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

Contents of /code/trunk/pcre_compile.c

Parent Directory Parent Directory | Revision Log Revision Log


Revision 635 - (show annotations) (download)
Sat Jul 23 16:19:50 2011 UTC (3 years, 1 month ago) by ph10
File MIME type: text/plain
File size: 241390 byte(s)
Document and comment tidies.

1 /*************************************************
2 * Perl-Compatible Regular Expressions *
3 *************************************************/
4
5 /* PCRE is a library of functions to support regular expressions whose syntax
6 and semantics are as close as possible to those of the Perl 5 language.
7
8 Written by Philip Hazel
9 Copyright (c) 1997-2011 University of Cambridge
10
11 -----------------------------------------------------------------------------
12 Redistribution and use in source and binary forms, with or without
13 modification, are permitted provided that the following conditions are met:
14
15 * Redistributions of source code must retain the above copyright notice,
16 this list of conditions and the following disclaimer.
17
18 * Redistributions in binary form must reproduce the above copyright
19 notice, this list of conditions and the following disclaimer in the
20 documentation and/or other materials provided with the distribution.
21
22 * Neither the name of the University of Cambridge nor the names of its
23 contributors may be used to endorse or promote products derived from
24 this software without specific prior written permission.
25
26 THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
27 AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28 IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29 ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE
30 LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
31 CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
32 SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
33 INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
34 CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
35 ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
36 POSSIBILITY OF SUCH DAMAGE.
37 -----------------------------------------------------------------------------
38 */
39
40
41 /* This module contains the external function pcre_compile(), along with
42 supporting internal functions that are not used by other modules. */
43
44
45 #ifdef HAVE_CONFIG_H
46 #include "config.h"
47 #endif
48
49 #define NLBLOCK cd /* Block containing newline information */
50 #define PSSTART start_pattern /* Field containing processed string start */
51 #define PSEND end_pattern /* Field containing processed string end */
52
53 #include "pcre_internal.h"
54
55
56 /* When PCRE_DEBUG is defined, we need the pcre_printint() function, which is
57 also used by pcretest. PCRE_DEBUG is not defined when building a production
58 library. */
59
60 #ifdef PCRE_DEBUG
61 #include "pcre_printint.src"
62 #endif
63
64
65 /* Macro for setting individual bits in class bitmaps. */
66
67 #define SETBIT(a,b) a[b/8] |= (1 << (b%8))
68
69 /* Maximum length value to check against when making sure that the integer that
70 holds the compiled pattern length does not overflow. We make it a bit less than
71 INT_MAX to allow for adding in group terminating bytes, so that we don't have
72 to check them every time. */
73
74 #define OFLOW_MAX (INT_MAX - 20)
75
76
77 /*************************************************
78 * Code parameters and static tables *
79 *************************************************/
80
81 /* This value specifies the size of stack workspace that is used during the
82 first pre-compile phase that determines how much memory is required. The regex
83 is partly compiled into this space, but the compiled parts are discarded as
84 soon as they can be, so that hopefully there will never be an overrun. The code
85 does, however, check for an overrun. The largest amount I've seen used is 218,
86 so this number is very generous.
87
88 The same workspace is used during the second, actual compile phase for
89 remembering forward references to groups so that they can be filled in at the
90 end. Each entry in this list occupies LINK_SIZE bytes, so even when LINK_SIZE
91 is 4 there is plenty of room. */
92
93 #define COMPILE_WORK_SIZE (4096)
94
95 /* The overrun tests check for a slightly smaller size so that they detect the
96 overrun before it actually does run off the end of the data block. */
97
98 #define WORK_SIZE_CHECK (COMPILE_WORK_SIZE - 100)
99
100
101 /* Table for handling escaped characters in the range '0'-'z'. Positive returns
102 are simple data values; negative values are for special things like \d and so
103 on. Zero means further processing is needed (for things like \x), or the escape
104 is invalid. */
105
106 #ifndef EBCDIC
107
108 /* This is the "normal" table for ASCII systems or for EBCDIC systems running
109 in UTF-8 mode. */
110
111 static const short int escapes[] = {
112 0, 0,
113 0, 0,
114 0, 0,
115 0, 0,
116 0, 0,
117 CHAR_COLON, CHAR_SEMICOLON,
118 CHAR_LESS_THAN_SIGN, CHAR_EQUALS_SIGN,
119 CHAR_GREATER_THAN_SIGN, CHAR_QUESTION_MARK,
120 CHAR_COMMERCIAL_AT, -ESC_A,
121 -ESC_B, -ESC_C,
122 -ESC_D, -ESC_E,
123 0, -ESC_G,
124 -ESC_H, 0,
125 0, -ESC_K,
126 0, 0,
127 -ESC_N, 0,
128 -ESC_P, -ESC_Q,
129 -ESC_R, -ESC_S,
130 0, 0,
131 -ESC_V, -ESC_W,
132 -ESC_X, 0,
133 -ESC_Z, CHAR_LEFT_SQUARE_BRACKET,
134 CHAR_BACKSLASH, CHAR_RIGHT_SQUARE_BRACKET,
135 CHAR_CIRCUMFLEX_ACCENT, CHAR_UNDERSCORE,
136 CHAR_GRAVE_ACCENT, 7,
137 -ESC_b, 0,
138 -ESC_d, ESC_e,
139 ESC_f, 0,
140 -ESC_h, 0,
141 0, -ESC_k,
142 0, 0,
143 ESC_n, 0,
144 -ESC_p, 0,
145 ESC_r, -ESC_s,
146 ESC_tee, 0,
147 -ESC_v, -ESC_w,
148 0, 0,
149 -ESC_z
150 };
151
152 #else
153
154 /* This is the "abnormal" table for EBCDIC systems without UTF-8 support. */
155
156 static const short int escapes[] = {
157 /* 48 */ 0, 0, 0, '.', '<', '(', '+', '|',
158 /* 50 */ '&', 0, 0, 0, 0, 0, 0, 0,
159 /* 58 */ 0, 0, '!', '$', '*', ')', ';', '~',
160 /* 60 */ '-', '/', 0, 0, 0, 0, 0, 0,
161 /* 68 */ 0, 0, '|', ',', '%', '_', '>', '?',
162 /* 70 */ 0, 0, 0, 0, 0, 0, 0, 0,
163 /* 78 */ 0, '`', ':', '#', '@', '\'', '=', '"',
164 /* 80 */ 0, 7, -ESC_b, 0, -ESC_d, ESC_e, ESC_f, 0,
165 /* 88 */-ESC_h, 0, 0, '{', 0, 0, 0, 0,
166 /* 90 */ 0, 0, -ESC_k, 'l', 0, ESC_n, 0, -ESC_p,
167 /* 98 */ 0, ESC_r, 0, '}', 0, 0, 0, 0,
168 /* A0 */ 0, '~', -ESC_s, ESC_tee, 0,-ESC_v, -ESC_w, 0,
169 /* A8 */ 0,-ESC_z, 0, 0, 0, '[', 0, 0,
170 /* B0 */ 0, 0, 0, 0, 0, 0, 0, 0,
171 /* B8 */ 0, 0, 0, 0, 0, ']', '=', '-',
172 /* C0 */ '{',-ESC_A, -ESC_B, -ESC_C, -ESC_D,-ESC_E, 0, -ESC_G,
173 /* C8 */-ESC_H, 0, 0, 0, 0, 0, 0, 0,
174 /* D0 */ '}', 0, -ESC_K, 0, 0,-ESC_N, 0, -ESC_P,
175 /* D8 */-ESC_Q,-ESC_R, 0, 0, 0, 0, 0, 0,
176 /* E0 */ '\\', 0, -ESC_S, 0, 0,-ESC_V, -ESC_W, -ESC_X,
177 /* E8 */ 0,-ESC_Z, 0, 0, 0, 0, 0, 0,
178 /* F0 */ 0, 0, 0, 0, 0, 0, 0, 0,
179 /* F8 */ 0, 0, 0, 0, 0, 0, 0, 0
180 };
181 #endif
182
183
184 /* Table of special "verbs" like (*PRUNE). This is a short table, so it is
185 searched linearly. Put all the names into a single string, in order to reduce
186 the number of relocations when a shared library is dynamically linked. The
187 string is built from string macros so that it works in UTF-8 mode on EBCDIC
188 platforms. */
189
190 typedef struct verbitem {
191 int len; /* Length of verb name */
192 int op; /* Op when no arg, or -1 if arg mandatory */
193 int op_arg; /* Op when arg present, or -1 if not allowed */
194 } verbitem;
195
196 static const char verbnames[] =
197 "\0" /* Empty name is a shorthand for MARK */
198 STRING_MARK0
199 STRING_ACCEPT0
200 STRING_COMMIT0
201 STRING_F0
202 STRING_FAIL0
203 STRING_PRUNE0
204 STRING_SKIP0
205 STRING_THEN;
206
207 static const verbitem verbs[] = {
208 { 0, -1, OP_MARK },
209 { 4, -1, OP_MARK },
210 { 6, OP_ACCEPT, -1 },
211 { 6, OP_COMMIT, -1 },
212 { 1, OP_FAIL, -1 },
213 { 4, OP_FAIL, -1 },
214 { 5, OP_PRUNE, OP_PRUNE_ARG },
215 { 4, OP_SKIP, OP_SKIP_ARG },
216 { 4, OP_THEN, OP_THEN_ARG }
217 };
218
219 static const int verbcount = sizeof(verbs)/sizeof(verbitem);
220
221
222 /* Tables of names of POSIX character classes and their lengths. The names are
223 now all in a single string, to reduce the number of relocations when a shared
224 library is dynamically loaded. The list of lengths is terminated by a zero
225 length entry. The first three must be alpha, lower, upper, as this is assumed
226 for handling case independence. */
227
228 static const char posix_names[] =
229 STRING_alpha0 STRING_lower0 STRING_upper0 STRING_alnum0
230 STRING_ascii0 STRING_blank0 STRING_cntrl0 STRING_digit0
231 STRING_graph0 STRING_print0 STRING_punct0 STRING_space0
232 STRING_word0 STRING_xdigit;
233
234 static const uschar posix_name_lengths[] = {
235 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 4, 6, 0 };
236
237 /* Table of class bit maps for each POSIX class. Each class is formed from a
238 base map, with an optional addition or removal of another map. Then, for some
239 classes, there is some additional tweaking: for [:blank:] the vertical space
240 characters are removed, and for [:alpha:] and [:alnum:] the underscore
241 character is removed. The triples in the table consist of the base map offset,
242 second map offset or -1 if no second map, and a non-negative value for map
243 addition or a negative value for map subtraction (if there are two maps). The
244 absolute value of the third field has these meanings: 0 => no tweaking, 1 =>
245 remove vertical space characters, 2 => remove underscore. */
246
247 static const int posix_class_maps[] = {
248 cbit_word, cbit_digit, -2, /* alpha */
249 cbit_lower, -1, 0, /* lower */
250 cbit_upper, -1, 0, /* upper */
251 cbit_word, -1, 2, /* alnum - word without underscore */
252 cbit_print, cbit_cntrl, 0, /* ascii */
253 cbit_space, -1, 1, /* blank - a GNU extension */
254 cbit_cntrl, -1, 0, /* cntrl */
255 cbit_digit, -1, 0, /* digit */
256 cbit_graph, -1, 0, /* graph */
257 cbit_print, -1, 0, /* print */
258 cbit_punct, -1, 0, /* punct */
259 cbit_space, -1, 0, /* space */
260 cbit_word, -1, 0, /* word - a Perl extension */
261 cbit_xdigit,-1, 0 /* xdigit */
262 };
263
264 /* Table of substitutes for \d etc when PCRE_UCP is set. The POSIX class
265 substitutes must be in the order of the names, defined above, and there are
266 both positive and negative cases. NULL means no substitute. */
267
268 #ifdef SUPPORT_UCP
269 static const uschar *substitutes[] = {
270 (uschar *)"\\P{Nd}", /* \D */
271 (uschar *)"\\p{Nd}", /* \d */
272 (uschar *)"\\P{Xsp}", /* \S */ /* NOTE: Xsp is Perl space */
273 (uschar *)"\\p{Xsp}", /* \s */
274 (uschar *)"\\P{Xwd}", /* \W */
275 (uschar *)"\\p{Xwd}" /* \w */
276 };
277
278 static const uschar *posix_substitutes[] = {
279 (uschar *)"\\p{L}", /* alpha */
280 (uschar *)"\\p{Ll}", /* lower */
281 (uschar *)"\\p{Lu}", /* upper */
282 (uschar *)"\\p{Xan}", /* alnum */
283 NULL, /* ascii */
284 (uschar *)"\\h", /* blank */
285 NULL, /* cntrl */
286 (uschar *)"\\p{Nd}", /* digit */
287 NULL, /* graph */
288 NULL, /* print */
289 NULL, /* punct */
290 (uschar *)"\\p{Xps}", /* space */ /* NOTE: Xps is POSIX space */
291 (uschar *)"\\p{Xwd}", /* word */
292 NULL, /* xdigit */
293 /* Negated cases */
294 (uschar *)"\\P{L}", /* ^alpha */
295 (uschar *)"\\P{Ll}", /* ^lower */
296 (uschar *)"\\P{Lu}", /* ^upper */
297 (uschar *)"\\P{Xan}", /* ^alnum */
298 NULL, /* ^ascii */
299 (uschar *)"\\H", /* ^blank */
300 NULL, /* ^cntrl */
301 (uschar *)"\\P{Nd}", /* ^digit */
302 NULL, /* ^graph */
303 NULL, /* ^print */
304 NULL, /* ^punct */
305 (uschar *)"\\P{Xps}", /* ^space */ /* NOTE: Xps is POSIX space */
306 (uschar *)"\\P{Xwd}", /* ^word */
307 NULL /* ^xdigit */
308 };
309 #define POSIX_SUBSIZE (sizeof(posix_substitutes)/sizeof(uschar *))
310 #endif
311
312 #define STRING(a) # a
313 #define XSTRING(s) STRING(s)
314
315 /* The texts of compile-time error messages. These are "char *" because they
316 are passed to the outside world. Do not ever re-use any error number, because
317 they are documented. Always add a new error instead. Messages marked DEAD below
318 are no longer used. This used to be a table of strings, but in order to reduce
319 the number of relocations needed when a shared library is loaded dynamically,
320 it is now one long string. We cannot use a table of offsets, because the
321 lengths of inserts such as XSTRING(MAX_NAME_SIZE) are not known. Instead, we
322 simply count through to the one we want - this isn't a performance issue
323 because these strings are used only when there is a compilation error.
324
325 Each substring ends with \0 to insert a null character. This includes the final
326 substring, so that the whole string ends with \0\0, which can be detected when
327 counting through. */
328
329 static const char error_texts[] =
330 "no error\0"
331 "\\ at end of pattern\0"
332 "\\c at end of pattern\0"
333 "unrecognized character follows \\\0"
334 "numbers out of order in {} quantifier\0"
335 /* 5 */
336 "number too big in {} quantifier\0"
337 "missing terminating ] for character class\0"
338 "invalid escape sequence in character class\0"
339 "range out of order in character class\0"
340 "nothing to repeat\0"
341 /* 10 */
342 "operand of unlimited repeat could match the empty string\0" /** DEAD **/
343 "internal error: unexpected repeat\0"
344 "unrecognized character after (? or (?-\0"
345 "POSIX named classes are supported only within a class\0"
346 "missing )\0"
347 /* 15 */
348 "reference to non-existent subpattern\0"
349 "erroffset passed as NULL\0"
350 "unknown option bit(s) set\0"
351 "missing ) after comment\0"
352 "parentheses nested too deeply\0" /** DEAD **/
353 /* 20 */
354 "regular expression is too large\0"
355 "failed to get memory\0"
356 "unmatched parentheses\0"
357 "internal error: code overflow\0"
358 "unrecognized character after (?<\0"
359 /* 25 */
360 "lookbehind assertion is not fixed length\0"
361 "malformed number or name after (?(\0"
362 "conditional group contains more than two branches\0"
363 "assertion expected after (?(\0"
364 "(?R or (?[+-]digits must be followed by )\0"
365 /* 30 */
366 "unknown POSIX class name\0"
367 "POSIX collating elements are not supported\0"
368 "this version of PCRE is not compiled with PCRE_UTF8 support\0"
369 "spare error\0" /** DEAD **/
370 "character value in \\x{...} sequence is too large\0"
371 /* 35 */
372 "invalid condition (?(0)\0"
373 "\\C not allowed in lookbehind assertion\0"
374 "PCRE does not support \\L, \\l, \\N{name}, \\U, or \\u\0"
375 "number after (?C is > 255\0"
376 "closing ) for (?C expected\0"
377 /* 40 */
378 "recursive call could loop indefinitely\0"
379 "unrecognized character after (?P\0"
380 "syntax error in subpattern name (missing terminator)\0"
381 "two named subpatterns have the same name\0"
382 "invalid UTF-8 string\0"
383 /* 45 */
384 "support for \\P, \\p, and \\X has not been compiled\0"
385 "malformed \\P or \\p sequence\0"
386 "unknown property name after \\P or \\p\0"
387 "subpattern name is too long (maximum " XSTRING(MAX_NAME_SIZE) " characters)\0"
388 "too many named subpatterns (maximum " XSTRING(MAX_NAME_COUNT) ")\0"
389 /* 50 */
390 "repeated subpattern is too long\0" /** DEAD **/
391 "octal value is greater than \\377 (not in UTF-8 mode)\0"
392 "internal error: overran compiling workspace\0"
393 "internal error: previously-checked referenced subpattern not found\0"
394 "DEFINE group contains more than one branch\0"
395 /* 55 */
396 "repeating a DEFINE group is not allowed\0"
397 "inconsistent NEWLINE options\0"
398 "\\g is not followed by a braced, angle-bracketed, or quoted name/number or by a plain number\0"
399 "a numbered reference must not be zero\0"
400 "an argument is not allowed for (*ACCEPT), (*FAIL), or (*COMMIT)\0"
401 /* 60 */
402 "(*VERB) not recognized\0"
403 "number is too big\0"
404 "subpattern name expected\0"
405 "digit expected after (?+\0"
406 "] is an invalid data character in JavaScript compatibility mode\0"
407 /* 65 */
408 "different names for subpatterns of the same number are not allowed\0"
409 "(*MARK) must have an argument\0"
410 "this version of PCRE is not compiled with PCRE_UCP support\0"
411 "\\c must be followed by an ASCII character\0"
412 "\\k is not followed by a braced, angle-bracketed, or quoted name\0"
413 ;
414
415 /* Table to identify digits and hex digits. This is used when compiling
416 patterns. Note that the tables in chartables are dependent on the locale, and
417 may mark arbitrary characters as digits - but the PCRE compiling code expects
418 to handle only 0-9, a-z, and A-Z as digits when compiling. That is why we have
419 a private table here. It costs 256 bytes, but it is a lot faster than doing
420 character value tests (at least in some simple cases I timed), and in some
421 applications one wants PCRE to compile efficiently as well as match
422 efficiently.
423
424 For convenience, we use the same bit definitions as in chartables:
425
426 0x04 decimal digit
427 0x08 hexadecimal digit
428
429 Then we can use ctype_digit and ctype_xdigit in the code. */
430
431 #ifndef EBCDIC
432
433 /* This is the "normal" case, for ASCII systems, and EBCDIC systems running in
434 UTF-8 mode. */
435
436 static const unsigned char digitab[] =
437 {
438 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 0- 7 */
439 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 8- 15 */
440 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 16- 23 */
441 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 24- 31 */
442 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - ' */
443 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* ( - / */
444 0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c, /* 0 - 7 */
445 0x0c,0x0c,0x00,0x00,0x00,0x00,0x00,0x00, /* 8 - ? */
446 0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* @ - G */
447 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* H - O */
448 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* P - W */
449 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* X - _ */
450 0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* ` - g */
451 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* h - o */
452 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* p - w */
453 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* x -127 */
454 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 128-135 */
455 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 136-143 */
456 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 144-151 */
457 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 152-159 */
458 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 160-167 */
459 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 168-175 */
460 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 176-183 */
461 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 184-191 */
462 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 192-199 */
463 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 200-207 */
464 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 208-215 */
465 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 216-223 */
466 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 224-231 */
467 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 232-239 */
468 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 240-247 */
469 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00};/* 248-255 */
470
471 #else
472
473 /* This is the "abnormal" case, for EBCDIC systems not running in UTF-8 mode. */
474
475 static const unsigned char digitab[] =
476 {
477 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 0- 7 0 */
478 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 8- 15 */
479 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 16- 23 10 */
480 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 24- 31 */
481 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 32- 39 20 */
482 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 40- 47 */
483 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 48- 55 30 */
484 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 56- 63 */
485 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - 71 40 */
486 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 72- | */
487 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* & - 87 50 */
488 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 88- 95 */
489 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - -103 60 */
490 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 104- ? */
491 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 112-119 70 */
492 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 120- " */
493 0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* 128- g 80 */
494 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* h -143 */
495 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 144- p 90 */
496 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* q -159 */
497 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 160- x A0 */
498 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* y -175 */
499 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* ^ -183 B0 */
500 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 184-191 */
501 0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* { - G C0 */
502 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* H -207 */
503 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* } - P D0 */
504 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* Q -223 */
505 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* \ - X E0 */
506 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* Y -239 */
507 0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c, /* 0 - 7 F0 */
508 0x0c,0x0c,0x00,0x00,0x00,0x00,0x00,0x00};/* 8 -255 */
509
510 static const unsigned char ebcdic_chartab[] = { /* chartable partial dup */
511 0x80,0x00,0x00,0x00,0x00,0x01,0x00,0x00, /* 0- 7 */
512 0x00,0x00,0x00,0x00,0x01,0x01,0x00,0x00, /* 8- 15 */
513 0x00,0x00,0x00,0x00,0x00,0x01,0x00,0x00, /* 16- 23 */
514 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 24- 31 */
515 0x00,0x00,0x00,0x00,0x00,0x01,0x00,0x00, /* 32- 39 */
516 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 40- 47 */
517 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 48- 55 */
518 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 56- 63 */
519 0x01,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - 71 */
520 0x00,0x00,0x00,0x80,0x00,0x80,0x80,0x80, /* 72- | */
521 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* & - 87 */
522 0x00,0x00,0x00,0x80,0x80,0x80,0x00,0x00, /* 88- 95 */
523 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - -103 */
524 0x00,0x00,0x00,0x00,0x00,0x10,0x00,0x80, /* 104- ? */
525 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 112-119 */
526 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 120- " */
527 0x00,0x1a,0x1a,0x1a,0x1a,0x1a,0x1a,0x12, /* 128- g */
528 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* h -143 */
529 0x00,0x12,0x12,0x12,0x12,0x12,0x12,0x12, /* 144- p */
530 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* q -159 */
531 0x00,0x00,0x12,0x12,0x12,0x12,0x12,0x12, /* 160- x */
532 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* y -175 */
533 0x80,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* ^ -183 */
534 0x00,0x00,0x80,0x00,0x00,0x00,0x00,0x00, /* 184-191 */
535 0x80,0x1a,0x1a,0x1a,0x1a,0x1a,0x1a,0x12, /* { - G */
536 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* H -207 */
537 0x00,0x12,0x12,0x12,0x12,0x12,0x12,0x12, /* } - P */
538 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* Q -223 */
539 0x00,0x00,0x12,0x12,0x12,0x12,0x12,0x12, /* \ - X */
540 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* Y -239 */
541 0x1c,0x1c,0x1c,0x1c,0x1c,0x1c,0x1c,0x1c, /* 0 - 7 */
542 0x1c,0x1c,0x00,0x00,0x00,0x00,0x00,0x00};/* 8 -255 */
543 #endif
544
545
546 /* Definition to allow mutual recursion */
547
548 static BOOL
549 compile_regex(int, uschar **, const uschar **, int *, BOOL, BOOL, int, int *,
550 int *, branch_chain *, compile_data *, int *);
551
552
553
554 /*************************************************
555 * Find an error text *
556 *************************************************/
557
558 /* The error texts are now all in one long string, to save on relocations. As
559 some of the text is of unknown length, we can't use a table of offsets.
560 Instead, just count through the strings. This is not a performance issue
561 because it happens only when there has been a compilation error.
562
563 Argument: the error number
564 Returns: pointer to the error string
565 */
566
567 static const char *
568 find_error_text(int n)
569 {
570 const char *s = error_texts;
571 for (; n > 0; n--)
572 {
573 while (*s++ != 0) {};
574 if (*s == 0) return "Error text not found (please report)";
575 }
576 return s;
577 }
578
579
580 /*************************************************
581 * Handle escapes *
582 *************************************************/
583
584 /* This function is called when a \ has been encountered. It either returns a
585 positive value for a simple escape such as \n, or a negative value which
586 encodes one of the more complicated things such as \d. A backreference to group
587 n is returned as -(ESC_REF + n); ESC_REF is the highest ESC_xxx macro. When
588 UTF-8 is enabled, a positive value greater than 255 may be returned. On entry,
589 ptr is pointing at the \. On exit, it is on the final character of the escape
590 sequence.
591
592 Arguments:
593 ptrptr points to the pattern position pointer
594 errorcodeptr points to the errorcode variable
595 bracount number of previous extracting brackets
596 options the options bits
597 isclass TRUE if inside a character class
598
599 Returns: zero or positive => a data character
600 negative => a special escape sequence
601 on error, errorcodeptr is set
602 */
603
604 static int
605 check_escape(const uschar **ptrptr, int *errorcodeptr, int bracount,
606 int options, BOOL isclass)
607 {
608 BOOL utf8 = (options & PCRE_UTF8) != 0;
609 const uschar *ptr = *ptrptr + 1;
610 int c, i;
611
612 GETCHARINCTEST(c, ptr); /* Get character value, increment pointer */
613 ptr--; /* Set pointer back to the last byte */
614
615 /* If backslash is at the end of the pattern, it's an error. */
616
617 if (c == 0) *errorcodeptr = ERR1;
618
619 /* Non-alphanumerics are literals. For digits or letters, do an initial lookup
620 in a table. A non-zero result is something that can be returned immediately.
621 Otherwise further processing may be required. */
622
623 #ifndef EBCDIC /* ASCII/UTF-8 coding */
624 else if (c < CHAR_0 || c > CHAR_z) {} /* Not alphanumeric */
625 else if ((i = escapes[c - CHAR_0]) != 0) c = i;
626
627 #else /* EBCDIC coding */
628 else if (c < 'a' || (ebcdic_chartab[c] & 0x0E) == 0) {} /* Not alphanumeric */
629 else if ((i = escapes[c - 0x48]) != 0) c = i;
630 #endif
631
632 /* Escapes that need further processing, or are illegal. */
633
634 else
635 {
636 const uschar *oldptr;
637 BOOL braced, negated;
638
639 switch (c)
640 {
641 /* A number of Perl escapes are not handled by PCRE. We give an explicit
642 error. */
643
644 case CHAR_l:
645 case CHAR_L:
646 case CHAR_u:
647 case CHAR_U:
648 *errorcodeptr = ERR37;
649 break;
650
651 /* \g must be followed by one of a number of specific things:
652
653 (1) A number, either plain or braced. If positive, it is an absolute
654 backreference. If negative, it is a relative backreference. This is a Perl
655 5.10 feature.
656
657 (2) Perl 5.10 also supports \g{name} as a reference to a named group. This
658 is part of Perl's movement towards a unified syntax for back references. As
659 this is synonymous with \k{name}, we fudge it up by pretending it really
660 was \k.
661
662 (3) For Oniguruma compatibility we also support \g followed by a name or a
663 number either in angle brackets or in single quotes. However, these are
664 (possibly recursive) subroutine calls, _not_ backreferences. Just return
665 the -ESC_g code (cf \k). */
666
667 case CHAR_g:
668 if (ptr[1] == CHAR_LESS_THAN_SIGN || ptr[1] == CHAR_APOSTROPHE)
669 {
670 c = -ESC_g;
671 break;
672 }
673
674 /* Handle the Perl-compatible cases */
675
676 if (ptr[1] == CHAR_LEFT_CURLY_BRACKET)
677 {
678 const uschar *p;
679 for (p = ptr+2; *p != 0 && *p != CHAR_RIGHT_CURLY_BRACKET; p++)
680 if (*p != CHAR_MINUS && (digitab[*p] & ctype_digit) == 0) break;
681 if (*p != 0 && *p != CHAR_RIGHT_CURLY_BRACKET)
682 {
683 c = -ESC_k;
684 break;
685 }
686 braced = TRUE;
687 ptr++;
688 }
689 else braced = FALSE;
690
691 if (ptr[1] == CHAR_MINUS)
692 {
693 negated = TRUE;
694 ptr++;
695 }
696 else negated = FALSE;
697
698 c = 0;
699 while ((digitab[ptr[1]] & ctype_digit) != 0)
700 c = c * 10 + *(++ptr) - CHAR_0;
701
702 if (c < 0) /* Integer overflow */
703 {
704 *errorcodeptr = ERR61;
705 break;
706 }
707
708 if (braced && *(++ptr) != CHAR_RIGHT_CURLY_BRACKET)
709 {
710 *errorcodeptr = ERR57;
711 break;
712 }
713
714 if (c == 0)
715 {
716 *errorcodeptr = ERR58;
717 break;
718 }
719
720 if (negated)
721 {
722 if (c > bracount)
723 {
724 *errorcodeptr = ERR15;
725 break;
726 }
727 c = bracount - (c - 1);
728 }
729
730 c = -(ESC_REF + c);
731 break;
732
733 /* The handling of escape sequences consisting of a string of digits
734 starting with one that is not zero is not straightforward. By experiment,
735 the way Perl works seems to be as follows:
736
737 Outside a character class, the digits are read as a decimal number. If the
738 number is less than 10, or if there are that many previous extracting
739 left brackets, then it is a back reference. Otherwise, up to three octal
740 digits are read to form an escaped byte. Thus \123 is likely to be octal
741 123 (cf \0123, which is octal 012 followed by the literal 3). If the octal
742 value is greater than 377, the least significant 8 bits are taken. Inside a
743 character class, \ followed by a digit is always an octal number. */
744
745 case CHAR_1: case CHAR_2: case CHAR_3: case CHAR_4: case CHAR_5:
746 case CHAR_6: case CHAR_7: case CHAR_8: case CHAR_9:
747
748 if (!isclass)
749 {
750 oldptr = ptr;
751 c -= CHAR_0;
752 while ((digitab[ptr[1]] & ctype_digit) != 0)
753 c = c * 10 + *(++ptr) - CHAR_0;
754 if (c < 0) /* Integer overflow */
755 {
756 *errorcodeptr = ERR61;
757 break;
758 }
759 if (c < 10 || c <= bracount)
760 {
761 c = -(ESC_REF + c);
762 break;
763 }
764 ptr = oldptr; /* Put the pointer back and fall through */
765 }
766
767 /* Handle an octal number following \. If the first digit is 8 or 9, Perl
768 generates a binary zero byte and treats the digit as a following literal.
769 Thus we have to pull back the pointer by one. */
770
771 if ((c = *ptr) >= CHAR_8)
772 {
773 ptr--;
774 c = 0;
775 break;
776 }
777
778 /* \0 always starts an octal number, but we may drop through to here with a
779 larger first octal digit. The original code used just to take the least
780 significant 8 bits of octal numbers (I think this is what early Perls used
781 to do). Nowadays we allow for larger numbers in UTF-8 mode, but no more
782 than 3 octal digits. */
783
784 case CHAR_0:
785 c -= CHAR_0;
786 while(i++ < 2 && ptr[1] >= CHAR_0 && ptr[1] <= CHAR_7)
787 c = c * 8 + *(++ptr) - CHAR_0;
788 if (!utf8 && c > 255) *errorcodeptr = ERR51;
789 break;
790
791 /* \x is complicated. \x{ddd} is a character number which can be greater
792 than 0xff in utf8 mode, but only if the ddd are hex digits. If not, { is
793 treated as a data character. */
794
795 case CHAR_x:
796 if (ptr[1] == CHAR_LEFT_CURLY_BRACKET)
797 {
798 const uschar *pt = ptr + 2;
799 int count = 0;
800
801 c = 0;
802 while ((digitab[*pt] & ctype_xdigit) != 0)
803 {
804 register int cc = *pt++;
805 if (c == 0 && cc == CHAR_0) continue; /* Leading zeroes */
806 count++;
807
808 #ifndef EBCDIC /* ASCII/UTF-8 coding */
809 if (cc >= CHAR_a) cc -= 32; /* Convert to upper case */
810 c = (c << 4) + cc - ((cc < CHAR_A)? CHAR_0 : (CHAR_A - 10));
811 #else /* EBCDIC coding */
812 if (cc >= CHAR_a && cc <= CHAR_z) cc += 64; /* Convert to upper case */
813 c = (c << 4) + cc - ((cc >= CHAR_0)? CHAR_0 : (CHAR_A - 10));
814 #endif
815 }
816
817 if (*pt == CHAR_RIGHT_CURLY_BRACKET)
818 {
819 if (c < 0 || count > (utf8? 8 : 2)) *errorcodeptr = ERR34;
820 ptr = pt;
821 break;
822 }
823
824 /* If the sequence of hex digits does not end with '}', then we don't
825 recognize this construct; fall through to the normal \x handling. */
826 }
827
828 /* Read just a single-byte hex-defined char */
829
830 c = 0;
831 while (i++ < 2 && (digitab[ptr[1]] & ctype_xdigit) != 0)
832 {
833 int cc; /* Some compilers don't like */
834 cc = *(++ptr); /* ++ in initializers */
835 #ifndef EBCDIC /* ASCII/UTF-8 coding */
836 if (cc >= CHAR_a) cc -= 32; /* Convert to upper case */
837 c = c * 16 + cc - ((cc < CHAR_A)? CHAR_0 : (CHAR_A - 10));
838 #else /* EBCDIC coding */
839 if (cc <= CHAR_z) cc += 64; /* Convert to upper case */
840 c = c * 16 + cc - ((cc >= CHAR_0)? CHAR_0 : (CHAR_A - 10));
841 #endif
842 }
843 break;
844
845 /* For \c, a following letter is upper-cased; then the 0x40 bit is flipped.
846 An error is given if the byte following \c is not an ASCII character. This
847 coding is ASCII-specific, but then the whole concept of \cx is
848 ASCII-specific. (However, an EBCDIC equivalent has now been added.) */
849
850 case CHAR_c:
851 c = *(++ptr);
852 if (c == 0)
853 {
854 *errorcodeptr = ERR2;
855 break;
856 }
857 #ifndef EBCDIC /* ASCII/UTF-8 coding */
858 if (c > 127) /* Excludes all non-ASCII in either mode */
859 {
860 *errorcodeptr = ERR68;
861 break;
862 }
863 if (c >= CHAR_a && c <= CHAR_z) c -= 32;
864 c ^= 0x40;
865 #else /* EBCDIC coding */
866 if (c >= CHAR_a && c <= CHAR_z) c += 64;
867 c ^= 0xC0;
868 #endif
869 break;
870
871 /* PCRE_EXTRA enables extensions to Perl in the matter of escapes. Any
872 other alphanumeric following \ is an error if PCRE_EXTRA was set;
873 otherwise, for Perl compatibility, it is a literal. This code looks a bit
874 odd, but there used to be some cases other than the default, and there may
875 be again in future, so I haven't "optimized" it. */
876
877 default:
878 if ((options & PCRE_EXTRA) != 0) switch(c)
879 {
880 default:
881 *errorcodeptr = ERR3;
882 break;
883 }
884 break;
885 }
886 }
887
888 /* Perl supports \N{name} for character names, as well as plain \N for "not
889 newline". PCRE does not support \N{name}. */
890
891 if (c == -ESC_N && ptr[1] == CHAR_LEFT_CURLY_BRACKET)
892 *errorcodeptr = ERR37;
893
894 /* If PCRE_UCP is set, we change the values for \d etc. */
895
896 if ((options & PCRE_UCP) != 0 && c <= -ESC_D && c >= -ESC_w)
897 c -= (ESC_DU - ESC_D);
898
899 /* Set the pointer to the final character before returning. */
900
901 *ptrptr = ptr;
902 return c;
903 }
904
905
906
907 #ifdef SUPPORT_UCP
908 /*************************************************
909 * Handle \P and \p *
910 *************************************************/
911
912 /* This function is called after \P or \p has been encountered, provided that
913 PCRE is compiled with support for Unicode properties. On entry, ptrptr is
914 pointing at the P or p. On exit, it is pointing at the final character of the
915 escape sequence.
916
917 Argument:
918 ptrptr points to the pattern position pointer
919 negptr points to a boolean that is set TRUE for negation else FALSE
920 dptr points to an int that is set to the detailed property value
921 errorcodeptr points to the error code variable
922
923 Returns: type value from ucp_type_table, or -1 for an invalid type
924 */
925
926 static int
927 get_ucp(const uschar **ptrptr, BOOL *negptr, int *dptr, int *errorcodeptr)
928 {
929 int c, i, bot, top;
930 const uschar *ptr = *ptrptr;
931 char name[32];
932
933 c = *(++ptr);
934 if (c == 0) goto ERROR_RETURN;
935
936 *negptr = FALSE;
937
938 /* \P or \p can be followed by a name in {}, optionally preceded by ^ for
939 negation. */
940
941 if (c == CHAR_LEFT_CURLY_BRACKET)
942 {
943 if (ptr[1] == CHAR_CIRCUMFLEX_ACCENT)
944 {
945 *negptr = TRUE;
946 ptr++;
947 }
948 for (i = 0; i < (int)sizeof(name) - 1; i++)
949 {
950 c = *(++ptr);
951 if (c == 0) goto ERROR_RETURN;
952 if (c == CHAR_RIGHT_CURLY_BRACKET) break;
953 name[i] = c;
954 }
955 if (c != CHAR_RIGHT_CURLY_BRACKET) goto ERROR_RETURN;
956 name[i] = 0;
957 }
958
959 /* Otherwise there is just one following character */
960
961 else
962 {
963 name[0] = c;
964 name[1] = 0;
965 }
966
967 *ptrptr = ptr;
968
969 /* Search for a recognized property name using binary chop */
970
971 bot = 0;
972 top = _pcre_utt_size;
973
974 while (bot < top)
975 {
976 i = (bot + top) >> 1;
977 c = strcmp(name, _pcre_utt_names + _pcre_utt[i].name_offset);
978 if (c == 0)
979 {
980 *dptr = _pcre_utt[i].value;
981 return _pcre_utt[i].type;
982 }
983 if (c > 0) bot = i + 1; else top = i;
984 }
985
986 *errorcodeptr = ERR47;
987 *ptrptr = ptr;
988 return -1;
989
990 ERROR_RETURN:
991 *errorcodeptr = ERR46;
992 *ptrptr = ptr;
993 return -1;
994 }
995 #endif
996
997
998
999
1000 /*************************************************
1001 * Check for counted repeat *
1002 *************************************************/
1003
1004 /* This function is called when a '{' is encountered in a place where it might
1005 start a quantifier. It looks ahead to see if it really is a quantifier or not.
1006 It is only a quantifier if it is one of the forms {ddd} {ddd,} or {ddd,ddd}
1007 where the ddds are digits.
1008
1009 Arguments:
1010 p pointer to the first char after '{'
1011
1012 Returns: TRUE or FALSE
1013 */
1014
1015 static BOOL
1016 is_counted_repeat(const uschar *p)
1017 {
1018 if ((digitab[*p++] & ctype_digit) == 0) return FALSE;
1019 while ((digitab[*p] & ctype_digit) != 0) p++;
1020 if (*p == CHAR_RIGHT_CURLY_BRACKET) return TRUE;
1021
1022 if (*p++ != CHAR_COMMA) return FALSE;
1023 if (*p == CHAR_RIGHT_CURLY_BRACKET) return TRUE;
1024
1025 if ((digitab[*p++] & ctype_digit) == 0) return FALSE;
1026 while ((digitab[*p] & ctype_digit) != 0) p++;
1027
1028 return (*p == CHAR_RIGHT_CURLY_BRACKET);
1029 }
1030
1031
1032
1033 /*************************************************
1034 * Read repeat counts *
1035 *************************************************/
1036
1037 /* Read an item of the form {n,m} and return the values. This is called only
1038 after is_counted_repeat() has confirmed that a repeat-count quantifier exists,
1039 so the syntax is guaranteed to be correct, but we need to check the values.
1040
1041 Arguments:
1042 p pointer to first char after '{'
1043 minp pointer to int for min
1044 maxp pointer to int for max
1045 returned as -1 if no max
1046 errorcodeptr points to error code variable
1047
1048 Returns: pointer to '}' on success;
1049 current ptr on error, with errorcodeptr set non-zero
1050 */
1051
1052 static const uschar *
1053 read_repeat_counts(const uschar *p, int *minp, int *maxp, int *errorcodeptr)
1054 {
1055 int min = 0;
1056 int max = -1;
1057
1058 /* Read the minimum value and do a paranoid check: a negative value indicates
1059 an integer overflow. */
1060
1061 while ((digitab[*p] & ctype_digit) != 0) min = min * 10 + *p++ - CHAR_0;
1062 if (min < 0 || min > 65535)
1063 {
1064 *errorcodeptr = ERR5;
1065 return p;
1066 }
1067
1068 /* Read the maximum value if there is one, and again do a paranoid on its size.
1069 Also, max must not be less than min. */
1070
1071 if (*p == CHAR_RIGHT_CURLY_BRACKET) max = min; else
1072 {
1073 if (*(++p) != CHAR_RIGHT_CURLY_BRACKET)
1074 {
1075 max = 0;
1076 while((digitab[*p] & ctype_digit) != 0) max = max * 10 + *p++ - CHAR_0;
1077 if (max < 0 || max > 65535)
1078 {
1079 *errorcodeptr = ERR5;
1080 return p;
1081 }
1082 if (max < min)
1083 {
1084 *errorcodeptr = ERR4;
1085 return p;
1086 }
1087 }
1088 }
1089
1090 /* Fill in the required variables, and pass back the pointer to the terminating
1091 '}'. */
1092
1093 *minp = min;
1094 *maxp = max;
1095 return p;
1096 }
1097
1098
1099
1100 /*************************************************
1101 * Subroutine for finding forward reference *
1102 *************************************************/
1103
1104 /* This recursive function is called only from find_parens() below. The
1105 top-level call starts at the beginning of the pattern. All other calls must
1106 start at a parenthesis. It scans along a pattern's text looking for capturing
1107 subpatterns, and counting them. If it finds a named pattern that matches the
1108 name it is given, it returns its number. Alternatively, if the name is NULL, it
1109 returns when it reaches a given numbered subpattern. Recursion is used to keep
1110 track of subpatterns that reset the capturing group numbers - the (?| feature.
1111
1112 This function was originally called only from the second pass, in which we know
1113 that if (?< or (?' or (?P< is encountered, the name will be correctly
1114 terminated because that is checked in the first pass. There is now one call to
1115 this function in the first pass, to check for a recursive back reference by
1116 name (so that we can make the whole group atomic). In this case, we need check
1117 only up to the current position in the pattern, and that is still OK because
1118 and previous occurrences will have been checked. To make this work, the test
1119 for "end of pattern" is a check against cd->end_pattern in the main loop,
1120 instead of looking for a binary zero. This means that the special first-pass
1121 call can adjust cd->end_pattern temporarily. (Checks for binary zero while
1122 processing items within the loop are OK, because afterwards the main loop will
1123 terminate.)
1124
1125 Arguments:
1126 ptrptr address of the current character pointer (updated)
1127 cd compile background data
1128 name name to seek, or NULL if seeking a numbered subpattern
1129 lorn name length, or subpattern number if name is NULL
1130 xmode TRUE if we are in /x mode
1131 utf8 TRUE if we are in UTF-8 mode
1132 count pointer to the current capturing subpattern number (updated)
1133
1134 Returns: the number of the named subpattern, or -1 if not found
1135 */
1136
1137 static int
1138 find_parens_sub(uschar **ptrptr, compile_data *cd, const uschar *name, int lorn,
1139 BOOL xmode, BOOL utf8, int *count)
1140 {
1141 uschar *ptr = *ptrptr;
1142 int start_count = *count;
1143 int hwm_count = start_count;
1144 BOOL dup_parens = FALSE;
1145
1146 /* If the first character is a parenthesis, check on the type of group we are
1147 dealing with. The very first call may not start with a parenthesis. */
1148
1149 if (ptr[0] == CHAR_LEFT_PARENTHESIS)
1150 {
1151 /* Handle specials such as (*SKIP) or (*UTF8) etc. */
1152
1153 if (ptr[1] == CHAR_ASTERISK) ptr += 2;
1154
1155 /* Handle a normal, unnamed capturing parenthesis. */
1156
1157 else if (ptr[1] != CHAR_QUESTION_MARK)
1158 {
1159 *count += 1;
1160 if (name == NULL && *count == lorn) return *count;
1161 ptr++;
1162 }
1163
1164 /* All cases now have (? at the start. Remember when we are in a group
1165 where the parenthesis numbers are duplicated. */
1166
1167 else if (ptr[2] == CHAR_VERTICAL_LINE)
1168 {
1169 ptr += 3;
1170 dup_parens = TRUE;
1171 }
1172
1173 /* Handle comments; all characters are allowed until a ket is reached. */
1174
1175 else if (ptr[2] == CHAR_NUMBER_SIGN)
1176 {
1177 for (ptr += 3; *ptr != 0; ptr++) if (*ptr == CHAR_RIGHT_PARENTHESIS) break;
1178 goto FAIL_EXIT;
1179 }
1180
1181 /* Handle a condition. If it is an assertion, just carry on so that it
1182 is processed as normal. If not, skip to the closing parenthesis of the
1183 condition (there can't be any nested parens). */
1184
1185 else if (ptr[2] == CHAR_LEFT_PARENTHESIS)
1186 {
1187 ptr += 2;
1188 if (ptr[1] != CHAR_QUESTION_MARK)
1189 {
1190 while (*ptr != 0 && *ptr != CHAR_RIGHT_PARENTHESIS) ptr++;
1191 if (*ptr != 0) ptr++;
1192 }
1193 }
1194
1195 /* Start with (? but not a condition. */
1196
1197 else
1198 {
1199 ptr += 2;
1200 if (*ptr == CHAR_P) ptr++; /* Allow optional P */
1201
1202 /* We have to disambiguate (?<! and (?<= from (?<name> for named groups */
1203
1204 if ((*ptr == CHAR_LESS_THAN_SIGN && ptr[1] != CHAR_EXCLAMATION_MARK &&
1205 ptr[1] != CHAR_EQUALS_SIGN) || *ptr == CHAR_APOSTROPHE)
1206 {
1207 int term;
1208 const uschar *thisname;
1209 *count += 1;
1210 if (name == NULL && *count == lorn) return *count;
1211 term = *ptr++;
1212 if (term == CHAR_LESS_THAN_SIGN) term = CHAR_GREATER_THAN_SIGN;
1213 thisname = ptr;
1214 while (*ptr != term) ptr++;
1215 if (name != NULL && lorn == ptr - thisname &&
1216 strncmp((const char *)name, (const char *)thisname, lorn) == 0)
1217 return *count;
1218 term++;
1219 }
1220 }
1221 }
1222
1223 /* Past any initial parenthesis handling, scan for parentheses or vertical
1224 bars. Stop if we get to cd->end_pattern. Note that this is important for the
1225 first-pass call when this value is temporarily adjusted to stop at the current
1226 position. So DO NOT change this to a test for binary zero. */
1227
1228 for (; ptr < cd->end_pattern; ptr++)
1229 {
1230 /* Skip over backslashed characters and also entire \Q...\E */
1231
1232 if (*ptr == CHAR_BACKSLASH)
1233 {
1234 if (*(++ptr) == 0) goto FAIL_EXIT;
1235 if (*ptr == CHAR_Q) for (;;)
1236 {
1237 while (*(++ptr) != 0 && *ptr != CHAR_BACKSLASH) {};
1238 if (*ptr == 0) goto FAIL_EXIT;
1239 if (*(++ptr) == CHAR_E) break;
1240 }
1241 continue;
1242 }
1243
1244 /* Skip over character classes; this logic must be similar to the way they
1245 are handled for real. If the first character is '^', skip it. Also, if the
1246 first few characters (either before or after ^) are \Q\E or \E we skip them
1247 too. This makes for compatibility with Perl. Note the use of STR macros to
1248 encode "Q\\E" so that it works in UTF-8 on EBCDIC platforms. */
1249
1250 if (*ptr == CHAR_LEFT_SQUARE_BRACKET)
1251 {
1252 BOOL negate_class = FALSE;
1253 for (;;)
1254 {
1255 if (ptr[1] == CHAR_BACKSLASH)
1256 {
1257 if (ptr[2] == CHAR_E)
1258 ptr+= 2;
1259 else if (strncmp((const char *)ptr+2,
1260 STR_Q STR_BACKSLASH STR_E, 3) == 0)
1261 ptr += 4;
1262 else
1263 break;
1264 }
1265 else if (!negate_class && ptr[1] == CHAR_CIRCUMFLEX_ACCENT)
1266 {
1267 negate_class = TRUE;
1268 ptr++;
1269 }
1270 else break;
1271 }
1272
1273 /* If the next character is ']', it is a data character that must be
1274 skipped, except in JavaScript compatibility mode. */
1275
1276 if (ptr[1] == CHAR_RIGHT_SQUARE_BRACKET &&
1277 (cd->external_options & PCRE_JAVASCRIPT_COMPAT) == 0)
1278 ptr++;
1279
1280 while (*(++ptr) != CHAR_RIGHT_SQUARE_BRACKET)
1281 {
1282 if (*ptr == 0) return -1;
1283 if (*ptr == CHAR_BACKSLASH)
1284 {
1285 if (*(++ptr) == 0) goto FAIL_EXIT;
1286 if (*ptr == CHAR_Q) for (;;)
1287 {
1288 while (*(++ptr) != 0 && *ptr != CHAR_BACKSLASH) {};
1289 if (*ptr == 0) goto FAIL_EXIT;
1290 if (*(++ptr) == CHAR_E) break;
1291 }
1292 continue;
1293 }
1294 }
1295 continue;
1296 }
1297
1298 /* Skip comments in /x mode */
1299
1300 if (xmode && *ptr == CHAR_NUMBER_SIGN)
1301 {
1302 ptr++;
1303 while (*ptr != 0)
1304 {
1305 if (IS_NEWLINE(ptr)) { ptr += cd->nllen - 1; break; }
1306 ptr++;
1307 #ifdef SUPPORT_UTF8
1308 if (utf8) while ((*ptr & 0xc0) == 0x80) ptr++;
1309 #endif
1310 }
1311 if (*ptr == 0) goto FAIL_EXIT;
1312 continue;
1313 }
1314
1315 /* Check for the special metacharacters */
1316
1317 if (*ptr == CHAR_LEFT_PARENTHESIS)
1318 {
1319 int rc = find_parens_sub(&ptr, cd, name, lorn, xmode, utf8, count);
1320 if (rc > 0) return rc;
1321 if (*ptr == 0) goto FAIL_EXIT;
1322 }
1323
1324 else if (*ptr == CHAR_RIGHT_PARENTHESIS)
1325 {
1326 if (dup_parens && *count < hwm_count) *count = hwm_count;
1327 goto FAIL_EXIT;
1328 }
1329
1330 else if (*ptr == CHAR_VERTICAL_LINE && dup_parens)
1331 {
1332 if (*count > hwm_count) hwm_count = *count;
1333 *count = start_count;
1334 }
1335 }
1336
1337 FAIL_EXIT:
1338 *ptrptr = ptr;
1339 return -1;
1340 }
1341
1342
1343
1344
1345 /*************************************************
1346 * Find forward referenced subpattern *
1347 *************************************************/
1348
1349 /* This function scans along a pattern's text looking for capturing
1350 subpatterns, and counting them. If it finds a named pattern that matches the
1351 name it is given, it returns its number. Alternatively, if the name is NULL, it
1352 returns when it reaches a given numbered subpattern. This is used for forward
1353 references to subpatterns. We used to be able to start this scan from the
1354 current compiling point, using the current count value from cd->bracount, and
1355 do it all in a single loop, but the addition of the possibility of duplicate
1356 subpattern numbers means that we have to scan from the very start, in order to
1357 take account of such duplicates, and to use a recursive function to keep track
1358 of the different types of group.
1359
1360 Arguments:
1361 cd compile background data
1362 name name to seek, or NULL if seeking a numbered subpattern
1363 lorn name length, or subpattern number if name is NULL
1364 xmode TRUE if we are in /x mode
1365 utf8 TRUE if we are in UTF-8 mode
1366
1367 Returns: the number of the found subpattern, or -1 if not found
1368 */
1369
1370 static int
1371 find_parens(compile_data *cd, const uschar *name, int lorn, BOOL xmode,
1372 BOOL utf8)
1373 {
1374 uschar *ptr = (uschar *)cd->start_pattern;
1375 int count = 0;
1376 int rc;
1377
1378 /* If the pattern does not start with an opening parenthesis, the first call
1379 to find_parens_sub() will scan right to the end (if necessary). However, if it
1380 does start with a parenthesis, find_parens_sub() will return when it hits the
1381 matching closing parens. That is why we have to have a loop. */
1382
1383 for (;;)
1384 {
1385 rc = find_parens_sub(&ptr, cd, name, lorn, xmode, utf8, &count);
1386 if (rc > 0 || *ptr++ == 0) break;
1387 }
1388
1389 return rc;
1390 }
1391
1392
1393
1394
1395 /*************************************************
1396 * Find first significant op code *
1397 *************************************************/
1398
1399 /* This is called by several functions that scan a compiled expression looking
1400 for a fixed first character, or an anchoring op code etc. It skips over things
1401 that do not influence this. For some calls, it makes sense to skip negative
1402 forward and all backward assertions, and also the \b assertion; for others it
1403 does not.
1404
1405 Arguments:
1406 code pointer to the start of the group
1407 skipassert TRUE if certain assertions are to be skipped
1408
1409 Returns: pointer to the first significant opcode
1410 */
1411
1412 static const uschar*
1413 first_significant_code(const uschar *code, BOOL skipassert)
1414 {
1415 for (;;)
1416 {
1417 switch ((int)*code)
1418 {
1419 case OP_ASSERT_NOT:
1420 case OP_ASSERTBACK:
1421 case OP_ASSERTBACK_NOT:
1422 if (!skipassert) return code;
1423 do code += GET(code, 1); while (*code == OP_ALT);
1424 code += _pcre_OP_lengths[*code];
1425 break;
1426
1427 case OP_WORD_BOUNDARY:
1428 case OP_NOT_WORD_BOUNDARY:
1429 if (!skipassert) return code;
1430 /* Fall through */
1431
1432 case OP_CALLOUT:
1433 case OP_CREF:
1434 case OP_NCREF:
1435 case OP_RREF:
1436 case OP_NRREF:
1437 case OP_DEF:
1438 code += _pcre_OP_lengths[*code];
1439 break;
1440
1441 default:
1442 return code;
1443 }
1444 }
1445 /* Control never reaches here */
1446 }
1447
1448
1449
1450
1451 /*************************************************
1452 * Find the fixed length of a branch *
1453 *************************************************/
1454
1455 /* Scan a branch and compute the fixed length of subject that will match it,
1456 if the length is fixed. This is needed for dealing with backward assertions.
1457 In UTF8 mode, the result is in characters rather than bytes. The branch is
1458 temporarily terminated with OP_END when this function is called.
1459
1460 This function is called when a backward assertion is encountered, so that if it
1461 fails, the error message can point to the correct place in the pattern.
1462 However, we cannot do this when the assertion contains subroutine calls,
1463 because they can be forward references. We solve this by remembering this case
1464 and doing the check at the end; a flag specifies which mode we are running in.
1465
1466 Arguments:
1467 code points to the start of the pattern (the bracket)
1468 utf8 TRUE in UTF-8 mode
1469 atend TRUE if called when the pattern is complete
1470 cd the "compile data" structure
1471
1472 Returns: the fixed length,
1473 or -1 if there is no fixed length,
1474 or -2 if \C was encountered
1475 or -3 if an OP_RECURSE item was encountered and atend is FALSE
1476 */
1477
1478 static int
1479 find_fixedlength(uschar *code, BOOL utf8, BOOL atend, compile_data *cd)
1480 {
1481 int length = -1;
1482
1483 register int branchlength = 0;
1484 register uschar *cc = code + 1 + LINK_SIZE;
1485
1486 /* Scan along the opcodes for this branch. If we get to the end of the
1487 branch, check the length against that of the other branches. */
1488
1489 for (;;)
1490 {
1491 int d;
1492 uschar *ce, *cs;
1493 register int op = *cc;
1494 switch (op)
1495 {
1496 /* We only need to continue for OP_CBRA (normal capturing bracket) and
1497 OP_BRA (normal non-capturing bracket) because the other variants of these
1498 opcodes are all concerned with unlimited repeated groups, which of course
1499 are not of fixed length. They will cause a -1 response from the default
1500 case of this switch. */
1501
1502 case OP_CBRA:
1503 case OP_BRA:
1504 case OP_ONCE:
1505 case OP_COND:
1506 d = find_fixedlength(cc + ((op == OP_CBRA)? 2:0), utf8, atend, cd);
1507 if (d < 0) return d;
1508 branchlength += d;
1509 do cc += GET(cc, 1); while (*cc == OP_ALT);
1510 cc += 1 + LINK_SIZE;
1511 break;
1512
1513 /* Reached end of a branch; if it's a ket it is the end of a nested
1514 call. If it's ALT it is an alternation in a nested call. If it is
1515 END it's the end of the outer call. All can be handled by the same code.
1516 Note that we must not include the OP_KETRxxx opcodes here, because they
1517 all imply an unlimited repeat. */
1518
1519 case OP_ALT:
1520 case OP_KET:
1521 case OP_END:
1522 if (length < 0) length = branchlength;
1523 else if (length != branchlength) return -1;
1524 if (*cc != OP_ALT) return length;
1525 cc += 1 + LINK_SIZE;
1526 branchlength = 0;
1527 break;
1528
1529 /* A true recursion implies not fixed length, but a subroutine call may
1530 be OK. If the subroutine is a forward reference, we can't deal with
1531 it until the end of the pattern, so return -3. */
1532
1533 case OP_RECURSE:
1534 if (!atend) return -3;
1535 cs = ce = (uschar *)cd->start_code + GET(cc, 1); /* Start subpattern */
1536 do ce += GET(ce, 1); while (*ce == OP_ALT); /* End subpattern */
1537 if (cc > cs && cc < ce) return -1; /* Recursion */
1538 d = find_fixedlength(cs + 2, utf8, atend, cd);
1539 if (d < 0) return d;
1540 branchlength += d;
1541 cc += 1 + LINK_SIZE;
1542 break;
1543
1544 /* Skip over assertive subpatterns */
1545
1546 case OP_ASSERT:
1547 case OP_ASSERT_NOT:
1548 case OP_ASSERTBACK:
1549 case OP_ASSERTBACK_NOT:
1550 do cc += GET(cc, 1); while (*cc == OP_ALT);
1551 /* Fall through */
1552
1553 /* Skip over things that don't match chars */
1554
1555 case OP_REVERSE:
1556 case OP_CREF:
1557 case OP_NCREF:
1558 case OP_RREF:
1559 case OP_NRREF:
1560 case OP_DEF:
1561 case OP_CALLOUT:
1562 case OP_SOD:
1563 case OP_SOM:
1564 case OP_SET_SOM:
1565 case OP_EOD:
1566 case OP_EODN:
1567 case OP_CIRC:
1568 case OP_CIRCM:
1569 case OP_DOLL:
1570 case OP_DOLLM:
1571 case OP_NOT_WORD_BOUNDARY:
1572 case OP_WORD_BOUNDARY:
1573 cc += _pcre_OP_lengths[*cc];
1574 break;
1575
1576 /* Handle literal characters */
1577
1578 case OP_CHAR:
1579 case OP_CHARI:
1580 case OP_NOT:
1581 case OP_NOTI:
1582 branchlength++;
1583 cc += 2;
1584 #ifdef SUPPORT_UTF8
1585 if (utf8 && cc[-1] >= 0xc0) cc += _pcre_utf8_table4[cc[-1] & 0x3f];
1586 #endif
1587 break;
1588
1589 /* Handle exact repetitions. The count is already in characters, but we
1590 need to skip over a multibyte character in UTF8 mode. */
1591
1592 case OP_EXACT:
1593 branchlength += GET2(cc,1);
1594 cc += 4;
1595 #ifdef SUPPORT_UTF8
1596 if (utf8 && cc[-1] >= 0xc0) cc += _pcre_utf8_table4[cc[-1] & 0x3f];
1597 #endif
1598 break;
1599
1600 case OP_TYPEEXACT:
1601 branchlength += GET2(cc,1);
1602 if (cc[3] == OP_PROP || cc[3] == OP_NOTPROP) cc += 2;
1603 cc += 4;
1604 break;
1605
1606 /* Handle single-char matchers */
1607
1608 case OP_PROP:
1609 case OP_NOTPROP:
1610 cc += 2;
1611 /* Fall through */
1612
1613 case OP_NOT_DIGIT:
1614 case OP_DIGIT:
1615 case OP_NOT_WHITESPACE:
1616 case OP_WHITESPACE:
1617 case OP_NOT_WORDCHAR:
1618 case OP_WORDCHAR:
1619 case OP_ANY:
1620 case OP_ALLANY:
1621 branchlength++;
1622 cc++;
1623 break;
1624
1625 /* The single-byte matcher isn't allowed */
1626
1627 case OP_ANYBYTE:
1628 return -2;
1629
1630 /* Check a class for variable quantification */
1631
1632 #ifdef SUPPORT_UTF8
1633 case OP_XCLASS:
1634 cc += GET(cc, 1) - 33;
1635 /* Fall through */
1636 #endif
1637
1638 case OP_CLASS:
1639 case OP_NCLASS:
1640 cc += 33;
1641
1642 switch (*cc)
1643 {
1644 case OP_CRSTAR:
1645 case OP_CRMINSTAR:
1646 case OP_CRQUERY:
1647 case OP_CRMINQUERY:
1648 return -1;
1649
1650 case OP_CRRANGE:
1651 case OP_CRMINRANGE:
1652 if (GET2(cc,1) != GET2(cc,3)) return -1;
1653 branchlength += GET2(cc,1);
1654 cc += 5;
1655 break;
1656
1657 default:
1658 branchlength++;
1659 }
1660 break;
1661
1662 /* Anything else is variable length */
1663
1664 default:
1665 return -1;
1666 }
1667 }
1668 /* Control never gets here */
1669 }
1670
1671
1672
1673
1674 /*************************************************
1675 * Scan compiled regex for specific bracket *
1676 *************************************************/
1677
1678 /* This little function scans through a compiled pattern until it finds a
1679 capturing bracket with the given number, or, if the number is negative, an
1680 instance of OP_REVERSE for a lookbehind. The function is global in the C sense
1681 so that it can be called from pcre_study() when finding the minimum matching
1682 length.
1683
1684 Arguments:
1685 code points to start of expression
1686 utf8 TRUE in UTF-8 mode
1687 number the required bracket number or negative to find a lookbehind
1688
1689 Returns: pointer to the opcode for the bracket, or NULL if not found
1690 */
1691
1692 const uschar *
1693 _pcre_find_bracket(const uschar *code, BOOL utf8, int number)
1694 {
1695 for (;;)
1696 {
1697 register int c = *code;
1698
1699 if (c == OP_END) return NULL;
1700
1701 /* XCLASS is used for classes that cannot be represented just by a bit
1702 map. This includes negated single high-valued characters. The length in
1703 the table is zero; the actual length is stored in the compiled code. */
1704
1705 if (c == OP_XCLASS) code += GET(code, 1);
1706
1707 /* Handle recursion */
1708
1709 else if (c == OP_REVERSE)
1710 {
1711 if (number < 0) return (uschar *)code;
1712 code += _pcre_OP_lengths[c];
1713 }
1714
1715 /* Handle capturing bracket */
1716
1717 else if (c == OP_CBRA || c == OP_SCBRA ||
1718 c == OP_CBRAPOS || c == OP_SCBRAPOS)
1719 {
1720 int n = GET2(code, 1+LINK_SIZE);
1721 if (n == number) return (uschar *)code;
1722 code += _pcre_OP_lengths[c];
1723 }
1724
1725 /* Otherwise, we can get the item's length from the table, except that for
1726 repeated character types, we have to test for \p and \P, which have an extra
1727 two bytes of parameters, and for MARK/PRUNE/SKIP/THEN with an argument, we
1728 must add in its length. */
1729
1730 else
1731 {
1732 switch(c)
1733 {
1734 case OP_TYPESTAR:
1735 case OP_TYPEMINSTAR:
1736 case OP_TYPEPLUS:
1737 case OP_TYPEMINPLUS:
1738 case OP_TYPEQUERY:
1739 case OP_TYPEMINQUERY:
1740 case OP_TYPEPOSSTAR:
1741 case OP_TYPEPOSPLUS:
1742 case OP_TYPEPOSQUERY:
1743 if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
1744 break;
1745
1746 case OP_TYPEUPTO:
1747 case OP_TYPEMINUPTO:
1748 case OP_TYPEEXACT:
1749 case OP_TYPEPOSUPTO:
1750 if (code[3] == OP_PROP || code[3] == OP_NOTPROP) code += 2;
1751 break;
1752
1753 case OP_MARK:
1754 case OP_PRUNE_ARG:
1755 case OP_SKIP_ARG:
1756 code += code[1];
1757 break;
1758
1759 case OP_THEN_ARG:
1760 code += code[1+LINK_SIZE];
1761 break;
1762 }
1763
1764 /* Add in the fixed length from the table */
1765
1766 code += _pcre_OP_lengths[c];
1767
1768 /* In UTF-8 mode, opcodes that are followed by a character may be followed by
1769 a multi-byte character. The length in the table is a minimum, so we have to
1770 arrange to skip the extra bytes. */
1771
1772 #ifdef SUPPORT_UTF8
1773 if (utf8) switch(c)
1774 {
1775 case OP_CHAR:
1776 case OP_CHARI:
1777 case OP_EXACT:
1778 case OP_EXACTI:
1779 case OP_UPTO:
1780 case OP_UPTOI:
1781 case OP_MINUPTO:
1782 case OP_MINUPTOI:
1783 case OP_POSUPTO:
1784 case OP_POSUPTOI:
1785 case OP_STAR:
1786 case OP_STARI:
1787 case OP_MINSTAR:
1788 case OP_MINSTARI:
1789 case OP_POSSTAR:
1790 case OP_POSSTARI:
1791 case OP_PLUS:
1792 case OP_PLUSI:
1793 case OP_MINPLUS:
1794 case OP_MINPLUSI:
1795 case OP_POSPLUS:
1796 case OP_POSPLUSI:
1797 case OP_QUERY:
1798 case OP_QUERYI:
1799 case OP_MINQUERY:
1800 case OP_MINQUERYI:
1801 case OP_POSQUERY:
1802 case OP_POSQUERYI:
1803 if (code[-1] >= 0xc0) code += _pcre_utf8_table4[code[-1] & 0x3f];
1804 break;
1805 }
1806 #else
1807 (void)(utf8); /* Keep compiler happy by referencing function argument */
1808 #endif
1809 }
1810 }
1811 }
1812
1813
1814
1815 /*************************************************
1816 * Scan compiled regex for recursion reference *
1817 *************************************************/
1818
1819 /* This little function scans through a compiled pattern until it finds an
1820 instance of OP_RECURSE.
1821
1822 Arguments:
1823 code points to start of expression
1824 utf8 TRUE in UTF-8 mode
1825
1826 Returns: pointer to the opcode for OP_RECURSE, or NULL if not found
1827 */
1828
1829 static const uschar *
1830 find_recurse(const uschar *code, BOOL utf8)
1831 {
1832 for (;;)
1833 {
1834 register int c = *code;
1835 if (c == OP_END) return NULL;
1836 if (c == OP_RECURSE) return code;
1837
1838 /* XCLASS is used for classes that cannot be represented just by a bit
1839 map. This includes negated single high-valued characters. The length in
1840 the table is zero; the actual length is stored in the compiled code. */
1841
1842 if (c == OP_XCLASS) code += GET(code, 1);
1843
1844 /* Otherwise, we can get the item's length from the table, except that for
1845 repeated character types, we have to test for \p and \P, which have an extra
1846 two bytes of parameters, and for MARK/PRUNE/SKIP/THEN with an argument, we
1847 must add in its length. */
1848
1849 else
1850 {
1851 switch(c)
1852 {
1853 case OP_TYPESTAR:
1854 case OP_TYPEMINSTAR:
1855 case OP_TYPEPLUS:
1856 case OP_TYPEMINPLUS:
1857 case OP_TYPEQUERY:
1858 case OP_TYPEMINQUERY:
1859 case OP_TYPEPOSSTAR:
1860 case OP_TYPEPOSPLUS:
1861 case OP_TYPEPOSQUERY:
1862 if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
1863 break;
1864
1865 case OP_TYPEPOSUPTO:
1866 case OP_TYPEUPTO:
1867 case OP_TYPEMINUPTO:
1868 case OP_TYPEEXACT:
1869 if (code[3] == OP_PROP || code[3] == OP_NOTPROP) code += 2;
1870 break;
1871
1872 case OP_MARK:
1873 case OP_PRUNE_ARG:
1874 case OP_SKIP_ARG:
1875 code += code[1];
1876 break;
1877
1878 case OP_THEN_ARG:
1879 code += code[1+LINK_SIZE];
1880 break;
1881 }
1882
1883 /* Add in the fixed length from the table */
1884
1885 code += _pcre_OP_lengths[c];
1886
1887 /* In UTF-8 mode, opcodes that are followed by a character may be followed
1888 by a multi-byte character. The length in the table is a minimum, so we have
1889 to arrange to skip the extra bytes. */
1890
1891 #ifdef SUPPORT_UTF8
1892 if (utf8) switch(c)
1893 {
1894 case OP_CHAR:
1895 case OP_CHARI:
1896 case OP_EXACT:
1897 case OP_EXACTI:
1898 case OP_UPTO:
1899 case OP_UPTOI:
1900 case OP_MINUPTO:
1901 case OP_MINUPTOI:
1902 case OP_POSUPTO:
1903 case OP_POSUPTOI:
1904 case OP_STAR:
1905 case OP_STARI:
1906 case OP_MINSTAR:
1907 case OP_MINSTARI:
1908 case OP_POSSTAR:
1909 case OP_POSSTARI:
1910 case OP_PLUS:
1911 case OP_PLUSI:
1912 case OP_MINPLUS:
1913 case OP_MINPLUSI:
1914 case OP_POSPLUS:
1915 case OP_POSPLUSI:
1916 case OP_QUERY:
1917 case OP_QUERYI:
1918 case OP_MINQUERY:
1919 case OP_MINQUERYI:
1920 case OP_POSQUERY:
1921 case OP_POSQUERYI:
1922 if (code[-1] >= 0xc0) code += _pcre_utf8_table4[code[-1] & 0x3f];
1923 break;
1924 }
1925 #else
1926 (void)(utf8); /* Keep compiler happy by referencing function argument */
1927 #endif
1928 }
1929 }
1930 }
1931
1932
1933
1934 /*************************************************
1935 * Scan compiled branch for non-emptiness *
1936 *************************************************/
1937
1938 /* This function scans through a branch of a compiled pattern to see whether it
1939 can match the empty string or not. It is called from could_be_empty()
1940 below and from compile_branch() when checking for an unlimited repeat of a
1941 group that can match nothing. Note that first_significant_code() skips over
1942 backward and negative forward assertions when its final argument is TRUE. If we
1943 hit an unclosed bracket, we return "empty" - this means we've struck an inner
1944 bracket whose current branch will already have been scanned.
1945
1946 Arguments:
1947 code points to start of search
1948 endcode points to where to stop
1949 utf8 TRUE if in UTF8 mode
1950 cd contains pointers to tables etc.
1951
1952 Returns: TRUE if what is matched could be empty
1953 */
1954
1955 static BOOL
1956 could_be_empty_branch(const uschar *code, const uschar *endcode, BOOL utf8,
1957 compile_data *cd)
1958 {
1959 register int c;
1960 for (code = first_significant_code(code + _pcre_OP_lengths[*code], TRUE);
1961 code < endcode;
1962 code = first_significant_code(code + _pcre_OP_lengths[c], TRUE))
1963 {
1964 const uschar *ccode;
1965
1966 c = *code;
1967
1968 /* Skip over forward assertions; the other assertions are skipped by
1969 first_significant_code() with a TRUE final argument. */
1970
1971 if (c == OP_ASSERT)
1972 {
1973 do code += GET(code, 1); while (*code == OP_ALT);
1974 c = *code;
1975 continue;
1976 }
1977
1978 /* For a recursion/subroutine call, if its end has been reached, which
1979 implies a backward reference subroutine call, we can scan it. If it's a
1980 forward reference subroutine call, we can't. To detect forward reference
1981 we have to scan up the list that is kept in the workspace. This function is
1982 called only when doing the real compile, not during the pre-compile that
1983 measures the size of the compiled pattern. */
1984
1985 if (c == OP_RECURSE)
1986 {
1987 const uschar *scode;
1988 BOOL empty_branch;
1989
1990 /* Test for forward reference */
1991
1992 for (scode = cd->start_workspace; scode < cd->hwm; scode += LINK_SIZE)
1993 if (GET(scode, 0) == code + 1 - cd->start_code) return TRUE;
1994
1995 /* Not a forward reference, test for completed backward reference */
1996
1997 empty_branch = FALSE;
1998 scode = cd->start_code + GET(code, 1);
1999 if (GET(scode, 1) == 0) return TRUE; /* Unclosed */
2000
2001 /* Completed backwards reference */
2002
2003 do
2004 {
2005 if (could_be_empty_branch(scode, endcode, utf8, cd))
2006 {
2007 empty_branch = TRUE;
2008 break;
2009 }
2010 scode += GET(scode, 1);
2011 }
2012 while (*scode == OP_ALT);
2013
2014 if (!empty_branch) return FALSE; /* All branches are non-empty */
2015 continue;
2016 }
2017
2018 /* Groups with zero repeats can of course be empty; skip them. */
2019
2020 if (c == OP_BRAZERO || c == OP_BRAMINZERO || c == OP_SKIPZERO ||
2021 c == OP_BRAPOSZERO)
2022 {
2023 code += _pcre_OP_lengths[c];
2024 do code += GET(code, 1); while (*code == OP_ALT);
2025 c = *code;
2026 continue;
2027 }
2028
2029 /* A nested group that is already marked as "could be empty" can just be
2030 skipped. */
2031
2032 if (c == OP_SBRA || c == OP_SBRAPOS ||
2033 c == OP_SCBRA || c == OP_SCBRAPOS)
2034 {
2035 do code += GET(code, 1); while (*code == OP_ALT);
2036 c = *code;
2037 continue;
2038 }
2039
2040 /* For other groups, scan the branches. */
2041
2042 if (c == OP_BRA || c == OP_BRAPOS ||
2043 c == OP_CBRA || c == OP_CBRAPOS ||
2044 c == OP_ONCE || c == OP_COND)
2045 {
2046 BOOL empty_branch;
2047 if (GET(code, 1) == 0) return TRUE; /* Hit unclosed bracket */
2048
2049 /* If a conditional group has only one branch, there is a second, implied,
2050 empty branch, so just skip over the conditional, because it could be empty.
2051 Otherwise, scan the individual branches of the group. */
2052
2053 if (c == OP_COND && code[GET(code, 1)] != OP_ALT)
2054 code += GET(code, 1);
2055 else
2056 {
2057 empty_branch = FALSE;
2058 do
2059 {
2060 if (!empty_branch && could_be_empty_branch(code, endcode, utf8, cd))
2061 empty_branch = TRUE;
2062 code += GET(code, 1);
2063 }
2064 while (*code == OP_ALT);
2065 if (!empty_branch) return FALSE; /* All branches are non-empty */
2066 }
2067
2068 c = *code;
2069 continue;
2070 }
2071
2072 /* Handle the other opcodes */
2073
2074 switch (c)
2075 {
2076 /* Check for quantifiers after a class. XCLASS is used for classes that
2077 cannot be represented just by a bit map. This includes negated single
2078 high-valued characters. The length in _pcre_OP_lengths[] is zero; the
2079 actual length is stored in the compiled code, so we must update "code"
2080 here. */
2081
2082 #ifdef SUPPORT_UTF8
2083 case OP_XCLASS:
2084 ccode = code += GET(code, 1);
2085 goto CHECK_CLASS_REPEAT;
2086 #endif
2087
2088 case OP_CLASS:
2089 case OP_NCLASS:
2090 ccode = code + 33;
2091
2092 #ifdef SUPPORT_UTF8
2093 CHECK_CLASS_REPEAT:
2094 #endif
2095
2096 switch (*ccode)
2097 {
2098 case OP_CRSTAR: /* These could be empty; continue */
2099 case OP_CRMINSTAR:
2100 case OP_CRQUERY:
2101 case OP_CRMINQUERY:
2102 break;
2103
2104 default: /* Non-repeat => class must match */
2105 case OP_CRPLUS: /* These repeats aren't empty */
2106 case OP_CRMINPLUS:
2107 return FALSE;
2108
2109 case OP_CRRANGE:
2110 case OP_CRMINRANGE:
2111 if (GET2(ccode, 1) > 0) return FALSE; /* Minimum > 0 */
2112 break;
2113 }
2114 break;
2115
2116 /* Opcodes that must match a character */
2117
2118 case OP_PROP:
2119 case OP_NOTPROP:
2120 case OP_EXTUNI:
2121 case OP_NOT_DIGIT:
2122 case OP_DIGIT:
2123 case OP_NOT_WHITESPACE:
2124 case OP_WHITESPACE:
2125 case OP_NOT_WORDCHAR:
2126 case OP_WORDCHAR:
2127 case OP_ANY:
2128 case OP_ALLANY:
2129 case OP_ANYBYTE:
2130 case OP_CHAR:
2131 case OP_CHARI:
2132 case OP_NOT:
2133 case OP_NOTI:
2134 case OP_PLUS:
2135 case OP_MINPLUS:
2136 case OP_POSPLUS:
2137 case OP_EXACT:
2138 case OP_NOTPLUS:
2139 case OP_NOTMINPLUS:
2140 case OP_NOTPOSPLUS:
2141 case OP_NOTEXACT:
2142 case OP_TYPEPLUS:
2143 case OP_TYPEMINPLUS:
2144 case OP_TYPEPOSPLUS:
2145 case OP_TYPEEXACT:
2146 return FALSE;
2147
2148 /* These are going to continue, as they may be empty, but we have to
2149 fudge the length for the \p and \P cases. */
2150
2151 case OP_TYPESTAR:
2152 case OP_TYPEMINSTAR:
2153 case OP_TYPEPOSSTAR:
2154 case OP_TYPEQUERY:
2155 case OP_TYPEMINQUERY:
2156 case OP_TYPEPOSQUERY:
2157 if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
2158 break;
2159
2160 /* Same for these */
2161
2162 case OP_TYPEUPTO:
2163 case OP_TYPEMINUPTO:
2164 case OP_TYPEPOSUPTO:
2165 if (code[3] == OP_PROP || code[3] == OP_NOTPROP) code += 2;
2166 break;
2167
2168 /* End of branch */
2169
2170 case OP_KET:
2171 case OP_KETRMAX:
2172 case OP_KETRMIN:
2173 case OP_KETRPOS:
2174 case OP_ALT:
2175 return TRUE;
2176
2177 /* In UTF-8 mode, STAR, MINSTAR, POSSTAR, QUERY, MINQUERY, POSQUERY, UPTO,
2178 MINUPTO, and POSUPTO may be followed by a multibyte character */
2179
2180 #ifdef SUPPORT_UTF8
2181 case OP_STAR:
2182 case OP_STARI:
2183 case OP_MINSTAR:
2184 case OP_MINSTARI:
2185 case OP_POSSTAR:
2186 case OP_POSSTARI:
2187 case OP_QUERY:
2188 case OP_QUERYI:
2189 case OP_MINQUERY:
2190 case OP_MINQUERYI:
2191 case OP_POSQUERY:
2192 case OP_POSQUERYI:
2193 if (utf8 && code[1] >= 0xc0) code += _pcre_utf8_table4[code[1] & 0x3f];
2194 break;
2195
2196 case OP_UPTO:
2197 case OP_UPTOI:
2198 case OP_MINUPTO:
2199 case OP_MINUPTOI:
2200 case OP_POSUPTO:
2201 case OP_POSUPTOI:
2202 if (utf8 && code[3] >= 0xc0) code += _pcre_utf8_table4[code[3] & 0x3f];
2203 break;
2204 #endif
2205
2206 /* MARK, and PRUNE/SKIP/THEN with an argument must skip over the argument
2207 string. */
2208
2209 case OP_MARK:
2210 case OP_PRUNE_ARG:
2211 case OP_SKIP_ARG:
2212 code += code[1];
2213 break;
2214
2215 case OP_THEN_ARG:
2216 code += code[1+LINK_SIZE];
2217 break;
2218
2219 /* None of the remaining opcodes are required to match a character. */
2220
2221 default:
2222 break;
2223 }
2224 }
2225
2226 return TRUE;
2227 }
2228
2229
2230
2231 /*************************************************
2232 * Scan compiled regex for non-emptiness *
2233 *************************************************/
2234
2235 /* This function is called to check for left recursive calls. We want to check
2236 the current branch of the current pattern to see if it could match the empty
2237 string. If it could, we must look outwards for branches at other levels,
2238 stopping when we pass beyond the bracket which is the subject of the recursion.
2239 This function is called only during the real compile, not during the
2240 pre-compile.
2241
2242 Arguments:
2243 code points to start of the recursion
2244 endcode points to where to stop (current RECURSE item)
2245 bcptr points to the chain of current (unclosed) branch starts
2246 utf8 TRUE if in UTF-8 mode
2247 cd pointers to tables etc
2248
2249 Returns: TRUE if what is matched could be empty
2250 */
2251
2252 static BOOL
2253 could_be_empty(const uschar *code, const uschar *endcode, branch_chain *bcptr,
2254 BOOL utf8, compile_data *cd)
2255 {
2256 while (bcptr != NULL && bcptr->current_branch >= code)
2257 {
2258 if (!could_be_empty_branch(bcptr->current_branch, endcode, utf8, cd))
2259 return FALSE;
2260 bcptr = bcptr->outer;
2261 }
2262 return TRUE;
2263 }
2264
2265
2266
2267 /*************************************************
2268 * Check for POSIX class syntax *
2269 *************************************************/
2270
2271 /* This function is called when the sequence "[:" or "[." or "[=" is
2272 encountered in a character class. It checks whether this is followed by a
2273 sequence of characters terminated by a matching ":]" or ".]" or "=]". If we
2274 reach an unescaped ']' without the special preceding character, return FALSE.
2275
2276 Originally, this function only recognized a sequence of letters between the
2277 terminators, but it seems that Perl recognizes any sequence of characters,
2278 though of course unknown POSIX names are subsequently rejected. Perl gives an
2279 "Unknown POSIX class" error for [:f\oo:] for example, where previously PCRE
2280 didn't consider this to be a POSIX class. Likewise for [:1234:].
2281
2282 The problem in trying to be exactly like Perl is in the handling of escapes. We
2283 have to be sure that [abc[:x\]pqr] is *not* treated as containing a POSIX
2284 class, but [abc[:x\]pqr:]] is (so that an error can be generated). The code
2285 below handles the special case of \], but does not try to do any other escape
2286 processing. This makes it different from Perl for cases such as [:l\ower:]
2287 where Perl recognizes it as the POSIX class "lower" but PCRE does not recognize
2288 "l\ower". This is a lesser evil that not diagnosing bad classes when Perl does,
2289 I think.
2290
2291 Arguments:
2292 ptr pointer to the initial [
2293 endptr where to return the end pointer
2294
2295 Returns: TRUE or FALSE
2296 */
2297
2298 static BOOL
2299 check_posix_syntax(const uschar *ptr, const uschar **endptr)
2300 {
2301 int terminator; /* Don't combine these lines; the Solaris cc */
2302 terminator = *(++ptr); /* compiler warns about "non-constant" initializer. */
2303 for (++ptr; *ptr != 0; ptr++)
2304 {
2305 if (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET) ptr++; else
2306 {
2307 if (*ptr == CHAR_RIGHT_SQUARE_BRACKET) return FALSE;
2308 if (*ptr == terminator && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET)
2309 {
2310 *endptr = ptr;
2311 return TRUE;
2312 }
2313 }
2314 }
2315 return FALSE;
2316 }
2317
2318
2319
2320
2321 /*************************************************
2322 * Check POSIX class name *
2323 *************************************************/
2324
2325 /* This function is called to check the name given in a POSIX-style class entry
2326 such as [:alnum:].
2327
2328 Arguments:
2329 ptr points to the first letter
2330 len the length of the name
2331
2332 Returns: a value representing the name, or -1 if unknown
2333 */
2334
2335 static int
2336 check_posix_name(const uschar *ptr, int len)
2337 {
2338 const char *pn = posix_names;
2339 register int yield = 0;
2340 while (posix_name_lengths[yield] != 0)
2341 {
2342 if (len == posix_name_lengths[yield] &&
2343 strncmp((const char *)ptr, pn, len) == 0) return yield;
2344 pn += posix_name_lengths[yield] + 1;
2345 yield++;
2346 }
2347 return -1;
2348 }
2349
2350
2351 /*************************************************
2352 * Adjust OP_RECURSE items in repeated group *
2353 *************************************************/
2354
2355 /* OP_RECURSE items contain an offset from the start of the regex to the group
2356 that is referenced. This means that groups can be replicated for fixed
2357 repetition simply by copying (because the recursion is allowed to refer to
2358 earlier groups that are outside the current group). However, when a group is
2359 optional (i.e. the minimum quantifier is zero), OP_BRAZERO or OP_SKIPZERO is
2360 inserted before it, after it has been compiled. This means that any OP_RECURSE
2361 items within it that refer to the group itself or any contained groups have to
2362 have their offsets adjusted. That one of the jobs of this function. Before it
2363 is called, the partially compiled regex must be temporarily terminated with
2364 OP_END.
2365
2366 This function has been extended with the possibility of forward references for
2367 recursions and subroutine calls. It must also check the list of such references
2368 for the group we are dealing with. If it finds that one of the recursions in
2369 the current group is on this list, it adjusts the offset in the list, not the
2370 value in the reference (which is a group number).
2371
2372 Arguments:
2373 group points to the start of the group
2374 adjust the amount by which the group is to be moved
2375 utf8 TRUE in UTF-8 mode
2376 cd contains pointers to tables etc.
2377 save_hwm the hwm forward reference pointer at the start of the group
2378
2379 Returns: nothing
2380 */
2381
2382 static void
2383 adjust_recurse(uschar *group, int adjust, BOOL utf8, compile_data *cd,
2384 uschar *save_hwm)
2385 {
2386 uschar *ptr = group;
2387
2388 while ((ptr = (uschar *)find_recurse(ptr, utf8)) != NULL)
2389 {
2390 int offset;
2391 uschar *hc;
2392
2393 /* See if this recursion is on the forward reference list. If so, adjust the
2394 reference. */
2395
2396 for (hc = save_hwm; hc < cd->hwm; hc += LINK_SIZE)
2397 {
2398 offset = GET(hc, 0);
2399 if (cd->start_code + offset == ptr + 1)
2400 {
2401 PUT(hc, 0, offset + adjust);
2402 break;
2403 }
2404 }
2405
2406 /* Otherwise, adjust the recursion offset if it's after the start of this
2407 group. */
2408
2409 if (hc >= cd->hwm)
2410 {
2411 offset = GET(ptr, 1);
2412 if (cd->start_code + offset >= group) PUT(ptr, 1, offset + adjust);
2413 }
2414
2415 ptr += 1 + LINK_SIZE;
2416 }
2417 }
2418
2419
2420
2421 /*************************************************
2422 * Insert an automatic callout point *
2423 *************************************************/
2424
2425 /* This function is called when the PCRE_AUTO_CALLOUT option is set, to insert
2426 callout points before each pattern item.
2427
2428 Arguments:
2429 code current code pointer
2430 ptr current pattern pointer
2431 cd pointers to tables etc
2432
2433 Returns: new code pointer
2434 */
2435
2436 static uschar *
2437 auto_callout(uschar *code, const uschar *ptr, compile_data *cd)
2438 {
2439 *code++ = OP_CALLOUT;
2440 *code++ = 255;
2441 PUT(code, 0, (int)(ptr - cd->start_pattern)); /* Pattern offset */
2442 PUT(code, LINK_SIZE, 0); /* Default length */
2443 return code + 2*LINK_SIZE;
2444 }
2445
2446
2447
2448 /*************************************************
2449 * Complete a callout item *
2450 *************************************************/
2451
2452 /* A callout item contains the length of the next item in the pattern, which
2453 we can't fill in till after we have reached the relevant point. This is used
2454 for both automatic and manual callouts.
2455
2456 Arguments:
2457 previous_callout points to previous callout item
2458 ptr current pattern pointer
2459 cd pointers to tables etc
2460
2461 Returns: nothing
2462 */
2463
2464 static void
2465 complete_callout(uschar *previous_callout, const uschar *ptr, compile_data *cd)
2466 {
2467 int length = (int)(ptr - cd->start_pattern - GET(previous_callout, 2));
2468 PUT(previous_callout, 2 + LINK_SIZE, length);
2469 }
2470
2471
2472
2473 #ifdef SUPPORT_UCP
2474 /*************************************************
2475 * Get othercase range *
2476 *************************************************/
2477
2478 /* This function is passed the start and end of a class range, in UTF-8 mode
2479 with UCP support. It searches up the characters, looking for internal ranges of
2480 characters in the "other" case. Each call returns the next one, updating the
2481 start address.
2482
2483 Arguments:
2484 cptr points to starting character value; updated
2485 d end value
2486 ocptr where to put start of othercase range
2487 odptr where to put end of othercase range
2488
2489 Yield: TRUE when range returned; FALSE when no more
2490 */
2491
2492 static BOOL
2493 get_othercase_range(unsigned int *cptr, unsigned int d, unsigned int *ocptr,
2494 unsigned int *odptr)
2495 {
2496 unsigned int c, othercase, next;
2497
2498 for (c = *cptr; c <= d; c++)
2499 { if ((othercase = UCD_OTHERCASE(c)) != c) break; }
2500
2501 if (c > d) return FALSE;
2502
2503 *ocptr = othercase;
2504 next = othercase + 1;
2505
2506 for (++c; c <= d; c++)
2507 {
2508 if (UCD_OTHERCASE(c) != next) break;
2509 next++;
2510 }
2511
2512 *odptr = next - 1;
2513 *cptr = c;
2514
2515 return TRUE;
2516 }
2517
2518
2519
2520 /*************************************************
2521 * Check a character and a property *
2522 *************************************************/
2523
2524 /* This function is called by check_auto_possessive() when a property item
2525 is adjacent to a fixed character.
2526
2527 Arguments:
2528 c the character
2529 ptype the property type
2530 pdata the data for the type
2531 negated TRUE if it's a negated property (\P or \p{^)
2532
2533 Returns: TRUE if auto-possessifying is OK
2534 */
2535
2536 static BOOL
2537 check_char_prop(int c, int ptype, int pdata, BOOL negated)
2538 {
2539 const ucd_record *prop = GET_UCD(c);
2540 switch(ptype)
2541 {
2542 case PT_LAMP:
2543 return (prop->chartype == ucp_Lu ||
2544 prop->chartype == ucp_Ll ||
2545 prop->chartype == ucp_Lt) == negated;
2546
2547 case PT_GC:
2548 return (pdata == _pcre_ucp_gentype[prop->chartype]) == negated;
2549
2550 case PT_PC:
2551 return (pdata == prop->chartype) == negated;
2552
2553 case PT_SC:
2554 return (pdata == prop->script) == negated;
2555
2556 /* These are specials */
2557
2558 case PT_ALNUM:
2559 return (_pcre_ucp_gentype[prop->chartype] == ucp_L ||
2560 _pcre_ucp_gentype[prop->chartype] == ucp_N) == negated;
2561
2562 case PT_SPACE: /* Perl space */
2563 return (_pcre_ucp_gentype[prop->chartype] == ucp_Z ||
2564 c == CHAR_HT || c == CHAR_NL || c == CHAR_FF || c == CHAR_CR)
2565 == negated;
2566
2567 case PT_PXSPACE: /* POSIX space */
2568 return (_pcre_ucp_gentype[prop->chartype] == ucp_Z ||
2569 c == CHAR_HT || c == CHAR_NL || c == CHAR_VT ||
2570 c == CHAR_FF || c == CHAR_CR)
2571 == negated;
2572
2573 case PT_WORD:
2574 return (_pcre_ucp_gentype[prop->chartype] == ucp_L ||
2575 _pcre_ucp_gentype[prop->chartype] == ucp_N ||
2576 c == CHAR_UNDERSCORE) == negated;
2577 }
2578 return FALSE;
2579 }
2580 #endif /* SUPPORT_UCP */
2581
2582
2583
2584 /*************************************************
2585 * Check if auto-possessifying is possible *
2586 *************************************************/
2587
2588 /* This function is called for unlimited repeats of certain items, to see
2589 whether the next thing could possibly match the repeated item. If not, it makes
2590 sense to automatically possessify the repeated item.
2591
2592 Arguments:
2593 previous pointer to the repeated opcode
2594 utf8 TRUE in UTF-8 mode
2595 ptr next character in pattern
2596 options options bits
2597 cd contains pointers to tables etc.
2598
2599 Returns: TRUE if possessifying is wanted
2600 */
2601
2602 static BOOL
2603 check_auto_possessive(const uschar *previous, BOOL utf8, const uschar *ptr,
2604 int options, compile_data *cd)
2605 {
2606 int c, next;
2607 int op_code = *previous++;
2608
2609 /* Skip whitespace and comments in extended mode */
2610
2611 if ((options & PCRE_EXTENDED) != 0)
2612 {
2613 for (;;)
2614 {
2615 while ((cd->ctypes[*ptr] & ctype_space) != 0) ptr++;
2616 if (*ptr == CHAR_NUMBER_SIGN)
2617 {
2618 ptr++;
2619 while (*ptr != 0)
2620 {
2621 if (IS_NEWLINE(ptr)) { ptr += cd->nllen; break; }
2622 ptr++;
2623 #ifdef SUPPORT_UTF8
2624 if (utf8) while ((*ptr & 0xc0) == 0x80) ptr++;
2625 #endif
2626 }
2627 }
2628 else break;
2629 }
2630 }
2631
2632 /* If the next item is one that we can handle, get its value. A non-negative
2633 value is a character, a negative value is an escape value. */
2634
2635 if (*ptr == CHAR_BACKSLASH)
2636 {
2637 int temperrorcode = 0;
2638 next = check_escape(&ptr, &temperrorcode, cd->bracount, options, FALSE);
2639 if (temperrorcode != 0) return FALSE;
2640 ptr++; /* Point after the escape sequence */
2641 }
2642
2643 else if ((cd->ctypes[*ptr] & ctype_meta) == 0)
2644 {
2645 #ifdef SUPPORT_UTF8
2646 if (utf8) { GETCHARINC(next, ptr); } else
2647 #endif
2648 next = *ptr++;
2649 }
2650
2651 else return FALSE;
2652
2653 /* Skip whitespace and comments in extended mode */
2654
2655 if ((options & PCRE_EXTENDED) != 0)
2656 {
2657 for (;;)
2658 {
2659 while ((cd->ctypes[*ptr] & ctype_space) != 0) ptr++;
2660 if (*ptr == CHAR_NUMBER_SIGN)
2661 {
2662 ptr++;
2663 while (*ptr != 0)
2664 {
2665 if (IS_NEWLINE(ptr)) { ptr += cd->nllen; break; }
2666 ptr++;
2667 #ifdef SUPPORT_UTF8
2668 if (utf8) while ((*ptr & 0xc0) == 0x80) ptr++;
2669 #endif
2670 }
2671 }
2672 else break;
2673 }
2674 }
2675
2676 /* If the next thing is itself optional, we have to give up. */
2677
2678 if (*ptr == CHAR_ASTERISK || *ptr == CHAR_QUESTION_MARK ||
2679 strncmp((char *)ptr, STR_LEFT_CURLY_BRACKET STR_0 STR_COMMA, 3) == 0)
2680 return FALSE;
2681
2682 /* Now compare the next item with the previous opcode. First, handle cases when
2683 the next item is a character. */
2684
2685 if (next >= 0) switch(op_code)
2686 {
2687 case OP_CHAR:
2688 #ifdef SUPPORT_UTF8
2689 GETCHARTEST(c, previous);
2690 #else
2691 c = *previous;
2692 #endif
2693 return c != next;
2694
2695 /* For CHARI (caseless character) we must check the other case. If we have
2696 Unicode property support, we can use it to test the other case of
2697 high-valued characters. */
2698
2699 case OP_CHARI:
2700 #ifdef SUPPORT_UTF8
2701 GETCHARTEST(c, previous);
2702 #else
2703 c = *previous;
2704 #endif
2705 if (c == next) return FALSE;
2706 #ifdef SUPPORT_UTF8
2707 if (utf8)
2708 {
2709 unsigned int othercase;
2710 if (next < 128) othercase = cd->fcc[next]; else
2711 #ifdef SUPPORT_UCP
2712 othercase = UCD_OTHERCASE((unsigned int)next);
2713 #else
2714 othercase = NOTACHAR;
2715 #endif
2716 return (unsigned int)c != othercase;
2717 }
2718 else
2719 #endif /* SUPPORT_UTF8 */
2720 return (c != cd->fcc[next]); /* Non-UTF-8 mode */
2721
2722 /* For OP_NOT and OP_NOTI, the data is always a single-byte character. These
2723 opcodes are not used for multi-byte characters, because they are coded using
2724 an XCLASS instead. */
2725
2726 case OP_NOT:
2727 return (c = *previous) == next;
2728
2729 case OP_NOTI:
2730 if ((c = *previous) == next) return TRUE;
2731 #ifdef SUPPORT_UTF8
2732 if (utf8)
2733 {
2734 unsigned int othercase;
2735 if (next < 128) othercase = cd->fcc[next]; else
2736 #ifdef SUPPORT_UCP
2737 othercase = UCD_OTHERCASE(next);
2738 #else
2739 othercase = NOTACHAR;
2740 #endif
2741 return (unsigned int)c == othercase;
2742 }
2743 else
2744 #endif /* SUPPORT_UTF8 */
2745 return (c == cd->fcc[next]); /* Non-UTF-8 mode */
2746
2747 /* Note that OP_DIGIT etc. are generated only when PCRE_UCP is *not* set.
2748 When it is set, \d etc. are converted into OP_(NOT_)PROP codes. */
2749
2750 case OP_DIGIT:
2751 return next > 127 || (cd->ctypes[next] & ctype_digit) == 0;
2752
2753 case OP_NOT_DIGIT:
2754 return next <= 127 && (cd->ctypes[next] & ctype_digit) != 0;
2755
2756 case OP_WHITESPACE:
2757 return next > 127 || (cd->ctypes[next] & ctype_space) == 0;
2758
2759 case OP_NOT_WHITESPACE:
2760 return next <= 127 && (cd->ctypes[next] & ctype_space) != 0;
2761
2762 case OP_WORDCHAR:
2763 return next > 127 || (cd->ctypes[next] & ctype_word) == 0;
2764
2765 case OP_NOT_WORDCHAR:
2766 return next <= 127 && (cd->ctypes[next] & ctype_word) != 0;
2767
2768 case OP_HSPACE:
2769 case OP_NOT_HSPACE:
2770 switch(next)
2771 {
2772 case 0x09:
2773 case 0x20:
2774 case 0xa0:
2775 case 0x1680:
2776 case 0x180e:
2777 case 0x2000:
2778 case 0x2001:
2779 case 0x2002:
2780 case 0x2003:
2781 case 0x2004:
2782 case 0x2005:
2783 case 0x2006:
2784 case 0x2007:
2785 case 0x2008:
2786 case 0x2009:
2787 case 0x200A:
2788 case 0x202f:
2789 case 0x205f:
2790 case 0x3000:
2791 return op_code == OP_NOT_HSPACE;
2792 default:
2793 return op_code != OP_NOT_HSPACE;
2794 }
2795
2796 case OP_ANYNL:
2797 case OP_VSPACE:
2798 case OP_NOT_VSPACE:
2799 switch(next)
2800 {
2801 case 0x0a:
2802 case 0x0b:
2803 case 0x0c:
2804 case 0x0d:
2805 case 0x85:
2806 case 0x2028:
2807 case 0x2029:
2808 return op_code == OP_NOT_VSPACE;
2809 default:
2810 return op_code != OP_NOT_VSPACE;
2811 }
2812
2813 #ifdef SUPPORT_UCP
2814 case OP_PROP:
2815 return check_char_prop(next, previous[0], previous[1], FALSE);
2816
2817 case OP_NOTPROP:
2818 return check_char_prop(next, previous[0], previous[1], TRUE);
2819 #endif
2820
2821 default:
2822 return FALSE;
2823 }
2824
2825
2826 /* Handle the case when the next item is \d, \s, etc. Note that when PCRE_UCP
2827 is set, \d turns into ESC_du rather than ESC_d, etc., so ESC_d etc. are
2828 generated only when PCRE_UCP is *not* set, that is, when only ASCII
2829 characteristics are recognized. Similarly, the opcodes OP_DIGIT etc. are
2830 replaced by OP_PROP codes when PCRE_UCP is set. */
2831
2832 switch(op_code)
2833 {
2834 case OP_CHAR:
2835 case OP_CHARI:
2836 #ifdef SUPPORT_UTF8
2837 GETCHARTEST(c, previous);
2838 #else
2839 c = *previous;
2840 #endif
2841 switch(-next)
2842 {
2843 case ESC_d:
2844 return c > 127 || (cd->ctypes[c] & ctype_digit) == 0;
2845
2846 case ESC_D:
2847 return c <= 127 && (cd->ctypes[c] & ctype_digit) != 0;
2848
2849 case ESC_s:
2850 return c > 127 || (cd->ctypes[c] & ctype_space) == 0;
2851
2852 case ESC_S:
2853 return c <= 127 && (cd->ctypes[c] & ctype_space) != 0;
2854
2855 case ESC_w:
2856 return c > 127 || (cd->ctypes[c] & ctype_word) == 0;
2857
2858 case ESC_W:
2859 return c <= 127 && (cd->ctypes[c] & ctype_word) != 0;
2860
2861 case ESC_h:
2862 case ESC_H:
2863 switch(c)
2864 {
2865 case 0x09:
2866 case 0x20:
2867 case 0xa0:
2868 case 0x1680:
2869 case 0x180e:
2870 case 0x2000:
2871 case 0x2001:
2872 case 0x2002:
2873 case 0x2003:
2874 case 0x2004:
2875 case 0x2005:
2876 case 0x2006:
2877 case 0x2007:
2878 case 0x2008:
2879 case 0x2009:
2880 case 0x200A:
2881 case 0x202f:
2882 case 0x205f:
2883 case 0x3000:
2884 return -next != ESC_h;
2885 default:
2886 return -next == ESC_h;
2887 }
2888
2889 case ESC_v:
2890 case ESC_V:
2891 switch(c)
2892 {
2893 case 0x0a:
2894 case 0x0b:
2895 case 0x0c:
2896 case 0x0d:
2897 case 0x85:
2898 case 0x2028:
2899 case 0x2029:
2900 return -next != ESC_v;
2901 default:
2902 return -next == ESC_v;
2903 }
2904
2905 /* When PCRE_UCP is set, these values get generated for \d etc. Find
2906 their substitutions and process them. The result will always be either
2907 -ESC_p or -ESC_P. Then fall through to process those values. */
2908
2909 #ifdef SUPPORT_UCP
2910 case ESC_du:
2911 case ESC_DU:
2912 case ESC_wu:
2913 case ESC_WU:
2914 case ESC_su:
2915 case ESC_SU:
2916 {
2917 int temperrorcode = 0;
2918 ptr = substitutes[-next - ESC_DU];
2919 next = check_escape(&ptr, &temperrorcode, 0, options, FALSE);
2920 if (temperrorcode != 0) return FALSE;
2921 ptr++; /* For compatibility */
2922 }
2923 /* Fall through */
2924
2925 case ESC_p:
2926 case ESC_P:
2927 {
2928 int ptype, pdata, errorcodeptr;
2929 BOOL negated;
2930
2931 ptr--; /* Make ptr point at the p or P */
2932 ptype = get_ucp(&ptr, &negated, &pdata, &errorcodeptr);
2933 if (ptype < 0) return FALSE;
2934 ptr++; /* Point past the final curly ket */
2935
2936 /* If the property item is optional, we have to give up. (When generated
2937 from \d etc by PCRE_UCP, this test will have been applied much earlier,
2938 to the original \d etc. At this point, ptr will point to a zero byte. */
2939
2940 if (*ptr == CHAR_ASTERISK || *ptr == CHAR_QUESTION_MARK ||
2941 strncmp((char *)ptr, STR_LEFT_CURLY_BRACKET STR_0 STR_COMMA, 3) == 0)
2942 return FALSE;
2943
2944 /* Do the property check. */
2945
2946 return check_char_prop(c, ptype, pdata, (next == -ESC_P) != negated);
2947 }
2948 #endif
2949
2950 default:
2951 return FALSE;
2952 }
2953
2954 /* In principle, support for Unicode properties should be integrated here as
2955 well. It means re-organizing the above code so as to get hold of the property
2956 values before switching on the op-code. However, I wonder how many patterns
2957 combine ASCII \d etc with Unicode properties? (Note that if PCRE_UCP is set,
2958 these op-codes are never generated.) */
2959
2960 case OP_DIGIT:
2961 return next == -ESC_D || next == -ESC_s || next == -ESC_W ||
2962 next == -ESC_h || next == -ESC_v || next == -ESC_R;
2963
2964 case OP_NOT_DIGIT:
2965 return next == -ESC_d;
2966
2967 case OP_WHITESPACE:
2968 return next == -ESC_S || next == -ESC_d || next == -ESC_w || next == -ESC_R;
2969
2970 case OP_NOT_WHITESPACE:
2971 return next == -ESC_s || next == -ESC_h || next == -ESC_v;
2972
2973 case OP_HSPACE:
2974 return next == -ESC_S || next == -ESC_H || next == -ESC_d ||
2975 next == -ESC_w || next == -ESC_v || next == -ESC_R;
2976
2977 case OP_NOT_HSPACE:
2978 return next == -ESC_h;
2979
2980 /* Can't have \S in here because VT matches \S (Perl anomaly) */
2981 case OP_ANYNL:
2982 case OP_VSPACE:
2983 return next == -ESC_V || next == -ESC_d || next == -ESC_w;
2984
2985 case OP_NOT_VSPACE:
2986 return next == -ESC_v || next == -ESC_R;
2987
2988 case OP_WORDCHAR:
2989 return next == -ESC_W || next == -ESC_s || next == -ESC_h ||
2990 next == -ESC_v || next == -ESC_R;
2991
2992 case OP_NOT_WORDCHAR:
2993 return next == -ESC_w || next == -ESC_d;
2994
2995 default:
2996 return FALSE;
2997 }
2998
2999 /* Control does not reach here */
3000 }
3001
3002
3003
3004 /*************************************************
3005 * Compile one branch *
3006 *************************************************/
3007
3008 /* Scan the pattern, compiling it into the a vector. If the options are
3009 changed during the branch, the pointer is used to change the external options
3010 bits. This function is used during the pre-compile phase when we are trying
3011 to find out the amount of memory needed, as well as during the real compile
3012 phase. The value of lengthptr distinguishes the two phases.
3013
3014 Arguments:
3015 optionsptr pointer to the option bits
3016 codeptr points to the pointer to the current code point
3017 ptrptr points to the current pattern pointer
3018 errorcodeptr points to error code variable
3019 firstbyteptr set to initial literal character, or < 0 (REQ_UNSET, REQ_NONE)
3020 reqbyteptr set to the last literal character required, else < 0
3021 bcptr points to current branch chain
3022 cd contains pointers to tables etc.
3023 lengthptr NULL during the real compile phase
3024 points to length accumulator during pre-compile phase
3025
3026 Returns: TRUE on success
3027 FALSE, with *errorcodeptr set non-zero on error
3028 */
3029
3030 static BOOL
3031 compile_branch(int *optionsptr, uschar **codeptr, const uschar **ptrptr,
3032 int *errorcodeptr, int *firstbyteptr, int *reqbyteptr, branch_chain *bcptr,
3033 compile_data *cd, int *lengthptr)
3034 {
3035 int repeat_type, op_type;
3036 int repeat_min = 0, repeat_max = 0; /* To please picky compilers */
3037 int bravalue = 0;
3038 int greedy_default, greedy_non_default;
3039 int firstbyte, reqbyte;
3040 int zeroreqbyte, zerofirstbyte;
3041 int req_caseopt, reqvary, tempreqvary;
3042 int options = *optionsptr; /* May change dynamically */
3043 int after_manual_callout = 0;
3044 int length_prevgroup = 0;
3045 register int c;
3046 register uschar *code = *codeptr;
3047 uschar *last_code = code;
3048 uschar *orig_code = code;
3049 uschar *tempcode;
3050 BOOL inescq = FALSE;
3051 BOOL groupsetfirstbyte = FALSE;
3052 const uschar *ptr = *ptrptr;
3053 const uschar *tempptr;
3054 const uschar *nestptr = NULL;
3055 uschar *previous = NULL;
3056 uschar *previous_callout = NULL;
3057 uschar *save_hwm = NULL;
3058 uschar classbits[32];
3059
3060 /* We can fish out the UTF-8 setting once and for all into a BOOL, but we
3061 must not do this for other options (e.g. PCRE_EXTENDED) because they may change
3062 dynamically as we process the pattern. */
3063
3064 #ifdef SUPPORT_UTF8
3065 BOOL class_utf8;
3066 BOOL utf8 = (options & PCRE_UTF8) != 0;
3067 uschar *class_utf8data;
3068 uschar *class_utf8data_base;
3069 uschar utf8_char[6];
3070 #else
3071 BOOL utf8 = FALSE;
3072 uschar *utf8_char = NULL;
3073 #endif
3074
3075 #ifdef PCRE_DEBUG
3076 if (lengthptr != NULL) DPRINTF((">> start branch\n"));
3077 #endif
3078
3079 /* Set up the default and non-default settings for greediness */
3080
3081 greedy_default = ((options & PCRE_UNGREEDY) != 0);
3082 greedy_non_default = greedy_default ^ 1;
3083
3084 /* Initialize no first byte, no required byte. REQ_UNSET means "no char
3085 matching encountered yet". It gets changed to REQ_NONE if we hit something that
3086 matches a non-fixed char first char; reqbyte just remains unset if we never
3087 find one.
3088
3089 When we hit a repeat whose minimum is zero, we may have to adjust these values
3090 to take the zero repeat into account. This is implemented by setting them to
3091 zerofirstbyte and zeroreqbyte when such a repeat is encountered. The individual
3092 item types that can be repeated set these backoff variables appropriately. */
3093
3094 firstbyte = reqbyte = zerofirstbyte = zeroreqbyte = REQ_UNSET;
3095
3096 /* The variable req_caseopt contains either the REQ_CASELESS value or zero,
3097 according to the current setting of the caseless flag. REQ_CASELESS is a bit
3098 value > 255. It is added into the firstbyte or reqbyte variables to record the
3099 case status of the value. This is used only for ASCII characters. */
3100
3101 req_caseopt = ((options & PCRE_CASELESS) != 0)? REQ_CASELESS : 0;
3102
3103 /* Switch on next character until the end of the branch */
3104
3105 for (;; ptr++)
3106 {
3107 BOOL negate_class;
3108 BOOL should_flip_negation;
3109 BOOL possessive_quantifier;
3110 BOOL is_quantifier;
3111 BOOL is_recurse;
3112 BOOL reset_bracount;
3113 int class_charcount;
3114 int class_lastchar;
3115 int newoptions;
3116 int recno;
3117 int refsign;
3118 int skipbytes;
3119 int subreqbyte;
3120 int subfirstbyte;
3121 int terminator;
3122 int mclength;
3123 uschar mcbuffer[8];
3124
3125 /* Get next byte in the pattern */
3126
3127 c = *ptr;
3128
3129 /* If we are at the end of a nested substitution, revert to the outer level
3130 string. Nesting only happens one level deep. */
3131
3132 if (c == 0 && nestptr != NULL)
3133 {
3134 ptr = nestptr;
3135 nestptr = NULL;
3136 c = *ptr;
3137 }
3138
3139 /* If we are in the pre-compile phase, accumulate the length used for the
3140 previous cycle of this loop. */
3141
3142 if (lengthptr != NULL)
3143 {
3144 #ifdef PCRE_DEBUG
3145 if (code > cd->hwm) cd->hwm = code; /* High water info */
3146 #endif
3147 if (code > cd->start_workspace + WORK_SIZE_CHECK) /* Check for overrun */
3148 {
3149 *errorcodeptr = ERR52;
3150 goto FAILED;
3151 }
3152
3153 /* There is at least one situation where code goes backwards: this is the
3154 case of a zero quantifier after a class (e.g. [ab]{0}). At compile time,
3155 the class is simply eliminated. However, it is created first, so we have to
3156 allow memory for it. Therefore, don't ever reduce the length at this point.
3157 */
3158
3159 if (code < last_code) code = last_code;
3160
3161 /* Paranoid check for integer overflow */
3162
3163 if (OFLOW_MAX - *lengthptr < code - last_code)
3164 {
3165 *errorcodeptr = ERR20;
3166 goto FAILED;
3167 }
3168
3169 *lengthptr += (int)(code - last_code);
3170 DPRINTF(("length=%d added %d c=%c\n", *lengthptr, code - last_code, c));
3171
3172 /* If "previous" is set and it is not at the start of the work space, move
3173 it back to there, in order to avoid filling up the work space. Otherwise,
3174 if "previous" is NULL, reset the current code pointer to the start. */
3175
3176 if (previous != NULL)
3177 {
3178 if (previous > orig_code)
3179 {
3180 memmove(orig_code, previous, code - previous);
3181 code -= previous - orig_code;
3182 previous = orig_code;
3183 }
3184 }
3185 else code = orig_code;
3186
3187 /* Remember where this code item starts so we can pick up the length
3188 next time round. */
3189
3190 last_code = code;
3191 }
3192
3193 /* In the real compile phase, just check the workspace used by the forward
3194 reference list. */
3195
3196 else if (cd->hwm > cd->start_workspace + WORK_SIZE_CHECK)
3197 {
3198 *errorcodeptr = ERR52;
3199 goto FAILED;
3200 }
3201
3202 /* If in \Q...\E, check for the end; if not, we have a literal */
3203
3204 if (inescq && c != 0)
3205 {
3206 if (c == CHAR_BACKSLASH && ptr[1] == CHAR_E)
3207 {
3208 inescq = FALSE;
3209 ptr++;
3210 continue;
3211 }
3212 else
3213 {
3214 if (previous_callout != NULL)
3215 {
3216 if (lengthptr == NULL) /* Don't attempt in pre-compile phase */
3217 complete_callout(previous_callout, ptr, cd);
3218 previous_callout = NULL;
3219 }
3220 if ((options & PCRE_AUTO_CALLOUT) != 0)
3221 {
3222 previous_callout = code;
3223 code = auto_callout(code, ptr, cd);
3224 }
3225 goto NORMAL_CHAR;
3226 }
3227 }
3228
3229 /* Fill in length of a previous callout, except when the next thing is
3230 a quantifier. */
3231
3232 is_quantifier =
3233 c == CHAR_ASTERISK || c == CHAR_PLUS || c == CHAR_QUESTION_MARK ||
3234 (c == CHAR_LEFT_CURLY_BRACKET && is_counted_repeat(ptr+1));
3235
3236 if (!is_quantifier && previous_callout != NULL &&
3237 after_manual_callout-- <= 0)
3238 {
3239 if (lengthptr == NULL) /* Don't attempt in pre-compile phase */
3240 complete_callout(previous_callout, ptr, cd);
3241 previous_callout = NULL;
3242 }
3243
3244 /* In extended mode, skip white space and comments. */
3245
3246 if ((options & PCRE_EXTENDED) != 0)
3247 {
3248 if ((cd->ctypes[c] & ctype_space) != 0) continue;
3249 if (c == CHAR_NUMBER_SIGN)
3250 {
3251 ptr++;
3252 while (*ptr != 0)
3253 {
3254 if (IS_NEWLINE(ptr)) { ptr += cd->nllen - 1; break; }
3255 ptr++;
3256 #ifdef SUPPORT_UTF8
3257 if (utf8) while ((*ptr & 0xc0) == 0x80) ptr++;
3258 #endif
3259 }
3260 if (*ptr != 0) continue;
3261
3262 /* Else fall through to handle end of string */
3263 c = 0;
3264 }
3265 }
3266
3267 /* No auto callout for quantifiers. */
3268
3269 if ((options & PCRE_AUTO_CALLOUT) != 0 && !is_quantifier)
3270 {
3271 previous_callout = code;
3272 code = auto_callout(code, ptr, cd);
3273 }
3274
3275 switch(c)
3276 {
3277 /* ===================================================================*/
3278 case 0: /* The branch terminates at string end */
3279 case CHAR_VERTICAL_LINE: /* or | or ) */
3280 case CHAR_RIGHT_PARENTHESIS:
3281 *firstbyteptr = firstbyte;
3282 *reqbyteptr = reqbyte;
3283 *codeptr = code;
3284 *ptrptr = ptr;
3285 if (lengthptr != NULL)
3286 {
3287 if (OFLOW_MAX - *lengthptr < code - last_code)
3288 {
3289 *errorcodeptr = ERR20;
3290 goto FAILED;
3291 }
3292 *lengthptr += (int)(code - last_code); /* To include callout length */
3293 DPRINTF((">> end branch\n"));
3294 }
3295 return TRUE;
3296
3297
3298 /* ===================================================================*/
3299 /* Handle single-character metacharacters. In multiline mode, ^ disables
3300 the setting of any following char as a first character. */
3301
3302 case CHAR_CIRCUMFLEX_ACCENT:
3303 previous = NULL;
3304 if ((options & PCRE_MULTILINE) != 0)
3305 {
3306 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
3307 *code++ = OP_CIRCM;
3308 }
3309 else *code++ = OP_CIRC;
3310 break;
3311
3312 case CHAR_DOLLAR_SIGN:
3313 previous = NULL;
3314 *code++ = ((options & PCRE_MULTILINE) != 0)? OP_DOLLM : OP_DOLL;
3315 break;
3316
3317 /* There can never be a first char if '.' is first, whatever happens about
3318 repeats. The value of reqbyte doesn't change either. */
3319
3320 case CHAR_DOT:
3321 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
3322 zerofirstbyte = firstbyte;
3323 zeroreqbyte = reqbyte;
3324 previous = code;
3325 *code++ = ((options & PCRE_DOTALL) != 0)? OP_ALLANY: OP_ANY;
3326 break;
3327
3328
3329 /* ===================================================================*/
3330 /* Character classes. If the included characters are all < 256, we build a
3331 32-byte bitmap of the permitted characters, except in the special case
3332 where there is only one such character. For negated classes, we build the
3333 map as usual, then invert it at the end. However, we use a different opcode
3334 so that data characters > 255 can be handled correctly.
3335
3336 If the class contains characters outside the 0-255 range, a different
3337 opcode is compiled. It may optionally have a bit map for characters < 256,
3338 but those above are are explicitly listed afterwards. A flag byte tells
3339 whether the bitmap is present, and whether this is a negated class or not.
3340
3341 In JavaScript compatibility mode, an isolated ']' causes an error. In
3342 default (Perl) mode, it is treated as a data character. */
3343
3344 case CHAR_RIGHT_SQUARE_BRACKET:
3345 if ((cd->external_options & PCRE_JAVASCRIPT_COMPAT) != 0)
3346 {
3347 *errorcodeptr = ERR64;
3348 goto FAILED;
3349 }
3350 goto NORMAL_CHAR;
3351
3352 case CHAR_LEFT_SQUARE_BRACKET:
3353 previous = code;
3354
3355 /* PCRE supports POSIX class stuff inside a class. Perl gives an error if
3356 they are encountered at the top level, so we'll do that too. */
3357
3358 if ((ptr[1] == CHAR_COLON || ptr[1] == CHAR_DOT ||
3359 ptr[1] == CHAR_EQUALS_SIGN) &&
3360 check_posix_syntax(ptr, &tempptr))
3361 {
3362 *errorcodeptr = (ptr[1] == CHAR_COLON)? ERR13 : ERR31;
3363 goto FAILED;
3364 }
3365
3366 /* If the first character is '^', set the negation flag and skip it. Also,
3367 if the first few characters (either before or after ^) are \Q\E or \E we
3368 skip them too. This makes for compatibility with Perl. */
3369
3370 negate_class = FALSE;
3371 for (;;)
3372 {
3373 c = *(++ptr);
3374 if (c == CHAR_BACKSLASH)
3375 {
3376 if (ptr[1] == CHAR_E)
3377 ptr++;
3378 else if (strncmp((const char *)ptr+1,
3379 STR_Q STR_BACKSLASH STR_E, 3) == 0)
3380 ptr += 3;
3381 else
3382 break;
3383 }
3384 else if (!negate_class && c == CHAR_CIRCUMFLEX_ACCENT)
3385 negate_class = TRUE;
3386 else break;
3387 }
3388
3389 /* Empty classes are allowed in JavaScript compatibility mode. Otherwise,
3390 an initial ']' is taken as a data character -- the code below handles
3391 that. In JS mode, [] must always fail, so generate OP_FAIL, whereas
3392 [^] must match any character, so generate OP_ALLANY. */
3393
3394 if (c == CHAR_RIGHT_SQUARE_BRACKET &&
3395 (cd->external_options & PCRE_JAVASCRIPT_COMPAT) != 0)
3396 {
3397 *code++ = negate_class? OP_ALLANY : OP_FAIL;
3398 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
3399 zerofirstbyte = firstbyte;
3400 break;
3401 }
3402
3403 /* If a class contains a negative special such as \S, we need to flip the
3404 negation flag at the end, so that support for characters > 255 works
3405 correctly (they are all included in the class). */
3406
3407 should_flip_negation = FALSE;
3408
3409 /* Keep a count of chars with values < 256 so that we can optimize the case
3410 of just a single character (as long as it's < 256). However, For higher
3411 valued UTF-8 characters, we don't yet do any optimization. */
3412
3413 class_charcount = 0;
3414 class_lastchar = -1;
3415
3416 /* Initialize the 32-char bit map to all zeros. We build the map in a
3417 temporary bit of memory, in case the class contains only 1 character (less
3418 than 256), because in that case the compiled code doesn't use the bit map.
3419 */
3420
3421 memset(classbits, 0, 32 * sizeof(uschar));
3422
3423 #ifdef SUPPORT_UTF8
3424 class_utf8 = FALSE; /* No chars >= 256 */
3425 class_utf8data = code + LINK_SIZE + 2; /* For UTF-8 items */
3426 class_utf8data_base = class_utf8data; /* For resetting in pass 1 */
3427 #endif
3428
3429 /* Process characters until ] is reached. By writing this as a "do" it
3430 means that an initial ] is taken as a data character. At the start of the
3431 loop, c contains the first byte of the character. */
3432
3433 if (c != 0) do
3434 {
3435 const uschar *oldptr;
3436
3437 #ifdef SUPPORT_UTF8
3438 if (utf8 && c > 127)
3439 { /* Braces are required because the */
3440 GETCHARLEN(c, ptr, ptr); /* macro generates multiple statements */
3441 }
3442
3443 /* In the pre-compile phase, accumulate the length of any UTF-8 extra
3444 data and reset the pointer. This is so that very large classes that
3445 contain a zillion UTF-8 characters no longer overwrite the work space
3446 (which is on the stack). */
3447
3448 if (lengthptr != NULL)
3449 {
3450 *lengthptr += class_utf8data - class_utf8data_base;
3451 class_utf8data = class_utf8data_base;
3452 }
3453
3454 #endif
3455
3456 /* Inside \Q...\E everything is literal except \E */
3457
3458 if (inescq)
3459 {
3460 if (c == CHAR_BACKSLASH && ptr[1] == CHAR_E) /* If we are at \E */
3461 {
3462 inescq = FALSE; /* Reset literal state */
3463 ptr++; /* Skip the 'E' */
3464 continue; /* Carry on with next */
3465 }
3466 goto CHECK_RANGE; /* Could be range if \E follows */
3467 }
3468
3469 /* Handle POSIX class names. Perl allows a negation extension of the
3470 form [:^name:]. A square bracket that doesn't match the syntax is
3471 treated as a literal. We also recognize the POSIX constructions
3472 [.ch.] and [=ch=] ("collating elements") and fault them, as Perl
3473 5.6 and 5.8 do. */
3474
3475 if (c == CHAR_LEFT_SQUARE_BRACKET &&
3476 (ptr[1] == CHAR_COLON || ptr[1] == CHAR_DOT ||
3477 ptr[1] == CHAR_EQUALS_SIGN) && check_posix_syntax(ptr, &tempptr))
3478 {
3479 BOOL local_negate = FALSE;
3480 int posix_class, taboffset, tabopt;
3481 register const uschar *cbits = cd->cbits;
3482 uschar pbits[32];
3483
3484 if (ptr[1] != CHAR_COLON)
3485 {
3486 *errorcodeptr = ERR31;
3487 goto FAILED;
3488 }
3489
3490 ptr += 2;
3491 if (*ptr == CHAR_CIRCUMFLEX_ACCENT)
3492 {
3493 local_negate = TRUE;
3494 should_flip_negation = TRUE; /* Note negative special */
3495 ptr++;
3496 }
3497
3498 posix_class = check_posix_name(ptr, (int)(tempptr - ptr));
3499 if (posix_class < 0)
3500 {
3501 *errorcodeptr = ERR30;
3502 goto FAILED;
3503 }
3504
3505 /* If matching is caseless, upper and lower are converted to
3506 alpha. This relies on the fact that the class table starts with
3507 alpha, lower, upper as the first 3 entries. */
3508
3509 if ((options & PCRE_CASELESS) != 0 && posix_class <= 2)
3510 posix_class = 0;
3511
3512 /* When PCRE_UCP is set, some of the POSIX classes are converted to
3513 different escape sequences that use Unicode properties. */
3514
3515 #ifdef SUPPORT_UCP
3516 if ((options & PCRE_UCP) != 0)
3517 {
3518 int pc = posix_class + ((local_negate)? POSIX_SUBSIZE/2 : 0);
3519 if (posix_substitutes[pc] != NULL)
3520 {
3521 nestptr = tempptr + 1;
3522 ptr = posix_substitutes[pc] - 1;
3523 continue;
3524 }
3525 }
3526 #endif
3527 /* In the non-UCP case, we build the bit map for the POSIX class in a
3528 chunk of local store because we may be adding and subtracting from it,
3529 and we don't want to subtract bits that may be in the main map already.
3530 At the end we or the result into the bit map that is being built. */
3531
3532 posix_class *= 3;
3533
3534 /* Copy in the first table (always present) */
3535
3536 memcpy(pbits, cbits + posix_class_maps[posix_class],
3537 32 * sizeof(uschar));
3538
3539 /* If there is a second table, add or remove it as required. */
3540
3541 taboffset = posix_class_maps[posix_class + 1];
3542 tabopt = posix_class_maps[posix_class + 2];
3543
3544 if (taboffset >= 0)
3545 {
3546 if (tabopt >= 0)
3547 for (c = 0; c < 32; c++) pbits[c] |= cbits[c + taboffset];
3548 else
3549 for (c = 0; c < 32; c++) pbits[c] &= ~cbits[c + taboffset];
3550 }
3551
3552 /* Not see if we need to remove any special characters. An option
3553 value of 1 removes vertical space and 2 removes underscore. */
3554
3555 if (tabopt < 0) tabopt = -tabopt;
3556 if (tabopt == 1) pbits[1] &= ~0x3c;
3557 else if (tabopt == 2) pbits[11] &= 0x7f;
3558
3559 /* Add the POSIX table or its complement into the main table that is
3560 being built and we are done. */
3561
3562 if (local_negate)
3563 for (c = 0; c < 32; c++) classbits[c] |= ~pbits[c];
3564 else
3565 for (c = 0; c < 32; c++) classbits[c] |= pbits[c];
3566
3567 ptr = tempptr + 1;
3568 class_charcount = 10; /* Set > 1; assumes more than 1 per class */
3569 continue; /* End of POSIX syntax handling */
3570 }
3571
3572 /* Backslash may introduce a single character, or it may introduce one
3573 of the specials, which just set a flag. The sequence \b is a special
3574 case. Inside a class (and only there) it is treated as backspace. We
3575 assume that other escapes have more than one character in them, so set
3576 class_charcount bigger than one. Unrecognized escapes fall through and
3577 are either treated as literal characters (by default), or are faulted if
3578 PCRE_EXTRA is set. */
3579
3580 if (c == CHAR_BACKSLASH)
3581 {
3582 c = check_escape(&ptr, errorcodeptr, cd->bracount, options, TRUE);
3583 if (*errorcodeptr != 0) goto FAILED;
3584
3585 if (-c == ESC_b) c = CHAR_BS; /* \b is backspace in a class */
3586 else if (-c == ESC_Q) /* Handle start of quoted string */
3587 {
3588 if (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
3589 {
3590 ptr += 2; /* avoid empty string */
3591 }
3592 else inescq = TRUE;
3593 continue;
3594 }
3595 else if (-c == ESC_E) continue; /* Ignore orphan \E */
3596
3597 if (c < 0)
3598 {
3599 register const uschar *cbits = cd->cbits;
3600 class_charcount += 2; /* Greater than 1 is what matters */
3601
3602 switch (-c)
3603 {
3604 #ifdef SUPPORT_UCP
3605 case ESC_du: /* These are the values given for \d etc */
3606 case ESC_DU: /* when PCRE_UCP is set. We replace the */
3607 case ESC_wu: /* escape sequence with an appropriate \p */
3608 case ESC_WU: /* or \P to test Unicode properties instead */
3609 case ESC_su: /* of the default ASCII testing. */
3610 case ESC_SU:
3611 nestptr = ptr;
3612 ptr = substitutes[-c - ESC_DU] - 1; /* Just before substitute */
3613 class_charcount -= 2; /* Undo! */
3614 continue;
3615 #endif
3616 case ESC_d:
3617 for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_digit];
3618 continue;
3619
3620 case ESC_D:
3621 should_flip_negation = TRUE;
3622 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_digit];
3623 continue;
3624
3625 case ESC_w:
3626 for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_word];
3627 continue;
3628
3629 case ESC_W:
3630 should_flip_negation = TRUE;
3631 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_word];
3632 continue;
3633
3634 /* Perl 5.004 onwards omits VT from \s, but we must preserve it
3635 if it was previously set by something earlier in the character
3636 class. */
3637
3638 case ESC_s:
3639 classbits[0] |= cbits[cbit_space];
3640 classbits[1] |= cbits[cbit_space+1] & ~0x08;
3641 for (c = 2; c < 32; c++) classbits[c] |= cbits[c+cbit_space];
3642 continue;
3643
3644 case ESC_S:
3645 should_flip_negation = TRUE;
3646 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_space];
3647 classbits[1] |= 0x08; /* Perl 5.004 onwards omits VT from \s */
3648 continue;
3649
3650 case ESC_h:
3651 SETBIT(classbits, 0x09); /* VT */
3652 SETBIT(classbits, 0x20); /* SPACE */
3653 SETBIT(classbits, 0xa0); /* NSBP */
3654 #ifdef SUPPORT_UTF8
3655 if (utf8)
3656 {
3657 class_utf8 = TRUE;
3658 *class_utf8data++ = XCL_SINGLE;
3659 class_utf8data += _pcre_ord2utf8(0x1680, class_utf8data);
3660 *class_utf8data++ = XCL_SINGLE;
3661 class_utf8data += _pcre_ord2utf8(0x180e, class_utf8data);
3662 *class_utf8data++ = XCL_RANGE;
3663 class_utf8data += _pcre_ord2utf8(0x2000, class_utf8data);
3664 class_utf8data += _pcre_ord2utf8(0x200A, class_utf8data);
3665 *class_utf8data++ = XCL_SINGLE;
3666 class_utf8data += _pcre_ord2utf8(0x202f, class_utf8data);
3667 *class_utf8data++ = XCL_SINGLE;
3668 class_utf8data += _pcre_ord2utf8(0x205f, class_utf8data);
3669 *class_utf8data++ = XCL_SINGLE;
3670 class_utf8data += _pcre_ord2utf8(0x3000, class_utf8data);
3671 }
3672 #endif
3673 continue;
3674
3675 case ESC_H:
3676 for (c = 0; c < 32; c++)
3677 {
3678 int x = 0xff;
3679 switch (c)
3680 {
3681 case 0x09/8: x ^= 1 << (0x09%8); break;
3682 case 0x20/8: x ^= 1 << (0x20%8); break;
3683 case 0xa0/8: x ^= 1 << (0xa0%8); break;
3684 default: break;
3685 }
3686 classbits[c] |= x;
3687 }
3688
3689 #ifdef SUPPORT_UTF8
3690 if (utf8)
3691 {
3692 class_utf8 = TRUE;
3693 *class_utf8data++ = XCL_RANGE;
3694 class_utf8data += _pcre_ord2utf8(0x0100, class_utf8data);
3695 class_utf8data += _pcre_ord2utf8(0x167f, class_utf8data);
3696 *class_utf8data++ = XCL_RANGE;
3697 class_utf8data += _pcre_ord2utf8(0x1681, class_utf8data);
3698 class_utf8data += _pcre_ord2utf8(0x180d, class_utf8data);
3699 *class_utf8data++ = XCL_RANGE;
3700 class_utf8data += _pcre_ord2utf8(0x180f, class_utf8data);
3701 class_utf8data += _pcre_ord2utf8(0x1fff, class_utf8data);
3702 *class_utf8data++ = XCL_RANGE;
3703 class_utf8data += _pcre_ord2utf8(0x200B, class_utf8data);
3704 class_utf8data += _pcre_ord2utf8(0x202e, class_utf8data);
3705 *class_utf8data++ = XCL_RANGE;
3706 class_utf8data += _pcre_ord2utf8(0x2030, class_utf8data);
3707 class_utf8data += _pcre_ord2utf8(0x205e, class_utf8data);
3708 *class_utf8data++ = XCL_RANGE;
3709 class_utf8data += _pcre_ord2utf8(0x2060, class_utf8data);
3710 class_utf8data += _pcre_ord2utf8(0x2fff, class_utf8data);
3711 *class_utf8data++ = XCL_RANGE;
3712 class_utf8data += _pcre_ord2utf8(0x3001, class_utf8data);
3713 class_utf8data += _pcre_ord2utf8(0x7fffffff, class_utf8data);
3714 }
3715 #endif
3716 continue;
3717
3718 case ESC_v:
3719 SETBIT(classbits, 0x0a); /* LF */
3720 SETBIT(classbits, 0x0b); /* VT */
3721 SETBIT(classbits, 0x0c); /* FF */
3722 SETBIT(classbits, 0x0d); /* CR */
3723 SETBIT(classbits, 0x85); /* NEL */
3724 #ifdef SUPPORT_UTF8
3725 if (utf8)
3726 {
3727 class_utf8 = TRUE;
3728 *class_utf8data++ = XCL_RANGE;
3729 class_utf8data += _pcre_ord2utf8(0x2028, class_utf8data);
3730 class_utf8data += _pcre_ord2utf8(0x2029, class_utf8data);
3731 }
3732 #endif
3733 continue;
3734
3735 case ESC_V:
3736 for (c = 0; c < 32; c++)
3737 {
3738 int x = 0xff;
3739 switch (c)
3740 {
3741 case 0x0a/8: x ^= 1 << (0x0a%8);
3742 x ^= 1 << (0x0b%8);
3743 x ^= 1 << (0x0c%8);
3744 x ^= 1 << (0x0d%8);
3745 break;
3746 case 0x85/8: x ^= 1 << (0x85%8); break;
3747 default: break;
3748 }
3749 classbits[c] |= x;
3750 }
3751
3752 #ifdef SUPPORT_UTF8
3753 if (utf8)
3754 {
3755 class_utf8 = TRUE;
3756 *class_utf8data++ = XCL_RANGE;
3757 class_utf8data += _pcre_ord2utf8(0x0100, class_utf8data);
3758 class_utf8data += _pcre_ord2utf8(0x2027, class_utf8data);
3759 *class_utf8data++ = XCL_RANGE;
3760 class_utf8data += _pcre_ord2utf8(0x2029, class_utf8data);
3761 class_utf8data += _pcre_ord2utf8(0x7fffffff, class_utf8data);
3762 }
3763 #endif
3764 continue;
3765
3766 #ifdef SUPPORT_UCP
3767 case ESC_p:
3768 case ESC_P:
3769 {
3770 BOOL negated;
3771 int pdata;
3772 int ptype = get_ucp(&ptr, &negated, &pdata, errorcodeptr);
3773 if (ptype < 0) goto FAILED;
3774 class_utf8 = TRUE;
3775 *class_utf8data++ = ((-c == ESC_p) != negated)?
3776 XCL_PROP : XCL_NOTPROP;
3777 *class_utf8data++ = ptype;
3778 *class_utf8data++ = pdata;
3779 class_charcount -= 2; /* Not a < 256 character */
3780 continue;
3781 }
3782 #endif
3783 /* Unrecognized escapes are faulted if PCRE is running in its
3784 strict mode. By default, for compatibility with Perl, they are
3785 treated as literals. */
3786
3787 default:
3788 if ((options & PCRE_EXTRA) != 0)
3789 {
3790 *errorcodeptr = ERR7;
3791 goto FAILED;
3792 }
3793 class_charcount -= 2; /* Undo the default count from above */
3794 c = *ptr; /* Get the final character and fall through */
3795 break;
3796 }
3797 }
3798
3799 /* Fall through if we have a single character (c >= 0). This may be
3800 greater than 256 in UTF-8 mode. */
3801
3802 } /* End of backslash handling */
3803
3804 /* A single character may be followed by '-' to form a range. However,
3805 Perl does not permit ']' to be the end of the range. A '-' character
3806 at the end is treated as a literal. Perl ignores orphaned \E sequences
3807 entirely. The code for handling \Q and \E is messy. */
3808
3809 CHECK_RANGE:
3810 while (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
3811 {
3812 inescq = FALSE;
3813 ptr += 2;
3814 }
3815
3816 oldptr = ptr;
3817
3818 /* Remember \r or \n */
3819
3820 if (c == CHAR_CR || c == CHAR_NL) cd->external_flags |= PCRE_HASCRORLF;
3821
3822 /* Check for range */
3823
3824 if (!inescq && ptr[1] == CHAR_MINUS)
3825 {
3826 int d;
3827 ptr += 2;
3828 while (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_E) ptr += 2;
3829
3830 /* If we hit \Q (not followed by \E) at this point, go into escaped
3831 mode. */
3832
3833 while (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_Q)
3834 {
3835 ptr += 2;
3836 if (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_E)
3837 { ptr += 2; continue; }
3838 inescq = TRUE;
3839 break;
3840 }
3841
3842 if (*ptr == 0 || (!inescq && *ptr == CHAR_RIGHT_SQUARE_BRACKET))
3843 {
3844 ptr = oldptr;
3845 goto LONE_SINGLE_CHARACTER;
3846 }
3847
3848 #ifdef SUPPORT_UTF8
3849 if (utf8)
3850 { /* Braces are required because the */
3851 GETCHARLEN(d, ptr, ptr); /* macro generates multiple statements */
3852 }
3853 else
3854 #endif
3855 d = *ptr; /* Not UTF-8 mode */
3856
3857 /* The second part of a range can be a single-character escape, but
3858 not any of the other escapes. Perl 5.6 treats a hyphen as a literal
3859 in such circumstances. */
3860
3861 if (!inescq && d == CHAR_BACKSLASH)
3862 {
3863 d = check_escape(&ptr, errorcodeptr, cd->bracount, options, TRUE);
3864 if (*errorcodeptr != 0) goto FAILED;
3865
3866 /* \b is backspace; any other special means the '-' was literal */
3867
3868 if (d < 0)
3869 {
3870 if (d == -ESC_b) d = CHAR_BS; else
3871 {
3872 ptr = oldptr;
3873 goto LONE_SINGLE_CHARACTER; /* A few lines below */
3874 }
3875 }
3876 }
3877
3878 /* Check that the two values are in the correct order. Optimize
3879 one-character ranges */
3880
3881 if (d < c)
3882 {
3883 *errorcodeptr = ERR8;
3884 goto FAILED;
3885 }
3886
3887 if (d == c) goto LONE_SINGLE_CHARACTER; /* A few lines below */
3888
3889 /* Remember \r or \n */
3890
3891 if (d == CHAR_CR || d == CHAR_NL) cd->external_flags |= PCRE_HASCRORLF;
3892
3893 /* In UTF-8 mode, if the upper limit is > 255, or > 127 for caseless
3894 matching, we have to use an XCLASS with extra data items. Caseless
3895 matching for characters > 127 is available only if UCP support is
3896 available. */
3897
3898 #ifdef SUPPORT_UTF8
3899 if (utf8 && (d > 255 || ((options & PCRE_CASELESS) != 0 && d > 127)))
3900 {
3901 class_utf8 = TRUE;
3902
3903 /* With UCP support, we can find the other case equivalents of
3904 the relevant characters. There may be several ranges. Optimize how
3905 they fit with the basic range. */
3906
3907 #ifdef SUPPORT_UCP
3908 if ((options & PCRE_CASELESS) != 0)
3909 {
3910 unsigned int occ, ocd;
3911 unsigned int cc = c;
3912 unsigned int origd = d;
3913 while (get_othercase_range(&cc, origd, &occ, &ocd))
3914 {
3915 if (occ >= (unsigned int)c &&
3916 ocd <= (unsigned int)d)
3917 continue; /* Skip embedded ranges */
3918
3919 if (occ < (unsigned int)c &&
3920 ocd >= (unsigned int)c - 1) /* Extend the basic range */
3921 { /* if there is overlap, */
3922 c = occ; /* noting that if occ < c */
3923 continue; /* we can't have ocd > d */
3924 } /* because a subrange is */
3925 if (ocd > (unsigned int)d &&
3926 occ <= (unsigned int)d + 1) /* always shorter than */
3927 { /* the basic range. */
3928 d = ocd;
3929 continue;
3930 }
3931
3932 if (occ == ocd)
3933 {
3934 *class_utf8data++ = XCL_SINGLE;
3935 }
3936 else
3937 {
3938 *class_utf8data++ = XCL_RANGE;
3939 class_utf8data += _pcre_ord2utf8(occ, class_utf8data);
3940 }
3941 class_utf8data += _pcre_ord2utf8(ocd, class_utf8data);
3942 }
3943 }
3944 #endif /* SUPPORT_UCP */
3945
3946 /* Now record the original range, possibly modified for UCP caseless
3947 overlapping ranges. */
3948
3949 *class_utf8data++ = XCL_RANGE;
3950 class_utf8data += _pcre_ord2utf8(c, class_utf8data);
3951 class_utf8data += _pcre_ord2utf8(d, class_utf8data);
3952
3953 /* With UCP support, we are done. Without UCP support, there is no
3954 caseless matching for UTF-8 characters > 127; we can use the bit map
3955 for the smaller ones. */
3956
3957 #ifdef SUPPORT_UCP
3958 continue; /* With next character in the class */
3959 #else
3960 if ((options & PCRE_CASELESS) == 0 || c > 127) continue;
3961
3962 /* Adjust upper limit and fall through to set up the map */
3963
3964 d = 127;
3965
3966 #endif /* SUPPORT_UCP */
3967 }
3968 #endif /* SUPPORT_UTF8 */
3969
3970 /* We use the bit map for all cases when not in UTF-8 mode; else
3971 ranges that lie entirely within 0-127 when there is UCP support; else
3972 for partial ranges without UCP support. */
3973
3974 class_charcount += d - c + 1;
3975 class_lastchar = d;
3976
3977 /* We can save a bit of time by skipping this in the pre-compile. */
3978
3979 if (lengthptr == NULL) for (; c <= d; c++)
3980 {
3981 classbits[c/8] |= (1 << (c&7));
3982 if ((options & PCRE_CASELESS) != 0)
3983 {
3984 int uc = cd->fcc[c]; /* flip case */
3985 classbits[uc/8] |= (1 << (uc&7));
3986 }
3987 }
3988
3989 continue; /* Go get the next char in the class */
3990 }
3991
3992 /* Handle a lone single character - we can get here for a normal
3993 non-escape char, or after \ that introduces a single character or for an
3994 apparent range that isn't. */
3995
3996 LONE_SINGLE_CHARACTER:
3997
3998 /* Handle a character that cannot go in the bit map */
3999
4000 #ifdef SUPPORT_UTF8
4001 if (utf8 && (c > 255 || ((options & PCRE_CASELESS) != 0 && c > 127)))
4002 {
4003 class_utf8 = TRUE;
4004 *class_utf8data++ = XCL_SINGLE;
4005 class_utf8data += _pcre_ord2utf8(c, class_utf8data);
4006
4007 #ifdef SUPPORT_UCP
4008 if ((options & PCRE_CASELESS) != 0)
4009 {
4010 unsigned int othercase;
4011 if ((othercase = UCD_OTHERCASE(c)) != c)
4012 {
4013 *class_utf8data++ = XCL_SINGLE;
4014 class_utf8data += _pcre_ord2utf8(othercase, class_utf8data);
4015 }
4016 }
4017 #endif /* SUPPORT_UCP */
4018
4019 }
4020 else
4021 #endif /* SUPPORT_UTF8 */
4022
4023 /* Handle a single-byte character */
4024 {
4025 classbits[c/8] |= (1 << (c&7));
4026 if ((options & PCRE_CASELESS) != 0)
4027 {
4028 c = cd->fcc[c]; /* flip case */
4029 classbits[c/8] |= (1 << (c&7));
4030 }
4031 class_charcount++;
4032 class_lastchar = c;
4033 }
4034 }
4035
4036 /* Loop until ']' reached. This "while" is the end of the "do" far above.
4037 If we are at the end of an internal nested string, revert to the outer
4038 string. */
4039
4040 while (((c = *(++ptr)) != 0 ||
4041 (nestptr != NULL &&
4042 (ptr = nestptr, nestptr = NULL, c = *(++ptr)) != 0)) &&
4043 (c != CHAR_RIGHT_SQUARE_BRACKET || inescq));
4044
4045 /* Check for missing terminating ']' */
4046
4047 if (c == 0)
4048 {
4049 *errorcodeptr = ERR6;
4050 goto FAILED;
4051 }
4052
4053 /* If class_charcount is 1, we saw precisely one character whose value is
4054 less than 256. As long as there were no characters >= 128 and there was no
4055 use of \p or \P, in other words, no use of any XCLASS features, we can
4056 optimize.
4057
4058 In UTF-8 mode, we can optimize the negative case only if there were no
4059 characters >= 128 because OP_NOT and the related opcodes like OP_NOTSTAR
4060 operate on single-bytes characters only. This is an historical hangover.
4061 Maybe one day we can tidy these opcodes to handle multi-byte characters.
4062
4063 The optimization throws away the bit map. We turn the item into a
4064 1-character OP_CHAR[I] if it's positive, or OP_NOT[I] if it's negative.
4065 Note that OP_NOT[I] does not support multibyte characters. In the positive
4066 case, it can cause firstbyte to be set. Otherwise, there can be no first
4067 char if this item is first, whatever repeat count may follow. In the case
4068 of reqbyte, save the previous value for reinstating. */
4069
4070 #ifdef SUPPORT_UTF8
4071 if (class_charcount == 1 && !class_utf8 &&
4072 (!utf8 || !negate_class || class_lastchar < 128))
4073 #else
4074 if (class_charcount == 1)
4075 #endif
4076 {
4077 zeroreqbyte = reqbyte;
4078
4079 /* The OP_NOT[I] opcodes work on one-byte characters only. */
4080
4081 if (negate_class)
4082 {
4083 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
4084 zerofirstbyte = firstbyte;
4085 *code++ = ((options & PCRE_CASELESS) != 0)? OP_NOTI: OP_NOT;
4086 *code++ = class_lastchar;
4087 break;
4088 }
4089
4090 /* For a single, positive character, get the value into mcbuffer, and
4091 then we can handle this with the normal one-character code. */
4092
4093 #ifdef SUPPORT_UTF8
4094 if (utf8 && class_lastchar > 127)
4095 mclength = _pcre_ord2utf8(class_lastchar, mcbuffer);
4096 else
4097 #endif
4098 {
4099 mcbuffer[0] = class_lastchar;
4100 mclength = 1;
4101 }
4102 goto ONE_CHAR;
4103 } /* End of 1-char optimization */
4104
4105 /* The general case - not the one-char optimization. If this is the first
4106 thing in the branch, there can be no first char setting, whatever the
4107 repeat count. Any reqbyte setting must remain unchanged after any kind of
4108 repeat. */
4109
4110 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
4111 zerofirstbyte = firstbyte;
4112 zeroreqbyte = reqbyte;
4113
4114 /* If there are characters with values > 255, we have to compile an
4115 extended class, with its own opcode, unless there was a negated special
4116 such as \S in the class, and PCRE_UCP is not set, because in that case all
4117 characters > 255 are in the class, so any that were explicitly given as
4118 well can be ignored. If (when there are explicit characters > 255 that must
4119 be listed) there are no characters < 256, we can omit the bitmap in the
4120 actual compiled code. */
4121
4122 #ifdef SUPPORT_UTF8
4123 if (class_utf8 && (!should_flip_negation || (options & PCRE_UCP) != 0))
4124 {
4125 *class_utf8data++ = XCL_END; /* Marks the end of extra data */
4126 *code++ = OP_XCLASS;
4127 code += LINK_SIZE;
4128 *code = negate_class? XCL_NOT : 0;
4129
4130 /* If the map is required, move up the extra data to make room for it;
4131 otherwise just move the code pointer to the end of the extra data. */
4132
4133 if (class_charcount > 0)
4134 {
4135 *code++ |= XCL_MAP;
4136 memmove(code + 32, code, class_utf8data - code);
4137 memcpy(code, classbits, 32);
4138 code = class_utf8data + 32;
4139 }
4140 else code = class_utf8data;
4141
4142 /* Now fill in the complete length of the item */
4143
4144 PUT(previous, 1, code - previous);
4145 break; /* End of class handling */
4146 }
4147 #endif
4148
4149 /* If there are no characters > 255, or they are all to be included or
4150 excluded, set the opcode to OP_CLASS or OP_NCLASS, depending on whether the
4151 whole class was negated and whether there were negative specials such as \S
4152 (non-UCP) in the class. Then copy the 32-byte map into the code vector,
4153 negating it if necessary. */
4154
4155 *code++ = (negate_class == should_flip_negation) ? OP_CLASS : OP_NCLASS;
4156 if (negate_class)
4157 {
4158 if (lengthptr == NULL) /* Save time in the pre-compile phase */
4159 for (c = 0; c < 32; c++) code[c] = ~classbits[c];
4160 }
4161 else
4162 {
4163 memcpy(code, classbits, 32);
4164 }
4165 code += 32;
4166 break;
4167
4168
4169 /* ===================================================================*/
4170 /* Various kinds of repeat; '{' is not necessarily a quantifier, but this
4171 has been tested above. */
4172
4173 case CHAR_LEFT_CURLY_BRACKET:
4174 if (!is_quantifier) goto NORMAL_CHAR;
4175 ptr = read_repeat_counts(ptr+1, &repeat_min, &repeat_max, errorcodeptr);
4176 if (*errorcodeptr != 0) goto FAILED;
4177 goto REPEAT;
4178
4179 case CHAR_ASTERISK:
4180 repeat_min = 0;
4181 repeat_max = -1;
4182 goto REPEAT;
4183
4184 case CHAR_PLUS:
4185 repeat_min = 1;
4186 repeat_max = -1;
4187 goto REPEAT;
4188
4189 case CHAR_QUESTION_MARK:
4190 repeat_min = 0;
4191 repeat_max = 1;
4192
4193 REPEAT:
4194 if (previous == NULL)
4195 {
4196 *errorcodeptr = ERR9;
4197 goto FAILED;
4198 }
4199
4200 if (repeat_min == 0)
4201 {
4202 firstbyte = zerofirstbyte; /* Adjust for zero repeat */
4203 reqbyte = zeroreqbyte; /* Ditto */
4204 }
4205
4206 /* Remember whether this is a variable length repeat */
4207
4208 reqvary = (repeat_min == repeat_max)? 0 : REQ_VARY;
4209
4210 op_type = 0; /* Default single-char op codes */
4211 possessive_quantifier = FALSE; /* Default not possessive quantifier */
4212
4213 /* Save start of previous item, in case we have to move it up to make space
4214 for an inserted OP_ONCE for the additional '+' extension. */
4215
4216 tempcode = previous;
4217
4218 /* If the next character is '+', we have a possessive quantifier. This
4219 implies greediness, whatever the setting of the PCRE_UNGREEDY option.
4220 If the next character is '?' this is a minimizing repeat, by default,
4221 but if PCRE_UNGREEDY is set, it works the other way round. We change the
4222 repeat type to the non-default. */
4223
4224 if (ptr[1] == CHAR_PLUS)
4225 {
4226 repeat_type = 0; /* Force greedy */
4227 possessive_quantifier = TRUE;
4228 ptr++;
4229 }
4230 else if (ptr[1] == CHAR_QUESTION_MARK)
4231 {
4232 repeat_type = greedy_non_default;
4233 ptr++;
4234 }
4235 else repeat_type = greedy_default;
4236
4237 /* If previous was a recursion call, wrap it in atomic brackets so that
4238 previous becomes the atomic group. All recursions were so wrapped in the
4239 past, but it no longer happens for non-repeated recursions. In fact, the
4240 repeated ones could be re-implemented independently so as not to need this,
4241 but for the moment we rely on the code for repeating groups. */
4242
4243 if (*previous == OP_RECURSE)
4244 {
4245 memmove(previous + 1 + LINK_SIZE, previous, 1 + LINK_SIZE);
4246 *previous = OP_ONCE;
4247 PUT(previous, 1, 2 + 2*LINK_SIZE);
4248 previous[2 + 2*LINK_SIZE] = OP_KET;
4249 PUT(previous, 3 + 2*LINK_SIZE, 2 + 2*LINK_SIZE);
4250 code += 2 + 2 * LINK_SIZE;
4251 length_prevgroup = 3 + 3*LINK_SIZE;
4252
4253 /* When actually compiling, we need to check whether this was a forward
4254 reference, and if so, adjust the offset. */
4255
4256 if (lengthptr == NULL && cd->hwm >= cd->start_workspace + LINK_SIZE)
4257 {
4258 int offset = GET(cd->hwm, -LINK_SIZE);
4259 if (offset == previous + 1 - cd->start_code)
4260 PUT(cd->hwm, -LINK_SIZE, offset + 1 + LINK_SIZE);
4261 }
4262 }
4263
4264 /* Now handle repetition for the different types of item. */
4265
4266 /* If previous was a character match, abolish the item and generate a
4267 repeat item instead. If a char item has a minumum of more than one, ensure
4268 that it is set in reqbyte - it might not be if a sequence such as x{3} is
4269 the first thing in a branch because the x will have gone into firstbyte
4270 instead. */
4271
4272 if (*previous == OP_CHAR || *previous == OP_CHARI)
4273 {
4274 op_type = (*previous == OP_CHAR)? 0 : OP_STARI - OP_STAR;
4275
4276 /* Deal with UTF-8 characters that take up more than one byte. It's
4277 easier to write this out separately than try to macrify it. Use c to
4278 hold the length of the character in bytes, plus 0x80 to flag that it's a
4279 length rather than a small character. */
4280
4281 #ifdef SUPPORT_UTF8
4282 if (utf8 && (code[-1] & 0x80) != 0)
4283 {
4284 uschar *lastchar = code - 1;
4285 while((*lastchar & 0xc0) == 0x80) lastchar--;
4286 c = code - lastchar; /* Length of UTF-8 character */
4287 memcpy(utf8_char, lastchar, c); /* Save the char */
4288 c |= 0x80; /* Flag c as a length */
4289 }
4290 else
4291 #endif
4292
4293 /* Handle the case of a single byte - either with no UTF8 support, or
4294 with UTF-8 disabled, or for a UTF-8 character < 128. */
4295
4296 {
4297 c = code[-1];
4298 if (repeat_min > 1) reqbyte = c | req_caseopt | cd->req_varyopt;
4299 }
4300
4301 /* If the repetition is unlimited, it pays to see if the next thing on
4302 the line is something that cannot possibly match this character. If so,
4303 automatically possessifying this item gains some performance in the case
4304 where the match fails. */
4305
4306 if (!possessive_quantifier &&
4307 repeat_max < 0 &&
4308 check_auto_possessive(previous, utf8, ptr + 1, options, cd))
4309 {
4310 repeat_type = 0; /* Force greedy */
4311 possessive_quantifier = TRUE;
4312 }
4313
4314 goto OUTPUT_SINGLE_REPEAT; /* Code shared with single character types */
4315 }
4316
4317 /* If previous was a single negated character ([^a] or similar), we use
4318 one of the special opcodes, replacing it. The code is shared with single-
4319 character repeats by setting opt_type to add a suitable offset into
4320 repeat_type. We can also test for auto-possessification. OP_NOT and OP_NOTI
4321 are currently used only for single-byte chars. */
4322
4323 else if (*previous == OP_NOT || *previous == OP_NOTI)
4324 {
4325 op_type = ((*previous == OP_NOT)? OP_NOTSTAR : OP_NOTSTARI) - OP_STAR;
4326 c = previous[1];
4327 if (!possessive_quantifier &&
4328 repeat_max < 0 &&
4329 check_auto_possessive(previous, utf8, ptr + 1, options, cd))
4330 {
4331 repeat_type = 0; /* Force greedy */
4332 possessive_quantifier = TRUE;
4333 }
4334 goto OUTPUT_SINGLE_REPEAT;
4335 }
4336
4337 /* If previous was a character type match (\d or similar), abolish it and
4338 create a suitable repeat item. The code is shared with single-character
4339 repeats by setting op_type to add a suitable offset into repeat_type. Note
4340 the the Unicode property types will be present only when SUPPORT_UCP is
4341 defined, but we don't wrap the little bits of code here because it just
4342 makes it horribly messy. */
4343
4344 else if (*previous < OP_EODN)
4345 {
4346 uschar *oldcode;
4347 int prop_type, prop_value;
4348 op_type = OP_TYPESTAR - OP_STAR; /* Use type opcodes */
4349 c = *previous;
4350
4351 if (!possessive_quantifier &&
4352 repeat_max < 0 &&
4353 check_auto_possessive(previous, utf8, ptr + 1, options, cd))
4354 {
4355 repeat_type = 0; /* Force greedy */
4356 possessive_quantifier = TRUE;
4357 }
4358
4359 OUTPUT_SINGLE_REPEAT:
4360 if (*previous == OP_PROP || *previous == OP_NOTPROP)
4361 {
4362 prop_type = previous[1];
4363 prop_value = previous[2];
4364 }
4365 else prop_type = prop_value = -1;
4366
4367 oldcode = code;
4368 code = previous; /* Usually overwrite previous item */
4369
4370 /* If the maximum is zero then the minimum must also be zero; Perl allows
4371 this case, so we do too - by simply omitting the item altogether. */
4372
4373 if (repeat_max == 0) goto END_REPEAT;
4374
4375 /*--------------------------------------------------------------------*/
4376 /* This code is obsolete from release 8.00; the restriction was finally
4377 removed: */
4378
4379 /* All real repeats make it impossible to handle partial matching (maybe
4380 one day we will be able to remove this restriction). */
4381
4382 /* if (repeat_max != 1) cd->external_flags |= PCRE_NOPARTIAL; */
4383 /*--------------------------------------------------------------------*/
4384
4385 /* Combine the op_type with the repeat_type */
4386
4387 repeat_type += op_type;
4388
4389 /* A minimum of zero is handled either as the special case * or ?, or as
4390 an UPTO, with the maximum given. */
4391
4392 if (repeat_min == 0)
4393 {
4394 if (repeat_max == -1) *code++ = OP_STAR + repeat_type;
4395 else if (repeat_max == 1) *code++ = OP_QUERY + repeat_type;
4396 else
4397 {
4398 *code++ = OP_UPTO + repeat_type;
4399 PUT2INC(code, 0, repeat_max);
4400 }
4401 }
4402
4403 /* A repeat minimum of 1 is optimized into some special cases. If the
4404 maximum is unlimited, we use OP_PLUS. Otherwise, the original item is
4405 left in place and, if the maximum is greater than 1, we use OP_UPTO with
4406 one less than the maximum. */
4407
4408 else if (repeat_min == 1)
4409 {
4410 if (repeat_max == -1)
4411 *code++ = OP_PLUS + repeat_type;
4412 else
4413 {
4414 code = oldcode; /* leave previous item in place */
4415 if (repeat_max == 1) goto END_REPEAT;
4416 *code++ = OP_UPTO + repeat_type;
4417 PUT2INC(code, 0, repeat_max - 1);
4418 }
4419 }
4420
4421 /* The case {n,n} is just an EXACT, while the general case {n,m} is
4422 handled as an EXACT followed by an UPTO. */
4423
4424 else
4425 {
4426 *code++ = OP_EXACT + op_type; /* NB EXACT doesn't have repeat_type */
4427 PUT2INC(code, 0, repeat_min);
4428
4429 /* If the maximum is unlimited, insert an OP_STAR. Before doing so,
4430 we have to insert the character for the previous code. For a repeated
4431 Unicode property match, there are two extra bytes that define the
4432 required property. In UTF-8 mode, long characters have their length in
4433 c, with the 0x80 bit as a flag. */
4434
4435 if (repeat_max < 0)
4436 {
4437 #ifdef SUPPORT_UTF8
4438 if (utf8 && c >= 128)
4439 {
4440 memcpy(code, utf8_char, c & 7);
4441 code += c & 7;
4442 }
4443 else
4444 #endif
4445 {
4446 *code++ = c;
4447 if (prop_type >= 0)
4448 {
4449 *code++ = prop_type;
4450 *code++ = prop_value;
4451 }
4452 }
4453 *code++ = OP_STAR + repeat_type;
4454 }
4455
4456 /* Else insert an UPTO if the max is greater than the min, again
4457 preceded by the character, for the previously inserted code. If the
4458 UPTO is just for 1 instance, we can use QUERY instead. */
4459
4460 else if (repeat_max != repeat_min)
4461 {
4462 #ifdef SUPPORT_UTF8
4463 if (utf8 && c >= 128)
4464 {
4465 memcpy(code, utf8_char, c & 7);
4466 code += c & 7;
4467 }
4468 else
4469 #endif
4470 *code++ = c;
4471 if (prop_type >= 0)
4472 {
4473 *code++ = prop_type;
4474 *code++ = prop_value;
4475 }
4476 repeat_max -= repeat_min;
4477
4478 if (repeat_max == 1)
4479 {
4480 *code++ = OP_QUERY + repeat_type;
4481 }
4482 else
4483 {
4484 *code++ = OP_UPTO + repeat_type;
4485 PUT2INC(code, 0, repeat_max);
4486 }
4487 }
4488 }
4489
4490 /* The character or character type itself comes last in all cases. */
4491
4492 #ifdef SUPPORT_UTF8
4493 if (utf8 && c >= 128)
4494 {
4495 memcpy(code, utf8_char, c & 7);
4496 code += c & 7;
4497 }
4498 else
4499 #endif
4500 *code++ = c;
4501
4502 /* For a repeated Unicode property match, there are two extra bytes that
4503 define the required property. */
4504
4505 #ifdef SUPPORT_UCP
4506 if (prop_type >= 0)
4507 {
4508 *code++ = prop_type;
4509 *code++ = prop_value;
4510 }
4511 #endif
4512 }
4513
4514 /* If previous was a character class or a back reference, we put the repeat
4515 stuff after it, but just skip the item if the repeat was {0,0}. */
4516
4517 else if (*previous == OP_CLASS ||
4518 *previous == OP_NCLASS ||
4519 #ifdef SUPPORT_UTF8
4520 *previous == OP_XCLASS ||
4521 #endif
4522 *previous == OP_REF ||
4523 *previous == OP_REFI)
4524 {
4525 if (repeat_max == 0)
4526 {
4527 code = previous;
4528 goto END_REPEAT;
4529 }
4530
4531 /*--------------------------------------------------------------------*/
4532 /* This code is obsolete from release 8.00; the restriction was finally
4533 removed: */
4534
4535 /* All real repeats make it impossible to handle partial matching (maybe
4536 one day we will be able to remove this restriction). */
4537
4538 /* if (repeat_max != 1) cd->external_flags |= PCRE_NOPARTIAL; */
4539 /*--------------------------------------------------------------------*/
4540
4541 if (repeat_min == 0 && repeat_max == -1)
4542 *code++ = OP_CRSTAR + repeat_type;
4543 else if (repeat_min == 1 && repeat_max == -1)
4544 *code++ = OP_CRPLUS + repeat_type;
4545 else if (repeat_min == 0 && repeat_max == 1)
4546 *code++ = OP_CRQUERY + repeat_type;
4547 else
4548 {
4549 *code++ = OP_CRRANGE + repeat_type;
4550 PUT2INC(code, 0, repeat_min);
4551 if (repeat_max == -1) repeat_max = 0; /* 2-byte encoding for max */
4552 PUT2INC(code, 0, repeat_max);
4553 }
4554 }
4555
4556 /* If previous was a bracket group, we may have to replicate it in certain
4557 cases. Note that at this point we can encounter only the "basic" BRA and
4558 KET opcodes, as this is the place where they get converted into the more
4559 special varieties. */
4560
4561 else if (*previous == OP_BRA || *previous == OP_CBRA ||
4562 *previous == OP_ONCE || *previous == OP_COND)
4563 {
4564 register int i;
4565 int len = (int)(code - previous);
4566 uschar *bralink = NULL;
4567 uschar *brazeroptr = NULL;
4568
4569 /* Repeating a DEFINE group is pointless */
4570
4571 if (*previous == OP_COND && previous[LINK_SIZE+1] == OP_DEF)
4572 {
4573 *errorcodeptr = ERR55;
4574 goto FAILED;
4575 }
4576
4577 /* The case of a zero minimum is special because of the need to stick
4578 OP_BRAZERO in front of it, and because the group appears once in the
4579 data, whereas in other cases it appears the minimum number of times. For
4580 this reason, it is simplest to treat this case separately, as otherwise
4581 the code gets far too messy. There are several special subcases when the
4582 minimum is zero. */
4583
4584 if (repeat_min == 0)
4585 {
4586 /* If the maximum is also zero, we used to just omit the group from the
4587 output altogether, like this:
4588
4589 ** if (repeat_max == 0)
4590 ** {
4591 ** code = previous;
4592 ** goto END_REPEAT;
4593 ** }
4594
4595 However, that fails when a group is referenced as a subroutine from
4596 elsewhere in the pattern, so now we stick in OP_SKIPZERO in front of it
4597 so that it is skipped on execution. As we don't have a list of which
4598 groups are referenced, we cannot do this selectively.
4599
4600 If the maximum is 1 or unlimited, we just have to stick in the BRAZERO
4601 and do no more at this point. However, we do need to adjust any
4602 OP_RECURSE calls inside the group that refer to the group itself or any
4603 internal or forward referenced group, because the offset is from the
4604 start of the whole regex. Temporarily terminate the pattern while doing
4605 this. */
4606
4607 if (repeat_max <= 1) /* Covers 0, 1, and unlimited */
4608 {
4609 *code = OP_END;
4610 adjust_recurse(previous, 1, utf8, cd, save_hwm);
4611 memmove(previous+1, previous, len);
4612 code++;
4613 if (repeat_max == 0)
4614 {
4615 *previous++ = OP_SKIPZERO;
4616 goto END_REPEAT;
4617 }
4618 brazeroptr = previous; /* Save for possessive optimizing */
4619 *previous++ = OP_BRAZERO + repeat_type;
4620 }
4621
4622 /* If the maximum is greater than 1 and limited, we have to replicate
4623 in a nested fashion, sticking OP_BRAZERO before each set of brackets.
4624 The first one has to be handled carefully because it's the original
4625 copy, which has to be moved up. The remainder can be handled by code
4626 that is common with the non-zero minimum case below. We have to
4627 adjust the value or repeat_max, since one less copy is required. Once
4628 again, we may have to adjust any OP_RECURSE calls inside the group. */
4629
4630 else
4631 {
4632 int offset;
4633 *code = OP_END;
4634 adjust_recurse(previous, 2 + LINK_SIZE, utf8, cd, save_hwm);
4635 memmove(previous + 2 + LINK_SIZE, previous, len);
4636 code += 2 + LINK_SIZE;
4637 *previous++ = OP_BRAZERO + repeat_type;
4638 *previous++ = OP_BRA;
4639
4640 /* We chain together the bracket offset fields that have to be
4641 filled in later when the ends of the brackets are reached. */
4642
4643 offset = (bralink == NULL)? 0 : (int)(previous - bralink);
4644 bralink = previous;
4645 PUTINC(previous, 0, offset);
4646 }
4647
4648 repeat_max--;
4649 }
4650
4651 /* If the minimum is greater than zero, replicate the group as many
4652 times as necessary, and adjust the maximum to the number of subsequent
4653 copies that we need. If we set a first char from the group, and didn't
4654 set a required char, copy the latter from the former. If there are any
4655 forward reference subroutine calls in the group, there will be entries on
4656 the workspace list; replicate these with an appropriate increment. */
4657
4658 else
4659 {
4660 if (repeat_min > 1)
4661 {
4662 /* In the pre-compile phase, we don't actually do the replication. We
4663 just adjust the length as if we had. Do some paranoid checks for
4664 potential integer overflow. The INT64_OR_DOUBLE type is a 64-bit
4665 integer type when available, otherwise double. */
4666
4667 if (lengthptr != NULL)
4668 {
4669 int delta = (repeat_min - 1)*length_prevgroup;
4670 if ((INT64_OR_DOUBLE)(repeat_min - 1)*
4671 (INT64_OR_DOUBLE)length_prevgroup >
4672 (INT64_OR_DOUBLE)INT_MAX ||
4673 OFLOW_MAX - *lengthptr < delta)
4674 {
4675 *errorcodeptr = ERR20;
4676 goto FAILED;
4677 }
4678 *lengthptr += delta;
4679 }
4680
4681 /* This is compiling for real */
4682
4683 else
4684 {
4685 if (groupsetfirstbyte && reqbyte < 0) reqbyte = firstbyte;
4686 for (i = 1; i < repeat_min; i++)
4687 {
4688 uschar *hc;
4689 uschar *this_hwm = cd->hwm;
4690 memcpy(code, previous, len);
4691 for (hc = save_hwm; hc < this_hwm; hc += LINK_SIZE)
4692 {
4693 PUT(cd->hwm, 0, GET(hc, 0) + len);
4694 cd->hwm += LINK_SIZE;
4695 }
4696 save_hwm = this_hwm;
4697 code += len;
4698 }
4699 }
4700 }
4701
4702 if (repeat_max > 0) repeat_max -= repeat_min;
4703 }
4704
4705 /* This code is common to both the zero and non-zero minimum cases. If
4706 the maximum is limited, it replicates the group in a nested fashion,
4707 remembering the bracket starts on a stack. In the case of a zero minimum,
4708 the first one was set up above. In all cases the repeat_max now specifies
4709 the number of additional copies needed. Again, we must remember to
4710 replicate entries on the forward reference list. */
4711
4712 if (repeat_max >= 0)
4713 {
4714 /* In the pre-compile phase, we don't actually do the replication. We
4715 just adjust the length as if we had. For each repetition we must add 1
4716 to the length for BRAZERO and for all but the last repetition we must
4717 add 2 + 2*LINKSIZE to allow for the nesting that occurs. Do some
4718 paranoid checks to avoid integer overflow. The INT64_OR_DOUBLE type is
4719 a 64-bit integer type when available, otherwise double. */
4720
4721 if (lengthptr != NULL && repeat_max > 0)
4722 {
4723 int delta = repeat_max * (length_prevgroup + 1 + 2 + 2*LINK_SIZE) -
4724 2 - 2*LINK_SIZE; /* Last one doesn't nest */
4725 if ((INT64_OR_DOUBLE)repeat_max *
4726 (INT64_OR_DOUBLE)(length_prevgroup + 1 + 2 + 2*LINK_SIZE)
4727 > (INT64_OR_DOUBLE)INT_MAX ||
4728 OFLOW_MAX - *lengthptr < delta)
4729 {
4730 *errorcodeptr = ERR20;
4731 goto FAILED;
4732 }
4733 *lengthptr += delta;
4734 }
4735
4736 /* This is compiling for real */
4737
4738 else for (i = repeat_max - 1; i >= 0; i--)
4739 {
4740 uschar *hc;
4741 uschar *this_hwm = cd->hwm;
4742
4743 *code++ = OP_BRAZERO + repeat_type;
4744
4745 /* All but the final copy start a new nesting, maintaining the
4746 chain of brackets outstanding. */
4747
4748 if (i != 0)
4749 {
4750 int offset;
4751 *code++ = OP_BRA;
4752 offset = (bralink == NULL)? 0 : (int)(code - bralink);
4753 bralink = code;
4754 PUTINC(code, 0, offset);
4755 }
4756
4757 memcpy(code, previous, len);
4758 for (hc = save_hwm; hc < this_hwm; hc += LINK_SIZE)
4759 {
4760 PUT(cd->hwm, 0, GET(hc, 0) + len + ((i != 0)? 2+LINK_SIZE : 1));
4761 cd->hwm += LINK_SIZE;
4762 }
4763 save_hwm = this_hwm;
4764 code += len;
4765 }
4766
4767 /* Now chain through the pending brackets, and fill in their length
4768 fields (which are holding the chain links pro tem). */
4769
4770 while (bralink != NULL)
4771 {
4772 int oldlinkoffset;
4773 int offset = (int)(code - bralink + 1);
4774 uschar *bra = code - offset;
4775 oldlinkoffset = GET(bra, 1);
4776 bralink = (oldlinkoffset == 0)? NULL : bralink - oldlinkoffset;
4777 *code++ = OP_KET;
4778 PUTINC(code, 0, offset);
4779 PUT(bra, 1, offset);
4780 }
4781 }
4782
4783 /* If the maximum is unlimited, set a repeater in the final copy. For
4784 ONCE brackets, that's all we need to do. However, possessively repeated
4785 ONCE brackets can be converted into non-capturing brackets, as the
4786 behaviour of (?:xx)++ is the same as (?>xx)++ and this saves having to
4787 deal with possessive ONCEs specially.
4788
4789 Otherwise, if the quantifier was possessive, we convert the BRA code to
4790 the POS form, and the KET code to KETRPOS. (It turns out to be convenient
4791 at runtime to detect this kind of subpattern at both the start and at the
4792 end.) The use of special opcodes makes it possible to reduce greatly the
4793 stack usage in pcre_exec(). If the group is preceded by OP_BRAZERO,
4794 convert this to OP_BRAPOSZERO. Then cancel the possessive flag so that
4795 the default action below, of wrapping everything inside atomic brackets,
4796 does not happen.
4797
4798 Then, when we are doing the actual compile phase, check to see whether
4799 this group is one that could match an empty string. If so, convert the
4800 initial operator to the S form (e.g. OP_BRA -> OP_SBRA) so that runtime
4801 checking can be done. [This check is also applied to ONCE groups at
4802 runtime, but in a different way.] */
4803
4804 else
4805 {
4806 uschar *ketcode = code - 1 - LINK_SIZE;
4807 uschar *bracode = ketcode - GET(ketcode, 1);
4808
4809 if (*bracode == OP_ONCE && possessive_quantifier) *bracode = OP_BRA;
4810 if (*bracode == OP_ONCE)
4811 *ketcode = OP_KETRMAX + repeat_type;
4812 else
4813 {
4814 if (possessive_quantifier)
4815 {
4816 *bracode += 1; /* Switch to xxxPOS opcodes */
4817 *ketcode = OP_KETRPOS;
4818 if (brazeroptr != NULL) *brazeroptr = OP_BRAPOSZERO;
4819 possessive_quantifier = FALSE;
4820 }
4821 else *ketcode = OP_KETRMAX + repeat_type;
4822
4823 if (lengthptr == NULL)
4824 {
4825 uschar *scode = bracode;
4826 do
4827 {
4828 if (could_be_empty_branch(scode, ketcode, utf8, cd))
4829 {
4830 *bracode += OP_SBRA - OP_BRA;
4831 break;
4832 }
4833 scode += GET(scode, 1);
4834 }
4835 while (*scode == OP_ALT);
4836 }
4837 }
4838 }
4839 }
4840
4841 /* If previous is OP_FAIL, it was generated by an empty class [] in
4842 JavaScript mode. The other ways in which OP_FAIL can be generated, that is
4843 by (*FAIL) or (?!) set previous to NULL, which gives a "nothing to repeat"
4844 error above. We can just ignore the repeat in JS case. */
4845
4846 else if (*previous == OP_FAIL) goto END_REPEAT;
4847
4848 /* Else there's some kind of shambles */
4849
4850 else
4851 {
4852 *errorcodeptr = ERR11;
4853 goto FAILED;
4854 }
4855
4856 /* If the character following a repeat is '+', or if certain optimization
4857 tests above succeeded, possessive_quantifier is TRUE. For some opcodes,
4858 there are special alternative opcodes for this case. For anything else, we
4859 wrap the entire repeated item inside OP_ONCE brackets. Logically, the '+'
4860 notation is just syntactic sugar, taken from Sun's Java package, but the
4861 special opcodes can optimize it.
4862
4863 Possessively repeated subpatterns have already been handled in the code
4864 just above, so possessive_quantifier is always FALSE for them at this
4865 stage.
4866
4867 Note that the repeated item starts at tempcode, not at previous, which
4868 might be the first part of a string whose (former) last char we repeated.
4869
4870 Possessifying an 'exact' quantifier has no effect, so we can ignore it. But
4871 an 'upto' may follow. We skip over an 'exact' item, and then test the
4872 length of what remains before proceeding. */
4873
4874 if (possessive_quantifier)
4875 {
4876 int len;
4877
4878 if (*tempcode == OP_TYPEEXACT)
4879 tempcode += _pcre_OP_lengths[*tempcode] +
4880 ((tempcode[3] == OP_PROP || tempcode[3] == OP_NOTPROP)? 2 : 0);
4881
4882 else if (*tempcode == OP_EXACT || *tempcode == OP_NOTEXACT)
4883 {
4884 tempcode += _pcre_OP_lengths[*tempcode];
4885 #ifdef SUPPORT_UTF8
4886 if (utf8 && tempcode[-1] >= 0xc0)
4887 tempcode += _pcre_utf8_table4[tempcode[-1] & 0x3f];
4888 #endif
4889 }
4890
4891 len = (int)(code - tempcode);
4892 if (len > 0) switch (*tempcode)
4893 {
4894 case OP_STAR: *tempcode = OP_POSSTAR; break;
4895 case OP_PLUS: *tempcode = OP_POSPLUS; break;
4896 case OP_QUERY: *tempcode = OP_POSQUERY; break;
4897 case OP_UPTO: *tempcode = OP_POSUPTO; break;
4898
4899 case OP_STARI: *tempcode = OP_POSSTARI; break;
4900 case OP_PLUSI: *tempcode = OP_POSPLUSI; break;
4901 case OP_QUERYI: *tempcode = OP_POSQUERYI; break;
4902 case OP_UPTOI: *tempcode = OP_POSUPTOI; break;
4903
4904 case OP_NOTSTAR: *tempcode = OP_NOTPOSSTAR; break;
4905 case OP_NOTPLUS: *tempcode = OP_NOTPOSPLUS; break;
4906 case OP_NOTQUERY: *tempcode = OP_NOTPOSQUERY; break;
4907 case OP_NOTUPTO: *tempcode = OP_NOTPOSUPTO; break;
4908
4909 case OP_NOTSTARI: *tempcode = OP_NOTPOSSTARI; break;
4910 case OP_NOTPLUSI: *tempcode = OP_NOTPOSPLUSI; break;
4911 case OP_NOTQUERYI: *tempcode = OP_NOTPOSQUERYI; break;
4912 case OP_NOTUPTOI: *tempcode = OP_NOTPOSUPTOI; break;
4913
4914 case OP_TYPESTAR: *tempcode = OP_TYPEPOSSTAR; break;
4915 case OP_TYPEPLUS: *tempcode = OP_TYPEPOSPLUS; break;
4916 case OP_TYPEQUERY: *tempcode = OP_TYPEPOSQUERY; break;
4917 case OP_TYPEUPTO: *tempcode = OP_TYPEPOSUPTO; break;
4918
4919 /* Because we are moving code along, we must ensure that any
4920 pending recursive references are updated. */
4921
4922 default:
4923 *code = OP_END;
4924 adjust_recurse(tempcode, 1 + LINK_SIZE, utf8, cd, save_hwm);
4925 memmove(tempcode + 1+LINK_SIZE, tempcode, len);
4926 code += 1 + LINK_SIZE;
4927 len += 1 + LINK_SIZE;
4928 tempcode[0] = OP_ONCE;
4929 *code++ = OP_KET;
4930 PUTINC(code, 0, len);
4931 PUT(tempcode, 1, len);
4932 break;
4933 }
4934 }
4935
4936 /* In all case we no longer have a previous item. We also set the
4937 "follows varying string" flag for subsequently encountered reqbytes if
4938 it isn't already set and we have just passed a varying length item. */
4939
4940 END_REPEAT:
4941 previous = NULL;
4942 cd->req_varyopt |= reqvary;
4943 break;
4944
4945
4946 /* ===================================================================*/
4947 /* Start of nested parenthesized sub-expression, or comment or lookahead or
4948 lookbehind or option setting or condition or all the other extended
4949 parenthesis forms. */
4950
4951 case CHAR_LEFT_PARENTHESIS:
4952 newoptions = options;
4953 skipbytes = 0;
4954 bravalue = OP_CBRA;
4955 save_hwm = cd->hwm;
4956 reset_bracount = FALSE;
4957
4958 /* First deal with various "verbs" that can be introduced by '*'. */
4959
4960 if (*(++ptr) == CHAR_ASTERISK &&
4961 ((cd->ctypes[ptr[1]] & ctype_letter) != 0 || ptr[1] == ':'))
4962 {
4963 int i, namelen;
4964 int arglen = 0;
4965 const char *vn = verbnames;
4966 const uschar *name = ptr + 1;
4967 const uschar *arg = NULL;
4968 previous = NULL;
4969 while ((cd->ctypes[*++ptr] & ctype_letter) != 0) {};
4970 namelen = (int)(ptr - name);
4971
4972 if (*ptr == CHAR_COLON)
4973 {
4974 arg = ++ptr;
4975 while ((cd->ctypes[*ptr] & (ctype_letter|ctype_digit)) != 0
4976 || *ptr == '_') ptr++;
4977 arglen = (int)(ptr - arg);
4978 }
4979
4980 if (*ptr != CHAR_RIGHT_PARENTHESIS)
4981 {
4982 *errorcodeptr = ERR60;
4983 goto FAILED;
4984 }
4985
4986 /* Scan the table of verb names */
4987
4988 for (i = 0; i < verbcount; i++)
4989 {
4990 if (namelen == verbs[i].len &&
4991 strncmp((char *)name, vn, namelen) == 0)
4992 {
4993 /* Check for open captures before ACCEPT and convert it to
4994 ASSERT_ACCEPT if in an assertion. */
4995
4996 if (verbs[i].op == OP_ACCEPT)
4997 {
4998 open_capitem *oc;
4999 if (arglen != 0)
5000 {
5001 *errorcodeptr = ERR59;
5002 goto FAILED;
5003 }
5004 cd->had_accept = TRUE;
5005 for (oc = cd->open_caps; oc != NULL; oc = oc->next)
5006 {
5007 *code++ = OP_CLOSE;
5008 PUT2INC(code, 0, oc->number);
5009 }
5010 *code++ = (cd->assert_depth > 0)? OP_ASSERT_ACCEPT : OP_ACCEPT;
5011 }
5012
5013 /* Handle other cases with/without an argument */
5014
5015 else if (arglen == 0)
5016 {
5017 if (verbs[i].op < 0) /* Argument is mandatory */
5018 {
5019 *errorcodeptr = ERR66;
5020 goto FAILED;
5021 }
5022 *code = verbs[i].op;
5023 if (*code++ == OP_THEN)
5024 {
5025 PUT(code, 0, code - bcptr->current_branch - 1);
5026 code += LINK_SIZE;
5027 }
5028 }
5029
5030 else
5031 {
5032 if (verbs[i].op_arg < 0) /* Argument is forbidden */
5033 {
5034 *errorcodeptr = ERR59;
5035 goto FAILED;
5036 }
5037 *code = verbs[i].op_arg;
5038 if (*code++ == OP_THEN_ARG)
5039 {
5040 PUT(code, 0, code - bcptr->current_branch - 1);
5041 code += LINK_SIZE;
5042 }
5043 *code++ = arglen;
5044 memcpy(code, arg, arglen);
5045 code += arglen;
5046 *code++ = 0;
5047 }
5048
5049 break; /* Found verb, exit loop */
5050 }
5051
5052 vn += verbs[i].len + 1;
5053 }
5054
5055 if (i < verbcount) continue; /* Successfully handled a verb */
5056 *errorcodeptr = ERR60; /* Verb not recognized */
5057 goto FAILED;
5058 }
5059
5060 /* Deal with the extended parentheses; all are introduced by '?', and the
5061 appearance of any of them means that this is not a capturing group. */
5062
5063 else if (*ptr == CHAR_QUESTION_MARK)
5064 {
5065 int i, set, unset, namelen;
5066 int *optset;
5067 const uschar *name;
5068 uschar *slot;
5069
5070 switch (*(++ptr))
5071 {
5072 case CHAR_NUMBER_SIGN: /* Comment; skip to ket */
5073 ptr++;
5074 while (*ptr != 0 && *ptr != CHAR_RIGHT_PARENTHESIS) ptr++;
5075 if (*ptr == 0)
5076 {
5077 *errorcodeptr = ERR18;
5078 goto FAILED;
5079 }
5080 continue;
5081
5082
5083 /* ------------------------------------------------------------ */
5084 case CHAR_VERTICAL_LINE: /* Reset capture count for each branch */
5085 reset_bracount = TRUE;
5086 /* Fall through */
5087
5088 /* ------------------------------------------------------------ */
5089 case CHAR_COLON: /* Non-capturing bracket */
5090 bravalue = OP_BRA;
5091 ptr++;
5092 break;
5093
5094
5095 /* ------------------------------------------------------------ */
5096 case CHAR_LEFT_PARENTHESIS:
5097 bravalue = OP_COND; /* Conditional group */
5098
5099 /* A condition can be an assertion, a number (referring to a numbered
5100 group), a name (referring to a named group), or 'R', referring to
5101 recursion. R<digits> and R&name are also permitted for recursion tests.
5102
5103 There are several syntaxes for testing a named group: (?(name)) is used
5104 by Python; Perl 5.10 onwards uses (?(<name>) or (?('name')).
5105
5106 There are two unfortunate ambiguities, caused by history. (a) 'R' can
5107 be the recursive thing or the name 'R' (and similarly for 'R' followed
5108 by digits), and (b) a number could be a name that consists of digits.
5109 In both cases, we look for a name first; if not found, we try the other
5110 cases. */
5111
5112 /* For conditions that are assertions, check the syntax, and then exit
5113 the switch. This will take control down to where bracketed groups,
5114 including assertions, are processed. */
5115
5116 if (ptr[1] == CHAR_QUESTION_MARK && (ptr[2] == CHAR_EQUALS_SIGN ||
5117 ptr[2] == CHAR_EXCLAMATION_MARK || ptr[2] == CHAR_LESS_THAN_SIGN))
5118 break;
5119
5120 /* Most other conditions use OP_CREF (a couple change to OP_RREF
5121 below), and all need to skip 3 bytes at the start of the group. */
5122
5123 code[1+LINK_SIZE] = OP_CREF;
5124 skipbytes = 3;
5125 refsign = -1;
5126
5127 /* Check for a test for recursion in a named group. */
5128
5129 if (ptr[1] == CHAR_R && ptr[2] == CHAR_AMPERSAND)
5130 {
5131 terminator = -1;
5132 ptr += 2;
5133 code[1+LINK_SIZE] = OP_RREF; /* Change the type of test */
5134 }
5135
5136 /* Check for a test for a named group's having been set, using the Perl
5137 syntax (?(<name>) or (?('name') */
5138
5139 else if (ptr[1] == CHAR_LESS_THAN_SIGN)
5140 {
5141 terminator = CHAR_GREATER_THAN_SIGN;
5142 ptr++;
5143 }
5144 else if (ptr[1] == CHAR_APOSTROPHE)
5145 {
5146 terminator = CHAR_APOSTROPHE;
5147 ptr++;
5148 }
5149 else
5150 {
5151 terminator = 0;
5152 if (ptr[1] == CHAR_MINUS || ptr[1] == CHAR_PLUS) refsign = *(++ptr);
5153 }
5154
5155 /* We now expect to read a name; any thing else is an error */
5156
5157 if ((cd->ctypes[ptr[1]] & ctype_word) == 0)
5158 {
5159 ptr += 1; /* To get the right offset */
5160 *errorcodeptr = ERR28;
5161 goto FAILED;
5162 }
5163
5164 /* Read the name, but also get it as a number if it's all digits */
5165
5166 recno = 0;
5167 name = ++ptr;
5168 while ((cd->ctypes[*ptr] & ctype_word) != 0)
5169 {
5170 if (recno >= 0)
5171 recno = ((digitab[*ptr] & ctype_digit) != 0)?
5172 recno * 10 + *ptr - CHAR_0 : -1;
5173 ptr++;
5174 }
5175 namelen = (int)(ptr - name);
5176
5177 if ((terminator > 0 && *ptr++ != terminator) ||
5178 *ptr++ != CHAR_RIGHT_PARENTHESIS)
5179 {
5180 ptr--; /* Error offset */
5181 *errorcodeptr = ERR26;
5182 goto FAILED;
5183 }
5184
5185 /* Do no further checking in the pre-compile phase. */
5186
5187 if (lengthptr != NULL) break;
5188
5189 /* In the real compile we do the work of looking for the actual
5190 reference. If the string started with "+" or "-" we require the rest to
5191 be digits, in which case recno will be set. */
5192
5193 if (refsign > 0)
5194 {
5195 if (recno <= 0)
5196 {
5197 *errorcodeptr = ERR58;
5198 goto FAILED;
5199 }
5200 recno = (refsign == CHAR_MINUS)?
5201 cd->bracount - recno + 1 : recno +cd->bracount;
5202 if (recno <= 0 || recno > cd->final_bracount)
5203 {
5204 *errorcodeptr = ERR15;
5205 goto FAILED;
5206 }
5207 PUT2(code, 2+LINK_SIZE, recno);
5208 break;
5209 }
5210
5211 /* Otherwise (did not start with "+" or "-"), start by looking for the
5212 name. If we find a name, add one to the opcode to change OP_CREF or
5213 OP_RREF into OP_NCREF or OP_NRREF. These behave exactly the same,
5214 except they record that the reference was originally to a name. The
5215 information is used to check duplicate names. */
5216
5217 slot = cd->name_table;
5218 for (i = 0; i < cd->names_found; i++)
5219 {
5220 if (strncmp((char *)name, (char *)slot+2, namelen) == 0) break;
5221 slot += cd->name_entry_size;
5222 }
5223
5224 /* Found a previous named subpattern */
5225
5226 if (i < cd->names_found)
5227 {
5228 recno = GET2(slot, 0);
5229 PUT2(code, 2+LINK_SIZE, recno);
5230 code[1+LINK_SIZE]++;
5231 }
5232
5233 /* Search the pattern for a forward reference */
5234
5235 else if ((i = find_parens(cd, name, namelen,
5236 (options & PCRE_EXTENDED) != 0, utf8)) > 0)
5237 {
5238 PUT2(code, 2+LINK_SIZE, i);
5239 code[1+LINK_SIZE]++;
5240 }
5241
5242 /* If terminator == 0 it means that the name followed directly after
5243 the opening parenthesis [e.g. (?(abc)...] and in this case there are
5244 some further alternatives to try. For the cases where terminator != 0
5245 [things like (?(<name>... or (?('name')... or (?(R&name)... ] we have
5246 now checked all the possibilities, so give an error. */
5247
5248 else if (terminator != 0)
5249 {
5250 *errorcodeptr = ERR15;
5251 goto FAILED;
5252 }
5253
5254 /* Check for (?(R) for recursion. Allow digits after R to specify a
5255 specific group number. */
5256
5257 else if (*name == CHAR_R)
5258 {
5259 recno = 0;
5260 for (i = 1; i < namelen; i++)
5261 {
5262 if ((digitab[name[i]] & ctype_digit) == 0)
5263 {
5264 *errorcodeptr = ERR15;
5265 goto FAILED;
5266 }
5267 recno = recno * 10 + name[i] - CHAR_0;
5268 }
5269 if (recno == 0) recno = RREF_ANY;
5270 code[1+LINK_SIZE] = OP_RREF; /* Change test type */
5271 PUT2(code, 2+LINK_SIZE, recno);
5272 }
5273
5274 /* Similarly, check for the (?(DEFINE) "condition", which is always
5275 false. */
5276
5277 else if (namelen == 6 && strncmp((char *)name, STRING_DEFINE, 6) == 0)
5278 {
5279 code[1+LINK_SIZE] = OP_DEF;
5280 skipbytes = 1;
5281 }
5282
5283 /* Check for the "name" actually being a subpattern number. We are
5284 in the second pass here, so final_bracount is set. */
5285
5286 else if (recno > 0 && recno <= cd->final_bracount)
5287 {
5288 PUT2(code, 2+LINK_SIZE, recno);
5289 }
5290
5291 /* Either an unidentified subpattern, or a reference to (?(0) */
5292
5293 else
5294 {
5295 *errorcodeptr = (recno == 0)? ERR35: ERR15;
5296 goto FAILED;
5297 }
5298 break;
5299
5300
5301 /* ------------------------------------------------------------ */
5302 case CHAR_EQUALS_SIGN: /* Positive lookahead */
5303 bravalue = OP_ASSERT;
5304 cd->assert_depth += 1;
5305 ptr++;
5306 break;
5307
5308
5309 /* ------------------------------------------------------------ */
5310 case CHAR_EXCLAMATION_MARK: /* Negative lookahead */
5311 ptr++;
5312 if (*ptr == CHAR_RIGHT_PARENTHESIS) /* Optimize (?!) */
5313 {
5314 *code++ = OP_FAIL;
5315 previous = NULL;
5316 continue;
5317 }
5318 bravalue = OP_ASSERT_NOT;
5319 cd->assert_depth += 1;
5320 break;
5321
5322
5323 /* ------------------------------------------------------------ */
5324 case CHAR_LESS_THAN_SIGN: /* Lookbehind or named define */
5325 switch (ptr[1])
5326 {
5327 case CHAR_EQUALS_SIGN: /* Positive lookbehind */
5328 bravalue = OP_ASSERTBACK;
5329 cd->assert_depth += 1;
5330 ptr += 2;
5331 break;
5332
5333 case CHAR_EXCLAMATION_MARK: /* Negative lookbehind */
5334 bravalue = OP_ASSERTBACK_NOT;
5335 cd->assert_depth += 1;
5336 ptr += 2;
5337 break;
5338
5339 default: /* Could be name define, else bad */
5340 if ((cd->ctypes[ptr[1]] & ctype_word) != 0) goto DEFINE_NAME;
5341 ptr++; /* Correct offset for error */
5342 *errorcodeptr = ERR24;
5343 goto FAILED;
5344 }
5345 break;
5346
5347
5348 /* ------------------------------------------------------------ */
5349 case CHAR_GREATER_THAN_SIGN: /* One-time brackets */
5350 bravalue = OP_ONCE;
5351 ptr++;
5352 break;
5353
5354
5355 /* ------------------------------------------------------------ */
5356 case CHAR_C: /* Callout - may be followed by digits; */
5357 previous_callout = code; /* Save for later completion */
5358 after_manual_callout = 1; /* Skip one item before completing */
5359 *code++ = OP_CALLOUT;
5360 {
5361 int n = 0;
5362 while ((digitab[*(++ptr)] & ctype_digit) != 0)
5363 n = n * 10 + *ptr - CHAR_0;
5364 if (*ptr != CHAR_RIGHT_PARENTHESIS)
5365 {
5366 *errorcodeptr = ERR39;
5367 goto FAILED;
5368 }
5369 if (n > 255)
5370 {
5371 *errorcodeptr = ERR38;
5372 goto FAILED;
5373 }
5374 *code++ = n;
5375 PUT(code, 0, (int)(ptr - cd->start_pattern + 1)); /* Pattern offset */
5376 PUT(code, LINK_SIZE, 0); /* Default length */
5377 code += 2 * LINK_SIZE;
5378 }
5379 previous = NULL;
5380 continue;
5381
5382
5383 /* ------------------------------------------------------------ */
5384 case CHAR_P: /* Python-style named subpattern handling */
5385 if (*(++ptr) == CHAR_EQUALS_SIGN ||
5386 *ptr == CHAR_GREATER_THAN_SIGN) /* Reference or recursion */
5387 {
5388 is_recurse = *ptr == CHAR_GREATER_THAN_SIGN;
5389 terminator = CHAR_RIGHT_PARENTHESIS;
5390 goto NAMED_REF_OR_RECURSE;
5391 }
5392 else if (*ptr != CHAR_LESS_THAN_SIGN) /* Test for Python-style defn */
5393 {
5394 *errorcodeptr = ERR41;
5395 goto FAILED;
5396 }
5397 /* Fall through to handle (?P< as (?< is handled */
5398
5399
5400 /* ------------------------------------------------------------ */
5401 DEFINE_NAME: /* Come here from (?< handling */
5402 case CHAR_APOSTROPHE:
5403 {
5404 terminator = (*ptr == CHAR_LESS_THAN_SIGN)?
5405 CHAR_GREATER_THAN_SIGN : CHAR_APOSTROPHE;
5406 name = ++ptr;
5407
5408 while ((cd->ctypes[*ptr] & ctype_word) != 0) ptr++;
5409 namelen = (int)(ptr - name);
5410
5411 /* In the pre-compile phase, just do a syntax check. */
5412
5413 if (lengthptr != NULL)
5414 {
5415 if (*ptr != terminator)
5416 {
5417 *errorcodeptr = ERR42;
5418 goto FAILED;
5419 }
5420 if (cd->names_found >= MAX_NAME_COUNT)
5421 {
5422 *errorcodeptr = ERR49;
5423 goto FAILED;
5424 }
5425 if (namelen + 3 > cd->name_entry_size)
5426 {
5427 cd->name_entry_size = namelen + 3;
5428 if (namelen > MAX_NAME_SIZE)
5429 {
5430 *errorcodeptr = ERR48;
5431 goto FAILED;
5432 }
5433 }
5434 }
5435
5436 /* In the real compile, create the entry in the table, maintaining
5437 alphabetical order. Duplicate names for different numbers are
5438 permitted only if PCRE_DUPNAMES is set. Duplicate names for the same
5439 number are always OK. (An existing number can be re-used if (?|
5440 appears in the pattern.) In either event, a duplicate name results in
5441 a duplicate entry in the table, even if the number is the same. This
5442 is because the number of names, and hence the table size, is computed
5443 in the pre-compile, and it affects various numbers and pointers which
5444 would all have to be modified, and the compiled code moved down, if
5445 duplicates with the same number were omitted from the table. This
5446 doesn't seem worth the hassle. However, *different* names for the
5447 same number are not permitted. */
5448
5449 else
5450 {
5451 BOOL dupname = FALSE;
5452 slot = cd->name_table;
5453
5454 for (i = 0; i < cd->names_found; i++)
5455 {
5456 int crc = memcmp(name, slot+2, namelen);
5457 if (crc == 0)
5458 {
5459 if (slot[2+namelen] == 0)
5460 {
5461 if (GET2(slot, 0) != cd->bracount + 1 &&
5462 (options & PCRE_DUPNAMES) == 0)
5463 {
5464 *errorcodeptr = ERR43;
5465 goto FAILED;
5466 }
5467 else dupname = TRUE;
5468 }
5469 else crc = -1; /* Current name is a substring */
5470 }
5471
5472 /* Make space in the table and break the loop for an earlier
5473 name. For a duplicate or later name, carry on. We do this for
5474 duplicates so that in the simple case (when ?(| is not used) they
5475 are in order of their numbers. */
5476
5477 if (crc < 0)
5478 {
5479 memmove(slot + cd->name_entry_size, slot,
5480 (cd->names_found - i) * cd->name_entry_size);
5481 break;
5482 }
5483
5484 /* Continue the loop for a later or duplicate name */
5485
5486 slot += cd->name_entry_size;
5487 }
5488
5489 /* For non-duplicate names, check for a duplicate number before
5490 adding the new name. */
5491
5492 if (!dupname)
5493 {
5494 uschar *cslot = cd->name_table;
5495 for (i = 0; i < cd->names_found; i++)
5496 {
5497 if (cslot != slot)
5498 {
5499 if (GET2(cslot, 0) == cd->bracount + 1)
5500 {
5501 *errorcodeptr = ERR65;
5502 goto FAILED;
5503 }
5504 }
5505 else i--;
5506 cslot += cd->name_entry_size;
5507 }
5508 }
5509
5510 PUT2(slot, 0, cd->bracount + 1);
5511 memcpy(slot + 2, name, namelen);
5512 slot[2+namelen] = 0;
5513 }
5514 }
5515
5516 /* In both pre-compile and compile, count the number of names we've
5517 encountered. */
5518
5519 cd->names_found++;
5520 ptr++; /* Move past > or ' */
5521 goto NUMBERED_GROUP;
5522
5523
5524 /* ------------------------------------------------------------ */
5525 case CHAR_AMPERSAND: /* Perl recursion/subroutine syntax */
5526 terminator = CHAR_RIGHT_PARENTHESIS;
5527 is_recurse = TRUE;
5528 /* Fall through */
5529
5530 /* We come here from the Python syntax above that handles both
5531 references (?P=name) and recursion (?P>name), as well as falling
5532 through from the Perl recursion syntax (?&name). We also come here from
5533 the Perl \k<name> or \k'name' back reference syntax and the \k{name}
5534 .NET syntax, and the Oniguruma \g<...> and \g'...' subroutine syntax. */
5535
5536 NAMED_REF_OR_RECURSE:
5537 name = ++ptr;
5538 while ((cd->ctypes[*ptr] & ctype_word) != 0) ptr++;
5539 namelen = (int)(ptr - name);
5540
5541 /* In the pre-compile phase, do a syntax check. We used to just set
5542 a dummy reference number, because it was not used in the first pass.
5543 However, with the change of recursive back references to be atomic,
5544 we have to look for the number so that this state can be identified, as
5545 otherwise the incorrect length is computed. If it's not a backwards
5546 reference, the dummy number will do. */
5547
5548 if (lengthptr != NULL)
5549 {
5550 const uschar *temp;
5551
5552 if (namelen == 0)
5553 {
5554 *errorcodeptr = ERR62;
5555 goto FAILED;
5556 }
5557 if (*ptr != terminator)
5558 {
5559 *errorcodeptr = ERR42;
5560 goto FAILED;
5561 }
5562 if (namelen > MAX_NAME_SIZE)
5563 {
5564 *errorcodeptr = ERR48;
5565 goto FAILED;
5566 }
5567
5568 /* The name table does not exist in the first pass, so we cannot
5569 do a simple search as in the code below. Instead, we have to scan the
5570 pattern to find the number. It is important that we scan it only as
5571 far as we have got because the syntax of named subpatterns has not
5572 been checked for the rest of the pattern, and find_parens() assumes
5573 correct syntax. In any case, it's a waste of resources to scan
5574 further. We stop the scan at the current point by temporarily
5575 adjusting the value of cd->endpattern. */
5576
5577 temp = cd->end_pattern;
5578 cd->end_pattern = ptr;
5579 recno = find_parens(cd, name, namelen,
5580 (options & PCRE_EXTENDED) != 0, utf8);
5581 cd->end_pattern = temp;
5582 if (recno < 0) recno = 0; /* Forward ref; set dummy number */
5583 }
5584
5585 /* In the real compile, seek the name in the table. We check the name
5586 first, and then check that we have reached the end of the name in the
5587 table. That way, if the name that is longer than any in the table,
5588 the comparison will fail without reading beyond the table entry. */
5589
5590 else
5591 {
5592 slot = cd->name_table;
5593 for (i = 0; i < cd->names_found; i++)
5594 {
5595 if (strncmp((char *)name, (char *)slot+2, namelen) == 0 &&
5596 slot[2+namelen] == 0)
5597 break;
5598 slot += cd->name_entry_size;
5599 }
5600
5601 if (i < cd->names_found) /* Back reference */
5602 {
5603 recno = GET2(slot, 0);
5604 }
5605 else if ((recno = /* Forward back reference */
5606 find_parens(cd, name, namelen,
5607 (options & PCRE_EXTENDED) != 0, utf8)) <= 0)
5608 {
5609 *errorcodeptr = ERR15;
5610 goto FAILED;
5611 }
5612 }
5613
5614 /* In both phases, we can now go to the code than handles numerical
5615 recursion or backreferences. */
5616
5617 if (is_recurse) goto HANDLE_RECURSION;
5618 else goto HANDLE_REFERENCE;
5619
5620
5621 /* ------------------------------------------------------------ */
5622 case CHAR_R: /* Recursion */
5623 ptr++; /* Same as (?0) */
5624 /* Fall through */
5625
5626
5627 /* ------------------------------------------------------------ */
5628 case CHAR_MINUS: case CHAR_PLUS: /* Recursion or subroutine */
5629 case CHAR_0: case CHAR_1: case CHAR_2: case CHAR_3: case CHAR_4:
5630 case CHAR_5: case CHAR_6: case CHAR_7: case CHAR_8: case CHAR_9:
5631 {
5632 const uschar *called;
5633 terminator = CHAR_RIGHT_PARENTHESIS;
5634
5635 /* Come here from the \g<...> and \g'...' code (Oniguruma
5636 compatibility). However, the syntax has been checked to ensure that
5637 the ... are a (signed) number, so that neither ERR63 nor ERR29 will
5638 be called on this path, nor with the jump to OTHER_CHAR_AFTER_QUERY
5639 ever be taken. */
5640
5641 HANDLE_NUMERICAL_RECURSION:
5642
5643 if ((refsign = *ptr) == CHAR_PLUS)
5644 {
5645 ptr++;
5646 if ((digitab[*ptr] & ctype_digit) == 0)
5647 {
5648 *errorcodeptr = ERR63;
5649 goto FAILED;
5650 }
5651 }
5652 else if (refsign == CHAR_MINUS)
5653 {
5654 if ((digitab[ptr[1]] & ctype_digit) == 0)
5655 goto OTHER_CHAR_AFTER_QUERY;
5656 ptr++;
5657 }
5658
5659 recno = 0;
5660 while((digitab[*ptr] & ctype_digit) != 0)
5661 recno = recno * 10 + *ptr++ - CHAR_0;
5662
5663 if (*ptr != terminator)
5664 {
5665 *errorcodeptr = ERR29;
5666 goto FAILED;
5667 }
5668
5669 if (refsign == CHAR_MINUS)
5670 {
5671 if (recno == 0)
5672 {
5673 *errorcodeptr = ERR58;
5674 goto FAILED;
5675 }
5676 recno = cd->bracount - recno + 1;
5677 if (recno <= 0)
5678 {
5679 *errorcodeptr = ERR15;
5680 goto FAILED;
5681 }
5682 }
5683 else if (refsign == CHAR_PLUS)
5684 {
5685 if (recno == 0)
5686 {
5687 *errorcodeptr = ERR58;
5688 goto FAILED;
5689 }
5690 recno += cd->bracount;
5691 }
5692
5693 /* Come here from code above that handles a named recursion */
5694
5695 HANDLE_RECURSION:
5696
5697 previous = code;
5698 called = cd->start_code;
5699
5700 /* When we are actually compiling, find the bracket that is being
5701 referenced. Temporarily end the regex in case it doesn't exist before
5702 this point. If we end up with a forward reference, first check that
5703 the bracket does occur later so we can give the error (and position)
5704 now. Then remember this forward reference in the workspace so it can
5705 be filled in at the end. */
5706
5707 if (lengthptr == NULL)
5708 {
5709 *code = OP_END;
5710 if (recno != 0)
5711 called = _pcre_find_bracket(cd->start_code, utf8, recno);
5712
5713 /* Forward reference */
5714
5715 if (called == NULL)
5716 {
5717 if (find_parens(cd, NULL, recno,
5718 (options & PCRE_EXTENDED) != 0, utf8) < 0)
5719 {
5720 *errorcodeptr = ERR15;
5721 goto FAILED;
5722 }
5723
5724 /* Fudge the value of "called" so that when it is inserted as an
5725 offset below, what it actually inserted is the reference number
5726 of the group. Then remember the forward reference. */
5727
5728 called = cd->start_code + recno;
5729 PUTINC(cd->hwm, 0, (int)(code + 1 - cd->start_code));
5730 }
5731
5732 /* If not a forward reference, and the subpattern is still open,
5733 this is a recursive call. We check to see if this is a left
5734 recursion that could loop for ever, and diagnose that case. */
5735
5736 else if (GET(called, 1) == 0 &&
5737 could_be_empty(called, code, bcptr, utf8, cd))
5738 {
5739 *errorcodeptr = ERR40;
5740 goto FAILED;
5741 }
5742 }
5743
5744 /* Insert the recursion/subroutine item. */
5745
5746 *code = OP_RECURSE;
5747 PUT(code, 1, (int)(called - cd->start_code));
5748 code += 1 + LINK_SIZE;
5749 }
5750
5751 /* Can't determine a first byte now */
5752
5753 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
5754 continue;
5755
5756
5757 /* ------------------------------------------------------------ */
5758 default: /* Other characters: check option setting */
5759 OTHER_CHAR_AFTER_QUERY:
5760 set = unset = 0;
5761 optset = &set;
5762
5763 while (*ptr != CHAR_RIGHT_PARENTHESIS && *ptr != CHAR_COLON)
5764 {
5765 switch (*ptr++)
5766 {
5767 case CHAR_MINUS: optset = &unset; break;
5768
5769 case CHAR_J: /* Record that it changed in the external options */
5770 *optset |= PCRE_DUPNAMES;
5771 cd->external_flags |= PCRE_JCHANGED;
5772 break;
5773
5774 case CHAR_i: *optset |= PCRE_CASELESS; break;
5775 case CHAR_m: *optset |= PCRE_MULTILINE; break;
5776 case CHAR_s: *optset |= PCRE_DOTALL; break;
5777 case CHAR_x: *optset |= PCRE_EXTENDED; break;
5778 case CHAR_U: *optset |= PCRE_UNGREEDY; break;
5779 case CHAR_X: *optset |= PCRE_EXTRA; break;
5780
5781 default: *errorcodeptr = ERR12;
5782 ptr--; /* Correct the offset */
5783 goto FAILED;
5784 }
5785 }
5786
5787 /* Set up the changed option bits, but don't change anything yet. */
5788
5789 newoptions = (options | set) & (~unset);
5790
5791 /* If the options ended with ')' this is not the start of a nested
5792 group with option changes, so the options change at this level. If this
5793 item is right at the start of the pattern, the options can be
5794 abstracted and made external in the pre-compile phase, and ignored in
5795 the compile phase. This can be helpful when matching -- for instance in
5796 caseless checking of required bytes.
5797
5798 If the code pointer is not (cd->start_code + 1 + LINK_SIZE), we are
5799 definitely *not* at the start of the pattern because something has been
5800 compiled. In the pre-compile phase, however, the code pointer can have
5801 that value after the start, because it gets reset as code is discarded
5802 during the pre-compile. However, this can happen only at top level - if
5803 we are within parentheses, the starting BRA will still be present. At
5804 any parenthesis level, the length value can be used to test if anything
5805 has been compiled at that level. Thus, a test for both these conditions
5806 is necessary to ensure we correctly detect the start of the pattern in
5807 both phases.
5808
5809 If we are not at the pattern start, reset the greedy defaults and the
5810 case value for firstbyte and reqbyte. */
5811
5812 if (*ptr == CHAR_RIGHT_PARENTHESIS)
5813 {
5814 if (code == cd->start_code + 1 + LINK_SIZE &&
5815 (lengthptr == NULL || *lengthptr == 2 + 2*LINK_SIZE))
5816 {
5817 cd->external_options = newoptions;
5818 }
5819 else
5820 {
5821 greedy_default = ((newoptions & PCRE_UNGREEDY) != 0);
5822 greedy_non_default = greedy_default ^ 1;
5823 req_caseopt = ((newoptions & PCRE_CASELESS) != 0)? REQ_CASELESS : 0;
5824 }
5825
5826 /* Change options at this level, and pass them back for use
5827 in subsequent branches. */
5828
5829 *optionsptr = options = newoptions;
5830 previous = NULL; /* This item can't be repeated */
5831 continue; /* It is complete */
5832 }
5833
5834 /* If the options ended with ':' we are heading into a nested group
5835 with possible change of options. Such groups are non-capturing and are
5836 not assertions of any kind. All we need to do is skip over the ':';
5837 the newoptions value is handled below. */
5838
5839 bravalue = OP_BRA;
5840 ptr++;
5841 } /* End of switch for character following (? */
5842 } /* End of (? handling */
5843
5844 /* Opening parenthesis not followed by '*' or '?'. If PCRE_NO_AUTO_CAPTURE
5845 is set, all unadorned brackets become non-capturing and behave like (?:...)
5846 brackets. */
5847
5848 else if ((options & PCRE_NO_AUTO_CAPTURE) != 0)
5849 {
5850 bravalue = OP_BRA;
5851 }
5852
5853 /* Else we have a capturing group. */
5854
5855 else
5856 {
5857 NUMBERED_GROUP:
5858 cd->bracount += 1;
5859 PUT2(code, 1+LINK_SIZE, cd->bracount);
5860 skipbytes = 2;
5861 }
5862
5863 /* Process nested bracketed regex. Assertions may not be repeated, but
5864 other kinds can be. All their opcodes are >= OP_ONCE. We copy code into a
5865 non-register variable (tempcode) in order to be able to pass its address
5866 because some compilers complain otherwise. */
5867
5868 previous = (bravalue >= OP_ONCE)? code : NULL;
5869 *code = bravalue;
5870 tempcode = code;
5871 tempreqvary = cd->req_varyopt; /* Save value before bracket */
5872 length_prevgroup = 0; /* Initialize for pre-compile phase */
5873
5874 if (!compile_regex(
5875 newoptions, /* The complete new option state */
5876 &tempcode, /* Where to put code (updated) */
5877 &ptr, /* Input pointer (updated) */
5878 errorcodeptr, /* Where to put an error message */
5879 (bravalue == OP_ASSERTBACK ||
5880 bravalue == OP_ASSERTBACK_NOT), /* TRUE if back assert */
5881 reset_bracount, /* True if (?| group */
5882 skipbytes, /* Skip over bracket number */
5883 &subfirstbyte, /* For possible first char */
5884 &subreqbyte, /* For possible last char */
5885 bcptr, /* Current branch chain */
5886 cd, /* Tables block */
5887 (lengthptr == NULL)? NULL : /* Actual compile phase */
5888 &length_prevgroup /* Pre-compile phase */
5889 ))
5890 goto FAILED;
5891
5892 if (bravalue >= OP_ASSERT && bravalue <= OP_ASSERTBACK_NOT)
5893 cd->assert_depth -= 1;
5894
5895 /* At the end of compiling, code is still pointing to the start of the
5896 group, while tempcode has been updated to point past the end of the group
5897 and any option resetting that may follow it. The pattern pointer (ptr)
5898 is on the bracket. */
5899
5900 /* If this is a conditional bracket, check that there are no more than
5901 two branches in the group, or just one if it's a DEFINE group. We do this
5902 in the real compile phase, not in the pre-pass, where the whole group may
5903 not be available. */
5904
5905 if (bravalue == OP_COND && lengthptr == NULL)
5906 {
5907 uschar *tc = code;
5908 int condcount = 0;
5909
5910 do {
5911 condcount++;
5912 tc += GET(tc,1);
5913 }
5914 while (*tc != OP_KET);
5915
5916 /* A DEFINE group is never obeyed inline (the "condition" is always
5917 false). It must have only one branch. */
5918
5919 if (code[LINK_SIZE+1] == OP_DEF)
5920 {
5921 if (condcount > 1)
5922 {
5923 *errorcodeptr = ERR54;
5924 goto FAILED;
5925 }
5926 bravalue = OP_DEF; /* Just a flag to suppress char handling below */
5927 }
5928
5929 /* A "normal" conditional group. If there is just one branch, we must not
5930 make use of its firstbyte or reqbyte, because this is equivalent to an
5931 empty second branch. */
5932
5933 else
5934 {
5935 if (condcount > 2)
5936 {
5937 *errorcodeptr = ERR27;
5938 goto FAILED;
5939 }
5940 if (condcount == 1) subfirstbyte = subreqbyte = REQ_NONE;
5941 }
5942 }
5943
5944 /* Error if hit end of pattern */
5945
5946 if (*ptr != CHAR_RIGHT_PARENTHESIS)
5947 {
5948 *errorcodeptr = ERR14;
5949 goto FAILED;
5950 }
5951
5952 /* In the pre-compile phase, update the length by the length of the group,
5953 less the brackets at either end. Then reduce the compiled code to just a
5954 set of non-capturing brackets so that it doesn't use much memory if it is
5955 duplicated by a quantifier.*/
5956
5957 if (lengthptr != NULL)
5958 {
5959 if (OFLOW_MAX - *lengthptr < length_prevgroup - 2 - 2*LINK_SIZE)
5960 {
5961 *errorcodeptr = ERR20;
5962 goto FAILED;
5963 }
5964 *lengthptr += length_prevgroup - 2 - 2*LINK_SIZE;
5965 code++; /* This already contains bravalue */
5966 PUTINC(code, 0, 1 + LINK_SIZE);
5967 *code++ = OP_KET;
5968 PUTINC(code, 0, 1 + LINK_SIZE);
5969 break; /* No need to waste time with special character handling */
5970 }
5971
5972 /* Otherwise update the main code pointer to the end of the group. */
5973
5974 code = tempcode;
5975
5976 /* For a DEFINE group, required and first character settings are not
5977 relevant. */
5978
5979 if (bravalue == OP_DEF) break;
5980
5981 /* Handle updating of the required and first characters for other types of
5982 group. Update for normal brackets of all kinds, and conditions with two
5983 branches (see code above). If the bracket is followed by a quantifier with
5984 zero repeat, we have to back off. Hence the definition of zeroreqbyte and
5985 zerofirstbyte outside the main loop so that they can be accessed for the
5986 back off. */
5987
5988 zeroreqbyte = reqbyte;
5989 zerofirstbyte = firstbyte;
5990 groupsetfirstbyte = FALSE;
5991
5992 if (bravalue >= OP_ONCE)
5993 {
5994 /* If we have not yet set a firstbyte in this branch, take it from the
5995 subpattern, remembering that it was set here so that a repeat of more
5996 than one can replicate it as reqbyte if necessary. If the subpattern has
5997 no firstbyte, set "none" for the whole branch. In both cases, a zero
5998 repeat forces firstbyte to "none". */
5999
6000 if (firstbyte == REQ_UNSET)
6001 {
6002 if (subfirstbyte >= 0)
6003 {
6004 firstbyte = subfirstbyte;
6005 groupsetfirstbyte = TRUE;
6006 }
6007 else firstbyte = REQ_NONE;
6008 zerofirstbyte = REQ_NONE;
6009 }
6010
6011 /* If firstbyte was previously set, convert the subpattern's firstbyte
6012 into reqbyte if there wasn't one, using the vary flag that was in
6013 existence beforehand. */
6014
6015 else if (subfirstbyte >= 0 && subreqbyte < 0)
6016 subreqbyte = subfirstbyte | tempreqvary;
6017
6018 /* If the subpattern set a required byte (or set a first byte that isn't
6019 really the first byte - see above), set it. */
6020
6021 if (subreqbyte >= 0) reqbyte = subreqbyte;
6022 }
6023
6024 /* For a forward assertion, we take the reqbyte, if set. This can be
6025 helpful if the pattern that follows the assertion doesn't set a different
6026 char. For example, it's useful for /(?=abcde).+/. We can't set firstbyte
6027 for an assertion, however because it leads to incorrect effect for patterns
6028 such as /(?=a)a.+/ when the "real" "a" would then become a reqbyte instead
6029 of a firstbyte. This is overcome by a scan at the end if there's no
6030 firstbyte, looking for an asserted first char. */
6031
6032 else if (bravalue == OP_ASSERT && subreqbyte >= 0) reqbyte = subreqbyte;
6033 break; /* End of processing '(' */
6034
6035
6036 /* ===================================================================*/
6037 /* Handle metasequences introduced by \. For ones like \d, the ESC_ values
6038 are arranged to be the negation of the corresponding OP_values in the
6039 default case when PCRE_UCP is not set. For the back references, the values
6040 are ESC_REF plus the reference number. Only back references and those types
6041 that consume a character may be repeated. We can test for values between
6042 ESC_b and ESC_Z for the latter; this may have to change if any new ones are
6043 ever created. */
6044
6045 case CHAR_BACKSLASH:
6046 tempptr = ptr;
6047 c = check_escape(&ptr, errorcodeptr, cd->bracount, options, FALSE);
6048 if (*errorcodeptr != 0) goto FAILED;
6049
6050 if (c < 0)
6051 {
6052 if (-c == ESC_Q) /* Handle start of quoted string */
6053 {
6054 if (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
6055 ptr += 2; /* avoid empty string */
6056 else inescq = TRUE;
6057 continue;
6058 }
6059
6060 if (-c == ESC_E) continue; /* Perl ignores an orphan \E */
6061
6062 /* For metasequences that actually match a character, we disable the
6063 setting of a first character if it hasn't already been set. */
6064
6065 if (firstbyte == REQ_UNSET && -c > ESC_b && -c < ESC_Z)
6066 firstbyte = REQ_NONE;
6067
6068 /* Set values to reset to if this is followed by a zero repeat. */
6069
6070 zerofirstbyte = firstbyte;
6071 zeroreqbyte = reqbyte;
6072
6073 /* \g<name> or \g'name' is a subroutine call by name and \g<n> or \g'n'
6074 is a subroutine call by number (Oniguruma syntax). In fact, the value
6075 -ESC_g is returned only for these cases. So we don't need to check for <
6076 or ' if the value is -ESC_g. For the Perl syntax \g{n} the value is
6077 -ESC_REF+n, and for the Perl syntax \g{name} the result is -ESC_k (as
6078 that is a synonym for a named back reference). */
6079
6080 if (-c == ESC_g)
6081 {
6082 const uschar *p;
6083 save_hwm = cd->hwm; /* Normally this is set when '(' is read */
6084 terminator = (*(++ptr) == CHAR_LESS_THAN_SIGN)?
6085 CHAR_GREATER_THAN_SIGN : CHAR_APOSTROPHE;
6086
6087 /* These two statements stop the compiler for warning about possibly
6088 unset variables caused by the jump to HANDLE_NUMERICAL_RECURSION. In
6089 fact, because we actually check for a number below, the paths that
6090 would actually be in error are never taken. */
6091
6092 skipbytes = 0;
6093 reset_bracount = FALSE;
6094
6095 /* Test for a name */
6096
6097 if (ptr[1] != CHAR_PLUS && ptr[1] != CHAR_MINUS)
6098 {
6099 BOOL isnumber = TRUE;
6100 for (p = ptr + 1; *p != 0 && *p != terminator; p++)
6101 {
6102 if ((cd->ctypes[*p] & ctype_digit) == 0) isnumber = FALSE;
6103 if ((cd->ctypes[*p] & ctype_word) == 0) break;
6104 }
6105 if (*p != terminator)
6106 {
6107 *errorcodeptr = ERR57;
6108 break;
6109 }
6110 if (isnumber)
6111 {
6112 ptr++;
6113 goto HANDLE_NUMERICAL_RECURSION;
6114 }
6115 is_recurse = TRUE;
6116 goto NAMED_REF_OR_RECURSE;
6117 }
6118
6119 /* Test a signed number in angle brackets or quotes. */
6120
6121 p = ptr + 2;
6122 while ((digitab[*p] & ctype_digit) != 0) p++;
6123 if (*p != terminator)
6124 {
6125 *errorcodeptr = ERR57;
6126 break;
6127 }
6128 ptr++;
6129 goto HANDLE_NUMERICAL_RECURSION;
6130 }
6131
6132 /* \k<name> or \k'name' is a back reference by name (Perl syntax).
6133 We also support \k{name} (.NET syntax). */
6134
6135 if (-c == ESC_k)
6136 {
6137 if ((ptr[1] != CHAR_LESS_THAN_SIGN &&
6138 ptr[1] != CHAR_APOSTROPHE && ptr[1] != CHAR_LEFT_CURLY_BRACKET))
6139 {
6140 *errorcodeptr = ERR69;
6141 break;
6142 }
6143 is_recurse = FALSE;
6144 terminator = (*(++ptr) == CHAR_LESS_THAN_SIGN)?
6145 CHAR_GREATER_THAN_SIGN : (*ptr == CHAR_APOSTROPHE)?
6146 CHAR_APOSTROPHE : CHAR_RIGHT_CURLY_BRACKET;
6147 goto NAMED_REF_OR_RECURSE;
6148 }
6149
6150 /* Back references are handled specially; must disable firstbyte if
6151 not set to cope with cases like (?=(\w+))\1: which would otherwise set
6152 ':' later. */
6153
6154 if (-c >= ESC_REF)
6155 {
6156 open_capitem *oc;
6157 recno = -c - ESC_REF;
6158
6159 HANDLE_REFERENCE: /* Come here from named backref handling */
6160 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
6161 previous = code;
6162 *code++ = ((options & PCRE_CASELESS) != 0)? OP_REFI : OP_REF;
6163 PUT2INC(code, 0, recno);
6164 cd->backref_map |= (recno < 32)? (1 << recno) : 1;
6165 if (recno > cd->top_backref) cd->top_backref = recno;
6166
6167 /* Check to see if this back reference is recursive, that it, it
6168 is inside the group that it references. A flag is set so that the
6169 group can be made atomic. */
6170
6171 for (oc = cd->open_caps; oc != NULL; oc = oc->next)
6172 {
6173 if (oc->number == recno)
6174 {
6175 oc->flag = TRUE;
6176 break;
6177 }
6178 }
6179 }
6180
6181 /* So are Unicode property matches, if supported. */
6182
6183 #ifdef SUPPORT_UCP
6184 else if (-c == ESC_P || -c == ESC_p)
6185 {
6186 BOOL negated;
6187 int pdata;
6188 int ptype = get_ucp(&ptr, &negated, &pdata, errorcodeptr);
6189 if (ptype < 0) goto FAILED;
6190 previous = code;
6191 *code++ = ((-c == ESC_p) != negated)? OP_PROP : OP_NOTPROP;
6192 *code++ = ptype;
6193 *code++ = pdata;
6194 }
6195 #else
6196
6197 /* If Unicode properties are not supported, \X, \P, and \p are not
6198 allowed. */
6199
6200 else if (-c == ESC_X || -c == ESC_P || -c == ESC_p)
6201 {
6202 *errorcodeptr = ERR45;
6203 goto FAILED;
6204 }
6205 #endif
6206
6207 /* For the rest (including \X when Unicode properties are supported), we
6208 can obtain the OP value by negating the escape value in the default
6209 situation when PCRE_UCP is not set. When it *is* set, we substitute
6210 Unicode property tests. */
6211
6212 else
6213 {
6214 #ifdef SUPPORT_UCP
6215 if (-c >= ESC_DU && -c <= ESC_wu)
6216 {
6217 nestptr = ptr + 1; /* Where to resume */
6218 ptr = substitutes[-c - ESC_DU] - 1; /* Just before substitute */
6219 }
6220 else
6221 #endif
6222 {
6223 previous = (-c > ESC_b && -c < ESC_Z)? code : NULL;
6224 *code++ = -c;
6225 }
6226 }
6227 continue;
6228 }
6229
6230 /* We have a data character whose value is in c. In UTF-8 mode it may have
6231 a value > 127. We set its representation in the length/buffer, and then
6232 handle it as a data character. */
6233
6234 #ifdef SUPPORT_UTF8
6235 if (utf8 && c > 127)
6236 mclength = _pcre_ord2utf8(c, mcbuffer);
6237 else
6238 #endif
6239
6240 {
6241 mcbuffer[0] = c;
6242 mclength = 1;
6243 }
6244 goto ONE_CHAR;
6245
6246
6247 /* ===================================================================*/
6248 /* Handle a literal character. It is guaranteed not to be whitespace or #
6249 when the extended flag is set. If we are in UTF-8 mode, it may be a
6250 multi-byte literal character. */
6251
6252 default:
6253 NORMAL_CHAR:
6254 mclength = 1;
6255 mcbuffer[0] = c;
6256
6257 #ifdef SUPPORT_UTF8
6258 if (utf8 && c >= 0xc0)
6259 {
6260 while ((ptr[1] & 0xc0) == 0x80)
6261 mcbuffer[mclength++] = *(++ptr);
6262 }
6263 #endif
6264
6265 /* At this point we have the character's bytes in mcbuffer, and the length
6266 in mclength. When not in UTF-8 mode, the length is always 1. */
6267
6268 ONE_CHAR:
6269 previous = code;
6270 *code++ = ((options & PCRE_CASELESS) != 0)? OP_CHARI : OP_CHAR;
6271 for (c = 0; c < mclength; c++) *code++ = mcbuffer[c];
6272
6273 /* Remember if \r or \n were seen */
6274
6275 if (mcbuffer[0] == CHAR_CR || mcbuffer[0] == CHAR_NL)
6276 cd->external_flags |= PCRE_HASCRORLF;
6277
6278 /* Set the first and required bytes appropriately. If no previous first
6279 byte, set it from this character, but revert to none on a zero repeat.
6280 Otherwise, leave the firstbyte value alone, and don't change it on a zero
6281 repeat. */
6282
6283 if (firstbyte == REQ_UNSET)
6284 {
6285 zerofirstbyte = REQ_NONE;
6286 zeroreqbyte = reqbyte;
6287
6288 /* If the character is more than one byte long, we can set firstbyte
6289 only if it is not to be matched caselessly. */
6290
6291 if (mclength == 1 || req_caseopt == 0)
6292 {
6293 firstbyte = mcbuffer[0] | req_caseopt;
6294 if (mclength != 1) reqbyte = code[-1] | cd->req_varyopt;
6295 }
6296 else firstbyte = reqbyte = REQ_NONE;
6297 }
6298
6299 /* firstbyte was previously set; we can set reqbyte only the length is
6300 1 or the matching is caseful. */
6301
6302 else
6303 {
6304 zerofirstbyte = firstbyte;
6305 zeroreqbyte = reqbyte;
6306 if (mclength == 1 || req_caseopt == 0)
6307 reqbyte = code[-1] | req_caseopt | cd->req_varyopt;
6308 }
6309
6310 break; /* End of literal character handling */
6311 }
6312 } /* end of big loop */
6313
6314
6315 /* Control never reaches here by falling through, only by a goto for all the
6316 error states. Pass back the position in the pattern so that it can be displayed
6317 to the user for diagnosing the error. */
6318
6319 FAILED:
6320 *ptrptr = ptr;
6321 return FALSE;
6322 }
6323
6324
6325
6326
6327 /*************************************************
6328 * Compile sequence of alternatives *
6329 *************************************************/
6330
6331 /* On entry, ptr is pointing past the bracket character, but on return it
6332 points to the closing bracket, or vertical bar, or end of string. The code
6333 variable is pointing at the byte into which the BRA operator has been stored.
6334 This function is used during the pre-compile phase when we are trying to find
6335 out the amount of memory needed, as well as during the real compile phase. The
6336 value of lengthptr distinguishes the two phases.
6337
6338 Arguments:
6339 options option bits, including any changes for this subpattern
6340 codeptr -> the address of the current code pointer
6341 ptrptr -> the address of the current pattern pointer
6342 errorcodeptr -> pointer to error code variable
6343 lookbehind TRUE if this is a lookbehind assertion
6344 reset_bracount TRUE to reset the count for each branch
6345 skipbytes skip this many bytes at start (for brackets and OP_COND)
6346 firstbyteptr place to put the first required character, or a negative number
6347 reqbyteptr place to put the last required character, or a negative number
6348 bcptr pointer to the chain of currently open branches
6349 cd points to the data block with tables pointers etc.
6350 lengthptr NULL during the real compile phase
6351 points to length accumulator during pre-compile phase
6352
6353 Returns: TRUE on success
6354 */
6355
6356 static BOOL
6357 compile_regex(int options, uschar **codeptr, const uschar **ptrptr,
6358 int *errorcodeptr, BOOL lookbehind, BOOL reset_bracount, int skipbytes,
6359 int *firstbyteptr, int *reqbyteptr, branch_chain *bcptr, compile_data *cd,
6360 int *lengthptr)
6361 {
6362 const uschar *ptr = *ptrptr;
6363 uschar *code = *codeptr;
6364 uschar *last_branch = code;
6365 uschar *start_bracket = code;
6366 uschar *reverse_count = NULL;
6367 open_capitem capitem;
6368 int capnumber = 0;
6369 int firstbyte, reqbyte;
6370 int branchfirstbyte, branchreqbyte;
6371 int length;
6372 int orig_bracount;
6373 int max_bracount;
6374 branch_chain bc;
6375
6376 bc.outer = bcptr;
6377 bc.current_branch = code;
6378
6379 firstbyte = reqbyte = REQ_UNSET;
6380
6381 /* Accumulate the length for use in the pre-compile phase. Start with the
6382 length of the BRA and KET and any extra bytes that are required at the
6383 beginning. We accumulate in a local variable to save frequent testing of
6384 lenthptr for NULL. We cannot do this by looking at the value of code at the
6385 start and end of each alternative, because compiled items are discarded during
6386 the pre-compile phase so that the work space is not exceeded. */
6387
6388 length = 2 + 2*LINK_SIZE + skipbytes;
6389
6390 /* WARNING: If the above line is changed for any reason, you must also change
6391 the code that abstracts option settings at the start of the pattern and makes
6392 them global. It tests the value of length for (2 + 2*LINK_SIZE) in the
6393 pre-compile phase to find out whether anything has yet been compiled or not. */
6394
6395 /* If this is a capturing subpattern, add to the chain of open capturing items
6396 so that we can detect them if (*ACCEPT) is encountered. This is also used to
6397 detect groups that contain recursive back references to themselves. Note that
6398 only OP_CBRA need be tested here; changing this opcode to one of its variants,
6399 e.g. OP_SCBRAPOS, happens later, after the group has been compiled. */
6400
6401 if (*code == OP_CBRA)
6402 {
6403 capnumber = GET2(code, 1 + LINK_SIZE);
6404 capitem.number = capnumber;
6405 capitem.next = cd->open_caps;
6406 capitem.flag = FALSE;
6407 cd->open_caps = &capitem;
6408 }
6409
6410 /* Offset is set zero to mark that this bracket is still open */
6411
6412 PUT(code, 1, 0);
6413 code += 1 + LINK_SIZE + skipbytes;
6414
6415 /* Loop for each alternative branch */
6416
6417 orig_bracount = max_bracount = cd->bracount;
6418 for (;;)
6419 {
6420 /* For a (?| group, reset the capturing bracket count so that each branch
6421 uses the same numbers. */
6422
6423 if (reset_bracount) cd->bracount = orig_bracount;
6424
6425 /* Set up dummy OP_REVERSE if lookbehind assertion */
6426
6427 if (lookbehind)
6428 {
6429 *code++ = OP_REVERSE;
6430 reverse_count = code;
6431 PUTINC(code, 0, 0);
6432 length += 1 + LINK_SIZE;
6433 }
6434
6435 /* Now compile the branch; in the pre-compile phase its length gets added
6436 into the length. */
6437
6438 if (!compile_branch(&options, &code, &ptr, errorcodeptr, &branchfirstbyte,
6439 &branchreqbyte, &bc, cd, (lengthptr == NULL)? NULL : &length))
6440 {
6441 *ptrptr = ptr;
6442 return FALSE;
6443 }
6444
6445 /* Keep the highest bracket count in case (?| was used and some branch
6446 has fewer than the rest. */
6447
6448 if (cd->bracount > max_bracount) max_bracount = cd->bracount;
6449
6450 /* In the real compile phase, there is some post-processing to be done. */
6451
6452 if (lengthptr == NULL)
6453 {
6454 /* If this is the first branch, the firstbyte and reqbyte values for the
6455 branch become the values for the regex. */
6456
6457 if (*last_branch != OP_ALT)
6458 {
6459 firstbyte = branchfirstbyte;
6460 reqbyte = branchreqbyte;
6461 }
6462
6463 /* If this is not the first branch, the first char and reqbyte have to
6464 match the values from all the previous branches, except that if the
6465 previous value for reqbyte didn't have REQ_VARY set, it can still match,
6466 and we set REQ_VARY for the regex. */
6467
6468 else
6469 {
6470 /* If we previously had a firstbyte, but it doesn't match the new branch,
6471 we have to abandon the firstbyte for the regex, but if there was
6472 previously no reqbyte, it takes on the value of the old firstbyte. */
6473
6474 if (firstbyte >= 0 && firstbyte != branchfirstbyte)
6475 {
6476 if (reqbyte < 0) reqbyte = firstbyte;
6477 firstbyte = REQ_NONE;
6478 }
6479
6480 /* If we (now or from before) have no firstbyte, a firstbyte from the
6481 branch becomes a reqbyte if there isn't a branch reqbyte. */
6482
6483 if (firstbyte < 0 && branchfirstbyte >= 0 && branchreqbyte < 0)
6484 branchreqbyte = branchfirstbyte;
6485
6486 /* Now ensure that the reqbytes match */
6487
6488 if ((reqbyte & ~REQ_VARY) != (branchreqbyte & ~REQ_VARY))
6489 reqbyte = REQ_NONE;
6490 else reqbyte |= branchreqbyte; /* To "or" REQ_VARY */
6491 }
6492
6493 /* If lookbehind, check that this branch matches a fixed-length string, and
6494 put the length into the OP_REVERSE item. Temporarily mark the end of the
6495 branch with OP_END. If the branch contains OP_RECURSE, the result is -3
6496 because there may be forward references that we can't check here. Set a
6497 flag to cause another lookbehind check at the end. Why not do it all at the
6498 end? Because common, erroneous checks are picked up here and the offset of
6499 the problem can be shown. */
6500
6501 if (lookbehind)
6502 {
6503 int fixed_length;
6504 *code = OP_END;
6505 fixed_length = find_fixedlength(last_branch, (options & PCRE_UTF8) != 0,
6506 FALSE, cd);
6507 DPRINTF(("fixed length = %d\n", fixed_length));
6508 if (fixed_length == -3)
6509 {
6510 cd->check_lookbehind = TRUE;
6511 }
6512 else if (fixed_length < 0)
6513 {
6514 *errorcodeptr = (fixed_length == -2)? ERR36 : ERR25;
6515 *ptrptr = ptr;
6516 return FALSE;
6517 }
6518 else { PUT(reverse_count, 0, fixed_length); }
6519 }
6520 }
6521
6522 /* Reached end of expression, either ')' or end of pattern. In the real
6523 compile phase, go back through the alternative branches and reverse the chain
6524 of offsets, with the field in the BRA item now becoming an offset to the
6525 first alternative. If there are no alternatives, it points to the end of the
6526 group. The length in the terminating ket is always the length of the whole
6527 bracketed item. Return leaving the pointer at the terminating char. */
6528
6529 if (*ptr != CHAR_VERTICAL_LINE)
6530 {
6531 if (lengthptr == NULL)
6532 {
6533 int branch_length = (int)(code - last_branch);
6534 do
6535 {
6536 int prev_length = GET(last_branch, 1);
6537 PUT(last_branch, 1, branch_length);
6538 branch_length = prev_length;
6539 last_branch -= branch_length;
6540 }
6541 while (branch_length > 0);
6542 }
6543
6544 /* Fill in the ket */
6545
6546 *code = OP_KET;
6547 PUT(code, 1, (int)(code - start_bracket));
6548 code += 1 + LINK_SIZE;
6549
6550 /* If it was a capturing subpattern, check to see if it contained any
6551 recursive back references. If so, we must wrap it in atomic brackets.
6552 In any event, remove the block from the chain. */
6553
6554 if (capnumber > 0)
6555 {
6556 if (cd->open_caps->flag)
6557 {
6558 memmove(start_bracket + 1 + LINK_SIZE, start_bracket,
6559 code - start_bracket);
6560 *start_bracket = OP_ONCE;
6561 code += 1 + LINK_SIZE;
6562 PUT(start_bracket, 1, (int)(code - start_bracket));
6563 *code = OP_KET;
6564 PUT(code, 1, (int)(code - start_bracket));
6565 code += 1 + LINK_SIZE;
6566 length += 2 + 2*LINK_SIZE;
6567 }
6568 cd->open_caps = cd->open_caps->next;
6569 }
6570
6571 /* Retain the highest bracket number, in case resetting was used. */
6572
6573 cd->bracount = max_bracount;
6574
6575 /* Set values to pass back */
6576
6577 *codeptr = code;
6578 *ptrptr = ptr;
6579 *firstbyteptr = firstbyte;
6580 *reqbyteptr = reqbyte;
6581 if (lengthptr != NULL)
6582 {
6583 if (OFLOW_MAX - *lengthptr < length)
6584 {
6585 *errorcodeptr = ERR20;
6586 return FALSE;
6587 }
6588 *lengthptr += length;
6589 }
6590 return TRUE;
6591 }
6592
6593 /* Another branch follows. In the pre-compile phase, we can move the code
6594 pointer back to where it was for the start of the first branch. (That is,
6595 pretend that each branch is the only one.)
6596
6597 In the real compile phase, insert an ALT node. Its length field points back
6598 to the previous branch while the bracket remains open. At the end the chain
6599 is reversed. It's done like this so that the start of the bracket has a
6600 zero offset until it is closed, making it possible to detect recursion. */
6601
6602 if (lengthptr != NULL)
6603 {
6604 code = *codeptr + 1 + LINK_SIZE + skipbytes;
6605 length += 1 + LINK_SIZE;
6606 }
6607 else
6608 {
6609 *code = OP_ALT;
6610 PUT(code, 1, (int)(code - last_branch));
6611 bc.current_branch = last_branch = code;
6612 code += 1 + LINK_SIZE;
6613 }
6614
6615 ptr++;
6616 }
6617 /* Control never reaches here */
6618 }
6619
6620
6621
6622
6623 /*************************************************
6624 * Check for anchored expression *
6625 *************************************************/
6626
6627 /* Try to find out if this is an anchored regular expression. Consider each
6628 alternative branch. If they all start with OP_SOD or OP_CIRC, or with a bracket
6629 all of whose alternatives start with OP_SOD or OP_CIRC (recurse ad lib), then
6630 it's anchored. However, if this is a multiline pattern, then only OP_SOD will
6631 be found, because ^ generates OP_CIRCM in that mode.
6632
6633 We can also consider a regex to be anchored if OP_SOM starts all its branches.
6634 This is the code for \G, which means "match at start of match position, taking
6635 into account the match offset".
6636
6637 A branch is also implicitly anchored if it starts with .* and DOTALL is set,
6638 because that will try the rest of the pattern at all possible matching points,
6639 so there is no point trying again.... er ....
6640
6641 .... except when the .* appears inside capturing parentheses, and there is a
6642 subsequent back reference to those parentheses. We haven't enough information
6643 to catch that case precisely.
6644
6645 At first, the best we could do was to detect when .* was in capturing brackets
6646 and the highest back reference was greater than or equal to that level.
6647 However, by keeping a bitmap of the first 31 back references, we can catch some
6648 of the more common cases more precisely.
6649
6650 Arguments:
6651 code points to start of expression (the bracket)
6652 bracket_map a bitmap of which brackets we are inside while testing; this
6653 handles up to substring 31; after that we just have to take
6654 the less precise approach
6655 backref_map the back reference bitmap
6656
6657 Returns: TRUE or FALSE
6658 */
6659
6660 static BOOL