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

Contents of /code/trunk/pcre_compile.c

Parent Directory Parent Directory | Revision Log Revision Log


Revision 530 - (show annotations) (download)
Tue Jun 1 13:42:06 2010 UTC (4 years, 4 months ago) by ph10
File MIME type: text/plain
File size: 229406 byte(s)
Added a lot of (int) casts to avoid compiler warnings in systems where      
size_t is 64-bit.

1 /*************************************************
2 * Perl-Compatible Regular Expressions *
3 *************************************************/
4
5 /* PCRE is a library of functions to support regular expressions whose syntax
6 and semantics are as close as possible to those of the Perl 5 language.
7
8 Written by Philip Hazel
9 Copyright (c) 1997-2010 University of Cambridge
10
11 -----------------------------------------------------------------------------
12 Redistribution and use in source and binary forms, with or without
13 modification, are permitted provided that the following conditions are met:
14
15 * Redistributions of source code must retain the above copyright notice,
16 this list of conditions and the following disclaimer.
17
18 * Redistributions in binary form must reproduce the above copyright
19 notice, this list of conditions and the following disclaimer in the
20 documentation and/or other materials provided with the distribution.
21
22 * Neither the name of the University of Cambridge nor the names of its
23 contributors may be used to endorse or promote products derived from
24 this software without specific prior written permission.
25
26 THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
27 AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28 IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29 ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE
30 LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
31 CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
32 SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
33 INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
34 CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
35 ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
36 POSSIBILITY OF SUCH DAMAGE.
37 -----------------------------------------------------------------------------
38 */
39
40
41 /* This module contains the external function pcre_compile(), along with
42 supporting internal functions that are not used by other modules. */
43
44
45 #ifdef HAVE_CONFIG_H
46 #include "config.h"
47 #endif
48
49 #define NLBLOCK cd /* Block containing newline information */
50 #define PSSTART start_pattern /* Field containing processed string start */
51 #define PSEND end_pattern /* Field containing processed string end */
52
53 #include "pcre_internal.h"
54
55
56 /* When PCRE_DEBUG is defined, we need the pcre_printint() function, which is
57 also used by pcretest. PCRE_DEBUG is not defined when building a production
58 library. */
59
60 #ifdef PCRE_DEBUG
61 #include "pcre_printint.src"
62 #endif
63
64
65 /* Macro for setting individual bits in class bitmaps. */
66
67 #define SETBIT(a,b) a[b/8] |= (1 << (b%8))
68
69 /* Maximum length value to check against when making sure that the integer that
70 holds the compiled pattern length does not overflow. We make it a bit less than
71 INT_MAX to allow for adding in group terminating bytes, so that we don't have
72 to check them every time. */
73
74 #define OFLOW_MAX (INT_MAX - 20)
75
76
77 /*************************************************
78 * Code parameters and static tables *
79 *************************************************/
80
81 /* This value specifies the size of stack workspace that is used during the
82 first pre-compile phase that determines how much memory is required. The regex
83 is partly compiled into this space, but the compiled parts are discarded as
84 soon as they can be, so that hopefully there will never be an overrun. The code
85 does, however, check for an overrun. The largest amount I've seen used is 218,
86 so this number is very generous.
87
88 The same workspace is used during the second, actual compile phase for
89 remembering forward references to groups so that they can be filled in at the
90 end. Each entry in this list occupies LINK_SIZE bytes, so even when LINK_SIZE
91 is 4 there is plenty of room. */
92
93 #define COMPILE_WORK_SIZE (4096)
94
95 /* The overrun tests check for a slightly smaller size so that they detect the
96 overrun before it actually does run off the end of the data block. */
97
98 #define WORK_SIZE_CHECK (COMPILE_WORK_SIZE - 100)
99
100
101 /* Table for handling escaped characters in the range '0'-'z'. Positive returns
102 are simple data values; negative values are for special things like \d and so
103 on. Zero means further processing is needed (for things like \x), or the escape
104 is invalid. */
105
106 #ifndef EBCDIC
107
108 /* This is the "normal" table for ASCII systems or for EBCDIC systems running
109 in UTF-8 mode. */
110
111 static const short int escapes[] = {
112 0, 0,
113 0, 0,
114 0, 0,
115 0, 0,
116 0, 0,
117 CHAR_COLON, CHAR_SEMICOLON,
118 CHAR_LESS_THAN_SIGN, CHAR_EQUALS_SIGN,
119 CHAR_GREATER_THAN_SIGN, CHAR_QUESTION_MARK,
120 CHAR_COMMERCIAL_AT, -ESC_A,
121 -ESC_B, -ESC_C,
122 -ESC_D, -ESC_E,
123 0, -ESC_G,
124 -ESC_H, 0,
125 0, -ESC_K,
126 0, 0,
127 -ESC_N, 0,
128 -ESC_P, -ESC_Q,
129 -ESC_R, -ESC_S,
130 0, 0,
131 -ESC_V, -ESC_W,
132 -ESC_X, 0,
133 -ESC_Z, CHAR_LEFT_SQUARE_BRACKET,
134 CHAR_BACKSLASH, CHAR_RIGHT_SQUARE_BRACKET,
135 CHAR_CIRCUMFLEX_ACCENT, CHAR_UNDERSCORE,
136 CHAR_GRAVE_ACCENT, 7,
137 -ESC_b, 0,
138 -ESC_d, ESC_e,
139 ESC_f, 0,
140 -ESC_h, 0,
141 0, -ESC_k,
142 0, 0,
143 ESC_n, 0,
144 -ESC_p, 0,
145 ESC_r, -ESC_s,
146 ESC_tee, 0,
147 -ESC_v, -ESC_w,
148 0, 0,
149 -ESC_z
150 };
151
152 #else
153
154 /* This is the "abnormal" table for EBCDIC systems without UTF-8 support. */
155
156 static const short int escapes[] = {
157 /* 48 */ 0, 0, 0, '.', '<', '(', '+', '|',
158 /* 50 */ '&', 0, 0, 0, 0, 0, 0, 0,
159 /* 58 */ 0, 0, '!', '$', '*', ')', ';', '~',
160 /* 60 */ '-', '/', 0, 0, 0, 0, 0, 0,
161 /* 68 */ 0, 0, '|', ',', '%', '_', '>', '?',
162 /* 70 */ 0, 0, 0, 0, 0, 0, 0, 0,
163 /* 78 */ 0, '`', ':', '#', '@', '\'', '=', '"',
164 /* 80 */ 0, 7, -ESC_b, 0, -ESC_d, ESC_e, ESC_f, 0,
165 /* 88 */-ESC_h, 0, 0, '{', 0, 0, 0, 0,
166 /* 90 */ 0, 0, -ESC_k, 'l', 0, ESC_n, 0, -ESC_p,
167 /* 98 */ 0, ESC_r, 0, '}', 0, 0, 0, 0,
168 /* A0 */ 0, '~', -ESC_s, ESC_tee, 0,-ESC_v, -ESC_w, 0,
169 /* A8 */ 0,-ESC_z, 0, 0, 0, '[', 0, 0,
170 /* B0 */ 0, 0, 0, 0, 0, 0, 0, 0,
171 /* B8 */ 0, 0, 0, 0, 0, ']', '=', '-',
172 /* C0 */ '{',-ESC_A, -ESC_B, -ESC_C, -ESC_D,-ESC_E, 0, -ESC_G,
173 /* C8 */-ESC_H, 0, 0, 0, 0, 0, 0, 0,
174 /* D0 */ '}', 0, -ESC_K, 0, 0,-ESC_N, 0, -ESC_P,
175 /* D8 */-ESC_Q,-ESC_R, 0, 0, 0, 0, 0, 0,
176 /* E0 */ '\\', 0, -ESC_S, 0, 0,-ESC_V, -ESC_W, -ESC_X,
177 /* E8 */ 0,-ESC_Z, 0, 0, 0, 0, 0, 0,
178 /* F0 */ 0, 0, 0, 0, 0, 0, 0, 0,
179 /* F8 */ 0, 0, 0, 0, 0, 0, 0, 0
180 };
181 #endif
182
183
184 /* Table of special "verbs" like (*PRUNE). This is a short table, so it is
185 searched linearly. Put all the names into a single string, in order to reduce
186 the number of relocations when a shared library is dynamically linked. The
187 string is built from string macros so that it works in UTF-8 mode on EBCDIC
188 platforms. */
189
190 typedef struct verbitem {
191 int len; /* Length of verb name */
192 int op; /* Op when no arg, or -1 if arg mandatory */
193 int op_arg; /* Op when arg present, or -1 if not allowed */
194 } verbitem;
195
196 static const char verbnames[] =
197 "\0" /* Empty name is a shorthand for MARK */
198 STRING_MARK0
199 STRING_ACCEPT0
200 STRING_COMMIT0
201 STRING_F0
202 STRING_FAIL0
203 STRING_PRUNE0
204 STRING_SKIP0
205 STRING_THEN;
206
207 static const verbitem verbs[] = {
208 { 0, -1, OP_MARK },
209 { 4, -1, OP_MARK },
210 { 6, OP_ACCEPT, -1 },
211 { 6, OP_COMMIT, -1 },
212 { 1, OP_FAIL, -1 },
213 { 4, OP_FAIL, -1 },
214 { 5, OP_PRUNE, OP_PRUNE_ARG },
215 { 4, OP_SKIP, OP_SKIP_ARG },
216 { 4, OP_THEN, OP_THEN_ARG }
217 };
218
219 static const int verbcount = sizeof(verbs)/sizeof(verbitem);
220
221
222 /* Tables of names of POSIX character classes and their lengths. The names are
223 now all in a single string, to reduce the number of relocations when a shared
224 library is dynamically loaded. The list of lengths is terminated by a zero
225 length entry. The first three must be alpha, lower, upper, as this is assumed
226 for handling case independence. */
227
228 static const char posix_names[] =
229 STRING_alpha0 STRING_lower0 STRING_upper0 STRING_alnum0
230 STRING_ascii0 STRING_blank0 STRING_cntrl0 STRING_digit0
231 STRING_graph0 STRING_print0 STRING_punct0 STRING_space0
232 STRING_word0 STRING_xdigit;
233
234 static const uschar posix_name_lengths[] = {
235 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 4, 6, 0 };
236
237 /* Table of class bit maps for each POSIX class. Each class is formed from a
238 base map, with an optional addition or removal of another map. Then, for some
239 classes, there is some additional tweaking: for [:blank:] the vertical space
240 characters are removed, and for [:alpha:] and [:alnum:] the underscore
241 character is removed. The triples in the table consist of the base map offset,
242 second map offset or -1 if no second map, and a non-negative value for map
243 addition or a negative value for map subtraction (if there are two maps). The
244 absolute value of the third field has these meanings: 0 => no tweaking, 1 =>
245 remove vertical space characters, 2 => remove underscore. */
246
247 static const int posix_class_maps[] = {
248 cbit_word, cbit_digit, -2, /* alpha */
249 cbit_lower, -1, 0, /* lower */
250 cbit_upper, -1, 0, /* upper */
251 cbit_word, -1, 2, /* alnum - word without underscore */
252 cbit_print, cbit_cntrl, 0, /* ascii */
253 cbit_space, -1, 1, /* blank - a GNU extension */
254 cbit_cntrl, -1, 0, /* cntrl */
255 cbit_digit, -1, 0, /* digit */
256 cbit_graph, -1, 0, /* graph */
257 cbit_print, -1, 0, /* print */
258 cbit_punct, -1, 0, /* punct */
259 cbit_space, -1, 0, /* space */
260 cbit_word, -1, 0, /* word - a Perl extension */
261 cbit_xdigit,-1, 0 /* xdigit */
262 };
263
264 /* Table of substitutes for \d etc when PCRE_UCP is set. The POSIX class
265 substitutes must be in the order of the names, defined above, and there are
266 both positive and negative cases. NULL means no substitute. */
267
268 #ifdef SUPPORT_UCP
269 static const uschar *substitutes[] = {
270 (uschar *)"\\P{Nd}", /* \D */
271 (uschar *)"\\p{Nd}", /* \d */
272 (uschar *)"\\P{Xsp}", /* \S */ /* NOTE: Xsp is Perl space */
273 (uschar *)"\\p{Xsp}", /* \s */
274 (uschar *)"\\P{Xwd}", /* \W */
275 (uschar *)"\\p{Xwd}" /* \w */
276 };
277
278 static const uschar *posix_substitutes[] = {
279 (uschar *)"\\p{L}", /* alpha */
280 (uschar *)"\\p{Ll}", /* lower */
281 (uschar *)"\\p{Lu}", /* upper */
282 (uschar *)"\\p{Xan}", /* alnum */
283 NULL, /* ascii */
284 (uschar *)"\\h", /* blank */
285 NULL, /* cntrl */
286 (uschar *)"\\p{Nd}", /* digit */
287 NULL, /* graph */
288 NULL, /* print */
289 NULL, /* punct */
290 (uschar *)"\\p{Xps}", /* space */ /* NOTE: Xps is POSIX space */
291 (uschar *)"\\p{Xwd}", /* word */
292 NULL, /* xdigit */
293 /* Negated cases */
294 (uschar *)"\\P{L}", /* ^alpha */
295 (uschar *)"\\P{Ll}", /* ^lower */
296 (uschar *)"\\P{Lu}", /* ^upper */
297 (uschar *)"\\P{Xan}", /* ^alnum */
298 NULL, /* ^ascii */
299 (uschar *)"\\H", /* ^blank */
300 NULL, /* ^cntrl */
301 (uschar *)"\\P{Nd}", /* ^digit */
302 NULL, /* ^graph */
303 NULL, /* ^print */
304 NULL, /* ^punct */
305 (uschar *)"\\P{Xps}", /* ^space */ /* NOTE: Xps is POSIX space */
306 (uschar *)"\\P{Xwd}", /* ^word */
307 NULL /* ^xdigit */
308 };
309 #define POSIX_SUBSIZE (sizeof(posix_substitutes)/sizeof(uschar *))
310 #endif
311
312 #define STRING(a) # a
313 #define XSTRING(s) STRING(s)
314
315 /* The texts of compile-time error messages. These are "char *" because they
316 are passed to the outside world. Do not ever re-use any error number, because
317 they are documented. Always add a new error instead. Messages marked DEAD below
318 are no longer used. This used to be a table of strings, but in order to reduce
319 the number of relocations needed when a shared library is loaded dynamically,
320 it is now one long string. We cannot use a table of offsets, because the
321 lengths of inserts such as XSTRING(MAX_NAME_SIZE) are not known. Instead, we
322 simply count through to the one we want - this isn't a performance issue
323 because these strings are used only when there is a compilation error.
324
325 Each substring ends with \0 to insert a null character. This includes the final
326 substring, so that the whole string ends with \0\0, which can be detected when
327 counting through. */
328
329 static const char error_texts[] =
330 "no error\0"
331 "\\ at end of pattern\0"
332 "\\c at end of pattern\0"
333 "unrecognized character follows \\\0"
334 "numbers out of order in {} quantifier\0"
335 /* 5 */
336 "number too big in {} quantifier\0"
337 "missing terminating ] for character class\0"
338 "invalid escape sequence in character class\0"
339 "range out of order in character class\0"
340 "nothing to repeat\0"
341 /* 10 */
342 "operand of unlimited repeat could match the empty string\0" /** DEAD **/
343 "internal error: unexpected repeat\0"
344 "unrecognized character after (? or (?-\0"
345 "POSIX named classes are supported only within a class\0"
346 "missing )\0"
347 /* 15 */
348 "reference to non-existent subpattern\0"
349 "erroffset passed as NULL\0"
350 "unknown option bit(s) set\0"
351 "missing ) after comment\0"
352 "parentheses nested too deeply\0" /** DEAD **/
353 /* 20 */
354 "regular expression is too large\0"
355 "failed to get memory\0"
356 "unmatched parentheses\0"
357 "internal error: code overflow\0"
358 "unrecognized character after (?<\0"
359 /* 25 */
360 "lookbehind assertion is not fixed length\0"
361 "malformed number or name after (?(\0"
362 "conditional group contains more than two branches\0"
363 "assertion expected after (?(\0"
364 "(?R or (?[+-]digits must be followed by )\0"
365 /* 30 */
366 "unknown POSIX class name\0"
367 "POSIX collating elements are not supported\0"
368 "this version of PCRE is not compiled with PCRE_UTF8 support\0"
369 "spare error\0" /** DEAD **/
370 "character value in \\x{...} sequence is too large\0"
371 /* 35 */
372 "invalid condition (?(0)\0"
373 "\\C not allowed in lookbehind assertion\0"
374 "PCRE does not support \\L, \\l, \\N{name}, \\U, or \\u\0"
375 "number after (?C is > 255\0"
376 "closing ) for (?C expected\0"
377 /* 40 */
378 "recursive call could loop indefinitely\0"
379 "unrecognized character after (?P\0"
380 "syntax error in subpattern name (missing terminator)\0"
381 "two named subpatterns have the same name\0"
382 "invalid UTF-8 string\0"
383 /* 45 */
384 "support for \\P, \\p, and \\X has not been compiled\0"
385 "malformed \\P or \\p sequence\0"
386 "unknown property name after \\P or \\p\0"
387 "subpattern name is too long (maximum " XSTRING(MAX_NAME_SIZE) " characters)\0"
388 "too many named subpatterns (maximum " XSTRING(MAX_NAME_COUNT) ")\0"
389 /* 50 */
390 "repeated subpattern is too long\0" /** DEAD **/
391 "octal value is greater than \\377 (not in UTF-8 mode)\0"
392 "internal error: overran compiling workspace\0"
393 "internal error: previously-checked referenced subpattern not found\0"
394 "DEFINE group contains more than one branch\0"
395 /* 55 */
396 "repeating a DEFINE group is not allowed\0"
397 "inconsistent NEWLINE options\0"
398 "\\g is not followed by a braced, angle-bracketed, or quoted name/number or by a plain number\0"
399 "a numbered reference must not be zero\0"
400 "an argument is not allowed for (*ACCEPT), (*FAIL), or (*COMMIT)\0"
401 /* 60 */
402 "(*VERB) not recognized\0"
403 "number is too big\0"
404 "subpattern name expected\0"
405 "digit expected after (?+\0"
406 "] is an invalid data character in JavaScript compatibility mode\0"
407 /* 65 */
408 "different names for subpatterns of the same number are not allowed\0"
409 "(*MARK) must have an argument\0"
410 "this version of PCRE is not compiled with PCRE_UCP support\0"
411 ;
412
413 /* Table to identify digits and hex digits. This is used when compiling
414 patterns. Note that the tables in chartables are dependent on the locale, and
415 may mark arbitrary characters as digits - but the PCRE compiling code expects
416 to handle only 0-9, a-z, and A-Z as digits when compiling. That is why we have
417 a private table here. It costs 256 bytes, but it is a lot faster than doing
418 character value tests (at least in some simple cases I timed), and in some
419 applications one wants PCRE to compile efficiently as well as match
420 efficiently.
421
422 For convenience, we use the same bit definitions as in chartables:
423
424 0x04 decimal digit
425 0x08 hexadecimal digit
426
427 Then we can use ctype_digit and ctype_xdigit in the code. */
428
429 #ifndef EBCDIC
430
431 /* This is the "normal" case, for ASCII systems, and EBCDIC systems running in
432 UTF-8 mode. */
433
434 static const unsigned char digitab[] =
435 {
436 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 0- 7 */
437 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 8- 15 */
438 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 16- 23 */
439 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 24- 31 */
440 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - ' */
441 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* ( - / */
442 0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c, /* 0 - 7 */
443 0x0c,0x0c,0x00,0x00,0x00,0x00,0x00,0x00, /* 8 - ? */
444 0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* @ - G */
445 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* H - O */
446 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* P - W */
447 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* X - _ */
448 0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* ` - g */
449 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* h - o */
450 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* p - w */
451 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* x -127 */
452 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 128-135 */
453 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 136-143 */
454 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 144-151 */
455 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 152-159 */
456 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 160-167 */
457 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 168-175 */
458 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 176-183 */
459 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 184-191 */
460 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 192-199 */
461 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 200-207 */
462 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 208-215 */
463 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 216-223 */
464 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 224-231 */
465 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 232-239 */
466 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 240-247 */
467 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00};/* 248-255 */
468
469 #else
470
471 /* This is the "abnormal" case, for EBCDIC systems not running in UTF-8 mode. */
472
473 static const unsigned char digitab[] =
474 {
475 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 0- 7 0 */
476 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 8- 15 */
477 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 16- 23 10 */
478 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 24- 31 */
479 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 32- 39 20 */
480 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 40- 47 */
481 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 48- 55 30 */
482 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 56- 63 */
483 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - 71 40 */
484 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 72- | */
485 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* & - 87 50 */
486 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 88- 95 */
487 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - -103 60 */
488 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 104- ? */
489 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 112-119 70 */
490 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 120- " */
491 0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* 128- g 80 */
492 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* h -143 */
493 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 144- p 90 */
494 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* q -159 */
495 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 160- x A0 */
496 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* y -175 */
497 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* ^ -183 B0 */
498 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 184-191 */
499 0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* { - G C0 */
500 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* H -207 */
501 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* } - P D0 */
502 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* Q -223 */
503 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* \ - X E0 */
504 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* Y -239 */
505 0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c, /* 0 - 7 F0 */
506 0x0c,0x0c,0x00,0x00,0x00,0x00,0x00,0x00};/* 8 -255 */
507
508 static const unsigned char ebcdic_chartab[] = { /* chartable partial dup */
509 0x80,0x00,0x00,0x00,0x00,0x01,0x00,0x00, /* 0- 7 */
510 0x00,0x00,0x00,0x00,0x01,0x01,0x00,0x00, /* 8- 15 */
511 0x00,0x00,0x00,0x00,0x00,0x01,0x00,0x00, /* 16- 23 */
512 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 24- 31 */
513 0x00,0x00,0x00,0x00,0x00,0x01,0x00,0x00, /* 32- 39 */
514 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 40- 47 */
515 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 48- 55 */
516 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 56- 63 */
517 0x01,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - 71 */
518 0x00,0x00,0x00,0x80,0x00,0x80,0x80,0x80, /* 72- | */
519 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* & - 87 */
520 0x00,0x00,0x00,0x80,0x80,0x80,0x00,0x00, /* 88- 95 */
521 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - -103 */
522 0x00,0x00,0x00,0x00,0x00,0x10,0x00,0x80, /* 104- ? */
523 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 112-119 */
524 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 120- " */
525 0x00,0x1a,0x1a,0x1a,0x1a,0x1a,0x1a,0x12, /* 128- g */
526 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* h -143 */
527 0x00,0x12,0x12,0x12,0x12,0x12,0x12,0x12, /* 144- p */
528 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* q -159 */
529 0x00,0x00,0x12,0x12,0x12,0x12,0x12,0x12, /* 160- x */
530 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* y -175 */
531 0x80,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* ^ -183 */
532 0x00,0x00,0x80,0x00,0x00,0x00,0x00,0x00, /* 184-191 */
533 0x80,0x1a,0x1a,0x1a,0x1a,0x1a,0x1a,0x12, /* { - G */
534 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* H -207 */
535 0x00,0x12,0x12,0x12,0x12,0x12,0x12,0x12, /* } - P */
536 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* Q -223 */
537 0x00,0x00,0x12,0x12,0x12,0x12,0x12,0x12, /* \ - X */
538 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* Y -239 */
539 0x1c,0x1c,0x1c,0x1c,0x1c,0x1c,0x1c,0x1c, /* 0 - 7 */
540 0x1c,0x1c,0x00,0x00,0x00,0x00,0x00,0x00};/* 8 -255 */
541 #endif
542
543
544 /* Definition to allow mutual recursion */
545
546 static BOOL
547 compile_regex(int, int, uschar **, const uschar **, int *, BOOL, BOOL, int,
548 int *, int *, branch_chain *, compile_data *, int *);
549
550
551
552 /*************************************************
553 * Find an error text *
554 *************************************************/
555
556 /* The error texts are now all in one long string, to save on relocations. As
557 some of the text is of unknown length, we can't use a table of offsets.
558 Instead, just count through the strings. This is not a performance issue
559 because it happens only when there has been a compilation error.
560
561 Argument: the error number
562 Returns: pointer to the error string
563 */
564
565 static const char *
566 find_error_text(int n)
567 {
568 const char *s = error_texts;
569 for (; n > 0; n--)
570 {
571 while (*s++ != 0) {};
572 if (*s == 0) return "Error text not found (please report)";
573 }
574 return s;
575 }
576
577
578 /*************************************************
579 * Handle escapes *
580 *************************************************/
581
582 /* This function is called when a \ has been encountered. It either returns a
583 positive value for a simple escape such as \n, or a negative value which
584 encodes one of the more complicated things such as \d. A backreference to group
585 n is returned as -(ESC_REF + n); ESC_REF is the highest ESC_xxx macro. When
586 UTF-8 is enabled, a positive value greater than 255 may be returned. On entry,
587 ptr is pointing at the \. On exit, it is on the final character of the escape
588 sequence.
589
590 Arguments:
591 ptrptr points to the pattern position pointer
592 errorcodeptr points to the errorcode variable
593 bracount number of previous extracting brackets
594 options the options bits
595 isclass TRUE if inside a character class
596
597 Returns: zero or positive => a data character
598 negative => a special escape sequence
599 on error, errorcodeptr is set
600 */
601
602 static int
603 check_escape(const uschar **ptrptr, int *errorcodeptr, int bracount,
604 int options, BOOL isclass)
605 {
606 BOOL utf8 = (options & PCRE_UTF8) != 0;
607 const uschar *ptr = *ptrptr + 1;
608 int c, i;
609
610 GETCHARINCTEST(c, ptr); /* Get character value, increment pointer */
611 ptr--; /* Set pointer back to the last byte */
612
613 /* If backslash is at the end of the pattern, it's an error. */
614
615 if (c == 0) *errorcodeptr = ERR1;
616
617 /* Non-alphanumerics are literals. For digits or letters, do an initial lookup
618 in a table. A non-zero result is something that can be returned immediately.
619 Otherwise further processing may be required. */
620
621 #ifndef EBCDIC /* ASCII/UTF-8 coding */
622 else if (c < CHAR_0 || c > CHAR_z) {} /* Not alphanumeric */
623 else if ((i = escapes[c - CHAR_0]) != 0) c = i;
624
625 #else /* EBCDIC coding */
626 else if (c < 'a' || (ebcdic_chartab[c] & 0x0E) == 0) {} /* Not alphanumeric */
627 else if ((i = escapes[c - 0x48]) != 0) c = i;
628 #endif
629
630 /* Escapes that need further processing, or are illegal. */
631
632 else
633 {
634 const uschar *oldptr;
635 BOOL braced, negated;
636
637 switch (c)
638 {
639 /* A number of Perl escapes are not handled by PCRE. We give an explicit
640 error. */
641
642 case CHAR_l:
643 case CHAR_L:
644 case CHAR_u:
645 case CHAR_U:
646 *errorcodeptr = ERR37;
647 break;
648
649 /* \g must be followed by one of a number of specific things:
650
651 (1) A number, either plain or braced. If positive, it is an absolute
652 backreference. If negative, it is a relative backreference. This is a Perl
653 5.10 feature.
654
655 (2) Perl 5.10 also supports \g{name} as a reference to a named group. This
656 is part of Perl's movement towards a unified syntax for back references. As
657 this is synonymous with \k{name}, we fudge it up by pretending it really
658 was \k.
659
660 (3) For Oniguruma compatibility we also support \g followed by a name or a
661 number either in angle brackets or in single quotes. However, these are
662 (possibly recursive) subroutine calls, _not_ backreferences. Just return
663 the -ESC_g code (cf \k). */
664
665 case CHAR_g:
666 if (ptr[1] == CHAR_LESS_THAN_SIGN || ptr[1] == CHAR_APOSTROPHE)
667 {
668 c = -ESC_g;
669 break;
670 }
671
672 /* Handle the Perl-compatible cases */
673
674 if (ptr[1] == CHAR_LEFT_CURLY_BRACKET)
675 {
676 const uschar *p;
677 for (p = ptr+2; *p != 0 && *p != CHAR_RIGHT_CURLY_BRACKET; p++)
678 if (*p != CHAR_MINUS && (digitab[*p] & ctype_digit) == 0) break;
679 if (*p != 0 && *p != CHAR_RIGHT_CURLY_BRACKET)
680 {
681 c = -ESC_k;
682 break;
683 }
684 braced = TRUE;
685 ptr++;
686 }
687 else braced = FALSE;
688
689 if (ptr[1] == CHAR_MINUS)
690 {
691 negated = TRUE;
692 ptr++;
693 }
694 else negated = FALSE;
695
696 c = 0;
697 while ((digitab[ptr[1]] & ctype_digit) != 0)
698 c = c * 10 + *(++ptr) - CHAR_0;
699
700 if (c < 0) /* Integer overflow */
701 {
702 *errorcodeptr = ERR61;
703 break;
704 }
705
706 if (braced && *(++ptr) != CHAR_RIGHT_CURLY_BRACKET)
707 {
708 *errorcodeptr = ERR57;
709 break;
710 }
711
712 if (c == 0)
713 {
714 *errorcodeptr = ERR58;
715 break;
716 }
717
718 if (negated)
719 {
720 if (c > bracount)
721 {
722 *errorcodeptr = ERR15;
723 break;
724 }
725 c = bracount - (c - 1);
726 }
727
728 c = -(ESC_REF + c);
729 break;
730
731 /* The handling of escape sequences consisting of a string of digits
732 starting with one that is not zero is not straightforward. By experiment,
733 the way Perl works seems to be as follows:
734
735 Outside a character class, the digits are read as a decimal number. If the
736 number is less than 10, or if there are that many previous extracting
737 left brackets, then it is a back reference. Otherwise, up to three octal
738 digits are read to form an escaped byte. Thus \123 is likely to be octal
739 123 (cf \0123, which is octal 012 followed by the literal 3). If the octal
740 value is greater than 377, the least significant 8 bits are taken. Inside a
741 character class, \ followed by a digit is always an octal number. */
742
743 case CHAR_1: case CHAR_2: case CHAR_3: case CHAR_4: case CHAR_5:
744 case CHAR_6: case CHAR_7: case CHAR_8: case CHAR_9:
745
746 if (!isclass)
747 {
748 oldptr = ptr;
749 c -= CHAR_0;
750 while ((digitab[ptr[1]] & ctype_digit) != 0)
751 c = c * 10 + *(++ptr) - CHAR_0;
752 if (c < 0) /* Integer overflow */
753 {
754 *errorcodeptr = ERR61;
755 break;
756 }
757 if (c < 10 || c <= bracount)
758 {
759 c = -(ESC_REF + c);
760 break;
761 }
762 ptr = oldptr; /* Put the pointer back and fall through */
763 }
764
765 /* Handle an octal number following \. If the first digit is 8 or 9, Perl
766 generates a binary zero byte and treats the digit as a following literal.
767 Thus we have to pull back the pointer by one. */
768
769 if ((c = *ptr) >= CHAR_8)
770 {
771 ptr--;
772 c = 0;
773 break;
774 }
775
776 /* \0 always starts an octal number, but we may drop through to here with a
777 larger first octal digit. The original code used just to take the least
778 significant 8 bits of octal numbers (I think this is what early Perls used
779 to do). Nowadays we allow for larger numbers in UTF-8 mode, but no more
780 than 3 octal digits. */
781
782 case CHAR_0:
783 c -= CHAR_0;
784 while(i++ < 2 && ptr[1] >= CHAR_0 && ptr[1] <= CHAR_7)
785 c = c * 8 + *(++ptr) - CHAR_0;
786 if (!utf8 && c > 255) *errorcodeptr = ERR51;
787 break;
788
789 /* \x is complicated. \x{ddd} is a character number which can be greater
790 than 0xff in utf8 mode, but only if the ddd are hex digits. If not, { is
791 treated as a data character. */
792
793 case CHAR_x:
794 if (ptr[1] == CHAR_LEFT_CURLY_BRACKET)
795 {
796 const uschar *pt = ptr + 2;
797 int count = 0;
798
799 c = 0;
800 while ((digitab[*pt] & ctype_xdigit) != 0)
801 {
802 register int cc = *pt++;
803 if (c == 0 && cc == CHAR_0) continue; /* Leading zeroes */
804 count++;
805
806 #ifndef EBCDIC /* ASCII/UTF-8 coding */
807 if (cc >= CHAR_a) cc -= 32; /* Convert to upper case */
808 c = (c << 4) + cc - ((cc < CHAR_A)? CHAR_0 : (CHAR_A - 10));
809 #else /* EBCDIC coding */
810 if (cc >= CHAR_a && cc <= CHAR_z) cc += 64; /* Convert to upper case */
811 c = (c << 4) + cc - ((cc >= CHAR_0)? CHAR_0 : (CHAR_A - 10));
812 #endif
813 }
814
815 if (*pt == CHAR_RIGHT_CURLY_BRACKET)
816 {
817 if (c < 0 || count > (utf8? 8 : 2)) *errorcodeptr = ERR34;
818 ptr = pt;
819 break;
820 }
821
822 /* If the sequence of hex digits does not end with '}', then we don't
823 recognize this construct; fall through to the normal \x handling. */
824 }
825
826 /* Read just a single-byte hex-defined char */
827
828 c = 0;
829 while (i++ < 2 && (digitab[ptr[1]] & ctype_xdigit) != 0)
830 {
831 int cc; /* Some compilers don't like */
832 cc = *(++ptr); /* ++ in initializers */
833 #ifndef EBCDIC /* ASCII/UTF-8 coding */
834 if (cc >= CHAR_a) cc -= 32; /* Convert to upper case */
835 c = c * 16 + cc - ((cc < CHAR_A)? CHAR_0 : (CHAR_A - 10));
836 #else /* EBCDIC coding */
837 if (cc <= CHAR_z) cc += 64; /* Convert to upper case */
838 c = c * 16 + cc - ((cc >= CHAR_0)? CHAR_0 : (CHAR_A - 10));
839 #endif
840 }
841 break;
842
843 /* For \c, a following letter is upper-cased; then the 0x40 bit is flipped.
844 This coding is ASCII-specific, but then the whole concept of \cx is
845 ASCII-specific. (However, an EBCDIC equivalent has now been added.) */
846
847 case CHAR_c:
848 c = *(++ptr);
849 if (c == 0)
850 {
851 *errorcodeptr = ERR2;
852 break;
853 }
854
855 #ifndef EBCDIC /* ASCII/UTF-8 coding */
856 if (c >= CHAR_a && c <= CHAR_z) c -= 32;
857 c ^= 0x40;
858 #else /* EBCDIC coding */
859 if (c >= CHAR_a && c <= CHAR_z) c += 64;
860 c ^= 0xC0;
861 #endif
862 break;
863
864 /* PCRE_EXTRA enables extensions to Perl in the matter of escapes. Any
865 other alphanumeric following \ is an error if PCRE_EXTRA was set;
866 otherwise, for Perl compatibility, it is a literal. This code looks a bit
867 odd, but there used to be some cases other than the default, and there may
868 be again in future, so I haven't "optimized" it. */
869
870 default:
871 if ((options & PCRE_EXTRA) != 0) switch(c)
872 {
873 default:
874 *errorcodeptr = ERR3;
875 break;
876 }
877 break;
878 }
879 }
880
881 /* Perl supports \N{name} for character names, as well as plain \N for "not
882 newline". PCRE does not support \N{name}. */
883
884 if (c == -ESC_N && ptr[1] == CHAR_LEFT_CURLY_BRACKET)
885 *errorcodeptr = ERR37;
886
887 /* If PCRE_UCP is set, we change the values for \d etc. */
888
889 if ((options & PCRE_UCP) != 0 && c <= -ESC_D && c >= -ESC_w)
890 c -= (ESC_DU - ESC_D);
891
892 /* Set the pointer to the final character before returning. */
893
894 *ptrptr = ptr;
895 return c;
896 }
897
898
899
900 #ifdef SUPPORT_UCP
901 /*************************************************
902 * Handle \P and \p *
903 *************************************************/
904
905 /* This function is called after \P or \p has been encountered, provided that
906 PCRE is compiled with support for Unicode properties. On entry, ptrptr is
907 pointing at the P or p. On exit, it is pointing at the final character of the
908 escape sequence.
909
910 Argument:
911 ptrptr points to the pattern position pointer
912 negptr points to a boolean that is set TRUE for negation else FALSE
913 dptr points to an int that is set to the detailed property value
914 errorcodeptr points to the error code variable
915
916 Returns: type value from ucp_type_table, or -1 for an invalid type
917 */
918
919 static int
920 get_ucp(const uschar **ptrptr, BOOL *negptr, int *dptr, int *errorcodeptr)
921 {
922 int c, i, bot, top;
923 const uschar *ptr = *ptrptr;
924 char name[32];
925
926 c = *(++ptr);
927 if (c == 0) goto ERROR_RETURN;
928
929 *negptr = FALSE;
930
931 /* \P or \p can be followed by a name in {}, optionally preceded by ^ for
932 negation. */
933
934 if (c == CHAR_LEFT_CURLY_BRACKET)
935 {
936 if (ptr[1] == CHAR_CIRCUMFLEX_ACCENT)
937 {
938 *negptr = TRUE;
939 ptr++;
940 }
941 for (i = 0; i < (int)sizeof(name) - 1; i++)
942 {
943 c = *(++ptr);
944 if (c == 0) goto ERROR_RETURN;
945 if (c == CHAR_RIGHT_CURLY_BRACKET) break;
946 name[i] = c;
947 }
948 if (c != CHAR_RIGHT_CURLY_BRACKET) goto ERROR_RETURN;
949 name[i] = 0;
950 }
951
952 /* Otherwise there is just one following character */
953
954 else
955 {
956 name[0] = c;
957 name[1] = 0;
958 }
959
960 *ptrptr = ptr;
961
962 /* Search for a recognized property name using binary chop */
963
964 bot = 0;
965 top = _pcre_utt_size;
966
967 while (bot < top)
968 {
969 i = (bot + top) >> 1;
970 c = strcmp(name, _pcre_utt_names + _pcre_utt[i].name_offset);
971 if (c == 0)
972 {
973 *dptr = _pcre_utt[i].value;
974 return _pcre_utt[i].type;
975 }
976 if (c > 0) bot = i + 1; else top = i;
977 }
978
979 *errorcodeptr = ERR47;
980 *ptrptr = ptr;
981 return -1;
982
983 ERROR_RETURN:
984 *errorcodeptr = ERR46;
985 *ptrptr = ptr;
986 return -1;
987 }
988 #endif
989
990
991
992
993 /*************************************************
994 * Check for counted repeat *
995 *************************************************/
996
997 /* This function is called when a '{' is encountered in a place where it might
998 start a quantifier. It looks ahead to see if it really is a quantifier or not.
999 It is only a quantifier if it is one of the forms {ddd} {ddd,} or {ddd,ddd}
1000 where the ddds are digits.
1001
1002 Arguments:
1003 p pointer to the first char after '{'
1004
1005 Returns: TRUE or FALSE
1006 */
1007
1008 static BOOL
1009 is_counted_repeat(const uschar *p)
1010 {
1011 if ((digitab[*p++] & ctype_digit) == 0) return FALSE;
1012 while ((digitab[*p] & ctype_digit) != 0) p++;
1013 if (*p == CHAR_RIGHT_CURLY_BRACKET) return TRUE;
1014
1015 if (*p++ != CHAR_COMMA) return FALSE;
1016 if (*p == CHAR_RIGHT_CURLY_BRACKET) return TRUE;
1017
1018 if ((digitab[*p++] & ctype_digit) == 0) return FALSE;
1019 while ((digitab[*p] & ctype_digit) != 0) p++;
1020
1021 return (*p == CHAR_RIGHT_CURLY_BRACKET);
1022 }
1023
1024
1025
1026 /*************************************************
1027 * Read repeat counts *
1028 *************************************************/
1029
1030 /* Read an item of the form {n,m} and return the values. This is called only
1031 after is_counted_repeat() has confirmed that a repeat-count quantifier exists,
1032 so the syntax is guaranteed to be correct, but we need to check the values.
1033
1034 Arguments:
1035 p pointer to first char after '{'
1036 minp pointer to int for min
1037 maxp pointer to int for max
1038 returned as -1 if no max
1039 errorcodeptr points to error code variable
1040
1041 Returns: pointer to '}' on success;
1042 current ptr on error, with errorcodeptr set non-zero
1043 */
1044
1045 static const uschar *
1046 read_repeat_counts(const uschar *p, int *minp, int *maxp, int *errorcodeptr)
1047 {
1048 int min = 0;
1049 int max = -1;
1050
1051 /* Read the minimum value and do a paranoid check: a negative value indicates
1052 an integer overflow. */
1053
1054 while ((digitab[*p] & ctype_digit) != 0) min = min * 10 + *p++ - CHAR_0;
1055 if (min < 0 || min > 65535)
1056 {
1057 *errorcodeptr = ERR5;
1058 return p;
1059 }
1060
1061 /* Read the maximum value if there is one, and again do a paranoid on its size.
1062 Also, max must not be less than min. */
1063
1064 if (*p == CHAR_RIGHT_CURLY_BRACKET) max = min; else
1065 {
1066 if (*(++p) != CHAR_RIGHT_CURLY_BRACKET)
1067 {
1068 max = 0;
1069 while((digitab[*p] & ctype_digit) != 0) max = max * 10 + *p++ - CHAR_0;
1070 if (max < 0 || max > 65535)
1071 {
1072 *errorcodeptr = ERR5;
1073 return p;
1074 }
1075 if (max < min)
1076 {
1077 *errorcodeptr = ERR4;
1078 return p;
1079 }
1080 }
1081 }
1082
1083 /* Fill in the required variables, and pass back the pointer to the terminating
1084 '}'. */
1085
1086 *minp = min;
1087 *maxp = max;
1088 return p;
1089 }
1090
1091
1092
1093 /*************************************************
1094 * Subroutine for finding forward reference *
1095 *************************************************/
1096
1097 /* This recursive function is called only from find_parens() below. The
1098 top-level call starts at the beginning of the pattern. All other calls must
1099 start at a parenthesis. It scans along a pattern's text looking for capturing
1100 subpatterns, and counting them. If it finds a named pattern that matches the
1101 name it is given, it returns its number. Alternatively, if the name is NULL, it
1102 returns when it reaches a given numbered subpattern. We know that if (?P< is
1103 encountered, the name will be terminated by '>' because that is checked in the
1104 first pass. Recursion is used to keep track of subpatterns that reset the
1105 capturing group numbers - the (?| feature.
1106
1107 Arguments:
1108 ptrptr address of the current character pointer (updated)
1109 cd compile background data
1110 name name to seek, or NULL if seeking a numbered subpattern
1111 lorn name length, or subpattern number if name is NULL
1112 xmode TRUE if we are in /x mode
1113 count pointer to the current capturing subpattern number (updated)
1114
1115 Returns: the number of the named subpattern, or -1 if not found
1116 */
1117
1118 static int
1119 find_parens_sub(uschar **ptrptr, compile_data *cd, const uschar *name, int lorn,
1120 BOOL xmode, int *count)
1121 {
1122 uschar *ptr = *ptrptr;
1123 int start_count = *count;
1124 int hwm_count = start_count;
1125 BOOL dup_parens = FALSE;
1126
1127 /* If the first character is a parenthesis, check on the type of group we are
1128 dealing with. The very first call may not start with a parenthesis. */
1129
1130 if (ptr[0] == CHAR_LEFT_PARENTHESIS)
1131 {
1132 if (ptr[1] == CHAR_QUESTION_MARK &&
1133 ptr[2] == CHAR_VERTICAL_LINE)
1134 {
1135 ptr += 3;
1136 dup_parens = TRUE;
1137 }
1138
1139 /* Handle a normal, unnamed capturing parenthesis */
1140
1141 else if (ptr[1] != CHAR_QUESTION_MARK && ptr[1] != CHAR_ASTERISK)
1142 {
1143 *count += 1;
1144 if (name == NULL && *count == lorn) return *count;
1145 ptr++;
1146 }
1147
1148 /* Handle a condition. If it is an assertion, just carry on so that it
1149 is processed as normal. If not, skip to the closing parenthesis of the
1150 condition (there can't be any nested parens. */
1151
1152 else if (ptr[2] == CHAR_LEFT_PARENTHESIS)
1153 {
1154 ptr += 2;
1155 if (ptr[1] != CHAR_QUESTION_MARK)
1156 {
1157 while (*ptr != 0 && *ptr != CHAR_RIGHT_PARENTHESIS) ptr++;
1158 if (*ptr != 0) ptr++;
1159 }
1160 }
1161
1162 /* We have either (? or (* and not a condition */
1163
1164 else
1165 {
1166 ptr += 2;
1167 if (*ptr == CHAR_P) ptr++; /* Allow optional P */
1168
1169 /* We have to disambiguate (?<! and (?<= from (?<name> for named groups */
1170
1171 if ((*ptr == CHAR_LESS_THAN_SIGN && ptr[1] != CHAR_EXCLAMATION_MARK &&
1172 ptr[1] != CHAR_EQUALS_SIGN) || *ptr == CHAR_APOSTROPHE)
1173 {
1174 int term;
1175 const uschar *thisname;
1176 *count += 1;
1177 if (name == NULL && *count == lorn) return *count;
1178 term = *ptr++;
1179 if (term == CHAR_LESS_THAN_SIGN) term = CHAR_GREATER_THAN_SIGN;
1180 thisname = ptr;
1181 while (*ptr != term) ptr++;
1182 if (name != NULL && lorn == ptr - thisname &&
1183 strncmp((const char *)name, (const char *)thisname, lorn) == 0)
1184 return *count;
1185 term++;
1186 }
1187 }
1188 }
1189
1190 /* Past any initial parenthesis handling, scan for parentheses or vertical
1191 bars. */
1192
1193 for (; *ptr != 0; ptr++)
1194 {
1195 /* Skip over backslashed characters and also entire \Q...\E */
1196
1197 if (*ptr == CHAR_BACKSLASH)
1198 {
1199 if (*(++ptr) == 0) goto FAIL_EXIT;
1200 if (*ptr == CHAR_Q) for (;;)
1201 {
1202 while (*(++ptr) != 0 && *ptr != CHAR_BACKSLASH) {};
1203 if (*ptr == 0) goto FAIL_EXIT;
1204 if (*(++ptr) == CHAR_E) break;
1205 }
1206 continue;
1207 }
1208
1209 /* Skip over character classes; this logic must be similar to the way they
1210 are handled for real. If the first character is '^', skip it. Also, if the
1211 first few characters (either before or after ^) are \Q\E or \E we skip them
1212 too. This makes for compatibility with Perl. Note the use of STR macros to
1213 encode "Q\\E" so that it works in UTF-8 on EBCDIC platforms. */
1214
1215 if (*ptr == CHAR_LEFT_SQUARE_BRACKET)
1216 {
1217 BOOL negate_class = FALSE;
1218 for (;;)
1219 {
1220 if (ptr[1] == CHAR_BACKSLASH)
1221 {
1222 if (ptr[2] == CHAR_E)
1223 ptr+= 2;
1224 else if (strncmp((const char *)ptr+2,
1225 STR_Q STR_BACKSLASH STR_E, 3) == 0)
1226 ptr += 4;
1227 else
1228 break;
1229 }
1230 else if (!negate_class && ptr[1] == CHAR_CIRCUMFLEX_ACCENT)
1231 {
1232 negate_class = TRUE;
1233 ptr++;
1234 }
1235 else break;
1236 }
1237
1238 /* If the next character is ']', it is a data character that must be
1239 skipped, except in JavaScript compatibility mode. */
1240
1241 if (ptr[1] == CHAR_RIGHT_SQUARE_BRACKET &&
1242 (cd->external_options & PCRE_JAVASCRIPT_COMPAT) == 0)
1243 ptr++;
1244
1245 while (*(++ptr) != CHAR_RIGHT_SQUARE_BRACKET)
1246 {
1247 if (*ptr == 0) return -1;
1248 if (*ptr == CHAR_BACKSLASH)
1249 {
1250 if (*(++ptr) == 0) goto FAIL_EXIT;
1251 if (*ptr == CHAR_Q) for (;;)
1252 {
1253 while (*(++ptr) != 0 && *ptr != CHAR_BACKSLASH) {};
1254 if (*ptr == 0) goto FAIL_EXIT;
1255 if (*(++ptr) == CHAR_E) break;
1256 }
1257 continue;
1258 }
1259 }
1260 continue;
1261 }
1262
1263 /* Skip comments in /x mode */
1264
1265 if (xmode && *ptr == CHAR_NUMBER_SIGN)
1266 {
1267 while (*(++ptr) != 0 && *ptr != CHAR_NL) {};
1268 if (*ptr == 0) goto FAIL_EXIT;
1269 continue;
1270 }
1271
1272 /* Check for the special metacharacters */
1273
1274 if (*ptr == CHAR_LEFT_PARENTHESIS)
1275 {
1276 int rc = find_parens_sub(&ptr, cd, name, lorn, xmode, count);
1277 if (rc > 0) return rc;
1278 if (*ptr == 0) goto FAIL_EXIT;
1279 }
1280
1281 else if (*ptr == CHAR_RIGHT_PARENTHESIS)
1282 {
1283 if (dup_parens && *count < hwm_count) *count = hwm_count;
1284 *ptrptr = ptr;
1285 return -1;
1286 }
1287
1288 else if (*ptr == CHAR_VERTICAL_LINE && dup_parens)
1289 {
1290 if (*count > hwm_count) hwm_count = *count;
1291 *count = start_count;
1292 }
1293 }
1294
1295 FAIL_EXIT:
1296 *ptrptr = ptr;
1297 return -1;
1298 }
1299
1300
1301
1302
1303 /*************************************************
1304 * Find forward referenced subpattern *
1305 *************************************************/
1306
1307 /* This function scans along a pattern's text looking for capturing
1308 subpatterns, and counting them. If it finds a named pattern that matches the
1309 name it is given, it returns its number. Alternatively, if the name is NULL, it
1310 returns when it reaches a given numbered subpattern. This is used for forward
1311 references to subpatterns. We used to be able to start this scan from the
1312 current compiling point, using the current count value from cd->bracount, and
1313 do it all in a single loop, but the addition of the possibility of duplicate
1314 subpattern numbers means that we have to scan from the very start, in order to
1315 take account of such duplicates, and to use a recursive function to keep track
1316 of the different types of group.
1317
1318 Arguments:
1319 cd compile background data
1320 name name to seek, or NULL if seeking a numbered subpattern
1321 lorn name length, or subpattern number if name is NULL
1322 xmode TRUE if we are in /x mode
1323
1324 Returns: the number of the found subpattern, or -1 if not found
1325 */
1326
1327 static int
1328 find_parens(compile_data *cd, const uschar *name, int lorn, BOOL xmode)
1329 {
1330 uschar *ptr = (uschar *)cd->start_pattern;
1331 int count = 0;
1332 int rc;
1333
1334 /* If the pattern does not start with an opening parenthesis, the first call
1335 to find_parens_sub() will scan right to the end (if necessary). However, if it
1336 does start with a parenthesis, find_parens_sub() will return when it hits the
1337 matching closing parens. That is why we have to have a loop. */
1338
1339 for (;;)
1340 {
1341 rc = find_parens_sub(&ptr, cd, name, lorn, xmode, &count);
1342 if (rc > 0 || *ptr++ == 0) break;
1343 }
1344
1345 return rc;
1346 }
1347
1348
1349
1350
1351 /*************************************************
1352 * Find first significant op code *
1353 *************************************************/
1354
1355 /* This is called by several functions that scan a compiled expression looking
1356 for a fixed first character, or an anchoring op code etc. It skips over things
1357 that do not influence this. For some calls, a change of option is important.
1358 For some calls, it makes sense to skip negative forward and all backward
1359 assertions, and also the \b assertion; for others it does not.
1360
1361 Arguments:
1362 code pointer to the start of the group
1363 options pointer to external options
1364 optbit the option bit whose changing is significant, or
1365 zero if none are
1366 skipassert TRUE if certain assertions are to be skipped
1367
1368 Returns: pointer to the first significant opcode
1369 */
1370
1371 static const uschar*
1372 first_significant_code(const uschar *code, int *options, int optbit,
1373 BOOL skipassert)
1374 {
1375 for (;;)
1376 {
1377 switch ((int)*code)
1378 {
1379 case OP_OPT:
1380 if (optbit > 0 && ((int)code[1] & optbit) != (*options & optbit))
1381 *options = (int)code[1];
1382 code += 2;
1383 break;
1384
1385 case OP_ASSERT_NOT:
1386 case OP_ASSERTBACK:
1387 case OP_ASSERTBACK_NOT:
1388 if (!skipassert) return code;
1389 do code += GET(code, 1); while (*code == OP_ALT);
1390 code += _pcre_OP_lengths[*code];
1391 break;
1392
1393 case OP_WORD_BOUNDARY:
1394 case OP_NOT_WORD_BOUNDARY:
1395 if (!skipassert) return code;
1396 /* Fall through */
1397
1398 case OP_CALLOUT:
1399 case OP_CREF:
1400 case OP_NCREF:
1401 case OP_RREF:
1402 case OP_NRREF:
1403 case OP_DEF:
1404 code += _pcre_OP_lengths[*code];
1405 break;
1406
1407 default:
1408 return code;
1409 }
1410 }
1411 /* Control never reaches here */
1412 }
1413
1414
1415
1416
1417 /*************************************************
1418 * Find the fixed length of a branch *
1419 *************************************************/
1420
1421 /* Scan a branch and compute the fixed length of subject that will match it,
1422 if the length is fixed. This is needed for dealing with backward assertions.
1423 In UTF8 mode, the result is in characters rather than bytes. The branch is
1424 temporarily terminated with OP_END when this function is called.
1425
1426 This function is called when a backward assertion is encountered, so that if it
1427 fails, the error message can point to the correct place in the pattern.
1428 However, we cannot do this when the assertion contains subroutine calls,
1429 because they can be forward references. We solve this by remembering this case
1430 and doing the check at the end; a flag specifies which mode we are running in.
1431
1432 Arguments:
1433 code points to the start of the pattern (the bracket)
1434 options the compiling options
1435 atend TRUE if called when the pattern is complete
1436 cd the "compile data" structure
1437
1438 Returns: the fixed length,
1439 or -1 if there is no fixed length,
1440 or -2 if \C was encountered
1441 or -3 if an OP_RECURSE item was encountered and atend is FALSE
1442 */
1443
1444 static int
1445 find_fixedlength(uschar *code, int options, BOOL atend, compile_data *cd)
1446 {
1447 int length = -1;
1448
1449 register int branchlength = 0;
1450 register uschar *cc = code + 1 + LINK_SIZE;
1451
1452 /* Scan along the opcodes for this branch. If we get to the end of the
1453 branch, check the length against that of the other branches. */
1454
1455 for (;;)
1456 {
1457 int d;
1458 uschar *ce, *cs;
1459 register int op = *cc;
1460 switch (op)
1461 {
1462 case OP_CBRA:
1463 case OP_BRA:
1464 case OP_ONCE:
1465 case OP_COND:
1466 d = find_fixedlength(cc + ((op == OP_CBRA)? 2:0), options, atend, cd);
1467 if (d < 0) return d;
1468 branchlength += d;
1469 do cc += GET(cc, 1); while (*cc == OP_ALT);
1470 cc += 1 + LINK_SIZE;
1471 break;
1472
1473 /* Reached end of a branch; if it's a ket it is the end of a nested
1474 call. If it's ALT it is an alternation in a nested call. If it is
1475 END it's the end of the outer call. All can be handled by the same code. */
1476
1477 case OP_ALT:
1478 case OP_KET:
1479 case OP_KETRMAX:
1480 case OP_KETRMIN:
1481 case OP_END:
1482 if (length < 0) length = branchlength;
1483 else if (length != branchlength) return -1;
1484 if (*cc != OP_ALT) return length;
1485 cc += 1 + LINK_SIZE;
1486 branchlength = 0;
1487 break;
1488
1489 /* A true recursion implies not fixed length, but a subroutine call may
1490 be OK. If the subroutine is a forward reference, we can't deal with
1491 it until the end of the pattern, so return -3. */
1492
1493 case OP_RECURSE:
1494 if (!atend) return -3;
1495 cs = ce = (uschar *)cd->start_code + GET(cc, 1); /* Start subpattern */
1496 do ce += GET(ce, 1); while (*ce == OP_ALT); /* End subpattern */
1497 if (cc > cs && cc < ce) return -1; /* Recursion */
1498 d = find_fixedlength(cs + 2, options, atend, cd);
1499 if (d < 0) return d;
1500 branchlength += d;
1501 cc += 1 + LINK_SIZE;
1502 break;
1503
1504 /* Skip over assertive subpatterns */
1505
1506 case OP_ASSERT:
1507 case OP_ASSERT_NOT:
1508 case OP_ASSERTBACK:
1509 case OP_ASSERTBACK_NOT:
1510 do cc += GET(cc, 1); while (*cc == OP_ALT);
1511 /* Fall through */
1512
1513 /* Skip over things that don't match chars */
1514
1515 case OP_REVERSE:
1516 case OP_CREF:
1517 case OP_NCREF:
1518 case OP_RREF:
1519 case OP_NRREF:
1520 case OP_DEF:
1521 case OP_OPT:
1522 case OP_CALLOUT:
1523 case OP_SOD:
1524 case OP_SOM:
1525 case OP_SET_SOM:
1526 case OP_EOD:
1527 case OP_EODN:
1528 case OP_CIRC:
1529 case OP_DOLL:
1530 case OP_NOT_WORD_BOUNDARY:
1531 case OP_WORD_BOUNDARY:
1532 cc += _pcre_OP_lengths[*cc];
1533 break;
1534
1535 /* Handle literal characters */
1536
1537 case OP_CHAR:
1538 case OP_CHARNC:
1539 case OP_NOT:
1540 branchlength++;
1541 cc += 2;
1542 #ifdef SUPPORT_UTF8
1543 if ((options & PCRE_UTF8) != 0 && cc[-1] >= 0xc0)
1544 cc += _pcre_utf8_table4[cc[-1] & 0x3f];
1545 #endif
1546 break;
1547
1548 /* Handle exact repetitions. The count is already in characters, but we
1549 need to skip over a multibyte character in UTF8 mode. */
1550
1551 case OP_EXACT:
1552 branchlength += GET2(cc,1);
1553 cc += 4;
1554 #ifdef SUPPORT_UTF8
1555 if ((options & PCRE_UTF8) != 0 && cc[-1] >= 0xc0)
1556 cc += _pcre_utf8_table4[cc[-1] & 0x3f];
1557 #endif
1558 break;
1559
1560 case OP_TYPEEXACT:
1561 branchlength += GET2(cc,1);
1562 if (cc[3] == OP_PROP || cc[3] == OP_NOTPROP) cc += 2;
1563 cc += 4;
1564 break;
1565
1566 /* Handle single-char matchers */
1567
1568 case OP_PROP:
1569 case OP_NOTPROP:
1570 cc += 2;
1571 /* Fall through */
1572
1573 case OP_NOT_DIGIT:
1574 case OP_DIGIT:
1575 case OP_NOT_WHITESPACE:
1576 case OP_WHITESPACE:
1577 case OP_NOT_WORDCHAR:
1578 case OP_WORDCHAR:
1579 case OP_ANY:
1580 case OP_ALLANY:
1581 branchlength++;
1582 cc++;
1583 break;
1584
1585 /* The single-byte matcher isn't allowed */
1586
1587 case OP_ANYBYTE:
1588 return -2;
1589
1590 /* Check a class for variable quantification */
1591
1592 #ifdef SUPPORT_UTF8
1593 case OP_XCLASS:
1594 cc += GET(cc, 1) - 33;
1595 /* Fall through */
1596 #endif
1597
1598 case OP_CLASS:
1599 case OP_NCLASS:
1600 cc += 33;
1601
1602 switch (*cc)
1603 {
1604 case OP_CRSTAR:
1605 case OP_CRMINSTAR:
1606 case OP_CRQUERY:
1607 case OP_CRMINQUERY:
1608 return -1;
1609
1610 case OP_CRRANGE:
1611 case OP_CRMINRANGE:
1612 if (GET2(cc,1) != GET2(cc,3)) return -1;
1613 branchlength += GET2(cc,1);
1614 cc += 5;
1615 break;
1616
1617 default:
1618 branchlength++;
1619 }
1620 break;
1621
1622 /* Anything else is variable length */
1623
1624 default:
1625 return -1;
1626 }
1627 }
1628 /* Control never gets here */
1629 }
1630
1631
1632
1633
1634 /*************************************************
1635 * Scan compiled regex for specific bracket *
1636 *************************************************/
1637
1638 /* This little function scans through a compiled pattern until it finds a
1639 capturing bracket with the given number, or, if the number is negative, an
1640 instance of OP_REVERSE for a lookbehind. The function is global in the C sense
1641 so that it can be called from pcre_study() when finding the minimum matching
1642 length.
1643
1644 Arguments:
1645 code points to start of expression
1646 utf8 TRUE in UTF-8 mode
1647 number the required bracket number or negative to find a lookbehind
1648
1649 Returns: pointer to the opcode for the bracket, or NULL if not found
1650 */
1651
1652 const uschar *
1653 _pcre_find_bracket(const uschar *code, BOOL utf8, int number)
1654 {
1655 for (;;)
1656 {
1657 register int c = *code;
1658 if (c == OP_END) return NULL;
1659
1660 /* XCLASS is used for classes that cannot be represented just by a bit
1661 map. This includes negated single high-valued characters. The length in
1662 the table is zero; the actual length is stored in the compiled code. */
1663
1664 if (c == OP_XCLASS) code += GET(code, 1);
1665
1666 /* Handle recursion */
1667
1668 else if (c == OP_REVERSE)
1669 {
1670 if (number < 0) return (uschar *)code;
1671 code += _pcre_OP_lengths[c];
1672 }
1673
1674 /* Handle capturing bracket */
1675
1676 else if (c == OP_CBRA)
1677 {
1678 int n = GET2(code, 1+LINK_SIZE);
1679 if (n == number) return (uschar *)code;
1680 code += _pcre_OP_lengths[c];
1681 }
1682
1683 /* Otherwise, we can get the item's length from the table, except that for
1684 repeated character types, we have to test for \p and \P, which have an extra
1685 two bytes of parameters, and for MARK/PRUNE/SKIP/THEN with an argument, we
1686 must add in its length. */
1687
1688 else
1689 {
1690 switch(c)
1691 {
1692 case OP_TYPESTAR:
1693 case OP_TYPEMINSTAR:
1694 case OP_TYPEPLUS:
1695 case OP_TYPEMINPLUS:
1696 case OP_TYPEQUERY:
1697 case OP_TYPEMINQUERY:
1698 case OP_TYPEPOSSTAR:
1699 case OP_TYPEPOSPLUS:
1700 case OP_TYPEPOSQUERY:
1701 if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
1702 break;
1703
1704 case OP_TYPEUPTO:
1705 case OP_TYPEMINUPTO:
1706 case OP_TYPEEXACT:
1707 case OP_TYPEPOSUPTO:
1708 if (code[3] == OP_PROP || code[3] == OP_NOTPROP) code += 2;
1709 break;
1710
1711 case OP_MARK:
1712 case OP_PRUNE_ARG:
1713 case OP_SKIP_ARG:
1714 case OP_THEN_ARG:
1715 code += code[1];
1716 break;
1717 }
1718
1719 /* Add in the fixed length from the table */
1720
1721 code += _pcre_OP_lengths[c];
1722
1723 /* In UTF-8 mode, opcodes that are followed by a character may be followed by
1724 a multi-byte character. The length in the table is a minimum, so we have to
1725 arrange to skip the extra bytes. */
1726
1727 #ifdef SUPPORT_UTF8
1728 if (utf8) switch(c)
1729 {
1730 case OP_CHAR:
1731 case OP_CHARNC:
1732 case OP_EXACT:
1733 case OP_UPTO:
1734 case OP_MINUPTO:
1735 case OP_POSUPTO:
1736 case OP_STAR:
1737 case OP_MINSTAR:
1738 case OP_POSSTAR:
1739 case OP_PLUS:
1740 case OP_MINPLUS:
1741 case OP_POSPLUS:
1742 case OP_QUERY:
1743 case OP_MINQUERY:
1744 case OP_POSQUERY:
1745 if (code[-1] >= 0xc0) code += _pcre_utf8_table4[code[-1] & 0x3f];
1746 break;
1747 }
1748 #else
1749 (void)(utf8); /* Keep compiler happy by referencing function argument */
1750 #endif
1751 }
1752 }
1753 }
1754
1755
1756
1757 /*************************************************
1758 * Scan compiled regex for recursion reference *
1759 *************************************************/
1760
1761 /* This little function scans through a compiled pattern until it finds an
1762 instance of OP_RECURSE.
1763
1764 Arguments:
1765 code points to start of expression
1766 utf8 TRUE in UTF-8 mode
1767
1768 Returns: pointer to the opcode for OP_RECURSE, or NULL if not found
1769 */
1770
1771 static const uschar *
1772 find_recurse(const uschar *code, BOOL utf8)
1773 {
1774 for (;;)
1775 {
1776 register int c = *code;
1777 if (c == OP_END) return NULL;
1778 if (c == OP_RECURSE) return code;
1779
1780 /* XCLASS is used for classes that cannot be represented just by a bit
1781 map. This includes negated single high-valued characters. The length in
1782 the table is zero; the actual length is stored in the compiled code. */
1783
1784 if (c == OP_XCLASS) code += GET(code, 1);
1785
1786 /* Otherwise, we can get the item's length from the table, except that for
1787 repeated character types, we have to test for \p and \P, which have an extra
1788 two bytes of parameters, and for MARK/PRUNE/SKIP/THEN with an argument, we
1789 must add in its length. */
1790
1791 else
1792 {
1793 switch(c)
1794 {
1795 case OP_TYPESTAR:
1796 case OP_TYPEMINSTAR:
1797 case OP_TYPEPLUS:
1798 case OP_TYPEMINPLUS:
1799 case OP_TYPEQUERY:
1800 case OP_TYPEMINQUERY:
1801 case OP_TYPEPOSSTAR:
1802 case OP_TYPEPOSPLUS:
1803 case OP_TYPEPOSQUERY:
1804 if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
1805 break;
1806
1807 case OP_TYPEPOSUPTO:
1808 case OP_TYPEUPTO:
1809 case OP_TYPEMINUPTO:
1810 case OP_TYPEEXACT:
1811 if (code[3] == OP_PROP || code[3] == OP_NOTPROP) code += 2;
1812 break;
1813
1814 case OP_MARK:
1815 case OP_PRUNE_ARG:
1816 case OP_SKIP_ARG:
1817 case OP_THEN_ARG:
1818 code += code[1];
1819 break;
1820 }
1821
1822 /* Add in the fixed length from the table */
1823
1824 code += _pcre_OP_lengths[c];
1825
1826 /* In UTF-8 mode, opcodes that are followed by a character may be followed
1827 by a multi-byte character. The length in the table is a minimum, so we have
1828 to arrange to skip the extra bytes. */
1829
1830 #ifdef SUPPORT_UTF8
1831 if (utf8) switch(c)
1832 {
1833 case OP_CHAR:
1834 case OP_CHARNC:
1835 case OP_EXACT:
1836 case OP_UPTO:
1837 case OP_MINUPTO:
1838 case OP_POSUPTO:
1839 case OP_STAR:
1840 case OP_MINSTAR:
1841 case OP_POSSTAR:
1842 case OP_PLUS:
1843 case OP_MINPLUS:
1844 case OP_POSPLUS:
1845 case OP_QUERY:
1846 case OP_MINQUERY:
1847 case OP_POSQUERY:
1848 if (code[-1] >= 0xc0) code += _pcre_utf8_table4[code[-1] & 0x3f];
1849 break;
1850 }
1851 #else
1852 (void)(utf8); /* Keep compiler happy by referencing function argument */
1853 #endif
1854 }
1855 }
1856 }
1857
1858
1859
1860 /*************************************************
1861 * Scan compiled branch for non-emptiness *
1862 *************************************************/
1863
1864 /* This function scans through a branch of a compiled pattern to see whether it
1865 can match the empty string or not. It is called from could_be_empty()
1866 below and from compile_branch() when checking for an unlimited repeat of a
1867 group that can match nothing. Note that first_significant_code() skips over
1868 backward and negative forward assertions when its final argument is TRUE. If we
1869 hit an unclosed bracket, we return "empty" - this means we've struck an inner
1870 bracket whose current branch will already have been scanned.
1871
1872 Arguments:
1873 code points to start of search
1874 endcode points to where to stop
1875 utf8 TRUE if in UTF8 mode
1876 cd contains pointers to tables etc.
1877
1878 Returns: TRUE if what is matched could be empty
1879 */
1880
1881 static BOOL
1882 could_be_empty_branch(const uschar *code, const uschar *endcode, BOOL utf8,
1883 compile_data *cd)
1884 {
1885 register int c;
1886 for (code = first_significant_code(code + _pcre_OP_lengths[*code], NULL, 0, TRUE);
1887 code < endcode;
1888 code = first_significant_code(code + _pcre_OP_lengths[c], NULL, 0, TRUE))
1889 {
1890 const uschar *ccode;
1891
1892 c = *code;
1893
1894 /* Skip over forward assertions; the other assertions are skipped by
1895 first_significant_code() with a TRUE final argument. */
1896
1897 if (c == OP_ASSERT)
1898 {
1899 do code += GET(code, 1); while (*code == OP_ALT);
1900 c = *code;
1901 continue;
1902 }
1903
1904 /* Groups with zero repeats can of course be empty; skip them. */
1905
1906 if (c == OP_BRAZERO || c == OP_BRAMINZERO || c == OP_SKIPZERO)
1907 {
1908 code += _pcre_OP_lengths[c];
1909 do code += GET(code, 1); while (*code == OP_ALT);
1910 c = *code;
1911 continue;
1912 }
1913
1914 /* For a recursion/subroutine call, if its end has been reached, which
1915 implies a subroutine call, we can scan it. */
1916
1917 if (c == OP_RECURSE)
1918 {
1919 BOOL empty_branch = FALSE;
1920 const uschar *scode = cd->start_code + GET(code, 1);
1921 if (GET(scode, 1) == 0) return TRUE; /* Unclosed */
1922 do
1923 {
1924 if (could_be_empty_branch(scode, endcode, utf8, cd))
1925 {
1926 empty_branch = TRUE;
1927 break;
1928 }
1929 scode += GET(scode, 1);
1930 }
1931 while (*scode == OP_ALT);
1932 if (!empty_branch) return FALSE; /* All branches are non-empty */
1933 continue;
1934 }
1935
1936 /* For other groups, scan the branches. */
1937
1938 if (c == OP_BRA || c == OP_CBRA || c == OP_ONCE || c == OP_COND)
1939 {
1940 BOOL empty_branch;
1941 if (GET(code, 1) == 0) return TRUE; /* Hit unclosed bracket */
1942
1943 /* If a conditional group has only one branch, there is a second, implied,
1944 empty branch, so just skip over the conditional, because it could be empty.
1945 Otherwise, scan the individual branches of the group. */
1946
1947 if (c == OP_COND && code[GET(code, 1)] != OP_ALT)
1948 code += GET(code, 1);
1949 else
1950 {
1951 empty_branch = FALSE;
1952 do
1953 {
1954 if (!empty_branch && could_be_empty_branch(code, endcode, utf8, cd))
1955 empty_branch = TRUE;
1956 code += GET(code, 1);
1957 }
1958 while (*code == OP_ALT);
1959 if (!empty_branch) return FALSE; /* All branches are non-empty */
1960 }
1961
1962 c = *code;
1963 continue;
1964 }
1965
1966 /* Handle the other opcodes */
1967
1968 switch (c)
1969 {
1970 /* Check for quantifiers after a class. XCLASS is used for classes that
1971 cannot be represented just by a bit map. This includes negated single
1972 high-valued characters. The length in _pcre_OP_lengths[] is zero; the
1973 actual length is stored in the compiled code, so we must update "code"
1974 here. */
1975
1976 #ifdef SUPPORT_UTF8
1977 case OP_XCLASS:
1978 ccode = code += GET(code, 1);
1979 goto CHECK_CLASS_REPEAT;
1980 #endif
1981
1982 case OP_CLASS:
1983 case OP_NCLASS:
1984 ccode = code + 33;
1985
1986 #ifdef SUPPORT_UTF8
1987 CHECK_CLASS_REPEAT:
1988 #endif
1989
1990 switch (*ccode)
1991 {
1992 case OP_CRSTAR: /* These could be empty; continue */
1993 case OP_CRMINSTAR:
1994 case OP_CRQUERY:
1995 case OP_CRMINQUERY:
1996 break;
1997
1998 default: /* Non-repeat => class must match */
1999 case OP_CRPLUS: /* These repeats aren't empty */
2000 case OP_CRMINPLUS:
2001 return FALSE;
2002
2003 case OP_CRRANGE:
2004 case OP_CRMINRANGE:
2005 if (GET2(ccode, 1) > 0) return FALSE; /* Minimum > 0 */
2006 break;
2007 }
2008 break;
2009
2010 /* Opcodes that must match a character */
2011
2012 case OP_PROP:
2013 case OP_NOTPROP:
2014 case OP_EXTUNI:
2015 case OP_NOT_DIGIT:
2016 case OP_DIGIT:
2017 case OP_NOT_WHITESPACE:
2018 case OP_WHITESPACE:
2019 case OP_NOT_WORDCHAR:
2020 case OP_WORDCHAR:
2021 case OP_ANY:
2022 case OP_ALLANY:
2023 case OP_ANYBYTE:
2024 case OP_CHAR:
2025 case OP_CHARNC:
2026 case OP_NOT:
2027 case OP_PLUS:
2028 case OP_MINPLUS:
2029 case OP_POSPLUS:
2030 case OP_EXACT:
2031 case OP_NOTPLUS:
2032 case OP_NOTMINPLUS:
2033 case OP_NOTPOSPLUS:
2034 case OP_NOTEXACT:
2035 case OP_TYPEPLUS:
2036 case OP_TYPEMINPLUS:
2037 case OP_TYPEPOSPLUS:
2038 case OP_TYPEEXACT:
2039 return FALSE;
2040
2041 /* These are going to continue, as they may be empty, but we have to
2042 fudge the length for the \p and \P cases. */
2043
2044 case OP_TYPESTAR:
2045 case OP_TYPEMINSTAR:
2046 case OP_TYPEPOSSTAR:
2047 case OP_TYPEQUERY:
2048 case OP_TYPEMINQUERY:
2049 case OP_TYPEPOSQUERY:
2050 if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
2051 break;
2052
2053 /* Same for these */
2054
2055 case OP_TYPEUPTO:
2056 case OP_TYPEMINUPTO:
2057 case OP_TYPEPOSUPTO:
2058 if (code[3] == OP_PROP || code[3] == OP_NOTPROP) code += 2;
2059 break;
2060
2061 /* End of branch */
2062
2063 case OP_KET:
2064 case OP_KETRMAX:
2065 case OP_KETRMIN:
2066 case OP_ALT:
2067 return TRUE;
2068
2069 /* In UTF-8 mode, STAR, MINSTAR, POSSTAR, QUERY, MINQUERY, POSQUERY, UPTO,
2070 MINUPTO, and POSUPTO may be followed by a multibyte character */
2071
2072 #ifdef SUPPORT_UTF8
2073 case OP_STAR:
2074 case OP_MINSTAR:
2075 case OP_POSSTAR:
2076 case OP_QUERY:
2077 case OP_MINQUERY:
2078 case OP_POSQUERY:
2079 if (utf8 && code[1] >= 0xc0) code += _pcre_utf8_table4[code[1] & 0x3f];
2080 break;
2081
2082 case OP_UPTO:
2083 case OP_MINUPTO:
2084 case OP_POSUPTO:
2085 if (utf8 && code[3] >= 0xc0) code += _pcre_utf8_table4[code[3] & 0x3f];
2086 break;
2087 #endif
2088
2089 /* MARK, and PRUNE/SKIP/THEN with an argument must skip over the argument
2090 string. */
2091
2092 case OP_MARK:
2093 case OP_PRUNE_ARG:
2094 case OP_SKIP_ARG:
2095 case OP_THEN_ARG:
2096 code += code[1];
2097 break;
2098
2099 /* None of the remaining opcodes are required to match a character. */
2100
2101 default:
2102 break;
2103 }
2104 }
2105
2106 return TRUE;
2107 }
2108
2109
2110
2111 /*************************************************
2112 * Scan compiled regex for non-emptiness *
2113 *************************************************/
2114
2115 /* This function is called to check for left recursive calls. We want to check
2116 the current branch of the current pattern to see if it could match the empty
2117 string. If it could, we must look outwards for branches at other levels,
2118 stopping when we pass beyond the bracket which is the subject of the recursion.
2119
2120 Arguments:
2121 code points to start of the recursion
2122 endcode points to where to stop (current RECURSE item)
2123 bcptr points to the chain of current (unclosed) branch starts
2124 utf8 TRUE if in UTF-8 mode
2125 cd pointers to tables etc
2126
2127 Returns: TRUE if what is matched could be empty
2128 */
2129
2130 static BOOL
2131 could_be_empty(const uschar *code, const uschar *endcode, branch_chain *bcptr,
2132 BOOL utf8, compile_data *cd)
2133 {
2134 while (bcptr != NULL && bcptr->current_branch >= code)
2135 {
2136 if (!could_be_empty_branch(bcptr->current_branch, endcode, utf8, cd))
2137 return FALSE;
2138 bcptr = bcptr->outer;
2139 }
2140 return TRUE;
2141 }
2142
2143
2144
2145 /*************************************************
2146 * Check for POSIX class syntax *
2147 *************************************************/
2148
2149 /* This function is called when the sequence "[:" or "[." or "[=" is
2150 encountered in a character class. It checks whether this is followed by a
2151 sequence of characters terminated by a matching ":]" or ".]" or "=]". If we
2152 reach an unescaped ']' without the special preceding character, return FALSE.
2153
2154 Originally, this function only recognized a sequence of letters between the
2155 terminators, but it seems that Perl recognizes any sequence of characters,
2156 though of course unknown POSIX names are subsequently rejected. Perl gives an
2157 "Unknown POSIX class" error for [:f\oo:] for example, where previously PCRE
2158 didn't consider this to be a POSIX class. Likewise for [:1234:].
2159
2160 The problem in trying to be exactly like Perl is in the handling of escapes. We
2161 have to be sure that [abc[:x\]pqr] is *not* treated as containing a POSIX
2162 class, but [abc[:x\]pqr:]] is (so that an error can be generated). The code
2163 below handles the special case of \], but does not try to do any other escape
2164 processing. This makes it different from Perl for cases such as [:l\ower:]
2165 where Perl recognizes it as the POSIX class "lower" but PCRE does not recognize
2166 "l\ower". This is a lesser evil that not diagnosing bad classes when Perl does,
2167 I think.
2168
2169 Arguments:
2170 ptr pointer to the initial [
2171 endptr where to return the end pointer
2172
2173 Returns: TRUE or FALSE
2174 */
2175
2176 static BOOL
2177 check_posix_syntax(const uschar *ptr, const uschar **endptr)
2178 {
2179 int terminator; /* Don't combine these lines; the Solaris cc */
2180 terminator = *(++ptr); /* compiler warns about "non-constant" initializer. */
2181 for (++ptr; *ptr != 0; ptr++)
2182 {
2183 if (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET) ptr++; else
2184 {
2185 if (*ptr == CHAR_RIGHT_SQUARE_BRACKET) return FALSE;
2186 if (*ptr == terminator && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET)
2187 {
2188 *endptr = ptr;
2189 return TRUE;
2190 }
2191 }
2192 }
2193 return FALSE;
2194 }
2195
2196
2197
2198
2199 /*************************************************
2200 * Check POSIX class name *
2201 *************************************************/
2202
2203 /* This function is called to check the name given in a POSIX-style class entry
2204 such as [:alnum:].
2205
2206 Arguments:
2207 ptr points to the first letter
2208 len the length of the name
2209
2210 Returns: a value representing the name, or -1 if unknown
2211 */
2212
2213 static int
2214 check_posix_name(const uschar *ptr, int len)
2215 {
2216 const char *pn = posix_names;
2217 register int yield = 0;
2218 while (posix_name_lengths[yield] != 0)
2219 {
2220 if (len == posix_name_lengths[yield] &&
2221 strncmp((const char *)ptr, pn, len) == 0) return yield;
2222 pn += posix_name_lengths[yield] + 1;
2223 yield++;
2224 }
2225 return -1;
2226 }
2227
2228
2229 /*************************************************
2230 * Adjust OP_RECURSE items in repeated group *
2231 *************************************************/
2232
2233 /* OP_RECURSE items contain an offset from the start of the regex to the group
2234 that is referenced. This means that groups can be replicated for fixed
2235 repetition simply by copying (because the recursion is allowed to refer to
2236 earlier groups that are outside the current group). However, when a group is
2237 optional (i.e. the minimum quantifier is zero), OP_BRAZERO or OP_SKIPZERO is
2238 inserted before it, after it has been compiled. This means that any OP_RECURSE
2239 items within it that refer to the group itself or any contained groups have to
2240 have their offsets adjusted. That one of the jobs of this function. Before it
2241 is called, the partially compiled regex must be temporarily terminated with
2242 OP_END.
2243
2244 This function has been extended with the possibility of forward references for
2245 recursions and subroutine calls. It must also check the list of such references
2246 for the group we are dealing with. If it finds that one of the recursions in
2247 the current group is on this list, it adjusts the offset in the list, not the
2248 value in the reference (which is a group number).
2249
2250 Arguments:
2251 group points to the start of the group
2252 adjust the amount by which the group is to be moved
2253 utf8 TRUE in UTF-8 mode
2254 cd contains pointers to tables etc.
2255 save_hwm the hwm forward reference pointer at the start of the group
2256
2257 Returns: nothing
2258 */
2259
2260 static void
2261 adjust_recurse(uschar *group, int adjust, BOOL utf8, compile_data *cd,
2262 uschar *save_hwm)
2263 {
2264 uschar *ptr = group;
2265
2266 while ((ptr = (uschar *)find_recurse(ptr, utf8)) != NULL)
2267 {
2268 int offset;
2269 uschar *hc;
2270
2271 /* See if this recursion is on the forward reference list. If so, adjust the
2272 reference. */
2273
2274 for (hc = save_hwm; hc < cd->hwm; hc += LINK_SIZE)
2275 {
2276 offset = GET(hc, 0);
2277 if (cd->start_code + offset == ptr + 1)
2278 {
2279 PUT(hc, 0, offset + adjust);
2280 break;
2281 }
2282 }
2283
2284 /* Otherwise, adjust the recursion offset if it's after the start of this
2285 group. */
2286
2287 if (hc >= cd->hwm)
2288 {
2289 offset = GET(ptr, 1);
2290 if (cd->start_code + offset >= group) PUT(ptr, 1, offset + adjust);
2291 }
2292
2293 ptr += 1 + LINK_SIZE;
2294 }
2295 }
2296
2297
2298
2299 /*************************************************
2300 * Insert an automatic callout point *
2301 *************************************************/
2302
2303 /* This function is called when the PCRE_AUTO_CALLOUT option is set, to insert
2304 callout points before each pattern item.
2305
2306 Arguments:
2307 code current code pointer
2308 ptr current pattern pointer
2309 cd pointers to tables etc
2310
2311 Returns: new code pointer
2312 */
2313
2314 static uschar *
2315 auto_callout(uschar *code, const uschar *ptr, compile_data *cd)
2316 {
2317 *code++ = OP_CALLOUT;
2318 *code++ = 255;
2319 PUT(code, 0, (int)(ptr - cd->start_pattern)); /* Pattern offset */
2320 PUT(code, LINK_SIZE, 0); /* Default length */
2321 return code + 2*LINK_SIZE;
2322 }
2323
2324
2325
2326 /*************************************************
2327 * Complete a callout item *
2328 *************************************************/
2329
2330 /* A callout item contains the length of the next item in the pattern, which
2331 we can't fill in till after we have reached the relevant point. This is used
2332 for both automatic and manual callouts.
2333
2334 Arguments:
2335 previous_callout points to previous callout item
2336 ptr current pattern pointer
2337 cd pointers to tables etc
2338
2339 Returns: nothing
2340 */
2341
2342 static void
2343 complete_callout(uschar *previous_callout, const uschar *ptr, compile_data *cd)
2344 {
2345 int length = (int)(ptr - cd->start_pattern - GET(previous_callout, 2));
2346 PUT(previous_callout, 2 + LINK_SIZE, length);
2347 }
2348
2349
2350
2351 #ifdef SUPPORT_UCP
2352 /*************************************************
2353 * Get othercase range *
2354 *************************************************/
2355
2356 /* This function is passed the start and end of a class range, in UTF-8 mode
2357 with UCP support. It searches up the characters, looking for internal ranges of
2358 characters in the "other" case. Each call returns the next one, updating the
2359 start address.
2360
2361 Arguments:
2362 cptr points to starting character value; updated
2363 d end value
2364 ocptr where to put start of othercase range
2365 odptr where to put end of othercase range
2366
2367 Yield: TRUE when range returned; FALSE when no more
2368 */
2369
2370 static BOOL
2371 get_othercase_range(unsigned int *cptr, unsigned int d, unsigned int *ocptr,
2372 unsigned int *odptr)
2373 {
2374 unsigned int c, othercase, next;
2375
2376 for (c = *cptr; c <= d; c++)
2377 { if ((othercase = UCD_OTHERCASE(c)) != c) break; }
2378
2379 if (c > d) return FALSE;
2380
2381 *ocptr = othercase;
2382 next = othercase + 1;
2383
2384 for (++c; c <= d; c++)
2385 {
2386 if (UCD_OTHERCASE(c) != next) break;
2387 next++;
2388 }
2389
2390 *odptr = next - 1;
2391 *cptr = c;
2392
2393 return TRUE;
2394 }
2395 #endif /* SUPPORT_UCP */
2396
2397
2398
2399 /*************************************************
2400 * Check if auto-possessifying is possible *
2401 *************************************************/
2402
2403 /* This function is called for unlimited repeats of certain items, to see
2404 whether the next thing could possibly match the repeated item. If not, it makes
2405 sense to automatically possessify the repeated item.
2406
2407 Arguments:
2408 op_code the repeated op code
2409 this data for this item, depends on the opcode
2410 utf8 TRUE in UTF-8 mode
2411 utf8_char used for utf8 character bytes, NULL if not relevant
2412 ptr next character in pattern
2413 options options bits
2414 cd contains pointers to tables etc.
2415
2416 Returns: TRUE if possessifying is wanted
2417 */
2418
2419 static BOOL
2420 check_auto_possessive(int op_code, int item, BOOL utf8, uschar *utf8_char,
2421 const uschar *ptr, int options, compile_data *cd)
2422 {
2423 int next;
2424
2425 /* Skip whitespace and comments in extended mode */
2426
2427 if ((options & PCRE_EXTENDED) != 0)
2428 {
2429 for (;;)
2430 {
2431 while ((cd->ctypes[*ptr] & ctype_space) != 0) ptr++;
2432 if (*ptr == CHAR_NUMBER_SIGN)
2433 {
2434 while (*(++ptr) != 0)
2435 if (IS_NEWLINE(ptr)) { ptr += cd->nllen; break; }
2436 }
2437 else break;
2438 }
2439 }
2440
2441 /* If the next item is one that we can handle, get its value. A non-negative
2442 value is a character, a negative value is an escape value. */
2443
2444 if (*ptr == CHAR_BACKSLASH)
2445 {
2446 int temperrorcode = 0;
2447 next = check_escape(&ptr, &temperrorcode, cd->bracount, options, FALSE);
2448 if (temperrorcode != 0) return FALSE;
2449 ptr++; /* Point after the escape sequence */
2450 }
2451
2452 else if ((cd->ctypes[*ptr] & ctype_meta) == 0)
2453 {
2454 #ifdef SUPPORT_UTF8
2455 if (utf8) { GETCHARINC(next, ptr); } else
2456 #endif
2457 next = *ptr++;
2458 }
2459
2460 else return FALSE;
2461
2462 /* Skip whitespace and comments in extended mode */
2463
2464 if ((options & PCRE_EXTENDED) != 0)
2465 {
2466 for (;;)
2467 {
2468 while ((cd->ctypes[*ptr] & ctype_space) != 0) ptr++;
2469 if (*ptr == CHAR_NUMBER_SIGN)
2470 {
2471 while (*(++ptr) != 0)
2472 if (IS_NEWLINE(ptr)) { ptr += cd->nllen; break; }
2473 }
2474 else break;
2475 }
2476 }
2477
2478 /* If the next thing is itself optional, we have to give up. */
2479
2480 if (*ptr == CHAR_ASTERISK || *ptr == CHAR_QUESTION_MARK ||
2481 strncmp((char *)ptr, STR_LEFT_CURLY_BRACKET STR_0 STR_COMMA, 3) == 0)
2482 return FALSE;
2483
2484 /* Now compare the next item with the previous opcode. If the previous is a
2485 positive single character match, "item" either contains the character or, if
2486 "item" is greater than 127 in utf8 mode, the character's bytes are in
2487 utf8_char. */
2488
2489
2490 /* Handle cases when the next item is a character. */
2491
2492 if (next >= 0) switch(op_code)
2493 {
2494 case OP_CHAR:
2495 #ifdef SUPPORT_UTF8
2496 if (utf8 && item > 127) { GETCHAR(item, utf8_char); }
2497 #else
2498 (void)(utf8_char); /* Keep compiler happy by referencing function argument */
2499 #endif
2500 return item != next;
2501
2502 /* For CHARNC (caseless character) we must check the other case. If we have
2503 Unicode property support, we can use it to test the other case of
2504 high-valued characters. */
2505
2506 case OP_CHARNC:
2507 #ifdef SUPPORT_UTF8
2508 if (utf8 && item > 127) { GETCHAR(item, utf8_char); }
2509 #endif
2510 if (item == next) return FALSE;
2511 #ifdef SUPPORT_UTF8
2512 if (utf8)
2513 {
2514 unsigned int othercase;
2515 if (next < 128) othercase = cd->fcc[next]; else
2516 #ifdef SUPPORT_UCP
2517 othercase = UCD_OTHERCASE((unsigned int)next);
2518 #else
2519 othercase = NOTACHAR;
2520 #endif
2521 return (unsigned int)item != othercase;
2522 }
2523 else
2524 #endif /* SUPPORT_UTF8 */
2525 return (item != cd->fcc[next]); /* Non-UTF-8 mode */
2526
2527 /* For OP_NOT, "item" must be a single-byte character. */
2528
2529 case OP_NOT:
2530 if (item == next) return TRUE;
2531 if ((options & PCRE_CASELESS) == 0) return FALSE;
2532 #ifdef SUPPORT_UTF8
2533 if (utf8)
2534 {
2535 unsigned int othercase;
2536 if (next < 128) othercase = cd->fcc[next]; else
2537 #ifdef SUPPORT_UCP
2538 othercase = UCD_OTHERCASE(next);
2539 #else
2540 othercase = NOTACHAR;
2541 #endif
2542 return (unsigned int)item == othercase;
2543 }
2544 else
2545 #endif /* SUPPORT_UTF8 */
2546 return (item == cd->fcc[next]); /* Non-UTF-8 mode */
2547
2548 /* Note that OP_DIGIT etc. are generated only when PCRE_UCP is *not* set.
2549 When it is set, \d etc. are converted into OP_(NOT_)PROP codes. */
2550
2551 case OP_DIGIT:
2552 return next > 127 || (cd->ctypes[next] & ctype_digit) == 0;
2553
2554 case OP_NOT_DIGIT:
2555 return next <= 127 && (cd->ctypes[next] & ctype_digit) != 0;
2556
2557 case OP_WHITESPACE:
2558 return next > 127 || (cd->ctypes[next] & ctype_space) == 0;
2559
2560 case OP_NOT_WHITESPACE:
2561 return next <= 127 && (cd->ctypes[next] & ctype_space) != 0;
2562
2563 case OP_WORDCHAR:
2564 return next > 127 || (cd->ctypes[next] & ctype_word) == 0;
2565
2566 case OP_NOT_WORDCHAR:
2567 return next <= 127 && (cd->ctypes[next] & ctype_word) != 0;
2568
2569 case OP_HSPACE:
2570 case OP_NOT_HSPACE:
2571 switch(next)
2572 {
2573 case 0x09:
2574 case 0x20:
2575 case 0xa0:
2576 case 0x1680:
2577 case 0x180e:
2578 case 0x2000:
2579 case 0x2001:
2580 case 0x2002:
2581 case 0x2003:
2582 case 0x2004:
2583 case 0x2005:
2584 case 0x2006:
2585 case 0x2007:
2586 case 0x2008:
2587 case 0x2009:
2588 case 0x200A:
2589 case 0x202f:
2590 case 0x205f:
2591 case 0x3000:
2592 return op_code == OP_NOT_HSPACE;
2593 default:
2594 return op_code != OP_NOT_HSPACE;
2595 }
2596
2597 case OP_ANYNL:
2598 case OP_VSPACE:
2599 case OP_NOT_VSPACE:
2600 switch(next)
2601 {
2602 case 0x0a:
2603 case 0x0b:
2604 case 0x0c:
2605 case 0x0d:
2606 case 0x85:
2607 case 0x2028:
2608 case 0x2029:
2609 return op_code == OP_NOT_VSPACE;
2610 default:
2611 return op_code != OP_NOT_VSPACE;
2612 }
2613
2614 default:
2615 return FALSE;
2616 }
2617
2618
2619 /* Handle the case when the next item is \d, \s, etc. Note that when PCRE_UCP
2620 is set, \d turns into ESC_du rather than ESC_d, etc., so ESC_d etc. are
2621 generated only when PCRE_UCP is *not* set, that is, when only ASCII
2622 characteristics are recognized. */
2623
2624 switch(op_code)
2625 {
2626 case OP_CHAR:
2627 case OP_CHARNC:
2628 #ifdef SUPPORT_UTF8
2629 if (utf8 && item > 127) { GETCHAR(item, utf8_char); }
2630 #endif
2631 switch(-next)
2632 {
2633 case ESC_d:
2634 return item > 127 || (cd->ctypes[item] & ctype_digit) == 0;
2635
2636 case ESC_D:
2637 return item <= 127 && (cd->ctypes[item] & ctype_digit) != 0;
2638
2639 case ESC_s:
2640 return item > 127 || (cd->ctypes[item] & ctype_space) == 0;
2641
2642 case ESC_S:
2643 return item <= 127 && (cd->ctypes[item] & ctype_space) != 0;
2644
2645 case ESC_w:
2646 return item > 127 || (cd->ctypes[item] & ctype_word) == 0;
2647
2648 case ESC_W:
2649 return item <= 127 && (cd->ctypes[item] & ctype_word) != 0;
2650
2651 case ESC_h:
2652 case ESC_H:
2653 switch(item)
2654 {
2655 case 0x09:
2656 case 0x20:
2657 case 0xa0:
2658 case 0x1680:
2659 case 0x180e:
2660 case 0x2000:
2661 case 0x2001:
2662 case 0x2002:
2663 case 0x2003:
2664 case 0x2004:
2665 case 0x2005:
2666 case 0x2006:
2667 case 0x2007:
2668 case 0x2008:
2669 case 0x2009:
2670 case 0x200A:
2671 case 0x202f:
2672 case 0x205f:
2673 case 0x3000:
2674 return -next != ESC_h;
2675 default:
2676 return -next == ESC_h;
2677 }
2678
2679 case ESC_v:
2680 case ESC_V:
2681 switch(item)
2682 {
2683 case 0x0a:
2684 case 0x0b:
2685 case 0x0c:
2686 case 0x0d:
2687 case 0x85:
2688 case 0x2028:
2689 case 0x2029:
2690 return -next != ESC_v;
2691 default:
2692 return -next == ESC_v;
2693 }
2694
2695 default:
2696 return FALSE;
2697 }
2698
2699 case OP_DIGIT:
2700 return next == -ESC_D || next == -ESC_s || next == -ESC_W ||
2701 next == -ESC_h || next == -ESC_v || next == -ESC_R;
2702
2703 case OP_NOT_DIGIT:
2704 return next == -ESC_d;
2705
2706 case OP_WHITESPACE:
2707 return next == -ESC_S || next == -ESC_d || next == -ESC_w || next == -ESC_R;
2708
2709 case OP_NOT_WHITESPACE:
2710 return next == -ESC_s || next == -ESC_h || next == -ESC_v;
2711
2712 case OP_HSPACE:
2713 return next == -ESC_S || next == -ESC_H || next == -ESC_d ||
2714 next == -ESC_w || next == -ESC_v || next == -ESC_R;
2715
2716 case OP_NOT_HSPACE:
2717 return next == -ESC_h;
2718
2719 /* Can't have \S in here because VT matches \S (Perl anomaly) */
2720 case OP_ANYNL:
2721 case OP_VSPACE:
2722 return next == -ESC_V || next == -ESC_d || next == -ESC_w;
2723
2724 case OP_NOT_VSPACE:
2725 return next == -ESC_v || next == -ESC_R;
2726
2727 case OP_WORDCHAR:
2728 return next == -ESC_W || next == -ESC_s || next == -ESC_h ||
2729 next == -ESC_v || next == -ESC_R;
2730
2731 case OP_NOT_WORDCHAR:
2732 return next == -ESC_w || next == -ESC_d;
2733
2734 default:
2735 return FALSE;
2736 }
2737
2738 /* Control does not reach here */
2739 }
2740
2741
2742
2743 /*************************************************
2744 * Compile one branch *
2745 *************************************************/
2746
2747 /* Scan the pattern, compiling it into the a vector. If the options are
2748 changed during the branch, the pointer is used to change the external options
2749 bits. This function is used during the pre-compile phase when we are trying
2750 to find out the amount of memory needed, as well as during the real compile
2751 phase. The value of lengthptr distinguishes the two phases.
2752
2753 Arguments:
2754 optionsptr pointer to the option bits
2755 codeptr points to the pointer to the current code point
2756 ptrptr points to the current pattern pointer
2757 errorcodeptr points to error code variable
2758 firstbyteptr set to initial literal character, or < 0 (REQ_UNSET, REQ_NONE)
2759 reqbyteptr set to the last literal character required, else < 0
2760 bcptr points to current branch chain
2761 cd contains pointers to tables etc.
2762 lengthptr NULL during the real compile phase
2763 points to length accumulator during pre-compile phase
2764
2765 Returns: TRUE on success
2766 FALSE, with *errorcodeptr set non-zero on error
2767 */
2768
2769 static BOOL
2770 compile_branch(int *optionsptr, uschar **codeptr, const uschar **ptrptr,
2771 int *errorcodeptr, int *firstbyteptr, int *reqbyteptr, branch_chain *bcptr,
2772 compile_data *cd, int *lengthptr)
2773 {
2774 int repeat_type, op_type;
2775 int repeat_min = 0, repeat_max = 0; /* To please picky compilers */
2776 int bravalue = 0;
2777 int greedy_default, greedy_non_default;
2778 int firstbyte, reqbyte;
2779 int zeroreqbyte, zerofirstbyte;
2780 int req_caseopt, reqvary, tempreqvary;
2781 int options = *optionsptr;
2782 int after_manual_callout = 0;
2783 int length_prevgroup = 0;
2784 register int c;
2785 register uschar *code = *codeptr;
2786 uschar *last_code = code;
2787 uschar *orig_code = code;
2788 uschar *tempcode;
2789 BOOL inescq = FALSE;
2790 BOOL groupsetfirstbyte = FALSE;
2791 const uschar *ptr = *ptrptr;
2792 const uschar *tempptr;
2793 const uschar *nestptr = NULL;
2794 uschar *previous = NULL;
2795 uschar *previous_callout = NULL;
2796 uschar *save_hwm = NULL;
2797 uschar classbits[32];
2798
2799 #ifdef SUPPORT_UTF8
2800 BOOL class_utf8;
2801 BOOL utf8 = (options & PCRE_UTF8) != 0;
2802 uschar *class_utf8data;
2803 uschar *class_utf8data_base;
2804 uschar utf8_char[6];
2805 #else
2806 BOOL utf8 = FALSE;
2807 uschar *utf8_char = NULL;
2808 #endif
2809
2810 #ifdef PCRE_DEBUG
2811 if (lengthptr != NULL) DPRINTF((">> start branch\n"));
2812 #endif
2813
2814 /* Set up the default and non-default settings for greediness */
2815
2816 greedy_default = ((options & PCRE_UNGREEDY) != 0);
2817 greedy_non_default = greedy_default ^ 1;
2818
2819 /* Initialize no first byte, no required byte. REQ_UNSET means "no char
2820 matching encountered yet". It gets changed to REQ_NONE if we hit something that
2821 matches a non-fixed char first char; reqbyte just remains unset if we never
2822 find one.
2823
2824 When we hit a repeat whose minimum is zero, we may have to adjust these values
2825 to take the zero repeat into account. This is implemented by setting them to
2826 zerofirstbyte and zeroreqbyte when such a repeat is encountered. The individual
2827 item types that can be repeated set these backoff variables appropriately. */
2828
2829 firstbyte = reqbyte = zerofirstbyte = zeroreqbyte = REQ_UNSET;
2830
2831 /* The variable req_caseopt contains either the REQ_CASELESS value or zero,
2832 according to the current setting of the caseless flag. REQ_CASELESS is a bit
2833 value > 255. It is added into the firstbyte or reqbyte variables to record the
2834 case status of the value. This is used only for ASCII characters. */
2835
2836 req_caseopt = ((options & PCRE_CASELESS) != 0)? REQ_CASELESS : 0;
2837
2838 /* Switch on next character until the end of the branch */
2839
2840 for (;; ptr++)
2841 {
2842 BOOL negate_class;
2843 BOOL should_flip_negation;
2844 BOOL possessive_quantifier;
2845 BOOL is_quantifier;
2846 BOOL is_recurse;
2847 BOOL reset_bracount;
2848 int class_charcount;
2849 int class_lastchar;
2850 int newoptions;
2851 int recno;
2852 int refsign;
2853 int skipbytes;
2854 int subreqbyte;
2855 int subfirstbyte;
2856 int terminator;
2857 int mclength;
2858 uschar mcbuffer[8];
2859
2860 /* Get next byte in the pattern */
2861
2862 c = *ptr;
2863
2864 /* If we are at the end of a nested substitution, revert to the outer level
2865 string. Nesting only happens one level deep. */
2866
2867 if (c == 0 && nestptr != NULL)
2868 {
2869 ptr = nestptr;
2870 nestptr = NULL;
2871 c = *ptr;
2872 }
2873
2874 /* If we are in the pre-compile phase, accumulate the length used for the
2875 previous cycle of this loop. */
2876
2877 if (lengthptr != NULL)
2878 {
2879 #ifdef PCRE_DEBUG
2880 if (code > cd->hwm) cd->hwm = code; /* High water info */
2881 #endif
2882 if (code > cd->start_workspace + WORK_SIZE_CHECK) /* Check for overrun */
2883 {
2884 *errorcodeptr = ERR52;
2885 goto FAILED;
2886 }
2887
2888 /* There is at least one situation where code goes backwards: this is the
2889 case of a zero quantifier after a class (e.g. [ab]{0}). At compile time,
2890 the class is simply eliminated. However, it is created first, so we have to
2891 allow memory for it. Therefore, don't ever reduce the length at this point.
2892 */
2893
2894 if (code < last_code) code = last_code;
2895
2896 /* Paranoid check for integer overflow */
2897
2898 if (OFLOW_MAX - *lengthptr < code - last_code)
2899 {
2900 *errorcodeptr = ERR20;
2901 goto FAILED;
2902 }
2903
2904 *lengthptr += (int)(code - last_code);
2905 DPRINTF(("length=%d added %d c=%c\n", *lengthptr, code - last_code, c));
2906
2907 /* If "previous" is set and it is not at the start of the work space, move
2908 it back to there, in order to avoid filling up the work space. Otherwise,
2909 if "previous" is NULL, reset the current code pointer to the start. */
2910
2911 if (previous != NULL)
2912 {
2913 if (previous > orig_code)
2914 {
2915 memmove(orig_code, previous, code - previous);
2916 code -= previous - orig_code;
2917 previous = orig_code;
2918 }
2919 }
2920 else code = orig_code;
2921
2922 /* Remember where this code item starts so we can pick up the length
2923 next time round. */
2924
2925 last_code = code;
2926 }
2927
2928 /* In the real compile phase, just check the workspace used by the forward
2929 reference list. */
2930
2931 else if (cd->hwm > cd->start_workspace + WORK_SIZE_CHECK)
2932 {
2933 *errorcodeptr = ERR52;
2934 goto FAILED;
2935 }
2936
2937 /* If in \Q...\E, check for the end; if not, we have a literal */
2938
2939 if (inescq && c != 0)
2940 {
2941 if (c == CHAR_BACKSLASH && ptr[1] == CHAR_E)
2942 {
2943 inescq = FALSE;
2944 ptr++;
2945 continue;
2946 }
2947 else
2948 {
2949 if (previous_callout != NULL)
2950 {
2951 if (lengthptr == NULL) /* Don't attempt in pre-compile phase */
2952 complete_callout(previous_callout, ptr, cd);
2953 previous_callout = NULL;
2954 }
2955 if ((options & PCRE_AUTO_CALLOUT) != 0)
2956 {
2957 previous_callout = code;
2958 code = auto_callout(code, ptr, cd);
2959 }
2960 goto NORMAL_CHAR;
2961 }
2962 }
2963
2964 /* Fill in length of a previous callout, except when the next thing is
2965 a quantifier. */
2966
2967 is_quantifier =
2968 c == CHAR_ASTERISK || c == CHAR_PLUS || c == CHAR_QUESTION_MARK ||
2969 (c == CHAR_LEFT_CURLY_BRACKET && is_counted_repeat(ptr+1));
2970
2971 if (!is_quantifier && previous_callout != NULL &&
2972 after_manual_callout-- <= 0)
2973 {
2974 if (lengthptr == NULL) /* Don't attempt in pre-compile phase */
2975 complete_callout(previous_callout, ptr, cd);
2976 previous_callout = NULL;
2977 }
2978
2979 /* In extended mode, skip white space and comments */
2980
2981 if ((options & PCRE_EXTENDED) != 0)
2982 {
2983 if ((cd->ctypes[c] & ctype_space) != 0) continue;
2984 if (c == CHAR_NUMBER_SIGN)
2985 {
2986 while (*(++ptr) != 0)
2987 {
2988 if (IS_NEWLINE(ptr)) { ptr += cd->nllen - 1; break; }
2989 }
2990 if (*ptr != 0) continue;
2991
2992 /* Else fall through to handle end of string */
2993 c = 0;
2994 }
2995 }
2996
2997 /* No auto callout for quantifiers. */
2998
2999 if ((options & PCRE_AUTO_CALLOUT) != 0 && !is_quantifier)
3000 {
3001 previous_callout = code;
3002 code = auto_callout(code, ptr, cd);
3003 }
3004
3005 switch(c)
3006 {
3007 /* ===================================================================*/
3008 case 0: /* The branch terminates at string end */
3009 case CHAR_VERTICAL_LINE: /* or | or ) */
3010 case CHAR_RIGHT_PARENTHESIS:
3011 *firstbyteptr = firstbyte;
3012 *reqbyteptr = reqbyte;
3013 *codeptr = code;
3014 *ptrptr = ptr;
3015 if (lengthptr != NULL)
3016 {
3017 if (OFLOW_MAX - *lengthptr < code - last_code)
3018 {
3019 *errorcodeptr = ERR20;
3020 goto FAILED;
3021 }
3022 *lengthptr += (int)(code - last_code); /* To include callout length */
3023 DPRINTF((">> end branch\n"));
3024 }
3025 return TRUE;
3026
3027
3028 /* ===================================================================*/
3029 /* Handle single-character metacharacters. In multiline mode, ^ disables
3030 the setting of any following char as a first character. */
3031
3032 case CHAR_CIRCUMFLEX_ACCENT:
3033 if ((options & PCRE_MULTILINE) != 0)
3034 {
3035 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
3036 }
3037 previous = NULL;
3038 *code++ = OP_CIRC;
3039 break;
3040
3041 case CHAR_DOLLAR_SIGN:
3042 previous = NULL;
3043 *code++ = OP_DOLL;
3044 break;
3045
3046 /* There can never be a first char if '.' is first, whatever happens about
3047 repeats. The value of reqbyte doesn't change either. */
3048
3049 case CHAR_DOT:
3050 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
3051 zerofirstbyte = firstbyte;
3052 zeroreqbyte = reqbyte;
3053 previous = code;
3054 *code++ = ((options & PCRE_DOTALL) != 0)? OP_ALLANY: OP_ANY;
3055 break;
3056
3057
3058 /* ===================================================================*/
3059 /* Character classes. If the included characters are all < 256, we build a
3060 32-byte bitmap of the permitted characters, except in the special case
3061 where there is only one such character. For negated classes, we build the
3062 map as usual, then invert it at the end. However, we use a different opcode
3063 so that data characters > 255 can be handled correctly.
3064
3065 If the class contains characters outside the 0-255 range, a different
3066 opcode is compiled. It may optionally have a bit map for characters < 256,
3067 but those above are are explicitly listed afterwards. A flag byte tells
3068 whether the bitmap is present, and whether this is a negated class or not.
3069
3070 In JavaScript compatibility mode, an isolated ']' causes an error. In
3071 default (Perl) mode, it is treated as a data character. */
3072
3073 case CHAR_RIGHT_SQUARE_BRACKET:
3074 if ((cd->external_options & PCRE_JAVASCRIPT_COMPAT) != 0)
3075 {
3076 *errorcodeptr = ERR64;
3077 goto FAILED;
3078 }
3079 goto NORMAL_CHAR;
3080
3081 case CHAR_LEFT_SQUARE_BRACKET:
3082 previous = code;
3083
3084 /* PCRE supports POSIX class stuff inside a class. Perl gives an error if
3085 they are encountered at the top level, so we'll do that too. */
3086
3087 if ((ptr[1] == CHAR_COLON || ptr[1] == CHAR_DOT ||
3088 ptr[1] == CHAR_EQUALS_SIGN) &&
3089 check_posix_syntax(ptr, &tempptr))
3090 {
3091 *errorcodeptr = (ptr[1] == CHAR_COLON)? ERR13 : ERR31;
3092 goto FAILED;
3093 }
3094
3095 /* If the first character is '^', set the negation flag and skip it. Also,
3096 if the first few characters (either before or after ^) are \Q\E or \E we
3097 skip them too. This makes for compatibility with Perl. */
3098
3099 negate_class = FALSE;
3100 for (;;)
3101 {
3102 c = *(++ptr);
3103 if (c == CHAR_BACKSLASH)
3104 {
3105 if (ptr[1] == CHAR_E)
3106 ptr++;
3107 else if (strncmp((const char *)ptr+1,
3108 STR_Q STR_BACKSLASH STR_E, 3) == 0)
3109 ptr += 3;
3110 else
3111 break;
3112 }
3113 else if (!negate_class && c == CHAR_CIRCUMFLEX_ACCENT)
3114 negate_class = TRUE;
3115 else break;
3116 }
3117
3118 /* Empty classes are allowed in JavaScript compatibility mode. Otherwise,
3119 an initial ']' is taken as a data character -- the code below handles
3120 that. In JS mode, [] must always fail, so generate OP_FAIL, whereas
3121 [^] must match any character, so generate OP_ALLANY. */
3122
3123 if (c == CHAR_RIGHT_SQUARE_BRACKET &&
3124 (cd->external_options & PCRE_JAVASCRIPT_COMPAT) != 0)
3125 {
3126 *code++ = negate_class? OP_ALLANY : OP_FAIL;
3127 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
3128 zerofirstbyte = firstbyte;
3129 break;
3130 }
3131
3132 /* If a class contains a negative special such as \S, we need to flip the
3133 negation flag at the end, so that support for characters > 255 works
3134 correctly (they are all included in the class). */
3135
3136 should_flip_negation = FALSE;
3137
3138 /* Keep a count of chars with values < 256 so that we can optimize the case
3139 of just a single character (as long as it's < 256). However, For higher
3140 valued UTF-8 characters, we don't yet do any optimization. */
3141
3142 class_charcount = 0;
3143 class_lastchar = -1;
3144
3145 /* Initialize the 32-char bit map to all zeros. We build the map in a
3146 temporary bit of memory, in case the class contains only 1 character (less
3147 than 256), because in that case the compiled code doesn't use the bit map.
3148 */
3149
3150 memset(classbits, 0, 32 * sizeof(uschar));
3151
3152 #ifdef SUPPORT_UTF8
3153 class_utf8 = FALSE; /* No chars >= 256 */
3154 class_utf8data = code + LINK_SIZE + 2; /* For UTF-8 items */
3155 class_utf8data_base = class_utf8data; /* For resetting in pass 1 */
3156 #endif
3157
3158 /* Process characters until ] is reached. By writing this as a "do" it
3159 means that an initial ] is taken as a data character. At the start of the
3160 loop, c contains the first byte of the character. */
3161
3162 if (c != 0) do
3163 {
3164 const uschar *oldptr;
3165
3166 #ifdef SUPPORT_UTF8
3167 if (utf8 && c > 127)
3168 { /* Braces are required because the */
3169 GETCHARLEN(c, ptr, ptr); /* macro generates multiple statements */
3170 }
3171
3172 /* In the pre-compile phase, accumulate the length of any UTF-8 extra
3173 data and reset the pointer. This is so that very large classes that
3174 contain a zillion UTF-8 characters no longer overwrite the work space
3175 (which is on the stack). */
3176
3177 if (lengthptr != NULL)
3178 {
3179 *lengthptr += class_utf8data - class_utf8data_base;
3180 class_utf8data = class_utf8data_base;
3181 }
3182
3183 #endif
3184
3185 /* Inside \Q...\E everything is literal except \E */
3186
3187 if (inescq)
3188 {
3189 if (c == CHAR_BACKSLASH && ptr[1] == CHAR_E) /* If we are at \E */
3190 {
3191 inescq = FALSE; /* Reset literal state */
3192 ptr++; /* Skip the 'E' */
3193 continue; /* Carry on with next */
3194 }
3195 goto CHECK_RANGE; /* Could be range if \E follows */
3196 }
3197
3198 /* Handle POSIX class names. Perl allows a negation extension of the
3199 form [:^name:]. A square bracket that doesn't match the syntax is
3200 treated as a literal. We also recognize the POSIX constructions
3201 [.ch.] and [=ch=] ("collating elements") and fault them, as Perl
3202 5.6 and 5.8 do. */
3203
3204 if (c == CHAR_LEFT_SQUARE_BRACKET &&
3205 (ptr[1] == CHAR_COLON || ptr[1] == CHAR_DOT ||
3206 ptr[1] == CHAR_EQUALS_SIGN) && check_posix_syntax(ptr, &tempptr))
3207 {
3208 BOOL local_negate = FALSE;
3209 int posix_class, taboffset, tabopt;
3210 register const uschar *cbits = cd->cbits;
3211 uschar pbits[32];
3212
3213 if (ptr[1] != CHAR_COLON)
3214 {
3215 *errorcodeptr = ERR31;
3216 goto FAILED;
3217 }
3218
3219 ptr += 2;
3220 if (*ptr == CHAR_CIRCUMFLEX_ACCENT)
3221 {
3222 local_negate = TRUE;
3223 should_flip_negation = TRUE; /* Note negative special */
3224 ptr++;
3225 }
3226
3227 posix_class = check_posix_name(ptr, (int)(tempptr - ptr));
3228 if (posix_class < 0)
3229 {
3230 *errorcodeptr = ERR30;
3231 goto FAILED;
3232 }
3233
3234 /* If matching is caseless, upper and lower are converted to
3235 alpha. This relies on the fact that the class table starts with
3236 alpha, lower, upper as the first 3 entries. */
3237
3238 if ((options & PCRE_CASELESS) != 0 && posix_class <= 2)
3239 posix_class = 0;
3240
3241 /* When PCRE_UCP is set, some of the POSIX classes are converted to
3242 different escape sequences that use Unicode properties. */
3243
3244 #ifdef SUPPORT_UCP
3245 if ((options & PCRE_UCP) != 0)
3246 {
3247 int pc = posix_class + ((local_negate)? POSIX_SUBSIZE/2 : 0);
3248 if (posix_substitutes[pc] != NULL)
3249 {
3250 nestptr = tempptr + 1;
3251 ptr = posix_substitutes[pc] - 1;
3252 continue;
3253 }
3254 }
3255 #endif
3256 /* In the non-UCP case, we build the bit map for the POSIX class in a
3257 chunk of local store because we may be adding and subtracting from it,
3258 and we don't want to subtract bits that may be in the main map already.
3259 At the end we or the result into the bit map that is being built. */
3260
3261 posix_class *= 3;
3262
3263 /* Copy in the first table (always present) */
3264
3265 memcpy(pbits, cbits + posix_class_maps[posix_class],
3266 32 * sizeof(uschar));
3267
3268 /* If there is a second table, add or remove it as required. */
3269
3270 taboffset = posix_class_maps[posix_class + 1];
3271 tabopt = posix_class_maps[posix_class + 2];
3272
3273 if (taboffset >= 0)
3274 {
3275 if (tabopt >= 0)
3276 for (c = 0; c < 32; c++) pbits[c] |= cbits[c + taboffset];
3277 else
3278 for (c = 0; c < 32; c++) pbits[c] &= ~cbits[c + taboffset];
3279 }
3280
3281 /* Not see if we need to remove any special characters. An option
3282 value of 1 removes vertical space and 2 removes underscore. */
3283
3284 if (tabopt < 0) tabopt = -tabopt;
3285 if (tabopt == 1) pbits[1] &= ~0x3c;
3286 else if (tabopt == 2) pbits[11] &= 0x7f;
3287
3288 /* Add the POSIX table or its complement into the main table that is
3289 being built and we are done. */
3290
3291 if (local_negate)
3292 for (c = 0; c < 32; c++) classbits[c] |= ~pbits[c];
3293 else
3294 for (c = 0; c < 32; c++) classbits[c] |= pbits[c];
3295
3296 ptr = tempptr + 1;
3297 class_charcount = 10; /* Set > 1; assumes more than 1 per class */
3298 continue; /* End of POSIX syntax handling */
3299 }
3300
3301 /* Backslash may introduce a single character, or it may introduce one
3302 of the specials, which just set a flag. The sequence \b is a special
3303 case. Inside a class (and only there) it is treated as backspace. We
3304 assume that other escapes have more than one character in them, so set
3305 class_charcount bigger than one. Unrecognized escapes fall through and
3306 are either treated as literal characters (by default), or are faulted if
3307 PCRE_EXTRA is set. */
3308
3309 if (c == CHAR_BACKSLASH)
3310 {
3311 c = check_escape(&ptr, errorcodeptr, cd->bracount, options, TRUE);
3312 if (*errorcodeptr != 0) goto FAILED;
3313
3314 if (-c == ESC_b) c = CHAR_BS; /* \b is backspace in a class */
3315 else if (-c == ESC_Q) /* Handle start of quoted string */
3316 {
3317 if (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
3318 {
3319 ptr += 2; /* avoid empty string */
3320 }
3321 else inescq = TRUE;
3322 continue;
3323 }
3324 else if (-c == ESC_E) continue; /* Ignore orphan \E */
3325
3326 if (c < 0)
3327 {
3328 register const uschar *cbits = cd->cbits;
3329 class_charcount += 2; /* Greater than 1 is what matters */
3330
3331 switch (-c)
3332 {
3333 #ifdef SUPPORT_UCP
3334 case ESC_du: /* These are the values given for \d etc */
3335 case ESC_DU: /* when PCRE_UCP is set. We replace the */
3336 case ESC_wu: /* escape sequence with an appropriate \p */
3337 case ESC_WU: /* or \P to test Unicode properties instead */
3338 case ESC_su: /* of the default ASCII testing. */
3339 case ESC_SU:
3340 nestptr = ptr;
3341 ptr = substitutes[-c - ESC_DU] - 1; /* Just before substitute */
3342 class_charcount -= 2; /* Undo! */
3343 continue;
3344 #endif
3345 case ESC_d:
3346 for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_digit];
3347 continue;
3348
3349 case ESC_D:
3350 should_flip_negation = TRUE;
3351 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_digit];
3352 continue;
3353
3354 case ESC_w:
3355 for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_word];
3356 continue;
3357
3358 case ESC_W:
3359 should_flip_negation = TRUE;
3360 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_word];
3361 continue;
3362
3363 case ESC_s:
3364 for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_space];
3365 classbits[1] &= ~0x08; /* Perl 5.004 onwards omits VT from \s */
3366 continue;
3367
3368 case ESC_S:
3369 should_flip_negation = TRUE;
3370 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_space];
3371 classbits[1] |= 0x08; /* Perl 5.004 onwards omits VT from \s */
3372 continue;
3373
3374 case ESC_h:
3375 SETBIT(classbits, 0x09); /* VT */
3376 SETBIT(classbits, 0x20); /* SPACE */
3377 SETBIT(classbits, 0xa0); /* NSBP */
3378 #ifdef SUPPORT_UTF8
3379 if (utf8)
3380 {
3381 class_utf8 = TRUE;
3382 *class_utf8data++ = XCL_SINGLE;
3383 class_utf8data += _pcre_ord2utf8(0x1680, class_utf8data);
3384 *class_utf8data++ = XCL_SINGLE;
3385 class_utf8data += _pcre_ord2utf8(0x180e, class_utf8data);
3386 *class_utf8data++ = XCL_RANGE;
3387 class_utf8data += _pcre_ord2utf8(0x2000, class_utf8data);
3388 class_utf8data += _pcre_ord2utf8(0x200A, class_utf8data);
3389 *class_utf8data++ = XCL_SINGLE;
3390 class_utf8data += _pcre_ord2utf8(0x202f, class_utf8data);
3391 *class_utf8data++ = XCL_SINGLE;
3392 class_utf8data += _pcre_ord2utf8(0x205f, class_utf8data);
3393 *class_utf8data++ = XCL_SINGLE;
3394 class_utf8data += _pcre_ord2utf8(0x3000, class_utf8data);
3395 }
3396 #endif
3397 continue;
3398
3399 case ESC_H:
3400 for (c = 0; c < 32; c++)
3401 {
3402 int x = 0xff;
3403 switch (c)
3404 {
3405 case 0x09/8: x ^= 1 << (0x09%8); break;
3406 case 0x20/8: x ^= 1 << (0x20%8); break;
3407 case 0xa0/8: x ^= 1 << (0xa0%8); break;
3408 default: break;
3409 }
3410 classbits[c] |= x;
3411 }
3412
3413 #ifdef SUPPORT_UTF8
3414 if (utf8)
3415 {
3416 class_utf8 = TRUE;
3417 *class_utf8data++ = XCL_RANGE;
3418 class_utf8data += _pcre_ord2utf8(0x0100, class_utf8data);
3419 class_utf8data += _pcre_ord2utf8(0x167f, class_utf8data);
3420 *class_utf8data++ = XCL_RANGE;
3421 class_utf8data += _pcre_ord2utf8(0x1681, class_utf8data);
3422 class_utf8data += _pcre_ord2utf8(0x180d, class_utf8data);
3423 *class_utf8data++ = XCL_RANGE;
3424 class_utf8data += _pcre_ord2utf8(0x180f, class_utf8data);
3425 class_utf8data += _pcre_ord2utf8(0x1fff, class_utf8data);
3426 *class_utf8data++ = XCL_RANGE;
3427 class_utf8data += _pcre_ord2utf8(0x200B, class_utf8data);
3428 class_utf8data += _pcre_ord2utf8(0x202e, class_utf8data);
3429 *class_utf8data++ = XCL_RANGE;
3430 class_utf8data += _pcre_ord2utf8(0x2030, class_utf8data);
3431 class_utf8data += _pcre_ord2utf8(0x205e, class_utf8data);
3432 *class_utf8data++ = XCL_RANGE;
3433 class_utf8data += _pcre_ord2utf8(0x2060, class_utf8data);
3434 class_utf8data += _pcre_ord2utf8(0x2fff, class_utf8data);
3435 *class_utf8data++ = XCL_RANGE;
3436 class_utf8data += _pcre_ord2utf8(0x3001, class_utf8data);
3437 class_utf8data += _pcre_ord2utf8(0x7fffffff, class_utf8data);
3438 }
3439 #endif
3440 continue;
3441
3442 case ESC_v:
3443 SETBIT(classbits, 0x0a); /* LF */
3444 SETBIT(classbits, 0x0b); /* VT */
3445 SETBIT(classbits, 0x0c); /* FF */
3446 SETBIT(classbits, 0x0d); /* CR */
3447 SETBIT(classbits, 0x85); /* NEL */
3448 #ifdef SUPPORT_UTF8
3449 if (utf8)
3450 {
3451 class_utf8 = TRUE;
3452 *class_utf8data++ = XCL_RANGE;
3453 class_utf8data += _pcre_ord2utf8(0x2028, class_utf8data);
3454 class_utf8data += _pcre_ord2utf8(0x2029, class_utf8data);
3455 }
3456 #endif
3457 continue;
3458
3459 case ESC_V:
3460 for (c = 0; c < 32; c++)
3461 {
3462 int x = 0xff;
3463 switch (c)
3464 {
3465 case 0x0a/8: x ^= 1 << (0x0a%8);
3466 x ^= 1 << (0x0b%8);
3467 x ^= 1 << (0x0c%8);
3468 x ^= 1 << (0x0d%8);
3469 break;
3470 case 0x85/8: x ^= 1 << (0x85%8); break;
3471 default: break;
3472 }
3473 classbits[c] |= x;
3474 }
3475
3476 #ifdef SUPPORT_UTF8
3477 if (utf8)
3478 {
3479 class_utf8 = TRUE;
3480 *class_utf8data++ = XCL_RANGE;
3481 class_utf8data += _pcre_ord2utf8(0x0100, class_utf8data);
3482 class_utf8data += _pcre_ord2utf8(0x2027, class_utf8data);
3483 *class_utf8data++ = XCL_RANGE;
3484 class_utf8data += _pcre_ord2utf8(0x2029, class_utf8data);
3485 class_utf8data += _pcre_ord2utf8(0x7fffffff, class_utf8data);
3486 }
3487 #endif
3488 continue;
3489
3490 #ifdef SUPPORT_UCP
3491 case ESC_p:
3492 case ESC_P:
3493 {
3494 BOOL negated;
3495 int pdata;
3496 int ptype = get_ucp(&ptr, &negated, &pdata, errorcodeptr);
3497 if (ptype < 0) goto FAILED;
3498 class_utf8 = TRUE;
3499 *class_utf8data++ = ((-c == ESC_p) != negated)?
3500 XCL_PROP : XCL_NOTPROP;
3501 *class_utf8data++ = ptype;
3502 *class_utf8data++ = pdata;
3503 class_charcount -= 2; /* Not a < 256 character */
3504 continue;
3505 }
3506 #endif
3507 /* Unrecognized escapes are faulted if PCRE is running in its
3508 strict mode. By default, for compatibility with Perl, they are
3509 treated as literals. */
3510
3511 default:
3512 if ((options & PCRE_EXTRA) != 0)
3513 {
3514 *errorcodeptr = ERR7;
3515 goto FAILED;
3516 }
3517 class_charcount -= 2; /* Undo the default count from above */
3518 c = *ptr; /* Get the final character and fall through */
3519 break;
3520 }
3521 }
3522
3523 /* Fall through if we have a single character (c >= 0). This may be
3524 greater than 256 in UTF-8 mode. */
3525
3526 } /* End of backslash handling */
3527
3528 /* A single character may be followed by '-' to form a range. However,
3529 Perl does not permit ']' to be the end of the range. A '-' character
3530 at the end is treated as a literal. Perl ignores orphaned \E sequences
3531 entirely. The code for handling \Q and \E is messy. */
3532
3533 CHECK_RANGE:
3534 while (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
3535 {
3536 inescq = FALSE;
3537 ptr += 2;
3538 }
3539
3540 oldptr = ptr;
3541
3542 /* Remember \r or \n */
3543
3544 if (c == CHAR_CR || c == CHAR_NL) cd->external_flags |= PCRE_HASCRORLF;
3545
3546 /* Check for range */
3547
3548 if (!inescq && ptr[1] == CHAR_MINUS)
3549 {
3550 int d;
3551 ptr += 2;
3552 while (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_E) ptr += 2;
3553
3554 /* If we hit \Q (not followed by \E) at this point, go into escaped
3555 mode. */
3556
3557 while (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_Q)
3558 {
3559 ptr += 2;
3560 if (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_E)
3561 { ptr += 2; continue; }
3562 inescq = TRUE;
3563 break;
3564 }
3565
3566 if (*ptr == 0 || (!inescq && *ptr == CHAR_RIGHT_SQUARE_BRACKET))
3567 {
3568 ptr = oldptr;
3569 goto LONE_SINGLE_CHARACTER;
3570 }
3571
3572 #ifdef SUPPORT_UTF8
3573 if (utf8)
3574 { /* Braces are required because the */
3575 GETCHARLEN(d, ptr, ptr); /* macro generates multiple statements */
3576 }
3577 else
3578 #endif
3579 d = *ptr; /* Not UTF-8 mode */
3580
3581 /* The second part of a range can be a single-character escape, but
3582 not any of the other escapes. Perl 5.6 treats a hyphen as a literal
3583 in such circumstances. */
3584
3585 if (!inescq && d == CHAR_BACKSLASH)
3586 {
3587 d = check_escape(&ptr, errorcodeptr, cd->bracount, options, TRUE);
3588 if (*errorcodeptr != 0) goto FAILED;
3589
3590 /* \b is backspace; any other special means the '-' was literal */
3591
3592 if (d < 0)
3593 {
3594 if (d == -ESC_b) d = CHAR_BS; else
3595 {
3596 ptr = oldptr;
3597 goto LONE_SINGLE_CHARACTER; /* A few lines below */
3598 }
3599 }
3600 }
3601
3602 /* Check that the two values are in the correct order. Optimize
3603 one-character ranges */
3604
3605 if (d < c)
3606 {
3607 *errorcodeptr = ERR8;
3608 goto FAILED;
3609 }
3610
3611 if (d == c) goto LONE_SINGLE_CHARACTER; /* A few lines below */
3612
3613 /* Remember \r or \n */
3614
3615 if (d == CHAR_CR || d == CHAR_NL) cd->external_flags |= PCRE_HASCRORLF;
3616
3617 /* In UTF-8 mode, if the upper limit is > 255, or > 127 for caseless
3618 matching, we have to use an XCLASS with extra data items. Caseless
3619 matching for characters > 127 is available only if UCP support is
3620 available. */
3621
3622 #ifdef SUPPORT_UTF8
3623 if (utf8 && (d > 255 || ((options & PCRE_CASELESS) != 0 && d > 127)))
3624 {
3625 class_utf8 = TRUE;
3626
3627 /* With UCP support, we can find the other case equivalents of
3628 the relevant characters. There may be several ranges. Optimize how
3629 they fit with the basic range. */
3630
3631 #ifdef SUPPORT_UCP
3632 if ((options & PCRE_CASELESS) != 0)
3633 {
3634 unsigned int occ, ocd;
3635 unsigned int cc = c;
3636 unsigned int origd = d;
3637 while (get_othercase_range(&cc, origd, &occ, &ocd))
3638 {
3639 if (occ >= (unsigned int)c &&
3640 ocd <= (unsigned int)d)
3641 continue; /* Skip embedded ranges */
3642
3643 if (occ < (unsigned int)c &&
3644 ocd >= (unsigned int)c - 1) /* Extend the basic range */
3645 { /* if there is overlap, */
3646 c = occ; /* noting that if occ < c */
3647 continue; /* we can't have ocd > d */
3648 } /* because a subrange is */
3649 if (ocd > (unsigned int)d &&
3650 occ <= (unsigned int)d + 1) /* always shorter than */
3651 { /* the basic range. */
3652 d = ocd;
3653 continue;
3654 }
3655
3656 if (occ == ocd)
3657 {
3658 *class_utf8data++ = XCL_SINGLE;
3659 }
3660 else
3661 {
3662 *class_utf8data++ = XCL_RANGE;
3663 class_utf8data += _pcre_ord2utf8(occ, class_utf8data);
3664 }
3665 class_utf8data += _pcre_ord2utf8(ocd, class_utf8data);
3666 }
3667 }
3668 #endif /* SUPPORT_UCP */
3669
3670 /* Now record the original range, possibly modified for UCP caseless
3671 overlapping ranges. */
3672
3673 *class_utf8data++ = XCL_RANGE;
3674 class_utf8data += _pcre_ord2utf8(c, class_utf8data);
3675 class_utf8data += _pcre_ord2utf8(d, class_utf8data);
3676
3677 /* With UCP support, we are done. Without UCP support, there is no
3678 caseless matching for UTF-8 characters > 127; we can use the bit map
3679 for the smaller ones. */
3680
3681 #ifdef SUPPORT_UCP
3682 continue; /* With next character in the class */
3683 #else
3684 if ((options & PCRE_CASELESS) == 0 || c > 127) continue;
3685
3686 /* Adjust upper limit and fall through to set up the map */
3687
3688 d = 127;
3689
3690 #endif /* SUPPORT_UCP */
3691 }
3692 #endif /* SUPPORT_UTF8 */
3693
3694 /* We use the bit map for all cases when not in UTF-8 mode; else
3695 ranges that lie entirely within 0-127 when there is UCP support; else
3696 for partial ranges without UCP support. */
3697
3698 class_charcount += d - c + 1;
3699 class_lastchar = d;
3700
3701 /* We can save a bit of time by skipping this in the pre-compile. */
3702
3703 if (lengthptr == NULL) for (; c <= d; c++)
3704 {
3705 classbits[c/8] |= (1 << (c&7));
3706 if ((options & PCRE_CASELESS) != 0)
3707 {
3708 int uc = cd->fcc[c]; /* flip case */
3709 classbits[uc/8] |= (1 << (uc&7));
3710 }
3711 }
3712
3713 continue; /* Go get the next char in the class */
3714 }
3715
3716 /* Handle a lone single character - we can get here for a normal
3717 non-escape char, or after \ that introduces a single character or for an
3718 apparent range that isn't. */
3719
3720 LONE_SINGLE_CHARACTER:
3721
3722 /* Handle a character that cannot go in the bit map */
3723
3724 #ifdef SUPPORT_UTF8
3725 if (utf8 && (c > 255 || ((options & PCRE_CASELESS) != 0 && c > 127)))
3726 {
3727 class_utf8 = TRUE;
3728 *class_utf8data++ = XCL_SINGLE;
3729 class_utf8data += _pcre_ord2utf8(c, class_utf8data);
3730
3731 #ifdef SUPPORT_UCP
3732 if ((options & PCRE_CASELESS) != 0)
3733 {
3734 unsigned int othercase;
3735 if ((othercase = UCD_OTHERCASE(c)) != c)
3736 {
3737 *class_utf8data++ = XCL_SINGLE;
3738 class_utf8data += _pcre_ord2utf8(othercase, class_utf8data);
3739 }
3740 }
3741 #endif /* SUPPORT_UCP */
3742
3743 }
3744 else
3745 #endif /* SUPPORT_UTF8 */
3746
3747 /* Handle a single-byte character */
3748 {
3749 classbits[c/8] |= (1 << (c&7));
3750 if ((options & PCRE_CASELESS) != 0)
3751 {
3752 c = cd->fcc[c]; /* flip case */
3753 classbits[c/8] |= (1 << (c&7));
3754 }
3755 class_charcount++;
3756 class_lastchar = c;
3757 }
3758 }
3759
3760 /* Loop until ']' reached. This "while" is the end of the "do" far above.
3761 If we are at the end of an internal nested string, revert to the outer
3762 string. */
3763
3764 while (((c = *(++ptr)) != 0 ||
3765 (nestptr != NULL &&
3766 (ptr = nestptr, nestptr = NULL, c = *(++ptr)) != 0)) &&
3767 (c != CHAR_RIGHT_SQUARE_BRACKET || inescq));
3768
3769 /* Check for missing terminating ']' */
3770
3771 if (c == 0)
3772 {
3773 *errorcodeptr = ERR6;
3774 goto FAILED;
3775 }
3776
3777 /* If class_charcount is 1, we saw precisely one character whose value is
3778 less than 256. As long as there were no characters >= 128 and there was no
3779 use of \p or \P, in other words, no use of any XCLASS features, we can
3780 optimize.
3781
3782 In UTF-8 mode, we can optimize the negative case only if there were no
3783 characters >= 128 because OP_NOT and the related opcodes like OP_NOTSTAR
3784 operate on single-bytes only. This is an historical hangover. Maybe one day
3785 we can tidy these opcodes to handle multi-byte characters.
3786
3787 The optimization throws away the bit map. We turn the item into a
3788 1-character OP_CHAR[NC] if it's positive, or OP_NOT if it's negative. Note
3789 that OP_NOT does not support multibyte characters. In the positive case, it
3790 can cause firstbyte to be set. Otherwise, there can be no first char if
3791 this item is first, whatever repeat count may follow. In the case of
3792 reqbyte, save the previous value for reinstating. */
3793
3794 #ifdef SUPPORT_UTF8
3795 if (class_charcount == 1 && !class_utf8 &&
3796 (!utf8 || !negate_class || class_lastchar < 128))
3797 #else
3798 if (class_charcount == 1)
3799 #endif
3800 {
3801 zeroreqbyte = reqbyte;
3802
3803 /* The OP_NOT opcode works on one-byte characters only. */
3804
3805 if (negate_class)
3806 {
3807 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
3808 zerofirstbyte = firstbyte;
3809 *code++ = OP_NOT;
3810 *code++ = class_lastchar;
3811 break;
3812 }
3813
3814 /* For a single, positive character, get the value into mcbuffer, and
3815 then we can handle this with the normal one-character code. */
3816
3817 #ifdef SUPPORT_UTF8
3818 if (utf8 && class_lastchar > 127)
3819 mclength = _pcre_ord2utf8(class_lastchar, mcbuffer);
3820 else
3821 #endif
3822 {
3823 mcbuffer[0] = class_lastchar;
3824 mclength = 1;
3825 }
3826 goto ONE_CHAR;
3827 } /* End of 1-char optimization */
3828
3829 /* The general case - not the one-char optimization. If this is the first
3830 thing in the branch, there can be no first char setting, whatever the
3831 repeat count. Any reqbyte setting must remain unchanged after any kind of
3832 repeat. */
3833
3834 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
3835 zerofirstbyte = firstbyte;
3836 zeroreqbyte = reqbyte;
3837
3838 /* If there are characters with values > 255, we have to compile an
3839 extended class, with its own opcode, unless there was a negated special
3840 such as \S in the class, and PCRE_UCP is not set, because in that case all
3841 characters > 255 are in the class, so any that were explicitly given as
3842 well can be ignored. If (when there are explicit characters > 255 that must
3843 be listed) there are no characters < 256, we can omit the bitmap in the
3844 actual compiled code. */
3845
3846 #ifdef SUPPORT_UTF8
3847 if (class_utf8 && (!should_flip_negation || (options & PCRE_UCP) != 0))
3848 {
3849 *class_utf8data++ = XCL_END; /* Marks the end of extra data */
3850 *code++ = OP_XCLASS;
3851 code += LINK_SIZE;
3852 *code = negate_class? XCL_NOT : 0;
3853
3854 /* If the map is required, move up the extra data to make room for it;
3855 otherwise just move the code pointer to the end of the extra data. */
3856
3857 if (class_charcount > 0)
3858 {
3859 *code++ |= XCL_MAP;
3860 memmove(code + 32, code, class_utf8data - code);
3861 memcpy(code, classbits, 32);
3862 code = class_utf8data + 32;
3863 }
3864 else code = class_utf8data;
3865
3866 /* Now fill in the complete length of the item */
3867
3868 PUT(previous, 1, code - previous);
3869 break; /* End of class handling */
3870 }
3871 #endif
3872
3873 /* If there are no characters > 255, or they are all to be included or
3874 excluded, set the opcode to OP_CLASS or OP_NCLASS, depending on whether the
3875 whole class was negated and whether there were negative specials such as \S
3876 (non-UCP) in the class. Then copy the 32-byte map into the code vector,
3877 negating it if necessary. */
3878
3879 *code++ = (negate_class == should_flip_negation) ? OP_CLASS : OP_NCLASS;
3880 if (negate_class)
3881 {
3882 if (lengthptr == NULL) /* Save time in the pre-compile phase */
3883 for (c = 0; c < 32; c++) code[c] = ~classbits[c];
3884 }
3885 else
3886 {
3887 memcpy(code, classbits, 32);
3888 }
3889 code += 32;
3890 break;
3891
3892
3893 /* ===================================================================*/
3894 /* Various kinds of repeat; '{' is not necessarily a quantifier, but this
3895 has been tested above. */
3896
3897 case CHAR_LEFT_CURLY_BRACKET:
3898 if (!is_quantifier) goto NORMAL_CHAR;
3899 ptr = read_repeat_counts(ptr+1, &repeat_min, &repeat_max, errorcodeptr);
3900 if (*errorcodeptr != 0) goto FAILED;
3901 goto REPEAT;
3902
3903 case CHAR_ASTERISK:
3904 repeat_min = 0;
3905 repeat_max = -1;
3906 goto REPEAT;
3907
3908 case CHAR_PLUS:
3909 repeat_min = 1;
3910 repeat_max = -1;
3911 goto REPEAT;
3912
3913 case CHAR_QUESTION_MARK:
3914 repeat_min = 0;
3915 repeat_max = 1;
3916
3917 REPEAT:
3918 if (previous == NULL)
3919 {
3920 *errorcodeptr = ERR9;
3921 goto FAILED;
3922 }
3923
3924 if (repeat_min == 0)
3925 {
3926 firstbyte = zerofirstbyte; /* Adjust for zero repeat */
3927 reqbyte = zeroreqbyte; /* Ditto */
3928 }
3929
3930 /* Remember whether this is a variable length repeat */
3931
3932 reqvary = (repeat_min == repeat_max)? 0 : REQ_VARY;
3933
3934 op_type = 0; /* Default single-char op codes */
3935 possessive_quantifier = FALSE; /* Default not possessive quantifier */
3936
3937 /* Save start of previous item, in case we have to move it up to make space
3938 for an inserted OP_ONCE for the additional '+' extension. */
3939
3940 tempcode = previous;
3941
3942 /* If the next character is '+', we have a possessive quantifier. This
3943 implies greediness, whatever the setting of the PCRE_UNGREEDY option.
3944 If the next character is '?' this is a minimizing repeat, by default,
3945 but if PCRE_UNGREEDY is set, it works the other way round. We change the
3946 repeat type to the non-default. */
3947
3948 if (ptr[1] == CHAR_PLUS)
3949 {
3950 repeat_type = 0; /* Force greedy */
3951 possessive_quantifier = TRUE;
3952 ptr++;
3953 }
3954 else if (ptr[1] == CHAR_QUESTION_MARK)
3955 {
3956 repeat_type = greedy_non_default;
3957 ptr++;
3958 }
3959 else repeat_type = greedy_default;
3960
3961 /* If previous was a character match, abolish the item and generate a
3962 repeat item instead. If a char item has a minumum of more than one, ensure
3963 that it is set in reqbyte - it might not be if a sequence such as x{3} is
3964 the first thing in a branch because the x will have gone into firstbyte
3965 instead. */
3966
3967 if (*previous == OP_CHAR || *previous == OP_CHARNC)
3968 {
3969 /* Deal with UTF-8 characters that take up more than one byte. It's
3970 easier to write this out separately than try to macrify it. Use c to
3971 hold the length of the character in bytes, plus 0x80 to flag that it's a
3972 length rather than a small character. */
3973
3974 #ifdef SUPPORT_UTF8
3975 if (utf8 && (code[-1] & 0x80) != 0)
3976 {
3977 uschar *lastchar = code - 1;
3978 while((*lastchar & 0xc0) == 0x80) lastchar--;
3979 c = code - lastchar; /* Length of UTF-8 character */
3980 memcpy(utf8_char, lastchar, c); /* Save the char */
3981 c |= 0x80; /* Flag c as a length */
3982 }
3983 else
3984 #endif
3985
3986 /* Handle the case of a single byte - either with no UTF8 support, or
3987 with UTF-8 disabled, or for a UTF-8 character < 128. */
3988
3989 {
3990 c = code[-1];
3991 if (repeat_min > 1) reqbyte = c | req_caseopt | cd->req_varyopt;
3992 }
3993
3994 /* If the repetition is unlimited, it pays to see if the next thing on
3995 the line is something that cannot possibly match this character. If so,
3996 automatically possessifying this item gains some performance in the case
3997 where the match fails. */
3998
3999 if (!possessive_quantifier &&
4000 repeat_max < 0 &&
4001 check_auto_possessive(*previous, c, utf8, utf8_char, ptr + 1,
4002 options, cd))
4003 {
4004 repeat_type = 0; /* Force greedy */
4005 possessive_quantifier = TRUE;
4006 }
4007
4008 goto OUTPUT_SINGLE_REPEAT; /* Code shared with single character types */
4009 }
4010
4011 /* If previous was a single negated character ([^a] or similar), we use
4012 one of the special opcodes, replacing it. The code is shared with single-
4013 character repeats by setting opt_type to add a suitable offset into
4014 repeat_type. We can also test for auto-possessification. OP_NOT is
4015 currently used only for single-byte chars. */
4016
4017 else if (*previous == OP_NOT)
4018 {
4019 op_type = OP_NOTSTAR - OP_STAR; /* Use "not" opcodes */
4020 c = previous[1];
4021 if (!possessive_quantifier &&
4022 repeat_max < 0 &&
4023 check_auto_possessive(OP_NOT, c, utf8, NULL, ptr + 1, options, cd))
4024 {
4025 repeat_type = 0; /* Force greedy */
4026 possessive_quantifier = TRUE;
4027 }
4028 goto OUTPUT_SINGLE_REPEAT;
4029 }
4030
4031 /* If previous was a character type match (\d or similar), abolish it and
4032 create a suitable repeat item. The code is shared with single-character
4033 repeats by setting op_type to add a suitable offset into repeat_type. Note
4034 the the Unicode property types will be present only when SUPPORT_UCP is
4035 defined, but we don't wrap the little bits of code here because it just
4036 makes it horribly messy. */
4037
4038 else if (*previous < OP_EODN)
4039 {
4040 uschar *oldcode;
4041 int prop_type, prop_value;
4042 op_type = OP_TYPESTAR - OP_STAR; /* Use type opcodes */
4043 c = *previous;
4044
4045 if (!possessive_quantifier &&
4046 repeat_max < 0 &&
4047 check_auto_possessive(c, 0, utf8, NULL, ptr + 1, options, cd))
4048 {
4049 repeat_type = 0; /* Force greedy */
4050 possessive_quantifier = TRUE;
4051 }
4052
4053 OUTPUT_SINGLE_REPEAT:
4054 if (*previous == OP_PROP || *previous == OP_NOTPROP)
4055 {
4056 prop_type = previous[1];
4057 prop_value = previous[2];
4058 }
4059 else prop_type = prop_value = -1;
4060
4061 oldcode = code;
4062 code = previous; /* Usually overwrite previous item */
4063
4064 /* If the maximum is zero then the minimum must also be zero; Perl allows
4065 this case, so we do too - by simply omitting the item altogether. */
4066
4067 if (repeat_max == 0) goto END_REPEAT;
4068
4069 /*--------------------------------------------------------------------*/
4070 /* This code is obsolete from release 8.00; the restriction was finally
4071 removed: */
4072
4073 /* All real repeats make it impossible to handle partial matching (maybe
4074 one day we will be able to remove this restriction). */
4075
4076 /* if (repeat_max != 1) cd->external_flags |= PCRE_NOPARTIAL; */
4077 /*--------------------------------------------------------------------*/
4078
4079 /* Combine the op_type with the repeat_type */
4080
4081 repeat_type += op_type;
4082
4083 /* A minimum of zero is handled either as the special case * or ?, or as
4084 an UPTO, with the maximum given. */
4085
4086 if (repeat_min == 0)
4087 {
4088 if (repeat_max == -1) *code++ = OP_STAR + repeat_type;
4089 else if (repeat_max == 1) *code++ = OP_QUERY + repeat_type;
4090 else
4091 {
4092 *code++ = OP_UPTO + repeat_type;
4093 PUT2INC(code, 0, repeat_max);
4094 }
4095 }
4096
4097 /* A repeat minimum of 1 is optimized into some special cases. If the
4098 maximum is unlimited, we use OP_PLUS. Otherwise, the original item is
4099 left in place and, if the maximum is greater than 1, we use OP_UPTO with
4100 one less than the maximum. */
4101
4102 else if (repeat_min == 1)
4103 {
4104 if (repeat_max == -1)
4105 *code++ = OP_PLUS + repeat_type;
4106 else
4107 {
4108 code = oldcode; /* leave previous item in place */
4109 if (repeat_max == 1) goto END_REPEAT;
4110 *code++ = OP_UPTO + repeat_type;
4111 PUT2INC(code, 0, repeat_max - 1);
4112 }
4113 }
4114
4115 /* The case {n,n} is just an EXACT, while the general case {n,m} is
4116 handled as an EXACT followed by an UPTO. */
4117
4118 else
4119 {
4120 *code++ = OP_EXACT + op_type; /* NB EXACT doesn't have repeat_type */
4121 PUT2INC(code, 0, repeat_min);
4122
4123 /* If the maximum is unlimited, insert an OP_STAR. Before doing so,
4124 we have to insert the character for the previous code. For a repeated
4125 Unicode property match, there are two extra bytes that define the
4126 required property. In UTF-8 mode, long characters have their length in
4127 c, with the 0x80 bit as a flag. */
4128
4129 if (repeat_max < 0)
4130 {
4131 #ifdef SUPPORT_UTF8
4132 if (utf8 && c >= 128)
4133 {
4134 memcpy(code, utf8_char, c & 7);
4135 code += c & 7;
4136 }
4137 else
4138 #endif
4139 {
4140 *code++ = c;
4141 if (prop_type >= 0)
4142 {
4143 *code++ = prop_type;
4144 *code++ = prop_value;
4145 }
4146 }
4147 *code++ = OP_STAR + repeat_type;
4148 }
4149
4150 /* Else insert an UPTO if the max is greater than the min, again
4151 preceded by the character, for the previously inserted code. If the
4152 UPTO is just for 1 instance, we can use QUERY instead. */
4153
4154 else if (repeat_max != repeat_min)
4155 {
4156 #ifdef SUPPORT_UTF8
4157 if (utf8 && c >= 128)
4158 {
4159 memcpy(code, utf8_char, c & 7);
4160 code += c & 7;
4161 }
4162 else
4163 #endif
4164 *code++ = c;
4165 if (prop_type >= 0)
4166 {
4167 *code++ = prop_type;
4168 *code++ = prop_value;
4169 }
4170 repeat_max -= repeat_min;
4171
4172 if (repeat_max == 1)
4173 {
4174 *code++ = OP_QUERY + repeat_type;
4175 }
4176 else
4177 {
4178 *code++ = OP_UPTO + repeat_type;
4179 PUT2INC(code, 0, repeat_max);
4180 }
4181 }
4182 }
4183
4184 /* The character or character type itself comes last in all cases. */
4185
4186 #ifdef SUPPORT_UTF8
4187 if (utf8 && c >= 128)
4188 {
4189 memcpy(code, utf8_char, c & 7);
4190 code += c & 7;
4191 }
4192 else
4193 #endif
4194 *code++ = c;
4195
4196 /* For a repeated Unicode property match, there are two extra bytes that
4197 define the required property. */
4198
4199 #ifdef SUPPORT_UCP
4200 if (prop_type >= 0)
4201 {
4202 *code++ = prop_type;
4203 *code++ = prop_value;
4204 }
4205 #endif
4206 }
4207
4208 /* If previous was a character class or a back reference, we put the repeat
4209 stuff after it, but just skip the item if the repeat was {0,0}. */
4210
4211 else if (*previous == OP_CLASS ||
4212 *previous == OP_NCLASS ||
4213 #ifdef SUPPORT_UTF8
4214 *previous == OP_XCLASS ||
4215 #endif
4216 *previous == OP_REF)
4217 {
4218 if (repeat_max == 0)
4219 {
4220 code = previous;
4221 goto END_REPEAT;
4222 }
4223
4224 /*--------------------------------------------------------------------*/
4225 /* This code is obsolete from release 8.00; the restriction was finally
4226 removed: */
4227
4228 /* All real repeats make it impossible to handle partial matching (maybe
4229 one day we will be able to remove this restriction). */
4230
4231 /* if (repeat_max != 1) cd->external_flags |= PCRE_NOPARTIAL; */
4232 /*--------------------------------------------------------------------*/
4233
4234 if (repeat_min == 0 && repeat_max == -1)
4235 *code++ = OP_CRSTAR + repeat_type;
4236 else if (repeat_min == 1 && repeat_max == -1)
4237 *code++ = OP_CRPLUS + repeat_type;
4238 else if (repeat_min == 0 && repeat_max == 1)
4239 *code++ = OP_CRQUERY + repeat_type;
4240 else
4241 {
4242 *code++ = OP_CRRANGE + repeat_type;
4243 PUT2INC(code, 0, repeat_min);
4244 if (repeat_max == -1) repeat_max = 0; /* 2-byte encoding for max */
4245 PUT2INC(code, 0, repeat_max);
4246 }
4247 }
4248
4249 /* If previous was a bracket group, we may have to replicate it in certain
4250 cases. */
4251
4252 else if (*previous == OP_BRA || *previous == OP_CBRA ||
4253 *previous == OP_ONCE || *previous == OP_COND)
4254 {
4255 register int i;
4256 int ketoffset = 0;
4257 int len = (int)(code - previous);
4258 uschar *bralink = NULL;
4259
4260 /* Repeating a DEFINE group is pointless */
4261
4262 if (*previous == OP_COND && previous[LINK_SIZE+1] == OP_DEF)
4263 {
4264 *errorcodeptr = ERR55;
4265 goto FAILED;
4266 }
4267
4268 /* If the maximum repeat count is unlimited, find the end of the bracket
4269 by scanning through from the start, and compute the offset back to it
4270 from the current code pointer. There may be an OP_OPT setting following
4271 the final KET, so we can't find the end just by going back from the code
4272 pointer. */
4273
4274 if (repeat_max == -1)
4275 {
4276 register uschar *ket = previous;
4277 do ket += GET(ket, 1); while (*ket != OP_KET);
4278 ketoffset = (int)(code - ket);
4279 }
4280
4281 /* The case of a zero minimum is special because of the need to stick
4282 OP_BRAZERO in front of it, and because the group appears once in the
4283 data, whereas in other cases it appears the minimum number of times. For
4284 this reason, it is simplest to treat this case separately, as otherwise
4285 the code gets far too messy. There are several special subcases when the
4286 minimum is zero. */
4287
4288 if (repeat_min == 0)
4289 {
4290 /* If the maximum is also zero, we used to just omit the group from the
4291 output altogether, like this:
4292
4293 ** if (repeat_max == 0)
4294 ** {
4295 ** code = previous;
4296 ** goto END_REPEAT;
4297 ** }
4298
4299 However, that fails when a group is referenced as a subroutine from
4300 elsewhere in the pattern, so now we stick in OP_SKIPZERO in front of it
4301 so that it is skipped on execution. As we don't have a list of which
4302 groups are referenced, we cannot do this selectively.
4303
4304 If the maximum is 1 or unlimited, we just have to stick in the BRAZERO
4305 and do no more at this point. However, we do need to adjust any
4306 OP_RECURSE calls inside the group that refer to the group itself or any
4307 internal or forward referenced group, because the offset is from the
4308 start of the whole regex. Temporarily terminate the pattern while doing
4309 this. */
4310
4311 if (repeat_max <= 1) /* Covers 0, 1, and unlimited */
4312 {
4313 *code = OP_END;
4314 adjust_recurse(previous, 1, utf8, cd, save_hwm);
4315 memmove(previous+1, previous, len);
4316 code++;
4317 if (repeat_max == 0)
4318 {
4319 *previous++ = OP_SKIPZERO;
4320 goto END_REPEAT;
4321 }
4322 *previous++ = OP_BRAZERO + repeat_type;
4323 }
4324
4325 /* If the maximum is greater than 1 and limited, we have to replicate
4326 in a nested fashion, sticking OP_BRAZERO before each set of brackets.
4327 The first one has to be handled carefully because it's the original
4328 copy, which has to be moved up. The remainder can be handled by code
4329 that is common with the non-zero minimum case below. We have to
4330 adjust the value or repeat_max, since one less copy is required. Once
4331 again, we may have to adjust any OP_RECURSE calls inside the group. */
4332
4333 else
4334 {
4335 int offset;
4336 *code = OP_END;
4337 adjust_recurse(previous, 2 + LINK_SIZE, utf8, cd, save_hwm);
4338 memmove(previous + 2 + LINK_SIZE, previous, len);
4339 code += 2 + LINK_SIZE;
4340 *previous++ = OP_BRAZERO + repeat_type;
4341 *previous++ = OP_BRA;
4342
4343 /* We chain together the bracket offset fields that have to be
4344 filled in later when the ends of the brackets are reached. */
4345
4346 offset = (bralink == NULL)? 0 : (int)(previous - bralink);
4347 bralink = previous;
4348 PUTINC(previous, 0, offset);
4349 }
4350
4351 repeat_max--;
4352 }
4353
4354 /* If the minimum is greater than zero, replicate the group as many
4355 times as necessary, and adjust the maximum to the number of subsequent
4356 copies that we need. If we set a first char from the group, and didn't
4357 set a required char, copy the latter from the former. If there are any
4358 forward reference subroutine calls in the group, there will be entries on
4359 the workspace list; replicate these with an appropriate increment. */
4360
4361 else
4362 {
4363 if (repeat_min > 1)
4364 {
4365 /* In the pre-compile phase, we don't actually do the replication. We
4366 just adjust the length as if we had. Do some paranoid checks for
4367 potential integer overflow. The INT64_OR_DOUBLE type is a 64-bit
4368 integer type when available, otherwise double. */
4369
4370 if (lengthptr != NULL)
4371 {
4372 int delta = (repeat_min - 1)*length_prevgroup;
4373 if ((INT64_OR_DOUBLE)(repeat_min - 1)*
4374 (INT64_OR_DOUBLE)length_prevgroup >
4375 (INT64_OR_DOUBLE)INT_MAX ||
4376 OFLOW_MAX - *lengthptr < delta)
4377 {
4378 *errorcodeptr = ERR20;
4379 goto FAILED;
4380 }
4381 *lengthptr += delta;
4382 }
4383
4384 /* This is compiling for real */
4385
4386 else
4387 {
4388 if (groupsetfirstbyte && reqbyte < 0) reqbyte = firstbyte;
4389 for (i = 1; i < repeat_min; i++)
4390 {
4391 uschar *hc;
4392 uschar *this_hwm = cd->hwm;
4393 memcpy(code, previous, len);
4394 for (hc = save_hwm; hc < this_hwm; hc += LINK_SIZE)
4395 {
4396 PUT(cd->hwm, 0, GET(hc, 0) + len);
4397 cd->hwm += LINK_SIZE;
4398 }
4399 save_hwm = this_hwm;
4400 code += len;
4401 }
4402 }
4403 }
4404
4405 if (repeat_max > 0) repeat_max -= repeat_min;
4406 }
4407
4408 /* This code is common to both the zero and non-zero minimum cases. If
4409 the maximum is limited, it replicates the group in a nested fashion,
4410 remembering the bracket starts on a stack. In the case of a zero minimum,
4411 the first one was set up above. In all cases the repeat_max now specifies
4412 the number of additional copies needed. Again, we must remember to
4413 replicate entries on the forward reference list. */
4414
4415 if (repeat_max >= 0)
4416 {
4417 /* In the pre-compile phase, we don't actually do the replication. We
4418 just adjust the length as if we had. For each repetition we must add 1
4419 to the length for BRAZERO and for all but the last repetition we must
4420 add 2 + 2*LINKSIZE to allow for the nesting that occurs. Do some
4421 paranoid checks to avoid integer overflow. The INT64_OR_DOUBLE type is
4422 a 64-bit integer type when available, otherwise double. */
4423
4424 if (lengthptr != NULL && repeat_max > 0)
4425 {
4426 int delta = repeat_max * (length_prevgroup + 1 + 2 + 2*LINK_SIZE) -
4427 2 - 2*LINK_SIZE; /* Last one doesn't nest */
4428 if ((INT64_OR_DOUBLE)repeat_max *
4429 (INT64_OR_DOUBLE)(length_prevgroup + 1 + 2 + 2*LINK_SIZE)
4430 > (INT64_OR_DOUBLE)INT_MAX ||
4431 OFLOW_MAX - *lengthptr < delta)
4432 {
4433 *errorcodeptr = ERR20;
4434 goto FAILED;
4435 }
4436 *lengthptr += delta;
4437 }
4438
4439 /* This is compiling for real */
4440
4441 else for (i = repeat_max - 1; i >= 0; i--)
4442 {
4443 uschar *hc;
4444 uschar *this_hwm = cd->hwm;
4445
4446 *code++ = OP_BRAZERO + repeat_type;
4447
4448 /* All but the final copy start a new nesting, maintaining the
4449 chain of brackets outstanding. */
4450
4451 if (i != 0)
4452 {
4453 int offset;
4454 *code++ = OP_BRA;
4455 offset = (bralink == NULL)? 0 : (int)(code - bralink);
4456 bralink = code;
4457 PUTINC(code, 0, offset);
4458 }
4459
4460 memcpy(code, previous, len);
4461 for (hc = save_hwm; hc < this_hwm; hc += LINK_SIZE)
4462 {
4463 PUT(cd->hwm, 0, GET(hc, 0) + len + ((i != 0)? 2+LINK_SIZE : 1));
4464 cd->hwm += LINK_SIZE;
4465 }
4466 save_hwm = this_hwm;
4467 code += len;
4468 }
4469
4470 /* Now chain through the pending brackets, and fill in their length
4471 fields (which are holding the chain links pro tem). */
4472
4473 while (bralink != NULL)
4474 {
4475 int oldlinkoffset;
4476 int offset = (int)(code - bralink + 1);
4477 uschar *bra = code - offset;
4478 oldlinkoffset = GET(bra, 1);
4479 bralink = (oldlinkoffset == 0)? NULL : bralink - oldlinkoffset;
4480 *code++ = OP_KET;
4481 PUTINC(code, 0, offset);
4482 PUT(bra, 1, offset);
4483 }
4484 }
4485
4486 /* If the maximum is unlimited, set a repeater in the final copy. We
4487 can't just offset backwards from the current code point, because we
4488 don't know if there's been an options resetting after the ket. The
4489 correct offset was computed above.
4490
4491 Then, when we are doing the actual compile phase, check to see whether
4492 this group is a non-atomic one that could match an empty string. If so,
4493 convert the initial operator to the S form (e.g. OP_BRA -> OP_SBRA) so
4494 that runtime checking can be done. [This check is also applied to
4495 atomic groups at runtime, but in a different way.] */
4496
4497 else
4498 {
4499 uschar *ketcode = code - ketoffset;
4500 uschar *bracode = ketcode - GET(ketcode, 1);
4501 *ketcode = OP_KETRMAX + repeat_type;
4502 if (lengthptr == NULL && *bracode != OP_ONCE)
4503 {
4504 uschar *scode = bracode;
4505 do
4506 {
4507 if (could_be_empty_branch(scode, ketcode, utf8, cd))
4508 {
4509 *bracode += OP_SBRA - OP_BRA;
4510 break;
4511 }
4512 scode += GET(scode, 1);
4513 }
4514 while (*scode == OP_ALT);
4515 }
4516 }
4517 }
4518
4519 /* If previous is OP_FAIL, it was generated by an empty class [] in
4520 JavaScript mode. The other ways in which OP_FAIL can be generated, that is
4521 by (*FAIL) or (?!) set previous to NULL, which gives a "nothing to repeat"
4522 error above. We can just ignore the repeat in JS case. */
4523
4524 else if (*previous == OP_FAIL) goto END_REPEAT;
4525
4526 /* Else there's some kind of shambles */
4527
4528 else
4529 {
4530 *errorcodeptr = ERR11;
4531 goto FAILED;
4532 }
4533
4534 /* If the character following a repeat is '+', or if certain optimization
4535 tests above succeeded, possessive_quantifier is TRUE. For some of the
4536 simpler opcodes, there is an special alternative opcode for this. For
4537 anything else, we wrap the entire repeated item inside OP_ONCE brackets.
4538 The '+' notation is just syntactic sugar, taken from Sun's Java package,
4539 but the special opcodes can optimize it a bit. The repeated item starts at
4540 tempcode, not at previous, which might be the first part of a string whose
4541 (former) last char we repeated.
4542
4543 Possessifying an 'exact' quantifier has no effect, so we can ignore it. But
4544 an 'upto' may follow. We skip over an 'exact' item, and then test the
4545 length of what remains before proceeding. */
4546
4547 if (possessive_quantifier)
4548 {
4549 int len;
4550
4551 if (*tempcode == OP_TYPEEXACT)
4552 tempcode += _pcre_OP_lengths[*tempcode] +
4553 ((tempcode[3] == OP_PROP || tempcode[3] == OP_NOTPROP)? 2 : 0);
4554
4555 else if (*tempcode == OP_EXACT || *tempcode == OP_NOTEXACT)
4556 {
4557 tempcode += _pcre_OP_lengths[*tempcode];
4558 #ifdef SUPPORT_UTF8
4559 if (utf8 && tempcode[-1] >= 0xc0)
4560 tempcode += _pcre_utf8_table4[tempcode[-1] & 0x3f];
4561 #endif
4562 }
4563
4564 len = (int)(code - tempcode);
4565 if (len > 0) switch (*tempcode)
4566 {
4567 case OP_STAR: *tempcode = OP_POSSTAR; break;
4568 case OP_PLUS: *tempcode = OP_POSPLUS; break;
4569 case OP_QUERY: *tempcode = OP_POSQUERY; break;
4570 case OP_UPTO: *tempcode = OP_POSUPTO; break;
4571
4572 case OP_TYPESTAR: *tempcode = OP_TYPEPOSSTAR; break;
4573 case OP_TYPEPLUS: *tempcode = OP_TYPEPOSPLUS; break;
4574 case OP_TYPEQUERY: *tempcode = OP_TYPEPOSQUERY; break;
4575 case OP_TYPEUPTO: *tempcode = OP_TYPEPOSUPTO; break;
4576
4577 case OP_NOTSTAR: *tempcode = OP_NOTPOSSTAR; break;
4578 case OP_NOTPLUS: *tempcode = OP_NOTPOSPLUS; break;
4579 case OP_NOTQUERY: *tempcode = OP_NOTPOSQUERY; break;
4580 case OP_NOTUPTO: *tempcode = OP_NOTPOSUPTO; break;
4581
4582 /* Because we are moving code along, we must ensure that any
4583 pending recursive references are updated. */
4584
4585 default:
4586 *code = OP_END;
4587 adjust_recurse(tempcode, 1 + LINK_SIZE, utf8, cd, save_hwm);
4588 memmove(tempcode + 1+LINK_SIZE, tempcode, len);
4589 code += 1 + LINK_SIZE;
4590 len += 1 + LINK_SIZE;
4591 tempcode[0] = OP_ONCE;
4592 *code++ = OP_KET;
4593 PUTINC(code, 0, len);
4594 PUT(tempcode, 1, len);
4595 break;
4596 }
4597 }
4598
4599 /* In all case we no longer have a previous item. We also set the
4600 "follows varying string" flag for subsequently encountered reqbytes if
4601 it isn't already set and we have just passed a varying length item. */
4602
4603 END_REPEAT:
4604 previous = NULL;
4605 cd->req_varyopt |= reqvary;
4606 break;
4607
4608
4609 /* ===================================================================*/
4610 /* Start of nested parenthesized sub-expression, or comment or lookahead or
4611 lookbehind or option setting or condition or all the other extended
4612 parenthesis forms. */
4613
4614 case CHAR_LEFT_PARENTHESIS:
4615 newoptions = options;
4616 skipbytes = 0;
4617 bravalue = OP_CBRA;
4618 save_hwm = cd->hwm;
4619 reset_bracount = FALSE;
4620
4621 /* First deal with various "verbs" that can be introduced by '*'. */
4622
4623 if (*(++ptr) == CHAR_ASTERISK &&
4624 ((cd->ctypes[ptr[1]] & ctype_letter) != 0 || ptr[1] == ':'))
4625 {
4626 int i, namelen;
4627 int arglen = 0;
4628 const char *vn = verbnames;
4629 const uschar *name = ptr + 1;
4630 const uschar *arg = NULL;
4631 previous = NULL;
4632 while ((cd->ctypes[*++ptr] & ctype_letter) != 0) {};
4633 namelen = (int)(ptr - name);
4634
4635 if (*ptr == CHAR_COLON)
4636 {
4637 arg = ++ptr;
4638 while ((cd->ctypes[*ptr] & (ctype_letter|ctype_digit)) != 0
4639 || *ptr == '_') ptr++;
4640 arglen = ptr - arg;
4641 }
4642
4643 if (*ptr != CHAR_RIGHT_PARENTHESIS)
4644 {
4645 *errorcodeptr = ERR60;
4646 goto FAILED;
4647 }
4648
4649 /* Scan the table of verb names */
4650
4651 for (i = 0; i < verbcount; i++)
4652 {
4653 if (namelen == verbs[i].len &&
4654 strncmp((char *)name, vn, namelen) == 0)
4655 {
4656 /* Check for open captures before ACCEPT */
4657
4658 if (verbs[i].op == OP_ACCEPT)
4659 {
4660 open_capitem *oc;
4661 cd->had_accept = TRUE;
4662 for (oc = cd->open_caps; oc != NULL; oc = oc->next)
4663 {
4664 *code++ = OP_CLOSE;
4665 PUT2INC(code, 0, oc->number);
4666 }
4667 }
4668
4669 /* Handle the cases with/without an argument */
4670
4671 if (arglen == 0)
4672 {
4673 if (verbs[i].op < 0) /* Argument is mandatory */
4674 {
4675 *errorcodeptr = ERR66;
4676 goto FAILED;
4677 }
4678 *code++ = verbs[i].op;
4679 }
4680
4681 else
4682 {
4683 if (verbs[i].op_arg < 0) /* Argument is forbidden */
4684 {
4685 *errorcodeptr = ERR59;
4686 goto FAILED;
4687 }
4688 *code++ = verbs[i].op_arg;
4689 *code++ = arglen;
4690 memcpy(code, arg, arglen);
4691 code += arglen;
4692 *code++ = 0;
4693 }
4694
4695 break; /* Found verb, exit loop */
4696 }
4697
4698 vn += verbs[i].len + 1;
4699 }
4700
4701 if (i < verbcount) continue; /* Successfully handled a verb */
4702 *errorcodeptr = ERR60; /* Verb not recognized */
4703 goto FAILED;
4704 }
4705
4706 /* Deal with the extended parentheses; all are introduced by '?', and the
4707 appearance of any of them means that this is not a capturing group. */
4708
4709 else if (*ptr == CHAR_QUESTION_MARK)
4710 {
4711 int i, set, unset, namelen;
4712 int *optset;
4713 const uschar *name;
4714 uschar *slot;
4715
4716 switch (*(++ptr))
4717 {
4718 case CHAR_NUMBER_SIGN: /* Comment; skip to ket */
4719 ptr++;
4720 while (*ptr != 0 && *ptr != CHAR_RIGHT_PARENTHESIS) ptr++;
4721 if (*ptr == 0)
4722 {
4723 *errorcodeptr = ERR18;
4724 goto FAILED;
4725 }
4726 continue;
4727
4728
4729 /* ------------------------------------------------------------ */
4730 case CHAR_VERTICAL_LINE: /* Reset capture count for each branch */
4731 reset_bracount = TRUE;
4732 /* Fall through */
4733
4734 /* ------------------------------------------------------------ */
4735 case CHAR_COLON: /* Non-capturing bracket */
4736 bravalue = OP_BRA;
4737 ptr++;
4738 break;
4739
4740
4741 /* ------------------------------------------------------------ */
4742 case CHAR_LEFT_PARENTHESIS:
4743 bravalue = OP_COND; /* Conditional group */
4744
4745 /* A condition can be an assertion, a number (referring to a numbered
4746 group), a name (referring to a named group), or 'R', referring to
4747 recursion. R<digits> and R&name are also permitted for recursion tests.
4748
4749 There are several syntaxes for testing a named group: (?(name)) is used
4750 by Python; Perl 5.10 onwards uses (?(<name>) or (?('name')).
4751
4752 There are two unfortunate ambiguities, caused by history. (a) 'R' can
4753 be the recursive thing or the name 'R' (and similarly for 'R' followed
4754 by digits), and (b) a number could be a name that consists of digits.
4755 In both cases, we look for a name first; if not found, we try the other
4756 cases. */
4757
4758 /* For conditions that are assertions, check the syntax, and then exit
4759 the switch. This will take control down to where bracketed groups,
4760 including assertions, are processed. */
4761
4762 if (ptr[1] == CHAR_QUESTION_MARK && (ptr[2] == CHAR_EQUALS_SIGN ||
4763 ptr[2] == CHAR_EXCLAMATION_MARK || ptr[2] == CHAR_LESS_THAN_SIGN))
4764 break;
4765
4766 /* Most other conditions use OP_CREF (a couple change to OP_RREF
4767 below), and all need to skip 3 bytes at the start of the group. */
4768
4769 code[1+LINK_SIZE] = OP_CREF;
4770 skipbytes = 3;
4771 refsign = -1;
4772
4773 /* Check for a test for recursion in a named group. */
4774
4775 if (ptr[1] == CHAR_R && ptr[2] == CHAR_AMPERSAND)
4776 {
4777 terminator = -1;
4778 ptr += 2;
4779 code[1+LINK_SIZE] = OP_RREF; /* Change the type of test */
4780 }
4781
4782 /* Check for a test for a named group's having been set, using the Perl
4783 syntax (?(<name>) or (?('name') */
4784
4785 else if (ptr[1] == CHAR_LESS_THAN_SIGN)
4786 {
4787 terminator = CHAR_GREATER_THAN_SIGN;
4788 ptr++;
4789 }
4790 else if (ptr[1] == CHAR_APOSTROPHE)
4791 {
4792 terminator = CHAR_APOSTROPHE;
4793 ptr++;
4794 }
4795 else
4796 {
4797 terminator = 0;
4798 if (ptr[1] == CHAR_MINUS || ptr[1] == CHAR_PLUS) refsign = *(++ptr);
4799 }
4800
4801 /* We now expect to read a name; any thing else is an error */
4802
4803 if ((cd->ctypes[ptr[1]] & ctype_word) == 0)
4804 {
4805 ptr += 1; /* To get the right offset */
4806 *errorcodeptr = ERR28;
4807 goto FAILED;
4808 }
4809
4810 /* Read the name, but also get it as a number if it's all digits */
4811
4812 recno = 0;
4813 name = ++ptr;
4814 while ((cd->ctypes[*ptr] & ctype_word) != 0)
4815 {
4816 if (recno >= 0)
4817 recno = ((digitab[*ptr] & ctype_digit) != 0)?
4818 recno * 10 + *ptr - CHAR_0 : -1;
4819 ptr++;
4820 }
4821 namelen = (int)(ptr - name);
4822
4823 if ((terminator > 0 && *ptr++ != terminator) ||
4824 *ptr++ != CHAR_RIGHT_PARENTHESIS)
4825 {
4826 ptr--; /* Error offset */
4827 *errorcodeptr = ERR26;
4828 goto FAILED;
4829 }
4830
4831 /* Do no further checking in the pre-compile phase. */
4832
4833 if (lengthptr != NULL) break;
4834
4835 /* In the real compile we do the work of looking for the actual
4836 reference. If the string started with "+" or "-" we require the rest to
4837 be digits, in which case recno will be set. */
4838
4839 if (refsign > 0)
4840 {
4841 if (recno <= 0)
4842 {
4843 *errorcodeptr = ERR58;
4844 goto FAILED;
4845 }
4846 recno = (refsign == CHAR_MINUS)?
4847 cd->bracount - recno + 1 : recno +cd->bracount;
4848 if (recno <= 0 || recno > cd->final_bracount)
4849 {
4850 *errorcodeptr = ERR15;
4851 goto FAILED;
4852 }
4853 PUT2(code, 2+LINK_SIZE, recno);
4854 break;
4855 }
4856
4857 /* Otherwise (did not start with "+" or "-"), start by looking for the
4858 name. If we find a name, add one to the opcode to change OP_CREF or
4859 OP_RREF into OP_NCREF or OP_NRREF. These behave exactly the same,
4860 except they record that the reference was originally to a name. The
4861 information is used to check duplicate names. */
4862
4863 slot = cd->name_table;
4864 for (i = 0; i < cd->names_found; i++)
4865 {
4866 if (strncmp((char *)name, (char *)slot+2, namelen) == 0) break;
4867 slot += cd->name_entry_size;
4868 }
4869
4870 /* Found a previous named subpattern */
4871
4872 if (i < cd->names_found)
4873 {
4874 recno = GET2(slot, 0);
4875 PUT2(code, 2+LINK_SIZE, recno);
4876 code[1+LINK_SIZE]++;
4877 }
4878
4879 /* Search the pattern for a forward reference */
4880
4881 else if ((i = find_parens(cd, name, namelen,
4882 (options & PCRE_EXTENDED) != 0)) > 0)
4883 {
4884 PUT2(code, 2+LINK_SIZE, i);
4885 code[1+LINK_SIZE]++;
4886 }
4887
4888 /* If terminator == 0 it means that the name followed directly after
4889 the opening parenthesis [e.g. (?(abc)...] and in this case there are
4890 some further alternatives to try. For the cases where terminator != 0
4891 [things like (?(<name>... or (?('name')... or (?(R&name)... ] we have
4892 now checked all the possibilities, so give an error. */
4893
4894 else if (terminator != 0)
4895 {
4896 *errorcodeptr = ERR15;
4897 goto FAILED;
4898 }
4899
4900 /* Check for (?(R) for recursion. Allow digits after R to specify a
4901 specific group number. */
4902
4903 else if (*name == CHAR_R)
4904 {
4905 recno = 0;
4906 for (i = 1; i < namelen; i++)
4907 {
4908 if ((digitab[name[i]] & ctype_digit) == 0)
4909 {
4910 *errorcodeptr = ERR15;
4911 goto FAILED;
4912 }
4913 recno = recno * 10 + name[i] - CHAR_0;
4914 }
4915 if (recno == 0) recno = RREF_ANY;
4916 code[1+LINK_SIZE] = OP_RREF; /* Change test type */
4917 PUT2(code, 2+LINK_SIZE, recno);
4918 }
4919
4920 /* Similarly, check for the (?(DEFINE) "condition", which is always
4921 false. */
4922
4923 else if (namelen == 6 && strncmp((char *)name, STRING_DEFINE, 6) == 0)
4924 {
4925 code[1+LINK_SIZE] = OP_DEF;
4926 skipbytes = 1;
4927 }
4928
4929 /* Check for the "name" actually being a subpattern number. We are
4930 in the second pass here, so final_bracount is set. */
4931
4932 else if (recno > 0 && recno <= cd->final_bracount)
4933 {
4934 PUT2(code, 2+LINK_SIZE, recno);
4935 }
4936
4937 /* Either an unidentified subpattern, or a reference to (?(0) */
4938
4939 else
4940 {
4941 *errorcodeptr = (recno == 0)? ERR35: ERR15;
4942 goto FAILED;
4943 }
4944 break;
4945
4946
4947 /* ------------------------------------------------------------ */
4948 case CHAR_EQUALS_SIGN: /* Positive lookahead */
4949 bravalue = OP_ASSERT;
4950 ptr++;
4951 break;
4952
4953
4954 /* ------------------------------------------------------------ */
4955 case CHAR_EXCLAMATION_MARK: /* Negative lookahead */
4956 ptr++;
4957 if (*ptr == CHAR_RIGHT_PARENTHESIS) /* Optimize (?!) */
4958 {
4959 *code++ = OP_FAIL;
4960 previous = NULL;
4961 continue;
4962 }
4963 bravalue = OP_ASSERT_NOT;
4964 break;
4965
4966
4967 /* ------------------------------------------------------------ */
4968 case CHAR_LESS_THAN_SIGN: /* Lookbehind or named define */
4969 switch (ptr[1])
4970 {
4971 case CHAR_EQUALS_SIGN: /* Positive lookbehind */
4972 bravalue = OP_ASSERTBACK;
4973 ptr += 2;
4974 break;
4975
4976 case CHAR_EXCLAMATION_MARK: /* Negative lookbehind */
4977 bravalue = OP_ASSERTBACK_NOT;
4978 ptr += 2;
4979 break;
4980
4981 default: /* Could be name define, else bad */
4982 if ((cd->ctypes[ptr[1]] & ctype_word) != 0) goto DEFINE_NAME;
4983 ptr++; /* Correct offset for error */
4984 *errorcodeptr = ERR24;
4985 goto FAILED;
4986 }
4987 break;
4988
4989
4990 /* ------------------------------------------------------------ */
4991 case CHAR_GREATER_THAN_SIGN: /* One-time brackets */
4992 bravalue = OP_ONCE;
4993 ptr++;
4994 break;
4995
4996
4997 /* ------------------------------------------------------------ */
4998 case CHAR_C: /* Callout - may be followed by digits; */
4999 previous_callout = code; /* Save for later completion */
5000 after_manual_callout = 1; /* Skip one item before completing */
5001 *code++ = OP_CALLOUT;
5002 {
5003 int n = 0;
5004 while ((digitab[*(++ptr)] & ctype_digit) != 0)
5005 n = n * 10 + *ptr - CHAR_0;
5006 if (*ptr != CHAR_RIGHT_PARENTHESIS)
5007 {
5008 *errorcodeptr = ERR39;
5009 goto FAILED;
5010 }
5011 if (n > 255)
5012 {
5013 *errorcodeptr = ERR38;
5014 goto FAILED;
5015 }
5016 *code++ = n;
5017 PUT(code, 0, (int)(ptr - cd->start_pattern + 1)); /* Pattern offset */
5018 PUT(code, LINK_SIZE, 0); /* Default length */
5019 code += 2 * LINK_SIZE;
5020 }
5021 previous = NULL;
5022 continue;
5023
5024
5025 /* ------------------------------------------------------------ */
5026 case CHAR_P: /* Python-style named subpattern handling */
5027 if (*(++ptr) == CHAR_EQUALS_SIGN ||
5028 *ptr == CHAR_GREATER_THAN_SIGN) /* Reference or recursion */
5029 {
5030 is_recurse = *ptr == CHAR_GREATER_THAN_SIGN;
5031 terminator = CHAR_RIGHT_PARENTHESIS;
5032 goto NAMED_REF_OR_RECURSE;
5033 }
5034 else if (*ptr != CHAR_LESS_THAN_SIGN) /* Test for Python-style defn */
5035 {
5036 *errorcodeptr = ERR41;
5037 goto FAILED;
5038 }
5039 /* Fall through to handle (?P< as (?< is handled */
5040
5041
5042 /* ------------------------------------------------------------ */
5043 DEFINE_NAME: /* Come here from (?< handling */
5044 case CHAR_APOSTROPHE:
5045 {
5046 terminator = (*ptr == CHAR_LESS_THAN_SIGN)?
5047 CHAR_GREATER_THAN_SIGN : CHAR_APOSTROPHE;
5048 name = ++ptr;
5049
5050 while ((cd->ctypes[*ptr] & ctype_word) != 0) ptr++;
5051 namelen = (int)(ptr - name);
5052
5053 /* In the pre-compile phase, just do a syntax check. */
5054
5055 if (lengthptr != NULL)
5056 {
5057 if (*ptr != terminator)
5058 {
5059 *errorcodeptr = ERR42;
5060 goto FAILED;
5061 }
5062 if (cd->names_found >= MAX_NAME_COUNT)
5063 {
5064 *errorcodeptr = ERR49;
5065 goto FAILED;
5066 }
5067 if (namelen + 3 > cd->name_entry_size)
5068 {
5069 cd->name_entry_size = namelen + 3;
5070 if (namelen > MAX_NAME_SIZE)
5071 {
5072 *errorcodeptr = ERR48;
5073 goto FAILED;
5074 }
5075 }
5076 }
5077
5078 /* In the real compile, create the entry in the table, maintaining
5079 alphabetical order. Duplicate names for different numbers are
5080 permitted only if PCRE_DUPNAMES is set. Duplicate names for the same
5081 number are always OK. (An existing number can be re-used if (?|
5082 appears in the pattern.) In either event, a duplicate name results in
5083 a duplicate entry in the table, even if the number is the same. This
5084 is because the number of names, and hence the table size, is computed
5085 in the pre-compile, and it affects various numbers and pointers which
5086 would all have to be modified, and the compiled code moved down, if
5087 duplicates with the same number were omitted from the table. This
5088 doesn't seem worth the hassle. However, *different* names for the
5089 same number are not permitted. */
5090
5091 else
5092 {
5093 BOOL dupname = FALSE;
5094 slot = cd->name_table;
5095
5096 for (i = 0; i < cd->names_found; i++)
5097 {
5098 int crc = memcmp(name, slot+2, namelen);
5099 if (crc == 0)
5100 {
5101 if (slot[2+namelen] == 0)
5102 {
5103 if (GET2(slot, 0) != cd->bracount + 1 &&
5104 (options & PCRE_DUPNAMES) == 0)
5105 {
5106 *errorcodeptr = ERR43;
5107 goto FAILED;
5108 }
5109 else dupname = TRUE;
5110 }
5111 else crc = -1; /* Current name is a substring */
5112 }
5113
5114 /* Make space in the table and break the loop for an earlier
5115 name. For a duplicate or later name, carry on. We do this for
5116 duplicates so that in the simple case (when ?(| is not used) they
5117 are in order of their numbers. */
5118
5119 if (crc < 0)
5120 {
5121 memmove(slot + cd->name_entry_size, slot,
5122 (cd->names_found - i) * cd->name_entry_size);
5123 break;
5124 }
5125
5126 /* Continue the loop for a later or duplicate name */
5127
5128 slot += cd->name_entry_size;
5129 }
5130
5131 /* For non-duplicate names, check for a duplicate number before
5132 adding the new name. */
5133
5134 if (!dupname)
5135 {
5136 uschar *cslot = cd->name_table;
5137 for (i = 0; i < cd->names_found; i++)
5138 {
5139 if (cslot != slot)
5140 {
5141 if (GET2(cslot, 0) == cd->bracount + 1)
5142 {
5143 *errorcodeptr = ERR65;
5144 goto FAILED;
5145 }
5146 }
5147 else i--;
5148 cslot += cd->name_entry_size;
5149 }
5150 }
5151
5152 PUT2(slot, 0, cd->bracount + 1);
5153 memcpy(slot + 2, name, namelen);
5154 slot[2+namelen] = 0;
5155 }
5156 }
5157
5158 /* In both pre-compile and compile, count the number of names we've
5159 encountered. */
5160
5161 cd->names_found++;
5162 ptr++; /* Move past > or ' */
5163 goto NUMBERED_GROUP;
5164
5165
5166 /* ------------------------------------------------------------ */
5167 case CHAR_AMPERSAND: /* Perl recursion/subroutine syntax */
5168 terminator = CHAR_RIGHT_PARENTHESIS;
5169 is_recurse = TRUE;
5170 /* Fall through */
5171
5172 /* We come here from the Python syntax above that handles both
5173 references (?P=name) and recursion (?P>name), as well as falling
5174 through from the Perl recursion syntax (?&name). We also come here from
5175 the Perl \k<name> or \k'name' back reference syntax and the \k{name}
5176 .NET syntax, and the Oniguruma \g<...> and \g'...' subroutine syntax. */
5177
5178 NAMED_REF_OR_RECURSE:
5179 name = ++ptr;
5180 while ((cd->ctypes[*ptr] & ctype_word) != 0) ptr++;
5181 namelen = (int)(ptr - name);
5182
5183 /* In the pre-compile phase, do a syntax check and set a dummy
5184 reference number. */
5185
5186 if (lengthptr != NULL)
5187 {
5188 if (namelen == 0)
5189 {
5190 *errorcodeptr = ERR62;
5191 goto FAILED;
5192 }
5193 if (*ptr != terminator)
5194 {
5195 *errorcodeptr = ERR42;
5196 goto FAILED;
5197 }
5198 if (namelen > MAX_NAME_SIZE)
5199 {
5200 *errorcodeptr = ERR48;
5201 goto FAILED;
5202 }
5203 recno = 0;
5204 }
5205
5206 /* In the real compile, seek the name in the table. We check the name
5207 first, and then check that we have reached the end of the name in the
5208 table. That way, if the name that is longer than any in the table,
5209 the comparison will fail without reading beyond the table entry. */
5210
5211 else
5212 {
5213 slot = cd->name_table;
5214 for (i = 0; i < cd->names_found; i++)
5215 {
5216 if (strncmp((char *)name, (char *)slot+2, namelen) == 0 &&
5217 slot[2+namelen] == 0)
5218 break;
5219 slot += cd->name_entry_size;
5220 }
5221
5222 if (i < cd->names_found) /* Back reference */
5223 {
5224 recno = GET2(slot, 0);
5225 }
5226 else if ((recno = /* Forward back reference */
5227 find_parens(cd, name, namelen,
5228 (options & PCRE_EXTENDED) != 0)) <= 0)
5229 {
5230 *errorcodeptr = ERR15;
5231 goto FAILED;
5232 }
5233 }
5234
5235 /* In both phases, we can now go to the code than handles numerical
5236 recursion or backreferences. */
5237
5238 if (is_recurse) goto HANDLE_RECURSION;
5239 else goto HANDLE_REFERENCE;
5240
5241
5242 /* ------------------------------------------------------------ */
5243 case CHAR_R: /* Recursion */
5244 ptr++; /* Same as (?0) */
5245 /* Fall through */
5246
5247
5248 /* ------------------------------------------------------------ */
5249 case CHAR_MINUS: case CHAR_PLUS: /* Recursion or subroutine */
5250 case CHAR_0: case CHAR_1: case CHAR_2: case CHAR_3: case CHAR_4:
5251 case CHAR_5: case CHAR_6: case CHAR_7: case CHAR_8: case CHAR_9:
5252 {
5253 const uschar *called;
5254 terminator = CHAR_RIGHT_PARENTHESIS;
5255
5256 /* Come here from the \g<...> and \g'...' code (Oniguruma
5257 compatibility). However, the syntax has been checked to ensure that
5258 the ... are a (signed) number, so that neither ERR63 nor ERR29 will
5259 be called on this path, nor with the jump to OTHER_CHAR_AFTER_QUERY
5260 ever be taken. */
5261
5262 HANDLE_NUMERICAL_RECURSION:
5263
5264 if ((refsign = *ptr) == CHAR_PLUS)
5265 {
5266 ptr++;
5267 if ((digitab[*ptr] & ctype_digit) == 0)
5268 {
5269 *errorcodeptr = ERR63;
5270 goto FAILED;
5271 }
5272 }
5273 else if (refsign == CHAR_MINUS)
5274 {
5275 if ((digitab[ptr[1]] & ctype_digit) == 0)
5276 goto OTHER_CHAR_AFTER_QUERY;
5277 ptr++;
5278 }
5279
5280 recno = 0;
5281 while((digitab[*ptr] & ctype_digit) != 0)
5282 recno = recno * 10 + *ptr++ - CHAR_0;
5283
5284 if (*ptr != terminator)
5285 {
5286 *errorcodeptr = ERR29;
5287 goto FAILED;
5288 }
5289
5290 if (refsign == CHAR_MINUS)
5291 {
5292 if (recno == 0)
5293 {
5294 *errorcodeptr = ERR58;
5295 goto FAILED;
5296 }
5297 recno = cd->bracount - recno + 1;
5298 if (recno <= 0)
5299 {
5300 *errorcodeptr = ERR15;
5301 goto FAILED;
5302 }
5303 }
5304 else if (refsign == CHAR_PLUS)
5305 {
5306 if (recno == 0)
5307 {
5308 *errorcodeptr = ERR58;
5309 goto FAILED;
5310 }
5311 recno += cd->bracount;
5312 }
5313
5314 /* Come here from code above that handles a named recursion */
5315
5316 HANDLE_RECURSION:
5317
5318 previous = code;
5319 called = cd->start_code;
5320
5321 /* When we are actually compiling, find the bracket that is being
5322 referenced. Temporarily end the regex in case it doesn't exist before
5323 this point. If we end up with a forward reference, first check that
5324 the bracket does occur later so we can give the error (and position)
5325 now. Then remember this forward reference in the workspace so it can
5326 be filled in at the end. */
5327
5328 if (lengthptr == NULL)
5329 {
5330 *code = OP_END;
5331 if (recno != 0)
5332 called = _pcre_find_bracket(cd->start_code, utf8, recno);
5333
5334 /* Forward reference */
5335
5336 if (called == NULL)
5337 {
5338 if (find_parens(cd, NULL, recno,
5339 (options & PCRE_EXTENDED) != 0) < 0)
5340 {
5341 *errorcodeptr = ERR15;
5342 goto FAILED;
5343 }
5344
5345 /* Fudge the value of "called" so that when it is inserted as an
5346 offset below, what it actually inserted is the reference number
5347 of the group. */
5348
5349 called = cd->start_code + recno;
5350 PUTINC(cd->hwm, 0, (int)(code + 2 + LINK_SIZE - cd->start_code));
5351 }
5352
5353 /* If not a forward reference, and the subpattern is still open,
5354 this is a recursive call. We check to see if this is a left
5355 recursion that could loop for ever, and diagnose that case. */
5356
5357 else if (GET(called, 1) == 0 &&
5358 could_be_empty(called, code, bcptr, utf8, cd))
5359 {
5360 *errorcodeptr = ERR40;
5361 goto FAILED;
5362 }
5363 }
5364
5365 /* Insert the recursion/subroutine item, automatically wrapped inside
5366 "once" brackets. Set up a "previous group" length so that a
5367 subsequent quantifier will work. */
5368
5369 *code = OP_ONCE;
5370 PUT(code, 1, 2 + 2*LINK_SIZE);
5371 code += 1 + LINK_SIZE;
5372
5373 *code = OP_RECURSE;
5374 PUT(code, 1, (int)(called - cd->start_code));
5375 code += 1 + LINK_SIZE;
5376
5377 *code = OP_KET;
5378 PUT(code, 1, 2 + 2*LINK_SIZE);
5379 code += 1 + LINK_SIZE;
5380
5381 length_prevgroup = 3 + 3*LINK_SIZE;
5382 }
5383
5384 /* Can't determine a first byte now */
5385
5386 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
5387 continue;
5388
5389
5390 /* ------------------------------------------------------------ */
5391 default: /* Other characters: check option setting */
5392 OTHER_CHAR_AFTER_QUERY:
5393 set = unset = 0;
5394 optset = &set;
5395
5396 while (*ptr != CHAR_RIGHT_PARENTHESIS && *ptr != CHAR_COLON)
5397 {
5398 switch (*ptr++)
5399 {
5400 case CHAR_MINUS: optset = &unset; break;
5401
5402 case CHAR_J: /* Record that it changed in the external options */
5403 *optset |= PCRE_DUPNAMES;
5404 cd->external_flags |= PCRE_JCHANGED;
5405 break;
5406
5407 case CHAR_i: *optset |= PCRE_CASELESS; break;
5408 case CHAR_m: *optset |= PCRE_MULTILINE; break;
5409 case CHAR_s: *optset |= PCRE_DOTALL; break;
5410 case CHAR_x: *optset |= PCRE_EXTENDED; break;
5411 case CHAR_U: *optset |= PCRE_UNGREEDY; break;
5412 case CHAR_X: *optset |= PCRE_EXTRA; break;
5413
5414 default: *errorcodeptr = ERR12;
5415 ptr--; /* Correct the offset */
5416 goto FAILED;
5417 }
5418 }
5419
5420 /* Set up the changed option bits, but don't change anything yet. */
5421
5422 newoptions = (options | set) & (~unset);
5423
5424 /* If the options ended with ')' this is not the start of a nested
5425 group with option changes, so the options change at this level. If this
5426 item is right at the start of the pattern, the options can be
5427 abstracted and made external in the pre-compile phase, and ignored in
5428 the compile phase. This can be helpful when matching -- for instance in
5429 caseless checking of required bytes.
5430
5431 If the code pointer is not (cd->start_code + 1 + LINK_SIZE), we are
5432 definitely *not* at the start of the pattern because something has been
5433 compiled. In the pre-compile phase, however, the code pointer can have
5434 that value after the start, because it gets reset as code is discarded
5435 during the pre-compile. However, this can happen only at top level - if
5436 we are within parentheses, the starting BRA will still be present. At
5437 any parenthesis level, the length value can be used to test if anything
5438 has been compiled at that level. Thus, a test for both these conditions
5439 is necessary to ensure we correctly detect the start of the pattern in
5440 both phases.
5441
5442 If we are not at the pattern start, compile code to change the ims
5443 options if this setting actually changes any of them, and reset the
5444 greedy defaults and the case value for firstbyte and reqbyte. */
5445
5446 if (*ptr == CHAR_RIGHT_PARENTHESIS)
5447 {
5448 if (code == cd->start_code + 1 + LINK_SIZE &&
5449 (lengthptr == NULL || *lengthptr == 2 + 2*LINK_SIZE))
5450 {
5451 cd->external_options = newoptions;
5452 }
5453 else
5454 {
5455 if ((options & PCRE_IMS) != (newoptions & PCRE_IMS))
5456 {
5457 *code++ = OP_OPT;
5458 *code++ = newoptions & PCRE_IMS;
5459 }
5460 greedy_default = ((newoptions & PCRE_UNGREEDY) != 0);
5461 greedy_non_default = greedy_default ^ 1;
5462 req_caseopt = ((newoptions & PCRE_CASELESS) != 0)? REQ_CASELESS : 0;
5463 }
5464
5465 /* Change options at this level, and pass them back for use
5466 in subsequent branches. When not at the start of the pattern, this
5467 information is also necessary so that a resetting item can be
5468 compiled at the end of a group (if we are in a group). */
5469
5470 *optionsptr = options = newoptions;
5471 previous = NULL; /* This item can't be repeated */
5472 continue; /* It is complete */
5473 }
5474
5475 /* If the options ended with ':' we are heading into a nested group
5476 with possible change of options. Such groups are non-capturing and are
5477 not assertions of any kind. All we need to do is skip over the ':';
5478 the newoptions value is handled below. */
5479
5480 bravalue = OP_BRA;
5481 ptr++;
5482 } /* End of switch for character following (? */
5483 } /* End of (? handling */
5484
5485 /* Opening parenthesis not followed by '*' or '?'. If PCRE_NO_AUTO_CAPTURE
5486 is set, all unadorned brackets become non-capturing and behave like (?:...)
5487 brackets. */
5488
5489 else if ((options & PCRE_NO_AUTO_CAPTURE) != 0)
5490 {
5491 bravalue = OP_BRA;
5492 }
5493
5494 /* Else we have a capturing group. */
5495
5496 else
5497 {
5498 NUMBERED_GROUP:
5499 cd->bracount += 1;
5500 PUT2(code, 1+LINK_SIZE, cd->bracount);
5501 skipbytes = 2;
5502 }
5503
5504 /* Process nested bracketed regex. Assertions may not be repeated, but
5505 other kinds can be. All their opcodes are >= OP_ONCE. We copy code into a
5506 non-register variable in order to be able to pass its address because some
5507 compilers complain otherwise. Pass in a new setting for the ims options if
5508 they have changed. */
5509
5510 previous = (bravalue >= OP_ONCE)? code : NULL;
5511 *code = bravalue;
5512 tempcode = code;
5513 tempreqvary = cd->req_varyopt; /* Save value before bracket */
5514 length_prevgroup = 0; /* Initialize for pre-compile phase */
5515
5516 if (!compile_regex(
5517 newoptions, /* The complete new option state */
5518 options & PCRE_IMS, /* The previous ims option state */
5519 &tempcode, /* Where to put code (updated) */
5520 &ptr, /* Input pointer (updated) */
5521 errorcodeptr, /* Where to put an error message */
5522 (bravalue == OP_ASSERTBACK ||
5523 bravalue == OP_ASSERTBACK_NOT), /* TRUE if back assert */
5524 reset_bracount, /* True if (?| group */
5525 skipbytes, /* Skip over bracket number */
5526 &subfirstbyte, /* For possible first char */
5527 &subreqbyte, /* For possible last char */
5528 bcptr, /* Current branch chain */
5529 cd, /* Tables block */
5530 (lengthptr == NULL)? NULL : /* Actual compile phase */
5531 &length_prevgroup /* Pre-compile phase */
5532 ))
5533 goto FAILED;
5534
5535 /* At the end of compiling, code is still pointing to the start of the
5536 group, while tempcode has been updated to point past the end of the group
5537 and any option resetting that may follow it. The pattern pointer (ptr)
5538 is on the bracket. */
5539
5540 /* If this is a conditional bracket, check that there are no more than
5541 two branches in the group, or just one if it's a DEFINE group. We do this
5542 in the real compile phase, not in the pre-pass, where the whole group may
5543 not be available. */
5544
5545 if (bravalue == OP_COND && lengthptr == NULL)
5546 {
5547 uschar *tc = code;
5548 int condcount = 0;
5549
5550 do {
5551 condcount++;
5552 tc += GET(tc,1);
5553 }
5554 while (*tc != OP_KET);
5555
5556 /* A DEFINE group is never obeyed inline (the "condition" is always
5557 false). It must have only one branch. */
5558
5559 if (code[LINK_SIZE+1] == OP_DEF)
5560 {
5561 if (condcount > 1)
5562 {
5563 *errorcodeptr = ERR54;
5564 goto FAILED;
5565 }
5566 bravalue = OP_DEF; /* Just a flag to suppress char handling below */
5567 }
5568
5569 /* A "normal" conditional group. If there is just one branch, we must not
5570 make use of its firstbyte or reqbyte, because this is equivalent to an
5571 empty second branch. */
5572
5573 else
5574 {
5575 if (condcount > 2)
5576 {
5577 *errorcodeptr = ERR27;
5578 goto FAILED;
5579 }
5580 if (condcount == 1) subfirstbyte = subreqbyte = REQ_NONE;
5581 }
5582 }
5583
5584 /* Error if hit end of pattern */
5585
5586 if (*ptr != CHAR_RIGHT_PARENTHESIS)
5587 {
5588 *errorcodeptr = ERR14;
5589 goto FAILED;
5590 }
5591
5592 /* In the pre-compile phase, update the length by the length of the group,
5593 less the brackets at either end. Then reduce the compiled code to just a
5594 set of non-capturing brackets so that it doesn't use much memory if it is
5595 duplicated by a quantifier.*/
5596
5597 if (lengthptr != NULL)
5598 {
5599 if (OFLOW_MAX - *lengthptr < length_prevgroup - 2 - 2*LINK_SIZE)
5600 {
5601 *errorcodeptr = ERR20;
5602 goto FAILED;
5603 }
5604 *lengthptr += length_prevgroup - 2 - 2*LINK_SIZE;
5605 *code++ = OP_BRA;
5606 PUTINC(code, 0, 1 + LINK_SIZE);
5607 *code++ = OP_KET;
5608 PUTINC(code, 0, 1 + LINK_SIZE);
5609 break; /* No need to waste time with special character handling */
5610 }
5611
5612 /* Otherwise update the main code pointer to the end of the group. */
5613
5614 code = tempcode;
5615
5616 /* For a DEFINE group, required and first character settings are not
5617 relevant. */
5618
5619 if (bravalue == OP_DEF) break;
5620
5621 /* Handle updating of the required and first characters for other types of
5622 group. Update for normal brackets of all kinds, and conditions with two
5623 branches (see code above). If the bracket is followed by a quantifier with
5624 zero repeat, we have to back off. Hence the definition of zeroreqbyte and
5625 zerofirstbyte outside the main loop so that they can be accessed for the
5626 back off. */
5627
5628 zeroreqbyte = reqbyte;
5629 zerofirstbyte = firstbyte;
5630 groupsetfirstbyte = FALSE;
5631
5632 if (bravalue >= OP_ONCE)
5633 {
5634 /* If we have not yet set a firstbyte in this branch, take it from the
5635 subpattern, remembering that it was set here so that a repeat of more
5636 than one can replicate it as reqbyte if necessary. If the subpattern has
5637 no firstbyte, set "none" for the whole branch. In both cases, a zero
5638 repeat forces firstbyte to "none". */
5639
5640 if (firstbyte == REQ_UNSET)
5641 {
5642 if (subfirstbyte >= 0)
5643 {
5644 firstbyte = subfirstbyte;
5645 groupsetfirstbyte = TRUE;
5646 }
5647 else firstbyte = REQ_NONE;
5648 zerofirstbyte = REQ_NONE;
5649 }
5650
5651 /* If firstbyte was previously set, convert the subpattern's firstbyte
5652 into reqbyte if there wasn't one, using the vary flag that was in
5653 existence beforehand. */
5654
5655 else if (subfirstbyte >= 0 && subreqbyte < 0)
5656 subreqbyte = subfirstbyte | tempreqvary;
5657
5658 /* If the subpattern set a required byte (or set a first byte that isn't
5659 really the first byte - see above), set it. */
5660
5661 if (subreqbyte >= 0) reqbyte = subreqbyte;
5662 }
5663
5664 /* For a forward assertion, we take the reqbyte, if set. This can be
5665 helpful if the pattern that follows the assertion doesn't set a different
5666 char. For example, it's useful for /(?=abcde).+/. We can't set firstbyte
5667 for an assertion, however because it leads to incorrect effect for patterns
5668 such as /(?=a)a.+/ when the "real" "a" would then become a reqbyte instead
5669 of a firstbyte. This is overcome by a scan at the end if there's no
5670 firstbyte, looking for an asserted first char. */
5671
5672 else if (bravalue == OP_ASSERT && subreqbyte >= 0) reqbyte = subreqbyte;
5673 break; /* End of processing '(' */
5674
5675
5676 /* ===================================================================*/
5677 /* Handle metasequences introduced by \. For ones like \d, the ESC_ values
5678 are arranged to be the negation of the corresponding OP_values in the
5679 default case when PCRE_UCP is not set. For the back references, the values
5680 are ESC_REF plus the reference number. Only back references and those types
5681 that consume a character may be repeated. We can test for values between
5682 ESC_b and ESC_Z for the latter; this may have to change if any new ones are
5683 ever created. */
5684
5685 case CHAR_BACKSLASH:
5686 tempptr = ptr;
5687 c = check_escape(&ptr, errorcodeptr, cd->bracount, options, FALSE);
5688 if (*errorcodeptr != 0) goto FAILED;
5689
5690 if (c < 0)
5691 {
5692 if (-c == ESC_Q) /* Handle start of quoted string */
5693 {
5694 if (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
5695 ptr += 2; /* avoid empty string */
5696 else inescq = TRUE;
5697 continue;
5698 }
5699
5700 if (-c == ESC_E) continue; /* Perl ignores an orphan \E */
5701
5702 /* For metasequences that actually match a character, we disable the
5703 setting of a first character if it hasn't already been set. */
5704
5705 if (firstbyte == REQ_UNSET && -c > ESC_b && -c < ESC_Z)
5706 firstbyte = REQ_NONE;
5707
5708 /* Set values to reset to if this is followed by a zero repeat. */
5709
5710 zerofirstbyte = firstbyte;
5711 zeroreqbyte = reqbyte;
5712
5713 /* \g<name> or \g'name' is a subroutine call by name and \g<n> or \g'n'
5714 is a subroutine call by number (Oniguruma syntax). In fact, the value
5715 -ESC_g is returned only for these cases. So we don't need to check for <
5716 or ' if the value is -ESC_g. For the Perl syntax \g{n} the value is
5717 -ESC_REF+n, and for the Perl syntax \g{name} the result is -ESC_k (as
5718 that is a synonym for a named back reference). */
5719
5720 if (-c == ESC_g)
5721 {
5722 const uschar *p;
5723 save_hwm = cd->hwm; /* Normally this is set when '(' is read */
5724 terminator = (*(++ptr) == CHAR_LESS_THAN_SIGN)?
5725 CHAR_GREATER_THAN_SIGN : CHAR_APOSTROPHE;
5726
5727 /* These two statements stop the compiler for warning about possibly
5728 unset variables caused by the jump to HANDLE_NUMERICAL_RECURSION. In
5729 fact, because we actually check for a number below, the paths that
5730 would actually be in error are never taken. */
5731
5732 skipbytes = 0;
5733 reset_bracount = FALSE;
5734
5735 /* Test for a name */
5736
5737 if (ptr[1] != CHAR_PLUS && ptr[1] != CHAR_MINUS)
5738 {
5739 BOOL isnumber = TRUE;
5740 for (p = ptr + 1; *p != 0 && *p != terminator; p++)
5741 {
5742 if ((cd->ctypes[*p] & ctype_digit) == 0) isnumber = FALSE;
5743 if ((cd->ctypes[*p] & ctype_word) == 0) break;
5744 }
5745 if (*p != terminator)
5746 {
5747 *errorcodeptr = ERR57;
5748 break;
5749 }
5750 if (isnumber)
5751 {
5752 ptr++;
5753 goto HANDLE_NUMERICAL_RECURSION;
5754 }
5755 is_recurse = TRUE;
5756 goto NAMED_REF_OR_RECURSE;
5757 }
5758
5759 /* Test a signed number in angle brackets or quotes. */
5760
5761 p = ptr + 2;
5762 while ((digitab[*p] & ctype_digit) != 0) p++;
5763 if (*p != terminator)
5764 {
5765 *errorcodeptr = ERR57;
5766 break;
5767 }
5768 ptr++;
5769 goto HANDLE_NUMERICAL_RECURSION;
5770 }
5771
5772 /* \k<name> or \k'name' is a back reference by name (Perl syntax).
5773 We also support \k{name} (.NET syntax) */
5774
5775 if (-c == ESC_k && (ptr[1] == CHAR_LESS_THAN_SIGN ||
5776 ptr[1] == CHAR_APOSTROPHE || ptr[1] == CHAR_LEFT_CURLY_BRACKET))
5777 {
5778 is_recurse = FALSE;
5779 terminator = (*(++ptr) == CHAR_LESS_THAN_SIGN)?
5780 CHAR_GREATER_THAN_SIGN : (*ptr == CHAR_APOSTROPHE)?
5781 CHAR_APOSTROPHE : CHAR_RIGHT_CURLY_BRACKET;
5782 goto NAMED_REF_OR_RECURSE;
5783 }
5784
5785 /* Back references are handled specially; must disable firstbyte if
5786 not set to cope with cases like (?=(\w+))\1: which would otherwise set
5787 ':' later. */
5788
5789 if (-c >= ESC_REF)
5790 {
5791 open_capitem *oc;
5792 recno = -c - ESC_REF;
5793
5794 HANDLE_REFERENCE: /* Come here from named backref handling */
5795 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
5796 previous = code;
5797 *code++ = OP_REF;
5798 PUT2INC(code, 0, recno);
5799 cd->backref_map |= (recno < 32)? (1 << recno) : 1;
5800 if (recno > cd->top_backref) cd->top_backref = recno;
5801
5802 /* Check to see if this back reference is recursive, that it, it
5803 is inside the group that it references. A flag is set so that the
5804 group can be made atomic. */
5805
5806 for (oc = cd->open_caps; oc != NULL; oc = oc->next)
5807 {
5808 if (oc->number == recno)
5809 {
5810 oc->flag = TRUE;
5811 break;
5812 }
5813 }
5814 }
5815
5816 /* So are Unicode property matches, if supported. */
5817
5818 #ifdef SUPPORT_UCP
5819 else if (-c == ESC_P || -c == ESC_p)
5820 {
5821 BOOL negated;
5822 int pdata;
5823 int ptype = get_ucp(&ptr, &negated, &pdata, errorcodeptr);
5824 if (ptype < 0) goto FAILED;
5825 previous = code;
5826 *code++ = ((-c == ESC_p) != negated)? OP_PROP : OP_NOTPROP;
5827 *code++ = ptype;
5828 *code++ = pdata;
5829 }
5830 #else
5831
5832 /* If Unicode properties are not supported, \X, \P, and \p are not
5833 allowed. */
5834
5835 else if (-c == ESC_X || -c == ESC_P || -c == ESC_p)
5836 {
5837 *errorcodeptr = ERR45;
5838 goto FAILED;
5839 }
5840 #endif
5841
5842 /* For the rest (including \X when Unicode properties are supported), we
5843 can obtain the OP value by negating the escape value in the default
5844 situation when PCRE_UCP is not set. When it *is* set, we substitute
5845 Unicode property tests. */
5846
5847 else
5848 {
5849 #ifdef SUPPORT_UCP
5850 if (-c >= ESC_DU && -c <= ESC_wu)
5851 {
5852 nestptr = ptr + 1; /* Where to resume */
5853 ptr = substitutes[-c - ESC_DU] - 1; /* Just before substitute */
5854 }
5855 else
5856 #endif
5857 {
5858 previous = (-c > ESC_b && -c < ESC_Z)? code : NULL;
5859 *code++ = -c;
5860 }
5861 }
5862 continue;
5863 }
5864
5865 /* We have a data character whose value is in c. In UTF-8 mode it may have
5866 a value > 127. We set its representation in the length/buffer, and then
5867 handle it as a data character. */
5868
5869 #ifdef SUPPORT_UTF8
5870 if (utf8 && c > 127)
5871 mclength = _pcre_ord2utf8(c, mcbuffer);
5872 else
5873 #endif
5874
5875 {
5876 mcbuffer[0] = c;
5877 mclength = 1;
5878 }
5879 goto ONE_CHAR;
5880
5881
5882 /* ===================================================================*/
5883 /* Handle a literal character. It is guaranteed not to be whitespace or #
5884 when the extended flag is set. If we are in UTF-8 mode, it may be a
5885 multi-byte literal character. */
5886
5887 default:
5888 NORMAL_CHAR:
5889 mclength = 1;
5890 mcbuffer[0] = c;
5891
5892 #ifdef SUPPORT_UTF8
5893 if (utf8 && c >= 0xc0)
5894 {
5895 while ((ptr[1] & 0xc0) == 0x80)
5896 mcbuffer[mclength++] = *(++ptr);
5897 }
5898 #endif
5899
5900 /* At this point we have the character's bytes in mcbuffer, and the length
5901 in mclength. When not in UTF-8 mode, the length is always 1. */
5902
5903 ONE_CHAR:
5904 previous = code;
5905 *code++ = ((options & PCRE_CASELESS) != 0)? OP_CHARNC : OP_CHAR;
5906 for (c = 0; c < mclength; c++) *code++ = mcbuffer[c];
5907
5908 /* Remember if \r or \n were seen */
5909
5910 if (mcbuffer[0] == CHAR_CR || mcbuffer[0] == CHAR_NL)
5911 cd->external_flags |= PCRE_HASCRORLF;
5912
5913 /* Set the first and required bytes appropriately. If no previous first
5914 byte, set it from this character, but revert to none on a zero repeat.
5915 Otherwise, leave the firstbyte value alone, and don't change it on a zero
5916 repeat. */
5917
5918 if (firstbyte == REQ_UNSET)
5919 {
5920 zerofirstbyte = REQ_NONE;
5921 zeroreqbyte = reqbyte;
5922
5923 /* If the character is more than one byte long, we can set firstbyte
5924 only if it is not to be matched caselessly. */
5925
5926 if (mclength == 1 || req_caseopt == 0)
5927 {
5928 firstbyte = mcbuffer[0] | req_caseopt;
5929 if (mclength != 1) reqbyte = code[-1] | cd->req_varyopt;
5930 }
5931 else firstbyte = reqbyte = REQ_NONE;
5932 }
5933
5934 /* firstbyte was previously set; we can set reqbyte only the length is
5935 1 or the matching is caseful. */
5936
5937 else
5938 {
5939 zerofirstbyte = firstbyte;
5940 zeroreqbyte = reqbyte;
5941 if (mclength == 1 || req_caseopt == 0)
5942 reqbyte = code[-1] | req_caseopt | cd->req_varyopt;
5943 }
5944
5945 break; /* End of literal character handling */
5946 }
5947 } /* end of big loop */
5948
5949
5950 /* Control never reaches here by falling through, only by a goto for all the
5951 error states. Pass back the position in the pattern so that it can be displayed
5952 to the user for diagnosing the error. */
5953
5954 FAILED:
5955 *ptrptr = ptr;
5956 return FALSE;
5957 }
5958
5959
5960
5961
5962 /*************************************************
5963 * Compile sequence of alternatives *
5964 *************************************************/
5965
5966 /* On entry, ptr is pointing past the bracket character, but on return it
5967 points to the closing bracket, or vertical bar, or end of string. The code
5968 variable is pointing at the byte into which the BRA operator has been stored.
5969 If the ims options are changed at the start (for a (?ims: group) or during any
5970 branch, we need to insert an OP_OPT item at the start of every following branch
5971 to ensure they get set correctly at run time, and also pass the new options
5972 into every subsequent branch compile.
5973
5974 This function is used during the pre-compile phase when we are trying to find
5975 out the amount of memory needed, as well as during the real compile phase. The
5976 value of lengthptr distinguishes the two phases.
5977
5978 Arguments:
5979 options option bits, including any changes for this subpattern
5980 oldims previous settings of ims option bits
5981 codeptr -> the address of the current code pointer
5982 ptrptr -> the address of the current pattern pointer
5983 errorcodeptr -> pointer to error code variable
5984 lookbehind TRUE if this is a lookbehind assertion
5985 reset_bracount TRUE to reset the count for each branch
5986 skipbytes skip this many bytes at start (for brackets and OP_COND)
5987 firstbyteptr place to put the first required character, or a negative number
5988 reqbyteptr place to put the last required character, or a negative number
5989 bcptr pointer to the chain of currently open branches
5990 cd points to the data block with tables pointers etc.
5991 lengthptr NULL during the real compile phase
5992 points to length accumulator during pre-compile phase
5993
5994 Returns: TRUE on success
5995 */
5996
5997 static BOOL
5998 compile_regex(int options, int oldims, uschar **codeptr, const uschar **ptrptr,
5999 int *errorcodeptr, BOOL lookbehind, BOOL reset_bracount, int skipbytes,
6000 int *firstbyteptr, int *reqbyteptr, branch_chain *bcptr, compile_data *cd,
6001 int *lengthptr)
6002 {
6003 const uschar *ptr = *ptrptr;
6004 uschar *code = *codeptr;
6005 uschar *last_branch = code;
6006 uschar *start_bracket = code;
6007 uschar *reverse_count = NULL;
6008 open_capitem capitem;
6009 int capnumber = 0;
6010 int firstbyte, reqbyte;
6011 int branchfirstbyte, branchreqbyte;
6012 int length;
6013 int orig_bracount;
6014 int max_bracount;
6015 int old_external_options = cd->external_options;
6016 branch_chain bc;
6017
6018 bc.outer = bcptr;
6019 bc.current_branch = code;
6020
6021 firstbyte = reqbyte = REQ_UNSET;
6022
6023 /* Accumulate the length for use in the pre-compile phase. Start with the
6024 length of the BRA and KET and any extra bytes that are required at the
6025 beginning. We accumulate in a local variable to save frequent testing of
6026 lenthptr for NULL. We cannot do this by looking at the value of code at the
6027 start and end of each alternative, because compiled items are discarded during
6028 the pre-compile phase so that the work space is not exceeded. */
6029
6030 length = 2 + 2*LINK_SIZE + skipbytes;
6031
6032 /* WARNING: If the above line is changed for any reason, you must also change
6033 the code that abstracts option settings at the start of the pattern and makes
6034 them global. It tests the value of length for (2 + 2*LINK_SIZE) in the
6035 pre-compile phase to find out whether anything has yet been compiled or not. */
6036
6037 /* If this is a capturing subpattern, add to the chain of open capturing items
6038 so that we can detect them if (*ACCEPT) is encountered. This is also used to
6039 detect groups that contain recursive back references to themselves. */
6040
6041 if (*code == OP_CBRA)
6042 {
6043 capnumber = GET2(code, 1 + LINK_SIZE);
6044 capitem.number = capnumber;
6045 capitem.next = cd->open_caps;
6046 capitem.flag = FALSE;
6047 cd->open_caps = &capitem;
6048 }
6049
6050 /* Offset is set zero to mark that this bracket is still open */
6051
6052 PUT(code, 1, 0);
6053 code += 1 + LINK_SIZE + skipbytes;
6054
6055 /* Loop for each alternative branch */
6056
6057 orig_bracount = max_bracount = cd->bracount;
6058 for (;;)
6059 {
6060 /* For a (?| group, reset the capturing bracket count so that each branch
6061 uses the same numbers. */
6062
6063 if (reset_bracount) cd->bracount = orig_bracount;
6064
6065 /* Handle a change of ims options at the start of the branch */
6066
6067 if ((options & PCRE_IMS) != oldims)
6068 {
6069 *code++ = OP_OPT;
6070 *code++ = options & PCRE_IMS;
6071 length += 2;
6072 }
6073
6074 /* Set up dummy OP_REVERSE if lookbehind assertion */
6075
6076 if (lookbehind)
6077 {
6078 *code++ = OP_REVERSE;
6079 reverse_count = code;
6080 PUTINC(code, 0, 0);
6081 length += 1 + LINK_SIZE;
6082 }
6083
6084 /* Now compile the branch; in the pre-compile phase its length gets added
6085 into the length. */
6086
6087 if (!compile_branch(&options, &code, &ptr, errorcodeptr, &branchfirstbyte,
6088 &branchreqbyte, &bc, cd, (lengthptr == NULL)? NULL : &length))
6089 {
6090 *ptrptr = ptr;
6091 return FALSE;
6092 }
6093
6094 /* If the external options have changed during this branch, it means that we
6095 are at the top level, and a leading option setting has been encountered. We
6096 need to re-set the original option values to take account of this so that,
6097 during the pre-compile phase, we know to allow for a re-set at the start of
6098 subsequent branches. */
6099
6100 if (old_external_options != cd->external_options)
6101 oldims = cd->external_options & PCRE_IMS;
6102
6103 /* Keep the highest bracket count in case (?| was used and some branch
6104 has fewer than the rest. */
6105
6106 if (cd->bracount > max_bracount) max_bracount = cd->bracount;
6107
6108 /* In the real compile phase, there is some post-processing to be done. */
6109
6110 if (lengthptr == NULL)
6111 {
6112 /* If this is the first branch, the firstbyte and reqbyte values for the
6113 branch become the values for the regex. */
6114
6115 if (*last_branch != OP_ALT)
6116 {
6117 firstbyte = branchfirstbyte;
6118 reqbyte = branchreqbyte;
6119 }
6120
6121 /* If this is not the first branch, the first char and reqbyte have to
6122 match the values from all the previous branches, except that if the
6123 previous value for reqbyte didn't have REQ_VARY set, it can still match,
6124 and we set REQ_VARY for the regex. */
6125
6126 else
6127 {
6128 /* If we previously had a firstbyte, but it doesn't match the new branch,
6129 we have to abandon the firstbyte for the regex, but if there was
6130 previously no reqbyte, it takes on the value of the old firstbyte. */
6131
6132 if (firstbyte >= 0 && firstbyte != branchfirstbyte)
6133 {
6134 if (reqbyte < 0) reqbyte = firstbyte;
6135 firstbyte = REQ_NONE;
6136 }
6137
6138 /* If we (now or from before) have no firstbyte, a firstbyte from the
6139 branch becomes a reqbyte if there isn't a branch reqbyte. */
6140
6141 if (firstbyte < 0 && branchfirstbyte >= 0 && branchreqbyte < 0)
6142 branchreqbyte = branchfirstbyte;
6143
6144 /* Now ensure that the reqbytes match */
6145
6146 if ((reqbyte & ~REQ_VARY) != (branchreqbyte & ~REQ_VARY))
6147 reqbyte = REQ_NONE;
6148 else reqbyte |= branchreqbyte; /* To "or" REQ_VARY */
6149 }
6150
6151 /* If lookbehind, check that this branch matches a fixed-length string, and
6152 put the length into the OP_REVERSE item. Temporarily mark the end of the
6153 branch with OP_END. If the branch contains OP_RECURSE, the result is -3
6154 because there may be forward references that we can't check here. Set a
6155 flag to cause another lookbehind check at the end. Why not do it all at the
6156 end? Because common, erroneous checks are picked up here and the offset of
6157 the problem can be shown. */
6158
6159 if (lookbehind)
6160 {
6161 int fixed_length;
6162 *code = OP_END;
6163 fixed_length = find_fixedlength(last_branch, options, FALSE, cd);
6164 DPRINTF(("fixed length = %d\n", fixed_length));
6165 if (fixed_length == -3)
6166 {
6167 cd->check_lookbehind = TRUE;
6168 }
6169 else if (fixed_length < 0)
6170 {
6171 *errorcodeptr = (fixed_length == -2)? ERR36 : ERR25;
6172 *ptrptr = ptr;
6173 return FALSE;
6174 }
6175 else { PUT(reverse_count, 0, fixed_length); }
6176 }
6177 }
6178
6179 /* Reached end of expression, either ')' or end of pattern. In the real
6180 compile phase, go back through the alternative branches and reverse the chain
6181 of offsets, with the field in the BRA item now becoming an offset to the
6182 first alternative. If there are no alternatives, it points to the end of the
6183 group. The length in the terminating ket is always the length of the whole
6184 bracketed item. If any of the ims options were changed inside the group,
6185 compile a resetting op-code following, except at the very end of the pattern.
6186 Return leaving the pointer at the terminating char. */
6187
6188 if (*ptr != CHAR_VERTICAL_LINE)
6189 {
6190 if (lengthptr == NULL)
6191 {
6192 int branch_length = (int)(code - last_branch);
6193 do
6194 {
6195 int prev_length = GET(last_branch, 1);
6196 PUT(last_branch, 1, branch_length);
6197 branch_length = prev_length;
6198 last_branch -= branch_length;
6199 }
6200 while (branch_length > 0);
6201 }
6202
6203 /* Fill in the ket */
6204
6205 *code = OP_KET;
6206 PUT(code, 1, (int)(code - start_bracket));
6207 code += 1 + LINK_SIZE;
6208
6209 /* If it was a capturing subpattern, check to see if it contained any
6210 recursive back references. If so, we must wrap it in atomic brackets.
6211 In any event, remove the block from the chain. */
6212
6213 if (capnumber > 0)
6214 {
6215 if (cd->open_caps->flag)
6216 {
6217 memmove(start_bracket + 1 + LINK_SIZE, start_bracket,
6218 code - start_bracket);
6219 *start_bracket = OP_ONCE;
6220 code += 1 + LINK_SIZE;
6221 PUT(start_bracket, 1, (int)(code - start_bracket));
6222 *code = OP_KET;
6223 PUT(code, 1, (int)(code - start_bracket));
6224 code += 1 + LINK_SIZE;
6225 length += 2 + 2*LINK_SIZE;
6226 }
6227 cd->open_caps = cd->open_caps->next;
6228 }
6229
6230 /* Reset options if needed. */
6231
6232 if ((options & PCRE_IMS) != oldims && *ptr == CHAR_RIGHT_PARENTHESIS)
6233 {
6234 *code++ = OP_OPT;
6235 *code++ = oldims;
6236 length += 2;
6237 }
6238
6239 /* Retain the highest bracket number, in case resetting was used. */
6240
6241 cd->bracount = max_bracount;
6242
6243 /* Set values to pass back */
6244
6245 *codeptr = code;
6246 *ptrptr = ptr;
6247 *firstbyteptr = firstbyte;
6248 *reqbyteptr = reqbyte;
6249 if (lengthptr != NULL)
6250 {
6251 if (OFLOW_MAX - *lengthptr < length)
6252 {
6253 *errorcodeptr = ERR20;
6254 return FALSE;
6255 }
6256 *lengthptr += length;
6257 }
6258 return TRUE;
6259 }
6260
6261 /* Another branch follows. In the pre-compile phase, we can move the code
6262 pointer back to where it was for the start of the first branch. (That is,
6263 pretend that each branch is the only one.)
6264
6265 In the real compile phase, insert an ALT node. Its length field points back
6266 to the previous branch while the bracket remains open. At the end the chain
6267 is reversed. It's done like this so that the start of the bracket has a
6268 zero offset until it is closed, making it possible to detect recursion. */
6269
6270 if (lengthptr != NULL)
6271 {
6272 code = *codeptr + 1 + LINK_SIZE + skipbytes;
6273 length += 1 + LINK_SIZE;
6274 }
6275 else
6276 {
6277 *code = OP_ALT;
6278 PUT(code, 1, (int)(code - last_branch));
6279 bc.current_branch = last_branch = code;
6280 code += 1 + LINK_SIZE;
6281 }
6282
6283 ptr++;
6284 }
6285 /* Control never reaches here */
6286 }
6287
6288
6289
6290
6291 /*************************************************
6292 * Check for anchored expression *
6293 *************************************************/
6294
6295 /* Try to find out if this is an anchored regular expression. Consider each
6296 alternative branch. If they all start with OP_SOD or OP_CIRC, or with a bracket
6297 all of whose alternatives start with OP_SOD or OP_CIRC (recurse ad lib), then
6298 it's anchored. However, if this is a multiline pattern, then only OP_SOD
6299 counts, since OP_CIRC can match in the middle.
6300
6301 We can also consider a regex to be anchored if OP_SOM starts all its branches.
6302 This is the code for \G, which means "match at start of match position, taking
6303 into account the match offset".
6304
6305 A branch is also implicitly anchored if it starts with .* and DOTALL is set,
6306 because that will try the rest of the pattern at all possible matching points,
6307 so there is no point trying again.... er ....
6308
6309 .... except when the .* appears inside capturing parentheses, and there is a
6310 subsequent back reference to those parentheses. We haven't enough information
6311 to catch that case precisely.
6312
6313 At first, the best we could do was to detect when .* was in capturing brackets
6314 and the highest back reference was greater than or equal to that level.
6315 However, by keeping a bitmap of the first 31 back references, we can catch some
6316 of the more common cases more precisely.
6317
6318 Arguments:
6319 code points to start of expression (the bracket)
6320 options points to the options setting
6321 bracket_map a bitmap of which brackets we are inside while testing; this
6322 handles up to substring 31; after that we just have to take
6323 the less precise approach
6324 backref_map the back reference bitmap
6325
6326 Returns: TRUE or FALSE
6327 */
6328
6329 static BOOL
6330 is_anchored(register const uschar *code, int *options, unsigned int bracket_map,
6331 unsigned int backref_map)
6332 {
6333 do {
6334 const uschar *scode = first_significant_code(code + _pcre_OP_lengths[*code],
6335 options, PCRE_MULTILINE, FALSE);
6336 register int op = *scode;
6337
6338 /* Non-capturing brackets */
6339
6340 if (op == OP_BRA)
6341 {
6342 if (!is_anchored(scode, options, bracket_map, backref_map)) return FALSE;
6343 }
6344
6345 /* Capturing brackets */
6346
6347 else if (op == OP_CBRA)
6348 {
6349 int n = GET2(scode, 1+LINK_SIZE);
6350 int new_map = bracket_map | ((n < 32)? (1 << n) : 1);
6351 if (!is_anchored(scode, options, new_map, backref_map)) return FALSE;
6352 }
6353
6354 /* Other brackets */
6355
6356 else if (op == OP_ASSERT || op == OP_ONCE || op == OP_COND)
6357 {
6358 if (!is_anchored(scode, options, bracket_map, backref_map)) return FALSE;
6359 }
6360
6361 /* .* is not anchored unless DOTALL is set (which generates OP_ALLANY) and
6362 it isn't in brackets that are or may be referenced. */
6363
6364 else if ((op == OP_TYPESTAR || op == OP_TYPEMINSTAR ||
6365 op == OP_TYPEPOSSTAR))
6366 {
6367 if (scode[1] != OP_ALLANY || (bracket_map & backref_map) != 0)
6368 return FALSE;
6369 }
6370
6371 /* Check for explicit anchoring */
6372
6373 else if (op != OP_SOD && op != OP_SOM &&
6374 ((*options & PCRE_MULTILINE) != 0 || op != OP_CIRC))
6375 return FALSE;
6376 code += GET(code, 1);
6377 }
6378 while (*code == OP_ALT); /* Loop for each alternative */
6379 return TRUE;
6380 }
6381
6382
6383
6384 /*************************************************
6385 * Check for starting with ^ or .* *
6386 *************************************************/
6387
6388 /* This is called to find out if every branch starts with ^ or .* so that
6389 "first char" processing can be done to speed things up in multiline
6390 matching and for non-DOTALL patterns that start with .* (which must start at
6391 the beginning or after \n). As in the case of is_anchored() (see above), we
6392 have to take account of back references to capturing brackets that contain .*
6393 because in that case we can't make the assumption.
6394
6395 Arguments:
6396 code points to start of expression (the bracket)
6397 bracket_map a bitmap of which brackets we are inside while testing; this
6398 handles up to substring 31; after that we just have to take
6399 the less precise approach
6400 backref_map the back reference bitmap
6401
6402 Returns: TRUE or FALSE
6403 */
6404
6405 static BOOL
6406 is_startline(const uschar *code, unsigned int bracket_map,
6407 unsigned int backref_map)
6408 {
6409 do {
6410 const uschar *scode = first_significant_code(code + _pcre_OP_lengths[*code],
6411 NULL, 0, FALSE);
6412 register int op = *scode;
6413
6414 /* If we are at the start of a conditional assertion group, *both* the
6415 conditional assertion *and* what follows the condition must satisfy the test
6416 for start of line. Other kinds of condition fail. Note that there may be an
6417 auto-callout at the start of a condition. */
6418
6419 if (op == OP_COND)
6420 {
6421 scode += 1 + LINK_SIZE;
6422 if (*scode == OP_CALLOUT) scode += _pcre_OP_lengths[OP_CALLOUT];
6423 switch (*scode)
6424 {
6425 case OP_CREF:
6426 case OP_NCREF:
6427 case OP_RREF:
6428 case OP_NRREF:
6429 case OP_DEF:
6430 return FALSE;
6431
6432 default: /* Assertion */
6433 if (!is_startline(scode, bracket_map, backref_map)) return FALSE;
6434 do scode += GET(scode, 1); while (*scode == OP_ALT);
6435 scode += 1 + LINK_SIZE;
6436 break;
6437 }
6438 scode = first_significant_code(scode, NULL, 0, FALSE);
6439 op = *scode;
6440 }
6441
6442 /* Non-capturing brackets */
6443
6444 if (op == OP_BRA)
6445 {
6446 if (!is_startline(scode, bracket_map, backref_map)) return FALSE;
6447 }
6448
6449 /* Capturing brackets */
6450
6451 else if (op == OP_CBRA)
6452 {
6453 int n = GET2(scode, 1+LINK_SIZE);
6454 int new_map = bracket_map | ((n < 32)? (1 << n) : 1);
6455 if (!is_startline(scode, new_map, backref_map)) return FALSE;
6456 }
6457
6458 /* Other brackets */
6459
6460 else if (op == OP_ASSERT || op == OP_ONCE)
6461 {
6462 if (!is_startline(scode, bracket_map, backref_map)) return FALSE;
6463 }
6464
6465 /* .* means "start at start or after \n" if it isn't in brackets that
6466 may be referenced. */
6467
6468 else if (op == OP_TYPESTAR || op == OP_TYPEMINSTAR || op == OP_TYPEPOSSTAR)
6469 {
6470 if (scode[1] != OP_ANY || (bracket_map & backref_map) != 0) return FALSE;
6471 }
6472
6473 /* Check for explicit circumflex */
6474
6475 else if (op != OP_CIRC) return FALSE;
6476
6477 /* Move on to the next alternative */
6478
6479 code += GET(code, 1);
6480 }
6481 while (*code == OP_ALT); /* Loop for each alternative */
6482 return TRUE;
6483 }
6484
6485
6486
6487 /*************************************************
6488 * Check for asserted fixed first char *
6489 *************************************************/
6490
6491 /* During compilation, the "first char" settings from forward assertions are
6492 discarded, because they can cause conflicts with actual literals that follow.
6493 However, if we end up without a first char setting for an unanchored pattern,
6494 it is worth scanning the regex to see if there is an initial asserted first
6495 char. If all branches start with the same asserted char, or with a bracket all
6496 of whose alternatives start with the same asserted char (recurse ad lib), then
6497 we return that char, otherwise -1.
6498
6499 Arguments:
6500 code points to start of expression (the bracket)
6501 options pointer to the options (used to check casing changes)
6502 inassert TRUE if in an assertion
6503
6504 Returns: -1 or the fixed first char
6505 */
6506
6507 static int
6508 find_firstassertedchar(const uschar *code, int *options, BOOL inassert)
6509 {
6510 register int c = -1;
6511 do {
6512 int d;
6513 const uschar *scode =
6514 first_significant_code(code + 1+LINK_SIZE, options, PCRE_CASELESS, TRUE);
6515 register int op = *scode;
6516
6517 switch(op)
6518 {
6519 default:
6520 return -1;
6521
6522 case OP_BRA:
6523 case OP_CBRA:
6524 case OP_ASSERT:
6525 case OP_ONCE:
6526 case OP_COND:
6527 if ((d = find_firstassertedchar(scode, options, op == OP_ASSERT)) < 0)
6528 return -1;
6529 if (c < 0) c = d; else if (c != d) return -1;
6530 break;
6531
6532 case OP_EXACT: /* Fall through */
6533 scode += 2;
6534
6535 case OP_CHAR:
6536 case OP_CHARNC:
6537 case OP_PLUS:
6538 case OP_MINPLUS:
6539 case OP_POSPLUS:
6540 if (!inassert) return -1;
6541 if (c < 0)
6542 {
6543 c = scode[1];
6544 if ((*options & PCRE_CASELESS) != 0) c |= REQ_CASELESS;
6545 }
6546 else if (c != scode[1]) return -1;
6547 break;
6548 }
6549
6550 code += GET(code, 1);
6551 }
6552 while (*code == OP_ALT);
6553 return c;
6554 }
6555
6556
6557
6558 /*************************************************
6559 * Compile a Regular Expression *
6560 *************************************************/
6561
6562 /* This function takes a string and returns a pointer to a block of store
6563 holding a compiled version of the expression. The original API for this
6564 function had no error code return variable; it is retained for backwards
6565 compatibility. The new function is given a new name.
6566
6567 Arguments:
6568 pattern the regular expression
6569 options various option bits
6570 errorcodeptr pointer to error code variable (pcre_compile2() only)
6571 can be NULL if you don't want a code value
6572 errorptr pointer to pointer to error text
6573 erroroffset ptr offset in pattern where error was detected
6574 tables pointer to character tables or NULL
6575
6576 Returns: pointer to compiled data block, or NULL on error,
6577 with errorptr and erroroffset set
6578 */
6579
6580 PCRE_EXP_DEFN pcre * PCRE_CALL_CONVENTION
6581 pcre_compile(const char *pattern, int options, const char **errorptr,
6582 int *erroroffset, const unsigned char *tables)
6583 {
6584 return pcre_compile2(pattern, options, NULL, errorptr, erroroffset, tables);
6585 }
6586
6587
6588 PCRE_EXP_DEFN pcre * PCRE_CALL_CONVENTION
6589 pcre_compile2(const char *pattern, int options, int *errorcodeptr,
6590 const char **errorptr, int *erroroffset, const unsigned char *tables)
6591 {
6592 real_pcre *re;
6593 int length = 1; /* For final END opcode */
6594 int firstbyte, reqbyte, newline;
6595 int errorcode = 0;
6596 int skipatstart = 0;
6597 BOOL utf8;
6598 size_t size;
6599 uschar *code;
6600 const uschar *codestart;
6601 const uschar *ptr;
6602 compile_data compile_block;
6603 compile_data *cd = &compile_block;
6604
6605 /* This space is used for "compiling" into during the first phase, when we are
6606 computing the amount of memory that is needed. Compiled items are thrown away
6607 as soon as possible, so that a fairly large buffer should be sufficient for
6608 this purpose. The same space is used in the second phase for remembering where
6609 to fill in forward references to subpatterns. */
6610
6611 uschar cworkspace[COMPILE_WORK_SIZE];
6612
6613 /* Set this early so that early errors get offset 0. */
6614
6615 ptr = (const uschar *)pattern;
6616
6617 /* We can't pass back an error message if errorptr is NULL; I guess the best we
6618 can do is just return NULL, but we can set a code value if there is a code
6619 pointer. */
6620
6621 if (errorptr == NULL)
6622 {
6623 if (errorcodeptr != NULL) *errorcodeptr = 99;
6624 return NULL;
6625 }
6626
6627 *errorptr = NULL;
6628 if (errorcodeptr != NULL) *errorcodeptr = ERR0;
6629
6630 /* However, we can give a message for this error */
6631
6632 if (erroroffset == NULL)
6633 {
6634 errorcode = ERR16;
6635 goto PCRE_EARLY_ERROR_RETURN2;
6636 }
6637
6638 *erroroffset = 0;
6639
6640 /* Set up pointers to the individual character tables */
6641
6642 if (tables == NULL) tables = _pcre_default_tables;
6643 cd->lcc = tables + lcc_offset;
6644 cd->fcc = tables + fcc_offset;
6645 cd->cbits = tables + cbits_offset;
6646 cd->ctypes = tables + ctypes_offset;
6647
6648 /* Check that all undefined public option bits are zero */
6649
6650 if ((options & ~PUBLIC_COMPILE_OPTIONS) != 0)
6651 {
6652 errorcode = ERR17;
6653 goto PCRE_EARLY_ERROR_RETURN;
6654 }
6655
6656 /* Check for global one-time settings at the start of the pattern, and remember
6657 the offset for later. */
6658
6659 while (ptr[skipatstart] == CHAR_LEFT_PARENTHESIS &&
6660 ptr[skipatstart+1] == CHAR_ASTERISK)
6661 {
6662 int newnl = 0;
6663 int newbsr = 0;
6664
6665 if (strn