1 /*************************************************
2 * Perl-Compatible Regular Expressions *
3 *************************************************/
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.
8 Written by Philip Hazel
9 Copyright (c) 1997-2008 University of Cambridge
11 -----------------------------------------------------------------------------
12 Redistribution and use in source and binary forms, with or without
13 modification, are permitted provided that the following conditions are met:
15 * Redistributions of source code must retain the above copyright notice,
16 this list of conditions and the following disclaimer.
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.
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.
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 -----------------------------------------------------------------------------
41 /* This module contains the external function pcre_compile(), along with
42 supporting internal functions that are not used by other modules. */
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 */
53 #include "pcre_internal.h"
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. */
60 #include "pcre_printint.src"
64 /* Macro for setting individual bits in class bitmaps. */
66 #define SETBIT(a,b) a[b/8] |= (1 << (b%8))
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. */
73 #define OFLOW_MAX (INT_MAX - 20)
76 /*************************************************
77 * Code parameters and static tables *
78 *************************************************/
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.
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. */
92 #define COMPILE_WORK_SIZE (4096)
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
100 #ifndef EBCDIC /* This is the "normal" table for ASCII systems */
101 static const short int escapes[] = {
102 0, 0, 0, 0, 0, 0, 0, 0, /* 0 - 7 */
103 0, 0, ':', ';', '<', '=', '>', '?', /* 8 - ? */
104 '@', -ESC_A, -ESC_B, -ESC_C, -ESC_D, -ESC_E, 0, -ESC_G, /* @ - G */
105 -ESC_H, 0, 0, -ESC_K, 0, 0, 0, 0, /* H - O */
106 -ESC_P, -ESC_Q, -ESC_R, -ESC_S, 0, 0, -ESC_V, -ESC_W, /* P - W */
107 -ESC_X, 0, -ESC_Z, '[', '\\', ']', '^', '_', /* X - _ */
108 '`', 7, -ESC_b, 0, -ESC_d, ESC_e, ESC_f, 0, /* ` - g */
109 -ESC_h, 0, 0, -ESC_k, 0, 0, ESC_n, 0, /* h - o */
110 -ESC_p, 0, ESC_r, -ESC_s, ESC_tee, 0, -ESC_v, -ESC_w, /* p - w */
111 0, 0, -ESC_z /* x - z */
114 #else /* This is the "abnormal" table for EBCDIC systems */
115 static const short int escapes[] = {
116 /* 48 */ 0, 0, 0, '.', '<', '(', '+', '|',
117 /* 50 */ '&', 0, 0, 0, 0, 0, 0, 0,
118 /* 58 */ 0, 0, '!', '$', '*', ')', ';', '~',
119 /* 60 */ '-', '/', 0, 0, 0, 0, 0, 0,
120 /* 68 */ 0, 0, '|', ',', '%', '_', '>', '?',
121 /* 70 */ 0, 0, 0, 0, 0, 0, 0, 0,
122 /* 78 */ 0, '`', ':', '#', '@', '\'', '=', '"',
123 /* 80 */ 0, 7, -ESC_b, 0, -ESC_d, ESC_e, ESC_f, 0,
124 /* 88 */-ESC_h, 0, 0, '{', 0, 0, 0, 0,
125 /* 90 */ 0, 0, -ESC_k, 'l', 0, ESC_n, 0, -ESC_p,
126 /* 98 */ 0, ESC_r, 0, '}', 0, 0, 0, 0,
127 /* A0 */ 0, '~', -ESC_s, ESC_tee, 0,-ESC_v, -ESC_w, 0,
128 /* A8 */ 0,-ESC_z, 0, 0, 0, '[', 0, 0,
129 /* B0 */ 0, 0, 0, 0, 0, 0, 0, 0,
130 /* B8 */ 0, 0, 0, 0, 0, ']', '=', '-',
131 /* C0 */ '{',-ESC_A, -ESC_B, -ESC_C, -ESC_D,-ESC_E, 0, -ESC_G,
132 /* C8 */-ESC_H, 0, 0, 0, 0, 0, 0, 0,
133 /* D0 */ '}', 0, -ESC_K, 0, 0, 0, 0, -ESC_P,
134 /* D8 */-ESC_Q,-ESC_R, 0, 0, 0, 0, 0, 0,
135 /* E0 */ '\\', 0, -ESC_S, 0, 0,-ESC_V, -ESC_W, -ESC_X,
136 /* E8 */ 0,-ESC_Z, 0, 0, 0, 0, 0, 0,
137 /* F0 */ 0, 0, 0, 0, 0, 0, 0, 0,
138 /* F8 */ 0, 0, 0, 0, 0, 0, 0, 0
143 /* Table of special "verbs" like (*PRUNE). This is a short table, so it is
144 searched linearly. Put all the names into a single string, in order to reduce
145 the number of relocations when a shared library is dynamically linked. */
147 typedef struct verbitem {
152 static const char verbnames[] =
161 static verbitem verbs[] = {
171 static int verbcount = sizeof(verbs)/sizeof(verbitem);
174 /* Tables of names of POSIX character classes and their lengths. The names are
175 now all in a single string, to reduce the number of relocations when a shared
176 library is dynamically loaded. The list of lengths is terminated by a zero
177 length entry. The first three must be alpha, lower, upper, as this is assumed
178 for handling case independence. */
180 static const char posix_names[] =
181 "alpha\0" "lower\0" "upper\0" "alnum\0" "ascii\0" "blank\0"
182 "cntrl\0" "digit\0" "graph\0" "print\0" "punct\0" "space\0"
185 static const uschar posix_name_lengths[] = {
186 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 4, 6, 0 };
188 /* Table of class bit maps for each POSIX class. Each class is formed from a
189 base map, with an optional addition or removal of another map. Then, for some
190 classes, there is some additional tweaking: for [:blank:] the vertical space
191 characters are removed, and for [:alpha:] and [:alnum:] the underscore
192 character is removed. The triples in the table consist of the base map offset,
193 second map offset or -1 if no second map, and a non-negative value for map
194 addition or a negative value for map subtraction (if there are two maps). The
195 absolute value of the third field has these meanings: 0 => no tweaking, 1 =>
196 remove vertical space characters, 2 => remove underscore. */
198 static const int posix_class_maps[] = {
199 cbit_word, cbit_digit, -2, /* alpha */
200 cbit_lower, -1, 0, /* lower */
201 cbit_upper, -1, 0, /* upper */
202 cbit_word, -1, 2, /* alnum - word without underscore */
203 cbit_print, cbit_cntrl, 0, /* ascii */
204 cbit_space, -1, 1, /* blank - a GNU extension */
205 cbit_cntrl, -1, 0, /* cntrl */
206 cbit_digit, -1, 0, /* digit */
207 cbit_graph, -1, 0, /* graph */
208 cbit_print, -1, 0, /* print */
209 cbit_punct, -1, 0, /* punct */
210 cbit_space, -1, 0, /* space */
211 cbit_word, -1, 0, /* word - a Perl extension */
212 cbit_xdigit,-1, 0 /* xdigit */
216 #define STRING(a) # a
217 #define XSTRING(s) STRING(s)
219 /* The texts of compile-time error messages. These are "char *" because they
220 are passed to the outside world. Do not ever re-use any error number, because
221 they are documented. Always add a new error instead. Messages marked DEAD below
222 are no longer used. This used to be a table of strings, but in order to reduce
223 the number of relocations needed when a shared library is loaded dynamically,
224 it is now one long string. We cannot use a table of offsets, because the
225 lengths of inserts such as XSTRING(MAX_NAME_SIZE) are not known. Instead, we
226 simply count through to the one we want - this isn't a performance issue
227 because these strings are used only when there is a compilation error. */
229 static const char error_texts[] =
231 "\\ at end of pattern\0"
232 "\\c at end of pattern\0"
233 "unrecognized character follows \\\0"
234 "numbers out of order in {} quantifier\0"
236 "number too big in {} quantifier\0"
237 "missing terminating ] for character class\0"
238 "invalid escape sequence in character class\0"
239 "range out of order in character class\0"
240 "nothing to repeat\0"
242 "operand of unlimited repeat could match the empty string\0" /** DEAD **/
243 "internal error: unexpected repeat\0"
244 "unrecognized character after (? or (?-\0"
245 "POSIX named classes are supported only within a class\0"
248 "reference to non-existent subpattern\0"
249 "erroffset passed as NULL\0"
250 "unknown option bit(s) set\0"
251 "missing ) after comment\0"
252 "parentheses nested too deeply\0" /** DEAD **/
254 "regular expression is too large\0"
255 "failed to get memory\0"
256 "unmatched parentheses\0"
257 "internal error: code overflow\0"
258 "unrecognized character after (?<\0"
260 "lookbehind assertion is not fixed length\0"
261 "malformed number or name after (?(\0"
262 "conditional group contains more than two branches\0"
263 "assertion expected after (?(\0"
264 "(?R or (?[+-]digits must be followed by )\0"
266 "unknown POSIX class name\0"
267 "POSIX collating elements are not supported\0"
268 "this version of PCRE is not compiled with PCRE_UTF8 support\0"
269 "spare error\0" /** DEAD **/
270 "character value in \\x{...} sequence is too large\0"
272 "invalid condition (?(0)\0"
273 "\\C not allowed in lookbehind assertion\0"
274 "PCRE does not support \\L, \\l, \\N, \\U, or \\u\0"
275 "number after (?C is > 255\0"
276 "closing ) for (?C expected\0"
278 "recursive call could loop indefinitely\0"
279 "unrecognized character after (?P\0"
280 "syntax error in subpattern name (missing terminator)\0"
281 "two named subpatterns have the same name\0"
282 "invalid UTF-8 string\0"
284 "support for \\P, \\p, and \\X has not been compiled\0"
285 "malformed \\P or \\p sequence\0"
286 "unknown property name after \\P or \\p\0"
287 "subpattern name is too long (maximum " XSTRING(MAX_NAME_SIZE) " characters)\0"
288 "too many named subpatterns (maximum " XSTRING(MAX_NAME_COUNT) ")\0"
290 "repeated subpattern is too long\0" /** DEAD **/
291 "octal value is greater than \\377 (not in UTF-8 mode)\0"
292 "internal error: overran compiling workspace\0"
293 "internal error: previously-checked referenced subpattern not found\0"
294 "DEFINE group contains more than one branch\0"
296 "repeating a DEFINE group is not allowed\0"
297 "inconsistent NEWLINE options\0"
298 "\\g is not followed by a braced name or an optionally braced non-zero number\0"
299 "(?+ or (?- or (?(+ or (?(- must be followed by a non-zero number\0"
300 "(*VERB) with an argument is not supported\0"
302 "(*VERB) not recognized\0"
303 "number is too big\0"
304 "subpattern name expected\0"
305 "digit expected after (?+";
308 /* Table to identify digits and hex digits. This is used when compiling
309 patterns. Note that the tables in chartables are dependent on the locale, and
310 may mark arbitrary characters as digits - but the PCRE compiling code expects
311 to handle only 0-9, a-z, and A-Z as digits when compiling. That is why we have
312 a private table here. It costs 256 bytes, but it is a lot faster than doing
313 character value tests (at least in some simple cases I timed), and in some
314 applications one wants PCRE to compile efficiently as well as match
317 For convenience, we use the same bit definitions as in chartables:
320 0x08 hexadecimal digit
322 Then we can use ctype_digit and ctype_xdigit in the code. */
324 #ifndef EBCDIC /* This is the "normal" case, for ASCII systems */
325 static const unsigned char digitab[] =
327 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 0- 7 */
328 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 8- 15 */
329 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 16- 23 */
330 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 24- 31 */
331 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - ' */
332 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* ( - / */
333 0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c, /* 0 - 7 */
334 0x0c,0x0c,0x00,0x00,0x00,0x00,0x00,0x00, /* 8 - ? */
335 0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* @ - G */
336 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* H - O */
337 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* P - W */
338 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* X - _ */
339 0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* ` - g */
340 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* h - o */
341 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* p - w */
342 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* x -127 */
343 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 128-135 */
344 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 136-143 */
345 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 144-151 */
346 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 152-159 */
347 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 160-167 */
348 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 168-175 */
349 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 176-183 */
350 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 184-191 */
351 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 192-199 */
352 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 200-207 */
353 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 208-215 */
354 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 216-223 */
355 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 224-231 */
356 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 232-239 */
357 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 240-247 */
358 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00};/* 248-255 */
360 #else /* This is the "abnormal" case, for EBCDIC systems */
361 static const unsigned char digitab[] =
363 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 0- 7 0 */
364 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 8- 15 */
365 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 16- 23 10 */
366 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 24- 31 */
367 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 32- 39 20 */
368 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 40- 47 */
369 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 48- 55 30 */
370 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 56- 63 */
371 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - 71 40 */
372 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 72- | */
373 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* & - 87 50 */
374 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 88- 95 */
375 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - -103 60 */
376 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 104- ? */
377 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 112-119 70 */
378 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 120- " */
379 0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* 128- g 80 */
380 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* h -143 */
381 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 144- p 90 */
382 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* q -159 */
383 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 160- x A0 */
384 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* y -175 */
385 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* ^ -183 B0 */
386 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 184-191 */
387 0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* { - G C0 */
388 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* H -207 */
389 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* } - P D0 */
390 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* Q -223 */
391 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* \ - X E0 */
392 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* Y -239 */
393 0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c, /* 0 - 7 F0 */
394 0x0c,0x0c,0x00,0x00,0x00,0x00,0x00,0x00};/* 8 -255 */
396 static const unsigned char ebcdic_chartab[] = { /* chartable partial dup */
397 0x80,0x00,0x00,0x00,0x00,0x01,0x00,0x00, /* 0- 7 */
398 0x00,0x00,0x00,0x00,0x01,0x01,0x00,0x00, /* 8- 15 */
399 0x00,0x00,0x00,0x00,0x00,0x01,0x00,0x00, /* 16- 23 */
400 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 24- 31 */
401 0x00,0x00,0x00,0x00,0x00,0x01,0x00,0x00, /* 32- 39 */
402 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 40- 47 */
403 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 48- 55 */
404 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 56- 63 */
405 0x01,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - 71 */
406 0x00,0x00,0x00,0x80,0x00,0x80,0x80,0x80, /* 72- | */
407 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* & - 87 */
408 0x00,0x00,0x00,0x80,0x80,0x80,0x00,0x00, /* 88- 95 */
409 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - -103 */
410 0x00,0x00,0x00,0x00,0x00,0x10,0x00,0x80, /* 104- ? */
411 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 112-119 */
412 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 120- " */
413 0x00,0x1a,0x1a,0x1a,0x1a,0x1a,0x1a,0x12, /* 128- g */
414 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* h -143 */
415 0x00,0x12,0x12,0x12,0x12,0x12,0x12,0x12, /* 144- p */
416 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* q -159 */
417 0x00,0x00,0x12,0x12,0x12,0x12,0x12,0x12, /* 160- x */
418 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* y -175 */
419 0x80,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* ^ -183 */
420 0x00,0x00,0x80,0x00,0x00,0x00,0x00,0x00, /* 184-191 */
421 0x80,0x1a,0x1a,0x1a,0x1a,0x1a,0x1a,0x12, /* { - G */
422 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* H -207 */
423 0x00,0x12,0x12,0x12,0x12,0x12,0x12,0x12, /* } - P */
424 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* Q -223 */
425 0x00,0x00,0x12,0x12,0x12,0x12,0x12,0x12, /* \ - X */
426 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* Y -239 */
427 0x1c,0x1c,0x1c,0x1c,0x1c,0x1c,0x1c,0x1c, /* 0 - 7 */
428 0x1c,0x1c,0x00,0x00,0x00,0x00,0x00,0x00};/* 8 -255 */
432 /* Definition to allow mutual recursion */
435 compile_regex(int, int, uschar **, const uschar **, int *, BOOL, BOOL, int,
436 int *, int *, branch_chain *, compile_data *, int *);
440 /*************************************************
441 * Find an error text *
442 *************************************************/
444 /* The error texts are now all in one long string, to save on relocations. As
445 some of the text is of unknown length, we can't use a table of offsets.
446 Instead, just count through the strings. This is not a performance issue
447 because it happens only when there has been a compilation error.
449 Argument: the error number
450 Returns: pointer to the error string
454 find_error_text(int n)
456 const char *s = error_texts;
457 for (; n > 0; n--) while (*s++ != 0);
462 /*************************************************
464 *************************************************/
466 /* This function is called when a \ has been encountered. It either returns a
467 positive value for a simple escape such as \n, or a negative value which
468 encodes one of the more complicated things such as \d. A backreference to group
469 n is returned as -(ESC_REF + n); ESC_REF is the highest ESC_xxx macro. When
470 UTF-8 is enabled, a positive value greater than 255 may be returned. On entry,
471 ptr is pointing at the \. On exit, it is on the final character of the escape
475 ptrptr points to the pattern position pointer
476 errorcodeptr points to the errorcode variable
477 bracount number of previous extracting brackets
478 options the options bits
479 isclass TRUE if inside a character class
481 Returns: zero or positive => a data character
482 negative => a special escape sequence
483 on error, errorcodeptr is set
487 check_escape(const uschar **ptrptr, int *errorcodeptr, int bracount,
488 int options, BOOL isclass)
490 BOOL utf8 = (options & PCRE_UTF8) != 0;
491 const uschar *ptr = *ptrptr + 1;
494 GETCHARINCTEST(c, ptr); /* Get character value, increment pointer */
495 ptr--; /* Set pointer back to the last byte */
497 /* If backslash is at the end of the pattern, it's an error. */
499 if (c == 0) *errorcodeptr = ERR1;
501 /* Non-alphanumerics are literals. For digits or letters, do an initial lookup
502 in a table. A non-zero result is something that can be returned immediately.
503 Otherwise further processing may be required. */
505 #ifndef EBCDIC /* ASCII coding */
506 else if (c < '0' || c > 'z') {} /* Not alphanumeric */
507 else if ((i = escapes[c - '0']) != 0) c = i;
509 #else /* EBCDIC coding */
510 else if (c < 'a' || (ebcdic_chartab[c] & 0x0E) == 0) {} /* Not alphanumeric */
511 else if ((i = escapes[c - 0x48]) != 0) c = i;
514 /* Escapes that need further processing, or are illegal. */
518 const uschar *oldptr;
519 BOOL braced, negated;
523 /* A number of Perl escapes are not handled by PCRE. We give an explicit
531 *errorcodeptr = ERR37;
534 /* \g must be followed by a number, either plain or braced. If positive, it
535 is an absolute backreference. If negative, it is a relative backreference.
536 This is a Perl 5.10 feature. Perl 5.10 also supports \g{name} as a
537 reference to a named group. This is part of Perl's movement towards a
538 unified syntax for back references. As this is synonymous with \k{name}, we
539 fudge it up by pretending it really was \k. */
545 for (p = ptr+2; *p != 0 && *p != '}'; p++)
546 if (*p != '-' && (digitab[*p] & ctype_digit) == 0) break;
547 if (*p != 0 && *p != '}')
562 else negated = FALSE;
565 while ((digitab[ptr[1]] & ctype_digit) != 0)
566 c = c * 10 + *(++ptr) - '0';
570 *errorcodeptr = ERR61;
574 if (c == 0 || (braced && *(++ptr) != '}'))
576 *errorcodeptr = ERR57;
584 *errorcodeptr = ERR15;
587 c = bracount - (c - 1);
593 /* The handling of escape sequences consisting of a string of digits
594 starting with one that is not zero is not straightforward. By experiment,
595 the way Perl works seems to be as follows:
597 Outside a character class, the digits are read as a decimal number. If the
598 number is less than 10, or if there are that many previous extracting
599 left brackets, then it is a back reference. Otherwise, up to three octal
600 digits are read to form an escaped byte. Thus \123 is likely to be octal
601 123 (cf \0123, which is octal 012 followed by the literal 3). If the octal
602 value is greater than 377, the least significant 8 bits are taken. Inside a
603 character class, \ followed by a digit is always an octal number. */
605 case '1': case '2': case '3': case '4': case '5':
606 case '6': case '7': case '8': case '9':
612 while ((digitab[ptr[1]] & ctype_digit) != 0)
613 c = c * 10 + *(++ptr) - '0';
616 *errorcodeptr = ERR61;
619 if (c < 10 || c <= bracount)
624 ptr = oldptr; /* Put the pointer back and fall through */
627 /* Handle an octal number following \. If the first digit is 8 or 9, Perl
628 generates a binary zero byte and treats the digit as a following literal.
629 Thus we have to pull back the pointer by one. */
631 if ((c = *ptr) >= '8')
638 /* \0 always starts an octal number, but we may drop through to here with a
639 larger first octal digit. The original code used just to take the least
640 significant 8 bits of octal numbers (I think this is what early Perls used
641 to do). Nowadays we allow for larger numbers in UTF-8 mode, but no more
642 than 3 octal digits. */
646 while(i++ < 2 && ptr[1] >= '0' && ptr[1] <= '7')
647 c = c * 8 + *(++ptr) - '0';
648 if (!utf8 && c > 255) *errorcodeptr = ERR51;
651 /* \x is complicated. \x{ddd} is a character number which can be greater
652 than 0xff in utf8 mode, but only if the ddd are hex digits. If not, { is
653 treated as a data character. */
658 const uschar *pt = ptr + 2;
662 while ((digitab[*pt] & ctype_xdigit) != 0)
664 register int cc = *pt++;
665 if (c == 0 && cc == '0') continue; /* Leading zeroes */
668 #ifndef EBCDIC /* ASCII coding */
669 if (cc >= 'a') cc -= 32; /* Convert to upper case */
670 c = (c << 4) + cc - ((cc < 'A')? '0' : ('A' - 10));
671 #else /* EBCDIC coding */
672 if (cc >= 'a' && cc <= 'z') cc += 64; /* Convert to upper case */
673 c = (c << 4) + cc - ((cc >= '0')? '0' : ('A' - 10));
679 if (c < 0 || count > (utf8? 8 : 2)) *errorcodeptr = ERR34;
684 /* If the sequence of hex digits does not end with '}', then we don't
685 recognize this construct; fall through to the normal \x handling. */
688 /* Read just a single-byte hex-defined char */
691 while (i++ < 2 && (digitab[ptr[1]] & ctype_xdigit) != 0)
693 int cc; /* Some compilers don't like ++ */
694 cc = *(++ptr); /* in initializers */
695 #ifndef EBCDIC /* ASCII coding */
696 if (cc >= 'a') cc -= 32; /* Convert to upper case */
697 c = c * 16 + cc - ((cc < 'A')? '0' : ('A' - 10));
698 #else /* EBCDIC coding */
699 if (cc <= 'z') cc += 64; /* Convert to upper case */
700 c = c * 16 + cc - ((cc >= '0')? '0' : ('A' - 10));
705 /* For \c, a following letter is upper-cased; then the 0x40 bit is flipped.
706 This coding is ASCII-specific, but then the whole concept of \cx is
707 ASCII-specific. (However, an EBCDIC equivalent has now been added.) */
713 *errorcodeptr = ERR2;
717 #ifndef EBCDIC /* ASCII coding */
718 if (c >= 'a' && c <= 'z') c -= 32;
720 #else /* EBCDIC coding */
721 if (c >= 'a' && c <= 'z') c += 64;
726 /* PCRE_EXTRA enables extensions to Perl in the matter of escapes. Any
727 other alphanumeric following \ is an error if PCRE_EXTRA was set;
728 otherwise, for Perl compatibility, it is a literal. This code looks a bit
729 odd, but there used to be some cases other than the default, and there may
730 be again in future, so I haven't "optimized" it. */
733 if ((options & PCRE_EXTRA) != 0) switch(c)
736 *errorcodeptr = ERR3;
750 /*************************************************
752 *************************************************/
754 /* This function is called after \P or \p has been encountered, provided that
755 PCRE is compiled with support for Unicode properties. On entry, ptrptr is
756 pointing at the P or p. On exit, it is pointing at the final character of the
760 ptrptr points to the pattern position pointer
761 negptr points to a boolean that is set TRUE for negation else FALSE
762 dptr points to an int that is set to the detailed property value
763 errorcodeptr points to the error code variable
765 Returns: type value from ucp_type_table, or -1 for an invalid type
769 get_ucp(const uschar **ptrptr, BOOL *negptr, int *dptr, int *errorcodeptr)
772 const uschar *ptr = *ptrptr;
776 if (c == 0) goto ERROR_RETURN;
780 /* \P or \p can be followed by a name in {}, optionally preceded by ^ for
790 for (i = 0; i < (int)sizeof(name) - 1; i++)
793 if (c == 0) goto ERROR_RETURN;
797 if (c !='}') goto ERROR_RETURN;
801 /* Otherwise there is just one following character */
811 /* Search for a recognized property name using binary chop */
814 top = _pcre_utt_size;
818 i = (bot + top) >> 1;
819 c = strcmp(name, _pcre_utt_names + _pcre_utt[i].name_offset);
822 *dptr = _pcre_utt[i].value;
823 return _pcre_utt[i].type;
825 if (c > 0) bot = i + 1; else top = i;
828 *errorcodeptr = ERR47;
833 *errorcodeptr = ERR46;
842 /*************************************************
843 * Check for counted repeat *
844 *************************************************/
846 /* This function is called when a '{' is encountered in a place where it might
847 start a quantifier. It looks ahead to see if it really is a quantifier or not.
848 It is only a quantifier if it is one of the forms {ddd} {ddd,} or {ddd,ddd}
849 where the ddds are digits.
852 p pointer to the first char after '{'
854 Returns: TRUE or FALSE
858 is_counted_repeat(const uschar *p)
860 if ((digitab[*p++] & ctype_digit) == 0) return FALSE;
861 while ((digitab[*p] & ctype_digit) != 0) p++;
862 if (*p == '}') return TRUE;
864 if (*p++ != ',') return FALSE;
865 if (*p == '}') return TRUE;
867 if ((digitab[*p++] & ctype_digit) == 0) return FALSE;
868 while ((digitab[*p] & ctype_digit) != 0) p++;
875 /*************************************************
876 * Read repeat counts *
877 *************************************************/
879 /* Read an item of the form {n,m} and return the values. This is called only
880 after is_counted_repeat() has confirmed that a repeat-count quantifier exists,
881 so the syntax is guaranteed to be correct, but we need to check the values.
884 p pointer to first char after '{'
885 minp pointer to int for min
886 maxp pointer to int for max
887 returned as -1 if no max
888 errorcodeptr points to error code variable
890 Returns: pointer to '}' on success;
891 current ptr on error, with errorcodeptr set non-zero
894 static const uschar *
895 read_repeat_counts(const uschar *p, int *minp, int *maxp, int *errorcodeptr)
900 /* Read the minimum value and do a paranoid check: a negative value indicates
901 an integer overflow. */
903 while ((digitab[*p] & ctype_digit) != 0) min = min * 10 + *p++ - '0';
904 if (min < 0 || min > 65535)
906 *errorcodeptr = ERR5;
910 /* Read the maximum value if there is one, and again do a paranoid on its size.
911 Also, max must not be less than min. */
913 if (*p == '}') max = min; else
918 while((digitab[*p] & ctype_digit) != 0) max = max * 10 + *p++ - '0';
919 if (max < 0 || max > 65535)
921 *errorcodeptr = ERR5;
926 *errorcodeptr = ERR4;
932 /* Fill in the required variables, and pass back the pointer to the terminating
942 /*************************************************
943 * Find forward referenced subpattern *
944 *************************************************/
946 /* This function scans along a pattern's text looking for capturing
947 subpatterns, and counting them. If it finds a named pattern that matches the
948 name it is given, it returns its number. Alternatively, if the name is NULL, it
949 returns when it reaches a given numbered subpattern. This is used for forward
950 references to subpatterns. We know that if (?P< is encountered, the name will
951 be terminated by '>' because that is checked in the first pass.
954 ptr current position in the pattern
955 count current count of capturing parens so far encountered
956 name name to seek, or NULL if seeking a numbered subpattern
957 lorn name length, or subpattern number if name is NULL
958 xmode TRUE if we are in /x mode
960 Returns: the number of the named subpattern, or -1 if not found
964 find_parens(const uschar *ptr, int count, const uschar *name, int lorn,
967 const uschar *thisname;
969 for (; *ptr != 0; ptr++)
973 /* Skip over backslashed characters and also entire \Q...\E */
977 if (*(++ptr) == 0) return -1;
978 if (*ptr == 'Q') for (;;)
980 while (*(++ptr) != 0 && *ptr != '\\');
981 if (*ptr == 0) return -1;
982 if (*(++ptr) == 'E') break;
987 /* Skip over character classes */
991 while (*(++ptr) != ']')
993 if (*ptr == 0) return -1;
996 if (*(++ptr) == 0) return -1;
997 if (*ptr == 'Q') for (;;)
999 while (*(++ptr) != 0 && *ptr != '\\');
1000 if (*ptr == 0) return -1;
1001 if (*(++ptr) == 'E') break;
1009 /* Skip comments in /x mode */
1011 if (xmode && *ptr == '#')
1013 while (*(++ptr) != 0 && *ptr != '\n');
1014 if (*ptr == 0) return -1;
1018 /* An opening parens must now be a real metacharacter */
1020 if (*ptr != '(') continue;
1021 if (ptr[1] != '?' && ptr[1] != '*')
1024 if (name == NULL && count == lorn) return count;
1029 if (*ptr == 'P') ptr++; /* Allow optional P */
1031 /* We have to disambiguate (?<! and (?<= from (?<name> */
1033 if ((*ptr != '<' || ptr[1] == '!' || ptr[1] == '=') &&
1039 if (name == NULL && count == lorn) return count;
1041 if (term == '<') term = '>';
1043 while (*ptr != term) ptr++;
1044 if (name != NULL && lorn == ptr - thisname &&
1045 strncmp((const char *)name, (const char *)thisname, lorn) == 0)
1054 /*************************************************
1055 * Find first significant op code *
1056 *************************************************/
1058 /* This is called by several functions that scan a compiled expression looking
1059 for a fixed first character, or an anchoring op code etc. It skips over things
1060 that do not influence this. For some calls, a change of option is important.
1061 For some calls, it makes sense to skip negative forward and all backward
1062 assertions, and also the \b assertion; for others it does not.
1065 code pointer to the start of the group
1066 options pointer to external options
1067 optbit the option bit whose changing is significant, or
1069 skipassert TRUE if certain assertions are to be skipped
1071 Returns: pointer to the first significant opcode
1074 static const uschar*
1075 first_significant_code(const uschar *code, int *options, int optbit,
1083 if (optbit > 0 && ((int)code[1] & optbit) != (*options & optbit))
1084 *options = (int)code[1];
1090 case OP_ASSERTBACK_NOT:
1091 if (!skipassert) return code;
1092 do code += GET(code, 1); while (*code == OP_ALT);
1093 code += _pcre_OP_lengths[*code];
1096 case OP_WORD_BOUNDARY:
1097 case OP_NOT_WORD_BOUNDARY:
1098 if (!skipassert) return code;
1105 code += _pcre_OP_lengths[*code];
1112 /* Control never reaches here */
1118 /*************************************************
1119 * Find the fixed length of a pattern *
1120 *************************************************/
1122 /* Scan a pattern and compute the fixed length of subject that will match it,
1123 if the length is fixed. This is needed for dealing with backward assertions.
1124 In UTF8 mode, the result is in characters rather than bytes.
1127 code points to the start of the pattern (the bracket)
1128 options the compiling options
1130 Returns: the fixed length, or -1 if there is no fixed length,
1131 or -2 if \C was encountered
1135 find_fixedlength(uschar *code, int options)
1139 register int branchlength = 0;
1140 register uschar *cc = code + 1 + LINK_SIZE;
1142 /* Scan along the opcodes for this branch. If we get to the end of the
1143 branch, check the length against that of the other branches. */
1148 register int op = *cc;
1155 d = find_fixedlength(cc + ((op == OP_CBRA)? 2:0), options);
1156 if (d < 0) return d;
1158 do cc += GET(cc, 1); while (*cc == OP_ALT);
1159 cc += 1 + LINK_SIZE;
1162 /* Reached end of a branch; if it's a ket it is the end of a nested
1163 call. If it's ALT it is an alternation in a nested call. If it is
1164 END it's the end of the outer call. All can be handled by the same code. */
1171 if (length < 0) length = branchlength;
1172 else if (length != branchlength) return -1;
1173 if (*cc != OP_ALT) return length;
1174 cc += 1 + LINK_SIZE;
1178 /* Skip over assertive subpatterns */
1183 case OP_ASSERTBACK_NOT:
1184 do cc += GET(cc, 1); while (*cc == OP_ALT);
1187 /* Skip over things that don't match chars */
1201 case OP_NOT_WORD_BOUNDARY:
1202 case OP_WORD_BOUNDARY:
1203 cc += _pcre_OP_lengths[*cc];
1206 /* Handle literal characters */
1214 if ((options & PCRE_UTF8) != 0)
1216 while ((*cc & 0xc0) == 0x80) cc++;
1221 /* Handle exact repetitions. The count is already in characters, but we
1222 need to skip over a multibyte character in UTF8 mode. */
1225 branchlength += GET2(cc,1);
1228 if ((options & PCRE_UTF8) != 0)
1230 while((*cc & 0x80) == 0x80) cc++;
1236 branchlength += GET2(cc,1);
1237 if (cc[3] == OP_PROP || cc[3] == OP_NOTPROP) cc += 2;
1241 /* Handle single-char matchers */
1250 case OP_NOT_WHITESPACE:
1252 case OP_NOT_WORDCHAR:
1259 /* The single-byte matcher isn't allowed */
1264 /* Check a class for variable quantification */
1268 cc += GET(cc, 1) - 33;
1286 if (GET2(cc,1) != GET2(cc,3)) return -1;
1287 branchlength += GET2(cc,1);
1296 /* Anything else is variable length */
1302 /* Control never gets here */
1308 /*************************************************
1309 * Scan compiled regex for numbered bracket *
1310 *************************************************/
1312 /* This little function scans through a compiled pattern until it finds a
1313 capturing bracket with the given number.
1316 code points to start of expression
1317 utf8 TRUE in UTF-8 mode
1318 number the required bracket number
1320 Returns: pointer to the opcode for the bracket, or NULL if not found
1323 static const uschar *
1324 find_bracket(const uschar *code, BOOL utf8, int number)
1328 register int c = *code;
1329 if (c == OP_END) return NULL;
1331 /* XCLASS is used for classes that cannot be represented just by a bit
1332 map. This includes negated single high-valued characters. The length in
1333 the table is zero; the actual length is stored in the compiled code. */
1335 if (c == OP_XCLASS) code += GET(code, 1);
1337 /* Handle capturing bracket */
1339 else if (c == OP_CBRA)
1341 int n = GET2(code, 1+LINK_SIZE);
1342 if (n == number) return (uschar *)code;
1343 code += _pcre_OP_lengths[c];
1346 /* Otherwise, we can get the item's length from the table, except that for
1347 repeated character types, we have to test for \p and \P, which have an extra
1348 two bytes of parameters. */
1355 case OP_TYPEMINSTAR:
1357 case OP_TYPEMINPLUS:
1359 case OP_TYPEMINQUERY:
1360 case OP_TYPEPOSSTAR:
1361 case OP_TYPEPOSPLUS:
1362 case OP_TYPEPOSQUERY:
1363 if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
1367 case OP_TYPEMINUPTO:
1369 case OP_TYPEPOSUPTO:
1370 if (code[3] == OP_PROP || code[3] == OP_NOTPROP) code += 2;
1374 /* Add in the fixed length from the table */
1376 code += _pcre_OP_lengths[c];
1378 /* In UTF-8 mode, opcodes that are followed by a character may be followed by
1379 a multi-byte character. The length in the table is a minimum, so we have to
1380 arrange to skip the extra bytes. */
1400 if (code[-1] >= 0xc0) code += _pcre_utf8_table4[code[-1] & 0x3f];
1410 /*************************************************
1411 * Scan compiled regex for recursion reference *
1412 *************************************************/
1414 /* This little function scans through a compiled pattern until it finds an
1415 instance of OP_RECURSE.
1418 code points to start of expression
1419 utf8 TRUE in UTF-8 mode
1421 Returns: pointer to the opcode for OP_RECURSE, or NULL if not found
1424 static const uschar *
1425 find_recurse(const uschar *code, BOOL utf8)
1429 register int c = *code;
1430 if (c == OP_END) return NULL;
1431 if (c == OP_RECURSE) return code;
1433 /* XCLASS is used for classes that cannot be represented just by a bit
1434 map. This includes negated single high-valued characters. The length in
1435 the table is zero; the actual length is stored in the compiled code. */
1437 if (c == OP_XCLASS) code += GET(code, 1);
1439 /* Otherwise, we can get the item's length from the table, except that for
1440 repeated character types, we have to test for \p and \P, which have an extra
1441 two bytes of parameters. */
1448 case OP_TYPEMINSTAR:
1450 case OP_TYPEMINPLUS:
1452 case OP_TYPEMINQUERY:
1453 case OP_TYPEPOSSTAR:
1454 case OP_TYPEPOSPLUS:
1455 case OP_TYPEPOSQUERY:
1456 if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
1459 case OP_TYPEPOSUPTO:
1461 case OP_TYPEMINUPTO:
1463 if (code[3] == OP_PROP || code[3] == OP_NOTPROP) code += 2;
1467 /* Add in the fixed length from the table */
1469 code += _pcre_OP_lengths[c];
1471 /* In UTF-8 mode, opcodes that are followed by a character may be followed
1472 by a multi-byte character. The length in the table is a minimum, so we have
1473 to arrange to skip the extra bytes. */
1493 if (code[-1] >= 0xc0) code += _pcre_utf8_table4[code[-1] & 0x3f];
1503 /*************************************************
1504 * Scan compiled branch for non-emptiness *
1505 *************************************************/
1507 /* This function scans through a branch of a compiled pattern to see whether it
1508 can match the empty string or not. It is called from could_be_empty()
1509 below and from compile_branch() when checking for an unlimited repeat of a
1510 group that can match nothing. Note that first_significant_code() skips over
1511 backward and negative forward assertions when its final argument is TRUE. If we
1512 hit an unclosed bracket, we return "empty" - this means we've struck an inner
1513 bracket whose current branch will already have been scanned.
1516 code points to start of search
1517 endcode points to where to stop
1518 utf8 TRUE if in UTF8 mode
1520 Returns: TRUE if what is matched could be empty
1524 could_be_empty_branch(const uschar *code, const uschar *endcode, BOOL utf8)
1527 for (code = first_significant_code(code + _pcre_OP_lengths[*code], NULL, 0, TRUE);
1529 code = first_significant_code(code + _pcre_OP_lengths[c], NULL, 0, TRUE))
1531 const uschar *ccode;
1535 /* Skip over forward assertions; the other assertions are skipped by
1536 first_significant_code() with a TRUE final argument. */
1540 do code += GET(code, 1); while (*code == OP_ALT);
1545 /* Groups with zero repeats can of course be empty; skip them. */
1547 if (c == OP_BRAZERO || c == OP_BRAMINZERO)
1549 code += _pcre_OP_lengths[c];
1550 do code += GET(code, 1); while (*code == OP_ALT);
1555 /* For other groups, scan the branches. */
1557 if (c == OP_BRA || c == OP_CBRA || c == OP_ONCE || c == OP_COND)
1560 if (GET(code, 1) == 0) return TRUE; /* Hit unclosed bracket */
1562 /* Scan a closed bracket */
1564 empty_branch = FALSE;
1567 if (!empty_branch && could_be_empty_branch(code, endcode, utf8))
1568 empty_branch = TRUE;
1569 code += GET(code, 1);
1571 while (*code == OP_ALT);
1572 if (!empty_branch) return FALSE; /* All branches are non-empty */
1577 /* Handle the other opcodes */
1581 /* Check for quantifiers after a class. XCLASS is used for classes that
1582 cannot be represented just by a bit map. This includes negated single
1583 high-valued characters. The length in _pcre_OP_lengths[] is zero; the
1584 actual length is stored in the compiled code, so we must update "code"
1589 ccode = code += GET(code, 1);
1590 goto CHECK_CLASS_REPEAT;
1603 case OP_CRSTAR: /* These could be empty; continue */
1609 default: /* Non-repeat => class must match */
1610 case OP_CRPLUS: /* These repeats aren't empty */
1616 if (GET2(ccode, 1) > 0) return FALSE; /* Minimum > 0 */
1621 /* Opcodes that must match a character */
1628 case OP_NOT_WHITESPACE:
1630 case OP_NOT_WORDCHAR:
1646 case OP_TYPEMINPLUS:
1647 case OP_TYPEPOSPLUS:
1651 /* These are going to continue, as they may be empty, but we have to
1652 fudge the length for the \p and \P cases. */
1655 case OP_TYPEMINSTAR:
1656 case OP_TYPEPOSSTAR:
1658 case OP_TYPEMINQUERY:
1659 case OP_TYPEPOSQUERY:
1660 if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
1663 /* Same for these */
1666 case OP_TYPEMINUPTO:
1667 case OP_TYPEPOSUPTO:
1668 if (code[3] == OP_PROP || code[3] == OP_NOTPROP) code += 2;
1679 /* In UTF-8 mode, STAR, MINSTAR, POSSTAR, QUERY, MINQUERY, POSQUERY, UPTO,
1680 MINUPTO, and POSUPTO may be followed by a multibyte character */
1692 if (utf8) while ((code[2] & 0xc0) == 0x80) code++;
1703 /*************************************************
1704 * Scan compiled regex for non-emptiness *
1705 *************************************************/
1707 /* This function is called to check for left recursive calls. We want to check
1708 the current branch of the current pattern to see if it could match the empty
1709 string. If it could, we must look outwards for branches at other levels,
1710 stopping when we pass beyond the bracket which is the subject of the recursion.
1713 code points to start of the recursion
1714 endcode points to where to stop (current RECURSE item)
1715 bcptr points to the chain of current (unclosed) branch starts
1716 utf8 TRUE if in UTF-8 mode
1718 Returns: TRUE if what is matched could be empty
1722 could_be_empty(const uschar *code, const uschar *endcode, branch_chain *bcptr,
1725 while (bcptr != NULL && bcptr->current >= code)
1727 if (!could_be_empty_branch(bcptr->current, endcode, utf8)) return FALSE;
1728 bcptr = bcptr->outer;
1735 /*************************************************
1736 * Check for POSIX class syntax *
1737 *************************************************/
1739 /* This function is called when the sequence "[:" or "[." or "[=" is
1740 encountered in a character class. It checks whether this is followed by a
1741 sequence of characters terminated by a matching ":]" or ".]" or "=]". If we
1742 reach an unescaped ']' without the special preceding character, return FALSE.
1744 Originally, this function only recognized a sequence of letters between the
1745 terminators, but it seems that Perl recognizes any sequence of characters,
1746 though of course unknown POSIX names are subsequently rejected. Perl gives an
1747 "Unknown POSIX class" error for [:f\oo:] for example, where previously PCRE
1748 didn't consider this to be a POSIX class. Likewise for [:1234:].
1750 The problem in trying to be exactly like Perl is in the handling of escapes. We
1751 have to be sure that [abc[:x\]pqr] is *not* treated as containing a POSIX
1752 class, but [abc[:x\]pqr:]] is (so that an error can be generated). The code
1753 below handles the special case of \], but does not try to do any other escape
1754 processing. This makes it different from Perl for cases such as [:l\ower:]
1755 where Perl recognizes it as the POSIX class "lower" but PCRE does not recognize
1756 "l\ower". This is a lesser evil that not diagnosing bad classes when Perl does,
1760 ptr pointer to the initial [
1761 endptr where to return the end pointer
1763 Returns: TRUE or FALSE
1767 check_posix_syntax(const uschar *ptr, const uschar **endptr)
1769 int terminator; /* Don't combine these lines; the Solaris cc */
1770 terminator = *(++ptr); /* compiler warns about "non-constant" initializer. */
1771 for (++ptr; *ptr != 0; ptr++)
1773 if (*ptr == '\\' && ptr[1] == ']') ptr++; else
1775 if (*ptr == ']') return FALSE;
1776 if (*ptr == terminator && ptr[1] == ']')
1789 /*************************************************
1790 * Check POSIX class name *
1791 *************************************************/
1793 /* This function is called to check the name given in a POSIX-style class entry
1797 ptr points to the first letter
1798 len the length of the name
1800 Returns: a value representing the name, or -1 if unknown
1804 check_posix_name(const uschar *ptr, int len)
1806 const char *pn = posix_names;
1807 register int yield = 0;
1808 while (posix_name_lengths[yield] != 0)
1810 if (len == posix_name_lengths[yield] &&
1811 strncmp((const char *)ptr, pn, len) == 0) return yield;
1812 pn += posix_name_lengths[yield] + 1;
1819 /*************************************************
1820 * Adjust OP_RECURSE items in repeated group *
1821 *************************************************/
1823 /* OP_RECURSE items contain an offset from the start of the regex to the group
1824 that is referenced. This means that groups can be replicated for fixed
1825 repetition simply by copying (because the recursion is allowed to refer to
1826 earlier groups that are outside the current group). However, when a group is
1827 optional (i.e. the minimum quantifier is zero), OP_BRAZERO is inserted before
1828 it, after it has been compiled. This means that any OP_RECURSE items within it
1829 that refer to the group itself or any contained groups have to have their
1830 offsets adjusted. That one of the jobs of this function. Before it is called,
1831 the partially compiled regex must be temporarily terminated with OP_END.
1833 This function has been extended with the possibility of forward references for
1834 recursions and subroutine calls. It must also check the list of such references
1835 for the group we are dealing with. If it finds that one of the recursions in
1836 the current group is on this list, it adjusts the offset in the list, not the
1837 value in the reference (which is a group number).
1840 group points to the start of the group
1841 adjust the amount by which the group is to be moved
1842 utf8 TRUE in UTF-8 mode
1843 cd contains pointers to tables etc.
1844 save_hwm the hwm forward reference pointer at the start of the group
1850 adjust_recurse(uschar *group, int adjust, BOOL utf8, compile_data *cd,
1853 uschar *ptr = group;
1855 while ((ptr = (uschar *)find_recurse(ptr, utf8)) != NULL)
1860 /* See if this recursion is on the forward reference list. If so, adjust the
1863 for (hc = save_hwm; hc < cd->hwm; hc += LINK_SIZE)
1865 offset = GET(hc, 0);
1866 if (cd->start_code + offset == ptr + 1)
1868 PUT(hc, 0, offset + adjust);
1873 /* Otherwise, adjust the recursion offset if it's after the start of this
1878 offset = GET(ptr, 1);
1879 if (cd->start_code + offset >= group) PUT(ptr, 1, offset + adjust);
1882 ptr += 1 + LINK_SIZE;
1888 /*************************************************
1889 * Insert an automatic callout point *
1890 *************************************************/
1892 /* This function is called when the PCRE_AUTO_CALLOUT option is set, to insert
1893 callout points before each pattern item.
1896 code current code pointer
1897 ptr current pattern pointer
1898 cd pointers to tables etc
1900 Returns: new code pointer
1904 auto_callout(uschar *code, const uschar *ptr, compile_data *cd)
1906 *code++ = OP_CALLOUT;
1908 PUT(code, 0, ptr - cd->start_pattern); /* Pattern offset */
1909 PUT(code, LINK_SIZE, 0); /* Default length */
1910 return code + 2*LINK_SIZE;
1915 /*************************************************
1916 * Complete a callout item *
1917 *************************************************/
1919 /* A callout item contains the length of the next item in the pattern, which
1920 we can't fill in till after we have reached the relevant point. This is used
1921 for both automatic and manual callouts.
1924 previous_callout points to previous callout item
1925 ptr current pattern pointer
1926 cd pointers to tables etc
1932 complete_callout(uschar *previous_callout, const uschar *ptr, compile_data *cd)
1934 int length = ptr - cd->start_pattern - GET(previous_callout, 2);
1935 PUT(previous_callout, 2 + LINK_SIZE, length);
1941 /*************************************************
1942 * Get othercase range *
1943 *************************************************/
1945 /* This function is passed the start and end of a class range, in UTF-8 mode
1946 with UCP support. It searches up the characters, looking for internal ranges of
1947 characters in the "other" case. Each call returns the next one, updating the
1951 cptr points to starting character value; updated
1953 ocptr where to put start of othercase range
1954 odptr where to put end of othercase range
1956 Yield: TRUE when range returned; FALSE when no more
1960 get_othercase_range(unsigned int *cptr, unsigned int d, unsigned int *ocptr,
1961 unsigned int *odptr)
1963 unsigned int c, othercase, next;
1965 for (c = *cptr; c <= d; c++)
1966 { if ((othercase = _pcre_ucp_othercase(c)) != NOTACHAR) break; }
1968 if (c > d) return FALSE;
1971 next = othercase + 1;
1973 for (++c; c <= d; c++)
1975 if (_pcre_ucp_othercase(c) != next) break;
1984 #endif /* SUPPORT_UCP */
1988 /*************************************************
1989 * Check if auto-possessifying is possible *
1990 *************************************************/
1992 /* This function is called for unlimited repeats of certain items, to see
1993 whether the next thing could possibly match the repeated item. If not, it makes
1994 sense to automatically possessify the repeated item.
1997 op_code the repeated op code
1998 this data for this item, depends on the opcode
1999 utf8 TRUE in UTF-8 mode
2000 utf8_char used for utf8 character bytes, NULL if not relevant
2001 ptr next character in pattern
2002 options options bits
2003 cd contains pointers to tables etc.
2005 Returns: TRUE if possessifying is wanted
2009 check_auto_possessive(int op_code, int item, BOOL utf8, uschar *utf8_char,
2010 const uschar *ptr, int options, compile_data *cd)
2014 /* Skip whitespace and comments in extended mode */
2016 if ((options & PCRE_EXTENDED) != 0)
2020 while ((cd->ctypes[*ptr] & ctype_space) != 0) ptr++;
2023 while (*(++ptr) != 0)
2024 if (IS_NEWLINE(ptr)) { ptr += cd->nllen; break; }
2030 /* If the next item is one that we can handle, get its value. A non-negative
2031 value is a character, a negative value is an escape value. */
2035 int temperrorcode = 0;
2036 next = check_escape(&ptr, &temperrorcode, cd->bracount, options, FALSE);
2037 if (temperrorcode != 0) return FALSE;
2038 ptr++; /* Point after the escape sequence */
2041 else if ((cd->ctypes[*ptr] & ctype_meta) == 0)
2044 if (utf8) { GETCHARINC(next, ptr); } else
2051 /* Skip whitespace and comments in extended mode */
2053 if ((options & PCRE_EXTENDED) != 0)
2057 while ((cd->ctypes[*ptr] & ctype_space) != 0) ptr++;
2060 while (*(++ptr) != 0)
2061 if (IS_NEWLINE(ptr)) { ptr += cd->nllen; break; }
2067 /* If the next thing is itself optional, we have to give up. */
2069 if (*ptr == '*' || *ptr == '?' || strncmp((char *)ptr, "{0,", 3) == 0)
2072 /* Now compare the next item with the previous opcode. If the previous is a
2073 positive single character match, "item" either contains the character or, if
2074 "item" is greater than 127 in utf8 mode, the character's bytes are in
2078 /* Handle cases when the next item is a character. */
2080 if (next >= 0) switch(op_code)
2084 if (utf8 && item > 127) { GETCHAR(item, utf8_char); }
2086 return item != next;
2088 /* For CHARNC (caseless character) we must check the other case. If we have
2089 Unicode property support, we can use it to test the other case of
2090 high-valued characters. */
2094 if (utf8 && item > 127) { GETCHAR(item, utf8_char); }
2096 if (item == next) return FALSE;
2100 unsigned int othercase;
2101 if (next < 128) othercase = cd->fcc[next]; else
2103 othercase = _pcre_ucp_othercase((unsigned int)next);
2105 othercase = NOTACHAR;
2107 return (unsigned int)item != othercase;
2110 #endif /* SUPPORT_UTF8 */
2111 return (item != cd->fcc[next]); /* Non-UTF-8 mode */
2113 /* For OP_NOT, "item" must be a single-byte character. */
2116 if (next < 0) return FALSE; /* Not a character */
2117 if (item == next) return TRUE;
2118 if ((options & PCRE_CASELESS) == 0) return FALSE;
2122 unsigned int othercase;
2123 if (next < 128) othercase = cd->fcc[next]; else
2125 othercase = _pcre_ucp_othercase(next);
2127 othercase = NOTACHAR;
2129 return (unsigned int)item == othercase;
2132 #endif /* SUPPORT_UTF8 */
2133 return (item == cd->fcc[next]); /* Non-UTF-8 mode */
2136 return next > 127 || (cd->ctypes[next] & ctype_digit) == 0;
2139 return next <= 127 && (cd->ctypes[next] & ctype_digit) != 0;
2142 return next > 127 || (cd->ctypes[next] & ctype_space) == 0;
2144 case OP_NOT_WHITESPACE:
2145 return next <= 127 && (cd->ctypes[next] & ctype_space) != 0;
2148 return next > 127 || (cd->ctypes[next] & ctype_word) == 0;
2150 case OP_NOT_WORDCHAR:
2151 return next <= 127 && (cd->ctypes[next] & ctype_word) != 0;
2176 return op_code != OP_HSPACE;
2178 return op_code == OP_HSPACE;
2192 return op_code != OP_VSPACE;
2194 return op_code == OP_VSPACE;
2202 /* Handle the case when the next item is \d, \s, etc. */
2209 if (utf8 && item > 127) { GETCHAR(item, utf8_char); }
2214 return item > 127 || (cd->ctypes[item] & ctype_digit) == 0;
2217 return item <= 127 && (cd->ctypes[item] & ctype_digit) != 0;
2220 return item > 127 || (cd->ctypes[item] & ctype_space) == 0;
2223 return item <= 127 && (cd->ctypes[item] & ctype_space) != 0;
2226 return item > 127 || (cd->ctypes[item] & ctype_word) == 0;
2229 return item <= 127 && (cd->ctypes[item] & ctype_word) != 0;
2254 return -next != ESC_h;
2256 return -next == ESC_h;
2270 return -next != ESC_v;
2272 return -next == ESC_v;
2280 return next == -ESC_D || next == -ESC_s || next == -ESC_W ||
2281 next == -ESC_h || next == -ESC_v;
2284 return next == -ESC_d;
2287 return next == -ESC_S || next == -ESC_d || next == -ESC_w;
2289 case OP_NOT_WHITESPACE:
2290 return next == -ESC_s || next == -ESC_h || next == -ESC_v;
2293 return next == -ESC_S || next == -ESC_H || next == -ESC_d || next == -ESC_w;
2296 return next == -ESC_h;
2298 /* Can't have \S in here because VT matches \S (Perl anomaly) */
2300 return next == -ESC_V || next == -ESC_d || next == -ESC_w;
2303 return next == -ESC_v;
2306 return next == -ESC_W || next == -ESC_s || next == -ESC_h || next == -ESC_v;
2308 case OP_NOT_WORDCHAR:
2309 return next == -ESC_w || next == -ESC_d;
2315 /* Control does not reach here */
2320 /*************************************************
2321 * Compile one branch *
2322 *************************************************/
2324 /* Scan the pattern, compiling it into the a vector. If the options are
2325 changed during the branch, the pointer is used to change the external options
2326 bits. This function is used during the pre-compile phase when we are trying
2327 to find out the amount of memory needed, as well as during the real compile
2328 phase. The value of lengthptr distinguishes the two phases.
2331 optionsptr pointer to the option bits
2332 codeptr points to the pointer to the current code point
2333 ptrptr points to the current pattern pointer
2334 errorcodeptr points to error code variable
2335 firstbyteptr set to initial literal character, or < 0 (REQ_UNSET, REQ_NONE)
2336 reqbyteptr set to the last literal character required, else < 0
2337 bcptr points to current branch chain
2338 cd contains pointers to tables etc.
2339 lengthptr NULL during the real compile phase
2340 points to length accumulator during pre-compile phase
2342 Returns: TRUE on success
2343 FALSE, with *errorcodeptr set non-zero on error
2347 compile_branch(int *optionsptr, uschar **codeptr, const uschar **ptrptr,
2348 int *errorcodeptr, int *firstbyteptr, int *reqbyteptr, branch_chain *bcptr,
2349 compile_data *cd, int *lengthptr)
2351 int repeat_type, op_type;
2352 int repeat_min = 0, repeat_max = 0; /* To please picky compilers */
2354 int greedy_default, greedy_non_default;
2355 int firstbyte, reqbyte;
2356 int zeroreqbyte, zerofirstbyte;
2357 int req_caseopt, reqvary, tempreqvary;
2358 int options = *optionsptr;
2359 int after_manual_callout = 0;
2360 int length_prevgroup = 0;
2362 register uschar *code = *codeptr;
2363 uschar *last_code = code;
2364 uschar *orig_code = code;
2366 BOOL inescq = FALSE;
2367 BOOL groupsetfirstbyte = FALSE;
2368 const uschar *ptr = *ptrptr;
2369 const uschar *tempptr;
2370 uschar *previous = NULL;
2371 uschar *previous_callout = NULL;
2372 uschar *save_hwm = NULL;
2373 uschar classbits[32];
2377 BOOL utf8 = (options & PCRE_UTF8) != 0;
2378 uschar *class_utf8data;
2379 uschar *class_utf8data_base;
2380 uschar utf8_char[6];
2383 uschar *utf8_char = NULL;
2387 if (lengthptr != NULL) DPRINTF((">> start branch\n"));
2390 /* Set up the default and non-default settings for greediness */
2392 greedy_default = ((options & PCRE_UNGREEDY) != 0);
2393 greedy_non_default = greedy_default ^ 1;
2395 /* Initialize no first byte, no required byte. REQ_UNSET means "no char
2396 matching encountered yet". It gets changed to REQ_NONE if we hit something that
2397 matches a non-fixed char first char; reqbyte just remains unset if we never
2400 When we hit a repeat whose minimum is zero, we may have to adjust these values
2401 to take the zero repeat into account. This is implemented by setting them to
2402 zerofirstbyte and zeroreqbyte when such a repeat is encountered. The individual
2403 item types that can be repeated set these backoff variables appropriately. */
2405 firstbyte = reqbyte = zerofirstbyte = zeroreqbyte = REQ_UNSET;
2407 /* The variable req_caseopt contains either the REQ_CASELESS value or zero,
2408 according to the current setting of the caseless flag. REQ_CASELESS is a bit
2409 value > 255. It is added into the firstbyte or reqbyte variables to record the
2410 case status of the value. This is used only for ASCII characters. */
2412 req_caseopt = ((options & PCRE_CASELESS) != 0)? REQ_CASELESS : 0;
2414 /* Switch on next character until the end of the branch */
2419 BOOL should_flip_negation;
2420 BOOL possessive_quantifier;
2423 BOOL reset_bracount;
2424 int class_charcount;
2436 /* Get next byte in the pattern */
2440 /* If we are in the pre-compile phase, accumulate the length used for the
2441 previous cycle of this loop. */
2443 if (lengthptr != NULL)
2446 if (code > cd->hwm) cd->hwm = code; /* High water info */
2448 if (code > cd->start_workspace + COMPILE_WORK_SIZE) /* Check for overrun */
2450 *errorcodeptr = ERR52;
2454 /* There is at least one situation where code goes backwards: this is the
2455 case of a zero quantifier after a class (e.g. [ab]{0}). At compile time,
2456 the class is simply eliminated. However, it is created first, so we have to
2457 allow memory for it. Therefore, don't ever reduce the length at this point.
2460 if (code < last_code) code = last_code;
2462 /* Paranoid check for integer overflow */
2464 if (OFLOW_MAX - *lengthptr < code - last_code)
2466 *errorcodeptr = ERR20;
2470 *lengthptr += code - last_code;
2471 DPRINTF(("length=%d added %d c=%c\n", *lengthptr, code - last_code, c));
2473 /* If "previous" is set and it is not at the start of the work space, move
2474 it back to there, in order to avoid filling up the work space. Otherwise,
2475 if "previous" is NULL, reset the current code pointer to the start. */
2477 if (previous != NULL)
2479 if (previous > orig_code)
2481 memmove(orig_code, previous, code - previous);
2482 code -= previous - orig_code;
2483 previous = orig_code;
2486 else code = orig_code;
2488 /* Remember where this code item starts so we can pick up the length
2494 /* In the real compile phase, just check the workspace used by the forward
2497 else if (cd->hwm > cd->start_workspace + COMPILE_WORK_SIZE)
2499 *errorcodeptr = ERR52;
2503 /* If in \Q...\E, check for the end; if not, we have a literal */
2505 if (inescq && c != 0)
2507 if (c == '\\' && ptr[1] == 'E')
2515 if (previous_callout != NULL)
2517 if (lengthptr == NULL) /* Don't attempt in pre-compile phase */
2518 complete_callout(previous_callout, ptr, cd);
2519 previous_callout = NULL;
2521 if ((options & PCRE_AUTO_CALLOUT) != 0)
2523 previous_callout = code;
2524 code = auto_callout(code, ptr, cd);
2530 /* Fill in length of a previous callout, except when the next thing is
2533 is_quantifier = c == '*' || c == '+' || c == '?' ||
2534 (c == '{' && is_counted_repeat(ptr+1));
2536 if (!is_quantifier && previous_callout != NULL &&
2537 after_manual_callout-- <= 0)
2539 if (lengthptr == NULL) /* Don't attempt in pre-compile phase */
2540 complete_callout(previous_callout, ptr, cd);
2541 previous_callout = NULL;
2544 /* In extended mode, skip white space and comments */
2546 if ((options & PCRE_EXTENDED) != 0)
2548 if ((cd->ctypes[c] & ctype_space) != 0) continue;
2551 while (*(++ptr) != 0)
2553 if (IS_NEWLINE(ptr)) { ptr += cd->nllen - 1; break; }
2555 if (*ptr != 0) continue;
2557 /* Else fall through to handle end of string */
2562 /* No auto callout for quantifiers. */
2564 if ((options & PCRE_AUTO_CALLOUT) != 0 && !is_quantifier)
2566 previous_callout = code;
2567 code = auto_callout(code, ptr, cd);
2572 /* ===================================================================*/
2573 case 0: /* The branch terminates at string end */
2574 case '|': /* or | or ) */
2576 *firstbyteptr = firstbyte;
2577 *reqbyteptr = reqbyte;
2580 if (lengthptr != NULL)
2582 if (OFLOW_MAX - *lengthptr < code - last_code)
2584 *errorcodeptr = ERR20;
2587 *lengthptr += code - last_code; /* To include callout length */
2588 DPRINTF((">> end branch\n"));
2593 /* ===================================================================*/
2594 /* Handle single-character metacharacters. In multiline mode, ^ disables
2595 the setting of any following char as a first character. */
2598 if ((options & PCRE_MULTILINE) != 0)
2600 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
2611 /* There can never be a first char if '.' is first, whatever happens about
2612 repeats. The value of reqbyte doesn't change either. */
2615 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
2616 zerofirstbyte = firstbyte;
2617 zeroreqbyte = reqbyte;
2623 /* ===================================================================*/
2624 /* Character classes. If the included characters are all < 256, we build a
2625 32-byte bitmap of the permitted characters, except in the special case
2626 where there is only one such character. For negated classes, we build the
2627 map as usual, then invert it at the end. However, we use a different opcode
2628 so that data characters > 255 can be handled correctly.
2630 If the class contains characters outside the 0-255 range, a different
2631 opcode is compiled. It may optionally have a bit map for characters < 256,
2632 but those above are are explicitly listed afterwards. A flag byte tells
2633 whether the bitmap is present, and whether this is a negated class or not.
2639 /* PCRE supports POSIX class stuff inside a class. Perl gives an error if
2640 they are encountered at the top level, so we'll do that too. */
2642 if ((ptr[1] == ':' || ptr[1] == '.' || ptr[1] == '=') &&
2643 check_posix_syntax(ptr, &tempptr))
2645 *errorcodeptr = (ptr[1] == ':')? ERR13 : ERR31;
2649 /* If the first character is '^', set the negation flag and skip it. Also,
2650 if the first few characters (either before or after ^) are \Q\E or \E we
2651 skip them too. This makes for compatibility with Perl. */
2653 negate_class = FALSE;
2659 if (ptr[1] == 'E') ptr++;
2660 else if (strncmp((const char *)ptr+1, "Q\\E", 3) == 0) ptr += 3;
2663 else if (!negate_class && c == '^')
2664 negate_class = TRUE;
2668 /* If a class contains a negative special such as \S, we need to flip the
2669 negation flag at the end, so that support for characters > 255 works
2670 correctly (they are all included in the class). */
2672 should_flip_negation = FALSE;
2674 /* Keep a count of chars with values < 256 so that we can optimize the case
2675 of just a single character (as long as it's < 256). However, For higher
2676 valued UTF-8 characters, we don't yet do any optimization. */
2678 class_charcount = 0;
2679 class_lastchar = -1;
2681 /* Initialize the 32-char bit map to all zeros. We build the map in a
2682 temporary bit of memory, in case the class contains only 1 character (less
2683 than 256), because in that case the compiled code doesn't use the bit map.
2686 memset(classbits, 0, 32 * sizeof(uschar));
2689 class_utf8 = FALSE; /* No chars >= 256 */
2690 class_utf8data = code + LINK_SIZE + 2; /* For UTF-8 items */
2691 class_utf8data_base = class_utf8data; /* For resetting in pass 1 */
2694 /* Process characters until ] is reached. By writing this as a "do" it
2695 means that an initial ] is taken as a data character. At the start of the
2696 loop, c contains the first byte of the character. */
2700 const uschar *oldptr;
2703 if (utf8 && c > 127)
2704 { /* Braces are required because the */
2705 GETCHARLEN(c, ptr, ptr); /* macro generates multiple statements */
2708 /* In the pre-compile phase, accumulate the length of any UTF-8 extra
2709 data and reset the pointer. This is so that very large classes that
2710 contain a zillion UTF-8 characters no longer overwrite the work space
2711 (which is on the stack). */
2713 if (lengthptr != NULL)
2715 *lengthptr += class_utf8data - class_utf8data_base;
2716 class_utf8data = class_utf8data_base;
2721 /* Inside \Q...\E everything is literal except \E */
2725 if (c == '\\' && ptr[1] == 'E') /* If we are at \E */
2727 inescq = FALSE; /* Reset literal state */
2728 ptr++; /* Skip the 'E' */
2729 continue; /* Carry on with next */
2731 goto CHECK_RANGE; /* Could be range if \E follows */
2734 /* Handle POSIX class names. Perl allows a negation extension of the
2735 form [:^name:]. A square bracket that doesn't match the syntax is
2736 treated as a literal. We also recognize the POSIX constructions
2737 [.ch.] and [=ch=] ("collating elements") and fault them, as Perl
2741 (ptr[1] == ':' || ptr[1] == '.' || ptr[1] == '=') &&
2742 check_posix_syntax(ptr, &tempptr))
2744 BOOL local_negate = FALSE;
2745 int posix_class, taboffset, tabopt;
2746 register const uschar *cbits = cd->cbits;
2751 *errorcodeptr = ERR31;
2758 local_negate = TRUE;
2759 should_flip_negation = TRUE; /* Note negative special */
2763 posix_class = check_posix_name(ptr, tempptr - ptr);
2764 if (posix_class < 0)
2766 *errorcodeptr = ERR30;
2770 /* If matching is caseless, upper and lower are converted to
2771 alpha. This relies on the fact that the class table starts with
2772 alpha, lower, upper as the first 3 entries. */
2774 if ((options & PCRE_CASELESS) != 0 && posix_class <= 2)
2777 /* We build the bit map for the POSIX class in a chunk of local store
2778 because we may be adding and subtracting from it, and we don't want to
2779 subtract bits that may be in the main map already. At the end we or the
2780 result into the bit map that is being built. */
2784 /* Copy in the first table (always present) */
2786 memcpy(pbits, cbits + posix_class_maps[posix_class],
2787 32 * sizeof(uschar));
2789 /* If there is a second table, add or remove it as required. */
2791 taboffset = posix_class_maps[posix_class + 1];
2792 tabopt = posix_class_maps[posix_class + 2];
2797 for (c = 0; c < 32; c++) pbits[c] |= cbits[c + taboffset];
2799 for (c = 0; c < 32; c++) pbits[c] &= ~cbits[c + taboffset];
2802 /* Not see if we need to remove any special characters. An option
2803 value of 1 removes vertical space and 2 removes underscore. */
2805 if (tabopt < 0) tabopt = -tabopt;
2806 if (tabopt == 1) pbits[1] &= ~0x3c;
2807 else if (tabopt == 2) pbits[11] &= 0x7f;
2809 /* Add the POSIX table or its complement into the main table that is
2810 being built and we are done. */
2813 for (c = 0; c < 32; c++) classbits[c] |= ~pbits[c];
2815 for (c = 0; c < 32; c++) classbits[c] |= pbits[c];
2818 class_charcount = 10; /* Set > 1; assumes more than 1 per class */
2819 continue; /* End of POSIX syntax handling */
2822 /* Backslash may introduce a single character, or it may introduce one
2823 of the specials, which just set a flag. The sequence \b is a special
2824 case. Inside a class (and only there) it is treated as backspace.
2825 Elsewhere it marks a word boundary. Other escapes have preset maps ready
2826 to 'or' into the one we are building. We assume they have more than one
2827 character in them, so set class_charcount bigger than one. */
2831 c = check_escape(&ptr, errorcodeptr, cd->bracount, options, TRUE);
2832 if (*errorcodeptr != 0) goto FAILED;
2834 if (-c == ESC_b) c = '\b'; /* \b is backspace in a class */
2835 else if (-c == ESC_X) c = 'X'; /* \X is literal X in a class */
2836 else if (-c == ESC_R) c = 'R'; /* \R is literal R in a class */
2837 else if (-c == ESC_Q) /* Handle start of quoted string */
2839 if (ptr[1] == '\\' && ptr[2] == 'E')
2841 ptr += 2; /* avoid empty string */
2846 else if (-c == ESC_E) continue; /* Ignore orphan \E */
2850 register const uschar *cbits = cd->cbits;
2851 class_charcount += 2; /* Greater than 1 is what matters */
2853 /* Save time by not doing this in the pre-compile phase. */
2855 if (lengthptr == NULL) switch (-c)
2858 for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_digit];
2862 should_flip_negation = TRUE;
2863 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_digit];
2867 for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_word];
2871 should_flip_negation = TRUE;
2872 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_word];
2876 for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_space];
2877 classbits[1] &= ~0x08; /* Perl 5.004 onwards omits VT from \s */
2881 should_flip_negation = TRUE;
2882 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_space];
2883 classbits[1] |= 0x08; /* Perl 5.004 onwards omits VT from \s */
2886 default: /* Not recognized; fall through */
2887 break; /* Need "default" setting to stop compiler warning. */
2890 /* In the pre-compile phase, just do the recognition. */
2892 else if (c == -ESC_d || c == -ESC_D || c == -ESC_w ||
2893 c == -ESC_W || c == -ESC_s || c == -ESC_S) continue;
2895 /* We need to deal with \H, \h, \V, and \v in both phases because
2896 they use extra memory. */
2900 SETBIT(classbits, 0x09); /* VT */
2901 SETBIT(classbits, 0x20); /* SPACE */
2902 SETBIT(classbits, 0xa0); /* NSBP */
2907 *class_utf8data++ = XCL_SINGLE;
2908 class_utf8data += _pcre_ord2utf8(0x1680, class_utf8data);
2909 *class_utf8data++ = XCL_SINGLE;
2910 class_utf8data += _pcre_ord2utf8(0x180e, class_utf8data);
2911 *class_utf8data++ = XCL_RANGE;
2912 class_utf8data += _pcre_ord2utf8(0x2000, class_utf8data);
2913 class_utf8data += _pcre_ord2utf8(0x200A, class_utf8data);
2914 *class_utf8data++ = XCL_SINGLE;
2915 class_utf8data += _pcre_ord2utf8(0x202f, class_utf8data);
2916 *class_utf8data++ = XCL_SINGLE;
2917 class_utf8data += _pcre_ord2utf8(0x205f, class_utf8data);
2918 *class_utf8data++ = XCL_SINGLE;
2919 class_utf8data += _pcre_ord2utf8(0x3000, class_utf8data);
2927 for (c = 0; c < 32; c++)
2932 case 0x09/8: x ^= 1 << (0x09%8); break;
2933 case 0x20/8: x ^= 1 << (0x20%8); break;
2934 case 0xa0/8: x ^= 1 << (0xa0%8); break;
2944 *class_utf8data++ = XCL_RANGE;
2945 class_utf8data += _pcre_ord2utf8(0x0100, class_utf8data);
2946 class_utf8data += _pcre_ord2utf8(0x167f, class_utf8data);
2947 *class_utf8data++ = XCL_RANGE;
2948 class_utf8data += _pcre_ord2utf8(0x1681, class_utf8data);
2949 class_utf8data += _pcre_ord2utf8(0x180d, class_utf8data);
2950 *class_utf8data++ = XCL_RANGE;
2951 class_utf8data += _pcre_ord2utf8(0x180f, class_utf8data);
2952 class_utf8data += _pcre_ord2utf8(0x1fff, class_utf8data);
2953 *class_utf8data++ = XCL_RANGE;
2954 class_utf8data += _pcre_ord2utf8(0x200B, class_utf8data);
2955 class_utf8data += _pcre_ord2utf8(0x202e, class_utf8data);
2956 *class_utf8data++ = XCL_RANGE;
2957 class_utf8data += _pcre_ord2utf8(0x2030, class_utf8data);
2958 class_utf8data += _pcre_ord2utf8(0x205e, class_utf8data);
2959 *class_utf8data++ = XCL_RANGE;
2960 class_utf8data += _pcre_ord2utf8(0x2060, class_utf8data);
2961 class_utf8data += _pcre_ord2utf8(0x2fff, class_utf8data);
2962 *class_utf8data++ = XCL_RANGE;
2963 class_utf8data += _pcre_ord2utf8(0x3001, class_utf8data);
2964 class_utf8data += _pcre_ord2utf8(0x7fffffff, class_utf8data);
2972 SETBIT(classbits, 0x0a); /* LF */
2973 SETBIT(classbits, 0x0b); /* VT */
2974 SETBIT(classbits, 0x0c); /* FF */
2975 SETBIT(classbits, 0x0d); /* CR */
2976 SETBIT(classbits, 0x85); /* NEL */
2981 *class_utf8data++ = XCL_RANGE;
2982 class_utf8data += _pcre_ord2utf8(0x2028, class_utf8data);
2983 class_utf8data += _pcre_ord2utf8(0x2029, class_utf8data);
2991 for (c = 0; c < 32; c++)
2996 case 0x0a/8: x ^= 1 << (0x0a%8);
3001 case 0x85/8: x ^= 1 << (0x85%8); break;
3011 *class_utf8data++ = XCL_RANGE;
3012 class_utf8data += _pcre_ord2utf8(0x0100, class_utf8data);
3013 class_utf8data += _pcre_ord2utf8(0x2027, class_utf8data);
3014 *class_utf8data++ = XCL_RANGE;
3015 class_utf8data += _pcre_ord2utf8(0x2029, class_utf8data);
3016 class_utf8data += _pcre_ord2utf8(0x7fffffff, class_utf8data);
3022 /* We need to deal with \P and \p in both phases. */
3025 if (-c == ESC_p || -c == ESC_P)
3029 int ptype = get_ucp(&ptr, &negated, &pdata, errorcodeptr);
3030 if (ptype < 0) goto FAILED;
3032 *class_utf8data++ = ((-c == ESC_p) != negated)?
3033 XCL_PROP : XCL_NOTPROP;
3034 *class_utf8data++ = ptype;
3035 *class_utf8data++ = pdata;
3036 class_charcount -= 2; /* Not a < 256 character */
3040 /* Unrecognized escapes are faulted if PCRE is running in its
3041 strict mode. By default, for compatibility with Perl, they are
3042 treated as literals. */
3044 if ((options & PCRE_EXTRA) != 0)
3046 *errorcodeptr = ERR7;
3050 class_charcount -= 2; /* Undo the default count from above */
3051 c = *ptr; /* Get the final character and fall through */
3054 /* Fall through if we have a single character (c >= 0). This may be
3055 greater than 256 in UTF-8 mode. */
3057 } /* End of backslash handling */
3059 /* A single character may be followed by '-' to form a range. However,
3060 Perl does not permit ']' to be the end of the range. A '-' character
3061 at the end is treated as a literal. Perl ignores orphaned \E sequences
3062 entirely. The code for handling \Q and \E is messy. */
3065 while (ptr[1] == '\\' && ptr[2] == 'E')
3073 /* Remember \r or \n */
3075 if (c == '\r' || c == '\n') cd->external_flags |= PCRE_HASCRORLF;
3077 /* Check for range */
3079 if (!inescq && ptr[1] == '-')
3083 while (*ptr == '\\' && ptr[1] == 'E') ptr += 2;
3085 /* If we hit \Q (not followed by \E) at this point, go into escaped
3088 while (*ptr == '\\' && ptr[1] == 'Q')
3091 if (*ptr == '\\' && ptr[1] == 'E') { ptr += 2; continue; }
3096 if (*ptr == 0 || (!inescq && *ptr == ']'))
3099 goto LONE_SINGLE_CHARACTER;
3104 { /* Braces are required because the */
3105 GETCHARLEN(d, ptr, ptr); /* macro generates multiple statements */
3109 d = *ptr; /* Not UTF-8 mode */
3111 /* The second part of a range can be a single-character escape, but
3112 not any of the other escapes. Perl 5.6 treats a hyphen as a literal
3113 in such circumstances. */
3115 if (!inescq && d == '\\')
3117 d = check_escape(&ptr, errorcodeptr, cd->bracount, options, TRUE);
3118 if (*errorcodeptr != 0) goto FAILED;
3120 /* \b is backspace; \X is literal X; \R is literal R; any other
3121 special means the '-' was literal */
3125 if (d == -ESC_b) d = '\b';
3126 else if (d == -ESC_X) d = 'X';
3127 else if (d == -ESC_R) d = 'R'; else
3130 goto LONE_SINGLE_CHARACTER; /* A few lines below */
3135 /* Check that the two values are in the correct order. Optimize
3136 one-character ranges */
3140 *errorcodeptr = ERR8;
3144 if (d == c) goto LONE_SINGLE_CHARACTER; /* A few lines below */
3146 /* Remember \r or \n */
3148 if (d == '\r' || d == '\n') cd->external_flags |= PCRE_HASCRORLF;
3150 /* In UTF-8 mode, if the upper limit is > 255, or > 127 for caseless
3151 matching, we have to use an XCLASS with extra data items. Caseless
3152 matching for characters > 127 is available only if UCP support is
3156 if (utf8 && (d > 255 || ((options & PCRE_CASELESS) != 0 && d > 127)))
3160 /* With UCP support, we can find the other case equivalents of
3161 the relevant characters. There may be several ranges. Optimize how
3162 they fit with the basic range. */
3165 if ((options & PCRE_CASELESS) != 0)
3167 unsigned int occ, ocd;
3168 unsigned int cc = c;
3169 unsigned int origd = d;
3170 while (get_othercase_range(&cc, origd, &occ, &ocd))
3172 if (occ >= (unsigned int)c &&
3173 ocd <= (unsigned int)d)
3174 continue; /* Skip embedded ranges */
3176 if (occ < (unsigned int)c &&
3177 ocd >= (unsigned int)c - 1) /* Extend the basic range */
3178 { /* if there is overlap, */
3179 c = occ; /* noting that if occ < c */
3180 continue; /* we can't have ocd > d */
3181 } /* because a subrange is */
3182 if (ocd > (unsigned int)d &&
3183 occ <= (unsigned int)d + 1) /* always shorter than */
3184 { /* the basic range. */
3191 *class_utf8data++ = XCL_SINGLE;
3195 *class_utf8data++ = XCL_RANGE;
3196 class_utf8data += _pcre_ord2utf8(occ, class_utf8data);
3198 class_utf8data += _pcre_ord2utf8(ocd, class_utf8data);
3201 #endif /* SUPPORT_UCP */
3203 /* Now record the original range, possibly modified for UCP caseless
3204 overlapping ranges. */
3206 *class_utf8data++ = XCL_RANGE;
3207 class_utf8data += _pcre_ord2utf8(c, class_utf8data);
3208 class_utf8data += _pcre_ord2utf8(d, class_utf8data);
3210 /* With UCP support, we are done. Without UCP support, there is no
3211 caseless matching for UTF-8 characters > 127; we can use the bit map
3212 for the smaller ones. */
3215 continue; /* With next character in the class */
3217 if ((options & PCRE_CASELESS) == 0 || c > 127) continue;
3219 /* Adjust upper limit and fall through to set up the map */
3223 #endif /* SUPPORT_UCP */
3225 #endif /* SUPPORT_UTF8 */
3227 /* We use the bit map for all cases when not in UTF-8 mode; else
3228 ranges that lie entirely within 0-127 when there is UCP support; else
3229 for partial ranges without UCP support. */
3231 class_charcount += d - c + 1;
3234 /* We can save a bit of time by skipping this in the pre-compile. */
3236 if (lengthptr == NULL) for (; c <= d; c++)
3238 classbits[c/8] |= (1 << (c&7));
3239 if ((options & PCRE_CASELESS) != 0)
3241 int uc = cd->fcc[c]; /* flip case */
3242 classbits[uc/8] |= (1 << (uc&7));
3246 continue; /* Go get the next char in the class */
3249 /* Handle a lone single character - we can get here for a normal
3250 non-escape char, or after \ that introduces a single character or for an
3251 apparent range that isn't. */
3253 LONE_SINGLE_CHARACTER:
3255 /* Handle a character that cannot go in the bit map */
3258 if (utf8 && (c > 255 || ((options & PCRE_CASELESS) != 0 && c > 127)))
3261 *class_utf8data++ = XCL_SINGLE;
3262 class_utf8data += _pcre_ord2utf8(c, class_utf8data);
3265 if ((options & PCRE_CASELESS) != 0)
3267 unsigned int othercase;
3268 if ((othercase = _pcre_ucp_othercase(c)) != NOTACHAR)
3270 *class_utf8data++ = XCL_SINGLE;
3271 class_utf8data += _pcre_ord2utf8(othercase, class_utf8data);
3274 #endif /* SUPPORT_UCP */
3278 #endif /* SUPPORT_UTF8 */
3280 /* Handle a single-byte character */
3282 classbits[c/8] |= (1 << (c&7));
3283 if ((options & PCRE_CASELESS) != 0)
3285 c = cd->fcc[c]; /* flip case */
3286 classbits[c/8] |= (1 << (c&7));
3293 /* Loop until ']' reached. This "while" is the end of the "do" above. */
3295 while ((c = *(++ptr)) != 0 && (c != ']' || inescq));
3297 if (c == 0) /* Missing terminating ']' */
3299 *errorcodeptr = ERR6;
3304 /* This code has been disabled because it would mean that \s counts as
3305 an explicit \r or \n reference, and that's not really what is wanted. Now
3306 we set the flag only if there is a literal "\r" or "\n" in the class. */
3309 /* Remember whether \r or \n are in this class */
3313 if ((classbits[1] & 0x24) != 0x24) cd->external_flags |= PCRE_HASCRORLF;
3317 if ((classbits[1] & 0x24) != 0) cd->external_flags |= PCRE_HASCRORLF;
3322 /* If class_charcount is 1, we saw precisely one character whose value is
3323 less than 256. As long as there were no characters >= 128 and there was no
3324 use of \p or \P, in other words, no use of any XCLASS features, we can
3327 In UTF-8 mode, we can optimize the negative case only if there were no
3328 characters >= 128 because OP_NOT and the related opcodes like OP_NOTSTAR
3329 operate on single-bytes only. This is an historical hangover. Maybe one day
3330 we can tidy these opcodes to handle multi-byte characters.
3332 The optimization throws away the bit map. We turn the item into a
3333 1-character OP_CHAR[NC] if it's positive, or OP_NOT if it's negative. Note
3334 that OP_NOT does not support multibyte characters. In the positive case, it
3335 can cause firstbyte to be set. Otherwise, there can be no first char if
3336 this item is first, whatever repeat count may follow. In the case of
3337 reqbyte, save the previous value for reinstating. */
3340 if (class_charcount == 1 && !class_utf8 &&
3341 (!utf8 || !negate_class || class_lastchar < 128))
3343 if (class_charcount == 1)
3346 zeroreqbyte = reqbyte;
3348 /* The OP_NOT opcode works on one-byte characters only. */
3352 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
3353 zerofirstbyte = firstbyte;
3355 *code++ = class_lastchar;
3359 /* For a single, positive character, get the value into mcbuffer, and
3360 then we can handle this with the normal one-character code. */
3363 if (utf8 && class_lastchar > 127)
3364 mclength = _pcre_ord2utf8(class_lastchar, mcbuffer);
3368 mcbuffer[0] = class_lastchar;
3372 } /* End of 1-char optimization */
3374 /* The general case - not the one-char optimization. If this is the first
3375 thing in the branch, there can be no first char setting, whatever the
3376 repeat count. Any reqbyte setting must remain unchanged after any kind of
3379 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
3380 zerofirstbyte = firstbyte;
3381 zeroreqbyte = reqbyte;
3383 /* If there are characters with values > 255, we have to compile an
3384 extended class, with its own opcode, unless there was a negated special
3385 such as \S in the class, because in that case all characters > 255 are in
3386 the class, so any that were explicitly given as well can be ignored. If
3387 (when there are explicit characters > 255 that must be listed) there are no
3388 characters < 256, we can omit the bitmap in the actual compiled code. */
3391 if (class_utf8 && !should_flip_negation)
3393 *class_utf8data++ = XCL_END; /* Marks the end of extra data */
3394 *code++ = OP_XCLASS;
3396 *code = negate_class? XCL_NOT : 0;
3398 /* If the map is required, move up the extra data to make room for it;
3399 otherwise just move the code pointer to the end of the extra data. */
3401 if (class_charcount > 0)
3404 memmove(code + 32, code, class_utf8data - code);
3405 memcpy(code, classbits, 32);
3406 code = class_utf8data + 32;
3408 else code = class_utf8data;
3410 /* Now fill in the complete length of the item */
3412 PUT(previous, 1, code - previous);
3413 break; /* End of class handling */
3417 /* If there are no characters > 255, set the opcode to OP_CLASS or
3418 OP_NCLASS, depending on whether the whole class was negated and whether
3419 there were negative specials such as \S in the class. Then copy the 32-byte
3420 map into the code vector, negating it if necessary. */
3422 *code++ = (negate_class == should_flip_negation) ? OP_CLASS : OP_NCLASS;
3425 if (lengthptr == NULL) /* Save time in the pre-compile phase */
3426 for (c = 0; c < 32; c++) code[c] = ~classbits[c];
3430 memcpy(code, classbits, 32);
3436 /* ===================================================================*/
3437 /* Various kinds of repeat; '{' is not necessarily a quantifier, but this
3438 has been tested above. */
3441 if (!is_quantifier) goto NORMAL_CHAR;
3442 ptr = read_repeat_counts(ptr+1, &repeat_min, &repeat_max, errorcodeptr);
3443 if (*errorcodeptr != 0) goto FAILED;
3461 if (previous == NULL)
3463 *errorcodeptr = ERR9;
3467 if (repeat_min == 0)
3469 firstbyte = zerofirstbyte; /* Adjust for zero repeat */
3470 reqbyte = zeroreqbyte; /* Ditto */
3473 /* Remember whether this is a variable length repeat */
3475 reqvary = (repeat_min == repeat_max)? 0 : REQ_VARY;
3477 op_type = 0; /* Default single-char op codes */
3478 possessive_quantifier = FALSE; /* Default not possessive quantifier */
3480 /* Save start of previous item, in case we have to move it up to make space
3481 for an inserted OP_ONCE for the additional '+' extension. */
3483 tempcode = previous;
3485 /* If the next character is '+', we have a possessive quantifier. This
3486 implies greediness, whatever the setting of the PCRE_UNGREEDY option.
3487 If the next character is '?' this is a minimizing repeat, by default,
3488 but if PCRE_UNGREEDY is set, it works the other way round. We change the
3489 repeat type to the non-default. */
3493 repeat_type = 0; /* Force greedy */
3494 possessive_quantifier = TRUE;
3497 else if (ptr[1] == '?')
3499 repeat_type = greedy_non_default;
3502 else repeat_type = greedy_default;
3504 /* If previous was a character match, abolish the item and generate a
3505 repeat item instead. If a char item has a minumum of more than one, ensure
3506 that it is set in reqbyte - it might not be if a sequence such as x{3} is
3507 the first thing in a branch because the x will have gone into firstbyte
3510 if (*previous == OP_CHAR || *previous == OP_CHARNC)
3512 /* Deal with UTF-8 characters that take up more than one byte. It's
3513 easier to write this out separately than try to macrify it. Use c to
3514 hold the length of the character in bytes, plus 0x80 to flag that it's a
3515 length rather than a small character. */
3518 if (utf8 && (code[-1] & 0x80) != 0)
3520 uschar *lastchar = code - 1;
3521 while((*lastchar & 0xc0) == 0x80) lastchar--;
3522 c = code - lastchar; /* Length of UTF-8 character */
3523 memcpy(utf8_char, lastchar, c); /* Save the char */
3524 c |= 0x80; /* Flag c as a length */
3529 /* Handle the case of a single byte - either with no UTF8 support, or
3530 with UTF-8 disabled, or for a UTF-8 character < 128. */
3534 if (repeat_min > 1) reqbyte = c | req_caseopt | cd->req_varyopt;
3537 /* If the repetition is unlimited, it pays to see if the next thing on
3538 the line is something that cannot possibly match this character. If so,
3539 automatically possessifying this item gains some performance in the case
3540 where the match fails. */
3542 if (!possessive_quantifier &&
3544 check_auto_possessive(*previous, c, utf8, utf8_char, ptr + 1,
3547 repeat_type = 0; /* Force greedy */
3548 possessive_quantifier = TRUE;
3551 goto OUTPUT_SINGLE_REPEAT; /* Code shared with single character types */
3554 /* If previous was a single negated character ([^a] or similar), we use
3555 one of the special opcodes, replacing it. The code is shared with single-
3556 character repeats by setting opt_type to add a suitable offset into
3557 repeat_type. We can also test for auto-possessification. OP_NOT is
3558 currently used only for single-byte chars. */
3560 else if (*previous == OP_NOT)
3562 op_type = OP_NOTSTAR - OP_STAR; /* Use "not" opcodes */
3564 if (!possessive_quantifier &&
3566 check_auto_possessive(OP_NOT, c, utf8, NULL, ptr + 1, options, cd))
3568 repeat_type = 0; /* Force greedy */
3569 possessive_quantifier = TRUE;
3571 goto OUTPUT_SINGLE_REPEAT;
3574 /* If previous was a character type match (\d or similar), abolish it and
3575 create a suitable repeat item. The code is shared with single-character
3576 repeats by setting op_type to add a suitable offset into repeat_type. Note
3577 the the Unicode property types will be present only when SUPPORT_UCP is
3578 defined, but we don't wrap the little bits of code here because it just
3579 makes it horribly messy. */
3581 else if (*previous < OP_EODN)
3584 int prop_type, prop_value;
3585 op_type = OP_TYPESTAR - OP_STAR; /* Use type opcodes */
3588 if (!possessive_quantifier &&
3590 check_auto_possessive(c, 0, utf8, NULL, ptr + 1, options, cd))
3592 repeat_type = 0; /* Force greedy */
3593 possessive_quantifier = TRUE;
3596 OUTPUT_SINGLE_REPEAT:
3597 if (*previous == OP_PROP || *previous == OP_NOTPROP)
3599 prop_type = previous[1];
3600 prop_value = previous[2];
3602 else prop_type = prop_value = -1;
3605 code = previous; /* Usually overwrite previous item */
3607 /* If the maximum is zero then the minimum must also be zero; Perl allows
3608 this case, so we do too - by simply omitting the item altogether. */
3610 if (repeat_max == 0) goto END_REPEAT;
3612 /* All real repeats make it impossible to handle partial matching (maybe
3613 one day we will be able to remove this restriction). */
3615 if (repeat_max != 1) cd->external_flags |= PCRE_NOPARTIAL;
3617 /* Combine the op_type with the repeat_type */
3619 repeat_type += op_type;
3621 /* A minimum of zero is handled either as the special case * or ?, or as
3622 an UPTO, with the maximum given. */
3624 if (repeat_min == 0)
3626 if (repeat_max == -1) *code++ = OP_STAR + repeat_type;
3627 else if (repeat_max == 1) *code++ = OP_QUERY + repeat_type;
3630 *code++ = OP_UPTO + repeat_type;
3631 PUT2INC(code, 0, repeat_max);
3635 /* A repeat minimum of 1 is optimized into some special cases. If the
3636 maximum is unlimited, we use OP_PLUS. Otherwise, the original item is
3637 left in place and, if the maximum is greater than 1, we use OP_UPTO with
3638 one less than the maximum. */
3640 else if (repeat_min == 1)
3642 if (repeat_max == -1)
3643 *code++ = OP_PLUS + repeat_type;
3646 code = oldcode; /* leave previous item in place */
3647 if (repeat_max == 1) goto END_REPEAT;
3648 *code++ = OP_UPTO + repeat_type;
3649 PUT2INC(code, 0, repeat_max - 1);
3653 /* The case {n,n} is just an EXACT, while the general case {n,m} is
3654 handled as an EXACT followed by an UPTO. */
3658 *code++ = OP_EXACT + op_type; /* NB EXACT doesn't have repeat_type */
3659 PUT2INC(code, 0, repeat_min);
3661 /* If the maximum is unlimited, insert an OP_STAR. Before doing so,
3662 we have to insert the character for the previous code. For a repeated
3663 Unicode property match, there are two extra bytes that define the
3664 required property. In UTF-8 mode, long characters have their length in
3665 c, with the 0x80 bit as a flag. */
3670 if (utf8 && c >= 128)
3672 memcpy(code, utf8_char, c & 7);
3681 *code++ = prop_type;
3682 *code++ = prop_value;
3685 *code++ = OP_STAR + repeat_type;
3688 /* Else insert an UPTO if the max is greater than the min, again
3689 preceded by the character, for the previously inserted code. If the
3690 UPTO is just for 1 instance, we can use QUERY instead. */
3692 else if (repeat_max != repeat_min)
3695 if (utf8 && c >= 128)
3697 memcpy(code, utf8_char, c & 7);
3705 *code++ = prop_type;
3706 *code++ = prop_value;
3708 repeat_max -= repeat_min;
3710 if (repeat_max == 1)
3712 *code++ = OP_QUERY + repeat_type;
3716 *code++ = OP_UPTO + repeat_type;
3717 PUT2INC(code, 0, repeat_max);
3722 /* The character or character type itself comes last in all cases. */
3725 if (utf8 && c >= 128)
3727 memcpy(code, utf8_char, c & 7);
3734 /* For a repeated Unicode property match, there are two extra bytes that
3735 define the required property. */
3740 *code++ = prop_type;
3741 *code++ = prop_value;
3746 /* If previous was a character class or a back reference, we put the repeat
3747 stuff after it, but just skip the item if the repeat was {0,0}. */
3749 else if (*previous == OP_CLASS ||
3750 *previous == OP_NCLASS ||
3752 *previous == OP_XCLASS ||
3754 *previous == OP_REF)
3756 if (repeat_max == 0)
3762 /* All real repeats make it impossible to handle partial matching (maybe
3763 one day we will be able to remove this restriction). */
3765 if (repeat_max != 1) cd->external_flags |= PCRE_NOPARTIAL;
3767 if (repeat_min == 0 && repeat_max == -1)
3768 *code++ = OP_CRSTAR + repeat_type;
3769 else if (repeat_min == 1 && repeat_max == -1)
3770 *code++ = OP_CRPLUS + repeat_type;
3771 else if (repeat_min == 0 && repeat_max == 1)
3772 *code++ = OP_CRQUERY + repeat_type;
3775 *code++ = OP_CRRANGE + repeat_type;
3776 PUT2INC(code, 0, repeat_min);
3777 if (repeat_max == -1) repeat_max = 0; /* 2-byte encoding for max */
3778 PUT2INC(code, 0, repeat_max);
3782 /* If previous was a bracket group, we may have to replicate it in certain
3785 else if (*previous == OP_BRA || *previous == OP_CBRA ||
3786 *previous == OP_ONCE || *previous == OP_COND)
3790 int len = code - previous;
3791 uschar *bralink = NULL;
3793 /* Repeating a DEFINE group is pointless */
3795 if (*previous == OP_COND && previous[LINK_SIZE+1] == OP_DEF)
3797 *errorcodeptr = ERR55;
3801 /* If the maximum repeat count is unlimited, find the end of the bracket
3802 by scanning through from the start, and compute the offset back to it
3803 from the current code pointer. There may be an OP_OPT setting following
3804 the final KET, so we can't find the end just by going back from the code
3807 if (repeat_max == -1)
3809 register uschar *ket = previous;
3810 do ket += GET(ket, 1); while (*ket != OP_KET);
3811 ketoffset = code - ket;
3814 /* The case of a zero minimum is special because of the need to stick
3815 OP_BRAZERO in front of it, and because the group appears once in the
3816 data, whereas in other cases it appears the minimum number of times. For
3817 this reason, it is simplest to treat this case separately, as otherwise
3818 the code gets far too messy. There are several special subcases when the
3821 if (repeat_min == 0)
3823 /* If the maximum is also zero, we just omit the group from the output
3826 if (repeat_max == 0)
3832 /* If the maximum is 1 or unlimited, we just have to stick in the
3833 BRAZERO and do no more at this point. However, we do need to adjust
3834 any OP_RECURSE calls inside the group that refer to the group itself or
3835 any internal or forward referenced group, because the offset is from
3836 the start of the whole regex. Temporarily terminate the pattern while
3839 if (repeat_max <= 1)
3842 adjust_recurse(previous, 1, utf8, cd, save_hwm);
3843 memmove(previous+1, previous, len);
3845 *previous++ = OP_BRAZERO + repeat_type;
3848 /* If the maximum is greater than 1 and limited, we have to replicate
3849 in a nested fashion, sticking OP_BRAZERO before each set of brackets.
3850 The first one has to be handled carefully because it's the original
3851 copy, which has to be moved up. The remainder can be handled by code
3852 that is common with the non-zero minimum case below. We have to
3853 adjust the value or repeat_max, since one less copy is required. Once
3854 again, we may have to adjust any OP_RECURSE calls inside the group. */
3860 adjust_recurse(previous, 2 + LINK_SIZE, utf8, cd, save_hwm);
3861 memmove(previous + 2 + LINK_SIZE, previous, len);
3862 code += 2 + LINK_SIZE;
3863 *previous++ = OP_BRAZERO + repeat_type;
3864 *previous++ = OP_BRA;
3866 /* We chain together the bracket offset fields that have to be
3867 filled in later when the ends of the brackets are reached. */
3869 offset = (bralink == NULL)? 0 : previous - bralink;
3871 PUTINC(previous, 0, offset);
3877 /* If the minimum is greater than zero, replicate the group as many
3878 times as necessary, and adjust the maximum to the number of subsequent
3879 copies that we need. If we set a first char from the group, and didn't
3880 set a required char, copy the latter from the former. If there are any
3881 forward reference subroutine calls in the group, there will be entries on
3882 the workspace list; replicate these with an appropriate increment. */
3888 /* In the pre-compile phase, we don't actually do the replication. We
3889 just adjust the length as if we had. Do some paranoid checks for
3890 potential integer overflow. */
3892 if (lengthptr != NULL)
3894 int delta = (repeat_min - 1)*length_prevgroup;
3895 if ((double)(repeat_min - 1)*(double)length_prevgroup >
3897 OFLOW_MAX - *lengthptr < delta)
3899 *errorcodeptr = ERR20;
3902 *lengthptr += delta;
3905 /* This is compiling for real */
3909 if (groupsetfirstbyte && reqbyte < 0) reqbyte = firstbyte;
3910 for (i = 1; i < repeat_min; i++)
3913 uschar *this_hwm = cd->hwm;
3914 memcpy(code, previous, len);
3915 for (hc = save_hwm; hc < this_hwm; hc += LINK_SIZE)
3917 PUT(cd->hwm, 0, GET(hc, 0) + len);
3918 cd->hwm += LINK_SIZE;
3920 save_hwm = this_hwm;
3926 if (repeat_max > 0) repeat_max -= repeat_min;
3929 /* This code is common to both the zero and non-zero minimum cases. If
3930 the maximum is limited, it replicates the group in a nested fashion,
3931 remembering the bracket starts on a stack. In the case of a zero minimum,
3932 the first one was set up above. In all cases the repeat_max now specifies
3933 the number of additional copies needed. Again, we must remember to
3934 replicate entries on the forward reference list. */
3936 if (repeat_max >= 0)
3938 /* In the pre-compile phase, we don't actually do the replication. We
3939 just adjust the length as if we had. For each repetition we must add 1
3940 to the length for BRAZERO and for all but the last repetition we must
3941 add 2 + 2*LINKSIZE to allow for the nesting that occurs. Do some
3942 paranoid checks to avoid integer overflow. */
3944 if (lengthptr != NULL && repeat_max > 0)
3946 int delta = repeat_max * (length_prevgroup + 1 + 2 + 2*LINK_SIZE) -
3947 2 - 2*LINK_SIZE; /* Last one doesn't nest */
3948 if ((double)repeat_max *
3949 (double)(length_prevgroup + 1 + 2 + 2*LINK_SIZE)
3950 > (double)INT_MAX ||
3951 OFLOW_MAX - *lengthptr < delta)
3953 *errorcodeptr = ERR20;
3956 *lengthptr += delta;
3959 /* This is compiling for real */
3961 else for (i = repeat_max - 1; i >= 0; i--)
3964 uschar *this_hwm = cd->hwm;
3966 *code++ = OP_BRAZERO + repeat_type;
3968 /* All but the final copy start a new nesting, maintaining the
3969 chain of brackets outstanding. */
3975 offset = (bralink == NULL)? 0 : code - bralink;
3977 PUTINC(code, 0, offset);
3980 memcpy(code, previous, len);
3981 for (hc = save_hwm; hc < this_hwm; hc += LINK_SIZE)
3983 PUT(cd->hwm, 0, GET(hc, 0) + len + ((i != 0)? 2+LINK_SIZE : 1));
3984 cd->hwm += LINK_SIZE;
3986 save_hwm = this_hwm;
3990 /* Now chain through the pending brackets, and fill in their length
3991 fields (which are holding the chain links pro tem). */
3993 while (bralink != NULL)
3996 int offset = code - bralink + 1;
3997 uschar *bra = code - offset;
3998 oldlinkoffset = GET(bra, 1);
3999 bralink = (oldlinkoffset == 0)? NULL : bralink - oldlinkoffset;
4001 PUTINC(code, 0, offset);
4002 PUT(bra, 1, offset);
4006 /* If the maximum is unlimited, set a repeater in the final copy. We
4007 can't just offset backwards from the current code point, because we
4008 don't know if there's been an options resetting after the ket. The
4009 correct offset was computed above.
4011 Then, when we are doing the actual compile phase, check to see whether
4012 this group is a non-atomic one that could match an empty string. If so,
4013 convert the initial operator to the S form (e.g. OP_BRA -> OP_SBRA) so
4014 that runtime checking can be done. [This check is also applied to
4015 atomic groups at runtime, but in a different way.] */
4019 uschar *ketcode = code - ketoffset;
4020 uschar *bracode = ketcode - GET(ketcode, 1);
4021 *ketcode = OP_KETRMAX + repeat_type;
4022 if (lengthptr == NULL && *bracode != OP_ONCE)
4024 uschar *scode = bracode;
4027 if (could_be_empty_branch(scode, ketcode, utf8))
4029 *bracode += OP_SBRA - OP_BRA;
4032 scode += GET(scode, 1);
4034 while (*scode == OP_ALT);
4039 /* Else there's some kind of shambles */
4043 *errorcodeptr = ERR11;
4047 /* If the character following a repeat is '+', or if certain optimization
4048 tests above succeeded, possessive_quantifier is TRUE. For some of the
4049 simpler opcodes, there is an special alternative opcode for this. For
4050 anything else, we wrap the entire repeated item inside OP_ONCE brackets.
4051 The '+' notation is just syntactic sugar, taken from Sun's Java package,
4052 but the special opcodes can optimize it a bit. The repeated item starts at
4053 tempcode, not at previous, which might be the first part of a string whose
4054 (former) last char we repeated.
4056 Possessifying an 'exact' quantifier has no effect, so we can ignore it. But
4057 an 'upto' may follow. We skip over an 'exact' item, and then test the
4058 length of what remains before proceeding. */
4060 if (possessive_quantifier)
4063 if (*tempcode == OP_EXACT || *tempcode == OP_TYPEEXACT ||
4064 *tempcode == OP_NOTEXACT)
4065 tempcode += _pcre_OP_lengths[*tempcode] +
4066 ((*tempcode == OP_TYPEEXACT &&
4067 (tempcode[3] == OP_PROP || tempcode[3] == OP_NOTPROP))? 2:0);
4068 len = code - tempcode;
4069 if (len > 0) switch (*tempcode)
4071 case OP_STAR: *tempcode = OP_POSSTAR; break;
4072 case OP_PLUS: *tempcode = OP_POSPLUS; break;
4073 case OP_QUERY: *tempcode = OP_POSQUERY; break;
4074 case OP_UPTO: *tempcode = OP_POSUPTO; break;
4076 case OP_TYPESTAR: *tempcode = OP_TYPEPOSSTAR; break;
4077 case OP_TYPEPLUS: *tempcode = OP_TYPEPOSPLUS; break;
4078 case OP_TYPEQUERY: *tempcode = OP_TYPEPOSQUERY; break;
4079 case OP_TYPEUPTO: *tempcode = OP_TYPEPOSUPTO; break;
4081 case OP_NOTSTAR: *tempcode = OP_NOTPOSSTAR; break;
4082 case OP_NOTPLUS: *tempcode = OP_NOTPOSPLUS; break;
4083 case OP_NOTQUERY: *tempcode = OP_NOTPOSQUERY; break;
4084 case OP_NOTUPTO: *tempcode = OP_NOTPOSUPTO; break;
4087 memmove(tempcode + 1+LINK_SIZE, tempcode, len);
4088 code += 1 + LINK_SIZE;
4089 len += 1 + LINK_SIZE;
4090 tempcode[0] = OP_ONCE;
4092 PUTINC(code, 0, len);
4093 PUT(tempcode, 1, len);
4098 /* In all case we no longer have a previous item. We also set the
4099 "follows varying string" flag for subsequently encountered reqbytes if
4100 it isn't already set and we have just passed a varying length item. */
4104 cd->req_varyopt |= reqvary;
4108 /* ===================================================================*/
4109 /* Start of nested parenthesized sub-expression, or comment or lookahead or
4110 lookbehind or option setting or condition or all the other extended
4111 parenthesis forms. */
4114 newoptions = options;
4118 reset_bracount = FALSE;
4120 /* First deal with various "verbs" that can be introduced by '*'. */
4122 if (*(++ptr) == '*' && (cd->ctypes[ptr[1]] & ctype_letter) != 0)
4125 const char *vn = verbnames;
4126 const uschar *name = ++ptr;
4128 while ((cd->ctypes[*++ptr] & ctype_letter) != 0);
4131 *errorcodeptr = ERR59; /* Not supported */
4136 *errorcodeptr = ERR60;
4139 namelen = ptr - name;
4140 for (i = 0; i < verbcount; i++)
4142 if (namelen == verbs[i].len &&
4143 strncmp((char *)name, vn, namelen) == 0)
4145 *code = verbs[i].op;
4146 if (*code++ == OP_ACCEPT) cd->had_accept = TRUE;
4149 vn += verbs[i].len + 1;
4151 if (i < verbcount) continue;
4152 *errorcodeptr = ERR60;
4156 /* Deal with the extended parentheses; all are introduced by '?', and the
4157 appearance of any of them means that this is not a capturing group. */
4159 else if (*ptr == '?')
4161 int i, set, unset, namelen;
4168 case '#': /* Comment; skip to ket */
4170 while (*ptr != 0 && *ptr != ')') ptr++;
4173 *errorcodeptr = ERR18;
4179 /* ------------------------------------------------------------ */
4180 case '|': /* Reset capture count for each branch */
4181 reset_bracount = TRUE;
4184 /* ------------------------------------------------------------ */
4185 case ':': /* Non-capturing bracket */
4191 /* ------------------------------------------------------------ */
4193 bravalue = OP_COND; /* Conditional group */
4195 /* A condition can be an assertion, a number (referring to a numbered
4196 group), a name (referring to a named group), or 'R', referring to
4197 recursion. R<digits> and R&name are also permitted for recursion tests.
4199 There are several syntaxes for testing a named group: (?(name)) is used
4200 by Python; Perl 5.10 onwards uses (?(<name>) or (?('name')).
4202 There are two unfortunate ambiguities, caused by history. (a) 'R' can
4203 be the recursive thing or the name 'R' (and similarly for 'R' followed
4204 by digits), and (b) a number could be a name that consists of digits.
4205 In both cases, we look for a name first; if not found, we try the other
4208 /* For conditions that are assertions, check the syntax, and then exit
4209 the switch. This will take control down to where bracketed groups,
4210 including assertions, are processed. */
4212 if (ptr[1] == '?' && (ptr[2] == '=' || ptr[2] == '!' || ptr[2] == '<'))
4215 /* Most other conditions use OP_CREF (a couple change to OP_RREF
4216 below), and all need to skip 3 bytes at the start of the group. */
4218 code[1+LINK_SIZE] = OP_CREF;
4222 /* Check for a test for recursion in a named group. */
4224 if (ptr[1] == 'R' && ptr[2] == '&')
4228 code[1+LINK_SIZE] = OP_RREF; /* Change the type of test */
4231 /* Check for a test for a named group's having been set, using the Perl
4232 syntax (?(<name>) or (?('name') */
4234 else if (ptr[1] == '<')
4239 else if (ptr[1] == '\'')
4247 if (ptr[1] == '-' || ptr[1] == '+') refsign = *(++ptr);
4250 /* We now expect to read a name; any thing else is an error */
4252 if ((cd->ctypes[ptr[1]] & ctype_word) == 0)
4254 ptr += 1; /* To get the right offset */
4255 *errorcodeptr = ERR28;
4259 /* Read the name, but also get it as a number if it's all digits */
4263 while ((cd->ctypes[*ptr] & ctype_word) != 0)
4266 recno = ((digitab[*ptr] & ctype_digit) != 0)?
4267 recno * 10 + *ptr - '0' : -1;
4270 namelen = ptr - name;
4272 if ((terminator > 0 && *ptr++ != terminator) || *ptr++ != ')')
4274 ptr--; /* Error offset */
4275 *errorcodeptr = ERR26;
4279 /* Do no further checking in the pre-compile phase. */
4281 if (lengthptr != NULL) break;
4283 /* In the real compile we do the work of looking for the actual
4284 reference. If the string started with "+" or "-" we require the rest to
4285 be digits, in which case recno will be set. */
4291 *errorcodeptr = ERR58;
4294 recno = (refsign == '-')?
4295 cd->bracount - recno + 1 : recno +cd->bracount;
4296 if (recno <= 0 || recno > cd->final_bracount)
4298 *errorcodeptr = ERR15;
4301 PUT2(code, 2+LINK_SIZE, recno);
4305 /* Otherwise (did not start with "+" or "-"), start by looking for the
4308 slot = cd->name_table;
4309 for (i = 0; i < cd->names_found; i++)
4311 if (strncmp((char *)name, (char *)slot+2, namelen) == 0) break;
4312 slot += cd->name_entry_size;
4315 /* Found a previous named subpattern */
4317 if (i < cd->names_found)
4319 recno = GET2(slot, 0);
4320 PUT2(code, 2+LINK_SIZE, recno);
4323 /* Search the pattern for a forward reference */
4325 else if ((i = find_parens(ptr, cd->bracount, name, namelen,
4326 (options & PCRE_EXTENDED) != 0)) > 0)
4328 PUT2(code, 2+LINK_SIZE, i);
4331 /* If terminator == 0 it means that the name followed directly after
4332 the opening parenthesis [e.g. (?(abc)...] and in this case there are
4333 some further alternatives to try. For the cases where terminator != 0
4334 [things like (?(<name>... or (?('name')... or (?(R&name)... ] we have
4335 now checked all the possibilities, so give an error. */
4337 else if (terminator != 0)
4339 *errorcodeptr = ERR15;
4343 /* Check for (?(R) for recursion. Allow digits after R to specify a
4344 specific group number. */
4346 else if (*name == 'R')
4349 for (i = 1; i < namelen; i++)
4351 if ((digitab[name[i]] & ctype_digit) == 0)
4353 *errorcodeptr = ERR15;
4356 recno = recno * 10 + name[i] - '0';
4358 if (recno == 0) recno = RREF_ANY;
4359 code[1+LINK_SIZE] = OP_RREF; /* Change test type */
4360 PUT2(code, 2+LINK_SIZE, recno);
4363 /* Similarly, check for the (?(DEFINE) "condition", which is always
4366 else if (namelen == 6 && strncmp((char *)name, "DEFINE", 6) == 0)
4368 code[1+LINK_SIZE] = OP_DEF;
4372 /* Check for the "name" actually being a subpattern number. We are
4373 in the second pass here, so final_bracount is set. */
4375 else if (recno > 0 && recno <= cd->final_bracount)
4377 PUT2(code, 2+LINK_SIZE, recno);
4380 /* Either an unidentified subpattern, or a reference to (?(0) */
4384 *errorcodeptr = (recno == 0)? ERR35: ERR15;
4390 /* ------------------------------------------------------------ */
4391 case '=': /* Positive lookahead */
4392 bravalue = OP_ASSERT;
4397 /* ------------------------------------------------------------ */
4398 case '!': /* Negative lookahead */
4400 if (*ptr == ')') /* Optimize (?!) */
4406 bravalue = OP_ASSERT_NOT;
4410 /* ------------------------------------------------------------ */
4411 case '<': /* Lookbehind or named define */
4414 case '=': /* Positive lookbehind */
4415 bravalue = OP_ASSERTBACK;
4419 case '!': /* Negative lookbehind */
4420 bravalue = OP_ASSERTBACK_NOT;
4424 default: /* Could be name define, else bad */
4425 if ((cd->ctypes[ptr[1]] & ctype_word) != 0) goto DEFINE_NAME;
4426 ptr++; /* Correct offset for error */
4427 *errorcodeptr = ERR24;
4433 /* ------------------------------------------------------------ */
4434 case '>': /* One-time brackets */
4440 /* ------------------------------------------------------------ */
4441 case 'C': /* Callout - may be followed by digits; */
4442 previous_callout = code; /* Save for later completion */
4443 after_manual_callout = 1; /* Skip one item before completing */
4444 *code++ = OP_CALLOUT;
4447 while ((digitab[*(++ptr)] & ctype_digit) != 0)
4448 n = n * 10 + *ptr - '0';
4451 *errorcodeptr = ERR39;
4456 *errorcodeptr = ERR38;
4460 PUT(code, 0, ptr - cd->start_pattern + 1); /* Pattern offset */
4461 PUT(code, LINK_SIZE, 0); /* Default length */
4462 code += 2 * LINK_SIZE;
4468 /* ------------------------------------------------------------ */
4469 case 'P': /* Python-style named subpattern handling */
4470 if (*(++ptr) == '=' || *ptr == '>') /* Reference or recursion */
4472 is_recurse = *ptr == '>';
4474 goto NAMED_REF_OR_RECURSE;
4476 else if (*ptr != '<') /* Test for Python-style definition */
4478 *errorcodeptr = ERR41;
4481 /* Fall through to handle (?P< as (?< is handled */
4484 /* ------------------------------------------------------------ */
4485 DEFINE_NAME: /* Come here from (?< handling */
4488 terminator = (*ptr == '<')? '>' : '\'';
4491 while ((cd->ctypes[*ptr] & ctype_word) != 0) ptr++;
4492 namelen = ptr - name;
4494 /* In the pre-compile phase, just do a syntax check. */
4496 if (lengthptr != NULL)
4498 if (*ptr != terminator)
4500 *errorcodeptr = ERR42;
4503 if (cd->names_found >= MAX_NAME_COUNT)
4505 *errorcodeptr = ERR49;
4508 if (namelen + 3 > cd->name_entry_size)
4510 cd->name_entry_size = namelen + 3;
4511 if (namelen > MAX_NAME_SIZE)
4513 *errorcodeptr = ERR48;
4519 /* In the real compile, create the entry in the table */
4523 slot = cd->name_table;
4524 for (i = 0; i < cd->names_found; i++)
4526 int crc = memcmp(name, slot+2, namelen);
4529 if (slot[2+namelen] == 0)
4531 if ((options & PCRE_DUPNAMES) == 0)
4533 *errorcodeptr = ERR43;
4537 else crc = -1; /* Current name is substring */
4541 memmove(slot + cd->name_entry_size, slot,
4542 (cd->names_found - i) * cd->name_entry_size);
4545 slot += cd->name_entry_size;
4548 PUT2(slot, 0, cd->bracount + 1);
4549 memcpy(slot + 2, name, namelen);
4550 slot[2+namelen] = 0;
4554 /* In both cases, count the number of names we've encountered. */
4556 ptr++; /* Move past > or ' */
4558 goto NUMBERED_GROUP;
4561 /* ------------------------------------------------------------ */
4562 case '&': /* Perl recursion/subroutine syntax */
4567 /* We come here from the Python syntax above that handles both
4568 references (?P=name) and recursion (?P>name), as well as falling
4569 through from the Perl recursion syntax (?&name). We also come here from
4570 the Perl \k<name> or \k'name' back reference syntax and the \k{name}
4573 NAMED_REF_OR_RECURSE:
4575 while ((cd->ctypes[*ptr] & ctype_word) != 0) ptr++;
4576 namelen = ptr - name;
4578 /* In the pre-compile phase, do a syntax check and set a dummy
4579 reference number. */
4581 if (lengthptr != NULL)
4585 *errorcodeptr = ERR62;
4588 if (*ptr != terminator)
4590 *errorcodeptr = ERR42;
4593 if (namelen > MAX_NAME_SIZE)
4595 *errorcodeptr = ERR48;
4601 /* In the real compile, seek the name in the table. We check the name
4602 first, and then check that we have reached the end of the name in the
4603 table. That way, if the name that is longer than any in the table,
4604 the comparison will fail without reading beyond the table entry. */
4608 slot = cd->name_table;
4609 for (i = 0; i < cd->names_found; i++)
4611 if (strncmp((char *)name, (char *)slot+2, namelen) == 0 &&
4612 slot[2+namelen] == 0)
4614 slot += cd->name_entry_size;
4617 if (i < cd->names_found) /* Back reference */
4619 recno = GET2(slot, 0);
4621 else if ((recno = /* Forward back reference */
4622 find_parens(ptr, cd->bracount, name, namelen,
4623 (options & PCRE_EXTENDED) != 0)) <= 0)
4625 *errorcodeptr = ERR15;
4630 /* In both phases, we can now go to the code than handles numerical
4631 recursion or backreferences. */
4633 if (is_recurse) goto HANDLE_RECURSION;
4634 else goto HANDLE_REFERENCE;
4637 /* ------------------------------------------------------------ */
4638 case 'R': /* Recursion */
4639 ptr++; /* Same as (?0) */
4643 /* ------------------------------------------------------------ */
4645 case '0': case '1': case '2': case '3': case '4': /* Recursion or */
4646 case '5': case '6': case '7': case '8': case '9': /* subroutine */
4648 const uschar *called;
4650 if ((refsign = *ptr) == '+')
4653 if ((digitab[*ptr] & ctype_digit) == 0)
4655 *errorcodeptr = ERR63;
4659 else if (refsign == '-')
4661 if ((digitab[ptr[1]] & ctype_digit) == 0)
4662 goto OTHER_CHAR_AFTER_QUERY;
4667 while((digitab[*ptr] & ctype_digit) != 0)
4668 recno = recno * 10 + *ptr++ - '0';
4672 *errorcodeptr = ERR29;
4680 *errorcodeptr = ERR58;
4683 recno = cd->bracount - recno + 1;
4686 *errorcodeptr = ERR15;
4690 else if (refsign == '+')
4694 *errorcodeptr = ERR58;
4697 recno += cd->bracount;
4700 /* Come here from code above that handles a named recursion */
4705 called = cd->start_code;
4707 /* When we are actually compiling, find the bracket that is being
4708 referenced. Temporarily end the regex in case it doesn't exist before
4709 this point. If we end up with a forward reference, first check that
4710 the bracket does occur later so we can give the error (and position)
4711 now. Then remember this forward reference in the workspace so it can
4712 be filled in at the end. */
4714 if (lengthptr == NULL)
4717 if (recno != 0) called = find_bracket(cd->start_code, utf8, recno);
4719 /* Forward reference */
4723 if (find_parens(ptr, cd->bracount, NULL, recno,
4724 (options & PCRE_EXTENDED) != 0) < 0)
4726 *errorcodeptr = ERR15;
4729 called = cd->start_code + recno;
4730 PUTINC(cd->hwm, 0, code + 2 + LINK_SIZE - cd->start_code);
4733 /* If not a forward reference, and the subpattern is still open,
4734 this is a recursive call. We check to see if this is a left
4735 recursion that could loop for ever, and diagnose that case. */
4737 else if (GET(called, 1) == 0 &&
4738 could_be_empty(called, code, bcptr, utf8))
4740 *errorcodeptr = ERR40;
4745 /* Insert the recursion/subroutine item, automatically wrapped inside
4746 "once" brackets. Set up a "previous group" length so that a
4747 subsequent quantifier will work. */
4750 PUT(code, 1, 2 + 2*LINK_SIZE);
4751 code += 1 + LINK_SIZE;
4754 PUT(code, 1, called - cd->start_code);
4755 code += 1 + LINK_SIZE;
4758 PUT(code, 1, 2 + 2*LINK_SIZE);
4759 code += 1 + LINK_SIZE;
4761 length_prevgroup = 3 + 3*LINK_SIZE;
4764 /* Can't determine a first byte now */
4766 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
4770 /* ------------------------------------------------------------ */
4771 default: /* Other characters: check option setting */
4772 OTHER_CHAR_AFTER_QUERY:
4776 while (*ptr != ')' && *ptr != ':')
4780 case '-': optset = &unset; break;
4782 case 'J': /* Record that it changed in the external options */
4783 *optset |= PCRE_DUPNAMES;
4784 cd->external_flags |= PCRE_JCHANGED;
4787 case 'i': *optset |= PCRE_CASELESS; break;
4788 case 'm': *optset |= PCRE_MULTILINE; break;
4789 case 's': *optset |= PCRE_DOTALL; break;
4790 case 'x': *optset |= PCRE_EXTENDED; break;
4791 case 'U': *optset |= PCRE_UNGREEDY; break;
4792 case 'X': *optset |= PCRE_EXTRA; break;
4794 default: *errorcodeptr = ERR12;
4795 ptr--; /* Correct the offset */
4800 /* Set up the changed option bits, but don't change anything yet. */
4802 newoptions = (options | set) & (~unset);
4804 /* If the options ended with ')' this is not the start of a nested
4805 group with option changes, so the options change at this level. If this
4806 item is right at the start of the pattern, the options can be
4807 abstracted and made external in the pre-compile phase, and ignored in
4808 the compile phase. This can be helpful when matching -- for instance in
4809 caseless checking of required bytes.
4811 If the code pointer is not (cd->start_code + 1 + LINK_SIZE), we are
4812 definitely *not* at the start of the pattern because something has been
4813 compiled. In the pre-compile phase, however, the code pointer can have
4814 that value after the start, because it gets reset as code is discarded
4815 during the pre-compile. However, this can happen only at top level - if
4816 we are within parentheses, the starting BRA will still be present. At
4817 any parenthesis level, the length value can be used to test if anything
4818 has been compiled at that level. Thus, a test for both these conditions
4819 is necessary to ensure we correctly detect the start of the pattern in
4822 If we are not at the pattern start, compile code to change the ims
4823 options if this setting actually changes any of them. We also pass the
4824 new setting back so that it can be put at the start of any following
4825 branches, and when this group ends (if we are in a group), a resetting
4826 item can be compiled. */
4830 if (code == cd->start_code + 1 + LINK_SIZE &&
4831 (lengthptr == NULL || *lengthptr == 2 + 2*LINK_SIZE))
4833 cd->external_options = newoptions;
4834 options = newoptions;
4838 if ((options & PCRE_IMS) != (newoptions & PCRE_IMS))
4841 *code++ = newoptions & PCRE_IMS;
4844 /* Change options at this level, and pass them back for use
4845 in subsequent branches. Reset the greedy defaults and the case
4846 value for firstbyte and reqbyte. */
4848 *optionsptr = options = newoptions;
4849 greedy_default = ((newoptions & PCRE_UNGREEDY) != 0);
4850 greedy_non_default = greedy_default ^ 1;
4851 req_caseopt = ((options & PCRE_CASELESS) != 0)? REQ_CASELESS : 0;
4854 previous = NULL; /* This item can't be repeated */
4855 continue; /* It is complete */
4858 /* If the options ended with ':' we are heading into a nested group
4859 with possible change of options. Such groups are non-capturing and are
4860 not assertions of any kind. All we need to do is skip over the ':';
4861 the newoptions value is handled below. */
4865 } /* End of switch for character following (? */
4866 } /* End of (? handling */
4868 /* Opening parenthesis not followed by '?'. If PCRE_NO_AUTO_CAPTURE is set,
4869 all unadorned brackets become non-capturing and behave like (?:...)
4872 else if ((options & PCRE_NO_AUTO_CAPTURE) != 0)
4877 /* Else we have a capturing group. */
4883 PUT2(code, 1+LINK_SIZE, cd->bracount);
4887 /* Process nested bracketed regex. Assertions may not be repeated, but
4888 other kinds can be. All their opcodes are >= OP_ONCE. We copy code into a
4889 non-register variable in order to be able to pass its address because some
4890 compilers complain otherwise. Pass in a new setting for the ims options if
4891 they have changed. */
4893 previous = (bravalue >= OP_ONCE)? code : NULL;
4896 tempreqvary = cd->req_varyopt; /* Save value before bracket */
4897 length_prevgroup = 0; /* Initialize for pre-compile phase */
4900 newoptions, /* The complete new option state */
4901 options & PCRE_IMS, /* The previous ims option state */
4902 &tempcode, /* Where to put code (updated) */
4903 &ptr, /* Input pointer (updated) */
4904 errorcodeptr, /* Where to put an error message */
4905 (bravalue == OP_ASSERTBACK ||
4906 bravalue == OP_ASSERTBACK_NOT), /* TRUE if back assert */
4907 reset_bracount, /* True if (?| group */
4908 skipbytes, /* Skip over bracket number */
4909 &subfirstbyte, /* For possible first char */
4910 &subreqbyte, /* For possible last char */
4911 bcptr, /* Current branch chain */
4912 cd, /* Tables block */
4913 (lengthptr == NULL)? NULL : /* Actual compile phase */
4914 &length_prevgroup /* Pre-compile phase */
4918 /* At the end of compiling, code is still pointing to the start of the
4919 group, while tempcode has been updated to point past the end of the group
4920 and any option resetting that may follow it. The pattern pointer (ptr)
4921 is on the bracket. */
4923 /* If this is a conditional bracket, check that there are no more than
4924 two branches in the group, or just one if it's a DEFINE group. We do this
4925 in the real compile phase, not in the pre-pass, where the whole group may
4926 not be available. */
4928 if (bravalue == OP_COND && lengthptr == NULL)
4937 while (*tc != OP_KET);
4939 /* A DEFINE group is never obeyed inline (the "condition" is always
4940 false). It must have only one branch. */
4942 if (code[LINK_SIZE+1] == OP_DEF)
4946 *errorcodeptr = ERR54;
4949 bravalue = OP_DEF; /* Just a flag to suppress char handling below */
4952 /* A "normal" conditional group. If there is just one branch, we must not
4953 make use of its firstbyte or reqbyte, because this is equivalent to an
4954 empty second branch. */
4960 *errorcodeptr = ERR27;
4963 if (condcount == 1) subfirstbyte = subreqbyte = REQ_NONE;
4967 /* Error if hit end of pattern */
4971 *errorcodeptr = ERR14;
4975 /* In the pre-compile phase, update the length by the length of the group,
4976 less the brackets at either end. Then reduce the compiled code to just a
4977 set of non-capturing brackets so that it doesn't use much memory if it is
4978 duplicated by a quantifier.*/
4980 if (lengthptr != NULL)
4982 if (OFLOW_MAX - *lengthptr < length_prevgroup - 2 - 2*LINK_SIZE)
4984 *errorcodeptr = ERR20;
4987 *lengthptr += length_prevgroup - 2 - 2*LINK_SIZE;
4989 PUTINC(code, 0, 1 + LINK_SIZE);
4991 PUTINC(code, 0, 1 + LINK_SIZE);
4992 break; /* No need to waste time with special character handling */
4995 /* Otherwise update the main code pointer to the end of the group. */
4999 /* For a DEFINE group, required and first character settings are not
5002 if (bravalue == OP_DEF) break;
5004 /* Handle updating of the required and first characters for other types of
5005 group. Update for normal brackets of all kinds, and conditions with two
5006 branches (see code above). If the bracket is followed by a quantifier with
5007 zero repeat, we have to back off. Hence the definition of zeroreqbyte and
5008 zerofirstbyte outside the main loop so that they can be accessed for the
5011 zeroreqbyte = reqbyte;
5012 zerofirstbyte = firstbyte;
5013 groupsetfirstbyte = FALSE;
5015 if (bravalue >= OP_ONCE)
5017 /* If we have not yet set a firstbyte in this branch, take it from the
5018 subpattern, remembering that it was set here so that a repeat of more
5019 than one can replicate it as reqbyte if necessary. If the subpattern has
5020 no firstbyte, set "none" for the whole branch. In both cases, a zero
5021 repeat forces firstbyte to "none". */
5023 if (firstbyte == REQ_UNSET)
5025 if (subfirstbyte >= 0)
5027 firstbyte = subfirstbyte;
5028 groupsetfirstbyte = TRUE;
5030 else firstbyte = REQ_NONE;
5031 zerofirstbyte = REQ_NONE;
5034 /* If firstbyte was previously set, convert the subpattern's firstbyte
5035 into reqbyte if there wasn't one, using the vary flag that was in
5036 existence beforehand. */
5038 else if (subfirstbyte >= 0 && subreqbyte < 0)
5039 subreqbyte = subfirstbyte | tempreqvary;
5041 /* If the subpattern set a required byte (or set a first byte that isn't
5042 really the first byte - see above), set it. */
5044 if (subreqbyte >= 0) reqbyte = subreqbyte;
5047 /* For a forward assertion, we take the reqbyte, if set. This can be
5048 helpful if the pattern that follows the assertion doesn't set a different
5049 char. For example, it's useful for /(?=abcde).+/. We can't set firstbyte
5050 for an assertion, however because it leads to incorrect effect for patterns
5051 such as /(?=a)a.+/ when the "real" "a" would then become a reqbyte instead
5052 of a firstbyte. This is overcome by a scan at the end if there's no
5053 firstbyte, looking for an asserted first char. */
5055 else if (bravalue == OP_ASSERT && subreqbyte >= 0) reqbyte = subreqbyte;
5056 break; /* End of processing '(' */
5059 /* ===================================================================*/
5060 /* Handle metasequences introduced by \. For ones like \d, the ESC_ values
5061 are arranged to be the negation of the corresponding OP_values. For the
5062 back references, the values are ESC_REF plus the reference number. Only
5063 back references and those types that consume a character may be repeated.
5064 We can test for values between ESC_b and ESC_Z for the latter; this may
5065 have to change if any new ones are ever created. */
5069 c = check_escape(&ptr, errorcodeptr, cd->bracount, options, FALSE);
5070 if (*errorcodeptr != 0) goto FAILED;
5074 if (-c == ESC_Q) /* Handle start of quoted string */
5076 if (ptr[1] == '\\' && ptr[2] == 'E') ptr += 2; /* avoid empty string */
5081 if (-c == ESC_E) continue; /* Perl ignores an orphan \E */
5083 /* For metasequences that actually match a character, we disable the
5084 setting of a first character if it hasn't already been set. */
5086 if (firstbyte == REQ_UNSET && -c > ESC_b && -c < ESC_Z)
5087 firstbyte = REQ_NONE;
5089 /* Set values to reset to if this is followed by a zero repeat. */
5091 zerofirstbyte = firstbyte;
5092 zeroreqbyte = reqbyte;
5094 /* \k<name> or \k'name' is a back reference by name (Perl syntax).
5095 We also support \k{name} (.NET syntax) */
5097 if (-c == ESC_k && (ptr[1] == '<' || ptr[1] == '\'' || ptr[1] == '{'))
5100 terminator = (*(++ptr) == '<')? '>' : (*ptr == '\'')? '\'' : '}';
5101 goto NAMED_REF_OR_RECURSE;
5104 /* Back references are handled specially; must disable firstbyte if
5105 not set to cope with cases like (?=(\w+))\1: which would otherwise set
5110 recno = -c - ESC_REF;
5112 HANDLE_REFERENCE: /* Come here from named backref handling */
5113 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
5116 PUT2INC(code, 0, recno);
5117 cd->backref_map |= (recno < 32)? (1 << recno) : 1;
5118 if (recno > cd->top_backref) cd->top_backref = recno;
5121 /* So are Unicode property matches, if supported. */
5124 else if (-c == ESC_P || -c == ESC_p)
5128 int ptype = get_ucp(&ptr, &negated, &pdata, errorcodeptr);
5129 if (ptype < 0) goto FAILED;
5131 *code++ = ((-c == ESC_p) != negated)? OP_PROP : OP_NOTPROP;
5137 /* If Unicode properties are not supported, \X, \P, and \p are not
5140 else if (-c == ESC_X || -c == ESC_P || -c == ESC_p)
5142 *errorcodeptr = ERR45;
5147 /* For the rest (including \X when Unicode properties are supported), we
5148 can obtain the OP value by negating the escape value. */
5152 previous = (-c > ESC_b && -c < ESC_Z)? code : NULL;
5158 /* We have a data character whose value is in c. In UTF-8 mode it may have
5159 a value > 127. We set its representation in the length/buffer, and then
5160 handle it as a data character. */
5163 if (utf8 && c > 127)
5164 mclength = _pcre_ord2utf8(c, mcbuffer);
5175 /* ===================================================================*/
5176 /* Handle a literal character. It is guaranteed not to be whitespace or #
5177 when the extended flag is set. If we are in UTF-8 mode, it may be a
5178 multi-byte literal character. */
5186 if (utf8 && c >= 0xc0)
5188 while ((ptr[1] & 0xc0) == 0x80)
5189 mcbuffer[mclength++] = *(++ptr);
5193 /* At this point we have the character's bytes in mcbuffer, and the length
5194 in mclength. When not in UTF-8 mode, the length is always 1. */
5198 *code++ = ((options & PCRE_CASELESS) != 0)? OP_CHARNC : OP_CHAR;
5199 for (c = 0; c < mclength; c++) *code++ = mcbuffer[c];
5201 /* Remember if \r or \n were seen */
5203 if (mcbuffer[0] == '\r' || mcbuffer[0] == '\n')
5204 cd->external_flags |= PCRE_HASCRORLF;
5206 /* Set the first and required bytes appropriately. If no previous first
5207 byte, set it from this character, but revert to none on a zero repeat.
5208 Otherwise, leave the firstbyte value alone, and don't change it on a zero
5211 if (firstbyte == REQ_UNSET)
5213 zerofirstbyte = REQ_NONE;
5214 zeroreqbyte = reqbyte;
5216 /* If the character is more than one byte long, we can set firstbyte
5217 only if it is not to be matched caselessly. */
5219 if (mclength == 1 || req_caseopt == 0)
5221 firstbyte = mcbuffer[0] | req_caseopt;
5222 if (mclength != 1) reqbyte = code[-1] | cd->req_varyopt;
5224 else firstbyte = reqbyte = REQ_NONE;
5227 /* firstbyte was previously set; we can set reqbyte only the length is
5228 1 or the matching is caseful. */
5232 zerofirstbyte = firstbyte;
5233 zeroreqbyte = reqbyte;
5234 if (mclength == 1 || req_caseopt == 0)
5235 reqbyte = code[-1] | req_caseopt | cd->req_varyopt;
5238 break; /* End of literal character handling */
5240 } /* end of big loop */
5243 /* Control never reaches here by falling through, only by a goto for all the
5244 error states. Pass back the position in the pattern so that it can be displayed
5245 to the user for diagnosing the error. */
5255 /*************************************************
5256 * Compile sequence of alternatives *
5257 *************************************************/
5259 /* On entry, ptr is pointing past the bracket character, but on return it
5260 points to the closing bracket, or vertical bar, or end of string. The code
5261 variable is pointing at the byte into which the BRA operator has been stored.
5262 If the ims options are changed at the start (for a (?ims: group) or during any
5263 branch, we need to insert an OP_OPT item at the start of every following branch
5264 to ensure they get set correctly at run time, and also pass the new options
5265 into every subsequent branch compile.
5267 This function is used during the pre-compile phase when we are trying to find
5268 out the amount of memory needed, as well as during the real compile phase. The
5269 value of lengthptr distinguishes the two phases.
5272 options option bits, including any changes for this subpattern
5273 oldims previous settings of ims option bits
5274 codeptr -> the address of the current code pointer
5275 ptrptr -> the address of the current pattern pointer
5276 errorcodeptr -> pointer to error code variable
5277 lookbehind TRUE if this is a lookbehind assertion
5278 reset_bracount TRUE to reset the count for each branch
5279 skipbytes skip this many bytes at start (for brackets and OP_COND)
5280 firstbyteptr place to put the first required character, or a negative number
5281 reqbyteptr place to put the last required character, or a negative number
5282 bcptr pointer to the chain of currently open branches
5283 cd points to the data block with tables pointers etc.
5284 lengthptr NULL during the real compile phase
5285 points to length accumulator during pre-compile phase
5287 Returns: TRUE on success
5291 compile_regex(int options, int oldims, uschar **codeptr, const uschar **ptrptr,
5292 int *errorcodeptr, BOOL lookbehind, BOOL reset_bracount, int skipbytes,
5293 int *firstbyteptr, int *reqbyteptr, branch_chain *bcptr, compile_data *cd,
5296 const uschar *ptr = *ptrptr;
5297 uschar *code = *codeptr;
5298 uschar *last_branch = code;
5299 uschar *start_bracket = code;
5300 uschar *reverse_count = NULL;
5301 int firstbyte, reqbyte;
5302 int branchfirstbyte, branchreqbyte;
5311 firstbyte = reqbyte = REQ_UNSET;
5313 /* Accumulate the length for use in the pre-compile phase. Start with the
5314 length of the BRA and KET and any extra bytes that are required at the
5315 beginning. We accumulate in a local variable to save frequent testing of
5316 lenthptr for NULL. We cannot do this by looking at the value of code at the
5317 start and end of each alternative, because compiled items are discarded during
5318 the pre-compile phase so that the work space is not exceeded. */
5320 length = 2 + 2*LINK_SIZE + skipbytes;
5322 /* WARNING: If the above line is changed for any reason, you must also change
5323 the code that abstracts option settings at the start of the pattern and makes
5324 them global. It tests the value of length for (2 + 2*LINK_SIZE) in the
5325 pre-compile phase to find out whether anything has yet been compiled or not. */
5327 /* Offset is set zero to mark that this bracket is still open */
5330 code += 1 + LINK_SIZE + skipbytes;
5332 /* Loop for each alternative branch */
5334 orig_bracount = max_bracount = cd->bracount;
5337 /* For a (?| group, reset the capturing bracket count so that each branch
5338 uses the same numbers. */
5340 if (reset_bracount) cd->bracount = orig_bracount;
5342 /* Handle a change of ims options at the start of the branch */
5344 if ((options & PCRE_IMS) != oldims)
5347 *code++ = options & PCRE_IMS;
5351 /* Set up dummy OP_REVERSE if lookbehind assertion */
5355 *code++ = OP_REVERSE;
5356 reverse_count = code;
5358 length += 1 + LINK_SIZE;
5361 /* Now compile the branch; in the pre-compile phase its length gets added
5364 if (!compile_branch(&options, &code, &ptr, errorcodeptr, &branchfirstbyte,
5365 &branchreqbyte, &bc, cd, (lengthptr == NULL)? NULL : &length))
5371 /* Keep the highest bracket count in case (?| was used and some branch
5372 has fewer than the rest. */
5374 if (cd->bracount > max_bracount) max_bracount = cd->bracount;
5376 /* In the real compile phase, there is some post-processing to be done. */
5378 if (lengthptr == NULL)
5380 /* If this is the first branch, the firstbyte and reqbyte values for the
5381 branch become the values for the regex. */
5383 if (*last_branch != OP_ALT)
5385 firstbyte = branchfirstbyte;
5386 reqbyte = branchreqbyte;
5389 /* If this is not the first branch, the first char and reqbyte have to
5390 match the values from all the previous branches, except that if the
5391 previous value for reqbyte didn't have REQ_VARY set, it can still match,
5392 and we set REQ_VARY for the regex. */
5396 /* If we previously had a firstbyte, but it doesn't match the new branch,
5397 we have to abandon the firstbyte for the regex, but if there was
5398 previously no reqbyte, it takes on the value of the old firstbyte. */
5400 if (firstbyte >= 0 && firstbyte != branchfirstbyte)
5402 if (reqbyte < 0) reqbyte = firstbyte;
5403 firstbyte = REQ_NONE;
5406 /* If we (now or from before) have no firstbyte, a firstbyte from the
5407 branch becomes a reqbyte if there isn't a branch reqbyte. */
5409 if (firstbyte < 0 && branchfirstbyte >= 0 && branchreqbyte < 0)
5410 branchreqbyte = branchfirstbyte;
5412 /* Now ensure that the reqbytes match */
5414 if ((reqbyte & ~REQ_VARY) != (branchreqbyte & ~REQ_VARY))
5416 else reqbyte |= branchreqbyte; /* To "or" REQ_VARY */
5419 /* If lookbehind, check that this branch matches a fixed-length string, and
5420 put the length into the OP_REVERSE item. Temporarily mark the end of the
5421 branch with OP_END. */
5427 fixed_length = find_fixedlength(last_branch, options);
5428 DPRINTF(("fixed length = %d\n", fixed_length));
5429 if (fixed_length < 0)
5431 *errorcodeptr = (fixed_length == -2)? ERR36 : ERR25;
5435 PUT(reverse_count, 0, fixed_length);
5439 /* Reached end of expression, either ')' or end of pattern. In the real
5440 compile phase, go back through the alternative branches and reverse the chain
5441 of offsets, with the field in the BRA item now becoming an offset to the
5442 first alternative. If there are no alternatives, it points to the end of the
5443 group. The length in the terminating ket is always the length of the whole
5444 bracketed item. If any of the ims options were changed inside the group,
5445 compile a resetting op-code following, except at the very end of the pattern.
5446 Return leaving the pointer at the terminating char. */
5450 if (lengthptr == NULL)
5452 int branch_length = code - last_branch;
5455 int prev_length = GET(last_branch, 1);
5456 PUT(last_branch, 1, branch_length);
5457 branch_length = prev_length;
5458 last_branch -= branch_length;
5460 while (branch_length > 0);
5463 /* Fill in the ket */
5466 PUT(code, 1, code - start_bracket);
5467 code += 1 + LINK_SIZE;
5469 /* Resetting option if needed */
5471 if ((options & PCRE_IMS) != oldims && *ptr == ')')
5478 /* Retain the highest bracket number, in case resetting was used. */
5480 cd->bracount = max_bracount;
5482 /* Set values to pass back */
5486 *firstbyteptr = firstbyte;
5487 *reqbyteptr = reqbyte;
5488 if (lengthptr != NULL)
5490 if (OFLOW_MAX - *lengthptr < length)
5492 *errorcodeptr = ERR20;
5495 *lengthptr += length;
5500 /* Another branch follows. In the pre-compile phase, we can move the code
5501 pointer back to where it was for the start of the first branch. (That is,
5502 pretend that each branch is the only one.)
5504 In the real compile phase, insert an ALT node. Its length field points back
5505 to the previous branch while the bracket remains open. At the end the chain
5506 is reversed. It's done like this so that the start of the bracket has a
5507 zero offset until it is closed, making it possible to detect recursion. */
5509 if (lengthptr != NULL)
5511 code = *codeptr + 1 + LINK_SIZE + skipbytes;
5512 length += 1 + LINK_SIZE;
5517 PUT(code, 1, code - last_branch);
5518 bc.current = last_branch = code;
5519 code += 1 + LINK_SIZE;
5524 /* Control never reaches here */
5530 /*************************************************
5531 * Check for anchored expression *
5532 *************************************************/
5534 /* Try to find out if this is an anchored regular expression. Consider each
5535 alternative branch. If they all start with OP_SOD or OP_CIRC, or with a bracket
5536 all of whose alternatives start with OP_SOD or OP_CIRC (recurse ad lib), then
5537 it's anchored. However, if this is a multiline pattern, then only OP_SOD
5538 counts, since OP_CIRC can match in the middle.
5540 We can also consider a regex to be anchored if OP_SOM starts all its branches.
5541 This is the code for \G, which means "match at start of match position, taking
5542 into account the match offset".
5544 A branch is also implicitly anchored if it starts with .* and DOTALL is set,
5545 because that will try the rest of the pattern at all possible matching points,
5546 so there is no point trying again.... er ....
5548 .... except when the .* appears inside capturing parentheses, and there is a
5549 subsequent back reference to those parentheses. We haven't enough information
5550 to catch that case precisely.
5552 At first, the best we could do was to detect when .* was in capturing brackets
5553 and the highest back reference was greater than or equal to that level.
5554 However, by keeping a bitmap of the first 31 back references, we can catch some
5555 of the more common cases more precisely.
5558 code points to start of expression (the bracket)
5559 options points to the options setting
5560 bracket_map a bitmap of which brackets we are inside while testing; this
5561 handles up to substring 31; after that we just have to take
5562 the less precise approach
5563 backref_map the back reference bitmap
5565 Returns: TRUE or FALSE
5569 is_anchored(register const uschar *code, int *options, unsigned int bracket_map,
5570 unsigned int backref_map)
5573 const uschar *scode = first_significant_code(code + _pcre_OP_lengths[*code],
5574 options, PCRE_MULTILINE, FALSE);
5575 register int op = *scode;
5577 /* Non-capturing brackets */
5581 if (!is_anchored(scode, options, bracket_map, backref_map)) return FALSE;
5584 /* Capturing brackets */
5586 else if (op == OP_CBRA)
5588 int n = GET2(scode, 1+LINK_SIZE);
5589 int new_map = bracket_map | ((n < 32)? (1 << n) : 1);
5590 if (!is_anchored(scode, options, new_map, backref_map)) return FALSE;
5593 /* Other brackets */
5595 else if (op == OP_ASSERT || op == OP_ONCE || op == OP_COND)
5597 if (!is_anchored(scode, options, bracket_map, backref_map)) return FALSE;
5600 /* .* is not anchored unless DOTALL is set and it isn't in brackets that
5601 are or may be referenced. */
5603 else if ((op == OP_TYPESTAR || op == OP_TYPEMINSTAR ||
5604 op == OP_TYPEPOSSTAR) &&
5605 (*options & PCRE_DOTALL) != 0)
5607 if (scode[1] != OP_ANY || (bracket_map & backref_map) != 0) return FALSE;
5610 /* Check for explicit anchoring */
5612 else if (op != OP_SOD && op != OP_SOM &&
5613 ((*options & PCRE_MULTILINE) != 0 || op != OP_CIRC))
5615 code += GET(code, 1);
5617 while (*code == OP_ALT); /* Loop for each alternative */
5623 /*************************************************
5624 * Check for starting with ^ or .* *
5625 *************************************************/
5627 /* This is called to find out if every branch starts with ^ or .* so that
5628 "first char" processing can be done to speed things up in multiline
5629 matching and for non-DOTALL patterns that start with .* (which must start at
5630 the beginning or after \n). As in the case of is_anchored() (see above), we
5631 have to take account of back references to capturing brackets that contain .*
5632 because in that case we can't make the assumption.
5635 code points to start of expression (the bracket)
5636 bracket_map a bitmap of which brackets we are inside while testing; this
5637 handles up to substring 31; after that we just have to take
5638 the less precise approach
5639 backref_map the back reference bitmap
5641 Returns: TRUE or FALSE
5645 is_startline(const uschar *code, unsigned int bracket_map,
5646 unsigned int backref_map)
5649 const uschar *scode = first_significant_code(code + _pcre_OP_lengths[*code],
5651 register int op = *scode;
5653 /* Non-capturing brackets */
5657 if (!is_startline(scode, bracket_map, backref_map)) return FALSE;
5660 /* Capturing brackets */
5662 else if (op == OP_CBRA)
5664 int n = GET2(scode, 1+LINK_SIZE);
5665 int new_map = bracket_map | ((n < 32)? (1 << n) : 1);
5666 if (!is_startline(scode, new_map, backref_map)) return FALSE;
5669 /* Other brackets */
5671 else if (op == OP_ASSERT || op == OP_ONCE || op == OP_COND)
5672 { if (!is_startline(scode, bracket_map, backref_map)) return FALSE; }
5674 /* .* means "start at start or after \n" if it isn't in brackets that
5675 may be referenced. */
5677 else if (op == OP_TYPESTAR || op == OP_TYPEMINSTAR || op == OP_TYPEPOSSTAR)
5679 if (scode[1] != OP_ANY || (bracket_map & backref_map) != 0) return FALSE;
5682 /* Check for explicit circumflex */
5684 else if (op != OP_CIRC) return FALSE;
5686 /* Move on to the next alternative */
5688 code += GET(code, 1);
5690 while (*code == OP_ALT); /* Loop for each alternative */
5696 /*************************************************
5697 * Check for asserted fixed first char *
5698 *************************************************/
5700 /* During compilation, the "first char" settings from forward assertions are
5701 discarded, because they can cause conflicts with actual literals that follow.
5702 However, if we end up without a first char setting for an unanchored pattern,
5703 it is worth scanning the regex to see if there is an initial asserted first
5704 char. If all branches start with the same asserted char, or with a bracket all
5705 of whose alternatives start with the same asserted char (recurse ad lib), then
5706 we return that char, otherwise -1.
5709 code points to start of expression (the bracket)
5710 options pointer to the options (used to check casing changes)
5711 inassert TRUE if in an assertion
5713 Returns: -1 or the fixed first char
5717 find_firstassertedchar(const uschar *code, int *options, BOOL inassert)
5719 register int c = -1;
5722 const uschar *scode =
5723 first_significant_code(code + 1+LINK_SIZE, options, PCRE_CASELESS, TRUE);
5724 register int op = *scode;
5736 if ((d = find_firstassertedchar(scode, options, op == OP_ASSERT)) < 0)
5738 if (c < 0) c = d; else if (c != d) return -1;
5741 case OP_EXACT: /* Fall through */
5749 if (!inassert) return -1;
5753 if ((*options & PCRE_CASELESS) != 0) c |= REQ_CASELESS;
5755 else if (c != scode[1]) return -1;
5759 code += GET(code, 1);
5761 while (*code == OP_ALT);
5767 /*************************************************
5768 * Compile a Regular Expression *
5769 *************************************************/
5771 /* This function takes a string and returns a pointer to a block of store
5772 holding a compiled version of the expression. The original API for this
5773 function had no error code return variable; it is retained for backwards
5774 compatibility. The new function is given a new name.
5777 pattern the regular expression
5778 options various option bits
5779 errorcodeptr pointer to error code variable (pcre_compile2() only)
5780 can be NULL if you don't want a code value
5781 errorptr pointer to pointer to error text
5782 erroroffset ptr offset in pattern where error was detected
5783 tables pointer to character tables or NULL
5785 Returns: pointer to compiled data block, or NULL on error,
5786 with errorptr and erroroffset set
5789 PCRE_EXP_DEFN pcre *
5790 pcre_compile(const char *pattern, int options, const char **errorptr,
5791 int *erroroffset, const unsigned char *tables)
5793 return pcre_compile2(pattern, options, NULL, errorptr, erroroffset, tables);
5797 PCRE_EXP_DEFN pcre *
5798 pcre_compile2(const char *pattern, int options, int *errorcodeptr,
5799 const char **errorptr, int *erroroffset, const unsigned char *tables)
5802 int length = 1; /* For final END opcode */
5803 int firstbyte, reqbyte, newline;
5805 int skipatstart = 0;
5811 const uschar *codestart;
5813 compile_data compile_block;
5814 compile_data *cd = &compile_block;
5816 /* This space is used for "compiling" into during the first phase, when we are
5817 computing the amount of memory that is needed. Compiled items are thrown away
5818 as soon as possible, so that a fairly large buffer should be sufficient for
5819 this purpose. The same space is used in the second phase for remembering where
5820 to fill in forward references to subpatterns. */
5822 uschar cworkspace[COMPILE_WORK_SIZE];
5824 /* Set this early so that early errors get offset 0. */
5826 ptr = (const uschar *)pattern;
5828 /* We can't pass back an error message if errorptr is NULL; I guess the best we
5829 can do is just return NULL, but we can set a code value if there is a code
5832 if (errorptr == NULL)
5834 if (errorcodeptr != NULL) *errorcodeptr = 99;
5839 if (errorcodeptr != NULL) *errorcodeptr = ERR0;
5841 /* However, we can give a message for this error */
5843 if (erroroffset == NULL)
5846 goto PCRE_EARLY_ERROR_RETURN2;
5851 /* Can't support UTF8 unless PCRE has been compiled to include the code. */
5854 utf8 = (options & PCRE_UTF8) != 0;
5855 if (utf8 && (options & PCRE_NO_UTF8_CHECK) == 0 &&
5856 (*erroroffset = _pcre_valid_utf8((uschar *)pattern, -1)) >= 0)
5859 goto PCRE_EARLY_ERROR_RETURN2;
5862 if ((options & PCRE_UTF8) != 0)
5865 goto PCRE_EARLY_ERROR_RETURN;
5869 if ((options & ~PUBLIC_OPTIONS) != 0)
5872 goto PCRE_EARLY_ERROR_RETURN;
5875 /* Set up pointers to the individual character tables */
5877 if (tables == NULL) tables = _pcre_default_tables;
5878 cd->lcc = tables + lcc_offset;
5879 cd->fcc = tables + fcc_offset;
5880 cd->cbits = tables + cbits_offset;
5881 cd->ctypes = tables + ctypes_offset;
5883 /* Check for global one-time settings at the start of the pattern, and remember
5884 the offset for later. */
5886 while (ptr[skipatstart] == '(' && ptr[skipatstart+1] == '*')
5891 if (strncmp((char *)(ptr+skipatstart+2), "CR)", 3) == 0)
5892 { skipatstart += 5; newnl = PCRE_NEWLINE_CR; }
5893 else if (strncmp((char *)(ptr+skipatstart+2), "LF)", 3) == 0)
5894 { skipatstart += 5; newnl = PCRE_NEWLINE_LF; }
5895 else if (strncmp((char *)(ptr+skipatstart+2), "CRLF)", 5) == 0)
5896 { skipatstart += 7; newnl = PCRE_NEWLINE_CR + PCRE_NEWLINE_LF; }
5897 else if (strncmp((char *)(ptr+skipatstart+2), "ANY)", 4) == 0)
5898 { skipatstart += 6; newnl = PCRE_NEWLINE_ANY; }
5899 else if (strncmp((char *)(ptr+skipatstart+2), "ANYCRLF)", 8) == 0)
5900 { skipatstart += 10; newnl = PCRE_NEWLINE_ANYCRLF; }
5902 else if (strncmp((char *)(ptr+skipatstart+2), "BSR_ANYCRLF)", 12) == 0)
5903 { skipatstart += 14; newbsr = PCRE_BSR_ANYCRLF; }
5904 else if (strncmp((char *)(ptr+skipatstart+2), "BSR_UNICODE)", 12) == 0)
5905 { skipatstart += 14; newbsr = PCRE_BSR_UNICODE; }
5908 options = (options & ~PCRE_NEWLINE_BITS) | newnl;
5909 else if (newbsr != 0)
5910 options = (options & ~(PCRE_BSR_ANYCRLF|PCRE_BSR_UNICODE)) | newbsr;
5914 /* Check validity of \R options. */
5916 switch (options & (PCRE_BSR_ANYCRLF|PCRE_BSR_UNICODE))
5919 case PCRE_BSR_ANYCRLF:
5920 case PCRE_BSR_UNICODE:
5922 default: errorcode = ERR56; goto PCRE_EARLY_ERROR_RETURN;
5925 /* Handle different types of newline. The three bits give seven cases. The
5926 current code allows for fixed one- or two-byte sequences, plus "any" and
5929 switch (options & PCRE_NEWLINE_BITS)
5931 case 0: newline = NEWLINE; break; /* Build-time default */
5932 case PCRE_NEWLINE_CR: newline = '\r'; break;
5933 case PCRE_NEWLINE_LF: newline = '\n'; break;
5934 case PCRE_NEWLINE_CR+
5935 PCRE_NEWLINE_LF: newline = ('\r' << 8) | '\n'; break;
5936 case PCRE_NEWLINE_ANY: newline = -1; break;
5937 case PCRE_NEWLINE_ANYCRLF: newline = -2; break;
5938 default: errorcode = ERR56; goto PCRE_EARLY_ERROR_RETURN;
5943 cd->nltype = NLTYPE_ANYCRLF;
5945 else if (newline < 0)
5947 cd->nltype = NLTYPE_ANY;
5951 cd->nltype = NLTYPE_FIXED;
5955 cd->nl[0] = (newline >> 8) & 255;
5956 cd->nl[1] = newline & 255;
5961 cd->nl[0] = newline;
5965 /* Maximum back reference and backref bitmap. The bitmap records up to 31 back
5966 references to help in deciding whether (.*) can be treated as anchored or not.
5969 cd->top_backref = 0;
5970 cd->backref_map = 0;
5972 /* Reflect pattern for debugging output */
5974 DPRINTF(("------------------------------------------------------------------\n"));
5975 DPRINTF(("%s\n", pattern));
5977 /* Pretend to compile the pattern while actually just accumulating the length
5978 of memory required. This behaviour is triggered by passing a non-NULL final
5979 argument to compile_regex(). We pass a block of workspace (cworkspace) for it
5980 to compile parts of the pattern into; the compiled code is discarded when it is
5981 no longer needed, so hopefully this workspace will never overflow, though there
5982 is a test for its doing so. */
5984 cd->bracount = cd->final_bracount = 0;
5985 cd->names_found = 0;
5986 cd->name_entry_size = 0;
5987 cd->name_table = NULL;
5988 cd->start_workspace = cworkspace;
5989 cd->start_code = cworkspace;
5990 cd->hwm = cworkspace;
5991 cd->start_pattern = (const uschar *)pattern;
5992 cd->end_pattern = (const uschar *)(pattern + strlen(pattern));
5993 cd->req_varyopt = 0;
5994 cd->external_options = options;
5995 cd->external_flags = 0;
5997 /* Now do the pre-compile. On error, errorcode will be set non-zero, so we
5998 don't need to look at the result of the function here. The initial options have
5999 been put into the cd block so that they can be changed if an option setting is
6000 found within the regex right at the beginning. Bringing initial option settings
6001 outside can help speed up starting point checks. */
6006 (void)compile_regex(cd->external_options, cd->external_options & PCRE_IMS,
6007 &code, &ptr, &errorcode, FALSE, FALSE, 0, &firstbyte, &reqbyte, NULL, cd,
6009 if (errorcode != 0) goto PCRE_EARLY_ERROR_RETURN;
6011 DPRINTF(("end pre-compile: length=%d workspace=%d\n", length,
6012 cd->hwm - cworkspace));
6014 if (length > MAX_PATTERN_SIZE)
6017 goto PCRE_EARLY_ERROR_RETURN;
6020 /* Compute the size of data block needed and get it, either from malloc or
6021 externally provided function. Integer overflow should no longer be possible
6022 because nowadays we limit the maximum value of cd->names_found and
6023 cd->name_entry_size. */
6025 size = length + sizeof(real_pcre) + cd->names_found * (cd->name_entry_size + 3);
6026 re = (real_pcre *)(pcre_malloc)(size);
6031 goto PCRE_EARLY_ERROR_RETURN;
6034 /* Put in the magic number, and save the sizes, initial options, internal
6035 flags, and character table pointer. NULL is used for the default character
6036 tables. The nullpad field is at the end; it's there to help in the case when a
6037 regex compiled on a system with 4-byte pointers is run on another with 8-byte
6040 re->magic_number = MAGIC_NUMBER;
6042 re->options = cd->external_options;
6043 re->flags = cd->external_flags;
6047 re->name_table_offset = sizeof(real_pcre);
6048 re->name_entry_size = cd->name_entry_size;
6049 re->name_count = cd->names_found;
6051 re->tables = (tables == _pcre_default_tables)? NULL : tables;
6054 /* The starting points of the name/number translation table and of the code are
6055 passed around in the compile data block. The start/end pattern and initial
6056 options are already set from the pre-compile phase, as is the name_entry_size
6057 field. Reset the bracket count and the names_found field. Also reset the hwm
6058 field; this time it's used for remembering forward references to subpatterns.
6061 cd->final_bracount = cd->bracount; /* Save for checking forward references */
6063 cd->names_found = 0;
6064 cd->name_table = (uschar *)re + re->name_table_offset;
6065 codestart = cd->name_table + re->name_entry_size * re->name_count;
6066 cd->start_code = codestart;
6067 cd->hwm = cworkspace;
6068 cd->req_varyopt = 0;
6069 cd->had_accept = FALSE;
6071 /* Set up a starting, non-extracting bracket, then compile the expression. On
6072 error, errorcode will be set non-zero, so we don't need to look at the result
6073 of the function here. */
6075 ptr = (const uschar *)pattern + skipatstart;
6076 code = (uschar *)codestart;
6078 (void)compile_regex(re->options, re->options & PCRE_IMS, &code, &ptr,
6079 &errorcode, FALSE, FALSE, 0, &firstbyte, &reqbyte, NULL, cd, NULL);
6080 re->top_bracket = cd->bracount;
6081 re->top_backref = cd->top_backref;
6082 re->flags = cd->external_flags;
6084 if (cd->had_accept) reqbyte = -1; /* Must disable after (*ACCEPT) */
6086 /* If not reached end of pattern on success, there's an excess bracket. */
6088 if (errorcode == 0 && *ptr != 0) errorcode = ERR22;
6090 /* Fill in the terminating state and check for disastrous overflow, but
6091 if debugging, leave the test till after things are printed out. */
6096 if (code - codestart > length) errorcode = ERR23;
6099 /* Fill in any forward references that are required. */
6101 while (errorcode == 0 && cd->hwm > cworkspace)
6104 const uschar *groupptr;
6105 cd->hwm -= LINK_SIZE;
6106 offset = GET(cd->hwm, 0);
6107 recno = GET(codestart, offset);
6108 groupptr = find_bracket(codestart, (re->options & PCRE_UTF8) != 0, recno);
6109 if (groupptr == NULL) errorcode = ERR53;
6110 else PUT(((uschar *)codestart), offset, groupptr - codestart);
6113 /* Give an error if there's back reference to a non-existent capturing
6116 if (errorcode == 0 && re->top_backref > re->top_bracket) errorcode = ERR15;
6118 /* Failed to compile, or error while post-processing */
6123 PCRE_EARLY_ERROR_RETURN:
6124 *erroroffset = ptr - (const uschar *)pattern;
6125 PCRE_EARLY_ERROR_RETURN2:
6126 *errorptr = find_error_text(errorcode);
6127 if (errorcodeptr != NULL) *errorcodeptr = errorcode;
6131 /* If the anchored option was not passed, set the flag if we can determine that
6132 the pattern is anchored by virtue of ^ characters or \A or anything else (such
6133 as starting with .* when DOTALL is set).
6135 Otherwise, if we know what the first byte has to be, save it, because that
6136 speeds up unanchored matches no end. If not, see if we can set the
6137 PCRE_STARTLINE flag. This is helpful for multiline matches when all branches
6138 start with ^. and also when all branches start with .* for non-DOTALL matches.
6141 if ((re->options & PCRE_ANCHORED) == 0)
6143 int temp_options = re->options; /* May get changed during these scans */
6144 if (is_anchored(codestart, &temp_options, 0, cd->backref_map))
6145 re->options |= PCRE_ANCHORED;
6149 firstbyte = find_firstassertedchar(codestart, &temp_options, FALSE);
6150 if (firstbyte >= 0) /* Remove caseless flag for non-caseable chars */
6152 int ch = firstbyte & 255;
6153 re->first_byte = ((firstbyte & REQ_CASELESS) != 0 &&
6154 cd->fcc[ch] == ch)? ch : firstbyte;
6155 re->flags |= PCRE_FIRSTSET;
6157 else if (is_startline(codestart, 0, cd->backref_map))
6158 re->flags |= PCRE_STARTLINE;
6162 /* For an anchored pattern, we use the "required byte" only if it follows a
6163 variable length item in the regex. Remove the caseless flag for non-caseable
6167 ((re->options & PCRE_ANCHORED) == 0 || (reqbyte & REQ_VARY) != 0))
6169 int ch = reqbyte & 255;
6170 re->req_byte = ((reqbyte & REQ_CASELESS) != 0 &&
6171 cd->fcc[ch] == ch)? (reqbyte & ~REQ_CASELESS) : reqbyte;
6172 re->flags |= PCRE_REQCHSET;
6175 /* Print out the compiled data if debugging is enabled. This is never the
6176 case when building a production library. */
6180 printf("Length = %d top_bracket = %d top_backref = %d\n",
6181 length, re->top_bracket, re->top_backref);
6183 printf("Options=%08x\n", re->options);
6185 if ((re->flags & PCRE_FIRSTSET) != 0)
6187 int ch = re->first_byte & 255;
6188 const char *caseless = ((re->first_byte & REQ_CASELESS) == 0)?
6190 if (isprint(ch)) printf("First char = %c%s\n", ch, caseless);
6191 else printf("First char = \\x%02x%s\n", ch, caseless);
6194 if ((re->flags & PCRE_REQCHSET) != 0)
6196 int ch = re->req_byte & 255;
6197 const char *caseless = ((re->req_byte & REQ_CASELESS) == 0)?
6199 if (isprint(ch)) printf("Req char = %c%s\n", ch, caseless);
6200 else printf("Req char = \\x%02x%s\n", ch, caseless);
6203 pcre_printint(re, stdout, TRUE);
6205 /* This check is done here in the debugging case so that the code that
6206 was compiled can be seen. */
6208 if (code - codestart > length)
6211 *errorptr = find_error_text(ERR23);
6212 *erroroffset = ptr - (uschar *)pattern;
6213 if (errorcodeptr != NULL) *errorcodeptr = ERR23;
6221 /* End of pcre_compile.c */