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

Contents of /code/trunk/pcre_compile.c

Parent Directory Parent Directory | Revision Log Revision Log


Revision 621 - (show annotations)
Mon Jul 18 10:14:09 2011 UTC (8 years, 4 months ago) by ph10
File MIME type: text/plain
File size: 240237 byte(s)
Remove atomic from single repeats; convert possessive atomic to possessive 
non-atomic (because they are the same).
1 /*************************************************
2 * Perl-Compatible Regular Expressions *
3 *************************************************/
4
5 /* PCRE is a library of functions to support regular expressions whose syntax
6 and semantics are as close as possible to those of the Perl 5 language.
7
8 Written by Philip Hazel
9 Copyright (c) 1997-2011 University of Cambridge
10
11 -----------------------------------------------------------------------------
12 Redistribution and use in source and binary forms, with or without
13 modification, are permitted provided that the following conditions are met:
14
15 * Redistributions of source code must retain the above copyright notice,
16 this list of conditions and the following disclaimer.
17
18 * Redistributions in binary form must reproduce the above copyright
19 notice, this list of conditions and the following disclaimer in the
20 documentation and/or other materials provided with the distribution.
21
22 * Neither the name of the University of Cambridge nor the names of its
23 contributors may be used to endorse or promote products derived from
24 this software without specific prior written permission.
25
26 THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
27 AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28 IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29 ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE
30 LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
31 CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
32 SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
33 INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
34 CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
35 ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
36 POSSIBILITY OF SUCH DAMAGE.
37 -----------------------------------------------------------------------------
38 */
39
40
41 /* This module contains the external function pcre_compile(), along with
42 supporting internal functions that are not used by other modules. */
43
44
45 #ifdef HAVE_CONFIG_H
46 #include "config.h"
47 #endif
48
49 #define NLBLOCK cd /* Block containing newline information */
50 #define PSSTART start_pattern /* Field containing processed string start */
51 #define PSEND end_pattern /* Field containing processed string end */
52
53 #include "pcre_internal.h"
54
55
56 /* When PCRE_DEBUG is defined, we need the pcre_printint() function, which is
57 also used by pcretest. PCRE_DEBUG is not defined when building a production
58 library. */
59
60 #ifdef PCRE_DEBUG
61 #include "pcre_printint.src"
62 #endif
63
64
65 /* Macro for setting individual bits in class bitmaps. */
66
67 #define SETBIT(a,b) a[b/8] |= (1 << (b%8))
68
69 /* Maximum length value to check against when making sure that the integer that
70 holds the compiled pattern length does not overflow. We make it a bit less than
71 INT_MAX to allow for adding in group terminating bytes, so that we don't have
72 to check them every time. */
73
74 #define OFLOW_MAX (INT_MAX - 20)
75
76
77 /*************************************************
78 * Code parameters and static tables *
79 *************************************************/
80
81 /* This value specifies the size of stack workspace that is used during the
82 first pre-compile phase that determines how much memory is required. The regex
83 is partly compiled into this space, but the compiled parts are discarded as
84 soon as they can be, so that hopefully there will never be an overrun. The code
85 does, however, check for an overrun. The largest amount I've seen used is 218,
86 so this number is very generous.
87
88 The same workspace is used during the second, actual compile phase for
89 remembering forward references to groups so that they can be filled in at the
90 end. Each entry in this list occupies LINK_SIZE bytes, so even when LINK_SIZE
91 is 4 there is plenty of room. */
92
93 #define COMPILE_WORK_SIZE (4096)
94
95 /* The overrun tests check for a slightly smaller size so that they detect the
96 overrun before it actually does run off the end of the data block. */
97
98 #define WORK_SIZE_CHECK (COMPILE_WORK_SIZE - 100)
99
100
101 /* Table for handling escaped characters in the range '0'-'z'. Positive returns
102 are simple data values; negative values are for special things like \d and so
103 on. Zero means further processing is needed (for things like \x), or the escape
104 is invalid. */
105
106 #ifndef EBCDIC
107
108 /* This is the "normal" table for ASCII systems or for EBCDIC systems running
109 in UTF-8 mode. */
110
111 static const short int escapes[] = {
112 0, 0,
113 0, 0,
114 0, 0,
115 0, 0,
116 0, 0,
117 CHAR_COLON, CHAR_SEMICOLON,
118 CHAR_LESS_THAN_SIGN, CHAR_EQUALS_SIGN,
119 CHAR_GREATER_THAN_SIGN, CHAR_QUESTION_MARK,
120 CHAR_COMMERCIAL_AT, -ESC_A,
121 -ESC_B, -ESC_C,
122 -ESC_D, -ESC_E,
123 0, -ESC_G,
124 -ESC_H, 0,
125 0, -ESC_K,
126 0, 0,
127 -ESC_N, 0,
128 -ESC_P, -ESC_Q,
129 -ESC_R, -ESC_S,
130 0, 0,
131 -ESC_V, -ESC_W,
132 -ESC_X, 0,
133 -ESC_Z, CHAR_LEFT_SQUARE_BRACKET,
134 CHAR_BACKSLASH, CHAR_RIGHT_SQUARE_BRACKET,
135 CHAR_CIRCUMFLEX_ACCENT, CHAR_UNDERSCORE,
136 CHAR_GRAVE_ACCENT, 7,
137 -ESC_b, 0,
138 -ESC_d, ESC_e,
139 ESC_f, 0,
140 -ESC_h, 0,
141 0, -ESC_k,
142 0, 0,
143 ESC_n, 0,
144 -ESC_p, 0,
145 ESC_r, -ESC_s,
146 ESC_tee, 0,
147 -ESC_v, -ESC_w,
148 0, 0,
149 -ESC_z
150 };
151
152 #else
153
154 /* This is the "abnormal" table for EBCDIC systems without UTF-8 support. */
155
156 static const short int escapes[] = {
157 /* 48 */ 0, 0, 0, '.', '<', '(', '+', '|',
158 /* 50 */ '&', 0, 0, 0, 0, 0, 0, 0,
159 /* 58 */ 0, 0, '!', '$', '*', ')', ';', '~',
160 /* 60 */ '-', '/', 0, 0, 0, 0, 0, 0,
161 /* 68 */ 0, 0, '|', ',', '%', '_', '>', '?',
162 /* 70 */ 0, 0, 0, 0, 0, 0, 0, 0,
163 /* 78 */ 0, '`', ':', '#', '@', '\'', '=', '"',
164 /* 80 */ 0, 7, -ESC_b, 0, -ESC_d, ESC_e, ESC_f, 0,
165 /* 88 */-ESC_h, 0, 0, '{', 0, 0, 0, 0,
166 /* 90 */ 0, 0, -ESC_k, 'l', 0, ESC_n, 0, -ESC_p,
167 /* 98 */ 0, ESC_r, 0, '}', 0, 0, 0, 0,
168 /* A0 */ 0, '~', -ESC_s, ESC_tee, 0,-ESC_v, -ESC_w, 0,
169 /* A8 */ 0,-ESC_z, 0, 0, 0, '[', 0, 0,
170 /* B0 */ 0, 0, 0, 0, 0, 0, 0, 0,
171 /* B8 */ 0, 0, 0, 0, 0, ']', '=', '-',
172 /* C0 */ '{',-ESC_A, -ESC_B, -ESC_C, -ESC_D,-ESC_E, 0, -ESC_G,
173 /* C8 */-ESC_H, 0, 0, 0, 0, 0, 0, 0,
174 /* D0 */ '}', 0, -ESC_K, 0, 0,-ESC_N, 0, -ESC_P,
175 /* D8 */-ESC_Q,-ESC_R, 0, 0, 0, 0, 0, 0,
176 /* E0 */ '\\', 0, -ESC_S, 0, 0,-ESC_V, -ESC_W, -ESC_X,
177 /* E8 */ 0,-ESC_Z, 0, 0, 0, 0, 0, 0,
178 /* F0 */ 0, 0, 0, 0, 0, 0, 0, 0,
179 /* F8 */ 0, 0, 0, 0, 0, 0, 0, 0
180 };
181 #endif
182
183
184 /* Table of special "verbs" like (*PRUNE). This is a short table, so it is
185 searched linearly. Put all the names into a single string, in order to reduce
186 the number of relocations when a shared library is dynamically linked. The
187 string is built from string macros so that it works in UTF-8 mode on EBCDIC
188 platforms. */
189
190 typedef struct verbitem {
191 int len; /* Length of verb name */
192 int op; /* Op when no arg, or -1 if arg mandatory */
193 int op_arg; /* Op when arg present, or -1 if not allowed */
194 } verbitem;
195
196 static const char verbnames[] =
197 "\0" /* Empty name is a shorthand for MARK */
198 STRING_MARK0
199 STRING_ACCEPT0
200 STRING_COMMIT0
201 STRING_F0
202 STRING_FAIL0
203 STRING_PRUNE0
204 STRING_SKIP0
205 STRING_THEN;
206
207 static const verbitem verbs[] = {
208 { 0, -1, OP_MARK },
209 { 4, -1, OP_MARK },
210 { 6, OP_ACCEPT, -1 },
211 { 6, OP_COMMIT, -1 },
212 { 1, OP_FAIL, -1 },
213 { 4, OP_FAIL, -1 },
214 { 5, OP_PRUNE, OP_PRUNE_ARG },
215 { 4, OP_SKIP, OP_SKIP_ARG },
216 { 4, OP_THEN, OP_THEN_ARG }
217 };
218
219 static const int verbcount = sizeof(verbs)/sizeof(verbitem);
220
221
222 /* Tables of names of POSIX character classes and their lengths. The names are
223 now all in a single string, to reduce the number of relocations when a shared
224 library is dynamically loaded. The list of lengths is terminated by a zero
225 length entry. The first three must be alpha, lower, upper, as this is assumed
226 for handling case independence. */
227
228 static const char posix_names[] =
229 STRING_alpha0 STRING_lower0 STRING_upper0 STRING_alnum0
230 STRING_ascii0 STRING_blank0 STRING_cntrl0 STRING_digit0
231 STRING_graph0 STRING_print0 STRING_punct0 STRING_space0
232 STRING_word0 STRING_xdigit;
233
234 static const uschar posix_name_lengths[] = {
235 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 4, 6, 0 };
236
237 /* Table of class bit maps for each POSIX class. Each class is formed from a
238 base map, with an optional addition or removal of another map. Then, for some
239 classes, there is some additional tweaking: for [:blank:] the vertical space
240 characters are removed, and for [:alpha:] and [:alnum:] the underscore
241 character is removed. The triples in the table consist of the base map offset,
242 second map offset or -1 if no second map, and a non-negative value for map
243 addition or a negative value for map subtraction (if there are two maps). The
244 absolute value of the third field has these meanings: 0 => no tweaking, 1 =>
245 remove vertical space characters, 2 => remove underscore. */
246
247 static const int posix_class_maps[] = {
248 cbit_word, cbit_digit, -2, /* alpha */
249 cbit_lower, -1, 0, /* lower */
250 cbit_upper, -1, 0, /* upper */
251 cbit_word, -1, 2, /* alnum - word without underscore */
252 cbit_print, cbit_cntrl, 0, /* ascii */
253 cbit_space, -1, 1, /* blank - a GNU extension */
254 cbit_cntrl, -1, 0, /* cntrl */
255 cbit_digit, -1, 0, /* digit */
256 cbit_graph, -1, 0, /* graph */
257 cbit_print, -1, 0, /* print */
258 cbit_punct, -1, 0, /* punct */
259 cbit_space, -1, 0, /* space */
260 cbit_word, -1, 0, /* word - a Perl extension */
261 cbit_xdigit,-1, 0 /* xdigit */
262 };
263
264 /* Table of substitutes for \d etc when PCRE_UCP is set. The POSIX class
265 substitutes must be in the order of the names, defined above, and there are
266 both positive and negative cases. NULL means no substitute. */
267
268 #ifdef SUPPORT_UCP
269 static const uschar *substitutes[] = {
270 (uschar *)"\\P{Nd}", /* \D */
271 (uschar *)"\\p{Nd}", /* \d */
272 (uschar *)"\\P{Xsp}", /* \S */ /* NOTE: Xsp is Perl space */
273 (uschar *)"\\p{Xsp}", /* \s */
274 (uschar *)"\\P{Xwd}", /* \W */
275 (uschar *)"\\p{Xwd}" /* \w */
276 };
277
278 static const uschar *posix_substitutes[] = {
279 (uschar *)"\\p{L}", /* alpha */
280 (uschar *)"\\p{Ll}", /* lower */
281 (uschar *)"\\p{Lu}", /* upper */
282 (uschar *)"\\p{Xan}", /* alnum */
283 NULL, /* ascii */
284 (uschar *)"\\h", /* blank */
285 NULL, /* cntrl */
286 (uschar *)"\\p{Nd}", /* digit */
287 NULL, /* graph */
288 NULL, /* print */
289 NULL, /* punct */
290 (uschar *)"\\p{Xps}", /* space */ /* NOTE: Xps is POSIX space */
291 (uschar *)"\\p{Xwd}", /* word */
292 NULL, /* xdigit */
293 /* Negated cases */
294 (uschar *)"\\P{L}", /* ^alpha */
295 (uschar *)"\\P{Ll}", /* ^lower */
296 (uschar *)"\\P{Lu}", /* ^upper */
297 (uschar *)"\\P{Xan}", /* ^alnum */
298 NULL, /* ^ascii */
299 (uschar *)"\\H", /* ^blank */
300 NULL, /* ^cntrl */
301 (uschar *)"\\P{Nd}", /* ^digit */
302 NULL, /* ^graph */
303 NULL, /* ^print */
304 NULL, /* ^punct */
305 (uschar *)"\\P{Xps}", /* ^space */ /* NOTE: Xps is POSIX space */
306 (uschar *)"\\P{Xwd}", /* ^word */
307 NULL /* ^xdigit */
308 };
309 #define POSIX_SUBSIZE (sizeof(posix_substitutes)/sizeof(uschar *))
310 #endif
311
312 #define STRING(a) # a
313 #define XSTRING(s) STRING(s)
314
315 /* The texts of compile-time error messages. These are "char *" because they
316 are passed to the outside world. Do not ever re-use any error number, because
317 they are documented. Always add a new error instead. Messages marked DEAD below
318 are no longer used. This used to be a table of strings, but in order to reduce
319 the number of relocations needed when a shared library is loaded dynamically,
320 it is now one long string. We cannot use a table of offsets, because the
321 lengths of inserts such as XSTRING(MAX_NAME_SIZE) are not known. Instead, we
322 simply count through to the one we want - this isn't a performance issue
323 because these strings are used only when there is a compilation error.
324
325 Each substring ends with \0 to insert a null character. This includes the final
326 substring, so that the whole string ends with \0\0, which can be detected when
327 counting through. */
328
329 static const char error_texts[] =
330 "no error\0"
331 "\\ at end of pattern\0"
332 "\\c at end of pattern\0"
333 "unrecognized character follows \\\0"
334 "numbers out of order in {} quantifier\0"
335 /* 5 */
336 "number too big in {} quantifier\0"
337 "missing terminating ] for character class\0"
338 "invalid escape sequence in character class\0"
339 "range out of order in character class\0"
340 "nothing to repeat\0"
341 /* 10 */
342 "operand of unlimited repeat could match the empty string\0" /** DEAD **/
343 "internal error: unexpected repeat\0"
344 "unrecognized character after (? or (?-\0"
345 "POSIX named classes are supported only within a class\0"
346 "missing )\0"
347 /* 15 */
348 "reference to non-existent subpattern\0"
349 "erroffset passed as NULL\0"
350 "unknown option bit(s) set\0"
351 "missing ) after comment\0"
352 "parentheses nested too deeply\0" /** DEAD **/
353 /* 20 */
354 "regular expression is too large\0"
355 "failed to get memory\0"
356 "unmatched parentheses\0"
357 "internal error: code overflow\0"
358 "unrecognized character after (?<\0"
359 /* 25 */
360 "lookbehind assertion is not fixed length\0"
361 "malformed number or name after (?(\0"
362 "conditional group contains more than two branches\0"
363 "assertion expected after (?(\0"
364 "(?R or (?[+-]digits must be followed by )\0"
365 /* 30 */
366 "unknown POSIX class name\0"
367 "POSIX collating elements are not supported\0"
368 "this version of PCRE is not compiled with PCRE_UTF8 support\0"
369 "spare error\0" /** DEAD **/
370 "character value in \\x{...} sequence is too large\0"
371 /* 35 */
372 "invalid condition (?(0)\0"
373 "\\C not allowed in lookbehind assertion\0"
374 "PCRE does not support \\L, \\l, \\N{name}, \\U, or \\u\0"
375 "number after (?C is > 255\0"
376 "closing ) for (?C expected\0"
377 /* 40 */
378 "recursive call could loop indefinitely\0"
379 "unrecognized character after (?P\0"
380 "syntax error in subpattern name (missing terminator)\0"
381 "two named subpatterns have the same name\0"
382 "invalid UTF-8 string\0"
383 /* 45 */
384 "support for \\P, \\p, and \\X has not been compiled\0"
385 "malformed \\P or \\p sequence\0"
386 "unknown property name after \\P or \\p\0"
387 "subpattern name is too long (maximum " XSTRING(MAX_NAME_SIZE) " characters)\0"
388 "too many named subpatterns (maximum " XSTRING(MAX_NAME_COUNT) ")\0"
389 /* 50 */
390 "repeated subpattern is too long\0" /** DEAD **/
391 "octal value is greater than \\377 (not in UTF-8 mode)\0"
392 "internal error: overran compiling workspace\0"
393 "internal error: previously-checked referenced subpattern not found\0"
394 "DEFINE group contains more than one branch\0"
395 /* 55 */
396 "repeating a DEFINE group is not allowed\0"
397 "inconsistent NEWLINE options\0"
398 "\\g is not followed by a braced, angle-bracketed, or quoted name/number or by a plain number\0"
399 "a numbered reference must not be zero\0"
400 "an argument is not allowed for (*ACCEPT), (*FAIL), or (*COMMIT)\0"
401 /* 60 */
402 "(*VERB) not recognized\0"
403 "number is too big\0"
404 "subpattern name expected\0"
405 "digit expected after (?+\0"
406 "] is an invalid data character in JavaScript compatibility mode\0"
407 /* 65 */
408 "different names for subpatterns of the same number are not allowed\0"
409 "(*MARK) must have an argument\0"
410 "this version of PCRE is not compiled with PCRE_UCP support\0"
411 "\\c must be followed by an ASCII character\0"
412 ;
413
414 /* Table to identify digits and hex digits. This is used when compiling
415 patterns. Note that the tables in chartables are dependent on the locale, and
416 may mark arbitrary characters as digits - but the PCRE compiling code expects
417 to handle only 0-9, a-z, and A-Z as digits when compiling. That is why we have
418 a private table here. It costs 256 bytes, but it is a lot faster than doing
419 character value tests (at least in some simple cases I timed), and in some
420 applications one wants PCRE to compile efficiently as well as match
421 efficiently.
422
423 For convenience, we use the same bit definitions as in chartables:
424
425 0x04 decimal digit
426 0x08 hexadecimal digit
427
428 Then we can use ctype_digit and ctype_xdigit in the code. */
429
430 #ifndef EBCDIC
431
432 /* This is the "normal" case, for ASCII systems, and EBCDIC systems running in
433 UTF-8 mode. */
434
435 static const unsigned char digitab[] =
436 {
437 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 0- 7 */
438 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 8- 15 */
439 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 16- 23 */
440 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 24- 31 */
441 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - ' */
442 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* ( - / */
443 0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c, /* 0 - 7 */
444 0x0c,0x0c,0x00,0x00,0x00,0x00,0x00,0x00, /* 8 - ? */
445 0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* @ - G */
446 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* H - O */
447 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* P - W */
448 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* X - _ */
449 0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* ` - g */
450 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* h - o */
451 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* p - w */
452 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* x -127 */
453 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 128-135 */
454 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 136-143 */
455 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 144-151 */
456 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 152-159 */
457 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 160-167 */
458 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 168-175 */
459 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 176-183 */
460 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 184-191 */
461 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 192-199 */
462 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 200-207 */
463 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 208-215 */
464 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 216-223 */
465 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 224-231 */
466 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 232-239 */
467 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 240-247 */
468 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00};/* 248-255 */
469
470 #else
471
472 /* This is the "abnormal" case, for EBCDIC systems not running in UTF-8 mode. */
473
474 static const unsigned char digitab[] =
475 {
476 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 0- 7 0 */
477 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 8- 15 */
478 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 16- 23 10 */
479 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 24- 31 */
480 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 32- 39 20 */
481 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 40- 47 */
482 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 48- 55 30 */
483 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 56- 63 */
484 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - 71 40 */
485 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 72- | */
486 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* & - 87 50 */
487 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 88- 95 */
488 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - -103 60 */
489 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 104- ? */
490 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 112-119 70 */
491 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 120- " */
492 0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* 128- g 80 */
493 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* h -143 */
494 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 144- p 90 */
495 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* q -159 */
496 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 160- x A0 */
497 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* y -175 */
498 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* ^ -183 B0 */
499 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 184-191 */
500 0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* { - G C0 */
501 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* H -207 */
502 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* } - P D0 */
503 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* Q -223 */
504 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* \ - X E0 */
505 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* Y -239 */
506 0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c, /* 0 - 7 F0 */
507 0x0c,0x0c,0x00,0x00,0x00,0x00,0x00,0x00};/* 8 -255 */
508
509 static const unsigned char ebcdic_chartab[] = { /* chartable partial dup */
510 0x80,0x00,0x00,0x00,0x00,0x01,0x00,0x00, /* 0- 7 */
511 0x00,0x00,0x00,0x00,0x01,0x01,0x00,0x00, /* 8- 15 */
512 0x00,0x00,0x00,0x00,0x00,0x01,0x00,0x00, /* 16- 23 */
513 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 24- 31 */
514 0x00,0x00,0x00,0x00,0x00,0x01,0x00,0x00, /* 32- 39 */
515 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 40- 47 */
516 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 48- 55 */
517 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 56- 63 */
518 0x01,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - 71 */
519 0x00,0x00,0x00,0x80,0x00,0x80,0x80,0x80, /* 72- | */
520 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* & - 87 */
521 0x00,0x00,0x00,0x80,0x80,0x80,0x00,0x00, /* 88- 95 */
522 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - -103 */
523 0x00,0x00,0x00,0x00,0x00,0x10,0x00,0x80, /* 104- ? */
524 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 112-119 */
525 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 120- " */
526 0x00,0x1a,0x1a,0x1a,0x1a,0x1a,0x1a,0x12, /* 128- g */
527 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* h -143 */
528 0x00,0x12,0x12,0x12,0x12,0x12,0x12,0x12, /* 144- p */
529 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* q -159 */
530 0x00,0x00,0x12,0x12,0x12,0x12,0x12,0x12, /* 160- x */
531 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* y -175 */
532 0x80,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* ^ -183 */
533 0x00,0x00,0x80,0x00,0x00,0x00,0x00,0x00, /* 184-191 */
534 0x80,0x1a,0x1a,0x1a,0x1a,0x1a,0x1a,0x12, /* { - G */
535 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* H -207 */
536 0x00,0x12,0x12,0x12,0x12,0x12,0x12,0x12, /* } - P */
537 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* Q -223 */
538 0x00,0x00,0x12,0x12,0x12,0x12,0x12,0x12, /* \ - X */
539 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* Y -239 */
540 0x1c,0x1c,0x1c,0x1c,0x1c,0x1c,0x1c,0x1c, /* 0 - 7 */
541 0x1c,0x1c,0x00,0x00,0x00,0x00,0x00,0x00};/* 8 -255 */
542 #endif
543
544
545 /* Definition to allow mutual recursion */
546
547 static BOOL
548 compile_regex(int, uschar **, const uschar **, int *, BOOL, BOOL, int, int *,
549 int *, branch_chain *, compile_data *, int *);
550
551
552
553 /*************************************************
554 * Find an error text *
555 *************************************************/
556
557 /* The error texts are now all in one long string, to save on relocations. As
558 some of the text is of unknown length, we can't use a table of offsets.
559 Instead, just count through the strings. This is not a performance issue
560 because it happens only when there has been a compilation error.
561
562 Argument: the error number
563 Returns: pointer to the error string
564 */
565
566 static const char *
567 find_error_text(int n)
568 {
569 const char *s = error_texts;
570 for (; n > 0; n--)
571 {
572 while (*s++ != 0) {};
573 if (*s == 0) return "Error text not found (please report)";
574 }
575 return s;
576 }
577
578
579 /*************************************************
580 * Handle escapes *
581 *************************************************/
582
583 /* This function is called when a \ has been encountered. It either returns a
584 positive value for a simple escape such as \n, or a negative value which
585 encodes one of the more complicated things such as \d. A backreference to group
586 n is returned as -(ESC_REF + n); ESC_REF is the highest ESC_xxx macro. When
587 UTF-8 is enabled, a positive value greater than 255 may be returned. On entry,
588 ptr is pointing at the \. On exit, it is on the final character of the escape
589 sequence.
590
591 Arguments:
592 ptrptr points to the pattern position pointer
593 errorcodeptr points to the errorcode variable
594 bracount number of previous extracting brackets
595 options the options bits
596 isclass TRUE if inside a character class
597
598 Returns: zero or positive => a data character
599 negative => a special escape sequence
600 on error, errorcodeptr is set
601 */
602
603 static int
604 check_escape(const uschar **ptrptr, int *errorcodeptr, int bracount,
605 int options, BOOL isclass)
606 {
607 BOOL utf8 = (options & PCRE_UTF8) != 0;
608 const uschar *ptr = *ptrptr + 1;
609 int c, i;
610
611 GETCHARINCTEST(c, ptr); /* Get character value, increment pointer */
612 ptr--; /* Set pointer back to the last byte */
613
614 /* If backslash is at the end of the pattern, it's an error. */
615
616 if (c == 0) *errorcodeptr = ERR1;
617
618 /* Non-alphanumerics are literals. For digits or letters, do an initial lookup
619 in a table. A non-zero result is something that can be returned immediately.
620 Otherwise further processing may be required. */
621
622 #ifndef EBCDIC /* ASCII/UTF-8 coding */
623 else if (c < CHAR_0 || c > CHAR_z) {} /* Not alphanumeric */
624 else if ((i = escapes[c - CHAR_0]) != 0) c = i;
625
626 #else /* EBCDIC coding */
627 else if (c < 'a' || (ebcdic_chartab[c] & 0x0E) == 0) {} /* Not alphanumeric */
628 else if ((i = escapes[c - 0x48]) != 0) c = i;
629 #endif
630
631 /* Escapes that need further processing, or are illegal. */
632
633 else
634 {
635 const uschar *oldptr;
636 BOOL braced, negated;
637
638 switch (c)
639 {
640 /* A number of Perl escapes are not handled by PCRE. We give an explicit
641 error. */
642
643 case CHAR_l:
644 case CHAR_L:
645 case CHAR_u:
646 case CHAR_U:
647 *errorcodeptr = ERR37;
648 break;
649
650 /* \g must be followed by one of a number of specific things:
651
652 (1) A number, either plain or braced. If positive, it is an absolute
653 backreference. If negative, it is a relative backreference. This is a Perl
654 5.10 feature.
655
656 (2) Perl 5.10 also supports \g{name} as a reference to a named group. This
657 is part of Perl's movement towards a unified syntax for back references. As
658 this is synonymous with \k{name}, we fudge it up by pretending it really
659 was \k.
660
661 (3) For Oniguruma compatibility we also support \g followed by a name or a
662 number either in angle brackets or in single quotes. However, these are
663 (possibly recursive) subroutine calls, _not_ backreferences. Just return
664 the -ESC_g code (cf \k). */
665
666 case CHAR_g:
667 if (ptr[1] == CHAR_LESS_THAN_SIGN || ptr[1] == CHAR_APOSTROPHE)
668 {
669 c = -ESC_g;
670 break;
671 }
672
673 /* Handle the Perl-compatible cases */
674
675 if (ptr[1] == CHAR_LEFT_CURLY_BRACKET)
676 {
677 const uschar *p;
678 for (p = ptr+2; *p != 0 && *p != CHAR_RIGHT_CURLY_BRACKET; p++)
679 if (*p != CHAR_MINUS && (digitab[*p] & ctype_digit) == 0) break;
680 if (*p != 0 && *p != CHAR_RIGHT_CURLY_BRACKET)
681 {
682 c = -ESC_k;
683 break;
684 }
685 braced = TRUE;
686 ptr++;
687 }
688 else braced = FALSE;
689
690 if (ptr[1] == CHAR_MINUS)
691 {
692 negated = TRUE;
693 ptr++;
694 }
695 else negated = FALSE;
696
697 c = 0;
698 while ((digitab[ptr[1]] & ctype_digit) != 0)
699 c = c * 10 + *(++ptr) - CHAR_0;
700
701 if (c < 0) /* Integer overflow */
702 {
703 *errorcodeptr = ERR61;
704 break;
705 }
706
707 if (braced && *(++ptr) != CHAR_RIGHT_CURLY_BRACKET)
708 {
709 *errorcodeptr = ERR57;
710 break;
711 }
712
713 if (c == 0)
714 {
715 *errorcodeptr = ERR58;
716 break;
717 }
718
719 if (negated)
720 {
721 if (c > bracount)
722 {
723 *errorcodeptr = ERR15;
724 break;
725 }
726 c = bracount - (c - 1);
727 }
728
729 c = -(ESC_REF + c);
730 break;
731
732 /* The handling of escape sequences consisting of a string of digits
733 starting with one that is not zero is not straightforward. By experiment,
734 the way Perl works seems to be as follows:
735
736 Outside a character class, the digits are read as a decimal number. If the
737 number is less than 10, or if there are that many previous extracting
738 left brackets, then it is a back reference. Otherwise, up to three octal
739 digits are read to form an escaped byte. Thus \123 is likely to be octal
740 123 (cf \0123, which is octal 012 followed by the literal 3). If the octal
741 value is greater than 377, the least significant 8 bits are taken. Inside a
742 character class, \ followed by a digit is always an octal number. */
743
744 case CHAR_1: case CHAR_2: case CHAR_3: case CHAR_4: case CHAR_5:
745 case CHAR_6: case CHAR_7: case CHAR_8: case CHAR_9:
746
747 if (!isclass)
748 {
749 oldptr = ptr;
750 c -= CHAR_0;
751 while ((digitab[ptr[1]] & ctype_digit) != 0)
752 c = c * 10 + *(++ptr) - CHAR_0;
753 if (c < 0) /* Integer overflow */
754 {
755 *errorcodeptr = ERR61;
756 break;
757 }
758 if (c < 10 || c <= bracount)
759 {
760 c = -(ESC_REF + c);
761 break;
762 }
763 ptr = oldptr; /* Put the pointer back and fall through */
764 }
765
766 /* Handle an octal number following \. If the first digit is 8 or 9, Perl
767 generates a binary zero byte and treats the digit as a following literal.
768 Thus we have to pull back the pointer by one. */
769
770 if ((c = *ptr) >= CHAR_8)
771 {
772 ptr--;
773 c = 0;
774 break;
775 }
776
777 /* \0 always starts an octal number, but we may drop through to here with a
778 larger first octal digit. The original code used just to take the least
779 significant 8 bits of octal numbers (I think this is what early Perls used
780 to do). Nowadays we allow for larger numbers in UTF-8 mode, but no more
781 than 3 octal digits. */
782
783 case CHAR_0:
784 c -= CHAR_0;
785 while(i++ < 2 && ptr[1] >= CHAR_0 && ptr[1] <= CHAR_7)
786 c = c * 8 + *(++ptr) - CHAR_0;
787 if (!utf8 && c > 255) *errorcodeptr = ERR51;
788 break;
789
790 /* \x is complicated. \x{ddd} is a character number which can be greater
791 than 0xff in utf8 mode, but only if the ddd are hex digits. If not, { is
792 treated as a data character. */
793
794 case CHAR_x:
795 if (ptr[1] == CHAR_LEFT_CURLY_BRACKET)
796 {
797 const uschar *pt = ptr + 2;
798 int count = 0;
799
800 c = 0;
801 while ((digitab[*pt] & ctype_xdigit) != 0)
802 {
803 register int cc = *pt++;
804 if (c == 0 && cc == CHAR_0) continue; /* Leading zeroes */
805 count++;
806
807 #ifndef EBCDIC /* ASCII/UTF-8 coding */
808 if (cc >= CHAR_a) cc -= 32; /* Convert to upper case */
809 c = (c << 4) + cc - ((cc < CHAR_A)? CHAR_0 : (CHAR_A - 10));
810 #else /* EBCDIC coding */
811 if (cc >= CHAR_a && cc <= CHAR_z) cc += 64; /* Convert to upper case */
812 c = (c << 4) + cc - ((cc >= CHAR_0)? CHAR_0 : (CHAR_A - 10));
813 #endif
814 }
815
816 if (*pt == CHAR_RIGHT_CURLY_BRACKET)
817 {
818 if (c < 0 || count > (utf8? 8 : 2)) *errorcodeptr = ERR34;
819 ptr = pt;
820 break;
821 }
822
823 /* If the sequence of hex digits does not end with '}', then we don't
824 recognize this construct; fall through to the normal \x handling. */
825 }
826
827 /* Read just a single-byte hex-defined char */
828
829 c = 0;
830 while (i++ < 2 && (digitab[ptr[1]] & ctype_xdigit) != 0)
831 {
832 int cc; /* Some compilers don't like */
833 cc = *(++ptr); /* ++ in initializers */
834 #ifndef EBCDIC /* ASCII/UTF-8 coding */
835 if (cc >= CHAR_a) cc -= 32; /* Convert to upper case */
836 c = c * 16 + cc - ((cc < CHAR_A)? CHAR_0 : (CHAR_A - 10));
837 #else /* EBCDIC coding */
838 if (cc <= CHAR_z) cc += 64; /* Convert to upper case */
839 c = c * 16 + cc - ((cc >= CHAR_0)? CHAR_0 : (CHAR_A - 10));
840 #endif
841 }
842 break;
843
844 /* For \c, a following letter is upper-cased; then the 0x40 bit is flipped.
845 An error is given if the byte following \c is not an ASCII character. This
846 coding is ASCII-specific, but then the whole concept of \cx is
847 ASCII-specific. (However, an EBCDIC equivalent has now been added.) */
848
849 case CHAR_c:
850 c = *(++ptr);
851 if (c == 0)
852 {
853 *errorcodeptr = ERR2;
854 break;
855 }
856 #ifndef EBCDIC /* ASCII/UTF-8 coding */
857 if (c > 127) /* Excludes all non-ASCII in either mode */
858 {
859 *errorcodeptr = ERR68;
860 break;
861 }
862 if (c >= CHAR_a && c <= CHAR_z) c -= 32;
863 c ^= 0x40;
864 #else /* EBCDIC coding */
865 if (c >= CHAR_a && c <= CHAR_z) c += 64;
866 c ^= 0xC0;
867 #endif
868 break;
869
870 /* PCRE_EXTRA enables extensions to Perl in the matter of escapes. Any
871 other alphanumeric following \ is an error if PCRE_EXTRA was set;
872 otherwise, for Perl compatibility, it is a literal. This code looks a bit
873 odd, but there used to be some cases other than the default, and there may
874 be again in future, so I haven't "optimized" it. */
875
876 default:
877 if ((options & PCRE_EXTRA) != 0) switch(c)
878 {
879 default:
880 *errorcodeptr = ERR3;
881 break;
882 }
883 break;
884 }
885 }
886
887 /* Perl supports \N{name} for character names, as well as plain \N for "not
888 newline". PCRE does not support \N{name}. */
889
890 if (c == -ESC_N && ptr[1] == CHAR_LEFT_CURLY_BRACKET)
891 *errorcodeptr = ERR37;
892
893 /* If PCRE_UCP is set, we change the values for \d etc. */
894
895 if ((options & PCRE_UCP) != 0 && c <= -ESC_D && c >= -ESC_w)
896 c -= (ESC_DU - ESC_D);
897
898 /* Set the pointer to the final character before returning. */
899
900 *ptrptr = ptr;
901 return c;
902 }
903
904
905
906 #ifdef SUPPORT_UCP
907 /*************************************************
908 * Handle \P and \p *
909 *************************************************/
910
911 /* This function is called after \P or \p has been encountered, provided that
912 PCRE is compiled with support for Unicode properties. On entry, ptrptr is
913 pointing at the P or p. On exit, it is pointing at the final character of the
914 escape sequence.
915
916 Argument:
917 ptrptr points to the pattern position pointer
918 negptr points to a boolean that is set TRUE for negation else FALSE
919 dptr points to an int that is set to the detailed property value
920 errorcodeptr points to the error code variable
921
922 Returns: type value from ucp_type_table, or -1 for an invalid type
923 */
924
925 static int
926 get_ucp(const uschar **ptrptr, BOOL *negptr, int *dptr, int *errorcodeptr)
927 {
928 int c, i, bot, top;
929 const uschar *ptr = *ptrptr;
930 char name[32];
931
932 c = *(++ptr);
933 if (c == 0) goto ERROR_RETURN;
934
935 *negptr = FALSE;
936
937 /* \P or \p can be followed by a name in {}, optionally preceded by ^ for
938 negation. */
939
940 if (c == CHAR_LEFT_CURLY_BRACKET)
941 {
942 if (ptr[1] == CHAR_CIRCUMFLEX_ACCENT)
943 {
944 *negptr = TRUE;
945 ptr++;
946 }
947 for (i = 0; i < (int)sizeof(name) - 1; i++)
948 {
949 c = *(++ptr);
950 if (c == 0) goto ERROR_RETURN;
951 if (c == CHAR_RIGHT_CURLY_BRACKET) break;
952 name[i] = c;
953 }
954 if (c != CHAR_RIGHT_CURLY_BRACKET) goto ERROR_RETURN;
955 name[i] = 0;
956 }
957
958 /* Otherwise there is just one following character */
959
960 else
961 {
962 name[0] = c;
963 name[1] = 0;
964 }
965
966 *ptrptr = ptr;
967
968 /* Search for a recognized property name using binary chop */
969
970 bot = 0;
971 top = _pcre_utt_size;
972
973 while (bot < top)
974 {
975 i = (bot + top) >> 1;
976 c = strcmp(name, _pcre_utt_names + _pcre_utt[i].name_offset);
977 if (c == 0)
978 {
979 *dptr = _pcre_utt[i].value;
980 return _pcre_utt[i].type;
981 }
982 if (c > 0) bot = i + 1; else top = i;
983 }
984
985 *errorcodeptr = ERR47;
986 *ptrptr = ptr;
987 return -1;
988
989 ERROR_RETURN:
990 *errorcodeptr = ERR46;
991 *ptrptr = ptr;
992 return -1;
993 }
994 #endif
995
996
997
998
999 /*************************************************
1000 * Check for counted repeat *
1001 *************************************************/
1002
1003 /* This function is called when a '{' is encountered in a place where it might
1004 start a quantifier. It looks ahead to see if it really is a quantifier or not.
1005 It is only a quantifier if it is one of the forms {ddd} {ddd,} or {ddd,ddd}
1006 where the ddds are digits.
1007
1008 Arguments:
1009 p pointer to the first char after '{'
1010
1011 Returns: TRUE or FALSE
1012 */
1013
1014 static BOOL
1015 is_counted_repeat(const uschar *p)
1016 {
1017 if ((digitab[*p++] & ctype_digit) == 0) return FALSE;
1018 while ((digitab[*p] & ctype_digit) != 0) p++;
1019 if (*p == CHAR_RIGHT_CURLY_BRACKET) return TRUE;
1020
1021 if (*p++ != CHAR_COMMA) return FALSE;
1022 if (*p == CHAR_RIGHT_CURLY_BRACKET) return TRUE;
1023
1024 if ((digitab[*p++] & ctype_digit) == 0) return FALSE;
1025 while ((digitab[*p] & ctype_digit) != 0) p++;
1026
1027 return (*p == CHAR_RIGHT_CURLY_BRACKET);
1028 }
1029
1030
1031
1032 /*************************************************
1033 * Read repeat counts *
1034 *************************************************/
1035
1036 /* Read an item of the form {n,m} and return the values. This is called only
1037 after is_counted_repeat() has confirmed that a repeat-count quantifier exists,
1038 so the syntax is guaranteed to be correct, but we need to check the values.
1039
1040 Arguments:
1041 p pointer to first char after '{'
1042 minp pointer to int for min
1043 maxp pointer to int for max
1044 returned as -1 if no max
1045 errorcodeptr points to error code variable
1046
1047 Returns: pointer to '}' on success;
1048 current ptr on error, with errorcodeptr set non-zero
1049 */
1050
1051 static const uschar *
1052 read_repeat_counts(const uschar *p, int *minp, int *maxp, int *errorcodeptr)
1053 {
1054 int min = 0;
1055 int max = -1;
1056
1057 /* Read the minimum value and do a paranoid check: a negative value indicates
1058 an integer overflow. */
1059
1060 while ((digitab[*p] & ctype_digit) != 0) min = min * 10 + *p++ - CHAR_0;
1061 if (min < 0 || min > 65535)
1062 {
1063 *errorcodeptr = ERR5;
1064 return p;
1065 }
1066
1067 /* Read the maximum value if there is one, and again do a paranoid on its size.
1068 Also, max must not be less than min. */
1069
1070 if (*p == CHAR_RIGHT_CURLY_BRACKET) max = min; else
1071 {
1072 if (*(++p) != CHAR_RIGHT_CURLY_BRACKET)
1073 {
1074 max = 0;
1075 while((digitab[*p] & ctype_digit) != 0) max = max * 10 + *p++ - CHAR_0;
1076 if (max < 0 || max > 65535)
1077 {
1078 *errorcodeptr = ERR5;
1079 return p;
1080 }
1081 if (max < min)
1082 {
1083 *errorcodeptr = ERR4;
1084 return p;
1085 }
1086 }
1087 }
1088
1089 /* Fill in the required variables, and pass back the pointer to the terminating
1090 '}'. */
1091
1092 *minp = min;
1093 *maxp = max;
1094 return p;
1095 }
1096
1097
1098
1099 /*************************************************
1100 * Subroutine for finding forward reference *
1101 *************************************************/
1102
1103 /* This recursive function is called only from find_parens() below. The
1104 top-level call starts at the beginning of the pattern. All other calls must
1105 start at a parenthesis. It scans along a pattern's text looking for capturing
1106 subpatterns, and counting them. If it finds a named pattern that matches the
1107 name it is given, it returns its number. Alternatively, if the name is NULL, it
1108 returns when it reaches a given numbered subpattern. Recursion is used to keep
1109 track of subpatterns that reset the capturing group numbers - the (?| feature.
1110
1111 This function was originally called only from the second pass, in which we know
1112 that if (?< or (?' or (?P< is encountered, the name will be correctly
1113 terminated because that is checked in the first pass. There is now one call to
1114 this function in the first pass, to check for a recursive back reference by
1115 name (so that we can make the whole group atomic). In this case, we need check
1116 only up to the current position in the pattern, and that is still OK because
1117 and previous occurrences will have been checked. To make this work, the test
1118 for "end of pattern" is a check against cd->end_pattern in the main loop,
1119 instead of looking for a binary zero. This means that the special first-pass
1120 call can adjust cd->end_pattern temporarily. (Checks for binary zero while
1121 processing items within the loop are OK, because afterwards the main loop will
1122 terminate.)
1123
1124 Arguments:
1125 ptrptr address of the current character pointer (updated)
1126 cd compile background data
1127 name name to seek, or NULL if seeking a numbered subpattern
1128 lorn name length, or subpattern number if name is NULL
1129 xmode TRUE if we are in /x mode
1130 utf8 TRUE if we are in UTF-8 mode
1131 count pointer to the current capturing subpattern number (updated)
1132
1133 Returns: the number of the named subpattern, or -1 if not found
1134 */
1135
1136 static int
1137 find_parens_sub(uschar **ptrptr, compile_data *cd, const uschar *name, int lorn,
1138 BOOL xmode, BOOL utf8, int *count)
1139 {
1140 uschar *ptr = *ptrptr;
1141 int start_count = *count;
1142 int hwm_count = start_count;
1143 BOOL dup_parens = FALSE;
1144
1145 /* If the first character is a parenthesis, check on the type of group we are
1146 dealing with. The very first call may not start with a parenthesis. */
1147
1148 if (ptr[0] == CHAR_LEFT_PARENTHESIS)
1149 {
1150 /* Handle specials such as (*SKIP) or (*UTF8) etc. */
1151
1152 if (ptr[1] == CHAR_ASTERISK) ptr += 2;
1153
1154 /* Handle a normal, unnamed capturing parenthesis. */
1155
1156 else if (ptr[1] != CHAR_QUESTION_MARK)
1157 {
1158 *count += 1;
1159 if (name == NULL && *count == lorn) return *count;
1160 ptr++;
1161 }
1162
1163 /* All cases now have (? at the start. Remember when we are in a group
1164 where the parenthesis numbers are duplicated. */
1165
1166 else if (ptr[2] == CHAR_VERTICAL_LINE)
1167 {
1168 ptr += 3;
1169 dup_parens = TRUE;
1170 }
1171
1172 /* Handle comments; all characters are allowed until a ket is reached. */
1173
1174 else if (ptr[2] == CHAR_NUMBER_SIGN)
1175 {
1176 for (ptr += 3; *ptr != 0; ptr++) if (*ptr == CHAR_RIGHT_PARENTHESIS) break;
1177 goto FAIL_EXIT;
1178 }
1179
1180 /* Handle a condition. If it is an assertion, just carry on so that it
1181 is processed as normal. If not, skip to the closing parenthesis of the
1182 condition (there can't be any nested parens). */
1183
1184 else if (ptr[2] == CHAR_LEFT_PARENTHESIS)
1185 {
1186 ptr += 2;
1187 if (ptr[1] != CHAR_QUESTION_MARK)
1188 {
1189 while (*ptr != 0 && *ptr != CHAR_RIGHT_PARENTHESIS) ptr++;
1190 if (*ptr != 0) ptr++;
1191 }
1192 }
1193
1194 /* Start with (? but not a condition. */
1195
1196 else
1197 {
1198 ptr += 2;
1199 if (*ptr == CHAR_P) ptr++; /* Allow optional P */
1200
1201 /* We have to disambiguate (?<! and (?<= from (?<name> for named groups */
1202
1203 if ((*ptr == CHAR_LESS_THAN_SIGN && ptr[1] != CHAR_EXCLAMATION_MARK &&
1204 ptr[1] != CHAR_EQUALS_SIGN) || *ptr == CHAR_APOSTROPHE)
1205 {
1206 int term;
1207 const uschar *thisname;
1208 *count += 1;
1209 if (name == NULL && *count == lorn) return *count;
1210 term = *ptr++;
1211 if (term == CHAR_LESS_THAN_SIGN) term = CHAR_GREATER_THAN_SIGN;
1212 thisname = ptr;
1213 while (*ptr != term) ptr++;
1214 if (name != NULL && lorn == ptr - thisname &&
1215 strncmp((const char *)name, (const char *)thisname, lorn) == 0)
1216 return *count;
1217 term++;
1218 }
1219 }
1220 }
1221
1222 /* Past any initial parenthesis handling, scan for parentheses or vertical
1223 bars. Stop if we get to cd->end_pattern. Note that this is important for the
1224 first-pass call when this value is temporarily adjusted to stop at the current
1225 position. So DO NOT change this to a test for binary zero. */
1226
1227 for (; ptr < cd->end_pattern; ptr++)
1228 {
1229 /* Skip over backslashed characters and also entire \Q...\E */
1230
1231 if (*ptr == CHAR_BACKSLASH)
1232 {
1233 if (*(++ptr) == 0) goto FAIL_EXIT;
1234 if (*ptr == CHAR_Q) for (;;)
1235 {
1236 while (*(++ptr) != 0 && *ptr != CHAR_BACKSLASH) {};
1237 if (*ptr == 0) goto FAIL_EXIT;
1238 if (*(++ptr) == CHAR_E) break;
1239 }
1240 continue;
1241 }
1242
1243 /* Skip over character classes; this logic must be similar to the way they
1244 are handled for real. If the first character is '^', skip it. Also, if the
1245 first few characters (either before or after ^) are \Q\E or \E we skip them
1246 too. This makes for compatibility with Perl. Note the use of STR macros to
1247 encode "Q\\E" so that it works in UTF-8 on EBCDIC platforms. */
1248
1249 if (*ptr == CHAR_LEFT_SQUARE_BRACKET)
1250 {
1251 BOOL negate_class = FALSE;
1252 for (;;)
1253 {
1254 if (ptr[1] == CHAR_BACKSLASH)
1255 {
1256 if (ptr[2] == CHAR_E)
1257 ptr+= 2;
1258 else if (strncmp((const char *)ptr+2,
1259 STR_Q STR_BACKSLASH STR_E, 3) == 0)
1260 ptr += 4;
1261 else
1262 break;
1263 }
1264 else if (!negate_class && ptr[1] == CHAR_CIRCUMFLEX_ACCENT)
1265 {
1266 negate_class = TRUE;
1267 ptr++;
1268 }
1269 else break;
1270 }
1271
1272 /* If the next character is ']', it is a data character that must be
1273 skipped, except in JavaScript compatibility mode. */
1274
1275 if (ptr[1] == CHAR_RIGHT_SQUARE_BRACKET &&
1276 (cd->external_options & PCRE_JAVASCRIPT_COMPAT) == 0)
1277 ptr++;
1278
1279 while (*(++ptr) != CHAR_RIGHT_SQUARE_BRACKET)
1280 {
1281 if (*ptr == 0) return -1;
1282 if (*ptr == CHAR_BACKSLASH)
1283 {
1284 if (*(++ptr) == 0) goto FAIL_EXIT;
1285 if (*ptr == CHAR_Q) for (;;)
1286 {
1287 while (*(++ptr) != 0 && *ptr != CHAR_BACKSLASH) {};
1288 if (*ptr == 0) goto FAIL_EXIT;
1289 if (*(++ptr) == CHAR_E) break;
1290 }
1291 continue;
1292 }
1293 }
1294 continue;
1295 }
1296
1297 /* Skip comments in /x mode */
1298
1299 if (xmode && *ptr == CHAR_NUMBER_SIGN)
1300 {
1301 ptr++;
1302 while (*ptr != 0)
1303 {
1304 if (IS_NEWLINE(ptr)) { ptr += cd->nllen - 1; break; }
1305 ptr++;
1306 #ifdef SUPPORT_UTF8
1307 if (utf8) while ((*ptr & 0xc0) == 0x80) ptr++;
1308 #endif
1309 }
1310 if (*ptr == 0) goto FAIL_EXIT;
1311 continue;
1312 }
1313
1314 /* Check for the special metacharacters */
1315
1316 if (*ptr == CHAR_LEFT_PARENTHESIS)
1317 {
1318 int rc = find_parens_sub(&ptr, cd, name, lorn, xmode, utf8, count);
1319 if (rc > 0) return rc;
1320 if (*ptr == 0) goto FAIL_EXIT;
1321 }
1322
1323 else if (*ptr == CHAR_RIGHT_PARENTHESIS)
1324 {
1325 if (dup_parens && *count < hwm_count) *count = hwm_count;
1326 goto FAIL_EXIT;
1327 }
1328
1329 else if (*ptr == CHAR_VERTICAL_LINE && dup_parens)
1330 {
1331 if (*count > hwm_count) hwm_count = *count;
1332 *count = start_count;
1333 }
1334 }
1335
1336 FAIL_EXIT:
1337 *ptrptr = ptr;
1338 return -1;
1339 }
1340
1341
1342
1343
1344 /*************************************************
1345 * Find forward referenced subpattern *
1346 *************************************************/
1347
1348 /* This function scans along a pattern's text looking for capturing
1349 subpatterns, and counting them. If it finds a named pattern that matches the
1350 name it is given, it returns its number. Alternatively, if the name is NULL, it
1351 returns when it reaches a given numbered subpattern. This is used for forward
1352 references to subpatterns. We used to be able to start this scan from the
1353 current compiling point, using the current count value from cd->bracount, and
1354 do it all in a single loop, but the addition of the possibility of duplicate
1355 subpattern numbers means that we have to scan from the very start, in order to
1356 take account of such duplicates, and to use a recursive function to keep track
1357 of the different types of group.
1358
1359 Arguments:
1360 cd compile background data
1361 name name to seek, or NULL if seeking a numbered subpattern
1362 lorn name length, or subpattern number if name is NULL
1363 xmode TRUE if we are in /x mode
1364 utf8 TRUE if we are in UTF-8 mode
1365
1366 Returns: the number of the found subpattern, or -1 if not found
1367 */
1368
1369 static int
1370 find_parens(compile_data *cd, const uschar *name, int lorn, BOOL xmode,
1371 BOOL utf8)
1372 {
1373 uschar *ptr = (uschar *)cd->start_pattern;
1374 int count = 0;
1375 int rc;
1376
1377 /* If the pattern does not start with an opening parenthesis, the first call
1378 to find_parens_sub() will scan right to the end (if necessary). However, if it
1379 does start with a parenthesis, find_parens_sub() will return when it hits the
1380 matching closing parens. That is why we have to have a loop. */
1381
1382 for (;;)
1383 {
1384 rc = find_parens_sub(&ptr, cd, name, lorn, xmode, utf8, &count);
1385 if (rc > 0 || *ptr++ == 0) break;
1386 }
1387
1388 return rc;
1389 }
1390
1391
1392
1393
1394 /*************************************************
1395 * Find first significant op code *
1396 *************************************************/
1397
1398 /* This is called by several functions that scan a compiled expression looking
1399 for a fixed first character, or an anchoring op code etc. It skips over things
1400 that do not influence this. For some calls, it makes sense to skip negative
1401 forward and all backward assertions, and also the \b assertion; for others it
1402 does not.
1403
1404 Arguments:
1405 code pointer to the start of the group
1406 skipassert TRUE if certain assertions are to be skipped
1407
1408 Returns: pointer to the first significant opcode
1409 */
1410
1411 static const uschar*
1412 first_significant_code(const uschar *code, BOOL skipassert)
1413 {
1414 for (;;)
1415 {
1416 switch ((int)*code)
1417 {
1418 case OP_ASSERT_NOT:
1419 case OP_ASSERTBACK:
1420 case OP_ASSERTBACK_NOT:
1421 if (!skipassert) return code;
1422 do code += GET(code, 1); while (*code == OP_ALT);
1423 code += _pcre_OP_lengths[*code];
1424 break;
1425
1426 case OP_WORD_BOUNDARY:
1427 case OP_NOT_WORD_BOUNDARY:
1428 if (!skipassert) return code;
1429 /* Fall through */
1430
1431 case OP_CALLOUT:
1432 case OP_CREF:
1433 case OP_NCREF:
1434 case OP_RREF:
1435 case OP_NRREF:
1436 case OP_DEF:
1437 code += _pcre_OP_lengths[*code];
1438 break;
1439
1440 default:
1441 return code;
1442 }
1443 }
1444 /* Control never reaches here */
1445 }
1446
1447
1448
1449
1450 /*************************************************
1451 * Find the fixed length of a branch *
1452 *************************************************/
1453
1454 /* Scan a branch and compute the fixed length of subject that will match it,
1455 if the length is fixed. This is needed for dealing with backward assertions.
1456 In UTF8 mode, the result is in characters rather than bytes. The branch is
1457 temporarily terminated with OP_END when this function is called.
1458
1459 This function is called when a backward assertion is encountered, so that if it
1460 fails, the error message can point to the correct place in the pattern.
1461 However, we cannot do this when the assertion contains subroutine calls,
1462 because they can be forward references. We solve this by remembering this case
1463 and doing the check at the end; a flag specifies which mode we are running in.
1464
1465 Arguments:
1466 code points to the start of the pattern (the bracket)
1467 utf8 TRUE in UTF-8 mode
1468 atend TRUE if called when the pattern is complete
1469 cd the "compile data" structure
1470
1471 Returns: the fixed length,
1472 or -1 if there is no fixed length,
1473 or -2 if \C was encountered
1474 or -3 if an OP_RECURSE item was encountered and atend is FALSE
1475 */
1476
1477 static int
1478 find_fixedlength(uschar *code, BOOL utf8, BOOL atend, compile_data *cd)
1479 {
1480 int length = -1;
1481
1482 register int branchlength = 0;
1483 register uschar *cc = code + 1 + LINK_SIZE;
1484
1485 /* Scan along the opcodes for this branch. If we get to the end of the
1486 branch, check the length against that of the other branches. */
1487
1488 for (;;)
1489 {
1490 int d;
1491 uschar *ce, *cs;
1492 register int op = *cc;
1493 switch (op)
1494 {
1495 /* We only need to continue for OP_CBRA (normal capturing bracket) and
1496 OP_BRA (normal non-capturing bracket) because the other variants of these
1497 opcodes are all concerned with unlimited repeated groups, which of course
1498 are not of fixed length. They will cause a -1 response from the default
1499 case of this switch. */
1500
1501 case OP_CBRA:
1502 case OP_BRA:
1503 case OP_ONCE:
1504 case OP_COND:
1505 d = find_fixedlength(cc + ((op == OP_CBRA)? 2:0), utf8, atend, cd);
1506 if (d < 0) return d;
1507 branchlength += d;
1508 do cc += GET(cc, 1); while (*cc == OP_ALT);
1509 cc += 1 + LINK_SIZE;
1510 break;
1511
1512 /* Reached end of a branch; if it's a ket it is the end of a nested
1513 call. If it's ALT it is an alternation in a nested call. If it is
1514 END it's the end of the outer call. All can be handled by the same code.
1515 Note that we must not include the OP_KETRxxx opcodes here, because they
1516 all imply an unlimited repeat. */
1517
1518 case OP_ALT:
1519 case OP_KET:
1520 case OP_END:
1521 if (length < 0) length = branchlength;
1522 else if (length != branchlength) return -1;
1523 if (*cc != OP_ALT) return length;
1524 cc += 1 + LINK_SIZE;
1525 branchlength = 0;
1526 break;
1527
1528 /* A true recursion implies not fixed length, but a subroutine call may
1529 be OK. If the subroutine is a forward reference, we can't deal with
1530 it until the end of the pattern, so return -3. */
1531
1532 case OP_RECURSE:
1533 if (!atend) return -3;
1534 cs = ce = (uschar *)cd->start_code + GET(cc, 1); /* Start subpattern */
1535 do ce += GET(ce, 1); while (*ce == OP_ALT); /* End subpattern */
1536 if (cc > cs && cc < ce) return -1; /* Recursion */
1537 d = find_fixedlength(cs + 2, utf8, atend, cd);
1538 if (d < 0) return d;
1539 branchlength += d;
1540 cc += 1 + LINK_SIZE;
1541 break;
1542
1543 /* Skip over assertive subpatterns */
1544
1545 case OP_ASSERT:
1546 case OP_ASSERT_NOT:
1547 case OP_ASSERTBACK:
1548 case OP_ASSERTBACK_NOT:
1549 do cc += GET(cc, 1); while (*cc == OP_ALT);
1550 /* Fall through */
1551
1552 /* Skip over things that don't match chars */
1553
1554 case OP_REVERSE:
1555 case OP_CREF:
1556 case OP_NCREF:
1557 case OP_RREF:
1558 case OP_NRREF:
1559 case OP_DEF:
1560 case OP_CALLOUT:
1561 case OP_SOD:
1562 case OP_SOM:
1563 case OP_SET_SOM:
1564 case OP_EOD:
1565 case OP_EODN:
1566 case OP_CIRC:
1567 case OP_CIRCM:
1568 case OP_DOLL:
1569 case OP_DOLLM:
1570 case OP_NOT_WORD_BOUNDARY:
1571 case OP_WORD_BOUNDARY:
1572 cc += _pcre_OP_lengths[*cc];
1573 break;
1574
1575 /* Handle literal characters */
1576
1577 case OP_CHAR:
1578 case OP_CHARI:
1579 case OP_NOT:
1580 case OP_NOTI:
1581 branchlength++;
1582 cc += 2;
1583 #ifdef SUPPORT_UTF8
1584 if (utf8 && cc[-1] >= 0xc0) cc += _pcre_utf8_table4[cc[-1] & 0x3f];
1585 #endif
1586 break;
1587
1588 /* Handle exact repetitions. The count is already in characters, but we
1589 need to skip over a multibyte character in UTF8 mode. */
1590
1591 case OP_EXACT:
1592 branchlength += GET2(cc,1);
1593 cc += 4;
1594 #ifdef SUPPORT_UTF8
1595 if (utf8 && cc[-1] >= 0xc0) cc += _pcre_utf8_table4[cc[-1] & 0x3f];
1596 #endif
1597 break;
1598
1599 case OP_TYPEEXACT:
1600 branchlength += GET2(cc,1);
1601 if (cc[3] == OP_PROP || cc[3] == OP_NOTPROP) cc += 2;
1602 cc += 4;
1603 break;
1604
1605 /* Handle single-char matchers */
1606
1607 case OP_PROP:
1608 case OP_NOTPROP:
1609 cc += 2;
1610 /* Fall through */
1611
1612 case OP_NOT_DIGIT:
1613 case OP_DIGIT:
1614 case OP_NOT_WHITESPACE:
1615 case OP_WHITESPACE:
1616 case OP_NOT_WORDCHAR:
1617 case OP_WORDCHAR:
1618 case OP_ANY:
1619 case OP_ALLANY:
1620 branchlength++;
1621 cc++;
1622 break;
1623
1624 /* The single-byte matcher isn't allowed */
1625
1626 case OP_ANYBYTE:
1627 return -2;
1628
1629 /* Check a class for variable quantification */
1630
1631 #ifdef SUPPORT_UTF8
1632 case OP_XCLASS:
1633 cc += GET(cc, 1) - 33;
1634 /* Fall through */
1635 #endif
1636
1637 case OP_CLASS:
1638 case OP_NCLASS:
1639 cc += 33;
1640
1641 switch (*cc)
1642 {
1643 case OP_CRSTAR:
1644 case OP_CRMINSTAR:
1645 case OP_CRQUERY:
1646 case OP_CRMINQUERY:
1647 return -1;
1648
1649 case OP_CRRANGE:
1650 case OP_CRMINRANGE:
1651 if (GET2(cc,1) != GET2(cc,3)) return -1;
1652 branchlength += GET2(cc,1);
1653 cc += 5;
1654 break;
1655
1656 default:
1657 branchlength++;
1658 }
1659 break;
1660
1661 /* Anything else is variable length */
1662
1663 default:
1664 return -1;
1665 }
1666 }
1667 /* Control never gets here */
1668 }
1669
1670
1671
1672
1673 /*************************************************
1674 * Scan compiled regex for specific bracket *
1675 *************************************************/
1676
1677 /* This little function scans through a compiled pattern until it finds a
1678 capturing bracket with the given number, or, if the number is negative, an
1679 instance of OP_REVERSE for a lookbehind. The function is global in the C sense
1680 so that it can be called from pcre_study() when finding the minimum matching
1681 length.
1682
1683 Arguments:
1684 code points to start of expression
1685 utf8 TRUE in UTF-8 mode
1686 number the required bracket number or negative to find a lookbehind
1687
1688 Returns: pointer to the opcode for the bracket, or NULL if not found
1689 */
1690
1691 const uschar *
1692 _pcre_find_bracket(const uschar *code, BOOL utf8, int number)
1693 {
1694 for (;;)
1695 {
1696 register int c = *code;
1697
1698 if (c == OP_END) return NULL;
1699
1700 /* XCLASS is used for classes that cannot be represented just by a bit
1701 map. This includes negated single high-valued characters. The length in
1702 the table is zero; the actual length is stored in the compiled code. */
1703
1704 if (c == OP_XCLASS) code += GET(code, 1);
1705
1706 /* Handle recursion */
1707
1708 else if (c == OP_REVERSE)
1709 {
1710 if (number < 0) return (uschar *)code;
1711 code += _pcre_OP_lengths[c];
1712 }
1713
1714 /* Handle capturing bracket */
1715
1716 else if (c == OP_CBRA || c == OP_SCBRA ||
1717 c == OP_CBRAPOS || c == OP_SCBRAPOS)
1718 {
1719 int n = GET2(code, 1+LINK_SIZE);
1720 if (n == number) return (uschar *)code;
1721 code += _pcre_OP_lengths[c];
1722 }
1723
1724 /* Otherwise, we can get the item's length from the table, except that for
1725 repeated character types, we have to test for \p and \P, which have an extra
1726 two bytes of parameters, and for MARK/PRUNE/SKIP/THEN with an argument, we
1727 must add in its length. */
1728
1729 else
1730 {
1731 switch(c)
1732 {
1733 case OP_TYPESTAR:
1734 case OP_TYPEMINSTAR:
1735 case OP_TYPEPLUS:
1736 case OP_TYPEMINPLUS:
1737 case OP_TYPEQUERY:
1738 case OP_TYPEMINQUERY:
1739 case OP_TYPEPOSSTAR:
1740 case OP_TYPEPOSPLUS:
1741 case OP_TYPEPOSQUERY:
1742 if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
1743 break;
1744
1745 case OP_TYPEUPTO:
1746 case OP_TYPEMINUPTO:
1747 case OP_TYPEEXACT:
1748 case OP_TYPEPOSUPTO:
1749 if (code[3] == OP_PROP || code[3] == OP_NOTPROP) code += 2;
1750 break;
1751
1752 case OP_MARK:
1753 case OP_PRUNE_ARG:
1754 case OP_SKIP_ARG:
1755 code += code[1];
1756 break;
1757
1758 case OP_THEN_ARG:
1759 code += code[1+LINK_SIZE];
1760 break;
1761 }
1762
1763 /* Add in the fixed length from the table */
1764
1765 code += _pcre_OP_lengths[c];
1766
1767 /* In UTF-8 mode, opcodes that are followed by a character may be followed by
1768 a multi-byte character. The length in the table is a minimum, so we have to
1769 arrange to skip the extra bytes. */
1770
1771 #ifdef SUPPORT_UTF8
1772 if (utf8) switch(c)
1773 {
1774 case OP_CHAR:
1775 case OP_CHARI:
1776 case OP_EXACT:
1777 case OP_EXACTI:
1778 case OP_UPTO:
1779 case OP_UPTOI:
1780 case OP_MINUPTO:
1781 case OP_MINUPTOI:
1782 case OP_POSUPTO:
1783 case OP_POSUPTOI:
1784 case OP_STAR:
1785 case OP_STARI:
1786 case OP_MINSTAR:
1787 case OP_MINSTARI:
1788 case OP_POSSTAR:
1789 case OP_POSSTARI:
1790 case OP_PLUS:
1791 case OP_PLUSI:
1792 case OP_MINPLUS:
1793 case OP_MINPLUSI:
1794 case OP_POSPLUS:
1795 case OP_POSPLUSI:
1796 case OP_QUERY:
1797 case OP_QUERYI:
1798 case OP_MINQUERY:
1799 case OP_MINQUERYI:
1800 case OP_POSQUERY:
1801 case OP_POSQUERYI:
1802 if (code[-1] >= 0xc0) code += _pcre_utf8_table4[code[-1] & 0x3f];
1803 break;
1804 }
1805 #else
1806 (void)(utf8); /* Keep compiler happy by referencing function argument */
1807 #endif
1808 }
1809 }
1810 }
1811
1812
1813
1814 /*************************************************
1815 * Scan compiled regex for recursion reference *
1816 *************************************************/
1817
1818 /* This little function scans through a compiled pattern until it finds an
1819 instance of OP_RECURSE.
1820
1821 Arguments:
1822 code points to start of expression
1823 utf8 TRUE in UTF-8 mode
1824
1825 Returns: pointer to the opcode for OP_RECURSE, or NULL if not found
1826 */
1827
1828 static const uschar *
1829 find_recurse(const uschar *code, BOOL utf8)
1830 {
1831 for (;;)
1832 {
1833 register int c = *code;
1834 if (c == OP_END) return NULL;
1835 if (c == OP_RECURSE) return code;
1836
1837 /* XCLASS is used for classes that cannot be represented just by a bit
1838 map. This includes negated single high-valued characters. The length in
1839 the table is zero; the actual length is stored in the compiled code. */
1840
1841 if (c == OP_XCLASS) code += GET(code, 1);
1842
1843 /* Otherwise, we can get the item's length from the table, except that for
1844 repeated character types, we have to test for \p and \P, which have an extra
1845 two bytes of parameters, and for MARK/PRUNE/SKIP/THEN with an argument, we
1846 must add in its length. */
1847
1848 else
1849 {
1850 switch(c)
1851 {
1852 case OP_TYPESTAR:
1853 case OP_TYPEMINSTAR:
1854 case OP_TYPEPLUS:
1855 case OP_TYPEMINPLUS:
1856 case OP_TYPEQUERY:
1857 case OP_TYPEMINQUERY:
1858 case OP_TYPEPOSSTAR:
1859 case OP_TYPEPOSPLUS:
1860 case OP_TYPEPOSQUERY:
1861 if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
1862 break;
1863
1864 case OP_TYPEPOSUPTO:
1865 case OP_TYPEUPTO:
1866 case OP_TYPEMINUPTO:
1867 case OP_TYPEEXACT:
1868 if (code[3] == OP_PROP || code[3] == OP_NOTPROP) code += 2;
1869 break;
1870
1871 case OP_MARK:
1872 case OP_PRUNE_ARG:
1873 case OP_SKIP_ARG:
1874 code += code[1];
1875 break;
1876
1877 case OP_THEN_ARG:
1878 code += code[1+LINK_SIZE];
1879 break;
1880 }
1881
1882 /* Add in the fixed length from the table */
1883
1884 code += _pcre_OP_lengths[c];
1885
1886 /* In UTF-8 mode, opcodes that are followed by a character may be followed
1887 by a multi-byte character. The length in the table is a minimum, so we have
1888 to arrange to skip the extra bytes. */
1889
1890 #ifdef SUPPORT_UTF8
1891 if (utf8) switch(c)
1892 {
1893 case OP_CHAR:
1894 case OP_CHARI:
1895 case OP_EXACT:
1896 case OP_EXACTI:
1897 case OP_UPTO:
1898 case OP_UPTOI:
1899 case OP_MINUPTO:
1900 case OP_MINUPTOI:
1901 case OP_POSUPTO:
1902 case OP_POSUPTOI:
1903 case OP_STAR:
1904 case OP_STARI:
1905 case OP_MINSTAR:
1906 case OP_MINSTARI:
1907 case OP_POSSTAR:
1908 case OP_POSSTARI:
1909 case OP_PLUS:
1910 case OP_PLUSI:
1911 case OP_MINPLUS:
1912 case OP_MINPLUSI:
1913 case OP_POSPLUS:
1914 case OP_POSPLUSI:
1915 case OP_QUERY:
1916 case OP_QUERYI:
1917 case OP_MINQUERY:
1918 case OP_MINQUERYI:
1919 case OP_POSQUERY:
1920 case OP_POSQUERYI:
1921 if (code[-1] >= 0xc0) code += _pcre_utf8_table4[code[-1] & 0x3f];
1922 break;
1923 }
1924 #else
1925 (void)(utf8); /* Keep compiler happy by referencing function argument */
1926 #endif
1927 }
1928 }
1929 }
1930
1931
1932
1933 /*************************************************
1934 * Scan compiled branch for non-emptiness *
1935 *************************************************/
1936
1937 /* This function scans through a branch of a compiled pattern to see whether it
1938 can match the empty string or not. It is called from could_be_empty()
1939 below and from compile_branch() when checking for an unlimited repeat of a
1940 group that can match nothing. Note that first_significant_code() skips over
1941 backward and negative forward assertions when its final argument is TRUE. If we
1942 hit an unclosed bracket, we return "empty" - this means we've struck an inner
1943 bracket whose current branch will already have been scanned.
1944
1945 Arguments:
1946 code points to start of search
1947 endcode points to where to stop
1948 utf8 TRUE if in UTF8 mode
1949 cd contains pointers to tables etc.
1950
1951 Returns: TRUE if what is matched could be empty
1952 */
1953
1954 static BOOL
1955 could_be_empty_branch(const uschar *code, const uschar *endcode, BOOL utf8,
1956 compile_data *cd)
1957 {
1958 register int c;
1959 for (code = first_significant_code(code + _pcre_OP_lengths[*code], TRUE);
1960 code < endcode;
1961 code = first_significant_code(code + _pcre_OP_lengths[c], TRUE))
1962 {
1963 const uschar *ccode;
1964
1965 c = *code;
1966
1967 /* Skip over forward assertions; the other assertions are skipped by
1968 first_significant_code() with a TRUE final argument. */
1969
1970 if (c == OP_ASSERT)
1971 {
1972 do code += GET(code, 1); while (*code == OP_ALT);
1973 c = *code;
1974 continue;
1975 }
1976
1977 /* For a recursion/subroutine call, if its end has been reached, which
1978 implies a subroutine call, we can scan it. */
1979
1980 if (c == OP_RECURSE)
1981 {
1982 BOOL empty_branch = FALSE;
1983 const uschar *scode = cd->start_code + GET(code, 1);
1984 if (GET(scode, 1) == 0) return TRUE; /* Unclosed */
1985 do
1986 {
1987 if (could_be_empty_branch(scode, endcode, utf8, cd))
1988 {
1989 empty_branch = TRUE;
1990 break;
1991 }
1992 scode += GET(scode, 1);
1993 }
1994 while (*scode == OP_ALT);
1995 if (!empty_branch) return FALSE; /* All branches are non-empty */
1996 continue;
1997 }
1998
1999 /* Groups with zero repeats can of course be empty; skip them. */
2000
2001 if (c == OP_BRAZERO || c == OP_BRAMINZERO || c == OP_SKIPZERO ||
2002 c == OP_BRAPOSZERO)
2003 {
2004 code += _pcre_OP_lengths[c];
2005 do code += GET(code, 1); while (*code == OP_ALT);
2006 c = *code;
2007 continue;
2008 }
2009
2010 /* A nested group that is already marked as "could be empty" can just be
2011 skipped. */
2012
2013 if (c == OP_SBRA || c == OP_SBRAPOS ||
2014 c == OP_SCBRA || c == OP_SCBRAPOS)
2015 {
2016 do code += GET(code, 1); while (*code == OP_ALT);
2017 c = *code;
2018 continue;
2019 }
2020
2021 /* For other groups, scan the branches. */
2022
2023 if (c == OP_BRA || c == OP_BRAPOS ||
2024 c == OP_CBRA || c == OP_CBRAPOS ||
2025 c == OP_ONCE || c == OP_COND)
2026 {
2027 BOOL empty_branch;
2028 if (GET(code, 1) == 0) return TRUE; /* Hit unclosed bracket */
2029
2030 /* If a conditional group has only one branch, there is a second, implied,
2031 empty branch, so just skip over the conditional, because it could be empty.
2032 Otherwise, scan the individual branches of the group. */
2033
2034 if (c == OP_COND && code[GET(code, 1)] != OP_ALT)
2035 code += GET(code, 1);
2036 else
2037 {
2038 empty_branch = FALSE;
2039 do
2040 {
2041 if (!empty_branch && could_be_empty_branch(code, endcode, utf8, cd))
2042 empty_branch = TRUE;
2043 code += GET(code, 1);
2044 }
2045 while (*code == OP_ALT);
2046 if (!empty_branch) return FALSE; /* All branches are non-empty */
2047 }
2048
2049 c = *code;
2050 continue;
2051 }
2052
2053 /* Handle the other opcodes */
2054
2055 switch (c)
2056 {
2057 /* Check for quantifiers after a class. XCLASS is used for classes that
2058 cannot be represented just by a bit map. This includes negated single
2059 high-valued characters. The length in _pcre_OP_lengths[] is zero; the
2060 actual length is stored in the compiled code, so we must update "code"
2061 here. */
2062
2063 #ifdef SUPPORT_UTF8
2064 case OP_XCLASS:
2065 ccode = code += GET(code, 1);
2066 goto CHECK_CLASS_REPEAT;
2067 #endif
2068
2069 case OP_CLASS:
2070 case OP_NCLASS:
2071 ccode = code + 33;
2072
2073 #ifdef SUPPORT_UTF8
2074 CHECK_CLASS_REPEAT:
2075 #endif
2076
2077 switch (*ccode)
2078 {
2079 case OP_CRSTAR: /* These could be empty; continue */
2080 case OP_CRMINSTAR:
2081 case OP_CRQUERY:
2082 case OP_CRMINQUERY:
2083 break;
2084
2085 default: /* Non-repeat => class must match */
2086 case OP_CRPLUS: /* These repeats aren't empty */
2087 case OP_CRMINPLUS:
2088 return FALSE;
2089
2090 case OP_CRRANGE:
2091 case OP_CRMINRANGE:
2092 if (GET2(ccode, 1) > 0) return FALSE; /* Minimum > 0 */
2093 break;
2094 }
2095 break;
2096
2097 /* Opcodes that must match a character */
2098
2099 case OP_PROP:
2100 case OP_NOTPROP:
2101 case OP_EXTUNI:
2102 case OP_NOT_DIGIT:
2103 case OP_DIGIT:
2104 case OP_NOT_WHITESPACE:
2105 case OP_WHITESPACE:
2106 case OP_NOT_WORDCHAR:
2107 case OP_WORDCHAR:
2108 case OP_ANY:
2109 case OP_ALLANY:
2110 case OP_ANYBYTE:
2111 case OP_CHAR:
2112 case OP_CHARI:
2113 case OP_NOT:
2114 case OP_NOTI:
2115 case OP_PLUS:
2116 case OP_MINPLUS:
2117 case OP_POSPLUS:
2118 case OP_EXACT:
2119 case OP_NOTPLUS:
2120 case OP_NOTMINPLUS:
2121 case OP_NOTPOSPLUS:
2122 case OP_NOTEXACT:
2123 case OP_TYPEPLUS:
2124 case OP_TYPEMINPLUS:
2125 case OP_TYPEPOSPLUS:
2126 case OP_TYPEEXACT:
2127 return FALSE;
2128
2129 /* These are going to continue, as they may be empty, but we have to
2130 fudge the length for the \p and \P cases. */
2131
2132 case OP_TYPESTAR:
2133 case OP_TYPEMINSTAR:
2134 case OP_TYPEPOSSTAR:
2135 case OP_TYPEQUERY:
2136 case OP_TYPEMINQUERY:
2137 case OP_TYPEPOSQUERY:
2138 if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
2139 break;
2140
2141 /* Same for these */
2142
2143 case OP_TYPEUPTO:
2144 case OP_TYPEMINUPTO:
2145 case OP_TYPEPOSUPTO:
2146 if (code[3] == OP_PROP || code[3] == OP_NOTPROP) code += 2;
2147 break;
2148
2149 /* End of branch */
2150
2151 case OP_KET:
2152 case OP_KETRMAX:
2153 case OP_KETRMIN:
2154 case OP_KETRPOS:
2155 case OP_ALT:
2156 return TRUE;
2157
2158 /* In UTF-8 mode, STAR, MINSTAR, POSSTAR, QUERY, MINQUERY, POSQUERY, UPTO,
2159 MINUPTO, and POSUPTO may be followed by a multibyte character */
2160
2161 #ifdef SUPPORT_UTF8
2162 case OP_STAR:
2163 case OP_STARI:
2164 case OP_MINSTAR:
2165 case OP_MINSTARI:
2166 case OP_POSSTAR:
2167 case OP_POSSTARI:
2168 case OP_QUERY:
2169 case OP_QUERYI:
2170 case OP_MINQUERY:
2171 case OP_MINQUERYI:
2172 case OP_POSQUERY:
2173 case OP_POSQUERYI:
2174 if (utf8 && code[1] >= 0xc0) code += _pcre_utf8_table4[code[1] & 0x3f];
2175 break;
2176
2177 case OP_UPTO:
2178 case OP_UPTOI:
2179 case OP_MINUPTO:
2180 case OP_MINUPTOI:
2181 case OP_POSUPTO:
2182 case OP_POSUPTOI:
2183 if (utf8 && code[3] >= 0xc0) code += _pcre_utf8_table4[code[3] & 0x3f];
2184 break;
2185 #endif
2186
2187 /* MARK, and PRUNE/SKIP/THEN with an argument must skip over the argument
2188 string. */
2189
2190 case OP_MARK:
2191 case OP_PRUNE_ARG:
2192 case OP_SKIP_ARG:
2193 code += code[1];
2194 break;
2195
2196 case OP_THEN_ARG:
2197 code += code[1+LINK_SIZE];
2198 break;
2199
2200 /* None of the remaining opcodes are required to match a character. */
2201
2202 default:
2203 break;
2204 }
2205 }
2206
2207 return TRUE;
2208 }
2209
2210
2211
2212 /*************************************************
2213 * Scan compiled regex for non-emptiness *
2214 *************************************************/
2215
2216 /* This function is called to check for left recursive calls. We want to check
2217 the current branch of the current pattern to see if it could match the empty
2218 string. If it could, we must look outwards for branches at other levels,
2219 stopping when we pass beyond the bracket which is the subject of the recursion.
2220
2221 Arguments:
2222 code points to start of the recursion
2223 endcode points to where to stop (current RECURSE item)
2224 bcptr points to the chain of current (unclosed) branch starts
2225 utf8 TRUE if in UTF-8 mode
2226 cd pointers to tables etc
2227
2228 Returns: TRUE if what is matched could be empty
2229 */
2230
2231 static BOOL
2232 could_be_empty(const uschar *code, const uschar *endcode, branch_chain *bcptr,
2233 BOOL utf8, compile_data *cd)
2234 {
2235 while (bcptr != NULL && bcptr->current_branch >= code)
2236 {
2237 if (!could_be_empty_branch(bcptr->current_branch, endcode, utf8, cd))
2238 return FALSE;
2239 bcptr = bcptr->outer;
2240 }
2241 return TRUE;
2242 }
2243
2244
2245
2246 /*************************************************
2247 * Check for POSIX class syntax *
2248 *************************************************/
2249
2250 /* This function is called when the sequence "[:" or "[." or "[=" is
2251 encountered in a character class. It checks whether this is followed by a
2252 sequence of characters terminated by a matching ":]" or ".]" or "=]". If we
2253 reach an unescaped ']' without the special preceding character, return FALSE.
2254
2255 Originally, this function only recognized a sequence of letters between the
2256 terminators, but it seems that Perl recognizes any sequence of characters,
2257 though of course unknown POSIX names are subsequently rejected. Perl gives an
2258 "Unknown POSIX class" error for [:f\oo:] for example, where previously PCRE
2259 didn't consider this to be a POSIX class. Likewise for [:1234:].
2260
2261 The problem in trying to be exactly like Perl is in the handling of escapes. We
2262 have to be sure that [abc[:x\]pqr] is *not* treated as containing a POSIX
2263 class, but [abc[:x\]pqr:]] is (so that an error can be generated). The code
2264 below handles the special case of \], but does not try to do any other escape
2265 processing. This makes it different from Perl for cases such as [:l\ower:]
2266 where Perl recognizes it as the POSIX class "lower" but PCRE does not recognize
2267 "l\ower". This is a lesser evil that not diagnosing bad classes when Perl does,
2268 I think.
2269
2270 Arguments:
2271 ptr pointer to the initial [
2272 endptr where to return the end pointer
2273
2274 Returns: TRUE or FALSE
2275 */
2276
2277 static BOOL
2278 check_posix_syntax(const uschar *ptr, const uschar **endptr)
2279 {
2280 int terminator; /* Don't combine these lines; the Solaris cc */
2281 terminator = *(++ptr); /* compiler warns about "non-constant" initializer. */
2282 for (++ptr; *ptr != 0; ptr++)
2283 {
2284 if (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET) ptr++; else
2285 {
2286 if (*ptr == CHAR_RIGHT_SQUARE_BRACKET) return FALSE;
2287 if (*ptr == terminator && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET)
2288 {
2289 *endptr = ptr;
2290 return TRUE;
2291 }
2292 }
2293 }
2294 return FALSE;
2295 }
2296
2297
2298
2299
2300 /*************************************************
2301 * Check POSIX class name *
2302 *************************************************/
2303
2304 /* This function is called to check the name given in a POSIX-style class entry
2305 such as [:alnum:].
2306
2307 Arguments:
2308 ptr points to the first letter
2309 len the length of the name
2310
2311 Returns: a value representing the name, or -1 if unknown
2312 */
2313
2314 static int
2315 check_posix_name(const uschar *ptr, int len)
2316 {
2317 const char *pn = posix_names;
2318 register int yield = 0;
2319 while (posix_name_lengths[yield] != 0)
2320 {
2321 if (len == posix_name_lengths[yield] &&
2322 strncmp((const char *)ptr, pn, len) == 0) return yield;
2323 pn += posix_name_lengths[yield] + 1;
2324 yield++;
2325 }
2326 return -1;
2327 }
2328
2329
2330 /*************************************************
2331 * Adjust OP_RECURSE items in repeated group *
2332 *************************************************/
2333
2334 /* OP_RECURSE items contain an offset from the start of the regex to the group
2335 that is referenced. This means that groups can be replicated for fixed
2336 repetition simply by copying (because the recursion is allowed to refer to
2337 earlier groups that are outside the current group). However, when a group is
2338 optional (i.e. the minimum quantifier is zero), OP_BRAZERO or OP_SKIPZERO is
2339 inserted before it, after it has been compiled. This means that any OP_RECURSE
2340 items within it that refer to the group itself or any contained groups have to
2341 have their offsets adjusted. That one of the jobs of this function. Before it
2342 is called, the partially compiled regex must be temporarily terminated with
2343 OP_END.
2344
2345 This function has been extended with the possibility of forward references for
2346 recursions and subroutine calls. It must also check the list of such references
2347 for the group we are dealing with. If it finds that one of the recursions in
2348 the current group is on this list, it adjusts the offset in the list, not the
2349 value in the reference (which is a group number).
2350
2351 Arguments:
2352 group points to the start of the group
2353 adjust the amount by which the group is to be moved
2354 utf8 TRUE in UTF-8 mode
2355 cd contains pointers to tables etc.
2356 save_hwm the hwm forward reference pointer at the start of the group
2357
2358 Returns: nothing
2359 */
2360
2361 static void
2362 adjust_recurse(uschar *group, int adjust, BOOL utf8, compile_data *cd,
2363 uschar *save_hwm)
2364 {
2365 uschar *ptr = group;
2366
2367 while ((ptr = (uschar *)find_recurse(ptr, utf8)) != NULL)
2368 {
2369 int offset;
2370 uschar *hc;
2371
2372 /* See if this recursion is on the forward reference list. If so, adjust the
2373 reference. */
2374
2375 for (hc = save_hwm; hc < cd->hwm; hc += LINK_SIZE)
2376 {
2377 offset = GET(hc, 0);
2378 if (cd->start_code + offset == ptr + 1)
2379 {
2380 PUT(hc, 0, offset + adjust);
2381 break;
2382 }
2383 }
2384
2385 /* Otherwise, adjust the recursion offset if it's after the start of this
2386 group. */
2387
2388 if (hc >= cd->hwm)
2389 {
2390 offset = GET(ptr, 1);
2391 if (cd->start_code + offset >= group) PUT(ptr, 1, offset + adjust);
2392 }
2393
2394 ptr += 1 + LINK_SIZE;
2395 }
2396 }
2397
2398
2399
2400 /*************************************************
2401 * Insert an automatic callout point *
2402 *************************************************/
2403
2404 /* This function is called when the PCRE_AUTO_CALLOUT option is set, to insert
2405 callout points before each pattern item.
2406
2407 Arguments:
2408 code current code pointer
2409 ptr current pattern pointer
2410 cd pointers to tables etc
2411
2412 Returns: new code pointer
2413 */
2414
2415 static uschar *
2416 auto_callout(uschar *code, const uschar *ptr, compile_data *cd)
2417 {
2418 *code++ = OP_CALLOUT;
2419 *code++ = 255;
2420 PUT(code, 0, (int)(ptr - cd->start_pattern)); /* Pattern offset */
2421 PUT(code, LINK_SIZE, 0); /* Default length */
2422 return code + 2*LINK_SIZE;
2423 }
2424
2425
2426
2427 /*************************************************
2428 * Complete a callout item *
2429 *************************************************/
2430
2431 /* A callout item contains the length of the next item in the pattern, which
2432 we can't fill in till after we have reached the relevant point. This is used
2433 for both automatic and manual callouts.
2434
2435 Arguments:
2436 previous_callout points to previous callout item
2437 ptr current pattern pointer
2438 cd pointers to tables etc
2439
2440 Returns: nothing
2441 */
2442
2443 static void
2444 complete_callout(uschar *previous_callout, const uschar *ptr, compile_data *cd)
2445 {
2446 int length = (int)(ptr - cd->start_pattern - GET(previous_callout, 2));
2447 PUT(previous_callout, 2 + LINK_SIZE, length);
2448 }
2449
2450
2451
2452 #ifdef SUPPORT_UCP
2453 /*************************************************
2454 * Get othercase range *
2455 *************************************************/
2456
2457 /* This function is passed the start and end of a class range, in UTF-8 mode
2458 with UCP support. It searches up the characters, looking for internal ranges of
2459 characters in the "other" case. Each call returns the next one, updating the
2460 start address.
2461
2462 Arguments:
2463 cptr points to starting character value; updated
2464 d end value
2465 ocptr where to put start of othercase range
2466 odptr where to put end of othercase range
2467
2468 Yield: TRUE when range returned; FALSE when no more
2469 */
2470
2471 static BOOL
2472 get_othercase_range(unsigned int *cptr, unsigned int d, unsigned int *ocptr,
2473 unsigned int *odptr)
2474 {
2475 unsigned int c, othercase, next;
2476
2477 for (c = *cptr; c <= d; c++)
2478 { if ((othercase = UCD_OTHERCASE(c)) != c) break; }
2479
2480 if (c > d) return FALSE;
2481
2482 *ocptr = othercase;
2483 next = othercase + 1;
2484
2485 for (++c; c <= d; c++)
2486 {
2487 if (UCD_OTHERCASE(c) != next) break;
2488 next++;
2489 }
2490
2491 *odptr = next - 1;
2492 *cptr = c;
2493
2494 return TRUE;
2495 }
2496
2497
2498
2499 /*************************************************
2500 * Check a character and a property *
2501 *************************************************/
2502
2503 /* This function is called by check_auto_possessive() when a property item
2504 is adjacent to a fixed character.
2505
2506 Arguments:
2507 c the character
2508 ptype the property type
2509 pdata the data for the type
2510 negated TRUE if it's a negated property (\P or \p{^)
2511
2512 Returns: TRUE if auto-possessifying is OK
2513 */
2514
2515 static BOOL
2516 check_char_prop(int c, int ptype, int pdata, BOOL negated)
2517 {
2518 const ucd_record *prop = GET_UCD(c);
2519 switch(ptype)
2520 {
2521 case PT_LAMP:
2522 return (prop->chartype == ucp_Lu ||
2523 prop->chartype == ucp_Ll ||
2524 prop->chartype == ucp_Lt) == negated;
2525
2526 case PT_GC:
2527 return (pdata == _pcre_ucp_gentype[prop->chartype]) == negated;
2528
2529 case PT_PC:
2530 return (pdata == prop->chartype) == negated;
2531
2532 case PT_SC:
2533 return (pdata == prop->script) == negated;
2534
2535 /* These are specials */
2536
2537 case PT_ALNUM:
2538 return (_pcre_ucp_gentype[prop->chartype] == ucp_L ||
2539 _pcre_ucp_gentype[prop->chartype] == ucp_N) == negated;
2540
2541 case PT_SPACE: /* Perl space */
2542 return (_pcre_ucp_gentype[prop->chartype] == ucp_Z ||
2543 c == CHAR_HT || c == CHAR_NL || c == CHAR_FF || c == CHAR_CR)
2544 == negated;
2545
2546 case PT_PXSPACE: /* POSIX space */
2547 return (_pcre_ucp_gentype[prop->chartype] == ucp_Z ||
2548 c == CHAR_HT || c == CHAR_NL || c == CHAR_VT ||
2549 c == CHAR_FF || c == CHAR_CR)
2550 == negated;
2551
2552 case PT_WORD:
2553 return (_pcre_ucp_gentype[prop->chartype] == ucp_L ||
2554 _pcre_ucp_gentype[prop->chartype] == ucp_N ||
2555 c == CHAR_UNDERSCORE) == negated;
2556 }
2557 return FALSE;
2558 }
2559 #endif /* SUPPORT_UCP */
2560
2561
2562
2563 /*************************************************
2564 * Check if auto-possessifying is possible *
2565 *************************************************/
2566
2567 /* This function is called for unlimited repeats of certain items, to see
2568 whether the next thing could possibly match the repeated item. If not, it makes
2569 sense to automatically possessify the repeated item.
2570
2571 Arguments:
2572 previous pointer to the repeated opcode
2573 utf8 TRUE in UTF-8 mode
2574 ptr next character in pattern
2575 options options bits
2576 cd contains pointers to tables etc.
2577
2578 Returns: TRUE if possessifying is wanted
2579 */
2580
2581 static BOOL
2582 check_auto_possessive(const uschar *previous, BOOL utf8, const uschar *ptr,
2583 int options, compile_data *cd)
2584 {
2585 int c, next;
2586 int op_code = *previous++;
2587
2588 /* Skip whitespace and comments in extended mode */
2589
2590 if ((options & PCRE_EXTENDED) != 0)
2591 {
2592 for (;;)
2593 {
2594 while ((cd->ctypes[*ptr] & ctype_space) != 0) ptr++;
2595 if (*ptr == CHAR_NUMBER_SIGN)
2596 {
2597 ptr++;
2598 while (*ptr != 0)
2599 {
2600 if (IS_NEWLINE(ptr)) { ptr += cd->nllen; break; }
2601 ptr++;
2602 #ifdef SUPPORT_UTF8
2603 if (utf8) while ((*ptr & 0xc0) == 0x80) ptr++;
2604 #endif
2605 }
2606 }
2607 else break;
2608 }
2609 }
2610
2611 /* If the next item is one that we can handle, get its value. A non-negative
2612 value is a character, a negative value is an escape value. */
2613
2614 if (*ptr == CHAR_BACKSLASH)
2615 {
2616 int temperrorcode = 0;
2617 next = check_escape(&ptr, &temperrorcode, cd->bracount, options, FALSE);
2618 if (temperrorcode != 0) return FALSE;
2619 ptr++; /* Point after the escape sequence */
2620 }
2621
2622 else if ((cd->ctypes[*ptr] & ctype_meta) == 0)
2623 {
2624 #ifdef SUPPORT_UTF8
2625 if (utf8) { GETCHARINC(next, ptr); } else
2626 #endif
2627 next = *ptr++;
2628 }
2629
2630 else return FALSE;
2631
2632 /* Skip whitespace and comments in extended mode */
2633
2634 if ((options & PCRE_EXTENDED) != 0)
2635 {
2636 for (;;)
2637 {
2638 while ((cd->ctypes[*ptr] & ctype_space) != 0) ptr++;
2639 if (*ptr == CHAR_NUMBER_SIGN)
2640 {
2641 ptr++;
2642 while (*ptr != 0)
2643 {
2644 if (IS_NEWLINE(ptr)) { ptr += cd->nllen; break; }
2645 ptr++;
2646 #ifdef SUPPORT_UTF8
2647 if (utf8) while ((*ptr & 0xc0) == 0x80) ptr++;
2648 #endif
2649 }
2650 }
2651 else break;
2652 }
2653 }
2654
2655 /* If the next thing is itself optional, we have to give up. */
2656
2657 if (*ptr == CHAR_ASTERISK || *ptr == CHAR_QUESTION_MARK ||
2658 strncmp((char *)ptr, STR_LEFT_CURLY_BRACKET STR_0 STR_COMMA, 3) == 0)
2659 return FALSE;
2660
2661 /* Now compare the next item with the previous opcode. First, handle cases when
2662 the next item is a character. */
2663
2664 if (next >= 0) switch(op_code)
2665 {
2666 case OP_CHAR:
2667 #ifdef SUPPORT_UTF8
2668 GETCHARTEST(c, previous);
2669 #else
2670 c = *previous;
2671 #endif
2672 return c != next;
2673
2674 /* For CHARI (caseless character) we must check the other case. If we have
2675 Unicode property support, we can use it to test the other case of
2676 high-valued characters. */
2677
2678 case OP_CHARI:
2679 #ifdef SUPPORT_UTF8
2680 GETCHARTEST(c, previous);
2681 #else
2682 c = *previous;
2683 #endif
2684 if (c == next) return FALSE;
2685 #ifdef SUPPORT_UTF8
2686 if (utf8)
2687 {
2688 unsigned int othercase;
2689 if (next < 128) othercase = cd->fcc[next]; else
2690 #ifdef SUPPORT_UCP
2691 othercase = UCD_OTHERCASE((unsigned int)next);
2692 #else
2693 othercase = NOTACHAR;
2694 #endif
2695 return (unsigned int)c != othercase;
2696 }
2697 else
2698 #endif /* SUPPORT_UTF8 */
2699 return (c != cd->fcc[next]); /* Non-UTF-8 mode */
2700
2701 /* For OP_NOT and OP_NOTI, the data is always a single-byte character. These
2702 opcodes are not used for multi-byte characters, because they are coded using
2703 an XCLASS instead. */
2704
2705 case OP_NOT:
2706 return (c = *previous) == next;
2707
2708 case OP_NOTI:
2709 if ((c = *previous) == next) return TRUE;
2710 #ifdef SUPPORT_UTF8
2711 if (utf8)
2712 {
2713 unsigned int othercase;
2714 if (next < 128) othercase = cd->fcc[next]; else
2715 #ifdef SUPPORT_UCP
2716 othercase = UCD_OTHERCASE(next);
2717 #else
2718 othercase = NOTACHAR;
2719 #endif
2720 return (unsigned int)c == othercase;
2721 }
2722 else
2723 #endif /* SUPPORT_UTF8 */
2724 return (c == cd->fcc[next]); /* Non-UTF-8 mode */
2725
2726 /* Note that OP_DIGIT etc. are generated only when PCRE_UCP is *not* set.
2727 When it is set, \d etc. are converted into OP_(NOT_)PROP codes. */
2728
2729 case OP_DIGIT:
2730 return next > 127 || (cd->ctypes[next] & ctype_digit) == 0;
2731
2732 case OP_NOT_DIGIT:
2733 return next <= 127 && (cd->ctypes[next] & ctype_digit) != 0;
2734
2735 case OP_WHITESPACE:
2736 return next > 127 || (cd->ctypes[next] & ctype_space) == 0;
2737
2738 case OP_NOT_WHITESPACE:
2739 return next <= 127 && (cd->ctypes[next] & ctype_space) != 0;
2740
2741 case OP_WORDCHAR:
2742 return next > 127 || (cd->ctypes[next] & ctype_word) == 0;
2743
2744 case OP_NOT_WORDCHAR:
2745 return next <= 127 && (cd->ctypes[next] & ctype_word) != 0;
2746
2747 case OP_HSPACE:
2748 case OP_NOT_HSPACE:
2749 switch(next)
2750 {
2751 case 0x09:
2752 case 0x20:
2753 case 0xa0:
2754 case 0x1680:
2755 case 0x180e:
2756 case 0x2000:
2757 case 0x2001:
2758 case 0x2002:
2759 case 0x2003:
2760 case 0x2004:
2761 case 0x2005:
2762 case 0x2006:
2763 case 0x2007:
2764 case 0x2008:
2765 case 0x2009:
2766 case 0x200A:
2767 case 0x202f:
2768 case 0x205f:
2769 case 0x3000:
2770 return op_code == OP_NOT_HSPACE;
2771 default:
2772 return op_code != OP_NOT_HSPACE;
2773 }
2774
2775 case OP_ANYNL:
2776 case OP_VSPACE:
2777 case OP_NOT_VSPACE:
2778 switch(next)
2779 {
2780 case 0x0a:
2781 case 0x0b:
2782 case 0x0c:
2783 case 0x0d:
2784 case 0x85:
2785 case 0x2028:
2786 case 0x2029:
2787 return op_code == OP_NOT_VSPACE;
2788 default:
2789 return op_code != OP_NOT_VSPACE;
2790 }
2791
2792 #ifdef SUPPORT_UCP
2793 case OP_PROP:
2794 return check_char_prop(next, previous[0], previous[1], FALSE);
2795
2796 case OP_NOTPROP:
2797 return check_char_prop(next, previous[0], previous[1], TRUE);
2798 #endif
2799
2800 default:
2801 return FALSE;
2802 }
2803
2804
2805 /* Handle the case when the next item is \d, \s, etc. Note that when PCRE_UCP
2806 is set, \d turns into ESC_du rather than ESC_d, etc., so ESC_d etc. are
2807 generated only when PCRE_UCP is *not* set, that is, when only ASCII
2808 characteristics are recognized. Similarly, the opcodes OP_DIGIT etc. are
2809 replaced by OP_PROP codes when PCRE_UCP is set. */
2810
2811 switch(op_code)
2812 {
2813 case OP_CHAR:
2814 case OP_CHARI:
2815 #ifdef SUPPORT_UTF8
2816 GETCHARTEST(c, previous);
2817 #else
2818 c = *previous;
2819 #endif
2820 switch(-next)
2821 {
2822 case ESC_d:
2823 return c > 127 || (cd->ctypes[c] & ctype_digit) == 0;
2824
2825 case ESC_D:
2826 return c <= 127 && (cd->ctypes[c] & ctype_digit) != 0;
2827
2828 case ESC_s:
2829 return c > 127 || (cd->ctypes[c] & ctype_space) == 0;
2830
2831 case ESC_S:
2832 return c <= 127 && (cd->ctypes[c] & ctype_space) != 0;
2833
2834 case ESC_w:
2835 return c > 127 || (cd->ctypes[c] & ctype_word) == 0;
2836
2837 case ESC_W:
2838 return c <= 127 && (cd->ctypes[c] & ctype_word) != 0;
2839
2840 case ESC_h:
2841 case ESC_H:
2842 switch(c)
2843 {
2844 case 0x09:
2845 case 0x20:
2846 case 0xa0:
2847 case 0x1680:
2848 case 0x180e:
2849 case 0x2000:
2850 case 0x2001:
2851 case 0x2002:
2852 case 0x2003:
2853 case 0x2004:
2854 case 0x2005:
2855 case 0x2006:
2856 case 0x2007:
2857 case 0x2008:
2858 case 0x2009:
2859 case 0x200A:
2860 case 0x202f:
2861 case 0x205f:
2862 case 0x3000:
2863 return -next != ESC_h;
2864 default:
2865 return -next == ESC_h;
2866 }
2867
2868 case ESC_v:
2869 case ESC_V:
2870 switch(c)
2871 {
2872 case 0x0a:
2873 case 0x0b:
2874 case 0x0c:
2875 case 0x0d:
2876 case 0x85:
2877 case 0x2028:
2878 case 0x2029:
2879 return -next != ESC_v;
2880 default:
2881 return -next == ESC_v;
2882 }
2883
2884 /* When PCRE_UCP is set, these values get generated for \d etc. Find
2885 their substitutions and process them. The result will always be either
2886 -ESC_p or -ESC_P. Then fall through to process those values. */
2887
2888 #ifdef SUPPORT_UCP
2889 case ESC_du:
2890 case ESC_DU:
2891 case ESC_wu:
2892 case ESC_WU:
2893 case ESC_su:
2894 case ESC_SU:
2895 {
2896 int temperrorcode = 0;
2897 ptr = substitutes[-next - ESC_DU];
2898 next = check_escape(&ptr, &temperrorcode, 0, options, FALSE);
2899 if (temperrorcode != 0) return FALSE;
2900 ptr++; /* For compatibility */
2901 }
2902 /* Fall through */
2903
2904 case ESC_p:
2905 case ESC_P:
2906 {
2907 int ptype, pdata, errorcodeptr;
2908 BOOL negated;
2909
2910 ptr--; /* Make ptr point at the p or P */
2911 ptype = get_ucp(&ptr, &negated, &pdata, &errorcodeptr);
2912 if (ptype < 0) return FALSE;
2913 ptr++; /* Point past the final curly ket */
2914
2915 /* If the property item is optional, we have to give up. (When generated
2916 from \d etc by PCRE_UCP, this test will have been applied much earlier,
2917 to the original \d etc. At this point, ptr will point to a zero byte. */
2918
2919 if (*ptr == CHAR_ASTERISK || *ptr == CHAR_QUESTION_MARK ||
2920 strncmp((char *)ptr, STR_LEFT_CURLY_BRACKET STR_0 STR_COMMA, 3) == 0)
2921 return FALSE;
2922
2923 /* Do the property check. */
2924
2925 return check_char_prop(c, ptype, pdata, (next == -ESC_P) != negated);
2926 }
2927 #endif
2928
2929 default:
2930 return FALSE;
2931 }
2932
2933 /* In principle, support for Unicode properties should be integrated here as
2934 well. It means re-organizing the above code so as to get hold of the property
2935 values before switching on the op-code. However, I wonder how many patterns
2936 combine ASCII \d etc with Unicode properties? (Note that if PCRE_UCP is set,
2937 these op-codes are never generated.) */
2938
2939 case OP_DIGIT:
2940 return next == -ESC_D || next == -ESC_s || next == -ESC_W ||
2941 next == -ESC_h || next == -ESC_v || next == -ESC_R;
2942
2943 case OP_NOT_DIGIT:
2944 return next == -ESC_d;
2945
2946 case OP_WHITESPACE:
2947 return next == -ESC_S || next == -ESC_d || next == -ESC_w || next == -ESC_R;
2948
2949 case OP_NOT_WHITESPACE:
2950 return next == -ESC_s || next == -ESC_h || next == -ESC_v;
2951
2952 case OP_HSPACE:
2953 return next == -ESC_S || next == -ESC_H || next == -ESC_d ||
2954 next == -ESC_w || next == -ESC_v || next == -ESC_R;
2955
2956 case OP_NOT_HSPACE:
2957 return next == -ESC_h;
2958
2959 /* Can't have \S in here because VT matches \S (Perl anomaly) */
2960 case OP_ANYNL:
2961 case OP_VSPACE:
2962 return next == -ESC_V || next == -ESC_d || next == -ESC_w;
2963
2964 case OP_NOT_VSPACE:
2965 return next == -ESC_v || next == -ESC_R;
2966
2967 case OP_WORDCHAR:
2968 return next == -ESC_W || next == -ESC_s || next == -ESC_h ||
2969 next == -ESC_v || next == -ESC_R;
2970
2971 case OP_NOT_WORDCHAR:
2972 return next == -ESC_w || next == -ESC_d;
2973
2974 default:
2975 return FALSE;
2976 }
2977
2978 /* Control does not reach here */
2979 }
2980
2981
2982
2983 /*************************************************
2984 * Compile one branch *
2985 *************************************************/
2986
2987 /* Scan the pattern, compiling it into the a vector. If the options are
2988 changed during the branch, the pointer is used to change the external options
2989 bits. This function is used during the pre-compile phase when we are trying
2990 to find out the amount of memory needed, as well as during the real compile
2991 phase. The value of lengthptr distinguishes the two phases.
2992
2993 Arguments:
2994 optionsptr pointer to the option bits
2995 codeptr points to the pointer to the current code point
2996 ptrptr points to the current pattern pointer
2997 errorcodeptr points to error code variable
2998 firstbyteptr set to initial literal character, or < 0 (REQ_UNSET, REQ_NONE)
2999 reqbyteptr set to the last literal character required, else < 0
3000 bcptr points to current branch chain
3001 cd contains pointers to tables etc.
3002 lengthptr NULL during the real compile phase
3003 points to length accumulator during pre-compile phase
3004
3005 Returns: TRUE on success
3006 FALSE, with *errorcodeptr set non-zero on error
3007 */
3008
3009 static BOOL
3010 compile_branch(int *optionsptr, uschar **codeptr, const uschar **ptrptr,
3011 int *errorcodeptr, int *firstbyteptr, int *reqbyteptr, branch_chain *bcptr,
3012 compile_data *cd, int *lengthptr)
3013 {
3014 int repeat_type, op_type;
3015 int repeat_min = 0, repeat_max = 0; /* To please picky compilers */
3016 int bravalue = 0;
3017 int greedy_default, greedy_non_default;
3018 int firstbyte, reqbyte;
3019 int zeroreqbyte, zerofirstbyte;
3020 int req_caseopt, reqvary, tempreqvary;
3021 int options = *optionsptr;
3022 int after_manual_callout = 0;
3023 int length_prevgroup = 0;
3024 register int c;
3025 register uschar *code = *codeptr;
3026 uschar *last_code = code;
3027 uschar *orig_code = code;
3028 uschar *tempcode;
3029 BOOL inescq = FALSE;
3030 BOOL groupsetfirstbyte = FALSE;
3031 const uschar *ptr = *ptrptr;
3032 const uschar *tempptr;
3033 const uschar *nestptr = NULL;
3034 uschar *previous = NULL;
3035 uschar *previous_callout = NULL;
3036 uschar *save_hwm = NULL;
3037 uschar classbits[32];
3038
3039 #ifdef SUPPORT_UTF8
3040 BOOL class_utf8;
3041 BOOL utf8 = (options & PCRE_UTF8) != 0;
3042 uschar *class_utf8data;
3043 uschar *class_utf8data_base;
3044 uschar utf8_char[6];
3045 #else
3046 BOOL utf8 = FALSE;
3047 uschar *utf8_char = NULL;
3048 #endif
3049
3050 #ifdef PCRE_DEBUG
3051 if (lengthptr != NULL) DPRINTF((">> start branch\n"));
3052 #endif
3053
3054 /* Set up the default and non-default settings for greediness */
3055
3056 greedy_default = ((options & PCRE_UNGREEDY) != 0);
3057 greedy_non_default = greedy_default ^ 1;
3058
3059 /* Initialize no first byte, no required byte. REQ_UNSET means "no char
3060 matching encountered yet". It gets changed to REQ_NONE if we hit something that
3061 matches a non-fixed char first char; reqbyte just remains unset if we never
3062 find one.
3063
3064 When we hit a repeat whose minimum is zero, we may have to adjust these values
3065 to take the zero repeat into account. This is implemented by setting them to
3066 zerofirstbyte and zeroreqbyte when such a repeat is encountered. The individual
3067 item types that can be repeated set these backoff variables appropriately. */
3068
3069 firstbyte = reqbyte = zerofirstbyte = zeroreqbyte = REQ_UNSET;
3070
3071 /* The variable req_caseopt contains either the REQ_CASELESS value or zero,
3072 according to the current setting of the caseless flag. REQ_CASELESS is a bit
3073 value > 255. It is added into the firstbyte or reqbyte variables to record the
3074 case status of the value. This is used only for ASCII characters. */
3075
3076 req_caseopt = ((options & PCRE_CASELESS) != 0)? REQ_CASELESS : 0;
3077
3078 /* Switch on next character until the end of the branch */
3079
3080 for (;; ptr++)
3081 {
3082 BOOL negate_class;
3083 BOOL should_flip_negation;
3084 BOOL possessive_quantifier;
3085 BOOL is_quantifier;
3086 BOOL is_recurse;
3087 BOOL reset_bracount;
3088 int class_charcount;
3089 int class_lastchar;
3090 int newoptions;
3091 int recno;
3092 int refsign;
3093 int skipbytes;
3094 int subreqbyte;
3095 int subfirstbyte;
3096 int terminator;
3097 int mclength;
3098 uschar mcbuffer[8];
3099
3100 /* Get next byte in the pattern */
3101
3102 c = *ptr;
3103
3104 /* If we are at the end of a nested substitution, revert to the outer level
3105 string. Nesting only happens one level deep. */
3106
3107 if (c == 0 && nestptr != NULL)
3108 {
3109 ptr = nestptr;
3110 nestptr = NULL;
3111 c = *ptr;
3112 }
3113
3114 /* If we are in the pre-compile phase, accumulate the length used for the
3115 previous cycle of this loop. */
3116
3117 if (lengthptr != NULL)
3118 {
3119 #ifdef PCRE_DEBUG
3120 if (code > cd->hwm) cd->hwm = code; /* High water info */
3121 #endif
3122 if (code > cd->start_workspace + WORK_SIZE_CHECK) /* Check for overrun */
3123 {
3124 *errorcodeptr = ERR52;
3125 goto FAILED;
3126 }
3127
3128 /* There is at least one situation where code goes backwards: this is the
3129 case of a zero quantifier after a class (e.g. [ab]{0}). At compile time,
3130 the class is simply eliminated. However, it is created first, so we have to
3131 allow memory for it. Therefore, don't ever reduce the length at this point.
3132 */
3133
3134 if (code < last_code) code = last_code;
3135
3136 /* Paranoid check for integer overflow */
3137
3138 if (OFLOW_MAX - *lengthptr < code - last_code)
3139 {
3140 *errorcodeptr = ERR20;
3141 goto FAILED;
3142 }
3143
3144 *lengthptr += (int)(code - last_code);
3145 DPRINTF(("length=%d added %d c=%c\n", *lengthptr, code - last_code, c));
3146
3147 /* If "previous" is set and it is not at the start of the work space, move
3148 it back to there, in order to avoid filling up the work space. Otherwise,
3149 if "previous" is NULL, reset the current code pointer to the start. */
3150
3151 if (previous != NULL)
3152 {
3153 if (previous > orig_code)
3154 {
3155 memmove(orig_code, previous, code - previous);
3156 code -= previous - orig_code;
3157 previous = orig_code;
3158 }
3159 }
3160 else code = orig_code;
3161
3162 /* Remember where this code item starts so we can pick up the length
3163 next time round. */
3164
3165 last_code = code;
3166 }
3167
3168 /* In the real compile phase, just check the workspace used by the forward
3169 reference list. */
3170
3171 else if (cd->hwm > cd->start_workspace + WORK_SIZE_CHECK)
3172 {
3173 *errorcodeptr = ERR52;
3174 goto FAILED;
3175 }
3176
3177 /* If in \Q...\E, check for the end; if not, we have a literal */
3178
3179 if (inescq && c != 0)
3180 {
3181 if (c == CHAR_BACKSLASH && ptr[1] == CHAR_E)
3182 {
3183 inescq = FALSE;
3184 ptr++;
3185 continue;
3186 }
3187 else
3188 {
3189 if (previous_callout != NULL)
3190 {
3191 if (lengthptr == NULL) /* Don't attempt in pre-compile phase */
3192 complete_callout(previous_callout, ptr, cd);
3193 previous_callout = NULL;
3194 }
3195 if ((options & PCRE_AUTO_CALLOUT) != 0)
3196 {
3197 previous_callout = code;
3198 code = auto_callout(code, ptr, cd);
3199 }
3200 goto NORMAL_CHAR;
3201 }
3202 }
3203
3204 /* Fill in length of a previous callout, except when the next thing is
3205 a quantifier. */
3206
3207 is_quantifier =
3208 c == CHAR_ASTERISK || c == CHAR_PLUS || c == CHAR_QUESTION_MARK ||
3209 (c == CHAR_LEFT_CURLY_BRACKET && is_counted_repeat(ptr+1));
3210
3211 if (!is_quantifier && previous_callout != NULL &&
3212 after_manual_callout-- <= 0)
3213 {
3214 if (lengthptr == NULL) /* Don't attempt in pre-compile phase */
3215 complete_callout(previous_callout, ptr, cd);
3216 previous_callout = NULL;
3217 }
3218
3219 /* In extended mode, skip white space and comments */
3220
3221 if ((options & PCRE_EXTENDED) != 0)
3222 {
3223 if ((cd->ctypes[c] & ctype_space) != 0) continue;
3224 if (c == CHAR_NUMBER_SIGN)
3225 {
3226 ptr++;
3227 while (*ptr != 0)
3228 {
3229 if (IS_NEWLINE(ptr)) { ptr += cd->nllen - 1; break; }
3230 ptr++;
3231 #ifdef SUPPORT_UTF8
3232 if (utf8) while ((*ptr & 0xc0) == 0x80) ptr++;
3233 #endif
3234 }
3235 if (*ptr != 0) continue;
3236
3237 /* Else fall through to handle end of string */
3238 c = 0;
3239 }
3240 }
3241
3242 /* No auto callout for quantifiers. */
3243
3244 if ((options & PCRE_AUTO_CALLOUT) != 0 && !is_quantifier)
3245 {
3246 previous_callout = code;
3247 code = auto_callout(code, ptr, cd);
3248 }
3249
3250 switch(c)
3251 {
3252 /* ===================================================================*/
3253 case 0: /* The branch terminates at string end */
3254 case CHAR_VERTICAL_LINE: /* or | or ) */
3255 case CHAR_RIGHT_PARENTHESIS:
3256 *firstbyteptr = firstbyte;
3257 *reqbyteptr = reqbyte;
3258 *codeptr = code;
3259 *ptrptr = ptr;
3260 if (lengthptr != NULL)
3261 {
3262 if (OFLOW_MAX - *lengthptr < code - last_code)
3263 {
3264 *errorcodeptr = ERR20;
3265 goto FAILED;
3266 }
3267 *lengthptr += (int)(code - last_code); /* To include callout length */
3268 DPRINTF((">> end branch\n"));
3269 }
3270 return TRUE;
3271
3272
3273 /* ===================================================================*/
3274 /* Handle single-character metacharacters. In multiline mode, ^ disables
3275 the setting of any following char as a first character. */
3276
3277 case CHAR_CIRCUMFLEX_ACCENT:
3278 previous = NULL;
3279 if ((options & PCRE_MULTILINE) != 0)
3280 {
3281 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
3282 *code++ = OP_CIRCM;
3283 }
3284 else *code++ = OP_CIRC;
3285 break;
3286
3287 case CHAR_DOLLAR_SIGN:
3288 previous = NULL;
3289 *code++ = ((options & PCRE_MULTILINE) != 0)? OP_DOLLM : OP_DOLL;
3290 break;
3291
3292 /* There can never be a first char if '.' is first, whatever happens about
3293 repeats. The value of reqbyte doesn't change either. */
3294
3295 case CHAR_DOT:
3296 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
3297 zerofirstbyte = firstbyte;
3298 zeroreqbyte = reqbyte;
3299 previous = code;
3300 *code++ = ((options & PCRE_DOTALL) != 0)? OP_ALLANY: OP_ANY;
3301 break;
3302
3303
3304 /* ===================================================================*/
3305 /* Character classes. If the included characters are all < 256, we build a
3306 32-byte bitmap of the permitted characters, except in the special case
3307 where there is only one such character. For negated classes, we build the
3308 map as usual, then invert it at the end. However, we use a different opcode
3309 so that data characters > 255 can be handled correctly.
3310
3311 If the class contains characters outside the 0-255 range, a different
3312 opcode is compiled. It may optionally have a bit map for characters < 256,
3313 but those above are are explicitly listed afterwards. A flag byte tells
3314 whether the bitmap is present, and whether this is a negated class or not.
3315
3316 In JavaScript compatibility mode, an isolated ']' causes an error. In
3317 default (Perl) mode, it is treated as a data character. */
3318
3319 case CHAR_RIGHT_SQUARE_BRACKET:
3320 if ((cd->external_options & PCRE_JAVASCRIPT_COMPAT) != 0)
3321 {
3322 *errorcodeptr = ERR64;
3323 goto FAILED;
3324 }
3325 goto NORMAL_CHAR;
3326
3327 case CHAR_LEFT_SQUARE_BRACKET:
3328 previous = code;
3329
3330 /* PCRE supports POSIX class stuff inside a class. Perl gives an error if
3331 they are encountered at the top level, so we'll do that too. */
3332
3333 if ((ptr[1] == CHAR_COLON || ptr[1] == CHAR_DOT ||
3334 ptr[1] == CHAR_EQUALS_SIGN) &&
3335 check_posix_syntax(ptr, &tempptr))
3336 {
3337 *errorcodeptr = (ptr[1] == CHAR_COLON)? ERR13 : ERR31;
3338 goto FAILED;
3339 }
3340
3341 /* If the first character is '^', set the negation flag and skip it. Also,
3342 if the first few characters (either before or after ^) are \Q\E or \E we
3343 skip them too. This makes for compatibility with Perl. */
3344
3345 negate_class = FALSE;
3346 for (;;)
3347 {
3348 c = *(++ptr);
3349 if (c == CHAR_BACKSLASH)
3350 {
3351 if (ptr[1] == CHAR_E)
3352 ptr++;
3353 else if (strncmp((const char *)ptr+1,
3354 STR_Q STR_BACKSLASH STR_E, 3) == 0)
3355 ptr += 3;
3356 else
3357 break;
3358 }
3359 else if (!negate_class && c == CHAR_CIRCUMFLEX_ACCENT)
3360 negate_class = TRUE;
3361 else break;
3362 }
3363
3364 /* Empty classes are allowed in JavaScript compatibility mode. Otherwise,
3365 an initial ']' is taken as a data character -- the code below handles
3366 that. In JS mode, [] must always fail, so generate OP_FAIL, whereas
3367 [^] must match any character, so generate OP_ALLANY. */
3368
3369 if (c == CHAR_RIGHT_SQUARE_BRACKET &&
3370 (cd->external_options & PCRE_JAVASCRIPT_COMPAT) != 0)
3371 {
3372 *code++ = negate_class? OP_ALLANY : OP_FAIL;
3373 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
3374 zerofirstbyte = firstbyte;
3375 break;
3376 }
3377
3378 /* If a class contains a negative special such as \S, we need to flip the
3379 negation flag at the end, so that support for characters > 255 works
3380 correctly (they are all included in the class). */
3381
3382 should_flip_negation = FALSE;
3383
3384 /* Keep a count of chars with values < 256 so that we can optimize the case
3385 of just a single character (as long as it's < 256). However, For higher
3386 valued UTF-8 characters, we don't yet do any optimization. */
3387
3388 class_charcount = 0;
3389 class_lastchar = -1;
3390
3391 /* Initialize the 32-char bit map to all zeros. We build the map in a
3392 temporary bit of memory, in case the class contains only 1 character (less
3393 than 256), because in that case the compiled code doesn't use the bit map.
3394 */
3395
3396 memset(classbits, 0, 32 * sizeof(uschar));
3397
3398 #ifdef SUPPORT_UTF8
3399 class_utf8 = FALSE; /* No chars >= 256 */
3400 class_utf8data = code + LINK_SIZE + 2; /* For UTF-8 items */
3401 class_utf8data_base = class_utf8data; /* For resetting in pass 1 */
3402 #endif
3403
3404 /* Process characters until ] is reached. By writing this as a "do" it
3405 means that an initial ] is taken as a data character. At the start of the
3406 loop, c contains the first byte of the character. */
3407
3408 if (c != 0) do
3409 {
3410 const uschar *oldptr;
3411
3412 #ifdef SUPPORT_UTF8
3413 if (utf8 && c > 127)
3414 { /* Braces are required because the */
3415 GETCHARLEN(c, ptr, ptr); /* macro generates multiple statements */
3416 }
3417
3418 /* In the pre-compile phase, accumulate the length of any UTF-8 extra
3419 data and reset the pointer. This is so that very large classes that
3420 contain a zillion UTF-8 characters no longer overwrite the work space
3421 (which is on the stack). */
3422
3423 if (lengthptr != NULL)
3424 {
3425 *lengthptr += class_utf8data - class_utf8data_base;
3426 class_utf8data = class_utf8data_base;
3427 }
3428
3429 #endif
3430
3431 /* Inside \Q...\E everything is literal except \E */
3432
3433 if (inescq)
3434 {
3435 if (c == CHAR_BACKSLASH && ptr[1] == CHAR_E) /* If we are at \E */
3436 {
3437 inescq = FALSE; /* Reset literal state */
3438 ptr++; /* Skip the 'E' */
3439 continue; /* Carry on with next */
3440 }
3441 goto CHECK_RANGE; /* Could be range if \E follows */
3442 }
3443
3444 /* Handle POSIX class names. Perl allows a negation extension of the
3445 form [:^name:]. A square bracket that doesn't match the syntax is
3446 treated as a literal. We also recognize the POSIX constructions
3447 [.ch.] and [=ch=] ("collating elements") and fault them, as Perl
3448 5.6 and 5.8 do. */
3449
3450 if (c == CHAR_LEFT_SQUARE_BRACKET &&
3451 (ptr[1] == CHAR_COLON || ptr[1] == CHAR_DOT ||
3452 ptr[1] == CHAR_EQUALS_SIGN) && check_posix_syntax(ptr, &tempptr))
3453 {
3454 BOOL local_negate = FALSE;
3455 int posix_class, taboffset, tabopt;
3456 register const uschar *cbits = cd->cbits;
3457 uschar pbits[32];
3458
3459 if (ptr[1] != CHAR_COLON)
3460 {
3461 *errorcodeptr = ERR31;
3462 goto FAILED;
3463 }
3464
3465 ptr += 2;
3466 if (*ptr == CHAR_CIRCUMFLEX_ACCENT)
3467 {
3468 local_negate = TRUE;
3469 should_flip_negation = TRUE; /* Note negative special */
3470 ptr++;
3471 }
3472
3473 posix_class = check_posix_name(ptr, (int)(tempptr - ptr));
3474 if (posix_class < 0)
3475 {
3476 *errorcodeptr = ERR30;
3477 goto FAILED;
3478 }
3479
3480 /* If matching is caseless, upper and lower are converted to
3481 alpha. This relies on the fact that the class table starts with
3482 alpha, lower, upper as the first 3 entries. */
3483
3484 if ((options & PCRE_CASELESS) != 0 && posix_class <= 2)
3485 posix_class = 0;
3486
3487 /* When PCRE_UCP is set, some of the POSIX classes are converted to
3488 different escape sequences that use Unicode properties. */
3489
3490 #ifdef SUPPORT_UCP
3491 if ((options & PCRE_UCP) != 0)
3492 {
3493 int pc = posix_class + ((local_negate)? POSIX_SUBSIZE/2 : 0);
3494 if (posix_substitutes[pc] != NULL)
3495 {
3496 nestptr = tempptr + 1;
3497 ptr = posix_substitutes[pc] - 1;
3498 continue;
3499 }
3500 }
3501 #endif
3502 /* In the non-UCP case, we build the bit map for the POSIX class in a
3503 chunk of local store because we may be adding and subtracting from it,
3504 and we don't want to subtract bits that may be in the main map already.
3505 At the end we or the result into the bit map that is being built. */
3506
3507 posix_class *= 3;
3508
3509 /* Copy in the first table (always present) */
3510
3511 memcpy(pbits, cbits + posix_class_maps[posix_class],
3512 32 * sizeof(uschar));
3513
3514 /* If there is a second table, add or remove it as required. */
3515
3516 taboffset = posix_class_maps[posix_class + 1];
3517 tabopt = posix_class_maps[posix_class + 2];
3518
3519 if (taboffset >= 0)
3520 {
3521 if (tabopt >= 0)
3522 for (c = 0; c < 32; c++) pbits[c] |= cbits[c + taboffset];
3523 else
3524 for (c = 0; c < 32; c++) pbits[c] &= ~cbits[c + taboffset];
3525 }
3526
3527 /* Not see if we need to remove any special characters. An option
3528 value of 1 removes vertical space and 2 removes underscore. */
3529
3530 if (tabopt < 0) tabopt = -tabopt;
3531 if (tabopt == 1) pbits[1] &= ~0x3c;
3532 else if (tabopt == 2) pbits[11] &= 0x7f;
3533
3534 /* Add the POSIX table or its complement into the main table that is
3535 being built and we are done. */
3536
3537 if (local_negate)
3538 for (c = 0; c < 32; c++) classbits[c] |= ~pbits[c];
3539 else
3540 for (c = 0; c < 32; c++) classbits[c] |= pbits[c];
3541
3542 ptr = tempptr + 1;
3543 class_charcount = 10; /* Set > 1; assumes more than 1 per class */
3544 continue; /* End of POSIX syntax handling */
3545 }
3546
3547 /* Backslash may introduce a single character, or it may introduce one
3548 of the specials, which just set a flag. The sequence \b is a special
3549 case. Inside a class (and only there) it is treated as backspace. We
3550 assume that other escapes have more than one character in them, so set
3551 class_charcount bigger than one. Unrecognized escapes fall through and
3552 are either treated as literal characters (by default), or are faulted if
3553 PCRE_EXTRA is set. */
3554
3555 if (c == CHAR_BACKSLASH)
3556 {
3557 c = check_escape(&ptr, errorcodeptr, cd->bracount, options, TRUE);
3558 if (*errorcodeptr != 0) goto FAILED;
3559
3560 if (-c == ESC_b) c = CHAR_BS; /* \b is backspace in a class */
3561 else if (-c == ESC_Q) /* Handle start of quoted string */
3562 {
3563 if (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
3564 {
3565 ptr += 2; /* avoid empty string */
3566 }
3567 else inescq = TRUE;
3568 continue;
3569 }
3570 else if (-c == ESC_E) continue; /* Ignore orphan \E */
3571
3572 if (c < 0)
3573 {
3574 register const uschar *cbits = cd->cbits;
3575 class_charcount += 2; /* Greater than 1 is what matters */
3576
3577 switch (-c)
3578 {
3579 #ifdef SUPPORT_UCP
3580 case ESC_du: /* These are the values given for \d etc */
3581 case ESC_DU: /* when PCRE_UCP is set. We replace the */
3582 case ESC_wu: /* escape sequence with an appropriate \p */
3583 case ESC_WU: /* or \P to test Unicode properties instead */
3584 case ESC_su: /* of the default ASCII testing. */
3585 case ESC_SU:
3586 nestptr = ptr;
3587 ptr = substitutes[-c - ESC_DU] - 1; /* Just before substitute */
3588 class_charcount -= 2; /* Undo! */
3589 continue;
3590 #endif
3591 case ESC_d:
3592 for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_digit];
3593 continue;
3594
3595 case ESC_D:
3596 should_flip_negation = TRUE;
3597 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_digit];
3598 continue;
3599
3600 case ESC_w:
3601 for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_word];
3602 continue;
3603
3604 case ESC_W:
3605 should_flip_negation = TRUE;
3606 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_word];
3607 continue;
3608
3609 /* Perl 5.004 onwards omits VT from \s, but we must preserve it
3610 if it was previously set by something earlier in the character
3611 class. */
3612
3613 case ESC_s:
3614 classbits[0] |= cbits[cbit_space];
3615 classbits[1] |= cbits[cbit_space+1] & ~0x08;
3616 for (c = 2; c < 32; c++) classbits[c] |= cbits[c+cbit_space];
3617 continue;
3618
3619 case ESC_S:
3620 should_flip_negation = TRUE;
3621 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_space];
3622 classbits[1] |= 0x08; /* Perl 5.004 onwards omits VT from \s */
3623 continue;
3624
3625 case ESC_h:
3626 SETBIT(classbits, 0x09); /* VT */
3627 SETBIT(classbits, 0x20); /* SPACE */
3628 SETBIT(classbits, 0xa0); /* NSBP */
3629 #ifdef SUPPORT_UTF8
3630 if (utf8)
3631 {
3632 class_utf8 = TRUE;
3633 *class_utf8data++ = XCL_SINGLE;
3634 class_utf8data += _pcre_ord2utf8(0x1680, class_utf8data);
3635 *class_utf8data++ = XCL_SINGLE;
3636 class_utf8data += _pcre_ord2utf8(0x180e, class_utf8data);
3637 *class_utf8data++ = XCL_RANGE;
3638 class_utf8data += _pcre_ord2utf8(0x2000, class_utf8data);
3639 class_utf8data += _pcre_ord2utf8(0x200A, class_utf8data);
3640 *class_utf8data++ = XCL_SINGLE;
3641 class_utf8data += _pcre_ord2utf8(0x202f, class_utf8data);
3642 *class_utf8data++ = XCL_SINGLE;
3643 class_utf8data += _pcre_ord2utf8(0x205f, class_utf8data);
3644 *class_utf8data++ = XCL_SINGLE;
3645 class_utf8data += _pcre_ord2utf8(0x3000, class_utf8data);
3646 }
3647 #endif
3648 continue;
3649
3650 case ESC_H:
3651 for (c = 0; c < 32; c++)
3652 {
3653 int x = 0xff;
3654 switch (c)
3655 {
3656 case 0x09/8: x ^= 1 << (0x09%8); break;
3657 case 0x20/8: x ^= 1 << (0x20%8); break;
3658 case 0xa0/8: x ^= 1 << (0xa0%8); break;
3659 default: break;
3660 }
3661 classbits[c] |= x;
3662 }
3663
3664 #ifdef SUPPORT_UTF8
3665 if (utf8)
3666 {
3667 class_utf8 = TRUE;
3668 *class_utf8data++ = XCL_RANGE;
3669 class_utf8data += _pcre_ord2utf8(0x0100, class_utf8data);
3670 class_utf8data += _pcre_ord2utf8(0x167f, class_utf8data);
3671 *class_utf8data++ = XCL_RANGE;
3672 class_utf8data += _pcre_ord2utf8(0x1681, class_utf8data);
3673 class_utf8data += _pcre_ord2utf8(0x180d, class_utf8data);
3674 *class_utf8data++ = XCL_RANGE;
3675 class_utf8data += _pcre_ord2utf8(0x180f, class_utf8data);
3676 class_utf8data += _pcre_ord2utf8(0x1fff, class_utf8data);
3677 *class_utf8data++ = XCL_RANGE;
3678 class_utf8data += _pcre_ord2utf8(0x200B, class_utf8data);
3679 class_utf8data += _pcre_ord2utf8(0x202e, class_utf8data);
3680 *class_utf8data++ = XCL_RANGE;
3681 class_utf8data += _pcre_ord2utf8(0x2030, class_utf8data);
3682 class_utf8data += _pcre_ord2utf8(0x205e, class_utf8data);
3683 *class_utf8data++ = XCL_RANGE;
3684 class_utf8data += _pcre_ord2utf8(0x2060, class_utf8data);
3685 class_utf8data += _pcre_ord2utf8(0x2fff, class_utf8data);
3686 *class_utf8data++ = XCL_RANGE;
3687 class_utf8data += _pcre_ord2utf8(0x3001, class_utf8data);
3688 class_utf8data += _pcre_ord2utf8(0x7fffffff, class_utf8data);
3689 }
3690 #endif
3691 continue;
3692
3693 case ESC_v:
3694 SETBIT(classbits, 0x0a); /* LF */
3695 SETBIT(classbits, 0x0b); /* VT */
3696 SETBIT(classbits, 0x0c); /* FF */
3697 SETBIT(classbits, 0x0d); /* CR */
3698 SETBIT(classbits, 0x85); /* NEL */
3699 #ifdef SUPPORT_UTF8
3700 if (utf8)
3701 {
3702 class_utf8 = TRUE;
3703 *class_utf8data++ = XCL_RANGE;
3704 class_utf8data += _pcre_ord2utf8(0x2028, class_utf8data);
3705 class_utf8data += _pcre_ord2utf8(0x2029, class_utf8data);
3706 }
3707 #endif
3708 continue;
3709
3710 case ESC_V:
3711 for (c = 0; c < 32; c++)
3712 {
3713 int x = 0xff;
3714 switch (c)
3715 {
3716 case 0x0a/8: x ^= 1 << (0x0a%8);
3717 x ^= 1 << (0x0b%8);
3718 x ^= 1 << (0x0c%8);
3719 x ^= 1 << (0x0d%8);
3720 break;
3721 case 0x85/8: x ^= 1 << (0x85%8); break;
3722 default: break;
3723 }
3724 classbits[c] |= x;
3725 }
3726
3727 #ifdef SUPPORT_UTF8
3728 if (utf8)
3729 {
3730 class_utf8 = TRUE;
3731 *class_utf8data++ = XCL_RANGE;
3732 class_utf8data += _pcre_ord2utf8(0x0100, class_utf8data);
3733 class_utf8data += _pcre_ord2utf8(0x2027, class_utf8data);
3734 *class_utf8data++ = XCL_RANGE;
3735 class_utf8data += _pcre_ord2utf8(0x2029, class_utf8data);
3736 class_utf8data += _pcre_ord2utf8(0x7fffffff, class_utf8data);
3737 }
3738 #endif
3739 continue;
3740
3741 #ifdef SUPPORT_UCP
3742 case ESC_p:
3743 case ESC_P:
3744 {
3745 BOOL negated;
3746 int pdata;
3747 int ptype = get_ucp(&ptr, &negated, &pdata, errorcodeptr);
3748 if (ptype < 0) goto FAILED;
3749 class_utf8 = TRUE;
3750 *class_utf8data++ = ((-c == ESC_p) != negated)?
3751 XCL_PROP : XCL_NOTPROP;
3752 *class_utf8data++ = ptype;
3753 *class_utf8data++ = pdata;
3754 class_charcount -= 2; /* Not a < 256 character */
3755 continue;
3756 }
3757 #endif
3758 /* Unrecognized escapes are faulted if PCRE is running in its
3759 strict mode. By default, for compatibility with Perl, they are
3760 treated as literals. */
3761
3762 default:
3763 if ((options & PCRE_EXTRA) != 0)
3764 {
3765 *errorcodeptr = ERR7;
3766 goto FAILED;
3767 }
3768 class_charcount -= 2; /* Undo the default count from above */
3769 c = *ptr; /* Get the final character and fall through */
3770 break;
3771 }
3772 }
3773
3774 /* Fall through if we have a single character (c >= 0). This may be
3775 greater than 256 in UTF-8 mode. */
3776
3777 } /* End of backslash handling */
3778
3779 /* A single character may be followed by '-' to form a range. However,
3780 Perl does not permit ']' to be the end of the range. A '-' character
3781 at the end is treated as a literal. Perl ignores orphaned \E sequences
3782 entirely. The code for handling \Q and \E is messy. */
3783
3784 CHECK_RANGE:
3785 while (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
3786 {
3787 inescq = FALSE;
3788 ptr += 2;
3789 }
3790
3791 oldptr = ptr;
3792
3793 /* Remember \r or \n */
3794
3795 if (c == CHAR_CR || c == CHAR_NL) cd->external_flags |= PCRE_HASCRORLF;
3796
3797 /* Check for range */
3798
3799 if (!inescq && ptr[1] == CHAR_MINUS)
3800 {
3801 int d;
3802 ptr += 2;
3803 while (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_E) ptr += 2;
3804
3805 /* If we hit \Q (not followed by \E) at this point, go into escaped
3806 mode. */
3807
3808 while (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_Q)
3809 {
3810 ptr += 2;
3811 if (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_E)
3812 { ptr += 2; continue; }
3813 inescq = TRUE;
3814 break;
3815 }
3816
3817 if (*ptr == 0 || (!inescq && *ptr == CHAR_RIGHT_SQUARE_BRACKET))
3818 {
3819 ptr = oldptr;
3820 goto LONE_SINGLE_CHARACTER;
3821 }
3822
3823 #ifdef SUPPORT_UTF8
3824 if (utf8)
3825 { /* Braces are required because the */
3826 GETCHARLEN(d, ptr, ptr); /* macro generates multiple statements */
3827 }
3828 else
3829 #endif
3830 d = *ptr; /* Not UTF-8 mode */
3831
3832 /* The second part of a range can be a single-character escape, but
3833 not any of the other escapes. Perl 5.6 treats a hyphen as a literal
3834 in such circumstances. */
3835
3836 if (!inescq && d == CHAR_BACKSLASH)
3837 {
3838 d = check_escape(&ptr, errorcodeptr, cd->bracount, options, TRUE);
3839 if (*errorcodeptr != 0) goto FAILED;
3840
3841 /* \b is backspace; any other special means the '-' was literal */
3842
3843 if (d < 0)
3844 {
3845 if (d == -ESC_b) d = CHAR_BS; else
3846 {
3847 ptr = oldptr;
3848 goto LONE_SINGLE_CHARACTER; /* A few lines below */
3849 }
3850 }
3851 }
3852
3853 /* Check that the two values are in the correct order. Optimize
3854 one-character ranges */
3855
3856 if (d < c)
3857 {
3858 *errorcodeptr = ERR8;
3859 goto FAILED;
3860 }
3861
3862 if (d == c) goto LONE_SINGLE_CHARACTER; /* A few lines below */
3863
3864 /* Remember \r or \n */
3865
3866 if (d == CHAR_CR || d == CHAR_NL) cd->external_flags |= PCRE_HASCRORLF;
3867
3868 /* In UTF-8 mode, if the upper limit is > 255, or > 127 for caseless
3869 matching, we have to use an XCLASS with extra data items. Caseless
3870 matching for characters > 127 is available only if UCP support is
3871 available. */
3872
3873 #ifdef SUPPORT_UTF8
3874 if (utf8 && (d > 255 || ((options & PCRE_CASELESS) != 0 && d > 127)))
3875 {
3876 class_utf8 = TRUE;
3877
3878 /* With UCP support, we can find the other case equivalents of
3879 the relevant characters. There may be several ranges. Optimize how
3880 they fit with the basic range. */
3881
3882 #ifdef SUPPORT_UCP
3883 if ((options & PCRE_CASELESS) != 0)
3884 {
3885 unsigned int occ, ocd;
3886 unsigned int cc = c;
3887 unsigned int origd = d;
3888 while (get_othercase_range(&cc, origd, &occ, &ocd))
3889 {
3890 if (occ >= (unsigned int)c &&
3891 ocd <= (unsigned int)d)
3892 continue; /* Skip embedded ranges */
3893
3894 if (occ < (unsigned int)c &&
3895 ocd >= (unsigned int)c - 1) /* Extend the basic range */
3896 { /* if there is overlap, */
3897 c = occ; /* noting that if occ < c */
3898 continue; /* we can't have ocd > d */
3899 } /* because a subrange is */
3900 if (ocd > (unsigned int)d &&
3901 occ <= (unsigned int)d + 1) /* always shorter than */
3902 { /* the basic range. */
3903 d = ocd;
3904 continue;
3905 }
3906
3907 if (occ == ocd)
3908 {
3909 *class_utf8data++ = XCL_SINGLE;
3910 }
3911 else
3912 {
3913 *class_utf8data++ = XCL_RANGE;
3914 class_utf8data += _pcre_ord2utf8(occ, class_utf8data);
3915 }
3916 class_utf8data += _pcre_ord2utf8(ocd, class_utf8data);
3917 }
3918 }
3919 #endif /* SUPPORT_UCP */
3920
3921 /* Now record the original range, possibly modified for UCP caseless
3922 overlapping ranges. */
3923
3924 *class_utf8data++ = XCL_RANGE;
3925 class_utf8data += _pcre_ord2utf8(c, class_utf8data);
3926 class_utf8data += _pcre_ord2utf8(d, class_utf8data);
3927
3928 /* With UCP support, we are done. Without UCP support, there is no
3929 caseless matching for UTF-8 characters > 127; we can use the bit map
3930 for the smaller ones. */
3931
3932 #ifdef SUPPORT_UCP
3933 continue; /* With next character in the class */
3934 #else
3935 if ((options & PCRE_CASELESS) == 0 || c > 127) continue;
3936
3937 /* Adjust upper limit and fall through to set up the map */
3938
3939 d = 127;
3940
3941 #endif /* SUPPORT_UCP */
3942 }
3943 #endif /* SUPPORT_UTF8 */
3944
3945 /* We use the bit map for all cases when not in UTF-8 mode; else
3946 ranges that lie entirely within 0-127 when there is UCP support; else
3947 for partial ranges without UCP support. */
3948
3949 class_charcount += d - c + 1;
3950 class_lastchar = d;
3951
3952 /* We can save a bit of time by skipping this in the pre-compile. */
3953
3954 if (lengthptr == NULL) for (; c <= d; c++)
3955 {
3956 classbits[c/8] |= (1 << (c&7));
3957 if ((options & PCRE_CASELESS) != 0)
3958 {
3959 int uc = cd->fcc[c]; /* flip case */
3960 classbits[uc/8] |= (1 << (uc&7));
3961 }
3962 }
3963
3964 continue; /* Go get the next char in the class */
3965 }
3966
3967 /* Handle a lone single character - we can get here for a normal
3968 non-escape char, or after \ that introduces a single character or for an
3969 apparent range that isn't. */
3970
3971 LONE_SINGLE_CHARACTER:
3972
3973 /* Handle a character that cannot go in the bit map */
3974
3975 #ifdef SUPPORT_UTF8
3976 if (utf8 && (c > 255 || ((options & PCRE_CASELESS) != 0 && c > 127)))
3977 {
3978 class_utf8 = TRUE;
3979 *class_utf8data++ = XCL_SINGLE;
3980 class_utf8data += _pcre_ord2utf8(c, class_utf8data);
3981
3982 #ifdef SUPPORT_UCP
3983 if ((options & PCRE_CASELESS) != 0)
3984 {
3985 unsigned int othercase;
3986 if ((othercase = UCD_OTHERCASE(c)) != c)
3987 {
3988 *class_utf8data++ = XCL_SINGLE;
3989 class_utf8data += _pcre_ord2utf8(othercase, class_utf8data);
3990 }
3991 }
3992 #endif /* SUPPORT_UCP */
3993
3994 }
3995 else
3996 #endif /* SUPPORT_UTF8 */
3997
3998 /* Handle a single-byte character */
3999 {
4000 classbits[c/8] |= (1 << (c&7));
4001 if ((options & PCRE_CASELESS) != 0)
4002 {
4003 c = cd->fcc[c]; /* flip case */
4004 classbits[c/8] |= (1 << (c&7));
4005 }
4006 class_charcount++;
4007 class_lastchar = c;
4008 }
4009 }
4010
4011 /* Loop until ']' reached. This "while" is the end of the "do" far above.
4012 If we are at the end of an internal nested string, revert to the outer
4013 string. */
4014
4015 while (((c = *(++ptr)) != 0 ||
4016 (nestptr != NULL &&
4017 (ptr = nestptr, nestptr = NULL, c = *(++ptr)) != 0)) &&
4018 (c != CHAR_RIGHT_SQUARE_BRACKET || inescq));
4019
4020 /* Check for missing terminating ']' */
4021
4022 if (c == 0)
4023 {
4024 *errorcodeptr = ERR6;
4025 goto FAILED;
4026 }
4027
4028 /* If class_charcount is 1, we saw precisely one character whose value is
4029 less than 256. As long as there were no characters >= 128 and there was no
4030 use of \p or \P, in other words, no use of any XCLASS features, we can
4031 optimize.
4032
4033 In UTF-8 mode, we can optimize the negative case only if there were no
4034 characters >= 128 because OP_NOT and the related opcodes like OP_NOTSTAR
4035 operate on single-bytes characters only. This is an historical hangover.
4036 Maybe one day we can tidy these opcodes to handle multi-byte characters.
4037
4038 The optimization throws away the bit map. We turn the item into a
4039 1-character OP_CHAR[I] if it's positive, or OP_NOT[I] if it's negative.
4040 Note that OP_NOT[I] does not support multibyte characters. In the positive
4041 case, it can cause firstbyte to be set. Otherwise, there can be no first
4042 char if this item is first, whatever repeat count may follow. In the case
4043 of reqbyte, save the previous value for reinstating. */
4044
4045 #ifdef SUPPORT_UTF8
4046 if (class_charcount == 1 && !class_utf8 &&
4047 (!utf8 || !negate_class || class_lastchar < 128))
4048 #else
4049 if (class_charcount == 1)
4050 #endif
4051 {
4052 zeroreqbyte = reqbyte;
4053
4054 /* The OP_NOT[I] opcodes work on one-byte characters only. */
4055
4056 if (negate_class)
4057 {
4058 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
4059 zerofirstbyte = firstbyte;
4060 *code++ = ((options & PCRE_CASELESS) != 0)? OP_NOTI: OP_NOT;
4061 *code++ = class_lastchar;
4062 break;
4063 }
4064
4065 /* For a single, positive character, get the value into mcbuffer, and
4066 then we can handle this with the normal one-character code. */
4067
4068 #ifdef SUPPORT_UTF8
4069 if (utf8 && class_lastchar > 127)
4070 mclength = _pcre_ord2utf8(class_lastchar, mcbuffer);
4071 else
4072 #endif
4073 {
4074 mcbuffer[0] = class_lastchar;
4075 mclength = 1;
4076 }
4077 goto ONE_CHAR;
4078 } /* End of 1-char optimization */
4079
4080 /* The general case - not the one-char optimization. If this is the first
4081 thing in the branch, there can be no first char setting, whatever the
4082 repeat count. Any reqbyte setting must remain unchanged after any kind of
4083 repeat. */
4084
4085 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
4086 zerofirstbyte = firstbyte;
4087 zeroreqbyte = reqbyte;
4088
4089 /* If there are characters with values > 255, we have to compile an
4090 extended class, with its own opcode, unless there was a negated special
4091 such as \S in the class, and PCRE_UCP is not set, because in that case all
4092 characters > 255 are in the class, so any that were explicitly given as
4093 well can be ignored. If (when there are explicit characters > 255 that must
4094 be listed) there are no characters < 256, we can omit the bitmap in the
4095 actual compiled code. */
4096
4097 #ifdef SUPPORT_UTF8
4098 if (class_utf8 && (!should_flip_negation || (options & PCRE_UCP) != 0))
4099 {
4100 *class_utf8data++ = XCL_END; /* Marks the end of extra data */
4101 *code++ = OP_XCLASS;
4102 code += LINK_SIZE;
4103 *code = negate_class? XCL_NOT : 0;
4104
4105 /* If the map is required, move up the extra data to make room for it;
4106 otherwise just move the code pointer to the end of the extra data. */
4107
4108 if (class_charcount > 0)
4109 {
4110 *code++ |= XCL_MAP;
4111 memmove(code + 32, code, class_utf8data - code);
4112 memcpy(code, classbits, 32);
4113 code = class_utf8data + 32;
4114 }
4115 else code = class_utf8data;
4116
4117 /* Now fill in the complete length of the item */
4118
4119 PUT(previous, 1, code - previous);
4120 break; /* End of class handling */
4121 }
4122 #endif
4123
4124 /* If there are no characters > 255, or they are all to be included or
4125 excluded, set the opcode to OP_CLASS or OP_NCLASS, depending on whether the
4126 whole class was negated and whether there were negative specials such as \S
4127 (non-UCP) in the class. Then copy the 32-byte map into the code vector,
4128 negating it if necessary. */
4129
4130 *code++ = (negate_class == should_flip_negation) ? OP_CLASS : OP_NCLASS;
4131 if (negate_class)
4132 {
4133 if (lengthptr == NULL) /* Save time in the pre-compile phase */
4134 for (c = 0; c < 32; c++) code[c] = ~classbits[c];
4135 }
4136 else
4137 {
4138 memcpy(code, classbits, 32);
4139 }
4140 code += 32;
4141 break;
4142
4143
4144 /* ===================================================================*/
4145 /* Various kinds of repeat; '{' is not necessarily a quantifier, but this
4146 has been tested above. */
4147
4148 case CHAR_LEFT_CURLY_BRACKET:
4149 if (!is_quantifier) goto NORMAL_CHAR;
4150 ptr = read_repeat_counts(ptr+1, &repeat_min, &repeat_max, errorcodeptr);
4151 if (*errorcodeptr != 0) goto FAILED;
4152 goto REPEAT;
4153
4154 case CHAR_ASTERISK:
4155 repeat_min = 0;
4156 repeat_max = -1;
4157 goto REPEAT;
4158
4159 case CHAR_PLUS:
4160 repeat_min = 1;
4161 repeat_max = -1;
4162 goto REPEAT;
4163
4164 case CHAR_QUESTION_MARK:
4165 repeat_min = 0;
4166 repeat_max = 1;
4167
4168 REPEAT:
4169 if (previous == NULL)
4170 {
4171 *errorcodeptr = ERR9;
4172 goto FAILED;
4173 }
4174
4175 if (repeat_min == 0)
4176 {
4177 firstbyte = zerofirstbyte; /* Adjust for zero repeat */
4178 reqbyte = zeroreqbyte; /* Ditto */
4179 }
4180
4181 /* Remember whether this is a variable length repeat */
4182
4183 reqvary = (repeat_min == repeat_max)? 0 : REQ_VARY;
4184
4185 op_type = 0; /* Default single-char op codes */
4186 possessive_quantifier = FALSE; /* Default not possessive quantifier */
4187
4188 /* Save start of previous item, in case we have to move it up to make space
4189 for an inserted OP_ONCE for the additional '+' extension. */
4190
4191 tempcode = previous;
4192
4193 /* If the next character is '+', we have a possessive quantifier. This
4194 implies greediness, whatever the setting of the PCRE_UNGREEDY option.
4195 If the next character is '?' this is a minimizing repeat, by default,
4196 but if PCRE_UNGREEDY is set, it works the other way round. We change the
4197 repeat type to the non-default. */
4198
4199 if (ptr[1] == CHAR_PLUS)
4200 {
4201 repeat_type = 0; /* Force greedy */
4202 possessive_quantifier = TRUE;
4203 ptr++;
4204 }
4205 else if (ptr[1] == CHAR_QUESTION_MARK)
4206 {
4207 repeat_type = greedy_non_default;
4208 ptr++;
4209 }
4210 else repeat_type = greedy_default;
4211
4212 /* If previous was a recursion call, wrap it in atomic brackets so that
4213 previous becomes the atomic group. All recursions were so wrapped in the
4214 past, but it no longer happens for non-repeated recursions. In fact, the
4215 repeated ones could be re-implemented independently so as not to need this,
4216 but for the moment we rely on the code for repeating groups. */
4217
4218 if (*previous == OP_RECURSE)
4219 {
4220 memmove(previous + 1 + LINK_SIZE, previous, 1 + LINK_SIZE);
4221 *previous = OP_ONCE;
4222 PUT(previous, 1, 2 + 2*LINK_SIZE);
4223 previous[2 + 2*LINK_SIZE] = OP_KET;
4224 PUT(previous, 3 + 2*LINK_SIZE, 2 + 2*LINK_SIZE);
4225 code += 2 + 2 * LINK_SIZE;
4226 length_prevgroup = 3 + 3*LINK_SIZE;
4227
4228 /* When actually compiling, we need to check whether this was a forward
4229 reference, and if so, adjust the offset. */
4230
4231 if (lengthptr == NULL && cd->hwm >= cd->start_workspace + LINK_SIZE)
4232 {
4233 int offset = GET(cd->hwm, -LINK_SIZE);
4234 if (offset == previous + 1 - cd->start_code)
4235 PUT(cd->hwm, -LINK_SIZE, offset + 1 + LINK_SIZE);
4236 }
4237 }
4238
4239 /* Now handle repetition for the different types of item. */
4240
4241 /* If previous was a character match, abolish the item and generate a
4242 repeat item instead. If a char item has a minumum of more than one, ensure
4243 that it is set in reqbyte - it might not be if a sequence such as x{3} is
4244 the first thing in a branch because the x will have gone into firstbyte
4245 instead. */
4246
4247 if (*previous == OP_CHAR || *previous == OP_CHARI)
4248 {
4249 op_type = (*previous == OP_CHAR)? 0 : OP_STARI - OP_STAR;
4250
4251 /* Deal with UTF-8 characters that take up more than one byte. It's
4252 easier to write this out separately than try to macrify it. Use c to
4253 hold the length of the character in bytes, plus 0x80 to flag that it's a
4254 length rather than a small character. */
4255
4256 #ifdef SUPPORT_UTF8
4257 if (utf8 && (code[-1] & 0x80) != 0)
4258 {
4259 uschar *lastchar = code - 1;
4260 while((*lastchar & 0xc0) == 0x80) lastchar--;
4261 c = code - lastchar; /* Length of UTF-8 character */
4262 memcpy(utf8_char, lastchar, c); /* Save the char */
4263 c |= 0x80; /* Flag c as a length */
4264 }
4265 else
4266 #endif
4267
4268 /* Handle the case of a single byte - either with no UTF8 support, or
4269 with UTF-8 disabled, or for a UTF-8 character < 128. */
4270
4271 {
4272 c = code[-1];
4273 if (repeat_min > 1) reqbyte = c | req_caseopt | cd->req_varyopt;
4274 }
4275
4276 /* If the repetition is unlimited, it pays to see if the next thing on
4277 the line is something that cannot possibly match this character. If so,
4278 automatically possessifying this item gains some performance in the case
4279 where the match fails. */
4280
4281 if (!possessive_quantifier &&
4282 repeat_max < 0 &&
4283 check_auto_possessive(previous, utf8, ptr + 1, options, cd))
4284 {
4285 repeat_type = 0; /* Force greedy */
4286 possessive_quantifier = TRUE;
4287 }
4288
4289 goto OUTPUT_SINGLE_REPEAT; /* Code shared with single character types */
4290 }
4291
4292 /* If previous was a single negated character ([^a] or similar), we use
4293 one of the special opcodes, replacing it. The code is shared with single-
4294 character repeats by setting opt_type to add a suitable offset into
4295 repeat_type. We can also test for auto-possessification. OP_NOT and OP_NOTI
4296 are currently used only for single-byte chars. */
4297
4298 else if (*previous == OP_NOT || *previous == OP_NOTI)
4299 {
4300 op_type = ((*previous == OP_NOT)? OP_NOTSTAR : OP_NOTSTARI) - OP_STAR;
4301 c = previous[1];
4302 if (!possessive_quantifier &&
4303 repeat_max < 0 &&
4304 check_auto_possessive(previous, utf8, ptr + 1, options, cd))
4305 {
4306 repeat_type = 0; /* Force greedy */
4307 possessive_quantifier = TRUE;
4308 }
4309 goto OUTPUT_SINGLE_REPEAT;
4310 }
4311
4312 /* If previous was a character type match (\d or similar), abolish it and
4313 create a suitable repeat item. The code is shared with single-character
4314 repeats by setting op_type to add a suitable offset into repeat_type. Note
4315 the the Unicode property types will be present only when SUPPORT_UCP is
4316 defined, but we don't wrap the little bits of code here because it just
4317 makes it horribly messy. */
4318
4319 else if (*previous < OP_EODN)
4320 {
4321 uschar *oldcode;
4322 int prop_type, prop_value;
4323 op_type = OP_TYPESTAR - OP_STAR; /* Use type opcodes */
4324 c = *previous;
4325
4326 if (!possessive_quantifier &&
4327 repeat_max < 0 &&
4328 check_auto_possessive(previous, utf8, ptr + 1, options, cd))
4329 {
4330 repeat_type = 0; /* Force greedy */
4331 possessive_quantifier = TRUE;
4332 }
4333
4334 OUTPUT_SINGLE_REPEAT:
4335 if (*previous == OP_PROP || *previous == OP_NOTPROP)
4336 {
4337 prop_type = previous[1];
4338 prop_value = previous[2];
4339 }
4340 else prop_type = prop_value = -1;
4341
4342 oldcode = code;
4343 code = previous; /* Usually overwrite previous item */
4344
4345 /* If the maximum is zero then the minimum must also be zero; Perl allows
4346 this case, so we do too - by simply omitting the item altogether. */
4347
4348 if (repeat_max == 0) goto END_REPEAT;
4349
4350 /*--------------------------------------------------------------------*/
4351 /* This code is obsolete from release 8.00; the restriction was finally
4352 removed: */
4353
4354 /* All real repeats make it impossible to handle partial matching (maybe
4355 one day we will be able to remove this restriction). */
4356
4357 /* if (repeat_max != 1) cd->external_flags |= PCRE_NOPARTIAL; */
4358 /*--------------------------------------------------------------------*/
4359
4360 /* Combine the op_type with the repeat_type */
4361
4362 repeat_type += op_type;
4363
4364 /* A minimum of zero is handled either as the special case * or ?, or as
4365 an UPTO, with the maximum given. */
4366
4367 if (repeat_min == 0)
4368 {
4369 if (repeat_max == -1) *code++ = OP_STAR + repeat_type;
4370 else if (repeat_max == 1) *code++ = OP_QUERY + repeat_type;
4371 else
4372 {
4373 *code++ = OP_UPTO + repeat_type;
4374 PUT2INC(code, 0, repeat_max);
4375 }
4376 }
4377
4378 /* A repeat minimum of 1 is optimized into some special cases. If the
4379 maximum is unlimited, we use OP_PLUS. Otherwise, the original item is
4380 left in place and, if the maximum is greater than 1, we use OP_UPTO with
4381 one less than the maximum. */
4382
4383 else if (repeat_min == 1)
4384 {
4385 if (repeat_max == -1)
4386 *code++ = OP_PLUS + repeat_type;
4387 else
4388 {
4389 code = oldcode; /* leave previous item in place */
4390 if (repeat_max == 1) goto END_REPEAT;
4391 *code++ = OP_UPTO + repeat_type;
4392 PUT2INC(code, 0, repeat_max - 1);
4393 }
4394 }
4395
4396 /* The case {n,n} is just an EXACT, while the general case {n,m} is
4397 handled as an EXACT followed by an UPTO. */
4398
4399 else
4400 {
4401 *code++ = OP_EXACT + op_type; /* NB EXACT doesn't have repeat_type */
4402 PUT2INC(code, 0, repeat_min);
4403
4404 /* If the maximum is unlimited, insert an OP_STAR. Before doing so,
4405 we have to insert the character for the previous code. For a repeated
4406 Unicode property match, there are two extra bytes that define the
4407 required property. In UTF-8 mode, long characters have their length in
4408 c, with the 0x80 bit as a flag. */
4409
4410 if (repeat_max < 0)
4411 {
4412 #ifdef SUPPORT_UTF8
4413 if (utf8 && c >= 128)
4414 {
4415 memcpy(code, utf8_char, c & 7);
4416 code += c & 7;
4417 }
4418 else
4419 #endif
4420 {
4421 *code++ = c;
4422 if (prop_type >= 0)
4423 {
4424 *code++ = prop_type;
4425 *code++ = prop_value;
4426 }
4427 }
4428 *code++ = OP_STAR + repeat_type;
4429 }
4430
4431 /* Else insert an UPTO if the max is greater than the min, again
4432 preceded by the character, for the previously inserted code. If the
4433 UPTO is just for 1 instance, we can use QUERY instead. */
4434
4435 else if (repeat_max != repeat_min)
4436 {
4437 #ifdef SUPPORT_UTF8
4438 if (utf8 && c >= 128)
4439 {
4440 memcpy(code, utf8_char, c & 7);
4441 code += c & 7;
4442 }
4443 else
4444 #endif
4445 *code++ = c;
4446 if (prop_type >= 0)
4447 {
4448 *code++ = prop_type;
4449 *code++ = prop_value;
4450 }
4451 repeat_max -= repeat_min;
4452
4453 if (repeat_max == 1)
4454 {
4455 *code++ = OP_QUERY + repeat_type;
4456 }
4457 else
4458 {
4459 *code++ = OP_UPTO + repeat_type;
4460 PUT2INC(code, 0, repeat_max);
4461 }
4462 }
4463 }
4464
4465 /* The character or character type itself comes last in all cases. */
4466
4467 #ifdef SUPPORT_UTF8
4468 if (utf8 && c >= 128)
4469 {
4470 memcpy(code, utf8_char, c & 7);
4471 code += c & 7;
4472 }
4473 else
4474 #endif
4475 *code++ = c;
4476
4477 /* For a repeated Unicode property match, there are two extra bytes that
4478 define the required property. */
4479
4480 #ifdef SUPPORT_UCP
4481 if (prop_type >= 0)
4482 {
4483 *code++ = prop_type;
4484 *code++ = prop_value;
4485 }
4486 #endif
4487 }
4488
4489 /* If previous was a character class or a back reference, we put the repeat
4490 stuff after it, but just skip the item if the repeat was {0,0}. */
4491
4492 else if (*previous == OP_CLASS ||
4493 *previous == OP_NCLASS ||
4494 #ifdef SUPPORT_UTF8
4495 *previous == OP_XCLASS ||
4496 #endif
4497 *previous == OP_REF ||
4498 *previous == OP_REFI)
4499 {
4500 if (repeat_max == 0)
4501 {
4502 code = previous;
4503 goto END_REPEAT;
4504 }
4505
4506 /*--------------------------------------------------------------------*/
4507 /* This code is obsolete from release 8.00; the restriction was finally
4508 removed: */
4509
4510 /* All real repeats make it impossible to handle partial matching (maybe
4511 one day we will be able to remove this restriction). */
4512
4513 /* if (repeat_max != 1) cd->external_flags |= PCRE_NOPARTIAL; */
4514 /*--------------------------------------------------------------------*/
4515
4516 if (repeat_min == 0 && repeat_max == -1)
4517 *code++ = OP_CRSTAR + repeat_type;
4518 else if (repeat_min == 1 && repeat_max == -1)
4519 *code++ = OP_CRPLUS + repeat_type;
4520 else if (repeat_min == 0 && repeat_max == 1)
4521 *code++ = OP_CRQUERY + repeat_type;
4522 else
4523 {
4524 *code++ = OP_CRRANGE + repeat_type;
4525 PUT2INC(code, 0, repeat_min);
4526 if (repeat_max == -1) repeat_max = 0; /* 2-byte encoding for max */
4527 PUT2INC(code, 0, repeat_max);
4528 }
4529 }
4530
4531 /* If previous was a bracket group, we may have to replicate it in certain
4532 cases. Note that at this point we can encounter only the "basic" BRA and
4533 KET opcodes, as this is the place where they get converted into the more
4534 special varieties. */
4535
4536 else if (*previous == OP_BRA || *previous == OP_CBRA ||
4537 *previous == OP_ONCE || *previous == OP_COND)
4538 {
4539 register int i;
4540 int len = (int)(code - previous);
4541 uschar *bralink = NULL;
4542 uschar *brazeroptr = NULL;
4543
4544 /* Repeating a DEFINE group is pointless */
4545
4546 if (*previous == OP_COND && previous[LINK_SIZE+1] == OP_DEF)
4547 {
4548 *errorcodeptr = ERR55;
4549 goto FAILED;
4550 }
4551
4552 /* The case of a zero minimum is special because of the need to stick
4553 OP_BRAZERO in front of it, and because the group appears once in the
4554 data, whereas in other cases it appears the minimum number of times. For
4555 this reason, it is simplest to treat this case separately, as otherwise
4556 the code gets far too messy. There are several special subcases when the
4557 minimum is zero. */
4558
4559 if (repeat_min == 0)
4560 {
4561 /* If the maximum is also zero, we used to just omit the group from the
4562 output altogether, like this:
4563
4564 ** if (repeat_max == 0)
4565 ** {
4566 ** code = previous;
4567 ** goto END_REPEAT;
4568 ** }
4569
4570 However, that fails when a group is referenced as a subroutine from
4571 elsewhere in the pattern, so now we stick in OP_SKIPZERO in front of it
4572 so that it is skipped on execution. As we don't have a list of which
4573 groups are referenced, we cannot do this selectively.
4574
4575 If the maximum is 1 or unlimited, we just have to stick in the BRAZERO
4576 and do no more at this point. However, we do need to adjust any
4577 OP_RECURSE calls inside the group that refer to the group itself or any
4578 internal or forward referenced group, because the offset is from the
4579 start of the whole regex. Temporarily terminate the pattern while doing
4580 this. */
4581
4582 if (repeat_max <= 1) /* Covers 0, 1, and unlimited */
4583 {
4584 *code = OP_END;
4585 adjust_recurse(previous, 1, utf8, cd, save_hwm);
4586 memmove(previous+1, previous, len);
4587 code++;
4588 if (repeat_max == 0)
4589 {
4590 *previous++ = OP_SKIPZERO;
4591 goto END_REPEAT;
4592 }
4593 brazeroptr = previous; /* Save for possessive optimizing */
4594 *previous++ = OP_BRAZERO + repeat_type;
4595 }
4596
4597 /* If the maximum is greater than 1 and limited, we have to replicate
4598 in a nested fashion, sticking OP_BRAZERO before each set of brackets.
4599 The first one has to be handled carefully because it's the original
4600 copy, which has to be moved up. The remainder can be handled by code
4601 that is common with the non-zero minimum case below. We have to
4602 adjust the value or repeat_max, since one less copy is required. Once
4603 again, we may have to adjust any OP_RECURSE calls inside the group. */
4604
4605 else
4606 {
4607 int offset;
4608 *code = OP_END;
4609 adjust_recurse(previous, 2 + LINK_SIZE, utf8, cd, save_hwm);
4610 memmove(previous + 2 + LINK_SIZE, previous, len);
4611 code += 2 + LINK_SIZE;
4612 *previous++ = OP_BRAZERO + repeat_type;
4613 *previous++ = OP_BRA;
4614
4615 /* We chain together the bracket offset fields that have to be
4616 filled in later when the ends of the brackets are reached. */
4617
4618 offset = (bralink == NULL)? 0 : (int)(previous - bralink);
4619 bralink = previous;
4620 PUTINC(previous, 0, offset);
4621 }
4622
4623 repeat_max--;
4624 }
4625
4626 /* If the minimum is greater than zero, replicate the group as many
4627 times as necessary, and adjust the maximum to the number of subsequent
4628 copies that we need. If we set a first char from the group, and didn't
4629 set a required char, copy the latter from the former. If there are any
4630 forward reference subroutine calls in the group, there will be entries on
4631 the workspace list; replicate these with an appropriate increment. */
4632
4633 else
4634 {
4635 if (repeat_min > 1)
4636 {
4637 /* In the pre-compile phase, we don't actually do the replication. We
4638 just adjust the length as if we had. Do some paranoid checks for
4639 potential integer overflow. The INT64_OR_DOUBLE type is a 64-bit
4640 integer type when available, otherwise double. */
4641
4642 if (lengthptr != NULL)
4643 {
4644 int delta = (repeat_min - 1)*length_prevgroup;
4645 if ((INT64_OR_DOUBLE)(repeat_min - 1)*
4646 (INT64_OR_DOUBLE)length_prevgroup >
4647 (INT64_OR_DOUBLE)INT_MAX ||
4648 OFLOW_MAX - *lengthptr < delta)
4649 {
4650 *errorcodeptr = ERR20;
4651 goto FAILED;
4652 }
4653 *lengthptr += delta;
4654 }
4655
4656 /* This is compiling for real */
4657
4658 else
4659 {
4660 if (groupsetfirstbyte && reqbyte < 0) reqbyte = firstbyte;
4661 for (i = 1; i < repeat_min; i++)
4662 {
4663 uschar *hc;
4664 uschar *this_hwm = cd->hwm;
4665 memcpy(code, previous, len);
4666 for (hc = save_hwm; hc < this_hwm; hc += LINK_SIZE)
4667 {
4668 PUT(cd->hwm, 0, GET(hc, 0) + len);
4669 cd->hwm += LINK_SIZE;
4670 }
4671 save_hwm = this_hwm;
4672 code += len;
4673 }
4674 }
4675 }
4676
4677 if (repeat_max > 0) repeat_max -= repeat_min;
4678 }
4679
4680 /* This code is common to both the zero and non-zero minimum cases. If
4681 the maximum is limited, it replicates the group in a nested fashion,
4682 remembering the bracket starts on a stack. In the case of a zero minimum,
4683 the first one was set up above. In all cases the repeat_max now specifies
4684 the number of additional copies needed. Again, we must remember to
4685 replicate entries on the forward reference list. */
4686
4687 if (repeat_max >= 0)
4688 {
4689 /* In the pre-compile phase, we don't actually do the replication. We
4690 just adjust the length as if we had. For each repetition we must add 1
4691 to the length for BRAZERO and for all but the last repetition we must
4692 add 2 + 2*LINKSIZE to allow for the nesting that occurs. Do some
4693 paranoid checks to avoid integer overflow. The INT64_OR_DOUBLE type is
4694 a 64-bit integer type when available, otherwise double. */
4695
4696 if (lengthptr != NULL && repeat_max > 0)
4697 {
4698 int delta = repeat_max * (length_prevgroup + 1 + 2 + 2*LINK_SIZE) -
4699 2 - 2*LINK_SIZE; /* Last one doesn't nest */
4700 if ((INT64_OR_DOUBLE)repeat_max *
4701 (INT64_OR_DOUBLE)(length_prevgroup + 1 + 2 + 2*LINK_SIZE)
4702 > (INT64_OR_DOUBLE)INT_MAX ||
4703 OFLOW_MAX - *lengthptr < delta)
4704 {
4705 *errorcodeptr = ERR20;
4706 goto FAILED;
4707 }
4708 *lengthptr += delta;
4709 }
4710
4711 /* This is compiling for real */
4712
4713 else for (i = repeat_max - 1; i >= 0; i--)
4714 {
4715 uschar *hc;
4716 uschar *this_hwm = cd->hwm;
4717
4718 *code++ = OP_BRAZERO + repeat_type;
4719
4720 /* All but the final copy start a new nesting, maintaining the
4721 chain of brackets outstanding. */
4722
4723 if (i != 0)
4724 {
4725 int offset;
4726 *code++ = OP_BRA;
4727 offset = (bralink == NULL)? 0 : (int)(code - bralink);
4728 bralink = code;
4729 PUTINC(code, 0, offset);
4730 }
4731
4732 memcpy(code, previous, len);
4733 for (hc = save_hwm; hc < this_hwm; hc += LINK_SIZE)
4734 {
4735 PUT(cd->hwm, 0, GET(hc, 0) + len + ((i != 0)? 2+LINK_SIZE : 1));
4736 cd->hwm += LINK_SIZE;
4737 }
4738 save_hwm = this_hwm;
4739 code += len;
4740 }
4741
4742 /* Now chain through the pending brackets, and fill in their length
4743 fields (which are holding the chain links pro tem). */
4744
4745 while (bralink != NULL)
4746 {
4747 int oldlinkoffset;
4748 int offset = (int)(code - bralink + 1);
4749 uschar *bra = code - offset;
4750 oldlinkoffset = GET(bra, 1);
4751 bralink = (oldlinkoffset == 0)? NULL : bralink - oldlinkoffset;
4752 *code++ = OP_KET;
4753 PUTINC(code, 0, offset);
4754 PUT(bra, 1, offset);
4755 }
4756 }
4757
4758 /* If the maximum is unlimited, set a repeater in the final copy. For
4759 ONCE brackets, that's all we need to do. However, possessively repeated
4760 ONCE brackets can be converted into non-capturing brackets, as the
4761 behaviour of (?:xx)++ is the same as (?>xx)++ and this saves having to
4762 deal with possessive ONCEs specially.
4763
4764 Otherwise, if the quantifier was possessive, we convert the BRA code to
4765 the POS form, and the KET code to KETRPOS. (It turns out to be convenient
4766 at runtime to detect this kind of subpattern at both the start and at the
4767 end.) The use of special opcodes makes it possible to reduce greatly the
4768 stack usage in pcre_exec(). If the group is preceded by OP_BRAZERO,
4769 convert this to OP_BRAPOSZERO. Then cancel the possessive flag so that
4770 the default action below, of wrapping everything inside atomic brackets,
4771 does not happen.
4772
4773 Then, when we are doing the actual compile phase, check to see whether
4774 this group is one that could match an empty string. If so, convert the
4775 initial operator to the S form (e.g. OP_BRA -> OP_SBRA) so that runtime
4776 checking can be done. [This check is also applied to ONCE groups at
4777 runtime, but in a different way.] */
4778
4779 else
4780 {
4781 uschar *ketcode = code - 1 - LINK_SIZE;
4782 uschar *bracode = ketcode - GET(ketcode, 1);
4783
4784 if (*bracode == OP_ONCE && possessive_quantifier) *bracode = OP_BRA;
4785 if (*bracode == OP_ONCE)
4786 *ketcode = OP_KETRMAX + repeat_type;
4787 else
4788 {
4789 if (possessive_quantifier)
4790 {
4791 *bracode += 1; /* Switch to xxxPOS opcodes */
4792 *ketcode = OP_KETRPOS;
4793 if (brazeroptr != NULL) *brazeroptr = OP_BRAPOSZERO;
4794 possessive_quantifier = FALSE;
4795 }
4796 else *ketcode = OP_KETRMAX + repeat_type;
4797
4798 if (lengthptr == NULL)
4799 {
4800 uschar *scode = bracode;
4801 do
4802 {
4803 if (could_be_empty_branch(scode, ketcode, utf8, cd))
4804 {
4805 *bracode += OP_SBRA - OP_BRA;
4806 break;
4807 }
4808 scode += GET(scode, 1);
4809 }
4810 while (*scode == OP_ALT);
4811 }
4812 }
4813 }
4814 }
4815
4816 /* If previous is OP_FAIL, it was generated by an empty class [] in
4817 JavaScript mode. The other ways in which OP_FAIL can be generated, that is
4818 by (*FAIL) or (?!) set previous to NULL, which gives a "nothing to repeat"
4819 error above. We can just ignore the repeat in JS case. */
4820
4821 else if (*previous == OP_FAIL) goto END_REPEAT;
4822
4823 /* Else there's some kind of shambles */
4824
4825 else
4826 {
4827 *errorcodeptr = ERR11;
4828 goto FAILED;
4829 }
4830
4831 /* If the character following a repeat is '+', or if certain optimization
4832 tests above succeeded, possessive_quantifier is TRUE. For some opcodes,
4833 there are special alternative opcodes for this case. For anything else, we
4834 wrap the entire repeated item inside OP_ONCE brackets. Logically, the '+'
4835 notation is just syntactic sugar, taken from Sun's Java package, but the
4836 special opcodes can optimize it.
4837
4838 Possessively repeated subpatterns have already been handled in the code
4839 just above, so possessive_quantifier is always FALSE for them at this
4840 stage.
4841
4842 Note that the repeated item starts at tempcode, not at previous, which
4843 might be the first part of a string whose (former) last char we repeated.
4844
4845 Possessifying an 'exact' quantifier has no effect, so we can ignore it. But
4846 an 'upto' may follow. We skip over an 'exact' item, and then test the
4847 length of what remains before proceeding. */
4848
4849 if (possessive_quantifier)
4850 {
4851 int len;
4852
4853 if (*tempcode == OP_TYPEEXACT)
4854 tempcode += _pcre_OP_lengths[*tempcode] +
4855 ((tempcode[3] == OP_PROP || tempcode[3] == OP_NOTPROP)? 2 : 0);
4856
4857 else if (*tempcode == OP_EXACT || *tempcode == OP_NOTEXACT)
4858 {
4859 tempcode += _pcre_OP_lengths[*tempcode];
4860 #ifdef SUPPORT_UTF8
4861 if (utf8 && tempcode[-1] >= 0xc0)
4862 tempcode += _pcre_utf8_table4[tempcode[-1] & 0x3f];
4863 #endif
4864 }
4865
4866 len = (int)(code - tempcode);
4867 if (len > 0) switch (*tempcode)
4868 {
4869 case OP_STAR: *tempcode = OP_POSSTAR; break;
4870 case OP_PLUS: *tempcode = OP_POSPLUS; break;
4871 case OP_QUERY: *tempcode = OP_POSQUERY; break;
4872 case OP_UPTO: *tempcode = OP_POSUPTO; break;
4873
4874 case OP_STARI: *tempcode = OP_POSSTARI; break;
4875 case OP_PLUSI: *tempcode = OP_POSPLUSI; break;
4876 case OP_QUERYI: *tempcode = OP_POSQUERYI; break;
4877 case OP_UPTOI: *tempcode = OP_POSUPTOI; break;
4878
4879 case OP_NOTSTAR: *tempcode = OP_NOTPOSSTAR; break;
4880 case OP_NOTPLUS: *tempcode = OP_NOTPOSPLUS; break;
4881 case OP_NOTQUERY: *tempcode = OP_NOTPOSQUERY; break;
4882 case OP_NOTUPTO: *tempcode = OP_NOTPOSUPTO; break;
4883
4884 case OP_NOTSTARI: *tempcode = OP_NOTPOSSTARI; break;
4885 case OP_NOTPLUSI: *tempcode = OP_NOTPOSPLUSI; break;
4886 case OP_NOTQUERYI: *tempcode = OP_NOTPOSQUERYI; break;
4887 case OP_NOTUPTOI: *tempcode = OP_NOTPOSUPTOI; break;
4888
4889 case OP_TYPESTAR: *tempcode = OP_TYPEPOSSTAR; break;
4890 case OP_TYPEPLUS: *tempcode = OP_TYPEPOSPLUS; break;
4891 case OP_TYPEQUERY: *tempcode = OP_TYPEPOSQUERY; break;
4892 case OP_TYPEUPTO: *tempcode = OP_TYPEPOSUPTO; break;
4893
4894 /* Because we are moving code along, we must ensure that any
4895 pending recursive references are updated. */
4896
4897 default:
4898 *code = OP_END;
4899 adjust_recurse(tempcode, 1 + LINK_SIZE, utf8, cd, save_hwm);
4900 memmove(tempcode + 1+LINK_SIZE, tempcode, len);
4901 code += 1 + LINK_SIZE;
4902 len += 1 + LINK_SIZE;
4903 tempcode[0] = OP_ONCE;
4904 *code++ = OP_KET;
4905 PUTINC(code, 0, len);
4906 PUT(tempcode, 1, len);
4907 break;
4908 }
4909 }
4910
4911 /* In all case we no longer have a previous item. We also set the
4912 "follows varying string" flag for subsequently encountered reqbytes if
4913 it isn't already set and we have just passed a varying length item. */
4914
4915 END_REPEAT:
4916 previous = NULL;
4917 cd->req_varyopt |= reqvary;
4918 break;
4919
4920
4921 /* ===================================================================*/
4922 /* Start of nested parenthesized sub-expression, or comment or lookahead or
4923 lookbehind or option setting or condition or all the other extended
4924 parenthesis forms. */
4925
4926 case CHAR_LEFT_PARENTHESIS:
4927 newoptions = options;
4928 skipbytes = 0;
4929 bravalue = OP_CBRA;
4930 save_hwm = cd->hwm;
4931 reset_bracount = FALSE;
4932
4933 /* First deal with various "verbs" that can be introduced by '*'. */
4934
4935 if (*(++ptr) == CHAR_ASTERISK &&
4936 ((cd->ctypes[ptr[1]] & ctype_letter) != 0 || ptr[1] == ':'))
4937 {
4938 int i, namelen;
4939 int arglen = 0;
4940 const char *vn = verbnames;
4941 const uschar *name = ptr + 1;
4942 const uschar *arg = NULL;
4943 previous = NULL;
4944 while ((cd->ctypes[*++ptr] & ctype_letter) != 0) {};
4945 namelen = (int)(ptr - name);
4946
4947 if (*ptr == CHAR_COLON)
4948 {
4949 arg = ++ptr;
4950 while ((cd->ctypes[*ptr] & (ctype_letter|ctype_digit)) != 0
4951 || *ptr == '_') ptr++;
4952 arglen = (int)(ptr - arg);
4953 }
4954
4955 if (*ptr != CHAR_RIGHT_PARENTHESIS)
4956 {
4957 *errorcodeptr = ERR60;
4958 goto FAILED;
4959 }
4960
4961 /* Scan the table of verb names */
4962
4963 for (i = 0; i < verbcount; i++)
4964 {
4965 if (namelen == verbs[i].len &&
4966 strncmp((char *)name, vn, namelen) == 0)
4967 {
4968 /* Check for open captures before ACCEPT and convert it to
4969 ASSERT_ACCEPT if in an assertion. */
4970
4971 if (verbs[i].op == OP_ACCEPT)
4972 {
4973 open_capitem *oc;
4974 if (arglen != 0)
4975 {
4976 *errorcodeptr = ERR59;
4977 goto FAILED;
4978 }
4979 cd->had_accept = TRUE;
4980 for (oc = cd->open_caps; oc != NULL; oc = oc->next)
4981 {
4982 *code++ = OP_CLOSE;
4983 PUT2INC(code, 0, oc->number);
4984 }
4985 *code++ = (cd->assert_depth > 0)? OP_ASSERT_ACCEPT : OP_ACCEPT;
4986 }
4987
4988 /* Handle other cases with/without an argument */
4989
4990 else if (arglen == 0)
4991 {
4992 if (verbs[i].op < 0) /* Argument is mandatory */
4993 {
4994 *errorcodeptr = ERR66;
4995 goto FAILED;
4996 }
4997 *code = verbs[i].op;
4998 if (*code++ == OP_THEN)
4999 {
5000 PUT(code, 0, code - bcptr->current_branch - 1);
5001 code += LINK_SIZE;
5002 }
5003 }
5004
5005 else
5006 {
5007 if (verbs[i].op_arg < 0) /* Argument is forbidden */
5008 {
5009 *errorcodeptr = ERR59;
5010 goto FAILED;
5011 }
5012 *code = verbs[i].op_arg;
5013 if (*code++ == OP_THEN_ARG)
5014 {
5015 PUT(code, 0, code - bcptr->current_branch - 1);
5016 code += LINK_SIZE;
5017 }
5018 *code++ = arglen;
5019 memcpy(code, arg, arglen);
5020 code += arglen;
5021 *code++ = 0;
5022 }
5023
5024 break; /* Found verb, exit loop */
5025 }
5026
5027 vn += verbs[i].len + 1;
5028 }
5029
5030 if (i < verbcount) continue; /* Successfully handled a verb */
5031 *errorcodeptr = ERR60; /* Verb not recognized */
5032 goto FAILED;
5033 }
5034
5035 /* Deal with the extended parentheses; all are introduced by '?', and the
5036 appearance of any of them means that this is not a capturing group. */
5037
5038 else if (*ptr == CHAR_QUESTION_MARK)
5039 {
5040 int i, set, unset, namelen;
5041 int *optset;
5042 const uschar *name;
5043 uschar *slot;
5044
5045 switch (*(++ptr))
5046 {
5047 case CHAR_NUMBER_SIGN: /* Comment; skip to ket */
5048 ptr++;
5049 while (*ptr != 0 && *ptr != CHAR_RIGHT_PARENTHESIS) ptr++;
5050 if (*ptr == 0)
5051 {
5052 *errorcodeptr = ERR18;
5053 goto FAILED;
5054 }
5055 continue;
5056
5057
5058 /* ------------------------------------------------------------ */
5059 case CHAR_VERTICAL_LINE: /* Reset capture count for each branch */
5060 reset_bracount = TRUE;
5061 /* Fall through */
5062
5063 /* ------------------------------------------------------------ */
5064 case CHAR_COLON: /* Non-capturing bracket */
5065 bravalue = OP_BRA;
5066 ptr++;
5067 break;
5068
5069
5070 /* ------------------------------------------------------------ */
5071 case CHAR_LEFT_PARENTHESIS:
5072 bravalue = OP_COND; /* Conditional group */
5073
5074 /* A condition can be an assertion, a number (referring to a numbered
5075 group), a name (referring to a named group), or 'R', referring to
5076 recursion. R<digits> and R&name are also permitted for recursion tests.
5077
5078 There are several syntaxes for testing a named group: (?(name)) is used
5079 by Python; Perl 5.10 onwards uses (?(<name>) or (?('name')).
5080
5081 There are two unfortunate ambiguities, caused by history. (a) 'R' can
5082 be the recursive thing or the name 'R' (and similarly for 'R' followed
5083 by digits), and (b) a number could be a name that consists of digits.
5084 In both cases, we look for a name first; if not found, we try the other
5085 cases. */
5086
5087 /* For conditions that are assertions, check the syntax, and then exit
5088 the switch. This will take control down to where bracketed groups,
5089 including assertions, are processed. */
5090
5091 if (ptr[1] == CHAR_QUESTION_MARK && (ptr[2] == CHAR_EQUALS_SIGN ||
5092 ptr[2] == CHAR_EXCLAMATION_MARK || ptr[2] == CHAR_LESS_THAN_SIGN))
5093 break;
5094
5095 /* Most other conditions use OP_CREF (a couple change to OP_RREF
5096 below), and all need to skip 3 bytes at the start of the group. */
5097
5098 code[1+LINK_SIZE] = OP_CREF;
5099 skipbytes = 3;
5100 refsign = -1;
5101
5102 /* Check for a test for recursion in a named group. */
5103
5104 if (ptr[1] == CHAR_R && ptr[2] == CHAR_AMPERSAND)
5105 {
5106 terminator = -1;
5107 ptr += 2;
5108 code[1+LINK_SIZE] = OP_RREF; /* Change the type of test */
5109 }
5110
5111 /* Check for a test for a named group's having been set, using the Perl
5112 syntax (?(<name>) or (?('name') */
5113
5114 else if (ptr[1] == CHAR_LESS_THAN_SIGN)
5115 {
5116 terminator = CHAR_GREATER_THAN_SIGN;
5117 ptr++;
5118 }
5119 else if (ptr[1] == CHAR_APOSTROPHE)
5120 {
5121 terminator = CHAR_APOSTROPHE;
5122 ptr++;
5123 }
5124 else
5125 {
5126 terminator = 0;
5127 if (ptr[1] == CHAR_MINUS || ptr[1] == CHAR_PLUS) refsign = *(++ptr);
5128 }
5129
5130 /* We now expect to read a name; any thing else is an error */
5131
5132 if ((cd->ctypes[ptr[1]] & ctype_word) == 0)
5133 {
5134 ptr += 1; /* To get the right offset */
5135 *errorcodeptr = ERR28;
5136 goto FAILED;
5137 }
5138
5139 /* Read the name, but also get it as a number if it's all digits */
5140
5141 recno = 0;
5142 name = ++ptr;
5143 while ((cd->ctypes[*ptr] & ctype_word) != 0)
5144 {
5145 if (recno >= 0)
5146 recno = ((digitab[*ptr] & ctype_digit) != 0)?
5147 recno * 10 + *ptr - CHAR_0 : -1;
5148 ptr++;
5149 }
5150 namelen = (int)(ptr - name);
5151
5152 if ((terminator > 0 && *ptr++ != terminator) ||
5153 *ptr++ != CHAR_RIGHT_PARENTHESIS)
5154 {
5155 ptr--; /* Error offset */
5156 *errorcodeptr = ERR26;
5157 goto FAILED;
5158 }
5159
5160 /* Do no further checking in the pre-compile phase. */
5161
5162 if (lengthptr != NULL) break;
5163
5164 /* In the real compile we do the work of looking for the actual
5165 reference. If the string started with "+" or "-" we require the rest to
5166 be digits, in which case recno will be set. */
5167
5168 if (refsign > 0)
5169 {
5170 if (recno <= 0)
5171 {
5172 *errorcodeptr = ERR58;
5173 goto FAILED;
5174 }
5175 recno = (refsign == CHAR_MINUS)?
5176 cd->bracount - recno + 1 : recno +cd->bracount;
5177 if (recno <= 0 || recno > cd->final_bracount)
5178 {
5179 *errorcodeptr = ERR15;
5180 goto FAILED;
5181 }
5182 PUT2(code, 2+LINK_SIZE, recno);
5183 break;
5184 }
5185
5186 /* Otherwise (did not start with "+" or "-"), start by looking for the
5187 name. If we find a name, add one to the opcode to change OP_CREF or
5188 OP_RREF into OP_NCREF or OP_NRREF. These behave exactly the same,
5189 except they record that the reference was originally to a name. The
5190 information is used to check duplicate names. */
5191
5192 slot = cd->name_table;
5193 for (i = 0; i < cd->names_found; i++)
5194 {
5195 if (strncmp((char *)name, (char *)slot+2, namelen) == 0) break;
5196 slot += cd->name_entry_size;
5197 }
5198
5199 /* Found a previous named subpattern */
5200
5201 if (i < cd->names_found)
5202 {
5203 recno = GET2(slot, 0);
5204 PUT2(code, 2+LINK_SIZE, recno);
5205 code[1+LINK_SIZE]++;
5206 }
5207
5208 /* Search the pattern for a forward reference */
5209
5210 else if ((i = find_parens(cd, name, namelen,
5211 (options & PCRE_EXTENDED) != 0, utf8)) > 0)
5212 {
5213 PUT2(code, 2+LINK_SIZE, i);
5214 code[1+LINK_SIZE]++;
5215 }
5216
5217 /* If terminator == 0 it means that the name followed directly after
5218 the opening parenthesis [e.g. (?(abc)...] and in this case there are
5219 some further alternatives to try. For the cases where terminator != 0
5220 [things like (?(<name>... or (?('name')... or (?(R&name)... ] we have
5221 now checked all the possibilities, so give an error. */
5222
5223 else if (terminator != 0)
5224 {
5225 *errorcodeptr = ERR15;
5226 goto FAILED;
5227 }
5228
5229 /* Check for (?(R) for recursion. Allow digits after R to specify a
5230 specific group number. */
5231
5232 else if (*name == CHAR_R)
5233 {
5234 recno = 0;
5235 for (i = 1; i < namelen; i++)
5236 {
5237 if ((digitab[name[i]] & ctype_digit) == 0)
5238 {
5239 *errorcodeptr = ERR15;
5240 goto FAILED;
5241 }
5242 recno = recno * 10 + name[i] - CHAR_0;
5243 }
5244 if (recno == 0) recno = RREF_ANY;
5245 code[1+LINK_SIZE] = OP_RREF; /* Change test type */
5246 PUT2(code, 2+LINK_SIZE, recno);
5247 }
5248
5249 /* Similarly, check for the (?(DEFINE) "condition", which is always
5250 false. */
5251
5252 else if (namelen == 6 && strncmp((char *)name, STRING_DEFINE, 6) == 0)
5253 {
5254 code[1+LINK_SIZE] = OP_DEF;
5255 skipbytes = 1;
5256 }
5257
5258 /* Check for the "name" actually being a subpattern number. We are
5259 in the second pass here, so final_bracount is set. */
5260
5261 else if (recno > 0 && recno <= cd->final_bracount)
5262 {
5263 PUT2(code, 2+LINK_SIZE, recno);
5264 }
5265
5266 /* Either an unidentified subpattern, or a reference to (?(0) */
5267
5268 else
5269 {
5270 *errorcodeptr = (recno == 0)? ERR35: ERR15;
5271 goto FAILED;
5272 }
5273 break;
5274
5275
5276 /* ------------------------------------------------------------ */
5277 case CHAR_EQUALS_SIGN: /* Positive lookahead */
5278 bravalue = OP_ASSERT;
5279 cd->assert_depth += 1;
5280 ptr++;
5281 break;
5282
5283
5284 /* ------------------------------------------------------------ */
5285 case CHAR_EXCLAMATION_MARK: /* Negative lookahead */
5286 ptr++;
5287 if (*ptr == CHAR_RIGHT_PARENTHESIS) /* Optimize (?!) */
5288 {
5289 *code++ = OP_FAIL;
5290 previous = NULL;
5291 continue;
5292 }
5293 bravalue = OP_ASSERT_NOT;
5294 cd->assert_depth += 1;
5295 break;
5296
5297
5298 /* ------------------------------------------------------------ */
5299 case CHAR_LESS_THAN_SIGN: /* Lookbehind or named define */
5300 switch (ptr[1])
5301 {
5302 case CHAR_EQUALS_SIGN: /* Positive lookbehind */
5303 bravalue = OP_ASSERTBACK;
5304 cd->assert_depth += 1;
5305 ptr += 2;
5306 break;
5307
5308 case CHAR_EXCLAMATION_MARK: /* Negative lookbehind */
5309 bravalue = OP_ASSERTBACK_NOT;
5310 cd->assert_depth += 1;
5311 ptr += 2;
5312 break;
5313
5314 default: /* Could be name define, else bad */
5315 if ((cd->ctypes[ptr[1]] & ctype_word) != 0) goto DEFINE_NAME;
5316 ptr++; /* Correct offset for error */
5317 *errorcodeptr = ERR24;
5318 goto FAILED;
5319 }
5320 break;
5321
5322
5323 /* ------------------------------------------------------------ */
5324 case CHAR_GREATER_THAN_SIGN: /* One-time brackets */
5325 bravalue = OP_ONCE;
5326 ptr++;
5327 break;
5328
5329
5330 /* ------------------------------------------------------------ */
5331 case CHAR_C: /* Callout - may be followed by digits; */
5332 previous_callout = code; /* Save for later completion */
5333 after_manual_callout = 1; /* Skip one item before completing */
5334 *code++ = OP_CALLOUT;
5335 {
5336 int n = 0;
5337 while ((digitab[*(++ptr)] & ctype_digit) != 0)
5338 n = n * 10 + *ptr - CHAR_0;
5339 if (*ptr != CHAR_RIGHT_PARENTHESIS)
5340 {
5341 *errorcodeptr = ERR39;
5342 goto FAILED;
5343 }
5344 if (n > 255)
5345 {
5346 *errorcodeptr = ERR38;
5347 goto FAILED;
5348 }
5349 *code++ = n;
5350 PUT(code, 0, (int)(ptr - cd->start_pattern + 1)); /* Pattern offset */
5351 PUT(code, LINK_SIZE, 0); /* Default length */
5352 code += 2 * LINK_SIZE;
5353 }
5354 previous = NULL;
5355 continue;
5356
5357
5358 /* ------------------------------------------------------------ */
5359 case CHAR_P: /* Python-style named subpattern handling */
5360 if (*(++ptr) == CHAR_EQUALS_SIGN ||
5361 *ptr == CHAR_GREATER_THAN_SIGN) /* Reference or recursion */
5362 {
5363 is_recurse = *ptr == CHAR_GREATER_THAN_SIGN;
5364 terminator = CHAR_RIGHT_PARENTHESIS;
5365 goto NAMED_REF_OR_RECURSE;
5366 }
5367 else if (*ptr != CHAR_LESS_THAN_SIGN) /* Test for Python-style defn */
5368 {
5369 *errorcodeptr = ERR41;
5370 goto FAILED;
5371 }
5372 /* Fall through to handle (?P< as (?< is handled */
5373
5374
5375 /* ------------------------------------------------------------ */
5376 DEFINE_NAME: /* Come here from (?< handling */
5377 case CHAR_APOSTROPHE:
5378 {
5379 terminator = (*ptr == CHAR_LESS_THAN_SIGN)?
5380 CHAR_GREATER_THAN_SIGN : CHAR_APOSTROPHE;
5381 name = ++ptr;
5382
5383 while ((cd->ctypes[*ptr] & ctype_word) != 0) ptr++;
5384 namelen = (int)(ptr - name);
5385
5386 /* In the pre-compile phase, just do a syntax check. */
5387
5388 if (lengthptr != NULL)
5389 {
5390 if (*ptr != terminator)
5391 {
5392 *errorcodeptr = ERR42;
5393 goto FAILED;
5394 }
5395 if (cd->names_found >= MAX_NAME_COUNT)
5396 {
5397 *errorcodeptr = ERR49;
5398 goto FAILED;
5399 }
5400 if (namelen + 3 > cd->name_entry_size)
5401 {
5402 cd->name_entry_size = namelen + 3;
5403 if (namelen > MAX_NAME_SIZE)
5404 {
5405 *errorcodeptr = ERR48;
5406 goto FAILED;
5407 }
5408 }
5409 }
5410
5411 /* In the real compile, create the entry in the table, maintaining
5412 alphabetical order. Duplicate names for different numbers are
5413 permitted only if PCRE_DUPNAMES is set. Duplicate names for the same
5414 number are always OK. (An existing number can be re-used if (?|
5415 appears in the pattern.) In either event, a duplicate name results in
5416 a duplicate entry in the table, even if the number is the same. This
5417 is because the number of names, and hence the table size, is computed
5418 in the pre-compile, and it affects various numbers and pointers which
5419 would all have to be modified, and the compiled code moved down, if
5420 duplicates with the same number were omitted from the table. This
5421 doesn't seem worth the hassle. However, *different* names for the
5422 same number are not permitted. */
5423
5424 else
5425 {
5426 BOOL dupname = FALSE;
5427 slot = cd->name_table;
5428
5429 for (i = 0; i < cd->names_found; i++)
5430 {
5431 int crc = memcmp(name, slot+2, namelen);
5432 if (crc == 0)
5433 {
5434 if (slot[2+namelen] == 0)
5435 {
5436 if (GET2(slot, 0) != cd->bracount + 1 &&
5437 (options & PCRE_DUPNAMES) == 0)
5438 {
5439 *errorcodeptr = ERR43;
5440 goto FAILED;
5441 }
5442 else dupname = TRUE;
5443 }
5444 else crc = -1; /* Current name is a substring */
5445 }
5446
5447 /* Make space in the table and break the loop for an earlier
5448 name. For a duplicate or later name, carry on. We do this for
5449 duplicates so that in the simple case (when ?(| is not used) they
5450 are in order of their numbers. */
5451
5452 if (crc < 0)
5453 {
5454 memmove(slot + cd->name_entry_size, slot,
5455 (cd->names_found - i) * cd->name_entry_size);
5456 break;
5457 }
5458
5459 /* Continue the loop for a later or duplicate name */
5460
5461 slot += cd->name_entry_size;
5462 }
5463
5464 /* For non-duplicate names, check for a duplicate number before
5465 adding the new name. */
5466
5467 if (!dupname)
5468 {
5469 uschar *cslot = cd->name_table;
5470 for (i = 0; i < cd->names_found; i++)
5471 {
5472 if (cslot != slot)
5473 {
5474 if (GET2(cslot, 0) == cd->bracount + 1)
5475 {
5476 *errorcodeptr = ERR65;
5477 goto FAILED;
5478 }
5479 }
5480 else i--;
5481 cslot += cd->name_entry_size;
5482 }
5483 }
5484
5485 PUT2(slot, 0, cd->bracount + 1);
5486 memcpy(slot + 2, name, namelen);
5487 slot[2+namelen] = 0;
5488 }
5489 }
5490
5491 /* In both pre-compile and compile, count the number of names we've
5492 encountered. */
5493
5494 cd->names_found++;
5495 ptr++; /* Move past > or ' */
5496 goto NUMBERED_GROUP;
5497
5498
5499 /* ------------------------------------------------------------ */
5500 case CHAR_AMPERSAND: /* Perl recursion/subroutine syntax */
5501 terminator = CHAR_RIGHT_PARENTHESIS;
5502 is_recurse = TRUE;
5503 /* Fall through */
5504
5505 /* We come here from the Python syntax above that handles both
5506 references (?P=name) and recursion (?P>name), as well as falling
5507 through from the Perl recursion syntax (?&name). We also come here from
5508 the Perl \k<name> or \k'name' back reference syntax and the \k{name}
5509 .NET syntax, and the Oniguruma \g<...> and \g'...' subroutine syntax. */
5510
5511 NAMED_REF_OR_RECURSE:
5512 name = ++ptr;
5513 while ((cd->ctypes[*ptr] & ctype_word) != 0) ptr++;
5514 namelen = (int)(ptr - name);
5515
5516 /* In the pre-compile phase, do a syntax check. We used to just set
5517 a dummy reference number, because it was not used in the first pass.
5518 However, with the change of recursive back references to be atomic,
5519 we have to look for the number so that this state can be identified, as
5520 otherwise the incorrect length is computed. If it's not a backwards
5521 reference, the dummy number will do. */
5522
5523 if (lengthptr != NULL)
5524 {
5525 const uschar *temp;
5526
5527 if (namelen == 0)
5528 {
5529 *errorcodeptr = ERR62;
5530 goto FAILED;
5531 }
5532 if (*ptr != terminator)
5533 {
5534 *errorcodeptr = ERR42;
5535 goto FAILED;
5536 }
5537 if (namelen > MAX_NAME_SIZE)
5538 {
5539 *errorcodeptr = ERR48;
5540 goto FAILED;
5541 }
5542
5543 /* The name table does not exist in the first pass, so we cannot
5544 do a simple search as in the code below. Instead, we have to scan the
5545 pattern to find the number. It is important that we scan it only as
5546 far as we have got because the syntax of named subpatterns has not
5547 been checked for the rest of the pattern, and find_parens() assumes
5548 correct syntax. In any case, it's a waste of resources to scan
5549 further. We stop the scan at the current point by temporarily
5550 adjusting the value of cd->endpattern. */
5551
5552 temp = cd->end_pattern;
5553 cd->end_pattern = ptr;
5554 recno = find_parens(cd, name, namelen,
5555 (options & PCRE_EXTENDED) != 0, utf8);
5556 cd->end_pattern = temp;
5557 if (recno < 0) recno = 0; /* Forward ref; set dummy number */
5558 }
5559
5560 /* In the real compile, seek the name in the table. We check the name
5561 first, and then check that we have reached the end of the name in the
5562 table. That way, if the name that is longer than any in the table,
5563 the comparison will fail without reading beyond the table entry. */
5564
5565 else
5566 {
5567 slot = cd->name_table;
5568 for (i = 0; i < cd->names_found; i++)
5569 {
5570 if (strncmp((char *)name, (char *)slot+2, namelen) == 0 &&
5571 slot[2+namelen] == 0)
5572 break;
5573 slot += cd->name_entry_size;
5574 }
5575
5576 if (i < cd->names_found) /* Back reference */
5577 {
5578 recno = GET2(slot, 0);
5579 }
5580 else if ((recno = /* Forward back reference */
5581 find_parens(cd, name, namelen,
5582 (options & PCRE_EXTENDED) != 0, utf8)) <= 0)
5583 {
5584 *errorcodeptr = ERR15;
5585 goto FAILED;
5586 }
5587 }
5588
5589 /* In both phases, we can now go to the code than handles numerical
5590 recursion or backreferences. */
5591
5592 if (is_recurse) goto HANDLE_RECURSION;
5593 else goto HANDLE_REFERENCE;
5594
5595
5596 /* ------------------------------------------------------------ */
5597 case CHAR_R: /* Recursion */
5598 ptr++; /* Same as (?0) */
5599 /* Fall through */
5600
5601
5602 /* ------------------------------------------------------------ */
5603 case CHAR_MINUS: case CHAR_PLUS: /* Recursion or subroutine */
5604 case CHAR_0: case CHAR_1: case CHAR_2: case CHAR_3: case CHAR_4:
5605 case CHAR_5: case CHAR_6: case CHAR_7: case CHAR_8: case CHAR_9:
5606 {
5607 const uschar *called;
5608 terminator = CHAR_RIGHT_PARENTHESIS;
5609
5610 /* Come here from the \g<...> and \g'...' code (Oniguruma
5611 compatibility). However, the syntax has been checked to ensure that
5612 the ... are a (signed) number, so that neither ERR63 nor ERR29 will
5613 be called on this path, nor with the jump to OTHER_CHAR_AFTER_QUERY
5614 ever be taken. */
5615
5616 HANDLE_NUMERICAL_RECURSION:
5617
5618 if ((refsign = *ptr) == CHAR_PLUS)
5619 {
5620 ptr++;
5621 if ((digitab[*ptr] & ctype_digit) == 0)
5622 {
5623 *errorcodeptr = ERR63;
5624 goto FAILED;
5625 }
5626 }
5627 else if (refsign == CHAR_MINUS)
5628 {
5629 if ((digitab[ptr[1]] & ctype_digit) == 0)
5630 goto OTHER_CHAR_AFTER_QUERY;
5631 ptr++;
5632 }
5633
5634 recno = 0;
5635 while((digitab[*ptr] & ctype_digit) != 0)
5636 recno = recno * 10 + *ptr++ - CHAR_0;
5637
5638 if (*ptr != terminator)
5639 {
5640 *errorcodeptr = ERR29;
5641 goto FAILED;
5642 }
5643
5644 if (refsign == CHAR_MINUS)
5645 {
5646 if (recno == 0)
5647 {
5648 *errorcodeptr = ERR58;
5649 goto FAILED;
5650 }
5651 recno = cd->bracount - recno + 1;
5652 if (recno <= 0)
5653 {
5654 *errorcodeptr = ERR15;
5655 goto FAILED;
5656 }
5657 }
5658 else if (refsign == CHAR_PLUS)
5659 {
5660 if (recno == 0)
5661 {
5662 *errorcodeptr = ERR58;
5663 goto FAILED;
5664 }
5665 recno += cd->bracount;
5666 }
5667
5668 /* Come here from code above that handles a named recursion */
5669
5670 HANDLE_RECURSION:
5671
5672 previous = code;
5673 called = cd->start_code;
5674
5675 /* When we are actually compiling, find the bracket that is being
5676 referenced. Temporarily end the regex in case it doesn't exist before
5677 this point. If we end up with a forward reference, first check that
5678 the bracket does occur later so we can give the error (and position)
5679 now. Then remember this forward reference in the workspace so it can
5680 be filled in at the end. */
5681
5682 if (lengthptr == NULL)
5683 {
5684 *code = OP_END;
5685 if (recno != 0)
5686 called = _pcre_find_bracket(cd->start_code, utf8, recno);
5687
5688 /* Forward reference */
5689
5690 if (called == NULL)
5691 {
5692 if (find_parens(cd, NULL, recno,
5693 (options & PCRE_EXTENDED) != 0, utf8) < 0)
5694 {
5695 *errorcodeptr = ERR15;
5696 goto FAILED;
5697 }
5698
5699 /* Fudge the value of "called" so that when it is inserted as an
5700 offset below, what it actually inserted is the reference number
5701 of the group. Then remember the forward reference. */
5702
5703 called = cd->start_code + recno;
5704 PUTINC(cd->hwm, 0, (int)(code + 1 - cd->start_code));
5705 }
5706
5707 /* If not a forward reference, and the subpattern is still open,
5708 this is a recursive call. We check to see if this is a left
5709 recursion that could loop for ever, and diagnose that case. */
5710
5711 else if (GET(called, 1) == 0 &&
5712 could_be_empty(called, code, bcptr, utf8, cd))
5713 {
5714 *errorcodeptr = ERR40;
5715 goto FAILED;
5716 }
5717 }
5718
5719 /* Insert the recursion/subroutine item. */
5720
5721 *code = OP_RECURSE;
5722 PUT(code, 1, (int)(called - cd->start_code));
5723 code += 1 + LINK_SIZE;
5724 }
5725
5726 /* Can't determine a first byte now */
5727
5728 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
5729 continue;
5730
5731
5732 /* ------------------------------------------------------------ */
5733 default: /* Other characters: check option setting */
5734 OTHER_CHAR_AFTER_QUERY:
5735 set = unset = 0;
5736 optset = &set;
5737
5738 while (*ptr != CHAR_RIGHT_PARENTHESIS && *ptr != CHAR_COLON)
5739 {
5740 switch (*ptr++)
5741 {
5742 case CHAR_MINUS: optset = &unset; break;
5743
5744 case CHAR_J: /* Record that it changed in the external options */
5745 *optset |= PCRE_DUPNAMES;
5746 cd->external_flags |= PCRE_JCHANGED;
5747 break;
5748
5749 case CHAR_i: *optset |= PCRE_CASELESS; break;
5750 case CHAR_m: *optset |= PCRE_MULTILINE; break;
5751 case CHAR_s: *optset |= PCRE_DOTALL; break;
5752 case CHAR_x: *optset |= PCRE_EXTENDED; break;
5753 case CHAR_U: *optset |= PCRE_UNGREEDY; break;
5754 case CHAR_X: *optset |= PCRE_EXTRA; break;
5755
5756 default: *errorcodeptr = ERR12;
5757 ptr--; /* Correct the offset */
5758 goto FAILED;
5759 }
5760 }
5761
5762 /* Set up the changed option bits, but don't change anything yet. */
5763
5764 newoptions = (options | set) & (~unset);
5765
5766 /* If the options ended with ')' this is not the start of a nested
5767 group with option changes, so the options change at this level. If this
5768 item is right at the start of the pattern, the options can be
5769 abstracted and made external in the pre-compile phase, and ignored in
5770 the compile phase. This can be helpful when matching -- for instance in
5771 caseless checking of required bytes.
5772
5773 If the code pointer is not (cd->start_code + 1 + LINK_SIZE), we are
5774 definitely *not* at the start of the pattern because something has been
5775 compiled. In the pre-compile phase, however, the code pointer can have
5776 that value after the start, because it gets reset as code is discarded
5777 during the pre-compile. However, this can happen only at top level - if
5778 we are within parentheses, the starting BRA will still be present. At
5779 any parenthesis level, the length value can be used to test if anything
5780 has been compiled at that level. Thus, a test for both these conditions
5781 is necessary to ensure we correctly detect the start of the pattern in
5782 both phases.
5783
5784 If we are not at the pattern start, reset the greedy defaults and the
5785 case value for firstbyte and reqbyte. */
5786
5787 if (*ptr == CHAR_RIGHT_PARENTHESIS)
5788 {
5789 if (code == cd->start_code + 1 + LINK_SIZE &&
5790 (lengthptr == NULL || *lengthptr == 2 + 2*LINK_SIZE))
5791 {
5792 cd->external_options = newoptions;
5793 }
5794 else
5795 {
5796 greedy_default = ((newoptions & PCRE_UNGREEDY) != 0);
5797 greedy_non_default = greedy_default ^ 1;
5798 req_caseopt = ((newoptions & PCRE_CASELESS) != 0)? REQ_CASELESS : 0;
5799 }
5800
5801 /* Change options at this level, and pass them back for use
5802 in subsequent branches. */
5803
5804 *optionsptr = options = newoptions;
5805 previous = NULL; /* This item can't be repeated */
5806 continue; /* It is complete */
5807 }
5808
5809 /* If the options ended with ':' we are heading into a nested group
5810 with possible change of options. Such groups are non-capturing and are
5811 not assertions of any kind. All we need to do is skip over the ':';
5812 the newoptions value is handled below. */
5813
5814 bravalue = OP_BRA;
5815 ptr++;
5816 } /* End of switch for character following (? */
5817 } /* End of (? handling */
5818
5819 /* Opening parenthesis not followed by '*' or '?'. If PCRE_NO_AUTO_CAPTURE
5820 is set, all unadorned brackets become non-capturing and behave like (?:...)
5821 brackets. */
5822
5823 else if ((options & PCRE_NO_AUTO_CAPTURE) != 0)
5824 {
5825 bravalue = OP_BRA;
5826 }
5827
5828 /* Else we have a capturing group. */
5829
5830 else
5831 {
5832 NUMBERED_GROUP:
5833 cd->bracount += 1;
5834 PUT2(code, 1+LINK_SIZE, cd->bracount);
5835 skipbytes = 2;
5836 }
5837
5838 /* Process nested bracketed regex. Assertions may not be repeated, but
5839 other kinds can be. All their opcodes are >= OP_ONCE. We copy code into a
5840 non-register variable (tempcode) in order to be able to pass its address
5841 because some compilers complain otherwise. */
5842
5843 previous = (bravalue >= OP_ONCE)? code : NULL;
5844 *code = bravalue;
5845 tempcode = code;
5846 tempreqvary = cd->req_varyopt; /* Save value before bracket */
5847 length_prevgroup = 0; /* Initialize for pre-compile phase */
5848
5849 if (!compile_regex(
5850 newoptions, /* The complete new option state */
5851 &tempcode, /* Where to put code (updated) */
5852 &ptr, /* Input pointer (updated) */
5853 errorcodeptr, /* Where to put an error message */
5854 (bravalue == OP_ASSERTBACK ||
5855 bravalue == OP_ASSERTBACK_NOT), /* TRUE if back assert */
5856 reset_bracount, /* True if (?| group */
5857 skipbytes, /* Skip over bracket number */
5858 &subfirstbyte, /* For possible first char */
5859 &subreqbyte, /* For possible last char */
5860 bcptr, /* Current branch chain */
5861 cd, /* Tables block */
5862 (lengthptr == NULL)? NULL : /* Actual compile phase */
5863 &length_prevgroup /* Pre-compile phase */
5864 ))
5865 goto FAILED;
5866
5867 if (bravalue >= OP_ASSERT && bravalue <= OP_ASSERTBACK_NOT)
5868 cd->assert_depth -= 1;
5869
5870 /* At the end of compiling, code is still pointing to the start of the
5871 group, while tempcode has been updated to point past the end of the group
5872 and any option resetting that may follow it. The pattern pointer (ptr)
5873 is on the bracket. */
5874
5875 /* If this is a conditional bracket, check that there are no more than
5876 two branches in the group, or just one if it's a DEFINE group. We do this
5877 in the real compile phase, not in the pre-pass, where the whole group may
5878 not be available. */
5879
5880 if (bravalue == OP_COND && lengthptr == NULL)
5881 {
5882 uschar *tc = code;
5883 int condcount = 0;
5884
5885 do {
5886 condcount++;
5887 tc += GET(tc,1);
5888 }
5889 while (*tc != OP_KET);
5890
5891 /* A DEFINE group is never obeyed inline (the "condition" is always
5892 false). It must have only one branch. */
5893
5894 if (code[LINK_SIZE+1] == OP_DEF)
5895 {
5896 if (condcount > 1)
5897 {
5898 *errorcodeptr = ERR54;
5899 goto FAILED;
5900 }
5901 bravalue = OP_DEF; /* Just a flag to suppress char handling below */
5902 }
5903
5904 /* A "normal" conditional group. If there is just one branch, we must not
5905 make use of its firstbyte or reqbyte, because this is equivalent to an
5906 empty second branch. */
5907
5908 else
5909 {
5910 if (condcount > 2)
5911 {
5912 *errorcodeptr = ERR27;
5913 goto FAILED;
5914 }
5915 if (condcount == 1) subfirstbyte = subreqbyte = REQ_NONE;
5916 }
5917 }
5918
5919 /* Error if hit end of pattern */
5920
5921 if (*ptr != CHAR_RIGHT_PARENTHESIS)
5922 {
5923 *errorcodeptr = ERR14;
5924 goto FAILED;
5925 }
5926
5927 /* In the pre-compile phase, update the length by the length of the group,
5928 less the brackets at either end. Then reduce the compiled code to just a
5929 set of non-capturing brackets so that it doesn't use much memory if it is
5930 duplicated by a quantifier.*/
5931
5932 if (lengthptr != NULL)
5933 {
5934 if (OFLOW_MAX - *lengthptr < length_prevgroup - 2 - 2*LINK_SIZE)
5935 {
5936 *errorcodeptr = ERR20;
5937 goto FAILED;
5938 }
5939 *lengthptr += length_prevgroup - 2 - 2*LINK_SIZE;
5940 code++; /* This already contains bravalue */
5941 PUTINC(code, 0, 1 + LINK_SIZE);
5942 *code++ = OP_KET;
5943 PUTINC(code, 0, 1 + LINK_SIZE);
5944 break; /* No need to waste time with special character handling */
5945 }
5946
5947 /* Otherwise update the main code pointer to the end of the group. */
5948
5949 code = tempcode;
5950
5951 /* For a DEFINE group, required and first character settings are not
5952 relevant. */
5953
5954 if (bravalue == OP_DEF) break;
5955
5956 /* Handle updating of the required and first characters for other types of
5957 group. Update for normal brackets of all kinds, and conditions with two
5958 branches (see code above). If the bracket is followed by a quantifier with
5959 zero repeat, we have to back off. Hence the definition of zeroreqbyte and
5960 zerofirstbyte outside the main loop so that they can be accessed for the
5961 back off. */
5962
5963 zeroreqbyte = reqbyte;
5964 zerofirstbyte = firstbyte;
5965 groupsetfirstbyte = FALSE;
5966
5967 if (bravalue >= OP_ONCE)
5968 {
5969 /* If we have not yet set a firstbyte in this branch, take it from the
5970 subpattern, remembering that it was set here so that a repeat of more
5971 than one can replicate it as reqbyte if necessary. If the subpattern has
5972 no firstbyte, set "none" for the whole branch. In both cases, a zero
5973 repeat forces firstbyte to "none". */
5974
5975 if (firstbyte == REQ_UNSET)
5976 {
5977 if (subfirstbyte >= 0)
5978 {
5979 firstbyte = subfirstbyte;
5980 groupsetfirstbyte = TRUE;
5981 }
5982 else firstbyte = REQ_NONE;
5983 zerofirstbyte = REQ_NONE;
5984 }
5985
5986 /* If firstbyte was previously set, convert the subpattern's firstbyte
5987 into reqbyte if there wasn't one, using the vary flag that was in
5988 existence beforehand. */
5989
5990 else if (subfirstbyte >= 0 && subreqbyte < 0)
5991 subreqbyte = subfirstbyte | tempreqvary;
5992
5993 /* If the subpattern set a required byte (or set a first byte that isn't
5994 really the first byte - see above), set it. */
5995
5996 if (subreqbyte >= 0) reqbyte = subreqbyte;
5997 }
5998
5999 /* For a forward assertion, we take the reqbyte, if set. This can be
6000 helpful if the pattern that follows the assertion doesn't set a different
6001 char. For example, it's useful for /(?=abcde).+/. We can't set firstbyte
6002 for an assertion, however because it leads to incorrect effect for patterns
6003 such as /(?=a)a.+/ when the "real" "a" would then become a reqbyte instead
6004 of a firstbyte. This is overcome by a scan at the end if there's no
6005 firstbyte, looking for an asserted first char. */
6006
6007 else if (bravalue == OP_ASSERT && subreqbyte >= 0) reqbyte = subreqbyte;
6008 break; /* End of processing '(' */
6009
6010
6011 /* ===================================================================*/
6012 /* Handle metasequences introduced by \. For ones like \d, the ESC_ values
6013 are arranged to be the negation of the corresponding OP_values in the
6014 default case when PCRE_UCP is not set. For the back references, the values
6015 are ESC_REF plus the reference number. Only back references and those types
6016 that consume a character may be repeated. We can test for values between
6017 ESC_b and ESC_Z for the latter; this may have to change if any new ones are
6018 ever created. */
6019
6020 case CHAR_BACKSLASH:
6021 tempptr = ptr;
6022 c = check_escape(&ptr, errorcodeptr, cd->bracount, options, FALSE);
6023 if (*errorcodeptr != 0) goto FAILED;
6024
6025 if (c < 0)
6026 {
6027 if (-c == ESC_Q) /* Handle start of quoted string */
6028 {
6029 if (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
6030 ptr += 2; /* avoid empty string */
6031 else inescq = TRUE;
6032 continue;
6033 }
6034
6035 if (-c == ESC_E) continue; /* Perl ignores an orphan \E */
6036
6037 /* For metasequences that actually match a character, we disable the
6038 setting of a first character if it hasn't already been set. */
6039
6040 if (firstbyte == REQ_UNSET && -c > ESC_b && -c < ESC_Z)
6041 firstbyte = REQ_NONE;
6042
6043 /* Set values to reset to if this is followed by a zero repeat. */
6044
6045 zerofirstbyte = firstbyte;
6046 zeroreqbyte = reqbyte;
6047
6048 /* \g<name> or \g'name' is a subroutine call by name and \g<n> or \g'n'
6049 is a subroutine call by number (Oniguruma syntax). In fact, the value
6050 -ESC_g is returned only for these cases. So we don't need to check for <
6051 or ' if the value is -ESC_g. For the Perl syntax \g{n} the value is
6052 -ESC_REF+n, and for the Perl syntax \g{name} the result is -ESC_k (as
6053 that is a synonym for a named back reference). */
6054
6055 if (-c == ESC_g)
6056 {
6057 const uschar *p;
6058 save_hwm = cd->hwm; /* Normally this is set when '(' is read */
6059 terminator = (*(++ptr) == CHAR_LESS_THAN_SIGN)?
6060 CHAR_GREATER_THAN_SIGN : CHAR_APOSTROPHE;
6061
6062 /* These two statements stop the compiler for warning about possibly
6063 unset variables caused by the jump to HANDLE_NUMERICAL_RECURSION. In
6064 fact, because we actually check for a number below, the paths that
6065 would actually be in error are never taken. */
6066
6067 skipbytes = 0;
6068 reset_bracount = FALSE;
6069
6070 /* Test for a name */
6071
6072 if (ptr[1] != CHAR_PLUS && ptr[1] != CHAR_MINUS)
6073 {
6074 BOOL isnumber = TRUE;
6075 for (p = ptr + 1; *p != 0 && *p != terminator; p++)
6076 {
6077 if ((cd->ctypes[*p] & ctype_digit) == 0) isnumber = FALSE;
6078 if ((cd->ctypes[*p] & ctype_word) == 0) break;
6079 }
6080 if (*p != terminator)
6081 {
6082 *errorcodeptr = ERR57;
6083 break;
6084 }
6085 if (isnumber)
6086 {
6087 ptr++;
6088 goto HANDLE_NUMERICAL_RECURSION;
6089 }
6090 is_recurse = TRUE;
6091 goto NAMED_REF_OR_RECURSE;
6092 }
6093
6094 /* Test a signed number in angle brackets or quotes. */
6095
6096 p = ptr + 2;
6097 while ((digitab[*p] & ctype_digit) != 0) p++;
6098 if (*p != terminator)
6099 {
6100 *errorcodeptr = ERR57;
6101 break;
6102 }
6103 ptr++;
6104 goto HANDLE_NUMERICAL_RECURSION;
6105 }
6106
6107 /* \k<name> or \k'name' is a back reference by name (Perl syntax).
6108 We also support \k{name} (.NET syntax) */
6109
6110 if (-c == ESC_k && (ptr[1] == CHAR_LESS_THAN_SIGN ||
6111 ptr[1] == CHAR_APOSTROPHE || ptr[1] == CHAR_LEFT_CURLY_BRACKET))
6112 {
6113 is_recurse = FALSE;
6114 terminator = (*(++ptr) == CHAR_LESS_THAN_SIGN)?
6115 CHAR_GREATER_THAN_SIGN : (*ptr == CHAR_APOSTROPHE)?
6116 CHAR_APOSTROPHE : CHAR_RIGHT_CURLY_BRACKET;
6117 goto NAMED_REF_OR_RECURSE;
6118 }
6119
6120 /* Back references are handled specially; must disable firstbyte if
6121 not set to cope with cases like (?=(\w+))\1: which would otherwise set
6122 ':' later. */
6123
6124 if (-c >= ESC_REF)
6125 {
6126 open_capitem *oc;
6127 recno = -c - ESC_REF;
6128
6129 HANDLE_REFERENCE: /* Come here from named backref handling */
6130 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
6131 previous = code;
6132 *code++ = ((options & PCRE_CASELESS) != 0)? OP_REFI : OP_REF;
6133 PUT2INC(code, 0, recno);
6134 cd->backref_map |= (recno < 32)? (1 << recno) : 1;
6135 if (recno > cd->top_backref) cd->top_backref = recno;
6136
6137 /* Check to see if this back reference is recursive, that it, it
6138 is inside the group that it references. A flag is set so that the
6139 group can be made atomic. */
6140
6141 for (oc = cd->open_caps; oc != NULL; oc = oc->next)
6142 {
6143 if (oc->number == recno)
6144 {
6145 oc->flag = TRUE;
6146 break;
6147 }
6148 }
6149 }
6150
6151 /* So are Unicode property matches, if supported. */
6152
6153 #ifdef SUPPORT_UCP
6154 else if (-c == ESC_P || -c == ESC_p)
6155 {
6156 BOOL negated;
6157 int pdata;
6158 int ptype = get_ucp(&ptr, &negated, &pdata, errorcodeptr);
6159 if (ptype < 0) goto FAILED;
6160 previous = code;
6161 *code++ = ((-c == ESC_p) != negated)? OP_PROP : OP_NOTPROP;
6162 *code++ = ptype;
6163 *code++ = pdata;
6164 }
6165 #else
6166
6167 /* If Unicode properties are not supported, \X, \P, and \p are not
6168 allowed. */
6169
6170 else if (-c == ESC_X || -c == ESC_P || -c == ESC_p)
6171 {
6172 *errorcodeptr = ERR45;
6173 goto FAILED;
6174 }
6175 #endif
6176
6177 /* For the rest (including \X when Unicode properties are supported), we
6178 can obtain the OP value by negating the escape value in the default
6179 situation when PCRE_UCP is not set. When it *is* set, we substitute
6180 Unicode property tests. */
6181
6182 else
6183 {
6184 #ifdef SUPPORT_UCP
6185 if (-c >= ESC_DU && -c <= ESC_wu)
6186 {
6187 nestptr = ptr + 1; /* Where to resume */
6188 ptr = substitutes[-c - ESC_DU] - 1; /* Just before substitute */
6189 }
6190 else
6191 #endif
6192 {
6193 previous = (-c > ESC_b && -c < ESC_Z)? code : NULL;
6194 *code++ = -c;
6195 }
6196 }
6197 continue;
6198 }
6199
6200 /* We have a data character whose value is in c. In UTF-8 mode it may have
6201 a value > 127. We set its representation in the length/buffer, and then
6202 handle it as a data character. */
6203
6204 #ifdef SUPPORT_UTF8
6205 if (utf8 && c > 127)
6206 mclength = _pcre_ord2utf8(c, mcbuffer);
6207 else
6208 #endif
6209
6210 {
6211 mcbuffer[0] = c;
6212 mclength = 1;
6213 }
6214 goto ONE_CHAR;
6215
6216
6217 /* ===================================================================*/
6218 /* Handle a literal character. It is guaranteed not to be whitespace or #
6219 when the extended flag is set. If we are in UTF-8 mode, it may be a
6220 multi-byte literal character. */
6221
6222 default:
6223 NORMAL_CHAR:
6224 mclength = 1;
6225 mcbuffer[0] = c;
6226
6227 #ifdef SUPPORT_UTF8
6228 if (utf8 && c >= 0xc0)
6229 {
6230 while ((ptr[1] & 0xc0) == 0x80)
6231 mcbuffer[mclength++] = *(++ptr);
6232 }
6233 #endif
6234
6235 /* At this point we have the character's bytes in mcbuffer, and the length
6236 in mclength. When not in UTF-8 mode, the length is always 1. */
6237
6238 ONE_CHAR:
6239 previous = code;
6240 *code++ = ((options & PCRE_CASELESS) != 0)? OP_CHARI : OP_CHAR;
6241 for (c = 0; c < mclength; c++) *code++ = mcbuffer[c];
6242
6243 /* Remember if \r or \n were seen */
6244
6245 if (mcbuffer[0] == CHAR_CR || mcbuffer[0] == CHAR_NL)
6246 cd->external_flags |= PCRE_HASCRORLF;
6247
6248 /* Set the first and required bytes appropriately. If no previous first
6249 byte, set it from this character, but revert to none on a zero repeat.
6250 Otherwise, leave the firstbyte value alone, and don't change it on a zero
6251 repeat. */
6252
6253 if (firstbyte == REQ_UNSET)
6254 {
6255 zerofirstbyte = REQ_NONE;
6256 zeroreqbyte = reqbyte;
6257
6258 /* If the character is more than one byte long, we can set firstbyte
6259 only if it is not to be matched caselessly. */
6260
6261 if (mclength == 1 || req_caseopt == 0)
6262 {
6263 firstbyte = mcbuffer[0] | req_caseopt;
6264 if (mclength != 1) reqbyte = code[-1] | cd->req_varyopt;
6265 }
6266 else firstbyte = reqbyte = REQ_NONE;
6267 }
6268
6269 /* firstbyte was previously set; we can set reqbyte only the length is
6270 1 or the matching is caseful. */
6271
6272 else
6273 {
6274 zerofirstbyte = firstbyte;
6275 zeroreqbyte = reqbyte;
6276 if (mclength == 1 || req_caseopt == 0)
6277 reqbyte = code[-1] | req_caseopt | cd->req_varyopt;
6278 }
6279
6280 break; /* End of literal character handling */
6281 }
6282 } /* end of big loop */
6283
6284
6285 /* Control never reaches here by falling through, only by a goto for all the
6286 error states. Pass back the position in the pattern so that it can be displayed
6287 to the user for diagnosing the error. */
6288
6289 FAILED:
6290 *ptrptr = ptr;
6291 return FALSE;
6292 }
6293
6294
6295
6296
6297 /*************************************************
6298 * Compile sequence of alternatives *
6299 *************************************************/
6300
6301 /* On entry, ptr is pointing past the bracket character, but on return it
6302 points to the closing bracket, or vertical bar, or end of string. The code
6303 variable is pointing at the byte into which the BRA operator has been stored.
6304 This function is used during the pre-compile phase when we are trying to find
6305 out the amount of memory needed, as well as during the real compile phase. The
6306 value of lengthptr distinguishes the two phases.
6307
6308 Arguments:
6309 options option bits, including any changes for this subpattern
6310 codeptr -> the address of the current code pointer
6311 ptrptr -> the address of the current pattern pointer
6312 errorcodeptr -> pointer to error code variable
6313 lookbehind TRUE if this is a lookbehind assertion
6314 reset_bracount TRUE to reset the count for each branch
6315 skipbytes skip this many bytes at start (for brackets and OP_COND)
6316 firstbyteptr place to put the first required character, or a negative number
6317 reqbyteptr place to put the last required character, or a negative number
6318 bcptr pointer to the chain of currently open branches
6319 cd points to the data block with tables pointers etc.
6320 lengthptr NULL during the real compile phase
6321 points to length accumulator during pre-compile phase
6322
6323 Returns: TRUE on success
6324 */
6325
6326 static BOOL
6327 compile_regex(int options, uschar **codeptr, const uschar **ptrptr,
6328 int *errorcodeptr, BOOL lookbehind, BOOL reset_bracount, int skipbytes,
6329 int *firstbyteptr, int *reqbyteptr, branch_chain *bcptr, compile_data *cd,
6330 int *lengthptr)
6331 {
6332 const uschar *ptr = *ptrptr;
6333 uschar *code = *codeptr;
6334 uschar *last_branch = code;
6335 uschar *start_bracket = code;
6336 uschar *reverse_count = NULL;
6337 open_capitem capitem;
6338 int capnumber = 0;
6339 int firstbyte, reqbyte;
6340 int branchfirstbyte, branchreqbyte;
6341 int length;
6342 int orig_bracount;
6343 int max_bracount;
6344 branch_chain bc;
6345
6346 bc.outer = bcptr;
6347 bc.current_branch = code;
6348
6349 firstbyte = reqbyte = REQ_UNSET;
6350
6351 /* Accumulate the length for use in the pre-compile phase. Start with the
6352 length of the BRA and KET and any extra bytes that are required at the
6353 beginning. We accumulate in a local variable to save frequent testing of
6354 lenthptr for NULL. We cannot do this by looking at the value of code at the
6355 start and end of each alternative, because compiled items are discarded during
6356 the pre-compile phase so that the work space is not exceeded. */
6357
6358 length = 2 + 2*LINK_SIZE + skipbytes;
6359
6360 /* WARNING: If the above line is changed for any reason, you must also change
6361 the code that abstracts option settings at the start of the pattern and makes
6362 them global. It tests the value of length for (2 + 2*LINK_SIZE) in the
6363 pre-compile phase to find out whether anything has yet been compiled or not. */
6364
6365 /* If this is a capturing subpattern, add to the chain of open capturing items
6366 so that we can detect them if (*ACCEPT) is encountered. This is also used to
6367 detect groups that contain recursive back references to themselves. Note that
6368 only OP_CBRA need be tested here; changing this opcode to one of its variants,
6369 e.g. OP_SCBRAPOS, happens later, after the group has been compiled. */
6370
6371 if (*code == OP_CBRA)
6372 {
6373 capnumber = GET2(code, 1 + LINK_SIZE);
6374 capitem.number = capnumber;
6375 capitem.next = cd->open_caps;
6376 capitem.flag = FALSE;
6377 cd->open_caps = &capitem;
6378 }
6379
6380 /* Offset is set zero to mark that this bracket is still open */
6381
6382 PUT(code, 1, 0);
6383 code += 1 + LINK_SIZE + skipbytes;
6384
6385 /* Loop for each alternative branch */
6386
6387 orig_bracount = max_bracount = cd->bracount;
6388 for (;;)
6389 {
6390 /* For a (?| group, reset the capturing bracket count so that each branch
6391 uses the same numbers. */
6392
6393 if (reset_bracount) cd->bracount = orig_bracount;
6394
6395 /* Set up dummy OP_REVERSE if lookbehind assertion */
6396
6397 if (lookbehind)
6398 {
6399 *code++ = OP_REVERSE;
6400 reverse_count = code;
6401 PUTINC(code, 0, 0);
6402 length += 1 + LINK_SIZE;
6403 }
6404
6405 /* Now compile the branch; in the pre-compile phase its length gets added
6406 into the length. */
6407
6408 if (!compile_branch(&options, &code, &ptr, errorcodeptr, &branchfirstbyte,
6409 &branchreqbyte, &bc, cd, (lengthptr == NULL)? NULL : &length))
6410 {
6411 *ptrptr = ptr;
6412 return FALSE;
6413 }
6414
6415 /* Keep the highest bracket count in case (?| was used and some branch
6416 has fewer than the rest. */
6417
6418 if (cd->bracount > max_bracount) max_bracount = cd->bracount;
6419
6420 /* In the real compile phase, there is some post-processing to be done. */
6421
6422 if (lengthptr == NULL)
6423 {
6424 /* If this is the first branch, the firstbyte and reqbyte values for the
6425 branch become the values for the regex. */
6426
6427 if (*last_branch != OP_ALT)
6428 {
6429 firstbyte = branchfirstbyte;
6430 reqbyte = branchreqbyte;
6431 }
6432
6433 /* If this is not the first branch, the first char and reqbyte have to
6434 match the values from all the previous branches, except that if the
6435 previous value for reqbyte didn't have REQ_VARY set, it can still match,
6436 and we set REQ_VARY for the regex. */
6437
6438 else
6439 {
6440 /* If we previously had a firstbyte, but it doesn't match the new branch,
6441 we have to abandon the firstbyte for the regex, but if there was
6442 previously no reqbyte, it takes on the value of the old firstbyte. */
6443
6444 if (firstbyte >= 0 && firstbyte != branchfirstbyte)
6445 {
6446 if (reqbyte < 0) reqbyte = firstbyte;
6447 firstbyte = REQ_NONE;
6448 }
6449
6450 /* If we (now or from before) have no firstbyte, a firstbyte from the
6451 branch becomes a reqbyte if there isn't a branch reqbyte. */
6452
6453 if (firstbyte < 0 && branchfirstbyte >= 0 && branchreqbyte < 0)
6454 branchreqbyte = branchfirstbyte;
6455
6456 /* Now ensure that the reqbytes match */
6457
6458 if ((reqbyte & ~REQ_VARY) != (branchreqbyte & ~REQ_VARY))
6459 reqbyte = REQ_NONE;
6460 else reqbyte |= branchreqbyte; /* To "or" REQ_VARY */
6461 }
6462
6463 /* If lookbehind, check that this branch matches a fixed-length string, and
6464 put the length into the OP_REVERSE item. Temporarily mark the end of the
6465 branch with OP_END. If the branch contains OP_RECURSE, the result is -3
6466 because there may be forward references that we can't check here. Set a
6467 flag to cause another lookbehind check at the end. Why not do it all at the
6468 end? Because common, erroneous checks are picked up here and the offset of
6469 the problem can be shown. */
6470
6471 if (lookbehind)
6472 {
6473 int fixed_length;
6474 *code = OP_END;
6475 fixed_length = find_fixedlength(last_branch, (options & PCRE_UTF8) != 0,
6476 FALSE, cd);
6477 DPRINTF(("fixed length = %d\n", fixed_length));
6478 if (fixed_length == -3)
6479 {
6480 cd->check_lookbehind = TRUE;
6481 }
6482 else if (fixed_length < 0)
6483 {
6484 *errorcodeptr = (fixed_length == -2)? ERR36 : ERR25;
6485 *ptrptr = ptr;
6486 return FALSE;
6487 }
6488 else { PUT(reverse_count, 0, fixed_length); }
6489 }
6490 }
6491
6492 /* Reached end of expression, either ')' or end of pattern. In the real
6493 compile phase, go back through the alternative branches and reverse the chain
6494 of offsets, with the field in the BRA item now becoming an offset to the
6495 first alternative. If there are no alternatives, it points to the end of the
6496 group. The length in the terminating ket is always the length of the whole
6497 bracketed item. Return leaving the pointer at the terminating char. */
6498
6499 if (*ptr != CHAR_VERTICAL_LINE)
6500 {
6501 if (lengthptr == NULL)
6502 {
6503 int branch_length = (int)(code - last_branch);
6504 do
6505 {
6506 int prev_length = GET(last_branch, 1);
6507 PUT(last_branch, 1, branch_length);
6508 branch_length = prev_length;
6509 last_branch -= branch_length;
6510 }
6511 while (branch_length > 0);
6512 }
6513
6514 /* Fill in the ket */
6515
6516 *code = OP_KET;
6517 PUT(code, 1, (int)(code - start_bracket));
6518 code += 1 + LINK_SIZE;
6519
6520 /* If it was a capturing subpattern, check to see if it contained any
6521 recursive back references. If so, we must wrap it in atomic brackets.
6522 In any event, remove the block from the chain. */
6523
6524 if (capnumber > 0)
6525 {
6526 if (cd->open_caps->flag)
6527 {
6528 memmove(start_bracket + 1 + LINK_SIZE, start_bracket,
6529 code - start_bracket);
6530 *start_bracket = OP_ONCE;
6531 code += 1 + LINK_SIZE;
6532 PUT(start_bracket, 1, (int)(code - start_bracket));
6533 *code = OP_KET;
6534 PUT(code, 1, (int)(code - start_bracket));
6535 code += 1 + LINK_SIZE;
6536 length += 2 + 2*LINK_SIZE;
6537 }
6538 cd->open_caps = cd->open_caps->next;
6539 }
6540
6541 /* Retain the highest bracket number, in case resetting was used. */
6542
6543 cd->bracount = max_bracount;
6544
6545 /* Set values to pass back */
6546
6547 *codeptr = code;
6548 *ptrptr = ptr;
6549 *firstbyteptr = firstbyte;
6550 *reqbyteptr = reqbyte;
6551 if (lengthptr != NULL)
6552 {
6553 if (OFLOW_MAX - *lengthptr < length)
6554 {
6555 *errorcodeptr = ERR20;
6556 return FALSE;
6557 }
6558 *lengthptr += length;
6559 }
6560 return TRUE;
6561 }
6562
6563 /* Another branch follows. In the pre-compile phase, we can move the code
6564 pointer back to where it was for the start of the first branch. (That is,
6565 pretend that each branch is the only one.)
6566
6567 In the real compile phase, insert an ALT node. Its length field points back
6568 to the previous branch while the bracket remains open. At the end the chain
6569 is reversed. It's done like this so that the start of the bracket has a
6570 zero offset until it is closed, making it possible to detect recursion. */
6571
6572 if (lengthptr != NULL)
6573 {
6574 code = *codeptr + 1 + LINK_SIZE + skipbytes;
6575 length += 1 + LINK_SIZE;
6576 }
6577 else
6578 {
6579 *code = OP_ALT;
6580 PUT(code, 1, (int)(code - last_branch));
6581 bc.current_branch = last_branch = code;
6582 code += 1 + LINK_SIZE;
6583 }
6584
6585 ptr++;
6586 }
6587 /* Control never reaches here */
6588 }
6589
6590
6591
6592
6593 /*************************************************
6594 * Check for anchored expression *
6595 *************************************************/
6596
6597 /* Try to find out if this is an anchored regular expression. Consider each
6598 alternative branch. If they all start with OP_SOD or OP_CIRC, or with a bracket
6599 all of whose alternatives start with OP_SOD or OP_CIRC (recurse ad lib), then
6600 it's anchored. However, if this is a multiline pattern, then only OP_SOD will
6601 be found, because ^ generates OP_CIRCM in that mode.
6602
6603 We can also consider a regex to be anchored if OP_SOM starts all its branches.
6604 This is the code for \G, which means "match at start of match position, taking
6605 into account the match offset".
6606
6607 A branch is also implicitly anchored if it starts with .* and DOTALL is set,
6608 because that will try the rest of the pattern at all possible matching points,
6609 so there is no point trying again.... er ....
6610
6611 .... except when the .* appears inside capturing parentheses, and there is a
6612 subsequent back reference to those parentheses. We haven't enough information
6613 to catch that case precisely.
6614
6615 At first, the best we could do was to detect when .* was in capturing brackets
6616 and the highest back reference was greater than or equal to that level.
6617 However, by keeping a bitmap of the first 31 back references, we can catch some
6618 of the more common cases more precisely.
6619
6620 Arguments:
6621 code points to start of expression (the bracket)
6622 bracket_map a bitmap of which brackets we are inside while testing; this
6623 handles up to substring 31; after that we just have to take
6624 the less precise approach
6625 backref_map the back reference bitmap
6626
6627 Returns: TRUE or FALSE
6628 */
6629
6630 static BOOL
6631 is_anchored(register const uschar *code, unsigned int bracket_map,
6632 unsigned int backref_map)
6633 {
6634 do {
6635 const uschar *scode = first_significant_code(code + _pcre_OP_lengths[*code],
6636 FALSE);
6637 register int op = *scode;
6638
6639 /* Non-capturing brackets */
6640
6641 if (op == OP_BRA || op == OP_BRAPOS ||
6642 op == OP_SBRA || op == OP_SBRAPOS)
6643 {
6644 if (!is_anchored(scode, bracket_map, backref_map)) return FALSE;
6645 }
6646
6647 /* Capturing brackets */
6648
6649 else if (op == OP_CBRA || op == OP_CBRAPOS ||
6650 op == OP_SCBRA || op == OP_SCBRAPOS)
6651 {
6652 int n = GET2(scode, 1+LINK_SIZE);
6653 int new_map = bracket_map | ((n < 32)? (1 << n) : 1);
6654 if (!is_anchored(scode, new_map, backref_map)) return FALSE;
6655