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

Contents of /code/trunk/pcre_compile.c

Parent Directory Parent Directory | Revision Log Revision Log


Revision 624 - (show annotations) (download)
Tue Jul 19 10:43:28 2011 UTC (3 years, 1 month ago) by ph10
File MIME type: text/plain
File size: 240983 byte(s)
Fix unset variable bug introduced during recursion refactoring.

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"
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 ;
413
414 /* Table to identify digits and hex digits. This is used when compiling
415 patterns. Note that the tables in chartables are dependent on the locale, and
416 may mark arbitrary characters as digits - but the PCRE compiling code expects
417 to handle only 0-9, a-z, and A-Z as digits when compiling. That is why we have
418 a private table here. It costs 256 bytes, but it is a lot faster than doing
419 character value tests (at least in some simple cases I timed), and in some
420 applications one wants PCRE to compile efficiently as well as match
421 efficiently.
422
423 For convenience, we use the same bit definitions as in chartables:
424
425 0x04 decimal digit
426 0x08 hexadecimal digit
427
428 Then we can use ctype_digit and ctype_xdigit in the code. */
429
430 #ifndef EBCDIC
431
432 /* This is the "normal" case, for ASCII systems, and EBCDIC systems running in
433 UTF-8 mode. */
434
435 static const unsigned char digitab[] =
436 {
437 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 0- 7 */
438 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 8- 15 */
439 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 16- 23 */
440 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 24- 31 */
441 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - ' */
442 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* ( - / */
443 0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c, /* 0 - 7 */
444 0x0c,0x0c,0x00,0x00,0x00,0x00,0x00,0x00, /* 8 - ? */
445 0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* @ - G */
446 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* H - O */
447 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* P - W */
448 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* X - _ */
449 0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* ` - g */
450 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* h - o */
451 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* p - w */
452 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* x -127 */
453 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 128-135 */
454 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 136-143 */
455 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 144-151 */
456 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 152-159 */
457 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 160-167 */
458 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 168-175 */
459 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 176-183 */
460 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 184-191 */
461 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 192-199 */
462 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 200-207 */
463 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 208-215 */
464 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 216-223 */
465 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 224-231 */
466 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 232-239 */
467 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 240-247 */
468 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00};/* 248-255 */
469
470 #else
471
472 /* This is the "abnormal" case, for EBCDIC systems not running in UTF-8 mode. */
473
474 static const unsigned char digitab[] =
475 {
476 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 0- 7 0 */
477 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 8- 15 */
478 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 16- 23 10 */
479 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 24- 31 */
480 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 32- 39 20 */
481 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 40- 47 */
482 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 48- 55 30 */
483 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 56- 63 */
484 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - 71 40 */
485 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 72- | */
486 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* & - 87 50 */
487 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 88- 95 */
488 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - -103 60 */
489 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 104- ? */
490 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 112-119 70 */
491 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 120- " */
492 0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* 128- g 80 */
493 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* h -143 */
494 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 144- p 90 */
495 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* q -159 */
496 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 160- x A0 */
497 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* y -175 */
498 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* ^ -183 B0 */
499 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 184-191 */
500 0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* { - G C0 */
501 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* H -207 */
502 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* } - P D0 */
503 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* Q -223 */
504 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* \ - X E0 */
505 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* Y -239 */
506 0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c, /* 0 - 7 F0 */
507 0x0c,0x0c,0x00,0x00,0x00,0x00,0x00,0x00};/* 8 -255 */
508
509 static const unsigned char ebcdic_chartab[] = { /* chartable partial dup */
510 0x80,0x00,0x00,0x00,0x00,0x01,0x00,0x00, /* 0- 7 */
511 0x00,0x00,0x00,0x00,0x01,0x01,0x00,0x00, /* 8- 15 */
512 0x00,0x00,0x00,0x00,0x00,0x01,0x00,0x00, /* 16- 23 */
513 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 24- 31 */
514 0x00,0x00,0x00,0x00,0x00,0x01,0x00,0x00, /* 32- 39 */
515 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 40- 47 */
516 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 48- 55 */
517 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 56- 63 */
518 0x01,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - 71 */
519 0x00,0x00,0x00,0x80,0x00,0x80,0x80,0x80, /* 72- | */
520 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* & - 87 */
521 0x00,0x00,0x00,0x80,0x80,0x80,0x00,0x00, /* 88- 95 */
522 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - -103 */
523 0x00,0x00,0x00,0x00,0x00,0x10,0x00,0x80, /* 104- ? */
524 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 112-119 */
525 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 120- " */
526 0x00,0x1a,0x1a,0x1a,0x1a,0x1a,0x1a,0x12, /* 128- g */
527 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* h -143 */
528 0x00,0x12,0x12,0x12,0x12,0x12,0x12,0x12, /* 144- p */
529 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* q -159 */
530 0x00,0x00,0x12,0x12,0x12,0x12,0x12,0x12, /* 160- x */
531 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* y -175 */
532 0x80,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* ^ -183 */
533 0x00,0x00,0x80,0x00,0x00,0x00,0x00,0x00, /* 184-191 */
534 0x80,0x1a,0x1a,0x1a,0x1a,0x1a,0x1a,0x12, /* { - G */
535 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* H -207 */
536 0x00,0x12,0x12,0x12,0x12,0x12,0x12,0x12, /* } - P */
537 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* Q -223 */
538 0x00,0x00,0x12,0x12,0x12,0x12,0x12,0x12, /* \ - X */
539 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* Y -239 */
540 0x1c,0x1c,0x1c,0x1c,0x1c,0x1c,0x1c,0x1c, /* 0 - 7 */
541 0x1c,0x1c,0x00,0x00,0x00,0x00,0x00,0x00};/* 8 -255 */
542 #endif
543
544
545 /* Definition to allow mutual recursion */
546
547 static BOOL
548 compile_regex(int, uschar **, const uschar **, int *, BOOL, BOOL, int, int *,
549 int *, branch_chain *, compile_data *, int *);
550
551
552
553 /*************************************************
554 * Find an error text *
555 *************************************************/
556
557 /* The error texts are now all in one long string, to save on relocations. As
558 some of the text is of unknown length, we can't use a table of offsets.
559 Instead, just count through the strings. This is not a performance issue
560 because it happens only when there has been a compilation error.
561
562 Argument: the error number
563 Returns: pointer to the error string
564 */
565
566 static const char *
567 find_error_text(int n)
568 {
569 const char *s = error_texts;
570 for (; n > 0; n--)
571 {
572 while (*s++ != 0) {};
573 if (*s == 0) return "Error text not found (please report)";
574 }
575 return s;
576 }
577
578
579 /*************************************************
580 * Handle escapes *
581 *************************************************/
582
583 /* This function is called when a \ has been encountered. It either returns a
584 positive value for a simple escape such as \n, or a negative value which
585 encodes one of the more complicated things such as \d. A backreference to group
586 n is returned as -(ESC_REF + n); ESC_REF is the highest ESC_xxx macro. When
587 UTF-8 is enabled, a positive value greater than 255 may be returned. On entry,
588 ptr is pointing at the \. On exit, it is on the final character of the escape
589 sequence.
590
591 Arguments:
592 ptrptr points to the pattern position pointer
593 errorcodeptr points to the errorcode variable
594 bracount number of previous extracting brackets
595 options the options bits
596 isclass TRUE if inside a character class
597
598 Returns: zero or positive => a data character
599 negative => a special escape sequence
600 on error, errorcodeptr is set
601 */
602
603 static int
604 check_escape(const uschar **ptrptr, int *errorcodeptr, int bracount,
605 int options, BOOL isclass)
606 {
607 BOOL utf8 = (options & PCRE_UTF8) != 0;
608 const uschar *ptr = *ptrptr + 1;
609 int c, i;
610
611 GETCHARINCTEST(c, ptr); /* Get character value, increment pointer */
612 ptr--; /* Set pointer back to the last byte */
613
614 /* If backslash is at the end of the pattern, it's an error. */
615
616 if (c == 0) *errorcodeptr = ERR1;
617
618 /* Non-alphanumerics are literals. For digits or letters, do an initial lookup
619 in a table. A non-zero result is something that can be returned immediately.
620 Otherwise further processing may be required. */
621
622 #ifndef EBCDIC /* ASCII/UTF-8 coding */
623 else if (c < CHAR_0 || c > CHAR_z) {} /* Not alphanumeric */
624 else if ((i = escapes[c - CHAR_0]) != 0) c = i;
625
626 #else /* EBCDIC coding */
627 else if (c < 'a' || (ebcdic_chartab[c] & 0x0E) == 0) {} /* Not alphanumeric */
628 else if ((i = escapes[c - 0x48]) != 0) c = i;
629 #endif
630
631 /* Escapes that need further processing, or are illegal. */
632
633 else
634 {
635 const uschar *oldptr;
636 BOOL braced, negated;
637
638 switch (c)
639 {
640 /* A number of Perl escapes are not handled by PCRE. We give an explicit
641 error. */
642
643 case CHAR_l:
644 case CHAR_L:
645 case CHAR_u:
646 case CHAR_U:
647 *errorcodeptr = ERR37;
648 break;
649
650 /* \g must be followed by one of a number of specific things:
651
652 (1) A number, either plain or braced. If positive, it is an absolute
653 backreference. If negative, it is a relative backreference. This is a Perl
654 5.10 feature.
655
656 (2) Perl 5.10 also supports \g{name} as a reference to a named group. This
657 is part of Perl's movement towards a unified syntax for back references. As
658 this is synonymous with \k{name}, we fudge it up by pretending it really
659 was \k.
660
661 (3) For Oniguruma compatibility we also support \g followed by a name or a
662 number either in angle brackets or in single quotes. However, these are
663 (possibly recursive) subroutine calls, _not_ backreferences. Just return
664 the -ESC_g code (cf \k). */
665
666 case CHAR_g:
667 if (ptr[1] == CHAR_LESS_THAN_SIGN || ptr[1] == CHAR_APOSTROPHE)
668 {
669 c = -ESC_g;
670 break;
671 }
672
673 /* Handle the Perl-compatible cases */
674
675 if (ptr[1] == CHAR_LEFT_CURLY_BRACKET)
676 {
677 const uschar *p;
678 for (p = ptr+2; *p != 0 && *p != CHAR_RIGHT_CURLY_BRACKET; p++)
679 if (*p != CHAR_MINUS && (digitab[*p] & ctype_digit) == 0) break;
680 if (*p != 0 && *p != CHAR_RIGHT_CURLY_BRACKET)
681 {
682 c = -ESC_k;
683 break;
684 }
685 braced = TRUE;
686 ptr++;
687 }
688 else braced = FALSE;
689
690 if (ptr[1] == CHAR_MINUS)
691 {
692 negated = TRUE;
693 ptr++;
694 }
695 else negated = FALSE;
696
697 c = 0;
698 while ((digitab[ptr[1]] & ctype_digit) != 0)
699 c = c * 10 + *(++ptr) - CHAR_0;
700
701 if (c < 0) /* Integer overflow */
702 {
703 *errorcodeptr = ERR61;
704 break;
705 }
706
707 if (braced && *(++ptr) != CHAR_RIGHT_CURLY_BRACKET)
708 {
709 *errorcodeptr = ERR57;
710 break;
711 }
712
713 if (c == 0)
714 {
715 *errorcodeptr = ERR58;
716 break;
717 }
718
719 if (negated)
720 {
721 if (c > bracount)
722 {
723 *errorcodeptr = ERR15;
724 break;
725 }
726 c = bracount - (c - 1);
727 }
728
729 c = -(ESC_REF + c);
730 break;
731
732 /* The handling of escape sequences consisting of a string of digits
733 starting with one that is not zero is not straightforward. By experiment,
734 the way Perl works seems to be as follows:
735
736 Outside a character class, the digits are read as a decimal number. If the
737 number is less than 10, or if there are that many previous extracting
738 left brackets, then it is a back reference. Otherwise, up to three octal
739 digits are read to form an escaped byte. Thus \123 is likely to be octal
740 123 (cf \0123, which is octal 012 followed by the literal 3). If the octal
741 value is greater than 377, the least significant 8 bits are taken. Inside a
742 character class, \ followed by a digit is always an octal number. */
743
744 case CHAR_1: case CHAR_2: case CHAR_3: case CHAR_4: case CHAR_5:
745 case CHAR_6: case CHAR_7: case CHAR_8: case CHAR_9:
746
747 if (!isclass)
748 {
749 oldptr = ptr;
750 c -= CHAR_0;
751 while ((digitab[ptr[1]] & ctype_digit) != 0)
752 c = c * 10 + *(++ptr) - CHAR_0;
753 if (c < 0) /* Integer overflow */
754 {
755 *errorcodeptr = ERR61;
756 break;
757 }
758 if (c < 10 || c <= bracount)
759 {
760 c = -(ESC_REF + c);
761 break;
762 }
763 ptr = oldptr; /* Put the pointer back and fall through */
764 }
765
766 /* Handle an octal number following \. If the first digit is 8 or 9, Perl
767 generates a binary zero byte and treats the digit as a following literal.
768 Thus we have to pull back the pointer by one. */
769
770 if ((c = *ptr) >= CHAR_8)
771 {
772 ptr--;
773 c = 0;
774 break;
775 }
776
777 /* \0 always starts an octal number, but we may drop through to here with a
778 larger first octal digit. The original code used just to take the least
779 significant 8 bits of octal numbers (I think this is what early Perls used
780 to do). Nowadays we allow for larger numbers in UTF-8 mode, but no more
781 than 3 octal digits. */
782
783 case CHAR_0:
784 c -= CHAR_0;
785 while(i++ < 2 && ptr[1] >= CHAR_0 && ptr[1] <= CHAR_7)
786 c = c * 8 + *(++ptr) - CHAR_0;
787 if (!utf8 && c > 255) *errorcodeptr = ERR51;
788 break;
789
790 /* \x is complicated. \x{ddd} is a character number which can be greater
791 than 0xff in utf8 mode, but only if the ddd are hex digits. If not, { is
792 treated as a data character. */
793
794 case CHAR_x:
795 if (ptr[1] == CHAR_LEFT_CURLY_BRACKET)
796 {
797 const uschar *pt = ptr + 2;
798 int count = 0;
799
800 c = 0;
801 while ((digitab[*pt] & ctype_xdigit) != 0)
802 {
803 register int cc = *pt++;
804 if (c == 0 && cc == CHAR_0) continue; /* Leading zeroes */
805 count++;
806
807 #ifndef EBCDIC /* ASCII/UTF-8 coding */
808 if (cc >= CHAR_a) cc -= 32; /* Convert to upper case */
809 c = (c << 4) + cc - ((cc < CHAR_A)? CHAR_0 : (CHAR_A - 10));
810 #else /* EBCDIC coding */
811 if (cc >= CHAR_a && cc <= CHAR_z) cc += 64; /* Convert to upper case */
812 c = (c << 4) + cc - ((cc >= CHAR_0)? CHAR_0 : (CHAR_A - 10));
813 #endif
814 }
815
816 if (*pt == CHAR_RIGHT_CURLY_BRACKET)
817 {
818 if (c < 0 || count > (utf8? 8 : 2)) *errorcodeptr = ERR34;
819 ptr = pt;
820 break;
821 }
822
823 /* If the sequence of hex digits does not end with '}', then we don't
824 recognize this construct; fall through to the normal \x handling. */
825 }
826
827 /* Read just a single-byte hex-defined char */
828
829 c = 0;
830 while (i++ < 2 && (digitab[ptr[1]] & ctype_xdigit) != 0)
831 {
832 int cc; /* Some compilers don't like */
833 cc = *(++ptr); /* ++ in initializers */
834 #ifndef EBCDIC /* ASCII/UTF-8 coding */
835 if (cc >= CHAR_a) cc -= 32; /* Convert to upper case */
836 c = c * 16 + cc - ((cc < CHAR_A)? CHAR_0 : (CHAR_A - 10));
837 #else /* EBCDIC coding */
838 if (cc <= CHAR_z) cc += 64; /* Convert to upper case */
839 c = c * 16 + cc - ((cc >= CHAR_0)? CHAR_0 : (CHAR_A - 10));
840 #endif
841 }
842 break;
843
844 /* For \c, a following letter is upper-cased; then the 0x40 bit is flipped.
845 An error is given if the byte following \c is not an ASCII character. This
846 coding is ASCII-specific, but then the whole concept of \cx is
847 ASCII-specific. (However, an EBCDIC equivalent has now been added.) */
848
849 case CHAR_c:
850 c = *(++ptr);
851 if (c == 0)
852 {
853 *errorcodeptr = ERR2;
854 break;
855 }
856 #ifndef EBCDIC /* ASCII/UTF-8 coding */
857 if (c > 127) /* Excludes all non-ASCII in either mode */
858 {
859 *errorcodeptr = ERR68;
860 break;
861 }
862 if (c >= CHAR_a && c <= CHAR_z) c -= 32;
863 c ^= 0x40;
864 #else /* EBCDIC coding */
865 if (c >= CHAR_a && c <= CHAR_z) c += 64;
866 c ^= 0xC0;
867 #endif
868 break;
869
870 /* PCRE_EXTRA enables extensions to Perl in the matter of escapes. Any
871 other alphanumeric following \ is an error if PCRE_EXTRA was set;
872 otherwise, for Perl compatibility, it is a literal. This code looks a bit
873 odd, but there used to be some cases other than the default, and there may
874 be again in future, so I haven't "optimized" it. */
875
876 default:
877 if ((options & PCRE_EXTRA) != 0) switch(c)
878 {
879 default:
880 *errorcodeptr = ERR3;
881 break;
882 }
883 break;
884 }
885 }
886
887 /* Perl supports \N{name} for character names, as well as plain \N for "not
888 newline". PCRE does not support \N{name}. */
889
890 if (c == -ESC_N && ptr[1] == CHAR_LEFT_CURLY_BRACKET)
891 *errorcodeptr = ERR37;
892
893 /* If PCRE_UCP is set, we change the values for \d etc. */
894
895 if ((options & PCRE_UCP) != 0 && c <= -ESC_D && c >= -ESC_w)
896 c -= (ESC_DU - ESC_D);
897
898 /* Set the pointer to the final character before returning. */
899
900 *ptrptr = ptr;
901 return c;
902 }
903
904
905
906 #ifdef SUPPORT_UCP
907 /*************************************************
908 * Handle \P and \p *
909 *************************************************/
910
911 /* This function is called after \P or \p has been encountered, provided that
912 PCRE is compiled with support for Unicode properties. On entry, ptrptr is
913 pointing at the P or p. On exit, it is pointing at the final character of the
914 escape sequence.
915
916 Argument:
917 ptrptr points to the pattern position pointer
918 negptr points to a boolean that is set TRUE for negation else FALSE
919 dptr points to an int that is set to the detailed property value
920 errorcodeptr points to the error code variable
921
922 Returns: type value from ucp_type_table, or -1 for an invalid type
923 */
924
925 static int
926 get_ucp(const uschar **ptrptr, BOOL *negptr, int *dptr, int *errorcodeptr)
927 {
928 int c, i, bot, top;
929 const uschar *ptr = *ptrptr;
930 char name[32];
931
932 c = *(++ptr);
933 if (c == 0) goto ERROR_RETURN;
934
935 *negptr = FALSE;
936
937 /* \P or \p can be followed by a name in {}, optionally preceded by ^ for
938 negation. */
939
940 if (c == CHAR_LEFT_CURLY_BRACKET)
941 {
942 if (ptr[1] == CHAR_CIRCUMFLEX_ACCENT)
943 {
944 *negptr = TRUE;
945 ptr++;
946 }
947 for (i = 0; i < (int)sizeof(name) - 1; i++)
948 {
949 c = *(++ptr);
950 if (c == 0) goto ERROR_RETURN;
951 if (c == CHAR_RIGHT_CURLY_BRACKET) break;
952 name[i] = c;
953 }
954 if (c != CHAR_RIGHT_CURLY_BRACKET) goto ERROR_RETURN;
955 name[i] = 0;
956 }
957
958 /* Otherwise there is just one following character */
959
960 else
961 {
962 name[0] = c;
963 name[1] = 0;
964 }
965
966 *ptrptr = ptr;
967
968 /* Search for a recognized property name using binary chop */
969
970 bot = 0;
971 top = _pcre_utt_size;
972
973 while (bot < top)
974 {
975 i = (bot + top) >> 1;
976 c = strcmp(name, _pcre_utt_names + _pcre_utt[i].name_offset);
977 if (c == 0)
978 {
979 *dptr = _pcre_utt[i].value;
980 return _pcre_utt[i].type;
981 }
982 if (c > 0) bot = i + 1; else top = i;
983 }
984
985 *errorcodeptr = ERR47;
986 *ptrptr = ptr;
987 return -1;
988
989 ERROR_RETURN:
990 *errorcodeptr = ERR46;
991 *ptrptr = ptr;
992 return -1;
993 }
994 #endif
995
996
997
998
999 /*************************************************
1000 * Check for counted repeat *
1001 *************************************************/
1002
1003 /* This function is called when a '{' is encountered in a place where it might
1004 start a quantifier. It looks ahead to see if it really is a quantifier or not.
1005 It is only a quantifier if it is one of the forms {ddd} {ddd,} or {ddd,ddd}
1006 where the ddds are digits.
1007
1008 Arguments:
1009 p pointer to the first char after '{'
1010
1011 Returns: TRUE or FALSE
1012 */
1013
1014 static BOOL
1015 is_counted_repeat(const uschar *p)
1016 {
1017 if ((digitab[*p++] & ctype_digit) == 0) return FALSE;
1018 while ((digitab[*p] & ctype_digit) != 0) p++;
1019 if (*p == CHAR_RIGHT_CURLY_BRACKET) return TRUE;
1020
1021 if (*p++ != CHAR_COMMA) return FALSE;
1022 if (*p == CHAR_RIGHT_CURLY_BRACKET) return TRUE;
1023
1024 if ((digitab[*p++] & ctype_digit) == 0) return FALSE;
1025 while ((digitab[*p] & ctype_digit) != 0) p++;
1026
1027 return (*p == CHAR_RIGHT_CURLY_BRACKET);
1028 }
1029
1030
1031
1032 /*************************************************
1033 * Read repeat counts *
1034 *************************************************/
1035
1036 /* Read an item of the form {n,m} and return the values. This is called only
1037 after is_counted_repeat() has confirmed that a repeat-count quantifier exists,
1038 so the syntax is guaranteed to be correct, but we need to check the values.
1039
1040 Arguments:
1041 p pointer to first char after '{'
1042 minp pointer to int for min
1043 maxp pointer to int for max
1044 returned as -1 if no max
1045 errorcodeptr points to error code variable
1046
1047 Returns: pointer to '}' on success;
1048 current ptr on error, with errorcodeptr set non-zero
1049 */
1050
1051 static const uschar *
1052 read_repeat_counts(const uschar *p, int *minp, int *maxp, int *errorcodeptr)
1053 {
1054 int min = 0;
1055 int max = -1;
1056
1057 /* Read the minimum value and do a paranoid check: a negative value indicates
1058 an integer overflow. */
1059
1060 while ((digitab[*p] & ctype_digit) != 0) min = min * 10 + *p++ - CHAR_0;
1061 if (min < 0 || min > 65535)
1062 {
1063 *errorcodeptr = ERR5;
1064 return p;
1065 }
1066
1067 /* Read the maximum value if there is one, and again do a paranoid on its size.
1068 Also, max must not be less than min. */
1069
1070 if (*p == CHAR_RIGHT_CURLY_BRACKET) max = min; else
1071 {
1072 if (*(++p) != CHAR_RIGHT_CURLY_BRACKET)
1073 {
1074 max = 0;
1075 while((digitab[*p] & ctype_digit) != 0) max = max * 10 + *p++ - CHAR_0;
1076 if (max < 0 || max > 65535)
1077 {
1078 *errorcodeptr = ERR5;
1079 return p;
1080 }
1081 if (max < min)
1082 {
1083 *errorcodeptr = ERR4;
1084 return p;
1085 }
1086 }
1087 }
1088
1089 /* Fill in the required variables, and pass back the pointer to the terminating
1090 '}'. */
1091
1092 *minp = min;
1093 *maxp = max;
1094 return p;
1095 }
1096
1097
1098
1099 /*************************************************
1100 * Subroutine for finding forward reference *
1101 *************************************************/
1102
1103 /* This recursive function is called only from find_parens() below. The
1104 top-level call starts at the beginning of the pattern. All other calls must
1105 start at a parenthesis. It scans along a pattern's text looking for capturing
1106 subpatterns, and counting them. If it finds a named pattern that matches the
1107 name it is given, it returns its number. Alternatively, if the name is NULL, it
1108 returns when it reaches a given numbered subpattern. Recursion is used to keep
1109 track of subpatterns that reset the capturing group numbers - the (?| feature.
1110
1111 This function was originally called only from the second pass, in which we know
1112 that if (?< or (?' or (?P< is encountered, the name will be correctly
1113 terminated because that is checked in the first pass. There is now one call to
1114 this function in the first pass, to check for a recursive back reference by
1115 name (so that we can make the whole group atomic). In this case, we need check
1116 only up to the current position in the pattern, and that is still OK because
1117 and previous occurrences will have been checked. To make this work, the test
1118 for "end of pattern" is a check against cd->end_pattern in the main loop,
1119 instead of looking for a binary zero. This means that the special first-pass
1120 call can adjust cd->end_pattern temporarily. (Checks for binary zero while
1121 processing items within the loop are OK, because afterwards the main loop will
1122 terminate.)
1123
1124 Arguments:
1125 ptrptr address of the current character pointer (updated)
1126 cd compile background data
1127 name name to seek, or NULL if seeking a numbered subpattern
1128 lorn name length, or subpattern number if name is NULL
1129 xmode TRUE if we are in /x mode
1130 utf8 TRUE if we are in UTF-8 mode
1131 count pointer to the current capturing subpattern number (updated)
1132
1133 Returns: the number of the named subpattern, or -1 if not found
1134 */
1135
1136 static int
1137 find_parens_sub(uschar **ptrptr, compile_data *cd, const uschar *name, int lorn,
1138 BOOL xmode, BOOL utf8, int *count)
1139 {
1140 uschar *ptr = *ptrptr;
1141 int start_count = *count;
1142 int hwm_count = start_count;
1143 BOOL dup_parens = FALSE;
1144
1145 /* If the first character is a parenthesis, check on the type of group we are
1146 dealing with. The very first call may not start with a parenthesis. */
1147
1148 if (ptr[0] == CHAR_LEFT_PARENTHESIS)
1149 {
1150 /* Handle specials such as (*SKIP) or (*UTF8) etc. */
1151
1152 if (ptr[1] == CHAR_ASTERISK) ptr += 2;
1153
1154 /* Handle a normal, unnamed capturing parenthesis. */
1155
1156 else if (ptr[1] != CHAR_QUESTION_MARK)
1157 {
1158 *count += 1;
1159 if (name == NULL && *count == lorn) return *count;
1160 ptr++;
1161 }
1162
1163 /* All cases now have (? at the start. Remember when we are in a group
1164 where the parenthesis numbers are duplicated. */
1165
1166 else if (ptr[2] == CHAR_VERTICAL_LINE)
1167 {
1168 ptr += 3;
1169 dup_parens = TRUE;
1170 }
1171
1172 /* Handle comments; all characters are allowed until a ket is reached. */
1173
1174 else if (ptr[2] == CHAR_NUMBER_SIGN)
1175 {
1176 for (ptr += 3; *ptr != 0; ptr++) if (*ptr == CHAR_RIGHT_PARENTHESIS) break;
1177 goto FAIL_EXIT;
1178 }
1179
1180 /* Handle a condition. If it is an assertion, just carry on so that it
1181 is processed as normal. If not, skip to the closing parenthesis of the
1182 condition (there can't be any nested parens). */
1183
1184 else if (ptr[2] == CHAR_LEFT_PARENTHESIS)
1185 {
1186 ptr += 2;
1187 if (ptr[1] != CHAR_QUESTION_MARK)
1188 {
1189 while (*ptr != 0 && *ptr != CHAR_RIGHT_PARENTHESIS) ptr++;
1190 if (*ptr != 0) ptr++;
1191 }
1192 }
1193
1194 /* Start with (? but not a condition. */
1195
1196 else
1197 {
1198 ptr += 2;
1199 if (*ptr == CHAR_P) ptr++; /* Allow optional P */
1200
1201 /* We have to disambiguate (?<! and (?<= from (?<name> for named groups */
1202
1203 if ((*ptr == CHAR_LESS_THAN_SIGN && ptr[1] != CHAR_EXCLAMATION_MARK &&
1204 ptr[1] != CHAR_EQUALS_SIGN) || *ptr == CHAR_APOSTROPHE)
1205 {
1206 int term;
1207 const uschar *thisname;
1208 *count += 1;
1209 if (name == NULL && *count == lorn) return *count;
1210 term = *ptr++;
1211 if (term == CHAR_LESS_THAN_SIGN) term = CHAR_GREATER_THAN_SIGN;
1212 thisname = ptr;
1213 while (*ptr != term) ptr++;
1214 if (name != NULL && lorn == ptr - thisname &&
1215 strncmp((const char *)name, (const char *)thisname, lorn) == 0)
1216 return *count;
1217 term++;
1218 }
1219 }
1220 }
1221
1222 /* Past any initial parenthesis handling, scan for parentheses or vertical
1223 bars. Stop if we get to cd->end_pattern. Note that this is important for the
1224 first-pass call when this value is temporarily adjusted to stop at the current
1225 position. So DO NOT change this to a test for binary zero. */
1226
1227 for (; ptr < cd->end_pattern; ptr++)
1228 {
1229 /* Skip over backslashed characters and also entire \Q...\E */
1230
1231 if (*ptr == CHAR_BACKSLASH)
1232 {
1233 if (*(++ptr) == 0) goto FAIL_EXIT;
1234 if (*ptr == CHAR_Q) for (;;)
1235 {
1236 while (*(++ptr) != 0 && *ptr != CHAR_BACKSLASH) {};
1237 if (*ptr == 0) goto FAIL_EXIT;
1238 if (*(++ptr) == CHAR_E) break;
1239 }
1240 continue;
1241 }
1242
1243 /* Skip over character classes; this logic must be similar to the way they
1244 are handled for real. If the first character is '^', skip it. Also, if the
1245 first few characters (either before or after ^) are \Q\E or \E we skip them
1246 too. This makes for compatibility with Perl. Note the use of STR macros to
1247 encode "Q\\E" so that it works in UTF-8 on EBCDIC platforms. */
1248
1249 if (*ptr == CHAR_LEFT_SQUARE_BRACKET)
1250 {
1251 BOOL negate_class = FALSE;
1252 for (;;)
1253 {
1254 if (ptr[1] == CHAR_BACKSLASH)
1255 {
1256 if (ptr[2] == CHAR_E)
1257 ptr+= 2;
1258 else if (strncmp((const char *)ptr+2,
1259 STR_Q STR_BACKSLASH STR_E, 3) == 0)
1260 ptr += 4;
1261 else
1262 break;
1263 }
1264 else if (!negate_class && ptr[1] == CHAR_CIRCUMFLEX_ACCENT)
1265 {
1266 negate_class = TRUE;
1267 ptr++;
1268 }
1269 else break;
1270 }
1271
1272 /* If the next character is ']', it is a data character that must be
1273 skipped, except in JavaScript compatibility mode. */
1274
1275 if (ptr[1] == CHAR_RIGHT_SQUARE_BRACKET &&
1276 (cd->external_options & PCRE_JAVASCRIPT_COMPAT) == 0)
1277 ptr++;
1278
1279 while (*(++ptr) != CHAR_RIGHT_SQUARE_BRACKET)
1280 {
1281 if (*ptr == 0) return -1;
1282 if (*ptr == CHAR_BACKSLASH)
1283 {
1284 if (*(++ptr) == 0) goto FAIL_EXIT;
1285 if (*ptr == CHAR_Q) for (;;)
1286 {
1287 while (*(++ptr) != 0 && *ptr != CHAR_BACKSLASH) {};
1288 if (*ptr == 0) goto FAIL_EXIT;
1289 if (*(++ptr) == CHAR_E) break;
1290 }
1291 continue;
1292 }
1293 }
1294 continue;
1295 }
1296
1297 /* Skip comments in /x mode */
1298
1299 if (xmode && *ptr == CHAR_NUMBER_SIGN)
1300 {
1301 ptr++;
1302 while (*ptr != 0)
1303 {
1304 if (IS_NEWLINE(ptr)) { ptr += cd->nllen - 1; break; }
1305 ptr++;
1306 #ifdef SUPPORT_UTF8
1307 if (utf8) while ((*ptr & 0xc0) == 0x80) ptr++;
1308 #endif
1309 }
1310 if (*ptr == 0) goto FAIL_EXIT;
1311 continue;
1312 }
1313
1314 /* Check for the special metacharacters */
1315
1316 if (*ptr == CHAR_LEFT_PARENTHESIS)
1317 {
1318 int rc = find_parens_sub(&ptr, cd, name, lorn, xmode, utf8, count);
1319 if (rc > 0) return rc;
1320 if (*ptr == 0) goto FAIL_EXIT;
1321 }
1322
1323 else if (*ptr == CHAR_RIGHT_PARENTHESIS)
1324 {
1325 if (dup_parens && *count < hwm_count) *count = hwm_count;
1326 goto FAIL_EXIT;
1327 }
1328
1329 else if (*ptr == CHAR_VERTICAL_LINE && dup_parens)
1330 {
1331 if (*count > hwm_count) hwm_count = *count;
1332 *count = start_count;
1333 }
1334 }
1335
1336 FAIL_EXIT:
1337 *ptrptr = ptr;
1338 return -1;
1339 }
1340
1341
1342
1343
1344 /*************************************************
1345 * Find forward referenced subpattern *
1346 *************************************************/
1347
1348 /* This function scans along a pattern's text looking for capturing
1349 subpatterns, and counting them. If it finds a named pattern that matches the
1350 name it is given, it returns its number. Alternatively, if the name is NULL, it
1351 returns when it reaches a given numbered subpattern. This is used for forward
1352 references to subpatterns. We used to be able to start this scan from the
1353 current compiling point, using the current count value from cd->bracount, and
1354 do it all in a single loop, but the addition of the possibility of duplicate
1355 subpattern numbers means that we have to scan from the very start, in order to
1356 take account of such duplicates, and to use a recursive function to keep track
1357 of the different types of group.
1358
1359 Arguments:
1360 cd compile background data
1361 name name to seek, or NULL if seeking a numbered subpattern
1362 lorn name length, or subpattern number if name is NULL
1363 xmode TRUE if we are in /x mode
1364 utf8 TRUE if we are in UTF-8 mode
1365
1366 Returns: the number of the found subpattern, or -1 if not found
1367 */
1368
1369 static int
1370 find_parens(compile_data *cd, const uschar *name, int lorn, BOOL xmode,
1371 BOOL utf8)
1372 {
1373 uschar *ptr = (uschar *)cd->start_pattern;
1374 int count = 0;
1375 int rc;
1376
1377 /* If the pattern does not start with an opening parenthesis, the first call
1378 to find_parens_sub() will scan right to the end (if necessary). However, if it
1379 does start with a parenthesis, find_parens_sub() will return when it hits the
1380 matching closing parens. That is why we have to have a loop. */
1381
1382 for (;;)
1383 {
1384 rc = find_parens_sub(&ptr, cd, name, lorn, xmode, utf8, &count);
1385 if (rc > 0 || *ptr++ == 0) break;
1386 }
1387
1388 return rc;
1389 }
1390
1391
1392
1393
1394 /*************************************************
1395 * Find first significant op code *
1396 *************************************************/
1397
1398 /* This is called by several functions that scan a compiled expression looking
1399 for a fixed first character, or an anchoring op code etc. It skips over things
1400 that do not influence this. For some calls, it makes sense to skip negative
1401 forward and all backward assertions, and also the \b assertion; for others it
1402 does not.
1403
1404 Arguments:
1405 code pointer to the start of the group
1406 skipassert TRUE if certain assertions are to be skipped
1407
1408 Returns: pointer to the first significant opcode
1409 */
1410
1411 static const uschar*
1412 first_significant_code(const uschar *code, BOOL skipassert)
1413 {
1414 for (;;)
1415 {
1416 switch ((int)*code)
1417 {
1418 case OP_ASSERT_NOT:
1419 case OP_ASSERTBACK:
1420 case OP_ASSERTBACK_NOT:
1421 if (!skipassert) return code;
1422 do code += GET(code, 1); while (*code == OP_ALT);
1423 code += _pcre_OP_lengths[*code];
1424 break;
1425
1426 case OP_WORD_BOUNDARY:
1427 case OP_NOT_WORD_BOUNDARY:
1428 if (!skipassert) return code;
1429 /* Fall through */
1430
1431 case OP_CALLOUT:
1432 case OP_CREF:
1433 case OP_NCREF:
1434 case OP_RREF:
1435 case OP_NRREF:
1436 case OP_DEF:
1437 code += _pcre_OP_lengths[*code];
1438 break;
1439
1440 default:
1441 return code;
1442 }
1443 }
1444 /* Control never reaches here */
1445 }
1446
1447
1448
1449
1450 /*************************************************
1451 * Find the fixed length of a branch *
1452 *************************************************/
1453
1454 /* Scan a branch and compute the fixed length of subject that will match it,
1455 if the length is fixed. This is needed for dealing with backward assertions.
1456 In UTF8 mode, the result is in characters rather than bytes. The branch is
1457 temporarily terminated with OP_END when this function is called.
1458
1459 This function is called when a backward assertion is encountered, so that if it
1460 fails, the error message can point to the correct place in the pattern.
1461 However, we cannot do this when the assertion contains subroutine calls,
1462 because they can be forward references. We solve this by remembering this case
1463 and doing the check at the end; a flag specifies which mode we are running in.
1464
1465 Arguments:
1466 code points to the start of the pattern (the bracket)
1467 utf8 TRUE in UTF-8 mode
1468 atend TRUE if called when the pattern is complete
1469 cd the "compile data" structure
1470
1471 Returns: the fixed length,
1472 or -1 if there is no fixed length,
1473 or -2 if \C was encountered
1474 or -3 if an OP_RECURSE item was encountered and atend is FALSE
1475 */
1476
1477 static int
1478 find_fixedlength(uschar *code, BOOL utf8, BOOL atend, compile_data *cd)
1479 {
1480 int length = -1;
1481
1482 register int branchlength = 0;
1483 register uschar *cc = code + 1 + LINK_SIZE;
1484
1485 /* Scan along the opcodes for this branch. If we get to the end of the
1486 branch, check the length against that of the other branches. */
1487
1488 for (;;)
1489 {
1490 int d;
1491 uschar *ce, *cs;
1492 register int op = *cc;
1493 switch (op)
1494 {
1495 /* We only need to continue for OP_CBRA (normal capturing bracket) and
1496 OP_BRA (normal non-capturing bracket) because the other variants of these
1497 opcodes are all concerned with unlimited repeated groups, which of course
1498 are not of fixed length. They will cause a -1 response from the default
1499 case of this switch. */
1500
1501 case OP_CBRA:
1502 case OP_BRA:
1503 case OP_ONCE:
1504 case OP_COND:
1505 d = find_fixedlength(cc + ((op == OP_CBRA)? 2:0), utf8, atend, cd);
1506 if (d < 0) return d;
1507 branchlength += d;
1508 do cc += GET(cc, 1); while (*cc == OP_ALT);
1509 cc += 1 + LINK_SIZE;
1510 break;
1511
1512 /* Reached end of a branch; if it's a ket it is the end of a nested
1513 call. If it's ALT it is an alternation in a nested call. If it is
1514 END it's the end of the outer call. All can be handled by the same code.
1515 Note that we must not include the OP_KETRxxx opcodes here, because they
1516 all imply an unlimited repeat. */
1517
1518 case OP_ALT:
1519 case OP_KET:
1520 case OP_END:
1521 if (length < 0) length = branchlength;
1522 else if (length != branchlength) return -1;
1523 if (*cc != OP_ALT) return length;
1524 cc += 1 + LINK_SIZE;
1525 branchlength = 0;
1526 break;
1527
1528 /* A true recursion implies not fixed length, but a subroutine call may
1529 be OK. If the subroutine is a forward reference, we can't deal with
1530 it until the end of the pattern, so return -3. */
1531
1532 case OP_RECURSE:
1533 if (!atend) return -3;
1534 cs = ce = (uschar *)cd->start_code + GET(cc, 1); /* Start subpattern */
1535 do ce += GET(ce, 1); while (*ce == OP_ALT); /* End subpattern */
1536 if (cc > cs && cc < ce) return -1; /* Recursion */
1537 d = find_fixedlength(cs + 2, utf8, atend, cd);
1538 if (d < 0) return d;
1539 branchlength += d;
1540 cc += 1 + LINK_SIZE;
1541 break;
1542
1543 /* Skip over assertive subpatterns */
1544
1545 case OP_ASSERT:
1546 case OP_ASSERT_NOT:
1547 case OP_ASSERTBACK:
1548 case OP_ASSERTBACK_NOT:
1549 do cc += GET(cc, 1); while (*cc == OP_ALT);
1550 /* Fall through */
1551
1552 /* Skip over things that don't match chars */
1553
1554 case OP_REVERSE:
1555 case OP_CREF:
1556 case OP_NCREF:
1557 case OP_RREF:
1558 case OP_NRREF:
1559 case OP_DEF:
1560 case OP_CALLOUT:
1561 case OP_SOD:
1562 case OP_SOM:
1563 case OP_SET_SOM:
1564 case OP_EOD:
1565 case OP_EODN:
1566 case OP_CIRC:
1567 case OP_CIRCM:
1568 case OP_DOLL:
1569 case OP_DOLLM:
1570 case OP_NOT_WORD_BOUNDARY:
1571 case OP_WORD_BOUNDARY:
1572 cc += _pcre_OP_lengths[*cc];
1573 break;
1574
1575 /* Handle literal characters */
1576
1577 case OP_CHAR:
1578 case OP_CHARI:
1579 case OP_NOT:
1580 case OP_NOTI:
1581 branchlength++;
1582 cc += 2;
1583 #ifdef SUPPORT_UTF8
1584 if (utf8 && cc[-1] >= 0xc0) cc += _pcre_utf8_table4[cc[-1] & 0x3f];
1585 #endif
1586 break;
1587
1588 /* Handle exact repetitions. The count is already in characters, but we
1589 need to skip over a multibyte character in UTF8 mode. */
1590
1591 case OP_EXACT:
1592 branchlength += GET2(cc,1);
1593 cc += 4;
1594 #ifdef SUPPORT_UTF8
1595 if (utf8 && cc[-1] >= 0xc0) cc += _pcre_utf8_table4[cc[-1] & 0x3f];
1596 #endif
1597 break;
1598
1599 case OP_TYPEEXACT:
1600 branchlength += GET2(cc,1);
1601 if (cc[3] == OP_PROP || cc[3] == OP_NOTPROP) cc += 2;
1602 cc += 4;
1603 break;
1604
1605 /* Handle single-char matchers */
1606
1607 case OP_PROP:
1608 case OP_NOTPROP:
1609 cc += 2;
1610 /* Fall through */
1611
1612 case OP_NOT_DIGIT:
1613 case OP_DIGIT:
1614 case OP_NOT_WHITESPACE:
1615 case OP_WHITESPACE:
1616 case OP_NOT_WORDCHAR:
1617 case OP_WORDCHAR:
1618 case OP_ANY:
1619 case OP_ALLANY:
1620 branchlength++;
1621 cc++;
1622 break;
1623
1624 /* The single-byte matcher isn't allowed */
1625
1626 case OP_ANYBYTE:
1627 return -2;
1628
1629 /* Check a class for variable quantification */
1630
1631 #ifdef SUPPORT_UTF8
1632 case OP_XCLASS:
1633 cc += GET(cc, 1) - 33;
1634 /* Fall through */
1635 #endif
1636
1637 case OP_CLASS:
1638 case OP_NCLASS:
1639 cc += 33;
1640
1641 switch (*cc)
1642 {
1643 case OP_CRSTAR:
1644 case OP_CRMINSTAR:
1645 case OP_CRQUERY:
1646 case OP_CRMINQUERY:
1647 return -1;
1648
1649 case OP_CRRANGE:
1650 case OP_CRMINRANGE:
1651 if (GET2(cc,1) != GET2(cc,3)) return -1;
1652 branchlength += GET2(cc,1);
1653 cc += 5;
1654 break;
1655
1656 default:
1657 branchlength++;
1658 }
1659 break;
1660
1661 /* Anything else is variable length */
1662
1663 default:
1664 return -1;
1665 }
1666 }
1667 /* Control never gets here */
1668 }
1669
1670
1671
1672
1673 /*************************************************
1674 * Scan compiled regex for specific bracket *
1675 *************************************************/
1676
1677 /* This little function scans through a compiled pattern until it finds a
1678 capturing bracket with the given number, or, if the number is negative, an
1679 instance of OP_REVERSE for a lookbehind. The function is global in the C sense
1680 so that it can be called from pcre_study() when finding the minimum matching
1681 length.
1682
1683 Arguments:
1684 code points to start of expression
1685 utf8 TRUE in UTF-8 mode
1686 number the required bracket number or negative to find a lookbehind
1687
1688 Returns: pointer to the opcode for the bracket, or NULL if not found
1689 */
1690
1691 const uschar *
1692 _pcre_find_bracket(const uschar *code, BOOL utf8, int number)
1693 {
1694 for (;;)
1695 {
1696 register int c = *code;
1697
1698 if (c == OP_END) return NULL;
1699
1700 /* XCLASS is used for classes that cannot be represented just by a bit
1701 map. This includes negated single high-valued characters. The length in
1702 the table is zero; the actual length is stored in the compiled code. */
1703
1704 if (c == OP_XCLASS) code += GET(code, 1);
1705
1706 /* Handle recursion */
1707
1708 else if (c == OP_REVERSE)
1709 {
1710 if (number < 0) return (uschar *)code;
1711 code += _pcre_OP_lengths[c];
1712 }
1713
1714 /* Handle capturing bracket */
1715
1716 else if (c == OP_CBRA || c == OP_SCBRA ||
1717 c == OP_CBRAPOS || c == OP_SCBRAPOS)
1718 {
1719 int n = GET2(code, 1+LINK_SIZE);
1720 if (n == number) return (uschar *)code;
1721 code += _pcre_OP_lengths[c];
1722 }
1723
1724 /* Otherwise, we can get the item's length from the table, except that for
1725 repeated character types, we have to test for \p and \P, which have an extra
1726 two bytes of parameters, and for MARK/PRUNE/SKIP/THEN with an argument, we
1727 must add in its length. */
1728
1729 else
1730 {
1731 switch(c)
1732 {
1733 case OP_TYPESTAR:
1734 case OP_TYPEMINSTAR:
1735 case OP_TYPEPLUS:
1736 case OP_TYPEMINPLUS:
1737 case OP_TYPEQUERY:
1738 case OP_TYPEMINQUERY:
1739 case OP_TYPEPOSSTAR:
1740 case OP_TYPEPOSPLUS:
1741 case OP_TYPEPOSQUERY:
1742 if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
1743 break;
1744
1745 case OP_TYPEUPTO:
1746 case OP_TYPEMINUPTO:
1747 case OP_TYPEEXACT:
1748 case OP_TYPEPOSUPTO:
1749 if (code[3] == OP_PROP || code[3] == OP_NOTPROP) code += 2;
1750 break;
1751
1752 case OP_MARK:
1753 case OP_PRUNE_ARG:
1754 case OP_SKIP_ARG:
1755 code += code[1];
1756 break;
1757
1758 case OP_THEN_ARG:
1759 code += code[1+LINK_SIZE];
1760 break;
1761 }
1762
1763 /* Add in the fixed length from the table */
1764
1765 code += _pcre_OP_lengths[c];
1766
1767 /* In UTF-8 mode, opcodes that are followed by a character may be followed by
1768 a multi-byte character. The length in the table is a minimum, so we have to
1769 arrange to skip the extra bytes. */
1770
1771 #ifdef SUPPORT_UTF8
1772 if (utf8) switch(c)
1773 {
1774 case OP_CHAR:
1775 case OP_CHARI:
1776 case OP_EXACT:
1777 case OP_EXACTI:
1778 case OP_UPTO:
1779 case OP_UPTOI:
1780 case OP_MINUPTO:
1781 case OP_MINUPTOI:
1782 case OP_POSUPTO:
1783 case OP_POSUPTOI:
1784 case OP_STAR:
1785 case OP_STARI:
1786 case OP_MINSTAR:
1787 case OP_MINSTARI:
1788 case OP_POSSTAR:
1789 case OP_POSSTARI:
1790 case OP_PLUS:
1791 case OP_PLUSI:
1792 case OP_MINPLUS:
1793 case OP_MINPLUSI:
1794 case OP_POSPLUS:
1795 case OP_POSPLUSI:
1796 case OP_QUERY:
1797 case OP_QUERYI:
1798 case OP_MINQUERY:
1799 case OP_MINQUERYI:
1800 case OP_POSQUERY:
1801 case OP_POSQUERYI:
1802 if (code[-1] >= 0xc0) code += _pcre_utf8_table4[code[-1] & 0x3f];
1803 break;
1804 }
1805 #else
1806 (void)(utf8); /* Keep compiler happy by referencing function argument */
1807 #endif
1808 }
1809 }
1810 }
1811
1812
1813
1814 /*************************************************
1815 * Scan compiled regex for recursion reference *
1816 *************************************************/
1817
1818 /* This little function scans through a compiled pattern until it finds an
1819 instance of OP_RECURSE.
1820
1821 Arguments:
1822 code points to start of expression
1823 utf8 TRUE in UTF-8 mode
1824
1825 Returns: pointer to the opcode for OP_RECURSE, or NULL if not found
1826 */
1827
1828 static const uschar *
1829 find_recurse(const uschar *code, BOOL utf8)
1830 {
1831 for (;;)
1832 {
1833 register int c = *code;
1834 if (c == OP_END) return NULL;
1835 if (c == OP_RECURSE) return code;
1836
1837 /* XCLASS is used for classes that cannot be represented just by a bit
1838 map. This includes negated single high-valued characters. The length in
1839 the table is zero; the actual length is stored in the compiled code. */
1840
1841 if (c == OP_XCLASS) code += GET(code, 1);
1842
1843 /* Otherwise, we can get the item's length from the table, except that for
1844 repeated character types, we have to test for \p and \P, which have an extra
1845 two bytes of parameters, and for MARK/PRUNE/SKIP/THEN with an argument, we
1846 must add in its length. */
1847
1848 else
1849 {
1850 switch(c)
1851 {
1852 case OP_TYPESTAR:
1853 case OP_TYPEMINSTAR:
1854 case OP_TYPEPLUS:
1855 case OP_TYPEMINPLUS:
1856 case OP_TYPEQUERY:
1857 case OP_TYPEMINQUERY:
1858 case OP_TYPEPOSSTAR:
1859 case OP_TYPEPOSPLUS:
1860 case OP_TYPEPOSQUERY:
1861 if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
1862 break;
1863
1864 case OP_TYPEPOSUPTO:
1865 case OP_TYPEUPTO:
1866 case OP_TYPEMINUPTO:
1867 case OP_TYPEEXACT:
1868 if (code[3] == OP_PROP || code[3] == OP_NOTPROP) code += 2;
1869 break;
1870
1871 case OP_MARK:
1872 case OP_PRUNE_ARG:
1873 case OP_SKIP_ARG:
1874 code += code[1];
1875 break;
1876
1877 case OP_THEN_ARG:
1878 code += code[1+LINK_SIZE];
1879 break;
1880 }
1881
1882 /* Add in the fixed length from the table */
1883
1884 code += _pcre_OP_lengths[c];
1885
1886 /* In UTF-8 mode, opcodes that are followed by a character may be followed
1887 by a multi-byte character. The length in the table is a minimum, so we have
1888 to arrange to skip the extra bytes. */
1889
1890 #ifdef SUPPORT_UTF8
1891 if (utf8) switch(c)
1892 {
1893 case OP_CHAR:
1894 case OP_CHARI:
1895 case OP_EXACT:
1896 case OP_EXACTI:
1897 case OP_UPTO:
1898 case OP_UPTOI:
1899 case OP_MINUPTO:
1900 case OP_MINUPTOI:
1901 case OP_POSUPTO:
1902 case OP_POSUPTOI:
1903 case OP_STAR:
1904 case OP_STARI:
1905 case OP_MINSTAR:
1906 case OP_MINSTARI:
1907 case OP_POSSTAR:
1908 case OP_POSSTARI:
1909 case OP_PLUS:
1910 case OP_PLUSI:
1911 case OP_MINPLUS:
1912 case OP_MINPLUSI:
1913 case OP_POSPLUS:
1914 case OP_POSPLUSI:
1915 case OP_QUERY:
1916 case OP_QUERYI:
1917 case OP_MINQUERY:
1918 case OP_MINQUERYI:
1919 case OP_POSQUERY:
1920 case OP_POSQUERYI:
1921 if (code[-1] >= 0xc0) code += _pcre_utf8_table4[code[-1] & 0x3f];
1922 break;
1923 }
1924 #else
1925 (void)(utf8); /* Keep compiler happy by referencing function argument */
1926 #endif
1927 }
1928 }
1929 }
1930
1931
1932
1933 /*************************************************
1934 * Scan compiled branch for non-emptiness *
1935 *************************************************/
1936
1937 /* This function scans through a branch of a compiled pattern to see whether it
1938 can match the empty string or not. It is called from could_be_empty()
1939 below and from compile_branch() when checking for an unlimited repeat of a
1940 group that can match nothing. Note that first_significant_code() skips over
1941 backward and negative forward assertions when its final argument is TRUE. If we
1942 hit an unclosed bracket, we return "empty" - this means we've struck an inner
1943 bracket whose current branch will already have been scanned.
1944
1945 Arguments:
1946 code points to start of search
1947 endcode points to where to stop
1948 utf8 TRUE if in UTF8 mode
1949 cd contains pointers to tables etc.
1950
1951 Returns: TRUE if what is matched could be empty
1952 */
1953
1954 static BOOL
1955 could_be_empty_branch(const uschar *code, const uschar *endcode, BOOL utf8,
1956 compile_data *cd)
1957 {
1958 register int c;
1959 for (code = first_significant_code(code + _pcre_OP_lengths[*code], TRUE);
1960 code < endcode;
1961 code = first_significant_code(code + _pcre_OP_lengths[c], TRUE))
1962 {
1963 const uschar *ccode;
1964
1965 c = *code;
1966
1967 /* Skip over forward assertions; the other assertions are skipped by
1968 first_significant_code() with a TRUE final argument. */
1969
1970 if (c == OP_ASSERT)
1971 {
1972 do code += GET(code, 1); while (*code == OP_ALT);
1973 c = *code;
1974 continue;
1975 }
1976
1977 /* For a recursion/subroutine call, if its end has been reached, which
1978 implies a backward reference subroutine call, we can scan it. If it's a
1979 forward reference subroutine call, we can't. To detect forward reference
1980 we have to scan up the list that is kept in the workspace. This function is
1981 called only when doing the real compile, not during the pre-compile that
1982 measures the size of the compiled pattern. */
1983
1984 if (c == OP_RECURSE)
1985 {
1986 const uschar *scode;
1987 BOOL empty_branch;
1988
1989 /* Test for forward reference */
1990
1991 for (scode = cd->start_workspace; scode < cd->hwm; scode += LINK_SIZE)
1992 if (GET(scode, 0) == code + 1 - cd->start_code) return TRUE;
1993
1994 /* Not a forward reference, test for completed backward reference */
1995
1996 empty_branch = FALSE;
1997 scode = cd->start_code + GET(code, 1);
1998 if (GET(scode, 1) == 0) return TRUE; /* Unclosed */
1999
2000 /* Completed backwards reference */
2001
2002 do
2003 {
2004 if (could_be_empty_branch(scode, endcode, utf8, cd))
2005 {
2006 empty_branch = TRUE;
2007 break;
2008 }
2009 scode += GET(scode, 1);
2010 }
2011 while (*scode == OP_ALT);
2012
2013 if (!empty_branch) return FALSE; /* All branches are non-empty */
2014 continue;
2015 }
2016
2017 /* Groups with zero repeats can of course be empty; skip them. */
2018
2019 if (c == OP_BRAZERO || c == OP_BRAMINZERO || c == OP_SKIPZERO ||
2020 c == OP_BRAPOSZERO)
2021 {
2022 code += _pcre_OP_lengths[c];
2023 do code += GET(code, 1); while (*code == OP_ALT);
2024 c = *code;
2025 continue;
2026 }
2027
2028 /* A nested group that is already marked as "could be empty" can just be
2029 skipped. */
2030
2031 if (c == OP_SBRA || c == OP_SBRAPOS ||
2032 c == OP_SCBRA || c == OP_SCBRAPOS)
2033 {
2034 do code += GET(code, 1); while (*code == OP_ALT);
2035 c = *code;
2036 continue;
2037 }
2038
2039 /* For other groups, scan the branches. */
2040
2041 if (c == OP_BRA || c == OP_BRAPOS ||
2042 c == OP_CBRA || c == OP_CBRAPOS ||
2043 c == OP_ONCE || c == OP_COND)
2044 {
2045 BOOL empty_branch;
2046 if (GET(code, 1) == 0) return TRUE; /* Hit unclosed bracket */
2047
2048 /* If a conditional group has only one branch, there is a second, implied,
2049 empty branch, so just skip over the conditional, because it could be empty.
2050 Otherwise, scan the individual branches of the group. */
2051
2052 if (c == OP_COND && code[GET(code, 1)] != OP_ALT)
2053 code += GET(code, 1);
2054 else
2055 {
2056 empty_branch = FALSE;
2057 do
2058 {
2059 if (!empty_branch && could_be_empty_branch(code, endcode, utf8, cd))
2060 empty_branch = TRUE;
2061 code += GET(code, 1);
2062 }
2063 while (*code == OP_ALT);
2064 if (!empty_branch) return FALSE; /* All branches are non-empty */
2065 }
2066
2067 c = *code;
2068 continue;
2069 }
2070
2071 /* Handle the other opcodes */
2072
2073 switch (c)
2074 {
2075 /* Check for quantifiers after a class. XCLASS is used for classes that
2076 cannot be represented just by a bit map. This includes negated single
2077 high-valued characters. The length in _pcre_OP_lengths[] is zero; the
2078 actual length is stored in the compiled code, so we must update "code"
2079 here. */
2080
2081 #ifdef SUPPORT_UTF8
2082 case OP_XCLASS:
2083 ccode = code += GET(code, 1);
2084 goto CHECK_CLASS_REPEAT;
2085 #endif
2086
2087 case OP_CLASS:
2088 case OP_NCLASS:
2089 ccode = code + 33;
2090
2091 #ifdef SUPPORT_UTF8
2092 CHECK_CLASS_REPEAT:
2093 #endif
2094
2095 switch (*ccode)
2096 {
2097 case OP_CRSTAR: /* These could be empty; continue */
2098 case OP_CRMINSTAR:
2099 case OP_CRQUERY:
2100 case OP_CRMINQUERY:
2101 break;
2102
2103 default: /* Non-repeat => class must match */
2104 case OP_CRPLUS: /* These repeats aren't empty */
2105 case OP_CRMINPLUS:
2106 return FALSE;
2107
2108 case OP_CRRANGE:
2109 case OP_CRMINRANGE:
2110 if (GET2(ccode, 1) > 0) return FALSE; /* Minimum > 0 */
2111 break;
2112 }
2113 break;
2114
2115 /* Opcodes that must match a character */
2116
2117 case OP_PROP:
2118 case OP_NOTPROP:
2119 case OP_EXTUNI:
2120 case OP_NOT_DIGIT:
2121 case OP_DIGIT:
2122 case OP_NOT_WHITESPACE:
2123 case OP_WHITESPACE:
2124 case OP_NOT_WORDCHAR:
2125 case OP_WORDCHAR:
2126 case OP_ANY:
2127 case OP_ALLANY:
2128 case OP_ANYBYTE:
2129 case OP_CHAR:
2130 case OP_CHARI:
2131 case OP_NOT:
2132 case OP_NOTI:
2133 case OP_PLUS:
2134 case OP_MINPLUS:
2135 case OP_POSPLUS:
2136 case OP_EXACT:
2137 case OP_NOTPLUS:
2138 case OP_NOTMINPLUS:
2139 case OP_NOTPOSPLUS:
2140 case OP_NOTEXACT:
2141 case OP_TYPEPLUS:
2142 case OP_TYPEMINPLUS:
2143 case OP_TYPEPOSPLUS:
2144 case OP_TYPEEXACT:
2145 return FALSE;
2146
2147 /* These are going to continue, as they may be empty, but we have to
2148 fudge the length for the \p and \P cases. */
2149
2150 case OP_TYPESTAR:
2151 case OP_TYPEMINSTAR:
2152 case OP_TYPEPOSSTAR:
2153 case OP_TYPEQUERY:
2154 case OP_TYPEMINQUERY:
2155 case OP_TYPEPOSQUERY:
2156 if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
2157 break;
2158
2159 /* Same for these */
2160
2161 case OP_TYPEUPTO:
2162 case OP_TYPEMINUPTO:
2163 case OP_TYPEPOSUPTO:
2164 if (code[3] == OP_PROP || code[3] == OP_NOTPROP) code += 2;
2165 break;
2166
2167 /* End of branch */
2168
2169 case OP_KET:
2170 case OP_KETRMAX:
2171 case OP_KETRMIN:
2172 case OP_KETRPOS:
2173 case OP_ALT:
2174 return TRUE;
2175
2176 /* In UTF-8 mode, STAR, MINSTAR, POSSTAR, QUERY, MINQUERY, POSQUERY, UPTO,
2177 MINUPTO, and POSUPTO may be followed by a multibyte character */
2178
2179 #ifdef SUPPORT_UTF8
2180 case OP_STAR:
2181 case OP_STARI:
2182 case OP_MINSTAR:
2183 case OP_MINSTARI:
2184 case OP_POSSTAR:
2185 case OP_POSSTARI:
2186 case OP_QUERY:
2187 case OP_QUERYI:
2188 case OP_MINQUERY:
2189 case OP_MINQUERYI:
2190 case OP_POSQUERY:
2191 case OP_POSQUERYI:
2192 if (utf8 && code[1] >= 0xc0) code += _pcre_utf8_table4[code[1] & 0x3f];
2193 break;
2194
2195 case OP_UPTO:
2196 case OP_UPTOI:
2197 case OP_MINUPTO:
2198 case OP_MINUPTOI:
2199 case OP_POSUPTO:
2200 case OP_POSUPTOI:
2201 if (utf8 && code[3] >= 0xc0) code += _pcre_utf8_table4[code[3] & 0x3f];
2202 break;
2203 #endif
2204
2205 /* MARK, and PRUNE/SKIP/THEN with an argument must skip over the argument
2206 string. */
2207
2208 case OP_MARK:
2209 case OP_PRUNE_ARG:
2210 case OP_SKIP_ARG:
2211 code += code[1];
2212 break;
2213
2214 case OP_THEN_ARG:
2215 code += code[1+LINK_SIZE];
2216 break;
2217
2218 /* None of the remaining opcodes are required to match a character. */
2219
2220 default:
2221 break;
2222 }
2223 }
2224
2225 return TRUE;
2226 }
2227
2228
2229
2230 /*************************************************
2231 * Scan compiled regex for non-emptiness *
2232 *************************************************/
2233
2234 /* This function is called to check for left recursive calls. We want to check
2235 the current branch of the current pattern to see if it could match the empty
2236 string. If it could, we must look outwards for branches at other levels,
2237 stopping when we pass beyond the bracket which is the subject of the recursion.
2238 This function is called only during the real compile, not during the
2239 pre-compile.
2240
2241 Arguments:
2242 code points to start of the recursion
2243 endcode points to where to stop (current RECURSE item)
2244 bcptr points to the chain of current (unclosed) branch starts
2245 utf8 TRUE if in UTF-8 mode
2246 cd pointers to tables etc
2247
2248 Returns: TRUE if what is matched could be empty
2249 */
2250
2251 static BOOL
2252 could_be_empty(const uschar *code, const uschar *endcode, branch_chain *bcptr,
2253 BOOL utf8, compile_data *cd)
2254 {
2255 while (bcptr != NULL && bcptr->current_branch >= code)
2256 {
2257 if (!could_be_empty_branch(bcptr->current_branch, endcode, utf8, cd))
2258 return FALSE;
2259 bcptr = bcptr->outer;
2260 }
2261 return TRUE;
2262 }
2263
2264
2265
2266 /*************************************************
2267 * Check for POSIX class syntax *
2268 *************************************************/
2269
2270 /* This function is called when the sequence "[:" or "[." or "[=" is
2271 encountered in a character class. It checks whether this is followed by a
2272 sequence of characters terminated by a matching ":]" or ".]" or "=]". If we
2273 reach an unescaped ']' without the special preceding character, return FALSE.
2274
2275 Originally, this function only recognized a sequence of letters between the
2276 terminators, but it seems that Perl recognizes any sequence of characters,
2277 though of course unknown POSIX names are subsequently rejected. Perl gives an
2278 "Unknown POSIX class" error for [:f\oo:] for example, where previously PCRE
2279 didn't consider this to be a POSIX class. Likewise for [:1234:].
2280
2281 The problem in trying to be exactly like Perl is in the handling of escapes. We
2282 have to be sure that [abc[:x\]pqr] is *not* treated as containing a POSIX
2283 class, but [abc[:x\]pqr:]] is (so that an error can be generated). The code
2284 below handles the special case of \], but does not try to do any other escape
2285 processing. This makes it different from Perl for cases such as [:l\ower:]
2286 where Perl recognizes it as the POSIX class "lower" but PCRE does not recognize
2287 "l\ower". This is a lesser evil that not diagnosing bad classes when Perl does,
2288 I think.
2289
2290 Arguments:
2291 ptr pointer to the initial [
2292 endptr where to return the end pointer
2293
2294 Returns: TRUE or FALSE
2295 */
2296
2297 static BOOL
2298 check_posix_syntax(const uschar *ptr, const uschar **endptr)
2299 {
2300 int terminator; /* Don't combine these lines; the Solaris cc */
2301 terminator = *(++ptr); /* compiler warns about "non-constant" initializer. */
2302 for (++ptr; *ptr != 0; ptr++)
2303 {
2304 if (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET) ptr++; else
2305 {
2306 if (*ptr == CHAR_RIGHT_SQUARE_BRACKET) return FALSE;
2307 if (*ptr == terminator && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET)
2308 {
2309 *endptr = ptr;
2310 return TRUE;
2311 }
2312 }
2313 }
2314 return FALSE;
2315 }
2316
2317
2318
2319
2320 /*************************************************
2321 * Check POSIX class name *
2322 *************************************************/
2323
2324 /* This function is called to check the name given in a POSIX-style class entry
2325 such as [:alnum:].
2326
2327 Arguments:
2328 ptr points to the first letter
2329 len the length of the name
2330
2331 Returns: a value representing the name, or -1 if unknown
2332 */
2333
2334 static int
2335 check_posix_name(const uschar *ptr, int len)
2336 {
2337 const char *pn = posix_names;
2338 register int yield = 0;
2339 while (posix_name_lengths[yield] != 0)
2340 {
2341 if (len == posix_name_lengths[yield] &&
2342 strncmp((const char *)ptr, pn, len) == 0) return yield;
2343 pn += posix_name_lengths[yield] + 1;
2344 yield++;
2345 }
2346 return -1;
2347 }
2348
2349
2350 /*************************************************
2351 * Adjust OP_RECURSE items in repeated group *
2352 *************************************************/
2353
2354 /* OP_RECURSE items contain an offset from the start of the regex to the group
2355 that is referenced. This means that groups can be replicated for fixed
2356 repetition simply by copying (because the recursion is allowed to refer to
2357 earlier groups that are outside the current group). However, when a group is
2358 optional (i.e. the minimum quantifier is zero), OP_BRAZERO or OP_SKIPZERO is
2359 inserted before it, after it has been compiled. This means that any OP_RECURSE
2360 items within it that refer to the group itself or any contained groups have to
2361 have their offsets adjusted. That one of the jobs of this function. Before it
2362 is called, the partially compiled regex must be temporarily terminated with
2363 OP_END.
2364
2365 This function has been extended with the possibility of forward references for
2366 recursions and subroutine calls. It must also check the list of such references
2367 for the group we are dealing with. If it finds that one of the recursions in
2368 the current group is on this list, it adjusts the offset in the list, not the
2369 value in the reference (which is a group number).
2370
2371 Arguments:
2372 group points to the start of the group
2373 adjust the amount by which the group is to be moved
2374 utf8 TRUE in UTF-8 mode
2375 cd contains pointers to tables etc.
2376 save_hwm the hwm forward reference pointer at the start of the group
2377
2378 Returns: nothing
2379 */
2380
2381 static void
2382 adjust_recurse(uschar *group, int adjust, BOOL utf8, compile_data *cd,
2383 uschar *save_hwm)
2384 {
2385 uschar *ptr = group;
2386
2387 while ((ptr = (uschar *)find_recurse(ptr, utf8)) != NULL)
2388 {
2389 int offset;
2390 uschar *hc;
2391
2392 /* See if this recursion is on the forward reference list. If so, adjust the
2393 reference. */
2394
2395 for (hc = save_hwm; hc < cd->hwm; hc += LINK_SIZE)
2396 {
2397 offset = GET(hc, 0);
2398 if (cd->start_code + offset == ptr + 1)
2399 {
2400 PUT(hc, 0, offset + adjust);
2401 break;
2402 }
2403 }
2404
2405 /* Otherwise, adjust the recursion offset if it's after the start of this
2406 group. */
2407
2408 if (hc >= cd->hwm)
2409 {
2410 offset = GET(ptr, 1);
2411 if (cd->start_code + offset >= group) PUT(ptr, 1, offset + adjust);
2412 }
2413
2414 ptr += 1 + LINK_SIZE;
2415 }
2416 }
2417
2418
2419
2420 /*************************************************
2421 * Insert an automatic callout point *
2422 *************************************************/
2423
2424 /* This function is called when the PCRE_AUTO_CALLOUT option is set, to insert
2425 callout points before each pattern item.
2426
2427 Arguments:
2428 code current code pointer
2429 ptr current pattern pointer
2430 cd pointers to tables etc
2431
2432 Returns: new code pointer
2433 */
2434
2435 static uschar *
2436 auto_callout(uschar *code, const uschar *ptr, compile_data *cd)
2437 {
2438 *code++ = OP_CALLOUT;
2439 *code++ = 255;
2440 PUT(code, 0, (int)(ptr - cd->start_pattern)); /* Pattern offset */
2441 PUT(code, LINK_SIZE, 0); /* Default length */
2442 return code + 2*LINK_SIZE;
2443 }
2444
2445
2446
2447 /*************************************************
2448 * Complete a callout item *
2449 *************************************************/
2450
2451 /* A callout item contains the length of the next item in the pattern, which
2452 we can't fill in till after we have reached the relevant point. This is used
2453 for both automatic and manual callouts.
2454
2455 Arguments:
2456 previous_callout points to previous callout item
2457 ptr current pattern pointer
2458 cd pointers to tables etc
2459
2460 Returns: nothing
2461 */
2462
2463 static void
2464 complete_callout(uschar *previous_callout, const uschar *ptr, compile_data *cd)
2465 {
2466 int length = (int)(ptr - cd->start_pattern - GET(previous_callout, 2));
2467 PUT(previous_callout, 2 + LINK_SIZE, length);
2468 }
2469
2470
2471
2472 #ifdef SUPPORT_UCP
2473 /*************************************************
2474 * Get othercase range *
2475 *************************************************/
2476
2477 /* This function is passed the start and end of a class range, in UTF-8 mode
2478 with UCP support. It searches up the characters, looking for internal ranges of
2479 characters in the "other" case. Each call returns the next one, updating the
2480 start address.
2481
2482 Arguments:
2483 cptr points to starting character value; updated
2484 d end value
2485 ocptr where to put start of othercase range
2486 odptr where to put end of othercase range
2487
2488 Yield: TRUE when range returned; FALSE when no more
2489 */
2490
2491 static BOOL
2492 get_othercase_range(unsigned int *cptr, unsigned int d, unsigned int *ocptr,
2493 unsigned int *odptr)
2494 {
2495 unsigned int c, othercase, next;
2496
2497 for (c = *cptr; c <= d; c++)
2498 { if ((othercase = UCD_OTHERCASE(c)) != c) break; }
2499
2500 if (c > d) return FALSE;
2501
2502 *ocptr = othercase;
2503 next = othercase + 1;
2504
2505 for (++c; c <= d; c++)
2506 {
2507 if (UCD_OTHERCASE(c) != next) break;
2508 next++;
2509 }
2510
2511 *odptr = next - 1;
2512 *cptr = c;
2513
2514 return TRUE;
2515 }
2516
2517
2518
2519 /*************************************************
2520 * Check a character and a property *
2521 *************************************************/
2522
2523 /* This function is called by check_auto_possessive() when a property item
2524 is adjacent to a fixed character.
2525
2526 Arguments:
2527 c the character
2528 ptype the property type
2529 pdata the data for the type
2530 negated TRUE if it's a negated property (\P or \p{^)
2531
2532 Returns: TRUE if auto-possessifying is OK
2533 */
2534
2535 static BOOL
2536 check_char_prop(int c, int ptype, int pdata, BOOL negated)
2537 {
2538 const ucd_record *prop = GET_UCD(c);
2539 switch(ptype)
2540 {
2541 case PT_LAMP:
2542 return (prop->chartype == ucp_Lu ||
2543 prop->chartype == ucp_Ll ||
2544 prop->chartype == ucp_Lt) == negated;
2545
2546 case PT_GC:
2547 return (pdata == _pcre_ucp_gentype[prop->chartype]) == negated;
2548
2549 case PT_PC:
2550 return (pdata == prop->chartype) == negated;
2551
2552 case PT_SC:
2553 return (pdata == prop->script) == negated;
2554
2555 /* These are specials */
2556
2557 case PT_ALNUM:
2558 return (_pcre_ucp_gentype[prop->chartype] == ucp_L ||
2559 _pcre_ucp_gentype[prop->chartype] == ucp_N) == negated;
2560
2561 case PT_SPACE: /* Perl space */
2562 return (_pcre_ucp_gentype[prop->chartype] == ucp_Z ||
2563 c == CHAR_HT || c == CHAR_NL || c == CHAR_FF || c == CHAR_CR)
2564 == negated;
2565
2566 case PT_PXSPACE: /* POSIX space */
2567 return (_pcre_ucp_gentype[prop->chartype] == ucp_Z ||
2568 c == CHAR_HT || c == CHAR_NL || c == CHAR_VT ||
2569 c == CHAR_FF || c == CHAR_CR)
2570 == negated;
2571
2572 case PT_WORD:
2573 return (_pcre_ucp_gentype[prop->chartype] == ucp_L ||
2574 _pcre_ucp_gentype[prop->chartype] == ucp_N ||
2575 c == CHAR_UNDERSCORE) == negated;
2576 }
2577 return FALSE;
2578 }
2579 #endif /* SUPPORT_UCP */
2580
2581
2582
2583 /*************************************************
2584 * Check if auto-possessifying is possible *
2585 *************************************************/
2586
2587 /* This function is called for unlimited repeats of certain items, to see
2588 whether the next thing could possibly match the repeated item. If not, it makes
2589 sense to automatically possessify the repeated item.
2590
2591 Arguments:
2592 previous pointer to the repeated opcode
2593 utf8 TRUE in UTF-8 mode
2594 ptr next character in pattern
2595 options options bits
2596 cd contains pointers to tables etc.
2597
2598 Returns: TRUE if possessifying is wanted
2599 */
2600
2601 static BOOL
2602 check_auto_possessive(const uschar *previous, BOOL utf8, const uschar *ptr,
2603 int options, compile_data *cd)
2604 {
2605 int c, next;
2606 int op_code = *previous++;
2607
2608 /* Skip whitespace and comments in extended mode */
2609
2610 if ((options & PCRE_EXTENDED) != 0)
2611 {
2612 for (;;)
2613 {
2614 while ((cd->ctypes[*ptr] & ctype_space) != 0) ptr++;
2615 if (*ptr == CHAR_NUMBER_SIGN)
2616 {
2617 ptr++;
2618 while (*ptr != 0)
2619 {
2620 if (IS_NEWLINE(ptr)) { ptr += cd->nllen; break; }
2621 ptr++;
2622 #ifdef SUPPORT_UTF8
2623 if (utf8) while ((*ptr & 0xc0) == 0x80) ptr++;
2624 #endif
2625 }
2626 }
2627 else break;
2628 }
2629 }
2630
2631 /* If the next item is one that we can handle, get its value. A non-negative
2632 value is a character, a negative value is an escape value. */
2633
2634 if (*ptr == CHAR_BACKSLASH)
2635 {
2636 int temperrorcode = 0;
2637 next = check_escape(&ptr, &temperrorcode, cd->bracount, options, FALSE);
2638 if (temperrorcode != 0) return FALSE;
2639 ptr++; /* Point after the escape sequence */
2640 }
2641
2642 else if ((cd->ctypes[*ptr] & ctype_meta) == 0)
2643 {
2644 #ifdef SUPPORT_UTF8
2645 if (utf8) { GETCHARINC(next, ptr); } else
2646 #endif
2647 next = *ptr++;
2648 }
2649
2650 else return FALSE;
2651
2652 /* Skip whitespace and comments in extended mode */
2653
2654 if ((options & PCRE_EXTENDED) != 0)
2655 {
2656 for (;;)
2657 {
2658 while ((cd->ctypes[*ptr] & ctype_space) != 0) ptr++;
2659 if (*ptr == CHAR_NUMBER_SIGN)
2660 {
2661 ptr++;
2662 while (*ptr != 0)
2663 {
2664 if (IS_NEWLINE(ptr)) { ptr += cd->nllen; break; }
2665 ptr++;
2666 #ifdef SUPPORT_UTF8
2667 if (utf8) while ((*ptr & 0xc0) == 0x80) ptr++;
2668 #endif
2669 }
2670 }
2671 else break;
2672 }
2673 }
2674
2675 /* If the next thing is itself optional, we have to give up. */
2676
2677 if (*ptr == CHAR_ASTERISK || *ptr == CHAR_QUESTION_MARK ||
2678 strncmp((char *)ptr, STR_LEFT_CURLY_BRACKET STR_0 STR_COMMA, 3) == 0)
2679 return FALSE;
2680
2681 /* Now compare the next item with the previous opcode. First, handle cases when
2682 the next item is a character. */
2683
2684 if (next >= 0) switch(op_code)
2685 {
2686 case OP_CHAR:
2687 #ifdef SUPPORT_UTF8
2688 GETCHARTEST(c, previous);
2689 #else
2690 c = *previous;
2691 #endif
2692 return c != next;
2693
2694 /* For CHARI (caseless character) we must check the other case. If we have
2695 Unicode property support, we can use it to test the other case of
2696 high-valued characters. */
2697
2698 case OP_CHARI:
2699 #ifdef SUPPORT_UTF8
2700 GETCHARTEST(c, previous);
2701 #else
2702 c = *previous;
2703 #endif
2704 if (c == next) return FALSE;
2705 #ifdef SUPPORT_UTF8
2706 if (utf8)
2707 {
2708 unsigned int othercase;
2709 if (next < 128) othercase = cd->fcc[next]; else
2710 #ifdef SUPPORT_UCP
2711 othercase = UCD_OTHERCASE((unsigned int)next);
2712 #else
2713 othercase = NOTACHAR;
2714 #endif
2715 return (unsigned int)c != othercase;
2716 }
2717 else
2718 #endif /* SUPPORT_UTF8 */
2719 return (c != cd->fcc[next]); /* Non-UTF-8 mode */
2720
2721 /* For OP_NOT and OP_NOTI, the data is always a single-byte character. These
2722 opcodes are not used for multi-byte characters, because they are coded using
2723 an XCLASS instead. */
2724
2725 case OP_NOT:
2726 return (c = *previous) == next;
2727
2728 case OP_NOTI:
2729 if ((c = *previous) == next) return TRUE;
2730 #ifdef SUPPORT_UTF8
2731 if (utf8)
2732 {
2733 unsigned int othercase;
2734 if (next < 128) othercase = cd->fcc[next]; else
2735 #ifdef SUPPORT_UCP
2736 othercase = UCD_OTHERCASE(next);
2737 #else
2738 othercase = NOTACHAR;
2739 #endif
2740 return (unsigned int)c == othercase;
2741 }
2742 else
2743 #endif /* SUPPORT_UTF8 */
2744 return (c == cd->fcc[next]); /* Non-UTF-8 mode */
2745
2746 /* Note that OP_DIGIT etc. are generated only when PCRE_UCP is *not* set.
2747 When it is set, \d etc. are converted into OP_(NOT_)PROP codes. */
2748
2749 case OP_DIGIT:
2750 return next > 127 || (cd->ctypes[next] & ctype_digit) == 0;
2751
2752 case OP_NOT_DIGIT:
2753 return next <= 127 && (cd->ctypes[next] & ctype_digit) != 0;
2754
2755 case OP_WHITESPACE:
2756 return next > 127 || (cd->ctypes[next] & ctype_space) == 0;
2757
2758 case OP_NOT_WHITESPACE:
2759 return next <= 127 && (cd->ctypes[next] & ctype_space) != 0;
2760
2761 case OP_WORDCHAR:
2762 return next > 127 || (cd->ctypes[next] & ctype_word) == 0;
2763
2764 case OP_NOT_WORDCHAR:
2765 return next <= 127 && (cd->ctypes[next] & ctype_word) != 0;
2766
2767 case OP_HSPACE:
2768 case OP_NOT_HSPACE:
2769 switch(next)
2770 {
2771 case 0x09:
2772 case 0x20:
2773 case 0xa0:
2774 case 0x1680:
2775 case 0x180e:
2776 case 0x2000:
2777 case 0x2001:
2778 case 0x2002:
2779 case 0x2003:
2780 case 0x2004:
2781 case 0x2005:
2782 case 0x2006:
2783 case 0x2007:
2784 case 0x2008:
2785 case 0x2009:
2786 case 0x200A:
2787 case 0x202f:
2788 case 0x205f:
2789 case 0x3000:
2790 return op_code == OP_NOT_HSPACE;
2791 default:
2792 return op_code != OP_NOT_HSPACE;
2793 }
2794
2795 case OP_ANYNL:
2796 case OP_VSPACE:
2797 case OP_NOT_VSPACE:
2798 switch(next)
2799 {
2800 case 0x0a:
2801 case 0x0b:
2802 case 0x0c:
2803 case 0x0d:
2804 case 0x85:
2805 case 0x2028:
2806 case 0x2029:
2807 return op_code == OP_NOT_VSPACE;
2808 default:
2809 return op_code != OP_NOT_VSPACE;
2810 }
2811
2812 #ifdef SUPPORT_UCP
2813 case OP_PROP:
2814 return check_char_prop(next, previous[0], previous[1], FALSE);
2815
2816 case OP_NOTPROP:
2817 return check_char_prop(next, previous[0], previous[1], TRUE);
2818 #endif
2819
2820 default:
2821 return FALSE;
2822 }
2823
2824
2825 /* Handle the case when the next item is \d, \s, etc. Note that when PCRE_UCP
2826 is set, \d turns into ESC_du rather than ESC_d, etc., so ESC_d etc. are
2827 generated only when PCRE_UCP is *not* set, that is, when only ASCII
2828 characteristics are recognized. Similarly, the opcodes OP_DIGIT etc. are
2829 replaced by OP_PROP codes when PCRE_UCP is set. */
2830
2831 switch(op_code)
2832 {
2833 case OP_CHAR:
2834 case OP_CHARI:
2835 #ifdef SUPPORT_UTF8
2836 GETCHARTEST(c, previous);
2837 #else
2838 c = *previous;
2839 #endif
2840 switch(-next)
2841 {
2842 case ESC_d:
2843 return c > 127 || (cd->ctypes[c] & ctype_digit) == 0;
2844
2845 case ESC_D:
2846 return c <= 127 && (cd->ctypes[c] & ctype_digit) != 0;
2847
2848 case ESC_s:
2849 return c > 127 || (cd->ctypes[c] & ctype_space) == 0;
2850
2851 case ESC_S:
2852 return c <= 127 && (cd->ctypes[c] & ctype_space) != 0;
2853
2854 case ESC_w:
2855 return c > 127 || (cd->ctypes[c] & ctype_word) == 0;
2856
2857 case ESC_W:
2858 return c <= 127 && (cd->ctypes[c] & ctype_word) != 0;
2859
2860 case ESC_h:
2861 case ESC_H:
2862 switch(c)
2863 {
2864 case 0x09:
2865 case 0x20:
2866 case 0xa0:
2867 case 0x1680:
2868 case 0x180e:
2869 case 0x2000:
2870 case 0x2001:
2871 case 0x2002:
2872 case 0x2003:
2873 case 0x2004:
2874 case 0x2005:
2875 case 0x2006:
2876 case 0x2007:
2877 case 0x2008:
2878 case 0x2009:
2879 case 0x200A:
2880 case 0x202f:
2881 case 0x205f:
2882 case 0x3000:
2883 return -next != ESC_h;
2884 default:
2885 return -next == ESC_h;
2886 }
2887
2888 case ESC_v:
2889 case ESC_V:
2890 switch(c)
2891 {
2892 case 0x0a:
2893 case 0x0b:
2894 case 0x0c:
2895 case 0x0d:
2896 case 0x85:
2897 case 0x2028:
2898 case 0x2029:
2899 return -next != ESC_v;
2900 default:
2901 return -next == ESC_v;
2902 }
2903
2904 /* When PCRE_UCP is set, these values get generated for \d etc. Find
2905 their substitutions and process them. The result will always be either
2906 -ESC_p or -ESC_P. Then fall through to process those values. */
2907
2908 #ifdef SUPPORT_UCP
2909 case ESC_du:
2910 case ESC_DU:
2911 case ESC_wu:
2912 case ESC_WU:
2913 case ESC_su:
2914 case ESC_SU:
2915 {
2916 int temperrorcode = 0;
2917 ptr = substitutes[-next - ESC_DU];
2918 next = check_escape(&ptr, &temperrorcode, 0, options, FALSE);
2919 if (temperrorcode != 0) return FALSE;
2920 ptr++; /* For compatibility */
2921 }
2922 /* Fall through */
2923
2924 case ESC_p:
2925 case ESC_P:
2926 {
2927 int ptype, pdata, errorcodeptr;
2928 BOOL negated;
2929
2930 ptr--; /* Make ptr point at the p or P */
2931 ptype = get_ucp(&ptr, &negated, &pdata, &errorcodeptr);
2932 if (ptype < 0) return FALSE;
2933 ptr++; /* Point past the final curly ket */
2934
2935 /* If the property item is optional, we have to give up. (When generated
2936 from \d etc by PCRE_UCP, this test will have been applied much earlier,
2937 to the original \d etc. At this point, ptr will point to a zero byte. */
2938
2939 if (*ptr == CHAR_ASTERISK || *ptr == CHAR_QUESTION_MARK ||
2940 strncmp((char *)ptr, STR_LEFT_CURLY_BRACKET STR_0 STR_COMMA, 3) == 0)
2941 return FALSE;
2942
2943 /* Do the property check. */
2944
2945 return check_char_prop(c, ptype, pdata, (next == -ESC_P) != negated);
2946 }
2947 #endif
2948
2949 default:
2950 return FALSE;
2951 }
2952
2953 /* In principle, support for Unicode properties should be integrated here as
2954 well. It means re-organizing the above code so as to get hold of the property
2955 values before switching on the op-code. However, I wonder how many patterns
2956 combine ASCII \d etc with Unicode properties? (Note that if PCRE_UCP is set,
2957 these op-codes are never generated.) */
2958
2959 case OP_DIGIT:
2960 return next == -ESC_D || next == -ESC_s || next == -ESC_W ||
2961 next == -ESC_h || next == -ESC_v || next == -ESC_R;
2962
2963 case OP_NOT_DIGIT:
2964 return next == -ESC_d;
2965
2966 case OP_WHITESPACE:
2967 return next == -ESC_S || next == -ESC_d || next == -ESC_w || next == -ESC_R;
2968
2969 case OP_NOT_WHITESPACE:
2970 return next == -ESC_s || next == -ESC_h || next == -ESC_v;
2971
2972 case OP_HSPACE:
2973 return next == -ESC_S || next == -ESC_H || next == -ESC_d ||
2974 next == -ESC_w || next == -ESC_v || next == -ESC_R;
2975
2976 case OP_NOT_HSPACE:
2977 return next == -ESC_h;
2978
2979 /* Can't have \S in here because VT matches \S (Perl anomaly) */
2980 case OP_ANYNL:
2981 case OP_VSPACE:
2982 return next == -ESC_V || next == -ESC_d || next == -ESC_w;
2983
2984 case OP_NOT_VSPACE:
2985 return next == -ESC_v || next == -ESC_R;
2986
2987 case OP_WORDCHAR:
2988 return next == -ESC_W || next == -ESC_s || next == -ESC_h ||
2989 next == -ESC_v || next == -ESC_R;
2990
2991 case OP_NOT_WORDCHAR:
2992 return next == -ESC_w || next == -ESC_d;
2993
2994 default:
2995 return FALSE;
2996 }
2997
2998 /* Control does not reach here */
2999 }
3000
3001
3002
3003 /*************************************************
3004 * Compile one branch *
3005 *************************************************/
3006
3007 /* Scan the pattern, compiling it into the a vector. If the options are
3008 changed during the branch, the pointer is used to change the external options
3009 bits. This function is used during the pre-compile phase when we are trying
3010 to find out the amount of memory needed, as well as during the real compile
3011 phase. The value of lengthptr distinguishes the two phases.
3012
3013 Arguments:
3014 optionsptr pointer to the option bits
3015 codeptr points to the pointer to the current code point
3016 ptrptr points to the current pattern pointer
3017 errorcodeptr points to error code variable
3018 firstbyteptr set to initial literal character, or < 0 (REQ_UNSET, REQ_NONE)
3019 reqbyteptr set to the last literal character required, else < 0
3020 bcptr points to current branch chain
3021 cd contains pointers to tables etc.
3022 lengthptr NULL during the real compile phase
3023 points to length accumulator during pre-compile phase
3024
3025 Returns: TRUE on success
3026 FALSE, with *errorcodeptr set non-zero on error
3027 */
3028
3029 static BOOL
3030 compile_branch(int *optionsptr, uschar **codeptr, const uschar **ptrptr,
3031 int *errorcodeptr, int *firstbyteptr, int *reqbyteptr, branch_chain *bcptr,
3032 compile_data *cd, int *lengthptr)
3033 {
3034 int repeat_type, op_type;
3035 int repeat_min = 0, repeat_max = 0; /* To please picky compilers */
3036 int bravalue = 0;
3037 int greedy_default, greedy_non_default;
3038 int firstbyte, reqbyte;
3039 int zeroreqbyte, zerofirstbyte;
3040 int req_caseopt, reqvary, tempreqvary;
3041 int options = *optionsptr;
3042 int after_manual_callout = 0;
3043 int length_prevgroup = 0;
3044 register int c;
3045 register uschar *code = *codeptr;
3046 uschar *last_code = code;
3047 uschar *orig_code = code;
3048 uschar *tempcode;
3049 BOOL inescq = FALSE;
3050 BOOL groupsetfirstbyte = FALSE;
3051 const uschar *ptr = *ptrptr;
3052 const uschar *tempptr;
3053 const uschar *nestptr = NULL;
3054 uschar *previous = NULL;
3055 uschar *previous_callout = NULL;
3056 uschar *save_hwm = NULL;
3057 uschar classbits[32];
3058
3059 #ifdef SUPPORT_UTF8
3060 BOOL class_utf8;
3061 BOOL utf8 = (options & PCRE_UTF8) != 0;
3062 uschar *class_utf8data;
3063 uschar *class_utf8data_base;
3064 uschar utf8_char[6];
3065 #else
3066 BOOL utf8 = FALSE;
3067 uschar *utf8_char = NULL;
3068 #endif
3069
3070 #ifdef PCRE_DEBUG
3071 if (lengthptr != NULL) DPRINTF((">> start branch\n"));
3072 #endif
3073
3074 /* Set up the default and non-default settings for greediness */
3075
3076 greedy_default = ((options & PCRE_UNGREEDY) != 0);
3077 greedy_non_default = greedy_default ^ 1;
3078
3079 /* Initialize no first byte, no required byte. REQ_UNSET means "no char
3080 matching encountered yet". It gets changed to REQ_NONE if we hit something that
3081 matches a non-fixed char first char; reqbyte just remains unset if we never
3082 find one.
3083
3084 When we hit a repeat whose minimum is zero, we may have to adjust these values
3085 to take the zero repeat into account. This is implemented by setting them to
3086 zerofirstbyte and zeroreqbyte when such a repeat is encountered. The individual
3087 item types that can be repeated set these backoff variables appropriately. */
3088
3089 firstbyte = reqbyte = zerofirstbyte = zeroreqbyte = REQ_UNSET;
3090
3091 /* The variable req_caseopt contains either the REQ_CASELESS value or zero,
3092 according to the current setting of the caseless flag. REQ_CASELESS is a bit
3093 value > 255. It is added into the firstbyte or reqbyte variables to record the
3094 case status of the value. This is used only for ASCII characters. */
3095
3096 req_caseopt = ((options & PCRE_CASELESS) != 0)? REQ_CASELESS : 0;
3097
3098 /* Switch on next character until the end of the branch */
3099
3100 for (;; ptr++)
3101 {
3102 BOOL negate_class;
3103 BOOL should_flip_negation;
3104 BOOL possessive_quantifier;
3105 BOOL is_quantifier;
3106 BOOL is_recurse;
3107 BOOL reset_bracount;
3108 int class_charcount;
3109 int class_lastchar;
3110 int newoptions;
3111 int recno;
3112 int refsign;
3113 int skipbytes;
3114 int subreqbyte;
3115 int subfirstbyte;
3116 int terminator;
3117 int mclength;
3118 uschar mcbuffer[8];
3119
3120 /* Get next byte in the pattern */
3121
3122 c = *ptr;
3123
3124 /* If we are at the end of a nested substitution, revert to the outer level
3125 string. Nesting only happens one level deep. */
3126
3127 if (c == 0 && nestptr != NULL)
3128 {
3129 ptr = nestptr;
3130 nestptr = NULL;
3131 c = *ptr;
3132 }
3133
3134 /* If we are in the pre-compile phase, accumulate the length used for the
3135 previous cycle of this loop. */
3136
3137 if (lengthptr != NULL)
3138 {
3139 #ifdef PCRE_DEBUG
3140 if (code > cd->hwm) cd->hwm = code; /* High water info */
3141 #endif
3142 if (code > cd->start_workspace + WORK_SIZE_CHECK) /* Check for overrun */
3143 {
3144 *errorcodeptr = ERR52;
3145 goto FAILED;
3146 }
3147
3148 /* There is at least one situation where code goes backwards: this is the
3149 case of a zero quantifier after a class (e.g. [ab]{0}). At compile time,
3150 the class is simply eliminated. However, it is created first, so we have to
3151 allow memory for it. Therefore, don't ever reduce the length at this point.
3152 */
3153
3154 if (code < last_code) code = last_code;
3155
3156 /* Paranoid check for integer overflow */
3157
3158 if (OFLOW_MAX - *lengthptr < code - last_code)
3159 {
3160 *errorcodeptr = ERR20;
3161 goto FAILED;
3162 }
3163
3164 *lengthptr += (int)(code - last_code);
3165 DPRINTF(("length=%d added %d c=%c\n", *lengthptr, code - last_code, c));
3166
3167 /* If "previous" is set and it is not at the start of the work space, move
3168 it back to there, in order to avoid filling up the work space. Otherwise,
3169 if "previous" is NULL, reset the current code pointer to the start. */
3170
3171 if (previous != NULL)
3172 {
3173 if (previous > orig_code)
3174 {
3175 memmove(orig_code, previous, code - previous);
3176 code -= previous - orig_code;
3177 previous = orig_code;
3178 }
3179 }
3180 else code = orig_code;
3181
3182 /* Remember where this code item starts so we can pick up the length
3183 next time round. */
3184
3185 last_code = code;
3186 }
3187
3188 /* In the real compile phase, just check the workspace used by the forward
3189 reference list. */
3190
3191 else if (cd->hwm > cd->start_workspace + WORK_SIZE_CHECK)
3192 {
3193 *errorcodeptr = ERR52;
3194 goto FAILED;
3195 }
3196
3197 /* If in \Q...\E, check for the end; if not, we have a literal */
3198
3199 if (inescq && c != 0)
3200 {
3201 if (c == CHAR_BACKSLASH && ptr[1] == CHAR_E)
3202 {
3203 inescq = FALSE;
3204 ptr++;
3205 continue;
3206 }
3207 else
3208 {
3209 if (previous_callout != NULL)
3210 {
3211 if (lengthptr == NULL) /* Don't attempt in pre-compile phase */
3212 complete_callout(previous_callout, ptr, cd);
3213 previous_callout = NULL;
3214 }
3215 if ((options & PCRE_AUTO_CALLOUT) != 0)
3216 {
3217 previous_callout = code;
3218 code = auto_callout(code, ptr, cd);
3219 }
3220 goto NORMAL_CHAR;
3221 }
3222 }
3223
3224 /* Fill in length of a previous callout, except when the next thing is
3225 a quantifier. */
3226
3227 is_quantifier =
3228 c == CHAR_ASTERISK || c == CHAR_PLUS || c == CHAR_QUESTION_MARK ||
3229 (c == CHAR_LEFT_CURLY_BRACKET && is_counted_repeat(ptr+1));
3230
3231 if (!is_quantifier && previous_callout != NULL &&
3232 after_manual_callout-- <= 0)
3233 {
3234 if (lengthptr == NULL) /* Don't attempt in pre-compile phase */
3235 complete_callout(previous_callout, ptr, cd);
3236 previous_callout = NULL;
3237 }
3238
3239 /* In extended mode, skip white space and comments */
3240
3241 if ((options & PCRE_EXTENDED) != 0)
3242 {
3243 if ((cd->ctypes[c] & ctype_space) != 0) continue;
3244 if (c == CHAR_NUMBER_SIGN)
3245 {
3246 ptr++;
3247 while (*ptr != 0)
3248 {
3249 if (IS_NEWLINE(ptr)) { ptr += cd->nllen - 1; break; }
3250 ptr++;
3251 #ifdef SUPPORT_UTF8
3252 if (utf8) while ((*ptr & 0xc0) == 0x80) ptr++;
3253 #endif
3254 }
3255 if (*ptr != 0) continue;
3256
3257 /* Else fall through to handle end of string */
3258 c = 0;
3259 }
3260 }
3261
3262 /* No auto callout for quantifiers. */
3263
3264 if ((options & PCRE_AUTO_CALLOUT) != 0 && !is_quantifier)
3265 {
3266 previous_callout = code;
3267 code = auto_callout(code, ptr, cd);
3268 }
3269
3270 switch(c)
3271 {
3272 /* ===================================================================*/
3273 case 0: /* The branch terminates at string end */
3274 case CHAR_VERTICAL_LINE: /* or | or ) */
3275 case CHAR_RIGHT_PARENTHESIS:
3276 *firstbyteptr = firstbyte;
3277 *reqbyteptr = reqbyte;
3278 *codeptr = code;
3279 *ptrptr = ptr;
3280 if (lengthptr != NULL)
3281 {
3282 if (OFLOW_MAX - *lengthptr < code - last_code)
3283 {
3284 *errorcodeptr = ERR20;
3285 goto FAILED;
3286 }
3287 *lengthptr += (int)(code - last_code); /* To include callout length */
3288 DPRINTF((">> end branch\n"));
3289 }
3290 return TRUE;
3291
3292
3293 /* ===================================================================*/
3294 /* Handle single-character metacharacters. In multiline mode, ^ disables
3295 the setting of any following char as a first character. */
3296
3297 case CHAR_CIRCUMFLEX_ACCENT:
3298 previous = NULL;
3299 if ((options & PCRE_MULTILINE) != 0)
3300 {
3301 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
3302 *code++ = OP_CIRCM;
3303 }
3304 else *code++ = OP_CIRC;
3305 break;
3306
3307 case CHAR_DOLLAR_SIGN:
3308 previous = NULL;
3309 *code++ = ((options & PCRE_MULTILINE) != 0)? OP_DOLLM : OP_DOLL;
3310 break;
3311
3312 /* There can never be a first char if '.' is first, whatever happens about
3313 repeats. The value of reqbyte doesn't change either. */
3314
3315 case CHAR_DOT:
3316 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
3317 zerofirstbyte = firstbyte;
3318 zeroreqbyte = reqbyte;
3319 previous = code;
3320 *code++ = ((options & PCRE_DOTALL) != 0)? OP_ALLANY: OP_ANY;
3321 break;
3322
3323
3324 /* ===================================================================*/
3325 /* Character classes. If the included characters are all < 256, we build a
3326 32-byte bitmap of the permitted characters, except in the special case
3327 where there is only one such character. For negated classes, we build the
3328 map as usual, then invert it at the end. However, we use a different opcode
3329 so that data characters > 255 can be handled correctly.
3330
3331 If the class contains characters outside the 0-255 range, a different
3332 opcode is compiled. It may optionally have a bit map for characters < 256,
3333 but those above are are explicitly listed afterwards. A flag byte tells
3334 whether the bitmap is present, and whether this is a negated class or not.
3335
3336 In JavaScript compatibility mode, an isolated ']' causes an error. In
3337 default (Perl) mode, it is treated as a data character. */
3338
3339 case CHAR_RIGHT_SQUARE_BRACKET:
3340 if ((cd->external_options & PCRE_JAVASCRIPT_COMPAT) != 0)
3341 {
3342 *errorcodeptr = ERR64;
3343 goto FAILED;
3344 }
3345 goto NORMAL_CHAR;
3346
3347 case CHAR_LEFT_SQUARE_BRACKET:
3348 previous = code;
3349
3350 /* PCRE supports POSIX class stuff inside a class. Perl gives an error if
3351 they are encountered at the top level, so we'll do that too. */
3352
3353 if ((ptr[1] == CHAR_COLON || ptr[1] == CHAR_DOT ||
3354 ptr[1] == CHAR_EQUALS_SIGN) &&
3355 check_posix_syntax(ptr, &tempptr))
3356 {
3357 *errorcodeptr = (ptr[1] == CHAR_COLON)? ERR13 : ERR31;
3358 goto FAILED;
3359 }
3360
3361 /* If the first character is '^', set the negation flag and skip it. Also,
3362 if the first few characters (either before or after ^) are \Q\E or \E we
3363 skip them too. This makes for compatibility with Perl. */
3364
3365 negate_class = FALSE;
3366 for (;;)
3367 {
3368 c = *(++ptr);
3369 if (c == CHAR_BACKSLASH)
3370 {
3371 if (ptr[1] == CHAR_E)
3372 ptr++;
3373 else if (strncmp((const char *)ptr+1,
3374 STR_Q STR_BACKSLASH STR_E, 3) == 0)
3375 ptr += 3;
3376 else
3377 break;
3378 }
3379 else if (!negate_class && c == CHAR_CIRCUMFLEX_ACCENT)
3380 negate_class = TRUE;
3381 else break;
3382 }
3383
3384 /* Empty classes are allowed in JavaScript compatibility mode. Otherwise,
3385 an initial ']' is taken as a data character -- the code below handles
3386 that. In JS mode, [] must always fail, so generate OP_FAIL, whereas
3387 [^] must match any character, so generate OP_ALLANY. */
3388
3389 if (c == CHAR_RIGHT_SQUARE_BRACKET &&
3390 (cd->external_options & PCRE_JAVASCRIPT_COMPAT) != 0)
3391 {
3392 *code++ = negate_class? OP_ALLANY : OP_FAIL;
3393 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
3394 zerofirstbyte = firstbyte;
3395 break;
3396 }
3397
3398 /* If a class contains a negative special such as \S, we need to flip the
3399 negation flag at the end, so that support for characters > 255 works
3400 correctly (they are all included in the class). */
3401
3402 should_flip_negation = FALSE;
3403
3404 /* Keep a count of chars with values < 256 so that we can optimize the case
3405 of just a single character (as long as it's < 256). However, For higher
3406 valued UTF-8 characters, we don't yet do any optimization. */
3407
3408 class_charcount = 0;
3409 class_lastchar = -1;
3410
3411 /* Initialize the 32-char bit map to all zeros. We build the map in a
3412 temporary bit of memory, in case the class contains only 1 character (less
3413 than 256), because in that case the compiled code doesn't use the bit map.
3414 */
3415
3416 memset(classbits, 0, 32 * sizeof(uschar));
3417
3418 #ifdef SUPPORT_UTF8
3419 class_utf8 = FALSE; /* No chars >= 256 */
3420 class_utf8data = code + LINK_SIZE + 2; /* For UTF-8 items */
3421 class_utf8data_base = class_utf8data; /* For resetting in pass 1 */
3422 #endif
3423
3424 /* Process characters until ] is reached. By writing this as a "do" it
3425 means that an initial ] is taken as a data character. At the start of the
3426 loop, c contains the first byte of the character. */
3427
3428 if (c != 0) do
3429 {
3430 const uschar *oldptr;
3431
3432 #ifdef SUPPORT_UTF8
3433 if (utf8 && c > 127)
3434 { /* Braces are required because the */
3435 GETCHARLEN(c, ptr, ptr); /* macro generates multiple statements */
3436 }
3437
3438 /* In the pre-compile phase, accumulate the length of any UTF-8 extra
3439 data and reset the pointer. This is so that very large classes that
3440 contain a zillion UTF-8 characters no longer overwrite the work space
3441 (which is on the stack). */
3442
3443 if (lengthptr != NULL)
3444 {
3445 *lengthptr += class_utf8data - class_utf8data_base;
3446 class_utf8data = class_utf8data_base;
3447 }
3448
3449 #endif
3450
3451 /* Inside \Q...\E everything is literal except \E */
3452
3453 if (inescq)
3454 {
3455 if (c == CHAR_BACKSLASH && ptr[1] == CHAR_E) /* If we are at \E */
3456 {
3457 inescq = FALSE; /* Reset literal state */
3458 ptr++; /* Skip the 'E' */
3459 continue; /* Carry on with next */
3460 }
3461 goto CHECK_RANGE; /* Could be range if \E follows */
3462 }
3463
3464 /* Handle POSIX class names. Perl allows a negation extension of the
3465 form [:^name:]. A square bracket that doesn't match the syntax is
3466 treated as a literal. We also recognize the POSIX constructions
3467 [.ch.] and [=ch=] ("collating elements") and fault them, as Perl
3468 5.6 and 5.8 do. */
3469
3470 if (c == CHAR_LEFT_SQUARE_BRACKET &&
3471 (ptr[1] == CHAR_COLON || ptr[1] == CHAR_DOT ||
3472 ptr[1] == CHAR_EQUALS_SIGN) && check_posix_syntax(ptr, &tempptr))
3473 {
3474 BOOL local_negate = FALSE;
3475 int posix_class, taboffset, tabopt;
3476 register const uschar *cbits = cd->cbits;
3477 uschar pbits[32];
3478
3479 if (ptr[1] != CHAR_COLON)
3480 {
3481 *errorcodeptr = ERR31;
3482 goto FAILED;
3483 }
3484
3485 ptr += 2;
3486 if (*ptr == CHAR_CIRCUMFLEX_ACCENT)
3487 {
3488 local_negate = TRUE;
3489 should_flip_negation = TRUE; /* Note negative special */
3490 ptr++;
3491 }
3492
3493 posix_class = check_posix_name(ptr, (int)(tempptr - ptr));
3494 if (posix_class < 0)
3495 {
3496 *errorcodeptr = ERR30;
3497 goto FAILED;
3498 }
3499
3500 /* If matching is caseless, upper and lower are converted to
3501 alpha. This relies on the fact that the class table starts with
3502 alpha, lower, upper as the first 3 entries. */
3503
3504 if ((options & PCRE_CASELESS) != 0 && posix_class <= 2)
3505 posix_class = 0;
3506
3507 /* When PCRE_UCP is set, some of the POSIX classes are converted to
3508 different escape sequences that use Unicode properties. */
3509
3510 #ifdef SUPPORT_UCP
3511 if ((options & PCRE_UCP) != 0)
3512 {
3513 int pc = posix_class + ((local_negate)? POSIX_SUBSIZE/2 : 0);
3514 if (posix_substitutes[pc] != NULL)
3515 {
3516 nestptr = tempptr + 1;
3517 ptr = posix_substitutes[pc] - 1;
3518 continue;
3519 }
3520 }
3521 #endif
3522 /* In the non-UCP case, we build the bit map for the POSIX class in a
3523 chunk of local store because we may be adding and subtracting from it,
3524 and we don't want to subtract bits that may be in the main map already.
3525 At the end we or the result into the bit map that is being built. */
3526
3527 posix_class *= 3;
3528
3529 /* Copy in the first table (always present) */
3530
3531 memcpy(pbits, cbits + posix_class_maps[posix_class],
3532 32 * sizeof(uschar));
3533
3534 /* If there is a second table, add or remove it as required. */
3535
3536 taboffset = posix_class_maps[posix_class + 1];
3537 tabopt = posix_class_maps[posix_class + 2];
3538
3539 if (taboffset >= 0)
3540 {
3541 if (tabopt >= 0)
3542 for (c = 0; c < 32; c++) pbits[c] |= cbits[c + taboffset];
3543 else
3544 for (c = 0; c < 32; c++) pbits[c] &= ~cbits[c + taboffset];
3545 }
3546
3547 /* Not see if we need to remove any special characters. An option
3548 value of 1 removes vertical space and 2 removes underscore. */
3549
3550 if (tabopt < 0) tabopt = -tabopt;
3551 if (tabopt == 1) pbits[1] &= ~0x3c;
3552 else if (tabopt == 2) pbits[11] &= 0x7f;
3553
3554 /* Add the POSIX table or its complement into the main table that is
3555 being built and we are done. */
3556
3557 if (local_negate)
3558 for (c = 0; c < 32; c++) classbits[c] |= ~pbits[c];
3559 else
3560 for (c = 0; c < 32; c++) classbits[c] |= pbits[c];
3561
3562 ptr = tempptr + 1;
3563 class_charcount = 10; /* Set > 1; assumes more than 1 per class */
3564 continue; /* End of POSIX syntax handling */
3565 }
3566
3567 /* Backslash may introduce a single character, or it may introduce one
3568 of the specials, which just set a flag. The sequence \b is a special
3569 case. Inside a class (and only there) it is treated as backspace. We
3570 assume that other escapes have more than one character in them, so set
3571 class_charcount bigger than one. Unrecognized escapes fall through and
3572 are either treated as literal characters (by default), or are faulted if
3573 PCRE_EXTRA is set. */
3574
3575 if (c == CHAR_BACKSLASH)
3576 {
3577 c = check_escape(&ptr, errorcodeptr, cd->bracount, options, TRUE);
3578 if (*errorcodeptr != 0) goto FAILED;
3579
3580 if (-c == ESC_b) c = CHAR_BS; /* \b is backspace in a class */
3581 else if (-c == ESC_Q) /* Handle start of quoted string */
3582 {
3583 if (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
3584 {
3585 ptr += 2; /* avoid empty string */
3586 }
3587 else inescq = TRUE;
3588 continue;
3589 }
3590 else if (-c == ESC_E) continue; /* Ignore orphan \E */
3591
3592 if (c < 0)
3593 {
3594 register const uschar *cbits = cd->cbits;
3595 class_charcount += 2; /* Greater than 1 is what matters */
3596
3597 switch (-c)
3598 {
3599 #ifdef SUPPORT_UCP
3600 case ESC_du: /* These are the values given for \d etc */
3601 case ESC_DU: /* when PCRE_UCP is set. We replace the */
3602 case ESC_wu: /* escape sequence with an appropriate \p */
3603 case ESC_WU: /* or \P to test Unicode properties instead */
3604 case ESC_su: /* of the default ASCII testing. */
3605 case ESC_SU:
3606 nestptr = ptr;
3607 ptr = substitutes[-c - ESC_DU] - 1; /* Just before substitute */
3608 class_charcount -= 2; /* Undo! */
3609 continue;
3610 #endif
3611 case ESC_d:
3612 for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_digit];
3613 continue;
3614
3615 case ESC_D:
3616 should_flip_negation = TRUE;
3617 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_digit];
3618 continue;
3619
3620 case ESC_w:
3621 for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_word];
3622 continue;
3623
3624 case ESC_W:
3625 should_flip_negation = TRUE;
3626 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_word];
3627 continue;
3628
3629 /* Perl 5.004 onwards omits VT from \s, but we must preserve it
3630 if it was previously set by something earlier in the character
3631 class. */
3632
3633 case ESC_s:
3634 classbits[0] |= cbits[cbit_space];
3635 classbits[1] |= cbits[cbit_space+1] & ~0x08;
3636 for (c = 2; c < 32; c++) classbits[c] |= cbits[c+cbit_space];
3637 continue;
3638
3639 case ESC_S:
3640 should_flip_negation = TRUE;
3641 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_space];
3642 classbits[1] |= 0x08; /* Perl 5.004 onwards omits VT from \s */
3643 continue;
3644
3645 case ESC_h:
3646 SETBIT(classbits, 0x09); /* VT */
3647 SETBIT(classbits, 0x20); /* SPACE */
3648 SETBIT(classbits, 0xa0); /* NSBP */
3649 #ifdef SUPPORT_UTF8
3650 if (utf8)
3651 {
3652 class_utf8 = TRUE;
3653 *class_utf8data++ = XCL_SINGLE;
3654 class_utf8data += _pcre_ord2utf8(0x1680, class_utf8data);
3655 *class_utf8data++ = XCL_SINGLE;
3656 class_utf8data += _pcre_ord2utf8(0x180e, class_utf8data);
3657 *class_utf8data++ = XCL_RANGE;
3658 class_utf8data += _pcre_ord2utf8(0x2000, class_utf8data);
3659 class_utf8data += _pcre_ord2utf8(0x200A, class_utf8data);
3660 *class_utf8data++ = XCL_SINGLE;
3661 class_utf8data += _pcre_ord2utf8(0x202f, class_utf8data);
3662 *class_utf8data++ = XCL_SINGLE;
3663 class_utf8data += _pcre_ord2utf8(0x205f, class_utf8data);
3664 *class_utf8data++ = XCL_SINGLE;
3665 class_utf8data += _pcre_ord2utf8(0x3000, class_utf8data);
3666 }
3667 #endif
3668 continue;
3669
3670 case ESC_H:
3671 for (c = 0; c < 32; c++)
3672 {
3673 int x = 0xff;
3674 switch (c)
3675 {
3676 case 0x09/8: x ^= 1 << (0x09%8); break;
3677 case 0x20/8: x ^= 1 << (0x20%8); break;
3678 case 0xa0/8: x ^= 1 << (0xa0%8); break;
3679 default: break;
3680 }
3681 classbits[c] |= x;
3682 }
3683
3684 #ifdef SUPPORT_UTF8
3685 if (utf8)
3686 {
3687 class_utf8 = TRUE;
3688 *class_utf8data++ = XCL_RANGE;
3689 class_utf8data += _pcre_ord2utf8(0x0100, class_utf8data);
3690 class_utf8data += _pcre_ord2utf8(0x167f, class_utf8data);
3691 *class_utf8data++ = XCL_RANGE;
3692 class_utf8data += _pcre_ord2utf8(0x1681, class_utf8data);
3693 class_utf8data += _pcre_ord2utf8(0x180d, class_utf8data);
3694 *class_utf8data++ = XCL_RANGE;
3695 class_utf8data += _pcre_ord2utf8(0x180f, class_utf8data);
3696 class_utf8data += _pcre_ord2utf8(0x1fff, class_utf8data);
3697 *class_utf8data++ = XCL_RANGE;
3698 class_utf8data += _pcre_ord2utf8(0x200B, class_utf8data);
3699 class_utf8data += _pcre_ord2utf8(0x202e, class_utf8data);
3700 *class_utf8data++ = XCL_RANGE;
3701 class_utf8data += _pcre_ord2utf8(0x2030, class_utf8data);
3702 class_utf8data += _pcre_ord2utf8(0x205e, class_utf8data);
3703 *class_utf8data++ = XCL_RANGE;
3704 class_utf8data += _pcre_ord2utf8(0x2060, class_utf8data);
3705 class_utf8data += _pcre_ord2utf8(0x2fff, class_utf8data);
3706 *class_utf8data++ = XCL_RANGE;
3707 class_utf8data += _pcre_ord2utf8(0x3001, class_utf8data);
3708 class_utf8data += _pcre_ord2utf8(0x7fffffff, class_utf8data);
3709 }
3710 #endif
3711 continue;
3712
3713 case ESC_v:
3714 SETBIT(classbits, 0x0a); /* LF */
3715 SETBIT(classbits, 0x0b); /* VT */
3716 SETBIT(classbits, 0x0c); /* FF */
3717 SETBIT(classbits, 0x0d); /* CR */
3718 SETBIT(classbits, 0x85); /* NEL */
3719 #ifdef SUPPORT_UTF8
3720 if (utf8)
3721 {
3722 class_utf8 = TRUE;
3723 *class_utf8data++ = XCL_RANGE;
3724 class_utf8data += _pcre_ord2utf8(0x2028, class_utf8data);
3725 class_utf8data += _pcre_ord2utf8(0x2029, class_utf8data);
3726 }
3727 #endif
3728 continue;
3729
3730 case ESC_V:
3731 for (c = 0; c < 32; c++)
3732 {
3733 int x = 0xff;
3734 switch (c)
3735 {
3736 case 0x0a/8: x ^= 1 << (0x0a%8);
3737 x ^= 1 << (0x0b%8);
3738 x ^= 1 << (0x0c%8);
3739 x ^= 1 << (0x0d%8);
3740 break;
3741 case 0x85/8: x ^= 1 << (0x85%8); break;
3742 default: break;
3743 }
3744 classbits[c] |= x;
3745 }
3746
3747 #ifdef SUPPORT_UTF8
3748 if (utf8)
3749 {
3750 class_utf8 = TRUE;
3751 *class_utf8data++ = XCL_RANGE;
3752 class_utf8data += _pcre_ord2utf8(0x0100, class_utf8data);
3753 class_utf8data += _pcre_ord2utf8(0x2027, class_utf8data);
3754 *class_utf8data++ = XCL_RANGE;
3755 class_utf8data += _pcre_ord2utf8(0x2029, class_utf8data);
3756 class_utf8data += _pcre_ord2utf8(0x7fffffff, class_utf8data);
3757 }
3758 #endif
3759 continue;
3760
3761 #ifdef SUPPORT_UCP
3762 case ESC_p:
3763 case ESC_P:
3764 {
3765 BOOL negated;
3766 int pdata;
3767 int ptype = get_ucp(&ptr, &negated, &pdata, errorcodeptr);
3768 if (ptype < 0) goto FAILED;
3769 class_utf8 = TRUE;
3770 *class_utf8data++ = ((-c == ESC_p) != negated)?
3771 XCL_PROP : XCL_NOTPROP;
3772 *class_utf8data++ = ptype;
3773 *class_utf8data++ = pdata;
3774 class_charcount -= 2; /* Not a < 256 character */
3775 continue;
3776 }
3777 #endif
3778 /* Unrecognized escapes are faulted if PCRE is running in its
3779 strict mode. By default, for compatibility with Perl, they are
3780 treated as literals. */
3781
3782 default:
3783 if ((options & PCRE_EXTRA) != 0)
3784 {
3785 *errorcodeptr = ERR7;
3786 goto FAILED;
3787 }
3788 class_charcount -= 2; /* Undo the default count from above */
3789 c = *ptr; /* Get the final character and fall through */
3790 break;
3791 }
3792 }
3793
3794 /* Fall through if we have a single character (c >= 0). This may be
3795 greater than 256 in UTF-8 mode. */
3796
3797 } /* End of backslash handling */
3798
3799 /* A single character may be followed by '-' to form a range. However,
3800 Perl does not permit ']' to be the end of the range. A '-' character
3801 at the end is treated as a literal. Perl ignores orphaned \E sequences
3802 entirely. The code for handling \Q and \E is messy. */
3803
3804 CHECK_RANGE:
3805 while (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
3806 {
3807 inescq = FALSE;
3808 ptr += 2;
3809 }
3810
3811 oldptr = ptr;
3812
3813 /* Remember \r or \n */
3814
3815 if (c == CHAR_CR || c == CHAR_NL) cd->external_flags |= PCRE_HASCRORLF;
3816
3817 /* Check for range */
3818
3819 if (!inescq && ptr[1] == CHAR_MINUS)
3820 {
3821 int d;
3822 ptr += 2;
3823 while (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_E) ptr += 2;
3824
3825 /* If we hit \Q (not followed by \E) at this point, go into escaped
3826 mode. */
3827
3828 while (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_Q)
3829 {
3830 ptr += 2;
3831 if (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_E)
3832 { ptr += 2; continue; }
3833 inescq = TRUE;
3834 break;
3835 }
3836
3837 if (*ptr == 0 || (!inescq && *ptr == CHAR_RIGHT_SQUARE_BRACKET))
3838 {
3839 ptr = oldptr;
3840 goto LONE_SINGLE_CHARACTER;
3841 }
3842
3843 #ifdef SUPPORT_UTF8
3844 if (utf8)
3845 { /* Braces are required because the */
3846 GETCHARLEN(d, ptr, ptr); /* macro generates multiple statements */
3847 }
3848 else
3849 #endif
3850 d = *ptr; /* Not UTF-8 mode */
3851
3852 /* The second part of a range can be a single-character escape, but
3853 not any of the other escapes. Perl 5.6 treats a hyphen as a literal
3854 in such circumstances. */
3855
3856 if (!inescq && d == CHAR_BACKSLASH)
3857 {
3858 d = check_escape(&ptr, errorcodeptr, cd->bracount, options, TRUE);
3859 if (*errorcodeptr != 0) goto FAILED;
3860
3861 /* \b is backspace; any other special means the '-' was literal */
3862
3863 if (d < 0)
3864 {
3865 if (d == -ESC_b) d = CHAR_BS; else
3866 {
3867 ptr = oldptr;
3868 goto LONE_SINGLE_CHARACTER; /* A few lines below */
3869 }
3870 }
3871 }
3872
3873 /* Check that the two values are in the correct order. Optimize
3874 one-character ranges */
3875
3876 if (d < c)
3877 {
3878 *errorcodeptr = ERR8;
3879 goto FAILED;
3880 }
3881
3882 if (d == c) goto LONE_SINGLE_CHARACTER; /* A few lines below */
3883
3884 /* Remember \r or \n */
3885
3886 if (d == CHAR_CR || d == CHAR_NL) cd->external_flags |= PCRE_HASCRORLF;
3887
3888 /* In UTF-8 mode, if the upper limit is > 255, or > 127 for caseless
3889 matching, we have to use an XCLASS with extra data items. Caseless
3890 matching for characters > 127 is available only if UCP support is
3891 available. */
3892
3893 #ifdef SUPPORT_UTF8
3894 if (utf8 && (d > 255 || ((options & PCRE_CASELESS) != 0 && d > 127)))
3895 {
3896 class_utf8 = TRUE;
3897
3898 /* With UCP support, we can find the other case equivalents of
3899 the relevant characters. There may be several ranges. Optimize how
3900 they fit with the basic range. */
3901
3902 #ifdef SUPPORT_UCP
3903 if ((options & PCRE_CASELESS) != 0)
3904 {
3905 unsigned int occ, ocd;
3906 unsigned int cc = c;
3907 unsigned int origd = d;
3908 while (get_othercase_range(&cc, origd, &occ, &ocd))
3909 {
3910 if (occ >= (unsigned int)c &&
3911 ocd <= (unsigned int)d)
3912 continue; /* Skip embedded ranges */
3913
3914 if (occ < (unsigned int)c &&
3915 ocd >= (unsigned int)c - 1) /* Extend the basic range */
3916 { /* if there is overlap, */
3917 c = occ; /* noting that if occ < c */
3918 continue; /* we can't have ocd > d */
3919 } /* because a subrange is */
3920 if (ocd > (unsigned int)d &&
3921 occ <= (unsigned int)d + 1) /* always shorter than */
3922 { /* the basic range. */
3923 d = ocd;
3924 continue;
3925 }
3926
3927 if (occ == ocd)
3928 {
3929 *class_utf8data++ = XCL_SINGLE;
3930 }
3931 else
3932 {
3933 *class_utf8data++ = XCL_RANGE;
3934 class_utf8data += _pcre_ord2utf8(occ, class_utf8data);
3935 }
3936 class_utf8data += _pcre_ord2utf8(ocd, class_utf8data);
3937 }
3938 }
3939 #endif /* SUPPORT_UCP */
3940
3941 /* Now record the original range, possibly modified for UCP caseless
3942 overlapping ranges. */
3943
3944 *class_utf8data++ = XCL_RANGE;
3945 class_utf8data += _pcre_ord2utf8(c, class_utf8data);
3946 class_utf8data += _pcre_ord2utf8(d, class_utf8data);
3947
3948 /* With UCP support, we are done. Without UCP support, there is no
3949 caseless matching for UTF-8 characters > 127; we can use the bit map
3950 for the smaller ones. */
3951
3952 #ifdef SUPPORT_UCP
3953 continue; /* With next character in the class */
3954 #else
3955 if ((options & PCRE_CASELESS) == 0 || c > 127) continue;
3956
3957 /* Adjust upper limit and fall through to set up the map */
3958
3959 d = 127;
3960
3961 #endif /* SUPPORT_UCP */
3962 }
3963 #endif /* SUPPORT_UTF8 */
3964
3965 /* We use the bit map for all cases when not in UTF-8 mode; else
3966 ranges that lie entirely within 0-127 when there is UCP support; else
3967 for partial ranges without UCP support. */
3968
3969 class_charcount += d - c + 1;
3970 class_lastchar = d;
3971
3972 /* We can save a bit of time by skipping this in the pre-compile. */
3973
3974 if (lengthptr == NULL) for (; c <= d; c++)
3975 {
3976 classbits[c/8] |= (1 << (c&7));
3977 if ((options & PCRE_CASELESS) != 0)
3978 {
3979 int uc = cd->fcc[c]; /* flip case */
3980 classbits[uc/8] |= (1 << (uc&7));
3981 }
3982 }
3983
3984 continue; /* Go get the next char in the class */
3985 }
3986
3987 /* Handle a lone single character - we can get here for a normal
3988 non-escape char, or after \ that introduces a single character or for an
3989 apparent range that isn't. */
3990
3991 LONE_SINGLE_CHARACTER:
3992
3993 /* Handle a character that cannot go in the bit map */
3994
3995 #ifdef SUPPORT_UTF8
3996 if (utf8 && (c > 255 || ((options & PCRE_CASELESS) != 0 && c > 127)))
3997 {
3998 class_utf8 = TRUE;
3999 *class_utf8data++ = XCL_SINGLE;
4000 class_utf8data += _pcre_ord2utf8(c, class_utf8data);
4001
4002 #ifdef SUPPORT_UCP
4003 if ((options & PCRE_CASELESS) != 0)
4004 {
4005 unsigned int othercase;
4006 if ((othercase = UCD_OTHERCASE(c)) != c)
4007 {
4008 *class_utf8data++ = XCL_SINGLE;
4009 class_utf8data += _pcre_ord2utf8(othercase, class_utf8data);
4010 }
4011 }
4012 #endif /* SUPPORT_UCP */
4013
4014 }
4015 else
4016 #endif /* SUPPORT_UTF8 */
4017
4018 /* Handle a single-byte character */
4019 {
4020 classbits[c/8] |= (1 << (c&7));
4021 if ((options & PCRE_CASELESS) != 0)
4022 {
4023 c = cd->fcc[c]; /* flip case */
4024 classbits[c/8] |= (1 << (c&7));
4025 }
4026 class_charcount++;
4027 class_lastchar = c;
4028 }
4029 }
4030
4031 /* Loop until ']' reached. This "while" is the end of the "do" far above.
4032 If we are at the end of an internal nested string, revert to the outer
4033 string. */
4034
4035 while (((c = *(++ptr)) != 0 ||
4036 (nestptr != NULL &&
4037 (ptr = nestptr, nestptr = NULL, c = *(++ptr)) != 0)) &&
4038 (c != CHAR_RIGHT_SQUARE_BRACKET || inescq));
4039
4040 /* Check for missing terminating ']' */
4041
4042 if (c == 0)
4043 {
4044 *errorcodeptr = ERR6;
4045 goto FAILED;
4046 }
4047
4048 /* If class_charcount is 1, we saw precisely one character whose value is
4049 less than 256. As long as there were no characters >= 128 and there was no
4050 use of \p or \P, in other words, no use of any XCLASS features, we can
4051 optimize.
4052
4053 In UTF-8 mode, we can optimize the negative case only if there were no
4054 characters >= 128 because OP_NOT and the related opcodes like OP_NOTSTAR
4055 operate on single-bytes characters only. This is an historical hangover.
4056 Maybe one day we can tidy these opcodes to handle multi-byte characters.
4057
4058 The optimization throws away the bit map. We turn the item into a
4059 1-character OP_CHAR[I] if it's positive, or OP_NOT[I] if it's negative.
4060 Note that OP_NOT[I] does not support multibyte characters. In the positive
4061 case, it can cause firstbyte to be set. Otherwise, there can be no first
4062 char if this item is first, whatever repeat count may follow. In the case
4063 of reqbyte, save the previous value for reinstating. */
4064
4065 #ifdef SUPPORT_UTF8
4066 if (class_charcount == 1 && !class_utf8 &&
4067 (!utf8 || !negate_class || class_lastchar < 128))
4068 #else
4069 if (class_charcount == 1)
4070 #endif
4071 {
4072 zeroreqbyte = reqbyte;
4073
4074 /* The OP_NOT[I] opcodes work on one-byte characters only. */
4075
4076 if (negate_class)
4077 {
4078 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
4079 zerofirstbyte = firstbyte;
4080 *code++ = ((options & PCRE_CASELESS) != 0)? OP_NOTI: OP_NOT;
4081 *code++ = class_lastchar;
4082 break;
4083 }
4084
4085 /* For a single, positive character, get the value into mcbuffer, and
4086 then we can handle this with the normal one-character code. */
4087
4088 #ifdef SUPPORT_UTF8
4089 if (utf8 && class_lastchar > 127)
4090 mclength = _pcre_ord2utf8(class_lastchar, mcbuffer);
4091 else
4092 #endif
4093 {
4094 mcbuffer[0] = class_lastchar;
4095 mclength = 1;
4096 }
4097 goto ONE_CHAR;
4098 } /* End of 1-char optimization */
4099
4100 /* The general case - not the one-char optimization. If this is the first
4101 thing in the branch, there can be no first char setting, whatever the
4102 repeat count. Any reqbyte setting must remain unchanged after any kind of
4103 repeat. */
4104
4105 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
4106 zerofirstbyte = firstbyte;
4107 zeroreqbyte = reqbyte;
4108
4109 /* If there are characters with values > 255, we have to compile an
4110 extended class, with its own opcode, unless there was a negated special
4111 such as \S in the class, and PCRE_UCP is not set, because in that case all
4112 characters > 255 are in the class, so any that were explicitly given as
4113 well can be ignored. If (when there are explicit characters > 255 that must
4114 be listed) there are no characters < 256, we can omit the bitmap in the
4115 actual compiled code. */
4116
4117 #ifdef SUPPORT_UTF8
4118 if (class_utf8 && (!should_flip_negation || (options & PCRE_UCP) != 0))
4119 {
4120 *class_utf8data++ = XCL_END; /* Marks the end of extra data */
4121 *code++ = OP_XCLASS;
4122 code += LINK_SIZE;
4123 *code = negate_class? XCL_NOT : 0;
4124
4125 /* If the map is required, move up the extra data to make room for it;
4126 otherwise just move the code pointer to the end of the extra data. */
4127
4128 if (class_charcount > 0)
4129 {
4130 *code++ |= XCL_MAP;
4131 memmove(code + 32, code, class_utf8data - code);
4132 memcpy(code, classbits, 32);
4133 code = class_utf8data + 32;
4134 }
4135 else code = class_utf8data;
4136
4137 /* Now fill in the complete length of the item */
4138
4139 PUT(previous, 1, code - previous);
4140 break; /* End of class handling */
4141 }
4142 #endif
4143
4144 /* If there are no characters > 255, or they are all to be included or
4145 excluded, set the opcode to OP_CLASS or OP_NCLASS, depending on whether the
4146 whole class was negated and whether there were negative specials such as \S
4147 (non-UCP) in the class. Then copy the 32-byte map into the code vector,
4148 negating it if necessary. */
4149
4150 *code++ = (negate_class == should_flip_negation) ? OP_CLASS : OP_NCLASS;
4151 if (negate_class)
4152 {
4153 if (lengthptr == NULL) /* Save time in the pre-compile phase */
4154 for (c = 0; c < 32; c++) code[c] = ~classbits[c];
4155 }
4156 else
4157 {
4158 memcpy(code, classbits, 32);
4159 }
4160 code += 32;
4161 break;
4162
4163
4164 /* ===================================================================*/
4165 /* Various kinds of repeat; '{' is not necessarily a quantifier, but this
4166 has been tested above. */
4167
4168 case CHAR_LEFT_CURLY_BRACKET:
4169 if (!is_quantifier) goto NORMAL_CHAR;
4170 ptr = read_repeat_counts(ptr+1, &repeat_min, &repeat_max, errorcodeptr);
4171 if (*errorcodeptr != 0) goto FAILED;
4172 goto REPEAT;
4173
4174 case CHAR_ASTERISK:
4175 repeat_min = 0;
4176 repeat_max = -1;
4177 goto REPEAT;
4178
4179 case CHAR_PLUS:
4180 repeat_min = 1;
4181 repeat_max = -1;
4182 goto REPEAT;
4183
4184 case CHAR_QUESTION_MARK:
4185 repeat_min = 0;
4186 repeat_max = 1;
4187
4188 REPEAT:
4189 if (previous == NULL)
4190 {
4191 *errorcodeptr = ERR9;
4192 goto FAILED;
4193 }
4194
4195 if (repeat_min == 0)
4196 {
4197 firstbyte = zerofirstbyte; /* Adjust for zero repeat */
4198 reqbyte = zeroreqbyte; /* Ditto */
4199 }
4200
4201 /* Remember whether this is a variable length repeat */
4202
4203 reqvary = (repeat_min == repeat_max)? 0 : REQ_VARY;
4204
4205 op_type = 0; /* Default single-char op codes */
4206 possessive_quantifier = FALSE; /* Default not possessive quantifier */
4207
4208 /* Save start of previous item, in case we have to move it up to make space
4209 for an inserted OP_ONCE for the additional '+' extension. */
4210
4211 tempcode = previous;
4212
4213 /* If the next character is '+', we have a possessive quantifier. This
4214 implies greediness, whatever the setting of the PCRE_UNGREEDY option.
4215 If the next character is '?' this is a minimizing repeat, by default,
4216 but if PCRE_UNGREEDY is set, it works the other way round. We change the
4217 repeat type to the non-default. */
4218
4219 if (ptr[1] == CHAR_PLUS)
4220 {
4221 repeat_type = 0; /* Force greedy */
4222 possessive_quantifier = TRUE;
4223 ptr++;
4224 }
4225 else if (ptr[1] == CHAR_QUESTION_MARK)
4226 {
4227 repeat_type = greedy_non_default;
4228 ptr++;
4229 }
4230 else repeat_type = greedy_default;
4231
4232 /* If previous was a recursion call, wrap it in atomic brackets so that
4233 previous becomes the atomic group. All recursions were so wrapped in the
4234 past, but it no longer happens for non-repeated recursions. In fact, the
4235 repeated ones could be re-implemented independently so as not to need this,
4236 but for the moment we rely on the code for repeating groups. */
4237
4238 if (*previous == OP_RECURSE)
4239 {
4240 memmove(previous + 1 + LINK_SIZE, previous, 1 + LINK_SIZE);
4241 *previous = OP_ONCE;
4242 PUT(previous, 1, 2 + 2*LINK_SIZE);
4243 previous[2 + 2*LINK_SIZE] = OP_KET;
4244 PUT(previous, 3 + 2*LINK_SIZE, 2 + 2*LINK_SIZE);
4245 code += 2 + 2 * LINK_SIZE;
4246 length_prevgroup = 3 + 3*LINK_SIZE;
4247
4248 /* When actually compiling, we need to check whether this was a forward
4249 reference, and if so, adjust the offset. */
4250
4251 if (lengthptr == NULL && cd->hwm >= cd->start_workspace + LINK_SIZE)
4252 {
4253 int offset = GET(cd->hwm, -LINK_SIZE);
4254 if (offset == previous + 1 - cd->start_code)
4255 PUT(cd->hwm, -LINK_SIZE, offset + 1 + LINK_SIZE);
4256 }
4257 }
4258
4259 /* Now handle repetition for the different types of item. */
4260
4261 /* If previous was a character match, abolish the item and generate a
4262 repeat item instead. If a char item has a minumum of more than one, ensure
4263 that it is set in reqbyte - it might not be if a sequence such as x{3} is
4264 the first thing in a branch because the x will have gone into firstbyte
4265 instead. */
4266
4267 if (*previous == OP_CHAR || *previous == OP_CHARI)
4268 {
4269 op_type = (*previous == OP_CHAR)? 0 : OP_STARI - OP_STAR;
4270
4271 /* Deal with UTF-8 characters that take up more than one byte. It's
4272 easier to write this out separately than try to macrify it. Use c to
4273 hold the length of the character in bytes, plus 0x80 to flag that it's a
4274 length rather than a small character. */
4275
4276 #ifdef SUPPORT_UTF8
4277 if (utf8 && (code[-1] & 0x80) != 0)
4278 {
4279 uschar *lastchar = code - 1;
4280 while((*lastchar & 0xc0) == 0x80) lastchar--;
4281 c = code - lastchar; /* Length of UTF-8 character */
4282 memcpy(utf8_char, lastchar, c); /* Save the char */
4283 c |= 0x80; /* Flag c as a length */
4284 }
4285 else
4286 #endif
4287
4288 /* Handle the case of a single byte - either with no UTF8 support, or
4289 with UTF-8 disabled, or for a UTF-8 character < 128. */
4290
4291 {
4292 c = code[-1];
4293 if (repeat_min > 1) reqbyte = c | req_caseopt | cd->req_varyopt;
4294 }
4295
4296 /* If the repetition is unlimited, it pays to see if the next thing on
4297 the line is something that cannot possibly match this character. If so,
4298 automatically possessifying this item gains some performance in the case
4299 where the match fails. */
4300
4301 if (!possessive_quantifier &&
4302 repeat_max < 0 &&
4303 check_auto_possessive(previous, utf8, ptr + 1, options, cd))
4304 {
4305 repeat_type = 0; /* Force greedy */
4306 possessive_quantifier = TRUE;
4307 }
4308
4309 goto OUTPUT_SINGLE_REPEAT; /* Code shared with single character types */
4310 }
4311
4312 /* If previous was a single negated character ([^a] or similar), we use
4313 one of the special opcodes, replacing it. The code is shared with single-
4314 character repeats by setting opt_type to add a suitable offset into
4315 repeat_type. We can also test for auto-possessification. OP_NOT and OP_NOTI
4316 are currently used only for single-byte chars. */
4317
4318 else if (*previous == OP_NOT || *previous == OP_NOTI)
4319 {
4320 op_type = ((*previous == OP_NOT)? OP_NOTSTAR : OP_NOTSTARI) - OP_STAR;
4321 c = previous[1];
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 goto OUTPUT_SINGLE_REPEAT;
4330 }
4331
4332 /* If previous was a character type match (\d or similar), abolish it and
4333 create a suitable repeat item. The code is shared with single-character
4334 repeats by setting op_type to add a suitable offset into repeat_type. Note
4335 the the Unicode property types will be present only when SUPPORT_UCP is
4336 defined, but we don't wrap the little bits of code here because it just
4337 makes it horribly messy. */
4338
4339 else if (*previous < OP_EODN)
4340 {
4341 uschar *oldcode;
4342 int prop_type, prop_value;
4343 op_type = OP_TYPESTAR - OP_STAR; /* Use type opcodes */
4344 c = *previous;
4345
4346 if (!possessive_quantifier &&
4347 repeat_max < 0 &&
4348 check_auto_possessive(previous, utf8, ptr + 1, options, cd))
4349 {
4350 repeat_type = 0; /* Force greedy */
4351 possessive_quantifier = TRUE;
4352 }
4353
4354 OUTPUT_SINGLE_REPEAT:
4355 if (*previous == OP_PROP || *previous == OP_NOTPROP)
4356 {
4357 prop_type = previous[1];
4358 prop_value = previous[2];
4359 }
4360 else prop_type = prop_value = -1;
4361
4362 oldcode = code;
4363 code = previous; /* Usually overwrite previous item */
4364
4365 /* If the maximum is zero then the minimum must also be zero; Perl allows
4366 this case, so we do too - by simply omitting the item altogether. */
4367
4368 if (repeat_max == 0) goto END_REPEAT;
4369
4370 /*--------------------------------------------------------------------*/
4371 /* This code is obsolete from release 8.00; the restriction was finally
4372 removed: */
4373
4374 /* All real repeats make it impossible to handle partial matching (maybe
4375 one day we will be able to remove this restriction). */
4376
4377 /* if (repeat_max != 1) cd->external_flags |= PCRE_NOPARTIAL; */
4378 /*--------------------------------------------------------------------*/
4379
4380 /* Combine the op_type with the repeat_type */
4381
4382 repeat_type += op_type;
4383
4384 /* A minimum of zero is handled either as the special case * or ?, or as
4385 an UPTO, with the maximum given. */
4386
4387 if (repeat_min == 0)
4388 {
4389 if (repeat_max == -1) *code++ = OP_STAR + repeat_type;
4390 else if (repeat_max == 1) *code++ = OP_QUERY + repeat_type;
4391 else
4392 {
4393 *code++ = OP_UPTO + repeat_type;
4394 PUT2INC(code, 0, repeat_max);
4395 }
4396 }
4397
4398 /* A repeat minimum of 1 is optimized into some special cases. If the
4399 maximum is unlimited, we use OP_PLUS. Otherwise, the original item is
4400 left in place and, if the maximum is greater than 1, we use OP_UPTO with
4401 one less than the maximum. */
4402
4403 else if (repeat_min == 1)
4404 {
4405 if (repeat_max == -1)
4406 *code++ = OP_PLUS + repeat_type;
4407 else
4408 {
4409 code = oldcode; /* leave previous item in place */
4410 if (repeat_max == 1) goto END_REPEAT;
4411 *code++ = OP_UPTO + repeat_type;
4412 PUT2INC(code, 0, repeat_max - 1);
4413 }
4414 }
4415
4416 /* The case {n,n} is just an EXACT, while the general case {n,m} is
4417 handled as an EXACT followed by an UPTO. */
4418
4419 else
4420 {
4421 *code++ = OP_EXACT + op_type; /* NB EXACT doesn't have repeat_type */
4422 PUT2INC(code, 0, repeat_min);
4423
4424 /* If the maximum is unlimited, insert an OP_STAR. Before doing so,
4425 we have to insert the character for the previous code. For a repeated
4426 Unicode property match, there are two extra bytes that define the
4427 required property. In UTF-8 mode, long characters have their length in
4428 c, with the 0x80 bit as a flag. */
4429
4430 if (repeat_max < 0)
4431 {
4432 #ifdef SUPPORT_UTF8
4433 if (utf8 && c >= 128)
4434 {
4435 memcpy(code, utf8_char, c & 7);
4436 code += c & 7;
4437 }
4438 else
4439 #endif
4440 {
4441 *code++ = c;
4442 if (prop_type >= 0)
4443 {
4444 *code++ = prop_type;
4445 *code++ = prop_value;
4446 }
4447 }
4448 *code++ = OP_STAR + repeat_type;
4449 }
4450
4451 /* Else insert an UPTO if the max is greater than the min, again
4452 preceded by the character, for the previously inserted code. If the
4453 UPTO is just for 1 instance, we can use QUERY instead. */
4454
4455 else if (repeat_max != repeat_min)
4456 {
4457 #ifdef SUPPORT_UTF8
4458 if (utf8 && c >= 128)
4459 {
4460 memcpy(code, utf8_char, c & 7);
4461 code += c & 7;
4462 }
4463 else
4464 #endif
4465 *code++ = c;
4466 if (prop_type >= 0)
4467 {
4468 *code++ = prop_type;
4469 *code++ = prop_value;
4470 }
4471 repeat_max -= repeat_min;
4472
4473 if (repeat_max == 1)
4474 {
4475 *code++ = OP_QUERY + repeat_type;
4476 }
4477 else
4478 {
4479 *code++ = OP_UPTO + repeat_type;
4480 PUT2INC(code, 0, repeat_max);
4481 }
4482 }
4483 }
4484
4485 /* The character or character type itself comes last in all cases. */
4486
4487 #ifdef SUPPORT_UTF8
4488 if (utf8 && c >= 128)
4489 {
4490 memcpy(code, utf8_char, c & 7);
4491 code += c & 7;
4492 }
4493 else
4494 #endif
4495 *code++ = c;
4496
4497 /* For a repeated Unicode property match, there are two extra bytes that
4498 define the required property. */
4499
4500 #ifdef SUPPORT_UCP
4501 if (prop_type >= 0)
4502 {
4503 *code++ = prop_type;
4504 *code++ = prop_value;
4505 }
4506 #endif
4507 }
4508
4509 /* If previous was a character class or a back reference, we put the repeat
4510 stuff after it, but just skip the item if the repeat was {0,0}. */
4511
4512 else if (*previous == OP_CLASS ||
4513 *previous == OP_NCLASS ||
4514 #ifdef SUPPORT_UTF8
4515 *previous == OP_XCLASS ||
4516 #endif
4517 *previous == OP_REF ||
4518 *previous == OP_REFI)
4519 {
4520 if (repeat_max == 0)
4521 {
4522 code = previous;
4523 goto END_REPEAT;
4524 }
4525
4526 /*--------------------------------------------------------------------*/
4527 /* This code is obsolete from release 8.00; the restriction was finally
4528 removed: */
4529
4530 /* All real repeats make it impossible to handle partial matching (maybe
4531 one day we will be able to remove this restriction). */
4532
4533 /* if (repeat_max != 1) cd->external_flags |= PCRE_NOPARTIAL; */
4534 /*--------------------------------------------------------------------*/
4535
4536 if (repeat_min == 0 && repeat_max == -1)
4537 *code++ = OP_CRSTAR + repeat_type;
4538 else if (repeat_min == 1 && repeat_max == -1)
4539 *code++ = OP_CRPLUS + repeat_type;
4540 else if (repeat_min == 0 && repeat_max == 1)
4541 *code++ = OP_CRQUERY + repeat_type;
4542 else
4543 {
4544 *code++ = OP_CRRANGE + repeat_type;
4545 PUT2INC(code, 0, repeat_min);
4546 if (repeat_max == -1) repeat_max = 0; /* 2-byte encoding for max */
4547 PUT2INC(code, 0, repeat_max);
4548 }
4549 }
4550
4551 /* If previous was a bracket group, we may have to replicate it in certain
4552 cases. Note that at this point we can encounter only the "basic" BRA and
4553 KET opcodes, as this is the place where they get converted into the more
4554 special varieties. */
4555
4556 else if (*previous == OP_BRA || *previous == OP_CBRA ||
4557 *previous == OP_ONCE || *previous == OP_COND)
4558 {
4559 register int i;
4560 int len = (int)(code - previous);
4561 uschar *bralink = NULL;
4562 uschar *brazeroptr = NULL;
4563
4564 /* Repeating a DEFINE group is pointless */
4565
4566 if (*previous == OP_COND && previous[LINK_SIZE+1] == OP_DEF)
4567 {
4568 *errorcodeptr = ERR55;
4569 goto FAILED;
4570 }
4571
4572 /* The case of a zero minimum is special because of the need to stick
4573 OP_BRAZERO in front of it, and because the group appears once in the
4574 data, whereas in other cases it appears the minimum number of times. For
4575 this reason, it is simplest to treat this case separately, as otherwise
4576 the code gets far too messy. There are several special subcases when the
4577 minimum is zero. */
4578
4579 if (repeat_min == 0)
4580 {
4581 /* If the maximum is also zero, we used to just omit the group from the
4582 output altogether, like this:
4583
4584 ** if (repeat_max == 0)
4585 ** {
4586 ** code = previous;
4587 ** goto END_REPEAT;
4588 ** }
4589
4590 However, that fails when a group is referenced as a subroutine from
4591 elsewhere in the pattern, so now we stick in OP_SKIPZERO in front of it
4592 so that it is skipped on execution. As we don't have a list of which
4593 groups are referenced, we cannot do this selectively.
4594
4595 If the maximum is 1 or unlimited, we just have to stick in the BRAZERO
4596 and do no more at this point. However, we do need to adjust any
4597 OP_RECURSE calls inside the group that refer to the group itself or any
4598 internal or forward referenced group, because the offset is from the
4599 start of the whole regex. Temporarily terminate the pattern while doing
4600 this. */
4601
4602 if (repeat_max <= 1) /* Covers 0, 1, and unlimited */
4603 {
4604 *code = OP_END;
4605 adjust_recurse(previous, 1, utf8, cd, save_hwm);
4606 memmove(previous+1, previous, len);
4607 code++;
4608 if (repeat_max == 0)
4609 {
4610 *previous++ = OP_SKIPZERO;
4611 goto END_REPEAT;
4612 }
4613 brazeroptr = previous; /* Save for possessive optimizing */
4614 *previous++ = OP_BRAZERO + repeat_type;
4615 }
4616
4617 /* If the maximum is greater than 1 and limited, we have to replicate
4618 in a nested fashion, sticking OP_BRAZERO before each set of brackets.
4619 The first one has to be handled carefully because it's the original
4620 copy, which has to be moved up. The remainder can be handled by code
4621 that is common with the non-zero minimum case below. We have to
4622 adjust the value or repeat_max, since one less copy is required. Once
4623 again, we may have to adjust any OP_RECURSE calls inside the group. */
4624
4625 else
4626 {
4627 int offset;
4628 *code = OP_END;
4629 adjust_recurse(previous, 2 + LINK_SIZE, utf8, cd, save_hwm);
4630 memmove(previous + 2 + LINK_SIZE, previous, len);
4631 code += 2 + LINK_SIZE;
4632 *previous++ = OP_BRAZERO + repeat_type;
4633 *previous++ = OP_BRA;
4634
4635 /* We chain together the bracket offset fields that have to be
4636 filled in later when the ends of the brackets are reached. */
4637
4638 offset = (bralink == NULL)? 0 : (int)(previous - bralink);
4639 bralink = previous;
4640 PUTINC(previous, 0, offset);
4641 }
4642
4643 repeat_max--;
4644 }
4645
4646 /* If the minimum is greater than zero, replicate the group as many
4647 times as necessary, and adjust the maximum to the number of subsequent
4648 copies that we need. If we set a first char from the group, and didn't
4649 set a required char, copy the latter from the former. If there are any
4650 forward reference subroutine calls in the group, there will be entries on
4651 the workspace list; replicate these with an appropriate increment. */
4652
4653 else
4654 {
4655 if (repeat_min > 1)
4656 {
4657 /* In the pre-compile phase, we don't actually do the replication. We
4658 just adjust the length as if we had. Do some paranoid checks for
4659 potential integer overflow. The INT64_OR_DOUBLE type is a 64-bit
4660 integer type when available, otherwise double. */
4661
4662 if (lengthptr != NULL)
4663 {
4664 int delta = (repeat_min - 1)*length_prevgroup;
4665 if ((INT64_OR_DOUBLE)(repeat_min - 1)*
4666 (INT64_OR_DOUBLE)length_prevgroup >
4667 (INT64_OR_DOUBLE)INT_MAX ||
4668 OFLOW_MAX - *lengthptr < delta)
4669 {
4670 *errorcodeptr = ERR20;
4671 goto FAILED;
4672 }
4673 *lengthptr += delta;
4674 }
4675
4676 /* This is compiling for real */
4677
4678 else
4679 {
4680 if (groupsetfirstbyte && reqbyte < 0) reqbyte = firstbyte;
4681 for (i = 1; i < repeat_min; i++)
4682 {
4683 uschar *hc;
4684 uschar *this_hwm = cd->hwm;
4685 memcpy(code, previous, len);
4686 for (hc = save_hwm; hc < this_hwm; hc += LINK_SIZE)
4687 {
4688 PUT(cd->hwm, 0, GET(hc, 0) + len);
4689 cd->hwm += LINK_SIZE;
4690 }
4691 save_hwm = this_hwm;
4692 code += len;
4693 }
4694 }
4695 }
4696
4697 if (repeat_max > 0) repeat_max -= repeat_min;
4698 }
4699
4700 /* This code is common to both the zero and non-zero minimum cases. If
4701 the maximum is limited, it replicates the group in a nested fashion,
4702 remembering the bracket starts on a stack. In the case of a zero minimum,
4703 the first one was set up above. In all cases the repeat_max now specifies
4704 the number of additional copies needed. Again, we must remember to
4705 replicate entries on the forward reference list. */
4706
4707 if (repeat_max >= 0)
4708 {
4709 /* In the pre-compile phase, we don't actually do the replication. We
4710 just adjust the length as if we had. For each repetition we must add 1
4711 to the length for BRAZERO and for all but the last repetition we must
4712 add 2 + 2*LINKSIZE to allow for the nesting that occurs. Do some
4713 paranoid checks to avoid integer overflow. The INT64_OR_DOUBLE type is
4714 a 64-bit integer type when available, otherwise double. */
4715
4716 if (lengthptr != NULL && repeat_max > 0)
4717 {
4718 int delta = repeat_max * (length_prevgroup + 1 + 2 + 2*LINK_SIZE) -
4719 2 - 2*LINK_SIZE; /* Last one doesn't nest */
4720 if ((INT64_OR_DOUBLE)repeat_max *
4721 (INT64_OR_DOUBLE)(length_prevgroup + 1 + 2 + 2*LINK_SIZE)
4722 > (INT64_OR_DOUBLE)INT_MAX ||
4723 OFLOW_MAX - *lengthptr < delta)
4724 {
4725 *errorcodeptr = ERR20;
4726 goto FAILED;
4727 }
4728 *lengthptr += delta;
4729 }
4730
4731 /* This is compiling for real */
4732
4733 else for (i = repeat_max - 1; i >= 0; i--)
4734 {
4735 uschar *hc;
4736 uschar *this_hwm = cd->hwm;
4737
4738 *code++ = OP_BRAZERO + repeat_type;
4739
4740 /* All but the final copy start a new nesting, maintaining the
4741 chain of brackets outstanding. */
4742
4743 if (i != 0)
4744 {
4745 int offset;
4746 *code++ = OP_BRA;
4747 offset = (bralink == NULL)? 0 : (int)(code - bralink);
4748 bralink = code;
4749 PUTINC(code, 0, offset);
4750 }
4751
4752 memcpy(code, previous, len);
4753 for (hc = save_hwm; hc < this_hwm; hc += LINK_SIZE)
4754 {
4755 PUT(cd->hwm, 0, GET(hc, 0) + len + ((i != 0)? 2+LINK_SIZE : 1));
4756 cd->hwm += LINK_SIZE;
4757 }
4758 save_hwm = this_hwm;
4759 code += len;
4760 }
4761
4762 /* Now chain through the pending brackets, and fill in their length
4763 fields (which are holding the chain links pro tem). */
4764
4765 while (bralink != NULL)
4766 {
4767 int oldlinkoffset;
4768 int offset = (int)(code - bralink + 1);
4769 uschar *bra = code - offset;
4770 oldlinkoffset = GET(bra, 1);
4771 bralink = (oldlinkoffset == 0)? NULL : bralink - oldlinkoffset;
4772 *code++ = OP_KET;
4773 PUTINC(code, 0, offset);
4774 PUT(bra, 1, offset);
4775 }
4776 }
4777
4778 /* If the maximum is unlimited, set a repeater in the final copy. For
4779 ONCE brackets, that's all we need to do. However, possessively repeated
4780 ONCE brackets can be converted into non-capturing brackets, as the
4781 behaviour of (?:xx)++ is the same as (?>xx)++ and this saves having to
4782 deal with possessive ONCEs specially.
4783
4784 Otherwise, if the quantifier was possessive, we convert the BRA code to
4785 the POS form, and the KET code to KETRPOS. (It turns out to be convenient
4786 at runtime to detect this kind of subpattern at both the start and at the
4787 end.) The use of special opcodes makes it possible to reduce greatly the
4788 stack usage in pcre_exec(). If the group is preceded by OP_BRAZERO,
4789 convert this to OP_BRAPOSZERO. Then cancel the possessive flag so that
4790 the default action below, of wrapping everything inside atomic brackets,
4791 does not happen.
4792
4793 Then, when we are doing the actual compile phase, check to see whether
4794 this group is one that could match an empty string. If so, convert the
4795 initial operator to the S form (e.g. OP_BRA -> OP_SBRA) so that runtime
4796 checking can be done. [This check is also applied to ONCE groups at
4797 runtime, but in a different way.] */
4798
4799 else
4800 {
4801 uschar *ketcode = code - 1 - LINK_SIZE;
4802 uschar *bracode = ketcode - GET(ketcode, 1);
4803
4804 if (*bracode == OP_ONCE && possessive_quantifier) *bracode = OP_BRA;
4805 if (*bracode == OP_ONCE)
4806 *ketcode = OP_KETRMAX + repeat_type;
4807 else
4808 {
4809 if (possessive_quantifier)
4810 {
4811 *bracode += 1; /* Switch to xxxPOS opcodes */
4812 *ketcode = OP_KETRPOS;
4813 if (brazeroptr != NULL) *brazeroptr = OP_BRAPOSZERO;
4814 possessive_quantifier = FALSE;
4815 }
4816 else *ketcode = OP_KETRMAX + repeat_type;
4817
4818 if (lengthptr == NULL)
4819 {
4820 uschar *scode = bracode;
4821 do
4822 {
4823 if (could_be_empty_branch(scode, ketcode, utf8, cd))
4824 {
4825 *bracode += OP_SBRA - OP_BRA;
4826 break;
4827 }
4828 scode += GET(scode, 1);
4829 }
4830 while (*scode == OP_ALT);
4831 }
4832 }
4833 }
4834 }
4835
4836 /* If previous is OP_FAIL, it was generated by an empty class [] in
4837 JavaScript mode. The other ways in which OP_FAIL can be generated, that is
4838 by (*FAIL) or (?!) set previous to NULL, which gives a "nothing to repeat"
4839 error above. We can just ignore the repeat in JS case. */
4840
4841 else if (*previous == OP_FAIL) goto END_REPEAT;
4842
4843 /* Else there's some kind of shambles */
4844
4845 else
4846 {
4847 *errorcodeptr = ERR11;
4848 goto FAILED;
4849 }
4850
4851 /* If the character following a repeat is '+', or if certain optimization
4852 tests above succeeded, possessive_quantifier is TRUE. For some opcodes,
4853 there are special alternative opcodes for this case. For anything else, we
4854 wrap the entire repeated item inside OP_ONCE brackets. Logically, the '+'
4855 notation is just syntactic sugar, taken from Sun's Java package, but the
4856 special opcodes can optimize it.
4857
4858 Possessively repeated subpatterns have already been handled in the code
4859 just above, so possessive_quantifier is always FALSE for them at this
4860 stage.
4861
4862 Note that the repeated item starts at tempcode, not at previous, which
4863 might be the first part of a string whose (former) last char we repeated.
4864
4865 Possessifying an 'exact' quantifier has no effect, so we can ignore it. But
4866 an 'upto' may follow. We skip over an 'exact' item, and then test the
4867 length of what remains before proceeding. */
4868
4869 if (possessive_quantifier)
4870 {
4871 int len;
4872
4873 if (*tempcode == OP_TYPEEXACT)
4874 tempcode += _pcre_OP_lengths[*tempcode] +
4875 ((tempcode[3] == OP_PROP || tempcode[3] == OP_NOTPROP)? 2 : 0);
4876
4877 else if (*tempcode == OP_EXACT || *tempcode == OP_NOTEXACT)
4878 {
4879 tempcode += _pcre_OP_lengths[*tempcode];
4880 #ifdef SUPPORT_UTF8
4881 if (utf8 && tempcode[-1] >= 0xc0)
4882 tempcode += _pcre_utf8_table4[tempcode[-1] & 0x3f];
4883 #endif
4884 }
4885
4886 len = (int)(code - tempcode);
4887 if (len > 0) switch (*tempcode)
4888 {
4889 case OP_STAR: *tempcode = OP_POSSTAR; break;
4890 case OP_PLUS: *tempcode = OP_POSPLUS; break;
4891 case OP_QUERY: *tempcode = OP_POSQUERY; break;
4892 case OP_UPTO: *tempcode = OP_POSUPTO; break;
4893
4894 case OP_STARI: *tempcode = OP_POSSTARI; break;
4895 case OP_PLUSI: *tempcode = OP_POSPLUSI; break;
4896 case OP_QUERYI: *tempcode = OP_POSQUERYI; break;
4897 case OP_UPTOI: *tempcode = OP_POSUPTOI; break;
4898
4899 case OP_NOTSTAR: *tempcode = OP_NOTPOSSTAR; break;
4900 case OP_NOTPLUS: *tempcode = OP_NOTPOSPLUS; break;
4901 case OP_NOTQUERY: *tempcode = OP_NOTPOSQUERY; break;
4902 case OP_NOTUPTO: *tempcode = OP_NOTPOSUPTO; break;
4903
4904 case OP_NOTSTARI: *tempcode = OP_NOTPOSSTARI; break;
4905 case OP_NOTPLUSI: *tempcode = OP_NOTPOSPLUSI; break;
4906 case OP_NOTQUERYI: *tempcode = OP_NOTPOSQUERYI; break;
4907 case OP_NOTUPTOI: *tempcode = OP_NOTPOSUPTOI; break;
4908
4909 case OP_TYPESTAR: *tempcode = OP_TYPEPOSSTAR; break;
4910 case OP_TYPEPLUS: *tempcode = OP_TYPEPOSPLUS; break;
4911 case OP_TYPEQUERY: *tempcode = OP_TYPEPOSQUERY; break;
4912 case OP_TYPEUPTO: *tempcode = OP_TYPEPOSUPTO; break;
4913
4914 /* Because we are moving code along, we must ensure that any
4915 pending recursive references are updated. */
4916
4917 default:
4918 *code = OP_END;
4919 adjust_recurse(tempcode, 1 + LINK_SIZE, utf8, cd, save_hwm);
4920 memmove(tempcode + 1+LINK_SIZE, tempcode, len);
4921 code += 1 + LINK_SIZE;
4922 len += 1 + LINK_SIZE;
4923 tempcode[0] = OP_ONCE;
4924 *code++ = OP_KET;
4925 PUTINC(code, 0, len);
4926 PUT(tempcode, 1, len);
4927 break;
4928 }
4929 }
4930
4931 /* In all case we no longer have a previous item. We also set the
4932 "follows varying string" flag for subsequently encountered reqbytes if
4933 it isn't already set and we have just passed a varying length item. */
4934
4935 END_REPEAT:
4936 previous = NULL;
4937 cd->req_varyopt |= reqvary;
4938 break;
4939
4940
4941 /* ===================================================================*/
4942 /* Start of nested parenthesized sub-expression, or comment or lookahead or
4943 lookbehind or option setting or condition or all the other extended
4944 parenthesis forms. */
4945
4946 case CHAR_LEFT_PARENTHESIS:
4947 newoptions = options;
4948 skipbytes = 0;
4949 bravalue = OP_CBRA;
4950 save_hwm = cd->hwm;
4951 reset_bracount = FALSE;
4952
4953 /* First deal with various "verbs" that can be introduced by '*'. */
4954
4955 if (*(++ptr) == CHAR_ASTERISK &&
4956 ((cd->ctypes[ptr[1]] & ctype_letter) != 0 || ptr[1] == ':'))
4957 {
4958 int i, namelen;
4959 int arglen = 0;
4960 const char *vn = verbnames;
4961 const uschar *name = ptr + 1;
4962 const uschar *arg = NULL;
4963 previous = NULL;
4964 while ((cd->ctypes[*++ptr] & ctype_letter) != 0) {};
4965 namelen = (int)(ptr - name);
4966
4967 if (*ptr == CHAR_COLON)
4968 {
4969 arg = ++ptr;
4970 while ((cd->ctypes[*ptr] & (ctype_letter|ctype_digit)) != 0
4971 || *ptr == '_') ptr++;
4972 arglen = (int)(ptr - arg);
4973 }
4974
4975 if (*ptr != CHAR_RIGHT_PARENTHESIS)
4976 {
4977 *errorcodeptr = ERR60;
4978 goto FAILED;
4979 }
4980
4981 /* Scan the table of verb names */
4982
4983 for (i = 0; i < verbcount; i++)
4984 {
4985 if (namelen == verbs[i].len &&
4986 strncmp((char *)name, vn, namelen) == 0)
4987 {
4988 /* Check for open captures before ACCEPT and convert it to
4989 ASSERT_ACCEPT if in an assertion. */
4990
4991 if (verbs[i].op == OP_ACCEPT)
4992 {
4993 open_capitem *oc;
4994 if (arglen != 0)
4995 {
4996 *errorcodeptr = ERR59;
4997 goto FAILED;
4998 }
4999 cd->had_accept = TRUE;
5000 for (oc = cd->open_caps; oc != NULL; oc = oc->next)
5001 {
5002 *code++ = OP_CLOSE;
5003 PUT2INC(code, 0, oc->number);
5004 }
5005 *code++ = (cd->assert_depth > 0)? OP_ASSERT_ACCEPT : OP_ACCEPT;
5006 }
5007
5008 /* Handle other cases with/without an argument */
5009
5010 else if (arglen == 0)
5011 {
5012 if (verbs[i].op < 0) /* Argument is mandatory */
5013 {
5014 *errorcodeptr = ERR66;
5015 goto FAILED;
5016 }
5017 *code = verbs[i].op;
5018 if (*code++ == OP_THEN)
5019 {
5020 PUT(code, 0, code - bcptr->current_branch - 1);
5021 code += LINK_SIZE;
5022 }
5023 }
5024
5025 else
5026 {
5027 if (verbs[i].op_arg < 0) /* Argument is forbidden */
5028 {
5029 *errorcodeptr = ERR59;
5030 goto FAILED;
5031 }
5032 *code = verbs[i].op_arg;
5033 if (*code++ == OP_THEN_ARG)
5034 {
5035 PUT(code, 0, code - bcptr->current_branch - 1);
5036 code += LINK_SIZE;
5037 }
5038 *code++ = arglen;
5039 memcpy(code, arg, arglen);
5040 code += arglen;
5041 *code++ = 0;
5042 }
5043
5044 break; /* Found verb, exit loop */
5045 }
5046
5047 vn += verbs[i].len + 1;
5048 }
5049
5050 if (i < verbcount) continue; /* Successfully handled a verb */
5051 *errorcodeptr = ERR60; /* Verb not recognized */
5052 goto FAILED;
5053 }
5054
5055 /* Deal with the extended parentheses; all are introduced by '?', and the
5056 appearance of any of them means that this is not a capturing group. */
5057
5058 else if (*ptr == CHAR_QUESTION_MARK)
5059 {
5060 int i, set, unset, namelen;
5061 int *optset;
5062 const uschar *name;
5063 uschar *slot;
5064
5065 switch (*(++ptr))
5066 {
5067 case CHAR_NUMBER_SIGN: /* Comment; skip to ket */
5068 ptr++;
5069 while (*ptr != 0 && *ptr != CHAR_RIGHT_PARENTHESIS) ptr++;
5070 if (*ptr == 0)
5071 {
5072 *errorcodeptr = ERR18;
5073 goto FAILED;
5074 }
5075 continue;
5076
5077
5078 /* ------------------------------------------------------------ */
5079 case CHAR_VERTICAL_LINE: /* Reset capture count for each branch */
5080 reset_bracount = TRUE;
5081 /* Fall through */
5082
5083 /* ------------------------------------------------------------ */
5084 case CHAR_COLON: /* Non-capturing bracket */
5085 bravalue = OP_BRA;
5086 ptr++;
5087 break;
5088
5089
5090 /* ------------------------------------------------------------ */
5091 case CHAR_LEFT_PARENTHESIS:
5092 bravalue = OP_COND; /* Conditional group */
5093
5094 /* A condition can be an assertion, a number (referring to a numbered
5095 group), a name (referring to a named group), or 'R', referring to
5096 recursion. R<digits> and R&name are also permitted for recursion tests.
5097
5098 There are several syntaxes for testing a named group: (?(name)) is used
5099 by Python; Perl 5.10 onwards uses (?(<name>) or (?('name')).
5100
5101 There are two unfortunate ambiguities, caused by history. (a) 'R' can
5102 be the recursive thing or the name 'R' (and similarly for 'R' followed
5103 by digits), and (b) a number could be a name that consists of digits.
5104 In both cases, we look for a name first; if not found, we try the other
5105 cases. */
5106
5107 /* For conditions that are assertions, check the syntax, and then exit
5108 the switch. This will take control down to where bracketed groups,
5109 including assertions, are processed. */
5110
5111 if (ptr[1] == CHAR_QUESTION_MARK && (ptr[2] == CHAR_EQUALS_SIGN ||
5112 ptr[2] == CHAR_EXCLAMATION_MARK || ptr[2] == CHAR_LESS_THAN_SIGN))
5113 break;
5114
5115 /* Most other conditions use OP_CREF (a couple change to OP_RREF
5116 below), and all need to skip 3 bytes at the start of the group. */
5117
5118 code[1+LINK_SIZE] = OP_CREF;
5119 skipbytes = 3;
5120 refsign = -1;
5121
5122 /* Check for a test for recursion in a named group. */
5123
5124 if (ptr[1] == CHAR_R && ptr[2] == CHAR_AMPERSAND)
5125 {
5126 terminator = -1;
5127 ptr += 2;
5128 code[1+LINK_SIZE] = OP_RREF; /* Change the type of test */
5129 }
5130
5131 /* Check for a test for a named group's having been set, using the Perl
5132 syntax (?(<name>) or (?('name') */
5133
5134 else if (ptr[1] == CHAR_LESS_THAN_SIGN)
5135 {
5136 terminator = CHAR_GREATER_THAN_SIGN;
5137 ptr++;
5138 }
5139 else if (ptr[1] == CHAR_APOSTROPHE)
5140 {
5141 terminator = CHAR_APOSTROPHE;
5142 ptr++;
5143 }
5144 else
5145 {
5146 terminator = 0;
5147 if (ptr[1] == CHAR_MINUS || ptr[1] == CHAR_PLUS) refsign = *(++ptr);
5148 }
5149
5150 /* We now expect to read a name; any thing else is an error */
5151
5152 if ((cd->ctypes[ptr[1]] & ctype_word) == 0)
5153 {
5154 ptr += 1; /* To get the right offset */
5155 *errorcodeptr = ERR28;
5156 goto FAILED;
5157 }
5158
5159 /* Read the name, but also get it as a number if it's all digits */
5160
5161 recno = 0;
5162 name = ++ptr;
5163 while ((cd->ctypes[*ptr] & ctype_word) != 0)
5164 {
5165 if (recno >= 0)
5166 recno = ((digitab[*ptr] & ctype_digit) != 0)?
5167 recno * 10 + *ptr - CHAR_0 : -1;
5168 ptr++;
5169 }
5170 namelen = (int)(ptr - name);
5171
5172 if ((terminator > 0 && *ptr++ != terminator) ||
5173 *ptr++ != CHAR_RIGHT_PARENTHESIS)
5174 {
5175 ptr--; /* Error offset */
5176 *errorcodeptr = ERR26;
5177 goto FAILED;
5178 }
5179
5180 /* Do no further checking in the pre-compile phase. */
5181
5182 if (lengthptr != NULL) break;
5183
5184 /* In the real compile we do the work of looking for the actual
5185 reference. If the string started with "+" or "-" we require the rest to
5186 be digits, in which case recno will be set. */
5187
5188 if (refsign > 0)
5189 {
5190 if (recno <= 0)
5191 {
5192 *errorcodeptr = ERR58;
5193 goto FAILED;
5194 }
5195 recno = (refsign == CHAR_MINUS)?
5196 cd->bracount - recno + 1 : recno +cd->bracount;
5197 if (recno <= 0 || recno > cd->final_bracount)
5198 {
5199 *errorcodeptr = ERR15;
5200 goto FAILED;
5201 }
5202 PUT2(code, 2+LINK_SIZE, recno);
5203 break;
5204 }
5205
5206 /* Otherwise (did not start with "+" or "-"), start by looking for the
5207 name. If we find a name, add one to the opcode to change OP_CREF or
5208 OP_RREF into OP_NCREF or OP_NRREF. These behave exactly the same,
5209 except they record that the reference was originally to a name. The
5210 information is used to check duplicate names. */
5211
5212 slot = cd->name_table;
5213 for (i = 0; i < cd->names_found; i++)
5214 {
5215 if (strncmp((char *)name, (char *)slot+2, namelen) == 0) break;
5216 slot += cd->name_entry_size;
5217 }
5218
5219 /* Found a previous named subpattern */
5220
5221 if (i < cd->names_found)
5222 {
5223 recno = GET2(slot, 0);
5224 PUT2(code, 2+LINK_SIZE, recno);
5225 code[1+LINK_SIZE]++;
5226 }
5227
5228 /* Search the pattern for a forward reference */
5229
5230 else if ((i = find_parens(cd, name, namelen,
5231 (options & PCRE_EXTENDED) != 0, utf8)) > 0)
5232 {
5233 PUT2(code, 2+LINK_SIZE, i);
5234 code[1+LINK_SIZE]++;
5235 }
5236
5237 /* If terminator == 0 it means that the name followed directly after
5238 the opening parenthesis [e.g. (?(abc)...] and in this case there are
5239 some further alternatives to try. For the cases where terminator != 0
5240 [things like (?(<name>... or (?('name')... or (?(R&name)... ] we have
5241 now checked all the possibilities, so give an error. */
5242
5243 else if (terminator != 0)
5244 {
5245 *errorcodeptr = ERR15;
5246 goto FAILED;
5247 }
5248
5249 /* Check for (?(R) for recursion. Allow digits after R to specify a
5250 specific group number. */
5251
5252 else if (*name == CHAR_R)
5253 {
5254 recno = 0;
5255 for (i = 1; i < namelen; i++)
5256 {
5257 if ((digitab[name[i]] & ctype_digit) == 0)
5258 {
5259 *errorcodeptr = ERR15;
5260 goto FAILED;
5261 }
5262 recno = recno * 10 + name[i] - CHAR_0;
5263 }
5264 if (recno == 0) recno = RREF_ANY;
5265 code[1+LINK_SIZE] = OP_RREF; /* Change test type */
5266 PUT2(code, 2+LINK_SIZE, recno);
5267 }
5268
5269 /* Similarly, check for the (?(DEFINE) "condition", which is always
5270 false. */
5271
5272 else if (namelen == 6 && strncmp((char *)name, STRING_DEFINE, 6) == 0)
5273 {
5274 code[1+LINK_SIZE] = OP_DEF;
5275 skipbytes = 1;
5276 }
5277
5278 /* Check for the "name" actually being a subpattern number. We are
5279 in the second pass here, so final_bracount is set. */
5280
5281 else if (recno > 0 && recno <= cd->final_bracount)
5282 {
5283 PUT2(code, 2+LINK_SIZE, recno);
5284 }
5285
5286 /* Either an unidentified subpattern, or a reference to (?(0) */
5287
5288 else
5289 {
5290 *errorcodeptr = (recno == 0)? ERR35: ERR15;
5291 goto FAILED;
5292 }
5293 break;
5294
5295
5296 /* ------------------------------------------------------------ */
5297 case CHAR_EQUALS_SIGN: /* Positive lookahead */
5298 bravalue = OP_ASSERT;
5299 cd->assert_depth += 1;
5300 ptr++;
5301 break;
5302
5303
5304 /* ------------------------------------------------------------ */
5305 case CHAR_EXCLAMATION_MARK: /* Negative lookahead */
5306 ptr++;
5307 if (*ptr == CHAR_RIGHT_PARENTHESIS) /* Optimize (?!) */
5308 {
5309 *code++ = OP_FAIL;
5310 previous = NULL;
5311 continue;
5312 }
5313 bravalue = OP_ASSERT_NOT;
5314 cd->assert_depth += 1;
5315 break;
5316
5317
5318 /* ------------------------------------------------------------ */
5319 case CHAR_LESS_THAN_SIGN: /* Lookbehind or named define */
5320 switch (ptr[1])
5321 {
5322 case CHAR_EQUALS_SIGN: /* Positive lookbehind */
5323 bravalue = OP_ASSERTBACK;
5324 cd->assert_depth += 1;
5325 ptr += 2;
5326 break;
5327
5328 case CHAR_EXCLAMATION_MARK: /* Negative lookbehind */
5329 bravalue = OP_ASSERTBACK_NOT;
5330 cd->assert_depth += 1;
5331 ptr += 2;
5332 break;
5333
5334 default: /* Could be name define, else bad */
5335 if ((cd->ctypes[ptr[1]] & ctype_word) != 0) goto DEFINE_NAME;
5336 ptr++; /* Correct offset for error */
5337 *errorcodeptr = ERR24;
5338 goto FAILED;
5339 }
5340 break;
5341
5342
5343 /* ------------------------------------------------------------ */
5344 case CHAR_GREATER_THAN_SIGN: /* One-time brackets */
5345 bravalue = OP_ONCE;
5346 ptr++;
5347 break;
5348
5349
5350 /* ------------------------------------------------------------ */
5351 case CHAR_C: /* Callout - may be followed by digits; */
5352 previous_callout = code; /* Save for later completion */
5353 after_manual_callout = 1; /* Skip one item before completing */
5354 *code++ = OP_CALLOUT;
5355 {
5356 int n = 0;
5357 while ((digitab[*(++ptr)] & ctype_digit) != 0)
5358 n = n * 10 + *ptr - CHAR_0;
5359 if (*ptr != CHAR_RIGHT_PARENTHESIS)
5360 {
5361 *errorcodeptr = ERR39;
5362 goto FAILED;
5363 }
5364 if (n > 255)
5365 {
5366 *errorcodeptr = ERR38;
5367 goto FAILED;
5368 }
5369 *code++ = n;
5370 PUT(code, 0, (int)(ptr - cd->start_pattern + 1)); /* Pattern offset */
5371 PUT(code, LINK_SIZE, 0); /* Default length */
5372 code += 2 * LINK_SIZE;
5373 }
5374 previous = NULL;
5375 continue;
5376
5377
5378 /* ------------------------------------------------------------ */
5379 case CHAR_P: /* Python-style named subpattern handling */
5380 if (*(++ptr) == CHAR_EQUALS_SIGN ||
5381 *ptr == CHAR_GREATER_THAN_SIGN) /* Reference or recursion */
5382 {
5383 is_recurse = *ptr == CHAR_GREATER_THAN_SIGN;
5384 terminator = CHAR_RIGHT_PARENTHESIS;
5385 goto NAMED_REF_OR_RECURSE;
5386 }
5387 else if (*ptr != CHAR_LESS_THAN_SIGN) /* Test for Python-style defn */
5388 {
5389 *errorcodeptr = ERR41;
5390 goto FAILED;
5391 }
5392 /* Fall through to handle (?P< as (?< is handled */
5393
5394
5395 /* ------------------------------------------------------------ */
5396 DEFINE_NAME: /* Come here from (?< handling */
5397 case CHAR_APOSTROPHE:
5398 {
5399 terminator = (*ptr == CHAR_LESS_THAN_SIGN)?
5400 CHAR_GREATER_THAN_SIGN : CHAR_APOSTROPHE;
5401 name = ++ptr;
5402
5403 while ((cd->ctypes[*ptr] & ctype_word) != 0) ptr++;
5404 namelen = (int)(ptr - name);
5405
5406 /* In the pre-compile phase, just do a syntax check. */
5407
5408 if (lengthptr != NULL)
5409 {
5410 if (*ptr != terminator)
5411 {
5412 *errorcodeptr = ERR42;
5413 goto FAILED;
5414 }
5415 if (cd->names_found >= MAX_NAME_COUNT)
5416 {
5417 *errorcodeptr = ERR49;
5418 goto FAILED;
5419 }
5420 if (namelen + 3 > cd->name_entry_size)
5421 {
5422 cd->name_entry_size = namelen + 3;
5423 if (namelen > MAX_NAME_SIZE)
5424 {
5425 *errorcodeptr = ERR48;
5426 goto FAILED;
5427 }
5428 }
5429 }
5430
5431 /* In the real compile, create the entry in the table, maintaining
5432 alphabetical order. Duplicate names for different numbers are
5433 permitted only if PCRE_DUPNAMES is set. Duplicate names for the same
5434 number are always OK. (An existing number can be re-used if (?|
5435 appears in the pattern.) In either event, a duplicate name results in
5436 a duplicate entry in the table, even if the number is the same. This
5437 is because the number of names, and hence the table size, is computed
5438 in the pre-compile, and it affects various numbers and pointers which
5439 would all have to be modified, and the compiled code moved down, if
5440 duplicates with the same number were omitted from the table. This
5441 doesn't seem worth the hassle. However, *different* names for the
5442 same number are not permitted. */
5443
5444 else
5445 {
5446 BOOL dupname = FALSE;
5447 slot = cd->name_table;
5448
5449 for (i = 0; i < cd->names_found; i++)
5450 {
5451 int crc = memcmp(name, slot+2, namelen);
5452 if (crc == 0)
5453 {
5454 if (slot[2+namelen] == 0)
5455 {
5456 if (GET2(slot, 0) != cd->bracount + 1 &&
5457 (options & PCRE_DUPNAMES) == 0)
5458 {
5459 *errorcodeptr = ERR43;
5460 goto FAILED;
5461 }
5462 else dupname = TRUE;
5463 }
5464 else crc = -1; /* Current name is a substring */
5465 }
5466
5467 /* Make space in the table and break the loop for an earlier
5468 name. For a duplicate or later name, carry on. We do this for
5469 duplicates so that in the simple case (when ?(| is not used) they
5470 are in order of their numbers. */
5471
5472 if (crc < 0)
5473 {
5474 memmove(slot + cd->name_entry_size, slot,
5475 (cd->names_found - i) * cd->name_entry_size);
5476 break;
5477 }
5478
5479 /* Continue the loop for a later or duplicate name */
5480
5481 slot += cd->name_entry_size;
5482 }
5483
5484 /* For non-duplicate names, check for a duplicate number before
5485 adding the new name. */
5486
5487 if (!dupname)
5488 {
5489 uschar *cslot = cd->name_table;
5490 for (i = 0; i < cd->names_found; i++)
5491 {
5492 if (cslot != slot)
5493 {
5494 if (GET2(cslot, 0) == cd->bracount + 1)
5495 {
5496 *errorcodeptr = ERR65;
5497 goto FAILED;
5498 }
5499 }
5500 else i--;
5501 cslot += cd->name_entry_size;
5502 }
5503 }
5504
5505 PUT2(slot, 0, cd->bracount + 1);
5506 memcpy(slot + 2, name, namelen);
5507 slot[2+namelen] = 0;
5508 }
5509 }
5510
5511 /* In both pre-compile and compile, count the number of names we've
5512 encountered. */
5513
5514 cd->names_found++;
5515 ptr++; /* Move past > or ' */
5516 goto NUMBERED_GROUP;
5517
5518
5519 /* ------------------------------------------------------------ */
5520 case CHAR_AMPERSAND: /* Perl recursion/subroutine syntax */
5521 terminator = CHAR_RIGHT_PARENTHESIS;
5522 is_recurse = TRUE;
5523 /* Fall through */
5524
5525 /* We come here from the Python syntax above that handles both
5526 references (?P=name) and recursion (?P>name), as well as falling
5527 through from the Perl recursion syntax (?&name). We also come here from
5528 the Perl \k<name> or \k'name' back reference syntax and the \k{name}
5529 .NET syntax, and the Oniguruma \g<...> and \g'...' subroutine syntax. */
5530
5531 NAMED_REF_OR_RECURSE:
5532 name = ++ptr;
5533 while ((cd->ctypes[*ptr] & ctype_word) != 0) ptr++;
5534 namelen = (int)(ptr - name);
5535
5536 /* In the pre-compile phase, do a syntax check. We used to just set
5537 a dummy reference number, because it was not used in the first pass.
5538 However, with the change of recursive back references to be atomic,
5539 we have to look for the number so that this state can be identified, as
5540 otherwise the incorrect length is computed. If it's not a backwards
5541 reference, the dummy number will do. */
5542
5543 if (lengthptr != NULL)
5544 {
5545 const uschar *temp;
5546
5547 if (namelen == 0)
5548 {
5549 *errorcodeptr = ERR62;
5550 goto FAILED;
5551 }
5552 if (*ptr != terminator)
5553 {
5554 *errorcodeptr = ERR42;
5555 goto FAILED;
5556 }
5557 if (namelen > MAX_NAME_SIZE)
5558 {
5559 *errorcodeptr = ERR48;
5560 goto FAILED;
5561 }
5562
5563 /* The name table does not exist in the first pass, so we cannot
5564 do a simple search as in the code below. Instead, we have to scan the
5565 pattern to find the number. It is important that we scan it only as
5566 far as we have got because the syntax of named subpatterns has not
5567 been checked for the rest of the pattern, and find_parens() assumes
5568 correct syntax. In any case, it's a waste of resources to scan
5569 further. We stop the scan at the current point by temporarily
5570 adjusting the value of cd->endpattern. */
5571
5572 temp = cd->end_pattern;
5573 cd->end_pattern = ptr;
5574 recno = find_parens(cd, name, namelen,
5575 (options & PCRE_EXTENDED) != 0, utf8);
5576 cd->end_pattern = temp;
5577 if (recno < 0) recno = 0; /* Forward ref; set dummy number */
5578 }
5579
5580 /* In the real compile, seek the name in the table. We check the name
5581 first, and then check that we have reached the end of the name in the
5582 table. That way, if the name that is longer than any in the table,
5583 the comparison will fail without reading beyond the table entry. */
5584
5585 else
5586 {
5587 slot = cd->name_table;
5588 for (i = 0; i < cd->names_found; i++)
5589 {
5590 if (strncmp((char *)name, (char *)slot+2, namelen) == 0 &&
5591 slot[2+namelen] == 0)
5592 break;
5593 slot += cd->name_entry_size;
5594 }
5595
5596 if (i < cd->names_found) /* Back reference */
5597 {
5598 recno = GET2(slot, 0);
5599 }
5600 else if ((recno = /* Forward back reference */
5601 find_parens(cd, name, namelen,
5602 (options & PCRE_EXTENDED) != 0, utf8)) <= 0)
5603 {
5604 *errorcodeptr = ERR15;
5605 goto FAILED;
5606 }
5607 }
5608
5609 /* In both phases, we can now go to the code than handles numerical
5610 recursion or backreferences. */
5611
5612 if (is_recurse) goto HANDLE_RECURSION;
5613 else goto HANDLE_REFERENCE;
5614
5615
5616 /* ------------------------------------------------------------ */
5617 case CHAR_R: /* Recursion */
5618 ptr++; /* Same as (?0) */
5619 /* Fall through */
5620
5621
5622 /* ------------------------------------------------------------ */
5623 case CHAR_MINUS: case CHAR_PLUS: /* Recursion or subroutine */
5624 case CHAR_0: case CHAR_1: case CHAR_2: case CHAR_3: case CHAR_4:
5625 case CHAR_5: case CHAR_6: case CHAR_7: case CHAR_8: case CHAR_9:
5626 {
5627 const uschar *called;
5628 terminator = CHAR_RIGHT_PARENTHESIS;
5629
5630 /* Come here from the \g<...> and \g'...' code (Oniguruma
5631 compatibility). However, the syntax has been checked to ensure that
5632 the ... are a (signed) number, so that neither ERR63 nor ERR29 will
5633 be called on this path, nor with the jump to OTHER_CHAR_AFTER_QUERY
5634 ever be taken. */
5635
5636 HANDLE_NUMERICAL_RECURSION:
5637
5638 if ((refsign = *ptr) == CHAR_PLUS)
5639 {
5640 ptr++;
5641 if ((digitab[*ptr] & ctype_digit) == 0)
5642 {
5643 *errorcodeptr = ERR63;
5644 goto FAILED;
5645 }
5646 }
5647 else if (refsign == CHAR_MINUS)
5648 {
5649 if ((digitab[ptr[1]] & ctype_digit) == 0)
5650 goto OTHER_CHAR_AFTER_QUERY;
5651 ptr++;
5652 }
5653
5654 recno = 0;
5655 while((digitab[*ptr] & ctype_digit) != 0)
5656 recno = recno * 10 + *ptr++ - CHAR_0;
5657
5658 if (*ptr != terminator)
5659 {
5660 *errorcodeptr = ERR29;
5661 goto FAILED;
5662 }
5663
5664 if (refsign == CHAR_MINUS)
5665 {
5666 if (recno == 0)
5667 {
5668 *errorcodeptr = ERR58;
5669 goto FAILED;
5670 }
5671 recno = cd->bracount - recno + 1;
5672 if (recno <= 0)
5673 {
5674 *errorcodeptr = ERR15;
5675 goto FAILED;
5676 }
5677 }
5678 else if (refsign == CHAR_PLUS)
5679 {
5680 if (recno == 0)
5681 {
5682 *errorcodeptr = ERR58;
5683 goto FAILED;
5684 }
5685 recno += cd->bracount;
5686 }
5687
5688 /* Come here from code above that handles a named recursion */
5689
5690 HANDLE_RECURSION:
5691
5692 previous = code;
5693 called = cd->start_code;
5694
5695 /* When we are actually compiling, find the bracket that is being
5696 referenced. Temporarily end the regex in case it doesn't exist before
5697 this point. If we end up with a forward reference, first check that
5698 the bracket does occur later so we can give the error (and position)
5699 now. Then remember this forward reference in the workspace so it can
5700 be filled in at the end. */
5701
5702 if (lengthptr == NULL)
5703 {
5704 *code = OP_END;
5705 if (recno != 0)
5706 called = _pcre_find_bracket(cd->start_code, utf8, recno);
5707
5708 /* Forward reference */
5709
5710 if (called == NULL)
5711 {
5712 if (find_parens(cd, NULL, recno,
5713 (options & PCRE_EXTENDED) != 0, utf8) < 0)
5714 {
5715 *errorcodeptr = ERR15;
5716 goto FAILED;
5717 }
5718
5719 /* Fudge the value of "called" so that when it is inserted as an
5720 offset below, what it actually inserted is the reference number
5721 of the group. Then remember the forward reference. */
5722
5723 called = cd->start_code + recno;
5724 PUTINC(cd->hwm, 0, (int)(code + 1 - cd->start_code));
5725 }
5726
5727 /* If not a forward reference, and the subpattern is still open,
5728 this is a recursive call. We check to see if this is a left
5729 recursion that could loop for ever, and diagnose that case. */
5730
5731 else if (GET(called, 1) == 0 &&
5732 could_be_empty(called, code, bcptr, utf8, cd))
5733 {
5734 *errorcodeptr = ERR40;
5735 goto FAILED;
5736 }
5737 }
5738
5739 /* Insert the recursion/subroutine item. */
5740
5741 *code = OP_RECURSE;
5742 PUT(code, 1, (int)(called - cd->start_code));
5743 code += 1 + LINK_SIZE;
5744 }
5745
5746 /* Can't determine a first byte now */
5747
5748 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
5749 continue;
5750
5751
5752 /* ------------------------------------------------------------ */
5753 default: /* Other characters: check option setting */
5754 OTHER_CHAR_AFTER_QUERY:
5755 set = unset = 0;
5756 optset = &set;
5757
5758 while (*ptr != CHAR_RIGHT_PARENTHESIS && *ptr != CHAR_COLON)
5759 {
5760 switch (*ptr++)
5761 {
5762 case CHAR_MINUS: optset = &unset; break;
5763
5764 case CHAR_J: /* Record that it changed in the external options */
5765 *optset |= PCRE_DUPNAMES;
5766 cd->external_flags |= PCRE_JCHANGED;
5767 break;
5768
5769 case CHAR_i: *optset |= PCRE_CASELESS; break;
5770 case CHAR_m: *optset |= PCRE_MULTILINE; break;
5771 case CHAR_s: *optset |= PCRE_DOTALL; break;
5772 case CHAR_x: *optset |= PCRE_EXTENDED; break;
5773 case CHAR_U: *optset |= PCRE_UNGREEDY; break;
5774 case CHAR_X: *optset |= PCRE_EXTRA; break;
5775
5776 default: *errorcodeptr = ERR12;
5777 ptr--; /* Correct the offset */
5778 goto FAILED;
5779 }
5780 }
5781
5782 /* Set up the changed option bits, but don't change anything yet. */
5783
5784 newoptions = (options | set) & (~unset);
5785
5786 /* If the options ended with ')' this is not the start of a nested
5787 group with option changes, so the options change at this level. If this
5788 item is right at the start of the pattern, the options can be
5789 abstracted and made external in the pre-compile phase, and ignored in
5790 the compile phase. This can be helpful when matching -- for instance in
5791 caseless checking of required bytes.
5792
5793 If the code pointer is not (cd->start_code + 1 + LINK_SIZE), we are
5794 definitely *not* at the start of the pattern because something has been
5795 compiled. In the pre-compile phase, however, the code pointer can have
5796 that value after the start, because it gets reset as code is discarded
5797 during the pre-compile. However, this can happen only at top level - if
5798 we are within parentheses, the starting BRA will still be present. At
5799 any parenthesis level, the length value can be used to test if anything
5800 has been compiled at that level. Thus, a test for both these conditions
5801 is necessary to ensure we correctly detect the start of the pattern in
5802 both phases.
5803
5804 If we are not at the pattern start, reset the greedy defaults and the
5805 case value for firstbyte and reqbyte. */
5806
5807 if (*ptr == CHAR_RIGHT_PARENTHESIS)
5808 {
5809 if (code == cd->start_code + 1 + LINK_SIZE &&
5810 (lengthptr == NULL || *lengthptr == 2 + 2*LINK_SIZE))
5811 {
5812 cd->external_options = newoptions;
5813 }
5814 else
5815 {
5816 greedy_default = ((newoptions & PCRE_UNGREEDY) != 0);
5817 greedy_non_default = greedy_default ^ 1;
5818 req_caseopt = ((newoptions & PCRE_CASELESS) != 0)? REQ_CASELESS : 0;
5819 }
5820
5821 /* Change options at this level, and pass them back for use
5822 in subsequent branches. */
5823
5824 *optionsptr = options = newoptions;
5825 previous = NULL; /* This item can't be repeated */
5826 continue; /* It is complete */
5827 }
5828
5829 /* If the options ended with ':' we are heading into a nested group
5830 with possible change of options. Such groups are non-capturing and are
5831 not assertions of any kind. All we need to do is skip over the ':';
5832 the newoptions value is handled below. */
5833
5834 bravalue = OP_BRA;
5835 ptr++;
5836 } /* End of switch for character following (? */
5837 } /* End of (? handling */
5838
5839 /* Opening parenthesis not followed by '*' or '?'. If PCRE_NO_AUTO_CAPTURE
5840 is set, all unadorned brackets become non-capturing and behave like (?:...)
5841 brackets. */
5842
5843 else if ((options & PCRE_NO_AUTO_CAPTURE) != 0)
5844 {
5845 bravalue = OP_BRA;
5846 }
5847
5848 /* Else we have a capturing group. */
5849
5850 else
5851 {
5852 NUMBERED_GROUP:
5853 cd->bracount += 1;
5854 PUT2(code, 1+LINK_SIZE, cd->bracount);
5855 skipbytes = 2;
5856 }
5857
5858 /* Process nested bracketed regex. Assertions may not be repeated, but
5859 other kinds can be. All their opcodes are >= OP_ONCE. We copy code into a
5860 non-register variable (tempcode) in order to be able to pass its address
5861 because some compilers complain otherwise. */
5862
5863 previous = (bravalue >= OP_ONCE)? code : NULL;
5864 *code = bravalue;
5865 tempcode = code;
5866 tempreqvary = cd->req_varyopt; /* Save value before bracket */
5867 length_prevgroup = 0; /* Initialize for pre-compile phase */
5868
5869 if (!compile_regex(
5870 newoptions, /* The complete new option state */
5871 &tempcode, /* Where to put code (updated) */
5872 &ptr, /* Input pointer (updated) */
5873 errorcodeptr, /* Where to put an error message */
5874 (bravalue == OP_ASSERTBACK ||
5875 bravalue == OP_ASSERTBACK_NOT), /* TRUE if back assert */
5876 reset_bracount, /* True if (?| group */
5877 skipbytes, /* Skip over bracket number */
5878 &subfirstbyte, /* For possible first char */
5879 &subreqbyte, /* For possible last char */
5880 bcptr, /* Current branch chain */
5881 cd, /* Tables block */
5882 (lengthptr == NULL)? NULL : /* Actual compile phase */
5883 &length_prevgroup /* Pre-compile phase */
5884 ))
5885 goto FAILED;
5886
5887 if (bravalue >= OP_ASSERT && bravalue <= OP_ASSERTBACK_NOT)
5888 cd->assert_depth -= 1;
5889
5890 /* At the end of compiling, code is still pointing to the start of the
5891 group, while tempcode has been updated to point past the end of the group
5892 and any option resetting that may follow it. The pattern pointer (ptr)
5893 is on the bracket. */
5894
5895 /* If this is a conditional bracket, check that there are no more than
5896 two branches in the group, or just one if it's a DEFINE group. We do this
5897 in the real compile phase, not in the pre-pass, where the whole group may
5898 not be available. */
5899
5900 if (bravalue == OP_COND && lengthptr == NULL)
5901 {
5902 uschar *tc = code;
5903 int condcount = 0;
5904
5905 do {
5906 condcount++;
5907 tc += GET(tc,1);
5908 }
5909 while (*tc != OP_KET);
5910
5911 /* A DEFINE group is never obeyed inline (the "condition" is always
5912 false). It must have only one branch. */
5913
5914 if (code[LINK_SIZE+1] == OP_DEF)
5915 {
5916 if (condcount > 1)
5917 {
5918 *errorcodeptr = ERR54;
5919 goto FAILED;
5920 }
5921 bravalue = OP_DEF; /* Just a flag to suppress char handling below */
5922 }
5923
5924 /* A "normal" conditional group. If there is just one branch, we must not
5925 make use of its firstbyte or reqbyte, because this is equivalent to an
5926 empty second branch. */
5927
5928 else
5929 {
5930 if (condcount > 2)
5931 {
5932 *errorcodeptr = ERR27;
5933 goto FAILED;
5934 }
5935 if (condcount == 1) subfirstbyte = subreqbyte = REQ_NONE;
5936 }
5937 }
5938
5939 /* Error if hit end of pattern */
5940
5941 if (*ptr != CHAR_RIGHT_PARENTHESIS)
5942 {
5943 *errorcodeptr = ERR14;
5944 goto FAILED;
5945 }
5946
5947 /* In the pre-compile phase, update the length by the length of the group,
5948 less the brackets at either end. Then reduce the compiled code to just a
5949 set of non-capturing brackets so that it doesn't use much memory if it is
5950 duplicated by a quantifier.*/
5951
5952 if (lengthptr != NULL)
5953 {
5954 if (OFLOW_MAX - *lengthptr < length_prevgroup - 2 - 2*LINK_SIZE)
5955 {
5956 *errorcodeptr = ERR20;
5957 goto FAILED;
5958 }
5959 *lengthptr += length_prevgroup - 2 - 2*LINK_SIZE;
5960 code++; /* This already contains bravalue */
5961 PUTINC(code, 0, 1 + LINK_SIZE);
5962 *code++ = OP_KET;
5963 PUTINC(code, 0, 1 + LINK_SIZE);
5964 break; /* No need to waste time with special character handling */
5965 }
5966
5967 /* Otherwise update the main code pointer to the end of the group. */
5968
5969 code = tempcode;
5970
5971 /* For a DEFINE group, required and first character settings are not
5972 relevant. */
5973
5974 if (bravalue == OP_DEF) break;
5975
5976 /* Handle updating of the required and first characters for other types of
5977 group. Update for normal brackets of all kinds, and conditions with two
5978 branches (see code above). If the bracket is followed by a quantifier with
5979 zero repeat, we have to back off. Hence the definition of zeroreqbyte and
5980 zerofirstbyte outside the main loop so that they can be accessed for the
5981 back off. */
5982
5983 zeroreqbyte = reqbyte;
5984 zerofirstbyte = firstbyte;
5985 groupsetfirstbyte = FALSE;
5986
5987 if (bravalue >= OP_ONCE)
5988 {
5989 /* If we have not yet set a firstbyte in this branch, take it from the
5990 subpattern, remembering that it was set here so that a repeat of more
5991 than one can replicate it as reqbyte if necessary. If the subpattern has
5992 no firstbyte, set "none" for the whole branch. In both cases, a zero
5993 repeat forces firstbyte to "none". */
5994
5995 if (firstbyte == REQ_UNSET)
5996 {
5997 if (subfirstbyte >= 0)
5998 {
5999 firstbyte = subfirstbyte;
6000 groupsetfirstbyte = TRUE;
6001 }
6002 else firstbyte = REQ_NONE;
6003 zerofirstbyte = REQ_NONE;
6004 }
6005
6006 /* If firstbyte was previously set, convert the subpattern's firstbyte
6007 into reqbyte if there wasn't one, using the vary flag that was in
6008 existence beforehand. */
6009
6010 else if (subfirstbyte >= 0 && subreqbyte < 0)
6011 subreqbyte = subfirstbyte | tempreqvary;
6012
6013 /* If the subpattern set a required byte (or set a first byte that isn't
6014 really the first byte - see above), set it. */
6015
6016 if (subreqbyte >= 0) reqbyte = subreqbyte;
6017 }
6018
6019 /* For a forward assertion, we take the reqbyte, if set. This can be
6020 helpful if the pattern that follows the assertion doesn't set a different
6021 char. For example, it's useful for /(?=abcde).+/. We can't set firstbyte
6022 for an assertion, however because it leads to incorrect effect for patterns
6023 such as /(?=a)a.+/ when the "real" "a" would then become a reqbyte instead
6024 of a firstbyte. This is overcome by a scan at the end if there's no
6025 firstbyte, looking for an asserted first char. */
6026
6027 else if (bravalue == OP_ASSERT && subreqbyte >= 0) reqbyte = subreqbyte;
6028 break; /* End of processing '(' */
6029
6030
6031 /* ===================================================================*/
6032 /* Handle metasequences introduced by \. For ones like \d, the ESC_ values
6033 are arranged to be the negation of the corresponding OP_values in the
6034 default case when PCRE_UCP is not set. For the back references, the values
6035 are ESC_REF plus the reference number. Only back references and those types
6036 that consume a character may be repeated. We can test for values between
6037 ESC_b and ESC_Z for the latter; this may have to change if any new ones are
6038 ever created. */
6039
6040 case CHAR_BACKSLASH:
6041 tempptr = ptr;
6042 c = check_escape(&ptr, errorcodeptr, cd->bracount, options, FALSE);
6043 if (*errorcodeptr != 0) goto FAILED;
6044
6045 if (c < 0)
6046 {
6047 if (-c == ESC_Q) /* Handle start of quoted string */
6048 {
6049 if (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
6050 ptr += 2; /* avoid empty string */
6051 else inescq = TRUE;
6052 continue;
6053 }
6054
6055 if (-c == ESC_E) continue; /* Perl ignores an orphan \E */
6056
6057 /* For metasequences that actually match a character, we disable the
6058 setting of a first character if it hasn't already been set. */
6059
6060 if (firstbyte == REQ_UNSET && -c > ESC_b && -c < ESC_Z)
6061 firstbyte = REQ_NONE;
6062
6063 /* Set values to reset to if this is followed by a zero repeat. */
6064
6065 zerofirstbyte = firstbyte;
6066 zeroreqbyte = reqbyte;
6067
6068 /* \g<name> or \g'name' is a subroutine call by name and \g<n> or \g'n'
6069 is a subroutine call by number (Oniguruma syntax). In fact, the value
6070 -ESC_g is returned only for these cases. So we don't need to check for <
6071 or ' if the value is -ESC_g. For the Perl syntax \g{n} the value is
6072 -ESC_REF+n, and for the Perl syntax \g{name} the result is -ESC_k (as
6073 that is a synonym for a named back reference). */
6074
6075 if (-c == ESC_g)
6076 {
6077 const uschar *p;
6078 save_hwm = cd->hwm; /* Normally this is set when '(' is read */
6079 terminator = (*(++ptr) == CHAR_LESS_THAN_SIGN)?
6080 CHAR_GREATER_THAN_SIGN : CHAR_APOSTROPHE;
6081
6082 /* These two statements stop the compiler for warning about possibly
6083 unset variables caused by the jump to HANDLE_NUMERICAL_RECURSION. In
6084 fact, because we actually check for a number below, the paths that
6085 would actually be in error are never taken. */
6086
6087 skipbytes = 0;
6088 reset_bracount = FALSE;
6089
6090 /* Test for a name */
6091
6092 if (ptr[1] != CHAR_PLUS && ptr[1] != CHAR_MINUS)
6093 {
6094 BOOL isnumber = TRUE;
6095 for (p = ptr + 1; *p != 0 && *p != terminator; p++)
6096 {
6097 if ((cd->ctypes[*p] & ctype_digit) == 0) isnumber = FALSE;
6098 if ((cd->ctypes[*p] & ctype_word) == 0) break;
6099 }
6100 if (*p != terminator)
6101 {
6102 *errorcodeptr = ERR57;
6103 break;
6104 }
6105 if (isnumber)
6106 {
6107 ptr++;
6108 goto HANDLE_NUMERICAL_RECURSION;
6109 }
6110 is_recurse = TRUE;
6111 goto NAMED_REF_OR_RECURSE;
6112 }
6113
6114 /* Test a signed number in angle brackets or quotes. */
6115
6116 p = ptr + 2;
6117 while ((digitab[*p] & ctype_digit) != 0) p++;
6118 if (*p != terminator)
6119 {
6120 *errorcodeptr = ERR57;
6121 break;
6122 }
6123 ptr++;
6124 goto HANDLE_NUMERICAL_RECURSION;
6125 }
6126
6127 /* \k<name> or \k'name' is a back reference by name (Perl syntax).
6128 We also support \k{name} (.NET syntax) */
6129
6130 if (-c == ESC_k && (ptr[1] == CHAR_LESS_THAN_SIGN ||
6131 ptr[1] == CHAR_APOSTROPHE || ptr[1] == CHAR_LEFT_CURLY_BRACKET))
6132 {
6133 is_recurse = FALSE;
6134 terminator = (*(++ptr) == CHAR_LESS_THAN_SIGN)?
6135 CHAR_GREATER_THAN_SIGN : (*ptr == CHAR_APOSTROPHE)?
6136 CHAR_APOSTROPHE : CHAR_RIGHT_CURLY_BRACKET;
6137 goto NAMED_REF_OR_RECURSE;
6138 }
6139
6140 /* Back references are handled specially; must disable firstbyte if
6141 not set to cope with cases like (?=(\w+))\1: which would otherwise set
6142 ':' later. */
6143
6144 if (-c >= ESC_REF)
6145 {
6146 open_capitem *oc;
6147 recno = -c - ESC_REF;
6148
6149 HANDLE_REFERENCE: /* Come here from named backref handling */
6150 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
6151 previous = code;
6152 *code++ = ((options & PCRE_CASELESS) != 0)? OP_REFI : OP_REF;
6153 PUT2INC(code, 0, recno);
6154 cd->backref_map |= (recno < 32)? (1 << recno) : 1;
6155 if (recno > cd->top_backref) cd->top_backref = recno;
6156
6157 /* Check to see if this back reference is recursive, that it, it
6158 is inside the group that it references. A flag is set so that the
6159 group can be made atomic. */
6160
6161 for (oc = cd->open_caps; oc != NULL; oc = oc->next)
6162 {
6163 if (oc->number == recno)
6164 {
6165 oc->flag = TRUE;
6166 break;
6167 }
6168 }
6169 }
6170
6171 /* So are Unicode property matches, if supported. */
6172
6173 #ifdef SUPPORT_UCP
6174 else if (-c == ESC_P || -c == ESC_p)
6175 {
6176 BOOL negated;
6177 int pdata;
6178 int ptype = get_ucp(&ptr, &negated, &pdata, errorcodeptr);
6179 if (ptype < 0) goto FAILED;
6180 previous = code;
6181 *code++ = ((-c == ESC_p) != negated)? OP_PROP : OP_NOTPROP;
6182 *code++ = ptype;
6183 *code++ = pdata;
6184 }
6185 #else
6186
6187 /* If Unicode properties are not supported, \X, \P, and \p are not
6188 allowed. */
6189
6190 else if (-c == ESC_X || -c == ESC_P || -c == ESC_p)
6191 {
6192 *errorcodeptr = ERR45;
6193 goto FAILED;
6194 }
6195 #endif
6196
6197 /* For the rest (including \X when Unicode properties are supported), we
6198 can obtain the OP value by negating the escape value in the default
6199 situation when PCRE_UCP is not set. When it *is* set, we substitute
6200 Unicode property tests. */
6201
6202 else
6203 {
6204 #ifdef SUPPORT_UCP
6205 if (-c >= ESC_DU && -c <= ESC_wu)
6206 {
6207 nestptr = ptr + 1; /* Where to resume */
6208 ptr = substitutes[-c - ESC_DU] - 1; /* Just before substitute */
6209 }
6210 else
6211 #endif
6212 {
6213 previous = (-c > ESC_b && -c < ESC_Z)? code : NULL;
6214 *code++ = -c;
6215 }
6216 }
6217 continue;
6218 }
6219
6220 /* We have a data character whose value is in c. In UTF-8 mode it may have
6221 a value > 127. We set its representation in the length/buffer, and then
6222 handle it as a data character. */
6223
6224 #ifdef SUPPORT_UTF8
6225 if (utf8 && c > 127)
6226 mclength = _pcre_ord2utf8(c, mcbuffer);
6227 else
6228 #endif
6229
6230 {
6231 mcbuffer[0] = c;
6232 mclength = 1;
6233 }
6234 goto ONE_CHAR;
6235
6236
6237 /* ===================================================================*/
6238 /* Handle a literal character. It is guaranteed not to be whitespace or #
6239 when the extended flag is set. If we are in UTF-8 mode, it may be a
6240 multi-byte literal character. */
6241
6242 default:
6243 NORMAL_CHAR:
6244 mclength = 1;
6245 mcbuffer[0] = c;
6246
6247 #ifdef SUPPORT_UTF8
6248 if (utf8 && c >= 0xc0)
6249 {
6250 while ((ptr[1] & 0xc0) == 0x80)
6251 mcbuffer[mclength++] = *(++ptr);
6252 }
6253 #endif
6254
6255 /* At this point we have the character's bytes in mcbuffer, and the length
6256 in mclength. When not in UTF-8 mode, the length is always 1. */
6257
6258 ONE_CHAR:
6259 previous = code;
6260 *code++ = ((options & PCRE_CASELESS) != 0)? OP_CHARI : OP_CHAR;
6261 for (c = 0; c < mclength; c++) *code++ = mcbuffer[c];
6262
6263 /* Remember if \r or \n were seen */
6264
6265 if (mcbuffer[0] == CHAR_CR || mcbuffer[0] == CHAR_NL)
6266 cd->external_flags |= PCRE_HASCRORLF;
6267
6268 /* Set the first and required bytes appropriately. If no previous first
6269 byte, set it from this character, but revert to none on a zero repeat.
6270 Otherwise, leave the firstbyte value alone, and don't change it on a zero
6271 repeat. */
6272
6273 if (firstbyte == REQ_UNSET)
6274 {
6275 zerofirstbyte = REQ_NONE;
6276 zeroreqbyte = reqbyte;
6277
6278 /* If the character is more than one byte long, we can set firstbyte
6279 only if it is not to be matched caselessly. */
6280
6281 if (mclength == 1 || req_caseopt == 0)
6282 {
6283 firstbyte = mcbuffer[0] | req_caseopt;
6284 if (mclength != 1) reqbyte = code[-1] | cd->req_varyopt;
6285 }
6286 else firstbyte = reqbyte = REQ_NONE;
6287 }
6288
6289 /* firstbyte was previously set; we can set reqbyte only the length is
6290 1 or the matching is caseful. */
6291
6292 else
6293 {
6294 zerofirstbyte = firstbyte;
6295 zeroreqbyte = reqbyte;
6296 if (mclength == 1 || req_caseopt == 0)
6297 reqbyte = code[-1] | req_caseopt | cd->req_varyopt;
6298 }
6299
6300 break; /* End of literal character handling */
6301 }
6302 } /* end of big loop */
6303
6304
6305 /* Control never reaches here by falling through, only by a goto for all the
6306 error states. Pass back the position in the pattern so that it can be displayed
6307 to the user for diagnosing the error. */
6308
6309 FAILED:
6310 *ptrptr = ptr;
6311 return FALSE;
6312 }
6313
6314
6315
6316
6317 /*************************************************
6318 * Compile sequence of alternatives *
6319 *************************************************/
6320
6321 /* On entry, ptr is pointing past the bracket character, but on return it
6322 points to the closing bracket, or vertical bar, or end of string. The code
6323 variable is pointing at the byte into which the BRA operator has been stored.
6324 This function is used during the pre-compile phase when we are trying to find
6325 out the amount of memory needed, as well as during the real compile phase. The
6326 value of lengthptr distinguishes the two phases.
6327
6328 Arguments:
6329 options option bits, including any changes for this subpattern
6330 codeptr -> the address of the current code pointer
6331 ptrptr -> the address of the current pattern pointer
6332 errorcodeptr -> pointer to error code variable
6333 lookbehind TRUE if this is a lookbehind assertion
6334 reset_bracount TRUE to reset the count for each branch
6335 skipbytes skip this many bytes at start (for brackets and OP_COND)
6336 firstbyteptr place to put the first required character, or a negative number
6337 reqbyteptr place to put the last required character, or a negative number
6338 bcptr pointer to the chain of currently open branches
6339 cd points to the data block with tables pointers etc.
6340 lengthptr NULL during the real compile phase
6341 points to length accumulator during pre-compile phase
6342
6343 Returns: TRUE on success
6344 */
6345
6346 static BOOL
6347 compile_regex(int options, uschar **codeptr, const uschar **ptrptr,
6348 int *errorcodeptr, BOOL lookbehind, BOOL reset_bracount, int skipbytes,
6349 int *firstbyteptr, int *reqbyteptr, branch_chain *bcptr, compile_data *cd,
6350 int *lengthptr)
6351 {
6352 const uschar *ptr = *ptrptr;
6353 uschar *code = *codeptr;
6354 uschar *last_branch = code;
6355 uschar *start_bracket = code;
6356 uschar *reverse_count = NULL;
6357 open_capitem capitem;
6358 int capnumber = 0;
6359 int firstbyte, reqbyte;
6360 int branchfirstbyte, branchreqbyte;
6361 int length;
6362 int orig_bracount;
6363 int max_bracount;
6364 branch_chain bc;
6365
6366 bc.outer = bcptr;
6367 bc.current_branch = code;
6368
6369 firstbyte = reqbyte = REQ_UNSET;
6370
6371 /* Accumulate the length for use in the pre-compile phase. Start with the
6372 length of the BRA and KET and any extra bytes that are required at the
6373 beginning. We accumulate in a local variable to save frequent testing of
6374 lenthptr for NULL. We cannot do this by looking at the value of code at the
6375 start and end of each alternative, because compiled items are discarded during
6376 the pre-compile phase so that the work space is not exceeded. */
6377
6378 length = 2 + 2*LINK_SIZE + skipbytes;
6379
6380 /* WARNING: If the above line is changed for any reason, you must also change
6381 the code that abstracts option settings at the start of the pattern and makes
6382 them global. It tests the value of length for (2 + 2*LINK_SIZE) in the
6383 pre-compile phase to find out whether anything has yet been compiled or not. */
6384
6385 /* If this is a capturing subpattern, add to the chain of open capturing items
6386 so that we can detect them if (*ACCEPT) is encountered. This is also used to
6387 detect groups that contain recursive back references to themselves. Note that
6388 only OP_CBRA need be tested here; changing this opcode to one of its variants,
6389 e.g. OP_SCBRAPOS, happens later, after the group has been compiled. */
6390
6391 if (*code == OP_CBRA)
6392 {
6393 capnumber = GET2(code, 1 + LINK_SIZE);
6394 capitem.number = capnumber;
6395 capitem.next = cd->open_caps;
6396 capitem.flag = FALSE;
6397 cd->open_caps = &capitem;
6398 }
6399
6400 /* Offset is set zero to mark that this bracket is still open */
6401
6402 PUT(code, 1, 0);
6403 code += 1 + LINK_SIZE + skipbytes;
6404
6405 /* Loop for each alternative branch */
6406
6407 orig_bracount = max_bracount = cd->bracount;
6408 for (;;)
6409 {
6410 /* For a (?| group, reset the capturing bracket count so that each branch
6411 uses the same numbers. */
6412
6413 if (reset_bracount) cd->bracount = orig_bracount;
6414
6415 /* Set up dummy OP_REVERSE if lookbehind assertion */
6416
6417 if (lookbehind)
6418 {
6419 *code++ = OP_REVERSE;
6420 reverse_count = code;
6421 PUTINC(code, 0, 0);
6422 length += 1 + LINK_SIZE;
6423 }
6424
6425 /* Now compile the branch; in the pre-compile phase its length gets added
6426 into the length. */
6427
6428 if (!compile_branch(&options, &code, &ptr, errorcodeptr, &branchfirstbyte,
6429 &branchreqbyte, &bc, cd, (lengthptr == NULL)? NULL : &length))
6430 {
6431 *ptrptr = ptr;
6432 return FALSE;
6433 }
6434
6435 /* Keep the highest bracket count in case (?| was used and some branch
6436 has fewer than the rest. */
6437
6438 if (cd->bracount > max_bracount) max_bracount = cd->bracount;
6439
6440 /* In the real compile phase, there is some post-processing to be done. */
6441
6442 if (lengthptr == NULL)
6443 {
6444 /* If this is the first branch, the firstbyte and reqbyte values for the
6445 branch become the values for the regex. */
6446
6447 if (*last_branch != OP_ALT)
6448 {
6449 firstbyte = branchfirstbyte;
6450 reqbyte = branchreqbyte;
6451 }
6452
6453 /* If this is not the first branch, the first char and reqbyte have to
6454 match the values from all the previous branches, except that if the
6455 previous value for reqbyte didn't have REQ_VARY set, it can still match,
6456 and we set REQ_VARY for the regex. */
6457
6458 else
6459 {
6460 /* If we previously had a firstbyte, but it doesn't match the new branch,
6461 we have to abandon the firstbyte for the regex, but if there was
6462 previously no reqbyte, it takes on the value of the old firstbyte. */
6463
6464 if (firstbyte >= 0 && firstbyte != branchfirstbyte)
6465 {
6466 if (reqbyte < 0) reqbyte = firstbyte;
6467 firstbyte = REQ_NONE;
6468 }
6469
6470 /* If we (now or from before) have no firstbyte, a firstbyte from the
6471 branch becomes a reqbyte if there isn't a branch reqbyte. */
6472
6473 if (firstbyte < 0 && branchfirstbyte >= 0 && branchreqbyte < 0)
6474 branchreqbyte = branchfirstbyte;
6475
6476 /* Now ensure that the reqbytes match */
6477
6478 if ((reqbyte & ~REQ_VARY) != (branchreqbyte & ~REQ_VARY))
6479 reqbyte = REQ_NONE;
6480 else reqbyte |= branchreqbyte; /* To "or" REQ_VARY */
6481 }
6482
6483 /* If lookbehind, check that this branch matches a fixed-length string, and
6484 put the length into the OP_REVERSE item. Temporarily mark the end of the
6485 branch with OP_END. If the branch contains OP_RECURSE, the result is -3
6486 because there may be forward references that we can't check here. Set a
6487 flag to cause another lookbehind check at the end. Why not do it all at the
6488 end? Because common, erroneous checks are picked up here and the offset of
6489 the problem can be shown. */
6490
6491 if (lookbehind)
6492 {
6493 int fixed_length;
6494 *code = OP_END;
6495 fixed_length = find_fixedlength(last_branch, (options & PCRE_UTF8) != 0,
6496 FALSE, cd);
6497 DPRINTF(("fixed length = %d\n", fixed_length));
6498 if (fixed_length == -3)
6499 {
6500 cd->check_lookbehind = TRUE;
6501 }
6502 else if (fixed_length < 0)
6503 {
6504 *errorcodeptr = (fixed_length == -2)? ERR36 : ERR25;
6505 *ptrptr = ptr;
6506 return FALSE;
6507 }
6508 else { PUT(reverse_count, 0, fixed_length); }
6509 }
6510 }
6511
6512 /* Reached end of expression, either ')' or end of pattern. In the real
6513 compile phase, go back through the alternative branches and reverse the chain
6514 of offsets, with the field in the BRA item now becoming an offset to the
6515 first alternative. If there are no alternatives, it points to the end of the
6516 group. The length in the terminating ket is always the length of the whole
6517 bracketed item. Return leaving the pointer at the terminating char. */
6518
6519 if (*ptr != CHAR_VERTICAL_LINE)
6520 {
6521 if (lengthptr == NULL)
6522 {
6523 int branch_length = (int)(code - last_branch);
6524 do
6525 {
6526 int prev_length = GET(last_branch, 1);
6527 PUT(last_branch, 1, branch_length);
6528 branch_length = prev_length;
6529 last_branch -= branch_length;
6530 }
6531 while (branch_length > 0);
6532 }
6533
6534 /* Fill in the ket */
6535
6536 *code = OP_KET;
6537 PUT(code, 1, (int)(code - start_bracket));
6538 code += 1 + LINK_SIZE;
6539
6540 /* If it was a capturing subpattern, check to see if it contained any
6541 recursive back references. If so, we must wrap it in atomic brackets.
6542 In any event, remove the block from the chain. */
6543
6544 if (capnumber > 0)
6545 {
6546 if (cd->open_caps->flag)
6547 {
6548 memmove(start_bracket + 1 + LINK_SIZE, start_bracket,
6549 code - start_bracket);
6550 *start_bracket = OP_ONCE;
6551 code += 1 + LINK_SIZE;
6552 PUT(start_bracket, 1, (int)(code - start_bracket));
6553 *code = OP_KET;
6554 PUT(code, 1, (int)(code - start_bracket));
6555 code += 1 + LINK_SIZE;
6556 length += 2 + 2*LINK_SIZE;
6557 }
6558 cd->open_caps = cd->open_caps->next;
6559 }
6560
6561 /* Retain the highest bracket number, in case resetting was used. */
6562
6563 cd->bracount = max_bracount;
6564
6565 /* Set values to pass back */
6566
6567 *codeptr = code;
6568 *ptrptr = ptr;
6569 *firstbyteptr = firstbyte;
6570 *reqbyteptr = reqbyte;
6571 if (lengthptr != NULL)
6572 {
6573 if (OFLOW_MAX - *lengthptr < length)
6574 {
6575 *errorcodeptr = ERR20;
6576 return FALSE;
6577 }
6578 *lengthptr += length;
6579 }
6580 return TRUE;
6581 }
6582
6583 /* Another branch follows. In the pre-compile phase, we can move the code
6584 pointer back to where it was for the start of the first branch. (That is,
6585 pretend that each branch is the only one.)
6586
6587 In the real compile phase, insert an ALT node. Its length field points back
6588 to the previous branch while the bracket remains open. At the end the chain
6589 is reversed. It's done like this so that the start of the bracket has a
6590 zero offset until it is closed, making it possible to detect recursion. */
6591
6592 if (lengthptr != NULL)
6593 {
6594 code = *codeptr + 1 + LINK_SIZE + skipbytes;
6595 length += 1 + LINK_SIZE;
6596 }
6597 else
6598 {
6599 *code = OP_ALT;
6600 PUT(code, 1, (int)(code - last_branch));
6601 bc.current_branch = last_branch = code;
6602 code += 1 + LINK_SIZE;
6603 }
6604
6605 ptr++;
6606 }
6607 /* Control never reaches here */
6608 }
6609
6610
6611
6612
6613 /*************************************************
6614 * Check for anchored expression *
6615 *************************************************/
6616
6617 /* Try to find out if this is an anchored regular expression. Consider each
6618 alternative branch. If they all start with OP_SOD or OP_CIRC, or with a bracket
6619 all of whose alternatives start