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

Contents of /code/trunk/pcre_compile.c

Parent Directory Parent Directory | Revision Log Revision Log


Revision 745 - (show annotations)
Mon Nov 14 11:41:03 2011 UTC (7 years, 10 months ago) by ph10
File MIME type: text/plain
File size: 245998 byte(s)
Error occurred while calculating annotation data.
Small tidies, and documentation update for JavaScript \x, \u, \U support.
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" /** DEAD **/
397 "inconsistent NEWLINE options\0"
398 "\\g is not followed by a braced, angle-bracketed, or quoted name/number or by a plain number\0"
399 "a numbered reference must not be zero\0"
400 "an argument is not allowed for (*ACCEPT), (*FAIL), or (*COMMIT)\0"
401 /* 60 */
402 "(*VERB) not recognized\0"
403 "number is too big\0"
404 "subpattern name expected\0"
405 "digit expected after (?+\0"
406 "] is an invalid data character in JavaScript compatibility mode\0"
407 /* 65 */
408 "different names for subpatterns of the same number are not allowed\0"
409 "(*MARK) must have an argument\0"
410 "this version of PCRE is not compiled with PCRE_UCP support\0"
411 "\\c must be followed by an ASCII character\0"
412 "\\k is not followed by a braced, angle-bracketed, or quoted name\0"
413 ;
414
415 /* Table to identify digits and hex digits. This is used when compiling
416 patterns. Note that the tables in chartables are dependent on the locale, and
417 may mark arbitrary characters as digits - but the PCRE compiling code expects
418 to handle only 0-9, a-z, and A-Z as digits when compiling. That is why we have
419 a private table here. It costs 256 bytes, but it is a lot faster than doing
420 character value tests (at least in some simple cases I timed), and in some
421 applications one wants PCRE to compile efficiently as well as match
422 efficiently.
423
424 For convenience, we use the same bit definitions as in chartables:
425
426 0x04 decimal digit
427 0x08 hexadecimal digit
428
429 Then we can use ctype_digit and ctype_xdigit in the code. */
430
431 #ifndef EBCDIC
432
433 /* This is the "normal" case, for ASCII systems, and EBCDIC systems running in
434 UTF-8 mode. */
435
436 static const unsigned char digitab[] =
437 {
438 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 0- 7 */
439 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 8- 15 */
440 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 16- 23 */
441 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 24- 31 */
442 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - ' */
443 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* ( - / */
444 0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c, /* 0 - 7 */
445 0x0c,0x0c,0x00,0x00,0x00,0x00,0x00,0x00, /* 8 - ? */
446 0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* @ - G */
447 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* H - O */
448 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* P - W */
449 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* X - _ */
450 0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* ` - g */
451 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* h - o */
452 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* p - w */
453 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* x -127 */
454 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 128-135 */
455 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 136-143 */
456 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 144-151 */
457 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 152-159 */
458 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 160-167 */
459 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 168-175 */
460 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 176-183 */
461 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 184-191 */
462 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 192-199 */
463 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 200-207 */
464 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 208-215 */
465 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 216-223 */
466 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 224-231 */
467 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 232-239 */
468 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 240-247 */
469 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00};/* 248-255 */
470
471 #else
472
473 /* This is the "abnormal" case, for EBCDIC systems not running in UTF-8 mode. */
474
475 static const unsigned char digitab[] =
476 {
477 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 0- 7 0 */
478 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 8- 15 */
479 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 16- 23 10 */
480 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 24- 31 */
481 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 32- 39 20 */
482 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 40- 47 */
483 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 48- 55 30 */
484 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 56- 63 */
485 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - 71 40 */
486 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 72- | */
487 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* & - 87 50 */
488 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 88- 95 */
489 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - -103 60 */
490 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 104- ? */
491 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 112-119 70 */
492 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 120- " */
493 0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* 128- g 80 */
494 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* h -143 */
495 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 144- p 90 */
496 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* q -159 */
497 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 160- x A0 */
498 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* y -175 */
499 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* ^ -183 B0 */
500 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 184-191 */
501 0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* { - G C0 */
502 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* H -207 */
503 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* } - P D0 */
504 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* Q -223 */
505 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* \ - X E0 */
506 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* Y -239 */
507 0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c, /* 0 - 7 F0 */
508 0x0c,0x0c,0x00,0x00,0x00,0x00,0x00,0x00};/* 8 -255 */
509
510 static const unsigned char ebcdic_chartab[] = { /* chartable partial dup */
511 0x80,0x00,0x00,0x00,0x00,0x01,0x00,0x00, /* 0- 7 */
512 0x00,0x00,0x00,0x00,0x01,0x01,0x00,0x00, /* 8- 15 */
513 0x00,0x00,0x00,0x00,0x00,0x01,0x00,0x00, /* 16- 23 */
514 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 24- 31 */
515 0x00,0x00,0x00,0x00,0x00,0x01,0x00,0x00, /* 32- 39 */
516 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 40- 47 */
517 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 48- 55 */
518 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 56- 63 */
519 0x01,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - 71 */
520 0x00,0x00,0x00,0x80,0x00,0x80,0x80,0x80, /* 72- | */
521 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* & - 87 */
522 0x00,0x00,0x00,0x80,0x80,0x80,0x00,0x00, /* 88- 95 */
523 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - -103 */
524 0x00,0x00,0x00,0x00,0x00,0x10,0x00,0x80, /* 104- ? */
525 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 112-119 */
526 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 120- " */
527 0x00,0x1a,0x1a,0x1a,0x1a,0x1a,0x1a,0x12, /* 128- g */
528 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* h -143 */
529 0x00,0x12,0x12,0x12,0x12,0x12,0x12,0x12, /* 144- p */
530 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* q -159 */
531 0x00,0x00,0x12,0x12,0x12,0x12,0x12,0x12, /* 160- x */
532 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* y -175 */
533 0x80,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* ^ -183 */
534 0x00,0x00,0x80,0x00,0x00,0x00,0x00,0x00, /* 184-191 */
535 0x80,0x1a,0x1a,0x1a,0x1a,0x1a,0x1a,0x12, /* { - G */
536 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* H -207 */
537 0x00,0x12,0x12,0x12,0x12,0x12,0x12,0x12, /* } - P */
538 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* Q -223 */
539 0x00,0x00,0x12,0x12,0x12,0x12,0x12,0x12, /* \ - X */
540 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* Y -239 */
541 0x1c,0x1c,0x1c,0x1c,0x1c,0x1c,0x1c,0x1c, /* 0 - 7 */
542 0x1c,0x1c,0x00,0x00,0x00,0x00,0x00,0x00};/* 8 -255 */
543 #endif
544
545
546 /* Definition to allow mutual recursion */
547
548 static BOOL
549 compile_regex(int, uschar **, const uschar **, int *, BOOL, BOOL, int, int,
550 int *, int *, branch_chain *, compile_data *, int *);
551
552
553
554 /*************************************************
555 * Find an error text *
556 *************************************************/
557
558 /* The error texts are now all in one long string, to save on relocations. As
559 some of the text is of unknown length, we can't use a table of offsets.
560 Instead, just count through the strings. This is not a performance issue
561 because it happens only when there has been a compilation error.
562
563 Argument: the error number
564 Returns: pointer to the error string
565 */
566
567 static const char *
568 find_error_text(int n)
569 {
570 const char *s = error_texts;
571 for (; n > 0; n--)
572 {
573 while (*s++ != 0) {};
574 if (*s == 0) return "Error text not found (please report)";
575 }
576 return s;
577 }
578
579
580 /*************************************************
581 * Check for counted repeat *
582 *************************************************/
583
584 /* This function is called when a '{' is encountered in a place where it might
585 start a quantifier. It looks ahead to see if it really is a quantifier or not.
586 It is only a quantifier if it is one of the forms {ddd} {ddd,} or {ddd,ddd}
587 where the ddds are digits.
588
589 Arguments:
590 p pointer to the first char after '{'
591
592 Returns: TRUE or FALSE
593 */
594
595 static BOOL
596 is_counted_repeat(const uschar *p)
597 {
598 if ((digitab[*p++] & ctype_digit) == 0) return FALSE;
599 while ((digitab[*p] & ctype_digit) != 0) p++;
600 if (*p == CHAR_RIGHT_CURLY_BRACKET) return TRUE;
601
602 if (*p++ != CHAR_COMMA) return FALSE;
603 if (*p == CHAR_RIGHT_CURLY_BRACKET) return TRUE;
604
605 if ((digitab[*p++] & ctype_digit) == 0) return FALSE;
606 while ((digitab[*p] & ctype_digit) != 0) p++;
607
608 return (*p == CHAR_RIGHT_CURLY_BRACKET);
609 }
610
611
612
613 /*************************************************
614 * Handle escapes *
615 *************************************************/
616
617 /* This function is called when a \ has been encountered. It either returns a
618 positive value for a simple escape such as \n, or a negative value which
619 encodes one of the more complicated things such as \d. A backreference to group
620 n is returned as -(ESC_REF + n); ESC_REF is the highest ESC_xxx macro. When
621 UTF-8 is enabled, a positive value greater than 255 may be returned. On entry,
622 ptr is pointing at the \. On exit, it is on the final character of the escape
623 sequence.
624
625 Arguments:
626 ptrptr points to the pattern position pointer
627 errorcodeptr points to the errorcode variable
628 bracount number of previous extracting brackets
629 options the options bits
630 isclass TRUE if inside a character class
631
632 Returns: zero or positive => a data character
633 negative => a special escape sequence
634 on error, errorcodeptr is set
635 */
636
637 static int
638 check_escape(const uschar **ptrptr, int *errorcodeptr, int bracount,
639 int options, BOOL isclass)
640 {
641 BOOL utf8 = (options & PCRE_UTF8) != 0;
642 const uschar *ptr = *ptrptr + 1;
643 int c, i;
644
645 GETCHARINCTEST(c, ptr); /* Get character value, increment pointer */
646 ptr--; /* Set pointer back to the last byte */
647
648 /* If backslash is at the end of the pattern, it's an error. */
649
650 if (c == 0) *errorcodeptr = ERR1;
651
652 /* Non-alphanumerics are literals. For digits or letters, do an initial lookup
653 in a table. A non-zero result is something that can be returned immediately.
654 Otherwise further processing may be required. */
655
656 #ifndef EBCDIC /* ASCII/UTF-8 coding */
657 else if (c < CHAR_0 || c > CHAR_z) {} /* Not alphanumeric */
658 else if ((i = escapes[c - CHAR_0]) != 0) c = i;
659
660 #else /* EBCDIC coding */
661 else if (c < 'a' || (ebcdic_chartab[c] & 0x0E) == 0) {} /* Not alphanumeric */
662 else if ((i = escapes[c - 0x48]) != 0) c = i;
663 #endif
664
665 /* Escapes that need further processing, or are illegal. */
666
667 else
668 {
669 const uschar *oldptr;
670 BOOL braced, negated;
671
672 switch (c)
673 {
674 /* A number of Perl escapes are not handled by PCRE. We give an explicit
675 error. */
676
677 case CHAR_l:
678 case CHAR_L:
679 *errorcodeptr = ERR37;
680 break;
681
682 case CHAR_u:
683 if ((options & PCRE_JAVASCRIPT_COMPAT) != 0)
684 {
685 /* In JavaScript, \u must be followed by four hexadecimal numbers.
686 Otherwise it is a lowercase u letter. */
687 if ((digitab[ptr[1]] & ctype_xdigit) != 0 && (digitab[ptr[2]] & ctype_xdigit) != 0
688 && (digitab[ptr[3]] & ctype_xdigit) != 0 && (digitab[ptr[4]] & ctype_xdigit) != 0)
689 {
690 c = 0;
691 for (i = 0; i < 4; ++i)
692 {
693 register int cc = *(++ptr);
694 #ifndef EBCDIC /* ASCII/UTF-8 coding */
695 if (cc >= CHAR_a) cc -= 32; /* Convert to upper case */
696 c = (c << 4) + cc - ((cc < CHAR_A)? CHAR_0 : (CHAR_A - 10));
697 #else /* EBCDIC coding */
698 if (cc >= CHAR_a && cc <= CHAR_z) cc += 64; /* Convert to upper case */
699 c = (c << 4) + cc - ((cc >= CHAR_0)? CHAR_0 : (CHAR_A - 10));
700 #endif
701 }
702 }
703 }
704 else
705 *errorcodeptr = ERR37;
706 break;
707
708 case CHAR_U:
709 /* In JavaScript, \U is an uppercase U letter. */
710 if ((options & PCRE_JAVASCRIPT_COMPAT) == 0) *errorcodeptr = ERR37;
711 break;
712
713 /* In a character class, \g is just a literal "g". Outside a character
714 class, \g must be followed by one of a number of specific things:
715
716 (1) A number, either plain or braced. If positive, it is an absolute
717 backreference. If negative, it is a relative backreference. This is a Perl
718 5.10 feature.
719
720 (2) Perl 5.10 also supports \g{name} as a reference to a named group. This
721 is part of Perl's movement towards a unified syntax for back references. As
722 this is synonymous with \k{name}, we fudge it up by pretending it really
723 was \k.
724
725 (3) For Oniguruma compatibility we also support \g followed by a name or a
726 number either in angle brackets or in single quotes. However, these are
727 (possibly recursive) subroutine calls, _not_ backreferences. Just return
728 the -ESC_g code (cf \k). */
729
730 case CHAR_g:
731 if (isclass) break;
732 if (ptr[1] == CHAR_LESS_THAN_SIGN || ptr[1] == CHAR_APOSTROPHE)
733 {
734 c = -ESC_g;
735 break;
736 }
737
738 /* Handle the Perl-compatible cases */
739
740 if (ptr[1] == CHAR_LEFT_CURLY_BRACKET)
741 {
742 const uschar *p;
743 for (p = ptr+2; *p != 0 && *p != CHAR_RIGHT_CURLY_BRACKET; p++)
744 if (*p != CHAR_MINUS && (digitab[*p] & ctype_digit) == 0) break;
745 if (*p != 0 && *p != CHAR_RIGHT_CURLY_BRACKET)
746 {
747 c = -ESC_k;
748 break;
749 }
750 braced = TRUE;
751 ptr++;
752 }
753 else braced = FALSE;
754
755 if (ptr[1] == CHAR_MINUS)
756 {
757 negated = TRUE;
758 ptr++;
759 }
760 else negated = FALSE;
761
762 c = 0;
763 while ((digitab[ptr[1]] & ctype_digit) != 0)
764 c = c * 10 + *(++ptr) - CHAR_0;
765
766 if (c < 0) /* Integer overflow */
767 {
768 *errorcodeptr = ERR61;
769 break;
770 }
771
772 if (braced && *(++ptr) != CHAR_RIGHT_CURLY_BRACKET)
773 {
774 *errorcodeptr = ERR57;
775 break;
776 }
777
778 if (c == 0)
779 {
780 *errorcodeptr = ERR58;
781 break;
782 }
783
784 if (negated)
785 {
786 if (c > bracount)
787 {
788 *errorcodeptr = ERR15;
789 break;
790 }
791 c = bracount - (c - 1);
792 }
793
794 c = -(ESC_REF + c);
795 break;
796
797 /* The handling of escape sequences consisting of a string of digits
798 starting with one that is not zero is not straightforward. By experiment,
799 the way Perl works seems to be as follows:
800
801 Outside a character class, the digits are read as a decimal number. If the
802 number is less than 10, or if there are that many previous extracting
803 left brackets, then it is a back reference. Otherwise, up to three octal
804 digits are read to form an escaped byte. Thus \123 is likely to be octal
805 123 (cf \0123, which is octal 012 followed by the literal 3). If the octal
806 value is greater than 377, the least significant 8 bits are taken. Inside a
807 character class, \ followed by a digit is always an octal number. */
808
809 case CHAR_1: case CHAR_2: case CHAR_3: case CHAR_4: case CHAR_5:
810 case CHAR_6: case CHAR_7: case CHAR_8: case CHAR_9:
811
812 if (!isclass)
813 {
814 oldptr = ptr;
815 c -= CHAR_0;
816 while ((digitab[ptr[1]] & ctype_digit) != 0)
817 c = c * 10 + *(++ptr) - CHAR_0;
818 if (c < 0) /* Integer overflow */
819 {
820 *errorcodeptr = ERR61;
821 break;
822 }
823 if (c < 10 || c <= bracount)
824 {
825 c = -(ESC_REF + c);
826 break;
827 }
828 ptr = oldptr; /* Put the pointer back and fall through */
829 }
830
831 /* Handle an octal number following \. If the first digit is 8 or 9, Perl
832 generates a binary zero byte and treats the digit as a following literal.
833 Thus we have to pull back the pointer by one. */
834
835 if ((c = *ptr) >= CHAR_8)
836 {
837 ptr--;
838 c = 0;
839 break;
840 }
841
842 /* \0 always starts an octal number, but we may drop through to here with a
843 larger first octal digit. The original code used just to take the least
844 significant 8 bits of octal numbers (I think this is what early Perls used
845 to do). Nowadays we allow for larger numbers in UTF-8 mode, but no more
846 than 3 octal digits. */
847
848 case CHAR_0:
849 c -= CHAR_0;
850 while(i++ < 2 && ptr[1] >= CHAR_0 && ptr[1] <= CHAR_7)
851 c = c * 8 + *(++ptr) - CHAR_0;
852 if (!utf8 && c > 255) *errorcodeptr = ERR51;
853 break;
854
855 /* \x is complicated. \x{ddd} is a character number which can be greater
856 than 0xff in utf8 mode, but only if the ddd are hex digits. If not, { is
857 treated as a data character. */
858
859 case CHAR_x:
860 if ((options & PCRE_JAVASCRIPT_COMPAT) != 0)
861 {
862 /* In JavaScript, \x must be followed by two hexadecimal numbers.
863 Otherwise it is a lowercase x letter. */
864 if ((digitab[ptr[1]] & ctype_xdigit) != 0 && (digitab[ptr[2]] & ctype_xdigit) != 0)
865 {
866 c = 0;
867 for (i = 0; i < 2; ++i)
868 {
869 register int cc = *(++ptr);
870 #ifndef EBCDIC /* ASCII/UTF-8 coding */
871 if (cc >= CHAR_a) cc -= 32; /* Convert to upper case */
872 c = (c << 4) + cc - ((cc < CHAR_A)? CHAR_0 : (CHAR_A - 10));
873 #else /* EBCDIC coding */
874 if (cc >= CHAR_a && cc <= CHAR_z) cc += 64; /* Convert to upper case */
875 c = (c << 4) + cc - ((cc >= CHAR_0)? CHAR_0 : (CHAR_A - 10));
876 #endif
877 }
878 }
879 break;
880 }
881
882 if (ptr[1] == CHAR_LEFT_CURLY_BRACKET)
883 {
884 const uschar *pt = ptr + 2;
885 int count = 0;
886
887 c = 0;
888 while ((digitab[*pt] & ctype_xdigit) != 0)
889 {
890 register int cc = *pt++;
891 if (c == 0 && cc == CHAR_0) continue; /* Leading zeroes */
892 count++;
893
894 #ifndef EBCDIC /* ASCII/UTF-8 coding */
895 if (cc >= CHAR_a) cc -= 32; /* Convert to upper case */
896 c = (c << 4) + cc - ((cc < CHAR_A)? CHAR_0 : (CHAR_A - 10));
897 #else /* EBCDIC coding */
898 if (cc >= CHAR_a && cc <= CHAR_z) cc += 64; /* Convert to upper case */
899 c = (c << 4) + cc - ((cc >= CHAR_0)? CHAR_0 : (CHAR_A - 10));
900 #endif
901 }
902
903 if (*pt == CHAR_RIGHT_CURLY_BRACKET)
904 {
905 if (c < 0 || count > (utf8? 8 : 2)) *errorcodeptr = ERR34;
906 ptr = pt;
907 break;
908 }
909
910 /* If the sequence of hex digits does not end with '}', then we don't
911 recognize this construct; fall through to the normal \x handling. */
912 }
913
914 /* Read just a single-byte hex-defined char */
915
916 c = 0;
917 while (i++ < 2 && (digitab[ptr[1]] & ctype_xdigit) != 0)
918 {
919 int cc; /* Some compilers don't like */
920 cc = *(++ptr); /* ++ in initializers */
921 #ifndef EBCDIC /* ASCII/UTF-8 coding */
922 if (cc >= CHAR_a) cc -= 32; /* Convert to upper case */
923 c = c * 16 + cc - ((cc < CHAR_A)? CHAR_0 : (CHAR_A - 10));
924 #else /* EBCDIC coding */
925 if (cc <= CHAR_z) cc += 64; /* Convert to upper case */
926 c = c * 16 + cc - ((cc >= CHAR_0)? CHAR_0 : (CHAR_A - 10));
927 #endif
928 }
929 break;
930
931 /* For \c, a following letter is upper-cased; then the 0x40 bit is flipped.
932 An error is given if the byte following \c is not an ASCII character. This
933 coding is ASCII-specific, but then the whole concept of \cx is
934 ASCII-specific. (However, an EBCDIC equivalent has now been added.) */
935
936 case CHAR_c:
937 c = *(++ptr);
938 if (c == 0)
939 {
940 *errorcodeptr = ERR2;
941 break;
942 }
943 #ifndef EBCDIC /* ASCII/UTF-8 coding */
944 if (c > 127) /* Excludes all non-ASCII in either mode */
945 {
946 *errorcodeptr = ERR68;
947 break;
948 }
949 if (c >= CHAR_a && c <= CHAR_z) c -= 32;
950 c ^= 0x40;
951 #else /* EBCDIC coding */
952 if (c >= CHAR_a && c <= CHAR_z) c += 64;
953 c ^= 0xC0;
954 #endif
955 break;
956
957 /* PCRE_EXTRA enables extensions to Perl in the matter of escapes. Any
958 other alphanumeric following \ is an error if PCRE_EXTRA was set;
959 otherwise, for Perl compatibility, it is a literal. This code looks a bit
960 odd, but there used to be some cases other than the default, and there may
961 be again in future, so I haven't "optimized" it. */
962
963 default:
964 if ((options & PCRE_EXTRA) != 0) switch(c)
965 {
966 default:
967 *errorcodeptr = ERR3;
968 break;
969 }
970 break;
971 }
972 }
973
974 /* Perl supports \N{name} for character names, as well as plain \N for "not
975 newline". PCRE does not support \N{name}. However, it does support
976 quantification such as \N{2,3}. */
977
978 if (c == -ESC_N && ptr[1] == CHAR_LEFT_CURLY_BRACKET &&
979 !is_counted_repeat(ptr+2))
980 *errorcodeptr = ERR37;
981
982 /* If PCRE_UCP is set, we change the values for \d etc. */
983
984 if ((options & PCRE_UCP) != 0 && c <= -ESC_D && c >= -ESC_w)
985 c -= (ESC_DU - ESC_D);
986
987 /* Set the pointer to the final character before returning. */
988
989 *ptrptr = ptr;
990 return c;
991 }
992
993
994
995 #ifdef SUPPORT_UCP
996 /*************************************************
997 * Handle \P and \p *
998 *************************************************/
999
1000 /* This function is called after \P or \p has been encountered, provided that
1001 PCRE is compiled with support for Unicode properties. On entry, ptrptr is
1002 pointing at the P or p. On exit, it is pointing at the final character of the
1003 escape sequence.
1004
1005 Argument:
1006 ptrptr points to the pattern position pointer
1007 negptr points to a boolean that is set TRUE for negation else FALSE
1008 dptr points to an int that is set to the detailed property value
1009 errorcodeptr points to the error code variable
1010
1011 Returns: type value from ucp_type_table, or -1 for an invalid type
1012 */
1013
1014 static int
1015 get_ucp(const uschar **ptrptr, BOOL *negptr, int *dptr, int *errorcodeptr)
1016 {
1017 int c, i, bot, top;
1018 const uschar *ptr = *ptrptr;
1019 char name[32];
1020
1021 c = *(++ptr);
1022 if (c == 0) goto ERROR_RETURN;
1023
1024 *negptr = FALSE;
1025
1026 /* \P or \p can be followed by a name in {}, optionally preceded by ^ for
1027 negation. */
1028
1029 if (c == CHAR_LEFT_CURLY_BRACKET)
1030 {
1031 if (ptr[1] == CHAR_CIRCUMFLEX_ACCENT)
1032 {
1033 *negptr = TRUE;
1034 ptr++;
1035 }
1036 for (i = 0; i < (int)sizeof(name) - 1; i++)
1037 {
1038 c = *(++ptr);
1039 if (c == 0) goto ERROR_RETURN;
1040 if (c == CHAR_RIGHT_CURLY_BRACKET) break;
1041 name[i] = c;
1042 }
1043 if (c != CHAR_RIGHT_CURLY_BRACKET) goto ERROR_RETURN;
1044 name[i] = 0;
1045 }
1046
1047 /* Otherwise there is just one following character */
1048
1049 else
1050 {
1051 name[0] = c;
1052 name[1] = 0;
1053 }
1054
1055 *ptrptr = ptr;
1056
1057 /* Search for a recognized property name using binary chop */
1058
1059 bot = 0;
1060 top = _pcre_utt_size;
1061
1062 while (bot < top)
1063 {
1064 i = (bot + top) >> 1;
1065 c = strcmp(name, _pcre_utt_names + _pcre_utt[i].name_offset);
1066 if (c == 0)
1067 {
1068 *dptr = _pcre_utt[i].value;
1069 return _pcre_utt[i].type;
1070 }
1071 if (c > 0) bot = i + 1; else top = i;
1072 }
1073
1074 *errorcodeptr = ERR47;
1075 *ptrptr = ptr;
1076 return -1;
1077
1078 ERROR_RETURN:
1079 *errorcodeptr = ERR46;
1080 *ptrptr = ptr;
1081 return -1;
1082 }
1083 #endif
1084
1085
1086
1087
1088 /*************************************************
1089 * Read repeat counts *
1090 *************************************************/
1091
1092 /* Read an item of the form {n,m} and return the values. This is called only
1093 after is_counted_repeat() has confirmed that a repeat-count quantifier exists,
1094 so the syntax is guaranteed to be correct, but we need to check the values.
1095
1096 Arguments:
1097 p pointer to first char after '{'
1098 minp pointer to int for min
1099 maxp pointer to int for max
1100 returned as -1 if no max
1101 errorcodeptr points to error code variable
1102
1103 Returns: pointer to '}' on success;
1104 current ptr on error, with errorcodeptr set non-zero
1105 */
1106
1107 static const uschar *
1108 read_repeat_counts(const uschar *p, int *minp, int *maxp, int *errorcodeptr)
1109 {
1110 int min = 0;
1111 int max = -1;
1112
1113 /* Read the minimum value and do a paranoid check: a negative value indicates
1114 an integer overflow. */
1115
1116 while ((digitab[*p] & ctype_digit) != 0) min = min * 10 + *p++ - CHAR_0;
1117 if (min < 0 || min > 65535)
1118 {
1119 *errorcodeptr = ERR5;
1120 return p;
1121 }
1122
1123 /* Read the maximum value if there is one, and again do a paranoid on its size.
1124 Also, max must not be less than min. */
1125
1126 if (*p == CHAR_RIGHT_CURLY_BRACKET) max = min; else
1127 {
1128 if (*(++p) != CHAR_RIGHT_CURLY_BRACKET)
1129 {
1130 max = 0;
1131 while((digitab[*p] & ctype_digit) != 0) max = max * 10 + *p++ - CHAR_0;
1132 if (max < 0 || max > 65535)
1133 {
1134 *errorcodeptr = ERR5;
1135 return p;
1136 }
1137 if (max < min)
1138 {
1139 *errorcodeptr = ERR4;
1140 return p;
1141 }
1142 }
1143 }
1144
1145 /* Fill in the required variables, and pass back the pointer to the terminating
1146 '}'. */
1147
1148 *minp = min;
1149 *maxp = max;
1150 return p;
1151 }
1152
1153
1154
1155 /*************************************************
1156 * Subroutine for finding forward reference *
1157 *************************************************/
1158
1159 /* This recursive function is called only from find_parens() below. The
1160 top-level call starts at the beginning of the pattern. All other calls must
1161 start at a parenthesis. It scans along a pattern's text looking for capturing
1162 subpatterns, and counting them. If it finds a named pattern that matches the
1163 name it is given, it returns its number. Alternatively, if the name is NULL, it
1164 returns when it reaches a given numbered subpattern. Recursion is used to keep
1165 track of subpatterns that reset the capturing group numbers - the (?| feature.
1166
1167 This function was originally called only from the second pass, in which we know
1168 that if (?< or (?' or (?P< is encountered, the name will be correctly
1169 terminated because that is checked in the first pass. There is now one call to
1170 this function in the first pass, to check for a recursive back reference by
1171 name (so that we can make the whole group atomic). In this case, we need check
1172 only up to the current position in the pattern, and that is still OK because
1173 and previous occurrences will have been checked. To make this work, the test
1174 for "end of pattern" is a check against cd->end_pattern in the main loop,
1175 instead of looking for a binary zero. This means that the special first-pass
1176 call can adjust cd->end_pattern temporarily. (Checks for binary zero while
1177 processing items within the loop are OK, because afterwards the main loop will
1178 terminate.)
1179
1180 Arguments:
1181 ptrptr address of the current character pointer (updated)
1182 cd compile background data
1183 name name to seek, or NULL if seeking a numbered subpattern
1184 lorn name length, or subpattern number if name is NULL
1185 xmode TRUE if we are in /x mode
1186 utf8 TRUE if we are in UTF-8 mode
1187 count pointer to the current capturing subpattern number (updated)
1188
1189 Returns: the number of the named subpattern, or -1 if not found
1190 */
1191
1192 static int
1193 find_parens_sub(uschar **ptrptr, compile_data *cd, const uschar *name, int lorn,
1194 BOOL xmode, BOOL utf8, int *count)
1195 {
1196 uschar *ptr = *ptrptr;
1197 int start_count = *count;
1198 int hwm_count = start_count;
1199 BOOL dup_parens = FALSE;
1200
1201 /* If the first character is a parenthesis, check on the type of group we are
1202 dealing with. The very first call may not start with a parenthesis. */
1203
1204 if (ptr[0] == CHAR_LEFT_PARENTHESIS)
1205 {
1206 /* Handle specials such as (*SKIP) or (*UTF8) etc. */
1207
1208 if (ptr[1] == CHAR_ASTERISK) ptr += 2;
1209
1210 /* Handle a normal, unnamed capturing parenthesis. */
1211
1212 else if (ptr[1] != CHAR_QUESTION_MARK)
1213 {
1214 *count += 1;
1215 if (name == NULL && *count == lorn) return *count;
1216 ptr++;
1217 }
1218
1219 /* All cases now have (? at the start. Remember when we are in a group
1220 where the parenthesis numbers are duplicated. */
1221
1222 else if (ptr[2] == CHAR_VERTICAL_LINE)
1223 {
1224 ptr += 3;
1225 dup_parens = TRUE;
1226 }
1227
1228 /* Handle comments; all characters are allowed until a ket is reached. */
1229
1230 else if (ptr[2] == CHAR_NUMBER_SIGN)
1231 {
1232 for (ptr += 3; *ptr != 0; ptr++) if (*ptr == CHAR_RIGHT_PARENTHESIS) break;
1233 goto FAIL_EXIT;
1234 }
1235
1236 /* Handle a condition. If it is an assertion, just carry on so that it
1237 is processed as normal. If not, skip to the closing parenthesis of the
1238 condition (there can't be any nested parens). */
1239
1240 else if (ptr[2] == CHAR_LEFT_PARENTHESIS)
1241 {
1242 ptr += 2;
1243 if (ptr[1] != CHAR_QUESTION_MARK)
1244 {
1245 while (*ptr != 0 && *ptr != CHAR_RIGHT_PARENTHESIS) ptr++;
1246 if (*ptr != 0) ptr++;
1247 }
1248 }
1249
1250 /* Start with (? but not a condition. */
1251
1252 else
1253 {
1254 ptr += 2;
1255 if (*ptr == CHAR_P) ptr++; /* Allow optional P */
1256
1257 /* We have to disambiguate (?<! and (?<= from (?<name> for named groups */
1258
1259 if ((*ptr == CHAR_LESS_THAN_SIGN && ptr[1] != CHAR_EXCLAMATION_MARK &&
1260 ptr[1] != CHAR_EQUALS_SIGN) || *ptr == CHAR_APOSTROPHE)
1261 {
1262 int term;
1263 const uschar *thisname;
1264 *count += 1;
1265 if (name == NULL && *count == lorn) return *count;
1266 term = *ptr++;
1267 if (term == CHAR_LESS_THAN_SIGN) term = CHAR_GREATER_THAN_SIGN;
1268 thisname = ptr;
1269 while (*ptr != term) ptr++;
1270 if (name != NULL && lorn == ptr - thisname &&
1271 strncmp((const char *)name, (const char *)thisname, lorn) == 0)
1272 return *count;
1273 term++;
1274 }
1275 }
1276 }
1277
1278 /* Past any initial parenthesis handling, scan for parentheses or vertical
1279 bars. Stop if we get to cd->end_pattern. Note that this is important for the
1280 first-pass call when this value is temporarily adjusted to stop at the current
1281 position. So DO NOT change this to a test for binary zero. */
1282
1283 for (; ptr < cd->end_pattern; ptr++)
1284 {
1285 /* Skip over backslashed characters and also entire \Q...\E */
1286
1287 if (*ptr == CHAR_BACKSLASH)
1288 {
1289 if (*(++ptr) == 0) goto FAIL_EXIT;
1290 if (*ptr == CHAR_Q) for (;;)
1291 {
1292 while (*(++ptr) != 0 && *ptr != CHAR_BACKSLASH) {};
1293 if (*ptr == 0) goto FAIL_EXIT;
1294 if (*(++ptr) == CHAR_E) break;
1295 }
1296 continue;
1297 }
1298
1299 /* Skip over character classes; this logic must be similar to the way they
1300 are handled for real. If the first character is '^', skip it. Also, if the
1301 first few characters (either before or after ^) are \Q\E or \E we skip them
1302 too. This makes for compatibility with Perl. Note the use of STR macros to
1303 encode "Q\\E" so that it works in UTF-8 on EBCDIC platforms. */
1304
1305 if (*ptr == CHAR_LEFT_SQUARE_BRACKET)
1306 {
1307 BOOL negate_class = FALSE;
1308 for (;;)
1309 {
1310 if (ptr[1] == CHAR_BACKSLASH)
1311 {
1312 if (ptr[2] == CHAR_E)
1313 ptr+= 2;
1314 else if (strncmp((const char *)ptr+2,
1315 STR_Q STR_BACKSLASH STR_E, 3) == 0)
1316 ptr += 4;
1317 else
1318 break;
1319 }
1320 else if (!negate_class && ptr[1] == CHAR_CIRCUMFLEX_ACCENT)
1321 {
1322 negate_class = TRUE;
1323 ptr++;
1324 }
1325 else break;
1326 }
1327
1328 /* If the next character is ']', it is a data character that must be
1329 skipped, except in JavaScript compatibility mode. */
1330
1331 if (ptr[1] == CHAR_RIGHT_SQUARE_BRACKET &&
1332 (cd->external_options & PCRE_JAVASCRIPT_COMPAT) == 0)
1333 ptr++;
1334
1335 while (*(++ptr) != CHAR_RIGHT_SQUARE_BRACKET)
1336 {
1337 if (*ptr == 0) return -1;
1338 if (*ptr == CHAR_BACKSLASH)
1339 {
1340 if (*(++ptr) == 0) goto FAIL_EXIT;
1341 if (*ptr == CHAR_Q) for (;;)
1342 {
1343 while (*(++ptr) != 0 && *ptr != CHAR_BACKSLASH) {};
1344 if (*ptr == 0) goto FAIL_EXIT;
1345 if (*(++ptr) == CHAR_E) break;
1346 }
1347 continue;
1348 }
1349 }
1350 continue;
1351 }
1352
1353 /* Skip comments in /x mode */
1354
1355 if (xmode && *ptr == CHAR_NUMBER_SIGN)
1356 {
1357 ptr++;
1358 while (*ptr != 0)
1359 {
1360 if (IS_NEWLINE(ptr)) { ptr += cd->nllen - 1; break; }
1361 ptr++;
1362 #ifdef SUPPORT_UTF8
1363 if (utf8) while ((*ptr & 0xc0) == 0x80) ptr++;
1364 #endif
1365 }
1366 if (*ptr == 0) goto FAIL_EXIT;
1367 continue;
1368 }
1369
1370 /* Check for the special metacharacters */
1371
1372 if (*ptr == CHAR_LEFT_PARENTHESIS)
1373 {
1374 int rc = find_parens_sub(&ptr, cd, name, lorn, xmode, utf8, count);
1375 if (rc > 0) return rc;
1376 if (*ptr == 0) goto FAIL_EXIT;
1377 }
1378
1379 else if (*ptr == CHAR_RIGHT_PARENTHESIS)
1380 {
1381 if (dup_parens && *count < hwm_count) *count = hwm_count;
1382 goto FAIL_EXIT;
1383 }
1384
1385 else if (*ptr == CHAR_VERTICAL_LINE && dup_parens)
1386 {
1387 if (*count > hwm_count) hwm_count = *count;
1388 *count = start_count;
1389 }
1390 }
1391
1392 FAIL_EXIT:
1393 *ptrptr = ptr;
1394 return -1;
1395 }
1396
1397
1398
1399
1400 /*************************************************
1401 * Find forward referenced subpattern *
1402 *************************************************/
1403
1404 /* This function scans along a pattern's text looking for capturing
1405 subpatterns, and counting them. If it finds a named pattern that matches the
1406 name it is given, it returns its number. Alternatively, if the name is NULL, it
1407 returns when it reaches a given numbered subpattern. This is used for forward
1408 references to subpatterns. We used to be able to start this scan from the
1409 current compiling point, using the current count value from cd->bracount, and
1410 do it all in a single loop, but the addition of the possibility of duplicate
1411 subpattern numbers means that we have to scan from the very start, in order to
1412 take account of such duplicates, and to use a recursive function to keep track
1413 of the different types of group.
1414
1415 Arguments:
1416 cd compile background data
1417 name name to seek, or NULL if seeking a numbered subpattern
1418 lorn name length, or subpattern number if name is NULL
1419 xmode TRUE if we are in /x mode
1420 utf8 TRUE if we are in UTF-8 mode
1421
1422 Returns: the number of the found subpattern, or -1 if not found
1423 */
1424
1425 static int
1426 find_parens(compile_data *cd, const uschar *name, int lorn, BOOL xmode,
1427 BOOL utf8)
1428 {
1429 uschar *ptr = (uschar *)cd->start_pattern;
1430 int count = 0;
1431 int rc;
1432
1433 /* If the pattern does not start with an opening parenthesis, the first call
1434 to find_parens_sub() will scan right to the end (if necessary). However, if it
1435 does start with a parenthesis, find_parens_sub() will return when it hits the
1436 matching closing parens. That is why we have to have a loop. */
1437
1438 for (;;)
1439 {
1440 rc = find_parens_sub(&ptr, cd, name, lorn, xmode, utf8, &count);
1441 if (rc > 0 || *ptr++ == 0) break;
1442 }
1443
1444 return rc;
1445 }
1446
1447
1448
1449
1450 /*************************************************
1451 * Find first significant op code *
1452 *************************************************/
1453
1454 /* This is called by several functions that scan a compiled expression looking
1455 for a fixed first character, or an anchoring op code etc. It skips over things
1456 that do not influence this. For some calls, it makes sense to skip negative
1457 forward and all backward assertions, and also the \b assertion; for others it
1458 does not.
1459
1460 Arguments:
1461 code pointer to the start of the group
1462 skipassert TRUE if certain assertions are to be skipped
1463
1464 Returns: pointer to the first significant opcode
1465 */
1466
1467 static const uschar*
1468 first_significant_code(const uschar *code, BOOL skipassert)
1469 {
1470 for (;;)
1471 {
1472 switch ((int)*code)
1473 {
1474 case OP_ASSERT_NOT:
1475 case OP_ASSERTBACK:
1476 case OP_ASSERTBACK_NOT:
1477 if (!skipassert) return code;
1478 do code += GET(code, 1); while (*code == OP_ALT);
1479 code += _pcre_OP_lengths[*code];
1480 break;
1481
1482 case OP_WORD_BOUNDARY:
1483 case OP_NOT_WORD_BOUNDARY:
1484 if (!skipassert) return code;
1485 /* Fall through */
1486
1487 case OP_CALLOUT:
1488 case OP_CREF:
1489 case OP_NCREF:
1490 case OP_RREF:
1491 case OP_NRREF:
1492 case OP_DEF:
1493 code += _pcre_OP_lengths[*code];
1494 break;
1495
1496 default:
1497 return code;
1498 }
1499 }
1500 /* Control never reaches here */
1501 }
1502
1503
1504
1505
1506 /*************************************************
1507 * Find the fixed length of a branch *
1508 *************************************************/
1509
1510 /* Scan a branch and compute the fixed length of subject that will match it,
1511 if the length is fixed. This is needed for dealing with backward assertions.
1512 In UTF8 mode, the result is in characters rather than bytes. The branch is
1513 temporarily terminated with OP_END when this function is called.
1514
1515 This function is called when a backward assertion is encountered, so that if it
1516 fails, the error message can point to the correct place in the pattern.
1517 However, we cannot do this when the assertion contains subroutine calls,
1518 because they can be forward references. We solve this by remembering this case
1519 and doing the check at the end; a flag specifies which mode we are running in.
1520
1521 Arguments:
1522 code points to the start of the pattern (the bracket)
1523 utf8 TRUE in UTF-8 mode
1524 atend TRUE if called when the pattern is complete
1525 cd the "compile data" structure
1526
1527 Returns: the fixed length,
1528 or -1 if there is no fixed length,
1529 or -2 if \C was encountered
1530 or -3 if an OP_RECURSE item was encountered and atend is FALSE
1531 */
1532
1533 static int
1534 find_fixedlength(uschar *code, BOOL utf8, BOOL atend, compile_data *cd)
1535 {
1536 int length = -1;
1537
1538 register int branchlength = 0;
1539 register uschar *cc = code + 1 + LINK_SIZE;
1540
1541 /* Scan along the opcodes for this branch. If we get to the end of the
1542 branch, check the length against that of the other branches. */
1543
1544 for (;;)
1545 {
1546 int d;
1547 uschar *ce, *cs;
1548 register int op = *cc;
1549 switch (op)
1550 {
1551 /* We only need to continue for OP_CBRA (normal capturing bracket) and
1552 OP_BRA (normal non-capturing bracket) because the other variants of these
1553 opcodes are all concerned with unlimited repeated groups, which of course
1554 are not of fixed length. They will cause a -1 response from the default
1555 case of this switch. */
1556
1557 case OP_CBRA:
1558 case OP_BRA:
1559 case OP_ONCE:
1560 case OP_ONCE_NC:
1561 case OP_COND:
1562 d = find_fixedlength(cc + ((op == OP_CBRA)? 2:0), utf8, atend, cd);
1563 if (d < 0) return d;
1564 branchlength += d;
1565 do cc += GET(cc, 1); while (*cc == OP_ALT);
1566 cc += 1 + LINK_SIZE;
1567 break;
1568
1569 /* Reached end of a branch; if it's a ket it is the end of a nested
1570 call. If it's ALT it is an alternation in a nested call. If it is
1571 END it's the end of the outer call. All can be handled by the same code.
1572 Note that we must not include the OP_KETRxxx opcodes here, because they
1573 all imply an unlimited repeat. */
1574
1575 case OP_ALT:
1576 case OP_KET:
1577 case OP_END:
1578 if (length < 0) length = branchlength;
1579 else if (length != branchlength) return -1;
1580 if (*cc != OP_ALT) return length;
1581 cc += 1 + LINK_SIZE;
1582 branchlength = 0;
1583 break;
1584
1585 /* A true recursion implies not fixed length, but a subroutine call may
1586 be OK. If the subroutine is a forward reference, we can't deal with
1587 it until the end of the pattern, so return -3. */
1588
1589 case OP_RECURSE:
1590 if (!atend) return -3;
1591 cs = ce = (uschar *)cd->start_code + GET(cc, 1); /* Start subpattern */
1592 do ce += GET(ce, 1); while (*ce == OP_ALT); /* End subpattern */
1593 if (cc > cs && cc < ce) return -1; /* Recursion */
1594 d = find_fixedlength(cs + 2, utf8, atend, cd);
1595 if (d < 0) return d;
1596 branchlength += d;
1597 cc += 1 + LINK_SIZE;
1598 break;
1599
1600 /* Skip over assertive subpatterns */
1601
1602 case OP_ASSERT:
1603 case OP_ASSERT_NOT:
1604 case OP_ASSERTBACK:
1605 case OP_ASSERTBACK_NOT:
1606 do cc += GET(cc, 1); while (*cc == OP_ALT);
1607 /* Fall through */
1608
1609 /* Skip over things that don't match chars */
1610
1611 case OP_REVERSE:
1612 case OP_CREF:
1613 case OP_NCREF:
1614 case OP_RREF:
1615 case OP_NRREF:
1616 case OP_DEF:
1617 case OP_CALLOUT:
1618 case OP_SOD:
1619 case OP_SOM:
1620 case OP_SET_SOM:
1621 case OP_EOD:
1622 case OP_EODN:
1623 case OP_CIRC:
1624 case OP_CIRCM:
1625 case OP_DOLL:
1626 case OP_DOLLM:
1627 case OP_NOT_WORD_BOUNDARY:
1628 case OP_WORD_BOUNDARY:
1629 cc += _pcre_OP_lengths[*cc];
1630 break;
1631
1632 /* Handle literal characters */
1633
1634 case OP_CHAR:
1635 case OP_CHARI:
1636 case OP_NOT:
1637 case OP_NOTI:
1638 branchlength++;
1639 cc += 2;
1640 #ifdef SUPPORT_UTF8
1641 if (utf8 && cc[-1] >= 0xc0) cc += _pcre_utf8_table4[cc[-1] & 0x3f];
1642 #endif
1643 break;
1644
1645 /* Handle exact repetitions. The count is already in characters, but we
1646 need to skip over a multibyte character in UTF8 mode. */
1647
1648 case OP_EXACT:
1649 branchlength += GET2(cc,1);
1650 cc += 4;
1651 #ifdef SUPPORT_UTF8
1652 if (utf8 && cc[-1] >= 0xc0) cc += _pcre_utf8_table4[cc[-1] & 0x3f];
1653 #endif
1654 break;
1655
1656 case OP_TYPEEXACT:
1657 branchlength += GET2(cc,1);
1658 if (cc[3] == OP_PROP || cc[3] == OP_NOTPROP) cc += 2;
1659 cc += 4;
1660 break;
1661
1662 /* Handle single-char matchers */
1663
1664 case OP_PROP:
1665 case OP_NOTPROP:
1666 cc += 2;
1667 /* Fall through */
1668
1669 case OP_NOT_DIGIT:
1670 case OP_DIGIT:
1671 case OP_NOT_WHITESPACE:
1672 case OP_WHITESPACE:
1673 case OP_NOT_WORDCHAR:
1674 case OP_WORDCHAR:
1675 case OP_ANY:
1676 case OP_ALLANY:
1677 branchlength++;
1678 cc++;
1679 break;
1680
1681 /* The single-byte matcher isn't allowed */
1682
1683 case OP_ANYBYTE:
1684 return -2;
1685
1686 /* Check a class for variable quantification */
1687
1688 #ifdef SUPPORT_UTF8
1689 case OP_XCLASS:
1690 cc += GET(cc, 1) - 33;
1691 /* Fall through */
1692 #endif
1693
1694 case OP_CLASS:
1695 case OP_NCLASS:
1696 cc += 33;
1697
1698 switch (*cc)
1699 {
1700 case OP_CRSTAR:
1701 case OP_CRMINSTAR:
1702 case OP_CRQUERY:
1703 case OP_CRMINQUERY:
1704 return -1;
1705
1706 case OP_CRRANGE:
1707 case OP_CRMINRANGE:
1708 if (GET2(cc,1) != GET2(cc,3)) return -1;
1709 branchlength += GET2(cc,1);
1710 cc += 5;
1711 break;
1712
1713 default:
1714 branchlength++;
1715 }
1716 break;
1717
1718 /* Anything else is variable length */
1719
1720 default:
1721 return -1;
1722 }
1723 }
1724 /* Control never gets here */
1725 }
1726
1727
1728
1729
1730 /*************************************************
1731 * Scan compiled regex for specific bracket *
1732 *************************************************/
1733
1734 /* This little function scans through a compiled pattern until it finds a
1735 capturing bracket with the given number, or, if the number is negative, an
1736 instance of OP_REVERSE for a lookbehind. The function is global in the C sense
1737 so that it can be called from pcre_study() when finding the minimum matching
1738 length.
1739
1740 Arguments:
1741 code points to start of expression
1742 utf8 TRUE in UTF-8 mode
1743 number the required bracket number or negative to find a lookbehind
1744
1745 Returns: pointer to the opcode for the bracket, or NULL if not found
1746 */
1747
1748 const uschar *
1749 _pcre_find_bracket(const uschar *code, BOOL utf8, int number)
1750 {
1751 for (;;)
1752 {
1753 register int c = *code;
1754
1755 if (c == OP_END) return NULL;
1756
1757 /* XCLASS is used for classes that cannot be represented just by a bit
1758 map. This includes negated single high-valued characters. The length in
1759 the table is zero; the actual length is stored in the compiled code. */
1760
1761 if (c == OP_XCLASS) code += GET(code, 1);
1762
1763 /* Handle recursion */
1764
1765 else if (c == OP_REVERSE)
1766 {
1767 if (number < 0) return (uschar *)code;
1768 code += _pcre_OP_lengths[c];
1769 }
1770
1771 /* Handle capturing bracket */
1772
1773 else if (c == OP_CBRA || c == OP_SCBRA ||
1774 c == OP_CBRAPOS || c == OP_SCBRAPOS)
1775 {
1776 int n = GET2(code, 1+LINK_SIZE);
1777 if (n == number) return (uschar *)code;
1778 code += _pcre_OP_lengths[c];
1779 }
1780
1781 /* Otherwise, we can get the item's length from the table, except that for
1782 repeated character types, we have to test for \p and \P, which have an extra
1783 two bytes of parameters, and for MARK/PRUNE/SKIP/THEN with an argument, we
1784 must add in its length. */
1785
1786 else
1787 {
1788 switch(c)
1789 {
1790 case OP_TYPESTAR:
1791 case OP_TYPEMINSTAR:
1792 case OP_TYPEPLUS:
1793 case OP_TYPEMINPLUS:
1794 case OP_TYPEQUERY:
1795 case OP_TYPEMINQUERY:
1796 case OP_TYPEPOSSTAR:
1797 case OP_TYPEPOSPLUS:
1798 case OP_TYPEPOSQUERY:
1799 if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
1800 break;
1801
1802 case OP_TYPEUPTO:
1803 case OP_TYPEMINUPTO:
1804 case OP_TYPEEXACT:
1805 case OP_TYPEPOSUPTO:
1806 if (code[3] == OP_PROP || code[3] == OP_NOTPROP) code += 2;
1807 break;
1808
1809 case OP_MARK:
1810 case OP_PRUNE_ARG:
1811 case OP_SKIP_ARG:
1812 code += code[1];
1813 break;
1814
1815 case OP_THEN_ARG:
1816 code += code[1];
1817 break;
1818 }
1819
1820 /* Add in the fixed length from the table */
1821
1822 code += _pcre_OP_lengths[c];
1823
1824 /* In UTF-8 mode, opcodes that are followed by a character may be followed by
1825 a multi-byte character. The length in the table is a minimum, so we have to
1826 arrange to skip the extra bytes. */
1827
1828 #ifdef SUPPORT_UTF8
1829 if (utf8) switch(c)
1830 {
1831 case OP_CHAR:
1832 case OP_CHARI:
1833 case OP_EXACT:
1834 case OP_EXACTI:
1835 case OP_UPTO:
1836 case OP_UPTOI:
1837 case OP_MINUPTO:
1838 case OP_MINUPTOI:
1839 case OP_POSUPTO:
1840 case OP_POSUPTOI:
1841 case OP_STAR:
1842 case OP_STARI:
1843 case OP_MINSTAR:
1844 case OP_MINSTARI:
1845 case OP_POSSTAR:
1846 case OP_POSSTARI:
1847 case OP_PLUS:
1848 case OP_PLUSI:
1849 case OP_MINPLUS:
1850 case OP_MINPLUSI:
1851 case OP_POSPLUS:
1852 case OP_POSPLUSI:
1853 case OP_QUERY:
1854 case OP_QUERYI:
1855 case OP_MINQUERY:
1856 case OP_MINQUERYI:
1857 case OP_POSQUERY:
1858 case OP_POSQUERYI:
1859 if (code[-1] >= 0xc0) code += _pcre_utf8_table4[code[-1] & 0x3f];
1860 break;
1861 }
1862 #else
1863 (void)(utf8); /* Keep compiler happy by referencing function argument */
1864 #endif
1865 }
1866 }
1867 }
1868
1869
1870
1871 /*************************************************
1872 * Scan compiled regex for recursion reference *
1873 *************************************************/
1874
1875 /* This little function scans through a compiled pattern until it finds an
1876 instance of OP_RECURSE.
1877
1878 Arguments:
1879 code points to start of expression
1880 utf8 TRUE in UTF-8 mode
1881
1882 Returns: pointer to the opcode for OP_RECURSE, or NULL if not found
1883 */
1884
1885 static const uschar *
1886 find_recurse(const uschar *code, BOOL utf8)
1887 {
1888 for (;;)
1889 {
1890 register int c = *code;
1891 if (c == OP_END) return NULL;
1892 if (c == OP_RECURSE) return code;
1893
1894 /* XCLASS is used for classes that cannot be represented just by a bit
1895 map. This includes negated single high-valued characters. The length in
1896 the table is zero; the actual length is stored in the compiled code. */
1897
1898 if (c == OP_XCLASS) code += GET(code, 1);
1899
1900 /* Otherwise, we can get the item's length from the table, except that for
1901 repeated character types, we have to test for \p and \P, which have an extra
1902 two bytes of parameters, and for MARK/PRUNE/SKIP/THEN with an argument, we
1903 must add in its length. */
1904
1905 else
1906 {
1907 switch(c)
1908 {
1909 case OP_TYPESTAR:
1910 case OP_TYPEMINSTAR:
1911 case OP_TYPEPLUS:
1912 case OP_TYPEMINPLUS:
1913 case OP_TYPEQUERY:
1914 case OP_TYPEMINQUERY:
1915 case OP_TYPEPOSSTAR:
1916 case OP_TYPEPOSPLUS:
1917 case OP_TYPEPOSQUERY:
1918 if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
1919 break;
1920
1921 case OP_TYPEPOSUPTO:
1922 case OP_TYPEUPTO:
1923 case OP_TYPEMINUPTO:
1924 case OP_TYPEEXACT:
1925 if (code[3] == OP_PROP || code[3] == OP_NOTPROP) code += 2;
1926 break;
1927
1928 case OP_MARK:
1929 case OP_PRUNE_ARG:
1930 case OP_SKIP_ARG:
1931 code += code[1];
1932 break;
1933
1934 case OP_THEN_ARG:
1935 code += code[1];
1936 break;
1937 }
1938
1939 /* Add in the fixed length from the table */
1940
1941 code += _pcre_OP_lengths[c];
1942
1943 /* In UTF-8 mode, opcodes that are followed by a character may be followed
1944 by a multi-byte character. The length in the table is a minimum, so we have
1945 to arrange to skip the extra bytes. */
1946
1947 #ifdef SUPPORT_UTF8
1948 if (utf8) switch(c)
1949 {
1950 case OP_CHAR:
1951 case OP_CHARI:
1952 case OP_EXACT:
1953 case OP_EXACTI:
1954 case OP_UPTO:
1955 case OP_UPTOI:
1956 case OP_MINUPTO:
1957 case OP_MINUPTOI:
1958 case OP_POSUPTO:
1959 case OP_POSUPTOI:
1960 case OP_STAR:
1961 case OP_STARI:
1962 case OP_MINSTAR:
1963 case OP_MINSTARI:
1964 case OP_POSSTAR:
1965 case OP_POSSTARI:
1966 case OP_PLUS:
1967 case OP_PLUSI:
1968 case OP_MINPLUS:
1969 case OP_MINPLUSI:
1970 case OP_POSPLUS:
1971 case OP_POSPLUSI:
1972 case OP_QUERY:
1973 case OP_QUERYI:
1974 case OP_MINQUERY:
1975 case OP_MINQUERYI:
1976 case OP_POSQUERY:
1977 case OP_POSQUERYI:
1978 if (code[-1] >= 0xc0) code += _pcre_utf8_table4[code[-1] & 0x3f];
1979 break;
1980 }
1981 #else
1982 (void)(utf8); /* Keep compiler happy by referencing function argument */
1983 #endif
1984 }
1985 }
1986 }
1987
1988
1989
1990 /*************************************************
1991 * Scan compiled branch for non-emptiness *
1992 *************************************************/
1993
1994 /* This function scans through a branch of a compiled pattern to see whether it
1995 can match the empty string or not. It is called from could_be_empty()
1996 below and from compile_branch() when checking for an unlimited repeat of a
1997 group that can match nothing. Note that first_significant_code() skips over
1998 backward and negative forward assertions when its final argument is TRUE. If we
1999 hit an unclosed bracket, we return "empty" - this means we've struck an inner
2000 bracket whose current branch will already have been scanned.
2001
2002 Arguments:
2003 code points to start of search
2004 endcode points to where to stop
2005 utf8 TRUE if in UTF8 mode
2006 cd contains pointers to tables etc.
2007
2008 Returns: TRUE if what is matched could be empty
2009 */
2010
2011 static BOOL
2012 could_be_empty_branch(const uschar *code, const uschar *endcode, BOOL utf8,
2013 compile_data *cd)
2014 {
2015 register int c;
2016 for (code = first_significant_code(code + _pcre_OP_lengths[*code], TRUE);
2017 code < endcode;
2018 code = first_significant_code(code + _pcre_OP_lengths[c], TRUE))
2019 {
2020 const uschar *ccode;
2021
2022 c = *code;
2023
2024 /* Skip over forward assertions; the other assertions are skipped by
2025 first_significant_code() with a TRUE final argument. */
2026
2027 if (c == OP_ASSERT)
2028 {
2029 do code += GET(code, 1); while (*code == OP_ALT);
2030 c = *code;
2031 continue;
2032 }
2033
2034 /* For a recursion/subroutine call, if its end has been reached, which
2035 implies a backward reference subroutine call, we can scan it. If it's a
2036 forward reference subroutine call, we can't. To detect forward reference
2037 we have to scan up the list that is kept in the workspace. This function is
2038 called only when doing the real compile, not during the pre-compile that
2039 measures the size of the compiled pattern. */
2040
2041 if (c == OP_RECURSE)
2042 {
2043 const uschar *scode;
2044 BOOL empty_branch;
2045
2046 /* Test for forward reference */
2047
2048 for (scode = cd->start_workspace; scode < cd->hwm; scode += LINK_SIZE)
2049 if (GET(scode, 0) == code + 1 - cd->start_code) return TRUE;
2050
2051 /* Not a forward reference, test for completed backward reference */
2052
2053 empty_branch = FALSE;
2054 scode = cd->start_code + GET(code, 1);
2055 if (GET(scode, 1) == 0) return TRUE; /* Unclosed */
2056
2057 /* Completed backwards reference */
2058
2059 do
2060 {
2061 if (could_be_empty_branch(scode, endcode, utf8, cd))
2062 {
2063 empty_branch = TRUE;
2064 break;
2065 }
2066 scode += GET(scode, 1);
2067 }
2068 while (*scode == OP_ALT);
2069
2070 if (!empty_branch) return FALSE; /* All branches are non-empty */
2071 continue;
2072 }
2073
2074 /* Groups with zero repeats can of course be empty; skip them. */
2075
2076 if (c == OP_BRAZERO || c == OP_BRAMINZERO || c == OP_SKIPZERO ||
2077 c == OP_BRAPOSZERO)
2078 {
2079 code += _pcre_OP_lengths[c];
2080 do code += GET(code, 1); while (*code == OP_ALT);
2081 c = *code;
2082 continue;
2083 }
2084
2085 /* A nested group that is already marked as "could be empty" can just be
2086 skipped. */
2087
2088 if (c == OP_SBRA || c == OP_SBRAPOS ||
2089 c == OP_SCBRA || c == OP_SCBRAPOS)
2090 {
2091 do code += GET(code, 1); while (*code == OP_ALT);
2092 c = *code;
2093 continue;
2094 }
2095
2096 /* For other groups, scan the branches. */
2097
2098 if (c == OP_BRA || c == OP_BRAPOS ||
2099 c == OP_CBRA || c == OP_CBRAPOS ||
2100 c == OP_ONCE || c == OP_ONCE_NC ||
2101 c == OP_COND)
2102 {
2103 BOOL empty_branch;
2104 if (GET(code, 1) == 0) return TRUE; /* Hit unclosed bracket */
2105
2106 /* If a conditional group has only one branch, there is a second, implied,
2107 empty branch, so just skip over the conditional, because it could be empty.
2108 Otherwise, scan the individual branches of the group. */
2109
2110 if (c == OP_COND && code[GET(code, 1)] != OP_ALT)
2111 code += GET(code, 1);
2112 else
2113 {
2114 empty_branch = FALSE;
2115 do
2116 {
2117 if (!empty_branch && could_be_empty_branch(code, endcode, utf8, cd))
2118 empty_branch = TRUE;
2119 code += GET(code, 1);
2120 }
2121 while (*code == OP_ALT);
2122 if (!empty_branch) return FALSE; /* All branches are non-empty */
2123 }
2124
2125 c = *code;
2126 continue;
2127 }
2128
2129 /* Handle the other opcodes */
2130
2131 switch (c)
2132 {
2133 /* Check for quantifiers after a class. XCLASS is used for classes that
2134 cannot be represented just by a bit map. This includes negated single
2135 high-valued characters. The length in _pcre_OP_lengths[] is zero; the
2136 actual length is stored in the compiled code, so we must update "code"
2137 here. */
2138
2139 #ifdef SUPPORT_UTF8
2140 case OP_XCLASS:
2141 ccode = code += GET(code, 1);
2142 goto CHECK_CLASS_REPEAT;
2143 #endif
2144
2145 case OP_CLASS:
2146 case OP_NCLASS:
2147 ccode = code + 33;
2148
2149 #ifdef SUPPORT_UTF8
2150 CHECK_CLASS_REPEAT:
2151 #endif
2152
2153 switch (*ccode)
2154 {
2155 case OP_CRSTAR: /* These could be empty; continue */
2156 case OP_CRMINSTAR:
2157 case OP_CRQUERY:
2158 case OP_CRMINQUERY:
2159 break;
2160
2161 default: /* Non-repeat => class must match */
2162 case OP_CRPLUS: /* These repeats aren't empty */
2163 case OP_CRMINPLUS:
2164 return FALSE;
2165
2166 case OP_CRRANGE:
2167 case OP_CRMINRANGE:
2168 if (GET2(ccode, 1) > 0) return FALSE; /* Minimum > 0 */
2169 break;
2170 }
2171 break;
2172
2173 /* Opcodes that must match a character */
2174
2175 case OP_PROP:
2176 case OP_NOTPROP:
2177 case OP_EXTUNI:
2178 case OP_NOT_DIGIT:
2179 case OP_DIGIT:
2180 case OP_NOT_WHITESPACE:
2181 case OP_WHITESPACE:
2182 case OP_NOT_WORDCHAR:
2183 case OP_WORDCHAR:
2184 case OP_ANY:
2185 case OP_ALLANY:
2186 case OP_ANYBYTE:
2187 case OP_CHAR:
2188 case OP_CHARI:
2189 case OP_NOT:
2190 case OP_NOTI:
2191 case OP_PLUS:
2192 case OP_MINPLUS:
2193 case OP_POSPLUS:
2194 case OP_EXACT:
2195 case OP_NOTPLUS:
2196 case OP_NOTMINPLUS:
2197 case OP_NOTPOSPLUS:
2198 case OP_NOTEXACT:
2199 case OP_TYPEPLUS:
2200 case OP_TYPEMINPLUS:
2201 case OP_TYPEPOSPLUS:
2202 case OP_TYPEEXACT:
2203 return FALSE;
2204
2205 /* These are going to continue, as they may be empty, but we have to
2206 fudge the length for the \p and \P cases. */
2207
2208 case OP_TYPESTAR:
2209 case OP_TYPEMINSTAR:
2210 case OP_TYPEPOSSTAR:
2211 case OP_TYPEQUERY:
2212 case OP_TYPEMINQUERY:
2213 case OP_TYPEPOSQUERY:
2214 if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
2215 break;
2216
2217 /* Same for these */
2218
2219 case OP_TYPEUPTO:
2220 case OP_TYPEMINUPTO:
2221 case OP_TYPEPOSUPTO:
2222 if (code[3] == OP_PROP || code[3] == OP_NOTPROP) code += 2;
2223 break;
2224
2225 /* End of branch */
2226
2227 case OP_KET:
2228 case OP_KETRMAX:
2229 case OP_KETRMIN:
2230 case OP_KETRPOS:
2231 case OP_ALT:
2232 return TRUE;
2233
2234 /* In UTF-8 mode, STAR, MINSTAR, POSSTAR, QUERY, MINQUERY, POSQUERY, UPTO,
2235 MINUPTO, and POSUPTO may be followed by a multibyte character */
2236
2237 #ifdef SUPPORT_UTF8
2238 case OP_STAR:
2239 case OP_STARI:
2240 case OP_MINSTAR:
2241 case OP_MINSTARI:
2242 case OP_POSSTAR:
2243 case OP_POSSTARI:
2244 case OP_QUERY:
2245 case OP_QUERYI:
2246 case OP_MINQUERY:
2247 case OP_MINQUERYI:
2248 case OP_POSQUERY:
2249 case OP_POSQUERYI:
2250 if (utf8 && code[1] >= 0xc0) code += _pcre_utf8_table4[code[1] & 0x3f];
2251 break;
2252
2253 case OP_UPTO:
2254 case OP_UPTOI:
2255 case OP_MINUPTO:
2256 case OP_MINUPTOI:
2257 case OP_POSUPTO:
2258 case OP_POSUPTOI:
2259 if (utf8 && code[3] >= 0xc0) code += _pcre_utf8_table4[code[3] & 0x3f];
2260 break;
2261 #endif
2262
2263 /* MARK, and PRUNE/SKIP/THEN with an argument must skip over the argument
2264 string. */
2265
2266 case OP_MARK:
2267 case OP_PRUNE_ARG:
2268 case OP_SKIP_ARG:
2269 code += code[1];
2270 break;
2271
2272 case OP_THEN_ARG:
2273 code += code[1];
2274 break;
2275
2276 /* None of the remaining opcodes are required to match a character. */
2277
2278 default:
2279 break;
2280 }
2281 }
2282
2283 return TRUE;
2284 }
2285
2286
2287
2288 /*************************************************
2289 * Scan compiled regex for non-emptiness *
2290 *************************************************/
2291
2292 /* This function is called to check for left recursive calls. We want to check
2293 the current branch of the current pattern to see if it could match the empty
2294 string. If it could, we must look outwards for branches at other levels,
2295 stopping when we pass beyond the bracket which is the subject of the recursion.
2296 This function is called only during the real compile, not during the
2297 pre-compile.
2298
2299 Arguments:
2300 code points to start of the recursion
2301 endcode points to where to stop (current RECURSE item)
2302 bcptr points to the chain of current (unclosed) branch starts
2303 utf8 TRUE if in UTF-8 mode
2304 cd pointers to tables etc
2305
2306 Returns: TRUE if what is matched could be empty
2307 */
2308
2309 static BOOL
2310 could_be_empty(const uschar *code, const uschar *endcode, branch_chain *bcptr,
2311 BOOL utf8, compile_data *cd)
2312 {
2313 while (bcptr != NULL && bcptr->current_branch >= code)
2314 {
2315 if (!could_be_empty_branch(bcptr->current_branch, endcode, utf8, cd))
2316 return FALSE;
2317 bcptr = bcptr->outer;
2318 }
2319 return TRUE;
2320 }
2321
2322
2323
2324 /*************************************************
2325 * Check for POSIX class syntax *
2326 *************************************************/
2327
2328 /* This function is called when the sequence "[:" or "[." or "[=" is
2329 encountered in a character class. It checks whether this is followed by a
2330 sequence of characters terminated by a matching ":]" or ".]" or "=]". If we
2331 reach an unescaped ']' without the special preceding character, return FALSE.
2332
2333 Originally, this function only recognized a sequence of letters between the
2334 terminators, but it seems that Perl recognizes any sequence of characters,
2335 though of course unknown POSIX names are subsequently rejected. Perl gives an
2336 "Unknown POSIX class" error for [:f\oo:] for example, where previously PCRE
2337 didn't consider this to be a POSIX class. Likewise for [:1234:].
2338
2339 The problem in trying to be exactly like Perl is in the handling of escapes. We
2340 have to be sure that [abc[:x\]pqr] is *not* treated as containing a POSIX
2341 class, but [abc[:x\]pqr:]] is (so that an error can be generated). The code
2342 below handles the special case of \], but does not try to do any other escape
2343 processing. This makes it different from Perl for cases such as [:l\ower:]
2344 where Perl recognizes it as the POSIX class "lower" but PCRE does not recognize
2345 "l\ower". This is a lesser evil that not diagnosing bad classes when Perl does,
2346 I think.
2347
2348 A user pointed out that PCRE was rejecting [:a[:digit:]] whereas Perl was not.
2349 It seems that the appearance of a nested POSIX class supersedes an apparent
2350 external class. For example, [:a[:digit:]b:] matches "a", "b", ":", or
2351 a digit.
2352
2353 In Perl, unescaped square brackets may also appear as part of class names. For
2354 example, [:a[:abc]b:] gives unknown POSIX class "[:abc]b:]". However, for
2355 [:a[:abc]b][b:] it gives unknown POSIX class "[:abc]b][b:]", which does not
2356 seem right at all. PCRE does not allow closing square brackets in POSIX class
2357 names.
2358
2359 Arguments:
2360 ptr pointer to the initial [
2361 endptr where to return the end pointer
2362
2363 Returns: TRUE or FALSE
2364 */
2365
2366 static BOOL
2367 check_posix_syntax(const uschar *ptr, const uschar **endptr)
2368 {
2369 int terminator; /* Don't combine these lines; the Solaris cc */
2370 terminator = *(++ptr); /* compiler warns about "non-constant" initializer. */
2371 for (++ptr; *ptr != 0; ptr++)
2372 {
2373 if (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET)
2374 ptr++;
2375 else if (*ptr == CHAR_RIGHT_SQUARE_BRACKET) return FALSE;
2376 else
2377 {
2378 if (*ptr == terminator && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET)
2379 {
2380 *endptr = ptr;
2381 return TRUE;
2382 }
2383 if (*ptr == CHAR_LEFT_SQUARE_BRACKET &&
2384 (ptr[1] == CHAR_COLON || ptr[1] == CHAR_DOT ||
2385 ptr[1] == CHAR_EQUALS_SIGN) &&
2386 check_posix_syntax(ptr, endptr))
2387 return FALSE;
2388 }
2389 }
2390 return FALSE;
2391 }
2392
2393
2394
2395
2396 /*************************************************
2397 * Check POSIX class name *
2398 *************************************************/
2399
2400 /* This function is called to check the name given in a POSIX-style class entry
2401 such as [:alnum:].
2402
2403 Arguments:
2404 ptr points to the first letter
2405 len the length of the name
2406
2407 Returns: a value representing the name, or -1 if unknown
2408 */
2409
2410 static int
2411 check_posix_name(const uschar *ptr, int len)
2412 {
2413 const char *pn = posix_names;
2414 register int yield = 0;
2415 while (posix_name_lengths[yield] != 0)
2416 {
2417 if (len == posix_name_lengths[yield] &&
2418 strncmp((const char *)ptr, pn, len) == 0) return yield;
2419 pn += posix_name_lengths[yield] + 1;
2420 yield++;
2421 }
2422 return -1;
2423 }
2424
2425
2426 /*************************************************
2427 * Adjust OP_RECURSE items in repeated group *
2428 *************************************************/
2429
2430 /* OP_RECURSE items contain an offset from the start of the regex to the group
2431 that is referenced. This means that groups can be replicated for fixed
2432 repetition simply by copying (because the recursion is allowed to refer to
2433 earlier groups that are outside the current group). However, when a group is
2434 optional (i.e. the minimum quantifier is zero), OP_BRAZERO or OP_SKIPZERO is
2435 inserted before it, after it has been compiled. This means that any OP_RECURSE
2436 items within it that refer to the group itself or any contained groups have to
2437 have their offsets adjusted. That one of the jobs of this function. Before it
2438 is called, the partially compiled regex must be temporarily terminated with
2439 OP_END.
2440
2441 This function has been extended with the possibility of forward references for
2442 recursions and subroutine calls. It must also check the list of such references
2443 for the group we are dealing with. If it finds that one of the recursions in
2444 the current group is on this list, it adjusts the offset in the list, not the
2445 value in the reference (which is a group number).
2446
2447 Arguments:
2448 group points to the start of the group
2449 adjust the amount by which the group is to be moved
2450 utf8 TRUE in UTF-8 mode
2451 cd contains pointers to tables etc.
2452 save_hwm the hwm forward reference pointer at the start of the group
2453
2454 Returns: nothing
2455 */
2456
2457 static void
2458 adjust_recurse(uschar *group, int adjust, BOOL utf8, compile_data *cd,
2459 uschar *save_hwm)
2460 {
2461 uschar *ptr = group;
2462
2463 while ((ptr = (uschar *)find_recurse(ptr, utf8)) != NULL)
2464 {
2465 int offset;
2466 uschar *hc;
2467
2468 /* See if this recursion is on the forward reference list. If so, adjust the
2469 reference. */
2470
2471 for (hc = save_hwm; hc < cd->hwm; hc += LINK_SIZE)
2472 {
2473 offset = GET(hc, 0);
2474 if (cd->start_code + offset == ptr + 1)
2475 {
2476 PUT(hc, 0, offset + adjust);
2477 break;
2478 }
2479 }
2480
2481 /* Otherwise, adjust the recursion offset if it's after the start of this
2482 group. */
2483
2484 if (hc >= cd->hwm)
2485 {
2486 offset = GET(ptr, 1);
2487 if (cd->start_code + offset >= group) PUT(ptr, 1, offset + adjust);
2488 }
2489
2490 ptr += 1 + LINK_SIZE;
2491 }
2492 }
2493
2494
2495
2496 /*************************************************
2497 * Insert an automatic callout point *
2498 *************************************************/
2499
2500 /* This function is called when the PCRE_AUTO_CALLOUT option is set, to insert
2501 callout points before each pattern item.
2502
2503 Arguments:
2504 code current code pointer
2505 ptr current pattern pointer
2506 cd pointers to tables etc
2507
2508 Returns: new code pointer
2509 */
2510
2511 static uschar *
2512 auto_callout(uschar *code, const uschar *ptr, compile_data *cd)
2513 {
2514 *code++ = OP_CALLOUT;
2515 *code++ = 255;
2516 PUT(code, 0, (int)(ptr - cd->start_pattern)); /* Pattern offset */
2517 PUT(code, LINK_SIZE, 0); /* Default length */
2518 return code + 2*LINK_SIZE;
2519 }
2520
2521
2522
2523 /*************************************************
2524 * Complete a callout item *
2525 *************************************************/
2526
2527 /* A callout item contains the length of the next item in the pattern, which
2528 we can't fill in till after we have reached the relevant point. This is used
2529 for both automatic and manual callouts.
2530
2531 Arguments:
2532 previous_callout points to previous callout item
2533 ptr current pattern pointer
2534 cd pointers to tables etc
2535
2536 Returns: nothing
2537 */
2538
2539 static void
2540 complete_callout(uschar *previous_callout, const uschar *ptr, compile_data *cd)
2541 {
2542 int length = (int)(ptr - cd->start_pattern - GET(previous_callout, 2));
2543 PUT(previous_callout, 2 + LINK_SIZE, length);
2544 }
2545
2546
2547
2548 #ifdef SUPPORT_UCP
2549 /*************************************************
2550 * Get othercase range *
2551 *************************************************/
2552
2553 /* This function is passed the start and end of a class range, in UTF-8 mode
2554 with UCP support. It searches up the characters, looking for internal ranges of
2555 characters in the "other" case. Each call returns the next one, updating the
2556 start address.
2557
2558 Arguments:
2559 cptr points to starting character value; updated
2560 d end value
2561 ocptr where to put start of othercase range
2562 odptr where to put end of othercase range
2563
2564 Yield: TRUE when range returned; FALSE when no more
2565 */
2566
2567 static BOOL
2568 get_othercase_range(unsigned int *cptr, unsigned int d, unsigned int *ocptr,
2569 unsigned int *odptr)
2570 {
2571 unsigned int c, othercase, next;
2572
2573 for (c = *cptr; c <= d; c++)
2574 { if ((othercase = UCD_OTHERCASE(c)) != c) break; }
2575
2576 if (c > d) return FALSE;
2577
2578 *ocptr = othercase;
2579 next = othercase + 1;
2580
2581 for (++c; c <= d; c++)
2582 {
2583 if (UCD_OTHERCASE(c) != next) break;
2584 next++;
2585 }
2586
2587 *odptr = next - 1;
2588 *cptr = c;
2589
2590 return TRUE;
2591 }
2592
2593
2594
2595 /*************************************************
2596 * Check a character and a property *
2597 *************************************************/
2598
2599 /* This function is called by check_auto_possessive() when a property item
2600 is adjacent to a fixed character.
2601
2602 Arguments:
2603 c the character
2604 ptype the property type
2605 pdata the data for the type
2606 negated TRUE if it's a negated property (\P or \p{^)
2607
2608 Returns: TRUE if auto-possessifying is OK
2609 */
2610
2611 static BOOL
2612 check_char_prop(int c, int ptype, int pdata, BOOL negated)
2613 {
2614 const ucd_record *prop = GET_UCD(c);
2615 switch(ptype)
2616 {
2617 case PT_LAMP:
2618 return (prop->chartype == ucp_Lu ||
2619 prop->chartype == ucp_Ll ||
2620 prop->chartype == ucp_Lt) == negated;
2621
2622 case PT_GC:
2623 return (pdata == _pcre_ucp_gentype[prop->chartype]) == negated;
2624
2625 case PT_PC:
2626 return (pdata == prop->chartype) == negated;
2627
2628 case PT_SC:
2629 return (pdata == prop->script) == negated;
2630
2631 /* These are specials */
2632
2633 case PT_ALNUM:
2634 return (_pcre_ucp_gentype[prop->chartype] == ucp_L ||
2635 _pcre_ucp_gentype[prop->chartype] == ucp_N) == negated;
2636
2637 case PT_SPACE: /* Perl space */
2638 return (_pcre_ucp_gentype[prop->chartype] == ucp_Z ||
2639 c == CHAR_HT || c == CHAR_NL || c == CHAR_FF || c == CHAR_CR)
2640 == negated;
2641
2642 case PT_PXSPACE: /* POSIX space */
2643 return (_pcre_ucp_gentype[prop->chartype] == ucp_Z ||
2644 c == CHAR_HT || c == CHAR_NL || c == CHAR_VT ||
2645 c == CHAR_FF || c == CHAR_CR)
2646 == negated;
2647
2648 case PT_WORD:
2649 return (_pcre_ucp_gentype[prop->chartype] == ucp_L ||
2650 _pcre_ucp_gentype[prop->chartype] == ucp_N ||
2651 c == CHAR_UNDERSCORE) == negated;
2652 }
2653 return FALSE;
2654 }
2655 #endif /* SUPPORT_UCP */
2656
2657
2658
2659 /*************************************************
2660 * Check if auto-possessifying is possible *
2661 *************************************************/
2662
2663 /* This function is called for unlimited repeats of certain items, to see
2664 whether the next thing could possibly match the repeated item. If not, it makes
2665 sense to automatically possessify the repeated item.
2666
2667 Arguments:
2668 previous pointer to the repeated opcode
2669 utf8 TRUE in UTF-8 mode
2670 ptr next character in pattern
2671 options options bits
2672 cd contains pointers to tables etc.
2673
2674 Returns: TRUE if possessifying is wanted
2675 */
2676
2677 static BOOL
2678 check_auto_possessive(const uschar *previous, BOOL utf8, const uschar *ptr,
2679 int options, compile_data *cd)
2680 {
2681 int c, next;
2682 int op_code = *previous++;
2683
2684 /* Skip whitespace and comments in extended mode */
2685
2686 if ((options & PCRE_EXTENDED) != 0)
2687 {
2688 for (;;)
2689 {
2690 while ((cd->ctypes[*ptr] & ctype_space) != 0) ptr++;
2691 if (*ptr == CHAR_NUMBER_SIGN)
2692 {
2693 ptr++;
2694 while (*ptr != 0)
2695 {
2696 if (IS_NEWLINE(ptr)) { ptr += cd->nllen; break; }
2697 ptr++;
2698 #ifdef SUPPORT_UTF8
2699 if (utf8) while ((*ptr & 0xc0) == 0x80) ptr++;
2700 #endif
2701 }
2702 }
2703 else break;
2704 }
2705 }
2706
2707 /* If the next item is one that we can handle, get its value. A non-negative
2708 value is a character, a negative value is an escape value. */
2709
2710 if (*ptr == CHAR_BACKSLASH)
2711 {
2712 int temperrorcode = 0;
2713 next = check_escape(&ptr, &temperrorcode, cd->bracount, options, FALSE);
2714 if (temperrorcode != 0) return FALSE;
2715 ptr++; /* Point after the escape sequence */
2716 }
2717
2718 else if ((cd->ctypes[*ptr] & ctype_meta) == 0)
2719 {
2720 #ifdef SUPPORT_UTF8
2721 if (utf8) { GETCHARINC(next, ptr); } else
2722 #endif
2723 next = *ptr++;
2724 }
2725
2726 else return FALSE;
2727
2728 /* Skip whitespace and comments in extended mode */
2729
2730 if ((options & PCRE_EXTENDED) != 0)
2731 {
2732 for (;;)
2733 {
2734 while ((cd->ctypes[*ptr] & ctype_space) != 0) ptr++;
2735 if (*ptr == CHAR_NUMBER_SIGN)
2736 {
2737 ptr++;
2738 while (*ptr != 0)
2739 {
2740 if (IS_NEWLINE(ptr)) { ptr += cd->nllen; break; }
2741 ptr++;
2742 #ifdef SUPPORT_UTF8
2743 if (utf8) while ((*ptr & 0xc0) == 0x80) ptr++;
2744 #endif
2745 }
2746 }
2747 else break;
2748 }
2749 }
2750
2751 /* If the next thing is itself optional, we have to give up. */
2752
2753 if (*ptr == CHAR_ASTERISK || *ptr == CHAR_QUESTION_MARK ||
2754 strncmp((char *)ptr, STR_LEFT_CURLY_BRACKET STR_0 STR_COMMA, 3) == 0)
2755 return FALSE;
2756
2757 /* Now compare the next item with the previous opcode. First, handle cases when
2758 the next item is a character. */
2759
2760 if (next >= 0) switch(op_code)
2761 {
2762 case OP_CHAR:
2763 #ifdef SUPPORT_UTF8
2764 GETCHARTEST(c, previous);
2765 #else
2766 c = *previous;
2767 #endif
2768 return c != next;
2769
2770 /* For CHARI (caseless character) we must check the other case. If we have
2771 Unicode property support, we can use it to test the other case of
2772 high-valued characters. */
2773
2774 case OP_CHARI:
2775 #ifdef SUPPORT_UTF8
2776 GETCHARTEST(c, previous);
2777 #else
2778 c = *previous;
2779 #endif
2780 if (c == next) return FALSE;
2781 #ifdef SUPPORT_UTF8
2782 if (utf8)
2783 {
2784 unsigned int othercase;
2785 if (next < 128) othercase = cd->fcc[next]; else
2786 #ifdef SUPPORT_UCP
2787 othercase = UCD_OTHERCASE((unsigned int)next);
2788 #else
2789 othercase = NOTACHAR;
2790 #endif
2791 return (unsigned int)c != othercase;
2792 }
2793 else
2794 #endif /* SUPPORT_UTF8 */
2795 return (c != cd->fcc[next]); /* Non-UTF-8 mode */
2796
2797 /* For OP_NOT and OP_NOTI, the data is always a single-byte character. These
2798 opcodes are not used for multi-byte characters, because they are coded using
2799 an XCLASS instead. */
2800
2801 case OP_NOT:
2802 return (c = *previous) == next;
2803
2804 case OP_NOTI:
2805 if ((c = *previous) == next) return TRUE;
2806 #ifdef SUPPORT_UTF8
2807 if (utf8)
2808 {
2809 unsigned int othercase;
2810 if (next < 128) othercase = cd->fcc[next]; else
2811 #ifdef SUPPORT_UCP
2812 othercase = UCD_OTHERCASE(next);
2813 #else
2814 othercase = NOTACHAR;
2815 #endif
2816 return (unsigned int)c == othercase;
2817 }
2818 else
2819 #endif /* SUPPORT_UTF8 */
2820 return (c == cd->fcc[next]); /* Non-UTF-8 mode */
2821
2822 /* Note that OP_DIGIT etc. are generated only when PCRE_UCP is *not* set.
2823 When it is set, \d etc. are converted into OP_(NOT_)PROP codes. */
2824
2825 case OP_DIGIT:
2826 return next > 127 || (cd->ctypes[next] & ctype_digit) == 0;
2827
2828 case OP_NOT_DIGIT:
2829 return next <= 127 && (cd->ctypes[next] & ctype_digit) != 0;
2830
2831 case OP_WHITESPACE:
2832 return next > 127 || (cd->ctypes[next] & ctype_space) == 0;
2833
2834 case OP_NOT_WHITESPACE:
2835 return next <= 127 && (cd->ctypes[next] & ctype_space) != 0;
2836
2837 case OP_WORDCHAR:
2838 return next > 127 || (cd->ctypes[next] & ctype_word) == 0;
2839
2840 case OP_NOT_WORDCHAR:
2841 return next <= 127 && (cd->ctypes[next] & ctype_word) != 0;
2842
2843 case OP_HSPACE:
2844 case OP_NOT_HSPACE:
2845 switch(next)
2846 {
2847 case 0x09:
2848 case 0x20:
2849 case 0xa0:
2850 case 0x1680:
2851 case 0x180e:
2852 case 0x2000:
2853 case 0x2001:
2854 case 0x2002:
2855 case 0x2003:
2856 case 0x2004:
2857 case 0x2005:
2858 case 0x2006:
2859 case 0x2007:
2860 case 0x2008:
2861 case 0x2009:
2862 case 0x200A:
2863 case 0x202f:
2864 case 0x205f:
2865 case 0x3000:
2866 return op_code == OP_NOT_HSPACE;
2867 default:
2868 return op_code != OP_NOT_HSPACE;
2869 }
2870
2871 case OP_ANYNL:
2872 case OP_VSPACE:
2873 case OP_NOT_VSPACE:
2874 switch(next)
2875 {
2876 case 0x0a:
2877 case 0x0b:
2878 case 0x0c:
2879 case 0x0d:
2880 case 0x85:
2881 case 0x2028:
2882 case 0x2029:
2883 return op_code == OP_NOT_VSPACE;
2884 default:
2885 return op_code != OP_NOT_VSPACE;
2886 }
2887
2888 #ifdef SUPPORT_UCP
2889 case OP_PROP:
2890 return check_char_prop(next, previous[0], previous[1], FALSE);
2891
2892 case OP_NOTPROP:
2893 return check_char_prop(next, previous[0], previous[1], TRUE);
2894 #endif
2895
2896 default:
2897 return FALSE;
2898 }
2899
2900
2901 /* Handle the case when the next item is \d, \s, etc. Note that when PCRE_UCP
2902 is set, \d turns into ESC_du rather than ESC_d, etc., so ESC_d etc. are
2903 generated only when PCRE_UCP is *not* set, that is, when only ASCII
2904 characteristics are recognized. Similarly, the opcodes OP_DIGIT etc. are
2905 replaced by OP_PROP codes when PCRE_UCP is set. */
2906
2907 switch(op_code)
2908 {
2909 case OP_CHAR:
2910 case OP_CHARI:
2911 #ifdef SUPPORT_UTF8
2912 GETCHARTEST(c, previous);
2913 #else
2914 c = *previous;
2915 #endif
2916 switch(-next)
2917 {
2918 case ESC_d:
2919 return c > 127 || (cd->ctypes[c] & ctype_digit) == 0;
2920
2921 case ESC_D:
2922 return c <= 127 && (cd->ctypes[c] & ctype_digit) != 0;
2923
2924 case ESC_s:
2925 return c > 127 || (cd->ctypes[c] & ctype_space) == 0;
2926
2927 case ESC_S:
2928 return c <= 127 && (cd->ctypes[c] & ctype_space) != 0;
2929
2930 case ESC_w:
2931 return c > 127 || (cd->ctypes[c] & ctype_word) == 0;
2932
2933 case ESC_W:
2934 return c <= 127 && (cd->ctypes[c] & ctype_word) != 0;
2935
2936 case ESC_h:
2937 case ESC_H:
2938 switch(c)
2939 {
2940 case 0x09:
2941 case 0x20:
2942 case 0xa0:
2943 case 0x1680:
2944 case 0x180e:
2945 case 0x2000:
2946 case 0x2001:
2947 case 0x2002:
2948 case 0x2003:
2949 case 0x2004:
2950 case 0x2005:
2951 case 0x2006:
2952 case 0x2007:
2953 case 0x2008:
2954 case 0x2009:
2955 case 0x200A:
2956 case 0x202f:
2957 case 0x205f:
2958 case 0x3000:
2959 return -next != ESC_h;
2960 default:
2961 return -next == ESC_h;
2962 }
2963
2964 case ESC_v:
2965 case ESC_V:
2966 switch(c)
2967 {
2968 case 0x0a:
2969 case 0x0b:
2970 case 0x0c:
2971 case 0x0d:
2972 case 0x85:
2973 case 0x2028:
2974 case 0x2029:
2975 return -next != ESC_v;
2976 default:
2977 return -next == ESC_v;
2978 }
2979
2980 /* When PCRE_UCP is set, these values get generated for \d etc. Find
2981 their substitutions and process them. The result will always be either
2982 -ESC_p or -ESC_P. Then fall through to process those values. */
2983
2984 #ifdef SUPPORT_UCP
2985 case ESC_du:
2986 case ESC_DU:
2987 case ESC_wu:
2988 case ESC_WU:
2989 case ESC_su:
2990 case ESC_SU:
2991 {
2992 int temperrorcode = 0;
2993 ptr = substitutes[-next - ESC_DU];
2994 next = check_escape(&ptr, &temperrorcode, 0, options, FALSE);
2995 if (temperrorcode != 0) return FALSE;
2996 ptr++; /* For compatibility */
2997 }
2998 /* Fall through */
2999
3000 case ESC_p:
3001 case ESC_P:
3002 {
3003 int ptype, pdata, errorcodeptr;
3004 BOOL negated;
3005
3006 ptr--; /* Make ptr point at the p or P */
3007 ptype = get_ucp(&ptr, &negated, &pdata, &errorcodeptr);
3008 if (ptype < 0) return FALSE;
3009 ptr++; /* Point past the final curly ket */
3010
3011 /* If the property item is optional, we have to give up. (When generated
3012 from \d etc by PCRE_UCP, this test will have been applied much earlier,
3013 to the original \d etc. At this point, ptr will point to a zero byte. */
3014
3015 if (*ptr == CHAR_ASTERISK || *ptr == CHAR_QUESTION_MARK ||
3016 strncmp((char *)ptr, STR_LEFT_CURLY_BRACKET STR_0 STR_COMMA, 3) == 0)
3017 return FALSE;
3018
3019 /* Do the property check. */
3020
3021 return check_char_prop(c, ptype, pdata, (next == -ESC_P) != negated);
3022 }
3023 #endif
3024
3025 default:
3026 return FALSE;
3027 }
3028
3029 /* In principle, support for Unicode properties should be integrated here as
3030 well. It means re-organizing the above code so as to get hold of the property
3031 values before switching on the op-code. However, I wonder how many patterns
3032 combine ASCII \d etc with Unicode properties? (Note that if PCRE_UCP is set,
3033 these op-codes are never generated.) */
3034
3035 case OP_DIGIT:
3036 return next == -ESC_D || next == -ESC_s || next == -ESC_W ||
3037 next == -ESC_h || next == -ESC_v || next == -ESC_R;
3038
3039 case OP_NOT_DIGIT:
3040 return next == -ESC_d;
3041
3042 case OP_WHITESPACE:
3043 return next == -ESC_S || next == -ESC_d || next == -ESC_w || next == -ESC_R;
3044
3045 case OP_NOT_WHITESPACE:
3046 return next == -ESC_s || next == -ESC_h || next == -ESC_v;
3047
3048 case OP_HSPACE:
3049 return next == -ESC_S || next == -ESC_H || next == -ESC_d ||
3050 next == -ESC_w || next == -ESC_v || next == -ESC_R;
3051
3052 case OP_NOT_HSPACE:
3053 return next == -ESC_h;
3054
3055 /* Can't have \S in here because VT matches \S (Perl anomaly) */
3056 case OP_ANYNL:
3057 case OP_VSPACE:
3058 return next == -ESC_V || next == -ESC_d || next == -ESC_w;
3059
3060 case OP_NOT_VSPACE:
3061 return next == -ESC_v || next == -ESC_R;
3062
3063 case OP_WORDCHAR:
3064 return next == -ESC_W || next == -ESC_s || next == -ESC_h ||
3065 next == -ESC_v || next == -ESC_R;
3066
3067 case OP_NOT_WORDCHAR:
3068 return next == -ESC_w || next == -ESC_d;
3069
3070 default:
3071 return FALSE;
3072 }
3073
3074 /* Control does not reach here */
3075 }
3076
3077
3078
3079 /*************************************************
3080 * Compile one branch *
3081 *************************************************/
3082
3083 /* Scan the pattern, compiling it into the a vector. If the options are
3084 changed during the branch, the pointer is used to change the external options
3085 bits. This function is used during the pre-compile phase when we are trying
3086 to find out the amount of memory needed, as well as during the real compile
3087 phase. The value of lengthptr distinguishes the two phases.
3088
3089 Arguments:
3090 optionsptr pointer to the option bits
3091 codeptr points to the pointer to the current code point
3092 ptrptr points to the current pattern pointer
3093 errorcodeptr points to error code variable
3094 firstbyteptr set to initial literal character, or < 0 (REQ_UNSET, REQ_NONE)
3095 reqbyteptr set to the last literal character required, else < 0
3096 bcptr points to current branch chain
3097 cond_depth conditional nesting depth
3098 cd contains pointers to tables etc.
3099 lengthptr NULL during the real compile phase
3100 points to length accumulator during pre-compile phase
3101
3102 Returns: TRUE on success
3103 FALSE, with *errorcodeptr set non-zero on error
3104 */
3105
3106 static BOOL
3107 compile_branch(int *optionsptr, uschar **codeptr, const uschar **ptrptr,
3108 int *errorcodeptr, int *firstbyteptr, int *reqbyteptr, branch_chain *bcptr,
3109 int cond_depth, compile_data *cd, int *lengthptr)
3110 {
3111 int repeat_type, op_type;
3112 int repeat_min = 0, repeat_max = 0; /* To please picky compilers */
3113 int bravalue = 0;
3114 int greedy_default, greedy_non_default;
3115 int firstbyte, reqbyte;
3116 int zeroreqbyte, zerofirstbyte;
3117 int req_caseopt, reqvary, tempreqvary;
3118 int options = *optionsptr; /* May change dynamically */
3119 int after_manual_callout = 0;
3120 int length_prevgroup = 0;
3121 register int c;
3122 register uschar *code = *codeptr;
3123 uschar *last_code = code;
3124 uschar *orig_code = code;
3125 uschar *tempcode;
3126 BOOL inescq = FALSE;
3127 BOOL groupsetfirstbyte = FALSE;
3128 const uschar *ptr = *ptrptr;
3129 const uschar *tempptr;
3130 const uschar *nestptr = NULL;
3131 uschar *previous = NULL;
3132 uschar *previous_callout = NULL;
3133 uschar *save_hwm = NULL;
3134 uschar classbits[32];
3135
3136 /* We can fish out the UTF-8 setting once and for all into a BOOL, but we
3137 must not do this for other options (e.g. PCRE_EXTENDED) because they may change
3138 dynamically as we process the pattern. */
3139
3140 #ifdef SUPPORT_UTF8
3141 BOOL class_utf8;
3142 BOOL utf8 = (options & PCRE_UTF8) != 0;
3143 uschar *class_utf8data;
3144 uschar *class_utf8data_base;
3145 uschar utf8_char[6];
3146 #else
3147 BOOL utf8 = FALSE;
3148 #endif
3149
3150 #ifdef PCRE_DEBUG
3151 if (lengthptr != NULL) DPRINTF((">> start branch\n"));
3152 #endif
3153
3154 /* Set up the default and non-default settings for greediness */
3155
3156 greedy_default = ((options & PCRE_UNGREEDY) != 0);
3157 greedy_non_default = greedy_default ^ 1;
3158
3159 /* Initialize no first byte, no required byte. REQ_UNSET means "no char
3160 matching encountered yet". It gets changed to REQ_NONE if we hit something that
3161 matches a non-fixed char first char; reqbyte just remains unset if we never
3162 find one.
3163
3164 When we hit a repeat whose minimum is zero, we may have to adjust these values
3165 to take the zero repeat into account. This is implemented by setting them to
3166 zerofirstbyte and zeroreqbyte when such a repeat is encountered. The individual
3167 item types that can be repeated set these backoff variables appropriately. */
3168
3169 firstbyte = reqbyte = zerofirstbyte = zeroreqbyte = REQ_UNSET;
3170
3171 /* The variable req_caseopt contains either the REQ_CASELESS value or zero,
3172 according to the current setting of the caseless flag. REQ_CASELESS is a bit
3173 value > 255. It is added into the firstbyte or reqbyte variables to record the
3174 case status of the value. This is used only for ASCII characters. */
3175
3176 req_caseopt = ((options & PCRE_CASELESS) != 0)? REQ_CASELESS : 0;
3177
3178 /* Switch on next character until the end of the branch */
3179
3180 for (;; ptr++)
3181 {
3182 BOOL negate_class;
3183 BOOL should_flip_negation;
3184 BOOL possessive_quantifier;
3185 BOOL is_quantifier;
3186 BOOL is_recurse;
3187 BOOL reset_bracount;
3188 int class_charcount;
3189 int class_lastchar;
3190 int newoptions;
3191 int recno;
3192 int refsign;
3193 int skipbytes;
3194 int subreqbyte;
3195 int subfirstbyte;
3196 int terminator;
3197 int mclength;
3198 int tempbracount;
3199 uschar mcbuffer[8];
3200
3201 /* Get next byte in the pattern */
3202
3203 c = *ptr;
3204
3205 /* If we are at the end of a nested substitution, revert to the outer level
3206 string. Nesting only happens one level deep. */
3207
3208 if (c == 0 && nestptr != NULL)
3209 {
3210 ptr = nestptr;
3211 nestptr = NULL;
3212 c = *ptr;
3213 }
3214
3215 /* If we are in the pre-compile phase, accumulate the length used for the
3216 previous cycle of this loop. */
3217
3218 if (lengthptr != NULL)
3219 {
3220 #ifdef PCRE_DEBUG
3221 if (code > cd->hwm) cd->hwm = code; /* High water info */
3222 #endif
3223 if (code > cd->start_workspace + WORK_SIZE_CHECK) /* Check for overrun */
3224 {
3225 *errorcodeptr = ERR52;
3226 goto FAILED;
3227 }
3228
3229 /* There is at least one situation where code goes backwards: this is the
3230 case of a zero quantifier after a class (e.g. [ab]{0}). At compile time,
3231 the class is simply eliminated. However, it is created first, so we have to
3232 allow memory for it. Therefore, don't ever reduce the length at this point.
3233 */
3234
3235 if (code < last_code) code = last_code;
3236
3237 /* Paranoid check for integer overflow */
3238
3239 if (OFLOW_MAX - *lengthptr < code - last_code)
3240 {
3241 *errorcodeptr = ERR20;
3242 goto FAILED;
3243 }
3244
3245 *lengthptr += (int)(code - last_code);
3246 DPRINTF(("length=%d added %d c=%c\n", *lengthptr, code - last_code, c));
3247
3248 /* If "previous" is set and it is not at the start of the work space, move
3249 it back to there, in order to avoid filling up the work space. Otherwise,
3250 if "previous" is NULL, reset the current code pointer to the start. */
3251
3252 if (previous != NULL)
3253 {
3254 if (previous > orig_code)
3255 {
3256 memmove(orig_code, previous, code - previous);
3257 code -= previous - orig_code;
3258 previous = orig_code;
3259 }
3260 }
3261 else code = orig_code;
3262
3263 /* Remember where this code item starts so we can pick up the length
3264 next time round. */
3265
3266 last_code = code;
3267 }
3268
3269 /* In the real compile phase, just check the workspace used by the forward
3270 reference list. */
3271
3272 else if (cd->hwm > cd->start_workspace + WORK_SIZE_CHECK)
3273 {
3274 *errorcodeptr = ERR52;
3275 goto FAILED;
3276 }
3277
3278 /* If in \Q...\E, check for the end; if not, we have a literal */
3279
3280 if (inescq && c != 0)
3281 {
3282 if (c == CHAR_BACKSLASH && ptr[1] == CHAR_E)
3283 {
3284 inescq = FALSE;
3285 ptr++;
3286 continue;
3287 }
3288 else
3289 {
3290 if (previous_callout != NULL)
3291 {
3292 if (lengthptr == NULL) /* Don't attempt in pre-compile phase */
3293 complete_callout(previous_callout, ptr, cd);
3294 previous_callout = NULL;
3295 }
3296 if ((options & PCRE_AUTO_CALLOUT) != 0)
3297 {
3298 previous_callout = code;
3299 code = auto_callout(code, ptr, cd);
3300 }
3301 goto NORMAL_CHAR;
3302 }
3303 }
3304
3305 /* Fill in length of a previous callout, except when the next thing is
3306 a quantifier. */
3307
3308 is_quantifier =
3309 c == CHAR_ASTERISK || c == CHAR_PLUS || c == CHAR_QUESTION_MARK ||
3310 (c == CHAR_LEFT_CURLY_BRACKET && is_counted_repeat(ptr+1));
3311
3312 if (!is_quantifier && previous_callout != NULL &&
3313 after_manual_callout-- <= 0)
3314 {
3315 if (lengthptr == NULL) /* Don't attempt in pre-compile phase */
3316 complete_callout(previous_callout, ptr, cd);
3317 previous_callout = NULL;
3318 }
3319
3320 /* In extended mode, skip white space and comments. */
3321
3322 if ((options & PCRE_EXTENDED) != 0)
3323 {
3324 if ((cd->ctypes[c] & ctype_space) != 0) continue;
3325 if (c == CHAR_NUMBER_SIGN)
3326 {
3327 ptr++;
3328 while (*ptr != 0)
3329 {
3330 if (IS_NEWLINE(ptr)) { ptr += cd->nllen - 1; break; }
3331 ptr++;
3332 #ifdef SUPPORT_UTF8
3333 if (utf8) while ((*ptr & 0xc0) == 0x80) ptr++;
3334 #endif
3335 }
3336 if (*ptr != 0) continue;
3337
3338 /* Else fall through to handle end of string */
3339 c = 0;
3340 }
3341 }
3342
3343 /* No auto callout for quantifiers. */
3344
3345 if ((options & PCRE_AUTO_CALLOUT) != 0 && !is_quantifier)
3346 {
3347 previous_callout = code;
3348 code = auto_callout(code, ptr, cd);
3349 }
3350
3351 switch(c)
3352 {
3353 /* ===================================================================*/
3354 case 0: /* The branch terminates at string end */
3355 case CHAR_VERTICAL_LINE: /* or | or ) */
3356 case CHAR_RIGHT_PARENTHESIS:
3357 *firstbyteptr = firstbyte;
3358 *reqbyteptr = reqbyte;
3359 *codeptr = code;
3360 *ptrptr = ptr;
3361 if (lengthptr != NULL)
3362 {
3363 if (OFLOW_MAX - *lengthptr < code - last_code)
3364 {
3365 *errorcodeptr = ERR20;
3366 goto FAILED;
3367 }
3368 *lengthptr += (int)(code - last_code); /* To include callout length */
3369 DPRINTF((">> end branch\n"));
3370 }
3371 return TRUE;
3372
3373
3374 /* ===================================================================*/
3375 /* Handle single-character metacharacters. In multiline mode, ^ disables
3376 the setting of any following char as a first character. */
3377
3378 case CHAR_CIRCUMFLEX_ACCENT:
3379 previous = NULL;
3380 if ((options & PCRE_MULTILINE) != 0)
3381 {
3382 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
3383 *code++ = OP_CIRCM;
3384 }
3385 else *code++ = OP_CIRC;
3386 break;
3387
3388 case CHAR_DOLLAR_SIGN:
3389 previous = NULL;
3390 *code++ = ((options & PCRE_MULTILINE) != 0)? OP_DOLLM : OP_DOLL;
3391 break;
3392
3393 /* There can never be a first char if '.' is first, whatever happens about
3394 repeats. The value of reqbyte doesn't change either. */
3395
3396 case CHAR_DOT:
3397 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
3398 zerofirstbyte = firstbyte;
3399 zeroreqbyte = reqbyte;
3400 previous = code;
3401 *code++ = ((options & PCRE_DOTALL) != 0)? OP_ALLANY: OP_ANY;
3402 break;
3403
3404
3405 /* ===================================================================*/
3406 /* Character classes. If the included characters are all < 256, we build a
3407 32-byte bitmap of the permitted characters, except in the special case
3408 where there is only one such character. For negated classes, we build the
3409 map as usual, then invert it at the end. However, we use a different opcode
3410 so that data characters > 255 can be handled correctly.
3411
3412 If the class contains characters outside the 0-255 range, a different
3413 opcode is compiled. It may optionally have a bit map for characters < 256,
3414 but those above are are explicitly listed afterwards. A flag byte tells
3415 whether the bitmap is present, and whether this is a negated class or not.
3416
3417 In JavaScript compatibility mode, an isolated ']' causes an error. In
3418 default (Perl) mode, it is treated as a data character. */
3419
3420 case CHAR_RIGHT_SQUARE_BRACKET:
3421 if ((cd->external_options & PCRE_JAVASCRIPT_COMPAT) != 0)
3422 {
3423 *errorcodeptr = ERR64;
3424 goto FAILED;
3425 }
3426 goto NORMAL_CHAR;
3427
3428 case CHAR_LEFT_SQUARE_BRACKET:
3429 previous = code;
3430
3431 /* PCRE supports POSIX class stuff inside a class. Perl gives an error if
3432 they are encountered at the top level, so we'll do that too. */
3433
3434 if ((ptr[1] == CHAR_COLON || ptr[1] == CHAR_DOT ||
3435 ptr[1] == CHAR_EQUALS_SIGN) &&
3436 check_posix_syntax(ptr, &tempptr))
3437 {
3438 *errorcodeptr = (ptr[1] == CHAR_COLON)? ERR13 : ERR31;
3439 goto FAILED;
3440 }
3441
3442 /* If the first character is '^', set the negation flag and skip it. Also,
3443 if the first few characters (either before or after ^) are \Q\E or \E we
3444 skip them too. This makes for compatibility with Perl. */
3445
3446 negate_class = FALSE;
3447 for (;;)
3448 {
3449 c = *(++ptr);
3450 if (c == CHAR_BACKSLASH)
3451 {
3452 if (ptr[1] == CHAR_E)
3453 ptr++;
3454 else if (strncmp((const char *)ptr+1,
3455 STR_Q STR_BACKSLASH STR_E, 3) == 0)
3456 ptr += 3;
3457 else
3458 break;
3459 }
3460 else if (!negate_class && c == CHAR_CIRCUMFLEX_ACCENT)
3461 negate_class = TRUE;
3462 else break;
3463 }
3464
3465 /* Empty classes are allowed in JavaScript compatibility mode. Otherwise,
3466 an initial ']' is taken as a data character -- the code below handles
3467 that. In JS mode, [] must always fail, so generate OP_FAIL, whereas
3468 [^] must match any character, so generate OP_ALLANY. */
3469
3470 if (c == CHAR_RIGHT_SQUARE_BRACKET &&
3471 (cd->external_options & PCRE_JAVASCRIPT_COMPAT) != 0)
3472 {
3473 *code++ = negate_class? OP_ALLANY : OP_FAIL;
3474 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
3475 zerofirstbyte = firstbyte;
3476 break;
3477 }
3478
3479 /* If a class contains a negative special such as \S, we need to flip the
3480 negation flag at the end, so that support for characters > 255 works
3481 correctly (they are all included in the class). */
3482
3483 should_flip_negation = FALSE;
3484
3485 /* Keep a count of chars with values < 256 so that we can optimize the case
3486 of just a single character (as long as it's < 256). However, For higher
3487 valued UTF-8 characters, we don't yet do any optimization. */
3488
3489 class_charcount = 0;
3490 class_lastchar = -1;
3491
3492 /* Initialize the 32-char bit map to all zeros. We build the map in a
3493 temporary bit of memory, in case the class contains only 1 character (less
3494 than 256), because in that case the compiled code doesn't use the bit map.
3495 */
3496
3497 memset(classbits, 0, 32 * sizeof(uschar));
3498
3499 #ifdef SUPPORT_UTF8
3500 class_utf8 = FALSE; /* No chars >= 256 */
3501 class_utf8data = code + LINK_SIZE + 2; /* For UTF-8 items */
3502 class_utf8data_base = class_utf8data; /* For resetting in pass 1 */
3503 #endif
3504
3505 /* Process characters until ] is reached. By writing this as a "do" it
3506 means that an initial ] is taken as a data character. At the start of the
3507 loop, c contains the first byte of the character. */
3508
3509 if (c != 0) do
3510 {
3511 const uschar *oldptr;
3512
3513 #ifdef SUPPORT_UTF8
3514 if (utf8 && c > 127)
3515 { /* Braces are required because the */
3516 GETCHARLEN(c, ptr, ptr); /* macro generates multiple statements */
3517 }
3518
3519 /* In the pre-compile phase, accumulate the length of any UTF-8 extra
3520 data and reset the pointer. This is so that very large classes that
3521 contain a zillion UTF-8 characters no longer overwrite the work space
3522 (which is on the stack). */
3523
3524 if (lengthptr != NULL)
3525 {
3526 *lengthptr += class_utf8data - class_utf8data_base;
3527 class_utf8data = class_utf8data_base;
3528 }
3529
3530 #endif
3531
3532 /* Inside \Q...\E everything is literal except \E */
3533
3534 if (inescq)
3535 {
3536 if (c == CHAR_BACKSLASH && ptr[1] == CHAR_E) /* If we are at \E */
3537 {
3538 inescq = FALSE; /* Reset literal state */
3539 ptr++; /* Skip the 'E' */
3540 continue; /* Carry on with next */
3541 }
3542 goto CHECK_RANGE; /* Could be range if \E follows */
3543 }
3544
3545 /* Handle POSIX class names. Perl allows a negation extension of the
3546 form [:^name:]. A square bracket that doesn't match the syntax is
3547 treated as a literal. We also recognize the POSIX constructions
3548 [.ch.] and [=ch=] ("collating elements") and fault them, as Perl
3549 5.6 and 5.8 do. */
3550
3551 if (c == CHAR_LEFT_SQUARE_BRACKET &&
3552 (ptr[1] == CHAR_COLON || ptr[1] == CHAR_DOT ||
3553 ptr[1] == CHAR_EQUALS_SIGN) && check_posix_syntax(ptr, &tempptr))
3554 {
3555 BOOL local_negate = FALSE;
3556 int posix_class, taboffset, tabopt;
3557 register const uschar *cbits = cd->cbits;
3558 uschar pbits[32];
3559
3560 if (ptr[1] != CHAR_COLON)
3561 {
3562 *errorcodeptr = ERR31;
3563 goto FAILED;
3564 }
3565
3566 ptr += 2;
3567 if (*ptr == CHAR_CIRCUMFLEX_ACCENT)
3568 {
3569 local_negate = TRUE;
3570 should_flip_negation = TRUE; /* Note negative special */
3571 ptr++;
3572 }
3573
3574 posix_class = check_posix_name(ptr, (int)(tempptr - ptr));
3575 if (posix_class < 0)
3576 {
3577 *errorcodeptr = ERR30;
3578 goto FAILED;
3579 }
3580
3581 /* If matching is caseless, upper and lower are converted to
3582 alpha. This relies on the fact that the class table starts with
3583 alpha, lower, upper as the first 3 entries. */
3584
3585 if ((options & PCRE_CASELESS) != 0 && posix_class <= 2)
3586 posix_class = 0;
3587
3588 /* When PCRE_UCP is set, some of the POSIX classes are converted to
3589 different escape sequences that use Unicode properties. */
3590
3591 #ifdef SUPPORT_UCP
3592 if ((options & PCRE_UCP) != 0)
3593 {
3594 int pc = posix_class + ((local_negate)? POSIX_SUBSIZE/2 : 0);
3595 if (posix_substitutes[pc] != NULL)
3596 {
3597 nestptr = tempptr + 1;
3598 ptr = posix_substitutes[pc] - 1;
3599 continue;
3600 }
3601 }
3602 #endif
3603 /* In the non-UCP case, we build the bit map for the POSIX class in a
3604 chunk of local store because we may be adding and subtracting from it,
3605 and we don't want to subtract bits that may be in the main map already.
3606 At the end we or the result into the bit map that is being built. */
3607
3608 posix_class *= 3;
3609
3610 /* Copy in the first table (always present) */
3611
3612 memcpy(pbits, cbits + posix_class_maps[posix_class],
3613 32 * sizeof(uschar));
3614
3615 /* If there is a second table, add or remove it as required. */
3616
3617 taboffset = posix_class_maps[posix_class + 1];
3618 tabopt = posix_class_maps[posix_class + 2];
3619
3620 if (taboffset >= 0)
3621 {
3622 if (tabopt >= 0)
3623 for (c = 0; c < 32; c++) pbits[c] |= cbits[c + taboffset];
3624 else
3625 for (c = 0; c < 32; c++) pbits[c] &= ~cbits[c + taboffset];
3626 }
3627
3628 /* Not see if we need to remove any special characters. An option
3629 value of 1 removes vertical space and 2 removes underscore. */
3630
3631 if (tabopt < 0) tabopt = -tabopt;
3632 if (tabopt == 1) pbits[1] &= ~0x3c;
3633 else if (tabopt == 2) pbits[11] &= 0x7f;
3634
3635 /* Add the POSIX table or its complement into the main table that is
3636 being built and we are done. */
3637
3638 if (local_negate)
3639 for (c = 0; c < 32; c++) classbits[c] |= ~pbits[c];
3640 else
3641 for (c = 0; c < 32; c++) classbits[c] |= pbits[c];
3642
3643 ptr = tempptr + 1;
3644 class_charcount = 10; /* Set > 1; assumes more than 1 per class */
3645 continue; /* End of POSIX syntax handling */
3646 }
3647
3648 /* Backslash may introduce a single character, or it may introduce one
3649 of the specials, which just set a flag. The sequence \b is a special
3650 case. Inside a class (and only there) it is treated as backspace. We
3651 assume that other escapes have more than one character in them, so set
3652 class_charcount bigger than one. Unrecognized escapes fall through and
3653 are either treated as literal characters (by default), or are faulted if
3654 PCRE_EXTRA is set. */
3655
3656 if (c == CHAR_BACKSLASH)
3657 {
3658 c = check_escape(&ptr, errorcodeptr, cd->bracount, options, TRUE);
3659 if (*errorcodeptr != 0) goto FAILED;
3660
3661 if (-c == ESC_b) c = CHAR_BS; /* \b is backspace in a class */
3662 else if (-c == ESC_Q) /* Handle start of quoted string */
3663 {
3664 if (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
3665 {
3666 ptr += 2; /* avoid empty string */
3667 }
3668 else inescq = TRUE;
3669 continue;
3670 }
3671 else if (-c == ESC_E) continue; /* Ignore orphan \E */
3672
3673 if (c < 0)
3674 {
3675 register const uschar *cbits = cd->cbits;
3676 class_charcount += 2; /* Greater than 1 is what matters */
3677
3678 switch (-c)
3679 {
3680 #ifdef SUPPORT_UCP
3681 case ESC_du: /* These are the values given for \d etc */
3682 case ESC_DU: /* when PCRE_UCP is set. We replace the */
3683 case ESC_wu: /* escape sequence with an appropriate \p */
3684 case ESC_WU: /* or \P to test Unicode properties instead */
3685 case ESC_su: /* of the default ASCII testing. */
3686 case ESC_SU:
3687 nestptr = ptr;
3688 ptr = substitutes[-c - ESC_DU] - 1; /* Just before substitute */
3689 class_charcount -= 2; /* Undo! */
3690 continue;
3691 #endif
3692 case ESC_d:
3693 for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_digit];
3694 continue;
3695
3696 case ESC_D:
3697 should_flip_negation = TRUE;
3698 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_digit];
3699 continue;
3700
3701 case ESC_w:
3702 for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_word];
3703 continue;
3704
3705 case ESC_W:
3706 should_flip_negation = TRUE;
3707 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_word];
3708 continue;
3709
3710 /* Perl 5.004 onwards omits VT from \s, but we must preserve it
3711 if it was previously set by something earlier in the character
3712 class. */
3713
3714 case ESC_s:
3715 classbits[0] |= cbits[cbit_space];
3716 classbits[1] |= cbits[cbit_space+1] & ~0x08;
3717 for (c = 2; c < 32; c++) classbits[c] |= cbits[c+cbit_space];
3718 continue;
3719
3720 case ESC_S:
3721 should_flip_negation = TRUE;
3722 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_space];
3723 classbits[1] |= 0x08; /* Perl 5.004 onwards omits VT from \s */
3724 continue;
3725
3726 case ESC_h:
3727 SETBIT(classbits, 0x09); /* VT */
3728 SETBIT(classbits, 0x20); /* SPACE */
3729 SETBIT(classbits, 0xa0); /* NSBP */
3730 #ifdef SUPPORT_UTF8
3731 if (utf8)
3732 {
3733 class_utf8 = TRUE;
3734 *class_utf8data++ = XCL_SINGLE;
3735 class_utf8data += _pcre_ord2utf8(0x1680, class_utf8data);
3736 *class_utf8data++ = XCL_SINGLE;
3737 class_utf8data += _pcre_ord2utf8(0x180e, class_utf8data);
3738 *class_utf8data++ = XCL_RANGE;
3739 class_utf8data += _pcre_ord2utf8(0x2000, class_utf8data);
3740 class_utf8data += _pcre_ord2utf8(0x200A, class_utf8data);
3741 *class_utf8data++ = XCL_SINGLE;
3742 class_utf8data += _pcre_ord2utf8(0x202f, class_utf8data);
3743 *class_utf8data++ = XCL_SINGLE;
3744 class_utf8data += _pcre_ord2utf8(0x205f, class_utf8data);
3745 *class_utf8data++ = XCL_SINGLE;
3746 class_utf8data += _pcre_ord2utf8(0x3000, class_utf8data);
3747 }
3748 #endif
3749 continue;
3750
3751 case ESC_H:
3752 for (c = 0; c < 32; c++)
3753 {
3754 int x = 0xff;
3755 switch (c)
3756 {
3757 case 0x09/8: x ^= 1 << (0x09%8); break;
3758 case 0x20/8: x ^= 1 << (0x20%8); break;
3759 case 0xa0/8: x ^= 1 << (0xa0%8); break;
3760 default: break;
3761 }
3762 classbits[c] |= x;
3763 }
3764
3765 #ifdef SUPPORT_UTF8
3766 if (utf8)
3767 {
3768 class_utf8 = TRUE;
3769 *class_utf8data++ = XCL_RANGE;
3770 class_utf8data += _pcre_ord2utf8(0x0100, class_utf8data);
3771 class_utf8data += _pcre_ord2utf8(0x167f, class_utf8data);
3772 *class_utf8data++ = XCL_RANGE;
3773 class_utf8data += _pcre_ord2utf8(0x1681, class_utf8data);
3774 class_utf8data += _pcre_ord2utf8(0x180d, class_utf8data);
3775 *class_utf8data++ = XCL_RANGE;
3776 class_utf8data += _pcre_ord2utf8(0x180f, class_utf8data);
3777 class_utf8data += _pcre_ord2utf8(0x1fff, class_utf8data);
3778 *class_utf8data++ = XCL_RANGE;
3779 class_utf8data += _pcre_ord2utf8(0x200B, class_utf8data);
3780 class_utf8data += _pcre_ord2utf8(0x202e, class_utf8data);
3781 *class_utf8data++ = XCL_RANGE;
3782 class_utf8data += _pcre_ord2utf8(0x2030, class_utf8data);
3783 class_utf8data += _pcre_ord2utf8(0x205e, class_utf8data);
3784 *class_utf8data++ = XCL_RANGE;
3785 class_utf8data += _pcre_ord2utf8(0x2060, class_utf8data);
3786 class_utf8data += _pcre_ord2utf8(0x2fff, class_utf8data);
3787 *class_utf8data++ = XCL_RANGE;
3788 class_utf8data += _pcre_ord2utf8(0x3001, class_utf8data);
3789 class_utf8data += _pcre_ord2utf8(0x7fffffff, class_utf8data);
3790 }
3791 #endif
3792 continue;
3793
3794 case ESC_v:
3795 SETBIT(classbits, 0x0a); /* LF */
3796 SETBIT(classbits, 0x0b); /* VT */
3797 SETBIT(classbits, 0x0c); /* FF */
3798 SETBIT(classbits, 0x0d); /* CR */
3799 SETBIT(classbits, 0x85); /* NEL */
3800 #ifdef SUPPORT_UTF8
3801 if (utf8)
3802 {
3803 class_utf8 = TRUE;
3804 *class_utf8data++ = XCL_RANGE;
3805 class_utf8data += _pcre_ord2utf8(0x2028, class_utf8data);
3806 class_utf8data += _pcre_ord2utf8(0x2029, class_utf8data);
3807 }
3808 #endif
3809 continue;
3810
3811 case ESC_V:
3812 for (c = 0; c < 32; c++)
3813 {
3814 int x = 0xff;
3815 switch (c)
3816 {
3817 case 0x0a/8: x ^= 1 << (0x0a%8);
3818 x ^= 1 << (0x0b%8);
3819 x ^= 1 << (0x0c%8);
3820 x ^= 1 << (0x0d%8);
3821 break;
3822 case 0x85/8: x ^= 1 << (0x85%8); break;
3823 default: break;
3824 }
3825 classbits[c] |= x;
3826 }
3827
3828 #ifdef SUPPORT_UTF8
3829 if (utf8)
3830 {
3831 class_utf8 = TRUE;
3832 *class_utf8data++ = XCL_RANGE;
3833 class_utf8data += _pcre_ord2utf8(0x0100, class_utf8data);
3834 class_utf8data += _pcre_ord2utf8(0x2027, class_utf8data);
3835 *class_utf8data++ = XCL_RANGE;
3836 class_utf8data += _pcre_ord2utf8(0x2029, class_utf8data);
3837 class_utf8data += _pcre_ord2utf8(0x7fffffff, class_utf8data);
3838 }
3839 #endif
3840 continue;
3841
3842 #ifdef SUPPORT_UCP
3843 case ESC_p:
3844 case ESC_P:
3845 {
3846 BOOL negated;
3847 int pdata;
3848 int ptype = get_ucp(&ptr, &negated, &pdata, errorcodeptr);
3849 if (ptype < 0) goto FAILED;
3850 class_utf8 = TRUE;
3851 *class_utf8data++ = ((-c == ESC_p) != negated)?
3852 XCL_PROP : XCL_NOTPROP;
3853 *class_utf8data++ = ptype;
3854 *class_utf8data++ = pdata;
3855 class_charcount -= 2; /* Not a < 256 character */
3856 continue;
3857 }
3858 #endif
3859 /* Unrecognized escapes are faulted if PCRE is running in its
3860 strict mode. By default, for compatibility with Perl, they are
3861 treated as literals. */
3862
3863 default:
3864 if ((options & PCRE_EXTRA) != 0)
3865 {
3866 *errorcodeptr = ERR7;
3867 goto FAILED;
3868 }
3869 class_charcount -= 2; /* Undo the default count from above */
3870 c = *ptr; /* Get the final character and fall through */
3871 break;
3872 }
3873 }
3874
3875 /* Fall through if we have a single character (c >= 0). This may be
3876 greater than 256 in UTF-8 mode. */
3877
3878 } /* End of backslash handling */
3879
3880 /* A single character may be followed by '-' to form a range. However,
3881 Perl does not permit ']' to be the end of the range. A '-' character
3882 at the end is treated as a literal. Perl ignores orphaned \E sequences
3883 entirely. The code for handling \Q and \E is messy. */
3884
3885 CHECK_RANGE:
3886 while (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
3887 {
3888 inescq = FALSE;
3889 ptr += 2;
3890 }
3891
3892 oldptr = ptr;
3893
3894 /* Remember \r or \n */
3895
3896 if (c == CHAR_CR || c == CHAR_NL) cd->external_flags |= PCRE_HASCRORLF;
3897
3898 /* Check for range */
3899
3900 if (!inescq && ptr[1] == CHAR_MINUS)
3901 {
3902 int d;
3903 ptr += 2;
3904 while (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_E) ptr += 2;
3905
3906 /* If we hit \Q (not followed by \E) at this point, go into escaped
3907 mode. */
3908
3909 while (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_Q)
3910 {
3911 ptr += 2;
3912 if (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_E)
3913 { ptr += 2; continue; }
3914 inescq = TRUE;
3915 break;
3916 }
3917
3918 if (*ptr == 0 || (!inescq && *ptr == CHAR_RIGHT_SQUARE_BRACKET))
3919 {
3920 ptr = oldptr;
3921 goto LONE_SINGLE_CHARACTER;
3922 }
3923
3924 #ifdef SUPPORT_UTF8
3925 if (utf8)
3926 { /* Braces are required because the */
3927 GETCHARLEN(d, ptr, ptr); /* macro generates multiple statements */
3928 }
3929 else
3930 #endif
3931 d = *ptr; /* Not UTF-8 mode */
3932
3933 /* The second part of a range can be a single-character escape, but
3934 not any of the other escapes. Perl 5.6 treats a hyphen as a literal
3935 in such circumstances. */
3936
3937 if (!inescq && d == CHAR_BACKSLASH)
3938 {
3939 d = check_escape(&ptr, errorcodeptr, cd->bracount, options, TRUE);
3940 if (*errorcodeptr != 0) goto FAILED;
3941
3942 /* \b is backspace; any other special means the '-' was literal */
3943
3944 if (d < 0)
3945 {
3946 if (d == -ESC_b) d = CHAR_BS; else
3947 {
3948 ptr = oldptr;
3949 goto LONE_SINGLE_CHARACTER; /* A few lines below */
3950 }
3951 }
3952 }
3953
3954 /* Check that the two values are in the correct order. Optimize
3955 one-character ranges */
3956
3957 if (d < c)
3958 {
3959 *errorcodeptr = ERR8;
3960 goto FAILED;
3961 }
3962
3963 if (d == c) goto LONE_SINGLE_CHARACTER; /* A few lines below */
3964
3965 /* Remember \r or \n */
3966
3967 if (d == CHAR_CR || d == CHAR_NL) cd->external_flags |= PCRE_HASCRORLF;
3968
3969 /* In UTF-8 mode, if the upper limit is > 255, or > 127 for caseless
3970 matching, we have to use an XCLASS with extra data items. Caseless
3971 matching for characters > 127 is available only if UCP support is
3972 available. */
3973
3974 #ifdef SUPPORT_UTF8
3975 if (utf8 && (d > 255 || ((options & PCRE_CASELESS) != 0 && d > 127)))
3976 {
3977 class_utf8 = TRUE;
3978
3979 /* With UCP support, we can find the other case equivalents of
3980 the relevant characters. There may be several ranges. Optimize how
3981 they fit with the basic range. */
3982
3983 #ifdef SUPPORT_UCP
3984 if ((options & PCRE_CASELESS) != 0)
3985 {
3986 unsigned int occ, ocd;
3987 unsigned int cc = c;
3988 unsigned int origd = d;
3989 while (get_othercase_range(&cc, origd, &occ, &ocd))
3990 {
3991 if (occ >= (unsigned int)c &&
3992 ocd <= (unsigned int)d)
3993 continue; /* Skip embedded ranges */
3994
3995 if (occ < (unsigned int)c &&
3996 ocd >= (unsigned int)c - 1) /* Extend the basic range */
3997 { /* if there is overlap, */
3998 c = occ; /* noting that if occ < c */
3999 continue; /* we can't have ocd > d */
4000 } /* because a subrange is */
4001 if (ocd > (unsigned int)d &&
4002 occ <= (unsigned int)d + 1) /* always shorter than */
4003 { /* the basic range. */
4004 d = ocd;
4005 continue;
4006 }
4007
4008 if (occ == ocd)
4009 {
4010 *class_utf8data++ = XCL_SINGLE;
4011 }
4012 else
4013 {
4014 *class_utf8data++ = XCL_RANGE;
4015 class_utf8data += _pcre_ord2utf8(occ, class_utf8data);
4016 }
4017 class_utf8data += _pcre_ord2utf8(ocd, class_utf8data);
4018 }
4019 }
4020 #endif /* SUPPORT_UCP */
4021
4022 /* Now record the original range, possibly modified for UCP caseless
4023 overlapping ranges. */
4024
4025 *class_utf8data++ = XCL_RANGE;
4026 class_utf8data += _pcre_ord2utf8(c, class_utf8data);
4027 class_utf8data += _pcre_ord2utf8(d, class_utf8data);
4028
4029 /* With UCP support, we are done. Without UCP support, there is no
4030 caseless matching for UTF-8 characters > 127; we can use the bit map
4031 for the smaller ones. */
4032
4033 #ifdef SUPPORT_UCP
4034 continue; /* With next character in the class */
4035 #else
4036 if ((options & PCRE_CASELESS) == 0 || c > 127) continue;
4037
4038 /* Adjust upper limit and fall through to set up the map */
4039
4040 d = 127;
4041
4042 #endif /* SUPPORT_UCP */
4043 }
4044 #endif /* SUPPORT_UTF8 */
4045
4046 /* We use the bit map for all cases when not in UTF-8 mode; else
4047 ranges that lie entirely within 0-127 when there is UCP support; else
4048 for partial ranges without UCP support. */
4049
4050 class_charcount += d - c + 1;
4051 class_lastchar = d;
4052
4053 /* We can save a bit of time by skipping this in the pre-compile. */
4054
4055 if (lengthptr == NULL) for (; c <= d; c++)
4056 {
4057 classbits[c/8] |= (1 << (c&7));
4058 if ((options & PCRE_CASELESS) != 0)
4059 {
4060 int uc = cd->fcc[c]; /* flip case */
4061 classbits[uc/8] |= (1 << (uc&7));
4062 }
4063 }
4064
4065 continue; /* Go get the next char in the class */
4066 }
4067
4068 /* Handle a lone single character - we can get here for a normal
4069 non-escape char, or after \ that introduces a single character or for an
4070 apparent range that isn't. */
4071
4072 LONE_SINGLE_CHARACTER:
4073
4074 /* Handle a character that cannot go in the bit map */
4075
4076 #ifdef SUPPORT_UTF8
4077 if (utf8 && (c > 255 || ((options & PCRE_CASELESS) != 0 && c > 127)))
4078 {
4079 class_utf8 = TRUE;
4080 *class_utf8data++ = XCL_SINGLE;
4081 class_utf8data += _pcre_ord2utf8(c, class_utf8data);
4082
4083 #ifdef SUPPORT_UCP
4084 if ((options & PCRE_CASELESS) != 0)
4085 {
4086 unsigned int othercase;
4087 if ((othercase = UCD_OTHERCASE(c)) != c)
4088 {
4089 *class_utf8data++ = XCL_SINGLE;
4090 class_utf8data += _pcre_ord2utf8(othercase, class_utf8data);
4091 }
4092 }
4093 #endif /* SUPPORT_UCP */
4094
4095 }
4096 else
4097 #endif /* SUPPORT_UTF8 */
4098
4099 /* Handle a single-byte character */
4100 {
4101 classbits[c/8] |= (1 << (c&7));
4102 if ((options & PCRE_CASELESS) != 0)
4103 {
4104 c = cd->fcc[c]; /* flip case */
4105 classbits[c/8] |= (1 << (c&7));
4106 }
4107 class_charcount++;
4108 class_lastchar = c;
4109 }
4110 }
4111
4112 /* Loop until ']' reached. This "while" is the end of the "do" far above.
4113 If we are at the end of an internal nested string, revert to the outer
4114 string. */
4115
4116 while (((c = *(++ptr)) != 0 ||
4117 (nestptr != NULL &&
4118 (ptr = nestptr, nestptr = NULL, c = *(++ptr)) != 0)) &&
4119 (c != CHAR_RIGHT_SQUARE_BRACKET || inescq));
4120
4121 /* Check for missing terminating ']' */
4122
4123 if (c == 0)
4124 {
4125 *errorcodeptr = ERR6;
4126 goto FAILED;
4127 }
4128
4129 /* If class_charcount is 1, we saw precisely one character whose value is
4130 less than 256. As long as there were no characters >= 128 and there was no
4131 use of \p or \P, in other words, no use of any XCLASS features, we can
4132 optimize.
4133
4134 In UTF-8 mode, we can optimize the negative case only if there were no
4135 characters >= 128 because OP_NOT and the related opcodes like OP_NOTSTAR
4136 operate on single-bytes characters only. This is an historical hangover.
4137 Maybe one day we can tidy these opcodes to handle multi-byte characters.
4138
4139 The optimization throws away the bit map. We turn the item into a
4140 1-character OP_CHAR[I] if it's positive, or OP_NOT[I] if it's negative.
4141 Note that OP_NOT[I] does not support multibyte characters. In the positive
4142 case, it can cause firstbyte to be set. Otherwise, there can be no first
4143 char if this item is first, whatever repeat count may follow. In the case
4144 of reqbyte, save the previous value for reinstating. */
4145
4146 #ifdef SUPPORT_UTF8
4147 if (class_charcount == 1 && !class_utf8 &&
4148 (!utf8 || !negate_class || class_lastchar < 128))
4149 #else
4150 if (class_charcount == 1)
4151 #endif
4152 {
4153 zeroreqbyte = reqbyte;
4154
4155 /* The OP_NOT[I] opcodes work on one-byte characters only. */
4156
4157 if (negate_class)
4158 {
4159 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
4160 zerofirstbyte = firstbyte;
4161 *code++ = ((options & PCRE_CASELESS) != 0)? OP_NOTI: OP_NOT;
4162 *code++ = class_lastchar;
4163 break;
4164 }
4165
4166 /* For a single, positive character, get the value into mcbuffer, and
4167 then we can handle this with the normal one-character code. */
4168
4169 #ifdef SUPPORT_UTF8
4170 if (utf8 && class_lastchar > 127)
4171 mclength = _pcre_ord2utf8(class_lastchar, mcbuffer);
4172 else
4173 #endif
4174 {
4175 mcbuffer[0] = class_lastchar;
4176 mclength = 1;
4177 }
4178 goto ONE_CHAR;
4179 } /* End of 1-char optimization */
4180
4181 /* The general case - not the one-char optimization. If this is the first
4182 thing in the branch, there can be no first char setting, whatever the
4183 repeat count. Any reqbyte setting must remain unchanged after any kind of
4184 repeat. */
4185
4186 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
4187 zerofirstbyte = firstbyte;
4188 zeroreqbyte = reqbyte;
4189
4190 /* If there are characters with values > 255, we have to compile an
4191 extended class, with its own opcode, unless there was a negated special
4192 such as \S in the class, and PCRE_UCP is not set, because in that case all
4193 characters > 255 are in the class, so any that were explicitly given as
4194 well can be ignored. If (when there are explicit characters > 255 that must
4195 be listed) there are no characters < 256, we can omit the bitmap in the
4196 actual compiled code. */
4197
4198 #ifdef SUPPORT_UTF8
4199 if (class_utf8 && (!should_flip_negation || (options & PCRE_UCP) != 0))
4200 {
4201 *class_utf8data++ = XCL_END; /* Marks the end of extra data */
4202 *code++ = OP_XCLASS;
4203 code += LINK_SIZE;
4204 *code = negate_class? XCL_NOT : 0;
4205
4206 /* If the map is required, move up the extra data to make room for it;
4207 otherwise just move the code pointer to the end of the extra data. */
4208
4209 if (class_charcount > 0)
4210 {
4211 *code++ |= XCL_MAP;
4212 memmove(code + 32, code, class_utf8data - code);
4213 memcpy(code, classbits, 32);
4214 code = class_utf8data + 32;
4215 }
4216 else code = class_utf8data;
4217
4218 /* Now fill in the complete length of the item */
4219
4220 PUT(previous, 1, code - previous);
4221 break; /* End of class handling */
4222 }
4223 #endif
4224
4225 /* If there are no characters > 255, or they are all to be included or
4226 excluded, set the opcode to OP_CLASS or OP_NCLASS, depending on whether the
4227 whole class was negated and whether there were negative specials such as \S
4228 (non-UCP) in the class. Then copy the 32-byte map into the code vector,
4229 negating it if necessary. */
4230
4231 *code++ = (negate_class == should_flip_negation) ? OP_CLASS : OP_NCLASS;
4232 if (negate_class)
4233 {
4234 if (lengthptr == NULL) /* Save time in the pre-compile phase */
4235 for (c = 0; c < 32; c++) code[c] = ~classbits[c];
4236 }
4237 else
4238 {
4239 memcpy(code, classbits, 32);
4240 }
4241 code += 32;
4242 break;
4243
4244
4245 /* ===================================================================*/
4246 /* Various kinds of repeat; '{' is not necessarily a quantifier, but this
4247 has been tested above. */
4248
4249 case CHAR_LEFT_CURLY_BRACKET:
4250 if (!is_quantifier) goto NORMAL_CHAR;
4251 ptr = read_repeat_counts(ptr+1, &repeat_min, &repeat_max, errorcodeptr);
4252 if (*errorcodeptr != 0) goto FAILED;
4253 goto REPEAT;
4254
4255 case CHAR_ASTERISK:
4256 repeat_min = 0;
4257 repeat_max = -1;
4258 goto REPEAT;
4259
4260 case CHAR_PLUS:
4261 repeat_min = 1;
4262 repeat_max = -1;
4263 goto REPEAT;
4264
4265 case CHAR_QUESTION_MARK:
4266 repeat_min = 0;
4267 repeat_max = 1;
4268
4269 REPEAT:
4270 if (previous == NULL)
4271 {
4272 *errorcodeptr = ERR9;
4273 goto FAILED;
4274 }
4275
4276 if (repeat_min == 0)
4277 {
4278 firstbyte = zerofirstbyte; /* Adjust for zero repeat */
4279 reqbyte = zeroreqbyte; /* Ditto */
4280 }
4281
4282 /* Remember whether this is a variable length repeat */
4283
4284 reqvary = (repeat_min == repeat_max)? 0 : REQ_VARY;
4285
4286 op_type = 0; /* Default single-char op codes */
4287 possessive_quantifier = FALSE; /* Default not possessive quantifier */
4288
4289 /* Save start of previous item, in case we have to move it up in order to
4290 insert something before it. */
4291
4292 tempcode = previous;
4293
4294 /* If the next character is '+', we have a possessive quantifier. This
4295 implies greediness, whatever the setting of the PCRE_UNGREEDY option.
4296 If the next character is '?' this is a minimizing repeat, by default,
4297 but if PCRE_UNGREEDY is set, it works the other way round. We change the
4298 repeat type to the non-default. */
4299
4300 if (ptr[1] == CHAR_PLUS)
4301 {
4302 repeat_type = 0; /* Force greedy */
4303 possessive_quantifier = TRUE;
4304 ptr++;
4305 }
4306 else if (ptr[1] == CHAR_QUESTION_MARK)
4307 {
4308 repeat_type = greedy_non_default;
4309 ptr++;
4310 }
4311 else repeat_type = greedy_default;
4312
4313 /* If previous was a recursion call, wrap it in atomic brackets so that
4314 previous becomes the atomic group. All recursions were so wrapped in the
4315 past, but it no longer happens for non-repeated recursions. In fact, the
4316 repeated ones could be re-implemented independently so as not to need this,
4317 but for the moment we rely on the code for repeating groups. */
4318
4319 if (*previous == OP_RECURSE)
4320 {
4321 memmove(previous + 1 + LINK_SIZE, previous, 1 + LINK_SIZE);
4322 *previous = OP_ONCE;
4323 PUT(previous, 1, 2 + 2*LINK_SIZE);
4324 previous[2 + 2*LINK_SIZE] = OP_KET;
4325 PUT(previous, 3 + 2*LINK_SIZE, 2 + 2*LINK_SIZE);
4326 code += 2 + 2 * LINK_SIZE;
4327 length_prevgroup = 3 + 3*LINK_SIZE;
4328
4329 /* When actually compiling, we need to check whether this was a forward
4330 reference, and if so, adjust the offset. */
4331
4332 if (lengthptr == NULL && cd->hwm >= cd->start_workspace + LINK_SIZE)
4333 {
4334 int offset = GET(cd->hwm, -LINK_SIZE);
4335 if (offset == previous + 1 - cd->start_code)
4336 PUT(cd->hwm, -LINK_SIZE, offset + 1 + LINK_SIZE);
4337 }
4338 }
4339
4340 /* Now handle repetition for the different types of item. */
4341
4342 /* If previous was a character match, abolish the item and generate a
4343 repeat item instead. If a char item has a minumum of more than one, ensure
4344 that it is set in reqbyte - it might not be if a sequence such as x{3} is
4345 the first thing in a branch because the x will have gone into firstbyte
4346 instead. */
4347
4348 if (*previous == OP_CHAR || *previous == OP_CHARI)
4349 {
4350 op_type = (*previous == OP_CHAR)? 0 : OP_STARI - OP_STAR;
4351
4352 /* Deal with UTF-8 characters that take up more than one byte. It's
4353 easier to write this out separately than try to macrify it. Use c to
4354 hold the length of the character in bytes, plus 0x80 to flag that it's a
4355 length rather than a small character. */
4356
4357 #ifdef SUPPORT_UTF8
4358 if (utf8 && (code[-1] & 0x80) != 0)
4359 {
4360 uschar *lastchar = code - 1;
4361 while((*lastchar & 0xc0) == 0x80) lastchar--;
4362 c = code - lastchar; /* Length of UTF-8 character */
4363 memcpy(utf8_char, lastchar, c); /* Save the char */
4364 c |= 0x80; /* Flag c as a length */
4365 }
4366 else
4367 #endif
4368
4369 /* Handle the case of a single byte - either with no UTF8 support, or
4370 with UTF-8 disabled, or for a UTF-8 character < 128. */
4371
4372 {
4373 c = code[-1];
4374 if (repeat_min > 1) reqbyte = c | req_caseopt | cd->req_varyopt;
4375 }
4376
4377 /* If the repetition is unlimited, it pays to see if the next thing on
4378 the line is something that cannot possibly match this character. If so,
4379 automatically possessifying this item gains some performance in the case
4380 where the match fails. */
4381
4382 if (!possessive_quantifier &&
4383 repeat_max < 0 &&
4384 check_auto_possessive(previous, utf8, ptr + 1, options, cd))
4385 {
4386 repeat_type = 0; /* Force greedy */
4387 possessive_quantifier = TRUE;
4388 }
4389
4390 goto OUTPUT_SINGLE_REPEAT; /* Code shared with single character types */
4391 }
4392
4393 /* If previous was a single negated character ([^a] or similar), we use
4394 one of the special opcodes, replacing it. The code is shared with single-
4395 character repeats by setting opt_type to add a suitable offset into
4396 repeat_type. We can also test for auto-possessification. OP_NOT and OP_NOTI
4397 are currently used only for single-byte chars. */
4398
4399 else if (*previous == OP_NOT || *previous == OP_NOTI)
4400 {
4401 op_type = ((*previous == OP_NOT)? OP_NOTSTAR : OP_NOTSTARI) - OP_STAR;
4402 c = previous[1];
4403 if (!possessive_quantifier &&
4404 repeat_max < 0 &&
4405 check_auto_possessive(previous, utf8, ptr + 1, options, cd))
4406 {
4407 repeat_type = 0; /* Force greedy */
4408 possessive_quantifier = TRUE;
4409 }
4410 goto OUTPUT_SINGLE_REPEAT;
4411 }
4412
4413 /* If previous was a character type match (\d or similar), abolish it and
4414 create a suitable repeat item. The code is shared with single-character
4415 repeats by setting op_type to add a suitable offset into repeat_type. Note
4416 the the Unicode property types will be present only when SUPPORT_UCP is
4417 defined, but we don't wrap the little bits of code here because it just
4418 makes it horribly messy. */
4419
4420 else if (*previous < OP_EODN)
4421 {
4422 uschar *oldcode;
4423 int prop_type, prop_value;
4424 op_type = OP_TYPESTAR - OP_STAR; /* Use type opcodes */
4425 c = *previous;
4426
4427 if (!possessive_quantifier &&
4428 repeat_max < 0 &&
4429 check_auto_possessive(previous, utf8, ptr + 1, options, cd))
4430 {
4431 repeat_type = 0; /* Force greedy */
4432 possessive_quantifier = TRUE;
4433 }
4434
4435 OUTPUT_SINGLE_REPEAT:
4436 if (*previous == OP_PROP || *previous == OP_NOTPROP)
4437 {
4438 prop_type = previous[1];
4439 prop_value = previous[2];
4440 }
4441 else prop_type = prop_value = -1;
4442
4443 oldcode = code;
4444 code = previous; /* Usually overwrite previous item */
4445
4446 /* If the maximum is zero then the minimum must also be zero; Perl allows
4447 this case, so we do too - by simply omitting the item altogether. */
4448
4449 if (repeat_max == 0) goto END_REPEAT;
4450
4451 /*--------------------------------------------------------------------*/
4452 /* This code is obsolete from release 8.00; the restriction was finally
4453 removed: */
4454
4455 /* All real repeats make it impossible to handle partial matching (maybe
4456 one day we will be able to remove this restriction). */
4457
4458 /* if (repeat_max != 1) cd->external_flags |= PCRE_NOPARTIAL; */
4459 /*--------------------------------------------------------------------*/
4460
4461 /* Combine the op_type with the repeat_type */
4462
4463 repeat_type += op_type;
4464
4465 /* A minimum of zero is handled either as the special case * or ?, or as
4466 an UPTO, with the maximum given. */
4467
4468 if (repeat_min == 0)
4469 {
4470 if (repeat_max == -1) *code++ = OP_STAR + repeat_type;
4471 else if (repeat_max == 1) *code++ = OP_QUERY + repeat_type;
4472 else
4473 {
4474 *code++ = OP_UPTO + repeat_type;
4475 PUT2INC(code, 0, repeat_max);
4476 }
4477 }
4478
4479 /* A repeat minimum of 1 is optimized into some special cases. If the
4480 maximum is unlimited, we use OP_PLUS. Otherwise, the original item is
4481 left in place and, if the maximum is greater than 1, we use OP_UPTO with
4482 one less than the maximum. */
4483
4484 else if (repeat_min == 1)
4485 {
4486 if (repeat_max == -1)
4487 *code++ = OP_PLUS + repeat_type;
4488 else
4489 {
4490 code = oldcode; /* leave previous item in place */
4491 if (repeat_max == 1) goto END_REPEAT;
4492 *code++ = OP_UPTO + repeat_type;
4493 PUT2INC(code, 0, repeat_max - 1);
4494 }
4495 }
4496
4497 /* The case {n,n} is just an EXACT, while the general case {n,m} is
4498 handled as an EXACT followed by an UPTO. */
4499
4500 else
4501 {
4502 *code++ = OP_EXACT + op_type; /* NB EXACT doesn't have repeat_type */
4503 PUT2INC(code, 0, repeat_min);
4504
4505 /* If the maximum is unlimited, insert an OP_STAR. Before doing so,
4506 we have to insert the character for the previous code. For a repeated
4507 Unicode property match, there are two extra bytes that define the
4508 required property. In UTF-8 mode, long characters have their length in
4509 c, with the 0x80 bit as a flag. */
4510
4511 if (repeat_max < 0)
4512 {
4513 #ifdef SUPPORT_UTF8
4514 if (utf8 && c >= 128)
4515 {
4516 memcpy(code, utf8_char, c & 7);
4517 code += c & 7;
4518 }
4519 else
4520 #endif
4521 {
4522 *code++ = c;
4523 if (prop_type >= 0)
4524 {
4525 *code++ = prop_type;
4526 *code++ = prop_value;
4527 }
4528 }
4529 *code++ = OP_STAR + repeat_type;
4530 }
4531
4532 /* Else insert an UPTO if the max is greater than the min, again
4533 preceded by the character, for the previously inserted code. If the
4534 UPTO is just for 1 instance, we can use QUERY instead. */
4535
4536 else if (repeat_max != repeat_min)
4537 {
4538 #ifdef SUPPORT_UTF8
4539 if (utf8 && c >= 128)
4540 {
4541 memcpy(code, utf8_char, c & 7);
4542 code += c & 7;
4543 }
4544 else
4545 #endif
4546 *code++ = c;
4547 if (prop_type >= 0)
4548 {
4549 *code++ = prop_type;
4550 *code++ = prop_value;
4551 }
4552 repeat_max -= repeat_min;
4553
4554 if (repeat_max == 1)
4555 {
4556 *code++ = OP_QUERY + repeat_type;
4557 }
4558 else
4559 {
4560 *code++ = OP_UPTO + repeat_type;
4561 PUT2INC(code, 0, repeat_max);
4562 }
4563 }
4564 }
4565
4566 /* The character or character type itself comes last in all cases. */
4567
4568 #ifdef SUPPORT_UTF8
4569 if (utf8 && c >= 128)
4570 {
4571 memcpy(code, utf8_char, c & 7);
4572 code += c & 7;
4573 }
4574 else
4575 #endif
4576 *code++ = c;
4577
4578 /* For a repeated Unicode property match, there are two extra bytes that
4579 define the required property. */
4580
4581 #ifdef SUPPORT_UCP
4582 if (prop_type >= 0)
4583 {
4584 *code++ = prop_type;
4585 *code++ = prop_value;
4586 }
4587 #endif
4588 }
4589
4590 /* If previous was a character class or a back reference, we put the repeat
4591 stuff after it, but just skip the item if the repeat was {0,0}. */
4592
4593 else if (*previous == OP_CLASS ||
4594 *previous == OP_NCLASS ||
4595 #ifdef SUPPORT_UTF8
4596 *previous == OP_XCLASS ||
4597 #endif
4598 *previous == OP_REF ||
4599 *previous == OP_REFI)
4600 {
4601 if (repeat_max == 0)
4602 {
4603 code = previous;
4604 goto END_REPEAT;
4605 }
4606
4607 /*--------------------------------------------------------------------*/
4608 /* This code is obsolete from release 8.00; the restriction was finally
4609 removed: */
4610
4611 /* All real repeats make it impossible to handle partial matching (maybe
4612 one day we will be able to remove this restriction). */
4613
4614 /* if (repeat_max != 1) cd->external_flags |= PCRE_NOPARTIAL; */
4615 /*--------------------------------------------------------------------*/
4616
4617 if (repeat_min == 0 && repeat_max == -1)
4618 *code++ = OP_CRSTAR + repeat_type;
4619 else if (repeat_min == 1 && repeat_max == -1)
4620 *code++ = OP_CRPLUS + repeat_type;
4621 else if (repeat_min == 0 && repeat_max == 1)
4622 *code++ = OP_CRQUERY + repeat_type;
4623 else
4624 {
4625 *code++ = OP_CRRANGE + repeat_type;
4626 PUT2INC(code, 0, repeat_min);
4627 if (repeat_max == -1) repeat_max = 0; /* 2-byte encoding for max */
4628 PUT2INC(code, 0, repeat_max);
4629 }
4630 }
4631
4632 /* If previous was a bracket group, we may have to replicate it in certain
4633 cases. Note that at this point we can encounter only the "basic" bracket
4634 opcodes such as BRA and CBRA, as this is the place where they get converted
4635 into the more special varieties such as BRAPOS and SBRA. A test for >=
4636 OP_ASSERT and <= OP_COND includes ASSERT, ASSERT_NOT, ASSERTBACK,
4637 ASSERTBACK_NOT, ONCE, BRA, CBRA, and COND. Originally, PCRE did not allow
4638 repetition of assertions, but now it does, for Perl compatibility. */
4639
4640 else if (*previous >= OP_ASSERT && *previous <= OP_COND)
4641 {
4642 register int i;
4643 int len = (int)(code - previous);
4644 uschar *bralink = NULL;
4645 uschar *brazeroptr = NULL;
4646
4647 /* Repeating a DEFINE group is pointless, but Perl allows the syntax, so
4648 we just ignore the repeat. */
4649
4650 if (*previous == OP_COND && previous[LINK_SIZE+1] == OP_DEF)
4651 goto END_REPEAT;
4652
4653 /* There is no sense in actually repeating assertions. The only potential
4654 use of repetition is in cases when the assertion is optional. Therefore,
4655 if the minimum is greater than zero, just ignore the repeat. If the
4656 maximum is not not zero or one, set it to 1. */
4657
4658 if (*previous < OP_ONCE) /* Assertion */
4659 {
4660 if (repeat_min > 0) goto END_REPEAT;
4661 if (repeat_max < 0 || repeat_max > 1) repeat_max = 1;
4662 }
4663
4664 /* The case of a zero minimum is special because of the need to stick
4665 OP_BRAZERO in front of it, and because the group appears once in the
4666 data, whereas in other cases it appears the minimum number of times. For
4667 this reason, it is simplest to treat this case separately, as otherwise
4668 the code gets far too messy. There are several special subcases when the
4669 minimum is zero. */
4670
4671 if (repeat_min == 0)
4672 {
4673 /* If the maximum is also zero, we used to just omit the group from the
4674 output altogether, like this:
4675
4676 ** if (repeat_max == 0)
4677 ** {
4678 ** code = previous;
4679 ** goto END_REPEAT;
4680 ** }
4681
4682 However, that fails when a group or a subgroup within it is referenced
4683 as a subroutine from elsewhere in the pattern, so now we stick in
4684 OP_SKIPZERO in front of it so that it is skipped on execution. As we
4685 don't have a list of which groups are referenced, we cannot do this
4686 selectively.
4687
4688 If the maximum is 1 or unlimited, we just have to stick in the BRAZERO
4689 and do no more at this point. However, we do need to adjust any
4690 OP_RECURSE calls inside the group that refer to the group itself or any
4691 internal or forward referenced group, because the offset is from the
4692 start of the whole regex. Temporarily terminate the pattern while doing
4693 this. */
4694
4695 if (repeat_max <= 1) /* Covers 0, 1, and unlimited */
4696 {
4697 *code = OP_END;
4698 adjust_recurse(previous, 1, utf8, cd, save_hwm);
4699 memmove(previous+1, previous, len);
4700 code++;
4701 if (repeat_max == 0)
4702 {
4703 *previous++ = OP_SKIPZERO;
4704 goto END_REPEAT;
4705 }
4706 brazeroptr = previous; /* Save for possessive optimizing */
4707 *previous++ = OP_BRAZERO + repeat_type;
4708 }
4709
4710 /* If the maximum is greater than 1 and limited, we have to replicate
4711 in a nested fashion, sticking OP_BRAZERO before each set of brackets.
4712 The first one has to be handled carefully because it's the original
4713 copy, which has to be moved up. The remainder can be handled by code
4714 that is common with the non-zero minimum case below. We have to
4715 adjust the value or repeat_max, since one less copy is required. Once
4716 again, we may have to adjust any OP_RECURSE calls inside the group. */
4717
4718 else
4719 {
4720 int offset;
4721 *code = OP_END;
4722 adjust_recurse(previous, 2 + LINK_SIZE, utf8, cd, save_hwm);
4723 memmove(previous + 2 + LINK_SIZE, previous, len);
4724 code += 2 + LINK_SIZE;
4725 *previous++ = OP_BRAZERO + repeat_type;
4726 *previous++ = OP_BRA;
4727
4728 /* We chain together the bracket offset fields that have to be
4729 filled in later when the ends of the brackets are reached. */
4730
4731 offset = (bralink == NULL)? 0 : (int)(previous - bralink);
4732 bralink = previous;
4733 PUTINC(previous, 0, offset);
4734 }
4735
4736 repeat_max--;
4737 }
4738
4739 /* If the minimum is greater than zero, replicate the group as many
4740 times as necessary, and adjust the maximum to the number of subsequent
4741 copies that we need. If we set a first char from the group, and didn't
4742 set a required char, copy the latter from the former. If there are any
4743 forward reference subroutine calls in the group, there will be entries on
4744 the workspace list; replicate these with an appropriate increment. */
4745
4746 else
4747 {
4748 if (repeat_min > 1)
4749 {
4750 /* In the pre-compile phase, we don't actually do the replication. We
4751 just adjust the length as if we had. Do some paranoid checks for
4752 potential integer overflow. The INT64_OR_DOUBLE type is a 64-bit
4753 integer type when available, otherwise double. */
4754
4755 if (lengthptr != NULL)
4756 {
4757 int delta = (repeat_min - 1)*length_prevgroup;
4758 if ((INT64_OR_DOUBLE)(repeat_min - 1)*
4759 (INT64_OR_DOUBLE)length_prevgroup >
4760 (INT64_OR_DOUBLE)INT_MAX ||
4761 OFLOW_MAX - *lengthptr < delta)
4762 {
4763 *errorcodeptr = ERR20;
4764 goto FAILED;
4765 }
4766 *lengthptr += delta;
4767 }
4768
4769 /* This is compiling for real */
4770
4771 else
4772 {
4773 if (groupsetfirstbyte && reqbyte < 0) reqbyte = firstbyte;
4774 for (i = 1; i < repeat_min; i++)
4775 {
4776 uschar *hc;
4777 uschar *this_hwm = cd->hwm;
4778 memcpy(code, previous, len);
4779 for (hc = save_hwm; hc < this_hwm; hc += LINK_SIZE)
4780 {
4781 PUT(cd->hwm, 0, GET(hc, 0) + len);
4782 cd->hwm += LINK_SIZE;
4783 }
4784 save_hwm = this_hwm;
4785 code += len;
4786 }
4787 }
4788 }
4789
4790 if (repeat_max > 0) repeat_max -= repeat_min;
4791 }
4792
4793 /* This code is common to both the zero and non-zero minimum cases. If
4794 the maximum is limited, it replicates the group in a nested fashion,
4795 remembering the bracket starts on a stack. In the case of a zero minimum,
4796 the first one was set up above. In all cases the repeat_max now specifies
4797 the number of additional copies needed. Again, we must remember to
4798 replicate entries on the forward reference list. */
4799
4800 if (repeat_max >= 0)
4801 {
4802 /* In the pre-compile phase, we don't actually do the replication. We
4803 just adjust the length as if we had. For each repetition we must add 1
4804 to the length for BRAZERO and for all but the last repetition we must
4805 add 2 + 2*LINKSIZE to allow for the nesting that occurs. Do some
4806 paranoid checks to avoid integer overflow. The INT64_OR_DOUBLE type is
4807 a 64-bit integer type when available, otherwise double. */
4808
4809 if (lengthptr != NULL && repeat_max > 0)
4810 {
4811 int delta = repeat_max * (length_prevgroup + 1 + 2 + 2*LINK_SIZE) -
4812 2 - 2*LINK_SIZE; /* Last one doesn't nest */
4813 if ((INT64_OR_DOUBLE)repeat_max *
4814 (INT64_OR_DOUBLE)(length_prevgroup + 1 + 2 + 2*LINK_SIZE)
4815 > (INT64_OR_DOUBLE)INT_MAX ||
4816 OFLOW_MAX - *lengthptr < delta)
4817 {
4818 *errorcodeptr = ERR20;
4819 goto FAILED;
4820 }
4821 *lengthptr += delta;
4822 }
4823
4824 /* This is compiling for real */
4825
4826 else for (i = repeat_max - 1; i >= 0; i--)
4827 {
4828 uschar *hc;
4829 uschar *this_hwm = cd->hwm;
4830
4831 *code++ = OP_BRAZERO + repeat_type;
4832
4833 /* All but the final copy start a new nesting, maintaining the
4834 chain of brackets outstanding. */
4835
4836 if (i != 0)
4837 {
4838 int offset;
4839 *code++ = OP_BRA;
4840 offset = (bralink == NULL)? 0 : (int)(code - bralink);
4841 bralink = code;
4842 PUTINC(code, 0, offset);
4843 }
4844
4845 memcpy(code, previous, len);
4846 for (hc = save_hwm; hc < this_hwm; hc += LINK_SIZE)
4847 {
4848 PUT(cd->hwm, 0, GET(hc, 0) + len + ((i != 0)? 2+LINK_SIZE : 1));
4849 cd->hwm += LINK_SIZE;
4850 }
4851 save_hwm = this_hwm;
4852 code += len;
4853 }
4854
4855 /* Now chain through the pending brackets, and fill in their length
4856 fields (which are holding the chain links pro tem). */
4857
4858 while (bralink != NULL)
4859 {
4860 int oldlinkoffset;
4861 int offset = (int)(code - bralink + 1);
4862 uschar *bra = code - offset;
4863 oldlinkoffset = GET(bra, 1);
4864 bralink = (oldlinkoffset == 0)? NULL : bralink - oldlinkoffset;
4865 *code++ = OP_KET;
4866 PUTINC(code, 0, offset);
4867 PUT(bra, 1, offset);
4868 }
4869 }
4870
4871 /* If the maximum is unlimited, set a repeater in the final copy. For
4872 ONCE brackets, that's all we need to do. However, possessively repeated
4873 ONCE brackets can be converted into non-capturing brackets, as the
4874 behaviour of (?:xx)++ is the same as (?>xx)++ and this saves having to
4875 deal with possessive ONCEs specially.
4876
4877 Otherwise, if the quantifier was possessive, we convert the BRA code to
4878 the POS form, and the KET code to KETRPOS. (It turns out to be convenient
4879 at runtime to detect this kind of subpattern at both the start and at the
4880 end.) The use of special opcodes makes it possible to reduce greatly the
4881 stack usage in pcre_exec(). If the group is preceded by OP_BRAZERO,
4882 convert this to OP_BRAPOSZERO. Then cancel the possessive flag so that
4883 the default action below, of wrapping everything inside atomic brackets,
4884 does not happen.
4885
4886 Then, when we are doing the actual compile phase, check to see whether
4887 this group is one that could match an empty string. If so, convert the
4888 initial operator to the S form (e.g. OP_BRA -> OP_SBRA) so that runtime
4889 checking can be done. [This check is also applied to ONCE groups at
4890 runtime, but in a different way.] */
4891
4892 else
4893 {
4894 uschar *ketcode = code - 1 - LINK_SIZE;
4895 uschar *bracode = ketcode - GET(ketcode, 1);
4896
4897 if ((*bracode == OP_ONCE || *bracode == OP_ONCE_NC) &&
4898 possessive_quantifier) *bracode = OP_BRA;
4899
4900 if (*bracode == OP_ONCE || *bracode == OP_ONCE_NC)
4901 *ketcode = OP_KETRMAX + repeat_type;
4902 else
4903 {
4904 if (possessive_quantifier)
4905 {
4906 *bracode += 1; /* Switch to xxxPOS opcodes */
4907 *ketcode = OP_KETRPOS;
4908 if (brazeroptr != NULL) *brazeroptr = OP_BRAPOSZERO;
4909 possessive_quantifier = FALSE;
4910 }
4911 else *ketcode = OP_KETRMAX + repeat_type;
4912
4913 if (lengthptr == NULL)
4914 {
4915 uschar *scode = bracode;
4916 do
4917 {
4918 if (could_be_empty_branch(scode, ketcode, utf8, cd))
4919 {
4920 *bracode += OP_SBRA - OP_BRA;
4921 break;
4922 }
4923 scode += GET(scode, 1);
4924 }
4925 while (*scode == OP_ALT);
4926 }
4927 }
4928 }
4929 }
4930
4931 /* If previous is OP_FAIL, it was generated by an empty class [] in
4932 JavaScript mode. The other ways in which OP_FAIL can be generated, that is
4933 by (*FAIL) or (?!) set previous to NULL, which gives a "nothing to repeat"
4934 error above. We can just ignore the repeat in JS case. */
4935
4936 else if (*previous == OP_FAIL) goto END_REPEAT;
4937
4938 /* Else there's some kind of shambles */
4939
4940 else
4941 {
4942 *errorcodeptr = ERR11;
4943 goto FAILED;
4944 }
4945
4946 /* If the character following a repeat is '+', or if certain optimization
4947 tests above succeeded, possessive_quantifier is TRUE. For some opcodes,
4948 there are special alternative opcodes for this case. For anything else, we
4949 wrap the entire repeated item inside OP_ONCE brackets. Logically, the '+'
4950 notation is just syntactic sugar, taken from Sun's Java package, but the
4951 special opcodes can optimize it.
4952
4953 Possessively repeated subpatterns have already been handled in the code
4954 just above, so possessive_quantifier is always FALSE for them at this
4955 stage.
4956
4957 Note that the repeated item starts at tempcode, not at previous, which
4958 might be the first part of a string whose (former) last char we repeated.
4959
4960 Possessifying an 'exact' quantifier has no effect, so we can ignore it. But
4961 an 'upto' may follow. We skip over an 'exact' item, and then test the
4962 length of what remains before proceeding. */
4963
4964 if (possessive_quantifier)
4965 {
4966 int len;
4967
4968 if (*tempcode == OP_TYPEEXACT)
4969 tempcode += _pcre_OP_lengths[*tempcode] +
4970 ((tempcode[3] == OP_PROP || tempcode[3] == OP_NOTPROP)? 2 : 0);
4971
4972 else if (*tempcode == OP_EXACT || *tempcode == OP_NOTEXACT)
4973 {
4974 tempcode += _pcre_OP_lengths[*tempcode];
4975 #ifdef SUPPORT_UTF8
4976 if (utf8 && tempcode[-1] >= 0xc0)
4977 tempcode += _pcre_utf8_table4[tempcode[-1] & 0x3f];
4978 #endif
4979 }
4980
4981 len = (int)(code - tempcode);
4982 if (len > 0) switch (*tempcode)
4983 {
4984 case OP_STAR: *tempcode = OP_POSSTAR; break;
4985 case OP_PLUS: *tempcode = OP_POSPLUS; break;
4986 case OP_QUERY: *tempcode = OP_POSQUERY; break;
4987 case OP_UPTO: *tempcode = OP_POSUPTO; break;
4988
4989 case OP_STARI: *tempcode = OP_POSSTARI; break;
4990 case OP_PLUSI: *tempcode = OP_POSPLUSI; break;
4991 case OP_QUERYI: *tempcode = OP_POSQUERYI; break;
4992 case OP_UPTOI: *tempcode = OP_POSUPTOI; break;
4993
4994 case OP_NOTSTAR: *tempcode = OP_NOTPOSSTAR; break;
4995 case OP_NOTPLUS: *tempcode = OP_NOTPOSPLUS; break;
4996 case OP_NOTQUERY: *tempcode = OP_NOTPOSQUERY; break;
4997 case OP_NOTUPTO: *tempcode = OP_NOTPOSUPTO; break;
4998
4999 case OP_NOTSTARI: *tempcode = OP_NOTPOSSTARI; break;
5000 case OP_NOTPLUSI: *tempcode = OP_NOTPOSPLUSI; break;
5001 case OP_NOTQUERYI: *tempcode = OP_NOTPOSQUERYI; break;
5002 case OP_NOTUPTOI: *tempcode = OP_NOTPOSUPTOI; break;
5003
5004 case OP_TYPESTAR: *tempcode = OP_TYPEPOSSTAR; break;
5005 case OP_TYPEPLUS: *tempcode = OP_TYPEPOSPLUS; break;
5006 case OP_TYPEQUERY: *tempcode = OP_TYPEPOSQUERY; break;
5007 case OP_TYPEUPTO: *tempcode = OP_TYPEPOSUPTO; break;
5008
5009 /* Because we are moving code along, we must ensure that any
5010 pending recursive references are updated. */
5011
5012 default:
5013 *code = OP_END;
5014 adjust_recurse(tempcode, 1 + LINK_SIZE, utf8, cd, save_hwm);
5015 memmove(tempcode + 1+LINK_SIZE, tempcode, len);
5016 code += 1 + LINK_SIZE;
5017 len += 1 + LINK_SIZE;
5018 tempcode[0] = OP_ONCE;
5019 *code++ = OP_KET;
5020 PUTINC(code, 0, len);
5021 PUT(tempcode, 1, len);
5022 break;
5023 }
5024 }
5025
5026 /* In all case we no longer have a previous item. We also set the
5027 "follows varying string" flag for subsequently encountered reqbytes if
5028 it isn't already set and we have just passed a varying length item. */
5029
5030 END_REPEAT:
5031 previous = NULL;
5032 cd->req_varyopt |= reqvary;
5033 break;
5034
5035
5036 /* ===================================================================*/
5037 /* Start of nested parenthesized sub-expression, or comment or lookahead or
5038 lookbehind or option setting or condition or all the other extended
5039 parenthesis forms. */
5040
5041 case CHAR_LEFT_PARENTHESIS:
5042 newoptions = options;
5043 skipbytes = 0;
5044 bravalue = OP_CBRA;
5045 save_hwm = cd->hwm;
5046 reset_bracount = FALSE;
5047
5048 /* First deal with various "verbs" that can be introduced by '*'. */
5049
5050 if (*(++ptr) == CHAR_ASTERISK &&
5051 ((cd->ctypes[ptr[1]] & ctype_letter) != 0 || ptr[1] == ':'))
5052 {
5053 int i, namelen;
5054 int arglen = 0;
5055 const char *vn = verbnames;
5056 const uschar *name = ptr + 1;
5057 const uschar *arg = NULL;
5058 previous = NULL;
5059 while ((cd->ctypes[*++ptr] & ctype_letter) != 0) {};
5060 namelen = (int)(ptr - name);
5061
5062 /* It appears that Perl allows any characters whatsoever, other than
5063 a closing parenthesis, to appear in arguments, so we no longer insist on
5064 letters, digits, and underscores. */
5065
5066 if (*ptr == CHAR_COLON)
5067 {
5068 arg = ++ptr;
5069 while (*ptr != 0 && *ptr != CHAR_RIGHT_PARENTHESIS) ptr++;
5070 arglen = (int)(ptr - arg);
5071 }
5072
5073 if (*ptr != CHAR_RIGHT_PARENTHESIS)
5074 {
5075 *errorcodeptr = ERR60;
5076 goto FAILED;
5077 }
5078
5079 /* Scan the table of verb names */
5080
5081 for (i = 0; i < verbcount; i++)
5082 {
5083 if (namelen == verbs[i].len &&
5084 strncmp((char *)name, vn, namelen) == 0)
5085 {
5086 /* Check for open captures before ACCEPT and convert it to
5087 ASSERT_ACCEPT if in an assertion. */
5088
5089 if (verbs[i].op == OP_ACCEPT)
5090 {
5091 open_capitem *oc;
5092 if (arglen != 0)
5093 {
5094 *errorcodeptr = ERR59;
5095 goto FAILED;
5096 }
5097 cd->had_accept = TRUE;
5098 for (oc = cd->open_caps; oc != NULL; oc = oc->next)
5099 {
5100 *code++ = OP_CLOSE;
5101 PUT2INC(code, 0, oc->number);
5102 }
5103 *code++ = (cd->assert_depth > 0)? OP_ASSERT_ACCEPT : OP_ACCEPT;
5104
5105 /* Do not set firstbyte after *ACCEPT */
5106 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
5107 }
5108
5109 /* Handle other cases with/without an argument */
5110
5111 else if (arglen == 0)
5112 {
5113 if (verbs[i].op < 0) /* Argument is mandatory */
5114 {
5115 *errorcodeptr = ERR66;
5116 goto FAILED;
5117 }
5118 *code = verbs[i].op;
5119 if (*code++ == OP_THEN) cd->external_flags |= PCRE_HASTHEN;
5120 }
5121
5122 else
5123 {
5124 if (verbs[i].op_arg < 0) /* Argument is forbidden */
5125 {
5126 *errorcodeptr = ERR59;
5127 goto FAILED;
5128 }
5129 *code = verbs[i].op_arg;
5130 if (*code++ == OP_THEN_ARG) cd->external_flags |= PCRE_HASTHEN;
5131 *code++ = arglen;
5132 memcpy(code, arg, arglen);
5133 code += arglen;
5134 *code++ = 0;
5135 }
5136
5137 break; /* Found verb, exit loop */
5138 }
5139
5140 vn += verbs[i].len + 1;
5141 }
5142
5143 if (i < verbcount) continue; /* Successfully handled a verb */
5144 *errorcodeptr = ERR60; /* Verb not recognized */
5145 goto FAILED;
5146 }
5147
5148 /* Deal with the extended parentheses; all are introduced by '?', and the
5149 appearance of any of them means that this is not a capturing group. */
5150
5151 else if (*ptr == CHAR_QUESTION_MARK)
5152 {
5153 int i, set, unset, namelen;
5154 int *optset;
5155 const uschar *name;
5156 uschar *slot;
5157
5158 switch (*(++ptr))
5159 {
5160 case CHAR_NUMBER_SIGN: /* Comment; skip to ket */
5161 ptr++;
5162 while (*ptr != 0 && *ptr != CHAR_RIGHT_PARENTHESIS) ptr++;
5163 if (*ptr == 0)
5164 {
5165 *errorcodeptr = ERR18;
5166 goto FAILED;
5167 }
5168 continue;
5169
5170
5171 /* ------------------------------------------------------------ */
5172 case CHAR_VERTICAL_LINE: /* Reset capture count for each branch */
5173 reset_bracount = TRUE;
5174 /* Fall through */
5175
5176 /* ------------------------------------------------------------ */
5177 case CHAR_COLON: /* Non-capturing bracket */
5178 bravalue = OP_BRA;
5179 ptr++;
5180 break;
5181
5182
5183 /* ------------------------------------------------------------ */
5184 case CHAR_LEFT_PARENTHESIS:
5185 bravalue = OP_COND; /* Conditional group */
5186
5187 /* A condition can be an assertion, a number (referring to a numbered
5188 group), a name (referring to a named group), or 'R', referring to
5189 recursion. R<digits> and R&name are also permitted for recursion tests.
5190
5191 There are several syntaxes for testing a named group: (?(name)) is used
5192 by Python; Perl 5.10 onwards uses (?(<name>) or (?('name')).
5193
5194 There are two unfortunate ambiguities, caused by history. (a) 'R' can
5195 be the recursive thing or the name 'R' (and similarly for 'R' followed
5196 by digits), and (b) a number could be a name that consists of digits.
5197 In both cases, we look for a name first; if not found, we try the other
5198 cases. */
5199
5200 /* For conditions that are assertions, check the syntax, and then exit
5201 the switch. This will take control down to where bracketed groups,
5202 including assertions, are processed. */
5203
5204 if (ptr[1] == CHAR_QUESTION_MARK && (ptr[2] == CHAR_EQUALS_SIGN ||
5205 ptr[2] == CHAR_EXCLAMATION_MARK || ptr[2] == CHAR_LESS_THAN_SIGN))
5206 break;
5207
5208 /* Most other conditions use OP_CREF (a couple change to OP_RREF
5209 below), and all need to skip 3 bytes at the start of the group. */
5210
5211 code[1+LINK_SIZE] = OP_CREF;
5212 skipbytes = 3;
5213 refsign = -1;
5214
5215 /* Check for a test for recursion in a named group. */
5216
5217 if (ptr[1] == CHAR_R && ptr[2] == CHAR_AMPERSAND)
5218 {
5219 terminator = -1;
5220 ptr += 2;
5221 code[1+LINK_SIZE] = OP_RREF; /* Change the type of test */
5222 }
5223
5224 /* Check for a test for a named group's having been set, using the Perl
5225 syntax (?(<name>) or (?('name') */
5226
5227 else if (ptr[1] == CHAR_LESS_THAN_SIGN)
5228 {
5229 terminator = CHAR_GREATER_THAN_SIGN;
5230 ptr++;
5231 }
5232 else if (ptr[1] == CHAR_APOSTROPHE)
5233 {
5234 terminator = CHAR_APOSTROPHE;
5235 ptr++;
5236 }
5237 else
5238 {
5239 terminator = 0;
5240 if (ptr[1] == CHAR_MINUS || ptr[1] == CHAR_PLUS) refsign = *(++ptr);
5241 }
5242
5243 /* We now expect to read a name; any thing else is an error */
5244
5245 if ((cd->ctypes[ptr[1]] & ctype_word) == 0)
5246 {
5247 ptr += 1; /* To get the right offset */
5248 *errorcodeptr = ERR28;
5249 goto FAILED;
5250 }
5251
5252 /* Read the name, but also get it as a number if it's all digits */
5253
5254 recno = 0;
5255 name = ++ptr;
5256 while ((cd->ctypes[*ptr] & ctype_word) != 0)
5257 {
5258 if (recno >= 0)
5259 recno = ((digitab[*ptr] & ctype_digit) != 0)?
5260 recno * 10 + *ptr - CHAR_0 : -1;
5261 ptr++;
5262 }
5263 namelen = (int)(ptr - name);
5264
5265 if ((terminator > 0 && *ptr++ != terminator) ||
5266 *ptr++ != CHAR_RIGHT_PARENTHESIS)
5267 {
5268 ptr--; /* Error offset */
5269 *errorcodeptr = ERR26;
5270 goto FAILED;
5271 }
5272
5273 /* Do no further checking in the pre-compile phase. */
5274
5275 if (lengthptr != NULL) break;
5276
5277 /* In the real compile we do the work of looking for the actual
5278 reference. If the string started with "+" or "-" we require the rest to
5279 be digits, in which case recno will be set. */
5280
5281 if (refsign > 0)
5282 {
5283 if (recno <= 0)
5284 {
5285 *errorcodeptr = ERR58;
5286 goto FAILED;
5287 }
5288 recno = (refsign == CHAR_MINUS)?
5289 cd->bracount - recno + 1 : recno +cd->bracount;
5290 if (recno <= 0 || recno > cd->final_bracount)
5291 {
5292 *errorcodeptr = ERR15;
5293 goto FAILED;
5294 }
5295 PUT2(code, 2+LINK_SIZE, recno);
5296 break;
5297 }
5298
5299 /* Otherwise (did not start with "+" or "-"), start by looking for the
5300 name. If we find a name, add one to the opcode to change OP_CREF or
5301 OP_RREF into OP_NCREF or OP_NRREF. These behave exactly the same,
5302 except they record that the reference was originally to a name. The
5303 information is used to check duplicate names. */
5304
5305 slot = cd->name_table;
5306 for (i = 0; i < cd->names_found; i++)
5307 {
5308 if (strncmp((char *)name, (char *)slot+2, namelen) == 0) break;
5309 slot += cd->name_entry_size;
5310 }
5311
5312 /* Found a previous named subpattern */
5313
5314 if (i < cd->names_found)
5315 {
5316 recno = GET2(slot, 0);
5317 PUT2(code, 2+LINK_SIZE, recno);
5318 code[1+LINK_SIZE]++;
5319 }
5320
5321 /* Search the pattern for a forward reference */
5322
5323 else if ((i = find_parens(cd, name, namelen,
5324 (options & PCRE_EXTENDED) != 0, utf8)) > 0)
5325 {
5326 PUT2(code, 2+LINK_SIZE, i);
5327 code[1+LINK_SIZE]++;
5328 }
5329
5330 /* If terminator == 0 it means that the name followed directly after
5331 the opening parenthesis [e.g. (?(abc)...] and in this case there are
5332 some further alternatives to try. For the cases where terminator != 0
5333 [things like (?(<name>... or (?('name')... or (?(R&name)... ] we have
5334 now checked all the possibilities, so give an error. */
5335
5336 else if (terminator != 0)
5337 {
5338 *errorcodeptr = ERR15;
5339 goto FAILED;
5340 }
5341
5342 /* Check for (?(R) for recursion. Allow digits after R to specify a
5343 specific group number. */
5344
5345 else if (*name == CHAR_R)
5346 {
5347 recno = 0;
5348 for (i = 1; i < namelen; i++)
5349 {
5350 if ((digitab[name[i]] & ctype_digit) == 0)
5351 {
5352 *errorcodeptr = ERR15;
5353 goto FAILED;
5354 }
5355 recno = recno * 10 + name[i] - CHAR_0;
5356 }
5357 if (recno == 0) recno = RREF_ANY;
5358 code[1+LINK_SIZE] = OP_RREF; /* Change test type */
5359 PUT2(code, 2+LINK_SIZE, recno);
5360 }
5361
5362 /* Similarly, check for the (?(DEFINE) "condition", which is always
5363 false. */
5364
5365 else if (namelen == 6 && strncmp((char *)name, STRING_DEFINE, 6) == 0)
5366 {
5367 code[1+LINK_SIZE] = OP_DEF;
5368 skipbytes = 1;
5369 }
5370
5371 /* Check for the "name" actually being a subpattern number. We are
5372 in the second pass here, so final_bracount is set. */
5373
5374 else if (recno > 0 && recno <= cd->final_bracount)
5375 {
5376 PUT2(code, 2+LINK_SIZE, recno);
5377 }
5378
5379 /* Either an unidentified subpattern, or a reference to (?(0) */
5380
5381 else
5382 {
5383 *errorcodeptr = (recno == 0)? ERR35: ERR15;
5384 goto FAILED;
5385 }
5386 break;
5387
5388
5389 /* ------------------------------------------------------------ */
5390 case CHAR_EQUALS_SIGN: /* Positive lookahead */
5391 bravalue = OP_ASSERT;
5392 cd->assert_depth += 1;
5393 ptr++;
5394 break;
5395
5396
5397 /* ------------------------------------------------------------ */
5398 case CHAR_EXCLAMATION_MARK: /* Negative lookahead */
5399 ptr++;
5400 if (*ptr == CHAR_RIGHT_PARENTHESIS) /* Optimize (?!) */
5401 {
5402 *code++ = OP_FAIL;
5403 previous = NULL;
5404 continue;
5405 }
5406 bravalue = OP_ASSERT_NOT;
5407 cd->assert_depth += 1;
5408 break;
5409
5410
5411 /* ------------------------------------------------------------ */
5412 case CHAR_LESS_THAN_SIGN: /* Lookbehind or named define */
5413 switch (ptr[1])
5414 {
5415 case CHAR_EQUALS_SIGN: /* Positive lookbehind */
5416 bravalue = OP_ASSERTBACK;
5417 cd->assert_depth += 1;
5418 ptr += 2;
5419 break;
5420
5421 case CHAR_EXCLAMATION_MARK: /* Negative lookbehind */
5422 bravalue = OP_ASSERTBACK_NOT;
5423 cd->assert_depth += 1;
5424 ptr += 2;
5425 break;
5426
5427 default: /* Could be name define, else bad */
5428 if ((cd->ctypes[ptr[1]] & ctype_word) != 0) goto DEFINE_NAME;
5429 ptr++; /* Correct offset for error */
5430 *errorcodeptr = ERR24;
5431 goto FAILED;
5432 }
5433 break;
5434
5435
5436 /* ------------------------------------------------------------ */
5437 case CHAR_GREATER_THAN_SIGN: /* One-time brackets */
5438 bravalue = OP_ONCE;
5439 ptr++;
5440 break;
5441
5442
5443 /* ------------------------------------------------------------ */
5444 case CHAR_C: /* Callout - may be followed by digits; */
5445 previous_callout = code; /* Save for later completion */
5446 after_manual_callout = 1; /* Skip one item before completing */
5447 *code++ = OP_CALLOUT;
5448 {
5449 int n = 0;
5450 while ((digitab[*(++ptr)] & ctype_digit) != 0)
5451 n = n * 10 + *ptr - CHAR_0;
5452 if (*ptr != CHAR_RIGHT_PARENTHESIS)
5453 {
5454 *errorcodeptr = ERR39;
5455 goto FAILED;
5456 }
5457 if (n > 255)
5458 {
5459 *errorcodeptr = ERR38;
5460 goto FAILED;
5461 }
5462 *code++ = n;
5463 PUT(code, 0, (int)(ptr - cd->start_pattern + 1)); /* Pattern offset */
5464 PUT(code, LINK_SIZE, 0); /* Default length */
5465 code += 2 * LINK_SIZE;
5466 }
5467 previous = NULL;
5468 continue;
5469
5470
5471 /* ------------------------------------------------------------ */
5472 case CHAR_P: /* Python-style named subpattern handling */
5473 if (*(++ptr) == CHAR_EQUALS_SIGN ||
5474 *ptr == CHAR_GREATER_THAN_SIGN) /* Reference or recursion */
5475 {
5476 is_recurse = *ptr == CHAR_GREATER_THAN_SIGN;
5477 terminator = CHAR_RIGHT_PARENTHESIS;
5478 goto NAMED_REF_OR_RECURSE;
5479 }
5480 else if (*ptr != CHAR_LESS_THAN_SIGN) /* Test for Python-style defn */
5481 {
5482 *errorcodeptr = ERR41;
5483 goto FAILED;
5484 }
5485 /* Fall through to handle (?P< as (?< is handled */
5486
5487
5488 /* ------------------------------------------------------------ */
5489 DEFINE_NAME: /* Come here from (?< handling */
5490 case CHAR_APOSTROPHE:
5491 {
5492 terminator = (*ptr == CHAR_LESS_THAN_SIGN)?
5493 CHAR_GREATER_THAN_SIGN : CHAR_APOSTROPHE;
5494 name = ++ptr;
5495
5496 while ((cd->ctypes[*ptr] & ctype_word) != 0) ptr++;
5497 namelen = (int)(ptr - name);
5498
5499 /* In the pre-compile phase, just do a syntax check. */
5500
5501 if (lengthptr != NULL)
5502 {
5503 if (*ptr != terminator)
5504 {
5505 *errorcodeptr = ERR42;
5506 goto FAILED;
5507 }
5508 if (cd->names_found >= MAX_NAME_COUNT)
5509 {
5510 *errorcodeptr = ERR49;
5511 goto FAILED;
5512 }
5513 if (namelen + 3 > cd->name_entry_size)
5514 {
5515 cd->name_entry_size = namelen + 3;
5516 if (namelen > MAX_NAME_SIZE)
5517 {
5518 *errorcodeptr = ERR48;
5519 goto FAILED;
5520 }
5521 }
5522 }
5523
5524 /* In the real compile, create the entry in the table, maintaining
5525 alphabetical order. Duplicate names for different numbers are
5526 permitted only if PCRE_DUPNAMES is set. Duplicate names for the same
5527 number are always OK. (An existing number can be re-used if (?|
5528 appears in the pattern.) In either event, a duplicate name results in
5529 a duplicate entry in the table, even if the number is the same. This
5530 is because the number of names, and hence the table size, is computed
5531 in the pre-compile, and it affects various numbers and pointers which
5532 would all have to be modified, and the compiled code moved down, if
5533 duplicates with the same number were omitted from the table. This
5534 doesn't seem worth the hassle. However, *different* names for the
5535 same number are not permitted. */
5536
5537 else
5538 {
5539 BOOL dupname = FALSE;
5540 slot = cd->name_table;
5541
5542 for (i = 0; i < cd->names_found; i++)
5543 {
5544 int crc = memcmp(name, slot+2, namelen);
5545 if (crc == 0)
5546 {
5547 if (slot[2+namelen] == 0)
5548 {
5549 if (GET2(slot, 0) != cd->bracount + 1 &&
5550 (options & PCRE_DUPNAMES) == 0)
5551 {
5552 *errorcodeptr = ERR43;
5553 goto FAILED;
5554 }
5555 else dupname = TRUE;
5556 }
5557 else crc = -1; /* Current name is a substring */
5558 }
5559
5560 /* Make space in the table and break the loop for an earlier
5561 name. For a duplicate or later name, carry on. We do this for
5562 duplicates so that in the simple case (when ?(| is not used) they
5563 are in order of their numbers. */
5564
5565 if (crc < 0)
5566 {
5567 memmove(slot + cd->name_entry_size, slot,
5568 (cd->names_found - i) * cd->name_entry_size);
5569 break;
5570 }
5571
5572 /* Continue the loop for a later or duplicate name */
5573
5574 slot += cd->name_entry_size;
5575 }
5576
5577 /* For non-duplicate names, check for a duplicate number before
5578 adding the new name. */
5579
5580 if (!dupname)
5581 {
5582 uschar *cslot = cd->name_table;
5583 for (i = 0; i < cd->names_found; i++)
5584 {
5585 if (cslot != slot)
5586 {
5587 if (GET2(cslot, 0) == cd->bracount + 1)
5588 {
5589 *errorcodeptr = ERR65;
5590 goto FAILED;
5591 }
5592 }
5593 else i--;
5594 cslot += cd->name_entry_size;
5595 }
5596 }
5597
5598 PUT2(slot, 0, cd->bracount + 1);
5599 memcpy(slot + 2, name, namelen);
5600 slot[2+namelen] = 0;
5601 }
5602 }
5603
5604 /* In both pre-compile and compile, count the number of names we've
5605 encountered. */
5606
5607 cd->names_found++;
5608 ptr++; /* Move past > or ' */
5609 goto NUMBERED_GROUP;
5610
5611
5612 /* ------------------------------------------------------------ */
5613 case CHAR_AMPERSAND: /* Perl recursion/subroutine syntax */
5614 terminator = CHAR_RIGHT_PARENTHESIS;
5615 is_recurse = TRUE;
5616 /* Fall through */
5617
5618 /* We come here from the Python syntax above that handles both
5619 references (?P=name) and recursion (?P>name), as well as falling
5620 through from the Perl recursion syntax (?&name). We also come here from
5621 the Perl \k<name> or \k'name' back reference syntax and the \k{name}
5622 .NET syntax, and the Oniguruma \g<...> and \g'...' subroutine syntax. */
5623
5624 NAMED_REF_OR_RECURSE:
5625 name = ++ptr;
5626 while ((cd->ctypes[*ptr] & ctype_word) != 0) ptr++;
5627 namelen = (int)(ptr - name);
5628
5629 /* In the pre-compile phase, do a syntax check. We used to just set
5630 a dummy reference number, because it was not used in the first pass.
5631 However, with the change of recursive back references to be atomic,
5632 we have to look for the number so that this state can be identified, as
5633 otherwise the incorrect length is computed. If it's not a backwards
5634 reference, the dummy number will do. */
5635
5636 if (lengthptr != NULL)
5637 {
5638 const uschar *temp;
5639
5640 if (namelen == 0)
5641 {
5642 *errorcodeptr = ERR62;
5643 goto FAILED;
5644 }
5645 if (*ptr != terminator)
5646 {
5647 *errorcodeptr = ERR42;
5648 goto FAILED;
5649 }
5650 if (namelen > MAX_NAME_SIZE)
5651 {
5652 *errorcodeptr = ERR48;
5653 goto FAILED;
5654 }
5655
5656 /* The name table does not exist in the first pass, so we cannot
5657 do a simple search as in the code below. Instead, we have to scan the
5658 pattern to find the number. It is important that we scan it only as
5659 far as we have got because the syntax of named subpatterns has not
5660 been checked for the rest of the pattern, and find_parens() assumes
5661 correct syntax. In any case, it's a waste of resources to scan
5662 further. We stop the scan at the current point by temporarily
5663 adjusting the value of cd->endpattern. */
5664
5665 temp = cd->end_pattern;
5666 cd->end_pattern = ptr;
5667 recno = find_parens(cd, name, namelen,
5668 (options & PCRE_EXTENDED) != 0, utf8);
5669 cd->end_pattern = temp;
5670 if (recno < 0) recno = 0; /* Forward ref; set dummy number */
5671 }
5672
5673 /* In the real compile, seek the name in the table. We check the name
5674 first, and then check that we have reached the end of the name in the
5675 table. That way, if the name that is longer than any in the table,
5676 the comparison will fail without reading beyond the table entry. */
5677
5678 else
5679 {
5680 slot = cd->name_table;
5681 for (i = 0; i < cd->names_found; i++)
5682 {
5683 if (strncmp((char *)name, (char *)slot+2, namelen) == 0 &&
5684 slot[2+namelen] == 0)
5685 break;
5686 slot += cd->name_entry_size;
5687 }
5688
5689 if (i < cd->names_found) /* Back reference */
5690 {
5691 recno = GET2(slot, 0);
5692 }
5693 else if ((recno = /* Forward back reference */
5694 find_parens(cd, name, namelen,
5695 (options & PCRE_EXTENDED) != 0, utf8)) <= 0)
5696 {
5697 *errorcodeptr = ERR15;
5698 goto FAILED;
5699 }
5700 }
5701
5702 /* In both phases, we can now go to the code than handles numerical
5703 recursion or backreferences. */
5704
5705 if (is_recurse) goto HANDLE_RECURSION;
5706 else goto HANDLE_REFERENCE;
5707
5708
5709 /* ------------------------------------------------------------ */
5710 case CHAR_R: /* Recursion */
5711 ptr++; /* Same as (?0) */
5712 /* Fall through */
5713
5714
5715 /* ------------------------------------------------------------ */
5716 case CHAR_MINUS: case CHAR_PLUS: /* Recursion or subroutine */
5717 case CHAR_0: case CHAR_1: case CHAR_2: case CHAR_3: case CHAR_4:
5718 case CHAR_5: case CHAR_6: case CHAR_7: case CHAR_8: case CHAR_9:
5719 {
5720 const uschar *called;
5721 terminator = CHAR_RIGHT_PARENTHESIS;
5722
5723 /* Come here from the \g<...> and \g'...' code (Oniguruma
5724 compatibility). However, the syntax has been checked to ensure that
5725 the ... are a (signed) number, so that neither ERR63 nor ERR29 will
5726 be called on this path, nor with the jump to OTHER_CHAR_AFTER_QUERY
5727 ever be taken. */
5728
5729 HANDLE_NUMERICAL_RECURSION:
5730
5731 if ((refsign = *ptr) == CHAR_PLUS)
5732 {
5733 ptr++;
5734 if ((digitab[*ptr] & ctype_digit) == 0)
5735 {
5736 *errorcodeptr = ERR63;
5737 goto FAILED;
5738 }
5739 }
5740 else if (refsign == CHAR_MINUS)
5741 {
5742 if ((digitab[ptr[1]] & ctype_digit) == 0)
5743 goto OTHER_CHAR_AFTER_QUERY;
5744 ptr++;
5745 }
5746
5747 recno = 0;
5748 while((digitab[*ptr] & ctype_digit) != 0)
5749 recno = recno * 10 + *ptr++ - CHAR_0;
5750
5751 if (*ptr != terminator)
5752 {
5753 *errorcodeptr = ERR29;
5754 goto FAILED;
5755 }
5756
5757 if (refsign == CHAR_MINUS)
5758 {
5759 if (recno == 0)
5760 {
5761 *errorcodeptr = ERR58;
5762 goto FAILED;
5763 }
5764 recno = cd->bracount - recno + 1;
5765 if (recno <= 0)
5766 {
5767 *errorcodeptr = ERR15;
5768 goto FAILED;
5769 }
5770 }
5771 else if (refsign == CHAR_PLUS)
5772 {
5773 if (recno == 0)
5774 {
5775 *errorcodeptr = ERR58;
5776 goto FAILED;
5777 }
5778 recno += cd->bracount;
5779 }
5780
5781 /* Come here from code above that handles a named recursion */
5782
5783 HANDLE_RECURSION:
5784
5785 previous = code;
5786 called = cd->start_code;
5787
5788 /* When we are actually compiling, find the bracket that is being
5789 referenced. Temporarily end the regex in case it doesn't exist before
5790 this point. If we end up with a forward reference, first check that
5791 the bracket does occur later so we can give the error (and position)
5792 now. Then remember this forward reference in the workspace so it can
5793 be filled in at the end. */
5794
5795 if (lengthptr == NULL)
5796 {
5797 *code = OP_END;
5798 if (recno != 0)
5799 called = _pcre_find_bracket(cd->start_code, utf8, recno);
5800
5801 /* Forward reference */
5802
5803 if (called == NULL)
5804 {
5805 if (find_parens(cd, NULL, recno,
5806 (options & PCRE_EXTENDED) != 0, utf8) < 0)
5807 {
5808 *errorcodeptr = ERR15;
5809 goto FAILED;
5810 }
5811
5812 /* Fudge the value of "called" so that when it is inserted as an
5813 offset below, what it actually inserted is the reference number
5814 of the group. Then remember the forward reference. */
5815
5816 called = cd->start_code + recno;
5817 PUTINC(cd->hwm, 0, (int)(code + 1 - cd->start_code));
5818 }
5819
5820 /* If not a forward reference, and the subpattern is still open,
5821 this is a recursive call. We check to see if this is a left
5822 recursion that could loop for ever, and diagnose that case. We
5823 must not, however, do this check if we are in a conditional
5824 subpattern because the condition might be testing for recursion in
5825 a pattern such as /(?(R)a+|(?R)b)/, which is perfectly valid.
5826 Forever loops are also detected at runtime, so those that occur in
5827 conditional subpatterns will be picked up then. */
5828
5829 else if (GET(called, 1) == 0 && cond_depth <= 0 &&
5830 could_be_empty(called, code, bcptr, utf8, cd))
5831 {
5832 *errorcodeptr = ERR40;
5833 goto FAILED;
5834 }
5835 }
5836
5837 /* Insert the recursion/subroutine item. */
5838
5839 *code = OP_RECURSE;
5840 PUT(code, 1, (int)(called - cd->start_code));
5841 code += 1 + LINK_SIZE;
5842 }
5843
5844 /* Can't determine a first byte now */
5845
5846 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
5847 continue;
5848
5849
5850 /* ------------------------------------------------------------ */
5851 default: /* Other characters: check option setting */
5852 OTHER_CHAR_AFTER_QUERY:
5853 set = unset = 0;
5854 optset = &set;
5855
5856 while (*ptr != CHAR_RIGHT_PARENTHESIS && *ptr != CHAR_COLON)
5857 {
5858 switch (*ptr++)
5859 {
5860 case CHAR_MINUS: optset = &unset; break;
5861
5862 case CHAR_J: /* Record that it changed in the external options */
5863 *optset |= PCRE_DUPNAMES;
5864 cd->external_flags |= PCRE_JCHANGED;
5865 break;
5866
5867 case CHAR_i: *optset |= PCRE_CASELESS; break;
5868 case CHAR_m: *optset |= PCRE_MULTILINE; break;
5869 case CHAR_s: *optset |= PCRE_DOTALL; break;
5870 case CHAR_x: *optset |= PCRE_EXTENDED; break;
5871 case CHAR_U: *optset |= PCRE_UNGREEDY; break;
5872 case CHAR_X: *optset |= PCRE_EXTRA; break;
5873
5874 default: *errorcodeptr = ERR12;
5875 ptr--; /* Correct the offset */
5876 goto FAILED;
5877 }
5878 }
5879
5880 /* Set up the changed option bits, but don't change anything yet. */
5881
5882 newoptions = (options | set) & (~unset);
5883
5884 /* If the options ended with ')' this is not the start of a nested
5885 group with option changes, so the options change at this level. If this
5886 item is right at the start of the pattern, the options can be
5887 abstracted and made external in the pre-compile phase, and ignored in
5888 the compile phase. This can be helpful when matching -- for instance in
5889 caseless checking of required bytes.
5890
5891 If the code pointer is not (cd->start_code + 1 + LINK_SIZE), we are
5892 definitely *not* at the start of the pattern because something has been
5893 compiled. In the pre-compile phase, however, the code pointer can have
5894 that value after the start, because it gets reset as code is discarded
5895 during the pre-compile. However, this can happen only at top level - if
5896 we are within parentheses, the starting BRA will still be present. At
5897 any parenthesis level, the length value can be used to test if anything
5898 has been compiled at that level. Thus, a test for both these conditions
5899 is necessary to ensure we correctly detect the start of the pattern in
5900 both phases.
5901
5902 If we are not at the pattern start, reset the greedy defaults and the
5903 case value for firstbyte and reqbyte. */
5904
5905 if (*ptr == CHAR_RIGHT_PARENTHESIS)
5906 {
5907 if (code == cd->start_code + 1 + LINK_SIZE &&
5908 (lengthptr == NULL || *lengthptr == 2 + 2*LINK_SIZE))
5909 {
5910 cd->external_options = newoptions;
5911 }
5912 else
5913 {
5914 greedy_default = ((newoptions & PCRE_UNGREEDY) != 0);
5915 greedy_non_default = greedy_default ^ 1;
5916 req_caseopt = ((newoptions & PCRE_CASELESS) != 0)? REQ_CASELESS : 0;
5917 }
5918
5919 /* Change options at this level, and pass them back for use
5920 in subsequent branches. */
5921
5922 *optionsptr = options = newoptions;
5923 previous = NULL; /* This item can't be repeated */
5924 continue; /* It is complete */
5925 }
5926
5927 /* If the options ended with ':' we are heading into a nested group
5928 with possible change of options. Such groups are non-capturing and are
5929 not assertions of any kind. All we need to do is skip over the ':';
5930 the newoptions value is handled below. */
5931
5932 bravalue = OP_BRA;
5933 ptr++;
5934 } /* End of switch for character following (? */
5935 } /* End of (? handling */
5936
5937 /* Opening parenthesis not followed by '*' or '?'. If PCRE_NO_AUTO_CAPTURE
5938 is set, all unadorned brackets become non-capturing and behave like (?:...)
5939 brackets. */
5940
5941 else if ((options & PCRE_NO_AUTO_CAPTURE) != 0)
5942 {
5943 bravalue = OP_BRA;
5944 }
5945
5946 /* Else we have a capturing group. */
5947
5948 else
5949 {
5950 NUMBERED_GROUP:
5951 cd->bracount += 1;
5952 PUT2(code, 1+LINK_SIZE, cd->bracount);
5953 skipbytes = 2;
5954 }
5955
5956 /* Process nested bracketed regex. Assertions used not to be repeatable,
5957 but this was changed for Perl compatibility, so all kinds can now be
5958 repeated. We copy code into a non-register variable (tempcode) in order to
5959 be able to pass its address because some compilers complain otherwise. */
5960
5961 previous = code; /* For handling repetition */
5962 *code = bravalue;
5963 tempcode = code;
5964 tempreqvary = cd->req_varyopt; /* Save value before bracket */
5965 tempbracount = cd->bracount; /* Save value before bracket */
5966 length_prevgroup = 0; /* Initialize for pre-compile phase */
5967
5968 if (!compile_regex(
5969 newoptions, /* The complete new option state */
5970 &tempcode, /* Where to put code (updated) */
5971 &ptr, /* Input pointer (updated) */
5972 errorcodeptr, /* Where to put an error message */
5973 (bravalue == OP_ASSERTBACK ||
5974 bravalue == OP_ASSERTBACK_NOT), /* TRUE if back assert */
5975 reset_bracount, /* True if (?| group */
5976 skipbytes, /* Skip over bracket number */
5977 cond_depth +
5978 ((bravalue == OP_COND)?1:0), /* Depth of condition subpatterns */
5979 &subfirstbyte, /* For possible first char */
5980 &subreqbyte, /* For possible last char */
5981 bcptr, /* Current branch chain */
5982 cd, /* Tables block */
5983 (lengthptr == NULL)? NULL : /* Actual compile phase */
5984 &length_prevgroup /* Pre-compile phase */
5985 ))
5986 goto FAILED;
5987
5988 /* If this was an atomic group and there are no capturing groups within it,
5989 generate OP_ONCE_NC instead of OP_ONCE. */
5990
5991 if (bravalue == OP_ONCE && cd->bracount <= tempbracount)
5992 *code = OP_ONCE_NC;
5993
5994 if (bravalue >= OP_ASSERT && bravalue <= OP_ASSERTBACK_NOT)
5995 cd->assert_depth -= 1;
5996
5997 /* At the end of compiling, code is still pointing to the start of the
5998 group, while tempcode has been updated to point past the end of the group.
5999 The pattern pointer (ptr) is on the bracket.
6000
6001 If this is a conditional bracket, check that there are no more than
6002 two branches in the group, or just one if it's a DEFINE group. We do this
6003 in the real compile phase, not in the pre-pass, where the whole group may
6004 not be available. */
6005
6006 if (bravalue == OP_COND && lengthptr == NULL)
6007 {
6008 uschar *tc = code;
6009 int condcount = 0;
6010
6011 do {
6012 condcount++;
6013 tc += GET(tc,1);
6014 }
6015 while (*tc != OP_KET);
6016
6017 /* A DEFINE group is never obeyed inline (the "condition" is always
6018 false). It must have only one branch. */
6019
6020 if (code[LINK_SIZE+1] == OP_DEF)
6021 {
6022 if (condcount > 1)
6023 {
6024 *errorcodeptr = ERR54;
6025 goto FAILED;
6026 }
6027 bravalue = OP_DEF; /* Just a flag to suppress char handling below */
6028 }
6029
6030 /* A "normal" conditional group. If there is just one branch, we must not
6031 make use of its firstbyte or reqbyte, because this is equivalent to an
6032 empty second branch. */
6033
6034 else
6035 {
6036 if (condcount > 2)
6037 {
6038 *errorcodeptr = ERR27;
6039 goto FAILED;
6040 }
6041 if (condcount == 1) subfirstbyte = subreqbyte = REQ_NONE;
6042 }
6043 }
6044
6045 /* Error if hit end of pattern */
6046
6047 if (*ptr != CHAR_RIGHT_PARENTHESIS)
6048 {
6049 *errorcodeptr = ERR14;
6050 goto FAILED;
6051 }
6052
6053 /* In the pre-compile phase, update the length by the length of the group,
6054 less the brackets at either end. Then reduce the compiled code to just a
6055 set of non-capturing brackets so that it doesn't use much memory if it is
6056 duplicated by a quantifier.*/
6057
6058 if (lengthptr != NULL)
6059 {
6060 if (OFLOW_MAX - *lengthptr < length_prevgroup - 2 - 2*LINK_SIZE)
6061 {
6062 *errorcodeptr = ERR20;
6063 goto FAILED;
6064 }
6065 *lengthptr += length_prevgroup - 2 - 2*LINK_SIZE;
6066 code++; /* This already contains bravalue */
6067 PUTINC(code, 0, 1 + LINK_SIZE);
6068 *code++ = OP_KET;
6069 PUTINC(code, 0, 1 + LINK_SIZE);
6070 break; /* No need to waste time with special character handling */
6071 }
6072
6073 /* Otherwise update the main code pointer to the end of the group. */
6074
6075 code = tempcode;
6076
6077 /* For a DEFINE group, required and first character settings are not
6078 relevant. */
6079
6080 if (bravalue == OP_DEF) break;
6081
6082 /* Handle updating of the required and first characters for other types of
6083 group. Update for normal brackets of all kinds, and conditions with two
6084 branches (see code above). If the bracket is followed by a quantifier with
6085 zero repeat, we have to back off. Hence the definition of zeroreqbyte and
6086 zerofirstbyte outside the main loop so that they can be accessed for the
6087 back off. */
6088
6089 zeroreqbyte = reqbyte;
6090 zerofirstbyte = firstbyte;
6091 groupsetfirstbyte = FALSE;
6092
6093 if (bravalue >= OP_ONCE)
6094 {
6095 /* If we have not yet set a firstbyte in this branch, take it from the
6096 subpattern, remembering that it was set here so that a repeat of more
6097 than one can replicate it as reqbyte if necessary. If the subpattern has
6098 no firstbyte, set "none" for the whole branch. In both cases, a zero
6099 repeat forces firstbyte to "none". */
6100
6101 if (firstbyte == REQ_UNSET)
6102 {
6103 if (subfirstbyte >= 0)
6104 {
6105 firstbyte = subfirstbyte;
6106 groupsetfirstbyte = TRUE;
6107 }
6108 else firstbyte = REQ_NONE;
6109 zerofirstbyte = REQ_NONE;
6110 }
6111
6112 /* If firstbyte was previously set, convert the subpattern's firstbyte
6113 into reqbyte if there wasn't one, using the vary flag that was in
6114 existence beforehand. */
6115
6116 else if (subfirstbyte >= 0 && subreqbyte < 0)
6117 subreqbyte = subfirstbyte | tempreqvary;
6118
6119 /* If the subpattern set a required byte (or set a first byte that isn't
6120 really the first byte - see above), set it. */
6121
6122 if (subreqbyte >= 0) reqbyte = subreqbyte;
6123 }
6124
6125 /* For a forward assertion, we take the reqbyte, if set. This can be
6126 helpful if the pattern that follows the assertion doesn't set a different
6127 char. For example, it's useful for /(?=abcde).+/. We can't set firstbyte
6128 for an assertion, however because it leads to incorrect effect for patterns
6129 such as /(?=a)a.+/ when the "real" "a" would then become a reqbyte instead
6130 of a firstbyte. This is overcome by a scan at the end if there's no
6131 firstbyte, looking for an asserted first char. */
6132
6133 else if (bravalue == OP_ASSERT && subreqbyte >= 0) reqbyte = subreqbyte;
6134 break; /* End of processing '(' */
6135
6136
6137 /* ===================================================================*/
6138 /* Handle metasequences introduced by \. For ones like \d, the ESC_ values
6139 are arranged to be the negation of the corresponding OP_values in the
6140 default case when PCRE_UCP is not set. For the back references, the values
6141 are ESC_REF plus the reference number. Only back references and those types
6142 that consume a character may be repeated. We can test for values between
6143 ESC_b and ESC_Z for the latter; this may have to change if any new ones are
6144 ever created. */
6145
6146 case CHAR_BACKSLASH:
6147 tempptr = ptr;
6148 c = check_escape(&ptr, errorcodeptr, cd->bracount, options, FALSE);
6149 if (*errorcodeptr != 0) goto FAILED;
6150
6151 if (c < 0)
6152 {
6153 if (-c == ESC_Q) /* Handle start of quoted string */
6154 {
6155 if (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
6156 ptr += 2; /* avoid empty string */
6157 else inescq = TRUE;
6158 continue;
6159 }
6160
6161 if (-c == ESC_E) continue; /* Perl ignores an orphan \E */
6162
6163 /* For metasequences that actually match a character, we disable the
6164 setting of a first character if it hasn't already been set. */
6165
6166 if (firstbyte == REQ_UNSET && -c > ESC_b && -c < ESC_Z)
6167 firstbyte = REQ_NONE;
6168
6169 /* Set values to reset to if this is followed by a zero repeat. */
6170
6171 zerofirstbyte = firstbyte;
6172 zeroreqbyte = reqbyte;
6173
6174 /* \g<name> or \g'name' is a subroutine call by name and \g<n> or \g'n'
6175 is a subroutine call by number (Oniguruma syntax). In fact, the value
6176 -ESC_g is returned only for these cases. So we don't need to check for <
6177 or ' if the value is -ESC_g. For the Perl syntax \g{n} the value is
6178 -ESC_REF+n, and for the Perl syntax \g{name} the result is -ESC_k (as
6179 that is a synonym for a named back reference). */
6180
6181 if (-c == ESC_g)
6182 {
6183 const uschar *p;
6184 save_hwm = cd->hwm; /* Normally this is set when '(' is read */
6185 terminator = (*(++ptr) == CHAR_LESS_THAN_SIGN)?
6186 CHAR_GREATER_THAN_SIGN : CHAR_APOSTROPHE;
6187
6188 /* These two statements stop the compiler for warning about possibly
6189 unset variables caused by the jump to HANDLE_NUMERICAL_RECURSION. In
6190 fact, because we actually check for a number below, the paths that
6191 would actually be in error are never taken. */
6192
6193 skipbytes = 0;
6194 reset_bracount = FALSE;
6195
6196 /* Test for a name */
6197
6198 if (ptr[1] != CHAR_PLUS && ptr[1] != CHAR_MINUS)
6199 {
6200 BOOL isnumber = TRUE;
6201 for (p = ptr + 1; *p != 0 && *p != terminator; p++)
6202 {
6203 if ((cd->ctypes[*p] & ctype_digit) == 0) isnumber = FALSE;
6204 if ((cd->ctypes[*p] & ctype_word) == 0) break;
6205 }
6206 if (*p != terminator)
6207 {
6208 *errorcodeptr = ERR57;
6209 break;
6210 }
6211 if (isnumber)
6212 {
6213 ptr++;
6214 goto HANDLE_NUMERICAL_RECURSION;
6215 }
6216 is_recurse = TRUE;
6217 goto NAMED_REF_OR_RECURSE;
6218 }
6219
6220 /* Test a signed number in angle brackets or quotes. */
6221
6222 p = ptr + 2;
6223 while ((digitab[*p] & ctype_digit) != 0) p++;
6224 if (*p != terminator)
6225 {
6226 *errorcodeptr = ERR57;
6227 break;
6228 }
6229 ptr++;
6230 goto HANDLE_NUMERICAL_RECURSION;
6231 }
6232
6233 /* \k<name> or \k'name' is a back reference by name (Perl syntax).
6234 We also support \k{name} (.NET syntax). */
6235
6236 if (-c == ESC_k)
6237 {
6238 if ((ptr[1] != CHAR_LESS_THAN_SIGN &&
6239 ptr[1] != CHAR_APOSTROPHE && ptr[1] != CHAR_LEFT_CURLY_BRACKET))
6240 {
6241 *errorcodeptr = ERR69;
6242 break;
6243 }
6244 is_recurse = FALSE;
6245 terminator = (*(++ptr) == CHAR_LESS_THAN_SIGN)?
6246 CHAR_GREATER_THAN_SIGN : (*ptr == CHAR_APOSTROPHE)?
6247 CHAR_APOSTROPHE : CHAR_RIGHT_CURLY_BRACKET;
6248 goto NAMED_REF_OR_RECURSE;
6249 }
6250
6251 /* Back references are handled specially; must disable firstbyte if
6252 not set to cope with cases like (?=(\w+))\1: which would otherwise set
6253 ':' later. */
6254
6255 if (-c >= ESC_REF)
6256 {
6257 open_capitem *oc;
6258 recno = -c - ESC_REF;
6259
6260 HANDLE_REFERENCE: /* Come here from named backref handling */
6261 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
6262 previous = code;
6263 *code++ = ((options & PCRE_CASELESS) != 0)? OP_REFI : OP_REF;
6264 PUT2INC(code, 0, recno);
6265 cd->backref_map |= (recno < 32)? (1 << recno) : 1;
6266 if (recno > cd->top_backref) cd->top_backref = recno;
6267
6268 /* Check to see if this back reference is recursive, that it, it
6269 is inside the group that it references. A flag is set so that the
6270 group can be made atomic. */
6271
6272 for (oc = cd->open_caps; oc != NULL; oc = oc->next)
6273 {
6274 if (oc->number == recno)
6275 {
6276 oc->flag = TRUE;
6277 break;
6278 }
6279 }
6280 }
6281
6282 /* So are Unicode property matches, if supported. */
6283
6284 #ifdef SUPPORT_UCP
6285 else if (-c == ESC_P || -c == ESC_p)
6286 {
6287 BOOL negated;
6288 int pdata;
6289 int ptype = get_ucp(&ptr, &negated, &pdata, errorcodeptr);
6290 if (ptype < 0) goto FAILED;
6291 previous = code;
6292 *code++ = ((-c == ESC_p) != negated)? OP_PROP : OP_NOTPROP;
6293 *code++ = ptype;
6294 *code++ = pdata;
6295 }
6296 #else
6297
6298 /* If Unicode properties are not supported, \X, \P, and \p are not
6299 allowed. */
6300
6301 else if (-c == ESC_X || -c == ESC_P || -c == ESC_p)
6302 {
6303 *errorcodeptr = ERR45;
6304 goto FAILED;
6305 }
6306 #endif
6307
6308 /* For the rest (including \X when Unicode properties are supported), we
6309 can obtain the OP value by negating the escape value in the default
6310 situation when PCRE_UCP is not set. When it *is* set, we substitute
6311 Unicode property tests. */
6312
6313 else
6314 {
6315 #ifdef SUPPORT_UCP
6316 if (-c >= ESC_DU && -c <= ESC_wu)
6317 {
6318 nestptr = ptr + 1; /* Where to resume */
6319 ptr = substitutes[-c - ESC_DU] - 1; /* Just before substitute */
6320 }
6321 else
6322 #endif
6323 {
6324 previous = (-c > ESC_b && -c < ESC_Z)? code : NULL;
6325 *code++ = -c;
6326 }
6327 }
6328 continue;
6329 }
6330
6331 /* We have a data character whose value is in c. In UTF-8 mode it may have
6332 a value > 127. We set its representation in the length/buffer, and then
6333 handle it as a data character. */
6334
6335 #ifdef SUPPORT_UTF8
6336 if (utf8 && c > 127)
6337 mclength = _pcre_ord2utf8(c, mcbuffer);
6338 else
6339 #endif
6340
6341 {
6342 mcbuffer[0] = c;
6343 mclength = 1;
6344 }
6345 goto ONE_CHAR;
6346
6347
6348 /* ===================================================================*/
6349 /* Handle a literal character. It is guaranteed not to be whitespace or #
6350 when the extended flag is set. If we are in UTF-8 mode, it may be a
6351 multi-byte literal character. */
6352
6353 default:
6354 NORMAL_CHAR:
6355 mclength = 1;
6356 mcbuffer[0] = c;
6357
6358 #ifdef SUPPORT_UTF8
6359 if (utf8 && c >= 0xc0)
6360 {
6361 while ((ptr[1] & 0xc0) == 0x80)
6362 mcbuffer[mclength++] = *(++ptr);
6363 }
6364 #endif
6365
6366 /* At this point we have the character's bytes in mcbuffer, and the length
6367 in mclength. When not in UTF-8 mode, the length is always 1. */
6368
6369 ONE_CHAR:
6370 previous = code;
6371 *code++ = ((options & PCRE_CASELESS) != 0)? OP_CHARI : OP_CHAR;
6372 for (c = 0; c < mclength; c++) *code++ = mcbuffer[c];
6373
6374 /* Remember if \r or \n were seen */
6375
6376 if (mcbuffer[0] == CHAR_CR || mcbuffer[0] == CHAR_NL)
6377 cd->external_flags |= PCRE_HASCRORLF;
6378
6379 /* Set the first and required bytes appropriately. If no previous first
6380 byte, set it from this character, but revert to none on a zero repeat.
6381 Otherwise, leave the firstbyte value alone, and don't change it on a zero
6382 repeat. */
6383
6384 if (firstbyte == REQ_UNSET)
6385 {
6386 zerofirstbyte = REQ_NONE;
6387 zeroreqbyte = reqbyte;
6388
6389 /* If the character is more than one byte long, we can set firstbyte
6390 only if it is not to be matched caselessly. */
6391
6392 if (mclength == 1 || req_caseopt == 0)
6393 {
6394 firstbyte = mcbuffer[0] | req_caseopt;
6395 if (mclength != 1) reqbyte = code[-1] | cd->req_varyopt;
6396 }
6397 else firstbyte = reqbyte = REQ_NONE;
6398 }
6399
6400 /* firstbyte was previously set; we can set reqbyte only if the length is
6401 1 or the matching is caseful. */
6402
6403 else
6404 {
6405 zerofirstbyte = firstbyte;
6406 zeroreqbyte = reqbyte;
6407 if (mclength == 1 || req_caseopt == 0)
6408 reqbyte = code[-1] | req_caseopt | cd->req_varyopt;
6409 }
6410
6411 break; /* End of literal character handling */
6412 }
6413 } /* end of big loop */
6414
6415
6416 /* Control never reaches here by falling through, only by a goto for all the
6417 error states. Pass back the position in the pattern so that it can be displayed
6418 to the user for diagnosing the error. */
6419
6420 FAILED:
6421 *ptrptr = ptr;
6422 return FALSE;
6423 }
6424
6425
6426
6427
6428 /*************************************************
6429 * Compile sequence of alternatives *
6430 *************************************************/
6431
6432 /* On entry, ptr is pointing past the bracket character, but on return it
6433 points to the closing bracket, or vertical bar, or end of string. The code
6434 variable is pointing at the byte into which the BRA operator has been stored.
6435 This function is used during the pre-compile phase when we are trying to find
6436 out the amount of memory needed, as well as during the real compile phase. The
6437 value of lengthptr distinguishes the two phases.
6438
6439 Arguments:
6440 options option bits, including any changes for this subpattern
6441 codeptr -> the address of the current code pointer
6442 ptrptr -> the address of the current pattern pointer
6443 errorcodeptr -> pointer to error code variable
6444 lookbehind TRUE if this is a lookbehind assertion
6445 reset_bracount TRUE to reset the count for each branch
6446 skipbytes skip this many bytes at start (for brackets and OP_COND)
6447 cond_depth depth of nesting for conditional subpatterns
6448 firstbyteptr place to put the first required character, or a negative number
6449 reqbyteptr place to put the last required character, or a negative number
6450 bcptr pointer to the chain of currently open branches
6451 cd points to the data block with tables pointers etc.
6452 lengthptr NULL during the real compile phase
6453 points to length accumulator during pre-compile phase
6454
6455 Returns: TRUE on success
6456 */
6457
6458 static BOOL
6459 compile_regex(int options, uschar **codeptr, const uschar **ptrptr,
6460 int *errorcodeptr, BOOL lookbehind, BOOL reset_bracount, int skipbytes,
6461 int cond_depth, int *firstbyteptr, int *reqbyteptr, branch_chain *bcptr,
6462 compile_data *cd, int *lengthptr)
6463 {
6464 const uschar *ptr = *ptrptr;
6465 uschar *code = *codeptr;
6466 uschar *last_branch = code;
6467 uschar *start_bracket = code;
6468 uschar *reverse_count = NULL;
6469 open_capitem capitem;
6470 int capnumber = 0;
6471 int firstbyte, reqbyte;
6472 int branchfirstbyte, branchreqbyte;
6473 int length;
6474 int orig_bracount;
6475 int max_bracount;
6476 branch_chain bc;
6477
6478 bc.outer = bcptr;
6479 bc.current_branch = code;
6480
6481 firstbyte = reqbyte = REQ_UNSET;
6482
6483 /* Accumulate the length for use in the pre-compile phase. Start with the
6484 length of the BRA and KET and any extra bytes that are required at the
6485 beginning. We accumulate in a local variable to save frequent testing of
6486 lenthptr for NULL. We cannot do this by looking at the value of code at the
6487 start and end of each alternative, because compiled items are discarded during
6488 the pre-compile phase so that the work space is not exceeded. */
6489
6490 length = 2 + 2*LINK_SIZE + skipbytes;
6491
6492 /* WARNING: If the above line is changed for any reason, you must also change
6493 the code that abstracts option settings at the start of the pattern and makes
6494 them global. It tests the value of length for (2 + 2*LINK_SIZE) in the
6495 pre-compile phase to find out whether anything has yet been compiled or not. */
6496
6497 /* If this is a capturing subpattern, add to the chain of open capturing items
6498 so that we can detect them if (*ACCEPT) is encountered. This is also used to
6499 detect groups that contain recursive back references to themselves. Note that
6500 only OP_CBRA need be tested here; changing this opcode to one of its variants,
6501 e.g. OP_SCBRAPOS, happens later, after the group has been compiled. */
6502
6503 if (*code == OP_CBRA)
6504 {
6505 capnumber = GET2(code, 1 + LINK_SIZE);
6506 capitem.number = capnumber;
6507 capitem.next = cd->open_caps;
6508 capitem.flag = FALSE;
6509 cd->open_caps = &capitem;
6510 }
6511
6512 /* Offset is set zero to mark that this bracket is still open */
6513
6514 PUT(code, 1, 0);
6515 code += 1 + LINK_SIZE + skipbytes;
6516
6517 /* Loop for each alternative branch */
6518
6519 orig_bracount = max_bracount = cd->bracount;
6520 for (;;)
6521 {
6522 /* For a (?| group, reset the capturing bracket count so that each branch
6523 uses the same numbers. */
6524
6525 if (reset_bracount) cd->bracount = orig_bracount;
6526
6527 /* Set up dummy OP_REVERSE if lookbehind assertion */
6528
6529 if (lookbehind)
6530 {
6531 *code++ = OP_REVERSE;
6532 reverse_count = code;
6533 PUTINC(code, 0, 0);
6534 length += 1 + LINK_SIZE;
6535 }
6536
6537 /* Now compile the branch; in the pre-compile phase its length gets added
6538 into the length. */
6539
6540 if (!compile_branch(&options, &code, &ptr, errorcodeptr, &branchfirstbyte,
6541 &branchreqbyte, &bc, cond_depth, cd,
6542 (lengthptr == NULL)? NULL : &length))
6543 {
6544 *ptrptr = ptr;
6545 return FALSE;
6546 }
6547
6548 /* Keep the highest bracket count in case (?| was used and some branch
6549 has fewer than the rest. */
6550
6551 if (cd->bracount > max_bracount) max_bracount = cd->bracount;
6552
6553 /* In the real compile phase, there is some post-processing to be done. */
6554
6555 if (lengthptr == NULL)
6556 {
6557 /* If this is the first branch, the firstbyte and reqbyte values for the
6558 branch become the values for the regex. */
6559
6560 if (*last_branch != OP_ALT)
6561 {
6562 firstbyte = branchfirstbyte;
6563 reqbyte = branchreqbyte;
6564 }
6565
6566 /* If this is not the first branch, the first char and reqbyte have to
6567 match the values from all the previous branches, except that if the
6568 previous value for reqbyte didn't have REQ_VARY set, it can still match,
6569 and we set REQ_VARY for the regex. */
6570
6571 else
6572 {
6573 /* If we previously had a firstbyte, but it doesn't match the new branch,
6574 we have to abandon the firstbyte for the regex, but if there was
6575 previously no reqbyte, it takes on the value of the old firstbyte. */
6576
6577 if (firstbyte >= 0 && firstbyte != branchfirstbyte)
6578 {
6579 if (reqbyte < 0) reqbyte = firstbyte;
6580 firstbyte = REQ_NONE;
6581 }
6582
6583 /* If we (now or from before) have no firstbyte, a firstbyte from the
6584 branch becomes a reqbyte if there isn't a branch reqbyte. */
6585
6586 if (firstbyte < 0 && branchfirstbyte >= 0 && branchreqbyte < 0)
6587 branchreqbyte = branchfirstbyte;
6588
6589 /* Now ensure that the reqbytes match */
6590
6591 if ((reqbyte & ~REQ_VARY) != (branchreqbyte & ~REQ_VARY))
6592 reqbyte = REQ_NONE;
6593 else reqbyte |= branchreqbyte; /* To "or" REQ_VARY */
6594 }
6595
6596 /* If lookbehind, check that this branch matches a fixed-length string, and
6597 put the length into the OP_REVERSE item. Temporarily mark the end of the
6598 branch with OP_END. If the branch contains OP_RECURSE, the result is -3
6599 because there may be forward references that we can't check here. Set a
6600 flag to cause another lookbehind check at the end. Why not do it all at the
6601 end? Because common, erroneous checks are picked up here and the offset of
6602 the problem can be shown. */
6603
6604 if (lookbehind)
6605 {
6606 int fixed_length;
6607 *code = OP_END;
6608 fixed_length = find_fixedlength(last_branch, (options & PCRE_UTF8) != 0,
6609 FALSE, cd);
6610 DPRINTF(("fixed length = %d\n", fixed_length));
6611 if (fixed_length == -3)
6612 {
6613 cd->check_lookbehind = TRUE;
6614 }
6615 else if (fixed_length < 0)
6616 {
6617 *errorcodeptr = (fixed_length == -2)? ERR36 : ERR25;
6618 *ptrptr = ptr;
6619 return FALSE;
6620 }
6621 else { PUT(reverse_count, 0, fixed_length); }
6622 }
6623 }
6624
6625 /* Reached end of expression, either ')' or end of pattern. In the real
6626 compile phase, go back through the alternative branches and reverse the chain
6627 of offsets, with the field in the BRA item now becoming an offset to the
6628 first alternative. If there are no alternatives, it points to the end of the
6629 group. The length in the terminating ket is always the length of the whole
6630 bracketed item. Return leaving the pointer at the terminating char. */
6631
6632 if (*ptr != CHAR_VERTICAL_LINE)
6633 {
6634 if (lengthptr == NULL)
6635 {
6636 int branch_length = (int)(code - last_branch);
6637 do
6638 {
6639 int prev_length = GET(last_branch, 1);
6640 PUT(last_branch, 1, branch_length);
6641 branch_length = prev_length;
6642 last_branch -= branch_length;
6643 }
6644 while (branch_length > 0);
6