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

Contents of /code/trunk/pcre_compile.c

Parent Directory Parent Directory | Revision Log Revision Log


Revision 512 - (show annotations) (download)
Tue Mar 30 11:11:52 2010 UTC (4 years, 6 months ago) by ph10
File MIME type: text/plain
File size: 225268 byte(s)
Fix compile problems when heap is in use

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.
3208 Elsewhere it marks a word boundary. Other escapes have preset maps ready
3209 to 'or' into the one we are building. We assume they have more than one
3210 character in them, so set class_charcount bigger than one. */
3211
3212 if (c == CHAR_BACKSLASH)
3213 {
3214 c = check_escape(&ptr, errorcodeptr, cd->bracount, options, TRUE);
3215 if (*errorcodeptr != 0) goto FAILED;
3216
3217 if (-c == ESC_b) c = CHAR_BS; /* \b is backspace in a class */
3218 else if (-c == ESC_X) c = CHAR_X; /* \X is literal X in a class */
3219 else if (-c == ESC_R) c = CHAR_R; /* \R is literal R in a class */
3220 else if (-c == ESC_Q) /* Handle start of quoted string */
3221 {
3222 if (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
3223 {
3224 ptr += 2; /* avoid empty string */
3225 }
3226 else inescq = TRUE;
3227 continue;
3228 }
3229 else if (-c == ESC_E) continue; /* Ignore orphan \E */
3230
3231 if (c < 0)
3232 {
3233 register const uschar *cbits = cd->cbits;
3234 class_charcount += 2; /* Greater than 1 is what matters */
3235
3236 /* Save time by not doing this in the pre-compile phase. */
3237
3238 if (lengthptr == NULL) switch (-c)
3239 {
3240 case ESC_d:
3241 for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_digit];
3242 continue;
3243
3244 case ESC_D:
3245 should_flip_negation = TRUE;
3246 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_digit];
3247 continue;
3248
3249 case ESC_w:
3250 for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_word];
3251 continue;
3252
3253 case ESC_W:
3254 should_flip_negation = TRUE;
3255 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_word];
3256 continue;
3257
3258 case ESC_s:
3259 for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_space];
3260 classbits[1] &= ~0x08; /* Perl 5.004 onwards omits VT from \s */
3261 continue;
3262
3263 case ESC_S:
3264 should_flip_negation = TRUE;
3265 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_space];
3266 classbits[1] |= 0x08; /* Perl 5.004 onwards omits VT from \s */
3267 continue;
3268
3269 default: /* Not recognized; fall through */
3270 break; /* Need "default" setting to stop compiler warning. */
3271 }
3272
3273 /* In the pre-compile phase, just do the recognition. */
3274
3275 else if (c == -ESC_d || c == -ESC_D || c == -ESC_w ||
3276 c == -ESC_W || c == -ESC_s || c == -ESC_S) continue;
3277
3278 /* We need to deal with \H, \h, \V, and \v in both phases because
3279 they use extra memory. */
3280
3281 if (-c == ESC_h)
3282 {
3283 SETBIT(classbits, 0x09); /* VT */
3284 SETBIT(classbits, 0x20); /* SPACE */
3285 SETBIT(classbits, 0xa0); /* NSBP */
3286 #ifdef SUPPORT_UTF8
3287 if (utf8)
3288 {
3289 class_utf8 = TRUE;
3290 *class_utf8data++ = XCL_SINGLE;
3291 class_utf8data += _pcre_ord2utf8(0x1680, class_utf8data);
3292 *class_utf8data++ = XCL_SINGLE;
3293 class_utf8data += _pcre_ord2utf8(0x180e, class_utf8data);
3294 *class_utf8data++ = XCL_RANGE;
3295 class_utf8data += _pcre_ord2utf8(0x2000, class_utf8data);
3296 class_utf8data += _pcre_ord2utf8(0x200A, class_utf8data);
3297 *class_utf8data++ = XCL_SINGLE;
3298 class_utf8data += _pcre_ord2utf8(0x202f, class_utf8data);
3299 *class_utf8data++ = XCL_SINGLE;
3300 class_utf8data += _pcre_ord2utf8(0x205f, class_utf8data);
3301 *class_utf8data++ = XCL_SINGLE;
3302 class_utf8data += _pcre_ord2utf8(0x3000, class_utf8data);
3303 }
3304 #endif
3305 continue;
3306 }
3307
3308 if (-c == ESC_H)
3309 {
3310 for (c = 0; c < 32; c++)
3311 {
3312 int x = 0xff;
3313 switch (c)
3314 {
3315 case 0x09/8: x ^= 1 << (0x09%8); break;
3316 case 0x20/8: x ^= 1 << (0x20%8); break;
3317 case 0xa0/8: x ^= 1 << (0xa0%8); break;
3318 default: break;
3319 }
3320 classbits[c] |= x;
3321 }
3322
3323 #ifdef SUPPORT_UTF8
3324 if (utf8)
3325 {
3326 class_utf8 = TRUE;
3327 *class_utf8data++ = XCL_RANGE;
3328 class_utf8data += _pcre_ord2utf8(0x0100, class_utf8data);
3329 class_utf8data += _pcre_ord2utf8(0x167f, class_utf8data);
3330 *class_utf8data++ = XCL_RANGE;
3331 class_utf8data += _pcre_ord2utf8(0x1681, class_utf8data);
3332 class_utf8data += _pcre_ord2utf8(0x180d, class_utf8data);
3333 *class_utf8data++ = XCL_RANGE;
3334 class_utf8data += _pcre_ord2utf8(0x180f, class_utf8data);
3335 class_utf8data += _pcre_ord2utf8(0x1fff, class_utf8data);
3336 *class_utf8data++ = XCL_RANGE;
3337 class_utf8data += _pcre_ord2utf8(0x200B, class_utf8data);
3338 class_utf8data += _pcre_ord2utf8(0x202e, class_utf8data);
3339 *class_utf8data++ = XCL_RANGE;
3340 class_utf8data += _pcre_ord2utf8(0x2030, class_utf8data);
3341 class_utf8data += _pcre_ord2utf8(0x205e, class_utf8data);
3342 *class_utf8data++ = XCL_RANGE;
3343 class_utf8data += _pcre_ord2utf8(0x2060, class_utf8data);
3344 class_utf8data += _pcre_ord2utf8(0x2fff, class_utf8data);
3345 *class_utf8data++ = XCL_RANGE;
3346 class_utf8data += _pcre_ord2utf8(0x3001, class_utf8data);
3347 class_utf8data += _pcre_ord2utf8(0x7fffffff, class_utf8data);
3348 }
3349 #endif
3350 continue;
3351 }
3352
3353 if (-c == ESC_v)
3354 {
3355 SETBIT(classbits, 0x0a); /* LF */
3356 SETBIT(classbits, 0x0b); /* VT */
3357 SETBIT(classbits, 0x0c); /* FF */
3358 SETBIT(classbits, 0x0d); /* CR */
3359 SETBIT(classbits, 0x85); /* NEL */
3360 #ifdef SUPPORT_UTF8
3361 if (utf8)
3362 {
3363 class_utf8 = TRUE;
3364 *class_utf8data++ = XCL_RANGE;
3365 class_utf8data += _pcre_ord2utf8(0x2028, class_utf8data);
3366 class_utf8data += _pcre_ord2utf8(0x2029, class_utf8data);
3367 }
3368 #endif
3369 continue;
3370 }
3371
3372 if (-c == ESC_V)
3373 {
3374 for (c = 0; c < 32; c++)
3375 {
3376 int x = 0xff;
3377 switch (c)
3378 {
3379 case 0x0a/8: x ^= 1 << (0x0a%8);
3380 x ^= 1 << (0x0b%8);
3381 x ^= 1 << (0x0c%8);
3382 x ^= 1 << (0x0d%8);
3383 break;
3384 case 0x85/8: x ^= 1 << (0x85%8); break;
3385 default: break;
3386 }
3387 classbits[c] |= x;
3388 }
3389
3390 #ifdef SUPPORT_UTF8
3391 if (utf8)
3392 {
3393 class_utf8 = TRUE;
3394 *class_utf8data++ = XCL_RANGE;
3395 class_utf8data += _pcre_ord2utf8(0x0100, class_utf8data);
3396 class_utf8data += _pcre_ord2utf8(0x2027, class_utf8data);
3397 *class_utf8data++ = XCL_RANGE;
3398 class_utf8data += _pcre_ord2utf8(0x2029, class_utf8data);
3399 class_utf8data += _pcre_ord2utf8(0x7fffffff, class_utf8data);
3400 }
3401 #endif
3402 continue;
3403 }
3404
3405 /* We need to deal with \P and \p in both phases. */
3406
3407 #ifdef SUPPORT_UCP
3408 if (-c == ESC_p || -c == ESC_P)
3409 {
3410 BOOL negated;
3411 int pdata;
3412 int ptype = get_ucp(&ptr, &negated, &pdata, errorcodeptr);
3413 if (ptype < 0) goto FAILED;
3414 class_utf8 = TRUE;
3415 *class_utf8data++ = ((-c == ESC_p) != negated)?
3416 XCL_PROP : XCL_NOTPROP;
3417 *class_utf8data++ = ptype;
3418 *class_utf8data++ = pdata;
3419 class_charcount -= 2; /* Not a < 256 character */
3420 continue;
3421 }
3422 #endif
3423 /* Unrecognized escapes are faulted if PCRE is running in its
3424 strict mode. By default, for compatibility with Perl, they are
3425 treated as literals. */
3426
3427 if ((options & PCRE_EXTRA) != 0)
3428 {
3429 *errorcodeptr = ERR7;
3430 goto FAILED;
3431 }
3432
3433 class_charcount -= 2; /* Undo the default count from above */
3434 c = *ptr; /* Get the final character and fall through */
3435 }
3436
3437 /* Fall through if we have a single character (c >= 0). This may be
3438 greater than 256 in UTF-8 mode. */
3439
3440 } /* End of backslash handling */
3441
3442 /* A single character may be followed by '-' to form a range. However,
3443 Perl does not permit ']' to be the end of the range. A '-' character
3444 at the end is treated as a literal. Perl ignores orphaned \E sequences
3445 entirely. The code for handling \Q and \E is messy. */
3446
3447 CHECK_RANGE:
3448 while (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
3449 {
3450 inescq = FALSE;
3451 ptr += 2;
3452 }
3453
3454 oldptr = ptr;
3455
3456 /* Remember \r or \n */
3457
3458 if (c == CHAR_CR || c == CHAR_NL) cd->external_flags |= PCRE_HASCRORLF;
3459
3460 /* Check for range */
3461
3462 if (!inescq && ptr[1] == CHAR_MINUS)
3463 {
3464 int d;
3465 ptr += 2;
3466 while (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_E) ptr += 2;
3467
3468 /* If we hit \Q (not followed by \E) at this point, go into escaped
3469 mode. */
3470
3471 while (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_Q)
3472 {
3473 ptr += 2;
3474 if (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_E)
3475 { ptr += 2; continue; }
3476 inescq = TRUE;
3477 break;
3478 }
3479
3480 if (*ptr == 0 || (!inescq && *ptr == CHAR_RIGHT_SQUARE_BRACKET))
3481 {
3482 ptr = oldptr;
3483 goto LONE_SINGLE_CHARACTER;
3484 }
3485
3486 #ifdef SUPPORT_UTF8
3487 if (utf8)
3488 { /* Braces are required because the */
3489 GETCHARLEN(d, ptr, ptr); /* macro generates multiple statements */
3490 }
3491 else
3492 #endif
3493 d = *ptr; /* Not UTF-8 mode */
3494
3495 /* The second part of a range can be a single-character escape, but
3496 not any of the other escapes. Perl 5.6 treats a hyphen as a literal
3497 in such circumstances. */
3498
3499 if (!inescq && d == CHAR_BACKSLASH)
3500 {
3501 d = check_escape(&ptr, errorcodeptr, cd->bracount, options, TRUE);
3502 if (*errorcodeptr != 0) goto FAILED;
3503
3504 /* \b is backspace; \X is literal X; \R is literal R; any other
3505 special means the '-' was literal */
3506
3507 if (d < 0)
3508 {
3509 if (d == -ESC_b) d = CHAR_BS;
3510 else if (d == -ESC_X) d = CHAR_X;
3511 else if (d == -ESC_R) d = CHAR_R; else
3512 {
3513 ptr = oldptr;
3514 goto LONE_SINGLE_CHARACTER; /* A few lines below */
3515 }
3516 }
3517 }
3518
3519 /* Check that the two values are in the correct order. Optimize
3520 one-character ranges */
3521
3522 if (d < c)
3523 {
3524 *errorcodeptr = ERR8;
3525 goto FAILED;
3526 }
3527
3528 if (d == c) goto LONE_SINGLE_CHARACTER; /* A few lines below */
3529
3530 /* Remember \r or \n */
3531
3532 if (d == CHAR_CR || d == CHAR_NL) cd->external_flags |= PCRE_HASCRORLF;
3533
3534 /* In UTF-8 mode, if the upper limit is > 255, or > 127 for caseless
3535 matching, we have to use an XCLASS with extra data items. Caseless
3536 matching for characters > 127 is available only if UCP support is
3537 available. */
3538
3539 #ifdef SUPPORT_UTF8
3540 if (utf8 && (d > 255 || ((options & PCRE_CASELESS) != 0 && d > 127)))
3541 {
3542 class_utf8 = TRUE;
3543
3544 /* With UCP support, we can find the other case equivalents of
3545 the relevant characters. There may be several ranges. Optimize how
3546 they fit with the basic range. */
3547
3548 #ifdef SUPPORT_UCP
3549 if ((options & PCRE_CASELESS) != 0)
3550 {
3551 unsigned int occ, ocd;
3552 unsigned int cc = c;
3553 unsigned int origd = d;
3554 while (get_othercase_range(&cc, origd, &occ, &ocd))
3555 {
3556 if (occ >= (unsigned int)c &&
3557 ocd <= (unsigned int)d)
3558 continue; /* Skip embedded ranges */
3559
3560 if (occ < (unsigned int)c &&
3561 ocd >= (unsigned int)c - 1) /* Extend the basic range */
3562 { /* if there is overlap, */
3563 c = occ; /* noting that if occ < c */
3564 continue; /* we can't have ocd > d */
3565 } /* because a subrange is */
3566 if (ocd > (unsigned int)d &&
3567 occ <= (unsigned int)d + 1) /* always shorter than */
3568 { /* the basic range. */
3569 d = ocd;
3570 continue;
3571 }
3572
3573 if (occ == ocd)
3574 {
3575 *class_utf8data++ = XCL_SINGLE;
3576 }
3577 else
3578 {
3579 *class_utf8data++ = XCL_RANGE;
3580 class_utf8data += _pcre_ord2utf8(occ, class_utf8data);
3581 }
3582 class_utf8data += _pcre_ord2utf8(ocd, class_utf8data);
3583 }
3584 }
3585 #endif /* SUPPORT_UCP */
3586
3587 /* Now record the original range, possibly modified for UCP caseless
3588 overlapping ranges. */
3589
3590 *class_utf8data++ = XCL_RANGE;
3591 class_utf8data += _pcre_ord2utf8(c, class_utf8data);
3592 class_utf8data += _pcre_ord2utf8(d, class_utf8data);
3593
3594 /* With UCP support, we are done. Without UCP support, there is no
3595 caseless matching for UTF-8 characters > 127; we can use the bit map
3596 for the smaller ones. */
3597
3598 #ifdef SUPPORT_UCP
3599 continue; /* With next character in the class */
3600 #else
3601 if ((options & PCRE_CASELESS) == 0 || c > 127) continue;
3602
3603 /* Adjust upper limit and fall through to set up the map */
3604
3605 d = 127;
3606
3607 #endif /* SUPPORT_UCP */
3608 }
3609 #endif /* SUPPORT_UTF8 */
3610
3611 /* We use the bit map for all cases when not in UTF-8 mode; else
3612 ranges that lie entirely within 0-127 when there is UCP support; else
3613 for partial ranges without UCP support. */
3614
3615 class_charcount += d - c + 1;
3616 class_lastchar = d;
3617
3618 /* We can save a bit of time by skipping this in the pre-compile. */
3619
3620 if (lengthptr == NULL) for (; c <= d; c++)
3621 {
3622 classbits[c/8] |= (1 << (c&7));
3623 if ((options & PCRE_CASELESS) != 0)
3624 {
3625 int uc = cd->fcc[c]; /* flip case */
3626 classbits[uc/8] |= (1 << (uc&7));
3627 }
3628 }
3629
3630 continue; /* Go get the next char in the class */
3631 }
3632
3633 /* Handle a lone single character - we can get here for a normal
3634 non-escape char, or after \ that introduces a single character or for an
3635 apparent range that isn't. */
3636
3637 LONE_SINGLE_CHARACTER:
3638
3639 /* Handle a character that cannot go in the bit map */
3640
3641 #ifdef SUPPORT_UTF8
3642 if (utf8 && (c > 255 || ((options & PCRE_CASELESS) != 0 && c > 127)))
3643 {
3644 class_utf8 = TRUE;
3645 *class_utf8data++ = XCL_SINGLE;
3646 class_utf8data += _pcre_ord2utf8(c, class_utf8data);
3647
3648 #ifdef SUPPORT_UCP
3649 if ((options & PCRE_CASELESS) != 0)
3650 {
3651 unsigned int othercase;
3652 if ((othercase = UCD_OTHERCASE(c)) != c)
3653 {
3654 *class_utf8data++ = XCL_SINGLE;
3655 class_utf8data += _pcre_ord2utf8(othercase, class_utf8data);
3656 }
3657 }
3658 #endif /* SUPPORT_UCP */
3659
3660 }
3661 else
3662 #endif /* SUPPORT_UTF8 */
3663
3664 /* Handle a single-byte character */
3665 {
3666 classbits[c/8] |= (1 << (c&7));
3667 if ((options & PCRE_CASELESS) != 0)
3668 {
3669 c = cd->fcc[c]; /* flip case */
3670 classbits[c/8] |= (1 << (c&7));
3671 }
3672 class_charcount++;
3673 class_lastchar = c;
3674 }
3675 }
3676
3677 /* Loop until ']' reached. This "while" is the end of the "do" above. */
3678
3679 while ((c = *(++ptr)) != 0 && (c != CHAR_RIGHT_SQUARE_BRACKET || inescq));
3680
3681 if (c == 0) /* Missing terminating ']' */
3682 {
3683 *errorcodeptr = ERR6;
3684 goto FAILED;
3685 }
3686
3687
3688 /* This code has been disabled because it would mean that \s counts as
3689 an explicit \r or \n reference, and that's not really what is wanted. Now
3690 we set the flag only if there is a literal "\r" or "\n" in the class. */
3691
3692 #if 0
3693 /* Remember whether \r or \n are in this class */
3694
3695 if (negate_class)
3696 {
3697 if ((classbits[1] & 0x24) != 0x24) cd->external_flags |= PCRE_HASCRORLF;
3698 }
3699 else
3700 {
3701 if ((classbits[1] & 0x24) != 0) cd->external_flags |= PCRE_HASCRORLF;
3702 }
3703 #endif
3704
3705
3706 /* If class_charcount is 1, we saw precisely one character whose value is
3707 less than 256. As long as there were no characters >= 128 and there was no
3708 use of \p or \P, in other words, no use of any XCLASS features, we can
3709 optimize.
3710
3711 In UTF-8 mode, we can optimize the negative case only if there were no
3712 characters >= 128 because OP_NOT and the related opcodes like OP_NOTSTAR
3713 operate on single-bytes only. This is an historical hangover. Maybe one day
3714 we can tidy these opcodes to handle multi-byte characters.
3715
3716 The optimization throws away the bit map. We turn the item into a
3717 1-character OP_CHAR[NC] if it's positive, or OP_NOT if it's negative. Note
3718 that OP_NOT does not support multibyte characters. In the positive case, it
3719 can cause firstbyte to be set. Otherwise, there can be no first char if
3720 this item is first, whatever repeat count may follow. In the case of
3721 reqbyte, save the previous value for reinstating. */
3722
3723 #ifdef SUPPORT_UTF8
3724 if (class_charcount == 1 && !class_utf8 &&
3725 (!utf8 || !negate_class || class_lastchar < 128))
3726 #else
3727 if (class_charcount == 1)
3728 #endif
3729 {
3730 zeroreqbyte = reqbyte;
3731
3732 /* The OP_NOT opcode works on one-byte characters only. */
3733
3734 if (negate_class)
3735 {
3736 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
3737 zerofirstbyte = firstbyte;
3738 *code++ = OP_NOT;
3739 *code++ = class_lastchar;
3740 break;
3741 }
3742
3743 /* For a single, positive character, get the value into mcbuffer, and
3744 then we can handle this with the normal one-character code. */
3745
3746 #ifdef SUPPORT_UTF8
3747 if (utf8 && class_lastchar > 127)
3748 mclength = _pcre_ord2utf8(class_lastchar, mcbuffer);
3749 else
3750 #endif
3751 {
3752 mcbuffer[0] = class_lastchar;
3753 mclength = 1;
3754 }
3755 goto ONE_CHAR;
3756 } /* End of 1-char optimization */
3757
3758 /* The general case - not the one-char optimization. If this is the first
3759 thing in the branch, there can be no first char setting, whatever the
3760 repeat count. Any reqbyte setting must remain unchanged after any kind of
3761 repeat. */
3762
3763 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
3764 zerofirstbyte = firstbyte;
3765 zeroreqbyte = reqbyte;
3766
3767 /* If there are characters with values > 255, we have to compile an
3768 extended class, with its own opcode, unless there was a negated special
3769 such as \S in the class, because in that case all characters > 255 are in
3770 the class, so any that were explicitly given as well can be ignored. If
3771 (when there are explicit characters > 255 that must be listed) there are no
3772 characters < 256, we can omit the bitmap in the actual compiled code. */
3773
3774 #ifdef SUPPORT_UTF8
3775 if (class_utf8 && !should_flip_negation)
3776 {
3777 *class_utf8data++ = XCL_END; /* Marks the end of extra data */
3778 *code++ = OP_XCLASS;
3779 code += LINK_SIZE;
3780 *code = negate_class? XCL_NOT : 0;
3781
3782 /* If the map is required, move up the extra data to make room for it;
3783 otherwise just move the code pointer to the end of the extra data. */
3784
3785 if (class_charcount > 0)
3786 {
3787 *code++ |= XCL_MAP;
3788 memmove(code + 32, code, class_utf8data - code);
3789 memcpy(code, classbits, 32);
3790 code = class_utf8data + 32;
3791 }
3792 else code = class_utf8data;
3793
3794 /* Now fill in the complete length of the item */
3795
3796 PUT(previous, 1, code - previous);
3797 break; /* End of class handling */
3798 }
3799 #endif
3800
3801 /* If there are no characters > 255, set the opcode to OP_CLASS or
3802 OP_NCLASS, depending on whether the whole class was negated and whether
3803 there were negative specials such as \S in the class. Then copy the 32-byte
3804 map into the code vector, negating it if necessary. */
3805
3806 *code++ = (negate_class == should_flip_negation) ? OP_CLASS : OP_NCLASS;
3807 if (negate_class)
3808 {
3809 if (lengthptr == NULL) /* Save time in the pre-compile phase */
3810 for (c = 0; c < 32; c++) code[c] = ~classbits[c];
3811 }
3812 else
3813 {
3814 memcpy(code, classbits, 32);
3815 }
3816 code += 32;
3817 break;
3818
3819
3820 /* ===================================================================*/
3821 /* Various kinds of repeat; '{' is not necessarily a quantifier, but this
3822 has been tested above. */
3823
3824 case CHAR_LEFT_CURLY_BRACKET:
3825 if (!is_quantifier) goto NORMAL_CHAR;
3826 ptr = read_repeat_counts(ptr+1, &repeat_min, &repeat_max, errorcodeptr);
3827 if (*errorcodeptr != 0) goto FAILED;
3828 goto REPEAT;
3829
3830 case CHAR_ASTERISK:
3831 repeat_min = 0;
3832 repeat_max = -1;
3833 goto REPEAT;
3834
3835 case CHAR_PLUS:
3836 repeat_min = 1;
3837 repeat_max = -1;
3838 goto REPEAT;
3839
3840 case CHAR_QUESTION_MARK:
3841 repeat_min = 0;
3842 repeat_max = 1;
3843
3844 REPEAT:
3845 if (previous == NULL)
3846 {
3847 *errorcodeptr = ERR9;
3848 goto FAILED;
3849 }
3850
3851 if (repeat_min == 0)
3852 {
3853 firstbyte = zerofirstbyte; /* Adjust for zero repeat */
3854 reqbyte = zeroreqbyte; /* Ditto */
3855 }
3856
3857 /* Remember whether this is a variable length repeat */
3858
3859 reqvary = (repeat_min == repeat_max)? 0 : REQ_VARY;
3860
3861 op_type = 0; /* Default single-char op codes */
3862 possessive_quantifier = FALSE; /* Default not possessive quantifier */
3863
3864 /* Save start of previous item, in case we have to move it up to make space
3865 for an inserted OP_ONCE for the additional '+' extension. */
3866
3867 tempcode = previous;
3868
3869 /* If the next character is '+', we have a possessive quantifier. This
3870 implies greediness, whatever the setting of the PCRE_UNGREEDY option.
3871 If the next character is '?' this is a minimizing repeat, by default,
3872 but if PCRE_UNGREEDY is set, it works the other way round. We change the
3873 repeat type to the non-default. */
3874
3875 if (ptr[1] == CHAR_PLUS)
3876 {
3877 repeat_type = 0; /* Force greedy */
3878 possessive_quantifier = TRUE;
3879 ptr++;
3880 }
3881 else if (ptr[1] == CHAR_QUESTION_MARK)
3882 {
3883 repeat_type = greedy_non_default;
3884 ptr++;
3885 }
3886 else repeat_type = greedy_default;
3887
3888 /* If previous was a character match, abolish the item and generate a
3889 repeat item instead. If a char item has a minumum of more than one, ensure
3890 that it is set in reqbyte - it might not be if a sequence such as x{3} is
3891 the first thing in a branch because the x will have gone into firstbyte
3892 instead. */
3893
3894 if (*previous == OP_CHAR || *previous == OP_CHARNC)
3895 {
3896 /* Deal with UTF-8 characters that take up more than one byte. It's
3897 easier to write this out separately than try to macrify it. Use c to
3898 hold the length of the character in bytes, plus 0x80 to flag that it's a
3899 length rather than a small character. */
3900
3901 #ifdef SUPPORT_UTF8
3902 if (utf8 && (code[-1] & 0x80) != 0)
3903 {
3904 uschar *lastchar = code - 1;
3905 while((*lastchar & 0xc0) == 0x80) lastchar--;
3906 c = code - lastchar; /* Length of UTF-8 character */
3907 memcpy(utf8_char, lastchar, c); /* Save the char */
3908 c |= 0x80; /* Flag c as a length */
3909 }
3910 else
3911 #endif
3912
3913 /* Handle the case of a single byte - either with no UTF8 support, or
3914 with UTF-8 disabled, or for a UTF-8 character < 128. */
3915
3916 {
3917 c = code[-1];
3918 if (repeat_min > 1) reqbyte = c | req_caseopt | cd->req_varyopt;
3919 }
3920
3921 /* If the repetition is unlimited, it pays to see if the next thing on
3922 the line is something that cannot possibly match this character. If so,
3923 automatically possessifying this item gains some performance in the case
3924 where the match fails. */
3925
3926 if (!possessive_quantifier &&
3927 repeat_max < 0 &&
3928 check_auto_possessive(*previous, c, utf8, utf8_char, ptr + 1,
3929 options, cd))
3930 {
3931 repeat_type = 0; /* Force greedy */
3932 possessive_quantifier = TRUE;
3933 }
3934
3935 goto OUTPUT_SINGLE_REPEAT; /* Code shared with single character types */
3936 }
3937
3938 /* If previous was a single negated character ([^a] or similar), we use
3939 one of the special opcodes, replacing it. The code is shared with single-
3940 character repeats by setting opt_type to add a suitable offset into
3941 repeat_type. We can also test for auto-possessification. OP_NOT is
3942 currently used only for single-byte chars. */
3943
3944 else if (*previous == OP_NOT)
3945 {
3946 op_type = OP_NOTSTAR - OP_STAR; /* Use "not" opcodes */
3947 c = previous[1];
3948 if (!possessive_quantifier &&
3949 repeat_max < 0 &&
3950 check_auto_possessive(OP_NOT, c, utf8, NULL, ptr + 1, options, cd))
3951 {
3952 repeat_type = 0; /* Force greedy */
3953 possessive_quantifier = TRUE;
3954 }
3955 goto OUTPUT_SINGLE_REPEAT;
3956 }
3957
3958 /* If previous was a character type match (\d or similar), abolish it and
3959 create a suitable repeat item. The code is shared with single-character
3960 repeats by setting op_type to add a suitable offset into repeat_type. Note
3961 the the Unicode property types will be present only when SUPPORT_UCP is
3962 defined, but we don't wrap the little bits of code here because it just
3963 makes it horribly messy. */
3964
3965 else if (*previous < OP_EODN)
3966 {
3967 uschar *oldcode;
3968 int prop_type, prop_value;
3969 op_type = OP_TYPESTAR - OP_STAR; /* Use type opcodes */
3970 c = *previous;
3971
3972 if (!possessive_quantifier &&
3973 repeat_max < 0 &&
3974 check_auto_possessive(c, 0, utf8, NULL, ptr + 1, options, cd))
3975 {
3976 repeat_type = 0; /* Force greedy */
3977 possessive_quantifier = TRUE;
3978 }
3979
3980 OUTPUT_SINGLE_REPEAT:
3981 if (*previous == OP_PROP || *previous == OP_NOTPROP)
3982 {
3983 prop_type = previous[1];
3984 prop_value = previous[2];
3985 }
3986 else prop_type = prop_value = -1;
3987
3988 oldcode = code;
3989 code = previous; /* Usually overwrite previous item */
3990
3991 /* If the maximum is zero then the minimum must also be zero; Perl allows
3992 this case, so we do too - by simply omitting the item altogether. */
3993
3994 if (repeat_max == 0) goto END_REPEAT;
3995
3996 /*--------------------------------------------------------------------*/
3997 /* This code is obsolete from release 8.00; the restriction was finally
3998 removed: */
3999
4000 /* All real repeats make it impossible to handle partial matching (maybe
4001 one day we will be able to remove this restriction). */
4002
4003 /* if (repeat_max != 1) cd->external_flags |= PCRE_NOPARTIAL; */
4004 /*--------------------------------------------------------------------*/
4005
4006 /* Combine the op_type with the repeat_type */
4007
4008 repeat_type += op_type;
4009
4010 /* A minimum of zero is handled either as the special case * or ?, or as
4011 an UPTO, with the maximum given. */
4012
4013 if (repeat_min == 0)
4014 {
4015 if (repeat_max == -1) *code++ = OP_STAR + repeat_type;
4016 else if (repeat_max == 1) *code++ = OP_QUERY + repeat_type;
4017 else
4018 {
4019 *code++ = OP_UPTO + repeat_type;
4020 PUT2INC(code, 0, repeat_max);
4021 }
4022 }
4023
4024 /* A repeat minimum of 1 is optimized into some special cases. If the
4025 maximum is unlimited, we use OP_PLUS. Otherwise, the original item is
4026 left in place and, if the maximum is greater than 1, we use OP_UPTO with
4027 one less than the maximum. */
4028
4029 else if (repeat_min == 1)
4030 {
4031 if (repeat_max == -1)
4032 *code++ = OP_PLUS + repeat_type;
4033 else
4034 {
4035 code = oldcode; /* leave previous item in place */
4036 if (repeat_max == 1) goto END_REPEAT;
4037 *code++ = OP_UPTO + repeat_type;
4038 PUT2INC(code, 0, repeat_max - 1);
4039 }
4040 }
4041
4042 /* The case {n,n} is just an EXACT, while the general case {n,m} is
4043 handled as an EXACT followed by an UPTO. */
4044
4045 else
4046 {
4047 *code++ = OP_EXACT + op_type; /* NB EXACT doesn't have repeat_type */
4048 PUT2INC(code, 0, repeat_min);
4049
4050 /* If the maximum is unlimited, insert an OP_STAR. Before doing so,
4051 we have to insert the character for the previous code. For a repeated
4052 Unicode property match, there are two extra bytes that define the
4053 required property. In UTF-8 mode, long characters have their length in
4054 c, with the 0x80 bit as a flag. */
4055
4056 if (repeat_max < 0)
4057 {
4058 #ifdef SUPPORT_UTF8
4059 if (utf8 && c >= 128)
4060 {
4061 memcpy(code, utf8_char, c & 7);
4062 code += c & 7;
4063 }
4064 else
4065 #endif
4066 {
4067 *code++ = c;
4068 if (prop_type >= 0)
4069 {
4070 *code++ = prop_type;
4071 *code++ = prop_value;
4072 }
4073 }
4074 *code++ = OP_STAR + repeat_type;
4075 }
4076
4077 /* Else insert an UPTO if the max is greater than the min, again
4078 preceded by the character, for the previously inserted code. If the
4079 UPTO is just for 1 instance, we can use QUERY instead. */
4080
4081 else if (repeat_max != repeat_min)
4082 {
4083 #ifdef SUPPORT_UTF8
4084 if (utf8 && c >= 128)
4085 {
4086 memcpy(code, utf8_char, c & 7);
4087 code += c & 7;
4088 }
4089 else
4090 #endif
4091 *code++ = c;
4092 if (prop_type >= 0)
4093 {
4094 *code++ = prop_type;
4095 *code++ = prop_value;
4096 }
4097 repeat_max -= repeat_min;
4098
4099 if (repeat_max == 1)
4100 {
4101 *code++ = OP_QUERY + repeat_type;
4102 }
4103 else
4104 {
4105 *code++ = OP_UPTO + repeat_type;
4106 PUT2INC(code, 0, repeat_max);
4107 }
4108 }
4109 }
4110
4111 /* The character or character type itself comes last in all cases. */
4112
4113 #ifdef SUPPORT_UTF8
4114 if (utf8 && c >= 128)
4115 {
4116 memcpy(code, utf8_char, c & 7);
4117 code += c & 7;
4118 }
4119 else
4120 #endif
4121 *code++ = c;
4122
4123 /* For a repeated Unicode property match, there are two extra bytes that
4124 define the required property. */
4125
4126 #ifdef SUPPORT_UCP
4127 if (prop_type >= 0)
4128 {
4129 *code++ = prop_type;
4130 *code++ = prop_value;
4131 }
4132 #endif
4133 }
4134
4135 /* If previous was a character class or a back reference, we put the repeat
4136 stuff after it, but just skip the item if the repeat was {0,0}. */
4137
4138 else if (*previous == OP_CLASS ||
4139 *previous == OP_NCLASS ||
4140 #ifdef SUPPORT_UTF8
4141 *previous == OP_XCLASS ||
4142 #endif
4143 *previous == OP_REF)
4144 {
4145 if (repeat_max == 0)
4146 {
4147 code = previous;
4148 goto END_REPEAT;
4149 }
4150
4151 /*--------------------------------------------------------------------*/
4152 /* This code is obsolete from release 8.00; the restriction was finally
4153 removed: */
4154
4155 /* All real repeats make it impossible to handle partial matching (maybe
4156 one day we will be able to remove this restriction). */
4157
4158 /* if (repeat_max != 1) cd->external_flags |= PCRE_NOPARTIAL; */
4159 /*--------------------------------------------------------------------*/
4160
4161 if (repeat_min == 0 && repeat_max == -1)
4162 *code++ = OP_CRSTAR + repeat_type;
4163 else if (repeat_min == 1 && repeat_max == -1)
4164 *code++ = OP_CRPLUS + repeat_type;
4165 else if (repeat_min == 0 && repeat_max == 1)
4166 *code++ = OP_CRQUERY + repeat_type;
4167 else
4168 {
4169 *code++ = OP_CRRANGE + repeat_type;
4170 PUT2INC(code, 0, repeat_min);
4171 if (repeat_max == -1) repeat_max = 0; /* 2-byte encoding for max */
4172 PUT2INC(code, 0, repeat_max);
4173 }
4174 }
4175
4176 /* If previous was a bracket group, we may have to replicate it in certain
4177 cases. */
4178
4179 else if (*previous == OP_BRA || *previous == OP_CBRA ||
4180 *previous == OP_ONCE || *previous == OP_COND)
4181 {
4182 register int i;
4183 int ketoffset = 0;
4184 int len = code - previous;
4185 uschar *bralink = NULL;
4186
4187 /* Repeating a DEFINE group is pointless */
4188
4189 if (*previous == OP_COND && previous[LINK_SIZE+1] == OP_DEF)
4190 {
4191 *errorcodeptr = ERR55;
4192 goto FAILED;
4193 }
4194
4195 /* If the maximum repeat count is unlimited, find the end of the bracket
4196 by scanning through from the start, and compute the offset back to it
4197 from the current code pointer. There may be an OP_OPT setting following
4198 the final KET, so we can't find the end just by going back from the code
4199 pointer. */
4200
4201 if (repeat_max == -1)
4202 {
4203 register uschar *ket = previous;
4204 do ket += GET(ket, 1); while (*ket != OP_KET);
4205 ketoffset = code - ket;
4206 }
4207
4208 /* The case of a zero minimum is special because of the need to stick
4209 OP_BRAZERO in front of it, and because the group appears once in the
4210 data, whereas in other cases it appears the minimum number of times. For
4211 this reason, it is simplest to treat this case separately, as otherwise
4212 the code gets far too messy. There are several special subcases when the
4213 minimum is zero. */
4214
4215 if (repeat_min == 0)
4216 {
4217 /* If the maximum is also zero, we used to just omit the group from the
4218 output altogether, like this:
4219
4220 ** if (repeat_max == 0)
4221 ** {
4222 ** code = previous;
4223 ** goto END_REPEAT;
4224 ** }
4225
4226 However, that fails when a group is referenced as a subroutine from
4227 elsewhere in the pattern, so now we stick in OP_SKIPZERO in front of it
4228 so that it is skipped on execution. As we don't have a list of which
4229 groups are referenced, we cannot do this selectively.
4230
4231 If the maximum is 1 or unlimited, we just have to stick in the BRAZERO
4232 and do no more at this point. However, we do need to adjust any
4233 OP_RECURSE calls inside the group that refer to the group itself or any
4234 internal or forward referenced group, because the offset is from the
4235 start of the whole regex. Temporarily terminate the pattern while doing
4236 this. */
4237
4238 if (repeat_max <= 1) /* Covers 0, 1, and unlimited */
4239 {
4240 *code = OP_END;
4241 adjust_recurse(previous, 1, utf8, cd, save_hwm);
4242 memmove(previous+1, previous, len);
4243 code++;
4244 if (repeat_max == 0)
4245 {
4246 *previous++ = OP_SKIPZERO;
4247 goto END_REPEAT;
4248 }
4249 *previous++ = OP_BRAZERO + repeat_type;
4250 }
4251
4252 /* If the maximum is greater than 1 and limited, we have to replicate
4253 in a nested fashion, sticking OP_BRAZERO before each set of brackets.
4254 The first one has to be handled carefully because it's the original
4255 copy, which has to be moved up. The remainder can be handled by code
4256 that is common with the non-zero minimum case below. We have to
4257 adjust the value or repeat_max, since one less copy is required. Once
4258 again, we may have to adjust any OP_RECURSE calls inside the group. */
4259
4260 else
4261 {
4262 int offset;
4263 *code = OP_END;
4264 adjust_recurse(previous, 2 + LINK_SIZE, utf8, cd, save_hwm);
4265 memmove(previous + 2 + LINK_SIZE, previous, len);
4266 code += 2 + LINK_SIZE;
4267 *previous++ = OP_BRAZERO + repeat_type;
4268 *previous++ = OP_BRA;
4269
4270 /* We chain together the bracket offset fields that have to be
4271 filled in later when the ends of the brackets are reached. */
4272
4273 offset = (bralink == NULL)? 0 : previous - bralink;
4274 bralink = previous;
4275 PUTINC(previous, 0, offset);
4276 }
4277
4278 repeat_max--;
4279 }
4280
4281 /* If the minimum is greater than zero, replicate the group as many
4282 times as necessary, and adjust the maximum to the number of subsequent
4283 copies that we need. If we set a first char from the group, and didn't
4284 set a required char, copy the latter from the former. If there are any
4285 forward reference subroutine calls in the group, there will be entries on
4286 the workspace list; replicate these with an appropriate increment. */
4287
4288 else
4289 {
4290 if (repeat_min > 1)
4291 {
4292 /* In the pre-compile phase, we don't actually do the replication. We
4293 just adjust the length as if we had. Do some paranoid checks for
4294 potential integer overflow. The INT64_OR_DOUBLE type is a 64-bit
4295 integer type when available, otherwise double. */
4296
4297 if (lengthptr != NULL)
4298 {
4299 int delta = (repeat_min - 1)*length_prevgroup;
4300 if ((INT64_OR_DOUBLE)(repeat_min - 1)*
4301 (INT64_OR_DOUBLE)length_prevgroup >
4302 (INT64_OR_DOUBLE)INT_MAX ||
4303 OFLOW_MAX - *lengthptr < delta)
4304 {
4305 *errorcodeptr = ERR20;
4306 goto FAILED;
4307 }
4308 *lengthptr += delta;
4309 }
4310
4311 /* This is compiling for real */
4312
4313 else
4314 {
4315 if (groupsetfirstbyte && reqbyte < 0) reqbyte = firstbyte;
4316 for (i = 1; i < repeat_min; i++)
4317 {
4318 uschar *hc;
4319 uschar *this_hwm = cd->hwm;
4320 memcpy(code, previous, len);
4321 for (hc = save_hwm; hc < this_hwm; hc += LINK_SIZE)
4322 {
4323 PUT(cd->hwm, 0, GET(hc, 0) + len);
4324 cd->hwm += LINK_SIZE;
4325 }
4326 save_hwm = this_hwm;
4327 code += len;
4328 }
4329 }
4330 }
4331
4332 if (repeat_max > 0) repeat_max -= repeat_min;
4333 }
4334
4335 /* This code is common to both the zero and non-zero minimum cases. If
4336 the maximum is limited, it replicates the group in a nested fashion,
4337 remembering the bracket starts on a stack. In the case of a zero minimum,
4338 the first one was set up above. In all cases the repeat_max now specifies
4339 the number of additional copies needed. Again, we must remember to
4340 replicate entries on the forward reference list. */
4341
4342 if (repeat_max >= 0)
4343 {
4344 /* In the pre-compile phase, we don't actually do the replication. We
4345 just adjust the length as if we had. For each repetition we must add 1
4346 to the length for BRAZERO and for all but the last repetition we must
4347 add 2 + 2*LINKSIZE to allow for the nesting that occurs. Do some
4348 paranoid checks to avoid integer overflow. The INT64_OR_DOUBLE type is
4349 a 64-bit integer type when available, otherwise double. */
4350
4351 if (lengthptr != NULL && repeat_max > 0)
4352 {
4353 int delta = repeat_max * (length_prevgroup + 1 + 2 + 2*LINK_SIZE) -
4354 2 - 2*LINK_SIZE; /* Last one doesn't nest */
4355 if ((INT64_OR_DOUBLE)repeat_max *
4356 (INT64_OR_DOUBLE)(length_prevgroup + 1 + 2 + 2*LINK_SIZE)
4357 > (INT64_OR_DOUBLE)INT_MAX ||
4358 OFLOW_MAX - *lengthptr < delta)
4359 {
4360 *errorcodeptr = ERR20;
4361 goto FAILED;
4362 }
4363 *lengthptr += delta;
4364 }
4365
4366 /* This is compiling for real */
4367
4368 else for (i = repeat_max - 1; i >= 0; i--)
4369 {
4370 uschar *hc;
4371 uschar *this_hwm = cd->hwm;
4372
4373 *code++ = OP_BRAZERO + repeat_type;
4374
4375 /* All but the final copy start a new nesting, maintaining the
4376 chain of brackets outstanding. */
4377
4378 if (i != 0)
4379 {
4380 int offset;
4381 *code++ = OP_BRA;
4382 offset = (bralink == NULL)? 0 : code - bralink;
4383 bralink = code;
4384 PUTINC(code, 0, offset);
4385 }
4386
4387 memcpy(code, previous, len);
4388 for (hc = save_hwm; hc < this_hwm; hc += LINK_SIZE)
4389 {
4390 PUT(cd->hwm, 0, GET(hc, 0) + len + ((i != 0)? 2+LINK_SIZE : 1));
4391 cd->hwm += LINK_SIZE;
4392 }
4393 save_hwm = this_hwm;
4394 code += len;
4395 }
4396
4397 /* Now chain through the pending brackets, and fill in their length
4398 fields (which are holding the chain links pro tem). */
4399
4400 while (bralink != NULL)
4401 {
4402 int oldlinkoffset;
4403 int offset = code - bralink + 1;
4404 uschar *bra = code - offset;
4405 oldlinkoffset = GET(bra, 1);
4406 bralink = (oldlinkoffset == 0)? NULL : bralink - oldlinkoffset;
4407 *code++ = OP_KET;
4408 PUTINC(code, 0, offset);
4409 PUT(bra, 1, offset);
4410 }
4411 }
4412
4413 /* If the maximum is unlimited, set a repeater in the final copy. We
4414 can't just offset backwards from the current code point, because we
4415 don't know if there's been an options resetting after the ket. The
4416 correct offset was computed above.
4417
4418 Then, when we are doing the actual compile phase, check to see whether
4419 this group is a non-atomic one that could match an empty string. If so,
4420 convert the initial operator to the S form (e.g. OP_BRA -> OP_SBRA) so
4421 that runtime checking can be done. [This check is also applied to
4422 atomic groups at runtime, but in a different way.] */
4423
4424 else
4425 {
4426 uschar *ketcode = code - ketoffset;
4427 uschar *bracode = ketcode - GET(ketcode, 1);
4428 *ketcode = OP_KETRMAX + repeat_type;
4429 if (lengthptr == NULL && *bracode != OP_ONCE)
4430 {
4431 uschar *scode = bracode;
4432 do
4433 {
4434 if (could_be_empty_branch(scode, ketcode, utf8, cd))
4435 {
4436 *bracode += OP_SBRA - OP_BRA;
4437 break;
4438 }
4439 scode += GET(scode, 1);
4440 }
4441 while (*scode == OP_ALT);
4442 }
4443 }
4444 }
4445
4446 /* If previous is OP_FAIL, it was generated by an empty class [] in
4447 JavaScript mode. The other ways in which OP_FAIL can be generated, that is
4448 by (*FAIL) or (?!) set previous to NULL, which gives a "nothing to repeat"
4449 error above. We can just ignore the repeat in JS case. */
4450
4451 else if (*previous == OP_FAIL) goto END_REPEAT;
4452
4453 /* Else there's some kind of shambles */
4454
4455 else
4456 {
4457 *errorcodeptr = ERR11;
4458 goto FAILED;
4459 }
4460
4461 /* If the character following a repeat is '+', or if certain optimization
4462 tests above succeeded, possessive_quantifier is TRUE. For some of the
4463 simpler opcodes, there is an special alternative opcode for this. For
4464 anything else, we wrap the entire repeated item inside OP_ONCE brackets.
4465 The '+' notation is just syntactic sugar, taken from Sun's Java package,
4466 but the special opcodes can optimize it a bit. The repeated item starts at
4467 tempcode, not at previous, which might be the first part of a string whose
4468 (former) last char we repeated.
4469
4470 Possessifying an 'exact' quantifier has no effect, so we can ignore it. But
4471 an 'upto' may follow. We skip over an 'exact' item, and then test the
4472 length of what remains before proceeding. */
4473
4474 if (possessive_quantifier)
4475 {
4476 int len;
4477
4478 if (*tempcode == OP_TYPEEXACT)
4479 tempcode += _pcre_OP_lengths[*tempcode] +
4480 ((tempcode[3] == OP_PROP || tempcode[3] == OP_NOTPROP)? 2 : 0);
4481
4482 else if (*tempcode == OP_EXACT || *tempcode == OP_NOTEXACT)
4483 {
4484 tempcode += _pcre_OP_lengths[*tempcode];
4485 #ifdef SUPPORT_UTF8
4486 if (utf8 && tempcode[-1] >= 0xc0)
4487 tempcode += _pcre_utf8_table4[tempcode[-1] & 0x3f];
4488 #endif
4489 }
4490
4491 len = code - tempcode;
4492 if (len > 0) switch (*tempcode)
4493 {
4494 case OP_STAR: *tempcode = OP_POSSTAR; break;
4495 case OP_PLUS: *tempcode = OP_POSPLUS; break;
4496 case OP_QUERY: *tempcode = OP_POSQUERY; break;
4497 case OP_UPTO: *tempcode = OP_POSUPTO; break;
4498
4499 case OP_TYPESTAR: *tempcode = OP_TYPEPOSSTAR; break;
4500 case OP_TYPEPLUS: *tempcode = OP_TYPEPOSPLUS; break;
4501 case OP_TYPEQUERY: *tempcode = OP_TYPEPOSQUERY; break;
4502 case OP_TYPEUPTO: *tempcode = OP_TYPEPOSUPTO; break;
4503
4504 case OP_NOTSTAR: *tempcode = OP_NOTPOSSTAR; break;
4505 case OP_NOTPLUS: *tempcode = OP_NOTPOSPLUS; break;
4506 case OP_NOTQUERY: *tempcode = OP_NOTPOSQUERY; break;
4507 case OP_NOTUPTO: *tempcode = OP_NOTPOSUPTO; break;
4508
4509 /* Because we are moving code along, we must ensure that any
4510 pending recursive references are updated. */
4511
4512 default:
4513 *code = OP_END;
4514 adjust_recurse(tempcode, 1 + LINK_SIZE, utf8, cd, save_hwm);
4515 memmove(tempcode + 1+LINK_SIZE, tempcode, len);
4516 code += 1 + LINK_SIZE;
4517 len += 1 + LINK_SIZE;
4518 tempcode[0] = OP_ONCE;
4519 *code++ = OP_KET;
4520 PUTINC(code, 0, len);
4521 PUT(tempcode, 1, len);
4522 break;
4523 }
4524 }
4525
4526 /* In all case we no longer have a previous item. We also set the
4527 "follows varying string" flag for subsequently encountered reqbytes if
4528 it isn't already set and we have just passed a varying length item. */
4529
4530 END_REPEAT:
4531 previous = NULL;
4532 cd->req_varyopt |= reqvary;
4533 break;
4534
4535
4536 /* ===================================================================*/
4537 /* Start of nested parenthesized sub-expression, or comment or lookahead or
4538 lookbehind or option setting or condition or all the other extended
4539 parenthesis forms. */
4540
4541 case CHAR_LEFT_PARENTHESIS:
4542 newoptions = options;
4543 skipbytes = 0;
4544 bravalue = OP_CBRA;
4545 save_hwm = cd->hwm;
4546 reset_bracount = FALSE;
4547
4548 /* First deal with various "verbs" that can be introduced by '*'. */
4549
4550 if (*(++ptr) == CHAR_ASTERISK &&
4551 ((cd->ctypes[ptr[1]] & ctype_letter) != 0 || ptr[1] == ':'))
4552 {
4553 int i, namelen;
4554 int arglen = 0;
4555 const char *vn = verbnames;
4556 const uschar *name = ptr + 1;
4557 const uschar *arg = NULL;
4558 previous = NULL;
4559 while ((cd->ctypes[*++ptr] & ctype_letter) != 0) {};
4560 namelen = ptr - name;
4561
4562 if (*ptr == CHAR_COLON)
4563 {
4564 arg = ++ptr;
4565 while ((cd->ctypes[*ptr] & (ctype_letter|ctype_digit)) != 0
4566 || *ptr == '_') ptr++;
4567 arglen = ptr - arg;
4568 }
4569
4570 if (*ptr != CHAR_RIGHT_PARENTHESIS)
4571 {
4572 *errorcodeptr = ERR60;
4573 goto FAILED;
4574 }
4575
4576 /* Scan the table of verb names */
4577
4578 for (i = 0; i < verbcount; i++)
4579 {
4580 if (namelen == verbs[i].len &&
4581 strncmp((char *)name, vn, namelen) == 0)
4582 {
4583 /* Check for open captures before ACCEPT */
4584
4585 if (verbs[i].op == OP_ACCEPT)
4586 {
4587 open_capitem *oc;
4588 cd->had_accept = TRUE;
4589 for (oc = cd->open_caps; oc != NULL; oc = oc->next)
4590 {
4591 *code++ = OP_CLOSE;
4592 PUT2INC(code, 0, oc->number);
4593 }
4594 }
4595
4596 /* Handle the cases with/without an argument */
4597
4598 if (arglen == 0)
4599 {
4600 if (verbs[i].op < 0) /* Argument is mandatory */
4601 {
4602 *errorcodeptr = ERR66;
4603 goto FAILED;
4604 }
4605 *code++ = verbs[i].op;
4606 }
4607
4608 else
4609 {
4610 if (verbs[i].op_arg < 0) /* Argument is forbidden */
4611 {
4612 *errorcodeptr = ERR59;
4613 goto FAILED;
4614 }
4615 *code++ = verbs[i].op_arg;
4616 *code++ = arglen;
4617 memcpy(code, arg, arglen);
4618 code += arglen;
4619 *code++ = 0;
4620 }
4621
4622 break; /* Found verb, exit loop */
4623 }
4624
4625 vn += verbs[i].len + 1;
4626 }
4627
4628 if (i < verbcount) continue; /* Successfully handled a verb */
4629 *errorcodeptr = ERR60; /* Verb not recognized */
4630 goto FAILED;
4631 }
4632
4633 /* Deal with the extended parentheses; all are introduced by '?', and the
4634 appearance of any of them means that this is not a capturing group. */
4635
4636 else if (*ptr == CHAR_QUESTION_MARK)
4637 {
4638 int i, set, unset, namelen;
4639 int *optset;
4640 const uschar *name;
4641 uschar *slot;
4642
4643 switch (*(++ptr))
4644 {
4645 case CHAR_NUMBER_SIGN: /* Comment; skip to ket */
4646 ptr++;
4647 while (*ptr != 0 && *ptr != CHAR_RIGHT_PARENTHESIS) ptr++;
4648 if (*ptr == 0)
4649 {
4650 *errorcodeptr = ERR18;
4651 goto FAILED;
4652 }
4653 continue;
4654
4655
4656 /* ------------------------------------------------------------ */
4657 case CHAR_VERTICAL_LINE: /* Reset capture count for each branch */
4658 reset_bracount = TRUE;
4659 /* Fall through */
4660
4661 /* ------------------------------------------------------------ */
4662 case CHAR_COLON: /* Non-capturing bracket */
4663 bravalue = OP_BRA;
4664 ptr++;
4665 break;
4666
4667
4668 /* ------------------------------------------------------------ */
4669 case CHAR_LEFT_PARENTHESIS:
4670 bravalue = OP_COND; /* Conditional group */
4671
4672 /* A condition can be an assertion, a number (referring to a numbered
4673 group), a name (referring to a named group), or 'R', referring to
4674 recursion. R<digits> and R&name are also permitted for recursion tests.
4675
4676 There are several syntaxes for testing a named group: (?(name)) is used
4677 by Python; Perl 5.10 onwards uses (?(<name>) or (?('name')).
4678
4679 There are two unfortunate ambiguities, caused by history. (a) 'R' can
4680 be the recursive thing or the name 'R' (and similarly for 'R' followed
4681 by digits), and (b) a number could be a name that consists of digits.
4682 In both cases, we look for a name first; if not found, we try the other
4683 cases. */
4684
4685 /* For conditions that are assertions, check the syntax, and then exit
4686 the switch. This will take control down to where bracketed groups,
4687 including assertions, are processed. */
4688
4689 if (ptr[1] == CHAR_QUESTION_MARK && (ptr[2] == CHAR_EQUALS_SIGN ||
4690 ptr[2] == CHAR_EXCLAMATION_MARK || ptr[2] == CHAR_LESS_THAN_SIGN))
4691 break;
4692
4693 /* Most other conditions use OP_CREF (a couple change to OP_RREF
4694 below), and all need to skip 3 bytes at the start of the group. */
4695
4696 code[1+LINK_SIZE] = OP_CREF;
4697 skipbytes = 3;
4698 refsign = -1;
4699
4700 /* Check for a test for recursion in a named group. */
4701
4702 if (ptr[1] == CHAR_R && ptr[2] == CHAR_AMPERSAND)
4703 {
4704 terminator = -1;
4705 ptr += 2;
4706 code[1+LINK_SIZE] = OP_RREF; /* Change the type of test */
4707 }
4708
4709 /* Check for a test for a named group's having been set, using the Perl
4710 syntax (?(<name>) or (?('name') */
4711
4712 else if (ptr[1] == CHAR_LESS_THAN_SIGN)
4713 {
4714 terminator = CHAR_GREATER_THAN_SIGN;
4715 ptr++;
4716 }
4717 else if (ptr[1] == CHAR_APOSTROPHE)
4718 {
4719 terminator = CHAR_APOSTROPHE;
4720 ptr++;
4721 }
4722 else
4723 {
4724 terminator = 0;
4725 if (ptr[1] == CHAR_MINUS || ptr[1] == CHAR_PLUS) refsign = *(++ptr);
4726 }
4727
4728 /* We now expect to read a name; any thing else is an error */
4729
4730 if ((cd->ctypes[ptr[1]] & ctype_word) == 0)
4731 {
4732 ptr += 1; /* To get the right offset */
4733 *errorcodeptr = ERR28;
4734 goto FAILED;
4735 }
4736
4737 /* Read the name, but also get it as a number if it's all digits */
4738
4739 recno = 0;
4740 name = ++ptr;
4741 while ((cd->ctypes[*ptr] & ctype_word) != 0)
4742 {
4743 if (recno >= 0)
4744 recno = ((digitab[*ptr] & ctype_digit) != 0)?
4745 recno * 10 + *ptr - CHAR_0 : -1;
4746 ptr++;
4747 }
4748 namelen = ptr - name;
4749
4750 if ((terminator > 0 && *ptr++ != terminator) ||
4751 *ptr++ != CHAR_RIGHT_PARENTHESIS)
4752 {
4753 ptr--; /* Error offset */
4754 *errorcodeptr = ERR26;
4755 goto FAILED;
4756 }
4757
4758 /* Do no further checking in the pre-compile phase. */
4759
4760 if (lengthptr != NULL) break;
4761
4762 /* In the real compile we do the work of looking for the actual
4763 reference. If the string started with "+" or "-" we require the rest to
4764 be digits, in which case recno will be set. */
4765
4766 if (refsign > 0)
4767 {
4768 if (recno <= 0)
4769 {
4770 *errorcodeptr = ERR58;
4771 goto FAILED;
4772 }
4773 recno = (refsign == CHAR_MINUS)?
4774 cd->bracount - recno + 1 : recno +cd->bracount;
4775 if (recno <= 0 || recno > cd->final_bracount)
4776 {
4777 *errorcodeptr = ERR15;
4778 goto FAILED;
4779 }
4780 PUT2(code, 2+LINK_SIZE, recno);
4781 break;
4782 }
4783
4784 /* Otherwise (did not start with "+" or "-"), start by looking for the
4785 name. If we find a name, add one to the opcode to change OP_CREF or
4786 OP_RREF into OP_NCREF or OP_NRREF. These behave exactly the same,
4787 except they record that the reference was originally to a name. The
4788 information is used to check duplicate names. */
4789
4790 slot = cd->name_table;
4791 for (i = 0; i < cd->names_found; i++)
4792 {
4793 if (strncmp((char *)name, (char *)slot+2, namelen) == 0) break;
4794 slot += cd->name_entry_size;
4795 }
4796
4797 /* Found a previous named subpattern */
4798
4799 if (i < cd->names_found)
4800 {
4801 recno = GET2(slot, 0);
4802 PUT2(code, 2+LINK_SIZE, recno);
4803 code[1+LINK_SIZE]++;
4804 }
4805
4806 /* Search the pattern for a forward reference */
4807
4808 else if ((i = find_parens(cd, name, namelen,
4809 (options & PCRE_EXTENDED) != 0)) > 0)
4810 {
4811 PUT2(code, 2+LINK_SIZE, i);
4812 code[1+LINK_SIZE]++;
4813 }
4814
4815 /* If terminator == 0 it means that the name followed directly after
4816 the opening parenthesis [e.g. (?(abc)...] and in this case there are
4817 some further alternatives to try. For the cases where terminator != 0
4818 [things like (?(<name>... or (?('name')... or (?(R&name)... ] we have
4819 now checked all the possibilities, so give an error. */
4820
4821 else if (terminator != 0)
4822 {
4823 *errorcodeptr = ERR15;
4824 goto FAILED;
4825 }
4826
4827 /* Check for (?(R) for recursion. Allow digits after R to specify a
4828 specific group number. */
4829
4830 else if (*name == CHAR_R)
4831 {
4832 recno = 0;
4833 for (i = 1; i < namelen; i++)
4834 {
4835 if ((digitab[name[i]] & ctype_digit) == 0)
4836 {
4837 *errorcodeptr = ERR15;
4838 goto FAILED;
4839 }
4840 recno = recno * 10 + name[i] - CHAR_0;
4841 }
4842 if (recno == 0) recno = RREF_ANY;
4843 code[1+LINK_SIZE] = OP_RREF; /* Change test type */
4844 PUT2(code, 2+LINK_SIZE, recno);
4845 }
4846
4847 /* Similarly, check for the (?(DEFINE) "condition", which is always
4848 false. */
4849
4850 else if (namelen == 6 && strncmp((char *)name, STRING_DEFINE, 6) == 0)
4851 {
4852 code[1+LINK_SIZE] = OP_DEF;
4853 skipbytes = 1;
4854 }
4855
4856 /* Check for the "name" actually being a subpattern number. We are
4857 in the second pass here, so final_bracount is set. */
4858
4859 else if (recno > 0 && recno <= cd->final_bracount)
4860 {
4861 PUT2(code, 2+LINK_SIZE, recno);
4862 }
4863
4864 /* Either an unidentified subpattern, or a reference to (?(0) */
4865
4866 else
4867 {
4868 *errorcodeptr = (recno == 0)? ERR35: ERR15;
4869 goto FAILED;
4870 }
4871 break;
4872
4873
4874 /* ------------------------------------------------------------ */
4875 case CHAR_EQUALS_SIGN: /* Positive lookahead */
4876 bravalue = OP_ASSERT;
4877 ptr++;
4878 break;
4879
4880
4881 /* ------------------------------------------------------------ */
4882 case CHAR_EXCLAMATION_MARK: /* Negative lookahead */
4883 ptr++;
4884 if (*ptr == CHAR_RIGHT_PARENTHESIS) /* Optimize (?!) */
4885 {
4886 *code++ = OP_FAIL;
4887 previous = NULL;
4888 continue;
4889 }
4890 bravalue = OP_ASSERT_NOT;
4891 break;
4892
4893
4894 /* ------------------------------------------------------------ */
4895 case CHAR_LESS_THAN_SIGN: /* Lookbehind or named define */
4896 switch (ptr[1])
4897 {
4898 case CHAR_EQUALS_SIGN: /* Positive lookbehind */
4899 bravalue = OP_ASSERTBACK;
4900 ptr += 2;
4901 break;
4902
4903 case CHAR_EXCLAMATION_MARK: /* Negative lookbehind */
4904 bravalue = OP_ASSERTBACK_NOT;
4905 ptr += 2;
4906 break;
4907
4908 default: /* Could be name define, else bad */
4909 if ((cd->ctypes[ptr[1]] & ctype_word) != 0) goto DEFINE_NAME;
4910 ptr++; /* Correct offset for error */
4911 *errorcodeptr = ERR24;
4912 goto FAILED;
4913 }
4914 break;
4915
4916
4917 /* ------------------------------------------------------------ */
4918 case CHAR_GREATER_THAN_SIGN: /* One-time brackets */
4919 bravalue = OP_ONCE;
4920 ptr++;
4921 break;
4922
4923
4924 /* ------------------------------------------------------------ */
4925 case CHAR_C: /* Callout - may be followed by digits; */
4926 previous_callout = code; /* Save for later completion */
4927 after_manual_callout = 1; /* Skip one item before completing */
4928 *code++ = OP_CALLOUT;
4929 {
4930 int n = 0;
4931 while ((digitab[*(++ptr)] & ctype_digit) != 0)
4932 n = n * 10 + *ptr - CHAR_0;
4933 if (*ptr != CHAR_RIGHT_PARENTHESIS)
4934 {
4935 *errorcodeptr = ERR39;
4936 goto FAILED;
4937 }
4938 if (n > 255)
4939 {
4940 *errorcodeptr = ERR38;
4941 goto FAILED;
4942 }
4943 *code++ = n;
4944 PUT(code, 0, ptr - cd->start_pattern + 1); /* Pattern offset */
4945 PUT(code, LINK_SIZE, 0); /* Default length */
4946 code += 2 * LINK_SIZE;
4947 }
4948 previous = NULL;
4949 continue;
4950
4951
4952 /* ------------------------------------------------------------ */
4953 case CHAR_P: /* Python-style named subpattern handling */
4954 if (*(++ptr) == CHAR_EQUALS_SIGN ||
4955 *ptr == CHAR_GREATER_THAN_SIGN) /* Reference or recursion */
4956 {
4957 is_recurse = *ptr == CHAR_GREATER_THAN_SIGN;
4958 terminator = CHAR_RIGHT_PARENTHESIS;
4959 goto NAMED_REF_OR_RECURSE;
4960 }
4961 else if (*ptr != CHAR_LESS_THAN_SIGN) /* Test for Python-style defn */
4962 {
4963 *errorcodeptr = ERR41;
4964 goto FAILED;
4965 }
4966 /* Fall through to handle (?P< as (?< is handled */
4967
4968
4969 /* ------------------------------------------------------------ */
4970 DEFINE_NAME: /* Come here from (?< handling */
4971 case CHAR_APOSTROPHE:
4972 {
4973 terminator = (*ptr == CHAR_LESS_THAN_SIGN)?
4974 CHAR_GREATER_THAN_SIGN : CHAR_APOSTROPHE;
4975 name = ++ptr;
4976
4977 while ((cd->ctypes[*ptr] & ctype_word) != 0) ptr++;
4978 namelen = ptr - name;
4979
4980 /* In the pre-compile phase, just do a syntax check. */
4981
4982 if (lengthptr != NULL)
4983 {
4984 if (*ptr != terminator)
4985 {
4986 *errorcodeptr = ERR42;
4987 goto FAILED;
4988 }
4989 if (cd->names_found >= MAX_NAME_COUNT)
4990 {
4991 *errorcodeptr = ERR49;
4992 goto FAILED;
4993 }
4994 if (namelen + 3 > cd->name_entry_size)
4995 {
4996 cd->name_entry_size = namelen + 3;
4997 if (namelen > MAX_NAME_SIZE)
4998 {
4999 *errorcodeptr = ERR48;
5000 goto FAILED;
5001 }
5002 }
5003 }
5004
5005 /* In the real compile, create the entry in the table, maintaining
5006 alphabetical order. Duplicate names for different numbers are
5007 permitted only if PCRE_DUPNAMES is set. Duplicate names for the same
5008 number are always OK. (An existing number can be re-used if (?|
5009 appears in the pattern.) In either event, a duplicate name results in
5010 a duplicate entry in the table, even if the number is the same. This
5011 is because the number of names, and hence the table size, is computed
5012 in the pre-compile, and it affects various numbers and pointers which
5013 would all have to be modified, and the compiled code moved down, if
5014 duplicates with the same number were omitted from the table. This
5015 doesn't seem worth the hassle. However, *different* names for the
5016 same number are not permitted. */
5017
5018 else
5019 {
5020 BOOL dupname = FALSE;
5021 slot = cd->name_table;
5022
5023 for (i = 0; i < cd->names_found; i++)
5024 {
5025 int crc = memcmp(name, slot+2, namelen);
5026 if (crc == 0)
5027 {
5028 if (slot[2+namelen] == 0)
5029 {
5030 if (GET2(slot, 0) != cd->bracount + 1 &&
5031 (options & PCRE_DUPNAMES) == 0)
5032 {
5033 *errorcodeptr = ERR43;
5034 goto FAILED;
5035 }
5036 else dupname = TRUE;
5037 }
5038 else crc = -1; /* Current name is a substring */
5039 }
5040
5041 /* Make space in the table and break the loop for an earlier
5042 name. For a duplicate or later name, carry on. We do this for
5043 duplicates so that in the simple case (when ?(| is not used) they
5044 are in order of their numbers. */
5045
5046 if (crc < 0)
5047 {
5048 memmove(slot + cd->name_entry_size, slot,
5049 (cd->names_found - i) * cd->name_entry_size);
5050 break;
5051 }
5052
5053 /* Continue the loop for a later or duplicate name */
5054
5055 slot += cd->name_entry_size;
5056 }
5057
5058 /* For non-duplicate names, check for a duplicate number before
5059 adding the new name. */
5060
5061 if (!dupname)
5062 {
5063 uschar *cslot = cd->name_table;
5064 for (i = 0; i < cd->names_found; i++)
5065 {
5066 if (cslot != slot)
5067 {
5068 if (GET2(cslot, 0) == cd->bracount + 1)
5069 {
5070 *errorcodeptr = ERR65;
5071 goto FAILED;
5072 }
5073 }
5074 else i--;
5075 cslot += cd->name_entry_size;
5076 }
5077 }
5078
5079 PUT2(slot, 0, cd->bracount + 1);
5080 memcpy(slot + 2, name, namelen);
5081 slot[2+namelen] = 0;
5082 }
5083 }
5084
5085 /* In both pre-compile and compile, count the number of names we've
5086 encountered. */
5087
5088 cd->names_found++;
5089 ptr++; /* Move past > or ' */
5090 goto NUMBERED_GROUP;
5091
5092
5093 /* ------------------------------------------------------------ */
5094 case CHAR_AMPERSAND: /* Perl recursion/subroutine syntax */
5095 terminator = CHAR_RIGHT_PARENTHESIS;
5096 is_recurse = TRUE;
5097 /* Fall through */
5098
5099 /* We come here from the Python syntax above that handles both
5100 references (?P=name) and recursion (?P>name), as well as falling
5101 through from the Perl recursion syntax (?&name). We also come here from
5102 the Perl \k<name> or \k'name' back reference syntax and the \k{name}
5103 .NET syntax, and the Oniguruma \g<...> and \g'...' subroutine syntax. */
5104
5105 NAMED_REF_OR_RECURSE:
5106 name = ++ptr;
5107 while ((cd->ctypes[*ptr] & ctype_word) != 0) ptr++;
5108 namelen = ptr - name;
5109
5110 /* In the pre-compile phase, do a syntax check and set a dummy
5111 reference number. */
5112
5113 if (lengthptr != NULL)
5114 {
5115 if (namelen == 0)
5116 {
5117 *errorcodeptr = ERR62;
5118 goto FAILED;
5119 }
5120 if (*ptr != terminator)
5121 {
5122 *errorcodeptr = ERR42;
5123 goto FAILED;
5124 }
5125 if (namelen > MAX_NAME_SIZE)
5126 {
5127 *errorcodeptr = ERR48;
5128 goto FAILED;
5129 }
5130 recno = 0;
5131 }
5132
5133 /* In the real compile, seek the name in the table. We check the name
5134 first, and then check that we have reached the end of the name in the
5135 table. That way, if the name that is longer than any in the table,
5136 the comparison will fail without reading beyond the table entry. */
5137
5138 else
5139 {
5140 slot = cd->name_table;
5141 for (i = 0; i < cd->names_found; i++)
5142 {
5143 if (strncmp((char *)name, (char *)slot+2, namelen) == 0 &&
5144 slot[2+namelen] == 0)
5145 break;
5146 slot += cd->name_entry_size;
5147 }
5148
5149 if (i < cd->names_found) /* Back reference */
5150 {
5151 recno = GET2(slot, 0);
5152 }
5153 else if ((recno = /* Forward back reference */
5154 find_parens(cd, name, namelen,
5155 (options & PCRE_EXTENDED) != 0)) <= 0)
5156 {
5157 *errorcodeptr = ERR15;
5158 goto FAILED;
5159 }
5160 }
5161
5162 /* In both phases, we can now go to the code than handles numerical
5163 recursion or backreferences. */
5164
5165 if (is_recurse) goto HANDLE_RECURSION;
5166 else goto HANDLE_REFERENCE;
5167
5168
5169 /* ------------------------------------------------------------ */
5170 case CHAR_R: /* Recursion */
5171 ptr++; /* Same as (?0) */
5172 /* Fall through */
5173
5174
5175 /* ------------------------------------------------------------ */
5176 case CHAR_MINUS: case CHAR_PLUS: /* Recursion or subroutine */
5177 case CHAR_0: case CHAR_1: case CHAR_2: case CHAR_3: case CHAR_4:
5178 case CHAR_5: case CHAR_6: case CHAR_7: case CHAR_8: case CHAR_9:
5179 {
5180 const uschar *called;
5181 terminator = CHAR_RIGHT_PARENTHESIS;
5182
5183 /* Come here from the \g<...> and \g'...' code (Oniguruma
5184 compatibility). However, the syntax has been checked to ensure that
5185 the ... are a (signed) number, so that neither ERR63 nor ERR29 will
5186 be called on this path, nor with the jump to OTHER_CHAR_AFTER_QUERY
5187 ever be taken. */
5188
5189 HANDLE_NUMERICAL_RECURSION:
5190
5191 if ((refsign = *ptr) == CHAR_PLUS)
5192 {
5193 ptr++;
5194 if ((digitab[*ptr] & ctype_digit) == 0)
5195 {
5196 *errorcodeptr = ERR63;
5197 goto FAILED;
5198 }
5199 }
5200 else if (refsign == CHAR_MINUS)
5201 {
5202 if ((digitab[ptr[1]] & ctype_digit) == 0)
5203 goto OTHER_CHAR_AFTER_QUERY;
5204 ptr++;
5205 }
5206
5207 recno = 0;
5208 while((digitab[*ptr] & ctype_digit) != 0)
5209 recno = recno * 10 + *ptr++ - CHAR_0;
5210
5211 if (*ptr != terminator)
5212 {
5213 *errorcodeptr = ERR29;
5214 goto FAILED;
5215 }
5216
5217 if (refsign == CHAR_MINUS)
5218 {
5219 if (recno == 0)
5220 {
5221 *errorcodeptr = ERR58;
5222 goto FAILED;
5223 }
5224 recno = cd->bracount - recno + 1;
5225 if (recno <= 0)
5226 {
5227 *errorcodeptr = ERR15;
5228 goto FAILED;
5229 }
5230 }
5231 else if (refsign == CHAR_PLUS)
5232 {
5233 if (recno == 0)
5234 {
5235 *errorcodeptr = ERR58;
5236 goto FAILED;
5237 }
5238 recno += cd->bracount;
5239 }
5240
5241 /* Come here from code above that handles a named recursion */
5242
5243 HANDLE_RECURSION:
5244
5245 previous = code;
5246 called = cd->start_code;
5247
5248 /* When we are actually compiling, find the bracket that is being
5249 referenced. Temporarily end the regex in case it doesn't exist before
5250 this point. If we end up with a forward reference, first check that
5251 the bracket does occur later so we can give the error (and position)
5252 now. Then remember this forward reference in the workspace so it can
5253 be filled in at the end. */
5254
5255 if (lengthptr == NULL)
5256 {
5257 *code = OP_END;
5258 if (recno != 0)
5259 called = _pcre_find_bracket(cd->start_code, utf8, recno);
5260
5261 /* Forward reference */
5262
5263 if (called == NULL)
5264 {
5265 if (find_parens(cd, NULL, recno,
5266 (options & PCRE_EXTENDED) != 0) < 0)
5267 {
5268 *errorcodeptr = ERR15;
5269 goto FAILED;
5270 }
5271
5272 /* Fudge the value of "called" so that when it is inserted as an
5273 offset below, what it actually inserted is the reference number
5274 of the group. */
5275
5276 called = cd->start_code + recno;
5277 PUTINC(cd->hwm, 0, code + 2 + LINK_SIZE - cd->start_code);
5278 }
5279
5280 /* If not a forward reference, and the subpattern is still open,
5281 this is a recursive call. We check to see if this is a left
5282 recursion that could loop for ever, and diagnose that case. */
5283
5284 else if (GET(called, 1) == 0 &&
5285 could_be_empty(called, code, bcptr, utf8, cd))
5286 {
5287 *errorcodeptr = ERR40;
5288 goto FAILED;
5289 }
5290 }
5291
5292 /* Insert the recursion/subroutine item, automatically wrapped inside
5293 "once" brackets. Set up a "previous group" length so that a
5294 subsequent quantifier will work. */
5295
5296 *code = OP_ONCE;
5297 PUT(code, 1, 2 + 2*LINK_SIZE);
5298 code += 1 + LINK_SIZE;
5299
5300 *code = OP_RECURSE;
5301 PUT(code, 1, called - cd->start_code);
5302 code += 1 + LINK_SIZE;
5303
5304 *code = OP_KET;
5305 PUT(code, 1, 2 + 2*LINK_SIZE);
5306 code += 1 + LINK_SIZE;
5307
5308 length_prevgroup = 3 + 3*LINK_SIZE;
5309 }
5310
5311 /* Can't determine a first byte now */
5312
5313 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
5314 continue;
5315
5316
5317 /* ------------------------------------------------------------ */
5318 default: /* Other characters: check option setting */
5319 OTHER_CHAR_AFTER_QUERY:
5320 set = unset = 0;
5321 optset = &set;
5322
5323 while (*ptr != CHAR_RIGHT_PARENTHESIS && *ptr != CHAR_COLON)
5324 {
5325 switch (*ptr++)
5326 {
5327 case CHAR_MINUS: optset = &unset; break;
5328
5329 case CHAR_J: /* Record that it changed in the external options */
5330 *optset |= PCRE_DUPNAMES;
5331 cd->external_flags |= PCRE_JCHANGED;
5332 break;
5333
5334 case CHAR_i: *optset |= PCRE_CASELESS; break;
5335 case CHAR_m: *optset |= PCRE_MULTILINE; break;
5336 case CHAR_s: *optset |= PCRE_DOTALL; break;
5337 case CHAR_x: *optset |= PCRE_EXTENDED; break;
5338 case CHAR_U: *optset |= PCRE_UNGREEDY; break;
5339 case CHAR_X: *optset |= PCRE_EXTRA; break;
5340
5341 default: *errorcodeptr = ERR12;
5342 ptr--; /* Correct the offset */
5343 goto FAILED;
5344 }
5345 }
5346
5347 /* Set up the changed option bits, but don't change anything yet. */
5348
5349 newoptions = (options | set) & (~unset);
5350
5351 /* If the options ended with ')' this is not the start of a nested
5352 group with option changes, so the options change at this level. If this
5353 item is right at the start of the pattern, the options can be
5354 abstracted and made external in the pre-compile phase, and ignored in
5355 the compile phase. This can be helpful when matching -- for instance in
5356 caseless checking of required bytes.
5357
5358 If the code pointer is not (cd->start_code + 1 + LINK_SIZE), we are
5359 definitely *not* at the start of the pattern because something has been
5360 compiled. In the pre-compile phase, however, the code pointer can have
5361 that value after the start, because it gets reset as code is discarded
5362 during the pre-compile. However, this can happen only at top level - if
5363 we are within parentheses, the starting BRA will still be present. At
5364 any parenthesis level, the length value can be used to test if anything
5365 has been compiled at that level. Thus, a test for both these conditions
5366 is necessary to ensure we correctly detect the start of the pattern in
5367 both phases.
5368
5369 If we are not at the pattern start, compile code to change the ims
5370 options if this setting actually changes any of them, and reset the
5371 greedy defaults and the case value for firstbyte and reqbyte. */
5372
5373 if (*ptr == CHAR_RIGHT_PARENTHESIS)
5374 {
5375 if (code == cd->start_code + 1 + LINK_SIZE &&
5376 (lengthptr == NULL || *lengthptr == 2 + 2*LINK_SIZE))
5377 {
5378 cd->external_options = newoptions;
5379 }
5380 else
5381 {
5382 if ((options & PCRE_IMS) != (newoptions & PCRE_IMS))
5383 {
5384 *code++ = OP_OPT;
5385 *code++ = newoptions & PCRE_IMS;
5386 }
5387 greedy_default = ((newoptions & PCRE_UNGREEDY) != 0);
5388 greedy_non_default = greedy_default ^ 1;
5389 req_caseopt = ((newoptions & PCRE_CASELESS) != 0)? REQ_CASELESS : 0;
5390 }
5391
5392 /* Change options at this level, and pass them back for use
5393 in subsequent branches. When not at the start of the pattern, this
5394 information is also necessary so that a resetting item can be
5395 compiled at the end of a group (if we are in a group). */
5396
5397 *optionsptr = options = newoptions;
5398 previous = NULL; /* This item can't be repeated */
5399 continue; /* It is complete */
5400 }
5401
5402 /* If the options ended with ':' we are heading into a nested group
5403 with possible change of options. Such groups are non-capturing and are
5404 not assertions of any kind. All we need to do is skip over the ':';
5405 the newoptions value is handled below. */
5406
5407 bravalue = OP_BRA;
5408 ptr++;
5409 } /* End of switch for character following (? */
5410 } /* End of (? handling */
5411
5412 /* Opening parenthesis not followed by '*' or '?'. If PCRE_NO_AUTO_CAPTURE
5413 is set, all unadorned brackets become non-capturing and behave like (?:...)
5414 brackets. */
5415
5416 else if ((options & PCRE_NO_AUTO_CAPTURE) != 0)
5417 {
5418 bravalue = OP_BRA;
5419 }
5420
5421 /* Else we have a capturing group. */
5422
5423 else
5424 {
5425 NUMBERED_GROUP:
5426 cd->bracount += 1;
5427 PUT2(code, 1+LINK_SIZE, cd->bracount);
5428 skipbytes = 2;
5429 }
5430
5431 /* Process nested bracketed regex. Assertions may not be repeated, but
5432 other kinds can be. All their opcodes are >= OP_ONCE. We copy code into a
5433 non-register variable in order to be able to pass its address because some
5434 compilers complain otherwise. Pass in a new setting for the ims options if
5435 they have changed. */
5436
5437 previous = (bravalue >= OP_ONCE)? code : NULL;
5438 *code = bravalue;
5439 tempcode = code;
5440 tempreqvary = cd->req_varyopt; /* Save value before bracket */
5441 length_prevgroup = 0; /* Initialize for pre-compile phase */
5442
5443 if (!compile_regex(
5444 newoptions, /* The complete new option state */
5445 options & PCRE_IMS, /* The previous ims option state */
5446 &tempcode, /* Where to put code (updated) */
5447 &ptr, /* Input pointer (updated) */
5448 errorcodeptr, /* Where to put an error message */
5449 (bravalue == OP_ASSERTBACK ||
5450 bravalue == OP_ASSERTBACK_NOT), /* TRUE if back assert */
5451 reset_bracount, /* True if (?| group */
5452 skipbytes, /* Skip over bracket number */
5453 &subfirstbyte, /* For possible first char */
5454 &subreqbyte, /* For possible last char */
5455 bcptr, /* Current branch chain */
5456 cd, /* Tables block */
5457 (lengthptr == NULL)? NULL : /* Actual compile phase */
5458 &length_prevgroup /* Pre-compile phase */
5459 ))
5460 goto FAILED;
5461
5462 /* At the end of compiling, code is still pointing to the start of the
5463 group, while tempcode has been updated to point past the end of the group
5464 and any option resetting that may follow it. The pattern pointer (ptr)
5465 is on the bracket. */
5466
5467 /* If this is a conditional bracket, check that there are no more than
5468 two branches in the group, or just one if it's a DEFINE group. We do this
5469 in the real compile phase, not in the pre-pass, where the whole group may
5470 not be available. */
5471
5472 if (bravalue == OP_COND && lengthptr == NULL)
5473 {
5474 uschar *tc = code;
5475 int condcount = 0;
5476
5477 do {
5478 condcount++;
5479 tc += GET(tc,1);
5480 }
5481 while (*tc != OP_KET);
5482
5483 /* A DEFINE group is never obeyed inline (the "condition" is always
5484 false). It must have only one branch. */
5485
5486 if (code[LINK_SIZE+1] == OP_DEF)
5487 {
5488 if (condcount > 1)
5489 {
5490 *errorcodeptr = ERR54;
5491 goto FAILED;
5492 }
5493 bravalue = OP_DEF; /* Just a flag to suppress char handling below */
5494 }
5495
5496 /* A "normal" conditional group. If there is just one branch, we must not
5497 make use of its firstbyte or reqbyte, because this is equivalent to an
5498 empty second branch. */
5499
5500 else
5501 {
5502 if (condcount > 2)
5503 {
5504 *errorcodeptr = ERR27;
5505 goto FAILED;
5506 }
5507 if (condcount == 1) subfirstbyte = subreqbyte = REQ_NONE;
5508 }
5509 }
5510
5511 /* Error if hit end of pattern */
5512
5513 if (*ptr != CHAR_RIGHT_PARENTHESIS)
5514 {
5515 *errorcodeptr = ERR14;
5516 goto FAILED;
5517 }
5518
5519 /* In the pre-compile phase, update the length by the length of the group,
5520 less the brackets at either end. Then reduce the compiled code to just a
5521 set of non-capturing brackets so that it doesn't use much memory if it is
5522 duplicated by a quantifier.*/
5523
5524 if (lengthptr != NULL)
5525 {
5526 if (OFLOW_MAX - *lengthptr < length_prevgroup - 2 - 2*LINK_SIZE)
5527 {
5528 *errorcodeptr = ERR20;
5529 goto FAILED;
5530 }
5531 *lengthptr += length_prevgroup - 2 - 2*LINK_SIZE;
5532 *code++ = OP_BRA;
5533 PUTINC(code, 0, 1 + LINK_SIZE);
5534 *code++ = OP_KET;
5535 PUTINC(code, 0, 1 + LINK_SIZE);
5536 break; /* No need to waste time with special character handling */
5537 }
5538
5539 /* Otherwise update the main code pointer to the end of the group. */
5540
5541 code = tempcode;
5542
5543 /* For a DEFINE group, required and first character settings are not
5544 relevant. */
5545
5546 if (bravalue == OP_DEF) break;
5547
5548 /* Handle updating of the required and first characters for other types of
5549 group. Update for normal brackets of all kinds, and conditions with two
5550 branches (see code above). If the bracket is followed by a quantifier with
5551 zero repeat, we have to back off. Hence the definition of zeroreqbyte and
5552 zerofirstbyte outside the main loop so that they can be accessed for the
5553 back off. */
5554
5555 zeroreqbyte = reqbyte;
5556 zerofirstbyte = firstbyte;
5557 groupsetfirstbyte = FALSE;
5558
5559 if (bravalue >= OP_ONCE)
5560 {
5561 /* If we have not yet set a firstbyte in this branch, take it from the
5562 subpattern, remembering that it was set here so that a repeat of more
5563 than one can replicate it as reqbyte if necessary. If the subpattern has
5564 no firstbyte, set "none" for the whole branch. In both cases, a zero
5565 repeat forces firstbyte to "none". */
5566
5567 if (firstbyte == REQ_UNSET)
5568 {
5569 if (subfirstbyte >= 0)
5570 {
5571 firstbyte = subfirstbyte;
5572 groupsetfirstbyte = TRUE;
5573 }
5574 else firstbyte = REQ_NONE;
5575 zerofirstbyte = REQ_NONE;
5576 }
5577
5578 /* If firstbyte was previously set, convert the subpattern's firstbyte
5579 into reqbyte if there wasn't one, using the vary flag that was in
5580 existence beforehand. */
5581
5582 else if (subfirstbyte >= 0 && subreqbyte < 0)
5583 subreqbyte = subfirstbyte | tempreqvary;
5584
5585 /* If the subpattern set a required byte (or set a first byte that isn't
5586 really the first byte - see above), set it. */
5587
5588 if (subreqbyte >= 0) reqbyte = subreqbyte;
5589 }
5590
5591 /* For a forward assertion, we take the reqbyte, if set. This can be
5592 helpful if the pattern that follows the assertion doesn't set a different
5593 char. For example, it's useful for /(?=abcde).+/. We can't set firstbyte
5594 for an assertion, however because it leads to incorrect effect for patterns
5595 such as /(?=a)a.+/ when the "real" "a" would then become a reqbyte instead
5596 of a firstbyte. This is overcome by a scan at the end if there's no
5597 firstbyte, looking for an asserted first char. */
5598
5599 else if (bravalue == OP_ASSERT && subreqbyte >= 0) reqbyte = subreqbyte;
5600 break; /* End of processing '(' */
5601
5602
5603 /* ===================================================================*/
5604 /* Handle metasequences introduced by \. For ones like \d, the ESC_ values
5605 are arranged to be the negation of the corresponding OP_values. For the
5606 back references, the values are ESC_REF plus the reference number. Only
5607 back references and those types that consume a character may be repeated.
5608 We can test for values between ESC_b and ESC_Z for the latter; this may
5609 have to change if any new ones are ever created. */
5610
5611 case CHAR_BACKSLASH:
5612 tempptr = ptr;
5613 c = check_escape(&ptr, errorcodeptr, cd->bracount, options, FALSE);
5614 if (*errorcodeptr != 0) goto FAILED;
5615
5616 if (c < 0)
5617 {
5618 if (-c == ESC_Q) /* Handle start of quoted string */
5619 {
5620 if (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
5621 ptr += 2; /* avoid empty string */
5622 else inescq = TRUE;
5623 continue;
5624 }
5625
5626 if (-c == ESC_E) continue; /* Perl ignores an orphan \E */
5627
5628 /* For metasequences that actually match a character, we disable the
5629 setting of a first character if it hasn't already been set. */
5630
5631 if (firstbyte == REQ_UNSET && -c > ESC_b && -c < ESC_Z)
5632 firstbyte = REQ_NONE;
5633
5634 /* Set values to reset to if this is followed by a zero repeat. */
5635
5636 zerofirstbyte = firstbyte;
5637 zeroreqbyte = reqbyte;
5638
5639 /* \g<name> or \g'name' is a subroutine call by name and \g<n> or \g'n'
5640 is a subroutine call by number (Oniguruma syntax). In fact, the value
5641 -ESC_g is returned only for these cases. So we don't need to check for <
5642 or ' if the value is -ESC_g. For the Perl syntax \g{n} the value is
5643 -ESC_REF+n, and for the Perl syntax \g{name} the result is -ESC_k (as
5644 that is a synonym for a named back reference). */
5645
5646 if (-c == ESC_g)
5647 {
5648 const uschar *p;
5649 save_hwm = cd->hwm; /* Normally this is set when '(' is read */
5650 terminator = (*(++ptr) == CHAR_LESS_THAN_SIGN)?
5651 CHAR_GREATER_THAN_SIGN : CHAR_APOSTROPHE;
5652
5653 /* These two statements stop the compiler for warning about possibly
5654 unset variables caused by the jump to HANDLE_NUMERICAL_RECURSION. In
5655 fact, because we actually check for a number below, the paths that
5656 would actually be in error are never taken. */
5657
5658 skipbytes = 0;
5659 reset_bracount = FALSE;
5660
5661 /* Test for a name */
5662
5663 if (ptr[1] != CHAR_PLUS && ptr[1] != CHAR_MINUS)
5664 {
5665 BOOL isnumber = TRUE;
5666 for (p = ptr + 1; *p != 0 && *p != terminator; p++)
5667 {
5668 if ((cd->ctypes[*p] & ctype_digit) == 0) isnumber = FALSE;
5669 if ((cd->ctypes[*p] & ctype_word) == 0) break;
5670 }
5671 if (*p != terminator)
5672 {
5673 *errorcodeptr = ERR57;
5674 break;
5675 }
5676 if (isnumber)
5677 {
5678 ptr++;
5679 goto HANDLE_NUMERICAL_RECURSION;
5680 }
5681 is_recurse = TRUE;
5682 goto NAMED_REF_OR_RECURSE;
5683 }
5684
5685 /* Test a signed number in angle brackets or quotes. */
5686
5687 p = ptr + 2;
5688 while ((digitab[*p] & ctype_digit) != 0) p++;
5689 if (*p != terminator)
5690 {
5691 *errorcodeptr = ERR57;
5692 break;
5693 }
5694 ptr++;
5695 goto HANDLE_NUMERICAL_RECURSION;
5696 }
5697
5698 /* \k<name> or \k'name' is a back reference by name (Perl syntax).
5699 We also support \k{name} (.NET syntax) */
5700
5701 if (-c == ESC_k && (ptr[1] == CHAR_LESS_THAN_SIGN ||
5702 ptr[1] == CHAR_APOSTROPHE || ptr[1] == CHAR_LEFT_CURLY_BRACKET))
5703 {
5704 is_recurse = FALSE;
5705 terminator = (*(++ptr) == CHAR_LESS_THAN_SIGN)?
5706 CHAR_GREATER_THAN_SIGN : (*ptr == CHAR_APOSTROPHE)?
5707 CHAR_APOSTROPHE : CHAR_RIGHT_CURLY_BRACKET;
5708 goto NAMED_REF_OR_RECURSE;
5709 }
5710
5711 /* Back references are handled specially; must disable firstbyte if
5712 not set to cope with cases like (?=(\w+))\1: which would otherwise set
5713 ':' later. */
5714
5715 if (-c >= ESC_REF)
5716 {
5717 open_capitem *oc;
5718 recno = -c - ESC_REF;
5719
5720 HANDLE_REFERENCE: /* Come here from named backref handling */
5721 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
5722 previous = code;
5723 *code++ = OP_REF;
5724 PUT2INC(code, 0, recno);
5725 cd->backref_map |= (recno < 32)? (1 << recno) : 1;
5726 if (recno > cd->top_backref) cd->top_backref = recno;
5727
5728 /* Check to see if this back reference is recursive, that it, it
5729 is inside the group that it references. A flag is set so that the
5730 group can be made atomic. */
5731
5732 for (oc = cd->open_caps; oc != NULL; oc = oc->next)
5733 {
5734 if (oc->number == recno)
5735 {
5736 oc->flag = TRUE;
5737 break;
5738 }
5739 }
5740 }
5741
5742 /* So are Unicode property matches, if supported. */
5743
5744 #ifdef SUPPORT_UCP
5745 else if (-c == ESC_P || -c == ESC_p)
5746 {
5747 BOOL negated;
5748 int pdata;
5749 int ptype = get_ucp(&ptr, &negated, &pdata, errorcodeptr);
5750 if (ptype < 0) goto FAILED;
5751 previous = code;
5752 *code++ = ((-c == ESC_p) != negated)? OP_PROP : OP_NOTPROP;
5753 *code++ = ptype;
5754 *code++ = pdata;
5755 }
5756 #else
5757
5758 /* If Unicode properties are not supported, \X, \P, and \p are not
5759 allowed. */
5760
5761 else if (-c == ESC_X || -c == ESC_P || -c == ESC_p)
5762 {
5763 *errorcodeptr = ERR45;
5764 goto FAILED;
5765 }
5766 #endif
5767
5768 /* For the rest (including \X when Unicode properties are supported), we
5769 can obtain the OP value by negating the escape value. */
5770
5771 else
5772 {
5773 previous = (-c > ESC_b && -c < ESC_Z)? code : NULL;
5774 *code++ = -c;
5775 }
5776 continue;
5777 }
5778
5779 /* We have a data character whose value is in c. In UTF-8 mode it may have
5780 a value > 127. We set its representation in the length/buffer, and then
5781 handle it as a data character. */
5782
5783 #ifdef SUPPORT_UTF8
5784 if (utf8 && c > 127)
5785 mclength = _pcre_ord2utf8(c, mcbuffer);
5786 else
5787 #endif
5788
5789 {
5790 mcbuffer[0] = c;
5791 mclength = 1;
5792 }
5793 goto ONE_CHAR;
5794
5795
5796 /* ===================================================================*/
5797 /* Handle a literal character. It is guaranteed not to be whitespace or #
5798 when the extended flag is set. If we are in UTF-8 mode, it may be a
5799 multi-byte literal character. */
5800
5801 default:
5802 NORMAL_CHAR:
5803 mclength = 1;
5804 mcbuffer[0] = c;
5805
5806 #ifdef SUPPORT_UTF8
5807 if (utf8 && c >= 0xc0)
5808 {
5809 while ((ptr[1] & 0xc0) == 0x80)
5810 mcbuffer[mclength++] = *(++ptr);
5811 }
5812 #endif
5813
5814 /* At this point we have the character's bytes in mcbuffer, and the length
5815 in mclength. When not in UTF-8 mode, the length is always 1. */
5816
5817 ONE_CHAR:
5818 previous = code;
5819 *code++ = ((options & PCRE_CASELESS) != 0)? OP_CHARNC : OP_CHAR;
5820 for (c = 0; c < mclength; c++) *code++ = mcbuffer[c];
5821
5822 /* Remember if \r or \n were seen */
5823
5824 if (mcbuffer[0] == CHAR_CR || mcbuffer[0] == CHAR_NL)
5825 cd->external_flags |= PCRE_HASCRORLF;
5826
5827 /* Set the first and required bytes appropriately. If no previous first
5828 byte, set it from this character, but revert to none on a zero repeat.
5829 Otherwise, leave the firstbyte value alone, and don't change it on a zero
5830 repeat. */
5831
5832 if (firstbyte == REQ_UNSET)
5833 {
5834 zerofirstbyte = REQ_NONE;
5835 zeroreqbyte = reqbyte;
5836
5837 /* If the character is more than one byte long, we can set firstbyte
5838 only if it is not to be matched caselessly. */
5839
5840 if (mclength == 1 || req_caseopt == 0)
5841 {
5842 firstbyte = mcbuffer[0] | req_caseopt;
5843 if (mclength != 1) reqbyte = code[-1] | cd->req_varyopt;
5844 }
5845 else firstbyte = reqbyte = REQ_NONE;
5846 }
5847
5848 /* firstbyte was previously set; we can set reqbyte only the length is
5849 1 or the matching is caseful. */
5850
5851 else
5852 {
5853 zerofirstbyte = firstbyte;
5854 zeroreqbyte = reqbyte;
5855 if (mclength == 1 || req_caseopt == 0)
5856 reqbyte = code[-1] | req_caseopt | cd->req_varyopt;
5857 }
5858
5859 break; /* End of literal character handling */
5860 }
5861 } /* end of big loop */
5862
5863
5864 /* Control never reaches here by falling through, only by a goto for all the
5865 error states. Pass back the position in the pattern so that it can be displayed
5866 to the user for diagnosing the error. */
5867
5868 FAILED:
5869 *ptrptr = ptr;
5870 return FALSE;
5871 }
5872
5873
5874
5875
5876 /*************************************************
5877 * Compile sequence of alternatives *
5878 *************************************************/
5879
5880 /* On entry, ptr is pointing past the bracket character, but on return it
5881 points to the closing bracket, or vertical bar, or end of string. The code
5882 variable is pointing at the byte into which the BRA operator has been stored.
5883 If the ims options are changed at the start (for a (?ims: group) or during any
5884 branch, we need to insert an OP_OPT item at the start of every following branch
5885 to ensure they get set correctly at run time, and also pass the new options
5886 into every subsequent branch compile.
5887
5888 This function is used during the pre-compile phase when we are trying to find
5889 out the amount of memory needed, as well as during the real compile phase. The
5890 value of lengthptr distinguishes the two phases.
5891
5892 Arguments:
5893 options option bits, including any changes for this subpattern
5894 oldims previous settings of ims option bits
5895 codeptr -> the address of the current code pointer
5896 ptrptr -> the address of the current pattern pointer
5897 errorcodeptr -> pointer to error code variable
5898 lookbehind TRUE if this is a lookbehind assertion
5899 reset_bracount TRUE to reset the count for each branch
5900 skipbytes skip this many bytes at start (for brackets and OP_COND)
5901 firstbyteptr place to put the first required character, or a negative number
5902 reqbyteptr place to put the last required character, or a negative number
5903 bcptr pointer to the chain of currently open branches
5904 cd points to the data block with tables pointers etc.
5905 lengthptr NULL during the real compile phase
5906 points to length accumulator during pre-compile phase
5907
5908 Returns: TRUE on success
5909 */
5910
5911 static BOOL
5912 compile_regex(int options, int oldims, uschar **codeptr, const uschar **ptrptr,
5913 int *errorcodeptr, BOOL lookbehind, BOOL reset_bracount, int skipbytes,
5914 int *firstbyteptr, int *reqbyteptr, branch_chain *bcptr, compile_data *cd,
5915 int *lengthptr)
5916 {
5917 const uschar *ptr = *ptrptr;
5918 uschar *code = *codeptr;
5919 uschar *last_branch = code;
5920 uschar *start_bracket = code;
5921 uschar *reverse_count = NULL;
5922 open_capitem capitem;
5923 int capnumber = 0;
5924 int firstbyte, reqbyte;
5925 int branchfirstbyte, branchreqbyte;
5926 int length;
5927 int orig_bracount;
5928 int max_bracount;
5929 int old_external_options = cd->external_options;
5930 branch_chain bc;
5931
5932 bc.outer = bcptr;
5933 bc.current_branch = code;
5934
5935 firstbyte = reqbyte = REQ_UNSET;
5936
5937 /* Accumulate the length for use in the pre-compile phase. Start with the
5938 length of the BRA and KET and any extra bytes that are required at the
5939 beginning. We accumulate in a local variable to save frequent testing of
5940 lenthptr for NULL. We cannot do this by looking at the value of code at the
5941 start and end of each alternative, because compiled items are discarded during
5942 the pre-compile phase so that the work space is not exceeded. */
5943
5944 length = 2 + 2*LINK_SIZE + skipbytes;
5945
5946 /* WARNING: If the above line is changed for any reason, you must also change
5947 the code that abstracts option settings at the start of the pattern and makes
5948 them global. It tests the value of length for (2 + 2*LINK_SIZE) in the
5949 pre-compile phase to find out whether anything has yet been compiled or not. */
5950
5951 /* If this is a capturing subpattern, add to the chain of open capturing items
5952 so that we can detect them if (*ACCEPT) is encountered. This is also used to
5953 detect groups that contain recursive back references to themselves. */
5954
5955 if (*code == OP_CBRA)
5956 {
5957 capnumber = GET2(code, 1 + LINK_SIZE);
5958 capitem.number = capnumber;
5959 capitem.next = cd->open_caps;
5960 capitem.flag = FALSE;
5961 cd->open_caps = &capitem;
5962 }
5963
5964 /* Offset is set zero to mark that this bracket is still open */
5965
5966 PUT(code, 1, 0);
5967 code += 1 + LINK_SIZE + skipbytes;
5968
5969 /* Loop for each alternative branch */
5970
5971 orig_bracount = max_bracount = cd->bracount;
5972 for (;;)
5973 {
5974 /* For a (?| group, reset the capturing bracket count so that each branch
5975 uses the same numbers. */
5976
5977 if (reset_bracount) cd->bracount = orig_bracount;
5978
5979 /* Handle a change of ims options at the start of the branch */
5980
5981 if ((options & PCRE_IMS) != oldims)
5982 {
5983 *code++ = OP_OPT;
5984 *code++ = options & PCRE_IMS;
5985 length += 2;
5986 }
5987
5988 /* Set up dummy OP_REVERSE if lookbehind assertion */
5989
5990 if (lookbehind)
5991 {
5992 *code++ = OP_REVERSE;
5993 reverse_count = code;
5994 PUTINC(code, 0, 0);
5995 length += 1 + LINK_SIZE;
5996 }
5997
5998 /* Now compile the branch; in the pre-compile phase its length gets added
5999 into the length. */
6000
6001 if (!compile_branch(&options, &code, &ptr, errorcodeptr, &branchfirstbyte,
6002 &branchreqbyte, &bc, cd, (lengthptr == NULL)? NULL : &length))
6003 {
6004 *ptrptr = ptr;
6005 return FALSE;
6006 }
6007
6008 /* If the external options have changed during this branch, it means that we
6009 are at the top level, and a leading option setting has been encountered. We
6010 need to re-set the original option values to take account of this so that,
6011 during the pre-compile phase, we know to allow for a re-set at the start of
6012 subsequent branches. */
6013
6014 if (old_external_options != cd->external_options)
6015 oldims = cd->external_options & PCRE_IMS;
6016
6017 /* Keep the highest bracket count in case (?| was used and some branch
6018 has fewer than the rest. */
6019
6020 if (cd->bracount > max_bracount) max_bracount = cd->bracount;
6021
6022 /* In the real compile phase, there is some post-processing to be done. */
6023
6024 if (lengthptr == NULL)
6025 {
6026 /* If this is the first branch, the firstbyte and reqbyte values for the
6027 branch become the values for the regex. */
6028
6029 if (*last_branch != OP_ALT)
6030 {
6031 firstbyte = branchfirstbyte;
6032 reqbyte = branchreqbyte;
6033 }
6034
6035 /* If this is not the first branch, the first char and reqbyte have to
6036 match the values from all the previous branches, except that if the
6037 previous value for reqbyte didn't have REQ_VARY set, it can still match,
6038 and we set REQ_VARY for the regex. */
6039
6040 else
6041 {
6042 /* If we previously had a firstbyte, but it doesn't match the new branch,
6043 we have to abandon the firstbyte for the regex, but if there was
6044 previously no reqbyte, it takes on the value of the old firstbyte. */
6045
6046 if (firstbyte >= 0 && firstbyte != branchfirstbyte)
6047 {
6048 if (reqbyte < 0) reqbyte = firstbyte;
6049 firstbyte = REQ_NONE;
6050 }
6051
6052 /* If we (now or from before) have no firstbyte, a firstbyte from the
6053 branch becomes a reqbyte if there isn't a branch reqbyte. */
6054
6055 if (firstbyte < 0 && branchfirstbyte >= 0 && branchreqbyte < 0)
6056 branchreqbyte = branchfirstbyte;
6057
6058 /* Now ensure that the reqbytes match */
6059
6060 if ((reqbyte & ~REQ_VARY) != (branchreqbyte & ~REQ_VARY))
6061 reqbyte = REQ_NONE;
6062 else reqbyte |= branchreqbyte; /* To "or" REQ_VARY */
6063 }
6064
6065 /* If lookbehind, check that this branch matches a fixed-length string, and
6066 put the length into the OP_REVERSE item. Temporarily mark the end of the
6067 branch with OP_END. If the branch contains OP_RECURSE, the result is -3
6068 because there may be forward references that we can't check here. Set a
6069 flag to cause another lookbehind check at the end. Why not do it all at the
6070 end? Because common, erroneous checks are picked up here and the offset of
6071 the problem can be shown. */
6072
6073 if (lookbehind)
6074 {
6075 int fixed_length;
6076 *code = OP_END;
6077 fixed_length = find_fixedlength(last_branch, options, FALSE, cd);
6078 DPRINTF(("fixed length = %d\n", fixed_length));
6079 if (fixed_length == -3)
6080 {
6081 cd->check_lookbehind = TRUE;
6082 }
6083 else if (fixed_length < 0)
6084 {
6085 *errorcodeptr = (fixed_length == -2)? ERR36 : ERR25;
6086 *ptrptr = ptr;
6087 return FALSE;
6088 }
6089 else { PUT(reverse_count, 0, fixed_length); }
6090 }
6091 }
6092
6093 /* Reached end of expression, either ')' or end of pattern. In the real
6094 compile phase, go back through the alternative branches and reverse the chain
6095 of offsets, with the field in the BRA item now becoming an offset to the
6096 first alternative. If there are no alternatives, it points to the end of the
6097 group. The length in the terminating ket is always the length of the whole
6098 bracketed item. If any of the ims options were changed inside the group,
6099 compile a resetting op-code following, except at the very end of the pattern.
6100 Return leaving the pointer at the terminating char. */
6101
6102 if (*ptr != CHAR_VERTICAL_LINE)
6103 {
6104 if (lengthptr == NULL)
6105 {
6106 int branch_length = code - last_branch;
6107 do
6108 {
6109 int prev_length = GET(last_branch, 1);
6110 PUT(last_branch, 1, branch_length);
6111 branch_length = prev_length;
6112 last_branch -= branch_length;
6113 }
6114 while (branch_length > 0);
6115 }
6116
6117 /* Fill in the ket */
6118
6119 *code = OP_KET;
6120 PUT(code, 1, code - start_bracket);
6121 code += 1 + LINK_SIZE;
6122
6123 /* If it was a capturing subpattern, check to see if it contained any
6124 recursive back references. If so, we must wrap it in atomic brackets.
6125 In any event, remove the block from the chain. */
6126
6127 if (capnumber > 0)
6128 {
6129 if (cd->open_caps->flag)
6130 {
6131 memmove(start_bracket + 1 + LINK_SIZE, start_bracket,
6132 code - start_bracket);
6133 *start_bracket = OP_ONCE;
6134 code += 1 + LINK_SIZE;
6135 PUT(start_bracket, 1, code - start_bracket);
6136 *code = OP_KET;
6137 PUT(code, 1, code - start_bracket);
6138 code += 1 + LINK_SIZE;
6139 length += 2 + 2*LINK_SIZE;
6140 }
6141 cd->open_caps = cd->open_caps->next;
6142 }
6143
6144 /* Reset options if needed. */
6145
6146 if ((options & PCRE_IMS) != oldims && *ptr == CHAR_RIGHT_PARENTHESIS)
6147 {
6148 *code++ = OP_OPT;
6149 *code++ = oldims;
6150 length += 2;
6151 }
6152
6153 /* Retain the highest bracket number, in case resetting was used. */
6154
6155 cd->bracount = max_bracount;
6156
6157 /* Set values to pass back */
6158
6159 *codeptr = code;
6160 *ptrptr = ptr;
6161 *firstbyteptr = firstbyte;
6162 *reqbyteptr = reqbyte;
6163 if (lengthptr != NULL)
6164 {
6165 if (OFLOW_MAX - *lengthptr < length)
6166 {
6167 *errorcodeptr = ERR20;
6168 return FALSE;
6169 }
6170 *lengthptr += length;
6171 }
6172 return TRUE;
6173 }
6174
6175 /* Another branch follows. In the pre-compile phase, we can move the code
6176 pointer back to where it was for the start of the first branch. (That is,
6177 pretend that each branch is the only one.)
6178
6179 In the real compile phase, insert an ALT node. Its length field points back
6180 to the previous branch while the bracket remains open. At the end the chain
6181 is reversed. It's done like this so that the start of the bracket has a
6182 zero offset until it is closed, making it possible to detect recursion. */
6183
6184 if (lengthptr != NULL)
6185 {
6186 code = *codeptr + 1 + LINK_SIZE + skipbytes;
6187 length += 1 + LINK_SIZE;
6188 }
6189 else
6190 {
6191 *code = OP_ALT;
6192 PUT(code, 1, code - last_branch);
6193 bc.current_branch = last_branch = code;
6194 code += 1 + LINK_SIZE;
6195 }
6196
6197 ptr++;
6198 }
6199 /* Control never reaches here */
6200 }
6201
6202
6203
6204
6205 /*************************************************
6206 * Check for anchored expression *
6207 *************************************************/
6208
6209 /* Try to find out if this is an anchored regular expression. Consider each
6210 alternative branch. If they all start with OP_SOD or OP_CIRC, or with a bracket
6211 all of whose alternatives start with OP_SOD or OP_CIRC (recurse ad lib), then
6212 it's anchored. However, if this is a multiline pattern, then only OP_SOD
6213 counts, since OP_CIRC can match in the middle.
6214
6215 We can also consider a regex to be anchored if OP_SOM starts all its branches.
6216 This is the code for \G, which means "match at start of match position, taking
6217 into account the match offset".
6218
6219 A branch is also implicitly anchored if it starts with .* and DOTALL is set,
6220 because that will try the rest of the pattern at all possible matching points,
6221 so there is no point trying again.... er ....
6222
6223 .... except when the .* appears inside capturing parentheses, and there is a
6224 subsequent back reference to those parentheses. We haven't enough information
6225 to catch that case precisely.
6226
6227 At first, the best we could do was to detect when .* was in capturing brackets
6228 and the highest back reference was greater than or equal to that level.
6229 However, by keeping a bitmap of the first 31 back references, we can catch some
6230 of the more common cases more precisely.
6231
6232 Arguments:
6233 code points to start of expression (the bracket)
6234 options points to the options setting
6235 bracket_map a bitmap of which brackets we are inside while testing; this
6236 handles up to substring 31; after that we just have to take
6237 the less precise approach
6238 backref_map the back reference bitmap
6239
6240 Returns: TRUE or FALSE
6241 */
6242
6243 static BOOL
6244 is_anchored(register const uschar *code, int *options, unsigned int bracket_map,
6245 unsigned int backref_map)
6246 {
6247 do {
6248 const uschar *scode = first_significant_code(code + _pcre_OP_lengths[*code],
6249 options, PCRE_MULTILINE, FALSE);
6250 register int op = *scode;
6251
6252 /* Non-capturing brackets */
6253
6254 if (op == OP_BRA)
6255 {
6256 if (!is_anchored(scode, options, bracket_map, backref_map)) return FALSE;
6257 }
6258
6259 /* Capturing brackets */
6260
6261 else if (op == OP_CBRA)
6262 {
6263 int n = GET2(scode, 1+LINK_SIZE);
6264 int new_map = bracket_map | ((n < 32)? (1 << n) : 1);
6265 if (!is_anchored(scode, options, new_map, backref_map)) return FALSE;
6266 }
6267
6268 /* Other brackets */
6269
6270 else if (op == OP_ASSERT || op == OP_ONCE || op == OP_COND)
6271 {
6272 if (!is_anchored(scode, options, bracket_map, backref_map)) return FALSE;
6273 }
6274
6275 /* .* is not anchored unless DOTALL is set (which generates OP_ALLANY) and
6276 it isn't in brackets that are or may be referenced. */
6277
6278 else if ((op == OP_TYPESTAR || op == OP_TYPEMINSTAR ||
6279 op == OP_TYPEPOSSTAR))
6280 {
6281 if (scode[1] != OP_ALLANY || (bracket_map & backref_map) != 0)
6282 return FALSE;
6283 }
6284
6285 /* Check for explicit anchoring */
6286
6287 else if (op != OP_SOD && op != OP_SOM &&
6288 ((*options & PCRE_MULTILINE) != 0 || op != OP_CIRC))
6289 return FALSE;
6290 code += GET(code, 1);
6291 }
6292 while (*code == OP_ALT); /* Loop for each alternative */
6293 return TRUE;
6294 }
6295
6296
6297
6298 /*************************************************
6299 * Check for starting with ^ or .* *
6300 *************************************************/
6301
6302 /* This is called to find out if every branch starts with ^ or .* so that
6303 "first char" processing can be done to speed things up in multiline
6304 matching and for non-DOTALL patterns that start with .* (which must start at
6305 the beginning or after \n). As in the case of is_anchored() (see above), we
6306 have to take account of back references to capturing brackets that contain .*
6307 because in that case we can't make the assumption.
6308
6309 Arguments:
6310 code points to start of expression (the bracket)
6311 bracket_map a bitmap of which brackets we are inside while testing; this
6312 handles up to substring 31; after that we just have to take
6313 the less precise approach
6314 backref_map the back reference bitmap
6315
6316 Returns: TRUE or FALSE
6317 */
6318
6319 static BOOL
6320 is_startline(const uschar *code, unsigned int bracket_map,
6321 unsigned int backref_map)
6322 {
6323 do {
6324 const uschar *scode = first_significant_code(code + _pcre_OP_lengths[*code],
6325 NULL, 0, FALSE);
6326 register int op = *scode;
6327
6328 /* If we are at the start of a conditional assertion group, *both* the
6329 conditional assertion *and* what follows the condition must satisfy the test
6330 for start of line. Other kinds of condition fail. Note that there may be an
6331 auto-callout at the start of a condition. */
6332
6333 if (op == OP_COND)
6334 {
6335 scode += 1 + LINK_SIZE;
6336 if (*scode == OP_CALLOUT) scode += _pcre_OP_lengths[OP_CALLOUT];
6337 switch (*scode)
6338 {
6339 case OP_CREF:
6340 case OP_NCREF:
6341 case OP_RREF:
6342 case OP_NRREF:
6343 case OP_DEF:
6344 return FALSE;
6345
6346 default: /* Assertion */
6347 if (!is_startline(scode, bracket_map, backref_map)) return FALSE;
6348 do scode += GET(scode, 1); while (*scode == OP_ALT);
6349 scode += 1 + LINK_SIZE;
6350 break;
6351 }
6352 scode = first_significant_code(scode, NULL, 0, FALSE);
6353 op = *scode;
6354 }
6355
6356 /* Non-capturing brackets */
6357
6358 if (op == OP_BRA)
6359 {
6360 if (!is_startline(scode, bracket_map, backref_map)) return FALSE;
6361 }
6362
6363 /* Capturing brackets */
6364
6365 else if (op == OP_CBRA)
6366 {
6367 int n = GET2(scode, 1+LINK_SIZE);
6368 int new_map = bracket_map | ((n < 32)? (1 << n) : 1);
6369 if (!is_startline(scode, new_map, backref_map)) return FALSE;
6370 }
6371
6372 /* Other brackets */
6373
6374 else if (op == OP_ASSERT || op == OP_ONCE)
6375 {
6376 if (!is_startline(scode, bracket_map, backref_map)) return FALSE;
6377 }
6378
6379 /* .* means "start at start or after \n" if it isn't in brackets that
6380 may be referenced. */
6381
6382 else if (op == OP_TYPESTAR || op == OP_TYPEMINSTAR || op == OP_TYPEPOSSTAR)
6383 {
6384 if (scode[1] != OP_ANY || (bracket_map & backref_map) != 0) return FALSE;
6385 }
6386
6387 /* Check for explicit circumflex */
6388
6389 else if (op != OP_CIRC) return FALSE;
6390
6391 /* Move on to the next alternative */
6392
6393 code += GET(code, 1);
6394 }
6395 while (*code == OP_ALT); /* Loop for each alternative */
6396 return TRUE;
6397 }
6398
6399
6400
6401 /*************************************************
6402 * Check for asserted fixed first char *
6403 *************************************************/
6404
6405 /* During compilation, the "first char" settings from forward assertions are
6406 discarded, because they can cause conflicts with actual literals that follow.
6407 However, if we end up without a first char setting for an unanchored pattern,
6408 it is worth scanning the regex to see if there is an initial asserted first
6409 char. If all branches start with the same asserted char, or with a bracket all
6410 of whose alternatives start with the same asserted char (recurse ad lib), then
6411 we return that char, otherwise -1.
6412
6413 Arguments:
6414 code points to start of expression (the bracket)
6415 options pointer to the options (used to check casing changes)
6416 inassert TRUE if in an assertion
6417
6418 Returns: -1 or the fixed first char
6419 */
6420
6421 static int
6422 find_firstassertedchar(const uschar *code, int *options, BOOL inassert)
6423 {
6424 register int c = -1;
6425 do {
6426 int d;
6427 const uschar *scode =
6428 first_significant_code(code + 1+LINK_SIZE, options, PCRE_CASELESS, TRUE);
6429 register int op = *scode;
6430
6431 switch(op)
6432 {
6433 default:
6434 return -1;
6435
6436 case OP_BRA:
6437 case OP_CBRA:
6438 case OP_ASSERT:
6439 case OP_ONCE:
6440 case OP_COND:
6441 if ((d = find_firstassertedchar(scode, options, op == OP_ASSERT)) < 0)
6442 return -1;
6443 if (c < 0) c = d; else if (c != d) return -1;
6444 break;
6445
6446 case OP_EXACT: /* Fall through */
6447 scode += 2;
6448
6449 case OP_CHAR:
6450 case OP_CHARNC:
6451 case OP_PLUS:
6452 case OP_MINPLUS:
6453 case OP_POSPLUS:
6454 if (!inassert) return -1;
6455 if (c < 0)
6456 {
6457 c = scode[1];
6458 if ((*options & PCRE_CASELESS) != 0) c |= REQ_CASELESS;
6459 }
6460 else if (c != scode[1]) return -1;
6461 break;
6462 }
6463
6464 code += GET(code, 1);
6465 }
6466 while (*code == OP_ALT);
6467 return c;
6468 }
6469
6470
6471
6472 /*************************************************
6473 * Compile a Regular Expression *
6474 *************************************************/
6475
6476 /* This function takes a string and returns a pointer to a block of store
6477 holding a compiled version of the expression. The original API for this
6478 function had no error code return variable; it is retained for backwards
6479 compatibility. The new function is given a new name.
6480
6481 Arguments:
6482 pattern the regular expression
6483 options various option bits
6484 errorcodeptr pointer to error code variable (pcre_compile2() only)
6485 can be NULL if you don't want a code value
6486 errorptr pointer to pointer to error text
6487 erroroffset ptr offset in pattern where error was detected
6488 tables pointer to character tables or NULL
6489
6490 Returns: pointer to compiled data block, or NULL on error,
6491 with errorptr and erroroffset set
6492 */
6493
6494 PCRE_EXP_DEFN pcre * PCRE_CALL_CONVENTION
6495 pcre_compile(const char *pattern, int options, const char **errorptr,
6496 int *erroroffset, const unsigned char *tables)
6497 {
6498 return pcre_compile2(pattern, options, NULL, errorptr, erroroffset, tables);
6499 }
6500
6501
6502 PCRE_EXP_DEFN pcre * PCRE_CALL_CONVENTION
6503 pcre_compile2(const char *pattern, int options, int *errorcodeptr,
6504 const char **errorptr, int *erroroffset, const unsigned char *tables)
6505 {
6506 real_pcre *re;
6507 int length = 1; /* For final END opcode */
6508 int firstbyte, reqbyte, newline;
6509 int errorcode = 0;
6510 int skipatstart = 0;
6511 BOOL utf8 = (options & PCRE_UTF8) != 0;
6512 size_t size;
6513 uschar *code;
6514 const uschar *codestart;
6515 const uschar *ptr;
6516 compile_data compile_block;
6517 compile_data *cd = &compile_block;
6518
6519 /* This space is used for "compiling" into during the first phase, when we are
6520 computing the amount of memory that is needed. Compiled items are thrown away
6521 as soon as possible, so that a fairly large buffer should be sufficient for
6522 this purpose. The same space is used in the second phase for remembering where
6523 to fill in forward references to subpatterns. */
6524
6525 uschar cworkspace[COMPILE_WORK_SIZE];
6526
6527 /* Set this early so that early errors get offset 0. */
6528
6529 ptr = (const uschar *)pattern;
6530
6531 /* We can't pass back an error message if errorptr is NULL; I guess the best we
6532 can do is just return NULL, but we can set a code value if there is a code
6533 pointer. */
6534
6535 if (errorptr == NULL)
6536 {
6537 if (errorcodeptr != NULL) *errorcodeptr = 99;
6538 return NULL;
6539 }
6540
6541 *errorptr = NULL;
6542 if (errorcodeptr != NULL) *errorcodeptr = ERR0;
6543
6544 /* However, we can give a message for this error */
6545
6546 if (erroroffset == NULL)
6547 {
6548 errorcode = ERR16;
6549 goto PCRE_EARLY_ERROR_RETURN2;
6550 }
6551
6552 *erroroffset = 0;
6553
6554 /* Set up pointers to the individual character tables */
6555
6556 if (tables == NULL) tables = _pcre_default_tables;
6557 cd->lcc = tables + lcc_offset;
6558 cd->fcc = tables + fcc_offset;
6559 cd->cbits = tables + cbits_offset;
6560 cd->ctypes = tables + ctypes_offset;
6561
6562 /* Check that all undefined public option bits are zero */
6563
6564 if ((options & ~PUBLIC_COMPILE_OPTIONS) != 0)
6565 {
6566 errorcode = ERR17;
6567 goto PCRE_EARLY_ERROR_RETURN;
6568 }
6569
6570 /* Check for global one-time settings at the start of the pattern, and remember
6571 the offset for later. */
6572
6573 while (ptr[skipatstart] == CHAR_LEFT_PARENTHESIS &&
6574 ptr[skipatstart+1] == CHAR_ASTERISK)
6575 {
6576 int newnl = 0;
6577 int newbsr = 0;
6578
6579 if (strncmp((char *)(ptr+skipatstart+2), STRING_UTF8_RIGHTPAR, 5) == 0)
6580 { skipatstart += 7; options |= PCRE_UTF8; continue; }
6581
6582 if (strncmp((char *)(ptr+skipatstart+2), STRING_CR_RIGHTPAR, 3) == 0)
6583 { skipatstart += 5; newnl = PCRE_NEWLINE_CR; }
6584 else if (strncmp((char *)(ptr+skipatstart+2), STRING_LF_RIGHTPAR, 3) == 0)
6585 { skipatstart += 5; newnl = PCRE_NEWLINE_LF; }
6586 else if (strncmp((char *)(ptr+skipatstart+2), STRING_CRLF_RIGHTPAR, 5) == 0)
6587 { skipatstart += 7; newnl = PCRE_NEWLINE_CR + PCRE_NEWLINE_LF; }
6588 else if (strncmp((char *)(ptr+skipatstart+2), STRING_ANY_RIGHTPAR, 4) == 0)
6589 { skipatstart += 6; newnl = PCRE_NEWLINE_ANY; }
6590 else if (strncmp((char *)(ptr+skipatstart+2), STRING_ANYCRLF_RIGHTPAR, 8) == 0)
6591 { skipatstart += 10; newnl = PCRE_NEWLINE_ANYCRLF; }
6592
6593 else if (strncmp((char *)(ptr+skipatstart+2), STRING_BSR_ANYCRLF_RIGHTPAR, 12) == 0)
6594 { skipatstart += 14; newbsr = PCRE_BSR_ANYCRLF; }
6595 else if (strncmp((char *)(ptr+skipatstart+2), STRING_BSR_UNICODE_RIGHTPAR, 12) == 0)
6596 { skipatstart += 14; newbsr = PCRE_BSR_UNICODE; }
6597
6598 if (newnl != 0)
6599 options = (options & ~PCRE_NEWLINE_BITS) | newnl;
6600 else if (newbsr != 0)
6601 options = (options & ~(PCRE_BSR_ANYCRLF|PCRE_BSR_UNICODE)) | newbsr;
6602 else break;
6603 }
6604
6605 /* Can't support UTF8 unless PCRE has been compiled to include the code. */
6606
6607 #ifdef SUPPORT_UTF8
6608 if (utf8 && (options & PCRE_NO_UTF8_CHECK) == 0 &&
6609 (*erroroffset = _pcre_valid_utf8((USPTR)pattern, -1)) >= 0)
6610 {
6611 errorcode = ERR44;
6612 goto PCRE_EARLY_ERROR_RETURN2;
6613 }
6614 #else
6615 if (utf8)
6616 {
6617 errorcode = ERR32;
6618 goto PCRE_EARLY_ERROR_RETURN;
6619 }
6620 #endif
6621
6622 /* Check validity of \R options. */
6623
6624 switch (options & (PCRE_BSR_ANYCRLF|PCRE_BSR_UNICODE))
6625 {
6626 case 0:
6627 case PCRE_BSR_ANYCRLF:
6628 case PCRE_BSR_UNICODE:
6629 break;
6630 default: errorcode = ERR56; goto PCRE_EARLY_ERROR_RETURN;
6631 }
6632
6633 /* Handle different types of newline. The three bits give seven cases. The
6634 current code allows for fixed one- or two-byte sequences, plus "any" and
6635 "anycrlf". */
6636
6637 switch (options & PCRE_NEWLINE_BITS)
6638 {
6639 case 0: newline = NEWLINE; break; /* Build-time default */
6640 case PCRE_NEWLINE_CR: newline = CHAR_CR; break;
6641 case PCRE_NEWLINE_LF: newline = CHAR_NL; break;
6642 case PCRE_NEWLINE_CR+
6643 PCRE_NEWLINE_LF: newline = (CHAR_CR << 8) | CHAR_NL; break;
6644 case PCRE_NEWLINE_ANY: newline = -1; break;
6645 case PCRE_NEWLINE_ANYCRLF: newline = -2; break;
6646 default: errorcode = ERR56; goto PCRE_EARLY_ERROR_RETURN;
6647 }
6648
6649 if (newline == -2)
6650 {
6651 cd->nltype = NLTYPE_ANYCRLF;
6652 }
6653 else if (newline < 0)
6654 {
6655 cd->nltype = NLTYPE_ANY;
6656 }
6657 else
6658 {
6659 cd->nltype = NLTYPE_FIXED;
6660 if (newline > 255)
6661 {
6662 cd->nllen = 2;
6663 cd->nl[0] = (newline >> 8) & 255;
6664 cd->nl[1] = newline & 255;
6665 }
6666 else
6667 {