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

Contents of /code/trunk/pcre_compile.c

Parent Directory Parent Directory | Revision Log Revision Log


Revision 426 - (show annotations) (download)
Wed Aug 26 15:38:32 2009 UTC (5 years, 1 month ago) by ph10
File MIME type: text/plain
File size: 212760 byte(s)
Remove restrictions on pcre_exec() partial matching.

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 }
1104 }
1105 }
1106
1107 /* Past any initial parenthesis handling, scan for parentheses or vertical
1108 bars. */
1109
1110 for (; *ptr != 0; ptr++)
1111 {
1112 /* Skip over backslashed characters and also entire \Q...\E */
1113
1114 if (*ptr == CHAR_BACKSLASH)
1115 {
1116 if (*(++ptr) == 0) goto FAIL_EXIT;
1117 if (*ptr == CHAR_Q) for (;;)
1118 {
1119 while (*(++ptr) != 0 && *ptr != CHAR_BACKSLASH) {};
1120 if (*ptr == 0) goto FAIL_EXIT;
1121 if (*(++ptr) == CHAR_E) break;
1122 }
1123 continue;
1124 }
1125
1126 /* Skip over character classes; this logic must be similar to the way they
1127 are handled for real. If the first character is '^', skip it. Also, if the
1128 first few characters (either before or after ^) are \Q\E or \E we skip them
1129 too. This makes for compatibility with Perl. Note the use of STR macros to
1130 encode "Q\\E" so that it works in UTF-8 on EBCDIC platforms. */
1131
1132 if (*ptr == CHAR_LEFT_SQUARE_BRACKET)
1133 {
1134 BOOL negate_class = FALSE;
1135 for (;;)
1136 {
1137 int c = *(++ptr);
1138 if (c == CHAR_BACKSLASH)
1139 {
1140 if (ptr[1] == CHAR_E)
1141 ptr++;
1142 else if (strncmp((const char *)ptr+1,
1143 STR_Q STR_BACKSLASH STR_E, 3) == 0)
1144 ptr += 3;
1145 else
1146 break;
1147 }
1148 else if (!negate_class && c == CHAR_CIRCUMFLEX_ACCENT)
1149 negate_class = TRUE;
1150 else break;
1151 }
1152
1153 /* If the next character is ']', it is a data character that must be
1154 skipped, except in JavaScript compatibility mode. */
1155
1156 if (ptr[1] == CHAR_RIGHT_SQUARE_BRACKET &&
1157 (cd->external_options & PCRE_JAVASCRIPT_COMPAT) == 0)
1158 ptr++;
1159
1160 while (*(++ptr) != CHAR_RIGHT_SQUARE_BRACKET)
1161 {
1162 if (*ptr == 0) return -1;
1163 if (*ptr == CHAR_BACKSLASH)
1164 {
1165 if (*(++ptr) == 0) goto FAIL_EXIT;
1166 if (*ptr == CHAR_Q) for (;;)
1167 {
1168 while (*(++ptr) != 0 && *ptr != CHAR_BACKSLASH) {};
1169 if (*ptr == 0) goto FAIL_EXIT;
1170 if (*(++ptr) == CHAR_E) break;
1171 }
1172 continue;
1173 }
1174 }
1175 continue;
1176 }
1177
1178 /* Skip comments in /x mode */
1179
1180 if (xmode && *ptr == CHAR_NUMBER_SIGN)
1181 {
1182 while (*(++ptr) != 0 && *ptr != CHAR_NL) {};
1183 if (*ptr == 0) goto FAIL_EXIT;
1184 continue;
1185 }
1186
1187 /* Check for the special metacharacters */
1188
1189 if (*ptr == CHAR_LEFT_PARENTHESIS)
1190 {
1191 int rc = find_parens_sub(&ptr, cd, name, lorn, xmode, count);
1192 if (rc > 0) return rc;
1193 if (*ptr == 0) goto FAIL_EXIT;
1194 }
1195
1196 else if (*ptr == CHAR_RIGHT_PARENTHESIS)
1197 {
1198 if (dup_parens && *count < hwm_count) *count = hwm_count;
1199 *ptrptr = ptr;
1200 return -1;
1201 }
1202
1203 else if (*ptr == CHAR_VERTICAL_LINE && dup_parens)
1204 {
1205 if (*count > hwm_count) hwm_count = *count;
1206 *count = start_count;
1207 }
1208 }
1209
1210 FAIL_EXIT:
1211 *ptrptr = ptr;
1212 return -1;
1213 }
1214
1215
1216
1217
1218 /*************************************************
1219 * Find forward referenced subpattern *
1220 *************************************************/
1221
1222 /* This function scans along a pattern's text looking for capturing
1223 subpatterns, and counting them. If it finds a named pattern that matches the
1224 name it is given, it returns its number. Alternatively, if the name is NULL, it
1225 returns when it reaches a given numbered subpattern. This is used for forward
1226 references to subpatterns. We used to be able to start this scan from the
1227 current compiling point, using the current count value from cd->bracount, and
1228 do it all in a single loop, but the addition of the possibility of duplicate
1229 subpattern numbers means that we have to scan from the very start, in order to
1230 take account of such duplicates, and to use a recursive function to keep track
1231 of the different types of group.
1232
1233 Arguments:
1234 cd compile background data
1235 name name to seek, or NULL if seeking a numbered subpattern
1236 lorn name length, or subpattern number if name is NULL
1237 xmode TRUE if we are in /x mode
1238
1239 Returns: the number of the found subpattern, or -1 if not found
1240 */
1241
1242 static int
1243 find_parens(compile_data *cd, const uschar *name, int lorn, BOOL xmode)
1244 {
1245 uschar *ptr = (uschar *)cd->start_pattern;
1246 int count = 0;
1247 int rc;
1248
1249 /* If the pattern does not start with an opening parenthesis, the first call
1250 to find_parens_sub() will scan right to the end (if necessary). However, if it
1251 does start with a parenthesis, find_parens_sub() will return when it hits the
1252 matching closing parens. That is why we have to have a loop. */
1253
1254 for (;;)
1255 {
1256 rc = find_parens_sub(&ptr, cd, name, lorn, xmode, &count);
1257 if (rc > 0 || *ptr++ == 0) break;
1258 }
1259
1260 return rc;
1261 }
1262
1263
1264
1265
1266 /*************************************************
1267 * Find first significant op code *
1268 *************************************************/
1269
1270 /* This is called by several functions that scan a compiled expression looking
1271 for a fixed first character, or an anchoring op code etc. It skips over things
1272 that do not influence this. For some calls, a change of option is important.
1273 For some calls, it makes sense to skip negative forward and all backward
1274 assertions, and also the \b assertion; for others it does not.
1275
1276 Arguments:
1277 code pointer to the start of the group
1278 options pointer to external options
1279 optbit the option bit whose changing is significant, or
1280 zero if none are
1281 skipassert TRUE if certain assertions are to be skipped
1282
1283 Returns: pointer to the first significant opcode
1284 */
1285
1286 static const uschar*
1287 first_significant_code(const uschar *code, int *options, int optbit,
1288 BOOL skipassert)
1289 {
1290 for (;;)
1291 {
1292 switch ((int)*code)
1293 {
1294 case OP_OPT:
1295 if (optbit > 0 && ((int)code[1] & optbit) != (*options & optbit))
1296 *options = (int)code[1];
1297 code += 2;
1298 break;
1299
1300 case OP_ASSERT_NOT:
1301 case OP_ASSERTBACK:
1302 case OP_ASSERTBACK_NOT:
1303 if (!skipassert) return code;
1304 do code += GET(code, 1); while (*code == OP_ALT);
1305 code += _pcre_OP_lengths[*code];
1306 break;
1307
1308 case OP_WORD_BOUNDARY:
1309 case OP_NOT_WORD_BOUNDARY:
1310 if (!skipassert) return code;
1311 /* Fall through */
1312
1313 case OP_CALLOUT:
1314 case OP_CREF:
1315 case OP_RREF:
1316 case OP_DEF:
1317 code += _pcre_OP_lengths[*code];
1318 break;
1319
1320 default:
1321 return code;
1322 }
1323 }
1324 /* Control never reaches here */
1325 }
1326
1327
1328
1329
1330 /*************************************************
1331 * Find the fixed length of a pattern *
1332 *************************************************/
1333
1334 /* Scan a pattern and compute the fixed length of subject that will match it,
1335 if the length is fixed. This is needed for dealing with backward assertions.
1336 In UTF8 mode, the result is in characters rather than bytes.
1337
1338 Arguments:
1339 code points to the start of the pattern (the bracket)
1340 options the compiling options
1341
1342 Returns: the fixed length, or -1 if there is no fixed length,
1343 or -2 if \C was encountered
1344 */
1345
1346 static int
1347 find_fixedlength(uschar *code, int options)
1348 {
1349 int length = -1;
1350
1351 register int branchlength = 0;
1352 register uschar *cc = code + 1 + LINK_SIZE;
1353
1354 /* Scan along the opcodes for this branch. If we get to the end of the
1355 branch, check the length against that of the other branches. */
1356
1357 for (;;)
1358 {
1359 int d;
1360 register int op = *cc;
1361 switch (op)
1362 {
1363 case OP_CBRA:
1364 case OP_BRA:
1365 case OP_ONCE:
1366 case OP_COND:
1367 d = find_fixedlength(cc + ((op == OP_CBRA)? 2:0), options);
1368 if (d < 0) return d;
1369 branchlength += d;
1370 do cc += GET(cc, 1); while (*cc == OP_ALT);
1371 cc += 1 + LINK_SIZE;
1372 break;
1373
1374 /* Reached end of a branch; if it's a ket it is the end of a nested
1375 call. If it's ALT it is an alternation in a nested call. If it is
1376 END it's the end of the outer call. All can be handled by the same code. */
1377
1378 case OP_ALT:
1379 case OP_KET:
1380 case OP_KETRMAX:
1381 case OP_KETRMIN:
1382 case OP_END:
1383 if (length < 0) length = branchlength;
1384 else if (length != branchlength) return -1;
1385 if (*cc != OP_ALT) return length;
1386 cc += 1 + LINK_SIZE;
1387 branchlength = 0;
1388 break;
1389
1390 /* Skip over assertive subpatterns */
1391
1392 case OP_ASSERT:
1393 case OP_ASSERT_NOT:
1394 case OP_ASSERTBACK:
1395 case OP_ASSERTBACK_NOT:
1396 do cc += GET(cc, 1); while (*cc == OP_ALT);
1397 /* Fall through */
1398
1399 /* Skip over things that don't match chars */
1400
1401 case OP_REVERSE:
1402 case OP_CREF:
1403 case OP_RREF:
1404 case OP_DEF:
1405 case OP_OPT:
1406 case OP_CALLOUT:
1407 case OP_SOD:
1408 case OP_SOM:
1409 case OP_EOD:
1410 case OP_EODN:
1411 case OP_CIRC:
1412 case OP_DOLL:
1413 case OP_NOT_WORD_BOUNDARY:
1414 case OP_WORD_BOUNDARY:
1415 cc += _pcre_OP_lengths[*cc];
1416 break;
1417
1418 /* Handle literal characters */
1419
1420 case OP_CHAR:
1421 case OP_CHARNC:
1422 case OP_NOT:
1423 branchlength++;
1424 cc += 2;
1425 #ifdef SUPPORT_UTF8
1426 if ((options & PCRE_UTF8) != 0 && cc[-1] >= 0xc0)
1427 cc += _pcre_utf8_table4[cc[-1] & 0x3f];
1428 #endif
1429 break;
1430
1431 /* Handle exact repetitions. The count is already in characters, but we
1432 need to skip over a multibyte character in UTF8 mode. */
1433
1434 case OP_EXACT:
1435 branchlength += GET2(cc,1);
1436 cc += 4;
1437 #ifdef SUPPORT_UTF8
1438 if ((options & PCRE_UTF8) != 0 && cc[-1] >= 0xc0)
1439 cc += _pcre_utf8_table4[cc[-1] & 0x3f];
1440 #endif
1441 break;
1442
1443 case OP_TYPEEXACT:
1444 branchlength += GET2(cc,1);
1445 if (cc[3] == OP_PROP || cc[3] == OP_NOTPROP) cc += 2;
1446 cc += 4;
1447 break;
1448
1449 /* Handle single-char matchers */
1450
1451 case OP_PROP:
1452 case OP_NOTPROP:
1453 cc += 2;
1454 /* Fall through */
1455
1456 case OP_NOT_DIGIT:
1457 case OP_DIGIT:
1458 case OP_NOT_WHITESPACE:
1459 case OP_WHITESPACE:
1460 case OP_NOT_WORDCHAR:
1461 case OP_WORDCHAR:
1462 case OP_ANY:
1463 case OP_ALLANY:
1464 branchlength++;
1465 cc++;
1466 break;
1467
1468 /* The single-byte matcher isn't allowed */
1469
1470 case OP_ANYBYTE:
1471 return -2;
1472
1473 /* Check a class for variable quantification */
1474
1475 #ifdef SUPPORT_UTF8
1476 case OP_XCLASS:
1477 cc += GET(cc, 1) - 33;
1478 /* Fall through */
1479 #endif
1480
1481 case OP_CLASS:
1482 case OP_NCLASS:
1483 cc += 33;
1484
1485 switch (*cc)
1486 {
1487 case OP_CRSTAR:
1488 case OP_CRMINSTAR:
1489 case OP_CRQUERY:
1490 case OP_CRMINQUERY:
1491 return -1;
1492
1493 case OP_CRRANGE:
1494 case OP_CRMINRANGE:
1495 if (GET2(cc,1) != GET2(cc,3)) return -1;
1496 branchlength += GET2(cc,1);
1497 cc += 5;
1498 break;
1499
1500 default:
1501 branchlength++;
1502 }
1503 break;
1504
1505 /* Anything else is variable length */
1506
1507 default:
1508 return -1;
1509 }
1510 }
1511 /* Control never gets here */
1512 }
1513
1514
1515
1516
1517 /*************************************************
1518 * Scan compiled regex for numbered bracket *
1519 *************************************************/
1520
1521 /* This little function scans through a compiled pattern until it finds a
1522 capturing bracket with the given number.
1523
1524 Arguments:
1525 code points to start of expression
1526 utf8 TRUE in UTF-8 mode
1527 number the required bracket number
1528
1529 Returns: pointer to the opcode for the bracket, or NULL if not found
1530 */
1531
1532 static const uschar *
1533 find_bracket(const uschar *code, BOOL utf8, int number)
1534 {
1535 for (;;)
1536 {
1537 register int c = *code;
1538 if (c == OP_END) return NULL;
1539
1540 /* XCLASS is used for classes that cannot be represented just by a bit
1541 map. This includes negated single high-valued characters. The length in
1542 the table is zero; the actual length is stored in the compiled code. */
1543
1544 if (c == OP_XCLASS) code += GET(code, 1);
1545
1546 /* Handle capturing bracket */
1547
1548 else if (c == OP_CBRA)
1549 {
1550 int n = GET2(code, 1+LINK_SIZE);
1551 if (n == number) return (uschar *)code;
1552 code += _pcre_OP_lengths[c];
1553 }
1554
1555 /* Otherwise, we can get the item's length from the table, except that for
1556 repeated character types, we have to test for \p and \P, which have an extra
1557 two bytes of parameters. */
1558
1559 else
1560 {
1561 switch(c)
1562 {
1563 case OP_TYPESTAR:
1564 case OP_TYPEMINSTAR:
1565 case OP_TYPEPLUS:
1566 case OP_TYPEMINPLUS:
1567 case OP_TYPEQUERY:
1568 case OP_TYPEMINQUERY:
1569 case OP_TYPEPOSSTAR:
1570 case OP_TYPEPOSPLUS:
1571 case OP_TYPEPOSQUERY:
1572 if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
1573 break;
1574
1575 case OP_TYPEUPTO:
1576 case OP_TYPEMINUPTO:
1577 case OP_TYPEEXACT:
1578 case OP_TYPEPOSUPTO:
1579 if (code[3] == OP_PROP || code[3] == OP_NOTPROP) code += 2;
1580 break;
1581 }
1582
1583 /* Add in the fixed length from the table */
1584
1585 code += _pcre_OP_lengths[c];
1586
1587 /* In UTF-8 mode, opcodes that are followed by a character may be followed by
1588 a multi-byte character. The length in the table is a minimum, so we have to
1589 arrange to skip the extra bytes. */
1590
1591 #ifdef SUPPORT_UTF8
1592 if (utf8) switch(c)
1593 {
1594 case OP_CHAR:
1595 case OP_CHARNC:
1596 case OP_EXACT:
1597 case OP_UPTO:
1598 case OP_MINUPTO:
1599 case OP_POSUPTO:
1600 case OP_STAR:
1601 case OP_MINSTAR:
1602 case OP_POSSTAR:
1603 case OP_PLUS:
1604 case OP_MINPLUS:
1605 case OP_POSPLUS:
1606 case OP_QUERY:
1607 case OP_MINQUERY:
1608 case OP_POSQUERY:
1609 if (code[-1] >= 0xc0) code += _pcre_utf8_table4[code[-1] & 0x3f];
1610 break;
1611 }
1612 #else
1613 (void)(utf8); /* Keep compiler happy by referencing function argument */
1614 #endif
1615 }
1616 }
1617 }
1618
1619
1620
1621 /*************************************************
1622 * Scan compiled regex for recursion reference *
1623 *************************************************/
1624
1625 /* This little function scans through a compiled pattern until it finds an
1626 instance of OP_RECURSE.
1627
1628 Arguments:
1629 code points to start of expression
1630 utf8 TRUE in UTF-8 mode
1631
1632 Returns: pointer to the opcode for OP_RECURSE, or NULL if not found
1633 */
1634
1635 static const uschar *
1636 find_recurse(const uschar *code, BOOL utf8)
1637 {
1638 for (;;)
1639 {
1640 register int c = *code;
1641 if (c == OP_END) return NULL;
1642 if (c == OP_RECURSE) return code;
1643
1644 /* XCLASS is used for classes that cannot be represented just by a bit
1645 map. This includes negated single high-valued characters. The length in
1646 the table is zero; the actual length is stored in the compiled code. */
1647
1648 if (c == OP_XCLASS) code += GET(code, 1);
1649
1650 /* Otherwise, we can get the item's length from the table, except that for
1651 repeated character types, we have to test for \p and \P, which have an extra
1652 two bytes of parameters. */
1653
1654 else
1655 {
1656 switch(c)
1657 {
1658 case OP_TYPESTAR:
1659 case OP_TYPEMINSTAR:
1660 case OP_TYPEPLUS:
1661 case OP_TYPEMINPLUS:
1662 case OP_TYPEQUERY:
1663 case OP_TYPEMINQUERY:
1664 case OP_TYPEPOSSTAR:
1665 case OP_TYPEPOSPLUS:
1666 case OP_TYPEPOSQUERY:
1667 if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
1668 break;
1669
1670 case OP_TYPEPOSUPTO:
1671 case OP_TYPEUPTO:
1672 case OP_TYPEMINUPTO:
1673 case OP_TYPEEXACT:
1674 if (code[3] == OP_PROP || code[3] == OP_NOTPROP) code += 2;
1675 break;
1676 }
1677
1678 /* Add in the fixed length from the table */
1679
1680 code += _pcre_OP_lengths[c];
1681
1682 /* In UTF-8 mode, opcodes that are followed by a character may be followed
1683 by a multi-byte character. The length in the table is a minimum, so we have
1684 to arrange to skip the extra bytes. */
1685
1686 #ifdef SUPPORT_UTF8
1687 if (utf8) switch(c)
1688 {
1689 case OP_CHAR:
1690 case OP_CHARNC:
1691 case OP_EXACT:
1692 case OP_UPTO:
1693 case OP_MINUPTO:
1694 case OP_POSUPTO:
1695 case OP_STAR:
1696 case OP_MINSTAR:
1697 case OP_POSSTAR:
1698 case OP_PLUS:
1699 case OP_MINPLUS:
1700 case OP_POSPLUS:
1701 case OP_QUERY:
1702 case OP_MINQUERY:
1703 case OP_POSQUERY:
1704 if (code[-1] >= 0xc0) code += _pcre_utf8_table4[code[-1] & 0x3f];
1705 break;
1706 }
1707 #else
1708 (void)(utf8); /* Keep compiler happy by referencing function argument */
1709 #endif
1710 }
1711 }
1712 }
1713
1714
1715
1716 /*************************************************
1717 * Scan compiled branch for non-emptiness *
1718 *************************************************/
1719
1720 /* This function scans through a branch of a compiled pattern to see whether it
1721 can match the empty string or not. It is called from could_be_empty()
1722 below and from compile_branch() when checking for an unlimited repeat of a
1723 group that can match nothing. Note that first_significant_code() skips over
1724 backward and negative forward assertions when its final argument is TRUE. If we
1725 hit an unclosed bracket, we return "empty" - this means we've struck an inner
1726 bracket whose current branch will already have been scanned.
1727
1728 Arguments:
1729 code points to start of search
1730 endcode points to where to stop
1731 utf8 TRUE if in UTF8 mode
1732
1733 Returns: TRUE if what is matched could be empty
1734 */
1735
1736 static BOOL
1737 could_be_empty_branch(const uschar *code, const uschar *endcode, BOOL utf8)
1738 {
1739 register int c;
1740 for (code = first_significant_code(code + _pcre_OP_lengths[*code], NULL, 0, TRUE);
1741 code < endcode;
1742 code = first_significant_code(code + _pcre_OP_lengths[c], NULL, 0, TRUE))
1743 {
1744 const uschar *ccode;
1745
1746 c = *code;
1747
1748 /* Skip over forward assertions; the other assertions are skipped by
1749 first_significant_code() with a TRUE final argument. */
1750
1751 if (c == OP_ASSERT)
1752 {
1753 do code += GET(code, 1); while (*code == OP_ALT);
1754 c = *code;
1755 continue;
1756 }
1757
1758 /* Groups with zero repeats can of course be empty; skip them. */
1759
1760 if (c == OP_BRAZERO || c == OP_BRAMINZERO || c == OP_SKIPZERO)
1761 {
1762 code += _pcre_OP_lengths[c];
1763 do code += GET(code, 1); while (*code == OP_ALT);
1764 c = *code;
1765 continue;
1766 }
1767
1768 /* For other groups, scan the branches. */
1769
1770 if (c == OP_BRA || c == OP_CBRA || c == OP_ONCE || c == OP_COND)
1771 {
1772 BOOL empty_branch;
1773 if (GET(code, 1) == 0) return TRUE; /* Hit unclosed bracket */
1774
1775 /* If a conditional group has only one branch, there is a second, implied,
1776 empty branch, so just skip over the conditional, because it could be empty.
1777 Otherwise, scan the individual branches of the group. */
1778
1779 if (c == OP_COND && code[GET(code, 1)] != OP_ALT)
1780 code += GET(code, 1);
1781 else
1782 {
1783 empty_branch = FALSE;
1784 do
1785 {
1786 if (!empty_branch && could_be_empty_branch(code, endcode, utf8))
1787 empty_branch = TRUE;
1788 code += GET(code, 1);
1789 }
1790 while (*code == OP_ALT);
1791 if (!empty_branch) return FALSE; /* All branches are non-empty */
1792 }
1793
1794 c = *code;
1795 continue;
1796 }
1797
1798 /* Handle the other opcodes */
1799
1800 switch (c)
1801 {
1802 /* Check for quantifiers after a class. XCLASS is used for classes that
1803 cannot be represented just by a bit map. This includes negated single
1804 high-valued characters. The length in _pcre_OP_lengths[] is zero; the
1805 actual length is stored in the compiled code, so we must update "code"
1806 here. */
1807
1808 #ifdef SUPPORT_UTF8
1809 case OP_XCLASS:
1810 ccode = code += GET(code, 1);
1811 goto CHECK_CLASS_REPEAT;
1812 #endif
1813
1814 case OP_CLASS:
1815 case OP_NCLASS:
1816 ccode = code + 33;
1817
1818 #ifdef SUPPORT_UTF8
1819 CHECK_CLASS_REPEAT:
1820 #endif
1821
1822 switch (*ccode)
1823 {
1824 case OP_CRSTAR: /* These could be empty; continue */
1825 case OP_CRMINSTAR:
1826 case OP_CRQUERY:
1827 case OP_CRMINQUERY:
1828 break;
1829
1830 default: /* Non-repeat => class must match */
1831 case OP_CRPLUS: /* These repeats aren't empty */
1832 case OP_CRMINPLUS:
1833 return FALSE;
1834
1835 case OP_CRRANGE:
1836 case OP_CRMINRANGE:
1837 if (GET2(ccode, 1) > 0) return FALSE; /* Minimum > 0 */
1838 break;
1839 }
1840 break;
1841
1842 /* Opcodes that must match a character */
1843
1844 case OP_PROP:
1845 case OP_NOTPROP:
1846 case OP_EXTUNI:
1847 case OP_NOT_DIGIT:
1848 case OP_DIGIT:
1849 case OP_NOT_WHITESPACE:
1850 case OP_WHITESPACE:
1851 case OP_NOT_WORDCHAR:
1852 case OP_WORDCHAR:
1853 case OP_ANY:
1854 case OP_ALLANY:
1855 case OP_ANYBYTE:
1856 case OP_CHAR:
1857 case OP_CHARNC:
1858 case OP_NOT:
1859 case OP_PLUS:
1860 case OP_MINPLUS:
1861 case OP_POSPLUS:
1862 case OP_EXACT:
1863 case OP_NOTPLUS:
1864 case OP_NOTMINPLUS:
1865 case OP_NOTPOSPLUS:
1866 case OP_NOTEXACT:
1867 case OP_TYPEPLUS:
1868 case OP_TYPEMINPLUS:
1869 case OP_TYPEPOSPLUS:
1870 case OP_TYPEEXACT:
1871 return FALSE;
1872
1873 /* These are going to continue, as they may be empty, but we have to
1874 fudge the length for the \p and \P cases. */
1875
1876 case OP_TYPESTAR:
1877 case OP_TYPEMINSTAR:
1878 case OP_TYPEPOSSTAR:
1879 case OP_TYPEQUERY:
1880 case OP_TYPEMINQUERY:
1881 case OP_TYPEPOSQUERY:
1882 if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
1883 break;
1884
1885 /* Same for these */
1886
1887 case OP_TYPEUPTO:
1888 case OP_TYPEMINUPTO:
1889 case OP_TYPEPOSUPTO:
1890 if (code[3] == OP_PROP || code[3] == OP_NOTPROP) code += 2;
1891 break;
1892
1893 /* End of branch */
1894
1895 case OP_KET:
1896 case OP_KETRMAX:
1897 case OP_KETRMIN:
1898 case OP_ALT:
1899 return TRUE;
1900
1901 /* In UTF-8 mode, STAR, MINSTAR, POSSTAR, QUERY, MINQUERY, POSQUERY, UPTO,
1902 MINUPTO, and POSUPTO may be followed by a multibyte character */
1903
1904 #ifdef SUPPORT_UTF8
1905 case OP_STAR:
1906 case OP_MINSTAR:
1907 case OP_POSSTAR:
1908 case OP_QUERY:
1909 case OP_MINQUERY:
1910 case OP_POSQUERY:
1911 if (utf8 && code[1] >= 0xc0) code += _pcre_utf8_table4[code[1] & 0x3f];
1912 break;
1913
1914 case OP_UPTO:
1915 case OP_MINUPTO:
1916 case OP_POSUPTO:
1917 if (utf8 && code[3] >= 0xc0) code += _pcre_utf8_table4[code[3] & 0x3f];
1918 break;
1919 #endif
1920 }
1921 }
1922
1923 return TRUE;
1924 }
1925
1926
1927
1928 /*************************************************
1929 * Scan compiled regex for non-emptiness *
1930 *************************************************/
1931
1932 /* This function is called to check for left recursive calls. We want to check
1933 the current branch of the current pattern to see if it could match the empty
1934 string. If it could, we must look outwards for branches at other levels,
1935 stopping when we pass beyond the bracket which is the subject of the recursion.
1936
1937 Arguments:
1938 code points to start of the recursion
1939 endcode points to where to stop (current RECURSE item)
1940 bcptr points to the chain of current (unclosed) branch starts
1941 utf8 TRUE if in UTF-8 mode
1942
1943 Returns: TRUE if what is matched could be empty
1944 */
1945
1946 static BOOL
1947 could_be_empty(const uschar *code, const uschar *endcode, branch_chain *bcptr,
1948 BOOL utf8)
1949 {
1950 while (bcptr != NULL && bcptr->current >= code)
1951 {
1952 if (!could_be_empty_branch(bcptr->current, endcode, utf8)) return FALSE;
1953 bcptr = bcptr->outer;
1954 }
1955 return TRUE;
1956 }
1957
1958
1959
1960 /*************************************************
1961 * Check for POSIX class syntax *
1962 *************************************************/
1963
1964 /* This function is called when the sequence "[:" or "[." or "[=" is
1965 encountered in a character class. It checks whether this is followed by a
1966 sequence of characters terminated by a matching ":]" or ".]" or "=]". If we
1967 reach an unescaped ']' without the special preceding character, return FALSE.
1968
1969 Originally, this function only recognized a sequence of letters between the
1970 terminators, but it seems that Perl recognizes any sequence of characters,
1971 though of course unknown POSIX names are subsequently rejected. Perl gives an
1972 "Unknown POSIX class" error for [:f\oo:] for example, where previously PCRE
1973 didn't consider this to be a POSIX class. Likewise for [:1234:].
1974
1975 The problem in trying to be exactly like Perl is in the handling of escapes. We
1976 have to be sure that [abc[:x\]pqr] is *not* treated as containing a POSIX
1977 class, but [abc[:x\]pqr:]] is (so that an error can be generated). The code
1978 below handles the special case of \], but does not try to do any other escape
1979 processing. This makes it different from Perl for cases such as [:l\ower:]
1980 where Perl recognizes it as the POSIX class "lower" but PCRE does not recognize
1981 "l\ower". This is a lesser evil that not diagnosing bad classes when Perl does,
1982 I think.
1983
1984 Arguments:
1985 ptr pointer to the initial [
1986 endptr where to return the end pointer
1987
1988 Returns: TRUE or FALSE
1989 */
1990
1991 static BOOL
1992 check_posix_syntax(const uschar *ptr, const uschar **endptr)
1993 {
1994 int terminator; /* Don't combine these lines; the Solaris cc */
1995 terminator = *(++ptr); /* compiler warns about "non-constant" initializer. */
1996 for (++ptr; *ptr != 0; ptr++)
1997 {
1998 if (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET) ptr++; else
1999 {
2000 if (*ptr == CHAR_RIGHT_SQUARE_BRACKET) return FALSE;
2001 if (*ptr == terminator && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET)
2002 {
2003 *endptr = ptr;
2004 return TRUE;
2005 }
2006 }
2007 }
2008 return FALSE;
2009 }
2010
2011
2012
2013
2014 /*************************************************
2015 * Check POSIX class name *
2016 *************************************************/
2017
2018 /* This function is called to check the name given in a POSIX-style class entry
2019 such as [:alnum:].
2020
2021 Arguments:
2022 ptr points to the first letter
2023 len the length of the name
2024
2025 Returns: a value representing the name, or -1 if unknown
2026 */
2027
2028 static int
2029 check_posix_name(const uschar *ptr, int len)
2030 {
2031 const char *pn = posix_names;
2032 register int yield = 0;
2033 while (posix_name_lengths[yield] != 0)
2034 {
2035 if (len == posix_name_lengths[yield] &&
2036 strncmp((const char *)ptr, pn, len) == 0) return yield;
2037 pn += posix_name_lengths[yield] + 1;
2038 yield++;
2039 }
2040 return -1;
2041 }
2042
2043
2044 /*************************************************
2045 * Adjust OP_RECURSE items in repeated group *
2046 *************************************************/
2047
2048 /* OP_RECURSE items contain an offset from the start of the regex to the group
2049 that is referenced. This means that groups can be replicated for fixed
2050 repetition simply by copying (because the recursion is allowed to refer to
2051 earlier groups that are outside the current group). However, when a group is
2052 optional (i.e. the minimum quantifier is zero), OP_BRAZERO or OP_SKIPZERO is
2053 inserted before it, after it has been compiled. This means that any OP_RECURSE
2054 items within it that refer to the group itself or any contained groups have to
2055 have their offsets adjusted. That one of the jobs of this function. Before it
2056 is called, the partially compiled regex must be temporarily terminated with
2057 OP_END.
2058
2059 This function has been extended with the possibility of forward references for
2060 recursions and subroutine calls. It must also check the list of such references
2061 for the group we are dealing with. If it finds that one of the recursions in
2062 the current group is on this list, it adjusts the offset in the list, not the
2063 value in the reference (which is a group number).
2064
2065 Arguments:
2066 group points to the start of the group
2067 adjust the amount by which the group is to be moved
2068 utf8 TRUE in UTF-8 mode
2069 cd contains pointers to tables etc.
2070 save_hwm the hwm forward reference pointer at the start of the group
2071
2072 Returns: nothing
2073 */
2074
2075 static void
2076 adjust_recurse(uschar *group, int adjust, BOOL utf8, compile_data *cd,
2077 uschar *save_hwm)
2078 {
2079 uschar *ptr = group;
2080
2081 while ((ptr = (uschar *)find_recurse(ptr, utf8)) != NULL)
2082 {
2083 int offset;
2084 uschar *hc;
2085
2086 /* See if this recursion is on the forward reference list. If so, adjust the
2087 reference. */
2088
2089 for (hc = save_hwm; hc < cd->hwm; hc += LINK_SIZE)
2090 {
2091 offset = GET(hc, 0);
2092 if (cd->start_code + offset == ptr + 1)
2093 {
2094 PUT(hc, 0, offset + adjust);
2095 break;
2096 }
2097 }
2098
2099 /* Otherwise, adjust the recursion offset if it's after the start of this
2100 group. */
2101
2102 if (hc >= cd->hwm)
2103 {
2104 offset = GET(ptr, 1);
2105 if (cd->start_code + offset >= group) PUT(ptr, 1, offset + adjust);
2106 }
2107
2108 ptr += 1 + LINK_SIZE;
2109 }
2110 }
2111
2112
2113
2114 /*************************************************
2115 * Insert an automatic callout point *
2116 *************************************************/
2117
2118 /* This function is called when the PCRE_AUTO_CALLOUT option is set, to insert
2119 callout points before each pattern item.
2120
2121 Arguments:
2122 code current code pointer
2123 ptr current pattern pointer
2124 cd pointers to tables etc
2125
2126 Returns: new code pointer
2127 */
2128
2129 static uschar *
2130 auto_callout(uschar *code, const uschar *ptr, compile_data *cd)
2131 {
2132 *code++ = OP_CALLOUT;
2133 *code++ = 255;
2134 PUT(code, 0, ptr - cd->start_pattern); /* Pattern offset */
2135 PUT(code, LINK_SIZE, 0); /* Default length */
2136 return code + 2*LINK_SIZE;
2137 }
2138
2139
2140
2141 /*************************************************
2142 * Complete a callout item *
2143 *************************************************/
2144
2145 /* A callout item contains the length of the next item in the pattern, which
2146 we can't fill in till after we have reached the relevant point. This is used
2147 for both automatic and manual callouts.
2148
2149 Arguments:
2150 previous_callout points to previous callout item
2151 ptr current pattern pointer
2152 cd pointers to tables etc
2153
2154 Returns: nothing
2155 */
2156
2157 static void
2158 complete_callout(uschar *previous_callout, const uschar *ptr, compile_data *cd)
2159 {
2160 int length = ptr - cd->start_pattern - GET(previous_callout, 2);
2161 PUT(previous_callout, 2 + LINK_SIZE, length);
2162 }
2163
2164
2165
2166 #ifdef SUPPORT_UCP
2167 /*************************************************
2168 * Get othercase range *
2169 *************************************************/
2170
2171 /* This function is passed the start and end of a class range, in UTF-8 mode
2172 with UCP support. It searches up the characters, looking for internal ranges of
2173 characters in the "other" case. Each call returns the next one, updating the
2174 start address.
2175
2176 Arguments:
2177 cptr points to starting character value; updated
2178 d end value
2179 ocptr where to put start of othercase range
2180 odptr where to put end of othercase range
2181
2182 Yield: TRUE when range returned; FALSE when no more
2183 */
2184
2185 static BOOL
2186 get_othercase_range(unsigned int *cptr, unsigned int d, unsigned int *ocptr,
2187 unsigned int *odptr)
2188 {
2189 unsigned int c, othercase, next;
2190
2191 for (c = *cptr; c <= d; c++)
2192 { if ((othercase = UCD_OTHERCASE(c)) != c) break; }
2193
2194 if (c > d) return FALSE;
2195
2196 *ocptr = othercase;
2197 next = othercase + 1;
2198
2199 for (++c; c <= d; c++)
2200 {
2201 if (UCD_OTHERCASE(c) != next) break;
2202 next++;
2203 }
2204
2205 *odptr = next - 1;
2206 *cptr = c;
2207
2208 return TRUE;
2209 }
2210 #endif /* SUPPORT_UCP */
2211
2212
2213
2214 /*************************************************
2215 * Check if auto-possessifying is possible *
2216 *************************************************/
2217
2218 /* This function is called for unlimited repeats of certain items, to see
2219 whether the next thing could possibly match the repeated item. If not, it makes
2220 sense to automatically possessify the repeated item.
2221
2222 Arguments:
2223 op_code the repeated op code
2224 this data for this item, depends on the opcode
2225 utf8 TRUE in UTF-8 mode
2226 utf8_char used for utf8 character bytes, NULL if not relevant
2227 ptr next character in pattern
2228 options options bits
2229 cd contains pointers to tables etc.
2230
2231 Returns: TRUE if possessifying is wanted
2232 */
2233
2234 static BOOL
2235 check_auto_possessive(int op_code, int item, BOOL utf8, uschar *utf8_char,
2236 const uschar *ptr, int options, compile_data *cd)
2237 {
2238 int next;
2239
2240 /* Skip whitespace and comments in extended mode */
2241
2242 if ((options & PCRE_EXTENDED) != 0)
2243 {
2244 for (;;)
2245 {
2246 while ((cd->ctypes[*ptr] & ctype_space) != 0) ptr++;
2247 if (*ptr == CHAR_NUMBER_SIGN)
2248 {
2249 while (*(++ptr) != 0)
2250 if (IS_NEWLINE(ptr)) { ptr += cd->nllen; break; }
2251 }
2252 else break;
2253 }
2254 }
2255
2256 /* If the next item is one that we can handle, get its value. A non-negative
2257 value is a character, a negative value is an escape value. */
2258
2259 if (*ptr == CHAR_BACKSLASH)
2260 {
2261 int temperrorcode = 0;
2262 next = check_escape(&ptr, &temperrorcode, cd->bracount, options, FALSE);
2263 if (temperrorcode != 0) return FALSE;
2264 ptr++; /* Point after the escape sequence */
2265 }
2266
2267 else if ((cd->ctypes[*ptr] & ctype_meta) == 0)
2268 {
2269 #ifdef SUPPORT_UTF8
2270 if (utf8) { GETCHARINC(next, ptr); } else
2271 #endif
2272 next = *ptr++;
2273 }
2274
2275 else return FALSE;
2276
2277 /* Skip whitespace and comments in extended mode */
2278
2279 if ((options & PCRE_EXTENDED) != 0)
2280 {
2281 for (;;)
2282 {
2283 while ((cd->ctypes[*ptr] & ctype_space) != 0) ptr++;
2284 if (*ptr == CHAR_NUMBER_SIGN)
2285 {
2286 while (*(++ptr) != 0)
2287 if (IS_NEWLINE(ptr)) { ptr += cd->nllen; break; }
2288 }
2289 else break;
2290 }
2291 }
2292
2293 /* If the next thing is itself optional, we have to give up. */
2294
2295 if (*ptr == CHAR_ASTERISK || *ptr == CHAR_QUESTION_MARK ||
2296 strncmp((char *)ptr, STR_LEFT_CURLY_BRACKET STR_0 STR_COMMA, 3) == 0)
2297 return FALSE;
2298
2299 /* Now compare the next item with the previous opcode. If the previous is a
2300 positive single character match, "item" either contains the character or, if
2301 "item" is greater than 127 in utf8 mode, the character's bytes are in
2302 utf8_char. */
2303
2304
2305 /* Handle cases when the next item is a character. */
2306
2307 if (next >= 0) switch(op_code)
2308 {
2309 case OP_CHAR:
2310 #ifdef SUPPORT_UTF8
2311 if (utf8 && item > 127) { GETCHAR(item, utf8_char); }
2312 #else
2313 (void)(utf8_char); /* Keep compiler happy by referencing function argument */
2314 #endif
2315 return item != next;
2316
2317 /* For CHARNC (caseless character) we must check the other case. If we have
2318 Unicode property support, we can use it to test the other case of
2319 high-valued characters. */
2320
2321 case OP_CHARNC:
2322 #ifdef SUPPORT_UTF8
2323 if (utf8 && item > 127) { GETCHAR(item, utf8_char); }
2324 #endif
2325 if (item == next) return FALSE;
2326 #ifdef SUPPORT_UTF8
2327 if (utf8)
2328 {
2329 unsigned int othercase;
2330 if (next < 128) othercase = cd->fcc[next]; else
2331 #ifdef SUPPORT_UCP
2332 othercase = UCD_OTHERCASE((unsigned int)next);
2333 #else
2334 othercase = NOTACHAR;
2335 #endif
2336 return (unsigned int)item != othercase;
2337 }
2338 else
2339 #endif /* SUPPORT_UTF8 */
2340 return (item != cd->fcc[next]); /* Non-UTF-8 mode */
2341
2342 /* For OP_NOT, "item" must be a single-byte character. */
2343
2344 case OP_NOT:
2345 if (item == next) return TRUE;
2346 if ((options & PCRE_CASELESS) == 0) return FALSE;
2347 #ifdef SUPPORT_UTF8
2348 if (utf8)
2349 {
2350 unsigned int othercase;
2351 if (next < 128) othercase = cd->fcc[next]; else
2352 #ifdef SUPPORT_UCP
2353 othercase = UCD_OTHERCASE(next);
2354 #else
2355 othercase = NOTACHAR;
2356 #endif
2357 return (unsigned int)item == othercase;
2358 }
2359 else
2360 #endif /* SUPPORT_UTF8 */
2361 return (item == cd->fcc[next]); /* Non-UTF-8 mode */
2362
2363 case OP_DIGIT:
2364 return next > 127 || (cd->ctypes[next] & ctype_digit) == 0;
2365
2366 case OP_NOT_DIGIT:
2367 return next <= 127 && (cd->ctypes[next] & ctype_digit) != 0;
2368
2369 case OP_WHITESPACE:
2370 return next > 127 || (cd->ctypes[next] & ctype_space) == 0;
2371
2372 case OP_NOT_WHITESPACE:
2373 return next <= 127 && (cd->ctypes[next] & ctype_space) != 0;
2374
2375 case OP_WORDCHAR:
2376 return next > 127 || (cd->ctypes[next] & ctype_word) == 0;
2377
2378 case OP_NOT_WORDCHAR:
2379 return next <= 127 && (cd->ctypes[next] & ctype_word) != 0;
2380
2381 case OP_HSPACE:
2382 case OP_NOT_HSPACE:
2383 switch(next)
2384 {
2385 case 0x09:
2386 case 0x20:
2387 case 0xa0:
2388 case 0x1680:
2389 case 0x180e:
2390 case 0x2000:
2391 case 0x2001:
2392 case 0x2002:
2393 case 0x2003:
2394 case 0x2004:
2395 case 0x2005:
2396 case 0x2006:
2397 case 0x2007:
2398 case 0x2008:
2399 case 0x2009:
2400 case 0x200A:
2401 case 0x202f:
2402 case 0x205f:
2403 case 0x3000:
2404 return op_code != OP_HSPACE;
2405 default:
2406 return op_code == OP_HSPACE;
2407 }
2408
2409 case OP_VSPACE:
2410 case OP_NOT_VSPACE:
2411 switch(next)
2412 {
2413 case 0x0a:
2414 case 0x0b:
2415 case 0x0c:
2416 case 0x0d:
2417 case 0x85:
2418 case 0x2028:
2419 case 0x2029:
2420 return op_code != OP_VSPACE;
2421 default:
2422 return op_code == OP_VSPACE;
2423 }
2424
2425 default:
2426 return FALSE;
2427 }
2428
2429
2430 /* Handle the case when the next item is \d, \s, etc. */
2431
2432 switch(op_code)
2433 {
2434 case OP_CHAR:
2435 case OP_CHARNC:
2436 #ifdef SUPPORT_UTF8
2437 if (utf8 && item > 127) { GETCHAR(item, utf8_char); }
2438 #endif
2439 switch(-next)
2440 {
2441 case ESC_d:
2442 return item > 127 || (cd->ctypes[item] & ctype_digit) == 0;
2443
2444 case ESC_D:
2445 return item <= 127 && (cd->ctypes[item] & ctype_digit) != 0;
2446
2447 case ESC_s:
2448 return item > 127 || (cd->ctypes[item] & ctype_space) == 0;
2449
2450 case ESC_S:
2451 return item <= 127 && (cd->ctypes[item] & ctype_space) != 0;
2452
2453 case ESC_w:
2454 return item > 127 || (cd->ctypes[item] & ctype_word) == 0;
2455
2456 case ESC_W:
2457 return item <= 127 && (cd->ctypes[item] & ctype_word) != 0;
2458
2459 case ESC_h:
2460 case ESC_H:
2461 switch(item)
2462 {
2463 case 0x09:
2464 case 0x20:
2465 case 0xa0:
2466 case 0x1680:
2467 case 0x180e:
2468 case 0x2000:
2469 case 0x2001:
2470 case 0x2002:
2471 case 0x2003:
2472 case 0x2004:
2473 case 0x2005:
2474 case 0x2006:
2475 case 0x2007:
2476 case 0x2008:
2477 case 0x2009:
2478 case 0x200A:
2479 case 0x202f:
2480 case 0x205f:
2481 case 0x3000:
2482 return -next != ESC_h;
2483 default:
2484 return -next == ESC_h;
2485 }
2486
2487 case ESC_v:
2488 case ESC_V:
2489 switch(item)
2490 {
2491 case 0x0a:
2492 case 0x0b:
2493 case 0x0c:
2494 case 0x0d:
2495 case 0x85:
2496 case 0x2028:
2497 case 0x2029:
2498 return -next != ESC_v;
2499 default:
2500 return -next == ESC_v;
2501 }
2502
2503 default:
2504 return FALSE;
2505 }
2506
2507 case OP_DIGIT:
2508 return next == -ESC_D || next == -ESC_s || next == -ESC_W ||
2509 next == -ESC_h || next == -ESC_v;
2510
2511 case OP_NOT_DIGIT:
2512 return next == -ESC_d;
2513
2514 case OP_WHITESPACE:
2515 return next == -ESC_S || next == -ESC_d || next == -ESC_w;
2516
2517 case OP_NOT_WHITESPACE:
2518 return next == -ESC_s || next == -ESC_h || next == -ESC_v;
2519
2520 case OP_HSPACE:
2521 return next == -ESC_S || next == -ESC_H || next == -ESC_d || next == -ESC_w;
2522
2523 case OP_NOT_HSPACE:
2524 return next == -ESC_h;
2525
2526 /* Can't have \S in here because VT matches \S (Perl anomaly) */
2527 case OP_VSPACE:
2528 return next == -ESC_V || next == -ESC_d || next == -ESC_w;
2529
2530 case OP_NOT_VSPACE:
2531 return next == -ESC_v;
2532
2533 case OP_WORDCHAR:
2534 return next == -ESC_W || next == -ESC_s || next == -ESC_h || next == -ESC_v;
2535
2536 case OP_NOT_WORDCHAR:
2537 return next == -ESC_w || next == -ESC_d;
2538
2539 default:
2540 return FALSE;
2541 }
2542
2543 /* Control does not reach here */
2544 }
2545
2546
2547
2548 /*************************************************
2549 * Compile one branch *
2550 *************************************************/
2551
2552 /* Scan the pattern, compiling it into the a vector. If the options are
2553 changed during the branch, the pointer is used to change the external options
2554 bits. This function is used during the pre-compile phase when we are trying
2555 to find out the amount of memory needed, as well as during the real compile
2556 phase. The value of lengthptr distinguishes the two phases.
2557
2558 Arguments:
2559 optionsptr pointer to the option bits
2560 codeptr points to the pointer to the current code point
2561 ptrptr points to the current pattern pointer
2562 errorcodeptr points to error code variable
2563 firstbyteptr set to initial literal character, or < 0 (REQ_UNSET, REQ_NONE)
2564 reqbyteptr set to the last literal character required, else < 0
2565 bcptr points to current branch chain
2566 cd contains pointers to tables etc.
2567 lengthptr NULL during the real compile phase
2568 points to length accumulator during pre-compile phase
2569
2570 Returns: TRUE on success
2571 FALSE, with *errorcodeptr set non-zero on error
2572 */
2573
2574 static BOOL
2575 compile_branch(int *optionsptr, uschar **codeptr, const uschar **ptrptr,
2576 int *errorcodeptr, int *firstbyteptr, int *reqbyteptr, branch_chain *bcptr,
2577 compile_data *cd, int *lengthptr)
2578 {
2579 int repeat_type, op_type;
2580 int repeat_min = 0, repeat_max = 0; /* To please picky compilers */
2581 int bravalue = 0;
2582 int greedy_default, greedy_non_default;
2583 int firstbyte, reqbyte;
2584 int zeroreqbyte, zerofirstbyte;
2585 int req_caseopt, reqvary, tempreqvary;
2586 int options = *optionsptr;
2587 int after_manual_callout = 0;
2588 int length_prevgroup = 0;
2589 register int c;
2590 register uschar *code = *codeptr;
2591 uschar *last_code = code;
2592 uschar *orig_code = code;
2593 uschar *tempcode;
2594 BOOL inescq = FALSE;
2595 BOOL groupsetfirstbyte = FALSE;
2596 const uschar *ptr = *ptrptr;
2597 const uschar *tempptr;
2598 uschar *previous = NULL;
2599 uschar *previous_callout = NULL;
2600 uschar *save_hwm = NULL;
2601 uschar classbits[32];
2602
2603 #ifdef SUPPORT_UTF8
2604 BOOL class_utf8;
2605 BOOL utf8 = (options & PCRE_UTF8) != 0;
2606 uschar *class_utf8data;
2607 uschar *class_utf8data_base;
2608 uschar utf8_char[6];
2609 #else
2610 BOOL utf8 = FALSE;
2611 uschar *utf8_char = NULL;
2612 #endif
2613
2614 #ifdef DEBUG
2615 if (lengthptr != NULL) DPRINTF((">> start branch\n"));
2616 #endif
2617
2618 /* Set up the default and non-default settings for greediness */
2619
2620 greedy_default = ((options & PCRE_UNGREEDY) != 0);
2621 greedy_non_default = greedy_default ^ 1;
2622
2623 /* Initialize no first byte, no required byte. REQ_UNSET means "no char
2624 matching encountered yet". It gets changed to REQ_NONE if we hit something that
2625 matches a non-fixed char first char; reqbyte just remains unset if we never
2626 find one.
2627
2628 When we hit a repeat whose minimum is zero, we may have to adjust these values
2629 to take the zero repeat into account. This is implemented by setting them to
2630 zerofirstbyte and zeroreqbyte when such a repeat is encountered. The individual
2631 item types that can be repeated set these backoff variables appropriately. */
2632
2633 firstbyte = reqbyte = zerofirstbyte = zeroreqbyte = REQ_UNSET;
2634
2635 /* The variable req_caseopt contains either the REQ_CASELESS value or zero,
2636 according to the current setting of the caseless flag. REQ_CASELESS is a bit
2637 value > 255. It is added into the firstbyte or reqbyte variables to record the
2638 case status of the value. This is used only for ASCII characters. */
2639
2640 req_caseopt = ((options & PCRE_CASELESS) != 0)? REQ_CASELESS : 0;
2641
2642 /* Switch on next character until the end of the branch */
2643
2644 for (;; ptr++)
2645 {
2646 BOOL negate_class;
2647 BOOL should_flip_negation;
2648 BOOL possessive_quantifier;
2649 BOOL is_quantifier;
2650 BOOL is_recurse;
2651 BOOL reset_bracount;
2652 int class_charcount;
2653 int class_lastchar;
2654 int newoptions;
2655 int recno;
2656 int refsign;
2657 int skipbytes;
2658 int subreqbyte;
2659 int subfirstbyte;
2660 int terminator;
2661 int mclength;
2662 uschar mcbuffer[8];
2663
2664 /* Get next byte in the pattern */
2665
2666 c = *ptr;
2667
2668 /* If we are in the pre-compile phase, accumulate the length used for the
2669 previous cycle of this loop. */
2670
2671 if (lengthptr != NULL)
2672 {
2673 #ifdef DEBUG
2674 if (code > cd->hwm) cd->hwm = code; /* High water info */
2675 #endif
2676 if (code > cd->start_workspace + COMPILE_WORK_SIZE) /* Check for overrun */
2677 {
2678 *errorcodeptr = ERR52;
2679 goto FAILED;
2680 }
2681
2682 /* There is at least one situation where code goes backwards: this is the
2683 case of a zero quantifier after a class (e.g. [ab]{0}). At compile time,
2684 the class is simply eliminated. However, it is created first, so we have to
2685 allow memory for it. Therefore, don't ever reduce the length at this point.
2686 */
2687
2688 if (code < last_code) code = last_code;
2689
2690 /* Paranoid check for integer overflow */
2691
2692 if (OFLOW_MAX - *lengthptr < code - last_code)
2693 {
2694 *errorcodeptr = ERR20;
2695 goto FAILED;
2696 }
2697
2698 *lengthptr += code - last_code;
2699 DPRINTF(("length=%d added %d c=%c\n", *lengthptr, code - last_code, c));
2700
2701 /* If "previous" is set and it is not at the start of the work space, move
2702 it back to there, in order to avoid filling up the work space. Otherwise,
2703 if "previous" is NULL, reset the current code pointer to the start. */
2704
2705 if (previous != NULL)
2706 {
2707 if (previous > orig_code)
2708 {
2709 memmove(orig_code, previous, code - previous);
2710 code -= previous - orig_code;
2711 previous = orig_code;
2712 }
2713 }
2714 else code = orig_code;
2715
2716 /* Remember where this code item starts so we can pick up the length
2717 next time round. */
2718
2719 last_code = code;
2720 }
2721
2722 /* In the real compile phase, just check the workspace used by the forward
2723 reference list. */
2724
2725 else if (cd->hwm > cd->start_workspace + COMPILE_WORK_SIZE)
2726 {
2727 *errorcodeptr = ERR52;
2728 goto FAILED;
2729 }
2730
2731 /* If in \Q...\E, check for the end; if not, we have a literal */
2732
2733 if (inescq && c != 0)
2734 {
2735 if (c == CHAR_BACKSLASH && ptr[1] == CHAR_E)
2736 {
2737 inescq = FALSE;
2738 ptr++;
2739 continue;
2740 }
2741 else
2742 {
2743 if (previous_callout != NULL)
2744 {
2745 if (lengthptr == NULL) /* Don't attempt in pre-compile phase */
2746 complete_callout(previous_callout, ptr, cd);
2747 previous_callout = NULL;
2748 }
2749 if ((options & PCRE_AUTO_CALLOUT) != 0)
2750 {
2751 previous_callout = code;
2752 code = auto_callout(code, ptr, cd);
2753 }
2754 goto NORMAL_CHAR;
2755 }
2756 }
2757
2758 /* Fill in length of a previous callout, except when the next thing is
2759 a quantifier. */
2760
2761 is_quantifier =
2762 c == CHAR_ASTERISK || c == CHAR_PLUS || c == CHAR_QUESTION_MARK ||
2763 (c == CHAR_LEFT_CURLY_BRACKET && is_counted_repeat(ptr+1));
2764
2765 if (!is_quantifier && previous_callout != NULL &&
2766 after_manual_callout-- <= 0)
2767 {
2768 if (lengthptr == NULL) /* Don't attempt in pre-compile phase */
2769 complete_callout(previous_callout, ptr, cd);
2770 previous_callout = NULL;
2771 }
2772
2773 /* In extended mode, skip white space and comments */
2774
2775 if ((options & PCRE_EXTENDED) != 0)
2776 {
2777 if ((cd->ctypes[c] & ctype_space) != 0) continue;
2778 if (c == CHAR_NUMBER_SIGN)
2779 {
2780 while (*(++ptr) != 0)
2781 {
2782 if (IS_NEWLINE(ptr)) { ptr += cd->nllen - 1; break; }
2783 }
2784 if (*ptr != 0) continue;
2785
2786 /* Else fall through to handle end of string */
2787 c = 0;
2788 }
2789 }
2790
2791 /* No auto callout for quantifiers. */
2792
2793 if ((options & PCRE_AUTO_CALLOUT) != 0 && !is_quantifier)
2794 {
2795 previous_callout = code;
2796 code = auto_callout(code, ptr, cd);
2797 }
2798
2799 switch(c)
2800 {
2801 /* ===================================================================*/
2802 case 0: /* The branch terminates at string end */
2803 case CHAR_VERTICAL_LINE: /* or | or ) */
2804 case CHAR_RIGHT_PARENTHESIS:
2805 *firstbyteptr = firstbyte;
2806 *reqbyteptr = reqbyte;
2807 *codeptr = code;
2808 *ptrptr = ptr;
2809 if (lengthptr != NULL)
2810 {
2811 if (OFLOW_MAX - *lengthptr < code - last_code)
2812 {
2813 *errorcodeptr = ERR20;
2814 goto FAILED;
2815 }
2816 *lengthptr += code - last_code; /* To include callout length */
2817 DPRINTF((">> end branch\n"));
2818 }
2819 return TRUE;
2820
2821
2822 /* ===================================================================*/
2823 /* Handle single-character metacharacters. In multiline mode, ^ disables
2824 the setting of any following char as a first character. */
2825
2826 case CHAR_CIRCUMFLEX_ACCENT:
2827 if ((options & PCRE_MULTILINE) != 0)
2828 {
2829 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
2830 }
2831 previous = NULL;
2832 *code++ = OP_CIRC;
2833 break;
2834
2835 case CHAR_DOLLAR_SIGN:
2836 previous = NULL;
2837 *code++ = OP_DOLL;
2838 break;
2839
2840 /* There can never be a first char if '.' is first, whatever happens about
2841 repeats. The value of reqbyte doesn't change either. */
2842
2843 case CHAR_DOT:
2844 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
2845 zerofirstbyte = firstbyte;
2846 zeroreqbyte = reqbyte;
2847 previous = code;
2848 *code++ = ((options & PCRE_DOTALL) != 0)? OP_ALLANY: OP_ANY;
2849 break;
2850
2851
2852 /* ===================================================================*/
2853 /* Character classes. If the included characters are all < 256, we build a
2854 32-byte bitmap of the permitted characters, except in the special case
2855 where there is only one such character. For negated classes, we build the
2856 map as usual, then invert it at the end. However, we use a different opcode
2857 so that data characters > 255 can be handled correctly.
2858
2859 If the class contains characters outside the 0-255 range, a different
2860 opcode is compiled. It may optionally have a bit map for characters < 256,
2861 but those above are are explicitly listed afterwards. A flag byte tells
2862 whether the bitmap is present, and whether this is a negated class or not.
2863
2864 In JavaScript compatibility mode, an isolated ']' causes an error. In
2865 default (Perl) mode, it is treated as a data character. */
2866
2867 case CHAR_RIGHT_SQUARE_BRACKET:
2868 if ((cd->external_options & PCRE_JAVASCRIPT_COMPAT) != 0)
2869 {
2870 *errorcodeptr = ERR64;
2871 goto FAILED;
2872 }
2873 goto NORMAL_CHAR;
2874
2875 case CHAR_LEFT_SQUARE_BRACKET:
2876 previous = code;
2877
2878 /* PCRE supports POSIX class stuff inside a class. Perl gives an error if
2879 they are encountered at the top level, so we'll do that too. */
2880
2881 if ((ptr[1] == CHAR_COLON || ptr[1] == CHAR_DOT ||
2882 ptr[1] == CHAR_EQUALS_SIGN) &&
2883 check_posix_syntax(ptr, &tempptr))
2884 {
2885 *errorcodeptr = (ptr[1] == CHAR_COLON)? ERR13 : ERR31;
2886 goto FAILED;
2887 }
2888
2889 /* If the first character is '^', set the negation flag and skip it. Also,
2890 if the first few characters (either before or after ^) are \Q\E or \E we
2891 skip them too. This makes for compatibility with Perl. */
2892
2893 negate_class = FALSE;
2894 for (;;)
2895 {
2896 c = *(++ptr);
2897 if (c == CHAR_BACKSLASH)
2898 {
2899 if (ptr[1] == CHAR_E)
2900 ptr++;
2901 else if (strncmp((const char *)ptr+1,
2902 STR_Q STR_BACKSLASH STR_E, 3) == 0)
2903 ptr += 3;
2904 else
2905 break;
2906 }
2907 else if (!negate_class && c == CHAR_CIRCUMFLEX_ACCENT)
2908 negate_class = TRUE;
2909 else break;
2910 }
2911
2912 /* Empty classes are allowed in JavaScript compatibility mode. Otherwise,
2913 an initial ']' is taken as a data character -- the code below handles
2914 that. In JS mode, [] must always fail, so generate OP_FAIL, whereas
2915 [^] must match any character, so generate OP_ALLANY. */
2916
2917 if (c == CHAR_RIGHT_SQUARE_BRACKET &&
2918 (cd->external_options & PCRE_JAVASCRIPT_COMPAT) != 0)
2919 {
2920 *code++ = negate_class? OP_ALLANY : OP_FAIL;
2921 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
2922 zerofirstbyte = firstbyte;
2923 break;
2924 }
2925
2926 /* If a class contains a negative special such as \S, we need to flip the
2927 negation flag at the end, so that support for characters > 255 works
2928 correctly (they are all included in the class). */
2929
2930 should_flip_negation = FALSE;
2931
2932 /* Keep a count of chars with values < 256 so that we can optimize the case
2933 of just a single character (as long as it's < 256). However, For higher
2934 valued UTF-8 characters, we don't yet do any optimization. */
2935
2936 class_charcount = 0;
2937 class_lastchar = -1;
2938
2939 /* Initialize the 32-char bit map to all zeros. We build the map in a
2940 temporary bit of memory, in case the class contains only 1 character (less
2941 than 256), because in that case the compiled code doesn't use the bit map.
2942 */
2943
2944 memset(classbits, 0, 32 * sizeof(uschar));
2945
2946 #ifdef SUPPORT_UTF8
2947 class_utf8 = FALSE; /* No chars >= 256 */
2948 class_utf8data = code + LINK_SIZE + 2; /* For UTF-8 items */
2949 class_utf8data_base = class_utf8data; /* For resetting in pass 1 */
2950 #endif
2951
2952 /* Process characters until ] is reached. By writing this as a "do" it
2953 means that an initial ] is taken as a data character. At the start of the
2954 loop, c contains the first byte of the character. */
2955
2956 if (c != 0) do
2957 {
2958 const uschar *oldptr;
2959
2960 #ifdef SUPPORT_UTF8
2961 if (utf8 && c > 127)
2962 { /* Braces are required because the */
2963 GETCHARLEN(c, ptr, ptr); /* macro generates multiple statements */
2964 }
2965
2966 /* In the pre-compile phase, accumulate the length of any UTF-8 extra
2967 data and reset the pointer. This is so that very large classes that
2968 contain a zillion UTF-8 characters no longer overwrite the work space
2969 (which is on the stack). */
2970
2971 if (lengthptr != NULL)
2972 {
2973 *lengthptr += class_utf8data - class_utf8data_base;
2974 class_utf8data = class_utf8data_base;
2975 }
2976
2977 #endif
2978
2979 /* Inside \Q...\E everything is literal except \E */
2980
2981 if (inescq)
2982 {
2983 if (c == CHAR_BACKSLASH && ptr[1] == CHAR_E) /* If we are at \E */
2984 {
2985 inescq = FALSE; /* Reset literal state */
2986 ptr++; /* Skip the 'E' */
2987 continue; /* Carry on with next */
2988 }
2989 goto CHECK_RANGE; /* Could be range if \E follows */
2990 }
2991
2992 /* Handle POSIX class names. Perl allows a negation extension of the
2993 form [:^name:]. A square bracket that doesn't match the syntax is
2994 treated as a literal. We also recognize the POSIX constructions
2995 [.ch.] and [=ch=] ("collating elements") and fault them, as Perl
2996 5.6 and 5.8 do. */
2997
2998 if (c == CHAR_LEFT_SQUARE_BRACKET &&
2999 (ptr[1] == CHAR_COLON || ptr[1] == CHAR_DOT ||
3000 ptr[1] == CHAR_EQUALS_SIGN) && check_posix_syntax(ptr, &tempptr))
3001 {
3002 BOOL local_negate = FALSE;
3003 int posix_class, taboffset, tabopt;
3004 register const uschar *cbits = cd->cbits;
3005 uschar pbits[32];
3006
3007 if (ptr[1] != CHAR_COLON)
3008 {
3009 *errorcodeptr = ERR31;
3010 goto FAILED;
3011 }
3012
3013 ptr += 2;
3014 if (*ptr == CHAR_CIRCUMFLEX_ACCENT)
3015 {
3016 local_negate = TRUE;
3017 should_flip_negation = TRUE; /* Note negative special */
3018 ptr++;
3019 }
3020
3021 posix_class = check_posix_name(ptr, tempptr - ptr);
3022 if (posix_class < 0)
3023 {
3024 *errorcodeptr = ERR30;
3025 goto FAILED;
3026 }
3027
3028 /* If matching is caseless, upper and lower are converted to
3029 alpha. This relies on the fact that the class table starts with
3030 alpha, lower, upper as the first 3 entries. */
3031
3032 if ((options & PCRE_CASELESS) != 0 && posix_class <= 2)
3033 posix_class = 0;
3034
3035 /* We build the bit map for the POSIX class in a chunk of local store
3036 because we may be adding and subtracting from it, and we don't want to
3037 subtract bits that may be in the main map already. At the end we or the
3038 result into the bit map that is being built. */
3039
3040 posix_class *= 3;
3041
3042 /* Copy in the first table (always present) */
3043
3044 memcpy(pbits, cbits + posix_class_maps[posix_class],
3045 32 * sizeof(uschar));
3046
3047 /* If there is a second table, add or remove it as required. */
3048
3049 taboffset = posix_class_maps[posix_class + 1];
3050 tabopt = posix_class_maps[posix_class + 2];
3051
3052 if (taboffset >= 0)
3053 {
3054 if (tabopt >= 0)
3055 for (c = 0; c < 32; c++) pbits[c] |= cbits[c + taboffset];
3056 else
3057 for (c = 0; c < 32; c++) pbits[c] &= ~cbits[c + taboffset];
3058 }
3059
3060 /* Not see if we need to remove any special characters. An option
3061 value of 1 removes vertical space and 2 removes underscore. */
3062
3063 if (tabopt < 0) tabopt = -tabopt;
3064 if (tabopt == 1) pbits[1] &= ~0x3c;
3065 else if (tabopt == 2) pbits[11] &= 0x7f;
3066
3067 /* Add the POSIX table or its complement into the main table that is
3068 being built and we are done. */
3069
3070 if (local_negate)
3071 for (c = 0; c < 32; c++) classbits[c] |= ~pbits[c];
3072 else
3073 for (c = 0; c < 32; c++) classbits[c] |= pbits[c];
3074
3075 ptr = tempptr + 1;
3076 class_charcount = 10; /* Set > 1; assumes more than 1 per class */
3077 continue; /* End of POSIX syntax handling */
3078 }
3079
3080 /* Backslash may introduce a single character, or it may introduce one
3081 of the specials, which just set a flag. The sequence \b is a special
3082 case. Inside a class (and only there) it is treated as backspace.
3083 Elsewhere it marks a word boundary. Other escapes have preset maps ready
3084 to 'or' into the one we are building. We assume they have more than one
3085 character in them, so set class_charcount bigger than one. */
3086
3087 if (c == CHAR_BACKSLASH)
3088 {
3089 c = check_escape(&ptr, errorcodeptr, cd->bracount, options, TRUE);
3090 if (*errorcodeptr != 0) goto FAILED;
3091
3092 if (-c == ESC_b) c = CHAR_BS; /* \b is backspace in a class */
3093 else if (-c == ESC_X) c = CHAR_X; /* \X is literal X in a class */
3094 else if (-c == ESC_R) c = CHAR_R; /* \R is literal R in a class */
3095 else if (-c == ESC_Q) /* Handle start of quoted string */
3096 {
3097 if (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
3098 {
3099 ptr += 2; /* avoid empty string */
3100 }
3101 else inescq = TRUE;
3102 continue;
3103 }
3104 else if (-c == ESC_E) continue; /* Ignore orphan \E */
3105
3106 if (c < 0)
3107 {
3108 register const uschar *cbits = cd->cbits;
3109 class_charcount += 2; /* Greater than 1 is what matters */
3110
3111 /* Save time by not doing this in the pre-compile phase. */
3112
3113 if (lengthptr == NULL) switch (-c)
3114 {
3115 case ESC_d:
3116 for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_digit];
3117 continue;
3118
3119 case ESC_D:
3120 should_flip_negation = TRUE;
3121 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_digit];
3122 continue;
3123
3124 case ESC_w:
3125 for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_word];
3126 continue;
3127
3128 case ESC_W:
3129 should_flip_negation = TRUE;
3130 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_word];
3131 continue;
3132
3133 case ESC_s:
3134 for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_space];
3135 classbits[1] &= ~0x08; /* Perl 5.004 onwards omits VT from \s */
3136 continue;
3137
3138 case ESC_S:
3139 should_flip_negation = TRUE;
3140 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_space];
3141 classbits[1] |= 0x08; /* Perl 5.004 onwards omits VT from \s */
3142 continue;
3143
3144 default: /* Not recognized; fall through */
3145 break; /* Need "default" setting to stop compiler warning. */
3146 }
3147
3148 /* In the pre-compile phase, just do the recognition. */
3149
3150 else if (c == -ESC_d || c == -ESC_D || c == -ESC_w ||
3151 c == -ESC_W || c == -ESC_s || c == -ESC_S) continue;
3152
3153 /* We need to deal with \H, \h, \V, and \v in both phases because
3154 they use extra memory. */
3155
3156 if (-c == ESC_h)
3157 {
3158 SETBIT(classbits, 0x09); /* VT */
3159 SETBIT(classbits, 0x20); /* SPACE */
3160 SETBIT(classbits, 0xa0); /* NSBP */
3161 #ifdef SUPPORT_UTF8
3162 if (utf8)
3163 {
3164 class_utf8 = TRUE;
3165 *class_utf8data++ = XCL_SINGLE;
3166 class_utf8data += _pcre_ord2utf8(0x1680, class_utf8data);
3167 *class_utf8data++ = XCL_SINGLE;
3168 class_utf8data += _pcre_ord2utf8(0x180e, class_utf8data);
3169 *class_utf8data++ = XCL_RANGE;
3170 class_utf8data += _pcre_ord2utf8(0x2000, class_utf8data);
3171 class_utf8data += _pcre_ord2utf8(0x200A, class_utf8data);
3172 *class_utf8data++ = XCL_SINGLE;
3173 class_utf8data += _pcre_ord2utf8(0x202f, class_utf8data);
3174 *class_utf8data++ = XCL_SINGLE;
3175 class_utf8data += _pcre_ord2utf8(0x205f, class_utf8data);
3176 *class_utf8data++ = XCL_SINGLE;
3177 class_utf8data += _pcre_ord2utf8(0x3000, class_utf8data);
3178 }
3179 #endif
3180 continue;
3181 }
3182
3183 if (-c == ESC_H)
3184 {
3185 for (c = 0; c < 32; c++)
3186 {
3187 int x = 0xff;
3188 switch (c)
3189 {
3190 case 0x09/8: x ^= 1 << (0x09%8); break;
3191 case 0x20/8: x ^= 1 << (0x20%8); break;
3192 case 0xa0/8: x ^= 1 << (0xa0%8); break;
3193 default: break;
3194 }
3195 classbits[c] |= x;
3196 }
3197
3198 #ifdef SUPPORT_UTF8
3199 if (utf8)
3200 {
3201 class_utf8 = TRUE;
3202 *class_utf8data++ = XCL_RANGE;
3203 class_utf8data += _pcre_ord2utf8(0x0100, class_utf8data);
3204 class_utf8data += _pcre_ord2utf8(0x167f, class_utf8data);
3205 *class_utf8data++ = XCL_RANGE;
3206 class_utf8data += _pcre_ord2utf8(0x1681, class_utf8data);
3207 class_utf8data += _pcre_ord2utf8(0x180d, class_utf8data);
3208 *class_utf8data++ = XCL_RANGE;
3209 class_utf8data += _pcre_ord2utf8(0x180f, class_utf8data);
3210 class_utf8data += _pcre_ord2utf8(0x1fff, class_utf8data);
3211 *class_utf8data++ = XCL_RANGE;
3212 class_utf8data += _pcre_ord2utf8(0x200B, class_utf8data);
3213 class_utf8data += _pcre_ord2utf8(0x202e, class_utf8data);
3214 *class_utf8data++ = XCL_RANGE;
3215 class_utf8data += _pcre_ord2utf8(0x2030, class_utf8data);
3216 class_utf8data += _pcre_ord2utf8(0x205e, class_utf8data);
3217 *class_utf8data++ = XCL_RANGE;
3218 class_utf8data += _pcre_ord2utf8(0x2060, class_utf8data);
3219 class_utf8data += _pcre_ord2utf8(0x2fff, class_utf8data);
3220 *class_utf8data++ = XCL_RANGE;
3221 class_utf8data += _pcre_ord2utf8(0x3001, class_utf8data);
3222 class_utf8data += _pcre_ord2utf8(0x7fffffff, class_utf8data);
3223 }
3224 #endif
3225 continue;
3226 }
3227
3228 if (-c == ESC_v)
3229 {
3230 SETBIT(classbits, 0x0a); /* LF */
3231 SETBIT(classbits, 0x0b); /* VT */
3232 SETBIT(classbits, 0x0c); /* FF */
3233 SETBIT(classbits, 0x0d); /* CR */
3234 SETBIT(classbits, 0x85); /* NEL */
3235 #ifdef SUPPORT_UTF8
3236 if (utf8)
3237 {
3238 class_utf8 = TRUE;
3239 *class_utf8data++ = XCL_RANGE;
3240 class_utf8data += _pcre_ord2utf8(0x2028, class_utf8data);
3241 class_utf8data += _pcre_ord2utf8(0x2029, class_utf8data);
3242 }
3243 #endif
3244 continue;
3245 }
3246
3247 if (-c == ESC_V)
3248 {
3249 for (c = 0; c < 32; c++)
3250 {
3251 int x = 0xff;
3252 switch (c)
3253 {
3254 case 0x0a/8: x ^= 1 << (0x0a%8);
3255 x ^= 1 << (0x0b%8);
3256 x ^= 1 << (0x0c%8);
3257 x ^= 1 << (0x0d%8);
3258 break;
3259 case 0x85/8: x ^= 1 << (0x85%8); break;
3260 default: break;
3261 }
3262 classbits[c] |= x;
3263 }
3264
3265 #ifdef SUPPORT_UTF8
3266 if (utf8)
3267 {
3268 class_utf8 = TRUE;
3269 *class_utf8data++ = XCL_RANGE;
3270 class_utf8data += _pcre_ord2utf8(0x0100, class_utf8data);
3271 class_utf8data += _pcre_ord2utf8(0x2027, class_utf8data);
3272 *class_utf8data++ = XCL_RANGE;
3273 class_utf8data += _pcre_ord2utf8(0x2029, class_utf8data);
3274 class_utf8data += _pcre_ord2utf8(0x7fffffff, class_utf8data);
3275 }
3276 #endif
3277 continue;
3278 }
3279
3280 /* We need to deal with \P and \p in both phases. */
3281
3282 #ifdef SUPPORT_UCP
3283 if (-c == ESC_p || -c == ESC_P)
3284 {
3285 BOOL negated;
3286 int pdata;
3287 int ptype = get_ucp(&ptr, &negated, &pdata, errorcodeptr);
3288 if (ptype < 0) goto FAILED;
3289 class_utf8 = TRUE;
3290 *class_utf8data++ = ((-c == ESC_p) != negated)?
3291 XCL_PROP : XCL_NOTPROP;
3292 *class_utf8data++ = ptype;
3293 *class_utf8data++ = pdata;
3294 class_charcount -= 2; /* Not a < 256 character */
3295 continue;
3296 }
3297 #endif
3298 /* Unrecognized escapes are faulted if PCRE is running in its
3299 strict mode. By default, for compatibility with Perl, they are
3300 treated as literals. */
3301
3302 if ((options & PCRE_EXTRA) != 0)
3303 {
3304 *errorcodeptr = ERR7;
3305 goto FAILED;
3306 }
3307
3308 class_charcount -= 2; /* Undo the default count from above */
3309 c = *ptr; /* Get the final character and fall through */
3310 }
3311
3312 /* Fall through if we have a single character (c >= 0). This may be
3313 greater than 256 in UTF-8 mode. */
3314
3315 } /* End of backslash handling */
3316
3317 /* A single character may be followed by '-' to form a range. However,
3318 Perl does not permit ']' to be the end of the range. A '-' character
3319 at the end is treated as a literal. Perl ignores orphaned \E sequences
3320 entirely. The code for handling \Q and \E is messy. */
3321
3322 CHECK_RANGE:
3323 while (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
3324 {
3325 inescq = FALSE;
3326 ptr += 2;
3327 }
3328
3329 oldptr = ptr;
3330
3331 /* Remember \r or \n */
3332
3333 if (c == CHAR_CR || c == CHAR_NL) cd->external_flags |= PCRE_HASCRORLF;
3334
3335 /* Check for range */
3336
3337 if (!inescq && ptr[1] == CHAR_MINUS)
3338 {
3339 int d;
3340 ptr += 2;
3341 while (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_E) ptr += 2;
3342
3343 /* If we hit \Q (not followed by \E) at this point, go into escaped
3344 mode. */
3345
3346 while (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_Q)
3347 {
3348 ptr += 2;
3349 if (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_E)
3350 { ptr += 2; continue; }
3351 inescq = TRUE;
3352 break;
3353 }
3354
3355 if (*ptr == 0 || (!inescq && *ptr == CHAR_RIGHT_SQUARE_BRACKET))
3356 {
3357 ptr = oldptr;
3358 goto LONE_SINGLE_CHARACTER;
3359 }
3360
3361 #ifdef SUPPORT_UTF8
3362 if (utf8)
3363 { /* Braces are required because the */
3364 GETCHARLEN(d, ptr, ptr); /* macro generates multiple statements */
3365 }
3366 else
3367 #endif
3368 d = *ptr; /* Not UTF-8 mode */
3369
3370 /* The second part of a range can be a single-character escape, but
3371 not any of the other escapes. Perl 5.6 treats a hyphen as a literal
3372 in such circumstances. */
3373
3374 if (!inescq && d == CHAR_BACKSLASH)
3375 {
3376 d = check_escape(&ptr, errorcodeptr, cd->bracount, options, TRUE);
3377 if (*errorcodeptr != 0) goto FAILED;
3378
3379 /* \b is backspace; \X is literal X; \R is literal R; any other
3380 special means the '-' was literal */
3381
3382 if (d < 0)
3383 {
3384 if (d == -ESC_b) d = CHAR_BS;
3385 else if (d == -ESC_X) d = CHAR_X;
3386 else if (d == -ESC_R) d = CHAR_R; else
3387 {
3388 ptr = oldptr;
3389 goto LONE_SINGLE_CHARACTER; /* A few lines below */
3390 }
3391 }
3392 }
3393
3394 /* Check that the two values are in the correct order. Optimize
3395 one-character ranges */
3396
3397 if (d < c)
3398 {
3399 *errorcodeptr = ERR8;
3400 goto FAILED;
3401 }
3402
3403 if (d == c) goto LONE_SINGLE_CHARACTER; /* A few lines below */
3404
3405 /* Remember \r or \n */
3406
3407 if (d == CHAR_CR || d == CHAR_NL) cd->external_flags |= PCRE_HASCRORLF;
3408
3409 /* In UTF-8 mode, if the upper limit is > 255, or > 127 for caseless
3410 matching, we have to use an XCLASS with extra data items. Caseless
3411 matching for characters > 127 is available only if UCP support is
3412 available. */
3413
3414 #ifdef SUPPORT_UTF8
3415 if (utf8 && (d > 255 || ((options & PCRE_CASELESS) != 0 && d > 127)))
3416 {
3417 class_utf8 = TRUE;
3418
3419 /* With UCP support, we can find the other case equivalents of
3420 the relevant characters. There may be several ranges. Optimize how
3421 they fit with the basic range. */
3422
3423 #ifdef SUPPORT_UCP
3424 if ((options & PCRE_CASELESS) != 0)
3425 {
3426 unsigned int occ, ocd;
3427 unsigned int cc = c;
3428 unsigned int origd = d;
3429 while (get_othercase_range(&cc, origd, &occ, &ocd))
3430 {
3431 if (occ >= (unsigned int)c &&
3432 ocd <= (unsigned int)d)
3433 continue; /* Skip embedded ranges */
3434
3435 if (occ < (unsigned int)c &&
3436 ocd >= (unsigned int)c - 1) /* Extend the basic range */
3437 { /* if there is overlap, */
3438 c = occ; /* noting that if occ < c */
3439 continue; /* we can't have ocd > d */
3440 } /* because a subrange is */
3441 if (ocd > (unsigned int)d &&
3442 occ <= (unsigned int)d + 1) /* always shorter than */
3443 { /* the basic range. */
3444 d = ocd;
3445 continue;
3446 }
3447
3448 if (occ == ocd)
3449 {
3450 *class_utf8data++ = XCL_SINGLE;
3451 }
3452 else
3453 {
3454 *class_utf8data++ = XCL_RANGE;
3455 class_utf8data += _pcre_ord2utf8(occ, class_utf8data);
3456 }
3457 class_utf8data += _pcre_ord2utf8(ocd, class_utf8data);
3458 }
3459 }
3460 #endif /* SUPPORT_UCP */
3461
3462 /* Now record the original range, possibly modified for UCP caseless
3463 overlapping ranges. */
3464
3465 *class_utf8data++ = XCL_RANGE;
3466 class_utf8data += _pcre_ord2utf8(c, class_utf8data);
3467 class_utf8data += _pcre_ord2utf8(d, class_utf8data);
3468
3469 /* With UCP support, we are done. Without UCP support, there is no
3470 caseless matching for UTF-8 characters > 127; we can use the bit map
3471 for the smaller ones. */
3472
3473 #ifdef SUPPORT_UCP
3474 continue; /* With next character in the class */
3475 #else
3476 if ((options & PCRE_CASELESS) == 0 || c > 127) continue;
3477
3478 /* Adjust upper limit and fall through to set up the map */
3479
3480 d = 127;
3481
3482 #endif /* SUPPORT_UCP */
3483 }
3484 #endif /* SUPPORT_UTF8 */
3485
3486 /* We use the bit map for all cases when not in UTF-8 mode; else
3487 ranges that lie entirely within 0-127 when there is UCP support; else
3488 for partial ranges without UCP support. */
3489
3490 class_charcount += d - c + 1;
3491 class_lastchar = d;
3492
3493 /* We can save a bit of time by skipping this in the pre-compile. */
3494
3495 if (lengthptr == NULL) for (; c <= d; c++)
3496 {
3497 classbits[c/8] |= (1 << (c&7));
3498 if ((options & PCRE_CASELESS) != 0)
3499 {
3500 int uc = cd->fcc[c]; /* flip case */
3501 classbits[uc/8] |= (1 << (uc&7));
3502 }
3503 }
3504
3505 continue; /* Go get the next char in the class */
3506 }
3507
3508 /* Handle a lone single character - we can get here for a normal
3509 non-escape char, or after \ that introduces a single character or for an
3510 apparent range that isn't. */
3511
3512 LONE_SINGLE_CHARACTER:
3513
3514 /* Handle a character that cannot go in the bit map */
3515
3516 #ifdef SUPPORT_UTF8
3517 if (utf8 && (c > 255 || ((options & PCRE_CASELESS) != 0 && c > 127)))
3518 {
3519 class_utf8 = TRUE;
3520 *class_utf8data++ = XCL_SINGLE;
3521 class_utf8data += _pcre_ord2utf8(c, class_utf8data);
3522
3523 #ifdef SUPPORT_UCP
3524 if ((options & PCRE_CASELESS) != 0)
3525 {
3526 unsigned int othercase;
3527 if ((othercase = UCD_OTHERCASE(c)) != c)
3528 {
3529 *class_utf8data++ = XCL_SINGLE;
3530 class_utf8data += _pcre_ord2utf8(othercase, class_utf8data);
3531 }
3532 }
3533 #endif /* SUPPORT_UCP */
3534
3535 }
3536 else
3537 #endif /* SUPPORT_UTF8 */
3538
3539 /* Handle a single-byte character */
3540 {
3541 classbits[c/8] |= (1 << (c&7));
3542 if ((options & PCRE_CASELESS) != 0)
3543 {
3544 c = cd->fcc[c]; /* flip case */
3545 classbits[c/8] |= (1 << (c&7));
3546 }
3547 class_charcount++;
3548 class_lastchar = c;
3549 }
3550 }
3551
3552 /* Loop until ']' reached. This "while" is the end of the "do" above. */
3553
3554 while ((c = *(++ptr)) != 0 && (c != CHAR_RIGHT_SQUARE_BRACKET || inescq));
3555
3556 if (c == 0) /* Missing terminating ']' */
3557 {
3558 *errorcodeptr = ERR6;
3559 goto FAILED;
3560 }
3561
3562
3563 /* This code has been disabled because it would mean that \s counts as
3564 an explicit \r or \n reference, and that's not really what is wanted. Now
3565 we set the flag only if there is a literal "\r" or "\n" in the class. */
3566
3567 #if 0
3568 /* Remember whether \r or \n are in this class */
3569
3570 if (negate_class)
3571 {
3572 if ((classbits[1] & 0x24) != 0x24) cd->external_flags |= PCRE_HASCRORLF;
3573 }
3574 else
3575 {
3576 if ((classbits[1] & 0x24) != 0) cd->external_flags |= PCRE_HASCRORLF;
3577 }
3578 #endif
3579
3580
3581 /* If class_charcount is 1, we saw precisely one character whose value is
3582 less than 256. As long as there were no characters >= 128 and there was no
3583 use of \p or \P, in other words, no use of any XCLASS features, we can
3584 optimize.
3585
3586 In UTF-8 mode, we can optimize the negative case only if there were no
3587 characters >= 128 because OP_NOT and the related opcodes like OP_NOTSTAR
3588 operate on single-bytes only. This is an historical hangover. Maybe one day
3589 we can tidy these opcodes to handle multi-byte characters.
3590
3591 The optimization throws away the bit map. We turn the item into a
3592 1-character OP_CHAR[NC] if it's positive, or OP_NOT if it's negative. Note
3593 that OP_NOT does not support multibyte characters. In the positive case, it
3594 can cause firstbyte to be set. Otherwise, there can be no first char if
3595 this item is first, whatever repeat count may follow. In the case of
3596 reqbyte, save the previous value for reinstating. */
3597
3598 #ifdef SUPPORT_UTF8
3599 if (class_charcount == 1 && !class_utf8 &&
3600 (!utf8 || !negate_class || class_lastchar < 128))
3601 #else
3602 if (class_charcount == 1)
3603 #endif
3604 {
3605 zeroreqbyte = reqbyte;
3606
3607 /* The OP_NOT opcode works on one-byte characters only. */
3608
3609 if (negate_class)
3610 {
3611 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
3612 zerofirstbyte = firstbyte;
3613 *code++ = OP_NOT;
3614 *code++ = class_lastchar;
3615 break;
3616 }
3617
3618 /* For a single, positive character, get the value into mcbuffer, and
3619 then we can handle this with the normal one-character code. */
3620
3621 #ifdef SUPPORT_UTF8
3622 if (utf8 && class_lastchar > 127)
3623 mclength = _pcre_ord2utf8(class_lastchar, mcbuffer);
3624 else
3625 #endif
3626 {
3627 mcbuffer[0] = class_lastchar;
3628 mclength = 1;
3629 }
3630 goto ONE_CHAR;
3631 } /* End of 1-char optimization */
3632
3633 /* The general case - not the one-char optimization. If this is the first
3634 thing in the branch, there can be no first char setting, whatever the
3635 repeat count. Any reqbyte setting must remain unchanged after any kind of
3636 repeat. */
3637
3638 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
3639 zerofirstbyte = firstbyte;
3640 zeroreqbyte = reqbyte;
3641
3642 /* If there are characters with values > 255, we have to compile an
3643 extended class, with its own opcode, unless there was a negated special
3644 such as \S in the class, because in that case all characters > 255 are in
3645 the class, so any that were explicitly given as well can be ignored. If
3646 (when there are explicit characters > 255 that must be listed) there are no
3647 characters < 256, we can omit the bitmap in the actual compiled code. */
3648
3649 #ifdef SUPPORT_UTF8
3650 if (class_utf8 && !should_flip_negation)
3651 {
3652 *class_utf8data++ = XCL_END; /* Marks the end of extra data */
3653 *code++ = OP_XCLASS;
3654 code += LINK_SIZE;
3655 *code = negate_class? XCL_NOT : 0;
3656
3657 /* If the map is required, move up the extra data to make room for it;
3658 otherwise just move the code pointer to the end of the extra data. */
3659
3660 if (class_charcount > 0)
3661 {
3662 *code++ |= XCL_MAP;
3663 memmove(code + 32, code, class_utf8data - code);
3664 memcpy(code, classbits, 32);
3665 code = class_utf8data + 32;
3666 }
3667 else code = class_utf8data;
3668
3669 /* Now fill in the complete length of the item */
3670
3671 PUT(previous, 1, code - previous);
3672 break; /* End of class handling */
3673 }
3674 #endif
3675
3676 /* If there are no characters > 255, set the opcode to OP_CLASS or
3677 OP_NCLASS, depending on whether the whole class was negated and whether
3678 there were negative specials such as \S in the class. Then copy the 32-byte
3679 map into the code vector, negating it if necessary. */
3680
3681 *code++ = (negate_class == should_flip_negation) ? OP_CLASS : OP_NCLASS;
3682 if (negate_class)
3683 {
3684 if (lengthptr == NULL) /* Save time in the pre-compile phase */
3685 for (c = 0; c < 32; c++) code[c] = ~classbits[c];
3686 }
3687 else
3688 {
3689 memcpy(code, classbits, 32);
3690 }
3691 code += 32;
3692 break;
3693
3694
3695 /* ===================================================================*/
3696 /* Various kinds of repeat; '{' is not necessarily a quantifier, but this
3697 has been tested above. */
3698
3699 case CHAR_LEFT_CURLY_BRACKET:
3700 if (!is_quantifier) goto NORMAL_CHAR;
3701 ptr = read_repeat_counts(ptr+1, &repeat_min, &repeat_max, errorcodeptr);
3702 if (*errorcodeptr != 0) goto FAILED;
3703 goto REPEAT;
3704
3705 case CHAR_ASTERISK:
3706 repeat_min = 0;
3707 repeat_max = -1;
3708 goto REPEAT;
3709
3710 case CHAR_PLUS:
3711 repeat_min = 1;
3712 repeat_max = -1;
3713 goto REPEAT;
3714
3715 case CHAR_QUESTION_MARK:
3716 repeat_min = 0;
3717 repeat_max = 1;
3718
3719 REPEAT:
3720 if (previous == NULL)
3721 {
3722 *errorcodeptr = ERR9;
3723 goto FAILED;
3724 }
3725
3726 if (repeat_min == 0)
3727 {
3728 firstbyte = zerofirstbyte; /* Adjust for zero repeat */
3729 reqbyte = zeroreqbyte; /* Ditto */
3730 }
3731
3732 /* Remember whether this is a variable length repeat */
3733
3734 reqvary = (repeat_min == repeat_max)? 0 : REQ_VARY;
3735
3736 op_type = 0; /* Default single-char op codes */
3737 possessive_quantifier = FALSE; /* Default not possessive quantifier */
3738
3739 /* Save start of previous item, in case we have to move it up to make space
3740 for an inserted OP_ONCE for the additional '+' extension. */
3741
3742 tempcode = previous;
3743
3744 /* If the next character is '+', we have a possessive quantifier. This
3745 implies greediness, whatever the setting of the PCRE_UNGREEDY option.
3746 If the next character is '?' this is a minimizing repeat, by default,
3747 but if PCRE_UNGREEDY is set, it works the other way round. We change the
3748 repeat type to the non-default. */
3749
3750 if (ptr[1] == CHAR_PLUS)
3751 {
3752 repeat_type = 0; /* Force greedy */
3753 possessive_quantifier = TRUE;
3754 ptr++;
3755 }
3756 else if (ptr[1] == CHAR_QUESTION_MARK)
3757 {
3758 repeat_type = greedy_non_default;
3759 ptr++;
3760 }
3761 else repeat_type = greedy_default;
3762
3763 /* If previous was a character match, abolish the item and generate a
3764 repeat item instead. If a char item has a minumum of more than one, ensure
3765 that it is set in reqbyte - it might not be if a sequence such as x{3} is
3766 the first thing in a branch because the x will have gone into firstbyte
3767 instead. */
3768
3769 if (*previous == OP_CHAR || *previous == OP_CHARNC)
3770 {
3771 /* Deal with UTF-8 characters that take up more than one byte. It's
3772 easier to write this out separately than try to macrify it. Use c to
3773 hold the length of the character in bytes, plus 0x80 to flag that it's a
3774 length rather than a small character. */
3775
3776 #ifdef SUPPORT_UTF8
3777 if (utf8 && (code[-1] & 0x80) != 0)
3778 {
3779 uschar *lastchar = code - 1;
3780 while((*lastchar & 0xc0) == 0x80) lastchar--;
3781 c = code - lastchar; /* Length of UTF-8 character */
3782 memcpy(utf8_char, lastchar, c); /* Save the char */
3783 c |= 0x80; /* Flag c as a length */
3784 }
3785 else
3786 #endif
3787
3788 /* Handle the case of a single byte - either with no UTF8 support, or
3789 with UTF-8 disabled, or for a UTF-8 character < 128. */
3790
3791 {
3792 c = code[-1];
3793 if (repeat_min > 1) reqbyte = c | req_caseopt | cd->req_varyopt;
3794 }
3795
3796 /* If the repetition is unlimited, it pays to see if the next thing on
3797 the line is something that cannot possibly match this character. If so,
3798 automatically possessifying this item gains some performance in the case
3799 where the match fails. */
3800
3801 if (!possessive_quantifier &&
3802 repeat_max < 0 &&
3803 check_auto_possessive(*previous, c, utf8, utf8_char, ptr + 1,
3804 options, cd))
3805 {
3806 repeat_type = 0; /* Force greedy */
3807 possessive_quantifier = TRUE;
3808 }
3809
3810 goto OUTPUT_SINGLE_REPEAT; /* Code shared with single character types */
3811 }
3812
3813 /* If previous was a single negated character ([^a] or similar), we use
3814 one of the special opcodes, replacing it. The code is shared with single-
3815 character repeats by setting opt_type to add a suitable offset into
3816 repeat_type. We can also test for auto-possessification. OP_NOT is
3817 currently used only for single-byte chars. */
3818
3819 else if (*previous == OP_NOT)
3820 {
3821 op_type = OP_NOTSTAR - OP_STAR; /* Use "not" opcodes */
3822 c = previous[1];
3823 if (!possessive_quantifier &&
3824 repeat_max < 0 &&
3825 check_auto_possessive(OP_NOT, c, utf8, NULL, ptr + 1, options, cd))
3826 {
3827 repeat_type = 0; /* Force greedy */
3828 possessive_quantifier = TRUE;
3829 }
3830 goto OUTPUT_SINGLE_REPEAT;
3831 }
3832
3833 /* If previous was a character type match (\d or similar), abolish it and
3834 create a suitable repeat item. The code is shared with single-character
3835 repeats by setting op_type to add a suitable offset into repeat_type. Note
3836 the the Unicode property types will be present only when SUPPORT_UCP is
3837 defined, but we don't wrap the little bits of code here because it just
3838 makes it horribly messy. */
3839
3840 else if (*previous < OP_EODN)
3841 {
3842 uschar *oldcode;
3843 int prop_type, prop_value;
3844 op_type = OP_TYPESTAR - OP_STAR; /* Use type opcodes */
3845 c = *previous;
3846
3847 if (!possessive_quantifier &&
3848 repeat_max < 0 &&
3849 check_auto_possessive(c, 0, utf8, NULL, ptr + 1, options, cd))
3850 {
3851 repeat_type = 0; /* Force greedy */
3852 possessive_quantifier = TRUE;
3853 }
3854
3855 OUTPUT_SINGLE_REPEAT:
3856 if (*previous == OP_PROP || *previous == OP_NOTPROP)
3857 {
3858 prop_type = previous[1];
3859 prop_value = previous[2];
3860 }
3861 else prop_type = prop_value = -1;
3862
3863 oldcode = code;
3864 code = previous; /* Usually overwrite previous item */
3865
3866 /* If the maximum is zero then the minimum must also be zero; Perl allows
3867 this case, so we do too - by simply omitting the item altogether. */
3868
3869 if (repeat_max == 0) goto END_REPEAT;
3870
3871 /*--------------------------------------------------------------------*/
3872 /* This code is obsolete from release 8.00; the restriction was finally
3873 removed: */
3874
3875 /* All real repeats make it impossible to handle partial matching (maybe
3876 one day we will be able to remove this restriction). */
3877
3878 /* if (repeat_max != 1) cd->external_flags |= PCRE_NOPARTIAL; */
3879 /*--------------------------------------------------------------------*/
3880
3881 /* Combine the op_type with the repeat_type */
3882
3883 repeat_type += op_type;
3884
3885 /* A minimum of zero is handled either as the special case * or ?, or as
3886 an UPTO, with the maximum given. */
3887
3888 if (repeat_min == 0)
3889 {
3890 if (repeat_max == -1) *code++ = OP_STAR + repeat_type;
3891 else if (repeat_max == 1) *code++ = OP_QUERY + repeat_type;
3892 else
3893 {
3894 *code++ = OP_UPTO + repeat_type;
3895 PUT2INC(code, 0, repeat_max);
3896 }
3897 }
3898
3899 /* A repeat minimum of 1 is optimized into some special cases. If the
3900 maximum is unlimited, we use OP_PLUS. Otherwise, the original item is
3901 left in place and, if the maximum is greater than 1, we use OP_UPTO with
3902 one less than the maximum. */
3903
3904 else if (repeat_min == 1)
3905 {
3906 if (repeat_max == -1)
3907 *code++ = OP_PLUS + repeat_type;
3908 else
3909 {
3910 code = oldcode; /* leave previous item in place */
3911 if (repeat_max == 1) goto END_REPEAT;
3912 *code++ = OP_UPTO + repeat_type;
3913 PUT2INC(code, 0, repeat_max - 1);
3914 }
3915 }
3916
3917 /* The case {n,n} is just an EXACT, while the general case {n,m} is
3918 handled as an EXACT followed by an UPTO. */
3919
3920 else
3921 {
3922 *code++ = OP_EXACT + op_type; /* NB EXACT doesn't have repeat_type */
3923 PUT2INC(code, 0, repeat_min);
3924
3925 /* If the maximum is unlimited, insert an OP_STAR. Before doing so,
3926 we have to insert the character for the previous code. For a repeated
3927 Unicode property match, there are two extra bytes that define the
3928 required property. In UTF-8 mode, long characters have their length in
3929 c, with the 0x80 bit as a flag. */
3930
3931 if (repeat_max < 0)
3932 {
3933 #ifdef SUPPORT_UTF8
3934 if (utf8 && c >= 128)
3935 {
3936 memcpy(code, utf8_char, c & 7);
3937 code += c & 7;
3938 }
3939 else
3940 #endif
3941 {
3942 *code++ = c;
3943 if (prop_type >= 0)
3944 {
3945 *code++ = prop_type;
3946 *code++ = prop_value;
3947 }
3948 }
3949 *code++ = OP_STAR + repeat_type;
3950 }
3951
3952 /* Else insert an UPTO if the max is greater than the min, again
3953 preceded by the character, for the previously inserted code. If the
3954 UPTO is just for 1 instance, we can use QUERY instead. */
3955
3956 else if (repeat_max != repeat_min)
3957 {
3958 #ifdef SUPPORT_UTF8
3959 if (utf8 && c >= 128)
3960 {
3961 memcpy(code, utf8_char, c & 7);
3962 code += c & 7;
3963 }
3964 else
3965 #endif
3966 *code++ = c;
3967 if (prop_type >= 0)
3968 {
3969 *code++ = prop_type;
3970 *code++ = prop_value;
3971 }
3972 repeat_max -= repeat_min;
3973
3974 if (repeat_max == 1)
3975 {
3976 *code++ = OP_QUERY + repeat_type;
3977 }
3978 else
3979 {
3980 *code++ = OP_UPTO + repeat_type;
3981 PUT2INC(code, 0, repeat_max);
3982 }
3983 }
3984 }
3985
3986 /* The character or character type itself comes last in all cases. */
3987
3988 #ifdef SUPPORT_UTF8
3989 if (utf8 && c >= 128)
3990 {
3991 memcpy(code, utf8_char, c & 7);
3992 code += c & 7;
3993 }
3994 else
3995 #endif
3996 *code++ = c;
3997
3998 /* For a repeated Unicode property match, there are two extra bytes that
3999 define the required property. */
4000
4001 #ifdef SUPPORT_UCP
4002 if (prop_type >= 0)
4003 {
4004 *code++ = prop_type;
4005 *code++ = prop_value;
4006 }
4007 #endif
4008 }
4009
4010 /* If previous was a character class or a back reference, we put the repeat
4011 stuff after it, but just skip the item if the repeat was {0,0}. */
4012
4013 else if (*previous == OP_CLASS ||
4014 *previous == OP_NCLASS ||
4015 #ifdef SUPPORT_UTF8
4016 *previous == OP_XCLASS ||
4017 #endif
4018 *previous == OP_REF)
4019 {
4020 if (repeat_max == 0)
4021 {
4022 code = previous;
4023 goto END_REPEAT;
4024 }
4025
4026 /*--------------------------------------------------------------------*/
4027 /* This code is obsolete from release 8.00; the restriction was finally
4028 removed: */
4029
4030 /* All real repeats make it impossible to handle partial matching (maybe
4031 one day we will be able to remove this restriction). */
4032
4033 /* if (repeat_max != 1) cd->external_flags |= PCRE_NOPARTIAL; */
4034 /*--------------------------------------------------------------------*/
4035
4036 if (repeat_min == 0 && repeat_max == -1)
4037 *code++ = OP_CRSTAR + repeat_type;
4038 else if (repeat_min == 1 && repeat_max == -1)
4039 *code++ = OP_CRPLUS + repeat_type;
4040 else if (repeat_min == 0 && repeat_max == 1)
4041 *code++ = OP_CRQUERY + repeat_type;
4042 else
4043 {
4044 *code++ = OP_CRRANGE + repeat_type;
4045 PUT2INC(code, 0, repeat_min);
4046 if (repeat_max == -1) repeat_max = 0; /* 2-byte encoding for max */
4047 PUT2INC(code, 0, repeat_max);
4048 }
4049 }
4050
4051 /* If previous was a bracket group, we may have to replicate it in certain
4052 cases. */
4053
4054 else if (*previous == OP_BRA || *previous == OP_CBRA ||
4055 *previous == OP_ONCE || *previous == OP_COND)
4056 {
4057 register int i;
4058 int ketoffset = 0;
4059 int len = code - previous;
4060 uschar *bralink = NULL;
4061
4062 /* Repeating a DEFINE group is pointless */
4063
4064 if (*previous == OP_COND && previous[LINK_SIZE+1] == OP_DEF)
4065 {
4066 *errorcodeptr = ERR55;
4067 goto FAILED;
4068 }
4069
4070 /* If the maximum repeat count is unlimited, find the end of the bracket
4071 by scanning through from the start, and compute the offset back to it
4072 from the current code pointer. There may be an OP_OPT setting following
4073 the final KET, so we can't find the end just by going back from the code
4074 pointer. */
4075
4076 if (repeat_max == -1)
4077 {
4078 register uschar *ket = previous;
4079 do ket += GET(ket, 1); while (*ket != OP_KET);
4080 ketoffset = code - ket;
4081 }
4082
4083 /* The case of a zero minimum is special because of the need to stick
4084 OP_BRAZERO in front of it, and because the group appears once in the
4085 data, whereas in other cases it appears the minimum number of times. For
4086 this reason, it is simplest to treat this case separately, as otherwise
4087 the code gets far too messy. There are several special subcases when the
4088 minimum is zero. */
4089
4090 if (repeat_min == 0)
4091 {
4092 /* If the maximum is also zero, we used to just omit the group from the
4093 output altogether, like this:
4094
4095 ** if (repeat_max == 0)
4096 ** {
4097 ** code = previous;
4098 ** goto END_REPEAT;
4099 ** }
4100
4101 However, that fails when a group is referenced as a subroutine from
4102 elsewhere in the pattern, so now we stick in OP_SKIPZERO in front of it
4103 so that it is skipped on execution. As we don't have a list of which
4104 groups are referenced, we cannot do this selectively.
4105
4106 If the maximum is 1 or unlimited, we just have to stick in the BRAZERO
4107 and do no more at this point. However, we do need to adjust any
4108 OP_RECURSE calls inside the group that refer to the group itself or any
4109 internal or forward referenced group, because the offset is from the
4110 start of the whole regex. Temporarily terminate the pattern while doing
4111 this. */
4112
4113 if (repeat_max <= 1) /* Covers 0, 1, and unlimited */
4114 {
4115 *code = OP_END;
4116 adjust_recurse(previous, 1, utf8, cd, save_hwm);
4117 memmove(previous+1, previous, len);
4118 code++;
4119 if (repeat_max == 0)
4120 {
4121 *previous++ = OP_SKIPZERO;
4122 goto END_REPEAT;
4123 }
4124 *previous++ = OP_BRAZERO + repeat_type;
4125 }
4126
4127 /* If the maximum is greater than 1 and limited, we have to replicate
4128 in a nested fashion, sticking OP_BRAZERO before each set of brackets.
4129 The first one has to be handled carefully because it's the original
4130 copy, which has to be moved up. The remainder can be handled by code
4131 that is common with the non-zero minimum case below. We have to
4132 adjust the value or repeat_max, since one less copy is required. Once
4133 again, we may have to adjust any OP_RECURSE calls inside the group. */
4134
4135 else
4136 {
4137 int offset;
4138 *code = OP_END;
4139 adjust_recurse(previous, 2 + LINK_SIZE, utf8, cd, save_hwm);
4140 memmove(previous + 2 + LINK_SIZE, previous, len);
4141 code += 2 + LINK_SIZE;
4142 *previous++ = OP_BRAZERO + repeat_type;
4143 *previous++ = OP_BRA;
4144
4145 /* We chain together the bracket offset fields that have to be
4146 filled in later when the ends of the brackets are reached. */
4147
4148 offset = (bralink == NULL)? 0 : previous - bralink;
4149 bralink = previous;
4150 PUTINC(previous, 0, offset);
4151 }
4152
4153 repeat_max--;
4154 }
4155
4156 /* If the minimum is greater than zero, replicate the group as many
4157 times as necessary, and adjust the maximum to the number of subsequent
4158 copies that we need. If we set a first char from the group, and didn't
4159 set a required char, copy the latter from the former. If there are any
4160 forward reference subroutine calls in the group, there will be entries on
4161 the workspace list; replicate these with an appropriate increment. */
4162
4163 else
4164 {
4165 if (repeat_min > 1)
4166 {
4167 /* In the pre-compile phase, we don't actually do the replication. We
4168 just adjust the length as if we had. Do some paranoid checks for
4169 potential integer overflow. */
4170
4171 if (lengthptr != NULL)
4172 {
4173 int delta = (repeat_min - 1)*length_prevgroup;
4174 if ((double)(repeat_min - 1)*(double)length_prevgroup >
4175 (double)INT_MAX ||
4176 OFLOW_MAX - *lengthptr < delta)
4177 {
4178 *errorcodeptr = ERR20;
4179 goto FAILED;
4180 }
4181 *lengthptr += delta;
4182 }
4183
4184 /* This is compiling for real */
4185
4186 else
4187 {
4188 if (groupsetfirstbyte && reqbyte < 0) reqbyte = firstbyte;
4189 for (i = 1; i < repeat_min; i++)
4190 {
4191 uschar *hc;
4192 uschar *this_hwm = cd->hwm;
4193 memcpy(code, previous, len);
4194 for (hc = save_hwm; hc < this_hwm; hc += LINK_SIZE)
4195 {
4196 PUT(cd->hwm, 0, GET(hc, 0) + len);
4197 cd->hwm += LINK_SIZE;
4198 }
4199 save_hwm = this_hwm;
4200 code += len;
4201 }
4202 }
4203 }
4204
4205 if (repeat_max > 0) repeat_max -= repeat_min;
4206 }
4207
4208 /* This code is common to both the zero and non-zero minimum cases. If
4209 the maximum is limited, it replicates the group in a nested fashion,
4210 remembering the bracket starts on a stack. In the case of a zero minimum,
4211 the first one was set up above. In all cases the repeat_max now specifies
4212 the number of additional copies needed. Again, we must remember to
4213 replicate entries on the forward reference list. */
4214
4215 if (repeat_max >= 0)
4216 {
4217 /* In the pre-compile phase, we don't actually do the replication. We
4218 just adjust the length as if we had. For each repetition we must add 1
4219 to the length for BRAZERO and for all but the last repetition we must
4220 add 2 + 2*LINKSIZE to allow for the nesting that occurs. Do some
4221 paranoid checks to avoid integer overflow. */
4222
4223 if (lengthptr != NULL && repeat_max > 0)
4224 {
4225 int delta = repeat_max * (length_prevgroup + 1 + 2 + 2*LINK_SIZE) -
4226 2 - 2*LINK_SIZE; /* Last one doesn't nest */
4227 if ((double)repeat_max *
4228 (double)(length_prevgroup + 1 + 2 + 2*LINK_SIZE)
4229 > (double)INT_MAX ||
4230 OFLOW_MAX - *lengthptr < delta)
4231 {
4232 *errorcodeptr = ERR20;
4233 goto FAILED;
4234 }
4235 *lengthptr += delta;
4236 }
4237
4238 /* This is compiling for real */
4239
4240 else for (i = repeat_max - 1; i >= 0; i--)
4241 {
4242 uschar *hc;
4243 uschar *this_hwm = cd->hwm;
4244
4245 *code++ = OP_BRAZERO + repeat_type;
4246
4247 /* All but the final copy start a new nesting, maintaining the
4248 chain of brackets outstanding. */
4249
4250 if (i != 0)
4251 {
4252 int offset;
4253 *code++ = OP_BRA;
4254 offset = (bralink == NULL)? 0 : code - bralink;
4255 bralink = code;
4256 PUTINC(code, 0, offset);
4257 }
4258
4259 memcpy(code, previous, len);
4260 for (hc = save_hwm; hc < this_hwm; hc += LINK_SIZE)
4261 {
4262 PUT(cd->hwm, 0, GET(hc, 0) + len + ((i != 0)? 2+LINK_SIZE : 1));
4263 cd->hwm += LINK_SIZE;
4264 }
4265 save_hwm = this_hwm;
4266 code += len;
4267 }
4268
4269 /* Now chain through the pending brackets, and fill in their length
4270 fields (which are holding the chain links pro tem). */
4271
4272 while (bralink != NULL)
4273 {
4274 int oldlinkoffset;
4275 int offset = code - bralink + 1;
4276 uschar *bra = code - offset;
4277 oldlinkoffset = GET(bra, 1);
4278 bralink = (oldlinkoffset == 0)? NULL : bralink - oldlinkoffset;
4279 *code++ = OP_KET;
4280 PUTINC(code, 0, offset);
4281 PUT(bra, 1, offset);
4282 }
4283 }
4284
4285 /* If the maximum is unlimited, set a repeater in the final copy. We
4286 can't just offset backwards from the current code point, because we
4287 don't know if there's been an options resetting after the ket. The
4288 correct offset was computed above.
4289
4290 Then, when we are doing the actual compile phase, check to see whether
4291 this group is a non-atomic one that could match an empty string. If so,
4292 convert the initial operator to the S form (e.g. OP_BRA -> OP_SBRA) so
4293 that runtime checking can be done. [This check is also applied to
4294 atomic groups at runtime, but in a different way.] */
4295
4296 else
4297 {
4298 uschar *ketcode = code - ketoffset;
4299 uschar *bracode = ketcode - GET(ketcode, 1);
4300 *ketcode = OP_KETRMAX + repeat_type;
4301 if (lengthptr == NULL && *bracode != OP_ONCE)
4302 {
4303 uschar *scode = bracode;
4304 do
4305 {
4306 if (could_be_empty_branch(scode, ketcode, utf8))
4307 {
4308 *bracode += OP_SBRA - OP_BRA;
4309 break;
4310 }
4311 scode += GET(scode, 1);
4312 }
4313 while (*scode == OP_ALT);
4314 }
4315 }
4316 }
4317
4318 /* If previous is OP_FAIL, it was generated by an empty class [] in
4319 JavaScript mode. The other ways in which OP_FAIL can be generated, that is
4320 by (*FAIL) or (?!) set previous to NULL, which gives a "nothing to repeat"
4321 error above. We can just ignore the repeat in JS case. */
4322
4323 else if (*previous == OP_FAIL) goto END_REPEAT;
4324
4325 /* Else there's some kind of shambles */
4326
4327 else
4328 {
4329 *errorcodeptr = ERR11;
4330 goto FAILED;
4331 }
4332
4333 /* If the character following a repeat is '+', or if certain optimization
4334 tests above succeeded, possessive_quantifier is TRUE. For some of the
4335 simpler opcodes, there is an special alternative opcode for this. For
4336 anything else, we wrap the entire repeated item inside OP_ONCE brackets.
4337 The '+' notation is just syntactic sugar, taken from Sun's Java package,
4338 but the special opcodes can optimize it a bit. The repeated item starts at
4339 tempcode, not at previous, which might be the first part of a string whose
4340 (former) last char we repeated.
4341
4342 Possessifying an 'exact' quantifier has no effect, so we can ignore it. But
4343 an 'upto' may follow. We skip over an 'exact' item, and then test the
4344 length of what remains before proceeding. */
4345
4346 if (possessive_quantifier)
4347 {
4348 int len;
4349
4350 if (*tempcode == OP_TYPEEXACT)
4351 tempcode += _pcre_OP_lengths[*tempcode] +
4352 ((tempcode[3] == OP_PROP || tempcode[3] == OP_NOTPROP)? 2 : 0);
4353
4354 else if (*tempcode == OP_EXACT || *tempcode == OP_NOTEXACT)
4355 {
4356 tempcode += _pcre_OP_lengths[*tempcode];
4357 #ifdef SUPPORT_UTF8
4358 if (utf8 && tempcode[-1] >= 0xc0)
4359 tempcode += _pcre_utf8_table4[tempcode[-1] & 0x3f];
4360 #endif
4361 }
4362
4363 len = code - tempcode;
4364 if (len > 0) switch (*tempcode)
4365 {
4366 case OP_STAR: *tempcode = OP_POSSTAR; break;
4367 case OP_PLUS: *tempcode = OP_POSPLUS; break;
4368 case OP_QUERY: *tempcode = OP_POSQUERY; break;
4369 case OP_UPTO: *tempcode = OP_POSUPTO; break;
4370
4371 case OP_TYPESTAR: *tempcode = OP_TYPEPOSSTAR; break;
4372 case OP_TYPEPLUS: *tempcode = OP_TYPEPOSPLUS; break;
4373 case OP_TYPEQUERY: *tempcode = OP_TYPEPOSQUERY; break;
4374 case OP_TYPEUPTO: *tempcode = OP_TYPEPOSUPTO; break;
4375
4376 case OP_NOTSTAR: *tempcode = OP_NOTPOSSTAR; break;
4377 case OP_NOTPLUS: *tempcode = OP_NOTPOSPLUS; break;
4378 case OP_NOTQUERY: *tempcode = OP_NOTPOSQUERY; break;
4379 case OP_NOTUPTO: *tempcode = OP_NOTPOSUPTO; break;
4380
4381 default:
4382 memmove(tempcode + 1+LINK_SIZE, tempcode, len);
4383 code += 1 + LINK_SIZE;
4384 len += 1 + LINK_SIZE;
4385 tempcode[0] = OP_ONCE;
4386 *code++ = OP_KET;
4387 PUTINC(code, 0, len);
4388 PUT(tempcode, 1, len);
4389 break;
4390 }
4391 }
4392
4393 /* In all case we no longer have a previous item. We also set the
4394 "follows varying string" flag for subsequently encountered reqbytes if
4395 it isn't already set and we have just passed a varying length item. */
4396
4397 END_REPEAT:
4398 previous = NULL;
4399 cd->req_varyopt |= reqvary;
4400 break;
4401
4402
4403 /* ===================================================================*/
4404 /* Start of nested parenthesized sub-expression, or comment or lookahead or
4405 lookbehind or option setting or condition or all the other extended
4406 parenthesis forms. */
4407
4408 case CHAR_LEFT_PARENTHESIS:
4409 newoptions = options;
4410 skipbytes = 0;
4411 bravalue = OP_CBRA;
4412 save_hwm = cd->hwm;
4413 reset_bracount = FALSE;
4414
4415 /* First deal with various "verbs" that can be introduced by '*'. */
4416
4417 if (*(++ptr) == CHAR_ASTERISK && (cd->ctypes[ptr[1]] & ctype_letter) != 0)
4418 {
4419 int i, namelen;
4420 const char *vn = verbnames;
4421 const uschar *name = ++ptr;
4422 previous = NULL;
4423 while ((cd->ctypes[*++ptr] & ctype_letter) != 0) {};
4424 if (*ptr == CHAR_COLON)
4425 {
4426 *errorcodeptr = ERR59; /* Not supported */
4427 goto FAILED;
4428 }
4429 if (*ptr != CHAR_RIGHT_PARENTHESIS)
4430 {
4431 *errorcodeptr = ERR60;
4432 goto FAILED;
4433 }
4434 namelen = ptr - name;
4435 for (i = 0; i < verbcount; i++)
4436 {
4437 if (namelen == verbs[i].len &&
4438 strncmp((char *)name, vn, namelen) == 0)
4439 {
4440 *code = verbs[i].op;
4441 if (*code++ == OP_ACCEPT) cd->had_accept = TRUE;
4442 break;
4443 }
4444 vn += verbs[i].len + 1;
4445 }
4446 if (i < verbcount) continue;
4447 *errorcodeptr = ERR60;
4448 goto FAILED;
4449 }
4450
4451 /* Deal with the extended parentheses; all are introduced by '?', and the
4452 appearance of any of them means that this is not a capturing group. */
4453
4454 else if (*ptr == CHAR_QUESTION_MARK)
4455 {
4456 int i, set, unset, namelen;
4457 int *optset;
4458 const uschar *name;
4459 uschar *slot;
4460
4461 switch (*(++ptr))
4462 {
4463 case CHAR_NUMBER_SIGN: /* Comment; skip to ket */
4464 ptr++;
4465 while (*ptr != 0 && *ptr != CHAR_RIGHT_PARENTHESIS) ptr++;
4466 if (*ptr == 0)
4467 {
4468 *errorcodeptr = ERR18;
4469 goto FAILED;
4470 }
4471 continue;
4472
4473
4474 /* ------------------------------------------------------------ */
4475 case CHAR_VERTICAL_LINE: /* Reset capture count for each branch */
4476 reset_bracount = TRUE;
4477 /* Fall through */
4478
4479 /* ------------------------------------------------------------ */
4480 case CHAR_COLON: /* Non-capturing bracket */
4481 bravalue = OP_BRA;
4482 ptr++;
4483 break;
4484
4485
4486 /* ------------------------------------------------------------ */
4487 case CHAR_LEFT_PARENTHESIS:
4488 bravalue = OP_COND; /* Conditional group */
4489
4490 /* A condition can be an assertion, a number (referring to a numbered
4491 group), a name (referring to a named group), or 'R', referring to
4492 recursion. R<digits> and R&name are also permitted for recursion tests.
4493
4494 There are several syntaxes for testing a named group: (?(name)) is used
4495 by Python; Perl 5.10 onwards uses (?(<name>) or (?('name')).
4496
4497 There are two unfortunate ambiguities, caused by history. (a) 'R' can
4498 be the recursive thing or the name 'R' (and similarly for 'R' followed
4499 by digits), and (b) a number could be a name that consists of digits.
4500 In both cases, we look for a name first; if not found, we try the other
4501 cases. */
4502
4503 /* For conditions that are assertions, check the syntax, and then exit
4504 the switch. This will take control down to where bracketed groups,
4505 including assertions, are processed. */
4506
4507 if (ptr[1] == CHAR_QUESTION_MARK && (ptr[2] == CHAR_EQUALS_SIGN ||
4508 ptr[2] == CHAR_EXCLAMATION_MARK || ptr[2] == CHAR_LESS_THAN_SIGN))
4509 break;
4510
4511 /* Most other conditions use OP_CREF (a couple change to OP_RREF
4512 below), and all need to skip 3 bytes at the start of the group. */
4513
4514 code[1+LINK_SIZE] = OP_CREF;
4515 skipbytes = 3;
4516 refsign = -1;
4517
4518 /* Check for a test for recursion in a named group. */
4519
4520 if (ptr[1] == CHAR_R && ptr[2] == CHAR_AMPERSAND)
4521 {
4522 terminator = -1;
4523 ptr += 2;
4524 code[1+LINK_SIZE] = OP_RREF; /* Change the type of test */
4525 }
4526
4527 /* Check for a test for a named group's having been set, using the Perl
4528 syntax (?(<name>) or (?('name') */
4529
4530 else if (ptr[1] == CHAR_LESS_THAN_SIGN)
4531 {
4532 terminator = CHAR_GREATER_THAN_SIGN;
4533 ptr++;
4534 }
4535 else if (ptr[1] == CHAR_APOSTROPHE)
4536 {
4537 terminator = CHAR_APOSTROPHE;
4538 ptr++;
4539 }
4540 else
4541 {
4542 terminator = 0;
4543 if (ptr[1] == CHAR_MINUS || ptr[1] == CHAR_PLUS) refsign = *(++ptr);
4544 }
4545
4546 /* We now expect to read a name; any thing else is an error */
4547
4548 if ((cd->ctypes[ptr[1]] & ctype_word) == 0)
4549 {
4550 ptr += 1; /* To get the right offset */
4551 *errorcodeptr = ERR28;
4552 goto FAILED;
4553 }
4554
4555 /* Read the name, but also get it as a number if it's all digits */
4556
4557 recno = 0;
4558 name = ++ptr;
4559 while ((cd->ctypes[*ptr] & ctype_word) != 0)
4560 {
4561 if (recno >= 0)
4562 recno = ((digitab[*ptr] & ctype_digit) != 0)?
4563 recno * 10 + *ptr - CHAR_0 : -1;
4564 ptr++;
4565 }
4566 namelen = ptr - name;
4567
4568 if ((terminator > 0 && *ptr++ != terminator) ||
4569 *ptr++ != CHAR_RIGHT_PARENTHESIS)
4570 {
4571 ptr--; /* Error offset */
4572 *errorcodeptr = ERR26;
4573 goto FAILED;
4574 }
4575
4576 /* Do no further checking in the pre-compile phase. */
4577
4578 if (lengthptr != NULL) break;
4579
4580 /* In the real compile we do the work of looking for the actual
4581 reference. If the string started with "+" or "-" we require the rest to
4582 be digits, in which case recno will be set. */
4583
4584 if (refsign > 0)
4585 {
4586 if (recno <= 0)
4587 {
4588 *errorcodeptr = ERR58;
4589 goto FAILED;
4590 }
4591 recno = (refsign == CHAR_MINUS)?
4592 cd->bracount - recno + 1 : recno +cd->bracount;
4593 if (recno <= 0 || recno > cd->final_bracount)
4594 {
4595 *errorcodeptr = ERR15;
4596 goto FAILED;
4597 }
4598 PUT2(code, 2+LINK_SIZE, recno);
4599 break;
4600 }
4601
4602 /* Otherwise (did not start with "+" or "-"), start by looking for the
4603 name. */
4604
4605 slot = cd->name_table;
4606 for (i = 0; i < cd->names_found; i++)
4607 {
4608 if (strncmp((char *)name, (char *)slot+2, namelen) == 0) break;
4609 slot += cd->name_entry_size;
4610 }
4611
4612 /* Found a previous named subpattern */
4613
4614 if (i < cd->names_found)
4615 {
4616 recno = GET2(slot, 0);
4617 PUT2(code, 2+LINK_SIZE, recno);
4618 }
4619
4620 /* Search the pattern for a forward reference */
4621
4622 else if ((i = find_parens(cd, name, namelen,
4623 (options & PCRE_EXTENDED) != 0)) > 0)
4624 {
4625 PUT2(code, 2+LINK_SIZE, i);
4626 }
4627
4628 /* If terminator == 0 it means that the name followed directly after
4629 the opening parenthesis [e.g. (?(abc)...] and in this case there are
4630 some further alternatives to try. For the cases where terminator != 0
4631 [things like (?(<name>... or (?('name')... or (?(R&name)... ] we have
4632 now checked all the possibilities, so give an error. */
4633
4634 else if (terminator != 0)
4635 {
4636 *errorcodeptr = ERR15;
4637 goto FAILED;
4638 }
4639
4640 /* Check for (?(R) for recursion. Allow digits after R to specify a
4641 specific group number. */
4642
4643 else if (*name == CHAR_R)
4644 {
4645 recno = 0;
4646 for (i = 1; i < namelen; i++)
4647 {
4648 if ((digitab[name[i]] & ctype_digit) == 0)
4649 {
4650 *errorcodeptr = ERR15;
4651 goto FAILED;
4652 }
4653 recno = recno * 10 + name[i] - CHAR_0;
4654 }
4655 if (recno == 0) recno = RREF_ANY;
4656 code[1+LINK_SIZE] = OP_RREF; /* Change test type */
4657 PUT2(code, 2+LINK_SIZE, recno);
4658 }
4659
4660 /* Similarly, check for the (?(DEFINE) "condition", which is always
4661 false. */
4662
4663 else if (namelen == 6 && strncmp((char *)name, STRING_DEFINE, 6) == 0)
4664 {
4665 code[1+LINK_SIZE] = OP_DEF;
4666 skipbytes = 1;
4667 }
4668
4669 /* Check for the "name" actually being a subpattern number. We are
4670 in the second pass here, so final_bracount is set. */
4671
4672 else if (recno > 0 && recno <= cd->final_bracount)
4673 {
4674 PUT2(code, 2+LINK_SIZE, recno);
4675 }
4676
4677 /* Either an unidentified subpattern, or a reference to (?(0) */
4678
4679 else
4680 {
4681 *errorcodeptr = (recno == 0)? ERR35: ERR15;
4682 goto FAILED;
4683 }
4684 break;
4685
4686
4687 /* ------------------------------------------------------------ */
4688 case CHAR_EQUALS_SIGN: /* Positive lookahead */
4689 bravalue = OP_ASSERT;
4690 ptr++;
4691 break;
4692
4693
4694 /* ------------------------------------------------------------ */
4695 case CHAR_EXCLAMATION_MARK: /* Negative lookahead */
4696 ptr++;
4697 if (*ptr == CHAR_RIGHT_PARENTHESIS) /* Optimize (?!) */
4698 {
4699 *code++ = OP_FAIL;
4700 previous = NULL;
4701 continue;
4702 }
4703 bravalue = OP_ASSERT_NOT;
4704 break;
4705
4706
4707 /* ------------------------------------------------------------ */
4708 case CHAR_LESS_THAN_SIGN: /* Lookbehind or named define */
4709 switch (ptr[1])
4710 {
4711 case CHAR_EQUALS_SIGN: /* Positive lookbehind */
4712 bravalue = OP_ASSERTBACK;
4713 ptr += 2;
4714 break;
4715
4716 case CHAR_EXCLAMATION_MARK: /* Negative lookbehind */
4717 bravalue = OP_ASSERTBACK_NOT;
4718 ptr += 2;
4719 break;
4720
4721 default: /* Could be name define, else bad */
4722 if ((cd->ctypes[ptr[1]] & ctype_word) != 0) goto DEFINE_NAME;
4723 ptr++; /* Correct offset for error */
4724 *errorcodeptr = ERR24;
4725 goto FAILED;
4726 }
4727 break;
4728
4729
4730 /* ------------------------------------------------------------ */
4731 case CHAR_GREATER_THAN_SIGN: /* One-time brackets */
4732 bravalue = OP_ONCE;
4733 ptr++;
4734 break;
4735
4736
4737 /* ------------------------------------------------------------ */
4738 case CHAR_C: /* Callout - may be followed by digits; */
4739 previous_callout = code; /* Save for later completion */
4740 after_manual_callout = 1; /* Skip one item before completing */
4741 *code++ = OP_CALLOUT;
4742 {
4743 int n = 0;
4744 while ((digitab[*(++ptr)] & ctype_digit) != 0)
4745 n = n * 10 + *ptr - CHAR_0;
4746 if (*ptr != CHAR_RIGHT_PARENTHESIS)
4747 {
4748 *errorcodeptr = ERR39;
4749 goto FAILED;
4750 }
4751 if (n > 255)
4752 {
4753 *errorcodeptr = ERR38;
4754 goto FAILED;
4755 }
4756 *code++ = n;
4757 PUT(code, 0, ptr - cd->start_pattern + 1); /* Pattern offset */
4758 PUT(code, LINK_SIZE, 0); /* Default length */
4759 code += 2 * LINK_SIZE;
4760 }
4761 previous = NULL;
4762 continue;
4763
4764
4765 /* ------------------------------------------------------------ */
4766 case CHAR_P: /* Python-style named subpattern handling */
4767 if (*(++ptr) == CHAR_EQUALS_SIGN ||
4768 *ptr == CHAR_GREATER_THAN_SIGN) /* Reference or recursion */
4769 {
4770 is_recurse = *ptr == CHAR_GREATER_THAN_SIGN;
4771 terminator = CHAR_RIGHT_PARENTHESIS;
4772 goto NAMED_REF_OR_RECURSE;
4773 }
4774 else if (*ptr != CHAR_LESS_THAN_SIGN) /* Test for Python-style defn */
4775 {
4776 *errorcodeptr = ERR41;
4777 goto FAILED;
4778 }
4779 /* Fall through to handle (?P< as (?< is handled */
4780
4781
4782 /* ------------------------------------------------------------ */
4783 DEFINE_NAME: /* Come here from (?< handling */
4784 case CHAR_APOSTROPHE:
4785 {
4786 terminator = (*ptr == CHAR_LESS_THAN_SIGN)?
4787 CHAR_GREATER_THAN_SIGN : CHAR_APOSTROPHE;
4788 name = ++ptr;
4789
4790 while ((cd->ctypes[*ptr] & ctype_word) != 0) ptr++;
4791 namelen = ptr - name;
4792
4793 /* In the pre-compile phase, just do a syntax check. */
4794
4795 if (lengthptr != NULL)
4796 {
4797 if (*ptr != terminator)
4798 {
4799 *errorcodeptr = ERR42;
4800 goto FAILED;
4801 }
4802 if (cd->names_found >= MAX_NAME_COUNT)
4803 {
4804 *errorcodeptr = ERR49;
4805 goto FAILED;
4806 }
4807 if (namelen + 3 > cd->name_entry_size)
4808 {
4809 cd->name_entry_size = namelen + 3;
4810 if (namelen > MAX_NAME_SIZE)
4811 {
4812 *errorcodeptr = ERR48;
4813 goto FAILED;
4814 }
4815 }
4816 }
4817
4818 /* In the real compile, create the entry in the table */
4819
4820 else
4821 {
4822 slot = cd->name_table;
4823 for (i = 0; i < cd->names_found; i++)
4824 {
4825 int crc = memcmp(name, slot+2, namelen);
4826 if (crc == 0)
4827 {
4828 if (slot[2+namelen] == 0)
4829 {
4830 if ((options & PCRE_DUPNAMES) == 0)
4831 {
4832 *errorcodeptr = ERR43;
4833 goto FAILED;
4834 }
4835 }
4836 else crc = -1; /* Current name is substring */
4837 }
4838 if (crc < 0)
4839 {
4840 memmove(slot + cd->name_entry_size, slot,
4841 (cd->names_found - i) * cd->name_entry_size);
4842 break;
4843 }
4844 slot += cd->name_entry_size;
4845 }
4846
4847 PUT2(slot, 0, cd->bracount + 1);
4848 memcpy(slot + 2, name, namelen);
4849 slot[2+namelen] = 0;
4850 }
4851 }
4852
4853 /* In both cases, count the number of names we've encountered. */
4854
4855 ptr++; /* Move past > or ' */
4856 cd->names_found++;
4857 goto NUMBERED_GROUP;
4858
4859
4860 /* ------------------------------------------------------------ */
4861 case CHAR_AMPERSAND: /* Perl recursion/subroutine syntax */
4862 terminator = CHAR_RIGHT_PARENTHESIS;
4863 is_recurse = TRUE;
4864 /* Fall through */
4865
4866 /* We come here from the Python syntax above that handles both
4867 references (?P=name) and recursion (?P>name), as well as falling
4868 through from the Perl recursion syntax (?&name). We also come here from
4869 the Perl \k<name> or \k'name' back reference syntax and the \k{name}
4870 .NET syntax, and the Oniguruma \g<...> and \g'...' subroutine syntax. */
4871
4872 NAMED_REF_OR_RECURSE:
4873 name = ++ptr;
4874 while ((cd->ctypes[*ptr] & ctype_word) != 0) ptr++;
4875 namelen = ptr - name;
4876
4877 /* In the pre-compile phase, do a syntax check and set a dummy
4878 reference number. */
4879
4880 if (lengthptr != NULL)
4881 {
4882 if (namelen == 0)
4883 {
4884 *errorcodeptr = ERR62;
4885 goto FAILED;
4886 }
4887 if (*ptr != terminator)
4888 {
4889 *errorcodeptr = ERR42;
4890 goto FAILED;
4891 }
4892 if (namelen > MAX_NAME_SIZE)
4893 {
4894 *errorcodeptr = ERR48;
4895 goto FAILED;
4896 }
4897 recno = 0;
4898 }
4899
4900 /* In the real compile, seek the name in the table. We check the name
4901 first, and then check that we have reached the end of the name in the
4902 table. That way, if the name that is longer than any in the table,
4903 the comparison will fail without reading beyond the table entry. */
4904
4905 else
4906 {
4907 slot = cd->name_table;
4908 for (i = 0; i < cd->names_found; i++)
4909 {
4910 if (strncmp((char *)name, (char *)slot+2, namelen) == 0 &&
4911 slot[2+namelen] == 0)
4912 break;
4913 slot += cd->name_entry_size;
4914 }
4915
4916 if (i < cd->names_found) /* Back reference */
4917 {
4918 recno = GET2(slot, 0);
4919 }
4920 else if ((recno = /* Forward back reference */
4921 find_parens(cd, name, namelen,
4922 (options & PCRE_EXTENDED) != 0)) <= 0)
4923 {
4924 *errorcodeptr = ERR15;
4925 goto FAILED;
4926 }
4927 }
4928
4929 /* In both phases, we can now go to the code than handles numerical
4930 recursion or backreferences. */
4931
4932 if (is_recurse) goto HANDLE_RECURSION;
4933 else goto HANDLE_REFERENCE;
4934
4935
4936 /* ------------------------------------------------------------ */
4937 case CHAR_R: /* Recursion */
4938 ptr++; /* Same as (?0) */
4939 /* Fall through */
4940
4941
4942 /* ------------------------------------------------------------ */
4943 case CHAR_MINUS: case CHAR_PLUS: /* Recursion or subroutine */
4944 case CHAR_0: case CHAR_1: case CHAR_2: case CHAR_3: case CHAR_4:
4945 case CHAR_5: case CHAR_6: case CHAR_7: case CHAR_8: case CHAR_9:
4946 {
4947 const uschar *called;
4948 terminator = CHAR_RIGHT_PARENTHESIS;
4949
4950 /* Come here from the \g<...> and \g'...' code (Oniguruma
4951 compatibility). However, the syntax has been checked to ensure that
4952 the ... are a (signed) number, so that neither ERR63 nor ERR29 will
4953 be called on this path, nor with the jump to OTHER_CHAR_AFTER_QUERY
4954 ever be taken. */
4955
4956 HANDLE_NUMERICAL_RECURSION:
4957
4958 if ((refsign = *ptr) == CHAR_PLUS)
4959 {
4960 ptr++;
4961 if ((digitab[*ptr] & ctype_digit) == 0)
4962 {
4963 *errorcodeptr = ERR63;
4964 goto FAILED;
4965 }
4966 }
4967 else if (refsign == CHAR_MINUS)
4968 {
4969 if ((digitab[ptr[1]] & ctype_digit) == 0)
4970 goto OTHER_CHAR_AFTER_QUERY;
4971 ptr++;
4972 }
4973
4974 recno = 0;
4975 while((digitab[*ptr] & ctype_digit) != 0)
4976 recno = recno * 10 + *ptr++ - CHAR_0;
4977
4978 if (*ptr != terminator)
4979 {
4980 *errorcodeptr = ERR29;
4981 goto FAILED;
4982 }
4983
4984 if (refsign == CHAR_MINUS)
4985 {
4986 if (recno == 0)
4987 {
4988 *errorcodeptr = ERR58;
4989 goto FAILED;
4990 }
4991 recno = cd->bracount - recno + 1;
4992 if (recno <= 0)
4993 {
4994 *errorcodeptr = ERR15;
4995 goto FAILED;
4996 }
4997 }
4998 else if (refsign == CHAR_PLUS)
4999 {
5000 if (recno == 0)
5001 {
5002 *errorcodeptr = ERR58;
5003 goto FAILED;
5004 }
5005 recno += cd->bracount;
5006 }
5007
5008 /* Come here from code above that handles a named recursion */
5009
5010 HANDLE_RECURSION:
5011
5012 previous = code;
5013 called = cd->start_code;
5014
5015 /* When we are actually compiling, find the bracket that is being
5016 referenced. Temporarily end the regex in case it doesn't exist before
5017 this point. If we end up with a forward reference, first check that
5018 the bracket does occur later so we can give the error (and position)
5019 now. Then remember this forward reference in the workspace so it can
5020 be filled in at the end. */
5021
5022 if (lengthptr == NULL)
5023 {
5024 *code = OP_END;
5025 if (recno != 0) called = find_bracket(cd->start_code, utf8, recno);
5026
5027 /* Forward reference */
5028
5029 if (called == NULL)
5030 {
5031 if (find_parens(cd, NULL, recno,
5032 (options & PCRE_EXTENDED) != 0) < 0)
5033 {
5034 *errorcodeptr = ERR15;
5035 goto FAILED;
5036 }
5037 called = cd->start_code + recno;
5038 PUTINC(cd->hwm, 0, code + 2 + LINK_SIZE - cd->start_code);
5039 }
5040
5041 /* If not a forward reference, and the subpattern is still open,
5042 this is a recursive call. We check to see if this is a left
5043 recursion that could loop for ever, and diagnose that case. */
5044
5045 else if (GET(called, 1) == 0 &&
5046 could_be_empty(called, code, bcptr, utf8))
5047 {
5048 *errorcodeptr = ERR40;
5049 goto FAILED;
5050 }
5051 }
5052
5053 /* Insert the recursion/subroutine item, automatically wrapped inside
5054 "once" brackets. Set up a "previous group" length so that a
5055 subsequent quantifier will work. */
5056
5057 *code = OP_ONCE;
5058 PUT(code, 1, 2 + 2*LINK_SIZE);
5059 code += 1 + LINK_SIZE;
5060
5061 *code = OP_RECURSE;
5062 PUT(code, 1, called - cd->start_code);
5063 code += 1 + LINK_SIZE;
5064
5065 *code = OP_KET;
5066 PUT(code, 1, 2 + 2*LINK_SIZE);
5067 code += 1 + LINK_SIZE;
5068
5069 length_prevgroup = 3 + 3*LINK_SIZE;
5070 }
5071
5072 /* Can't determine a first byte now */
5073
5074 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
5075 continue;
5076
5077
5078 /* ------------------------------------------------------------ */
5079 default: /* Other characters: check option setting */
5080 OTHER_CHAR_AFTER_QUERY:
5081 set = unset = 0;
5082 optset = &set;
5083
5084 while (*ptr != CHAR_RIGHT_PARENTHESIS && *ptr != CHAR_COLON)
5085 {
5086 switch (*ptr++)
5087 {
5088 case CHAR_MINUS: optset = &unset; break;
5089
5090 case CHAR_J: /* Record that it changed in the external options */
5091 *optset |= PCRE_DUPNAMES;
5092 cd->external_flags |= PCRE_JCHANGED;
5093 break;
5094
5095 case CHAR_i: *optset |= PCRE_CASELESS; break;
5096 case CHAR_m: *optset |= PCRE_MULTILINE; break;
5097 case CHAR_s: *optset |= PCRE_DOTALL; break;
5098 case CHAR_x: *optset |= PCRE_EXTENDED; break;
5099 case CHAR_U: *optset |= PCRE_UNGREEDY; break;
5100 case CHAR_X: *optset |= PCRE_EXTRA; break;
5101
5102 default: *errorcodeptr = ERR12;
5103 ptr--; /* Correct the offset */
5104 goto FAILED;
5105 }
5106 }
5107
5108 /* Set up the changed option bits, but don't change anything yet. */
5109
5110 newoptions = (options | set) & (~unset);
5111
5112 /* If the options ended with ')' this is not the start of a nested
5113 group with option changes, so the options change at this level. If this
5114 item is right at the start of the pattern, the options can be
5115 abstracted and made external in the pre-compile phase, and ignored in
5116 the compile phase. This can be helpful when matching -- for instance in
5117 caseless checking of required bytes.
5118
5119 If the code pointer is not (cd->start_code + 1 + LINK_SIZE), we are
5120 definitely *not* at the start of the pattern because something has been
5121 compiled. In the pre-compile phase, however, the code pointer can have
5122 that value after the start, because it gets reset as code is discarded
5123 during the pre-compile. However, this can happen only at top level - if
5124 we are within parentheses, the starting BRA will still be present. At
5125 any parenthesis level, the length value can be used to test if anything
5126 has been compiled at that level. Thus, a test for both these conditions
5127 is necessary to ensure we correctly detect the start of the pattern in
5128 both phases.
5129
5130 If we are not at the pattern start, compile code to change the ims
5131 options if this setting actually changes any of them, and reset the
5132 greedy defaults and the case value for firstbyte and reqbyte. */
5133
5134 if (*ptr == CHAR_RIGHT_PARENTHESIS)
5135 {
5136 if (code == cd->start_code + 1 + LINK_SIZE &&
5137 (lengthptr == NULL || *lengthptr == 2 + 2*LINK_SIZE))
5138 {
5139 cd->external_options = newoptions;
5140 }
5141 else
5142 {
5143 if ((options & PCRE_IMS) != (newoptions & PCRE_IMS))
5144 {
5145 *code++ = OP_OPT;
5146 *code++ = newoptions & PCRE_IMS;
5147 }
5148 greedy_default = ((newoptions & PCRE_UNGREEDY) != 0);
5149 greedy_non_default = greedy_default ^ 1;
5150 req_caseopt = ((newoptions & PCRE_CASELESS) != 0)? REQ_CASELESS : 0;
5151 }
5152
5153 /* Change options at this level, and pass them back for use
5154 in subsequent branches. When not at the start of the pattern, this
5155 information is also necessary so that a resetting item can be
5156 compiled at the end of a group (if we are in a group). */
5157
5158 *optionsptr = options = newoptions;
5159 previous = NULL; /* This item can't be repeated */
5160 continue; /* It is complete */
5161 }
5162
5163 /* If the options ended with ':' we are heading into a nested group
5164 with possible change of options. Such groups are non-capturing and are
5165 not assertions of any kind. All we need to do is skip over the ':';
5166 the newoptions value is handled below. */
5167
5168 bravalue = OP_BRA;
5169 ptr++;
5170 } /* End of switch for character following (? */
5171 } /* End of (? handling */
5172
5173 /* Opening parenthesis not followed by '?'. If PCRE_NO_AUTO_CAPTURE is set,
5174 all unadorned brackets become non-capturing and behave like (?:...)
5175 brackets. */
5176
5177 else if ((options & PCRE_NO_AUTO_CAPTURE) != 0)
5178 {
5179 bravalue = OP_BRA;
5180 }
5181
5182 /* Else we have a capturing group. */
5183
5184 else
5185 {
5186 NUMBERED_GROUP:
5187 cd->bracount += 1;
5188 PUT2(code, 1+LINK_SIZE, cd->bracount);
5189 skipbytes = 2;
5190 }
5191
5192 /* Process nested bracketed regex. Assertions may not be repeated, but
5193 other kinds can be. All their opcodes are >= OP_ONCE. We copy code into a
5194 non-register variable in order to be able to pass its address because some
5195 compilers complain otherwise. Pass in a new setting for the ims options if
5196 they have changed. */
5197
5198 previous = (bravalue >= OP_ONCE)? code : NULL;
5199 *code = bravalue;
5200 tempcode = code;
5201 tempreqvary = cd->req_varyopt; /* Save value before bracket */
5202 length_prevgroup = 0; /* Initialize for pre-compile phase */
5203
5204 if (!compile_regex(
5205 newoptions, /* The complete new option state */
5206 options & PCRE_IMS, /* The previous ims option state */
5207 &tempcode, /* Where to put code (updated) */
5208 &ptr, /* Input pointer (updated) */
5209 errorcodeptr, /* Where to put an error message */
5210 (bravalue == OP_ASSERTBACK ||
5211 bravalue == OP_ASSERTBACK_NOT), /* TRUE if back assert */
5212 reset_bracount, /* True if (?| group */
5213 skipbytes, /* Skip over bracket number */
5214 &subfirstbyte, /* For possible first char */
5215 &subreqbyte, /* For possible last char */
5216 bcptr, /* Current branch chain */
5217 cd, /* Tables block */
5218 (lengthptr == NULL)? NULL : /* Actual compile phase */
5219 &length_prevgroup /* Pre-compile phase */
5220 ))
5221 goto FAILED;
5222
5223 /* At the end of compiling, code is still pointing to the start of the
5224 group, while tempcode has been updated to point past the end of the group
5225 and any option resetting that may follow it. The pattern pointer (ptr)
5226 is on the bracket. */
5227
5228 /* If this is a conditional bracket, check that there are no more than
5229 two branches in the group, or just one if it's a DEFINE group. We do this
5230 in the real compile phase, not in the pre-pass, where the whole group may
5231 not be available. */
5232
5233 if (bravalue == OP_COND && lengthptr == NULL)
5234 {
5235 uschar *tc = code;
5236 int condcount = 0;
5237
5238 do {
5239 condcount++;
5240 tc += GET(tc,1);
5241 }
5242 while (*tc != OP_KET);
5243
5244 /* A DEFINE group is never obeyed inline (the "condition" is always
5245 false). It must have only one branch. */
5246
5247 if (code[LINK_SIZE+1] == OP_DEF)
5248 {
5249 if (condcount > 1)
5250 {
5251 *errorcodeptr = ERR54;
5252 goto FAILED;
5253 }
5254 bravalue = OP_DEF; /* Just a flag to suppress char handling below */
5255 }
5256
5257 /* A "normal" conditional group. If there is just one branch, we must not
5258 make use of its firstbyte or reqbyte, because this is equivalent to an
5259 empty second branch. */
5260
5261 else
5262 {
5263 if (condcount > 2)
5264 {
5265 *errorcodeptr = ERR27;
5266 goto FAILED;
5267 }
5268 if (condcount == 1) subfirstbyte = subreqbyte = REQ_NONE;
5269 }
5270 }
5271
5272 /* Error if hit end of pattern */
5273
5274 if (*ptr != CHAR_RIGHT_PARENTHESIS)
5275 {
5276 *errorcodeptr = ERR14;
5277 goto FAILED;
5278 }
5279
5280 /* In the pre-compile phase, update the length by the length of the group,
5281 less the brackets at either end. Then reduce the compiled code to just a
5282 set of non-capturing brackets so that it doesn't use much memory if it is
5283 duplicated by a quantifier.*/
5284
5285 if (lengthptr != NULL)
5286 {
5287 if (OFLOW_MAX - *lengthptr < length_prevgroup - 2 - 2*LINK_SIZE)
5288 {
5289 *errorcodeptr = ERR20;
5290 goto FAILED;
5291 }
5292 *lengthptr += length_prevgroup - 2 - 2*LINK_SIZE;
5293 *code++ = OP_BRA;
5294 PUTINC(code, 0, 1 + LINK_SIZE);
5295 *code++ = OP_KET;
5296 PUTINC(code, 0, 1 + LINK_SIZE);
5297 break; /* No need to waste time with special character handling */
5298 }
5299
5300 /* Otherwise update the main code pointer to the end of the group. */
5301
5302 code = tempcode;
5303
5304 /* For a DEFINE group, required and first character settings are not
5305 relevant. */
5306
5307 if (bravalue == OP_DEF) break;
5308
5309 /* Handle updating of the required and first characters for other types of
5310 group. Update for normal brackets of all kinds, and conditions with two
5311 branches (see code above). If the bracket is followed by a quantifier with
5312 zero repeat, we have to back off. Hence the definition of zeroreqbyte and
5313 zerofirstbyte outside the main loop so that they can be accessed for the
5314 back off. */
5315
5316 zeroreqbyte = reqbyte;
5317 zerofirstbyte = firstbyte;
5318 groupsetfirstbyte = FALSE;
5319
5320 if (bravalue >= OP_ONCE)
5321 {
5322 /* If we have not yet set a firstbyte in this branch, take it from the
5323 subpattern, remembering that it was set here so that a repeat of more
5324 than one can replicate it as reqbyte if necessary. If the subpattern has
5325 no firstbyte, set "none" for the whole branch. In both cases, a zero
5326 repeat forces firstbyte to "none". */
5327
5328 if (firstbyte == REQ_UNSET)
5329 {
5330 if (subfirstbyte >= 0)
5331 {
5332 firstbyte = subfirstbyte;
5333 groupsetfirstbyte = TRUE;
5334 }
5335 else firstbyte = REQ_NONE;
5336 zerofirstbyte = REQ_NONE;
5337 }
5338
5339 /* If firstbyte was previously set, convert the subpattern's firstbyte
5340 into reqbyte if there wasn't one, using the vary flag that was in
5341 existence beforehand. */
5342
5343 else if (subfirstbyte >= 0 && subreqbyte < 0)
5344 subreqbyte = subfirstbyte | tempreqvary;
5345
5346 /* If the subpattern set a required byte (or set a first byte that isn't
5347 really the first byte - see above), set it. */
5348
5349 if (subreqbyte >= 0) reqbyte = subreqbyte;
5350 }
5351
5352 /* For a forward assertion, we take the reqbyte, if set. This can be
5353 helpful if the pattern that follows the assertion doesn't set a different
5354 char. For example, it's useful for /(?=abcde).+/. We can't set firstbyte
5355 for an assertion, however because it leads to incorrect effect for patterns
5356 such as /(?=a)a.+/ when the "real" "a" would then become a reqbyte instead
5357 of a firstbyte. This is overcome by a scan at the end if there's no
5358 firstbyte, looking for an asserted first char. */
5359
5360 else if (bravalue == OP_ASSERT && subreqbyte >= 0) reqbyte = subreqbyte;
5361 break; /* End of processing '(' */
5362
5363
5364 /* ===================================================================*/
5365 /* Handle metasequences introduced by \. For ones like \d, the ESC_ values
5366 are arranged to be the negation of the corresponding OP_values. For the
5367 back references, the values are ESC_REF plus the reference number. Only
5368 back references and those types that consume a character may be repeated.
5369 We can test for values between ESC_b and ESC_Z for the latter; this may
5370 have to change if any new ones are ever created. */
5371
5372 case CHAR_BACKSLASH:
5373 tempptr = ptr;
5374 c = check_escape(&ptr, errorcodeptr, cd->bracount, options, FALSE);
5375 if (*errorcodeptr != 0) goto FAILED;
5376
5377 if (c < 0)
5378 {
5379 if (-c == ESC_Q) /* Handle start of quoted string */
5380 {
5381 if (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
5382 ptr += 2; /* avoid empty string */
5383 else inescq = TRUE;
5384 continue;
5385 }
5386
5387 if (-c == ESC_E) continue; /* Perl ignores an orphan \E */
5388
5389 /* For metasequences that actually match a character, we disable the
5390 setting of a first character if it hasn't already been set. */
5391
5392 if (firstbyte == REQ_UNSET && -c > ESC_b && -c < ESC_Z)
5393 firstbyte = REQ_NONE;
5394
5395 /* Set values to reset to if this is followed by a zero repeat. */
5396
5397 zerofirstbyte = firstbyte;
5398 zeroreqbyte = reqbyte;
5399
5400 /* \g<name> or \g'name' is a subroutine call by name and \g<n> or \g'n'
5401 is a subroutine call by number (Oniguruma syntax). In fact, the value
5402 -ESC_g is returned only for these cases. So we don't need to check for <
5403 or ' if the value is -ESC_g. For the Perl syntax \g{n} the value is
5404 -ESC_REF+n, and for the Perl syntax \g{name} the result is -ESC_k (as
5405 that is a synonym for a named back reference). */
5406
5407 if (-c == ESC_g)
5408 {
5409 const uschar *p;
5410 save_hwm = cd->hwm; /* Normally this is set when '(' is read */
5411 terminator = (*(++ptr) == CHAR_LESS_THAN_SIGN)?
5412 CHAR_GREATER_THAN_SIGN : CHAR_APOSTROPHE;
5413
5414 /* These two statements stop the compiler for warning about possibly
5415 unset variables caused by the jump to HANDLE_NUMERICAL_RECURSION. In
5416 fact, because we actually check for a number below, the paths that
5417 would actually be in error are never taken. */
5418
5419 skipbytes = 0;
5420 reset_bracount = FALSE;
5421
5422 /* Test for a name */
5423
5424 if (ptr[1] != CHAR_PLUS && ptr[1] != CHAR_MINUS)
5425 {
5426 BOOL isnumber = TRUE;
5427 for (p = ptr + 1; *p != 0 && *p != terminator; p++)
5428 {
5429 if ((cd->ctypes[*p] & ctype_digit) == 0) isnumber = FALSE;
5430 if ((cd->ctypes[*p] & ctype_word) == 0) break;
5431 }
5432 if (*p != terminator)
5433 {
5434 *errorcodeptr = ERR57;
5435 break;
5436 }
5437 if (isnumber)
5438 {
5439 ptr++;
5440 goto HANDLE_NUMERICAL_RECURSION;
5441 }
5442 is_recurse = TRUE;
5443 goto NAMED_REF_OR_RECURSE;
5444 }
5445
5446 /* Test a signed number in angle brackets or quotes. */
5447
5448 p = ptr + 2;
5449 while ((digitab[*p] & ctype_digit) != 0) p++;
5450 if (*p != terminator)
5451 {
5452 *errorcodeptr = ERR57;
5453 break;
5454 }
5455 ptr++;
5456 goto HANDLE_NUMERICAL_RECURSION;
5457 }
5458
5459 /* \k<name> or \k'name' is a back reference by name (Perl syntax).
5460 We also support \k{name} (.NET syntax) */
5461
5462 if (-c == ESC_k && (ptr[1] == CHAR_LESS_THAN_SIGN ||
5463 ptr[1] == CHAR_APOSTROPHE || ptr[1] == CHAR_LEFT_CURLY_BRACKET))
5464 {
5465 is_recurse = FALSE;
5466 terminator = (*(++ptr) == CHAR_LESS_THAN_SIGN)?
5467 CHAR_GREATER_THAN_SIGN : (*ptr == CHAR_APOSTROPHE)?
5468 CHAR_APOSTROPHE : CHAR_RIGHT_CURLY_BRACKET;
5469 goto NAMED_REF_OR_RECURSE;
5470 }
5471
5472 /* Back references are handled specially; must disable firstbyte if
5473 not set to cope with cases like (?=(\w+))\1: which would otherwise set
5474 ':' later. */
5475
5476 if (-c >= ESC_REF)
5477 {
5478 recno = -c - ESC_REF;
5479
5480 HANDLE_REFERENCE: /* Come here from named backref handling */
5481 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
5482 previous = code;
5483 *code++ = OP_REF;
5484 PUT2INC(code, 0, recno);
5485 cd->backref_map |= (recno < 32)? (1 << recno) : 1;
5486 if (recno > cd->top_backref) cd->top_backref = recno;
5487 }
5488
5489 /* So are Unicode property matches, if supported. */
5490
5491 #ifdef SUPPORT_UCP
5492 else if (-c == ESC_P || -c == ESC_p)
5493 {
5494 BOOL negated;
5495 int pdata;
5496 int ptype = get_ucp(&ptr, &negated, &pdata, errorcodeptr);
5497 if (ptype < 0) goto FAILED;
5498 previous = code;
5499 *code++ = ((-c == ESC_p) != negated)? OP_PROP : OP_NOTPROP;
5500 *code++ = ptype;
5501 *code++ = pdata;
5502 }
5503 #else
5504
5505 /* If Unicode properties are not supported, \X, \P, and \p are not
5506 allowed. */
5507
5508 else if (-c == ESC_X || -c == ESC_P || -c == ESC_p)
5509 {
5510 *errorcodeptr = ERR45;
5511 goto FAILED;
5512 }
5513 #endif
5514
5515 /* For the rest (including \X when Unicode properties are supported), we
5516 can obtain the OP value by negating the escape value. */
5517
5518 else
5519 {
5520 previous = (-c > ESC_b && -c < ESC_Z)? code : NULL;
5521 *code++ = -c;
5522 }
5523 continue;
5524 }
5525
5526 /* We have a data character whose value is in c. In UTF-8 mode it may have
5527 a value > 127. We set its representation in the length/buffer, and then
5528 handle it as a data character. */
5529
5530 #ifdef SUPPORT_UTF8
5531 if (utf8 && c > 127)
5532 mclength = _pcre_ord2utf8(c, mcbuffer);
5533 else
5534 #endif
5535
5536 {
5537 mcbuffer[0] = c;
5538 mclength = 1;
5539 }
5540 goto ONE_CHAR;
5541
5542
5543 /* ===================================================================*/
5544 /* Handle a literal character. It is guaranteed not to be whitespace or #
5545 when the extended flag is set. If we are in UTF-8 mode, it may be a
5546 multi-byte literal character. */
5547
5548 default:
5549 NORMAL_CHAR:
5550 mclength = 1;
5551 mcbuffer[0] = c;
5552
5553 #ifdef SUPPORT_UTF8
5554 if (utf8 && c >= 0xc0)
5555 {
5556 while ((ptr[1] & 0xc0) == 0x80)
5557 mcbuffer[mclength++] = *(++ptr);
5558 }
5559 #endif
5560
5561 /* At this point we have the character's bytes in mcbuffer, and the length
5562 in mclength. When not in UTF-8 mode, the length is always 1. */
5563
5564 ONE_CHAR:
5565 previous = code;
5566 *code++ = ((options & PCRE_CASELESS) != 0)? OP_CHARNC : OP_CHAR;
5567 for (c = 0; c < mclength; c++) *code++ = mcbuffer[c];
5568
5569 /* Remember if \r or \n were seen */
5570
5571 if (mcbuffer[0] == CHAR_CR || mcbuffer[0] == CHAR_NL)
5572 cd->external_flags |= PCRE_HASCRORLF;
5573
5574 /* Set the first and required bytes appropriately. If no previous first
5575 byte, set it from this character, but revert to none on a zero repeat.
5576 Otherwise, leave the firstbyte value alone, and don't change it on a zero
5577 repeat. */
5578
5579 if (firstbyte == REQ_UNSET)
5580 {
5581 zerofirstbyte = REQ_NONE;
5582 zeroreqbyte = reqbyte;
5583
5584 /* If the character is more than one byte long, we can set firstbyte
5585 only if it is not to be matched caselessly. */
5586
5587 if (mclength == 1 || req_caseopt == 0)
5588 {
5589 firstbyte = mcbuffer[0] | req_caseopt;
5590 if (mclength != 1) reqbyte = code[-1] | cd->req_varyopt;
5591 }
5592 else firstbyte = reqbyte = REQ_NONE;
5593 }
5594
5595 /* firstbyte was previously set; we can set reqbyte only the length is
5596 1 or the matching is caseful. */
5597
5598 else
5599 {
5600 zerofirstbyte = firstbyte;
5601 zeroreqbyte = reqbyte;
5602 if (mclength == 1 || req_caseopt == 0)
5603 reqbyte = code[-1] | req_caseopt | cd->req_varyopt;
5604 }
5605
5606 break; /* End of literal character handling */
5607 }
5608 } /* end of big loop */
5609
5610
5611 /* Control never reaches here by falling through, only by a goto for all the
5612 error states. Pass back the position in the pattern so that it can be displayed
5613 to the user for diagnosing the error. */
5614
5615 FAILED:
5616 *ptrptr = ptr;
5617 return FALSE;
5618 }
5619
5620
5621
5622
5623 /*************************************************
5624 * Compile sequence of alternatives *
5625 *************************************************/
5626
5627 /* On entry, ptr is pointing past the bracket character, but on return it
5628 points to the closing bracket, or vertical bar, or end of string. The code
5629 variable is pointing at the byte into which the BRA operator has been stored.
5630 If the ims options are changed at the start (for a (?ims: group) or during any
5631 branch, we need to insert an OP_OPT item at the start of every following branch
5632 to ensure they get set correctly at run time, and also pass the new options
5633 into every subsequent branch compile.
5634
5635 This function is used during the pre-compile phase when we are trying to find
5636 out the amount of memory needed, as well as during the real compile phase. The
5637 value of lengthptr distinguishes the two phases.
5638
5639 Arguments:
5640 options option bits, including any changes for this subpattern
5641 oldims previous settings of ims option bits
5642 codeptr -> the address of the current code pointer
5643 ptrptr -> the address of the current pattern pointer
5644 errorcodeptr -> pointer to error code variable
5645 lookbehind TRUE if this is a lookbehind assertion
5646 reset_bracount TRUE to reset the count for each branch
5647 skipbytes skip this many bytes at start (for brackets and OP_COND)
5648 firstbyteptr place to put the first required character, or a negative number
5649 reqbyteptr place to put the last required character, or a negative number
5650 bcptr pointer to the chain of currently open branches
5651 cd points to the data block with tables pointers etc.
5652 lengthptr NULL during the real compile phase
5653 points to length accumulator during pre-compile phase
5654
5655 Returns: TRUE on success
5656 */
5657
5658 static BOOL
5659 compile_regex(int options, int oldims, uschar **codeptr, const uschar **ptrptr,
5660 int *errorcodeptr, BOOL lookbehind, BOOL reset_bracount, int skipbytes,
5661 int *firstbyteptr, int *reqbyteptr, branch_chain *bcptr, compile_data *cd,
5662 int *lengthptr)
5663 {
5664 const uschar *ptr = *ptrptr;
5665 uschar *code = *codeptr;
5666 uschar *last_branch = code;
5667 uschar *start_bracket = code;
5668 uschar *reverse_count = NULL;
5669 int firstbyte, reqbyte;
5670 int branchfirstbyte, branchreqbyte;
5671 int length;
5672 int orig_bracount;
5673 int max_bracount;
5674 branch_chain bc;
5675
5676 bc.outer = bcptr;
5677 bc.current = code;
5678
5679 firstbyte = reqbyte = REQ_UNSET;
5680
5681 /* Accumulate the length for use in the pre-compile phase. Start with the
5682 length of the BRA and KET and any extra bytes that are required at the
5683 beginning. We accumulate in a local variable to save frequent testing of
5684 lenthptr for NULL. We cannot do this by looking at the value of code at the
5685 start and end of each alternative, because compiled items are discarded during
5686 the pre-compile phase so that the work space is not exceeded. */
5687
5688 length = 2 + 2*LINK_SIZE + skipbytes;
5689
5690 /* WARNING: If the above line is changed for any reason, you must also change
5691 the code that abstracts option settings at the start of the pattern and makes
5692 them global. It tests the value of length for (2 + 2*LINK_SIZE) in the
5693 pre-compile phase to find out whether anything has yet been compiled or not. */
5694
5695 /* Offset is set zero to mark that this bracket is still open */
5696
5697 PUT(code, 1, 0);
5698 code += 1 + LINK_SIZE + skipbytes;
5699
5700 /* Loop for each alternative branch */
5701
5702 orig_bracount = max_bracount = cd->bracount;
5703 for (;;)
5704 {
5705 /* For a (?| group, reset the capturing bracket count so that each branch
5706 uses the same numbers. */
5707
5708 if (reset_bracount) cd->bracount = orig_bracount;
5709
5710 /* Handle a change of ims options at the start of the branch */
5711
5712 if ((options & PCRE_IMS) != oldims)
5713 {
5714 *code++ = OP_OPT;
5715 *code++ = options & PCRE_IMS;
5716 length += 2;
5717 }
5718
5719 /* Set up dummy OP_REVERSE if lookbehind assertion */
5720
5721 if (lookbehind)
5722 {
5723 *code++ = OP_REVERSE;
5724 reverse_count = code;
5725 PUTINC(code, 0, 0);
5726 length += 1 + LINK_SIZE;
5727 }
5728
5729 /* Now compile the branch; in the pre-compile phase its length gets added
5730 into the length. */
5731
5732 if (!compile_branch(&options, &code, &ptr, errorcodeptr, &branchfirstbyte,
5733 &branchreqbyte, &bc, cd, (lengthptr == NULL)? NULL : &length))
5734 {
5735 *ptrptr = ptr;
5736 return FALSE;
5737 }
5738
5739 /* Keep the highest bracket count in case (?| was used and some branch
5740 has fewer than the rest. */
5741
5742 if (cd->bracount > max_bracount) max_bracount = cd->bracount;
5743
5744 /* In the real compile phase, there is some post-processing to be done. */
5745
5746 if (lengthptr == NULL)
5747 {
5748 /* If this is the first branch, the firstbyte and reqbyte values for the
5749 branch become the values for the regex. */
5750
5751 if (*last_branch != OP_ALT)
5752 {
5753 firstbyte = branchfirstbyte;
5754 reqbyte = branchreqbyte;
5755 }
5756
5757 /* If this is not the first branch, the first char and reqbyte have to
5758 match the values from all the previous branches, except that if the
5759 previous value for reqbyte didn't have REQ_VARY set, it can still match,
5760 and we set REQ_VARY for the regex. */
5761
5762 else
5763 {
5764 /* If we previously had a firstbyte, but it doesn't match the new branch,
5765 we have to abandon the firstbyte for the regex, but if there was
5766 previously no reqbyte, it takes on the value of the old firstbyte. */
5767
5768 if (firstbyte >= 0 && firstbyte != branchfirstbyte)
5769 {
5770 if (reqbyte < 0) reqbyte = firstbyte;
5771 firstbyte = REQ_NONE;
5772 }
5773
5774 /* If we (now or from before) have no firstbyte, a firstbyte from the
5775 branch becomes a reqbyte if there isn't a branch reqbyte. */
5776
5777 if (firstbyte < 0 && branchfirstbyte >= 0 && branchreqbyte < 0)
5778 branchreqbyte = branchfirstbyte;
5779
5780 /* Now ensure that the reqbytes match */
5781
5782 if ((reqbyte & ~REQ_VARY) != (branchreqbyte & ~REQ_VARY))
5783 reqbyte = REQ_NONE;
5784 else reqbyte |= branchreqbyte; /* To "or" REQ_VARY */
5785 }
5786
5787 /* If lookbehind, check that this branch matches a fixed-length string, and
5788 put the length into the OP_REVERSE item. Temporarily mark the end of the
5789 branch with OP_END. */
5790
5791 if (lookbehind)
5792 {
5793 int fixed_length;
5794 *code = OP_END;
5795 fixed_length = find_fixedlength(last_branch, options);
5796 DPRINTF(("fixed length = %d\n", fixed_length));
5797 if (fixed_length < 0)
5798 {
5799 *errorcodeptr = (fixed_length == -2)? ERR36 : ERR25;
5800 *ptrptr = ptr;
5801 return FALSE;
5802 }
5803 PUT(reverse_count, 0, fixed_length);
5804 }
5805 }
5806
5807 /* Reached end of expression, either ')' or end of pattern. In the real
5808 compile phase, go back through the alternative branches and reverse the chain
5809 of offsets, with the field in the BRA item now becoming an offset to the
5810 first alternative. If there are no alternatives, it points to the end of the
5811 group. The length in the terminating ket is always the length of the whole
5812 bracketed item. If any of the ims options were changed inside the group,
5813 compile a resetting op-code following, except at the very end of the pattern.
5814 Return leaving the pointer at the terminating char. */
5815
5816 if (*ptr != CHAR_VERTICAL_LINE)
5817 {
5818 if (lengthptr == NULL)
5819 {
5820 int branch_length = code - last_branch;
5821 do
5822 {
5823 int prev_length = GET(last_branch, 1);
5824 PUT(last_branch, 1, branch_length);
5825 branch_length = prev_length;
5826 last_branch -= branch_length;
5827 }
5828 while (branch_length > 0);
5829 }
5830
5831 /* Fill in the ket */
5832
5833 *code = OP_KET;
5834 PUT(code, 1, code - start_bracket);
5835 code += 1 + LINK_SIZE;
5836
5837 /* Resetting option if needed */
5838
5839 if ((options & PCRE_IMS) != oldims && *ptr == CHAR_RIGHT_PARENTHESIS)
5840 {
5841 *code++ = OP_OPT;
5842 *code++ = oldims;
5843 length += 2;
5844 }
5845
5846 /* Retain the highest bracket number, in case resetting was used. */
5847
5848 cd->bracount = max_bracount;
5849
5850 /* Set values to pass back */
5851
5852 *codeptr = code;
5853 *ptrptr = ptr;
5854 *firstbyteptr = firstbyte;
5855 *reqbyteptr = reqbyte;
5856 if (lengthptr != NULL)
5857 {
5858 if (OFLOW_MAX - *lengthptr < length)
5859 {
5860 *errorcodeptr = ERR20;
5861 return FALSE;
5862 }
5863 *lengthptr += length;
5864 }
5865 return TRUE;
5866 }
5867
5868 /* Another branch follows. In the pre-compile phase, we can move the code
5869 pointer back to where it was for the start of the first branch. (That is,
5870 pretend that each branch is the only one.)
5871
5872 In the real compile phase, insert an ALT node. Its length field points back
5873 to the previous branch while the bracket remains open. At the end the chain
5874 is reversed. It's done like this so that the start of the bracket has a
5875 zero offset until it is closed, making it possible to detect recursion. */
5876
5877 if (lengthptr != NULL)
5878 {
5879 code = *codeptr + 1 + LINK_SIZE + skipbytes;
5880 length += 1 + LINK_SIZE;
5881 }
5882 else
5883 {
5884 *code = OP_ALT;
5885 PUT(code, 1, code - last_branch);
5886 bc.current = last_branch = code;
5887 code += 1 + LINK_SIZE;
5888 }
5889
5890 ptr++;
5891 }
5892 /* Control never reaches here */
5893 }
5894
5895
5896
5897
5898 /*************************************************
5899 * Check for anchored expression *
5900 *************************************************/
5901
5902 /* Try to find out if this is an anchored regular expression. Consider each
5903 alternative branch. If they all start with OP_SOD or OP_CIRC, or with a bracket
5904 all of whose alternatives start with OP_SOD or OP_CIRC (recurse ad lib), then
5905 it's anchored. However, if this is a multiline pattern, then only OP_SOD
5906 counts, since OP_CIRC can match in the middle.
5907
5908 We can also consider a regex to be anchored if OP_SOM starts all its branches.
5909 This is the code for \G, which means "match at start of match position, taking
5910 into account the match offset".
5911
5912 A branch is also implicitly anchored if it starts with .* and DOTALL is set,
5913 because that will try the rest of the pattern at all possible matching points,
5914 so there is no point trying again.... er ....
5915
5916 .... except when the .* appears inside capturing parentheses, and there is a
5917 subsequent back reference to those parentheses. We haven't enough information
5918 to catch that case precisely.
5919
5920 At first, the best we could do was to detect when .* was in capturing brackets
5921 and the highest back reference was greater than or equal to that level.
5922 However, by keeping a bitmap of the first 31 back references, we can catch some
5923 of the more common cases more precisely.
5924
5925 Arguments:
5926 code points to start of expression (the bracket)
5927 options points to the options setting
5928 bracket_map a bitmap of which brackets we are inside while testing; this
5929 handles up to substring 31; after that we just have to take
5930 the less precise approach
5931 backref_map the back reference bitmap
5932
5933 Returns: TRUE or FALSE
5934 */
5935
5936 static BOOL
5937 is_anchored(register const uschar *code, int *options, unsigned int bracket_map,
5938 unsigned int backref_map)
5939 {
5940 do {
5941 const uschar *scode = first_significant_code(code + _pcre_OP_lengths[*code],
5942 options, PCRE_MULTILINE, FALSE);
5943 register int op = *scode;
5944
5945 /* Non-capturing brackets */
5946
5947 if (op == OP_BRA)
5948 {
5949 if (!is_anchored(scode, options, bracket_map, backref_map)) return FALSE;
5950 }
5951
5952 /* Capturing brackets */
5953
5954 else if (op == OP_CBRA)
5955 {
5956 int n = GET2(scode, 1+LINK_SIZE);
5957 int new_map = bracket_map | ((n < 32)? (1 << n) : 1);
5958 if (!is_anchored(scode, options, new_map, backref_map)) return FALSE;
5959 }
5960
5961 /* Other brackets */
5962
5963 else if (op == OP_ASSERT || op == OP_ONCE || op == OP_COND)
5964 {
5965 if (!is_anchored(scode, options, bracket_map, backref_map)) return FALSE;
5966 }
5967
5968 /* .* is not anchored unless DOTALL is set (which generates OP_ALLANY) and
5969 it isn't in brackets that are or may be referenced. */
5970
5971 else if ((op == OP_TYPESTAR || op == OP_TYPEMINSTAR ||
5972 op == OP_TYPEPOSSTAR))
5973 {
5974 if (scode[1] != OP_ALLANY || (bracket_map & backref_map) != 0)
5975 return FALSE;
5976 }
5977
5978 /* Check for explicit anchoring */
5979
5980 else if (op != OP_SOD && op != OP_SOM &&
5981 ((*options & PCRE_MULTILINE) != 0 || op != OP_CIRC))
5982 return FALSE;
5983 code += GET(code, 1);
5984 }
5985 while (*code == OP_ALT); /* Loop for each alternative */
5986 return TRUE;
5987 }
5988
5989
5990
5991 /*************************************************
5992 * Check for starting with ^ or .* *
5993 *************************************************/
5994
5995 /* This is called to find out if every branch starts with ^ or .* so that
5996 "first char" processing can be done to speed things up in multiline
5997 matching and for non-DOTALL patterns that start with .* (which must start at
5998 the beginning or after \n). As in the case of is_anchored() (see above), we
5999 have to take account of back references to capturing brackets that contain .*
6000 because in that case we can't make the assumption.
6001
6002 Arguments:
6003 code points to start of expression (the bracket)
6004 bracket_map a bitmap of which brackets we are inside while testing; this
6005 handles up to substring 31; after that we just have to take
6006 the less precise approach
6007 backref_map the back reference bitmap
6008
6009 Returns: TRUE or FALSE
6010 */
6011
6012 static BOOL
6013 is_startline(const uschar *code, unsigned int bracket_map,
6014 unsigned int backref_map)
6015 {
6016 do {
6017 const uschar *scode = first_significant_code(code + _pcre_OP_lengths[*code],
6018 NULL, 0, FALSE);
6019 register int op = *scode;
6020
6021 /* If we are at the start of a conditional assertion group, *both* the
6022 conditional assertion *and* what follows the condition must satisfy the test
6023 for start of line. Other kinds of condition fail. Note that there may be an
6024 auto-callout at the start of a condition. */
6025
6026 if (op == OP_COND)
6027 {
6028 scode += 1 + LINK_SIZE;
6029 if (*scode == OP_CALLOUT) scode += _pcre_OP_lengths[OP_CALLOUT];
6030 switch (*scode)
6031 {
6032 case OP_CREF:
6033 case OP_RREF:
6034 case OP_DEF:
6035 return FALSE;
6036
6037 default: /* Assertion */
6038 if (!is_startline(scode, bracket_map, backref_map)) return FALSE;
6039 do scode += GET(scode, 1); while (*scode == OP_ALT);
6040 scode += 1 + LINK_SIZE;
6041 break;
6042 }
6043 scode = first_significant_code(scode, NULL, 0, FALSE);
6044 op = *scode;
6045 }
6046
6047 /* Non-capturing brackets */
6048
6049 if (op == OP_BRA)
6050 {
6051 if (!is_startline(scode, bracket_map, backref_map)) return FALSE;
6052 }
6053
6054 /* Capturing brackets */
6055
6056 else if (op == OP_CBRA)
6057 {
6058 int n = GET2(scode, 1+LINK_SIZE);
6059 int new_map = bracket_map | ((n < 32)? (1 << n) : 1);
6060 if (!is_startline(scode, new_map, backref_map)) return FALSE;
6061 }
6062
6063 /* Other brackets */
6064
6065 else if (op == OP_ASSERT || op == OP_ONCE)
6066 {
6067 if (!is_startline(scode, bracket_map, backref_map)) return FALSE;
6068 }
6069
6070 /* .* means "start at start or after \n" if it isn't in brackets that
6071 may be referenced. */
6072
6073 else if (op == OP_TYPESTAR || op == OP_TYPEMINSTAR || op == OP_TYPEPOSSTAR)
6074 {
6075 if (scode[1] != OP_ANY || (bracket_map & backref_map) != 0) return FALSE;
6076 }
6077
6078 /* Check for explicit circumflex */
6079
6080 else if (op != OP_CIRC) return FALSE;
6081
6082 /* Move on to the next alternative */
6083
6084 code += GET(code, 1);
6085 }
6086 while (*code == OP_ALT); /* Loop for each alternative */
6087 return TRUE;
6088 }
6089
6090
6091
6092 /*************************************************
6093 * Check for asserted fixed first char *
6094 *************************************************/
6095
6096 /* During compilation, the "first char" settings from forward assertions are
6097 discarded, because they can cause conflicts with actual literals that follow.
6098 However, if we end up without a first char setting for an unanchored pattern,
6099 it is worth scanning the regex to see if there is an initial asserted first
6100 char. If all branches start with the same asserted char, or with a bracket all
6101 of whose alternatives start with the same asserted char (recurse ad lib), then
6102 we return that char, otherwise -1.
6103
6104 Arguments:
6105 code points to start of expression (the bracket)
6106 options pointer to the options (used to check casing changes)
6107 inassert TRUE if in an assertion
6108
6109 Returns: -1 or the fixed first char
6110 */
6111
6112 static int
6113 find_firstassertedchar(const uschar *code, int *options, BOOL inassert)
6114 {
6115 register int c = -1;
6116 do {
6117 int d;
6118 const uschar *scode =
6119 first_significant_code(code + 1+LINK_SIZE, options, PCRE_CASELESS, TRUE);
6120 register int op = *scode;
6121
6122 switch(op)
6123 {
6124 default:
6125 return -1;
6126
6127 case OP_BRA:
6128 case OP_CBRA:
6129 case OP_ASSERT:
6130 case OP_ONCE:
6131 case OP_COND:
6132 if ((d = find_firstassertedchar(scode, options, op == OP_ASSERT)) < 0)
6133 return -1;
6134 if (c < 0) c = d; else if (c != d) return -1;
6135 break;
6136
6137 case OP_EXACT: /* Fall through */
6138 scode += 2;
6139
6140 case OP_CHAR:
6141 case OP_CHARNC:
6142 case OP_PLUS:
6143 case OP_MINPLUS:
6144 case OP_POSPLUS:
6145 if (!inassert) return -1;
6146 if (c < 0)
6147 {
6148 c = scode[1];
6149 if ((*options & PCRE_CASELESS) != 0) c |= REQ_CASELESS;
6150 }
6151 else if (c != scode[1]) return -1;
6152 break;
6153 }
6154
6155 code += GET(code, 1);
6156 }
6157 while (*code == OP_ALT);
6158 return c;
6159 }
6160
6161
6162
6163 /*************************************************
6164 * Compile a Regular Expression *
6165 *************************************************/
6166
6167 /* This function takes a string and returns a pointer to a block of store
6168 holding a compiled version of the expression. The original API for this
6169 function had no error code return variable; it is retained for backwards
6170 compatibility. The new function is given a new name.
6171
6172 Arguments:
6173 pattern the regular expression
6174 options various option bits
6175 errorcodeptr pointer to error code variable (pcre_compile2() only)
6176 can be NULL if you don't want a code value
6177 errorptr pointer to pointer to error text
6178 erroroffset ptr offset in pattern where error was detected
6179 tables pointer to character tables or NULL
6180
6181 Returns: pointer to compiled data block, or NULL on error,
6182 with errorptr and erroroffset set
6183 */
6184
6185 PCRE_EXP_DEFN pcre * PCRE_CALL_CONVENTION
6186 pcre_compile(const char *pattern, int options, const char **errorptr,
6187 int *erroroffset, const unsigned char *tables)
6188 {
6189 return pcre_compile2(pattern, options, NULL, errorptr, erroroffset, tables);
6190 }
6191
6192
6193 PCRE_EXP_DEFN pcre * PCRE_CALL_CONVENTION
6194 pcre_compile2(const char *pattern, int options, int *errorcodeptr,
6195 const char **errorptr, int *erroroffset, const unsigned char *tables)
6196 {
6197 real_pcre *re;
6198 int length = 1; /* For final END opcode */
6199 int firstbyte, reqbyte, newline;
6200 int errorcode = 0;
6201 int skipatstart = 0;
6202 #ifdef SUPPORT_UTF8
6203 BOOL utf8;
6204 #endif
6205 size_t size;
6206 uschar *code;
6207 const uschar *codestart;
6208 const uschar *ptr;
6209 compile_data compile_block;
6210 compile_data *cd = &compile_block;
6211
6212 /* This space is used for "compiling" into during the first phase, when we are
6213 computing the amount of memory that is needed. Compiled items are thrown away
6214 as soon as possible, so that a fairly large buffer should be sufficient for
6215 this purpose. The same space is used in the second phase for remembering where
6216 to fill in forward references to subpatterns. */
6217
6218 uschar cworkspace[COMPILE_WORK_SIZE];
6219
6220 /* Set this early so that early errors get offset 0. */
6221
6222 ptr = (const uschar *)pattern;
6223
6224 /* We can't pass back an error message if errorptr is NULL; I guess the best we
6225 can do is just return NULL, but we can set a code value if there is a code
6226 pointer. */
6227
6228 if (errorptr == NULL)
6229 {
6230 if (errorcodeptr != NULL) *errorcodeptr = 99;
6231 return NULL;
6232 }
6233
6234 *errorptr = NULL;
6235 if (errorcodeptr != NULL) *errorcodeptr = ERR0;
6236
6237 /* However, we can give a message for this error */
6238
6239 if (erroroffset == NULL)
6240 {
6241 errorcode = ERR16;
6242 goto PCRE_EARLY_ERROR_RETURN2;
6243 }
6244
6245 *erroroffset = 0;
6246
6247 /* Set up pointers to the individual character tables */
6248
6249 if (tables == NULL) tables = _pcre_default_tables;
6250 cd->lcc = tables + lcc_offset;
6251 cd->fcc = tables + fcc_offset;
6252 cd->cbits = tables + cbits_offset;
6253 cd->ctypes = tables + ctypes_offset;
6254
6255 /* Check that all undefined public option bits are zero */
6256
6257 if ((options & ~PUBLIC_COMPILE_OPTIONS) != 0)
6258 {
6259 errorcode = ERR17;
6260 goto PCRE_EARLY_ERROR_RETURN;
6261 }
6262
6263 /* Check for global one-time settings at the start of the pattern, and remember
6264 the offset for later. */
6265
6266 while (ptr[skipatstart] == CHAR_LEFT_PARENTHESIS &&
6267 ptr[skipatstart+1] == CHAR_ASTERISK)
6268 {
6269 int newnl = 0;
6270 int newbsr = 0;
6271
6272 if (strncmp((char *)(ptr+skipatstart+2), STRING_UTF8_RIGHTPAR, 5) == 0)
6273 { skipatstart += 7; options |= PCRE_UTF8; continue; }
6274
6275 if (strncmp((char *)(ptr+skipatstart+2), STRING_CR_RIGHTPAR, 3) == 0)
6276 { skipatstart += 5; newnl = PCRE_NEWLINE_CR; }
6277 else if (strncmp((char *)(ptr+skipatstart+2), STRING_LF_RIGHTPAR, 3) == 0)
6278 { skipatstart += 5; newnl = PCRE_NEWLINE_LF; }
6279 else if (strncmp((char *)(ptr+skipatstart+2), STRING_CRLF_RIGHTPAR, 5) == 0)
6280 { skipatstart += 7; newnl = PCRE_NEWLINE_CR + PCRE_NEWLINE_LF; }
6281 else if (strncmp((char *)(ptr+skipatstart+2), STRING_ANY_RIGHTPAR, 4) == 0)
6282 { skipatstart += 6; newnl = PCRE_NEWLINE_ANY; }
6283 else if (strncmp((char *)(ptr+skipatstart+2), STRING_ANYCRLF_RIGHTPAR, 8) == 0)
6284 { skipatstart += 10; newnl = PCRE_NEWLINE_ANYCRLF; }
6285
6286 else if (strncmp((char *)(ptr+skipatstart+2), STRING_BSR_ANYCRLF_RIGHTPAR, 12) == 0)
6287 { skipatstart += 14; newbsr = PCRE_BSR_ANYCRLF; }
6288 else if (strncmp((char *)(ptr+skipatstart+2), STRING_BSR_UNICODE_RIGHTPAR, 12) == 0)
6289 { skipatstart += 14; newbsr = PCRE_BSR_UNICODE; }
6290
6291 if (newnl != 0)
6292 options = (options & ~PCRE_NEWLINE_BITS) | newnl;
6293 else if (newbsr != 0)
6294 options = (options & ~(PCRE_BSR_ANYCRLF|PCRE_BSR_UNICODE)) | newbsr;
6295 else break;
6296 }
6297
6298 /* Can't support UTF8 unless PCRE has been compiled to include the code. */
6299
6300 #ifdef SUPPORT_UTF8
6301 utf8 = (options & PCRE_UTF8) != 0;
6302 if (utf8 && (options & PCRE_NO_UTF8_CHECK) == 0 &&
6303 (*erroroffset = _pcre_valid_utf8((uschar *)pattern, -1)) >= 0)
6304 {
6305 errorcode = ERR44;
6306 goto PCRE_EARLY_ERROR_RETURN2;
6307 }
6308 #else
6309 if ((options & PCRE_UTF8) != 0)
6310 {
6311 errorcode = ERR32;
6312 goto PCRE_EARLY_ERROR_RETURN;
6313 }
6314 #endif
6315
6316 /* Check validity of \R options. */
6317
6318 switch (options & (PCRE_BSR_ANYCRLF|PCRE_BSR_UNICODE))
6319 {
6320 case 0:
6321 case PCRE_BSR_ANYCRLF:
6322 case PCRE_BSR_UNICODE:
6323 break;
6324 default: errorcode = ERR56; goto PCRE_EARLY_ERROR_RETURN;
6325 }
6326
6327 /* Handle different types of newline. The three bits give seven cases. The
6328 current code allows for fixed one- or two-byte sequences, plus "any" and
6329 "anycrlf". */
6330
6331 switch (options & PCRE_NEWLINE_BITS)
6332 {
6333 case 0: newline = NEWLINE; break; /* Build-time default */
6334 case PCRE_NEWLINE_CR: newline = CHAR_CR; break;
6335 case PCRE_NEWLINE_LF: newline = CHAR_NL; break;
6336 case PCRE_NEWLINE_CR+
6337 PCRE_NEWLINE_LF: newline = (CHAR_CR << 8) | CHAR_NL; break;
6338 case PCRE_NEWLINE_ANY: newline = -1; break;
6339 case PCRE_NEWLINE_ANYCRLF: newline = -2; break;
6340 default: errorcode = ERR56; goto PCRE_EARLY_ERROR_RETURN;
6341 }
6342
6343 if (newline == -2)
6344 {
6345 cd->nltype = NLTYPE_ANYCRLF;
6346 }
6347 else if (newline < 0)
6348 {
6349 cd->nltype = NLTYPE_ANY;
6350 }
6351 else
6352 {
6353 cd->nltype = NLTYPE_FIXED;
6354 if (newline > 255)
6355 {
6356 cd->nllen = 2;
6357 cd->nl[0] = (newline >> 8) & 255;
6358 cd->nl[1] = newline & 255;
6359 }
6360 else
6361 {
6362 cd->nllen = 1;
6363 cd->nl[0] = newline;
6364 }
6365 }
6366
6367 /* Maximum back reference and backref bitmap. The bitmap records up to 31 back
6368 references to help in deciding whether (.*) can be treated as anchored or not.
6369 */
6370
6371 cd->top_backref = 0;
6372 cd->backref_map = 0;
6373
6374 /* Reflect pattern for debugging output */
6375
6376 DPRINTF(("------------------------------------------------------------------\n"));
6377 DPRINTF(("%s\n", pattern));
6378
6379 /* Pretend to compile the pattern while actually just accumulating the length
6380 of memory required. This behaviour is triggered by passing a non-NULL final
6381 argument to compile_regex(). We pass a block of workspace (cworkspace) for it
6382 to compile parts of the pattern into; the compiled code is discarded when it is
6383 no longer needed, so hopefully this workspace will never overflow, though there
6384 is a test for its doing so. */
6385
6386 cd->bracount = cd->final_bracount = 0;
6387 cd->names_found = 0;
6388 cd->name_entry_size = 0;
6389 cd->name_table = NULL;
6390 cd->start_workspace = cworkspace;
6391 cd->start_code = cworkspace;
6392 cd->hwm = cworkspace;
6393 cd->start_pattern = (const uschar *)pattern;
6394 cd->end_pattern = (const uschar *)(pattern + strlen(pattern));
6395 cd->req_varyopt = 0;
6396 cd->external_options = options;
6397 cd->external_flags = 0;
6398
6399 /* Now do the pre-compile. On error, errorcode will be set non-zero, so we
6400 don't need to look at the result of the function here. The initial options have
6401 been put into the cd block so that they can be changed if an option setting is
6402 found within the regex right at the beginning. Bringing initial option settings
6403 outside can help speed up starting point checks. */
6404
6405 ptr += skipatstart;
6406 code = cworkspace;
6407 *code = OP_BRA;
6408 (void)compile_regex(cd->external_options, cd->external_options & PCRE_IMS,
6409 &code, &ptr, &errorcode, FALSE, FALSE, 0, &firstbyte, &reqbyte, NULL, cd,
6410 &length);
6411 if (errorcode != 0) goto PCRE_EARLY_ERROR_RETURN;
6412
6413 DPRINTF(("end pre-compile: length=%d workspace=%d\n", length,
6414 cd->hwm - cworkspace));
6415
6416 if (length > MAX_PATTERN_SIZE)
6417 {
6418 errorcode = ERR20;
6419 goto PCRE_EARLY_ERROR_RETURN;
6420 }
6421
6422 /* Compute the size of data block needed and get it, either from malloc or
6423 externally provided function. Integer overflow should no longer be possible
6424 because nowadays we limit the maximum value of cd->names_found and
6425 cd->name_entry_size. */
6426
6427 size = length + sizeof(real_pcre) + cd->names_found * (cd->name_entry_size + 3);
6428 re = (real_pcre *)(pcre_malloc)(size);
6429
6430 if (re == NULL)
6431 {
6432 errorcode = ERR21;
6433 goto PCRE_EARLY_ERROR_RETURN;
6434 }
6435
6436 /* Put in the magic number, and save the sizes, initial options, internal
6437 flags, and character table pointer. NULL is used for the default character
6438 tables. The nullpad field is at the end; it's there to help in the case when a
6439 regex compiled on a system with 4-byte pointers is run on another with 8-byte
6440 pointers. */
6441
6442 re->magic_number = MAGIC_NUMBER;
6443 re->size = size;
6444 re->options = cd->external_options;
6445 re->flags = cd->external_flags;
6446 re->dummy1 = 0;
6447 re->first_byte = 0;
6448 re->req_byte = 0;
6449 re->name_table_offset = sizeof(real_pcre);
6450 re->name_entry_size = cd->name_entry_size;
6451 re->name_count = cd->names_found;
6452 re->ref_count = 0;
6453 re->tables = (tables == _pcre_default_tables)? NULL : tables;
6454 re->nullpad = NULL;
6455
6456 /* The starting points of the name/number translation table and of the code are
6457 passed around in the compile data block. The start/end pattern and initial
6458 options are already set from the pre-compile phase, as is the name_entry_size
6459 field. Reset the bracket count and the names_found field. Also reset the hwm
6460 field; this time it's used for remembering forward references to subpatterns.
6461 */
6462
6463 cd->final_bracount = cd->bracount; /* Save for checking forward references */
6464 cd->bracount = 0;
6465 cd->names_found = 0;
6466 cd->name_table = (uschar *)re + re->name_table_offset;
6467 codestart = cd->name_table + re->name_entry_size * re->name_count;
6468 cd->start_code = codestart;
6469 cd->hwm = cworkspace;
6470 cd->req_varyopt = 0;
6471 cd->had_accept = FALSE;
6472
6473 /* Set up a starting, non-extracting bracket, then compile the expression. On
6474 error, errorcode will be set non-zero, so we don't need to look at the result
6475 of the function here. */
6476
6477 ptr = (const uschar *)pattern + skipatstart;
6478 code = (uschar *)codestart;
6479 *code = OP_BRA;
6480 (void)compile_regex(re->options, re->options & PCRE_IMS, &code, &ptr,
6481 &errorcode, FALSE, FALSE, 0, &firstbyte, &reqbyte, NULL, cd, NULL);
6482 re->top_bracket = cd->bracount;
6483 re->top_backref = cd->top_backref;
6484 re->flags = cd->external_flags;
6485
6486 if (cd->had_accept) reqbyte = -1; /* Must disable after (*ACCEPT) */
6487
6488 /* If not reached end of pattern on success, there's an excess bracket. */
6489
6490 if (errorcode == 0 && *ptr != 0) errorcode = ERR22;
6491
6492 /* Fill in the terminating state and check for disastrous overflow, but
6493 if debugging, leave the test till after things are printed out. */
6494
6495 *code++ = OP_END;
6496
6497 #ifndef DEBUG
6498 if (code - codestart > length) errorcode = ERR23;
6499 #endif
6500
6501 /* Fill in any forward references that are required. */
6502
6503 while (errorcode == 0 && cd->hwm > cworkspace)
6504 {
6505 int offset, recno;
6506 const uschar *groupptr;
6507 cd->hwm -= LINK_SIZE;
6508 offset = GET(cd->hwm, 0);
6509 recno = GET(codestart, offset);
6510 groupptr = find_bracket(codestart, (re->options & PCRE_UTF8) != 0, recno);
6511 if (groupptr == NULL) errorcode = ERR53;
6512 else PUT(((uschar *)codestart), offset, groupptr - codestart);
6513 }
6514
6515 /* Give an error if there's back reference to a non-existent capturing
6516 subpattern. */
6517
6518 if (errorcode == 0 && re->top_backref > re->top_bracket) errorcode = ERR15;
6519
6520 /* Failed to compile, or error while post-processing */
6521
6522 if (errorcode != 0)
6523 {
6524 (pcre_free)(re);
6525 PCRE_EARLY_ERROR_RETURN:
6526 *erroroffset = ptr - (const uschar *)pattern;
6527 PCRE_EARLY_ERROR_RETURN2:
6528 *errorptr = find_error_text(errorcode);
6529 if (errorcodeptr != NULL) *errorcodeptr = errorcode;
6530 return NULL;
6531 }
6532
6533 /* If the anchored option was not passed, set the flag if we can determine that
6534 the pattern is anchored by virtue of ^ characters or \A or anything else (such
6535 as starting with .* when DOTALL is set).
6536
6537 Otherwise, if we know what the first byte has to be, save it, because that
6538 speeds up unanchored matches no end. If not, see if we can set the
6539 PCRE_STARTLINE flag. This is helpful for multiline matches when all branches
6540 start with ^. and also when all branches start with .* for non-DOTALL matches.
6541 */
6542
6543 if ((re->options & PCRE_ANCHORED) == 0)
6544 {
6545 int temp_options = re->options; /* May get changed during these scans */
6546 if (is_anchored(codestart, &temp_options, 0, cd->backref_map))
6547 re->options |= PCRE_ANCHORED;
6548 else
6549 {
6550 if (firstbyte < 0)
6551 firstbyte = find_firstassertedchar(codestart, &temp_options, FALSE);
6552 if (firstbyte >= 0) /* Remove caseless flag for non-caseable chars */
6553 {
6554 int ch = firstbyte & 255;
6555 re->first_byte = ((firstbyte & REQ_CASELESS) != 0 &&
6556 cd->fcc[ch] == ch)? ch : firstbyte;
6557 re->flags |= PCRE_FIRSTSET;
6558 }
6559 else if (is_startline(codestart, 0, cd->backref_map))
6560 re->flags |= PCRE_STARTLINE;
6561 }
6562 }
6563
6564 /* For an anchored pattern, we use the "required byte" only if it follows a
6565 variable length item in the regex. Remove the caseless flag for non-caseable
6566 bytes. */
6567
6568 if (reqbyte >= 0 &&
6569 ((re->options & PCRE_ANCHORED) == 0 || (reqbyte & REQ_VARY) != 0))
6570 {
6571 int ch = reqbyte & 255;
6572 re->req_byte = ((reqbyte & REQ_CASELESS) != 0 &&
6573 cd->fcc[ch] == ch)? (reqbyte & ~REQ_CASELESS) : reqbyte;
6574 re->flags |= PCRE_REQCHSET;
6575 }
6576
6577 /* Print out the compiled data if debugging is enabled. This is never the
6578 case when building a production library. */
6579
6580 #ifdef DEBUG
6581
6582 printf("Length = %d top_bracket = %d top_backref = %d\n",
6583 length, re->top_bracket, re->top_backref);
6584
6585 printf("Options=%08x\n", re->options);
6586
6587 if ((re->flags & PCRE_FIRSTSET) != 0)
6588 {
6589 int ch = re->first_byte & 255;
6590 const char *caseless = ((re->first_byte & REQ_CASELESS) == 0)?
6591 "" : " (caseless)";
6592 if (isprint(ch)) printf("First char = %c%s\n", ch, caseless);
6593 else printf("First char = \\x%02x%s\n", ch, caseless);
6594 }
6595
6596 if ((re->flags & PCRE_REQCHSET) != 0)
6597 {
6598 int ch = re->req_byte & 255;
6599 const char *caseless = ((re->req_byte & REQ_CASELESS) == 0)?
6600 "" : " (caseless)";
6601 if (isprint(ch)) printf("Req char = %c%s\n", ch, caseless);
6602 else printf("Req char = \\x%02x%s\n", ch, caseless);
6603 }
6604
6605 pcre_printint(re, stdout, TRUE);
6606
6607 /* This check is done here in the debugging case so that the code that
6608 was compiled can be seen. */
6609
6610 if (code - codestart > length)
6611 {
6612 (pcre_free)(re);
6613 *errorptr = find_error_text(ERR23);
6614 *erroroffset = ptr - (uschar *)pattern;
6615 if (errorcodeptr != NULL) *errorcodeptr = ERR23;
6616 return NULL;
6617 }
6618 #endif /* DEBUG */
6619
6620 return (pcre *)re;
6621 }
6622
6623 /* End of pcre_compile.c */

Properties

Name Value
svn:eol-style native
svn:keywords "Author Date Id Revision Url"

webmaster@exim.org
ViewVC Help
Powered by ViewVC 1.1.12