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

Contents of /code/trunk/pcre_compile.c

Parent Directory Parent Directory | Revision Log Revision Log


Revision 518 - (show annotations)
Tue May 18 15:47:01 2010 UTC (9 years, 5 months ago) by ph10
File MIME type: text/plain
File size: 228686 byte(s)
Error occurred while calculating annotation data.
Added PCRE_UCP and related stuff to make \w etc use Unicode properties.
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, 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 = 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 case OP_DIGIT:
2549 return next > 127 || (cd->ctypes[next] & ctype_digit) == 0;
2550
2551 case OP_NOT_DIGIT:
2552 return next <= 127 && (cd->ctypes[next] & ctype_digit) != 0;
2553
2554 case OP_WHITESPACE:
2555 return next > 127 || (cd->ctypes[next] & ctype_space) == 0;
2556
2557 case OP_NOT_WHITESPACE:
2558 return next <= 127 && (cd->ctypes[next] & ctype_space) != 0;
2559
2560 case OP_WORDCHAR:
2561 return next > 127 || (cd->ctypes[next] & ctype_word) == 0;
2562
2563 case OP_NOT_WORDCHAR:
2564 return next <= 127 && (cd->ctypes[next] & ctype_word) != 0;
2565
2566 case OP_HSPACE:
2567 case OP_NOT_HSPACE:
2568 switch(next)
2569 {
2570 case 0x09:
2571 case 0x20:
2572 case 0xa0:
2573 case 0x1680:
2574 case 0x180e:
2575 case 0x2000:
2576 case 0x2001:
2577 case 0x2002:
2578 case 0x2003:
2579 case 0x2004:
2580 case 0x2005:
2581 case 0x2006:
2582 case 0x2007:
2583 case 0x2008:
2584 case 0x2009:
2585 case 0x200A:
2586 case 0x202f:
2587 case 0x205f:
2588 case 0x3000:
2589 return op_code != OP_HSPACE;
2590 default:
2591 return op_code == OP_HSPACE;
2592 }
2593
2594 case OP_VSPACE:
2595 case OP_NOT_VSPACE:
2596 switch(next)
2597 {
2598 case 0x0a:
2599 case 0x0b:
2600 case 0x0c:
2601 case 0x0d:
2602 case 0x85:
2603 case 0x2028:
2604 case 0x2029:
2605 return op_code != OP_VSPACE;
2606 default:
2607 return op_code == OP_VSPACE;
2608 }
2609
2610 default:
2611 return FALSE;
2612 }
2613
2614
2615 /* Handle the case when the next item is \d, \s, etc. */
2616
2617 switch(op_code)
2618 {
2619 case OP_CHAR:
2620 case OP_CHARNC:
2621 #ifdef SUPPORT_UTF8
2622 if (utf8 && item > 127) { GETCHAR(item, utf8_char); }
2623 #endif
2624 switch(-next)
2625 {
2626 case ESC_d:
2627 return item > 127 || (cd->ctypes[item] & ctype_digit) == 0;
2628
2629 case ESC_D:
2630 return item <= 127 && (cd->ctypes[item] & ctype_digit) != 0;
2631
2632 case ESC_s:
2633 return item > 127 || (cd->ctypes[item] & ctype_space) == 0;
2634
2635 case ESC_S:
2636 return item <= 127 && (cd->ctypes[item] & ctype_space) != 0;
2637
2638 case ESC_w:
2639 return item > 127 || (cd->ctypes[item] & ctype_word) == 0;
2640
2641 case ESC_W:
2642 return item <= 127 && (cd->ctypes[item] & ctype_word) != 0;
2643
2644 case ESC_h:
2645 case ESC_H:
2646 switch(item)
2647 {
2648 case 0x09:
2649 case 0x20:
2650 case 0xa0:
2651 case 0x1680:
2652 case 0x180e:
2653 case 0x2000:
2654 case 0x2001:
2655 case 0x2002:
2656 case 0x2003:
2657 case 0x2004:
2658 case 0x2005:
2659 case 0x2006:
2660 case 0x2007:
2661 case 0x2008:
2662 case 0x2009:
2663 case 0x200A:
2664 case 0x202f:
2665 case 0x205f:
2666 case 0x3000:
2667 return -next != ESC_h;
2668 default:
2669 return -next == ESC_h;
2670 }
2671
2672 case ESC_v:
2673 case ESC_V:
2674 switch(item)
2675 {
2676 case 0x0a:
2677 case 0x0b:
2678 case 0x0c:
2679 case 0x0d:
2680 case 0x85:
2681 case 0x2028:
2682 case 0x2029:
2683 return -next != ESC_v;
2684 default:
2685 return -next == ESC_v;
2686 }
2687
2688 default:
2689 return FALSE;
2690 }
2691
2692 case OP_DIGIT:
2693 return next == -ESC_D || next == -ESC_s || next == -ESC_W ||
2694 next == -ESC_h || next == -ESC_v;
2695
2696 case OP_NOT_DIGIT:
2697 return next == -ESC_d;
2698
2699 case OP_WHITESPACE:
2700 return next == -ESC_S || next == -ESC_d || next == -ESC_w;
2701
2702 case OP_NOT_WHITESPACE:
2703 return next == -ESC_s || next == -ESC_h || next == -ESC_v;
2704
2705 case OP_HSPACE:
2706 return next == -ESC_S || next == -ESC_H || next == -ESC_d || next == -ESC_w;
2707
2708 case OP_NOT_HSPACE:
2709 return next == -ESC_h;
2710
2711 /* Can't have \S in here because VT matches \S (Perl anomaly) */
2712 case OP_VSPACE:
2713 return next == -ESC_V || next == -ESC_d || next == -ESC_w;
2714
2715 case OP_NOT_VSPACE:
2716 return next == -ESC_v;
2717
2718 case OP_WORDCHAR:
2719 return next == -ESC_W || next == -ESC_s || next == -ESC_h || next == -ESC_v;
2720
2721 case OP_NOT_WORDCHAR:
2722 return next == -ESC_w || next == -ESC_d;
2723
2724 default:
2725 return FALSE;
2726 }
2727
2728 /* Control does not reach here */
2729 }
2730
2731
2732
2733 /*************************************************
2734 * Compile one branch *
2735 *************************************************/
2736
2737 /* Scan the pattern, compiling it into the a vector. If the options are
2738 changed during the branch, the pointer is used to change the external options
2739 bits. This function is used during the pre-compile phase when we are trying
2740 to find out the amount of memory needed, as well as during the real compile
2741 phase. The value of lengthptr distinguishes the two phases.
2742
2743 Arguments:
2744 optionsptr pointer to the option bits
2745 codeptr points to the pointer to the current code point
2746 ptrptr points to the current pattern pointer
2747 errorcodeptr points to error code variable
2748 firstbyteptr set to initial literal character, or < 0 (REQ_UNSET, REQ_NONE)
2749 reqbyteptr set to the last literal character required, else < 0
2750 bcptr points to current branch chain
2751 cd contains pointers to tables etc.
2752 lengthptr NULL during the real compile phase
2753 points to length accumulator during pre-compile phase
2754
2755 Returns: TRUE on success
2756 FALSE, with *errorcodeptr set non-zero on error
2757 */
2758
2759 static BOOL
2760 compile_branch(int *optionsptr, uschar **codeptr, const uschar **ptrptr,
2761 int *errorcodeptr, int *firstbyteptr, int *reqbyteptr, branch_chain *bcptr,
2762 compile_data *cd, int *lengthptr)
2763 {
2764 int repeat_type, op_type;
2765 int repeat_min = 0, repeat_max = 0; /* To please picky compilers */
2766 int bravalue = 0;
2767 int greedy_default, greedy_non_default;
2768 int firstbyte, reqbyte;
2769 int zeroreqbyte, zerofirstbyte;
2770 int req_caseopt, reqvary, tempreqvary;
2771 int options = *optionsptr;
2772 int after_manual_callout = 0;
2773 int length_prevgroup = 0;
2774 register int c;
2775 register uschar *code = *codeptr;
2776 uschar *last_code = code;
2777 uschar *orig_code = code;
2778 uschar *tempcode;
2779 BOOL inescq = FALSE;
2780 BOOL groupsetfirstbyte = FALSE;
2781 const uschar *ptr = *ptrptr;
2782 const uschar *tempptr;
2783 const uschar *nestptr = NULL;
2784 uschar *previous = NULL;
2785 uschar *previous_callout = NULL;
2786 uschar *save_hwm = NULL;
2787 uschar classbits[32];
2788
2789 #ifdef SUPPORT_UTF8
2790 BOOL class_utf8;
2791 BOOL utf8 = (options & PCRE_UTF8) != 0;
2792 uschar *class_utf8data;
2793 uschar *class_utf8data_base;
2794 uschar utf8_char[6];
2795 #else
2796 BOOL utf8 = FALSE;
2797 uschar *utf8_char = NULL;
2798 #endif
2799
2800 #ifdef PCRE_DEBUG
2801 if (lengthptr != NULL) DPRINTF((">> start branch\n"));
2802 #endif
2803
2804 /* Set up the default and non-default settings for greediness */
2805
2806 greedy_default = ((options & PCRE_UNGREEDY) != 0);
2807 greedy_non_default = greedy_default ^ 1;
2808
2809 /* Initialize no first byte, no required byte. REQ_UNSET means "no char
2810 matching encountered yet". It gets changed to REQ_NONE if we hit something that
2811 matches a non-fixed char first char; reqbyte just remains unset if we never
2812 find one.
2813
2814 When we hit a repeat whose minimum is zero, we may have to adjust these values
2815 to take the zero repeat into account. This is implemented by setting them to
2816 zerofirstbyte and zeroreqbyte when such a repeat is encountered. The individual
2817 item types that can be repeated set these backoff variables appropriately. */
2818
2819 firstbyte = reqbyte = zerofirstbyte = zeroreqbyte = REQ_UNSET;
2820
2821 /* The variable req_caseopt contains either the REQ_CASELESS value or zero,
2822 according to the current setting of the caseless flag. REQ_CASELESS is a bit
2823 value > 255. It is added into the firstbyte or reqbyte variables to record the
2824 case status of the value. This is used only for ASCII characters. */
2825
2826 req_caseopt = ((options & PCRE_CASELESS) != 0)? REQ_CASELESS : 0;
2827
2828 /* Switch on next character until the end of the branch */
2829
2830 for (;; ptr++)
2831 {
2832 BOOL negate_class;
2833 BOOL should_flip_negation;
2834 BOOL possessive_quantifier;
2835 BOOL is_quantifier;
2836 BOOL is_recurse;
2837 BOOL reset_bracount;
2838 int class_charcount;
2839 int class_lastchar;
2840 int newoptions;
2841 int recno;
2842 int refsign;
2843 int skipbytes;
2844 int subreqbyte;
2845 int subfirstbyte;
2846 int terminator;
2847 int mclength;
2848 uschar mcbuffer[8];
2849
2850 /* Get next byte in the pattern */
2851
2852 c = *ptr;
2853
2854 /* If we are at the end of a nested substitution, revert to the outer level
2855 string. Nesting only happens one level deep. */
2856
2857 if (c == 0 && nestptr != NULL)
2858 {
2859 ptr = nestptr;
2860 nestptr = NULL;
2861 c = *ptr;
2862 }
2863
2864 /* If we are in the pre-compile phase, accumulate the length used for the
2865 previous cycle of this loop. */
2866
2867 if (lengthptr != NULL)
2868 {
2869 #ifdef PCRE_DEBUG
2870 if (code > cd->hwm) cd->hwm = code; /* High water info */
2871 #endif
2872 if (code > cd->start_workspace + WORK_SIZE_CHECK) /* Check for overrun */
2873 {
2874 *errorcodeptr = ERR52;
2875 goto FAILED;
2876 }
2877
2878 /* There is at least one situation where code goes backwards: this is the
2879 case of a zero quantifier after a class (e.g. [ab]{0}). At compile time,
2880 the class is simply eliminated. However, it is created first, so we have to
2881 allow memory for it. Therefore, don't ever reduce the length at this point.
2882 */
2883
2884 if (code < last_code) code = last_code;
2885
2886 /* Paranoid check for integer overflow */
2887
2888 if (OFLOW_MAX - *lengthptr < code - last_code)
2889 {
2890 *errorcodeptr = ERR20;
2891 goto FAILED;
2892 }
2893
2894 *lengthptr += code - last_code;
2895 DPRINTF(("length=%d added %d c=%c\n", *lengthptr, code - last_code, c));
2896
2897 /* If "previous" is set and it is not at the start of the work space, move
2898 it back to there, in order to avoid filling up the work space. Otherwise,
2899 if "previous" is NULL, reset the current code pointer to the start. */
2900
2901 if (previous != NULL)
2902 {
2903 if (previous > orig_code)
2904 {
2905 memmove(orig_code, previous, code - previous);
2906 code -= previous - orig_code;
2907 previous = orig_code;
2908 }
2909 }
2910 else code = orig_code;
2911
2912 /* Remember where this code item starts so we can pick up the length
2913 next time round. */
2914
2915 last_code = code;
2916 }
2917
2918 /* In the real compile phase, just check the workspace used by the forward
2919 reference list. */
2920
2921 else if (cd->hwm > cd->start_workspace + WORK_SIZE_CHECK)
2922 {
2923 *errorcodeptr = ERR52;
2924 goto FAILED;
2925 }
2926
2927 /* If in \Q...\E, check for the end; if not, we have a literal */
2928
2929 if (inescq && c != 0)
2930 {
2931 if (c == CHAR_BACKSLASH && ptr[1] == CHAR_E)
2932 {
2933 inescq = FALSE;
2934 ptr++;
2935 continue;
2936 }
2937 else
2938 {
2939 if (previous_callout != NULL)
2940 {
2941 if (lengthptr == NULL) /* Don't attempt in pre-compile phase */
2942 complete_callout(previous_callout, ptr, cd);
2943 previous_callout = NULL;
2944 }
2945 if ((options & PCRE_AUTO_CALLOUT) != 0)
2946 {
2947 previous_callout = code;
2948 code = auto_callout(code, ptr, cd);
2949 }
2950 goto NORMAL_CHAR;
2951 }
2952 }
2953
2954 /* Fill in length of a previous callout, except when the next thing is
2955 a quantifier. */
2956
2957 is_quantifier =
2958 c == CHAR_ASTERISK || c == CHAR_PLUS || c == CHAR_QUESTION_MARK ||
2959 (c == CHAR_LEFT_CURLY_BRACKET && is_counted_repeat(ptr+1));
2960
2961 if (!is_quantifier && previous_callout != NULL &&
2962 after_manual_callout-- <= 0)
2963 {
2964 if (lengthptr == NULL) /* Don't attempt in pre-compile phase */
2965 complete_callout(previous_callout, ptr, cd);
2966 previous_callout = NULL;
2967 }
2968
2969 /* In extended mode, skip white space and comments */
2970
2971 if ((options & PCRE_EXTENDED) != 0)
2972 {
2973 if ((cd->ctypes[c] & ctype_space) != 0) continue;
2974 if (c == CHAR_NUMBER_SIGN)
2975 {
2976 while (*(++ptr) != 0)
2977 {
2978 if (IS_NEWLINE(ptr)) { ptr += cd->nllen - 1; break; }
2979 }
2980 if (*ptr != 0) continue;
2981
2982 /* Else fall through to handle end of string */
2983 c = 0;
2984 }
2985 }
2986
2987 /* No auto callout for quantifiers. */
2988
2989 if ((options & PCRE_AUTO_CALLOUT) != 0 && !is_quantifier)
2990 {
2991 previous_callout = code;
2992 code = auto_callout(code, ptr, cd);
2993 }
2994
2995 switch(c)
2996 {
2997 /* ===================================================================*/
2998 case 0: /* The branch terminates at string end */
2999 case CHAR_VERTICAL_LINE: /* or | or ) */
3000 case CHAR_RIGHT_PARENTHESIS:
3001 *firstbyteptr = firstbyte;
3002 *reqbyteptr = reqbyte;
3003 *codeptr = code;
3004 *ptrptr = ptr;
3005 if (lengthptr != NULL)
3006 {
3007 if (OFLOW_MAX - *lengthptr < code - last_code)
3008 {
3009 *errorcodeptr = ERR20;
3010 goto FAILED;
3011 }
3012 *lengthptr += code - last_code; /* To include callout length */
3013 DPRINTF((">> end branch\n"));
3014 }
3015 return TRUE;
3016
3017
3018 /* ===================================================================*/
3019 /* Handle single-character metacharacters. In multiline mode, ^ disables
3020 the setting of any following char as a first character. */
3021
3022 case CHAR_CIRCUMFLEX_ACCENT:
3023 if ((options & PCRE_MULTILINE) != 0)
3024 {
3025 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
3026 }
3027 previous = NULL;
3028 *code++ = OP_CIRC;
3029 break;
3030
3031 case CHAR_DOLLAR_SIGN:
3032 previous = NULL;
3033 *code++ = OP_DOLL;
3034 break;
3035
3036 /* There can never be a first char if '.' is first, whatever happens about
3037 repeats. The value of reqbyte doesn't change either. */
3038
3039 case CHAR_DOT:
3040 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
3041 zerofirstbyte = firstbyte;
3042 zeroreqbyte = reqbyte;
3043 previous = code;
3044 *code++ = ((options & PCRE_DOTALL) != 0)? OP_ALLANY: OP_ANY;
3045 break;
3046
3047
3048 /* ===================================================================*/
3049 /* Character classes. If the included characters are all < 256, we build a
3050 32-byte bitmap of the permitted characters, except in the special case
3051 where there is only one such character. For negated classes, we build the
3052 map as usual, then invert it at the end. However, we use a different opcode
3053 so that data characters > 255 can be handled correctly.
3054
3055 If the class contains characters outside the 0-255 range, a different
3056 opcode is compiled. It may optionally have a bit map for characters < 256,
3057 but those above are are explicitly listed afterwards. A flag byte tells
3058 whether the bitmap is present, and whether this is a negated class or not.
3059
3060 In JavaScript compatibility mode, an isolated ']' causes an error. In
3061 default (Perl) mode, it is treated as a data character. */
3062
3063 case CHAR_RIGHT_SQUARE_BRACKET:
3064 if ((cd->external_options & PCRE_JAVASCRIPT_COMPAT) != 0)
3065 {
3066 *errorcodeptr = ERR64;
3067 goto FAILED;
3068 }
3069 goto NORMAL_CHAR;
3070
3071 case CHAR_LEFT_SQUARE_BRACKET:
3072 previous = code;
3073
3074 /* PCRE supports POSIX class stuff inside a class. Perl gives an error if
3075 they are encountered at the top level, so we'll do that too. */
3076
3077 if ((ptr[1] == CHAR_COLON || ptr[1] == CHAR_DOT ||
3078 ptr[1] == CHAR_EQUALS_SIGN) &&
3079 check_posix_syntax(ptr, &tempptr))
3080 {
3081 *errorcodeptr = (ptr[1] == CHAR_COLON)? ERR13 : ERR31;
3082 goto FAILED;
3083 }
3084
3085 /* If the first character is '^', set the negation flag and skip it. Also,
3086 if the first few characters (either before or after ^) are \Q\E or \E we
3087 skip them too. This makes for compatibility with Perl. */
3088
3089 negate_class = FALSE;
3090 for (;;)
3091 {
3092 c = *(++ptr);
3093 if (c == CHAR_BACKSLASH)
3094 {
3095 if (ptr[1] == CHAR_E)
3096 ptr++;
3097 else if (strncmp((const char *)ptr+1,
3098 STR_Q STR_BACKSLASH STR_E, 3) == 0)
3099 ptr += 3;
3100 else
3101 break;
3102 }
3103 else if (!negate_class && c == CHAR_CIRCUMFLEX_ACCENT)
3104 negate_class = TRUE;
3105 else break;
3106 }
3107
3108 /* Empty classes are allowed in JavaScript compatibility mode. Otherwise,
3109 an initial ']' is taken as a data character -- the code below handles
3110 that. In JS mode, [] must always fail, so generate OP_FAIL, whereas
3111 [^] must match any character, so generate OP_ALLANY. */
3112
3113 if (c == CHAR_RIGHT_SQUARE_BRACKET &&
3114 (cd->external_options & PCRE_JAVASCRIPT_COMPAT) != 0)
3115 {
3116 *code++ = negate_class? OP_ALLANY : OP_FAIL;
3117 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
3118 zerofirstbyte = firstbyte;
3119 break;
3120 }
3121
3122 /* If a class contains a negative special such as \S, we need to flip the
3123 negation flag at the end, so that support for characters > 255 works
3124 correctly (they are all included in the class). */
3125
3126 should_flip_negation = FALSE;
3127
3128 /* Keep a count of chars with values < 256 so that we can optimize the case
3129 of just a single character (as long as it's < 256). However, For higher
3130 valued UTF-8 characters, we don't yet do any optimization. */
3131
3132 class_charcount = 0;
3133 class_lastchar = -1;
3134
3135 /* Initialize the 32-char bit map to all zeros. We build the map in a
3136 temporary bit of memory, in case the class contains only 1 character (less
3137 than 256), because in that case the compiled code doesn't use the bit map.
3138 */
3139
3140 memset(classbits, 0, 32 * sizeof(uschar));
3141
3142 #ifdef SUPPORT_UTF8
3143 class_utf8 = FALSE; /* No chars >= 256 */
3144 class_utf8data = code + LINK_SIZE + 2; /* For UTF-8 items */
3145 class_utf8data_base = class_utf8data; /* For resetting in pass 1 */
3146 #endif
3147
3148 /* Process characters until ] is reached. By writing this as a "do" it
3149 means that an initial ] is taken as a data character. At the start of the
3150 loop, c contains the first byte of the character. */
3151
3152 if (c != 0) do
3153 {
3154 const uschar *oldptr;
3155
3156 #ifdef SUPPORT_UTF8
3157 if (utf8 && c > 127)
3158 { /* Braces are required because the */
3159 GETCHARLEN(c, ptr, ptr); /* macro generates multiple statements */
3160 }
3161
3162 /* In the pre-compile phase, accumulate the length of any UTF-8 extra
3163 data and reset the pointer. This is so that very large classes that
3164 contain a zillion UTF-8 characters no longer overwrite the work space
3165 (which is on the stack). */
3166
3167 if (lengthptr != NULL)
3168 {
3169 *lengthptr += class_utf8data - class_utf8data_base;
3170 class_utf8data = class_utf8data_base;
3171 }
3172
3173 #endif
3174
3175 /* Inside \Q...\E everything is literal except \E */
3176
3177 if (inescq)
3178 {
3179 if (c == CHAR_BACKSLASH && ptr[1] == CHAR_E) /* If we are at \E */
3180 {
3181 inescq = FALSE; /* Reset literal state */
3182 ptr++; /* Skip the 'E' */
3183 continue; /* Carry on with next */
3184 }
3185 goto CHECK_RANGE; /* Could be range if \E follows */
3186 }
3187
3188 /* Handle POSIX class names. Perl allows a negation extension of the
3189 form [:^name:]. A square bracket that doesn't match the syntax is
3190 treated as a literal. We also recognize the POSIX constructions
3191 [.ch.] and [=ch=] ("collating elements") and fault them, as Perl
3192 5.6 and 5.8 do. */
3193
3194 if (c == CHAR_LEFT_SQUARE_BRACKET &&
3195 (ptr[1] == CHAR_COLON || ptr[1] == CHAR_DOT ||
3196 ptr[1] == CHAR_EQUALS_SIGN) && check_posix_syntax(ptr, &tempptr))
3197 {
3198 BOOL local_negate = FALSE;
3199 int posix_class, taboffset, tabopt;
3200 register const uschar *cbits = cd->cbits;
3201 uschar pbits[32];
3202
3203 if (ptr[1] != CHAR_COLON)
3204 {
3205 *errorcodeptr = ERR31;
3206 goto FAILED;
3207 }
3208
3209 ptr += 2;
3210 if (*ptr == CHAR_CIRCUMFLEX_ACCENT)
3211 {
3212 local_negate = TRUE;
3213 should_flip_negation = TRUE; /* Note negative special */
3214 ptr++;
3215 }
3216
3217 posix_class = check_posix_name(ptr, tempptr - ptr);
3218 if (posix_class < 0)
3219 {
3220 *errorcodeptr = ERR30;
3221 goto FAILED;
3222 }
3223
3224 /* If matching is caseless, upper and lower are converted to
3225 alpha. This relies on the fact that the class table starts with
3226 alpha, lower, upper as the first 3 entries. */
3227
3228 if ((options & PCRE_CASELESS) != 0 && posix_class <= 2)
3229 posix_class = 0;
3230
3231 /* When PCRE_UCP is set, some of the POSIX classes are converted to
3232 different escape sequences that use Unicode properties. */
3233
3234 #ifdef SUPPORT_UCP
3235 if ((options & PCRE_UCP) != 0)
3236 {
3237 int pc = posix_class + ((local_negate)? POSIX_SUBSIZE/2 : 0);
3238 if (posix_substitutes[pc] != NULL)
3239 {
3240 nestptr = tempptr + 1;
3241 ptr = posix_substitutes[pc] - 1;
3242 continue;
3243 }
3244 }
3245 #endif
3246 /* In the non-UCP case, we build the bit map for the POSIX class in a
3247 chunk of local store because we may be adding and subtracting from it,
3248 and we don't want to subtract bits that may be in the main map already.
3249 At the end we or the result into the bit map that is being built. */
3250
3251 posix_class *= 3;
3252
3253 /* Copy in the first table (always present) */
3254
3255 memcpy(pbits, cbits + posix_class_maps[posix_class],
3256 32 * sizeof(uschar));
3257
3258 /* If there is a second table, add or remove it as required. */
3259
3260 taboffset = posix_class_maps[posix_class + 1];
3261 tabopt = posix_class_maps[posix_class + 2];
3262
3263 if (taboffset >= 0)
3264 {
3265 if (tabopt >= 0)
3266 for (c = 0; c < 32; c++) pbits[c] |= cbits[c + taboffset];
3267 else
3268 for (c = 0; c < 32; c++) pbits[c] &= ~cbits[c + taboffset];
3269 }
3270
3271 /* Not see if we need to remove any special characters. An option
3272 value of 1 removes vertical space and 2 removes underscore. */
3273
3274 if (tabopt < 0) tabopt = -tabopt;
3275 if (tabopt == 1) pbits[1] &= ~0x3c;
3276 else if (tabopt == 2) pbits[11] &= 0x7f;
3277
3278 /* Add the POSIX table or its complement into the main table that is
3279 being built and we are done. */
3280
3281 if (local_negate)
3282 for (c = 0; c < 32; c++) classbits[c] |= ~pbits[c];
3283 else
3284 for (c = 0; c < 32; c++) classbits[c] |= pbits[c];
3285
3286 ptr = tempptr + 1;
3287 class_charcount = 10; /* Set > 1; assumes more than 1 per class */
3288 continue; /* End of POSIX syntax handling */
3289 }
3290
3291 /* Backslash may introduce a single character, or it may introduce one
3292 of the specials, which just set a flag. The sequence \b is a special
3293 case. Inside a class (and only there) it is treated as backspace. We
3294 assume that other escapes have more than one character in them, so set
3295 class_charcount bigger than one. Unrecognized escapes fall through and
3296 are either treated as literal characters (by default), or are faulted if
3297 PCRE_EXTRA is set. */
3298
3299 if (c == CHAR_BACKSLASH)
3300 {
3301 c = check_escape(&ptr, errorcodeptr, cd->bracount, options, TRUE);
3302 if (*errorcodeptr != 0) goto FAILED;
3303
3304 if (-c == ESC_b) c = CHAR_BS; /* \b is backspace in a class */
3305 else if (-c == ESC_Q) /* Handle start of quoted string */
3306 {
3307 if (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
3308 {
3309 ptr += 2; /* avoid empty string */
3310 }
3311 else inescq = TRUE;
3312 continue;
3313 }
3314 else if (-c == ESC_E) continue; /* Ignore orphan \E */
3315
3316 if (c < 0)
3317 {
3318 register const uschar *cbits = cd->cbits;
3319 class_charcount += 2; /* Greater than 1 is what matters */
3320
3321 switch (-c)
3322 {
3323 #ifdef SUPPORT_UCP
3324 case ESC_du: /* These are the values given for \d etc */
3325 case ESC_DU: /* when PCRE_UCP is set. We replace the */
3326 case ESC_wu: /* escape sequence with an appropriate \p */
3327 case ESC_WU: /* or \P to test Unicode properties instead */
3328 case ESC_su: /* of the default ASCII testing. */
3329 case ESC_SU:
3330 nestptr = ptr;
3331 ptr = substitutes[-c - ESC_DU] - 1; /* Just before substitute */
3332 class_charcount -= 2; /* Undo! */
3333 continue;
3334 #endif
3335 case ESC_d:
3336 for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_digit];
3337 continue;
3338
3339 case ESC_D:
3340 should_flip_negation = TRUE;
3341 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_digit];
3342 continue;
3343
3344 case ESC_w:
3345 for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_word];
3346 continue;
3347
3348 case ESC_W:
3349 should_flip_negation = TRUE;
3350 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_word];
3351 continue;
3352
3353 case ESC_s:
3354 for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_space];
3355 classbits[1] &= ~0x08; /* Perl 5.004 onwards omits VT from \s */
3356 continue;
3357
3358 case ESC_S:
3359 should_flip_negation = TRUE;
3360 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_space];
3361 classbits[1] |= 0x08; /* Perl 5.004 onwards omits VT from \s */
3362 continue;
3363
3364 case ESC_h:
3365 SETBIT(classbits, 0x09); /* VT */
3366 SETBIT(classbits, 0x20); /* SPACE */
3367 SETBIT(classbits, 0xa0); /* NSBP */
3368 #ifdef SUPPORT_UTF8
3369 if (utf8)
3370 {
3371 class_utf8 = TRUE;
3372 *class_utf8data++ = XCL_SINGLE;
3373 class_utf8data += _pcre_ord2utf8(0x1680, class_utf8data);
3374 *class_utf8data++ = XCL_SINGLE;
3375 class_utf8data += _pcre_ord2utf8(0x180e, class_utf8data);
3376 *class_utf8data++ = XCL_RANGE;
3377 class_utf8data += _pcre_ord2utf8(0x2000, class_utf8data);
3378 class_utf8data += _pcre_ord2utf8(0x200A, class_utf8data);
3379 *class_utf8data++ = XCL_SINGLE;
3380 class_utf8data += _pcre_ord2utf8(0x202f, class_utf8data);
3381 *class_utf8data++ = XCL_SINGLE;
3382 class_utf8data += _pcre_ord2utf8(0x205f, class_utf8data);
3383 *class_utf8data++ = XCL_SINGLE;
3384 class_utf8data += _pcre_ord2utf8(0x3000, class_utf8data);
3385 }
3386 #endif
3387 continue;
3388
3389 case ESC_H:
3390 for (c = 0; c < 32; c++)
3391 {
3392 int x = 0xff;
3393 switch (c)
3394 {
3395 case 0x09/8: x ^= 1 << (0x09%8); break;
3396 case 0x20/8: x ^= 1 << (0x20%8); break;
3397 case 0xa0/8: x ^= 1 << (0xa0%8); break;
3398 default: break;
3399 }
3400 classbits[c] |= x;
3401 }
3402
3403 #ifdef SUPPORT_UTF8
3404 if (utf8)
3405 {
3406 class_utf8 = TRUE;
3407 *class_utf8data++ = XCL_RANGE;
3408 class_utf8data += _pcre_ord2utf8(0x0100, class_utf8data);
3409 class_utf8data += _pcre_ord2utf8(0x167f, class_utf8data);
3410 *class_utf8data++ = XCL_RANGE;
3411 class_utf8data += _pcre_ord2utf8(0x1681, class_utf8data);
3412 class_utf8data += _pcre_ord2utf8(0x180d, class_utf8data);
3413 *class_utf8data++ = XCL_RANGE;
3414 class_utf8data += _pcre_ord2utf8(0x180f, class_utf8data);
3415 class_utf8data += _pcre_ord2utf8(0x1fff, class_utf8data);
3416 *class_utf8data++ = XCL_RANGE;
3417 class_utf8data += _pcre_ord2utf8(0x200B, class_utf8data);
3418 class_utf8data += _pcre_ord2utf8(0x202e, class_utf8data);
3419 *class_utf8data++ = XCL_RANGE;
3420 class_utf8data += _pcre_ord2utf8(0x2030, class_utf8data);
3421 class_utf8data += _pcre_ord2utf8(0x205e, class_utf8data);
3422 *class_utf8data++ = XCL_RANGE;
3423 class_utf8data += _pcre_ord2utf8(0x2060, class_utf8data);
3424 class_utf8data += _pcre_ord2utf8(0x2fff, class_utf8data);
3425 *class_utf8data++ = XCL_RANGE;
3426 class_utf8data += _pcre_ord2utf8(0x3001, class_utf8data);
3427 class_utf8data += _pcre_ord2utf8(0x7fffffff, class_utf8data);
3428 }
3429 #endif
3430 continue;
3431
3432 case ESC_v:
3433 SETBIT(classbits, 0x0a); /* LF */
3434 SETBIT(classbits, 0x0b); /* VT */
3435 SETBIT(classbits, 0x0c); /* FF */
3436 SETBIT(classbits, 0x0d); /* CR */
3437 SETBIT(classbits, 0x85); /* NEL */
3438 #ifdef SUPPORT_UTF8
3439 if (utf8)
3440 {
3441 class_utf8 = TRUE;
3442 *class_utf8data++ = XCL_RANGE;
3443 class_utf8data += _pcre_ord2utf8(0x2028, class_utf8data);
3444 class_utf8data += _pcre_ord2utf8(0x2029, class_utf8data);
3445 }
3446 #endif
3447 continue;
3448
3449 case ESC_V:
3450 for (c = 0; c < 32; c++)
3451 {
3452 int x = 0xff;
3453 switch (c)
3454 {
3455 case 0x0a/8: x ^= 1 << (0x0a%8);
3456 x ^= 1 << (0x0b%8);
3457 x ^= 1 << (0x0c%8);
3458 x ^= 1 << (0x0d%8);
3459 break;
3460 case 0x85/8: x ^= 1 << (0x85%8); break;
3461 default: break;
3462 }
3463 classbits[c] |= x;
3464 }
3465
3466 #ifdef SUPPORT_UTF8
3467 if (utf8)
3468 {
3469 class_utf8 = TRUE;
3470 *class_utf8data++ = XCL_RANGE;
3471 class_utf8data += _pcre_ord2utf8(0x0100, class_utf8data);
3472 class_utf8data += _pcre_ord2utf8(0x2027, class_utf8data);
3473 *class_utf8data++ = XCL_RANGE;
3474 class_utf8data += _pcre_ord2utf8(0x2029, class_utf8data);
3475 class_utf8data += _pcre_ord2utf8(0x7fffffff, class_utf8data);
3476 }
3477 #endif
3478 continue;
3479
3480 #ifdef SUPPORT_UCP
3481 case ESC_p:
3482 case ESC_P:
3483 {
3484 BOOL negated;
3485 int pdata;
3486 int ptype = get_ucp(&ptr, &negated, &pdata, errorcodeptr);
3487 if (ptype < 0) goto FAILED;
3488 class_utf8 = TRUE;
3489 *class_utf8data++ = ((-c == ESC_p) != negated)?
3490 XCL_PROP : XCL_NOTPROP;
3491 *class_utf8data++ = ptype;
3492 *class_utf8data++ = pdata;
3493 class_charcount -= 2; /* Not a < 256 character */
3494 continue;
3495 }
3496 #endif
3497 /* Unrecognized escapes are faulted if PCRE is running in its
3498 strict mode. By default, for compatibility with Perl, they are
3499 treated as literals. */
3500
3501 default:
3502 if ((options & PCRE_EXTRA) != 0)
3503 {
3504 *errorcodeptr = ERR7;
3505 goto FAILED;
3506 }
3507 class_charcount -= 2; /* Undo the default count from above */
3508 c = *ptr; /* Get the final character and fall through */
3509 break;
3510 }
3511 }
3512
3513 /* Fall through if we have a single character (c >= 0). This may be
3514 greater than 256 in UTF-8 mode. */
3515
3516 } /* End of backslash handling */
3517
3518 /* A single character may be followed by '-' to form a range. However,
3519 Perl does not permit ']' to be the end of the range. A '-' character
3520 at the end is treated as a literal. Perl ignores orphaned \E sequences
3521 entirely. The code for handling \Q and \E is messy. */
3522
3523 CHECK_RANGE:
3524 while (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
3525 {
3526 inescq = FALSE;
3527 ptr += 2;
3528 }
3529
3530 oldptr = ptr;
3531
3532 /* Remember \r or \n */
3533
3534 if (c == CHAR_CR || c == CHAR_NL) cd->external_flags |= PCRE_HASCRORLF;
3535
3536 /* Check for range */
3537
3538 if (!inescq && ptr[1] == CHAR_MINUS)
3539 {
3540 int d;
3541 ptr += 2;
3542 while (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_E) ptr += 2;
3543
3544 /* If we hit \Q (not followed by \E) at this point, go into escaped
3545 mode. */
3546
3547 while (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_Q)
3548 {
3549 ptr += 2;
3550 if (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_E)
3551 { ptr += 2; continue; }
3552 inescq = TRUE;
3553 break;
3554 }
3555
3556 if (*ptr == 0 || (!inescq && *ptr == CHAR_RIGHT_SQUARE_BRACKET))
3557 {
3558 ptr = oldptr;
3559 goto LONE_SINGLE_CHARACTER;
3560 }
3561
3562 #ifdef SUPPORT_UTF8
3563 if (utf8)
3564 { /* Braces are required because the */
3565 GETCHARLEN(d, ptr, ptr); /* macro generates multiple statements */
3566 }
3567 else
3568 #endif
3569 d = *ptr; /* Not UTF-8 mode */
3570
3571 /* The second part of a range can be a single-character escape, but
3572 not any of the other escapes. Perl 5.6 treats a hyphen as a literal
3573 in such circumstances. */
3574
3575 if (!inescq && d == CHAR_BACKSLASH)
3576 {
3577 d = check_escape(&ptr, errorcodeptr, cd->bracount, options, TRUE);
3578 if (*errorcodeptr != 0) goto FAILED;
3579
3580 /* \b is backspace; any other special means the '-' was literal */
3581
3582 if (d < 0)
3583 {
3584 if (d == -ESC_b) d = CHAR_BS; else
3585 {
3586 ptr = oldptr;
3587 goto LONE_SINGLE_CHARACTER; /* A few lines below */
3588 }
3589 }
3590 }
3591
3592 /* Check that the two values are in the correct order. Optimize
3593 one-character ranges */
3594
3595 if (d < c)
3596 {
3597 *errorcodeptr = ERR8;
3598 goto FAILED;
3599 }
3600
3601 if (d == c) goto LONE_SINGLE_CHARACTER; /* A few lines below */
3602
3603 /* Remember \r or \n */
3604
3605 if (d == CHAR_CR || d == CHAR_NL) cd->external_flags |= PCRE_HASCRORLF;
3606
3607 /* In UTF-8 mode, if the upper limit is > 255, or > 127 for caseless
3608 matching, we have to use an XCLASS with extra data items. Caseless
3609 matching for characters > 127 is available only if UCP support is
3610 available. */
3611
3612 #ifdef SUPPORT_UTF8
3613 if (utf8 && (d > 255 || ((options & PCRE_CASELESS) != 0 && d > 127)))
3614 {
3615 class_utf8 = TRUE;
3616
3617 /* With UCP support, we can find the other case equivalents of
3618 the relevant characters. There may be several ranges. Optimize how
3619 they fit with the basic range. */
3620
3621 #ifdef SUPPORT_UCP
3622 if ((options & PCRE_CASELESS) != 0)
3623 {
3624 unsigned int occ, ocd;
3625 unsigned int cc = c;
3626 unsigned int origd = d;
3627 while (get_othercase_range(&cc, origd, &occ, &ocd))
3628 {
3629 if (occ >= (unsigned int)c &&
3630 ocd <= (unsigned int)d)
3631 continue; /* Skip embedded ranges */
3632
3633 if (occ < (unsigned int)c &&
3634 ocd >= (unsigned int)c - 1) /* Extend the basic range */
3635 { /* if there is overlap, */
3636 c = occ; /* noting that if occ < c */
3637 continue; /* we can't have ocd > d */
3638 } /* because a subrange is */
3639 if (ocd > (unsigned int)d &&
3640 occ <= (unsigned int)d + 1) /* always shorter than */
3641 { /* the basic range. */
3642 d = ocd;
3643 continue;
3644 }
3645
3646 if (occ == ocd)
3647 {
3648 *class_utf8data++ = XCL_SINGLE;
3649 }
3650 else
3651 {
3652 *class_utf8data++ = XCL_RANGE;
3653 class_utf8data += _pcre_ord2utf8(occ, class_utf8data);
3654 }
3655 class_utf8data += _pcre_ord2utf8(ocd, class_utf8data);
3656 }
3657 }
3658 #endif /* SUPPORT_UCP */
3659
3660 /* Now record the original range, possibly modified for UCP caseless
3661 overlapping ranges. */
3662
3663 *class_utf8data++ = XCL_RANGE;
3664 class_utf8data += _pcre_ord2utf8(c, class_utf8data);
3665 class_utf8data += _pcre_ord2utf8(d, class_utf8data);
3666
3667 /* With UCP support, we are done. Without UCP support, there is no
3668 caseless matching for UTF-8 characters > 127; we can use the bit map
3669 for the smaller ones. */
3670
3671 #ifdef SUPPORT_UCP
3672 continue; /* With next character in the class */
3673 #else
3674 if ((options & PCRE_CASELESS) == 0 || c > 127) continue;
3675
3676 /* Adjust upper limit and fall through to set up the map */
3677
3678 d = 127;
3679
3680 #endif /* SUPPORT_UCP */
3681 }
3682 #endif /* SUPPORT_UTF8 */
3683
3684 /* We use the bit map for all cases when not in UTF-8 mode; else
3685 ranges that lie entirely within 0-127 when there is UCP support; else
3686 for partial ranges without UCP support. */
3687
3688 class_charcount += d - c + 1;
3689 class_lastchar = d;
3690
3691 /* We can save a bit of time by skipping this in the pre-compile. */
3692
3693 if (lengthptr == NULL) for (; c <= d; c++)
3694 {
3695 classbits[c/8] |= (1 << (c&7));
3696 if ((options & PCRE_CASELESS) != 0)
3697 {
3698 int uc = cd->fcc[c]; /* flip case */
3699 classbits[uc/8] |= (1 << (uc&7));
3700 }
3701 }
3702
3703 continue; /* Go get the next char in the class */
3704 }
3705
3706 /* Handle a lone single character - we can get here for a normal
3707 non-escape char, or after \ that introduces a single character or for an
3708 apparent range that isn't. */
3709
3710 LONE_SINGLE_CHARACTER:
3711
3712 /* Handle a character that cannot go in the bit map */
3713
3714 #ifdef SUPPORT_UTF8
3715 if (utf8 && (c > 255 || ((options & PCRE_CASELESS) != 0 && c > 127)))
3716 {
3717 class_utf8 = TRUE;
3718 *class_utf8data++ = XCL_SINGLE;
3719 class_utf8data += _pcre_ord2utf8(c, class_utf8data);
3720
3721 #ifdef SUPPORT_UCP
3722 if ((options & PCRE_CASELESS) != 0)
3723 {
3724 unsigned int othercase;
3725 if ((othercase = UCD_OTHERCASE(c)) != c)
3726 {
3727 *class_utf8data++ = XCL_SINGLE;
3728 class_utf8data += _pcre_ord2utf8(othercase, class_utf8data);
3729 }
3730 }
3731 #endif /* SUPPORT_UCP */
3732
3733 }
3734 else
3735 #endif /* SUPPORT_UTF8 */
3736
3737 /* Handle a single-byte character */
3738 {
3739 classbits[c/8] |= (1 << (c&7));
3740 if ((options & PCRE_CASELESS) != 0)
3741 {
3742 c = cd->fcc[c]; /* flip case */
3743 classbits[c/8] |= (1 << (c&7));
3744 }
3745 class_charcount++;
3746 class_lastchar = c;
3747 }
3748 }
3749
3750 /* Loop until ']' reached. This "while" is the end of the "do" far above.
3751 If we are at the end of an internal nested string, revert to the outer
3752 string. */
3753
3754 while (((c = *(++ptr)) != 0 ||
3755 (nestptr != NULL &&
3756 (ptr = nestptr, nestptr = NULL, c = *(++ptr)) != 0)) &&
3757 (c != CHAR_RIGHT_SQUARE_BRACKET || inescq));
3758
3759 /* Check for missing terminating ']' */
3760
3761 if (c == 0)
3762 {
3763 *errorcodeptr = ERR6;
3764 goto FAILED;
3765 }
3766
3767 /* If class_charcount is 1, we saw precisely one character whose value is
3768 less than 256. As long as there were no characters >= 128 and there was no
3769 use of \p or \P, in other words, no use of any XCLASS features, we can
3770 optimize.
3771
3772 In UTF-8 mode, we can optimize the negative case only if there were no
3773 characters >= 128 because OP_NOT and the related opcodes like OP_NOTSTAR
3774 operate on single-bytes only. This is an historical hangover. Maybe one day
3775 we can tidy these opcodes to handle multi-byte characters.
3776
3777 The optimization throws away the bit map. We turn the item into a
3778 1-character OP_CHAR[NC] if it's positive, or OP_NOT if it's negative. Note
3779 that OP_NOT does not support multibyte characters. In the positive case, it
3780 can cause firstbyte to be set. Otherwise, there can be no first char if
3781 this item is first, whatever repeat count may follow. In the case of
3782 reqbyte, save the previous value for reinstating. */
3783
3784 #ifdef SUPPORT_UTF8
3785 if (class_charcount == 1 && !class_utf8 &&
3786 (!utf8 || !negate_class || class_lastchar < 128))
3787 #else
3788 if (class_charcount == 1)
3789 #endif
3790 {
3791 zeroreqbyte = reqbyte;
3792
3793 /* The OP_NOT opcode works on one-byte characters only. */
3794
3795 if (negate_class)
3796 {
3797 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
3798 zerofirstbyte = firstbyte;
3799 *code++ = OP_NOT;
3800 *code++ = class_lastchar;
3801 break;
3802 }
3803
3804 /* For a single, positive character, get the value into mcbuffer, and
3805 then we can handle this with the normal one-character code. */
3806
3807 #ifdef SUPPORT_UTF8
3808 if (utf8 && class_lastchar > 127)
3809 mclength = _pcre_ord2utf8(class_lastchar, mcbuffer);
3810 else
3811 #endif
3812 {
3813 mcbuffer[0] = class_lastchar;
3814 mclength = 1;
3815 }
3816 goto ONE_CHAR;
3817 } /* End of 1-char optimization */
3818
3819 /* The general case - not the one-char optimization. If this is the first
3820 thing in the branch, there can be no first char setting, whatever the
3821 repeat count. Any reqbyte setting must remain unchanged after any kind of
3822 repeat. */
3823
3824 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
3825 zerofirstbyte = firstbyte;
3826 zeroreqbyte = reqbyte;
3827
3828 /* If there are characters with values > 255, we have to compile an
3829 extended class, with its own opcode, unless there was a negated special
3830 such as \S in the class, and PCRE_UCP is not set, because in that case all
3831 characters > 255 are in the class, so any that were explicitly given as
3832 well can be ignored. If (when there are explicit characters > 255 that must
3833 be listed) there are no characters < 256, we can omit the bitmap in the
3834 actual compiled code. */
3835
3836 #ifdef SUPPORT_UTF8
3837 if (class_utf8 && (!should_flip_negation || (options & PCRE_UCP) != 0))
3838 {
3839 *class_utf8data++ = XCL_END; /* Marks the end of extra data */
3840 *code++ = OP_XCLASS;
3841 code += LINK_SIZE;
3842 *code = negate_class? XCL_NOT : 0;
3843
3844 /* If the map is required, move up the extra data to make room for it;
3845 otherwise just move the code pointer to the end of the extra data. */
3846
3847 if (class_charcount > 0)
3848 {
3849 *code++ |= XCL_MAP;
3850 memmove(code + 32, code, class_utf8data - code);
3851 memcpy(code, classbits, 32);
3852 code = class_utf8data + 32;
3853 }
3854 else code = class_utf8data;
3855
3856 /* Now fill in the complete length of the item */
3857
3858 PUT(previous, 1, code - previous);
3859 break; /* End of class handling */
3860 }
3861 #endif
3862
3863 /* If there are no characters > 255, or they are all to be included or
3864 excluded, set the opcode to OP_CLASS or OP_NCLASS, depending on whether the
3865 whole class was negated and whether there were negative specials such as \S
3866 (non-UCP) in the class. Then copy the 32-byte map into the code vector,
3867 negating it if necessary. */
3868
3869 *code++ = (negate_class == should_flip_negation) ? OP_CLASS : OP_NCLASS;
3870 if (negate_class)
3871 {
3872 if (lengthptr == NULL) /* Save time in the pre-compile phase */
3873 for (c = 0; c < 32; c++) code[c] = ~classbits[c];
3874 }
3875 else
3876 {
3877 memcpy(code, classbits, 32);
3878 }
3879 code += 32;
3880 break;
3881
3882
3883 /* ===================================================================*/
3884 /* Various kinds of repeat; '{' is not necessarily a quantifier, but this
3885 has been tested above. */
3886
3887 case CHAR_LEFT_CURLY_BRACKET:
3888 if (!is_quantifier) goto NORMAL_CHAR;
3889 ptr = read_repeat_counts(ptr+1, &repeat_min, &repeat_max, errorcodeptr);
3890 if (*errorcodeptr != 0) goto FAILED;
3891 goto REPEAT;
3892
3893 case CHAR_ASTERISK:
3894 repeat_min = 0;
3895 repeat_max = -1;
3896 goto REPEAT;
3897
3898 case CHAR_PLUS:
3899 repeat_min = 1;
3900 repeat_max = -1;
3901 goto REPEAT;
3902
3903 case CHAR_QUESTION_MARK:
3904 repeat_min = 0;
3905 repeat_max = 1;
3906
3907 REPEAT:
3908 if (previous == NULL)
3909 {
3910 *errorcodeptr = ERR9;
3911 goto FAILED;
3912 }
3913
3914 if (repeat_min == 0)
3915 {
3916 firstbyte = zerofirstbyte; /* Adjust for zero repeat */
3917 reqbyte = zeroreqbyte; /* Ditto */
3918 }
3919
3920 /* Remember whether this is a variable length repeat */
3921
3922 reqvary = (repeat_min == repeat_max)? 0 : REQ_VARY;
3923
3924 op_type = 0; /* Default single-char op codes */
3925 possessive_quantifier = FALSE; /* Default not possessive quantifier */
3926
3927 /* Save start of previous item, in case we have to move it up to make space
3928 for an inserted OP_ONCE for the additional '+' extension. */
3929
3930 tempcode = previous;
3931
3932 /* If the next character is '+', we have a possessive quantifier. This
3933 implies greediness, whatever the setting of the PCRE_UNGREEDY option.
3934 If the next character is '?' this is a minimizing repeat, by default,
3935 but if PCRE_UNGREEDY is set, it works the other way round. We change the
3936 repeat type to the non-default. */
3937
3938 if (ptr[1] == CHAR_PLUS)
3939 {
3940 repeat_type = 0; /* Force greedy */
3941 possessive_quantifier = TRUE;
3942 ptr++;
3943 }
3944 else if (ptr[1] == CHAR_QUESTION_MARK)
3945 {
3946 repeat_type = greedy_non_default;
3947 ptr++;
3948 }
3949 else repeat_type = greedy_default;
3950
3951 /* If previous was a character match, abolish the item and generate a
3952 repeat item instead. If a char item has a minumum of more than one, ensure
3953 that it is set in reqbyte - it might not be if a sequence such as x{3} is
3954 the first thing in a branch because the x will have gone into firstbyte
3955 instead. */
3956
3957 if (*previous == OP_CHAR || *previous == OP_CHARNC)
3958 {
3959 /* Deal with UTF-8 characters that take up more than one byte. It's
3960 easier to write this out separately than try to macrify it. Use c to
3961 hold the length of the character in bytes, plus 0x80 to flag that it's a
3962 length rather than a small character. */
3963
3964 #ifdef SUPPORT_UTF8
3965 if (utf8 && (code[-1] & 0x80) != 0)
3966 {
3967 uschar *lastchar = code - 1;
3968 while((*lastchar & 0xc0) == 0x80) lastchar--;
3969 c = code - lastchar; /* Length of UTF-8 character */
3970 memcpy(utf8_char, lastchar, c); /* Save the char */
3971 c |= 0x80; /* Flag c as a length */
3972 }
3973 else
3974 #endif
3975
3976 /* Handle the case of a single byte - either with no UTF8 support, or
3977 with UTF-8 disabled, or for a UTF-8 character < 128. */
3978
3979 {
3980 c = code[-1];
3981 if (repeat_min > 1) reqbyte = c | req_caseopt | cd->req_varyopt;
3982 }
3983
3984 /* If the repetition is unlimited, it pays to see if the next thing on
3985 the line is something that cannot possibly match this character. If so,
3986 automatically possessifying this item gains some performance in the case
3987 where the match fails. */
3988
3989 if (!possessive_quantifier &&
3990 repeat_max < 0 &&
3991 check_auto_possessive(*previous, c, utf8, utf8_char, ptr + 1,
3992 options, cd))
3993 {
3994 repeat_type = 0; /* Force greedy */
3995 possessive_quantifier = TRUE;
3996 }
3997
3998 goto OUTPUT_SINGLE_REPEAT; /* Code shared with single character types */
3999 }
4000
4001 /* If previous was a single negated character ([^a] or similar), we use
4002 one of the special opcodes, replacing it. The code is shared with single-
4003 character repeats by setting opt_type to add a suitable offset into
4004 repeat_type. We can also test for auto-possessification. OP_NOT is
4005 currently used only for single-byte chars. */
4006
4007 else if (*previous == OP_NOT)
4008 {
4009 op_type = OP_NOTSTAR - OP_STAR; /* Use "not" opcodes */
4010 c = previous[1];
4011 if (!possessive_quantifier &&
4012 repeat_max < 0 &&
4013 check_auto_possessive(OP_NOT, c, utf8, NULL, ptr + 1, options, cd))
4014 {
4015 repeat_type = 0; /* Force greedy */
4016 possessive_quantifier = TRUE;
4017 }
4018 goto OUTPUT_SINGLE_REPEAT;
4019 }
4020
4021 /* If previous was a character type match (\d or similar), abolish it and
4022 create a suitable repeat item. The code is shared with single-character
4023 repeats by setting op_type to add a suitable offset into repeat_type. Note
4024 the the Unicode property types will be present only when SUPPORT_UCP is
4025 defined, but we don't wrap the little bits of code here because it just
4026 makes it horribly messy. */
4027
4028 else if (*previous < OP_EODN)
4029 {
4030 uschar *oldcode;
4031 int prop_type, prop_value;
4032 op_type = OP_TYPESTAR - OP_STAR; /* Use type opcodes */
4033 c = *previous;
4034
4035 if (!possessive_quantifier &&
4036 repeat_max < 0 &&
4037 check_auto_possessive(c, 0, utf8, NULL, ptr + 1, options, cd))
4038 {
4039 repeat_type = 0; /* Force greedy */
4040 possessive_quantifier = TRUE;
4041 }
4042
4043 OUTPUT_SINGLE_REPEAT:
4044 if (*previous == OP_PROP || *previous == OP_NOTPROP)
4045 {
4046 prop_type = previous[1];
4047 prop_value = previous[2];
4048 }
4049 else prop_type = prop_value = -1;
4050
4051 oldcode = code;
4052 code = previous; /* Usually overwrite previous item */
4053
4054 /* If the maximum is zero then the minimum must also be zero; Perl allows
4055 this case, so we do too - by simply omitting the item altogether. */
4056
4057 if (repeat_max == 0) goto END_REPEAT;
4058
4059 /*--------------------------------------------------------------------*/
4060 /* This code is obsolete from release 8.00; the restriction was finally
4061 removed: */
4062
4063 /* All real repeats make it impossible to handle partial matching (maybe
4064 one day we will be able to remove this restriction). */
4065
4066 /* if (repeat_max != 1) cd->external_flags |= PCRE_NOPARTIAL; */
4067 /*--------------------------------------------------------------------*/
4068
4069 /* Combine the op_type with the repeat_type */
4070
4071 repeat_type += op_type;
4072
4073 /* A minimum of zero is handled either as the special case * or ?, or as
4074 an UPTO, with the maximum given. */
4075
4076 if (repeat_min == 0)
4077 {
4078 if (repeat_max == -1) *code++ = OP_STAR + repeat_type;
4079 else if (repeat_max == 1) *code++ = OP_QUERY + repeat_type;
4080 else
4081 {
4082 *code++ = OP_UPTO + repeat_type;
4083 PUT2INC(code, 0, repeat_max);
4084 }
4085 }
4086
4087 /* A repeat minimum of 1 is optimized into some special cases. If the
4088 maximum is unlimited, we use OP_PLUS. Otherwise, the original item is
4089 left in place and, if the maximum is greater than 1, we use OP_UPTO with
4090 one less than the maximum. */
4091
4092 else if (repeat_min == 1)
4093 {
4094 if (repeat_max == -1)
4095 *code++ = OP_PLUS + repeat_type;
4096 else
4097 {
4098 code = oldcode; /* leave previous item in place */
4099 if (repeat_max == 1) goto END_REPEAT;
4100 *code++ = OP_UPTO + repeat_type;
4101 PUT2INC(code, 0, repeat_max - 1);
4102 }
4103 }
4104
4105 /* The case {n,n} is just an EXACT, while the general case {n,m} is
4106 handled as an EXACT followed by an UPTO. */
4107
4108 else
4109 {
4110 *code++ = OP_EXACT + op_type; /* NB EXACT doesn't have repeat_type */
4111 PUT2INC(code, 0, repeat_min);
4112
4113 /* If the maximum is unlimited, insert an OP_STAR. Before doing so,
4114 we have to insert the character for the previous code. For a repeated
4115 Unicode property match, there are two extra bytes that define the
4116 required property. In UTF-8 mode, long characters have their length in
4117 c, with the 0x80 bit as a flag. */
4118
4119 if (repeat_max < 0)
4120 {
4121 #ifdef SUPPORT_UTF8
4122 if (utf8 && c >= 128)
4123 {
4124 memcpy(code, utf8_char, c & 7);
4125 code += c & 7;
4126 }
4127 else
4128 #endif
4129 {
4130 *code++ = c;
4131 if (prop_type >= 0)
4132 {
4133 *code++ = prop_type;
4134 *code++ = prop_value;
4135 }
4136 }
4137 *code++ = OP_STAR + repeat_type;
4138 }
4139
4140 /* Else insert an UPTO if the max is greater than the min, again
4141 preceded by the character, for the previously inserted code. If the
4142 UPTO is just for 1 instance, we can use QUERY instead. */
4143
4144 else if (repeat_max != repeat_min)
4145 {
4146 #ifdef SUPPORT_UTF8
4147 if (utf8 && c >= 128)
4148 {
4149 memcpy(code, utf8_char, c & 7);
4150 code += c & 7;
4151 }
4152 else
4153 #endif
4154 *code++ = c;
4155 if (prop_type >= 0)
4156 {
4157 *code++ = prop_type;
4158 *code++ = prop_value;
4159 }
4160 repeat_max -= repeat_min;
4161
4162 if (repeat_max == 1)
4163 {
4164 *code++ = OP_QUERY + repeat_type;
4165 }
4166 else
4167 {
4168 *code++ = OP_UPTO + repeat_type;
4169 PUT2INC(code, 0, repeat_max);
4170 }
4171 }
4172 }
4173
4174 /* The character or character type itself comes last in all cases. */
4175
4176 #ifdef SUPPORT_UTF8
4177 if (utf8 && c >= 128)
4178 {
4179 memcpy(code, utf8_char, c & 7);
4180 code += c & 7;
4181 }
4182 else
4183 #endif
4184 *code++ = c;
4185
4186 /* For a repeated Unicode property match, there are two extra bytes that
4187 define the required property. */
4188
4189 #ifdef SUPPORT_UCP
4190 if (prop_type >= 0)
4191 {
4192 *code++ = prop_type;
4193 *code++ = prop_value;
4194 }
4195 #endif
4196 }
4197
4198 /* If previous was a character class or a back reference, we put the repeat
4199 stuff after it, but just skip the item if the repeat was {0,0}. */
4200
4201 else if (*previous == OP_CLASS ||
4202 *previous == OP_NCLASS ||
4203 #ifdef SUPPORT_UTF8
4204 *previous == OP_XCLASS ||
4205 #endif
4206 *previous == OP_REF)
4207 {
4208 if (repeat_max == 0)
4209 {
4210 code = previous;
4211 goto END_REPEAT;
4212 }
4213
4214 /*--------------------------------------------------------------------*/
4215 /* This code is obsolete from release 8.00; the restriction was finally
4216 removed: */
4217
4218 /* All real repeats make it impossible to handle partial matching (maybe
4219 one day we will be able to remove this restriction). */
4220
4221 /* if (repeat_max != 1) cd->external_flags |= PCRE_NOPARTIAL; */
4222 /*--------------------------------------------------------------------*/
4223
4224 if (repeat_min == 0 && repeat_max == -1)
4225 *code++ = OP_CRSTAR + repeat_type;
4226 else if (repeat_min == 1 && repeat_max == -1)
4227 *code++ = OP_CRPLUS + repeat_type;
4228 else if (repeat_min == 0 && repeat_max == 1)
4229 *code++ = OP_CRQUERY + repeat_type;
4230 else
4231 {
4232 *code++ = OP_CRRANGE + repeat_type;
4233 PUT2INC(code, 0, repeat_min);
4234 if (repeat_max == -1) repeat_max = 0; /* 2-byte encoding for max */
4235 PUT2INC(code, 0, repeat_max);
4236 }
4237 }
4238
4239 /* If previous was a bracket group, we may have to replicate it in certain
4240 cases. */
4241
4242 else if (*previous == OP_BRA || *previous == OP_CBRA ||
4243 *previous == OP_ONCE || *previous == OP_COND)
4244 {
4245 register int i;
4246 int ketoffset = 0;
4247 int len = code - previous;
4248 uschar *bralink = NULL;
4249
4250 /* Repeating a DEFINE group is pointless */
4251
4252 if (*previous == OP_COND && previous[LINK_SIZE+1] == OP_DEF)
4253 {
4254 *errorcodeptr = ERR55;
4255 goto FAILED;
4256 }
4257
4258 /* If the maximum repeat count is unlimited, find the end of the bracket
4259 by scanning through from the start, and compute the offset back to it
4260 from the current code pointer. There may be an OP_OPT setting following
4261 the final KET, so we can't find the end just by going back from the code
4262 pointer. */
4263
4264 if (repeat_max == -1)
4265 {
4266 register uschar *ket = previous;
4267 do ket += GET(ket, 1); while (*ket != OP_KET);
4268 ketoffset = code - ket;
4269 }
4270
4271 /* The case of a zero minimum is special because of the need to stick
4272 OP_BRAZERO in front of it, and because the group appears once in the
4273 data, whereas in other cases it appears the minimum number of times. For
4274 this reason, it is simplest to treat this case separately, as otherwise
4275 the code gets far too messy. There are several special subcases when the
4276 minimum is zero. */
4277
4278 if (repeat_min == 0)
4279 {
4280 /* If the maximum is also zero, we used to just omit the group from the
4281 output altogether, like this:
4282
4283 ** if (repeat_max == 0)
4284 ** {
4285 ** code = previous;
4286 ** goto END_REPEAT;
4287 ** }
4288
4289 However, that fails when a group is referenced as a subroutine from
4290 elsewhere in the pattern, so now we stick in OP_SKIPZERO in front of it
4291 so that it is skipped on execution. As we don't have a list of which
4292 groups are referenced, we cannot do this selectively.
4293
4294 If the maximum is 1 or unlimited, we just have to stick in the BRAZERO
4295 and do no more at this point. However, we do need to adjust any
4296 OP_RECURSE calls inside the group that refer to the group itself or any
4297 internal or forward referenced group, because the offset is from the
4298 start of the whole regex. Temporarily terminate the pattern while doing
4299 this. */
4300
4301 if (repeat_max <= 1) /* Covers 0, 1, and unlimited */
4302 {
4303 *code = OP_END;
4304 adjust_recurse(previous, 1, utf8, cd, save_hwm);
4305 memmove(previous+1, previous, len);
4306 code++;
4307 if (repeat_max == 0)
4308 {
4309 *previous++ = OP_SKIPZERO;
4310 goto END_REPEAT;
4311 }
4312 *previous++ = OP_BRAZERO + repeat_type;
4313 }
4314
4315 /* If the maximum is greater than 1 and limited, we have to replicate
4316 in a nested fashion, sticking OP_BRAZERO before each set of brackets.
4317 The first one has to be handled carefully because it's the original
4318 copy, which has to be moved up. The remainder can be handled by code
4319 that is common with the non-zero minimum case below. We have to
4320 adjust the value or repeat_max, since one less copy is required. Once
4321 again, we may have to adjust any OP_RECURSE calls inside the group. */
4322
4323 else
4324 {
4325 int offset;
4326 *code = OP_END;
4327 adjust_recurse(previous, 2 + LINK_SIZE, utf8, cd, save_hwm);
4328 memmove(previous + 2 + LINK_SIZE, previous, len);
4329 code += 2 + LINK_SIZE;
4330 *previous++ = OP_BRAZERO + repeat_type;
4331 *previous++ = OP_BRA;
4332
4333 /* We chain together the bracket offset fields that have to be
4334 filled in later when the ends of the brackets are reached. */
4335
4336 offset = (bralink == NULL)? 0 : previous - bralink;
4337 bralink = previous;
4338 PUTINC(previous, 0, offset);
4339 }
4340
4341 repeat_max--;
4342 }
4343
4344 /* If the minimum is greater than zero, replicate the group as many
4345 times as necessary, and adjust the maximum to the number of subsequent
4346 copies that we need. If we set a first char from the group, and didn't
4347 set a required char, copy the latter from the former. If there are any
4348 forward reference subroutine calls in the group, there will be entries on
4349 the workspace list; replicate these with an appropriate increment. */
4350
4351 else
4352 {
4353 if (repeat_min > 1)
4354 {
4355 /* In the pre-compile phase, we don't actually do the replication. We
4356 just adjust the length as if we had. Do some paranoid checks for
4357 potential integer overflow. The INT64_OR_DOUBLE type is a 64-bit
4358 integer type when available, otherwise double. */
4359
4360 if (lengthptr != NULL)
4361 {
4362 int delta = (repeat_min - 1)*length_prevgroup;
4363 if ((INT64_OR_DOUBLE)(repeat_min - 1)*
4364 (INT64_OR_DOUBLE)length_prevgroup >
4365 (INT64_OR_DOUBLE)INT_MAX ||
4366 OFLOW_MAX - *lengthptr < delta)
4367 {
4368 *errorcodeptr = ERR20;
4369 goto FAILED;
4370 }
4371 *lengthptr += delta;
4372 }
4373
4374 /* This is compiling for real */
4375
4376 else
4377 {
4378 if (groupsetfirstbyte && reqbyte < 0) reqbyte = firstbyte;
4379 for (i = 1; i < repeat_min; i++)
4380 {
4381 uschar *hc;
4382 uschar *this_hwm = cd->hwm;
4383 memcpy(code, previous, len);
4384 for (hc = save_hwm; hc < this_hwm; hc += LINK_SIZE)
4385 {
4386 PUT(cd->hwm, 0, GET(hc, 0) + len);
4387 cd->hwm += LINK_SIZE;
4388 }
4389 save_hwm = this_hwm;
4390 code += len;
4391 }
4392 }
4393 }
4394
4395 if (repeat_max > 0) repeat_max -= repeat_min;
4396 }
4397
4398 /* This code is common to both the zero and non-zero minimum cases. If
4399 the maximum is limited, it replicates the group in a nested fashion,
4400 remembering the bracket starts on a stack. In the case of a zero minimum,
4401 the first one was set up above. In all cases the repeat_max now specifies
4402 the number of additional copies needed. Again, we must remember to
4403 replicate entries on the forward reference list. */
4404
4405 if (repeat_max >= 0)
4406 {
4407 /* In the pre-compile phase, we don't actually do the replication. We
4408 just adjust the length as if we had. For each repetition we must add 1
4409 to the length for BRAZERO and for all but the last repetition we must
4410 add 2 + 2*LINKSIZE to allow for the nesting that occurs. Do some
4411 paranoid checks to avoid integer overflow. The INT64_OR_DOUBLE type is
4412 a 64-bit integer type when available, otherwise double. */
4413
4414 if (lengthptr != NULL && repeat_max > 0)
4415 {
4416 int delta = repeat_max * (length_prevgroup + 1 + 2 + 2*LINK_SIZE) -
4417 2 - 2*LINK_SIZE; /* Last one doesn't nest */
4418 if ((INT64_OR_DOUBLE)repeat_max *
4419 (INT64_OR_DOUBLE)(length_prevgroup + 1 + 2 + 2*LINK_SIZE)
4420 > (INT64_OR_DOUBLE)INT_MAX ||
4421 OFLOW_MAX - *lengthptr < delta)
4422 {
4423 *errorcodeptr = ERR20;
4424 goto FAILED;
4425 }
4426 *lengthptr += delta;
4427 }
4428
4429 /* This is compiling for real */
4430
4431 else for (i = repeat_max - 1; i >= 0; i--)
4432 {
4433 uschar *hc;
4434 uschar *this_hwm = cd->hwm;
4435
4436 *code++ = OP_BRAZERO + repeat_type;
4437
4438 /* All but the final copy start a new nesting, maintaining the
4439 chain of brackets outstanding. */
4440
4441 if (i != 0)
4442 {
4443 int offset;
4444 *code++ = OP_BRA;
4445 offset = (bralink == NULL)? 0 : code - bralink;
4446 bralink = code;
4447 PUTINC(code, 0, offset);
4448 }
4449
4450 memcpy(code, previous, len);
4451 for (hc = save_hwm; hc < this_hwm; hc += LINK_SIZE)
4452 {
4453 PUT(cd->hwm, 0, GET(hc, 0) + len + ((i != 0)? 2+LINK_SIZE : 1));
4454 cd->hwm += LINK_SIZE;
4455 }
4456 save_hwm = this_hwm;
4457 code += len;
4458 }
4459
4460 /* Now chain through the pending brackets, and fill in their length
4461 fields (which are holding the chain links pro tem). */
4462
4463 while (bralink != NULL)
4464 {
4465 int oldlinkoffset;
4466 int offset = code - bralink + 1;
4467 uschar *bra = code - offset;
4468 oldlinkoffset = GET(bra, 1);
4469 bralink = (oldlinkoffset == 0)? NULL : bralink - oldlinkoffset;
4470 *code++ = OP_KET;
4471 PUTINC(code, 0, offset);
4472 PUT(bra, 1, offset);
4473 }
4474 }
4475
4476 /* If the maximum is unlimited, set a repeater in the final copy. We
4477 can't just offset backwards from the current code point, because we
4478 don't know if there's been an options resetting after the ket. The
4479 correct offset was computed above.
4480
4481 Then, when we are doing the actual compile phase, check to see whether
4482 this group is a non-atomic one that could match an empty string. If so,
4483 convert the initial operator to the S form (e.g. OP_BRA -> OP_SBRA) so
4484 that runtime checking can be done. [This check is also applied to
4485 atomic groups at runtime, but in a different way.] */
4486
4487 else
4488 {
4489 uschar *ketcode = code - ketoffset;
4490 uschar *bracode = ketcode - GET(ketcode, 1);
4491 *ketcode = OP_KETRMAX + repeat_type;
4492 if (lengthptr == NULL && *bracode != OP_ONCE)
4493 {
4494 uschar *scode = bracode;
4495 do
4496 {
4497 if (could_be_empty_branch(scode, ketcode, utf8, cd))
4498 {
4499 *bracode += OP_SBRA - OP_BRA;
4500 break;
4501 }
4502 scode += GET(scode, 1);
4503 }
4504 while (*scode == OP_ALT);
4505 }
4506 }
4507 }
4508
4509 /* If previous is OP_FAIL, it was generated by an empty class [] in
4510 JavaScript mode. The other ways in which OP_FAIL can be generated, that is
4511 by (*FAIL) or (?!) set previous to NULL, which gives a "nothing to repeat"
4512 error above. We can just ignore the repeat in JS case. */
4513
4514 else if (*previous == OP_FAIL) goto END_REPEAT;
4515
4516 /* Else there's some kind of shambles */
4517
4518 else
4519 {
4520 *errorcodeptr = ERR11;
4521 goto FAILED;
4522 }
4523
4524 /* If the character following a repeat is '+', or if certain optimization
4525 tests above succeeded, possessive_quantifier is TRUE. For some of the
4526 simpler opcodes, there is an special alternative opcode for this. For
4527 anything else, we wrap the entire repeated item inside OP_ONCE brackets.
4528 The '+' notation is just syntactic sugar, taken from Sun's Java package,
4529 but the special opcodes can optimize it a bit. The repeated item starts at
4530 tempcode, not at previous, which might be the first part of a string whose
4531 (former) last char we repeated.
4532
4533 Possessifying an 'exact' quantifier has no effect, so we can ignore it. But
4534 an 'upto' may follow. We skip over an 'exact' item, and then test the
4535 length of what remains before proceeding. */
4536
4537 if (possessive_quantifier)
4538 {
4539 int len;
4540
4541 if (*tempcode == OP_TYPEEXACT)
4542 tempcode += _pcre_OP_lengths[*tempcode] +
4543 ((tempcode[3] == OP_PROP || tempcode[3] == OP_NOTPROP)? 2 : 0);
4544
4545 else if (*tempcode == OP_EXACT || *tempcode == OP_NOTEXACT)
4546 {
4547 tempcode += _pcre_OP_lengths[*tempcode];
4548 #ifdef SUPPORT_UTF8
4549 if (utf8 && tempcode[-1] >= 0xc0)
4550 tempcode += _pcre_utf8_table4[tempcode[-1] & 0x3f];
4551 #endif
4552 }
4553
4554 len = code - tempcode;
4555 if (len > 0) switch (*tempcode)
4556 {
4557 case OP_STAR: *tempcode = OP_POSSTAR; break;
4558 case OP_PLUS: *tempcode = OP_POSPLUS; break;
4559 case OP_QUERY: *tempcode = OP_POSQUERY; break;
4560 case OP_UPTO: *tempcode = OP_POSUPTO; break;
4561
4562 case OP_TYPESTAR: *tempcode = OP_TYPEPOSSTAR; break;
4563 case OP_TYPEPLUS: *tempcode = OP_TYPEPOSPLUS; break;
4564 case OP_TYPEQUERY: *tempcode = OP_TYPEPOSQUERY; break;
4565 case OP_TYPEUPTO: *tempcode = OP_TYPEPOSUPTO; break;
4566
4567 case OP_NOTSTAR: *tempcode = OP_NOTPOSSTAR; break;
4568 case OP_NOTPLUS: *tempcode = OP_NOTPOSPLUS; break;
4569 case OP_NOTQUERY: *tempcode = OP_NOTPOSQUERY; break;
4570 case OP_NOTUPTO: *tempcode = OP_NOTPOSUPTO; break;
4571
4572 /* Because we are moving code along, we must ensure that any
4573 pending recursive references are updated. */
4574
4575 default:
4576 *code = OP_END;
4577 adjust_recurse(tempcode, 1 + LINK_SIZE, utf8, cd, save_hwm);
4578 memmove(tempcode + 1+LINK_SIZE, tempcode, len);
4579 code += 1 + LINK_SIZE;
4580 len += 1 + LINK_SIZE;
4581 tempcode[0] = OP_ONCE;
4582 *code++ = OP_KET;
4583 PUTINC(code, 0, len);
4584 PUT(tempcode, 1, len);
4585 break;
4586 }
4587 }
4588
4589 /* In all case we no longer have a previous item. We also set the
4590 "follows varying string" flag for subsequently encountered reqbytes if
4591 it isn't already set and we have just passed a varying length item. */
4592
4593 END_REPEAT:
4594 previous = NULL;
4595 cd->req_varyopt |= reqvary;
4596 break;
4597
4598
4599 /* ===================================================================*/
4600 /* Start of nested parenthesized sub-expression, or comment or lookahead or
4601 lookbehind or option setting or condition or all the other extended
4602 parenthesis forms. */
4603
4604 case CHAR_LEFT_PARENTHESIS:
4605 newoptions = options;
4606 skipbytes = 0;
4607 bravalue = OP_CBRA;
4608 save_hwm = cd->hwm;
4609 reset_bracount = FALSE;
4610
4611 /* First deal with various "verbs" that can be introduced by '*'. */
4612
4613 if (*(++ptr) == CHAR_ASTERISK &&
4614 ((cd->ctypes[ptr[1]] & ctype_letter) != 0 || ptr[1] == ':'))
4615 {
4616 int i, namelen;
4617 int arglen = 0;
4618 const char *vn = verbnames;
4619 const uschar *name = ptr + 1;
4620 const uschar *arg = NULL;
4621 previous = NULL;
4622 while ((cd->ctypes[*++ptr] & ctype_letter) != 0) {};
4623 namelen = ptr - name;
4624
4625 if (*ptr == CHAR_COLON)
4626 {
4627 arg = ++ptr;
4628 while ((cd->ctypes[*ptr] & (ctype_letter|ctype_digit)) != 0
4629 || *ptr == '_') ptr++;
4630 arglen = ptr - arg;
4631 }
4632
4633 if (*ptr != CHAR_RIGHT_PARENTHESIS)
4634 {
4635 *errorcodeptr = ERR60;
4636 goto FAILED;
4637 }
4638
4639 /* Scan the table of verb names */
4640
4641 for (i = 0; i < verbcount; i++)
4642 {
4643 if (namelen == verbs[i].len &&
4644 strncmp((char *)name, vn, namelen) == 0)
4645 {
4646 /* Check for open captures before ACCEPT */
4647
4648 if (verbs[i].op == OP_ACCEPT)
4649 {
4650 open_capitem *oc;
4651 cd->had_accept = TRUE;
4652 for (oc = cd->open_caps; oc != NULL; oc = oc->next)
4653 {
4654 *code++ = OP_CLOSE;
4655 PUT2INC(code, 0, oc->number);
4656 }
4657 }
4658
4659 /* Handle the cases with/without an argument */
4660
4661 if (arglen == 0)
4662 {
4663 if (verbs[i].op < 0) /* Argument is mandatory */
4664 {
4665 *errorcodeptr = ERR66;
4666 goto FAILED;
4667 }
4668 *code++ = verbs[i].op;
4669 }
4670
4671 else
4672 {
4673 if (verbs[i].op_arg < 0) /* Argument is forbidden */
4674 {
4675 *errorcodeptr = ERR59;
4676 goto FAILED;
4677 }
4678 *code++ = verbs[i].op_arg;
4679 *code++ = arglen;
4680 memcpy(code, arg, arglen);
4681 code += arglen;
4682 *code++ = 0;
4683 }
4684
4685 break; /* Found verb, exit loop */
4686 }
4687
4688 vn += verbs[i].len + 1;
4689 }
4690
4691 if (i < verbcount) continue; /* Successfully handled a verb */
4692 *errorcodeptr = ERR60; /* Verb not recognized */
4693 goto FAILED;
4694 }
4695
4696 /* Deal with the extended parentheses; all are introduced by '?', and the
4697 appearance of any of them means that this is not a capturing group. */
4698
4699 else if (*ptr == CHAR_QUESTION_MARK)
4700 {
4701 int i, set, unset, namelen;
4702 int *optset;
4703 const uschar *name;
4704 uschar *slot;
4705
4706 switch (*(++ptr))
4707 {
4708 case CHAR_NUMBER_SIGN: /* Comment; skip to ket */
4709 ptr++;
4710 while (*ptr != 0 && *ptr != CHAR_RIGHT_PARENTHESIS) ptr++;
4711 if (*ptr == 0)
4712 {
4713 *errorcodeptr = ERR18;
4714 goto FAILED;
4715 }
4716 continue;
4717
4718
4719 /* ------------------------------------------------------------ */
4720 case CHAR_VERTICAL_LINE: /* Reset capture count for each branch */
4721 reset_bracount = TRUE;
4722 /* Fall through */
4723
4724 /* ------------------------------------------------------------ */
4725 case CHAR_COLON: /* Non-capturing bracket */
4726 bravalue = OP_BRA;
4727 ptr++;
4728 break;
4729
4730
4731 /* ------------------------------------------------------------ */
4732 case CHAR_LEFT_PARENTHESIS:
4733 bravalue = OP_COND; /* Conditional group */
4734
4735 /* A condition can be an assertion, a number (referring to a numbered
4736 group), a name (referring to a named group), or 'R', referring to
4737 recursion. R<digits> and R&name are also permitted for recursion tests.
4738
4739 There are several syntaxes for testing a named group: (?(name)) is used
4740 by Python; Perl 5.10 onwards uses (?(<name>) or (?('name')).
4741
4742 There are two unfortunate ambiguities, caused by history. (a) 'R' can
4743 be the recursive thing or the name 'R' (and similarly for 'R' followed
4744 by digits), and (b) a number could be a name that consists of digits.
4745 In both cases, we look for a name first; if not found, we try the other
4746 cases. */
4747
4748 /* For conditions that are assertions, check the syntax, and then exit
4749 the switch. This will take control down to where bracketed groups,
4750 including assertions, are processed. */
4751
4752 if (ptr[1] == CHAR_QUESTION_MARK && (ptr[2] == CHAR_EQUALS_SIGN ||
4753 ptr[2] == CHAR_EXCLAMATION_MARK || ptr[2] == CHAR_LESS_THAN_SIGN))
4754 break;
4755
4756 /* Most other conditions use OP_CREF (a couple change to OP_RREF
4757 below), and all need to skip 3 bytes at the start of the group. */
4758
4759 code[1+LINK_SIZE] = OP_CREF;
4760 skipbytes = 3;
4761 refsign = -1;
4762
4763 /* Check for a test for recursion in a named group. */
4764
4765 if (ptr[1] == CHAR_R && ptr[2] == CHAR_AMPERSAND)
4766 {
4767 terminator = -1;
4768 ptr += 2;
4769 code[1+LINK_SIZE] = OP_RREF; /* Change the type of test */
4770 }
4771
4772 /* Check for a test for a named group's having been set, using the Perl
4773 syntax (?(<name>) or (?('name') */
4774
4775 else if (ptr[1] == CHAR_LESS_THAN_SIGN)
4776 {
4777 terminator = CHAR_GREATER_THAN_SIGN;
4778 ptr++;
4779 }
4780 else if (ptr[1] == CHAR_APOSTROPHE)
4781 {
4782 terminator = CHAR_APOSTROPHE;
4783 ptr++;
4784 }
4785 else
4786 {
4787 terminator = 0;
4788 if (ptr[1] == CHAR_MINUS || ptr[1] == CHAR_PLUS) refsign = *(++ptr);
4789 }
4790
4791 /* We now expect to read a name; any thing else is an error */
4792
4793 if ((cd->ctypes[ptr[1]] & ctype_word) == 0)
4794 {
4795 ptr += 1; /* To get the right offset */
4796 *errorcodeptr = ERR28;
4797 goto FAILED;
4798 }
4799
4800 /* Read the name, but also get it as a number if it's all digits */
4801
4802 recno = 0;
4803 name = ++ptr;
4804 while ((cd->ctypes[*ptr] & ctype_word) != 0)
4805 {
4806 if (recno >= 0)
4807 recno = ((digitab[*ptr] & ctype_digit) != 0)?
4808 recno * 10 + *ptr - CHAR_0 : -1;
4809 ptr++;
4810 }
4811 namelen = ptr - name;
4812
4813 if ((terminator > 0 && *ptr++ != terminator) ||
4814 *ptr++ != CHAR_RIGHT_PARENTHESIS)
4815 {
4816 ptr--; /* Error offset */
4817 *errorcodeptr = ERR26;
4818 goto FAILED;
4819 }
4820
4821 /* Do no further checking in the pre-compile phase. */
4822
4823 if (lengthptr != NULL) break;
4824
4825 /* In the real compile we do the work of looking for the actual
4826 reference. If the string started with "+" or "-" we require the rest to
4827 be digits, in which case recno will be set. */
4828
4829 if (refsign > 0)
4830 {
4831 if (recno <= 0)
4832 {
4833 *errorcodeptr = ERR58;
4834 goto FAILED;
4835 }
4836 recno = (refsign == CHAR_MINUS)?
4837 cd->bracount - recno + 1 : recno +cd->bracount;
4838 if (recno <= 0 || recno > cd->final_bracount)
4839 {
4840 *errorcodeptr = ERR15;
4841 goto FAILED;
4842 }
4843 PUT2(code, 2+LINK_SIZE, recno);
4844 break;
4845 }
4846
4847 /* Otherwise (did not start with "+" or "-"), start by looking for the
4848 name. If we find a name, add one to the opcode to change OP_CREF or
4849 OP_RREF into OP_NCREF or OP_NRREF. These behave exactly the same,
4850 except they record that the reference was originally to a name. The
4851 information is used to check duplicate names. */
4852
4853 slot = cd->name_table;
4854 for (i = 0; i < cd->names_found; i++)
4855 {
4856 if (strncmp((char *)name, (char *)slot+2, namelen) == 0) break;
4857 slot += cd->name_entry_size;
4858 }
4859
4860 /* Found a previous named subpattern */
4861
4862 if (i < cd->names_found)
4863 {
4864 recno = GET2(slot, 0);
4865 PUT2(code, 2+LINK_SIZE, recno);
4866 code[1+LINK_SIZE]++;
4867 }
4868
4869 /* Search the pattern for a forward reference */
4870
4871 else if ((i = find_parens(cd, name, namelen,
4872 (options & PCRE_EXTENDED) != 0)) > 0)
4873 {
4874 PUT2(code, 2+LINK_SIZE, i);
4875 code[1+LINK_SIZE]++;
4876 }
4877
4878 /* If terminator == 0 it means that the name followed directly after
4879 the opening parenthesis [e.g. (?(abc)...] and in this case there are
4880 some further alternatives to try. For the cases where terminator != 0
4881 [things like (?(<name>... or (?('name')... or (?(R&name)... ] we have
4882 now checked all the possibilities, so give an error. */
4883
4884 else if (terminator != 0)
4885 {
4886 *errorcodeptr = ERR15;
4887 goto FAILED;
4888 }
4889
4890 /* Check for (?(R) for recursion. Allow digits after R to specify a
4891 specific group number. */
4892
4893 else if (*name == CHAR_R)
4894 {
4895 recno = 0;
4896 for (i = 1; i < namelen; i++)
4897 {
4898 if ((digitab[name[i]] & ctype_digit) == 0)
4899 {
4900 *errorcodeptr = ERR15;
4901 goto FAILED;
4902 }
4903 recno = recno * 10 + name[i] - CHAR_0;
4904 }
4905 if (recno == 0) recno = RREF_ANY;
4906 code[1+LINK_SIZE] = OP_RREF; /* Change test type */
4907 PUT2(code, 2+LINK_SIZE, recno);
4908 }
4909
4910 /* Similarly, check for the (?(DEFINE) "condition", which is always
4911 false. */
4912
4913 else if (namelen == 6 && strncmp((char *)name, STRING_DEFINE, 6) == 0)
4914 {
4915 code[1+LINK_SIZE] = OP_DEF;
4916 skipbytes = 1;
4917 }
4918
4919 /* Check for the "name" actually being a subpattern number. We are
4920 in the second pass here, so final_bracount is set. */
4921
4922 else if (recno > 0 && recno <= cd->final_bracount)
4923 {
4924 PUT2(code, 2+LINK_SIZE, recno);
4925 }
4926
4927 /* Either an unidentified subpattern, or a reference to (?(0) */
4928
4929 else
4930 {
4931 *errorcodeptr = (recno == 0)? ERR35: ERR15;
4932 goto FAILED;
4933 }
4934 break;
4935
4936
4937 /* ------------------------------------------------------------ */
4938 case CHAR_EQUALS_SIGN: /* Positive lookahead */
4939 bravalue = OP_ASSERT;
4940 ptr++;
4941 break;
4942
4943
4944 /* ------------------------------------------------------------ */
4945 case CHAR_EXCLAMATION_MARK: /* Negative lookahead */
4946 ptr++;
4947 if (*ptr == CHAR_RIGHT_PARENTHESIS) /* Optimize (?!) */
4948 {
4949 *code++ = OP_FAIL;
4950 previous = NULL;
4951 continue;
4952 }
4953 bravalue = OP_ASSERT_NOT;
4954 break;
4955
4956
4957 /* ------------------------------------------------------------ */
4958 case CHAR_LESS_THAN_SIGN: /* Lookbehind or named define */
4959 switch (ptr[1])
4960 {
4961 case CHAR_EQUALS_SIGN: /* Positive lookbehind */
4962 bravalue = OP_ASSERTBACK;
4963 ptr += 2;
4964 break;
4965
4966 case CHAR_EXCLAMATION_MARK: /* Negative lookbehind */
4967 bravalue = OP_ASSERTBACK_NOT;
4968 ptr += 2;
4969 break;
4970
4971 default: /* Could be name define, else bad */
4972 if ((cd->ctypes[ptr[1]] & ctype_word) != 0) goto DEFINE_NAME;
4973 ptr++; /* Correct offset for error */
4974 *errorcodeptr = ERR24;
4975 goto FAILED;
4976 }
4977 break;
4978
4979
4980 /* ------------------------------------------------------------ */
4981 case CHAR_GREATER_THAN_SIGN: /* One-time brackets */
4982 bravalue = OP_ONCE;
4983 ptr++;
4984 break;
4985
4986
4987 /* ------------------------------------------------------------ */
4988 case CHAR_C: /* Callout - may be followed by digits; */
4989 previous_callout = code; /* Save for later completion */
4990 after_manual_callout = 1; /* Skip one item before completing */
4991 *code++ = OP_CALLOUT;
4992 {
4993 int n = 0;
4994 while ((digitab[*(++ptr)] & ctype_digit) != 0)
4995 n = n * 10 + *ptr - CHAR_0;
4996 if (*ptr != CHAR_RIGHT_PARENTHESIS)
4997 {
4998 *errorcodeptr = ERR39;
4999 goto FAILED;
5000 }
5001 if (n > 255)
5002 {
5003 *errorcodeptr = ERR38;
5004 goto FAILED;
5005 }
5006 *code++ = n;
5007 PUT(code, 0, ptr - cd->start_pattern + 1); /* Pattern offset */
5008 PUT(code, LINK_SIZE, 0); /* Default length */
5009 code += 2 * LINK_SIZE;
5010 }
5011 previous = NULL;
5012 continue;
5013
5014
5015 /* ------------------------------------------------------------ */
5016 case CHAR_P: /* Python-style named subpattern handling */
5017 if (*(++ptr) == CHAR_EQUALS_SIGN ||
5018 *ptr == CHAR_GREATER_THAN_SIGN) /* Reference or recursion */
5019 {
5020 is_recurse = *ptr == CHAR_GREATER_THAN_SIGN;
5021 terminator = CHAR_RIGHT_PARENTHESIS;
5022 goto NAMED_REF_OR_RECURSE;
5023 }
5024 else if (*ptr != CHAR_LESS_THAN_SIGN) /* Test for Python-style defn */
5025 {
5026 *errorcodeptr = ERR41;
5027 goto FAILED;
5028 }
5029 /* Fall through to handle (?P< as (?< is handled */
5030
5031
5032 /* ------------------------------------------------------------ */
5033 DEFINE_NAME: /* Come here from (?< handling */
5034 case CHAR_APOSTROPHE:
5035 {
5036 terminator = (*ptr == CHAR_LESS_THAN_SIGN)?
5037 CHAR_GREATER_THAN_SIGN : CHAR_APOSTROPHE;
5038 name = ++ptr;
5039
5040 while ((cd->ctypes[*ptr] & ctype_word) != 0) ptr++;
5041 namelen = ptr - name;
5042
5043 /* In the pre-compile phase, just do a syntax check. */
5044
5045 if (lengthptr != NULL)
5046 {
5047 if (*ptr != terminator)
5048 {
5049 *errorcodeptr = ERR42;
5050 goto FAILED;
5051 }
5052 if (cd->names_found >= MAX_NAME_COUNT)
5053 {
5054 *errorcodeptr = ERR49;
5055 goto FAILED;
5056 }
5057 if (namelen + 3 > cd->name_entry_size)
5058 {
5059 cd->name_entry_size = namelen + 3;
5060 if (namelen > MAX_NAME_SIZE)
5061 {
5062 *errorcodeptr = ERR48;
5063 goto FAILED;
5064 }
5065 }
5066 }
5067
5068 /* In the real compile, create the entry in the table, maintaining
5069 alphabetical order. Duplicate names for different numbers are
5070 permitted only if PCRE_DUPNAMES is set. Duplicate names for the same
5071 number are always OK. (An existing number can be re-used if (?|
5072 appears in the pattern.) In either event, a duplicate name results in
5073 a duplicate entry in the table, even if the number is the same. This
5074 is because the number of names, and hence the table size, is computed
5075 in the pre-compile, and it affects various numbers and pointers which
5076 would all have to be modified, and the compiled code moved down, if
5077 duplicates with the same number were omitted from the table. This
5078 doesn't seem worth the hassle. However, *different* names for the
5079 same number are not permitted. */
5080
5081 else
5082 {
5083 BOOL dupname = FALSE;
5084 slot = cd->name_table;
5085
5086 for (i = 0; i < cd->names_found; i++)
5087 {
5088 int crc = memcmp(name, slot+2, namelen);
5089 if (crc == 0)
5090 {
5091 if (slot[2+namelen] == 0)
5092 {
5093 if (GET2(slot, 0) != cd->bracount + 1 &&
5094 (options & PCRE_DUPNAMES) == 0)
5095 {
5096 *errorcodeptr = ERR43;
5097 goto FAILED;
5098 }
5099 else dupname = TRUE;
5100 }
5101 else crc = -1; /* Current name is a substring */
5102 }
5103
5104 /* Make space in the table and break the loop for an earlier
5105 name. For a duplicate or later name, carry on. We do this for
5106 duplicates so that in the simple case (when ?(| is not used) they
5107 are in order of their numbers. */
5108
5109 if (crc < 0)
5110 {
5111 memmove(slot + cd->name_entry_size, slot,
5112 (cd->names_found - i) * cd->name_entry_size);
5113 break;
5114 }
5115
5116 /* Continue the loop for a later or duplicate name */
5117
5118 slot += cd->name_entry_size;
5119 }
5120
5121 /* For non-duplicate names, check for a duplicate number before
5122 adding the new name. */
5123
5124 if (!dupname)
5125 {
5126 uschar *cslot = cd->name_table;
5127 for (i = 0; i < cd->names_found; i++)
5128 {
5129 if (cslot != slot)
5130 {
5131 if (GET2(cslot, 0) == cd->bracount + 1)
5132 {
5133 *errorcodeptr = ERR65;
5134 goto FAILED;
5135 }
5136 }
5137 else i--;
5138 cslot += cd->name_entry_size;
5139 }
5140 }
5141
5142 PUT2(slot, 0, cd->bracount + 1);
5143 memcpy(slot + 2, name, namelen);
5144 slot[2+namelen] = 0;
5145 }
5146 }
5147
5148 /* In both pre-compile and compile, count the number of names we've
5149 encountered. */
5150
5151 cd->names_found++;
5152 ptr++; /* Move past > or ' */
5153 goto NUMBERED_GROUP;
5154
5155
5156 /* ------------------------------------------------------------ */
5157 case CHAR_AMPERSAND: /* Perl recursion/subroutine syntax */
5158 terminator = CHAR_RIGHT_PARENTHESIS;
5159 is_recurse = TRUE;
5160 /* Fall through */
5161
5162 /* We come here from the Python syntax above that handles both
5163 references (?P=name) and recursion (?P>name), as well as falling
5164 through from the Perl recursion syntax (?&name). We also come here from
5165 the Perl \k<name> or \k'name' back reference syntax and the \k{name}
5166 .NET syntax, and the Oniguruma \g<...> and \g'...' subroutine syntax. */
5167
5168 NAMED_REF_OR_RECURSE:
5169 name = ++ptr;
5170 while ((cd->ctypes[*ptr] & ctype_word) != 0) ptr++;
5171 namelen = ptr - name;
5172
5173 /* In the pre-compile phase, do a syntax check and set a dummy
5174 reference number. */
5175
5176 if (lengthptr != NULL)
5177 {
5178 if (namelen == 0)
5179 {
5180 *errorcodeptr = ERR62;
5181 goto FAILED;
5182 }
5183 if (*ptr != terminator)
5184 {
5185 *errorcodeptr = ERR42;
5186 goto FAILED;
5187 }
5188 if (namelen > MAX_NAME_SIZE)
5189 {
5190 *errorcodeptr = ERR48;
5191 goto FAILED;
5192 }
5193 recno = 0;
5194 }
5195
5196 /* In the real compile, seek the name in the table. We check the name
5197 first, and then check that we have reached the end of the name in the
5198 table. That way, if the name that is longer than any in the table,
5199 the comparison will fail without reading beyond the table entry. */
5200
5201 else
5202 {
5203 slot = cd->name_table;
5204 for (i = 0; i < cd->names_found; i++)
5205 {
5206 if (strncmp((char *)name, (char *)slot+2, namelen) == 0 &&
5207 slot[2+namelen] == 0)
5208 break;
5209 slot += cd->name_entry_size;
5210 }
5211
5212 if (i < cd->names_found) /* Back reference */
5213 {
5214 recno = GET2(slot, 0);
5215 }
5216 else if ((recno = /* Forward back reference */
5217 find_parens(cd, name, namelen,
5218 (options & PCRE_EXTENDED) != 0)) <= 0)
5219 {
5220 *errorcodeptr = ERR15;
5221 goto FAILED;
5222 }
5223 }
5224
5225 /* In both phases, we can now go to the code than handles numerical
5226 recursion or backreferences. */
5227
5228 if (is_recurse) goto HANDLE_RECURSION;
5229 else goto HANDLE_REFERENCE;
5230
5231
5232 /* ------------------------------------------------------------ */
5233 case CHAR_R: /* Recursion */
5234 ptr++; /* Same as (?0) */
5235 /* Fall through */
5236
5237
5238 /* ------------------------------------------------------------ */
5239 case CHAR_MINUS: case CHAR_PLUS: /* Recursion or subroutine */
5240 case CHAR_0: case CHAR_1: case CHAR_2: case CHAR_3: case CHAR_4:
5241 case CHAR_5: case CHAR_6: case CHAR_7: case CHAR_8: case CHAR_9:
5242 {
5243 const uschar *called;
5244 terminator = CHAR_RIGHT_PARENTHESIS;
5245
5246 /* Come here from the \g<...> and \g'...' code (Oniguruma
5247 compatibility). However, the syntax has been checked to ensure that
5248 the ... are a (signed) number, so that neither ERR63 nor ERR29 will
5249 be called on this path, nor with the jump to OTHER_CHAR_AFTER_QUERY
5250 ever be taken. */
5251
5252 HANDLE_NUMERICAL_RECURSION:
5253
5254 if ((refsign = *ptr) == CHAR_PLUS)
5255 {
5256 ptr++;
5257 if ((digitab[*ptr] & ctype_digit) == 0)
5258 {
5259 *errorcodeptr = ERR63;
5260 goto FAILED;
5261 }
5262 }
5263 else if (refsign == CHAR_MINUS)
5264 {
5265 if ((digitab[ptr[1]] & ctype_digit) == 0)
5266 goto OTHER_CHAR_AFTER_QUERY;
5267 ptr++;
5268 }
5269
5270 recno = 0;
5271 while((digitab[*ptr] & ctype_digit) != 0)
5272 recno = recno * 10 + *ptr++ - CHAR_0;
5273
5274 if (*ptr != terminator)
5275 {
5276 *errorcodeptr = ERR29;
5277 goto FAILED;
5278 }
5279
5280 if (refsign == CHAR_MINUS)
5281 {
5282 if (recno == 0)
5283 {
5284 *errorcodeptr = ERR58;
5285 goto FAILED;
5286 }
5287 recno = cd->bracount - recno + 1;
5288 if (recno <= 0)
5289 {
5290 *errorcodeptr = ERR15;
5291 goto FAILED;
5292 }
5293 }
5294 else if (refsign == CHAR_PLUS)
5295 {
5296 if (recno == 0)
5297 {
5298 *errorcodeptr = ERR58;
5299 goto FAILED;
5300 }
5301 recno += cd->bracount;
5302 }
5303
5304 /* Come here from code above that handles a named recursion */
5305
5306 HANDLE_RECURSION:
5307
5308 previous = code;
5309 called = cd->start_code;
5310
5311 /* When we are actually compiling, find the bracket that is being
5312 referenced. Temporarily end the regex in case it doesn't exist before
5313 this point. If we end up with a forward reference, first check that
5314 the bracket does occur later so we can give the error (and position)
5315 now. Then remember this forward reference in the workspace so it can
5316 be filled in at the end. */
5317
5318 if (lengthptr == NULL)
5319 {
5320 *code = OP_END;
5321 if (recno != 0)
5322 called = _pcre_find_bracket(cd->start_code, utf8, recno);
5323
5324 /* Forward reference */
5325
5326 if (called == NULL)
5327 {
5328 if (find_parens(cd, NULL, recno,
5329 (options & PCRE_EXTENDED) != 0) < 0)
5330 {
5331 *errorcodeptr = ERR15;
5332 goto FAILED;
5333 }
5334
5335 /* Fudge the value of "called" so that when it is inserted as an
5336 offset below, what it actually inserted is the reference number
5337 of the group. */
5338
5339 called = cd->start_code + recno;
5340 PUTINC(cd->hwm, 0, code + 2 + LINK_SIZE - cd->start_code);
5341 }
5342
5343 /* If not a forward reference, and the subpattern is still open,
5344 this is a recursive call. We check to see if this is a left
5345 recursion that could loop for ever, and diagnose that case. */
5346
5347 else if (GET(called, 1) == 0 &&
5348 could_be_empty(called, code, bcptr, utf8, cd))
5349 {
5350 *errorcodeptr = ERR40;
5351 goto FAILED;
5352 }
5353 }
5354
5355 /* Insert the recursion/subroutine item, automatically wrapped inside
5356 "once" brackets. Set up a "previous group" length so that a
5357 subsequent quantifier will work. */
5358
5359 *code = OP_ONCE;
5360 PUT(code, 1, 2 + 2*LINK_SIZE);
5361 code += 1 + LINK_SIZE;
5362
5363 *code = OP_RECURSE;
5364 PUT(code, 1, called - cd->start_code);
5365 code += 1 + LINK_SIZE;
5366
5367 *code = OP_KET;
5368 PUT(code, 1, 2 + 2*LINK_SIZE);
5369 code += 1 + LINK_SIZE;
5370
5371 length_prevgroup = 3 + 3*LINK_SIZE;
5372 }
5373
5374 /* Can't determine a first byte now */
5375
5376 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
5377 continue;
5378
5379
5380 /* ------------------------------------------------------------ */
5381 default: /* Other characters: check option setting */
5382 OTHER_CHAR_AFTER_QUERY:
5383 set = unset = 0;
5384 optset = &set;
5385
5386 while (*ptr != CHAR_RIGHT_PARENTHESIS && *ptr != CHAR_COLON)
5387 {
5388 switch (*ptr++)
5389 {
5390 case CHAR_MINUS: optset = &unset; break;
5391
5392 case CHAR_J: /* Record that it changed in the external options */
5393 *optset |= PCRE_DUPNAMES;
5394 cd->external_flags |= PCRE_JCHANGED;
5395 break;
5396
5397 case CHAR_i: *optset |= PCRE_CASELESS; break;
5398 case CHAR_m: *optset |= PCRE_MULTILINE; break;
5399 case CHAR_s: *optset |= PCRE_DOTALL; break;
5400 case CHAR_x: *optset |= PCRE_EXTENDED; break;
5401 case CHAR_U: *optset |= PCRE_UNGREEDY; break;
5402 case CHAR_X: *optset |= PCRE_EXTRA; break;
5403
5404 default: *errorcodeptr = ERR12;
5405 ptr--; /* Correct the offset */
5406 goto FAILED;
5407 }
5408 }
5409
5410 /* Set up the changed option bits, but don't change anything yet. */
5411
5412 newoptions = (options | set) & (~unset);
5413
5414 /* If the options ended with ')' this is not the start of a nested
5415 group with option changes, so the options change at this level. If this
5416 item is right at the start of the pattern, the options can be
5417 abstracted and made external in the pre-compile phase, and ignored in
5418 the compile phase. This can be helpful when matching -- for instance in
5419 caseless checking of required bytes.
5420
5421 If the code pointer is not (cd->start_code + 1 + LINK_SIZE), we are
5422 definitely *not* at the start of the pattern because something has been
5423 compiled. In the pre-compile phase, however, the code pointer can have
5424 that value after the start, because it gets reset as code is discarded
5425 during the pre-compile. However, this can happen only at top level - if
5426 we are within parentheses, the starting BRA will still be present. At
5427 any parenthesis level, the length value can be used to test if anything
5428 has been compiled at that level. Thus, a test for both these conditions
5429 is necessary to ensure we correctly detect the start of the pattern in
5430 both phases.
5431
5432 If we are not at the pattern start, compile code to change the ims
5433 options if this setting actually changes any of them, and reset the
5434 greedy defaults and the case value for firstbyte and reqbyte. */
5435
5436 if (*ptr == CHAR_RIGHT_PARENTHESIS)
5437 {
5438 if (code == cd->start_code + 1 + LINK_SIZE &&
5439 (lengthptr == NULL || *lengthptr == 2 + 2*LINK_SIZE))
5440 {
5441 cd->external_options = newoptions;
5442 }
5443 else
5444 {
5445 if ((options & PCRE_IMS) != (newoptions & PCRE_IMS))
5446 {
5447 *code++ = OP_OPT;
5448 *code++ = newoptions & PCRE_IMS;
5449 }
5450 greedy_default = ((newoptions & PCRE_UNGREEDY) != 0);
5451 greedy_non_default = greedy_default ^ 1;
5452 req_caseopt = ((newoptions & PCRE_CASELESS) != 0)? REQ_CASELESS : 0;
5453 }
5454
5455 /* Change options at this level, and pass them back for use
5456 in subsequent branches. When not at the start of the pattern, this
5457 information is also necessary so that a resetting item can be
5458 compiled at the end of a group (if we are in a group). */
5459
5460 *optionsptr = options = newoptions;
5461 previous = NULL; /* This item can't be repeated */
5462 continue; /* It is complete */
5463 }
5464
5465 /* If the options ended with ':' we are heading into a nested group
5466 with possible change of options. Such groups are non-capturing and are
5467 not assertions of any kind. All we need to do is skip over the ':';
5468 the newoptions value is handled below. */
5469
5470 bravalue = OP_BRA;
5471 ptr++;
5472 } /* End of switch for character following (? */
5473 } /* End of (? handling */
5474
5475 /* Opening parenthesis not followed by '*' or '?'. If PCRE_NO_AUTO_CAPTURE
5476 is set, all unadorned brackets become non-capturing and behave like (?:...)
5477 brackets. */
5478
5479 else if ((options & PCRE_NO_AUTO_CAPTURE) != 0)
5480 {
5481 bravalue = OP_BRA;
5482 }
5483
5484 /* Else we have a capturing group. */
5485
5486 else
5487 {
5488 NUMBERED_GROUP:
5489 cd->bracount += 1;
5490 PUT2(code, 1+LINK_SIZE, cd->bracount);
5491 skipbytes = 2;
5492 }
5493
5494 /* Process nested bracketed regex. Assertions may not be repeated, but
5495 other kinds can be. All their opcodes are >= OP_ONCE. We copy code into a
5496 non-register variable in order to be able to pass its address because some
5497 compilers complain otherwise. Pass in a new setting for the ims options if
5498 they have changed. */
5499
5500 previous = (bravalue >= OP_ONCE)? code : NULL;
5501 *code = bravalue;
5502 tempcode = code;
5503 tempreqvary = cd->req_varyopt; /* Save value before bracket */
5504 length_prevgroup = 0; /* Initialize for pre-compile phase */
5505
5506 if (!compile_regex(
5507 newoptions, /* The complete new option state */
5508 options & PCRE_IMS, /* The previous ims option state */
5509 &tempcode, /* Where to put code (updated) */
5510 &ptr, /* Input pointer (updated) */
5511 errorcodeptr, /* Where to put an error message */
5512 (bravalue == OP_ASSERTBACK ||
5513 bravalue == OP_ASSERTBACK_NOT), /* TRUE if back assert */
5514 reset_bracount, /* True if (?| group */
5515 skipbytes, /* Skip over bracket number */
5516 &subfirstbyte, /* For possible first char */
5517 &subreqbyte, /* For possible last char */
5518 bcptr, /* Current branch chain */
5519 cd, /* Tables block */
5520 (lengthptr == NULL)? NULL : /* Actual compile phase */
5521 &length_prevgroup /* Pre-compile phase */
5522 ))
5523 goto FAILED;
5524
5525 /* At the end of compiling, code is still pointing to the start of the
5526 group, while tempcode has been updated to point past the end of the group
5527 and any option resetting that may follow it. The pattern pointer (ptr)
5528 is on the bracket. */
5529
5530 /* If this is a conditional bracket, check that there are no more than
5531 two branches in the group, or just one if it's a DEFINE group. We do this
5532 in the real compile phase, not in the pre-pass, where the whole group may
5533 not be available. */
5534
5535 if (bravalue == OP_COND && lengthptr == NULL)
5536 {
5537 uschar *tc = code;
5538 int condcount = 0;
5539
5540 do {
5541 condcount++;
5542 tc += GET(tc,1);
5543 }
5544 while (*tc != OP_KET);
5545
5546 /* A DEFINE group is never obeyed inline (the "condition" is always
5547 false). It must have only one branch. */
5548
5549 if (code[LINK_SIZE+1] == OP_DEF)
5550 {
5551 if (condcount > 1)
5552 {
5553 *errorcodeptr = ERR54;
5554 goto FAILED;
5555 }
5556 bravalue = OP_DEF; /* Just a flag to suppress char handling below */
5557 }
5558
5559 /* A "normal" conditional group. If there is just one branch, we must not
5560 make use of its firstbyte or reqbyte, because this is equivalent to an
5561 empty second branch. */
5562
5563 else
5564 {
5565 if (condcount > 2)
5566 {
5567 *errorcodeptr = ERR27;
5568 goto FAILED;
5569 }
5570 if (condcount == 1) subfirstbyte = subreqbyte = REQ_NONE;
5571 }
5572 }
5573
5574 /* Error if hit end of pattern */
5575
5576 if (*ptr != CHAR_RIGHT_PARENTHESIS)
5577 {
5578 *errorcodeptr = ERR14;
5579 goto FAILED;
5580 }
5581
5582 /* In the pre-compile phase, update the length by the length of the group,
5583 less the brackets at either end. Then reduce the compiled code to just a
5584 set of non-capturing brackets so that it doesn't use much memory if it is
5585 duplicated by a quantifier.*/
5586
5587 if (lengthptr != NULL)
5588 {
5589 if (OFLOW_MAX - *lengthptr < length_prevgroup - 2 - 2*LINK_SIZE)
5590 {
5591 *errorcodeptr = ERR20;
5592 goto FAILED;
5593 }
5594 *lengthptr += length_prevgroup - 2 - 2*LINK_SIZE;
5595 *code++ = OP_BRA;
5596 PUTINC(code, 0, 1 + LINK_SIZE);
5597 *code++ = OP_KET;
5598 PUTINC(code, 0, 1 + LINK_SIZE);
5599 break; /* No need to waste time with special character handling */
5600 }
5601
5602 /* Otherwise update the main code pointer to the end of the group. */
5603
5604 code = tempcode;
5605
5606 /* For a DEFINE group, required and first character settings are not
5607 relevant. */
5608
5609 if (bravalue == OP_DEF) break;
5610
5611 /* Handle updating of the required and first characters for other types of
5612 group. Update for normal brackets of all kinds, and conditions with two
5613 branches (see code above). If the bracket is followed by a quantifier with
5614 zero repeat, we have to back off. Hence the definition of zeroreqbyte and
5615 zerofirstbyte outside the main loop so that they can be accessed for the
5616 back off. */
5617
5618 zeroreqbyte = reqbyte;
5619 zerofirstbyte = firstbyte;
5620 groupsetfirstbyte = FALSE;
5621
5622 if (bravalue >= OP_ONCE)
5623 {
5624 /* If we have not yet set a firstbyte in this branch, take it from the
5625 subpattern, remembering that it was set here so that a repeat of more
5626 than one can replicate it as reqbyte if necessary. If the subpattern has
5627 no firstbyte, set "none" for the whole branch. In both cases, a zero
5628 repeat forces firstbyte to "none". */
5629
5630 if (firstbyte == REQ_UNSET)
5631 {
5632 if (subfirstbyte >= 0)
5633 {
5634 firstbyte = subfirstbyte;
5635 groupsetfirstbyte = TRUE;
5636 }
5637 else firstbyte = REQ_NONE;
5638 zerofirstbyte = REQ_NONE;
5639 }
5640
5641 /* If firstbyte was previously set, convert the subpattern's firstbyte
5642 into reqbyte if there wasn't one, using the vary flag that was in
5643 existence beforehand. */
5644
5645 else if (subfirstbyte >= 0 && subreqbyte < 0)
5646 subreqbyte = subfirstbyte | tempreqvary;
5647
5648 /* If the subpattern set a required byte (or set a first byte that isn't
5649 really the first byte - see above), set it. */
5650
5651 if (subreqbyte >= 0) reqbyte = subreqbyte;
5652 }
5653
5654 /* For a forward assertion, we take the reqbyte, if set. This can be
5655 helpful if the pattern that follows the assertion doesn't set a different
5656 char. For example, it's useful for /(?=abcde).+/. We can't set firstbyte
5657 for an assertion, however because it leads to incorrect effect for patterns
5658 such as /(?=a)a.+/ when the "real" "a" would then become a reqbyte instead
5659 of a firstbyte. This is overcome by a scan at the end if there's no
5660 firstbyte, looking for an asserted first char. */
5661
5662 else if (bravalue == OP_ASSERT && subreqbyte >= 0) reqbyte = subreqbyte;
5663 break; /* End of processing '(' */
5664
5665
5666 /* ===================================================================*/
5667 /* Handle metasequences introduced by \. For ones like \d, the ESC_ values
5668 are arranged to be the negation of the corresponding OP_values in the
5669 default case when PCRE_UCP is not set. For the back references, the values
5670 are ESC_REF plus the reference number. Only back references and those types
5671 that consume a character may be repeated. We can test for values between
5672 ESC_b and ESC_Z for the latter; this may have to change if any new ones are
5673 ever created. */
5674
5675 case CHAR_BACKSLASH:
5676 tempptr = ptr;
5677 c = check_escape(&ptr, errorcodeptr, cd->bracount, options, FALSE);
5678 if (*errorcodeptr != 0) goto FAILED;
5679
5680 if (c < 0)
5681 {
5682 if (-c == ESC_Q) /* Handle start of quoted string */
5683 {
5684 if (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
5685 ptr += 2; /* avoid empty string */
5686 else inescq = TRUE;
5687 continue;
5688 }
5689
5690 if (-c == ESC_E) continue; /* Perl ignores an orphan \E */
5691
5692 /* For metasequences that actually match a character, we disable the
5693 setting of a first character if it hasn't already been set. */
5694
5695 if (firstbyte == REQ_UNSET && -c > ESC_b && -c < ESC_Z)
5696 firstbyte = REQ_NONE;
5697
5698 /* Set values to reset to if this is followed by a zero repeat. */
5699
5700 zerofirstbyte = firstbyte;
5701 zeroreqbyte = reqbyte;
5702
5703 /* \g<name> or \g'name' is a subroutine call by name and \g<n> or \g'n'
5704 is a subroutine call by number (Oniguruma syntax). In fact, the value
5705 -ESC_g is returned only for these cases. So we don't need to check for <
5706 or ' if the value is -ESC_g. For the Perl syntax \g{n} the value is
5707 -ESC_REF+n, and for the Perl syntax \g{name} the result is -ESC_k (as
5708 that is a synonym for a named back reference). */
5709
5710 if (-c == ESC_g)
5711 {
5712 const uschar *p;
5713 save_hwm = cd->hwm; /* Normally this is set when '(' is read */
5714 terminator = (*(++ptr) == CHAR_LESS_THAN_SIGN)?
5715 CHAR_GREATER_THAN_SIGN : CHAR_APOSTROPHE;
5716
5717 /* These two statements stop the compiler for warning about possibly
5718 unset variables caused by the jump to HANDLE_NUMERICAL_RECURSION. In
5719 fact, because we actually check for a number below, the paths that
5720 would actually be in error are never taken. */
5721
5722 skipbytes = 0;
5723 reset_bracount = FALSE;
5724
5725 /* Test for a name */
5726
5727 if (ptr[1] != CHAR_PLUS && ptr[1] != CHAR_MINUS)
5728 {
5729 BOOL isnumber = TRUE;
5730 for (p = ptr + 1; *p != 0 && *p != terminator; p++)
5731 {
5732 if ((cd->ctypes[*p] & ctype_digit) == 0) isnumber = FALSE;
5733 if ((cd->ctypes[*p] & ctype_word) == 0) break;
5734 }
5735 if (*p != terminator)
5736 {
5737 *errorcodeptr = ERR57;
5738 break;
5739 }
5740 if (isnumber)
5741 {
5742 ptr++;
5743 goto HANDLE_NUMERICAL_RECURSION;
5744 }
5745 is_recurse = TRUE;
5746 goto NAMED_REF_OR_RECURSE;
5747 }
5748
5749 /* Test a signed number in angle brackets or quotes. */
5750
5751 p = ptr + 2;
5752 while ((digitab[*p] & ctype_digit) != 0) p++;
5753 if (*p != terminator)
5754 {
5755 *errorcodeptr = ERR57;
5756 break;
5757 }
5758 ptr++;
5759 goto HANDLE_NUMERICAL_RECURSION;
5760 }
5761
5762 /* \k<name> or \k'name' is a back reference by name (Perl syntax).
5763 We also support \k{name} (.NET syntax) */
5764
5765 if (-c == ESC_k && (ptr[1] == CHAR_LESS_THAN_SIGN ||
5766 ptr[1] == CHAR_APOSTROPHE || ptr[1] == CHAR_LEFT_CURLY_BRACKET))
5767 {
5768 is_recurse = FALSE;
5769 terminator = (*(++ptr) == CHAR_LESS_THAN_SIGN)?
5770 CHAR_GREATER_THAN_SIGN : (*ptr == CHAR_APOSTROPHE)?
5771 CHAR_APOSTROPHE : CHAR_RIGHT_CURLY_BRACKET;
5772 goto NAMED_REF_OR_RECURSE;
5773 }
5774
5775 /* Back references are handled specially; must disable firstbyte if
5776 not set to cope with cases like (?=(\w+))\1: which would otherwise set
5777 ':' later. */
5778
5779 if (-c >= ESC_REF)
5780 {
5781 open_capitem *oc;
5782 recno = -c - ESC_REF;
5783
5784 HANDLE_REFERENCE: /* Come here from named backref handling */
5785 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
5786 previous = code;
5787 *code++ = OP_REF;
5788 PUT2INC(code, 0, recno);
5789 cd->backref_map |= (recno < 32)? (1 << recno) : 1;
5790 if (recno > cd->top_backref) cd->top_backref = recno;
5791
5792 /* Check to see if this back reference is recursive, that it, it
5793 is inside the group that it references. A flag is set so that the
5794 group can be made atomic. */
5795
5796 for (oc = cd->open_caps; oc != NULL; oc = oc->next)
5797 {
5798 if (oc->number == recno)
5799 {
5800 oc->flag = TRUE;
5801 break;
5802 }
5803 }
5804 }
5805
5806 /* So are Unicode property matches, if supported. */
5807
5808 #ifdef SUPPORT_UCP
5809 else if (-c == ESC_P || -c == ESC_p)
5810 {
5811 BOOL negated;
5812 int pdata;
5813 int ptype = get_ucp(&ptr, &negated, &pdata, errorcodeptr);
5814 if (ptype < 0) goto FAILED;
5815 previous = code;
5816 *code++ = ((-c == ESC_p) != negated)? OP_PROP : OP_NOTPROP;
5817 *code++ = ptype;
5818 *code++ = pdata;
5819 }
5820 #else
5821
5822 /* If Unicode properties are not supported, \X, \P, and \p are not
5823 allowed. */
5824
5825 else if (-c == ESC_X || -c == ESC_P || -c == ESC_p)
5826 {
5827 *errorcodeptr = ERR45;
5828 goto FAILED;
5829 }
5830 #endif
5831
5832 /* For the rest (including \X when Unicode properties are supported), we
5833 can obtain the OP value by negating the escape value in the default
5834 situation when PCRE_UCP is not set. When it *is* set, we substitute
5835 Unicode property tests. */
5836
5837 else
5838 {
5839 #ifdef SUPPORT_UCP
5840 if (-c >= ESC_DU && -c <= ESC_wu)
5841 {
5842 nestptr = ptr + 1; /* Where to resume */
5843 ptr = substitutes[-c - ESC_DU] - 1; /* Just before substitute */
5844 }
5845 else
5846 #endif
5847 {
5848 previous = (-c > ESC_b && -c < ESC_Z)? code : NULL;
5849 *code++ = -c;
5850 }
5851 }
5852 continue;
5853 }
5854
5855 /* We have a data character whose value is in c. In UTF-8 mode it may have
5856 a value > 127. We set its representation in the length/buffer, and then
5857 handle it as a data character. */
5858
5859 #ifdef SUPPORT_UTF8
5860 if (utf8 && c > 127)
5861 mclength = _pcre_ord2utf8(c, mcbuffer);
5862 else
5863 #endif
5864
5865 {
5866 mcbuffer[0] = c;
5867 mclength = 1;
5868 }
5869 goto ONE_CHAR;
5870
5871
5872 /* ===================================================================*/
5873 /* Handle a literal character. It is guaranteed not to be whitespace or #
5874 when the extended flag is set. If we are in UTF-8 mode, it may be a
5875 multi-byte literal character. */
5876
5877 default:
5878 NORMAL_CHAR:
5879 mclength = 1;
5880 mcbuffer[0] = c;
5881
5882 #ifdef SUPPORT_UTF8
5883 if (utf8 && c >= 0xc0)
5884 {
5885 while ((ptr[1] & 0xc0) == 0x80)
5886 mcbuffer[mclength++] = *(++ptr);
5887 }
5888 #endif
5889
5890 /* At this point we have the character's bytes in mcbuffer, and the length
5891 in mclength. When not in UTF-8 mode, the length is always 1. */
5892
5893 ONE_CHAR:
5894 previous = code;
5895 *code++ = ((options & PCRE_CASELESS) != 0)? OP_CHARNC : OP_CHAR;
5896 for (c = 0; c < mclength; c++) *code++ = mcbuffer[c];
5897
5898 /* Remember if \r or \n were seen */
5899
5900 if (mcbuffer[0] == CHAR_CR || mcbuffer[0] == CHAR_NL)
5901 cd->external_flags |= PCRE_HASCRORLF;
5902
5903 /* Set the first and required bytes appropriately. If no previous first
5904 byte, set it from this character, but revert to none on a zero repeat.
5905 Otherwise, leave the firstbyte value alone, and don't change it on a zero
5906 repeat. */
5907
5908 if (firstbyte == REQ_UNSET)
5909 {
5910 zerofirstbyte = REQ_NONE;
5911 zeroreqbyte = reqbyte;
5912
5913 /* If the character is more than one byte long, we can set firstbyte
5914 only if it is not to be matched caselessly. */
5915
5916 if (mclength == 1 || req_caseopt == 0)
5917 {
5918 firstbyte = mcbuffer[0] | req_caseopt;
5919 if (mclength != 1) reqbyte = code[-1] | cd->req_varyopt;
5920 }
5921 else firstbyte = reqbyte = REQ_NONE;
5922 }
5923
5924 /* firstbyte was previously set; we can set reqbyte only the length is
5925 1 or the matching is caseful. */
5926
5927 else
5928 {
5929 zerofirstbyte = firstbyte;
5930 zeroreqbyte = reqbyte;
5931 if (mclength == 1 || req_caseopt == 0)
5932 reqbyte = code[-1] | req_caseopt | cd->req_varyopt;
5933 }
5934
5935 break; /* End of literal character handling */
5936 }
5937 } /* end of big loop */
5938
5939
5940 /* Control never reaches here by falling through, only by a goto for all the
5941 error states. Pass back the position in the pattern so that it can be displayed
5942 to the user for diagnosing the error. */
5943
5944 FAILED:
5945 *ptrptr = ptr;
5946 return FALSE;
5947 }
5948
5949
5950
5951
5952 /*************************************************
5953 * Compile sequence of alternatives *
5954 *************************************************/
5955
5956 /* On entry, ptr is pointing past the bracket character, but on return it
5957 points to the closing bracket, or vertical bar, or end of string. The code
5958 variable is pointing at the byte into which the BRA operator has been stored.
5959 If the ims options are changed at the start (for a (?ims: group) or during any
5960 branch, we need to insert an OP_OPT item at the start of every following branch
5961 to ensure they get set correctly at run time, and also pass the new options
5962 into every subsequent branch compile.
5963
5964 This function is used during the pre-compile phase when we are trying to find
5965 out the amount of memory needed, as well as during the real compile phase. The
5966 value of lengthptr distinguishes the two phases.
5967
5968 Arguments:
5969 options option bits, including any changes for this subpattern
5970 oldims previous settings of ims option bits
5971 codeptr -> the address of the current code pointer
5972 ptrptr -> the address of the current pattern pointer
5973 errorcodeptr -> pointer to error code variable
5974 lookbehind TRUE if this is a lookbehind assertion
5975 reset_bracount TRUE to reset the count for each branch
5976 skipbytes skip this many bytes at start (for brackets and OP_COND)
5977 firstbyteptr place to put the first required character, or a negative number
5978 reqbyteptr place to put the last required character, or a negative number
5979 bcptr pointer to the chain of currently open branches
5980 cd points to the data block with tables pointers etc.
5981 lengthptr NULL during the real compile phase
5982 points to length accumulator during pre-compile phase
5983
5984 Returns: TRUE on success
5985 */
5986
5987 static BOOL
5988 compile_regex(int options, int oldims, uschar **codeptr, const uschar **ptrptr,
5989 int *errorcodeptr, BOOL lookbehind, BOOL reset_bracount, int skipbytes,
5990 int *firstbyteptr, int *reqbyteptr, branch_chain *bcptr, compile_data *cd,
5991 int *lengthptr)
5992 {
5993 const uschar *ptr = *ptrptr;
5994 uschar *code = *codeptr;
5995 uschar *last_branch = code;
5996 uschar *start_bracket = code;
5997 uschar *reverse_count = NULL;
5998 open_capitem capitem;
5999 int capnumber = 0;
6000 int firstbyte, reqbyte;
6001 int branchfirstbyte, branchreqbyte;
6002 int length;
6003 int orig_bracount;
6004 int max_bracount;
6005 int old_external_options = cd->external_options;
6006 branch_chain bc;
6007
6008 bc.outer = bcptr;
6009 bc.current_branch = code;
6010
6011 firstbyte = reqbyte = REQ_UNSET;
6012
6013 /* Accumulate the length for use in the pre-compile phase. Start with the
6014 length of the BRA and KET and any extra bytes that are required at the
6015 beginning. We accumulate in a local variable to save frequent testing of
6016 lenthptr for NULL. We cannot do this by looking at the value of code at the
6017 start and end of each alternative, because compiled items are discarded during
6018 the pre-compile phase so that the work space is not exceeded. */
6019
6020 length = 2 + 2*LINK_SIZE + skipbytes;
6021
6022 /* WARNING: If the above line is changed for any reason, you must also change
6023 the code that abstracts option settings at the start of the pattern and makes
6024 them global. It tests the value of length for (2 + 2*LINK_SIZE) in the
6025 pre-compile phase to find out whether anything has yet been compiled or not. */
6026
6027 /* If this is a capturing subpattern, add to the chain of open capturing items
6028 so that we can detect them if (*ACCEPT) is encountered. This is also used to
6029 detect groups that contain recursive back references to themselves. */
6030
6031 if (*code == OP_CBRA)
6032 {
6033 capnumber = GET2(code, 1 + LINK_SIZE);
6034 capitem.number = capnumber;
6035 capitem.next = cd->open_caps;
6036 capitem.flag = FALSE;
6037 cd->open_caps = &capitem;
6038 }
6039
6040 /* Offset is set zero to mark that this bracket is still open */
6041
6042 PUT(code, 1, 0);
6043 code += 1 + LINK_SIZE + skipbytes;
6044
6045 /* Loop for each alternative branch */
6046
6047 orig_bracount = max_bracount = cd->bracount;
6048 for (;;)
6049 {
6050 /* For a (?| group, reset the capturing bracket count so that each branch
6051 uses the same numbers. */
6052
6053 if (reset_bracount) cd->bracount = orig_bracount;
6054
6055 /* Handle a change of ims options at the start of the branch */
6056
6057 if ((options & PCRE_IMS) != oldims)
6058 {
6059 *code++ = OP_OPT;
6060 *code++ = options & PCRE_IMS;
6061 length += 2;
6062 }
6063
6064 /* Set up dummy OP_REVERSE if lookbehind assertion */
6065
6066 if (lookbehind)
6067 {
6068 *code++ = OP_REVERSE;
6069 reverse_count = code;
6070 PUTINC(code, 0, 0);
6071 length += 1 + LINK_SIZE;
6072 }
6073
6074 /* Now compile the branch; in the pre-compile phase its length gets added
6075 into the length. */
6076
6077 if (!compile_branch(&options, &code, &ptr, errorcodeptr, &branchfirstbyte,
6078 &branchreqbyte, &bc, cd, (lengthptr == NULL)? NULL : &length))
6079 {
6080 *ptrptr = ptr;
6081 return FALSE;
6082 }
6083
6084 /* If the external options have changed during this branch, it means that we
6085 are at the top level, and a leading option setting has been encountered. We
6086 need to re-set the original option values to take account of this so that,
6087 during the pre-compile phase, we know to allow for a re-set at the start of
6088 subsequent branches. */
6089
6090 if (old_external_options != cd->external_options)
6091 oldims = cd->external_options & PCRE_IMS;
6092
6093 /* Keep the highest bracket count in case (?| was used and some branch
6094 has fewer than the rest. */
6095
6096 if (cd->bracount > max_bracount) max_bracount = cd->bracount;
6097
6098 /* In the real compile phase, there is some post-processing to be done. */
6099
6100 if (lengthptr == NULL)
6101 {
6102 /* If this is the first branch, the firstbyte and reqbyte values for the
6103 branch become the values for the regex. */
6104
6105 if (*last_branch != OP_ALT)
6106 {
6107 firstbyte = branchfirstbyte;
6108 reqbyte = branchreqbyte;
6109 }
6110
6111 /* If this is not the first branch, the first char and reqbyte have to
6112 match the values from all the previous branches, except that if the
6113 previous value for reqbyte didn't have REQ_VARY set, it can still match,
6114 and we set REQ_VARY for the regex. */
6115
6116 else
6117 {
6118 /* If we previously had a firstbyte, but it doesn't match the new branch,
6119 we have to abandon the firstbyte for the regex, but if there was
6120 previously no reqbyte, it takes on the value of the old firstbyte. */
6121
6122 if (firstbyte >= 0 && firstbyte != branchfirstbyte)
6123 {
6124 if (reqbyte < 0) reqbyte = firstbyte;
6125 firstbyte = REQ_NONE;
6126 }
6127
6128 /* If we (now or from before) have no firstbyte, a firstbyte from the
6129 branch becomes a reqbyte if there isn't a branch reqbyte. */
6130
6131 if (firstbyte < 0 && branchfirstbyte >= 0 && branchreqbyte < 0)
6132 branchreqbyte = branchfirstbyte;
6133
6134 /* Now ensure that the reqbytes match */
6135
6136 if ((reqbyte & ~REQ_VARY) != (branchreqbyte & ~REQ_VARY))
6137 reqbyte = REQ_NONE;
6138 else reqbyte |= branchreqbyte; /* To "or" REQ_VARY */
6139 }
6140
6141 /* If lookbehind, check that this branch matches a fixed-length string, and
6142 put the length into the OP_REVERSE item. Temporarily mark the end of the
6143 branch with OP_END. If the branch contains OP_RECURSE, the result is -3
6144 because there may be forward references that we can't check here. Set a
6145 flag to cause another lookbehind check at the end. Why not do it all at the
6146 end? Because common, erroneous checks are picked up here and the offset of
6147 the problem can be shown. */
6148
6149 if (lookbehind)
6150 {
6151 int fixed_length;
6152 *code = OP_END;
6153 fixed_length = find_fixedlength(last_branch, options, FALSE, cd);
6154 DPRINTF(("fixed length = %d\n", fixed_length));
6155 if (fixed_length == -3)
6156 {
6157 cd->check_lookbehind = TRUE;
6158 }
6159 else if (fixed_length < 0)
6160 {
6161 *errorcodeptr = (fixed_length == -2)? ERR36 : ERR25;
6162 *ptrptr = ptr;
6163 return FALSE;
6164 }
6165 else { PUT(reverse_count, 0, fixed_length); }
6166 }
6167 }
6168
6169 /* Reached end of expression, either ')' or end of pattern. In the real
6170 compile phase, go back through the alternative branches and reverse the chain
6171 of offsets, with the field in the BRA item now becoming an offset to the
6172 first alternative. If there are no alternatives, it points to the end of the
6173 group. The length in the terminating ket is always the length of the whole
6174 bracketed item. If any of the ims options were changed inside the group,
6175 compile a resetting op-code following, except at the very end of the pattern.
6176 Return leaving the pointer at the terminating char. */
6177
6178 if (*ptr != CHAR_VERTICAL_LINE)
6179 {
6180 if (lengthptr == NULL)
6181 {
6182 int branch_length = code - last_branch;
6183 do
6184 {
6185 int prev_length = GET(last_branch, 1);
6186 PUT(last_branch, 1, branch_length);
6187 branch_length = prev_length;
6188 last_branch -= branch_length;
6189 }
6190 while (branch_length > 0);
6191 }
6192
6193 /* Fill in the ket */
6194
6195 *code = OP_KET;
6196 PUT(code, 1, code - start_bracket);
6197 code += 1 + LINK_SIZE;
6198
6199 /* If it was a capturing subpattern, check to see if it contained any
6200 recursive back references. If so, we must wrap it in atomic brackets.
6201 In any event, remove the block from the chain. */
6202
6203 if (capnumber > 0)
6204 {
6205 if (cd->open_caps->flag)
6206 {
6207 memmove(start_bracket + 1 + LINK_SIZE, start_bracket,
6208 code - start_bracket);
6209 *start_bracket = OP_ONCE;
6210 code += 1 + LINK_SIZE;
6211 PUT(start_bracket, 1, code - start_bracket);
6212 *code = OP_KET;
6213 PUT(code, 1, code - start_bracket);
6214 code += 1 + LINK_SIZE;
6215 length += 2 + 2*LINK_SIZE;
6216 }
6217 cd->open_caps = cd->open_caps->next;
6218 }
6219
6220 /* Reset options if needed. */
6221
6222 if ((options & PCRE_IMS) != oldims && *ptr == CHAR_RIGHT_PARENTHESIS)
6223 {
6224 *code++ = OP_OPT;
6225 *code++ = oldims;
6226 length += 2;
6227 }
6228
6229 /* Retain the highest bracket number, in case resetting was used. */
6230
6231 cd->bracount = max_bracount;
6232
6233 /* Set values to pass back */
6234
6235 *codeptr = code;
6236 *ptrptr = ptr;
6237 *firstbyteptr = firstbyte;
6238 *reqbyteptr = reqbyte;
6239 if (lengthptr != NULL)
6240 {
6241 if (OFLOW_MAX - *lengthptr < length)
6242 {
6243 *errorcodeptr = ERR20;
6244 return FALSE;
6245 }
6246 *lengthptr += length;
6247 }
6248 return TRUE;
6249 }
6250
6251 /* Another branch follows. In the pre-compile phase, we can move the code
6252 pointer back to where it was for the start of the first branch. (That is,
6253 pretend that each branch is the only one.)
6254
6255 In the real compile phase, insert an ALT node. Its length field points back
6256 to the previous branch while the bracket remains open. At the end the chain
6257 is reversed. It's done like this so that the start of the bracket has a
6258 zero offset until it is closed, making it possible to detect recursion. */
6259
6260 if (lengthptr != NULL)
6261 {
6262 code = *codeptr + 1 + LINK_SIZE + skipbytes;
6263 length += 1 + LINK_SIZE;
6264 }
6265 else
6266 {
6267 *code = OP_ALT;
6268 PUT(code, 1, code - last_branch);
6269 bc.current_branch = last_branch = code;
6270 code += 1 + LINK_SIZE;
6271 }
6272
6273 ptr++;
6274 }
6275 /* Control never reaches here */
6276 }
6277
6278
6279
6280
6281 /*************************************************
6282 * Check for anchored expression *
6283 *************************************************/
6284
6285 /* Try to find out if this is an anchored regular expression. Consider each
6286 alternative branch. If they all start with OP_SOD or OP_CIRC, or with a bracket
6287 all of whose alternatives start with OP_SOD or OP_CIRC (recurse ad lib), then
6288 it's anchored. However, if this is a multiline pattern, then only OP_SOD
6289 counts, since OP_CIRC can match in the middle.
6290
6291 We can also consider a regex to be anchored if OP_SOM starts all its branches.
6292 This is the code for \G, which means "match at start of match position, taking
6293 into account the match offset".
6294
6295 A branch is also implicitly anchored if it starts with .* and DOTALL is set,
6296 because that will try the rest of the pattern at all possible matching points,
6297 so there is no point trying again.... er ....
6298
6299 .... except when the .* appears inside capturing parentheses, and there is a
6300 subsequent back reference to those parentheses. We haven't enough information
6301 to catch that case precisely.
6302
6303 At first, the best we could do was to detect when .* was in capturing brackets
6304 and the highest back reference was greater than or equal to that level.
6305 However, by keeping a bitmap of the first 31 back references, we can catch some
6306 of the more common cases more precisely.
6307
6308 Arguments:
6309 code points to start of expression (the bracket)
6310 options points to the options setting
6311 bracket_map a bitmap of which brackets we are inside while testing; this
6312 handles up to substring 31; after that we just have to take
6313 the less precise approach
6314 backref_map the back reference bitmap
6315
6316 Returns: TRUE or FALSE
6317 */
6318
6319 static BOOL
6320 is_anchored(register const uschar *code, int *options, unsigned int bracket_map,
6321 unsigned int backref_map)
6322 {
6323 do {
6324 const uschar *scode = first_significant_code(code + _pcre_OP_lengths[*code],
6325 options, PCRE_MULTILINE, FALSE);
6326 register int op = *scode;
6327
6328 /* Non-capturing brackets */
6329
6330 if (op == OP_BRA)
6331 {
6332 if (!is_anchored(scode, options, bracket_map, backref_map)) return FALSE;
6333 }
6334
6335 /* Capturing brackets */
6336
6337 else if (op == OP_CBRA)
6338 {
6339 int n = GET2(scode, 1+LINK_SIZE);
6340 int new_map = bracket_map | ((n < 32)? (1 << n) : 1);
6341 if (!is_anchored(scode, options, new_map, backref_map)) return FALSE;
6342 }
6343
6344 /* Other brackets */
6345
6346 else if (op == OP_ASSERT || op == OP_ONCE || op == OP_COND)
6347 {
6348 if (!is_anchored(scode, options, bracket_map, backref_map)) return FALSE;
6349 }
6350
6351 /* .* is not anchored unless DOTALL is set (which generates OP_ALLANY) and
6352 it isn't in brackets that are or may be referenced. */
6353
6354 else if ((op == OP_TYPESTAR || op == OP_TYPEMINSTAR ||
6355 op == OP_TYPEPOSSTAR))
6356 {
6357 if (scode[1] != OP_ALLANY || (bracket_map & backref_map) != 0)
6358 return FALSE;
6359 }
6360
6361 /* Check for explicit anchoring */
6362
6363 else if (op != OP_SOD && op != OP_SOM &&
6364 ((*options & PCRE_MULTILINE) != 0 || op != OP_CIRC))
6365 return FALSE;
6366 code += GET(code, 1);
6367 }
6368 while (*code == OP_ALT); /* Loop for each alternative */
6369 return TRUE;
6370 }
6371
6372
6373
6374 /*************************************************
6375 * Check for starting with ^ or .* *
6376 *************************************************/
6377
6378 /* This is called to find out if every branch starts with ^ or .* so that
6379 "first char" processing can be done to speed things up in multiline
6380 matching and for non-DOTALL patterns that start with .* (which must start at
6381 the beginning or after \n). As in the case of is_anchored() (see above), we
6382 have to take account of back references to capturing brackets that contain .*
6383 because in that case we can't make the assumption.
6384
6385 Arguments:
6386 code points to start of expression (the bracket)
6387 bracket_map a bitmap of which brackets we are inside while testing; this
6388 handles up to substring 31; after that we just have to take
6389 the less precise approach
6390 backref_map the back reference bitmap
6391
6392 Returns: TRUE or FALSE
6393 */
6394
6395 static BOOL
6396 is_startline(const uschar *code, unsigned int bracket_map,
6397 unsigned int backref_map)
6398 {
6399 do {
6400 const uschar *scode = first_significant_code(code + _pcre_OP_lengths[*code],
6401 NULL, 0, FALSE);
6402 register int op = *scode;
6403
6404 /* If we are at the start of a conditional assertion group, *both* the
6405 conditional assertion *and* what follows the condition must satisfy the test
6406 for start of line. Other kinds of condition fail. Note that there may be an
6407 auto-callout at the start of a condition. */
6408
6409 if (op == OP_COND)
6410 {
6411 scode += 1 + LINK_SIZE;
6412 if (*scode == OP_CALLOUT) scode += _pcre_OP_lengths[OP_CALLOUT];
6413 switch (*scode)
6414 {
6415 case OP_CREF:
6416 case OP_NCREF:
6417 case OP_RREF:
6418 case OP_NRREF:
6419 case OP_DEF:
6420 return FALSE;
6421
6422 default: /* Assertion */
6423 if (!is_startline(scode, bracket_map, backref_map)) return FALSE;
6424 do scode += GET(scode, 1); while (*scode == OP_ALT);
6425 scode += 1 + LINK_SIZE;
6426 break;
6427 }
6428 scode = first_significant_code(scode, NULL, 0, FALSE);
6429 op = *scode;
6430 }
6431
6432 /* Non-capturing brackets */
6433
6434 if (op == OP_BRA)
6435 {
6436 if (!is_startline(scode, bracket_map, backref_map)) return FALSE;
6437 }
6438
6439 /* Capturing brackets */
6440
6441 else if (op == OP_CBRA)
6442 {
6443 int n = GET2(scode, 1+LINK_SIZE);
6444 int new_map = bracket_map | ((n < 32)? (1 << n) : 1);
6445 if (!is_startline(scode, new_map, backref_map)) return FALSE;
6446 }
6447
6448 /* Other brackets */
6449
6450 else if (op == OP_ASSERT || op == OP_ONCE)
6451 {
6452 if (!is_startline(scode, bracket_map, backref_map)) return FALSE;
6453 }
6454
6455 /* .* means "start at start or after \n" if it isn't in brackets that
6456 may be referenced. */
6457
6458 else if (op == OP_TYPESTAR || op == OP_TYPEMINSTAR || op == OP_TYPEPOSSTAR)
6459 {
6460 if (scode[1] != OP_ANY || (bracket_map & backref_map) != 0) return FALSE;
6461 }
6462
6463 /* Check for explicit circumflex */
6464
6465 else if (op != OP_CIRC) return FALSE;
6466
6467 /* Move on to the next alternative */
6468
6469 code += GET(code, 1);
6470 }
6471 while (*code == OP_ALT); /* Loop for each alternative */
6472 return TRUE;
6473 }
6474
6475
6476
6477 /*************************************************
6478 * Check for asserted fixed first char *
6479 *************************************************/
6480
6481 /* During compilation, the "first char" settings from forward assertions are
6482 discarded, because they can cause conflicts with actual literals that follow.
6483 However, if we end up without a first char setting for an unanchored pattern,
6484 it is worth scanning the regex to see if there is an initial asserted first
6485 char. If all branches start with the same asserted char, or with a bracket all
6486 of whose alternatives start with the same asserted char (recurse ad lib), then
6487 we return that char, otherwise -1.
6488
6489 Arguments:
6490 code points to start of expression (the bracket)
6491 options pointer to the options (used to check casing changes)
6492 inassert TRUE if in an assertion
6493
6494 Returns: -1 or the fixed first char
6495 */
6496
6497 static int
6498 find_firstassertedchar(const uschar *code, int *options, BOOL inassert)
6499 {
6500 register int c = -1;
6501 do {
6502 int d;
6503 const uschar *scode =
6504 first_significant_code(code + 1+LINK_SIZE, options, PCRE_CASELESS, TRUE);
6505 register int op = *scode;
6506
6507 switch(op)
6508 {
6509 default:
6510 return -1;
6511
6512 case OP_BRA:
6513 case OP_CBRA:
6514 case OP_ASSERT:
6515 case OP_ONCE:
6516 case OP_COND:
6517 if ((d = find_firstassertedchar(scode, options, op == OP_ASSERT)) < 0)
6518 return -1;
6519 if (c < 0) c = d; else if (c != d) return -1;
6520 break;
6521
6522 case OP_EXACT: /* Fall through */
6523 scode += 2;
6524
6525 case OP_CHAR:
6526 case OP_CHARNC:
6527 case OP_PLUS:
6528 case OP_MINPLUS:
6529 case OP_POSPLUS:
6530 if (!inassert) return -1;
6531 if (c < 0)
6532 {
6533 c = scode[1];
6534 if ((*options & PCRE_CASELESS) != 0) c |= REQ_CASELESS;
6535 }
6536 else if (c != scode[1]) return -1;
6537 break;
6538 }
6539
6540 code += GET(code, 1);
6541 }
6542 while (*code == OP_ALT);
6543 return c;
6544 }
6545
6546
6547
6548 /*************************************************
6549 * Compile a Regular Expression *
6550 *************************************************/
6551
6552 /* This function takes a string and returns a pointer to a block of store
6553 holding a compiled version of the expression. The original API for this
6554 function had no error code return variable; it is retained for backwards
6555 compatibility. The new function is given a new name.
6556
6557 Arguments:
6558 pattern the regular expression
6559 options various option bits
6560 errorcodeptr pointer to error code variable (pcre_compile2() only)
6561 can be NULL if you don't want a code value
6562 errorptr pointer to pointer to error text
6563 erroroffset ptr offset in pattern where error was detected
6564 tables pointer to character tables or NULL
6565
6566 Returns: pointer to compiled data block, or NULL on error,
6567 with errorptr and erroroffset set
6568 */
6569
6570 PCRE_EXP_DEFN pcre * PCRE_CALL_CONVENTION
6571 pcre_compile(const char *pattern, int options, const char **errorptr,
6572 int *erroroffset, const unsigned char *tables)
6573 {
6574 return pcre_compile2(pattern, options, NULL, errorptr, erroroffset, tables);
6575 }
6576
6577
6578 PCRE_EXP_DEFN pcre * PCRE_CALL_CONVENTION
6579 pcre_compile2(const char *pattern, int options, int *errorcodeptr,
6580 const char **errorptr, int *erroroffset, const unsigned char *tables)
6581 {
6582 real_pcre *re;
6583 int length = 1; /* For final END opcode */
6584 int firstbyte, reqbyte, newline;
6585 int errorcode = 0;
6586 int skipatstart = 0;
6587 BOOL utf8;
6588 size_t size;
6589 uschar *code;
6590 const uschar *codestart;
6591 const uschar *ptr;
6592 compile_data compile_block;
6593 compile_data *cd = &compile_block;
6594
6595 /* This space is used for "compiling" into during the first phase, when we are
6596 computing the amount of memory that is needed. Compiled items are thrown away
6597 as soon as possible, so that a fairly large buffer should be sufficient for
6598 this purpose. The same space is used in the second phase for remembering where
6599 to fill in forward references to subpatterns. */
6600
6601 uschar cworkspace[COMPILE_WORK_SIZE];
6602
6603 /* Set this early so that early errors get offset 0. */
6604
6605 ptr = (const uschar *)pattern;
6606
6607 /* We can't pass back an error message if errorptr is NULL; I guess the best we
6608 can do is just return NULL, but we can set a code value if there is a code
6609 pointer. */
6610
6611 if (errorptr == NULL)
6612 {
6613 if (errorcodeptr != NULL) *errorcodeptr = 99;
6614 return NULL;
6615 }
6616
6617 *errorptr = NULL;
6618 if (errorcodeptr != NULL) *errorcodeptr = ERR0;
6619
6620 /* However, we can give a message for this error */
6621
6622 if (erroroffset == NULL)
6623 {
6624 errorcode = ERR16;
6625 goto PCRE_EARLY_ERROR_RETURN2;
6626 }
6627
6628 *erroroffset = 0;
6629
6630 /* Set up pointers to the individual character tables */
6631
6632 if (tables == NULL) tables = _pcre_default_tables;
6633 cd->lcc = tables + lcc_offset;
6634 cd->fcc = tables + fcc_offset;
6635 cd->cbits = tables + cbits_offset;
6636 cd->ctypes = tables + ctypes_offset;
6637
6638 /* Check that all undefined public option bits are zero */
6639
6640 if ((options & ~PUBLIC_COMPILE_OPTIONS) != 0)
6641 {
6642 errorcode = ERR17;
6643 goto PCRE_EARLY_ERROR_RETURN;
6644 }
6645
6646 /* Check for global one-time settings at the start of the pattern, and remember
6647 the offset for later. */
6648
6649 while (ptr[skipatstart] == CHAR_LEFT_PARENTHESIS &&
6650 ptr[skipatstart+1] == CHAR_ASTERISK)
6651 {
6652 int newnl = 0;
6653 int newbsr = 0;
6654
6655 if (strncmp((char *)(ptr+skipatstart+2), STRING_UTF8_RIGHTPAR, 5) == 0)
6656 { skipatstart += 7; options |= PCRE_UTF8; continue; }
6657 else if (strncmp((char *)(ptr+skipatstart+2), STRING_UCP_RIGHTPAR, 4) == 0)
6658 { skipatstart += 6; options |= PCRE_UCP; continue; }
6659
6660 if (strncmp((char *)(ptr+skipatstart+2), STRING_CR_RIGHTPAR, 3) == 0)
6661 { skipatstart += 5; newnl = PCRE_NEWLINE_CR; }
6662 else if (strncmp((char *)(ptr+skipatstart+2), STRING_LF_RIGHTPAR, 3) == 0)
6663 { skipatstart += 5; newnl = PCRE_NEWLINE_LF; }
6664 else if (strncmp((char *)(ptr+skipatstart+2), STRING_CRLF_RIGHTPAR, 5) == 0)
6665 { skipatstart += 7; newnl = PCRE_NEWLINE_CR + PCRE_NEWLINE_LF; }