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

Contents of /code/trunk/pcre_compile.c

Parent Directory Parent Directory | Revision Log Revision Log


Revision 454 - (show annotations) (download)
Tue Sep 22 09:42:11 2009 UTC (5 years, 2 months ago) by ph10
File MIME type: text/plain
File size: 217155 byte(s)
Allow fixed-length subroutine calls in lookbehinds.

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) 1997-2009 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 DEBUG is defined, we need the pcre_printint() function, which is also
57 used by pcretest. DEBUG is not defined when building a production library. */
58
59 #ifdef DEBUG
60 #include "pcre_printint.src"
61 #endif
62
63
64 /* Macro for setting individual bits in class bitmaps. */
65
66 #define SETBIT(a,b) a[b/8] |= (1 << (b%8))
67
68 /* Maximum length value to check against when making sure that the integer that
69 holds the compiled pattern length does not overflow. We make it a bit less than
70 INT_MAX to allow for adding in group terminating bytes, so that we don't have
71 to check them every time. */
72
73 #define OFLOW_MAX (INT_MAX - 20)
74
75
76 /*************************************************
77 * Code parameters and static tables *
78 *************************************************/
79
80 /* This value specifies the size of stack workspace that is used during the
81 first pre-compile phase that determines how much memory is required. The regex
82 is partly compiled into this space, but the compiled parts are discarded as
83 soon as they can be, so that hopefully there will never be an overrun. The code
84 does, however, check for an overrun. The largest amount I've seen used is 218,
85 so this number is very generous.
86
87 The same workspace is used during the second, actual compile phase for
88 remembering forward references to groups so that they can be filled in at the
89 end. Each entry in this list occupies LINK_SIZE bytes, so even when LINK_SIZE
90 is 4 there is plenty of room. */
91
92 #define COMPILE_WORK_SIZE (4096)
93
94
95 /* Table for handling escaped characters in the range '0'-'z'. Positive returns
96 are simple data values; negative values are for special things like \d and so
97 on. Zero means further processing is needed (for things like \x), or the escape
98 is invalid. */
99
100 #ifndef EBCDIC
101
102 /* This is the "normal" table for ASCII systems or for EBCDIC systems running
103 in UTF-8 mode. */
104
105 static const short int escapes[] = {
106 0, 0,
107 0, 0,
108 0, 0,
109 0, 0,
110 0, 0,
111 CHAR_COLON, CHAR_SEMICOLON,
112 CHAR_LESS_THAN_SIGN, CHAR_EQUALS_SIGN,
113 CHAR_GREATER_THAN_SIGN, CHAR_QUESTION_MARK,
114 CHAR_COMMERCIAL_AT, -ESC_A,
115 -ESC_B, -ESC_C,
116 -ESC_D, -ESC_E,
117 0, -ESC_G,
118 -ESC_H, 0,
119 0, -ESC_K,
120 0, 0,
121 0, 0,
122 -ESC_P, -ESC_Q,
123 -ESC_R, -ESC_S,
124 0, 0,
125 -ESC_V, -ESC_W,
126 -ESC_X, 0,
127 -ESC_Z, CHAR_LEFT_SQUARE_BRACKET,
128 CHAR_BACKSLASH, CHAR_RIGHT_SQUARE_BRACKET,
129 CHAR_CIRCUMFLEX_ACCENT, CHAR_UNDERSCORE,
130 CHAR_GRAVE_ACCENT, 7,
131 -ESC_b, 0,
132 -ESC_d, ESC_e,
133 ESC_f, 0,
134 -ESC_h, 0,
135 0, -ESC_k,
136 0, 0,
137 ESC_n, 0,
138 -ESC_p, 0,
139 ESC_r, -ESC_s,
140 ESC_tee, 0,
141 -ESC_v, -ESC_w,
142 0, 0,
143 -ESC_z
144 };
145
146 #else
147
148 /* This is the "abnormal" table for EBCDIC systems without UTF-8 support. */
149
150 static const short int escapes[] = {
151 /* 48 */ 0, 0, 0, '.', '<', '(', '+', '|',
152 /* 50 */ '&', 0, 0, 0, 0, 0, 0, 0,
153 /* 58 */ 0, 0, '!', '$', '*', ')', ';', '~',
154 /* 60 */ '-', '/', 0, 0, 0, 0, 0, 0,
155 /* 68 */ 0, 0, '|', ',', '%', '_', '>', '?',
156 /* 70 */ 0, 0, 0, 0, 0, 0, 0, 0,
157 /* 78 */ 0, '`', ':', '#', '@', '\'', '=', '"',
158 /* 80 */ 0, 7, -ESC_b, 0, -ESC_d, ESC_e, ESC_f, 0,
159 /* 88 */-ESC_h, 0, 0, '{', 0, 0, 0, 0,
160 /* 90 */ 0, 0, -ESC_k, 'l', 0, ESC_n, 0, -ESC_p,
161 /* 98 */ 0, ESC_r, 0, '}', 0, 0, 0, 0,
162 /* A0 */ 0, '~', -ESC_s, ESC_tee, 0,-ESC_v, -ESC_w, 0,
163 /* A8 */ 0,-ESC_z, 0, 0, 0, '[', 0, 0,
164 /* B0 */ 0, 0, 0, 0, 0, 0, 0, 0,
165 /* B8 */ 0, 0, 0, 0, 0, ']', '=', '-',
166 /* C0 */ '{',-ESC_A, -ESC_B, -ESC_C, -ESC_D,-ESC_E, 0, -ESC_G,
167 /* C8 */-ESC_H, 0, 0, 0, 0, 0, 0, 0,
168 /* D0 */ '}', 0, -ESC_K, 0, 0, 0, 0, -ESC_P,
169 /* D8 */-ESC_Q,-ESC_R, 0, 0, 0, 0, 0, 0,
170 /* E0 */ '\\', 0, -ESC_S, 0, 0,-ESC_V, -ESC_W, -ESC_X,
171 /* E8 */ 0,-ESC_Z, 0, 0, 0, 0, 0, 0,
172 /* F0 */ 0, 0, 0, 0, 0, 0, 0, 0,
173 /* F8 */ 0, 0, 0, 0, 0, 0, 0, 0
174 };
175 #endif
176
177
178 /* Table of special "verbs" like (*PRUNE). This is a short table, so it is
179 searched linearly. Put all the names into a single string, in order to reduce
180 the number of relocations when a shared library is dynamically linked. The
181 string is built from string macros so that it works in UTF-8 mode on EBCDIC
182 platforms. */
183
184 typedef struct verbitem {
185 int len;
186 int op;
187 } verbitem;
188
189 static const char verbnames[] =
190 STRING_ACCEPT0
191 STRING_COMMIT0
192 STRING_F0
193 STRING_FAIL0
194 STRING_PRUNE0
195 STRING_SKIP0
196 STRING_THEN;
197
198 static const verbitem verbs[] = {
199 { 6, OP_ACCEPT },
200 { 6, OP_COMMIT },
201 { 1, OP_FAIL },
202 { 4, OP_FAIL },
203 { 5, OP_PRUNE },
204 { 4, OP_SKIP },
205 { 4, OP_THEN }
206 };
207
208 static const int verbcount = sizeof(verbs)/sizeof(verbitem);
209
210
211 /* Tables of names of POSIX character classes and their lengths. The names are
212 now all in a single string, to reduce the number of relocations when a shared
213 library is dynamically loaded. The list of lengths is terminated by a zero
214 length entry. The first three must be alpha, lower, upper, as this is assumed
215 for handling case independence. */
216
217 static const char posix_names[] =
218 STRING_alpha0 STRING_lower0 STRING_upper0 STRING_alnum0
219 STRING_ascii0 STRING_blank0 STRING_cntrl0 STRING_digit0
220 STRING_graph0 STRING_print0 STRING_punct0 STRING_space0
221 STRING_word0 STRING_xdigit;
222
223 static const uschar posix_name_lengths[] = {
224 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 4, 6, 0 };
225
226 /* Table of class bit maps for each POSIX class. Each class is formed from a
227 base map, with an optional addition or removal of another map. Then, for some
228 classes, there is some additional tweaking: for [:blank:] the vertical space
229 characters are removed, and for [:alpha:] and [:alnum:] the underscore
230 character is removed. The triples in the table consist of the base map offset,
231 second map offset or -1 if no second map, and a non-negative value for map
232 addition or a negative value for map subtraction (if there are two maps). The
233 absolute value of the third field has these meanings: 0 => no tweaking, 1 =>
234 remove vertical space characters, 2 => remove underscore. */
235
236 static const int posix_class_maps[] = {
237 cbit_word, cbit_digit, -2, /* alpha */
238 cbit_lower, -1, 0, /* lower */
239 cbit_upper, -1, 0, /* upper */
240 cbit_word, -1, 2, /* alnum - word without underscore */
241 cbit_print, cbit_cntrl, 0, /* ascii */
242 cbit_space, -1, 1, /* blank - a GNU extension */
243 cbit_cntrl, -1, 0, /* cntrl */
244 cbit_digit, -1, 0, /* digit */
245 cbit_graph, -1, 0, /* graph */
246 cbit_print, -1, 0, /* print */
247 cbit_punct, -1, 0, /* punct */
248 cbit_space, -1, 0, /* space */
249 cbit_word, -1, 0, /* word - a Perl extension */
250 cbit_xdigit,-1, 0 /* xdigit */
251 };
252
253
254 #define STRING(a) # a
255 #define XSTRING(s) STRING(s)
256
257 /* The texts of compile-time error messages. These are "char *" because they
258 are passed to the outside world. Do not ever re-use any error number, because
259 they are documented. Always add a new error instead. Messages marked DEAD below
260 are no longer used. This used to be a table of strings, but in order to reduce
261 the number of relocations needed when a shared library is loaded dynamically,
262 it is now one long string. We cannot use a table of offsets, because the
263 lengths of inserts such as XSTRING(MAX_NAME_SIZE) are not known. Instead, we
264 simply count through to the one we want - this isn't a performance issue
265 because these strings are used only when there is a compilation error. */
266
267 static const char error_texts[] =
268 "no error\0"
269 "\\ at end of pattern\0"
270 "\\c at end of pattern\0"
271 "unrecognized character follows \\\0"
272 "numbers out of order in {} quantifier\0"
273 /* 5 */
274 "number too big in {} quantifier\0"
275 "missing terminating ] for character class\0"
276 "invalid escape sequence in character class\0"
277 "range out of order in character class\0"
278 "nothing to repeat\0"
279 /* 10 */
280 "operand of unlimited repeat could match the empty string\0" /** DEAD **/
281 "internal error: unexpected repeat\0"
282 "unrecognized character after (? or (?-\0"
283 "POSIX named classes are supported only within a class\0"
284 "missing )\0"
285 /* 15 */
286 "reference to non-existent subpattern\0"
287 "erroffset passed as NULL\0"
288 "unknown option bit(s) set\0"
289 "missing ) after comment\0"
290 "parentheses nested too deeply\0" /** DEAD **/
291 /* 20 */
292 "regular expression is too large\0"
293 "failed to get memory\0"
294 "unmatched parentheses\0"
295 "internal error: code overflow\0"
296 "unrecognized character after (?<\0"
297 /* 25 */
298 "lookbehind assertion is not fixed length\0"
299 "malformed number or name after (?(\0"
300 "conditional group contains more than two branches\0"
301 "assertion expected after (?(\0"
302 "(?R or (?[+-]digits must be followed by )\0"
303 /* 30 */
304 "unknown POSIX class name\0"
305 "POSIX collating elements are not supported\0"
306 "this version of PCRE is not compiled with PCRE_UTF8 support\0"
307 "spare error\0" /** DEAD **/
308 "character value in \\x{...} sequence is too large\0"
309 /* 35 */
310 "invalid condition (?(0)\0"
311 "\\C not allowed in lookbehind assertion\0"
312 "PCRE does not support \\L, \\l, \\N, \\U, or \\u\0"
313 "number after (?C is > 255\0"
314 "closing ) for (?C expected\0"
315 /* 40 */
316 "recursive call could loop indefinitely\0"
317 "unrecognized character after (?P\0"
318 "syntax error in subpattern name (missing terminator)\0"
319 "two named subpatterns have the same name\0"
320 "invalid UTF-8 string\0"
321 /* 45 */
322 "support for \\P, \\p, and \\X has not been compiled\0"
323 "malformed \\P or \\p sequence\0"
324 "unknown property name after \\P or \\p\0"
325 "subpattern name is too long (maximum " XSTRING(MAX_NAME_SIZE) " characters)\0"
326 "too many named subpatterns (maximum " XSTRING(MAX_NAME_COUNT) ")\0"
327 /* 50 */
328 "repeated subpattern is too long\0" /** DEAD **/
329 "octal value is greater than \\377 (not in UTF-8 mode)\0"
330 "internal error: overran compiling workspace\0"
331 "internal error: previously-checked referenced subpattern not found\0"
332 "DEFINE group contains more than one branch\0"
333 /* 55 */
334 "repeating a DEFINE group is not allowed\0"
335 "inconsistent NEWLINE options\0"
336 "\\g is not followed by a braced, angle-bracketed, or quoted name/number or by a plain number\0"
337 "a numbered reference must not be zero\0"
338 "(*VERB) with an argument is not supported\0"
339 /* 60 */
340 "(*VERB) not recognized\0"
341 "number is too big\0"
342 "subpattern name expected\0"
343 "digit expected after (?+\0"
344 "] is an invalid data character in JavaScript compatibility mode";
345
346
347 /* Table to identify digits and hex digits. This is used when compiling
348 patterns. Note that the tables in chartables are dependent on the locale, and
349 may mark arbitrary characters as digits - but the PCRE compiling code expects
350 to handle only 0-9, a-z, and A-Z as digits when compiling. That is why we have
351 a private table here. It costs 256 bytes, but it is a lot faster than doing
352 character value tests (at least in some simple cases I timed), and in some
353 applications one wants PCRE to compile efficiently as well as match
354 efficiently.
355
356 For convenience, we use the same bit definitions as in chartables:
357
358 0x04 decimal digit
359 0x08 hexadecimal digit
360
361 Then we can use ctype_digit and ctype_xdigit in the code. */
362
363 #ifndef EBCDIC
364
365 /* This is the "normal" case, for ASCII systems, and EBCDIC systems running in
366 UTF-8 mode. */
367
368 static const unsigned char digitab[] =
369 {
370 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 0- 7 */
371 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 8- 15 */
372 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 16- 23 */
373 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 24- 31 */
374 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - ' */
375 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* ( - / */
376 0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c, /* 0 - 7 */
377 0x0c,0x0c,0x00,0x00,0x00,0x00,0x00,0x00, /* 8 - ? */
378 0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* @ - G */
379 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* H - O */
380 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* P - W */
381 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* X - _ */
382 0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* ` - g */
383 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* h - o */
384 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* p - w */
385 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* x -127 */
386 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 128-135 */
387 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 136-143 */
388 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 144-151 */
389 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 152-159 */
390 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 160-167 */
391 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 168-175 */
392 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 176-183 */
393 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 184-191 */
394 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 192-199 */
395 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 200-207 */
396 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 208-215 */
397 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 216-223 */
398 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 224-231 */
399 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 232-239 */
400 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 240-247 */
401 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00};/* 248-255 */
402
403 #else
404
405 /* This is the "abnormal" case, for EBCDIC systems not running in UTF-8 mode. */
406
407 static const unsigned char digitab[] =
408 {
409 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 0- 7 0 */
410 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 8- 15 */
411 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 16- 23 10 */
412 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 24- 31 */
413 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 32- 39 20 */
414 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 40- 47 */
415 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 48- 55 30 */
416 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 56- 63 */
417 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - 71 40 */
418 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 72- | */
419 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* & - 87 50 */
420 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 88- 95 */
421 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - -103 60 */
422 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 104- ? */
423 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 112-119 70 */
424 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 120- " */
425 0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* 128- g 80 */
426 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* h -143 */
427 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 144- p 90 */
428 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* q -159 */
429 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 160- x A0 */
430 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* y -175 */
431 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* ^ -183 B0 */
432 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 184-191 */
433 0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* { - G C0 */
434 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* H -207 */
435 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* } - P D0 */
436 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* Q -223 */
437 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* \ - X E0 */
438 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* Y -239 */
439 0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c, /* 0 - 7 F0 */
440 0x0c,0x0c,0x00,0x00,0x00,0x00,0x00,0x00};/* 8 -255 */
441
442 static const unsigned char ebcdic_chartab[] = { /* chartable partial dup */
443 0x80,0x00,0x00,0x00,0x00,0x01,0x00,0x00, /* 0- 7 */
444 0x00,0x00,0x00,0x00,0x01,0x01,0x00,0x00, /* 8- 15 */
445 0x00,0x00,0x00,0x00,0x00,0x01,0x00,0x00, /* 16- 23 */
446 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 24- 31 */
447 0x00,0x00,0x00,0x00,0x00,0x01,0x00,0x00, /* 32- 39 */
448 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 40- 47 */
449 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 48- 55 */
450 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 56- 63 */
451 0x01,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - 71 */
452 0x00,0x00,0x00,0x80,0x00,0x80,0x80,0x80, /* 72- | */
453 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* & - 87 */
454 0x00,0x00,0x00,0x80,0x80,0x80,0x00,0x00, /* 88- 95 */
455 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - -103 */
456 0x00,0x00,0x00,0x00,0x00,0x10,0x00,0x80, /* 104- ? */
457 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 112-119 */
458 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 120- " */
459 0x00,0x1a,0x1a,0x1a,0x1a,0x1a,0x1a,0x12, /* 128- g */
460 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* h -143 */
461 0x00,0x12,0x12,0x12,0x12,0x12,0x12,0x12, /* 144- p */
462 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* q -159 */
463 0x00,0x00,0x12,0x12,0x12,0x12,0x12,0x12, /* 160- x */
464 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* y -175 */
465 0x80,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* ^ -183 */
466 0x00,0x00,0x80,0x00,0x00,0x00,0x00,0x00, /* 184-191 */
467 0x80,0x1a,0x1a,0x1a,0x1a,0x1a,0x1a,0x12, /* { - G */
468 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* H -207 */
469 0x00,0x12,0x12,0x12,0x12,0x12,0x12,0x12, /* } - P */
470 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* Q -223 */
471 0x00,0x00,0x12,0x12,0x12,0x12,0x12,0x12, /* \ - X */
472 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* Y -239 */
473 0x1c,0x1c,0x1c,0x1c,0x1c,0x1c,0x1c,0x1c, /* 0 - 7 */
474 0x1c,0x1c,0x00,0x00,0x00,0x00,0x00,0x00};/* 8 -255 */
475 #endif
476
477
478 /* Definition to allow mutual recursion */
479
480 static BOOL
481 compile_regex(int, int, uschar **, const uschar **, int *, BOOL, BOOL, int,
482 int *, int *, branch_chain *, compile_data *, int *);
483
484
485
486 /*************************************************
487 * Find an error text *
488 *************************************************/
489
490 /* The error texts are now all in one long string, to save on relocations. As
491 some of the text is of unknown length, we can't use a table of offsets.
492 Instead, just count through the strings. This is not a performance issue
493 because it happens only when there has been a compilation error.
494
495 Argument: the error number
496 Returns: pointer to the error string
497 */
498
499 static const char *
500 find_error_text(int n)
501 {
502 const char *s = error_texts;
503 for (; n > 0; n--) while (*s++ != 0) {};
504 return s;
505 }
506
507
508 /*************************************************
509 * Handle escapes *
510 *************************************************/
511
512 /* This function is called when a \ has been encountered. It either returns a
513 positive value for a simple escape such as \n, or a negative value which
514 encodes one of the more complicated things such as \d. A backreference to group
515 n is returned as -(ESC_REF + n); ESC_REF is the highest ESC_xxx macro. When
516 UTF-8 is enabled, a positive value greater than 255 may be returned. On entry,
517 ptr is pointing at the \. On exit, it is on the final character of the escape
518 sequence.
519
520 Arguments:
521 ptrptr points to the pattern position pointer
522 errorcodeptr points to the errorcode variable
523 bracount number of previous extracting brackets
524 options the options bits
525 isclass TRUE if inside a character class
526
527 Returns: zero or positive => a data character
528 negative => a special escape sequence
529 on error, errorcodeptr is set
530 */
531
532 static int
533 check_escape(const uschar **ptrptr, int *errorcodeptr, int bracount,
534 int options, BOOL isclass)
535 {
536 BOOL utf8 = (options & PCRE_UTF8) != 0;
537 const uschar *ptr = *ptrptr + 1;
538 int c, i;
539
540 GETCHARINCTEST(c, ptr); /* Get character value, increment pointer */
541 ptr--; /* Set pointer back to the last byte */
542
543 /* If backslash is at the end of the pattern, it's an error. */
544
545 if (c == 0) *errorcodeptr = ERR1;
546
547 /* Non-alphanumerics are literals. For digits or letters, do an initial lookup
548 in a table. A non-zero result is something that can be returned immediately.
549 Otherwise further processing may be required. */
550
551 #ifndef EBCDIC /* ASCII/UTF-8 coding */
552 else if (c < CHAR_0 || c > CHAR_z) {} /* Not alphanumeric */
553 else if ((i = escapes[c - CHAR_0]) != 0) c = i;
554
555 #else /* EBCDIC coding */
556 else if (c < 'a' || (ebcdic_chartab[c] & 0x0E) == 0) {} /* Not alphanumeric */
557 else if ((i = escapes[c - 0x48]) != 0) c = i;
558 #endif
559
560 /* Escapes that need further processing, or are illegal. */
561
562 else
563 {
564 const uschar *oldptr;
565 BOOL braced, negated;
566
567 switch (c)
568 {
569 /* A number of Perl escapes are not handled by PCRE. We give an explicit
570 error. */
571
572 case CHAR_l:
573 case CHAR_L:
574 case CHAR_N:
575 case CHAR_u:
576 case CHAR_U:
577 *errorcodeptr = ERR37;
578 break;
579
580 /* \g must be followed by one of a number of specific things:
581
582 (1) A number, either plain or braced. If positive, it is an absolute
583 backreference. If negative, it is a relative backreference. This is a Perl
584 5.10 feature.
585
586 (2) Perl 5.10 also supports \g{name} as a reference to a named group. This
587 is part of Perl's movement towards a unified syntax for back references. As
588 this is synonymous with \k{name}, we fudge it up by pretending it really
589 was \k.
590
591 (3) For Oniguruma compatibility we also support \g followed by a name or a
592 number either in angle brackets or in single quotes. However, these are
593 (possibly recursive) subroutine calls, _not_ backreferences. Just return
594 the -ESC_g code (cf \k). */
595
596 case CHAR_g:
597 if (ptr[1] == CHAR_LESS_THAN_SIGN || ptr[1] == CHAR_APOSTROPHE)
598 {
599 c = -ESC_g;
600 break;
601 }
602
603 /* Handle the Perl-compatible cases */
604
605 if (ptr[1] == CHAR_LEFT_CURLY_BRACKET)
606 {
607 const uschar *p;
608 for (p = ptr+2; *p != 0 && *p != CHAR_RIGHT_CURLY_BRACKET; p++)
609 if (*p != CHAR_MINUS && (digitab[*p] & ctype_digit) == 0) break;
610 if (*p != 0 && *p != CHAR_RIGHT_CURLY_BRACKET)
611 {
612 c = -ESC_k;
613 break;
614 }
615 braced = TRUE;
616 ptr++;
617 }
618 else braced = FALSE;
619
620 if (ptr[1] == CHAR_MINUS)
621 {
622 negated = TRUE;
623 ptr++;
624 }
625 else negated = FALSE;
626
627 c = 0;
628 while ((digitab[ptr[1]] & ctype_digit) != 0)
629 c = c * 10 + *(++ptr) - CHAR_0;
630
631 if (c < 0) /* Integer overflow */
632 {
633 *errorcodeptr = ERR61;
634 break;
635 }
636
637 if (braced && *(++ptr) != CHAR_RIGHT_CURLY_BRACKET)
638 {
639 *errorcodeptr = ERR57;
640 break;
641 }
642
643 if (c == 0)
644 {
645 *errorcodeptr = ERR58;
646 break;
647 }
648
649 if (negated)
650 {
651 if (c > bracount)
652 {
653 *errorcodeptr = ERR15;
654 break;
655 }
656 c = bracount - (c - 1);
657 }
658
659 c = -(ESC_REF + c);
660 break;
661
662 /* The handling of escape sequences consisting of a string of digits
663 starting with one that is not zero is not straightforward. By experiment,
664 the way Perl works seems to be as follows:
665
666 Outside a character class, the digits are read as a decimal number. If the
667 number is less than 10, or if there are that many previous extracting
668 left brackets, then it is a back reference. Otherwise, up to three octal
669 digits are read to form an escaped byte. Thus \123 is likely to be octal
670 123 (cf \0123, which is octal 012 followed by the literal 3). If the octal
671 value is greater than 377, the least significant 8 bits are taken. Inside a
672 character class, \ followed by a digit is always an octal number. */
673
674 case CHAR_1: case CHAR_2: case CHAR_3: case CHAR_4: case CHAR_5:
675 case CHAR_6: case CHAR_7: case CHAR_8: case CHAR_9:
676
677 if (!isclass)
678 {
679 oldptr = ptr;
680 c -= CHAR_0;
681 while ((digitab[ptr[1]] & ctype_digit) != 0)
682 c = c * 10 + *(++ptr) - CHAR_0;
683 if (c < 0) /* Integer overflow */
684 {
685 *errorcodeptr = ERR61;
686 break;
687 }
688 if (c < 10 || c <= bracount)
689 {
690 c = -(ESC_REF + c);
691 break;
692 }
693 ptr = oldptr; /* Put the pointer back and fall through */
694 }
695
696 /* Handle an octal number following \. If the first digit is 8 or 9, Perl
697 generates a binary zero byte and treats the digit as a following literal.
698 Thus we have to pull back the pointer by one. */
699
700 if ((c = *ptr) >= CHAR_8)
701 {
702 ptr--;
703 c = 0;
704 break;
705 }
706
707 /* \0 always starts an octal number, but we may drop through to here with a
708 larger first octal digit. The original code used just to take the least
709 significant 8 bits of octal numbers (I think this is what early Perls used
710 to do). Nowadays we allow for larger numbers in UTF-8 mode, but no more
711 than 3 octal digits. */
712
713 case CHAR_0:
714 c -= CHAR_0;
715 while(i++ < 2 && ptr[1] >= CHAR_0 && ptr[1] <= CHAR_7)
716 c = c * 8 + *(++ptr) - CHAR_0;
717 if (!utf8 && c > 255) *errorcodeptr = ERR51;
718 break;
719
720 /* \x is complicated. \x{ddd} is a character number which can be greater
721 than 0xff in utf8 mode, but only if the ddd are hex digits. If not, { is
722 treated as a data character. */
723
724 case CHAR_x:
725 if (ptr[1] == CHAR_LEFT_CURLY_BRACKET)
726 {
727 const uschar *pt = ptr + 2;
728 int count = 0;
729
730 c = 0;
731 while ((digitab[*pt] & ctype_xdigit) != 0)
732 {
733 register int cc = *pt++;
734 if (c == 0 && cc == CHAR_0) continue; /* Leading zeroes */
735 count++;
736
737 #ifndef EBCDIC /* ASCII/UTF-8 coding */
738 if (cc >= CHAR_a) cc -= 32; /* Convert to upper case */
739 c = (c << 4) + cc - ((cc < CHAR_A)? CHAR_0 : (CHAR_A - 10));
740 #else /* EBCDIC coding */
741 if (cc >= CHAR_a && cc <= CHAR_z) cc += 64; /* Convert to upper case */
742 c = (c << 4) + cc - ((cc >= CHAR_0)? CHAR_0 : (CHAR_A - 10));
743 #endif
744 }
745
746 if (*pt == CHAR_RIGHT_CURLY_BRACKET)
747 {
748 if (c < 0 || count > (utf8? 8 : 2)) *errorcodeptr = ERR34;
749 ptr = pt;
750 break;
751 }
752
753 /* If the sequence of hex digits does not end with '}', then we don't
754 recognize this construct; fall through to the normal \x handling. */
755 }
756
757 /* Read just a single-byte hex-defined char */
758
759 c = 0;
760 while (i++ < 2 && (digitab[ptr[1]] & ctype_xdigit) != 0)
761 {
762 int cc; /* Some compilers don't like */
763 cc = *(++ptr); /* ++ in initializers */
764 #ifndef EBCDIC /* ASCII/UTF-8 coding */
765 if (cc >= CHAR_a) cc -= 32; /* Convert to upper case */
766 c = c * 16 + cc - ((cc < CHAR_A)? CHAR_0 : (CHAR_A - 10));
767 #else /* EBCDIC coding */
768 if (cc <= CHAR_z) cc += 64; /* Convert to upper case */
769 c = c * 16 + cc - ((cc >= CHAR_0)? CHAR_0 : (CHAR_A - 10));
770 #endif
771 }
772 break;
773
774 /* For \c, a following letter is upper-cased; then the 0x40 bit is flipped.
775 This coding is ASCII-specific, but then the whole concept of \cx is
776 ASCII-specific. (However, an EBCDIC equivalent has now been added.) */
777
778 case CHAR_c:
779 c = *(++ptr);
780 if (c == 0)
781 {
782 *errorcodeptr = ERR2;
783 break;
784 }
785
786 #ifndef EBCDIC /* ASCII/UTF-8 coding */
787 if (c >= CHAR_a && c <= CHAR_z) c -= 32;
788 c ^= 0x40;
789 #else /* EBCDIC coding */
790 if (c >= CHAR_a && c <= CHAR_z) c += 64;
791 c ^= 0xC0;
792 #endif
793 break;
794
795 /* PCRE_EXTRA enables extensions to Perl in the matter of escapes. Any
796 other alphanumeric following \ is an error if PCRE_EXTRA was set;
797 otherwise, for Perl compatibility, it is a literal. This code looks a bit
798 odd, but there used to be some cases other than the default, and there may
799 be again in future, so I haven't "optimized" it. */
800
801 default:
802 if ((options & PCRE_EXTRA) != 0) switch(c)
803 {
804 default:
805 *errorcodeptr = ERR3;
806 break;
807 }
808 break;
809 }
810 }
811
812 *ptrptr = ptr;
813 return c;
814 }
815
816
817
818 #ifdef SUPPORT_UCP
819 /*************************************************
820 * Handle \P and \p *
821 *************************************************/
822
823 /* This function is called after \P or \p has been encountered, provided that
824 PCRE is compiled with support for Unicode properties. On entry, ptrptr is
825 pointing at the P or p. On exit, it is pointing at the final character of the
826 escape sequence.
827
828 Argument:
829 ptrptr points to the pattern position pointer
830 negptr points to a boolean that is set TRUE for negation else FALSE
831 dptr points to an int that is set to the detailed property value
832 errorcodeptr points to the error code variable
833
834 Returns: type value from ucp_type_table, or -1 for an invalid type
835 */
836
837 static int
838 get_ucp(const uschar **ptrptr, BOOL *negptr, int *dptr, int *errorcodeptr)
839 {
840 int c, i, bot, top;
841 const uschar *ptr = *ptrptr;
842 char name[32];
843
844 c = *(++ptr);
845 if (c == 0) goto ERROR_RETURN;
846
847 *negptr = FALSE;
848
849 /* \P or \p can be followed by a name in {}, optionally preceded by ^ for
850 negation. */
851
852 if (c == CHAR_LEFT_CURLY_BRACKET)
853 {
854 if (ptr[1] == CHAR_CIRCUMFLEX_ACCENT)
855 {
856 *negptr = TRUE;
857 ptr++;
858 }
859 for (i = 0; i < (int)sizeof(name) - 1; i++)
860 {
861 c = *(++ptr);
862 if (c == 0) goto ERROR_RETURN;
863 if (c == CHAR_RIGHT_CURLY_BRACKET) break;
864 name[i] = c;
865 }
866 if (c != CHAR_RIGHT_CURLY_BRACKET) goto ERROR_RETURN;
867 name[i] = 0;
868 }
869
870 /* Otherwise there is just one following character */
871
872 else
873 {
874 name[0] = c;
875 name[1] = 0;
876 }
877
878 *ptrptr = ptr;
879
880 /* Search for a recognized property name using binary chop */
881
882 bot = 0;
883 top = _pcre_utt_size;
884
885 while (bot < top)
886 {
887 i = (bot + top) >> 1;
888 c = strcmp(name, _pcre_utt_names + _pcre_utt[i].name_offset);
889 if (c == 0)
890 {
891 *dptr = _pcre_utt[i].value;
892 return _pcre_utt[i].type;
893 }
894 if (c > 0) bot = i + 1; else top = i;
895 }
896
897 *errorcodeptr = ERR47;
898 *ptrptr = ptr;
899 return -1;
900
901 ERROR_RETURN:
902 *errorcodeptr = ERR46;
903 *ptrptr = ptr;
904 return -1;
905 }
906 #endif
907
908
909
910
911 /*************************************************
912 * Check for counted repeat *
913 *************************************************/
914
915 /* This function is called when a '{' is encountered in a place where it might
916 start a quantifier. It looks ahead to see if it really is a quantifier or not.
917 It is only a quantifier if it is one of the forms {ddd} {ddd,} or {ddd,ddd}
918 where the ddds are digits.
919
920 Arguments:
921 p pointer to the first char after '{'
922
923 Returns: TRUE or FALSE
924 */
925
926 static BOOL
927 is_counted_repeat(const uschar *p)
928 {
929 if ((digitab[*p++] & ctype_digit) == 0) return FALSE;
930 while ((digitab[*p] & ctype_digit) != 0) p++;
931 if (*p == CHAR_RIGHT_CURLY_BRACKET) return TRUE;
932
933 if (*p++ != CHAR_COMMA) return FALSE;
934 if (*p == CHAR_RIGHT_CURLY_BRACKET) return TRUE;
935
936 if ((digitab[*p++] & ctype_digit) == 0) return FALSE;
937 while ((digitab[*p] & ctype_digit) != 0) p++;
938
939 return (*p == CHAR_RIGHT_CURLY_BRACKET);
940 }
941
942
943
944 /*************************************************
945 * Read repeat counts *
946 *************************************************/
947
948 /* Read an item of the form {n,m} and return the values. This is called only
949 after is_counted_repeat() has confirmed that a repeat-count quantifier exists,
950 so the syntax is guaranteed to be correct, but we need to check the values.
951
952 Arguments:
953 p pointer to first char after '{'
954 minp pointer to int for min
955 maxp pointer to int for max
956 returned as -1 if no max
957 errorcodeptr points to error code variable
958
959 Returns: pointer to '}' on success;
960 current ptr on error, with errorcodeptr set non-zero
961 */
962
963 static const uschar *
964 read_repeat_counts(const uschar *p, int *minp, int *maxp, int *errorcodeptr)
965 {
966 int min = 0;
967 int max = -1;
968
969 /* Read the minimum value and do a paranoid check: a negative value indicates
970 an integer overflow. */
971
972 while ((digitab[*p] & ctype_digit) != 0) min = min * 10 + *p++ - CHAR_0;
973 if (min < 0 || min > 65535)
974 {
975 *errorcodeptr = ERR5;
976 return p;
977 }
978
979 /* Read the maximum value if there is one, and again do a paranoid on its size.
980 Also, max must not be less than min. */
981
982 if (*p == CHAR_RIGHT_CURLY_BRACKET) max = min; else
983 {
984 if (*(++p) != CHAR_RIGHT_CURLY_BRACKET)
985 {
986 max = 0;
987 while((digitab[*p] & ctype_digit) != 0) max = max * 10 + *p++ - CHAR_0;
988 if (max < 0 || max > 65535)
989 {
990 *errorcodeptr = ERR5;
991 return p;
992 }
993 if (max < min)
994 {
995 *errorcodeptr = ERR4;
996 return p;
997 }
998 }
999 }
1000
1001 /* Fill in the required variables, and pass back the pointer to the terminating
1002 '}'. */
1003
1004 *minp = min;
1005 *maxp = max;
1006 return p;
1007 }
1008
1009
1010
1011 /*************************************************
1012 * Subroutine for finding forward reference *
1013 *************************************************/
1014
1015 /* This recursive function is called only from find_parens() below. The
1016 top-level call starts at the beginning of the pattern. All other calls must
1017 start at a parenthesis. It scans along a pattern's text looking for capturing
1018 subpatterns, and counting them. If it finds a named pattern that matches the
1019 name it is given, it returns its number. Alternatively, if the name is NULL, it
1020 returns when it reaches a given numbered subpattern. We know that if (?P< is
1021 encountered, the name will be terminated by '>' because that is checked in the
1022 first pass. Recursion is used to keep track of subpatterns that reset the
1023 capturing group numbers - the (?| feature.
1024
1025 Arguments:
1026 ptrptr address of the current character pointer (updated)
1027 cd compile background data
1028 name name to seek, or NULL if seeking a numbered subpattern
1029 lorn name length, or subpattern number if name is NULL
1030 xmode TRUE if we are in /x mode
1031 count pointer to the current capturing subpattern number (updated)
1032
1033 Returns: the number of the named subpattern, or -1 if not found
1034 */
1035
1036 static int
1037 find_parens_sub(uschar **ptrptr, compile_data *cd, const uschar *name, int lorn,
1038 BOOL xmode, int *count)
1039 {
1040 uschar *ptr = *ptrptr;
1041 int start_count = *count;
1042 int hwm_count = start_count;
1043 BOOL dup_parens = FALSE;
1044
1045 /* If the first character is a parenthesis, check on the type of group we are
1046 dealing with. The very first call may not start with a parenthesis. */
1047
1048 if (ptr[0] == CHAR_LEFT_PARENTHESIS)
1049 {
1050 if (ptr[1] == CHAR_QUESTION_MARK &&
1051 ptr[2] == CHAR_VERTICAL_LINE)
1052 {
1053 ptr += 3;
1054 dup_parens = TRUE;
1055 }
1056
1057 /* Handle a normal, unnamed capturing parenthesis */
1058
1059 else if (ptr[1] != CHAR_QUESTION_MARK && ptr[1] != CHAR_ASTERISK)
1060 {
1061 *count += 1;
1062 if (name == NULL && *count == lorn) return *count;
1063 ptr++;
1064 }
1065
1066 /* Handle a condition. If it is an assertion, just carry on so that it
1067 is processed as normal. If not, skip to the closing parenthesis of the
1068 condition (there can't be any nested parens. */
1069
1070 else if (ptr[2] == CHAR_LEFT_PARENTHESIS)
1071 {
1072 ptr += 2;
1073 if (ptr[1] != CHAR_QUESTION_MARK)
1074 {
1075 while (*ptr != 0 && *ptr != CHAR_RIGHT_PARENTHESIS) ptr++;
1076 if (*ptr != 0) ptr++;
1077 }
1078 }
1079
1080 /* We have either (? or (* and not a condition */
1081
1082 else
1083 {
1084 ptr += 2;
1085 if (*ptr == CHAR_P) ptr++; /* Allow optional P */
1086
1087 /* We have to disambiguate (?<! and (?<= from (?<name> for named groups */
1088
1089 if ((*ptr == CHAR_LESS_THAN_SIGN && ptr[1] != CHAR_EXCLAMATION_MARK &&
1090 ptr[1] != CHAR_EQUALS_SIGN) || *ptr == CHAR_APOSTROPHE)
1091 {
1092 int term;
1093 const uschar *thisname;
1094 *count += 1;
1095 if (name == NULL && *count == lorn) return *count;
1096 term = *ptr++;
1097 if (term == CHAR_LESS_THAN_SIGN) term = CHAR_GREATER_THAN_SIGN;
1098 thisname = ptr;
1099 while (*ptr != term) ptr++;
1100 if (name != NULL && lorn == ptr - thisname &&
1101 strncmp((const char *)name, (const char *)thisname, lorn) == 0)
1102 return *count;
1103 term++;
1104 }
1105 }
1106 }
1107
1108 /* Past any initial parenthesis handling, scan for parentheses or vertical
1109 bars. */
1110
1111 for (; *ptr != 0; ptr++)
1112 {
1113 /* Skip over backslashed characters and also entire \Q...\E */
1114
1115 if (*ptr == CHAR_BACKSLASH)
1116 {
1117 if (*(++ptr) == 0) goto FAIL_EXIT;
1118 if (*ptr == CHAR_Q) for (;;)
1119 {
1120 while (*(++ptr) != 0 && *ptr != CHAR_BACKSLASH) {};
1121 if (*ptr == 0) goto FAIL_EXIT;
1122 if (*(++ptr) == CHAR_E) break;
1123 }
1124 continue;
1125 }
1126
1127 /* Skip over character classes; this logic must be similar to the way they
1128 are handled for real. If the first character is '^', skip it. Also, if the
1129 first few characters (either before or after ^) are \Q\E or \E we skip them
1130 too. This makes for compatibility with Perl. Note the use of STR macros to
1131 encode "Q\\E" so that it works in UTF-8 on EBCDIC platforms. */
1132
1133 if (*ptr == CHAR_LEFT_SQUARE_BRACKET)
1134 {
1135 BOOL negate_class = FALSE;
1136 for (;;)
1137 {
1138 if (ptr[1] == CHAR_BACKSLASH)
1139 {
1140 if (ptr[2] == CHAR_E)
1141 ptr+= 2;
1142 else if (strncmp((const char *)ptr+2,
1143 STR_Q STR_BACKSLASH STR_E, 3) == 0)
1144 ptr += 4;
1145 else
1146 break;
1147 }
1148 else if (!negate_class && ptr[1] == CHAR_CIRCUMFLEX_ACCENT)
1149 {
1150 negate_class = TRUE;
1151 ptr++;
1152 }
1153 else break;
1154 }
1155
1156 /* If the next character is ']', it is a data character that must be
1157 skipped, except in JavaScript compatibility mode. */
1158
1159 if (ptr[1] == CHAR_RIGHT_SQUARE_BRACKET &&
1160 (cd->external_options & PCRE_JAVASCRIPT_COMPAT) == 0)
1161 ptr++;
1162
1163 while (*(++ptr) != CHAR_RIGHT_SQUARE_BRACKET)
1164 {
1165 if (*ptr == 0) return -1;
1166 if (*ptr == CHAR_BACKSLASH)
1167 {
1168 if (*(++ptr) == 0) goto FAIL_EXIT;
1169 if (*ptr == CHAR_Q) for (;;)
1170 {
1171 while (*(++ptr) != 0 && *ptr != CHAR_BACKSLASH) {};
1172 if (*ptr == 0) goto FAIL_EXIT;
1173 if (*(++ptr) == CHAR_E) break;
1174 }
1175 continue;
1176 }
1177 }
1178 continue;
1179 }
1180
1181 /* Skip comments in /x mode */
1182
1183 if (xmode && *ptr == CHAR_NUMBER_SIGN)
1184 {
1185 while (*(++ptr) != 0 && *ptr != CHAR_NL) {};
1186 if (*ptr == 0) goto FAIL_EXIT;
1187 continue;
1188 }
1189
1190 /* Check for the special metacharacters */
1191
1192 if (*ptr == CHAR_LEFT_PARENTHESIS)
1193 {
1194 int rc = find_parens_sub(&ptr, cd, name, lorn, xmode, count);
1195 if (rc > 0) return rc;
1196 if (*ptr == 0) goto FAIL_EXIT;
1197 }
1198
1199 else if (*ptr == CHAR_RIGHT_PARENTHESIS)
1200 {
1201 if (dup_parens && *count < hwm_count) *count = hwm_count;
1202 *ptrptr = ptr;
1203 return -1;
1204 }
1205
1206 else if (*ptr == CHAR_VERTICAL_LINE && dup_parens)
1207 {
1208 if (*count > hwm_count) hwm_count = *count;
1209 *count = start_count;
1210 }
1211 }
1212
1213 FAIL_EXIT:
1214 *ptrptr = ptr;
1215 return -1;
1216 }
1217
1218
1219
1220
1221 /*************************************************
1222 * Find forward referenced subpattern *
1223 *************************************************/
1224
1225 /* This function scans along a pattern's text looking for capturing
1226 subpatterns, and counting them. If it finds a named pattern that matches the
1227 name it is given, it returns its number. Alternatively, if the name is NULL, it
1228 returns when it reaches a given numbered subpattern. This is used for forward
1229 references to subpatterns. We used to be able to start this scan from the
1230 current compiling point, using the current count value from cd->bracount, and
1231 do it all in a single loop, but the addition of the possibility of duplicate
1232 subpattern numbers means that we have to scan from the very start, in order to
1233 take account of such duplicates, and to use a recursive function to keep track
1234 of the different types of group.
1235
1236 Arguments:
1237 cd compile background data
1238 name name to seek, or NULL if seeking a numbered subpattern
1239 lorn name length, or subpattern number if name is NULL
1240 xmode TRUE if we are in /x mode
1241
1242 Returns: the number of the found subpattern, or -1 if not found
1243 */
1244
1245 static int
1246 find_parens(compile_data *cd, const uschar *name, int lorn, BOOL xmode)
1247 {
1248 uschar *ptr = (uschar *)cd->start_pattern;
1249 int count = 0;
1250 int rc;
1251
1252 /* If the pattern does not start with an opening parenthesis, the first call
1253 to find_parens_sub() will scan right to the end (if necessary). However, if it
1254 does start with a parenthesis, find_parens_sub() will return when it hits the
1255 matching closing parens. That is why we have to have a loop. */
1256
1257 for (;;)
1258 {
1259 rc = find_parens_sub(&ptr, cd, name, lorn, xmode, &count);
1260 if (rc > 0 || *ptr++ == 0) break;
1261 }
1262
1263 return rc;
1264 }
1265
1266
1267
1268
1269 /*************************************************
1270 * Find first significant op code *
1271 *************************************************/
1272
1273 /* This is called by several functions that scan a compiled expression looking
1274 for a fixed first character, or an anchoring op code etc. It skips over things
1275 that do not influence this. For some calls, a change of option is important.
1276 For some calls, it makes sense to skip negative forward and all backward
1277 assertions, and also the \b assertion; for others it does not.
1278
1279 Arguments:
1280 code pointer to the start of the group
1281 options pointer to external options
1282 optbit the option bit whose changing is significant, or
1283 zero if none are
1284 skipassert TRUE if certain assertions are to be skipped
1285
1286 Returns: pointer to the first significant opcode
1287 */
1288
1289 static const uschar*
1290 first_significant_code(const uschar *code, int *options, int optbit,
1291 BOOL skipassert)
1292 {
1293 for (;;)
1294 {
1295 switch ((int)*code)
1296 {
1297 case OP_OPT:
1298 if (optbit > 0 && ((int)code[1] & optbit) != (*options & optbit))
1299 *options = (int)code[1];
1300 code += 2;
1301 break;
1302
1303 case OP_ASSERT_NOT:
1304 case OP_ASSERTBACK:
1305 case OP_ASSERTBACK_NOT:
1306 if (!skipassert) return code;
1307 do code += GET(code, 1); while (*code == OP_ALT);
1308 code += _pcre_OP_lengths[*code];
1309 break;
1310
1311 case OP_WORD_BOUNDARY:
1312 case OP_NOT_WORD_BOUNDARY:
1313 if (!skipassert) return code;
1314 /* Fall through */
1315
1316 case OP_CALLOUT:
1317 case OP_CREF:
1318 case OP_RREF:
1319 case OP_DEF:
1320 code += _pcre_OP_lengths[*code];
1321 break;
1322
1323 default:
1324 return code;
1325 }
1326 }
1327 /* Control never reaches here */
1328 }
1329
1330
1331
1332
1333 /*************************************************
1334 * Find the fixed length of a branch *
1335 *************************************************/
1336
1337 /* Scan a branch and compute the fixed length of subject that will match it,
1338 if the length is fixed. This is needed for dealing with backward assertions.
1339 In UTF8 mode, the result is in characters rather than bytes. The branch is
1340 temporarily terminated with OP_END when this function is called.
1341
1342 This function is called when a backward assertion is encountered, so that if it
1343 fails, the error message can point to the correct place in the pattern.
1344 However, we cannot do this when the assertion contains subroutine calls,
1345 because they can be forward references. We solve this by remembering this case
1346 and doing the check at the end; a flag specifies which mode we are running in.
1347
1348 Arguments:
1349 code points to the start of the pattern (the bracket)
1350 options the compiling options
1351 atend TRUE if called when the pattern is complete
1352 cd the "compile data" structure
1353
1354 Returns: the fixed length,
1355 or -1 if there is no fixed length,
1356 or -2 if \C was encountered
1357 or -3 if an OP_RECURSE item was encountered and atend is FALSE
1358 */
1359
1360 static int
1361 find_fixedlength(uschar *code, int options, BOOL atend, compile_data *cd)
1362 {
1363 int length = -1;
1364
1365 register int branchlength = 0;
1366 register uschar *cc = code + 1 + LINK_SIZE;
1367
1368 /* Scan along the opcodes for this branch. If we get to the end of the
1369 branch, check the length against that of the other branches. */
1370
1371 for (;;)
1372 {
1373 int d;
1374 uschar *ce, *cs;
1375 register int op = *cc;
1376 switch (op)
1377 {
1378 case OP_CBRA:
1379 case OP_BRA:
1380 case OP_ONCE:
1381 case OP_COND:
1382 d = find_fixedlength(cc + ((op == OP_CBRA)? 2:0), options, atend, cd);
1383 if (d < 0) return d;
1384 branchlength += d;
1385 do cc += GET(cc, 1); while (*cc == OP_ALT);
1386 cc += 1 + LINK_SIZE;
1387 break;
1388
1389 /* Reached end of a branch; if it's a ket it is the end of a nested
1390 call. If it's ALT it is an alternation in a nested call. If it is
1391 END it's the end of the outer call. All can be handled by the same code. */
1392
1393 case OP_ALT:
1394 case OP_KET:
1395 case OP_KETRMAX:
1396 case OP_KETRMIN:
1397 case OP_END:
1398 if (length < 0) length = branchlength;
1399 else if (length != branchlength) return -1;
1400 if (*cc != OP_ALT) return length;
1401 cc += 1 + LINK_SIZE;
1402 branchlength = 0;
1403 break;
1404
1405 /* A true recursion implies not fixed length, but a subroutine call may
1406 be OK. If the subroutine is a forward reference, we can't deal with
1407 it until the end of the pattern, so return -3. */
1408
1409 case OP_RECURSE:
1410 if (!atend) return -3;
1411 cs = ce = (uschar *)cd->start_code + GET(cc, 1); /* Start subpattern */
1412 do ce += GET(ce, 1); while (*ce == OP_ALT); /* End subpattern */
1413 if (cc > cs && cc < ce) return -1; /* Recursion */
1414 d = find_fixedlength(cs + 2, options, atend, cd);
1415 if (d < 0) return d;
1416 branchlength += d;
1417 cc += 1 + LINK_SIZE;
1418 break;
1419
1420 /* Skip over assertive subpatterns */
1421
1422 case OP_ASSERT:
1423 case OP_ASSERT_NOT:
1424 case OP_ASSERTBACK:
1425 case OP_ASSERTBACK_NOT:
1426 do cc += GET(cc, 1); while (*cc == OP_ALT);
1427 /* Fall through */
1428
1429 /* Skip over things that don't match chars */
1430
1431 case OP_REVERSE:
1432 case OP_CREF:
1433 case OP_RREF:
1434 case OP_DEF:
1435 case OP_OPT:
1436 case OP_CALLOUT:
1437 case OP_SOD:
1438 case OP_SOM:
1439 case OP_EOD:
1440 case OP_EODN:
1441 case OP_CIRC:
1442 case OP_DOLL:
1443 case OP_NOT_WORD_BOUNDARY:
1444 case OP_WORD_BOUNDARY:
1445 cc += _pcre_OP_lengths[*cc];
1446 break;
1447
1448 /* Handle literal characters */
1449
1450 case OP_CHAR:
1451 case OP_CHARNC:
1452 case OP_NOT:
1453 branchlength++;
1454 cc += 2;
1455 #ifdef SUPPORT_UTF8
1456 if ((options & PCRE_UTF8) != 0 && cc[-1] >= 0xc0)
1457 cc += _pcre_utf8_table4[cc[-1] & 0x3f];
1458 #endif
1459 break;
1460
1461 /* Handle exact repetitions. The count is already in characters, but we
1462 need to skip over a multibyte character in UTF8 mode. */
1463
1464 case OP_EXACT:
1465 branchlength += GET2(cc,1);
1466 cc += 4;
1467 #ifdef SUPPORT_UTF8
1468 if ((options & PCRE_UTF8) != 0 && cc[-1] >= 0xc0)
1469 cc += _pcre_utf8_table4[cc[-1] & 0x3f];
1470 #endif
1471 break;
1472
1473 case OP_TYPEEXACT:
1474 branchlength += GET2(cc,1);
1475 if (cc[3] == OP_PROP || cc[3] == OP_NOTPROP) cc += 2;
1476 cc += 4;
1477 break;
1478
1479 /* Handle single-char matchers */
1480
1481 case OP_PROP:
1482 case OP_NOTPROP:
1483 cc += 2;
1484 /* Fall through */
1485
1486 case OP_NOT_DIGIT:
1487 case OP_DIGIT:
1488 case OP_NOT_WHITESPACE:
1489 case OP_WHITESPACE:
1490 case OP_NOT_WORDCHAR:
1491 case OP_WORDCHAR:
1492 case OP_ANY:
1493 case OP_ALLANY:
1494 branchlength++;
1495 cc++;
1496 break;
1497
1498 /* The single-byte matcher isn't allowed */
1499
1500 case OP_ANYBYTE:
1501 return -2;
1502
1503 /* Check a class for variable quantification */
1504
1505 #ifdef SUPPORT_UTF8
1506 case OP_XCLASS:
1507 cc += GET(cc, 1) - 33;
1508 /* Fall through */
1509 #endif
1510
1511 case OP_CLASS:
1512 case OP_NCLASS:
1513 cc += 33;
1514
1515 switch (*cc)
1516 {
1517 case OP_CRSTAR:
1518 case OP_CRMINSTAR:
1519 case OP_CRQUERY:
1520 case OP_CRMINQUERY:
1521 return -1;
1522
1523 case OP_CRRANGE:
1524 case OP_CRMINRANGE:
1525 if (GET2(cc,1) != GET2(cc,3)) return -1;
1526 branchlength += GET2(cc,1);
1527 cc += 5;
1528 break;
1529
1530 default:
1531 branchlength++;
1532 }
1533 break;
1534
1535 /* Anything else is variable length */
1536
1537 default:
1538 return -1;
1539 }
1540 }
1541 /* Control never gets here */
1542 }
1543
1544
1545
1546
1547 /*************************************************
1548 * Scan compiled regex for specific bracket *
1549 *************************************************/
1550
1551 /* This little function scans through a compiled pattern until it finds a
1552 capturing bracket with the given number, or, if the number is negative, an
1553 instance of OP_REVERSE for a lookbehind.
1554
1555 Arguments:
1556 code points to start of expression
1557 utf8 TRUE in UTF-8 mode
1558 number the required bracket number or negative to find a lookbehind
1559
1560 Returns: pointer to the opcode for the bracket, or NULL if not found
1561 */
1562
1563 static const uschar *
1564 find_bracket(const uschar *code, BOOL utf8, int number)
1565 {
1566 for (;;)
1567 {
1568 register int c = *code;
1569 if (c == OP_END) return NULL;
1570
1571 /* XCLASS is used for classes that cannot be represented just by a bit
1572 map. This includes negated single high-valued characters. The length in
1573 the table is zero; the actual length is stored in the compiled code. */
1574
1575 if (c == OP_XCLASS) code += GET(code, 1);
1576
1577 /* Handle recursion */
1578
1579 else if (c == OP_REVERSE)
1580 {
1581 if (number < 0) return (uschar *)code;
1582 code += _pcre_OP_lengths[c];
1583 }
1584
1585 /* Handle capturing bracket */
1586
1587 else if (c == OP_CBRA)
1588 {
1589 int n = GET2(code, 1+LINK_SIZE);
1590 if (n == number) return (uschar *)code;
1591 code += _pcre_OP_lengths[c];
1592 }
1593
1594 /* Otherwise, we can get the item's length from the table, except that for
1595 repeated character types, we have to test for \p and \P, which have an extra
1596 two bytes of parameters. */
1597
1598 else
1599 {
1600 switch(c)
1601 {
1602 case OP_TYPESTAR:
1603 case OP_TYPEMINSTAR:
1604 case OP_TYPEPLUS:
1605 case OP_TYPEMINPLUS:
1606 case OP_TYPEQUERY:
1607 case OP_TYPEMINQUERY:
1608 case OP_TYPEPOSSTAR:
1609 case OP_TYPEPOSPLUS:
1610 case OP_TYPEPOSQUERY:
1611 if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
1612 break;
1613
1614 case OP_TYPEUPTO:
1615 case OP_TYPEMINUPTO:
1616 case OP_TYPEEXACT:
1617 case OP_TYPEPOSUPTO:
1618 if (code[3] == OP_PROP || code[3] == OP_NOTPROP) code += 2;
1619 break;
1620 }
1621
1622 /* Add in the fixed length from the table */
1623
1624 code += _pcre_OP_lengths[c];
1625
1626 /* In UTF-8 mode, opcodes that are followed by a character may be followed by
1627 a multi-byte character. The length in the table is a minimum, so we have to
1628 arrange to skip the extra bytes. */
1629
1630 #ifdef SUPPORT_UTF8
1631 if (utf8) switch(c)
1632 {
1633 case OP_CHAR:
1634 case OP_CHARNC:
1635 case OP_EXACT:
1636 case OP_UPTO:
1637 case OP_MINUPTO:
1638 case OP_POSUPTO:
1639 case OP_STAR:
1640 case OP_MINSTAR:
1641 case OP_POSSTAR:
1642 case OP_PLUS:
1643 case OP_MINPLUS:
1644 case OP_POSPLUS:
1645 case OP_QUERY:
1646 case OP_MINQUERY:
1647 case OP_POSQUERY:
1648 if (code[-1] >= 0xc0) code += _pcre_utf8_table4[code[-1] & 0x3f];
1649 break;
1650 }
1651 #else
1652 (void)(utf8); /* Keep compiler happy by referencing function argument */
1653 #endif
1654 }
1655 }
1656 }
1657
1658
1659
1660 /*************************************************
1661 * Scan compiled regex for recursion reference *
1662 *************************************************/
1663
1664 /* This little function scans through a compiled pattern until it finds an
1665 instance of OP_RECURSE.
1666
1667 Arguments:
1668 code points to start of expression
1669 utf8 TRUE in UTF-8 mode
1670
1671 Returns: pointer to the opcode for OP_RECURSE, or NULL if not found
1672 */
1673
1674 static const uschar *
1675 find_recurse(const uschar *code, BOOL utf8)
1676 {
1677 for (;;)
1678 {
1679 register int c = *code;
1680 if (c == OP_END) return NULL;
1681 if (c == OP_RECURSE) return code;
1682
1683 /* XCLASS is used for classes that cannot be represented just by a bit
1684 map. This includes negated single high-valued characters. The length in
1685 the table is zero; the actual length is stored in the compiled code. */
1686
1687 if (c == OP_XCLASS) code += GET(code, 1);
1688
1689 /* Otherwise, we can get the item's length from the table, except that for
1690 repeated character types, we have to test for \p and \P, which have an extra
1691 two bytes of parameters. */
1692
1693 else
1694 {
1695 switch(c)
1696 {
1697 case OP_TYPESTAR:
1698 case OP_TYPEMINSTAR:
1699 case OP_TYPEPLUS:
1700 case OP_TYPEMINPLUS:
1701 case OP_TYPEQUERY:
1702 case OP_TYPEMINQUERY:
1703 case OP_TYPEPOSSTAR:
1704 case OP_TYPEPOSPLUS:
1705 case OP_TYPEPOSQUERY:
1706 if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
1707 break;
1708
1709 case OP_TYPEPOSUPTO:
1710 case OP_TYPEUPTO:
1711 case OP_TYPEMINUPTO:
1712 case OP_TYPEEXACT:
1713 if (code[3] == OP_PROP || code[3] == OP_NOTPROP) code += 2;
1714 break;
1715 }
1716
1717 /* Add in the fixed length from the table */
1718
1719 code += _pcre_OP_lengths[c];
1720
1721 /* In UTF-8 mode, opcodes that are followed by a character may be followed
1722 by a multi-byte character. The length in the table is a minimum, so we have
1723 to arrange to skip the extra bytes. */
1724
1725 #ifdef SUPPORT_UTF8
1726 if (utf8) switch(c)
1727 {
1728 case OP_CHAR:
1729 case OP_CHARNC:
1730 case OP_EXACT:
1731 case OP_UPTO:
1732 case OP_MINUPTO:
1733 case OP_POSUPTO:
1734 case OP_STAR:
1735 case OP_MINSTAR:
1736 case OP_POSSTAR:
1737 case OP_PLUS:
1738 case OP_MINPLUS:
1739 case OP_POSPLUS:
1740 case OP_QUERY:
1741 case OP_MINQUERY:
1742 case OP_POSQUERY:
1743 if (code[-1] >= 0xc0) code += _pcre_utf8_table4[code[-1] & 0x3f];
1744 break;
1745 }
1746 #else
1747 (void)(utf8); /* Keep compiler happy by referencing function argument */
1748 #endif
1749 }
1750 }
1751 }
1752
1753
1754
1755 /*************************************************
1756 * Scan compiled branch for non-emptiness *
1757 *************************************************/
1758
1759 /* This function scans through a branch of a compiled pattern to see whether it
1760 can match the empty string or not. It is called from could_be_empty()
1761 below and from compile_branch() when checking for an unlimited repeat of a
1762 group that can match nothing. Note that first_significant_code() skips over
1763 backward and negative forward assertions when its final argument is TRUE. If we
1764 hit an unclosed bracket, we return "empty" - this means we've struck an inner
1765 bracket whose current branch will already have been scanned.
1766
1767 Arguments:
1768 code points to start of search
1769 endcode points to where to stop
1770 utf8 TRUE if in UTF8 mode
1771
1772 Returns: TRUE if what is matched could be empty
1773 */
1774
1775 static BOOL
1776 could_be_empty_branch(const uschar *code, const uschar *endcode, BOOL utf8)
1777 {
1778 register int c;
1779 for (code = first_significant_code(code + _pcre_OP_lengths[*code], NULL, 0, TRUE);
1780 code < endcode;
1781 code = first_significant_code(code + _pcre_OP_lengths[c], NULL, 0, TRUE))
1782 {
1783 const uschar *ccode;
1784
1785 c = *code;
1786
1787 /* Skip over forward assertions; the other assertions are skipped by
1788 first_significant_code() with a TRUE final argument. */
1789
1790 if (c == OP_ASSERT)
1791 {
1792 do code += GET(code, 1); while (*code == OP_ALT);
1793 c = *code;
1794 continue;
1795 }
1796
1797 /* Groups with zero repeats can of course be empty; skip them. */
1798
1799 if (c == OP_BRAZERO || c == OP_BRAMINZERO || c == OP_SKIPZERO)
1800 {
1801 code += _pcre_OP_lengths[c];
1802 do code += GET(code, 1); while (*code == OP_ALT);
1803 c = *code;
1804 continue;
1805 }
1806
1807 /* For other groups, scan the branches. */
1808
1809 if (c == OP_BRA || c == OP_CBRA || c == OP_ONCE || c == OP_COND)
1810 {
1811 BOOL empty_branch;
1812 if (GET(code, 1) == 0) return TRUE; /* Hit unclosed bracket */
1813
1814 /* If a conditional group has only one branch, there is a second, implied,
1815 empty branch, so just skip over the conditional, because it could be empty.
1816 Otherwise, scan the individual branches of the group. */
1817
1818 if (c == OP_COND && code[GET(code, 1)] != OP_ALT)
1819 code += GET(code, 1);
1820 else
1821 {
1822 empty_branch = FALSE;
1823 do
1824 {
1825 if (!empty_branch && could_be_empty_branch(code, endcode, utf8))
1826 empty_branch = TRUE;
1827 code += GET(code, 1);
1828 }
1829 while (*code == OP_ALT);
1830 if (!empty_branch) return FALSE; /* All branches are non-empty */
1831 }
1832
1833 c = *code;
1834 continue;
1835 }
1836
1837 /* Handle the other opcodes */
1838
1839 switch (c)
1840 {
1841 /* Check for quantifiers after a class. XCLASS is used for classes that
1842 cannot be represented just by a bit map. This includes negated single
1843 high-valued characters. The length in _pcre_OP_lengths[] is zero; the
1844 actual length is stored in the compiled code, so we must update "code"
1845 here. */
1846
1847 #ifdef SUPPORT_UTF8
1848 case OP_XCLASS:
1849 ccode = code += GET(code, 1);
1850 goto CHECK_CLASS_REPEAT;
1851 #endif
1852
1853 case OP_CLASS:
1854 case OP_NCLASS:
1855 ccode = code + 33;
1856
1857 #ifdef SUPPORT_UTF8
1858 CHECK_CLASS_REPEAT:
1859 #endif
1860
1861 switch (*ccode)
1862 {
1863 case OP_CRSTAR: /* These could be empty; continue */
1864 case OP_CRMINSTAR:
1865 case OP_CRQUERY:
1866 case OP_CRMINQUERY:
1867 break;
1868
1869 default: /* Non-repeat => class must match */
1870 case OP_CRPLUS: /* These repeats aren't empty */
1871 case OP_CRMINPLUS:
1872 return FALSE;
1873
1874 case OP_CRRANGE:
1875 case OP_CRMINRANGE:
1876 if (GET2(ccode, 1) > 0) return FALSE; /* Minimum > 0 */
1877 break;
1878 }
1879 break;
1880
1881 /* Opcodes that must match a character */
1882
1883 case OP_PROP:
1884 case OP_NOTPROP:
1885 case OP_EXTUNI:
1886 case OP_NOT_DIGIT:
1887 case OP_DIGIT:
1888 case OP_NOT_WHITESPACE:
1889 case OP_WHITESPACE:
1890 case OP_NOT_WORDCHAR:
1891 case OP_WORDCHAR:
1892 case OP_ANY:
1893 case OP_ALLANY:
1894 case OP_ANYBYTE:
1895 case OP_CHAR:
1896 case OP_CHARNC:
1897 case OP_NOT:
1898 case OP_PLUS:
1899 case OP_MINPLUS:
1900 case OP_POSPLUS:
1901 case OP_EXACT:
1902 case OP_NOTPLUS:
1903 case OP_NOTMINPLUS:
1904 case OP_NOTPOSPLUS:
1905 case OP_NOTEXACT:
1906 case OP_TYPEPLUS:
1907 case OP_TYPEMINPLUS:
1908 case OP_TYPEPOSPLUS:
1909 case OP_TYPEEXACT:
1910 return FALSE;
1911
1912 /* These are going to continue, as they may be empty, but we have to
1913 fudge the length for the \p and \P cases. */
1914
1915 case OP_TYPESTAR:
1916 case OP_TYPEMINSTAR:
1917 case OP_TYPEPOSSTAR:
1918 case OP_TYPEQUERY:
1919 case OP_TYPEMINQUERY:
1920 case OP_TYPEPOSQUERY:
1921 if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
1922 break;
1923
1924 /* Same for these */
1925
1926 case OP_TYPEUPTO:
1927 case OP_TYPEMINUPTO:
1928 case OP_TYPEPOSUPTO:
1929 if (code[3] == OP_PROP || code[3] == OP_NOTPROP) code += 2;
1930 break;
1931
1932 /* End of branch */
1933
1934 case OP_KET:
1935 case OP_KETRMAX:
1936 case OP_KETRMIN:
1937 case OP_ALT:
1938 return TRUE;
1939
1940 /* In UTF-8 mode, STAR, MINSTAR, POSSTAR, QUERY, MINQUERY, POSQUERY, UPTO,
1941 MINUPTO, and POSUPTO may be followed by a multibyte character */
1942
1943 #ifdef SUPPORT_UTF8
1944 case OP_STAR:
1945 case OP_MINSTAR:
1946 case OP_POSSTAR:
1947 case OP_QUERY:
1948 case OP_MINQUERY:
1949 case OP_POSQUERY:
1950 if (utf8 && code[1] >= 0xc0) code += _pcre_utf8_table4[code[1] & 0x3f];
1951 break;
1952
1953 case OP_UPTO:
1954 case OP_MINUPTO:
1955 case OP_POSUPTO:
1956 if (utf8 && code[3] >= 0xc0) code += _pcre_utf8_table4[code[3] & 0x3f];
1957 break;
1958 #endif
1959 }
1960 }
1961
1962 return TRUE;
1963 }
1964
1965
1966
1967 /*************************************************
1968 * Scan compiled regex for non-emptiness *
1969 *************************************************/
1970
1971 /* This function is called to check for left recursive calls. We want to check
1972 the current branch of the current pattern to see if it could match the empty
1973 string. If it could, we must look outwards for branches at other levels,
1974 stopping when we pass beyond the bracket which is the subject of the recursion.
1975
1976 Arguments:
1977 code points to start of the recursion
1978 endcode points to where to stop (current RECURSE item)
1979 bcptr points to the chain of current (unclosed) branch starts
1980 utf8 TRUE if in UTF-8 mode
1981
1982 Returns: TRUE if what is matched could be empty
1983 */
1984
1985 static BOOL
1986 could_be_empty(const uschar *code, const uschar *endcode, branch_chain *bcptr,
1987 BOOL utf8)
1988 {
1989 while (bcptr != NULL && bcptr->current >= code)
1990 {
1991 if (!could_be_empty_branch(bcptr->current, endcode, utf8)) return FALSE;
1992 bcptr = bcptr->outer;
1993 }
1994 return TRUE;
1995 }
1996
1997
1998
1999 /*************************************************
2000 * Check for POSIX class syntax *
2001 *************************************************/
2002
2003 /* This function is called when the sequence "[:" or "[." or "[=" is
2004 encountered in a character class. It checks whether this is followed by a
2005 sequence of characters terminated by a matching ":]" or ".]" or "=]". If we
2006 reach an unescaped ']' without the special preceding character, return FALSE.
2007
2008 Originally, this function only recognized a sequence of letters between the
2009 terminators, but it seems that Perl recognizes any sequence of characters,
2010 though of course unknown POSIX names are subsequently rejected. Perl gives an
2011 "Unknown POSIX class" error for [:f\oo:] for example, where previously PCRE
2012 didn't consider this to be a POSIX class. Likewise for [:1234:].
2013
2014 The problem in trying to be exactly like Perl is in the handling of escapes. We
2015 have to be sure that [abc[:x\]pqr] is *not* treated as containing a POSIX
2016 class, but [abc[:x\]pqr:]] is (so that an error can be generated). The code
2017 below handles the special case of \], but does not try to do any other escape
2018 processing. This makes it different from Perl for cases such as [:l\ower:]
2019 where Perl recognizes it as the POSIX class "lower" but PCRE does not recognize
2020 "l\ower". This is a lesser evil that not diagnosing bad classes when Perl does,
2021 I think.
2022
2023 Arguments:
2024 ptr pointer to the initial [
2025 endptr where to return the end pointer
2026
2027 Returns: TRUE or FALSE
2028 */
2029
2030 static BOOL
2031 check_posix_syntax(const uschar *ptr, const uschar **endptr)
2032 {
2033 int terminator; /* Don't combine these lines; the Solaris cc */
2034 terminator = *(++ptr); /* compiler warns about "non-constant" initializer. */
2035 for (++ptr; *ptr != 0; ptr++)
2036 {
2037 if (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET) ptr++; else
2038 {
2039 if (*ptr == CHAR_RIGHT_SQUARE_BRACKET) return FALSE;
2040 if (*ptr == terminator && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET)
2041 {
2042 *endptr = ptr;
2043 return TRUE;
2044 }
2045 }
2046 }
2047 return FALSE;
2048 }
2049
2050
2051
2052
2053 /*************************************************
2054 * Check POSIX class name *
2055 *************************************************/
2056
2057 /* This function is called to check the name given in a POSIX-style class entry
2058 such as [:alnum:].
2059
2060 Arguments:
2061 ptr points to the first letter
2062 len the length of the name
2063
2064 Returns: a value representing the name, or -1 if unknown
2065 */
2066
2067 static int
2068 check_posix_name(const uschar *ptr, int len)
2069 {
2070 const char *pn = posix_names;
2071 register int yield = 0;
2072 while (posix_name_lengths[yield] != 0)
2073 {
2074 if (len == posix_name_lengths[yield] &&
2075 strncmp((const char *)ptr, pn, len) == 0) return yield;
2076 pn += posix_name_lengths[yield] + 1;
2077 yield++;
2078 }
2079 return -1;
2080 }
2081
2082
2083 /*************************************************
2084 * Adjust OP_RECURSE items in repeated group *
2085 *************************************************/
2086
2087 /* OP_RECURSE items contain an offset from the start of the regex to the group
2088 that is referenced. This means that groups can be replicated for fixed
2089 repetition simply by copying (because the recursion is allowed to refer to
2090 earlier groups that are outside the current group). However, when a group is
2091 optional (i.e. the minimum quantifier is zero), OP_BRAZERO or OP_SKIPZERO is
2092 inserted before it, after it has been compiled. This means that any OP_RECURSE
2093 items within it that refer to the group itself or any contained groups have to
2094 have their offsets adjusted. That one of the jobs of this function. Before it
2095 is called, the partially compiled regex must be temporarily terminated with
2096 OP_END.
2097
2098 This function has been extended with the possibility of forward references for
2099 recursions and subroutine calls. It must also check the list of such references
2100 for the group we are dealing with. If it finds that one of the recursions in
2101 the current group is on this list, it adjusts the offset in the list, not the
2102 value in the reference (which is a group number).
2103
2104 Arguments:
2105 group points to the start of the group
2106 adjust the amount by which the group is to be moved
2107 utf8 TRUE in UTF-8 mode
2108 cd contains pointers to tables etc.
2109 save_hwm the hwm forward reference pointer at the start of the group
2110
2111 Returns: nothing
2112 */
2113
2114 static void
2115 adjust_recurse(uschar *group, int adjust, BOOL utf8, compile_data *cd,
2116 uschar *save_hwm)
2117 {
2118 uschar *ptr = group;
2119
2120 while ((ptr = (uschar *)find_recurse(ptr, utf8)) != NULL)
2121 {
2122 int offset;
2123 uschar *hc;
2124
2125 /* See if this recursion is on the forward reference list. If so, adjust the
2126 reference. */
2127
2128 for (hc = save_hwm; hc < cd->hwm; hc += LINK_SIZE)
2129 {
2130 offset = GET(hc, 0);
2131 if (cd->start_code + offset == ptr + 1)
2132 {
2133 PUT(hc, 0, offset + adjust);
2134 break;
2135 }
2136 }
2137
2138 /* Otherwise, adjust the recursion offset if it's after the start of this
2139 group. */
2140
2141 if (hc >= cd->hwm)
2142 {
2143 offset = GET(ptr, 1);
2144 if (cd->start_code + offset >= group) PUT(ptr, 1, offset + adjust);
2145 }
2146
2147 ptr += 1 + LINK_SIZE;
2148 }
2149 }
2150
2151
2152
2153 /*************************************************
2154 * Insert an automatic callout point *
2155 *************************************************/
2156
2157 /* This function is called when the PCRE_AUTO_CALLOUT option is set, to insert
2158 callout points before each pattern item.
2159
2160 Arguments:
2161 code current code pointer
2162 ptr current pattern pointer
2163 cd pointers to tables etc
2164
2165 Returns: new code pointer
2166 */
2167
2168 static uschar *
2169 auto_callout(uschar *code, const uschar *ptr, compile_data *cd)
2170 {
2171 *code++ = OP_CALLOUT;
2172 *code++ = 255;
2173 PUT(code, 0, ptr - cd->start_pattern); /* Pattern offset */
2174 PUT(code, LINK_SIZE, 0); /* Default length */
2175 return code + 2*LINK_SIZE;
2176 }
2177
2178
2179
2180 /*************************************************
2181 * Complete a callout item *
2182 *************************************************/
2183
2184 /* A callout item contains the length of the next item in the pattern, which
2185 we can't fill in till after we have reached the relevant point. This is used
2186 for both automatic and manual callouts.
2187
2188 Arguments:
2189 previous_callout points to previous callout item
2190 ptr current pattern pointer
2191 cd pointers to tables etc
2192
2193 Returns: nothing
2194 */
2195
2196 static void
2197 complete_callout(uschar *previous_callout, const uschar *ptr, compile_data *cd)
2198 {
2199 int length = ptr - cd->start_pattern - GET(previous_callout, 2);
2200 PUT(previous_callout, 2 + LINK_SIZE, length);
2201 }
2202
2203
2204
2205 #ifdef SUPPORT_UCP
2206 /*************************************************
2207 * Get othercase range *
2208 *************************************************/
2209
2210 /* This function is passed the start and end of a class range, in UTF-8 mode
2211 with UCP support. It searches up the characters, looking for internal ranges of
2212 characters in the "other" case. Each call returns the next one, updating the
2213 start address.
2214
2215 Arguments:
2216 cptr points to starting character value; updated
2217 d end value
2218 ocptr where to put start of othercase range
2219 odptr where to put end of othercase range
2220
2221 Yield: TRUE when range returned; FALSE when no more
2222 */
2223
2224 static BOOL
2225 get_othercase_range(unsigned int *cptr, unsigned int d, unsigned int *ocptr,
2226 unsigned int *odptr)
2227 {
2228 unsigned int c, othercase, next;
2229
2230 for (c = *cptr; c <= d; c++)
2231 { if ((othercase = UCD_OTHERCASE(c)) != c) break; }
2232
2233 if (c > d) return FALSE;
2234
2235 *ocptr = othercase;
2236 next = othercase + 1;
2237
2238 for (++c; c <= d; c++)
2239 {
2240 if (UCD_OTHERCASE(c) != next) break;
2241 next++;
2242 }
2243
2244 *odptr = next - 1;
2245 *cptr = c;
2246
2247 return TRUE;
2248 }
2249 #endif /* SUPPORT_UCP */
2250
2251
2252
2253 /*************************************************
2254 * Check if auto-possessifying is possible *
2255 *************************************************/
2256
2257 /* This function is called for unlimited repeats of certain items, to see
2258 whether the next thing could possibly match the repeated item. If not, it makes
2259 sense to automatically possessify the repeated item.
2260
2261 Arguments:
2262 op_code the repeated op code
2263 this data for this item, depends on the opcode
2264 utf8 TRUE in UTF-8 mode
2265 utf8_char used for utf8 character bytes, NULL if not relevant
2266 ptr next character in pattern
2267 options options bits
2268 cd contains pointers to tables etc.
2269
2270 Returns: TRUE if possessifying is wanted
2271 */
2272
2273 static BOOL
2274 check_auto_possessive(int op_code, int item, BOOL utf8, uschar *utf8_char,
2275 const uschar *ptr, int options, compile_data *cd)
2276 {
2277 int next;
2278
2279 /* Skip whitespace and comments in extended mode */
2280
2281 if ((options & PCRE_EXTENDED) != 0)
2282 {
2283 for (;;)
2284 {
2285 while ((cd->ctypes[*ptr] & ctype_space) != 0) ptr++;
2286 if (*ptr == CHAR_NUMBER_SIGN)
2287 {
2288 while (*(++ptr) != 0)
2289 if (IS_NEWLINE(ptr)) { ptr += cd->nllen; break; }
2290 }
2291 else break;
2292 }
2293 }
2294
2295 /* If the next item is one that we can handle, get its value. A non-negative
2296 value is a character, a negative value is an escape value. */
2297
2298 if (*ptr == CHAR_BACKSLASH)
2299 {
2300 int temperrorcode = 0;
2301 next = check_escape(&ptr, &temperrorcode, cd->bracount, options, FALSE);
2302 if (temperrorcode != 0) return FALSE;
2303 ptr++; /* Point after the escape sequence */
2304 }
2305
2306 else if ((cd->ctypes[*ptr] & ctype_meta) == 0)
2307 {
2308 #ifdef SUPPORT_UTF8
2309 if (utf8) { GETCHARINC(next, ptr); } else
2310 #endif
2311 next = *ptr++;
2312 }
2313
2314 else return FALSE;
2315
2316 /* Skip whitespace and comments in extended mode */
2317
2318 if ((options & PCRE_EXTENDED) != 0)
2319 {
2320 for (;;)
2321 {
2322 while ((cd->ctypes[*ptr] & ctype_space) != 0) ptr++;
2323 if (*ptr == CHAR_NUMBER_SIGN)
2324 {
2325 while (*(++ptr) != 0)
2326 if (IS_NEWLINE(ptr)) { ptr += cd->nllen; break; }
2327 }
2328 else break;
2329 }
2330 }
2331
2332 /* If the next thing is itself optional, we have to give up. */
2333
2334 if (*ptr == CHAR_ASTERISK || *ptr == CHAR_QUESTION_MARK ||
2335 strncmp((char *)ptr, STR_LEFT_CURLY_BRACKET STR_0 STR_COMMA, 3) == 0)
2336 return FALSE;
2337
2338 /* Now compare the next item with the previous opcode. If the previous is a
2339 positive single character match, "item" either contains the character or, if
2340 "item" is greater than 127 in utf8 mode, the character's bytes are in
2341 utf8_char. */
2342
2343
2344 /* Handle cases when the next item is a character. */
2345
2346 if (next >= 0) switch(op_code)
2347 {
2348 case OP_CHAR:
2349 #ifdef SUPPORT_UTF8
2350 if (utf8 && item > 127) { GETCHAR(item, utf8_char); }
2351 #else
2352 (void)(utf8_char); /* Keep compiler happy by referencing function argument */
2353 #endif
2354 return item != next;
2355
2356 /* For CHARNC (caseless character) we must check the other case. If we have
2357 Unicode property support, we can use it to test the other case of
2358 high-valued characters. */
2359
2360 case OP_CHARNC:
2361 #ifdef SUPPORT_UTF8
2362 if (utf8 && item > 127) { GETCHAR(item, utf8_char); }
2363 #endif
2364 if (item == next) return FALSE;
2365 #ifdef SUPPORT_UTF8
2366 if (utf8)
2367 {
2368 unsigned int othercase;
2369 if (next < 128) othercase = cd->fcc[next]; else
2370 #ifdef SUPPORT_UCP
2371 othercase = UCD_OTHERCASE((unsigned int)next);
2372 #else
2373 othercase = NOTACHAR;
2374 #endif
2375 return (unsigned int)item != othercase;
2376 }
2377 else
2378 #endif /* SUPPORT_UTF8 */
2379 return (item != cd->fcc[next]); /* Non-UTF-8 mode */
2380
2381 /* For OP_NOT, "item" must be a single-byte character. */
2382
2383 case OP_NOT:
2384 if (item == next) return TRUE;
2385 if ((options & PCRE_CASELESS) == 0) return FALSE;
2386 #ifdef SUPPORT_UTF8
2387 if (utf8)
2388 {
2389 unsigned int othercase;
2390 if (next < 128) othercase = cd->fcc[next]; else
2391 #ifdef SUPPORT_UCP
2392 othercase = UCD_OTHERCASE(next);
2393 #else
2394 othercase = NOTACHAR;
2395 #endif
2396 return (unsigned int)item == othercase;
2397 }
2398 else
2399 #endif /* SUPPORT_UTF8 */
2400 return (item == cd->fcc[next]); /* Non-UTF-8 mode */
2401
2402 case OP_DIGIT:
2403 return next > 127 || (cd->ctypes[next] & ctype_digit) == 0;
2404
2405 case OP_NOT_DIGIT:
2406 return next <= 127 && (cd->ctypes[next] & ctype_digit) != 0;
2407
2408 case OP_WHITESPACE:
2409 return next > 127 || (cd->ctypes[next] & ctype_space) == 0;
2410
2411 case OP_NOT_WHITESPACE:
2412 return next <= 127 && (cd->ctypes[next] & ctype_space) != 0;
2413
2414 case OP_WORDCHAR:
2415 return next > 127 || (cd->ctypes[next] & ctype_word) == 0;
2416
2417 case OP_NOT_WORDCHAR:
2418 return next <= 127 && (cd->ctypes[next] & ctype_word) != 0;
2419
2420 case OP_HSPACE:
2421 case OP_NOT_HSPACE:
2422 switch(next)
2423 {
2424 case 0x09:
2425 case 0x20:
2426 case 0xa0:
2427 case 0x1680:
2428 case 0x180e:
2429 case 0x2000:
2430 case 0x2001:
2431 case 0x2002:
2432 case 0x2003:
2433 case 0x2004:
2434 case 0x2005:
2435 case 0x2006:
2436 case 0x2007:
2437 case 0x2008:
2438 case 0x2009:
2439 case 0x200A:
2440 case 0x202f:
2441 case 0x205f:
2442 case 0x3000:
2443 return op_code != OP_HSPACE;
2444 default:
2445 return op_code == OP_HSPACE;
2446 }
2447
2448 case OP_VSPACE:
2449 case OP_NOT_VSPACE:
2450 switch(next)
2451 {
2452 case 0x0a:
2453 case 0x0b:
2454 case 0x0c:
2455 case 0x0d:
2456 case 0x85:
2457 case 0x2028:
2458 case 0x2029:
2459 return op_code != OP_VSPACE;
2460 default:
2461 return op_code == OP_VSPACE;
2462 }
2463
2464 default:
2465 return FALSE;
2466 }
2467
2468
2469 /* Handle the case when the next item is \d, \s, etc. */
2470
2471 switch(op_code)
2472 {
2473 case OP_CHAR:
2474 case OP_CHARNC:
2475 #ifdef SUPPORT_UTF8
2476 if (utf8 && item > 127) { GETCHAR(item, utf8_char); }
2477 #endif
2478 switch(-next)
2479 {
2480 case ESC_d:
2481 return item > 127 || (cd->ctypes[item] & ctype_digit) == 0;
2482
2483 case ESC_D:
2484 return item <= 127 && (cd->ctypes[item] & ctype_digit) != 0;
2485
2486 case ESC_s:
2487 return item > 127 || (cd->ctypes[item] & ctype_space) == 0;
2488
2489 case ESC_S:
2490 return item <= 127 && (cd->ctypes[item] & ctype_space) != 0;
2491
2492 case ESC_w:
2493 return item > 127 || (cd->ctypes[item] & ctype_word) == 0;
2494
2495 case ESC_W:
2496 return item <= 127 && (cd->ctypes[item] & ctype_word) != 0;
2497
2498 case ESC_h:
2499 case ESC_H:
2500 switch(item)
2501 {
2502 case 0x09:
2503 case 0x20:
2504 case 0xa0:
2505 case 0x1680:
2506 case 0x180e:
2507 case 0x2000:
2508 case 0x2001:
2509 case 0x2002:
2510 case 0x2003:
2511 case 0x2004:
2512 case 0x2005:
2513 case 0x2006:
2514 case 0x2007:
2515 case 0x2008:
2516 case 0x2009:
2517 case 0x200A:
2518 case 0x202f:
2519 case 0x205f:
2520 case 0x3000:
2521 return -next != ESC_h;
2522 default:
2523 return -next == ESC_h;
2524 }
2525
2526 case ESC_v:
2527 case ESC_V:
2528 switch(item)
2529 {
2530 case 0x0a:
2531 case 0x0b:
2532 case 0x0c:
2533 case 0x0d:
2534 case 0x85:
2535 case 0x2028:
2536 case 0x2029:
2537 return -next != ESC_v;
2538 default:
2539 return -next == ESC_v;
2540 }
2541
2542 default:
2543 return FALSE;
2544 }
2545
2546 case OP_DIGIT:
2547 return next == -ESC_D || next == -ESC_s || next == -ESC_W ||
2548 next == -ESC_h || next == -ESC_v;
2549
2550 case OP_NOT_DIGIT:
2551 return next == -ESC_d;
2552
2553 case OP_WHITESPACE:
2554 return next == -ESC_S || next == -ESC_d || next == -ESC_w;
2555
2556 case OP_NOT_WHITESPACE:
2557 return next == -ESC_s || next == -ESC_h || next == -ESC_v;
2558
2559 case OP_HSPACE:
2560 return next == -ESC_S || next == -ESC_H || next == -ESC_d || next == -ESC_w;
2561
2562 case OP_NOT_HSPACE:
2563 return next == -ESC_h;
2564
2565 /* Can't have \S in here because VT matches \S (Perl anomaly) */
2566 case OP_VSPACE:
2567 return next == -ESC_V || next == -ESC_d || next == -ESC_w;
2568
2569 case OP_NOT_VSPACE:
2570 return next == -ESC_v;
2571
2572 case OP_WORDCHAR:
2573 return next == -ESC_W || next == -ESC_s || next == -ESC_h || next == -ESC_v;
2574
2575 case OP_NOT_WORDCHAR:
2576 return next == -ESC_w || next == -ESC_d;
2577
2578 default:
2579 return FALSE;
2580 }
2581
2582 /* Control does not reach here */
2583 }
2584
2585
2586
2587 /*************************************************
2588 * Compile one branch *
2589 *************************************************/
2590
2591 /* Scan the pattern, compiling it into the a vector. If the options are
2592 changed during the branch, the pointer is used to change the external options
2593 bits. This function is used during the pre-compile phase when we are trying
2594 to find out the amount of memory needed, as well as during the real compile
2595 phase. The value of lengthptr distinguishes the two phases.
2596
2597 Arguments:
2598 optionsptr pointer to the option bits
2599 codeptr points to the pointer to the current code point
2600 ptrptr points to the current pattern pointer
2601 errorcodeptr points to error code variable
2602 firstbyteptr set to initial literal character, or < 0 (REQ_UNSET, REQ_NONE)
2603 reqbyteptr set to the last literal character required, else < 0
2604 bcptr points to current branch chain
2605 cd contains pointers to tables etc.
2606 lengthptr NULL during the real compile phase
2607 points to length accumulator during pre-compile phase
2608
2609 Returns: TRUE on success
2610 FALSE, with *errorcodeptr set non-zero on error
2611 */
2612
2613 static BOOL
2614 compile_branch(int *optionsptr, uschar **codeptr, const uschar **ptrptr,
2615 int *errorcodeptr, int *firstbyteptr, int *reqbyteptr, branch_chain *bcptr,
2616 compile_data *cd, int *lengthptr)
2617 {
2618 int repeat_type, op_type;
2619 int repeat_min = 0, repeat_max = 0; /* To please picky compilers */
2620 int bravalue = 0;
2621 int greedy_default, greedy_non_default;
2622 int firstbyte, reqbyte;
2623 int zeroreqbyte, zerofirstbyte;
2624 int req_caseopt, reqvary, tempreqvary;
2625 int options = *optionsptr;
2626 int after_manual_callout = 0;
2627 int length_prevgroup = 0;
2628 register int c;
2629 register uschar *code = *codeptr;
2630 uschar *last_code = code;
2631 uschar *orig_code = code;
2632 uschar *tempcode;
2633 BOOL inescq = FALSE;
2634 BOOL groupsetfirstbyte = FALSE;
2635 const uschar *ptr = *ptrptr;
2636 const uschar *tempptr;
2637 uschar *previous = NULL;
2638 uschar *previous_callout = NULL;
2639 uschar *save_hwm = NULL;
2640 uschar classbits[32];
2641
2642 #ifdef SUPPORT_UTF8
2643 BOOL class_utf8;
2644 BOOL utf8 = (options & PCRE_UTF8) != 0;
2645 uschar *class_utf8data;
2646 uschar *class_utf8data_base;
2647 uschar utf8_char[6];
2648 #else
2649 BOOL utf8 = FALSE;
2650 uschar *utf8_char = NULL;
2651 #endif
2652
2653 #ifdef DEBUG
2654 if (lengthptr != NULL) DPRINTF((">> start branch\n"));
2655 #endif
2656
2657 /* Set up the default and non-default settings for greediness */
2658
2659 greedy_default = ((options & PCRE_UNGREEDY) != 0);
2660 greedy_non_default = greedy_default ^ 1;
2661
2662 /* Initialize no first byte, no required byte. REQ_UNSET means "no char
2663 matching encountered yet". It gets changed to REQ_NONE if we hit something that
2664 matches a non-fixed char first char; reqbyte just remains unset if we never
2665 find one.
2666
2667 When we hit a repeat whose minimum is zero, we may have to adjust these values
2668 to take the zero repeat into account. This is implemented by setting them to
2669 zerofirstbyte and zeroreqbyte when such a repeat is encountered. The individual
2670 item types that can be repeated set these backoff variables appropriately. */
2671
2672 firstbyte = reqbyte = zerofirstbyte = zeroreqbyte = REQ_UNSET;
2673
2674 /* The variable req_caseopt contains either the REQ_CASELESS value or zero,
2675 according to the current setting of the caseless flag. REQ_CASELESS is a bit
2676 value > 255. It is added into the firstbyte or reqbyte variables to record the
2677 case status of the value. This is used only for ASCII characters. */
2678
2679 req_caseopt = ((options & PCRE_CASELESS) != 0)? REQ_CASELESS : 0;
2680
2681 /* Switch on next character until the end of the branch */
2682
2683 for (;; ptr++)
2684 {
2685 BOOL negate_class;
2686 BOOL should_flip_negation;
2687 BOOL possessive_quantifier;
2688 BOOL is_quantifier;
2689 BOOL is_recurse;
2690 BOOL reset_bracount;
2691 int class_charcount;
2692 int class_lastchar;
2693 int newoptions;
2694 int recno;
2695 int refsign;
2696 int skipbytes;
2697 int subreqbyte;
2698 int subfirstbyte;
2699 int terminator;
2700 int mclength;
2701 uschar mcbuffer[8];
2702
2703 /* Get next byte in the pattern */
2704
2705 c = *ptr;
2706
2707 /* If we are in the pre-compile phase, accumulate the length used for the
2708 previous cycle of this loop. */
2709
2710 if (lengthptr != NULL)
2711 {
2712 #ifdef DEBUG
2713 if (code > cd->hwm) cd->hwm = code; /* High water info */
2714 #endif
2715 if (code > cd->start_workspace + COMPILE_WORK_SIZE) /* Check for overrun */
2716 {
2717 *errorcodeptr = ERR52;
2718 goto FAILED;
2719 }
2720
2721 /* There is at least one situation where code goes backwards: this is the
2722 case of a zero quantifier after a class (e.g. [ab]{0}). At compile time,
2723 the class is simply eliminated. However, it is created first, so we have to
2724 allow memory for it. Therefore, don't ever reduce the length at this point.
2725 */
2726
2727 if (code < last_code) code = last_code;
2728
2729 /* Paranoid check for integer overflow */
2730
2731 if (OFLOW_MAX - *lengthptr < code - last_code)
2732 {
2733 *errorcodeptr = ERR20;
2734 goto FAILED;
2735 }
2736
2737 *lengthptr += code - last_code;
2738 DPRINTF(("length=%d added %d c=%c\n", *lengthptr, code - last_code, c));
2739
2740 /* If "previous" is set and it is not at the start of the work space, move
2741 it back to there, in order to avoid filling up the work space. Otherwise,
2742 if "previous" is NULL, reset the current code pointer to the start. */
2743
2744 if (previous != NULL)
2745 {
2746 if (previous > orig_code)
2747 {
2748 memmove(orig_code, previous, code - previous);
2749 code -= previous - orig_code;
2750 previous = orig_code;
2751 }
2752 }
2753 else code = orig_code;
2754
2755 /* Remember where this code item starts so we can pick up the length
2756 next time round. */
2757
2758 last_code = code;
2759 }
2760
2761 /* In the real compile phase, just check the workspace used by the forward
2762 reference list. */
2763
2764 else if (cd->hwm > cd->start_workspace + COMPILE_WORK_SIZE)
2765 {
2766 *errorcodeptr = ERR52;
2767 goto FAILED;
2768 }
2769
2770 /* If in \Q...\E, check for the end; if not, we have a literal */
2771
2772 if (inescq && c != 0)
2773 {
2774 if (c == CHAR_BACKSLASH && ptr[1] == CHAR_E)
2775 {
2776 inescq = FALSE;
2777 ptr++;
2778 continue;
2779 }
2780 else
2781 {
2782 if (previous_callout != NULL)
2783 {
2784 if (lengthptr == NULL) /* Don't attempt in pre-compile phase */
2785 complete_callout(previous_callout, ptr, cd);
2786 previous_callout = NULL;
2787 }
2788 if ((options & PCRE_AUTO_CALLOUT) != 0)
2789 {
2790 previous_callout = code;
2791 code = auto_callout(code, ptr, cd);
2792 }
2793 goto NORMAL_CHAR;
2794 }
2795 }
2796
2797 /* Fill in length of a previous callout, except when the next thing is
2798 a quantifier. */
2799
2800 is_quantifier =
2801 c == CHAR_ASTERISK || c == CHAR_PLUS || c == CHAR_QUESTION_MARK ||
2802 (c == CHAR_LEFT_CURLY_BRACKET && is_counted_repeat(ptr+1));
2803
2804 if (!is_quantifier && previous_callout != NULL &&
2805 after_manual_callout-- <= 0)
2806 {
2807 if (lengthptr == NULL) /* Don't attempt in pre-compile phase */
2808 complete_callout(previous_callout, ptr, cd);
2809 previous_callout = NULL;
2810 }
2811
2812 /* In extended mode, skip white space and comments */
2813
2814 if ((options & PCRE_EXTENDED) != 0)
2815 {
2816 if ((cd->ctypes[c] & ctype_space) != 0) continue;
2817 if (c == CHAR_NUMBER_SIGN)
2818 {
2819 while (*(++ptr) != 0)
2820 {
2821 if (IS_NEWLINE(ptr)) { ptr += cd->nllen - 1; break; }
2822 }
2823 if (*ptr != 0) continue;
2824
2825 /* Else fall through to handle end of string */
2826 c = 0;
2827 }
2828 }
2829
2830 /* No auto callout for quantifiers. */
2831
2832 if ((options & PCRE_AUTO_CALLOUT) != 0 && !is_quantifier)
2833 {
2834 previous_callout = code;
2835 code = auto_callout(code, ptr, cd);
2836 }
2837
2838 switch(c)
2839 {
2840 /* ===================================================================*/
2841 case 0: /* The branch terminates at string end */
2842 case CHAR_VERTICAL_LINE: /* or | or ) */
2843 case CHAR_RIGHT_PARENTHESIS:
2844 *firstbyteptr = firstbyte;
2845 *reqbyteptr = reqbyte;
2846 *codeptr = code;
2847 *ptrptr = ptr;
2848 if (lengthptr != NULL)
2849 {
2850 if (OFLOW_MAX - *lengthptr < code - last_code)
2851 {
2852 *errorcodeptr = ERR20;
2853 goto FAILED;
2854 }
2855 *lengthptr += code - last_code; /* To include callout length */
2856 DPRINTF((">> end branch\n"));
2857 }
2858 return TRUE;
2859
2860
2861 /* ===================================================================*/
2862 /* Handle single-character metacharacters. In multiline mode, ^ disables
2863 the setting of any following char as a first character. */
2864
2865 case CHAR_CIRCUMFLEX_ACCENT:
2866 if ((options & PCRE_MULTILINE) != 0)
2867 {
2868 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
2869 }
2870 previous = NULL;
2871 *code++ = OP_CIRC;
2872 break;
2873
2874 case CHAR_DOLLAR_SIGN:
2875 previous = NULL;
2876 *code++ = OP_DOLL;
2877 break;
2878
2879 /* There can never be a first char if '.' is first, whatever happens about
2880 repeats. The value of reqbyte doesn't change either. */
2881
2882 case CHAR_DOT:
2883 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
2884 zerofirstbyte = firstbyte;
2885 zeroreqbyte = reqbyte;
2886 previous = code;
2887 *code++ = ((options & PCRE_DOTALL) != 0)? OP_ALLANY: OP_ANY;
2888 break;
2889
2890
2891 /* ===================================================================*/
2892 /* Character classes. If the included characters are all < 256, we build a
2893 32-byte bitmap of the permitted characters, except in the special case
2894 where there is only one such character. For negated classes, we build the
2895 map as usual, then invert it at the end. However, we use a different opcode
2896 so that data characters > 255 can be handled correctly.
2897
2898 If the class contains characters outside the 0-255 range, a different
2899 opcode is compiled. It may optionally have a bit map for characters < 256,
2900 but those above are are explicitly listed afterwards. A flag byte tells
2901 whether the bitmap is present, and whether this is a negated class or not.
2902
2903 In JavaScript compatibility mode, an isolated ']' causes an error. In
2904 default (Perl) mode, it is treated as a data character. */
2905
2906 case CHAR_RIGHT_SQUARE_BRACKET:
2907 if ((cd->external_options & PCRE_JAVASCRIPT_COMPAT) != 0)
2908 {
2909 *errorcodeptr = ERR64;
2910 goto FAILED;
2911 }
2912 goto NORMAL_CHAR;
2913
2914 case CHAR_LEFT_SQUARE_BRACKET:
2915 previous = code;
2916
2917 /* PCRE supports POSIX class stuff inside a class. Perl gives an error if
2918 they are encountered at the top level, so we'll do that too. */
2919
2920 if ((ptr[1] == CHAR_COLON || ptr[1] == CHAR_DOT ||
2921 ptr[1] == CHAR_EQUALS_SIGN) &&
2922 check_posix_syntax(ptr, &tempptr))
2923 {
2924 *errorcodeptr = (ptr[1] == CHAR_COLON)? ERR13 : ERR31;
2925 goto FAILED;
2926 }
2927
2928 /* If the first character is '^', set the negation flag and skip it. Also,
2929 if the first few characters (either before or after ^) are \Q\E or \E we
2930 skip them too. This makes for compatibility with Perl. */
2931
2932 negate_class = FALSE;
2933 for (;;)
2934 {
2935 c = *(++ptr);
2936 if (c == CHAR_BACKSLASH)
2937 {
2938 if (ptr[1] == CHAR_E)
2939 ptr++;
2940 else if (strncmp((const char *)ptr+1,
2941 STR_Q STR_BACKSLASH STR_E, 3) == 0)
2942 ptr += 3;
2943 else
2944 break;
2945 }
2946 else if (!negate_class && c == CHAR_CIRCUMFLEX_ACCENT)
2947 negate_class = TRUE;
2948 else break;
2949 }
2950
2951 /* Empty classes are allowed in JavaScript compatibility mode. Otherwise,
2952 an initial ']' is taken as a data character -- the code below handles
2953 that. In JS mode, [] must always fail, so generate OP_FAIL, whereas
2954 [^] must match any character, so generate OP_ALLANY. */
2955
2956 if (c == CHAR_RIGHT_SQUARE_BRACKET &&
2957 (cd->external_options & PCRE_JAVASCRIPT_COMPAT) != 0)
2958 {
2959 *code++ = negate_class? OP_ALLANY : OP_FAIL;
2960 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
2961 zerofirstbyte = firstbyte;
2962 break;
2963 }
2964
2965 /* If a class contains a negative special such as \S, we need to flip the
2966 negation flag at the end, so that support for characters > 255 works
2967 correctly (they are all included in the class). */
2968
2969 should_flip_negation = FALSE;
2970
2971 /* Keep a count of chars with values < 256 so that we can optimize the case
2972 of just a single character (as long as it's < 256). However, For higher
2973 valued UTF-8 characters, we don't yet do any optimization. */
2974
2975 class_charcount = 0;
2976 class_lastchar = -1;
2977
2978 /* Initialize the 32-char bit map to all zeros. We build the map in a
2979 temporary bit of memory, in case the class contains only 1 character (less
2980 than 256), because in that case the compiled code doesn't use the bit map.
2981 */
2982
2983 memset(classbits, 0, 32 * sizeof(uschar));
2984
2985 #ifdef SUPPORT_UTF8
2986 class_utf8 = FALSE; /* No chars >= 256 */
2987 class_utf8data = code + LINK_SIZE + 2; /* For UTF-8 items */
2988 class_utf8data_base = class_utf8data; /* For resetting in pass 1 */
2989 #endif
2990
2991 /* Process characters until ] is reached. By writing this as a "do" it
2992 means that an initial ] is taken as a data character. At the start of the
2993 loop, c contains the first byte of the character. */
2994
2995 if (c != 0) do
2996 {
2997 const uschar *oldptr;
2998
2999 #ifdef SUPPORT_UTF8
3000 if (utf8 && c > 127)
3001 { /* Braces are required because the */
3002 GETCHARLEN(c, ptr, ptr); /* macro generates multiple statements */
3003 }
3004
3005 /* In the pre-compile phase, accumulate the length of any UTF-8 extra
3006 data and reset the pointer. This is so that very large classes that
3007 contain a zillion UTF-8 characters no longer overwrite the work space
3008 (which is on the stack). */
3009
3010 if (lengthptr != NULL)
3011 {
3012 *lengthptr += class_utf8data - class_utf8data_base;
3013 class_utf8data = class_utf8data_base;
3014 }
3015
3016 #endif
3017
3018 /* Inside \Q...\E everything is literal except \E */
3019
3020 if (inescq)
3021 {
3022 if (c == CHAR_BACKSLASH && ptr[1] == CHAR_E) /* If we are at \E */
3023 {
3024 inescq = FALSE; /* Reset literal state */
3025 ptr++; /* Skip the 'E' */
3026 continue; /* Carry on with next */
3027 }
3028 goto CHECK_RANGE; /* Could be range if \E follows */
3029 }
3030
3031 /* Handle POSIX class names. Perl allows a negation extension of the
3032 form [:^name:]. A square bracket that doesn't match the syntax is
3033 treated as a literal. We also recognize the POSIX constructions
3034 [.ch.] and [=ch=] ("collating elements") and fault them, as Perl
3035 5.6 and 5.8 do. */
3036
3037 if (c == CHAR_LEFT_SQUARE_BRACKET &&
3038 (ptr[1] == CHAR_COLON || ptr[1] == CHAR_DOT ||
3039 ptr[1] == CHAR_EQUALS_SIGN) && check_posix_syntax(ptr, &tempptr))
3040 {
3041 BOOL local_negate = FALSE;
3042 int posix_class, taboffset, tabopt;
3043 register const uschar *cbits = cd->cbits;
3044 uschar pbits[32];
3045
3046 if (ptr[1] != CHAR_COLON)
3047 {
3048 *errorcodeptr = ERR31;
3049 goto FAILED;
3050 }
3051
3052 ptr += 2;
3053 if (*ptr == CHAR_CIRCUMFLEX_ACCENT)
3054 {
3055 local_negate = TRUE;
3056 should_flip_negation = TRUE; /* Note negative special */
3057 ptr++;
3058 }
3059
3060 posix_class = check_posix_name(ptr, tempptr - ptr);
3061 if (posix_class < 0)
3062 {
3063 *errorcodeptr = ERR30;
3064 goto FAILED;
3065 }
3066
3067 /* If matching is caseless, upper and lower are converted to
3068 alpha. This relies on the fact that the class table starts with
3069 alpha, lower, upper as the first 3 entries. */
3070
3071 if ((options & PCRE_CASELESS) != 0 && posix_class <= 2)
3072 posix_class = 0;
3073
3074 /* We build the bit map for the POSIX class in a chunk of local store
3075 because we may be adding and subtracting from it, and we don't want to
3076 subtract bits that may be in the main map already. At the end we or the
3077 result into the bit map that is being built. */
3078
3079 posix_class *= 3;
3080
3081 /* Copy in the first table (always present) */
3082
3083 memcpy(pbits, cbits + posix_class_maps[posix_class],
3084 32 * sizeof(uschar));
3085
3086 /* If there is a second table, add or remove it as required. */
3087
3088 taboffset = posix_class_maps[posix_class + 1];
3089 tabopt = posix_class_maps[posix_class + 2];
3090
3091 if (taboffset >= 0)
3092 {
3093 if (tabopt >= 0)
3094 for (c = 0; c < 32; c++) pbits[c] |= cbits[c + taboffset];
3095 else
3096 for (c = 0; c < 32; c++) pbits[c] &= ~cbits[c + taboffset];
3097 }
3098
3099 /* Not see if we need to remove any special characters. An option
3100 value of 1 removes vertical space and 2 removes underscore. */
3101
3102 if (tabopt < 0) tabopt = -tabopt;
3103 if (tabopt == 1) pbits[1] &= ~0x3c;
3104 else if (tabopt == 2) pbits[11] &= 0x7f;
3105
3106 /* Add the POSIX table or its complement into the main table that is
3107 being built and we are done. */
3108
3109 if (local_negate)
3110 for (c = 0; c < 32; c++) classbits[c] |= ~pbits[c];
3111 else
3112 for (c = 0; c < 32; c++) classbits[c] |= pbits[c];
3113
3114 ptr = tempptr + 1;
3115 class_charcount = 10; /* Set > 1; assumes more than 1 per class */
3116 continue; /* End of POSIX syntax handling */
3117 }
3118
3119 /* Backslash may introduce a single character, or it may introduce one
3120 of the specials, which just set a flag. The sequence \b is a special
3121 case. Inside a class (and only there) it is treated as backspace.
3122 Elsewhere it marks a word boundary. Other escapes have preset maps ready
3123 to 'or' into the one we are building. We assume they have more than one
3124 character in them, so set class_charcount bigger than one. */
3125
3126 if (c == CHAR_BACKSLASH)
3127 {
3128 c = check_escape(&ptr, errorcodeptr, cd->bracount, options, TRUE);
3129 if (*errorcodeptr != 0) goto FAILED;
3130
3131 if (-c == ESC_b) c = CHAR_BS; /* \b is backspace in a class */
3132 else if (-c == ESC_X) c = CHAR_X; /* \X is literal X in a class */
3133 else if (-c == ESC_R) c = CHAR_R; /* \R is literal R in a class */
3134 else if (-c == ESC_Q) /* Handle start of quoted string */
3135 {
3136 if (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
3137 {
3138 ptr += 2; /* avoid empty string */
3139 }
3140 else inescq = TRUE;
3141 continue;
3142 }
3143 else if (-c == ESC_E) continue; /* Ignore orphan \E */
3144
3145 if (c < 0)
3146 {
3147 register const uschar *cbits = cd->cbits;
3148 class_charcount += 2; /* Greater than 1 is what matters */
3149
3150 /* Save time by not doing this in the pre-compile phase. */
3151
3152 if (lengthptr == NULL) switch (-c)
3153 {
3154 case ESC_d:
3155 for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_digit];
3156 continue;
3157
3158 case ESC_D:
3159 should_flip_negation = TRUE;
3160 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_digit];
3161 continue;
3162
3163 case ESC_w:
3164 for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_word];
3165 continue;
3166
3167 case ESC_W:
3168 should_flip_negation = TRUE;
3169 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_word];
3170 continue;
3171
3172 case ESC_s:
3173 for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_space];
3174 classbits[1] &= ~0x08; /* Perl 5.004 onwards omits VT from \s */
3175 continue;
3176
3177 case ESC_S:
3178 should_flip_negation = TRUE;
3179 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_space];
3180 classbits[1] |= 0x08; /* Perl 5.004 onwards omits VT from \s */
3181 continue;
3182
3183 default: /* Not recognized; fall through */
3184 break; /* Need "default" setting to stop compiler warning. */
3185 }
3186
3187 /* In the pre-compile phase, just do the recognition. */
3188
3189 else if (c == -ESC_d || c == -ESC_D || c == -ESC_w ||
3190 c == -ESC_W || c == -ESC_s || c == -ESC_S) continue;
3191
3192 /* We need to deal with \H, \h, \V, and \v in both phases because
3193 they use extra memory. */
3194
3195 if (-c == ESC_h)
3196 {
3197 SETBIT(classbits, 0x09); /* VT */
3198 SETBIT(classbits, 0x20); /* SPACE */
3199 SETBIT(classbits, 0xa0); /* NSBP */
3200 #ifdef SUPPORT_UTF8
3201 if (utf8)
3202 {
3203 class_utf8 = TRUE;
3204 *class_utf8data++ = XCL_SINGLE;
3205 class_utf8data += _pcre_ord2utf8(0x1680, class_utf8data);
3206 *class_utf8data++ = XCL_SINGLE;
3207 class_utf8data += _pcre_ord2utf8(0x180e, class_utf8data);
3208 *class_utf8data++ = XCL_RANGE;
3209 class_utf8data += _pcre_ord2utf8(0x2000, class_utf8data);
3210 class_utf8data += _pcre_ord2utf8(0x200A, class_utf8data);
3211 *class_utf8data++ = XCL_SINGLE;
3212 class_utf8data += _pcre_ord2utf8(0x202f, class_utf8data);
3213 *class_utf8data++ = XCL_SINGLE;
3214 class_utf8data += _pcre_ord2utf8(0x205f, class_utf8data);
3215 *class_utf8data++ = XCL_SINGLE;
3216 class_utf8data += _pcre_ord2utf8(0x3000, class_utf8data);
3217 }
3218 #endif
3219 continue;
3220 }
3221
3222 if (-c == ESC_H)
3223 {
3224 for (c = 0; c < 32; c++)
3225 {
3226 int x = 0xff;
3227 switch (c)
3228 {
3229 case 0x09/8: x ^= 1 << (0x09%8); break;
3230 case 0x20/8: x ^= 1 << (0x20%8); break;
3231 case 0xa0/8: x ^= 1 << (0xa0%8); break;
3232 default: break;
3233 }
3234 classbits[c] |= x;
3235 }
3236
3237 #ifdef SUPPORT_UTF8
3238 if (utf8)
3239 {
3240 class_utf8 = TRUE;
3241 *class_utf8data++ = XCL_RANGE;
3242 class_utf8data += _pcre_ord2utf8(0x0100, class_utf8data);
3243 class_utf8data += _pcre_ord2utf8(0x167f, class_utf8data);
3244 *class_utf8data++ = XCL_RANGE;
3245 class_utf8data += _pcre_ord2utf8(0x1681, class_utf8data);
3246 class_utf8data += _pcre_ord2utf8(0x180d, class_utf8data);
3247 *class_utf8data++ = XCL_RANGE;
3248 class_utf8data += _pcre_ord2utf8(0x180f, class_utf8data);
3249 class_utf8data += _pcre_ord2utf8(0x1fff, class_utf8data);
3250 *class_utf8data++ = XCL_RANGE;
3251 class_utf8data += _pcre_ord2utf8(0x200B, class_utf8data);
3252 class_utf8data += _pcre_ord2utf8(0x202e, class_utf8data);
3253 *class_utf8data++ = XCL_RANGE;
3254 class_utf8data += _pcre_ord2utf8(0x2030, class_utf8data);
3255 class_utf8data += _pcre_ord2utf8(0x205e, class_utf8data);
3256 *class_utf8data++ = XCL_RANGE;
3257 class_utf8data += _pcre_ord2utf8(0x2060, class_utf8data);
3258 class_utf8data += _pcre_ord2utf8(0x2fff, class_utf8data);
3259 *class_utf8data++ = XCL_RANGE;
3260 class_utf8data += _pcre_ord2utf8(0x3001, class_utf8data);
3261 class_utf8data += _pcre_ord2utf8(0x7fffffff, class_utf8data);
3262 }
3263 #endif
3264 continue;
3265 }
3266
3267 if (-c == ESC_v)
3268 {
3269 SETBIT(classbits, 0x0a); /* LF */
3270 SETBIT(classbits, 0x0b); /* VT */
3271 SETBIT(classbits, 0x0c); /* FF */
3272 SETBIT(classbits, 0x0d); /* CR */
3273 SETBIT(classbits, 0x85); /* NEL */
3274 #ifdef SUPPORT_UTF8
3275 if (utf8)
3276 {
3277 class_utf8 = TRUE;
3278 *class_utf8data++ = XCL_RANGE;
3279 class_utf8data += _pcre_ord2utf8(0x2028, class_utf8data);
3280 class_utf8data += _pcre_ord2utf8(0x2029, class_utf8data);
3281 }
3282 #endif
3283 continue;
3284 }
3285
3286 if (-c == ESC_V)
3287 {
3288 for (c = 0; c < 32; c++)
3289 {
3290 int x = 0xff;
3291 switch (c)
3292 {
3293 case 0x0a/8: x ^= 1 << (0x0a%8);
3294 x ^= 1 << (0x0b%8);
3295 x ^= 1 << (0x0c%8);
3296 x ^= 1 << (0x0d%8);
3297 break;
3298 case 0x85/8: x ^= 1 << (0x85%8); break;
3299 default: break;
3300 }
3301 classbits[c] |= x;
3302 }
3303
3304 #ifdef SUPPORT_UTF8
3305 if (utf8)
3306 {
3307 class_utf8 = TRUE;
3308 *class_utf8data++ = XCL_RANGE;
3309 class_utf8data += _pcre_ord2utf8(0x0100, class_utf8data);
3310 class_utf8data += _pcre_ord2utf8(0x2027, class_utf8data);
3311 *class_utf8data++ = XCL_RANGE;
3312 class_utf8data += _pcre_ord2utf8(0x2029, class_utf8data);
3313 class_utf8data += _pcre_ord2utf8(0x7fffffff, class_utf8data);
3314 }
3315 #endif
3316 continue;
3317 }
3318
3319 /* We need to deal with \P and \p in both phases. */
3320
3321 #ifdef SUPPORT_UCP
3322 if (-c == ESC_p || -c == ESC_P)
3323 {
3324 BOOL negated;
3325 int pdata;
3326 int ptype = get_ucp(&ptr, &negated, &pdata, errorcodeptr);
3327 if (ptype < 0) goto FAILED;
3328 class_utf8 = TRUE;
3329 *class_utf8data++ = ((-c == ESC_p) != negated)?
3330 XCL_PROP : XCL_NOTPROP;
3331 *class_utf8data++ = ptype;
3332 *class_utf8data++ = pdata;
3333 class_charcount -= 2; /* Not a < 256 character */
3334 continue;
3335 }
3336 #endif
3337 /* Unrecognized escapes are faulted if PCRE is running in its
3338 strict mode. By default, for compatibility with Perl, they are
3339 treated as literals. */
3340
3341 if ((options & PCRE_EXTRA) != 0)
3342 {
3343 *errorcodeptr = ERR7;
3344 goto FAILED;
3345 }
3346
3347 class_charcount -= 2; /* Undo the default count from above */
3348 c = *ptr; /* Get the final character and fall through */
3349 }
3350
3351 /* Fall through if we have a single character (c >= 0). This may be
3352 greater than 256 in UTF-8 mode. */
3353
3354 } /* End of backslash handling */
3355
3356 /* A single character may be followed by '-' to form a range. However,
3357 Perl does not permit ']' to be the end of the range. A '-' character
3358 at the end is treated as a literal. Perl ignores orphaned \E sequences
3359 entirely. The code for handling \Q and \E is messy. */
3360
3361 CHECK_RANGE:
3362 while (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
3363 {
3364 inescq = FALSE;
3365 ptr += 2;
3366 }
3367
3368 oldptr = ptr;
3369
3370 /* Remember \r or \n */
3371
3372 if (c == CHAR_CR || c == CHAR_NL) cd->external_flags |= PCRE_HASCRORLF;
3373
3374 /* Check for range */
3375
3376 if (!inescq && ptr[1] == CHAR_MINUS)
3377 {
3378 int d;
3379 ptr += 2;
3380 while (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_E) ptr += 2;
3381
3382 /* If we hit \Q (not followed by \E) at this point, go into escaped
3383 mode. */
3384
3385 while (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_Q)
3386 {
3387 ptr += 2;
3388 if (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_E)
3389 { ptr += 2; continue; }
3390 inescq = TRUE;
3391 break;
3392 }
3393
3394 if (*ptr == 0 || (!inescq && *ptr == CHAR_RIGHT_SQUARE_BRACKET))
3395 {
3396 ptr = oldptr;
3397 goto LONE_SINGLE_CHARACTER;
3398 }
3399
3400 #ifdef SUPPORT_UTF8
3401 if (utf8)
3402 { /* Braces are required because the */
3403 GETCHARLEN(d, ptr, ptr); /* macro generates multiple statements */
3404 }
3405 else
3406 #endif
3407 d = *ptr; /* Not UTF-8 mode */
3408
3409 /* The second part of a range can be a single-character escape, but
3410 not any of the other escapes. Perl 5.6 treats a hyphen as a literal
3411 in such circumstances. */
3412
3413 if (!inescq && d == CHAR_BACKSLASH)
3414 {
3415 d = check_escape(&ptr, errorcodeptr, cd->bracount, options, TRUE);
3416 if (*errorcodeptr != 0) goto FAILED;
3417
3418 /* \b is backspace; \X is literal X; \R is literal R; any other
3419 special means the '-' was literal */
3420
3421 if (d < 0)
3422 {
3423 if (d == -ESC_b) d = CHAR_BS;
3424 else if (d == -ESC_X) d = CHAR_X;
3425 else if (d == -ESC_R) d = CHAR_R; else
3426 {
3427 ptr = oldptr;
3428 goto LONE_SINGLE_CHARACTER; /* A few lines below */
3429 }
3430 }
3431 }
3432
3433 /* Check that the two values are in the correct order. Optimize
3434 one-character ranges */
3435
3436 if (d < c)
3437 {
3438 *errorcodeptr = ERR8;
3439 goto FAILED;
3440 }
3441
3442 if (d == c) goto LONE_SINGLE_CHARACTER; /* A few lines below */
3443
3444 /* Remember \r or \n */
3445
3446 if (d == CHAR_CR || d == CHAR_NL) cd->external_flags |= PCRE_HASCRORLF;
3447
3448 /* In UTF-8 mode, if the upper limit is > 255, or > 127 for caseless
3449 matching, we have to use an XCLASS with extra data items. Caseless
3450 matching for characters > 127 is available only if UCP support is
3451 available. */
3452
3453 #ifdef SUPPORT_UTF8
3454 if (utf8 && (d > 255 || ((options & PCRE_CASELESS) != 0 && d > 127)))
3455 {
3456 class_utf8 = TRUE;
3457
3458 /* With UCP support, we can find the other case equivalents of
3459 the relevant characters. There may be several ranges. Optimize how
3460 they fit with the basic range. */
3461
3462 #ifdef SUPPORT_UCP
3463 if ((options & PCRE_CASELESS) != 0)
3464 {
3465 unsigned int occ, ocd;
3466 unsigned int cc = c;
3467 unsigned int origd = d;
3468 while (get_othercase_range(&cc, origd, &occ, &ocd))
3469 {
3470 if (occ >= (unsigned int)c &&
3471 ocd <= (unsigned int)d)
3472 continue; /* Skip embedded ranges */
3473
3474 if (occ < (unsigned int)c &&
3475 ocd >= (unsigned int)c - 1) /* Extend the basic range */
3476 { /* if there is overlap, */
3477 c = occ; /* noting that if occ < c */
3478 continue; /* we can't have ocd > d */
3479 } /* because a subrange is */
3480 if (ocd > (unsigned int)d &&
3481 occ <= (unsigned int)d + 1) /* always shorter than */
3482 { /* the basic range. */
3483 d = ocd;
3484 continue;
3485 }
3486
3487 if (occ == ocd)
3488 {
3489 *class_utf8data++ = XCL_SINGLE;
3490 }
3491 else
3492 {
3493 *class_utf8data++ = XCL_RANGE;
3494 class_utf8data += _pcre_ord2utf8(occ, class_utf8data);
3495 }
3496 class_utf8data += _pcre_ord2utf8(ocd, class_utf8data);
3497 }
3498 }
3499 #endif /* SUPPORT_UCP */
3500
3501 /* Now record the original range, possibly modified for UCP caseless
3502 overlapping ranges. */
3503
3504 *class_utf8data++ = XCL_RANGE;
3505 class_utf8data += _pcre_ord2utf8(c, class_utf8data);
3506 class_utf8data += _pcre_ord2utf8(d, class_utf8data);
3507
3508 /* With UCP support, we are done. Without UCP support, there is no
3509 caseless matching for UTF-8 characters > 127; we can use the bit map
3510 for the smaller ones. */
3511
3512 #ifdef SUPPORT_UCP
3513 continue; /* With next character in the class */
3514 #else
3515 if ((options & PCRE_CASELESS) == 0 || c > 127) continue;
3516
3517 /* Adjust upper limit and fall through to set up the map */
3518
3519 d = 127;
3520
3521 #endif /* SUPPORT_UCP */
3522 }
3523 #endif /* SUPPORT_UTF8 */
3524
3525 /* We use the bit map for all cases when not in UTF-8 mode; else
3526 ranges that lie entirely within 0-127 when there is UCP support; else
3527 for partial ranges without UCP support. */
3528
3529 class_charcount += d - c + 1;
3530 class_lastchar = d;
3531
3532 /* We can save a bit of time by skipping this in the pre-compile. */
3533
3534 if (lengthptr == NULL) for (; c <= d; c++)
3535 {
3536 classbits[c/8] |= (1 << (c&7));
3537 if ((options & PCRE_CASELESS) != 0)
3538 {
3539 int uc = cd->fcc[c]; /* flip case */
3540 classbits[uc/8] |= (1 << (uc&7));
3541 }
3542 }
3543
3544 continue; /* Go get the next char in the class */
3545 }
3546
3547 /* Handle a lone single character - we can get here for a normal
3548 non-escape char, or after \ that introduces a single character or for an
3549 apparent range that isn't. */
3550
3551 LONE_SINGLE_CHARACTER:
3552
3553 /* Handle a character that cannot go in the bit map */
3554
3555 #ifdef SUPPORT_UTF8
3556 if (utf8 && (c > 255 || ((options & PCRE_CASELESS) != 0 && c > 127)))
3557 {
3558 class_utf8 = TRUE;
3559 *class_utf8data++ = XCL_SINGLE;
3560 class_utf8data += _pcre_ord2utf8(c, class_utf8data);
3561
3562 #ifdef SUPPORT_UCP
3563 if ((options & PCRE_CASELESS) != 0)
3564 {
3565 unsigned int othercase;
3566 if ((othercase = UCD_OTHERCASE(c)) != c)
3567 {
3568 *class_utf8data++ = XCL_SINGLE;
3569 class_utf8data += _pcre_ord2utf8(othercase, class_utf8data);
3570 }
3571 }
3572 #endif /* SUPPORT_UCP */
3573
3574 }
3575 else
3576 #endif /* SUPPORT_UTF8 */
3577
3578 /* Handle a single-byte character */
3579 {
3580 classbits[c/8] |= (1 << (c&7));
3581 if ((options & PCRE_CASELESS) != 0)
3582 {
3583 c = cd->fcc[c]; /* flip case */
3584 classbits[c/8] |= (1 << (c&7));
3585 }
3586 class_charcount++;
3587 class_lastchar = c;
3588 }
3589 }
3590
3591 /* Loop until ']' reached. This "while" is the end of the "do" above. */
3592
3593 while ((c = *(++ptr)) != 0 && (c != CHAR_RIGHT_SQUARE_BRACKET || inescq));
3594
3595 if (c == 0) /* Missing terminating ']' */
3596 {
3597 *errorcodeptr = ERR6;
3598 goto FAILED;
3599 }
3600
3601
3602 /* This code has been disabled because it would mean that \s counts as
3603 an explicit \r or \n reference, and that's not really what is wanted. Now
3604 we set the flag only if there is a literal "\r" or "\n" in the class. */
3605
3606 #if 0
3607 /* Remember whether \r or \n are in this class */
3608
3609 if (negate_class)
3610 {
3611 if ((classbits[1] & 0x24) != 0x24) cd->external_flags |= PCRE_HASCRORLF;
3612 }
3613 else
3614 {
3615 if ((classbits[1] & 0x24) != 0) cd->external_flags |= PCRE_HASCRORLF;
3616 }
3617 #endif
3618
3619
3620 /* If class_charcount is 1, we saw precisely one character whose value is
3621 less than 256. As long as there were no characters >= 128 and there was no
3622 use of \p or \P, in other words, no use of any XCLASS features, we can
3623 optimize.
3624
3625 In UTF-8 mode, we can optimize the negative case only if there were no
3626 characters >= 128 because OP_NOT and the related opcodes like OP_NOTSTAR
3627 operate on single-bytes only. This is an historical hangover. Maybe one day
3628 we can tidy these opcodes to handle multi-byte characters.
3629
3630 The optimization throws away the bit map. We turn the item into a
3631 1-character OP_CHAR[NC] if it's positive, or OP_NOT if it's negative. Note
3632 that OP_NOT does not support multibyte characters. In the positive case, it
3633 can cause firstbyte to be set. Otherwise, there can be no first char if
3634 this item is first, whatever repeat count may follow. In the case of
3635 reqbyte, save the previous value for reinstating. */
3636
3637 #ifdef SUPPORT_UTF8
3638 if (class_charcount == 1 && !class_utf8 &&
3639 (!utf8 || !negate_class || class_lastchar < 128))
3640 #else
3641 if (class_charcount == 1)
3642 #endif
3643 {
3644 zeroreqbyte = reqbyte;
3645
3646 /* The OP_NOT opcode works on one-byte characters only. */
3647
3648 if (negate_class)
3649 {
3650 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
3651 zerofirstbyte = firstbyte;
3652 *code++ = OP_NOT;
3653 *code++ = class_lastchar;
3654 break;
3655 }
3656
3657 /* For a single, positive character, get the value into mcbuffer, and
3658 then we can handle this with the normal one-character code. */
3659
3660 #ifdef SUPPORT_UTF8
3661 if (utf8 && class_lastchar > 127)
3662 mclength = _pcre_ord2utf8(class_lastchar, mcbuffer);
3663 else
3664 #endif
3665 {
3666 mcbuffer[0] = class_lastchar;
3667 mclength = 1;
3668 }
3669 goto ONE_CHAR;
3670 } /* End of 1-char optimization */
3671
3672 /* The general case - not the one-char optimization. If this is the first
3673 thing in the branch, there can be no first char setting, whatever the
3674 repeat count. Any reqbyte setting must remain unchanged after any kind of
3675 repeat. */
3676
3677 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
3678 zerofirstbyte = firstbyte;
3679 zeroreqbyte = reqbyte;
3680
3681 /* If there are characters with values > 255, we have to compile an
3682 extended class, with its own opcode, unless there was a negated special
3683 such as \S in the class, because in that case all characters > 255 are in
3684 the class, so any that were explicitly given as well can be ignored. If
3685 (when there are explicit characters > 255 that must be listed) there are no
3686 characters < 256, we can omit the bitmap in the actual compiled code. */
3687
3688 #ifdef SUPPORT_UTF8
3689 if (class_utf8 && !should_flip_negation)
3690 {
3691 *class_utf8data++ = XCL_END; /* Marks the end of extra data */
3692 *code++ = OP_XCLASS;
3693 code += LINK_SIZE;
3694 *code = negate_class? XCL_NOT : 0;
3695
3696 /* If the map is required, move up the extra data to make room for it;
3697 otherwise just move the code pointer to the end of the extra data. */
3698
3699 if (class_charcount > 0)
3700 {
3701 *code++ |= XCL_MAP;
3702 memmove(code + 32, code, class_utf8data - code);
3703 memcpy(code, classbits, 32);
3704 code = class_utf8data + 32;
3705 }
3706 else code = class_utf8data;
3707
3708 /* Now fill in the complete length of the item */
3709
3710 PUT(previous, 1, code - previous);
3711 break; /* End of class handling */
3712 }
3713 #endif
3714
3715 /* If there are no characters > 255, set the opcode to OP_CLASS or
3716 OP_NCLASS, depending on whether the whole class was negated and whether
3717 there were negative specials such as \S in the class. Then copy the 32-byte
3718 map into the code vector, negating it if necessary. */
3719
3720 *code++ = (negate_class == should_flip_negation) ? OP_CLASS : OP_NCLASS;
3721 if (negate_class)
3722 {
3723 if (lengthptr == NULL) /* Save time in the pre-compile phase */
3724 for (c = 0; c < 32; c++) code[c] = ~classbits[c];
3725 }
3726 else
3727 {
3728 memcpy(code, classbits, 32);
3729 }
3730 code += 32;
3731 break;
3732
3733
3734 /* ===================================================================*/
3735 /* Various kinds of repeat; '{' is not necessarily a quantifier, but this
3736 has been tested above. */
3737
3738 case CHAR_LEFT_CURLY_BRACKET:
3739 if (!is_quantifier) goto NORMAL_CHAR;
3740 ptr = read_repeat_counts(ptr+1, &repeat_min, &repeat_max, errorcodeptr);
3741 if (*errorcodeptr != 0) goto FAILED;
3742 goto REPEAT;
3743
3744 case CHAR_ASTERISK:
3745 repeat_min = 0;
3746 repeat_max = -1;
3747 goto REPEAT;
3748
3749 case CHAR_PLUS:
3750 repeat_min = 1;
3751 repeat_max = -1;
3752 goto REPEAT;
3753
3754 case CHAR_QUESTION_MARK:
3755 repeat_min = 0;
3756 repeat_max = 1;
3757
3758 REPEAT:
3759 if (previous == NULL)
3760 {
3761 *errorcodeptr = ERR9;
3762 goto FAILED;
3763 }
3764
3765 if (repeat_min == 0)
3766 {
3767 firstbyte = zerofirstbyte; /* Adjust for zero repeat */
3768 reqbyte = zeroreqbyte; /* Ditto */
3769 }
3770
3771 /* Remember whether this is a variable length repeat */
3772
3773 reqvary = (repeat_min == repeat_max)? 0 : REQ_VARY;
3774
3775 op_type = 0; /* Default single-char op codes */
3776 possessive_quantifier = FALSE; /* Default not possessive quantifier */
3777
3778 /* Save start of previous item, in case we have to move it up to make space
3779 for an inserted OP_ONCE for the additional '+' extension. */
3780
3781 tempcode = previous;
3782
3783 /* If the next character is '+', we have a possessive quantifier. This
3784 implies greediness, whatever the setting of the PCRE_UNGREEDY option.
3785 If the next character is '?' this is a minimizing repeat, by default,
3786 but if PCRE_UNGREEDY is set, it works the other way round. We change the
3787 repeat type to the non-default. */
3788
3789 if (ptr[1] == CHAR_PLUS)
3790 {
3791 repeat_type = 0; /* Force greedy */
3792 possessive_quantifier = TRUE;
3793 ptr++;
3794 }
3795 else if (ptr[1] == CHAR_QUESTION_MARK)
3796 {
3797 repeat_type = greedy_non_default;
3798 ptr++;
3799 }
3800 else repeat_type = greedy_default;
3801
3802 /* If previous was a character match, abolish the item and generate a
3803 repeat item instead. If a char item has a minumum of more than one, ensure
3804 that it is set in reqbyte - it might not be if a sequence such as x{3} is
3805 the first thing in a branch because the x will have gone into firstbyte
3806 instead. */
3807
3808 if (*previous == OP_CHAR || *previous == OP_CHARNC)
3809 {
3810 /* Deal with UTF-8 characters that take up more than one byte. It's
3811 easier to write this out separately than try to macrify it. Use c to
3812 hold the length of the character in bytes, plus 0x80 to flag that it's a
3813 length rather than a small character. */
3814
3815 #ifdef SUPPORT_UTF8
3816 if (utf8 && (code[-1] & 0x80) != 0)
3817 {
3818 uschar *lastchar = code - 1;
3819 while((*lastchar & 0xc0) == 0x80) lastchar--;
3820 c = code - lastchar; /* Length of UTF-8 character */
3821 memcpy(utf8_char, lastchar, c); /* Save the char */
3822 c |= 0x80; /* Flag c as a length */
3823 }
3824 else
3825 #endif
3826
3827 /* Handle the case of a single byte - either with no UTF8 support, or
3828 with UTF-8 disabled, or for a UTF-8 character < 128. */
3829
3830 {
3831 c = code[-1];
3832 if (repeat_min > 1) reqbyte = c | req_caseopt | cd->req_varyopt;
3833 }
3834
3835 /* If the repetition is unlimited, it pays to see if the next thing on
3836 the line is something that cannot possibly match this character. If so,
3837 automatically possessifying this item gains some performance in the case
3838 where the match fails. */
3839
3840 if (!possessive_quantifier &&
3841 repeat_max < 0 &&
3842 check_auto_possessive(*previous, c, utf8, utf8_char, ptr + 1,
3843 options, cd))
3844 {
3845 repeat_type = 0; /* Force greedy */
3846 possessive_quantifier = TRUE;
3847 }
3848
3849 goto OUTPUT_SINGLE_REPEAT; /* Code shared with single character types */
3850 }
3851
3852 /* If previous was a single negated character ([^a] or similar), we use
3853 one of the special opcodes, replacing it. The code is shared with single-
3854 character repeats by setting opt_type to add a suitable offset into
3855 repeat_type. We can also test for auto-possessification. OP_NOT is
3856 currently used only for single-byte chars. */
3857
3858 else if (*previous == OP_NOT)
3859 {
3860 op_type = OP_NOTSTAR - OP_STAR; /* Use "not" opcodes */
3861 c = previous[1];
3862 if (!possessive_quantifier &&
3863 repeat_max < 0 &&
3864 check_auto_possessive(OP_NOT, c, utf8, NULL, ptr + 1, options, cd))
3865 {
3866 repeat_type = 0; /* Force greedy */
3867 possessive_quantifier = TRUE;
3868 }
3869 goto OUTPUT_SINGLE_REPEAT;
3870 }
3871
3872 /* If previous was a character type match (\d or similar), abolish it and
3873 create a suitable repeat item. The code is shared with single-character
3874 repeats by setting op_type to add a suitable offset into repeat_type. Note
3875 the the Unicode property types will be present only when SUPPORT_UCP is
3876 defined, but we don't wrap the little bits of code here because it just
3877 makes it horribly messy. */
3878
3879 else if (*previous < OP_EODN)
3880 {
3881 uschar *oldcode;
3882 int prop_type, prop_value;
3883 op_type = OP_TYPESTAR - OP_STAR; /* Use type opcodes */
3884 c = *previous;
3885
3886 if (!possessive_quantifier &&
3887 repeat_max < 0 &&
3888 check_auto_possessive(c, 0, utf8, NULL, ptr + 1, options, cd))
3889 {
3890 repeat_type = 0; /* Force greedy */
3891 possessive_quantifier = TRUE;
3892 }
3893
3894 OUTPUT_SINGLE_REPEAT:
3895 if (*previous == OP_PROP || *previous == OP_NOTPROP)
3896 {
3897 prop_type = previous[1];
3898 prop_value = previous[2];
3899 }
3900 else prop_type = prop_value = -1;
3901
3902 oldcode = code;
3903 code = previous; /* Usually overwrite previous item */
3904
3905 /* If the maximum is zero then the minimum must also be zero; Perl allows
3906 this case, so we do too - by simply omitting the item altogether. */
3907
3908 if (repeat_max == 0) goto END_REPEAT;
3909
3910 /*--------------------------------------------------------------------*/
3911 /* This code is obsolete from release 8.00; the restriction was finally
3912 removed: */
3913
3914 /* All real repeats make it impossible to handle partial matching (maybe
3915 one day we will be able to remove this restriction). */
3916
3917 /* if (repeat_max != 1) cd->external_flags |= PCRE_NOPARTIAL; */
3918 /*--------------------------------------------------------------------*/
3919
3920 /* Combine the op_type with the repeat_type */
3921
3922 repeat_type += op_type;
3923
3924 /* A minimum of zero is handled either as the special case * or ?, or as
3925 an UPTO, with the maximum given. */
3926
3927 if (repeat_min == 0)
3928 {
3929 if (repeat_max == -1) *code++ = OP_STAR + repeat_type;
3930 else if (repeat_max == 1) *code++ = OP_QUERY + repeat_type;
3931 else
3932 {
3933 *code++ = OP_UPTO + repeat_type;
3934 PUT2INC(code, 0, repeat_max);
3935 }
3936 }
3937
3938 /* A repeat minimum of 1 is optimized into some special cases. If the
3939 maximum is unlimited, we use OP_PLUS. Otherwise, the original item is
3940 left in place and, if the maximum is greater than 1, we use OP_UPTO with
3941 one less than the maximum. */
3942
3943 else if (repeat_min == 1)
3944 {
3945 if (repeat_max == -1)
3946 *code++ = OP_PLUS + repeat_type;
3947 else
3948 {
3949 code = oldcode; /* leave previous item in place */
3950 if (repeat_max == 1) goto END_REPEAT;
3951 *code++ = OP_UPTO + repeat_type;
3952 PUT2INC(code, 0, repeat_max - 1);
3953 }
3954 }
3955
3956 /* The case {n,n} is just an EXACT, while the general case {n,m} is
3957 handled as an EXACT followed by an UPTO. */
3958
3959 else
3960 {
3961 *code++ = OP_EXACT + op_type; /* NB EXACT doesn't have repeat_type */
3962 PUT2INC(code, 0, repeat_min);
3963
3964 /* If the maximum is unlimited, insert an OP_STAR. Before doing so,
3965 we have to insert the character for the previous code. For a repeated
3966 Unicode property match, there are two extra bytes that define the
3967 required property. In UTF-8 mode, long characters have their length in
3968 c, with the 0x80 bit as a flag. */
3969
3970 if (repeat_max < 0)
3971 {
3972 #ifdef SUPPORT_UTF8
3973 if (utf8 && c >= 128)
3974 {
3975 memcpy(code, utf8_char, c & 7);
3976 code += c & 7;
3977 }
3978 else
3979 #endif
3980 {
3981 *code++ = c;
3982 if (prop_type >= 0)
3983 {
3984 *code++ = prop_type;
3985 *code++ = prop_value;
3986 }
3987 }
3988 *code++ = OP_STAR + repeat_type;
3989 }
3990
3991 /* Else insert an UPTO if the max is greater than the min, again
3992 preceded by the character, for the previously inserted code. If the
3993 UPTO is just for 1 instance, we can use QUERY instead. */
3994
3995 else if (repeat_max != repeat_min)
3996 {
3997 #ifdef SUPPORT_UTF8
3998 if (utf8 && c >= 128)
3999 {
4000 memcpy(code, utf8_char, c & 7);
4001 code += c & 7;
4002 }
4003 else
4004 #endif
4005 *code++ = c;
4006 if (prop_type >= 0)
4007 {
4008 *code++ = prop_type;
4009 *code++ = prop_value;
4010 }
4011 repeat_max -= repeat_min;
4012
4013 if (repeat_max == 1)
4014 {
4015 *code++ = OP_QUERY + repeat_type;
4016 }
4017 else
4018 {
4019 *code++ = OP_UPTO + repeat_type;
4020 PUT2INC(code, 0, repeat_max);
4021 }
4022 }
4023 }
4024
4025 /* The character or character type itself comes last in all cases. */
4026
4027 #ifdef SUPPORT_UTF8
4028 if (utf8 && c >= 128)
4029 {
4030 memcpy(code, utf8_char, c & 7);
4031 code += c & 7;
4032 }
4033 else
4034 #endif
4035 *code++ = c;
4036
4037 /* For a repeated Unicode property match, there are two extra bytes that
4038 define the required property. */
4039
4040 #ifdef SUPPORT_UCP
4041 if (prop_type >= 0)
4042 {
4043 *code++ = prop_type;
4044 *code++ = prop_value;
4045 }
4046 #endif
4047 }
4048
4049 /* If previous was a character class or a back reference, we put the repeat
4050 stuff after it, but just skip the item if the repeat was {0,0}. */
4051
4052 else if (*previous == OP_CLASS ||
4053 *previous == OP_NCLASS ||
4054 #ifdef SUPPORT_UTF8
4055 *previous == OP_XCLASS ||
4056 #endif
4057 *previous == OP_REF)
4058 {
4059 if (repeat_max == 0)
4060 {
4061 code = previous;
4062 goto END_REPEAT;
4063 }
4064
4065 /*--------------------------------------------------------------------*/
4066 /* This code is obsolete from release 8.00; the restriction was finally
4067 removed: */
4068
4069 /* All real repeats make it impossible to handle partial matching (maybe
4070 one day we will be able to remove this restriction). */
4071
4072 /* if (repeat_max != 1) cd->external_flags |= PCRE_NOPARTIAL; */
4073 /*--------------------------------------------------------------------*/
4074
4075 if (repeat_min == 0 && repeat_max == -1)
4076 *code++ = OP_CRSTAR + repeat_type;
4077 else if (repeat_min == 1 && repeat_max == -1)
4078 *code++ = OP_CRPLUS + repeat_type;
4079 else if (repeat_min == 0 && repeat_max == 1)
4080 *code++ = OP_CRQUERY + repeat_type;
4081 else
4082 {
4083 *code++ = OP_CRRANGE + repeat_type;
4084 PUT2INC(code, 0, repeat_min);
4085 if (repeat_max == -1) repeat_max = 0; /* 2-byte encoding for max */
4086 PUT2INC(code, 0, repeat_max);
4087 }
4088 }
4089
4090 /* If previous was a bracket group, we may have to replicate it in certain
4091 cases. */
4092
4093 else if (*previous == OP_BRA || *previous == OP_CBRA ||
4094 *previous == OP_ONCE || *previous == OP_COND)
4095 {
4096 register int i;
4097 int ketoffset = 0;
4098 int len = code - previous;
4099 uschar *bralink = NULL;
4100
4101 /* Repeating a DEFINE group is pointless */
4102
4103 if (*previous == OP_COND && previous[LINK_SIZE+1] == OP_DEF)
4104 {
4105 *errorcodeptr = ERR55;
4106 goto FAILED;
4107 }
4108
4109 /* If the maximum repeat count is unlimited, find the end of the bracket
4110 by scanning through from the start, and compute the offset back to it
4111 from the current code pointer. There may be an OP_OPT setting following
4112 the final KET, so we can't find the end just by going back from the code
4113 pointer. */
4114
4115 if (repeat_max == -1)
4116 {
4117 register uschar *ket = previous;
4118 do ket += GET(ket, 1); while (*ket != OP_KET);
4119 ketoffset = code - ket;
4120 }
4121
4122 /* The case of a zero minimum is special because of the need to stick
4123 OP_BRAZERO in front of it, and because the group appears once in the
4124 data, whereas in other cases it appears the minimum number of times. For
4125 this reason, it is simplest to treat this case separately, as otherwise
4126 the code gets far too messy. There are several special subcases when the
4127 minimum is zero. */
4128
4129 if (repeat_min == 0)
4130 {
4131 /* If the maximum is also zero, we used to just omit the group from the
4132 output altogether, like this:
4133
4134 ** if (repeat_max == 0)
4135 ** {
4136 ** code = previous;
4137 ** goto END_REPEAT;
4138 ** }
4139
4140 However, that fails when a group is referenced as a subroutine from
4141 elsewhere in the pattern, so now we stick in OP_SKIPZERO in front of it
4142 so that it is skipped on execution. As we don't have a list of which
4143 groups are referenced, we cannot do this selectively.
4144
4145 If the maximum is 1 or unlimited, we just have to stick in the BRAZERO
4146 and do no more at this point. However, we do need to adjust any
4147 OP_RECURSE calls inside the group that refer to the group itself or any
4148 internal or forward referenced group, because the offset is from the
4149 start of the whole regex. Temporarily terminate the pattern while doing
4150 this. */
4151
4152 if (repeat_max <= 1) /* Covers 0, 1, and unlimited */
4153 {
4154 *code = OP_END;
4155 adjust_recurse(previous, 1, utf8, cd, save_hwm);
4156 memmove(previous+1, previous, len);
4157 code++;
4158 if (repeat_max == 0)
4159 {
4160 *previous++ = OP_SKIPZERO;
4161 goto END_REPEAT;
4162 }
4163 *previous++ = OP_BRAZERO + repeat_type;
4164 }
4165
4166 /* If the maximum is greater than 1 and limited, we have to replicate
4167 in a nested fashion, sticking OP_BRAZERO before each set of brackets.
4168 The first one has to be handled carefully because it's the original
4169 copy, which has to be moved up. The remainder can be handled by code
4170 that is common with the non-zero minimum case below. We have to
4171 adjust the value or repeat_max, since one less copy is required. Once
4172 again, we may have to adjust any OP_RECURSE calls inside the group. */
4173
4174 else
4175 {
4176 int offset;
4177 *code = OP_END;
4178 adjust_recurse(previous, 2 + LINK_SIZE, utf8, cd, save_hwm);
4179 memmove(previous + 2 + LINK_SIZE, previous, len);
4180 code += 2 + LINK_SIZE;
4181 *previous++ = OP_BRAZERO + repeat_type;
4182 *previous++ = OP_BRA;
4183
4184 /* We chain together the bracket offset fields that have to be
4185 filled in later when the ends of the brackets are reached. */
4186
4187 offset = (bralink == NULL)? 0 : previous - bralink;
4188 bralink = previous;
4189 PUTINC(previous, 0, offset);
4190 }
4191
4192 repeat_max--;
4193 }
4194
4195 /* If the minimum is greater than zero, replicate the group as many
4196 times as necessary, and adjust the maximum to the number of subsequent
4197 copies that we need. If we set a first char from the group, and didn't
4198 set a required char, copy the latter from the former. If there are any
4199 forward reference subroutine calls in the group, there will be entries on
4200 the workspace list; replicate these with an appropriate increment. */
4201
4202 else
4203 {
4204 if (repeat_min > 1)
4205 {
4206 /* In the pre-compile phase, we don't actually do the replication. We
4207 just adjust the length as if we had. Do some paranoid checks for
4208 potential integer overflow. */
4209
4210 if (lengthptr != NULL)
4211 {
4212 int delta = (repeat_min - 1)*length_prevgroup;
4213 if ((double)(repeat_min - 1)*(double)length_prevgroup >
4214 (double)INT_MAX ||
4215 OFLOW_MAX - *lengthptr < delta)
4216 {
4217 *errorcodeptr = ERR20;
4218 goto FAILED;
4219 }
4220 *lengthptr += delta;
4221 }
4222
4223 /* This is compiling for real */
4224
4225 else
4226 {
4227 if (groupsetfirstbyte && reqbyte < 0) reqbyte = firstbyte;
4228 for (i = 1; i < repeat_min; i++)
4229 {
4230 uschar *hc;
4231 uschar *this_hwm = cd->hwm;
4232 memcpy(code, previous, len);
4233 for (hc = save_hwm; hc < this_hwm; hc += LINK_SIZE)
4234 {
4235 PUT(cd->hwm, 0, GET(hc, 0) + len);
4236 cd->hwm += LINK_SIZE;
4237 }
4238 save_hwm = this_hwm;
4239 code += len;
4240 }
4241 }
4242 }
4243
4244 if (repeat_max > 0) repeat_max -= repeat_min;
4245 }
4246
4247 /* This code is common to both the zero and non-zero minimum cases. If
4248 the maximum is limited, it replicates the group in a nested fashion,
4249 remembering the bracket starts on a stack. In the case of a zero minimum,
4250 the first one was set up above. In all cases the repeat_max now specifies
4251 the number of additional copies needed. Again, we must remember to
4252 replicate entries on the forward reference list. */
4253
4254 if (repeat_max >= 0)
4255 {
4256 /* In the pre-compile phase, we don't actually do the replication. We
4257 just adjust the length as if we had. For each repetition we must add 1
4258 to the length for BRAZERO and for all but the last repetition we must
4259 add 2 + 2*LINKSIZE to allow for the nesting that occurs. Do some
4260 paranoid checks to avoid integer overflow. */
4261
4262 if (lengthptr != NULL && repeat_max > 0)
4263 {
4264 int delta = repeat_max * (length_prevgroup + 1 + 2 + 2*LINK_SIZE) -
4265 2 - 2*LINK_SIZE; /* Last one doesn't nest */
4266 if ((double)repeat_max *
4267 (double)(length_prevgroup + 1 + 2 + 2*LINK_SIZE)
4268 > (double)INT_MAX ||
4269 OFLOW_MAX - *lengthptr < delta)
4270 {
4271 *errorcodeptr = ERR20;
4272 goto FAILED;
4273 }
4274 *lengthptr += delta;
4275 }
4276
4277 /* This is compiling for real */
4278
4279 else for (i = repeat_max - 1; i >= 0; i--)
4280 {
4281 uschar *hc;
4282 uschar *this_hwm = cd->hwm;
4283
4284 *code++ = OP_BRAZERO + repeat_type;
4285
4286 /* All but the final copy start a new nesting, maintaining the
4287 chain of brackets outstanding. */
4288
4289 if (i != 0)
4290 {
4291 int offset;
4292 *code++ = OP_BRA;
4293 offset = (bralink == NULL)? 0 : code - bralink;
4294 bralink = code;
4295 PUTINC(code, 0, offset);
4296 }
4297
4298 memcpy(code, previous, len);
4299 for (hc = save_hwm; hc < this_hwm; hc += LINK_SIZE)
4300 {
4301 PUT(cd->hwm, 0, GET(hc, 0) + len + ((i != 0)? 2+LINK_SIZE : 1));
4302 cd->hwm += LINK_SIZE;
4303 }
4304 save_hwm = this_hwm;
4305 code += len;
4306 }
4307
4308 /* Now chain through the pending brackets, and fill in their length
4309 fields (which are holding the chain links pro tem). */
4310
4311 while (bralink != NULL)
4312 {
4313 int oldlinkoffset;
4314 int offset = code - bralink + 1;
4315 uschar *bra = code - offset;
4316 oldlinkoffset = GET(bra, 1);
4317 bralink = (oldlinkoffset == 0)? NULL : bralink - oldlinkoffset;
4318 *code++ = OP_KET;
4319 PUTINC(code, 0, offset);
4320 PUT(bra, 1, offset);
4321 }
4322 }
4323
4324 /* If the maximum is unlimited, set a repeater in the final copy. We
4325 can't just offset backwards from the current code point, because we
4326 don't know if there's been an options resetting after the ket. The
4327 correct offset was computed above.
4328
4329 Then, when we are doing the actual compile phase, check to see whether
4330 this group is a non-atomic one that could match an empty string. If so,
4331 convert the initial operator to the S form (e.g. OP_BRA -> OP_SBRA) so
4332 that runtime checking can be done. [This check is also applied to
4333 atomic groups at runtime, but in a different way.] */
4334
4335 else
4336 {
4337 uschar *ketcode = code - ketoffset;
4338 uschar *bracode = ketcode - GET(ketcode, 1);
4339 *ketcode = OP_KETRMAX + repeat_type;
4340 if (lengthptr == NULL && *bracode != OP_ONCE)
4341 {
4342 uschar *scode = bracode;
4343 do
4344 {
4345 if (could_be_empty_branch(scode, ketcode, utf8))
4346 {
4347 *bracode += OP_SBRA - OP_BRA;
4348 break;
4349 }
4350 scode += GET(scode, 1);
4351 }
4352 while (*scode == OP_ALT);
4353 }
4354 }
4355 }
4356
4357 /* If previous is OP_FAIL, it was generated by an empty class [] in
4358 JavaScript mode. The other ways in which OP_FAIL can be generated, that is
4359 by (*FAIL) or (?!) set previous to NULL, which gives a "nothing to repeat"
4360 error above. We can just ignore the repeat in JS case. */
4361
4362 else if (*previous == OP_FAIL) goto END_REPEAT;
4363
4364 /* Else there's some kind of shambles */
4365
4366 else
4367 {
4368 *errorcodeptr = ERR11;
4369 goto FAILED;
4370 }
4371
4372 /* If the character following a repeat is '+', or if certain optimization
4373 tests above succeeded, possessive_quantifier is TRUE. For some of the
4374 simpler opcodes, there is an special alternative opcode for this. For
4375 anything else, we wrap the entire repeated item inside OP_ONCE brackets.
4376 The '+' notation is just syntactic sugar, taken from Sun's Java package,
4377 but the special opcodes can optimize it a bit. The repeated item starts at
4378 tempcode, not at previous, which might be the first part of a string whose
4379 (former) last char we repeated.
4380
4381 Possessifying an 'exact' quantifier has no effect, so we can ignore it. But
4382 an 'upto' may follow. We skip over an 'exact' item, and then test the
4383 length of what remains before proceeding. */
4384
4385 if (possessive_quantifier)
4386 {
4387 int len;
4388
4389 if (*tempcode == OP_TYPEEXACT)
4390 tempcode += _pcre_OP_lengths[*tempcode] +
4391 ((tempcode[3] == OP_PROP || tempcode[3] == OP_NOTPROP)? 2 : 0);
4392
4393 else if (*tempcode == OP_EXACT || *tempcode == OP_NOTEXACT)
4394 {
4395 tempcode += _pcre_OP_lengths[*tempcode];
4396 #ifdef SUPPORT_UTF8
4397 if (utf8 && tempcode[-1] >= 0xc0)
4398 tempcode += _pcre_utf8_table4[tempcode[-1] & 0x3f];
4399 #endif
4400 }
4401
4402 len = code - tempcode;
4403 if (len > 0) switch (*tempcode)
4404 {
4405 case OP_STAR: *tempcode = OP_POSSTAR; break;
4406 case OP_PLUS: *tempcode = OP_POSPLUS; break;
4407 case OP_QUERY: *tempcode = OP_POSQUERY; break;
4408 case OP_UPTO: *tempcode = OP_POSUPTO; break;
4409
4410 case OP_TYPESTAR: *tempcode = OP_TYPEPOSSTAR; break;
4411 case OP_TYPEPLUS: *tempcode = OP_TYPEPOSPLUS; break;
4412 case OP_TYPEQUERY: *tempcode = OP_TYPEPOSQUERY; break;
4413 case OP_TYPEUPTO: *tempcode = OP_TYPEPOSUPTO; break;
4414
4415 case OP_NOTSTAR: *tempcode = OP_NOTPOSSTAR; break;
4416 case OP_NOTPLUS: *tempcode = OP_NOTPOSPLUS; break;
4417 case OP_NOTQUERY: *tempcode = OP_NOTPOSQUERY; break;
4418 case OP_NOTUPTO: *tempcode = OP_NOTPOSUPTO; break;
4419
4420 default:
4421 memmove(tempcode + 1+LINK_SIZE, tempcode, len);
4422 code += 1 + LINK_SIZE;
4423 len += 1 + LINK_SIZE;
4424 tempcode[0] = OP_ONCE;
4425 *code++ = OP_KET;
4426 PUTINC(code, 0, len);
4427 PUT(tempcode, 1, len);
4428 break;
4429 }
4430 }
4431
4432 /* In all case we no longer have a previous item. We also set the
4433 "follows varying string" flag for subsequently encountered reqbytes if
4434 it isn't already set and we have just passed a varying length item. */
4435
4436 END_REPEAT:
4437 previous = NULL;
4438 cd->req_varyopt |= reqvary;
4439 break;
4440
4441
4442 /* ===================================================================*/
4443 /* Start of nested parenthesized sub-expression, or comment or lookahead or
4444 lookbehind or option setting or condition or all the other extended
4445 parenthesis forms. */
4446
4447 case CHAR_LEFT_PARENTHESIS:
4448 newoptions = options;
4449 skipbytes = 0;
4450 bravalue = OP_CBRA;
4451 save_hwm = cd->hwm;
4452 reset_bracount = FALSE;
4453
4454 /* First deal with various "verbs" that can be introduced by '*'. */
4455
4456 if (*(++ptr) == CHAR_ASTERISK && (cd->ctypes[ptr[1]] & ctype_letter) != 0)
4457 {
4458 int i, namelen;
4459 const char *vn = verbnames;
4460 const uschar *name = ++ptr;
4461 previous = NULL;
4462 while ((cd->ctypes[*++ptr] & ctype_letter) != 0) {};
4463 if (*ptr == CHAR_COLON)
4464 {
4465 *errorcodeptr = ERR59; /* Not supported */
4466 goto FAILED;
4467 }
4468 if (*ptr != CHAR_RIGHT_PARENTHESIS)
4469 {
4470 *errorcodeptr = ERR60;
4471 goto FAILED;
4472 }
4473 namelen = ptr - name;
4474 for (i = 0; i < verbcount; i++)
4475 {
4476 if (namelen == verbs[i].len &&
4477 strncmp((char *)name, vn, namelen) == 0)
4478 {
4479 /* Check for open captures before ACCEPT */
4480
4481 if (verbs[i].op == OP_ACCEPT)
4482 {
4483 open_capitem *oc;
4484 cd->had_accept = TRUE;
4485 for (oc = cd->open_caps; oc != NULL; oc = oc->next)
4486 {
4487 *code++ = OP_CLOSE;
4488 PUT2INC(code, 0, oc->number);
4489 }
4490 }
4491 *code++ = verbs[i].op;
4492 break;
4493 }
4494 vn += verbs[i].len + 1;
4495 }
4496 if (i < verbcount) continue;
4497 *errorcodeptr = ERR60;
4498 goto FAILED;
4499 }
4500
4501 /* Deal with the extended parentheses; all are introduced by '?', and the
4502 appearance of any of them means that this is not a capturing group. */
4503
4504 else if (*ptr == CHAR_QUESTION_MARK)
4505 {
4506 int i, set, unset, namelen;
4507 int *optset;
4508 const uschar *name;
4509 uschar *slot;
4510
4511 switch (*(++ptr))
4512 {
4513 case CHAR_NUMBER_SIGN: /* Comment; skip to ket */
4514 ptr++;
4515 while (*ptr != 0 && *ptr != CHAR_RIGHT_PARENTHESIS) ptr++;
4516 if (*ptr == 0)
4517 {
4518 *errorcodeptr = ERR18;
4519 goto FAILED;
4520 }
4521 continue;
4522
4523
4524 /* ------------------------------------------------------------ */
4525 case CHAR_VERTICAL_LINE: /* Reset capture count for each branch */
4526 reset_bracount = TRUE;
4527 /* Fall through */
4528
4529 /* ------------------------------------------------------------ */
4530 case CHAR_COLON: /* Non-capturing bracket */
4531 bravalue = OP_BRA;
4532 ptr++;
4533 break;
4534
4535
4536 /* ------------------------------------------------------------ */
4537 case CHAR_LEFT_PARENTHESIS:
4538 bravalue = OP_COND; /* Conditional group */
4539
4540 /* A condition can be an assertion, a number (referring to a numbered
4541 group), a name (referring to a named group), or 'R', referring to
4542 recursion. R<digits> and R&name are also permitted for recursion tests.
4543
4544 There are several syntaxes for testing a named group: (?(name)) is used
4545 by Python; Perl 5.10 onwards uses (?(<name>) or (?('name')).
4546
4547 There are two unfortunate ambiguities, caused by history. (a) 'R' can
4548 be the recursive thing or the name 'R' (and similarly for 'R' followed
4549 by digits), and (b) a number could be a name that consists of digits.
4550 In both cases, we look for a name first; if not found, we try the other
4551 cases. */
4552
4553 /* For conditions that are assertions, check the syntax, and then exit
4554 the switch. This will take control down to where bracketed groups,
4555 including assertions, are processed. */
4556
4557 if (ptr[1] == CHAR_QUESTION_MARK && (ptr[2] == CHAR_EQUALS_SIGN ||
4558 ptr[2] == CHAR_EXCLAMATION_MARK || ptr[2] == CHAR_LESS_THAN_SIGN))
4559 break;
4560
4561 /* Most other conditions use OP_CREF (a couple change to OP_RREF
4562 below), and all need to skip 3 bytes at the start of the group. */
4563
4564 code[1+LINK_SIZE] = OP_CREF;
4565 skipbytes = 3;
4566 refsign = -1;
4567
4568 /* Check for a test for recursion in a named group. */
4569
4570 if (ptr[1] == CHAR_R && ptr[2] == CHAR_AMPERSAND)
4571 {
4572 terminator = -1;
4573 ptr += 2;
4574 code[1+LINK_SIZE] = OP_RREF; /* Change the type of test */
4575 }
4576
4577 /* Check for a test for a named group's having been set, using the Perl
4578 syntax (?(<name>) or (?('name') */
4579
4580 else if (ptr[1] == CHAR_LESS_THAN_SIGN)
4581 {
4582 terminator = CHAR_GREATER_THAN_SIGN;
4583 ptr++;
4584 }
4585 else if (ptr[1] == CHAR_APOSTROPHE)
4586 {
4587 terminator = CHAR_APOSTROPHE;
4588 ptr++;
4589 }
4590 else
4591 {
4592 terminator = 0;
4593 if (ptr[1] == CHAR_MINUS || ptr[1] == CHAR_PLUS) refsign = *(++ptr);
4594 }
4595
4596 /* We now expect to read a name; any thing else is an error */
4597
4598 if ((cd->ctypes[ptr[1]] & ctype_word) == 0)
4599 {
4600 ptr += 1; /* To get the right offset */
4601 *errorcodeptr = ERR28;
4602 goto FAILED;
4603 }
4604
4605 /* Read the name, but also get it as a number if it's all digits */
4606
4607 recno = 0;
4608 name = ++ptr;
4609 while ((cd->ctypes[*ptr] & ctype_word) != 0)
4610 {
4611 if (recno >= 0)
4612 recno = ((digitab[*ptr] & ctype_digit) != 0)?
4613 recno * 10 + *ptr - CHAR_0 : -1;
4614 ptr++;
4615 }
4616 namelen = ptr - name;
4617
4618 if ((terminator > 0 && *ptr++ != terminator) ||
4619 *ptr++ != CHAR_RIGHT_PARENTHESIS)
4620 {
4621 ptr--; /* Error offset */
4622 *errorcodeptr = ERR26;
4623 goto FAILED;
4624 }
4625
4626 /* Do no further checking in the pre-compile phase. */
4627
4628 if (lengthptr != NULL) break;
4629
4630 /* In the real compile we do the work of looking for the actual
4631 reference. If the string started with "+" or "-" we require the rest to
4632 be digits, in which case recno will be set. */
4633
4634 if (refsign > 0)
4635 {
4636 if (recno <= 0)
4637 {
4638 *errorcodeptr = ERR58;
4639 goto FAILED;
4640 }
4641 recno = (refsign == CHAR_MINUS)?
4642 cd->bracount - recno + 1 : recno +cd->bracount;
4643 if (recno <= 0 || recno > cd->final_bracount)
4644 {
4645 *errorcodeptr = ERR15;
4646 goto FAILED;
4647 }
4648 PUT2(code, 2+LINK_SIZE, recno);
4649 break;
4650 }
4651
4652 /* Otherwise (did not start with "+" or "-"), start by looking for the
4653 name. */
4654
4655 slot = cd->name_table;
4656 for (i = 0; i < cd->names_found; i++)
4657 {
4658 if (strncmp((char *)name, (char *)slot+2, namelen) == 0) break;
4659 slot += cd->name_entry_size;
4660 }
4661
4662 /* Found a previous named subpattern */
4663
4664 if (i < cd->names_found)
4665 {
4666 recno = GET2(slot, 0);
4667 PUT2(code, 2+LINK_SIZE, recno);
4668 }
4669
4670 /* Search the pattern for a forward reference */
4671
4672 else if ((i = find_parens(cd, name, namelen,
4673 (options & PCRE_EXTENDED) != 0)) > 0)
4674 {
4675 PUT2(code, 2+LINK_SIZE, i);
4676 }
4677
4678 /* If terminator == 0 it means that the name followed directly after
4679 the opening parenthesis [e.g. (?(abc)...] and in this case there are
4680 some further alternatives to try. For the cases where terminator != 0
4681 [things like (?(<name>... or (?('name')... or (?(R&name)... ] we have
4682 now checked all the possibilities, so give an error. */
4683
4684 else if (terminator != 0)
4685 {
4686 *errorcodeptr = ERR15;
4687 goto FAILED;
4688 }
4689
4690 /* Check for (?(R) for recursion. Allow digits after R to specify a
4691 specific group number. */
4692
4693 else if (*name == CHAR_R)
4694 {
4695 recno = 0;
4696 for (i = 1; i < namelen; i++)
4697 {
4698 if ((digitab[name[i]] & ctype_digit) == 0)
4699 {
4700 *errorcodeptr = ERR15;
4701 goto FAILED;
4702 }
4703 recno = recno * 10 + name[i] - CHAR_0;
4704 }
4705 if (recno == 0) recno = RREF_ANY;
4706 code[1+LINK_SIZE] = OP_RREF; /* Change test type */
4707 PUT2(code, 2+LINK_SIZE, recno);
4708 }
4709
4710 /* Similarly, check for the (?(DEFINE) "condition", which is always
4711 false. */
4712
4713 else if (namelen == 6 && strncmp((char *)name, STRING_DEFINE, 6) == 0)
4714 {
4715 code[1+LINK_SIZE] = OP_DEF;
4716 skipbytes = 1;
4717 }
4718
4719 /* Check for the "name" actually being a subpattern number. We are
4720 in the second pass here, so final_bracount is set. */
4721
4722 else if (recno > 0 && recno <= cd->final_bracount)
4723 {
4724 PUT2(code, 2+LINK_SIZE, recno);
4725 }
4726
4727 /* Either an unidentified subpattern, or a reference to (?(0) */
4728
4729 else
4730 {
4731 *errorcodeptr = (recno == 0)? ERR35: ERR15;
4732 goto FAILED;
4733 }
4734 break;
4735
4736
4737 /* ------------------------------------------------------------ */
4738 case CHAR_EQUALS_SIGN: /* Positive lookahead */
4739 bravalue = OP_ASSERT;
4740 ptr++;
4741 break;
4742
4743
4744 /* ------------------------------------------------------------ */
4745 case CHAR_EXCLAMATION_MARK: /* Negative lookahead */
4746 ptr++;
4747 if (*ptr == CHAR_RIGHT_PARENTHESIS) /* Optimize (?!) */
4748 {
4749 *code++ = OP_FAIL;
4750 previous = NULL;
4751 continue;
4752 }
4753 bravalue = OP_ASSERT_NOT;
4754 break;
4755
4756
4757 /* ------------------------------------------------------------ */
4758 case CHAR_LESS_THAN_SIGN: /* Lookbehind or named define */
4759 switch (ptr[1])
4760 {
4761 case CHAR_EQUALS_SIGN: /* Positive lookbehind */
4762 bravalue = OP_ASSERTBACK;
4763 ptr += 2;
4764 break;
4765
4766 case CHAR_EXCLAMATION_MARK: /* Negative lookbehind */
4767 bravalue = OP_ASSERTBACK_NOT;
4768 ptr += 2;
4769 break;
4770
4771 default: /* Could be name define, else bad */
4772 if ((cd->ctypes[ptr[1]] & ctype_word) != 0) goto DEFINE_NAME;
4773 ptr++; /* Correct offset for error */
4774 *errorcodeptr = ERR24;
4775 goto FAILED;
4776 }
4777 break;
4778
4779
4780 /* ------------------------------------------------------------ */
4781 case CHAR_GREATER_THAN_SIGN: /* One-time brackets */
4782 bravalue = OP_ONCE;
4783 ptr++;
4784 break;
4785
4786
4787 /* ------------------------------------------------------------ */
4788 case CHAR_C: /* Callout - may be followed by digits; */
4789 previous_callout = code; /* Save for later completion */
4790 after_manual_callout = 1; /* Skip one item before completing */
4791 *code++ = OP_CALLOUT;
4792 {
4793 int n = 0;
4794 while ((digitab[*(++ptr)] & ctype_digit) != 0)
4795 n = n * 10 + *ptr - CHAR_0;
4796 if (*ptr != CHAR_RIGHT_PARENTHESIS)
4797 {
4798 *errorcodeptr = ERR39;
4799 goto FAILED;
4800 }
4801 if (n > 255)
4802 {
4803 *errorcodeptr = ERR38;
4804 goto FAILED;
4805 }
4806 *code++ = n;
4807 PUT(code, 0, ptr - cd->start_pattern + 1); /* Pattern offset */
4808 PUT(code, LINK_SIZE, 0); /* Default length */
4809 code += 2 * LINK_SIZE;
4810 }
4811 previous = NULL;
4812 continue;
4813
4814
4815 /* ------------------------------------------------------------ */
4816 case CHAR_P: /* Python-style named subpattern handling */
4817 if (*(++ptr) == CHAR_EQUALS_SIGN ||
4818 *ptr == CHAR_GREATER_THAN_SIGN) /* Reference or recursion */
4819 {
4820 is_recurse = *ptr == CHAR_GREATER_THAN_SIGN;
4821 terminator = CHAR_RIGHT_PARENTHESIS;
4822 goto NAMED_REF_OR_RECURSE;
4823 }
4824 else if (*ptr != CHAR_LESS_THAN_SIGN) /* Test for Python-style defn */
4825 {
4826 *errorcodeptr = ERR41;
4827 goto FAILED;
4828 }
4829 /* Fall through to handle (?P< as (?< is handled */
4830
4831
4832 /* ------------------------------------------------------------ */
4833 DEFINE_NAME: /* Come here from (?< handling */
4834 case CHAR_APOSTROPHE:
4835 {
4836 terminator = (*ptr == CHAR_LESS_THAN_SIGN)?
4837 CHAR_GREATER_THAN_SIGN : CHAR_APOSTROPHE;
4838 name = ++ptr;
4839
4840 while ((cd->ctypes[*ptr] & ctype_word) != 0) ptr++;
4841 namelen = ptr - name;
4842
4843 /* In the pre-compile phase, just do a syntax check. */
4844
4845 if (lengthptr != NULL)
4846 {
4847 if (*ptr != terminator)
4848 {
4849 *errorcodeptr = ERR42;
4850 goto FAILED;
4851 }
4852 if (cd->names_found >= MAX_NAME_COUNT)
4853 {
4854 *errorcodeptr = ERR49;
4855 goto FAILED;
4856 }
4857 if (namelen + 3 > cd->name_entry_size)
4858 {
4859 cd->name_entry_size = namelen + 3;
4860 if (namelen > MAX_NAME_SIZE)
4861 {
4862 *errorcodeptr = ERR48;
4863 goto FAILED;
4864 }
4865 }
4866 }
4867
4868 /* In the real compile, create the entry in the table */
4869
4870 else
4871 {
4872 slot = cd->name_table;
4873 for (i = 0; i < cd->names_found; i++)
4874 {
4875 int crc = memcmp(name, slot+2, namelen);
4876 if (crc == 0)
4877 {
4878 if (slot[2+namelen] == 0)
4879 {
4880 if ((options & PCRE_DUPNAMES) == 0)
4881 {
4882 *errorcodeptr = ERR43;
4883 goto FAILED;
4884 }
4885 }
4886 else crc = -1; /* Current name is substring */
4887 }
4888 if (crc < 0)
4889 {
4890 memmove(slot + cd->name_entry_size, slot,
4891 (cd->names_found - i) * cd->name_entry_size);
4892 break;
4893 }
4894 slot += cd->name_entry_size;
4895 }
4896
4897 PUT2(slot, 0, cd->bracount + 1);
4898 memcpy(slot + 2, name, namelen);
4899 slot[2+namelen] = 0;
4900 }
4901 }
4902
4903 /* In both cases, count the number of names we've encountered. */
4904
4905 ptr++; /* Move past > or ' */
4906 cd->names_found++;
4907 goto NUMBERED_GROUP;
4908
4909
4910 /* ------------------------------------------------------------ */
4911 case CHAR_AMPERSAND: /* Perl recursion/subroutine syntax */
4912 terminator = CHAR_RIGHT_PARENTHESIS;
4913 is_recurse = TRUE;
4914 /* Fall through */
4915
4916 /* We come here from the Python syntax above that handles both
4917 references (?P=name) and recursion (?P>name), as well as falling
4918 through from the Perl recursion syntax (?&name). We also come here from
4919 the Perl \k<name> or \k'name' back reference syntax and the \k{name}
4920 .NET syntax, and the Oniguruma \g<...> and \g'...' subroutine syntax. */
4921
4922 NAMED_REF_OR_RECURSE:
4923 name = ++ptr;
4924 while ((cd->ctypes[*ptr] & ctype_word) != 0) ptr++;
4925 namelen = ptr - name;
4926
4927 /* In the pre-compile phase, do a syntax check and set a dummy
4928 reference number. */
4929
4930 if (lengthptr != NULL)
4931 {
4932 if (namelen == 0)
4933 {
4934 *errorcodeptr = ERR62;
4935 goto FAILED;
4936 }
4937 if (*ptr != terminator)
4938 {
4939 *errorcodeptr = ERR42;
4940 goto FAILED;
4941 }
4942 if (namelen > MAX_NAME_SIZE)
4943 {
4944 *errorcodeptr = ERR48;
4945 goto FAILED;
4946 }
4947 recno = 0;
4948 }
4949
4950 /* In the real compile, seek the name in the table. We check the name
4951 first, and then check that we have reached the end of the name in the
4952 table. That way, if the name that is longer than any in the table,
4953 the comparison will fail without reading beyond the table entry. */
4954
4955 else
4956 {
4957 slot = cd->name_table;
4958 for (i = 0; i < cd->names_found; i++)
4959 {
4960 if (strncmp((char *)name, (char *)slot+2, namelen) == 0 &&
4961 slot[2+namelen] == 0)
4962 break;
4963 slot += cd->name_entry_size;
4964 }
4965
4966 if (i < cd->names_found) /* Back reference */
4967 {
4968 recno = GET2(slot, 0);
4969 }
4970 else if ((recno = /* Forward back reference */
4971 find_parens(cd, name, namelen,
4972 (options & PCRE_EXTENDED) != 0)) <= 0)
4973 {
4974 *errorcodeptr = ERR15;
4975 goto FAILED;
4976 }
4977 }
4978
4979 /* In both phases, we can now go to the code than handles numerical
4980 recursion or backreferences. */
4981
4982 if (is_recurse) goto HANDLE_RECURSION;
4983 else goto HANDLE_REFERENCE;
4984
4985
4986 /* ------------------------------------------------------------ */
4987 case CHAR_R: /* Recursion */
4988 ptr++; /* Same as (?0) */
4989 /* Fall through */
4990
4991
4992 /* ------------------------------------------------------------ */
4993 case CHAR_MINUS: case CHAR_PLUS: /* Recursion or subroutine */
4994 case CHAR_0: case CHAR_1: case CHAR_2: case CHAR_3: case CHAR_4:
4995 case CHAR_5: case CHAR_6: case CHAR_7: case CHAR_8: case CHAR_9:
4996 {
4997 const uschar *called;
4998 terminator = CHAR_RIGHT_PARENTHESIS;
4999
5000 /* Come here from the \g<...> and \g'...' code (Oniguruma
5001 compatibility). However, the syntax has been checked to ensure that
5002 the ... are a (signed) number, so that neither ERR63 nor ERR29 will
5003 be called on this path, nor with the jump to OTHER_CHAR_AFTER_QUERY
5004 ever be taken. */
5005
5006 HANDLE_NUMERICAL_RECURSION:
5007
5008 if ((refsign = *ptr) == CHAR_PLUS)
5009 {
5010 ptr++;
5011 if ((digitab[*ptr] & ctype_digit) == 0)
5012 {
5013 *errorcodeptr = ERR63;
5014 goto FAILED;
5015 }
5016 }
5017 else if (refsign == CHAR_MINUS)
5018 {
5019 if ((digitab[ptr[1]] & ctype_digit) == 0)
5020 goto OTHER_CHAR_AFTER_QUERY;
5021 ptr++;
5022 }
5023
5024 recno = 0;
5025 while((digitab[*ptr] & ctype_digit) != 0)
5026 recno = recno * 10 + *ptr++ - CHAR_0;
5027
5028 if (*ptr != terminator)
5029 {
5030 *errorcodeptr = ERR29;
5031 goto FAILED;
5032 }
5033
5034 if (refsign == CHAR_MINUS)
5035 {
5036 if (recno == 0)
5037 {
5038 *errorcodeptr = ERR58;
5039 goto FAILED;
5040 }
5041 recno = cd->bracount - recno + 1;
5042 if (recno <= 0)
5043 {
5044 *errorcodeptr = ERR15;
5045 goto FAILED;
5046 }
5047 }
5048 else if (refsign == CHAR_PLUS)
5049 {
5050 if (recno == 0)
5051 {
5052 *errorcodeptr = ERR58;
5053 goto FAILED;
5054 }
5055 recno += cd->bracount;
5056 }
5057
5058 /* Come here from code above that handles a named recursion */
5059
5060 HANDLE_RECURSION:
5061
5062 previous = code;
5063 called = cd->start_code;
5064
5065 /* When we are actually compiling, find the bracket that is being
5066 referenced. Temporarily end the regex in case it doesn't exist before
5067 this point. If we end up with a forward reference, first check that
5068 the bracket does occur later so we can give the error (and position)
5069 now. Then remember this forward reference in the workspace so it can
5070 be filled in at the end. */
5071
5072 if (lengthptr == NULL)
5073 {
5074 *code = OP_END;
5075 if (recno != 0) called = find_bracket(cd->start_code, utf8, recno);
5076
5077 /* Forward reference */
5078
5079 if (called == NULL)
5080 {
5081 if (find_parens(cd, NULL, recno,
5082 (options & PCRE_EXTENDED) != 0) < 0)
5083 {
5084 *errorcodeptr = ERR15;
5085 goto FAILED;
5086 }
5087 called = cd->start_code + recno;
5088 PUTINC(cd->hwm, 0, code + 2 + LINK_SIZE - cd->start_code);
5089 }
5090
5091 /* If not a forward reference, and the subpattern is still open,
5092 this is a recursive call. We check to see if this is a left
5093 recursion that could loop for ever, and diagnose that case. */
5094
5095 else if (GET(called, 1) == 0 &&
5096 could_be_empty(called, code, bcptr, utf8))
5097 {
5098 *errorcodeptr = ERR40;
5099 goto FAILED;
5100 }
5101 }
5102
5103 /* Insert the recursion/subroutine item, automatically wrapped inside
5104 "once" brackets. Set up a "previous group" length so that a
5105 subsequent quantifier will work. */
5106
5107 *code = OP_ONCE;
5108 PUT(code, 1, 2 + 2*LINK_SIZE);
5109 code += 1 + LINK_SIZE;
5110
5111 *code = OP_RECURSE;
5112 PUT(code, 1, called - cd->start_code);
5113 code += 1 + LINK_SIZE;
5114
5115 *code = OP_KET;
5116 PUT(code, 1, 2 + 2*LINK_SIZE);
5117 code += 1 + LINK_SIZE;
5118
5119 length_prevgroup = 3 + 3*LINK_SIZE;
5120 }
5121
5122 /* Can't determine a first byte now */
5123
5124 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
5125 continue;
5126
5127
5128 /* ------------------------------------------------------------ */
5129 default: /* Other characters: check option setting */
5130 OTHER_CHAR_AFTER_QUERY:
5131 set = unset = 0;
5132 optset = &set;
5133
5134 while (*ptr != CHAR_RIGHT_PARENTHESIS && *ptr != CHAR_COLON)
5135 {
5136 switch (*ptr++)
5137 {
5138 case CHAR_MINUS: optset = &unset; break;
5139
5140 case CHAR_J: /* Record that it changed in the external options */
5141 *optset |= PCRE_DUPNAMES;
5142 cd->external_flags |= PCRE_JCHANGED;
5143 break;
5144
5145 case CHAR_i: *optset |= PCRE_CASELESS; break;
5146 case CHAR_m: *optset |= PCRE_MULTILINE; break;
5147 case CHAR_s: *optset |= PCRE_DOTALL; break;
5148 case CHAR_x: *optset |= PCRE_EXTENDED; break;
5149 case CHAR_U: *optset |= PCRE_UNGREEDY; break;
5150 case CHAR_X: *optset |= PCRE_EXTRA; break;
5151
5152 default: *errorcodeptr = ERR12;
5153 ptr--; /* Correct the offset */
5154 goto FAILED;
5155 }
5156 }
5157
5158 /* Set up the changed option bits, but don't change anything yet. */
5159
5160 newoptions = (options | set) & (~unset);
5161
5162 /* If the options ended with ')' this is not the start of a nested
5163 group with option changes, so the options change at this level. If this
5164 item is right at the start of the pattern, the options can be
5165 abstracted and made external in the pre-compile phase, and ignored in
5166 the compile phase. This can be helpful when matching -- for instance in
5167 caseless checking of required bytes.
5168
5169 If the code pointer is not (cd->start_code + 1 + LINK_SIZE), we are
5170 definitely *not* at the start of the pattern because something has been
5171 compiled. In the pre-compile phase, however, the code pointer can have
5172 that value after the start, because it gets reset as code is discarded
5173 during the pre-compile. However, this can happen only at top level - if
5174 we are within parentheses, the starting BRA will still be present. At
5175 any parenthesis level, the length value can be used to test if anything
5176 has been compiled at that level. Thus, a test for both these conditions
5177 is necessary to ensure we correctly detect the start of the pattern in
5178 both phases.
5179
5180 If we are not at the pattern start, compile code to change the ims
5181 options if this setting actually changes any of them, and reset the
5182 greedy defaults and the case value for firstbyte and reqbyte. */
5183
5184 if (*ptr == CHAR_RIGHT_PARENTHESIS)
5185 {
5186 if (code == cd->start_code + 1 + LINK_SIZE &&
5187 (lengthptr == NULL || *lengthptr == 2 + 2*LINK_SIZE))
5188 {
5189 cd->external_options = newoptions;
5190 }
5191 else
5192 {
5193 if ((options & PCRE_IMS) != (newoptions & PCRE_IMS))
5194 {
5195 *code++ = OP_OPT;
5196 *code++ = newoptions & PCRE_IMS;
5197 }
5198 greedy_default = ((newoptions & PCRE_UNGREEDY) != 0);
5199 greedy_non_default = greedy_default ^ 1;
5200 req_caseopt = ((newoptions & PCRE_CASELESS) != 0)? REQ_CASELESS : 0;
5201 }
5202
5203 /* Change options at this level, and pass them back for use
5204 in subsequent branches. When not at the start of the pattern, this
5205 information is also necessary so that a resetting item can be
5206 compiled at the end of a group (if we are in a group). */
5207
5208 *optionsptr = options = newoptions;
5209 previous = NULL; /* This item can't be repeated */
5210 continue; /* It is complete */
5211 }
5212
5213 /* If the options ended with ':' we are heading into a nested group
5214 with possible change of options. Such groups are non-capturing and are
5215 not assertions of any kind. All we need to do is skip over the ':';
5216 the newoptions value is handled below. */
5217
5218 bravalue = OP_BRA;
5219 ptr++;
5220 } /* End of switch for character following (? */
5221 } /* End of (? handling */
5222
5223 /* Opening parenthesis not followed by '?'. If PCRE_NO_AUTO_CAPTURE is set,
5224 all unadorned brackets become non-capturing and behave like (?:...)
5225 brackets. */
5226
5227 else if ((options & PCRE_NO_AUTO_CAPTURE) != 0)
5228 {
5229 bravalue = OP_BRA;
5230 }
5231
5232 /* Else we have a capturing group. */
5233
5234 else
5235 {
5236 NUMBERED_GROUP:
5237 cd->bracount += 1;
5238 PUT2(code, 1+LINK_SIZE, cd->bracount);
5239 skipbytes = 2;
5240 }
5241
5242 /* Process nested bracketed regex. Assertions may not be repeated, but
5243 other kinds can be. All their opcodes are >= OP_ONCE. We copy code into a
5244 non-register variable in order to be able to pass its address because some
5245 compilers complain otherwise. Pass in a new setting for the ims options if
5246 they have changed. */
5247
5248 previous = (bravalue >= OP_ONCE)? code : NULL;
5249 *code = bravalue;
5250 tempcode = code;
5251 tempreqvary = cd->req_varyopt; /* Save value before bracket */
5252 length_prevgroup = 0; /* Initialize for pre-compile phase */
5253
5254 if (!compile_regex(
5255 newoptions, /* The complete new option state */
5256 options & PCRE_IMS, /* The previous ims option state */
5257 &tempcode, /* Where to put code (updated) */
5258 &ptr, /* Input pointer (updated) */
5259 errorcodeptr, /* Where to put an error message */
5260 (bravalue == OP_ASSERTBACK ||
5261 bravalue == OP_ASSERTBACK_NOT), /* TRUE if back assert */
5262 reset_bracount, /* True if (?| group */
5263 skipbytes, /* Skip over bracket number */
5264 &subfirstbyte, /* For possible first char */
5265 &subreqbyte, /* For possible last char */
5266 bcptr, /* Current branch chain */
5267 cd, /* Tables block */
5268 (lengthptr == NULL)? NULL : /* Actual compile phase */
5269 &length_prevgroup /* Pre-compile phase */
5270 ))
5271 goto FAILED;
5272
5273 /* At the end of compiling, code is still pointing to the start of the
5274 group, while tempcode has been updated to point past the end of the group
5275 and any option resetting that may follow it. The pattern pointer (ptr)
5276 is on the bracket. */
5277
5278 /* If this is a conditional bracket, check that there are no more than
5279 two branches in the group, or just one if it's a DEFINE group. We do this
5280 in the real compile phase, not in the pre-pass, where the whole group may
5281 not be available. */
5282
5283 if (bravalue == OP_COND && lengthptr == NULL)
5284 {
5285 uschar *tc = code;
5286 int condcount = 0;
5287
5288 do {
5289 condcount++;
5290 tc += GET(tc,1);
5291 }
5292 while (*tc != OP_KET);
5293
5294 /* A DEFINE group is never obeyed inline (the "condition" is always
5295 false). It must have only one branch. */
5296
5297 if (code[LINK_SIZE+1] == OP_DEF)
5298 {
5299 if (condcount > 1)
5300 {
5301 *errorcodeptr = ERR54;
5302 goto FAILED;
5303 }
5304 bravalue = OP_DEF; /* Just a flag to suppress char handling below */
5305 }
5306
5307 /* A "normal" conditional group. If there is just one branch, we must not
5308 make use of its firstbyte or reqbyte, because this is equivalent to an
5309 empty second branch. */
5310
5311 else
5312 {
5313 if (condcount > 2)
5314 {
5315 *errorcodeptr = ERR27;
5316 goto FAILED;
5317 }
5318 if (condcount == 1) subfirstbyte = subreqbyte = REQ_NONE;
5319 }
5320 }
5321
5322 /* Error if hit end of pattern */
5323
5324 if (*ptr != CHAR_RIGHT_PARENTHESIS)
5325 {
5326 *errorcodeptr = ERR14;
5327 goto FAILED;
5328 }
5329
5330 /* In the pre-compile phase, update the length by the length of the group,
5331 less the brackets at either end. Then reduce the compiled code to just a
5332 set of non-capturing brackets so that it doesn't use much memory if it is
5333 duplicated by a quantifier.*/
5334
5335 if (lengthptr != NULL)
5336 {
5337 if (OFLOW_MAX - *lengthptr < length_prevgroup - 2 - 2*LINK_SIZE)
5338 {
5339 *errorcodeptr = ERR20;
5340 goto FAILED;
5341 }
5342 *lengthptr += length_prevgroup - 2 - 2*LINK_SIZE;
5343 *code++ = OP_BRA;
5344 PUTINC(code, 0, 1 + LINK_SIZE);
5345 *code++ = OP_KET;
5346 PUTINC(code, 0, 1 + LINK_SIZE);
5347 break; /* No need to waste time with special character handling */
5348 }
5349
5350 /* Otherwise update the main code pointer to the end of the group. */
5351
5352 code = tempcode;
5353
5354 /* For a DEFINE group, required and first character settings are not
5355 relevant. */
5356
5357 if (bravalue == OP_DEF) break;
5358
5359 /* Handle updating of the required and first characters for other types of
5360 group. Update for normal brackets of all kinds, and conditions with two
5361 branches (see code above). If the bracket is followed by a quantifier with
5362 zero repeat, we have to back off. Hence the definition of zeroreqbyte and
5363 zerofirstbyte outside the main loop so that they can be accessed for the
5364 back off. */
5365
5366 zeroreqbyte = reqbyte;
5367 zerofirstbyte = firstbyte;
5368 groupsetfirstbyte = FALSE;
5369
5370 if (bravalue >= OP_ONCE)
5371 {
5372 /* If we have not yet set a firstbyte in this branch, take it from the
5373 subpattern, remembering that it was set here so that a repeat of more
5374 than one can replicate it as reqbyte if necessary. If the subpattern has
5375 no firstbyte, set "none" for the whole branch. In both cases, a zero
5376 repeat forces firstbyte to "none". */
5377
5378 if (firstbyte == REQ_UNSET)
5379 {
5380 if (subfirstbyte >= 0)
5381 {
5382 firstbyte = subfirstbyte;
5383 groupsetfirstbyte = TRUE;
5384 }
5385 else firstbyte = REQ_NONE;
5386 zerofirstbyte = REQ_NONE;
5387 }
5388
5389 /* If firstbyte was previously set, convert the subpattern's firstbyte
5390 into reqbyte if there wasn't one, using the vary flag that was in
5391 existence beforehand. */
5392
5393 else if (subfirstbyte >= 0 && subreqbyte < 0)
5394 subreqbyte = subfirstbyte | tempreqvary;
5395
5396 /* If the subpattern set a required byte (or set a first byte that isn't
5397 really the first byte - see above), set it. */
5398
5399 if (subreqbyte >= 0) reqbyte = subreqbyte;
5400 }
5401
5402 /* For a forward assertion, we take the reqbyte, if set. This can be
5403 helpful if the pattern that follows the assertion doesn't set a different
5404 char. For example, it's useful for /(?=abcde).+/. We can't set firstbyte
5405 for an assertion, however because it leads to incorrect effect for patterns
5406 such as /(?=a)a.+/ when the "real" "a" would then become a reqbyte instead
5407 of a firstbyte. This is overcome by a scan at the end if there's no
5408 firstbyte, looking for an asserted first char. */
5409
5410 else if (bravalue == OP_ASSERT && subreqbyte >= 0) reqbyte = subreqbyte;
5411 break; /* End of processing '(' */
5412
5413
5414 /* ===================================================================*/
5415 /* Handle metasequences introduced by \. For ones like \d, the ESC_ values
5416 are arranged to be the negation of the corresponding OP_values. For the
5417 back references, the values are ESC_REF plus the reference number. Only
5418 back references and those types that consume a character may be repeated.
5419 We can test for values between ESC_b and ESC_Z for the latter; this may
5420 have to change if any new ones are ever created. */
5421
5422 case CHAR_BACKSLASH:
5423 tempptr = ptr;
5424 c = check_escape(&ptr, errorcodeptr, cd->bracount, options, FALSE);
5425 if (*errorcodeptr != 0) goto FAILED;
5426
5427 if (c < 0)
5428 {
5429 if (-c == ESC_Q) /* Handle start of quoted string */
5430 {
5431 if (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
5432 ptr += 2; /* avoid empty string */
5433 else inescq = TRUE;
5434 continue;
5435 }
5436
5437 if (-c == ESC_E) continue; /* Perl ignores an orphan \E */
5438
5439 /* For metasequences that actually match a character, we disable the
5440 setting of a first character if it hasn't already been set. */
5441
5442 if (firstbyte == REQ_UNSET && -c > ESC_b && -c < ESC_Z)
5443 firstbyte = REQ_NONE;
5444
5445 /* Set values to reset to if this is followed by a zero repeat. */
5446
5447 zerofirstbyte = firstbyte;
5448 zeroreqbyte = reqbyte;
5449
5450 /* \g<name> or \g'name' is a subroutine call by name and \g<n> or \g'n'
5451 is a subroutine call by number (Oniguruma syntax). In fact, the value
5452 -ESC_g is returned only for these cases. So we don't need to check for <
5453 or ' if the value is -ESC_g. For the Perl syntax \g{n} the value is
5454 -ESC_REF+n, and for the Perl syntax \g{name} the result is -ESC_k (as
5455 that is a synonym for a named back reference). */
5456
5457 if (-c == ESC_g)
5458 {
5459 const uschar *p;
5460 save_hwm = cd->hwm; /* Normally this is set when '(' is read */
5461 terminator = (*(++ptr) == CHAR_LESS_THAN_SIGN)?
5462 CHAR_GREATER_THAN_SIGN : CHAR_APOSTROPHE;
5463
5464 /* These two statements stop the compiler for warning about possibly
5465 unset variables caused by the jump to HANDLE_NUMERICAL_RECURSION. In
5466 fact, because we actually check for a number below, the paths that
5467 would actually be in error are never taken. */
5468
5469 skipbytes = 0;
5470 reset_bracount = FALSE;
5471
5472 /* Test for a name */
5473
5474 if (ptr[1] != CHAR_PLUS && ptr[1] != CHAR_MINUS)
5475 {
5476 BOOL isnumber = TRUE;
5477 for (p = ptr + 1; *p != 0 && *p != terminator; p++)
5478 {
5479 if ((cd->ctypes[*p] & ctype_digit) == 0) isnumber = FALSE;
5480 if ((cd->ctypes[*p] & ctype_word) == 0) break;
5481 }
5482 if (*p != terminator)
5483 {
5484 *errorcodeptr = ERR57;
5485 break;
5486 }
5487 if (isnumber)
5488 {
5489 ptr++;
5490 goto HANDLE_NUMERICAL_RECURSION;
5491 }
5492 is_recurse = TRUE;
5493 goto NAMED_REF_OR_RECURSE;
5494 }
5495
5496 /* Test a signed number in angle brackets or quotes. */
5497
5498 p = ptr + 2;
5499 while ((digitab[*p] & ctype_digit) != 0) p++;
5500 if (*p != terminator)
5501 {
5502 *errorcodeptr = ERR57;
5503 break;
5504 }
5505 ptr++;
5506 goto HANDLE_NUMERICAL_RECURSION;
5507 }
5508
5509 /* \k<name> or \k'name' is a back reference by name (Perl syntax).
5510 We also support \k{name} (.NET syntax) */
5511
5512 if (-c == ESC_k && (ptr[1] == CHAR_LESS_THAN_SIGN ||
5513 ptr[1] == CHAR_APOSTROPHE || ptr[1] == CHAR_LEFT_CURLY_BRACKET))
5514 {
5515 is_recurse = FALSE;
5516 terminator = (*(++ptr) == CHAR_LESS_THAN_SIGN)?
5517 CHAR_GREATER_THAN_SIGN : (*ptr == CHAR_APOSTROPHE)?
5518 CHAR_APOSTROPHE : CHAR_RIGHT_CURLY_BRACKET;
5519 goto NAMED_REF_OR_RECURSE;
5520 }
5521
5522 /* Back references are handled specially; must disable firstbyte if
5523 not set to cope with cases like (?=(\w+))\1: which would otherwise set
5524 ':' later. */
5525
5526 if (-c >= ESC_REF)
5527 {
5528 recno = -c - ESC_REF;
5529
5530 HANDLE_REFERENCE: /* Come here from named backref handling */
5531 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
5532 previous = code;
5533 *code++ = OP_REF;
5534 PUT2INC(code, 0, recno);
5535 cd->backref_map |= (recno < 32)? (1 << recno) : 1;
5536 if (recno > cd->top_backref) cd->top_backref = recno;
5537 }
5538
5539 /* So are Unicode property matches, if supported. */
5540
5541 #ifdef SUPPORT_UCP
5542 else if (-c == ESC_P || -c == ESC_p)
5543 {
5544 BOOL negated;
5545 int pdata;
5546 int ptype = get_ucp(&ptr, &negated, &pdata, errorcodeptr);
5547 if (ptype < 0) goto FAILED;
5548 previous = code;
5549 *code++ = ((-c == ESC_p) != negated)? OP_PROP : OP_NOTPROP;
5550 *code++ = ptype;
5551 *code++ = pdata;
5552 }
5553 #else
5554
5555 /* If Unicode properties are not supported, \X, \P, and \p are not
5556 allowed. */
5557
5558 else if (-c == ESC_X || -c == ESC_P || -c == ESC_p)
5559 {
5560 *errorcodeptr = ERR45;
5561 goto FAILED;
5562 }
5563 #endif
5564
5565 /* For the rest (including \X when Unicode properties are supported), we
5566 can obtain the OP value by negating the escape value. */
5567
5568 else
5569 {
5570 previous = (-c > ESC_b && -c < ESC_Z)? code : NULL;
5571 *code++ = -c;
5572 }
5573 continue;
5574 }
5575
5576 /* We have a data character whose value is in c. In UTF-8 mode it may have
5577 a value > 127. We set its representation in the length/buffer, and then
5578 handle it as a data character. */
5579
5580 #ifdef SUPPORT_UTF8
5581 if (utf8 && c > 127)
5582 mclength = _pcre_ord2utf8(c, mcbuffer);
5583 else
5584 #endif
5585
5586 {
5587 mcbuffer[0] = c;
5588 mclength = 1;
5589 }
5590 goto ONE_CHAR;
5591
5592
5593 /* ===================================================================*/
5594 /* Handle a literal character. It is guaranteed not to be whitespace or #
5595 when the extended flag is set. If we are in UTF-8 mode, it may be a
5596 multi-byte literal character. */
5597
5598 default:
5599 NORMAL_CHAR:
5600 mclength = 1;
5601 mcbuffer[0] = c;
5602
5603 #ifdef SUPPORT_UTF8
5604 if (utf8 && c >= 0xc0)
5605 {
5606 while ((ptr[1] & 0xc0) == 0x80)
5607 mcbuffer[mclength++] = *(++ptr);
5608 }
5609 #endif
5610
5611 /* At this point we have the character's bytes in mcbuffer, and the length
5612 in mclength. When not in UTF-8 mode, the length is always 1. */
5613
5614 ONE_CHAR:
5615 previous = code;
5616 *code++ = ((options & PCRE_CASELESS) != 0)? OP_CHARNC : OP_CHAR;
5617 for (c = 0; c < mclength; c++) *code++ = mcbuffer[c];
5618
5619 /* Remember if \r or \n were seen */
5620
5621 if (mcbuffer[0] == CHAR_CR || mcbuffer[0] == CHAR_NL)
5622 cd->external_flags |= PCRE_HASCRORLF;
5623
5624 /* Set the first and required bytes appropriately. If no previous first
5625 byte, set it from this character, but revert to none on a zero repeat.
5626 Otherwise, leave the firstbyte value alone, and don't change it on a zero
5627 repeat. */
5628
5629 if (firstbyte == REQ_UNSET)
5630 {
5631 zerofirstbyte = REQ_NONE;
5632 zeroreqbyte = reqbyte;
5633
5634 /* If the character is more than one byte long, we can set firstbyte
5635 only if it is not to be matched caselessly. */
5636
5637 if (mclength == 1 || req_caseopt == 0)
5638 {
5639 firstbyte = mcbuffer[0] | req_caseopt;
5640 if (mclength != 1) reqbyte = code[-1] | cd->req_varyopt;
5641 }
5642 else firstbyte = reqbyte = REQ_NONE;
5643 }
5644
5645 /* firstbyte was previously set; we can set reqbyte only the length is
5646 1 or the matching is caseful. */
5647
5648 else
5649 {
5650 zerofirstbyte = firstbyte;
5651 zeroreqbyte = reqbyte;
5652 if (mclength == 1 || req_caseopt == 0)
5653 reqbyte = code[-1] | req_caseopt | cd->req_varyopt;
5654 }
5655
5656 break; /* End of literal character handling */
5657 }
5658 } /* end of big loop */
5659
5660
5661 /* Control never reaches here by falling through, only by a goto for all the
5662 error states. Pass back the position in the pattern so that it can be displayed
5663 to the user for diagnosing the error. */
5664
5665 FAILED:
5666 *ptrptr = ptr;
5667 return FALSE;
5668 }
5669
5670
5671
5672
5673 /*************************************************
5674 * Compile sequence of alternatives *
5675 *************************************************/
5676
5677 /* On entry, ptr is pointing past the bracket character, but on return it
5678 points to the closing bracket, or vertical bar, or end of string. The code
5679 variable is pointing at the byte into which the BRA operator has been stored.
5680 If the ims options are changed at the start (for a (?ims: group) or during any
5681 branch, we need to insert an OP_OPT item at the start of every following branch
5682 to ensure they get set correctly at run time, and also pass the new options
5683 into every subsequent branch compile.
5684
5685 This function is used during the pre-compile phase when we are trying to find
5686 out the amount of memory needed, as well as during the real compile phase. The
5687 value of lengthptr distinguishes the two phases.
5688
5689 Arguments:
5690 options option bits, including any changes for this subpattern
5691 oldims previous settings of ims option bits
5692 codeptr -> the address of the current code pointer
5693 ptrptr -> the address of the current pattern pointer
5694 errorcodeptr -> pointer to error code variable
5695 lookbehind TRUE if this is a lookbehind assertion
5696 reset_bracount TRUE to reset the count for each branch
5697 skipbytes skip this many bytes at start (for brackets and OP_COND)
5698 firstbyteptr place to put the first required character, or a negative number
5699 reqbyteptr place to put the last required character, or a negative number
5700 bcptr pointer to the chain of currently open branches
5701 cd points to the data block with tables pointers etc.
5702 lengthptr NULL during the real compile phase
5703 points to length accumulator during pre-compile phase
5704
5705 Returns: TRUE on success
5706 */
5707
5708 static BOOL
5709 compile_regex(int options, int oldims, uschar **codeptr, const uschar **ptrptr,
5710 int *errorcodeptr, BOOL lookbehind, BOOL reset_bracount, int skipbytes,
5711 int *firstbyteptr, int *reqbyteptr, branch_chain *bcptr, compile_data *cd,
5712 int *lengthptr)
5713 {
5714 const uschar *ptr = *ptrptr;
5715 uschar *code = *codeptr;
5716 uschar *last_branch = code;
5717 uschar *start_bracket = code;
5718 uschar *reverse_count = NULL;
5719 open_capitem capitem;
5720 int capnumber = 0;
5721 int firstbyte, reqbyte;
5722 int branchfirstbyte, branchreqbyte;
5723 int length;
5724 int orig_bracount;
5725 int max_bracount;
5726 branch_chain bc;
5727
5728 bc.outer = bcptr;
5729 bc.current = code;
5730
5731 firstbyte = reqbyte = REQ_UNSET;
5732
5733 /* Accumulate the length for use in the pre-compile phase. Start with the
5734 length of the BRA and KET and any extra bytes that are required at the
5735 beginning. We accumulate in a local variable to save frequent testing of
5736 lenthptr for NULL. We cannot do this by looking at the value of code at the
5737 start and end of each alternative, because compiled items are discarded during
5738 the pre-compile phase so that the work space is not exceeded. */
5739
5740 length = 2 + 2*LINK_SIZE + skipbytes;
5741
5742 /* WARNING: If the above line is changed for any reason, you must also change
5743 the code that abstracts option settings at the start of the pattern and makes
5744 them global. It tests the value of length for (2 + 2*LINK_SIZE) in the
5745 pre-compile phase to find out whether anything has yet been compiled or not. */
5746
5747 /* If this is a capturing subpattern, add to the chain of open capturing items
5748 so that we can detect them if (*ACCEPT) is encountered. */
5749
5750 if (*code == OP_CBRA)
5751 {
5752 capnumber = GET2(code, 1 + LINK_SIZE);
5753 capitem.number = capnumber;
5754 capitem.next = cd->open_caps;
5755 cd->open_caps = &capitem;
5756 }
5757
5758 /* Offset is set zero to mark that this bracket is still open */
5759
5760 PUT(code, 1, 0);
5761 code += 1 + LINK_SIZE + skipbytes;
5762
5763 /* Loop for each alternative branch */
5764
5765 orig_bracount = max_bracount = cd->bracount;
5766 for (;;)
5767 {
5768 /* For a (?| group, reset the capturing bracket count so that each branch
5769 uses the same numbers. */
5770
5771 if (reset_bracount) cd->bracount = orig_bracount;
5772
5773 /* Handle a change of ims options at the start of the branch */
5774
5775 if ((options & PCRE_IMS) != oldims)
5776 {
5777 *code++ = OP_OPT;
5778 *code++ = options & PCRE_IMS;
5779 length += 2;
5780 }
5781
5782 /* Set up dummy OP_REVERSE if lookbehind assertion */
5783
5784 if (lookbehind)
5785 {
5786 *code++ = OP_REVERSE;
5787 reverse_count = code;
5788 PUTINC(code, 0, 0);
5789 length += 1 + LINK_SIZE;
5790 }
5791
5792 /* Now compile the branch; in the pre-compile phase its length gets added
5793 into the length. */
5794
5795 if (!compile_branch(&options, &code, &ptr, errorcodeptr, &branchfirstbyte,
5796 &branchreqbyte, &bc, cd, (lengthptr == NULL)? NULL : &length))
5797 {
5798 *ptrptr = ptr;
5799 return FALSE;
5800 }
5801
5802 /* Keep the highest bracket count in case (?| was used and some branch
5803 has fewer than the rest. */
5804
5805 if (cd->bracount > max_bracount) max_bracount = cd->bracount;
5806
5807 /* In the real compile phase, there is some post-processing to be done. */
5808
5809 if (lengthptr == NULL)
5810 {
5811 /* If this is the first branch, the firstbyte and reqbyte values for the
5812 branch become the values for the regex. */
5813
5814 if (*last_branch != OP_ALT)
5815 {
5816 firstbyte = branchfirstbyte;
5817 reqbyte = branchreqbyte;
5818 }
5819
5820 /* If this is not the first branch, the first char and reqbyte have to
5821 match the values from all the previous branches, except that if the
5822 previous value for reqbyte didn't have REQ_VARY set, it can still match,
5823 and we set REQ_VARY for the regex. */
5824
5825 else
5826 {
5827 /* If we previously had a firstbyte, but it doesn't match the new branch,
5828 we have to abandon the firstbyte for the regex, but if there was
5829 previously no reqbyte, it takes on the value of the old firstbyte. */
5830
5831 if (firstbyte >= 0 && firstbyte != branchfirstbyte)
5832 {
5833 if (reqbyte < 0) reqbyte = firstbyte;
5834 firstbyte = REQ_NONE;
5835 }
5836
5837 /* If we (now or from before) have no firstbyte, a firstbyte from the
5838 branch becomes a reqbyte if there isn't a branch reqbyte. */
5839
5840 if (firstbyte < 0 && branchfirstbyte >= 0 && branchreqbyte < 0)
5841 branchreqbyte = branchfirstbyte;
5842
5843 /* Now ensure that the reqbytes match */
5844
5845 if ((reqbyte & ~REQ_VARY) != (branchreqbyte & ~REQ_VARY))
5846 reqbyte = REQ_NONE;
5847 else reqbyte |= branchreqbyte; /* To "or" REQ_VARY */
5848 }
5849
5850 /* If lookbehind, check that this branch matches a fixed-length string, and
5851 put the length into the OP_REVERSE item. Temporarily mark the end of the
5852 branch with OP_END. If the branch contains OP_RECURSE, the result is -3
5853 because there may be forward references that we can't check here. Set a
5854 flag to cause another lookbehind check at the end. Why not do it all at the
5855 end? Because common, erroneous checks are picked up here and the offset of
5856 the problem can be shown. */
5857
5858 if (lookbehind)
5859 {
5860 int fixed_length;
5861 *code = OP_END;
5862 fixed_length = find_fixedlength(last_branch, options, FALSE, cd);
5863 DPRINTF(("fixed length = %d\n", fixed_length));
5864 if (fixed_length == -3)
5865 {
5866 cd->check_lookbehind = TRUE;
5867 }
5868 else if (fixed_length < 0)
5869 {
5870 *errorcodeptr = (fixed_length == -2)? ERR36 : ERR25;
5871 *ptrptr = ptr;
5872 return FALSE;
5873 }
5874 else { PUT(reverse_count, 0, fixed_length); }
5875 }
5876 }
5877
5878 /* Reached end of expression, either ')' or end of pattern. In the real
5879 compile phase, go back through the alternative branches and reverse the chain
5880 of offsets, with the field in the BRA item now becoming an offset to the
5881 first alternative. If there are no alternatives, it points to the end of the
5882 group. The length in the terminating ket is always the length of the whole
5883 bracketed item. If any of the ims options were changed inside the group,
5884 compile a resetting op-code following, except at the very end of the pattern.
5885 Return leaving the pointer at the terminating char. */
5886
5887 if (*ptr != CHAR_VERTICAL_LINE)
5888 {
5889 if (lengthptr == NULL)
5890 {
5891 int branch_length = code - last_branch;
5892 do
5893 {
5894 int prev_length = GET(last_branch, 1);
5895 PUT(last_branch, 1, branch_length);
5896 branch_length = prev_length;
5897 last_branch -= branch_length;
5898 }
5899 while (branch_length > 0);
5900 }
5901
5902 /* If it was a capturing subpattern, remove it from the chain. */
5903
5904 if (capnumber > 0) cd->open_caps = cd->open_caps->next;
5905
5906 /* Fill in the ket */
5907
5908 *code = OP_KET;
5909 PUT(code, 1, code - start_bracket);
5910 code += 1 + LINK_SIZE;
5911
5912 /* Resetting option if needed */
5913
5914 if ((options & PCRE_IMS) != oldims && *ptr == CHAR_RIGHT_PARENTHESIS)
5915 {
5916 *code++ = OP_OPT;
5917 *code++ = oldims;
5918 length += 2;
5919 }
5920
5921 /* Retain the highest bracket number, in case resetting was used. */
5922
5923 cd->bracount = max_bracount;
5924
5925 /* Set values to pass back */
5926
5927 *codeptr = code;
5928 *ptrptr = ptr;
5929 *firstbyteptr = firstbyte;
5930 *reqbyteptr = reqbyte;
5931 if (lengthptr != NULL)
5932 {
5933 if (OFLOW_MAX - *lengthptr < length)
5934 {
5935 *errorcodeptr = ERR20;
5936 return FALSE;
5937 }
5938 *lengthptr += length;
5939 }
5940 return TRUE;
5941 }
5942
5943 /* Another branch follows. In the pre-compile phase, we can move the code
5944 pointer back to where it was for the start of the first branch. (That is,
5945 pretend that each branch is the only one.)
5946
5947 In the real compile phase, insert an ALT node. Its length field points back
5948 to the previous branch while the bracket remains open. At the end the chain
5949 is reversed. It's done like this so that the start of the bracket has a
5950 zero offset until it is closed, making it possible to detect recursion. */
5951
5952 if (lengthptr != NULL)
5953 {
5954 code = *codeptr + 1 + LINK_SIZE + skipbytes;
5955 length += 1 + LINK_SIZE;
5956 }
5957 else
5958 {
5959 *code = OP_ALT;
5960 PUT(code, 1, code - last_branch);
5961 bc.current = last_branch = code;
5962 code += 1 + LINK_SIZE;
5963 }
5964
5965 ptr++;
5966 }
5967 /* Control never reaches here */
5968 }
5969
5970
5971
5972
5973 /*************************************************
5974 * Check for anchored expression *
5975 *************************************************/
5976
5977 /* Try to find out if this is an anchored regular expression. Consider each
5978 alternative branch. If they all start with OP_SOD or OP_CIRC, or with a bracket
5979 all of whose alternatives start with OP_SOD or OP_CIRC (recurse ad lib), then
5980 it's anchored. However, if this is a multiline pattern, then only OP_SOD
5981 counts, since OP_CIRC can match in the middle.
5982
5983 We can also consider a regex to be anchored if OP_SOM starts all its branches.
5984 This is the code for \G, which means "match at start of match position, taking
5985 into account the match offset".
5986
5987 A branch is also implicitly anchored if it starts with .* and DOTALL is set,
5988 because that will try the rest of the pattern at all possible matching points,
5989 so there is no point trying again.... er ....
5990
5991 .... except when the .* appears inside capturing parentheses, and there is a
5992 subsequent back reference to those parentheses. We haven't enough information
5993 to catch that case precisely.
5994
5995 At first, the best we could do was to detect when .* was in capturing brackets
5996 and the highest back reference was greater than or equal to that level.
5997 However, by keeping a bitmap of the first 31 back references, we can catch some
5998 of the more common cases more precisely.
5999
6000 Arguments:
6001 code points to start of expression (the bracket)
6002 options points to the options setting
6003 bracket_map a bitmap of which brackets we are inside while testing; this
6004 handles up to substring 31; after that we just have to take
6005 the less precise approach
6006 backref_map the back reference bitmap
6007
6008 Returns: TRUE or FALSE
6009 */
6010
6011 static BOOL
6012 is_anchored(register const uschar *code, int *options, unsigned int bracket_map,
6013 unsigned int backref_map)
6014 {
6015 do {
6016 const uschar *scode = first_significant_code(code + _pcre_OP_lengths[*code],
6017 options, PCRE_MULTILINE, FALSE);
6018 register int op = *scode;
6019
6020 /* Non-capturing brackets */
6021
6022 if (op == OP_BRA)
6023 {
6024 if (!is_anchored(scode, options, bracket_map, backref_map)) return FALSE;
6025 }
6026
6027 /* Capturing brackets */
6028
6029 else if (op == OP_CBRA)
6030 {
6031 int n = GET2(scode, 1+LINK_SIZE);
6032 int new_map = bracket_map | ((n < 32)? (1 << n) : 1);
6033 if (!is_anchored(scode, options, new_map, backref_map)) return FALSE;
6034 }
6035
6036 /* Other brackets */
6037
6038 else if (op == OP_ASSERT || op == OP_ONCE || op == OP_COND)
6039 {
6040 if (!is_anchored(scode, options, bracket_map, backref_map)) return FALSE;
6041 }
6042
6043 /* .* is not anchored unless DOTALL is set (which generates OP_ALLANY) and
6044 it isn't in brackets that are or may be referenced. */
6045
6046 else if ((op == OP_TYPESTAR || op == OP_TYPEMINSTAR ||
6047 op == OP_TYPEPOSSTAR))
6048 {
6049 if (scode[1] != OP_ALLANY || (bracket_map & backref_map) != 0)
6050 return FALSE;
6051 }
6052
6053 /* Check for explicit anchoring */
6054
6055 else if (op != OP_SOD && op != OP_SOM &&
6056 ((*options & PCRE_MULTILINE) != 0 || op != OP_CIRC))
6057 return FALSE;
6058 code += GET(code, 1);
6059 }
6060 while (*code == OP_ALT); /* Loop for each alternative */
6061 return TRUE;
6062 }
6063
6064
6065
6066 /*************************************************
6067 * Check for starting with ^ or .* *
6068 *************************************************/
6069
6070 /* This is called to find out if every branch starts with ^ or .* so that
6071 "first char" processing can be done to speed things up in multiline
6072 matching and for non-DOTALL patterns that start with .* (which must start at
6073 the beginning or after \n). As in the case of is_anchored() (see above), we
6074 have to take account of back references to capturing brackets that contain .*
6075 because in that case we can't make the assumption.
6076
6077 Arguments:
6078 code points to start of expression (the bracket)
6079 bracket_map a bitmap of which brackets we are inside while testing; this
6080 handles up to substring 31; after that we just have to take
6081 the less precise approach
6082 backref_map the back reference bitmap
6083
6084 Returns: TRUE or FALSE
6085 */
6086
6087 static BOOL
6088 is_startline(const uschar *code, unsigned int bracket_map,
6089 unsigned int backref_map)
6090 {
6091 do {
6092 const uschar *scode = first_significant_code(code + _pcre_OP_lengths[*code],
6093 NULL, 0, FALSE);
6094 register int op = *scode;
6095
6096 /* If we are at the start of a conditional assertion group, *both* the
6097 conditional assertion *and* what follows the condition must satisfy the test
6098 for start of line. Other kinds of condition fail. Note that there may be an
6099 auto-callout at the start of a condition. */
6100
6101 if (op == OP_COND)
6102 {
6103 scode += 1 + LINK_SIZE;
6104 if (*scode == OP_CALLOUT) scode += _pcre_OP_lengths[OP_CALLOUT];
6105 switch (*scode)
6106 {
6107 case OP_CREF:
6108 case OP_RREF:
6109 case OP_DEF:
6110 return FALSE;
6111
6112 default: /* Assertion */
6113 if (!is_startline(scode, bracket_map, backref_map)) return FALSE;
6114 do scode += GET(scode, 1); while (*scode == OP_ALT);
6115 scode += 1 + LINK_SIZE;
6116 break;
6117 }
6118 scode = first_significant_code(scode, NULL, 0, FALSE);
6119 op = *scode;
6120 }
6121
6122 /* Non-capturing brackets */
6123
6124 if (op == OP_BRA)
6125 {
6126 if (!is_startline(scode, bracket_map, backref_map)) return FALSE;
6127 }
6128
6129 /* Capturing brackets */
6130
6131 else if (op == OP_CBRA)
6132 {
6133 int n = GET2(scode, 1+LINK_SIZE);
6134 int new_map = bracket_map | ((n < 32)? (1 << n) : 1);
6135 if (!is_startline(scode, new_map, backref_map)) return FALSE;
6136 }
6137
6138 /* Other brackets */
6139
6140 else if (op == OP_ASSERT || op == OP_ONCE)
6141 {
6142 if (!is_startline(scode, bracket_map, backref_map)) return FALSE;
6143 }
6144
6145 /* .* means "start at start or after \n" if it isn't in brackets that
6146 may be referenced. */
6147
6148 else if (op == OP_TYPESTAR || op == OP_TYPEMINSTAR || op == OP_TYPEPOSSTAR)
6149 {
6150 if (scode[1] != OP_ANY || (bracket_map & backref_map) != 0) return FALSE;
6151 }
6152
6153 /* Check for explicit circumflex */
6154
6155 else if (op != OP_CIRC) return FALSE;
6156
6157 /* Move on to the next alternative */
6158
6159 code += GET(code, 1);
6160 }
6161 while (*code == OP_ALT); /* Loop for each alternative */
6162 return TRUE;
6163 }
6164
6165
6166
6167 /*************************************************
6168 * Check for asserted fixed first char *
6169 *************************************************/
6170
6171 /* During compilation, the "first char" settings from forward assertions are
6172 discarded, because they can cause conflicts with actual literals that follow.
6173 However, if we end up without a first char setting for an unanchored pattern,
6174 it is worth scanning the regex to see if there is an initial asserted first
6175 char. If all branches start with the same asserted char, or with a bracket all
6176 of whose alternatives start with the same asserted char (recurse ad lib), then
6177 we return that char, otherwise -1.
6178
6179 Arguments:
6180 code points to start of expression (the bracket)
6181 options pointer to the options (used to check casing changes)
6182 inassert TRUE if in an assertion
6183
6184 Returns: -1 or the fixed first char
6185 */
6186
6187 static int
6188 find_firstassertedchar(const uschar *code, int *options, BOOL inassert)
6189 {
6190 register int c = -1;
6191 do {
6192 int d;
6193 const uschar *scode =
6194 first_significant_code(code + 1+LINK_SIZE, options, PCRE_CASELESS, TRUE);
6195 register int op = *scode;
6196
6197 switch(op)
6198 {
6199 default:
6200 return -1;
6201
6202 case OP_BRA:
6203 case OP_CBRA:
6204 case OP_ASSERT:
6205 case OP_ONCE:
6206 case OP_COND:
6207 if ((d = find_firstassertedchar(scode, options, op == OP_ASSERT)) < 0)
6208 return -1;
6209 if (c < 0) c = d; else if (c != d) return -1;
6210 break;
6211
6212 case OP_EXACT: /* Fall through */
6213 scode += 2;
6214
6215 case OP_CHAR:
6216 case OP_CHARNC:
6217 case OP_PLUS:
6218 case OP_MINPLUS:
6219 case OP_POSPLUS:
6220 if (!inassert) return -1;
6221 if (c < 0)
6222 {
6223 c = scode[1];
6224 if ((*options & PCRE_CASELESS) != 0) c |= REQ_CASELESS;
6225 }
6226 else if (c != scode[1]) return -1;
6227 break;
6228 }
6229
6230 code += GET(code, 1);
6231 }
6232 while (*code == OP_ALT);
6233 return c;
6234 }
6235
6236
6237
6238 /*************************************************
6239 * Compile a Regular Expression *
6240 *************************************************/
6241
6242 /* This function takes a string and returns a pointer to a block of store
6243 holding a compiled version of the expression. The original API for this
6244 function had no error code return variable; it is retained for backwards
6245 compatibility. The new function is given a new name.
6246
6247 Arguments:
6248 pattern the regular expression
6249 options various option bits
6250 errorcodeptr pointer to error code variable (pcre_compile2() only)
6251 can be NULL if you don't want a code value
6252 errorptr pointer to pointer to error text
6253 erroroffset ptr offset in pattern where error was detected
6254 tables pointer to character tables or NULL
6255
6256 Returns: pointer to compiled data block, or NULL on error,
6257 with errorptr and erroroffset set
6258 */
6259
6260 PCRE_EXP_DEFN pcre * PCRE_CALL_CONVENTION
6261 pcre_compile(const char *pattern, int options, const char **errorptr,
6262 int *erroroffset, const unsigned char *tables)
6263 {
6264 return pcre_compile2(pattern, options, NULL, errorptr, erroroffset, tables);
6265 }
6266
6267
6268 PCRE_EXP_DEFN pcre * PCRE_CALL_CONVENTION
6269 pcre_compile2(const char *pattern, int options, int *errorcodeptr,
6270 const char **errorptr, int *erroroffset, const unsigned char *tables)
6271 {
6272 real_pcre *re;
6273 int length = 1; /* For final END opcode */
6274 int firstbyte, reqbyte, newline;
6275 int errorcode = 0;
6276 int skipatstart = 0;
6277 BOOL utf8 = (options & PCRE_UTF8) != 0;
6278 size_t size;
6279 uschar *code;
6280 const uschar *codestart;
6281 const uschar *ptr;
6282 compile_data compile_block;
6283 compile_data *cd = &compile_block;
6284
6285 /* This space is used for "compiling" into during the first phase, when we are
6286 computing the amount of memory that is needed. Compiled items are thrown away
6287 as soon as possible, so that a fairly large buffer should be sufficient for
6288 this purpose. The same space is used in the second phase for remembering where
6289 to fill in forward references to subpatterns. */
6290
6291 uschar cworkspace[COMPILE_WORK_SIZE];
6292
6293 /* Set this early so that early errors get offset 0. */
6294
6295 ptr = (const uschar *)pattern;
6296
6297 /* We can't pass back an error message if errorptr is NULL; I guess the best we
6298 can do is just return NULL, but we can set a code value if there is a code
6299 pointer. */
6300
6301 if (errorptr == NULL)
6302 {
6303 if (errorcodeptr != NULL) *errorcodeptr = 99;
6304 return NULL;
6305 }
6306
6307 *errorptr = NULL;
6308 if (errorcodeptr != NULL) *errorcodeptr = ERR0;
6309
6310 /* However, we can give a message for this error */
6311
6312 if (erroroffset == NULL)
6313 {
6314 errorcode = ERR16;
6315 goto PCRE_EARLY_ERROR_RETURN2;
6316 }
6317
6318 *erroroffset = 0;
6319
6320 /* Set up pointers to the individual character tables */
6321
6322 if (tables == NULL) tables = _pcre_default_tables;
6323 cd->lcc = tables + lcc_offset;
6324 cd->fcc = tables + fcc_offset;
6325 cd->cbits = tables + cbits_offset;
6326 cd->ctypes = tables + ctypes_offset;
6327
6328 /* Check that all undefined public option bits are zero */
6329
6330 if ((options & ~PUBLIC_COMPILE_OPTIONS) != 0)
6331 {
6332 errorcode = ERR17;
6333 goto PCRE_EARLY_ERROR_RETURN;
6334 }
6335
6336 /* Check for global one-time settings at the start of the pattern, and remember
6337 the offset for later. */
6338
6339 while (ptr[skipatstart] == CHAR_LEFT_PARENTHESIS &&
6340 ptr[skipatstart+1] == CHAR_ASTERISK)
6341 {
6342 int newnl = 0;
6343 int newbsr = 0;
6344
6345 if (strncmp((char *)(ptr+skipatstart+2), STRING_UTF8_RIGHTPAR, 5) == 0)
6346 { skipatstart += 7; options |= PCRE_UTF8; continue; }
6347
6348 if (strncmp((char *)(ptr+skipatstart+2), STRING_CR_RIGHTPAR, 3) == 0)
6349 { skipatstart += 5; newnl = PCRE_NEWLINE_CR; }
6350 else if (strncmp((char *)(ptr+skipatstart+2), STRING_LF_RIGHTPAR, 3) == 0)
6351 { skipatstart += 5; newnl = PCRE_NEWLINE_LF; }
6352 else if (strncmp((char *)(ptr+skipatstart+2), STRING_CRLF_RIGHTPAR, 5) == 0)
6353 { skipatstart += 7; newnl = PCRE_NEWLINE_CR + PCRE_NEWLINE_LF; }
6354 else if (strncmp((char *)(ptr+skipatstart+2), STRING_ANY_RIGHTPAR, 4) == 0)
6355 { skipatstart += 6; newnl = PCRE_NEWLINE_ANY; }
6356 else if (strncmp((char *)(ptr+skipatstart+2), STRING_ANYCRLF_RIGHTPAR, 8) == 0)
6357 { skipatstart += 10; newnl = PCRE_NEWLINE_ANYCRLF; }
6358
6359 else if (strncmp((char *)(ptr+skipatstart+2), STRING_BSR_ANYCRLF_RIGHTPAR, 12) == 0)
6360 { skipatstart += 14; newbsr = PCRE_BSR_ANYCRLF; }
6361 else if (strncmp((char *)(ptr+skipatstart+2), STRING_BSR_UNICODE_RIGHTPAR, 12) == 0)
6362 { skipatstart += 14; newbsr = PCRE_BSR_UNICODE; }
6363
6364 if (newnl != 0)
6365 options = (options & ~PCRE_NEWLINE_BITS) | newnl;
6366 else if (newbsr != 0)
6367 options = (options & ~(PCRE_BSR_ANYCRLF|PCRE_BSR_UNICODE)) | newbsr;
6368 else break;
6369 }
6370
6371 /* Can't support UTF8 unless PCRE has been compiled to include the code. */
6372
6373 #ifdef SUPPORT_UTF8
6374 if (utf8 && (options & PCRE_NO_UTF8_CHECK) == 0 &&
6375 (*erroroffset = _pcre_valid_utf8((uschar *)pattern, -1)) >= 0)
6376 {
6377 errorcode = ERR44;
6378 goto PCRE_EARLY_ERROR_RETURN2;
6379 }
6380 #else
6381 if (utf8)
6382 {
6383 errorcode = ERR32;
6384 goto PCRE_EARLY_ERROR_RETURN;
6385 }
6386 #endif
6387
6388 /* Check validity of \R options. */
6389
6390 switch (options & (PCRE_BSR_ANYCRLF|PCRE_BSR_UNICODE))
6391 {
6392 case 0:
6393 case PCRE_BSR_ANYCRLF:
6394 case PCRE_BSR_UNICODE:
6395 break;
6396 default: errorcode = ERR56; goto PCRE_EARLY_ERROR_RETURN;
6397 }
6398
6399 /* Handle different types of newline. The three bits give seven cases. The
6400 current code allows for fixed one- or two-byte sequences, plus "any" and
6401 "anycrlf". */
6402
6403 switch (options & PCRE_NEWLINE_BITS)
6404 {
6405 case 0: newline = NEWLINE; break; /* Build-time default */
6406 case PCRE_NEWLINE_CR: newline = CHAR_CR; break;
6407 case PCRE_NEWLINE_LF: newline = CHAR_NL; break;
6408 case PCRE_NEWLINE_CR+
6409 PCRE_NEWLINE_LF: newline = (CHAR_CR << 8) | CHAR_NL; break;
6410 case PCRE_NEWLINE_ANY: newline = -1; break;
6411 case PCRE_NEWLINE_ANYCRLF: newline = -2; break;
6412 default: errorcode = ERR56; goto PCRE_EARLY_ERROR_RETURN;
6413 }
6414
6415 if (newline == -2)
6416 {
6417 cd->nltype = NLTYPE_ANYCRLF;
6418 }
6419 else if (newline < 0)
6420 {
6421 cd->nltype = NLTYPE_ANY;
6422 }
6423 else
6424 {
6425 cd->nltype = NLTYPE_FIXED;
6426 if (newline > 255)
6427 {
6428 cd->nllen = 2;
6429 cd->nl[0] = (newline >> 8) & 255;
6430 cd->nl[1] = newline & 255;
6431 }
6432 else
6433 {
6434 cd->nllen = 1;
6435 cd->nl[0] = newline;
6436 }
6437 }
6438
6439 /* Maximum back reference and backref bitmap. The bitmap records up to 31 back
6440 references to help in deciding whether (.*) can be treated as anchored or not.
6441 */
6442
6443 cd->top_backref = 0;
6444 cd->backref_map = 0;
6445
6446 /* Reflect pattern for debugging output */
6447
6448 DPRINTF(("------------------------------------------------------------------\n"));
6449 DPRINTF(("%s\n", pattern));
6450
6451 /* Pretend to compile the pattern while actually just accumulating the length
6452 of memory required. This behaviour is triggered by passing a non-NULL final
6453 argument to compile_regex(). We pass a block of workspace (cworkspace) for it
6454 to compile parts of the pattern into; the compiled code is discarded when it is
6455 no longer needed, so hopefully this workspace will never overflow, though there
6456 is a test for its doing so. */
6457
6458 cd->bracount = cd->final_bracount = 0;
6459 cd->names_found = 0;
6460 cd->name_entry_size = 0;
6461 cd->name_table = NULL;
6462 cd->start_workspace = cworkspace;
6463 cd->start_code = cworkspace;
6464 cd->hwm = cworkspace;
6465 cd->start_pattern = (const uschar *)pattern;
6466 cd->end_pattern = (const uschar *)(pattern + strlen(pattern));
6467 cd->req_varyopt = 0;
6468 cd->external_options = options;
6469 cd->external_flags = 0;
6470 cd->open_caps = NULL;
6471
6472 /* Now do the pre-compile. On error, errorcode will be set non-zero, so we
6473 don't need to look at the result of the function here. The initial options have
6474 been put into the cd block so that they can be changed if an option setting is
6475 found within the regex right at the beginning. Bringing initial option settings
6476 outside can help speed up starting point checks. */
6477
6478 ptr += skipatstart;
6479 code = cworkspace;
6480 *code = OP_BRA;
6481 (void)compile_regex(cd->external_options, cd->external_options & PCRE_IMS,
6482 &code, &ptr, &errorcode, FALSE, FALSE, 0, &firstbyte, &reqbyte, NULL, cd,
6483 &length);
6484 if (errorcode != 0) goto PCRE_EARLY_ERROR_RETURN;
6485
6486 DPRINTF(("end pre-compile: length=%d workspace=%d\n", length,
6487 cd->hwm - cworkspace));
6488
6489 if (length > MAX_PATTERN_SIZE)
6490 {
6491 errorcode = ERR20;
6492 goto PCRE_EARLY_ERROR_RETURN;
6493 }
6494
6495 /* Compute the size of data block needed and get it, either from malloc or
6496 externally provided function. Integer overflow should no longer be possible
6497 because nowadays we limit the maximum value of cd->names_found and
6498 cd->name_entry_size. */
6499
6500 size = length + sizeof(real_pcre) + cd->names_found * (cd->name_entry_size + 3);
6501 re = (real_pcre *)(pcre_malloc)(size);
6502
6503 if (re == NULL)
6504 {
6505 errorcode = ERR21;
6506 goto PCRE_EARLY_ERROR_RETURN;
6507 }
6508
6509 /* Put in the magic number, and save the sizes, initial options, internal
6510 flags, and character table pointer. NULL is used for the default character
6511 tables. The nullpad field is at the end; it's there to help in the case when a
6512 regex compiled on a system with 4-byte pointers is run on another with 8-byte
6513 pointers. */
6514
6515 re->magic_number = MAGIC_NUMBER;
6516 re->size = size;
6517 re->options = cd->external_options;
6518 re->flags = cd->external_flags;
6519 re->dummy1 = 0;
6520 re->first_byte = 0;
6521 re->req_byte = 0;
6522 re->name_table_offset = sizeof(real_pcre);
6523 re->name_entry_size = cd->name_entry_size;
6524 re->name_count = cd->names_found;
6525 re->ref_count = 0;
6526 re->tables = (tables == _pcre_default_tables)? NULL : tables;
6527 re->nullpad = NULL;
6528
6529 /* The starting points of the name/number translation table and of the code are
6530 passed around in the compile data block. The start/end pattern and initial
6531 options are already set from the pre-compile phase, as is the name_entry_size
6532 field. Reset the bracket count and the names_found field. Also reset the hwm
6533 field; this time it's used for remembering forward references to subpatterns.
6534 */
6535
6536 cd->final_bracount = cd->bracount; /* Save for checking forward references */
6537 cd->bracount = 0;
6538 cd->names_found = 0;
6539 cd->name_table = (uschar *)re + re->name_table_offset;
6540 codestart = cd->name_table + re->name_entry_size * re->name_count;
6541 cd->start_code = codestart;
6542 cd->hwm = cworkspace;
6543 cd->req_varyopt = 0;
6544 cd->had_accept = FALSE;
6545 cd->check_lookbehind = FALSE;
6546 cd->open_caps = NULL;
6547
6548 /* Set up a starting, non-extracting bracket, then compile the expression. On
6549 error, errorcode will be set non-zero, so we don't need to look at the result
6550 of the function here. */
6551
6552 ptr = (const uschar *)pattern + skipatstart;
6553 code = (uschar *)codestart;
6554 *code = OP_BRA;
6555 (void)compile_regex(re->options, re->options & PCRE_IMS, &code, &ptr,
6556 &errorcode, FALSE, FALSE, 0, &firstbyte, &reqbyte, NULL, cd, NULL);
6557 re->top_bracket = cd->bracount;
6558 re->top_backref = cd->top_backref;
6559 re->flags = cd->external_flags;
6560
6561 if (cd->had_accept) reqbyte = -1; /* Must disable after (*ACCEPT) */
6562
6563 /* If not reached end of pattern on success, there's an excess bracket. */
6564
6565 if (errorcode == 0 && *ptr != 0) errorcode = ERR22;
6566
6567 /* Fill in the terminating state and check for disastrous overflow, but
6568 if debugging, leave the test till after things are printed out. */
6569
6570 *code++ = OP_END;
6571
6572 #ifndef DEBUG
6573 if (code - codestart > length) errorcode = ERR23;
6574 #endif
6575
6576 /* Fill in any forward references that are required. */
6577
6578 while (errorcode == 0 && cd->hwm > cworkspace)
6579 {
6580 int offset, recno;
6581 const uschar *groupptr;
6582 cd->hwm -= LINK_SIZE;
6583 offset = GET(cd->hwm, 0);
6584 recno = GET(codestart, offset);
6585 groupptr = find_bracket(codestart, utf8, recno);
6586 if (groupptr == NULL) errorcode = ERR53;
6587 else PUT(((uschar *)codestart), offset, groupptr - codestart);
6588 }
6589
6590 /* Give an error if there's back reference to a non-existent capturing
6591 subpattern. */
6592
6593 if (errorcode == 0 && re->top_backref > re->top_bracket) errorcode = ERR15;
6594
6595 /* If there were any lookbehind assertions that contained OP_RECURSE
6596 (recursions or subroutine calls), a flag is set for them to be checked here,
6597 because they may contain forward references. Actual recursions can't be fixed
6598 length, but subroutine calls can. It is done like this so that those without
6599 OP_RECURSE that are not fixed length get a diagnosic with a useful offset. The
6600 exceptional ones forgo this. We scan the pattern to check that they are fixed
6601 length, and set their lengths. */
6602
6603 if (cd->check_lookbehind)
6604 {
6605 uschar *cc = (uschar *)codestart;
6606
6607 /* Loop, searching for OP_REVERSE items, and process those that do not have
6608 their length set. (Actually, it will also re-process any that have a length
6609 of zero, but that is a pathological case, and it does no harm.) When we find
6610 one, we temporarily terminate the branch it is in while we scan it. */
6611
6612 for (cc = (uschar *)find_bracket(codestart, utf8, -1);
6613 cc != NULL;
6614 cc = (uschar *)find_bracket(cc, utf8, -1))
6615 {
6616 if (GET(cc, 1) == 0)
6617 {
6618 int fixed_length;
6619 uschar *be = cc - 1 - LINK_SIZE + GET(cc, -LINK_SIZE);
6620 int end_op = *be;
6621 *be = OP_END;
6622 fixed_length = find_fixedlength(cc, re->options, TRUE, cd);
6623 *be = end_op;
6624 DPRINTF(("fixed length = %d\n", fixed_length));
6625 if (fixed_length < 0)
6626 {
6627 errorcode = (fixed_length == -2)? ERR36 : ERR25;
6628 break;
6629 }
6630 PUT(cc, 1, fixed_length);
6631 }
6632 cc += 1 + LINK_SIZE;
6633 }
6634 }
6635
6636 /* Failed to compile, or error while post-processing */
6637
6638 if (errorcode != 0)
6639 {
6640 (pcre_free)(re);
6641 PCRE_EARLY_ERROR_RETURN:
6642 *erroroffset = ptr - (const uschar *)pattern;
6643 PCRE_EARLY_ERROR_RETURN2:
6644 *errorptr = find_error_text(errorcode);
6645 if (errorcodeptr != NULL) *errorcodeptr = errorcode;
6646 return NULL;
6647 }
6648
6649 /* If the anchored option was not passed, set the flag if we can determine that
6650 the pattern is anchored by virtue of ^ characters or \A or anything else (such
6651 as starting with .* when DOTALL is set).
6652
6653 Otherwise, if we know what the first byte has to be, save it, because that
6654 speeds up unanchored matches no end. If not, see if we can set the
6655 PCRE_STARTLINE flag. This is helpful for multiline matches when all branches
6656 start with ^. and also when all branches start with .* for non-DOTALL matches.
6657 */
6658
6659 if