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

Contents of /code/trunk/pcre_compile.c

Parent Directory Parent Directory | Revision Log Revision Log


Revision 602 - (show annotations) (download)
Wed May 25 08:29:03 2011 UTC (3 years, 3 months ago) by ph10
File MIME type: text/plain
File size: 237764 byte(s)
Remove OP_OPT by handling /i and /m entirely at compile time. Fixes bug with 
patterns like /(?i:([^b]))(?1)/, where the /i option was mishandled.

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