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

Contents of /code/trunk/pcre_compile.c

Parent Directory Parent Directory | Revision Log Revision Log


Revision 640 - (show annotations) (download)
Mon Jul 25 10:50:28 2011 UTC (3 years, 2 months ago) by ph10
File MIME type: text/plain
File size: 242872 byte(s)
Fix three compile-time bugs (Bugzilla #1123).

</
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 *, branch_chain *, compile_data *, int *);
551
552
553
554 /*************************************************
555 * Find an error text *
556 *************************************************/
557
558 /* The error texts are now all in one long string, to save on relocations. As
559 some of the text is of unknown length, we can't use a table of offsets.
560 Instead, just count through the strings. This is not a performance issue
561 because it happens only when there has been a compilation error.
562
563 Argument: the error number
564 Returns: pointer to the error string
565 */
566
567 static const char *
568 find_error_text(int n)
569 {
570 const char *s = error_texts;
571 for (; n > 0; n--)
572 {
573 while (*s++ != 0) {};
574 if (*s == 0) return "Error text not found (please report)";
575 }
576 return s;
577 }
578
579
580 /*************************************************
581 * Check for counted repeat *
582 *************************************************/
583
584 /* This function is called when a '{' is encountered in a place where it might
585 start a quantifier. It looks ahead to see if it really is a quantifier or not.
586 It is only a quantifier if it is one of the forms {ddd} {ddd,} or {ddd,ddd}
587 where the ddds are digits.
588
589 Arguments:
590 p pointer to the first char after '{'
591
592 Returns: TRUE or FALSE
593 */
594
595 static BOOL
596 is_counted_repeat(const uschar *p)
597 {
598 if ((digitab[*p++] & ctype_digit) == 0) return FALSE;
599 while ((digitab[*p] & ctype_digit) != 0) p++;
600 if (*p == CHAR_RIGHT_CURLY_BRACKET) return TRUE;
601
602 if (*p++ != CHAR_COMMA) return FALSE;
603 if (*p == CHAR_RIGHT_CURLY_BRACKET) return TRUE;
604
605 if ((digitab[*p++] & ctype_digit) == 0) return FALSE;
606 while ((digitab[*p] & ctype_digit) != 0) p++;
607
608 return (*p == CHAR_RIGHT_CURLY_BRACKET);
609 }
610
611
612
613 /*************************************************
614 * Handle escapes *
615 *************************************************/
616
617 /* This function is called when a \ has been encountered. It either returns a
618 positive value for a simple escape such as \n, or a negative value which
619 encodes one of the more complicated things such as \d. A backreference to group
620 n is returned as -(ESC_REF + n); ESC_REF is the highest ESC_xxx macro. When
621 UTF-8 is enabled, a positive value greater than 255 may be returned. On entry,
622 ptr is pointing at the \. On exit, it is on the final character of the escape
623 sequence.
624
625 Arguments:
626 ptrptr points to the pattern position pointer
627 errorcodeptr points to the errorcode variable
628 bracount number of previous extracting brackets
629 options the options bits
630 isclass TRUE if inside a character class
631
632 Returns: zero or positive => a data character
633 negative => a special escape sequence
634 on error, errorcodeptr is set
635 */
636
637 static int
638 check_escape(const uschar **ptrptr, int *errorcodeptr, int bracount,
639 int options, BOOL isclass)
640 {
641 BOOL utf8 = (options & PCRE_UTF8) != 0;
642 const uschar *ptr = *ptrptr + 1;
643 int c, i;
644
645 GETCHARINCTEST(c, ptr); /* Get character value, increment pointer */
646 ptr--; /* Set pointer back to the last byte */
647
648 /* If backslash is at the end of the pattern, it's an error. */
649
650 if (c == 0) *errorcodeptr = ERR1;
651
652 /* Non-alphanumerics are literals. For digits or letters, do an initial lookup
653 in a table. A non-zero result is something that can be returned immediately.
654 Otherwise further processing may be required. */
655
656 #ifndef EBCDIC /* ASCII/UTF-8 coding */
657 else if (c < CHAR_0 || c > CHAR_z) {} /* Not alphanumeric */
658 else if ((i = escapes[c - CHAR_0]) != 0) c = i;
659
660 #else /* EBCDIC coding */
661 else if (c < 'a' || (ebcdic_chartab[c] & 0x0E) == 0) {} /* Not alphanumeric */
662 else if ((i = escapes[c - 0x48]) != 0) c = i;
663 #endif
664
665 /* Escapes that need further processing, or are illegal. */
666
667 else
668 {
669 const uschar *oldptr;
670 BOOL braced, negated;
671
672 switch (c)
673 {
674 /* A number of Perl escapes are not handled by PCRE. We give an explicit
675 error. */
676
677 case CHAR_l:
678 case CHAR_L:
679 case CHAR_u:
680 case CHAR_U:
681 *errorcodeptr = ERR37;
682 break;
683
684 /* In a character class, \g is just a literal "g". Outside a character
685 class, \g must be followed by one of a number of specific things:
686
687 (1) A number, either plain or braced. If positive, it is an absolute
688 backreference. If negative, it is a relative backreference. This is a Perl
689 5.10 feature.
690
691 (2) Perl 5.10 also supports \g{name} as a reference to a named group. This
692 is part of Perl's movement towards a unified syntax for back references. As
693 this is synonymous with \k{name}, we fudge it up by pretending it really
694 was \k.
695
696 (3) For Oniguruma compatibility we also support \g followed by a name or a
697 number either in angle brackets or in single quotes. However, these are
698 (possibly recursive) subroutine calls, _not_ backreferences. Just return
699 the -ESC_g code (cf \k). */
700
701 case CHAR_g:
702 if (isclass) break;
703 if (ptr[1] == CHAR_LESS_THAN_SIGN || ptr[1] == CHAR_APOSTROPHE)
704 {
705 c = -ESC_g;
706 break;
707 }
708
709 /* Handle the Perl-compatible cases */
710
711 if (ptr[1] == CHAR_LEFT_CURLY_BRACKET)
712 {
713 const uschar *p;
714 for (p = ptr+2; *p != 0 && *p != CHAR_RIGHT_CURLY_BRACKET; p++)
715 if (*p != CHAR_MINUS && (digitab[*p] & ctype_digit) == 0) break;
716 if (*p != 0 && *p != CHAR_RIGHT_CURLY_BRACKET)
717 {
718 c = -ESC_k;
719 break;
720 }
721 braced = TRUE;
722 ptr++;
723 }
724 else braced = FALSE;
725
726 if (ptr[1] == CHAR_MINUS)
727 {
728 negated = TRUE;
729 ptr++;
730 }
731 else negated = FALSE;
732
733 c = 0;
734 while ((digitab[ptr[1]] & ctype_digit) != 0)
735 c = c * 10 + *(++ptr) - CHAR_0;
736
737 if (c < 0) /* Integer overflow */
738 {
739 *errorcodeptr = ERR61;
740 break;
741 }
742
743 if (braced && *(++ptr) != CHAR_RIGHT_CURLY_BRACKET)
744 {
745 *errorcodeptr = ERR57;
746 break;
747 }
748
749 if (c == 0)
750 {
751 *errorcodeptr = ERR58;
752 break;
753 }
754
755 if (negated)
756 {
757 if (c > bracount)
758 {
759 *errorcodeptr = ERR15;
760 break;
761 }
762 c = bracount - (c - 1);
763 }
764
765 c = -(ESC_REF + c);
766 break;
767
768 /* The handling of escape sequences consisting of a string of digits
769 starting with one that is not zero is not straightforward. By experiment,
770 the way Perl works seems to be as follows:
771
772 Outside a character class, the digits are read as a decimal number. If the
773 number is less than 10, or if there are that many previous extracting
774 left brackets, then it is a back reference. Otherwise, up to three octal
775 digits are read to form an escaped byte. Thus \123 is likely to be octal
776 123 (cf \0123, which is octal 012 followed by the literal 3). If the octal
777 value is greater than 377, the least significant 8 bits are taken. Inside a
778 character class, \ followed by a digit is always an octal number. */
779
780 case CHAR_1: case CHAR_2: case CHAR_3: case CHAR_4: case CHAR_5:
781 case CHAR_6: case CHAR_7: case CHAR_8: case CHAR_9:
782
783 if (!isclass)
784 {
785 oldptr = ptr;
786 c -= CHAR_0;
787 while ((digitab[ptr[1]] & ctype_digit) != 0)
788 c = c * 10 + *(++ptr) - CHAR_0;
789 if (c < 0) /* Integer overflow */
790 {
791 *errorcodeptr = ERR61;
792 break;
793 }
794 if (c < 10 || c <= bracount)
795 {
796 c = -(ESC_REF + c);
797 break;
798 }
799 ptr = oldptr; /* Put the pointer back and fall through */
800 }
801
802 /* Handle an octal number following \. If the first digit is 8 or 9, Perl
803 generates a binary zero byte and treats the digit as a following literal.
804 Thus we have to pull back the pointer by one. */
805
806 if ((c = *ptr) >= CHAR_8)
807 {
808 ptr--;
809 c = 0;
810 break;
811 }
812
813 /* \0 always starts an octal number, but we may drop through to here with a
814 larger first octal digit. The original code used just to take the least
815 significant 8 bits of octal numbers (I think this is what early Perls used
816 to do). Nowadays we allow for larger numbers in UTF-8 mode, but no more
817 than 3 octal digits. */
818
819 case CHAR_0:
820 c -= CHAR_0;
821 while(i++ < 2 && ptr[1] >= CHAR_0 && ptr[1] <= CHAR_7)
822 c = c * 8 + *(++ptr) - CHAR_0;
823 if (!utf8 && c > 255) *errorcodeptr = ERR51;
824 break;
825
826 /* \x is complicated. \x{ddd} is a character number which can be greater
827 than 0xff in utf8 mode, but only if the ddd are hex digits. If not, { is
828 treated as a data character. */
829
830 case CHAR_x:
831 if (ptr[1] == CHAR_LEFT_CURLY_BRACKET)
832 {
833 const uschar *pt = ptr + 2;
834 int count = 0;
835
836 c = 0;
837 while ((digitab[*pt] & ctype_xdigit) != 0)
838 {
839 register int cc = *pt++;
840 if (c == 0 && cc == CHAR_0) continue; /* Leading zeroes */
841 count++;
842
843 #ifndef EBCDIC /* ASCII/UTF-8 coding */
844 if (cc >= CHAR_a) cc -= 32; /* Convert to upper case */
845 c = (c << 4) + cc - ((cc < CHAR_A)? CHAR_0 : (CHAR_A - 10));
846 #else /* EBCDIC coding */
847 if (cc >= CHAR_a && cc <= CHAR_z) cc += 64; /* Convert to upper case */
848 c = (c << 4) + cc - ((cc >= CHAR_0)? CHAR_0 : (CHAR_A - 10));
849 #endif
850 }
851
852 if (*pt == CHAR_RIGHT_CURLY_BRACKET)
853 {
854 if (c < 0 || count > (utf8? 8 : 2)) *errorcodeptr = ERR34;
855 ptr = pt;
856 break;
857 }
858
859 /* If the sequence of hex digits does not end with '}', then we don't
860 recognize this construct; fall through to the normal \x handling. */
861 }
862
863 /* Read just a single-byte hex-defined char */
864
865 c = 0;
866 while (i++ < 2 && (digitab[ptr[1]] & ctype_xdigit) != 0)
867 {
868 int cc; /* Some compilers don't like */
869 cc = *(++ptr); /* ++ in initializers */
870 #ifndef EBCDIC /* ASCII/UTF-8 coding */
871 if (cc >= CHAR_a) cc -= 32; /* Convert to upper case */
872 c = c * 16 + cc - ((cc < CHAR_A)? CHAR_0 : (CHAR_A - 10));
873 #else /* EBCDIC coding */
874 if (cc <= CHAR_z) cc += 64; /* Convert to upper case */
875 c = c * 16 + cc - ((cc >= CHAR_0)? CHAR_0 : (CHAR_A - 10));
876 #endif
877 }
878 break;
879
880 /* For \c, a following letter is upper-cased; then the 0x40 bit is flipped.
881 An error is given if the byte following \c is not an ASCII character. This
882 coding is ASCII-specific, but then the whole concept of \cx is
883 ASCII-specific. (However, an EBCDIC equivalent has now been added.) */
884
885 case CHAR_c:
886 c = *(++ptr);
887 if (c == 0)
888 {
889 *errorcodeptr = ERR2;
890 break;
891 }
892 #ifndef EBCDIC /* ASCII/UTF-8 coding */
893 if (c > 127) /* Excludes all non-ASCII in either mode */
894 {
895 *errorcodeptr = ERR68;
896 break;
897 }
898 if (c >= CHAR_a && c <= CHAR_z) c -= 32;
899 c ^= 0x40;
900 #else /* EBCDIC coding */
901 if (c >= CHAR_a && c <= CHAR_z) c += 64;
902 c ^= 0xC0;
903 #endif
904 break;
905
906 /* PCRE_EXTRA enables extensions to Perl in the matter of escapes. Any
907 other alphanumeric following \ is an error if PCRE_EXTRA was set;
908 otherwise, for Perl compatibility, it is a literal. This code looks a bit
909 odd, but there used to be some cases other than the default, and there may
910 be again in future, so I haven't "optimized" it. */
911
912 default:
913 if ((options & PCRE_EXTRA) != 0) switch(c)
914 {
915 default:
916 *errorcodeptr = ERR3;
917 break;
918 }
919 break;
920 }
921 }
922
923 /* Perl supports \N{name} for character names, as well as plain \N for "not
924 newline". PCRE does not support \N{name}. However, it does support
925 quantification such as \N{2,3}. */
926
927 if (c == -ESC_N && ptr[1] == CHAR_LEFT_CURLY_BRACKET &&
928 !is_counted_repeat(ptr+2))
929 *errorcodeptr = ERR37;
930
931 /* If PCRE_UCP is set, we change the values for \d etc. */
932
933 if ((options & PCRE_UCP) != 0 && c <= -ESC_D && c >= -ESC_w)
934 c -= (ESC_DU - ESC_D);
935
936 /* Set the pointer to the final character before returning. */
937
938 *ptrptr = ptr;
939 return c;
940 }
941
942
943
944 #ifdef SUPPORT_UCP
945 /*************************************************
946 * Handle \P and \p *
947 *************************************************/
948
949 /* This function is called after \P or \p has been encountered, provided that
950 PCRE is compiled with support for Unicode properties. On entry, ptrptr is
951 pointing at the P or p. On exit, it is pointing at the final character of the
952 escape sequence.
953
954 Argument:
955 ptrptr points to the pattern position pointer
956 negptr points to a boolean that is set TRUE for negation else FALSE
957 dptr points to an int that is set to the detailed property value
958 errorcodeptr points to the error code variable
959
960 Returns: type value from ucp_type_table, or -1 for an invalid type
961 */
962
963 static int
964 get_ucp(const uschar **ptrptr, BOOL *negptr, int *dptr, int *errorcodeptr)
965 {
966 int c, i, bot, top;
967 const uschar *ptr = *ptrptr;
968 char name[32];
969
970 c = *(++ptr);
971 if (c == 0) goto ERROR_RETURN;
972
973 *negptr = FALSE;
974
975 /* \P or \p can be followed by a name in {}, optionally preceded by ^ for
976 negation. */
977
978 if (c == CHAR_LEFT_CURLY_BRACKET)
979 {
980 if (ptr[1] == CHAR_CIRCUMFLEX_ACCENT)
981 {
982 *negptr = TRUE;
983 ptr++;
984 }
985 for (i = 0; i < (int)sizeof(name) - 1; i++)
986 {
987 c = *(++ptr);
988 if (c == 0) goto ERROR_RETURN;
989 if (c == CHAR_RIGHT_CURLY_BRACKET) break;
990 name[i] = c;
991 }
992 if (c != CHAR_RIGHT_CURLY_BRACKET) goto ERROR_RETURN;
993 name[i] = 0;
994 }
995
996 /* Otherwise there is just one following character */
997
998 else
999 {
1000 name[0] = c;
1001 name[1] = 0;
1002 }
1003
1004 *ptrptr = ptr;
1005
1006 /* Search for a recognized property name using binary chop */
1007
1008 bot = 0;
1009 top = _pcre_utt_size;
1010
1011 while (bot < top)
1012 {
1013 i = (bot + top) >> 1;
1014 c = strcmp(name, _pcre_utt_names + _pcre_utt[i].name_offset);
1015 if (c == 0)
1016 {
1017 *dptr = _pcre_utt[i].value;
1018 return _pcre_utt[i].type;
1019 }
1020 if (c > 0) bot = i + 1; else top = i;
1021 }
1022
1023 *errorcodeptr = ERR47;
1024 *ptrptr = ptr;
1025 return -1;
1026
1027 ERROR_RETURN:
1028 *errorcodeptr = ERR46;
1029 *ptrptr = ptr;
1030 return -1;
1031 }
1032 #endif
1033
1034
1035
1036
1037 /*************************************************
1038 * Read repeat counts *
1039 *************************************************/
1040
1041 /* Read an item of the form {n,m} and return the values. This is called only
1042 after is_counted_repeat() has confirmed that a repeat-count quantifier exists,
1043 so the syntax is guaranteed to be correct, but we need to check the values.
1044
1045 Arguments:
1046 p pointer to first char after '{'
1047 minp pointer to int for min
1048 maxp pointer to int for max
1049 returned as -1 if no max
1050 errorcodeptr points to error code variable
1051
1052 Returns: pointer to '}' on success;
1053 current ptr on error, with errorcodeptr set non-zero
1054 */
1055
1056 static const uschar *
1057 read_repeat_counts(const uschar *p, int *minp, int *maxp, int *errorcodeptr)
1058 {
1059 int min = 0;
1060 int max = -1;
1061
1062 /* Read the minimum value and do a paranoid check: a negative value indicates
1063 an integer overflow. */
1064
1065 while ((digitab[*p] & ctype_digit) != 0) min = min * 10 + *p++ - CHAR_0;
1066 if (min < 0 || min > 65535)
1067 {
1068 *errorcodeptr = ERR5;
1069 return p;
1070 }
1071
1072 /* Read the maximum value if there is one, and again do a paranoid on its size.
1073 Also, max must not be less than min. */
1074
1075 if (*p == CHAR_RIGHT_CURLY_BRACKET) max = min; else
1076 {
1077 if (*(++p) != CHAR_RIGHT_CURLY_BRACKET)
1078 {
1079 max = 0;
1080 while((digitab[*p] & ctype_digit) != 0) max = max * 10 + *p++ - CHAR_0;
1081 if (max < 0 || max > 65535)
1082 {
1083 *errorcodeptr = ERR5;
1084 return p;
1085 }
1086 if (max < min)
1087 {
1088 *errorcodeptr = ERR4;
1089 return p;
1090 }
1091 }
1092 }
1093
1094 /* Fill in the required variables, and pass back the pointer to the terminating
1095 '}'. */
1096
1097 *minp = min;
1098 *maxp = max;
1099 return p;
1100 }
1101
1102
1103
1104 /*************************************************
1105 * Subroutine for finding forward reference *
1106 *************************************************/
1107
1108 /* This recursive function is called only from find_parens() below. The
1109 top-level call starts at the beginning of the pattern. All other calls must
1110 start at a parenthesis. It scans along a pattern's text looking for capturing
1111 subpatterns, and counting them. If it finds a named pattern that matches the
1112 name it is given, it returns its number. Alternatively, if the name is NULL, it
1113 returns when it reaches a given numbered subpattern. Recursion is used to keep
1114 track of subpatterns that reset the capturing group numbers - the (?| feature.
1115
1116 This function was originally called only from the second pass, in which we know
1117 that if (?< or (?' or (?P< is encountered, the name will be correctly
1118 terminated because that is checked in the first pass. There is now one call to
1119 this function in the first pass, to check for a recursive back reference by
1120 name (so that we can make the whole group atomic). In this case, we need check
1121 only up to the current position in the pattern, and that is still OK because
1122 and previous occurrences will have been checked. To make this work, the test
1123 for "end of pattern" is a check against cd->end_pattern in the main loop,
1124 instead of looking for a binary zero. This means that the special first-pass
1125 call can adjust cd->end_pattern temporarily. (Checks for binary zero while
1126 processing items within the loop are OK, because afterwards the main loop will
1127 terminate.)
1128
1129 Arguments:
1130 ptrptr address of the current character pointer (updated)
1131 cd compile background data
1132 name name to seek, or NULL if seeking a numbered subpattern
1133 lorn name length, or subpattern number if name is NULL
1134 xmode TRUE if we are in /x mode
1135 utf8 TRUE if we are in UTF-8 mode
1136 count pointer to the current capturing subpattern number (updated)
1137
1138 Returns: the number of the named subpattern, or -1 if not found
1139 */
1140
1141 static int
1142 find_parens_sub(uschar **ptrptr, compile_data *cd, const uschar *name, int lorn,
1143 BOOL xmode, BOOL utf8, int *count)
1144 {
1145 uschar *ptr = *ptrptr;
1146 int start_count = *count;
1147 int hwm_count = start_count;
1148 BOOL dup_parens = FALSE;
1149
1150 /* If the first character is a parenthesis, check on the type of group we are
1151 dealing with. The very first call may not start with a parenthesis. */
1152
1153 if (ptr[0] == CHAR_LEFT_PARENTHESIS)
1154 {
1155 /* Handle specials such as (*SKIP) or (*UTF8) etc. */
1156
1157 if (ptr[1] == CHAR_ASTERISK) ptr += 2;
1158
1159 /* Handle a normal, unnamed capturing parenthesis. */
1160
1161 else if (ptr[1] != CHAR_QUESTION_MARK)
1162 {
1163 *count += 1;
1164 if (name == NULL && *count == lorn) return *count;
1165 ptr++;
1166 }
1167
1168 /* All cases now have (? at the start. Remember when we are in a group
1169 where the parenthesis numbers are duplicated. */
1170
1171 else if (ptr[2] == CHAR_VERTICAL_LINE)
1172 {
1173 ptr += 3;
1174 dup_parens = TRUE;
1175 }
1176
1177 /* Handle comments; all characters are allowed until a ket is reached. */
1178
1179 else if (ptr[2] == CHAR_NUMBER_SIGN)
1180 {
1181 for (ptr += 3; *ptr != 0; ptr++) if (*ptr == CHAR_RIGHT_PARENTHESIS) break;
1182 goto FAIL_EXIT;
1183 }
1184
1185 /* Handle a condition. If it is an assertion, just carry on so that it
1186 is processed as normal. If not, skip to the closing parenthesis of the
1187 condition (there can't be any nested parens). */
1188
1189 else if (ptr[2] == CHAR_LEFT_PARENTHESIS)
1190 {
1191 ptr += 2;
1192 if (ptr[1] != CHAR_QUESTION_MARK)
1193 {
1194 while (*ptr != 0 && *ptr != CHAR_RIGHT_PARENTHESIS) ptr++;
1195 if (*ptr != 0) ptr++;
1196 }
1197 }
1198
1199 /* Start with (? but not a condition. */
1200
1201 else
1202 {
1203 ptr += 2;
1204 if (*ptr == CHAR_P) ptr++; /* Allow optional P */
1205
1206 /* We have to disambiguate (?<! and (?<= from (?<name> for named groups */
1207
1208 if ((*ptr == CHAR_LESS_THAN_SIGN && ptr[1] != CHAR_EXCLAMATION_MARK &&
1209 ptr[1] != CHAR_EQUALS_SIGN) || *ptr == CHAR_APOSTROPHE)
1210 {
1211 int term;
1212 const uschar *thisname;
1213 *count += 1;
1214 if (name == NULL && *count == lorn) return *count;
1215 term = *ptr++;
1216 if (term == CHAR_LESS_THAN_SIGN) term = CHAR_GREATER_THAN_SIGN;
1217 thisname = ptr;
1218 while (*ptr != term) ptr++;
1219 if (name != NULL && lorn == ptr - thisname &&
1220 strncmp((const char *)name, (const char *)thisname, lorn) == 0)
1221 return *count;
1222 term++;
1223 }
1224 }
1225 }
1226
1227 /* Past any initial parenthesis handling, scan for parentheses or vertical
1228 bars. Stop if we get to cd->end_pattern. Note that this is important for the
1229 first-pass call when this value is temporarily adjusted to stop at the current
1230 position. So DO NOT change this to a test for binary zero. */
1231
1232 for (; ptr < cd->end_pattern; ptr++)
1233 {
1234 /* Skip over backslashed characters and also entire \Q...\E */
1235
1236 if (*ptr == CHAR_BACKSLASH)
1237 {
1238 if (*(++ptr) == 0) goto FAIL_EXIT;
1239 if (*ptr == CHAR_Q) for (;;)
1240 {
1241 while (*(++ptr) != 0 && *ptr != CHAR_BACKSLASH) {};
1242 if (*ptr == 0) goto FAIL_EXIT;
1243 if (*(++ptr) == CHAR_E) break;
1244 }
1245 continue;
1246 }
1247
1248 /* Skip over character classes; this logic must be similar to the way they
1249 are handled for real. If the first character is '^', skip it. Also, if the
1250 first few characters (either before or after ^) are \Q\E or \E we skip them
1251 too. This makes for compatibility with Perl. Note the use of STR macros to
1252 encode "Q\\E" so that it works in UTF-8 on EBCDIC platforms. */
1253
1254 if (*ptr == CHAR_LEFT_SQUARE_BRACKET)
1255 {
1256 BOOL negate_class = FALSE;
1257 for (;;)
1258 {
1259 if (ptr[1] == CHAR_BACKSLASH)
1260 {
1261 if (ptr[2] == CHAR_E)
1262 ptr+= 2;
1263 else if (strncmp((const char *)ptr+2,
1264 STR_Q STR_BACKSLASH STR_E, 3) == 0)
1265 ptr += 4;
1266 else
1267 break;
1268 }
1269 else if (!negate_class && ptr[1] == CHAR_CIRCUMFLEX_ACCENT)
1270 {
1271 negate_class = TRUE;
1272 ptr++;
1273 }
1274 else break;
1275 }
1276
1277 /* If the next character is ']', it is a data character that must be
1278 skipped, except in JavaScript compatibility mode. */
1279
1280 if (ptr[1] == CHAR_RIGHT_SQUARE_BRACKET &&
1281 (cd->external_options & PCRE_JAVASCRIPT_COMPAT) == 0)
1282 ptr++;
1283
1284 while (*(++ptr) != CHAR_RIGHT_SQUARE_BRACKET)
1285 {
1286 if (*ptr == 0) return -1;
1287 if (*ptr == CHAR_BACKSLASH)
1288 {
1289 if (*(++ptr) == 0) goto FAIL_EXIT;
1290 if (*ptr == CHAR_Q) for (;;)
1291 {
1292 while (*(++ptr) != 0 && *ptr != CHAR_BACKSLASH) {};
1293 if (*ptr == 0) goto FAIL_EXIT;
1294 if (*(++ptr) == CHAR_E) break;
1295 }
1296 continue;
1297 }
1298 }
1299 continue;
1300 }
1301
1302 /* Skip comments in /x mode */
1303
1304 if (xmode && *ptr == CHAR_NUMBER_SIGN)
1305 {
1306 ptr++;
1307 while (*ptr != 0)
1308 {
1309 if (IS_NEWLINE(ptr)) { ptr += cd->nllen - 1; break; }
1310 ptr++;
1311 #ifdef SUPPORT_UTF8
1312 if (utf8) while ((*ptr & 0xc0) == 0x80) ptr++;
1313 #endif
1314 }
1315 if (*ptr == 0) goto FAIL_EXIT;
1316 continue;
1317 }
1318
1319 /* Check for the special metacharacters */
1320
1321 if (*ptr == CHAR_LEFT_PARENTHESIS)
1322 {
1323 int rc = find_parens_sub(&ptr, cd, name, lorn, xmode, utf8, count);
1324 if (rc > 0) return rc;
1325 if (*ptr == 0) goto FAIL_EXIT;
1326 }
1327
1328 else if (*ptr == CHAR_RIGHT_PARENTHESIS)
1329 {
1330 if (dup_parens && *count < hwm_count) *count = hwm_count;
1331 goto FAIL_EXIT;
1332 }
1333
1334 else if (*ptr == CHAR_VERTICAL_LINE && dup_parens)
1335 {
1336 if (*count > hwm_count) hwm_count = *count;
1337 *count = start_count;
1338 }
1339 }
1340
1341 FAIL_EXIT:
1342 *ptrptr = ptr;
1343 return -1;
1344 }
1345
1346
1347
1348
1349 /*************************************************
1350 * Find forward referenced subpattern *
1351 *************************************************/
1352
1353 /* This function scans along a pattern's text looking for capturing
1354 subpatterns, and counting them. If it finds a named pattern that matches the
1355 name it is given, it returns its number. Alternatively, if the name is NULL, it
1356 returns when it reaches a given numbered subpattern. This is used for forward
1357 references to subpatterns. We used to be able to start this scan from the
1358 current compiling point, using the current count value from cd->bracount, and
1359 do it all in a single loop, but the addition of the possibility of duplicate
1360 subpattern numbers means that we have to scan from the very start, in order to
1361 take account of such duplicates, and to use a recursive function to keep track
1362 of the different types of group.
1363
1364 Arguments:
1365 cd compile background data
1366 name name to seek, or NULL if seeking a numbered subpattern
1367 lorn name length, or subpattern number if name is NULL
1368 xmode TRUE if we are in /x mode
1369 utf8 TRUE if we are in UTF-8 mode
1370
1371 Returns: the number of the found subpattern, or -1 if not found
1372 */
1373
1374 static int
1375 find_parens(compile_data *cd, const uschar *name, int lorn, BOOL xmode,
1376 BOOL utf8)
1377 {
1378 uschar *ptr = (uschar *)cd->start_pattern;
1379 int count = 0;
1380 int rc;
1381
1382 /* If the pattern does not start with an opening parenthesis, the first call
1383 to find_parens_sub() will scan right to the end (if necessary). However, if it
1384 does start with a parenthesis, find_parens_sub() will return when it hits the
1385 matching closing parens. That is why we have to have a loop. */
1386
1387 for (;;)
1388 {
1389 rc = find_parens_sub(&ptr, cd, name, lorn, xmode, utf8, &count);
1390 if (rc > 0 || *ptr++ == 0) break;
1391 }
1392
1393 return rc;
1394 }
1395
1396
1397
1398
1399 /*************************************************
1400 * Find first significant op code *
1401 *************************************************/
1402
1403 /* This is called by several functions that scan a compiled expression looking
1404 for a fixed first character, or an anchoring op code etc. It skips over things
1405 that do not influence this. For some calls, it makes sense to skip negative
1406 forward and all backward assertions, and also the \b assertion; for others it
1407 does not.
1408
1409 Arguments:
1410 code pointer to the start of the group
1411 skipassert TRUE if certain assertions are to be skipped
1412
1413 Returns: pointer to the first significant opcode
1414 */
1415
1416 static const uschar*
1417 first_significant_code(const uschar *code, BOOL skipassert)
1418 {
1419 for (;;)
1420 {
1421 switch ((int)*code)
1422 {
1423 case OP_ASSERT_NOT:
1424 case OP_ASSERTBACK:
1425 case OP_ASSERTBACK_NOT:
1426 if (!skipassert) return code;
1427 do code += GET(code, 1); while (*code == OP_ALT);
1428 code += _pcre_OP_lengths[*code];
1429 break;
1430
1431 case OP_WORD_BOUNDARY:
1432 case OP_NOT_WORD_BOUNDARY:
1433 if (!skipassert) return code;
1434 /* Fall through */
1435
1436 case OP_CALLOUT:
1437 case OP_CREF:
1438 case OP_NCREF:
1439 case OP_RREF:
1440 case OP_NRREF:
1441 case OP_DEF:
1442 code += _pcre_OP_lengths[*code];
1443 break;
1444
1445 default:
1446 return code;
1447 }
1448 }
1449 /* Control never reaches here */
1450 }
1451
1452
1453
1454
1455 /*************************************************
1456 * Find the fixed length of a branch *
1457 *************************************************/
1458
1459 /* Scan a branch and compute the fixed length of subject that will match it,
1460 if the length is fixed. This is needed for dealing with backward assertions.
1461 In UTF8 mode, the result is in characters rather than bytes. The branch is
1462 temporarily terminated with OP_END when this function is called.
1463
1464 This function is called when a backward assertion is encountered, so that if it
1465 fails, the error message can point to the correct place in the pattern.
1466 However, we cannot do this when the assertion contains subroutine calls,
1467 because they can be forward references. We solve this by remembering this case
1468 and doing the check at the end; a flag specifies which mode we are running in.
1469
1470 Arguments:
1471 code points to the start of the pattern (the bracket)
1472 utf8 TRUE in UTF-8 mode
1473 atend TRUE if called when the pattern is complete
1474 cd the "compile data" structure
1475
1476 Returns: the fixed length,
1477 or -1 if there is no fixed length,
1478 or -2 if \C was encountered
1479 or -3 if an OP_RECURSE item was encountered and atend is FALSE
1480 */
1481
1482 static int
1483 find_fixedlength(uschar *code, BOOL utf8, BOOL atend, compile_data *cd)
1484 {
1485 int length = -1;
1486
1487 register int branchlength = 0;
1488 register uschar *cc = code + 1 + LINK_SIZE;
1489
1490 /* Scan along the opcodes for this branch. If we get to the end of the
1491 branch, check the length against that of the other branches. */
1492
1493 for (;;)
1494 {
1495 int d;
1496 uschar *ce, *cs;
1497 register int op = *cc;
1498 switch (op)
1499 {
1500 /* We only need to continue for OP_CBRA (normal capturing bracket) and
1501 OP_BRA (normal non-capturing bracket) because the other variants of these
1502 opcodes are all concerned with unlimited repeated groups, which of course
1503 are not of fixed length. They will cause a -1 response from the default
1504 case of this switch. */
1505
1506 case OP_CBRA:
1507 case OP_BRA:
1508 case OP_ONCE:
1509 case OP_COND:
1510 d = find_fixedlength(cc + ((op == OP_CBRA)? 2:0), utf8, atend, cd);
1511 if (d < 0) return d;
1512 branchlength += d;
1513 do cc += GET(cc, 1); while (*cc == OP_ALT);
1514 cc += 1 + LINK_SIZE;
1515 break;
1516
1517 /* Reached end of a branch; if it's a ket it is the end of a nested
1518 call. If it's ALT it is an alternation in a nested call. If it is
1519 END it's the end of the outer call. All can be handled by the same code.
1520 Note that we must not include the OP_KETRxxx opcodes here, because they
1521 all imply an unlimited repeat. */
1522
1523 case OP_ALT:
1524 case OP_KET:
1525 case OP_END:
1526 if (length < 0) length = branchlength;
1527 else if (length != branchlength) return -1;
1528 if (*cc != OP_ALT) return length;
1529 cc += 1 + LINK_SIZE;
1530 branchlength = 0;
1531 break;
1532
1533 /* A true recursion implies not fixed length, but a subroutine call may
1534 be OK. If the subroutine is a forward reference, we can't deal with
1535 it until the end of the pattern, so return -3. */
1536
1537 case OP_RECURSE:
1538 if (!atend) return -3;
1539 cs = ce = (uschar *)cd->start_code + GET(cc, 1); /* Start subpattern */
1540 do ce += GET(ce, 1); while (*ce == OP_ALT); /* End subpattern */
1541 if (cc > cs && cc < ce) return -1; /* Recursion */
1542 d = find_fixedlength(cs + 2, utf8, atend, cd);
1543 if (d < 0) return d;
1544 branchlength += d;
1545 cc += 1 + LINK_SIZE;
1546 break;
1547
1548 /* Skip over assertive subpatterns */
1549
1550 case OP_ASSERT:
1551 case OP_ASSERT_NOT:
1552 case OP_ASSERTBACK:
1553 case OP_ASSERTBACK_NOT:
1554 do cc += GET(cc, 1); while (*cc == OP_ALT);
1555 /* Fall through */
1556
1557 /* Skip over things that don't match chars */
1558
1559 case OP_REVERSE:
1560 case OP_CREF:
1561 case OP_NCREF:
1562 case OP_RREF:
1563 case OP_NRREF:
1564 case OP_DEF:
1565 case OP_CALLOUT:
1566 case OP_SOD:
1567 case OP_SOM:
1568 case OP_SET_SOM:
1569 case OP_EOD:
1570 case OP_EODN:
1571 case OP_CIRC:
1572 case OP_CIRCM:
1573 case OP_DOLL:
1574 case OP_DOLLM:
1575 case OP_NOT_WORD_BOUNDARY:
1576 case OP_WORD_BOUNDARY:
1577 cc += _pcre_OP_lengths[*cc];
1578 break;
1579
1580 /* Handle literal characters */
1581
1582 case OP_CHAR:
1583 case OP_CHARI:
1584 case OP_NOT:
1585 case OP_NOTI:
1586 branchlength++;
1587 cc += 2;
1588 #ifdef SUPPORT_UTF8
1589 if (utf8 && cc[-1] >= 0xc0) cc += _pcre_utf8_table4[cc[-1] & 0x3f];
1590 #endif
1591 break;
1592
1593 /* Handle exact repetitions. The count is already in characters, but we
1594 need to skip over a multibyte character in UTF8 mode. */
1595
1596 case OP_EXACT:
1597 branchlength += GET2(cc,1);
1598 cc += 4;
1599 #ifdef SUPPORT_UTF8
1600 if (utf8 && cc[-1] >= 0xc0) cc += _pcre_utf8_table4[cc[-1] & 0x3f];
1601 #endif
1602 break;
1603
1604 case OP_TYPEEXACT:
1605 branchlength += GET2(cc,1);
1606 if (cc[3] == OP_PROP || cc[3] == OP_NOTPROP) cc += 2;
1607 cc += 4;
1608 break;
1609
1610 /* Handle single-char matchers */
1611
1612 case OP_PROP:
1613 case OP_NOTPROP:
1614 cc += 2;
1615 /* Fall through */
1616
1617 case OP_NOT_DIGIT:
1618 case OP_DIGIT:
1619 case OP_NOT_WHITESPACE:
1620 case OP_WHITESPACE:
1621 case OP_NOT_WORDCHAR:
1622 case OP_WORDCHAR:
1623 case OP_ANY:
1624 case OP_ALLANY:
1625 branchlength++;
1626 cc++;
1627 break;
1628
1629 /* The single-byte matcher isn't allowed */
1630
1631 case OP_ANYBYTE:
1632 return -2;
1633
1634 /* Check a class for variable quantification */
1635
1636 #ifdef SUPPORT_UTF8
1637 case OP_XCLASS:
1638 cc += GET(cc, 1) - 33;
1639 /* Fall through */
1640 #endif
1641
1642 case OP_CLASS:
1643 case OP_NCLASS:
1644 cc += 33;
1645
1646 switch (*cc)
1647 {
1648 case OP_CRSTAR:
1649 case OP_CRMINSTAR:
1650 case OP_CRQUERY:
1651 case OP_CRMINQUERY:
1652 return -1;
1653
1654 case OP_CRRANGE:
1655 case OP_CRMINRANGE:
1656 if (GET2(cc,1) != GET2(cc,3)) return -1;
1657 branchlength += GET2(cc,1);
1658 cc += 5;
1659 break;
1660
1661 default:
1662 branchlength++;
1663 }
1664 break;
1665
1666 /* Anything else is variable length */
1667
1668 default:
1669 return -1;
1670 }
1671 }
1672 /* Control never gets here */
1673 }
1674
1675
1676
1677
1678 /*************************************************
1679 * Scan compiled regex for specific bracket *
1680 *************************************************/
1681
1682 /* This little function scans through a compiled pattern until it finds a
1683 capturing bracket with the given number, or, if the number is negative, an
1684 instance of OP_REVERSE for a lookbehind. The function is global in the C sense
1685 so that it can be called from pcre_study() when finding the minimum matching
1686 length.
1687
1688 Arguments:
1689 code points to start of expression
1690 utf8 TRUE in UTF-8 mode
1691 number the required bracket number or negative to find a lookbehind
1692
1693 Returns: pointer to the opcode for the bracket, or NULL if not found
1694 */
1695
1696 const uschar *
1697 _pcre_find_bracket(const uschar *code, BOOL utf8, int number)
1698 {
1699 for (;;)
1700 {
1701 register int c = *code;
1702
1703 if (c == OP_END) return NULL;
1704
1705 /* XCLASS is used for classes that cannot be represented just by a bit
1706 map. This includes negated single high-valued characters. The length in
1707 the table is zero; the actual length is stored in the compiled code. */
1708
1709 if (c == OP_XCLASS) code += GET(code, 1);
1710
1711 /* Handle recursion */
1712
1713 else if (c == OP_REVERSE)
1714 {
1715 if (number < 0) return (uschar *)code;
1716 code += _pcre_OP_lengths[c];
1717 }
1718
1719 /* Handle capturing bracket */
1720
1721 else if (c == OP_CBRA || c == OP_SCBRA ||
1722 c == OP_CBRAPOS || c == OP_SCBRAPOS)
1723 {
1724 int n = GET2(code, 1+LINK_SIZE);
1725 if (n == number) return (uschar *)code;
1726 code += _pcre_OP_lengths[c];
1727 }
1728
1729 /* Otherwise, we can get the item's length from the table, except that for
1730 repeated character types, we have to test for \p and \P, which have an extra
1731 two bytes of parameters, and for MARK/PRUNE/SKIP/THEN with an argument, we
1732 must add in its length. */
1733
1734 else
1735 {
1736 switch(c)
1737 {
1738 case OP_TYPESTAR:
1739 case OP_TYPEMINSTAR:
1740 case OP_TYPEPLUS:
1741 case OP_TYPEMINPLUS:
1742 case OP_TYPEQUERY:
1743 case OP_TYPEMINQUERY:
1744 case OP_TYPEPOSSTAR:
1745 case OP_TYPEPOSPLUS:
1746 case OP_TYPEPOSQUERY:
1747 if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
1748 break;
1749
1750 case OP_TYPEUPTO:
1751 case OP_TYPEMINUPTO:
1752 case OP_TYPEEXACT:
1753 case OP_TYPEPOSUPTO:
1754 if (code[3] == OP_PROP || code[3] == OP_NOTPROP) code += 2;
1755 break;
1756
1757 case OP_MARK:
1758 case OP_PRUNE_ARG:
1759 case OP_SKIP_ARG:
1760 code += code[1];
1761 break;
1762
1763 case OP_THEN_ARG:
1764 code += code[1+LINK_SIZE];
1765 break;
1766 }
1767
1768 /* Add in the fixed length from the table */
1769
1770 code += _pcre_OP_lengths[c];
1771
1772 /* In UTF-8 mode, opcodes that are followed by a character may be followed by
1773 a multi-byte character. The length in the table is a minimum, so we have to
1774 arrange to skip the extra bytes. */
1775
1776 #ifdef SUPPORT_UTF8
1777 if (utf8) switch(c)
1778 {
1779 case OP_CHAR:
1780 case OP_CHARI:
1781 case OP_EXACT:
1782 case OP_EXACTI:
1783 case OP_UPTO:
1784 case OP_UPTOI:
1785 case OP_MINUPTO:
1786 case OP_MINUPTOI:
1787 case OP_POSUPTO:
1788 case OP_POSUPTOI:
1789 case OP_STAR:
1790 case OP_STARI:
1791 case OP_MINSTAR:
1792 case OP_MINSTARI:
1793 case OP_POSSTAR:
1794 case OP_POSSTARI:
1795 case OP_PLUS:
1796 case OP_PLUSI:
1797 case OP_MINPLUS:
1798 case OP_MINPLUSI:
1799 case OP_POSPLUS:
1800 case OP_POSPLUSI:
1801 case OP_QUERY:
1802 case OP_QUERYI:
1803 case OP_MINQUERY:
1804 case OP_MINQUERYI:
1805 case OP_POSQUERY:
1806 case OP_POSQUERYI:
1807 if (code[-1] >= 0xc0) code += _pcre_utf8_table4[code[-1] & 0x3f];
1808 break;
1809 }
1810 #else
1811 (void)(utf8); /* Keep compiler happy by referencing function argument */
1812 #endif
1813 }
1814 }
1815 }
1816
1817
1818
1819 /*************************************************
1820 * Scan compiled regex for recursion reference *
1821 *************************************************/
1822
1823 /* This little function scans through a compiled pattern until it finds an
1824 instance of OP_RECURSE.
1825
1826 Arguments:
1827 code points to start of expression
1828 utf8 TRUE in UTF-8 mode
1829
1830 Returns: pointer to the opcode for OP_RECURSE, or NULL if not found
1831 */
1832
1833 static const uschar *
1834 find_recurse(const uschar *code, BOOL utf8)
1835 {
1836 for (;;)
1837 {
1838 register int c = *code;
1839 if (c == OP_END) return NULL;
1840 if (c == OP_RECURSE) return code;
1841
1842 /* XCLASS is used for classes that cannot be represented just by a bit
1843 map. This includes negated single high-valued characters. The length in
1844 the table is zero; the actual length is stored in the compiled code. */
1845
1846 if (c == OP_XCLASS) code += GET(code, 1);
1847
1848 /* Otherwise, we can get the item's length from the table, except that for
1849 repeated character types, we have to test for \p and \P, which have an extra
1850 two bytes of parameters, and for MARK/PRUNE/SKIP/THEN with an argument, we
1851 must add in its length. */
1852
1853 else
1854 {
1855 switch(c)
1856 {
1857 case OP_TYPESTAR:
1858 case OP_TYPEMINSTAR:
1859 case OP_TYPEPLUS:
1860 case OP_TYPEMINPLUS:
1861 case OP_TYPEQUERY:
1862 case OP_TYPEMINQUERY:
1863 case OP_TYPEPOSSTAR:
1864 case OP_TYPEPOSPLUS:
1865 case OP_TYPEPOSQUERY:
1866 if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
1867 break;
1868
1869 case OP_TYPEPOSUPTO:
1870 case OP_TYPEUPTO:
1871 case OP_TYPEMINUPTO:
1872 case OP_TYPEEXACT:
1873 if (code[3] == OP_PROP || code[3] == OP_NOTPROP) code += 2;
1874 break;
1875
1876 case OP_MARK:
1877 case OP_PRUNE_ARG:
1878 case OP_SKIP_ARG:
1879 code += code[1];
1880 break;
1881
1882 case OP_THEN_ARG:
1883 code += code[1+LINK_SIZE];
1884 break;
1885 }
1886
1887 /* Add in the fixed length from the table */
1888
1889 code += _pcre_OP_lengths[c];
1890
1891 /* In UTF-8 mode, opcodes that are followed by a character may be followed
1892 by a multi-byte character. The length in the table is a minimum, so we have
1893 to arrange to skip the extra bytes. */
1894
1895 #ifdef SUPPORT_UTF8
1896 if (utf8) switch(c)
1897 {
1898 case OP_CHAR:
1899 case OP_CHARI:
1900 case OP_EXACT:
1901 case OP_EXACTI:
1902 case OP_UPTO:
1903 case OP_UPTOI:
1904 case OP_MINUPTO:
1905 case OP_MINUPTOI:
1906 case OP_POSUPTO:
1907 case OP_POSUPTOI:
1908 case OP_STAR:
1909 case OP_STARI:
1910 case OP_MINSTAR:
1911 case OP_MINSTARI:
1912 case OP_POSSTAR:
1913 case OP_POSSTARI:
1914 case OP_PLUS:
1915 case OP_PLUSI:
1916 case OP_MINPLUS:
1917 case OP_MINPLUSI:
1918 case OP_POSPLUS:
1919 case OP_POSPLUSI:
1920 case OP_QUERY:
1921 case OP_QUERYI:
1922 case OP_MINQUERY:
1923 case OP_MINQUERYI:
1924 case OP_POSQUERY:
1925 case OP_POSQUERYI:
1926 if (code[-1] >= 0xc0) code += _pcre_utf8_table4[code[-1] & 0x3f];
1927 break;
1928 }
1929 #else
1930 (void)(utf8); /* Keep compiler happy by referencing function argument */
1931 #endif
1932 }
1933 }
1934 }
1935
1936
1937
1938 /*************************************************
1939 * Scan compiled branch for non-emptiness *
1940 *************************************************/
1941
1942 /* This function scans through a branch of a compiled pattern to see whether it
1943 can match the empty string or not. It is called from could_be_empty()
1944 below and from compile_branch() when checking for an unlimited repeat of a
1945 group that can match nothing. Note that first_significant_code() skips over
1946 backward and negative forward assertions when its final argument is TRUE. If we
1947 hit an unclosed bracket, we return "empty" - this means we've struck an inner
1948 bracket whose current branch will already have been scanned.
1949
1950 Arguments:
1951 code points to start of search
1952 endcode points to where to stop
1953 utf8 TRUE if in UTF8 mode
1954 cd contains pointers to tables etc.
1955
1956 Returns: TRUE if what is matched could be empty
1957 */
1958
1959 static BOOL
1960 could_be_empty_branch(const uschar *code, const uschar *endcode, BOOL utf8,
1961 compile_data *cd)
1962 {
1963 register int c;
1964 for (code = first_significant_code(code + _pcre_OP_lengths[*code], TRUE);
1965 code < endcode;
1966 code = first_significant_code(code + _pcre_OP_lengths[c], TRUE))
1967 {
1968 const uschar *ccode;
1969
1970 c = *code;
1971
1972 /* Skip over forward assertions; the other assertions are skipped by
1973 first_significant_code() with a TRUE final argument. */
1974
1975 if (c == OP_ASSERT)
1976 {
1977 do code += GET(code, 1); while (*code == OP_ALT);
1978 c = *code;
1979 continue;
1980 }
1981
1982 /* For a recursion/subroutine call, if its end has been reached, which
1983 implies a backward reference subroutine call, we can scan it. If it's a
1984 forward reference subroutine call, we can't. To detect forward reference
1985 we have to scan up the list that is kept in the workspace. This function is
1986 called only when doing the real compile, not during the pre-compile that
1987 measures the size of the compiled pattern. */
1988
1989 if (c == OP_RECURSE)
1990 {
1991 const uschar *scode;
1992 BOOL empty_branch;
1993
1994 /* Test for forward reference */
1995
1996 for (scode = cd->start_workspace; scode < cd->hwm; scode += LINK_SIZE)
1997 if (GET(scode, 0) == code + 1 - cd->start_code) return TRUE;
1998
1999 /* Not a forward reference, test for completed backward reference */
2000
2001 empty_branch = FALSE;
2002 scode = cd->start_code + GET(code, 1);
2003 if (GET(scode, 1) == 0) return TRUE; /* Unclosed */
2004
2005 /* Completed backwards reference */
2006
2007 do
2008 {
2009 if (could_be_empty_branch(scode, endcode, utf8, cd))
2010 {
2011 empty_branch = TRUE;
2012 break;
2013 }
2014 scode += GET(scode, 1);
2015 }
2016 while (*scode == OP_ALT);
2017
2018 if (!empty_branch) return FALSE; /* All branches are non-empty */
2019 continue;
2020 }
2021
2022 /* Groups with zero repeats can of course be empty; skip them. */
2023
2024 if (c == OP_BRAZERO || c == OP_BRAMINZERO || c == OP_SKIPZERO ||
2025 c == OP_BRAPOSZERO)
2026 {
2027 code += _pcre_OP_lengths[c];
2028 do code += GET(code, 1); while (*code == OP_ALT);
2029 c = *code;
2030 continue;
2031 }
2032
2033 /* A nested group that is already marked as "could be empty" can just be
2034 skipped. */
2035
2036 if (c == OP_SBRA || c == OP_SBRAPOS ||
2037 c == OP_SCBRA || c == OP_SCBRAPOS)
2038 {
2039 do code += GET(code, 1); while (*code == OP_ALT);
2040 c = *code;
2041 continue;
2042 }
2043
2044 /* For other groups, scan the branches. */
2045
2046 if (c == OP_BRA || c == OP_BRAPOS ||
2047 c == OP_CBRA || c == OP_CBRAPOS ||
2048 c == OP_ONCE || c == OP_COND)
2049 {
2050 BOOL empty_branch;
2051 if (GET(code, 1) == 0) return TRUE; /* Hit unclosed bracket */
2052
2053 /* If a conditional group has only one branch, there is a second, implied,
2054 empty branch, so just skip over the conditional, because it could be empty.
2055 Otherwise, scan the individual branches of the group. */
2056
2057 if (c == OP_COND && code[GET(code, 1)] != OP_ALT)
2058 code += GET(code, 1);
2059 else
2060 {
2061 empty_branch = FALSE;
2062 do
2063 {
2064 if (!empty_branch && could_be_empty_branch(code, endcode, utf8, cd))
2065 empty_branch = TRUE;
2066 code += GET(code, 1);
2067 }
2068 while (*code == OP_ALT);
2069 if (!empty_branch) return FALSE; /* All branches are non-empty */
2070 }
2071
2072 c = *code;
2073 continue;
2074 }
2075
2076 /* Handle the other opcodes */
2077
2078 switch (c)
2079 {
2080 /* Check for quantifiers after a class. XCLASS is used for classes that
2081 cannot be represented just by a bit map. This includes negated single
2082 high-valued characters. The length in _pcre_OP_lengths[] is zero; the
2083 actual length is stored in the compiled code, so we must update "code"
2084 here. */
2085
2086 #ifdef SUPPORT_UTF8
2087 case OP_XCLASS:
2088 ccode = code += GET(code, 1);
2089 goto CHECK_CLASS_REPEAT;
2090 #endif
2091
2092 case OP_CLASS:
2093 case OP_NCLASS:
2094 ccode = code + 33;
2095
2096 #ifdef SUPPORT_UTF8
2097 CHECK_CLASS_REPEAT:
2098 #endif
2099
2100 switch (*ccode)
2101 {
2102 case OP_CRSTAR: /* These could be empty; continue */
2103 case OP_CRMINSTAR:
2104 case OP_CRQUERY:
2105 case OP_CRMINQUERY:
2106 break;
2107
2108 default: /* Non-repeat => class must match */
2109 case OP_CRPLUS: /* These repeats aren't empty */
2110 case OP_CRMINPLUS:
2111 return FALSE;
2112
2113 case OP_CRRANGE:
2114 case OP_CRMINRANGE:
2115 if (GET2(ccode, 1) > 0) return FALSE; /* Minimum > 0 */
2116 break;
2117 }
2118 break;
2119
2120 /* Opcodes that must match a character */
2121
2122 case OP_PROP:
2123 case OP_NOTPROP:
2124 case OP_EXTUNI:
2125 case OP_NOT_DIGIT:
2126 case OP_DIGIT:
2127 case OP_NOT_WHITESPACE:
2128 case OP_WHITESPACE:
2129 case OP_NOT_WORDCHAR:
2130 case OP_WORDCHAR:
2131 case OP_ANY:
2132 case OP_ALLANY:
2133 case OP_ANYBYTE:
2134 case OP_CHAR:
2135 case OP_CHARI:
2136 case OP_NOT:
2137 case OP_NOTI:
2138 case OP_PLUS:
2139 case OP_MINPLUS:
2140 case OP_POSPLUS:
2141 case OP_EXACT:
2142 case OP_NOTPLUS:
2143 case OP_NOTMINPLUS:
2144 case OP_NOTPOSPLUS:
2145 case OP_NOTEXACT:
2146 case OP_TYPEPLUS:
2147 case OP_TYPEMINPLUS:
2148 case OP_TYPEPOSPLUS:
2149 case OP_TYPEEXACT:
2150 return FALSE;
2151
2152 /* These are going to continue, as they may be empty, but we have to
2153 fudge the length for the \p and \P cases. */
2154
2155 case OP_TYPESTAR:
2156 case OP_TYPEMINSTAR:
2157 case OP_TYPEPOSSTAR:
2158 case OP_TYPEQUERY:
2159 case OP_TYPEMINQUERY:
2160 case OP_TYPEPOSQUERY:
2161 if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
2162 break;
2163
2164 /* Same for these */
2165
2166 case OP_TYPEUPTO:
2167 case OP_TYPEMINUPTO:
2168 case OP_TYPEPOSUPTO:
2169 if (code[3] == OP_PROP || code[3] == OP_NOTPROP) code += 2;
2170 break;
2171
2172 /* End of branch */
2173
2174 case OP_KET:
2175 case OP_KETRMAX:
2176 case OP_KETRMIN:
2177 case OP_KETRPOS:
2178 case OP_ALT:
2179 return TRUE;
2180
2181 /* In UTF-8 mode, STAR, MINSTAR, POSSTAR, QUERY, MINQUERY, POSQUERY, UPTO,
2182 MINUPTO, and POSUPTO may be followed by a multibyte character */
2183
2184 #ifdef SUPPORT_UTF8
2185 case OP_STAR:
2186 case OP_STARI:
2187 case OP_MINSTAR:
2188 case OP_MINSTARI:
2189 case OP_POSSTAR:
2190 case OP_POSSTARI:
2191 case OP_QUERY:
2192 case OP_QUERYI:
2193 case OP_MINQUERY:
2194 case OP_MINQUERYI:
2195 case OP_POSQUERY:
2196 case OP_POSQUERYI:
2197 if (utf8 && code[1] >= 0xc0) code += _pcre_utf8_table4[code[1] & 0x3f];
2198 break;
2199
2200 case OP_UPTO:
2201 case OP_UPTOI:
2202 case OP_MINUPTO:
2203 case OP_MINUPTOI:
2204 case OP_POSUPTO:
2205 case OP_POSUPTOI:
2206 if (utf8 && code[3] >= 0xc0) code += _pcre_utf8_table4[code[3] & 0x3f];
2207 break;
2208 #endif
2209
2210 /* MARK, and PRUNE/SKIP/THEN with an argument must skip over the argument
2211 string. */
2212
2213 case OP_MARK:
2214 case OP_PRUNE_ARG:
2215 case OP_SKIP_ARG:
2216 code += code[1];
2217 break;
2218
2219 case OP_THEN_ARG:
2220 code += code[1+LINK_SIZE];
2221 break;
2222
2223 /* None of the remaining opcodes are required to match a character. */
2224
2225 default:
2226 break;
2227 }
2228 }
2229
2230 return TRUE;
2231 }
2232
2233
2234
2235 /*************************************************
2236 * Scan compiled regex for non-emptiness *
2237 *************************************************/
2238
2239 /* This function is called to check for left recursive calls. We want to check
2240 the current branch of the current pattern to see if it could match the empty
2241 string. If it could, we must look outwards for branches at other levels,
2242 stopping when we pass beyond the bracket which is the subject of the recursion.
2243 This function is called only during the real compile, not during the
2244 pre-compile.
2245
2246 Arguments:
2247 code points to start of the recursion
2248 endcode points to where to stop (current RECURSE item)
2249 bcptr points to the chain of current (unclosed) branch starts
2250 utf8 TRUE if in UTF-8 mode
2251 cd pointers to tables etc
2252
2253 Returns: TRUE if what is matched could be empty
2254 */
2255
2256 static BOOL
2257 could_be_empty(const uschar *code, const uschar *endcode, branch_chain *bcptr,
2258 BOOL utf8, compile_data *cd)
2259 {
2260 while (bcptr != NULL && bcptr->current_branch >= code)
2261 {
2262 if (!could_be_empty_branch(bcptr->current_branch, endcode, utf8, cd))
2263 return FALSE;
2264 bcptr = bcptr->outer;
2265 }
2266 return TRUE;
2267 }
2268
2269
2270
2271 /*************************************************
2272 * Check for POSIX class syntax *
2273 *************************************************/
2274
2275 /* This function is called when the sequence "[:" or "[." or "[=" is
2276 encountered in a character class. It checks whether this is followed by a
2277 sequence of characters terminated by a matching ":]" or ".]" or "=]". If we
2278 reach an unescaped ']' without the special preceding character, return FALSE.
2279
2280 Originally, this function only recognized a sequence of letters between the
2281 terminators, but it seems that Perl recognizes any sequence of characters,
2282 though of course unknown POSIX names are subsequently rejected. Perl gives an
2283 "Unknown POSIX class" error for [:f\oo:] for example, where previously PCRE
2284 didn't consider this to be a POSIX class. Likewise for [:1234:].
2285
2286 The problem in trying to be exactly like Perl is in the handling of escapes. We
2287 have to be sure that [abc[:x\]pqr] is *not* treated as containing a POSIX
2288 class, but [abc[:x\]pqr:]] is (so that an error can be generated). The code
2289 below handles the special case of \], but does not try to do any other escape
2290 processing. This makes it different from Perl for cases such as [:l\ower:]
2291 where Perl recognizes it as the POSIX class "lower" but PCRE does not recognize
2292 "l\ower". This is a lesser evil that not diagnosing bad classes when Perl does,
2293 I think.
2294
2295 A user pointed out that PCRE was rejecting [:a[:digit:]] whereas Perl was not.
2296 It seems that the appearance of a nested POSIX class supersedes an apparent
2297 external class. For example, [:a[:digit:]b:] matches "a", "b", ":", or
2298 a digit. Also, unescaped square brackets may also appear as part of class
2299 names. For example, [:a[:abc]b:] gives unknown class "[:abc]b:]"in Perl.
2300
2301 Arguments:
2302 ptr pointer to the initial [
2303 endptr where to return the end pointer
2304
2305 Returns: TRUE or FALSE
2306 */
2307
2308 static BOOL
2309 check_posix_syntax(const uschar *ptr, const uschar **endptr)
2310 {
2311 int terminator; /* Don't combine these lines; the Solaris cc */
2312 terminator = *(++ptr); /* compiler warns about "non-constant" initializer. */
2313 for (++ptr; *ptr != 0; ptr++)
2314 {
2315 if (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET)
2316 ptr++;
2317 else
2318 {
2319 if (*ptr == terminator && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET)
2320 {
2321 *endptr = ptr;
2322 return TRUE;
2323 }
2324 if (*ptr == CHAR_LEFT_SQUARE_BRACKET &&
2325 (ptr[1] == CHAR_COLON || ptr[1] == CHAR_DOT ||
2326 ptr[1] == CHAR_EQUALS_SIGN) &&
2327 check_posix_syntax(ptr, endptr))
2328 return FALSE;
2329 }
2330 }
2331 return FALSE;
2332 }
2333
2334
2335
2336
2337 /*************************************************
2338 * Check POSIX class name *
2339 *************************************************/
2340
2341 /* This function is called to check the name given in a POSIX-style class entry
2342 such as [:alnum:].
2343
2344 Arguments:
2345 ptr points to the first letter
2346 len the length of the name
2347
2348 Returns: a value representing the name, or -1 if unknown
2349 */
2350
2351 static int
2352 check_posix_name(const uschar *ptr, int len)
2353 {
2354 const char *pn = posix_names;
2355 register int yield = 0;
2356 while (posix_name_lengths[yield] != 0)
2357 {
2358 if (len == posix_name_lengths[yield] &&
2359 strncmp((const char *)ptr, pn, len) == 0) return yield;
2360 pn += posix_name_lengths[yield] + 1;
2361 yield++;
2362 }
2363 return -1;
2364 }
2365
2366
2367 /*************************************************
2368 * Adjust OP_RECURSE items in repeated group *
2369 *************************************************/
2370
2371 /* OP_RECURSE items contain an offset from the start of the regex to the group
2372 that is referenced. This means that groups can be replicated for fixed
2373 repetition simply by copying (because the recursion is allowed to refer to
2374 earlier groups that are outside the current group). However, when a group is
2375 optional (i.e. the minimum quantifier is zero), OP_BRAZERO or OP_SKIPZERO is
2376 inserted before it, after it has been compiled. This means that any OP_RECURSE
2377 items within it that refer to the group itself or any contained groups have to
2378 have their offsets adjusted. That one of the jobs of this function. Before it
2379 is called, the partially compiled regex must be temporarily terminated with
2380 OP_END.
2381
2382 This function has been extended with the possibility of forward references for
2383 recursions and subroutine calls. It must also check the list of such references
2384 for the group we are dealing with. If it finds that one of the recursions in
2385 the current group is on this list, it adjusts the offset in the list, not the
2386 value in the reference (which is a group number).
2387
2388 Arguments:
2389 group points to the start of the group
2390 adjust the amount by which the group is to be moved
2391 utf8 TRUE in UTF-8 mode
2392 cd contains pointers to tables etc.
2393 save_hwm the hwm forward reference pointer at the start of the group
2394
2395 Returns: nothing
2396 */
2397
2398 static void
2399 adjust_recurse(uschar *group, int adjust, BOOL utf8, compile_data *cd,
2400 uschar *save_hwm)
2401 {
2402 uschar *ptr = group;
2403
2404 while ((ptr = (uschar *)find_recurse(ptr, utf8)) != NULL)
2405 {
2406 int offset;
2407 uschar *hc;
2408
2409 /* See if this recursion is on the forward reference list. If so, adjust the
2410 reference. */
2411
2412 for (hc = save_hwm; hc < cd->hwm; hc += LINK_SIZE)
2413 {
2414 offset = GET(hc, 0);
2415 if (cd->start_code + offset == ptr + 1)
2416 {
2417 PUT(hc, 0, offset + adjust);
2418 break;
2419 }
2420 }
2421
2422 /* Otherwise, adjust the recursion offset if it's after the start of this
2423 group. */
2424
2425 if (hc >= cd->hwm)
2426 {
2427 offset = GET(ptr, 1);
2428 if (cd->start_code + offset >= group) PUT(ptr, 1, offset + adjust);
2429 }
2430
2431 ptr += 1 + LINK_SIZE;
2432 }
2433 }
2434
2435
2436
2437 /*************************************************
2438 * Insert an automatic callout point *
2439 *************************************************/
2440
2441 /* This function is called when the PCRE_AUTO_CALLOUT option is set, to insert
2442 callout points before each pattern item.
2443
2444 Arguments:
2445 code current code pointer
2446 ptr current pattern pointer
2447 cd pointers to tables etc
2448
2449 Returns: new code pointer
2450 */
2451
2452 static uschar *
2453 auto_callout(uschar *code, const uschar *ptr, compile_data *cd)
2454 {
2455 *code++ = OP_CALLOUT;
2456 *code++ = 255;
2457 PUT(code, 0, (int)(ptr - cd->start_pattern)); /* Pattern offset */
2458 PUT(code, LINK_SIZE, 0); /* Default length */
2459 return code + 2*LINK_SIZE;
2460 }
2461
2462
2463
2464 /*************************************************
2465 * Complete a callout item *
2466 *************************************************/
2467
2468 /* A callout item contains the length of the next item in the pattern, which
2469 we can't fill in till after we have reached the relevant point. This is used
2470 for both automatic and manual callouts.
2471
2472 Arguments:
2473 previous_callout points to previous callout item
2474 ptr current pattern pointer
2475 cd pointers to tables etc
2476
2477 Returns: nothing
2478 */
2479
2480 static void
2481 complete_callout(uschar *previous_callout, const uschar *ptr, compile_data *cd)
2482 {
2483 int length = (int)(ptr - cd->start_pattern - GET(previous_callout, 2));
2484 PUT(previous_callout, 2 + LINK_SIZE, length);
2485 }
2486
2487
2488
2489 #ifdef SUPPORT_UCP
2490 /*************************************************
2491 * Get othercase range *
2492 *************************************************/
2493
2494 /* This function is passed the start and end of a class range, in UTF-8 mode
2495 with UCP support. It searches up the characters, looking for internal ranges of
2496 characters in the "other" case. Each call returns the next one, updating the
2497 start address.
2498
2499 Arguments:
2500 cptr points to starting character value; updated
2501 d end value
2502 ocptr where to put start of othercase range
2503 odptr where to put end of othercase range
2504
2505 Yield: TRUE when range returned; FALSE when no more
2506 */
2507
2508 static BOOL
2509 get_othercase_range(unsigned int *cptr, unsigned int d, unsigned int *ocptr,
2510 unsigned int *odptr)
2511 {
2512 unsigned int c, othercase, next;
2513
2514 for (c = *cptr; c <= d; c++)
2515 { if ((othercase = UCD_OTHERCASE(c)) != c) break; }
2516
2517 if (c > d) return FALSE;
2518
2519 *ocptr = othercase;
2520 next = othercase + 1;
2521
2522 for (++c; c <= d; c++)
2523 {
2524 if (UCD_OTHERCASE(c) != next) break;
2525 next++;
2526 }
2527
2528 *odptr = next - 1;
2529 *cptr = c;
2530
2531 return TRUE;
2532 }
2533
2534
2535
2536 /*************************************************
2537 * Check a character and a property *
2538 *************************************************/
2539
2540 /* This function is called by check_auto_possessive() when a property item
2541 is adjacent to a fixed character.
2542
2543 Arguments:
2544 c the character
2545 ptype the property type
2546 pdata the data for the type
2547 negated TRUE if it's a negated property (\P or \p{^)
2548
2549 Returns: TRUE if auto-possessifying is OK
2550 */
2551
2552 static BOOL
2553 check_char_prop(int c, int ptype, int pdata, BOOL negated)
2554 {
2555 const ucd_record *prop = GET_UCD(c);
2556 switch(ptype)
2557 {
2558 case PT_LAMP:
2559 return (prop->chartype == ucp_Lu ||
2560 prop->chartype == ucp_Ll ||
2561 prop->chartype == ucp_Lt) == negated;
2562
2563 case PT_GC:
2564 return (pdata == _pcre_ucp_gentype[prop->chartype]) == negated;
2565
2566 case PT_PC:
2567 return (pdata == prop->chartype) == negated;
2568
2569 case PT_SC:
2570 return (pdata == prop->script) == negated;
2571
2572 /* These are specials */
2573
2574 case PT_ALNUM:
2575 return (_pcre_ucp_gentype[prop->chartype] == ucp_L ||
2576 _pcre_ucp_gentype[prop->chartype] == ucp_N) == negated;
2577
2578 case PT_SPACE: /* Perl space */
2579 return (_pcre_ucp_gentype[prop->chartype] == ucp_Z ||
2580 c == CHAR_HT || c == CHAR_NL || c == CHAR_FF || c == CHAR_CR)
2581 == negated;
2582
2583 case PT_PXSPACE: /* POSIX space */
2584 return (_pcre_ucp_gentype[prop->chartype] == ucp_Z ||
2585 c == CHAR_HT || c == CHAR_NL || c == CHAR_VT ||
2586 c == CHAR_FF || c == CHAR_CR)
2587 == negated;
2588
2589 case PT_WORD:
2590 return (_pcre_ucp_gentype[prop->chartype] == ucp_L ||
2591 _pcre_ucp_gentype[prop->chartype] == ucp_N ||
2592 c == CHAR_UNDERSCORE) == negated;
2593 }
2594 return FALSE;
2595 }
2596 #endif /* SUPPORT_UCP */
2597
2598
2599
2600 /*************************************************
2601 * Check if auto-possessifying is possible *
2602 *************************************************/
2603
2604 /* This function is called for unlimited repeats of certain items, to see
2605 whether the next thing could possibly match the repeated item. If not, it makes
2606 sense to automatically possessify the repeated item.
2607
2608 Arguments:
2609 previous pointer to the repeated opcode
2610 utf8 TRUE in UTF-8 mode
2611 ptr next character in pattern
2612 options options bits
2613 cd contains pointers to tables etc.
2614
2615 Returns: TRUE if possessifying is wanted
2616 */
2617
2618 static BOOL
2619 check_auto_possessive(const uschar *previous, BOOL utf8, const uschar *ptr,
2620 int options, compile_data *cd)
2621 {
2622 int c, next;
2623 int op_code = *previous++;
2624
2625 /* Skip whitespace and comments in extended mode */
2626
2627 if ((options & PCRE_EXTENDED) != 0)
2628 {
2629 for (;;)
2630 {
2631 while ((cd->ctypes[*ptr] & ctype_space) != 0) ptr++;
2632 if (*ptr == CHAR_NUMBER_SIGN)
2633 {
2634 ptr++;
2635 while (*ptr != 0)
2636 {
2637 if (IS_NEWLINE(ptr)) { ptr += cd->nllen; break; }
2638 ptr++;
2639 #ifdef SUPPORT_UTF8
2640 if (utf8) while ((*ptr & 0xc0) == 0x80) ptr++;
2641 #endif
2642 }
2643 }
2644 else break;
2645 }
2646 }
2647
2648 /* If the next item is one that we can handle, get its value. A non-negative
2649 value is a character, a negative value is an escape value. */
2650
2651 if (*ptr == CHAR_BACKSLASH)
2652 {
2653 int temperrorcode = 0;
2654 next = check_escape(&ptr, &temperrorcode, cd->bracount, options, FALSE);
2655 if (temperrorcode != 0) return FALSE;
2656 ptr++; /* Point after the escape sequence */
2657 }
2658
2659 else if ((cd->ctypes[*ptr] & ctype_meta) == 0)
2660 {
2661 #ifdef SUPPORT_UTF8
2662 if (utf8) { GETCHARINC(next, ptr); } else
2663 #endif
2664 next = *ptr++;
2665 }
2666
2667 else return FALSE;
2668
2669 /* Skip whitespace and comments in extended mode */
2670
2671 if ((options & PCRE_EXTENDED) != 0)
2672 {
2673 for (;;)
2674 {
2675 while ((cd->ctypes[*ptr] & ctype_space) != 0) ptr++;
2676 if (*ptr == CHAR_NUMBER_SIGN)
2677 {
2678 ptr++;
2679 while (*ptr != 0)
2680 {
2681 if (IS_NEWLINE(ptr)) { ptr += cd->nllen; break; }
2682 ptr++;
2683 #ifdef SUPPORT_UTF8
2684 if (utf8) while ((*ptr & 0xc0) == 0x80) ptr++;
2685 #endif
2686 }
2687 }
2688 else break;
2689 }
2690 }
2691
2692 /* If the next thing is itself optional, we have to give up. */
2693
2694 if (*ptr == CHAR_ASTERISK || *ptr == CHAR_QUESTION_MARK ||
2695 strncmp((char *)ptr, STR_LEFT_CURLY_BRACKET STR_0 STR_COMMA, 3) == 0)
2696 return FALSE;
2697
2698 /* Now compare the next item with the previous opcode. First, handle cases when
2699 the next item is a character. */
2700
2701 if (next >= 0) switch(op_code)
2702 {
2703 case OP_CHAR:
2704 #ifdef SUPPORT_UTF8
2705 GETCHARTEST(c, previous);
2706 #else
2707 c = *previous;
2708 #endif
2709 return c != next;
2710
2711 /* For CHARI (caseless character) we must check the other case. If we have
2712 Unicode property support, we can use it to test the other case of
2713 high-valued characters. */
2714
2715 case OP_CHARI:
2716 #ifdef SUPPORT_UTF8
2717 GETCHARTEST(c, previous);
2718 #else
2719 c = *previous;
2720 #endif
2721 if (c == next) return FALSE;
2722 #ifdef SUPPORT_UTF8
2723 if (utf8)
2724 {
2725 unsigned int othercase;
2726 if (next < 128) othercase = cd->fcc[next]; else
2727 #ifdef SUPPORT_UCP
2728 othercase = UCD_OTHERCASE((unsigned int)next);
2729 #else
2730 othercase = NOTACHAR;
2731 #endif
2732 return (unsigned int)c != othercase;
2733 }
2734 else
2735 #endif /* SUPPORT_UTF8 */
2736 return (c != cd->fcc[next]); /* Non-UTF-8 mode */
2737
2738 /* For OP_NOT and OP_NOTI, the data is always a single-byte character. These
2739 opcodes are not used for multi-byte characters, because they are coded using
2740 an XCLASS instead. */
2741
2742 case OP_NOT:
2743 return (c = *previous) == next;
2744
2745 case OP_NOTI:
2746 if ((c = *previous) == next) return TRUE;
2747 #ifdef SUPPORT_UTF8
2748 if (utf8)
2749 {
2750 unsigned int othercase;
2751 if (next < 128) othercase = cd->fcc[next]; else
2752 #ifdef SUPPORT_UCP
2753 othercase = UCD_OTHERCASE(next);
2754 #else
2755 othercase = NOTACHAR;
2756 #endif
2757 return (unsigned int)c == othercase;
2758 }
2759 else
2760 #endif /* SUPPORT_UTF8 */
2761 return (c == cd->fcc[next]); /* Non-UTF-8 mode */
2762
2763 /* Note that OP_DIGIT etc. are generated only when PCRE_UCP is *not* set.
2764 When it is set, \d etc. are converted into OP_(NOT_)PROP codes. */
2765
2766 case OP_DIGIT:
2767 return next > 127 || (cd->ctypes[next] & ctype_digit) == 0;
2768
2769 case OP_NOT_DIGIT:
2770 return next <= 127 && (cd->ctypes[next] & ctype_digit) != 0;
2771
2772 case OP_WHITESPACE:
2773 return next > 127 || (cd->ctypes[next] & ctype_space) == 0;
2774
2775 case OP_NOT_WHITESPACE:
2776 return next <= 127 && (cd->ctypes[next] & ctype_space) != 0;
2777
2778 case OP_WORDCHAR:
2779 return next > 127 || (cd->ctypes[next] & ctype_word) == 0;
2780
2781 case OP_NOT_WORDCHAR:
2782 return next <= 127 && (cd->ctypes[next] & ctype_word) != 0;
2783
2784 case OP_HSPACE:
2785 case OP_NOT_HSPACE:
2786 switch(next)
2787 {
2788 case 0x09:
2789 case 0x20:
2790 case 0xa0:
2791 case 0x1680:
2792 case 0x180e:
2793 case 0x2000:
2794 case 0x2001:
2795 case 0x2002:
2796 case 0x2003:
2797 case 0x2004:
2798 case 0x2005:
2799 case 0x2006:
2800 case 0x2007:
2801 case 0x2008:
2802 case 0x2009:
2803 case 0x200A:
2804 case 0x202f:
2805 case 0x205f:
2806 case 0x3000:
2807 return op_code == OP_NOT_HSPACE;
2808 default:
2809 return op_code != OP_NOT_HSPACE;
2810 }
2811
2812 case OP_ANYNL:
2813 case OP_VSPACE:
2814 case OP_NOT_VSPACE:
2815 switch(next)
2816 {
2817 case 0x0a:
2818 case 0x0b:
2819 case 0x0c:
2820 case 0x0d:
2821 case 0x85:
2822 case 0x2028:
2823 case 0x2029:
2824 return op_code == OP_NOT_VSPACE;
2825 default:
2826 return op_code != OP_NOT_VSPACE;
2827 }
2828
2829 #ifdef SUPPORT_UCP
2830 case OP_PROP:
2831 return check_char_prop(next, previous[0], previous[1], FALSE);
2832
2833 case OP_NOTPROP:
2834 return check_char_prop(next, previous[0], previous[1], TRUE);
2835 #endif
2836
2837 default:
2838 return FALSE;
2839 }
2840
2841
2842 /* Handle the case when the next item is \d, \s, etc. Note that when PCRE_UCP
2843 is set, \d turns into ESC_du rather than ESC_d, etc., so ESC_d etc. are
2844 generated only when PCRE_UCP is *not* set, that is, when only ASCII
2845 characteristics are recognized. Similarly, the opcodes OP_DIGIT etc. are
2846 replaced by OP_PROP codes when PCRE_UCP is set. */
2847
2848 switch(op_code)
2849 {
2850 case OP_CHAR:
2851 case OP_CHARI:
2852 #ifdef SUPPORT_UTF8
2853 GETCHARTEST(c, previous);
2854 #else
2855 c = *previous;
2856 #endif
2857 switch(-next)
2858 {
2859 case ESC_d:
2860 return c > 127 || (cd->ctypes[c] & ctype_digit) == 0;
2861
2862 case ESC_D:
2863 return c <= 127 && (cd->ctypes[c] & ctype_digit) != 0;
2864
2865 case ESC_s:
2866 return c > 127 || (cd->ctypes[c] & ctype_space) == 0;
2867
2868 case ESC_S:
2869 return c <= 127 && (cd->ctypes[c] & ctype_space) != 0;
2870
2871 case ESC_w:
2872 return c > 127 || (cd->ctypes[c] & ctype_word) == 0;
2873
2874 case ESC_W:
2875 return c <= 127 && (cd->ctypes[c] & ctype_word) != 0;
2876
2877 case ESC_h:
2878 case ESC_H:
2879 switch(c)
2880 {
2881 case 0x09:
2882 case 0x20:
2883 case 0xa0:
2884 case 0x1680:
2885 case 0x180e:
2886 case 0x2000:
2887 case 0x2001:
2888 case 0x2002:
2889 case 0x2003:
2890 case 0x2004:
2891 case 0x2005:
2892 case 0x2006:
2893 case 0x2007:
2894 case 0x2008:
2895 case 0x2009:
2896 case 0x200A:
2897 case 0x202f:
2898 case 0x205f:
2899 case 0x3000:
2900 return -next != ESC_h;
2901 default:
2902 return -next == ESC_h;
2903 }
2904
2905 case ESC_v:
2906 case ESC_V:
2907 switch(c)
2908 {
2909 case 0x0a:
2910 case 0x0b:
2911 case 0x0c:
2912 case 0x0d:
2913 case 0x85:
2914 case 0x2028:
2915 case 0x2029:
2916 return -next != ESC_v;
2917 default:
2918 return -next == ESC_v;
2919 }
2920
2921 /* When PCRE_UCP is set, these values get generated for \d etc. Find
2922 their substitutions and process them. The result will always be either
2923 -ESC_p or -ESC_P. Then fall through to process those values. */
2924
2925 #ifdef SUPPORT_UCP
2926 case ESC_du:
2927 case ESC_DU:
2928 case ESC_wu:
2929 case ESC_WU:
2930 case ESC_su:
2931 case ESC_SU:
2932 {
2933 int temperrorcode = 0;
2934 ptr = substitutes[-next - ESC_DU];
2935 next = check_escape(&ptr, &temperrorcode, 0, options, FALSE);
2936 if (temperrorcode != 0) return FALSE;
2937 ptr++; /* For compatibility */
2938 }
2939 /* Fall through */
2940
2941 case ESC_p:
2942 case ESC_P:
2943 {
2944 int ptype, pdata, errorcodeptr;
2945 BOOL negated;
2946
2947 ptr--; /* Make ptr point at the p or P */
2948 ptype = get_ucp(&ptr, &negated, &pdata, &errorcodeptr);
2949 if (ptype < 0) return FALSE;
2950 ptr++; /* Point past the final curly ket */
2951
2952 /* If the property item is optional, we have to give up. (When generated
2953 from \d etc by PCRE_UCP, this test will have been applied much earlier,
2954 to the original \d etc. At this point, ptr will point to a zero byte. */
2955
2956 if (*ptr == CHAR_ASTERISK || *ptr == CHAR_QUESTION_MARK ||
2957 strncmp((char *)ptr, STR_LEFT_CURLY_BRACKET STR_0 STR_COMMA, 3) == 0)
2958 return FALSE;
2959
2960 /* Do the property check. */
2961
2962 return check_char_prop(c, ptype, pdata, (next == -ESC_P) != negated);
2963 }
2964 #endif
2965
2966 default:
2967 return FALSE;
2968 }
2969
2970 /* In principle, support for Unicode properties should be integrated here as
2971 well. It means re-organizing the above code so as to get hold of the property
2972 values before switching on the op-code. However, I wonder how many patterns
2973 combine ASCII \d etc with Unicode properties? (Note that if PCRE_UCP is set,
2974 these op-codes are never generated.) */
2975
2976 case OP_DIGIT:
2977 return next == -ESC_D || next == -ESC_s || next == -ESC_W ||
2978 next == -ESC_h || next == -ESC_v || next == -ESC_R;
2979
2980 case OP_NOT_DIGIT:
2981 return next == -ESC_d;
2982
2983 case OP_WHITESPACE:
2984 return next == -ESC_S || next == -ESC_d || next == -ESC_w || next == -ESC_R;
2985
2986 case OP_NOT_WHITESPACE:
2987 return next == -ESC_s || next == -ESC_h || next == -ESC_v;
2988
2989 case OP_HSPACE:
2990 return next == -ESC_S || next == -ESC_H || next == -ESC_d ||
2991 next == -ESC_w || next == -ESC_v || next == -ESC_R;
2992
2993 case OP_NOT_HSPACE:
2994 return next == -ESC_h;
2995
2996 /* Can't have \S in here because VT matches \S (Perl anomaly) */
2997 case OP_ANYNL:
2998 case OP_VSPACE:
2999 return next == -ESC_V || next == -ESC_d || next == -ESC_w;
3000
3001 case OP_NOT_VSPACE:
3002 return next == -ESC_v || next == -ESC_R;
3003
3004 case OP_WORDCHAR:
3005 return next == -ESC_W || next == -ESC_s || next == -ESC_h ||
3006 next == -ESC_v || next == -ESC_R;
3007
3008 case OP_NOT_WORDCHAR:
3009 return next == -ESC_w || next == -ESC_d;
3010
3011 default:
3012 return FALSE;
3013 }
3014
3015 /* Control does not reach here */
3016 }
3017
3018
3019
3020 /*************************************************
3021 * Compile one branch *
3022 *************************************************/
3023
3024 /* Scan the pattern, compiling it into the a vector. If the options are
3025 changed during the branch, the pointer is used to change the external options
3026 bits. This function is used during the pre-compile phase when we are trying
3027 to find out the amount of memory needed, as well as during the real compile
3028 phase. The value of lengthptr distinguishes the two phases.
3029
3030 Arguments:
3031 optionsptr pointer to the option bits
3032 codeptr points to the pointer to the current code point
3033 ptrptr points to the current pattern pointer
3034 errorcodeptr points to error code variable
3035 firstbyteptr set to initial literal character, or < 0 (REQ_UNSET, REQ_NONE)
3036 reqbyteptr set to the last literal character required, else < 0
3037 bcptr points to current branch chain
3038 cd contains pointers to tables etc.
3039 lengthptr NULL during the real compile phase
3040 points to length accumulator during pre-compile phase
3041
3042 Returns: TRUE on success
3043 FALSE, with *errorcodeptr set non-zero on error
3044 */
3045
3046 static BOOL
3047 compile_branch(int *optionsptr, uschar **codeptr, const uschar **ptrptr,
3048 int *errorcodeptr, int *firstbyteptr, int *reqbyteptr, branch_chain *bcptr,
3049 compile_data *cd, int *lengthptr)
3050 {
3051 int repeat_type, op_type;
3052 int repeat_min = 0, repeat_max = 0; /* To please picky compilers */
3053 int bravalue = 0;
3054 int greedy_default, greedy_non_default;
3055 int firstbyte, reqbyte;
3056 int zeroreqbyte, zerofirstbyte;
3057 int req_caseopt, reqvary, tempreqvary;
3058 int options = *optionsptr; /* May change dynamically */
3059 int after_manual_callout = 0;
3060 int length_prevgroup = 0;
3061 register int c;
3062 register uschar *code = *codeptr;
3063 uschar *last_code = code;
3064 uschar *orig_code = code;
3065 uschar *tempcode;
3066 BOOL inescq = FALSE;
3067 BOOL groupsetfirstbyte = FALSE;
3068 const uschar *ptr = *ptrptr;
3069 const uschar *tempptr;
3070 const uschar *nestptr = NULL;
3071 uschar *previous = NULL;
3072 uschar *previous_callout = NULL;
3073 uschar *save_hwm = NULL;
3074 uschar classbits[32];
3075
3076 /* We can fish out the UTF-8 setting once and for all into a BOOL, but we
3077 must not do this for other options (e.g. PCRE_EXTENDED) because they may change
3078 dynamically as we process the pattern. */
3079
3080 #ifdef SUPPORT_UTF8
3081 BOOL class_utf8;
3082 BOOL utf8 = (options & PCRE_UTF8) != 0;
3083 uschar *class_utf8data;
3084 uschar *class_utf8data_base;
3085 uschar utf8_char[6];
3086 #else
3087 BOOL utf8 = FALSE;
3088 uschar *utf8_char = NULL;
3089 #endif
3090
3091 #ifdef PCRE_DEBUG
3092 if (lengthptr != NULL) DPRINTF((">> start branch\n"));
3093 #endif
3094
3095 /* Set up the default and non-default settings for greediness */
3096
3097 greedy_default = ((options & PCRE_UNGREEDY) != 0);
3098 greedy_non_default = greedy_default ^ 1;
3099
3100 /* Initialize no first byte, no required byte. REQ_UNSET means "no char
3101 matching encountered yet". It gets changed to REQ_NONE if we hit something that
3102 matches a non-fixed char first char; reqbyte just remains unset if we never
3103 find one.
3104
3105 When we hit a repeat whose minimum is zero, we may have to adjust these values
3106 to take the zero repeat into account. This is implemented by setting them to
3107 zerofirstbyte and zeroreqbyte when such a repeat is encountered. The individual
3108 item types that can be repeated set these backoff variables appropriately. */
3109
3110 firstbyte = reqbyte = zerofirstbyte = zeroreqbyte = REQ_UNSET;
3111
3112 /* The variable req_caseopt contains either the REQ_CASELESS value or zero,
3113 according to the current setting of the caseless flag. REQ_CASELESS is a bit
3114 value > 255. It is added into the firstbyte or reqbyte variables to record the
3115 case status of the value. This is used only for ASCII characters. */
3116
3117 req_caseopt = ((options & PCRE_CASELESS) != 0)? REQ_CASELESS : 0;
3118
3119 /* Switch on next character until the end of the branch */
3120
3121 for (;; ptr++)
3122 {
3123 BOOL negate_class;
3124 BOOL should_flip_negation;
3125 BOOL possessive_quantifier;
3126 BOOL is_quantifier;
3127 BOOL is_recurse;
3128 BOOL reset_bracount;
3129 int class_charcount;
3130 int class_lastchar;
3131 int newoptions;
3132 int recno;
3133 int refsign;
3134 int skipbytes;
3135 int subreqbyte;
3136 int subfirstbyte;
3137 int terminator;
3138 int mclength;
3139 uschar mcbuffer[8];
3140
3141 /* Get next byte in the pattern */
3142
3143 c = *ptr;
3144
3145 /* If we are at the end of a nested substitution, revert to the outer level
3146 string. Nesting only happens one level deep. */
3147
3148 if (c == 0 && nestptr != NULL)
3149 {
3150 ptr = nestptr;
3151 nestptr = NULL;
3152 c = *ptr;
3153 }
3154
3155 /* If we are in the pre-compile phase, accumulate the length used for the
3156 previous cycle of this loop. */
3157
3158 if (lengthptr != NULL)
3159 {
3160 #ifdef PCRE_DEBUG
3161 if (code > cd->hwm) cd->hwm = code; /* High water info */
3162 #endif
3163 if (code > cd->start_workspace + WORK_SIZE_CHECK) /* Check for overrun */
3164 {
3165 *errorcodeptr = ERR52;
3166 goto FAILED;
3167 }
3168
3169 /* There is at least one situation where code goes backwards: this is the
3170 case of a zero quantifier after a class (e.g. [ab]{0}). At compile time,
3171 the class is simply eliminated. However, it is created first, so we have to
3172 allow memory for it. Therefore, don't ever reduce the length at this point.
3173 */
3174
3175 if (code < last_code) code = last_code;
3176
3177 /* Paranoid check for integer overflow */
3178
3179 if (OFLOW_MAX - *lengthptr < code - last_code)
3180 {
3181 *errorcodeptr = ERR20;
3182 goto FAILED;
3183 }
3184
3185 *lengthptr += (int)(code - last_code);
3186 DPRINTF(("length=%d added %d c=%c\n", *lengthptr, code - last_code, c));
3187
3188 /* If "previous" is set and it is not at the start of the work space, move
3189 it back to there, in order to avoid filling up the work space. Otherwise,
3190 if "previous" is NULL, reset the current code pointer to the start. */
3191
3192 if (previous != NULL)
3193 {
3194 if (previous > orig_code)
3195 {
3196 memmove(orig_code, previous, code - previous);
3197 code -= previous - orig_code;
3198 previous = orig_code;
3199 }
3200 }
3201 else code = orig_code;
3202
3203 /* Remember where this code item starts so we can pick up the length
3204 next time round. */
3205
3206 last_code = code;
3207 }
3208
3209 /* In the real compile phase, just check the workspace used by the forward
3210 reference list. */
3211
3212 else if (cd->hwm > cd->start_workspace + WORK_SIZE_CHECK)
3213 {
3214 *errorcodeptr = ERR52;
3215 goto FAILED;
3216 }
3217
3218 /* If in \Q...\E, check for the end; if not, we have a literal */
3219
3220 if (inescq && c != 0)
3221 {
3222 if (c == CHAR_BACKSLASH && ptr[1] == CHAR_E)
3223 {
3224 inescq = FALSE;
3225 ptr++;
3226 continue;
3227 }
3228 else
3229 {
3230 if (previous_callout != NULL)
3231 {
3232 if (lengthptr == NULL) /* Don't attempt in pre-compile phase */
3233 complete_callout(previous_callout, ptr, cd);
3234 previous_callout = NULL;
3235 }
3236 if ((options & PCRE_AUTO_CALLOUT) != 0)
3237 {
3238 previous_callout = code;
3239 code = auto_callout(code, ptr, cd);
3240 }
3241 goto NORMAL_CHAR;
3242 }
3243 }
3244
3245 /* Fill in length of a previous callout, except when the next thing is
3246 a quantifier. */
3247
3248 is_quantifier =
3249 c == CHAR_ASTERISK || c == CHAR_PLUS || c == CHAR_QUESTION_MARK ||
3250 (c == CHAR_LEFT_CURLY_BRACKET && is_counted_repeat(ptr+1));
3251
3252 if (!is_quantifier && previous_callout != NULL &&
3253 after_manual_callout-- <= 0)
3254 {
3255 if (lengthptr == NULL) /* Don't attempt in pre-compile phase */
3256 complete_callout(previous_callout, ptr, cd);
3257 previous_callout = NULL;
3258 }
3259
3260 /* In extended mode, skip white space and comments. */
3261
3262 if ((options & PCRE_EXTENDED) != 0)
3263 {
3264 if ((cd->ctypes[c] & ctype_space) != 0) continue;
3265 if (c == CHAR_NUMBER_SIGN)
3266 {
3267 ptr++;
3268 while (*ptr != 0)
3269 {
3270 if (IS_NEWLINE(ptr)) { ptr += cd->nllen - 1; break; }
3271 ptr++;
3272 #ifdef SUPPORT_UTF8
3273 if (utf8) while ((*ptr & 0xc0) == 0x80) ptr++;
3274 #endif
3275 }
3276 if (*ptr != 0) continue;
3277
3278 /* Else fall through to handle end of string */
3279 c = 0;
3280 }
3281 }
3282
3283 /* No auto callout for quantifiers. */
3284
3285 if ((options & PCRE_AUTO_CALLOUT) != 0 && !is_quantifier)
3286 {
3287 previous_callout = code;
3288 code = auto_callout(code, ptr, cd);
3289 }
3290
3291 switch(c)
3292 {
3293 /* ===================================================================*/
3294 case 0: /* The branch terminates at string end */
3295 case CHAR_VERTICAL_LINE: /* or | or ) */
3296 case CHAR_RIGHT_PARENTHESIS:
3297 *firstbyteptr = firstbyte;
3298 *reqbyteptr = reqbyte;
3299 *codeptr = code;
3300 *ptrptr = ptr;
3301 if (lengthptr != NULL)
3302 {
3303 if (OFLOW_MAX - *lengthptr < code - last_code)
3304 {
3305 *errorcodeptr = ERR20;
3306 goto FAILED;
3307 }
3308 *lengthptr += (int)(code - last_code); /* To include callout length */
3309 DPRINTF((">> end branch\n"));
3310 }
3311 return TRUE;
3312
3313
3314 /* ===================================================================*/
3315 /* Handle single-character metacharacters. In multiline mode, ^ disables
3316 the setting of any following char as a first character. */
3317
3318 case CHAR_CIRCUMFLEX_ACCENT:
3319 previous = NULL;
3320 if ((options & PCRE_MULTILINE) != 0)
3321 {
3322 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
3323 *code++ = OP_CIRCM;
3324 }
3325 else *code++ = OP_CIRC;
3326 break;
3327
3328 case CHAR_DOLLAR_SIGN:
3329 previous = NULL;
3330 *code++ = ((options & PCRE_MULTILINE) != 0)? OP_DOLLM : OP_DOLL;
3331 break;
3332
3333 /* There can never be a first char if '.' is first, whatever happens about
3334 repeats. The value of reqbyte doesn't change either. */
3335
3336 case CHAR_DOT:
3337 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
3338 zerofirstbyte = firstbyte;
3339 zeroreqbyte = reqbyte;
3340 previous = code;
3341 *code++ = ((options & PCRE_DOTALL) != 0)? OP_ALLANY: OP_ANY;
3342 break;
3343
3344
3345 /* ===================================================================*/
3346 /* Character classes. If the included characters are all < 256, we build a
3347 32-byte bitmap of the permitted characters, except in the special case
3348 where there is only one such character. For negated classes, we build the
3349 map as usual, then invert it at the end. However, we use a different opcode
3350 so that data characters > 255 can be handled correctly.
3351
3352 If the class contains characters outside the 0-255 range, a different
3353 opcode is compiled. It may optionally have a bit map for characters < 256,
3354 but those above are are explicitly listed afterwards. A flag byte tells
3355 whether the bitmap is present, and whether this is a negated class or not.
3356
3357 In JavaScript compatibility mode, an isolated ']' causes an error. In
3358 default (Perl) mode, it is treated as a data character. */
3359
3360 case CHAR_RIGHT_SQUARE_BRACKET:
3361 if ((cd->external_options & PCRE_JAVASCRIPT_COMPAT) != 0)
3362 {
3363 *errorcodeptr = ERR64;
3364 goto FAILED;
3365 }
3366 goto NORMAL_CHAR;
3367
3368 case CHAR_LEFT_SQUARE_BRACKET:
3369 previous = code;
3370
3371 /* PCRE supports POSIX class stuff inside a class. Perl gives an error if
3372 they are encountered at the top level, so we'll do that too. */
3373
3374 if ((ptr[1] == CHAR_COLON || ptr[1] == CHAR_DOT ||
3375 ptr[1] == CHAR_EQUALS_SIGN) &&
3376 check_posix_syntax(ptr, &tempptr))
3377 {
3378 *errorcodeptr = (ptr[1] == CHAR_COLON)? ERR13 : ERR31;
3379 goto FAILED;
3380 }
3381
3382 /* If the first character is '^', set the negation flag and skip it. Also,
3383 if the first few characters (either before or after ^) are \Q\E or \E we
3384 skip them too. This makes for compatibility with Perl. */
3385
3386 negate_class = FALSE;
3387 for (;;)
3388 {
3389 c = *(++ptr);
3390 if (c == CHAR_BACKSLASH)
3391 {
3392 if (ptr[1] == CHAR_E)
3393 ptr++;
3394 else if (strncmp((const char *)ptr+1,
3395 STR_Q STR_BACKSLASH STR_E, 3) == 0)
3396 ptr += 3;
3397 else
3398 break;
3399 }
3400 else if (!negate_class && c == CHAR_CIRCUMFLEX_ACCENT)
3401 negate_class = TRUE;
3402 else break;
3403 }
3404
3405 /* Empty classes are allowed in JavaScript compatibility mode. Otherwise,
3406 an initial ']' is taken as a data character -- the code below handles
3407 that. In JS mode, [] must always fail, so generate OP_FAIL, whereas
3408 [^] must match any character, so generate OP_ALLANY. */
3409
3410 if (c == CHAR_RIGHT_SQUARE_BRACKET &&
3411 (cd->external_options & PCRE_JAVASCRIPT_COMPAT) != 0)
3412 {
3413 *code++ = negate_class? OP_ALLANY : OP_FAIL;
3414 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
3415 zerofirstbyte = firstbyte;
3416 break;
3417 }
3418
3419 /* If a class contains a negative special such as \S, we need to flip the
3420 negation flag at the end, so that support for characters > 255 works
3421 correctly (they are all included in the class). */
3422
3423 should_flip_negation = FALSE;
3424
3425 /* Keep a count of chars with values < 256 so that we can optimize the case
3426 of just a single character (as long as it's < 256). However, For higher
3427 valued UTF-8 characters, we don't yet do any optimization. */
3428
3429 class_charcount = 0;
3430 class_lastchar = -1;
3431
3432 /* Initialize the 32-char bit map to all zeros. We build the map in a
3433 temporary bit of memory, in case the class contains only 1 character (less
3434 than 256), because in that case the compiled code doesn't use the bit map.
3435 */
3436
3437 memset(classbits, 0, 32 * sizeof(uschar));
3438
3439 #ifdef SUPPORT_UTF8
3440 class_utf8 = FALSE; /* No chars >= 256 */
3441 class_utf8data = code + LINK_SIZE + 2; /* For UTF-8 items */
3442 class_utf8data_base = class_utf8data; /* For resetting in pass 1 */
3443 #endif
3444
3445 /* Process characters until ] is reached. By writing this as a "do" it
3446 means that an initial ] is taken as a data character. At the start of the
3447 loop, c contains the first byte of the character. */
3448
3449 if (c != 0) do
3450 {
3451 const uschar *oldptr;
3452
3453 #ifdef SUPPORT_UTF8
3454 if (utf8 && c > 127)
3455 { /* Braces are required because the */
3456 GETCHARLEN(c, ptr, ptr); /* macro generates multiple statements */
3457 }
3458
3459 /* In the pre-compile phase, accumulate the length of any UTF-8 extra
3460 data and reset the pointer. This is so that very large classes that
3461 contain a zillion UTF-8 characters no longer overwrite the work space
3462 (which is on the stack). */
3463
3464 if (lengthptr != NULL)
3465 {
3466 *lengthptr += class_utf8data - class_utf8data_base;
3467 class_utf8data = class_utf8data_base;
3468 }
3469
3470 #endif
3471
3472 /* Inside \Q...\E everything is literal except \E */
3473
3474 if (inescq)
3475 {
3476 if (c == CHAR_BACKSLASH && ptr[1] == CHAR_E) /* If we are at \E */
3477 {
3478 inescq = FALSE; /* Reset literal state */
3479 ptr++; /* Skip the 'E' */
3480 continue; /* Carry on with next */
3481 }
3482 goto CHECK_RANGE; /* Could be range if \E follows */
3483 }
3484
3485 /* Handle POSIX class names. Perl allows a negation extension of the
3486 form [:^name:]. A square bracket that doesn't match the syntax is
3487 treated as a literal. We also recognize the POSIX constructions
3488 [.ch.] and [=ch=] ("collating elements") and fault them, as Perl
3489 5.6 and 5.8 do. */
3490
3491 if (c == CHAR_LEFT_SQUARE_BRACKET &&
3492 (ptr[1] == CHAR_COLON || ptr[1] == CHAR_DOT ||
3493 ptr[1] == CHAR_EQUALS_SIGN) && check_posix_syntax(ptr, &tempptr))
3494 {
3495 BOOL local_negate = FALSE;
3496 int posix_class, taboffset, tabopt;
3497 register const uschar *cbits = cd->cbits;
3498 uschar pbits[32];
3499
3500 if (ptr[1] != CHAR_COLON)
3501 {
3502 *errorcodeptr = ERR31;
3503 goto FAILED;
3504 }
3505
3506 ptr += 2;
3507 if (*ptr == CHAR_CIRCUMFLEX_ACCENT)
3508 {
3509 local_negate = TRUE;
3510 should_flip_negation = TRUE; /* Note negative special */
3511 ptr++;
3512 }
3513
3514 posix_class = check_posix_name(ptr, (int)(tempptr - ptr));
3515 if (posix_class < 0)
3516 {
3517 *errorcodeptr = ERR30;
3518 goto FAILED;
3519 }
3520
3521 /* If matching is caseless, upper and lower are converted to
3522 alpha. This relies on the fact that the class table starts with
3523 alpha, lower, upper as the first 3 entries. */
3524
3525 if ((options & PCRE_CASELESS) != 0 && posix_class <= 2)
3526 posix_class = 0;
3527
3528 /* When PCRE_UCP is set, some of the POSIX classes are converted to
3529 different escape sequences that use Unicode properties. */
3530
3531 #ifdef SUPPORT_UCP
3532 if ((options & PCRE_UCP) != 0)
3533 {
3534 int pc = posix_class + ((local_negate)? POSIX_SUBSIZE/2 : 0);
3535 if (posix_substitutes[pc] != NULL)
3536 {
3537 nestptr = tempptr + 1;
3538 ptr = posix_substitutes[pc] - 1;
3539 continue;
3540 }
3541 }
3542 #endif
3543 /* In the non-UCP case, we build the bit map for the POSIX class in a
3544 chunk of local store because we may be adding and subtracting from it,
3545 and we don't want to subtract bits that may be in the main map already.
3546 At the end we or the result into the bit map that is being built. */
3547
3548 posix_class *= 3;
3549
3550 /* Copy in the first table (always present) */
3551
3552 memcpy(pbits, cbits + posix_class_maps[posix_class],
3553 32 * sizeof(uschar));
3554
3555 /* If there is a second table, add or remove it as required. */
3556
3557 taboffset = posix_class_maps[posix_class + 1];
3558 tabopt = posix_class_maps[posix_class + 2];
3559
3560 if (taboffset >= 0)
3561 {
3562 if (tabopt >= 0)
3563 for (c = 0; c < 32; c++) pbits[c] |= cbits[c + taboffset];
3564 else
3565 for (c = 0; c < 32; c++) pbits[c] &= ~cbits[c + taboffset];
3566 }
3567
3568 /* Not see if we need to remove any special characters. An option
3569 value of 1 removes vertical space and 2 removes underscore. */
3570
3571 if (tabopt < 0) tabopt = -tabopt;
3572 if (tabopt == 1) pbits[1] &= ~0x3c;
3573 else if (tabopt == 2) pbits[11] &= 0x7f;
3574
3575 /* Add the POSIX table or its complement into the main table that is
3576 being built and we are done. */
3577
3578 if (local_negate)
3579 for (c = 0; c < 32; c++) classbits[c] |= ~pbits[c];
3580 else
3581 for (c = 0; c < 32; c++) classbits[c] |= pbits[c];
3582
3583 ptr = tempptr + 1;
3584 class_charcount = 10; /* Set > 1; assumes more than 1 per class */
3585 continue; /* End of POSIX syntax handling */
3586 }
3587
3588 /* Backslash may introduce a single character, or it may introduce one
3589 of the specials, which just set a flag. The sequence \b is a special
3590 case. Inside a class (and only there) it is treated as backspace. We
3591 assume that other escapes have more than one character in them, so set
3592 class_charcount bigger than one. Unrecognized escapes fall through and
3593 are either treated as literal characters (by default), or are faulted if
3594 PCRE_EXTRA is set. */
3595
3596 if (c == CHAR_BACKSLASH)
3597 {
3598 c = check_escape(&ptr, errorcodeptr, cd->bracount, options, TRUE);
3599 if (*errorcodeptr != 0) goto FAILED;
3600
3601 if (-c == ESC_b) c = CHAR_BS; /* \b is backspace in a class */
3602 else if (-c == ESC_Q) /* Handle start of quoted string */
3603 {
3604 if (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
3605 {
3606 ptr += 2; /* avoid empty string */
3607 }
3608 else inescq = TRUE;
3609 continue;
3610 }
3611 else if (-c == ESC_E) continue; /* Ignore orphan \E */
3612
3613 if (c < 0)
3614 {
3615 register const uschar *cbits = cd->cbits;
3616 class_charcount += 2; /* Greater than 1 is what matters */
3617
3618 switch (-c)
3619 {
3620 #ifdef SUPPORT_UCP
3621 case ESC_du: /* These are the values given for \d etc */
3622 case ESC_DU: /* when PCRE_UCP is set. We replace the */
3623 case ESC_wu: /* escape sequence with an appropriate \p */
3624 case ESC_WU: /* or \P to test Unicode properties instead */
3625 case ESC_su: /* of the default ASCII testing. */
3626 case ESC_SU:
3627 nestptr = ptr;
3628 ptr = substitutes[-c - ESC_DU] - 1; /* Just before substitute */
3629 class_charcount -= 2; /* Undo! */
3630 continue;
3631 #endif
3632 case ESC_d:
3633 for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_digit];
3634 continue;
3635
3636 case ESC_D:
3637 should_flip_negation = TRUE;
3638 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_digit];
3639 continue;
3640
3641 case ESC_w:
3642 for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_word];
3643 continue;
3644
3645 case ESC_W:
3646 should_flip_negation = TRUE;
3647 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_word];
3648 continue;
3649
3650 /* Perl 5.004 onwards omits VT from \s, but we must preserve it
3651 if it was previously set by something earlier in the character
3652 class. */
3653
3654 case ESC_s:
3655 classbits[0] |= cbits[cbit_space];
3656 classbits[1] |= cbits[cbit_space+1] & ~0x08;
3657 for (c = 2; c < 32; c++) classbits[c] |= cbits[c+cbit_space];
3658 continue;
3659
3660 case ESC_S:
3661 should_flip_negation = TRUE;
3662 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_space];
3663 classbits[1] |= 0x08; /* Perl 5.004 onwards omits VT from \s */
3664 continue;
3665
3666 case ESC_h:
3667 SETBIT(classbits, 0x09); /* VT */
3668 SETBIT(classbits, 0x20); /* SPACE */
3669 SETBIT(classbits, 0xa0); /* NSBP */
3670 #ifdef SUPPORT_UTF8
3671 if (utf8)
3672 {
3673 class_utf8 = TRUE;
3674 *class_utf8data++ = XCL_SINGLE;
3675 class_utf8data += _pcre_ord2utf8(0x1680, class_utf8data);
3676 *class_utf8data++ = XCL_SINGLE;
3677 class_utf8data += _pcre_ord2utf8(0x180e, class_utf8data);
3678 *class_utf8data++ = XCL_RANGE;
3679 class_utf8data += _pcre_ord2utf8(0x2000, class_utf8data);
3680 class_utf8data += _pcre_ord2utf8(0x200A, class_utf8data);
3681 *class_utf8data++ = XCL_SINGLE;
3682 class_utf8data += _pcre_ord2utf8(0x202f, class_utf8data);
3683 *class_utf8data++ = XCL_SINGLE;
3684 class_utf8data += _pcre_ord2utf8(0x205f, class_utf8data);
3685 *class_utf8data++ = XCL_SINGLE;
3686 class_utf8data += _pcre_ord2utf8(0x3000, class_utf8data);
3687 }
3688 #endif
3689 continue;
3690
3691 case ESC_H:
3692 for (c = 0; c < 32; c++)
3693 {
3694 int x = 0xff;
3695 switch (c)
3696 {
3697 case 0x09/8: x ^= 1 << (0x09%8); break;
3698 case 0x20/8: x ^= 1 << (0x20%8); break;
3699 case 0xa0/8: x ^= 1 << (0xa0%8); break;
3700 default: break;
3701 }
3702 classbits[c] |= x;
3703 }
3704
3705 #ifdef SUPPORT_UTF8
3706 if (utf8)
3707 {
3708 class_utf8 = TRUE;
3709 *class_utf8data++ = XCL_RANGE;
3710 class_utf8data += _pcre_ord2utf8(0x0100, class_utf8data);
3711 class_utf8data += _pcre_ord2utf8(0x167f, class_utf8data);
3712 *class_utf8data++ = XCL_RANGE;
3713 class_utf8data += _pcre_ord2utf8(0x1681, class_utf8data);
3714 class_utf8data += _pcre_ord2utf8(0x180d, class_utf8data);
3715 *class_utf8data++ = XCL_RANGE;
3716 class_utf8data += _pcre_ord2utf8(0x180f, class_utf8data);
3717 class_utf8data += _pcre_ord2utf8(0x1fff, class_utf8data);
3718 *class_utf8data++ = XCL_RANGE;
3719 class_utf8data += _pcre_ord2utf8(0x200B, class_utf8data);
3720 class_utf8data += _pcre_ord2utf8(0x202e, class_utf8data);
3721 *class_utf8data++ = XCL_RANGE;
3722 class_utf8data += _pcre_ord2utf8(0x2030, class_utf8data);
3723 class_utf8data += _pcre_ord2utf8(0x205e, class_utf8data);
3724 *class_utf8data++ = XCL_RANGE;
3725 class_utf8data += _pcre_ord2utf8(0x2060, class_utf8data);
3726 class_utf8data += _pcre_ord2utf8(0x2fff, class_utf8data);
3727 *class_utf8data++ = XCL_RANGE;
3728 class_utf8data += _pcre_ord2utf8(0x3001, class_utf8data);
3729 class_utf8data += _pcre_ord2utf8(0x7fffffff, class_utf8data);
3730 }
3731 #endif
3732 continue;
3733
3734 case ESC_v:
3735 SETBIT(classbits, 0x0a); /* LF */
3736 SETBIT(classbits, 0x0b); /* VT */
3737 SETBIT(classbits, 0x0c); /* FF */
3738 SETBIT(classbits, 0x0d); /* CR */
3739 SETBIT(classbits, 0x85); /* NEL */
3740 #ifdef SUPPORT_UTF8
3741 if (utf8)
3742 {
3743 class_utf8 = TRUE;
3744 *class_utf8data++ = XCL_RANGE;
3745 class_utf8data += _pcre_ord2utf8(0x2028, class_utf8data);
3746 class_utf8data += _pcre_ord2utf8(0x2029, class_utf8data);
3747 }
3748 #endif
3749 continue;
3750
3751 case ESC_V:
3752 for (c = 0; c < 32; c++)
3753 {
3754 int x = 0xff;
3755 switch (c)
3756 {
3757 case 0x0a/8: x ^= 1 << (0x0a%8);
3758 x ^= 1 << (0x0b%8);
3759 x ^= 1 << (0x0c%8);
3760 x ^= 1 << (0x0d%8);
3761 break;
3762 case 0x85/8: x ^= 1 << (0x85%8); break;
3763 default: break;
3764 }
3765 classbits[c] |= x;
3766 }
3767
3768 #ifdef SUPPORT_UTF8
3769 if (utf8)
3770 {
3771 class_utf8 = TRUE;
3772 *class_utf8data++ = XCL_RANGE;
3773 class_utf8data += _pcre_ord2utf8(0x0100, class_utf8data);
3774 class_utf8data += _pcre_ord2utf8(0x2027, class_utf8data);
3775 *class_utf8data++ = XCL_RANGE;
3776 class_utf8data += _pcre_ord2utf8(0x2029, class_utf8data);
3777 class_utf8data += _pcre_ord2utf8(0x7fffffff, class_utf8data);
3778 }
3779 #endif
3780 continue;
3781
3782 #ifdef SUPPORT_UCP
3783 case ESC_p:
3784 case ESC_P:
3785 {
3786 BOOL negated;
3787 int pdata;
3788 int ptype = get_ucp(&ptr, &negated, &pdata, errorcodeptr);
3789 if (ptype < 0) goto FAILED;
3790 class_utf8 = TRUE;
3791 *class_utf8data++ = ((-c == ESC_p) != negated)?
3792 XCL_PROP : XCL_NOTPROP;
3793 *class_utf8data++ = ptype;
3794 *class_utf8data++ = pdata;
3795 class_charcount -= 2; /* Not a < 256 character */
3796 continue;
3797 }
3798 #endif
3799 /* Unrecognized escapes are faulted if PCRE is running in its
3800 strict mode. By default, for compatibility with Perl, they are
3801 treated as literals. */
3802
3803 default:
3804 if ((options & PCRE_EXTRA) != 0)
3805 {
3806 *errorcodeptr = ERR7;
3807 goto FAILED;
3808 }
3809 class_charcount -= 2; /* Undo the default count from above */
3810 c = *ptr; /* Get the final character and fall through */
3811 break;
3812 }
3813 }
3814
3815 /* Fall through if we have a single character (c >= 0). This may be
3816 greater than 256 in UTF-8 mode. */
3817
3818 } /* End of backslash handling */
3819
3820 /* A single character may be followed by '-' to form a range. However,
3821 Perl does not permit ']' to be the end of the range. A '-' character
3822 at the end is treated as a literal. Perl ignores orphaned \E sequences
3823 entirely. The code for handling \Q and \E is messy. */
3824
3825 CHECK_RANGE:
3826 while (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
3827 {
3828 inescq = FALSE;
3829 ptr += 2;
3830 }
3831
3832 oldptr = ptr;
3833
3834 /* Remember \r or \n */
3835
3836 if (c == CHAR_CR || c == CHAR_NL) cd->external_flags |= PCRE_HASCRORLF;
3837
3838 /* Check for range */
3839
3840 if (!inescq && ptr[1] == CHAR_MINUS)
3841 {
3842 int d;
3843 ptr += 2;
3844 while (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_E) ptr += 2;
3845
3846 /* If we hit \Q (not followed by \E) at this point, go into escaped
3847 mode. */
3848
3849 while (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_Q)
3850 {
3851 ptr += 2;
3852 if (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_E)
3853 { ptr += 2; continue; }
3854 inescq = TRUE;
3855 break;
3856 }
3857
3858 if (*ptr == 0 || (!inescq && *ptr == CHAR_RIGHT_SQUARE_BRACKET))
3859 {
3860 ptr = oldptr;
3861 goto LONE_SINGLE_CHARACTER;
3862 }
3863
3864 #ifdef SUPPORT_UTF8
3865 if (utf8)
3866 { /* Braces are required because the */
3867 GETCHARLEN(d, ptr, ptr); /* macro generates multiple statements */
3868 }
3869 else
3870 #endif
3871 d = *ptr; /* Not UTF-8 mode */
3872
3873 /* The second part of a range can be a single-character escape, but
3874 not any of the other escapes. Perl 5.6 treats a hyphen as a literal
3875 in such circumstances. */
3876
3877 if (!inescq && d == CHAR_BACKSLASH)
3878 {
3879 d = check_escape(&ptr, errorcodeptr, cd->bracount, options, TRUE);
3880 if (*errorcodeptr != 0) goto FAILED;
3881
3882 /* \b is backspace; any other special means the '-' was literal */
3883
3884 if (d < 0)
3885 {
3886 if (d == -ESC_b) d = CHAR_BS; else
3887 {
3888 ptr = oldptr;
3889 goto LONE_SINGLE_CHARACTER; /* A few lines below */
3890 }
3891 }
3892 }
3893
3894 /* Check that the two values are in the correct order. Optimize
3895 one-character ranges */
3896
3897 if (d < c)
3898 {
3899 *errorcodeptr = ERR8;
3900 goto FAILED;
3901 }
3902
3903 if (d == c) goto LONE_SINGLE_CHARACTER; /* A few lines below */
3904
3905 /* Remember \r or \n */
3906
3907 if (d == CHAR_CR || d == CHAR_NL) cd->external_flags |= PCRE_HASCRORLF;
3908
3909 /* In UTF-8 mode, if the upper limit is > 255, or > 127 for caseless
3910 matching, we have to use an XCLASS with extra data items. Caseless
3911 matching for characters > 127 is available only if UCP support is
3912 available. */
3913
3914 #ifdef SUPPORT_UTF8
3915 if (utf8 && (d > 255 || ((options & PCRE_CASELESS) != 0 && d > 127)))
3916 {
3917 class_utf8 = TRUE;
3918
3919 /* With UCP support, we can find the other case equivalents of
3920 the relevant characters. There may be several ranges. Optimize how
3921 they fit with the basic range. */
3922
3923 #ifdef SUPPORT_UCP
3924 if ((options & PCRE_CASELESS) != 0)
3925 {
3926 unsigned int occ, ocd;
3927 unsigned int cc = c;
3928 unsigned int origd = d;
3929 while (get_othercase_range(&cc, origd, &occ, &ocd))
3930 {
3931 if (occ >= (unsigned int)c &&
3932 ocd <= (unsigned int)d)
3933 continue; /* Skip embedded ranges */
3934
3935 if (occ < (unsigned int)c &&
3936 ocd >= (unsigned int)c - 1) /* Extend the basic range */
3937 { /* if there is overlap, */
3938 c = occ; /* noting that if occ < c */
3939 continue; /* we can't have ocd > d */
3940 } /* because a subrange is */
3941 if (ocd > (unsigned int)d &&
3942 occ <= (unsigned int)d + 1) /* always shorter than */
3943 { /* the basic range. */
3944 d = ocd;
3945 continue;
3946 }
3947
3948 if (occ == ocd)
3949 {
3950 *class_utf8data++ = XCL_SINGLE;
3951 }
3952 else
3953 {
3954 *class_utf8data++ = XCL_RANGE;
3955 class_utf8data += _pcre_ord2utf8(occ, class_utf8data);
3956 }
3957 class_utf8data += _pcre_ord2utf8(ocd, class_utf8data);
3958 }
3959 }
3960 #endif /* SUPPORT_UCP */
3961
3962 /* Now record the original range, possibly modified for UCP caseless
3963 overlapping ranges. */
3964
3965 *class_utf8data++ = XCL_RANGE;
3966 class_utf8data += _pcre_ord2utf8(c, class_utf8data);
3967 class_utf8data += _pcre_ord2utf8(d, class_utf8data);
3968
3969 /* With UCP support, we are done. Without UCP support, there is no
3970 caseless matching for UTF-8 characters > 127; we can use the bit map
3971 for the smaller ones. */
3972
3973 #ifdef SUPPORT_UCP
3974 continue; /* With next character in the class */
3975 #else
3976 if ((options & PCRE_CASELESS) == 0 || c > 127) continue;
3977
3978 /* Adjust upper limit and fall through to set up the map */
3979
3980 d = 127;
3981
3982 #endif /* SUPPORT_UCP */
3983 }
3984 #endif /* SUPPORT_UTF8 */
3985
3986 /* We use the bit map for all cases when not in UTF-8 mode; else
3987 ranges that lie entirely within 0-127 when there is UCP support; else
3988 for partial ranges without UCP support. */
3989
3990 class_charcount += d - c + 1;
3991 class_lastchar = d;
3992
3993 /* We can save a bit of time by skipping this in the pre-compile. */
3994
3995 if (lengthptr == NULL) for (; c <= d; c++)
3996 {
3997 classbits[c/8] |= (1 << (c&7));
3998 if ((options & PCRE_CASELESS) != 0)
3999 {
4000 int uc = cd->fcc[c]; /* flip case */
4001 classbits[uc/8] |= (1 << (uc&7));
4002 }
4003 }
4004
4005 continue; /* Go get the next char in the class */
4006 }
4007
4008 /* Handle a lone single character - we can get here for a normal
4009 non-escape char, or after \ that introduces a single character or for an
4010 apparent range that isn't. */
4011
4012 LONE_SINGLE_CHARACTER:
4013
4014 /* Handle a character that cannot go in the bit map */
4015
4016 #ifdef SUPPORT_UTF8
4017 if (utf8 && (c > 255 || ((options & PCRE_CASELESS) != 0 && c > 127)))
4018 {
4019 class_utf8 = TRUE;
4020 *class_utf8data++ = XCL_SINGLE;
4021 class_utf8data += _pcre_ord2utf8(c, class_utf8data);
4022
4023 #ifdef SUPPORT_UCP
4024 if ((options & PCRE_CASELESS) != 0)
4025 {
4026 unsigned int othercase;
4027 if ((othercase = UCD_OTHERCASE(c)) != c)
4028 {
4029 *class_utf8data++ = XCL_SINGLE;
4030 class_utf8data += _pcre_ord2utf8(othercase, class_utf8data);
4031 }
4032 }
4033 #endif /* SUPPORT_UCP */
4034
4035 }
4036 else
4037 #endif /* SUPPORT_UTF8 */
4038
4039 /* Handle a single-byte character */
4040 {
4041 classbits[c/8] |= (1 << (c&7));
4042 if ((options & PCRE_CASELESS) != 0)
4043 {
4044 c = cd->fcc[c]; /* flip case */
4045 classbits[c/8] |= (1 << (c&7));
4046 }
4047 class_charcount++;
4048 class_lastchar = c;
4049 }
4050 }
4051
4052 /* Loop until ']' reached. This "while" is the end of the "do" far above.
4053 If we are at the end of an internal nested string, revert to the outer
4054 string. */
4055
4056 while (((c = *(++ptr)) != 0 ||
4057 (nestptr != NULL &&
4058 (ptr = nestptr, nestptr = NULL, c = *(++ptr)) != 0)) &&
4059 (c != CHAR_RIGHT_SQUARE_BRACKET || inescq));
4060
4061 /* Check for missing terminating ']' */
4062
4063 if (c == 0)
4064 {
4065 *errorcodeptr = ERR6;
4066 goto FAILED;
4067 }
4068
4069 /* If class_charcount is 1, we saw precisely one character whose value is
4070 less than 256. As long as there were no characters >= 128 and there was no
4071 use of \p or \P, in other words, no use of any XCLASS features, we can
4072 optimize.
4073
4074 In UTF-8 mode, we can optimize the negative case only if there were no
4075 characters >= 128 because OP_NOT and the related opcodes like OP_NOTSTAR
4076 operate on single-bytes characters only. This is an historical hangover.
4077 Maybe one day we can tidy these opcodes to handle multi-byte characters.
4078
4079 The optimization throws away the bit map. We turn the item into a
4080 1-character OP_CHAR[I] if it's positive, or OP_NOT[I] if it's negative.
4081 Note that OP_NOT[I] does not support multibyte characters. In the positive
4082 case, it can cause firstbyte to be set. Otherwise, there can be no first
4083 char if this item is first, whatever repeat count may follow. In the case
4084 of reqbyte, save the previous value for reinstating. */
4085
4086 #ifdef SUPPORT_UTF8
4087 if (class_charcount == 1 && !class_utf8 &&
4088 (!utf8 || !negate_class || class_lastchar < 128))
4089 #else
4090 if (class_charcount == 1)
4091 #endif
4092 {
4093 zeroreqbyte = reqbyte;
4094
4095 /* The OP_NOT[I] opcodes work on one-byte characters only. */
4096
4097 if (negate_class)
4098 {
4099 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
4100 zerofirstbyte = firstbyte;
4101 *code++ = ((options & PCRE_CASELESS) != 0)? OP_NOTI: OP_NOT;
4102 *code++ = class_lastchar;
4103 break;
4104 }
4105
4106 /* For a single, positive character, get the value into mcbuffer, and
4107 then we can handle this with the normal one-character code. */
4108
4109 #ifdef SUPPORT_UTF8
4110 if (utf8 && class_lastchar > 127)
4111 mclength = _pcre_ord2utf8(class_lastchar, mcbuffer);
4112 else
4113 #endif
4114 {
4115 mcbuffer[0] = class_lastchar;
4116 mclength = 1;
4117 }
4118 goto ONE_CHAR;
4119 } /* End of 1-char optimization */
4120
4121 /* The general case - not the one-char optimization. If this is the first
4122 thing in the branch, there can be no first char setting, whatever the
4123 repeat count. Any reqbyte setting must remain unchanged after any kind of
4124 repeat. */
4125
4126 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
4127 zerofirstbyte = firstbyte;
4128 zeroreqbyte = reqbyte;
4129
4130 /* If there are characters with values > 255, we have to compile an
4131 extended class, with its own opcode, unless there was a negated special
4132 such as \S in the class, and PCRE_UCP is not set, because in that case all
4133 characters > 255 are in the class, so any that were explicitly given as
4134 well can be ignored. If (when there are explicit characters > 255 that must
4135 be listed) there are no characters < 256, we can omit the bitmap in the
4136 actual compiled code. */
4137
4138 #ifdef SUPPORT_UTF8
4139 if (class_utf8 && (!should_flip_negation || (options & PCRE_UCP) != 0))
4140 {
4141 *class_utf8data++ = XCL_END; /* Marks the end of extra data */
4142 *code++ = OP_XCLASS;
4143 code += LINK_SIZE;
4144 *code = negate_class? XCL_NOT : 0;
4145
4146 /* If the map is required, move up the extra data to make room for it;
4147 otherwise just move the code pointer to the end of the extra data. */
4148
4149 if (class_charcount > 0)
4150 {
4151 *code++ |= XCL_MAP;
4152 memmove(code + 32, code, class_utf8data - code);
4153 memcpy(code, classbits, 32);
4154 code = class_utf8data + 32;
4155 }
4156 else code = class_utf8data;
4157
4158 /* Now fill in the complete length of the item */
4159
4160 PUT(previous, 1, code - previous);
4161 break; /* End of class handling */
4162 }
4163 #endif
4164
4165 /* If there are no characters > 255, or they are all to be included or
4166 excluded, set the opcode to OP_CLASS or OP_NCLASS, depending on whether the
4167 whole class was negated and whether there were negative specials such as \S
4168 (non-UCP) in the class. Then copy the 32-byte map into the code vector,
4169 negating it if necessary. */
4170
4171 *code++ = (negate_class == should_flip_negation) ? OP_CLASS : OP_NCLASS;
4172 if (negate_class)
4173 {
4174 if (lengthptr == NULL) /* Save time in the pre-compile phase */
4175 for (c = 0; c < 32; c++) code[c] = ~classbits[c];
4176 }
4177 else
4178 {
4179 memcpy(code, classbits, 32);
4180 }
4181 code += 32;
4182 break;
4183
4184
4185 /* ===================================================================*/
4186 /* Various kinds of repeat; '{' is not necessarily a quantifier, but this
4187 has been tested above. */
4188
4189 case CHAR_LEFT_CURLY_BRACKET:
4190 if (!is_quantifier) goto NORMAL_CHAR;
4191 ptr = read_repeat_counts(ptr+1, &repeat_min, &repeat_max, errorcodeptr);
4192 if (*errorcodeptr != 0) goto FAILED;
4193 goto REPEAT;
4194
4195 case CHAR_ASTERISK:
4196 repeat_min = 0;
4197 repeat_max = -1;
4198 goto REPEAT;
4199
4200 case CHAR_PLUS:
4201 repeat_min = 1;
4202 repeat_max = -1;
4203 goto REPEAT;
4204
4205 case CHAR_QUESTION_MARK:
4206 repeat_min = 0;
4207 repeat_max = 1;
4208
4209 REPEAT:
4210 if (previous == NULL)
4211 {
4212 *errorcodeptr = ERR9;
4213 goto FAILED;
4214 }
4215
4216 if (repeat_min == 0)
4217 {
4218 firstbyte = zerofirstbyte; /* Adjust for zero repeat */
4219 reqbyte = zeroreqbyte; /* Ditto */
4220 }
4221
4222 /* Remember whether this is a variable length repeat */
4223
4224 reqvary = (repeat_min == repeat_max)? 0 : REQ_VARY;
4225
4226 op_type = 0; /* Default single-char op codes */
4227 possessive_quantifier = FALSE; /* Default not possessive quantifier */
4228
4229 /* Save start of previous item, in case we have to move it up in order to
4230 insert something before it. */
4231
4232 tempcode = previous;
4233
4234 /* If the next character is '+', we have a possessive quantifier. This
4235 implies greediness, whatever the setting of the PCRE_UNGREEDY option.
4236 If the next character is '?' this is a minimizing repeat, by default,
4237 but if PCRE_UNGREEDY is set, it works the other way round. We change the
4238 repeat type to the non-default. */
4239
4240 if (ptr[1] == CHAR_PLUS)
4241 {
4242 repeat_type = 0; /* Force greedy */
4243 possessive_quantifier = TRUE;
4244 ptr++;
4245 }
4246 else if (ptr[1] == CHAR_QUESTION_MARK)
4247 {
4248 repeat_type = greedy_non_default;
4249 ptr++;
4250 }
4251 else repeat_type = greedy_default;
4252
4253 /* If previous was a recursion call, wrap it in atomic brackets so that
4254 previous becomes the atomic group. All recursions were so wrapped in the
4255 past, but it no longer happens for non-repeated recursions. In fact, the
4256 repeated ones could be re-implemented independently so as not to need this,
4257 but for the moment we rely on the code for repeating groups. */
4258
4259 if (*previous == OP_RECURSE)
4260 {
4261 memmove(previous + 1 + LINK_SIZE, previous, 1 + LINK_SIZE);
4262 *previous = OP_ONCE;
4263 PUT(previous, 1, 2 + 2*LINK_SIZE);
4264 previous[2 + 2*LINK_SIZE] = OP_KET;
4265 PUT(previous, 3 + 2*LINK_SIZE, 2 + 2*LINK_SIZE);
4266 code += 2 + 2 * LINK_SIZE;
4267 length_prevgroup = 3 + 3*LINK_SIZE;
4268
4269 /* When actually compiling, we need to check whether this was a forward
4270 reference, and if so, adjust the offset. */
4271
4272 if (lengthptr == NULL && cd->hwm >= cd->start_workspace + LINK_SIZE)
4273 {
4274 int offset = GET(cd->hwm, -LINK_SIZE);
4275 if (offset == previous + 1 - cd->start_code)
4276 PUT(cd->hwm, -LINK_SIZE, offset + 1 + LINK_SIZE);
4277 }
4278 }
4279
4280 /* Now handle repetition for the different types of item. */
4281
4282 /* If previous was a character match, abolish the item and generate a
4283 repeat item instead. If a char item has a minumum of more than one, ensure
4284 that it is set in reqbyte - it might not be if a sequence such as x{3} is
4285 the first thing in a branch because the x will have gone into firstbyte
4286 instead. */
4287
4288 if (*previous == OP_CHAR || *previous == OP_CHARI)
4289 {
4290 op_type = (*previous == OP_CHAR)? 0 : OP_STARI - OP_STAR;
4291
4292 /* Deal with UTF-8 characters that take up more than one byte. It's
4293 easier to write this out separately than try to macrify it. Use c to
4294 hold the length of the character in bytes, plus 0x80 to flag that it's a
4295 length rather than a small character. */
4296
4297 #ifdef SUPPORT_UTF8
4298 if (utf8 && (code[-1] & 0x80) != 0)
4299 {
4300 uschar *lastchar = code - 1;
4301 while((*lastchar & 0xc0) == 0x80) lastchar--;
4302 c = code - lastchar; /* Length of UTF-8 character */
4303 memcpy(utf8_char, lastchar, c); /* Save the char */
4304 c |= 0x80; /* Flag c as a length */
4305 }
4306 else
4307 #endif
4308
4309 /* Handle the case of a single byte - either with no UTF8 support, or
4310 with UTF-8 disabled, or for a UTF-8 character < 128. */
4311
4312 {
4313 c = code[-1];
4314 if (repeat_min > 1) reqbyte = c | req_caseopt | cd->req_varyopt;
4315 }
4316
4317 /* If the repetition is unlimited, it pays to see if the next thing on
4318 the line is something that cannot possibly match this character. If so,
4319 automatically possessifying this item gains some performance in the case
4320 where the match fails. */
4321
4322 if (!possessive_quantifier &&
4323 repeat_max < 0 &&
4324 check_auto_possessive(previous, utf8, ptr + 1, options, cd))
4325 {
4326 repeat_type = 0; /* Force greedy */
4327 possessive_quantifier = TRUE;
4328 }
4329
4330 goto OUTPUT_SINGLE_REPEAT; /* Code shared with single character types */
4331 }
4332
4333 /* If previous was a single negated character ([^a] or similar), we use
4334 one of the special opcodes, replacing it. The code is shared with single-
4335 character repeats by setting opt_type to add a suitable offset into
4336 repeat_type. We can also test for auto-possessification. OP_NOT and OP_NOTI
4337 are currently used only for single-byte chars. */
4338
4339 else if (*previous == OP_NOT || *previous == OP_NOTI)
4340 {
4341 op_type = ((*previous == OP_NOT)? OP_NOTSTAR : OP_NOTSTARI) - OP_STAR;
4342 c = previous[1];
4343 if (!possessive_quantifier &&
4344 repeat_max < 0 &&
4345 check_auto_possessive(previous, utf8, ptr + 1, options, cd))
4346 {
4347 repeat_type = 0; /* Force greedy */
4348 possessive_quantifier = TRUE;
4349 }
4350 goto OUTPUT_SINGLE_REPEAT;
4351 }
4352
4353 /* If previous was a character type match (\d or similar), abolish it and
4354 create a suitable repeat item. The code is shared with single-character
4355 repeats by setting op_type to add a suitable offset into repeat_type. Note
4356 the the Unicode property types will be present only when SUPPORT_UCP is
4357 defined, but we don't wrap the little bits of code here because it just
4358 makes it horribly messy. */
4359
4360 else if (*previous < OP_EODN)
4361 {
4362 uschar *oldcode;
4363 int prop_type, prop_value;
4364 op_type = OP_TYPESTAR - OP_STAR; /* Use type opcodes */
4365 c = *previous;
4366
4367 if (!possessive_quantifier &&
4368 repeat_max < 0 &&
4369 check_auto_possessive(previous, utf8, ptr + 1, options, cd))
4370 {
4371 repeat_type = 0; /* Force greedy */
4372 possessive_quantifier = TRUE;
4373 }
4374
4375 OUTPUT_SINGLE_REPEAT:
4376 if (*previous == OP_PROP || *previous == OP_NOTPROP)
4377 {
4378 prop_type = previous[1];
4379 prop_value = previous[2];
4380 }
4381 else prop_type = prop_value = -1;
4382
4383 oldcode = code;
4384 code = previous; /* Usually overwrite previous item */
4385
4386 /* If the maximum is zero then the minimum must also be zero; Perl allows
4387 this case, so we do too - by simply omitting the item altogether. */
4388
4389 if (repeat_max == 0) goto END_REPEAT;
4390
4391 /*--------------------------------------------------------------------*/
4392 /* This code is obsolete from release 8.00; the restriction was finally
4393 removed: */
4394
4395 /* All real repeats make it impossible to handle partial matching (maybe
4396 one day we will be able to remove this restriction). */
4397
4398 /* if (repeat_max != 1) cd->external_flags |= PCRE_NOPARTIAL; */
4399 /*--------------------------------------------------------------------*/
4400
4401 /* Combine the op_type with the repeat_type */
4402
4403 repeat_type += op_type;
4404
4405 /* A minimum of zero is handled either as the special case * or ?, or as
4406 an UPTO, with the maximum given. */
4407
4408 if (repeat_min == 0)
4409 {
4410 if (repeat_max == -1) *code++ = OP_STAR + repeat_type;
4411 else if (repeat_max == 1) *code++ = OP_QUERY + repeat_type;
4412 else
4413 {
4414 *code++ = OP_UPTO + repeat_type;
4415 PUT2INC(code, 0, repeat_max);
4416 }
4417 }
4418
4419 /* A repeat minimum of 1 is optimized into some special cases. If the
4420 maximum is unlimited, we use OP_PLUS. Otherwise, the original item is
4421 left in place and, if the maximum is greater than 1, we use OP_UPTO with
4422 one less than the maximum. */
4423
4424 else if (repeat_min == 1)
4425 {
4426 if (repeat_max == -1)
4427 *code++ = OP_PLUS + repeat_type;
4428 else
4429 {
4430 code = oldcode; /* leave previous item in place */
4431 if (repeat_max == 1) goto END_REPEAT;
4432 *code++ = OP_UPTO + repeat_type;
4433 PUT2INC(code, 0, repeat_max - 1);
4434 }
4435 }
4436
4437 /* The case {n,n} is just an EXACT, while the general case {n,m} is
4438 handled as an EXACT followed by an UPTO. */
4439
4440 else
4441 {
4442 *code++ = OP_EXACT + op_type; /* NB EXACT doesn't have repeat_type */
4443 PUT2INC(code, 0, repeat_min);
4444
4445 /* If the maximum is unlimited, insert an OP_STAR. Before doing so,
4446 we have to insert the character for the previous code. For a repeated
4447 Unicode property match, there are two extra bytes that define the
4448 required property. In UTF-8 mode, long characters have their length in
4449 c, with the 0x80 bit as a flag. */
4450
4451 if (repeat_max < 0)
4452 {
4453 #ifdef SUPPORT_UTF8
4454 if (utf8 && c >= 128)
4455 {
4456 memcpy(code, utf8_char, c & 7);
4457 code += c & 7;
4458 }
4459 else
4460 #endif
4461 {
4462 *code++ = c;
4463 if (prop_type >= 0)
4464 {
4465 *code++ = prop_type;
4466 *code++ = prop_value;
4467 }
4468 }
4469 *code++ = OP_STAR + repeat_type;
4470 }
4471
4472 /* Else insert an UPTO if the max is greater than the min, again
4473 preceded by the character, for the previously inserted code. If the
4474 UPTO is just for 1 instance, we can use QUERY instead. */
4475
4476 else if (repeat_max != repeat_min)
4477 {
4478 #ifdef SUPPORT_UTF8
4479 if (utf8 && c >= 128)
4480 {
4481 memcpy(code, utf8_char, c & 7);
4482 code += c & 7;
4483 }
4484 else
4485 #endif
4486 *code++ = c;
4487 if (prop_type >= 0)
4488 {
4489 *code++ = prop_type;
4490 *code++ = prop_value;
4491 }
4492 repeat_max -= repeat_min;
4493
4494 if (repeat_max == 1)
4495 {
4496 *code++ = OP_QUERY + repeat_type;
4497 }
4498 else
4499 {
4500 *code++ = OP_UPTO + repeat_type;
4501 PUT2INC(code, 0, repeat_max);
4502 }
4503 }
4504 }
4505
4506 /* The character or character type itself comes last in all cases. */
4507
4508 #ifdef SUPPORT_UTF8
4509 if (utf8 && c >= 128)
4510 {
4511 memcpy(code, utf8_char, c & 7);
4512 code += c & 7;
4513 }
4514 else
4515 #endif
4516 *code++ = c;
4517
4518 /* For a repeated Unicode property match, there are two extra bytes that
4519 define the required property. */
4520
4521 #ifdef SUPPORT_UCP
4522 if (prop_type >= 0)
4523 {
4524 *code++ = prop_type;
4525 *code++ = prop_value;
4526 }
4527 #endif
4528 }
4529
4530 /* If previous was a character class or a back reference, we put the repeat
4531 stuff after it, but just skip the item if the repeat was {0,0}. */
4532
4533 else if (*previous == OP_CLASS ||
4534 *previous == OP_NCLASS ||
4535 #ifdef SUPPORT_UTF8
4536 *previous == OP_XCLASS ||
4537 #endif
4538 *previous == OP_REF ||
4539 *previous == OP_REFI)
4540 {
4541 if (repeat_max == 0)
4542 {
4543 code = previous;
4544 goto END_REPEAT;
4545 }
4546
4547 /*--------------------------------------------------------------------*/
4548 /* This code is obsolete from release 8.00; the restriction was finally
4549 removed: */
4550
4551 /* All real repeats make it impossible to handle partial matching (maybe
4552 one day we will be able to remove this restriction). */
4553
4554 /* if (repeat_max != 1) cd->external_flags |= PCRE_NOPARTIAL; */
4555 /*--------------------------------------------------------------------*/
4556
4557 if (repeat_min == 0 && repeat_max == -1)
4558 *code++ = OP_CRSTAR + repeat_type;
4559 else if (repeat_min == 1 && repeat_max == -1)
4560 *code++ = OP_CRPLUS + repeat_type;
4561 else if (repeat_min == 0 && repeat_max == 1)
4562 *code++ = OP_CRQUERY + repeat_type;
4563 else
4564 {
4565 *code++ = OP_CRRANGE + repeat_type;
4566 PUT2INC(code, 0, repeat_min);
4567 if (repeat_max == -1) repeat_max = 0; /* 2-byte encoding for max */
4568 PUT2INC(code, 0, repeat_max);
4569 }
4570 }
4571
4572 /* If previous was a bracket group, we may have to replicate it in certain
4573 cases. Note that at this point we can encounter only the "basic" bracket
4574 opcodes such as BRA and CBRA, as this is the place where they get converted
4575 into the more special varieties such as BRAPOS and SBRA. A test for >=
4576 OP_ASSERT and <= OP_COND includes ASSERT, ASSERT_NOT, ASSERTBACK,
4577 ASSERTBACK_NOT, ONCE, BRA, CBRA, and COND. Originally, PCRE did not allow
4578 repetition of assertions, but now it does, for Perl compatibility. */
4579
4580 else if (*previous >= OP_ASSERT && *previous <= OP_COND)
4581 {
4582 register int i;
4583 int len = (int)(code - previous);
4584 uschar *bralink = NULL;
4585 uschar *brazeroptr = NULL;
4586
4587 /* Repeating a DEFINE group is pointless, but Perl allows the syntax, so
4588 we just ignore the repeat. */
4589
4590 if (*previous == OP_COND && previous[LINK_SIZE+1] == OP_DEF)
4591 goto END_REPEAT;
4592
4593 /* There is no sense in actually repeating assertions. The only potential
4594 use of repetition is in cases when the assertion is optional. Therefore,
4595 if the minimum is greater than zero, just ignore the repeat. If the
4596 maximum is not not zero or one, set it to 1. */
4597
4598 if (*previous < OP_ONCE) /* Assertion */
4599 {
4600 if (repeat_min > 0) goto END_REPEAT;
4601 if (repeat_max < 0 || repeat_max > 1) repeat_max = 1;
4602 }
4603
4604 /* The case of a zero minimum is special because of the need to stick
4605 OP_BRAZERO in front of it, and because the group appears once in the
4606 data, whereas in other cases it appears the minimum number of times. For
4607 this reason, it is simplest to treat this case separately, as otherwise
4608 the code gets far too messy. There are several special subcases when the
4609 minimum is zero. */
4610
4611 if (repeat_min == 0)
4612 {
4613 /* If the maximum is also zero, we used to just omit the group from the
4614 output altogether, like this:
4615
4616 ** if (repeat_max == 0)
4617 ** {
4618 ** code = previous;
4619 ** goto END_REPEAT;
4620 ** }
4621
4622 However, that fails when a group or a subgroup within it is referenced
4623 as a subroutine from elsewhere in the pattern, so now we stick in
4624 OP_SKIPZERO in front of it so that it is skipped on execution. As we
4625 don't have a list of which groups are referenced, we cannot do this
4626 selectively.
4627
4628 If the maximum is 1 or unlimited, we just have to stick in the BRAZERO
4629 and do no more at this point. However, we do need to adjust any
4630 OP_RECURSE calls inside the group that refer to the group itself or any
4631 internal or forward referenced group, because the offset is from the
4632 start of the whole regex. Temporarily terminate the pattern while doing
4633 this. */
4634
4635 if (repeat_max <= 1) /* Covers 0, 1, and unlimited */
4636 {
4637 *code = OP_END;
4638 adjust_recurse(previous, 1, utf8, cd, save_hwm);
4639 memmove(previous+1, previous, len);
4640 code++;
4641 if (repeat_max == 0)
4642 {
4643 *previous++ = OP_SKIPZERO;
4644 goto END_REPEAT;
4645 }
4646 brazeroptr = previous; /* Save for possessive optimizing */
4647 *previous++ = OP_BRAZERO + repeat_type;
4648 }
4649
4650 /* If the maximum is greater than 1 and limited, we have to replicate
4651 in a nested fashion, sticking OP_BRAZERO before each set of brackets.
4652 The first one has to be handled carefully because it's the original
4653 copy, which has to be moved up. The remainder can be handled by code
4654 that is common with the non-zero minimum case below. We have to
4655 adjust the value or repeat_max, since one less copy is required. Once
4656 again, we may have to adjust any OP_RECURSE calls inside the group. */
4657
4658 else
4659 {
4660 int offset;
4661 *code = OP_END;
4662 adjust_recurse(previous, 2 + LINK_SIZE, utf8, cd, save_hwm);
4663 memmove(previous + 2 + LINK_SIZE, previous, len);
4664 code += 2 + LINK_SIZE;
4665 *previous++ = OP_BRAZERO + repeat_type;
4666 *previous++ = OP_BRA;
4667
4668 /* We chain together the bracket offset fields that have to be
4669 filled in later when the ends of the brackets are reached. */
4670
4671 offset = (bralink == NULL)? 0 : (int)(previous - bralink);
4672 bralink = previous;
4673 PUTINC(previous, 0, offset);
4674 }
4675
4676 repeat_max--;
4677 }
4678
4679 /* If the minimum is greater than zero, replicate the group as many
4680 times as necessary, and adjust the maximum to the number of subsequent
4681 copies that we need. If we set a first char from the group, and didn't
4682 set a required char, copy the latter from the former. If there are any
4683 forward reference subroutine calls in the group, there will be entries on
4684 the workspace list; replicate these with an appropriate increment. */
4685
4686 else
4687 {
4688 if (repeat_min > 1)
4689 {
4690 /* In the pre-compile phase, we don't actually do the replication. We
4691 just adjust the length as if we had. Do some paranoid checks for
4692 potential integer overflow. The INT64_OR_DOUBLE type is a 64-bit
4693 integer type when available, otherwise double. */
4694
4695 if (lengthptr != NULL)
4696 {
4697 int delta = (repeat_min - 1)*length_prevgroup;
4698 if ((INT64_OR_DOUBLE)(repeat_min - 1)*
4699 (INT64_OR_DOUBLE)length_prevgroup >
4700 (INT64_OR_DOUBLE)INT_MAX ||
4701 OFLOW_MAX - *lengthptr < delta)
4702 {
4703 *errorcodeptr = ERR20;
4704 goto FAILED;
4705 }
4706 *lengthptr += delta;
4707 }
4708
4709 /* This is compiling for real */
4710
4711 else
4712 {
4713 if (groupsetfirstbyte && reqbyte < 0) reqbyte = firstbyte;
4714 for (i = 1; i < repeat_min; i++)
4715 {
4716 uschar *hc;
4717 uschar *this_hwm = cd->hwm;
4718 memcpy(code, previous, len);
4719 for (hc = save_hwm; hc < this_hwm; hc += LINK_SIZE)
4720 {
4721 PUT(cd->hwm, 0, GET(hc, 0) + len);
4722 cd->hwm += LINK_SIZE;
4723 }
4724 save_hwm = this_hwm;
4725 code += len;
4726 }
4727 }
4728 }
4729
4730 if (repeat_max > 0) repeat_max -= repeat_min;
4731 }
4732
4733 /* This code is common to both the zero and non-zero minimum cases. If
4734 the maximum is limited, it replicates the group in a nested fashion,
4735 remembering the bracket starts on a stack. In the case of a zero minimum,
4736 the first one was set up above. In all cases the repeat_max now specifies
4737 the number of additional copies needed. Again, we must remember to
4738 replicate entries on the forward reference list. */
4739
4740 if (repeat_max >= 0)
4741 {
4742 /* In the pre-compile phase, we don't actually do the replication. We
4743 just adjust the length as if we had. For each repetition we must add 1
4744 to the length for BRAZERO and for all but the last repetition we must
4745 add 2 + 2*LINKSIZE to allow for the nesting that occurs. Do some
4746 paranoid checks to avoid integer overflow. The INT64_OR_DOUBLE type is
4747 a 64-bit integer type when available, otherwise double. */
4748
4749 if (lengthptr != NULL && repeat_max > 0)
4750 {
4751 int delta = repeat_max * (length_prevgroup + 1 + 2 + 2*LINK_SIZE) -
4752 2 - 2*LINK_SIZE; /* Last one doesn't nest */
4753 if ((INT64_OR_DOUBLE)repeat_max *
4754 (INT64_OR_DOUBLE)(length_prevgroup + 1 + 2 + 2*LINK_SIZE)
4755 > (INT64_OR_DOUBLE)INT_MAX ||
4756 OFLOW_MAX - *lengthptr < delta)
4757 {
4758 *errorcodeptr = ERR20;
4759 goto FAILED;
4760 }
4761 *lengthptr += delta;
4762 }
4763
4764 /* This is compiling for real */
4765
4766 else for (i = repeat_max - 1; i >= 0; i--)
4767 {
4768 uschar *hc;
4769 uschar *this_hwm = cd->hwm;
4770
4771 *code++ = OP_BRAZERO + repeat_type;
4772
4773 /* All but the final copy start a new nesting, maintaining the
4774 chain of brackets outstanding. */
4775
4776 if (i != 0)
4777 {
4778 int offset;
4779 *code++ = OP_BRA;
4780 offset = (bralink == NULL)? 0 : (int)(code - bralink);
4781 bralink = code;
4782 PUTINC(code, 0, offset);
4783 }
4784
4785 memcpy(code, previous, len);
4786 for (hc = save_hwm; hc < this_hwm; hc += LINK_SIZE)
4787 {
4788 PUT(cd->hwm, 0, GET(hc, 0) + len + ((i != 0)? 2+LINK_SIZE : 1));
4789 cd->hwm += LINK_SIZE;
4790 }
4791 save_hwm = this_hwm;
4792 code += len;
4793 }
4794
4795 /* Now chain through the pending brackets, and fill in their length
4796 fields (which are holding the chain links pro tem). */
4797
4798 while (bralink != NULL)
4799 {
4800 int oldlinkoffset;
4801 int offset = (int)(code - bralink + 1);
4802 uschar *bra = code - offset;
4803 oldlinkoffset = GET(bra, 1);
4804 bralink = (oldlinkoffset == 0)? NULL : bralink - oldlinkoffset;
4805 *code++ = OP_KET;
4806 PUTINC(code, 0, offset);
4807 PUT(bra, 1, offset);
4808 }
4809 }
4810
4811 /* If the maximum is unlimited, set a repeater in the final copy. For
4812 ONCE brackets, that's all we need to do. However, possessively repeated
4813 ONCE brackets can be converted into non-capturing brackets, as the
4814 behaviour of (?:xx)++ is the same as (?>xx)++ and this saves having to
4815 deal with possessive ONCEs specially.
4816
4817 Otherwise, if the quantifier was possessive, we convert the BRA code to
4818 the POS form, and the KET code to KETRPOS. (It turns out to be convenient
4819 at runtime to detect this kind of subpattern at both the start and at the
4820 end.) The use of special opcodes makes it possible to reduce greatly the
4821 stack usage in pcre_exec(). If the group is preceded by OP_BRAZERO,
4822 convert this to OP_BRAPOSZERO. Then cancel the possessive flag so that
4823 the default action below, of wrapping everything inside atomic brackets,
4824 does not happen.
4825
4826 Then, when we are doing the actual compile phase, check to see whether
4827 this group is one that could match an empty string. If so, convert the
4828 initial operator to the S form (e.g. OP_BRA -> OP_SBRA) so that runtime
4829 checking can be done. [This check is also applied to ONCE groups at
4830 runtime, but in a different way.] */
4831
4832 else
4833 {
4834 uschar *ketcode = code - 1 - LINK_SIZE;
4835 uschar *bracode = ketcode - GET(ketcode, 1);
4836
4837 if (*bracode == OP_ONCE && possessive_quantifier) *bracode = OP_BRA;
4838 if (*bracode == OP_ONCE)
4839 *ketcode = OP_KETRMAX + repeat_type;
4840 else
4841 {
4842 if (possessive_quantifier)
4843 {
4844 *bracode += 1; /* Switch to xxxPOS opcodes */
4845 *ketcode = OP_KETRPOS;
4846 if (brazeroptr != NULL) *brazeroptr = OP_BRAPOSZERO;
4847 possessive_quantifier = FALSE;
4848 }
4849 else *ketcode = OP_KETRMAX + repeat_type;
4850
4851 if (lengthptr == NULL)
4852 {
4853 uschar *scode = bracode;
4854 do
4855 {
4856 if (could_be_empty_branch(scode, ketcode, utf8, cd))
4857 {
4858 *bracode += OP_SBRA - OP_BRA;
4859 break;
4860 }
4861 scode += GET(scode, 1);
4862 }
4863 while (*scode == OP_ALT);
4864 }
4865 }
4866 }
4867 }
4868
4869 /* If previous is OP_FAIL, it was generated by an empty class [] in
4870 JavaScript mode. The other ways in which OP_FAIL can be generated, that is
4871 by (*FAIL) or (?!) set previous to NULL, which gives a "nothing to repeat"
4872 error above. We can just ignore the repeat in JS case. */
4873
4874 else if (*previous == OP_FAIL) goto END_REPEAT;
4875
4876 /* Else there's some kind of shambles */
4877
4878 else
4879 {
4880 *errorcodeptr = ERR11;
4881 goto FAILED;
4882 }
4883
4884 /* If the character following a repeat is '+', or if certain optimization
4885 tests above succeeded, possessive_quantifier is TRUE. For some opcodes,
4886 there are special alternative opcodes for this case. For anything else, we
4887 wrap the entire repeated item inside OP_ONCE brackets. Logically, the '+'
4888 notation is just syntactic sugar, taken from Sun's Java package, but the
4889 special opcodes can optimize it.
4890
4891 Possessively repeated subpatterns have already been handled in the code
4892 just above, so possessive_quantifier is always FALSE for them at this
4893 stage.
4894
4895 Note that the repeated item starts at tempcode, not at previous, which
4896 might be the first part of a string whose (former) last char we repeated.
4897
4898 Possessifying an 'exact' quantifier has no effect, so we can ignore it. But
4899 an 'upto' may follow. We skip over an 'exact' item, and then test the
4900 length of what remains before proceeding. */
4901
4902 if (possessive_quantifier)
4903 {
4904 int len;
4905
4906 if (*tempcode == OP_TYPEEXACT)
4907 tempcode += _pcre_OP_lengths[*tempcode] +
4908 ((tempcode[3] == OP_PROP || tempcode[3] == OP_NOTPROP)? 2 : 0);
4909
4910 else if (*tempcode == OP_EXACT || *tempcode == OP_NOTEXACT)
4911 {
4912 tempcode += _pcre_OP_lengths[*tempcode];
4913 #ifdef SUPPORT_UTF8
4914 if (utf8 && tempcode[-1] >= 0xc0)
4915 tempcode += _pcre_utf8_table4[tempcode[-1] & 0x3f];
4916 #endif
4917 }
4918
4919 len = (int)(code - tempcode);
4920 if (len > 0) switch (*tempcode)
4921 {
4922 case OP_STAR: *tempcode = OP_POSSTAR; break;
4923 case OP_PLUS: *tempcode = OP_POSPLUS; break;
4924 case OP_QUERY: *tempcode = OP_POSQUERY; break;
4925 case OP_UPTO: *tempcode = OP_POSUPTO; break;
4926
4927 case OP_STARI: *tempcode = OP_POSSTARI; break;
4928 case OP_PLUSI: *tempcode = OP_POSPLUSI; break;
4929 case OP_QUERYI: *tempcode = OP_POSQUERYI; break;
4930 case OP_UPTOI: *tempcode = OP_POSUPTOI; break;
4931
4932 case OP_NOTSTAR: *tempcode = OP_NOTPOSSTAR; break;
4933 case OP_NOTPLUS: *tempcode = OP_NOTPOSPLUS; break;
4934 case OP_NOTQUERY: *tempcode = OP_NOTPOSQUERY; break;
4935 case OP_NOTUPTO: *tempcode = OP_NOTPOSUPTO; break;
4936
4937 case OP_NOTSTARI: *tempcode = OP_NOTPOSSTARI; break;
4938 case OP_NOTPLUSI: *tempcode = OP_NOTPOSPLUSI; break;
4939 case OP_NOTQUERYI: *tempcode = OP_NOTPOSQUERYI; break;
4940 case OP_NOTUPTOI: *tempcode = OP_NOTPOSUPTOI; break;
4941
4942 case OP_TYPESTAR: *tempcode = OP_TYPEPOSSTAR; break;
4943 case OP_TYPEPLUS: *tempcode = OP_TYPEPOSPLUS; break;
4944 case OP_TYPEQUERY: *tempcode = OP_TYPEPOSQUERY; break;
4945 case OP_TYPEUPTO: *tempcode = OP_TYPEPOSUPTO; break;
4946
4947 /* Because we are moving code along, we must ensure that any
4948 pending recursive references are updated. */
4949
4950 default:
4951 *code = OP_END;
4952 adjust_recurse(tempcode, 1 + LINK_SIZE, utf8, cd, save_hwm);
4953 memmove(tempcode + 1+LINK_SIZE, tempcode, len);
4954 code += 1 + LINK_SIZE;
4955 len += 1 + LINK_SIZE;
4956 tempcode[0] = OP_ONCE;
4957 *code++ = OP_KET;
4958 PUTINC(code, 0, len);
4959 PUT(tempcode, 1, len);
4960 break;
4961 }
4962 }
4963
4964 /* In all case we no longer have a previous item. We also set the
4965 "follows varying string" flag for subsequently encountered reqbytes if
4966 it isn't already set and we have just passed a varying length item. */
4967
4968 END_REPEAT:
4969 previous = NULL;
4970 cd->req_varyopt |= reqvary;
4971 break;
4972
4973
4974 /* ===================================================================*/
4975 /* Start of nested parenthesized sub-expression, or comment or lookahead or
4976 lookbehind or option setting or condition or all the other extended
4977 parenthesis forms. */
4978
4979 case CHAR_LEFT_PARENTHESIS:
4980 newoptions = options;
4981 skipbytes = 0;
4982 bravalue = OP_CBRA;
4983 save_hwm = cd->hwm;
4984 reset_bracount = FALSE;
4985
4986 /* First deal with various "verbs" that can be introduced by '*'. */
4987
4988 if (*(++ptr) == CHAR_ASTERISK &&
4989 ((cd->ctypes[ptr[1]] & ctype_letter) != 0 || ptr[1] == ':'))
4990 {
4991 int i, namelen;
4992 int arglen = 0;
4993 const char *vn = verbnames;
4994 const uschar *name = ptr + 1;
4995 const uschar *arg = NULL;
4996 previous = NULL;
4997 while ((cd->ctypes[*++ptr] & ctype_letter) != 0) {};
4998 namelen = (int)(ptr - name);
4999
5000 if (*ptr == CHAR_COLON)
5001 {
5002 arg = ++ptr;
5003 while ((cd->ctypes[*ptr] & (ctype_letter|ctype_digit)) != 0
5004 || *ptr == '_') ptr++;
5005 arglen = (int)(ptr - arg);
5006 }
5007
5008 if (*ptr != CHAR_RIGHT_PARENTHESIS)
5009 {
5010 *errorcodeptr = ERR60;
5011 goto FAILED;
5012 }
5013
5014 /* Scan the table of verb names */
5015
5016 for (i = 0; i < verbcount; i++)
5017 {
5018 if (namelen == verbs[i].len &&
5019 strncmp((char *)name, vn, namelen) == 0)
5020 {
5021 /* Check for open captures before ACCEPT and convert it to
5022 ASSERT_ACCEPT if in an assertion. */
5023
5024 if (verbs[i].op == OP_ACCEPT)
5025 {
5026 open_capitem *oc;
5027 if (arglen != 0)
5028 {
5029 *errorcodeptr = ERR59;
5030 goto FAILED;
5031 }
5032 cd->had_accept = TRUE;
5033 for (oc = cd->open_caps; oc != NULL; oc = oc->next)
5034 {
5035 *code++ = OP_CLOSE;
5036 PUT2INC(code, 0, oc->number);
5037 }
5038 *code++ = (cd->assert_depth > 0)? OP_ASSERT_ACCEPT : OP_ACCEPT;
5039 }
5040
5041 /* Handle other cases with/without an argument */
5042
5043 else if (arglen == 0)
5044 {
5045 if (verbs[i].op < 0) /* Argument is mandatory */
5046 {
5047 *errorcodeptr = ERR66;
5048 goto FAILED;
5049 }
5050 *code = verbs[i].op;
5051 if (*code++ == OP_THEN)
5052 {
5053 PUT(code, 0, code - bcptr->current_branch - 1);
5054 code += LINK_SIZE;
5055 }
5056 }
5057
5058 else
5059 {
5060 if (verbs[i].op_arg < 0) /* Argument is forbidden */
5061 {
5062 *errorcodeptr = ERR59;
5063 goto FAILED;
5064 }
5065 *code = verbs[i].op_arg;
5066 if (*code++ == OP_THEN_ARG)
5067 {
5068 PUT(code, 0, code - bcptr->current_branch - 1);
5069 code += LINK_SIZE;
5070 }
5071 *code++ = arglen;
5072 memcpy(code, arg, arglen);
5073 code += arglen;
5074 *code++ = 0;
5075 }
5076
5077 break; /* Found verb, exit loop */
5078 }
5079
5080 vn += verbs[i].len + 1;
5081 }
5082
5083 if (i < verbcount) continue; /* Successfully handled a verb */
5084 *errorcodeptr = ERR60; /* Verb not recognized */
5085 goto FAILED;
5086 }
5087
5088 /* Deal with the extended parentheses; all are introduced by '?', and the
5089 appearance of any of them means that this is not a capturing group. */
5090
5091 else if (*ptr == CHAR_QUESTION_MARK)
5092 {
5093 int i, set, unset, namelen;
5094 int *optset;
5095 const uschar *name;
5096 uschar *slot;
5097
5098 switch (*(++ptr))
5099 {
5100 case CHAR_NUMBER_SIGN: /* Comment; skip to ket */
5101 ptr++;
5102 while (*ptr != 0 && *ptr != CHAR_RIGHT_PARENTHESIS) ptr++;
5103 if (*ptr == 0)
5104 {
5105 *errorcodeptr = ERR18;
5106 goto FAILED;
5107 }
5108 continue;
5109
5110
5111 /* ------------------------------------------------------------ */
5112 case CHAR_VERTICAL_LINE: /* Reset capture count for each branch */
5113 reset_bracount = TRUE;
5114 /* Fall through */
5115
5116 /* ------------------------------------------------------------ */
5117 case CHAR_COLON: /* Non-capturing bracket */
5118 bravalue = OP_BRA;
5119 ptr++;
5120 break;
5121
5122
5123 /* ------------------------------------------------------------ */
5124 case CHAR_LEFT_PARENTHESIS:
5125 bravalue = OP_COND; /* Conditional group */
5126
5127 /* A condition can be an assertion, a number (referring to a numbered
5128 group), a name (referring to a named group), or 'R', referring to
5129 recursion. R<digits> and R&name are also permitted for recursion tests.
5130
5131 There are several syntaxes for testing a named group: (?(name)) is used
5132 by Python; Perl 5.10 onwards uses (?(<name>) or (?('name')).
5133
5134 There are two unfortunate ambiguities, caused by history. (a) 'R' can
5135 be the recursive thing or the name 'R' (and similarly for 'R' followed
5136 by digits), and (b) a number could be a name that consists of digits.
5137 In both cases, we look for a name first; if not found, we try the other
5138 cases. */
5139
5140 /* For conditions that are assertions, check the syntax, and then exit
5141 the switch. This will take control down to where bracketed groups,
5142 including assertions, are processed. */
5143
5144 if (ptr[1] == CHAR_QUESTION_MARK && (ptr[2] == CHAR_EQUALS_SIGN ||
5145 ptr[2] == CHAR_EXCLAMATION_MARK || ptr[2] == CHAR_LESS_THAN_SIGN))
5146 break;
5147
5148 /* Most other conditions use OP_CREF (a couple change to OP_RREF
5149 below), and all need to skip 3 bytes at the start of the group. */
5150
5151 code[1+LINK_SIZE] = OP_CREF;
5152 skipbytes = 3;
5153 refsign = -1;
5154
5155 /* Check for a test for recursion in a named group. */
5156
5157 if (ptr[1] == CHAR_R && ptr[2] == CHAR_AMPERSAND)
5158 {
5159 terminator = -1;
5160 ptr += 2;
5161 code[1+LINK_SIZE] = OP_RREF; /* Change the type of test */
5162 }
5163
5164 /* Check for a test for a named group's having been set, using the Perl
5165 syntax (?(<name>) or (?('name') */
5166
5167 else if (ptr[1] == CHAR_LESS_THAN_SIGN)
5168 {
5169 terminator = CHAR_GREATER_THAN_SIGN;
5170 ptr++;
5171 }
5172 else if (ptr[1] == CHAR_APOSTROPHE)
5173 {
5174 terminator = CHAR_APOSTROPHE;
5175 ptr++;
5176 }
5177 else
5178 {
5179 terminator = 0;
5180 if (ptr[1] == CHAR_MINUS || ptr[1] == CHAR_PLUS) refsign = *(++ptr);
5181 }
5182
5183 /* We now expect to read a name; any thing else is an error */
5184
5185 if ((cd->ctypes[ptr[1]] & ctype_word) == 0)
5186 {
5187 ptr += 1; /* To get the right offset */
5188 *errorcodeptr = ERR28;
5189 goto FAILED;
5190 }
5191
5192 /* Read the name, but also get it as a number if it's all digits */
5193
5194 recno = 0;
5195 name = ++ptr;
5196 while ((cd->ctypes[*ptr] & ctype_word) != 0)
5197 {
5198 if (recno >= 0)
5199 recno = ((digitab[*ptr] & ctype_digit) != 0)?
5200 recno * 10 + *ptr - CHAR_0 : -1;
5201 ptr++;
5202 }
5203 namelen = (int)(ptr - name);
5204
5205 if ((terminator > 0 && *ptr++ != terminator) ||
5206 *ptr++ != CHAR_RIGHT_PARENTHESIS)
5207 {
5208 ptr--; /* Error offset */
5209 *errorcodeptr = ERR26;
5210 goto FAILED;
5211 }
5212
5213 /* Do no further checking in the pre-compile phase. */
5214
5215 if (lengthptr != NULL) break;
5216
5217 /* In the real compile we do the work of looking for the actual
5218 reference. If the string started with "+" or "-" we require the rest to
5219 be digits, in which case recno will be set. */
5220
5221 if (refsign > 0)
5222 {
5223 if (recno <= 0)
5224 {
5225 *errorcodeptr = ERR58;
5226 goto FAILED;
5227 }
5228 recno = (refsign == CHAR_MINUS)?
5229 cd->bracount - recno + 1 : recno +cd->bracount;
5230 if (recno <= 0 || recno > cd->final_bracount)
5231 {
5232 *errorcodeptr = ERR15;
5233 goto FAILED;
5234 }
5235 PUT2(code, 2+LINK_SIZE, recno);
5236 break;
5237 }
5238
5239 /* Otherwise (did not start with "+" or "-"), start by looking for the
5240 name. If we find a name, add one to the opcode to change OP_CREF or
5241 OP_RREF into OP_NCREF or OP_NRREF. These behave exactly the same,
5242 except they record that the reference was originally to a name. The
5243 information is used to check duplicate names. */
5244
5245 slot = cd->name_table;
5246 for (i = 0; i < cd->names_found; i++)
5247 {
5248 if (strncmp((char *)name, (char *)slot+2, namelen) == 0) break;
5249 slot += cd->name_entry_size;
5250 }
5251
5252 /* Found a previous named subpattern */
5253
5254 if (i < cd->names_found)
5255 {
5256 recno = GET2(slot, 0);
5257 PUT2(code, 2+LINK_SIZE, recno);
5258 code[1+LINK_SIZE]++;
5259 }
5260
5261 /* Search the pattern for a forward reference */
5262
5263 else if ((i = find_parens(cd, name, namelen,
5264 (options & PCRE_EXTENDED) != 0, utf8)) > 0)
5265 {
5266 PUT2(code, 2+LINK_SIZE, i);
5267 code[1+LINK_SIZE]++;
5268 }
5269
5270 /* If terminator == 0 it means that the name followed directly after
5271 the opening parenthesis [e.g. (?(abc)...] and in this case there are
5272 some further alternatives to try. For the cases where terminator != 0
5273 [things like (?(<name>... or (?('name')... or (?(R&name)... ] we have
5274 now checked all the possibilities, so give an error. */
5275
5276 else if (terminator != 0)
5277 {
5278 *errorcodeptr = ERR15;
5279 goto FAILED;
5280 }
5281
5282 /* Check for (?(R) for recursion. Allow digits after R to specify a
5283 specific group number. */
5284
5285 else if (*name == CHAR_R)
5286 {
5287 recno = 0;
5288 for (i = 1; i < namelen; i++)
5289 {
5290 if ((digitab[name[i]] & ctype_digit) == 0)
5291 {
5292 *errorcodeptr = ERR15;
5293 goto FAILED;
5294 }
5295 recno = recno * 10 + name[i] - CHAR_0;
5296 }
5297 if (recno == 0) recno = RREF_ANY;
5298 code[1+LINK_SIZE] = OP_RREF; /* Change test type */
5299 PUT2(code, 2+LINK_SIZE, recno);
5300 }
5301
5302 /* Similarly, check for the (?(DEFINE) "condition", which is always
5303 false. */
5304
5305 else if (namelen == 6 && strncmp((char *)name, STRING_DEFINE, 6) == 0)
5306 {
5307 code[1+LINK_SIZE] = OP_DEF;
5308 skipbytes = 1;
5309 }
5310
5311 /* Check for the "name" actually being a subpattern number. We are
5312 in the second pass here, so final_bracount is set. */
5313
5314 else if (recno > 0 && recno <= cd->final_bracount)
5315 {
5316 PUT2(code, 2+LINK_SIZE, recno);
5317 }
5318
5319 /* Either an unidentified subpattern, or a reference to (?(0) */
5320
5321 else
5322 {
5323 *errorcodeptr = (recno == 0)? ERR35: ERR15;
5324 goto FAILED;
5325 }
5326 break;
5327
5328
5329 /* ------------------------------------------------------------ */
5330 case CHAR_EQUALS_SIGN: /* Positive lookahead */
5331 bravalue = OP_ASSERT;
5332 cd->assert_depth += 1;
5333 ptr++;
5334 break;
5335
5336
5337 /* ------------------------------------------------------------ */
5338 case CHAR_EXCLAMATION_MARK: /* Negative lookahead */
5339 ptr++;
5340 if (*ptr == CHAR_RIGHT_PARENTHESIS) /* Optimize (?!) */
5341 {
5342 *code++ = OP_FAIL;
5343 previous = NULL;
5344 continue;
5345 }
5346 bravalue = OP_ASSERT_NOT;
5347 cd->assert_depth += 1;
5348 break;
5349
5350
5351 /* ------------------------------------------------------------ */
5352 case CHAR_LESS_THAN_SIGN: /* Lookbehind or named define */
5353 switch (ptr[1])
5354 {
5355 case CHAR_EQUALS_SIGN: /* Positive lookbehind */
5356 bravalue = OP_ASSERTBACK;
5357 cd->assert_depth += 1;
5358 ptr += 2;
5359 break;
5360
5361 case CHAR_EXCLAMATION_MARK: /* Negative lookbehind */
5362 bravalue = OP_ASSERTBACK_NOT;
5363 cd->assert_depth += 1;
5364 ptr += 2;
5365 break;
5366
5367 default: /* Could be name define, else bad */
5368 if ((cd->ctypes[ptr[1]] & ctype_word) != 0) goto DEFINE_NAME;
5369 ptr++; /* Correct offset for error */
5370 *errorcodeptr = ERR24;
5371 goto FAILED;
5372 }
5373 break;
5374
5375
5376 /* ------------------------------------------------------------ */
5377 case CHAR_GREATER_THAN_SIGN: /* One-time brackets */
5378 bravalue = OP_ONCE;
5379 ptr++;
5380 break;
5381
5382
5383 /* ------------------------------------------------------------ */
5384 case CHAR_C: /* Callout - may be followed by digits; */
5385 previous_callout = code; /* Save for later completion */
5386 after_manual_callout = 1; /* Skip one item before completing */
5387 *code++ = OP_CALLOUT;
5388 {
5389 int n = 0;
5390 while ((digitab[*(++ptr)] & ctype_digit) != 0)
5391 n = n * 10 + *ptr - CHAR_0;
5392 if (*ptr != CHAR_RIGHT_PARENTHESIS)
5393 {
5394 *errorcodeptr = ERR39;
5395 goto FAILED;
5396 }
5397 if (n > 255)
5398 {
5399 *errorcodeptr = ERR38;
5400 goto FAILED;
5401 }
5402 *code++ = n;
5403 PUT(code, 0, (int)(ptr - cd->start_pattern + 1)); /* Pattern offset */
5404 PUT(code, LINK_SIZE, 0); /* Default length */
5405 code += 2 * LINK_SIZE;
5406 }
5407 previous = NULL;
5408 continue;
5409
5410
5411 /* ------------------------------------------------------------ */
5412 case CHAR_P: /* Python-style named subpattern handling */
5413 if (*(++ptr) == CHAR_EQUALS_SIGN ||
5414 *ptr == CHAR_GREATER_THAN_SIGN) /* Reference or recursion */
5415 {
5416 is_recurse = *ptr == CHAR_GREATER_THAN_SIGN;
5417 terminator = CHAR_RIGHT_PARENTHESIS;
5418 goto NAMED_REF_OR_RECURSE;
5419 }
5420 else if (*ptr != CHAR_LESS_THAN_SIGN) /* Test for Python-style defn */
5421 {
5422 *errorcodeptr = ERR41;
5423 goto FAILED;
5424 }
5425 /* Fall through to handle (?P< as (?< is handled */
5426
5427
5428 /* ------------------------------------------------------------ */
5429 DEFINE_NAME: /* Come here from (?< handling */
5430 case CHAR_APOSTROPHE:
5431 {
5432 terminator = (*ptr == CHAR_LESS_THAN_SIGN)?
5433 CHAR_GREATER_THAN_SIGN : CHAR_APOSTROPHE;
5434 name = ++ptr;
5435
5436 while ((cd->ctypes[*ptr] & ctype_word) != 0) ptr++;
5437 namelen = (int)(ptr - name);
5438
5439 /* In the pre-compile phase, just do a syntax check. */
5440
5441 if (lengthptr != NULL)
5442 {
5443 if (*ptr != terminator)
5444 {
5445 *errorcodeptr = ERR42;
5446 goto FAILED;
5447 }
5448 if (cd->names_found >= MAX_NAME_COUNT)
5449 {
5450 *errorcodeptr = ERR49;
5451 goto FAILED;
5452 }
5453 if (namelen + 3 > cd->name_entry_size)
5454 {
5455 cd->name_entry_size = namelen + 3;
5456 if (namelen > MAX_NAME_SIZE)
5457 {
5458 *errorcodeptr = ERR48;
5459 goto FAILED;
5460 }
5461 }
5462 }
5463
5464 /* In the real compile, create the entry in the table, maintaining
5465 alphabetical order. Duplicate names for different numbers are
5466 permitted only if PCRE_DUPNAMES is set. Duplicate names for the same
5467 number are always OK. (An existing number can be re-used if (?|
5468 appears in the pattern.) In either event, a duplicate name results in
5469 a duplicate entry in the table, even if the number is the same. This
5470 is because the number of names, and hence the table size, is computed
5471 in the pre-compile, and it affects various numbers and pointers which
5472 would all have to be modified, and the compiled code moved down, if
5473 duplicates with the same number were omitted from the table. This
5474 doesn't seem worth the hassle. However, *different* names for the
5475 same number are not permitted. */
5476
5477 else
5478 {
5479 BOOL dupname = FALSE;
5480 slot = cd->name_table;
5481
5482 for (i = 0; i < cd->names_found; i++)
5483 {
5484 int crc = memcmp(name, slot+2, namelen);
5485 if (crc == 0)
5486 {
5487 if (slot[2+namelen] == 0)
5488 {
5489 if (GET2(slot, 0) != cd->bracount + 1 &&
5490 (options & PCRE_DUPNAMES) == 0)
5491 {
5492 *errorcodeptr = ERR43;
5493 goto FAILED;
5494 }
5495 else dupname = TRUE;
5496 }
5497 else crc = -1; /* Current name is a substring */
5498 }
5499
5500 /* Make space in the table and break the loop for an earlier
5501 name. For a duplicate or later name, carry on. We do this for
5502 duplicates so that in the simple case (when ?(| is not used) they
5503 are in order of their numbers. */
5504
5505 if (crc < 0)
5506 {
5507 memmove(slot + cd->name_entry_size, slot,
5508 (cd->names_found - i) * cd->name_entry_size);
5509 break;
5510 }
5511
5512 /* Continue the loop for a later or duplicate name */
5513
5514 slot += cd->name_entry_size;
5515 }
5516
5517 /* For non-duplicate names, check for a duplicate number before
5518 adding the new name. */
5519
5520 if (!dupname)
5521 {
5522 uschar *cslot = cd->name_table;
5523 for (i = 0; i < cd->names_found; i++)
5524 {
5525 if (cslot != slot)
5526 {
5527 if (GET2(cslot, 0) == cd->bracount + 1)
5528 {
5529 *errorcodeptr = ERR65;
5530 goto FAILED;
5531 }
5532 }
5533 else i--;
5534 cslot += cd->name_entry_size;
5535 }
5536 }
5537
5538 PUT2(slot, 0, cd->bracount + 1);
5539 memcpy(slot + 2, name, namelen);
5540 slot[2+namelen] = 0;
5541 }
5542 }
5543
5544 /* In both pre-compile and compile, count the number of names we've
5545 encountered. */
5546
5547 cd->names_found++;
5548 ptr++; /* Move past > or ' */
5549 goto NUMBERED_GROUP;
5550
5551
5552 /* ------------------------------------------------------------ */
5553 case CHAR_AMPERSAND: /* Perl recursion/subroutine syntax */
5554 terminator = CHAR_RIGHT_PARENTHESIS;
5555 is_recurse = TRUE;
5556 /* Fall through */
5557
5558 /* We come here from the Python syntax above that handles both
5559 references (?P=name) and recursion (?P>name), as well as falling
5560 through from the Perl recursion syntax (?&name). We also come here from
5561 the Perl \k<name> or \k'name' back reference syntax and the \k{name}
5562 .NET syntax, and the Oniguruma \g<...> and \g'...' subroutine syntax. */
5563
5564 NAMED_REF_OR_RECURSE:
5565 name = ++ptr;
5566 while ((cd->ctypes[*ptr] & ctype_word) != 0) ptr++;
5567 namelen = (int)(ptr - name);
5568
5569 /* In the pre-compile phase, do a syntax check. We used to just set
5570 a dummy reference number, because it was not used in the first pass.
5571 However, with the change of recursive back references to be atomic,
5572 we have to look for the number so that this state can be identified, as
5573 otherwise the incorrect length is computed. If it's not a backwards
5574 reference, the dummy number will do. */
5575
5576 if (lengthptr != NULL)
5577 {
5578 const uschar *temp;
5579
5580 if (namelen == 0)
5581 {
5582 *errorcodeptr = ERR62;
5583 goto FAILED;
5584 }
5585 if (*ptr != terminator)
5586 {
5587 *errorcodeptr = ERR42;
5588 goto FAILED;
5589 }
5590 if (namelen > MAX_NAME_SIZE)
5591 {
5592 *errorcodeptr = ERR48;
5593 goto FAILED;
5594 }
5595
5596 /* The name table does not exist in the first pass, so we cannot
5597 do a simple search as in the code below. Instead, we have to scan the
5598 pattern to find the number. It is important that we scan it only as
5599 far as we have got because the syntax of named subpatterns has not
5600 been checked for the rest of the pattern, and find_parens() assumes
5601 correct syntax. In any case, it's a waste of resources to scan
5602 further. We stop the scan at the current point by temporarily
5603 adjusting the value of cd->endpattern. */
5604
5605 temp = cd->end_pattern;
5606 cd->end_pattern = ptr;
5607 recno = find_parens(cd, name, namelen,
5608 (options & PCRE_EXTENDED) != 0, utf8);
5609 cd->end_pattern = temp;
5610 if (recno < 0) recno = 0; /* Forward ref; set dummy number */
5611 }
5612
5613 /* In the real compile, seek the name in the table. We check the name
5614 first, and then check that we have reached the end of the name in the
5615 table. That way, if the name that is longer than any in the table,
5616 the comparison will fail without reading beyond the table entry. */
5617
5618 else
5619 {
5620 slot = cd->name_table;
5621 for (i = 0; i < cd->names_found; i++)
5622 {
5623 if (strncmp((char *)name, (char *)slot+2, namelen) == 0 &&
5624 slot[2+namelen] == 0)
5625 break;
5626 slot += cd->name_entry_size;
5627 }
5628
5629 if (i < cd->names_found) /* Back reference */
5630 {
5631 recno = GET2(slot, 0);
5632 }
5633 else if ((recno = /* Forward back reference */
5634 find_parens(cd, name, namelen,
5635 (options & PCRE_EXTENDED) != 0, utf8)) <= 0)
5636 {
5637 *errorcodeptr = ERR15;
5638 goto FAILED;
5639 }
5640 }
5641
5642 /* In both phases, we can now go to the code than handles numerical
5643 recursion or backreferences. */
5644
5645 if (is_recurse) goto HANDLE_RECURSION;
5646 else goto HANDLE_REFERENCE;
5647
5648
5649 /* ------------------------------------------------------------ */
5650 case CHAR_R: /* Recursion */
5651 ptr++; /* Same as (?0) */
5652 /* Fall through */
5653
5654
5655 /* ------------------------------------------------------------ */
5656 case CHAR_MINUS: case CHAR_PLUS: /* Recursion or subroutine */
5657 case CHAR_0: case CHAR_1: case CHAR_2: case CHAR_3: case CHAR_4:
5658 case CHAR_5: case CHAR_6: case CHAR_7: case CHAR_8: case CHAR_9:
5659 {
5660 const uschar *called;
5661 terminator = CHAR_RIGHT_PARENTHESIS;
5662
5663 /* Come here from the \g<...> and \g'...' code (Oniguruma
5664 compatibility). However, the syntax has been checked to ensure that
5665 the ... are a (signed) number, so that neither ERR63 nor ERR29 will
5666 be called on this path, nor with the jump to OTHER_CHAR_AFTER_QUERY
5667 ever be taken. */
5668
5669 HANDLE_NUMERICAL_RECURSION:
5670
5671 if ((refsign = *ptr) == CHAR_PLUS)
5672 {
5673 ptr++;
5674 if ((digitab[*ptr] & ctype_digit) == 0)
5675 {
5676 *errorcodeptr = ERR63;
5677 goto FAILED;
5678 }
5679 }
5680 else if (refsign == CHAR_MINUS)
5681 {
5682 if ((digitab[ptr[1]] & ctype_digit) == 0)
5683 goto OTHER_CHAR_AFTER_QUERY;
5684 ptr++;
5685 }
5686
5687 recno = 0;
5688 while((digitab[*ptr] & ctype_digit) != 0)
5689 recno = recno * 10 + *ptr++ - CHAR_0;
5690
5691 if (*ptr != terminator)
5692 {
5693 *errorcodeptr = ERR29;
5694 goto FAILED;
5695 }
5696
5697 if (refsign == CHAR_MINUS)
5698 {
5699 if (recno == 0)
5700 {
5701 *errorcodeptr = ERR58;
5702 goto FAILED;
5703 }
5704 recno = cd->bracount - recno + 1;
5705 if (recno <= 0)
5706 {
5707 *errorcodeptr = ERR15;
5708 goto FAILED;
5709 }
5710 }
5711 else if (refsign == CHAR_PLUS)
5712 {
5713 if (recno == 0)
5714 {
5715 *errorcodeptr = ERR58;
5716 goto FAILED;
5717 }
5718 recno += cd->bracount;
5719 }
5720
5721 /* Come here from code above that handles a named recursion */
5722
5723 HANDLE_RECURSION:
5724
5725 previous = code;
5726 called = cd->start_code;
5727
5728 /* When we are actually compiling, find the bracket that is being
5729 referenced. Temporarily end the regex in case it doesn't exist before
5730 this point. If we end up with a forward reference, first check that
5731 the bracket does occur later so we can give the error (and position)
5732 now. Then remember this forward reference in the workspace so it can
5733 be filled in at the end. */
5734
5735 if (lengthptr == NULL)
5736 {
5737 *code = OP_END;
5738 if (recno != 0)
5739 called = _pcre_find_bracket(cd->start_code, utf8, recno);
5740
5741 /* Forward reference */
5742
5743 if (called == NULL)
5744 {
5745 if (find_parens(cd, NULL, recno,
5746 (options & PCRE_EXTENDED) != 0, utf8) < 0)
5747 {
5748 *errorcodeptr = ERR15;
5749 goto FAILED;
5750 }
5751
5752 /* Fudge the value of "called" so that when it is inserted as an
5753 offset below, what it actually inserted is the reference number
5754 of the group. Then remember the forward reference. */
5755
5756 called = cd->start_code + recno;
5757 PUTINC(cd->hwm, 0, (int)(code + 1 - cd->start_code));
5758 }
5759
5760 /* If not a forward reference, and the subpattern is still open,
5761 this is a recursive call. We check to see if this is a left
5762 recursion that could loop for ever, and diagnose that case. */
5763
5764 else if (GET(called, 1) == 0 &&
5765 could_be_empty(called, code, bcptr, utf8, cd))
5766 {
5767 *errorcodeptr = ERR40;
5768 goto FAILED;
5769 }
5770 }
5771
5772 /* Insert the recursion/subroutine item. */
5773
5774 *code = OP_RECURSE;
5775 PUT(code, 1, (int)(called - cd->start_code));
5776 code += 1 + LINK_SIZE;
5777 }
5778
5779 /* Can't determine a first byte now */
5780
5781 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
5782 continue;
5783
5784
5785 /* ------------------------------------------------------------ */
5786 default: /* Other characters: check option setting */
5787 OTHER_CHAR_AFTER_QUERY:
5788 set = unset = 0;
5789 optset = &set;
5790
5791 while (*ptr != CHAR_RIGHT_PARENTHESIS && *ptr != CHAR_COLON)
5792 {
5793 switch (*ptr++)
5794 {
5795 case CHAR_MINUS: optset = &unset; break;
5796
5797 case CHAR_J: /* Record that it changed in the external options */
5798 *optset |= PCRE_DUPNAMES;
5799 cd->external_flags |= PCRE_JCHANGED;
5800 break;
5801
5802 case CHAR_i: *optset |= PCRE_CASELESS; break;
5803 case CHAR_m: *optset |= PCRE_MULTILINE; break;
5804 case CHAR_s: *optset |= PCRE_DOTALL; break;
5805 case CHAR_x: *optset |= PCRE_EXTENDED; break;
5806 case CHAR_U: *optset |= PCRE_UNGREEDY; break;
5807 case CHAR_X: *optset |= PCRE_EXTRA; break;
5808
5809 default: *errorcodeptr = ERR12;
5810 ptr--; /* Correct the offset */
5811 goto FAILED;
5812 }
5813 }
5814
5815 /* Set up the changed option bits, but don't change anything yet. */
5816
5817 newoptions = (options | set) & (~unset);
5818
5819 /* If the options ended with ')' this is not the start of a nested
5820 group with option changes, so the options change at this level. If this
5821 item is right at the start of the pattern, the options can be
5822 abstracted and made external in the pre-compile phase, and ignored in
5823 the compile phase. This can be helpful when matching -- for instance in
5824 caseless checking of required bytes.
5825
5826 If the code pointer is not (cd->start_code + 1 + LINK_SIZE), we are
5827 definitely *not* at the start of the pattern because something has been
5828 compiled. In the pre-compile phase, however, the code pointer can have
5829 that value after the start, because it gets reset as code is discarded
5830 during the pre-compile. However, this can happen only at top level - if
5831 we are within parentheses, the starting BRA will still be present. At
5832 any parenthesis level, the length value can be used to test if anything
5833 has been compiled at that level. Thus, a test for both these conditions
5834 is necessary to ensure we correctly detect the start of the pattern in
5835 both phases.
5836
5837 If we are not at the pattern start, reset the greedy defaults and the
5838 case value for firstbyte and reqbyte. */
5839
5840 if (*ptr == CHAR_RIGHT_PARENTHESIS)
5841 {
5842 if (code == cd->start_code + 1 + LINK_SIZE &&
5843 (lengthptr == NULL || *lengthptr == 2 + 2*LINK_SIZE))
5844 {
5845 cd->external_options = newoptions;
5846 }
5847 else
5848 {
5849 greedy_default = ((newoptions & PCRE_UNGREEDY) != 0);
5850 greedy_non_default = greedy_default ^ 1;
5851 req_caseopt = ((newoptions & PCRE_CASELESS) != 0)? REQ_CASELESS : 0;
5852 }
5853
5854 /* Change options at this level, and pass them back for use
5855 in subsequent branches. */
5856
5857 *optionsptr = options = newoptions;
5858 previous = NULL; /* This item can't be repeated */
5859 continue; /* It is complete */
5860 }
5861
5862 /* If the options ended with ':' we are heading into a nested group
5863 with possible change of options. Such groups are non-capturing and are
5864 not assertions of any kind. All we need to do is skip over the ':';
5865 the newoptions value is handled below. */
5866
5867 bravalue = OP_BRA;
5868 ptr++;
5869 } /* End of switch for character following (? */
5870 } /* End of (? handling */
5871
5872 /* Opening parenthesis not followed by '*' or '?'. If PCRE_NO_AUTO_CAPTURE
5873 is set, all unadorned brackets become non-capturing and behave like (?:...)
5874 brackets. */
5875
5876 else if ((options & PCRE_NO_AUTO_CAPTURE) != 0)
5877 {
5878 bravalue = OP_BRA;
5879 }
5880
5881 /* Else we have a capturing group. */
5882
5883 else
5884 {
5885 NUMBERED_GROUP:
5886 cd->bracount += 1;
5887 PUT2(code, 1+LINK_SIZE, cd->bracount);
5888 skipbytes = 2;
5889 }
5890
5891 /* Process nested bracketed regex. Assertions used not to be repeatable,
5892 but this was changed for Perl compatibility, so all kinds can now be
5893 repeated. We copy code into a non-register variable (tempcode) in order to
5894 be able to pass its address because some compilers complain otherwise. */
5895
5896 previous = code; /* For handling repetition */
5897 *code = bravalue;
5898 tempcode = code;
5899 tempreqvary = cd->req_varyopt; /* Save value before bracket */
5900 length_prevgroup = 0; /* Initialize for pre-compile phase */
5901
5902 if (!compile_regex(
5903 newoptions, /* The complete new option state */
5904 &tempcode, /* Where to put code (updated) */
5905 &ptr, /* Input pointer (updated) */
5906 errorcodeptr, /* Where to put an error message */
5907 (bravalue == OP_ASSERTBACK ||
5908 bravalue == OP_ASSERTBACK_NOT), /* TRUE if back assert */
5909 reset_bracount, /* True if (?| group */
5910 skipbytes, /* Skip over bracket number */
5911 &subfirstbyte, /* For possible first char */
5912 &subreqbyte, /* For possible last char */
5913 bcptr, /* Current branch chain */
5914 cd, /* Tables block */
5915 (lengthptr == NULL)? NULL : /* Actual compile phase */
5916 &length_prevgroup /* Pre-compile phase */
5917 ))
5918 goto FAILED;
5919
5920 if (bravalue >= OP_ASSERT && bravalue <= OP_ASSERTBACK_NOT)
5921 cd->assert_depth -= 1;
5922
5923 /* At the end of compiling, code is still pointing to the start of the
5924 group, while tempcode has been updated to point past the end of the group
5925 and any option resetting that may follow it. The pattern pointer (ptr)
5926 is on the bracket. */
5927
5928 /* If this is a conditional bracket, check that there are no more than
5929 two branches in the group, or just one if it's a DEFINE group. We do this
5930 in the real compile phase, not in the pre-pass, where the whole group may
5931 not be available. */
5932
5933 if (bravalue == OP_COND && lengthptr == NULL)
5934 {
5935 uschar *tc = code;
5936 int condcount = 0;
5937
5938 do {
5939 condcount++;
5940 tc += GET(tc,1);
5941 }
5942 while (*tc != OP_KET);
5943
5944 /* A DEFINE group is never obeyed inline (the "condition" is always
5945 false). It must have only one branch. */
5946
5947 if (code[LINK_SIZE+1] == OP_DEF)
5948 {
5949 if (condcount > 1)
5950 {
5951 *errorcodeptr = ERR54;
5952 goto FAILED;
5953 }
5954 bravalue = OP_DEF; /* Just a flag to suppress char handling below */
5955 }
5956
5957 /* A "normal" conditional group. If there is just one branch, we must not
5958 make use of its firstbyte or reqbyte, because this is equivalent to an
5959 empty second branch. */
5960
5961 else
5962 {
5963 if (condcount > 2)
5964 {
5965 *errorcodeptr = ERR27;
5966 goto FAILED;
5967 }
5968 if (condcount == 1) subfirstbyte = subreqbyte = REQ_NONE;
5969 }
5970 }
5971
5972 /* Error if hit end of pattern */
5973
5974 if (*ptr != CHAR_RIGHT_PARENTHESIS)
5975 {
5976 *errorcodeptr = ERR14;
5977 goto FAILED;
5978 }
5979
5980 /* In the pre-compile phase, update the length by the length of the group,
5981 less the brackets at either end. Then reduce the compiled code to just a
5982 set of non-capturing brackets so that it doesn't use much memory if it is
5983 duplicated by a quantifier.*/
5984
5985 if (lengthptr != NULL)
5986 {
5987 if (OFLOW_MAX - *lengthptr < length_prevgroup - 2 - 2*LINK_SIZE)
5988 {
5989 *errorcodeptr = ERR20;
5990 goto FAILED;
5991 }
5992 *lengthptr += length_prevgroup - 2 - 2*LINK_SIZE;
5993 code++; /* This already contains bravalue */
5994 PUTINC(code, 0, 1 + LINK_SIZE);
5995 *code++ = OP_KET;
5996 PUTINC(code, 0, 1 + LINK_SIZE);
5997 break; /* No need to waste time with special character handling */
5998 }
5999
6000 /* Otherwise update the main code pointer to the end of the group. */
6001
6002 code = tempcode;
6003
6004 /* For a DEFINE group, required and first character settings are not
6005 relevant. */
6006
6007 if (bravalue == OP_DEF) break;
6008
6009 /* Handle updating of the required and first characters for other types of
6010 group. Update for normal brackets of all kinds, and conditions with two
6011 branches (see code above). If the bracket is followed by a quantifier with
6012 zero repeat, we have to back off. Hence the definition of zeroreqbyte and
6013 zerofirstbyte outside the main loop so that they can be accessed for the
6014 back off. */
6015
6016 zeroreqbyte = reqbyte;
6017 zerofirstbyte = firstbyte;
6018 groupsetfirstbyte = FALSE;
6019
6020 if (bravalue >= OP_ONCE)
6021 {
6022 /* If we have not yet set a firstbyte in this branch, take it from the
6023 subpattern, remembering that it was set here so that a repeat of more
6024 than one can replicate it as reqbyte if necessary. If the subpattern has
6025 no firstbyte, set "none" for the whole branch. In both cases, a zero
6026 repeat forces firstbyte to "none". */
6027
6028 if (firstbyte == REQ_UNSET)
6029 {
6030 if (subfirstbyte >= 0)
6031 {
6032 firstbyte = subfirstbyte;
6033 groupsetfirstbyte = TRUE;
6034 }
6035 else firstbyte = REQ_NONE;
6036 zerofirstbyte = REQ_NONE;
6037 }
6038
6039 /* If firstbyte was previously set, convert the subpattern's firstbyte
6040 into reqbyte if there wasn't one, using the vary flag that was in
6041 existence beforehand. */
6042
6043 else if (subfirstbyte >= 0 && subreqbyte < 0)
6044 subreqbyte = subfirstbyte | tempreqvary;
6045
6046 /* If the subpattern set a required byte (or set a first byte that isn't
6047 really the first byte - see above), set it. */
6048
6049 if (subreqbyte >= 0) reqbyte = subreqbyte;
6050 }
6051
6052 /* For a forward assertion, we take the reqbyte, if set. This can be
6053 helpful if the pattern that follows the assertion doesn't set a different
6054 char. For example, it's useful for /(?=abcde).+/. We can't set firstbyte
6055 for an assertion, however because it leads to incorrect effect for patterns
6056 such as /(?=a)a.+/ when the "real" "a" would then become a reqbyte instead
6057 of a firstbyte. This is overcome by a scan at the end if there's no
6058 firstbyte, looking for an asserted first char. */
6059
6060 else if (bravalue == OP_ASSERT && subreqbyte >= 0) reqbyte = subreqbyte;
6061 break; /* End of processing '(' */
6062
6063
6064 /* ===================================================================*/
6065 /* Handle metasequences introduced by \. For ones like \d, the ESC_ values
6066 are arranged to be the negation of the corresponding OP_values in the
6067 default case when PCRE_UCP is not set. For the back references, the values
6068 are ESC_REF plus the reference number. Only back references and those types
6069 that consume a character may be repeated. We can test for values between
6070 ESC_b and ESC_Z for the latter; this may have to change if any new ones are
6071 ever created. */
6072
6073 case CHAR_BACKSLASH:
6074 tempptr = ptr;
6075 c = check_escape(&ptr, errorcodeptr, cd->bracount, options, FALSE);
6076 if (*errorcodeptr != 0) goto FAILED;
6077
6078 if (c < 0)
6079 {
6080 if (-c == ESC_Q) /* Handle start of quoted string */
6081 {
6082 if (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
6083 ptr += 2; /* avoid empty string */
6084 else inescq = TRUE;
6085 continue;
6086 }
6087
6088 if (-c == ESC_E) continue; /* Perl ignores an orphan \E */
6089
6090 /* For metasequences that actually match a character, we disable the
6091 setting of a first character if it hasn't already been set. */
6092
6093 if (firstbyte == REQ_UNSET && -c > ESC_b && -c < ESC_Z)
6094 firstbyte = REQ_NONE;
6095
6096 /* Set values to reset to if this is followed by a zero repeat. */
6097
6098 zerofirstbyte = firstbyte;
6099 zeroreqbyte = reqbyte;
6100
6101 /* \g<name> or \g'name' is a subroutine call by name and \g<n> or \g'n'
6102 is a subroutine call by number (Oniguruma syntax). In fact, the value
6103 -ESC_g is returned only for these cases. So we don't need to check for <
6104 or ' if the value is -ESC_g. For the Perl syntax \g{n} the value is
6105 -ESC_REF+n, and for the Perl syntax \g{name} the result is -ESC_k (as
6106 that is a synonym for a named back reference). */
6107
6108 if (-c == ESC_g)
6109 {
6110 const uschar *p;
6111 save_hwm = cd->hwm; /* Normally this is set when '(' is read */
6112 terminator = (*(++ptr) == CHAR_LESS_THAN_SIGN)?
6113 CHAR_GREATER_THAN_SIGN : CHAR_APOSTROPHE;
6114
6115 /* These two statements stop the compiler for warning about possibly
6116 unset variables caused by the jump to HANDLE_NUMERICAL_RECURSION. In
6117 fact, because we actually check for a number below, the paths that
6118 would actually be in error are never taken. */
6119
6120 skipbytes = 0;
6121 reset_bracount = FALSE;
6122
6123 /* Test for a name */
6124
6125 if (ptr[1] != CHAR_PLUS && ptr[1] != CHAR_MINUS)
6126 {
6127 BOOL isnumber = TRUE;
6128 for (p = ptr + 1; *p != 0 && *p != terminator; p++)
6129 {
6130 if ((cd->ctypes[*p] & ctype_digit) == 0) isnumber = FALSE;
6131 if ((cd->ctypes[*p] & ctype_word) == 0) break;
6132 }
6133 if (*p != terminator)
6134 {
6135 *errorcodeptr = ERR57;
6136 break;
6137 }
6138 if (isnumber)
6139 {
6140 ptr++;
6141 goto HANDLE_NUMERICAL_RECURSION;
6142 }
6143 is_recurse = TRUE;
6144 goto NAMED_REF_OR_RECURSE;
6145 }
6146
6147 /* Test a signed number in angle brackets or quotes. */
6148
6149 p = ptr + 2;
6150 while ((digitab[*p] & ctype_digit) != 0) p++;
6151 if (*p != terminator)
6152 {
6153 *errorcodeptr = ERR57;
6154 break;
6155 }
6156 ptr++;
6157 goto HANDLE_NUMERICAL_RECURSION;
6158 }
6159
6160 /* \k<name> or \k'name' is a back reference by name (Perl syntax).
6161 We also support \k{name} (.NET syntax). */
6162
6163 if (-c == ESC_k)
6164 {
6165 if ((ptr[1] != CHAR_LESS_THAN_SIGN &&
6166 ptr[1] != CHAR_APOSTROPHE && ptr[1] != CHAR_LEFT_CURLY_BRACKET))
6167 {
6168 *errorcodeptr = ERR69;
6169 break;
6170 }
6171 is_recurse = FALSE;
6172 terminator = (*(++ptr) == CHAR_LESS_THAN_SIGN)?
6173 CHAR_GREATER_THAN_SIGN : (*ptr == CHAR_APOSTROPHE)?
6174 CHAR_APOSTROPHE : CHAR_RIGHT_CURLY_BRACKET;
6175 goto NAMED_REF_OR_RECURSE;
6176 }
6177
6178 /* Back references are handled specially; must disable firstbyte if
6179 not set to cope with cases like (?=(\w+))\1: which would otherwise set
6180 ':' later. */
6181
6182 if (-c >= ESC_REF)
6183 {
6184 open_capitem *oc;
6185 recno = -c - ESC_REF;
6186
6187 HANDLE_REFERENCE: /* Come here from named backref handling */
6188 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
6189 previous = code;
6190 *code++ = ((options & PCRE_CASELESS) != 0)? OP_REFI : OP_REF;
6191 PUT2INC(code, 0, recno);
6192 cd->backref_map |= (recno < 32)? (1 << recno) : 1;
6193 if (recno > cd->top_backref) cd->top_backref = recno;
6194
6195 /* Check to see if this back reference is recursive, that it, it
6196 is inside the group that it references. A flag is set so that the
6197 group can be made atomic. */
6198
6199 for (oc = cd->open_caps; oc != NULL; oc = oc->next)
6200 {
6201 if (oc->number == recno)
6202 {
6203 oc->flag = TRUE;
6204 break;
6205 }
6206 }
6207 }
6208
6209 /* So are Unicode property matches, if supported. */
6210
6211 #ifdef SUPPORT_UCP
6212 else if (-c == ESC_P || -c == ESC_p)
6213 {
6214 BOOL negated;
6215 int pdata;
6216 int ptype = get_ucp(&ptr, &negated, &pdata, errorcodeptr);
6217 if (ptype < 0) goto FAILED;
6218 previous = code;
6219 *code++ = ((-c == ESC_p) != negated)? OP_PROP : OP_NOTPROP;
6220 *code++ = ptype;
6221 *code++ = pdata;
6222 }
6223 #else
6224
6225 /* If Unicode properties are not supported, \X, \P, and \p are not
6226 allowed. */
6227
6228 else if (-c == ESC_X || -c == ESC_P || -c == ESC_p)
6229 {
6230 *errorcodeptr = ERR45;
6231 goto FAILED;
6232 }
6233 #endif
6234
6235 /* For the rest (including \X when Unicode properties are supported), we
6236 can obtain the OP value by negating the escape value in the default
6237 situation when PCRE_UCP is not set. When it *is* set, we substitute
6238 Unicode property tests. */
6239
6240 else
6241 {
6242 #ifdef SUPPORT_UCP
6243 if (-c >= ESC_DU && -c <= ESC_wu)
6244 {
6245 nestptr = ptr + 1; /* Where to resume */
6246 ptr = substitutes[-c - ESC_DU] - 1; /* Just before substitute */
6247 }
6248 else
6249 #endif
6250 {
6251 previous = (-c > ESC_b && -c < ESC_Z)? code : NULL;
6252 *code++ = -c;
6253 }
6254 }
6255 continue;
6256 }
6257
6258 /* We have a data character whose value is in c. In UTF-8 mode it may have
6259 a value > 127. We set its representation in the length/buffer, and then
6260 handle it as a data character. */
6261
6262 #ifdef SUPPORT_UTF8
6263 if (utf8 && c > 127)
6264 mclength = _pcre_ord2utf8(c, mcbuffer);
6265 else
6266 #endif
6267
6268 {
6269 mcbuffer[0] = c;
6270 mclength = 1;
6271 }
6272 goto ONE_CHAR;
6273
6274
6275 /* ===================================================================*/
6276 /* Handle a literal character. It is guaranteed not to be whitespace or #
6277 when the extended flag is set. If we are in UTF-8 mode, it may be a
6278 multi-byte literal character. */
6279
6280 default:
6281 NORMAL_CHAR:
6282 mclength = 1;
6283 mcbuffer[0] = c;
6284
6285 #ifdef SUPPORT_UTF8
6286 if (utf8 && c >= 0xc0)
6287 {
6288 while ((ptr[1] & 0xc0) == 0x80)
6289 mcbuffer[mclength++] = *(++ptr);
6290 }
6291 #endif
6292
6293 /* At this point we have the character's bytes in mcbuffer, and the length
6294 in mclength. When not in UTF-8 mode, the length is always 1. */
6295
6296 ONE_CHAR:
6297 previous = code;
6298 *code++ = ((options & PCRE_CASELESS) != 0)? OP_CHARI : OP_CHAR;
6299 for (c = 0; c < mclength; c++) *code++ = mcbuffer[c];
6300
6301 /* Remember if \r or \n were seen */
6302
6303 if (mcbuffer[0] == CHAR_CR || mcbuffer[0] == CHAR_NL)
6304 cd->external_flags |= PCRE_HASCRORLF;
6305
6306 /* Set the first and required bytes appropriately. If no previous first
6307 byte, set it from this character, but revert to none on a zero repeat.
6308 Otherwise, leave the firstbyte value alone, and don't change it on a zero
6309 repeat. */
6310
6311 if (firstbyte == REQ_UNSET)
6312 {
6313 zerofirstbyte = REQ_NONE;
6314 zeroreqbyte = reqbyte;
6315
6316 /* If the character is more than one byte long, we can set firstbyte
6317 only if it is not to be matched caselessly. */
6318
6319 if (mclength == 1 || req_caseopt == 0)
6320 {
6321 firstbyte = mcbuffer[0] | req_caseopt;
6322 if (mclength != 1) reqbyte = code[-1] | cd->req_varyopt;
6323 }
6324 else firstbyte = reqbyte = REQ_NONE;
6325 }
6326
6327 /* firstbyte was previously set; we can set reqbyte only the length is
6328 1 or the matching is caseful. */
6329
6330 else
6331 {
6332 zerofirstbyte = firstbyte;
6333 zeroreqbyte = reqbyte;
6334 if (mclength == 1 || req_caseopt == 0)
6335 reqbyte = code[-1] | req_caseopt | cd->req_varyopt;
6336 }
6337
6338 break; /* End of literal character handling */
6339 }
6340 } /* end of big loop */
6341
6342
6343 /* Control never reaches here by falling through, only by a goto for all the
6344 error states. Pass back the position in the pattern so that it can be displayed
6345 to the user for diagnosing the error. */
6346
6347 FAILED:
6348 *ptrptr = ptr;
6349 return FALSE;
6350 }
6351
6352
6353
6354
6355 /*************************************************
6356 * Compile sequence of alternatives *
6357 *************************************************/
6358
6359 /* On entry, ptr is pointing past the bracket character, but on return it
6360 points to the closing bracket, or vertical bar, or end of string. The code
6361 variable is pointing at the byte into which the BRA operator has been stored.
6362 This function is used during the pre-compile phase when we are trying to find
6363 out the amount of memory needed, as well as during the real compile phase. The
6364 value of lengthptr distinguishes the two phases.
6365
6366 Arguments:
6367 options option bits, including any changes for this subpattern
6368 codeptr -> the address of the current code pointer
6369 ptrptr -> the address of the current pattern pointer
6370 errorcodeptr -> pointer to error code variable
6371 lookbehind TRUE if this is a lookbehind assertion
6372 reset_bracount TRUE to reset the count for each branch
6373 skipbytes skip this many bytes at start (for brackets and OP_COND)
6374 firstbyteptr place to put the first required character, or a negative number
6375 reqbyteptr place to put the last required character, or a negative number
6376 bcptr pointer to the chain of currently open branches
6377 cd points to the data block with tables pointers etc.
6378 lengthptr NULL during the real compile phase
6379 points to length accumulator during pre-compile phase
6380
6381 Returns: TRUE on success
6382 */
6383
6384 static BOOL
6385 compile_regex(int options, uschar **codeptr, const uschar **ptrptr,
6386 int *errorcodeptr, BOOL lookbehind, BOOL reset_bracount, int skipbytes,
6387 int *firstbyteptr, int *reqbyteptr, branch_chain *bcptr, compile_data *cd,
6388 int *lengthptr)
6389 {
6390 const uschar *ptr = *ptrptr;
6391 uschar *code = *codeptr;
6392 uschar *last_branch = code;
6393 uschar *start_bracket = code;
6394 uschar *reverse_count = NULL;
6395 open_capitem capitem;
6396 int capnumber = 0;
6397 int firstbyte, reqbyte;
6398 int branchfirstbyte, branchreqbyte;
6399 int length;
6400 int orig_bracount;
6401 int max_bracount;
6402 branch_chain bc;
6403
6404 bc.outer = bcptr;
6405 bc.current_branch = code;
6406
6407 firstbyte = reqbyte = REQ_UNSET;
6408
6409 /* Accumulate the length for use in the pre-compile phase. Start with the
6410 length of the BRA and KET and any extra bytes that are required at the
6411 beginning. We accumulate in a local variable to save frequent testing of
6412 lenthptr for NULL. We cannot do this by looking at the value of code at the
6413 start and end of each alternative, because compiled items are discarded during
6414 the pre-compile phase so that the work space is not exceeded. */
6415
6416 length = 2 + 2*LINK_SIZE + skipbytes;
6417
6418 /* WARNING: If the above line is changed for any reason, you must also change
6419 the code that abstracts option settings at the start of the pattern and makes
6420 them global. It tests the value of length for (2 + 2*LINK_SIZE) in the
6421 pre-compile phase to find out whether anything has yet been compiled or not. */
6422
6423 /* If this is a capturing subpattern, add to the chain of open capturing items
6424 so that we can detect them if (*ACCEPT) is encountered. This is also used to
6425 detect groups that contain recursive back references to themselves. Note that
6426 only OP_CBRA need be tested here; changing this opcode to one of its variants,
6427 e.g. OP_SCBRAPOS, happens later, after the group has been compiled. */
6428
6429 if (*code == OP_CBRA)
6430 {
6431 capnumber = GET2(code, 1 + LINK_SIZE);
6432 capitem.number = capnumber;
6433 capitem.next = cd->open_caps;
6434 capitem.flag = FALSE;
6435 cd->open_caps = &capitem;
6436 }
6437
6438 /* Offset is set zero to mark that this bracket is still open */
6439
6440 PUT(code, 1, 0);
6441 code += 1 + LINK_SIZE + skipbytes;
6442
6443 /* Loop for each alternative branch */
6444
6445 orig_bracount = max_bracount = cd->bracount;
6446 for (;;)
6447 {
6448 /* For a (?| group, reset the capturing bracket count so that each branch
6449 uses the same numbers. */
6450
6451 if (reset_bracount) cd->bracount = orig_bracount;
6452
6453 /* Set up dummy OP_REVERSE if lookbehind assertion */
6454
6455 if (lookbehind)
6456 {
6457 *code++ = OP_REVERSE;
6458 reverse_count = code;
6459 PUTINC(code, 0, 0);
6460 length += 1 + LINK_SIZE;
6461 }
6462
6463 /* Now compile the branch; in the pre-compile phase its length gets added
6464 into the length. */
6465
6466 if (!compile_branch(&options, &code, &ptr, errorcodeptr, &branchfirstbyte,
6467 &branchreqbyte, &bc, cd, (lengthptr == NULL)? NULL : &length))
6468 {
6469 *ptrptr = ptr;
6470 return FALSE;
6471 }
6472
6473 /* Keep the highest bracket count in case (?| was used and some branch
6474 has fewer than the rest. */
6475
6476 if (cd->bracount > max_bracount) max_bracount = cd->bracount;
6477
6478 /* In the real compile phase, there is some post-processing to be done. */
6479
6480 if (lengthptr == NULL)
6481 {
6482 /* If this is the first branch, the firstbyte and reqbyte values for the
6483 branch become the values for the regex. */
6484
6485 if (*last_branch != OP_ALT)
6486 {
6487 firstbyte = branchfirstbyte;
6488 reqbyte = branchreqbyte;
6489 }
6490
6491 /* If this is not the first branch, the first char and reqbyte have to
6492 match the values from all the previous branches, except that if the
6493 previous value for reqbyte didn't have REQ_VARY set, it can still match,
6494 and we set REQ_VARY for the regex. */
6495
6496 else
6497 {
6498 /* If we previously had a firstbyte, but it doesn't match the new branch,
6499 we have to abandon the firstbyte for the regex, but if there was
6500 previously no reqbyte, it takes on the value of the old firstbyte. */
6501
6502 if (firstbyte >= 0 && firstbyte != branchfirstbyte)
6503 {
6504 if (reqbyte < 0) reqbyte = firstbyte;
6505 firstbyte = REQ_NONE;
6506 }
6507
6508 /* If we (now or from before) have no firstbyte, a firstbyte from the
6509 branch becomes a reqbyte if there isn't a branch reqbyte. */
6510
6511 if (firstbyte < 0 && branchfirstbyte >= 0 && branchreqbyte < 0)
6512 branchreqbyte = branchfirstbyte;
6513
6514 /* Now ensure that the reqbytes match */
6515
6516 if ((reqbyte & ~REQ_VARY) != (branchreqbyte & ~REQ_VARY))
6517 reqbyte = REQ_NONE;
6518 else reqbyte |= branchreqbyte; /* To "or" REQ_VARY */
6519 }
6520
6521 /* If lookbehind, check that this branch matches a fixed-length string, and
6522 put the length into the OP_REVERSE item. Temporarily mark the end of the
6523 branch with OP_END. If the branch contains OP_RECURSE, the result is -3
6524 because there may be forward references that we can't check here. Set a
6525 flag to cause another lookbehind check at the end. Why not do it all at the
6526 end? Because common, erroneous checks are picked up here and the offset of
6527 the problem can be shown. */
6528
6529 if (lookbehind)
6530 {
6531 int fixed_length;
6532 *code = OP_END;
6533 fixed_length = find_fixedlength(last_branch, (options & PCRE_UTF8) != 0,
6534 FALSE, cd);
6535 DPRINTF(("fixed length = %d\n", fixed_length));
6536 if (fixed_length == -3)
6537 {
6538 cd->check_lookbehind = TRUE;
6539 }
6540 else if (fixed_length < 0)
6541 {
6542 *errorcodeptr = (fixed_length == -2)? ERR36 : ERR25;
6543 *ptrptr = ptr;
6544 return FALSE;
6545 }
6546 else { PUT(reverse_count, 0, fixed_length); }
6547 }
6548 }
6549
6550 /* Reached end of expression, either ')' or end of pattern. In the real
6551 compile phase, go back through the alternative branches and reverse the chain
6552 of offsets, with the field in the BRA item now becoming an offset to the
6553 first alternative. If there are no alternatives, it points to the end of the
6554 group. The length in the terminating ket is always the length of the whole
6555 bracketed item. Return leaving the pointer at the terminating char. */
6556
6557 if (*ptr != CHAR_VERTICAL_LINE)
6558 {
6559 if (lengthptr == NULL)
6560 {
6561 int branch_length = (int)(code - last_branch);
6562 do
6563 {
6564 int prev_length = GET(last_branch, 1);
6565 PUT(last_branch, 1, branch_length);
6566 branch_length = prev_length;
6567 last_branch -= branch_length;
6568 }
6569 while (branch_length > 0);
6570 }
6571
6572 /* Fill in the ket */
6573
6574 *code = OP_KET;
6575 PUT(code, 1, (int)(code - start_bracket));
6576 code += 1 + LINK_SIZE;
6577
6578 /* If it was a capturing subpattern, check to see if it contained any
6579 recursive back references. If so, we must wrap it in atomic brackets.
6580 In any event, remove the block from the chain. */
6581
6582 if (capnumber > 0)
6583 {
6584 if (cd->open_caps->flag)
6585 {
6586 memmove(start_bracket + 1 + LINK_SIZE, start_bracket,
6587 code - start_bracket);
6588 *start_bracket = OP_ONCE;
6589 code += 1 + LINK_SIZE;
6590 PUT(start_bracket, 1, (int)(code - start_bracket));
6591 *code = OP_KET;
6592 PUT(code, 1, (int)(code - start_bracket));
6593 code += 1 + LINK_SIZE;
6594 length += 2 + 2*LINK_SIZE;
6595 }
6596 cd->open_caps = cd->open_caps->next;
6597 }
6598
6599 /* Retain the highest bracket number, in case resetting was used. */
6600
6601 cd->bracount = max_bracount;
6602
6603 /* Set values to pass back */
6604
6605 *codeptr = code;
6606 *ptrptr = ptr;
6607 *firstbyteptr = firstbyte;
6608 *reqbyteptr = reqbyte;
6609 if (lengthptr != NULL)
6610 {
6611 if (OFLOW_MAX - *lengthptr < length)
6612 {
6613 *errorcodeptr = ERR20;
6614 return FALSE;
6615 }
6616 *lengthptr += length;
6617 }
6618 return TRUE;
6619 }
6620
6621 /* Another branch follows. In the pre-compile phase, we can move the code
6622 pointer back to where it was for the start of the first branch. (That is,
6623 pretend that each branch is the only one.)
6624
6625 In the real compile phase, insert an ALT node. Its length field points back
6626 to the previous branch while the bracket remains open. At the end the chain
6627 is reversed. It's done like this so that the start of the bracket has a
6628 zero offset until it is closed, making it possible to detect recursion. */
6629
6630 if (lengthptr != NULL)
6631 {
6632 code = *codeptr + 1 + LINK_SIZE + skipbytes;
6633 length += 1 + LINK_SIZE;
6634 }
6635 else