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

Contents of /code/trunk/pcre_compile.c

Parent Directory Parent Directory | Revision Log Revision Log


Revision 513 - (show annotations) (download)
Mon May 3 11:13:37 2010 UTC (4 years, 2 months ago) by ph10
File MIME type: text/plain
File size: 225150 byte(s)
Make \R and \X in a character class behave more like Perl

1 /*************************************************
2 * Perl-Compatible Regular Expressions *
3 *************************************************/
4
5 /* PCRE is a library of functions to support regular expressions whose syntax
6 and semantics are as close as possible to those of the Perl 5 language.
7
8 Written by Philip Hazel
9 Copyright (c) 1997-2010 University of Cambridge
10
11 -----------------------------------------------------------------------------
12 Redistribution and use in source and binary forms, with or without
13 modification, are permitted provided that the following conditions are met:
14
15 * Redistributions of source code must retain the above copyright notice,
16 this list of conditions and the following disclaimer.
17
18 * Redistributions in binary form must reproduce the above copyright
19 notice, this list of conditions and the following disclaimer in the
20 documentation and/or other materials provided with the distribution.
21
22 * Neither the name of the University of Cambridge nor the names of its
23 contributors may be used to endorse or promote products derived from
24 this software without specific prior written permission.
25
26 THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
27 AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28 IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29 ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE
30 LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
31 CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
32 SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
33 INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
34 CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
35 ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
36 POSSIBILITY OF SUCH DAMAGE.
37 -----------------------------------------------------------------------------
38 */
39
40
41 /* This module contains the external function pcre_compile(), along with
42 supporting internal functions that are not used by other modules. */
43
44
45 #ifdef HAVE_CONFIG_H
46 #include "config.h"
47 #endif
48
49 #define NLBLOCK cd /* Block containing newline information */
50 #define PSSTART start_pattern /* Field containing processed string start */
51 #define PSEND end_pattern /* Field containing processed string end */
52
53 #include "pcre_internal.h"
54
55
56 /* When PCRE_DEBUG is defined, we need the pcre_printint() function, which is
57 also used by pcretest. PCRE_DEBUG is not defined when building a production
58 library. */
59
60 #ifdef PCRE_DEBUG
61 #include "pcre_printint.src"
62 #endif
63
64
65 /* Macro for setting individual bits in class bitmaps. */
66
67 #define SETBIT(a,b) a[b/8] |= (1 << (b%8))
68
69 /* Maximum length value to check against when making sure that the integer that
70 holds the compiled pattern length does not overflow. We make it a bit less than
71 INT_MAX to allow for adding in group terminating bytes, so that we don't have
72 to check them every time. */
73
74 #define OFLOW_MAX (INT_MAX - 20)
75
76
77 /*************************************************
78 * Code parameters and static tables *
79 *************************************************/
80
81 /* This value specifies the size of stack workspace that is used during the
82 first pre-compile phase that determines how much memory is required. The regex
83 is partly compiled into this space, but the compiled parts are discarded as
84 soon as they can be, so that hopefully there will never be an overrun. The code
85 does, however, check for an overrun. The largest amount I've seen used is 218,
86 so this number is very generous.
87
88 The same workspace is used during the second, actual compile phase for
89 remembering forward references to groups so that they can be filled in at the
90 end. Each entry in this list occupies LINK_SIZE bytes, so even when LINK_SIZE
91 is 4 there is plenty of room. */
92
93 #define COMPILE_WORK_SIZE (4096)
94
95 /* The overrun tests check for a slightly smaller size so that they detect the
96 overrun before it actually does run off the end of the data block. */
97
98 #define WORK_SIZE_CHECK (COMPILE_WORK_SIZE - 100)
99
100
101 /* Table for handling escaped characters in the range '0'-'z'. Positive returns
102 are simple data values; negative values are for special things like \d and so
103 on. Zero means further processing is needed (for things like \x), or the escape
104 is invalid. */
105
106 #ifndef EBCDIC
107
108 /* This is the "normal" table for ASCII systems or for EBCDIC systems running
109 in UTF-8 mode. */
110
111 static const short int escapes[] = {
112 0, 0,
113 0, 0,
114 0, 0,
115 0, 0,
116 0, 0,
117 CHAR_COLON, CHAR_SEMICOLON,
118 CHAR_LESS_THAN_SIGN, CHAR_EQUALS_SIGN,
119 CHAR_GREATER_THAN_SIGN, CHAR_QUESTION_MARK,
120 CHAR_COMMERCIAL_AT, -ESC_A,
121 -ESC_B, -ESC_C,
122 -ESC_D, -ESC_E,
123 0, -ESC_G,
124 -ESC_H, 0,
125 0, -ESC_K,
126 0, 0,
127 0, 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, 0, 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
265 #define STRING(a) # a
266 #define XSTRING(s) STRING(s)
267
268 /* The texts of compile-time error messages. These are "char *" because they
269 are passed to the outside world. Do not ever re-use any error number, because
270 they are documented. Always add a new error instead. Messages marked DEAD below
271 are no longer used. This used to be a table of strings, but in order to reduce
272 the number of relocations needed when a shared library is loaded dynamically,
273 it is now one long string. We cannot use a table of offsets, because the
274 lengths of inserts such as XSTRING(MAX_NAME_SIZE) are not known. Instead, we
275 simply count through to the one we want - this isn't a performance issue
276 because these strings are used only when there is a compilation error.
277
278 Each substring ends with \0 to insert a null character. This includes the final
279 substring, so that the whole string ends with \0\0, which can be detected when
280 counting through. */
281
282 static const char error_texts[] =
283 "no error\0"
284 "\\ at end of pattern\0"
285 "\\c at end of pattern\0"
286 "unrecognized character follows \\\0"
287 "numbers out of order in {} quantifier\0"
288 /* 5 */
289 "number too big in {} quantifier\0"
290 "missing terminating ] for character class\0"
291 "invalid escape sequence in character class\0"
292 "range out of order in character class\0"
293 "nothing to repeat\0"
294 /* 10 */
295 "operand of unlimited repeat could match the empty string\0" /** DEAD **/
296 "internal error: unexpected repeat\0"
297 "unrecognized character after (? or (?-\0"
298 "POSIX named classes are supported only within a class\0"
299 "missing )\0"
300 /* 15 */
301 "reference to non-existent subpattern\0"
302 "erroffset passed as NULL\0"
303 "unknown option bit(s) set\0"
304 "missing ) after comment\0"
305 "parentheses nested too deeply\0" /** DEAD **/
306 /* 20 */
307 "regular expression is too large\0"
308 "failed to get memory\0"
309 "unmatched parentheses\0"
310 "internal error: code overflow\0"
311 "unrecognized character after (?<\0"
312 /* 25 */
313 "lookbehind assertion is not fixed length\0"
314 "malformed number or name after (?(\0"
315 "conditional group contains more than two branches\0"
316 "assertion expected after (?(\0"
317 "(?R or (?[+-]digits must be followed by )\0"
318 /* 30 */
319 "unknown POSIX class name\0"
320 "POSIX collating elements are not supported\0"
321 "this version of PCRE is not compiled with PCRE_UTF8 support\0"
322 "spare error\0" /** DEAD **/
323 "character value in \\x{...} sequence is too large\0"
324 /* 35 */
325 "invalid condition (?(0)\0"
326 "\\C not allowed in lookbehind assertion\0"
327 "PCRE does not support \\L, \\l, \\N, \\U, or \\u\0"
328 "number after (?C is > 255\0"
329 "closing ) for (?C expected\0"
330 /* 40 */
331 "recursive call could loop indefinitely\0"
332 "unrecognized character after (?P\0"
333 "syntax error in subpattern name (missing terminator)\0"
334 "two named subpatterns have the same name\0"
335 "invalid UTF-8 string\0"
336 /* 45 */
337 "support for \\P, \\p, and \\X has not been compiled\0"
338 "malformed \\P or \\p sequence\0"
339 "unknown property name after \\P or \\p\0"
340 "subpattern name is too long (maximum " XSTRING(MAX_NAME_SIZE) " characters)\0"
341 "too many named subpatterns (maximum " XSTRING(MAX_NAME_COUNT) ")\0"
342 /* 50 */
343 "repeated subpattern is too long\0" /** DEAD **/
344 "octal value is greater than \\377 (not in UTF-8 mode)\0"
345 "internal error: overran compiling workspace\0"
346 "internal error: previously-checked referenced subpattern not found\0"
347 "DEFINE group contains more than one branch\0"
348 /* 55 */
349 "repeating a DEFINE group is not allowed\0"
350 "inconsistent NEWLINE options\0"
351 "\\g is not followed by a braced, angle-bracketed, or quoted name/number or by a plain number\0"
352 "a numbered reference must not be zero\0"
353 "an argument is not allowed for (*ACCEPT), (*FAIL), or (*COMMIT)\0"
354 /* 60 */
355 "(*VERB) not recognized\0"
356 "number is too big\0"
357 "subpattern name expected\0"
358 "digit expected after (?+\0"
359 "] is an invalid data character in JavaScript compatibility mode\0"
360 /* 65 */
361 "different names for subpatterns of the same number are not allowed\0"
362 "(*MARK) must have an argument\0"
363 ;
364
365 /* Table to identify digits and hex digits. This is used when compiling
366 patterns. Note that the tables in chartables are dependent on the locale, and
367 may mark arbitrary characters as digits - but the PCRE compiling code expects
368 to handle only 0-9, a-z, and A-Z as digits when compiling. That is why we have
369 a private table here. It costs 256 bytes, but it is a lot faster than doing
370 character value tests (at least in some simple cases I timed), and in some
371 applications one wants PCRE to compile efficiently as well as match
372 efficiently.
373
374 For convenience, we use the same bit definitions as in chartables:
375
376 0x04 decimal digit
377 0x08 hexadecimal digit
378
379 Then we can use ctype_digit and ctype_xdigit in the code. */
380
381 #ifndef EBCDIC
382
383 /* This is the "normal" case, for ASCII systems, and EBCDIC systems running in
384 UTF-8 mode. */
385
386 static const unsigned char digitab[] =
387 {
388 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 0- 7 */
389 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 8- 15 */
390 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 16- 23 */
391 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 24- 31 */
392 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - ' */
393 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* ( - / */
394 0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c, /* 0 - 7 */
395 0x0c,0x0c,0x00,0x00,0x00,0x00,0x00,0x00, /* 8 - ? */
396 0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* @ - G */
397 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* H - O */
398 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* P - W */
399 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* X - _ */
400 0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* ` - g */
401 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* h - o */
402 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* p - w */
403 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* x -127 */
404 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 128-135 */
405 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 136-143 */
406 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 144-151 */
407 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 152-159 */
408 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 160-167 */
409 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 168-175 */
410 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 176-183 */
411 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 184-191 */
412 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 192-199 */
413 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 200-207 */
414 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 208-215 */
415 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 216-223 */
416 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 224-231 */
417 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 232-239 */
418 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 240-247 */
419 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00};/* 248-255 */
420
421 #else
422
423 /* This is the "abnormal" case, for EBCDIC systems not running in UTF-8 mode. */
424
425 static const unsigned char digitab[] =
426 {
427 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 0- 7 0 */
428 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 8- 15 */
429 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 16- 23 10 */
430 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 24- 31 */
431 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 32- 39 20 */
432 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 40- 47 */
433 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 48- 55 30 */
434 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 56- 63 */
435 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - 71 40 */
436 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 72- | */
437 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* & - 87 50 */
438 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 88- 95 */
439 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - -103 60 */
440 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 104- ? */
441 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 112-119 70 */
442 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 120- " */
443 0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* 128- g 80 */
444 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* h -143 */
445 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 144- p 90 */
446 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* q -159 */
447 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 160- x A0 */
448 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* y -175 */
449 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* ^ -183 B0 */
450 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 184-191 */
451 0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* { - G C0 */
452 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* H -207 */
453 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* } - P D0 */
454 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* Q -223 */
455 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* \ - X E0 */
456 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* Y -239 */
457 0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c, /* 0 - 7 F0 */
458 0x0c,0x0c,0x00,0x00,0x00,0x00,0x00,0x00};/* 8 -255 */
459
460 static const unsigned char ebcdic_chartab[] = { /* chartable partial dup */
461 0x80,0x00,0x00,0x00,0x00,0x01,0x00,0x00, /* 0- 7 */
462 0x00,0x00,0x00,0x00,0x01,0x01,0x00,0x00, /* 8- 15 */
463 0x00,0x00,0x00,0x00,0x00,0x01,0x00,0x00, /* 16- 23 */
464 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 24- 31 */
465 0x00,0x00,0x00,0x00,0x00,0x01,0x00,0x00, /* 32- 39 */
466 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 40- 47 */
467 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 48- 55 */
468 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 56- 63 */
469 0x01,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - 71 */
470 0x00,0x00,0x00,0x80,0x00,0x80,0x80,0x80, /* 72- | */
471 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* & - 87 */
472 0x00,0x00,0x00,0x80,0x80,0x80,0x00,0x00, /* 88- 95 */
473 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - -103 */
474 0x00,0x00,0x00,0x00,0x00,0x10,0x00,0x80, /* 104- ? */
475 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 112-119 */
476 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 120- " */
477 0x00,0x1a,0x1a,0x1a,0x1a,0x1a,0x1a,0x12, /* 128- g */
478 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* h -143 */
479 0x00,0x12,0x12,0x12,0x12,0x12,0x12,0x12, /* 144- p */
480 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* q -159 */
481 0x00,0x00,0x12,0x12,0x12,0x12,0x12,0x12, /* 160- x */
482 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* y -175 */
483 0x80,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* ^ -183 */
484 0x00,0x00,0x80,0x00,0x00,0x00,0x00,0x00, /* 184-191 */
485 0x80,0x1a,0x1a,0x1a,0x1a,0x1a,0x1a,0x12, /* { - G */
486 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* H -207 */
487 0x00,0x12,0x12,0x12,0x12,0x12,0x12,0x12, /* } - P */
488 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* Q -223 */
489 0x00,0x00,0x12,0x12,0x12,0x12,0x12,0x12, /* \ - X */
490 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* Y -239 */
491 0x1c,0x1c,0x1c,0x1c,0x1c,0x1c,0x1c,0x1c, /* 0 - 7 */
492 0x1c,0x1c,0x00,0x00,0x00,0x00,0x00,0x00};/* 8 -255 */
493 #endif
494
495
496 /* Definition to allow mutual recursion */
497
498 static BOOL
499 compile_regex(int, int, uschar **, const uschar **, int *, BOOL, BOOL, int,
500 int *, int *, branch_chain *, compile_data *, int *);
501
502
503
504 /*************************************************
505 * Find an error text *
506 *************************************************/
507
508 /* The error texts are now all in one long string, to save on relocations. As
509 some of the text is of unknown length, we can't use a table of offsets.
510 Instead, just count through the strings. This is not a performance issue
511 because it happens only when there has been a compilation error.
512
513 Argument: the error number
514 Returns: pointer to the error string
515 */
516
517 static const char *
518 find_error_text(int n)
519 {
520 const char *s = error_texts;
521 for (; n > 0; n--)
522 {
523 while (*s++ != 0) {};
524 if (*s == 0) return "Error text not found (please report)";
525 }
526 return s;
527 }
528
529
530 /*************************************************
531 * Handle escapes *
532 *************************************************/
533
534 /* This function is called when a \ has been encountered. It either returns a
535 positive value for a simple escape such as \n, or a negative value which
536 encodes one of the more complicated things such as \d. A backreference to group
537 n is returned as -(ESC_REF + n); ESC_REF is the highest ESC_xxx macro. When
538 UTF-8 is enabled, a positive value greater than 255 may be returned. On entry,
539 ptr is pointing at the \. On exit, it is on the final character of the escape
540 sequence.
541
542 Arguments:
543 ptrptr points to the pattern position pointer
544 errorcodeptr points to the errorcode variable
545 bracount number of previous extracting brackets
546 options the options bits
547 isclass TRUE if inside a character class
548
549 Returns: zero or positive => a data character
550 negative => a special escape sequence
551 on error, errorcodeptr is set
552 */
553
554 static int
555 check_escape(const uschar **ptrptr, int *errorcodeptr, int bracount,
556 int options, BOOL isclass)
557 {
558 BOOL utf8 = (options & PCRE_UTF8) != 0;
559 const uschar *ptr = *ptrptr + 1;
560 int c, i;
561
562 GETCHARINCTEST(c, ptr); /* Get character value, increment pointer */
563 ptr--; /* Set pointer back to the last byte */
564
565 /* If backslash is at the end of the pattern, it's an error. */
566
567 if (c == 0) *errorcodeptr = ERR1;
568
569 /* Non-alphanumerics are literals. For digits or letters, do an initial lookup
570 in a table. A non-zero result is something that can be returned immediately.
571 Otherwise further processing may be required. */
572
573 #ifndef EBCDIC /* ASCII/UTF-8 coding */
574 else if (c < CHAR_0 || c > CHAR_z) {} /* Not alphanumeric */
575 else if ((i = escapes[c - CHAR_0]) != 0) c = i;
576
577 #else /* EBCDIC coding */
578 else if (c < 'a' || (ebcdic_chartab[c] & 0x0E) == 0) {} /* Not alphanumeric */
579 else if ((i = escapes[c - 0x48]) != 0) c = i;
580 #endif
581
582 /* Escapes that need further processing, or are illegal. */
583
584 else
585 {
586 const uschar *oldptr;
587 BOOL braced, negated;
588
589 switch (c)
590 {
591 /* A number of Perl escapes are not handled by PCRE. We give an explicit
592 error. */
593
594 case CHAR_l:
595 case CHAR_L:
596 case CHAR_N:
597 case CHAR_u:
598 case CHAR_U:
599 *errorcodeptr = ERR37;
600 break;
601
602 /* \g must be followed by one of a number of specific things:
603
604 (1) A number, either plain or braced. If positive, it is an absolute
605 backreference. If negative, it is a relative backreference. This is a Perl
606 5.10 feature.
607
608 (2) Perl 5.10 also supports \g{name} as a reference to a named group. This
609 is part of Perl's movement towards a unified syntax for back references. As
610 this is synonymous with \k{name}, we fudge it up by pretending it really
611 was \k.
612
613 (3) For Oniguruma compatibility we also support \g followed by a name or a
614 number either in angle brackets or in single quotes. However, these are
615 (possibly recursive) subroutine calls, _not_ backreferences. Just return
616 the -ESC_g code (cf \k). */
617
618 case CHAR_g:
619 if (ptr[1] == CHAR_LESS_THAN_SIGN || ptr[1] == CHAR_APOSTROPHE)
620 {
621 c = -ESC_g;
622 break;
623 }
624
625 /* Handle the Perl-compatible cases */
626
627 if (ptr[1] == CHAR_LEFT_CURLY_BRACKET)
628 {
629 const uschar *p;
630 for (p = ptr+2; *p != 0 && *p != CHAR_RIGHT_CURLY_BRACKET; p++)
631 if (*p != CHAR_MINUS && (digitab[*p] & ctype_digit) == 0) break;
632 if (*p != 0 && *p != CHAR_RIGHT_CURLY_BRACKET)
633 {
634 c = -ESC_k;
635 break;
636 }
637 braced = TRUE;
638 ptr++;
639 }
640 else braced = FALSE;
641
642 if (ptr[1] == CHAR_MINUS)
643 {
644 negated = TRUE;
645 ptr++;
646 }
647 else negated = FALSE;
648
649 c = 0;
650 while ((digitab[ptr[1]] & ctype_digit) != 0)
651 c = c * 10 + *(++ptr) - CHAR_0;
652
653 if (c < 0) /* Integer overflow */
654 {
655 *errorcodeptr = ERR61;
656 break;
657 }
658
659 if (braced && *(++ptr) != CHAR_RIGHT_CURLY_BRACKET)
660 {
661 *errorcodeptr = ERR57;
662 break;
663 }
664
665 if (c == 0)
666 {
667 *errorcodeptr = ERR58;
668 break;
669 }
670
671 if (negated)
672 {
673 if (c > bracount)
674 {
675 *errorcodeptr = ERR15;
676 break;
677 }
678 c = bracount - (c - 1);
679 }
680
681 c = -(ESC_REF + c);
682 break;
683
684 /* The handling of escape sequences consisting of a string of digits
685 starting with one that is not zero is not straightforward. By experiment,
686 the way Perl works seems to be as follows:
687
688 Outside a character class, the digits are read as a decimal number. If the
689 number is less than 10, or if there are that many previous extracting
690 left brackets, then it is a back reference. Otherwise, up to three octal
691 digits are read to form an escaped byte. Thus \123 is likely to be octal
692 123 (cf \0123, which is octal 012 followed by the literal 3). If the octal
693 value is greater than 377, the least significant 8 bits are taken. Inside a
694 character class, \ followed by a digit is always an octal number. */
695
696 case CHAR_1: case CHAR_2: case CHAR_3: case CHAR_4: case CHAR_5:
697 case CHAR_6: case CHAR_7: case CHAR_8: case CHAR_9:
698
699 if (!isclass)
700 {
701 oldptr = ptr;
702 c -= CHAR_0;
703 while ((digitab[ptr[1]] & ctype_digit) != 0)
704 c = c * 10 + *(++ptr) - CHAR_0;
705 if (c < 0) /* Integer overflow */
706 {
707 *errorcodeptr = ERR61;
708 break;
709 }
710 if (c < 10 || c <= bracount)
711 {
712 c = -(ESC_REF + c);
713 break;
714 }
715 ptr = oldptr; /* Put the pointer back and fall through */
716 }
717
718 /* Handle an octal number following \. If the first digit is 8 or 9, Perl
719 generates a binary zero byte and treats the digit as a following literal.
720 Thus we have to pull back the pointer by one. */
721
722 if ((c = *ptr) >= CHAR_8)
723 {
724 ptr--;
725 c = 0;
726 break;
727 }
728
729 /* \0 always starts an octal number, but we may drop through to here with a
730 larger first octal digit. The original code used just to take the least
731 significant 8 bits of octal numbers (I think this is what early Perls used
732 to do). Nowadays we allow for larger numbers in UTF-8 mode, but no more
733 than 3 octal digits. */
734
735 case CHAR_0:
736 c -= CHAR_0;
737 while(i++ < 2 && ptr[1] >= CHAR_0 && ptr[1] <= CHAR_7)
738 c = c * 8 + *(++ptr) - CHAR_0;
739 if (!utf8 && c > 255) *errorcodeptr = ERR51;
740 break;
741
742 /* \x is complicated. \x{ddd} is a character number which can be greater
743 than 0xff in utf8 mode, but only if the ddd are hex digits. If not, { is
744 treated as a data character. */
745
746 case CHAR_x:
747 if (ptr[1] == CHAR_LEFT_CURLY_BRACKET)
748 {
749 const uschar *pt = ptr + 2;
750 int count = 0;
751
752 c = 0;
753 while ((digitab[*pt] & ctype_xdigit) != 0)
754 {
755 register int cc = *pt++;
756 if (c == 0 && cc == CHAR_0) continue; /* Leading zeroes */
757 count++;
758
759 #ifndef EBCDIC /* ASCII/UTF-8 coding */
760 if (cc >= CHAR_a) cc -= 32; /* Convert to upper case */
761 c = (c << 4) + cc - ((cc < CHAR_A)? CHAR_0 : (CHAR_A - 10));
762 #else /* EBCDIC coding */
763 if (cc >= CHAR_a && cc <= CHAR_z) cc += 64; /* Convert to upper case */
764 c = (c << 4) + cc - ((cc >= CHAR_0)? CHAR_0 : (CHAR_A - 10));
765 #endif
766 }
767
768 if (*pt == CHAR_RIGHT_CURLY_BRACKET)
769 {
770 if (c < 0 || count > (utf8? 8 : 2)) *errorcodeptr = ERR34;
771 ptr = pt;
772 break;
773 }
774
775 /* If the sequence of hex digits does not end with '}', then we don't
776 recognize this construct; fall through to the normal \x handling. */
777 }
778
779 /* Read just a single-byte hex-defined char */
780
781 c = 0;
782 while (i++ < 2 && (digitab[ptr[1]] & ctype_xdigit) != 0)
783 {
784 int cc; /* Some compilers don't like */
785 cc = *(++ptr); /* ++ in initializers */
786 #ifndef EBCDIC /* ASCII/UTF-8 coding */
787 if (cc >= CHAR_a) cc -= 32; /* Convert to upper case */
788 c = c * 16 + cc - ((cc < CHAR_A)? CHAR_0 : (CHAR_A - 10));
789 #else /* EBCDIC coding */
790 if (cc <= CHAR_z) cc += 64; /* Convert to upper case */
791 c = c * 16 + cc - ((cc >= CHAR_0)? CHAR_0 : (CHAR_A - 10));
792 #endif
793 }
794 break;
795
796 /* For \c, a following letter is upper-cased; then the 0x40 bit is flipped.
797 This coding is ASCII-specific, but then the whole concept of \cx is
798 ASCII-specific. (However, an EBCDIC equivalent has now been added.) */
799
800 case CHAR_c:
801 c = *(++ptr);
802 if (c == 0)
803 {
804 *errorcodeptr = ERR2;
805 break;
806 }
807
808 #ifndef EBCDIC /* ASCII/UTF-8 coding */
809 if (c >= CHAR_a && c <= CHAR_z) c -= 32;
810 c ^= 0x40;
811 #else /* EBCDIC coding */
812 if (c >= CHAR_a && c <= CHAR_z) c += 64;
813 c ^= 0xC0;
814 #endif
815 break;
816
817 /* PCRE_EXTRA enables extensions to Perl in the matter of escapes. Any
818 other alphanumeric following \ is an error if PCRE_EXTRA was set;
819 otherwise, for Perl compatibility, it is a literal. This code looks a bit
820 odd, but there used to be some cases other than the default, and there may
821 be again in future, so I haven't "optimized" it. */
822
823 default:
824 if ((options & PCRE_EXTRA) != 0) switch(c)
825 {
826 default:
827 *errorcodeptr = ERR3;
828 break;
829 }
830 break;
831 }
832 }
833
834 *ptrptr = ptr;
835 return c;
836 }
837
838
839
840 #ifdef SUPPORT_UCP
841 /*************************************************
842 * Handle \P and \p *
843 *************************************************/
844
845 /* This function is called after \P or \p has been encountered, provided that
846 PCRE is compiled with support for Unicode properties. On entry, ptrptr is
847 pointing at the P or p. On exit, it is pointing at the final character of the
848 escape sequence.
849
850 Argument:
851 ptrptr points to the pattern position pointer
852 negptr points to a boolean that is set TRUE for negation else FALSE
853 dptr points to an int that is set to the detailed property value
854 errorcodeptr points to the error code variable
855
856 Returns: type value from ucp_type_table, or -1 for an invalid type
857 */
858
859 static int
860 get_ucp(const uschar **ptrptr, BOOL *negptr, int *dptr, int *errorcodeptr)
861 {
862 int c, i, bot, top;
863 const uschar *ptr = *ptrptr;
864 char name[32];
865
866 c = *(++ptr);
867 if (c == 0) goto ERROR_RETURN;
868
869 *negptr = FALSE;
870
871 /* \P or \p can be followed by a name in {}, optionally preceded by ^ for
872 negation. */
873
874 if (c == CHAR_LEFT_CURLY_BRACKET)
875 {
876 if (ptr[1] == CHAR_CIRCUMFLEX_ACCENT)
877 {
878 *negptr = TRUE;
879 ptr++;
880 }
881 for (i = 0; i < (int)sizeof(name) - 1; i++)
882 {
883 c = *(++ptr);
884 if (c == 0) goto ERROR_RETURN;
885 if (c == CHAR_RIGHT_CURLY_BRACKET) break;
886 name[i] = c;
887 }
888 if (c != CHAR_RIGHT_CURLY_BRACKET) goto ERROR_RETURN;
889 name[i] = 0;
890 }
891
892 /* Otherwise there is just one following character */
893
894 else
895 {
896 name[0] = c;
897 name[1] = 0;
898 }
899
900 *ptrptr = ptr;
901
902 /* Search for a recognized property name using binary chop */
903
904 bot = 0;
905 top = _pcre_utt_size;
906
907 while (bot < top)
908 {
909 i = (bot + top) >> 1;
910 c = strcmp(name, _pcre_utt_names + _pcre_utt[i].name_offset);
911 if (c == 0)
912 {
913 *dptr = _pcre_utt[i].value;
914 return _pcre_utt[i].type;
915 }
916 if (c > 0) bot = i + 1; else top = i;
917 }
918
919 *errorcodeptr = ERR47;
920 *ptrptr = ptr;
921 return -1;
922
923 ERROR_RETURN:
924 *errorcodeptr = ERR46;
925 *ptrptr = ptr;
926 return -1;
927 }
928 #endif
929
930
931
932
933 /*************************************************
934 * Check for counted repeat *
935 *************************************************/
936
937 /* This function is called when a '{' is encountered in a place where it might
938 start a quantifier. It looks ahead to see if it really is a quantifier or not.
939 It is only a quantifier if it is one of the forms {ddd} {ddd,} or {ddd,ddd}
940 where the ddds are digits.
941
942 Arguments:
943 p pointer to the first char after '{'
944
945 Returns: TRUE or FALSE
946 */
947
948 static BOOL
949 is_counted_repeat(const uschar *p)
950 {
951 if ((digitab[*p++] & ctype_digit) == 0) return FALSE;
952 while ((digitab[*p] & ctype_digit) != 0) p++;
953 if (*p == CHAR_RIGHT_CURLY_BRACKET) return TRUE;
954
955 if (*p++ != CHAR_COMMA) return FALSE;
956 if (*p == CHAR_RIGHT_CURLY_BRACKET) return TRUE;
957
958 if ((digitab[*p++] & ctype_digit) == 0) return FALSE;
959 while ((digitab[*p] & ctype_digit) != 0) p++;
960
961 return (*p == CHAR_RIGHT_CURLY_BRACKET);
962 }
963
964
965
966 /*************************************************
967 * Read repeat counts *
968 *************************************************/
969
970 /* Read an item of the form {n,m} and return the values. This is called only
971 after is_counted_repeat() has confirmed that a repeat-count quantifier exists,
972 so the syntax is guaranteed to be correct, but we need to check the values.
973
974 Arguments:
975 p pointer to first char after '{'
976 minp pointer to int for min
977 maxp pointer to int for max
978 returned as -1 if no max
979 errorcodeptr points to error code variable
980
981 Returns: pointer to '}' on success;
982 current ptr on error, with errorcodeptr set non-zero
983 */
984
985 static const uschar *
986 read_repeat_counts(const uschar *p, int *minp, int *maxp, int *errorcodeptr)
987 {
988 int min = 0;
989 int max = -1;
990
991 /* Read the minimum value and do a paranoid check: a negative value indicates
992 an integer overflow. */
993
994 while ((digitab[*p] & ctype_digit) != 0) min = min * 10 + *p++ - CHAR_0;
995 if (min < 0 || min > 65535)
996 {
997 *errorcodeptr = ERR5;
998 return p;
999 }
1000
1001 /* Read the maximum value if there is one, and again do a paranoid on its size.
1002 Also, max must not be less than min. */
1003
1004 if (*p == CHAR_RIGHT_CURLY_BRACKET) max = min; else
1005 {
1006 if (*(++p) != CHAR_RIGHT_CURLY_BRACKET)
1007 {
1008 max = 0;
1009 while((digitab[*p] & ctype_digit) != 0) max = max * 10 + *p++ - CHAR_0;
1010 if (max < 0 || max > 65535)
1011 {
1012 *errorcodeptr = ERR5;
1013 return p;
1014 }
1015 if (max < min)
1016 {
1017 *errorcodeptr = ERR4;
1018 return p;
1019 }
1020 }
1021 }
1022
1023 /* Fill in the required variables, and pass back the pointer to the terminating
1024 '}'. */
1025
1026 *minp = min;
1027 *maxp = max;
1028 return p;
1029 }
1030
1031
1032
1033 /*************************************************
1034 * Subroutine for finding forward reference *
1035 *************************************************/
1036
1037 /* This recursive function is called only from find_parens() below. The
1038 top-level call starts at the beginning of the pattern. All other calls must
1039 start at a parenthesis. It scans along a pattern's text looking for capturing
1040 subpatterns, and counting them. If it finds a named pattern that matches the
1041 name it is given, it returns its number. Alternatively, if the name is NULL, it
1042 returns when it reaches a given numbered subpattern. We know that if (?P< is
1043 encountered, the name will be terminated by '>' because that is checked in the
1044 first pass. Recursion is used to keep track of subpatterns that reset the
1045 capturing group numbers - the (?| feature.
1046
1047 Arguments:
1048 ptrptr address of the current character pointer (updated)
1049 cd compile background data
1050 name name to seek, or NULL if seeking a numbered subpattern
1051 lorn name length, or subpattern number if name is NULL
1052 xmode TRUE if we are in /x mode
1053 count pointer to the current capturing subpattern number (updated)
1054
1055 Returns: the number of the named subpattern, or -1 if not found
1056 */
1057
1058 static int
1059 find_parens_sub(uschar **ptrptr, compile_data *cd, const uschar *name, int lorn,
1060 BOOL xmode, int *count)
1061 {
1062 uschar *ptr = *ptrptr;
1063 int start_count = *count;
1064 int hwm_count = start_count;
1065 BOOL dup_parens = FALSE;
1066
1067 /* If the first character is a parenthesis, check on the type of group we are
1068 dealing with. The very first call may not start with a parenthesis. */
1069
1070 if (ptr[0] == CHAR_LEFT_PARENTHESIS)
1071 {
1072 if (ptr[1] == CHAR_QUESTION_MARK &&
1073 ptr[2] == CHAR_VERTICAL_LINE)
1074 {
1075 ptr += 3;
1076 dup_parens = TRUE;
1077 }
1078
1079 /* Handle a normal, unnamed capturing parenthesis */
1080
1081 else if (ptr[1] != CHAR_QUESTION_MARK && ptr[1] != CHAR_ASTERISK)
1082 {
1083 *count += 1;
1084 if (name == NULL && *count == lorn) return *count;
1085 ptr++;
1086 }
1087
1088 /* Handle a condition. If it is an assertion, just carry on so that it
1089 is processed as normal. If not, skip to the closing parenthesis of the
1090 condition (there can't be any nested parens. */
1091
1092 else if (ptr[2] == CHAR_LEFT_PARENTHESIS)
1093 {
1094 ptr += 2;
1095 if (ptr[1] != CHAR_QUESTION_MARK)
1096 {
1097 while (*ptr != 0 && *ptr != CHAR_RIGHT_PARENTHESIS) ptr++;
1098 if (*ptr != 0) ptr++;
1099 }
1100 }
1101
1102 /* We have either (? or (* and not a condition */
1103
1104 else
1105 {
1106 ptr += 2;
1107 if (*ptr == CHAR_P) ptr++; /* Allow optional P */
1108
1109 /* We have to disambiguate (?<! and (?<= from (?<name> for named groups */
1110
1111 if ((*ptr == CHAR_LESS_THAN_SIGN && ptr[1] != CHAR_EXCLAMATION_MARK &&
1112 ptr[1] != CHAR_EQUALS_SIGN) || *ptr == CHAR_APOSTROPHE)
1113 {
1114 int term;
1115 const uschar *thisname;
1116 *count += 1;
1117 if (name == NULL && *count == lorn) return *count;
1118 term = *ptr++;
1119 if (term == CHAR_LESS_THAN_SIGN) term = CHAR_GREATER_THAN_SIGN;
1120 thisname = ptr;
1121 while (*ptr != term) ptr++;
1122 if (name != NULL && lorn == ptr - thisname &&
1123 strncmp((const char *)name, (const char *)thisname, lorn) == 0)
1124 return *count;
1125 term++;
1126 }
1127 }
1128 }
1129
1130 /* Past any initial parenthesis handling, scan for parentheses or vertical
1131 bars. */
1132
1133 for (; *ptr != 0; ptr++)
1134 {
1135 /* Skip over backslashed characters and also entire \Q...\E */
1136
1137 if (*ptr == CHAR_BACKSLASH)
1138 {
1139 if (*(++ptr) == 0) goto FAIL_EXIT;
1140 if (*ptr == CHAR_Q) for (;;)
1141 {
1142 while (*(++ptr) != 0 && *ptr != CHAR_BACKSLASH) {};
1143 if (*ptr == 0) goto FAIL_EXIT;
1144 if (*(++ptr) == CHAR_E) break;
1145 }
1146 continue;
1147 }
1148
1149 /* Skip over character classes; this logic must be similar to the way they
1150 are handled for real. If the first character is '^', skip it. Also, if the
1151 first few characters (either before or after ^) are \Q\E or \E we skip them
1152 too. This makes for compatibility with Perl. Note the use of STR macros to
1153 encode "Q\\E" so that it works in UTF-8 on EBCDIC platforms. */
1154
1155 if (*ptr == CHAR_LEFT_SQUARE_BRACKET)
1156 {
1157 BOOL negate_class = FALSE;
1158 for (;;)
1159 {
1160 if (ptr[1] == CHAR_BACKSLASH)
1161 {
1162 if (ptr[2] == CHAR_E)
1163 ptr+= 2;
1164 else if (strncmp((const char *)ptr+2,
1165 STR_Q STR_BACKSLASH STR_E, 3) == 0)
1166 ptr += 4;
1167 else
1168 break;
1169 }
1170 else if (!negate_class && ptr[1] == CHAR_CIRCUMFLEX_ACCENT)
1171 {
1172 negate_class = TRUE;
1173 ptr++;
1174 }
1175 else break;
1176 }
1177
1178 /* If the next character is ']', it is a data character that must be
1179 skipped, except in JavaScript compatibility mode. */
1180
1181 if (ptr[1] == CHAR_RIGHT_SQUARE_BRACKET &&
1182 (cd->external_options & PCRE_JAVASCRIPT_COMPAT) == 0)
1183 ptr++;
1184
1185 while (*(++ptr) != CHAR_RIGHT_SQUARE_BRACKET)
1186 {
1187 if (*ptr == 0) return -1;
1188 if (*ptr == CHAR_BACKSLASH)
1189 {
1190 if (*(++ptr) == 0) goto FAIL_EXIT;
1191 if (*ptr == CHAR_Q) for (;;)
1192 {
1193 while (*(++ptr) != 0 && *ptr != CHAR_BACKSLASH) {};
1194 if (*ptr == 0) goto FAIL_EXIT;
1195 if (*(++ptr) == CHAR_E) break;
1196 }
1197 continue;
1198 }
1199 }
1200 continue;
1201 }
1202
1203 /* Skip comments in /x mode */
1204
1205 if (xmode && *ptr == CHAR_NUMBER_SIGN)
1206 {
1207 while (*(++ptr) != 0 && *ptr != CHAR_NL) {};
1208 if (*ptr == 0) goto FAIL_EXIT;
1209 continue;
1210 }
1211
1212 /* Check for the special metacharacters */
1213
1214 if (*ptr == CHAR_LEFT_PARENTHESIS)
1215 {
1216 int rc = find_parens_sub(&ptr, cd, name, lorn, xmode, count);
1217 if (rc > 0) return rc;
1218 if (*ptr == 0) goto FAIL_EXIT;
1219 }
1220
1221 else if (*ptr == CHAR_RIGHT_PARENTHESIS)
1222 {
1223 if (dup_parens && *count < hwm_count) *count = hwm_count;
1224 *ptrptr = ptr;
1225 return -1;
1226 }
1227
1228 else if (*ptr == CHAR_VERTICAL_LINE && dup_parens)
1229 {
1230 if (*count > hwm_count) hwm_count = *count;
1231 *count = start_count;
1232 }
1233 }
1234
1235 FAIL_EXIT:
1236 *ptrptr = ptr;
1237 return -1;
1238 }
1239
1240
1241
1242
1243 /*************************************************
1244 * Find forward referenced subpattern *
1245 *************************************************/
1246
1247 /* This function scans along a pattern's text looking for capturing
1248 subpatterns, and counting them. If it finds a named pattern that matches the
1249 name it is given, it returns its number. Alternatively, if the name is NULL, it
1250 returns when it reaches a given numbered subpattern. This is used for forward
1251 references to subpatterns. We used to be able to start this scan from the
1252 current compiling point, using the current count value from cd->bracount, and
1253 do it all in a single loop, but the addition of the possibility of duplicate
1254 subpattern numbers means that we have to scan from the very start, in order to
1255 take account of such duplicates, and to use a recursive function to keep track
1256 of the different types of group.
1257
1258 Arguments:
1259 cd compile background data
1260 name name to seek, or NULL if seeking a numbered subpattern
1261 lorn name length, or subpattern number if name is NULL
1262 xmode TRUE if we are in /x mode
1263
1264 Returns: the number of the found subpattern, or -1 if not found
1265 */
1266
1267 static int
1268 find_parens(compile_data *cd, const uschar *name, int lorn, BOOL xmode)
1269 {
1270 uschar *ptr = (uschar *)cd->start_pattern;
1271 int count = 0;
1272 int rc;
1273
1274 /* If the pattern does not start with an opening parenthesis, the first call
1275 to find_parens_sub() will scan right to the end (if necessary). However, if it
1276 does start with a parenthesis, find_parens_sub() will return when it hits the
1277 matching closing parens. That is why we have to have a loop. */
1278
1279 for (;;)
1280 {
1281 rc = find_parens_sub(&ptr, cd, name, lorn, xmode, &count);
1282 if (rc > 0 || *ptr++ == 0) break;
1283 }
1284
1285 return rc;
1286 }
1287
1288
1289
1290
1291 /*************************************************
1292 * Find first significant op code *
1293 *************************************************/
1294
1295 /* This is called by several functions that scan a compiled expression looking
1296 for a fixed first character, or an anchoring op code etc. It skips over things
1297 that do not influence this. For some calls, a change of option is important.
1298 For some calls, it makes sense to skip negative forward and all backward
1299 assertions, and also the \b assertion; for others it does not.
1300
1301 Arguments:
1302 code pointer to the start of the group
1303 options pointer to external options
1304 optbit the option bit whose changing is significant, or
1305 zero if none are
1306 skipassert TRUE if certain assertions are to be skipped
1307
1308 Returns: pointer to the first significant opcode
1309 */
1310
1311 static const uschar*
1312 first_significant_code(const uschar *code, int *options, int optbit,
1313 BOOL skipassert)
1314 {
1315 for (;;)
1316 {
1317 switch ((int)*code)
1318 {
1319 case OP_OPT:
1320 if (optbit > 0 && ((int)code[1] & optbit) != (*options & optbit))
1321 *options = (int)code[1];
1322 code += 2;
1323 break;
1324
1325 case OP_ASSERT_NOT:
1326 case OP_ASSERTBACK:
1327 case OP_ASSERTBACK_NOT:
1328 if (!skipassert) return code;
1329 do code += GET(code, 1); while (*code == OP_ALT);
1330 code += _pcre_OP_lengths[*code];
1331 break;
1332
1333 case OP_WORD_BOUNDARY:
1334 case OP_NOT_WORD_BOUNDARY:
1335 if (!skipassert) return code;
1336 /* Fall through */
1337
1338 case OP_CALLOUT:
1339 case OP_CREF:
1340 case OP_NCREF:
1341 case OP_RREF:
1342 case OP_NRREF:
1343 case OP_DEF:
1344 code += _pcre_OP_lengths[*code];
1345 break;
1346
1347 default:
1348 return code;
1349 }
1350 }
1351 /* Control never reaches here */
1352 }
1353
1354
1355
1356
1357 /*************************************************
1358 * Find the fixed length of a branch *
1359 *************************************************/
1360
1361 /* Scan a branch and compute the fixed length of subject that will match it,
1362 if the length is fixed. This is needed for dealing with backward assertions.
1363 In UTF8 mode, the result is in characters rather than bytes. The branch is
1364 temporarily terminated with OP_END when this function is called.
1365
1366 This function is called when a backward assertion is encountered, so that if it
1367 fails, the error message can point to the correct place in the pattern.
1368 However, we cannot do this when the assertion contains subroutine calls,
1369 because they can be forward references. We solve this by remembering this case
1370 and doing the check at the end; a flag specifies which mode we are running in.
1371
1372 Arguments:
1373 code points to the start of the pattern (the bracket)
1374 options the compiling options
1375 atend TRUE if called when the pattern is complete
1376 cd the "compile data" structure
1377
1378 Returns: the fixed length,
1379 or -1 if there is no fixed length,
1380 or -2 if \C was encountered
1381 or -3 if an OP_RECURSE item was encountered and atend is FALSE
1382 */
1383
1384 static int
1385 find_fixedlength(uschar *code, int options, BOOL atend, compile_data *cd)
1386 {
1387 int length = -1;
1388
1389 register int branchlength = 0;
1390 register uschar *cc = code + 1 + LINK_SIZE;
1391
1392 /* Scan along the opcodes for this branch. If we get to the end of the
1393 branch, check the length against that of the other branches. */
1394
1395 for (;;)
1396 {
1397 int d;
1398 uschar *ce, *cs;
1399 register int op = *cc;
1400 switch (op)
1401 {
1402 case OP_CBRA:
1403 case OP_BRA:
1404 case OP_ONCE:
1405 case OP_COND:
1406 d = find_fixedlength(cc + ((op == OP_CBRA)? 2:0), options, atend, cd);
1407 if (d < 0) return d;
1408 branchlength += d;
1409 do cc += GET(cc, 1); while (*cc == OP_ALT);
1410 cc += 1 + LINK_SIZE;
1411 break;
1412
1413 /* Reached end of a branch; if it's a ket it is the end of a nested
1414 call. If it's ALT it is an alternation in a nested call. If it is
1415 END it's the end of the outer call. All can be handled by the same code. */
1416
1417 case OP_ALT:
1418 case OP_KET:
1419 case OP_KETRMAX:
1420 case OP_KETRMIN:
1421 case OP_END:
1422 if (length < 0) length = branchlength;
1423 else if (length != branchlength) return -1;
1424 if (*cc != OP_ALT) return length;
1425 cc += 1 + LINK_SIZE;
1426 branchlength = 0;
1427 break;
1428
1429 /* A true recursion implies not fixed length, but a subroutine call may
1430 be OK. If the subroutine is a forward reference, we can't deal with
1431 it until the end of the pattern, so return -3. */
1432
1433 case OP_RECURSE:
1434 if (!atend) return -3;
1435 cs = ce = (uschar *)cd->start_code + GET(cc, 1); /* Start subpattern */
1436 do ce += GET(ce, 1); while (*ce == OP_ALT); /* End subpattern */
1437 if (cc > cs && cc < ce) return -1; /* Recursion */
1438 d = find_fixedlength(cs + 2, options, atend, cd);
1439 if (d < 0) return d;
1440 branchlength += d;
1441 cc += 1 + LINK_SIZE;
1442 break;
1443
1444 /* Skip over assertive subpatterns */
1445
1446 case OP_ASSERT:
1447 case OP_ASSERT_NOT:
1448 case OP_ASSERTBACK:
1449 case OP_ASSERTBACK_NOT:
1450 do cc += GET(cc, 1); while (*cc == OP_ALT);
1451 /* Fall through */
1452
1453 /* Skip over things that don't match chars */
1454
1455 case OP_REVERSE:
1456 case OP_CREF:
1457 case OP_NCREF:
1458 case OP_RREF:
1459 case OP_NRREF:
1460 case OP_DEF:
1461 case OP_OPT:
1462 case OP_CALLOUT:
1463 case OP_SOD:
1464 case OP_SOM:
1465 case OP_SET_SOM:
1466 case OP_EOD:
1467 case OP_EODN:
1468 case OP_CIRC:
1469 case OP_DOLL:
1470 case OP_NOT_WORD_BOUNDARY:
1471 case OP_WORD_BOUNDARY:
1472 cc += _pcre_OP_lengths[*cc];
1473 break;
1474
1475 /* Handle literal characters */
1476
1477 case OP_CHAR:
1478 case OP_CHARNC:
1479 case OP_NOT:
1480 branchlength++;
1481 cc += 2;
1482 #ifdef SUPPORT_UTF8
1483 if ((options & PCRE_UTF8) != 0 && cc[-1] >= 0xc0)
1484 cc += _pcre_utf8_table4[cc[-1] & 0x3f];
1485 #endif
1486 break;
1487
1488 /* Handle exact repetitions. The count is already in characters, but we
1489 need to skip over a multibyte character in UTF8 mode. */
1490
1491 case OP_EXACT:
1492 branchlength += GET2(cc,1);
1493 cc += 4;
1494 #ifdef SUPPORT_UTF8
1495 if ((options & PCRE_UTF8) != 0 && cc[-1] >= 0xc0)
1496 cc += _pcre_utf8_table4[cc[-1] & 0x3f];
1497 #endif
1498 break;
1499
1500 case OP_TYPEEXACT:
1501 branchlength += GET2(cc,1);
1502 if (cc[3] == OP_PROP || cc[3] == OP_NOTPROP) cc += 2;
1503 cc += 4;
1504 break;
1505
1506 /* Handle single-char matchers */
1507
1508 case OP_PROP:
1509 case OP_NOTPROP:
1510 cc += 2;
1511 /* Fall through */
1512
1513 case OP_NOT_DIGIT:
1514 case OP_DIGIT:
1515 case OP_NOT_WHITESPACE:
1516 case OP_WHITESPACE:
1517 case OP_NOT_WORDCHAR:
1518 case OP_WORDCHAR:
1519 case OP_ANY:
1520 case OP_ALLANY:
1521 branchlength++;
1522 cc++;
1523 break;
1524
1525 /* The single-byte matcher isn't allowed */
1526
1527 case OP_ANYBYTE:
1528 return -2;
1529
1530 /* Check a class for variable quantification */
1531
1532 #ifdef SUPPORT_UTF8
1533 case OP_XCLASS:
1534 cc += GET(cc, 1) - 33;
1535 /* Fall through */
1536 #endif
1537
1538 case OP_CLASS:
1539 case OP_NCLASS:
1540 cc += 33;
1541
1542 switch (*cc)
1543 {
1544 case OP_CRSTAR:
1545 case OP_CRMINSTAR:
1546 case OP_CRQUERY:
1547 case OP_CRMINQUERY:
1548 return -1;
1549
1550 case OP_CRRANGE:
1551 case OP_CRMINRANGE:
1552 if (GET2(cc,1) != GET2(cc,3)) return -1;
1553 branchlength += GET2(cc,1);
1554 cc += 5;
1555 break;
1556
1557 default:
1558 branchlength++;
1559 }
1560 break;
1561
1562 /* Anything else is variable length */
1563
1564 default:
1565 return -1;
1566 }
1567 }
1568 /* Control never gets here */
1569 }
1570
1571
1572
1573
1574 /*************************************************
1575 * Scan compiled regex for specific bracket *
1576 *************************************************/
1577
1578 /* This little function scans through a compiled pattern until it finds a
1579 capturing bracket with the given number, or, if the number is negative, an
1580 instance of OP_REVERSE for a lookbehind. The function is global in the C sense
1581 so that it can be called from pcre_study() when finding the minimum matching
1582 length.
1583
1584 Arguments:
1585 code points to start of expression
1586 utf8 TRUE in UTF-8 mode
1587 number the required bracket number or negative to find a lookbehind
1588
1589 Returns: pointer to the opcode for the bracket, or NULL if not found
1590 */
1591
1592 const uschar *
1593 _pcre_find_bracket(const uschar *code, BOOL utf8, int number)
1594 {
1595 for (;;)
1596 {
1597 register int c = *code;
1598 if (c == OP_END) return NULL;
1599
1600 /* XCLASS is used for classes that cannot be represented just by a bit
1601 map. This includes negated single high-valued characters. The length in
1602 the table is zero; the actual length is stored in the compiled code. */
1603
1604 if (c == OP_XCLASS) code += GET(code, 1);
1605
1606 /* Handle recursion */
1607
1608 else if (c == OP_REVERSE)
1609 {
1610 if (number < 0) return (uschar *)code;
1611 code += _pcre_OP_lengths[c];
1612 }
1613
1614 /* Handle capturing bracket */
1615
1616 else if (c == OP_CBRA)
1617 {
1618 int n = GET2(code, 1+LINK_SIZE);
1619 if (n == number) return (uschar *)code;
1620 code += _pcre_OP_lengths[c];
1621 }
1622
1623 /* Otherwise, we can get the item's length from the table, except that for
1624 repeated character types, we have to test for \p and \P, which have an extra
1625 two bytes of parameters, and for MARK/PRUNE/SKIP/THEN with an argument, we
1626 must add in its length. */
1627
1628 else
1629 {
1630 switch(c)
1631 {
1632 case OP_TYPESTAR:
1633 case OP_TYPEMINSTAR:
1634 case OP_TYPEPLUS:
1635 case OP_TYPEMINPLUS:
1636 case OP_TYPEQUERY:
1637 case OP_TYPEMINQUERY:
1638 case OP_TYPEPOSSTAR:
1639 case OP_TYPEPOSPLUS:
1640 case OP_TYPEPOSQUERY:
1641 if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
1642 break;
1643
1644 case OP_TYPEUPTO:
1645 case OP_TYPEMINUPTO:
1646 case OP_TYPEEXACT:
1647 case OP_TYPEPOSUPTO:
1648 if (code[3] == OP_PROP || code[3] == OP_NOTPROP) code += 2;
1649 break;
1650
1651 case OP_MARK:
1652 case OP_PRUNE_ARG:
1653 case OP_SKIP_ARG:
1654 case OP_THEN_ARG:
1655 code += code[1];
1656 break;
1657 }
1658
1659 /* Add in the fixed length from the table */
1660
1661 code += _pcre_OP_lengths[c];
1662
1663 /* In UTF-8 mode, opcodes that are followed by a character may be followed by
1664 a multi-byte character. The length in the table is a minimum, so we have to
1665 arrange to skip the extra bytes. */
1666
1667 #ifdef SUPPORT_UTF8
1668 if (utf8) switch(c)
1669 {
1670 case OP_CHAR:
1671 case OP_CHARNC:
1672 case OP_EXACT:
1673 case OP_UPTO:
1674 case OP_MINUPTO:
1675 case OP_POSUPTO:
1676 case OP_STAR:
1677 case OP_MINSTAR:
1678 case OP_POSSTAR:
1679 case OP_PLUS:
1680 case OP_MINPLUS:
1681 case OP_POSPLUS:
1682 case OP_QUERY:
1683 case OP_MINQUERY:
1684 case OP_POSQUERY:
1685 if (code[-1] >= 0xc0) code += _pcre_utf8_table4[code[-1] & 0x3f];
1686 break;
1687 }
1688 #else
1689 (void)(utf8); /* Keep compiler happy by referencing function argument */
1690 #endif
1691 }
1692 }
1693 }
1694
1695
1696
1697 /*************************************************
1698 * Scan compiled regex for recursion reference *
1699 *************************************************/
1700
1701 /* This little function scans through a compiled pattern until it finds an
1702 instance of OP_RECURSE.
1703
1704 Arguments:
1705 code points to start of expression
1706 utf8 TRUE in UTF-8 mode
1707
1708 Returns: pointer to the opcode for OP_RECURSE, or NULL if not found
1709 */
1710
1711 static const uschar *
1712 find_recurse(const uschar *code, BOOL utf8)
1713 {
1714 for (;;)
1715 {
1716 register int c = *code;
1717 if (c == OP_END) return NULL;
1718 if (c == OP_RECURSE) return code;
1719
1720 /* XCLASS is used for classes that cannot be represented just by a bit
1721 map. This includes negated single high-valued characters. The length in
1722 the table is zero; the actual length is stored in the compiled code. */
1723
1724 if (c == OP_XCLASS) code += GET(code, 1);
1725
1726 /* Otherwise, we can get the item's length from the table, except that for
1727 repeated character types, we have to test for \p and \P, which have an extra
1728 two bytes of parameters, and for MARK/PRUNE/SKIP/THEN with an argument, we
1729 must add in its length. */
1730
1731 else
1732 {
1733 switch(c)
1734 {
1735 case OP_TYPESTAR:
1736 case OP_TYPEMINSTAR:
1737 case OP_TYPEPLUS:
1738 case OP_TYPEMINPLUS:
1739 case OP_TYPEQUERY:
1740 case OP_TYPEMINQUERY:
1741 case OP_TYPEPOSSTAR:
1742 case OP_TYPEPOSPLUS:
1743 case OP_TYPEPOSQUERY:
1744 if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
1745 break;
1746
1747 case OP_TYPEPOSUPTO:
1748 case OP_TYPEUPTO:
1749 case OP_TYPEMINUPTO:
1750 case OP_TYPEEXACT:
1751 if (code[3] == OP_PROP || code[3] == OP_NOTPROP) code += 2;
1752 break;
1753
1754 case OP_MARK:
1755 case OP_PRUNE_ARG:
1756 case OP_SKIP_ARG:
1757 case OP_THEN_ARG:
1758 code += code[1];
1759 break;
1760 }
1761
1762 /* Add in the fixed length from the table */
1763
1764 code += _pcre_OP_lengths[c];
1765
1766 /* In UTF-8 mode, opcodes that are followed by a character may be followed
1767 by a multi-byte character. The length in the table is a minimum, so we have
1768 to arrange to skip the extra bytes. */
1769
1770 #ifdef SUPPORT_UTF8
1771 if (utf8) switch(c)
1772 {
1773 case OP_CHAR:
1774 case OP_CHARNC:
1775 case OP_EXACT:
1776 case OP_UPTO:
1777 case OP_MINUPTO:
1778 case OP_POSUPTO:
1779 case OP_STAR:
1780 case OP_MINSTAR:
1781 case OP_POSSTAR:
1782 case OP_PLUS:
1783 case OP_MINPLUS:
1784 case OP_POSPLUS:
1785 case OP_QUERY:
1786 case OP_MINQUERY:
1787 case OP_POSQUERY:
1788 if (code[-1] >= 0xc0) code += _pcre_utf8_table4[code[-1] & 0x3f];
1789 break;
1790 }
1791 #else
1792 (void)(utf8); /* Keep compiler happy by referencing function argument */
1793 #endif
1794 }
1795 }
1796 }
1797
1798
1799
1800 /*************************************************
1801 * Scan compiled branch for non-emptiness *
1802 *************************************************/
1803
1804 /* This function scans through a branch of a compiled pattern to see whether it
1805 can match the empty string or not. It is called from could_be_empty()
1806 below and from compile_branch() when checking for an unlimited repeat of a
1807 group that can match nothing. Note that first_significant_code() skips over
1808 backward and negative forward assertions when its final argument is TRUE. If we
1809 hit an unclosed bracket, we return "empty" - this means we've struck an inner
1810 bracket whose current branch will already have been scanned.
1811
1812 Arguments:
1813 code points to start of search
1814 endcode points to where to stop
1815 utf8 TRUE if in UTF8 mode
1816 cd contains pointers to tables etc.
1817
1818 Returns: TRUE if what is matched could be empty
1819 */
1820
1821 static BOOL
1822 could_be_empty_branch(const uschar *code, const uschar *endcode, BOOL utf8,
1823 compile_data *cd)
1824 {
1825 register int c;
1826 for (code = first_significant_code(code + _pcre_OP_lengths[*code], NULL, 0, TRUE);
1827 code < endcode;
1828 code = first_significant_code(code + _pcre_OP_lengths[c], NULL, 0, TRUE))
1829 {
1830 const uschar *ccode;
1831
1832 c = *code;
1833
1834 /* Skip over forward assertions; the other assertions are skipped by
1835 first_significant_code() with a TRUE final argument. */
1836
1837 if (c == OP_ASSERT)
1838 {
1839 do code += GET(code, 1); while (*code == OP_ALT);
1840 c = *code;
1841 continue;
1842 }
1843
1844 /* Groups with zero repeats can of course be empty; skip them. */
1845
1846 if (c == OP_BRAZERO || c == OP_BRAMINZERO || c == OP_SKIPZERO)
1847 {
1848 code += _pcre_OP_lengths[c];
1849 do code += GET(code, 1); while (*code == OP_ALT);
1850 c = *code;
1851 continue;
1852 }
1853
1854 /* For a recursion/subroutine call, if its end has been reached, which
1855 implies a subroutine call, we can scan it. */
1856
1857 if (c == OP_RECURSE)
1858 {
1859 BOOL empty_branch = FALSE;
1860 const uschar *scode = cd->start_code + GET(code, 1);
1861 if (GET(scode, 1) == 0) return TRUE; /* Unclosed */
1862 do
1863 {
1864 if (could_be_empty_branch(scode, endcode, utf8, cd))
1865 {
1866 empty_branch = TRUE;
1867 break;
1868 }
1869 scode += GET(scode, 1);
1870 }
1871 while (*scode == OP_ALT);
1872 if (!empty_branch) return FALSE; /* All branches are non-empty */
1873 continue;
1874 }
1875
1876 /* For other groups, scan the branches. */
1877
1878 if (c == OP_BRA || c == OP_CBRA || c == OP_ONCE || c == OP_COND)
1879 {
1880 BOOL empty_branch;
1881 if (GET(code, 1) == 0) return TRUE; /* Hit unclosed bracket */
1882
1883 /* If a conditional group has only one branch, there is a second, implied,
1884 empty branch, so just skip over the conditional, because it could be empty.
1885 Otherwise, scan the individual branches of the group. */
1886
1887 if (c == OP_COND && code[GET(code, 1)] != OP_ALT)
1888 code += GET(code, 1);
1889 else
1890 {
1891 empty_branch = FALSE;
1892 do
1893 {
1894 if (!empty_branch && could_be_empty_branch(code, endcode, utf8, cd))
1895 empty_branch = TRUE;
1896 code += GET(code, 1);
1897 }
1898 while (*code == OP_ALT);
1899 if (!empty_branch) return FALSE; /* All branches are non-empty */
1900 }
1901
1902 c = *code;
1903 continue;
1904 }
1905
1906 /* Handle the other opcodes */
1907
1908 switch (c)
1909 {
1910 /* Check for quantifiers after a class. XCLASS is used for classes that
1911 cannot be represented just by a bit map. This includes negated single
1912 high-valued characters. The length in _pcre_OP_lengths[] is zero; the
1913 actual length is stored in the compiled code, so we must update "code"
1914 here. */
1915
1916 #ifdef SUPPORT_UTF8
1917 case OP_XCLASS:
1918 ccode = code += GET(code, 1);
1919 goto CHECK_CLASS_REPEAT;
1920 #endif
1921
1922 case OP_CLASS:
1923 case OP_NCLASS:
1924 ccode = code + 33;
1925
1926 #ifdef SUPPORT_UTF8
1927 CHECK_CLASS_REPEAT:
1928 #endif
1929
1930 switch (*ccode)
1931 {
1932 case OP_CRSTAR: /* These could be empty; continue */
1933 case OP_CRMINSTAR:
1934 case OP_CRQUERY:
1935 case OP_CRMINQUERY:
1936 break;
1937
1938 default: /* Non-repeat => class must match */
1939 case OP_CRPLUS: /* These repeats aren't empty */
1940 case OP_CRMINPLUS:
1941 return FALSE;
1942
1943 case OP_CRRANGE:
1944 case OP_CRMINRANGE:
1945 if (GET2(ccode, 1) > 0) return FALSE; /* Minimum > 0 */
1946 break;
1947 }
1948 break;
1949
1950 /* Opcodes that must match a character */
1951
1952 case OP_PROP:
1953 case OP_NOTPROP:
1954 case OP_EXTUNI:
1955 case OP_NOT_DIGIT:
1956 case OP_DIGIT:
1957 case OP_NOT_WHITESPACE:
1958 case OP_WHITESPACE:
1959 case OP_NOT_WORDCHAR:
1960 case OP_WORDCHAR:
1961 case OP_ANY:
1962 case OP_ALLANY:
1963 case OP_ANYBYTE:
1964 case OP_CHAR:
1965 case OP_CHARNC:
1966 case OP_NOT:
1967 case OP_PLUS:
1968 case OP_MINPLUS:
1969 case OP_POSPLUS:
1970 case OP_EXACT:
1971 case OP_NOTPLUS:
1972 case OP_NOTMINPLUS:
1973 case OP_NOTPOSPLUS:
1974 case OP_NOTEXACT:
1975 case OP_TYPEPLUS:
1976 case OP_TYPEMINPLUS:
1977 case OP_TYPEPOSPLUS:
1978 case OP_TYPEEXACT:
1979 return FALSE;
1980
1981 /* These are going to continue, as they may be empty, but we have to
1982 fudge the length for the \p and \P cases. */
1983
1984 case OP_TYPESTAR:
1985 case OP_TYPEMINSTAR:
1986 case OP_TYPEPOSSTAR:
1987 case OP_TYPEQUERY:
1988 case OP_TYPEMINQUERY:
1989 case OP_TYPEPOSQUERY:
1990 if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
1991 break;
1992
1993 /* Same for these */
1994
1995 case OP_TYPEUPTO:
1996 case OP_TYPEMINUPTO:
1997 case OP_TYPEPOSUPTO:
1998 if (code[3] == OP_PROP || code[3] == OP_NOTPROP) code += 2;
1999 break;
2000
2001 /* End of branch */
2002
2003 case OP_KET:
2004 case OP_KETRMAX:
2005 case OP_KETRMIN:
2006 case OP_ALT:
2007 return TRUE;
2008
2009 /* In UTF-8 mode, STAR, MINSTAR, POSSTAR, QUERY, MINQUERY, POSQUERY, UPTO,
2010 MINUPTO, and POSUPTO may be followed by a multibyte character */
2011
2012 #ifdef SUPPORT_UTF8
2013 case OP_STAR:
2014 case OP_MINSTAR:
2015 case OP_POSSTAR:
2016 case OP_QUERY:
2017 case OP_MINQUERY:
2018 case OP_POSQUERY:
2019 if (utf8 && code[1] >= 0xc0) code += _pcre_utf8_table4[code[1] & 0x3f];
2020 break;
2021
2022 case OP_UPTO:
2023 case OP_MINUPTO:
2024 case OP_POSUPTO:
2025 if (utf8 && code[3] >= 0xc0) code += _pcre_utf8_table4[code[3] & 0x3f];
2026 break;
2027 #endif
2028
2029 /* MARK, and PRUNE/SKIP/THEN with an argument must skip over the argument
2030 string. */
2031
2032 case OP_MARK:
2033 case OP_PRUNE_ARG:
2034 case OP_SKIP_ARG:
2035 case OP_THEN_ARG:
2036 code += code[1];
2037 break;
2038
2039 /* None of the remaining opcodes are required to match a character. */
2040
2041 default:
2042 break;
2043 }
2044 }
2045
2046 return TRUE;
2047 }
2048
2049
2050
2051 /*************************************************
2052 * Scan compiled regex for non-emptiness *
2053 *************************************************/
2054
2055 /* This function is called to check for left recursive calls. We want to check
2056 the current branch of the current pattern to see if it could match the empty
2057 string. If it could, we must look outwards for branches at other levels,
2058 stopping when we pass beyond the bracket which is the subject of the recursion.
2059
2060 Arguments:
2061 code points to start of the recursion
2062 endcode points to where to stop (current RECURSE item)
2063 bcptr points to the chain of current (unclosed) branch starts
2064 utf8 TRUE if in UTF-8 mode
2065 cd pointers to tables etc
2066
2067 Returns: TRUE if what is matched could be empty
2068 */
2069
2070 static BOOL
2071 could_be_empty(const uschar *code, const uschar *endcode, branch_chain *bcptr,
2072 BOOL utf8, compile_data *cd)
2073 {
2074 while (bcptr != NULL && bcptr->current_branch >= code)
2075 {
2076 if (!could_be_empty_branch(bcptr->current_branch, endcode, utf8, cd))
2077 return FALSE;
2078 bcptr = bcptr->outer;
2079 }
2080 return TRUE;
2081 }
2082
2083
2084
2085 /*************************************************
2086 * Check for POSIX class syntax *
2087 *************************************************/
2088
2089 /* This function is called when the sequence "[:" or "[." or "[=" is
2090 encountered in a character class. It checks whether this is followed by a
2091 sequence of characters terminated by a matching ":]" or ".]" or "=]". If we
2092 reach an unescaped ']' without the special preceding character, return FALSE.
2093
2094 Originally, this function only recognized a sequence of letters between the
2095 terminators, but it seems that Perl recognizes any sequence of characters,
2096 though of course unknown POSIX names are subsequently rejected. Perl gives an
2097 "Unknown POSIX class" error for [:f\oo:] for example, where previously PCRE
2098 didn't consider this to be a POSIX class. Likewise for [:1234:].
2099
2100 The problem in trying to be exactly like Perl is in the handling of escapes. We
2101 have to be sure that [abc[:x\]pqr] is *not* treated as containing a POSIX
2102 class, but [abc[:x\]pqr:]] is (so that an error can be generated). The code
2103 below handles the special case of \], but does not try to do any other escape
2104 processing. This makes it different from Perl for cases such as [:l\ower:]
2105 where Perl recognizes it as the POSIX class "lower" but PCRE does not recognize
2106 "l\ower". This is a lesser evil that not diagnosing bad classes when Perl does,
2107 I think.
2108
2109 Arguments:
2110 ptr pointer to the initial [
2111 endptr where to return the end pointer
2112
2113 Returns: TRUE or FALSE
2114 */
2115
2116 static BOOL
2117 check_posix_syntax(const uschar *ptr, const uschar **endptr)
2118 {
2119 int terminator; /* Don't combine these lines; the Solaris cc */
2120 terminator = *(++ptr); /* compiler warns about "non-constant" initializer. */
2121 for (++ptr; *ptr != 0; ptr++)
2122 {
2123 if (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET) ptr++; else
2124 {
2125 if (*ptr == CHAR_RIGHT_SQUARE_BRACKET) return FALSE;
2126 if (*ptr == terminator && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET)
2127 {
2128 *endptr = ptr;
2129 return TRUE;
2130 }
2131 }
2132 }
2133 return FALSE;
2134 }
2135
2136
2137
2138
2139 /*************************************************
2140 * Check POSIX class name *
2141 *************************************************/
2142
2143 /* This function is called to check the name given in a POSIX-style class entry
2144 such as [:alnum:].
2145
2146 Arguments:
2147 ptr points to the first letter
2148 len the length of the name
2149
2150 Returns: a value representing the name, or -1 if unknown
2151 */
2152
2153 static int
2154 check_posix_name(const uschar *ptr, int len)
2155 {
2156 const char *pn = posix_names;
2157 register int yield = 0;
2158 while (posix_name_lengths[yield] != 0)
2159 {
2160 if (len == posix_name_lengths[yield] &&
2161 strncmp((const char *)ptr, pn, len) == 0) return yield;
2162 pn += posix_name_lengths[yield] + 1;
2163 yield++;
2164 }
2165 return -1;
2166 }
2167
2168
2169 /*************************************************
2170 * Adjust OP_RECURSE items in repeated group *
2171 *************************************************/
2172
2173 /* OP_RECURSE items contain an offset from the start of the regex to the group
2174 that is referenced. This means that groups can be replicated for fixed
2175 repetition simply by copying (because the recursion is allowed to refer to
2176 earlier groups that are outside the current group). However, when a group is
2177 optional (i.e. the minimum quantifier is zero), OP_BRAZERO or OP_SKIPZERO is
2178 inserted before it, after it has been compiled. This means that any OP_RECURSE
2179 items within it that refer to the group itself or any contained groups have to
2180 have their offsets adjusted. That one of the jobs of this function. Before it
2181 is called, the partially compiled regex must be temporarily terminated with
2182 OP_END.
2183
2184 This function has been extended with the possibility of forward references for
2185 recursions and subroutine calls. It must also check the list of such references
2186 for the group we are dealing with. If it finds that one of the recursions in
2187 the current group is on this list, it adjusts the offset in the list, not the
2188 value in the reference (which is a group number).
2189
2190 Arguments:
2191 group points to the start of the group
2192 adjust the amount by which the group is to be moved
2193 utf8 TRUE in UTF-8 mode
2194 cd contains pointers to tables etc.
2195 save_hwm the hwm forward reference pointer at the start of the group
2196
2197 Returns: nothing
2198 */
2199
2200 static void
2201 adjust_recurse(uschar *group, int adjust, BOOL utf8, compile_data *cd,
2202 uschar *save_hwm)
2203 {
2204 uschar *ptr = group;
2205
2206 while ((ptr = (uschar *)find_recurse(ptr, utf8)) != NULL)
2207 {
2208 int offset;
2209 uschar *hc;
2210
2211 /* See if this recursion is on the forward reference list. If so, adjust the
2212 reference. */
2213
2214 for (hc = save_hwm; hc < cd->hwm; hc += LINK_SIZE)
2215 {
2216 offset = GET(hc, 0);
2217 if (cd->start_code + offset == ptr + 1)
2218 {
2219 PUT(hc, 0, offset + adjust);
2220 break;
2221 }
2222 }
2223
2224 /* Otherwise, adjust the recursion offset if it's after the start of this
2225 group. */
2226
2227 if (hc >= cd->hwm)
2228 {
2229 offset = GET(ptr, 1);
2230 if (cd->start_code + offset >= group) PUT(ptr, 1, offset + adjust);
2231 }
2232
2233 ptr += 1 + LINK_SIZE;
2234 }
2235 }
2236
2237
2238
2239 /*************************************************
2240 * Insert an automatic callout point *
2241 *************************************************/
2242
2243 /* This function is called when the PCRE_AUTO_CALLOUT option is set, to insert
2244 callout points before each pattern item.
2245
2246 Arguments:
2247 code current code pointer
2248 ptr current pattern pointer
2249 cd pointers to tables etc
2250
2251 Returns: new code pointer
2252 */
2253
2254 static uschar *
2255 auto_callout(uschar *code, const uschar *ptr, compile_data *cd)
2256 {
2257 *code++ = OP_CALLOUT;
2258 *code++ = 255;
2259 PUT(code, 0, ptr - cd->start_pattern); /* Pattern offset */
2260 PUT(code, LINK_SIZE, 0); /* Default length */
2261 return code + 2*LINK_SIZE;
2262 }
2263
2264
2265
2266 /*************************************************
2267 * Complete a callout item *
2268 *************************************************/
2269
2270 /* A callout item contains the length of the next item in the pattern, which
2271 we can't fill in till after we have reached the relevant point. This is used
2272 for both automatic and manual callouts.
2273
2274 Arguments:
2275 previous_callout points to previous callout item
2276 ptr current pattern pointer
2277 cd pointers to tables etc
2278
2279 Returns: nothing
2280 */
2281
2282 static void
2283 complete_callout(uschar *previous_callout, const uschar *ptr, compile_data *cd)
2284 {
2285 int length = ptr - cd->start_pattern - GET(previous_callout, 2);
2286 PUT(previous_callout, 2 + LINK_SIZE, length);
2287 }
2288
2289
2290
2291 #ifdef SUPPORT_UCP
2292 /*************************************************
2293 * Get othercase range *
2294 *************************************************/
2295
2296 /* This function is passed the start and end of a class range, in UTF-8 mode
2297 with UCP support. It searches up the characters, looking for internal ranges of
2298 characters in the "other" case. Each call returns the next one, updating the
2299 start address.
2300
2301 Arguments:
2302 cptr points to starting character value; updated
2303 d end value
2304 ocptr where to put start of othercase range
2305 odptr where to put end of othercase range
2306
2307 Yield: TRUE when range returned; FALSE when no more
2308 */
2309
2310 static BOOL
2311 get_othercase_range(unsigned int *cptr, unsigned int d, unsigned int *ocptr,
2312 unsigned int *odptr)
2313 {
2314 unsigned int c, othercase, next;
2315
2316 for (c = *cptr; c <= d; c++)
2317 { if ((othercase = UCD_OTHERCASE(c)) != c) break; }
2318
2319 if (c > d) return FALSE;
2320
2321 *ocptr = othercase;
2322 next = othercase + 1;
2323
2324 for (++c; c <= d; c++)
2325 {
2326 if (UCD_OTHERCASE(c) != next) break;
2327 next++;
2328 }
2329
2330 *odptr = next - 1;
2331 *cptr = c;
2332
2333 return TRUE;
2334 }
2335 #endif /* SUPPORT_UCP */
2336
2337
2338
2339 /*************************************************
2340 * Check if auto-possessifying is possible *
2341 *************************************************/
2342
2343 /* This function is called for unlimited repeats of certain items, to see
2344 whether the next thing could possibly match the repeated item. If not, it makes
2345 sense to automatically possessify the repeated item.
2346
2347 Arguments:
2348 op_code the repeated op code
2349 this data for this item, depends on the opcode
2350 utf8 TRUE in UTF-8 mode
2351 utf8_char used for utf8 character bytes, NULL if not relevant
2352 ptr next character in pattern
2353 options options bits
2354 cd contains pointers to tables etc.
2355
2356 Returns: TRUE if possessifying is wanted
2357 */
2358
2359 static BOOL
2360 check_auto_possessive(int op_code, int item, BOOL utf8, uschar *utf8_char,
2361 const uschar *ptr, int options, compile_data *cd)
2362 {
2363 int next;
2364
2365 /* Skip whitespace and comments in extended mode */
2366
2367 if ((options & PCRE_EXTENDED) != 0)
2368 {
2369 for (;;)
2370 {
2371 while ((cd->ctypes[*ptr] & ctype_space) != 0) ptr++;
2372 if (*ptr == CHAR_NUMBER_SIGN)
2373 {
2374 while (*(++ptr) != 0)
2375 if (IS_NEWLINE(ptr)) { ptr += cd->nllen; break; }
2376 }
2377 else break;
2378 }
2379 }
2380
2381 /* If the next item is one that we can handle, get its value. A non-negative
2382 value is a character, a negative value is an escape value. */
2383
2384 if (*ptr == CHAR_BACKSLASH)
2385 {
2386 int temperrorcode = 0;
2387 next = check_escape(&ptr, &temperrorcode, cd->bracount, options, FALSE);
2388 if (temperrorcode != 0) return FALSE;
2389 ptr++; /* Point after the escape sequence */
2390 }
2391
2392 else if ((cd->ctypes[*ptr] & ctype_meta) == 0)
2393 {
2394 #ifdef SUPPORT_UTF8
2395 if (utf8) { GETCHARINC(next, ptr); } else
2396 #endif
2397 next = *ptr++;
2398 }
2399
2400 else return FALSE;
2401
2402 /* Skip whitespace and comments in extended mode */
2403
2404 if ((options & PCRE_EXTENDED) != 0)
2405 {
2406 for (;;)
2407 {
2408 while ((cd->ctypes[*ptr] & ctype_space) != 0) ptr++;
2409 if (*ptr == CHAR_NUMBER_SIGN)
2410 {
2411 while (*(++ptr) != 0)
2412 if (IS_NEWLINE(ptr)) { ptr += cd->nllen; break; }
2413 }
2414 else break;
2415 }
2416 }
2417
2418 /* If the next thing is itself optional, we have to give up. */
2419
2420 if (*ptr == CHAR_ASTERISK || *ptr == CHAR_QUESTION_MARK ||
2421 strncmp((char *)ptr, STR_LEFT_CURLY_BRACKET STR_0 STR_COMMA, 3) == 0)
2422 return FALSE;
2423
2424 /* Now compare the next item with the previous opcode. If the previous is a
2425 positive single character match, "item" either contains the character or, if
2426 "item" is greater than 127 in utf8 mode, the character's bytes are in
2427 utf8_char. */
2428
2429
2430 /* Handle cases when the next item is a character. */
2431
2432 if (next >= 0) switch(op_code)
2433 {
2434 case OP_CHAR:
2435 #ifdef SUPPORT_UTF8
2436 if (utf8 && item > 127) { GETCHAR(item, utf8_char); }
2437 #else
2438 (void)(utf8_char); /* Keep compiler happy by referencing function argument */
2439 #endif
2440 return item != next;
2441
2442 /* For CHARNC (caseless character) we must check the other case. If we have
2443 Unicode property support, we can use it to test the other case of
2444 high-valued characters. */
2445
2446 case OP_CHARNC:
2447 #ifdef SUPPORT_UTF8
2448 if (utf8 && item > 127) { GETCHAR(item, utf8_char); }
2449 #endif
2450 if (item == next) return FALSE;
2451 #ifdef SUPPORT_UTF8
2452 if (utf8)
2453 {
2454 unsigned int othercase;
2455 if (next < 128) othercase = cd->fcc[next]; else
2456 #ifdef SUPPORT_UCP
2457 othercase = UCD_OTHERCASE((unsigned int)next);
2458 #else
2459 othercase = NOTACHAR;
2460 #endif
2461 return (unsigned int)item != othercase;
2462 }
2463 else
2464 #endif /* SUPPORT_UTF8 */
2465 return (item != cd->fcc[next]); /* Non-UTF-8 mode */
2466
2467 /* For OP_NOT, "item" must be a single-byte character. */
2468
2469 case OP_NOT:
2470 if (item == next) return TRUE;
2471 if ((options & PCRE_CASELESS) == 0) return FALSE;
2472 #ifdef SUPPORT_UTF8
2473 if (utf8)
2474 {
2475 unsigned int othercase;
2476 if (next < 128) othercase = cd->fcc[next]; else
2477 #ifdef SUPPORT_UCP
2478 othercase = UCD_OTHERCASE(next);
2479 #else
2480 othercase = NOTACHAR;
2481 #endif
2482 return (unsigned int)item == othercase;
2483 }
2484 else
2485 #endif /* SUPPORT_UTF8 */
2486 return (item == cd->fcc[next]); /* Non-UTF-8 mode */
2487
2488 case OP_DIGIT:
2489 return next > 127 || (cd->ctypes[next] & ctype_digit) == 0;
2490
2491 case OP_NOT_DIGIT:
2492 return next <= 127 && (cd->ctypes[next] & ctype_digit) != 0;
2493
2494 case OP_WHITESPACE:
2495 return next > 127 || (cd->ctypes[next] & ctype_space) == 0;
2496
2497 case OP_NOT_WHITESPACE:
2498 return next <= 127 && (cd->ctypes[next] & ctype_space) != 0;
2499
2500 case OP_WORDCHAR:
2501 return next > 127 || (cd->ctypes[next] & ctype_word) == 0;
2502
2503 case OP_NOT_WORDCHAR:
2504 return next <= 127 && (cd->ctypes[next] & ctype_word) != 0;
2505
2506 case OP_HSPACE:
2507 case OP_NOT_HSPACE:
2508 switch(next)
2509 {
2510 case 0x09:
2511 case 0x20:
2512 case 0xa0:
2513 case 0x1680:
2514 case 0x180e:
2515 case 0x2000:
2516 case 0x2001:
2517 case 0x2002:
2518 case 0x2003:
2519 case 0x2004:
2520 case 0x2005:
2521 case 0x2006:
2522 case 0x2007:
2523 case 0x2008:
2524 case 0x2009:
2525 case 0x200A:
2526 case 0x202f:
2527 case 0x205f:
2528 case 0x3000:
2529 return op_code != OP_HSPACE;
2530 default:
2531 return op_code == OP_HSPACE;
2532 }
2533
2534 case OP_VSPACE:
2535 case OP_NOT_VSPACE:
2536 switch(next)
2537 {
2538 case 0x0a:
2539 case 0x0b:
2540 case 0x0c:
2541 case 0x0d:
2542 case 0x85:
2543 case 0x2028:
2544 case 0x2029:
2545 return op_code != OP_VSPACE;
2546 default:
2547 return op_code == OP_VSPACE;
2548 }
2549
2550 default:
2551 return FALSE;
2552 }
2553
2554
2555 /* Handle the case when the next item is \d, \s, etc. */
2556
2557 switch(op_code)
2558 {
2559 case OP_CHAR:
2560 case OP_CHARNC:
2561 #ifdef SUPPORT_UTF8
2562 if (utf8 && item > 127) { GETCHAR(item, utf8_char); }
2563 #endif
2564 switch(-next)
2565 {
2566 case ESC_d:
2567 return item > 127 || (cd->ctypes[item] & ctype_digit) == 0;
2568
2569 case ESC_D:
2570 return item <= 127 && (cd->ctypes[item] & ctype_digit) != 0;
2571
2572 case ESC_s:
2573 return item > 127 || (cd->ctypes[item] & ctype_space) == 0;
2574
2575 case ESC_S:
2576 return item <= 127 && (cd->ctypes[item] & ctype_space) != 0;
2577
2578 case ESC_w:
2579 return item > 127 || (cd->ctypes[item] & ctype_word) == 0;
2580
2581 case ESC_W:
2582 return item <= 127 && (cd->ctypes[item] & ctype_word) != 0;
2583
2584 case ESC_h:
2585 case ESC_H:
2586 switch(item)
2587 {
2588 case 0x09:
2589 case 0x20:
2590 case 0xa0:
2591 case 0x1680:
2592 case 0x180e:
2593 case 0x2000:
2594 case 0x2001:
2595 case 0x2002:
2596 case 0x2003:
2597 case 0x2004:
2598 case 0x2005:
2599 case 0x2006:
2600 case 0x2007:
2601 case 0x2008:
2602 case 0x2009:
2603 case 0x200A:
2604 case 0x202f:
2605 case 0x205f:
2606 case 0x3000:
2607 return -next != ESC_h;
2608 default:
2609 return -next == ESC_h;
2610 }
2611
2612 case ESC_v:
2613 case ESC_V:
2614 switch(item)
2615 {
2616 case 0x0a:
2617 case 0x0b:
2618 case 0x0c:
2619 case 0x0d:
2620 case 0x85:
2621 case 0x2028:
2622 case 0x2029:
2623 return -next != ESC_v;
2624 default:
2625 return -next == ESC_v;
2626 }
2627
2628 default:
2629 return FALSE;
2630 }
2631
2632 case OP_DIGIT:
2633 return next == -ESC_D || next == -ESC_s || next == -ESC_W ||
2634 next == -ESC_h || next == -ESC_v;
2635
2636 case OP_NOT_DIGIT:
2637 return next == -ESC_d;
2638
2639 case OP_WHITESPACE:
2640 return next == -ESC_S || next == -ESC_d || next == -ESC_w;
2641
2642 case OP_NOT_WHITESPACE:
2643 return next == -ESC_s || next == -ESC_h || next == -ESC_v;
2644
2645 case OP_HSPACE:
2646 return next == -ESC_S || next == -ESC_H || next == -ESC_d || next == -ESC_w;
2647
2648 case OP_NOT_HSPACE:
2649 return next == -ESC_h;
2650
2651 /* Can't have \S in here because VT matches \S (Perl anomaly) */
2652 case OP_VSPACE:
2653 return next == -ESC_V || next == -ESC_d || next == -ESC_w;
2654
2655 case OP_NOT_VSPACE:
2656 return next == -ESC_v;
2657
2658 case OP_WORDCHAR:
2659 return next == -ESC_W || next == -ESC_s || next == -ESC_h || next == -ESC_v;
2660
2661 case OP_NOT_WORDCHAR:
2662 return next == -ESC_w || next == -ESC_d;
2663
2664 default:
2665 return FALSE;
2666 }
2667
2668 /* Control does not reach here */
2669 }
2670
2671
2672
2673 /*************************************************
2674 * Compile one branch *
2675 *************************************************/
2676
2677 /* Scan the pattern, compiling it into the a vector. If the options are
2678 changed during the branch, the pointer is used to change the external options
2679 bits. This function is used during the pre-compile phase when we are trying
2680 to find out the amount of memory needed, as well as during the real compile
2681 phase. The value of lengthptr distinguishes the two phases.
2682
2683 Arguments:
2684 optionsptr pointer to the option bits
2685 codeptr points to the pointer to the current code point
2686 ptrptr points to the current pattern pointer
2687 errorcodeptr points to error code variable
2688 firstbyteptr set to initial literal character, or < 0 (REQ_UNSET, REQ_NONE)
2689 reqbyteptr set to the last literal character required, else < 0
2690 bcptr points to current branch chain
2691 cd contains pointers to tables etc.
2692 lengthptr NULL during the real compile phase
2693 points to length accumulator during pre-compile phase
2694
2695 Returns: TRUE on success
2696 FALSE, with *errorcodeptr set non-zero on error
2697 */
2698
2699 static BOOL
2700 compile_branch(int *optionsptr, uschar **codeptr, const uschar **ptrptr,
2701 int *errorcodeptr, int *firstbyteptr, int *reqbyteptr, branch_chain *bcptr,
2702 compile_data *cd, int *lengthptr)
2703 {
2704 int repeat_type, op_type;
2705 int repeat_min = 0, repeat_max = 0; /* To please picky compilers */
2706 int bravalue = 0;
2707 int greedy_default, greedy_non_default;
2708 int firstbyte, reqbyte;
2709 int zeroreqbyte, zerofirstbyte;
2710 int req_caseopt, reqvary, tempreqvary;
2711 int options = *optionsptr;
2712 int after_manual_callout = 0;
2713 int length_prevgroup = 0;
2714 register int c;
2715 register uschar *code = *codeptr;
2716 uschar *last_code = code;
2717 uschar *orig_code = code;
2718 uschar *tempcode;
2719 BOOL inescq = FALSE;
2720 BOOL groupsetfirstbyte = FALSE;
2721 const uschar *ptr = *ptrptr;
2722 const uschar *tempptr;
2723 uschar *previous = NULL;
2724 uschar *previous_callout = NULL;
2725 uschar *save_hwm = NULL;
2726 uschar classbits[32];
2727
2728 #ifdef SUPPORT_UTF8
2729 BOOL class_utf8;
2730 BOOL utf8 = (options & PCRE_UTF8) != 0;
2731 uschar *class_utf8data;
2732 uschar *class_utf8data_base;
2733 uschar utf8_char[6];
2734 #else
2735 BOOL utf8 = FALSE;
2736 uschar *utf8_char = NULL;
2737 #endif
2738
2739 #ifdef PCRE_DEBUG
2740 if (lengthptr != NULL) DPRINTF((">> start branch\n"));
2741 #endif
2742
2743 /* Set up the default and non-default settings for greediness */
2744
2745 greedy_default = ((options & PCRE_UNGREEDY) != 0);
2746 greedy_non_default = greedy_default ^ 1;
2747
2748 /* Initialize no first byte, no required byte. REQ_UNSET means "no char
2749 matching encountered yet". It gets changed to REQ_NONE if we hit something that
2750 matches a non-fixed char first char; reqbyte just remains unset if we never
2751 find one.
2752
2753 When we hit a repeat whose minimum is zero, we may have to adjust these values
2754 to take the zero repeat into account. This is implemented by setting them to
2755 zerofirstbyte and zeroreqbyte when such a repeat is encountered. The individual
2756 item types that can be repeated set these backoff variables appropriately. */
2757
2758 firstbyte = reqbyte = zerofirstbyte = zeroreqbyte = REQ_UNSET;
2759
2760 /* The variable req_caseopt contains either the REQ_CASELESS value or zero,
2761 according to the current setting of the caseless flag. REQ_CASELESS is a bit
2762 value > 255. It is added into the firstbyte or reqbyte variables to record the
2763 case status of the value. This is used only for ASCII characters. */
2764
2765 req_caseopt = ((options & PCRE_CASELESS) != 0)? REQ_CASELESS : 0;
2766
2767 /* Switch on next character until the end of the branch */
2768
2769 for (;; ptr++)
2770 {
2771 BOOL negate_class;
2772 BOOL should_flip_negation;
2773 BOOL possessive_quantifier;
2774 BOOL is_quantifier;
2775 BOOL is_recurse;
2776 BOOL reset_bracount;
2777 int class_charcount;
2778 int class_lastchar;
2779 int newoptions;
2780 int recno;
2781 int refsign;
2782 int skipbytes;
2783 int subreqbyte;
2784 int subfirstbyte;
2785 int terminator;
2786 int mclength;
2787 uschar mcbuffer[8];
2788
2789 /* Get next byte in the pattern */
2790
2791 c = *ptr;
2792
2793 /* If we are in the pre-compile phase, accumulate the length used for the
2794 previous cycle of this loop. */
2795
2796 if (lengthptr != NULL)
2797 {
2798 #ifdef PCRE_DEBUG
2799 if (code > cd->hwm) cd->hwm = code; /* High water info */
2800 #endif
2801 if (code > cd->start_workspace + WORK_SIZE_CHECK) /* Check for overrun */
2802 {
2803 *errorcodeptr = ERR52;
2804 goto FAILED;
2805 }
2806
2807 /* There is at least one situation where code goes backwards: this is the
2808 case of a zero quantifier after a class (e.g. [ab]{0}). At compile time,
2809 the class is simply eliminated. However, it is created first, so we have to
2810 allow memory for it. Therefore, don't ever reduce the length at this point.
2811 */
2812
2813 if (code < last_code) code = last_code;
2814
2815 /* Paranoid check for integer overflow */
2816
2817 if (OFLOW_MAX - *lengthptr < code - last_code)
2818 {
2819 *errorcodeptr = ERR20;
2820 goto FAILED;
2821 }
2822
2823 *lengthptr += code - last_code;
2824 DPRINTF(("length=%d added %d c=%c\n", *lengthptr, code - last_code, c));
2825
2826 /* If "previous" is set and it is not at the start of the work space, move
2827 it back to there, in order to avoid filling up the work space. Otherwise,
2828 if "previous" is NULL, reset the current code pointer to the start. */
2829
2830 if (previous != NULL)
2831 {
2832 if (previous > orig_code)
2833 {
2834 memmove(orig_code, previous, code - previous);
2835 code -= previous - orig_code;
2836 previous = orig_code;
2837 }
2838 }
2839 else code = orig_code;
2840
2841 /* Remember where this code item starts so we can pick up the length
2842 next time round. */
2843
2844 last_code = code;
2845 }
2846
2847 /* In the real compile phase, just check the workspace used by the forward
2848 reference list. */
2849
2850 else if (cd->hwm > cd->start_workspace + WORK_SIZE_CHECK)
2851 {
2852 *errorcodeptr = ERR52;
2853 goto FAILED;
2854 }
2855
2856 /* If in \Q...\E, check for the end; if not, we have a literal */
2857
2858 if (inescq && c != 0)
2859 {
2860 if (c == CHAR_BACKSLASH && ptr[1] == CHAR_E)
2861 {
2862 inescq = FALSE;
2863 ptr++;
2864 continue;
2865 }
2866 else
2867 {
2868 if (previous_callout != NULL)
2869 {
2870 if (lengthptr == NULL) /* Don't attempt in pre-compile phase */
2871 complete_callout(previous_callout, ptr, cd);
2872 previous_callout = NULL;
2873 }
2874 if ((options & PCRE_AUTO_CALLOUT) != 0)
2875 {
2876 previous_callout = code;
2877 code = auto_callout(code, ptr, cd);
2878 }
2879 goto NORMAL_CHAR;
2880 }
2881 }
2882
2883 /* Fill in length of a previous callout, except when the next thing is
2884 a quantifier. */
2885
2886 is_quantifier =
2887 c == CHAR_ASTERISK || c == CHAR_PLUS || c == CHAR_QUESTION_MARK ||
2888 (c == CHAR_LEFT_CURLY_BRACKET && is_counted_repeat(ptr+1));
2889
2890 if (!is_quantifier && previous_callout != NULL &&
2891 after_manual_callout-- <= 0)
2892 {
2893 if (lengthptr == NULL) /* Don't attempt in pre-compile phase */
2894 complete_callout(previous_callout, ptr, cd);
2895 previous_callout = NULL;
2896 }
2897
2898 /* In extended mode, skip white space and comments */
2899
2900 if ((options & PCRE_EXTENDED) != 0)
2901 {
2902 if ((cd->ctypes[c] & ctype_space) != 0) continue;
2903 if (c == CHAR_NUMBER_SIGN)
2904 {
2905 while (*(++ptr) != 0)
2906 {
2907 if (IS_NEWLINE(ptr)) { ptr += cd->nllen - 1; break; }
2908 }
2909 if (*ptr != 0) continue;
2910
2911 /* Else fall through to handle end of string */
2912 c = 0;
2913 }
2914 }
2915
2916 /* No auto callout for quantifiers. */
2917
2918 if ((options & PCRE_AUTO_CALLOUT) != 0 && !is_quantifier)
2919 {
2920 previous_callout = code;
2921 code = auto_callout(code, ptr, cd);
2922 }
2923
2924 switch(c)
2925 {
2926 /* ===================================================================*/
2927 case 0: /* The branch terminates at string end */
2928 case CHAR_VERTICAL_LINE: /* or | or ) */
2929 case CHAR_RIGHT_PARENTHESIS:
2930 *firstbyteptr = firstbyte;
2931 *reqbyteptr = reqbyte;
2932 *codeptr = code;
2933 *ptrptr = ptr;
2934 if (lengthptr != NULL)
2935 {
2936 if (OFLOW_MAX - *lengthptr < code - last_code)
2937 {
2938 *errorcodeptr = ERR20;
2939 goto FAILED;
2940 }
2941 *lengthptr += code - last_code; /* To include callout length */
2942 DPRINTF((">> end branch\n"));
2943 }
2944 return TRUE;
2945
2946
2947 /* ===================================================================*/
2948 /* Handle single-character metacharacters. In multiline mode, ^ disables
2949 the setting of any following char as a first character. */
2950
2951 case CHAR_CIRCUMFLEX_ACCENT:
2952 if ((options & PCRE_MULTILINE) != 0)
2953 {
2954 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
2955 }
2956 previous = NULL;
2957 *code++ = OP_CIRC;
2958 break;
2959
2960 case CHAR_DOLLAR_SIGN:
2961 previous = NULL;
2962 *code++ = OP_DOLL;
2963 break;
2964
2965 /* There can never be a first char if '.' is first, whatever happens about
2966 repeats. The value of reqbyte doesn't change either. */
2967
2968 case CHAR_DOT:
2969 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
2970 zerofirstbyte = firstbyte;
2971 zeroreqbyte = reqbyte;
2972 previous = code;
2973 *code++ = ((options & PCRE_DOTALL) != 0)? OP_ALLANY: OP_ANY;
2974 break;
2975
2976
2977 /* ===================================================================*/
2978 /* Character classes. If the included characters are all < 256, we build a
2979 32-byte bitmap of the permitted characters, except in the special case
2980 where there is only one such character. For negated classes, we build the
2981 map as usual, then invert it at the end. However, we use a different opcode
2982 so that data characters > 255 can be handled correctly.
2983
2984 If the class contains characters outside the 0-255 range, a different
2985 opcode is compiled. It may optionally have a bit map for characters < 256,
2986 but those above are are explicitly listed afterwards. A flag byte tells
2987 whether the bitmap is present, and whether this is a negated class or not.
2988
2989 In JavaScript compatibility mode, an isolated ']' causes an error. In
2990 default (Perl) mode, it is treated as a data character. */
2991
2992 case CHAR_RIGHT_SQUARE_BRACKET:
2993 if ((cd->external_options & PCRE_JAVASCRIPT_COMPAT) != 0)
2994 {
2995 *errorcodeptr = ERR64;
2996 goto FAILED;
2997 }
2998 goto NORMAL_CHAR;
2999
3000 case CHAR_LEFT_SQUARE_BRACKET:
3001 previous = code;
3002
3003 /* PCRE supports POSIX class stuff inside a class. Perl gives an error if
3004 they are encountered at the top level, so we'll do that too. */
3005
3006 if ((ptr[1] == CHAR_COLON || ptr[1] == CHAR_DOT ||
3007 ptr[1] == CHAR_EQUALS_SIGN) &&
3008 check_posix_syntax(ptr, &tempptr))
3009 {
3010 *errorcodeptr = (ptr[1] == CHAR_COLON)? ERR13 : ERR31;
3011 goto FAILED;
3012 }
3013
3014 /* If the first character is '^', set the negation flag and skip it. Also,
3015 if the first few characters (either before or after ^) are \Q\E or \E we
3016 skip them too. This makes for compatibility with Perl. */
3017
3018 negate_class = FALSE;
3019 for (;;)
3020 {
3021 c = *(++ptr);
3022 if (c == CHAR_BACKSLASH)
3023 {
3024 if (ptr[1] == CHAR_E)
3025 ptr++;
3026 else if (strncmp((const char *)ptr+1,
3027 STR_Q STR_BACKSLASH STR_E, 3) == 0)
3028 ptr += 3;
3029 else
3030 break;
3031 }
3032 else if (!negate_class && c == CHAR_CIRCUMFLEX_ACCENT)
3033 negate_class = TRUE;
3034 else break;
3035 }
3036
3037 /* Empty classes are allowed in JavaScript compatibility mode. Otherwise,
3038 an initial ']' is taken as a data character -- the code below handles
3039 that. In JS mode, [] must always fail, so generate OP_FAIL, whereas
3040 [^] must match any character, so generate OP_ALLANY. */
3041
3042 if (c == CHAR_RIGHT_SQUARE_BRACKET &&
3043 (cd->external_options & PCRE_JAVASCRIPT_COMPAT) != 0)
3044 {
3045 *code++ = negate_class? OP_ALLANY : OP_FAIL;
3046 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
3047 zerofirstbyte = firstbyte;
3048 break;
3049 }
3050
3051 /* If a class contains a negative special such as \S, we need to flip the
3052 negation flag at the end, so that support for characters > 255 works
3053 correctly (they are all included in the class). */
3054
3055 should_flip_negation = FALSE;
3056
3057 /* Keep a count of chars with values < 256 so that we can optimize the case
3058 of just a single character (as long as it's < 256). However, For higher
3059 valued UTF-8 characters, we don't yet do any optimization. */
3060
3061 class_charcount = 0;
3062 class_lastchar = -1;
3063
3064 /* Initialize the 32-char bit map to all zeros. We build the map in a
3065 temporary bit of memory, in case the class contains only 1 character (less
3066 than 256), because in that case the compiled code doesn't use the bit map.
3067 */
3068
3069 memset(classbits, 0, 32 * sizeof(uschar));
3070
3071 #ifdef SUPPORT_UTF8
3072 class_utf8 = FALSE; /* No chars >= 256 */
3073 class_utf8data = code + LINK_SIZE + 2; /* For UTF-8 items */
3074 class_utf8data_base = class_utf8data; /* For resetting in pass 1 */
3075 #endif
3076
3077 /* Process characters until ] is reached. By writing this as a "do" it
3078 means that an initial ] is taken as a data character. At the start of the
3079 loop, c contains the first byte of the character. */
3080
3081 if (c != 0) do
3082 {
3083 const uschar *oldptr;
3084
3085 #ifdef SUPPORT_UTF8
3086 if (utf8 && c > 127)
3087 { /* Braces are required because the */
3088 GETCHARLEN(c, ptr, ptr); /* macro generates multiple statements */
3089 }
3090
3091 /* In the pre-compile phase, accumulate the length of any UTF-8 extra
3092 data and reset the pointer. This is so that very large classes that
3093 contain a zillion UTF-8 characters no longer overwrite the work space
3094 (which is on the stack). */
3095
3096 if (lengthptr != NULL)
3097 {
3098 *lengthptr += class_utf8data - class_utf8data_base;
3099 class_utf8data = class_utf8data_base;
3100 }
3101
3102 #endif
3103
3104 /* Inside \Q...\E everything is literal except \E */
3105
3106 if (inescq)
3107 {
3108 if (c == CHAR_BACKSLASH && ptr[1] == CHAR_E) /* If we are at \E */
3109 {
3110 inescq = FALSE; /* Reset literal state */
3111 ptr++; /* Skip the 'E' */
3112 continue; /* Carry on with next */
3113 }
3114 goto CHECK_RANGE; /* Could be range if \E follows */
3115 }
3116
3117 /* Handle POSIX class names. Perl allows a negation extension of the
3118 form [:^name:]. A square bracket that doesn't match the syntax is
3119 treated as a literal. We also recognize the POSIX constructions
3120 [.ch.] and [=ch=] ("collating elements") and fault them, as Perl
3121 5.6 and 5.8 do. */
3122
3123 if (c == CHAR_LEFT_SQUARE_BRACKET &&
3124 (ptr[1] == CHAR_COLON || ptr[1] == CHAR_DOT ||
3125 ptr[1] == CHAR_EQUALS_SIGN) && check_posix_syntax(ptr, &tempptr))
3126 {
3127 BOOL local_negate = FALSE;
3128 int posix_class, taboffset, tabopt;
3129 register const uschar *cbits = cd->cbits;
3130 uschar pbits[32];
3131
3132 if (ptr[1] != CHAR_COLON)
3133 {
3134 *errorcodeptr = ERR31;
3135 goto FAILED;
3136 }
3137
3138 ptr += 2;
3139 if (*ptr == CHAR_CIRCUMFLEX_ACCENT)
3140 {
3141 local_negate = TRUE;
3142 should_flip_negation = TRUE; /* Note negative special */
3143 ptr++;
3144 }
3145
3146 posix_class = check_posix_name(ptr, tempptr - ptr);
3147 if (posix_class < 0)
3148 {
3149 *errorcodeptr = ERR30;
3150 goto FAILED;
3151 }
3152
3153 /* If matching is caseless, upper and lower are converted to
3154 alpha. This relies on the fact that the class table starts with
3155 alpha, lower, upper as the first 3 entries. */
3156
3157 if ((options & PCRE_CASELESS) != 0 && posix_class <= 2)
3158 posix_class = 0;
3159
3160 /* We build the bit map for the POSIX class in a chunk of local store
3161 because we may be adding and subtracting from it, and we don't want to
3162 subtract bits that may be in the main map already. At the end we or the
3163 result into the bit map that is being built. */
3164
3165 posix_class *= 3;
3166
3167 /* Copy in the first table (always present) */
3168
3169 memcpy(pbits, cbits + posix_class_maps[posix_class],
3170 32 * sizeof(uschar));
3171
3172 /* If there is a second table, add or remove it as required. */
3173
3174 taboffset = posix_class_maps[posix_class + 1];
3175 tabopt = posix_class_maps[posix_class + 2];
3176
3177 if (taboffset >= 0)
3178 {
3179 if (tabopt >= 0)
3180 for (c = 0; c < 32; c++) pbits[c] |= cbits[c + taboffset];
3181 else
3182 for (c = 0; c < 32; c++) pbits[c] &= ~cbits[c + taboffset];
3183 }
3184
3185 /* Not see if we need to remove any special characters. An option
3186 value of 1 removes vertical space and 2 removes underscore. */
3187
3188 if (tabopt < 0) tabopt = -tabopt;
3189 if (tabopt == 1) pbits[1] &= ~0x3c;
3190 else if (tabopt == 2) pbits[11] &= 0x7f;
3191
3192 /* Add the POSIX table or its complement into the main table that is
3193 being built and we are done. */
3194
3195 if (local_negate)
3196 for (c = 0; c < 32; c++) classbits[c] |= ~pbits[c];
3197 else
3198 for (c = 0; c < 32; c++) classbits[c] |= pbits[c];
3199
3200 ptr = tempptr + 1;
3201 class_charcount = 10; /* Set > 1; assumes more than 1 per class */
3202 continue; /* End of POSIX syntax handling */
3203 }
3204
3205 /* Backslash may introduce a single character, or it may introduce one
3206 of the specials, which just set a flag. The sequence \b is a special
3207 case. Inside a class (and only there) it is treated as backspace. We
3208 assume that other escapes have more than one character in them, so set
3209 class_charcount bigger than one. Unrecognized escapes fall through and
3210 are either treated as literal characters (by default), or are faulted if
3211 PCRE_EXTRA is set. */
3212
3213 if (c == CHAR_BACKSLASH)
3214 {
3215 c = check_escape(&ptr, errorcodeptr, cd->bracount, options, TRUE);
3216 if (*errorcodeptr != 0) goto FAILED;
3217
3218 if (-c == ESC_b) c = CHAR_BS; /* \b is backspace in a class */
3219 else if (-c == ESC_Q) /* Handle start of quoted string */
3220 {
3221 if (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
3222 {
3223 ptr += 2; /* avoid empty string */
3224 }
3225 else inescq = TRUE;
3226 continue;
3227 }
3228 else if (-c == ESC_E) continue; /* Ignore orphan \E */
3229
3230 if (c < 0)
3231 {
3232 register const uschar *cbits = cd->cbits;
3233 class_charcount += 2; /* Greater than 1 is what matters */
3234
3235 /* Save time by not doing this in the pre-compile phase. */
3236
3237 if (lengthptr == NULL) switch (-c)
3238 {
3239 case ESC_d:
3240 for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_digit];
3241 continue;
3242
3243 case ESC_D:
3244 should_flip_negation = TRUE;
3245 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_digit];
3246 continue;
3247
3248 case ESC_w:
3249 for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_word];
3250 continue;
3251
3252 case ESC_W:
3253 should_flip_negation = TRUE;
3254 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_word];
3255 continue;
3256
3257 case ESC_s:
3258 for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_space];
3259 classbits[1] &= ~0x08; /* Perl 5.004 onwards omits VT from \s */
3260 continue;
3261
3262 case ESC_S:
3263 should_flip_negation = TRUE;
3264 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_space];
3265 classbits[1] |= 0x08; /* Perl 5.004 onwards omits VT from \s */
3266 continue;
3267
3268 default: /* Not recognized; fall through */
3269 break; /* Need "default" setting to stop compiler warning. */
3270 }
3271
3272 /* In the pre-compile phase, just do the recognition. */
3273
3274 else if (c == -ESC_d || c == -ESC_D || c == -ESC_w ||
3275 c == -ESC_W || c == -ESC_s || c == -ESC_S) continue;
3276
3277 /* We need to deal with \H, \h, \V, and \v in both phases because
3278 they use extra memory. */
3279
3280 if (-c == ESC_h)
3281 {
3282 SETBIT(classbits, 0x09); /* VT */
3283 SETBIT(classbits, 0x20); /* SPACE */
3284 SETBIT(classbits, 0xa0); /* NSBP */
3285 #ifdef SUPPORT_UTF8
3286 if (utf8)
3287 {
3288 class_utf8 = TRUE;
3289 *class_utf8data++ = XCL_SINGLE;
3290 class_utf8data += _pcre_ord2utf8(0x1680, class_utf8data);
3291 *class_utf8data++ = XCL_SINGLE;
3292 class_utf8data += _pcre_ord2utf8(0x180e, class_utf8data);
3293 *class_utf8data++ = XCL_RANGE;
3294 class_utf8data += _pcre_ord2utf8(0x2000, class_utf8data);
3295 class_utf8data += _pcre_ord2utf8(0x200A, class_utf8data);
3296 *class_utf8data++ = XCL_SINGLE;
3297 class_utf8data += _pcre_ord2utf8(0x202f, class_utf8data);
3298 *class_utf8data++ = XCL_SINGLE;
3299 class_utf8data += _pcre_ord2utf8(0x205f, class_utf8data);
3300 *class_utf8data++ = XCL_SINGLE;
3301 class_utf8data += _pcre_ord2utf8(0x3000, class_utf8data);
3302 }
3303 #endif
3304 continue;
3305 }
3306
3307 if (-c == ESC_H)
3308 {
3309 for (c = 0; c < 32; c++)
3310 {
3311 int x = 0xff;
3312 switch (c)
3313 {
3314 case 0x09/8: x ^= 1 << (0x09%8); break;
3315 case 0x20/8: x ^= 1 << (0x20%8); break;
3316 case 0xa0/8: x ^= 1 << (0xa0%8); break;
3317 default: break;
3318 }
3319 classbits[c] |= x;
3320 }
3321
3322 #ifdef SUPPORT_UTF8
3323 if (utf8)
3324 {
3325 class_utf8 = TRUE;
3326 *class_utf8data++ = XCL_RANGE;
3327 class_utf8data += _pcre_ord2utf8(0x0100, class_utf8data);
3328 class_utf8data += _pcre_ord2utf8(0x167f, class_utf8data);
3329 *class_utf8data++ = XCL_RANGE;
3330 class_utf8data += _pcre_ord2utf8(0x1681, class_utf8data);
3331 class_utf8data += _pcre_ord2utf8(0x180d, class_utf8data);
3332 *class_utf8data++ = XCL_RANGE;
3333 class_utf8data += _pcre_ord2utf8(0x180f, class_utf8data);
3334 class_utf8data += _pcre_ord2utf8(0x1fff, class_utf8data);
3335 *class_utf8data++ = XCL_RANGE;
3336 class_utf8data += _pcre_ord2utf8(0x200B, class_utf8data);
3337 class_utf8data += _pcre_ord2utf8(0x202e, class_utf8data);
3338 *class_utf8data++ = XCL_RANGE;
3339 class_utf8data += _pcre_ord2utf8(0x2030, class_utf8data);
3340 class_utf8data += _pcre_ord2utf8(0x205e, class_utf8data);
3341 *class_utf8data++ = XCL_RANGE;
3342 class_utf8data += _pcre_ord2utf8(0x2060, class_utf8data);
3343 class_utf8data += _pcre_ord2utf8(0x2fff, class_utf8data);
3344 *class_utf8data++ = XCL_RANGE;
3345 class_utf8data += _pcre_ord2utf8(0x3001, class_utf8data);
3346 class_utf8data += _pcre_ord2utf8(0x7fffffff, class_utf8data);
3347 }
3348 #endif
3349 continue;
3350 }
3351
3352 if (-c == ESC_v)
3353 {
3354 SETBIT(classbits, 0x0a); /* LF */
3355 SETBIT(classbits, 0x0b); /* VT */
3356 SETBIT(classbits, 0x0c); /* FF */
3357 SETBIT(classbits, 0x0d); /* CR */
3358 SETBIT(classbits, 0x85); /* NEL */
3359 #ifdef SUPPORT_UTF8
3360 if (utf8)
3361 {
3362 class_utf8 = TRUE;
3363 *class_utf8data++ = XCL_RANGE;
3364 class_utf8data += _pcre_ord2utf8(0x2028, class_utf8data);
3365 class_utf8data += _pcre_ord2utf8(0x2029, class_utf8data);
3366 }
3367 #endif
3368 continue;
3369 }
3370
3371 if (-c == ESC_V)
3372 {
3373 for (c = 0; c < 32; c++)
3374 {
3375 int x = 0xff;
3376 switch (c)
3377 {
3378 case 0x0a/8: x ^= 1 << (0x0a%8);
3379 x ^= 1 << (0x0b%8);
3380 x ^= 1 << (0x0c%8);
3381 x ^= 1 << (0x0d%8);
3382 break;
3383 case 0x85/8: x ^= 1 << (0x85%8); break;
3384 default: break;
3385 }
3386 classbits[c] |= x;
3387 }
3388
3389 #ifdef SUPPORT_UTF8
3390 if (utf8)
3391 {
3392 class_utf8 = TRUE;
3393 *class_utf8data++ = XCL_RANGE;
3394 class_utf8data += _pcre_ord2utf8(0x0100, class_utf8data);
3395 class_utf8data += _pcre_ord2utf8(0x2027, class_utf8data);
3396 *class_utf8data++ = XCL_RANGE;
3397 class_utf8data += _pcre_ord2utf8(0x2029, class_utf8data);
3398 class_utf8data += _pcre_ord2utf8(0x7fffffff, class_utf8data);
3399 }
3400 #endif
3401 continue;
3402 }
3403
3404 /* We need to deal with \P and \p in both phases. */
3405
3406 #ifdef SUPPORT_UCP
3407 if (-c == ESC_p || -c == ESC_P)
3408 {
3409 BOOL negated;
3410 int pdata;
3411 int ptype = get_ucp(&ptr, &negated, &pdata, errorcodeptr);
3412 if (ptype < 0) goto FAILED;
3413 class_utf8 = TRUE;
3414 *class_utf8data++ = ((-c == ESC_p) != negated)?
3415 XCL_PROP : XCL_NOTPROP;
3416 *class_utf8data++ = ptype;
3417 *class_utf8data++ = pdata;
3418 class_charcount -= 2; /* Not a < 256 character */
3419 continue;
3420 }
3421 #endif
3422 /* Unrecognized escapes are faulted if PCRE is running in its
3423 strict mode. By default, for compatibility with Perl, they are
3424 treated as literals. */
3425
3426 if ((options & PCRE_EXTRA) != 0)
3427 {
3428 *errorcodeptr = ERR7;
3429 goto FAILED;
3430 }
3431
3432 class_charcount -= 2; /* Undo the default count from above */
3433 c = *ptr; /* Get the final character and fall through */
3434 }
3435
3436 /* Fall through if we have a single character (c >= 0). This may be
3437 greater than 256 in UTF-8 mode. */
3438
3439 } /* End of backslash handling */
3440
3441 /* A single character may be followed by '-' to form a range. However,
3442 Perl does not permit ']' to be the end of the range. A '-' character
3443 at the end is treated as a literal. Perl ignores orphaned \E sequences
3444 entirely. The code for handling \Q and \E is messy. */
3445
3446 CHECK_RANGE:
3447 while (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
3448 {
3449 inescq = FALSE;
3450 ptr += 2;
3451 }
3452
3453 oldptr = ptr;
3454
3455 /* Remember \r or \n */
3456
3457 if (c == CHAR_CR || c == CHAR_NL) cd->external_flags |= PCRE_HASCRORLF;
3458
3459 /* Check for range */
3460
3461 if (!inescq && ptr[1] == CHAR_MINUS)
3462 {
3463 int d;
3464 ptr += 2;
3465 while (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_E) ptr += 2;
3466
3467 /* If we hit \Q (not followed by \E) at this point, go into escaped
3468 mode. */
3469
3470 while (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_Q)
3471 {
3472 ptr += 2;
3473 if (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_E)
3474 { ptr += 2; continue; }
3475 inescq = TRUE;
3476 break;
3477 }
3478
3479 if (*ptr == 0 || (!inescq && *ptr == CHAR_RIGHT_SQUARE_BRACKET))
3480 {
3481 ptr = oldptr;
3482 goto LONE_SINGLE_CHARACTER;
3483 }
3484
3485 #ifdef SUPPORT_UTF8
3486 if (utf8)
3487 { /* Braces are required because the */
3488 GETCHARLEN(d, ptr, ptr); /* macro generates multiple statements */
3489 }
3490 else
3491 #endif
3492 d = *ptr; /* Not UTF-8 mode */
3493
3494 /* The second part of a range can be a single-character escape, but
3495 not any of the other escapes. Perl 5.6 treats a hyphen as a literal
3496 in such circumstances. */
3497
3498 if (!inescq && d == CHAR_BACKSLASH)
3499 {
3500 d = check_escape(&ptr, errorcodeptr, cd->bracount, options, TRUE);
3501 if (*errorcodeptr != 0) goto FAILED;
3502
3503 /* \b is backspace; \X is literal X; \R is literal R; any other
3504 special means the '-' was literal */
3505
3506 if (d < 0)
3507 {
3508 if (d == -ESC_b) d = CHAR_BS;
3509 else if (d == -ESC_X) d = CHAR_X;
3510 else if (d == -ESC_R) d = CHAR_R; else
3511 {
3512 ptr = oldptr;
3513 goto LONE_SINGLE_CHARACTER; /* A few lines below */
3514 }
3515 }
3516 }
3517
3518 /* Check that the two values are in the correct order. Optimize
3519 one-character ranges */
3520
3521 if (d < c)
3522 {
3523 *errorcodeptr = ERR8;
3524 goto FAILED;
3525 }
3526
3527 if (d == c) goto LONE_SINGLE_CHARACTER; /* A few lines below */
3528
3529 /* Remember \r or \n */
3530
3531 if (d == CHAR_CR || d == CHAR_NL) cd->external_flags |= PCRE_HASCRORLF;
3532
3533 /* In UTF-8 mode, if the upper limit is > 255, or > 127 for caseless
3534 matching, we have to use an XCLASS with extra data items. Caseless
3535 matching for characters > 127 is available only if UCP support is
3536 available. */
3537
3538 #ifdef SUPPORT_UTF8
3539 if (utf8 && (d > 255 || ((options & PCRE_CASELESS) != 0 && d > 127)))
3540 {
3541 class_utf8 = TRUE;
3542
3543 /* With UCP support, we can find the other case equivalents of
3544 the relevant characters. There may be several ranges. Optimize how
3545 they fit with the basic range. */
3546
3547 #ifdef SUPPORT_UCP
3548 if ((options & PCRE_CASELESS) != 0)
3549 {
3550 unsigned int occ, ocd;
3551 unsigned int cc = c;
3552 unsigned int origd = d;
3553 while (get_othercase_range(&cc, origd, &occ, &ocd))
3554 {
3555 if (occ >= (unsigned int)c &&
3556 ocd <= (unsigned int)d)
3557 continue; /* Skip embedded ranges */
3558
3559 if (occ < (unsigned int)c &&
3560 ocd >= (unsigned int)c - 1) /* Extend the basic range */
3561 { /* if there is overlap, */
3562 c = occ; /* noting that if occ < c */
3563 continue; /* we can't have ocd > d */
3564 } /* because a subrange is */
3565 if (ocd > (unsigned int)d &&
3566 occ <= (unsigned int)d + 1) /* always shorter than */
3567 { /* the basic range. */
3568 d = ocd;
3569 continue;
3570 }
3571
3572 if (occ == ocd)
3573 {
3574 *class_utf8data++ = XCL_SINGLE;
3575 }
3576 else
3577 {
3578 *class_utf8data++ = XCL_RANGE;
3579 class_utf8data += _pcre_ord2utf8(occ, class_utf8data);
3580 }
3581 class_utf8data += _pcre_ord2utf8(ocd, class_utf8data);
3582 }
3583 }
3584 #endif /* SUPPORT_UCP */
3585
3586 /* Now record the original range, possibly modified for UCP caseless
3587 overlapping ranges. */
3588
3589 *class_utf8data++ = XCL_RANGE;
3590 class_utf8data += _pcre_ord2utf8(c, class_utf8data);
3591 class_utf8data += _pcre_ord2utf8(d, class_utf8data);
3592
3593 /* With UCP support, we are done. Without UCP support, there is no
3594 caseless matching for UTF-8 characters > 127; we can use the bit map
3595 for the smaller ones. */
3596
3597 #ifdef SUPPORT_UCP
3598 continue; /* With next character in the class */
3599 #else
3600 if ((options & PCRE_CASELESS) == 0 || c > 127) continue;
3601
3602 /* Adjust upper limit and fall through to set up the map */
3603
3604 d = 127;
3605
3606 #endif /* SUPPORT_UCP */
3607 }
3608 #endif /* SUPPORT_UTF8 */
3609
3610 /* We use the bit map for all cases when not in UTF-8 mode; else
3611 ranges that lie entirely within 0-127 when there is UCP support; else
3612 for partial ranges without UCP support. */
3613
3614 class_charcount += d - c + 1;
3615 class_lastchar = d;
3616
3617 /* We can save a bit of time by skipping this in the pre-compile. */
3618
3619 if (lengthptr == NULL) for (; c <= d; c++)
3620 {
3621 classbits[c/8] |= (1 << (c&7));
3622 if ((options & PCRE_CASELESS) != 0)
3623 {
3624 int uc = cd->fcc[c]; /* flip case */
3625 classbits[uc/8] |= (1 << (uc&7));
3626 }
3627 }
3628
3629 continue; /* Go get the next char in the class */
3630 }
3631
3632 /* Handle a lone single character - we can get here for a normal
3633 non-escape char, or after \ that introduces a single character or for an
3634 apparent range that isn't. */
3635
3636 LONE_SINGLE_CHARACTER:
3637
3638 /* Handle a character that cannot go in the bit map */
3639
3640 #ifdef SUPPORT_UTF8
3641 if (utf8 && (c > 255 || ((options & PCRE_CASELESS) != 0 && c > 127)))
3642 {
3643 class_utf8 = TRUE;
3644 *class_utf8data++ = XCL_SINGLE;
3645 class_utf8data += _pcre_ord2utf8(c, class_utf8data);
3646
3647 #ifdef SUPPORT_UCP
3648 if ((options & PCRE_CASELESS) != 0)
3649 {
3650 unsigned int othercase;
3651 if ((othercase = UCD_OTHERCASE(c)) != c)
3652 {
3653 *class_utf8data++ = XCL_SINGLE;
3654 class_utf8data += _pcre_ord2utf8(othercase, class_utf8data);
3655 }
3656 }
3657 #endif /* SUPPORT_UCP */
3658
3659 }
3660 else
3661 #endif /* SUPPORT_UTF8 */
3662
3663 /* Handle a single-byte character */
3664 {
3665 classbits[c/8] |= (1 << (c&7));
3666 if ((options & PCRE_CASELESS) != 0)
3667 {
3668 c = cd->fcc[c]; /* flip case */
3669 classbits[c/8] |= (1 << (c&7));
3670 }
3671 class_charcount++;
3672 class_lastchar = c;
3673 }
3674 }
3675
3676 /* Loop until ']' reached. This "while" is the end of the "do" above. */
3677
3678 while ((c = *(++ptr)) != 0 && (c != CHAR_RIGHT_SQUARE_BRACKET || inescq));
3679
3680 if (c == 0) /* Missing terminating ']' */
3681 {
3682 *errorcodeptr = ERR6;
3683 goto FAILED;
3684 }
3685
3686
3687 /* This code has been disabled because it would mean that \s counts as
3688 an explicit \r or \n reference, and that's not really what is wanted. Now
3689 we set the flag only if there is a literal "\r" or "\n" in the class. */
3690
3691 #if 0
3692 /* Remember whether \r or \n are in this class */
3693
3694 if (negate_class)
3695 {
3696 if ((classbits[1] & 0x24) != 0x24) cd->external_flags |= PCRE_HASCRORLF;
3697 }
3698 else
3699 {
3700 if ((classbits[1] & 0x24) != 0) cd->external_flags |= PCRE_HASCRORLF;
3701 }
3702 #endif
3703
3704
3705 /* If class_charcount is 1, we saw precisely one character whose value is
3706 less than 256. As long as there were no characters >= 128 and there was no
3707 use of \p or \P, in other words, no use of any XCLASS features, we can
3708 optimize.
3709
3710 In UTF-8 mode, we can optimize the negative case only if there were no
3711 characters >= 128 because OP_NOT and the related opcodes like OP_NOTSTAR
3712 operate on single-bytes only. This is an historical hangover. Maybe one day
3713 we can tidy these opcodes to handle multi-byte characters.
3714
3715 The optimization throws away the bit map. We turn the item into a
3716 1-character OP_CHAR[NC] if it's positive, or OP_NOT if it's negative. Note
3717 that OP_NOT does not support multibyte characters. In the positive case, it
3718 can cause firstbyte to be set. Otherwise, there can be no first char if
3719 this item is first, whatever repeat count may follow. In the case of
3720 reqbyte, save the previous value for reinstating. */
3721
3722 #ifdef SUPPORT_UTF8
3723 if (class_charcount == 1 && !class_utf8 &&
3724 (!utf8 || !negate_class || class_lastchar < 128))
3725 #else
3726 if (class_charcount == 1)
3727 #endif
3728 {
3729 zeroreqbyte = reqbyte;
3730
3731 /* The OP_NOT opcode works on one-byte characters only. */
3732
3733 if (negate_class)
3734 {
3735 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
3736 zerofirstbyte = firstbyte;
3737 *code++ = OP_NOT;
3738 *code++ = class_lastchar;
3739 break;
3740 }
3741
3742 /* For a single, positive character, get the value into mcbuffer, and
3743 then we can handle this with the normal one-character code. */
3744
3745 #ifdef SUPPORT_UTF8
3746 if (utf8 && class_lastchar > 127)
3747 mclength = _pcre_ord2utf8(class_lastchar, mcbuffer);
3748 else
3749 #endif
3750 {
3751 mcbuffer[0] = class_lastchar;
3752 mclength = 1;
3753 }
3754 goto ONE_CHAR;
3755 } /* End of 1-char optimization */
3756
3757 /* The general case - not the one-char optimization. If this is the first
3758 thing in the branch, there can be no first char setting, whatever the
3759 repeat count. Any reqbyte setting must remain unchanged after any kind of
3760 repeat. */
3761
3762 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
3763 zerofirstbyte = firstbyte;
3764 zeroreqbyte = reqbyte;
3765
3766 /* If there are characters with values > 255, we have to compile an
3767 extended class, with its own opcode, unless there was a negated special
3768 such as \S in the class, because in that case all characters > 255 are in
3769 the class, so any that were explicitly given as well can be ignored. If
3770 (when there are explicit characters > 255 that must be listed) there are no
3771 characters < 256, we can omit the bitmap in the actual compiled code. */
3772
3773 #ifdef SUPPORT_UTF8
3774 if (class_utf8 && !should_flip_negation)
3775 {
3776 *class_utf8data++ = XCL_END; /* Marks the end of extra data */
3777 *code++ = OP_XCLASS;
3778 code += LINK_SIZE;
3779 *code = negate_class? XCL_NOT : 0;
3780
3781 /* If the map is required, move up the extra data to make room for it;
3782 otherwise just move the code pointer to the end of the extra data. */
3783
3784 if (class_charcount > 0)
3785 {
3786 *code++ |= XCL_MAP;
3787 memmove(code + 32, code, class_utf8data - code);
3788 memcpy(code, classbits, 32);
3789 code = class_utf8data + 32;
3790 }
3791 else code = class_utf8data;
3792
3793 /* Now fill in the complete length of the item */
3794
3795 PUT(previous, 1, code - previous);
3796 break; /* End of class handling */
3797 }
3798 #endif
3799
3800 /* If there are no characters > 255, set the opcode to OP_CLASS or
3801 OP_NCLASS, depending on whether the whole class was negated and whether
3802 there were negative specials such as \S in the class. Then copy the 32-byte
3803 map into the code vector, negating it if necessary. */
3804
3805 *code++ = (negate_class == should_flip_negation) ? OP_CLASS : OP_NCLASS;
3806 if (negate_class)
3807 {
3808 if (lengthptr == NULL) /* Save time in the pre-compile phase */
3809 for (c = 0; c < 32; c++) code[c] = ~classbits[c];
3810 }
3811 else
3812 {
3813 memcpy(code, classbits, 32);
3814 }
3815 code += 32;
3816 break;
3817
3818
3819 /* ===================================================================*/
3820 /* Various kinds of repeat; '{' is not necessarily a quantifier, but this
3821 has been tested above. */
3822
3823 case CHAR_LEFT_CURLY_BRACKET:
3824 if (!is_quantifier) goto NORMAL_CHAR;
3825 ptr = read_repeat_counts(ptr+1, &repeat_min, &repeat_max, errorcodeptr);
3826 if (*errorcodeptr != 0) goto FAILED;
3827 goto REPEAT;
3828
3829 case CHAR_ASTERISK:
3830 repeat_min = 0;
3831 repeat_max = -1;
3832 goto REPEAT;
3833
3834 case CHAR_PLUS:
3835 repeat_min = 1;
3836 repeat_max = -1;
3837 goto REPEAT;
3838
3839 case CHAR_QUESTION_MARK:
3840 repeat_min = 0;
3841 repeat_max = 1;
3842
3843 REPEAT:
3844 if (previous == NULL)
3845 {
3846 *errorcodeptr = ERR9;
3847 goto FAILED;
3848 }
3849
3850 if (repeat_min == 0)
3851 {
3852 firstbyte = zerofirstbyte; /* Adjust for zero repeat */
3853 reqbyte = zeroreqbyte; /* Ditto */
3854 }
3855
3856 /* Remember whether this is a variable length repeat */
3857
3858 reqvary = (repeat_min == repeat_max)? 0 : REQ_VARY;
3859
3860 op_type = 0; /* Default single-char op codes */
3861 possessive_quantifier = FALSE; /* Default not possessive quantifier */
3862
3863 /* Save start of previous item, in case we have to move it up to make space
3864 for an inserted OP_ONCE for the additional '+' extension. */
3865
3866 tempcode = previous;
3867
3868 /* If the next character is '+', we have a possessive quantifier. This
3869 implies greediness, whatever the setting of the PCRE_UNGREEDY option.
3870 If the next character is '?' this is a minimizing repeat, by default,
3871 but if PCRE_UNGREEDY is set, it works the other way round. We change the
3872 repeat type to the non-default. */
3873
3874 if (ptr[1] == CHAR_PLUS)
3875 {
3876 repeat_type = 0; /* Force greedy */
3877 possessive_quantifier = TRUE;
3878 ptr++;
3879 }
3880 else if (ptr[1] == CHAR_QUESTION_MARK)
3881 {
3882 repeat_type = greedy_non_default;
3883 ptr++;
3884 }
3885 else repeat_type = greedy_default;
3886
3887 /* If previous was a character match, abolish the item and generate a
3888 repeat item instead. If a char item has a minumum of more than one, ensure
3889 that it is set in reqbyte - it might not be if a sequence such as x{3} is
3890 the first thing in a branch because the x will have gone into firstbyte
3891 instead. */
3892
3893 if (*previous == OP_CHAR || *previous == OP_CHARNC)
3894 {
3895 /* Deal with UTF-8 characters that take up more than one byte. It's
3896 easier to write this out separately than try to macrify it. Use c to
3897 hold the length of the character in bytes, plus 0x80 to flag that it's a
3898 length rather than a small character. */
3899
3900 #ifdef SUPPORT_UTF8
3901 if (utf8 && (code[-1] & 0x80) != 0)
3902 {
3903 uschar *lastchar = code - 1;
3904 while((*lastchar & 0xc0) == 0x80) lastchar--;
3905 c = code - lastchar; /* Length of UTF-8 character */
3906 memcpy(utf8_char, lastchar, c); /* Save the char */
3907 c |= 0x80; /* Flag c as a length */
3908 }
3909 else
3910 #endif
3911
3912 /* Handle the case of a single byte - either with no UTF8 support, or
3913 with UTF-8 disabled, or for a UTF-8 character < 128. */
3914
3915 {
3916 c = code[-1];
3917 if (repeat_min > 1) reqbyte = c | req_caseopt | cd->req_varyopt;
3918 }
3919
3920 /* If the repetition is unlimited, it pays to see if the next thing on
3921 the line is something that cannot possibly match this character. If so,
3922 automatically possessifying this item gains some performance in the case
3923 where the match fails. */
3924
3925 if (!possessive_quantifier &&
3926 repeat_max < 0 &&
3927 check_auto_possessive(*previous, c, utf8, utf8_char, ptr + 1,
3928 options, cd))
3929 {
3930 repeat_type = 0; /* Force greedy */
3931 possessive_quantifier = TRUE;
3932 }
3933
3934 goto OUTPUT_SINGLE_REPEAT; /* Code shared with single character types */
3935 }
3936
3937 /* If previous was a single negated character ([^a] or similar), we use
3938 one of the special opcodes, replacing it. The code is shared with single-
3939 character repeats by setting opt_type to add a suitable offset into
3940 repeat_type. We can also test for auto-possessification. OP_NOT is
3941 currently used only for single-byte chars. */
3942
3943 else if (*previous == OP_NOT)
3944 {
3945 op_type = OP_NOTSTAR - OP_STAR; /* Use "not" opcodes */
3946 c = previous[1];
3947 if (!possessive_quantifier &&
3948 repeat_max < 0 &&
3949 check_auto_possessive(OP_NOT, c, utf8, NULL, ptr + 1, options, cd))
3950 {
3951 repeat_type = 0; /* Force greedy */
3952 possessive_quantifier = TRUE;
3953 }
3954 goto OUTPUT_SINGLE_REPEAT;
3955 }
3956
3957 /* If previous was a character type match (\d or similar), abolish it and
3958 create a suitable repeat item. The code is shared with single-character
3959 repeats by setting op_type to add a suitable offset into repeat_type. Note
3960 the the Unicode property types will be present only when SUPPORT_UCP is
3961 defined, but we don't wrap the little bits of code here because it just
3962 makes it horribly messy. */
3963
3964 else if (*previous < OP_EODN)
3965 {
3966 uschar *oldcode;
3967 int prop_type, prop_value;
3968 op_type = OP_TYPESTAR - OP_STAR; /* Use type opcodes */
3969 c = *previous;
3970
3971 if (!possessive_quantifier &&
3972 repeat_max < 0 &&
3973 check_auto_possessive(c, 0, utf8, NULL, ptr + 1, options, cd))
3974 {
3975 repeat_type = 0; /* Force greedy */
3976 possessive_quantifier = TRUE;
3977 }
3978
3979 OUTPUT_SINGLE_REPEAT:
3980 if (*previous == OP_PROP || *previous == OP_NOTPROP)
3981 {
3982 prop_type = previous[1];
3983 prop_value = previous[2];
3984 }
3985 else prop_type = prop_value = -1;
3986
3987 oldcode = code;
3988 code = previous; /* Usually overwrite previous item */
3989
3990 /* If the maximum is zero then the minimum must also be zero; Perl allows
3991 this case, so we do too - by simply omitting the item altogether. */
3992
3993 if (repeat_max == 0) goto END_REPEAT;
3994
3995 /*--------------------------------------------------------------------*/
3996 /* This code is obsolete from release 8.00; the restriction was finally
3997 removed: */
3998
3999 /* All real repeats make it impossible to handle partial matching (maybe
4000 one day we will be able to remove this restriction). */
4001
4002 /* if (repeat_max != 1) cd->external_flags |= PCRE_NOPARTIAL; */
4003 /*--------------------------------------------------------------------*/
4004
4005 /* Combine the op_type with the repeat_type */
4006
4007 repeat_type += op_type;
4008
4009 /* A minimum of zero is handled either as the special case * or ?, or as
4010 an UPTO, with the maximum given. */
4011
4012 if (repeat_min == 0)
4013 {
4014 if (repeat_max == -1) *code++ = OP_STAR + repeat_type;
4015 else if (repeat_max == 1) *code++ = OP_QUERY + repeat_type;
4016 else
4017 {
4018 *code++ = OP_UPTO + repeat_type;
4019 PUT2INC(code, 0, repeat_max);
4020 }
4021 }
4022
4023 /* A repeat minimum of 1 is optimized into some special cases. If the
4024 maximum is unlimited, we use OP_PLUS. Otherwise, the original item is
4025 left in place and, if the maximum is greater than 1, we use OP_UPTO with
4026 one less than the maximum. */
4027
4028 else if (repeat_min == 1)
4029 {
4030 if (repeat_max == -1)
4031 *code++ = OP_PLUS + repeat_type;
4032 else
4033 {
4034 code = oldcode; /* leave previous item in place */
4035 if (repeat_max == 1) goto END_REPEAT;
4036 *code++ = OP_UPTO + repeat_type;
4037 PUT2INC(code, 0, repeat_max - 1);
4038 }
4039 }
4040
4041 /* The case {n,n} is just an EXACT, while the general case {n,m} is
4042 handled as an EXACT followed by an UPTO. */
4043
4044 else
4045 {
4046 *code++ = OP_EXACT + op_type; /* NB EXACT doesn't have repeat_type */
4047 PUT2INC(code, 0, repeat_min);
4048
4049 /* If the maximum is unlimited, insert an OP_STAR. Before doing so,
4050 we have to insert the character for the previous code. For a repeated
4051 Unicode property match, there are two extra bytes that define the
4052 required property. In UTF-8 mode, long characters have their length in
4053 c, with the 0x80 bit as a flag. */
4054
4055 if (repeat_max < 0)
4056 {
4057 #ifdef SUPPORT_UTF8
4058 if (utf8 && c >= 128)
4059 {
4060 memcpy(code, utf8_char, c & 7);
4061 code += c & 7;
4062 }
4063 else
4064 #endif
4065 {
4066 *code++ = c;
4067 if (prop_type >= 0)
4068 {
4069 *code++ = prop_type;
4070 *code++ = prop_value;
4071 }
4072 }
4073 *code++ = OP_STAR + repeat_type;
4074 }
4075
4076 /* Else insert an UPTO if the max is greater than the min, again
4077 preceded by the character, for the previously inserted code. If the
4078 UPTO is just for 1 instance, we can use QUERY instead. */
4079
4080 else if (repeat_max != repeat_min)
4081 {
4082 #ifdef SUPPORT_UTF8
4083 if (utf8 && c >= 128)
4084 {
4085 memcpy(code, utf8_char, c & 7);
4086 code += c & 7;
4087 }
4088 else
4089 #endif
4090 *code++ = c;
4091 if (prop_type >= 0)
4092 {
4093 *code++ = prop_type;
4094 *code++ = prop_value;
4095 }
4096 repeat_max -= repeat_min;
4097
4098 if (repeat_max == 1)
4099 {
4100 *code++ = OP_QUERY + repeat_type;
4101 }
4102 else
4103 {
4104 *code++ = OP_UPTO + repeat_type;
4105 PUT2INC(code, 0, repeat_max);
4106 }
4107 }
4108 }
4109
4110 /* The character or character type itself comes last in all cases. */
4111
4112 #ifdef SUPPORT_UTF8
4113 if (utf8 && c >= 128)
4114 {
4115 memcpy(code, utf8_char, c & 7);
4116 code += c & 7;
4117 }
4118 else
4119 #endif
4120 *code++ = c;
4121
4122 /* For a repeated Unicode property match, there are two extra bytes that
4123 define the required property. */
4124
4125 #ifdef SUPPORT_UCP
4126 if (prop_type >= 0)
4127 {
4128 *code++ = prop_type;
4129 *code++ = prop_value;
4130 }
4131 #endif
4132 }
4133
4134 /* If previous was a character class or a back reference, we put the repeat
4135 stuff after it, but just skip the item if the repeat was {0,0}. */
4136
4137 else if (*previous == OP_CLASS ||
4138 *previous == OP_NCLASS ||
4139 #ifdef SUPPORT_UTF8
4140 *previous == OP_XCLASS ||
4141 #endif
4142 *previous == OP_REF)
4143 {
4144 if (repeat_max == 0)
4145 {
4146 code = previous;
4147 goto END_REPEAT;
4148 }
4149
4150 /*--------------------------------------------------------------------*/
4151 /* This code is obsolete from release 8.00; the restriction was finally
4152 removed: */
4153
4154 /* All real repeats make it impossible to handle partial matching (maybe
4155 one day we will be able to remove this restriction). */
4156
4157 /* if (repeat_max != 1) cd->external_flags |= PCRE_NOPARTIAL; */
4158 /*--------------------------------------------------------------------*/
4159
4160 if (repeat_min == 0 && repeat_max == -1)
4161 *code++ = OP_CRSTAR + repeat_type;
4162 else if (repeat_min == 1 && repeat_max == -1)
4163 *code++ = OP_CRPLUS + repeat_type;
4164 else if (repeat_min == 0 && repeat_max == 1)
4165 *code++ = OP_CRQUERY + repeat_type;
4166 else
4167 {
4168 *code++ = OP_CRRANGE + repeat_type;
4169 PUT2INC(code, 0, repeat_min);
4170 if (repeat_max == -1) repeat_max = 0; /* 2-byte encoding for max */
4171 PUT2INC(code, 0, repeat_max);
4172 }
4173 }
4174
4175 /* If previous was a bracket group, we may have to replicate it in certain
4176 cases. */
4177
4178 else if (*previous == OP_BRA || *previous == OP_CBRA ||
4179 *previous == OP_ONCE || *previous == OP_COND)
4180 {
4181 register int i;
4182 int ketoffset = 0;
4183 int len = code - previous;
4184 uschar *bralink = NULL;
4185
4186 /* Repeating a DEFINE group is pointless */
4187
4188 if (*previous == OP_COND && previous[LINK_SIZE+1] == OP_DEF)
4189 {
4190 *errorcodeptr = ERR55;
4191 goto FAILED;
4192 }
4193
4194 /* If the maximum repeat count is unlimited, find the end of the bracket
4195 by scanning through from the start, and compute the offset back to it
4196 from the current code pointer. There may be an OP_OPT setting following
4197 the final KET, so we can't find the end just by going back from the code
4198 pointer. */
4199
4200 if (repeat_max == -1)
4201 {
4202 register uschar *ket = previous;
4203 do ket += GET(ket, 1); while (*ket != OP_KET);
4204 ketoffset = code - ket;
4205 }
4206
4207 /* The case of a zero minimum is special because of the need to stick
4208 OP_BRAZERO in front of it, and because the group appears once in the
4209 data, whereas in other cases it appears the minimum number of times. For
4210 this reason, it is simplest to treat this case separately, as otherwise
4211 the code gets far too messy. There are several special subcases when the
4212 minimum is zero. */
4213
4214 if (repeat_min == 0)
4215 {
4216 /* If the maximum is also zero, we used to just omit the group from the
4217 output altogether, like this:
4218
4219 ** if (repeat_max == 0)
4220 ** {
4221 ** code = previous;
4222 ** goto END_REPEAT;
4223 ** }
4224
4225 However, that fails when a group is referenced as a subroutine from
4226 elsewhere in the pattern, so now we stick in OP_SKIPZERO in front of it
4227 so that it is skipped on execution. As we don't have a list of which
4228 groups are referenced, we cannot do this selectively.
4229
4230 If the maximum is 1 or unlimited, we just have to stick in the BRAZERO
4231 and do no more at this point. However, we do need to adjust any
4232 OP_RECURSE calls inside the group that refer to the group itself or any
4233 internal or forward referenced group, because the offset is from the
4234 start of the whole regex. Temporarily terminate the pattern while doing
4235 this. */
4236
4237 if (repeat_max <= 1) /* Covers 0, 1, and unlimited */
4238 {
4239 *code = OP_END;
4240 adjust_recurse(previous, 1, utf8, cd, save_hwm);
4241 memmove(previous+1, previous, len);
4242 code++;
4243 if (repeat_max == 0)
4244 {
4245 *previous++ = OP_SKIPZERO;
4246 goto END_REPEAT;
4247 }
4248 *previous++ = OP_BRAZERO + repeat_type;
4249 }
4250
4251 /* If the maximum is greater than 1 and limited, we have to replicate
4252 in a nested fashion, sticking OP_BRAZERO before each set of brackets.
4253 The first one has to be handled carefully because it's the original
4254 copy, which has to be moved up. The remainder can be handled by code
4255 that is common with the non-zero minimum case below. We have to
4256 adjust the value or repeat_max, since one less copy is required. Once
4257 again, we may have to adjust any OP_RECURSE calls inside the group. */
4258
4259 else
4260 {
4261 int offset;
4262 *code = OP_END;
4263 adjust_recurse(previous, 2 + LINK_SIZE, utf8, cd, save_hwm);
4264 memmove(previous + 2 + LINK_SIZE, previous, len);
4265 code += 2 + LINK_SIZE;
4266 *previous++ = OP_BRAZERO + repeat_type;
4267 *previous++ = OP_BRA;
4268
4269 /* We chain together the bracket offset fields that have to be
4270 filled in later when the ends of the brackets are reached. */
4271
4272 offset = (bralink == NULL)? 0 : previous - bralink;
4273 bralink = previous;
4274 PUTINC(previous, 0, offset);
4275 }
4276
4277 repeat_max--;
4278 }
4279
4280 /* If the minimum is greater than zero, replicate the group as many
4281 times as necessary, and adjust the maximum to the number of subsequent
4282 copies that we need. If we set a first char from the group, and didn't
4283 set a required char, copy the latter from the former. If there are any
4284 forward reference subroutine calls in the group, there will be entries on
4285 the workspace list; replicate these with an appropriate increment. */
4286
4287 else
4288 {
4289 if (repeat_min > 1)
4290 {
4291 /* In the pre-compile phase, we don't actually do the replication. We
4292 just adjust the length as if we had. Do some paranoid checks for
4293 potential integer overflow. The INT64_OR_DOUBLE type is a 64-bit
4294 integer type when available, otherwise double. */
4295
4296 if (lengthptr != NULL)
4297 {
4298 int delta = (repeat_min - 1)*length_prevgroup;
4299 if ((INT64_OR_DOUBLE)(repeat_min - 1)*
4300 (INT64_OR_DOUBLE)length_prevgroup >
4301 (INT64_OR_DOUBLE)INT_MAX ||
4302 OFLOW_MAX - *lengthptr < delta)
4303 {
4304 *errorcodeptr = ERR20;
4305 goto FAILED;
4306 }
4307 *lengthptr += delta;
4308 }
4309
4310 /* This is compiling for real */
4311
4312 else
4313 {
4314 if (groupsetfirstbyte && reqbyte < 0) reqbyte = firstbyte;
4315 for (i = 1; i < repeat_min; i++)
4316 {
4317 uschar *hc;
4318 uschar *this_hwm = cd->hwm;
4319 memcpy(code, previous, len);
4320 for (hc = save_hwm; hc < this_hwm; hc += LINK_SIZE)
4321 {
4322 PUT(cd->hwm, 0, GET(hc, 0) + len);
4323 cd->hwm += LINK_SIZE;
4324 }
4325 save_hwm = this_hwm;
4326 code += len;
4327 }
4328 }
4329 }
4330
4331 if (repeat_max > 0) repeat_max -= repeat_min;
4332 }
4333
4334 /* This code is common to both the zero and non-zero minimum cases. If
4335 the maximum is limited, it replicates the group in a nested fashion,
4336 remembering the bracket starts on a stack. In the case of a zero minimum,
4337 the first one was set up above. In all cases the repeat_max now specifies
4338 the number of additional copies needed. Again, we must remember to
4339 replicate entries on the forward reference list. */
4340
4341 if (repeat_max >= 0)
4342 {
4343 /* In the pre-compile phase, we don't actually do the replication. We
4344 just adjust the length as if we had. For each repetition we must add 1
4345 to the length for BRAZERO and for all but the last repetition we must
4346 add 2 + 2*LINKSIZE to allow for the nesting that occurs. Do some
4347 paranoid checks to avoid integer overflow. The INT64_OR_DOUBLE type is
4348 a 64-bit integer type when available, otherwise double. */
4349
4350 if (lengthptr != NULL && repeat_max > 0)
4351 {
4352 int delta = repeat_max * (length_prevgroup + 1 + 2 + 2*LINK_SIZE) -
4353 2 - 2*LINK_SIZE; /* Last one doesn't nest */
4354 if ((INT64_OR_DOUBLE)repeat_max *
4355 (INT64_OR_DOUBLE)(length_prevgroup + 1 + 2 + 2*LINK_SIZE)
4356 > (INT64_OR_DOUBLE)INT_MAX ||
4357 OFLOW_MAX - *lengthptr < delta)
4358 {
4359 *errorcodeptr = ERR20;
4360 goto FAILED;
4361 }
4362 *lengthptr += delta;
4363 }
4364
4365 /* This is compiling for real */
4366
4367 else for (i = repeat_max - 1; i >= 0; i--)
4368 {
4369 uschar *hc;
4370 uschar *this_hwm = cd->hwm;
4371
4372 *code++ = OP_BRAZERO + repeat_type;
4373
4374 /* All but the final copy start a new nesting, maintaining the
4375 chain of brackets outstanding. */
4376
4377 if (i != 0)
4378 {
4379 int offset;
4380 *code++ = OP_BRA;
4381 offset = (bralink == NULL)? 0 : code - bralink;
4382 bralink = code;
4383 PUTINC(code, 0, offset);
4384 }
4385
4386 memcpy(code, previous, len);
4387 for (hc = save_hwm; hc < this_hwm; hc += LINK_SIZE)
4388 {
4389 PUT(cd->hwm, 0, GET(hc, 0) + len + ((i != 0)? 2+LINK_SIZE : 1));
4390 cd->hwm += LINK_SIZE;
4391 }
4392 save_hwm = this_hwm;
4393 code += len;
4394 }
4395
4396 /* Now chain through the pending brackets, and fill in their length
4397 fields (which are holding the chain links pro tem). */
4398
4399 while (bralink != NULL)
4400 {
4401 int oldlinkoffset;
4402 int offset = code - bralink + 1;
4403 uschar *bra = code - offset;
4404 oldlinkoffset = GET(bra, 1);
4405 bralink = (oldlinkoffset == 0)? NULL : bralink - oldlinkoffset;
4406 *code++ = OP_KET;
4407 PUTINC(code, 0, offset);
4408 PUT(bra, 1, offset);
4409 }
4410 }
4411
4412 /* If the maximum is unlimited, set a repeater in the final copy. We
4413 can't just offset backwards from the current code point, because we
4414 don't know if there's been an options resetting after the ket. The
4415 correct offset was computed above.
4416
4417 Then, when we are doing the actual compile phase, check to see whether
4418 this group is a non-atomic one that could match an empty string. If so,
4419 convert the initial operator to the S form (e.g. OP_BRA -> OP_SBRA) so
4420 that runtime checking can be done. [This check is also applied to
4421 atomic groups at runtime, but in a different way.] */
4422
4423 else
4424 {
4425 uschar *ketcode = code - ketoffset;
4426 uschar *bracode = ketcode - GET(ketcode, 1);
4427 *ketcode = OP_KETRMAX + repeat_type;
4428 if (lengthptr == NULL && *bracode != OP_ONCE)
4429 {
4430 uschar *scode = bracode;
4431 do
4432 {
4433 if (could_be_empty_branch(scode, ketcode, utf8, cd))
4434 {
4435 *bracode += OP_SBRA - OP_BRA;
4436 break;
4437 }
4438 scode += GET(scode, 1);
4439 }
4440 while (*scode == OP_ALT);
4441 }
4442 }
4443 }
4444
4445 /* If previous is OP_FAIL, it was generated by an empty class [] in
4446 JavaScript mode. The other ways in which OP_FAIL can be generated, that is
4447 by (*FAIL) or (?!) set previous to NULL, which gives a "nothing to repeat"
4448 error above. We can just ignore the repeat in JS case. */
4449
4450 else if (*previous == OP_FAIL) goto END_REPEAT;
4451
4452 /* Else there's some kind of shambles */
4453
4454 else
4455 {
4456 *errorcodeptr = ERR11;
4457 goto FAILED;
4458 }
4459
4460 /* If the character following a repeat is '+', or if certain optimization
4461 tests above succeeded, possessive_quantifier is TRUE. For some of the
4462 simpler opcodes, there is an special alternative opcode for this. For
4463 anything else, we wrap the entire repeated item inside OP_ONCE brackets.
4464 The '+' notation is just syntactic sugar, taken from Sun's Java package,
4465 but the special opcodes can optimize it a bit. The repeated item starts at
4466 tempcode, not at previous, which might be the first part of a string whose
4467 (former) last char we repeated.
4468
4469 Possessifying an 'exact' quantifier has no effect, so we can ignore it. But
4470 an 'upto' may follow. We skip over an 'exact' item, and then test the
4471 length of what remains before proceeding. */
4472
4473 if (possessive_quantifier)
4474 {
4475 int len;
4476
4477 if (*tempcode == OP_TYPEEXACT)
4478 tempcode += _pcre_OP_lengths[*tempcode] +
4479 ((tempcode[3] == OP_PROP || tempcode[3] == OP_NOTPROP)? 2 : 0);
4480
4481 else if (*tempcode == OP_EXACT || *tempcode == OP_NOTEXACT)
4482 {
4483 tempcode += _pcre_OP_lengths[*tempcode];
4484 #ifdef SUPPORT_UTF8
4485 if (utf8 && tempcode[-1] >= 0xc0)
4486 tempcode += _pcre_utf8_table4[tempcode[-1] & 0x3f];
4487 #endif
4488 }
4489
4490 len = code - tempcode;
4491 if (len > 0) switch (*tempcode)
4492 {
4493 case OP_STAR: *tempcode = OP_POSSTAR; break;
4494 case OP_PLUS: *tempcode = OP_POSPLUS; break;
4495 case OP_QUERY: *tempcode = OP_POSQUERY; break;
4496 case OP_UPTO: *tempcode = OP_POSUPTO; break;
4497
4498 case OP_TYPESTAR: *tempcode = OP_TYPEPOSSTAR; break;
4499 case OP_TYPEPLUS: *tempcode = OP_TYPEPOSPLUS; break;
4500 case OP_TYPEQUERY: *tempcode = OP_TYPEPOSQUERY; break;
4501 case OP_TYPEUPTO: *tempcode = OP_TYPEPOSUPTO; break;
4502
4503 case OP_NOTSTAR: *tempcode = OP_NOTPOSSTAR; break;
4504 case OP_NOTPLUS: *tempcode = OP_NOTPOSPLUS; break;
4505 case OP_NOTQUERY: *tempcode = OP_NOTPOSQUERY; break;
4506 case OP_NOTUPTO: *tempcode = OP_NOTPOSUPTO; break;
4507
4508 /* Because we are moving code along, we must ensure that any
4509 pending recursive references are updated. */
4510
4511 default:
4512 *code = OP_END;
4513 adjust_recurse(tempcode, 1 + LINK_SIZE, utf8, cd, save_hwm);
4514 memmove(tempcode + 1+LINK_SIZE, tempcode, len);
4515 code += 1 + LINK_SIZE;
4516 len += 1 + LINK_SIZE;
4517 tempcode[0] = OP_ONCE;
4518 *code++ = OP_KET;
4519 PUTINC(code, 0, len);
4520 PUT(tempcode, 1, len);
4521 break;
4522 }
4523 }
4524
4525 /* In all case we no longer have a previous item. We also set the
4526 "follows varying string" flag for subsequently encountered reqbytes if
4527 it isn't already set and we have just passed a varying length item. */
4528
4529 END_REPEAT:
4530 previous = NULL;
4531 cd->req_varyopt |= reqvary;
4532 break;
4533
4534
4535 /* ===================================================================*/
4536 /* Start of nested parenthesized sub-expression, or comment or lookahead or
4537 lookbehind or option setting or condition or all the other extended
4538 parenthesis forms. */
4539
4540 case CHAR_LEFT_PARENTHESIS:
4541 newoptions = options;
4542 skipbytes = 0;
4543 bravalue = OP_CBRA;
4544 save_hwm = cd->hwm;
4545 reset_bracount = FALSE;
4546
4547 /* First deal with various "verbs" that can be introduced by '*'. */
4548
4549 if (*(++ptr) == CHAR_ASTERISK &&
4550 ((cd->ctypes[ptr[1]] & ctype_letter) != 0 || ptr[1] == ':'))
4551 {
4552 int i, namelen;
4553 int arglen = 0;
4554 const char *vn = verbnames;
4555 const uschar *name = ptr + 1;
4556 const uschar *arg = NULL;
4557 previous = NULL;
4558 while ((cd->ctypes[*++ptr] & ctype_letter) != 0) {};
4559 namelen = ptr - name;
4560
4561 if (*ptr == CHAR_COLON)
4562 {
4563 arg = ++ptr;
4564 while ((cd->ctypes[*ptr] & (ctype_letter|ctype_digit)) != 0
4565 || *ptr == '_') ptr++;
4566 arglen = ptr - arg;
4567 }
4568
4569 if (*ptr != CHAR_RIGHT_PARENTHESIS)
4570 {
4571 *errorcodeptr = ERR60;
4572 goto FAILED;
4573 }
4574
4575 /* Scan the table of verb names */
4576
4577 for (i = 0; i < verbcount; i++)
4578 {
4579 if (namelen == verbs[i].len &&
4580 strncmp((char *)name, vn, namelen) == 0)
4581 {
4582 /* Check for open captures before ACCEPT */
4583
4584 if (verbs[i].op == OP_ACCEPT)
4585 {
4586 open_capitem *oc;
4587 cd->had_accept = TRUE;
4588 for (oc = cd->open_caps; oc != NULL; oc = oc->next)
4589 {
4590 *code++ = OP_CLOSE;
4591 PUT2INC(code, 0, oc->number);
4592 }
4593 }
4594
4595 /* Handle the cases with/without an argument */
4596
4597 if (arglen == 0)
4598 {
4599 if (verbs[i].op < 0) /* Argument is mandatory */
4600 {
4601 *errorcodeptr = ERR66;
4602 goto FAILED;
4603 }
4604 *code++ = verbs[i].op;
4605 }
4606
4607 else
4608 {
4609 if (verbs[i].op_arg < 0) /* Argument is forbidden */
4610 {
4611 *errorcodeptr = ERR59;
4612 goto FAILED;
4613 }
4614 *code++ = verbs[i].op_arg;
4615 *code++ = arglen;
4616 memcpy(code, arg, arglen);
4617 code += arglen;
4618 *code++ = 0;
4619 }
4620
4621 break; /* Found verb, exit loop */
4622 }
4623
4624 vn += verbs[i].len + 1;
4625 }
4626
4627 if (i < verbcount) continue; /* Successfully handled a verb */
4628 *errorcodeptr = ERR60; /* Verb not recognized */
4629 goto FAILED;
4630 }
4631
4632 /* Deal with the extended parentheses; all are introduced by '?', and the
4633 appearance of any of them means that this is not a capturing group. */
4634
4635 else if (*ptr == CHAR_QUESTION_MARK)
4636 {
4637 int i, set, unset, namelen;
4638 int *optset;
4639 const uschar *name;
4640 uschar *slot;
4641
4642 switch (*(++ptr))
4643 {
4644 case CHAR_NUMBER_SIGN: /* Comment; skip to ket */
4645 ptr++;
4646 while (*ptr != 0 && *ptr != CHAR_RIGHT_PARENTHESIS) ptr++;
4647 if (*ptr == 0)
4648 {
4649 *errorcodeptr = ERR18;
4650 goto FAILED;
4651 }
4652 continue;
4653
4654
4655 /* ------------------------------------------------------------ */
4656 case CHAR_VERTICAL_LINE: /* Reset capture count for each branch */
4657 reset_bracount = TRUE;
4658 /* Fall through */
4659
4660 /* ------------------------------------------------------------ */
4661 case CHAR_COLON: /* Non-capturing bracket */
4662 bravalue = OP_BRA;
4663 ptr++;
4664 break;
4665
4666
4667 /* ------------------------------------------------------------ */
4668 case CHAR_LEFT_PARENTHESIS:
4669 bravalue = OP_COND; /* Conditional group */
4670
4671 /* A condition can be an assertion, a number (referring to a numbered
4672 group), a name (referring to a named group), or 'R', referring to
4673 recursion. R<digits> and R&name are also permitted for recursion tests.
4674
4675 There are several syntaxes for testing a named group: (?(name)) is used
4676 by Python; Perl 5.10 onwards uses (?(<name>) or (?('name')).
4677
4678 There are two unfortunate ambiguities, caused by history. (a) 'R' can
4679 be the recursive thing or the name 'R' (and similarly for 'R' followed
4680 by digits), and (b) a number could be a name that consists of digits.
4681 In both cases, we look for a name first; if not found, we try the other
4682 cases. */
4683
4684 /* For conditions that are assertions, check the syntax, and then exit
4685 the switch. This will take control down to where bracketed groups,
4686 including assertions, are processed. */
4687
4688 if (ptr[1] == CHAR_QUESTION_MARK && (ptr[2] == CHAR_EQUALS_SIGN ||
4689 ptr[2] == CHAR_EXCLAMATION_MARK || ptr[2] == CHAR_LESS_THAN_SIGN))
4690 break;
4691
4692 /* Most other conditions use OP_CREF (a couple change to OP_RREF
4693 below), and all need to skip 3 bytes at the start of the group. */
4694
4695 code[1+LINK_SIZE] = OP_CREF;
4696 skipbytes = 3;
4697 refsign = -1;
4698
4699 /* Check for a test for recursion in a named group. */
4700
4701 if (ptr[1] == CHAR_R && ptr[2] == CHAR_AMPERSAND)
4702 {
4703 terminator = -1;
4704 ptr += 2;
4705 code[1+LINK_SIZE] = OP_RREF; /* Change the type of test */
4706 }
4707
4708 /* Check for a test for a named group's having been set, using the Perl
4709 syntax (?(<name>) or (?('name') */
4710
4711 else if (ptr[1] == CHAR_LESS_THAN_SIGN)
4712 {
4713 terminator = CHAR_GREATER_THAN_SIGN;
4714 ptr++;
4715 }
4716 else if (ptr[1] == CHAR_APOSTROPHE)
4717 {
4718 terminator = CHAR_APOSTROPHE;
4719 ptr++;
4720 }
4721 else
4722 {
4723 terminator = 0;
4724 if (ptr[1] == CHAR_MINUS || ptr[1] == CHAR_PLUS) refsign = *(++ptr);
4725 }
4726
4727 /* We now expect to read a name; any thing else is an error */
4728
4729 if ((cd->ctypes[ptr[1]] & ctype_word) == 0)
4730 {
4731 ptr += 1; /* To get the right offset */
4732 *errorcodeptr = ERR28;
4733 goto FAILED;
4734 }
4735
4736 /* Read the name, but also get it as a number if it's all digits */
4737
4738 recno = 0;
4739 name = ++ptr;
4740 while ((cd->ctypes[*ptr] & ctype_word) != 0)
4741 {
4742 if (recno >= 0)
4743 recno = ((digitab[*ptr] & ctype_digit) != 0)?
4744 recno * 10 + *ptr - CHAR_0 : -1;
4745 ptr++;
4746 }
4747 namelen = ptr - name;
4748
4749 if ((terminator > 0 && *ptr++ != terminator) ||
4750 *ptr++ != CHAR_RIGHT_PARENTHESIS)
4751 {
4752 ptr--; /* Error offset */
4753 *errorcodeptr = ERR26;
4754 goto FAILED;
4755 }
4756
4757 /* Do no further checking in the pre-compile phase. */
4758
4759 if (lengthptr != NULL) break;
4760
4761 /* In the real compile we do the work of looking for the actual
4762 reference. If the string started with "+" or "-" we require the rest to
4763 be digits, in which case recno will be set. */
4764
4765 if (refsign > 0)
4766 {
4767 if (recno <= 0)
4768 {
4769 *errorcodeptr = ERR58;
4770 goto FAILED;
4771 }
4772 recno = (refsign == CHAR_MINUS)?
4773 cd->bracount - recno + 1 : recno +cd->bracount;
4774 if (recno <= 0 || recno > cd->final_bracount)
4775 {
4776 *errorcodeptr = ERR15;
4777 goto FAILED;
4778 }
4779 PUT2(code, 2+LINK_SIZE, recno);
4780 break;
4781 }
4782
4783 /* Otherwise (did not start with "+" or "-"), start by looking for the
4784 name. If we find a name, add one to the opcode to change OP_CREF or
4785 OP_RREF into OP_NCREF or OP_NRREF. These behave exactly the same,
4786 except they record that the reference was originally to a name. The
4787 information is used to check duplicate names. */
4788
4789 slot = cd->name_table;
4790 for (i = 0; i < cd->names_found; i++)
4791 {
4792 if (strncmp((char *)name, (char *)slot+2, namelen) == 0) break;
4793 slot += cd->name_entry_size;
4794 }
4795
4796 /* Found a previous named subpattern */
4797
4798 if (i < cd->names_found)
4799 {
4800 recno = GET2(slot, 0);
4801 PUT2(code, 2+LINK_SIZE, recno);
4802 code[1+LINK_SIZE]++;
4803 }
4804
4805 /* Search the pattern for a forward reference */
4806
4807 else if ((i = find_parens(cd, name, namelen,
4808 (options & PCRE_EXTENDED) != 0)) > 0)
4809 {
4810 PUT2(code, 2+LINK_SIZE, i);
4811 code[1+LINK_SIZE]++;
4812 }
4813
4814 /* If terminator == 0 it means that the name followed directly after
4815 the opening parenthesis [e.g. (?(abc)...] and in this case there are
4816 some further alternatives to try. For the cases where terminator != 0
4817 [things like (?(<name>... or (?('name')... or (?(R&name)... ] we have
4818 now checked all the possibilities, so give an error. */
4819
4820 else if (terminator != 0)
4821 {
4822 *errorcodeptr = ERR15;
4823 goto FAILED;
4824 }
4825
4826 /* Check for (?(R) for recursion. Allow digits after R to specify a
4827 specific group number. */
4828
4829 else if (*name == CHAR_R)
4830 {
4831 recno = 0;
4832 for (i = 1; i < namelen; i++)
4833 {
4834 if ((digitab[name[i]] & ctype_digit) == 0)
4835 {
4836 *errorcodeptr = ERR15;
4837 goto FAILED;
4838 }
4839 recno = recno * 10 + name[i] - CHAR_0;
4840 }
4841 if (recno == 0) recno = RREF_ANY;
4842 code[1+LINK_SIZE] = OP_RREF; /* Change test type */
4843 PUT2(code, 2+LINK_SIZE, recno);
4844 }
4845
4846 /* Similarly, check for the (?(DEFINE) "condition", which is always
4847 false. */
4848
4849 else if (namelen == 6 && strncmp((char *)name, STRING_DEFINE, 6) == 0)
4850 {
4851 code[1+LINK_SIZE] = OP_DEF;
4852 skipbytes = 1;
4853 }
4854
4855 /* Check for the "name" actually being a subpattern number. We are
4856 in the second pass here, so final_bracount is set. */
4857
4858 else if (recno > 0 && recno <= cd->final_bracount)
4859 {
4860 PUT2(code, 2+LINK_SIZE, recno);
4861 }
4862
4863 /* Either an unidentified subpattern, or a reference to (?(0) */
4864
4865 else
4866 {
4867 *errorcodeptr = (recno == 0)? ERR35: ERR15;
4868 goto FAILED;
4869 }
4870 break;
4871
4872
4873 /* ------------------------------------------------------------ */
4874 case CHAR_EQUALS_SIGN: /* Positive lookahead */
4875 bravalue = OP_ASSERT;
4876 ptr++;
4877 break;
4878
4879
4880 /* ------------------------------------------------------------ */
4881 case CHAR_EXCLAMATION_MARK: /* Negative lookahead */
4882 ptr++;
4883 if (*ptr == CHAR_RIGHT_PARENTHESIS) /* Optimize (?!) */
4884 {
4885 *code++ = OP_FAIL;
4886 previous = NULL;
4887 continue;
4888 }
4889 bravalue = OP_ASSERT_NOT;
4890 break;
4891
4892
4893 /* ------------------------------------------------------------ */
4894 case CHAR_LESS_THAN_SIGN: /* Lookbehind or named define */
4895 switch (ptr[1])
4896 {
4897 case CHAR_EQUALS_SIGN: /* Positive lookbehind */
4898 bravalue = OP_ASSERTBACK;
4899 ptr += 2;
4900 break;
4901
4902 case CHAR_EXCLAMATION_MARK: /* Negative lookbehind */
4903 bravalue = OP_ASSERTBACK_NOT;
4904 ptr += 2;
4905 break;
4906
4907 default: /* Could be name define, else bad */
4908 if ((cd->ctypes[ptr[1]] & ctype_word) != 0) goto DEFINE_NAME;
4909 ptr++; /* Correct offset for error */
4910 *errorcodeptr = ERR24;
4911 goto FAILED;
4912 }
4913 break;
4914
4915
4916 /* ------------------------------------------------------------ */
4917 case CHAR_GREATER_THAN_SIGN: /* One-time brackets */
4918 bravalue = OP_ONCE;
4919 ptr++;
4920 break;
4921
4922
4923 /* ------------------------------------------------------------ */
4924 case CHAR_C: /* Callout - may be followed by digits; */
4925 previous_callout = code; /* Save for later completion */
4926 after_manual_callout = 1; /* Skip one item before completing */
4927 *code++ = OP_CALLOUT;
4928 {
4929 int n = 0;
4930 while ((digitab[*(++ptr)] & ctype_digit) != 0)
4931 n = n * 10 + *ptr - CHAR_0;
4932 if (*ptr != CHAR_RIGHT_PARENTHESIS)
4933 {
4934 *errorcodeptr = ERR39;
4935 goto FAILED;
4936 }
4937 if (n > 255)
4938 {
4939 *errorcodeptr = ERR38;
4940 goto FAILED;
4941 }
4942 *code++ = n;
4943 PUT(code, 0, ptr - cd->start_pattern + 1); /* Pattern offset */
4944 PUT(code, LINK_SIZE, 0); /* Default length */
4945 code += 2 * LINK_SIZE;
4946 }
4947 previous = NULL;
4948 continue;
4949
4950
4951 /* ------------------------------------------------------------ */
4952 case CHAR_P: /* Python-style named subpattern handling */
4953 if (*(++ptr) == CHAR_EQUALS_SIGN ||
4954 *ptr == CHAR_GREATER_THAN_SIGN) /* Reference or recursion */
4955 {
4956 is_recurse = *ptr == CHAR_GREATER_THAN_SIGN;
4957 terminator = CHAR_RIGHT_PARENTHESIS;
4958 goto NAMED_REF_OR_RECURSE;
4959 }
4960 else if (*ptr != CHAR_LESS_THAN_SIGN) /* Test for Python-style defn */
4961 {
4962 *errorcodeptr = ERR41;
4963 goto FAILED;
4964 }
4965 /* Fall through to handle (?P< as (?< is handled */
4966
4967
4968 /* ------------------------------------------------------------ */
4969 DEFINE_NAME: /* Come here from (?< handling */
4970 case CHAR_APOSTROPHE:
4971 {
4972 terminator = (*ptr == CHAR_LESS_THAN_SIGN)?
4973 CHAR_GREATER_THAN_SIGN : CHAR_APOSTROPHE;
4974 name = ++ptr;
4975
4976 while ((cd->ctypes[*ptr] & ctype_word) != 0) ptr++;
4977 namelen = ptr - name;
4978
4979 /* In the pre-compile phase, just do a syntax check. */
4980
4981 if (lengthptr != NULL)
4982 {
4983 if (*ptr != terminator)
4984 {
4985 *errorcodeptr = ERR42;
4986 goto FAILED;
4987 }
4988 if (cd->names_found >= MAX_NAME_COUNT)
4989 {
4990 *errorcodeptr = ERR49;
4991 goto FAILED;
4992 }
4993 if (namelen + 3 > cd->name_entry_size)
4994 {
4995 cd->name_entry_size = namelen + 3;
4996 if (namelen > MAX_NAME_SIZE)
4997 {
4998 *errorcodeptr = ERR48;
4999 goto FAILED;
5000 }
5001 }
5002 }
5003
5004 /* In the real compile, create the entry in the table, maintaining
5005 alphabetical order. Duplicate names for different numbers are
5006 permitted only if PCRE_DUPNAMES is set. Duplicate names for the same
5007 number are always OK. (An existing number can be re-used if (?|
5008 appears in the pattern.) In either event, a duplicate name results in
5009 a duplicate entry in the table, even if the number is the same. This
5010 is because the number of names, and hence the table size, is computed
5011 in the pre-compile, and it affects various numbers and pointers which
5012 would all have to be modified, and the compiled code moved down, if
5013 duplicates with the same number were omitted from the table. This
5014 doesn't seem worth the hassle. However, *different* names for the
5015 same number are not permitted. */
5016
5017 else
5018 {
5019 BOOL dupname = FALSE;
5020 slot = cd->name_table;
5021
5022 for (i = 0; i < cd->names_found; i++)
5023 {
5024 int crc = memcmp(name, slot+2, namelen);
5025 if (crc == 0)
5026 {
5027 if (slot[2+namelen] == 0)
5028 {
5029 if (GET2(slot, 0) != cd->bracount + 1 &&
5030 (options & PCRE_DUPNAMES) == 0)
5031 {
5032 *errorcodeptr = ERR43;
5033 goto FAILED;
5034 }
5035 else dupname = TRUE;
5036 }
5037 else crc = -1; /* Current name is a substring */
5038 }
5039
5040 /* Make space in the table and break the loop for an earlier
5041 name. For a duplicate or later name, carry on. We do this for
5042 duplicates so that in the simple case (when ?(| is not used) they
5043 are in order of their numbers. */
5044
5045 if (crc < 0)
5046 {
5047 memmove(slot + cd->name_entry_size, slot,
5048 (cd->names_found - i) * cd->name_entry_size);
5049 break;
5050 }
5051
5052 /* Continue the loop for a later or duplicate name */
5053
5054 slot += cd->name_entry_size;
5055 }
5056
5057 /* For non-duplicate names, check for a duplicate number before
5058 adding the new name. */
5059
5060 if (!dupname)
5061 {
5062 uschar *cslot = cd->name_table;
5063 for (i = 0; i < cd->names_found; i++)
5064 {
5065 if (cslot != slot)
5066 {
5067 if (GET2(cslot, 0) == cd->bracount + 1)
5068 {
5069 *errorcodeptr = ERR65;
5070 goto FAILED;
5071 }
5072 }
5073 else i--;
5074 cslot += cd->name_entry_size;
5075 }
5076 }
5077
5078 PUT2(slot, 0, cd->bracount + 1);
5079 memcpy(slot + 2, name, namelen);
5080 slot[2+namelen] = 0;
5081 }
5082 }
5083
5084 /* In both pre-compile and compile, count the number of names we've
5085 encountered. */
5086
5087 cd->names_found++;
5088 ptr++; /* Move past > or ' */
5089 goto NUMBERED_GROUP;
5090
5091
5092 /* ------------------------------------------------------------ */
5093 case CHAR_AMPERSAND: /* Perl recursion/subroutine syntax */
5094 terminator = CHAR_RIGHT_PARENTHESIS;
5095 is_recurse = TRUE;
5096 /* Fall through */
5097
5098 /* We come here from the Python syntax above that handles both
5099 references (?P=name) and recursion (?P>name), as well as falling
5100 through from the Perl recursion syntax (?&name). We also come here from
5101 the Perl \k<name> or \k'name' back reference syntax and the \k{name}
5102 .NET syntax, and the Oniguruma \g<...> and \g'...' subroutine syntax. */
5103
5104 NAMED_REF_OR_RECURSE:
5105 name = ++ptr;
5106 while ((cd->ctypes[*ptr] & ctype_word) != 0) ptr++;
5107 namelen = ptr - name;
5108
5109 /* In the pre-compile phase, do a syntax check and set a dummy
5110 reference number. */
5111
5112 if (lengthptr != NULL)
5113 {
5114 if (namelen == 0)
5115 {
5116 *errorcodeptr = ERR62;
5117 goto FAILED;
5118 }
5119 if (*ptr != terminator)
5120 {
5121 *errorcodeptr = ERR42;
5122 goto FAILED;
5123 }
5124 if (namelen > MAX_NAME_SIZE)
5125 {
5126 *errorcodeptr = ERR48;
5127 goto FAILED;
5128 }
5129 recno = 0;
5130 }
5131
5132 /* In the real compile, seek the name in the table. We check the name
5133 first, and then check that we have reached the end of the name in the
5134 table. That way, if the name that is longer than any in the table,
5135 the comparison will fail without reading beyond the table entry. */
5136
5137 else
5138 {
5139 slot = cd->name_table;
5140 for (i = 0; i < cd->names_found; i++)
5141 {
5142 if (strncmp((char *)name, (char *)slot+2, namelen) == 0 &&
5143 slot[2+namelen] == 0)
5144 break;
5145 slot += cd->name_entry_size;
5146 }
5147
5148 if (i < cd->names_found) /* Back reference */
5149 {
5150 recno = GET2(slot, 0);
5151 }
5152 else if ((recno = /* Forward back reference */
5153 find_parens(cd, name, namelen,
5154 (options & PCRE_EXTENDED) != 0)) <= 0)
5155 {
5156 *errorcodeptr = ERR15;
5157 goto FAILED;
5158 }
5159 }
5160
5161 /* In both phases, we can now go to the code than handles numerical
5162 recursion or backreferences. */
5163
5164 if (is_recurse) goto HANDLE_RECURSION;
5165 else goto HANDLE_REFERENCE;
5166
5167
5168 /* ------------------------------------------------------------ */
5169 case CHAR_R: /* Recursion */
5170 ptr++; /* Same as (?0) */
5171 /* Fall through */
5172
5173
5174 /* ------------------------------------------------------------ */
5175 case CHAR_MINUS: case CHAR_PLUS: /* Recursion or subroutine */
5176 case CHAR_0: case CHAR_1: case CHAR_2: case CHAR_3: case CHAR_4:
5177 case CHAR_5: case CHAR_6: case CHAR_7: case CHAR_8: case CHAR_9:
5178 {
5179 const uschar *called;
5180 terminator = CHAR_RIGHT_PARENTHESIS;
5181
5182 /* Come here from the \g<...> and \g'...' code (Oniguruma
5183 compatibility). However, the syntax has been checked to ensure that
5184 the ... are a (signed) number, so that neither ERR63 nor ERR29 will
5185 be called on this path, nor with the jump to OTHER_CHAR_AFTER_QUERY
5186 ever be taken. */
5187
5188 HANDLE_NUMERICAL_RECURSION:
5189
5190 if ((refsign = *ptr) == CHAR_PLUS)
5191 {
5192 ptr++;
5193 if ((digitab[*ptr] & ctype_digit) == 0)
5194 {
5195 *errorcodeptr = ERR63;
5196 goto FAILED;
5197 }
5198 }
5199 else if (refsign == CHAR_MINUS)
5200 {
5201 if ((digitab[ptr[1]] & ctype_digit) == 0)
5202 goto OTHER_CHAR_AFTER_QUERY;
5203 ptr++;
5204 }
5205
5206 recno = 0;
5207 while((digitab[*ptr] & ctype_digit) != 0)
5208 recno = recno * 10 + *ptr++ - CHAR_0;
5209
5210 if (*ptr != terminator)
5211 {
5212 *errorcodeptr = ERR29;
5213 goto FAILED;
5214 }
5215
5216 if (refsign == CHAR_MINUS)
5217 {
5218 if (recno == 0)
5219 {
5220 *errorcodeptr = ERR58;
5221 goto FAILED;
5222 }
5223 recno = cd->bracount - recno + 1;
5224 if (recno <= 0)
5225 {
5226 *errorcodeptr = ERR15;
5227 goto FAILED;
5228 }
5229 }
5230 else if (refsign == CHAR_PLUS)
5231 {
5232 if (recno == 0)
5233 {
5234 *errorcodeptr = ERR58;
5235 goto FAILED;
5236 }
5237 recno += cd->bracount;
5238 }
5239
5240 /* Come here from code above that handles a named recursion */
5241
5242 HANDLE_RECURSION:
5243
5244 previous = code;
5245 called = cd->start_code;
5246
5247 /* When we are actually compiling, find the bracket that is being
5248 referenced. Temporarily end the regex in case it doesn't exist before
5249 this point. If we end up with a forward reference, first check that
5250 the bracket does occur later so we can give the error (and position)
5251 now. Then remember this forward reference in the workspace so it can
5252 be filled in at the end. */
5253
5254 if (lengthptr == NULL)
5255 {
5256 *code = OP_END;
5257 if (recno != 0)
5258 called = _pcre_find_bracket(cd->start_code, utf8, recno);
5259
5260 /* Forward reference */
5261
5262 if (called == NULL)
5263 {
5264 if (find_parens(cd, NULL, recno,
5265 (options & PCRE_EXTENDED) != 0) < 0)
5266 {
5267 *errorcodeptr = ERR15;
5268 goto FAILED;
5269 }
5270
5271 /* Fudge the value of "called" so that when it is inserted as an
5272 offset below, what it actually inserted is the reference number
5273 of the group. */
5274
5275 called = cd->start_code + recno;
5276 PUTINC(cd->hwm, 0, code + 2 + LINK_SIZE - cd->start_code);
5277 }
5278
5279 /* If not a forward reference, and the subpattern is still open,
5280 this is a recursive call. We check to see if this is a left
5281 recursion that could loop for ever, and diagnose that case. */
5282
5283 else if (GET(called, 1) == 0 &&
5284 could_be_empty(called, code, bcptr, utf8, cd))
5285 {
5286 *errorcodeptr = ERR40;
5287 goto FAILED;
5288 }
5289 }
5290
5291 /* Insert the recursion/subroutine item, automatically wrapped inside
5292 "once" brackets. Set up a "previous group" length so that a
5293 subsequent quantifier will work. */
5294
5295 *code = OP_ONCE;
5296 PUT(code, 1, 2 + 2*LINK_SIZE);
5297 code += 1 + LINK_SIZE;
5298
5299 *code = OP_RECURSE;
5300 PUT(code, 1, called - cd->start_code);
5301 code += 1 + LINK_SIZE;
5302
5303 *code = OP_KET;
5304 PUT(code, 1, 2 + 2*LINK_SIZE);
5305 code += 1 + LINK_SIZE;
5306
5307 length_prevgroup = 3 + 3*LINK_SIZE;
5308 }
5309
5310 /* Can't determine a first byte now */
5311
5312 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
5313 continue;
5314
5315
5316 /* ------------------------------------------------------------ */
5317 default: /* Other characters: check option setting */
5318 OTHER_CHAR_AFTER_QUERY:
5319 set = unset = 0;
5320 optset = &set;
5321
5322 while (*ptr != CHAR_RIGHT_PARENTHESIS && *ptr != CHAR_COLON)
5323 {
5324 switch (*ptr++)
5325 {
5326 case CHAR_MINUS: optset = &unset; break;
5327
5328 case CHAR_J: /* Record that it changed in the external options */
5329 *optset |= PCRE_DUPNAMES;
5330 cd->external_flags |= PCRE_JCHANGED;
5331 break;
5332
5333 case CHAR_i: *optset |= PCRE_CASELESS; break;
5334 case CHAR_m: *optset |= PCRE_MULTILINE; break;
5335 case CHAR_s: *optset |= PCRE_DOTALL; break;
5336 case CHAR_x: *optset |= PCRE_EXTENDED; break;
5337 case CHAR_U: *optset |= PCRE_UNGREEDY; break;
5338 case CHAR_X: *optset |= PCRE_EXTRA; break;
5339
5340 default: *errorcodeptr = ERR12;
5341 ptr--; /* Correct the offset */
5342 goto FAILED;
5343 }
5344 }
5345
5346 /* Set up the changed option bits, but don't change anything yet. */
5347
5348 newoptions = (options | set) & (~unset);
5349
5350 /* If the options ended with ')' this is not the start of a nested
5351 group with option changes, so the options change at this level. If this
5352 item is right at the start of the pattern, the options can be
5353 abstracted and made external in the pre-compile phase, and ignored in
5354 the compile phase. This can be helpful when matching -- for instance in
5355 caseless checking of required bytes.
5356
5357 If the code pointer is not (cd->start_code + 1 + LINK_SIZE), we are
5358 definitely *not* at the start of the pattern because something has been
5359 compiled. In the pre-compile phase, however, the code pointer can have
5360 that value after the start, because it gets reset as code is discarded
5361 during the pre-compile. However, this can happen only at top level - if
5362 we are within parentheses, the starting BRA will still be present. At
5363 any parenthesis level, the length value can be used to test if anything
5364 has been compiled at that level. Thus, a test for both these conditions
5365 is necessary to ensure we correctly detect the start of the pattern in
5366 both phases.
5367
5368 If we are not at the pattern start, compile code to change the ims
5369 options if this setting actually changes any of them, and reset the
5370 greedy defaults and the case value for firstbyte and reqbyte. */
5371
5372 if (*ptr == CHAR_RIGHT_PARENTHESIS)
5373 {
5374 if (code == cd->start_code + 1 + LINK_SIZE &&
5375 (lengthptr == NULL || *lengthptr == 2 + 2*LINK_SIZE))
5376 {
5377 cd->external_options = newoptions;
5378 }
5379 else
5380 {
5381 if ((options & PCRE_IMS) != (newoptions & PCRE_IMS))
5382 {
5383 *code++ = OP_OPT;
5384 *code++ = newoptions & PCRE_IMS;
5385 }
5386 greedy_default = ((newoptions & PCRE_UNGREEDY) != 0);
5387 greedy_non_default = greedy_default ^ 1;
5388 req_caseopt = ((newoptions & PCRE_CASELESS) != 0)? REQ_CASELESS : 0;
5389 }
5390
5391 /* Change options at this level, and pass them back for use
5392 in subsequent branches. When not at the start of the pattern, this
5393 information is also necessary so that a resetting item can be
5394 compiled at the end of a group (if we are in a group). */
5395
5396 *optionsptr = options = newoptions;
5397 previous = NULL; /* This item can't be repeated */
5398 continue; /* It is complete */
5399 }
5400
5401 /* If the options ended with ':' we are heading into a nested group
5402 with possible change of options. Such groups are non-capturing and are
5403 not assertions of any kind. All we need to do is skip over the ':';
5404 the newoptions value is handled below. */
5405
5406 bravalue = OP_BRA;
5407 ptr++;
5408 } /* End of switch for character following (? */
5409 } /* End of (? handling */
5410
5411 /* Opening parenthesis not followed by '*' or '?'. If PCRE_NO_AUTO_CAPTURE
5412 is set, all unadorned brackets become non-capturing and behave like (?:...)
5413 brackets. */
5414
5415 else if ((options & PCRE_NO_AUTO_CAPTURE) != 0)
5416 {
5417 bravalue = OP_BRA;
5418 }
5419
5420 /* Else we have a capturing group. */
5421
5422 else
5423 {
5424 NUMBERED_GROUP:
5425 cd->bracount += 1;
5426 PUT2(code, 1+LINK_SIZE, cd->bracount);
5427 skipbytes = 2;
5428 }
5429
5430 /* Process nested bracketed regex. Assertions may not be repeated, but
5431 other kinds can be. All their opcodes are >= OP_ONCE. We copy code into a
5432 non-register variable in order to be able to pass its address because some
5433 compilers complain otherwise. Pass in a new setting for the ims options if
5434 they have changed. */
5435
5436 previous = (bravalue >= OP_ONCE)? code : NULL;
5437 *code = bravalue;
5438 tempcode = code;
5439 tempreqvary = cd->req_varyopt; /* Save value before bracket */
5440 length_prevgroup = 0; /* Initialize for pre-compile phase */
5441
5442 if (!compile_regex(
5443 newoptions, /* The complete new option state */
5444 options & PCRE_IMS, /* The previous ims option state */
5445 &tempcode, /* Where to put code (updated) */
5446 &ptr, /* Input pointer (updated) */
5447 errorcodeptr, /* Where to put an error message */
5448 (bravalue == OP_ASSERTBACK ||
5449 bravalue == OP_ASSERTBACK_NOT), /* TRUE if back assert */
5450 reset_bracount, /* True if (?| group */
5451 skipbytes, /* Skip over bracket number */
5452 &subfirstbyte, /* For possible first char */
5453 &subreqbyte, /* For possible last char */
5454 bcptr, /* Current branch chain */
5455 cd, /* Tables block */
5456 (lengthptr == NULL)? NULL : /* Actual compile phase */
5457 &length_prevgroup /* Pre-compile phase */
5458 ))
5459 goto FAILED;
5460
5461 /* At the end of compiling, code is still pointing to the start of the
5462 group, while tempcode has been updated to point past the end of the group
5463 and any option resetting that may follow it. The pattern pointer (ptr)
5464 is on the bracket. */
5465
5466 /* If this is a conditional bracket, check that there are no more than
5467 two branches in the group, or just one if it's a DEFINE group. We do this
5468 in the real compile phase, not in the pre-pass, where the whole group may
5469 not be available. */
5470
5471 if (bravalue == OP_COND && lengthptr == NULL)
5472 {
5473 uschar *tc = code;
5474 int condcount = 0;
5475
5476 do {
5477 condcount++;
5478 tc += GET(tc,1);
5479 }
5480 while (*tc != OP_KET);
5481
5482 /* A DEFINE group is never obeyed inline (the "condition" is always
5483 false). It must have only one branch. */
5484
5485 if (code[LINK_SIZE+1] == OP_DEF)
5486 {
5487 if (condcount > 1)
5488 {
5489 *errorcodeptr = ERR54;
5490 goto FAILED;
5491 }
5492 bravalue = OP_DEF; /* Just a flag to suppress char handling below */
5493 }
5494
5495 /* A "normal" conditional group. If there is just one branch, we must not
5496 make use of its firstbyte or reqbyte, because this is equivalent to an
5497 empty second branch. */
5498
5499 else
5500 {
5501 if (condcount > 2)
5502 {
5503 *errorcodeptr = ERR27;
5504 goto FAILED;
5505 }
5506 if (condcount == 1) subfirstbyte = subreqbyte = REQ_NONE;
5507 }
5508 }
5509
5510 /* Error if hit end of pattern */
5511
5512 if (*ptr != CHAR_RIGHT_PARENTHESIS)
5513 {
5514 *errorcodeptr = ERR14;
5515 goto FAILED;
5516 }
5517
5518 /* In the pre-compile phase, update the length by the length of the group,
5519 less the brackets at either end. Then reduce the compiled code to just a
5520 set of non-capturing brackets so that it doesn't use much memory if it is
5521 duplicated by a quantifier.*/
5522
5523 if (lengthptr != NULL)
5524 {
5525 if (OFLOW_MAX - *lengthptr < length_prevgroup - 2 - 2*LINK_SIZE)
5526 {
5527 *errorcodeptr = ERR20;
5528 goto FAILED;
5529 }
5530 *lengthptr += length_prevgroup - 2 - 2*LINK_SIZE;
5531 *code++ = OP_BRA;
5532 PUTINC(code, 0, 1 + LINK_SIZE);
5533 *code++ = OP_KET;
5534 PUTINC(code, 0, 1 + LINK_SIZE);
5535 break; /* No need to waste time with special character handling */
5536 }
5537
5538 /* Otherwise update the main code pointer to the end of the group. */
5539
5540 code = tempcode;
5541
5542 /* For a DEFINE group, required and first character settings are not
5543 relevant. */
5544
5545 if (bravalue == OP_DEF) break;
5546
5547 /* Handle updating of the required and first characters for other types of
5548 group. Update for normal brackets of all kinds, and conditions with two
5549 branches (see code above). If the bracket is followed by a quantifier with
5550 zero repeat, we have to back off. Hence the definition of zeroreqbyte and
5551 zerofirstbyte outside the main loop so that they can be accessed for the
5552 back off. */
5553
5554 zeroreqbyte = reqbyte;
5555 zerofirstbyte = firstbyte;
5556 groupsetfirstbyte = FALSE;
5557
5558 if (bravalue >= OP_ONCE)
5559 {
5560 /* If we have not yet set a firstbyte in this branch, take it from the
5561 subpattern, remembering that it was set here so that a repeat of more
5562 than one can replicate it as reqbyte if necessary. If the subpattern has
5563 no firstbyte, set "none" for the whole branch. In both cases, a zero
5564 repeat forces firstbyte to "none". */
5565
5566 if (firstbyte == REQ_UNSET)
5567 {
5568 if (subfirstbyte >= 0)
5569 {
5570 firstbyte = subfirstbyte;
5571 groupsetfirstbyte = TRUE;
5572 }
5573 else firstbyte = REQ_NONE;
5574 zerofirstbyte = REQ_NONE;
5575 }
5576
5577 /* If firstbyte was previously set, convert the subpattern's firstbyte
5578 into reqbyte if there wasn't one, using the vary flag that was in
5579 existence beforehand. */
5580
5581 else if (subfirstbyte >= 0 && subreqbyte < 0)
5582 subreqbyte = subfirstbyte | tempreqvary;
5583
5584 /* If the subpattern set a required byte (or set a first byte that isn't
5585 really the first byte - see above), set it. */
5586
5587 if (subreqbyte >= 0) reqbyte = subreqbyte;
5588 }
5589
5590 /* For a forward assertion, we take the reqbyte, if set. This can be
5591 helpful if the pattern that follows the assertion doesn't set a different
5592 char. For example, it's useful for /(?=abcde).+/. We can't set firstbyte
5593 for an assertion, however because it leads to incorrect effect for patterns
5594 such as /(?=a)a.+/ when the "real" "a" would then become a reqbyte instead
5595 of a firstbyte. This is overcome by a scan at the end if there's no
5596 firstbyte, looking for an asserted first char. */
5597
5598 else if (bravalue == OP_ASSERT && subreqbyte >= 0) reqbyte = subreqbyte;
5599 break; /* End of processing '(' */
5600
5601
5602 /* ===================================================================*/
5603 /* Handle metasequences introduced by \. For ones like \d, the ESC_ values
5604 are arranged to be the negation of the corresponding OP_values. For the
5605 back references, the values are ESC_REF plus the reference number. Only
5606 back references and those types that consume a character may be repeated.
5607 We can test for values between ESC_b and ESC_Z for the latter; this may
5608 have to change if any new ones are ever created. */
5609
5610 case CHAR_BACKSLASH:
5611 tempptr = ptr;
5612 c = check_escape(&ptr, errorcodeptr, cd->bracount, options, FALSE);
5613 if (*errorcodeptr != 0) goto FAILED;
5614
5615 if (c < 0)
5616 {
5617 if (-c == ESC_Q) /* Handle start of quoted string */
5618 {
5619 if (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
5620 ptr += 2; /* avoid empty string */
5621 else inescq = TRUE;
5622 continue;
5623 }
5624
5625 if (-c == ESC_E) continue; /* Perl ignores an orphan \E */
5626
5627 /* For metasequences that actually match a character, we disable the
5628 setting of a first character if it hasn't already been set. */
5629
5630 if (firstbyte == REQ_UNSET && -c > ESC_b && -c < ESC_Z)
5631 firstbyte = REQ_NONE;
5632
5633 /* Set values to reset to if this is followed by a zero repeat. */
5634
5635 zerofirstbyte = firstbyte;
5636 zeroreqbyte = reqbyte;
5637
5638 /* \g<name> or \g'name' is a subroutine call by name and \g<n> or \g'n'
5639 is a subroutine call by number (Oniguruma syntax). In fact, the value
5640 -ESC_g is returned only for these cases. So we don't need to check for <
5641 or ' if the value is -ESC_g. For the Perl syntax \g{n} the value is
5642 -ESC_REF+n, and for the Perl syntax \g{name} the result is -ESC_k (as
5643 that is a synonym for a named back reference). */
5644
5645 if (-c == ESC_g)
5646 {
5647 const uschar *p;
5648 save_hwm = cd->hwm; /* Normally this is set when '(' is read */
5649 terminator = (*(++ptr) == CHAR_LESS_THAN_SIGN)?
5650 CHAR_GREATER_THAN_SIGN : CHAR_APOSTROPHE;
5651
5652 /* These two statements stop the compiler for warning about possibly
5653 unset variables caused by the jump to HANDLE_NUMERICAL_RECURSION. In
5654 fact, because we actually check for a number below, the paths that
5655 would actually be in error are never taken. */
5656
5657 skipbytes = 0;
5658 reset_bracount = FALSE;
5659
5660 /* Test for a name */
5661
5662 if (ptr[1] != CHAR_PLUS && ptr[1] != CHAR_MINUS)
5663 {
5664 BOOL isnumber = TRUE;
5665 for (p = ptr + 1; *p != 0 && *p != terminator; p++)
5666 {
5667 if ((cd->ctypes[*p] & ctype_digit) == 0) isnumber = FALSE;
5668 if ((cd->ctypes[*p] & ctype_word) == 0) break;
5669 }
5670 if (*p != terminator)
5671 {
5672 *errorcodeptr = ERR57;
5673 break;
5674 }
5675 if (isnumber)
5676 {
5677 ptr++;
5678 goto HANDLE_NUMERICAL_RECURSION;
5679 }
5680 is_recurse = TRUE;
5681 goto NAMED_REF_OR_RECURSE;
5682 }
5683
5684 /* Test a signed number in angle brackets or quotes. */
5685
5686 p = ptr + 2;
5687 while ((digitab[*p] & ctype_digit) != 0) p++;
5688 if (*p != terminator)
5689 {
5690 *errorcodeptr = ERR57;
5691 break;
5692 }
5693 ptr++;
5694 goto HANDLE_NUMERICAL_RECURSION;
5695 }
5696
5697 /* \k<name> or \k'name' is a back reference by name (Perl syntax).
5698 We also support \k{name} (.NET syntax) */
5699
5700 if (-c == ESC_k && (ptr[1] == CHAR_LESS_THAN_SIGN ||
5701 ptr[1] == CHAR_APOSTROPHE || ptr[1] == CHAR_LEFT_CURLY_BRACKET))
5702 {
5703 is_recurse = FALSE;
5704 terminator = (*(++ptr) == CHAR_LESS_THAN_SIGN)?
5705 CHAR_GREATER_THAN_SIGN : (*ptr == CHAR_APOSTROPHE)?
5706 CHAR_APOSTROPHE : CHAR_RIGHT_CURLY_BRACKET;
5707 goto NAMED_REF_OR_RECURSE;
5708 }
5709
5710 /* Back references are handled specially; must disable firstbyte if
5711 not set to cope with cases like (?=(\w+))\1: which would otherwise set
5712 ':' later. */
5713
5714 if (-c >= ESC_REF)
5715 {
5716 open_capitem *oc;
5717 recno = -c - ESC_REF;
5718
5719 HANDLE_REFERENCE: /* Come here from named backref handling */
5720 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
5721 previous = code;
5722 *code++ = OP_REF;
5723 PUT2INC(code, 0, recno);
5724 cd->backref_map |= (recno < 32)? (1 << recno) : 1;
5725 if (recno > cd->top_backref) cd->top_backref = recno;
5726
5727 /* Check to see if this back reference is recursive, that it, it
5728 is inside the group that it references. A flag is set so that the
5729 group can be made atomic. */
5730
5731 for (oc = cd->open_caps; oc != NULL; oc = oc->next)
5732 {
5733 if (oc->number == recno)
5734 {
5735 oc->flag = TRUE;
5736 break;
5737 }
5738 }
5739 }
5740
5741 /* So are Unicode property matches, if supported. */
5742
5743 #ifdef SUPPORT_UCP
5744 else if (-c == ESC_P || -c == ESC_p)
5745 {
5746 BOOL negated;
5747 int pdata;
5748 int ptype = get_ucp(&ptr, &negated, &pdata, errorcodeptr);
5749 if (ptype < 0) goto FAILED;
5750 previous = code;
5751 *code++ = ((-c == ESC_p) != negated)? OP_PROP : OP_NOTPROP;
5752 *code++ = ptype;
5753 *code++ = pdata;
5754 }
5755 #else
5756
5757 /* If Unicode properties are not supported, \X, \P, and \p are not
5758 allowed. */
5759
5760 else if (-c == ESC_X || -c == ESC_P || -c == ESC_p)
5761 {
5762 *errorcodeptr = ERR45;
5763 goto FAILED;
5764 }
5765 #endif
5766
5767 /* For the rest (including \X when Unicode properties are supported), we
5768 can obtain the OP value by negating the escape value. */
5769
5770 else
5771 {
5772 previous = (-c > ESC_b && -c < ESC_Z)? code : NULL;
5773 *code++ = -c;
5774 }
5775 continue;
5776 }
5777
5778 /* We have a data character whose value is in c. In UTF-8 mode it may have
5779 a value > 127. We set its representation in the length/buffer, and then
5780 handle it as a data character. */
5781
5782 #ifdef SUPPORT_UTF8
5783 if (utf8 && c > 127)
5784 mclength = _pcre_ord2utf8(c, mcbuffer);
5785 else
5786 #endif
5787
5788 {
5789 mcbuffer[0] = c;
5790 mclength = 1;
5791 }
5792 goto ONE_CHAR;
5793
5794
5795 /* ===================================================================*/
5796 /* Handle a literal character. It is guaranteed not to be whitespace or #
5797 when the extended flag is set. If we are in UTF-8 mode, it may be a
5798 multi-byte literal character. */
5799
5800 default:
5801 NORMAL_CHAR:
5802 mclength = 1;
5803 mcbuffer[0] = c;
5804
5805 #ifdef SUPPORT_UTF8
5806 if (utf8 && c >= 0xc0)
5807 {
5808 while ((ptr[1] & 0xc0) == 0x80)
5809 mcbuffer[mclength++] = *(++ptr);
5810 }
5811 #endif
5812
5813 /* At this point we have the character's bytes in mcbuffer, and the length
5814 in mclength. When not in UTF-8 mode, the length is always 1. */
5815
5816 ONE_CHAR:
5817 previous = code;
5818 *code++ = ((options & PCRE_CASELESS) != 0)? OP_CHARNC : OP_CHAR;
5819 for (c = 0; c < mclength; c++) *code++ = mcbuffer[c];
5820
5821 /* Remember if \r or \n were seen */
5822
5823 if (mcbuffer[0] == CHAR_CR || mcbuffer[0] == CHAR_NL)
5824 cd->external_flags |= PCRE_HASCRORLF;
5825
5826 /* Set the first and required bytes appropriately. If no previous first
5827 byte, set it from this character, but revert to none on a zero repeat.
5828 Otherwise, leave the firstbyte value alone, and don't change it on a zero
5829 repeat. */
5830
5831 if (firstbyte == REQ_UNSET)
5832 {
5833 zerofirstbyte = REQ_NONE;
5834 zeroreqbyte = reqbyte;
5835
5836 /* If the character is more than one byte long, we can set firstbyte
5837 only if it is not to be matched caselessly. */
5838
5839 if (mclength == 1 || req_caseopt == 0)
5840 {
5841 firstbyte = mcbuffer[0] | req_caseopt;
5842 if (mclength != 1) reqbyte = code[-1] | cd->req_varyopt;
5843 }
5844 else firstbyte = reqbyte = REQ_NONE;
5845 }
5846
5847 /* firstbyte was previously set; we can set reqbyte only the length is
5848 1 or the matching is caseful. */
5849
5850 else
5851 {
5852 zerofirstbyte = firstbyte;
5853 zeroreqbyte = reqbyte;
5854 if (mclength == 1 || req_caseopt == 0)
5855 reqbyte = code[-1] | req_caseopt | cd->req_varyopt;
5856 }
5857
5858 break; /* End of literal character handling */
5859 }
5860 } /* end of big loop */
5861
5862
5863 /* Control never reaches here by falling through, only by a goto for all the
5864 error states. Pass back the position in the pattern so that it can be displayed
5865 to the user for diagnosing the error. */
5866
5867 FAILED:
5868 *ptrptr = ptr;
5869 return FALSE;
5870 }
5871
5872
5873
5874
5875 /*************************************************
5876 * Compile sequence of alternatives *
5877 *************************************************/
5878
5879 /* On entry, ptr is pointing past the bracket character, but on return it
5880 points to the closing bracket, or vertical bar, or end of string. The code
5881 variable is pointing at the byte into which the BRA operator has been stored.
5882 If the ims options are changed at the start (for a (?ims: group) or during any
5883 branch, we need to insert an OP_OPT item at the start of every following branch
5884 to ensure they get set correctly at run time, and also pass the new options
5885 into every subsequent branch compile.
5886
5887 This function is used during the pre-compile phase when we are trying to find
5888 out the amount of memory needed, as well as during the real compile phase. The
5889 value of lengthptr distinguishes the two phases.
5890
5891 Arguments:
5892 options option bits, including any changes for this subpattern
5893 oldims previous settings of ims option bits
5894 codeptr -> the address of the current code pointer
5895 ptrptr -> the address of the current pattern pointer
5896 errorcodeptr -> pointer to error code variable
5897 lookbehind TRUE if this is a lookbehind assertion
5898 reset_bracount TRUE to reset the count for each branch
5899 skipbytes skip this many bytes at start (for brackets and OP_COND)
5900 firstbyteptr place to put the first required character, or a negative number
5901 reqbyteptr place to put the last required character, or a negative number
5902 bcptr pointer to the chain of currently open branches
5903 cd points to the data block with tables pointers etc.
5904 lengthptr NULL during the real compile phase
5905 points to length accumulator during pre-compile phase
5906
5907 Returns: TRUE on success
5908 */
5909
5910 static BOOL
5911 compile_regex(int options, int oldims, uschar **codeptr, const uschar **ptrptr,
5912 int *errorcodeptr, BOOL lookbehind, BOOL reset_bracount, int skipbytes,
5913 int *firstbyteptr, int *reqbyteptr, branch_chain *bcptr, compile_data *cd,
5914 int *lengthptr)
5915 {
5916 const uschar *ptr = *ptrptr;
5917 uschar *code = *codeptr;
5918 uschar *last_branch = code;
5919 uschar *start_bracket = code;
5920 uschar *reverse_count = NULL;
5921 open_capitem capitem;
5922 int capnumber = 0;
5923 int firstbyte, reqbyte;
5924 int branchfirstbyte, branchreqbyte;
5925 int length;
5926 int orig_bracount;
5927 int max_bracount;
5928 int old_external_options = cd->external_options;
5929 branch_chain bc;
5930
5931 bc.outer = bcptr;
5932 bc.current_branch = code;
5933
5934 firstbyte = reqbyte = REQ_UNSET;
5935
5936 /* Accumulate the length for use in the pre-compile phase. Start with the
5937 length of the BRA and KET and any extra bytes that are required at the
5938 beginning. We accumulate in a local variable to save frequent testing of
5939 lenthptr for NULL. We cannot do this by looking at the value of code at the
5940 start and end of each alternative, because compiled items are discarded during
5941 the pre-compile phase so that the work space is not exceeded. */
5942
5943 length = 2 + 2*LINK_SIZE + skipbytes;
5944
5945 /* WARNING: If the above line is changed for any reason, you must also change
5946 the code that abstracts option settings at the start of the pattern and makes
5947 them global. It tests the value of length for (2 + 2*LINK_SIZE) in the
5948 pre-compile phase to find out whether anything has yet been compiled or not. */
5949
5950 /* If this is a capturing subpattern, add to the chain of open capturing items
5951 so that we can detect them if (*ACCEPT) is encountered. This is also used to
5952 detect groups that contain recursive back references to themselves. */
5953
5954 if (*code == OP_CBRA)
5955 {
5956 capnumber = GET2(code, 1 + LINK_SIZE);
5957 capitem.number = capnumber;
5958 capitem.next = cd->open_caps;
5959 capitem.flag = FALSE;
5960 cd->open_caps = &capitem;
5961 }
5962
5963 /* Offset is set zero to mark that this bracket is still open */
5964
5965 PUT(code, 1, 0);
5966 code += 1 + LINK_SIZE + skipbytes;
5967
5968 /* Loop for each alternative branch */
5969
5970 orig_bracount = max_bracount = cd->bracount;
5971 for (;;)
5972 {
5973 /* For a (?| group, reset the capturing bracket count so that each branch
5974 uses the same numbers. */
5975
5976 if (reset_bracount) cd->bracount = orig_bracount;
5977
5978 /* Handle a change of ims options at the start of the branch */
5979
5980 if ((options & PCRE_IMS) != oldims)
5981 {
5982 *code++ = OP_OPT;
5983 *code++ = options & PCRE_IMS;
5984 length += 2;
5985 }
5986
5987 /* Set up dummy OP_REVERSE if lookbehind assertion */
5988
5989 if (lookbehind)
5990 {
5991 *code++ = OP_REVERSE;
5992 reverse_count = code;
5993 PUTINC(code, 0, 0);
5994 length += 1 + LINK_SIZE;
5995 }
5996
5997 /* Now compile the branch; in the pre-compile phase its length gets added
5998 into the length. */
5999
6000 if (!compile_branch(&options, &code, &ptr, errorcodeptr, &branchfirstbyte,
6001 &branchreqbyte, &bc, cd, (lengthptr == NULL)? NULL : &length))
6002 {
6003 *ptrptr = ptr;
6004 return FALSE;
6005 }
6006
6007 /* If the external options have changed during this branch, it means that we
6008 are at the top level, and a leading option setting has been encountered. We
6009 need to re-set the original option values to take account of this so that,
6010 during the pre-compile phase, we know to allow for a re-set at the start of
6011 subsequent branches. */
6012
6013 if (old_external_options != cd->external_options)
6014 oldims = cd->external_options & PCRE_IMS;
6015
6016 /* Keep the highest bracket count in case (?| was used and some branch
6017 has fewer than the rest. */
6018
6019 if (cd->bracount > max_bracount) max_bracount = cd->bracount;
6020
6021 /* In the real compile phase, there is some post-processing to be done. */
6022
6023 if (lengthptr == NULL)
6024 {
6025 /* If this is the first branch, the firstbyte and reqbyte values for the
6026 branch become the values for the regex. */
6027
6028 if (*last_branch != OP_ALT)
6029 {
6030 firstbyte = branchfirstbyte;
6031 reqbyte = branchreqbyte;
6032 }
6033
6034 /* If this is not the first branch, the first char and reqbyte have to
6035 match the values from all the previous branches, except that if the
6036 previous value for reqbyte didn't have REQ_VARY set, it can still match,
6037 and we set REQ_VARY for the regex. */
6038
6039 else
6040 {
6041 /* If we previously had a firstbyte, but it doesn't match the new branch,
6042 we have to abandon the firstbyte for the regex, but if there was
6043 previously no reqbyte, it takes on the value of the old firstbyte. */
6044
6045 if (firstbyte >= 0 && firstbyte != branchfirstbyte)
6046 {
6047 if (reqbyte < 0) reqbyte = firstbyte;
6048 firstbyte = REQ_NONE;
6049 }
6050
6051 /* If we (now or from before) have no firstbyte, a firstbyte from the
6052 branch becomes a reqbyte if there isn't a branch reqbyte. */
6053
6054 if (firstbyte < 0 && branchfirstbyte >= 0 && branchreqbyte < 0)
6055 branchreqbyte = branchfirstbyte;
6056
6057 /* Now ensure that the reqbytes match */
6058
6059 if ((reqbyte & ~REQ_VARY) != (branchreqbyte & ~REQ_VARY))
6060 reqbyte = REQ_NONE;
6061 else reqbyte |= branchreqbyte; /* To "or" REQ_VARY */
6062 }
6063
6064 /* If lookbehind, check that this branch matches a fixed-length string, and
6065 put the length into the OP_REVERSE item. Temporarily mark the end of the
6066 branch with OP_END. If the branch contains OP_RECURSE, the result is -3
6067 because there may be forward references that we can't check here. Set a
6068 flag to cause another lookbehind check at the end. Why not do it all at the
6069 end? Because common, erroneous checks are picked up here and the offset of
6070 the problem can be shown. */
6071
6072 if (lookbehind)
6073 {
6074 int fixed_length;
6075 *code = OP_END;
6076 fixed_length = find_fixedlength(last_branch, options, FALSE, cd);
6077 DPRINTF(("fixed length = %d\n", fixed_length));
6078 if (fixed_length == -3)
6079 {
6080 cd->check_lookbehind = TRUE;
6081 }
6082 else if (fixed_length < 0)
6083 {
6084 *errorcodeptr = (fixed_length == -2)? ERR36 : ERR25;
6085 *ptrptr = ptr;
6086 return FALSE;
6087 }
6088 else { PUT(reverse_count, 0, fixed_length); }
6089 }
6090 }
6091
6092 /* Reached end of expression, either ')' or end of pattern. In the real
6093 compile phase, go back through the alternative branches and reverse the chain
6094 of offsets, with the field in the BRA item now becoming an offset to the
6095 first alternative. If there are no alternatives, it points to the end of the
6096 group. The length in the terminating ket is always the length of the whole
6097 bracketed item. If any of the ims options were changed inside the group,
6098 compile a resetting op-code following, except at the very end of the pattern.
6099 Return leaving the pointer at the terminating char. */
6100
6101 if (*ptr != CHAR_VERTICAL_LINE)
6102 {
6103 if (lengthptr == NULL)
6104 {
6105 int branch_length = code - last_branch;
6106 do
6107 {
6108 int prev_length = GET(last_branch, 1);
6109 PUT(last_branch, 1, branch_length);
6110 branch_length = prev_length;
6111 last_branch -= branch_length;
6112 }
6113 while (branch_length > 0);
6114 }
6115
6116 /* Fill in the ket */
6117
6118 *code = OP_KET;
6119 PUT(code, 1, code - start_bracket);
6120 code += 1 + LINK_SIZE;
6121
6122 /* If it was a capturing subpattern, check to see if it contained any
6123 recursive back references. If so, we must wrap it in atomic brackets.
6124 In any event, remove the block from the chain. */
6125
6126 if (capnumber > 0)
6127 {
6128 if (cd->open_caps->flag)
6129 {
6130 memmove(start_bracket + 1 + LINK_SIZE, start_bracket,
6131 code - start_bracket);
6132 *start_bracket = OP_ONCE;
6133 code += 1 + LINK_SIZE;
6134 PUT(start_bracket, 1, code - start_bracket);
6135 *code = OP_KET;
6136 PUT(code, 1, code - start_bracket);
6137 code += 1 + LINK_SIZE;
6138 length += 2 + 2*LINK_SIZE;
6139 }
6140 cd->open_caps = cd->open_caps->next;
6141 }
6142
6143 /* Reset options if needed. */
6144
6145 if ((options & PCRE_IMS) != oldims && *ptr == CHAR_RIGHT_PARENTHESIS)
6146 {
6147 *code++ = OP_OPT;
6148 *code++ = oldims;
6149 length += 2;
6150 }
6151
6152 /* Retain the highest bracket number, in case resetting was used. */
6153
6154 cd->bracount = max_bracount;
6155
6156 /* Set values to pass back */
6157
6158 *codeptr = code;
6159 *ptrptr = ptr;
6160 *firstbyteptr = firstbyte;
6161 *reqbyteptr = reqbyte;
6162 if (lengthptr != NULL)
6163 {
6164 if (OFLOW_MAX - *lengthptr < length)
6165 {
6166 *errorcodeptr = ERR20;
6167 return FALSE;
6168 }
6169 *lengthptr += length;
6170 }
6171 return TRUE;
6172 }
6173
6174 /* Another branch follows. In the pre-compile phase, we can move the code
6175 pointer back to where it was for the start of the first branch. (That is,
6176 pretend that each branch is the only one.)
6177
6178 In the real compile phase, insert an ALT node. Its length field points back
6179 to the previous branch while the bracket remains open. At the end the chain
6180 is reversed. It's done like this so that the start of the bracket has a
6181 zero offset until it is closed, making it possible to detect recursion. */
6182
6183 if (lengthptr != NULL)
6184 {
6185 code = *codeptr + 1 + LINK_SIZE + skipbytes;
6186 length += 1 + LINK_SIZE;
6187 }
6188 else
6189 {
6190 *code = OP_ALT;
6191 PUT(code, 1, code - last_branch);
6192 bc.current_branch = last_branch = code;
6193 code += 1 + LINK_SIZE;
6194 }
6195
6196 ptr++;
6197 }
6198 /* Control never reaches here */
6199 }
6200
6201
6202
6203
6204 /*************************************************
6205 * Check for anchored expression *
6206 *************************************************/
6207
6208 /* Try to find out if this is an anchored regular expression. Consider each
6209 alternative branch. If they all start with OP_SOD or OP_CIRC, or with a bracket
6210 all of whose alternatives start with OP_SOD or OP_CIRC (recurse ad lib), then
6211 it's anchored. However, if this is a multiline pattern, then only OP_SOD
6212 counts, since OP_CIRC can match in the middle.
6213
6214 We can also consider a regex to be anchored if OP_SOM starts all its branches.
6215 This is the code for \G, which means "match at start of match position, taking
6216 into account the match offset".
6217
6218 A branch is also implicitly anchored if it starts with .* and DOTALL is set,
6219 because that will try the rest of the pattern at all possible matching points,
6220 so there is no point trying again.... er ....
6221
6222 .... except when the .* appears inside capturing parentheses, and there is a
6223 subsequent back reference to those parentheses. We haven't enough information
6224 to catch that case precisely.
6225
6226 At first, the best we could do was to detect when .* was in capturing brackets
6227 and the highest back reference was greater than or equal to that level.
6228 However, by keeping a bitmap of the first 31 back references, we can catch some
6229 of the more common cases more precisely.
6230
6231 Arguments:
6232 code points to start of expression (the bracket)
6233 options points to the options setting
6234 bracket_map a bitmap of which brackets we are inside while testing; this
6235 handles up to substring 31; after that we just have to take
6236 the less precise approach
6237 backref_map the back reference bitmap
6238
6239 Returns: TRUE or FALSE
6240 */
6241
6242 static BOOL
6243 is_anchored(register const uschar *code, int *options, unsigned int bracket_map,
6244 unsigned int backref_map)
6245 {
6246 do {
6247 const uschar *scode = first_significant_code(code + _pcre_OP_lengths[*code],
6248 options, PCRE_MULTILINE, FALSE);
6249 register int op = *scode;
6250
6251 /* Non-capturing brackets */
6252
6253 if (op == OP_BRA)
6254 {
6255 if (!is_anchored(scode, options, bracket_map, backref_map)) return FALSE;
6256 }
6257
6258 /* Capturing brackets */
6259
6260 else if (op == OP_CBRA)
6261 {
6262 int n = GET2(scode, 1+LINK_SIZE);
6263 int new_map = bracket_map | ((n < 32)? (1 << n) : 1);
6264 if (!is_anchored(scode, options, new_map, backref_map)) return FALSE;
6265 }
6266
6267 /* Other brackets */
6268
6269 else if (op == OP_ASSERT || op == OP_ONCE || op == OP_COND)
6270 {
6271 if (!is_anchored(scode, options, bracket_map, backref_map)) return FALSE;
6272 }
6273
6274 /* .* is not anchored unless DOTALL is set (which generates OP_ALLANY) and
6275 it isn't in brackets that are or may be referenced. */
6276
6277 else if ((op == OP_TYPESTAR || op == OP_TYPEMINSTAR ||
6278 op == OP_TYPEPOSSTAR))
6279 {
6280 if (scode[1] != OP_ALLANY || (bracket_map & backref_map) != 0)
6281 return FALSE;
6282 }
6283
6284 /* Check for explicit anchoring */
6285
6286 else if (op != OP_SOD && op != OP_SOM &&
6287 ((*options & PCRE_MULTILINE) != 0 || op != OP_CIRC))
6288 return FALSE;
6289 code += GET(code, 1);
6290 }
6291 while (*code == OP_ALT); /* Loop for each alternative */
6292 return TRUE;
6293 }
6294
6295
6296
6297 /*************************************************
6298 * Check for starting with ^ or .* *
6299 *************************************************/
6300
6301 /* This is called to find out if every branch starts with ^ or .* so that
6302 "first char" processing can be done to speed things up in multiline
6303 matching and for non-DOTALL patterns that start with .* (which must start at
6304 the beginning or after \n). As in the case of is_anchored() (see above), we
6305 have to take account of back references to capturing brackets that contain .*
6306 because in that case we can't make the assumption.
6307
6308 Arguments:
6309 code points to start of expression (the bracket)
6310 bracket_map a bitmap of which brackets we are inside while testing; this
6311 handles up to substring 31; after that we just have to take
6312 the less precise approach
6313 backref_map the back reference bitmap
6314
6315 Returns: TRUE or FALSE
6316 */
6317
6318 static BOOL
6319 is_startline(const uschar *code, unsigned int bracket_map,
6320 unsigned int backref_map)
6321 {
6322 do {
6323 const uschar *scode = first_significant_code(code + _pcre_OP_lengths[*code],
6324 NULL, 0, FALSE);
6325 register int op = *scode;
6326
6327 /* If we are at the start of a conditional assertion group, *both* the
6328 conditional assertion *and* what follows the condition must satisfy the test
6329 for start of line. Other kinds of condition fail. Note that there may be an
6330 auto-callout at the start of a condition. */
6331
6332 if (op == OP_COND)
6333 {
6334 scode += 1 + LINK_SIZE;
6335 if (*scode == OP_CALLOUT) scode += _pcre_OP_lengths[OP_CALLOUT];
6336 switch (*scode)
6337 {
6338 case OP_CREF:
6339 case OP_NCREF:
6340 case OP_RREF:
6341 case OP_NRREF:
6342 case OP_DEF:
6343 return FALSE;
6344
6345 default: /* Assertion */
6346 if (!is_startline(scode, bracket_map, backref_map)) return FALSE;
6347 do scode += GET(scode, 1); while (*scode == OP_ALT);
6348 scode += 1 + LINK_SIZE;
6349 break;
6350 }
6351 scode = first_significant_code(scode, NULL, 0, FALSE);
6352 op = *scode;
6353 }
6354
6355 /* Non-capturing brackets */
6356
6357 if (op == OP_BRA)
6358 {
6359 if (!is_startline(scode, bracket_map, backref_map)) return FALSE;
6360 }
6361
6362 /* Capturing brackets */
6363
6364 else if (op == OP_CBRA)
6365 {
6366 int n = GET2(scode, 1+LINK_SIZE);
6367 int new_map = bracket_map | ((n < 32)? (1 << n) : 1);
6368 if (!is_startline(scode, new_map, backref_map)) return FALSE;
6369 }
6370
6371 /* Other brackets */
6372
6373 else if (op == OP_ASSERT || op == OP_ONCE)
6374 {
6375 if (!is_startline(scode, bracket_map, backref_map)) return FALSE;
6376 }
6377
6378 /* .* means "start at start or after \n" if it isn't in brackets that
6379 may be referenced. */
6380
6381 else if (op == OP_TYPESTAR || op == OP_TYPEMINSTAR || op == OP_TYPEPOSSTAR)
6382 {
6383 if (scode[1] != OP_ANY || (bracket_map & backref_map) != 0) return FALSE;
6384 }
6385
6386 /* Check for explicit circumflex */
6387
6388 else if (op != OP_CIRC) return FALSE;
6389
6390 /* Move on to the next alternative */
6391
6392 code += GET(code, 1);
6393 }
6394 while (*code == OP_ALT); /* Loop for each alternative */
6395 return TRUE;
6396 }
6397
6398
6399
6400 /*************************************************
6401 * Check for asserted fixed first char *
6402 *************************************************/
6403
6404 /* During compilation, the "first char" settings from forward assertions are
6405 discarded, because they can cause conflicts with actual literals that follow.
6406 However, if we end up without a first char setting for an unanchored pattern,
6407 it is worth scanning the regex to see if there is an initial asserted first
6408 char. If all branches start with the same asserted char, or with a bracket all
6409 of whose alternatives start with the same asserted char (recurse ad lib), then
6410 we return that char, otherwise -1.
6411
6412 Arguments:
6413 code points to start of expression (the bracket)
6414 options pointer to the options (used to check casing changes)
6415 inassert TRUE if in an assertion
6416
6417 Returns: -1 or the fixed first char
6418 */
6419
6420 static int
6421 find_firstassertedchar(const uschar *code, int *options, BOOL inassert)
6422 {
6423 register int c = -1;
6424 do {
6425 int d;
6426 const uschar *scode =
6427 first_significant_code(code + 1+LINK_SIZE, options, PCRE_CASELESS, TRUE);
6428 register int op = *scode;
6429
6430 switch(op)
6431 {
6432 default:
6433 return -1;
6434
6435 case OP_BRA:
6436 case OP_CBRA:
6437 case OP_ASSERT:
6438 case OP_ONCE:
6439 case OP_COND:
6440 if ((d = find_firstassertedchar(scode, options, op == OP_ASSERT)) < 0)
6441 return -1;
6442 if (c < 0) c = d; else if (c != d) return -1;
6443 break;
6444
6445 case OP_EXACT: /* Fall through */
6446 scode += 2;
6447
6448 case OP_CHAR:
6449 case OP_CHARNC:
6450 case OP_PLUS:
6451 case OP_MINPLUS:
6452 case OP_POSPLUS:
6453 if (!inassert) return -1;
6454 if (c < 0)
6455 {
6456 c = scode[1];
6457 if ((*options & PCRE_CASELESS) != 0) c |= REQ_CASELESS;
6458 }
6459 else if (c != scode[1]) return -1;
6460 break;
6461 }
6462
6463 code += GET(code, 1);
6464 }
6465 while (*code == OP_ALT);
6466 return c;
6467 }
6468
6469
6470
6471 /*************************************************
6472 * Compile a Regular Expression *
6473 *************************************************/
6474
6475 /* This function takes a string and returns a pointer to a block of store
6476 holding a compiled version of the expression. The original API for this
6477 function had no error code return variable; it is retained for backwards
6478 compatibility. The new function is given a new name.
6479
6480 Arguments:
6481 pattern the regular expression
6482 options various option bits
6483 errorcodeptr pointer to error code variable (pcre_compile2() only)
6484 can be NULL if you don't want a code value
6485 errorptr pointer to pointer to error text
6486 erroroffset ptr offset in pattern where error was detected
6487 tables pointer to character tables or NULL
6488
6489 Returns: pointer to compiled data block, or NULL on error,
6490 with errorptr and erroroffset set
6491 */
6492
6493 PCRE_EXP_DEFN pcre * PCRE_CALL_CONVENTION
6494 pcre_compile(const char *pattern, int options, const char **errorptr,
6495 int *erroroffset, const unsigned char *tables)
6496 {
6497 return pcre_compile2(pattern, options, NULL, errorptr, erroroffset, tables);
6498 }
6499
6500
6501 PCRE_EXP_DEFN pcre * PCRE_CALL_CONVENTION
6502 pcre_compile2(const char *pattern, int options, int *errorcodeptr,
6503 const char **errorptr, int *erroroffset, const unsigned char *tables)
6504 {
6505 real_pcre *re;
6506 int length = 1; /* For final END opcode */
6507 int firstbyte, reqbyte, newline;
6508 int errorcode = 0;
6509 int skipatstart = 0;
6510 BOOL utf8 = (options & PCRE_UTF8) != 0;
6511 size_t size;
6512 uschar *code;
6513 const uschar *codestart;
6514 const uschar *ptr;
6515 compile_data compile_block;
6516 compile_data *cd = &compile_block;
6517
6518 /* This space is used for "compiling" into during the first phase, when we are
6519 computing the amount of memory that is needed. Compiled items are thrown away
6520 as soon as possible, so that a fairly large buffer should be sufficient for
6521 this purpose. The same space is used in the second phase for remembering where
6522 to fill in forward references to subpatterns. */
6523
6524 uschar cworkspace[COMPILE_WORK_SIZE];
6525
6526 /* Set this early so that early errors get offset 0. */
6527
6528 ptr = (const uschar *)pattern;
6529
6530 /* We can't pass back an error message if errorptr is NULL; I guess the best we
6531 can do is just return NULL, but we can set a code value if there is a code
6532 pointer. */
6533
6534 if (errorptr == NULL)
6535 {
6536 if (errorcodeptr != NULL) *errorcodeptr = 99;
6537 return NULL;
6538 }
6539
6540 *errorptr = NULL;
6541 if (errorcodeptr != NULL) *errorcodeptr = ERR0;
6542
6543 /* However, we can give a message for this error */
6544
6545 if (erroroffset == NULL)
6546 {
6547 errorcode = ERR16;
6548 goto PCRE_EARLY_ERROR_RETURN2;
6549 }
6550
6551 *erroroffset = 0;
6552
6553 /* Set up pointers to the individual character tables */
6554
6555 if (tables == NULL) tables = _pcre_default_tables;
6556 cd->lcc = tables + lcc_offset;
6557 cd->fcc = tables + fcc_offset;
6558 cd->cbits = tables + cbits_offset;
6559 cd->ctypes = tables + ctypes_offset;
6560
6561 /* Check that all undefined public option bits are zero */
6562
6563 if ((options & ~PUBLIC_COMPILE_OPTIONS) != 0)
6564 {
6565 errorcode = ERR17;
6566 goto PCRE_EARLY_ERROR_RETURN;
6567 }
6568
6569 /* Check for global one-time settings at the start of the pattern, and remember
6570 the offset for later. */
6571
6572 while (ptr[skipatstart] == CHAR_LEFT_PARENTHESIS &&
6573 ptr[skipatstart+1] == CHAR_ASTERISK)
6574 {
6575 int newnl = 0;
6576 int newbsr = 0;
6577
6578 if (strncmp((char *)(ptr+skipatstart+2), STRING_UTF8_RIGHTPAR, 5) == 0)
6579 { skipatstart += 7; options |= PCRE_UTF8; continue; }
6580
6581 if (strncmp((char *)(ptr+skipatstart+2), STRING_CR_RIGHTPAR, 3) == 0)
6582 { skipatstart += 5; newnl = PCRE_NEWLINE_CR; }
6583 else if (strncmp((char *)(ptr+skipatstart+2), STRING_LF_RIGHTPAR, 3) == 0)
6584 { skipatstart += 5; newnl = PCRE_NEWLINE_LF; }
6585 else if (strncmp((char *)(ptr+skipatstart+2), STRING_CRLF_RIGHTPAR, 5) == 0)
6586 { skipatstart += 7; newnl = PCRE_NEWLINE_CR + PCRE_NEWLINE_LF; }
6587 else if (strncmp((char *)(ptr+skipatstart+2), STRING_ANY_RIGHTPAR, 4) == 0)
6588 { skipatstart += 6; newnl = PCRE_NEWLINE_ANY; }
6589 else if (strncmp((char *)(ptr+skipatstart+2), STRING_ANYCRLF_RIGHTPAR, 8) == 0)
6590 { skipatstart += 10; newnl = PCRE_NEWLINE_ANYCRLF; }
6591
6592 else if (strncmp((char *)(ptr+skipatstart+2), STRING_BSR_ANYCRLF_RIGHTPAR, 12) == 0)
6593 { skipatstart += 14; newbsr = PCRE_BSR_ANYCRLF; }
6594 else if (strncmp((char *)(ptr+skipatstart+2), STRING_BSR_UNICODE_RIGHTPAR, 12) == 0)
6595 { skipatstart += 14; newbsr = PCRE_BSR_UNICODE; }
6596
6597 if (newnl != 0)
6598 options = (options & ~PCRE_NEWLINE_BITS) | newnl;
6599 else if (newbsr != 0)
6600 options = (options & ~(PCRE_BSR_ANYCRLF|PCRE_BSR_UNICODE)) | newbsr;
6601 else break;
6602 }
6603
6604 /* Can't support UTF8 unless PCRE has been compiled to include the code. */
6605
6606 #ifdef SUPPORT_UTF8
6607 if (utf8 && (options & PCRE_NO_UTF8_CHECK) == 0 &&
6608 (*erroroffset = _pcre_valid_utf8((USPTR)pattern, -1)) >= 0)
6609 {
6610 errorcode = ERR44;
6611 goto PCRE_EARLY_ERROR_RETURN2;
6612 }
6613 #else
6614 if (utf8)
6615 {
6616 errorcode = ERR32;
6617 goto PCRE_EARLY_ERROR_RETURN;
6618 }
6619 #endif
6620
6621 /* Check validity of \R options. */
6622
6623 switch (options & (PCRE_BSR_ANYCRLF|PCRE_BSR_UNICODE))
6624 {
6625 case 0:
6626 case PCRE_BSR_ANYCRLF:
6627 case PCRE_BSR_UNICODE:
6628 break;
6629 default: errorcode = ERR56; goto PCRE_EARLY_ERROR_RETURN;
6630 }
6631
6632 /* Handle different types of newline. The three bits give seven cases. The
6633 current code allows for fixed one- or two-byte sequences, plus "any" and
6634 "anycrlf". */
6635