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

Contents of /code/trunk/pcre_compile.c

Parent Directory Parent Directory | Revision Log Revision Log


Revision 574 - (show annotations) (download)
Sat Nov 20 17:47:27 2010 UTC (3 years, 8 months ago) by ph10
File MIME type: text/plain
File size: 234633 byte(s)
Give error if \c is followed by a byte > 127 (in ASCII/UTF-8 modes).

1 /*************************************************
2 * Perl-Compatible Regular Expressions *
3 *************************************************/
4
5 /* PCRE is a library of functions to support regular expressions whose syntax
6 and semantics are as close as possible to those of the Perl 5 language.
7
8 Written by Philip Hazel
9 Copyright (c) 1997-2010 University of Cambridge
10
11 -----------------------------------------------------------------------------
12 Redistribution and use in source and binary forms, with or without
13 modification, are permitted provided that the following conditions are met:
14
15 * Redistributions of source code must retain the above copyright notice,
16 this list of conditions and the following disclaimer.
17
18 * Redistributions in binary form must reproduce the above copyright
19 notice, this list of conditions and the following disclaimer in the
20 documentation and/or other materials provided with the distribution.
21
22 * Neither the name of the University of Cambridge nor the names of its
23 contributors may be used to endorse or promote products derived from
24 this software without specific prior written permission.
25
26 THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
27 AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28 IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29 ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE
30 LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
31 CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
32 SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
33 INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
34 CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
35 ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
36 POSSIBILITY OF SUCH DAMAGE.
37 -----------------------------------------------------------------------------
38 */
39
40
41 /* This module contains the external function pcre_compile(), along with
42 supporting internal functions that are not used by other modules. */
43
44
45 #ifdef HAVE_CONFIG_H
46 #include "config.h"
47 #endif
48
49 #define NLBLOCK cd /* Block containing newline information */
50 #define PSSTART start_pattern /* Field containing processed string start */
51 #define PSEND end_pattern /* Field containing processed string end */
52
53 #include "pcre_internal.h"
54
55
56 /* When PCRE_DEBUG is defined, we need the pcre_printint() function, which is
57 also used by pcretest. PCRE_DEBUG is not defined when building a production
58 library. */
59
60 #ifdef PCRE_DEBUG
61 #include "pcre_printint.src"
62 #endif
63
64
65 /* Macro for setting individual bits in class bitmaps. */
66
67 #define SETBIT(a,b) a[b/8] |= (1 << (b%8))
68
69 /* Maximum length value to check against when making sure that the integer that
70 holds the compiled pattern length does not overflow. We make it a bit less than
71 INT_MAX to allow for adding in group terminating bytes, so that we don't have
72 to check them every time. */
73
74 #define OFLOW_MAX (INT_MAX - 20)
75
76
77 /*************************************************
78 * Code parameters and static tables *
79 *************************************************/
80
81 /* This value specifies the size of stack workspace that is used during the
82 first pre-compile phase that determines how much memory is required. The regex
83 is partly compiled into this space, but the compiled parts are discarded as
84 soon as they can be, so that hopefully there will never be an overrun. The code
85 does, however, check for an overrun. The largest amount I've seen used is 218,
86 so this number is very generous.
87
88 The same workspace is used during the second, actual compile phase for
89 remembering forward references to groups so that they can be filled in at the
90 end. Each entry in this list occupies LINK_SIZE bytes, so even when LINK_SIZE
91 is 4 there is plenty of room. */
92
93 #define COMPILE_WORK_SIZE (4096)
94
95 /* The overrun tests check for a slightly smaller size so that they detect the
96 overrun before it actually does run off the end of the data block. */
97
98 #define WORK_SIZE_CHECK (COMPILE_WORK_SIZE - 100)
99
100
101 /* Table for handling escaped characters in the range '0'-'z'. Positive returns
102 are simple data values; negative values are for special things like \d and so
103 on. Zero means further processing is needed (for things like \x), or the escape
104 is invalid. */
105
106 #ifndef EBCDIC
107
108 /* This is the "normal" table for ASCII systems or for EBCDIC systems running
109 in UTF-8 mode. */
110
111 static const short int escapes[] = {
112 0, 0,
113 0, 0,
114 0, 0,
115 0, 0,
116 0, 0,
117 CHAR_COLON, CHAR_SEMICOLON,
118 CHAR_LESS_THAN_SIGN, CHAR_EQUALS_SIGN,
119 CHAR_GREATER_THAN_SIGN, CHAR_QUESTION_MARK,
120 CHAR_COMMERCIAL_AT, -ESC_A,
121 -ESC_B, -ESC_C,
122 -ESC_D, -ESC_E,
123 0, -ESC_G,
124 -ESC_H, 0,
125 0, -ESC_K,
126 0, 0,
127 -ESC_N, 0,
128 -ESC_P, -ESC_Q,
129 -ESC_R, -ESC_S,
130 0, 0,
131 -ESC_V, -ESC_W,
132 -ESC_X, 0,
133 -ESC_Z, CHAR_LEFT_SQUARE_BRACKET,
134 CHAR_BACKSLASH, CHAR_RIGHT_SQUARE_BRACKET,
135 CHAR_CIRCUMFLEX_ACCENT, CHAR_UNDERSCORE,
136 CHAR_GRAVE_ACCENT, 7,
137 -ESC_b, 0,
138 -ESC_d, ESC_e,
139 ESC_f, 0,
140 -ESC_h, 0,
141 0, -ESC_k,
142 0, 0,
143 ESC_n, 0,
144 -ESC_p, 0,
145 ESC_r, -ESC_s,
146 ESC_tee, 0,
147 -ESC_v, -ESC_w,
148 0, 0,
149 -ESC_z
150 };
151
152 #else
153
154 /* This is the "abnormal" table for EBCDIC systems without UTF-8 support. */
155
156 static const short int escapes[] = {
157 /* 48 */ 0, 0, 0, '.', '<', '(', '+', '|',
158 /* 50 */ '&', 0, 0, 0, 0, 0, 0, 0,
159 /* 58 */ 0, 0, '!', '$', '*', ')', ';', '~',
160 /* 60 */ '-', '/', 0, 0, 0, 0, 0, 0,
161 /* 68 */ 0, 0, '|', ',', '%', '_', '>', '?',
162 /* 70 */ 0, 0, 0, 0, 0, 0, 0, 0,
163 /* 78 */ 0, '`', ':', '#', '@', '\'', '=', '"',
164 /* 80 */ 0, 7, -ESC_b, 0, -ESC_d, ESC_e, ESC_f, 0,
165 /* 88 */-ESC_h, 0, 0, '{', 0, 0, 0, 0,
166 /* 90 */ 0, 0, -ESC_k, 'l', 0, ESC_n, 0, -ESC_p,
167 /* 98 */ 0, ESC_r, 0, '}', 0, 0, 0, 0,
168 /* A0 */ 0, '~', -ESC_s, ESC_tee, 0,-ESC_v, -ESC_w, 0,
169 /* A8 */ 0,-ESC_z, 0, 0, 0, '[', 0, 0,
170 /* B0 */ 0, 0, 0, 0, 0, 0, 0, 0,
171 /* B8 */ 0, 0, 0, 0, 0, ']', '=', '-',
172 /* C0 */ '{',-ESC_A, -ESC_B, -ESC_C, -ESC_D,-ESC_E, 0, -ESC_G,
173 /* C8 */-ESC_H, 0, 0, 0, 0, 0, 0, 0,
174 /* D0 */ '}', 0, -ESC_K, 0, 0,-ESC_N, 0, -ESC_P,
175 /* D8 */-ESC_Q,-ESC_R, 0, 0, 0, 0, 0, 0,
176 /* E0 */ '\\', 0, -ESC_S, 0, 0,-ESC_V, -ESC_W, -ESC_X,
177 /* E8 */ 0,-ESC_Z, 0, 0, 0, 0, 0, 0,
178 /* F0 */ 0, 0, 0, 0, 0, 0, 0, 0,
179 /* F8 */ 0, 0, 0, 0, 0, 0, 0, 0
180 };
181 #endif
182
183
184 /* Table of special "verbs" like (*PRUNE). This is a short table, so it is
185 searched linearly. Put all the names into a single string, in order to reduce
186 the number of relocations when a shared library is dynamically linked. The
187 string is built from string macros so that it works in UTF-8 mode on EBCDIC
188 platforms. */
189
190 typedef struct verbitem {
191 int len; /* Length of verb name */
192 int op; /* Op when no arg, or -1 if arg mandatory */
193 int op_arg; /* Op when arg present, or -1 if not allowed */
194 } verbitem;
195
196 static const char verbnames[] =
197 "\0" /* Empty name is a shorthand for MARK */
198 STRING_MARK0
199 STRING_ACCEPT0
200 STRING_COMMIT0
201 STRING_F0
202 STRING_FAIL0
203 STRING_PRUNE0
204 STRING_SKIP0
205 STRING_THEN;
206
207 static const verbitem verbs[] = {
208 { 0, -1, OP_MARK },
209 { 4, -1, OP_MARK },
210 { 6, OP_ACCEPT, -1 },
211 { 6, OP_COMMIT, -1 },
212 { 1, OP_FAIL, -1 },
213 { 4, OP_FAIL, -1 },
214 { 5, OP_PRUNE, OP_PRUNE_ARG },
215 { 4, OP_SKIP, OP_SKIP_ARG },
216 { 4, OP_THEN, OP_THEN_ARG }
217 };
218
219 static const int verbcount = sizeof(verbs)/sizeof(verbitem);
220
221
222 /* Tables of names of POSIX character classes and their lengths. The names are
223 now all in a single string, to reduce the number of relocations when a shared
224 library is dynamically loaded. The list of lengths is terminated by a zero
225 length entry. The first three must be alpha, lower, upper, as this is assumed
226 for handling case independence. */
227
228 static const char posix_names[] =
229 STRING_alpha0 STRING_lower0 STRING_upper0 STRING_alnum0
230 STRING_ascii0 STRING_blank0 STRING_cntrl0 STRING_digit0
231 STRING_graph0 STRING_print0 STRING_punct0 STRING_space0
232 STRING_word0 STRING_xdigit;
233
234 static const uschar posix_name_lengths[] = {
235 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 4, 6, 0 };
236
237 /* Table of class bit maps for each POSIX class. Each class is formed from a
238 base map, with an optional addition or removal of another map. Then, for some
239 classes, there is some additional tweaking: for [:blank:] the vertical space
240 characters are removed, and for [:alpha:] and [:alnum:] the underscore
241 character is removed. The triples in the table consist of the base map offset,
242 second map offset or -1 if no second map, and a non-negative value for map
243 addition or a negative value for map subtraction (if there are two maps). The
244 absolute value of the third field has these meanings: 0 => no tweaking, 1 =>
245 remove vertical space characters, 2 => remove underscore. */
246
247 static const int posix_class_maps[] = {
248 cbit_word, cbit_digit, -2, /* alpha */
249 cbit_lower, -1, 0, /* lower */
250 cbit_upper, -1, 0, /* upper */
251 cbit_word, -1, 2, /* alnum - word without underscore */
252 cbit_print, cbit_cntrl, 0, /* ascii */
253 cbit_space, -1, 1, /* blank - a GNU extension */
254 cbit_cntrl, -1, 0, /* cntrl */
255 cbit_digit, -1, 0, /* digit */
256 cbit_graph, -1, 0, /* graph */
257 cbit_print, -1, 0, /* print */
258 cbit_punct, -1, 0, /* punct */
259 cbit_space, -1, 0, /* space */
260 cbit_word, -1, 0, /* word - a Perl extension */
261 cbit_xdigit,-1, 0 /* xdigit */
262 };
263
264 /* Table of substitutes for \d etc when PCRE_UCP is set. The POSIX class
265 substitutes must be in the order of the names, defined above, and there are
266 both positive and negative cases. NULL means no substitute. */
267
268 #ifdef SUPPORT_UCP
269 static const uschar *substitutes[] = {
270 (uschar *)"\\P{Nd}", /* \D */
271 (uschar *)"\\p{Nd}", /* \d */
272 (uschar *)"\\P{Xsp}", /* \S */ /* NOTE: Xsp is Perl space */
273 (uschar *)"\\p{Xsp}", /* \s */
274 (uschar *)"\\P{Xwd}", /* \W */
275 (uschar *)"\\p{Xwd}" /* \w */
276 };
277
278 static const uschar *posix_substitutes[] = {
279 (uschar *)"\\p{L}", /* alpha */
280 (uschar *)"\\p{Ll}", /* lower */
281 (uschar *)"\\p{Lu}", /* upper */
282 (uschar *)"\\p{Xan}", /* alnum */
283 NULL, /* ascii */
284 (uschar *)"\\h", /* blank */
285 NULL, /* cntrl */
286 (uschar *)"\\p{Nd}", /* digit */
287 NULL, /* graph */
288 NULL, /* print */
289 NULL, /* punct */
290 (uschar *)"\\p{Xps}", /* space */ /* NOTE: Xps is POSIX space */
291 (uschar *)"\\p{Xwd}", /* word */
292 NULL, /* xdigit */
293 /* Negated cases */
294 (uschar *)"\\P{L}", /* ^alpha */
295 (uschar *)"\\P{Ll}", /* ^lower */
296 (uschar *)"\\P{Lu}", /* ^upper */
297 (uschar *)"\\P{Xan}", /* ^alnum */
298 NULL, /* ^ascii */
299 (uschar *)"\\H", /* ^blank */
300 NULL, /* ^cntrl */
301 (uschar *)"\\P{Nd}", /* ^digit */
302 NULL, /* ^graph */
303 NULL, /* ^print */
304 NULL, /* ^punct */
305 (uschar *)"\\P{Xps}", /* ^space */ /* NOTE: Xps is POSIX space */
306 (uschar *)"\\P{Xwd}", /* ^word */
307 NULL /* ^xdigit */
308 };
309 #define POSIX_SUBSIZE (sizeof(posix_substitutes)/sizeof(uschar *))
310 #endif
311
312 #define STRING(a) # a
313 #define XSTRING(s) STRING(s)
314
315 /* The texts of compile-time error messages. These are "char *" because they
316 are passed to the outside world. Do not ever re-use any error number, because
317 they are documented. Always add a new error instead. Messages marked DEAD below
318 are no longer used. This used to be a table of strings, but in order to reduce
319 the number of relocations needed when a shared library is loaded dynamically,
320 it is now one long string. We cannot use a table of offsets, because the
321 lengths of inserts such as XSTRING(MAX_NAME_SIZE) are not known. Instead, we
322 simply count through to the one we want - this isn't a performance issue
323 because these strings are used only when there is a compilation error.
324
325 Each substring ends with \0 to insert a null character. This includes the final
326 substring, so that the whole string ends with \0\0, which can be detected when
327 counting through. */
328
329 static const char error_texts[] =
330 "no error\0"
331 "\\ at end of pattern\0"
332 "\\c at end of pattern\0"
333 "unrecognized character follows \\\0"
334 "numbers out of order in {} quantifier\0"
335 /* 5 */
336 "number too big in {} quantifier\0"
337 "missing terminating ] for character class\0"
338 "invalid escape sequence in character class\0"
339 "range out of order in character class\0"
340 "nothing to repeat\0"
341 /* 10 */
342 "operand of unlimited repeat could match the empty string\0" /** DEAD **/
343 "internal error: unexpected repeat\0"
344 "unrecognized character after (? or (?-\0"
345 "POSIX named classes are supported only within a class\0"
346 "missing )\0"
347 /* 15 */
348 "reference to non-existent subpattern\0"
349 "erroffset passed as NULL\0"
350 "unknown option bit(s) set\0"
351 "missing ) after comment\0"
352 "parentheses nested too deeply\0" /** DEAD **/
353 /* 20 */
354 "regular expression is too large\0"
355 "failed to get memory\0"
356 "unmatched parentheses\0"
357 "internal error: code overflow\0"
358 "unrecognized character after (?<\0"
359 /* 25 */
360 "lookbehind assertion is not fixed length\0"
361 "malformed number or name after (?(\0"
362 "conditional group contains more than two branches\0"
363 "assertion expected after (?(\0"
364 "(?R or (?[+-]digits must be followed by )\0"
365 /* 30 */
366 "unknown POSIX class name\0"
367 "POSIX collating elements are not supported\0"
368 "this version of PCRE is not compiled with PCRE_UTF8 support\0"
369 "spare error\0" /** DEAD **/
370 "character value in \\x{...} sequence is too large\0"
371 /* 35 */
372 "invalid condition (?(0)\0"
373 "\\C not allowed in lookbehind assertion\0"
374 "PCRE does not support \\L, \\l, \\N{name}, \\U, or \\u\0"
375 "number after (?C is > 255\0"
376 "closing ) for (?C expected\0"
377 /* 40 */
378 "recursive call could loop indefinitely\0"
379 "unrecognized character after (?P\0"
380 "syntax error in subpattern name (missing terminator)\0"
381 "two named subpatterns have the same name\0"
382 "invalid UTF-8 string\0"
383 /* 45 */
384 "support for \\P, \\p, and \\X has not been compiled\0"
385 "malformed \\P or \\p sequence\0"
386 "unknown property name after \\P or \\p\0"
387 "subpattern name is too long (maximum " XSTRING(MAX_NAME_SIZE) " characters)\0"
388 "too many named subpatterns (maximum " XSTRING(MAX_NAME_COUNT) ")\0"
389 /* 50 */
390 "repeated subpattern is too long\0" /** DEAD **/
391 "octal value is greater than \\377 (not in UTF-8 mode)\0"
392 "internal error: overran compiling workspace\0"
393 "internal error: previously-checked referenced subpattern not found\0"
394 "DEFINE group contains more than one branch\0"
395 /* 55 */
396 "repeating a DEFINE group is not allowed\0"
397 "inconsistent NEWLINE options\0"
398 "\\g is not followed by a braced, angle-bracketed, or quoted name/number or by a plain number\0"
399 "a numbered reference must not be zero\0"
400 "an argument is not allowed for (*ACCEPT), (*FAIL), or (*COMMIT)\0"
401 /* 60 */
402 "(*VERB) not recognized\0"
403 "number is too big\0"
404 "subpattern name expected\0"
405 "digit expected after (?+\0"
406 "] is an invalid data character in JavaScript compatibility mode\0"
407 /* 65 */
408 "different names for subpatterns of the same number are not allowed\0"
409 "(*MARK) must have an argument\0"
410 "this version of PCRE is not compiled with PCRE_UCP support\0"
411 "\\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, int, uschar **, const uschar **, int *, BOOL, BOOL, int,
549 int *, 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. We know that if (?P< is
1109 encountered, the name will be terminated by '>' because that is checked in the
1110 first pass. Recursion is used to keep track of subpatterns that reset the
1111 capturing group numbers - the (?| feature.
1112
1113 Arguments:
1114 ptrptr address of the current character pointer (updated)
1115 cd compile background data
1116 name name to seek, or NULL if seeking a numbered subpattern
1117 lorn name length, or subpattern number if name is NULL
1118 xmode TRUE if we are in /x mode
1119 utf8 TRUE if we are in UTF-8 mode
1120 count pointer to the current capturing subpattern number (updated)
1121
1122 Returns: the number of the named subpattern, or -1 if not found
1123 */
1124
1125 static int
1126 find_parens_sub(uschar **ptrptr, compile_data *cd, const uschar *name, int lorn,
1127 BOOL xmode, BOOL utf8, int *count)
1128 {
1129 uschar *ptr = *ptrptr;
1130 int start_count = *count;
1131 int hwm_count = start_count;
1132 BOOL dup_parens = FALSE;
1133
1134 /* If the first character is a parenthesis, check on the type of group we are
1135 dealing with. The very first call may not start with a parenthesis. */
1136
1137 if (ptr[0] == CHAR_LEFT_PARENTHESIS)
1138 {
1139 /* Handle specials such as (*SKIP) or (*UTF8) etc. */
1140
1141 if (ptr[1] == CHAR_ASTERISK) ptr += 2;
1142
1143 /* Handle a normal, unnamed capturing parenthesis. */
1144
1145 else if (ptr[1] != CHAR_QUESTION_MARK)
1146 {
1147 *count += 1;
1148 if (name == NULL && *count == lorn) return *count;
1149 ptr++;
1150 }
1151
1152 /* All cases now have (? at the start. Remember when we are in a group
1153 where the parenthesis numbers are duplicated. */
1154
1155 else if (ptr[2] == CHAR_VERTICAL_LINE)
1156 {
1157 ptr += 3;
1158 dup_parens = TRUE;
1159 }
1160
1161 /* Handle comments; all characters are allowed until a ket is reached. */
1162
1163 else if (ptr[2] == CHAR_NUMBER_SIGN)
1164 {
1165 for (ptr += 3; *ptr != 0; ptr++) if (*ptr == CHAR_RIGHT_PARENTHESIS) break;
1166 goto FAIL_EXIT;
1167 }
1168
1169 /* Handle a condition. If it is an assertion, just carry on so that it
1170 is processed as normal. If not, skip to the closing parenthesis of the
1171 condition (there can't be any nested parens). */
1172
1173 else if (ptr[2] == CHAR_LEFT_PARENTHESIS)
1174 {
1175 ptr += 2;
1176 if (ptr[1] != CHAR_QUESTION_MARK)
1177 {
1178 while (*ptr != 0 && *ptr != CHAR_RIGHT_PARENTHESIS) ptr++;
1179 if (*ptr != 0) ptr++;
1180 }
1181 }
1182
1183 /* Start with (? but not a condition. */
1184
1185 else
1186 {
1187 ptr += 2;
1188 if (*ptr == CHAR_P) ptr++; /* Allow optional P */
1189
1190 /* We have to disambiguate (?<! and (?<= from (?<name> for named groups */
1191
1192 if ((*ptr == CHAR_LESS_THAN_SIGN && ptr[1] != CHAR_EXCLAMATION_MARK &&
1193 ptr[1] != CHAR_EQUALS_SIGN) || *ptr == CHAR_APOSTROPHE)
1194 {
1195 int term;
1196 const uschar *thisname;
1197 *count += 1;
1198 if (name == NULL && *count == lorn) return *count;
1199 term = *ptr++;
1200 if (term == CHAR_LESS_THAN_SIGN) term = CHAR_GREATER_THAN_SIGN;
1201 thisname = ptr;
1202 while (*ptr != term) ptr++;
1203 if (name != NULL && lorn == ptr - thisname &&
1204 strncmp((const char *)name, (const char *)thisname, lorn) == 0)
1205 return *count;
1206 term++;
1207 }
1208 }
1209 }
1210
1211 /* Past any initial parenthesis handling, scan for parentheses or vertical
1212 bars. */
1213
1214 for (; *ptr != 0; ptr++)
1215 {
1216 /* Skip over backslashed characters and also entire \Q...\E */
1217
1218 if (*ptr == CHAR_BACKSLASH)
1219 {
1220 if (*(++ptr) == 0) goto FAIL_EXIT;
1221 if (*ptr == CHAR_Q) for (;;)
1222 {
1223 while (*(++ptr) != 0 && *ptr != CHAR_BACKSLASH) {};
1224 if (*ptr == 0) goto FAIL_EXIT;
1225 if (*(++ptr) == CHAR_E) break;
1226 }
1227 continue;
1228 }
1229
1230 /* Skip over character classes; this logic must be similar to the way they
1231 are handled for real. If the first character is '^', skip it. Also, if the
1232 first few characters (either before or after ^) are \Q\E or \E we skip them
1233 too. This makes for compatibility with Perl. Note the use of STR macros to
1234 encode "Q\\E" so that it works in UTF-8 on EBCDIC platforms. */
1235
1236 if (*ptr == CHAR_LEFT_SQUARE_BRACKET)
1237 {
1238 BOOL negate_class = FALSE;
1239 for (;;)
1240 {
1241 if (ptr[1] == CHAR_BACKSLASH)
1242 {
1243 if (ptr[2] == CHAR_E)
1244 ptr+= 2;
1245 else if (strncmp((const char *)ptr+2,
1246 STR_Q STR_BACKSLASH STR_E, 3) == 0)
1247 ptr += 4;
1248 else
1249 break;
1250 }
1251 else if (!negate_class && ptr[1] == CHAR_CIRCUMFLEX_ACCENT)
1252 {
1253 negate_class = TRUE;
1254 ptr++;
1255 }
1256 else break;
1257 }
1258
1259 /* If the next character is ']', it is a data character that must be
1260 skipped, except in JavaScript compatibility mode. */
1261
1262 if (ptr[1] == CHAR_RIGHT_SQUARE_BRACKET &&
1263 (cd->external_options & PCRE_JAVASCRIPT_COMPAT) == 0)
1264 ptr++;
1265
1266 while (*(++ptr) != CHAR_RIGHT_SQUARE_BRACKET)
1267 {
1268 if (*ptr == 0) return -1;
1269 if (*ptr == CHAR_BACKSLASH)
1270 {
1271 if (*(++ptr) == 0) goto FAIL_EXIT;
1272 if (*ptr == CHAR_Q) for (;;)
1273 {
1274 while (*(++ptr) != 0 && *ptr != CHAR_BACKSLASH) {};
1275 if (*ptr == 0) goto FAIL_EXIT;
1276 if (*(++ptr) == CHAR_E) break;
1277 }
1278 continue;
1279 }
1280 }
1281 continue;
1282 }
1283
1284 /* Skip comments in /x mode */
1285
1286 if (xmode && *ptr == CHAR_NUMBER_SIGN)
1287 {
1288 ptr++;
1289 while (*ptr != 0)
1290 {
1291 if (IS_NEWLINE(ptr)) { ptr += cd->nllen - 1; break; }
1292 ptr++;
1293 #ifdef SUPPORT_UTF8
1294 if (utf8) while ((*ptr & 0xc0) == 0x80) ptr++;
1295 #endif
1296 }
1297 if (*ptr == 0) goto FAIL_EXIT;
1298 continue;
1299 }
1300
1301 /* Check for the special metacharacters */
1302
1303 if (*ptr == CHAR_LEFT_PARENTHESIS)
1304 {
1305 int rc = find_parens_sub(&ptr, cd, name, lorn, xmode, utf8, count);
1306 if (rc > 0) return rc;
1307 if (*ptr == 0) goto FAIL_EXIT;
1308 }
1309
1310 else if (*ptr == CHAR_RIGHT_PARENTHESIS)
1311 {
1312 if (dup_parens && *count < hwm_count) *count = hwm_count;
1313 goto FAIL_EXIT;
1314 }
1315
1316 else if (*ptr == CHAR_VERTICAL_LINE && dup_parens)
1317 {
1318 if (*count > hwm_count) hwm_count = *count;
1319 *count = start_count;
1320 }
1321 }
1322
1323 FAIL_EXIT:
1324 *ptrptr = ptr;
1325 return -1;
1326 }
1327
1328
1329
1330
1331 /*************************************************
1332 * Find forward referenced subpattern *
1333 *************************************************/
1334
1335 /* This function scans along a pattern's text looking for capturing
1336 subpatterns, and counting them. If it finds a named pattern that matches the
1337 name it is given, it returns its number. Alternatively, if the name is NULL, it
1338 returns when it reaches a given numbered subpattern. This is used for forward
1339 references to subpatterns. We used to be able to start this scan from the
1340 current compiling point, using the current count value from cd->bracount, and
1341 do it all in a single loop, but the addition of the possibility of duplicate
1342 subpattern numbers means that we have to scan from the very start, in order to
1343 take account of such duplicates, and to use a recursive function to keep track
1344 of the different types of group.
1345
1346 Arguments:
1347 cd compile background data
1348 name name to seek, or NULL if seeking a numbered subpattern
1349 lorn name length, or subpattern number if name is NULL
1350 xmode TRUE if we are in /x mode
1351 utf8 TRUE if we are in UTF-8 mode
1352
1353 Returns: the number of the found subpattern, or -1 if not found
1354 */
1355
1356 static int
1357 find_parens(compile_data *cd, const uschar *name, int lorn, BOOL xmode,
1358 BOOL utf8)
1359 {
1360 uschar *ptr = (uschar *)cd->start_pattern;
1361 int count = 0;
1362 int rc;
1363
1364 /* If the pattern does not start with an opening parenthesis, the first call
1365 to find_parens_sub() will scan right to the end (if necessary). However, if it
1366 does start with a parenthesis, find_parens_sub() will return when it hits the
1367 matching closing parens. That is why we have to have a loop. */
1368
1369 for (;;)
1370 {
1371 rc = find_parens_sub(&ptr, cd, name, lorn, xmode, utf8, &count);
1372 if (rc > 0 || *ptr++ == 0) break;
1373 }
1374
1375 return rc;
1376 }
1377
1378
1379
1380
1381 /*************************************************
1382 * Find first significant op code *
1383 *************************************************/
1384
1385 /* This is called by several functions that scan a compiled expression looking
1386 for a fixed first character, or an anchoring op code etc. It skips over things
1387 that do not influence this. For some calls, a change of option is important.
1388 For some calls, it makes sense to skip negative forward and all backward
1389 assertions, and also the \b assertion; for others it does not.
1390
1391 Arguments:
1392 code pointer to the start of the group
1393 options pointer to external options
1394 optbit the option bit whose changing is significant, or
1395 zero if none are
1396 skipassert TRUE if certain assertions are to be skipped
1397
1398 Returns: pointer to the first significant opcode
1399 */
1400
1401 static const uschar*
1402 first_significant_code(const uschar *code, int *options, int optbit,
1403 BOOL skipassert)
1404 {
1405 for (;;)
1406 {
1407 switch ((int)*code)
1408 {
1409 case OP_OPT:
1410 if (optbit > 0 && ((int)code[1] & optbit) != (*options & optbit))
1411 *options = (int)code[1];
1412 code += 2;
1413 break;
1414
1415 case OP_ASSERT_NOT:
1416 case OP_ASSERTBACK:
1417 case OP_ASSERTBACK_NOT:
1418 if (!skipassert) return code;
1419 do code += GET(code, 1); while (*code == OP_ALT);
1420 code += _pcre_OP_lengths[*code];
1421 break;
1422
1423 case OP_WORD_BOUNDARY:
1424 case OP_NOT_WORD_BOUNDARY:
1425 if (!skipassert) return code;
1426 /* Fall through */
1427
1428 case OP_CALLOUT:
1429 case OP_CREF:
1430 case OP_NCREF:
1431 case OP_RREF:
1432 case OP_NRREF:
1433 case OP_DEF:
1434 code += _pcre_OP_lengths[*code];
1435 break;
1436
1437 default:
1438 return code;
1439 }
1440 }
1441 /* Control never reaches here */
1442 }
1443
1444
1445
1446
1447 /*************************************************
1448 * Find the fixed length of a branch *
1449 *************************************************/
1450
1451 /* Scan a branch and compute the fixed length of subject that will match it,
1452 if the length is fixed. This is needed for dealing with backward assertions.
1453 In UTF8 mode, the result is in characters rather than bytes. The branch is
1454 temporarily terminated with OP_END when this function is called.
1455
1456 This function is called when a backward assertion is encountered, so that if it
1457 fails, the error message can point to the correct place in the pattern.
1458 However, we cannot do this when the assertion contains subroutine calls,
1459 because they can be forward references. We solve this by remembering this case
1460 and doing the check at the end; a flag specifies which mode we are running in.
1461
1462 Arguments:
1463 code points to the start of the pattern (the bracket)
1464 options the compiling options
1465 atend TRUE if called when the pattern is complete
1466 cd the "compile data" structure
1467
1468 Returns: the fixed length,
1469 or -1 if there is no fixed length,
1470 or -2 if \C was encountered
1471 or -3 if an OP_RECURSE item was encountered and atend is FALSE
1472 */
1473
1474 static int
1475 find_fixedlength(uschar *code, int options, BOOL atend, compile_data *cd)
1476 {
1477 int length = -1;
1478
1479 register int branchlength = 0;
1480 register uschar *cc = code + 1 + LINK_SIZE;
1481
1482 /* Scan along the opcodes for this branch. If we get to the end of the
1483 branch, check the length against that of the other branches. */
1484
1485 for (;;)
1486 {
1487 int d;
1488 uschar *ce, *cs;
1489 register int op = *cc;
1490 switch (op)
1491 {
1492 case OP_CBRA:
1493 case OP_BRA:
1494 case OP_ONCE:
1495 case OP_COND:
1496 d = find_fixedlength(cc + ((op == OP_CBRA)? 2:0), options, atend, cd);
1497 if (d < 0) return d;
1498 branchlength += d;
1499 do cc += GET(cc, 1); while (*cc == OP_ALT);
1500 cc += 1 + LINK_SIZE;
1501 break;
1502
1503 /* Reached end of a branch; if it's a ket it is the end of a nested
1504 call. If it's ALT it is an alternation in a nested call. If it is
1505 END it's the end of the outer call. All can be handled by the same code. */
1506
1507 case OP_ALT:
1508 case OP_KET:
1509 case OP_KETRMAX:
1510 case OP_KETRMIN:
1511 case OP_END:
1512 if (length < 0) length = branchlength;
1513 else if (length != branchlength) return -1;
1514 if (*cc != OP_ALT) return length;
1515 cc += 1 + LINK_SIZE;
1516 branchlength = 0;
1517 break;
1518
1519 /* A true recursion implies not fixed length, but a subroutine call may
1520 be OK. If the subroutine is a forward reference, we can't deal with
1521 it until the end of the pattern, so return -3. */
1522
1523 case OP_RECURSE:
1524 if (!atend) return -3;
1525 cs = ce = (uschar *)cd->start_code + GET(cc, 1); /* Start subpattern */
1526 do ce += GET(ce, 1); while (*ce == OP_ALT); /* End subpattern */
1527 if (cc > cs && cc < ce) return -1; /* Recursion */
1528 d = find_fixedlength(cs + 2, options, atend, cd);
1529 if (d < 0) return d;
1530 branchlength += d;
1531 cc += 1 + LINK_SIZE;
1532 break;
1533
1534 /* Skip over assertive subpatterns */
1535
1536 case OP_ASSERT:
1537 case OP_ASSERT_NOT:
1538 case OP_ASSERTBACK:
1539 case OP_ASSERTBACK_NOT:
1540 do cc += GET(cc, 1); while (*cc == OP_ALT);
1541 /* Fall through */
1542
1543 /* Skip over things that don't match chars */
1544
1545 case OP_REVERSE:
1546 case OP_CREF:
1547 case OP_NCREF:
1548 case OP_RREF:
1549 case OP_NRREF:
1550 case OP_DEF:
1551 case OP_OPT:
1552 case OP_CALLOUT:
1553 case OP_SOD:
1554 case OP_SOM:
1555 case OP_SET_SOM:
1556 case OP_EOD:
1557 case OP_EODN:
1558 case OP_CIRC:
1559 case OP_DOLL:
1560 case OP_NOT_WORD_BOUNDARY:
1561 case OP_WORD_BOUNDARY:
1562 cc += _pcre_OP_lengths[*cc];
1563 break;
1564
1565 /* Handle literal characters */
1566
1567 case OP_CHAR:
1568 case OP_CHARNC:
1569 case OP_NOT:
1570 branchlength++;
1571 cc += 2;
1572 #ifdef SUPPORT_UTF8
1573 if ((options & PCRE_UTF8) != 0 && cc[-1] >= 0xc0)
1574 cc += _pcre_utf8_table4[cc[-1] & 0x3f];
1575 #endif
1576 break;
1577
1578 /* Handle exact repetitions. The count is already in characters, but we
1579 need to skip over a multibyte character in UTF8 mode. */
1580
1581 case OP_EXACT:
1582 branchlength += GET2(cc,1);
1583 cc += 4;
1584 #ifdef SUPPORT_UTF8
1585 if ((options & PCRE_UTF8) != 0 && cc[-1] >= 0xc0)
1586 cc += _pcre_utf8_table4[cc[-1] & 0x3f];
1587 #endif
1588 break;
1589
1590 case OP_TYPEEXACT:
1591 branchlength += GET2(cc,1);
1592 if (cc[3] == OP_PROP || cc[3] == OP_NOTPROP) cc += 2;
1593 cc += 4;
1594 break;
1595
1596 /* Handle single-char matchers */
1597
1598 case OP_PROP:
1599 case OP_NOTPROP:
1600 cc += 2;
1601 /* Fall through */
1602
1603 case OP_NOT_DIGIT:
1604 case OP_DIGIT:
1605 case OP_NOT_WHITESPACE:
1606 case OP_WHITESPACE:
1607 case OP_NOT_WORDCHAR:
1608 case OP_WORDCHAR:
1609 case OP_ANY:
1610 case OP_ALLANY:
1611 branchlength++;
1612 cc++;
1613 break;
1614
1615 /* The single-byte matcher isn't allowed */
1616
1617 case OP_ANYBYTE:
1618 return -2;
1619
1620 /* Check a class for variable quantification */
1621
1622 #ifdef SUPPORT_UTF8
1623 case OP_XCLASS:
1624 cc += GET(cc, 1) - 33;
1625 /* Fall through */
1626 #endif
1627
1628 case OP_CLASS:
1629 case OP_NCLASS:
1630 cc += 33;
1631
1632 switch (*cc)
1633 {
1634 case OP_CRSTAR:
1635 case OP_CRMINSTAR:
1636 case OP_CRQUERY:
1637 case OP_CRMINQUERY:
1638 return -1;
1639
1640 case OP_CRRANGE:
1641 case OP_CRMINRANGE:
1642 if (GET2(cc,1) != GET2(cc,3)) return -1;
1643 branchlength += GET2(cc,1);
1644 cc += 5;
1645 break;
1646
1647 default:
1648 branchlength++;
1649 }
1650 break;
1651
1652 /* Anything else is variable length */
1653
1654 default:
1655 return -1;
1656 }
1657 }
1658 /* Control never gets here */
1659 }
1660
1661
1662
1663
1664 /*************************************************
1665 * Scan compiled regex for specific bracket *
1666 *************************************************/
1667
1668 /* This little function scans through a compiled pattern until it finds a
1669 capturing bracket with the given number, or, if the number is negative, an
1670 instance of OP_REVERSE for a lookbehind. The function is global in the C sense
1671 so that it can be called from pcre_study() when finding the minimum matching
1672 length.
1673
1674 Arguments:
1675 code points to start of expression
1676 utf8 TRUE in UTF-8 mode
1677 number the required bracket number or negative to find a lookbehind
1678
1679 Returns: pointer to the opcode for the bracket, or NULL if not found
1680 */
1681
1682 const uschar *
1683 _pcre_find_bracket(const uschar *code, BOOL utf8, int number)
1684 {
1685 for (;;)
1686 {
1687 register int c = *code;
1688 if (c == OP_END) return NULL;
1689
1690 /* XCLASS is used for classes that cannot be represented just by a bit
1691 map. This includes negated single high-valued characters. The length in
1692 the table is zero; the actual length is stored in the compiled code. */
1693
1694 if (c == OP_XCLASS) code += GET(code, 1);
1695
1696 /* Handle recursion */
1697
1698 else if (c == OP_REVERSE)
1699 {
1700 if (number < 0) return (uschar *)code;
1701 code += _pcre_OP_lengths[c];
1702 }
1703
1704 /* Handle capturing bracket */
1705
1706 else if (c == OP_CBRA)
1707 {
1708 int n = GET2(code, 1+LINK_SIZE);
1709 if (n == number) return (uschar *)code;
1710 code += _pcre_OP_lengths[c];
1711 }
1712
1713 /* Otherwise, we can get the item's length from the table, except that for
1714 repeated character types, we have to test for \p and \P, which have an extra
1715 two bytes of parameters, and for MARK/PRUNE/SKIP/THEN with an argument, we
1716 must add in its length. */
1717
1718 else
1719 {
1720 switch(c)
1721 {
1722 case OP_TYPESTAR:
1723 case OP_TYPEMINSTAR:
1724 case OP_TYPEPLUS:
1725 case OP_TYPEMINPLUS:
1726 case OP_TYPEQUERY:
1727 case OP_TYPEMINQUERY:
1728 case OP_TYPEPOSSTAR:
1729 case OP_TYPEPOSPLUS:
1730 case OP_TYPEPOSQUERY:
1731 if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
1732 break;
1733
1734 case OP_TYPEUPTO:
1735 case OP_TYPEMINUPTO:
1736 case OP_TYPEEXACT:
1737 case OP_TYPEPOSUPTO:
1738 if (code[3] == OP_PROP || code[3] == OP_NOTPROP) code += 2;
1739 break;
1740
1741 case OP_MARK:
1742 case OP_PRUNE_ARG:
1743 case OP_SKIP_ARG:
1744 code += code[1];
1745 break;
1746
1747 case OP_THEN_ARG:
1748 code += code[1+LINK_SIZE];
1749 break;
1750 }
1751
1752 /* Add in the fixed length from the table */
1753
1754 code += _pcre_OP_lengths[c];
1755
1756 /* In UTF-8 mode, opcodes that are followed by a character may be followed by
1757 a multi-byte character. The length in the table is a minimum, so we have to
1758 arrange to skip the extra bytes. */
1759
1760 #ifdef SUPPORT_UTF8
1761 if (utf8) switch(c)
1762 {
1763 case OP_CHAR:
1764 case OP_CHARNC:
1765 case OP_EXACT:
1766 case OP_UPTO:
1767 case OP_MINUPTO:
1768 case OP_POSUPTO:
1769 case OP_STAR:
1770 case OP_MINSTAR:
1771 case OP_POSSTAR:
1772 case OP_PLUS:
1773 case OP_MINPLUS:
1774 case OP_POSPLUS:
1775 case OP_QUERY:
1776 case OP_MINQUERY:
1777 case OP_POSQUERY:
1778 if (code[-1] >= 0xc0) code += _pcre_utf8_table4[code[-1] & 0x3f];
1779 break;
1780 }
1781 #else
1782 (void)(utf8); /* Keep compiler happy by referencing function argument */
1783 #endif
1784 }
1785 }
1786 }
1787
1788
1789
1790 /*************************************************
1791 * Scan compiled regex for recursion reference *
1792 *************************************************/
1793
1794 /* This little function scans through a compiled pattern until it finds an
1795 instance of OP_RECURSE.
1796
1797 Arguments:
1798 code points to start of expression
1799 utf8 TRUE in UTF-8 mode
1800
1801 Returns: pointer to the opcode for OP_RECURSE, or NULL if not found
1802 */
1803
1804 static const uschar *
1805 find_recurse(const uschar *code, BOOL utf8)
1806 {
1807 for (;;)
1808 {
1809 register int c = *code;
1810 if (c == OP_END) return NULL;
1811 if (c == OP_RECURSE) return code;
1812
1813 /* XCLASS is used for classes that cannot be represented just by a bit
1814 map. This includes negated single high-valued characters. The length in
1815 the table is zero; the actual length is stored in the compiled code. */
1816
1817 if (c == OP_XCLASS) code += GET(code, 1);
1818
1819 /* Otherwise, we can get the item's length from the table, except that for
1820 repeated character types, we have to test for \p and \P, which have an extra
1821 two bytes of parameters, and for MARK/PRUNE/SKIP/THEN with an argument, we
1822 must add in its length. */
1823
1824 else
1825 {
1826 switch(c)
1827 {
1828 case OP_TYPESTAR:
1829 case OP_TYPEMINSTAR:
1830 case OP_TYPEPLUS:
1831 case OP_TYPEMINPLUS:
1832 case OP_TYPEQUERY:
1833 case OP_TYPEMINQUERY:
1834 case OP_TYPEPOSSTAR:
1835 case OP_TYPEPOSPLUS:
1836 case OP_TYPEPOSQUERY:
1837 if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
1838 break;
1839
1840 case OP_TYPEPOSUPTO:
1841 case OP_TYPEUPTO:
1842 case OP_TYPEMINUPTO:
1843 case OP_TYPEEXACT:
1844 if (code[3] == OP_PROP || code[3] == OP_NOTPROP) code += 2;
1845 break;
1846
1847 case OP_MARK:
1848 case OP_PRUNE_ARG:
1849 case OP_SKIP_ARG:
1850 code += code[1];
1851 break;
1852
1853 case OP_THEN_ARG:
1854 code += code[1+LINK_SIZE];
1855 break;
1856 }
1857
1858 /* Add in the fixed length from the table */
1859
1860 code += _pcre_OP_lengths[c];
1861
1862 /* In UTF-8 mode, opcodes that are followed by a character may be followed
1863 by a multi-byte character. The length in the table is a minimum, so we have
1864 to arrange to skip the extra bytes. */
1865
1866 #ifdef SUPPORT_UTF8
1867 if (utf8) switch(c)
1868 {
1869 case OP_CHAR:
1870 case OP_CHARNC:
1871 case OP_EXACT:
1872 case OP_UPTO:
1873 case OP_MINUPTO:
1874 case OP_POSUPTO:
1875 case OP_STAR:
1876 case OP_MINSTAR:
1877 case OP_POSSTAR:
1878 case OP_PLUS:
1879 case OP_MINPLUS:
1880 case OP_POSPLUS:
1881 case OP_QUERY:
1882 case OP_MINQUERY:
1883 case OP_POSQUERY:
1884 if (code[-1] >= 0xc0) code += _pcre_utf8_table4[code[-1] & 0x3f];
1885 break;
1886 }
1887 #else
1888 (void)(utf8); /* Keep compiler happy by referencing function argument */
1889 #endif
1890 }
1891 }
1892 }
1893
1894
1895
1896 /*************************************************
1897 * Scan compiled branch for non-emptiness *
1898 *************************************************/
1899
1900 /* This function scans through a branch of a compiled pattern to see whether it
1901 can match the empty string or not. It is called from could_be_empty()
1902 below and from compile_branch() when checking for an unlimited repeat of a
1903 group that can match nothing. Note that first_significant_code() skips over
1904 backward and negative forward assertions when its final argument is TRUE. If we
1905 hit an unclosed bracket, we return "empty" - this means we've struck an inner
1906 bracket whose current branch will already have been scanned.
1907
1908 Arguments:
1909 code points to start of search
1910 endcode points to where to stop
1911 utf8 TRUE if in UTF8 mode
1912 cd contains pointers to tables etc.
1913
1914 Returns: TRUE if what is matched could be empty
1915 */
1916
1917 static BOOL
1918 could_be_empty_branch(const uschar *code, const uschar *endcode, BOOL utf8,
1919 compile_data *cd)
1920 {
1921 register int c;
1922 for (code = first_significant_code(code + _pcre_OP_lengths[*code], NULL, 0, TRUE);
1923 code < endcode;
1924 code = first_significant_code(code + _pcre_OP_lengths[c], NULL, 0, TRUE))
1925 {
1926 const uschar *ccode;
1927
1928 c = *code;
1929
1930 /* Skip over forward assertions; the other assertions are skipped by
1931 first_significant_code() with a TRUE final argument. */
1932
1933 if (c == OP_ASSERT)
1934 {
1935 do code += GET(code, 1); while (*code == OP_ALT);
1936 c = *code;
1937 continue;
1938 }
1939
1940 /* Groups with zero repeats can of course be empty; skip them. */
1941
1942 if (c == OP_BRAZERO || c == OP_BRAMINZERO || c == OP_SKIPZERO)
1943 {
1944 code += _pcre_OP_lengths[c];
1945 do code += GET(code, 1); while (*code == OP_ALT);
1946 c = *code;
1947 continue;
1948 }
1949
1950 /* For a recursion/subroutine call, if its end has been reached, which
1951 implies a subroutine call, we can scan it. */
1952
1953 if (c == OP_RECURSE)
1954 {
1955 BOOL empty_branch = FALSE;
1956 const uschar *scode = cd->start_code + GET(code, 1);
1957 if (GET(scode, 1) == 0) return TRUE; /* Unclosed */
1958 do
1959 {
1960 if (could_be_empty_branch(scode, endcode, utf8, cd))
1961 {
1962 empty_branch = TRUE;
1963 break;
1964 }
1965 scode += GET(scode, 1);
1966 }
1967 while (*scode == OP_ALT);
1968 if (!empty_branch) return FALSE; /* All branches are non-empty */
1969 continue;
1970 }
1971
1972 /* For other groups, scan the branches. */
1973
1974 if (c == OP_BRA || c == OP_CBRA || c == OP_ONCE || c == OP_COND)
1975 {
1976 BOOL empty_branch;
1977 if (GET(code, 1) == 0) return TRUE; /* Hit unclosed bracket */
1978
1979 /* If a conditional group has only one branch, there is a second, implied,
1980 empty branch, so just skip over the conditional, because it could be empty.
1981 Otherwise, scan the individual branches of the group. */
1982
1983 if (c == OP_COND && code[GET(code, 1)] != OP_ALT)
1984 code += GET(code, 1);
1985 else
1986 {
1987 empty_branch = FALSE;
1988 do
1989 {
1990 if (!empty_branch && could_be_empty_branch(code, endcode, utf8, cd))
1991 empty_branch = TRUE;
1992 code += GET(code, 1);
1993 }
1994 while (*code == OP_ALT);
1995 if (!empty_branch) return FALSE; /* All branches are non-empty */
1996 }
1997
1998 c = *code;
1999 continue;
2000 }
2001
2002 /* Handle the other opcodes */
2003
2004 switch (c)
2005 {
2006 /* Check for quantifiers after a class. XCLASS is used for classes that
2007 cannot be represented just by a bit map. This includes negated single
2008 high-valued characters. The length in _pcre_OP_lengths[] is zero; the
2009 actual length is stored in the compiled code, so we must update "code"
2010 here. */
2011
2012 #ifdef SUPPORT_UTF8
2013 case OP_XCLASS:
2014 ccode = code += GET(code, 1);
2015 goto CHECK_CLASS_REPEAT;
2016 #endif
2017
2018 case OP_CLASS:
2019 case OP_NCLASS:
2020 ccode = code + 33;
2021
2022 #ifdef SUPPORT_UTF8
2023 CHECK_CLASS_REPEAT:
2024 #endif
2025
2026 switch (*ccode)
2027 {
2028 case OP_CRSTAR: /* These could be empty; continue */
2029 case OP_CRMINSTAR:
2030 case OP_CRQUERY:
2031 case OP_CRMINQUERY:
2032 break;
2033
2034 default: /* Non-repeat => class must match */
2035 case OP_CRPLUS: /* These repeats aren't empty */
2036 case OP_CRMINPLUS:
2037 return FALSE;
2038
2039 case OP_CRRANGE:
2040 case OP_CRMINRANGE:
2041 if (GET2(ccode, 1) > 0) return FALSE; /* Minimum > 0 */
2042 break;
2043 }
2044 break;
2045
2046 /* Opcodes that must match a character */
2047
2048 case OP_PROP:
2049 case OP_NOTPROP:
2050 case OP_EXTUNI:
2051 case OP_NOT_DIGIT:
2052 case OP_DIGIT:
2053 case OP_NOT_WHITESPACE:
2054 case OP_WHITESPACE:
2055 case OP_NOT_WORDCHAR:
2056 case OP_WORDCHAR:
2057 case OP_ANY:
2058 case OP_ALLANY:
2059 case OP_ANYBYTE:
2060 case OP_CHAR:
2061 case OP_CHARNC:
2062 case OP_NOT:
2063 case OP_PLUS:
2064 case OP_MINPLUS:
2065 case OP_POSPLUS:
2066 case OP_EXACT:
2067 case OP_NOTPLUS:
2068 case OP_NOTMINPLUS:
2069 case OP_NOTPOSPLUS:
2070 case OP_NOTEXACT:
2071 case OP_TYPEPLUS:
2072 case OP_TYPEMINPLUS:
2073 case OP_TYPEPOSPLUS:
2074 case OP_TYPEEXACT:
2075 return FALSE;
2076
2077 /* These are going to continue, as they may be empty, but we have to
2078 fudge the length for the \p and \P cases. */
2079
2080 case OP_TYPESTAR:
2081 case OP_TYPEMINSTAR:
2082 case OP_TYPEPOSSTAR:
2083 case OP_TYPEQUERY:
2084 case OP_TYPEMINQUERY:
2085 case OP_TYPEPOSQUERY:
2086 if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
2087 break;
2088
2089 /* Same for these */
2090
2091 case OP_TYPEUPTO:
2092 case OP_TYPEMINUPTO:
2093 case OP_TYPEPOSUPTO:
2094 if (code[3] == OP_PROP || code[3] == OP_NOTPROP) code += 2;
2095 break;
2096
2097 /* End of branch */
2098
2099 case OP_KET:
2100 case OP_KETRMAX:
2101 case OP_KETRMIN:
2102 case OP_ALT:
2103 return TRUE;
2104
2105 /* In UTF-8 mode, STAR, MINSTAR, POSSTAR, QUERY, MINQUERY, POSQUERY, UPTO,
2106 MINUPTO, and POSUPTO may be followed by a multibyte character */
2107
2108 #ifdef SUPPORT_UTF8
2109 case OP_STAR:
2110 case OP_MINSTAR:
2111 case OP_POSSTAR:
2112 case OP_QUERY:
2113 case OP_MINQUERY:
2114 case OP_POSQUERY:
2115 if (utf8 && code[1] >= 0xc0) code += _pcre_utf8_table4[code[1] & 0x3f];
2116 break;
2117
2118 case OP_UPTO:
2119 case OP_MINUPTO:
2120 case OP_POSUPTO:
2121 if (utf8 && code[3] >= 0xc0) code += _pcre_utf8_table4[code[3] & 0x3f];
2122 break;
2123 #endif
2124
2125 /* MARK, and PRUNE/SKIP/THEN with an argument must skip over the argument
2126 string. */
2127
2128 case OP_MARK:
2129 case OP_PRUNE_ARG:
2130 case OP_SKIP_ARG:
2131 code += code[1];
2132 break;
2133
2134 case OP_THEN_ARG:
2135 code += code[1+LINK_SIZE];
2136 break;
2137
2138 /* None of the remaining opcodes are required to match a character. */
2139
2140 default:
2141 break;
2142 }
2143 }
2144
2145 return TRUE;
2146 }
2147
2148
2149
2150 /*************************************************
2151 * Scan compiled regex for non-emptiness *
2152 *************************************************/
2153
2154 /* This function is called to check for left recursive calls. We want to check
2155 the current branch of the current pattern to see if it could match the empty
2156 string. If it could, we must look outwards for branches at other levels,
2157 stopping when we pass beyond the bracket which is the subject of the recursion.
2158
2159 Arguments:
2160 code points to start of the recursion
2161 endcode points to where to stop (current RECURSE item)
2162 bcptr points to the chain of current (unclosed) branch starts
2163 utf8 TRUE if in UTF-8 mode
2164 cd pointers to tables etc
2165
2166 Returns: TRUE if what is matched could be empty
2167 */
2168
2169 static BOOL
2170 could_be_empty(const uschar *code, const uschar *endcode, branch_chain *bcptr,
2171 BOOL utf8, compile_data *cd)
2172 {
2173 while (bcptr != NULL && bcptr->current_branch >= code)
2174 {
2175 if (!could_be_empty_branch(bcptr->current_branch, endcode, utf8, cd))
2176 return FALSE;
2177 bcptr = bcptr->outer;
2178 }
2179 return TRUE;
2180 }
2181
2182
2183
2184 /*************************************************
2185 * Check for POSIX class syntax *
2186 *************************************************/
2187
2188 /* This function is called when the sequence "[:" or "[." or "[=" is
2189 encountered in a character class. It checks whether this is followed by a
2190 sequence of characters terminated by a matching ":]" or ".]" or "=]". If we
2191 reach an unescaped ']' without the special preceding character, return FALSE.
2192
2193 Originally, this function only recognized a sequence of letters between the
2194 terminators, but it seems that Perl recognizes any sequence of characters,
2195 though of course unknown POSIX names are subsequently rejected. Perl gives an
2196 "Unknown POSIX class" error for [:f\oo:] for example, where previously PCRE
2197 didn't consider this to be a POSIX class. Likewise for [:1234:].
2198
2199 The problem in trying to be exactly like Perl is in the handling of escapes. We
2200 have to be sure that [abc[:x\]pqr] is *not* treated as containing a POSIX
2201 class, but [abc[:x\]pqr:]] is (so that an error can be generated). The code
2202 below handles the special case of \], but does not try to do any other escape
2203 processing. This makes it different from Perl for cases such as [:l\ower:]
2204 where Perl recognizes it as the POSIX class "lower" but PCRE does not recognize
2205 "l\ower". This is a lesser evil that not diagnosing bad classes when Perl does,
2206 I think.
2207
2208 Arguments:
2209 ptr pointer to the initial [
2210 endptr where to return the end pointer
2211
2212 Returns: TRUE or FALSE
2213 */
2214
2215 static BOOL
2216 check_posix_syntax(const uschar *ptr, const uschar **endptr)
2217 {
2218 int terminator; /* Don't combine these lines; the Solaris cc */
2219 terminator = *(++ptr); /* compiler warns about "non-constant" initializer. */
2220 for (++ptr; *ptr != 0; ptr++)
2221 {
2222 if (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET) ptr++; else
2223 {
2224 if (*ptr == CHAR_RIGHT_SQUARE_BRACKET) return FALSE;
2225 if (*ptr == terminator && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET)
2226 {
2227 *endptr = ptr;
2228 return TRUE;
2229 }
2230 }
2231 }
2232 return FALSE;
2233 }
2234
2235
2236
2237
2238 /*************************************************
2239 * Check POSIX class name *
2240 *************************************************/
2241
2242 /* This function is called to check the name given in a POSIX-style class entry
2243 such as [:alnum:].
2244
2245 Arguments:
2246 ptr points to the first letter
2247 len the length of the name
2248
2249 Returns: a value representing the name, or -1 if unknown
2250 */
2251
2252 static int
2253 check_posix_name(const uschar *ptr, int len)
2254 {
2255 const char *pn = posix_names;
2256 register int yield = 0;
2257 while (posix_name_lengths[yield] != 0)
2258 {
2259 if (len == posix_name_lengths[yield] &&
2260 strncmp((const char *)ptr, pn, len) == 0) return yield;
2261 pn += posix_name_lengths[yield] + 1;
2262 yield++;
2263 }
2264 return -1;
2265 }
2266
2267
2268 /*************************************************
2269 * Adjust OP_RECURSE items in repeated group *
2270 *************************************************/
2271
2272 /* OP_RECURSE items contain an offset from the start of the regex to the group
2273 that is referenced. This means that groups can be replicated for fixed
2274 repetition simply by copying (because the recursion is allowed to refer to
2275 earlier groups that are outside the current group). However, when a group is
2276 optional (i.e. the minimum quantifier is zero), OP_BRAZERO or OP_SKIPZERO is
2277 inserted before it, after it has been compiled. This means that any OP_RECURSE
2278 items within it that refer to the group itself or any contained groups have to
2279 have their offsets adjusted. That one of the jobs of this function. Before it
2280 is called, the partially compiled regex must be temporarily terminated with
2281 OP_END.
2282
2283 This function has been extended with the possibility of forward references for
2284 recursions and subroutine calls. It must also check the list of such references
2285 for the group we are dealing with. If it finds that one of the recursions in
2286 the current group is on this list, it adjusts the offset in the list, not the
2287 value in the reference (which is a group number).
2288
2289 Arguments:
2290 group points to the start of the group
2291 adjust the amount by which the group is to be moved
2292 utf8 TRUE in UTF-8 mode
2293 cd contains pointers to tables etc.
2294 save_hwm the hwm forward reference pointer at the start of the group
2295
2296 Returns: nothing
2297 */
2298
2299 static void
2300 adjust_recurse(uschar *group, int adjust, BOOL utf8, compile_data *cd,
2301 uschar *save_hwm)
2302 {
2303 uschar *ptr = group;
2304
2305 while ((ptr = (uschar *)find_recurse(ptr, utf8)) != NULL)
2306 {
2307 int offset;
2308 uschar *hc;
2309
2310 /* See if this recursion is on the forward reference list. If so, adjust the
2311 reference. */
2312
2313 for (hc = save_hwm; hc < cd->hwm; hc += LINK_SIZE)
2314 {
2315 offset = GET(hc, 0);
2316 if (cd->start_code + offset == ptr + 1)
2317 {
2318 PUT(hc, 0, offset + adjust);
2319 break;
2320 }
2321 }
2322
2323 /* Otherwise, adjust the recursion offset if it's after the start of this
2324 group. */
2325
2326 if (hc >= cd->hwm)
2327 {
2328 offset = GET(ptr, 1);
2329 if (cd->start_code + offset >= group) PUT(ptr, 1, offset + adjust);
2330 }
2331
2332 ptr += 1 + LINK_SIZE;
2333 }
2334 }
2335
2336
2337
2338 /*************************************************
2339 * Insert an automatic callout point *
2340 *************************************************/
2341
2342 /* This function is called when the PCRE_AUTO_CALLOUT option is set, to insert
2343 callout points before each pattern item.
2344
2345 Arguments:
2346 code current code pointer
2347 ptr current pattern pointer
2348 cd pointers to tables etc
2349
2350 Returns: new code pointer
2351 */
2352
2353 static uschar *
2354 auto_callout(uschar *code, const uschar *ptr, compile_data *cd)
2355 {
2356 *code++ = OP_CALLOUT;
2357 *code++ = 255;
2358 PUT(code, 0, (int)(ptr - cd->start_pattern)); /* Pattern offset */
2359 PUT(code, LINK_SIZE, 0); /* Default length */
2360 return code + 2*LINK_SIZE;
2361 }
2362
2363
2364
2365 /*************************************************
2366 * Complete a callout item *
2367 *************************************************/
2368
2369 /* A callout item contains the length of the next item in the pattern, which
2370 we can't fill in till after we have reached the relevant point. This is used
2371 for both automatic and manual callouts.
2372
2373 Arguments:
2374 previous_callout points to previous callout item
2375 ptr current pattern pointer
2376 cd pointers to tables etc
2377
2378 Returns: nothing
2379 */
2380
2381 static void
2382 complete_callout(uschar *previous_callout, const uschar *ptr, compile_data *cd)
2383 {
2384 int length = (int)(ptr - cd->start_pattern - GET(previous_callout, 2));
2385 PUT(previous_callout, 2 + LINK_SIZE, length);
2386 }
2387
2388
2389
2390 #ifdef SUPPORT_UCP
2391 /*************************************************
2392 * Get othercase range *
2393 *************************************************/
2394
2395 /* This function is passed the start and end of a class range, in UTF-8 mode
2396 with UCP support. It searches up the characters, looking for internal ranges of
2397 characters in the "other" case. Each call returns the next one, updating the
2398 start address.
2399
2400 Arguments:
2401 cptr points to starting character value; updated
2402 d end value
2403 ocptr where to put start of othercase range
2404 odptr where to put end of othercase range
2405
2406 Yield: TRUE when range returned; FALSE when no more
2407 */
2408
2409 static BOOL
2410 get_othercase_range(unsigned int *cptr, unsigned int d, unsigned int *ocptr,
2411 unsigned int *odptr)
2412 {
2413 unsigned int c, othercase, next;
2414
2415 for (c = *cptr; c <= d; c++)
2416 { if ((othercase = UCD_OTHERCASE(c)) != c) break; }
2417
2418 if (c > d) return FALSE;
2419
2420 *ocptr = othercase;
2421 next = othercase + 1;
2422
2423 for (++c; c <= d; c++)
2424 {
2425 if (UCD_OTHERCASE(c) != next) break;
2426 next++;
2427 }
2428
2429 *odptr = next - 1;
2430 *cptr = c;
2431
2432 return TRUE;
2433 }
2434
2435
2436
2437 /*************************************************
2438 * Check a character and a property *
2439 *************************************************/
2440
2441 /* This function is called by check_auto_possessive() when a property item
2442 is adjacent to a fixed character.
2443
2444 Arguments:
2445 c the character
2446 ptype the property type
2447 pdata the data for the type
2448 negated TRUE if it's a negated property (\P or \p{^)
2449
2450 Returns: TRUE if auto-possessifying is OK
2451 */
2452
2453 static BOOL
2454 check_char_prop(int c, int ptype, int pdata, BOOL negated)
2455 {
2456 const ucd_record *prop = GET_UCD(c);
2457 switch(ptype)
2458 {
2459 case PT_LAMP:
2460 return (prop->chartype == ucp_Lu ||
2461 prop->chartype == ucp_Ll ||
2462 prop->chartype == ucp_Lt) == negated;
2463
2464 case PT_GC:
2465 return (pdata == _pcre_ucp_gentype[prop->chartype]) == negated;
2466
2467 case PT_PC:
2468 return (pdata == prop->chartype) == negated;
2469
2470 case PT_SC:
2471 return (pdata == prop->script) == negated;
2472
2473 /* These are specials */
2474
2475 case PT_ALNUM:
2476 return (_pcre_ucp_gentype[prop->chartype] == ucp_L ||
2477 _pcre_ucp_gentype[prop->chartype] == ucp_N) == negated;
2478
2479 case PT_SPACE: /* Perl space */
2480 return (_pcre_ucp_gentype[prop->chartype] == ucp_Z ||
2481 c == CHAR_HT || c == CHAR_NL || c == CHAR_FF || c == CHAR_CR)
2482 == negated;
2483
2484 case PT_PXSPACE: /* POSIX space */
2485 return (_pcre_ucp_gentype[prop->chartype] == ucp_Z ||
2486 c == CHAR_HT || c == CHAR_NL || c == CHAR_VT ||
2487 c == CHAR_FF || c == CHAR_CR)
2488 == negated;
2489
2490 case PT_WORD:
2491 return (_pcre_ucp_gentype[prop->chartype] == ucp_L ||
2492 _pcre_ucp_gentype[prop->chartype] == ucp_N ||
2493 c == CHAR_UNDERSCORE) == negated;
2494 }
2495 return FALSE;
2496 }
2497 #endif /* SUPPORT_UCP */
2498
2499
2500
2501 /*************************************************
2502 * Check if auto-possessifying is possible *
2503 *************************************************/
2504
2505 /* This function is called for unlimited repeats of certain items, to see
2506 whether the next thing could possibly match the repeated item. If not, it makes
2507 sense to automatically possessify the repeated item.
2508
2509 Arguments:
2510 previous pointer to the repeated opcode
2511 utf8 TRUE in UTF-8 mode
2512 ptr next character in pattern
2513 options options bits
2514 cd contains pointers to tables etc.
2515
2516 Returns: TRUE if possessifying is wanted
2517 */
2518
2519 static BOOL
2520 check_auto_possessive(const uschar *previous, BOOL utf8, const uschar *ptr,
2521 int options, compile_data *cd)
2522 {
2523 int c, next;
2524 int op_code = *previous++;
2525
2526 /* Skip whitespace and comments in extended mode */
2527
2528 if ((options & PCRE_EXTENDED) != 0)
2529 {
2530 for (;;)
2531 {
2532 while ((cd->ctypes[*ptr] & ctype_space) != 0) ptr++;
2533 if (*ptr == CHAR_NUMBER_SIGN)
2534 {
2535 ptr++;
2536 while (*ptr != 0)
2537 {
2538 if (IS_NEWLINE(ptr)) { ptr += cd->nllen; break; }
2539 ptr++;
2540 #ifdef SUPPORT_UTF8
2541 if (utf8) while ((*ptr & 0xc0) == 0x80) ptr++;
2542 #endif
2543 }
2544 }
2545 else break;
2546 }
2547 }
2548
2549 /* If the next item is one that we can handle, get its value. A non-negative
2550 value is a character, a negative value is an escape value. */
2551
2552 if (*ptr == CHAR_BACKSLASH)
2553 {
2554 int temperrorcode = 0;
2555 next = check_escape(&ptr, &temperrorcode, cd->bracount, options, FALSE);
2556 if (temperrorcode != 0) return FALSE;
2557 ptr++; /* Point after the escape sequence */
2558 }
2559
2560 else if ((cd->ctypes[*ptr] & ctype_meta) == 0)
2561 {
2562 #ifdef SUPPORT_UTF8
2563 if (utf8) { GETCHARINC(next, ptr); } else
2564 #endif
2565 next = *ptr++;
2566 }
2567
2568 else return FALSE;
2569
2570 /* Skip whitespace and comments in extended mode */
2571
2572 if ((options & PCRE_EXTENDED) != 0)
2573 {
2574 for (;;)
2575 {
2576 while ((cd->ctypes[*ptr] & ctype_space) != 0) ptr++;
2577 if (*ptr == CHAR_NUMBER_SIGN)
2578 {
2579 ptr++;
2580 while (*ptr != 0)
2581 {
2582 if (IS_NEWLINE(ptr)) { ptr += cd->nllen; break; }
2583 ptr++;
2584 #ifdef SUPPORT_UTF8
2585 if (utf8) while ((*ptr & 0xc0) == 0x80) ptr++;
2586 #endif
2587 }
2588 }
2589 else break;
2590 }
2591 }
2592
2593 /* If the next thing is itself optional, we have to give up. */
2594
2595 if (*ptr == CHAR_ASTERISK || *ptr == CHAR_QUESTION_MARK ||
2596 strncmp((char *)ptr, STR_LEFT_CURLY_BRACKET STR_0 STR_COMMA, 3) == 0)
2597 return FALSE;
2598
2599 /* Now compare the next item with the previous opcode. First, handle cases when
2600 the next item is a character. */
2601
2602 if (next >= 0) switch(op_code)
2603 {
2604 case OP_CHAR:
2605 #ifdef SUPPORT_UTF8
2606 GETCHARTEST(c, previous);
2607 #else
2608 c = *previous;
2609 #endif
2610 return c != next;
2611
2612 /* For CHARNC (caseless character) we must check the other case. If we have
2613 Unicode property support, we can use it to test the other case of
2614 high-valued characters. */
2615
2616 case OP_CHARNC:
2617 #ifdef SUPPORT_UTF8
2618 GETCHARTEST(c, previous);
2619 #else
2620 c = *previous;
2621 #endif
2622 if (c == next) return FALSE;
2623 #ifdef SUPPORT_UTF8
2624 if (utf8)
2625 {
2626 unsigned int othercase;
2627 if (next < 128) othercase = cd->fcc[next]; else
2628 #ifdef SUPPORT_UCP
2629 othercase = UCD_OTHERCASE((unsigned int)next);
2630 #else
2631 othercase = NOTACHAR;
2632 #endif
2633 return (unsigned int)c != othercase;
2634 }
2635 else
2636 #endif /* SUPPORT_UTF8 */
2637 return (c != cd->fcc[next]); /* Non-UTF-8 mode */
2638
2639 /* For OP_NOT, its data is always a single-byte character. */
2640
2641 case OP_NOT:
2642 if ((c = *previous) == next) return TRUE;
2643 if ((options & PCRE_CASELESS) == 0) return FALSE;
2644 #ifdef SUPPORT_UTF8
2645 if (utf8)
2646 {
2647 unsigned int othercase;
2648 if (next < 128) othercase = cd->fcc[next]; else
2649 #ifdef SUPPORT_UCP
2650 othercase = UCD_OTHERCASE(next);
2651 #else
2652 othercase = NOTACHAR;
2653 #endif
2654 return (unsigned int)c == othercase;
2655 }
2656 else
2657 #endif /* SUPPORT_UTF8 */
2658 return (c == cd->fcc[next]); /* Non-UTF-8 mode */
2659
2660 /* Note that OP_DIGIT etc. are generated only when PCRE_UCP is *not* set.
2661 When it is set, \d etc. are converted into OP_(NOT_)PROP codes. */
2662
2663 case OP_DIGIT:
2664 return next > 127 || (cd->ctypes[next] & ctype_digit) == 0;
2665
2666 case OP_NOT_DIGIT:
2667 return next <= 127 && (cd->ctypes[next] & ctype_digit) != 0;
2668
2669 case OP_WHITESPACE:
2670 return next > 127 || (cd->ctypes[next] & ctype_space) == 0;
2671
2672 case OP_NOT_WHITESPACE:
2673 return next <= 127 && (cd->ctypes[next] & ctype_space) != 0;
2674
2675 case OP_WORDCHAR:
2676 return next > 127 || (cd->ctypes[next] & ctype_word) == 0;
2677
2678 case OP_NOT_WORDCHAR:
2679 return next <= 127 && (cd->ctypes[next] & ctype_word) != 0;
2680
2681 case OP_HSPACE:
2682 case OP_NOT_HSPACE:
2683 switch(next)
2684 {
2685 case 0x09:
2686 case 0x20:
2687 case 0xa0:
2688 case 0x1680:
2689 case 0x180e:
2690 case 0x2000:
2691 case 0x2001:
2692 case 0x2002:
2693 case 0x2003:
2694 case 0x2004:
2695 case 0x2005:
2696 case 0x2006:
2697 case 0x2007:
2698 case 0x2008:
2699 case 0x2009:
2700 case 0x200A:
2701 case 0x202f:
2702 case 0x205f:
2703 case 0x3000:
2704 return op_code == OP_NOT_HSPACE;
2705 default:
2706 return op_code != OP_NOT_HSPACE;
2707 }
2708
2709 case OP_ANYNL:
2710 case OP_VSPACE:
2711 case OP_NOT_VSPACE:
2712 switch(next)
2713 {
2714 case 0x0a:
2715 case 0x0b:
2716 case 0x0c:
2717 case 0x0d:
2718 case 0x85:
2719 case 0x2028:
2720 case 0x2029:
2721 return op_code == OP_NOT_VSPACE;
2722 default:
2723 return op_code != OP_NOT_VSPACE;
2724 }
2725
2726 #ifdef SUPPORT_UCP
2727 case OP_PROP:
2728 return check_char_prop(next, previous[0], previous[1], FALSE);
2729
2730 case OP_NOTPROP:
2731 return check_char_prop(next, previous[0], previous[1], TRUE);
2732 #endif
2733
2734 default:
2735 return FALSE;
2736 }
2737
2738
2739 /* Handle the case when the next item is \d, \s, etc. Note that when PCRE_UCP
2740 is set, \d turns into ESC_du rather than ESC_d, etc., so ESC_d etc. are
2741 generated only when PCRE_UCP is *not* set, that is, when only ASCII
2742 characteristics are recognized. Similarly, the opcodes OP_DIGIT etc. are
2743 replaced by OP_PROP codes when PCRE_UCP is set. */
2744
2745 switch(op_code)
2746 {
2747 case OP_CHAR:
2748 case OP_CHARNC:
2749 #ifdef SUPPORT_UTF8
2750 GETCHARTEST(c, previous);
2751 #else
2752 c = *previous;
2753 #endif
2754 switch(-next)
2755 {
2756 case ESC_d:
2757 return c > 127 || (cd->ctypes[c] & ctype_digit) == 0;
2758
2759 case ESC_D:
2760 return c <= 127 && (cd->ctypes[c] & ctype_digit) != 0;
2761
2762 case ESC_s:
2763 return c > 127 || (cd->ctypes[c] & ctype_space) == 0;
2764
2765 case ESC_S:
2766 return c <= 127 && (cd->ctypes[c] & ctype_space) != 0;
2767
2768 case ESC_w:
2769 return c > 127 || (cd->ctypes[c] & ctype_word) == 0;
2770
2771 case ESC_W:
2772 return c <= 127 && (cd->ctypes[c] & ctype_word) != 0;
2773
2774 case ESC_h:
2775 case ESC_H:
2776 switch(c)
2777 {
2778 case 0x09:
2779 case 0x20:
2780 case 0xa0:
2781 case 0x1680:
2782 case 0x180e:
2783 case 0x2000:
2784 case 0x2001:
2785 case 0x2002:
2786 case 0x2003:
2787 case 0x2004:
2788 case 0x2005:
2789 case 0x2006:
2790 case 0x2007:
2791 case 0x2008:
2792 case 0x2009:
2793 case 0x200A:
2794 case 0x202f:
2795 case 0x205f:
2796 case 0x3000:
2797 return -next != ESC_h;
2798 default:
2799 return -next == ESC_h;
2800 }
2801
2802 case ESC_v:
2803 case ESC_V:
2804 switch(c)
2805 {
2806 case 0x0a:
2807 case 0x0b:
2808 case 0x0c:
2809 case 0x0d:
2810 case 0x85:
2811 case 0x2028:
2812 case 0x2029:
2813 return -next != ESC_v;
2814 default:
2815 return -next == ESC_v;
2816 }
2817
2818 /* When PCRE_UCP is set, these values get generated for \d etc. Find
2819 their substitutions and process them. The result will always be either
2820 -ESC_p or -ESC_P. Then fall through to process those values. */
2821
2822 #ifdef SUPPORT_UCP
2823 case ESC_du:
2824 case ESC_DU:
2825 case ESC_wu:
2826 case ESC_WU:
2827 case ESC_su:
2828 case ESC_SU:
2829 {
2830 int temperrorcode = 0;
2831 ptr = substitutes[-next - ESC_DU];
2832 next = check_escape(&ptr, &temperrorcode, 0, options, FALSE);
2833 if (temperrorcode != 0) return FALSE;
2834 ptr++; /* For compatibility */
2835 }
2836 /* Fall through */
2837
2838 case ESC_p:
2839 case ESC_P:
2840 {
2841 int ptype, pdata, errorcodeptr;
2842 BOOL negated;
2843
2844 ptr--; /* Make ptr point at the p or P */
2845 ptype = get_ucp(&ptr, &negated, &pdata, &errorcodeptr);
2846 if (ptype < 0) return FALSE;
2847 ptr++; /* Point past the final curly ket */
2848
2849 /* If the property item is optional, we have to give up. (When generated
2850 from \d etc by PCRE_UCP, this test will have been applied much earlier,
2851 to the original \d etc. At this point, ptr will point to a zero byte. */
2852
2853 if (*ptr == CHAR_ASTERISK || *ptr == CHAR_QUESTION_MARK ||
2854 strncmp((char *)ptr, STR_LEFT_CURLY_BRACKET STR_0 STR_COMMA, 3) == 0)
2855 return FALSE;
2856
2857 /* Do the property check. */
2858
2859 return check_char_prop(c, ptype, pdata, (next == -ESC_P) != negated);
2860 }
2861 #endif
2862
2863 default:
2864 return FALSE;
2865 }
2866
2867 /* In principle, support for Unicode properties should be integrated here as
2868 well. It means re-organizing the above code so as to get hold of the property
2869 values before switching on the op-code. However, I wonder how many patterns
2870 combine ASCII \d etc with Unicode properties? (Note that if PCRE_UCP is set,
2871 these op-codes are never generated.) */
2872
2873 case OP_DIGIT:
2874 return next == -ESC_D || next == -ESC_s || next == -ESC_W ||
2875 next == -ESC_h || next == -ESC_v || next == -ESC_R;
2876
2877 case OP_NOT_DIGIT:
2878 return next == -ESC_d;
2879
2880 case OP_WHITESPACE:
2881 return next == -ESC_S || next == -ESC_d || next == -ESC_w || next == -ESC_R;
2882
2883 case OP_NOT_WHITESPACE:
2884 return next == -ESC_s || next == -ESC_h || next == -ESC_v;
2885
2886 case OP_HSPACE:
2887 return next == -ESC_S || next == -ESC_H || next == -ESC_d ||
2888 next == -ESC_w || next == -ESC_v || next == -ESC_R;
2889
2890 case OP_NOT_HSPACE:
2891 return next == -ESC_h;
2892
2893 /* Can't have \S in here because VT matches \S (Perl anomaly) */
2894 case OP_ANYNL:
2895 case OP_VSPACE:
2896 return next == -ESC_V || next == -ESC_d || next == -ESC_w;
2897
2898 case OP_NOT_VSPACE:
2899 return next == -ESC_v || next == -ESC_R;
2900
2901 case OP_WORDCHAR:
2902 return next == -ESC_W || next == -ESC_s || next == -ESC_h ||
2903 next == -ESC_v || next == -ESC_R;
2904
2905 case OP_NOT_WORDCHAR:
2906 return next == -ESC_w || next == -ESC_d;
2907
2908 default:
2909 return FALSE;
2910 }
2911
2912 /* Control does not reach here */
2913 }
2914
2915
2916
2917 /*************************************************
2918 * Compile one branch *
2919 *************************************************/
2920
2921 /* Scan the pattern, compiling it into the a vector. If the options are
2922 changed during the branch, the pointer is used to change the external options
2923 bits. This function is used during the pre-compile phase when we are trying
2924 to find out the amount of memory needed, as well as during the real compile
2925 phase. The value of lengthptr distinguishes the two phases.
2926
2927 Arguments:
2928 optionsptr pointer to the option bits
2929 codeptr points to the pointer to the current code point
2930 ptrptr points to the current pattern pointer
2931 errorcodeptr points to error code variable
2932 firstbyteptr set to initial literal character, or < 0 (REQ_UNSET, REQ_NONE)
2933 reqbyteptr set to the last literal character required, else < 0
2934 bcptr points to current branch chain
2935 cd contains pointers to tables etc.
2936 lengthptr NULL during the real compile phase
2937 points to length accumulator during pre-compile phase
2938
2939 Returns: TRUE on success
2940 FALSE, with *errorcodeptr set non-zero on error
2941 */
2942
2943 static BOOL
2944 compile_branch(int *optionsptr, uschar **codeptr, const uschar **ptrptr,
2945 int *errorcodeptr, int *firstbyteptr, int *reqbyteptr, branch_chain *bcptr,
2946 compile_data *cd, int *lengthptr)
2947 {
2948 int repeat_type, op_type;
2949 int repeat_min = 0, repeat_max = 0; /* To please picky compilers */
2950 int bravalue = 0;
2951 int greedy_default, greedy_non_default;
2952 int firstbyte, reqbyte;
2953 int zeroreqbyte, zerofirstbyte;
2954 int req_caseopt, reqvary, tempreqvary;
2955 int options = *optionsptr;
2956 int after_manual_callout = 0;
2957 int length_prevgroup = 0;
2958 register int c;
2959 register uschar *code = *codeptr;
2960 uschar *last_code = code;
2961 uschar *orig_code = code;
2962 uschar *tempcode;
2963 BOOL inescq = FALSE;
2964 BOOL groupsetfirstbyte = FALSE;
2965 const uschar *ptr = *ptrptr;
2966 const uschar *tempptr;
2967 const uschar *nestptr = NULL;
2968 uschar *previous = NULL;
2969 uschar *previous_callout = NULL;
2970 uschar *save_hwm = NULL;
2971 uschar classbits[32];
2972
2973 #ifdef SUPPORT_UTF8
2974 BOOL class_utf8;
2975 BOOL utf8 = (options & PCRE_UTF8) != 0;
2976 uschar *class_utf8data;
2977 uschar *class_utf8data_base;
2978 uschar utf8_char[6];
2979 #else
2980 BOOL utf8 = FALSE;
2981 uschar *utf8_char = NULL;
2982 #endif
2983
2984 #ifdef PCRE_DEBUG
2985 if (lengthptr != NULL) DPRINTF((">> start branch\n"));
2986 #endif
2987
2988 /* Set up the default and non-default settings for greediness */
2989
2990 greedy_default = ((options & PCRE_UNGREEDY) != 0);
2991 greedy_non_default = greedy_default ^ 1;
2992
2993 /* Initialize no first byte, no required byte. REQ_UNSET means "no char
2994 matching encountered yet". It gets changed to REQ_NONE if we hit something that
2995 matches a non-fixed char first char; reqbyte just remains unset if we never
2996 find one.
2997
2998 When we hit a repeat whose minimum is zero, we may have to adjust these values
2999 to take the zero repeat into account. This is implemented by setting them to
3000 zerofirstbyte and zeroreqbyte when such a repeat is encountered. The individual
3001 item types that can be repeated set these backoff variables appropriately. */
3002
3003 firstbyte = reqbyte = zerofirstbyte = zeroreqbyte = REQ_UNSET;
3004
3005 /* The variable req_caseopt contains either the REQ_CASELESS value or zero,
3006 according to the current setting of the caseless flag. REQ_CASELESS is a bit
3007 value > 255. It is added into the firstbyte or reqbyte variables to record the
3008 case status of the value. This is used only for ASCII characters. */
3009
3010 req_caseopt = ((options & PCRE_CASELESS) != 0)? REQ_CASELESS : 0;
3011
3012 /* Switch on next character until the end of the branch */
3013
3014 for (;; ptr++)
3015 {
3016 BOOL negate_class;
3017 BOOL should_flip_negation;
3018 BOOL possessive_quantifier;
3019 BOOL is_quantifier;
3020 BOOL is_recurse;
3021 BOOL reset_bracount;
3022 int class_charcount;
3023 int class_lastchar;
3024 int newoptions;
3025 int recno;
3026 int refsign;
3027 int skipbytes;
3028 int subreqbyte;
3029 int subfirstbyte;
3030 int terminator;
3031 int mclength;
3032 uschar mcbuffer[8];
3033
3034 /* Get next byte in the pattern */
3035
3036 c = *ptr;
3037
3038 /* If we are at the end of a nested substitution, revert to the outer level
3039 string. Nesting only happens one level deep. */
3040
3041 if (c == 0 && nestptr != NULL)
3042 {
3043 ptr = nestptr;
3044 nestptr = NULL;
3045 c = *ptr;
3046 }
3047
3048 /* If we are in the pre-compile phase, accumulate the length used for the
3049 previous cycle of this loop. */
3050
3051 if (lengthptr != NULL)
3052 {
3053 #ifdef PCRE_DEBUG
3054 if (code > cd->hwm) cd->hwm = code; /* High water info */
3055 #endif
3056 if (code > cd->start_workspace + WORK_SIZE_CHECK) /* Check for overrun */
3057 {
3058 *errorcodeptr = ERR52;
3059 goto FAILED;
3060 }
3061
3062 /* There is at least one situation where code goes backwards: this is the
3063 case of a zero quantifier after a class (e.g. [ab]{0}). At compile time,
3064 the class is simply eliminated. However, it is created first, so we have to
3065 allow memory for it. Therefore, don't ever reduce the length at this point.
3066 */
3067
3068 if (code < last_code) code = last_code;
3069
3070 /* Paranoid check for integer overflow */
3071
3072 if (OFLOW_MAX - *lengthptr < code - last_code)
3073 {
3074 *errorcodeptr = ERR20;
3075 goto FAILED;
3076 }
3077
3078 *lengthptr += (int)(code - last_code);
3079 DPRINTF(("length=%d added %d c=%c\n", *lengthptr, code - last_code, c));
3080
3081 /* If "previous" is set and it is not at the start of the work space, move
3082 it back to there, in order to avoid filling up the work space. Otherwise,
3083 if "previous" is NULL, reset the current code pointer to the start. */
3084
3085 if (previous != NULL)
3086 {
3087 if (previous > orig_code)
3088 {
3089 memmove(orig_code, previous, code - previous);
3090 code -= previous - orig_code;
3091 previous = orig_code;
3092 }
3093 }
3094 else code = orig_code;
3095
3096 /* Remember where this code item starts so we can pick up the length
3097 next time round. */
3098
3099 last_code = code;
3100 }
3101
3102 /* In the real compile phase, just check the workspace used by the forward
3103 reference list. */
3104
3105 else if (cd->hwm > cd->start_workspace + WORK_SIZE_CHECK)
3106 {
3107 *errorcodeptr = ERR52;
3108 goto FAILED;
3109 }
3110
3111 /* If in \Q...\E, check for the end; if not, we have a literal */
3112
3113 if (inescq && c != 0)
3114 {
3115 if (c == CHAR_BACKSLASH && ptr[1] == CHAR_E)
3116 {
3117 inescq = FALSE;
3118 ptr++;
3119 continue;
3120 }
3121 else
3122 {
3123 if (previous_callout != NULL)
3124 {
3125 if (lengthptr == NULL) /* Don't attempt in pre-compile phase */
3126 complete_callout(previous_callout, ptr, cd);
3127 previous_callout = NULL;
3128 }
3129 if ((options & PCRE_AUTO_CALLOUT) != 0)
3130 {
3131 previous_callout = code;
3132 code = auto_callout(code, ptr, cd);
3133 }
3134 goto NORMAL_CHAR;
3135 }
3136 }
3137
3138 /* Fill in length of a previous callout, except when the next thing is
3139 a quantifier. */
3140
3141 is_quantifier =
3142 c == CHAR_ASTERISK || c == CHAR_PLUS || c == CHAR_QUESTION_MARK ||
3143 (c == CHAR_LEFT_CURLY_BRACKET && is_counted_repeat(ptr+1));
3144
3145 if (!is_quantifier && previous_callout != NULL &&
3146 after_manual_callout-- <= 0)
3147 {
3148 if (lengthptr == NULL) /* Don't attempt in pre-compile phase */
3149 complete_callout(previous_callout, ptr, cd);
3150 previous_callout = NULL;
3151 }
3152
3153 /* In extended mode, skip white space and comments */
3154
3155 if ((options & PCRE_EXTENDED) != 0)
3156 {
3157 if ((cd->ctypes[c] & ctype_space) != 0) continue;
3158 if (c == CHAR_NUMBER_SIGN)
3159 {
3160 ptr++;
3161 while (*ptr != 0)
3162 {
3163 if (IS_NEWLINE(ptr)) { ptr += cd->nllen - 1; break; }
3164 ptr++;
3165 #ifdef SUPPORT_UTF8
3166 if (utf8) while ((*ptr & 0xc0) == 0x80) ptr++;
3167 #endif
3168 }
3169 if (*ptr != 0) continue;
3170
3171 /* Else fall through to handle end of string */
3172 c = 0;
3173 }
3174 }
3175
3176 /* No auto callout for quantifiers. */
3177
3178 if ((options & PCRE_AUTO_CALLOUT) != 0 && !is_quantifier)
3179 {
3180 previous_callout = code;
3181 code = auto_callout(code, ptr, cd);
3182 }
3183
3184 switch(c)
3185 {
3186 /* ===================================================================*/
3187 case 0: /* The branch terminates at string end */
3188 case CHAR_VERTICAL_LINE: /* or | or ) */
3189 case CHAR_RIGHT_PARENTHESIS:
3190 *firstbyteptr = firstbyte;
3191 *reqbyteptr = reqbyte;
3192 *codeptr = code;
3193 *ptrptr = ptr;
3194 if (lengthptr != NULL)
3195 {
3196 if (OFLOW_MAX - *lengthptr < code - last_code)
3197 {
3198 *errorcodeptr = ERR20;
3199 goto FAILED;
3200 }
3201 *lengthptr += (int)(code - last_code); /* To include callout length */
3202 DPRINTF((">> end branch\n"));
3203 }
3204 return TRUE;
3205
3206
3207 /* ===================================================================*/
3208 /* Handle single-character metacharacters. In multiline mode, ^ disables
3209 the setting of any following char as a first character. */
3210
3211 case CHAR_CIRCUMFLEX_ACCENT:
3212 if ((options & PCRE_MULTILINE) != 0)
3213 {
3214 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
3215 }
3216 previous = NULL;
3217 *code++ = OP_CIRC;
3218 break;
3219
3220 case CHAR_DOLLAR_SIGN:
3221 previous = NULL;
3222 *code++ = OP_DOLL;
3223 break;
3224
3225 /* There can never be a first char if '.' is first, whatever happens about
3226 repeats. The value of reqbyte doesn't change either. */
3227
3228 case CHAR_DOT:
3229 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
3230 zerofirstbyte = firstbyte;
3231 zeroreqbyte = reqbyte;
3232 previous = code;
3233 *code++ = ((options & PCRE_DOTALL) != 0)? OP_ALLANY: OP_ANY;
3234 break;
3235
3236
3237 /* ===================================================================*/
3238 /* Character classes. If the included characters are all < 256, we build a
3239 32-byte bitmap of the permitted characters, except in the special case
3240 where there is only one such character. For negated classes, we build the
3241 map as usual, then invert it at the end. However, we use a different opcode
3242 so that data characters > 255 can be handled correctly.
3243
3244 If the class contains characters outside the 0-255 range, a different
3245 opcode is compiled. It may optionally have a bit map for characters < 256,
3246 but those above are are explicitly listed afterwards. A flag byte tells
3247 whether the bitmap is present, and whether this is a negated class or not.
3248
3249 In JavaScript compatibility mode, an isolated ']' causes an error. In
3250 default (Perl) mode, it is treated as a data character. */
3251
3252 case CHAR_RIGHT_SQUARE_BRACKET:
3253 if ((cd->external_options & PCRE_JAVASCRIPT_COMPAT) != 0)
3254 {
3255 *errorcodeptr = ERR64;
3256 goto FAILED;
3257 }
3258 goto NORMAL_CHAR;
3259
3260 case CHAR_LEFT_SQUARE_BRACKET:
3261 previous = code;
3262
3263 /* PCRE supports POSIX class stuff inside a class. Perl gives an error if
3264 they are encountered at the top level, so we'll do that too. */
3265
3266 if ((ptr[1] == CHAR_COLON || ptr[1] == CHAR_DOT ||
3267 ptr[1] == CHAR_EQUALS_SIGN) &&
3268 check_posix_syntax(ptr, &tempptr))
3269 {
3270 *errorcodeptr = (ptr[1] == CHAR_COLON)? ERR13 : ERR31;
3271 goto FAILED;
3272 }
3273
3274 /* If the first character is '^', set the negation flag and skip it. Also,
3275 if the first few characters (either before or after ^) are \Q\E or \E we
3276 skip them too. This makes for compatibility with Perl. */
3277
3278 negate_class = FALSE;
3279 for (;;)
3280 {
3281 c = *(++ptr);
3282 if (c == CHAR_BACKSLASH)
3283 {
3284 if (ptr[1] == CHAR_E)
3285 ptr++;
3286 else if (strncmp((const char *)ptr+1,
3287 STR_Q STR_BACKSLASH STR_E, 3) == 0)
3288 ptr += 3;
3289 else
3290 break;
3291 }
3292 else if (!negate_class && c == CHAR_CIRCUMFLEX_ACCENT)
3293 negate_class = TRUE;
3294 else break;
3295 }
3296
3297 /* Empty classes are allowed in JavaScript compatibility mode. Otherwise,
3298 an initial ']' is taken as a data character -- the code below handles
3299 that. In JS mode, [] must always fail, so generate OP_FAIL, whereas
3300 [^] must match any character, so generate OP_ALLANY. */
3301
3302 if (c == CHAR_RIGHT_SQUARE_BRACKET &&
3303 (cd->external_options & PCRE_JAVASCRIPT_COMPAT) != 0)
3304 {
3305 *code++ = negate_class? OP_ALLANY : OP_FAIL;
3306 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
3307 zerofirstbyte = firstbyte;
3308 break;
3309 }
3310
3311 /* If a class contains a negative special such as \S, we need to flip the
3312 negation flag at the end, so that support for characters > 255 works
3313 correctly (they are all included in the class). */
3314
3315 should_flip_negation = FALSE;
3316
3317 /* Keep a count of chars with values < 256 so that we can optimize the case
3318 of just a single character (as long as it's < 256). However, For higher
3319 valued UTF-8 characters, we don't yet do any optimization. */
3320
3321 class_charcount = 0;
3322 class_lastchar = -1;
3323
3324 /* Initialize the 32-char bit map to all zeros. We build the map in a
3325 temporary bit of memory, in case the class contains only 1 character (less
3326 than 256), because in that case the compiled code doesn't use the bit map.
3327 */
3328
3329 memset(classbits, 0, 32 * sizeof(uschar));
3330
3331 #ifdef SUPPORT_UTF8
3332 class_utf8 = FALSE; /* No chars >= 256 */
3333 class_utf8data = code + LINK_SIZE + 2; /* For UTF-8 items */
3334 class_utf8data_base = class_utf8data; /* For resetting in pass 1 */
3335 #endif
3336
3337 /* Process characters until ] is reached. By writing this as a "do" it
3338 means that an initial ] is taken as a data character. At the start of the
3339 loop, c contains the first byte of the character. */
3340
3341 if (c != 0) do
3342 {
3343 const uschar *oldptr;
3344
3345 #ifdef SUPPORT_UTF8
3346 if (utf8 && c > 127)
3347 { /* Braces are required because the */
3348 GETCHARLEN(c, ptr, ptr); /* macro generates multiple statements */
3349 }
3350
3351 /* In the pre-compile phase, accumulate the length of any UTF-8 extra
3352 data and reset the pointer. This is so that very large classes that
3353 contain a zillion UTF-8 characters no longer overwrite the work space
3354 (which is on the stack). */
3355
3356 if (lengthptr != NULL)
3357 {
3358 *lengthptr += class_utf8data - class_utf8data_base;
3359 class_utf8data = class_utf8data_base;
3360 }
3361
3362 #endif
3363
3364 /* Inside \Q...\E everything is literal except \E */
3365
3366 if (inescq)
3367 {
3368 if (c == CHAR_BACKSLASH && ptr[1] == CHAR_E) /* If we are at \E */
3369 {
3370 inescq = FALSE; /* Reset literal state */
3371 ptr++; /* Skip the 'E' */
3372 continue; /* Carry on with next */
3373 }
3374 goto CHECK_RANGE; /* Could be range if \E follows */
3375 }
3376
3377 /* Handle POSIX class names. Perl allows a negation extension of the
3378 form [:^name:]. A square bracket that doesn't match the syntax is
3379 treated as a literal. We also recognize the POSIX constructions
3380 [.ch.] and [=ch=] ("collating elements") and fault them, as Perl
3381 5.6 and 5.8 do. */
3382
3383 if (c == CHAR_LEFT_SQUARE_BRACKET &&
3384 (ptr[1] == CHAR_COLON || ptr[1] == CHAR_DOT ||
3385 ptr[1] == CHAR_EQUALS_SIGN) && check_posix_syntax(ptr, &tempptr))
3386 {
3387 BOOL local_negate = FALSE;
3388 int posix_class, taboffset, tabopt;
3389 register const uschar *cbits = cd->cbits;
3390 uschar pbits[32];
3391
3392 if (ptr[1] != CHAR_COLON)
3393 {
3394 *errorcodeptr = ERR31;
3395 goto FAILED;
3396 }
3397
3398 ptr += 2;
3399 if (*ptr == CHAR_CIRCUMFLEX_ACCENT)
3400 {
3401 local_negate = TRUE;
3402 should_flip_negation = TRUE; /* Note negative special */
3403 ptr++;
3404 }
3405
3406 posix_class = check_posix_name(ptr, (int)(tempptr - ptr));
3407 if (posix_class < 0)
3408 {
3409 *errorcodeptr = ERR30;
3410 goto FAILED;
3411 }
3412
3413 /* If matching is caseless, upper and lower are converted to
3414 alpha. This relies on the fact that the class table starts with
3415 alpha, lower, upper as the first 3 entries. */
3416
3417 if ((options & PCRE_CASELESS) != 0 && posix_class <= 2)
3418 posix_class = 0;
3419
3420 /* When PCRE_UCP is set, some of the POSIX classes are converted to
3421 different escape sequences that use Unicode properties. */
3422
3423 #ifdef SUPPORT_UCP
3424 if ((options & PCRE_UCP) != 0)
3425 {
3426 int pc = posix_class + ((local_negate)? POSIX_SUBSIZE/2 : 0);
3427 if (posix_substitutes[pc] != NULL)
3428 {
3429 nestptr = tempptr + 1;
3430 ptr = posix_substitutes[pc] - 1;
3431 continue;
3432 }
3433 }
3434 #endif
3435 /* In the non-UCP case, we build the bit map for the POSIX class in a
3436 chunk of local store because we may be adding and subtracting from it,
3437 and we don't want to subtract bits that may be in the main map already.
3438 At the end we or the result into the bit map that is being built. */
3439
3440 posix_class *= 3;
3441
3442 /* Copy in the first table (always present) */
3443
3444 memcpy(pbits, cbits + posix_class_maps[posix_class],
3445 32 * sizeof(uschar));
3446
3447 /* If there is a second table, add or remove it as required. */
3448
3449 taboffset = posix_class_maps[posix_class + 1];
3450 tabopt = posix_class_maps[posix_class + 2];
3451
3452 if (taboffset >= 0)
3453 {
3454 if (tabopt >= 0)
3455 for (c = 0; c < 32; c++) pbits[c] |= cbits[c + taboffset];
3456 else
3457 for (c = 0; c < 32; c++) pbits[c] &= ~cbits[c + taboffset];
3458 }
3459
3460 /* Not see if we need to remove any special characters. An option
3461 value of 1 removes vertical space and 2 removes underscore. */
3462
3463 if (tabopt < 0) tabopt = -tabopt;
3464 if (tabopt == 1) pbits[1] &= ~0x3c;
3465 else if (tabopt == 2) pbits[11] &= 0x7f;
3466
3467 /* Add the POSIX table or its complement into the main table that is
3468 being built and we are done. */
3469
3470 if (local_negate)
3471 for (c = 0; c < 32; c++) classbits[c] |= ~pbits[c];
3472 else
3473 for (c = 0; c < 32; c++) classbits[c] |= pbits[c];
3474
3475 ptr = tempptr + 1;
3476 class_charcount = 10; /* Set > 1; assumes more than 1 per class */
3477 continue; /* End of POSIX syntax handling */
3478 }
3479
3480 /* Backslash may introduce a single character, or it may introduce one
3481 of the specials, which just set a flag. The sequence \b is a special
3482 case. Inside a class (and only there) it is treated as backspace. We
3483 assume that other escapes have more than one character in them, so set
3484 class_charcount bigger than one. Unrecognized escapes fall through and
3485 are either treated as literal characters (by default), or are faulted if
3486 PCRE_EXTRA is set. */
3487
3488 if (c == CHAR_BACKSLASH)
3489 {
3490 c = check_escape(&ptr, errorcodeptr, cd->bracount, options, TRUE);
3491 if (*errorcodeptr != 0) goto FAILED;
3492
3493 if (-c == ESC_b) c = CHAR_BS; /* \b is backspace in a class */
3494 else if (-c == ESC_Q) /* Handle start of quoted string */
3495 {
3496 if (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
3497 {
3498 ptr += 2; /* avoid empty string */
3499 }
3500 else inescq = TRUE;
3501 continue;
3502 }
3503 else if (-c == ESC_E) continue; /* Ignore orphan \E */
3504
3505 if (c < 0)
3506 {
3507 register const uschar *cbits = cd->cbits;
3508 class_charcount += 2; /* Greater than 1 is what matters */
3509
3510 switch (-c)
3511 {
3512 #ifdef SUPPORT_UCP
3513 case ESC_du: /* These are the values given for \d etc */
3514 case ESC_DU: /* when PCRE_UCP is set. We replace the */
3515 case ESC_wu: /* escape sequence with an appropriate \p */
3516 case ESC_WU: /* or \P to test Unicode properties instead */
3517 case ESC_su: /* of the default ASCII testing. */
3518 case ESC_SU:
3519 nestptr = ptr;
3520 ptr = substitutes[-c - ESC_DU] - 1; /* Just before substitute */
3521 class_charcount -= 2; /* Undo! */
3522 continue;
3523 #endif
3524 case ESC_d:
3525 for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_digit];
3526 continue;
3527
3528 case ESC_D:
3529 should_flip_negation = TRUE;
3530 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_digit];
3531 continue;
3532
3533 case ESC_w:
3534 for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_word];
3535 continue;
3536
3537 case ESC_W:
3538 should_flip_negation = TRUE;
3539 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_word];
3540 continue;
3541
3542 /* Perl 5.004 onwards omits VT from \s, but we must preserve it
3543 if it was previously set by something earlier in the character
3544 class. */
3545
3546 case ESC_s:
3547 classbits[0] |= cbits[cbit_space];
3548 classbits[1] |= cbits[cbit_space+1] & ~0x08;
3549 for (c = 2; c < 32; c++) classbits[c] |= cbits[c+cbit_space];
3550 continue;
3551
3552 case ESC_S:
3553 should_flip_negation = TRUE;
3554 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_space];
3555 classbits[1] |= 0x08; /* Perl 5.004 onwards omits VT from \s */
3556 continue;
3557
3558 case ESC_h:
3559 SETBIT(classbits, 0x09); /* VT */
3560 SETBIT(classbits, 0x20); /* SPACE */
3561 SETBIT(classbits, 0xa0); /* NSBP */
3562 #ifdef SUPPORT_UTF8
3563 if (utf8)
3564 {
3565 class_utf8 = TRUE;
3566 *class_utf8data++ = XCL_SINGLE;
3567 class_utf8data += _pcre_ord2utf8(0x1680, class_utf8data);
3568 *class_utf8data++ = XCL_SINGLE;
3569 class_utf8data += _pcre_ord2utf8(0x180e, class_utf8data);
3570 *class_utf8data++ = XCL_RANGE;
3571 class_utf8data += _pcre_ord2utf8(0x2000, class_utf8data);
3572 class_utf8data += _pcre_ord2utf8(0x200A, class_utf8data);
3573 *class_utf8data++ = XCL_SINGLE;
3574 class_utf8data += _pcre_ord2utf8(0x202f, class_utf8data);
3575 *class_utf8data++ = XCL_SINGLE;
3576 class_utf8data += _pcre_ord2utf8(0x205f, class_utf8data);
3577 *class_utf8data++ = XCL_SINGLE;
3578 class_utf8data += _pcre_ord2utf8(0x3000, class_utf8data);
3579 }
3580 #endif
3581 continue;
3582
3583 case ESC_H:
3584 for (c = 0; c < 32; c++)
3585 {
3586 int x = 0xff;
3587 switch (c)
3588 {
3589 case 0x09/8: x ^= 1 << (0x09%8); break;
3590 case 0x20/8: x ^= 1 << (0x20%8); break;
3591 case 0xa0/8: x ^= 1 << (0xa0%8); break;
3592 default: break;
3593 }
3594 classbits[c] |= x;
3595 }
3596
3597 #ifdef SUPPORT_UTF8
3598 if (utf8)
3599 {
3600 class_utf8 = TRUE;
3601 *class_utf8data++ = XCL_RANGE;
3602 class_utf8data += _pcre_ord2utf8(0x0100, class_utf8data);
3603 class_utf8data += _pcre_ord2utf8(0x167f, class_utf8data);
3604 *class_utf8data++ = XCL_RANGE;
3605 class_utf8data += _pcre_ord2utf8(0x1681, class_utf8data);
3606 class_utf8data += _pcre_ord2utf8(0x180d, class_utf8data);
3607 *class_utf8data++ = XCL_RANGE;
3608 class_utf8data += _pcre_ord2utf8(0x180f, class_utf8data);
3609 class_utf8data += _pcre_ord2utf8(0x1fff, class_utf8data);
3610 *class_utf8data++ = XCL_RANGE;
3611 class_utf8data += _pcre_ord2utf8(0x200B, class_utf8data);
3612 class_utf8data += _pcre_ord2utf8(0x202e, class_utf8data);
3613 *class_utf8data++ = XCL_RANGE;
3614 class_utf8data += _pcre_ord2utf8(0x2030, class_utf8data);
3615 class_utf8data += _pcre_ord2utf8(0x205e, class_utf8data);
3616 *class_utf8data++ = XCL_RANGE;
3617 class_utf8data += _pcre_ord2utf8(0x2060, class_utf8data);
3618 class_utf8data += _pcre_ord2utf8(0x2fff, class_utf8data);
3619 *class_utf8data++ = XCL_RANGE;
3620 class_utf8data += _pcre_ord2utf8(0x3001, class_utf8data);
3621 class_utf8data += _pcre_ord2utf8(0x7fffffff, class_utf8data);
3622 }
3623 #endif
3624 continue;
3625
3626 case ESC_v:
3627 SETBIT(classbits, 0x0a); /* LF */
3628 SETBIT(classbits, 0x0b); /* VT */
3629 SETBIT(classbits, 0x0c); /* FF */
3630 SETBIT(classbits, 0x0d); /* CR */
3631 SETBIT(classbits, 0x85); /* NEL */
3632 #ifdef SUPPORT_UTF8
3633 if (utf8)
3634 {
3635 class_utf8 = TRUE;
3636 *class_utf8data++ = XCL_RANGE;
3637 class_utf8data += _pcre_ord2utf8(0x2028, class_utf8data);
3638 class_utf8data += _pcre_ord2utf8(0x2029, class_utf8data);
3639 }
3640 #endif
3641 continue;
3642
3643 case ESC_V:
3644 for (c = 0; c < 32; c++)
3645 {
3646 int x = 0xff;
3647 switch (c)
3648 {
3649 case 0x0a/8: x ^= 1 << (0x0a%8);
3650 x ^= 1 << (0x0b%8);
3651 x ^= 1 << (0x0c%8);
3652 x ^= 1 << (0x0d%8);
3653 break;
3654 case 0x85/8: x ^= 1 << (0x85%8); break;
3655 default: break;
3656 }
3657 classbits[c] |= x;
3658 }
3659
3660 #ifdef SUPPORT_UTF8
3661 if (utf8)
3662 {
3663 class_utf8 = TRUE;
3664 *class_utf8data++ = XCL_RANGE;
3665 class_utf8data += _pcre_ord2utf8(0x0100, class_utf8data);
3666 class_utf8data += _pcre_ord2utf8(0x2027, class_utf8data);
3667 *class_utf8data++ = XCL_RANGE;
3668 class_utf8data += _pcre_ord2utf8(0x2029, class_utf8data);
3669 class_utf8data += _pcre_ord2utf8(0x7fffffff, class_utf8data);
3670 }
3671 #endif
3672 continue;
3673
3674 #ifdef SUPPORT_UCP
3675 case ESC_p:
3676 case ESC_P:
3677 {
3678 BOOL negated;
3679 int pdata;
3680 int ptype = get_ucp(&ptr, &negated, &pdata, errorcodeptr);
3681 if (ptype < 0) goto FAILED;
3682 class_utf8 = TRUE;
3683 *class_utf8data++ = ((-c == ESC_p) != negated)?
3684 XCL_PROP : XCL_NOTPROP;
3685 *class_utf8data++ = ptype;
3686 *class_utf8data++ = pdata;
3687 class_charcount -= 2; /* Not a < 256 character */
3688 continue;
3689 }
3690 #endif
3691 /* Unrecognized escapes are faulted if PCRE is running in its
3692 strict mode. By default, for compatibility with Perl, they are
3693 treated as literals. */
3694
3695 default:
3696 if ((options & PCRE_EXTRA) != 0)
3697 {
3698 *errorcodeptr = ERR7;
3699 goto FAILED;
3700 }
3701 class_charcount -= 2; /* Undo the default count from above */
3702 c = *ptr; /* Get the final character and fall through */
3703 break;
3704 }
3705 }
3706
3707 /* Fall through if we have a single character (c >= 0). This may be
3708 greater than 256 in UTF-8 mode. */
3709
3710 } /* End of backslash handling */
3711
3712 /* A single character may be followed by '-' to form a range. However,
3713 Perl does not permit ']' to be the end of the range. A '-' character
3714 at the end is treated as a literal. Perl ignores orphaned \E sequences
3715 entirely. The code for handling \Q and \E is messy. */
3716
3717 CHECK_RANGE:
3718 while (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
3719 {
3720 inescq = FALSE;
3721 ptr += 2;
3722 }
3723
3724 oldptr = ptr;
3725
3726 /* Remember \r or \n */
3727
3728 if (c == CHAR_CR || c == CHAR_NL) cd->external_flags |= PCRE_HASCRORLF;
3729
3730 /* Check for range */
3731
3732 if (!inescq && ptr[1] == CHAR_MINUS)
3733 {
3734 int d;
3735 ptr += 2;
3736 while (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_E) ptr += 2;
3737
3738 /* If we hit \Q (not followed by \E) at this point, go into escaped
3739 mode. */
3740
3741 while (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_Q)
3742 {
3743 ptr += 2;
3744 if (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_E)
3745 { ptr += 2; continue; }
3746 inescq = TRUE;
3747 break;
3748 }
3749
3750 if (*ptr == 0 || (!inescq && *ptr == CHAR_RIGHT_SQUARE_BRACKET))
3751 {
3752 ptr = oldptr;
3753 goto LONE_SINGLE_CHARACTER;
3754 }
3755
3756 #ifdef SUPPORT_UTF8
3757 if (utf8)
3758 { /* Braces are required because the */
3759 GETCHARLEN(d, ptr, ptr); /* macro generates multiple statements */
3760 }
3761 else
3762 #endif
3763 d = *ptr; /* Not UTF-8 mode */
3764
3765 /* The second part of a range can be a single-character escape, but
3766 not any of the other escapes. Perl 5.6 treats a hyphen as a literal
3767 in such circumstances. */
3768
3769 if (!inescq && d == CHAR_BACKSLASH)
3770 {
3771 d = check_escape(&ptr, errorcodeptr, cd->bracount, options, TRUE);
3772 if (*errorcodeptr != 0) goto FAILED;
3773
3774 /* \b is backspace; any other special means the '-' was literal */
3775
3776 if (d < 0)
3777 {
3778 if (d == -ESC_b) d = CHAR_BS; else
3779 {
3780 ptr = oldptr;
3781 goto LONE_SINGLE_CHARACTER; /* A few lines below */
3782 }
3783 }
3784 }
3785
3786 /* Check that the two values are in the correct order. Optimize
3787 one-character ranges */
3788
3789 if (d < c)
3790 {
3791 *errorcodeptr = ERR8;
3792 goto FAILED;
3793 }
3794
3795 if (d == c) goto LONE_SINGLE_CHARACTER; /* A few lines below */
3796
3797 /* Remember \r or \n */
3798
3799 if (d == CHAR_CR || d == CHAR_NL) cd->external_flags |= PCRE_HASCRORLF;
3800
3801 /* In UTF-8 mode, if the upper limit is > 255, or > 127 for caseless
3802 matching, we have to use an XCLASS with extra data items. Caseless
3803 matching for characters > 127 is available only if UCP support is
3804 available. */
3805
3806 #ifdef SUPPORT_UTF8
3807 if (utf8 && (d > 255 || ((options & PCRE_CASELESS) != 0 && d > 127)))
3808 {
3809 class_utf8 = TRUE;
3810
3811 /* With UCP support, we can find the other case equivalents of
3812 the relevant characters. There may be several ranges. Optimize how
3813 they fit with the basic range. */
3814
3815 #ifdef SUPPORT_UCP
3816 if ((options & PCRE_CASELESS) != 0)
3817 {
3818 unsigned int occ, ocd;
3819 unsigned int cc = c;
3820 unsigned int origd = d;
3821 while (get_othercase_range(&cc, origd, &occ, &ocd))
3822 {
3823 if (occ >= (unsigned int)c &&
3824 ocd <= (unsigned int)d)
3825 continue; /* Skip embedded ranges */
3826
3827 if (occ < (unsigned int)c &&
3828 ocd >= (unsigned int)c - 1) /* Extend the basic range */
3829 { /* if there is overlap, */
3830 c = occ; /* noting that if occ < c */
3831 continue; /* we can't have ocd > d */
3832 } /* because a subrange is */
3833 if (ocd > (unsigned int)d &&
3834 occ <= (unsigned int)d + 1) /* always shorter than */
3835 { /* the basic range. */
3836 d = ocd;
3837 continue;
3838 }
3839
3840 if (occ == ocd)
3841 {
3842 *class_utf8data++ = XCL_SINGLE;
3843 }
3844 else
3845 {
3846 *class_utf8data++ = XCL_RANGE;
3847 class_utf8data += _pcre_ord2utf8(occ, class_utf8data);
3848 }
3849 class_utf8data += _pcre_ord2utf8(ocd, class_utf8data);
3850 }
3851 }
3852 #endif /* SUPPORT_UCP */
3853
3854 /* Now record the original range, possibly modified for UCP caseless
3855 overlapping ranges. */
3856
3857 *class_utf8data++ = XCL_RANGE;
3858 class_utf8data += _pcre_ord2utf8(c, class_utf8data);
3859 class_utf8data += _pcre_ord2utf8(d, class_utf8data);
3860
3861 /* With UCP support, we are done. Without UCP support, there is no
3862 caseless matching for UTF-8 characters > 127; we can use the bit map
3863 for the smaller ones. */
3864
3865 #ifdef SUPPORT_UCP
3866 continue; /* With next character in the class */
3867 #else
3868 if ((options & PCRE_CASELESS) == 0 || c > 127) continue;
3869
3870 /* Adjust upper limit and fall through to set up the map */
3871
3872 d = 127;
3873
3874 #endif /* SUPPORT_UCP */
3875 }
3876 #endif /* SUPPORT_UTF8 */
3877
3878 /* We use the bit map for all cases when not in UTF-8 mode; else
3879 ranges that lie entirely within 0-127 when there is UCP support; else
3880 for partial ranges without UCP support. */
3881
3882 class_charcount += d - c + 1;
3883 class_lastchar = d;
3884
3885 /* We can save a bit of time by skipping this in the pre-compile. */
3886
3887 if (lengthptr == NULL) for (; c <= d; c++)
3888 {
3889 classbits[c/8] |= (1 << (c&7));
3890 if ((options & PCRE_CASELESS) != 0)
3891 {
3892 int uc = cd->fcc[c]; /* flip case */
3893 classbits[uc/8] |= (1 << (uc&7));
3894 }
3895 }
3896
3897 continue; /* Go get the next char in the class */
3898 }
3899
3900 /* Handle a lone single character - we can get here for a normal
3901 non-escape char, or after \ that introduces a single character or for an
3902 apparent range that isn't. */
3903
3904 LONE_SINGLE_CHARACTER:
3905
3906 /* Handle a character that cannot go in the bit map */
3907
3908 #ifdef SUPPORT_UTF8
3909 if (utf8 && (c > 255 || ((options & PCRE_CASELESS) != 0 && c > 127)))
3910 {
3911 class_utf8 = TRUE;
3912 *class_utf8data++ = XCL_SINGLE;
3913 class_utf8data += _pcre_ord2utf8(c, class_utf8data);
3914
3915 #ifdef SUPPORT_UCP
3916 if ((options & PCRE_CASELESS) != 0)
3917 {
3918 unsigned int othercase;
3919 if ((othercase = UCD_OTHERCASE(c)) != c)
3920 {
3921 *class_utf8data++ = XCL_SINGLE;
3922 class_utf8data += _pcre_ord2utf8(othercase, class_utf8data);
3923 }
3924 }
3925 #endif /* SUPPORT_UCP */
3926
3927 }
3928 else
3929 #endif /* SUPPORT_UTF8 */
3930
3931 /* Handle a single-byte character */
3932 {
3933 classbits[c/8] |= (1 << (c&7));
3934 if ((options & PCRE_CASELESS) != 0)
3935 {
3936 c = cd->fcc[c]; /* flip case */
3937 classbits[c/8] |= (1 << (c&7));
3938 }
3939 class_charcount++;
3940 class_lastchar = c;
3941 }
3942 }
3943
3944 /* Loop until ']' reached. This "while" is the end of the "do" far above.
3945 If we are at the end of an internal nested string, revert to the outer
3946 string. */
3947
3948 while (((c = *(++ptr)) != 0 ||
3949 (nestptr != NULL &&
3950 (ptr = nestptr, nestptr = NULL, c = *(++ptr)) != 0)) &&
3951 (c != CHAR_RIGHT_SQUARE_BRACKET || inescq));
3952
3953 /* Check for missing terminating ']' */
3954
3955 if (c == 0)
3956 {
3957 *errorcodeptr = ERR6;
3958 goto FAILED;
3959 }
3960
3961 /* If class_charcount is 1, we saw precisely one character whose value is
3962 less than 256. As long as there were no characters >= 128 and there was no
3963 use of \p or \P, in other words, no use of any XCLASS features, we can
3964 optimize.
3965
3966 In UTF-8 mode, we can optimize the negative case only if there were no
3967 characters >= 128 because OP_NOT and the related opcodes like OP_NOTSTAR
3968 operate on single-bytes only. This is an historical hangover. Maybe one day
3969 we can tidy these opcodes to handle multi-byte characters.
3970
3971 The optimization throws away the bit map. We turn the item into a
3972 1-character OP_CHAR[NC] if it's positive, or OP_NOT if it's negative. Note
3973 that OP_NOT does not support multibyte characters. In the positive case, it
3974 can cause firstbyte to be set. Otherwise, there can be no first char if
3975 this item is first, whatever repeat count may follow. In the case of
3976 reqbyte, save the previous value for reinstating. */
3977
3978 #ifdef SUPPORT_UTF8
3979 if (class_charcount == 1 && !class_utf8 &&
3980 (!utf8 || !negate_class || class_lastchar < 128))
3981 #else
3982 if (class_charcount == 1)
3983 #endif
3984 {
3985 zeroreqbyte = reqbyte;
3986
3987 /* The OP_NOT opcode works on one-byte characters only. */
3988
3989 if (negate_class)
3990 {
3991 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
3992 zerofirstbyte = firstbyte;
3993 *code++ = OP_NOT;
3994 *code++ = class_lastchar;
3995 break;
3996 }
3997
3998 /* For a single, positive character, get the value into mcbuffer, and
3999 then we can handle this with the normal one-character code. */
4000
4001 #ifdef SUPPORT_UTF8
4002 if (utf8 && class_lastchar > 127)
4003 mclength = _pcre_ord2utf8(class_lastchar, mcbuffer);
4004 else
4005 #endif
4006 {
4007 mcbuffer[0] = class_lastchar;
4008 mclength = 1;
4009 }
4010 goto ONE_CHAR;
4011 } /* End of 1-char optimization */
4012
4013 /* The general case - not the one-char optimization. If this is the first
4014 thing in the branch, there can be no first char setting, whatever the
4015 repeat count. Any reqbyte setting must remain unchanged after any kind of
4016 repeat. */
4017
4018 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
4019 zerofirstbyte = firstbyte;
4020 zeroreqbyte = reqbyte;
4021
4022 /* If there are characters with values > 255, we have to compile an
4023 extended class, with its own opcode, unless there was a negated special
4024 such as \S in the class, and PCRE_UCP is not set, because in that case all
4025 characters > 255 are in the class, so any that were explicitly given as
4026 well can be ignored. If (when there are explicit characters > 255 that must
4027 be listed) there are no characters < 256, we can omit the bitmap in the
4028 actual compiled code. */
4029
4030 #ifdef SUPPORT_UTF8
4031 if (class_utf8 && (!should_flip_negation || (options & PCRE_UCP) != 0))
4032 {
4033 *class_utf8data++ = XCL_END; /* Marks the end of extra data */
4034 *code++ = OP_XCLASS;
4035 code += LINK_SIZE;
4036 *code = negate_class? XCL_NOT : 0;
4037
4038 /* If the map is required, move up the extra data to make room for it;
4039 otherwise just move the code pointer to the end of the extra data. */
4040
4041 if (class_charcount > 0)
4042 {
4043 *code++ |= XCL_MAP;
4044 memmove(code + 32, code, class_utf8data - code);
4045 memcpy(code, classbits, 32);
4046 code = class_utf8data + 32;
4047 }
4048 else code = class_utf8data;
4049
4050 /* Now fill in the complete length of the item */
4051
4052 PUT(previous, 1, code - previous);
4053 break; /* End of class handling */
4054 }
4055 #endif
4056
4057 /* If there are no characters > 255, or they are all to be included or
4058 excluded, set the opcode to OP_CLASS or OP_NCLASS, depending on whether the
4059 whole class was negated and whether there were negative specials such as \S
4060 (non-UCP) in the class. Then copy the 32-byte map into the code vector,
4061 negating it if necessary. */
4062
4063 *code++ = (negate_class == should_flip_negation) ? OP_CLASS : OP_NCLASS;
4064 if (negate_class)
4065 {
4066 if (lengthptr == NULL) /* Save time in the pre-compile phase */
4067 for (c = 0; c < 32; c++) code[c] = ~classbits[c];
4068 }
4069 else
4070 {
4071 memcpy(code, classbits, 32);
4072 }
4073 code += 32;
4074 break;
4075
4076
4077 /* ===================================================================*/
4078 /* Various kinds of repeat; '{' is not necessarily a quantifier, but this
4079 has been tested above. */
4080
4081 case CHAR_LEFT_CURLY_BRACKET:
4082 if (!is_quantifier) goto NORMAL_CHAR;
4083 ptr = read_repeat_counts(ptr+1, &repeat_min, &repeat_max, errorcodeptr);
4084 if (*errorcodeptr != 0) goto FAILED;
4085 goto REPEAT;
4086
4087 case CHAR_ASTERISK:
4088 repeat_min = 0;
4089 repeat_max = -1;
4090 goto REPEAT;
4091
4092 case CHAR_PLUS:
4093 repeat_min = 1;
4094 repeat_max = -1;
4095 goto REPEAT;
4096
4097 case CHAR_QUESTION_MARK:
4098 repeat_min = 0;
4099 repeat_max = 1;
4100
4101 REPEAT:
4102 if (previous == NULL)
4103 {
4104 *errorcodeptr = ERR9;
4105 goto FAILED;
4106 }
4107
4108 if (repeat_min == 0)
4109 {
4110 firstbyte = zerofirstbyte; /* Adjust for zero repeat */
4111 reqbyte = zeroreqbyte; /* Ditto */
4112 }
4113
4114 /* Remember whether this is a variable length repeat */
4115
4116 reqvary = (repeat_min == repeat_max)? 0 : REQ_VARY;
4117
4118 op_type = 0; /* Default single-char op codes */
4119 possessive_quantifier = FALSE; /* Default not possessive quantifier */
4120
4121 /* Save start of previous item, in case we have to move it up to make space
4122 for an inserted OP_ONCE for the additional '+' extension. */
4123
4124 tempcode = previous;
4125
4126 /* If the next character is '+', we have a possessive quantifier. This
4127 implies greediness, whatever the setting of the PCRE_UNGREEDY option.
4128 If the next character is '?' this is a minimizing repeat, by default,
4129 but if PCRE_UNGREEDY is set, it works the other way round. We change the
4130 repeat type to the non-default. */
4131
4132 if (ptr[1] == CHAR_PLUS)
4133 {
4134 repeat_type = 0; /* Force greedy */
4135 possessive_quantifier = TRUE;
4136 ptr++;
4137 }
4138 else if (ptr[1] == CHAR_QUESTION_MARK)
4139 {
4140 repeat_type = greedy_non_default;
4141 ptr++;
4142 }
4143 else repeat_type = greedy_default;
4144
4145 /* If previous was a character match, abolish the item and generate a
4146 repeat item instead. If a char item has a minumum of more than one, ensure
4147 that it is set in reqbyte - it might not be if a sequence such as x{3} is
4148 the first thing in a branch because the x will have gone into firstbyte
4149 instead. */
4150
4151 if (*previous == OP_CHAR || *previous == OP_CHARNC)
4152 {
4153 /* Deal with UTF-8 characters that take up more than one byte. It's
4154 easier to write this out separately than try to macrify it. Use c to
4155 hold the length of the character in bytes, plus 0x80 to flag that it's a
4156 length rather than a small character. */
4157
4158 #ifdef SUPPORT_UTF8
4159 if (utf8 && (code[-1] & 0x80) != 0)
4160 {
4161 uschar *lastchar = code - 1;
4162 while((*lastchar & 0xc0) == 0x80) lastchar--;
4163 c = code - lastchar; /* Length of UTF-8 character */
4164 memcpy(utf8_char, lastchar, c); /* Save the char */
4165 c |= 0x80; /* Flag c as a length */
4166 }
4167 else
4168 #endif
4169
4170 /* Handle the case of a single byte - either with no UTF8 support, or
4171 with UTF-8 disabled, or for a UTF-8 character < 128. */
4172
4173 {
4174 c = code[-1];
4175 if (repeat_min > 1) reqbyte = c | req_caseopt | cd->req_varyopt;
4176 }
4177
4178 /* If the repetition is unlimited, it pays to see if the next thing on
4179 the line is something that cannot possibly match this character. If so,
4180 automatically possessifying this item gains some performance in the case
4181 where the match fails. */
4182
4183 if (!possessive_quantifier &&
4184 repeat_max < 0 &&
4185 check_auto_possessive(previous, utf8, ptr + 1, options, cd))
4186 {
4187 repeat_type = 0; /* Force greedy */
4188 possessive_quantifier = TRUE;
4189 }
4190
4191 goto OUTPUT_SINGLE_REPEAT; /* Code shared with single character types */
4192 }
4193
4194 /* If previous was a single negated character ([^a] or similar), we use
4195 one of the special opcodes, replacing it. The code is shared with single-
4196 character repeats by setting opt_type to add a suitable offset into
4197 repeat_type. We can also test for auto-possessification. OP_NOT is
4198 currently used only for single-byte chars. */
4199
4200 else if (*previous == OP_NOT)
4201 {
4202 op_type = OP_NOTSTAR - OP_STAR; /* Use "not" opcodes */
4203 c = previous[1];
4204 if (!possessive_quantifier &&
4205 repeat_max < 0 &&
4206 check_auto_possessive(previous, utf8, ptr + 1, options, cd))
4207 {
4208 repeat_type = 0; /* Force greedy */
4209 possessive_quantifier = TRUE;
4210 }
4211 goto OUTPUT_SINGLE_REPEAT;
4212 }
4213
4214 /* If previous was a character type match (\d or similar), abolish it and
4215 create a suitable repeat item. The code is shared with single-character
4216 repeats by setting op_type to add a suitable offset into repeat_type. Note
4217 the the Unicode property types will be present only when SUPPORT_UCP is
4218 defined, but we don't wrap the little bits of code here because it just
4219 makes it horribly messy. */
4220
4221 else if (*previous < OP_EODN)
4222 {
4223 uschar *oldcode;
4224 int prop_type, prop_value;
4225 op_type = OP_TYPESTAR - OP_STAR; /* Use type opcodes */
4226 c = *previous;
4227
4228 if (!possessive_quantifier &&
4229 repeat_max < 0 &&
4230 check_auto_possessive(previous, utf8, ptr + 1, options, cd))
4231 {
4232 repeat_type = 0; /* Force greedy */
4233 possessive_quantifier = TRUE;
4234 }
4235
4236 OUTPUT_SINGLE_REPEAT:
4237 if (*previous == OP_PROP || *previous == OP_NOTPROP)
4238 {
4239 prop_type = previous[1];
4240 prop_value = previous[2];
4241 }
4242 else prop_type = prop_value = -1;
4243
4244 oldcode = code;
4245 code = previous; /* Usually overwrite previous item */
4246
4247 /* If the maximum is zero then the minimum must also be zero; Perl allows
4248 this case, so we do too - by simply omitting the item altogether. */
4249
4250 if (repeat_max == 0) goto END_REPEAT;
4251
4252 /*--------------------------------------------------------------------*/
4253 /* This code is obsolete from release 8.00; the restriction was finally
4254 removed: */
4255
4256 /* All real repeats make it impossible to handle partial matching (maybe
4257 one day we will be able to remove this restriction). */
4258
4259 /* if (repeat_max != 1) cd->external_flags |= PCRE_NOPARTIAL; */
4260 /*--------------------------------------------------------------------*/
4261
4262 /* Combine the op_type with the repeat_type */
4263
4264 repeat_type += op_type;
4265
4266 /* A minimum of zero is handled either as the special case * or ?, or as
4267 an UPTO, with the maximum given. */
4268
4269 if (repeat_min == 0)
4270 {
4271 if (repeat_max == -1) *code++ = OP_STAR + repeat_type;
4272 else if (repeat_max == 1) *code++ = OP_QUERY + repeat_type;
4273 else
4274 {
4275 *code++ = OP_UPTO + repeat_type;
4276 PUT2INC(code, 0, repeat_max);
4277 }
4278 }
4279
4280 /* A repeat minimum of 1 is optimized into some special cases. If the
4281 maximum is unlimited, we use OP_PLUS. Otherwise, the original item is
4282 left in place and, if the maximum is greater than 1, we use OP_UPTO with
4283 one less than the maximum. */
4284
4285 else if (repeat_min == 1)
4286 {
4287 if (repeat_max == -1)
4288 *code++ = OP_PLUS + repeat_type;
4289 else
4290 {
4291 code = oldcode; /* leave previous item in place */
4292 if (repeat_max == 1) goto END_REPEAT;
4293 *code++ = OP_UPTO + repeat_type;
4294 PUT2INC(code, 0, repeat_max - 1);
4295 }
4296 }
4297
4298 /* The case {n,n} is just an EXACT, while the general case {n,m} is
4299 handled as an EXACT followed by an UPTO. */
4300
4301 else
4302 {
4303 *code++ = OP_EXACT + op_type; /* NB EXACT doesn't have repeat_type */
4304 PUT2INC(code, 0, repeat_min);
4305
4306 /* If the maximum is unlimited, insert an OP_STAR. Before doing so,
4307 we have to insert the character for the previous code. For a repeated
4308 Unicode property match, there are two extra bytes that define the
4309 required property. In UTF-8 mode, long characters have their length in
4310 c, with the 0x80 bit as a flag. */
4311
4312 if (repeat_max < 0)
4313 {
4314 #ifdef SUPPORT_UTF8
4315 if (utf8 && c >= 128)
4316 {
4317 memcpy(code, utf8_char, c & 7);
4318 code += c & 7;
4319 }
4320 else
4321 #endif
4322 {
4323 *code++ = c;
4324 if (prop_type >= 0)
4325 {
4326 *code++ = prop_type;
4327 *code++ = prop_value;
4328 }
4329 }
4330 *code++ = OP_STAR + repeat_type;
4331 }
4332
4333 /* Else insert an UPTO if the max is greater than the min, again
4334 preceded by the character, for the previously inserted code. If the
4335 UPTO is just for 1 instance, we can use QUERY instead. */
4336
4337 else if (repeat_max != repeat_min)
4338 {
4339 #ifdef SUPPORT_UTF8
4340 if (utf8 && c >= 128)
4341 {
4342 memcpy(code, utf8_char, c & 7);
4343 code += c & 7;
4344 }
4345 else
4346 #endif
4347 *code++ = c;
4348 if (prop_type >= 0)
4349 {
4350 *code++ = prop_type;
4351 *code++ = prop_value;
4352 }
4353 repeat_max -= repeat_min;
4354
4355 if (repeat_max == 1)
4356 {
4357 *code++ = OP_QUERY + repeat_type;
4358 }
4359 else
4360 {
4361 *code++ = OP_UPTO + repeat_type;
4362 PUT2INC(code, 0, repeat_max);
4363 }
4364 }
4365 }
4366
4367 /* The character or character type itself comes last in all cases. */
4368
4369 #ifdef SUPPORT_UTF8
4370 if (utf8 && c >= 128)
4371 {
4372 memcpy(code, utf8_char, c & 7);
4373 code += c & 7;
4374 }
4375 else
4376 #endif
4377 *code++ = c;
4378
4379 /* For a repeated Unicode property match, there are two extra bytes that
4380 define the required property. */
4381
4382 #ifdef SUPPORT_UCP
4383 if (prop_type >= 0)
4384 {
4385 *code++ = prop_type;
4386 *code++ = prop_value;
4387 }
4388 #endif
4389 }
4390
4391 /* If previous was a character class or a back reference, we put the repeat
4392 stuff after it, but just skip the item if the repeat was {0,0}. */
4393
4394 else if (*previous == OP_CLASS ||
4395 *previous == OP_NCLASS ||
4396 #ifdef SUPPORT_UTF8
4397 *previous == OP_XCLASS ||
4398 #endif
4399 *previous == OP_REF)
4400 {
4401 if (repeat_max == 0)
4402 {
4403 code = previous;
4404 goto END_REPEAT;
4405 }
4406
4407 /*--------------------------------------------------------------------*/
4408 /* This code is obsolete from release 8.00; the restriction was finally
4409 removed: */
4410
4411 /* All real repeats make it impossible to handle partial matching (maybe
4412 one day we will be able to remove this restriction). */
4413
4414 /* if (repeat_max != 1) cd->external_flags |= PCRE_NOPARTIAL; */
4415 /*--------------------------------------------------------------------*/
4416
4417 if (repeat_min == 0 && repeat_max == -1)
4418 *code++ = OP_CRSTAR + repeat_type;
4419 else if (repeat_min == 1 && repeat_max == -1)
4420 *code++ = OP_CRPLUS + repeat_type;
4421 else if (repeat_min == 0 && repeat_max == 1)
4422 *code++ = OP_CRQUERY + repeat_type;
4423 else
4424 {
4425 *code++ = OP_CRRANGE + repeat_type;
4426 PUT2INC(code, 0, repeat_min);
4427 if (repeat_max == -1) repeat_max = 0; /* 2-byte encoding for max */
4428 PUT2INC(code, 0, repeat_max);
4429 }
4430 }
4431
4432 /* If previous was a bracket group, we may have to replicate it in certain
4433 cases. */
4434
4435 else if (*previous == OP_BRA || *previous == OP_CBRA ||
4436 *previous == OP_ONCE || *previous == OP_COND)
4437 {
4438 register int i;
4439 int ketoffset = 0;
4440 int len = (int)(code - previous);
4441 uschar *bralink = NULL;
4442
4443 /* Repeating a DEFINE group is pointless */
4444
4445 if (*previous == OP_COND && previous[LINK_SIZE+1] == OP_DEF)
4446 {
4447 *errorcodeptr = ERR55;
4448 goto FAILED;
4449 }
4450
4451 /* If the maximum repeat count is unlimited, find the end of the bracket
4452 by scanning through from the start, and compute the offset back to it
4453 from the current code pointer. There may be an OP_OPT setting following
4454 the final KET, so we can't find the end just by going back from the code
4455 pointer. */
4456
4457 if (repeat_max == -1)
4458 {
4459 register uschar *ket = previous;
4460 do ket += GET(ket, 1); while (*ket != OP_KET);
4461 ketoffset = (int)(code - ket);
4462 }
4463
4464 /* The case of a zero minimum is special because of the need to stick
4465 OP_BRAZERO in front of it, and because the group appears once in the
4466 data, whereas in other cases it appears the minimum number of times. For
4467 this reason, it is simplest to treat this case separately, as otherwise
4468 the code gets far too messy. There are several special subcases when the
4469 minimum is zero. */
4470
4471 if (repeat_min == 0)
4472 {
4473 /* If the maximum is also zero, we used to just omit the group from the
4474 output altogether, like this:
4475
4476 ** if (repeat_max == 0)
4477 ** {
4478 ** code = previous;
4479 ** goto END_REPEAT;
4480 ** }
4481
4482 However, that fails when a group is referenced as a subroutine from
4483 elsewhere in the pattern, so now we stick in OP_SKIPZERO in front of it
4484 so that it is skipped on execution. As we don't have a list of which
4485 groups are referenced, we cannot do this selectively.
4486
4487 If the maximum is 1 or unlimited, we just have to stick in the BRAZERO
4488 and do no more at this point. However, we do need to adjust any
4489 OP_RECURSE calls inside the group that refer to the group itself or any
4490 internal or forward referenced group, because the offset is from the
4491 start of the whole regex. Temporarily terminate the pattern while doing
4492 this. */
4493
4494 if (repeat_max <= 1) /* Covers 0, 1, and unlimited */
4495 {
4496 *code = OP_END;
4497 adjust_recurse(previous, 1, utf8, cd, save_hwm);
4498 memmove(previous+1, previous, len);
4499 code++;
4500 if (repeat_max == 0)
4501 {
4502 *previous++ = OP_SKIPZERO;
4503 goto END_REPEAT;
4504 }
4505 *previous++ = OP_BRAZERO + repeat_type;
4506 }
4507
4508 /* If the maximum is greater than 1 and limited, we have to replicate
4509 in a nested fashion, sticking OP_BRAZERO before each set of brackets.
4510 The first one has to be handled carefully because it's the original
4511 copy, which has to be moved up. The remainder can be handled by code
4512 that is common with the non-zero minimum case below. We have to
4513 adjust the value or repeat_max, since one less copy is required. Once
4514 again, we may have to adjust any OP_RECURSE calls inside the group. */
4515
4516 else
4517 {
4518 int offset;
4519 *code = OP_END;
4520 adjust_recurse(previous, 2 + LINK_SIZE, utf8, cd, save_hwm);
4521 memmove(previous + 2 + LINK_SIZE, previous, len);
4522 code += 2 + LINK_SIZE;
4523 *previous++ = OP_BRAZERO + repeat_type;
4524 *previous++ = OP_BRA;
4525
4526 /* We chain together the bracket offset fields that have to be
4527 filled in later when the ends of the brackets are reached. */
4528
4529 offset = (bralink == NULL)? 0 : (int)(previous - bralink);
4530 bralink = previous;
4531 PUTINC(previous, 0, offset);
4532 }
4533
4534 repeat_max--;
4535 }
4536
4537 /* If the minimum is greater than zero, replicate the group as many
4538 times as necessary, and adjust the maximum to the number of subsequent
4539 copies that we need. If we set a first char from the group, and didn't
4540 set a required char, copy the latter from the former. If there are any
4541 forward reference subroutine calls in the group, there will be entries on
4542 the workspace list; replicate these with an appropriate increment. */
4543
4544 else
4545 {
4546 if (repeat_min > 1)
4547 {
4548 /* In the pre-compile phase, we don't actually do the replication. We
4549 just adjust the length as if we had. Do some paranoid checks for
4550 potential integer overflow. The INT64_OR_DOUBLE type is a 64-bit
4551 integer type when available, otherwise double. */
4552
4553 if (lengthptr != NULL)
4554 {
4555 int delta = (repeat_min - 1)*length_prevgroup;
4556 if ((INT64_OR_DOUBLE)(repeat_min - 1)*
4557 (INT64_OR_DOUBLE)length_prevgroup >
4558 (INT64_OR_DOUBLE)INT_MAX ||
4559 OFLOW_MAX - *lengthptr < delta)
4560 {
4561 *errorcodeptr = ERR20;
4562 goto FAILED;
4563 }
4564 *lengthptr += delta;
4565 }
4566
4567 /* This is compiling for real */
4568
4569 else
4570 {
4571 if (groupsetfirstbyte && reqbyte < 0) reqbyte = firstbyte;
4572 for (i = 1; i < repeat_min; i++)
4573 {
4574 uschar *hc;
4575 uschar *this_hwm = cd->hwm;
4576 memcpy(code, previous, len);
4577 for (hc = save_hwm; hc < this_hwm; hc += LINK_SIZE)
4578 {
4579 PUT(cd->hwm, 0, GET(hc, 0) + len);
4580 cd->hwm += LINK_SIZE;
4581 }
4582 save_hwm = this_hwm;
4583 code += len;
4584 }
4585 }
4586 }
4587
4588 if (repeat_max > 0) repeat_max -= repeat_min;
4589 }
4590
4591 /* This code is common to both the zero and non-zero minimum cases. If
4592 the maximum is limited, it replicates the group in a nested fashion,
4593 remembering the bracket starts on a stack. In the case of a zero minimum,
4594 the first one was set up above. In all cases the repeat_max now specifies
4595 the number of additional copies needed. Again, we must remember to
4596 replicate entries on the forward reference list. */
4597
4598 if (repeat_max >= 0)
4599 {
4600 /* In the pre-compile phase, we don't actually do the replication. We
4601 just adjust the length as if we had. For each repetition we must add 1
4602 to the length for BRAZERO and for all but the last repetition we must
4603 add 2 + 2*LINKSIZE to allow for the nesting that occurs. Do some
4604 paranoid checks to avoid integer overflow. The INT64_OR_DOUBLE type is
4605 a 64-bit integer type when available, otherwise double. */
4606
4607 if (lengthptr != NULL && repeat_max > 0)
4608 {
4609 int delta = repeat_max * (length_prevgroup + 1 + 2 + 2*LINK_SIZE) -
4610 2 - 2*LINK_SIZE; /* Last one doesn't nest */
4611 if ((INT64_OR_DOUBLE)repeat_max *
4612 (INT64_OR_DOUBLE)(length_prevgroup + 1 + 2 + 2*LINK_SIZE)
4613 > (INT64_OR_DOUBLE)INT_MAX ||
4614 OFLOW_MAX - *lengthptr < delta)
4615 {
4616 *errorcodeptr = ERR20;
4617 goto FAILED;
4618 }
4619 *lengthptr += delta;
4620 }
4621
4622 /* This is compiling for real */
4623
4624 else for (i = repeat_max - 1; i >= 0; i--)
4625 {
4626 uschar *hc;
4627 uschar *this_hwm = cd->hwm;
4628
4629 *code++ = OP_BRAZERO + repeat_type;
4630
4631 /* All but the final copy start a new nesting, maintaining the
4632 chain of brackets outstanding. */
4633
4634 if (i != 0)
4635 {
4636 int offset;
4637 *code++ = OP_BRA;
4638 offset = (bralink == NULL)? 0 : (int)(code - bralink);
4639 bralink = code;
4640 PUTINC(code, 0, offset);
4641 }
4642
4643 memcpy(code, previous, len);
4644 for (hc = save_hwm; hc < this_hwm; hc += LINK_SIZE)
4645 {
4646 PUT(cd->hwm, 0, GET(hc, 0) + len + ((i != 0)? 2+LINK_SIZE : 1));
4647 cd->hwm += LINK_SIZE;
4648 }
4649 save_hwm = this_hwm;
4650 code += len;
4651 }
4652
4653 /* Now chain through the pending brackets, and fill in their length
4654 fields (which are holding the chain links pro tem). */
4655
4656 while (bralink != NULL)
4657 {
4658 int oldlinkoffset;
4659 int offset = (int)(code - bralink + 1);
4660 uschar *bra = code - offset;
4661 oldlinkoffset = GET(bra, 1);
4662 bralink = (oldlinkoffset == 0)? NULL : bralink - oldlinkoffset;
4663 *code++ = OP_KET;
4664 PUTINC(code, 0, offset);
4665 PUT(bra, 1, offset);
4666 }
4667 }
4668
4669 /* If the maximum is unlimited, set a repeater in the final copy. We
4670 can't just offset backwards from the current code point, because we
4671 don't know if there's been an options resetting after the ket. The
4672 correct offset was computed above.
4673
4674 Then, when we are doing the actual compile phase, check to see whether
4675 this group is a non-atomic one that could match an empty string. If so,
4676 convert the initial operator to the S form (e.g. OP_BRA -> OP_SBRA) so
4677 that runtime checking can be done. [This check is also applied to
4678 atomic groups at runtime, but in a different way.] */
4679
4680 else
4681 {
4682 uschar *ketcode = code - ketoffset;
4683 uschar *bracode = ketcode - GET(ketcode, 1);
4684 *ketcode = OP_KETRMAX + repeat_type;
4685 if (lengthptr == NULL && *bracode != OP_ONCE)
4686 {
4687 uschar *scode = bracode;
4688 do
4689 {
4690 if (could_be_empty_branch(scode, ketcode, utf8, cd))
4691 {
4692 *bracode += OP_SBRA - OP_BRA;
4693 break;
4694 }
4695 scode += GET(scode, 1);
4696 }
4697 while (*scode == OP_ALT);
4698 }
4699 }
4700 }
4701
4702 /* If previous is OP_FAIL, it was generated by an empty class [] in
4703 JavaScript mode. The other ways in which OP_FAIL can be generated, that is
4704 by (*FAIL) or (?!) set previous to NULL, which gives a "nothing to repeat"
4705 error above. We can just ignore the repeat in JS case. */
4706
4707 else if (*previous == OP_FAIL) goto END_REPEAT;
4708
4709 /* Else there's some kind of shambles */
4710
4711 else
4712 {
4713 *errorcodeptr = ERR11;
4714 goto FAILED;
4715 }
4716
4717 /* If the character following a repeat is '+', or if certain optimization
4718 tests above succeeded, possessive_quantifier is TRUE. For some of the
4719 simpler opcodes, there is an special alternative opcode for this. For
4720 anything else, we wrap the entire repeated item inside OP_ONCE brackets.
4721 The '+' notation is just syntactic sugar, taken from Sun's Java package,
4722 but the special opcodes can optimize it a bit. The repeated item starts at
4723 tempcode, not at previous, which might be the first part of a string whose
4724 (former) last char we repeated.
4725
4726 Possessifying an 'exact' quantifier has no effect, so we can ignore it. But
4727 an 'upto' may follow. We skip over an 'exact' item, and then test the
4728 length of what remains before proceeding. */
4729
4730 if (possessive_quantifier)
4731 {
4732 int len;
4733
4734 if (*tempcode == OP_TYPEEXACT)
4735 tempcode += _pcre_OP_lengths[*tempcode] +
4736 ((tempcode[3] == OP_PROP || tempcode[3] == OP_NOTPROP)? 2 : 0);
4737
4738 else if (*tempcode == OP_EXACT || *tempcode == OP_NOTEXACT)
4739 {
4740 tempcode += _pcre_OP_lengths[*tempcode];
4741 #ifdef SUPPORT_UTF8
4742 if (utf8 && tempcode[-1] >= 0xc0)
4743 tempcode += _pcre_utf8_table4[tempcode[-1] & 0x3f];
4744 #endif
4745 }
4746
4747 len = (int)(code - tempcode);
4748 if (len > 0) switch (*tempcode)
4749 {
4750 case OP_STAR: *tempcode = OP_POSSTAR; break;
4751 case OP_PLUS: *tempcode = OP_POSPLUS; break;
4752 case OP_QUERY: *tempcode = OP_POSQUERY; break;
4753 case OP_UPTO: *tempcode = OP_POSUPTO; break;
4754
4755 case OP_TYPESTAR: *tempcode = OP_TYPEPOSSTAR; break;
4756 case OP_TYPEPLUS: *tempcode = OP_TYPEPOSPLUS; break;
4757 case OP_TYPEQUERY: *tempcode = OP_TYPEPOSQUERY; break;
4758 case OP_TYPEUPTO: *tempcode = OP_TYPEPOSUPTO; break;
4759
4760 case OP_NOTSTAR: *tempcode = OP_NOTPOSSTAR; break;
4761 case OP_NOTPLUS: *tempcode = OP_NOTPOSPLUS; break;
4762 case OP_NOTQUERY: *tempcode = OP_NOTPOSQUERY; break;
4763 case OP_NOTUPTO: *tempcode = OP_NOTPOSUPTO; break;
4764
4765 /* Because we are moving code along, we must ensure that any
4766 pending recursive references are updated. */
4767
4768 default:
4769 *code = OP_END;
4770 adjust_recurse(tempcode, 1 + LINK_SIZE, utf8, cd, save_hwm);
4771 memmove(tempcode + 1+LINK_SIZE, tempcode, len);
4772 code += 1 + LINK_SIZE;
4773 len += 1 + LINK_SIZE;
4774 tempcode[0] = OP_ONCE;
4775 *code++ = OP_KET;
4776 PUTINC(code, 0, len);
4777 PUT(tempcode, 1, len);
4778 break;
4779 }
4780 }
4781
4782 /* In all case we no longer have a previous item. We also set the
4783 "follows varying string" flag for subsequently encountered reqbytes if
4784 it isn't already set and we have just passed a varying length item. */
4785
4786 END_REPEAT:
4787 previous = NULL;
4788 cd->req_varyopt |= reqvary;
4789 break;
4790
4791
4792 /* ===================================================================*/
4793 /* Start of nested parenthesized sub-expression, or comment or lookahead or
4794 lookbehind or option setting or condition or all the other extended
4795 parenthesis forms. */
4796
4797 case CHAR_LEFT_PARENTHESIS:
4798 newoptions = options;
4799 skipbytes = 0;
4800 bravalue = OP_CBRA;
4801 save_hwm = cd->hwm;
4802 reset_bracount = FALSE;
4803
4804 /* First deal with various "verbs" that can be introduced by '*'. */
4805
4806 if (*(++ptr) == CHAR_ASTERISK &&
4807 ((cd->ctypes[ptr[1]] & ctype_letter) != 0 || ptr[1] == ':'))
4808 {
4809 int i, namelen;
4810 int arglen = 0;
4811 const char *vn = verbnames;
4812 const uschar *name = ptr + 1;
4813 const uschar *arg = NULL;
4814 previous = NULL;
4815 while ((cd->ctypes[*++ptr] & ctype_letter) != 0) {};
4816 namelen = (int)(ptr - name);
4817
4818 if (*ptr == CHAR_COLON)
4819 {
4820 arg = ++ptr;
4821 while ((cd->ctypes[*ptr] & (ctype_letter|ctype_digit)) != 0
4822 || *ptr == '_') ptr++;
4823 arglen = (int)(ptr - arg);
4824 }
4825
4826 if (*ptr != CHAR_RIGHT_PARENTHESIS)
4827 {
4828 *errorcodeptr = ERR60;
4829 goto FAILED;
4830 }
4831
4832 /* Scan the table of verb names */
4833
4834 for (i = 0; i < verbcount; i++)
4835 {
4836 if (namelen == verbs[i].len &&
4837 strncmp((char *)name, vn, namelen) == 0)
4838 {
4839 /* Check for open captures before ACCEPT */
4840
4841 if (verbs[i].op == OP_ACCEPT)
4842 {
4843 open_capitem *oc;
4844 cd->had_accept = TRUE;
4845 for (oc = cd->open_caps; oc != NULL; oc = oc->next)
4846 {
4847 *code++ = OP_CLOSE;
4848 PUT2INC(code, 0, oc->number);
4849 }
4850 }
4851
4852 /* Handle the cases with/without an argument */
4853
4854 if (arglen == 0)
4855 {
4856 if (verbs[i].op < 0) /* Argument is mandatory */
4857 {
4858 *errorcodeptr = ERR66;
4859 goto FAILED;
4860 }
4861 *code = verbs[i].op;
4862 if (*code++ == OP_THEN)
4863 {
4864 PUT(code, 0, code - bcptr->current_branch - 1);
4865 code += LINK_SIZE;
4866 }
4867 }
4868
4869 else
4870 {
4871 if (verbs[i].op_arg < 0) /* Argument is forbidden */
4872 {
4873 *errorcodeptr = ERR59;
4874 goto FAILED;
4875 }
4876 *code = verbs[i].op_arg;
4877 if (*code++ == OP_THEN_ARG)
4878 {
4879 PUT(code, 0, code - bcptr->current_branch - 1);
4880 code += LINK_SIZE;
4881 }
4882 *code++ = arglen;
4883 memcpy(code, arg, arglen);
4884 code += arglen;
4885 *code++ = 0;
4886 }
4887
4888 break; /* Found verb, exit loop */
4889 }
4890
4891 vn += verbs[i].len + 1;
4892 }
4893
4894 if (i < verbcount) continue; /* Successfully handled a verb */
4895 *errorcodeptr = ERR60; /* Verb not recognized */
4896 goto FAILED;
4897 }
4898
4899 /* Deal with the extended parentheses; all are introduced by '?', and the
4900 appearance of any of them means that this is not a capturing group. */
4901
4902 else if (*ptr == CHAR_QUESTION_MARK)
4903 {
4904 int i, set, unset, namelen;
4905 int *optset;
4906 const uschar *name;
4907 uschar *slot;
4908
4909 switch (*(++ptr))
4910 {
4911 case CHAR_NUMBER_SIGN: /* Comment; skip to ket */
4912 ptr++;
4913 while (*ptr != 0 && *ptr != CHAR_RIGHT_PARENTHESIS) ptr++;
4914 if (*ptr == 0)
4915 {
4916 *errorcodeptr = ERR18;
4917 goto FAILED;
4918 }
4919 continue;
4920
4921
4922 /* ------------------------------------------------------------ */
4923 case CHAR_VERTICAL_LINE: /* Reset capture count for each branch */
4924 reset_bracount = TRUE;
4925 /* Fall through */
4926
4927 /* ------------------------------------------------------------ */
4928 case CHAR_COLON: /* Non-capturing bracket */
4929 bravalue = OP_BRA;
4930 ptr++;
4931 break;
4932
4933
4934 /* ------------------------------------------------------------ */
4935 case CHAR_LEFT_PARENTHESIS:
4936 bravalue = OP_COND; /* Conditional group */
4937
4938 /* A condition can be an assertion, a number (referring to a numbered
4939 group), a name (referring to a named group), or 'R', referring to
4940 recursion. R<digits> and R&name are also permitted for recursion tests.
4941
4942 There are several syntaxes for testing a named group: (?(name)) is used
4943 by Python; Perl 5.10 onwards uses (?(<name>) or (?('name')).
4944
4945 There are two unfortunate ambiguities, caused by history. (a) 'R' can
4946 be the recursive thing or the name 'R' (and similarly for 'R' followed
4947 by digits), and (b) a number could be a name that consists of digits.
4948 In both cases, we look for a name first; if not found, we try the other
4949 cases. */
4950
4951 /* For conditions that are assertions, check the syntax, and then exit
4952 the switch. This will take control down to where bracketed groups,
4953 including assertions, are processed. */
4954
4955 if (ptr[1] == CHAR_QUESTION_MARK && (ptr[2] == CHAR_EQUALS_SIGN ||
4956 ptr[2] == CHAR_EXCLAMATION_MARK || ptr[2] == CHAR_LESS_THAN_SIGN))
4957 break;
4958
4959 /* Most other conditions use OP_CREF (a couple change to OP_RREF
4960 below), and all need to skip 3 bytes at the start of the group. */
4961
4962 code[1+LINK_SIZE] = OP_CREF;
4963 skipbytes = 3;
4964 refsign = -1;
4965
4966 /* Check for a test for recursion in a named group. */
4967
4968 if (ptr[1] == CHAR_R && ptr[2] == CHAR_AMPERSAND)
4969 {
4970 terminator = -1;
4971 ptr += 2;
4972 code[1+LINK_SIZE] = OP_RREF; /* Change the type of test */
4973 }
4974
4975 /* Check for a test for a named group's having been set, using the Perl
4976 syntax (?(<name>) or (?('name') */
4977
4978 else if (ptr[1] == CHAR_LESS_THAN_SIGN)
4979 {
4980 terminator = CHAR_GREATER_THAN_SIGN;
4981 ptr++;
4982 }
4983 else if (ptr[1] == CHAR_APOSTROPHE)
4984 {
4985 terminator = CHAR_APOSTROPHE;
4986 ptr++;
4987 }
4988 else
4989 {
4990 terminator = 0;
4991 if (ptr[1] == CHAR_MINUS || ptr[1] == CHAR_PLUS) refsign = *(++ptr);
4992 }
4993
4994 /* We now expect to read a name; any thing else is an error */
4995
4996 if ((cd->ctypes[ptr[1]] & ctype_word) == 0)
4997 {
4998 ptr += 1; /* To get the right offset */
4999 *errorcodeptr = ERR28;
5000 goto FAILED;
5001 }
5002
5003 /* Read the name, but also get it as a number if it's all digits */
5004
5005 recno = 0;
5006 name = ++ptr;
5007 while ((cd->ctypes[*ptr] & ctype_word) != 0)
5008 {
5009 if (recno >= 0)
5010 recno = ((digitab[*ptr] & ctype_digit) != 0)?
5011 recno * 10 + *ptr - CHAR_0 : -1;
5012 ptr++;
5013 }
5014 namelen = (int)(ptr - name);
5015
5016 if ((terminator > 0 && *ptr++ != terminator) ||
5017 *ptr++ != CHAR_RIGHT_PARENTHESIS)
5018 {
5019 ptr--; /* Error offset */
5020 *errorcodeptr = ERR26;
5021 goto FAILED;
5022 }
5023
5024 /* Do no further checking in the pre-compile phase. */
5025
5026 if (lengthptr != NULL) break;
5027
5028 /* In the real compile we do the work of looking for the actual
5029 reference. If the string started with "+" or "-" we require the rest to
5030 be digits, in which case recno will be set. */
5031
5032 if (refsign > 0)
5033 {
5034 if (recno <= 0)
5035 {
5036 *errorcodeptr = ERR58;
5037 goto FAILED;
5038 }
5039 recno = (refsign == CHAR_MINUS)?
5040 cd->bracount - recno + 1 : recno +cd->bracount;
5041 if (recno <= 0 || recno > cd->final_bracount)
5042 {
5043 *errorcodeptr = ERR15;
5044 goto FAILED;
5045 }
5046 PUT2(code, 2+LINK_SIZE, recno);
5047 break;
5048 }
5049
5050 /* Otherwise (did not start with "+" or "-"), start by looking for the
5051 name. If we find a name, add one to the opcode to change OP_CREF or
5052 OP_RREF into OP_NCREF or OP_NRREF. These behave exactly the same,
5053 except they record that the reference was originally to a name. The
5054 information is used to check duplicate names. */
5055
5056 slot = cd->name_table;
5057 for (i = 0; i < cd->names_found; i++)
5058 {
5059 if (strncmp((char *)name, (char *)slot+2, namelen) == 0) break;
5060 slot += cd->name_entry_size;
5061 }
5062
5063 /* Found a previous named subpattern */
5064
5065 if (i < cd->names_found)
5066 {
5067 recno = GET2(slot, 0);
5068 PUT2(code, 2+LINK_SIZE, recno);
5069 code[1+LINK_SIZE]++;
5070 }
5071
5072 /* Search the pattern for a forward reference */
5073
5074 else if ((i = find_parens(cd, name, namelen,
5075 (options & PCRE_EXTENDED) != 0, utf8)) > 0)
5076 {
5077 PUT2(code, 2+LINK_SIZE, i);
5078 code[1+LINK_SIZE]++;
5079 }
5080
5081 /* If terminator == 0 it means that the name followed directly after
5082 the opening parenthesis [e.g. (?(abc)...] and in this case there are
5083 some further alternatives to try. For the cases where terminator != 0
5084 [things like (?(<name>... or (?('name')... or (?(R&name)... ] we have
5085 now checked all the possibilities, so give an error. */
5086
5087 else if (terminator != 0)
5088 {
5089 *errorcodeptr = ERR15;
5090 goto FAILED;
5091 }
5092
5093 /* Check for (?(R) for recursion. Allow digits after R to specify a
5094 specific group number. */
5095
5096 else if (*name == CHAR_R)
5097 {
5098 recno = 0;
5099 for (i = 1; i < namelen; i++)
5100 {
5101 if ((digitab[name[i]] & ctype_digit) == 0)
5102 {
5103 *errorcodeptr = ERR15;
5104 goto FAILED;
5105 }
5106 recno = recno * 10 + name[i] - CHAR_0;
5107 }
5108 if (recno == 0) recno = RREF_ANY;
5109 code[1+LINK_SIZE] = OP_RREF; /* Change test type */
5110 PUT2(code, 2+LINK_SIZE, recno);
5111 }
5112
5113 /* Similarly, check for the (?(DEFINE) "condition", which is always
5114 false. */
5115
5116 else if (namelen == 6 && strncmp((char *)name, STRING_DEFINE, 6) == 0)
5117 {
5118 code[1+LINK_SIZE] = OP_DEF;
5119 skipbytes = 1;
5120 }
5121
5122 /* Check for the "name" actually being a subpattern number. We are
5123 in the second pass here, so final_bracount is set. */
5124
5125 else if (recno > 0 && recno <= cd->final_bracount)
5126 {
5127 PUT2(code, 2+LINK_SIZE, recno);
5128 }
5129
5130 /* Either an unidentified subpattern, or a reference to (?(0) */
5131
5132 else
5133 {
5134 *errorcodeptr = (recno == 0)? ERR35: ERR15;
5135 goto FAILED;
5136 }
5137 break;
5138
5139
5140 /* ------------------------------------------------------------ */
5141 case CHAR_EQUALS_SIGN: /* Positive lookahead */
5142 bravalue = OP_ASSERT;
5143 ptr++;
5144 break;
5145
5146
5147 /* ------------------------------------------------------------ */
5148 case CHAR_EXCLAMATION_MARK: /* Negative lookahead */
5149 ptr++;
5150 if (*ptr == CHAR_RIGHT_PARENTHESIS) /* Optimize (?!) */
5151 {
5152 *code++ = OP_FAIL;
5153 previous = NULL;
5154 continue;
5155 }
5156 bravalue = OP_ASSERT_NOT;
5157 break;
5158
5159
5160 /* ------------------------------------------------------------ */
5161 case CHAR_LESS_THAN_SIGN: /* Lookbehind or named define */
5162 switch (ptr[1])
5163 {
5164 case CHAR_EQUALS_SIGN: /* Positive lookbehind */
5165 bravalue = OP_ASSERTBACK;
5166 ptr += 2;
5167 break;
5168
5169 case CHAR_EXCLAMATION_MARK: /* Negative lookbehind */
5170 bravalue = OP_ASSERTBACK_NOT;
5171 ptr += 2;
5172 break;
5173
5174 default: /* Could be name define, else bad */
5175 if ((cd->ctypes[ptr[1]] & ctype_word) != 0) goto DEFINE_NAME;
5176 ptr++; /* Correct offset for error */
5177 *errorcodeptr = ERR24;
5178 goto FAILED;
5179 }
5180 break;
5181
5182
5183 /* ------------------------------------------------------------ */
5184 case CHAR_GREATER_THAN_SIGN: /* One-time brackets */
5185 bravalue = OP_ONCE;
5186 ptr++;
5187 break;
5188
5189
5190 /* ------------------------------------------------------------ */
5191 case CHAR_C: /* Callout - may be followed by digits; */
5192 previous_callout = code; /* Save for later completion */
5193 after_manual_callout = 1; /* Skip one item before completing */
5194 *code++ = OP_CALLOUT;
5195 {
5196 int n = 0;
5197 while ((digitab[*(++ptr)] & ctype_digit) != 0)
5198 n = n * 10 + *ptr - CHAR_0;
5199 if (*ptr != CHAR_RIGHT_PARENTHESIS)
5200 {
5201 *errorcodeptr = ERR39;
5202 goto FAILED;
5203 }
5204 if (n > 255)
5205 {
5206 *errorcodeptr = ERR38;
5207 goto FAILED;
5208 }
5209 *code++ = n;
5210 PUT(code, 0, (int)(ptr - cd->start_pattern + 1)); /* Pattern offset */
5211 PUT(code, LINK_SIZE, 0); /* Default length */
5212 code += 2 * LINK_SIZE;
5213 }
5214 previous = NULL;
5215 continue;
5216
5217
5218 /* ------------------------------------------------------------ */
5219 case CHAR_P: /* Python-style named subpattern handling */
5220 if (*(++ptr) == CHAR_EQUALS_SIGN ||
5221 *ptr == CHAR_GREATER_THAN_SIGN) /* Reference or recursion */
5222 {
5223 is_recurse = *ptr == CHAR_GREATER_THAN_SIGN;
5224 terminator = CHAR_RIGHT_PARENTHESIS;
5225 goto NAMED_REF_OR_RECURSE;
5226 }
5227 else if (*ptr != CHAR_LESS_THAN_SIGN) /* Test for Python-style defn */
5228 {
5229 *errorcodeptr = ERR41;
5230 goto FAILED;
5231 }
5232 /* Fall through to handle (?P< as (?< is handled */
5233
5234
5235 /* ------------------------------------------------------------ */
5236 DEFINE_NAME: /* Come here from (?< handling */
5237 case CHAR_APOSTROPHE:
5238 {
5239 terminator = (*ptr == CHAR_LESS_THAN_SIGN)?
5240 CHAR_GREATER_THAN_SIGN : CHAR_APOSTROPHE;
5241 name = ++ptr;
5242
5243 while ((cd->ctypes[*ptr] & ctype_word) != 0) ptr++;
5244 namelen = (int)(ptr - name);
5245
5246 /* In the pre-compile phase, just do a syntax check. */
5247
5248 if (lengthptr != NULL)
5249 {
5250 if (*ptr != terminator)
5251 {
5252 *errorcodeptr = ERR42;
5253 goto FAILED;
5254 }
5255 if (cd->names_found >= MAX_NAME_COUNT)
5256 {
5257 *errorcodeptr = ERR49;
5258 goto FAILED;
5259 }
5260 if (namelen + 3 > cd->name_entry_size)
5261 {
5262 cd->name_entry_size = namelen + 3;
5263 if (namelen > MAX_NAME_SIZE)
5264 {
5265 *errorcodeptr = ERR48;
5266 goto FAILED;
5267 }
5268 }
5269 }
5270
5271 /* In the real compile, create the entry in the table, maintaining
5272 alphabetical order. Duplicate names for different numbers are
5273 permitted only if PCRE_DUPNAMES is set. Duplicate names for the same
5274 number are always OK. (An existing number can be re-used if (?|
5275 appears in the pattern.) In either event, a duplicate name results in
5276 a duplicate entry in the table, even if the number is the same. This
5277 is because the number of names, and hence the table size, is computed
5278 in the pre-compile, and it affects various numbers and pointers which
5279 would all have to be modified, and the compiled code moved down, if
5280 duplicates with the same number were omitted from the table. This
5281 doesn't seem worth the hassle. However, *different* names for the
5282 same number are not permitted. */
5283
5284 else
5285 {
5286 BOOL dupname = FALSE;
5287 slot = cd->name_table;
5288
5289 for (i = 0; i < cd->names_found; i++)
5290 {
5291 int crc = memcmp(name, slot+2, namelen);
5292 if (crc == 0)
5293 {
5294 if (slot[2+namelen] == 0)
5295 {
5296 if (GET2(slot, 0) != cd->bracount + 1 &&
5297 (options & PCRE_DUPNAMES) == 0)
5298 {
5299 *errorcodeptr = ERR43;
5300 goto FAILED;
5301 }
5302 else dupname = TRUE;
5303 }
5304 else crc = -1; /* Current name is a substring */
5305 }
5306
5307 /* Make space in the table and break the loop for an earlier
5308 name. For a duplicate or later name, carry on. We do this for
5309 duplicates so that in the simple case (when ?(| is not used) they
5310 are in order of their numbers. */
5311
5312 if (crc < 0)
5313 {
5314 memmove(slot + cd->name_entry_size, slot,
5315 (cd->names_found - i) * cd->name_entry_size);
5316 break;
5317 }
5318
5319 /* Continue the loop for a later or duplicate name */
5320
5321 slot += cd->name_entry_size;
5322 }
5323
5324 /* For non-duplicate names, check for a duplicate number before
5325 adding the new name. */
5326
5327 if (!dupname)
5328 {
5329 uschar *cslot = cd->name_table;
5330 for (i = 0; i < cd->names_found; i++)
5331 {
5332 if (cslot != slot)
5333 {
5334 if (GET2(cslot, 0) == cd->bracount + 1)
5335 {
5336 *errorcodeptr = ERR65;
5337 goto FAILED;
5338 }
5339 }
5340 else i--;
5341 cslot += cd->name_entry_size;
5342 }
5343 }
5344
5345 PUT2(slot, 0, cd->bracount + 1);
5346 memcpy(slot + 2, name, namelen);
5347 slot[2+namelen] = 0;
5348 }
5349 }
5350
5351 /* In both pre-compile and compile, count the number of names we've
5352 encountered. */
5353
5354 cd->names_found++;
5355 ptr++; /* Move past > or ' */
5356 goto NUMBERED_GROUP;
5357
5358
5359 /* ------------------------------------------------------------ */
5360 case CHAR_AMPERSAND: /* Perl recursion/subroutine syntax */
5361 terminator = CHAR_RIGHT_PARENTHESIS;
5362 is_recurse = TRUE;
5363 /* Fall through */
5364
5365 /* We come here from the Python syntax above that handles both
5366 references (?P=name) and recursion (?P>name), as well as falling
5367 through from the Perl recursion syntax (?&name). We also come here from
5368 the Perl \k<name> or \k'name' back reference syntax and the \k{name}
5369 .NET syntax, and the Oniguruma \g<...> and \g'...' subroutine syntax. */
5370
5371 NAMED_REF_OR_RECURSE:
5372 name = ++ptr;
5373 while ((cd->ctypes[*ptr] & ctype_word) != 0) ptr++;
5374 namelen = (int)(ptr - name);
5375
5376 /* In the pre-compile phase, do a syntax check and set a dummy
5377 reference number. */
5378
5379 if (lengthptr != NULL)
5380 {
5381 if (namelen == 0)
5382 {
5383 *errorcodeptr = ERR62;
5384 goto FAILED;
5385 }
5386 if (*ptr != terminator)
5387 {
5388 *errorcodeptr = ERR42;
5389 goto FAILED;
5390 }
5391 if (namelen > MAX_NAME_SIZE)
5392 {
5393 *errorcodeptr = ERR48;
5394 goto FAILED;
5395 }
5396 recno = 0;
5397 }
5398
5399 /* In the real compile, seek the name in the table. We check the name
5400 first, and then check that we have reached the end of the name in the
5401 table. That way, if the name that is longer than any in the table,
5402 the comparison will fail without reading beyond the table entry. */
5403
5404 else
5405 {
5406 slot = cd->name_table;
5407 for (i = 0; i < cd->names_found; i++)
5408 {
5409 if (strncmp((char *)name, (char *)slot+2, namelen) == 0 &&
5410 slot[2+namelen] == 0)
5411 break;
5412 slot += cd->name_entry_size;
5413 }
5414
5415 if (i < cd->names_found) /* Back reference */
5416 {
5417 recno = GET2(slot, 0);
5418 }
5419 else if ((recno = /* Forward back reference */
5420 find_parens(cd, name, namelen,
5421 (options & PCRE_EXTENDED) != 0, utf8)) <= 0)
5422 {
5423 *errorcodeptr = ERR15;
5424 goto FAILED;
5425 }
5426 }
5427
5428 /* In both phases, we can now go to the code than handles numerical
5429 recursion or backreferences. */
5430
5431 if (is_recurse) goto HANDLE_RECURSION;
5432 else goto HANDLE_REFERENCE;
5433
5434
5435 /* ------------------------------------------------------------ */
5436 case CHAR_R: /* Recursion */
5437 ptr++; /* Same as (?0) */
5438 /* Fall through */
5439
5440
5441 /* ------------------------------------------------------------ */
5442 case CHAR_MINUS: case CHAR_PLUS: /* Recursion or subroutine */
5443 case CHAR_0: case CHAR_1: case CHAR_2: case CHAR_3: case CHAR_4:
5444 case CHAR_5: case CHAR_6: case CHAR_7: case CHAR_8: case CHAR_9:
5445 {
5446 const uschar *called;
5447 terminator = CHAR_RIGHT_PARENTHESIS;
5448
5449 /* Come here from the \g<...> and \g'...' code (Oniguruma
5450 compatibility). However, the syntax has been checked to ensure that
5451 the ... are a (signed) number, so that neither ERR63 nor ERR29 will
5452 be called on this path, nor with the jump to OTHER_CHAR_AFTER_QUERY
5453 ever be taken. */
5454
5455 HANDLE_NUMERICAL_RECURSION:
5456
5457 if ((refsign = *ptr) == CHAR_PLUS)
5458 {
5459 ptr++;
5460 if ((digitab[*ptr] & ctype_digit) == 0)
5461 {
5462 *errorcodeptr = ERR63;
5463 goto FAILED;
5464 }
5465 }
5466 else if (refsign == CHAR_MINUS)
5467 {
5468 if ((digitab[ptr[1]] & ctype_digit) == 0)
5469 goto OTHER_CHAR_AFTER_QUERY;
5470 ptr++;
5471 }
5472
5473 recno = 0;
5474 while((digitab[*ptr] & ctype_digit) != 0)
5475 recno = recno * 10 + *ptr++ - CHAR_0;
5476
5477 if (*ptr != terminator)
5478 {
5479 *errorcodeptr = ERR29;
5480 goto FAILED;
5481 }
5482
5483 if (refsign == CHAR_MINUS)
5484 {
5485 if (recno == 0)
5486 {
5487 *errorcodeptr = ERR58;
5488 goto FAILED;
5489 }
5490 recno = cd->bracount - recno + 1;
5491 if (recno <= 0)
5492 {
5493 *errorcodeptr = ERR15;
5494 goto FAILED;
5495 }
5496 }
5497 else if (refsign == CHAR_PLUS)
5498 {
5499 if (recno == 0)
5500 {
5501 *errorcodeptr = ERR58;
5502 goto FAILED;
5503 }
5504 recno += cd->bracount;
5505 }
5506
5507 /* Come here from code above that handles a named recursion */
5508
5509 HANDLE_RECURSION:
5510
5511 previous = code;
5512 called = cd->start_code;
5513
5514 /* When we are actually compiling, find the bracket that is being
5515 referenced. Temporarily end the regex in case it doesn't exist before
5516 this point. If we end up with a forward reference, first check that
5517 the bracket does occur later so we can give the error (and position)
5518 now. Then remember this forward reference in the workspace so it can
5519 be filled in at the end. */
5520
5521 if (lengthptr == NULL)
5522 {
5523 *code = OP_END;
5524 if (recno != 0)
5525 called = _pcre_find_bracket(cd->start_code, utf8, recno);
5526
5527 /* Forward reference */
5528
5529 if (called == NULL)
5530 {
5531 if (find_parens(cd, NULL, recno,
5532 (options & PCRE_EXTENDED) != 0, utf8) < 0)
5533 {
5534 *errorcodeptr = ERR15;
5535 goto FAILED;
5536 }
5537
5538 /* Fudge the value of "called" so that when it is inserted as an
5539 offset below, what it actually inserted is the reference number
5540 of the group. */
5541
5542 called = cd->start_code + recno;
5543 PUTINC(cd->hwm, 0, (int)(code + 2 + LINK_SIZE - cd->start_code));
5544 }
5545
5546 /* If not a forward reference, and the subpattern is still open,
5547 this is a recursive call. We check to see if this is a left
5548 recursion that could loop for ever, and diagnose that case. */
5549
5550 else if (GET(called, 1) == 0 &&
5551 could_be_empty(called, code, bcptr, utf8, cd))
5552 {
5553 *errorcodeptr = ERR40;
5554 goto FAILED;
5555 }
5556 }
5557
5558 /* Insert the recursion/subroutine item, automatically wrapped inside
5559 "once" brackets. Set up a "previous group" length so that a
5560 subsequent quantifier will work. */
5561
5562 *code = OP_ONCE;
5563 PUT(code, 1, 2 + 2*LINK_SIZE);
5564 code += 1 + LINK_SIZE;
5565
5566 *code = OP_RECURSE;
5567 PUT(code, 1, (int)(called - cd->start_code));
5568 code += 1 + LINK_SIZE;
5569
5570 *code = OP_KET;
5571 PUT(code, 1, 2 + 2*LINK_SIZE);
5572 code += 1 + LINK_SIZE;
5573
5574 length_prevgroup = 3 + 3*LINK_SIZE;
5575 }
5576
5577 /* Can't determine a first byte now */
5578
5579 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
5580 continue;
5581
5582
5583 /* ------------------------------------------------------------ */
5584 default: /* Other characters: check option setting */
5585 OTHER_CHAR_AFTER_QUERY:
5586 set = unset = 0;
5587 optset = &set;
5588
5589 while (*ptr != CHAR_RIGHT_PARENTHESIS && *ptr != CHAR_COLON)
5590 {
5591 switch (*ptr++)
5592 {
5593 case CHAR_MINUS: optset = &unset; break;
5594
5595 case CHAR_J: /* Record that it changed in the external options */
5596 *optset |= PCRE_DUPNAMES;
5597 cd->external_flags |= PCRE_JCHANGED;
5598 break;
5599
5600 case CHAR_i: *optset |= PCRE_CASELESS; break;
5601 case CHAR_m: *optset |= PCRE_MULTILINE; break;
5602 case CHAR_s: *optset |= PCRE_DOTALL; break;
5603 case CHAR_x: *optset |= PCRE_EXTENDED; break;
5604 case CHAR_U: *optset |= PCRE_UNGREEDY; break;
5605 case CHAR_X: *optset |= PCRE_EXTRA; break;
5606
5607 default: *errorcodeptr = ERR12;
5608 ptr--; /* Correct the offset */
5609 goto FAILED;
5610 }
5611 }
5612
5613 /* Set up the changed option bits, but don't change anything yet. */
5614
5615 newoptions = (options | set) & (~unset);
5616
5617 /* If the options ended with ')' this is not the start of a nested
5618 group with option changes, so the options change at this level. If this
5619 item is right at the start of the pattern, the options can be
5620 abstracted and made external in the pre-compile phase, and ignored in
5621 the compile phase. This can be helpful when matching -- for instance in
5622 caseless checking of required bytes.
5623
5624 If the code pointer is not (cd->start_code + 1 + LINK_SIZE), we are
5625 definitely *not* at the start of the pattern because something has been
5626 compiled. In the pre-compile phase, however, the code pointer can have
5627 that value after the start, because it gets reset as code is discarded
5628 during the pre-compile. However, this can happen only at top level - if
5629 we are within parentheses, the starting BRA will still be present. At
5630 any parenthesis level, the length value can be used to test if anything
5631 has been compiled at that level. Thus, a test for both these conditions
5632 is necessary to ensure we correctly detect the start of the pattern in
5633 both phases.
5634
5635 If we are not at the pattern start, compile code to change the ims
5636 options if this setting actually changes any of them, and reset the
5637 greedy defaults and the case value for firstbyte and reqbyte. */
5638
5639 if (*ptr == CHAR_RIGHT_PARENTHESIS)
5640 {
5641 if (code == cd->start_code + 1 + LINK_SIZE &&
5642 (lengthptr == NULL || *lengthptr == 2 + 2*LINK_SIZE))
5643 {
5644 cd->external_options = newoptions;
5645 }
5646 else
5647 {
5648 if ((options & PCRE_IMS) != (newoptions & PCRE_IMS))
5649 {
5650 *code++ = OP_OPT;
5651 *code++ = newoptions & PCRE_IMS;
5652 }
5653 greedy_default = ((newoptions & PCRE_UNGREEDY) != 0);
5654 greedy_non_default = greedy_default ^ 1;
5655 req_caseopt = ((newoptions & PCRE_CASELESS) != 0)? REQ_CASELESS : 0;
5656 }
5657
5658 /* Change options at this level, and pass them back for use
5659 in subsequent branches. When not at the start of the pattern, this
5660 information is also necessary so that a resetting item can be
5661 compiled at the end of a group (if we are in a group). */
5662
5663 *optionsptr = options = newoptions;
5664 previous = NULL; /* This item can't be repeated */
5665 continue; /* It is complete */
5666 }
5667
5668 /* If the options ended with ':' we are heading into a nested group
5669 with possible change of options. Such groups are non-capturing and are
5670 not assertions of any kind. All we need to do is skip over the ':';
5671 the newoptions value is handled below. */
5672
5673 bravalue = OP_BRA;
5674 ptr++;
5675 } /* End of switch for character following (? */
5676 } /* End of (? handling */
5677
5678 /* Opening parenthesis not followed by '*' or '?'. If PCRE_NO_AUTO_CAPTURE
5679 is set, all unadorned brackets become non-capturing and behave like (?:...)
5680 brackets. */
5681
5682 else if ((options & PCRE_NO_AUTO_CAPTURE) != 0)
5683 {
5684 bravalue = OP_BRA;
5685 }
5686
5687 /* Else we have a capturing group. */
5688
5689 else
5690 {
5691 NUMBERED_GROUP:
5692 cd->bracount += 1;
5693 PUT2(code, 1+LINK_SIZE, cd->bracount);
5694 skipbytes = 2;
5695 }
5696
5697 /* Process nested bracketed regex. Assertions may not be repeated, but
5698 other kinds can be. All their opcodes are >= OP_ONCE. We copy code into a
5699 non-register variable in order to be able to pass its address because some
5700 compilers complain otherwise. Pass in a new setting for the ims options if
5701 they have changed. */
5702
5703 previous = (bravalue >= OP_ONCE)? code : NULL;
5704 *code = bravalue;
5705 tempcode = code;
5706 tempreqvary = cd->req_varyopt; /* Save value before bracket */
5707 length_prevgroup = 0; /* Initialize for pre-compile phase */
5708
5709 if (!compile_regex(
5710 newoptions, /* The complete new option state */
5711 options & PCRE_IMS, /* The previous ims option state */
5712 &tempcode, /* Where to put code (updated) */
5713 &ptr, /* Input pointer (updated) */
5714 errorcodeptr, /* Where to put an error message */
5715 (bravalue == OP_ASSERTBACK ||
5716 bravalue == OP_ASSERTBACK_NOT), /* TRUE if back assert */
5717 reset_bracount, /* True if (?| group */
5718 skipbytes, /* Skip over bracket number */
5719 &subfirstbyte, /* For possible first char */
5720 &subreqbyte, /* For possible last char */
5721 bcptr, /* Current branch chain */
5722 cd, /* Tables block */
5723 (lengthptr == NULL)? NULL : /* Actual compile phase */
5724 &length_prevgroup /* Pre-compile phase */
5725 ))
5726 goto FAILED;
5727
5728 /* At the end of compiling, code is still pointing to the start of the
5729 group, while tempcode has been updated to point past the end of the group
5730 and any option resetting that may follow it. The pattern pointer (ptr)
5731 is on the bracket. */
5732
5733 /* If this is a conditional bracket, check that there are no more than
5734 two branches in the group, or just one if it's a DEFINE group. We do this
5735 in the real compile phase, not in the pre-pass, where the whole group may
5736 not be available. */
5737
5738 if (bravalue == OP_COND && lengthptr == NULL)
5739 {
5740 uschar *tc = code;
5741 int condcount = 0;
5742
5743 do {
5744 condcount++;
5745 tc += GET(tc,1);
5746 }
5747 while (*tc != OP_KET);
5748
5749 /* A DEFINE group is never obeyed inline (the "condition" is always
5750 false). It must have only one branch. */
5751
5752 if (code[LINK_SIZE+1] == OP_DEF)
5753 {
5754 if (condcount > 1)
5755 {
5756 *errorcodeptr = ERR54;
5757 goto FAILED;
5758 }
5759 bravalue = OP_DEF; /* Just a flag to suppress char handling below */
5760 }
5761
5762 /* A "normal" conditional group. If there is just one branch, we must not
5763 make use of its firstbyte or reqbyte, because this is equivalent to an
5764 empty second branch. */
5765
5766 else
5767 {
5768 if (condcount > 2)
5769 {
5770 *errorcodeptr = ERR27;
5771 goto FAILED;
5772 }
5773 if (condcount == 1) subfirstbyte = subreqbyte = REQ_NONE;
5774 }
5775 }
5776
5777 /* Error if hit end of pattern */
5778
5779 if (*ptr != CHAR_RIGHT_PARENTHESIS)
5780 {
5781 *errorcodeptr = ERR14;
5782 goto FAILED;
5783 }
5784
5785 /* In the pre-compile phase, update the length by the length of the group,
5786 less the brackets at either end. Then reduce the compiled code to just a
5787 set of non-capturing brackets so that it doesn't use much memory if it is
5788 duplicated by a quantifier.*/
5789
5790 if (lengthptr != NULL)
5791 {
5792 if (OFLOW_MAX - *lengthptr < length_prevgroup - 2 - 2*LINK_SIZE)
5793 {
5794 *errorcodeptr = ERR20;
5795 goto FAILED;
5796 }
5797 *lengthptr += length_prevgroup - 2 - 2*LINK_SIZE;
5798 *code++ = OP_BRA;
5799 PUTINC(code, 0, 1 + LINK_SIZE);
5800 *code++ = OP_KET;
5801 PUTINC(code, 0, 1 + LINK_SIZE);
5802 break; /* No need to waste time with special character handling */
5803 }
5804
5805 /* Otherwise update the main code pointer to the end of the group. */
5806
5807 code = tempcode;
5808
5809 /* For a DEFINE group, required and first character settings are not
5810 relevant. */
5811
5812 if (bravalue == OP_DEF) break;
5813
5814 /* Handle updating of the required and first characters for other types of
5815 group. Update for normal brackets of all kinds, and conditions with two
5816 branches (see code above). If the bracket is followed by a quantifier with
5817 zero repeat, we have to back off. Hence the definition of zeroreqbyte and
5818 zerofirstbyte outside the main loop so that they can be accessed for the
5819 back off. */
5820
5821 zeroreqbyte = reqbyte;
5822 zerofirstbyte = firstbyte;
5823 groupsetfirstbyte = FALSE;
5824
5825 if (bravalue >= OP_ONCE)
5826 {
5827 /* If we have not yet set a firstbyte in this branch, take it from the
5828 subpattern, remembering that it was set here so that a repeat of more
5829 than one can replicate it as reqbyte if necessary. If the subpattern has
5830 no firstbyte, set "none" for the whole branch. In both cases, a zero
5831 repeat forces firstbyte to "none". */
5832
5833 if (firstbyte == REQ_UNSET)
5834 {
5835 if (subfirstbyte >= 0)
5836 {
5837 firstbyte = subfirstbyte;
5838 groupsetfirstbyte = TRUE;
5839 }
5840 else firstbyte = REQ_NONE;
5841 zerofirstbyte = REQ_NONE;
5842 }
5843
5844 /* If firstbyte was previously set, convert the subpattern's firstbyte
5845 into reqbyte if there wasn't one, using the vary flag that was in
5846 existence beforehand. */
5847
5848 else if (subfirstbyte >= 0 && subreqbyte < 0)
5849 subreqbyte = subfirstbyte | tempreqvary;
5850
5851 /* If the subpattern set a required byte (or set a first byte that isn't
5852 really the first byte - see above), set it. */
5853
5854 if (subreqbyte >= 0) reqbyte = subreqbyte;
5855 }
5856
5857 /* For a forward assertion, we take the reqbyte, if set. This can be
5858 helpful if the pattern that follows the assertion doesn't set a different
5859 char. For example, it's useful for /(?=abcde).+/. We can't set firstbyte
5860 for an assertion, however because it leads to incorrect effect for patterns
5861 such as /(?=a)a.+/ when the "real" "a" would then become a reqbyte instead
5862 of a firstbyte. This is overcome by a scan at the end if there's no
5863 firstbyte, looking for an asserted first char. */
5864
5865 else if (bravalue == OP_ASSERT && subreqbyte >= 0) reqbyte = subreqbyte;
5866 break; /* End of processing '(' */
5867
5868
5869 /* ===================================================================*/
5870 /* Handle metasequences introduced by \. For ones like \d, the ESC_ values
5871 are arranged to be the negation of the corresponding OP_values in the
5872 default case when PCRE_UCP is not set. For the back references, the values
5873 are ESC_REF plus the reference number. Only back references and those types
5874 that consume a character may be repeated. We can test for values between
5875 ESC_b and ESC_Z for the latter; this may have to change if any new ones are
5876 ever created. */
5877
5878 case CHAR_BACKSLASH:
5879 tempptr = ptr;
5880 c = check_escape(&ptr, errorcodeptr, cd->bracount, options, FALSE);
5881 if (*errorcodeptr != 0) goto FAILED;
5882
5883 if (c < 0)
5884 {
5885 if (-c == ESC_Q) /* Handle start of quoted string */
5886 {
5887 if (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
5888 ptr += 2; /* avoid empty string */
5889 else inescq = TRUE;
5890 continue;
5891 }
5892
5893 if (-c == ESC_E) continue; /* Perl ignores an orphan \E */
5894
5895 /* For metasequences that actually match a character, we disable the
5896 setting of a first character if it hasn't already been set. */
5897
5898 if (firstbyte == REQ_UNSET && -c > ESC_b && -c < ESC_Z)
5899 firstbyte = REQ_NONE;
5900
5901 /* Set values to reset to if this is followed by a zero repeat. */
5902
5903 zerofirstbyte = firstbyte;
5904 zeroreqbyte = reqbyte;
5905
5906 /* \g<name> or \g'name' is a subroutine call by name and \g<n> or \g'n'
5907 is a subroutine call by number (Oniguruma syntax). In fact, the value
5908 -ESC_g is returned only for these cases. So we don't need to check for <
5909 or ' if the value is -ESC_g. For the Perl syntax \g{n} the value is
5910 -ESC_REF+n, and for the Perl syntax \g{name} the result is -ESC_k (as
5911 that is a synonym for a named back reference). */
5912
5913 if (-c == ESC_g)
5914 {
5915 const uschar *p;
5916 save_hwm = cd->hwm; /* Normally this is set when '(' is read */
5917 terminator = (*(++ptr) == CHAR_LESS_THAN_SIGN)?
5918 CHAR_GREATER_THAN_SIGN : CHAR_APOSTROPHE;
5919
5920 /* These two statements stop the compiler for warning about possibly
5921 unset variables caused by the jump to HANDLE_NUMERICAL_RECURSION. In
5922 fact, because we actually check for a number below, the paths that
5923 would actually be in error are never taken. */
5924
5925 skipbytes = 0;
5926 reset_bracount = FALSE;
5927
5928 /* Test for a name */
5929
5930 if (ptr[1] != CHAR_PLUS && ptr[1] != CHAR_MINUS)
5931 {
5932 BOOL isnumber = TRUE;
5933 for (p = ptr + 1; *p != 0 && *p != terminator; p++)
5934 {
5935 if ((cd->ctypes[*p] & ctype_digit) == 0) isnumber = FALSE;
5936 if ((cd->ctypes[*p] & ctype_word) == 0) break;
5937 }
5938 if (*p != terminator)
5939 {
5940 *errorcodeptr = ERR57;
5941 break;
5942 }
5943 if (isnumber)
5944 {
5945 ptr++;
5946 goto HANDLE_NUMERICAL_RECURSION;
5947 }
5948 is_recurse = TRUE;
5949 goto NAMED_REF_OR_RECURSE;
5950 }
5951
5952 /* Test a signed number in angle brackets or quotes. */
5953
5954 p = ptr + 2;
5955 while ((digitab[*p] & ctype_digit) != 0) p++;
5956 if (*p != terminator)
5957 {
5958 *errorcodeptr = ERR57;
5959 break;
5960 }
5961 ptr++;
5962 goto HANDLE_NUMERICAL_RECURSION;
5963 }
5964
5965 /* \k<name> or \k'name' is a back reference by name (Perl syntax).
5966 We also support \k{name} (.NET syntax) */
5967
5968 if (-c == ESC_k && (ptr[1] == CHAR_LESS_THAN_SIGN ||
5969 ptr[1] == CHAR_APOSTROPHE || ptr[1] == CHAR_LEFT_CURLY_BRACKET))
5970 {
5971 is_recurse = FALSE;
5972 terminator = (*(++ptr) == CHAR_LESS_THAN_SIGN)?
5973 CHAR_GREATER_THAN_SIGN : (*ptr == CHAR_APOSTROPHE)?
5974 CHAR_APOSTROPHE : CHAR_RIGHT_CURLY_BRACKET;
5975 goto NAMED_REF_OR_RECURSE;
5976 }
5977
5978 /* Back references are handled specially; must disable firstbyte if
5979 not set to cope with cases like (?=(\w+))\1: which would otherwise set
5980 ':' later. */
5981
5982 if (-c >= ESC_REF)
5983 {
5984 open_capitem *oc;
5985 recno = -c - ESC_REF;
5986
5987 HANDLE_REFERENCE: /* Come here from named backref handling */
5988 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
5989 previous = code;
5990 *code++ = OP_REF;
5991 PUT2INC(code, 0, recno);
5992 cd->backref_map |= (recno < 32)? (1 << recno) : 1;
5993 if (recno > cd->top_backref) cd->top_backref = recno;
5994
5995 /* Check to see if this back reference is recursive, that it, it
5996 is inside the group that it references. A flag is set so that the
5997 group can be made atomic. */
5998
5999 for (oc = cd->open_caps; oc != NULL; oc = oc->next)
6000 {
6001 if (oc->number == recno)
6002 {
6003 oc->flag = TRUE;
6004 break;
6005 }
6006 }
6007 }
6008
6009 /* So are Unicode property matches, if supported. */
6010
6011 #ifdef SUPPORT_UCP
6012 else if (-c == ESC_P || -c == ESC_p)
6013 {
6014 BOOL negated;
6015 int pdata;
6016 int ptype = get_ucp(&ptr, &negated, &pdata, errorcodeptr);
6017 if (ptype < 0) goto FAILED;
6018 previous = code;
6019 *code++ = ((-c == ESC_p) != negated)? OP_PROP : OP_NOTPROP;
6020 *code++ = ptype;
6021 *code++ = pdata;
6022 }
6023 #else
6024
6025 /* If Unicode properties are not supported, \X, \P, and \p are not
6026 allowed. */
6027
6028 else if (-c == ESC_X || -c == ESC_P || -c == ESC_p)
6029 {
6030 *errorcodeptr = ERR45;
6031 goto FAILED;
6032 }
6033 #endif
6034
6035 /* For the rest (including \X when Unicode properties are supported), we
6036 can obtain the OP value by negating the escape value in the default
6037 situation when PCRE_UCP is not set. When it *is* set, we substitute
6038 Unicode property tests. */
6039
6040 else
6041 {
6042 #ifdef SUPPORT_UCP
6043 if (-c >= ESC_DU && -c <= ESC_wu)
6044 {
6045 nestptr = ptr + 1; /* Where to resume */
6046 ptr = substitutes[-c - ESC_DU] - 1; /* Just before substitute */
6047 }
6048 else
6049 #endif
6050 {
6051 previous = (-c > ESC_b && -c < ESC_Z)? code : NULL;
6052 *code++ = -c;
6053 }
6054 }
6055 continue;
6056 }
6057
6058 /* We have a data character whose value is in c. In UTF-8 mode it may have
6059 a value > 127. We set its representation in the length/buffer, and then
6060 handle it as a data character. */
6061
6062 #ifdef SUPPORT_UTF8
6063 if (utf8 && c > 127)
6064 mclength = _pcre_ord2utf8(c, mcbuffer);
6065 else
6066 #endif
6067
6068 {
6069 mcbuffer[0] = c;
6070 mclength = 1;
6071 }
6072 goto ONE_CHAR;
6073
6074
6075 /* ===================================================================*/
6076 /* Handle a literal character. It is guaranteed not to be whitespace or #
6077 when the extended flag is set. If we are in UTF-8 mode, it may be a
6078 multi-byte literal character. */
6079
6080 default:
6081 NORMAL_CHAR:
6082 mclength = 1;
6083 mcbuffer[0] = c;
6084
6085 #ifdef SUPPORT_UTF8
6086 if (utf8 && c >= 0xc0)
6087 {
6088 while ((ptr[1] & 0xc0) == 0x80)
6089 mcbuffer[mclength++] = *(++ptr);
6090 }
6091 #endif
6092
6093 /* At this point we have the character's bytes in mcbuffer, and the length
6094 in mclength. When not in UTF-8 mode, the length is always 1. */
6095
6096 ONE_CHAR:
6097 previous = code;
6098 *code++ = ((options & PCRE_CASELESS) != 0)? OP_CHARNC : OP_CHAR;
6099 for (c = 0; c < mclength; c++) *code++ = mcbuffer[c];
6100
6101 /* Remember if \r or \n were seen */
6102
6103 if (mcbuffer[0] == CHAR_CR || mcbuffer[0] == CHAR_NL)
6104 cd->external_flags |= PCRE_HASCRORLF;
6105
6106 /* Set the first and required bytes appropriately. If no previous first
6107 byte, set it from this character, but revert to none on a zero repeat.
6108 Otherwise, leave the firstbyte value alone, and don't change it on a zero
6109 repeat. */
6110
6111 if (firstbyte == REQ_UNSET)
6112 {
6113 zerofirstbyte = REQ_NONE;
6114 zeroreqbyte = reqbyte;
6115
6116 /* If the character is more than one byte long, we can set firstbyte
6117 only if it is not to be matched caselessly. */
6118
6119 if (mclength == 1 || req_caseopt == 0)
6120 {
6121 firstbyte = mcbuffer[0] | req_caseopt;
6122 if (mclength != 1) reqbyte = code[-1] | cd->req_varyopt;
6123 }
6124 else firstbyte = reqbyte = REQ_NONE;
6125 }
6126
6127 /* firstbyte was previously set; we can set reqbyte only the length is
6128 1 or the matching is caseful. */
6129
6130 else
6131 {
6132 zerofirstbyte = firstbyte;
6133 zeroreqbyte = reqbyte;
6134 if (mclength == 1 || req_caseopt == 0)
6135 reqbyte = code[-1] | req_caseopt | cd->req_varyopt;
6136 }
6137
6138 break; /* End of literal character handling */
6139 }
6140 } /* end of big loop */
6141
6142
6143 /* Control never reaches here by falling through, only by a goto for all the
6144 error states. Pass back the position in the pattern so that it can be displayed
6145 to the user for diagnosing the error. */
6146
6147 FAILED:
6148 *ptrptr = ptr;
6149 return FALSE;
6150 }
6151
6152
6153
6154
6155 /*************************************************
6156 * Compile sequence of alternatives *
6157 *************************************************/
6158
6159 /* On entry, ptr is pointing past the bracket character, but on return it
6160 points to the closing bracket, or vertical bar, or end of string. The code
6161 variable is pointing at the byte into which the BRA operator has been stored.
6162 If the ims options are changed at the start (for a (?ims: group) or during any
6163 branch, we need to insert an OP_OPT item at the start of every following branch
6164 to ensure they get set correctly at run time, and also pass the new options
6165 into every subsequent branch compile.
6166
6167 This function is used during the pre-compile phase when we are trying to find
6168 out the amount of memory needed, as well as during the real compile phase. The
6169 value of lengthptr distinguishes the two phases.
6170
6171 Arguments:
6172 options option bits, including any changes for this subpattern
6173 oldims previous settings of ims option bits
6174 codeptr -> the address of the current code pointer
6175 ptrptr -> the address of the current pattern pointer
6176 errorcodeptr -> pointer to error code variable
6177 lookbehind TRUE if this is a lookbehind assertion
6178 reset_bracount TRUE to reset the count for each branch
6179 skipbytes skip this many bytes at start (for brackets and OP_COND)
6180 firstbyteptr place to put the first required character, or a negative number
6181 reqbyteptr place to put the last required character, or a negative number
6182 bcptr pointer to the chain of currently open branches
6183 cd points to the data block with tables pointers etc.
6184 lengthptr NULL during the real compile phase
6185 points to length accumulator during pre-compile phase
6186
6187 Returns: TRUE on success
6188 */
6189
6190 static BOOL
6191 compile_regex(int options, int oldims, uschar **codeptr, const uschar **ptrptr,
6192 int *errorcodeptr, BOOL lookbehind, BOOL reset_bracount, int skipbytes,
6193 int *firstbyteptr, int *reqbyteptr, branch_chain *bcptr, compile_data *cd,
6194 int *lengthptr)
6195 {
6196 const uschar *ptr = *ptrptr;
6197 uschar *code = *codeptr;
6198 uschar *last_branch = code;
6199 uschar *start_bracket = code;
6200 uschar *reverse_count = NULL;
6201 open_capitem capitem;
6202 int capnumber = 0;
6203 int firstbyte, reqbyte;
6204 int branchfirstbyte, branchreqbyte;
6205 int length;
6206 int orig_bracount;
6207 int max_bracount;
6208 int old_external_options = cd->external_options;
6209 branch_chain bc;
6210
6211 bc.outer = bcptr;
6212 bc.current_branch = code;
6213
6214 firstbyte = reqbyte = REQ_UNSET;
6215
6216 /* Accumulate the length for use in the pre-compile phase. Start with the
6217 length of the BRA and KET and any extra bytes that are required at the
6218 beginning. We accumulate in a local variable to save frequent testing of
6219 lenthptr for NULL. We cannot do this by looking at the value of code at the
6220 start and end of each alternative, because compiled items are discarded during
6221 the pre-compile phase so that the work space is not exceeded. */
6222
6223 length = 2 + 2*LINK_SIZE + skipbytes;
6224
6225 /* WARNING: If the above line is changed for any reason, you must also change
6226 the code that abstracts option settings at the start of the pattern and makes
6227 them global. It tests the value of length for (2 + 2*LINK_SIZE) in the
6228 pre-compile phase to find out whether anything has yet been compiled or not. */
6229
6230 /* If this is a capturing subpattern, add to the chain of open capturing items
6231 so that we can detect them if (*ACCEPT) is encountered. This is also used to
6232 detect groups that contain recursive back references to themselves. */
6233
6234 if (*code == OP_CBRA)
6235 {
6236 capnumber = GET2(code, 1 + LINK_SIZE);
6237 capitem.number = capnumber;
6238 capitem.next = cd->open_caps;
6239 capitem.flag = FALSE;
6240 cd->open_caps = &capitem;
6241 }
6242
6243 /* Offset is set zero to mark that this bracket is still open */
6244
6245 PUT(code, 1, 0);
6246 code += 1 + LINK_SIZE + skipbytes;
6247
6248 /* Loop for each alternative branch */
6249
6250 orig_bracount = max_bracount = cd->bracount;
6251 for (;;)
6252 {
6253 /* For a (?| group, reset the capturing bracket count so that each branch
6254 uses the same numbers. */
6255
6256 if (reset_bracount) cd->bracount = orig_bracount;
6257
6258 /* Handle a change of ims options at the start of the branch */
6259
6260 if ((options & PCRE_IMS) != oldims)
6261 {
6262 *code++ = OP_OPT;
6263 *code++ = options & PCRE_IMS;
6264 length += 2;
6265 }
6266
6267 /* Set up dummy OP_REVERSE if lookbehind assertion */
6268
6269 if (lookbehind)
6270 {
6271 *code++ = OP_REVERSE;
6272 reverse_count = code;
6273 PUTINC(code, 0, 0);
6274 length += 1 + LINK_SIZE;
6275 }
6276
6277 /* Now compile the branch; in the pre-compile phase its length gets added
6278 into the length. */
6279
6280 if (!compile_branch(&options, &code, &ptr, errorcodeptr, &branchfirstbyte,
6281 &branchreqbyte, &bc, cd, (lengthptr == NULL)? NULL : &length))
6282 {
6283 *ptrptr = ptr;
6284 return FALSE;
6285 }
6286
6287 /* If the external options have changed during this branch, it means that we
6288 are at the top level, and a leading option setting has been encountered. We
6289 need to re-set the original option values to take account of this so that,
6290 during the pre-compile phase, we know to allow for a re-set at the start of
6291 subsequent branches. */
6292
6293 if (old_external_options != cd->external_options)
6294 oldims = cd->external_options & PCRE_IMS;
6295
6296 /* Keep the highest bracket count in case (?| was used and some branch
6297 has fewer than the rest. */
6298
6299 if (cd->bracount > max_bracount) max_bracount = cd->bracount;
6300
6301 /* In the real compile phase, there is some post-processing to be done. */
6302
6303 if (lengthptr == NULL)
6304 {
6305 /* If this is the first branch, the firstbyte and reqbyte values for the
6306 branch become the values for the regex. */
6307
6308 if (*last_branch != OP_ALT)
6309 {
6310 firstbyte = branchfirstbyte;
6311 reqbyte = branchreqbyte;
6312 }
6313
6314 /* If this is not the first branch, the first char and reqbyte have to
6315 match the values from all the previous branches, except that if the
6316 previous value for reqbyte didn't have REQ_VARY set, it can still match,
6317 and we set REQ_VARY for the regex. */
6318
6319 else
6320 {
6321 /* If we previously had a firstbyte, but it doesn't match the new branch,
6322 we have to abandon the firstbyte for the regex, but if there was
6323 previously no reqbyte, it takes on the value of the old firstbyte. */
6324
6325 if (firstbyte >= 0 && firstbyte != branchfirstbyte)
6326 {
6327 if (reqbyte < 0) reqbyte = firstbyte;
6328 firstbyte = REQ_NONE;
6329 }
6330
6331 /* If we (now or from before) have no firstbyte, a firstbyte from the
6332 branch becomes a reqbyte if there isn't a branch reqbyte. */
6333
6334 if (firstbyte < 0 && branchfirstbyte >= 0 && branchreqbyte < 0)
6335 branchreqbyte = branchfirstbyte;
6336
6337 /* Now ensure that the reqbytes match */
6338
6339 if ((reqbyte & ~REQ_VARY) != (branchreqbyte & ~REQ_VARY))
6340 reqbyte = REQ_NONE;
6341 else reqbyte |= branchreqbyte; /* To "or" REQ_VARY */
6342 }
6343
6344 /* If lookbehind, check that this branch matches a fixed-length string, and
6345 put the length into the OP_REVERSE item. Temporarily mark the end of the
6346 branch with OP_END. If the branch contains OP_RECURSE, the result is -3
6347 because there may be forward references that we can't check here. Set a
6348 flag to cause another lookbehind check at the end. Why not do it all at the
6349 end? Because common, erroneous checks are picked up here and the offset of
6350 the problem can be shown. */
6351
6352 if (lookbehind)
6353 {
6354 int fixed_length;
6355 *code = OP_END;
6356 fixed_length = find_fixedlength(last_branch, options, FALSE, cd);
6357 DPRINTF(("fixed length = %d\n", fixed_length));
6358 if (fixed_length == -3)
6359 {
6360 cd->check_lookbehind = TRUE;
6361 }
6362 else if (fixed_length < 0)
6363 {
6364 *errorcodeptr = (fixed_length == -2)? ERR36 : ERR25;
6365 *ptrptr = ptr;
6366 return FALSE;
6367 }
6368 else { PUT(reverse_count, 0, fixed_length); }
6369 }
6370 }
6371
6372 /* Reached end of expression, either ')' or end of pattern. In the real
6373 compile phase, go back through the alternative branches and reverse the chain
6374 of offsets, with the field in the BRA item now becoming an offset to the
6375 first alternative. If there are no alternatives, it points to the end of the
6376 group. The length in the terminating ket is always the length of the whole
6377 bracketed item. If any of the ims options were changed inside the group,
6378 compile a resetting op-code following, except at the very end of the pattern.
6379 Return leaving the pointer at the terminating char. */
6380
6381 if (*ptr != CHAR_VERTICAL_LINE)
6382 {
6383 if (lengthptr == NULL)
6384 {
6385 int branch_length = (int)(code - last_branch);
6386 do
6387 {
6388 int prev_length = GET(last_branch, 1);
6389 PUT(last_branch, 1, branch_length);
6390 branch_length = prev_length;
6391 last_branch -= branch_length;
6392 }
6393 while (branch_length > 0);
6394 }
6395
6396 /* Fill in the ket */
6397
6398 *code = OP_KET;
6399 PUT(code, 1, (int)(code - start_bracket));
6400 code += 1 + LINK_SIZE;
6401
6402 /* If it was a capturing subpattern, check to see if it contained any
6403 recursive back references. If so, we must wrap it in atomic brackets.
6404 In any event, remove the block from the chain. */
6405
6406 if (capnumber > 0)
6407 {
6408 if (cd->open_caps->flag)
6409 {
6410 memmove(start_bracket + 1 + LINK_SIZE, start_bracket,
6411 code - start_bracket);
6412 *start_bracket = OP_ONCE;
6413 code += 1 + LINK_SIZE;
6414 PUT(start_bracket, 1, (int)(code - start_bracket));
6415 *code = OP_KET;
6416 PUT(code, 1, (int)(code - start_bracket));
6417 code += 1 + LINK_SIZE;
6418 length += 2 + 2*LINK_SIZE;
6419 }
6420 cd->open_caps = cd->open_caps->next;
6421 }
6422
6423 /* Reset options if needed. */
6424
6425 if ((options & PCRE_IMS) != oldims && *ptr == CHAR_RIGHT_PARENTHESIS)
6426 {
6427 *code++ = OP_OPT;
6428 *code++ = oldims;
6429 length += 2;
6430 }
6431
6432 /* Retain the highest bracket number, in case resetting was used. */
6433
6434 cd->bracount = max_bracount;
6435
6436 /* Set values to pass back */
6437
6438 *codeptr = code;
6439 *ptrptr = ptr;
6440 *firstbyteptr = firstbyte;
6441 *reqbyteptr = reqbyte;
6442 if (lengthptr != NULL)
6443 {
6444 if (OFLOW_MAX - *lengthptr < length)
6445 {
6446 *errorcodeptr = ERR20;
6447 return FALSE;
6448 }
6449 *lengthptr += length;
6450 }
6451 return TRUE;
6452 }
6453
6454 /* Another branch follows. In the pre-compile phase, we can move the code
6455 pointer back to where it was for the start of the first branch. (That is,
6456 pretend that each branch is the only one.)
6457
6458 In the real compile phase, insert an ALT node. Its length field points back
6459 to the previous branch while the bracket remains open. At the end the chain
6460 is reversed. It's done like this so that the start of the bracket has a
6461 zero offset until it is closed, making it possible to detect recursion. */
6462
6463 if (lengthptr != NULL)
6464 {
6465 code = *codeptr + 1 + LINK_SIZE + skipbytes;
6466 length += 1 + LINK_SIZE;
6467 }
6468 else
6469 {
6470 *code = OP_ALT;
6471 PUT(code, 1, (int)(code - last_branch));
6472 bc.current_branch = last_branch = code;
6473 code += 1 + LINK_SIZE;
6474 }
6475
6476 ptr++;
6477 }
6478 /* Control never reaches here */
6479 }
6480
6481
6482
6483
6484 /*************************************************
6485 * Check for anchored expression *
6486 *************************************************/
6487
6488 /* Try to find out if this is an anchored regular expression. Consider each
6489 alternative branch. If they all start with OP_SOD or OP_CIRC, or with a bracket
6490 all of whose alternatives start with OP_SOD or OP_CIRC (recurse ad lib), then
6491 it's anchored. However, if this is a multiline pattern, then only OP_SOD
6492 counts, since OP_CIRC can match in the middle.
6493
6494 We can also consider a regex to be anchored if OP_SOM starts all its branches.
6495 This is the code for \G, which means "match at start of match position, taking
6496 into account the match offset".
6497
6498 A branch is also implicitly anchored if it starts with .* and DOTALL is set,
6499 because that will try the rest of the pattern at all possible matching points,
6500 so there is no point trying again.... er ....
6501
6502 .... except when the .* appears inside capturing parentheses, and there is a
6503 subsequent back reference to those parentheses. We haven't enough information
6504 to catch that case precisely.
6505
6506 At first, the best we could do was to detect when .* was in capturing brackets
6507 and the highest back reference was greater than or equal to that level.
6508 However, by keeping a bitmap of the first 31 back references, we can catch some
6509 of the more common cases more precisely.
6510
6511 Arguments:
6512 code points to start of expression (the bracket)
6513 options points to the options setting
6514 bracket_map a bitmap of which brackets we are inside while testing; this
6515 handles up to substring 31; after that we just have to take
6516 the less precise approach
6517 backref_map the back reference bitmap
6518
6519 Returns: TRUE or FALSE
6520 */
6521
6522 static BOOL
6523 is_anchored(register const uschar *code, int *options, unsigned int bracket_map,
6524 unsigned int backref_map)
6525 {
6526 do {
6527 const uschar *scode = first_significant_code(code + _pcre_OP_lengths[*code],
6528 options, PCRE_MULTILINE, FALSE);
6529 register int op = *scode;
6530
6531 /* Non-capturing brackets */
6532
6533 if (op == OP_BRA)
6534 {
6535 if (!is_anchored(scode, options, bracket_map, backref_map)) return FALSE;
6536 }
6537
6538 /* Capturing brackets */
6539
6540 else if (op == OP_CBRA)
6541 {
6542 int n = GET2(scode, 1+LINK_SIZE);
6543 int new_map = bracket_map | ((n < 32)? (1 << n) : 1);
6544 if (!is_anchored(scode, options, new_map, backref_map)) return FALSE;
6545 }
6546
6547 /* Other brackets */
6548
6549 else if (op == OP_ASSERT || op == OP_ONCE || op == OP_COND)
6550 {
6551 if (!is_anchored(scode, options, bracket_map, backref_map)) return FALSE;
6552 }
6553
6554 /* .* is not anchored unless DOTALL is set (which generates OP_ALLANY) and
6555 it isn't in brackets that are or may be referenced. */
6556
6557 else if ((op == OP_TYPESTAR || op == OP_TYPEMINSTAR ||
6558 op == OP_TYPEPOSSTAR))
6559 {
6560 if (scode[1] != OP_ALLANY || (bracket_map & backref_map) != 0)
6561 return FALSE;
6562 }
6563
6564 /* Check for explicit anchoring */
6565
6566 else if (op != OP_SOD && op != OP_SOM &&
6567 ((*options & PCRE_MULTILINE) != 0 || op != OP_CIRC))
6568 return FALSE;
6569 code += GET(code, 1);
6570 }
6571 while (*code == OP_ALT); /* Loop for each alternative */
6572 return TRUE;
6573 }
6574
6575
6576
6577 /*************************************************
6578 * Check for starting with ^ or .* *
6579 *************************************************/
6580
6581 /* This is called to find out if every branch starts with ^ or .* so that
6582 "first char" processing can be done to speed things up in multiline
6583 matching and for non-DOTALL patterns that start with .* (which must start at
6584 the beginning or after \n). As in the case of is_anchored() (see above), we
6585 have to take account of back references to capturing brackets that contain .*
6586 because in that case we can't make the assumption.
6587
6588 Arguments:
6589 code points to start of expression (the bracket)
6590 bracket_map a bitmap of which brackets we are inside while testing; this
6591 handles up to substring 31; after that we just have to take
6592 the less precise approach
6593 backref_map the back reference bitmap
6594
6595 Returns: TRUE or FALSE
6596 */
6597
6598 static BOOL
6599 is_startline(const uschar *code, unsigned int bracket_map,
6600 unsigned int backref_map)
6601 {
6602 do {
6603 const uschar *scode = first_significant_code(code + _pcre_OP_lengths[*code],
6604 NULL, 0, FALSE);
6605 register int op = *scode;
6606
6607 /* If we are at the start of a conditional assertion group, *both* the
6608 conditional assertion *and* what follows the condition must satisfy the test
6609 for start of line. Other kinds of condition fail. Note that there may be an
6610 auto-callout at the start of a condition. */
6611
6612 if (op == OP_COND)
6613 {
6614 scode += 1 + LINK_SIZE;
6615 if (*scode == OP_CALLOUT) scode += _pcre_OP_lengths[OP_CALLOUT];
6616 switch (*scode)
6617 {
6618 case OP_CREF:
6619 case OP_NCREF:
6620 case OP_RREF:
6621 case OP_NRREF:
6622 case OP_DEF:
6623 return FALSE;
6624
6625 default: /* Assertion */
6626 if (!is_startline(scode, bracket_map, backref_map)) return FALSE;
6627 do scode += GET(scode, 1); while (*scode == OP_ALT);
6628 scode += 1 + LINK_SIZE;
6629 break;
6630 }
6631 scode = first_significant_code(scode, NULL, 0, FALSE);
6632 op = *scode;
6633 }
6634
6635 /* Non-capturing brackets */
6636
6637 if (op == OP_BRA)
6638 {
6639 if (!is_startline(scode, bracket_map, backref_map)) return FALSE;
6640 }
6641
6642 /* Capturing brackets */
6643
6644 else if (op == OP_CBRA)
6645 {
6646 int n = GET2(scode, 1+LINK_SIZE);
6647 int new_map = bracket_map | ((n < 32)? (1 << n) : 1);
6648 if (!is_startline(scode, new_map, backref_map)) return FALSE;
6649 }
6650
6651 /* Other brackets */
6652
6653 else if (op == OP_ASSERT || op == OP_ONCE)
6654 {
6655 if (!is_startline(scode, bracket_map, backref_map)) return FALSE;
6656 }
6657
6658 /* .* means "start at start or after \n" if it isn't in brackets that
6659 may be referenced. */
6660
6661 else if (op == OP_TYPESTAR || op == OP_TYPEMINSTAR || op == OP_TYPEPOSSTAR)
6662 {
6663 if (scode[1] != OP_ANY || (bracket_map & backref_map) != 0) return FALSE;
6664 }
6665
6666 /* Check for explicit circumflex */
6667