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

Contents of /code/trunk/pcre_compile.c

Parent Directory Parent Directory | Revision Log Revision Log


Revision 691 - (show annotations) (download)
Sun Sep 11 14:31:21 2011 UTC (3 years ago) by ph10
File MIME type: text/plain
File size: 243760 byte(s)
Final source and document tidies for 8.20-RC1.

1 /*************************************************
2 * Perl-Compatible Regular Expressions *
3 *************************************************/
4
5 /* PCRE is a library of functions to support regular expressions whose syntax
6 and semantics are as close as possible to those of the Perl 5 language.
7
8 Written by Philip Hazel
9 Copyright (c) 1997-2011 University of Cambridge
10
11 -----------------------------------------------------------------------------
12 Redistribution and use in source and binary forms, with or without
13 modification, are permitted provided that the following conditions are met:
14
15 * Redistributions of source code must retain the above copyright notice,
16 this list of conditions and the following disclaimer.
17
18 * Redistributions in binary form must reproduce the above copyright
19 notice, this list of conditions and the following disclaimer in the
20 documentation and/or other materials provided with the distribution.
21
22 * Neither the name of the University of Cambridge nor the names of its
23 contributors may be used to endorse or promote products derived from
24 this software without specific prior written permission.
25
26 THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
27 AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28 IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29 ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE
30 LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
31 CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
32 SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
33 INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
34 CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
35 ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
36 POSSIBILITY OF SUCH DAMAGE.
37 -----------------------------------------------------------------------------
38 */
39
40
41 /* This module contains the external function pcre_compile(), along with
42 supporting internal functions that are not used by other modules. */
43
44
45 #ifdef HAVE_CONFIG_H
46 #include "config.h"
47 #endif
48
49 #define NLBLOCK cd /* Block containing newline information */
50 #define PSSTART start_pattern /* Field containing processed string start */
51 #define PSEND end_pattern /* Field containing processed string end */
52
53 #include "pcre_internal.h"
54
55
56 /* When PCRE_DEBUG is defined, we need the pcre_printint() function, which is
57 also used by pcretest. PCRE_DEBUG is not defined when building a production
58 library. */
59
60 #ifdef PCRE_DEBUG
61 #include "pcre_printint.src"
62 #endif
63
64
65 /* Macro for setting individual bits in class bitmaps. */
66
67 #define SETBIT(a,b) a[b/8] |= (1 << (b%8))
68
69 /* Maximum length value to check against when making sure that the integer that
70 holds the compiled pattern length does not overflow. We make it a bit less than
71 INT_MAX to allow for adding in group terminating bytes, so that we don't have
72 to check them every time. */
73
74 #define OFLOW_MAX (INT_MAX - 20)
75
76
77 /*************************************************
78 * Code parameters and static tables *
79 *************************************************/
80
81 /* This value specifies the size of stack workspace that is used during the
82 first pre-compile phase that determines how much memory is required. The regex
83 is partly compiled into this space, but the compiled parts are discarded as
84 soon as they can be, so that hopefully there will never be an overrun. The code
85 does, however, check for an overrun. The largest amount I've seen used is 218,
86 so this number is very generous.
87
88 The same workspace is used during the second, actual compile phase for
89 remembering forward references to groups so that they can be filled in at the
90 end. Each entry in this list occupies LINK_SIZE bytes, so even when LINK_SIZE
91 is 4 there is plenty of room. */
92
93 #define COMPILE_WORK_SIZE (4096)
94
95 /* The overrun tests check for a slightly smaller size so that they detect the
96 overrun before it actually does run off the end of the data block. */
97
98 #define WORK_SIZE_CHECK (COMPILE_WORK_SIZE - 100)
99
100
101 /* Table for handling escaped characters in the range '0'-'z'. Positive returns
102 are simple data values; negative values are for special things like \d and so
103 on. Zero means further processing is needed (for things like \x), or the escape
104 is invalid. */
105
106 #ifndef EBCDIC
107
108 /* This is the "normal" table for ASCII systems or for EBCDIC systems running
109 in UTF-8 mode. */
110
111 static const short int escapes[] = {
112 0, 0,
113 0, 0,
114 0, 0,
115 0, 0,
116 0, 0,
117 CHAR_COLON, CHAR_SEMICOLON,
118 CHAR_LESS_THAN_SIGN, CHAR_EQUALS_SIGN,
119 CHAR_GREATER_THAN_SIGN, CHAR_QUESTION_MARK,
120 CHAR_COMMERCIAL_AT, -ESC_A,
121 -ESC_B, -ESC_C,
122 -ESC_D, -ESC_E,
123 0, -ESC_G,
124 -ESC_H, 0,
125 0, -ESC_K,
126 0, 0,
127 -ESC_N, 0,
128 -ESC_P, -ESC_Q,
129 -ESC_R, -ESC_S,
130 0, 0,
131 -ESC_V, -ESC_W,
132 -ESC_X, 0,
133 -ESC_Z, CHAR_LEFT_SQUARE_BRACKET,
134 CHAR_BACKSLASH, CHAR_RIGHT_SQUARE_BRACKET,
135 CHAR_CIRCUMFLEX_ACCENT, CHAR_UNDERSCORE,
136 CHAR_GRAVE_ACCENT, 7,
137 -ESC_b, 0,
138 -ESC_d, ESC_e,
139 ESC_f, 0,
140 -ESC_h, 0,
141 0, -ESC_k,
142 0, 0,
143 ESC_n, 0,
144 -ESC_p, 0,
145 ESC_r, -ESC_s,
146 ESC_tee, 0,
147 -ESC_v, -ESC_w,
148 0, 0,
149 -ESC_z
150 };
151
152 #else
153
154 /* This is the "abnormal" table for EBCDIC systems without UTF-8 support. */
155
156 static const short int escapes[] = {
157 /* 48 */ 0, 0, 0, '.', '<', '(', '+', '|',
158 /* 50 */ '&', 0, 0, 0, 0, 0, 0, 0,
159 /* 58 */ 0, 0, '!', '$', '*', ')', ';', '~',
160 /* 60 */ '-', '/', 0, 0, 0, 0, 0, 0,
161 /* 68 */ 0, 0, '|', ',', '%', '_', '>', '?',
162 /* 70 */ 0, 0, 0, 0, 0, 0, 0, 0,
163 /* 78 */ 0, '`', ':', '#', '@', '\'', '=', '"',
164 /* 80 */ 0, 7, -ESC_b, 0, -ESC_d, ESC_e, ESC_f, 0,
165 /* 88 */-ESC_h, 0, 0, '{', 0, 0, 0, 0,
166 /* 90 */ 0, 0, -ESC_k, 'l', 0, ESC_n, 0, -ESC_p,
167 /* 98 */ 0, ESC_r, 0, '}', 0, 0, 0, 0,
168 /* A0 */ 0, '~', -ESC_s, ESC_tee, 0,-ESC_v, -ESC_w, 0,
169 /* A8 */ 0,-ESC_z, 0, 0, 0, '[', 0, 0,
170 /* B0 */ 0, 0, 0, 0, 0, 0, 0, 0,
171 /* B8 */ 0, 0, 0, 0, 0, ']', '=', '-',
172 /* C0 */ '{',-ESC_A, -ESC_B, -ESC_C, -ESC_D,-ESC_E, 0, -ESC_G,
173 /* C8 */-ESC_H, 0, 0, 0, 0, 0, 0, 0,
174 /* D0 */ '}', 0, -ESC_K, 0, 0,-ESC_N, 0, -ESC_P,
175 /* D8 */-ESC_Q,-ESC_R, 0, 0, 0, 0, 0, 0,
176 /* E0 */ '\\', 0, -ESC_S, 0, 0,-ESC_V, -ESC_W, -ESC_X,
177 /* E8 */ 0,-ESC_Z, 0, 0, 0, 0, 0, 0,
178 /* F0 */ 0, 0, 0, 0, 0, 0, 0, 0,
179 /* F8 */ 0, 0, 0, 0, 0, 0, 0, 0
180 };
181 #endif
182
183
184 /* Table of special "verbs" like (*PRUNE). This is a short table, so it is
185 searched linearly. Put all the names into a single string, in order to reduce
186 the number of relocations when a shared library is dynamically linked. The
187 string is built from string macros so that it works in UTF-8 mode on EBCDIC
188 platforms. */
189
190 typedef struct verbitem {
191 int len; /* Length of verb name */
192 int op; /* Op when no arg, or -1 if arg mandatory */
193 int op_arg; /* Op when arg present, or -1 if not allowed */
194 } verbitem;
195
196 static const char verbnames[] =
197 "\0" /* Empty name is a shorthand for MARK */
198 STRING_MARK0
199 STRING_ACCEPT0
200 STRING_COMMIT0
201 STRING_F0
202 STRING_FAIL0
203 STRING_PRUNE0
204 STRING_SKIP0
205 STRING_THEN;
206
207 static const verbitem verbs[] = {
208 { 0, -1, OP_MARK },
209 { 4, -1, OP_MARK },
210 { 6, OP_ACCEPT, -1 },
211 { 6, OP_COMMIT, -1 },
212 { 1, OP_FAIL, -1 },
213 { 4, OP_FAIL, -1 },
214 { 5, OP_PRUNE, OP_PRUNE_ARG },
215 { 4, OP_SKIP, OP_SKIP_ARG },
216 { 4, OP_THEN, OP_THEN_ARG }
217 };
218
219 static const int verbcount = sizeof(verbs)/sizeof(verbitem);
220
221
222 /* Tables of names of POSIX character classes and their lengths. The names are
223 now all in a single string, to reduce the number of relocations when a shared
224 library is dynamically loaded. The list of lengths is terminated by a zero
225 length entry. The first three must be alpha, lower, upper, as this is assumed
226 for handling case independence. */
227
228 static const char posix_names[] =
229 STRING_alpha0 STRING_lower0 STRING_upper0 STRING_alnum0
230 STRING_ascii0 STRING_blank0 STRING_cntrl0 STRING_digit0
231 STRING_graph0 STRING_print0 STRING_punct0 STRING_space0
232 STRING_word0 STRING_xdigit;
233
234 static const 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" /** DEAD **/
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 "\\c must be followed by an ASCII character\0"
412 "\\k is not followed by a braced, angle-bracketed, or quoted name\0"
413 ;
414
415 /* Table to identify digits and hex digits. This is used when compiling
416 patterns. Note that the tables in chartables are dependent on the locale, and
417 may mark arbitrary characters as digits - but the PCRE compiling code expects
418 to handle only 0-9, a-z, and A-Z as digits when compiling. That is why we have
419 a private table here. It costs 256 bytes, but it is a lot faster than doing
420 character value tests (at least in some simple cases I timed), and in some
421 applications one wants PCRE to compile efficiently as well as match
422 efficiently.
423
424 For convenience, we use the same bit definitions as in chartables:
425
426 0x04 decimal digit
427 0x08 hexadecimal digit
428
429 Then we can use ctype_digit and ctype_xdigit in the code. */
430
431 #ifndef EBCDIC
432
433 /* This is the "normal" case, for ASCII systems, and EBCDIC systems running in
434 UTF-8 mode. */
435
436 static const unsigned char digitab[] =
437 {
438 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 0- 7 */
439 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 8- 15 */
440 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 16- 23 */
441 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 24- 31 */
442 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - ' */
443 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* ( - / */
444 0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c, /* 0 - 7 */
445 0x0c,0x0c,0x00,0x00,0x00,0x00,0x00,0x00, /* 8 - ? */
446 0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* @ - G */
447 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* H - O */
448 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* P - W */
449 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* X - _ */
450 0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* ` - g */
451 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* h - o */
452 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* p - w */
453 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* x -127 */
454 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 128-135 */
455 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 136-143 */
456 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 144-151 */
457 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 152-159 */
458 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 160-167 */
459 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 168-175 */
460 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 176-183 */
461 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 184-191 */
462 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 192-199 */
463 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 200-207 */
464 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 208-215 */
465 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 216-223 */
466 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 224-231 */
467 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 232-239 */
468 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 240-247 */
469 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00};/* 248-255 */
470
471 #else
472
473 /* This is the "abnormal" case, for EBCDIC systems not running in UTF-8 mode. */
474
475 static const unsigned char digitab[] =
476 {
477 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 0- 7 0 */
478 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 8- 15 */
479 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 16- 23 10 */
480 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 24- 31 */
481 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 32- 39 20 */
482 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 40- 47 */
483 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 48- 55 30 */
484 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 56- 63 */
485 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - 71 40 */
486 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 72- | */
487 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* & - 87 50 */
488 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 88- 95 */
489 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - -103 60 */
490 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 104- ? */
491 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 112-119 70 */
492 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 120- " */
493 0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* 128- g 80 */
494 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* h -143 */
495 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 144- p 90 */
496 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* q -159 */
497 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 160- x A0 */
498 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* y -175 */
499 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* ^ -183 B0 */
500 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 184-191 */
501 0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* { - G C0 */
502 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* H -207 */
503 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* } - P D0 */
504 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* Q -223 */
505 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* \ - X E0 */
506 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* Y -239 */
507 0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c, /* 0 - 7 F0 */
508 0x0c,0x0c,0x00,0x00,0x00,0x00,0x00,0x00};/* 8 -255 */
509
510 static const unsigned char ebcdic_chartab[] = { /* chartable partial dup */
511 0x80,0x00,0x00,0x00,0x00,0x01,0x00,0x00, /* 0- 7 */
512 0x00,0x00,0x00,0x00,0x01,0x01,0x00,0x00, /* 8- 15 */
513 0x00,0x00,0x00,0x00,0x00,0x01,0x00,0x00, /* 16- 23 */
514 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 24- 31 */
515 0x00,0x00,0x00,0x00,0x00,0x01,0x00,0x00, /* 32- 39 */
516 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 40- 47 */
517 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 48- 55 */
518 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 56- 63 */
519 0x01,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - 71 */
520 0x00,0x00,0x00,0x80,0x00,0x80,0x80,0x80, /* 72- | */
521 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* & - 87 */
522 0x00,0x00,0x00,0x80,0x80,0x80,0x00,0x00, /* 88- 95 */
523 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - -103 */
524 0x00,0x00,0x00,0x00,0x00,0x10,0x00,0x80, /* 104- ? */
525 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 112-119 */
526 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 120- " */
527 0x00,0x1a,0x1a,0x1a,0x1a,0x1a,0x1a,0x12, /* 128- g */
528 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* h -143 */
529 0x00,0x12,0x12,0x12,0x12,0x12,0x12,0x12, /* 144- p */
530 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* q -159 */
531 0x00,0x00,0x12,0x12,0x12,0x12,0x12,0x12, /* 160- x */
532 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* y -175 */
533 0x80,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* ^ -183 */
534 0x00,0x00,0x80,0x00,0x00,0x00,0x00,0x00, /* 184-191 */
535 0x80,0x1a,0x1a,0x1a,0x1a,0x1a,0x1a,0x12, /* { - G */
536 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* H -207 */
537 0x00,0x12,0x12,0x12,0x12,0x12,0x12,0x12, /* } - P */
538 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* Q -223 */
539 0x00,0x00,0x12,0x12,0x12,0x12,0x12,0x12, /* \ - X */
540 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* Y -239 */
541 0x1c,0x1c,0x1c,0x1c,0x1c,0x1c,0x1c,0x1c, /* 0 - 7 */
542 0x1c,0x1c,0x00,0x00,0x00,0x00,0x00,0x00};/* 8 -255 */
543 #endif
544
545
546 /* Definition to allow mutual recursion */
547
548 static BOOL
549 compile_regex(int, uschar **, const uschar **, int *, BOOL, BOOL, int, int,
550 int *, int *, branch_chain *, compile_data *, int *);
551
552
553
554 /*************************************************
555 * Find an error text *
556 *************************************************/
557
558 /* The error texts are now all in one long string, to save on relocations. As
559 some of the text is of unknown length, we can't use a table of offsets.
560 Instead, just count through the strings. This is not a performance issue
561 because it happens only when there has been a compilation error.
562
563 Argument: the error number
564 Returns: pointer to the error string
565 */
566
567 static const char *
568 find_error_text(int n)
569 {
570 const char *s = error_texts;
571 for (; n > 0; n--)
572 {
573 while (*s++ != 0) {};
574 if (*s == 0) return "Error text not found (please report)";
575 }
576 return s;
577 }
578
579
580 /*************************************************
581 * Check for counted repeat *
582 *************************************************/
583
584 /* This function is called when a '{' is encountered in a place where it might
585 start a quantifier. It looks ahead to see if it really is a quantifier or not.
586 It is only a quantifier if it is one of the forms {ddd} {ddd,} or {ddd,ddd}
587 where the ddds are digits.
588
589 Arguments:
590 p pointer to the first char after '{'
591
592 Returns: TRUE or FALSE
593 */
594
595 static BOOL
596 is_counted_repeat(const uschar *p)
597 {
598 if ((digitab[*p++] & ctype_digit) == 0) return FALSE;
599 while ((digitab[*p] & ctype_digit) != 0) p++;
600 if (*p == CHAR_RIGHT_CURLY_BRACKET) return TRUE;
601
602 if (*p++ != CHAR_COMMA) return FALSE;
603 if (*p == CHAR_RIGHT_CURLY_BRACKET) return TRUE;
604
605 if ((digitab[*p++] & ctype_digit) == 0) return FALSE;
606 while ((digitab[*p] & ctype_digit) != 0) p++;
607
608 return (*p == CHAR_RIGHT_CURLY_BRACKET);
609 }
610
611
612
613 /*************************************************
614 * Handle escapes *
615 *************************************************/
616
617 /* This function is called when a \ has been encountered. It either returns a
618 positive value for a simple escape such as \n, or a negative value which
619 encodes one of the more complicated things such as \d. A backreference to group
620 n is returned as -(ESC_REF + n); ESC_REF is the highest ESC_xxx macro. When
621 UTF-8 is enabled, a positive value greater than 255 may be returned. On entry,
622 ptr is pointing at the \. On exit, it is on the final character of the escape
623 sequence.
624
625 Arguments:
626 ptrptr points to the pattern position pointer
627 errorcodeptr points to the errorcode variable
628 bracount number of previous extracting brackets
629 options the options bits
630 isclass TRUE if inside a character class
631
632 Returns: zero or positive => a data character
633 negative => a special escape sequence
634 on error, errorcodeptr is set
635 */
636
637 static int
638 check_escape(const uschar **ptrptr, int *errorcodeptr, int bracount,
639 int options, BOOL isclass)
640 {
641 BOOL utf8 = (options & PCRE_UTF8) != 0;
642 const uschar *ptr = *ptrptr + 1;
643 int c, i;
644
645 GETCHARINCTEST(c, ptr); /* Get character value, increment pointer */
646 ptr--; /* Set pointer back to the last byte */
647
648 /* If backslash is at the end of the pattern, it's an error. */
649
650 if (c == 0) *errorcodeptr = ERR1;
651
652 /* Non-alphanumerics are literals. For digits or letters, do an initial lookup
653 in a table. A non-zero result is something that can be returned immediately.
654 Otherwise further processing may be required. */
655
656 #ifndef EBCDIC /* ASCII/UTF-8 coding */
657 else if (c < CHAR_0 || c > CHAR_z) {} /* Not alphanumeric */
658 else if ((i = escapes[c - CHAR_0]) != 0) c = i;
659
660 #else /* EBCDIC coding */
661 else if (c < 'a' || (ebcdic_chartab[c] & 0x0E) == 0) {} /* Not alphanumeric */
662 else if ((i = escapes[c - 0x48]) != 0) c = i;
663 #endif
664
665 /* Escapes that need further processing, or are illegal. */
666
667 else
668 {
669 const uschar *oldptr;
670 BOOL braced, negated;
671
672 switch (c)
673 {
674 /* A number of Perl escapes are not handled by PCRE. We give an explicit
675 error. */
676
677 case CHAR_l:
678 case CHAR_L:
679 case CHAR_u:
680 case CHAR_U:
681 *errorcodeptr = ERR37;
682 break;
683
684 /* In a character class, \g is just a literal "g". Outside a character
685 class, \g must be followed by one of a number of specific things:
686
687 (1) A number, either plain or braced. If positive, it is an absolute
688 backreference. If negative, it is a relative backreference. This is a Perl
689 5.10 feature.
690
691 (2) Perl 5.10 also supports \g{name} as a reference to a named group. This
692 is part of Perl's movement towards a unified syntax for back references. As
693 this is synonymous with \k{name}, we fudge it up by pretending it really
694 was \k.
695
696 (3) For Oniguruma compatibility we also support \g followed by a name or a
697 number either in angle brackets or in single quotes. However, these are
698 (possibly recursive) subroutine calls, _not_ backreferences. Just return
699 the -ESC_g code (cf \k). */
700
701 case CHAR_g:
702 if (isclass) break;
703 if (ptr[1] == CHAR_LESS_THAN_SIGN || ptr[1] == CHAR_APOSTROPHE)
704 {
705 c = -ESC_g;
706 break;
707 }
708
709 /* Handle the Perl-compatible cases */
710
711 if (ptr[1] == CHAR_LEFT_CURLY_BRACKET)
712 {
713 const uschar *p;
714 for (p = ptr+2; *p != 0 && *p != CHAR_RIGHT_CURLY_BRACKET; p++)
715 if (*p != CHAR_MINUS && (digitab[*p] & ctype_digit) == 0) break;
716 if (*p != 0 && *p != CHAR_RIGHT_CURLY_BRACKET)
717 {
718 c = -ESC_k;
719 break;
720 }
721 braced = TRUE;
722 ptr++;
723 }
724 else braced = FALSE;
725
726 if (ptr[1] == CHAR_MINUS)
727 {
728 negated = TRUE;
729 ptr++;
730 }
731 else negated = FALSE;
732
733 c = 0;
734 while ((digitab[ptr[1]] & ctype_digit) != 0)
735 c = c * 10 + *(++ptr) - CHAR_0;
736
737 if (c < 0) /* Integer overflow */
738 {
739 *errorcodeptr = ERR61;
740 break;
741 }
742
743 if (braced && *(++ptr) != CHAR_RIGHT_CURLY_BRACKET)
744 {
745 *errorcodeptr = ERR57;
746 break;
747 }
748
749 if (c == 0)
750 {
751 *errorcodeptr = ERR58;
752 break;
753 }
754
755 if (negated)
756 {
757 if (c > bracount)
758 {
759 *errorcodeptr = ERR15;
760 break;
761 }
762 c = bracount - (c - 1);
763 }
764
765 c = -(ESC_REF + c);
766 break;
767
768 /* The handling of escape sequences consisting of a string of digits
769 starting with one that is not zero is not straightforward. By experiment,
770 the way Perl works seems to be as follows:
771
772 Outside a character class, the digits are read as a decimal number. If the
773 number is less than 10, or if there are that many previous extracting
774 left brackets, then it is a back reference. Otherwise, up to three octal
775 digits are read to form an escaped byte. Thus \123 is likely to be octal
776 123 (cf \0123, which is octal 012 followed by the literal 3). If the octal
777 value is greater than 377, the least significant 8 bits are taken. Inside a
778 character class, \ followed by a digit is always an octal number. */
779
780 case CHAR_1: case CHAR_2: case CHAR_3: case CHAR_4: case CHAR_5:
781 case CHAR_6: case CHAR_7: case CHAR_8: case CHAR_9:
782
783 if (!isclass)
784 {
785 oldptr = ptr;
786 c -= CHAR_0;
787 while ((digitab[ptr[1]] & ctype_digit) != 0)
788 c = c * 10 + *(++ptr) - CHAR_0;
789 if (c < 0) /* Integer overflow */
790 {
791 *errorcodeptr = ERR61;
792 break;
793 }
794 if (c < 10 || c <= bracount)
795 {
796 c = -(ESC_REF + c);
797 break;
798 }
799 ptr = oldptr; /* Put the pointer back and fall through */
800 }
801
802 /* Handle an octal number following \. If the first digit is 8 or 9, Perl
803 generates a binary zero byte and treats the digit as a following literal.
804 Thus we have to pull back the pointer by one. */
805
806 if ((c = *ptr) >= CHAR_8)
807 {
808 ptr--;
809 c = 0;
810 break;
811 }
812
813 /* \0 always starts an octal number, but we may drop through to here with a
814 larger first octal digit. The original code used just to take the least
815 significant 8 bits of octal numbers (I think this is what early Perls used
816 to do). Nowadays we allow for larger numbers in UTF-8 mode, but no more
817 than 3 octal digits. */
818
819 case CHAR_0:
820 c -= CHAR_0;
821 while(i++ < 2 && ptr[1] >= CHAR_0 && ptr[1] <= CHAR_7)
822 c = c * 8 + *(++ptr) - CHAR_0;
823 if (!utf8 && c > 255) *errorcodeptr = ERR51;
824 break;
825
826 /* \x is complicated. \x{ddd} is a character number which can be greater
827 than 0xff in utf8 mode, but only if the ddd are hex digits. If not, { is
828 treated as a data character. */
829
830 case CHAR_x:
831 if (ptr[1] == CHAR_LEFT_CURLY_BRACKET)
832 {
833 const uschar *pt = ptr + 2;
834 int count = 0;
835
836 c = 0;
837 while ((digitab[*pt] & ctype_xdigit) != 0)
838 {
839 register int cc = *pt++;
840 if (c == 0 && cc == CHAR_0) continue; /* Leading zeroes */
841 count++;
842
843 #ifndef EBCDIC /* ASCII/UTF-8 coding */
844 if (cc >= CHAR_a) cc -= 32; /* Convert to upper case */
845 c = (c << 4) + cc - ((cc < CHAR_A)? CHAR_0 : (CHAR_A - 10));
846 #else /* EBCDIC coding */
847 if (cc >= CHAR_a && cc <= CHAR_z) cc += 64; /* Convert to upper case */
848 c = (c << 4) + cc - ((cc >= CHAR_0)? CHAR_0 : (CHAR_A - 10));
849 #endif
850 }
851
852 if (*pt == CHAR_RIGHT_CURLY_BRACKET)
853 {
854 if (c < 0 || count > (utf8? 8 : 2)) *errorcodeptr = ERR34;
855 ptr = pt;
856 break;
857 }
858
859 /* If the sequence of hex digits does not end with '}', then we don't
860 recognize this construct; fall through to the normal \x handling. */
861 }
862
863 /* Read just a single-byte hex-defined char */
864
865 c = 0;
866 while (i++ < 2 && (digitab[ptr[1]] & ctype_xdigit) != 0)
867 {
868 int cc; /* Some compilers don't like */
869 cc = *(++ptr); /* ++ in initializers */
870 #ifndef EBCDIC /* ASCII/UTF-8 coding */
871 if (cc >= CHAR_a) cc -= 32; /* Convert to upper case */
872 c = c * 16 + cc - ((cc < CHAR_A)? CHAR_0 : (CHAR_A - 10));
873 #else /* EBCDIC coding */
874 if (cc <= CHAR_z) cc += 64; /* Convert to upper case */
875 c = c * 16 + cc - ((cc >= CHAR_0)? CHAR_0 : (CHAR_A - 10));
876 #endif
877 }
878 break;
879
880 /* For \c, a following letter is upper-cased; then the 0x40 bit is flipped.
881 An error is given if the byte following \c is not an ASCII character. This
882 coding is ASCII-specific, but then the whole concept of \cx is
883 ASCII-specific. (However, an EBCDIC equivalent has now been added.) */
884
885 case CHAR_c:
886 c = *(++ptr);
887 if (c == 0)
888 {
889 *errorcodeptr = ERR2;
890 break;
891 }
892 #ifndef EBCDIC /* ASCII/UTF-8 coding */
893 if (c > 127) /* Excludes all non-ASCII in either mode */
894 {
895 *errorcodeptr = ERR68;
896 break;
897 }
898 if (c >= CHAR_a && c <= CHAR_z) c -= 32;
899 c ^= 0x40;
900 #else /* EBCDIC coding */
901 if (c >= CHAR_a && c <= CHAR_z) c += 64;
902 c ^= 0xC0;
903 #endif
904 break;
905
906 /* PCRE_EXTRA enables extensions to Perl in the matter of escapes. Any
907 other alphanumeric following \ is an error if PCRE_EXTRA was set;
908 otherwise, for Perl compatibility, it is a literal. This code looks a bit
909 odd, but there used to be some cases other than the default, and there may
910 be again in future, so I haven't "optimized" it. */
911
912 default:
913 if ((options & PCRE_EXTRA) != 0) switch(c)
914 {
915 default:
916 *errorcodeptr = ERR3;
917 break;
918 }
919 break;
920 }
921 }
922
923 /* Perl supports \N{name} for character names, as well as plain \N for "not
924 newline". PCRE does not support \N{name}. However, it does support
925 quantification such as \N{2,3}. */
926
927 if (c == -ESC_N && ptr[1] == CHAR_LEFT_CURLY_BRACKET &&
928 !is_counted_repeat(ptr+2))
929 *errorcodeptr = ERR37;
930
931 /* If PCRE_UCP is set, we change the values for \d etc. */
932
933 if ((options & PCRE_UCP) != 0 && c <= -ESC_D && c >= -ESC_w)
934 c -= (ESC_DU - ESC_D);
935
936 /* Set the pointer to the final character before returning. */
937
938 *ptrptr = ptr;
939 return c;
940 }
941
942
943
944 #ifdef SUPPORT_UCP
945 /*************************************************
946 * Handle \P and \p *
947 *************************************************/
948
949 /* This function is called after \P or \p has been encountered, provided that
950 PCRE is compiled with support for Unicode properties. On entry, ptrptr is
951 pointing at the P or p. On exit, it is pointing at the final character of the
952 escape sequence.
953
954 Argument:
955 ptrptr points to the pattern position pointer
956 negptr points to a boolean that is set TRUE for negation else FALSE
957 dptr points to an int that is set to the detailed property value
958 errorcodeptr points to the error code variable
959
960 Returns: type value from ucp_type_table, or -1 for an invalid type
961 */
962
963 static int
964 get_ucp(const uschar **ptrptr, BOOL *negptr, int *dptr, int *errorcodeptr)
965 {
966 int c, i, bot, top;
967 const uschar *ptr = *ptrptr;
968 char name[32];
969
970 c = *(++ptr);
971 if (c == 0) goto ERROR_RETURN;
972
973 *negptr = FALSE;
974
975 /* \P or \p can be followed by a name in {}, optionally preceded by ^ for
976 negation. */
977
978 if (c == CHAR_LEFT_CURLY_BRACKET)
979 {
980 if (ptr[1] == CHAR_CIRCUMFLEX_ACCENT)
981 {
982 *negptr = TRUE;
983 ptr++;
984 }
985 for (i = 0; i < (int)sizeof(name) - 1; i++)
986 {
987 c = *(++ptr);
988 if (c == 0) goto ERROR_RETURN;
989 if (c == CHAR_RIGHT_CURLY_BRACKET) break;
990 name[i] = c;
991 }
992 if (c != CHAR_RIGHT_CURLY_BRACKET) goto ERROR_RETURN;
993 name[i] = 0;
994 }
995
996 /* Otherwise there is just one following character */
997
998 else
999 {
1000 name[0] = c;
1001 name[1] = 0;
1002 }
1003
1004 *ptrptr = ptr;
1005
1006 /* Search for a recognized property name using binary chop */
1007
1008 bot = 0;
1009 top = _pcre_utt_size;
1010
1011 while (bot < top)
1012 {
1013 i = (bot + top) >> 1;
1014 c = strcmp(name, _pcre_utt_names + _pcre_utt[i].name_offset);
1015 if (c == 0)
1016 {
1017 *dptr = _pcre_utt[i].value;
1018 return _pcre_utt[i].type;
1019 }
1020 if (c > 0) bot = i + 1; else top = i;
1021 }
1022
1023 *errorcodeptr = ERR47;
1024 *ptrptr = ptr;
1025 return -1;
1026
1027 ERROR_RETURN:
1028 *errorcodeptr = ERR46;
1029 *ptrptr = ptr;
1030 return -1;
1031 }
1032 #endif
1033
1034
1035
1036
1037 /*************************************************
1038 * Read repeat counts *
1039 *************************************************/
1040
1041 /* Read an item of the form {n,m} and return the values. This is called only
1042 after is_counted_repeat() has confirmed that a repeat-count quantifier exists,
1043 so the syntax is guaranteed to be correct, but we need to check the values.
1044
1045 Arguments:
1046 p pointer to first char after '{'
1047 minp pointer to int for min
1048 maxp pointer to int for max
1049 returned as -1 if no max
1050 errorcodeptr points to error code variable
1051
1052 Returns: pointer to '}' on success;
1053 current ptr on error, with errorcodeptr set non-zero
1054 */
1055
1056 static const uschar *
1057 read_repeat_counts(const uschar *p, int *minp, int *maxp, int *errorcodeptr)
1058 {
1059 int min = 0;
1060 int max = -1;
1061
1062 /* Read the minimum value and do a paranoid check: a negative value indicates
1063 an integer overflow. */
1064
1065 while ((digitab[*p] & ctype_digit) != 0) min = min * 10 + *p++ - CHAR_0;
1066 if (min < 0 || min > 65535)
1067 {
1068 *errorcodeptr = ERR5;
1069 return p;
1070 }
1071
1072 /* Read the maximum value if there is one, and again do a paranoid on its size.
1073 Also, max must not be less than min. */
1074
1075 if (*p == CHAR_RIGHT_CURLY_BRACKET) max = min; else
1076 {
1077 if (*(++p) != CHAR_RIGHT_CURLY_BRACKET)
1078 {
1079 max = 0;
1080 while((digitab[*p] & ctype_digit) != 0) max = max * 10 + *p++ - CHAR_0;
1081 if (max < 0 || max > 65535)
1082 {
1083 *errorcodeptr = ERR5;
1084 return p;
1085 }
1086 if (max < min)
1087 {
1088 *errorcodeptr = ERR4;
1089 return p;
1090 }
1091 }
1092 }
1093
1094 /* Fill in the required variables, and pass back the pointer to the terminating
1095 '}'. */
1096
1097 *minp = min;
1098 *maxp = max;
1099 return p;
1100 }
1101
1102
1103
1104 /*************************************************
1105 * Subroutine for finding forward reference *
1106 *************************************************/
1107
1108 /* This recursive function is called only from find_parens() below. The
1109 top-level call starts at the beginning of the pattern. All other calls must
1110 start at a parenthesis. It scans along a pattern's text looking for capturing
1111 subpatterns, and counting them. If it finds a named pattern that matches the
1112 name it is given, it returns its number. Alternatively, if the name is NULL, it
1113 returns when it reaches a given numbered subpattern. Recursion is used to keep
1114 track of subpatterns that reset the capturing group numbers - the (?| feature.
1115
1116 This function was originally called only from the second pass, in which we know
1117 that if (?< or (?' or (?P< is encountered, the name will be correctly
1118 terminated because that is checked in the first pass. There is now one call to
1119 this function in the first pass, to check for a recursive back reference by
1120 name (so that we can make the whole group atomic). In this case, we need check
1121 only up to the current position in the pattern, and that is still OK because
1122 and previous occurrences will have been checked. To make this work, the test
1123 for "end of pattern" is a check against cd->end_pattern in the main loop,
1124 instead of looking for a binary zero. This means that the special first-pass
1125 call can adjust cd->end_pattern temporarily. (Checks for binary zero while
1126 processing items within the loop are OK, because afterwards the main loop will
1127 terminate.)
1128
1129 Arguments:
1130 ptrptr address of the current character pointer (updated)
1131 cd compile background data
1132 name name to seek, or NULL if seeking a numbered subpattern
1133 lorn name length, or subpattern number if name is NULL
1134 xmode TRUE if we are in /x mode
1135 utf8 TRUE if we are in UTF-8 mode
1136 count pointer to the current capturing subpattern number (updated)
1137
1138 Returns: the number of the named subpattern, or -1 if not found
1139 */
1140
1141 static int
1142 find_parens_sub(uschar **ptrptr, compile_data *cd, const uschar *name, int lorn,
1143 BOOL xmode, BOOL utf8, int *count)
1144 {
1145 uschar *ptr = *ptrptr;
1146 int start_count = *count;
1147 int hwm_count = start_count;
1148 BOOL dup_parens = FALSE;
1149
1150 /* If the first character is a parenthesis, check on the type of group we are
1151 dealing with. The very first call may not start with a parenthesis. */
1152
1153 if (ptr[0] == CHAR_LEFT_PARENTHESIS)
1154 {
1155 /* Handle specials such as (*SKIP) or (*UTF8) etc. */
1156
1157 if (ptr[1] == CHAR_ASTERISK) ptr += 2;
1158
1159 /* Handle a normal, unnamed capturing parenthesis. */
1160
1161 else if (ptr[1] != CHAR_QUESTION_MARK)
1162 {
1163 *count += 1;
1164 if (name == NULL && *count == lorn) return *count;
1165 ptr++;
1166 }
1167
1168 /* All cases now have (? at the start. Remember when we are in a group
1169 where the parenthesis numbers are duplicated. */
1170
1171 else if (ptr[2] == CHAR_VERTICAL_LINE)
1172 {
1173 ptr += 3;
1174 dup_parens = TRUE;
1175 }
1176
1177 /* Handle comments; all characters are allowed until a ket is reached. */
1178
1179 else if (ptr[2] == CHAR_NUMBER_SIGN)
1180 {
1181 for (ptr += 3; *ptr != 0; ptr++) if (*ptr == CHAR_RIGHT_PARENTHESIS) break;
1182 goto FAIL_EXIT;
1183 }
1184
1185 /* Handle a condition. If it is an assertion, just carry on so that it
1186 is processed as normal. If not, skip to the closing parenthesis of the
1187 condition (there can't be any nested parens). */
1188
1189 else if (ptr[2] == CHAR_LEFT_PARENTHESIS)
1190 {
1191 ptr += 2;
1192 if (ptr[1] != CHAR_QUESTION_MARK)
1193 {
1194 while (*ptr != 0 && *ptr != CHAR_RIGHT_PARENTHESIS) ptr++;
1195 if (*ptr != 0) ptr++;
1196 }
1197 }
1198
1199 /* Start with (? but not a condition. */
1200
1201 else
1202 {
1203 ptr += 2;
1204 if (*ptr == CHAR_P) ptr++; /* Allow optional P */
1205
1206 /* We have to disambiguate (?<! and (?<= from (?<name> for named groups */
1207
1208 if ((*ptr == CHAR_LESS_THAN_SIGN && ptr[1] != CHAR_EXCLAMATION_MARK &&
1209 ptr[1] != CHAR_EQUALS_SIGN) || *ptr == CHAR_APOSTROPHE)
1210 {
1211 int term;
1212 const uschar *thisname;
1213 *count += 1;
1214 if (name == NULL && *count == lorn) return *count;
1215 term = *ptr++;
1216 if (term == CHAR_LESS_THAN_SIGN) term = CHAR_GREATER_THAN_SIGN;
1217 thisname = ptr;
1218 while (*ptr != term) ptr++;
1219 if (name != NULL && lorn == ptr - thisname &&
1220 strncmp((const char *)name, (const char *)thisname, lorn) == 0)
1221 return *count;
1222 term++;
1223 }
1224 }
1225 }
1226
1227 /* Past any initial parenthesis handling, scan for parentheses or vertical
1228 bars. Stop if we get to cd->end_pattern. Note that this is important for the
1229 first-pass call when this value is temporarily adjusted to stop at the current
1230 position. So DO NOT change this to a test for binary zero. */
1231
1232 for (; ptr < cd->end_pattern; ptr++)
1233 {
1234 /* Skip over backslashed characters and also entire \Q...\E */
1235
1236 if (*ptr == CHAR_BACKSLASH)
1237 {
1238 if (*(++ptr) == 0) goto FAIL_EXIT;
1239 if (*ptr == CHAR_Q) for (;;)
1240 {
1241 while (*(++ptr) != 0 && *ptr != CHAR_BACKSLASH) {};
1242 if (*ptr == 0) goto FAIL_EXIT;
1243 if (*(++ptr) == CHAR_E) break;
1244 }
1245 continue;
1246 }
1247
1248 /* Skip over character classes; this logic must be similar to the way they
1249 are handled for real. If the first character is '^', skip it. Also, if the
1250 first few characters (either before or after ^) are \Q\E or \E we skip them
1251 too. This makes for compatibility with Perl. Note the use of STR macros to
1252 encode "Q\\E" so that it works in UTF-8 on EBCDIC platforms. */
1253
1254 if (*ptr == CHAR_LEFT_SQUARE_BRACKET)
1255 {
1256 BOOL negate_class = FALSE;
1257 for (;;)
1258 {
1259 if (ptr[1] == CHAR_BACKSLASH)
1260 {
1261 if (ptr[2] == CHAR_E)
1262 ptr+= 2;
1263 else if (strncmp((const char *)ptr+2,
1264 STR_Q STR_BACKSLASH STR_E, 3) == 0)
1265 ptr += 4;
1266 else
1267 break;
1268 }
1269 else if (!negate_class && ptr[1] == CHAR_CIRCUMFLEX_ACCENT)
1270 {
1271 negate_class = TRUE;
1272 ptr++;
1273 }
1274 else break;
1275 }
1276
1277 /* If the next character is ']', it is a data character that must be
1278 skipped, except in JavaScript compatibility mode. */
1279
1280 if (ptr[1] == CHAR_RIGHT_SQUARE_BRACKET &&
1281 (cd->external_options & PCRE_JAVASCRIPT_COMPAT) == 0)
1282 ptr++;
1283
1284 while (*(++ptr) != CHAR_RIGHT_SQUARE_BRACKET)
1285 {
1286 if (*ptr == 0) return -1;
1287 if (*ptr == CHAR_BACKSLASH)
1288 {
1289 if (*(++ptr) == 0) goto FAIL_EXIT;
1290 if (*ptr == CHAR_Q) for (;;)
1291 {
1292 while (*(++ptr) != 0 && *ptr != CHAR_BACKSLASH) {};
1293 if (*ptr == 0) goto FAIL_EXIT;
1294 if (*(++ptr) == CHAR_E) break;
1295 }
1296 continue;
1297 }
1298 }
1299 continue;
1300 }
1301
1302 /* Skip comments in /x mode */
1303
1304 if (xmode && *ptr == CHAR_NUMBER_SIGN)
1305 {
1306 ptr++;
1307 while (*ptr != 0)
1308 {
1309 if (IS_NEWLINE(ptr)) { ptr += cd->nllen - 1; break; }
1310 ptr++;
1311 #ifdef SUPPORT_UTF8
1312 if (utf8) while ((*ptr & 0xc0) == 0x80) ptr++;
1313 #endif
1314 }
1315 if (*ptr == 0) goto FAIL_EXIT;
1316 continue;
1317 }
1318
1319 /* Check for the special metacharacters */
1320
1321 if (*ptr == CHAR_LEFT_PARENTHESIS)
1322 {
1323 int rc = find_parens_sub(&ptr, cd, name, lorn, xmode, utf8, count);
1324 if (rc > 0) return rc;
1325 if (*ptr == 0) goto FAIL_EXIT;
1326 }
1327
1328 else if (*ptr == CHAR_RIGHT_PARENTHESIS)
1329 {
1330 if (dup_parens && *count < hwm_count) *count = hwm_count;
1331 goto FAIL_EXIT;
1332 }
1333
1334 else if (*ptr == CHAR_VERTICAL_LINE && dup_parens)
1335 {
1336 if (*count > hwm_count) hwm_count = *count;
1337 *count = start_count;
1338 }
1339 }
1340
1341 FAIL_EXIT:
1342 *ptrptr = ptr;
1343 return -1;
1344 }
1345
1346
1347
1348
1349 /*************************************************
1350 * Find forward referenced subpattern *
1351 *************************************************/
1352
1353 /* This function scans along a pattern's text looking for capturing
1354 subpatterns, and counting them. If it finds a named pattern that matches the
1355 name it is given, it returns its number. Alternatively, if the name is NULL, it
1356 returns when it reaches a given numbered subpattern. This is used for forward
1357 references to subpatterns. We used to be able to start this scan from the
1358 current compiling point, using the current count value from cd->bracount, and
1359 do it all in a single loop, but the addition of the possibility of duplicate
1360 subpattern numbers means that we have to scan from the very start, in order to
1361 take account of such duplicates, and to use a recursive function to keep track
1362 of the different types of group.
1363
1364 Arguments:
1365 cd compile background data
1366 name name to seek, or NULL if seeking a numbered subpattern
1367 lorn name length, or subpattern number if name is NULL
1368 xmode TRUE if we are in /x mode
1369 utf8 TRUE if we are in UTF-8 mode
1370
1371 Returns: the number of the found subpattern, or -1 if not found
1372 */
1373
1374 static int
1375 find_parens(compile_data *cd, const uschar *name, int lorn, BOOL xmode,
1376 BOOL utf8)
1377 {
1378 uschar *ptr = (uschar *)cd->start_pattern;
1379 int count = 0;
1380 int rc;
1381
1382 /* If the pattern does not start with an opening parenthesis, the first call
1383 to find_parens_sub() will scan right to the end (if necessary). However, if it
1384 does start with a parenthesis, find_parens_sub() will return when it hits the
1385 matching closing parens. That is why we have to have a loop. */
1386
1387 for (;;)
1388 {
1389 rc = find_parens_sub(&ptr, cd, name, lorn, xmode, utf8, &count);
1390 if (rc > 0 || *ptr++ == 0) break;
1391 }
1392
1393 return rc;
1394 }
1395
1396
1397
1398
1399 /*************************************************
1400 * Find first significant op code *
1401 *************************************************/
1402
1403 /* This is called by several functions that scan a compiled expression looking
1404 for a fixed first character, or an anchoring op code etc. It skips over things
1405 that do not influence this. For some calls, it makes sense to skip negative
1406 forward and all backward assertions, and also the \b assertion; for others it
1407 does not.
1408
1409 Arguments:
1410 code pointer to the start of the group
1411 skipassert TRUE if certain assertions are to be skipped
1412
1413 Returns: pointer to the first significant opcode
1414 */
1415
1416 static const uschar*
1417 first_significant_code(const uschar *code, BOOL skipassert)
1418 {
1419 for (;;)
1420 {
1421 switch ((int)*code)
1422 {
1423 case OP_ASSERT_NOT:
1424 case OP_ASSERTBACK:
1425 case OP_ASSERTBACK_NOT:
1426 if (!skipassert) return code;
1427 do code += GET(code, 1); while (*code == OP_ALT);
1428 code += _pcre_OP_lengths[*code];
1429 break;
1430
1431 case OP_WORD_BOUNDARY:
1432 case OP_NOT_WORD_BOUNDARY:
1433 if (!skipassert) return code;
1434 /* Fall through */
1435
1436 case OP_CALLOUT:
1437 case OP_CREF:
1438 case OP_NCREF:
1439 case OP_RREF:
1440 case OP_NRREF:
1441 case OP_DEF:
1442 code += _pcre_OP_lengths[*code];
1443 break;
1444
1445 default:
1446 return code;
1447 }
1448 }
1449 /* Control never reaches here */
1450 }
1451
1452
1453
1454
1455 /*************************************************
1456 * Find the fixed length of a branch *
1457 *************************************************/
1458
1459 /* Scan a branch and compute the fixed length of subject that will match it,
1460 if the length is fixed. This is needed for dealing with backward assertions.
1461 In UTF8 mode, the result is in characters rather than bytes. The branch is
1462 temporarily terminated with OP_END when this function is called.
1463
1464 This function is called when a backward assertion is encountered, so that if it
1465 fails, the error message can point to the correct place in the pattern.
1466 However, we cannot do this when the assertion contains subroutine calls,
1467 because they can be forward references. We solve this by remembering this case
1468 and doing the check at the end; a flag specifies which mode we are running in.
1469
1470 Arguments:
1471 code points to the start of the pattern (the bracket)
1472 utf8 TRUE in UTF-8 mode
1473 atend TRUE if called when the pattern is complete
1474 cd the "compile data" structure
1475
1476 Returns: the fixed length,
1477 or -1 if there is no fixed length,
1478 or -2 if \C was encountered
1479 or -3 if an OP_RECURSE item was encountered and atend is FALSE
1480 */
1481
1482 static int
1483 find_fixedlength(uschar *code, BOOL utf8, BOOL atend, compile_data *cd)
1484 {
1485 int length = -1;
1486
1487 register int branchlength = 0;
1488 register uschar *cc = code + 1 + LINK_SIZE;
1489
1490 /* Scan along the opcodes for this branch. If we get to the end of the
1491 branch, check the length against that of the other branches. */
1492
1493 for (;;)
1494 {
1495 int d;
1496 uschar *ce, *cs;
1497 register int op = *cc;
1498 switch (op)
1499 {
1500 /* We only need to continue for OP_CBRA (normal capturing bracket) and
1501 OP_BRA (normal non-capturing bracket) because the other variants of these
1502 opcodes are all concerned with unlimited repeated groups, which of course
1503 are not of fixed length. They will cause a -1 response from the default
1504 case of this switch. */
1505
1506 case OP_CBRA:
1507 case OP_BRA:
1508 case OP_ONCE:
1509 case OP_COND:
1510 d = find_fixedlength(cc + ((op == OP_CBRA)? 2:0), utf8, atend, cd);
1511 if (d < 0) return d;
1512 branchlength += d;
1513 do cc += GET(cc, 1); while (*cc == OP_ALT);
1514 cc += 1 + LINK_SIZE;
1515 break;
1516
1517 /* Reached end of a branch; if it's a ket it is the end of a nested
1518 call. If it's ALT it is an alternation in a nested call. If it is
1519 END it's the end of the outer call. All can be handled by the same code.
1520 Note that we must not include the OP_KETRxxx opcodes here, because they
1521 all imply an unlimited repeat. */
1522
1523 case OP_ALT:
1524 case OP_KET:
1525 case OP_END:
1526 if (length < 0) length = branchlength;
1527 else if (length != branchlength) return -1;
1528 if (*cc != OP_ALT) return length;
1529 cc += 1 + LINK_SIZE;
1530 branchlength = 0;
1531 break;
1532
1533 /* A true recursion implies not fixed length, but a subroutine call may
1534 be OK. If the subroutine is a forward reference, we can't deal with
1535 it until the end of the pattern, so return -3. */
1536
1537 case OP_RECURSE:
1538 if (!atend) return -3;
1539 cs = ce = (uschar *)cd->start_code + GET(cc, 1); /* Start subpattern */
1540 do ce += GET(ce, 1); while (*ce == OP_ALT); /* End subpattern */
1541 if (cc > cs && cc < ce) return -1; /* Recursion */
1542 d = find_fixedlength(cs + 2, utf8, atend, cd);
1543 if (d < 0) return d;
1544 branchlength += d;
1545 cc += 1 + LINK_SIZE;
1546 break;
1547
1548 /* Skip over assertive subpatterns */
1549
1550 case OP_ASSERT:
1551 case OP_ASSERT_NOT:
1552 case OP_ASSERTBACK:
1553 case OP_ASSERTBACK_NOT:
1554 do cc += GET(cc, 1); while (*cc == OP_ALT);
1555 /* Fall through */
1556
1557 /* Skip over things that don't match chars */
1558
1559 case OP_REVERSE:
1560 case OP_CREF:
1561 case OP_NCREF:
1562 case OP_RREF:
1563 case OP_NRREF:
1564 case OP_DEF:
1565 case OP_CALLOUT:
1566 case OP_SOD:
1567 case OP_SOM:
1568 case OP_SET_SOM:
1569 case OP_EOD:
1570 case OP_EODN:
1571 case OP_CIRC:
1572 case OP_CIRCM:
1573 case OP_DOLL:
1574 case OP_DOLLM:
1575 case OP_NOT_WORD_BOUNDARY:
1576 case OP_WORD_BOUNDARY:
1577 cc += _pcre_OP_lengths[*cc];
1578 break;
1579
1580 /* Handle literal characters */
1581
1582 case OP_CHAR:
1583 case OP_CHARI:
1584 case OP_NOT:
1585 case OP_NOTI:
1586 branchlength++;
1587 cc += 2;
1588 #ifdef SUPPORT_UTF8
1589 if (utf8 && cc[-1] >= 0xc0) cc += _pcre_utf8_table4[cc[-1] & 0x3f];
1590 #endif
1591 break;
1592
1593 /* Handle exact repetitions. The count is already in characters, but we
1594 need to skip over a multibyte character in UTF8 mode. */
1595
1596 case OP_EXACT:
1597 branchlength += GET2(cc,1);
1598 cc += 4;
1599 #ifdef SUPPORT_UTF8
1600 if (utf8 && cc[-1] >= 0xc0) cc += _pcre_utf8_table4[cc[-1] & 0x3f];
1601 #endif
1602 break;
1603
1604 case OP_TYPEEXACT:
1605 branchlength += GET2(cc,1);
1606 if (cc[3] == OP_PROP || cc[3] == OP_NOTPROP) cc += 2;
1607 cc += 4;
1608 break;
1609
1610 /* Handle single-char matchers */
1611
1612 case OP_PROP:
1613 case OP_NOTPROP:
1614 cc += 2;
1615 /* Fall through */
1616
1617 case OP_NOT_DIGIT:
1618 case OP_DIGIT:
1619 case OP_NOT_WHITESPACE:
1620 case OP_WHITESPACE:
1621 case OP_NOT_WORDCHAR:
1622 case OP_WORDCHAR:
1623 case OP_ANY:
1624 case OP_ALLANY:
1625 branchlength++;
1626 cc++;
1627 break;
1628
1629 /* The single-byte matcher isn't allowed */
1630
1631 case OP_ANYBYTE:
1632 return -2;
1633
1634 /* Check a class for variable quantification */
1635
1636 #ifdef SUPPORT_UTF8
1637 case OP_XCLASS:
1638 cc += GET(cc, 1) - 33;
1639 /* Fall through */
1640 #endif
1641
1642 case OP_CLASS:
1643 case OP_NCLASS:
1644 cc += 33;
1645
1646 switch (*cc)
1647 {
1648 case OP_CRSTAR:
1649 case OP_CRMINSTAR:
1650 case OP_CRQUERY:
1651 case OP_CRMINQUERY:
1652 return -1;
1653
1654 case OP_CRRANGE:
1655 case OP_CRMINRANGE:
1656 if (GET2(cc,1) != GET2(cc,3)) return -1;
1657 branchlength += GET2(cc,1);
1658 cc += 5;
1659 break;
1660
1661 default:
1662 branchlength++;
1663 }
1664 break;
1665
1666 /* Anything else is variable length */
1667
1668 default:
1669 return -1;
1670 }
1671 }
1672 /* Control never gets here */
1673 }
1674
1675
1676
1677
1678 /*************************************************
1679 * Scan compiled regex for specific bracket *
1680 *************************************************/
1681
1682 /* This little function scans through a compiled pattern until it finds a
1683 capturing bracket with the given number, or, if the number is negative, an
1684 instance of OP_REVERSE for a lookbehind. The function is global in the C sense
1685 so that it can be called from pcre_study() when finding the minimum matching
1686 length.
1687
1688 Arguments:
1689 code points to start of expression
1690 utf8 TRUE in UTF-8 mode
1691 number the required bracket number or negative to find a lookbehind
1692
1693 Returns: pointer to the opcode for the bracket, or NULL if not found
1694 */
1695
1696 const uschar *
1697 _pcre_find_bracket(const uschar *code, BOOL utf8, int number)
1698 {
1699 for (;;)
1700 {
1701 register int c = *code;
1702
1703 if (c == OP_END) return NULL;
1704
1705 /* XCLASS is used for classes that cannot be represented just by a bit
1706 map. This includes negated single high-valued characters. The length in
1707 the table is zero; the actual length is stored in the compiled code. */
1708
1709 if (c == OP_XCLASS) code += GET(code, 1);
1710
1711 /* Handle recursion */
1712
1713 else if (c == OP_REVERSE)
1714 {
1715 if (number < 0) return (uschar *)code;
1716 code += _pcre_OP_lengths[c];
1717 }
1718
1719 /* Handle capturing bracket */
1720
1721 else if (c == OP_CBRA || c == OP_SCBRA ||
1722 c == OP_CBRAPOS || c == OP_SCBRAPOS)
1723 {
1724 int n = GET2(code, 1+LINK_SIZE);
1725 if (n == number) return (uschar *)code;
1726 code += _pcre_OP_lengths[c];
1727 }
1728
1729 /* Otherwise, we can get the item's length from the table, except that for
1730 repeated character types, we have to test for \p and \P, which have an extra
1731 two bytes of parameters, and for MARK/PRUNE/SKIP/THEN with an argument, we
1732 must add in its length. */
1733
1734 else
1735 {
1736 switch(c)
1737 {
1738 case OP_TYPESTAR:
1739 case OP_TYPEMINSTAR:
1740 case OP_TYPEPLUS:
1741 case OP_TYPEMINPLUS:
1742 case OP_TYPEQUERY:
1743 case OP_TYPEMINQUERY:
1744 case OP_TYPEPOSSTAR:
1745 case OP_TYPEPOSPLUS:
1746 case OP_TYPEPOSQUERY:
1747 if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
1748 break;
1749
1750 case OP_TYPEUPTO:
1751 case OP_TYPEMINUPTO:
1752 case OP_TYPEEXACT:
1753 case OP_TYPEPOSUPTO:
1754 if (code[3] == OP_PROP || code[3] == OP_NOTPROP) code += 2;
1755 break;
1756
1757 case OP_MARK:
1758 case OP_PRUNE_ARG:
1759 case OP_SKIP_ARG:
1760 code += code[1];
1761 break;
1762
1763 case OP_THEN_ARG:
1764 code += code[1+LINK_SIZE];
1765 break;
1766 }
1767
1768 /* Add in the fixed length from the table */
1769
1770 code += _pcre_OP_lengths[c];
1771
1772 /* In UTF-8 mode, opcodes that are followed by a character may be followed by
1773 a multi-byte character. The length in the table is a minimum, so we have to
1774 arrange to skip the extra bytes. */
1775
1776 #ifdef SUPPORT_UTF8
1777 if (utf8) switch(c)
1778 {
1779 case OP_CHAR:
1780 case OP_CHARI:
1781 case OP_EXACT:
1782 case OP_EXACTI:
1783 case OP_UPTO:
1784 case OP_UPTOI:
1785 case OP_MINUPTO:
1786 case OP_MINUPTOI:
1787 case OP_POSUPTO:
1788 case OP_POSUPTOI:
1789 case OP_STAR:
1790 case OP_STARI:
1791 case OP_MINSTAR:
1792 case OP_MINSTARI:
1793 case OP_POSSTAR:
1794 case OP_POSSTARI:
1795 case OP_PLUS:
1796 case OP_PLUSI:
1797 case OP_MINPLUS:
1798 case OP_MINPLUSI:
1799 case OP_POSPLUS:
1800 case OP_POSPLUSI:
1801 case OP_QUERY:
1802 case OP_QUERYI:
1803 case OP_MINQUERY:
1804 case OP_MINQUERYI:
1805 case OP_POSQUERY:
1806 case OP_POSQUERYI:
1807 if (code[-1] >= 0xc0) code += _pcre_utf8_table4[code[-1] & 0x3f];
1808 break;
1809 }
1810 #else
1811 (void)(utf8); /* Keep compiler happy by referencing function argument */
1812 #endif
1813 }
1814 }
1815 }
1816
1817
1818
1819 /*************************************************
1820 * Scan compiled regex for recursion reference *
1821 *************************************************/
1822
1823 /* This little function scans through a compiled pattern until it finds an
1824 instance of OP_RECURSE.
1825
1826 Arguments:
1827 code points to start of expression
1828 utf8 TRUE in UTF-8 mode
1829
1830 Returns: pointer to the opcode for OP_RECURSE, or NULL if not found
1831 */
1832
1833 static const uschar *
1834 find_recurse(const uschar *code, BOOL utf8)
1835 {
1836 for (;;)
1837 {
1838 register int c = *code;
1839 if (c == OP_END) return NULL;
1840 if (c == OP_RECURSE) return code;
1841
1842 /* XCLASS is used for classes that cannot be represented just by a bit
1843 map. This includes negated single high-valued characters. The length in
1844 the table is zero; the actual length is stored in the compiled code. */
1845
1846 if (c == OP_XCLASS) code += GET(code, 1);
1847
1848 /* Otherwise, we can get the item's length from the table, except that for
1849 repeated character types, we have to test for \p and \P, which have an extra
1850 two bytes of parameters, and for MARK/PRUNE/SKIP/THEN with an argument, we
1851 must add in its length. */
1852
1853 else
1854 {
1855 switch(c)
1856 {
1857 case OP_TYPESTAR:
1858 case OP_TYPEMINSTAR:
1859 case OP_TYPEPLUS:
1860 case OP_TYPEMINPLUS:
1861 case OP_TYPEQUERY:
1862 case OP_TYPEMINQUERY:
1863 case OP_TYPEPOSSTAR:
1864 case OP_TYPEPOSPLUS:
1865 case OP_TYPEPOSQUERY:
1866 if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
1867 break;
1868
1869 case OP_TYPEPOSUPTO:
1870 case OP_TYPEUPTO:
1871 case OP_TYPEMINUPTO:
1872 case OP_TYPEEXACT:
1873 if (code[3] == OP_PROP || code[3] == OP_NOTPROP) code += 2;
1874 break;
1875
1876 case OP_MARK:
1877 case OP_PRUNE_ARG:
1878 case OP_SKIP_ARG:
1879 code += code[1];
1880 break;
1881
1882 case OP_THEN_ARG:
1883 code += code[1+LINK_SIZE];
1884 break;
1885 }
1886
1887 /* Add in the fixed length from the table */
1888
1889 code += _pcre_OP_lengths[c];
1890
1891 /* In UTF-8 mode, opcodes that are followed by a character may be followed
1892 by a multi-byte character. The length in the table is a minimum, so we have
1893 to arrange to skip the extra bytes. */
1894
1895 #ifdef SUPPORT_UTF8
1896 if (utf8) switch(c)
1897 {
1898 case OP_CHAR:
1899 case OP_CHARI:
1900 case OP_EXACT:
1901 case OP_EXACTI:
1902 case OP_UPTO:
1903 case OP_UPTOI:
1904 case OP_MINUPTO:
1905 case OP_MINUPTOI:
1906 case OP_POSUPTO:
1907 case OP_POSUPTOI:
1908 case OP_STAR:
1909 case OP_STARI:
1910 case OP_MINSTAR:
1911 case OP_MINSTARI:
1912 case OP_POSSTAR:
1913 case OP_POSSTARI:
1914 case OP_PLUS:
1915 case OP_PLUSI:
1916 case OP_MINPLUS:
1917 case OP_MINPLUSI:
1918 case OP_POSPLUS:
1919 case OP_POSPLUSI:
1920 case OP_QUERY:
1921 case OP_QUERYI:
1922 case OP_MINQUERY:
1923 case OP_MINQUERYI:
1924 case OP_POSQUERY:
1925 case OP_POSQUERYI:
1926 if (code[-1] >= 0xc0) code += _pcre_utf8_table4[code[-1] & 0x3f];
1927 break;
1928 }
1929 #else
1930 (void)(utf8); /* Keep compiler happy by referencing function argument */
1931 #endif
1932 }
1933 }
1934 }
1935
1936
1937
1938 /*************************************************
1939 * Scan compiled branch for non-emptiness *
1940 *************************************************/
1941
1942 /* This function scans through a branch of a compiled pattern to see whether it
1943 can match the empty string or not. It is called from could_be_empty()
1944 below and from compile_branch() when checking for an unlimited repeat of a
1945 group that can match nothing. Note that first_significant_code() skips over
1946 backward and negative forward assertions when its final argument is TRUE. If we
1947 hit an unclosed bracket, we return "empty" - this means we've struck an inner
1948 bracket whose current branch will already have been scanned.
1949
1950 Arguments:
1951 code points to start of search
1952 endcode points to where to stop
1953 utf8 TRUE if in UTF8 mode
1954 cd contains pointers to tables etc.
1955
1956 Returns: TRUE if what is matched could be empty
1957 */
1958
1959 static BOOL
1960 could_be_empty_branch(const uschar *code, const uschar *endcode, BOOL utf8,
1961 compile_data *cd)
1962 {
1963 register int c;
1964 for (code = first_significant_code(code + _pcre_OP_lengths[*code], TRUE);
1965 code < endcode;
1966 code = first_significant_code(code + _pcre_OP_lengths[c], TRUE))
1967 {
1968 const uschar *ccode;
1969
1970 c = *code;
1971
1972 /* Skip over forward assertions; the other assertions are skipped by
1973 first_significant_code() with a TRUE final argument. */
1974
1975 if (c == OP_ASSERT)
1976 {
1977 do code += GET(code, 1); while (*code == OP_ALT);
1978 c = *code;
1979 continue;
1980 }
1981
1982 /* For a recursion/subroutine call, if its end has been reached, which
1983 implies a backward reference subroutine call, we can scan it. If it's a
1984 forward reference subroutine call, we can't. To detect forward reference
1985 we have to scan up the list that is kept in the workspace. This function is
1986 called only when doing the real compile, not during the pre-compile that
1987 measures the size of the compiled pattern. */
1988
1989 if (c == OP_RECURSE)
1990 {
1991 const uschar *scode;
1992 BOOL empty_branch;
1993
1994 /* Test for forward reference */
1995
1996 for (scode = cd->start_workspace; scode < cd->hwm; scode += LINK_SIZE)
1997 if (GET(scode, 0) == code + 1 - cd->start_code) return TRUE;
1998
1999 /* Not a forward reference, test for completed backward reference */
2000
2001 empty_branch = FALSE;
2002 scode = cd->start_code + GET(code, 1);
2003 if (GET(scode, 1) == 0) return TRUE; /* Unclosed */
2004
2005 /* Completed backwards reference */
2006
2007 do
2008 {
2009 if (could_be_empty_branch(scode, endcode, utf8, cd))
2010 {
2011 empty_branch = TRUE;
2012 break;
2013 }
2014 scode += GET(scode, 1);
2015 }
2016 while (*scode == OP_ALT);
2017
2018 if (!empty_branch) return FALSE; /* All branches are non-empty */
2019 continue;
2020 }
2021
2022 /* Groups with zero repeats can of course be empty; skip them. */
2023
2024 if (c == OP_BRAZERO || c == OP_BRAMINZERO || c == OP_SKIPZERO ||
2025 c == OP_BRAPOSZERO)
2026 {
2027 code += _pcre_OP_lengths[c];
2028 do code += GET(code, 1); while (*code == OP_ALT);
2029 c = *code;
2030 continue;
2031 }
2032
2033 /* A nested group that is already marked as "could be empty" can just be
2034 skipped. */
2035
2036 if (c == OP_SBRA || c == OP_SBRAPOS ||
2037 c == OP_SCBRA || c == OP_SCBRAPOS)
2038 {
2039 do code += GET(code, 1); while (*code == OP_ALT);
2040 c = *code;
2041 continue;
2042 }
2043
2044 /* For other groups, scan the branches. */
2045
2046 if (c == OP_BRA || c == OP_BRAPOS ||
2047 c == OP_CBRA || c == OP_CBRAPOS ||
2048 c == OP_ONCE || c == OP_COND)
2049 {
2050 BOOL empty_branch;
2051 if (GET(code, 1) == 0) return TRUE; /* Hit unclosed bracket */
2052
2053 /* If a conditional group has only one branch, there is a second, implied,
2054 empty branch, so just skip over the conditional, because it could be empty.
2055 Otherwise, scan the individual branches of the group. */
2056
2057 if (c == OP_COND && code[GET(code, 1)] != OP_ALT)
2058 code += GET(code, 1);
2059 else
2060 {
2061 empty_branch = FALSE;
2062 do
2063 {
2064 if (!empty_branch && could_be_empty_branch(code, endcode, utf8, cd))
2065 empty_branch = TRUE;
2066 code += GET(code, 1);
2067 }
2068 while (*code == OP_ALT);
2069 if (!empty_branch) return FALSE; /* All branches are non-empty */
2070 }
2071
2072 c = *code;
2073 continue;
2074 }
2075
2076 /* Handle the other opcodes */
2077
2078 switch (c)
2079 {
2080 /* Check for quantifiers after a class. XCLASS is used for classes that
2081 cannot be represented just by a bit map. This includes negated single
2082 high-valued characters. The length in _pcre_OP_lengths[] is zero; the
2083 actual length is stored in the compiled code, so we must update "code"
2084 here. */
2085
2086 #ifdef SUPPORT_UTF8
2087 case OP_XCLASS:
2088 ccode = code += GET(code, 1);
2089 goto CHECK_CLASS_REPEAT;
2090 #endif
2091
2092 case OP_CLASS:
2093 case OP_NCLASS:
2094 ccode = code + 33;
2095
2096 #ifdef SUPPORT_UTF8
2097 CHECK_CLASS_REPEAT:
2098 #endif
2099
2100 switch (*ccode)
2101 {
2102 case OP_CRSTAR: /* These could be empty; continue */
2103 case OP_CRMINSTAR:
2104 case OP_CRQUERY:
2105 case OP_CRMINQUERY:
2106 break;
2107
2108 default: /* Non-repeat => class must match */
2109 case OP_CRPLUS: /* These repeats aren't empty */
2110 case OP_CRMINPLUS:
2111 return FALSE;
2112
2113 case OP_CRRANGE:
2114 case OP_CRMINRANGE:
2115 if (GET2(ccode, 1) > 0) return FALSE; /* Minimum > 0 */
2116 break;
2117 }
2118 break;
2119
2120 /* Opcodes that must match a character */
2121
2122 case OP_PROP:
2123 case OP_NOTPROP:
2124 case OP_EXTUNI:
2125 case OP_NOT_DIGIT:
2126 case OP_DIGIT:
2127 case OP_NOT_WHITESPACE:
2128 case OP_WHITESPACE:
2129 case OP_NOT_WORDCHAR:
2130 case OP_WORDCHAR:
2131 case OP_ANY:
2132 case OP_ALLANY:
2133 case OP_ANYBYTE:
2134 case OP_CHAR:
2135 case OP_CHARI:
2136 case OP_NOT:
2137 case OP_NOTI:
2138 case OP_PLUS:
2139 case OP_MINPLUS:
2140 case OP_POSPLUS:
2141 case OP_EXACT:
2142 case OP_NOTPLUS:
2143 case OP_NOTMINPLUS:
2144 case OP_NOTPOSPLUS:
2145 case OP_NOTEXACT:
2146 case OP_TYPEPLUS:
2147 case OP_TYPEMINPLUS:
2148 case OP_TYPEPOSPLUS:
2149 case OP_TYPEEXACT:
2150 return FALSE;
2151
2152 /* These are going to continue, as they may be empty, but we have to
2153 fudge the length for the \p and \P cases. */
2154
2155 case OP_TYPESTAR:
2156 case OP_TYPEMINSTAR:
2157 case OP_TYPEPOSSTAR:
2158 case OP_TYPEQUERY:
2159 case OP_TYPEMINQUERY:
2160 case OP_TYPEPOSQUERY:
2161 if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
2162 break;
2163
2164 /* Same for these */
2165
2166 case OP_TYPEUPTO:
2167 case OP_TYPEMINUPTO:
2168 case OP_TYPEPOSUPTO:
2169 if (code[3] == OP_PROP || code[3] == OP_NOTPROP) code += 2;
2170 break;
2171
2172 /* End of branch */
2173
2174 case OP_KET:
2175 case OP_KETRMAX:
2176 case OP_KETRMIN:
2177 case OP_KETRPOS:
2178 case OP_ALT:
2179 return TRUE;
2180
2181 /* In UTF-8 mode, STAR, MINSTAR, POSSTAR, QUERY, MINQUERY, POSQUERY, UPTO,
2182 MINUPTO, and POSUPTO may be followed by a multibyte character */
2183
2184 #ifdef SUPPORT_UTF8
2185 case OP_STAR:
2186 case OP_STARI:
2187 case OP_MINSTAR:
2188 case OP_MINSTARI:
2189 case OP_POSSTAR:
2190 case OP_POSSTARI:
2191 case OP_QUERY:
2192 case OP_QUERYI:
2193 case OP_MINQUERY:
2194 case OP_MINQUERYI:
2195 case OP_POSQUERY:
2196 case OP_POSQUERYI:
2197 if (utf8 && code[1] >= 0xc0) code += _pcre_utf8_table4[code[1] & 0x3f];
2198 break;
2199
2200 case OP_UPTO:
2201 case OP_UPTOI:
2202 case OP_MINUPTO:
2203 case OP_MINUPTOI:
2204 case OP_POSUPTO:
2205 case OP_POSUPTOI:
2206 if (utf8 && code[3] >= 0xc0) code += _pcre_utf8_table4[code[3] & 0x3f];
2207 break;
2208 #endif
2209
2210 /* MARK, and PRUNE/SKIP/THEN with an argument must skip over the argument
2211 string. */
2212
2213 case OP_MARK:
2214 case OP_PRUNE_ARG:
2215 case OP_SKIP_ARG:
2216 code += code[1];
2217 break;
2218
2219 case OP_THEN_ARG:
2220 code += code[1+LINK_SIZE];
2221 break;
2222
2223 /* None of the remaining opcodes are required to match a character. */
2224
2225 default:
2226 break;
2227 }
2228 }
2229
2230 return TRUE;
2231 }
2232
2233
2234
2235 /*************************************************
2236 * Scan compiled regex for non-emptiness *
2237 *************************************************/
2238
2239 /* This function is called to check for left recursive calls. We want to check
2240 the current branch of the current pattern to see if it could match the empty
2241 string. If it could, we must look outwards for branches at other levels,
2242 stopping when we pass beyond the bracket which is the subject of the recursion.
2243 This function is called only during the real compile, not during the
2244 pre-compile.
2245
2246 Arguments:
2247 code points to start of the recursion
2248 endcode points to where to stop (current RECURSE item)
2249 bcptr points to the chain of current (unclosed) branch starts
2250 utf8 TRUE if in UTF-8 mode
2251 cd pointers to tables etc
2252
2253 Returns: TRUE if what is matched could be empty
2254 */
2255
2256 static BOOL
2257 could_be_empty(const uschar *code, const uschar *endcode, branch_chain *bcptr,
2258 BOOL utf8, compile_data *cd)
2259 {
2260 while (bcptr != NULL && bcptr->current_branch >= code)
2261 {
2262 if (!could_be_empty_branch(bcptr->current_branch, endcode, utf8, cd))
2263 return FALSE;
2264 bcptr = bcptr->outer;
2265 }
2266 return TRUE;
2267 }
2268
2269
2270
2271 /*************************************************
2272 * Check for POSIX class syntax *
2273 *************************************************/
2274
2275 /* This function is called when the sequence "[:" or "[." or "[=" is
2276 encountered in a character class. It checks whether this is followed by a
2277 sequence of characters terminated by a matching ":]" or ".]" or "=]". If we
2278 reach an unescaped ']' without the special preceding character, return FALSE.
2279
2280 Originally, this function only recognized a sequence of letters between the
2281 terminators, but it seems that Perl recognizes any sequence of characters,
2282 though of course unknown POSIX names are subsequently rejected. Perl gives an
2283 "Unknown POSIX class" error for [:f\oo:] for example, where previously PCRE
2284 didn't consider this to be a POSIX class. Likewise for [:1234:].
2285
2286 The problem in trying to be exactly like Perl is in the handling of escapes. We
2287 have to be sure that [abc[:x\]pqr] is *not* treated as containing a POSIX
2288 class, but [abc[:x\]pqr:]] is (so that an error can be generated). The code
2289 below handles the special case of \], but does not try to do any other escape
2290 processing. This makes it different from Perl for cases such as [:l\ower:]
2291 where Perl recognizes it as the POSIX class "lower" but PCRE does not recognize
2292 "l\ower". This is a lesser evil that not diagnosing bad classes when Perl does,
2293 I think.
2294
2295 A user pointed out that PCRE was rejecting [:a[:digit:]] whereas Perl was not.
2296 It seems that the appearance of a nested POSIX class supersedes an apparent
2297 external class. For example, [:a[:digit:]b:] matches "a", "b", ":", or
2298 a digit.
2299
2300 In Perl, unescaped square brackets may also appear as part of class names. For
2301 example, [:a[:abc]b:] gives unknown POSIX class "[:abc]b:]". However, for
2302 [:a[:abc]b][b:] it gives unknown POSIX class "[:abc]b][b:]", which does not
2303 seem right at all. PCRE does not allow closing square brackets in POSIX class
2304 names.
2305
2306 Arguments:
2307 ptr pointer to the initial [
2308 endptr where to return the end pointer
2309
2310 Returns: TRUE or FALSE
2311 */
2312
2313 static BOOL
2314 check_posix_syntax(const uschar *ptr, const uschar **endptr)
2315 {
2316 int terminator; /* Don't combine these lines; the Solaris cc */
2317 terminator = *(++ptr); /* compiler warns about "non-constant" initializer. */
2318 for (++ptr; *ptr != 0; ptr++)
2319 {
2320 if (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET)
2321 ptr++;
2322 else if (*ptr == CHAR_RIGHT_SQUARE_BRACKET) return FALSE;
2323 else
2324 {
2325 if (*ptr == terminator && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET)
2326 {
2327 *endptr = ptr;
2328 return TRUE;
2329 }
2330 if (*ptr == CHAR_LEFT_SQUARE_BRACKET &&
2331 (ptr[1] == CHAR_COLON || ptr[1] == CHAR_DOT ||
2332 ptr[1] == CHAR_EQUALS_SIGN) &&
2333 check_posix_syntax(ptr, endptr))
2334 return FALSE;
2335 }
2336 }
2337 return FALSE;
2338 }
2339
2340
2341
2342
2343 /*************************************************
2344 * Check POSIX class name *
2345 *************************************************/
2346
2347 /* This function is called to check the name given in a POSIX-style class entry
2348 such as [:alnum:].
2349
2350 Arguments:
2351 ptr points to the first letter
2352 len the length of the name
2353
2354 Returns: a value representing the name, or -1 if unknown
2355 */
2356
2357 static int
2358 check_posix_name(const uschar *ptr, int len)
2359 {
2360 const char *pn = posix_names;
2361 register int yield = 0;
2362 while (posix_name_lengths[yield] != 0)
2363 {
2364 if (len == posix_name_lengths[yield] &&
2365 strncmp((const char *)ptr, pn, len) == 0) return yield;
2366 pn += posix_name_lengths[yield] + 1;
2367 yield++;
2368 }
2369 return -1;
2370 }
2371
2372
2373 /*************************************************
2374 * Adjust OP_RECURSE items in repeated group *
2375 *************************************************/
2376
2377 /* OP_RECURSE items contain an offset from the start of the regex to the group
2378 that is referenced. This means that groups can be replicated for fixed
2379 repetition simply by copying (because the recursion is allowed to refer to
2380 earlier groups that are outside the current group). However, when a group is
2381 optional (i.e. the minimum quantifier is zero), OP_BRAZERO or OP_SKIPZERO is
2382 inserted before it, after it has been compiled. This means that any OP_RECURSE
2383 items within it that refer to the group itself or any contained groups have to
2384 have their offsets adjusted. That one of the jobs of this function. Before it
2385 is called, the partially compiled regex must be temporarily terminated with
2386 OP_END.
2387
2388 This function has been extended with the possibility of forward references for
2389 recursions and subroutine calls. It must also check the list of such references
2390 for the group we are dealing with. If it finds that one of the recursions in
2391 the current group is on this list, it adjusts the offset in the list, not the
2392 value in the reference (which is a group number).
2393
2394 Arguments:
2395 group points to the start of the group
2396 adjust the amount by which the group is to be moved
2397 utf8 TRUE in UTF-8 mode
2398 cd contains pointers to tables etc.
2399 save_hwm the hwm forward reference pointer at the start of the group
2400
2401 Returns: nothing
2402 */
2403
2404 static void
2405 adjust_recurse(uschar *group, int adjust, BOOL utf8, compile_data *cd,
2406 uschar *save_hwm)
2407 {
2408 uschar *ptr = group;
2409
2410 while ((ptr = (uschar *)find_recurse(ptr, utf8)) != NULL)
2411 {
2412 int offset;
2413 uschar *hc;
2414
2415 /* See if this recursion is on the forward reference list. If so, adjust the
2416 reference. */
2417
2418 for (hc = save_hwm; hc < cd->hwm; hc += LINK_SIZE)
2419 {
2420 offset = GET(hc, 0);
2421 if (cd->start_code + offset == ptr + 1)
2422 {
2423 PUT(hc, 0, offset + adjust);
2424 break;
2425 }
2426 }
2427
2428 /* Otherwise, adjust the recursion offset if it's after the start of this
2429 group. */
2430
2431 if (hc >= cd->hwm)
2432 {
2433 offset = GET(ptr, 1);
2434 if (cd->start_code + offset >= group) PUT(ptr, 1, offset + adjust);
2435 }
2436
2437 ptr += 1 + LINK_SIZE;
2438 }
2439 }
2440
2441
2442
2443 /*************************************************
2444 * Insert an automatic callout point *
2445 *************************************************/
2446
2447 /* This function is called when the PCRE_AUTO_CALLOUT option is set, to insert
2448 callout points before each pattern item.
2449
2450 Arguments:
2451 code current code pointer
2452 ptr current pattern pointer
2453 cd pointers to tables etc
2454
2455 Returns: new code pointer
2456 */
2457
2458 static uschar *
2459 auto_callout(uschar *code, const uschar *ptr, compile_data *cd)
2460 {
2461 *code++ = OP_CALLOUT;
2462 *code++ = 255;
2463 PUT(code, 0, (int)(ptr - cd->start_pattern)); /* Pattern offset */
2464 PUT(code, LINK_SIZE, 0); /* Default length */
2465 return code + 2*LINK_SIZE;
2466 }
2467
2468
2469
2470 /*************************************************
2471 * Complete a callout item *
2472 *************************************************/
2473
2474 /* A callout item contains the length of the next item in the pattern, which
2475 we can't fill in till after we have reached the relevant point. This is used
2476 for both automatic and manual callouts.
2477
2478 Arguments:
2479 previous_callout points to previous callout item
2480 ptr current pattern pointer
2481 cd pointers to tables etc
2482
2483 Returns: nothing
2484 */
2485
2486 static void
2487 complete_callout(uschar *previous_callout, const uschar *ptr, compile_data *cd)
2488 {
2489 int length = (int)(ptr - cd->start_pattern - GET(previous_callout, 2));
2490 PUT(previous_callout, 2 + LINK_SIZE, length);
2491 }
2492
2493
2494
2495 #ifdef SUPPORT_UCP
2496 /*************************************************
2497 * Get othercase range *
2498 *************************************************/
2499
2500 /* This function is passed the start and end of a class range, in UTF-8 mode
2501 with UCP support. It searches up the characters, looking for internal ranges of
2502 characters in the "other" case. Each call returns the next one, updating the
2503 start address.
2504
2505 Arguments:
2506 cptr points to starting character value; updated
2507 d end value
2508 ocptr where to put start of othercase range
2509 odptr where to put end of othercase range
2510
2511 Yield: TRUE when range returned; FALSE when no more
2512 */
2513
2514 static BOOL
2515 get_othercase_range(unsigned int *cptr, unsigned int d, unsigned int *ocptr,
2516 unsigned int *odptr)
2517 {
2518 unsigned int c, othercase, next;
2519
2520 for (c = *cptr; c <= d; c++)
2521 { if ((othercase = UCD_OTHERCASE(c)) != c) break; }
2522
2523 if (c > d) return FALSE;
2524
2525 *ocptr = othercase;
2526 next = othercase + 1;
2527
2528 for (++c; c <= d; c++)
2529 {
2530 if (UCD_OTHERCASE(c) != next) break;
2531 next++;
2532 }
2533
2534 *odptr = next - 1;
2535 *cptr = c;
2536
2537 return TRUE;
2538 }
2539
2540
2541
2542 /*************************************************
2543 * Check a character and a property *
2544 *************************************************/
2545
2546 /* This function is called by check_auto_possessive() when a property item
2547 is adjacent to a fixed character.
2548
2549 Arguments:
2550 c the character
2551 ptype the property type
2552 pdata the data for the type
2553 negated TRUE if it's a negated property (\P or \p{^)
2554
2555 Returns: TRUE if auto-possessifying is OK
2556 */
2557
2558 static BOOL
2559 check_char_prop(int c, int ptype, int pdata, BOOL negated)
2560 {
2561 const ucd_record *prop = GET_UCD(c);
2562 switch(ptype)
2563 {
2564 case PT_LAMP:
2565 return (prop->chartype == ucp_Lu ||
2566 prop->chartype == ucp_Ll ||
2567 prop->chartype == ucp_Lt) == negated;
2568
2569 case PT_GC:
2570 return (pdata == _pcre_ucp_gentype[prop->chartype]) == negated;
2571
2572 case PT_PC:
2573 return (pdata == prop->chartype) == negated;
2574
2575 case PT_SC:
2576 return (pdata == prop->script) == negated;
2577
2578 /* These are specials */
2579
2580 case PT_ALNUM:
2581 return (_pcre_ucp_gentype[prop->chartype] == ucp_L ||
2582 _pcre_ucp_gentype[prop->chartype] == ucp_N) == negated;
2583
2584 case PT_SPACE: /* Perl space */
2585 return (_pcre_ucp_gentype[prop->chartype] == ucp_Z ||
2586 c == CHAR_HT || c == CHAR_NL || c == CHAR_FF || c == CHAR_CR)
2587 == negated;
2588
2589 case PT_PXSPACE: /* POSIX space */
2590 return (_pcre_ucp_gentype[prop->chartype] == ucp_Z ||
2591 c == CHAR_HT || c == CHAR_NL || c == CHAR_VT ||
2592 c == CHAR_FF || c == CHAR_CR)
2593 == negated;
2594
2595 case PT_WORD:
2596 return (_pcre_ucp_gentype[prop->chartype] == ucp_L ||
2597 _pcre_ucp_gentype[prop->chartype] == ucp_N ||
2598 c == CHAR_UNDERSCORE) == negated;
2599 }
2600 return FALSE;
2601 }
2602 #endif /* SUPPORT_UCP */
2603
2604
2605
2606 /*************************************************
2607 * Check if auto-possessifying is possible *
2608 *************************************************/
2609
2610 /* This function is called for unlimited repeats of certain items, to see
2611 whether the next thing could possibly match the repeated item. If not, it makes
2612 sense to automatically possessify the repeated item.
2613
2614 Arguments:
2615 previous pointer to the repeated opcode
2616 utf8 TRUE in UTF-8 mode
2617 ptr next character in pattern
2618 options options bits
2619 cd contains pointers to tables etc.
2620
2621 Returns: TRUE if possessifying is wanted
2622 */
2623
2624 static BOOL
2625 check_auto_possessive(const uschar *previous, BOOL utf8, const uschar *ptr,
2626 int options, compile_data *cd)
2627 {
2628 int c, next;
2629 int op_code = *previous++;
2630
2631 /* Skip whitespace and comments in extended mode */
2632
2633 if ((options & PCRE_EXTENDED) != 0)
2634 {
2635 for (;;)
2636 {
2637 while ((cd->ctypes[*ptr] & ctype_space) != 0) ptr++;
2638 if (*ptr == CHAR_NUMBER_SIGN)
2639 {
2640 ptr++;
2641 while (*ptr != 0)
2642 {
2643 if (IS_NEWLINE(ptr)) { ptr += cd->nllen; break; }
2644 ptr++;
2645 #ifdef SUPPORT_UTF8
2646 if (utf8) while ((*ptr & 0xc0) == 0x80) ptr++;
2647 #endif
2648 }
2649 }
2650 else break;
2651 }
2652 }
2653
2654 /* If the next item is one that we can handle, get its value. A non-negative
2655 value is a character, a negative value is an escape value. */
2656
2657 if (*ptr == CHAR_BACKSLASH)
2658 {
2659 int temperrorcode = 0;
2660 next = check_escape(&ptr, &temperrorcode, cd->bracount, options, FALSE);
2661 if (temperrorcode != 0) return FALSE;
2662 ptr++; /* Point after the escape sequence */
2663 }
2664
2665 else if ((cd->ctypes[*ptr] & ctype_meta) == 0)
2666 {
2667 #ifdef SUPPORT_UTF8
2668 if (utf8) { GETCHARINC(next, ptr); } else
2669 #endif
2670 next = *ptr++;
2671 }
2672
2673 else return FALSE;
2674
2675 /* Skip whitespace and comments in extended mode */
2676
2677 if ((options & PCRE_EXTENDED) != 0)
2678 {
2679 for (;;)
2680 {
2681 while ((cd->ctypes[*ptr] & ctype_space) != 0) ptr++;
2682 if (*ptr == CHAR_NUMBER_SIGN)
2683 {
2684 ptr++;
2685 while (*ptr != 0)
2686 {
2687 if (IS_NEWLINE(ptr)) { ptr += cd->nllen; break; }
2688 ptr++;
2689 #ifdef SUPPORT_UTF8
2690 if (utf8) while ((*ptr & 0xc0) == 0x80) ptr++;
2691 #endif
2692 }
2693 }
2694 else break;
2695 }
2696 }
2697
2698 /* If the next thing is itself optional, we have to give up. */
2699
2700 if (*ptr == CHAR_ASTERISK || *ptr == CHAR_QUESTION_MARK ||
2701 strncmp((char *)ptr, STR_LEFT_CURLY_BRACKET STR_0 STR_COMMA, 3) == 0)
2702 return FALSE;
2703
2704 /* Now compare the next item with the previous opcode. First, handle cases when
2705 the next item is a character. */
2706
2707 if (next >= 0) switch(op_code)
2708 {
2709 case OP_CHAR:
2710 #ifdef SUPPORT_UTF8
2711 GETCHARTEST(c, previous);
2712 #else
2713 c = *previous;
2714 #endif
2715 return c != next;
2716
2717 /* For CHARI (caseless character) we must check the other case. If we have
2718 Unicode property support, we can use it to test the other case of
2719 high-valued characters. */
2720
2721 case OP_CHARI:
2722 #ifdef SUPPORT_UTF8
2723 GETCHARTEST(c, previous);
2724 #else
2725 c = *previous;
2726 #endif
2727 if (c == next) return FALSE;
2728 #ifdef SUPPORT_UTF8
2729 if (utf8)
2730 {
2731 unsigned int othercase;
2732 if (next < 128) othercase = cd->fcc[next]; else
2733 #ifdef SUPPORT_UCP
2734 othercase = UCD_OTHERCASE((unsigned int)next);
2735 #else
2736 othercase = NOTACHAR;
2737 #endif
2738 return (unsigned int)c != othercase;
2739 }
2740 else
2741 #endif /* SUPPORT_UTF8 */
2742 return (c != cd->fcc[next]); /* Non-UTF-8 mode */
2743
2744 /* For OP_NOT and OP_NOTI, the data is always a single-byte character. These
2745 opcodes are not used for multi-byte characters, because they are coded using
2746 an XCLASS instead. */
2747
2748 case OP_NOT:
2749 return (c = *previous) == next;
2750
2751 case OP_NOTI:
2752 if ((c = *previous) == next) return TRUE;
2753 #ifdef SUPPORT_UTF8
2754 if (utf8)
2755 {
2756 unsigned int othercase;
2757 if (next < 128) othercase = cd->fcc[next]; else
2758 #ifdef SUPPORT_UCP
2759 othercase = UCD_OTHERCASE(next);
2760 #else
2761 othercase = NOTACHAR;
2762 #endif
2763 return (unsigned int)c == othercase;
2764 }
2765 else
2766 #endif /* SUPPORT_UTF8 */
2767 return (c == cd->fcc[next]); /* Non-UTF-8 mode */
2768
2769 /* Note that OP_DIGIT etc. are generated only when PCRE_UCP is *not* set.
2770 When it is set, \d etc. are converted into OP_(NOT_)PROP codes. */
2771
2772 case OP_DIGIT:
2773 return next > 127 || (cd->ctypes[next] & ctype_digit) == 0;
2774
2775 case OP_NOT_DIGIT:
2776 return next <= 127 && (cd->ctypes[next] & ctype_digit) != 0;
2777
2778 case OP_WHITESPACE:
2779 return next > 127 || (cd->ctypes[next] & ctype_space) == 0;
2780
2781 case OP_NOT_WHITESPACE:
2782 return next <= 127 && (cd->ctypes[next] & ctype_space) != 0;
2783
2784 case OP_WORDCHAR:
2785 return next > 127 || (cd->ctypes[next] & ctype_word) == 0;
2786
2787 case OP_NOT_WORDCHAR:
2788 return next <= 127 && (cd->ctypes[next] & ctype_word) != 0;
2789
2790 case OP_HSPACE:
2791 case OP_NOT_HSPACE:
2792 switch(next)
2793 {
2794 case 0x09:
2795 case 0x20:
2796 case 0xa0:
2797 case 0x1680:
2798 case 0x180e:
2799 case 0x2000:
2800 case 0x2001:
2801 case 0x2002:
2802 case 0x2003:
2803 case 0x2004:
2804 case 0x2005:
2805 case 0x2006:
2806 case 0x2007:
2807 case 0x2008:
2808 case 0x2009:
2809 case 0x200A:
2810 case 0x202f:
2811 case 0x205f:
2812 case 0x3000:
2813 return op_code == OP_NOT_HSPACE;
2814 default:
2815 return op_code != OP_NOT_HSPACE;
2816 }
2817
2818 case OP_ANYNL:
2819 case OP_VSPACE:
2820 case OP_NOT_VSPACE:
2821 switch(next)
2822 {
2823 case 0x0a:
2824 case 0x0b:
2825 case 0x0c:
2826 case 0x0d:
2827 case 0x85:
2828 case 0x2028:
2829 case 0x2029:
2830 return op_code == OP_NOT_VSPACE;
2831 default:
2832 return op_code != OP_NOT_VSPACE;
2833 }
2834
2835 #ifdef SUPPORT_UCP
2836 case OP_PROP:
2837 return check_char_prop(next, previous[0], previous[1], FALSE);
2838
2839 case OP_NOTPROP:
2840 return check_char_prop(next, previous[0], previous[1], TRUE);
2841 #endif
2842
2843 default:
2844 return FALSE;
2845 }
2846
2847
2848 /* Handle the case when the next item is \d, \s, etc. Note that when PCRE_UCP
2849 is set, \d turns into ESC_du rather than ESC_d, etc., so ESC_d etc. are
2850 generated only when PCRE_UCP is *not* set, that is, when only ASCII
2851 characteristics are recognized. Similarly, the opcodes OP_DIGIT etc. are
2852 replaced by OP_PROP codes when PCRE_UCP is set. */
2853
2854 switch(op_code)
2855 {
2856 case OP_CHAR:
2857 case OP_CHARI:
2858 #ifdef SUPPORT_UTF8
2859 GETCHARTEST(c, previous);
2860 #else
2861 c = *previous;
2862 #endif
2863 switch(-next)
2864 {
2865 case ESC_d:
2866 return c > 127 || (cd->ctypes[c] & ctype_digit) == 0;
2867
2868 case ESC_D:
2869 return c <= 127 && (cd->ctypes[c] & ctype_digit) != 0;
2870
2871 case ESC_s:
2872 return c > 127 || (cd->ctypes[c] & ctype_space) == 0;
2873
2874 case ESC_S:
2875 return c <= 127 && (cd->ctypes[c] & ctype_space) != 0;
2876
2877 case ESC_w:
2878 return c > 127 || (cd->ctypes[c] & ctype_word) == 0;
2879
2880 case ESC_W:
2881 return c <= 127 && (cd->ctypes[c] & ctype_word) != 0;
2882
2883 case ESC_h:
2884 case ESC_H:
2885 switch(c)
2886 {
2887 case 0x09:
2888 case 0x20:
2889 case 0xa0:
2890 case 0x1680:
2891 case 0x180e:
2892 case 0x2000:
2893 case 0x2001:
2894 case 0x2002:
2895 case 0x2003:
2896 case 0x2004:
2897 case 0x2005:
2898 case 0x2006:
2899 case 0x2007:
2900 case 0x2008:
2901 case 0x2009:
2902 case 0x200A:
2903 case 0x202f:
2904 case 0x205f:
2905 case 0x3000:
2906 return -next != ESC_h;
2907 default:
2908 return -next == ESC_h;
2909 }
2910
2911 case ESC_v:
2912 case ESC_V:
2913 switch(c)
2914 {
2915 case 0x0a:
2916 case 0x0b:
2917 case 0x0c:
2918 case 0x0d:
2919 case 0x85:
2920 case 0x2028:
2921 case 0x2029:
2922 return -next != ESC_v;
2923 default:
2924 return -next == ESC_v;
2925 }
2926
2927 /* When PCRE_UCP is set, these values get generated for \d etc. Find
2928 their substitutions and process them. The result will always be either
2929 -ESC_p or -ESC_P. Then fall through to process those values. */
2930
2931 #ifdef SUPPORT_UCP
2932 case ESC_du:
2933 case ESC_DU:
2934 case ESC_wu:
2935 case ESC_WU:
2936 case ESC_su:
2937 case ESC_SU:
2938 {
2939 int temperrorcode = 0;
2940 ptr = substitutes[-next - ESC_DU];
2941 next = check_escape(&ptr, &temperrorcode, 0, options, FALSE);
2942 if (temperrorcode != 0) return FALSE;
2943 ptr++; /* For compatibility */
2944 }
2945 /* Fall through */
2946
2947 case ESC_p:
2948 case ESC_P:
2949 {
2950 int ptype, pdata, errorcodeptr;
2951 BOOL negated;
2952
2953 ptr--; /* Make ptr point at the p or P */
2954 ptype = get_ucp(&ptr, &negated, &pdata, &errorcodeptr);
2955 if (ptype < 0) return FALSE;
2956 ptr++; /* Point past the final curly ket */
2957
2958 /* If the property item is optional, we have to give up. (When generated
2959 from \d etc by PCRE_UCP, this test will have been applied much earlier,
2960 to the original \d etc. At this point, ptr will point to a zero byte. */
2961
2962 if (*ptr == CHAR_ASTERISK || *ptr == CHAR_QUESTION_MARK ||
2963 strncmp((char *)ptr, STR_LEFT_CURLY_BRACKET STR_0 STR_COMMA, 3) == 0)
2964 return FALSE;
2965
2966 /* Do the property check. */
2967
2968 return check_char_prop(c, ptype, pdata, (next == -ESC_P) != negated);
2969 }
2970 #endif
2971
2972 default:
2973 return FALSE;
2974 }
2975
2976 /* In principle, support for Unicode properties should be integrated here as
2977 well. It means re-organizing the above code so as to get hold of the property
2978 values before switching on the op-code. However, I wonder how many patterns
2979 combine ASCII \d etc with Unicode properties? (Note that if PCRE_UCP is set,
2980 these op-codes are never generated.) */
2981
2982 case OP_DIGIT:
2983 return next == -ESC_D || next == -ESC_s || next == -ESC_W ||
2984 next == -ESC_h || next == -ESC_v || next == -ESC_R;
2985
2986 case OP_NOT_DIGIT:
2987 return next == -ESC_d;
2988
2989 case OP_WHITESPACE:
2990 return next == -ESC_S || next == -ESC_d || next == -ESC_w || next == -ESC_R;
2991
2992 case OP_NOT_WHITESPACE:
2993 return next == -ESC_s || next == -ESC_h || next == -ESC_v;
2994
2995 case OP_HSPACE:
2996 return next == -ESC_S || next == -ESC_H || next == -ESC_d ||
2997 next == -ESC_w || next == -ESC_v || next == -ESC_R;
2998
2999 case OP_NOT_HSPACE:
3000 return next == -ESC_h;
3001
3002 /* Can't have \S in here because VT matches \S (Perl anomaly) */
3003 case OP_ANYNL:
3004 case OP_VSPACE:
3005 return next == -ESC_V || next == -ESC_d || next == -ESC_w;
3006
3007 case OP_NOT_VSPACE:
3008 return next == -ESC_v || next == -ESC_R;
3009
3010 case OP_WORDCHAR:
3011 return next == -ESC_W || next == -ESC_s || next == -ESC_h ||
3012 next == -ESC_v || next == -ESC_R;
3013
3014 case OP_NOT_WORDCHAR:
3015 return next == -ESC_w || next == -ESC_d;
3016
3017 default:
3018 return FALSE;
3019 }
3020
3021 /* Control does not reach here */
3022 }
3023
3024
3025
3026 /*************************************************
3027 * Compile one branch *
3028 *************************************************/
3029
3030 /* Scan the pattern, compiling it into the a vector. If the options are
3031 changed during the branch, the pointer is used to change the external options
3032 bits. This function is used during the pre-compile phase when we are trying
3033 to find out the amount of memory needed, as well as during the real compile
3034 phase. The value of lengthptr distinguishes the two phases.
3035
3036 Arguments:
3037 optionsptr pointer to the option bits
3038 codeptr points to the pointer to the current code point
3039 ptrptr points to the current pattern pointer
3040 errorcodeptr points to error code variable
3041 firstbyteptr set to initial literal character, or < 0 (REQ_UNSET, REQ_NONE)
3042 reqbyteptr set to the last literal character required, else < 0
3043 bcptr points to current branch chain
3044 cond_depth conditional nesting depth
3045 cd contains pointers to tables etc.
3046 lengthptr NULL during the real compile phase
3047 points to length accumulator during pre-compile phase
3048
3049 Returns: TRUE on success
3050 FALSE, with *errorcodeptr set non-zero on error
3051 */
3052
3053 static BOOL
3054 compile_branch(int *optionsptr, uschar **codeptr, const uschar **ptrptr,
3055 int *errorcodeptr, int *firstbyteptr, int *reqbyteptr, branch_chain *bcptr,
3056 int cond_depth, compile_data *cd, int *lengthptr)
3057 {
3058 int repeat_type, op_type;
3059 int repeat_min = 0, repeat_max = 0; /* To please picky compilers */
3060 int bravalue = 0;
3061 int greedy_default, greedy_non_default;
3062 int firstbyte, reqbyte;
3063 int zeroreqbyte, zerofirstbyte;
3064 int req_caseopt, reqvary, tempreqvary;
3065 int options = *optionsptr; /* May change dynamically */
3066 int after_manual_callout = 0;
3067 int length_prevgroup = 0;
3068 register int c;
3069 register uschar *code = *codeptr;
3070 uschar *last_code = code;
3071 uschar *orig_code = code;
3072 uschar *tempcode;
3073 BOOL inescq = FALSE;
3074 BOOL groupsetfirstbyte = FALSE;
3075 const uschar *ptr = *ptrptr;
3076 const uschar *tempptr;
3077 const uschar *nestptr = NULL;
3078 uschar *previous = NULL;
3079 uschar *previous_callout = NULL;
3080 uschar *save_hwm = NULL;
3081 uschar classbits[32];
3082
3083 /* We can fish out the UTF-8 setting once and for all into a BOOL, but we
3084 must not do this for other options (e.g. PCRE_EXTENDED) because they may change
3085 dynamically as we process the pattern. */
3086
3087 #ifdef SUPPORT_UTF8
3088 BOOL class_utf8;
3089 BOOL utf8 = (options & PCRE_UTF8) != 0;
3090 uschar *class_utf8data;
3091 uschar *class_utf8data_base;
3092 uschar utf8_char[6];
3093 #else
3094 BOOL utf8 = FALSE;
3095 #endif
3096
3097 #ifdef PCRE_DEBUG
3098 if (lengthptr != NULL) DPRINTF((">> start branch\n"));
3099 #endif
3100
3101 /* Set up the default and non-default settings for greediness */
3102
3103 greedy_default = ((options & PCRE_UNGREEDY) != 0);
3104 greedy_non_default = greedy_default ^ 1;
3105
3106 /* Initialize no first byte, no required byte. REQ_UNSET means "no char
3107 matching encountered yet". It gets changed to REQ_NONE if we hit something that
3108 matches a non-fixed char first char; reqbyte just remains unset if we never
3109 find one.
3110
3111 When we hit a repeat whose minimum is zero, we may have to adjust these values
3112 to take the zero repeat into account. This is implemented by setting them to
3113 zerofirstbyte and zeroreqbyte when such a repeat is encountered. The individual
3114 item types that can be repeated set these backoff variables appropriately. */
3115
3116 firstbyte = reqbyte = zerofirstbyte = zeroreqbyte = REQ_UNSET;
3117
3118 /* The variable req_caseopt contains either the REQ_CASELESS value or zero,
3119 according to the current setting of the caseless flag. REQ_CASELESS is a bit
3120 value > 255. It is added into the firstbyte or reqbyte variables to record the
3121 case status of the value. This is used only for ASCII characters. */
3122
3123 req_caseopt = ((options & PCRE_CASELESS) != 0)? REQ_CASELESS : 0;
3124
3125 /* Switch on next character until the end of the branch */
3126
3127 for (;; ptr++)
3128 {
3129 BOOL negate_class;
3130 BOOL should_flip_negation;
3131 BOOL possessive_quantifier;
3132 BOOL is_quantifier;
3133 BOOL is_recurse;
3134 BOOL reset_bracount;
3135 int class_charcount;
3136 int class_lastchar;
3137 int newoptions;
3138 int recno;
3139 int refsign;
3140 int skipbytes;
3141 int subreqbyte;
3142 int subfirstbyte;
3143 int terminator;
3144 int mclength;
3145 uschar mcbuffer[8];
3146
3147 /* Get next byte in the pattern */
3148
3149 c = *ptr;
3150
3151 /* If we are at the end of a nested substitution, revert to the outer level
3152 string. Nesting only happens one level deep. */
3153
3154 if (c == 0 && nestptr != NULL)
3155 {
3156 ptr = nestptr;
3157 nestptr = NULL;
3158 c = *ptr;
3159 }
3160
3161 /* If we are in the pre-compile phase, accumulate the length used for the
3162 previous cycle of this loop. */
3163
3164 if (lengthptr != NULL)
3165 {
3166 #ifdef PCRE_DEBUG
3167 if (code > cd->hwm) cd->hwm = code; /* High water info */
3168 #endif
3169 if (code > cd->start_workspace + WORK_SIZE_CHECK) /* Check for overrun */
3170 {
3171 *errorcodeptr = ERR52;
3172 goto FAILED;
3173 }
3174
3175 /* There is at least one situation where code goes backwards: this is the
3176 case of a zero quantifier after a class (e.g. [ab]{0}). At compile time,
3177 the class is simply eliminated. However, it is created first, so we have to
3178 allow memory for it. Therefore, don't ever reduce the length at this point.
3179 */
3180
3181 if (code < last_code) code = last_code;
3182
3183 /* Paranoid check for integer overflow */
3184
3185 if (OFLOW_MAX - *lengthptr < code - last_code)
3186 {
3187 *errorcodeptr = ERR20;
3188 goto FAILED;
3189 }
3190
3191 *lengthptr += (int)(code - last_code);
3192 DPRINTF(("length=%d added %d c=%c\n", *lengthptr, code - last_code, c));
3193
3194 /* If "previous" is set and it is not at the start of the work space, move
3195 it back to there, in order to avoid filling up the work space. Otherwise,
3196 if "previous" is NULL, reset the current code pointer to the start. */
3197
3198 if (previous != NULL)
3199 {
3200 if (previous > orig_code)
3201 {
3202 memmove(orig_code, previous, code - previous);
3203 code -= previous - orig_code;
3204 previous = orig_code;
3205 }
3206 }
3207 else code = orig_code;
3208
3209 /* Remember where this code item starts so we can pick up the length
3210 next time round. */
3211
3212 last_code = code;
3213 }
3214
3215 /* In the real compile phase, just check the workspace used by the forward
3216 reference list. */
3217
3218 else if (cd->hwm > cd->start_workspace + WORK_SIZE_CHECK)
3219 {
3220 *errorcodeptr = ERR52;
3221 goto FAILED;
3222 }
3223
3224 /* If in \Q...\E, check for the end; if not, we have a literal */
3225
3226 if (inescq && c != 0)
3227 {
3228 if (c == CHAR_BACKSLASH && ptr[1] == CHAR_E)
3229 {
3230 inescq = FALSE;
3231 ptr++;
3232 continue;
3233 }
3234 else
3235 {
3236 if (previous_callout != NULL)
3237 {
3238 if (lengthptr == NULL) /* Don't attempt in pre-compile phase */
3239 complete_callout(previous_callout, ptr, cd);
3240 previous_callout = NULL;
3241 }
3242 if ((options & PCRE_AUTO_CALLOUT) != 0)
3243 {
3244 previous_callout = code;
3245 code = auto_callout(code, ptr, cd);
3246 }
3247 goto NORMAL_CHAR;
3248 }
3249 }
3250
3251 /* Fill in length of a previous callout, except when the next thing is
3252 a quantifier. */
3253
3254 is_quantifier =
3255 c == CHAR_ASTERISK || c == CHAR_PLUS || c == CHAR_QUESTION_MARK ||
3256 (c == CHAR_LEFT_CURLY_BRACKET && is_counted_repeat(ptr+1));
3257
3258 if (!is_quantifier && previous_callout != NULL &&
3259 after_manual_callout-- <= 0)
3260 {
3261 if (lengthptr == NULL) /* Don't attempt in pre-compile phase */
3262 complete_callout(previous_callout, ptr, cd);
3263 previous_callout = NULL;
3264 }
3265
3266 /* In extended mode, skip white space and comments. */
3267
3268 if ((options & PCRE_EXTENDED) != 0)
3269 {
3270 if ((cd->ctypes[c] & ctype_space) != 0) continue;
3271 if (c == CHAR_NUMBER_SIGN)
3272 {
3273 ptr++;
3274 while (*ptr != 0)
3275 {
3276 if (IS_NEWLINE(ptr)) { ptr += cd->nllen - 1; break; }
3277 ptr++;
3278 #ifdef SUPPORT_UTF8
3279 if (utf8) while ((*ptr & 0xc0) == 0x80) ptr++;
3280 #endif
3281 }
3282 if (*ptr != 0) continue;
3283
3284 /* Else fall through to handle end of string */
3285 c = 0;
3286 }
3287 }
3288
3289 /* No auto callout for quantifiers. */
3290
3291 if ((options & PCRE_AUTO_CALLOUT) != 0 && !is_quantifier)
3292 {
3293 previous_callout = code;
3294 code = auto_callout(code, ptr, cd);
3295 }
3296
3297 switch(c)
3298 {
3299 /* ===================================================================*/
3300 case 0: /* The branch terminates at string end */
3301 case CHAR_VERTICAL_LINE: /* or | or ) */
3302 case CHAR_RIGHT_PARENTHESIS:
3303 *firstbyteptr = firstbyte;
3304 *reqbyteptr = reqbyte;
3305 *codeptr = code;
3306 *ptrptr = ptr;
3307 if (lengthptr != NULL)
3308 {
3309 if (OFLOW_MAX - *lengthptr < code - last_code)
3310 {
3311 *errorcodeptr = ERR20;
3312 goto FAILED;
3313 }
3314 *lengthptr += (int)(code - last_code); /* To include callout length */
3315 DPRINTF((">> end branch\n"));
3316 }
3317 return TRUE;
3318
3319
3320 /* ===================================================================*/
3321 /* Handle single-character metacharacters. In multiline mode, ^ disables
3322 the setting of any following char as a first character. */
3323
3324 case CHAR_CIRCUMFLEX_ACCENT:
3325 previous = NULL;
3326 if ((options & PCRE_MULTILINE) != 0)
3327 {
3328 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
3329 *code++ = OP_CIRCM;
3330 }
3331 else *code++ = OP_CIRC;
3332 break;
3333
3334 case CHAR_DOLLAR_SIGN:
3335 previous = NULL;
3336 *code++ = ((options & PCRE_MULTILINE) != 0)? OP_DOLLM : OP_DOLL;
3337 break;
3338
3339 /* There can never be a first char if '.' is first, whatever happens about
3340 repeats. The value of reqbyte doesn't change either. */
3341
3342 case CHAR_DOT:
3343 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
3344 zerofirstbyte = firstbyte;
3345 zeroreqbyte = reqbyte;
3346 previous = code;
3347 *code++ = ((options & PCRE_DOTALL) != 0)? OP_ALLANY: OP_ANY;
3348 break;
3349
3350
3351 /* ===================================================================*/
3352 /* Character classes. If the included characters are all < 256, we build a
3353 32-byte bitmap of the permitted characters, except in the special case
3354 where there is only one such character. For negated classes, we build the
3355 map as usual, then invert it at the end. However, we use a different opcode
3356 so that data characters > 255 can be handled correctly.
3357
3358 If the class contains characters outside the 0-255 range, a different
3359 opcode is compiled. It may optionally have a bit map for characters < 256,
3360 but those above are are explicitly listed afterwards. A flag byte tells
3361 whether the bitmap is present, and whether this is a negated class or not.
3362
3363 In JavaScript compatibility mode, an isolated ']' causes an error. In
3364 default (Perl) mode, it is treated as a data character. */
3365
3366 case CHAR_RIGHT_SQUARE_BRACKET:
3367 if ((cd->external_options & PCRE_JAVASCRIPT_COMPAT) != 0)
3368 {
3369 *errorcodeptr = ERR64;
3370 goto FAILED;
3371 }
3372 goto NORMAL_CHAR;
3373
3374 case CHAR_LEFT_SQUARE_BRACKET:
3375 previous = code;
3376
3377 /* PCRE supports POSIX class stuff inside a class. Perl gives an error if
3378 they are encountered at the top level, so we'll do that too. */
3379
3380 if ((ptr[1] == CHAR_COLON || ptr[1] == CHAR_DOT ||
3381 ptr[1] == CHAR_EQUALS_SIGN) &&
3382 check_posix_syntax(ptr, &tempptr))
3383 {
3384 *errorcodeptr = (ptr[1] == CHAR_COLON)? ERR13 : ERR31;
3385 goto FAILED;
3386 }
3387
3388 /* If the first character is '^', set the negation flag and skip it. Also,
3389 if the first few characters (either before or after ^) are \Q\E or \E we
3390 skip them too. This makes for compatibility with Perl. */
3391
3392 negate_class = FALSE;
3393 for (;;)
3394 {
3395 c = *(++ptr);
3396 if (c == CHAR_BACKSLASH)
3397 {
3398 if (ptr[1] == CHAR_E)
3399 ptr++;
3400 else if (strncmp((const char *)ptr+1,
3401 STR_Q STR_BACKSLASH STR_E, 3) == 0)
3402 ptr += 3;
3403 else
3404 break;
3405 }
3406 else if (!negate_class && c == CHAR_CIRCUMFLEX_ACCENT)
3407 negate_class = TRUE;
3408 else break;
3409 }
3410
3411 /* Empty classes are allowed in JavaScript compatibility mode. Otherwise,
3412 an initial ']' is taken as a data character -- the code below handles
3413 that. In JS mode, [] must always fail, so generate OP_FAIL, whereas
3414 [^] must match any character, so generate OP_ALLANY. */
3415
3416 if (c == CHAR_RIGHT_SQUARE_BRACKET &&
3417 (cd->external_options & PCRE_JAVASCRIPT_COMPAT) != 0)
3418 {
3419 *code++ = negate_class? OP_ALLANY : OP_FAIL;
3420 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
3421 zerofirstbyte = firstbyte;
3422 break;
3423 }
3424
3425 /* If a class contains a negative special such as \S, we need to flip the
3426 negation flag at the end, so that support for characters > 255 works
3427 correctly (they are all included in the class). */
3428
3429 should_flip_negation = FALSE;
3430
3431 /* Keep a count of chars with values < 256 so that we can optimize the case
3432 of just a single character (as long as it's < 256). However, For higher
3433 valued UTF-8 characters, we don't yet do any optimization. */
3434
3435 class_charcount = 0;
3436 class_lastchar = -1;
3437
3438 /* Initialize the 32-char bit map to all zeros. We build the map in a
3439 temporary bit of memory, in case the class contains only 1 character (less
3440 than 256), because in that case the compiled code doesn't use the bit map.
3441 */
3442
3443 memset(classbits, 0, 32 * sizeof(uschar));
3444
3445 #ifdef SUPPORT_UTF8
3446 class_utf8 = FALSE; /* No chars >= 256 */
3447 class_utf8data = code + LINK_SIZE + 2; /* For UTF-8 items */
3448 class_utf8data_base = class_utf8data; /* For resetting in pass 1 */
3449 #endif
3450
3451 /* Process characters until ] is reached. By writing this as a "do" it
3452 means that an initial ] is taken as a data character. At the start of the
3453 loop, c contains the first byte of the character. */
3454
3455 if (c != 0) do
3456 {
3457 const uschar *oldptr;
3458
3459 #ifdef SUPPORT_UTF8
3460 if (utf8 && c > 127)
3461 { /* Braces are required because the */
3462 GETCHARLEN(c, ptr, ptr); /* macro generates multiple statements */
3463 }
3464
3465 /* In the pre-compile phase, accumulate the length of any UTF-8 extra
3466 data and reset the pointer. This is so that very large classes that
3467 contain a zillion UTF-8 characters no longer overwrite the work space
3468 (which is on the stack). */
3469
3470 if (lengthptr != NULL)
3471 {
3472 *lengthptr += class_utf8data - class_utf8data_base;
3473 class_utf8data = class_utf8data_base;
3474 }
3475
3476 #endif
3477
3478 /* Inside \Q...\E everything is literal except \E */
3479
3480 if (inescq)
3481 {
3482 if (c == CHAR_BACKSLASH && ptr[1] == CHAR_E) /* If we are at \E */
3483 {
3484 inescq = FALSE; /* Reset literal state */
3485 ptr++; /* Skip the 'E' */
3486 continue; /* Carry on with next */
3487 }
3488 goto CHECK_RANGE; /* Could be range if \E follows */
3489 }
3490
3491 /* Handle POSIX class names. Perl allows a negation extension of the
3492 form [:^name:]. A square bracket that doesn't match the syntax is
3493 treated as a literal. We also recognize the POSIX constructions
3494 [.ch.] and [=ch=] ("collating elements") and fault them, as Perl
3495 5.6 and 5.8 do. */
3496
3497 if (c == CHAR_LEFT_SQUARE_BRACKET &&
3498 (ptr[1] == CHAR_COLON || ptr[1] == CHAR_DOT ||
3499 ptr[1] == CHAR_EQUALS_SIGN) && check_posix_syntax(ptr, &tempptr))
3500 {
3501 BOOL local_negate = FALSE;
3502 int posix_class, taboffset, tabopt;
3503 register const uschar *cbits = cd->cbits;
3504 uschar pbits[32];
3505
3506 if (ptr[1] != CHAR_COLON)
3507 {
3508 *errorcodeptr = ERR31;
3509 goto FAILED;
3510 }
3511
3512 ptr += 2;
3513 if (*ptr == CHAR_CIRCUMFLEX_ACCENT)
3514 {
3515 local_negate = TRUE;
3516 should_flip_negation = TRUE; /* Note negative special */
3517 ptr++;
3518 }
3519
3520 posix_class = check_posix_name(ptr, (int)(tempptr - ptr));
3521 if (posix_class < 0)
3522 {
3523 *errorcodeptr = ERR30;
3524 goto FAILED;
3525 }
3526
3527 /* If matching is caseless, upper and lower are converted to
3528 alpha. This relies on the fact that the class table starts with
3529 alpha, lower, upper as the first 3 entries. */
3530
3531 if ((options & PCRE_CASELESS) != 0 && posix_class <= 2)
3532 posix_class = 0;
3533
3534 /* When PCRE_UCP is set, some of the POSIX classes are converted to
3535 different escape sequences that use Unicode properties. */
3536
3537 #ifdef SUPPORT_UCP
3538 if ((options & PCRE_UCP) != 0)
3539 {
3540 int pc = posix_class + ((local_negate)? POSIX_SUBSIZE/2 : 0);
3541 if (posix_substitutes[pc] != NULL)
3542 {
3543 nestptr = tempptr + 1;
3544 ptr = posix_substitutes[pc] - 1;
3545 continue;
3546 }
3547 }
3548 #endif
3549 /* In the non-UCP case, we build the bit map for the POSIX class in a
3550 chunk of local store because we may be adding and subtracting from it,
3551 and we don't want to subtract bits that may be in the main map already.
3552 At the end we or the result into the bit map that is being built. */
3553
3554 posix_class *= 3;
3555
3556 /* Copy in the first table (always present) */
3557
3558 memcpy(pbits, cbits + posix_class_maps[posix_class],
3559 32 * sizeof(uschar));
3560
3561 /* If there is a second table, add or remove it as required. */
3562
3563 taboffset = posix_class_maps[posix_class + 1];
3564 tabopt = posix_class_maps[posix_class + 2];
3565
3566 if (taboffset >= 0)
3567 {
3568 if (tabopt >= 0)
3569 for (c = 0; c < 32; c++) pbits[c] |= cbits[c + taboffset];
3570 else
3571 for (c = 0; c < 32; c++) pbits[c] &= ~cbits[c + taboffset];
3572 }
3573
3574 /* Not see if we need to remove any special characters. An option
3575 value of 1 removes vertical space and 2 removes underscore. */
3576
3577 if (tabopt < 0) tabopt = -tabopt;
3578 if (tabopt == 1) pbits[1] &= ~0x3c;
3579 else if (tabopt == 2) pbits[11] &= 0x7f;
3580
3581 /* Add the POSIX table or its complement into the main table that is
3582 being built and we are done. */
3583
3584 if (local_negate)
3585 for (c = 0; c < 32; c++) classbits[c] |= ~pbits[c];
3586 else
3587 for (c = 0; c < 32; c++) classbits[c] |= pbits[c];
3588
3589 ptr = tempptr + 1;
3590 class_charcount = 10; /* Set > 1; assumes more than 1 per class */
3591 continue; /* End of POSIX syntax handling */
3592 }
3593
3594 /* Backslash may introduce a single character, or it may introduce one
3595 of the specials, which just set a flag. The sequence \b is a special
3596 case. Inside a class (and only there) it is treated as backspace. We
3597 assume that other escapes have more than one character in them, so set
3598 class_charcount bigger than one. Unrecognized escapes fall through and
3599 are either treated as literal characters (by default), or are faulted if
3600 PCRE_EXTRA is set. */
3601
3602 if (c == CHAR_BACKSLASH)
3603 {
3604 c = check_escape(&ptr, errorcodeptr, cd->bracount, options, TRUE);
3605 if (*errorcodeptr != 0) goto FAILED;
3606
3607 if (-c == ESC_b) c = CHAR_BS; /* \b is backspace in a class */
3608 else if (-c == ESC_Q) /* Handle start of quoted string */
3609 {
3610 if (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
3611 {
3612 ptr += 2; /* avoid empty string */
3613 }
3614 else inescq = TRUE;
3615 continue;
3616 }
3617 else if (-c == ESC_E) continue; /* Ignore orphan \E */
3618
3619 if (c < 0)
3620 {
3621 register const uschar *cbits = cd->cbits;
3622 class_charcount += 2; /* Greater than 1 is what matters */
3623
3624 switch (-c)
3625 {
3626 #ifdef SUPPORT_UCP
3627 case ESC_du: /* These are the values given for \d etc */
3628 case ESC_DU: /* when PCRE_UCP is set. We replace the */
3629 case ESC_wu: /* escape sequence with an appropriate \p */
3630 case ESC_WU: /* or \P to test Unicode properties instead */
3631 case ESC_su: /* of the default ASCII testing. */
3632 case ESC_SU:
3633 nestptr = ptr;
3634 ptr = substitutes[-c - ESC_DU] - 1; /* Just before substitute */
3635 class_charcount -= 2; /* Undo! */
3636 continue;
3637 #endif
3638 case ESC_d:
3639 for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_digit];
3640 continue;
3641
3642 case ESC_D:
3643 should_flip_negation = TRUE;
3644 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_digit];
3645 continue;
3646
3647 case ESC_w:
3648 for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_word];
3649 continue;
3650
3651 case ESC_W:
3652 should_flip_negation = TRUE;
3653 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_word];
3654 continue;
3655
3656 /* Perl 5.004 onwards omits VT from \s, but we must preserve it
3657 if it was previously set by something earlier in the character
3658 class. */
3659
3660 case ESC_s:
3661 classbits[0] |= cbits[cbit_space];
3662 classbits[1] |= cbits[cbit_space+1] & ~0x08;
3663 for (c = 2; c < 32; c++) classbits[c] |= cbits[c+cbit_space];
3664 continue;
3665
3666 case ESC_S:
3667 should_flip_negation = TRUE;
3668 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_space];
3669 classbits[1] |= 0x08; /* Perl 5.004 onwards omits VT from \s */
3670 continue;
3671
3672 case ESC_h:
3673 SETBIT(classbits, 0x09); /* VT */
3674 SETBIT(classbits, 0x20); /* SPACE */
3675 SETBIT(classbits, 0xa0); /* NSBP */
3676 #ifdef SUPPORT_UTF8
3677 if (utf8)
3678 {
3679 class_utf8 = TRUE;
3680 *class_utf8data++ = XCL_SINGLE;
3681 class_utf8data += _pcre_ord2utf8(0x1680, class_utf8data);
3682 *class_utf8data++ = XCL_SINGLE;
3683 class_utf8data += _pcre_ord2utf8(0x180e, class_utf8data);
3684 *class_utf8data++ = XCL_RANGE;
3685 class_utf8data += _pcre_ord2utf8(0x2000, class_utf8data);
3686 class_utf8data += _pcre_ord2utf8(0x200A, class_utf8data);
3687 *class_utf8data++ = XCL_SINGLE;
3688 class_utf8data += _pcre_ord2utf8(0x202f, class_utf8data);
3689 *class_utf8data++ = XCL_SINGLE;
3690 class_utf8data += _pcre_ord2utf8(0x205f, class_utf8data);
3691 *class_utf8data++ = XCL_SINGLE;
3692 class_utf8data += _pcre_ord2utf8(0x3000, class_utf8data);
3693 }
3694 #endif
3695 continue;
3696
3697 case ESC_H:
3698 for (c = 0; c < 32; c++)
3699 {
3700 int x = 0xff;
3701 switch (c)
3702 {
3703 case 0x09/8: x ^= 1 << (0x09%8); break;
3704 case 0x20/8: x ^= 1 << (0x20%8); break;
3705 case 0xa0/8: x ^= 1 << (0xa0%8); break;
3706 default: break;
3707 }
3708 classbits[c] |= x;
3709 }
3710
3711 #ifdef SUPPORT_UTF8
3712 if (utf8)
3713 {
3714 class_utf8 = TRUE;
3715 *class_utf8data++ = XCL_RANGE;
3716 class_utf8data += _pcre_ord2utf8(0x0100, class_utf8data);
3717 class_utf8data += _pcre_ord2utf8(0x167f, class_utf8data);
3718 *class_utf8data++ = XCL_RANGE;
3719 class_utf8data += _pcre_ord2utf8(0x1681, class_utf8data);
3720 class_utf8data += _pcre_ord2utf8(0x180d, class_utf8data);
3721 *class_utf8data++ = XCL_RANGE;
3722 class_utf8data += _pcre_ord2utf8(0x180f, class_utf8data);
3723 class_utf8data += _pcre_ord2utf8(0x1fff, class_utf8data);
3724 *class_utf8data++ = XCL_RANGE;
3725 class_utf8data += _pcre_ord2utf8(0x200B, class_utf8data);
3726 class_utf8data += _pcre_ord2utf8(0x202e, class_utf8data);
3727 *class_utf8data++ = XCL_RANGE;
3728 class_utf8data += _pcre_ord2utf8(0x2030, class_utf8data);
3729 class_utf8data += _pcre_ord2utf8(0x205e, class_utf8data);
3730 *class_utf8data++ = XCL_RANGE;
3731 class_utf8data += _pcre_ord2utf8(0x2060, class_utf8data);
3732 class_utf8data += _pcre_ord2utf8(0x2fff, class_utf8data);
3733 *class_utf8data++ = XCL_RANGE;
3734 class_utf8data += _pcre_ord2utf8(0x3001, class_utf8data);
3735 class_utf8data += _pcre_ord2utf8(0x7fffffff, class_utf8data);
3736 }
3737 #endif
3738 continue;
3739
3740 case ESC_v:
3741 SETBIT(classbits, 0x0a); /* LF */
3742 SETBIT(classbits, 0x0b); /* VT */
3743 SETBIT(classbits, 0x0c); /* FF */
3744 SETBIT(classbits, 0x0d); /* CR */
3745 SETBIT(classbits, 0x85); /* NEL */
3746 #ifdef SUPPORT_UTF8
3747 if (utf8)
3748 {
3749 class_utf8 = TRUE;
3750 *class_utf8data++ = XCL_RANGE;
3751 class_utf8data += _pcre_ord2utf8(0x2028, class_utf8data);
3752 class_utf8data += _pcre_ord2utf8(0x2029, class_utf8data);
3753 }
3754 #endif
3755 continue;
3756
3757 case ESC_V:
3758 for (c = 0; c < 32; c++)
3759 {
3760 int x = 0xff;
3761 switch (c)
3762 {
3763 case 0x0a/8: x ^= 1 << (0x0a%8);
3764 x ^= 1 << (0x0b%8);
3765 x ^= 1 << (0x0c%8);
3766 x ^= 1 << (0x0d%8);
3767 break;
3768 case 0x85/8: x ^= 1 << (0x85%8); break;
3769 default: break;
3770 }
3771 classbits[c] |= x;
3772 }
3773
3774 #ifdef SUPPORT_UTF8
3775 if (utf8)
3776 {
3777 class_utf8 = TRUE;
3778 *class_utf8data++ = XCL_RANGE;
3779 class_utf8data += _pcre_ord2utf8(0x0100, class_utf8data);
3780 class_utf8data += _pcre_ord2utf8(0x2027, class_utf8data);
3781 *class_utf8data++ = XCL_RANGE;
3782 class_utf8data += _pcre_ord2utf8(0x2029, class_utf8data);
3783 class_utf8data += _pcre_ord2utf8(0x7fffffff, class_utf8data);
3784 }
3785 #endif
3786 continue;
3787
3788 #ifdef SUPPORT_UCP
3789 case ESC_p:
3790 case ESC_P:
3791 {
3792 BOOL negated;
3793 int pdata;
3794 int ptype = get_ucp(&ptr, &negated, &pdata, errorcodeptr);
3795 if (ptype < 0) goto FAILED;
3796 class_utf8 = TRUE;
3797 *class_utf8data++ = ((-c == ESC_p) != negated)?
3798 XCL_PROP : XCL_NOTPROP;
3799 *class_utf8data++ = ptype;
3800 *class_utf8data++ = pdata;
3801 class_charcount -= 2; /* Not a < 256 character */
3802 continue;
3803 }
3804 #endif
3805 /* Unrecognized escapes are faulted if PCRE is running in its
3806 strict mode. By default, for compatibility with Perl, they are
3807 treated as literals. */
3808
3809 default:
3810 if ((options & PCRE_EXTRA) != 0)
3811 {
3812 *errorcodeptr = ERR7;
3813 goto FAILED;
3814 }
3815 class_charcount -= 2; /* Undo the default count from above */
3816 c = *ptr; /* Get the final character and fall through */
3817 break;
3818 }
3819 }
3820
3821 /* Fall through if we have a single character (c >= 0). This may be
3822 greater than 256 in UTF-8 mode. */
3823
3824 } /* End of backslash handling */
3825
3826 /* A single character may be followed by '-' to form a range. However,
3827 Perl does not permit ']' to be the end of the range. A '-' character
3828 at the end is treated as a literal. Perl ignores orphaned \E sequences
3829 entirely. The code for handling \Q and \E is messy. */
3830
3831 CHECK_RANGE:
3832 while (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
3833 {
3834 inescq = FALSE;
3835 ptr += 2;
3836 }
3837
3838 oldptr = ptr;
3839
3840 /* Remember \r or \n */
3841
3842 if (c == CHAR_CR || c == CHAR_NL) cd->external_flags |= PCRE_HASCRORLF;
3843
3844 /* Check for range */
3845
3846 if (!inescq && ptr[1] == CHAR_MINUS)
3847 {
3848 int d;
3849 ptr += 2;
3850 while (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_E) ptr += 2;
3851
3852 /* If we hit \Q (not followed by \E) at this point, go into escaped
3853 mode. */
3854
3855 while (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_Q)
3856 {
3857 ptr += 2;
3858 if (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_E)
3859 { ptr += 2; continue; }
3860 inescq = TRUE;
3861 break;
3862 }
3863
3864 if (*ptr == 0 || (!inescq && *ptr == CHAR_RIGHT_SQUARE_BRACKET))
3865 {
3866 ptr = oldptr;
3867 goto LONE_SINGLE_CHARACTER;
3868 }
3869
3870 #ifdef SUPPORT_UTF8
3871 if (utf8)
3872 { /* Braces are required because the */
3873 GETCHARLEN(d, ptr, ptr); /* macro generates multiple statements */
3874 }
3875 else
3876 #endif
3877 d = *ptr; /* Not UTF-8 mode */
3878
3879 /* The second part of a range can be a single-character escape, but
3880 not any of the other escapes. Perl 5.6 treats a hyphen as a literal
3881 in such circumstances. */
3882
3883 if (!inescq && d == CHAR_BACKSLASH)
3884 {
3885 d = check_escape(&ptr, errorcodeptr, cd->bracount, options, TRUE);
3886 if (*errorcodeptr != 0) goto FAILED;
3887
3888 /* \b is backspace; any other special means the '-' was literal */
3889
3890 if (d < 0)
3891 {
3892 if (d == -ESC_b) d = CHAR_BS; else
3893 {
3894 ptr = oldptr;
3895 goto LONE_SINGLE_CHARACTER; /* A few lines below */
3896 }
3897 }
3898 }
3899
3900 /* Check that the two values are in the correct order. Optimize
3901 one-character ranges */
3902
3903 if (d < c)
3904 {
3905 *errorcodeptr = ERR8;
3906 goto FAILED;
3907 }
3908
3909 if (d == c) goto LONE_SINGLE_CHARACTER; /* A few lines below */
3910
3911 /* Remember \r or \n */
3912
3913 if (d == CHAR_CR || d == CHAR_NL) cd->external_flags |= PCRE_HASCRORLF;
3914
3915 /* In UTF-8 mode, if the upper limit is > 255, or > 127 for caseless
3916 matching, we have to use an XCLASS with extra data items. Caseless
3917 matching for characters > 127 is available only if UCP support is
3918 available. */
3919
3920 #ifdef SUPPORT_UTF8
3921 if (utf8 && (d > 255 || ((options & PCRE_CASELESS) != 0 && d > 127)))
3922 {
3923 class_utf8 = TRUE;
3924
3925 /* With UCP support, we can find the other case equivalents of
3926 the relevant characters. There may be several ranges. Optimize how
3927 they fit with the basic range. */
3928
3929 #ifdef SUPPORT_UCP
3930 if ((options & PCRE_CASELESS) != 0)
3931 {
3932 unsigned int occ, ocd;
3933 unsigned int cc = c;
3934 unsigned int origd = d;
3935 while (get_othercase_range(&cc, origd, &occ, &ocd))
3936 {
3937 if (occ >= (unsigned int)c &&
3938 ocd <= (unsigned int)d)
3939 continue; /* Skip embedded ranges */
3940
3941 if (occ < (unsigned int)c &&
3942 ocd >= (unsigned int)c - 1) /* Extend the basic range */
3943 { /* if there is overlap, */
3944 c = occ; /* noting that if occ < c */
3945 continue; /* we can't have ocd > d */
3946 } /* because a subrange is */
3947 if (ocd > (unsigned int)d &&
3948 occ <= (unsigned int)d + 1) /* always shorter than */
3949 { /* the basic range. */
3950 d = ocd;
3951 continue;
3952 }
3953
3954 if (occ == ocd)
3955 {
3956 *class_utf8data++ = XCL_SINGLE;
3957 }
3958 else
3959 {
3960 *class_utf8data++ = XCL_RANGE;
3961 class_utf8data += _pcre_ord2utf8(occ, class_utf8data);
3962 }
3963 class_utf8data += _pcre_ord2utf8(ocd, class_utf8data);
3964 }
3965 }
3966 #endif /* SUPPORT_UCP */
3967
3968 /* Now record the original range, possibly modified for UCP caseless
3969 overlapping ranges. */
3970
3971 *class_utf8data++ = XCL_RANGE;
3972 class_utf8data += _pcre_ord2utf8(c, class_utf8data);
3973 class_utf8data += _pcre_ord2utf8(d, class_utf8data);
3974
3975 /* With UCP support, we are done. Without UCP support, there is no
3976 caseless matching for UTF-8 characters > 127; we can use the bit map
3977 for the smaller ones. */
3978
3979 #ifdef SUPPORT_UCP
3980 continue; /* With next character in the class */
3981 #else
3982 if ((options & PCRE_CASELESS) == 0 || c > 127) continue;
3983
3984 /* Adjust upper limit and fall through to set up the map */
3985
3986 d = 127;
3987
3988 #endif /* SUPPORT_UCP */
3989 }
3990 #endif /* SUPPORT_UTF8 */
3991
3992 /* We use the bit map for all cases when not in UTF-8 mode; else
3993 ranges that lie entirely within 0-127 when there is UCP support; else
3994 for partial ranges without UCP support. */
3995
3996 class_charcount += d - c + 1;
3997 class_lastchar = d;
3998
3999 /* We can save a bit of time by skipping this in the pre-compile. */
4000
4001 if (lengthptr == NULL) for (; c <= d; c++)
4002 {
4003 classbits[c/8] |= (1 << (c&7));
4004 if ((options & PCRE_CASELESS) != 0)
4005 {
4006 int uc = cd->fcc[c]; /* flip case */
4007 classbits[uc/8] |= (1 << (uc&7));
4008 }
4009 }
4010
4011 continue; /* Go get the next char in the class */
4012 }
4013
4014 /* Handle a lone single character - we can get here for a normal
4015 non-escape char, or after \ that introduces a single character or for an
4016 apparent range that isn't. */
4017
4018 LONE_SINGLE_CHARACTER:
4019
4020 /* Handle a character that cannot go in the bit map */
4021
4022 #ifdef SUPPORT_UTF8
4023 if (utf8 && (c > 255 || ((options & PCRE_CASELESS) != 0 && c > 127)))
4024 {
4025 class_utf8 = TRUE;
4026 *class_utf8data++ = XCL_SINGLE;
4027 class_utf8data += _pcre_ord2utf8(c, class_utf8data);
4028
4029 #ifdef SUPPORT_UCP
4030 if ((options & PCRE_CASELESS) != 0)
4031 {
4032 unsigned int othercase;
4033 if ((othercase = UCD_OTHERCASE(c)) != c)
4034 {
4035 *class_utf8data++ = XCL_SINGLE;
4036 class_utf8data += _pcre_ord2utf8(othercase, class_utf8data);
4037 }
4038 }
4039 #endif /* SUPPORT_UCP */
4040
4041 }
4042 else
4043 #endif /* SUPPORT_UTF8 */
4044
4045 /* Handle a single-byte character */
4046 {
4047 classbits[c/8] |= (1 << (c&7));
4048 if ((options & PCRE_CASELESS) != 0)
4049 {
4050 c = cd->fcc[c]; /* flip case */
4051 classbits[c/8] |= (1 << (c&7));
4052 }
4053 class_charcount++;
4054 class_lastchar = c;
4055 }
4056 }
4057
4058 /* Loop until ']' reached. This "while" is the end of the "do" far above.
4059 If we are at the end of an internal nested string, revert to the outer
4060 string. */
4061
4062 while (((c = *(++ptr)) != 0 ||
4063 (nestptr != NULL &&
4064 (ptr = nestptr, nestptr = NULL, c = *(++ptr)) != 0)) &&
4065 (c != CHAR_RIGHT_SQUARE_BRACKET || inescq));
4066
4067 /* Check for missing terminating ']' */
4068
4069 if (c == 0)
4070 {
4071 *errorcodeptr = ERR6;
4072 goto FAILED;
4073 }
4074
4075 /* If class_charcount is 1, we saw precisely one character whose value is
4076 less than 256. As long as there were no characters >= 128 and there was no
4077 use of \p or \P, in other words, no use of any XCLASS features, we can
4078 optimize.
4079
4080 In UTF-8 mode, we can optimize the negative case only if there were no
4081 characters >= 128 because OP_NOT and the related opcodes like OP_NOTSTAR
4082 operate on single-bytes characters only. This is an historical hangover.
4083 Maybe one day we can tidy these opcodes to handle multi-byte characters.
4084
4085 The optimization throws away the bit map. We turn the item into a
4086 1-character OP_CHAR[I] if it's positive, or OP_NOT[I] if it's negative.
4087 Note that OP_NOT[I] does not support multibyte characters. In the positive
4088 case, it can cause firstbyte to be set. Otherwise, there can be no first
4089 char if this item is first, whatever repeat count may follow. In the case
4090 of reqbyte, save the previous value for reinstating. */
4091
4092 #ifdef SUPPORT_UTF8
4093 if (class_charcount == 1 && !class_utf8 &&
4094 (!utf8 || !negate_class || class_lastchar < 128))
4095 #else
4096 if (class_charcount == 1)
4097 #endif
4098 {
4099 zeroreqbyte = reqbyte;
4100
4101 /* The OP_NOT[I] opcodes work on one-byte characters only. */
4102
4103 if (negate_class)
4104 {
4105 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
4106 zerofirstbyte = firstbyte;
4107 *code++ = ((options & PCRE_CASELESS) != 0)? OP_NOTI: OP_NOT;
4108 *code++ = class_lastchar;
4109 break;
4110 }
4111
4112 /* For a single, positive character, get the value into mcbuffer, and
4113 then we can handle this with the normal one-character code. */
4114
4115 #ifdef SUPPORT_UTF8
4116 if (utf8 && class_lastchar > 127)
4117 mclength = _pcre_ord2utf8(class_lastchar, mcbuffer);
4118 else
4119 #endif
4120 {
4121 mcbuffer[0] = class_lastchar;
4122 mclength = 1;
4123 }
4124 goto ONE_CHAR;
4125 } /* End of 1-char optimization */
4126
4127 /* The general case - not the one-char optimization. If this is the first
4128 thing in the branch, there can be no first char setting, whatever the
4129 repeat count. Any reqbyte setting must remain unchanged after any kind of
4130 repeat. */
4131
4132 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
4133 zerofirstbyte = firstbyte;
4134 zeroreqbyte = reqbyte;
4135
4136 /* If there are characters with values > 255, we have to compile an
4137 extended class, with its own opcode, unless there was a negated special
4138 such as \S in the class, and PCRE_UCP is not set, because in that case all
4139 characters > 255 are in the class, so any that were explicitly given as
4140 well can be ignored. If (when there are explicit characters > 255 that must
4141 be listed) there are no characters < 256, we can omit the bitmap in the
4142 actual compiled code. */
4143
4144 #ifdef SUPPORT_UTF8
4145 if (class_utf8 && (!should_flip_negation || (options & PCRE_UCP) != 0))
4146 {
4147 *class_utf8data++ = XCL_END; /* Marks the end of extra data */
4148 *code++ = OP_XCLASS;
4149 code += LINK_SIZE;
4150 *code = negate_class? XCL_NOT : 0;
4151
4152 /* If the map is required, move up the extra data to make room for it;
4153 otherwise just move the code pointer to the end of the extra data. */
4154
4155 if (class_charcount > 0)
4156 {
4157 *code++ |= XCL_MAP;
4158 memmove(code + 32, code, class_utf8data - code);
4159 memcpy(code, classbits, 32);
4160 code = class_utf8data + 32;
4161 }
4162 else code = class_utf8data;
4163
4164 /* Now fill in the complete length of the item */
4165
4166 PUT(previous, 1, code - previous);
4167 break; /* End of class handling */
4168 }
4169 #endif
4170
4171 /* If there are no characters > 255, or they are all to be included or
4172 excluded, set the opcode to OP_CLASS or OP_NCLASS, depending on whether the
4173 whole class was negated and whether there were negative specials such as \S
4174 (non-UCP) in the class. Then copy the 32-byte map into the code vector,
4175 negating it if necessary. */
4176
4177 *code++ = (negate_class == should_flip_negation) ? OP_CLASS : OP_NCLASS;
4178 if (negate_class)
4179 {
4180 if (lengthptr == NULL) /* Save time in the pre-compile phase */
4181 for (c = 0; c < 32; c++) code[c] = ~classbits[c];
4182 }
4183 else
4184 {
4185 memcpy(code, classbits, 32);
4186 }
4187 code += 32;
4188 break;
4189
4190
4191 /* ===================================================================*/
4192 /* Various kinds of repeat; '{' is not necessarily a quantifier, but this
4193 has been tested above. */
4194
4195 case CHAR_LEFT_CURLY_BRACKET:
4196 if (!is_quantifier) goto NORMAL_CHAR;
4197 ptr = read_repeat_counts(ptr+1, &repeat_min, &repeat_max, errorcodeptr);
4198 if (*errorcodeptr != 0) goto FAILED;
4199 goto REPEAT;
4200
4201 case CHAR_ASTERISK:
4202 repeat_min = 0;
4203 repeat_max = -1;
4204 goto REPEAT;
4205
4206 case CHAR_PLUS:
4207 repeat_min = 1;
4208 repeat_max = -1;
4209 goto REPEAT;
4210
4211 case CHAR_QUESTION_MARK:
4212 repeat_min = 0;
4213 repeat_max = 1;
4214
4215 REPEAT:
4216 if (previous == NULL)
4217 {
4218 *errorcodeptr = ERR9;
4219 goto FAILED;
4220 }
4221
4222 if (repeat_min == 0)
4223 {
4224 firstbyte = zerofirstbyte; /* Adjust for zero repeat */
4225 reqbyte = zeroreqbyte; /* Ditto */
4226 }
4227
4228 /* Remember whether this is a variable length repeat */
4229
4230 reqvary = (repeat_min == repeat_max)? 0 : REQ_VARY;
4231
4232 op_type = 0; /* Default single-char op codes */
4233 possessive_quantifier = FALSE; /* Default not possessive quantifier */
4234
4235 /* Save start of previous item, in case we have to move it up in order to
4236 insert something before it. */
4237
4238 tempcode = previous;
4239
4240 /* If the next character is '+', we have a possessive quantifier. This
4241 implies greediness, whatever the setting of the PCRE_UNGREEDY option.
4242 If the next character is '?' this is a minimizing repeat, by default,
4243 but if PCRE_UNGREEDY is set, it works the other way round. We change the
4244 repeat type to the non-default. */
4245
4246 if (ptr[1] == CHAR_PLUS)
4247 {
4248 repeat_type = 0; /* Force greedy */
4249 possessive_quantifier = TRUE;
4250 ptr++;
4251 }
4252 else if (ptr[1] == CHAR_QUESTION_MARK)
4253 {
4254 repeat_type = greedy_non_default;
4255 ptr++;
4256 }
4257 else repeat_type = greedy_default;
4258
4259 /* If previous was a recursion call, wrap it in atomic brackets so that
4260 previous becomes the atomic group. All recursions were so wrapped in the
4261 past, but it no longer happens for non-repeated recursions. In fact, the
4262 repeated ones could be re-implemented independently so as not to need this,
4263 but for the moment we rely on the code for repeating groups. */
4264
4265 if (*previous == OP_RECURSE)
4266 {
4267 memmove(previous + 1 + LINK_SIZE, previous, 1 + LINK_SIZE);
4268 *previous = OP_ONCE;
4269 PUT(previous, 1, 2 + 2*LINK_SIZE);
4270 previous[2 + 2*LINK_SIZE] = OP_KET;
4271 PUT(previous, 3 + 2*LINK_SIZE, 2 + 2*LINK_SIZE);
4272 code += 2 + 2 * LINK_SIZE;
4273 length_prevgroup = 3 + 3*LINK_SIZE;
4274
4275 /* When actually compiling, we need to check whether this was a forward
4276 reference, and if so, adjust the offset. */
4277
4278 if (lengthptr == NULL && cd->hwm >= cd->start_workspace + LINK_SIZE)
4279 {
4280 int offset = GET(cd->hwm, -LINK_SIZE);
4281 if (offset == previous + 1 - cd->start_code)
4282 PUT(cd->hwm, -LINK_SIZE, offset + 1 + LINK_SIZE);
4283 }
4284 }
4285
4286 /* Now handle repetition for the different types of item. */
4287
4288 /* If previous was a character match, abolish the item and generate a
4289 repeat item instead. If a char item has a minumum of more than one, ensure
4290 that it is set in reqbyte - it might not be if a sequence such as x{3} is
4291 the first thing in a branch because the x will have gone into firstbyte
4292 instead. */
4293
4294 if (*previous == OP_CHAR || *previous == OP_CHARI)
4295 {
4296 op_type = (*previous == OP_CHAR)? 0 : OP_STARI - OP_STAR;
4297
4298 /* Deal with UTF-8 characters that take up more than one byte. It's
4299 easier to write this out separately than try to macrify it. Use c to
4300 hold the length of the character in bytes, plus 0x80 to flag that it's a
4301 length rather than a small character. */
4302
4303 #ifdef SUPPORT_UTF8
4304 if (utf8 && (code[-1] & 0x80) != 0)
4305 {
4306 uschar *lastchar = code - 1;
4307 while((*lastchar & 0xc0) == 0x80) lastchar--;
4308 c = code - lastchar; /* Length of UTF-8 character */
4309 memcpy(utf8_char, lastchar, c); /* Save the char */
4310 c |= 0x80; /* Flag c as a length */
4311 }
4312 else
4313 #endif
4314
4315 /* Handle the case of a single byte - either with no UTF8 support, or
4316 with UTF-8 disabled, or for a UTF-8 character < 128. */
4317
4318 {
4319 c = code[-1];
4320 if (repeat_min > 1) reqbyte = c | req_caseopt | cd->req_varyopt;
4321 }
4322
4323 /* If the repetition is unlimited, it pays to see if the next thing on
4324 the line is something that cannot possibly match this character. If so,
4325 automatically possessifying this item gains some performance in the case
4326 where the match fails. */
4327
4328 if (!possessive_quantifier &&
4329 repeat_max < 0 &&
4330 check_auto_possessive(previous, utf8, ptr + 1, options, cd))
4331 {
4332 repeat_type = 0; /* Force greedy */
4333 possessive_quantifier = TRUE;
4334 }
4335
4336 goto OUTPUT_SINGLE_REPEAT; /* Code shared with single character types */
4337 }
4338
4339 /* If previous was a single negated character ([^a] or similar), we use
4340 one of the special opcodes, replacing it. The code is shared with single-
4341 character repeats by setting opt_type to add a suitable offset into
4342 repeat_type. We can also test for auto-possessification. OP_NOT and OP_NOTI
4343 are currently used only for single-byte chars. */
4344
4345 else if (*previous == OP_NOT || *previous == OP_NOTI)
4346 {
4347 op_type = ((*previous == OP_NOT)? OP_NOTSTAR : OP_NOTSTARI) - OP_STAR;
4348 c = previous[1];
4349 if (!possessive_quantifier &&
4350 repeat_max < 0 &&
4351 check_auto_possessive(previous, utf8, ptr + 1, options, cd))
4352 {
4353 repeat_type = 0; /* Force greedy */
4354 possessive_quantifier = TRUE;
4355 }
4356 goto OUTPUT_SINGLE_REPEAT;
4357 }
4358
4359 /* If previous was a character type match (\d or similar), abolish it and
4360 create a suitable repeat item. The code is shared with single-character
4361 repeats by setting op_type to add a suitable offset into repeat_type. Note
4362 the the Unicode property types will be present only when SUPPORT_UCP is
4363 defined, but we don't wrap the little bits of code here because it just
4364 makes it horribly messy. */
4365
4366 else if (*previous < OP_EODN)
4367 {
4368 uschar *oldcode;
4369 int prop_type, prop_value;
4370 op_type = OP_TYPESTAR - OP_STAR; /* Use type opcodes */
4371 c = *previous;
4372
4373 if (!possessive_quantifier &&
4374 repeat_max < 0 &&
4375 check_auto_possessive(previous, utf8, ptr + 1, options, cd))
4376 {
4377 repeat_type = 0; /* Force greedy */
4378 possessive_quantifier = TRUE;
4379 }
4380
4381 OUTPUT_SINGLE_REPEAT:
4382 if (*previous == OP_PROP || *previous == OP_NOTPROP)
4383 {
4384 prop_type = previous[1];
4385 prop_value = previous[2];
4386 }
4387 else prop_type = prop_value = -1;
4388
4389 oldcode = code;
4390 code = previous; /* Usually overwrite previous item */
4391
4392 /* If the maximum is zero then the minimum must also be zero; Perl allows
4393 this case, so we do too - by simply omitting the item altogether. */
4394
4395 if (repeat_max == 0) goto END_REPEAT;
4396
4397 /*--------------------------------------------------------------------*/
4398 /* This code is obsolete from release 8.00; the restriction was finally
4399 removed: */
4400
4401 /* All real repeats make it impossible to handle partial matching (maybe
4402 one day we will be able to remove this restriction). */
4403
4404 /* if (repeat_max != 1) cd->external_flags |= PCRE_NOPARTIAL; */
4405 /*--------------------------------------------------------------------*/
4406
4407 /* Combine the op_type with the repeat_type */
4408
4409 repeat_type += op_type;
4410
4411 /* A minimum of zero is handled either as the special case * or ?, or as
4412 an UPTO, with the maximum given. */
4413
4414 if (repeat_min == 0)
4415 {
4416 if (repeat_max == -1) *code++ = OP_STAR + repeat_type;
4417 else if (repeat_max == 1) *code++ = OP_QUERY + repeat_type;
4418 else
4419 {
4420 *code++ = OP_UPTO + repeat_type;
4421 PUT2INC(code, 0, repeat_max);
4422 }
4423 }
4424
4425 /* A repeat minimum of 1 is optimized into some special cases. If the
4426 maximum is unlimited, we use OP_PLUS. Otherwise, the original item is
4427 left in place and, if the maximum is greater than 1, we use OP_UPTO with
4428 one less than the maximum. */
4429
4430 else if (repeat_min == 1)
4431 {
4432 if (repeat_max == -1)
4433 *code++ = OP_PLUS + repeat_type;
4434 else
4435 {
4436 code = oldcode; /* leave previous item in place */
4437 if (repeat_max == 1) goto END_REPEAT;
4438 *code++ = OP_UPTO + repeat_type;
4439 PUT2INC(code, 0, repeat_max - 1);
4440 }
4441 }
4442
4443 /* The case {n,n} is just an EXACT, while the general case {n,m} is
4444 handled as an EXACT followed by an UPTO. */
4445
4446 else
4447 {
4448 *code++ = OP_EXACT + op_type; /* NB EXACT doesn't have repeat_type */
4449 PUT2INC(code, 0, repeat_min);
4450
4451 /* If the maximum is unlimited, insert an OP_STAR. Before doing so,
4452 we have to insert the character for the previous code. For a repeated
4453 Unicode property match, there are two extra bytes that define the
4454 required property. In UTF-8 mode, long characters have their length in
4455 c, with the 0x80 bit as a flag. */
4456
4457 if (repeat_max < 0)
4458 {
4459 #ifdef SUPPORT_UTF8
4460 if (utf8 && c >= 128)
4461 {
4462 memcpy(code, utf8_char, c & 7);
4463 code += c & 7;
4464 }
4465 else
4466 #endif
4467 {
4468 *code++ = c;
4469 if (prop_type >= 0)
4470 {
4471 *code++ = prop_type;
4472 *code++ = prop_value;
4473 }
4474 }
4475 *code++ = OP_STAR + repeat_type;
4476 }
4477
4478 /* Else insert an UPTO if the max is greater than the min, again
4479 preceded by the character, for the previously inserted code. If the
4480 UPTO is just for 1 instance, we can use QUERY instead. */
4481
4482 else if (repeat_max != repeat_min)
4483 {
4484 #ifdef SUPPORT_UTF8
4485 if (utf8 && c >= 128)
4486 {
4487 memcpy(code, utf8_char, c & 7);
4488 code += c & 7;
4489 }
4490 else
4491 #endif
4492 *code++ = c;
4493 if (prop_type >= 0)
4494 {
4495 *code++ = prop_type;
4496 *code++ = prop_value;
4497 }
4498 repeat_max -= repeat_min;
4499
4500 if (repeat_max == 1)
4501 {
4502 *code++ = OP_QUERY + repeat_type;
4503 }
4504 else
4505 {
4506 *code++ = OP_UPTO + repeat_type;
4507 PUT2INC(code, 0, repeat_max);
4508 }
4509 }
4510 }
4511
4512 /* The character or character type itself comes last in all cases. */
4513
4514 #ifdef SUPPORT_UTF8
4515 if (utf8 && c >= 128)
4516 {
4517 memcpy(code, utf8_char, c & 7);
4518 code += c & 7;
4519 }
4520 else
4521 #endif
4522 *code++ = c;
4523
4524 /* For a repeated Unicode property match, there are two extra bytes that
4525 define the required property. */
4526
4527 #ifdef SUPPORT_UCP
4528 if (prop_type >= 0)
4529 {
4530 *code++ = prop_type;
4531 *code++ = prop_value;
4532 }
4533 #endif
4534 }
4535
4536 /* If previous was a character class or a back reference, we put the repeat
4537 stuff after it, but just skip the item if the repeat was {0,0}. */
4538
4539 else if (*previous == OP_CLASS ||
4540 *previous == OP_NCLASS ||
4541 #ifdef SUPPORT_UTF8
4542 *previous == OP_XCLASS ||
4543 #endif
4544 *previous == OP_REF ||
4545 *previous == OP_REFI)
4546 {
4547 if (repeat_max == 0)
4548 {
4549 code = previous;
4550 goto END_REPEAT;
4551 }
4552
4553 /*--------------------------------------------------------------------*/
4554 /* This code is obsolete from release 8.00; the restriction was finally
4555 removed: */
4556
4557 /* All real repeats make it impossible to handle partial matching (maybe
4558 one day we will be able to remove this restriction). */
4559
4560 /* if (repeat_max != 1) cd->external_flags |= PCRE_NOPARTIAL; */
4561 /*--------------------------------------------------------------------*/
4562
4563 if (repeat_min == 0 && repeat_max == -1)
4564 *code++ = OP_CRSTAR + repeat_type;
4565 else if (repeat_min == 1 && repeat_max == -1)
4566 *code++ = OP_CRPLUS + repeat_type;
4567 else if (repeat_min == 0 && repeat_max == 1)
4568 *code++ = OP_CRQUERY + repeat_type;
4569 else
4570 {
4571 *code++ = OP_CRRANGE + repeat_type;
4572 PUT2INC(code, 0, repeat_min);
4573 if (repeat_max == -1) repeat_max = 0; /* 2-byte encoding for max */
4574 PUT2INC(code, 0, repeat_max);
4575 }
4576 }
4577
4578 /* If previous was a bracket group, we may have to replicate it in certain
4579 cases. Note that at this point we can encounter only the "basic" bracket
4580 opcodes such as BRA and CBRA, as this is the place where they get converted
4581 into the more special varieties such as BRAPOS and SBRA. A test for >=
4582 OP_ASSERT and <= OP_COND includes ASSERT, ASSERT_NOT, ASSERTBACK,
4583 ASSERTBACK_NOT, ONCE, BRA, CBRA, and COND. Originally, PCRE did not allow
4584 repetition of assertions, but now it does, for Perl compatibility. */
4585
4586 else if (*previous >= OP_ASSERT && *previous <= OP_COND)
4587 {
4588 register int i;
4589 int len = (int)(code - previous);
4590 uschar *bralink = NULL;
4591 uschar *brazeroptr = NULL;
4592
4593 /* Repeating a DEFINE group is pointless, but Perl allows the syntax, so
4594 we just ignore the repeat. */
4595
4596 if (*previous == OP_COND && previous[LINK_SIZE+1] == OP_DEF)
4597 goto END_REPEAT;
4598
4599 /* There is no sense in actually repeating assertions. The only potential
4600 use of repetition is in cases when the assertion is optional. Therefore,
4601 if the minimum is greater than zero, just ignore the repeat. If the
4602 maximum is not not zero or one, set it to 1. */
4603
4604 if (*previous < OP_ONCE) /* Assertion */
4605 {
4606 if (repeat_min > 0) goto END_REPEAT;
4607 if (repeat_max < 0 || repeat_max > 1) repeat_max = 1;
4608 }
4609
4610 /* The case of a zero minimum is special because of the need to stick
4611 OP_BRAZERO in front of it, and because the group appears once in the
4612 data, whereas in other cases it appears the minimum number of times. For
4613 this reason, it is simplest to treat this case separately, as otherwise
4614 the code gets far too messy. There are several special subcases when the
4615 minimum is zero. */
4616
4617 if (repeat_min == 0)
4618 {
4619 /* If the maximum is also zero, we used to just omit the group from the
4620 output altogether, like this:
4621
4622 ** if (repeat_max == 0)
4623 ** {
4624 ** code = previous;
4625 ** goto END_REPEAT;
4626 ** }
4627
4628 However, that fails when a group or a subgroup within it is referenced
4629 as a subroutine from elsewhere in the pattern, so now we stick in
4630 OP_SKIPZERO in front of it so that it is skipped on execution. As we
4631 don't have a list of which groups are referenced, we cannot do this
4632 selectively.
4633
4634 If the maximum is 1 or unlimited, we just have to stick in the BRAZERO
4635 and do no more at this point. However, we do need to adjust any
4636 OP_RECURSE calls inside the group that refer to the group itself or any
4637 internal or forward referenced group, because the offset is from the
4638 start of the whole regex. Temporarily terminate the pattern while doing
4639 this. */
4640
4641 if (repeat_max <= 1) /* Covers 0, 1, and unlimited */
4642 {
4643 *code = OP_END;
4644 adjust_recurse(previous, 1, utf8, cd, save_hwm);
4645 memmove(previous+1, previous, len);
4646 code++;
4647 if (repeat_max == 0)
4648 {
4649 *previous++ = OP_SKIPZERO;
4650 goto END_REPEAT;
4651 }
4652 brazeroptr = previous; /* Save for possessive optimizing */
4653 *previous++ = OP_BRAZERO + repeat_type;
4654 }
4655
4656 /* If the maximum is greater than 1 and limited, we have to replicate
4657 in a nested fashion, sticking OP_BRAZERO before each set of brackets.
4658 The first one has to be handled carefully because it's the original
4659 copy, which has to be moved up. The remainder can be handled by code
4660 that is common with the non-zero minimum case below. We have to
4661 adjust the value or repeat_max, since one less copy is required. Once
4662 again, we may have to adjust any OP_RECURSE calls inside the group. */
4663
4664 else
4665 {
4666 int offset;
4667 *code = OP_END;
4668 adjust_recurse(previous, 2 + LINK_SIZE, utf8, cd, save_hwm);
4669 memmove(previous + 2 + LINK_SIZE, previous, len);
4670 code += 2 + LINK_SIZE;
4671 *previous++ = OP_BRAZERO + repeat_type;
4672 *previous++ = OP_BRA;
4673
4674 /* We chain together the bracket offset fields that have to be
4675 filled in later when the ends of the brackets are reached. */
4676
4677 offset = (bralink == NULL)? 0 : (int)(previous - bralink);
4678 bralink = previous;
4679 PUTINC(previous, 0, offset);
4680 }
4681
4682 repeat_max--;
4683 }
4684
4685 /* If the minimum is greater than zero, replicate the group as many
4686 times as necessary, and adjust the maximum to the number of subsequent
4687 copies that we need. If we set a first char from the group, and didn't
4688 set a required char, copy the latter from the former. If there are any
4689 forward reference subroutine calls in the group, there will be entries on
4690 the workspace list; replicate these with an appropriate increment. */
4691
4692 else
4693 {
4694 if (repeat_min > 1)
4695 {
4696 /* In the pre-compile phase, we don't actually do the replication. We
4697 just adjust the length as if we had. Do some paranoid checks for
4698 potential integer overflow. The INT64_OR_DOUBLE type is a 64-bit
4699 integer type when available, otherwise double. */
4700
4701 if (lengthptr != NULL)
4702 {
4703 int delta = (repeat_min - 1)*length_prevgroup;
4704 if ((INT64_OR_DOUBLE)(repeat_min - 1)*
4705 (INT64_OR_DOUBLE)length_prevgroup >
4706 (INT64_OR_DOUBLE)INT_MAX ||
4707 OFLOW_MAX - *lengthptr < delta)
4708 {
4709 *errorcodeptr = ERR20;
4710 goto FAILED;
4711 }
4712 *lengthptr += delta;
4713 }
4714
4715 /* This is compiling for real */
4716
4717 else
4718 {
4719 if (groupsetfirstbyte && reqbyte < 0) reqbyte = firstbyte;
4720 for (i = 1; i < repeat_min; i++)
4721 {
4722 uschar *hc;
4723 uschar *this_hwm = cd->hwm;
4724 memcpy(code, previous, len);
4725 for (hc = save_hwm; hc < this_hwm; hc += LINK_SIZE)
4726 {
4727 PUT(cd->hwm, 0, GET(hc, 0) + len);
4728 cd->hwm += LINK_SIZE;
4729 }
4730 save_hwm = this_hwm;
4731 code += len;
4732 }
4733 }
4734 }
4735
4736 if (repeat_max > 0) repeat_max -= repeat_min;
4737 }
4738
4739 /* This code is common to both the zero and non-zero minimum cases. If
4740 the maximum is limited, it replicates the group in a nested fashion,
4741 remembering the bracket starts on a stack. In the case of a zero minimum,
4742 the first one was set up above. In all cases the repeat_max now specifies
4743 the number of additional copies needed. Again, we must remember to
4744 replicate entries on the forward reference list. */
4745
4746 if (repeat_max >= 0)
4747 {
4748 /* In the pre-compile phase, we don't actually do the replication. We
4749 just adjust the length as if we had. For each repetition we must add 1
4750 to the length for BRAZERO and for all but the last repetition we must
4751 add 2 + 2*LINKSIZE to allow for the nesting that occurs. Do some
4752 paranoid checks to avoid integer overflow. The INT64_OR_DOUBLE type is
4753 a 64-bit integer type when available, otherwise double. */
4754
4755 if (lengthptr != NULL && repeat_max > 0)
4756 {
4757 int delta = repeat_max * (length_prevgroup + 1 + 2 + 2*LINK_SIZE) -
4758 2 - 2*LINK_SIZE; /* Last one doesn't nest */
4759 if ((INT64_OR_DOUBLE)repeat_max *
4760 (INT64_OR_DOUBLE)(length_prevgroup + 1 + 2 + 2*LINK_SIZE)
4761 > (INT64_OR_DOUBLE)INT_MAX ||
4762 OFLOW_MAX - *lengthptr < delta)
4763 {
4764 *errorcodeptr = ERR20;
4765 goto FAILED;
4766 }
4767 *lengthptr += delta;
4768 }
4769
4770 /* This is compiling for real */
4771
4772 else for (i = repeat_max - 1; i >= 0; i--)
4773 {
4774 uschar *hc;
4775 uschar *this_hwm = cd->hwm;
4776
4777 *code++ = OP_BRAZERO + repeat_type;
4778
4779 /* All but the final copy start a new nesting, maintaining the
4780 chain of brackets outstanding. */
4781
4782 if (i != 0)
4783 {
4784 int offset;
4785 *code++ = OP_BRA;
4786 offset = (bralink == NULL)? 0 : (int)(code - bralink);
4787 bralink = code;
4788 PUTINC(code, 0, offset);
4789 }
4790
4791 memcpy(code, previous, len);
4792 for (hc = save_hwm; hc < this_hwm; hc += LINK_SIZE)
4793 {
4794 PUT(cd->hwm, 0, GET(hc, 0) + len + ((i != 0)? 2+LINK_SIZE : 1));
4795 cd->hwm += LINK_SIZE;
4796 }
4797 save_hwm = this_hwm;
4798 code += len;
4799 }
4800
4801 /* Now chain through the pending brackets, and fill in their length
4802 fields (which are holding the chain links pro tem). */
4803
4804 while (bralink != NULL)
4805 {
4806 int oldlinkoffset;
4807 int offset = (int)(code - bralink + 1);
4808 uschar *bra = code - offset;
4809 oldlinkoffset = GET(bra, 1);
4810 bralink = (oldlinkoffset == 0)? NULL : bralink - oldlinkoffset;
4811 *code++ = OP_KET;
4812 PUTINC(code, 0, offset);
4813 PUT(bra, 1, offset);
4814 }
4815 }
4816
4817 /* If the maximum is unlimited, set a repeater in the final copy. For
4818 ONCE brackets, that's all we need to do. However, possessively repeated
4819 ONCE brackets can be converted into non-capturing brackets, as the
4820 behaviour of (?:xx)++ is the same as (?>xx)++ and this saves having to
4821 deal with possessive ONCEs specially.
4822
4823 Otherwise, if the quantifier was possessive, we convert the BRA code to
4824 the POS form, and the KET code to KETRPOS. (It turns out to be convenient
4825 at runtime to detect this kind of subpattern at both the start and at the
4826 end.) The use of special opcodes makes it possible to reduce greatly the
4827 stack usage in pcre_exec(). If the group is preceded by OP_BRAZERO,
4828 convert this to OP_BRAPOSZERO. Then cancel the possessive flag so that
4829 the default action below, of wrapping everything inside atomic brackets,
4830 does not happen.
4831
4832 Then, when we are doing the actual compile phase, check to see whether
4833 this group is one that could match an empty string. If so, convert the
4834 initial operator to the S form (e.g. OP_BRA -> OP_SBRA) so that runtime
4835 checking can be done. [This check is also applied to ONCE groups at
4836 runtime, but in a different way.] */
4837
4838 else
4839 {
4840 uschar *ketcode = code - 1 - LINK_SIZE;
4841 uschar *bracode = ketcode - GET(ketcode, 1);
4842
4843 if (*bracode == OP_ONCE && possessive_quantifier) *bracode = OP_BRA;
4844 if (*bracode == OP_ONCE)
4845 *ketcode = OP_KETRMAX + repeat_type;
4846 else
4847 {
4848 if (possessive_quantifier)
4849 {
4850 *bracode += 1; /* Switch to xxxPOS opcodes */
4851 *ketcode = OP_KETRPOS;
4852 if (brazeroptr != NULL) *brazeroptr = OP_BRAPOSZERO;
4853 possessive_quantifier = FALSE;
4854 }
4855 else *ketcode = OP_KETRMAX + repeat_type;
4856
4857 if (lengthptr == NULL)
4858 {
4859 uschar *scode = bracode;
4860 do
4861 {
4862 if (could_be_empty_branch(scode, ketcode, utf8, cd))
4863 {
4864 *bracode += OP_SBRA - OP_BRA;
4865 break;
4866 }
4867 scode += GET(scode, 1);
4868 }
4869 while (*scode == OP_ALT);
4870 }
4871 }
4872 }
4873 }
4874
4875 /* If previous is OP_FAIL, it was generated by an empty class [] in
4876 JavaScript mode. The other ways in which OP_FAIL can be generated, that is
4877 by (*FAIL) or (?!) set previous to NULL, which gives a "nothing to repeat"
4878 error above. We can just ignore the repeat in JS case. */
4879
4880 else if (*previous == OP_FAIL) goto END_REPEAT;
4881
4882 /* Else there's some kind of shambles */
4883
4884 else
4885 {
4886 *errorcodeptr = ERR11;
4887 goto FAILED;
4888 }
4889
4890 /* If the character following a repeat is '+', or if certain optimization
4891 tests above succeeded, possessive_quantifier is TRUE. For some opcodes,
4892 there are special alternative opcodes for this case. For anything else, we
4893 wrap the entire repeated item inside OP_ONCE brackets. Logically, the '+'
4894 notation is just syntactic sugar, taken from Sun's Java package, but the
4895 special opcodes can optimize it.
4896
4897 Possessively repeated subpatterns have already been handled in the code
4898 just above, so possessive_quantifier is always FALSE for them at this
4899 stage.
4900
4901 Note that the repeated item starts at tempcode, not at previous, which
4902 might be the first part of a string whose (former) last char we repeated.
4903
4904 Possessifying an 'exact' quantifier has no effect, so we can ignore it. But
4905 an 'upto' may follow. We skip over an 'exact' item, and then test the
4906 length of what remains before proceeding. */
4907
4908 if (possessive_quantifier)
4909 {
4910 int len;
4911
4912 if (*tempcode == OP_TYPEEXACT)
4913 tempcode += _pcre_OP_lengths[*tempcode] +
4914 ((tempcode[3] == OP_PROP || tempcode[3] == OP_NOTPROP)? 2 : 0);
4915
4916 else if (*tempcode == OP_EXACT || *tempcode == OP_NOTEXACT)
4917 {
4918 tempcode += _pcre_OP_lengths[*tempcode];
4919 #ifdef SUPPORT_UTF8
4920 if (utf8 && tempcode[-1] >= 0xc0)
4921 tempcode += _pcre_utf8_table4[tempcode[-1] & 0x3f];
4922 #endif
4923 }
4924
4925 len = (int)(code - tempcode);
4926 if (len > 0) switch (*tempcode)
4927 {
4928 case OP_STAR: *tempcode = OP_POSSTAR; break;
4929 case OP_PLUS: *tempcode = OP_POSPLUS; break;
4930 case OP_QUERY: *tempcode = OP_POSQUERY; break;
4931 case OP_UPTO: *tempcode = OP_POSUPTO; break;
4932
4933 case OP_STARI: *tempcode = OP_POSSTARI; break;
4934 case OP_PLUSI: *tempcode = OP_POSPLUSI; break;
4935 case OP_QUERYI: *tempcode = OP_POSQUERYI; break;
4936 case OP_UPTOI: *tempcode = OP_POSUPTOI; break;
4937
4938 case OP_NOTSTAR: *tempcode = OP_NOTPOSSTAR; break;
4939 case OP_NOTPLUS: *tempcode = OP_NOTPOSPLUS; break;
4940 case OP_NOTQUERY: *tempcode = OP_NOTPOSQUERY; break;
4941 case OP_NOTUPTO: *tempcode = OP_NOTPOSUPTO; break;
4942
4943 case OP_NOTSTARI: *tempcode = OP_NOTPOSSTARI; break;
4944 case OP_NOTPLUSI: *tempcode = OP_NOTPOSPLUSI; break;
4945 case OP_NOTQUERYI: *tempcode = OP_NOTPOSQUERYI; break;
4946 case OP_NOTUPTOI: *tempcode = OP_NOTPOSUPTOI; break;
4947
4948 case OP_TYPESTAR: *tempcode = OP_TYPEPOSSTAR; break;
4949 case OP_TYPEPLUS: *tempcode = OP_TYPEPOSPLUS; break;
4950 case OP_TYPEQUERY: *tempcode = OP_TYPEPOSQUERY; break;
4951 case OP_TYPEUPTO: *tempcode = OP_TYPEPOSUPTO; break;
4952
4953 /* Because we are moving code along, we must ensure that any
4954 pending recursive references are updated. */
4955
4956 default:
4957 *code = OP_END;
4958 adjust_recurse(tempcode, 1 + LINK_SIZE, utf8, cd, save_hwm);
4959 memmove(tempcode + 1+LINK_SIZE, tempcode, len);
4960 code += 1 + LINK_SIZE;
4961 len += 1 + LINK_SIZE;
4962 tempcode[0] = OP_ONCE;
4963 *code++ = OP_KET;
4964 PUTINC(code, 0, len);
4965 PUT(tempcode, 1, len);
4966 break;
4967 }
4968 }
4969
4970 /* In all case we no longer have a previous item. We also set the
4971 "follows varying string" flag for subsequently encountered reqbytes if
4972 it isn't already set and we have just passed a varying length item. */
4973
4974 END_REPEAT:
4975 previous = NULL;
4976 cd->req_varyopt |= reqvary;
4977 break;
4978
4979
4980 /* ===================================================================*/
4981 /* Start of nested parenthesized sub-expression, or comment or lookahead or
4982 lookbehind or option setting or condition or all the other extended
4983 parenthesis forms. */
4984
4985 case CHAR_LEFT_PARENTHESIS:
4986 newoptions = options;
4987 skipbytes = 0;
4988 bravalue = OP_CBRA;
4989 save_hwm = cd->hwm;
4990 reset_bracount = FALSE;
4991
4992 /* First deal with various "verbs" that can be introduced by '*'. */
4993
4994 if (*(++ptr) == CHAR_ASTERISK &&
4995 ((cd->ctypes[ptr[1]] & ctype_letter) != 0 || ptr[1] == ':'))
4996 {
4997 int i, namelen;
4998 int arglen = 0;
4999 const char *vn = verbnames;
5000 const uschar *name = ptr + 1;
5001 const uschar *arg = NULL;
5002 previous = NULL;
5003 while ((cd->ctypes[*++ptr] & ctype_letter) != 0) {};
5004 namelen = (int)(ptr - name);
5005
5006 /* It appears that Perl allows any characters whatsoever, other than
5007 a closing parenthesis, to appear in arguments, so we no longer insist on
5008 letters, digits, and underscores. */
5009
5010 if (*ptr == CHAR_COLON)
5011 {
5012 arg = ++ptr;
5013 while (*ptr != 0 && *ptr != CHAR_RIGHT_PARENTHESIS) ptr++;
5014 arglen = (int)(ptr - arg);
5015 }
5016
5017 if (*ptr != CHAR_RIGHT_PARENTHESIS)
5018 {
5019 *errorcodeptr = ERR60;
5020 goto FAILED;
5021 }
5022
5023 /* Scan the table of verb names */
5024
5025 for (i = 0; i < verbcount; i++)
5026 {
5027 if (namelen == verbs[i].len &&
5028 strncmp((char *)name, vn, namelen) == 0)
5029 {
5030 /* Check for open captures before ACCEPT and convert it to
5031 ASSERT_ACCEPT if in an assertion. */
5032
5033 if (verbs[i].op == OP_ACCEPT)
5034 {
5035 open_capitem *oc;
5036 if (arglen != 0)
5037 {
5038 *errorcodeptr = ERR59;
5039 goto FAILED;
5040 }
5041 cd->had_accept = TRUE;
5042 for (oc = cd->open_caps; oc != NULL; oc = oc->next)
5043 {
5044 *code++ = OP_CLOSE;
5045 PUT2INC(code, 0, oc->number);
5046 }
5047 *code++ = (cd->assert_depth > 0)? OP_ASSERT_ACCEPT : OP_ACCEPT;
5048 }
5049
5050 /* Handle other cases with/without an argument */
5051
5052 else if (arglen == 0)
5053 {
5054 if (verbs[i].op < 0) /* Argument is mandatory */
5055 {
5056 *errorcodeptr = ERR66;
5057 goto FAILED;
5058 }
5059 *code = verbs[i].op;
5060 if (*code++ == OP_THEN)
5061 {
5062 PUT(code, 0, code - bcptr->current_branch - 1);
5063 code += LINK_SIZE;
5064 }
5065 }
5066
5067 else
5068 {
5069 if (verbs[i].op_arg < 0) /* Argument is forbidden */
5070 {
5071 *errorcodeptr = ERR59;
5072 goto FAILED;
5073 }
5074 *code = verbs[i].op_arg;
5075 if (*code++ == OP_THEN_ARG)
5076 {
5077 PUT(code, 0, code - bcptr->current_branch - 1);
5078 code += LINK_SIZE;
5079 }
5080 *code++ = arglen;
5081 memcpy(code, arg, arglen);
5082 code += arglen;
5083 *code++ = 0;
5084 }
5085
5086 break; /* Found verb, exit loop */
5087 }
5088
5089 vn += verbs[i].len + 1;
5090 }
5091
5092 if (i < verbcount) continue; /* Successfully handled a verb */
5093 *errorcodeptr = ERR60; /* Verb not recognized */
5094 goto FAILED;
5095 }
5096
5097 /* Deal with the extended parentheses; all are introduced by '?', and the
5098 appearance of any of them means that this is not a capturing group. */
5099
5100 else if (*ptr == CHAR_QUESTION_MARK)
5101 {
5102 int i, set, unset, namelen;
5103 int *optset;
5104 const uschar *name;
5105 uschar *slot;
5106
5107 switch (*(++ptr))
5108 {
5109 case CHAR_NUMBER_SIGN: /* Comment; skip to ket */
5110 ptr++;
5111 while (*ptr != 0 && *ptr != CHAR_RIGHT_PARENTHESIS) ptr++;
5112 if (*ptr == 0)
5113 {
5114 *errorcodeptr = ERR18;
5115 goto FAILED;
5116 }
5117 continue;
5118
5119
5120 /* ------------------------------------------------------------ */
5121 case CHAR_VERTICAL_LINE: /* Reset capture count for each branch */
5122 reset_bracount = TRUE;
5123 /* Fall through */
5124
5125 /* ------------------------------------------------------------ */
5126 case CHAR_COLON: /* Non-capturing bracket */
5127 bravalue = OP_BRA;
5128 ptr++;
5129 break;
5130
5131
5132 /* ------------------------------------------------------------ */
5133 case CHAR_LEFT_PARENTHESIS:
5134 bravalue = OP_COND; /* Conditional group */
5135
5136 /* A condition can be an assertion, a number (referring to a numbered
5137 group), a name (referring to a named group), or 'R', referring to
5138 recursion. R<digits> and R&name are also permitted for recursion tests.
5139
5140 There are several syntaxes for testing a named group: (?(name)) is used
5141 by Python; Perl 5.10 onwards uses (?(<name>) or (?('name')).
5142
5143 There are two unfortunate ambiguities, caused by history. (a) 'R' can
5144 be the recursive thing or the name 'R' (and similarly for 'R' followed
5145 by digits), and (b) a number could be a name that consists of digits.
5146 In both cases, we look for a name first; if not found, we try the other
5147 cases. */
5148
5149 /* For conditions that are assertions, check the syntax, and then exit
5150 the switch. This will take control down to where bracketed groups,
5151 including assertions, are processed. */
5152
5153 if (ptr[1] == CHAR_QUESTION_MARK && (ptr[2] == CHAR_EQUALS_SIGN ||
5154 ptr[2] == CHAR_EXCLAMATION_MARK || ptr[2] == CHAR_LESS_THAN_SIGN))
5155 break;
5156
5157 /* Most other conditions use OP_CREF (a couple change to OP_RREF
5158 below), and all need to skip 3 bytes at the start of the group. */
5159
5160 code[1+LINK_SIZE] = OP_CREF;
5161 skipbytes = 3;
5162 refsign = -1;
5163
5164 /* Check for a test for recursion in a named group. */
5165
5166 if (ptr[1] == CHAR_R && ptr[2] == CHAR_AMPERSAND)
5167 {
5168 terminator = -1;
5169 ptr += 2;
5170 code[1+LINK_SIZE] = OP_RREF; /* Change the type of test */
5171 }
5172
5173 /* Check for a test for a named group's having been set, using the Perl
5174 syntax (?(<name>) or (?('name') */
5175
5176 else if (ptr[1] == CHAR_LESS_THAN_SIGN)
5177 {
5178 terminator = CHAR_GREATER_THAN_SIGN;
5179 ptr++;
5180 }
5181 else if (ptr[1] == CHAR_APOSTROPHE)
5182 {
5183 terminator = CHAR_APOSTROPHE;
5184 ptr++;
5185 }
5186 else
5187 {
5188 terminator = 0;
5189 if (ptr[1] == CHAR_MINUS || ptr[1] == CHAR_PLUS) refsign = *(++ptr);
5190 }
5191
5192 /* We now expect to read a name; any thing else is an error */
5193
5194 if ((cd->ctypes[ptr[1]] & ctype_word) == 0)
5195 {
5196 ptr += 1; /* To get the right offset */
5197 *errorcodeptr = ERR28;
5198 goto FAILED;
5199 }
5200
5201 /* Read the name, but also get it as a number if it's all digits */
5202
5203 recno = 0;
5204 name = ++ptr;
5205 while ((cd->ctypes[*ptr] & ctype_word) != 0)
5206 {
5207 if (recno >= 0)
5208 recno = ((digitab[*ptr] & ctype_digit) != 0)?
5209 recno * 10 + *ptr - CHAR_0 : -1;
5210 ptr++;
5211 }
5212 namelen = (int)(ptr - name);
5213
5214 if ((terminator > 0 && *ptr++ != terminator) ||
5215 *ptr++ != CHAR_RIGHT_PARENTHESIS)
5216 {
5217 ptr--; /* Error offset */
5218 *errorcodeptr = ERR26;
5219 goto FAILED;
5220 }
5221
5222 /* Do no further checking in the pre-compile phase. */
5223
5224 if (lengthptr != NULL) break;
5225
5226 /* In the real compile we do the work of looking for the actual
5227 reference. If the string started with "+" or "-" we require the rest to
5228 be digits, in which case recno will be set. */
5229
5230 if (refsign > 0)
5231 {
5232 if (recno <= 0)
5233 {
5234 *errorcodeptr = ERR58;
5235 goto FAILED;
5236 }
5237 recno = (refsign == CHAR_MINUS)?
5238 cd->bracount - recno + 1 : recno +cd->bracount;
5239 if (recno <= 0 || recno > cd->final_bracount)
5240 {
5241 *errorcodeptr = ERR15;
5242 goto FAILED;
5243 }
5244 PUT2(code, 2+LINK_SIZE, recno);
5245 break;
5246 }
5247
5248 /* Otherwise (did not start with "+" or "-"), start by looking for the
5249 name. If we find a name, add one to the opcode to change OP_CREF or
5250 OP_RREF into OP_NCREF or OP_NRREF. These behave exactly the same,
5251 except they record that the reference was originally to a name. The
5252 information is used to check duplicate names. */
5253
5254 slot = cd->name_table;
5255 for (i = 0; i < cd->names_found; i++)
5256 {
5257 if (strncmp((char *)name, (char *)slot+2, namelen) == 0) break;
5258 slot += cd->name_entry_size;
5259 }
5260
5261 /* Found a previous named subpattern */
5262
5263 if (i < cd->names_found)
5264 {
5265 recno = GET2(slot, 0);
5266 PUT2(code, 2+LINK_SIZE, recno);
5267 code[1+LINK_SIZE]++;
5268 }
5269
5270 /* Search the pattern for a forward reference */
5271
5272 else if ((i = find_parens(cd, name, namelen,
5273 (options & PCRE_EXTENDED) != 0, utf8)) > 0)
5274 {
5275 PUT2(code, 2+LINK_SIZE, i);
5276 code[1+LINK_SIZE]++;
5277 }
5278
5279 /* If terminator == 0 it means that the name followed directly after
5280 the opening parenthesis [e.g. (?(abc)...] and in this case there are
5281 some further alternatives to try. For the cases where terminator != 0
5282 [things like (?(<name>... or (?('name')... or (?(R&name)... ] we have
5283 now checked all the possibilities, so give an error. */
5284
5285 else if (terminator != 0)
5286 {
5287 *errorcodeptr = ERR15;
5288 goto FAILED;
5289 }
5290
5291 /* Check for (?(R) for recursion. Allow digits after R to specify a
5292 specific group number. */
5293
5294 else if (*name == CHAR_R)
5295 {
5296 recno = 0;
5297 for (i = 1; i < namelen; i++)
5298 {
5299 if ((digitab[name[i]] & ctype_digit) == 0)
5300 {
5301 *errorcodeptr = ERR15;
5302 goto FAILED;
5303 }
5304 recno = recno * 10 + name[i] - CHAR_0;
5305 }
5306 if (recno == 0) recno = RREF_ANY;
5307 code[1+LINK_SIZE] = OP_RREF; /* Change test type */
5308 PUT2(code, 2+LINK_SIZE, recno);
5309 }
5310
5311 /* Similarly, check for the (?(DEFINE) "condition", which is always
5312 false. */
5313
5314 else if (namelen == 6 && strncmp((char *)name, STRING_DEFINE, 6) == 0)
5315 {
5316 code[1+LINK_SIZE] = OP_DEF;
5317 skipbytes = 1;
5318 }
5319
5320 /* Check for the "name" actually being a subpattern number. We are
5321 in the second pass here, so final_bracount is set. */
5322
5323 else if (recno > 0 && recno <= cd->final_bracount)
5324 {
5325 PUT2(code, 2+LINK_SIZE, recno);
5326 }
5327
5328 /* Either an unidentified subpattern, or a reference to (?(0) */
5329
5330 else
5331 {
5332 *errorcodeptr = (recno == 0)? ERR35: ERR15;
5333 goto FAILED;
5334 }
5335 break;
5336
5337
5338 /* ------------------------------------------------------------ */
5339 case CHAR_EQUALS_SIGN: /* Positive lookahead */
5340 bravalue = OP_ASSERT;
5341 cd->assert_depth += 1;
5342 ptr++;
5343 break;
5344
5345
5346 /* ------------------------------------------------------------ */
5347 case CHAR_EXCLAMATION_MARK: /* Negative lookahead */
5348 ptr++;
5349 if (*ptr == CHAR_RIGHT_PARENTHESIS) /* Optimize (?!) */
5350 {
5351 *code++ = OP_FAIL;
5352 previous = NULL;
5353 continue;
5354 }
5355 bravalue = OP_ASSERT_NOT;
5356 cd->assert_depth += 1;
5357 break;
5358
5359
5360 /* ------------------------------------------------------------ */
5361 case CHAR_LESS_THAN_SIGN: /* Lookbehind or named define */
5362 switch (ptr[1])
5363 {
5364 case CHAR_EQUALS_SIGN: /* Positive lookbehind */
5365 bravalue = OP_ASSERTBACK;
5366 cd->assert_depth += 1;
5367 ptr += 2;
5368 break;
5369
5370 case CHAR_EXCLAMATION_MARK: /* Negative lookbehind */
5371 bravalue = OP_ASSERTBACK_NOT;
5372 cd->assert_depth += 1;
5373 ptr += 2;
5374 break;
5375
5376 default: /* Could be name define, else bad */
5377 if ((cd->ctypes[ptr[1]] & ctype_word) != 0) goto DEFINE_NAME;
5378 ptr++; /* Correct offset for error */
5379 *errorcodeptr = ERR24;
5380 goto FAILED;
5381 }
5382 break;
5383
5384
5385 /* ------------------------------------------------------------ */
5386 case CHAR_GREATER_THAN_SIGN: /* One-time brackets */
5387 bravalue = OP_ONCE;
5388 ptr++;
5389 break;
5390
5391
5392 /* ------------------------------------------------------------ */
5393 case CHAR_C: /* Callout - may be followed by digits; */
5394 previous_callout = code; /* Save for later completion */
5395 after_manual_callout = 1; /* Skip one item before completing */
5396 *code++ = OP_CALLOUT;
5397 {
5398 int n = 0;
5399 while ((digitab[*(++ptr)] & ctype_digit) != 0)
5400 n = n * 10 + *ptr - CHAR_0;
5401 if (*ptr != CHAR_RIGHT_PARENTHESIS)
5402 {
5403 *errorcodeptr = ERR39;
5404 goto FAILED;
5405 }
5406 if (n > 255)
5407 {
5408 *errorcodeptr = ERR38;
5409 goto FAILED;
5410 }
5411 *code++ = n;
5412 PUT(code, 0, (int)(ptr - cd->start_pattern + 1)); /* Pattern offset */
5413 PUT(code, LINK_SIZE, 0); /* Default length */
5414 code += 2 * LINK_SIZE;
5415 }
5416 previous = NULL;
5417 continue;
5418
5419
5420 /* ------------------------------------------------------------ */
5421 case CHAR_P: /* Python-style named subpattern handling */
5422 if (*(++ptr) == CHAR_EQUALS_SIGN ||
5423 *ptr == CHAR_GREATER_THAN_SIGN) /* Reference or recursion */
5424 {
5425 is_recurse = *ptr == CHAR_GREATER_THAN_SIGN;
5426 terminator = CHAR_RIGHT_PARENTHESIS;
5427 goto NAMED_REF_OR_RECURSE;
5428 }
5429 else if (*ptr != CHAR_LESS_THAN_SIGN) /* Test for Python-style defn */
5430 {
5431 *errorcodeptr = ERR41;
5432 goto FAILED;
5433 }
5434 /* Fall through to handle (?P< as (?< is handled */
5435
5436
5437 /* ------------------------------------------------------------ */
5438 DEFINE_NAME: /* Come here from (?< handling */
5439 case CHAR_APOSTROPHE:
5440 {
5441 terminator = (*ptr == CHAR_LESS_THAN_SIGN)?
5442 CHAR_GREATER_THAN_SIGN : CHAR_APOSTROPHE;
5443 name = ++ptr;
5444
5445 while ((cd->ctypes[*ptr] & ctype_word) != 0) ptr++;
5446 namelen = (int)(ptr - name);
5447
5448 /* In the pre-compile phase, just do a syntax check. */
5449
5450 if (lengthptr != NULL)
5451 {
5452 if (*ptr != terminator)
5453 {
5454 *errorcodeptr = ERR42;
5455 goto FAILED;
5456 }
5457 if (cd->names_found >= MAX_NAME_COUNT)
5458 {
5459 *errorcodeptr = ERR49;
5460 goto FAILED;
5461 }
5462 if (namelen + 3 > cd->name_entry_size)
5463 {
5464 cd->name_entry_size = namelen + 3;
5465 if (namelen > MAX_NAME_SIZE)
5466 {
5467 *errorcodeptr = ERR48;
5468 goto FAILED;
5469 }
5470 }
5471 }
5472
5473 /* In the real compile, create the entry in the table, maintaining
5474 alphabetical order. Duplicate names for different numbers are
5475 permitted only if PCRE_DUPNAMES is set. Duplicate names for the same
5476 number are always OK. (An existing number can be re-used if (?|
5477 appears in the pattern.) In either event, a duplicate name results in
5478 a duplicate entry in the table, even if the number is the same. This
5479 is because the number of names, and hence the table size, is computed
5480 in the pre-compile, and it affects various numbers and pointers which
5481 would all have to be modified, and the compiled code moved down, if
5482 duplicates with the same number were omitted from the table. This
5483 doesn't seem worth the hassle. However, *different* names for the
5484 same number are not permitted. */
5485
5486 else
5487 {
5488 BOOL dupname = FALSE;
5489 slot = cd->name_table;
5490
5491 for (i = 0; i < cd->names_found; i++)
5492 {
5493 int crc = memcmp(name, slot+2, namelen);
5494 if (crc == 0)
5495 {
5496 if (slot[2+namelen] == 0)
5497 {
5498 if (GET2(slot, 0) != cd->bracount + 1 &&
5499 (options & PCRE_DUPNAMES) == 0)
5500 {
5501 *errorcodeptr = ERR43;
5502 goto FAILED;
5503 }
5504 else dupname = TRUE;
5505 }
5506 else crc = -1; /* Current name is a substring */
5507 }
5508
5509 /* Make space in the table and break the loop for an earlier
5510 name. For a duplicate or later name, carry on. We do this for
5511 duplicates so that in the simple case (when ?(| is not used) they
5512 are in order of their numbers. */
5513
5514 if (crc < 0)
5515 {
5516 memmove(slot + cd->name_entry_size, slot,
5517 (cd->names_found - i) * cd->name_entry_size);
5518 break;
5519 }
5520
5521 /* Continue the loop for a later or duplicate name */
5522
5523 slot += cd->name_entry_size;
5524 }
5525
5526 /* For non-duplicate names, check for a duplicate number before
5527 adding the new name. */
5528
5529 if (!dupname)
5530 {
5531 uschar *cslot = cd->name_table;
5532 for (i = 0; i < cd->names_found; i++)
5533 {
5534 if (cslot != slot)
5535 {
5536 if (GET2(cslot, 0) == cd->bracount + 1)
5537 {
5538 *errorcodeptr = ERR65;
5539 goto FAILED;
5540 }
5541 }
5542 else i--;
5543 cslot += cd->name_entry_size;
5544 }
5545 }
5546
5547 PUT2(slot, 0, cd->bracount + 1);
5548 memcpy(slot + 2, name, namelen);
5549 slot[2+namelen] = 0;
5550 }
5551 }
5552
5553 /* In both pre-compile and compile, count the number of names we've
5554 encountered. */
5555
5556 cd->names_found++;
5557 ptr++; /* Move past > or ' */
5558 goto NUMBERED_GROUP;
5559
5560
5561 /* ------------------------------------------------------------ */
5562 case CHAR_AMPERSAND: /* Perl recursion/subroutine syntax */
5563 terminator = CHAR_RIGHT_PARENTHESIS;
5564 is_recurse = TRUE;
5565 /* Fall through */
5566
5567 /* We come here from the Python syntax above that handles both
5568 references (?P=name) and recursion (?P>name), as well as falling
5569 through from the Perl recursion syntax (?&name). We also come here from
5570 the Perl \k<name> or \k'name' back reference syntax and the \k{name}
5571 .NET syntax, and the Oniguruma \g<...> and \g'...' subroutine syntax. */
5572
5573 NAMED_REF_OR_RECURSE:
5574 name = ++ptr;
5575 while ((cd->ctypes[*ptr] & ctype_word) != 0) ptr++;
5576 namelen = (int)(ptr - name);
5577
5578 /* In the pre-compile phase, do a syntax check. We used to just set
5579 a dummy reference number, because it was not used in the first pass.
5580 However, with the change of recursive back references to be atomic,
5581 we have to look for the number so that this state can be identified, as
5582 otherwise the incorrect length is computed. If it's not a backwards
5583 reference, the dummy number will do. */
5584
5585 if (lengthptr != NULL)
5586 {
5587 const uschar *temp;
5588
5589 if (namelen == 0)
5590 {
5591 *errorcodeptr = ERR62;
5592 goto FAILED;
5593 }
5594 if (*ptr != terminator)
5595 {
5596 *errorcodeptr = ERR42;
5597 goto FAILED;
5598 }
5599 if (namelen > MAX_NAME_SIZE)
5600 {
5601 *errorcodeptr = ERR48;
5602 goto FAILED;
5603 }
5604
5605 /* The name table does not exist in the first pass, so we cannot
5606 do a simple search as in the code below. Instead, we have to scan the
5607 pattern to find the number. It is important that we scan it only as
5608 far as we have got because the syntax of named subpatterns has not
5609 been checked for the rest of the pattern, and find_parens() assumes
5610 correct syntax. In any case, it's a waste of resources to scan
5611 further. We stop the scan at the current point by temporarily
5612 adjusting the value of cd->endpattern. */
5613
5614 temp = cd->end_pattern;
5615 cd->end_pattern = ptr;
5616 recno = find_parens(cd, name, namelen,
5617 (options & PCRE_EXTENDED) != 0, utf8);
5618 cd->end_pattern = temp;
5619 if (recno < 0) recno = 0; /* Forward ref; set dummy number */
5620 }
5621
5622 /* In the real compile, seek the name in the table. We check the name
5623 first, and then check that we have reached the end of the name in the
5624 table. That way, if the name that is longer than any in the table,
5625 the comparison will fail without reading beyond the table entry. */
5626
5627 else
5628 {
5629 slot = cd->name_table;
5630 for (i = 0; i < cd->names_found; i++)
5631 {
5632 if (strncmp((char *)name, (char *)slot+2, namelen) == 0 &&
5633 slot[2+namelen] == 0)
5634 break;
5635 slot += cd->name_entry_size;
5636 }
5637
5638 if (i < cd->names_found) /* Back reference */
5639 {
5640 recno = GET2(slot, 0);
5641 }
5642 else if ((recno = /* Forward back reference */
5643 find_parens(cd, name, namelen,
5644 (options & PCRE_EXTENDED) != 0, utf8)) <= 0)
5645 {
5646 *errorcodeptr = ERR15;
5647 goto FAILED;
5648 }
5649 }
5650
5651 /* In both phases, we can now go to the code than handles numerical
5652 recursion or backreferences. */
5653
5654 if (is_recurse) goto HANDLE_RECURSION;
5655 else goto HANDLE_REFERENCE;
5656
5657
5658 /* ------------------------------------------------------------ */
5659 case CHAR_R: /* Recursion */
5660 ptr++; /* Same as (?0) */
5661 /* Fall through */
5662
5663
5664 /* ------------------------------------------------------------ */
5665 case CHAR_MINUS: case CHAR_PLUS: /* Recursion or subroutine */
5666 case CHAR_0: case CHAR_1: case CHAR_2: case CHAR_3: case CHAR_4:
5667 case CHAR_5: case CHAR_6: case CHAR_7: case CHAR_8: case CHAR_9:
5668 {
5669 const uschar *called;
5670 terminator = CHAR_RIGHT_PARENTHESIS;
5671
5672 /* Come here from the \g<...> and \g'...' code (Oniguruma
5673 compatibility). However, the syntax has been checked to ensure that
5674 the ... are a (signed) number, so that neither ERR63 nor ERR29 will
5675 be called on this path, nor with the jump to OTHER_CHAR_AFTER_QUERY
5676 ever be taken. */
5677
5678 HANDLE_NUMERICAL_RECURSION:
5679
5680 if ((refsign = *ptr) == CHAR_PLUS)
5681 {
5682 ptr++;
5683 if ((digitab[*ptr] & ctype_digit) == 0)
5684 {
5685 *errorcodeptr = ERR63;
5686 goto FAILED;
5687 }
5688 }
5689 else if (refsign == CHAR_MINUS)
5690 {
5691 if ((digitab[ptr[1]] & ctype_digit) == 0)
5692 goto OTHER_CHAR_AFTER_QUERY;
5693 ptr++;
5694 }
5695
5696 recno = 0;
5697 while((digitab[*ptr] & ctype_digit) != 0)
5698 recno = recno * 10 + *ptr++ - CHAR_0;
5699
5700 if (*ptr != terminator)
5701 {
5702 *errorcodeptr = ERR29;
5703 goto FAILED;
5704 }
5705
5706 if (refsign == CHAR_MINUS)
5707 {
5708 if (recno == 0)
5709 {
5710 *errorcodeptr = ERR58;
5711 goto FAILED;
5712 }
5713 recno = cd->bracount - recno + 1;
5714 if (recno <= 0)
5715 {
5716 *errorcodeptr = ERR15;
5717 goto FAILED;
5718 }
5719 }
5720 else if (refsign == CHAR_PLUS)
5721 {
5722 if (recno == 0)
5723 {
5724 *errorcodeptr = ERR58;
5725 goto FAILED;
5726 }
5727 recno += cd->bracount;
5728 }
5729
5730 /* Come here from code above that handles a named recursion */
5731
5732 HANDLE_RECURSION:
5733
5734 previous = code;
5735 called = cd->start_code;
5736
5737 /* When we are actually compiling, find the bracket that is being
5738 referenced. Temporarily end the regex in case it doesn't exist before
5739 this point. If we end up with a forward reference, first check that
5740 the bracket does occur later so we can give the error (and position)
5741 now. Then remember this forward reference in the workspace so it can
5742 be filled in at the end. */
5743
5744 if (lengthptr == NULL)
5745 {
5746 *code = OP_END;
5747 if (recno != 0)
5748 called = _pcre_find_bracket(cd->start_code, utf8, recno);
5749
5750 /* Forward reference */
5751
5752 if (called == NULL)
5753 {
5754 if (find_parens(cd, NULL, recno,
5755 (options & PCRE_EXTENDED) != 0, utf8) < 0)
5756 {
5757 *errorcodeptr = ERR15;
5758 goto FAILED;
5759 }
5760
5761 /* Fudge the value of "called" so that when it is inserted as an
5762 offset below, what it actually inserted is the reference number
5763 of the group. Then remember the forward reference. */
5764
5765 called = cd->start_code + recno;
5766 PUTINC(cd->hwm, 0, (int)(code + 1 - cd->start_code));
5767 }
5768
5769 /* If not a forward reference, and the subpattern is still open,
5770 this is a recursive call. We check to see if this is a left
5771 recursion that could loop for ever, and diagnose that case. We
5772 must not, however, do this check if we are in a conditional
5773 subpattern because the condition might be testing for recursion in
5774 a pattern such as /(?(R)a+|(?R)b)/, which is perfectly valid.
5775 Forever loops are also detected at runtime, so those that occur in
5776 conditional subpatterns will be picked up then. */
5777
5778 else if (GET(called, 1) == 0 && cond_depth <= 0 &&
5779 could_be_empty(called, code, bcptr, utf8, cd))
5780 {
5781 *errorcodeptr = ERR40;
5782 goto FAILED;
5783 }
5784 }
5785
5786 /* Insert the recursion/subroutine item. */
5787
5788 *code = OP_RECURSE;
5789 PUT(code, 1, (int)(called - cd->start_code));
5790 code += 1 + LINK_SIZE;
5791 }
5792
5793 /* Can't determine a first byte now */
5794
5795 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
5796 continue;
5797
5798
5799 /* ------------------------------------------------------------ */
5800 default: /* Other characters: check option setting */
5801 OTHER_CHAR_AFTER_QUERY:
5802 set = unset = 0;
5803 optset = &set;
5804
5805 while (*ptr != CHAR_RIGHT_PARENTHESIS && *ptr != CHAR_COLON)
5806 {
5807 switch (*ptr++)
5808 {
5809 case CHAR_MINUS: optset = &unset; break;
5810
5811 case CHAR_J: /* Record that it changed in the external options */
5812 *optset |= PCRE_DUPNAMES;
5813 cd->external_flags |= PCRE_JCHANGED;
5814 break;
5815
5816 case CHAR_i: *optset |= PCRE_CASELESS; break;
5817 case CHAR_m: *optset |= PCRE_MULTILINE; break;
5818 case CHAR_s: *optset |= PCRE_DOTALL; break;
5819 case CHAR_x: *optset |= PCRE_EXTENDED; break;
5820 case CHAR_U: *optset |= PCRE_UNGREEDY; break;
5821 case CHAR_X: *optset |= PCRE_EXTRA; break;
5822
5823 default: *errorcodeptr = ERR12;
5824 ptr--; /* Correct the offset */
5825 goto FAILED;
5826 }
5827 }
5828
5829 /* Set up the changed option bits, but don't change anything yet. */
5830
5831 newoptions = (options | set) & (~unset);
5832
5833 /* If the options ended with ')' this is not the start of a nested
5834 group with option changes, so the options change at this level. If this
5835 item is right at the start of the pattern, the options can be
5836 abstracted and made external in the pre-compile phase, and ignored in
5837 the compile phase. This can be helpful when matching -- for instance in
5838 caseless checking of required bytes.
5839
5840 If the code pointer is not (cd->start_code + 1 + LINK_SIZE), we are
5841 definitely *not* at the start of the pattern because something has been
5842 compiled. In the pre-compile phase, however, the code pointer can have
5843 that value after the start, because it gets reset as code is discarded
5844 during the pre-compile. However, this can happen only at top level - if
5845 we are within parentheses, the starting BRA will still be present. At
5846 any parenthesis level, the length value can be used to test if anything
5847 has been compiled at that level. Thus, a test for both these conditions
5848 is necessary to ensure we correctly detect the start of the pattern in
5849 both phases.
5850
5851 If we are not at the pattern start, reset the greedy defaults and the
5852 case value for firstbyte and reqbyte. */
5853
5854 if (*ptr == CHAR_RIGHT_PARENTHESIS)
5855 {
5856 if (code == cd->start_code + 1 + LINK_SIZE &&
5857 (lengthptr == NULL || *lengthptr == 2 + 2*LINK_SIZE))
5858 {
5859 cd->external_options = newoptions;
5860 }
5861 else
5862 {
5863 greedy_default = ((newoptions & PCRE_UNGREEDY) != 0);
5864 greedy_non_default = greedy_default ^ 1;
5865 req_caseopt = ((newoptions & PCRE_CASELESS) != 0)? REQ_CASELESS : 0;
5866 }
5867
5868 /* Change options at this level, and pass them back for use
5869 in subsequent branches. */
5870
5871 *optionsptr = options = newoptions;
5872 previous = NULL; /* This item can't be repeated */
5873 continue; /* It is complete */
5874 }
5875
5876 /* If the options ended with ':' we are heading into a nested group
5877 with possible change of options. Such groups are non-capturing and are
5878 not assertions of any kind. All we need to do is skip over the ':';
5879 the newoptions value is handled below. */
5880
5881 bravalue = OP_BRA;
5882 ptr++;
5883 } /* End of switch for character following (? */
5884 } /* End of (? handling */
5885
5886 /* Opening parenthesis not followed by '*' or '?'. If PCRE_NO_AUTO_CAPTURE
5887 is set, all unadorned brackets become non-capturing and behave like (?:...)
5888 brackets. */
5889
5890 else if ((options & PCRE_NO_AUTO_CAPTURE) != 0)
5891 {
5892 bravalue = OP_BRA;
5893 }
5894
5895 /* Else we have a capturing group. */
5896
5897 else
5898 {
5899 NUMBERED_GROUP:
5900 cd->bracount += 1;
5901 PUT2(code, 1+LINK_SIZE, cd->bracount);
5902 skipbytes = 2;
5903 }
5904
5905 /* Process nested bracketed regex. Assertions used not to be repeatable,
5906 but this was changed for Perl compatibility, so all kinds can now be
5907 repeated. We copy code into a non-register variable (tempcode) in order to
5908 be able to pass its address because some compilers complain otherwise. */
5909
5910 previous = code; /* For handling repetition */
5911 *code = bravalue;
5912 tempcode = code;
5913 tempreqvary = cd->req_varyopt; /* Save value before bracket */
5914 length_prevgroup = 0; /* Initialize for pre-compile phase */
5915
5916 if (!compile_regex(
5917 newoptions, /* The complete new option state */
5918 &tempcode, /* Where to put code (updated) */
5919 &ptr, /* Input pointer (updated) */
5920 errorcodeptr, /* Where to put an error message */
5921 (bravalue == OP_ASSERTBACK ||
5922 bravalue == OP_ASSERTBACK_NOT), /* TRUE if back assert */
5923 reset_bracount, /* True if (?| group */
5924 skipbytes, /* Skip over bracket number */
5925 cond_depth +
5926 ((bravalue == OP_COND)?1:0), /* Depth of condition subpatterns */
5927 &subfirstbyte, /* For possible first char */
5928 &subreqbyte, /* For possible last char */
5929 bcptr, /* Current branch chain */
5930 cd, /* Tables block */
5931 (lengthptr == NULL)? NULL : /* Actual compile phase */
5932 &length_prevgroup /* Pre-compile phase */
5933 ))
5934 goto FAILED;
5935
5936 if (bravalue >= OP_ASSERT && bravalue <= OP_ASSERTBACK_NOT)
5937 cd->assert_depth -= 1;
5938
5939 /* At the end of compiling, code is still pointing to the start of the
5940 group, while tempcode has been updated to point past the end of the group
5941 and any option resetting that may follow it. The pattern pointer (ptr)
5942 is on the bracket. */
5943
5944 /* If this is a conditional bracket, check that there are no more than
5945 two branches in the group, or just one if it's a DEFINE group. We do this
5946 in the real compile phase, not in the pre-pass, where the whole group may
5947 not be available. */
5948
5949 if (bravalue == OP_COND && lengthptr == NULL)
5950 {
5951 uschar *tc = code;
5952 int condcount = 0;
5953
5954 do {
5955 condcount++;
5956 tc += GET(tc,1);
5957 }
5958 while (*tc != OP_KET);
5959
5960 /* A DEFINE group is never obeyed inline (the "condition" is always
5961 false). It must have only one branch. */
5962
5963 if (code[LINK_SIZE+1] == OP_DEF)
5964 {
5965 if (condcount > 1)
5966 {
5967 *errorcodeptr = ERR54;
5968 goto FAILED;
5969 }
5970 bravalue = OP_DEF; /* Just a flag to suppress char handling below */
5971 }
5972
5973 /* A "normal" conditional group. If there is just one branch, we must not
5974 make use of its firstbyte or reqbyte, because this is equivalent to an
5975 empty second branch. */
5976
5977 else
5978 {
5979 if (condcount > 2)
5980 {
5981 *errorcodeptr = ERR27;
5982 goto FAILED;
5983 }
5984 if (condcount == 1) subfirstbyte = subreqbyte = REQ_NONE;
5985 }
5986 }
5987
5988 /* Error if hit end of pattern */
5989
5990 if (*ptr != CHAR_RIGHT_PARENTHESIS)
5991 {
5992 *errorcodeptr = ERR14;
5993 goto FAILED;
5994 }
5995
5996 /* In the pre-compile phase, update the length by the length of the group,
5997 less the brackets at either end. Then reduce the compiled code to just a
5998 set of non-capturing brackets so that it doesn't use much memory if it is
5999 duplicated by a quantifier.*/
6000
6001 if (lengthptr != NULL)
6002 {
6003 if (OFLOW_MAX - *lengthptr < length_prevgroup - 2 - 2*LINK_SIZE)
6004 {
6005 *errorcodeptr = ERR20;
6006 goto FAILED;
6007 }
6008 *lengthptr += length_prevgroup - 2 - 2*LINK_SIZE;
6009 code++; /* This already contains bravalue */
6010 PUTINC(code, 0, 1 + LINK_SIZE);
6011 *code++ = OP_KET;
6012 PUTINC(code, 0, 1 + LINK_SIZE);
6013 break; /* No need to waste time with special character handling */
6014 }
6015
6016 /* Otherwise update the main code pointer to the end of the group. */
6017
6018 code = tempcode;
6019
6020 /* For a DEFINE group, required and first character settings are not
6021 relevant. */
6022
6023 if (bravalue == OP_DEF) break;
6024
6025 /* Handle updating of the required and first characters for other types of
6026 group. Update for normal brackets of all kinds, and conditions with two
6027 branches (see code above). If the bracket is followed by a quantifier with
6028 zero repeat, we have to back off. Hence the definition of zeroreqbyte and
6029 zerofirstbyte outside the main loop so that they can be accessed for the
6030 back off. */
6031
6032 zeroreqbyte = reqbyte;
6033 zerofirstbyte = firstbyte;
6034 groupsetfirstbyte = FALSE;
6035
6036 if (bravalue >= OP_ONCE)
6037 {
6038 /* If we have not yet set a firstbyte in this branch, take it from the
6039 subpattern, remembering that it was set here so that a repeat of more
6040 than one can replicate it as reqbyte if necessary. If the subpattern has
6041 no firstbyte, set "none" for the whole branch. In both cases, a zero
6042 repeat forces firstbyte to "none". */
6043
6044 if (firstbyte == REQ_UNSET)
6045 {
6046 if (subfirstbyte >= 0)
6047 {
6048 firstbyte = subfirstbyte;
6049 groupsetfirstbyte = TRUE;
6050 }
6051 else firstbyte = REQ_NONE;
6052 zerofirstbyte = REQ_NONE;
6053 }
6054
6055 /* If firstbyte was previously set, convert the subpattern's firstbyte
6056 into reqbyte if there wasn't one, using the vary flag that was in
6057 existence beforehand. */
6058
6059 else if (subfirstbyte >= 0 && subreqbyte < 0)
6060 subreqbyte = subfirstbyte | tempreqvary;
6061
6062 /* If the subpattern set a required byte (or set a first byte that isn't
6063 really the first byte - see above), set it. */
6064
6065 if (subreqbyte >= 0) reqbyte = subreqbyte;
6066 }
6067
6068 /* For a forward assertion, we take the reqbyte, if set. This can be
6069 helpful if the pattern that follows the assertion doesn't set a different
6070 char. For example, it's useful for /(?=abcde).+/. We can't set firstbyte
6071 for an assertion, however because it leads to incorrect effect for patterns
6072 such as /(?=a)a.+/ when the "real" "a" would then become a reqbyte instead
6073 of a firstbyte. This is overcome by a scan at the end if there's no
6074 firstbyte, looking for an asserted first char. */
6075
6076 else if (bravalue == OP_ASSERT && subreqbyte >= 0) reqbyte = subreqbyte;
6077 break; /* End of processing '(' */
6078
6079
6080 /* ===================================================================*/
6081 /* Handle metasequences introduced by \. For ones like \d, the ESC_ values
6082 are arranged to be the negation of the corresponding OP_values in the
6083 default case when PCRE_UCP is not set. For the back references, the values
6084 are ESC_REF plus the reference number. Only back references and those types
6085 that consume a character may be repeated. We can test for values between
6086 ESC_b and ESC_Z for the latter; this may have to change if any new ones are
6087 ever created. */
6088
6089 case CHAR_BACKSLASH:
6090 tempptr = ptr;
6091 c = check_escape(&ptr, errorcodeptr, cd->bracount, options, FALSE);
6092 if (*errorcodeptr != 0) goto FAILED;
6093
6094 if (c < 0)
6095 {
6096 if (-c == ESC_Q) /* Handle start of quoted string */
6097 {
6098 if (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
6099 ptr += 2; /* avoid empty string */
6100 else inescq = TRUE;
6101 continue;
6102 }
6103
6104 if (-c == ESC_E) continue; /* Perl ignores an orphan \E */
6105
6106 /* For metasequences that actually match a character, we disable the
6107 setting of a first character if it hasn't already been set. */
6108
6109 if (firstbyte == REQ_UNSET && -c > ESC_b && -c < ESC_Z)
6110 firstbyte = REQ_NONE;
6111
6112 /* Set values to reset to if this is followed by a zero repeat. */
6113
6114 zerofirstbyte = firstbyte;
6115 zeroreqbyte = reqbyte;
6116
6117 /* \g<name> or \g'name' is a subroutine call by name and \g<n> or \g'n'
6118 is a subroutine call by number (Oniguruma syntax). In fact, the value
6119 -ESC_g is returned only for these cases. So we don't need to check for <
6120 or ' if the value is -ESC_g. For the Perl syntax \g{n} the value is
6121 -ESC_REF+n, and for the Perl syntax \g{name} the result is -ESC_k (as
6122 that is a synonym for a named back reference). */
6123
6124 if (-c == ESC_g)
6125 {
6126 const uschar *p;
6127 save_hwm = cd->hwm; /* Normally this is set when '(' is read */
6128 terminator = (*(++ptr) == CHAR_LESS_THAN_SIGN)?
6129 CHAR_GREATER_THAN_SIGN : CHAR_APOSTROPHE;
6130
6131 /* These two statements stop the compiler for warning about possibly
6132 unset variables caused by the jump to HANDLE_NUMERICAL_RECURSION. In
6133 fact, because we actually check for a number below, the paths that
6134 would actually be in error are never taken. */
6135
6136 skipbytes = 0;
6137 reset_bracount = FALSE;
6138
6139 /* Test for a name */
6140
6141 if (ptr[1] != CHAR_PLUS && ptr[1] != CHAR_MINUS)
6142 {
6143 BOOL isnumber = TRUE;
6144 for (p = ptr + 1; *p != 0 && *p != terminator; p++)
6145 {
6146 if ((cd->ctypes[*p] & ctype_digit) == 0) isnumber = FALSE;
6147 if ((cd->ctypes[*p] & ctype_word) == 0) break;
6148 }
6149 if (*p != terminator)
6150 {
6151 *errorcodeptr = ERR57;
6152 break;
6153 }
6154 if (isnumber)
6155 {
6156 ptr++;
6157 goto HANDLE_NUMERICAL_RECURSION;
6158 }
6159 is_recurse = TRUE;
6160 goto NAMED_REF_OR_RECURSE;
6161 }
6162
6163 /* Test a signed number in angle brackets or quotes. */
6164
6165 p = ptr + 2;
6166 while ((digitab[*p] & ctype_digit) != 0) p++;
6167 if (*p != terminator)
6168 {
6169 *errorcodeptr = ERR57;
6170 break;
6171 }
6172 ptr++;
6173 goto HANDLE_NUMERICAL_RECURSION;
6174 }
6175
6176 /* \k<name> or \k'name' is a back reference by name (Perl syntax).
6177 We also support \k{name} (.NET syntax). */
6178
6179 if (-c == ESC_k)
6180 {
6181 if ((ptr[1] != CHAR_LESS_THAN_SIGN &&
6182 ptr[1] != CHAR_APOSTROPHE && ptr[1] != CHAR_LEFT_CURLY_BRACKET))
6183 {
6184 *errorcodeptr = ERR69;
6185 break;
6186 }
6187 is_recurse = FALSE;
6188 terminator = (*(++ptr) == CHAR_LESS_THAN_SIGN)?
6189 CHAR_GREATER_THAN_SIGN : (*ptr == CHAR_APOSTROPHE)?
6190 CHAR_APOSTROPHE : CHAR_RIGHT_CURLY_BRACKET;
6191 goto NAMED_REF_OR_RECURSE;
6192 }
6193
6194 /* Back references are handled specially; must disable firstbyte if
6195 not set to cope with cases like (?=(\w+))\1: which would otherwise set
6196 ':' later. */
6197
6198 if (-c >= ESC_REF)
6199 {
6200 open_capitem *oc;
6201 recno = -c - ESC_REF;
6202
6203 HANDLE_REFERENCE: /* Come here from named backref handling */
6204 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
6205 previous = code;
6206 *code++ = ((options & PCRE_CASELESS) != 0)? OP_REFI : OP_REF;
6207 PUT2INC(code, 0, recno);
6208 cd->backref_map |= (recno < 32)? (1 << recno) : 1;
6209 if (recno > cd->top_backref) cd->top_backref = recno;
6210
6211 /* Check to see if this back reference is recursive, that it, it
6212 is inside the group that it references. A flag is set so that the
6213 group can be made atomic. */
6214
6215 for (oc = cd->open_caps; oc != NULL; oc = oc->next)
6216 {
6217 if (oc->number == recno)
6218 {
6219 oc->flag = TRUE;
6220 break;
6221 }
6222 }
6223 }
6224
6225 /* So are Unicode property matches, if supported. */
6226
6227 #ifdef SUPPORT_UCP
6228 else if (-c == ESC_P || -c == ESC_p)
6229 {
6230 BOOL negated;
6231 int pdata;
6232 int ptype = get_ucp(&ptr, &negated, &pdata, errorcodeptr);
6233 if (ptype < 0) goto FAILED;
6234 previous = code;
6235 *code++ = ((-c == ESC_p) != negated)? OP_PROP : OP_NOTPROP;
6236 *code++ = ptype;
6237 *code++ = pdata;
6238 }
6239 #else
6240
6241 /* If Unicode properties are not supported, \X, \P, and \p are not
6242 allowed. */
6243
6244 else if (-c == ESC_X || -c == ESC_P || -c == ESC_p)
6245 {
6246 *errorcodeptr = ERR45;
6247 goto FAILED;
6248 }
6249 #endif
6250
6251 /* For the rest (including \X when Unicode properties are supported), we
6252 can obtain the OP value by negating the escape value in the default
6253 situation when PCRE_UCP is not set. When it *is* set, we substitute
6254 Unicode property tests. */
6255
6256 else
6257 {
6258 #ifdef SUPPORT_UCP
6259 if (-c >= ESC_DU && -c <= ESC_wu)
6260 {
6261 nestptr = ptr + 1; /* Where to resume */
6262 ptr = substitutes[-c - ESC_DU] - 1; /* Just before substitute */
6263 }
6264 else
6265 #endif
6266 {
6267 previous = (-c > ESC_b && -c < ESC_Z)? code : NULL;
6268 *code++ = -c;
6269 }
6270 }
6271 continue;
6272 }
6273
6274 /* We have a data character whose value is in c. In UTF-8 mode it may have
6275 a value > 127. We set its representation in the length/buffer, and then
6276 handle it as a data character. */
6277
6278 #ifdef SUPPORT_UTF8
6279 if (utf8 && c > 127)
6280 mclength = _pcre_ord2utf8(c, mcbuffer);
6281 else
6282 #endif
6283
6284 {
6285 mcbuffer[0] = c;
6286 mclength = 1;
6287 }
6288 goto ONE_CHAR;
6289
6290
6291 /* ===================================================================*/
6292 /* Handle a literal character. It is guaranteed not to be whitespace or #
6293 when the extended flag is set. If we are in UTF-8 mode, it may be a
6294 multi-byte literal character. */
6295
6296 default:
6297 NORMAL_CHAR:
6298 mclength = 1;
6299 mcbuffer[0] = c;
6300
6301 #ifdef SUPPORT_UTF8
6302 if (utf8 && c >= 0xc0)
6303 {
6304 while ((ptr[1] & 0xc0) == 0x80)
6305 mcbuffer[mclength++] = *(++ptr);
6306 }
6307 #endif
6308
6309 /* At this point we have the character's bytes in mcbuffer, and the length
6310 in mclength. When not in UTF-8 mode, the length is always 1. */
6311
6312 ONE_CHAR:
6313 previous = code;
6314 *code++ = ((options & PCRE_CASELESS) != 0)? OP_CHARI : OP_CHAR;
6315 for (c = 0; c < mclength; c++) *code++ = mcbuffer[c];
6316
6317 /* Remember if \r or \n were seen */
6318
6319 if (mcbuffer[0] == CHAR_CR || mcbuffer[0] == CHAR_NL)
6320 cd->external_flags |= PCRE_HASCRORLF;
6321
6322 /* Set the first and required bytes appropriately. If no previous first
6323 byte, set it from this character, but revert to none on a zero repeat.
6324 Otherwise, leave the firstbyte value alone, and don't change it on a zero
6325 repeat. */
6326
6327 if (firstbyte == REQ_UNSET)
6328 {
6329 zerofirstbyte = REQ_NONE;
6330 zeroreqbyte = reqbyte;
6331
6332 /* If the character is more than one byte long, we can set firstbyte
6333 only if it is not to be matched caselessly. */
6334
6335 if (mclength == 1 || req_caseopt == 0)
6336 {
6337 firstbyte = mcbuffer[0] | req_caseopt;
6338 if (mclength != 1) reqbyte = code[-1] | cd->req_varyopt;
6339 }
6340 else firstbyte = reqbyte = REQ_NONE;
6341 }
6342
6343 /* firstbyte was previously set; we can set reqbyte only the length is
6344 1 or the matching is caseful. */
6345
6346 else
6347 {
6348 zerofirstbyte = firstbyte;
6349 zeroreqbyte = reqbyte;
6350 if (mclength == 1 || req_caseopt == 0)
6351 reqbyte = code[-1] | req_caseopt | cd->req_varyopt;
6352 }
6353
6354 break; /* End of literal character handling */
6355 }
6356 } /* end of big loop */
6357
6358
6359 /* Control never reaches here by falling through, only by a goto for all the
6360 error states. Pass back the position in the pattern so that it can be displayed
6361 to the user for diagnosing the error. */
6362
6363 FAILED:
6364 *ptrptr = ptr;
6365 return FALSE;
6366 }
6367
6368
6369
6370
6371 /*************************************************
6372 * Compile sequence of alternatives *
6373 *************************************************/
6374
6375 /* On entry, ptr is pointing past the bracket character, but on return it
6376 points to the closing bracket, or vertical bar, or end of string. The code
6377 variable is pointing at the byte into which the BRA operator has been stored.
6378 This function is used during the pre-compile phase when we are trying to find
6379 out the amount of memory needed, as well as during the real compile phase. The
6380 value of lengthptr distinguishes the two phases.
6381
6382 Arguments:
6383 options option bits, including any changes for this subpattern
6384 codeptr -> the address of the current code pointer
6385 ptrptr -> the address of the current pattern pointer
6386 errorcodeptr -> pointer to error code variable
6387 lookbehind TRUE if this is a lookbehind assertion
6388 reset_bracount TRUE to reset the count for each branch
6389 skipbytes skip this many bytes at start (for brackets and OP_COND)
6390 cond_depth depth of nesting for conditional subpatterns
6391 firstbyteptr place to put the first required character, or a negative number
6392 reqbyteptr place to put the last required character, or a negative number
6393 bcptr pointer to the chain of currently open branches
6394 cd points to the data block with tables pointers etc.
6395 lengthptr NULL during the real compile phase
6396 points to length accumulator during pre-compile phase
6397
6398 Returns: TRUE on success
6399 */
6400
6401 static BOOL
6402 compile_regex(int options, uschar **codeptr, const uschar **ptrptr,
6403 int *errorcodeptr, BOOL lookbehind, BOOL reset_bracount, int skipbytes,
6404 int cond_depth, int *firstbyteptr, int *reqbyteptr, branch_chain *bcptr,
6405 compile_data *cd, int *lengthptr)
6406 {
6407 const uschar *ptr = *ptrptr;
6408 uschar *code = *codeptr;
6409 uschar *last_branch = code;
6410 uschar *start_bracket = code;
6411 uschar *reverse_count = NULL;
6412 open_capitem capitem;
6413 int capnumber = 0;
6414 int firstbyte, reqbyte;
6415 int branchfirstbyte, branchreqbyte;
6416 int length;
6417 int orig_bracount;
6418 int max_bracount;
6419 branch_chain bc;
6420
6421 bc.outer = bcptr;
6422 bc.current_branch = code;
6423
6424 firstbyte = reqbyte = REQ_UNSET;
6425
6426 /* Accumulate the length for use in the pre-compile phase. Start with the
6427 length of the BRA and KET and any extra bytes that are required at the
6428 beginning. We accumulate in a local variable to save frequent testing of
6429 lenthptr for NULL. We cannot do this by looking at the value of code at the
6430 start and end of each alternative, because compiled items are discarded during
6431 the pre-compile phase so that the work space is not exceeded. */
6432
6433 length = 2 + 2*LINK_SIZE + skipbytes;
6434
6435 /* WARNING: If the above line is changed for any reason, you must also change
6436 the code that abstracts option settings at the start of the pattern and makes
6437 them global. It tests the value of length for (2 + 2*LINK_SIZE) in the
6438 pre-compile phase to find out whether anything has yet been compiled or not. */
6439
6440 /* If this is a capturing subpattern, add to the chain of open capturing items
6441 so that we can detect them if (*ACCEPT) is encountered. This is also used to
6442 detect groups that contain recursive back references to themselves. Note that
6443 only OP_CBRA need be tested here; changing this opcode to one of its variants,
6444 e.g. OP_SCBRAPOS, happens later, after the group has been compiled. */
6445
6446 if (*code == OP_CBRA)
6447 {
6448 capnumber = GET2(code, 1 + LINK_SIZE);
6449 capitem.number = capnumber;
6450 capitem.next = cd->open_caps;
6451 capitem.flag = FALSE;
6452 cd->open_caps = &capitem;
6453 }
6454
6455 /* Offset is set zero to mark that this bracket is still open */
6456
6457 PUT(code, 1, 0);
6458 code += 1 + LINK_SIZE + skipbytes;
6459
6460 /* Loop for each alternative branch */
6461
6462 orig_bracount = max_bracount = cd->bracount;
6463 for (;;)
6464 {
6465 /* For a (?| group, reset the capturing bracket count so that each branch
6466 uses the same numbers. */
6467
6468 if (reset_bracount) cd->bracount = orig_bracount;
6469
6470 /* Set up dummy OP_REVERSE if lookbehind assertion */
6471
6472 if (lookbehind)
6473 {
6474 *code++ = OP_REVERSE;
6475 reverse_count = code;
6476 PUTINC(code, 0, 0);
6477 length += 1 + LINK_SIZE;
6478 }
6479
6480 /* Now compile the branch; in the pre-compile phase its length gets added
6481 into the length. */
6482
6483 if (!compile_branch(&options, &code, &ptr, errorcodeptr, &branchfirstbyte,
6484 &branchreqbyte, &bc, cond_depth, cd,
6485 (lengthptr == NULL)? NULL : &length))
6486 {
6487 *ptrptr = ptr;
6488 return FALSE;
6489 }
6490
6491 /* Keep the highest bracket count in case (?| was used and some branch
6492 has fewer than the rest. */
6493
6494 if (cd->bracount > max_bracount) max_bracount = cd->bracount;
6495
6496 /* In the real compile phase, there is some post-processing to be done. */
6497
6498 if (lengthptr == NULL)
6499 {
6500 /* If this is the first branch, the firstbyte and reqbyte values for the
6501 branch become the values for the regex. */
6502
6503 if (*last_branch != OP_ALT)
6504 {
6505 firstbyte = branchfirstbyte;
6506 reqbyte = branchreqbyte;
6507 }
6508
6509 /* If this is not the first branch, the first char and reqbyte have to
6510 match the values from all the previous branches, except that if the
6511 previous value for reqbyte didn't have REQ_VARY set, it can still match,
6512 and we set REQ_VARY for the regex. */
6513
6514 else
6515 {
6516 /* If we previously had a firstbyte, but it doesn't match the new branch,
6517 we have to abandon the firstbyte for the regex, but if there was
6518 previously no reqbyte, it takes on the value of the old firstbyte. */
6519
6520 if (firstbyte >= 0 && firstbyte != branchfirstbyte)
6521 {
6522 if (reqbyte < 0) reqbyte = firstbyte;
6523 firstbyte = REQ_NONE;
6524 }
6525
6526 /* If we (now or from before) have no firstbyte, a firstbyte from the
6527 branch becomes a reqbyte if there isn't a branch reqbyte. */
6528
6529 if (firstbyte < 0 && branchfirstbyte >= 0 && branchreqbyte < 0)
6530 branchreqbyte = branchfirstbyte;
6531
6532 /* Now ensure that the reqbytes match */
6533
6534 if ((reqbyte & ~REQ_VARY) != (branchreqbyte & ~REQ_VARY))
6535 reqbyte = REQ_NONE;
6536 else reqbyte |= branchreqbyte; /* To "or" REQ_VARY */
6537 }
6538
6539 /* If lookbehind, check that this branch matches a fixed-length string, and
6540 put the length into the OP_REVERSE item. Temporarily mark the end of the
6541 branch with OP_END. If the branch contains OP_RECURSE, the result is -3
6542 because there may be forward references that we can't check here. Set a
6543 flag to cause another lookbehind check at the end. Why not do it all at the
6544 end? Because common, erroneous checks are picked up here and the offset of
6545 the problem can be shown. */
6546
6547 if (lookbehind)
6548 {
6549 int fixed_length;
6550 *code = OP_END;
6551 fixed_length = find_fixedlength(last_branch, (options & PCRE_UTF8) != 0,
6552 FALSE, cd);
6553 DPRINTF(("fixed length = %d\n", fixed_length));
6554 if (fixed_length == -3)
6555 {
6556 cd->check_lookbehind = TRUE;
6557 }
6558 else if (fixed_length < 0)
6559 {
6560 *errorcodeptr = (fixed_length == -2)? ERR36 : ERR25;
6561 *ptrptr = ptr;
6562 return FALSE;
6563 }
6564 else { PUT(reverse_count, 0, fixed_length); }
6565 }
6566 }
6567
6568 /* Reached end of expression, either ')' or end of pattern. In the real
6569 compile phase, go back through the alternative branches and reverse the chain
6570 of offsets, with the field in the BRA item now becoming an offset to the
6571 first alternative. If there are no alternatives, it points to the end of the
6572 group. The length in the terminating ket is always the length of the whole
6573 bracketed item. Return leaving the pointer at the terminating char. */
6574
6575 if (*ptr != CHAR_VERTICAL_LINE)
6576 {
6577 if (lengthptr == NULL)
6578 {
6579 int branch_length = (int)(code - last_branch);
6580 do
6581 {
6582 int prev_length = GET(last_branch, 1);
6583 PUT(last_branch, 1, branch_length);
6584 branch_length = prev_length;
6585 last_branch -= branch_length;
6586 }
6587 while (branch_length > 0);
6588 }
6589
6590 /* Fill in the ket */
6591
6592 *code = OP_KET;
6593 PUT(code, 1, (int)(code - start_bracket));
6594 code += 1 + LINK_SIZE;
6595
6596 /* If it was a capturing subpattern, check to see if it contained any
6597 recursive back references. If so, we must wrap it in atomic brackets.
6598 In any event, remove the block from the chain. */
6599
6600 if (capnumber > 0)
6601 {
6602 if (cd->open_caps->flag)
6603 {
6604 memmove(start_bracket + 1 + LINK_SIZE, start_bracket,
6605 code - start_bracket);
6606 *start_bracket = OP_ONCE;