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

Contents of /code/trunk/pcre_compile.c

Parent Directory Parent Directory | Revision Log Revision Log


Revision 579 - (show annotations) (download)
Wed Nov 24 17:39:25 2010 UTC (3 years, 9 months ago) by ph10
File MIME type: text/plain
File size: 236901 byte(s)
Tidies of documenation and code while preparing for release.

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