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

Contents of /code/trunk/pcre_compile.c

Parent Directory Parent Directory | Revision Log Revision Log


Revision 723 - (show annotations) (download)
Sat Oct 8 15:55:23 2011 UTC (2 years, 10 months ago) by ph10
File MIME type: text/plain
File size: 244148 byte(s)
Revert handling of atomic groups that do not include captures to the old way of 
handling, thus reducing stack usage.

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_ONCE_NC:
1510 case OP_COND:
1511 d = find_fixedlength(cc + ((op == OP_CBRA)? 2:0), utf8, atend, cd);
1512 if (d < 0) return d;
1513 branchlength += d;
1514 do cc += GET(cc, 1); while (*cc == OP_ALT);
1515 cc += 1 + LINK_SIZE;
1516 break;
1517
1518 /* Reached end of a branch; if it's a ket it is the end of a nested
1519 call. If it's ALT it is an alternation in a nested call. If it is
1520 END it's the end of the outer call. All can be handled by the same code.
1521 Note that we must not include the OP_KETRxxx opcodes here, because they
1522 all imply an unlimited repeat. */
1523
1524 case OP_ALT:
1525 case OP_KET:
1526 case OP_END:
1527 if (length < 0) length = branchlength;
1528 else if (length != branchlength) return -1;
1529 if (*cc != OP_ALT) return length;
1530 cc += 1 + LINK_SIZE;
1531 branchlength = 0;
1532 break;
1533
1534 /* A true recursion implies not fixed length, but a subroutine call may
1535 be OK. If the subroutine is a forward reference, we can't deal with
1536 it until the end of the pattern, so return -3. */
1537
1538 case OP_RECURSE:
1539 if (!atend) return -3;
1540 cs = ce = (uschar *)cd->start_code + GET(cc, 1); /* Start subpattern */
1541 do ce += GET(ce, 1); while (*ce == OP_ALT); /* End subpattern */
1542 if (cc > cs && cc < ce) return -1; /* Recursion */
1543 d = find_fixedlength(cs + 2, utf8, atend, cd);
1544 if (d < 0) return d;
1545 branchlength += d;
1546 cc += 1 + LINK_SIZE;
1547 break;
1548
1549 /* Skip over assertive subpatterns */
1550
1551 case OP_ASSERT:
1552 case OP_ASSERT_NOT:
1553 case OP_ASSERTBACK:
1554 case OP_ASSERTBACK_NOT:
1555 do cc += GET(cc, 1); while (*cc == OP_ALT);
1556 /* Fall through */
1557
1558 /* Skip over things that don't match chars */
1559
1560 case OP_REVERSE:
1561 case OP_CREF:
1562 case OP_NCREF:
1563 case OP_RREF:
1564 case OP_NRREF:
1565 case OP_DEF:
1566 case OP_CALLOUT:
1567 case OP_SOD:
1568 case OP_SOM:
1569 case OP_SET_SOM:
1570 case OP_EOD:
1571 case OP_EODN:
1572 case OP_CIRC:
1573 case OP_CIRCM:
1574 case OP_DOLL:
1575 case OP_DOLLM:
1576 case OP_NOT_WORD_BOUNDARY:
1577 case OP_WORD_BOUNDARY:
1578 cc += _pcre_OP_lengths[*cc];
1579 break;
1580
1581 /* Handle literal characters */
1582
1583 case OP_CHAR:
1584 case OP_CHARI:
1585 case OP_NOT:
1586 case OP_NOTI:
1587 branchlength++;
1588 cc += 2;
1589 #ifdef SUPPORT_UTF8
1590 if (utf8 && cc[-1] >= 0xc0) cc += _pcre_utf8_table4[cc[-1] & 0x3f];
1591 #endif
1592 break;
1593
1594 /* Handle exact repetitions. The count is already in characters, but we
1595 need to skip over a multibyte character in UTF8 mode. */
1596
1597 case OP_EXACT:
1598 branchlength += GET2(cc,1);
1599 cc += 4;
1600 #ifdef SUPPORT_UTF8
1601 if (utf8 && cc[-1] >= 0xc0) cc += _pcre_utf8_table4[cc[-1] & 0x3f];
1602 #endif
1603 break;
1604
1605 case OP_TYPEEXACT:
1606 branchlength += GET2(cc,1);
1607 if (cc[3] == OP_PROP || cc[3] == OP_NOTPROP) cc += 2;
1608 cc += 4;
1609 break;
1610
1611 /* Handle single-char matchers */
1612
1613 case OP_PROP:
1614 case OP_NOTPROP:
1615 cc += 2;
1616 /* Fall through */
1617
1618 case OP_NOT_DIGIT:
1619 case OP_DIGIT:
1620 case OP_NOT_WHITESPACE:
1621 case OP_WHITESPACE:
1622 case OP_NOT_WORDCHAR:
1623 case OP_WORDCHAR:
1624 case OP_ANY:
1625 case OP_ALLANY:
1626 branchlength++;
1627 cc++;
1628 break;
1629
1630 /* The single-byte matcher isn't allowed */
1631
1632 case OP_ANYBYTE:
1633 return -2;
1634
1635 /* Check a class for variable quantification */
1636
1637 #ifdef SUPPORT_UTF8
1638 case OP_XCLASS:
1639 cc += GET(cc, 1) - 33;
1640 /* Fall through */
1641 #endif
1642
1643 case OP_CLASS:
1644 case OP_NCLASS:
1645 cc += 33;
1646
1647 switch (*cc)
1648 {
1649 case OP_CRSTAR:
1650 case OP_CRMINSTAR:
1651 case OP_CRQUERY:
1652 case OP_CRMINQUERY:
1653 return -1;
1654
1655 case OP_CRRANGE:
1656 case OP_CRMINRANGE:
1657 if (GET2(cc,1) != GET2(cc,3)) return -1;
1658 branchlength += GET2(cc,1);
1659 cc += 5;
1660 break;
1661
1662 default:
1663 branchlength++;
1664 }
1665 break;
1666
1667 /* Anything else is variable length */
1668
1669 default:
1670 return -1;
1671 }
1672 }
1673 /* Control never gets here */
1674 }
1675
1676
1677
1678
1679 /*************************************************
1680 * Scan compiled regex for specific bracket *
1681 *************************************************/
1682
1683 /* This little function scans through a compiled pattern until it finds a
1684 capturing bracket with the given number, or, if the number is negative, an
1685 instance of OP_REVERSE for a lookbehind. The function is global in the C sense
1686 so that it can be called from pcre_study() when finding the minimum matching
1687 length.
1688
1689 Arguments:
1690 code points to start of expression
1691 utf8 TRUE in UTF-8 mode
1692 number the required bracket number or negative to find a lookbehind
1693
1694 Returns: pointer to the opcode for the bracket, or NULL if not found
1695 */
1696
1697 const uschar *
1698 _pcre_find_bracket(const uschar *code, BOOL utf8, int number)
1699 {
1700 for (;;)
1701 {
1702 register int c = *code;
1703
1704 if (c == OP_END) return NULL;
1705
1706 /* XCLASS is used for classes that cannot be represented just by a bit
1707 map. This includes negated single high-valued characters. The length in
1708 the table is zero; the actual length is stored in the compiled code. */
1709
1710 if (c == OP_XCLASS) code += GET(code, 1);
1711
1712 /* Handle recursion */
1713
1714 else if (c == OP_REVERSE)
1715 {
1716 if (number < 0) return (uschar *)code;
1717 code += _pcre_OP_lengths[c];
1718 }
1719
1720 /* Handle capturing bracket */
1721
1722 else if (c == OP_CBRA || c == OP_SCBRA ||
1723 c == OP_CBRAPOS || c == OP_SCBRAPOS)
1724 {
1725 int n = GET2(code, 1+LINK_SIZE);
1726 if (n == number) return (uschar *)code;
1727 code += _pcre_OP_lengths[c];
1728 }
1729
1730 /* Otherwise, we can get the item's length from the table, except that for
1731 repeated character types, we have to test for \p and \P, which have an extra
1732 two bytes of parameters, and for MARK/PRUNE/SKIP/THEN with an argument, we
1733 must add in its length. */
1734
1735 else
1736 {
1737 switch(c)
1738 {
1739 case OP_TYPESTAR:
1740 case OP_TYPEMINSTAR:
1741 case OP_TYPEPLUS:
1742 case OP_TYPEMINPLUS:
1743 case OP_TYPEQUERY:
1744 case OP_TYPEMINQUERY:
1745 case OP_TYPEPOSSTAR:
1746 case OP_TYPEPOSPLUS:
1747 case OP_TYPEPOSQUERY:
1748 if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
1749 break;
1750
1751 case OP_TYPEUPTO:
1752 case OP_TYPEMINUPTO:
1753 case OP_TYPEEXACT:
1754 case OP_TYPEPOSUPTO:
1755 if (code[3] == OP_PROP || code[3] == OP_NOTPROP) code += 2;
1756 break;
1757
1758 case OP_MARK:
1759 case OP_PRUNE_ARG:
1760 case OP_SKIP_ARG:
1761 code += code[1];
1762 break;
1763
1764 case OP_THEN_ARG:
1765 code += code[1];
1766 break;
1767 }
1768
1769 /* Add in the fixed length from the table */
1770
1771 code += _pcre_OP_lengths[c];
1772
1773 /* In UTF-8 mode, opcodes that are followed by a character may be followed by
1774 a multi-byte character. The length in the table is a minimum, so we have to
1775 arrange to skip the extra bytes. */
1776
1777 #ifdef SUPPORT_UTF8
1778 if (utf8) switch(c)
1779 {
1780 case OP_CHAR:
1781 case OP_CHARI:
1782 case OP_EXACT:
1783 case OP_EXACTI:
1784 case OP_UPTO:
1785 case OP_UPTOI:
1786 case OP_MINUPTO:
1787 case OP_MINUPTOI:
1788 case OP_POSUPTO:
1789 case OP_POSUPTOI:
1790 case OP_STAR:
1791 case OP_STARI:
1792 case OP_MINSTAR:
1793 case OP_MINSTARI:
1794 case OP_POSSTAR:
1795 case OP_POSSTARI:
1796 case OP_PLUS:
1797 case OP_PLUSI:
1798 case OP_MINPLUS:
1799 case OP_MINPLUSI:
1800 case OP_POSPLUS:
1801 case OP_POSPLUSI:
1802 case OP_QUERY:
1803 case OP_QUERYI:
1804 case OP_MINQUERY:
1805 case OP_MINQUERYI:
1806 case OP_POSQUERY:
1807 case OP_POSQUERYI:
1808 if (code[-1] >= 0xc0) code += _pcre_utf8_table4[code[-1] & 0x3f];
1809 break;
1810 }
1811 #else
1812 (void)(utf8); /* Keep compiler happy by referencing function argument */
1813 #endif
1814 }
1815 }
1816 }
1817
1818
1819
1820 /*************************************************
1821 * Scan compiled regex for recursion reference *
1822 *************************************************/
1823
1824 /* This little function scans through a compiled pattern until it finds an
1825 instance of OP_RECURSE.
1826
1827 Arguments:
1828 code points to start of expression
1829 utf8 TRUE in UTF-8 mode
1830
1831 Returns: pointer to the opcode for OP_RECURSE, or NULL if not found
1832 */
1833
1834 static const uschar *
1835 find_recurse(const uschar *code, BOOL utf8)
1836 {
1837 for (;;)
1838 {
1839 register int c = *code;
1840 if (c == OP_END) return NULL;
1841 if (c == OP_RECURSE) return code;
1842
1843 /* XCLASS is used for classes that cannot be represented just by a bit
1844 map. This includes negated single high-valued characters. The length in
1845 the table is zero; the actual length is stored in the compiled code. */
1846
1847 if (c == OP_XCLASS) code += GET(code, 1);
1848
1849 /* Otherwise, we can get the item's length from the table, except that for
1850 repeated character types, we have to test for \p and \P, which have an extra
1851 two bytes of parameters, and for MARK/PRUNE/SKIP/THEN with an argument, we
1852 must add in its length. */
1853
1854 else
1855 {
1856 switch(c)
1857 {
1858 case OP_TYPESTAR:
1859 case OP_TYPEMINSTAR:
1860 case OP_TYPEPLUS:
1861 case OP_TYPEMINPLUS:
1862 case OP_TYPEQUERY:
1863 case OP_TYPEMINQUERY:
1864 case OP_TYPEPOSSTAR:
1865 case OP_TYPEPOSPLUS:
1866 case OP_TYPEPOSQUERY:
1867 if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
1868 break;
1869
1870 case OP_TYPEPOSUPTO:
1871 case OP_TYPEUPTO:
1872 case OP_TYPEMINUPTO:
1873 case OP_TYPEEXACT:
1874 if (code[3] == OP_PROP || code[3] == OP_NOTPROP) code += 2;
1875 break;
1876
1877 case OP_MARK:
1878 case OP_PRUNE_ARG:
1879 case OP_SKIP_ARG:
1880 code += code[1];
1881 break;
1882
1883 case OP_THEN_ARG:
1884 code += code[1];
1885 break;
1886 }
1887
1888 /* Add in the fixed length from the table */
1889
1890 code += _pcre_OP_lengths[c];
1891
1892 /* In UTF-8 mode, opcodes that are followed by a character may be followed
1893 by a multi-byte character. The length in the table is a minimum, so we have
1894 to arrange to skip the extra bytes. */
1895
1896 #ifdef SUPPORT_UTF8
1897 if (utf8) switch(c)
1898 {
1899 case OP_CHAR:
1900 case OP_CHARI:
1901 case OP_EXACT:
1902 case OP_EXACTI:
1903 case OP_UPTO:
1904 case OP_UPTOI:
1905 case OP_MINUPTO:
1906 case OP_MINUPTOI:
1907 case OP_POSUPTO:
1908 case OP_POSUPTOI:
1909 case OP_STAR:
1910 case OP_STARI:
1911 case OP_MINSTAR:
1912 case OP_MINSTARI:
1913 case OP_POSSTAR:
1914 case OP_POSSTARI:
1915 case OP_PLUS:
1916 case OP_PLUSI:
1917 case OP_MINPLUS:
1918 case OP_MINPLUSI:
1919 case OP_POSPLUS:
1920 case OP_POSPLUSI:
1921 case OP_QUERY:
1922 case OP_QUERYI:
1923 case OP_MINQUERY:
1924 case OP_MINQUERYI:
1925 case OP_POSQUERY:
1926 case OP_POSQUERYI:
1927 if (code[-1] >= 0xc0) code += _pcre_utf8_table4[code[-1] & 0x3f];
1928 break;
1929 }
1930 #else
1931 (void)(utf8); /* Keep compiler happy by referencing function argument */
1932 #endif
1933 }
1934 }
1935 }
1936
1937
1938
1939 /*************************************************
1940 * Scan compiled branch for non-emptiness *
1941 *************************************************/
1942
1943 /* This function scans through a branch of a compiled pattern to see whether it
1944 can match the empty string or not. It is called from could_be_empty()
1945 below and from compile_branch() when checking for an unlimited repeat of a
1946 group that can match nothing. Note that first_significant_code() skips over
1947 backward and negative forward assertions when its final argument is TRUE. If we
1948 hit an unclosed bracket, we return "empty" - this means we've struck an inner
1949 bracket whose current branch will already have been scanned.
1950
1951 Arguments:
1952 code points to start of search
1953 endcode points to where to stop
1954 utf8 TRUE if in UTF8 mode
1955 cd contains pointers to tables etc.
1956
1957 Returns: TRUE if what is matched could be empty
1958 */
1959
1960 static BOOL
1961 could_be_empty_branch(const uschar *code, const uschar *endcode, BOOL utf8,
1962 compile_data *cd)
1963 {
1964 register int c;
1965 for (code = first_significant_code(code + _pcre_OP_lengths[*code], TRUE);
1966 code < endcode;
1967 code = first_significant_code(code + _pcre_OP_lengths[c], TRUE))
1968 {
1969 const uschar *ccode;
1970
1971 c = *code;
1972
1973 /* Skip over forward assertions; the other assertions are skipped by
1974 first_significant_code() with a TRUE final argument. */
1975
1976 if (c == OP_ASSERT)
1977 {
1978 do code += GET(code, 1); while (*code == OP_ALT);
1979 c = *code;
1980 continue;
1981 }
1982
1983 /* For a recursion/subroutine call, if its end has been reached, which
1984 implies a backward reference subroutine call, we can scan it. If it's a
1985 forward reference subroutine call, we can't. To detect forward reference
1986 we have to scan up the list that is kept in the workspace. This function is
1987 called only when doing the real compile, not during the pre-compile that
1988 measures the size of the compiled pattern. */
1989
1990 if (c == OP_RECURSE)
1991 {
1992 const uschar *scode;
1993 BOOL empty_branch;
1994
1995 /* Test for forward reference */
1996
1997 for (scode = cd->start_workspace; scode < cd->hwm; scode += LINK_SIZE)
1998 if (GET(scode, 0) == code + 1 - cd->start_code) return TRUE;
1999
2000 /* Not a forward reference, test for completed backward reference */
2001
2002 empty_branch = FALSE;
2003 scode = cd->start_code + GET(code, 1);
2004 if (GET(scode, 1) == 0) return TRUE; /* Unclosed */
2005
2006 /* Completed backwards reference */
2007
2008 do
2009 {
2010 if (could_be_empty_branch(scode, endcode, utf8, cd))
2011 {
2012 empty_branch = TRUE;
2013 break;
2014 }
2015 scode += GET(scode, 1);
2016 }
2017 while (*scode == OP_ALT);
2018
2019 if (!empty_branch) return FALSE; /* All branches are non-empty */
2020 continue;
2021 }
2022
2023 /* Groups with zero repeats can of course be empty; skip them. */
2024
2025 if (c == OP_BRAZERO || c == OP_BRAMINZERO || c == OP_SKIPZERO ||
2026 c == OP_BRAPOSZERO)
2027 {
2028 code += _pcre_OP_lengths[c];
2029 do code += GET(code, 1); while (*code == OP_ALT);
2030 c = *code;
2031 continue;
2032 }
2033
2034 /* A nested group that is already marked as "could be empty" can just be
2035 skipped. */
2036
2037 if (c == OP_SBRA || c == OP_SBRAPOS ||
2038 c == OP_SCBRA || c == OP_SCBRAPOS)
2039 {
2040 do code += GET(code, 1); while (*code == OP_ALT);
2041 c = *code;
2042 continue;
2043 }
2044
2045 /* For other groups, scan the branches. */
2046
2047 if (c == OP_BRA || c == OP_BRAPOS ||
2048 c == OP_CBRA || c == OP_CBRAPOS ||
2049 c == OP_ONCE || c == OP_ONCE_NC ||
2050 c == OP_COND)
2051 {
2052 BOOL empty_branch;
2053 if (GET(code, 1) == 0) return TRUE; /* Hit unclosed bracket */
2054
2055 /* If a conditional group has only one branch, there is a second, implied,
2056 empty branch, so just skip over the conditional, because it could be empty.
2057 Otherwise, scan the individual branches of the group. */
2058
2059 if (c == OP_COND && code[GET(code, 1)] != OP_ALT)
2060 code += GET(code, 1);
2061 else
2062 {
2063 empty_branch = FALSE;
2064 do
2065 {
2066 if (!empty_branch && could_be_empty_branch(code, endcode, utf8, cd))
2067 empty_branch = TRUE;
2068 code += GET(code, 1);
2069 }
2070 while (*code == OP_ALT);
2071 if (!empty_branch) return FALSE; /* All branches are non-empty */
2072 }
2073
2074 c = *code;
2075 continue;
2076 }
2077
2078 /* Handle the other opcodes */
2079
2080 switch (c)
2081 {
2082 /* Check for quantifiers after a class. XCLASS is used for classes that
2083 cannot be represented just by a bit map. This includes negated single
2084 high-valued characters. The length in _pcre_OP_lengths[] is zero; the
2085 actual length is stored in the compiled code, so we must update "code"
2086 here. */
2087
2088 #ifdef SUPPORT_UTF8
2089 case OP_XCLASS:
2090 ccode = code += GET(code, 1);
2091 goto CHECK_CLASS_REPEAT;
2092 #endif
2093
2094 case OP_CLASS:
2095 case OP_NCLASS:
2096 ccode = code + 33;
2097
2098 #ifdef SUPPORT_UTF8
2099 CHECK_CLASS_REPEAT:
2100 #endif
2101
2102 switch (*ccode)
2103 {
2104 case OP_CRSTAR: /* These could be empty; continue */
2105 case OP_CRMINSTAR:
2106 case OP_CRQUERY:
2107 case OP_CRMINQUERY:
2108 break;
2109
2110 default: /* Non-repeat => class must match */
2111 case OP_CRPLUS: /* These repeats aren't empty */
2112 case OP_CRMINPLUS:
2113 return FALSE;
2114
2115 case OP_CRRANGE:
2116 case OP_CRMINRANGE:
2117 if (GET2(ccode, 1) > 0) return FALSE; /* Minimum > 0 */
2118 break;
2119 }
2120 break;
2121
2122 /* Opcodes that must match a character */
2123
2124 case OP_PROP:
2125 case OP_NOTPROP:
2126 case OP_EXTUNI:
2127 case OP_NOT_DIGIT:
2128 case OP_DIGIT:
2129 case OP_NOT_WHITESPACE:
2130 case OP_WHITESPACE:
2131 case OP_NOT_WORDCHAR:
2132 case OP_WORDCHAR:
2133 case OP_ANY:
2134 case OP_ALLANY:
2135 case OP_ANYBYTE:
2136 case OP_CHAR:
2137 case OP_CHARI:
2138 case OP_NOT:
2139 case OP_NOTI:
2140 case OP_PLUS:
2141 case OP_MINPLUS:
2142 case OP_POSPLUS:
2143 case OP_EXACT:
2144 case OP_NOTPLUS:
2145 case OP_NOTMINPLUS:
2146 case OP_NOTPOSPLUS:
2147 case OP_NOTEXACT:
2148 case OP_TYPEPLUS:
2149 case OP_TYPEMINPLUS:
2150 case OP_TYPEPOSPLUS:
2151 case OP_TYPEEXACT:
2152 return FALSE;
2153
2154 /* These are going to continue, as they may be empty, but we have to
2155 fudge the length for the \p and \P cases. */
2156
2157 case OP_TYPESTAR:
2158 case OP_TYPEMINSTAR:
2159 case OP_TYPEPOSSTAR:
2160 case OP_TYPEQUERY:
2161 case OP_TYPEMINQUERY:
2162 case OP_TYPEPOSQUERY:
2163 if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
2164 break;
2165
2166 /* Same for these */
2167
2168 case OP_TYPEUPTO:
2169 case OP_TYPEMINUPTO:
2170 case OP_TYPEPOSUPTO:
2171 if (code[3] == OP_PROP || code[3] == OP_NOTPROP) code += 2;
2172 break;
2173
2174 /* End of branch */
2175
2176 case OP_KET:
2177 case OP_KETRMAX:
2178 case OP_KETRMIN:
2179 case OP_KETRPOS:
2180 case OP_ALT:
2181 return TRUE;
2182
2183 /* In UTF-8 mode, STAR, MINSTAR, POSSTAR, QUERY, MINQUERY, POSQUERY, UPTO,
2184 MINUPTO, and POSUPTO may be followed by a multibyte character */
2185
2186 #ifdef SUPPORT_UTF8
2187 case OP_STAR:
2188 case OP_STARI:
2189 case OP_MINSTAR:
2190 case OP_MINSTARI:
2191 case OP_POSSTAR:
2192 case OP_POSSTARI:
2193 case OP_QUERY:
2194 case OP_QUERYI:
2195 case OP_MINQUERY:
2196 case OP_MINQUERYI:
2197 case OP_POSQUERY:
2198 case OP_POSQUERYI:
2199 if (utf8 && code[1] >= 0xc0) code += _pcre_utf8_table4[code[1] & 0x3f];
2200 break;
2201
2202 case OP_UPTO:
2203 case OP_UPTOI:
2204 case OP_MINUPTO:
2205 case OP_MINUPTOI:
2206 case OP_POSUPTO:
2207 case OP_POSUPTOI:
2208 if (utf8 && code[3] >= 0xc0) code += _pcre_utf8_table4[code[3] & 0x3f];
2209 break;
2210 #endif
2211
2212 /* MARK, and PRUNE/SKIP/THEN with an argument must skip over the argument
2213 string. */
2214
2215 case OP_MARK:
2216 case OP_PRUNE_ARG:
2217 case OP_SKIP_ARG:
2218 code += code[1];
2219 break;
2220
2221 case OP_THEN_ARG:
2222 code += code[1];
2223 break;
2224
2225 /* None of the remaining opcodes are required to match a character. */
2226
2227 default:
2228 break;
2229 }
2230 }
2231
2232 return TRUE;
2233 }
2234
2235
2236
2237 /*************************************************
2238 * Scan compiled regex for non-emptiness *
2239 *************************************************/
2240
2241 /* This function is called to check for left recursive calls. We want to check
2242 the current branch of the current pattern to see if it could match the empty
2243 string. If it could, we must look outwards for branches at other levels,
2244 stopping when we pass beyond the bracket which is the subject of the recursion.
2245 This function is called only during the real compile, not during the
2246 pre-compile.
2247
2248 Arguments:
2249 code points to start of the recursion
2250 endcode points to where to stop (current RECURSE item)
2251 bcptr points to the chain of current (unclosed) branch starts
2252 utf8 TRUE if in UTF-8 mode
2253 cd pointers to tables etc
2254
2255 Returns: TRUE if what is matched could be empty
2256 */
2257
2258 static BOOL
2259 could_be_empty(const uschar *code, const uschar *endcode, branch_chain *bcptr,
2260 BOOL utf8, compile_data *cd)
2261 {
2262 while (bcptr != NULL && bcptr->current_branch >= code)
2263 {
2264 if (!could_be_empty_branch(bcptr->current_branch, endcode, utf8, cd))
2265 return FALSE;
2266 bcptr = bcptr->outer;
2267 }
2268 return TRUE;
2269 }
2270
2271
2272
2273 /*************************************************
2274 * Check for POSIX class syntax *
2275 *************************************************/
2276
2277 /* This function is called when the sequence "[:" or "[." or "[=" is
2278 encountered in a character class. It checks whether this is followed by a
2279 sequence of characters terminated by a matching ":]" or ".]" or "=]". If we
2280 reach an unescaped ']' without the special preceding character, return FALSE.
2281
2282 Originally, this function only recognized a sequence of letters between the
2283 terminators, but it seems that Perl recognizes any sequence of characters,
2284 though of course unknown POSIX names are subsequently rejected. Perl gives an
2285 "Unknown POSIX class" error for [:f\oo:] for example, where previously PCRE
2286 didn't consider this to be a POSIX class. Likewise for [:1234:].
2287
2288 The problem in trying to be exactly like Perl is in the handling of escapes. We
2289 have to be sure that [abc[:x\]pqr] is *not* treated as containing a POSIX
2290 class, but [abc[:x\]pqr:]] is (so that an error can be generated). The code
2291 below handles the special case of \], but does not try to do any other escape
2292 processing. This makes it different from Perl for cases such as [:l\ower:]
2293 where Perl recognizes it as the POSIX class "lower" but PCRE does not recognize
2294 "l\ower". This is a lesser evil that not diagnosing bad classes when Perl does,
2295 I think.
2296
2297 A user pointed out that PCRE was rejecting [:a[:digit:]] whereas Perl was not.
2298 It seems that the appearance of a nested POSIX class supersedes an apparent
2299 external class. For example, [:a[:digit:]b:] matches "a", "b", ":", or
2300 a digit.
2301
2302 In Perl, unescaped square brackets may also appear as part of class names. For
2303 example, [:a[:abc]b:] gives unknown POSIX class "[:abc]b:]". However, for
2304 [:a[:abc]b][b:] it gives unknown POSIX class "[:abc]b][b:]", which does not
2305 seem right at all. PCRE does not allow closing square brackets in POSIX class
2306 names.
2307
2308 Arguments:
2309 ptr pointer to the initial [
2310 endptr where to return the end pointer
2311
2312 Returns: TRUE or FALSE
2313 */
2314
2315 static BOOL
2316 check_posix_syntax(const uschar *ptr, const uschar **endptr)
2317 {
2318 int terminator; /* Don't combine these lines; the Solaris cc */
2319 terminator = *(++ptr); /* compiler warns about "non-constant" initializer. */
2320 for (++ptr; *ptr != 0; ptr++)
2321 {
2322 if (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET)
2323 ptr++;
2324 else if (*ptr == CHAR_RIGHT_SQUARE_BRACKET) return FALSE;
2325 else
2326 {
2327 if (*ptr == terminator && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET)
2328 {
2329 *endptr = ptr;
2330 return TRUE;
2331 }
2332 if (*ptr == CHAR_LEFT_SQUARE_BRACKET &&
2333 (ptr[1] == CHAR_COLON || ptr[1] == CHAR_DOT ||
2334 ptr[1] == CHAR_EQUALS_SIGN) &&
2335 check_posix_syntax(ptr, endptr))
2336 return FALSE;
2337 }
2338 }
2339 return FALSE;
2340 }
2341
2342
2343
2344
2345 /*************************************************
2346 * Check POSIX class name *
2347 *************************************************/
2348
2349 /* This function is called to check the name given in a POSIX-style class entry
2350 such as [:alnum:].
2351
2352 Arguments:
2353 ptr points to the first letter
2354 len the length of the name
2355
2356 Returns: a value representing the name, or -1 if unknown
2357 */
2358
2359 static int
2360 check_posix_name(const uschar *ptr, int len)
2361 {
2362 const char *pn = posix_names;
2363 register int yield = 0;
2364 while (posix_name_lengths[yield] != 0)
2365 {
2366 if (len == posix_name_lengths[yield] &&
2367 strncmp((const char *)ptr, pn, len) == 0) return yield;
2368 pn += posix_name_lengths[yield] + 1;
2369 yield++;
2370 }
2371 return -1;
2372 }
2373
2374
2375 /*************************************************
2376 * Adjust OP_RECURSE items in repeated group *
2377 *************************************************/
2378
2379 /* OP_RECURSE items contain an offset from the start of the regex to the group
2380 that is referenced. This means that groups can be replicated for fixed
2381 repetition simply by copying (because the recursion is allowed to refer to
2382 earlier groups that are outside the current group). However, when a group is
2383 optional (i.e. the minimum quantifier is zero), OP_BRAZERO or OP_SKIPZERO is
2384 inserted before it, after it has been compiled. This means that any OP_RECURSE
2385 items within it that refer to the group itself or any contained groups have to
2386 have their offsets adjusted. That one of the jobs of this function. Before it
2387 is called, the partially compiled regex must be temporarily terminated with
2388 OP_END.
2389
2390 This function has been extended with the possibility of forward references for
2391 recursions and subroutine calls. It must also check the list of such references
2392 for the group we are dealing with. If it finds that one of the recursions in
2393 the current group is on this list, it adjusts the offset in the list, not the
2394 value in the reference (which is a group number).
2395
2396 Arguments:
2397 group points to the start of the group
2398 adjust the amount by which the group is to be moved
2399 utf8 TRUE in UTF-8 mode
2400 cd contains pointers to tables etc.
2401 save_hwm the hwm forward reference pointer at the start of the group
2402
2403 Returns: nothing
2404 */
2405
2406 static void
2407 adjust_recurse(uschar *group, int adjust, BOOL utf8, compile_data *cd,
2408 uschar *save_hwm)
2409 {
2410 uschar *ptr = group;
2411
2412 while ((ptr = (uschar *)find_recurse(ptr, utf8)) != NULL)
2413 {
2414 int offset;
2415 uschar *hc;
2416
2417 /* See if this recursion is on the forward reference list. If so, adjust the
2418 reference. */
2419
2420 for (hc = save_hwm; hc < cd->hwm; hc += LINK_SIZE)
2421 {
2422 offset = GET(hc, 0);
2423 if (cd->start_code + offset == ptr + 1)
2424 {
2425 PUT(hc, 0, offset + adjust);
2426 break;
2427 }
2428 }
2429
2430 /* Otherwise, adjust the recursion offset if it's after the start of this
2431 group. */
2432
2433 if (hc >= cd->hwm)
2434 {
2435 offset = GET(ptr, 1);
2436 if (cd->start_code + offset >= group) PUT(ptr, 1, offset + adjust);
2437 }
2438
2439 ptr += 1 + LINK_SIZE;
2440 }
2441 }
2442
2443
2444
2445 /*************************************************
2446 * Insert an automatic callout point *
2447 *************************************************/
2448
2449 /* This function is called when the PCRE_AUTO_CALLOUT option is set, to insert
2450 callout points before each pattern item.
2451
2452 Arguments:
2453 code current code pointer
2454 ptr current pattern pointer
2455 cd pointers to tables etc
2456
2457 Returns: new code pointer
2458 */
2459
2460 static uschar *
2461 auto_callout(uschar *code, const uschar *ptr, compile_data *cd)
2462 {
2463 *code++ = OP_CALLOUT;
2464 *code++ = 255;
2465 PUT(code, 0, (int)(ptr - cd->start_pattern)); /* Pattern offset */
2466 PUT(code, LINK_SIZE, 0); /* Default length */
2467 return code + 2*LINK_SIZE;
2468 }
2469
2470
2471
2472 /*************************************************
2473 * Complete a callout item *
2474 *************************************************/
2475
2476 /* A callout item contains the length of the next item in the pattern, which
2477 we can't fill in till after we have reached the relevant point. This is used
2478 for both automatic and manual callouts.
2479
2480 Arguments:
2481 previous_callout points to previous callout item
2482 ptr current pattern pointer
2483 cd pointers to tables etc
2484
2485 Returns: nothing
2486 */
2487
2488 static void
2489 complete_callout(uschar *previous_callout, const uschar *ptr, compile_data *cd)
2490 {
2491 int length = (int)(ptr - cd->start_pattern - GET(previous_callout, 2));
2492 PUT(previous_callout, 2 + LINK_SIZE, length);
2493 }
2494
2495
2496
2497 #ifdef SUPPORT_UCP
2498 /*************************************************
2499 * Get othercase range *
2500 *************************************************/
2501
2502 /* This function is passed the start and end of a class range, in UTF-8 mode
2503 with UCP support. It searches up the characters, looking for internal ranges of
2504 characters in the "other" case. Each call returns the next one, updating the
2505 start address.
2506
2507 Arguments:
2508 cptr points to starting character value; updated
2509 d end value
2510 ocptr where to put start of othercase range
2511 odptr where to put end of othercase range
2512
2513 Yield: TRUE when range returned; FALSE when no more
2514 */
2515
2516 static BOOL
2517 get_othercase_range(unsigned int *cptr, unsigned int d, unsigned int *ocptr,
2518 unsigned int *odptr)
2519 {
2520 unsigned int c, othercase, next;
2521
2522 for (c = *cptr; c <= d; c++)
2523 { if ((othercase = UCD_OTHERCASE(c)) != c) break; }
2524
2525 if (c > d) return FALSE;
2526
2527 *ocptr = othercase;
2528 next = othercase + 1;
2529
2530 for (++c; c <= d; c++)
2531 {
2532 if (UCD_OTHERCASE(c) != next) break;
2533 next++;
2534 }
2535
2536 *odptr = next - 1;
2537 *cptr = c;
2538
2539 return TRUE;
2540 }
2541
2542
2543
2544 /*************************************************
2545 * Check a character and a property *
2546 *************************************************/
2547
2548 /* This function is called by check_auto_possessive() when a property item
2549 is adjacent to a fixed character.
2550
2551 Arguments:
2552 c the character
2553 ptype the property type
2554 pdata the data for the type
2555 negated TRUE if it's a negated property (\P or \p{^)
2556
2557 Returns: TRUE if auto-possessifying is OK
2558 */
2559
2560 static BOOL
2561 check_char_prop(int c, int ptype, int pdata, BOOL negated)
2562 {
2563 const ucd_record *prop = GET_UCD(c);
2564 switch(ptype)
2565 {
2566 case PT_LAMP:
2567 return (prop->chartype == ucp_Lu ||
2568 prop->chartype == ucp_Ll ||
2569 prop->chartype == ucp_Lt) == negated;
2570
2571 case PT_GC:
2572 return (pdata == _pcre_ucp_gentype[prop->chartype]) == negated;
2573
2574 case PT_PC:
2575 return (pdata == prop->chartype) == negated;
2576
2577 case PT_SC:
2578 return (pdata == prop->script) == negated;
2579
2580 /* These are specials */
2581
2582 case PT_ALNUM:
2583 return (_pcre_ucp_gentype[prop->chartype] == ucp_L ||
2584 _pcre_ucp_gentype[prop->chartype] == ucp_N) == negated;
2585
2586 case PT_SPACE: /* Perl space */
2587 return (_pcre_ucp_gentype[prop->chartype] == ucp_Z ||
2588 c == CHAR_HT || c == CHAR_NL || c == CHAR_FF || c == CHAR_CR)
2589 == negated;
2590
2591 case PT_PXSPACE: /* POSIX space */
2592 return (_pcre_ucp_gentype[prop->chartype] == ucp_Z ||
2593 c == CHAR_HT || c == CHAR_NL || c == CHAR_VT ||
2594 c == CHAR_FF || c == CHAR_CR)
2595 == negated;
2596
2597 case PT_WORD:
2598 return (_pcre_ucp_gentype[prop->chartype] == ucp_L ||
2599 _pcre_ucp_gentype[prop->chartype] == ucp_N ||
2600 c == CHAR_UNDERSCORE) == negated;
2601 }
2602 return FALSE;
2603 }
2604 #endif /* SUPPORT_UCP */
2605
2606
2607
2608 /*************************************************
2609 * Check if auto-possessifying is possible *
2610 *************************************************/
2611
2612 /* This function is called for unlimited repeats of certain items, to see
2613 whether the next thing could possibly match the repeated item. If not, it makes
2614 sense to automatically possessify the repeated item.
2615
2616 Arguments:
2617 previous pointer to the repeated opcode
2618 utf8 TRUE in UTF-8 mode
2619 ptr next character in pattern
2620 options options bits
2621 cd contains pointers to tables etc.
2622
2623 Returns: TRUE if possessifying is wanted
2624 */
2625
2626 static BOOL
2627 check_auto_possessive(const uschar *previous, BOOL utf8, const uschar *ptr,
2628 int options, compile_data *cd)
2629 {
2630 int c, next;
2631 int op_code = *previous++;
2632
2633 /* Skip whitespace and comments in extended mode */
2634
2635 if ((options & PCRE_EXTENDED) != 0)
2636 {
2637 for (;;)
2638 {
2639 while ((cd->ctypes[*ptr] & ctype_space) != 0) ptr++;
2640 if (*ptr == CHAR_NUMBER_SIGN)
2641 {
2642 ptr++;
2643 while (*ptr != 0)
2644 {
2645 if (IS_NEWLINE(ptr)) { ptr += cd->nllen; break; }
2646 ptr++;
2647 #ifdef SUPPORT_UTF8
2648 if (utf8) while ((*ptr & 0xc0) == 0x80) ptr++;
2649 #endif
2650 }
2651 }
2652 else break;
2653 }
2654 }
2655
2656 /* If the next item is one that we can handle, get its value. A non-negative
2657 value is a character, a negative value is an escape value. */
2658
2659 if (*ptr == CHAR_BACKSLASH)
2660 {
2661 int temperrorcode = 0;
2662 next = check_escape(&ptr, &temperrorcode, cd->bracount, options, FALSE);
2663 if (temperrorcode != 0) return FALSE;
2664 ptr++; /* Point after the escape sequence */
2665 }
2666
2667 else if ((cd->ctypes[*ptr] & ctype_meta) == 0)
2668 {
2669 #ifdef SUPPORT_UTF8
2670 if (utf8) { GETCHARINC(next, ptr); } else
2671 #endif
2672 next = *ptr++;
2673 }
2674
2675 else return FALSE;
2676
2677 /* Skip whitespace and comments in extended mode */
2678
2679 if ((options & PCRE_EXTENDED) != 0)
2680 {
2681 for (;;)
2682 {
2683 while ((cd->ctypes[*ptr] & ctype_space) != 0) ptr++;
2684 if (*ptr == CHAR_NUMBER_SIGN)
2685 {
2686 ptr++;
2687 while (*ptr != 0)
2688 {
2689 if (IS_NEWLINE(ptr)) { ptr += cd->nllen; break; }
2690 ptr++;
2691 #ifdef SUPPORT_UTF8
2692 if (utf8) while ((*ptr & 0xc0) == 0x80) ptr++;
2693 #endif
2694 }
2695 }
2696 else break;
2697 }
2698 }
2699
2700 /* If the next thing is itself optional, we have to give up. */
2701
2702 if (*ptr == CHAR_ASTERISK || *ptr == CHAR_QUESTION_MARK ||
2703 strncmp((char *)ptr, STR_LEFT_CURLY_BRACKET STR_0 STR_COMMA, 3) == 0)
2704 return FALSE;
2705
2706 /* Now compare the next item with the previous opcode. First, handle cases when
2707 the next item is a character. */
2708
2709 if (next >= 0) switch(op_code)
2710 {
2711 case OP_CHAR:
2712 #ifdef SUPPORT_UTF8
2713 GETCHARTEST(c, previous);
2714 #else
2715 c = *previous;
2716 #endif
2717 return c != next;
2718
2719 /* For CHARI (caseless character) we must check the other case. If we have
2720 Unicode property support, we can use it to test the other case of
2721 high-valued characters. */
2722
2723 case OP_CHARI:
2724 #ifdef SUPPORT_UTF8
2725 GETCHARTEST(c, previous);
2726 #else
2727 c = *previous;
2728 #endif
2729 if (c == next) return FALSE;
2730 #ifdef SUPPORT_UTF8
2731 if (utf8)
2732 {
2733 unsigned int othercase;
2734 if (next < 128) othercase = cd->fcc[next]; else
2735 #ifdef SUPPORT_UCP
2736 othercase = UCD_OTHERCASE((unsigned int)next);
2737 #else
2738 othercase = NOTACHAR;
2739 #endif
2740 return (unsigned int)c != othercase;
2741 }
2742 else
2743 #endif /* SUPPORT_UTF8 */
2744 return (c != cd->fcc[next]); /* Non-UTF-8 mode */
2745
2746 /* For OP_NOT and OP_NOTI, the data is always a single-byte character. These
2747 opcodes are not used for multi-byte characters, because they are coded using
2748 an XCLASS instead. */
2749
2750 case OP_NOT:
2751 return (c = *previous) == next;
2752
2753 case OP_NOTI:
2754 if ((c = *previous) == next) return TRUE;
2755 #ifdef SUPPORT_UTF8
2756 if (utf8)
2757 {
2758 unsigned int othercase;
2759 if (next < 128) othercase = cd->fcc[next]; else
2760 #ifdef SUPPORT_UCP
2761 othercase = UCD_OTHERCASE(next);
2762 #else
2763 othercase = NOTACHAR;
2764 #endif
2765 return (unsigned int)c == othercase;
2766 }
2767 else
2768 #endif /* SUPPORT_UTF8 */
2769 return (c == cd->fcc[next]); /* Non-UTF-8 mode */
2770
2771 /* Note that OP_DIGIT etc. are generated only when PCRE_UCP is *not* set.
2772 When it is set, \d etc. are converted into OP_(NOT_)PROP codes. */
2773
2774 case OP_DIGIT:
2775 return next > 127 || (cd->ctypes[next] & ctype_digit) == 0;
2776
2777 case OP_NOT_DIGIT:
2778 return next <= 127 && (cd->ctypes[next] & ctype_digit) != 0;
2779
2780 case OP_WHITESPACE:
2781 return next > 127 || (cd->ctypes[next] & ctype_space) == 0;
2782
2783 case OP_NOT_WHITESPACE:
2784 return next <= 127 && (cd->ctypes[next] & ctype_space) != 0;
2785
2786 case OP_WORDCHAR:
2787 return next > 127 || (cd->ctypes[next] & ctype_word) == 0;
2788
2789 case OP_NOT_WORDCHAR:
2790 return next <= 127 && (cd->ctypes[next] & ctype_word) != 0;
2791
2792 case OP_HSPACE:
2793 case OP_NOT_HSPACE:
2794 switch(next)
2795 {
2796 case 0x09:
2797 case 0x20:
2798 case 0xa0:
2799 case 0x1680:
2800 case 0x180e:
2801 case 0x2000:
2802 case 0x2001:
2803 case 0x2002:
2804 case 0x2003:
2805 case 0x2004:
2806 case 0x2005:
2807 case 0x2006:
2808 case 0x2007:
2809 case 0x2008:
2810 case 0x2009:
2811 case 0x200A:
2812 case 0x202f:
2813 case 0x205f:
2814 case 0x3000:
2815 return op_code == OP_NOT_HSPACE;
2816 default:
2817 return op_code != OP_NOT_HSPACE;
2818 }
2819
2820 case OP_ANYNL:
2821 case OP_VSPACE:
2822 case OP_NOT_VSPACE:
2823 switch(next)
2824 {
2825 case 0x0a:
2826 case 0x0b:
2827 case 0x0c:
2828 case 0x0d:
2829 case 0x85:
2830 case 0x2028:
2831 case 0x2029:
2832 return op_code == OP_NOT_VSPACE;
2833 default:
2834 return op_code != OP_NOT_VSPACE;
2835 }
2836
2837 #ifdef SUPPORT_UCP
2838 case OP_PROP:
2839 return check_char_prop(next, previous[0], previous[1], FALSE);
2840
2841 case OP_NOTPROP:
2842 return check_char_prop(next, previous[0], previous[1], TRUE);
2843 #endif
2844
2845 default:
2846 return FALSE;
2847 }
2848
2849
2850 /* Handle the case when the next item is \d, \s, etc. Note that when PCRE_UCP
2851 is set, \d turns into ESC_du rather than ESC_d, etc., so ESC_d etc. are
2852 generated only when PCRE_UCP is *not* set, that is, when only ASCII
2853 characteristics are recognized. Similarly, the opcodes OP_DIGIT etc. are
2854 replaced by OP_PROP codes when PCRE_UCP is set. */
2855
2856 switch(op_code)
2857 {
2858 case OP_CHAR:
2859 case OP_CHARI:
2860 #ifdef SUPPORT_UTF8
2861 GETCHARTEST(c, previous);
2862 #else
2863 c = *previous;
2864 #endif
2865 switch(-next)
2866 {
2867 case ESC_d:
2868 return c > 127 || (cd->ctypes[c] & ctype_digit) == 0;
2869
2870 case ESC_D:
2871 return c <= 127 && (cd->ctypes[c] & ctype_digit) != 0;
2872
2873 case ESC_s:
2874 return c > 127 || (cd->ctypes[c] & ctype_space) == 0;
2875
2876 case ESC_S:
2877 return c <= 127 && (cd->ctypes[c] & ctype_space) != 0;
2878
2879 case ESC_w:
2880 return c > 127 || (cd->ctypes[c] & ctype_word) == 0;
2881
2882 case ESC_W:
2883 return c <= 127 && (cd->ctypes[c] & ctype_word) != 0;
2884
2885 case ESC_h:
2886 case ESC_H:
2887 switch(c)
2888 {
2889 case 0x09:
2890 case 0x20:
2891 case 0xa0:
2892 case 0x1680:
2893 case 0x180e:
2894 case 0x2000:
2895 case 0x2001:
2896 case 0x2002:
2897 case 0x2003:
2898 case 0x2004:
2899 case 0x2005:
2900 case 0x2006:
2901 case 0x2007:
2902 case 0x2008:
2903 case 0x2009:
2904 case 0x200A:
2905 case 0x202f:
2906 case 0x205f:
2907 case 0x3000:
2908 return -next != ESC_h;
2909 default:
2910 return -next == ESC_h;
2911 }
2912
2913 case ESC_v:
2914 case ESC_V:
2915 switch(c)
2916 {
2917 case 0x0a:
2918 case 0x0b:
2919 case 0x0c:
2920 case 0x0d:
2921 case 0x85:
2922 case 0x2028:
2923 case 0x2029:
2924 return -next != ESC_v;
2925 default:
2926 return -next == ESC_v;
2927 }
2928
2929 /* When PCRE_UCP is set, these values get generated for \d etc. Find
2930 their substitutions and process them. The result will always be either
2931 -ESC_p or -ESC_P. Then fall through to process those values. */
2932
2933 #ifdef SUPPORT_UCP
2934 case ESC_du:
2935 case ESC_DU:
2936 case ESC_wu:
2937 case ESC_WU:
2938 case ESC_su:
2939 case ESC_SU:
2940 {
2941 int temperrorcode = 0;
2942 ptr = substitutes[-next - ESC_DU];
2943 next = check_escape(&ptr, &temperrorcode, 0, options, FALSE);
2944 if (temperrorcode != 0) return FALSE;
2945 ptr++; /* For compatibility */
2946 }
2947 /* Fall through */
2948
2949 case ESC_p:
2950 case ESC_P:
2951 {
2952 int ptype, pdata, errorcodeptr;
2953 BOOL negated;
2954
2955 ptr--; /* Make ptr point at the p or P */
2956 ptype = get_ucp(&ptr, &negated, &pdata, &errorcodeptr);
2957 if (ptype < 0) return FALSE;
2958 ptr++; /* Point past the final curly ket */
2959
2960 /* If the property item is optional, we have to give up. (When generated
2961 from \d etc by PCRE_UCP, this test will have been applied much earlier,
2962 to the original \d etc. At this point, ptr will point to a zero byte. */
2963
2964 if (*ptr == CHAR_ASTERISK || *ptr == CHAR_QUESTION_MARK ||
2965 strncmp((char *)ptr, STR_LEFT_CURLY_BRACKET STR_0 STR_COMMA, 3) == 0)
2966 return FALSE;
2967
2968 /* Do the property check. */
2969
2970 return check_char_prop(c, ptype, pdata, (next == -ESC_P) != negated);
2971 }
2972 #endif
2973
2974 default:
2975 return FALSE;
2976 }
2977
2978 /* In principle, support for Unicode properties should be integrated here as
2979 well. It means re-organizing the above code so as to get hold of the property
2980 values before switching on the op-code. However, I wonder how many patterns
2981 combine ASCII \d etc with Unicode properties? (Note that if PCRE_UCP is set,
2982 these op-codes are never generated.) */
2983
2984 case OP_DIGIT:
2985 return next == -ESC_D || next == -ESC_s || next == -ESC_W ||
2986 next == -ESC_h || next == -ESC_v || next == -ESC_R;
2987
2988 case OP_NOT_DIGIT:
2989 return next == -ESC_d;
2990
2991 case OP_WHITESPACE:
2992 return next == -ESC_S || next == -ESC_d || next == -ESC_w || next == -ESC_R;
2993
2994 case OP_NOT_WHITESPACE:
2995 return next == -ESC_s || next == -ESC_h || next == -ESC_v;
2996
2997 case OP_HSPACE:
2998 return next == -ESC_S || next == -ESC_H || next == -ESC_d ||
2999 next == -ESC_w || next == -ESC_v || next == -ESC_R;
3000
3001 case OP_NOT_HSPACE:
3002 return next == -ESC_h;
3003
3004 /* Can't have \S in here because VT matches \S (Perl anomaly) */
3005 case OP_ANYNL:
3006 case OP_VSPACE:
3007 return next == -ESC_V || next == -ESC_d || next == -ESC_w;
3008
3009 case OP_NOT_VSPACE:
3010 return next == -ESC_v || next == -ESC_R;
3011
3012 case OP_WORDCHAR:
3013 return next == -ESC_W || next == -ESC_s || next == -ESC_h ||
3014 next == -ESC_v || next == -ESC_R;
3015
3016 case OP_NOT_WORDCHAR:
3017 return next == -ESC_w || next == -ESC_d;
3018
3019 default:
3020 return FALSE;
3021 }
3022
3023 /* Control does not reach here */
3024 }
3025
3026
3027
3028 /*************************************************
3029 * Compile one branch *
3030 *************************************************/
3031
3032 /* Scan the pattern, compiling it into the a vector. If the options are
3033 changed during the branch, the pointer is used to change the external options
3034 bits. This function is used during the pre-compile phase when we are trying
3035 to find out the amount of memory needed, as well as during the real compile
3036 phase. The value of lengthptr distinguishes the two phases.
3037
3038 Arguments:
3039 optionsptr pointer to the option bits
3040 codeptr points to the pointer to the current code point
3041 ptrptr points to the current pattern pointer
3042 errorcodeptr points to error code variable
3043 firstbyteptr set to initial literal character, or < 0 (REQ_UNSET, REQ_NONE)
3044 reqbyteptr set to the last literal character required, else < 0
3045 bcptr points to current branch chain
3046 cond_depth conditional nesting depth
3047 cd contains pointers to tables etc.
3048 lengthptr NULL during the real compile phase
3049 points to length accumulator during pre-compile phase
3050
3051 Returns: TRUE on success
3052 FALSE, with *errorcodeptr set non-zero on error
3053 */
3054
3055 static BOOL
3056 compile_branch(int *optionsptr, uschar **codeptr, const uschar **ptrptr,
3057 int *errorcodeptr, int *firstbyteptr, int *reqbyteptr, branch_chain *bcptr,
3058 int cond_depth, compile_data *cd, int *lengthptr)
3059 {
3060 int repeat_type, op_type;
3061 int repeat_min = 0, repeat_max = 0; /* To please picky compilers */
3062 int bravalue = 0;
3063 int greedy_default, greedy_non_default;
3064 int firstbyte, reqbyte;
3065 int zeroreqbyte, zerofirstbyte;
3066 int req_caseopt, reqvary, tempreqvary;
3067 int options = *optionsptr; /* May change dynamically */
3068 int after_manual_callout = 0;
3069 int length_prevgroup = 0;
3070 register int c;
3071 register uschar *code = *codeptr;
3072 uschar *last_code = code;
3073 uschar *orig_code = code;
3074 uschar *tempcode;
3075 BOOL inescq = FALSE;
3076 BOOL groupsetfirstbyte = FALSE;
3077 const uschar *ptr = *ptrptr;
3078 const uschar *tempptr;
3079 const uschar *nestptr = NULL;
3080 uschar *previous = NULL;
3081 uschar *previous_callout = NULL;
3082 uschar *save_hwm = NULL;
3083 uschar classbits[32];
3084
3085 /* We can fish out the UTF-8 setting once and for all into a BOOL, but we
3086 must not do this for other options (e.g. PCRE_EXTENDED) because they may change
3087 dynamically as we process the pattern. */
3088
3089 #ifdef SUPPORT_UTF8
3090 BOOL class_utf8;
3091 BOOL utf8 = (options & PCRE_UTF8) != 0;
3092 uschar *class_utf8data;
3093 uschar *class_utf8data_base;
3094 uschar utf8_char[6];
3095 #else
3096 BOOL utf8 = FALSE;
3097 #endif
3098
3099 #ifdef PCRE_DEBUG
3100 if (lengthptr != NULL) DPRINTF((">> start branch\n"));
3101 #endif
3102
3103 /* Set up the default and non-default settings for greediness */
3104
3105 greedy_default = ((options & PCRE_UNGREEDY) != 0);
3106 greedy_non_default = greedy_default ^ 1;
3107
3108 /* Initialize no first byte, no required byte. REQ_UNSET means "no char
3109 matching encountered yet". It gets changed to REQ_NONE if we hit something that
3110 matches a non-fixed char first char; reqbyte just remains unset if we never
3111 find one.
3112
3113 When we hit a repeat whose minimum is zero, we may have to adjust these values
3114 to take the zero repeat into account. This is implemented by setting them to
3115 zerofirstbyte and zeroreqbyte when such a repeat is encountered. The individual
3116 item types that can be repeated set these backoff variables appropriately. */
3117
3118 firstbyte = reqbyte = zerofirstbyte = zeroreqbyte = REQ_UNSET;
3119
3120 /* The variable req_caseopt contains either the REQ_CASELESS value or zero,
3121 according to the current setting of the caseless flag. REQ_CASELESS is a bit
3122 value > 255. It is added into the firstbyte or reqbyte variables to record the
3123 case status of the value. This is used only for ASCII characters. */
3124
3125 req_caseopt = ((options & PCRE_CASELESS) != 0)? REQ_CASELESS : 0;
3126
3127 /* Switch on next character until the end of the branch */
3128
3129 for (;; ptr++)
3130 {
3131 BOOL negate_class;
3132 BOOL should_flip_negation;
3133 BOOL possessive_quantifier;
3134 BOOL is_quantifier;
3135 BOOL is_recurse;
3136 BOOL reset_bracount;
3137 int class_charcount;
3138 int class_lastchar;
3139 int newoptions;
3140 int recno;
3141 int refsign;
3142 int skipbytes;
3143 int subreqbyte;
3144 int subfirstbyte;
3145 int terminator;
3146 int mclength;
3147 int tempbracount;
3148 uschar mcbuffer[8];
3149
3150 /* Get next byte in the pattern */
3151
3152 c = *ptr;
3153
3154 /* If we are at the end of a nested substitution, revert to the outer level
3155 string. Nesting only happens one level deep. */
3156
3157 if (c == 0 && nestptr != NULL)
3158 {
3159 ptr = nestptr;
3160 nestptr = NULL;
3161 c = *ptr;
3162 }
3163
3164 /* If we are in the pre-compile phase, accumulate the length used for the
3165 previous cycle of this loop. */
3166
3167 if (lengthptr != NULL)
3168 {
3169 #ifdef PCRE_DEBUG
3170 if (code > cd->hwm) cd->hwm = code; /* High water info */
3171 #endif
3172 if (code > cd->start_workspace + WORK_SIZE_CHECK) /* Check for overrun */
3173 {
3174 *errorcodeptr = ERR52;
3175 goto FAILED;
3176 }
3177
3178 /* There is at least one situation where code goes backwards: this is the
3179 case of a zero quantifier after a class (e.g. [ab]{0}). At compile time,
3180 the class is simply eliminated. However, it is created first, so we have to
3181 allow memory for it. Therefore, don't ever reduce the length at this point.
3182 */
3183
3184 if (code < last_code) code = last_code;
3185
3186 /* Paranoid check for integer overflow */
3187
3188 if (OFLOW_MAX - *lengthptr < code - last_code)
3189 {
3190 *errorcodeptr = ERR20;
3191 goto FAILED;
3192 }
3193
3194 *lengthptr += (int)(code - last_code);
3195 DPRINTF(("length=%d added %d c=%c\n", *lengthptr, code - last_code, c));
3196
3197 /* If "previous" is set and it is not at the start of the work space, move
3198 it back to there, in order to avoid filling up the work space. Otherwise,
3199 if "previous" is NULL, reset the current code pointer to the start. */
3200
3201 if (previous != NULL)
3202 {
3203 if (previous > orig_code)
3204 {
3205 memmove(orig_code, previous, code - previous);
3206 code -= previous - orig_code;
3207 previous = orig_code;
3208 }
3209 }
3210 else code = orig_code;
3211
3212 /* Remember where this code item starts so we can pick up the length
3213 next time round. */
3214
3215 last_code = code;
3216 }
3217
3218 /* In the real compile phase, just check the workspace used by the forward
3219 reference list. */
3220
3221 else if (cd->hwm > cd->start_workspace + WORK_SIZE_CHECK)
3222 {
3223 *errorcodeptr = ERR52;
3224 goto FAILED;
3225 }
3226
3227 /* If in \Q...\E, check for the end; if not, we have a literal */
3228
3229 if (inescq && c != 0)
3230 {
3231 if (c == CHAR_BACKSLASH && ptr[1] == CHAR_E)
3232 {
3233 inescq = FALSE;
3234 ptr++;
3235 continue;
3236 }
3237 else
3238 {
3239 if (previous_callout != NULL)
3240 {
3241 if (lengthptr == NULL) /* Don't attempt in pre-compile phase */
3242 complete_callout(previous_callout, ptr, cd);
3243 previous_callout = NULL;
3244 }
3245 if ((options & PCRE_AUTO_CALLOUT) != 0)
3246 {
3247 previous_callout = code;
3248 code = auto_callout(code, ptr, cd);
3249 }
3250 goto NORMAL_CHAR;
3251 }
3252 }
3253
3254 /* Fill in length of a previous callout, except when the next thing is
3255 a quantifier. */
3256
3257 is_quantifier =
3258 c == CHAR_ASTERISK || c == CHAR_PLUS || c == CHAR_QUESTION_MARK ||
3259 (c == CHAR_LEFT_CURLY_BRACKET && is_counted_repeat(ptr+1));
3260
3261 if (!is_quantifier && previous_callout != NULL &&
3262 after_manual_callout-- <= 0)
3263 {
3264 if (lengthptr == NULL) /* Don't attempt in pre-compile phase */
3265 complete_callout(previous_callout, ptr, cd);
3266 previous_callout = NULL;
3267 }
3268
3269 /* In extended mode, skip white space and comments. */
3270
3271 if ((options & PCRE_EXTENDED) != 0)
3272 {
3273 if ((cd->ctypes[c] & ctype_space) != 0) continue;
3274 if (c == CHAR_NUMBER_SIGN)
3275 {
3276 ptr++;
3277 while (*ptr != 0)
3278 {
3279 if (IS_NEWLINE(ptr)) { ptr += cd->nllen - 1; break; }
3280 ptr++;
3281 #ifdef SUPPORT_UTF8
3282 if (utf8) while ((*ptr & 0xc0) == 0x80) ptr++;
3283 #endif
3284 }
3285 if (*ptr != 0) continue;
3286
3287 /* Else fall through to handle end of string */
3288 c = 0;
3289 }
3290 }
3291
3292 /* No auto callout for quantifiers. */
3293
3294 if ((options & PCRE_AUTO_CALLOUT) != 0 && !is_quantifier)
3295 {
3296 previous_callout = code;
3297 code = auto_callout(code, ptr, cd);
3298 }
3299
3300 switch(c)
3301 {
3302 /* ===================================================================*/
3303 case 0: /* The branch terminates at string end */
3304 case CHAR_VERTICAL_LINE: /* or | or ) */
3305 case CHAR_RIGHT_PARENTHESIS:
3306 *firstbyteptr = firstbyte;
3307 *reqbyteptr = reqbyte;
3308 *codeptr = code;
3309 *ptrptr = ptr;
3310 if (lengthptr != NULL)
3311 {
3312 if (OFLOW_MAX - *lengthptr < code - last_code)
3313 {
3314 *errorcodeptr = ERR20;
3315 goto FAILED;
3316 }
3317 *lengthptr += (int)(code - last_code); /* To include callout length */
3318 DPRINTF((">> end branch\n"));
3319 }
3320 return TRUE;
3321
3322
3323 /* ===================================================================*/
3324 /* Handle single-character metacharacters. In multiline mode, ^ disables
3325 the setting of any following char as a first character. */
3326
3327 case CHAR_CIRCUMFLEX_ACCENT:
3328 previous = NULL;
3329 if ((options & PCRE_MULTILINE) != 0)
3330 {
3331 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
3332 *code++ = OP_CIRCM;
3333 }
3334 else *code++ = OP_CIRC;
3335 break;
3336
3337 case CHAR_DOLLAR_SIGN:
3338 previous = NULL;
3339 *code++ = ((options & PCRE_MULTILINE) != 0)? OP_DOLLM : OP_DOLL;
3340 break;
3341
3342 /* There can never be a first char if '.' is first, whatever happens about
3343 repeats. The value of reqbyte doesn't change either. */
3344
3345 case CHAR_DOT:
3346 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
3347 zerofirstbyte = firstbyte;
3348 zeroreqbyte = reqbyte;
3349 previous = code;
3350 *code++ = ((options & PCRE_DOTALL) != 0)? OP_ALLANY: OP_ANY;
3351 break;
3352
3353
3354 /* ===================================================================*/
3355 /* Character classes. If the included characters are all < 256, we build a
3356 32-byte bitmap of the permitted characters, except in the special case
3357 where there is only one such character. For negated classes, we build the
3358 map as usual, then invert it at the end. However, we use a different opcode
3359 so that data characters > 255 can be handled correctly.
3360
3361 If the class contains characters outside the 0-255 range, a different
3362 opcode is compiled. It may optionally have a bit map for characters < 256,
3363 but those above are are explicitly listed afterwards. A flag byte tells
3364 whether the bitmap is present, and whether this is a negated class or not.
3365
3366 In JavaScript compatibility mode, an isolated ']' causes an error. In
3367 default (Perl) mode, it is treated as a data character. */
3368
3369 case CHAR_RIGHT_SQUARE_BRACKET:
3370 if ((cd->external_options & PCRE_JAVASCRIPT_COMPAT) != 0)
3371 {
3372 *errorcodeptr = ERR64;
3373 goto FAILED;
3374 }
3375 goto NORMAL_CHAR;
3376
3377 case CHAR_LEFT_SQUARE_BRACKET:
3378 previous = code;
3379
3380 /* PCRE supports POSIX class stuff inside a class. Perl gives an error if
3381 they are encountered at the top level, so we'll do that too. */
3382
3383 if ((ptr[1] == CHAR_COLON || ptr[1] == CHAR_DOT ||
3384 ptr[1] == CHAR_EQUALS_SIGN) &&
3385 check_posix_syntax(ptr, &tempptr))
3386 {
3387 *errorcodeptr = (ptr[1] == CHAR_COLON)? ERR13 : ERR31;
3388 goto FAILED;
3389 }
3390
3391 /* If the first character is '^', set the negation flag and skip it. Also,
3392 if the first few characters (either before or after ^) are \Q\E or \E we
3393 skip them too. This makes for compatibility with Perl. */
3394
3395 negate_class = FALSE;
3396 for (;;)
3397 {
3398 c = *(++ptr);
3399 if (c == CHAR_BACKSLASH)
3400 {
3401 if (ptr[1] == CHAR_E)
3402 ptr++;
3403 else if (strncmp((const char *)ptr+1,
3404 STR_Q STR_BACKSLASH STR_E, 3) == 0)
3405 ptr += 3;
3406 else
3407 break;
3408 }
3409 else if (!negate_class && c == CHAR_CIRCUMFLEX_ACCENT)
3410 negate_class = TRUE;
3411 else break;
3412 }
3413
3414 /* Empty classes are allowed in JavaScript compatibility mode. Otherwise,
3415 an initial ']' is taken as a data character -- the code below handles
3416 that. In JS mode, [] must always fail, so generate OP_FAIL, whereas
3417 [^] must match any character, so generate OP_ALLANY. */
3418
3419 if (c == CHAR_RIGHT_SQUARE_BRACKET &&
3420 (cd->external_options & PCRE_JAVASCRIPT_COMPAT) != 0)
3421 {
3422 *code++ = negate_class? OP_ALLANY : OP_FAIL;
3423 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
3424 zerofirstbyte = firstbyte;
3425 break;
3426 }
3427
3428 /* If a class contains a negative special such as \S, we need to flip the
3429 negation flag at the end, so that support for characters > 255 works
3430 correctly (they are all included in the class). */
3431
3432 should_flip_negation = FALSE;
3433
3434 /* Keep a count of chars with values < 256 so that we can optimize the case
3435 of just a single character (as long as it's < 256). However, For higher
3436 valued UTF-8 characters, we don't yet do any optimization. */
3437
3438 class_charcount = 0;
3439 class_lastchar = -1;
3440
3441 /* Initialize the 32-char bit map to all zeros. We build the map in a
3442 temporary bit of memory, in case the class contains only 1 character (less
3443 than 256), because in that case the compiled code doesn't use the bit map.
3444 */
3445
3446 memset(classbits, 0, 32 * sizeof(uschar));
3447
3448 #ifdef SUPPORT_UTF8
3449 class_utf8 = FALSE; /* No chars >= 256 */
3450 class_utf8data = code + LINK_SIZE + 2; /* For UTF-8 items */
3451 class_utf8data_base = class_utf8data; /* For resetting in pass 1 */
3452 #endif
3453
3454 /* Process characters until ] is reached. By writing this as a "do" it
3455 means that an initial ] is taken as a data character. At the start of the
3456 loop, c contains the first byte of the character. */
3457
3458 if (c != 0) do
3459 {
3460 const uschar *oldptr;
3461
3462 #ifdef SUPPORT_UTF8
3463 if (utf8 && c > 127)
3464 { /* Braces are required because the */
3465 GETCHARLEN(c, ptr, ptr); /* macro generates multiple statements */
3466 }
3467
3468 /* In the pre-compile phase, accumulate the length of any UTF-8 extra
3469 data and reset the pointer. This is so that very large classes that
3470 contain a zillion UTF-8 characters no longer overwrite the work space
3471 (which is on the stack). */
3472
3473 if (lengthptr != NULL)
3474 {
3475 *lengthptr += class_utf8data - class_utf8data_base;
3476 class_utf8data = class_utf8data_base;
3477 }
3478
3479 #endif
3480
3481 /* Inside \Q...\E everything is literal except \E */
3482
3483 if (inescq)
3484 {
3485 if (c == CHAR_BACKSLASH && ptr[1] == CHAR_E) /* If we are at \E */
3486 {
3487 inescq = FALSE; /* Reset literal state */
3488 ptr++; /* Skip the 'E' */
3489 continue; /* Carry on with next */
3490 }
3491 goto CHECK_RANGE; /* Could be range if \E follows */
3492 }
3493
3494 /* Handle POSIX class names. Perl allows a negation extension of the
3495 form [:^name:]. A square bracket that doesn't match the syntax is
3496 treated as a literal. We also recognize the POSIX constructions
3497 [.ch.] and [=ch=] ("collating elements") and fault them, as Perl
3498 5.6 and 5.8 do. */
3499
3500 if (c == CHAR_LEFT_SQUARE_BRACKET &&
3501 (ptr[1] == CHAR_COLON || ptr[1] == CHAR_DOT ||
3502 ptr[1] == CHAR_EQUALS_SIGN) && check_posix_syntax(ptr, &tempptr))
3503 {
3504 BOOL local_negate = FALSE;
3505 int posix_class, taboffset, tabopt;
3506 register const uschar *cbits = cd->cbits;
3507 uschar pbits[32];
3508
3509 if (ptr[1] != CHAR_COLON)
3510 {
3511 *errorcodeptr = ERR31;
3512 goto FAILED;
3513 }
3514
3515 ptr += 2;
3516 if (*ptr == CHAR_CIRCUMFLEX_ACCENT)
3517 {
3518 local_negate = TRUE;
3519 should_flip_negation = TRUE; /* Note negative special */
3520 ptr++;
3521 }
3522
3523 posix_class = check_posix_name(ptr, (int)(tempptr - ptr));
3524 if (posix_class < 0)
3525 {
3526 *errorcodeptr = ERR30;
3527 goto FAILED;
3528 }
3529
3530 /* If matching is caseless, upper and lower are converted to
3531 alpha. This relies on the fact that the class table starts with
3532 alpha, lower, upper as the first 3 entries. */
3533
3534 if ((options & PCRE_CASELESS) != 0 && posix_class <= 2)
3535 posix_class = 0;
3536
3537 /* When PCRE_UCP is set, some of the POSIX classes are converted to
3538 different escape sequences that use Unicode properties. */
3539
3540 #ifdef SUPPORT_UCP
3541 if ((options & PCRE_UCP) != 0)
3542 {
3543 int pc = posix_class + ((local_negate)? POSIX_SUBSIZE/2 : 0);
3544 if (posix_substitutes[pc] != NULL)
3545 {
3546 nestptr = tempptr + 1;
3547 ptr = posix_substitutes[pc] - 1;
3548 continue;
3549 }
3550 }
3551 #endif
3552 /* In the non-UCP case, we build the bit map for the POSIX class in a
3553 chunk of local store because we may be adding and subtracting from it,
3554 and we don't want to subtract bits that may be in the main map already.
3555 At the end we or the result into the bit map that is being built. */
3556
3557 posix_class *= 3;
3558
3559 /* Copy in the first table (always present) */
3560
3561 memcpy(pbits, cbits + posix_class_maps[posix_class],
3562 32 * sizeof(uschar));
3563
3564 /* If there is a second table, add or remove it as required. */
3565
3566 taboffset = posix_class_maps[posix_class + 1];
3567 tabopt = posix_class_maps[posix_class + 2];
3568
3569 if (taboffset >= 0)
3570 {
3571 if (tabopt >= 0)
3572 for (c = 0; c < 32; c++) pbits[c] |= cbits[c + taboffset];
3573 else
3574 for (c = 0; c < 32; c++) pbits[c] &= ~cbits[c + taboffset];
3575 }
3576
3577 /* Not see if we need to remove any special characters. An option
3578 value of 1 removes vertical space and 2 removes underscore. */
3579
3580 if (tabopt < 0) tabopt = -tabopt;
3581 if (tabopt == 1) pbits[1] &= ~0x3c;
3582 else if (tabopt == 2) pbits[11] &= 0x7f;
3583
3584 /* Add the POSIX table or its complement into the main table that is
3585 being built and we are done. */
3586
3587 if (local_negate)
3588 for (c = 0; c < 32; c++) classbits[c] |= ~pbits[c];
3589 else
3590 for (c = 0; c < 32; c++) classbits[c] |= pbits[c];
3591
3592 ptr = tempptr + 1;
3593 class_charcount = 10; /* Set > 1; assumes more than 1 per class */
3594 continue; /* End of POSIX syntax handling */
3595 }
3596
3597 /* Backslash may introduce a single character, or it may introduce one
3598 of the specials, which just set a flag. The sequence \b is a special
3599 case. Inside a class (and only there) it is treated as backspace. We
3600 assume that other escapes have more than one character in them, so set
3601 class_charcount bigger than one. Unrecognized escapes fall through and
3602 are either treated as literal characters (by default), or are faulted if
3603 PCRE_EXTRA is set. */
3604
3605 if (c == CHAR_BACKSLASH)
3606 {
3607 c = check_escape(&ptr, errorcodeptr, cd->bracount, options, TRUE);
3608 if (*errorcodeptr != 0) goto FAILED;
3609
3610 if (-c == ESC_b) c = CHAR_BS; /* \b is backspace in a class */
3611 else if (-c == ESC_Q) /* Handle start of quoted string */
3612 {
3613 if (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
3614 {
3615 ptr += 2; /* avoid empty string */
3616 }
3617 else inescq = TRUE;
3618 continue;
3619 }
3620 else if (-c == ESC_E) continue; /* Ignore orphan \E */
3621
3622 if (c < 0)
3623 {
3624 register const uschar *cbits = cd->cbits;
3625 class_charcount += 2; /* Greater than 1 is what matters */
3626
3627 switch (-c)
3628 {
3629 #ifdef SUPPORT_UCP
3630 case ESC_du: /* These are the values given for \d etc */
3631 case ESC_DU: /* when PCRE_UCP is set. We replace the */
3632 case ESC_wu: /* escape sequence with an appropriate \p */
3633 case ESC_WU: /* or \P to test Unicode properties instead */
3634 case ESC_su: /* of the default ASCII testing. */
3635 case ESC_SU:
3636 nestptr = ptr;
3637 ptr = substitutes[-c - ESC_DU] - 1; /* Just before substitute */
3638 class_charcount -= 2; /* Undo! */
3639 continue;
3640 #endif
3641 case ESC_d:
3642 for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_digit];
3643 continue;
3644
3645 case ESC_D:
3646 should_flip_negation = TRUE;
3647 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_digit];
3648 continue;
3649
3650 case ESC_w:
3651 for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_word];
3652 continue;
3653
3654 case ESC_W:
3655 should_flip_negation = TRUE;
3656 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_word];
3657 continue;
3658
3659 /* Perl 5.004 onwards omits VT from \s, but we must preserve it
3660 if it was previously set by something earlier in the character
3661 class. */
3662
3663 case ESC_s:
3664 classbits[0] |= cbits[cbit_space];
3665 classbits[1] |= cbits[cbit_space+1] & ~0x08;
3666 for (c = 2; c < 32; c++) classbits[c] |= cbits[c+cbit_space];
3667 continue;
3668
3669 case ESC_S:
3670 should_flip_negation = TRUE;
3671 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_space];
3672 classbits[1] |= 0x08; /* Perl 5.004 onwards omits VT from \s */
3673 continue;
3674
3675 case ESC_h:
3676 SETBIT(classbits, 0x09); /* VT */
3677 SETBIT(classbits, 0x20); /* SPACE */
3678 SETBIT(classbits, 0xa0); /* NSBP */
3679 #ifdef SUPPORT_UTF8
3680 if (utf8)
3681 {
3682 class_utf8 = TRUE;
3683 *class_utf8data++ = XCL_SINGLE;
3684 class_utf8data += _pcre_ord2utf8(0x1680, class_utf8data);
3685 *class_utf8data++ = XCL_SINGLE;
3686 class_utf8data += _pcre_ord2utf8(0x180e, class_utf8data);
3687 *class_utf8data++ = XCL_RANGE;
3688 class_utf8data += _pcre_ord2utf8(0x2000, class_utf8data);
3689 class_utf8data += _pcre_ord2utf8(0x200A, class_utf8data);
3690 *class_utf8data++ = XCL_SINGLE;
3691 class_utf8data += _pcre_ord2utf8(0x202f, class_utf8data);
3692 *class_utf8data++ = XCL_SINGLE;
3693 class_utf8data += _pcre_ord2utf8(0x205f, class_utf8data);
3694 *class_utf8data++ = XCL_SINGLE;
3695 class_utf8data += _pcre_ord2utf8(0x3000, class_utf8data);
3696 }
3697 #endif
3698 continue;
3699
3700 case ESC_H:
3701 for (c = 0; c < 32; c++)
3702 {
3703 int x = 0xff;
3704 switch (c)
3705 {
3706 case 0x09/8: x ^= 1 << (0x09%8); break;
3707 case 0x20/8: x ^= 1 << (0x20%8); break;
3708 case 0xa0/8: x ^= 1 << (0xa0%8); break;
3709 default: break;
3710 }
3711 classbits[c] |= x;
3712 }
3713
3714 #ifdef SUPPORT_UTF8
3715 if (utf8)
3716 {
3717 class_utf8 = TRUE;
3718 *class_utf8data++ = XCL_RANGE;
3719 class_utf8data += _pcre_ord2utf8(0x0100, class_utf8data);
3720 class_utf8data += _pcre_ord2utf8(0x167f, class_utf8data);
3721 *class_utf8data++ = XCL_RANGE;
3722 class_utf8data += _pcre_ord2utf8(0x1681, class_utf8data);
3723 class_utf8data += _pcre_ord2utf8(0x180d, class_utf8data);
3724 *class_utf8data++ = XCL_RANGE;
3725 class_utf8data += _pcre_ord2utf8(0x180f, class_utf8data);
3726 class_utf8data += _pcre_ord2utf8(0x1fff, class_utf8data);
3727 *class_utf8data++ = XCL_RANGE;
3728 class_utf8data += _pcre_ord2utf8(0x200B, class_utf8data);
3729 class_utf8data += _pcre_ord2utf8(0x202e, class_utf8data);
3730 *class_utf8data++ = XCL_RANGE;
3731 class_utf8data += _pcre_ord2utf8(0x2030, class_utf8data);
3732 class_utf8data += _pcre_ord2utf8(0x205e, class_utf8data);
3733 *class_utf8data++ = XCL_RANGE;
3734 class_utf8data += _pcre_ord2utf8(0x2060, class_utf8data);
3735 class_utf8data += _pcre_ord2utf8(0x2fff, class_utf8data);
3736 *class_utf8data++ = XCL_RANGE;
3737 class_utf8data += _pcre_ord2utf8(0x3001, class_utf8data);
3738 class_utf8data += _pcre_ord2utf8(0x7fffffff, class_utf8data);
3739 }
3740 #endif
3741 continue;
3742
3743 case ESC_v:
3744 SETBIT(classbits, 0x0a); /* LF */
3745 SETBIT(classbits, 0x0b); /* VT */
3746 SETBIT(classbits, 0x0c); /* FF */
3747 SETBIT(classbits, 0x0d); /* CR */
3748 SETBIT(classbits, 0x85); /* NEL */
3749 #ifdef SUPPORT_UTF8
3750 if (utf8)
3751 {
3752 class_utf8 = TRUE;
3753 *class_utf8data++ = XCL_RANGE;
3754 class_utf8data += _pcre_ord2utf8(0x2028, class_utf8data);
3755 class_utf8data += _pcre_ord2utf8(0x2029, class_utf8data);
3756 }
3757 #endif
3758 continue;
3759
3760 case ESC_V:
3761 for (c = 0; c < 32; c++)
3762 {
3763 int x = 0xff;
3764 switch (c)
3765 {
3766 case 0x0a/8: x ^= 1 << (0x0a%8);
3767 x ^= 1 << (0x0b%8);
3768 x ^= 1 << (0x0c%8);
3769 x ^= 1 << (0x0d%8);
3770 break;
3771 case 0x85/8: x ^= 1 << (0x85%8); break;
3772 default: break;
3773 }
3774 classbits[c] |= x;
3775 }
3776
3777 #ifdef SUPPORT_UTF8
3778 if (utf8)
3779 {
3780 class_utf8 = TRUE;
3781 *class_utf8data++ = XCL_RANGE;
3782 class_utf8data += _pcre_ord2utf8(0x0100, class_utf8data);
3783 class_utf8data += _pcre_ord2utf8(0x2027, class_utf8data);
3784 *class_utf8data++ = XCL_RANGE;
3785 class_utf8data += _pcre_ord2utf8(0x2029, class_utf8data);
3786 class_utf8data += _pcre_ord2utf8(0x7fffffff, class_utf8data);
3787 }
3788 #endif
3789 continue;
3790
3791 #ifdef SUPPORT_UCP
3792 case ESC_p:
3793 case ESC_P:
3794 {
3795 BOOL negated;
3796 int pdata;
3797 int ptype = get_ucp(&ptr, &negated, &pdata, errorcodeptr);
3798 if (ptype < 0) goto FAILED;
3799 class_utf8 = TRUE;
3800 *class_utf8data++ = ((-c == ESC_p) != negated)?
3801 XCL_PROP : XCL_NOTPROP;
3802 *class_utf8data++ = ptype;
3803 *class_utf8data++ = pdata;
3804 class_charcount -= 2; /* Not a < 256 character */
3805 continue;
3806 }
3807 #endif
3808 /* Unrecognized escapes are faulted if PCRE is running in its
3809 strict mode. By default, for compatibility with Perl, they are
3810 treated as literals. */
3811
3812 default:
3813 if ((options & PCRE_EXTRA) != 0)
3814 {
3815 *errorcodeptr = ERR7;
3816 goto FAILED;
3817 }
3818 class_charcount -= 2; /* Undo the default count from above */
3819 c = *ptr; /* Get the final character and fall through */
3820 break;
3821 }
3822 }
3823
3824 /* Fall through if we have a single character (c >= 0). This may be
3825 greater than 256 in UTF-8 mode. */
3826
3827 } /* End of backslash handling */
3828
3829 /* A single character may be followed by '-' to form a range. However,
3830 Perl does not permit ']' to be the end of the range. A '-' character
3831 at the end is treated as a literal. Perl ignores orphaned \E sequences
3832 entirely. The code for handling \Q and \E is messy. */
3833
3834 CHECK_RANGE:
3835 while (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
3836 {
3837 inescq = FALSE;
3838 ptr += 2;
3839 }
3840
3841 oldptr = ptr;
3842
3843 /* Remember \r or \n */
3844
3845 if (c == CHAR_CR || c == CHAR_NL) cd->external_flags |= PCRE_HASCRORLF;
3846
3847 /* Check for range */
3848
3849 if (!inescq && ptr[1] == CHAR_MINUS)
3850 {
3851 int d;
3852 ptr += 2;
3853 while (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_E) ptr += 2;
3854
3855 /* If we hit \Q (not followed by \E) at this point, go into escaped
3856 mode. */
3857
3858 while (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_Q)
3859 {
3860 ptr += 2;
3861 if (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_E)
3862 { ptr += 2; continue; }
3863 inescq = TRUE;
3864 break;
3865 }
3866
3867 if (*ptr == 0 || (!inescq && *ptr == CHAR_RIGHT_SQUARE_BRACKET))
3868 {
3869 ptr = oldptr;
3870 goto LONE_SINGLE_CHARACTER;
3871 }
3872
3873 #ifdef SUPPORT_UTF8
3874 if (utf8)
3875 { /* Braces are required because the */
3876 GETCHARLEN(d, ptr, ptr); /* macro generates multiple statements */
3877 }
3878 else
3879 #endif
3880 d = *ptr; /* Not UTF-8 mode */
3881
3882 /* The second part of a range can be a single-character escape, but
3883 not any of the other escapes. Perl 5.6 treats a hyphen as a literal
3884 in such circumstances. */
3885
3886 if (!inescq && d == CHAR_BACKSLASH)
3887 {
3888 d = check_escape(&ptr, errorcodeptr, cd->bracount, options, TRUE);
3889 if (*errorcodeptr != 0) goto FAILED;
3890
3891 /* \b is backspace; any other special means the '-' was literal */
3892
3893 if (d < 0)
3894 {
3895 if (d == -ESC_b) d = CHAR_BS; else
3896 {
3897 ptr = oldptr;
3898 goto LONE_SINGLE_CHARACTER; /* A few lines below */
3899 }
3900 }
3901 }
3902
3903 /* Check that the two values are in the correct order. Optimize
3904 one-character ranges */
3905
3906 if (d < c)
3907 {
3908 *errorcodeptr = ERR8;
3909 goto FAILED;
3910 }
3911
3912 if (d == c) goto LONE_SINGLE_CHARACTER; /* A few lines below */
3913
3914 /* Remember \r or \n */
3915
3916 if (d == CHAR_CR || d == CHAR_NL) cd->external_flags |= PCRE_HASCRORLF;
3917
3918 /* In UTF-8 mode, if the upper limit is > 255, or > 127 for caseless
3919 matching, we have to use an XCLASS with extra data items. Caseless
3920 matching for characters > 127 is available only if UCP support is
3921 available. */
3922
3923 #ifdef SUPPORT_UTF8
3924 if (utf8 && (d > 255 || ((options & PCRE_CASELESS) != 0 && d > 127)))
3925 {
3926 class_utf8 = TRUE;
3927
3928 /* With UCP support, we can find the other case equivalents of
3929 the relevant characters. There may be several ranges. Optimize how
3930 they fit with the basic range. */
3931
3932 #ifdef SUPPORT_UCP
3933 if ((options & PCRE_CASELESS) != 0)
3934 {
3935 unsigned int occ, ocd;
3936 unsigned int cc = c;
3937 unsigned int origd = d;
3938 while (get_othercase_range(&cc, origd, &occ, &ocd))
3939 {
3940 if (occ >= (unsigned int)c &&
3941 ocd <= (unsigned int)d)
3942 continue; /* Skip embedded ranges */
3943
3944 if (occ < (unsigned int)c &&
3945 ocd >= (unsigned int)c - 1) /* Extend the basic range */
3946 { /* if there is overlap, */
3947 c = occ; /* noting that if occ < c */
3948 continue; /* we can't have ocd > d */
3949 } /* because a subrange is */
3950 if (ocd > (unsigned int)d &&
3951 occ <= (unsigned int)d + 1) /* always shorter than */
3952 { /* the basic range. */
3953 d = ocd;
3954 continue;
3955 }
3956
3957 if (occ == ocd)
3958 {
3959 *class_utf8data++ = XCL_SINGLE;
3960 }
3961 else
3962 {
3963 *class_utf8data++ = XCL_RANGE;
3964 class_utf8data += _pcre_ord2utf8(occ, class_utf8data);
3965 }
3966 class_utf8data += _pcre_ord2utf8(ocd, class_utf8data);
3967 }
3968 }
3969 #endif /* SUPPORT_UCP */
3970
3971 /* Now record the original range, possibly modified for UCP caseless
3972 overlapping ranges. */
3973
3974 *class_utf8data++ = XCL_RANGE;
3975 class_utf8data += _pcre_ord2utf8(c, class_utf8data);
3976 class_utf8data += _pcre_ord2utf8(d, class_utf8data);
3977
3978 /* With UCP support, we are done. Without UCP support, there is no
3979 caseless matching for UTF-8 characters > 127; we can use the bit map
3980 for the smaller ones. */
3981
3982 #ifdef SUPPORT_UCP
3983 continue; /* With next character in the class */
3984 #else
3985 if ((options & PCRE_CASELESS) == 0 || c > 127) continue;
3986
3987 /* Adjust upper limit and fall through to set up the map */
3988
3989 d = 127;
3990
3991 #endif /* SUPPORT_UCP */
3992 }
3993 #endif /* SUPPORT_UTF8 */
3994
3995 /* We use the bit map for all cases when not in UTF-8 mode; else
3996 ranges that lie entirely within 0-127 when there is UCP support; else
3997 for partial ranges without UCP support. */
3998
3999 class_charcount += d - c + 1;
4000 class_lastchar = d;
4001
4002 /* We can save a bit of time by skipping this in the pre-compile. */
4003
4004 if (lengthptr == NULL) for (; c <= d; c++)
4005 {
4006 classbits[c/8] |= (1 << (c&7));
4007 if ((options & PCRE_CASELESS) != 0)
4008 {
4009 int uc = cd->fcc[c]; /* flip case */
4010 classbits[uc/8] |= (1 << (uc&7));
4011 }
4012 }
4013
4014 continue; /* Go get the next char in the class */
4015 }
4016
4017 /* Handle a lone single character - we can get here for a normal
4018 non-escape char, or after \ that introduces a single character or for an
4019 apparent range that isn't. */
4020
4021 LONE_SINGLE_CHARACTER:
4022
4023 /* Handle a character that cannot go in the bit map */
4024
4025 #ifdef SUPPORT_UTF8
4026 if (utf8 && (c > 255 || ((options & PCRE_CASELESS) != 0 && c > 127)))
4027 {
4028 class_utf8 = TRUE;
4029 *class_utf8data++ = XCL_SINGLE;
4030 class_utf8data += _pcre_ord2utf8(c, class_utf8data);
4031
4032 #ifdef SUPPORT_UCP
4033 if ((options & PCRE_CASELESS) != 0)
4034 {
4035 unsigned int othercase;
4036 if ((othercase = UCD_OTHERCASE(c)) != c)
4037 {
4038 *class_utf8data++ = XCL_SINGLE;
4039 class_utf8data += _pcre_ord2utf8(othercase, class_utf8data);
4040 }
4041 }
4042 #endif /* SUPPORT_UCP */
4043
4044 }
4045 else
4046 #endif /* SUPPORT_UTF8 */
4047
4048 /* Handle a single-byte character */
4049 {
4050 classbits[c/8] |= (1 << (c&7));
4051 if ((options & PCRE_CASELESS) != 0)
4052 {
4053 c = cd->fcc[c]; /* flip case */
4054 classbits[c/8] |= (1 << (c&7));
4055 }
4056 class_charcount++;
4057 class_lastchar = c;
4058 }
4059 }
4060
4061 /* Loop until ']' reached. This "while" is the end of the "do" far above.
4062 If we are at the end of an internal nested string, revert to the outer
4063 string. */
4064
4065 while (((c = *(++ptr)) != 0 ||
4066 (nestptr != NULL &&
4067 (ptr = nestptr, nestptr = NULL, c = *(++ptr)) != 0)) &&
4068 (c != CHAR_RIGHT_SQUARE_BRACKET || inescq));
4069
4070 /* Check for missing terminating ']' */
4071
4072 if (c == 0)
4073 {
4074 *errorcodeptr = ERR6;
4075 goto FAILED;
4076 }
4077
4078 /* If class_charcount is 1, we saw precisely one character whose value is
4079 less than 256. As long as there were no characters >= 128 and there was no
4080 use of \p or \P, in other words, no use of any XCLASS features, we can
4081 optimize.
4082
4083 In UTF-8 mode, we can optimize the negative case only if there were no
4084 characters >= 128 because OP_NOT and the related opcodes like OP_NOTSTAR
4085 operate on single-bytes characters only. This is an historical hangover.
4086 Maybe one day we can tidy these opcodes to handle multi-byte characters.
4087
4088 The optimization throws away the bit map. We turn the item into a
4089 1-character OP_CHAR[I] if it's positive, or OP_NOT[I] if it's negative.
4090 Note that OP_NOT[I] does not support multibyte characters. In the positive
4091 case, it can cause firstbyte to be set. Otherwise, there can be no first
4092 char if this item is first, whatever repeat count may follow. In the case
4093 of reqbyte, save the previous value for reinstating. */
4094
4095 #ifdef SUPPORT_UTF8
4096 if (class_charcount == 1 && !class_utf8 &&
4097 (!utf8 || !negate_class || class_lastchar < 128))
4098 #else
4099 if (class_charcount == 1)
4100 #endif
4101 {
4102 zeroreqbyte = reqbyte;
4103
4104 /* The OP_NOT[I] opcodes work on one-byte characters only. */
4105
4106 if (negate_class)
4107 {
4108 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
4109 zerofirstbyte = firstbyte;
4110 *code++ = ((options & PCRE_CASELESS) != 0)? OP_NOTI: OP_NOT;
4111 *code++ = class_lastchar;
4112 break;
4113 }
4114
4115 /* For a single, positive character, get the value into mcbuffer, and
4116 then we can handle this with the normal one-character code. */
4117
4118 #ifdef SUPPORT_UTF8
4119 if (utf8 && class_lastchar > 127)
4120 mclength = _pcre_ord2utf8(class_lastchar, mcbuffer);
4121 else
4122 #endif
4123 {
4124 mcbuffer[0] = class_lastchar;
4125 mclength = 1;
4126 }
4127 goto ONE_CHAR;
4128 } /* End of 1-char optimization */
4129
4130 /* The general case - not the one-char optimization. If this is the first
4131 thing in the branch, there can be no first char setting, whatever the
4132 repeat count. Any reqbyte setting must remain unchanged after any kind of
4133 repeat. */
4134
4135 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
4136 zerofirstbyte = firstbyte;
4137 zeroreqbyte = reqbyte;
4138
4139 /* If there are characters with values > 255, we have to compile an
4140 extended class, with its own opcode, unless there was a negated special
4141 such as \S in the class, and PCRE_UCP is not set, because in that case all
4142 characters > 255 are in the class, so any that were explicitly given as
4143 well can be ignored. If (when there are explicit characters > 255 that must
4144 be listed) there are no characters < 256, we can omit the bitmap in the
4145 actual compiled code. */
4146
4147 #ifdef SUPPORT_UTF8
4148 if (class_utf8 && (!should_flip_negation || (options & PCRE_UCP) != 0))
4149 {
4150 *class_utf8data++ = XCL_END; /* Marks the end of extra data */
4151 *code++ = OP_XCLASS;
4152 code += LINK_SIZE;
4153 *code = negate_class? XCL_NOT : 0;
4154
4155 /* If the map is required, move up the extra data to make room for it;
4156 otherwise just move the code pointer to the end of the extra data. */
4157
4158 if (class_charcount > 0)
4159 {
4160 *code++ |= XCL_MAP;
4161 memmove(code + 32, code, class_utf8data - code);
4162 memcpy(code, classbits, 32);
4163 code = class_utf8data + 32;
4164 }
4165 else code = class_utf8data;
4166
4167 /* Now fill in the complete length of the item */
4168
4169 PUT(previous, 1, code - previous);
4170 break; /* End of class handling */
4171 }
4172 #endif
4173
4174 /* If there are no characters > 255, or they are all to be included or
4175 excluded, set the opcode to OP_CLASS or OP_NCLASS, depending on whether the
4176 whole class was negated and whether there were negative specials such as \S
4177 (non-UCP) in the class. Then copy the 32-byte map into the code vector,
4178 negating it if necessary. */
4179
4180 *code++ = (negate_class == should_flip_negation) ? OP_CLASS : OP_NCLASS;
4181 if (negate_class)
4182 {
4183 if (lengthptr == NULL) /* Save time in the pre-compile phase */
4184 for (c = 0; c < 32; c++) code[c] = ~classbits[c];
4185 }
4186 else
4187 {
4188 memcpy(code, classbits, 32);
4189 }
4190 code += 32;
4191 break;
4192
4193
4194 /* ===================================================================*/
4195 /* Various kinds of repeat; '{' is not necessarily a quantifier, but this
4196 has been tested above. */
4197
4198 case CHAR_LEFT_CURLY_BRACKET:
4199 if (!is_quantifier) goto NORMAL_CHAR;
4200 ptr = read_repeat_counts(ptr+1, &repeat_min, &repeat_max, errorcodeptr);
4201 if (*errorcodeptr != 0) goto FAILED;
4202 goto REPEAT;
4203
4204 case CHAR_ASTERISK:
4205 repeat_min = 0;
4206 repeat_max = -1;
4207 goto REPEAT;
4208
4209 case CHAR_PLUS:
4210 repeat_min = 1;
4211 repeat_max = -1;
4212 goto REPEAT;
4213
4214 case CHAR_QUESTION_MARK:
4215 repeat_min = 0;
4216 repeat_max = 1;
4217
4218 REPEAT:
4219 if (previous == NULL)
4220 {
4221 *errorcodeptr = ERR9;
4222 goto FAILED;
4223 }
4224
4225 if (repeat_min == 0)
4226 {
4227 firstbyte = zerofirstbyte; /* Adjust for zero repeat */
4228 reqbyte = zeroreqbyte; /* Ditto */
4229 }
4230
4231 /* Remember whether this is a variable length repeat */
4232
4233 reqvary = (repeat_min == repeat_max)? 0 : REQ_VARY;
4234
4235 op_type = 0; /* Default single-char op codes */
4236 possessive_quantifier = FALSE; /* Default not possessive quantifier */
4237
4238 /* Save start of previous item, in case we have to move it up in order to
4239 insert something before it. */
4240
4241 tempcode = previous;
4242
4243 /* If the next character is '+', we have a possessive quantifier. This
4244 implies greediness, whatever the setting of the PCRE_UNGREEDY option.
4245 If the next character is '?' this is a minimizing repeat, by default,
4246 but if PCRE_UNGREEDY is set, it works the other way round. We change the
4247 repeat type to the non-default. */
4248
4249 if (ptr[1] == CHAR_PLUS)
4250 {
4251 repeat_type = 0; /* Force greedy */
4252 possessive_quantifier = TRUE;
4253 ptr++;
4254 }
4255 else if (ptr[1] == CHAR_QUESTION_MARK)
4256 {
4257 repeat_type = greedy_non_default;
4258 ptr++;
4259 }
4260 else repeat_type = greedy_default;
4261
4262 /* If previous was a recursion call, wrap it in atomic brackets so that
4263 previous becomes the atomic group. All recursions were so wrapped in the
4264 past, but it no longer happens for non-repeated recursions. In fact, the
4265 repeated ones could be re-implemented independently so as not to need this,
4266 but for the moment we rely on the code for repeating groups. */
4267
4268 if (*previous == OP_RECURSE)
4269 {
4270 memmove(previous + 1 + LINK_SIZE, previous, 1 + LINK_SIZE);
4271 *previous = OP_ONCE;
4272 PUT(previous, 1, 2 + 2*LINK_SIZE);
4273 previous[2 + 2*LINK_SIZE] = OP_KET;
4274 PUT(previous, 3 + 2*LINK_SIZE, 2 + 2*LINK_SIZE);
4275 code += 2 + 2 * LINK_SIZE;
4276 length_prevgroup = 3 + 3*LINK_SIZE;
4277
4278 /* When actually compiling, we need to check whether this was a forward
4279 reference, and if so, adjust the offset. */
4280
4281 if (lengthptr == NULL && cd->hwm >= cd->start_workspace + LINK_SIZE)
4282 {
4283 int offset = GET(cd->hwm, -LINK_SIZE);
4284 if (offset == previous + 1 - cd->start_code)
4285 PUT(cd->hwm, -LINK_SIZE, offset + 1 + LINK_SIZE);
4286 }
4287 }
4288
4289 /* Now handle repetition for the different types of item. */
4290
4291 /* If previous was a character match, abolish the item and generate a
4292 repeat item instead. If a char item has a minumum of more than one, ensure
4293 that it is set in reqbyte - it might not be if a sequence such as x{3} is
4294 the first thing in a branch because the x will have gone into firstbyte
4295 instead. */
4296
4297 if (*previous == OP_CHAR || *previous == OP_CHARI)
4298 {
4299 op_type = (*previous == OP_CHAR)? 0 : OP_STARI - OP_STAR;
4300
4301 /* Deal with UTF-8 characters that take up more than one byte. It's
4302 easier to write this out separately than try to macrify it. Use c to
4303 hold the length of the character in bytes, plus 0x80 to flag that it's a
4304 length rather than a small character. */
4305
4306 #ifdef SUPPORT_UTF8
4307 if (utf8 && (code[-1] & 0x80) != 0)
4308 {
4309 uschar *lastchar = code - 1;
4310 while((*lastchar & 0xc0) == 0x80) lastchar--;
4311 c = code - lastchar; /* Length of UTF-8 character */
4312 memcpy(utf8_char, lastchar, c); /* Save the char */
4313 c |= 0x80; /* Flag c as a length */
4314 }
4315 else
4316 #endif
4317
4318 /* Handle the case of a single byte - either with no UTF8 support, or
4319 with UTF-8 disabled, or for a UTF-8 character < 128. */
4320
4321 {
4322 c = code[-1];
4323 if (repeat_min > 1) reqbyte = c | req_caseopt | cd->req_varyopt;
4324 }
4325
4326 /* If the repetition is unlimited, it pays to see if the next thing on
4327 the line is something that cannot possibly match this character. If so,
4328 automatically possessifying this item gains some performance in the case
4329 where the match fails. */
4330
4331 if (!possessive_quantifier &&
4332 repeat_max < 0 &&
4333 check_auto_possessive(previous, utf8, ptr + 1, options, cd))
4334 {
4335 repeat_type = 0; /* Force greedy */
4336 possessive_quantifier = TRUE;
4337 }
4338
4339 goto OUTPUT_SINGLE_REPEAT; /* Code shared with single character types */
4340 }
4341
4342 /* If previous was a single negated character ([^a] or similar), we use
4343 one of the special opcodes, replacing it. The code is shared with single-
4344 character repeats by setting opt_type to add a suitable offset into
4345 repeat_type. We can also test for auto-possessification. OP_NOT and OP_NOTI
4346 are currently used only for single-byte chars. */
4347
4348 else if (*previous == OP_NOT || *previous == OP_NOTI)
4349 {
4350 op_type = ((*previous == OP_NOT)? OP_NOTSTAR : OP_NOTSTARI) - OP_STAR;
4351 c = previous[1];
4352 if (!possessive_quantifier &&
4353 repeat_max < 0 &&
4354 check_auto_possessive(previous, utf8, ptr + 1, options, cd))
4355 {
4356 repeat_type = 0; /* Force greedy */
4357 possessive_quantifier = TRUE;
4358 }
4359 goto OUTPUT_SINGLE_REPEAT;
4360 }
4361
4362 /* If previous was a character type match (\d or similar), abolish it and
4363 create a suitable repeat item. The code is shared with single-character
4364 repeats by setting op_type to add a suitable offset into repeat_type. Note
4365 the the Unicode property types will be present only when SUPPORT_UCP is
4366 defined, but we don't wrap the little bits of code here because it just
4367 makes it horribly messy. */
4368
4369 else if (*previous < OP_EODN)
4370 {
4371 uschar *oldcode;
4372 int prop_type, prop_value;
4373 op_type = OP_TYPESTAR - OP_STAR; /* Use type opcodes */
4374 c = *previous;
4375
4376 if (!possessive_quantifier &&
4377 repeat_max < 0 &&
4378 check_auto_possessive(previous, utf8, ptr + 1, options, cd))
4379 {
4380 repeat_type = 0; /* Force greedy */
4381 possessive_quantifier = TRUE;
4382 }
4383
4384 OUTPUT_SINGLE_REPEAT:
4385 if (*previous == OP_PROP || *previous == OP_NOTPROP)
4386 {
4387 prop_type = previous[1];
4388 prop_value = previous[2];
4389 }
4390 else prop_type = prop_value = -1;
4391
4392 oldcode = code;
4393 code = previous; /* Usually overwrite previous item */
4394
4395 /* If the maximum is zero then the minimum must also be zero; Perl allows
4396 this case, so we do too - by simply omitting the item altogether. */
4397
4398 if (repeat_max == 0) goto END_REPEAT;
4399
4400 /*--------------------------------------------------------------------*/
4401 /* This code is obsolete from release 8.00; the restriction was finally
4402 removed: */
4403
4404 /* All real repeats make it impossible to handle partial matching (maybe
4405 one day we will be able to remove this restriction). */
4406
4407 /* if (repeat_max != 1) cd->external_flags |= PCRE_NOPARTIAL; */
4408 /*--------------------------------------------------------------------*/
4409
4410 /* Combine the op_type with the repeat_type */
4411
4412 repeat_type += op_type;
4413
4414 /* A minimum of zero is handled either as the special case * or ?, or as
4415 an UPTO, with the maximum given. */
4416
4417 if (repeat_min == 0)
4418 {
4419 if (repeat_max == -1) *code++ = OP_STAR + repeat_type;
4420 else if (repeat_max == 1) *code++ = OP_QUERY + repeat_type;
4421 else
4422 {
4423 *code++ = OP_UPTO + repeat_type;
4424 PUT2INC(code, 0, repeat_max);
4425 }
4426 }
4427
4428 /* A repeat minimum of 1 is optimized into some special cases. If the
4429 maximum is unlimited, we use OP_PLUS. Otherwise, the original item is
4430 left in place and, if the maximum is greater than 1, we use OP_UPTO with
4431 one less than the maximum. */
4432
4433 else if (repeat_min == 1)
4434 {
4435 if (repeat_max == -1)
4436 *code++ = OP_PLUS + repeat_type;
4437 else
4438 {
4439 code = oldcode; /* leave previous item in place */
4440 if (repeat_max == 1) goto END_REPEAT;
4441 *code++ = OP_UPTO + repeat_type;
4442 PUT2INC(code, 0, repeat_max - 1);
4443 }
4444 }
4445
4446 /* The case {n,n} is just an EXACT, while the general case {n,m} is
4447 handled as an EXACT followed by an UPTO. */
4448
4449 else
4450 {
4451 *code++ = OP_EXACT + op_type; /* NB EXACT doesn't have repeat_type */
4452 PUT2INC(code, 0, repeat_min);
4453
4454 /* If the maximum is unlimited, insert an OP_STAR. Before doing so,
4455 we have to insert the character for the previous code. For a repeated
4456 Unicode property match, there are two extra bytes that define the
4457 required property. In UTF-8 mode, long characters have their length in
4458 c, with the 0x80 bit as a flag. */
4459
4460 if (repeat_max < 0)
4461 {
4462 #ifdef SUPPORT_UTF8
4463 if (utf8 && c >= 128)
4464 {
4465 memcpy(code, utf8_char, c & 7);
4466 code += c & 7;
4467 }
4468 else
4469 #endif
4470 {
4471 *code++ = c;
4472 if (prop_type >= 0)
4473 {
4474 *code++ = prop_type;
4475 *code++ = prop_value;
4476 }
4477 }
4478 *code++ = OP_STAR + repeat_type;
4479 }
4480
4481 /* Else insert an UPTO if the max is greater than the min, again
4482 preceded by the character, for the previously inserted code. If the
4483 UPTO is just for 1 instance, we can use QUERY instead. */
4484
4485 else if (repeat_max != repeat_min)
4486 {
4487 #ifdef SUPPORT_UTF8
4488 if (utf8 && c >= 128)
4489 {
4490 memcpy(code, utf8_char, c & 7);
4491 code += c & 7;
4492 }
4493 else
4494 #endif
4495 *code++ = c;
4496 if (prop_type >= 0)
4497 {
4498 *code++ = prop_type;
4499 *code++ = prop_value;
4500 }
4501 repeat_max -= repeat_min;
4502
4503 if (repeat_max == 1)
4504 {
4505 *code++ = OP_QUERY + repeat_type;
4506 }
4507 else
4508 {
4509 *code++ = OP_UPTO + repeat_type;
4510 PUT2INC(code, 0, repeat_max);
4511 }
4512 }
4513 }
4514
4515 /* The character or character type itself comes last in all cases. */
4516
4517 #ifdef SUPPORT_UTF8
4518 if (utf8 && c >= 128)
4519 {
4520 memcpy(code, utf8_char, c & 7);
4521 code += c & 7;
4522 }
4523 else
4524 #endif
4525 *code++ = c;
4526
4527 /* For a repeated Unicode property match, there are two extra bytes that
4528 define the required property. */
4529
4530 #ifdef SUPPORT_UCP
4531 if (prop_type >= 0)
4532 {
4533 *code++ = prop_type;
4534 *code++ = prop_value;
4535 }
4536 #endif
4537 }
4538
4539 /* If previous was a character class or a back reference, we put the repeat
4540 stuff after it, but just skip the item if the repeat was {0,0}. */
4541
4542 else if (*previous == OP_CLASS ||
4543 *previous == OP_NCLASS ||
4544 #ifdef SUPPORT_UTF8
4545 *previous == OP_XCLASS ||
4546 #endif
4547 *previous == OP_REF ||
4548 *previous == OP_REFI)
4549 {
4550 if (repeat_max == 0)
4551 {
4552 code = previous;
4553 goto END_REPEAT;
4554 }
4555
4556 /*--------------------------------------------------------------------*/
4557 /* This code is obsolete from release 8.00; the restriction was finally
4558 removed: */
4559
4560 /* All real repeats make it impossible to handle partial matching (maybe
4561 one day we will be able to remove this restriction). */
4562
4563 /* if (repeat_max != 1) cd->external_flags |= PCRE_NOPARTIAL; */
4564 /*--------------------------------------------------------------------*/
4565
4566 if (repeat_min == 0 && repeat_max == -1)
4567 *code++ = OP_CRSTAR + repeat_type;
4568 else if (repeat_min == 1 && repeat_max == -1)
4569 *code++ = OP_CRPLUS + repeat_type;
4570 else if (repeat_min == 0 && repeat_max == 1)
4571 *code++ = OP_CRQUERY + repeat_type;
4572 else
4573 {
4574 *code++ = OP_CRRANGE + repeat_type;
4575 PUT2INC(code, 0, repeat_min);
4576 if (repeat_max == -1) repeat_max = 0; /* 2-byte encoding for max */
4577 PUT2INC(code, 0, repeat_max);
4578 }
4579 }
4580
4581 /* If previous was a bracket group, we may have to replicate it in certain
4582 cases. Note that at this point we can encounter only the "basic" bracket
4583 opcodes such as BRA and CBRA, as this is the place where they get converted
4584 into the more special varieties such as BRAPOS and SBRA. A test for >=
4585 OP_ASSERT and <= OP_COND includes ASSERT, ASSERT_NOT, ASSERTBACK,
4586 ASSERTBACK_NOT, ONCE, BRA, CBRA, and COND. Originally, PCRE did not allow
4587 repetition of assertions, but now it does, for Perl compatibility. */
4588
4589 else if (*previous >= OP_ASSERT && *previous <= OP_COND)
4590 {
4591 register int i;
4592 int len = (int)(code - previous);
4593 uschar *bralink = NULL;
4594 uschar *brazeroptr = NULL;
4595
4596 /* Repeating a DEFINE group is pointless, but Perl allows the syntax, so
4597 we just ignore the repeat. */
4598
4599 if (*previous == OP_COND && previous[LINK_SIZE+1] == OP_DEF)
4600 goto END_REPEAT;
4601
4602 /* There is no sense in actually repeating assertions. The only potential
4603 use of repetition is in cases when the assertion is optional. Therefore,
4604 if the minimum is greater than zero, just ignore the repeat. If the
4605 maximum is not not zero or one, set it to 1. */
4606
4607 if (*previous < OP_ONCE) /* Assertion */
4608 {
4609 if (repeat_min > 0) goto END_REPEAT;
4610 if (repeat_max < 0 || repeat_max > 1) repeat_max = 1;
4611 }
4612
4613 /* The case of a zero minimum is special because of the need to stick
4614 OP_BRAZERO in front of it, and because the group appears once in the
4615 data, whereas in other cases it appears the minimum number of times. For
4616 this reason, it is simplest to treat this case separately, as otherwise
4617 the code gets far too messy. There are several special subcases when the
4618 minimum is zero. */
4619
4620 if (repeat_min == 0)
4621 {
4622 /* If the maximum is also zero, we used to just omit the group from the
4623 output altogether, like this:
4624
4625 ** if (repeat_max == 0)
4626 ** {
4627 ** code = previous;
4628 ** goto END_REPEAT;
4629 ** }
4630
4631 However, that fails when a group or a subgroup within it is referenced
4632 as a subroutine from elsewhere in the pattern, so now we stick in
4633 OP_SKIPZERO in front of it so that it is skipped on execution. As we
4634 don't have a list of which groups are referenced, we cannot do this
4635 selectively.
4636
4637 If the maximum is 1 or unlimited, we just have to stick in the BRAZERO
4638 and do no more at this point. However, we do need to adjust any
4639 OP_RECURSE calls inside the group that refer to the group itself or any
4640 internal or forward referenced group, because the offset is from the
4641 start of the whole regex. Temporarily terminate the pattern while doing
4642 this. */
4643
4644 if (repeat_max <= 1) /* Covers 0, 1, and unlimited */
4645 {
4646 *code = OP_END;
4647 adjust_recurse(previous, 1, utf8, cd, save_hwm);
4648 memmove(previous+1, previous, len);
4649 code++;
4650 if (repeat_max == 0)
4651 {
4652 *previous++ = OP_SKIPZERO;
4653 goto END_REPEAT;
4654 }
4655 brazeroptr = previous; /* Save for possessive optimizing */
4656 *previous++ = OP_BRAZERO + repeat_type;
4657 }
4658
4659 /* If the maximum is greater than 1 and limited, we have to replicate
4660 in a nested fashion, sticking OP_BRAZERO before each set of brackets.
4661 The first one has to be handled carefully because it's the original
4662 copy, which has to be moved up. The remainder can be handled by code
4663 that is common with the non-zero minimum case below. We have to
4664 adjust the value or repeat_max, since one less copy is required. Once
4665 again, we may have to adjust any OP_RECURSE calls inside the group. */
4666
4667 else
4668 {
4669 int offset;
4670 *code = OP_END;
4671 adjust_recurse(previous, 2 + LINK_SIZE, utf8, cd, save_hwm);
4672 memmove(previous + 2 + LINK_SIZE, previous, len);
4673 code += 2 + LINK_SIZE;
4674 *previous++ = OP_BRAZERO + repeat_type;
4675 *previous++ = OP_BRA;
4676
4677 /* We chain together the bracket offset fields that have to be
4678 filled in later when the ends of the brackets are reached. */
4679
4680 offset = (bralink == NULL)? 0 : (int)(previous - bralink);
4681 bralink = previous;
4682 PUTINC(previous, 0, offset);
4683 }
4684
4685 repeat_max--;
4686 }
4687
4688 /* If the minimum is greater than zero, replicate the group as many
4689 times as necessary, and adjust the maximum to the number of subsequent
4690 copies that we need. If we set a first char from the group, and didn't
4691 set a required char, copy the latter from the former. If there are any
4692 forward reference subroutine calls in the group, there will be entries on
4693 the workspace list; replicate these with an appropriate increment. */
4694
4695 else
4696 {
4697 if (repeat_min > 1)
4698 {
4699 /* In the pre-compile phase, we don't actually do the replication. We
4700 just adjust the length as if we had. Do some paranoid checks for
4701 potential integer overflow. The INT64_OR_DOUBLE type is a 64-bit
4702 integer type when available, otherwise double. */
4703
4704 if (lengthptr != NULL)
4705 {
4706 int delta = (repeat_min - 1)*length_prevgroup;
4707 if ((INT64_OR_DOUBLE)(repeat_min - 1)*
4708 (INT64_OR_DOUBLE)length_prevgroup >
4709 (INT64_OR_DOUBLE)INT_MAX ||
4710 OFLOW_MAX - *lengthptr < delta)
4711 {
4712 *errorcodeptr = ERR20;
4713 goto FAILED;
4714 }
4715 *lengthptr += delta;
4716 }
4717
4718 /* This is compiling for real */
4719
4720 else
4721 {
4722 if (groupsetfirstbyte && reqbyte < 0) reqbyte = firstbyte;
4723 for (i = 1; i < repeat_min; i++)
4724 {
4725 uschar *hc;
4726 uschar *this_hwm = cd->hwm;
4727 memcpy(code, previous, len);
4728 for (hc = save_hwm; hc < this_hwm; hc += LINK_SIZE)
4729 {
4730 PUT(cd->hwm, 0, GET(hc, 0) + len);
4731 cd->hwm += LINK_SIZE;
4732 }
4733 save_hwm = this_hwm;
4734 code += len;
4735 }
4736 }
4737 }
4738
4739 if (repeat_max > 0) repeat_max -= repeat_min;
4740 }
4741
4742 /* This code is common to both the zero and non-zero minimum cases. If
4743 the maximum is limited, it replicates the group in a nested fashion,
4744 remembering the bracket starts on a stack. In the case of a zero minimum,
4745 the first one was set up above. In all cases the repeat_max now specifies
4746 the number of additional copies needed. Again, we must remember to
4747 replicate entries on the forward reference list. */
4748
4749 if (repeat_max >= 0)
4750 {
4751 /* In the pre-compile phase, we don't actually do the replication. We
4752 just adjust the length as if we had. For each repetition we must add 1
4753 to the length for BRAZERO and for all but the last repetition we must
4754 add 2 + 2*LINKSIZE to allow for the nesting that occurs. Do some
4755 paranoid checks to avoid integer overflow. The INT64_OR_DOUBLE type is
4756 a 64-bit integer type when available, otherwise double. */
4757
4758 if (lengthptr != NULL && repeat_max > 0)
4759 {
4760 int delta = repeat_max * (length_prevgroup + 1 + 2 + 2*LINK_SIZE) -
4761 2 - 2*LINK_SIZE; /* Last one doesn't nest */
4762 if ((INT64_OR_DOUBLE)repeat_max *
4763 (INT64_OR_DOUBLE)(length_prevgroup + 1 + 2 + 2*LINK_SIZE)
4764 > (INT64_OR_DOUBLE)INT_MAX ||
4765 OFLOW_MAX - *lengthptr < delta)
4766 {
4767 *errorcodeptr = ERR20;
4768 goto FAILED;
4769 }
4770 *lengthptr += delta;
4771 }
4772
4773 /* This is compiling for real */
4774
4775 else for (i = repeat_max - 1; i >= 0; i--)
4776 {
4777 uschar *hc;
4778 uschar *this_hwm = cd->hwm;
4779
4780 *code++ = OP_BRAZERO + repeat_type;
4781
4782 /* All but the final copy start a new nesting, maintaining the
4783 chain of brackets outstanding. */
4784
4785 if (i != 0)
4786 {
4787 int offset;
4788 *code++ = OP_BRA;
4789 offset = (bralink == NULL)? 0 : (int)(code - bralink);
4790 bralink = code;
4791 PUTINC(code, 0, offset);
4792 }
4793
4794 memcpy(code, previous, len);
4795 for (hc = save_hwm; hc < this_hwm; hc += LINK_SIZE)
4796 {
4797 PUT(cd->hwm, 0, GET(hc, 0) + len + ((i != 0)? 2+LINK_SIZE : 1));
4798 cd->hwm += LINK_SIZE;
4799 }
4800 save_hwm = this_hwm;
4801 code += len;
4802 }
4803
4804 /* Now chain through the pending brackets, and fill in their length
4805 fields (which are holding the chain links pro tem). */
4806
4807 while (bralink != NULL)
4808 {
4809 int oldlinkoffset;
4810 int offset = (int)(code - bralink + 1);
4811 uschar *bra = code - offset;
4812 oldlinkoffset = GET(bra, 1);
4813 bralink = (oldlinkoffset == 0)? NULL : bralink - oldlinkoffset;
4814 *code++ = OP_KET;
4815 PUTINC(code, 0, offset);
4816 PUT(bra, 1, offset);
4817 }
4818 }
4819
4820 /* If the maximum is unlimited, set a repeater in the final copy. For
4821 ONCE brackets, that's all we need to do. However, possessively repeated
4822 ONCE brackets can be converted into non-capturing brackets, as the
4823 behaviour of (?:xx)++ is the same as (?>xx)++ and this saves having to
4824 deal with possessive ONCEs specially.
4825
4826 Otherwise, if the quantifier was possessive, we convert the BRA code to
4827 the POS form, and the KET code to KETRPOS. (It turns out to be convenient
4828 at runtime to detect this kind of subpattern at both the start and at the
4829 end.) The use of special opcodes makes it possible to reduce greatly the
4830 stack usage in pcre_exec(). If the group is preceded by OP_BRAZERO,
4831 convert this to OP_BRAPOSZERO. Then cancel the possessive flag so that
4832 the default action below, of wrapping everything inside atomic brackets,
4833 does not happen.
4834
4835 Then, when we are doing the actual compile phase, check to see whether
4836 this group is one that could match an empty string. If so, convert the
4837 initial operator to the S form (e.g. OP_BRA -> OP_SBRA) so that runtime
4838 checking can be done. [This check is also applied to ONCE groups at
4839 runtime, but in a different way.] */
4840
4841 else
4842 {
4843 uschar *ketcode = code - 1 - LINK_SIZE;
4844 uschar *bracode = ketcode - GET(ketcode, 1);
4845
4846 if ((*bracode == OP_ONCE || *bracode == OP_ONCE_NC) &&
4847 possessive_quantifier) *bracode = OP_BRA;
4848
4849 if (*bracode == OP_ONCE || *bracode == OP_ONCE_NC)
4850 *ketcode = OP_KETRMAX + repeat_type;
4851 else
4852 {
4853 if (possessive_quantifier)
4854 {
4855 *bracode += 1; /* Switch to xxxPOS opcodes */
4856 *ketcode = OP_KETRPOS;
4857 if (brazeroptr != NULL) *brazeroptr = OP_BRAPOSZERO;
4858 possessive_quantifier = FALSE;
4859 }
4860 else *ketcode = OP_KETRMAX + repeat_type;
4861
4862 if (lengthptr == NULL)
4863 {
4864 uschar *scode = bracode;
4865 do
4866 {
4867 if (could_be_empty_branch(scode, ketcode, utf8, cd))
4868 {
4869 *bracode += OP_SBRA - OP_BRA;
4870 break;
4871 }
4872 scode += GET(scode, 1);
4873 }
4874 while (*scode == OP_ALT);
4875 }
4876 }
4877 }
4878 }
4879
4880 /* If previous is OP_FAIL, it was generated by an empty class [] in
4881 JavaScript mode. The other ways in which OP_FAIL can be generated, that is
4882 by (*FAIL) or (?!) set previous to NULL, which gives a "nothing to repeat"
4883 error above. We can just ignore the repeat in JS case. */
4884
4885 else if (*previous == OP_FAIL) goto END_REPEAT;
4886
4887 /* Else there's some kind of shambles */
4888
4889 else
4890 {
4891 *errorcodeptr = ERR11;
4892 goto FAILED;
4893 }
4894
4895 /* If the character following a repeat is '+', or if certain optimization
4896 tests above succeeded, possessive_quantifier is TRUE. For some opcodes,
4897 there are special alternative opcodes for this case. For anything else, we
4898 wrap the entire repeated item inside OP_ONCE brackets. Logically, the '+'
4899 notation is just syntactic sugar, taken from Sun's Java package, but the
4900 special opcodes can optimize it.
4901
4902 Possessively repeated subpatterns have already been handled in the code
4903 just above, so possessive_quantifier is always FALSE for them at this
4904 stage.
4905
4906 Note that the repeated item starts at tempcode, not at previous, which
4907 might be the first part of a string whose (former) last char we repeated.
4908
4909 Possessifying an 'exact' quantifier has no effect, so we can ignore it. But
4910 an 'upto' may follow. We skip over an 'exact' item, and then test the
4911 length of what remains before proceeding. */
4912
4913 if (possessive_quantifier)
4914 {
4915 int len;
4916
4917 if (*tempcode == OP_TYPEEXACT)
4918 tempcode += _pcre_OP_lengths[*tempcode] +
4919 ((tempcode[3] == OP_PROP || tempcode[3] == OP_NOTPROP)? 2 : 0);
4920
4921 else if (*tempcode == OP_EXACT || *tempcode == OP_NOTEXACT)
4922 {
4923 tempcode += _pcre_OP_lengths[*tempcode];
4924 #ifdef SUPPORT_UTF8
4925 if (utf8 && tempcode[-1] >= 0xc0)
4926 tempcode += _pcre_utf8_table4[tempcode[-1] & 0x3f];
4927 #endif
4928 }
4929
4930 len = (int)(code - tempcode);
4931 if (len > 0) switch (*tempcode)
4932 {
4933 case OP_STAR: *tempcode = OP_POSSTAR; break;
4934 case OP_PLUS: *tempcode = OP_POSPLUS; break;
4935 case OP_QUERY: *tempcode = OP_POSQUERY; break;
4936 case OP_UPTO: *tempcode = OP_POSUPTO; break;
4937
4938 case OP_STARI: *tempcode = OP_POSSTARI; break;
4939 case OP_PLUSI: *tempcode = OP_POSPLUSI; break;
4940 case OP_QUERYI: *tempcode = OP_POSQUERYI; break;
4941 case OP_UPTOI: *tempcode = OP_POSUPTOI; break;
4942
4943 case OP_NOTSTAR: *tempcode = OP_NOTPOSSTAR; break;
4944 case OP_NOTPLUS: *tempcode = OP_NOTPOSPLUS; break;
4945 case OP_NOTQUERY: *tempcode = OP_NOTPOSQUERY; break;
4946 case OP_NOTUPTO: *tempcode = OP_NOTPOSUPTO; break;
4947
4948 case OP_NOTSTARI: *tempcode = OP_NOTPOSSTARI; break;
4949 case OP_NOTPLUSI: *tempcode = OP_NOTPOSPLUSI; break;
4950 case OP_NOTQUERYI: *tempcode = OP_NOTPOSQUERYI; break;
4951 case OP_NOTUPTOI: *tempcode = OP_NOTPOSUPTOI; break;
4952
4953 case OP_TYPESTAR: *tempcode = OP_TYPEPOSSTAR; break;
4954 case OP_TYPEPLUS: *tempcode = OP_TYPEPOSPLUS; break;
4955 case OP_TYPEQUERY: *tempcode = OP_TYPEPOSQUERY; break;
4956 case OP_TYPEUPTO: *tempcode = OP_TYPEPOSUPTO; break;
4957
4958 /* Because we are moving code along, we must ensure that any
4959 pending recursive references are updated. */
4960
4961 default:
4962 *code = OP_END;
4963 adjust_recurse(tempcode, 1 + LINK_SIZE, utf8, cd, save_hwm);
4964 memmove(tempcode + 1+LINK_SIZE, tempcode, len);
4965 code += 1 + LINK_SIZE;
4966 len += 1 + LINK_SIZE;
4967 tempcode[0] = OP_ONCE;
4968 *code++ = OP_KET;
4969 PUTINC(code, 0, len);
4970 PUT(tempcode, 1, len);
4971 break;
4972 }
4973 }
4974
4975 /* In all case we no longer have a previous item. We also set the
4976 "follows varying string" flag for subsequently encountered reqbytes if
4977 it isn't already set and we have just passed a varying length item. */
4978
4979 END_REPEAT:
4980 previous = NULL;
4981 cd->req_varyopt |= reqvary;
4982 break;
4983
4984
4985 /* ===================================================================*/
4986 /* Start of nested parenthesized sub-expression, or comment or lookahead or
4987 lookbehind or option setting or condition or all the other extended
4988 parenthesis forms. */
4989
4990 case CHAR_LEFT_PARENTHESIS:
4991 newoptions = options;
4992 skipbytes = 0;
4993 bravalue = OP_CBRA;
4994 save_hwm = cd->hwm;
4995 reset_bracount = FALSE;
4996
4997 /* First deal with various "verbs" that can be introduced by '*'. */
4998
4999 if (*(++ptr) == CHAR_ASTERISK &&
5000 ((cd->ctypes[ptr[1]] & ctype_letter) != 0 || ptr[1] == ':'))
5001 {
5002 int i, namelen;
5003 int arglen = 0;
5004 const char *vn = verbnames;
5005 const uschar *name = ptr + 1;
5006 const uschar *arg = NULL;
5007 previous = NULL;
5008 while ((cd->ctypes[*++ptr] & ctype_letter) != 0) {};
5009 namelen = (int)(ptr - name);
5010
5011 /* It appears that Perl allows any characters whatsoever, other than
5012 a closing parenthesis, to appear in arguments, so we no longer insist on
5013 letters, digits, and underscores. */
5014
5015 if (*ptr == CHAR_COLON)
5016 {
5017 arg = ++ptr;
5018 while (*ptr != 0 && *ptr != CHAR_RIGHT_PARENTHESIS) ptr++;
5019 arglen = (int)(ptr - arg);
5020 }
5021
5022 if (*ptr != CHAR_RIGHT_PARENTHESIS)
5023 {
5024 *errorcodeptr = ERR60;
5025 goto FAILED;
5026 }
5027
5028 /* Scan the table of verb names */
5029
5030 for (i = 0; i < verbcount; i++)
5031 {
5032 if (namelen == verbs[i].len &&
5033 strncmp((char *)name, vn, namelen) == 0)
5034 {
5035 /* Check for open captures before ACCEPT and convert it to
5036 ASSERT_ACCEPT if in an assertion. */
5037
5038 if (verbs[i].op == OP_ACCEPT)
5039 {
5040 open_capitem *oc;
5041 if (arglen != 0)
5042 {
5043 *errorcodeptr = ERR59;
5044 goto FAILED;
5045 }
5046 cd->had_accept = TRUE;
5047 for (oc = cd->open_caps; oc != NULL; oc = oc->next)
5048 {
5049 *code++ = OP_CLOSE;
5050 PUT2INC(code, 0, oc->number);
5051 }
5052 *code++ = (cd->assert_depth > 0)? OP_ASSERT_ACCEPT : OP_ACCEPT;
5053
5054 /* Do not set firstbyte after *ACCEPT */
5055 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
5056 }
5057
5058 /* Handle other cases with/without an argument */
5059
5060 else if (arglen == 0)
5061 {
5062 if (verbs[i].op < 0) /* Argument is mandatory */
5063 {
5064 *errorcodeptr = ERR66;
5065 goto FAILED;
5066 }
5067 *code = verbs[i].op;
5068 if (*code++ == OP_THEN) cd->external_flags |= PCRE_HASTHEN;
5069 }
5070
5071 else
5072 {
5073 if (verbs[i].op_arg < 0) /* Argument is forbidden */
5074 {
5075 *errorcodeptr = ERR59;
5076 goto FAILED;
5077 }
5078 *code = verbs[i].op_arg;
5079 if (*code++ == OP_THEN_ARG) cd->external_flags |= PCRE_HASTHEN;
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 tempbracount = cd->bracount; /* Save value before bracket */
5915 length_prevgroup = 0; /* Initialize for pre-compile phase */
5916
5917 if (!compile_regex(
5918 newoptions, /* The complete new option state */
5919 &tempcode, /* Where to put code (updated) */
5920 &ptr, /* Input pointer (updated) */
5921 errorcodeptr, /* Where to put an error message */
5922 (bravalue == OP_ASSERTBACK ||
5923 bravalue == OP_ASSERTBACK_NOT), /* TRUE if back assert */
5924 reset_bracount, /* True if (?| group */
5925 skipbytes, /* Skip over bracket number */
5926 cond_depth +
5927 ((bravalue == OP_COND)?1:0), /* Depth of condition subpatterns */
5928 &subfirstbyte, /* For possible first char */
5929 &subreqbyte, /* For possible last char */
5930 bcptr, /* Current branch chain */
5931 cd, /* Tables block */
5932 (lengthptr == NULL)? NULL : /* Actual compile phase */
5933 &length_prevgroup /* Pre-compile phase */
5934 ))
5935 goto FAILED;
5936
5937 /* If this was an atomic group and there are no capturing groups within it,
5938 generate OP_ONCE_NC instead of OP_ONCE. */
5939
5940 if (bravalue == OP_ONCE && cd->bracount <= tempbracount)
5941 *code = OP_ONCE_NC;
5942
5943 if (bravalue >= OP_ASSERT && bravalue <= OP_ASSERTBACK_NOT)
5944 cd->assert_depth -= 1;
5945
5946 /* At the end of compiling, code is still pointing to the start of the
5947 group, while tempcode has been updated to point past the end of the group.
5948 The pattern pointer (ptr) is on the bracket.
5949
5950 If this is a conditional bracket, check that there are no more than
5951 two branches in the group, or just one if it's a DEFINE group. We do this
5952 in the real compile phase, not in the pre-pass, where the whole group may
5953 not be available. */
5954
5955 if (bravalue == OP_COND && lengthptr == NULL)
5956 {
5957 uschar *tc = code;
5958 int condcount = 0;
5959
5960 do {
5961 condcount++;
5962 tc += GET(tc,1);
5963 }
5964 while (*tc != OP_KET);
5965
5966 /* A DEFINE group is never obeyed inline (the "condition" is always
5967 false). It must have only one branch. */
5968
5969 if (code[LINK_SIZE+1] == OP_DEF)
5970 {
5971 if (condcount > 1)
5972 {
5973 *errorcodeptr = ERR54;
5974 goto FAILED;
5975 }
5976 bravalue = OP_DEF; /* Just a flag to suppress char handling below */
5977 }
5978
5979 /* A "normal" conditional group. If there is just one branch, we must not
5980 make use of its firstbyte or reqbyte, because this is equivalent to an
5981 empty second branch. */
5982
5983 else
5984 {
5985 if (condcount > 2)
5986 {
5987 *errorcodeptr = ERR27;
5988 goto FAILED;
5989 }
5990 if (condcount == 1) subfirstbyte = subreqbyte = REQ_NONE;
5991 }
5992 }
5993
5994 /* Error if hit end of pattern */
5995
5996 if (*ptr != CHAR_RIGHT_PARENTHESIS)
5997 {
5998 *errorcodeptr = ERR14;
5999 goto FAILED;
6000 }
6001
6002 /* In the pre-compile phase, update the length by the length of the group,
6003 less the brackets at either end. Then reduce the compiled code to just a
6004 set of non-capturing brackets so that it doesn't use much memory if it is
6005 duplicated by a quantifier.*/
6006
6007 if (lengthptr != NULL)
6008 {
6009 if (OFLOW_MAX - *lengthptr < length_prevgroup - 2 - 2*LINK_SIZE)
6010 {
6011 *errorcodeptr = ERR20;
6012 goto FAILED;
6013 }
6014 *lengthptr += length_prevgroup - 2 - 2*LINK_SIZE;
6015 code++; /* This already contains bravalue */
6016 PUTINC(code, 0, 1 + LINK_SIZE);
6017 *code++ = OP_KET;
6018 PUTINC(code, 0, 1 + LINK_SIZE);
6019 break; /* No need to waste time with special character handling */
6020 }
6021
6022 /* Otherwise update the main code pointer to the end of the group. */
6023
6024 code = tempcode;
6025
6026 /* For a DEFINE group, required and first character settings are not
6027 relevant. */
6028
6029 if (bravalue == OP_DEF) break;
6030
6031 /* Handle updating of the required and first characters for other types of
6032 group. Update for normal brackets of all kinds, and conditions with two
6033 branches (see code above). If the bracket is followed by a quantifier with
6034 zero repeat, we have to back off. Hence the definition of zeroreqbyte and
6035 zerofirstbyte outside the main loop so that they can be accessed for the
6036 back off. */
6037
6038 zeroreqbyte = reqbyte;
6039 zerofirstbyte = firstbyte;
6040 groupsetfirstbyte = FALSE;
6041
6042 if (bravalue >= OP_ONCE)
6043 {
6044 /* If we have not yet set a firstbyte in this branch, take it from the
6045 subpattern, remembering that it was set here so that a repeat of more
6046 than one can replicate it as reqbyte if necessary. If the subpattern has
6047 no firstbyte, set "none" for the whole branch. In both cases, a zero
6048 repeat forces firstbyte to "none". */
6049
6050 if (firstbyte == REQ_UNSET)
6051 {
6052 if (subfirstbyte >= 0)
6053 {
6054 firstbyte = subfirstbyte;
6055 groupsetfirstbyte = TRUE;
6056 }
6057 else firstbyte = REQ_NONE;
6058 zerofirstbyte = REQ_NONE;
6059 }
6060
6061 /* If firstbyte was previously set, convert the subpattern's firstbyte
6062 into reqbyte if there wasn't one, using the vary flag that was in
6063 existence beforehand. */
6064
6065 else if (subfirstbyte >= 0 && subreqbyte < 0)
6066 subreqbyte = subfirstbyte | tempreqvary;
6067
6068 /* If the subpattern set a required byte (or set a first byte that isn't
6069 really the first byte - see above), set it. */
6070
6071 if (subreqbyte >= 0) reqbyte = subreqbyte;
6072 }
6073
6074 /* For a forward assertion, we take the reqbyte, if set. This can be
6075 helpful if the pattern that follows the assertion doesn't set a different
6076 char. For example, it's useful for /(?=abcde).+/. We can't set firstbyte
6077 for an assertion, however because it leads to incorrect effect for patterns
6078 such as /(?=a)a.+/ when the "real" "a" would then become a reqbyte instead
6079 of a firstbyte. This is overcome by a scan at the end if there's no
6080 firstbyte, looking for an asserted first char. */
6081
6082 else if (bravalue == OP_ASSERT && subreqbyte >= 0) reqbyte = subreqbyte;
6083 break; /* End of processing '(' */
6084
6085
6086 /* ===================================================================*/
6087 /* Handle metasequences introduced by \. For ones like \d, the ESC_ values
6088 are arranged to be the negation of the corresponding OP_values in the
6089 default case when PCRE_UCP is not set. For the back references, the values
6090 are ESC_REF plus the reference number. Only back references and those types
6091 that consume a character may be repeated. We can test for values between
6092 ESC_b and ESC_Z for the latter; this may have to change if any new ones are
6093 ever created. */
6094
6095 case CHAR_BACKSLASH:
6096 tempptr = ptr;
6097 c = check_escape(&ptr, errorcodeptr, cd->bracount, options, FALSE);
6098 if (*errorcodeptr != 0) goto FAILED;
6099
6100 if (c < 0)
6101 {
6102 if (-c == ESC_Q) /* Handle start of quoted string */
6103 {
6104 if (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
6105 ptr += 2; /* avoid empty string */
6106 else inescq = TRUE;
6107 continue;
6108 }
6109
6110 if (-c == ESC_E) continue; /* Perl ignores an orphan \E */
6111
6112 /* For metasequences that actually match a character, we disable the
6113 setting of a first character if it hasn't already been set. */
6114
6115 if (firstbyte == REQ_UNSET && -c > ESC_b && -c < ESC_Z)
6116 firstbyte = REQ_NONE;
6117
6118 /* Set values to reset to if this is followed by a zero repeat. */
6119
6120 zerofirstbyte = firstbyte;
6121 zeroreqbyte = reqbyte;
6122
6123 /* \g<name> or \g'name' is a subroutine call by name and \g<n> or \g'n'
6124 is a subroutine call by number (Oniguruma syntax). In fact, the value
6125 -ESC_g is returned only for these cases. So we don't need to check for <
6126 or ' if the value is -ESC_g. For the Perl syntax \g{n} the value is
6127 -ESC_REF+n, and for the Perl syntax \g{name} the result is -ESC_k (as
6128 that is a synonym for a named back reference). */
6129
6130 if (-c == ESC_g)
6131 {
6132 const uschar *p;
6133 save_hwm = cd->hwm; /* Normally this is set when '(' is read */
6134 terminator = (*(++ptr) == CHAR_LESS_THAN_SIGN)?
6135 CHAR_GREATER_THAN_SIGN : CHAR_APOSTROPHE;
6136
6137 /* These two statements stop the compiler for warning about possibly
6138 unset variables caused by the jump to HANDLE_NUMERICAL_RECURSION. In
6139 fact, because we actually check for a number below, the paths that
6140 would actually be in error are never taken. */
6141
6142 skipbytes = 0;
6143 reset_bracount = FALSE;
6144
6145 /* Test for a name */
6146
6147 if (ptr[1] != CHAR_PLUS && ptr[1] != CHAR_MINUS)
6148 {
6149 BOOL isnumber = TRUE;
6150 for (p = ptr + 1; *p != 0 && *p != terminator; p++)
6151 {
6152 if ((cd->ctypes[*p] & ctype_digit) == 0) isnumber = FALSE;
6153 if ((cd->ctypes[*p] & ctype_word) == 0) break;
6154 }
6155 if (*p != terminator)
6156 {
6157 *errorcodeptr = ERR57;
6158 break;
6159 }
6160 if (isnumber)
6161 {
6162 ptr++;
6163 goto HANDLE_NUMERICAL_RECURSION;
6164 }
6165 is_recurse = TRUE;
6166 goto NAMED_REF_OR_RECURSE;
6167 }
6168
6169 /* Test a signed number in angle brackets or quotes. */
6170
6171 p = ptr + 2;
6172 while ((digitab[*p] & ctype_digit) != 0) p++;
6173 if (*p != terminator)
6174 {
6175 *errorcodeptr = ERR57;
6176 break;
6177 }
6178 ptr++;
6179 goto HANDLE_NUMERICAL_RECURSION;
6180 }
6181
6182 /* \k<name> or \k'name' is a back reference by name (Perl syntax).
6183 We also support \k{name} (.NET syntax). */
6184
6185 if (-c == ESC_k)
6186 {
6187 if ((ptr[1] != CHAR_LESS_THAN_SIGN &&
6188 ptr[1] != CHAR_APOSTROPHE && ptr[1] != CHAR_LEFT_CURLY_BRACKET))
6189 {
6190 *errorcodeptr = ERR69;
6191 break;
6192 }
6193 is_recurse = FALSE;
6194 terminator = (*(++ptr) == CHAR_LESS_THAN_SIGN)?
6195 CHAR_GREATER_THAN_SIGN : (*ptr == CHAR_APOSTROPHE)?
6196 CHAR_APOSTROPHE : CHAR_RIGHT_CURLY_BRACKET;
6197 goto NAMED_REF_OR_RECURSE;
6198 }
6199
6200 /* Back references are handled specially; must disable firstbyte if
6201 not set to cope with cases like (?=(\w+))\1: which would otherwise set
6202 ':' later. */
6203
6204 if (-c >= ESC_REF)
6205 {
6206 open_capitem *oc;
6207 recno = -c - ESC_REF;
6208
6209 HANDLE_REFERENCE: /* Come here from named backref handling */
6210 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
6211 previous = code;
6212 *code++ = ((options & PCRE_CASELESS) != 0)? OP_REFI : OP_REF;
6213 PUT2INC(code, 0, recno);
6214 cd->backref_map |= (recno < 32)? (1 << recno) : 1;
6215 if (recno > cd->top_backref) cd->top_backref = recno;
6216
6217 /* Check to see if this back reference is recursive, that it, it
6218 is inside the group that it references. A flag is set so that the
6219 group can be made atomic. */
6220
6221 for (oc = cd->open_caps; oc != NULL; oc = oc->next)
6222 {
6223 if (oc->number == recno)
6224 {
6225 oc->flag = TRUE;
6226 break;
6227 }
6228 }
6229 }
6230
6231 /* So are Unicode property matches, if supported. */
6232
6233 #ifdef SUPPORT_UCP
6234 else if (-c == ESC_P || -c == ESC_p)
6235 {
6236 BOOL negated;
6237 int pdata;
6238 int ptype = get_ucp(&ptr, &negated, &pdata, errorcodeptr);
6239 if (ptype < 0) goto FAILED;
6240 previous = code;
6241 *code++ = ((-c == ESC_p) != negated)? OP_PROP : OP_NOTPROP;
6242 *code++ = ptype;
6243 *code++ = pdata;
6244 }
6245 #else
6246
6247 /* If Unicode properties are not supported, \X, \P, and \p are not
6248 allowed. */
6249
6250 else if (-c == ESC_X || -c == ESC_P || -c == ESC_p)
6251 {
6252 *errorcodeptr = ERR45;
6253 goto FAILED;
6254 }
6255 #endif
6256
6257 /* For the rest (including \X when Unicode properties are supported), we
6258 can obtain the OP value by negating the escape value in the default
6259 situation when PCRE_UCP is not set. When it *is* set, we substitute
6260 Unicode property tests. */
6261
6262 else
6263 {
6264 #ifdef SUPPORT_UCP
6265 if (-c >= ESC_DU && -c <= ESC_wu)
6266 {
6267 nestptr = ptr + 1; /* Where to resume */
6268 ptr = substitutes[-c - ESC_DU] - 1; /* Just before substitute */
6269 }
6270 else
6271 #endif
6272 {
6273 previous = (-c > ESC_b && -c < ESC_Z)? code : NULL;
6274 *code++ = -c;
6275 }
6276 }
6277 continue;
6278 }
6279
6280 /* We have a data character whose value is in c. In UTF-8 mode it may have
6281 a value > 127. We set its representation in the length/buffer, and then
6282 handle it as a data character. */
6283
6284 #ifdef SUPPORT_UTF8
6285 if (utf8 && c > 127)
6286 mclength = _pcre_ord2utf8(c, mcbuffer);
6287 else
6288 #endif
6289
6290 {
6291 mcbuffer[0] = c;
6292 mclength = 1;
6293 }
6294 goto ONE_CHAR;
6295
6296
6297 /* ===================================================================*/
6298 /* Handle a literal character. It is guaranteed not to be whitespace or #
6299 when the extended flag is set. If we are in UTF-8 mode, it may be a
6300 multi-byte literal character. */
6301
6302 default:
6303 NORMAL_CHAR:
6304 mclength = 1;
6305 mcbuffer[0] = c;
6306
6307 #ifdef SUPPORT_UTF8
6308 if (utf8 && c >= 0xc0)
6309 {
6310 while ((ptr[1] & 0xc0) == 0x80)
6311 mcbuffer[mclength++] = *(++ptr);
6312 }
6313 #endif
6314
6315 /* At this point we have the character's bytes in mcbuffer, and the length
6316 in mclength. When not in UTF-8 mode, the length is always 1. */
6317
6318 ONE_CHAR:
6319 previous = code;
6320 *code++ = ((options & PCRE_CASELESS) != 0)? OP_CHARI : OP_CHAR;
6321 for (c = 0; c < mclength; c++) *code++ = mcbuffer[c];
6322
6323 /* Remember if \r or \n were seen */
6324
6325 if (mcbuffer[0] == CHAR_CR || mcbuffer[0] == CHAR_NL)
6326 cd->external_flags |= PCRE_HASCRORLF;
6327
6328 /* Set the first and required bytes appropriately. If no previous first
6329 byte, set it from this character, but revert to none on a zero repeat.
6330 Otherwise, leave the firstbyte value alone, and don't change it on a zero
6331 repeat. */
6332
6333 if (firstbyte == REQ_UNSET)
6334 {
6335 zerofirstbyte = REQ_NONE;
6336 zeroreqbyte = reqbyte;
6337
6338 /* If the character is more than one byte long, we can set firstbyte
6339 only if it is not to be matched caselessly. */
6340
6341 if (mclength == 1 || req_caseopt == 0)
6342 {
6343 firstbyte = mcbuffer[0] | req_caseopt;
6344 if (mclength != 1) reqbyte = code[-1] | cd->req_varyopt;
6345 }
6346 else firstbyte = reqbyte = REQ_NONE;
6347 }
6348
6349 /* firstbyte was previously set; we can set reqbyte only if the length is
6350 1 or the matching is caseful. */
6351
6352 else
6353 {
6354 zerofirstbyte = firstbyte;
6355 zeroreqbyte = reqbyte;
6356 if (mclength == 1 || req_caseopt == 0)
6357 reqbyte = code[-1] | req_caseopt | cd->req_varyopt;
6358 }
6359
6360 break; /* End of literal character handling */
6361 }
6362 } /* end of big loop */
6363
6364
6365 /* Control never reaches here by falling through, only by a goto for all the
6366 error states. Pass back the position in the pattern so that it can be displayed
6367 to the user for diagnosing the error. */
6368
6369 FAILED:
6370 *ptrptr = ptr;
6371 return FALSE;
6372 }
6373
6374
6375
6376
6377 /*************************************************
6378 * Compile sequence of alternatives *
6379 *************************************************/
6380
6381 /* On entry, ptr is pointing past the bracket character, but on return it
6382 points to the closing bracket, or vertical bar, or end of string. The code
6383 variable is pointing at the byte into which the BRA operator has been stored.
6384 This function is used during the pre-compile phase when we are trying to find
6385 out the amount of memory needed, as well as during the real compile phase. The
6386 value of lengthptr distinguishes the two phases.
6387
6388 Arguments:
6389 options option bits, including any changes for this subpattern
6390 codeptr -> the address of the current code pointer
6391 ptrptr -> the address of the current pattern pointer
6392 errorcodeptr -> pointer to error code variable
6393 lookbehind TRUE if this is a lookbehind assertion
6394 reset_bracount TRUE to reset the count for each branch
6395 skipbytes skip this many bytes at start (for brackets and OP_COND)
6396 cond_depth depth of nesting for conditional subpatterns
6397 firstbyteptr place to put the first required character, or a negative number
6398 reqbyteptr place to put the last required character, or a negative number
6399 bcptr pointer to the chain of currently open branches
6400 cd points to the data block with tables pointers etc.
6401 lengthptr NULL during the real compile phase
6402 points to length accumulator during pre-compile phase
6403
6404 Returns: TRUE on success
6405 */
6406
6407 static BOOL
6408 compile_regex(int options, uschar **codeptr, const uschar **ptrptr,
6409 int *errorcodeptr, BOOL lookbehind, BOOL reset_bracount, int skipbytes,
6410 int cond_depth, int *firstbyteptr, int *reqbyteptr, branch_chain *bcptr,
6411 compile_data *cd, int *lengthptr)
6412 {
6413 const uschar *ptr = *ptrptr;
6414 uschar *code = *codeptr;
6415 uschar *last_branch = code;
6416 uschar *start_bracket = code;
6417 uschar *reverse_count = NULL;
6418 open_capitem capitem;
6419 int capnumber = 0;
6420 int firstbyte, reqbyte;
6421 int branchfirstbyte, branchreqbyte;
6422 int length;
6423 int orig_bracount;
6424 int max_bracount;
6425 branch_chain bc;
6426
6427 bc.outer = bcptr;
6428 bc.current_branch = code;
6429
6430 firstbyte = reqbyte = REQ_UNSET;
6431
6432 /* Accumulate the length for use in the pre-compile phase. Start with the
6433 length of the BRA and KET and any extra bytes that are required at the
6434 beginning. We accumulate in a local variable to save frequent testing of
6435 lenthptr for NULL. We cannot do this by looking at the value of code at the
6436 start and end of each alternative, because compiled items are discarded during
6437 the pre-compile phase so that the work space is not exceeded. */
6438
6439 length = 2 + 2*LINK_SIZE + skipbytes;
6440
6441 /* WARNING: If the above line is changed for any reason, you must also change
6442 the code that abstracts option settings at the start of the pattern and makes
6443 them global. It tests the value of length for (2 + 2*LINK_SIZE) in the
6444 pre-compile phase to find out whether anything has yet been compiled or not. */
6445
6446 /* If this is a capturing subpattern, add to the chain of open capturing items
6447 so that we can detect them if (*ACCEPT) is encountered. This is also used to
6448 detect groups that contain recursive back references to themselves. Note that
6449 only OP_CBRA need be tested here; changing this opcode to one of its variants,
6450 e.g. OP_SCBRAPOS, happens later, after the group has been compiled. */
6451
6452 if (*code == OP_CBRA)
6453 {
6454 capnumber = GET2(code, 1 + LINK_SIZE);
6455 capitem.number = capnumber;
6456 capitem.next = cd->open_caps;
6457 capitem.flag = FALSE;
6458 cd->open_caps = &capitem;
6459 }
6460
6461 /* Offset is set zero to mark that this bracket is still open */
6462
6463 PUT(code, 1, 0);
6464 code += 1 + LINK_SIZE + skipbytes;
6465
6466 /* Loop for each alternative branch */
6467
6468 orig_bracount = max_bracount = cd->bracount;
6469 for (;;)
6470 {
6471 /* For a (?| group, reset the capturing bracket count so that each branch
6472 uses the same numbers. */
6473
6474 if (reset_bracount) cd->bracount = orig_bracount;
6475
6476 /* Set up dummy OP_REVERSE if lookbehind assertion */
6477
6478 if (lookbehind)
6479 {
6480 *code++ = OP_REVERSE;
6481 reverse_count = code;
6482 PUTINC(code, 0, 0);
6483 length += 1 + LINK_SIZE;
6484 }
6485
6486 /* Now compile the branch; in the pre-compile phase its length gets added
6487 into the length. */
6488
6489 if (!compile_branch(&options, &code, &ptr, errorcodeptr, &branchfirstbyte,
6490 &branchreqbyte, &bc, cond_depth, cd,
6491 (lengthptr == NULL)? NULL : &length))
6492 {
6493 *ptrptr = ptr;
6494 return FALSE;
6495 }
6496
6497 /* Keep the highest bracket count in case (?| was used and some branch
6498 has fewer than the rest. */
6499
6500 if (cd->bracount > max_bracount) max_bracount = cd->bracount;
6501
6502 /* In the real compile phase, there is some post-processing to be done. */
6503
6504 if (lengthptr == NULL)
6505 {
6506 /* If this is the first branch, the firstbyte and reqbyte values for the
6507 branch become the values for the regex. */
6508
6509 if (*last_branch != OP_ALT)
6510 {
6511 firstbyte = branchfirstbyte;
6512 reqbyte = branchreqbyte;
6513 }
6514
6515 /* If this is not the first branch, the first char and reqbyte have to
6516 match the values from all the previous branches, except that if the
6517 previous value for reqbyte didn't have REQ_VARY set, it can still match,
6518 and we set REQ_VARY for the regex. */
6519
6520 else
6521 {
6522 /* If we previously had a firstbyte, but it doesn't match the new branch,
6523 we have to abandon the firstbyte for the regex, but if there was
6524 previously no reqbyte, it takes on the value of the old firstbyte. */
6525
6526 if (firstbyte >= 0 && firstbyte != branchfirstbyte)
6527 {
6528 if (reqbyte < 0) reqbyte = firstbyte;
6529 firstbyte = REQ_NONE;
6530 }
6531
6532 /* If we (now or from before) have no firstbyte, a firstbyte from the
6533 branch becomes a reqbyte if there isn't a branch reqbyte. */
6534
6535 if (firstbyte < 0 && branchfirstbyte >= 0 && branchreqbyte < 0)
6536 branchreqbyte = branchfirstbyte;
6537
6538 /* Now ensure that the reqbytes match */
6539
6540 if ((reqbyte & ~REQ_VARY) != (branchreqbyte & ~REQ_VARY))
6541 reqbyte = REQ_NONE;
6542 else reqbyte |= branchreqbyte; /* To "or" REQ_VARY */
6543 }
6544
6545 /* If lookbehind, check that this branch matches a fixed-length string, and
6546 put the length into the OP_REVERSE item. Temporarily mark the end of the
6547 branch with OP_END. If the branch contains OP_RECURSE, the result is -3
6548 because there may be forward references that we can't check here. Set a
6549 flag to cause another lookbehind check at the end. Why not do it all at the
6550 end? Because common, erroneous checks are picked up here and the offset of
6551 the problem can be shown. */
6552
6553 if (lookbehind)
6554 {
6555 int fixed_length;
6556 *code = OP_END;
6557 fixed_length = find_fixedlength(last_branch, (options & PCRE_UTF8) != 0,
6558 FALSE, cd);
6559 DPRINTF(("fixed length = %d\n", fixed_length));
6560 if (fixed_length == -3)
6561 {
6562 cd->check_lookbehind = TRUE;
6563 }
6564 else if (fixed_length < 0)
6565 {
6566 *errorcodeptr = (fixed_length == -2)? ERR36 : ERR25;
6567 *ptrptr = ptr;
6568 return FALSE;
6569 }
6570 else { PUT(reverse_count, 0, fixed_length); }
6571 }
6572 }
6573
6574 /* Reached end of expression, either ')' or end of pattern. In the real
6575 compile phase, go back through the alternative branches and reverse the chain
6576 of offsets, with the field in the BRA item now becoming an offset to the
6577 first alternative. If there are no alternatives, it points to the end of the
6578 group. The length in the terminating ket is always the length of the whole
6579 bracketed item. Return leaving the pointer at the terminating char. */
6580
6581 if (*ptr != CHAR_VERTICAL_LINE)
6582 {
6583 if (lengthptr == NULL)
6584 {
6585 int branch_length = (int)(code - last_branch);
6586 do
6587 {
6588 int prev_length = GET(last_branch, 1);
6589 PUT(last_branch, 1, branch_length);
6590 branch_length = prev_length;
6591 last_branch -= branch_length;
6592 }
6593 while (branch_length > 0);
6594 }
6595
6596 /* Fill in the ket */
6597
6598 *code = OP_KET;
6599 PUT(code, 1, (int)(code - start_bracket));
6600 code += 1 + LINK_SIZE;
6601
6602 /* If it was a capturing subpattern, check to see if it contained any
6603 recursive back references. If so, we must wrap it in atomic brackets.
6604 In any event, remove the block from the chain. */
6605
6606 if (capnumber > 0)
6607 {
6608 if (cd->open_caps->flag)
6609 {
6610 memmove(start_bracket + 1 + LINK_SIZE, start_bracket,
6611 code - start_bracket);
6612 *start_bracket = OP_ONCE;
6613 code += 1 + LINK_SIZE;
6614 PUT(start_bracket, 1, (int)(code - start_bracket));
6615 *code = OP_KET;
6616 PUT(code, 1, (int)(code - start_bracket));
6617 code += 1 + LINK_SIZE;
6618 length += 2 + 2*LINK_SIZE;
6619 }
6620 cd->open_caps = cd->open_caps->next;
6621 }
6622
6623 /* Retain the highest bracket number, in case resetting was used. */
6624
6625 cd->bracount = max_bracount;
6626
6627 /* Set values to pass back */
6628
6629 *codeptr = code;
6630 *ptrptr = ptr;
6631 *firstbyteptr = firstbyte;
6632 *reqbyteptr = reqbyte;
6633 if (lengthptr != NULL)
6634 {
6635 if (OFLOW_MAX - *lengthptr < length)
6636 {
6637 *errorcodeptr = ERR20;
6638 return FALSE;
6639 }
6640 *lengthptr += length;
6641 }
6642 return TRUE;
6643 }
6644
6645 /* Another branch follows. In the pre-compile phase, we can move the code
6646 pointer back to where it was for the start of the first branch. (That is,
6647 pretend that each branch is the only one.)
6648
6649 In the real compile phase, insert an ALT node. Its length field points back
6650 to the previous branch while the bracket remains open. At the end the chain
6651 is reversed. It's done like this so that the start of the bracket has a
6652 zero offset until it is closed, making it possible to detect recursion. */
6653