 0) goto FAILED;
3434
3435 if (-c == ESC_b) c = CHAR_BS; /* \b is backspace in a class */
3436 else if (-c == ESC_Q) /* Handle start of quoted string */
3437 {
3438 if (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
3439 {
3440 ptr += 2; /* avoid empty string */
3441 }
3442 else inescq = TRUE;
3443 continue;
3444 }
3445 else if (-c == ESC_E) continue; /* Ignore orphan \E */
3446
3447 if (c < 0)
3448 {
3449 register const uschar *cbits = cd->cbits;
3450 class_charcount += 2; /* Greater than 1 is what matters */
3451
3452 switch (-c)
3453 {
3454 #ifdef SUPPORT_UCP
3455 case ESC_du: /* These are the values given for \d etc */
3456 case ESC_DU: /* when PCRE_UCP is set. We replace the */
3457 case ESC_wu: /* escape sequence with an appropriate \p */
3458 case ESC_WU: /* or \P to test Unicode properties instead */
3459 case ESC_su: /* of the default ASCII testing. */
3460 case ESC_SU:
3461 nestptr = ptr;
3462 ptr = substitutes[-c - ESC_DU] - 1; /* Just before substitute */
3463 class_charcount -= 2; /* Undo! */
3464 continue;
3465 #endif
3466 case ESC_d:
3467 for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_digit];
3468 continue;
3469
3470 case ESC_D:
3471 should_flip_negation = TRUE;
3472 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_digit];
3473 continue;
3474
3475 case ESC_w:
3476 for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_word];
3477 continue;
3478
3479 case ESC_W:
3480 should_flip_negation = TRUE;
3481 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_word];
3482 continue;
3483
3484 case ESC_s:
3485 for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_space];
3486 classbits[1] &= ~0x08; /* Perl 5.004 onwards omits VT from \s */
3487 continue;
3488
3489 case ESC_S:
3490 should_flip_negation = TRUE;
3491 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_space];
3492 classbits[1] |= 0x08; /* Perl 5.004 onwards omits VT from \s */
3493 continue;
3494
3495 case ESC_h:
3496 SETBIT(classbits, 0x09); /* VT */
3497 SETBIT(classbits, 0x20); /* SPACE */
3498 SETBIT(classbits, 0xa0); /* NSBP */
3499 #ifdef SUPPORT_UTF8
3500 if (utf8)
3501 {
3502 class_utf8 = TRUE;
3503 *class_utf8data++ = XCL_SINGLE;
3504 class_utf8data += _pcre_ord2utf8(0x1680, class_utf8data);
3505 *class_utf8data++ = XCL_SINGLE;
3506 class_utf8data += _pcre_ord2utf8(0x180e, class_utf8data);
3507 *class_utf8data++ = XCL_RANGE;
3508 class_utf8data += _pcre_ord2utf8(0x2000, class_utf8data);
3509 class_utf8data += _pcre_ord2utf8(0x200A, class_utf8data);
3510 *class_utf8data++ = XCL_SINGLE;
3511 class_utf8data += _pcre_ord2utf8(0x202f, class_utf8data);
3512 *class_utf8data++ = XCL_SINGLE;
3513 class_utf8data += _pcre_ord2utf8(0x205f, class_utf8data);
3514 *class_utf8data++ = XCL_SINGLE;
3515 class_utf8data += _pcre_ord2utf8(0x3000, class_utf8data);
3516 }
3517 #endif
3518 continue;
3519
3520 case ESC_H:
3521 for (c = 0; c < 32; c++)
3522 {
3523 int x = 0xff;
3524 switch (c)
3525 {
3526 case 0x09/8: x ^= 1 << (0x09%8); break;
3527 case 0x20/8: x ^= 1 << (0x20%8); break;
3528 case 0xa0/8: x ^= 1 << (0xa0%8); break;
3529 default: break;
3530 }
3531 classbits[c] |= x;
3532 }
3533
3534 #ifdef SUPPORT_UTF8
3535 if (utf8)
3536 {
3537 class_utf8 = TRUE;
3538 *class_utf8data++ = XCL_RANGE;
3539 class_utf8data += _pcre_ord2utf8(0x0100, class_utf8data);
3540 class_utf8data += _pcre_ord2utf8(0x167f, class_utf8data);
3541 *class_utf8data++ = XCL_RANGE;
3542 class_utf8data += _pcre_ord2utf8(0x1681, class_utf8data);
3543 class_utf8data += _pcre_ord2utf8(0x180d, class_utf8data);
3544 *class_utf8data++ = XCL_RANGE;
3545 class_utf8data += _pcre_ord2utf8(0x180f, class_utf8data);
3546 class_utf8data += _pcre_ord2utf8(0x1fff, class_utf8data);
3547 *class_utf8data++ = XCL_RANGE;
3548 class_utf8data += _pcre_ord2utf8(0x200B, class_utf8data);
3549 class_utf8data += _pcre_ord2utf8(0x202e, class_utf8data);
3550 *class_utf8data++ = XCL_RANGE;
3551 class_utf8data += _pcre_ord2utf8(0x2030, class_utf8data);
3552 class_utf8data += _pcre_ord2utf8(0x205e, class_utf8data);
3553 *class_utf8data++ = XCL_RANGE;
3554 class_utf8data += _pcre_ord2utf8(0x2060, class_utf8data);
3555 class_utf8data += _pcre_ord2utf8(0x2fff, class_utf8data);
3556 *class_utf8data++ = XCL_RANGE;
3557 class_utf8data += _pcre_ord2utf8(0x3001, class_utf8data);
3558 class_utf8data += _pcre_ord2utf8(0x7fffffff, class_utf8data);
3559 }
3560 #endif
3561 continue;
3562
3563 case ESC_v:
3564 SETBIT(classbits, 0x0a); /* LF */
3565 SETBIT(classbits, 0x0b); /* VT */
3566 SETBIT(classbits, 0x0c); /* FF */
3567 SETBIT(classbits, 0x0d); /* CR */
3568 SETBIT(classbits, 0x85); /* NEL */
3569 #ifdef SUPPORT_UTF8
3570 if (utf8)
3571 {
3572 class_utf8 = TRUE;
3573 *class_utf8data++ = XCL_RANGE;
3574 class_utf8data += _pcre_ord2utf8(0x2028, class_utf8data);
3575 class_utf8data += _pcre_ord2utf8(0x2029, class_utf8data);
3576 }
3577 #endif
3578 continue;
3579
3580 case ESC_V:
3581 for (c = 0; c < 32; c++)
3582 {
3583 int x = 0xff;
3584 switch (c)
3585 {
3586 case 0x0a/8: x ^= 1 << (0x0a%8);
3587 x ^= 1 << (0x0b%8);
3588 x ^= 1 << (0x0c%8);
3589 x ^= 1 << (0x0d%8);
3590 break;
3591 case 0x85/8: x ^= 1 << (0x85%8); break;
3592 default: break;
3593 }
3594 classbits[c] |= x;
3595 }
3596
3597 #ifdef SUPPORT_UTF8
3598 if (utf8)
3599 {
3600 class_utf8 = TRUE;
3601 *class_utf8data++ = XCL_RANGE;
3602 class_utf8data += _pcre_ord2utf8(0x0100, class_utf8data);
3603 class_utf8data += _pcre_ord2utf8(0x2027, class_utf8data);
3604 *class_utf8data++ = XCL_RANGE;
3605 class_utf8data += _pcre_ord2utf8(0x2029, class_utf8data);
3606 class_utf8data += _pcre_ord2utf8(0x7fffffff, class_utf8data);
3607 }
3608 #endif
3609 continue;
3610
3611 #ifdef SUPPORT_UCP
3612 case ESC_p:
3613 case ESC_P:
3614 {
3615 BOOL negated;
3616 int pdata;
3617 int ptype = get_ucp(&ptr, &negated, &pdata, errorcodeptr);
3618 if (ptype < 0) goto FAILED;
3619 class_utf8 = TRUE;
3620 *class_utf8data++ = ((-c == ESC_p) != negated)?
3621 XCL_PROP : XCL_NOTPROP;
3622 *class_utf8data++ = ptype;
3623 *class_utf8data++ = pdata;
3624 class_charcount -= 2; /* Not a < 256 character */
3625 continue;
3626 }
3627 #endif
3628 /* Unrecognized escapes are faulted if PCRE is running in its
3629 strict mode. By default, for compatibility with Perl, they are
3630 treated as literals. */
3631
3632 default:
3633 if ((options & PCRE_EXTRA) != 0)
3634 {
3635 *errorcodeptr = ERR7;
3636 goto FAILED;
3637 }
3638 class_charcount -= 2; /* Undo the default count from above */
3639 c = *ptr; /* Get the final character and fall through */
3640 break;
3641 }
3642 }
3643
3644 /* Fall through if we have a single character (c >= 0). This may be
3645 greater than 256 in UTF-8 mode. */
3646
3647 } /* End of backslash handling */
3648
3649 /* A single character may be followed by '-' to form a range. However,
3650 Perl does not permit ']' to be the end of the range. A '-' character
3651 at the end is treated as a literal. Perl ignores orphaned \E sequences
3652 entirely. The code for handling \Q and \E is messy. */
3653
3654 CHECK_RANGE:
3655 while (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
3656 {
3657 inescq = FALSE;
3658 ptr += 2;
3659 }
3660
3661 oldptr = ptr;
3662
3663 /* Remember \r or \n */
3664
3665 if (c == CHAR_CR || c == CHAR_NL) cd->external_flags |= PCRE_HASCRORLF;
3666
3667 /* Check for range */
3668
3669 if (!inescq && ptr[1] == CHAR_MINUS)
3670 {
3671 int d;
3672 ptr += 2;
3673 while (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_E) ptr += 2;
3674
3675 /* If we hit \Q (not followed by \E) at this point, go into escaped
3676 mode. */
3677
3678 while (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_Q)
3679 {
3680 ptr += 2;
3681 if (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_E)
3682 { ptr += 2; continue; }
3683 inescq = TRUE;
3684 break;
3685 }
3686
3687 if (*ptr == 0 || (!inescq && *ptr == CHAR_RIGHT_SQUARE_BRACKET))
3688 {
3689 ptr = oldptr;
3690 goto LONE_SINGLE_CHARACTER;
3691 }
3692
3693 #ifdef SUPPORT_UTF8
3694 if (utf8)
3695 { /* Braces are required because the */
3696 GETCHARLEN(d, ptr, ptr); /* macro generates multiple statements */
3697 }
3698 else
3699 #endif
3700 d = *ptr; /* Not UTF-8 mode */
3701
3702 /* The second part of a range can be a single-character escape, but
3703 not any of the other escapes. Perl 5.6 treats a hyphen as a literal
3704 in such circumstances. */
3705
3706 if (!inescq && d == CHAR_BACKSLASH)
3707 {
3708 d = check_escape(&ptr, errorcodeptr, cd->bracount, options, TRUE);
3709 if (*errorcodeptr != 0) goto FAILED;
3710
3711 /* \b is backspace; any other special means the '-' was literal */
3712
3713 if (d < 0)
3714 {
3715 if (d == -ESC_b) d = CHAR_BS; else
3716 {
3717 ptr = oldptr;
3718 goto LONE_SINGLE_CHARACTER; /* A few lines below */
3719 }
3720 }
3721 }
3722
3723 /* Check that the two values are in the correct order. Optimize
3724 one-character ranges */
3725
3726 if (d < c)
3727 {
3728 *errorcodeptr = ERR8;
3729 goto FAILED;
3730 }
3731
3732 if (d == c) goto LONE_SINGLE_CHARACTER; /* A few lines below */
3733
3734 /* Remember \r or \n */
3735
3736 if (d == CHAR_CR || d == CHAR_NL) cd->external_flags |= PCRE_HASCRORLF;
3737
3738 /* In UTF-8 mode, if the upper limit is > 255, or > 127 for caseless
3739 matching, we have to use an XCLASS with extra data items. Caseless
3740 matching for characters > 127 is available only if UCP support is
3741 available. */
3742
3743 #ifdef SUPPORT_UTF8
3744 if (utf8 && (d > 255 || ((options & PCRE_CASELESS) != 0 && d > 127)))
3745 {
3746 class_utf8 = TRUE;
3747
3748 /* With UCP support, we can find the other case equivalents of
3749 the relevant characters. There may be several ranges. Optimize how
3750 they fit with the basic range. */
3751
3752 #ifdef SUPPORT_UCP
3753 if ((options & PCRE_CASELESS) != 0)
3754 {
3755 unsigned int occ, ocd;
3756 unsigned int cc = c;
3757 unsigned int origd = d;
3758 while (get_othercase_range(&cc, origd, &occ, &ocd))
3759 {
3760 if (occ >= (unsigned int)c &&
3761 ocd <= (unsigned int)d)
3762 continue; /* Skip embedded ranges */
3763
3764 if (occ < (unsigned int)c &&
3765 ocd >= (unsigned int)c - 1) /* Extend the basic range */
3766 { /* if there is overlap, */
3767 c = occ; /* noting that if occ < c */
3768 continue; /* we can't have ocd > d */
3769 } /* because a subrange is */
3770 if (ocd > (unsigned int)d &&
3771 occ <= (unsigned int)d + 1) /* always shorter than */
3772 { /* the basic range. */
3773 d = ocd;
3774 continue;
3775 }
3776
3777 if (occ == ocd)
3778 {
3779 *class_utf8data++ = XCL_SINGLE;
3780 }
3781 else
3782 {
3783 *class_utf8data++ = XCL_RANGE;
3784 class_utf8data += _pcre_ord2utf8(occ, class_utf8data);
3785 }
3786 class_utf8data += _pcre_ord2utf8(ocd, class_utf8data);
3787 }
3788 }
3789 #endif /* SUPPORT_UCP */
3790
3791 /* Now record the original range, possibly modified for UCP caseless
3792 overlapping ranges. */
3793
3794 *class_utf8data++ = XCL_RANGE;
3795 class_utf8data += _pcre_ord2utf8(c, class_utf8data);
3796 class_utf8data += _pcre_ord2utf8(d, class_utf8data);
3797
3798 /* With UCP support, we are done. Without UCP support, there is no
3799 caseless matching for UTF-8 characters > 127; we can use the bit map
3800 for the smaller ones. */
3801
3802 #ifdef SUPPORT_UCP
3803 continue; /* With next character in the class */
3804 #else
3805 if ((options & PCRE_CASELESS) == 0 || c > 127) continue;
3806
3807 /* Adjust upper limit and fall through to set up the map */
3808
3809 d = 127;
3810
3811 #endif /* SUPPORT_UCP */
3812 }
3813 #endif /* SUPPORT_UTF8 */
3814
3815 /* We use the bit map for all cases when not in UTF-8 mode; else
3816 ranges that lie entirely within 0-127 when there is UCP support; else
3817 for partial ranges without UCP support. */
3818
3819 class_charcount += d - c + 1;
3820 class_lastchar = d;
3821
3822 /* We can save a bit of time by skipping this in the pre-compile. */
3823
3824 if (lengthptr == NULL) for (; c <= d; c++)
3825 {
3826 classbits[c/8] |= (1 << (c&7));
3827 if ((options & PCRE_CASELESS) != 0)
3828 {
3829 int uc = cd->fcc[c]; /* flip case */
3830 classbits[uc/8] |= (1 << (uc&7));
3831 }
3832 }
3833
3834 continue; /* Go get the next char in the class */
3835 }
3836
3837 /* Handle a lone single character - we can get here for a normal
3838 non-escape char, or after \ that introduces a single character or for an
3839 apparent range that isn't. */
3840
3841 LONE_SINGLE_CHARACTER:
3842
3843 /* Handle a character that cannot go in the bit map */
3844
3845 #ifdef SUPPORT_UTF8
3846 if (utf8 && (c > 255 || ((options & PCRE_CASELESS) != 0 && c > 127)))
3847 {
3848 class_utf8 = TRUE;
3849 *class_utf8data++ = XCL_SINGLE;
3850 class_utf8data += _pcre_ord2utf8(c, class_utf8data);
3851
3852 #ifdef SUPPORT_UCP
3853 if ((options & PCRE_CASELESS) != 0)
3854 {
3855 unsigned int othercase;
3856 if ((othercase = UCD_OTHERCASE(c)) != c)
3857 {
3858 *class_utf8data++ = XCL_SINGLE;
3859 class_utf8data += _pcre_ord2utf8(othercase, class_utf8data);
3860 }
3861 }
3862 #endif /* SUPPORT_UCP */
3863
3864 }
3865 else
3866 #endif /* SUPPORT_UTF8 */
3867
3868 /* Handle a single-byte character */
3869 {
3870 classbits[c/8] |= (1 << (c&7));
3871 if ((options & PCRE_CASELESS) != 0)
3872 {
3873 c = cd->fcc[c]; /* flip case */
3874 classbits[c/8] |= (1 << (c&7));
3875 }
3876 class_charcount++;
3877 class_lastchar = c;
3878 }
3879 }
3880
3881 /* Loop until ']' reached. This "while" is the end of the "do" far above.
3882 If we are at the end of an internal nested string, revert to the outer
3883 string. */
3884
3885 while (((c = *(++ptr)) != 0 ||
3886 (nestptr != NULL &&
3887 (ptr = nestptr, nestptr = NULL, c = *(++ptr)) != 0)) &&
3888 (c != CHAR_RIGHT_SQUARE_BRACKET || inescq));
3889
3890 /* Check for missing terminating ']' */
3891
3892 if (c == 0)
3893 {
3894 *errorcodeptr = ERR6;
3895 goto FAILED;
3896 }
3897
3898 /* If class_charcount is 1, we saw precisely one character whose value is
3899 less than 256. As long as there were no characters >= 128 and there was no
3900 use of \p or \P, in other words, no use of any XCLASS features, we can
3901 optimize.
3902
3903 In UTF-8 mode, we can optimize the negative case only if there were no
3904 characters >= 128 because OP_NOT and the related opcodes like OP_NOTSTAR
3905 operate on single-bytes only. This is an historical hangover. Maybe one day
3906 we can tidy these opcodes to handle multi-byte characters.
3907
3908 The optimization throws away the bit map. We turn the item into a
3909 1-character OP_CHAR[NC] if it's positive, or OP_NOT if it's negative. Note
3910 that OP_NOT does not support multibyte characters. In the positive case, it
3911 can cause firstbyte to be set. Otherwise, there can be no first char if
3912 this item is first, whatever repeat count may follow. In the case of
3913 reqbyte, save the previous value for reinstating. */
3914
3915 #ifdef SUPPORT_UTF8
3916 if (class_charcount == 1 && !class_utf8 &&
3917 (!utf8 || !negate_class || class_lastchar < 128))
3918 #else
3919 if (class_charcount == 1)
3920 #endif
3921 {
3922 zeroreqbyte = reqbyte;
3923
3924 /* The OP_NOT opcode works on one-byte characters only. */
3925
3926 if (negate_class)
3927 {
3928 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
3929 zerofirstbyte = firstbyte;
3930 *code++ = OP_NOT;
3931 *code++ = class_lastchar;
3932 break;
3933 }
3934
3935 /* For a single, positive character, get the value into mcbuffer, and
3936 then we can handle this with the normal one-character code. */
3937
3938 #ifdef SUPPORT_UTF8
3939 if (utf8 && class_lastchar > 127)
3940 mclength = _pcre_ord2utf8(class_lastchar, mcbuffer);
3941 else
3942 #endif
3943 {
3944 mcbuffer[0] = class_lastchar;
3945 mclength = 1;
3946 }
3947 goto ONE_CHAR;
3948 } /* End of 1-char optimization */
3949
3950 /* The general case - not the one-char optimization. If this is the first
3951 thing in the branch, there can be no first char setting, whatever the
3952 repeat count. Any reqbyte setting must remain unchanged after any kind of
3953 repeat. */
3954
3955 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
3956 zerofirstbyte = firstbyte;
3957 zeroreqbyte = reqbyte;
3958
3959 /* If there are characters with values > 255, we have to compile an
3960 extended class, with its own opcode, unless there was a negated special
3961 such as \S in the class, and PCRE_UCP is not set, because in that case all
3962 characters > 255 are in the class, so any that were explicitly given as
3963 well can be ignored. If (when there are explicit characters > 255 that must
3964 be listed) there are no characters < 256, we can omit the bitmap in the
3965 actual compiled code. */
3966
3967 #ifdef SUPPORT_UTF8
3968 if (class_utf8 && (!should_flip_negation || (options & PCRE_UCP) != 0))
3969 {
3970 *class_utf8data++ = XCL_END; /* Marks the end of extra data */
3971 *code++ = OP_XCLASS;
3972 code += LINK_SIZE;
3973 *code = negate_class? XCL_NOT : 0;
3974
3975 /* If the map is required, move up the extra data to make room for it;
3976 otherwise just move the code pointer to the end of the extra data. */
3977
3978 if (class_charcount > 0)
3979 {
3980 *code++ |= XCL_MAP;
3981 memmove(code + 32, code, class_utf8data - code);
3982 memcpy(code, classbits, 32);
3983 code = class_utf8data + 32;
3984 }
3985 else code = class_utf8data;
3986
3987 /* Now fill in the complete length of the item */
3988
3989 PUT(previous, 1, code - previous);
3990 break; /* End of class handling */
3991 }
3992 #endif
3993
3994 /* If there are no characters > 255, or they are all to be included or
3995 excluded, set the opcode to OP_CLASS or OP_NCLASS, depending on whether the
3996 whole class was negated and whether there were negative specials such as \S
3997 (non-UCP) in the class. Then copy the 32-byte map into the code vector,
3998 negating it if necessary. */
3999
4000 *code++ = (negate_class == should_flip_negation) ? OP_CLASS : OP_NCLASS;
4001 if (negate_class)
4002 {
4003 if (lengthptr == NULL) /* Save time in the pre-compile phase */
4004 for (c = 0; c < 32; c++) code[c] = ~classbits[c];
4005 }
4006 else
4007 {
4008 memcpy(code, classbits, 32);
4009 }
4010 code += 32;
4011 break;
4012
4013
4014 /* ===================================================================*/
4015 /* Various kinds of repeat; '{' is not necessarily a quantifier, but this
4016 has been tested above. */
4017
4018 case CHAR_LEFT_CURLY_BRACKET:
4019 if (!is_quantifier) goto NORMAL_CHAR;
4020 ptr = read_repeat_counts(ptr+1, &repeat_min, &repeat_max, errorcodeptr);
4021 if (*errorcodeptr != 0) goto FAILED;
4022 goto REPEAT;
4023
4024 case CHAR_ASTERISK:
4025 repeat_min = 0;
4026 repeat_max = -1;
4027 goto REPEAT;
4028
4029 case CHAR_PLUS:
4030 repeat_min = 1;
4031 repeat_max = -1;
4032 goto REPEAT;
4033
4034 case CHAR_QUESTION_MARK:
4035 repeat_min = 0;
4036 repeat_max = 1;
4037
4038 REPEAT:
4039 if (previous == NULL)
4040 {
4041 *errorcodeptr = ERR9;
4042 goto FAILED;
4043 }
4044
4045 if (repeat_min == 0)
4046 {
4047 firstbyte = zerofirstbyte; /* Adjust for zero repeat */
4048 reqbyte = zeroreqbyte; /* Ditto */
4049 }
4050
4051 /* Remember whether this is a variable length repeat */
4052
4053 reqvary = (repeat_min == repeat_max)? 0 : REQ_VARY;
4054
4055 op_type = 0; /* Default single-char op codes */
4056 possessive_quantifier = FALSE; /* Default not possessive quantifier */
4057
4058 /* Save start of previous item, in case we have to move it up to make space
4059 for an inserted OP_ONCE for the additional '+' extension. */
4060
4061 tempcode = previous;
4062
4063 /* If the next character is '+', we have a possessive quantifier. This
4064 implies greediness, whatever the setting of the PCRE_UNGREEDY option.
4065 If the next character is '?' this is a minimizing repeat, by default,
4066 but if PCRE_UNGREEDY is set, it works the other way round. We change the
4067 repeat type to the non-default. */
4068
4069 if (ptr[1] == CHAR_PLUS)
4070 {
4071 repeat_type = 0; /* Force greedy */
4072 possessive_quantifier = TRUE;
4073 ptr++;
4074 }
4075 else if (ptr[1] == CHAR_QUESTION_MARK)
4076 {
4077 repeat_type = greedy_non_default;
4078 ptr++;
4079 }
4080 else repeat_type = greedy_default;
4081
4082 /* If previous was a character match, abolish the item and generate a
4083 repeat item instead. If a char item has a minumum of more than one, ensure
4084 that it is set in reqbyte - it might not be if a sequence such as x{3} is
4085 the first thing in a branch because the x will have gone into firstbyte
4086 instead. */
4087
4088 if (*previous == OP_CHAR || *previous == OP_CHARNC)
4089 {
4090 /* Deal with UTF-8 characters that take up more than one byte. It's
4091 easier to write this out separately than try to macrify it. Use c to
4092 hold the length of the character in bytes, plus 0x80 to flag that it's a
4093 length rather than a small character. */
4094
4095 #ifdef SUPPORT_UTF8
4096 if (utf8 && (code[-1] & 0x80) != 0)
4097 {
4098 uschar *lastchar = code - 1;
4099 while((*lastchar & 0xc0) == 0x80) lastchar--;
4100 c = code - lastchar; /* Length of UTF-8 character */
4101 memcpy(utf8_char, lastchar, c); /* Save the char */
4102 c |= 0x80; /* Flag c as a length */
4103 }
4104 else
4105 #endif
4106
4107 /* Handle the case of a single byte - either with no UTF8 support, or
4108 with UTF-8 disabled, or for a UTF-8 character < 128. */
4109
4110 {
4111 c = code[-1];
4112 if (repeat_min > 1) reqbyte = c | req_caseopt | cd->req_varyopt;
4113 }
4114
4115 /* If the repetition is unlimited, it pays to see if the next thing on
4116 the line is something that cannot possibly match this character. If so,
4117 automatically possessifying this item gains some performance in the case
4118 where the match fails. */
4119
4120 if (!possessive_quantifier &&
4121 repeat_max < 0 &&
4122 check_auto_possessive(previous, utf8, ptr + 1, options, cd))
4123 {
4124 repeat_type = 0; /* Force greedy */
4125 possessive_quantifier = TRUE;
4126 }
4127
4128 goto OUTPUT_SINGLE_REPEAT; /* Code shared with single character types */
4129 }
4130
4131 /* If previous was a single negated character ([^a] or similar), we use
4132 one of the special opcodes, replacing it. The code is shared with single-
4133 character repeats by setting opt_type to add a suitable offset into
4134 repeat_type. We can also test for auto-possessification. OP_NOT is
4135 currently used only for single-byte chars. */
4136
4137 else if (*previous == OP_NOT)
4138 {
4139 op_type = OP_NOTSTAR - OP_STAR; /* Use "not" opcodes */
4140 c = previous[1];
4141 if (!possessive_quantifier &&
4142 repeat_max < 0 &&
4143 check_auto_possessive(previous, utf8, ptr + 1, options, cd))
4144 {
4145 repeat_type = 0; /* Force greedy */
4146 possessive_quantifier = TRUE;
4147 }
4148 goto OUTPUT_SINGLE_REPEAT;
4149 }
4150
4151 /* If previous was a character type match (\d or similar), abolish it and
4152 create a suitable repeat item. The code is shared with single-character
4153 repeats by setting op_type to add a suitable offset into repeat_type. Note
4154 the the Unicode property types will be present only when SUPPORT_UCP is
4155 defined, but we don't wrap the little bits of code here because it just
4156 makes it horribly messy. */
4157
4158 else if (*previous < OP_EODN)
4159 {
4160 uschar *oldcode;
4161 int prop_type, prop_value;
4162 op_type = OP_TYPESTAR - OP_STAR; /* Use type opcodes */
4163 c = *previous;
4164
4165 if (!possessive_quantifier &&
4166 repeat_max < 0 &&
4167 check_auto_possessive(previous, utf8, ptr + 1, options, cd))
4168 {
4169 repeat_type = 0; /* Force greedy */
4170 possessive_quantifier = TRUE;
4171 }
4172
4173 OUTPUT_SINGLE_REPEAT:
4174 if (*previous == OP_PROP || *previous == OP_NOTPROP)
4175 {
4176 prop_type = previous[1];
4177 prop_value = previous[2];
4178 }
4179 else prop_type = prop_value = -1;
4180
4181 oldcode = code;
4182 code = previous; /* Usually overwrite previous item */
4183
4184 /* If the maximum is zero then the minimum must also be zero; Perl allows
4185 this case, so we do too - by simply omitting the item altogether. */
4186
4187 if (repeat_max == 0) goto END_REPEAT;
4188
4189 /*--------------------------------------------------------------------*/
4190 /* This code is obsolete from release 8.00; the restriction was finally
4191 removed: */
4192
4193 /* All real repeats make it impossible to handle partial matching (maybe
4194 one day we will be able to remove this restriction). */
4195
4196 /* if (repeat_max != 1) cd->external_flags |= PCRE_NOPARTIAL; */
4197 /*--------------------------------------------------------------------*/
4198
4199 /* Combine the op_type with the repeat_type */
4200
4201 repeat_type += op_type;
4202
4203 /* A minimum of zero is handled either as the special case * or ?, or as
4204 an UPTO, with the maximum given. */
4205
4206 if (repeat_min == 0)
4207 {
4208 if (repeat_max == -1) *code++ = OP_STAR + repeat_type;
4209 else if (repeat_max == 1) *code++ = OP_QUERY + repeat_type;
4210 else
4211 {
4212 *code++ = OP_UPTO + repeat_type;
4213 PUT2INC(code, 0, repeat_max);
4214 }
4215 }
4216
4217 /* A repeat minimum of 1 is optimized into some special cases. If the
4218 maximum is unlimited, we use OP_PLUS. Otherwise, the original item is
4219 left in place and, if the maximum is greater than 1, we use OP_UPTO with
4220 one less than the maximum. */
4221
4222 else if (repeat_min == 1)
4223 {
4224 if (repeat_max == -1)
4225 *code++ = OP_PLUS + repeat_type;
4226 else
4227 {
4228 code = oldcode; /* leave previous item in place */
4229 if (repeat_max == 1) goto END_REPEAT;
4230 *code++ = OP_UPTO + repeat_type;
4231 PUT2INC(code, 0, repeat_max - 1);
4232 }
4233 }
4234
4235 /* The case {n,n} is just an EXACT, while the general case {n,m} is
4236 handled as an EXACT followed by an UPTO. */
4237
4238 else
4239 {
4240 *code++ = OP_EXACT + op_type; /* NB EXACT doesn't have repeat_type */
4241 PUT2INC(code, 0, repeat_min);
4242
4243 /* If the maximum is unlimited, insert an OP_STAR. Before doing so,
4244 we have to insert the character for the previous code. For a repeated
4245 Unicode property match, there are two extra bytes that define the
4246 required property. In UTF-8 mode, long characters have their length in
4247 c, with the 0x80 bit as a flag. */
4248
4249 if (repeat_max < 0)
4250 {
4251 #ifdef SUPPORT_UTF8
4252 if (utf8 && c >= 128)
4253 {
4254 memcpy(code, utf8_char, c & 7);
4255 code += c & 7;
4256 }
4257 else
4258 #endif
4259 {
4260 *code++ = c;
4261 if (prop_type >= 0)
4262 {
4263 *code++ = prop_type;
4264 *code++ = prop_value;
4265 }
4266 }
4267 *code++ = OP_STAR + repeat_type;
4268 }
4269
4270 /* Else insert an UPTO if the max is greater than the min, again
4271 preceded by the character, for the previously inserted code. If the
4272 UPTO is just for 1 instance, we can use QUERY instead. */
4273
4274 else if (repeat_max != repeat_min)
4275 {
4276 #ifdef SUPPORT_UTF8
4277 if (utf8 && c >= 128)
4278 {
4279 memcpy(code, utf8_char, c & 7);
4280 code += c & 7;
4281 }
4282 else
4283 #endif
4284 *code++ = c;
4285 if (prop_type >= 0)
4286 {
4287 *code++ = prop_type;
4288 *code++ = prop_value;
4289 }
4290 repeat_max -= repeat_min;
4291
4292 if (repeat_max == 1)
4293 {
4294 *code++ = OP_QUERY + repeat_type;
4295 }
4296 else
4297 {
4298 *code++ = OP_UPTO + repeat_type;
4299 PUT2INC(code, 0, repeat_max);
4300 }
4301 }
4302 }
4303
4304 /* The character or character type itself comes last in all cases. */
4305
4306 #ifdef SUPPORT_UTF8
4307 if (utf8 && c >= 128)
4308 {
4309 memcpy(code, utf8_char, c & 7);
4310 code += c & 7;
4311 }
4312 else
4313 #endif
4314 *code++ = c;
4315
4316 /* For a repeated Unicode property match, there are two extra bytes that
4317 define the required property. */
4318
4319 #ifdef SUPPORT_UCP
4320 if (prop_type >= 0)
4321 {
4322 *code++ = prop_type;
4323 *code++ = prop_value;
4324 }
4325 #endif
4326 }
4327
4328 /* If previous was a character class or a back reference, we put the repeat
4329 stuff after it, but just skip the item if the repeat was {0,0}. */
4330
4331 else if (*previous == OP_CLASS ||
4332 *previous == OP_NCLASS ||
4333 #ifdef SUPPORT_UTF8
4334 *previous == OP_XCLASS ||
4335 #endif
4336 *previous == OP_REF)
4337 {
4338 if (repeat_max == 0)
4339 {
4340 code = previous;
4341 goto END_REPEAT;
4342 }
4343
4344 /*--------------------------------------------------------------------*/
4345 /* This code is obsolete from release 8.00; the restriction was finally
4346 removed: */
4347
4348 /* All real repeats make it impossible to handle partial matching (maybe
4349 one day we will be able to remove this restriction). */
4350
4351 /* if (repeat_max != 1) cd->external_flags |= PCRE_NOPARTIAL; */
4352 /*--------------------------------------------------------------------*/
4353
4354 if (repeat_min == 0 && repeat_max == -1)
4355 *code++ = OP_CRSTAR + repeat_type;
4356 else if (repeat_min == 1 && repeat_max == -1)
4357 *code++ = OP_CRPLUS + repeat_type;
4358 else if (repeat_min == 0 && repeat_max == 1)
4359 *code++ = OP_CRQUERY + repeat_type;
4360 else
4361 {
4362 *code++ = OP_CRRANGE + repeat_type;
4363 PUT2INC(code, 0, repeat_min);
4364 if (repeat_max == -1) repeat_max = 0; /* 2-byte encoding for max */
4365 PUT2INC(code, 0, repeat_max);
4366 }
4367 }
4368
4369 /* If previous was a bracket group, we may have to replicate it in certain
4370 cases. */
4371
4372 else if (*previous == OP_BRA || *previous == OP_CBRA ||
4373 *previous == OP_ONCE || *previous == OP_COND)
4374 {
4375 register int i;
4376 int ketoffset = 0;
4377 int len = (int)(code - previous);
4378 uschar *bralink = NULL;
4379
4380 /* Repeating a DEFINE group is pointless */
4381
4382 if (*previous == OP_COND && previous[LINK_SIZE+1] == OP_DEF)
4383 {
4384 *errorcodeptr = ERR55;
4385 goto FAILED;
4386 }
4387
4388 /* If the maximum repeat count is unlimited, find the end of the bracket
4389 by scanning through from the start, and compute the offset back to it
4390 from the current code pointer. There may be an OP_OPT setting following
4391 the final KET, so we can't find the end just by going back from the code
4392 pointer. */
4393
4394 if (repeat_max == -1)
4395 {
4396 register uschar *ket = previous;
4397 do ket += GET(ket, 1); while (*ket != OP_KET);
4398 ketoffset = (int)(code - ket);
4399 }
4400
4401 /* The case of a zero minimum is special because of the need to stick
4402 OP_BRAZERO in front of it, and because the group appears once in the
4403 data, whereas in other cases it appears the minimum number of times. For
4404 this reason, it is simplest to treat this case separately, as otherwise
4405 the code gets far too messy. There are several special subcases when the
4406 minimum is zero. */
4407
4408 if (repeat_min == 0)
4409 {
4410 /* If the maximum is also zero, we used to just omit the group from the
4411 output altogether, like this:
4412
4413 ** if (repeat_max == 0)
4414 ** {
4415 ** code = previous;
4416 ** goto END_REPEAT;
4417 ** }
4418
4419 However, that fails when a group is referenced as a subroutine from
4420 elsewhere in the pattern, so now we stick in OP_SKIPZERO in front of it
4421 so that it is skipped on execution. As we don't have a list of which
4422 groups are referenced, we cannot do this selectively.
4423
4424 If the maximum is 1 or unlimited, we just have to stick in the BRAZERO
4425 and do no more at this point. However, we do need to adjust any
4426 OP_RECURSE calls inside the group that refer to the group itself or any
4427 internal or forward referenced group, because the offset is from the
4428 start of the whole regex. Temporarily terminate the pattern while doing
4429 this. */
4430
4431 if (repeat_max <= 1) /* Covers 0, 1, and unlimited */
4432 {
4433 *code = OP_END;
4434 adjust_recurse(previous, 1, utf8, cd, save_hwm);
4435 memmove(previous+1, previous, len);
4436 code++;
4437 if (repeat_max == 0)
4438 {
4439 *previous++ = OP_SKIPZERO;
4440 goto END_REPEAT;
4441 }
4442 *previous++ = OP_BRAZERO + repeat_type;
4443 }
4444
4445 /* If the maximum is greater than 1 and limited, we have to replicate
4446 in a nested fashion, sticking OP_BRAZERO before each set of brackets.
4447 The first one has to be handled carefully because it's the original
4448 copy, which has to be moved up. The remainder can be handled by code
4449 that is common with the non-zero minimum case below. We have to
4450 adjust the value or repeat_max, since one less copy is required. Once
4451 again, we may have to adjust any OP_RECURSE calls inside the group. */
4452
4453 else
4454 {
4455 int offset;
4456 *code = OP_END;
4457 adjust_recurse(previous, 2 + LINK_SIZE, utf8, cd, save_hwm);
4458 memmove(previous + 2 + LINK_SIZE, previous, len);
4459 code += 2 + LINK_SIZE;
4460 *previous++ = OP_BRAZERO + repeat_type;
4461 *previous++ = OP_BRA;
4462
4463 /* We chain together the bracket offset fields that have to be
4464 filled in later when the ends of the brackets are reached. */
4465
4466 offset = (bralink == NULL)? 0 : (int)(previous - bralink);
4467 bralink = previous;
4468 PUTINC(previous, 0, offset);
4469 }
4470
4471 repeat_max--;
4472 }
4473
4474 /* If the minimum is greater than zero, replicate the group as many
4475 times as necessary, and adjust the maximum to the number of subsequent
4476 copies that we need. If we set a first char from the group, and didn't
4477 set a required char, copy the latter from the former. If there are any
4478 forward reference subroutine calls in the group, there will be entries on
4479 the workspace list; replicate these with an appropriate increment. */
4480
4481 else
4482 {
4483 if (repeat_min > 1)
4484 {
4485 /* In the pre-compile phase, we don't actually do the replication. We
4486 just adjust the length as if we had. Do some paranoid checks for
4487 potential integer overflow. The INT64_OR_DOUBLE type is a 64-bit
4488 integer type when available, otherwise double. */
4489
4490 if (lengthptr != NULL)
4491 {
4492 int delta = (repeat_min - 1)*length_prevgroup;
4493 if ((INT64_OR_DOUBLE)(repeat_min - 1)*
4494 (INT64_OR_DOUBLE)length_prevgroup >
4495 (INT64_OR_DOUBLE)INT_MAX ||
4496 OFLOW_MAX - *lengthptr < delta)
4497 {
4498 *errorcodeptr = ERR20;
4499 goto FAILED;
4500 }
4501 *lengthptr += delta;
4502 }
4503
4504 /* This is compiling for real */
4505
4506 else
4507 {
4508 if (groupsetfirstbyte && reqbyte < 0) reqbyte = firstbyte;
4509 for (i = 1; i < repeat_min; i++)
4510 {
4511 uschar *hc;
4512 uschar *this_hwm = cd->hwm;
4513 memcpy(code, previous, len);
4514 for (hc = save_hwm; hc < this_hwm; hc += LINK_SIZE)
4515 {
4516 PUT(cd->hwm, 0, GET(hc, 0) + len);
4517 cd->hwm += LINK_SIZE;
4518 }
4519 save_hwm = this_hwm;
4520 code += len;
4521 }
4522 }
4523 }
4524
4525 if (repeat_max > 0) repeat_max -= repeat_min;
4526 }
4527
4528 /* This code is common to both the zero and non-zero minimum cases. If
4529 the maximum is limited, it replicates the group in a nested fashion,
4530 remembering the bracket starts on a stack. In the case of a zero minimum,
4531 the first one was set up above. In all cases the repeat_max now specifies
4532 the number of additional copies needed. Again, we must remember to
4533 replicate entries on the forward reference list. */
4534
4535 if (repeat_max >= 0)
4536 {
4537 /* In the pre-compile phase, we don't actually do the replication. We
4538 just adjust the length as if we had. For each repetition we must add 1
4539 to the length for BRAZERO and for all but the last repetition we must
4540 add 2 + 2*LINKSIZE to allow for the nesting that occurs. Do some
4541 paranoid checks to avoid integer overflow. The INT64_OR_DOUBLE type is
4542 a 64-bit integer type when available, otherwise double. */
4543
4544 if (lengthptr != NULL && repeat_max > 0)
4545 {
4546 int delta = repeat_max * (length_prevgroup + 1 + 2 + 2*LINK_SIZE) -
4547 2 - 2*LINK_SIZE; /* Last one doesn't nest */
4548 if ((INT64_OR_DOUBLE)repeat_max *
4549 (INT64_OR_DOUBLE)(length_prevgroup + 1 + 2 + 2*LINK_SIZE)
4550 > (INT64_OR_DOUBLE)INT_MAX ||
4551 OFLOW_MAX - *lengthptr < delta)
4552 {
4553 *errorcodeptr = ERR20;
4554 goto FAILED;
4555 }
4556 *lengthptr += delta;
4557 }
4558
4559 /* This is compiling for real */
4560
4561 else for (i = repeat_max - 1; i >= 0; i--)
4562 {
4563 uschar *hc;
4564 uschar *this_hwm = cd->hwm;
4565
4566 *code++ = OP_BRAZERO + repeat_type;
4567
4568 /* All but the final copy start a new nesting, maintaining the
4569 chain of brackets outstanding. */
4570
4571 if (i != 0)
4572 {
4573 int offset;
4574 *code++ = OP_BRA;
4575 offset = (bralink == NULL)? 0 : (int)(code - bralink);
4576 bralink = code;
4577 PUTINC(code, 0, offset);
4578 }
4579
4580 memcpy(code, previous, len);
4581 for (hc = save_hwm; hc < this_hwm; hc += LINK_SIZE)
4582 {
4583 PUT(cd->hwm, 0, GET(hc, 0) + len + ((i != 0)? 2+LINK_SIZE : 1));
4584 cd->hwm += LINK_SIZE;
4585 }
4586 save_hwm = this_hwm;
4587 code += len;
4588 }
4589
4590 /* Now chain through the pending brackets, and fill in their length
4591 fields (which are holding the chain links pro tem). */
4592
4593 while (bralink != NULL)
4594 {
4595 int oldlinkoffset;
4596 int offset = (int)(code - bralink + 1);
4597 uschar *bra = code - offset;
4598 oldlinkoffset = GET(bra, 1);
4599 bralink = (oldlinkoffset == 0)? NULL : bralink - oldlinkoffset;
4600 *code++ = OP_KET;
4601 PUTINC(code, 0, offset);
4602 PUT(bra, 1, offset);
4603 }
4604 }
4605
4606 /* If the maximum is unlimited, set a repeater in the final copy. We
4607 can't just offset backwards from the current code point, because we
4608 don't know if there's been an options resetting after the ket. The
4609 correct offset was computed above.
4610
4611 Then, when we are doing the actual compile phase, check to see whether
4612 this group is a non-atomic one that could match an empty string. If so,
4613 convert the initial operator to the S form (e.g. OP_BRA -> OP_SBRA) so
4614 that runtime checking can be done. [This check is also applied to
4615 atomic groups at runtime, but in a different way.] */
4616
4617 else
4618 {
4619 uschar *ketcode = code - ketoffset;
4620 uschar *bracode = ketcode - GET(ketcode, 1);
4621 *ketcode = OP_KETRMAX + repeat_type;
4622 if (lengthptr == NULL && *bracode != OP_ONCE)
4623 {
4624 uschar *scode = bracode;
4625 do
4626 {
4627 if (could_be_empty_branch(scode, ketcode, utf8, cd))
4628 {
4629 *bracode += OP_SBRA - OP_BRA;
4630 break;
4631 }
4632 scode += GET(scode, 1);
4633 }
4634 while (*scode == OP_ALT);
4635 }
4636 }
4637 }
4638
4639 /* If previous is OP_FAIL, it was generated by an empty class [] in
4640 JavaScript mode. The other ways in which OP_FAIL can be generated, that is
4641 by (*FAIL) or (?!) set previous to NULL, which gives a "nothing to repeat"
4642 error above. We can just ignore the repeat in JS case. */
4643
4644 else if (*previous == OP_FAIL) goto END_REPEAT;
4645
4646 /* Else there's some kind of shambles */
4647
4648 else
4649 {
4650 *errorcodeptr = ERR11;
4651 goto FAILED;
4652 }
4653
4654 /* If the character following a repeat is '+', or if certain optimization
4655 tests above succeeded, possessive_quantifier is TRUE. For some of the
4656 simpler opcodes, there is an special alternative opcode for this. For
4657 anything else, we wrap the entire repeated item inside OP_ONCE brackets.
4658 The '+' notation is just syntactic sugar, taken from Sun's Java package,
4659 but the special opcodes can optimize it a bit. The repeated item starts at
4660 tempcode, not at previous, which might be the first part of a string whose
4661 (former) last char we repeated.
4662
4663 Possessifying an 'exact' quantifier has no effect, so we can ignore it. But
4664 an 'upto' may follow. We skip over an 'exact' item, and then test the
4665 length of what remains before proceeding. */
4666
4667 if (possessive_quantifier)
4668 {
4669 int len;
4670
4671 if (*tempcode == OP_TYPEEXACT)
4672 tempcode += _pcre_OP_lengths[*tempcode] +
4673 ((tempcode[3] == OP_PROP || tempcode[3] == OP_NOTPROP)? 2 : 0);
4674
4675 else if (*tempcode == OP_EXACT || *tempcode == OP_NOTEXACT)
4676 {
4677 tempcode += _pcre_OP_lengths[*tempcode];
4678 #ifdef SUPPORT_UTF8
4679 if (utf8 && tempcode[-1] >= 0xc0)
4680 tempcode += _pcre_utf8_table4[tempcode[-1] & 0x3f];
4681 #endif
4682 }
4683
4684 len = (int)(code - tempcode);
4685 if (len > 0) switch (*tempcode)
4686 {
4687 case OP_STAR: *tempcode = OP_POSSTAR; break;
4688 case OP_PLUS: *tempcode = OP_POSPLUS; break;
4689 case OP_QUERY: *tempcode = OP_POSQUERY; break;
4690 case OP_UPTO: *tempcode = OP_POSUPTO; break;
4691
4692 case OP_TYPESTAR: *tempcode = OP_TYPEPOSSTAR; break;
4693 case OP_TYPEPLUS: *tempcode = OP_TYPEPOSPLUS; break;
4694 case OP_TYPEQUERY: *tempcode = OP_TYPEPOSQUERY; break;
4695 case OP_TYPEUPTO: *tempcode = OP_TYPEPOSUPTO; break;
4696
4697 case OP_NOTSTAR: *tempcode = OP_NOTPOSSTAR; break;
4698 case OP_NOTPLUS: *tempcode = OP_NOTPOSPLUS; break;
4699 case OP_NOTQUERY: *tempcode = OP_NOTPOSQUERY; break;
4700 case OP_NOTUPTO: *tempcode = OP_NOTPOSUPTO; break;
4701
4702 /* Because we are moving code along, we must ensure that any
4703 pending recursive references are updated. */
4704
4705 default:
4706 *code = OP_END;
4707 adjust_recurse(tempcode, 1 + LINK_SIZE, utf8, cd, save_hwm);
4708 memmove(tempcode + 1+LINK_SIZE, tempcode, len);
4709 code += 1 + LINK_SIZE;
4710 len += 1 + LINK_SIZE;
4711 tempcode[0] = OP_ONCE;
4712 *code++ = OP_KET;
4713 PUTINC(code, 0, len);
4714 PUT(tempcode, 1, len);
4715 break;
4716 }
4717 }
4718
4719 /* In all case we no longer have a previous item. We also set the
4720 "follows varying string" flag for subsequently encountered reqbytes if
4721 it isn't already set and we have just passed a varying length item. */
4722
4723 END_REPEAT:
4724 previous = NULL;
4725 cd->req_varyopt |= reqvary;
4726 break;
4727
4728
4729 /* ===================================================================*/
4730 /* Start of nested parenthesized sub-expression, or comment or lookahead or
4731 lookbehind or option setting or condition or all the other extended
4732 parenthesis forms. */
4733
4734 case CHAR_LEFT_PARENTHESIS:
4735 newoptions = options;
4736 skipbytes = 0;
4737 bravalue = OP_CBRA;
4738 save_hwm = cd->hwm;
4739 reset_bracount = FALSE;
4740
4741 /* First deal with various "verbs" that can be introduced by '*'. */
4742
4743 if (*(++ptr) == CHAR_ASTERISK &&
4744 ((cd->ctypes[ptr[1]] & ctype_letter) != 0 || ptr[1] == ':'))
4745 {
4746 int i, namelen;
4747 int arglen = 0;
4748 const char *vn = verbnames;
4749 const uschar *name = ptr + 1;
4750 const uschar *arg = NULL;
4751 previous = NULL;
4752 while ((cd->ctypes[*++ptr] & ctype_letter) != 0) {};
4753 namelen = (int)(ptr - name);
4754
4755 if (*ptr == CHAR_COLON)
4756 {
4757 arg = ++ptr;
4758 while ((cd->ctypes[*ptr] & (ctype_letter|ctype_digit)) != 0
4759 || *ptr == '_') ptr++;
4760 arglen = ptr - arg;
4761 }
4762
4763 if (*ptr != CHAR_RIGHT_PARENTHESIS)
4764 {
4765 *errorcodeptr = ERR60;
4766 goto FAILED;
4767 }
4768
4769 /* Scan the table of verb names */
4770
4771 for (i = 0; i < verbcount; i++)
4772 {
4773 if (namelen == verbs[i].len &&
4774 strncmp((char *)name, vn, namelen) == 0)
4775 {
4776 /* Check for open captures before ACCEPT */
4777
4778 if (verbs[i].op == OP_ACCEPT)
4779 {
4780 open_capitem *oc;
4781 cd->had_accept = TRUE;
4782 for (oc = cd->open_caps; oc != NULL; oc = oc->next)
4783 {
4784 *code++ = OP_CLOSE;
4785 PUT2INC(code, 0, oc->number);
4786 }
4787 }
4788
4789 /* Handle the cases with/without an argument */
4790
4791 if (arglen == 0)
4792 {
4793 if (verbs[i].op < 0) /* Argument is mandatory */
4794 {
4795 *errorcodeptr = ERR66;
4796 goto FAILED;
4797 }
4798 *code++ = verbs[i].op;
4799 }
4800
4801 else
4802 {
4803 if (verbs[i].op_arg < 0) /* Argument is forbidden */
4804 {
4805 *errorcodeptr = ERR59;
4806 goto FAILED;
4807 }
4808 *code++ = verbs[i].op_arg;
4809 *code++ = arglen;
4810 memcpy(code, arg, arglen);
4811 code += arglen;
4812 *code++ = 0;
4813 }
4814
4815 break; /* Found verb, exit loop */
4816 }
4817
4818 vn += verbs[i].len + 1;
4819 }
4820
4821 if (i < verbcount) continue; /* Successfully handled a verb */
4822 *errorcodeptr = ERR60; /* Verb not recognized */
4823 goto FAILED;
4824 }
4825
4826 /* Deal with the extended parentheses; all are introduced by '?', and the
4827 appearance of any of them means that this is not a capturing group. */
4828
4829 else if (*ptr == CHAR_QUESTION_MARK)
4830 {
4831 int i, set, unset, namelen;
4832 int *optset;
4833 const uschar *name;
4834 uschar *slot;
4835
4836 switch (*(++ptr))
4837 {
4838 case CHAR_NUMBER_SIGN: /* Comment; skip to ket */
4839 ptr++;
4840 while (*ptr != 0 && *ptr != CHAR_RIGHT_PARENTHESIS) ptr++;
4841 if (*ptr == 0)
4842 {
4843 *errorcodeptr = ERR18;
4844 goto FAILED;
4845 }
4846 continue;
4847
4848
4849 /* ------------------------------------------------------------ */
4850 case CHAR_VERTICAL_LINE: /* Reset capture count for each branch */
4851 reset_bracount = TRUE;
4852 /* Fall through */
4853
4854 /* ------------------------------------------------------------ */
4855 case CHAR_COLON: /* Non-capturing bracket */
4856 bravalue = OP_BRA;
4857 ptr++;
4858 break;
4859
4860
4861 /* ------------------------------------------------------------ */
4862 case CHAR_LEFT_PARENTHESIS:
4863 bravalue = OP_COND; /* Conditional group */
4864
4865 /* A condition can be an assertion, a number (referring to a numbered
4866 group), a name (referring to a named group), or 'R', referring to
4867 recursion. R<digits> and R&name are also permitted for recursion tests.
4868
4869 There are several syntaxes for testing a named group: (?(name)) is used
4870 by Python; Perl 5.10 onwards uses (?(<name>) or (?('name')).
4871
4872 There are two unfortunate ambiguities, caused by history. (a) 'R' can
4873 be the recursive thing or the name 'R' (and similarly for 'R' followed
4874 by digits), and (b) a number could be a name that consists of digits.
4875 In both cases, we look for a name first; if not found, we try the other
4876 cases. */
4877
4878 /* For conditions that are assertions, check the syntax, and then exit
4879 the switch. This will take control down to where bracketed groups,
4880 including assertions, are processed. */
4881
4882 if (ptr[1] == CHAR_QUESTION_MARK && (ptr[2] == CHAR_EQUALS_SIGN ||
4883 ptr[2] == CHAR_EXCLAMATION_MARK || ptr[2] == CHAR_LESS_THAN_SIGN))
4884 break;
4885
4886 /* Most other conditions use OP_CREF (a couple change to OP_RREF
4887 below), and all need to skip 3 bytes at the start of the group. */
4888
4889 code[1+LINK_SIZE] = OP_CREF;
4890 skipbytes = 3;
4891 refsign = -1;
4892
4893 /* Check for a test for recursion in a named group. */
4894
4895 if (ptr[1] == CHAR_R && ptr[2] == CHAR_AMPERSAND)
4896 {
4897 terminator = -1;
4898 ptr += 2;
4899 code[1+LINK_SIZE] = OP_RREF; /* Change the type of test */
4900 }
4901
4902 /* Check for a test for a named group's having been set, using the Perl
4903 syntax (?(<name>) or (?('name') */
4904
4905 else if (ptr[1] == CHAR_LESS_THAN_SIGN)
4906 {
4907 terminator = CHAR_GREATER_THAN_SIGN;
4908 ptr++;
4909 }
4910 else if (ptr[1] == CHAR_APOSTROPHE)
4911 {
4912 terminator = CHAR_APOSTROPHE;
4913 ptr++;
4914 }
4915 else
4916 {
4917 terminator = 0;
4918 if (ptr[1] == CHAR_MINUS || ptr[1] == CHAR_PLUS) refsign = *(++ptr);
4919 }
4920
4921 /* We now expect to read a name; any thing else is an error */
4922
4923 if ((cd->ctypes[ptr[1]] & ctype_word) == 0)
4924 {
4925 ptr += 1; /* To get the right offset */
4926 *errorcodeptr = ERR28;
4927 goto FAILED;
4928 }
4929
4930 /* Read the name, but also get it as a number if it's all digits */
4931
4932 recno = 0;
4933 name = ++ptr;
4934 while ((cd->ctypes[*ptr] & ctype_word) != 0)
4935 {
4936 if (recno >= 0)
4937 recno = ((digitab[*ptr] & ctype_digit) != 0)?
4938 recno * 10 + *ptr - CHAR_0 : -1;
4939 ptr++;
4940 }
4941 namelen = (int)(ptr - name);
4942
4943 if ((terminator > 0 && *ptr++ != terminator) ||
4944 *ptr++ != CHAR_RIGHT_PARENTHESIS)
4945 {
4946 ptr--; /* Error offset */
4947 *errorcodeptr = ERR26;
4948 goto FAILED;
4949 }
4950
4951 /* Do no further checking in the pre-compile phase. */
4952
4953 if (lengthptr != NULL) break;
4954
4955 /* In the real compile we do the work of looking for the actual
4956 reference. If the string started with "+" or "-" we require the rest to
4957 be digits, in which case recno will be set. */
4958
4959 if (refsign > 0)
4960 {
4961 if (recno <= 0)
4962 {
4963 *errorcodeptr = ERR58;
4964 goto FAILED;
4965 }
4966 recno = (refsign == CHAR_MINUS)?
4967 cd->bracount - recno + 1 : recno +cd->bracount;
4968 if (recno <= 0 || recno > cd->final_bracount)
4969 {
4970 *errorcodeptr = ERR15;
4971 goto FAILED;
4972 }
4973 PUT2(code, 2+LINK_SIZE, recno);
4974 break;
4975 }
4976
4977 /* Otherwise (did not start with "+" or "-"), start by looking for the
4978 name. If we find a name, add one to the opcode to change OP_CREF or
4979 OP_RREF into OP_NCREF or OP_NRREF. These behave exactly the same,
4980 except they record that the reference was originally to a name. The
4981 information is used to check duplicate names. */
4982
4983 slot = cd->name_table;
4984 for (i = 0; i < cd->names_found; i++)
4985 {
4986 if (strncmp((char *)name, (char *)slot+2, namelen) == 0) break;
4987 slot += cd->name_entry_size;
4988 }
4989
4990 /* Found a previous named subpattern */
4991
4992 if (i < cd->names_found)
4993 {
4994 recno = GET2(slot, 0);
4995 PUT2(code, 2+LINK_SIZE, recno);
4996 code[1+LINK_SIZE]++;
4997 }
4998
4999 /* Search the pattern for a forward reference */
5000
5001 else if ((i = find_parens(cd, name, namelen,
5002 (options & PCRE_EXTENDED) != 0)) > 0)
5003 {
5004 PUT2(code, 2+LINK_SIZE, i);
5005 code[1+LINK_SIZE]++;
5006 }
5007
5008 /* If terminator == 0 it means that the name followed directly after
5009 the opening parenthesis [e.g. (?(abc)...] and in this case there are
5010 some further alternatives to try. For the cases where terminator != 0
5011 [things like (?(<name>... or (?('name')... or (?(R&name)... ] we have
5012 now checked all the possibilities, so give an error. */
5013
5014 else if (terminator != 0)
5015 {
5016 *errorcodeptr = ERR15;
5017 goto FAILED;
5018 }
5019
5020 /* Check for (?(R) for recursion. Allow digits after R to specify a
5021 specific group number. */
5022
5023 else if (*name == CHAR_R)
5024 {
5025 recno = 0;
5026 for (i = 1; i < namelen; i++)
5027 {
5028 if ((digitab[name[i]] & ctype_digit) == 0)
5029 {
5030 *errorcodeptr = ERR15;
5031 goto FAILED;
5032 }
5033 recno = recno * 10 + name[i] - CHAR_0;
5034 }
5035 if (recno == 0) recno = RREF_ANY;
5036 code[1+LINK_SIZE] = OP_RREF; /* Change test type */
5037 PUT2(code, 2+LINK_SIZE, recno);
5038 }
5039
5040 /* Similarly, check for the (?(DEFINE) "condition", which is always
5041 false. */
5042
5043 else if (namelen == 6 && strncmp((char *)name, STRING_DEFINE, 6) == 0)
5044 {
5045 code[1+LINK_SIZE] = OP_DEF;
5046 skipbytes = 1;
5047 }
5048
5049 /* Check for the "name" actually being a subpattern number. We are
5050 in the second pass here, so final_bracount is set. */
5051
5052 else if (recno > 0 && recno <= cd->final_bracount)
5053 {
5054 PUT2(code, 2+LINK_SIZE, recno);
5055 }
5056
5057 /* Either an unidentified subpattern, or a reference to (?(0) */
5058
5059 else
5060 {
5061 *errorcodeptr = (recno == 0)? ERR35: ERR15;
5062 goto FAILED;
5063 }
5064 break;
5065
5066
5067 /* ------------------------------------------------------------ */
5068 case CHAR_EQUALS_SIGN: /* Positive lookahead */
5069 bravalue = OP_ASSERT;
5070 ptr++;
5071 break;
5072
5073
5074 /* ------------------------------------------------------------ */
5075 case CHAR_EXCLAMATION_MARK: /* Negative lookahead */
5076 ptr++;
5077 if (*ptr == CHAR_RIGHT_PARENTHESIS) /* Optimize (?!) */
5078 {
5079 *code++ = OP_FAIL;
5080 previous = NULL;
5081 continue;
5082 }
5083 bravalue = OP_ASSERT_NOT;
5084 break;
5085
5086
5087 /* ------------------------------------------------------------ */
5088 case CHAR_LESS_THAN_SIGN: /* Lookbehind or named define */
5089 switch (ptr[1])
5090 {
5091 case CHAR_EQUALS_SIGN: /* Positive lookbehind */
5092 bravalue = OP_ASSERTBACK;
5093 ptr += 2;
5094 break;
5095
5096 case CHAR_EXCLAMATION_MARK: /* Negative lookbehind */
5097 bravalue = OP_ASSERTBACK_NOT;
5098 ptr += 2;
5099 break;
5100
5101 default: /* Could be name define, else bad */
5102 if ((cd->ctypes[ptr[1]] & ctype_word) != 0) goto DEFINE_NAME;
5103 ptr++; /* Correct offset for error */
5104 *errorcodeptr = ERR24;
5105 goto FAILED;
5106 }
5107 break;
5108
5109
5110 /* ------------------------------------------------------------ */
5111 case CHAR_GREATER_THAN_SIGN: /* One-time brackets */
5112 bravalue = OP_ONCE;
5113 ptr++;
5114 break;
5115
5116
5117 /* ------------------------------------------------------------ */
5118 case CHAR_C: /* Callout - may be followed by digits; */
5119 previous_callout = code; /* Save for later completion */
5120 after_manual_callout = 1; /* Skip one item before completing */
5121 *code++ = OP_CALLOUT;
5122 {
5123 int n = 0;
5124 while ((digitab[*(++ptr)] & ctype_digit) != 0)
5125 n = n * 10 + *ptr - CHAR_0;
5126 if (*ptr != CHAR_RIGHT_PARENTHESIS)
5127 {
5128 *errorcodeptr = ERR39;
5129 goto FAILED;
5130 }
5131 if (n > 255)
5132 {
5133 *errorcodeptr = ERR38;
5134 goto FAILED;
5135 }
5136 *code++ = n;
5137 PUT(code, 0, (int)(ptr - cd->start_pattern + 1)); /* Pattern offset */
5138 PUT(code, LINK_SIZE, 0); /* Default length */
5139 code += 2 * LINK_SIZE;
5140 }
5141 previous = NULL;
5142 continue;
5143
5144
5145 /* ------------------------------------------------------------ */
5146 case CHAR_P: /* Python-style named subpattern handling */
5147 if (*(++ptr) == CHAR_EQUALS_SIGN ||
5148 *ptr == CHAR_GREATER_THAN_SIGN) /* Reference or recursion */
5149 {
5150 is_recurse = *ptr == CHAR_GREATER_THAN_SIGN;
5151 terminator = CHAR_RIGHT_PARENTHESIS;
5152 goto NAMED_REF_OR_RECURSE;
5153 }
5154 else if (*ptr != CHAR_LESS_THAN_SIGN) /* Test for Python-style defn */
5155 {
5156 *errorcodeptr = ERR41;
5157 goto FAILED;
5158 }
5159 /* Fall through to handle (?P< as (?< is handled */
5160
5161
5162 /* ------------------------------------------------------------ */
5163 DEFINE_NAME: /* Come here from (?< handling */
5164 case CHAR_APOSTROPHE:
5165 {
5166 terminator = (*ptr == CHAR_LESS_THAN_SIGN)?
5167 CHAR_GREATER_THAN_SIGN : CHAR_APOSTROPHE;
5168 name = ++ptr;
5169
5170 while ((cd->ctypes[*ptr] & ctype_word) != 0) ptr++;
5171 namelen = (int)(ptr - name);
5172
5173 /* In the pre-compile phase, just do a syntax check. */
5174
5175 if (lengthptr != NULL)
5176 {
5177 if (*ptr != terminator)
5178 {
5179 *errorcodeptr = ERR42;
5180 goto FAILED;
5181 }
5182 if (cd->names_found >= MAX_NAME_COUNT)
5183 {
5184 *errorcodeptr = ERR49;
5185 goto FAILED;
5186 }
5187 if (namelen + 3 > cd->name_entry_size)
5188 {
5189 cd->name_entry_size = namelen + 3;
5190 if (namelen > MAX_NAME_SIZE)
5191 {
5192 *errorcodeptr = ERR48;
5193 goto FAILED;
5194 }
5195 }
5196 }
5197
5198 /* In the real compile, create the entry in the table, maintaining
5199 alphabetical order. Duplicate names for different numbers are
5200 permitted only if PCRE_DUPNAMES is set. Duplicate names for the same
5201 number are always OK. (An existing number can be re-used if (?|
5202 appears in the pattern.) In either event, a duplicate name results in
5203 a duplicate entry in the table, even if the number is the same. This
5204 is because the number of names, and hence the table size, is computed
5205 in the pre-compile, and it affects various numbers and pointers which
5206 would all have to be modified, and the compiled code moved down, if
5207 duplicates with the same number were omitted from the table. This
5208 doesn't seem worth the hassle. However, *different* names for the
5209 same number are not permitted. */
5210
5211 else
5212 {
5213 BOOL dupname = FALSE;
5214 slot = cd->name_table;
5215
5216 for (i = 0; i < cd->names_found; i++)
5217 {
5218 int crc = memcmp(name, slot+2, namelen);
5219 if (crc == 0)
5220 {
5221 if (slot[2+namelen] == 0)
5222 {
5223 if (GET2(slot, 0) != cd->bracount + 1 &&
5224 (options & PCRE_DUPNAMES) == 0)
5225 {
5226 *errorcodeptr = ERR43;
5227 goto FAILED;
5228 }
5229 else dupname = TRUE;
5230 }
5231 else crc = -1; /* Current name is a substring */
5232 }
5233
5234 /* Make space in the table and break the loop for an earlier
5235 name. For a duplicate or later name, carry on. We do this for
5236 duplicates so that in the simple case (when ?(| is not used) they
5237 are in order of their numbers. */
5238
5239 if (crc < 0)
5240 {
5241 memmove(slot + cd->name_entry_size, slot,
5242 (cd->names_found - i) * cd->name_entry_size);
5243 break;
5244 }
5245
5246 /* Continue the loop for a later or duplicate name */
5247
5248 slot += cd->name_entry_size;
5249 }
5250
5251 /* For non-duplicate names, check for a duplicate number before
5252 adding the new name. */
5253
5254 if (!dupname)
5255 {
5256 uschar *cslot = cd->name_table;
5257 for (i = 0; i < cd->names_found; i++)
5258 {
5259 if (cslot != slot)
5260 {
5261 if (GET2(cslot, 0) == cd->bracount + 1)
5262 {
5263 *errorcodeptr = ERR65;
5264 goto FAILED;
5265 }
5266 }
5267 else i--;
5268 cslot += cd->name_entry_size;
5269 }
5270 }
5271
5272 PUT2(slot, 0, cd->bracount + 1);
5273 memcpy(slot + 2, name, namelen);
5274 slot[2+namelen] = 0;
5275 }
5276 }
5277
5278 /* In both pre-compile and compile, count the number of names we've
5279 encountered. */
5280
5281 cd->names_found++;
5282 ptr++; /* Move past > or ' */
5283 goto NUMBERED_GROUP;
5284
5285
5286 /* ------------------------------------------------------------ */
5287 case CHAR_AMPERSAND: /* Perl recursion/subroutine syntax */
5288 terminator = CHAR_RIGHT_PARENTHESIS;
5289 is_recurse = TRUE;
5290 /* Fall through */
5291
5292 /* We come here from the Python syntax above that handles both
5293 references (?P=name) and recursion (?P>name), as well as falling
5294 through from the Perl recursion syntax (?&name). We also come here from
5295 the Perl \k<name> or \k'name' back reference syntax and the \k{name}
5296 .NET syntax, and the Oniguruma \g<...> and \g'...' subroutine syntax. */
5297
5298 NAMED_REF_OR_RECURSE:
5299 name = ++ptr;
5300 while ((cd->ctypes[*ptr] & ctype_word) != 0) ptr++;
5301 namelen = (int)(ptr - name);
5302
5303 /* In the pre-compile phase, do a syntax check and set a dummy
5304 reference number. */
5305
5306 if (lengthptr != NULL)
5307 {
5308 if (namelen == 0)
5309 {
5310 *errorcodeptr = ERR62;
5311 goto FAILED;
5312 }
5313 if (*ptr != terminator)
5314 {
5315 *errorcodeptr = ERR42;
5316 goto FAILED;
5317 }
5318 if (namelen > MAX_NAME_SIZE)
5319 {
5320 *errorcodeptr = ERR48;
5321 goto FAILED;
5322 }
5323 recno = 0;
5324 }
5325
5326 /* In the real compile, seek the name in the table. We check the name
5327 first, and then check that we have reached the end of the name in the
5328 table. That way, if the name that is longer than any in the table,
5329 the comparison will fail without reading beyond the table entry. */
5330
5331 else
5332 {
5333 slot = cd->name_table;
5334 for (i = 0; i < cd->names_found; i++)
5335 {
5336 if (strncmp((char *)name, (char *)slot+2, namelen) == 0 &&
5337 slot[2+namelen] == 0)
5338 break;
5339 slot += cd->name_entry_size;
5340 }
5341
5342 if (i < cd->names_found) /* Back reference */
5343 {
5344 recno = GET2(slot, 0);
5345 }
5346 else if ((recno = /* Forward back reference */
5347 find_parens(cd, name, namelen,
5348 (options & PCRE_EXTENDED) != 0)) <= 0)
5349 {
5350 *errorcodeptr = ERR15;
5351 goto FAILED;
5352 }
5353 }
5354
5355 /* In both phases, we can now go to the code than handles numerical
5356 recursion or backreferences. */
5357
5358 if (is_recurse) goto HANDLE_RECURSION;
5359 else goto HANDLE_REFERENCE;
5360
5361
5362 /* ------------------------------------------------------------ */
5363 case CHAR_R: /* Recursion */
5364 ptr++; /* Same as (?0) */
5365 /* Fall through */
5366
5367
5368 /* ------------------------------------------------------------ */
5369 case CHAR_MINUS: case CHAR_PLUS: /* Recursion or subroutine */
5370 case CHAR_0: case CHAR_1: case CHAR_2: case CHAR_3: case CHAR_4:
5371 case CHAR_5: case CHAR_6: case CHAR_7: case CHAR_8: case CHAR_9:
5372 {
5373 const uschar *called;
5374 terminator = CHAR_RIGHT_PARENTHESIS;
5375
5376 /* Come here from the \g<...> and \g'...' code (Oniguruma
5377 compatibility). However, the syntax has been checked to ensure that
5378 the ... are a (signed) number, so that neither ERR63 nor ERR29 will
5379 be called on this path, nor with the jump to OTHER_CHAR_AFTER_QUERY
5380 ever be taken. */
5381
5382 HANDLE_NUMERICAL_RECURSION:
5383
5384 if ((refsign = *ptr) == CHAR_PLUS)
5385 {
5386 ptr++;
5387 if ((digitab[*ptr] & ctype_digit) == 0)
5388 {
5389 *errorcodeptr = ERR63;
5390 goto FAILED;
5391 }
5392 }
5393 else if (refsign == CHAR_MINUS)
5394 {
5395 if ((digitab[ptr[1]] & ctype_digit) == 0)
5396 goto OTHER_CHAR_AFTER_QUERY;
5397 ptr++;
5398 }
5399
5400 recno = 0;
5401 while((digitab[*ptr] & ctype_digit) != 0)
5402 recno = recno * 10 + *ptr++ - CHAR_0;
5403
5404 if (*ptr != terminator)
5405 {
5406 *errorcodeptr = ERR29;
5407 goto FAILED;
5408 }
5409
5410 if (refsign == CHAR_MINUS)
5411 {
5412 if (recno == 0)
5413 {
5414 *errorcodeptr = ERR58;
5415 goto FAILED;
5416 }
5417 recno = cd->bracount - recno + 1;
5418 if (recno <= 0)
5419 {
5420 *errorcodeptr = ERR15;
5421 goto FAILED;
5422 }
5423 }
5424 else if (refsign == CHAR_PLUS)
5425 {
5426 if (recno == 0)
5427 {
5428 *errorcodeptr = ERR58;
5429 goto FAILED;
5430 }
5431 recno += cd->bracount;
5432 }
5433
5434 /* Come here from code above that handles a named recursion */
5435
5436 HANDLE_RECURSION:
5437
5438 previous = code;
5439 called = cd->start_code;
5440
5441 /* When we are actually compiling, find the bracket that is being
5442 referenced. Temporarily end the regex in case it doesn't exist before
5443 this point. If we end up with a forward reference, first check that
5444 the bracket does occur later so we can give the error (and position)
5445 now. Then remember this forward reference in the workspace so it can
5446 be filled in at the end. */
5447
5448 if (lengthptr == NULL)
5449 {
5450 *code = OP_END;
5451 if (recno != 0)
5452 called = _pcre_find_bracket(cd->start_code, utf8, recno);
5453
5454 /* Forward reference */
5455
5456 if (called == NULL)
5457 {
5458 if (find_parens(cd, NULL, recno,
5459 (options & PCRE_EXTENDED) != 0) < 0)
5460 {
5461 *errorcodeptr = ERR15;
5462 goto FAILED;
5463 }
5464
5465 /* Fudge the value of "called" so that when it is inserted as an
5466 offset below, what it actually inserted is the reference number
5467 of the group. */
5468
5469 called = cd->start_code + recno;
5470 PUTINC(cd->hwm, 0, (int)(code + 2 + LINK_SIZE - cd->start_code));
5471 }
5472
5473 /* If not a forward reference, and the subpattern is still open,
5474 this is a recursive call. We check to see if this is a left
5475 recursion that could loop for ever, and diagnose that case. */
5476
5477 else if (GET(called, 1) == 0 &&
5478 could_be_empty(called, code, bcptr, utf8, cd))
5479 {
5480 *errorcodeptr = ERR40;
5481 goto FAILED;
5482 }
5483 }
5484
5485 /* Insert the recursion/subroutine item, automatically wrapped inside
5486 "once" brackets. Set up a "previous group" length so that a
5487 subsequent quantifier will work. */
5488
5489 *code = OP_ONCE;
5490 PUT(code, 1, 2 + 2*LINK_SIZE);
5491 code += 1 + LINK_SIZE;
5492
5493 *code = OP_RECURSE;
5494 PUT(code, 1, (int)(called - cd->start_code));
5495 code += 1 + LINK_SIZE;
5496
5497 *code = OP_KET;
5498 PUT(code, 1, 2 + 2*LINK_SIZE);
5499 code += 1 + LINK_SIZE;
5500
5501 length_prevgroup = 3 + 3*LINK_SIZE;
5502 }
5503
5504 /* Can't determine a first byte now */
5505
5506 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
5507 continue;
5508
5509
5510 /* ------------------------------------------------------------ */
5511 default: /* Other characters: check option setting */
5512 OTHER_CHAR_AFTER_QUERY:
5513 set = unset = 0;
5514 optset = &set;
5515
5516 while (*ptr != CHAR_RIGHT_PARENTHESIS && *ptr != CHAR_COLON)
5517 {
5518 switch (*ptr++)
5519 {
5520 case CHAR_MINUS: optset = &unset; break;
5521
5522 case CHAR_J: /* Record that it changed in the external options */
5523 *optset |= PCRE_DUPNAMES;
5524 cd->external_flags |= PCRE_JCHANGED;
5525 break;
5526
5527 case CHAR_i: *optset |= PCRE_CASELESS; break;
5528 case CHAR_m: *optset |= PCRE_MULTILINE; break;
5529 case CHAR_s: *optset |= PCRE_DOTALL; break;
5530 case CHAR_x: *optset |= PCRE_EXTENDED; break;
5531 case CHAR_U: *optset |= PCRE_UNGREEDY; break;
5532 case CHAR_X: *optset |= PCRE_EXTRA; break;
5533
5534 default: *errorcodeptr = ERR12;
5535 ptr--; /* Correct the offset */
5536 goto FAILED;
5537 }
5538 }
5539
5540 /* Set up the changed option bits, but don't change anything yet. */
5541
5542 newoptions = (options | set) & (~unset);
5543
5544 /* If the options ended with ')' this is not the start of a nested
5545 group with option changes, so the options change at this level. If this
5546 item is right at the start of the pattern, the options can be
5547 abstracted and made external in the pre-compile phase, and ignored in
5548 the compile phase. This can be helpful when matching -- for instance in
5549 caseless checking of required bytes.
5550
5551 If the code pointer is not (cd->start_code + 1 + LINK_SIZE), we are
5552 definitely *not* at the start of the pattern because something has been
5553 compiled. In the pre-compile phase, however, the code pointer can have
5554 that value after the start, because it gets reset as code is discarded
5555 during the pre-compile. However, this can happen only at top level - if
5556 we are within parentheses, the starting BRA will still be present. At
5557 any parenthesis level, the length value can be used to test if anything
5558 has been compiled at that level. Thus, a test for both these conditions
5559 is necessary to ensure we correctly detect the start of the pattern in
5560 both phases.
5561
5562 If we are not at the pattern start, compile code to change the ims
5563 options if this setting actually changes any of them, and reset the
5564 greedy defaults and the case value for firstbyte and reqbyte. */
5565
5566 if (*ptr == CHAR_RIGHT_PARENTHESIS)
5567 {
5568 if (code == cd->start_code + 1 + LINK_SIZE &&
5569 (lengthptr == NULL || *lengthptr == 2 + 2*LINK_SIZE))
5570 {
5571 cd->external_options = newoptions;
5572 }
5573 else
5574 {
5575 if ((options & PCRE_IMS) != (newoptions & PCRE_IMS))
5576 {
5577 *code++ = OP_OPT;
5578 *code++ = newoptions & PCRE_IMS;
5579 }
5580 greedy_default = ((newoptions & PCRE_UNGREEDY) != 0);
5581 greedy_non_default = greedy_default ^ 1;
5582 req_caseopt = ((newoptions & PCRE_CASELESS) != 0)? REQ_CASELESS : 0;
5583 }
5584
5585 /* Change options at this level, and pass them back for use
5586 in subsequent branches. When not at the start of the pattern, this
5587 information is also necessary so that a resetting item can be
5588 compiled at the end of a group (if we are in a group). */
5589
5590 *optionsptr = options = newoptions;
5591 previous = NULL; /* This item can't be repeated */
5592 continue; /* It is complete */
5593 }
5594
5595 /* If the options ended with ':' we are heading into a nested group
5596 with possible change of options. Such groups are non-capturing and are
5597 not assertions of any kind. All we need to do is skip over the ':';
5598 the newoptions value is handled below. */
5599
5600 bravalue = OP_BRA;
5601 ptr++;
5602 } /* End of switch for character following (? */
5603 } /* End of (? handling */
5604
5605 /* Opening parenthesis not followed by '*' or '?'. If PCRE_NO_AUTO_CAPTURE
5606 is set, all unadorned brackets become non-capturing and behave like (?:...)
5607 brackets. */
5608
5609 else if ((options & PCRE_NO_AUTO_CAPTURE) != 0)
5610 {
5611 bravalue = OP_BRA;
5612 }
5613
5614 /* Else we have a capturing group. */
5615
5616 else
5617 {
5618 NUMBERED_GROUP:
5619 cd->bracount += 1;
5620 PUT2(code, 1+LINK_SIZE, cd->bracount);
5621 skipbytes = 2;
5622 }
5623
5624 /* Process nested bracketed regex. Assertions may not be repeated, but
5625 other kinds can be. All their opcodes are >= OP_ONCE. We copy code into a
5626 non-register variable in order to be able to pass its address because some
5627 compilers complain otherwise. Pass in a new setting for the ims options if
5628 they have changed. */
5629
5630 previous = (bravalue >= OP_ONCE)? code : NULL;
5631 *code = bravalue;
5632 tempcode = code;
5633 tempreqvary = cd->req_varyopt; /* Save value before bracket */
5634 length_prevgroup = 0; /* Initialize for pre-compile phase */
5635
5636 if (!compile_regex(
5637 newoptions, /* The complete new option state */
5638 options & PCRE_IMS, /* The previous ims option state */
5639 &tempcode, /* Where to put code (updated) */
5640 &ptr, /* Input pointer (updated) */
5641 errorcodeptr, /* Where to put an error message */
5642 (bravalue == OP_ASSERTBACK ||
5643 bravalue == OP_ASSERTBACK_NOT), /* TRUE if back assert */
5644 reset_bracount, /* True if (?| group */
5645 skipbytes, /* Skip over bracket number */
5646 &subfirstbyte, /* For possible first char */
5647 &subreqbyte, /* For possible last char */
5648 bcptr, /* Current branch chain */
5649 cd, /* Tables block */
5650 (lengthptr == NULL)? NULL : /* Actual compile phase */
5651 &length_prevgroup /* Pre-compile phase */
5652 ))
5653 goto FAILED;
5654
5655 /* At the end of compiling, code is still pointing to the start of the
5656 group, while tempcode has been updated to point past the end of the group
5657 and any option resetting that may follow it. The pattern pointer (ptr)
5658 is on the bracket. */
5659
5660 /* If this is a conditional bracket, check that there are no more than
5661 two branches in the group, or just one if it's a DEFINE group. We do this
5662 in the real compile phase, not in the pre-pass, where the whole group may
5663 not be available. */
5664
5665 if (bravalue == OP_COND && lengthptr == NULL)
5666 {
5667 uschar *tc = code;
5668 int condcount = 0;
5669
5670 do {
5671 condcount++;
5672 tc += GET(tc,1);
5673 }
5674 while (*tc != OP_KET);
5675
5676 /* A DEFINE group is never obeyed inline (the "condition" is always
5677 false). It must have only one branch. */
5678
5679 if (code[LINK_SIZE+1] == OP_DEF)
5680 {
5681 if (condcount > 1)
5682 {
5683 *errorcodeptr = ERR54;
5684 goto FAILED;
5685 }
5686 bravalue = OP_DEF; /* Just a flag to suppress char handling below */
5687 }
5688
5689 /* A "normal" conditional group. If there is just one branch, we must not
5690 make use of its firstbyte or reqbyte, because this is equivalent to an
5691 empty second branch. */
5692
5693 else
5694 {
5695 if (condcount > 2)
5696 {
5697 *errorcodeptr = ERR27;
5698 goto FAILED;
5699 }
5700 if (condcount == 1) subfirstbyte = subreqbyte = REQ_NONE;
5701 }
5702 }
5703
5704 /* Error if hit end of pattern */
5705
5706 if (*ptr != CHAR_RIGHT_PARENTHESIS)
5707 {
5708 *errorcodeptr = ERR14;
5709 goto FAILED;
5710 }
5711
5712 /* In the pre-compile phase, update the length by the length of the group,
5713 less the brackets at either end. Then reduce the compiled code to just a
5714 set of non-capturing brackets so that it doesn't use much memory if it is
5715 duplicated by a quantifier.*/
5716
5717 if (lengthptr != NULL)
5718 {
5719 if (OFLOW_MAX - *lengthptr < length_prevgroup - 2 - 2*LINK_SIZE)
5720 {
5721 *errorcodeptr = ERR20;
5722 goto FAILED;
5723 }
5724 *lengthptr += length_prevgroup - 2 - 2*LINK_SIZE;
5725 *code++ = OP_BRA;
5726 PUTINC(code, 0, 1 + LINK_SIZE);
5727 *code++ = OP_KET;
5728 PUTINC(code, 0, 1 + LINK_SIZE);
5729 break; /* No need to waste time with special character handling */
5730 }
5731
5732 /* Otherwise update the main code pointer to the end of the group. */
5733
5734 code = tempcode;
5735
5736 /* For a DEFINE group, required and first character settings are not
5737 relevant. */
5738
5739 if (bravalue == OP_DEF) break;
5740
5741 /* Handle updating of the required and first characters for other types of
5742 group. Update for normal brackets of all kinds, and conditions with two
5743 branches (see code above). If the bracket is followed by a quantifier with
5744 zero repeat, we have to back off. Hence the definition of zeroreqbyte and
5745 zerofirstbyte outside the main loop so that they can be accessed for the
5746 back off. */
5747
5748 zeroreqbyte = reqbyte;
5749 zerofirstbyte = firstbyte;
5750 groupsetfirstbyte = FALSE;
5751
5752 if (bravalue >= OP_ONCE)
5753 {
5754 /* If we have not yet set a firstbyte in this branch, take it from the
5755 subpattern, remembering that it was set here so that a repeat of more
5756 than one can replicate it as reqbyte if necessary. If the subpattern has
5757 no firstbyte, set "none" for the whole branch. In both cases, a zero
5758 repeat forces firstbyte to "none". */
5759
5760 if (firstbyte == REQ_UNSET)
5761 {
5762 if (subfirstbyte >= 0)
5763 {
5764 firstbyte = subfirstbyte;
5765 groupsetfirstbyte = TRUE;
5766 }
5767 else firstbyte = REQ_NONE;
5768 zerofirstbyte = REQ_NONE;
5769 }
5770
5771 /* If firstbyte was previously set, convert the subpattern's firstbyte
5772 into reqbyte if there wasn't one, using the vary flag that was in
5773 existence beforehand. */
5774
5775 else if (subfirstbyte >= 0 && subreqbyte < 0)
5776 subreqbyte = subfirstbyte | tempreqvary;
5777
5778 /* If the subpattern set a required byte (or set a first byte that isn't
5779 really the first byte - see above), set it. */
5780
5781 if (subreqbyte >= 0) reqbyte = subreqbyte;
5782 }
5783
5784 /* For a forward assertion, we take the reqbyte, if set. This can be
5785 helpful if the pattern that follows the assertion doesn't set a different
5786 char. For example, it's useful for /(?=abcde).+/. We can't set firstbyte
5787 for an assertion, however because it leads to incorrect effect for patterns
5788 such as /(?=a)a.+/ when the "real" "a" would then become a reqbyte instead
5789 of a firstbyte. This is overcome by a scan at the end if there's no
5790 firstbyte, looking for an asserted first char. */
5791
5792 else if (bravalue == OP_ASSERT && subreqbyte >= 0) reqbyte = subreqbyte;
5793 break; /* End of processing '(' */
5794
5795
5796 /* ===================================================================*/
5797 /* Handle metasequences introduced by \. For ones like \d, the ESC_ values
5798 are arranged to be the negation of the corresponding OP_values in the
5799 default case when PCRE_UCP is not set. For the back references, the values
5800 are ESC_REF plus the reference number. Only back references and those types
5801 that consume a character may be repeated. We can test for values between
5802 ESC_b and ESC_Z for the latter; this may have to change if any new ones are
5803 ever created. */
5804
5805 case CHAR_BACKSLASH:
5806 tempptr = ptr;
5807 c = check_escape(&ptr, errorcodeptr, cd->bracount, options, FALSE);
5808 if (*errorcodeptr != 0) goto FAILED;
5809
5810 if (c < 0)
5811 {
5812 if (-c == ESC_Q) /* Handle start of quoted string */
5813 {
5814 if (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
5815 ptr += 2; /* avoid empty string */
5816 else inescq = TRUE;
5817 continue;
5818 }
5819
5820 if (-c == ESC_E) continue; /* Perl ignores an orphan \E */
5821
5822 /* For metasequences that actually match a character, we disable the
5823 setting of a first character if it hasn't already been set. */
5824
5825 if (firstbyte == REQ_UNSET && -c > ESC_b && -c < ESC_Z)
5826 firstbyte = REQ_NONE;
5827
5828 /* Set values to reset to if this is followed by a zero repeat. */
5829
5830 zerofirstbyte = firstbyte;
5831 zeroreqbyte = reqbyte;
5832
5833 /* \g<name> or \g'name' is a subroutine call by name and \g<n> or \g'n'
5834 is a subroutine call by number (Oniguruma syntax). In fact, the value
5835 -ESC_g is returned only for these cases. So we don't need to check for <
5836 or ' if the value is -ESC_g. For the Perl syntax \g{n} the value is
5837 -ESC_REF+n, and for the Perl syntax \g{name} the result is -ESC_k (as
5838 that is a synonym for a named back reference). */
5839
5840 if (-c == ESC_g)
5841 {
5842 const uschar *p;
5843 save_hwm = cd->hwm; /* Normally this is set when '(' is read */
5844 terminator = (*(++ptr) == CHAR_LESS_THAN_SIGN)?
5845 CHAR_GREATER_THAN_SIGN : CHAR_APOSTROPHE;
5846
5847 /* These two statements stop the compiler for warning about possibly
5848 unset variables caused by the jump to HANDLE_NUMERICAL_RECURSION. In
5849 fact, because we actually check for a number below, the paths that
5850 would actually be in error are never taken. */
5851
5852 skipbytes = 0;
5853 reset_bracount = FALSE;
5854
5855 /* Test for a name */
5856
5857 if (ptr[1] != CHAR_PLUS && ptr[1] != CHAR_MINUS)
5858 {
5859 BOOL isnumber = TRUE;
5860 for (p = ptr + 1; *p != 0 && *p != terminator; p++)
5861 {
5862 if ((cd->ctypes[*p] & ctype_digit) == 0) isnumber = FALSE;
5863 if ((cd->ctypes[*p] & ctype_word) == 0) break;
5864 }
5865 if (*p != terminator)
5866 {
5867 *errorcodeptr = ERR57;
5868 break;
5869 }
5870 if (isnumber)
5871 {
5872 ptr++;
5873 goto HANDLE_NUMERICAL_RECURSION;
5874 }
5875 is_recurse = TRUE;
5876 goto NAMED_REF_OR_RECURSE;
5877 }
5878
5879 /* Test a signed number in angle brackets or quotes. */
5880
5881 p = ptr + 2;
5882 while ((digitab[*p] & ctype_digit) != 0) p++;
5883 if (*p != terminator)
5884 {
5885 *errorcodeptr = ERR57;
5886 break;
5887 }
5888 ptr++;
5889 goto HANDLE_NUMERICAL_RECURSION;
5890 }
5891
5892 /* \k<name> or \k'name' is a back reference by name (Perl syntax).
5893 We also support \k{name} (.NET syntax) */
5894
5895 if (-c == ESC_k && (ptr[1] == CHAR_LESS_THAN_SIGN ||
5896 ptr[1] == CHAR_APOSTROPHE || ptr[1] == CHAR_LEFT_CURLY_BRACKET))
5897 {
5898 is_recurse = FALSE;
5899 terminator = (*(++ptr) == CHAR_LESS_THAN_SIGN)?
5900 CHAR_GREATER_THAN_SIGN : (*ptr == CHAR_APOSTROPHE)?
5901 CHAR_APOSTROPHE : CHAR_RIGHT_CURLY_BRACKET;
5902 goto NAMED_REF_OR_RECURSE;
5903 }
5904
5905 /* Back references are handled specially; must disable firstbyte if
5906 not set to cope with cases like (?=(\w+))\1: which would otherwise set
5907 ':' later. */
5908
5909 if (-c >= ESC_REF)
5910 {
5911 open_capitem *oc;
5912 recno = -c - ESC_REF;
5913
5914 HANDLE_REFERENCE: /* Come here from named backref handling */
5915 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
5916 previous = code;
5917 *code++ = OP_REF;
5918 PUT2INC(code, 0, recno);
5919 cd->backref_map |= (recno < 32)? (1 << recno) : 1;
5920 if (recno > cd->top_backref) cd->top_backref = recno;
5921
5922 /* Check to see if this back reference is recursive, that it, it
5923 is inside the group that it references. A flag is set so that the
5924 group can be made atomic. */
5925
5926 for (oc = cd->open_caps; oc != NULL; oc = oc->next)
5927 {
5928 if (oc->number == recno)
5929 {
5930 oc->flag = TRUE;
5931 break;
5932 }
5933 }
5934 }
5935
5936 /* So are Unicode property matches, if supported. */
5937
5938 #ifdef SUPPORT_UCP
5939 else if (-c == ESC_P || -c == ESC_p)
5940 {
5941 BOOL negated;
5942 int pdata;
5943 int ptype = get_ucp(&ptr, &negated, &pdata, errorcodeptr);
5944 if (ptype < 0) goto FAILED;
5945 previous = code;
5946 *code++ = ((-c == ESC_p) != negated)? OP_PROP : OP_NOTPROP;
5947 *code++ = ptype;
5948 *code++ = pdata;
5949 }
5950 #else
5951
5952 /* If Unicode properties are not supported, \X, \P, and \p are not
5953 allowed. */
5954
5955 else if (-c == ESC_X || -c == ESC_P || -c == ESC_p)
5956 {
5957 *errorcodeptr = ERR45;
5958 goto FAILED;
5959 }
5960 #endif
5961
5962 /* For the rest (including \X when Unicode properties are supported), we
5963 can obtain the OP value by negating the escape value in the default
5964 situation when PCRE_UCP is not set. When it *is* set, we substitute
5965 Unicode property tests. */
5966
5967 else
5968 {
5969 #ifdef SUPPORT_UCP
5970 if (-c >= ESC_DU && -c <= ESC_wu)
5971 {
5972 nestptr = ptr + 1; /* Where to resume */
5973 ptr = substitutes[-c - ESC_DU] - 1; /* Just before substitute */
5974 }
5975 else
5976 #endif
5977 {
5978 previous = (-c > ESC_b && -c < ESC_Z)? code : NULL;
5979 *code++ = -c;
5980 }
5981 }
5982 continue;
5983 }
5984
5985 /* We have a data character whose value is in c. In UTF-8 mode it may have
5986 a value > 127. We set its representation in the length/buffer, and then
5987 handle it as a data character. */
5988
5989 #ifdef SUPPORT_UTF8
5990 if (utf8 && c > 127)
5991 mclength = _pcre_ord2utf8(c, mcbuffer);
5992 else
5993 #endif
5994
5995 {
5996 mcbuffer[0] = c;
5997 mclength = 1;
5998 }
5999 goto ONE_CHAR;
6000
6001
6002 /* ===================================================================*/
6003 /* Handle a literal character. It is guaranteed not to be whitespace or #
6004 when the extended flag is set. If we are in UTF-8 mode, it may be a
6005 multi-byte literal character. */
6006
6007 default:
6008 NORMAL_CHAR:
6009 mclength = 1;
6010 mcbuffer[0] = c;
6011
6012 #ifdef SUPPORT_UTF8
6013 if (utf8 && c >= 0xc0)
6014 {
6015 while ((ptr[1] & 0xc0) == 0x80)
6016 mcbuffer[mclength++] = *(++ptr);
6017 }
6018 #endif
6019
6020 /* At this point we have the character's bytes in mcbuffer, and the length
6021 in mclength. When not in UTF-8 mode, the length is always 1. */
6022
6023 ONE_CHAR:
6024 previous = code;
6025 *code++ = ((options & PCRE_CASELESS) != 0)? OP_CHARNC : OP_CHAR;
6026 for (c = 0; c < mclength; c++) *code++ = mcbuffer[c];
6027
6028 /* Remember if \r or \n were seen */
6029
6030 if (mcbuffer[0] == CHAR_CR || mcbuffer[0] == CHAR_NL)
6031 cd->external_flags |= PCRE_HASCRORLF;
6032
6033 /* Set the first and required bytes appropriately. If no previous first
6034 byte, set it from this character, but revert to none on a zero repeat.
6035 Otherwise, leave the firstbyte value alone, and don't change it on a zero
6036 repeat. */
6037
6038 if (firstbyte == REQ_UNSET)
6039 {
6040 zerofirstbyte = REQ_NONE;
6041 zeroreqbyte = reqbyte;
6042
6043 /* If the character is more than one byte long, we can set firstbyte
6044 only if it is not to be matched caselessly. */
6045
6046 if (mclength == 1 || req_caseopt == 0)
6047 {
6048 firstbyte = mcbuffer[0] | req_caseopt;
6049 if (mclength != 1) reqbyte = code[-1] | cd->req_varyopt;
6050 }
6051 else firstbyte = reqbyte = REQ_NONE;
6052 }
6053
6054 /* firstbyte was previously set; we can set reqbyte only the length is
6055 1 or the matching is caseful. */
6056
6057 else
6058 {
6059 zerofirstbyte = firstbyte;
6060 zeroreqbyte = reqbyte;
6061 if (mclength == 1 || req_caseopt == 0)
6062 reqbyte = code[-1] | req_caseopt | cd->req_varyopt;
6063 }
6064
6065 break; /* End of literal character handling */
6066 }
6067 } /* end of big loop */
6068
6069
6070 /* Control never reaches here by falling through, only by a goto for all the
6071 error states. Pass back the position in the pattern so that it can be displayed
6072 to the user for diagnosing the error. */
6073
6074 FAILED:
6075 *ptrptr = ptr;
6076 return FALSE;
6077 }
6078
6079
6080
6081
6082 /*************************************************
6083 * Compile sequence of alternatives *
6084 *************************************************/
6085
6086 /* On entry, ptr is pointing past the bracket character, but on return it
6087 points to the closing bracket, or vertical bar, or end of string. The code
6088 variable is pointing at the byte into which the BRA operator has been stored.
6089 If the ims options are changed at the start (for a (?ims: group) or during any
6090 branch, we need to insert an OP_OPT item at the start of every following branch
6091 to ensure they get set correctly at run time, and also pass the new options
6092 into every subsequent branch compile.
6093
6094 This function is used during the pre-compile phase when we are trying to find
6095 out the amount of memory needed, as well as during the real compile phase. The
6096 value of lengthptr distinguishes the two phases.
6097
6098 Arguments:
6099 options option bits, including any changes for this subpattern
6100 oldims previous settings of ims option bits
6101 codeptr -> the address of the current code pointer
6102 ptrptr -> the address of the current pattern pointer
6103 errorcodeptr -> pointer to error code variable
6104 lookbehind TRUE if this is a lookbehind assertion
6105 reset_bracount TRUE to reset the count for each branch
6106 skipbytes skip this many bytes at start (for brackets and OP_COND)
6107 firstbyteptr place to put the first required character, or a negative number
6108 reqbyteptr place to put the last required character, or a negative number
6109 bcptr pointer to the chain of currently open branches
6110 cd points to the data block with tables pointers etc.
6111 lengthptr NULL during the real compile phase
6112 points to length accumulator during pre-compile phase
6113
6114 Returns: TRUE on success
6115 */
6116
6117 static BOOL
6118 compile_regex(int options, int oldims, uschar **codeptr, const uschar **ptrptr,
6119 int *errorcodeptr, BOOL lookbehind, BOOL reset_bracount, int skipbytes,
6120 int *firstbyteptr, int *reqbyteptr, branch_chain *bcptr, compile_data *cd,
6121 int *lengthptr)
6122 {
6123 const uschar *ptr = *ptrptr;
6124 uschar *code = *codeptr;
6125 uschar *last_branch = code;
6126 uschar *start_bracket = code;
6127 uschar *reverse_count = NULL;
6128 open_capitem capitem;
6129 int capnumber = 0;
6130 int firstbyte, reqbyte;
6131 int branchfirstbyte, branchreqbyte;
6132 int length;
6133 int orig_bracount;
6134 int max_bracount;
6135 int old_external_options = cd->external_options;
6136 branch_chain bc;
6137
6138 bc.outer = bcptr;
6139 bc.current_branch = code;
6140
6141 firstbyte = reqbyte = REQ_UNSET;
6142
6143 /* Accumulate the length for use in the pre-compile phase. Start with the
6144 length of the BRA and KET and any extra bytes that are required at the
6145 beginning. We accumulate in a local variable to save frequent testing of
6146 lenthptr for NULL. We cannot do this by looking at the value of code at the
6147 start and end of each alternative, because compiled items are discarded during
6148 the pre-compile phase so that the work space is not exceeded. */
6149
6150 length = 2 + 2*LINK_SIZE + skipbytes;
6151
6152 /* WARNING: If the above line is changed for any reason, you must also change
6153 the code that abstracts option settings at the start of the pattern and makes
6154 them global. It tests the value of length for (2 + 2*LINK_SIZE) in the
6155 pre-compile phase to find out whether anything has yet been compiled or not. */
6156
6157 /* If this is a capturing subpattern, add to the chain of open capturing items
6158 so that we can detect them if (*ACCEPT) is encountered. This is also used to
6159 detect groups that contain recursive back references to themselves. */
6160
6161 if (*code == OP_CBRA)
6162 {
6163 capnumber = GET2(code, 1 + LINK_SIZE);
6164 capitem.number = capnumber;
6165 capitem.next = cd->open_caps;
6166 capitem.flag = FALSE;
6167 cd->open_caps = &capitem;
6168 }
6169
6170 /* Offset is set zero to mark that this bracket is still open */
6171
6172 PUT(code, 1, 0);
6173 code += 1 + LINK_SIZE + skipbytes;
6174
6175 /* Loop for each alternative branch */
6176
6177 orig_bracount = max_bracount = cd->bracount;
6178 for (;;)
6179 {
6180 /* For a (?| group, reset the capturing bracket count so that each branch
6181 uses the same numbers. */
6182
6183 if (reset_bracount) cd->bracount = orig_bracount;
6184
6185 /* Handle a change of ims options at the start of the branch */
6186
6187 if ((options & PCRE_IMS) != oldims)
6188 {
6189 *code++ = OP_OPT;
6190 *code++ = options & PCRE_IMS;
6191 length += 2;
6192 }
6193
6194 /* Set up dummy OP_REVERSE if lookbehind assertion */
6195
6196 if (lookbehind)
6197 {
6198 *code++ = OP_REVERSE;
6199 reverse_count = code;
6200 PUTINC(code, 0, 0);
6201 length += 1 + LINK_SIZE;
6202 }
6203
6204 /* Now compile the branch; in the pre-compile phase its length gets added
6205 into the length. */
6206
6207 if (!compile_branch(&options, &code, &ptr, errorcodeptr, &branchfirstbyte,
6208 &branchreqbyte, &bc, cd, (lengthptr == NULL)? NULL : &length))
6209 {
6210 *ptrptr = ptr;
6211 return FALSE;
6212 }
6213
6214 /* If the external options have changed during this branch, it means that we
6215 are at the top level, and a leading option setting has been encountered. We
6216 need to re-set the original option values to take account of this so that,
6217 during the pre-compile phase, we know to allow for a re-set at the start of
6218 subsequent branches. */
6219
6220 if (old_external_options != cd->external_options)
6221 oldims = cd->external_options & PCRE_IMS;
6222
6223 /* Keep the highest bracket count in case (?| was used and some branch
6224 has fewer than the rest. */
6225
6226 if (cd->bracount > max_bracount) max_bracount = cd->bracount;
6227
6228 /* In the real compile phase, there is some post-processing to be done. */
6229
6230 if (lengthptr == NULL)
6231 {
6232 /* If this is the first branch, the firstbyte and reqbyte values for the
6233 branch become the values for the regex. */
6234
6235 if (*last_branch != OP_ALT)
6236 {
6237 firstbyte = branchfirstbyte;
6238 reqbyte = branchreqbyte;
6239 }
6240
6241 /* If this is not the first branch, the first char and reqbyte have to
6242 match the values from all the previous branches, except that if the
6243 previous value for reqbyte didn't have REQ_VARY set, it can still match,
6244 and we set REQ_VARY for the regex. */
6245
6246 else
6247 {
6248 /* If we previously had a firstbyte, but it doesn't match the new branch,
6249 we have to abandon the firstbyte for the regex, but if there was
6250 previously no reqbyte, it takes on the value of the old firstbyte. */
6251
6252 if (firstbyte >= 0 && firstbyte != branchfirstbyte)
6253 {
6254 if (reqbyte < 0) reqbyte = firstbyte;
6255 firstbyte = REQ_NONE;
6256 }
6257
6258 /* If we (now or from before) have no firstbyte, a firstbyte from the
6259 branch becomes a reqbyte if there isn't a branch reqbyte. */
6260
6261 if (firstbyte < 0 && branchfirstbyte >= 0 && branchreqbyte < 0)
6262 branchreqbyte = branchfirstbyte;
6263
6264 /* Now ensure that the reqbytes match */
6265
6266 if ((reqbyte & ~REQ_VARY) != (branchreqbyte & ~REQ_VARY))
6267 reqbyte = REQ_NONE;
6268 else reqbyte |= branchreqbyte; /* To "or" REQ_VARY */
6269 }
6270
6271 /* If lookbehind, check that this branch matches a fixed-length string, and
6272 put the length into the OP_REVERSE item. Temporarily mark the end of the
6273 branch with OP_END. If the branch contains OP_RECURSE, the result is -3
6274 because there may be forward references that we can't check here. Set a
6275 flag to cause another lookbehind check at the end. Why not do it all at the
6276 end? Because common, erroneous checks are picked up here and the offset of
6277 the problem can be shown. */
6278
6279 if (lookbehind)
6280 {
6281 int fixed_length;
6282 *code = OP_END;
6283 fixed_length = find_fixedlength(last_branch, options, FALSE, cd);
6284 DPRINTF(("fixed length = %d\n", fixed_length));
6285 if (fixed_length == -3)
6286 {
6287 cd->check_lookbehind = TRUE;
6288 }
6289 else if (fixed_length < 0)
6290 {
6291 *errorcodeptr = (fixed_length == -2)? ERR36 : ERR25;
6292 *ptrptr = ptr;
6293 return FALSE;
6294 }
6295 else { PUT(reverse_count, 0, fixed_length); }
6296 }
6297 }
6298
6299 /* Reached end of expression, either ')' or end of pattern. In the real
6300 compile phase, go back through the alternative branches and reverse the chain
6301 of offsets, with the field in the BRA item now becoming an offset to the
6302 first alternative. If there are no alternatives, it points to the end of the
6303 group. The length in the terminating ket is always the length of the whole
6304 bracketed item. If any of the ims options were changed inside the group,
6305 compile a resetting op-code following, except at the very end of the pattern.
6306 Return leaving the pointer at the terminating char. */
6307
6308 if (*ptr != CHAR_VERTICAL_LINE)
6309 {
6310 if (lengthptr == NULL)
6311 {
6312 int branch_length = (int)(code - last_branch);
6313 do
6314 {
6315 int prev_length = GET(last_branch, 1);
6316 PUT(last_branch, 1, branch_length);
6317 branch_length = prev_length;
6318 last_branch -= branch_length;
6319 }
6320 while (branch_length > 0);
6321 }
6322
6323 /* Fill in the ket */
6324
6325 *code = OP_KET;
6326 PUT(code, 1, (int)(code - start_bracket));
6327 code += 1 + LINK_SIZE;
6328
6329 /* If it was a capturing subpattern, check to see if it contained any
6330 recursive back references. If so, we must wrap it in atomic brackets.
6331 In any event, remove the block from the chain. */
6332
6333 if (capnumber > 0)
6334 {
6335 if (cd->open_caps->flag)
6336 {
6337 memmove(start_bracket + 1 + LINK_SIZE, start_bracket,
6338 code - start_bracket);
6339 *start_bracket = OP_ONCE;
6340 code += 1 + LINK_SIZE;
6341 PUT(start_bracket, 1, (int)(code - start_bracket));
6342 *code = OP_KET;
6343 PUT(code, 1, (int)(code - start_bracket));
6344 code += 1 + LINK_SIZE;
6345 length += 2 + 2*LINK_SIZE;
6346 }
6347 cd->open_caps = cd->open_caps->next;
6348 }
6349
6350 /* Reset options if needed. */
6351
6352 if ((options & PCRE_IMS) != oldims && *ptr == CHAR_RIGHT_PARENTHESIS)
6353 {
6354 *code++ = OP_OPT;
6355 *code++ = oldims;
6356 length += 2;
6357 }
6358
6359 /* Retain the highest bracket number, in case resetting was used. */
6360
6361 cd->bracount = max_bracount;
6362
6363 /* Set values to pass back */
6364
6365 *codeptr = code;
6366 *ptrptr = ptr;
6367 *firstbyteptr = firstbyte;
6368 *reqbyteptr = reqbyte;
6369 if (lengthptr != NULL)
6370 {
6371 if (OFLOW_MAX - *lengthptr < length)
6372 {
6373 *errorcodeptr = ERR20;
6374 return FALSE;
6375 }
6376 *lengthptr += length;
6377 }
6378 return TRUE;
6379 }
6380
6381 /* Another branch follows. In the pre-compile phase, we can move the code
6382 pointer back to where it was for the start of the first branch. (That is,
6383 pretend that each branch is the only one.)
6384
6385 In the real compile phase, insert an ALT node. Its length field points back
6386 to the previous branch while the bracket remains open. At the end the chain
6387 is reversed. It's done like this so that the start of the bracket has a
6388 zero offset until it is closed, making it possible to detect recursion. */
6389
6390 if (lengthptr != NULL)
6391 {
6392 code = *codeptr + 1 + LINK_SIZE + skipbytes;
6393 length += 1 + LINK_SIZE;
6394 }
6395 else
6396 {
6397 *code = OP_ALT;
6398 PUT(code, 1, (int)(code - last_branch));
6399 bc.current_branch = last_branch = code;
6400 code += 1 + LINK_SIZE;
6401 }
6402
6403 ptr++;
6404 }
6405 /* Control never reaches here */
6406 }
6407
6408
6409
6410
6411 /*************************************************
6412 * Check for anchored expression *
6413 *************************************************/
6414
6415 /* Try to find out if this is an anchored regular expression. Consider each
6416 alternative branch. If they all start with OP_SOD or OP_CIRC, or with a bracket
6417 all of whose alternatives start with OP_SOD or OP_CIRC (recurse ad lib), then
6418 it's anchored. However, if this is a multiline pattern, then only OP_SOD
6419 counts, since OP_CIRC can match in the middle.
6420
6421 We can also consider a regex to be anchored if OP_SOM starts all its branches.
6422 This is the code for \G, which means "match at start of match position, taking
6423 into account the match offset".
6424
6425 A branch is also implicitly anchored if it starts with .* and DOTALL is set,
6426 because that will try the rest of the pattern at all possible matching points,
6427 so there is no point trying again.... er ....
6428
6429 .... except when the .* appears inside capturing parentheses, and there is a
6430 subsequent back reference to those parentheses. We haven't enough information
6431 to catch that case precisely.
6432
6433 At first, the best we could do was to detect when .* was in capturing brackets
6434 and the highest back reference was greater than or equal to that level.
6435 However, by keeping a bitmap of the first 31 back references, we can catch some
6436 of the more common cases more precisely.
6437
6438 Arguments:
6439 code points to start of expression (the bracket)
6440 options points to the options setting
6441 bracket_map a bitmap of which brackets we are inside while testing; this
6442 handles up to substring 31; after that we just have to take
6443 the less precise approach
6444 backref_map the back reference bitmap
6445
6446 Returns: TRUE or FALSE
6447 */
6448
6449 static BOOL
6450 is_anchored(register const uschar *code, int *options, unsigned int bracket_map,
6451 unsigned int backref_map)
6452 {
6453 do {
6454 const uschar *scode = first_significant_code(code + _pcre_OP_lengths[*code],
6455 options, PCRE_MULTILINE, FALSE);
6456 register int op = *scode;
6457
6458 /* Non-capturing brackets */
6459
6460 if (op == OP_BRA)
6461 {
6462 if (!is_anchored(scode, options, bracket_map, backref_map)) return FALSE;
6463 }
6464
6465 /* Capturing brackets */
6466
6467 else if (op == OP_CBRA)
6468 {
6469 int n = GET2(scode, 1+LINK_SIZE);
6470 int new_map = bracket_map | ((n < 32)? (1 << n) : 1);
6471 if (!is_anchored(scode, options, new_map, backref_map)) return FALSE;
6472 }
6473
6474 /* Other brackets */
6475
6476 else if (op == OP_ASSERT || op == OP_ONCE || op == OP_COND)
6477 {
6478 if (!is_anchored(scode, options, bracket_map, backref_map)) return FALSE;
6479 }
6480
6481 /* .* is not anchored unless DOTALL is set (which generates OP_ALLANY) and
6482 it isn't in brackets that are or may be referenced. */
6483
6484 else if ((op == OP_TYPESTAR || op == OP_TYPEMINSTAR ||
6485 op == OP_TYPEPOSSTAR))
6486 {
6487 if (scode[1] != OP_ALLANY || (bracket_map & backref_map) != 0)
6488 return FALSE;
6489 }
6490
6491 /* Check for explicit anchoring */
6492
6493 else if (op != OP_SOD && op != OP_SOM &&
6494 ((*options & PCRE_MULTILINE) != 0 || op != OP_CIRC))
6495 return FALSE;
6496 code += GET(code, 1);
6497 }
6498 while (*code == OP_ALT); /* Loop for each alternative */
6499 return TRUE;
6500 }
6501
6502
6503
6504 /*************************************************
6505 * Check for starting with ^ or .* *
6506 *************************************************/
6507
6508 /* This is called to find out if every branch starts with ^ or .* so that
6509 "first char" processing can be done to speed things up in multiline
6510 matching and for non-DOTALL patterns that start with .* (which must start at
6511 the beginning or after \n). As in the case of is_anchored() (see above), we
6512 have to take account of back references to capturing brackets that contain .*
6513 because in that case we can't make the assumption.
6514
6515 Arguments:
6516 code points to start of expression (the bracket)
6517 bracket_map a bitmap of which brackets we are inside while testing; this
6518 handles up to substring 31; after that we just have to take
6519 the less precise approach
6520 backref_map the back reference bitmap
6521
6522 Returns: TRUE or FALSE
6523 */
6524
6525 static BOOL
6526 is_startline(const uschar *code, unsigned int bracket_map,
6527 unsigned int backref_map)
6528 {
6529 do {
6530 const uschar *scode = first_significant_code(code + _pcre_OP_lengths[*code],
6531 NULL, 0, FALSE);
6532 register int op = *scode;
6533
6534 /* If we are at the start of a conditional assertion group, *both* the
6535 conditional assertion *and* what follows the condition must satisfy the test
6536 for start of line. Other kinds of condition fail. Note that there may be an
6537 auto-callout at the start of a condition. */
6538
6539 if (op == OP_COND)
6540 {
6541 scode += 1 + LINK_SIZE;
6542 if (*scode == OP_CALLOUT) scode += _pcre_OP_lengths[OP_CALLOUT];
6543 switch (*scode)
6544 {
6545 case OP_CREF:
6546 case OP_NCREF:
6547 case OP_RREF:
6548 case OP_NRREF:
6549 case OP_DEF:
6550 return FALSE;
6551
6552 default: /* Assertion */
6553 if (!is_startline(scode, bracket_map, backref_map)) return FALSE;
6554 do scode += GET(scode, 1); while (*scode == OP_ALT);
6555 scode += 1 + LINK_SIZE;
6556 break;
6557 }
6558 scode = first_significant_code(scode, NULL, 0, FALSE);
6559 op = *scode;
6560 }
6561
6562 /* Non-capturing brackets */
6563
6564 if (op == OP_BRA)
6565 {
6566 if (!is_startline(scode, bracket_map, backref_map)) return FALSE;
6567 }
6568
6569 /* Capturing brackets */
6570
6571 else if (op == OP_CBRA)
6572 {
6573 int n = GET2(scode, 1+LINK_SIZE);
6574 int new_map = bracket_map | ((n < 32)? (1 << n) : 1);
6575 if (!is_startline(scode, new_map, backref_map)) return FALSE;
6576 }
6577
6578 /* Other brackets */
6579
6580 else if (op == OP_ASSERT || op == OP_ONCE)
6581 {
6582 if (!is_startline(scode, bracket_map, backref_map)) return FALSE;
6583 }
6584
6585 /* .* means "start at start or after \n" if it isn't in brackets that
6586 may be referenced. */
6587
6588 else if (op == OP_TYPESTAR || op == OP_TYPEMINSTAR || op == OP_TYPEPOSSTAR)
6589 {
6590 if (scode[1] != OP_ANY || (bracket_map & backref_map) != 0) return FALSE;
6591 }
6592
6593 /* Check for explicit circumflex */
6594
6595 else if (op != OP_CIRC) return FALSE;
6596
6597 /* Move on to the next alternative */
6598
6599 code += GET(code, 1);
6600 }
6601 while (*code == OP_ALT); /* Loop for each alternative */
6602 return TRUE;
6603 }
6604
6605
6606
6607 /*************************************************
6608 * Check for asserted fixed first char *
6609 *************************************************/
6610
6611 /* During compilation, the "first char" settings from forward assertions are
6612 discarded, because they can cause conflicts with actual literals that follow.
6613 However, if we end up without a first char setting for an unanchored pattern,
6614 it is worth scanning the regex to see if there is an initial asserted first
6615 char. If all branches start with the same asserted char, or with a bracket all
6616 of whose alternatives start with the same asserted char (recurse ad lib), then
6617 we return that char, otherwise -1.
6618
6619 Arguments:
6620 code points to start of expression (the bracket)
6621 options pointer to the options (used to check casing changes)
6622 inassert TRUE if in an assertion
6623
6624 Returns: -1 or the fixed first char
6625 */
6626
6627 static int
6628 find_firstassertedchar(const uschar *code, int *options, BOOL inassert)
6629 {
6630 register int c = -1;
6631 do {
6632 int d;
6633 const uschar *scode =
6634 first_significant_code(code + 1+LINK_SIZE, options, PCRE_CASELESS, TRUE);
6635 register int op = *scode;
6636
6637 switch(op)
6638 {
6639 default:
6640 return -1;
6641
6642 case OP_BRA:
6643 case OP_CBRA:
6644 case OP_ASSERT:
6645 case OP_ONCE:
6646 case OP_COND:
6647 if ((d = find_firstassertedchar(scode, options, op == OP_ASSERT)) < 0)
6648 return -1;
6649 if (c < 0) c = d; else if (c != d) return -1;
6650 break;
6651
6652 case OP_EXACT: /* Fall through */
6653 scode += 2;
6654
6655 case OP_CHAR:
6656 case OP_CHARNC:
6657 case OP_PLUS:
6658 case OP_MINPLUS:
6659 case OP_POSPLUS:
6660 if (!inassert) return -1;
6661 if (c < 0)
6662 {
6663 c = scode[1];
6664 if ((*options & PCRE_CASELESS) != 0) c |= REQ_CASELESS;
6665 }
6666 else if (c != scode[1]) return -1;