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) 1997-2010 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 PCRE_DEBUG is defined, we need the pcre_printint() function, which is
57 also used by pcretest. PCRE_DEBUG is not defined when building a production
61 #include "pcre_printint.src"
65 /* Macro for setting individual bits in class bitmaps. */
67 #define SETBIT(a,b) a[b/8] |= (1 << (b%8))
69 /* Maximum length value to check against when making sure that the integer that
70 holds the compiled pattern length does not overflow. We make it a bit less than
71 INT_MAX to allow for adding in group terminating bytes, so that we don't have
72 to check them every time. */
74 #define OFLOW_MAX (INT_MAX - 20)
77 /*************************************************
78 * Code parameters and static tables *
79 *************************************************/
81 /* This value specifies the size of stack workspace that is used during the
82 first pre-compile phase that determines how much memory is required. The regex
83 is partly compiled into this space, but the compiled parts are discarded as
84 soon as they can be, so that hopefully there will never be an overrun. The code
85 does, however, check for an overrun. The largest amount I've seen used is 218,
86 so this number is very generous.
88 The same workspace is used during the second, actual compile phase for
89 remembering forward references to groups so that they can be filled in at the
90 end. Each entry in this list occupies LINK_SIZE bytes, so even when LINK_SIZE
91 is 4 there is plenty of room. */
93 #define COMPILE_WORK_SIZE (4096)
95 /* The overrun tests check for a slightly smaller size so that they detect the
96 overrun before it actually does run off the end of the data block. */
98 #define WORK_SIZE_CHECK (COMPILE_WORK_SIZE - 100)
101 /* Table for handling escaped characters in the range '0'-'z'. Positive returns
102 are simple data values; negative values are for special things like \d and so
103 on. Zero means further processing is needed (for things like \x), or the escape
108 /* This is the "normal" table for ASCII systems or for EBCDIC systems running
111 static const short int escapes[] = {
117 CHAR_COLON, CHAR_SEMICOLON,
118 CHAR_LESS_THAN_SIGN, CHAR_EQUALS_SIGN,
119 CHAR_GREATER_THAN_SIGN, CHAR_QUESTION_MARK,
120 CHAR_COMMERCIAL_AT, -ESC_A,
133 -ESC_Z, CHAR_LEFT_SQUARE_BRACKET,
134 CHAR_BACKSLASH, CHAR_RIGHT_SQUARE_BRACKET,
135 CHAR_CIRCUMFLEX_ACCENT, CHAR_UNDERSCORE,
136 CHAR_GRAVE_ACCENT, 7,
154 /* This is the "abnormal" table for EBCDIC systems without UTF-8 support. */
156 static const short int escapes[] = {
157 /* 48 */ 0, 0, 0, '.', '<', '(', '+', '|',
158 /* 50 */ '&', 0, 0, 0, 0, 0, 0, 0,
159 /* 58 */ 0, 0, '!', '$', '*', ')', ';', '~',
160 /* 60 */ '-', '/', 0, 0, 0, 0, 0, 0,
161 /* 68 */ 0, 0, '|', ',', '%', '_', '>', '?',
162 /* 70 */ 0, 0, 0, 0, 0, 0, 0, 0,
163 /* 78 */ 0, '`', ':', '#', '@', '\'', '=', '"',
164 /* 80 */ 0, 7, -ESC_b, 0, -ESC_d, ESC_e, ESC_f, 0,
165 /* 88 */-ESC_h, 0, 0, '{', 0, 0, 0, 0,
166 /* 90 */ 0, 0, -ESC_k, 'l', 0, ESC_n, 0, -ESC_p,
167 /* 98 */ 0, ESC_r, 0, '}', 0, 0, 0, 0,
168 /* A0 */ 0, '~', -ESC_s, ESC_tee, 0,-ESC_v, -ESC_w, 0,
169 /* A8 */ 0,-ESC_z, 0, 0, 0, '[', 0, 0,
170 /* B0 */ 0, 0, 0, 0, 0, 0, 0, 0,
171 /* B8 */ 0, 0, 0, 0, 0, ']', '=', '-',
172 /* C0 */ '{',-ESC_A, -ESC_B, -ESC_C, -ESC_D,-ESC_E, 0, -ESC_G,
173 /* C8 */-ESC_H, 0, 0, 0, 0, 0, 0, 0,
174 /* D0 */ '}', 0, -ESC_K, 0, 0,-ESC_N, 0, -ESC_P,
175 /* D8 */-ESC_Q,-ESC_R, 0, 0, 0, 0, 0, 0,
176 /* E0 */ '\\', 0, -ESC_S, 0, 0,-ESC_V, -ESC_W, -ESC_X,
177 /* E8 */ 0,-ESC_Z, 0, 0, 0, 0, 0, 0,
178 /* F0 */ 0, 0, 0, 0, 0, 0, 0, 0,
179 /* F8 */ 0, 0, 0, 0, 0, 0, 0, 0
184 /* Table of special "verbs" like (*PRUNE). This is a short table, so it is
185 searched linearly. Put all the names into a single string, in order to reduce
186 the number of relocations when a shared library is dynamically linked. The
187 string is built from string macros so that it works in UTF-8 mode on EBCDIC
190 typedef struct verbitem {
191 int len; /* Length of verb name */
192 int op; /* Op when no arg, or -1 if arg mandatory */
193 int op_arg; /* Op when arg present, or -1 if not allowed */
196 static const char verbnames[] =
197 "\0" /* Empty name is a shorthand for MARK */
207 static const verbitem verbs[] = {
210 { 6, OP_ACCEPT, -1 },
211 { 6, OP_COMMIT, -1 },
214 { 5, OP_PRUNE, OP_PRUNE_ARG },
215 { 4, OP_SKIP, OP_SKIP_ARG },
216 { 4, OP_THEN, OP_THEN_ARG }
219 static const int verbcount = sizeof(verbs)/sizeof(verbitem);
222 /* Tables of names of POSIX character classes and their lengths. The names are
223 now all in a single string, to reduce the number of relocations when a shared
224 library is dynamically loaded. The list of lengths is terminated by a zero
225 length entry. The first three must be alpha, lower, upper, as this is assumed
226 for handling case independence. */
228 static const char posix_names[] =
229 STRING_alpha0 STRING_lower0 STRING_upper0 STRING_alnum0
230 STRING_ascii0 STRING_blank0 STRING_cntrl0 STRING_digit0
231 STRING_graph0 STRING_print0 STRING_punct0 STRING_space0
232 STRING_word0 STRING_xdigit;
234 static const uschar posix_name_lengths[] = {
235 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 4, 6, 0 };
237 /* Table of class bit maps for each POSIX class. Each class is formed from a
238 base map, with an optional addition or removal of another map. Then, for some
239 classes, there is some additional tweaking: for [:blank:] the vertical space
240 characters are removed, and for [:alpha:] and [:alnum:] the underscore
241 character is removed. The triples in the table consist of the base map offset,
242 second map offset or -1 if no second map, and a non-negative value for map
243 addition or a negative value for map subtraction (if there are two maps). The
244 absolute value of the third field has these meanings: 0 => no tweaking, 1 =>
245 remove vertical space characters, 2 => remove underscore. */
247 static const int posix_class_maps[] = {
248 cbit_word, cbit_digit, -2, /* alpha */
249 cbit_lower, -1, 0, /* lower */
250 cbit_upper, -1, 0, /* upper */
251 cbit_word, -1, 2, /* alnum - word without underscore */
252 cbit_print, cbit_cntrl, 0, /* ascii */
253 cbit_space, -1, 1, /* blank - a GNU extension */
254 cbit_cntrl, -1, 0, /* cntrl */
255 cbit_digit, -1, 0, /* digit */
256 cbit_graph, -1, 0, /* graph */
257 cbit_print, -1, 0, /* print */
258 cbit_punct, -1, 0, /* punct */
259 cbit_space, -1, 0, /* space */
260 cbit_word, -1, 0, /* word - a Perl extension */
261 cbit_xdigit,-1, 0 /* xdigit */
264 /* Table of substitutes for \d etc when PCRE_UCP is set. The POSIX class
265 substitutes must be in the order of the names, defined above, and there are
266 both positive and negative cases. NULL means no substitute. */
269 static const uschar *substitutes[] = {
270 (uschar *)"\\P{Nd}", /* \D */
271 (uschar *)"\\p{Nd}", /* \d */
272 (uschar *)"\\P{Xsp}", /* \S */ /* NOTE: Xsp is Perl space */
273 (uschar *)"\\p{Xsp}", /* \s */
274 (uschar *)"\\P{Xwd}", /* \W */
275 (uschar *)"\\p{Xwd}" /* \w */
278 static const uschar *posix_substitutes[] = {
279 (uschar *)"\\p{L}", /* alpha */
280 (uschar *)"\\p{Ll}", /* lower */
281 (uschar *)"\\p{Lu}", /* upper */
282 (uschar *)"\\p{Xan}", /* alnum */
284 (uschar *)"\\h", /* blank */
286 (uschar *)"\\p{Nd}", /* digit */
290 (uschar *)"\\p{Xps}", /* space */ /* NOTE: Xps is POSIX space */
291 (uschar *)"\\p{Xwd}", /* word */
294 (uschar *)"\\P{L}", /* ^alpha */
295 (uschar *)"\\P{Ll}", /* ^lower */
296 (uschar *)"\\P{Lu}", /* ^upper */
297 (uschar *)"\\P{Xan}", /* ^alnum */
299 (uschar *)"\\H", /* ^blank */
301 (uschar *)"\\P{Nd}", /* ^digit */
305 (uschar *)"\\P{Xps}", /* ^space */ /* NOTE: Xps is POSIX space */
306 (uschar *)"\\P{Xwd}", /* ^word */
309 #define POSIX_SUBSIZE (sizeof(posix_substitutes)/sizeof(uschar *))
312 #define STRING(a) # a
313 #define XSTRING(s) STRING(s)
315 /* The texts of compile-time error messages. These are "char *" because they
316 are passed to the outside world. Do not ever re-use any error number, because
317 they are documented. Always add a new error instead. Messages marked DEAD below
318 are no longer used. This used to be a table of strings, but in order to reduce
319 the number of relocations needed when a shared library is loaded dynamically,
320 it is now one long string. We cannot use a table of offsets, because the
321 lengths of inserts such as XSTRING(MAX_NAME_SIZE) are not known. Instead, we
322 simply count through to the one we want - this isn't a performance issue
323 because these strings are used only when there is a compilation error.
325 Each substring ends with \0 to insert a null character. This includes the final
326 substring, so that the whole string ends with \0\0, which can be detected when
329 static const char error_texts[] =
331 "\\ at end of pattern\0"
332 "\\c at end of pattern\0"
333 "unrecognized character follows \\\0"
334 "numbers out of order in {} quantifier\0"
336 "number too big in {} quantifier\0"
337 "missing terminating ] for character class\0"
338 "invalid escape sequence in character class\0"
339 "range out of order in character class\0"
340 "nothing to repeat\0"
342 "operand of unlimited repeat could match the empty string\0" /** DEAD **/
343 "internal error: unexpected repeat\0"
344 "unrecognized character after (? or (?-\0"
345 "POSIX named classes are supported only within a class\0"
348 "reference to non-existent subpattern\0"
349 "erroffset passed as NULL\0"
350 "unknown option bit(s) set\0"
351 "missing ) after comment\0"
352 "parentheses nested too deeply\0" /** DEAD **/
354 "regular expression is too large\0"
355 "failed to get memory\0"
356 "unmatched parentheses\0"
357 "internal error: code overflow\0"
358 "unrecognized character after (?<\0"
360 "lookbehind assertion is not fixed length\0"
361 "malformed number or name after (?(\0"
362 "conditional group contains more than two branches\0"
363 "assertion expected after (?(\0"
364 "(?R or (?[+-]digits must be followed by )\0"
366 "unknown POSIX class name\0"
367 "POSIX collating elements are not supported\0"
368 "this version of PCRE is not compiled with PCRE_UTF8 support\0"
369 "spare error\0" /** DEAD **/
370 "character value in \\x{...} sequence is too large\0"
372 "invalid condition (?(0)\0"
373 "\\C not allowed in lookbehind assertion\0"
374 "PCRE does not support \\L, \\l, \\N{name}, \\U, or \\u\0"
375 "number after (?C is > 255\0"
376 "closing ) for (?C expected\0"
378 "recursive call could loop indefinitely\0"
379 "unrecognized character after (?P\0"
380 "syntax error in subpattern name (missing terminator)\0"
381 "two named subpatterns have the same name\0"
382 "invalid UTF-8 string\0"
384 "support for \\P, \\p, and \\X has not been compiled\0"
385 "malformed \\P or \\p sequence\0"
386 "unknown property name after \\P or \\p\0"
387 "subpattern name is too long (maximum " XSTRING(MAX_NAME_SIZE) " characters)\0"
388 "too many named subpatterns (maximum " XSTRING(MAX_NAME_COUNT) ")\0"
390 "repeated subpattern is too long\0" /** DEAD **/
391 "octal value is greater than \\377 (not in UTF-8 mode)\0"
392 "internal error: overran compiling workspace\0"
393 "internal error: previously-checked referenced subpattern not found\0"
394 "DEFINE group contains more than one branch\0"
396 "repeating a DEFINE group is not allowed\0"
397 "inconsistent NEWLINE options\0"
398 "\\g is not followed by a braced, angle-bracketed, or quoted name/number or by a plain number\0"
399 "a numbered reference must not be zero\0"
400 "an argument is not allowed for (*ACCEPT), (*FAIL), or (*COMMIT)\0"
402 "(*VERB) not recognized\0"
403 "number is too big\0"
404 "subpattern name expected\0"
405 "digit expected after (?+\0"
406 "] is an invalid data character in JavaScript compatibility mode\0"
408 "different names for subpatterns of the same number are not allowed\0"
409 "(*MARK) must have an argument\0"
410 "this version of PCRE is not compiled with PCRE_UCP support\0"
411 "\\c must be followed by an ASCII character\0"
414 /* Table to identify digits and hex digits. This is used when compiling
415 patterns. Note that the tables in chartables are dependent on the locale, and
416 may mark arbitrary characters as digits - but the PCRE compiling code expects
417 to handle only 0-9, a-z, and A-Z as digits when compiling. That is why we have
418 a private table here. It costs 256 bytes, but it is a lot faster than doing
419 character value tests (at least in some simple cases I timed), and in some
420 applications one wants PCRE to compile efficiently as well as match
423 For convenience, we use the same bit definitions as in chartables:
426 0x08 hexadecimal digit
428 Then we can use ctype_digit and ctype_xdigit in the code. */
432 /* This is the "normal" case, for ASCII systems, and EBCDIC systems running in
435 static const unsigned char digitab[] =
437 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 0- 7 */
438 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 8- 15 */
439 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 16- 23 */
440 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 24- 31 */
441 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - ' */
442 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* ( - / */
443 0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c, /* 0 - 7 */
444 0x0c,0x0c,0x00,0x00,0x00,0x00,0x00,0x00, /* 8 - ? */
445 0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* @ - G */
446 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* H - O */
447 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* P - W */
448 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* X - _ */
449 0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* ` - g */
450 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* h - o */
451 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* p - w */
452 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* x -127 */
453 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 128-135 */
454 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 136-143 */
455 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 144-151 */
456 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 152-159 */
457 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 160-167 */
458 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 168-175 */
459 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 176-183 */
460 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 184-191 */
461 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 192-199 */
462 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 200-207 */
463 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 208-215 */
464 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 216-223 */
465 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 224-231 */
466 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 232-239 */
467 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 240-247 */
468 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00};/* 248-255 */
472 /* This is the "abnormal" case, for EBCDIC systems not running in UTF-8 mode. */
474 static const unsigned char digitab[] =
476 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 0- 7 0 */
477 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 8- 15 */
478 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 16- 23 10 */
479 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 24- 31 */
480 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 32- 39 20 */
481 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 40- 47 */
482 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 48- 55 30 */
483 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 56- 63 */
484 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - 71 40 */
485 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 72- | */
486 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* & - 87 50 */
487 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 88- 95 */
488 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - -103 60 */
489 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 104- ? */
490 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 112-119 70 */
491 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 120- " */
492 0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* 128- g 80 */
493 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* h -143 */
494 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 144- p 90 */
495 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* q -159 */
496 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 160- x A0 */
497 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* y -175 */
498 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* ^ -183 B0 */
499 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 184-191 */
500 0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* { - G C0 */
501 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* H -207 */
502 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* } - P D0 */
503 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* Q -223 */
504 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* \ - X E0 */
505 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* Y -239 */
506 0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c, /* 0 - 7 F0 */
507 0x0c,0x0c,0x00,0x00,0x00,0x00,0x00,0x00};/* 8 -255 */
509 static const unsigned char ebcdic_chartab[] = { /* chartable partial dup */
510 0x80,0x00,0x00,0x00,0x00,0x01,0x00,0x00, /* 0- 7 */
511 0x00,0x00,0x00,0x00,0x01,0x01,0x00,0x00, /* 8- 15 */
512 0x00,0x00,0x00,0x00,0x00,0x01,0x00,0x00, /* 16- 23 */
513 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 24- 31 */
514 0x00,0x00,0x00,0x00,0x00,0x01,0x00,0x00, /* 32- 39 */
515 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 40- 47 */
516 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 48- 55 */
517 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 56- 63 */
518 0x01,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - 71 */
519 0x00,0x00,0x00,0x80,0x00,0x80,0x80,0x80, /* 72- | */
520 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* & - 87 */
521 0x00,0x00,0x00,0x80,0x80,0x80,0x00,0x00, /* 88- 95 */
522 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - -103 */
523 0x00,0x00,0x00,0x00,0x00,0x10,0x00,0x80, /* 104- ? */
524 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 112-119 */
525 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 120- " */
526 0x00,0x1a,0x1a,0x1a,0x1a,0x1a,0x1a,0x12, /* 128- g */
527 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* h -143 */
528 0x00,0x12,0x12,0x12,0x12,0x12,0x12,0x12, /* 144- p */
529 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* q -159 */
530 0x00,0x00,0x12,0x12,0x12,0x12,0x12,0x12, /* 160- x */
531 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* y -175 */
532 0x80,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* ^ -183 */
533 0x00,0x00,0x80,0x00,0x00,0x00,0x00,0x00, /* 184-191 */
534 0x80,0x1a,0x1a,0x1a,0x1a,0x1a,0x1a,0x12, /* { - G */
535 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* H -207 */
536 0x00,0x12,0x12,0x12,0x12,0x12,0x12,0x12, /* } - P */
537 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* Q -223 */
538 0x00,0x00,0x12,0x12,0x12,0x12,0x12,0x12, /* \ - X */
539 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* Y -239 */
540 0x1c,0x1c,0x1c,0x1c,0x1c,0x1c,0x1c,0x1c, /* 0 - 7 */
541 0x1c,0x1c,0x00,0x00,0x00,0x00,0x00,0x00};/* 8 -255 */
545 /* Definition to allow mutual recursion */
548 compile_regex(int, int, uschar **, const uschar **, int *, BOOL, BOOL, int,
549 int *, int *, branch_chain *, compile_data *, int *);
553 /*************************************************
554 * Find an error text *
555 *************************************************/
557 /* The error texts are now all in one long string, to save on relocations. As
558 some of the text is of unknown length, we can't use a table of offsets.
559 Instead, just count through the strings. This is not a performance issue
560 because it happens only when there has been a compilation error.
562 Argument: the error number
563 Returns: pointer to the error string
567 find_error_text(int n)
569 const char *s = error_texts;
572 while (*s++ != 0) {};
573 if (*s == 0) return "Error text not found (please report)";
579 /*************************************************
581 *************************************************/
583 /* This function is called when a \ has been encountered. It either returns a
584 positive value for a simple escape such as \n, or a negative value which
585 encodes one of the more complicated things such as \d. A backreference to group
586 n is returned as -(ESC_REF + n); ESC_REF is the highest ESC_xxx macro. When
587 UTF-8 is enabled, a positive value greater than 255 may be returned. On entry,
588 ptr is pointing at the \. On exit, it is on the final character of the escape
592 ptrptr points to the pattern position pointer
593 errorcodeptr points to the errorcode variable
594 bracount number of previous extracting brackets
595 options the options bits
596 isclass TRUE if inside a character class
598 Returns: zero or positive => a data character
599 negative => a special escape sequence
600 on error, errorcodeptr is set
604 check_escape(const uschar **ptrptr, int *errorcodeptr, int bracount,
605 int options, BOOL isclass)
607 BOOL utf8 = (options & PCRE_UTF8) != 0;
608 const uschar *ptr = *ptrptr + 1;
611 GETCHARINCTEST(c, ptr); /* Get character value, increment pointer */
612 ptr--; /* Set pointer back to the last byte */
614 /* If backslash is at the end of the pattern, it's an error. */
616 if (c == 0) *errorcodeptr = ERR1;
618 /* Non-alphanumerics are literals. For digits or letters, do an initial lookup
619 in a table. A non-zero result is something that can be returned immediately.
620 Otherwise further processing may be required. */
622 #ifndef EBCDIC /* ASCII/UTF-8 coding */
623 else if (c < CHAR_0 || c > CHAR_z) {} /* Not alphanumeric */
624 else if ((i = escapes[c - CHAR_0]) != 0) c = i;
626 #else /* EBCDIC coding */
627 else if (c < 'a' || (ebcdic_chartab[c] & 0x0E) == 0) {} /* Not alphanumeric */
628 else if ((i = escapes[c - 0x48]) != 0) c = i;
631 /* Escapes that need further processing, or are illegal. */
635 const uschar *oldptr;
636 BOOL braced, negated;
640 /* A number of Perl escapes are not handled by PCRE. We give an explicit
647 *errorcodeptr = ERR37;
650 /* \g must be followed by one of a number of specific things:
652 (1) A number, either plain or braced. If positive, it is an absolute
653 backreference. If negative, it is a relative backreference. This is a Perl
656 (2) Perl 5.10 also supports \g{name} as a reference to a named group. This
657 is part of Perl's movement towards a unified syntax for back references. As
658 this is synonymous with \k{name}, we fudge it up by pretending it really
661 (3) For Oniguruma compatibility we also support \g followed by a name or a
662 number either in angle brackets or in single quotes. However, these are
663 (possibly recursive) subroutine calls, _not_ backreferences. Just return
664 the -ESC_g code (cf \k). */
667 if (ptr[1] == CHAR_LESS_THAN_SIGN || ptr[1] == CHAR_APOSTROPHE)
673 /* Handle the Perl-compatible cases */
675 if (ptr[1] == CHAR_LEFT_CURLY_BRACKET)
678 for (p = ptr+2; *p != 0 && *p != CHAR_RIGHT_CURLY_BRACKET; p++)
679 if (*p != CHAR_MINUS && (digitab[*p] & ctype_digit) == 0) break;
680 if (*p != 0 && *p != CHAR_RIGHT_CURLY_BRACKET)
690 if (ptr[1] == CHAR_MINUS)
695 else negated = FALSE;
698 while ((digitab[ptr[1]] & ctype_digit) != 0)
699 c = c * 10 + *(++ptr) - CHAR_0;
701 if (c < 0) /* Integer overflow */
703 *errorcodeptr = ERR61;
707 if (braced && *(++ptr) != CHAR_RIGHT_CURLY_BRACKET)
709 *errorcodeptr = ERR57;
715 *errorcodeptr = ERR58;
723 *errorcodeptr = ERR15;
726 c = bracount - (c - 1);
732 /* The handling of escape sequences consisting of a string of digits
733 starting with one that is not zero is not straightforward. By experiment,
734 the way Perl works seems to be as follows:
736 Outside a character class, the digits are read as a decimal number. If the
737 number is less than 10, or if there are that many previous extracting
738 left brackets, then it is a back reference. Otherwise, up to three octal
739 digits are read to form an escaped byte. Thus \123 is likely to be octal
740 123 (cf \0123, which is octal 012 followed by the literal 3). If the octal
741 value is greater than 377, the least significant 8 bits are taken. Inside a
742 character class, \ followed by a digit is always an octal number. */
744 case CHAR_1: case CHAR_2: case CHAR_3: case CHAR_4: case CHAR_5:
745 case CHAR_6: case CHAR_7: case CHAR_8: case CHAR_9:
751 while ((digitab[ptr[1]] & ctype_digit) != 0)
752 c = c * 10 + *(++ptr) - CHAR_0;
753 if (c < 0) /* Integer overflow */
755 *errorcodeptr = ERR61;
758 if (c < 10 || c <= bracount)
763 ptr = oldptr; /* Put the pointer back and fall through */
766 /* Handle an octal number following \. If the first digit is 8 or 9, Perl
767 generates a binary zero byte and treats the digit as a following literal.
768 Thus we have to pull back the pointer by one. */
770 if ((c = *ptr) >= CHAR_8)
777 /* \0 always starts an octal number, but we may drop through to here with a
778 larger first octal digit. The original code used just to take the least
779 significant 8 bits of octal numbers (I think this is what early Perls used
780 to do). Nowadays we allow for larger numbers in UTF-8 mode, but no more
781 than 3 octal digits. */
785 while(i++ < 2 && ptr[1] >= CHAR_0 && ptr[1] <= CHAR_7)
786 c = c * 8 + *(++ptr) - CHAR_0;
787 if (!utf8 && c > 255) *errorcodeptr = ERR51;
790 /* \x is complicated. \x{ddd} is a character number which can be greater
791 than 0xff in utf8 mode, but only if the ddd are hex digits. If not, { is
792 treated as a data character. */
795 if (ptr[1] == CHAR_LEFT_CURLY_BRACKET)
797 const uschar *pt = ptr + 2;
801 while ((digitab[*pt] & ctype_xdigit) != 0)
803 register int cc = *pt++;
804 if (c == 0 && cc == CHAR_0) continue; /* Leading zeroes */
807 #ifndef EBCDIC /* ASCII/UTF-8 coding */
808 if (cc >= CHAR_a) cc -= 32; /* Convert to upper case */
809 c = (c << 4) + cc - ((cc < CHAR_A)? CHAR_0 : (CHAR_A - 10));
810 #else /* EBCDIC coding */
811 if (cc >= CHAR_a && cc <= CHAR_z) cc += 64; /* Convert to upper case */
812 c = (c << 4) + cc - ((cc >= CHAR_0)? CHAR_0 : (CHAR_A - 10));
816 if (*pt == CHAR_RIGHT_CURLY_BRACKET)
818 if (c < 0 || count > (utf8? 8 : 2)) *errorcodeptr = ERR34;
823 /* If the sequence of hex digits does not end with '}', then we don't
824 recognize this construct; fall through to the normal \x handling. */
827 /* Read just a single-byte hex-defined char */
830 while (i++ < 2 && (digitab[ptr[1]] & ctype_xdigit) != 0)
832 int cc; /* Some compilers don't like */
833 cc = *(++ptr); /* ++ in initializers */
834 #ifndef EBCDIC /* ASCII/UTF-8 coding */
835 if (cc >= CHAR_a) cc -= 32; /* Convert to upper case */
836 c = c * 16 + cc - ((cc < CHAR_A)? CHAR_0 : (CHAR_A - 10));
837 #else /* EBCDIC coding */
838 if (cc <= CHAR_z) cc += 64; /* Convert to upper case */
839 c = c * 16 + cc - ((cc >= CHAR_0)? CHAR_0 : (CHAR_A - 10));
844 /* For \c, a following letter is upper-cased; then the 0x40 bit is flipped.
845 An error is given if the byte following \c is not an ASCII character. This
846 coding is ASCII-specific, but then the whole concept of \cx is
847 ASCII-specific. (However, an EBCDIC equivalent has now been added.) */
853 *errorcodeptr = ERR2;
856 #ifndef EBCDIC /* ASCII/UTF-8 coding */
857 if (c > 127) /* Excludes all non-ASCII in either mode */
859 *errorcodeptr = ERR68;
862 if (c >= CHAR_a && c <= CHAR_z) c -= 32;
864 #else /* EBCDIC coding */
865 if (c >= CHAR_a && c <= CHAR_z) c += 64;
870 /* PCRE_EXTRA enables extensions to Perl in the matter of escapes. Any
871 other alphanumeric following \ is an error if PCRE_EXTRA was set;
872 otherwise, for Perl compatibility, it is a literal. This code looks a bit
873 odd, but there used to be some cases other than the default, and there may
874 be again in future, so I haven't "optimized" it. */
877 if ((options & PCRE_EXTRA) != 0) switch(c)
880 *errorcodeptr = ERR3;
887 /* Perl supports \N{name} for character names, as well as plain \N for "not
888 newline". PCRE does not support \N{name}. */
890 if (c == -ESC_N && ptr[1] == CHAR_LEFT_CURLY_BRACKET)
891 *errorcodeptr = ERR37;
893 /* If PCRE_UCP is set, we change the values for \d etc. */
895 if ((options & PCRE_UCP) != 0 && c <= -ESC_D && c >= -ESC_w)
896 c -= (ESC_DU - ESC_D);
898 /* Set the pointer to the final character before returning. */
907 /*************************************************
909 *************************************************/
911 /* This function is called after \P or \p has been encountered, provided that
912 PCRE is compiled with support for Unicode properties. On entry, ptrptr is
913 pointing at the P or p. On exit, it is pointing at the final character of the
917 ptrptr points to the pattern position pointer
918 negptr points to a boolean that is set TRUE for negation else FALSE
919 dptr points to an int that is set to the detailed property value
920 errorcodeptr points to the error code variable
922 Returns: type value from ucp_type_table, or -1 for an invalid type
926 get_ucp(const uschar **ptrptr, BOOL *negptr, int *dptr, int *errorcodeptr)
929 const uschar *ptr = *ptrptr;
933 if (c == 0) goto ERROR_RETURN;
937 /* \P or \p can be followed by a name in {}, optionally preceded by ^ for
940 if (c == CHAR_LEFT_CURLY_BRACKET)
942 if (ptr[1] == CHAR_CIRCUMFLEX_ACCENT)
947 for (i = 0; i < (int)sizeof(name) - 1; i++)
950 if (c == 0) goto ERROR_RETURN;
951 if (c == CHAR_RIGHT_CURLY_BRACKET) break;
954 if (c != CHAR_RIGHT_CURLY_BRACKET) goto ERROR_RETURN;
958 /* Otherwise there is just one following character */
968 /* Search for a recognized property name using binary chop */
971 top = _pcre_utt_size;
975 i = (bot + top) >> 1;
976 c = strcmp(name, _pcre_utt_names + _pcre_utt[i].name_offset);
979 *dptr = _pcre_utt[i].value;
980 return _pcre_utt[i].type;
982 if (c > 0) bot = i + 1; else top = i;
985 *errorcodeptr = ERR47;
990 *errorcodeptr = ERR46;
999 /*************************************************
1000 * Check for counted repeat *
1001 *************************************************/
1003 /* This function is called when a '{' is encountered in a place where it might
1004 start a quantifier. It looks ahead to see if it really is a quantifier or not.
1005 It is only a quantifier if it is one of the forms {ddd} {ddd,} or {ddd,ddd}
1006 where the ddds are digits.
1009 p pointer to the first char after '{'
1011 Returns: TRUE or FALSE
1015 is_counted_repeat(const uschar *p)
1017 if ((digitab[*p++] & ctype_digit) == 0) return FALSE;
1018 while ((digitab[*p] & ctype_digit) != 0) p++;
1019 if (*p == CHAR_RIGHT_CURLY_BRACKET) return TRUE;
1021 if (*p++ != CHAR_COMMA) return FALSE;
1022 if (*p == CHAR_RIGHT_CURLY_BRACKET) return TRUE;
1024 if ((digitab[*p++] & ctype_digit) == 0) return FALSE;
1025 while ((digitab[*p] & ctype_digit) != 0) p++;
1027 return (*p == CHAR_RIGHT_CURLY_BRACKET);
1032 /*************************************************
1033 * Read repeat counts *
1034 *************************************************/
1036 /* Read an item of the form {n,m} and return the values. This is called only
1037 after is_counted_repeat() has confirmed that a repeat-count quantifier exists,
1038 so the syntax is guaranteed to be correct, but we need to check the values.
1041 p pointer to first char after '{'
1042 minp pointer to int for min
1043 maxp pointer to int for max
1044 returned as -1 if no max
1045 errorcodeptr points to error code variable
1047 Returns: pointer to '}' on success;
1048 current ptr on error, with errorcodeptr set non-zero
1051 static const uschar *
1052 read_repeat_counts(const uschar *p, int *minp, int *maxp, int *errorcodeptr)
1057 /* Read the minimum value and do a paranoid check: a negative value indicates
1058 an integer overflow. */
1060 while ((digitab[*p] & ctype_digit) != 0) min = min * 10 + *p++ - CHAR_0;
1061 if (min < 0 || min > 65535)
1063 *errorcodeptr = ERR5;
1067 /* Read the maximum value if there is one, and again do a paranoid on its size.
1068 Also, max must not be less than min. */
1070 if (*p == CHAR_RIGHT_CURLY_BRACKET) max = min; else
1072 if (*(++p) != CHAR_RIGHT_CURLY_BRACKET)
1075 while((digitab[*p] & ctype_digit) != 0) max = max * 10 + *p++ - CHAR_0;
1076 if (max < 0 || max > 65535)
1078 *errorcodeptr = ERR5;
1083 *errorcodeptr = ERR4;
1089 /* Fill in the required variables, and pass back the pointer to the terminating
1099 /*************************************************
1100 * Subroutine for finding forward reference *
1101 *************************************************/
1103 /* This recursive function is called only from find_parens() below. The
1104 top-level call starts at the beginning of the pattern. All other calls must
1105 start at a parenthesis. It scans along a pattern's text looking for capturing
1106 subpatterns, and counting them. If it finds a named pattern that matches the
1107 name it is given, it returns its number. Alternatively, if the name is NULL, it
1108 returns when it reaches a given numbered subpattern. Recursion is used to keep
1109 track of subpatterns that reset the capturing group numbers - the (?| feature.
1111 This function was originally called only from the second pass, in which we know
1112 that if (?< or (?' or (?P< is encountered, the name will be correctly
1113 terminated because that is checked in the first pass. There is now one call to
1114 this function in the first pass, to check for a recursive back reference by
1115 name (so that we can make the whole group atomic). In this case, we need check
1116 only up to the current position in the pattern, and that is still OK because
1117 and previous occurrences will have been checked. To make this work, the test
1118 for "end of pattern" is a check against cd->end_pattern in the main loop,
1119 instead of looking for a binary zero. This means that the special first-pass
1120 call can adjust cd->end_pattern temporarily. (Checks for binary zero while
1121 processing items within the loop are OK, because afterwards the main loop will
1125 ptrptr address of the current character pointer (updated)
1126 cd compile background data
1127 name name to seek, or NULL if seeking a numbered subpattern
1128 lorn name length, or subpattern number if name is NULL
1129 xmode TRUE if we are in /x mode
1130 utf8 TRUE if we are in UTF-8 mode
1131 count pointer to the current capturing subpattern number (updated)
1133 Returns: the number of the named subpattern, or -1 if not found
1137 find_parens_sub(uschar **ptrptr, compile_data *cd, const uschar *name, int lorn,
1138 BOOL xmode, BOOL utf8, int *count)
1140 uschar *ptr = *ptrptr;
1141 int start_count = *count;
1142 int hwm_count = start_count;
1143 BOOL dup_parens = FALSE;
1145 /* If the first character is a parenthesis, check on the type of group we are
1146 dealing with. The very first call may not start with a parenthesis. */
1148 if (ptr[0] == CHAR_LEFT_PARENTHESIS)
1150 /* Handle specials such as (*SKIP) or (*UTF8) etc. */
1152 if (ptr[1] == CHAR_ASTERISK) ptr += 2;
1154 /* Handle a normal, unnamed capturing parenthesis. */
1156 else if (ptr[1] != CHAR_QUESTION_MARK)
1159 if (name == NULL && *count == lorn) return *count;
1163 /* All cases now have (? at the start. Remember when we are in a group
1164 where the parenthesis numbers are duplicated. */
1166 else if (ptr[2] == CHAR_VERTICAL_LINE)
1172 /* Handle comments; all characters are allowed until a ket is reached. */
1174 else if (ptr[2] == CHAR_NUMBER_SIGN)
1176 for (ptr += 3; *ptr != 0; ptr++) if (*ptr == CHAR_RIGHT_PARENTHESIS) break;
1180 /* Handle a condition. If it is an assertion, just carry on so that it
1181 is processed as normal. If not, skip to the closing parenthesis of the
1182 condition (there can't be any nested parens). */
1184 else if (ptr[2] == CHAR_LEFT_PARENTHESIS)
1187 if (ptr[1] != CHAR_QUESTION_MARK)
1189 while (*ptr != 0 && *ptr != CHAR_RIGHT_PARENTHESIS) ptr++;
1190 if (*ptr != 0) ptr++;
1194 /* Start with (? but not a condition. */
1199 if (*ptr == CHAR_P) ptr++; /* Allow optional P */
1201 /* We have to disambiguate (?<! and (?<= from (?<name> for named groups */
1203 if ((*ptr == CHAR_LESS_THAN_SIGN && ptr[1] != CHAR_EXCLAMATION_MARK &&
1204 ptr[1] != CHAR_EQUALS_SIGN) || *ptr == CHAR_APOSTROPHE)
1207 const uschar *thisname;
1209 if (name == NULL && *count == lorn) return *count;
1211 if (term == CHAR_LESS_THAN_SIGN) term = CHAR_GREATER_THAN_SIGN;
1213 while (*ptr != term) ptr++;
1214 if (name != NULL && lorn == ptr - thisname &&
1215 strncmp((const char *)name, (const char *)thisname, lorn) == 0)
1222 /* Past any initial parenthesis handling, scan for parentheses or vertical
1223 bars. Stop if we get to cd->end_pattern. Note that this is important for the
1224 first-pass call when this value is temporarily adjusted to stop at the current
1225 position. So DO NOT change this to a test for binary zero. */
1227 for (; ptr < cd->end_pattern; ptr++)
1229 /* Skip over backslashed characters and also entire \Q...\E */
1231 if (*ptr == CHAR_BACKSLASH)
1233 if (*(++ptr) == 0) goto FAIL_EXIT;
1234 if (*ptr == CHAR_Q) for (;;)
1236 while (*(++ptr) != 0 && *ptr != CHAR_BACKSLASH) {};
1237 if (*ptr == 0) goto FAIL_EXIT;
1238 if (*(++ptr) == CHAR_E) break;
1243 /* Skip over character classes; this logic must be similar to the way they
1244 are handled for real. If the first character is '^', skip it. Also, if the
1245 first few characters (either before or after ^) are \Q\E or \E we skip them
1246 too. This makes for compatibility with Perl. Note the use of STR macros to
1247 encode "Q\\E" so that it works in UTF-8 on EBCDIC platforms. */
1249 if (*ptr == CHAR_LEFT_SQUARE_BRACKET)
1251 BOOL negate_class = FALSE;
1254 if (ptr[1] == CHAR_BACKSLASH)
1256 if (ptr[2] == CHAR_E)
1258 else if (strncmp((const char *)ptr+2,
1259 STR_Q STR_BACKSLASH STR_E, 3) == 0)
1264 else if (!negate_class && ptr[1] == CHAR_CIRCUMFLEX_ACCENT)
1266 negate_class = TRUE;
1272 /* If the next character is ']', it is a data character that must be
1273 skipped, except in JavaScript compatibility mode. */
1275 if (ptr[1] == CHAR_RIGHT_SQUARE_BRACKET &&
1276 (cd->external_options & PCRE_JAVASCRIPT_COMPAT) == 0)
1279 while (*(++ptr) != CHAR_RIGHT_SQUARE_BRACKET)
1281 if (*ptr == 0) return -1;
1282 if (*ptr == CHAR_BACKSLASH)
1284 if (*(++ptr) == 0) goto FAIL_EXIT;
1285 if (*ptr == CHAR_Q) for (;;)
1287 while (*(++ptr) != 0 && *ptr != CHAR_BACKSLASH) {};
1288 if (*ptr == 0) goto FAIL_EXIT;
1289 if (*(++ptr) == CHAR_E) break;
1297 /* Skip comments in /x mode */
1299 if (xmode && *ptr == CHAR_NUMBER_SIGN)
1304 if (IS_NEWLINE(ptr)) { ptr += cd->nllen - 1; break; }
1307 if (utf8) while ((*ptr & 0xc0) == 0x80) ptr++;
1310 if (*ptr == 0) goto FAIL_EXIT;
1314 /* Check for the special metacharacters */
1316 if (*ptr == CHAR_LEFT_PARENTHESIS)
1318 int rc = find_parens_sub(&ptr, cd, name, lorn, xmode, utf8, count);
1319 if (rc > 0) return rc;
1320 if (*ptr == 0) goto FAIL_EXIT;
1323 else if (*ptr == CHAR_RIGHT_PARENTHESIS)
1325 if (dup_parens && *count < hwm_count) *count = hwm_count;
1329 else if (*ptr == CHAR_VERTICAL_LINE && dup_parens)
1331 if (*count > hwm_count) hwm_count = *count;
1332 *count = start_count;
1344 /*************************************************
1345 * Find forward referenced subpattern *
1346 *************************************************/
1348 /* This function scans along a pattern's text looking for capturing
1349 subpatterns, and counting them. If it finds a named pattern that matches the
1350 name it is given, it returns its number. Alternatively, if the name is NULL, it
1351 returns when it reaches a given numbered subpattern. This is used for forward
1352 references to subpatterns. We used to be able to start this scan from the
1353 current compiling point, using the current count value from cd->bracount, and
1354 do it all in a single loop, but the addition of the possibility of duplicate
1355 subpattern numbers means that we have to scan from the very start, in order to
1356 take account of such duplicates, and to use a recursive function to keep track
1357 of the different types of group.
1360 cd compile background data
1361 name name to seek, or NULL if seeking a numbered subpattern
1362 lorn name length, or subpattern number if name is NULL
1363 xmode TRUE if we are in /x mode
1364 utf8 TRUE if we are in UTF-8 mode
1366 Returns: the number of the found subpattern, or -1 if not found
1370 find_parens(compile_data *cd, const uschar *name, int lorn, BOOL xmode,
1373 uschar *ptr = (uschar *)cd->start_pattern;
1377 /* If the pattern does not start with an opening parenthesis, the first call
1378 to find_parens_sub() will scan right to the end (if necessary). However, if it
1379 does start with a parenthesis, find_parens_sub() will return when it hits the
1380 matching closing parens. That is why we have to have a loop. */
1384 rc = find_parens_sub(&ptr, cd, name, lorn, xmode, utf8, &count);
1385 if (rc > 0 || *ptr++ == 0) break;
1394 /*************************************************
1395 * Find first significant op code *
1396 *************************************************/
1398 /* This is called by several functions that scan a compiled expression looking
1399 for a fixed first character, or an anchoring op code etc. It skips over things
1400 that do not influence this. For some calls, a change of option is important.
1401 For some calls, it makes sense to skip negative forward and all backward
1402 assertions, and also the \b assertion; for others it does not.
1405 code pointer to the start of the group
1406 options pointer to external options
1407 optbit the option bit whose changing is significant, or
1409 skipassert TRUE if certain assertions are to be skipped
1411 Returns: pointer to the first significant opcode
1414 static const uschar*
1415 first_significant_code(const uschar *code, int *options, int optbit,
1423 if (optbit > 0 && ((int)code[1] & optbit) != (*options & optbit))
1424 *options = (int)code[1];
1430 case OP_ASSERTBACK_NOT:
1431 if (!skipassert) return code;
1432 do code += GET(code, 1); while (*code == OP_ALT);
1433 code += _pcre_OP_lengths[*code];
1436 case OP_WORD_BOUNDARY:
1437 case OP_NOT_WORD_BOUNDARY:
1438 if (!skipassert) return code;
1447 code += _pcre_OP_lengths[*code];
1454 /* Control never reaches here */
1460 /*************************************************
1461 * Find the fixed length of a branch *
1462 *************************************************/
1464 /* Scan a branch and compute the fixed length of subject that will match it,
1465 if the length is fixed. This is needed for dealing with backward assertions.
1466 In UTF8 mode, the result is in characters rather than bytes. The branch is
1467 temporarily terminated with OP_END when this function is called.
1469 This function is called when a backward assertion is encountered, so that if it
1470 fails, the error message can point to the correct place in the pattern.
1471 However, we cannot do this when the assertion contains subroutine calls,
1472 because they can be forward references. We solve this by remembering this case
1473 and doing the check at the end; a flag specifies which mode we are running in.
1476 code points to the start of the pattern (the bracket)
1477 options the compiling options
1478 atend TRUE if called when the pattern is complete
1479 cd the "compile data" structure
1481 Returns: the fixed length,
1482 or -1 if there is no fixed length,
1483 or -2 if \C was encountered
1484 or -3 if an OP_RECURSE item was encountered and atend is FALSE
1488 find_fixedlength(uschar *code, int options, BOOL atend, compile_data *cd)
1492 register int branchlength = 0;
1493 register uschar *cc = code + 1 + LINK_SIZE;
1495 /* Scan along the opcodes for this branch. If we get to the end of the
1496 branch, check the length against that of the other branches. */
1502 register int op = *cc;
1509 d = find_fixedlength(cc + ((op == OP_CBRA)? 2:0), options, atend, cd);
1510 if (d < 0) return d;
1512 do cc += GET(cc, 1); while (*cc == OP_ALT);
1513 cc += 1 + LINK_SIZE;
1516 /* Reached end of a branch; if it's a ket it is the end of a nested
1517 call. If it's ALT it is an alternation in a nested call. If it is
1518 END it's the end of the outer call. All can be handled by the same code. */
1525 if (length < 0) length = branchlength;
1526 else if (length != branchlength) return -1;
1527 if (*cc != OP_ALT) return length;
1528 cc += 1 + LINK_SIZE;
1532 /* A true recursion implies not fixed length, but a subroutine call may
1533 be OK. If the subroutine is a forward reference, we can't deal with
1534 it until the end of the pattern, so return -3. */
1537 if (!atend) return -3;
1538 cs = ce = (uschar *)cd->start_code + GET(cc, 1); /* Start subpattern */
1539 do ce += GET(ce, 1); while (*ce == OP_ALT); /* End subpattern */
1540 if (cc > cs && cc < ce) return -1; /* Recursion */
1541 d = find_fixedlength(cs + 2, options, atend, cd);
1542 if (d < 0) return d;
1544 cc += 1 + LINK_SIZE;
1547 /* Skip over assertive subpatterns */
1552 case OP_ASSERTBACK_NOT:
1553 do cc += GET(cc, 1); while (*cc == OP_ALT);
1556 /* Skip over things that don't match chars */
1573 case OP_NOT_WORD_BOUNDARY:
1574 case OP_WORD_BOUNDARY:
1575 cc += _pcre_OP_lengths[*cc];
1578 /* Handle literal characters */
1586 if ((options & PCRE_UTF8) != 0 && cc[-1] >= 0xc0)
1587 cc += _pcre_utf8_table4[cc[-1] & 0x3f];
1591 /* Handle exact repetitions. The count is already in characters, but we
1592 need to skip over a multibyte character in UTF8 mode. */
1595 branchlength += GET2(cc,1);
1598 if ((options & PCRE_UTF8) != 0 && cc[-1] >= 0xc0)
1599 cc += _pcre_utf8_table4[cc[-1] & 0x3f];
1604 branchlength += GET2(cc,1);
1605 if (cc[3] == OP_PROP || cc[3] == OP_NOTPROP) cc += 2;
1609 /* Handle single-char matchers */
1618 case OP_NOT_WHITESPACE:
1620 case OP_NOT_WORDCHAR:
1628 /* The single-byte matcher isn't allowed */
1633 /* Check a class for variable quantification */
1637 cc += GET(cc, 1) - 33;
1655 if (GET2(cc,1) != GET2(cc,3)) return -1;
1656 branchlength += GET2(cc,1);
1665 /* Anything else is variable length */
1671 /* Control never gets here */
1677 /*************************************************
1678 * Scan compiled regex for specific bracket *
1679 *************************************************/
1681 /* This little function scans through a compiled pattern until it finds a
1682 capturing bracket with the given number, or, if the number is negative, an
1683 instance of OP_REVERSE for a lookbehind. The function is global in the C sense
1684 so that it can be called from pcre_study() when finding the minimum matching
1688 code points to start of expression
1689 utf8 TRUE in UTF-8 mode
1690 number the required bracket number or negative to find a lookbehind
1692 Returns: pointer to the opcode for the bracket, or NULL if not found
1696 _pcre_find_bracket(const uschar *code, BOOL utf8, int number)
1700 register int c = *code;
1701 if (c == OP_END) return NULL;
1703 /* XCLASS is used for classes that cannot be represented just by a bit
1704 map. This includes negated single high-valued characters. The length in
1705 the table is zero; the actual length is stored in the compiled code. */
1707 if (c == OP_XCLASS) code += GET(code, 1);
1709 /* Handle recursion */
1711 else if (c == OP_REVERSE)
1713 if (number < 0) return (uschar *)code;
1714 code += _pcre_OP_lengths[c];
1717 /* Handle capturing bracket */
1719 else if (c == OP_CBRA)
1721 int n = GET2(code, 1+LINK_SIZE);
1722 if (n == number) return (uschar *)code;
1723 code += _pcre_OP_lengths[c];
1726 /* Otherwise, we can get the item's length from the table, except that for
1727 repeated character types, we have to test for \p and \P, which have an extra
1728 two bytes of parameters, and for MARK/PRUNE/SKIP/THEN with an argument, we
1729 must add in its length. */
1736 case OP_TYPEMINSTAR:
1738 case OP_TYPEMINPLUS:
1740 case OP_TYPEMINQUERY:
1741 case OP_TYPEPOSSTAR:
1742 case OP_TYPEPOSPLUS:
1743 case OP_TYPEPOSQUERY:
1744 if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
1748 case OP_TYPEMINUPTO:
1750 case OP_TYPEPOSUPTO:
1751 if (code[3] == OP_PROP || code[3] == OP_NOTPROP) code += 2;
1761 code += code[1+LINK_SIZE];
1765 /* Add in the fixed length from the table */
1767 code += _pcre_OP_lengths[c];
1769 /* In UTF-8 mode, opcodes that are followed by a character may be followed by
1770 a multi-byte character. The length in the table is a minimum, so we have to
1771 arrange to skip the extra bytes. */
1791 if (code[-1] >= 0xc0) code += _pcre_utf8_table4[code[-1] & 0x3f];
1795 (void)(utf8); /* Keep compiler happy by referencing function argument */
1803 /*************************************************
1804 * Scan compiled regex for recursion reference *
1805 *************************************************/
1807 /* This little function scans through a compiled pattern until it finds an
1808 instance of OP_RECURSE.
1811 code points to start of expression
1812 utf8 TRUE in UTF-8 mode
1814 Returns: pointer to the opcode for OP_RECURSE, or NULL if not found
1817 static const uschar *
1818 find_recurse(const uschar *code, BOOL utf8)
1822 register int c = *code;
1823 if (c == OP_END) return NULL;
1824 if (c == OP_RECURSE) return code;
1826 /* XCLASS is used for classes that cannot be represented just by a bit
1827 map. This includes negated single high-valued characters. The length in
1828 the table is zero; the actual length is stored in the compiled code. */
1830 if (c == OP_XCLASS) code += GET(code, 1);
1832 /* Otherwise, we can get the item's length from the table, except that for
1833 repeated character types, we have to test for \p and \P, which have an extra
1834 two bytes of parameters, and for MARK/PRUNE/SKIP/THEN with an argument, we
1835 must add in its length. */
1842 case OP_TYPEMINSTAR:
1844 case OP_TYPEMINPLUS:
1846 case OP_TYPEMINQUERY:
1847 case OP_TYPEPOSSTAR:
1848 case OP_TYPEPOSPLUS:
1849 case OP_TYPEPOSQUERY:
1850 if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
1853 case OP_TYPEPOSUPTO:
1855 case OP_TYPEMINUPTO:
1857 if (code[3] == OP_PROP || code[3] == OP_NOTPROP) code += 2;
1867 code += code[1+LINK_SIZE];
1871 /* Add in the fixed length from the table */
1873 code += _pcre_OP_lengths[c];
1875 /* In UTF-8 mode, opcodes that are followed by a character may be followed
1876 by a multi-byte character. The length in the table is a minimum, so we have
1877 to arrange to skip the extra bytes. */
1897 if (code[-1] >= 0xc0) code += _pcre_utf8_table4[code[-1] & 0x3f];
1901 (void)(utf8); /* Keep compiler happy by referencing function argument */
1909 /*************************************************
1910 * Scan compiled branch for non-emptiness *
1911 *************************************************/
1913 /* This function scans through a branch of a compiled pattern to see whether it
1914 can match the empty string or not. It is called from could_be_empty()
1915 below and from compile_branch() when checking for an unlimited repeat of a
1916 group that can match nothing. Note that first_significant_code() skips over
1917 backward and negative forward assertions when its final argument is TRUE. If we
1918 hit an unclosed bracket, we return "empty" - this means we've struck an inner
1919 bracket whose current branch will already have been scanned.
1922 code points to start of search
1923 endcode points to where to stop
1924 utf8 TRUE if in UTF8 mode
1925 cd contains pointers to tables etc.
1927 Returns: TRUE if what is matched could be empty
1931 could_be_empty_branch(const uschar *code, const uschar *endcode, BOOL utf8,
1935 for (code = first_significant_code(code + _pcre_OP_lengths[*code], NULL, 0, TRUE);
1937 code = first_significant_code(code + _pcre_OP_lengths[c], NULL, 0, TRUE))
1939 const uschar *ccode;
1943 /* Skip over forward assertions; the other assertions are skipped by
1944 first_significant_code() with a TRUE final argument. */
1948 do code += GET(code, 1); while (*code == OP_ALT);
1953 /* Groups with zero repeats can of course be empty; skip them. */
1955 if (c == OP_BRAZERO || c == OP_BRAMINZERO || c == OP_SKIPZERO)
1957 code += _pcre_OP_lengths[c];
1958 do code += GET(code, 1); while (*code == OP_ALT);
1963 /* For a recursion/subroutine call, if its end has been reached, which
1964 implies a subroutine call, we can scan it. */
1966 if (c == OP_RECURSE)
1968 BOOL empty_branch = FALSE;
1969 const uschar *scode = cd->start_code + GET(code, 1);
1970 if (GET(scode, 1) == 0) return TRUE; /* Unclosed */
1973 if (could_be_empty_branch(scode, endcode, utf8, cd))
1975 empty_branch = TRUE;
1978 scode += GET(scode, 1);
1980 while (*scode == OP_ALT);
1981 if (!empty_branch) return FALSE; /* All branches are non-empty */
1985 /* For other groups, scan the branches. */
1987 if (c == OP_BRA || c == OP_CBRA || c == OP_ONCE || c == OP_COND)
1990 if (GET(code, 1) == 0) return TRUE; /* Hit unclosed bracket */
1992 /* If a conditional group has only one branch, there is a second, implied,
1993 empty branch, so just skip over the conditional, because it could be empty.
1994 Otherwise, scan the individual branches of the group. */
1996 if (c == OP_COND && code[GET(code, 1)] != OP_ALT)
1997 code += GET(code, 1);
2000 empty_branch = FALSE;
2003 if (!empty_branch && could_be_empty_branch(code, endcode, utf8, cd))
2004 empty_branch = TRUE;
2005 code += GET(code, 1);
2007 while (*code == OP_ALT);
2008 if (!empty_branch) return FALSE; /* All branches are non-empty */
2015 /* Handle the other opcodes */
2019 /* Check for quantifiers after a class. XCLASS is used for classes that
2020 cannot be represented just by a bit map. This includes negated single
2021 high-valued characters. The length in _pcre_OP_lengths[] is zero; the
2022 actual length is stored in the compiled code, so we must update "code"
2027 ccode = code += GET(code, 1);
2028 goto CHECK_CLASS_REPEAT;
2041 case OP_CRSTAR: /* These could be empty; continue */
2047 default: /* Non-repeat => class must match */
2048 case OP_CRPLUS: /* These repeats aren't empty */
2054 if (GET2(ccode, 1) > 0) return FALSE; /* Minimum > 0 */
2059 /* Opcodes that must match a character */
2066 case OP_NOT_WHITESPACE:
2068 case OP_NOT_WORDCHAR:
2085 case OP_TYPEMINPLUS:
2086 case OP_TYPEPOSPLUS:
2090 /* These are going to continue, as they may be empty, but we have to
2091 fudge the length for the \p and \P cases. */
2094 case OP_TYPEMINSTAR:
2095 case OP_TYPEPOSSTAR:
2097 case OP_TYPEMINQUERY:
2098 case OP_TYPEPOSQUERY:
2099 if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
2102 /* Same for these */
2105 case OP_TYPEMINUPTO:
2106 case OP_TYPEPOSUPTO:
2107 if (code[3] == OP_PROP || code[3] == OP_NOTPROP) code += 2;
2118 /* In UTF-8 mode, STAR, MINSTAR, POSSTAR, QUERY, MINQUERY, POSQUERY, UPTO,
2119 MINUPTO, and POSUPTO may be followed by a multibyte character */
2128 if (utf8 && code[1] >= 0xc0) code += _pcre_utf8_table4[code[1] & 0x3f];
2134 if (utf8 && code[3] >= 0xc0) code += _pcre_utf8_table4[code[3] & 0x3f];
2138 /* MARK, and PRUNE/SKIP/THEN with an argument must skip over the argument
2148 code += code[1+LINK_SIZE];
2151 /* None of the remaining opcodes are required to match a character. */
2163 /*************************************************
2164 * Scan compiled regex for non-emptiness *
2165 *************************************************/
2167 /* This function is called to check for left recursive calls. We want to check
2168 the current branch of the current pattern to see if it could match the empty
2169 string. If it could, we must look outwards for branches at other levels,
2170 stopping when we pass beyond the bracket which is the subject of the recursion.
2173 code points to start of the recursion
2174 endcode points to where to stop (current RECURSE item)
2175 bcptr points to the chain of current (unclosed) branch starts
2176 utf8 TRUE if in UTF-8 mode
2177 cd pointers to tables etc
2179 Returns: TRUE if what is matched could be empty
2183 could_be_empty(const uschar *code, const uschar *endcode, branch_chain *bcptr,
2184 BOOL utf8, compile_data *cd)
2186 while (bcptr != NULL && bcptr->current_branch >= code)
2188 if (!could_be_empty_branch(bcptr->current_branch, endcode, utf8, cd))
2190 bcptr = bcptr->outer;
2197 /*************************************************
2198 * Check for POSIX class syntax *
2199 *************************************************/
2201 /* This function is called when the sequence "[:" or "[." or "[=" is
2202 encountered in a character class. It checks whether this is followed by a
2203 sequence of characters terminated by a matching ":]" or ".]" or "=]". If we
2204 reach an unescaped ']' without the special preceding character, return FALSE.
2206 Originally, this function only recognized a sequence of letters between the
2207 terminators, but it seems that Perl recognizes any sequence of characters,
2208 though of course unknown POSIX names are subsequently rejected. Perl gives an
2209 "Unknown POSIX class" error for [:f\oo:] for example, where previously PCRE
2210 didn't consider this to be a POSIX class. Likewise for [:1234:].
2212 The problem in trying to be exactly like Perl is in the handling of escapes. We
2213 have to be sure that [abc[:x\]pqr] is *not* treated as containing a POSIX
2214 class, but [abc[:x\]pqr:]] is (so that an error can be generated). The code
2215 below handles the special case of \], but does not try to do any other escape
2216 processing. This makes it different from Perl for cases such as [:l\ower:]
2217 where Perl recognizes it as the POSIX class "lower" but PCRE does not recognize
2218 "l\ower". This is a lesser evil that not diagnosing bad classes when Perl does,
2222 ptr pointer to the initial [
2223 endptr where to return the end pointer
2225 Returns: TRUE or FALSE
2229 check_posix_syntax(const uschar *ptr, const uschar **endptr)
2231 int terminator; /* Don't combine these lines; the Solaris cc */
2232 terminator = *(++ptr); /* compiler warns about "non-constant" initializer. */
2233 for (++ptr; *ptr != 0; ptr++)
2235 if (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET) ptr++; else
2237 if (*ptr == CHAR_RIGHT_SQUARE_BRACKET) return FALSE;
2238 if (*ptr == terminator && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET)
2251 /*************************************************
2252 * Check POSIX class name *
2253 *************************************************/
2255 /* This function is called to check the name given in a POSIX-style class entry
2259 ptr points to the first letter
2260 len the length of the name
2262 Returns: a value representing the name, or -1 if unknown
2266 check_posix_name(const uschar *ptr, int len)
2268 const char *pn = posix_names;
2269 register int yield = 0;
2270 while (posix_name_lengths[yield] != 0)
2272 if (len == posix_name_lengths[yield] &&
2273 strncmp((const char *)ptr, pn, len) == 0) return yield;
2274 pn += posix_name_lengths[yield] + 1;
2281 /*************************************************
2282 * Adjust OP_RECURSE items in repeated group *
2283 *************************************************/
2285 /* OP_RECURSE items contain an offset from the start of the regex to the group
2286 that is referenced. This means that groups can be replicated for fixed
2287 repetition simply by copying (because the recursion is allowed to refer to
2288 earlier groups that are outside the current group). However, when a group is
2289 optional (i.e. the minimum quantifier is zero), OP_BRAZERO or OP_SKIPZERO is
2290 inserted before it, after it has been compiled. This means that any OP_RECURSE
2291 items within it that refer to the group itself or any contained groups have to
2292 have their offsets adjusted. That one of the jobs of this function. Before it
2293 is called, the partially compiled regex must be temporarily terminated with
2296 This function has been extended with the possibility of forward references for
2297 recursions and subroutine calls. It must also check the list of such references
2298 for the group we are dealing with. If it finds that one of the recursions in
2299 the current group is on this list, it adjusts the offset in the list, not the
2300 value in the reference (which is a group number).
2303 group points to the start of the group
2304 adjust the amount by which the group is to be moved
2305 utf8 TRUE in UTF-8 mode
2306 cd contains pointers to tables etc.
2307 save_hwm the hwm forward reference pointer at the start of the group
2313 adjust_recurse(uschar *group, int adjust, BOOL utf8, compile_data *cd,
2316 uschar *ptr = group;
2318 while ((ptr = (uschar *)find_recurse(ptr, utf8)) != NULL)
2323 /* See if this recursion is on the forward reference list. If so, adjust the
2326 for (hc = save_hwm; hc < cd->hwm; hc += LINK_SIZE)
2328 offset = GET(hc, 0);
2329 if (cd->start_code + offset == ptr + 1)
2331 PUT(hc, 0, offset + adjust);
2336 /* Otherwise, adjust the recursion offset if it's after the start of this
2341 offset = GET(ptr, 1);
2342 if (cd->start_code + offset >= group) PUT(ptr, 1, offset + adjust);
2345 ptr += 1 + LINK_SIZE;
2351 /*************************************************
2352 * Insert an automatic callout point *
2353 *************************************************/
2355 /* This function is called when the PCRE_AUTO_CALLOUT option is set, to insert
2356 callout points before each pattern item.
2359 code current code pointer
2360 ptr current pattern pointer
2361 cd pointers to tables etc
2363 Returns: new code pointer
2367 auto_callout(uschar *code, const uschar *ptr, compile_data *cd)
2369 *code++ = OP_CALLOUT;
2371 PUT(code, 0, (int)(ptr - cd->start_pattern)); /* Pattern offset */
2372 PUT(code, LINK_SIZE, 0); /* Default length */
2373 return code + 2*LINK_SIZE;
2378 /*************************************************
2379 * Complete a callout item *
2380 *************************************************/
2382 /* A callout item contains the length of the next item in the pattern, which
2383 we can't fill in till after we have reached the relevant point. This is used
2384 for both automatic and manual callouts.
2387 previous_callout points to previous callout item
2388 ptr current pattern pointer
2389 cd pointers to tables etc
2395 complete_callout(uschar *previous_callout, const uschar *ptr, compile_data *cd)
2397 int length = (int)(ptr - cd->start_pattern - GET(previous_callout, 2));
2398 PUT(previous_callout, 2 + LINK_SIZE, length);
2404 /*************************************************
2405 * Get othercase range *
2406 *************************************************/
2408 /* This function is passed the start and end of a class range, in UTF-8 mode
2409 with UCP support. It searches up the characters, looking for internal ranges of
2410 characters in the "other" case. Each call returns the next one, updating the
2414 cptr points to starting character value; updated
2416 ocptr where to put start of othercase range
2417 odptr where to put end of othercase range
2419 Yield: TRUE when range returned; FALSE when no more
2423 get_othercase_range(unsigned int *cptr, unsigned int d, unsigned int *ocptr,
2424 unsigned int *odptr)
2426 unsigned int c, othercase, next;
2428 for (c = *cptr; c <= d; c++)
2429 { if ((othercase = UCD_OTHERCASE(c)) != c) break; }
2431 if (c > d) return FALSE;
2434 next = othercase + 1;
2436 for (++c; c <= d; c++)
2438 if (UCD_OTHERCASE(c) != next) break;
2450 /*************************************************
2451 * Check a character and a property *
2452 *************************************************/
2454 /* This function is called by check_auto_possessive() when a property item
2455 is adjacent to a fixed character.
2459 ptype the property type
2460 pdata the data for the type
2461 negated TRUE if it's a negated property (\P or \p{^)
2463 Returns: TRUE if auto-possessifying is OK
2467 check_char_prop(int c, int ptype, int pdata, BOOL negated)
2469 const ucd_record *prop = GET_UCD(c);
2473 return (prop->chartype == ucp_Lu ||
2474 prop->chartype == ucp_Ll ||
2475 prop->chartype == ucp_Lt) == negated;
2478 return (pdata == _pcre_ucp_gentype[prop->chartype]) == negated;
2481 return (pdata == prop->chartype) == negated;
2484 return (pdata == prop->script) == negated;
2486 /* These are specials */
2489 return (_pcre_ucp_gentype[prop->chartype] == ucp_L ||
2490 _pcre_ucp_gentype[prop->chartype] == ucp_N) == negated;
2492 case PT_SPACE: /* Perl space */
2493 return (_pcre_ucp_gentype[prop->chartype] == ucp_Z ||
2494 c == CHAR_HT || c == CHAR_NL || c == CHAR_FF || c == CHAR_CR)
2497 case PT_PXSPACE: /* POSIX space */
2498 return (_pcre_ucp_gentype[prop->chartype] == ucp_Z ||
2499 c == CHAR_HT || c == CHAR_NL || c == CHAR_VT ||
2500 c == CHAR_FF || c == CHAR_CR)
2504 return (_pcre_ucp_gentype[prop->chartype] == ucp_L ||
2505 _pcre_ucp_gentype[prop->chartype] == ucp_N ||
2506 c == CHAR_UNDERSCORE) == negated;
2510 #endif /* SUPPORT_UCP */
2514 /*************************************************
2515 * Check if auto-possessifying is possible *
2516 *************************************************/
2518 /* This function is called for unlimited repeats of certain items, to see
2519 whether the next thing could possibly match the repeated item. If not, it makes
2520 sense to automatically possessify the repeated item.
2523 previous pointer to the repeated opcode
2524 utf8 TRUE in UTF-8 mode
2525 ptr next character in pattern
2526 options options bits
2527 cd contains pointers to tables etc.
2529 Returns: TRUE if possessifying is wanted
2533 check_auto_possessive(const uschar *previous, BOOL utf8, const uschar *ptr,
2534 int options, compile_data *cd)
2537 int op_code = *previous++;
2539 /* Skip whitespace and comments in extended mode */
2541 if ((options & PCRE_EXTENDED) != 0)
2545 while ((cd->ctypes[*ptr] & ctype_space) != 0) ptr++;
2546 if (*ptr == CHAR_NUMBER_SIGN)
2551 if (IS_NEWLINE(ptr)) { ptr += cd->nllen; break; }
2554 if (utf8) while ((*ptr & 0xc0) == 0x80) ptr++;
2562 /* If the next item is one that we can handle, get its value. A non-negative
2563 value is a character, a negative value is an escape value. */
2565 if (*ptr == CHAR_BACKSLASH)
2567 int temperrorcode = 0;
2568 next = check_escape(&ptr, &temperrorcode, cd->bracount, options, FALSE);
2569 if (temperrorcode != 0) return FALSE;
2570 ptr++; /* Point after the escape sequence */
2573 else if ((cd->ctypes[*ptr] & ctype_meta) == 0)
2576 if (utf8) { GETCHARINC(next, ptr); } else
2583 /* Skip whitespace and comments in extended mode */
2585 if ((options & PCRE_EXTENDED) != 0)
2589 while ((cd->ctypes[*ptr] & ctype_space) != 0) ptr++;
2590 if (*ptr == CHAR_NUMBER_SIGN)
2595 if (IS_NEWLINE(ptr)) { ptr += cd->nllen; break; }
2598 if (utf8) while ((*ptr & 0xc0) == 0x80) ptr++;
2606 /* If the next thing is itself optional, we have to give up. */
2608 if (*ptr == CHAR_ASTERISK || *ptr == CHAR_QUESTION_MARK ||
2609 strncmp((char *)ptr, STR_LEFT_CURLY_BRACKET STR_0 STR_COMMA, 3) == 0)
2612 /* Now compare the next item with the previous opcode. First, handle cases when
2613 the next item is a character. */
2615 if (next >= 0) switch(op_code)
2619 GETCHARTEST(c, previous);
2625 /* For CHARNC (caseless character) we must check the other case. If we have
2626 Unicode property support, we can use it to test the other case of
2627 high-valued characters. */
2631 GETCHARTEST(c, previous);
2635 if (c == next) return FALSE;
2639 unsigned int othercase;
2640 if (next < 128) othercase = cd->fcc[next]; else
2642 othercase = UCD_OTHERCASE((unsigned int)next);
2644 othercase = NOTACHAR;
2646 return (unsigned int)c != othercase;
2649 #endif /* SUPPORT_UTF8 */
2650 return (c != cd->fcc[next]); /* Non-UTF-8 mode */
2652 /* For OP_NOT, its data is always a single-byte character. */
2655 if ((c = *previous) == next) return TRUE;
2656 if ((options & PCRE_CASELESS) == 0) return FALSE;
2660 unsigned int othercase;
2661 if (next < 128) othercase = cd->fcc[next]; else
2663 othercase = UCD_OTHERCASE(next);
2665 othercase = NOTACHAR;
2667 return (unsigned int)c == othercase;
2670 #endif /* SUPPORT_UTF8 */
2671 return (c == cd->fcc[next]); /* Non-UTF-8 mode */
2673 /* Note that OP_DIGIT etc. are generated only when PCRE_UCP is *not* set.
2674 When it is set, \d etc. are converted into OP_(NOT_)PROP codes. */
2677 return next > 127 || (cd->ctypes[next] & ctype_digit) == 0;
2680 return next <= 127 && (cd->ctypes[next] & ctype_digit) != 0;
2683 return next > 127 || (cd->ctypes[next] & ctype_space) == 0;
2685 case OP_NOT_WHITESPACE:
2686 return next <= 127 && (cd->ctypes[next] & ctype_space) != 0;
2689 return next > 127 || (cd->ctypes[next] & ctype_word) == 0;
2691 case OP_NOT_WORDCHAR:
2692 return next <= 127 && (cd->ctypes[next] & ctype_word) != 0;
2717 return op_code == OP_NOT_HSPACE;
2719 return op_code != OP_NOT_HSPACE;
2734 return op_code == OP_NOT_VSPACE;
2736 return op_code != OP_NOT_VSPACE;
2741 return check_char_prop(next, previous[0], previous[1], FALSE);
2744 return check_char_prop(next, previous[0], previous[1], TRUE);
2752 /* Handle the case when the next item is \d, \s, etc. Note that when PCRE_UCP
2753 is set, \d turns into ESC_du rather than ESC_d, etc., so ESC_d etc. are
2754 generated only when PCRE_UCP is *not* set, that is, when only ASCII
2755 characteristics are recognized. Similarly, the opcodes OP_DIGIT etc. are
2756 replaced by OP_PROP codes when PCRE_UCP is set. */
2763 GETCHARTEST(c, previous);
2770 return c > 127 || (cd->ctypes[c] & ctype_digit) == 0;
2773 return c <= 127 && (cd->ctypes[c] & ctype_digit) != 0;
2776 return c > 127 || (cd->ctypes[c] & ctype_space) == 0;
2779 return c <= 127 && (cd->ctypes[c] & ctype_space) != 0;
2782 return c > 127 || (cd->ctypes[c] & ctype_word) == 0;
2785 return c <= 127 && (cd->ctypes[c] & ctype_word) != 0;
2810 return -next != ESC_h;
2812 return -next == ESC_h;
2826 return -next != ESC_v;
2828 return -next == ESC_v;
2831 /* When PCRE_UCP is set, these values get generated for \d etc. Find
2832 their substitutions and process them. The result will always be either
2833 -ESC_p or -ESC_P. Then fall through to process those values. */
2843 int temperrorcode = 0;
2844 ptr = substitutes[-next - ESC_DU];
2845 next = check_escape(&ptr, &temperrorcode, 0, options, FALSE);
2846 if (temperrorcode != 0) return FALSE;
2847 ptr++; /* For compatibility */
2854 int ptype, pdata, errorcodeptr;
2857 ptr--; /* Make ptr point at the p or P */
2858 ptype = get_ucp(&ptr, &negated, &pdata, &errorcodeptr);
2859 if (ptype < 0) return FALSE;
2860 ptr++; /* Point past the final curly ket */
2862 /* If the property item is optional, we have to give up. (When generated
2863 from \d etc by PCRE_UCP, this test will have been applied much earlier,
2864 to the original \d etc. At this point, ptr will point to a zero byte. */
2866 if (*ptr == CHAR_ASTERISK || *ptr == CHAR_QUESTION_MARK ||
2867 strncmp((char *)ptr, STR_LEFT_CURLY_BRACKET STR_0 STR_COMMA, 3) == 0)
2870 /* Do the property check. */
2872 return check_char_prop(c, ptype, pdata, (next == -ESC_P) != negated);
2880 /* In principle, support for Unicode properties should be integrated here as
2881 well. It means re-organizing the above code so as to get hold of the property
2882 values before switching on the op-code. However, I wonder how many patterns
2883 combine ASCII \d etc with Unicode properties? (Note that if PCRE_UCP is set,
2884 these op-codes are never generated.) */
2887 return next == -ESC_D || next == -ESC_s || next == -ESC_W ||
2888 next == -ESC_h || next == -ESC_v || next == -ESC_R;
2891 return next == -ESC_d;
2894 return next == -ESC_S || next == -ESC_d || next == -ESC_w || next == -ESC_R;
2896 case OP_NOT_WHITESPACE:
2897 return next == -ESC_s || next == -ESC_h || next == -ESC_v;
2900 return next == -ESC_S || next == -ESC_H || next == -ESC_d ||
2901 next == -ESC_w || next == -ESC_v || next == -ESC_R;
2904 return next == -ESC_h;
2906 /* Can't have \S in here because VT matches \S (Perl anomaly) */
2909 return next == -ESC_V || next == -ESC_d || next == -ESC_w;
2912 return next == -ESC_v || next == -ESC_R;
2915 return next == -ESC_W || next == -ESC_s || next == -ESC_h ||
2916 next == -ESC_v || next == -ESC_R;
2918 case OP_NOT_WORDCHAR:
2919 return next == -ESC_w || next == -ESC_d;
2925 /* Control does not reach here */
2930 /*************************************************
2931 * Compile one branch *
2932 *************************************************/
2934 /* Scan the pattern, compiling it into the a vector. If the options are
2935 changed during the branch, the pointer is used to change the external options
2936 bits. This function is used during the pre-compile phase when we are trying
2937 to find out the amount of memory needed, as well as during the real compile
2938 phase. The value of lengthptr distinguishes the two phases.
2941 optionsptr pointer to the option bits
2942 codeptr points to the pointer to the current code point
2943 ptrptr points to the current pattern pointer
2944 errorcodeptr points to error code variable
2945 firstbyteptr set to initial literal character, or < 0 (REQ_UNSET, REQ_NONE)
2946 reqbyteptr set to the last literal character required, else < 0
2947 bcptr points to current branch chain
2948 cd contains pointers to tables etc.
2949 lengthptr NULL during the real compile phase
2950 points to length accumulator during pre-compile phase
2952 Returns: TRUE on success
2953 FALSE, with *errorcodeptr set non-zero on error
2957 compile_branch(int *optionsptr, uschar **codeptr, const uschar **ptrptr,
2958 int *errorcodeptr, int *firstbyteptr, int *reqbyteptr, branch_chain *bcptr,
2959 compile_data *cd, int *lengthptr)
2961 int repeat_type, op_type;
2962 int repeat_min = 0, repeat_max = 0; /* To please picky compilers */
2964 int greedy_default, greedy_non_default;
2965 int firstbyte, reqbyte;
2966 int zeroreqbyte, zerofirstbyte;
2967 int req_caseopt, reqvary, tempreqvary;
2968 int options = *optionsptr;
2969 int after_manual_callout = 0;
2970 int length_prevgroup = 0;
2972 register uschar *code = *codeptr;
2973 uschar *last_code = code;
2974 uschar *orig_code = code;
2976 BOOL inescq = FALSE;
2977 BOOL groupsetfirstbyte = FALSE;
2978 const uschar *ptr = *ptrptr;
2979 const uschar *tempptr;
2980 const uschar *nestptr = NULL;
2981 uschar *previous = NULL;
2982 uschar *previous_callout = NULL;
2983 uschar *save_hwm = NULL;
2984 uschar classbits[32];
2988 BOOL utf8 = (options & PCRE_UTF8) != 0;
2989 uschar *class_utf8data;
2990 uschar *class_utf8data_base;
2991 uschar utf8_char[6];
2994 uschar *utf8_char = NULL;
2998 if (lengthptr != NULL) DPRINTF((">> start branch\n"));
3001 /* Set up the default and non-default settings for greediness */
3003 greedy_default = ((options & PCRE_UNGREEDY) != 0);
3004 greedy_non_default = greedy_default ^ 1;
3006 /* Initialize no first byte, no required byte. REQ_UNSET means "no char
3007 matching encountered yet". It gets changed to REQ_NONE if we hit something that
3008 matches a non-fixed char first char; reqbyte just remains unset if we never
3011 When we hit a repeat whose minimum is zero, we may have to adjust these values
3012 to take the zero repeat into account. This is implemented by setting them to
3013 zerofirstbyte and zeroreqbyte when such a repeat is encountered. The individual
3014 item types that can be repeated set these backoff variables appropriately. */
3016 firstbyte = reqbyte = zerofirstbyte = zeroreqbyte = REQ_UNSET;
3018 /* The variable req_caseopt contains either the REQ_CASELESS value or zero,
3019 according to the current setting of the caseless flag. REQ_CASELESS is a bit
3020 value > 255. It is added into the firstbyte or reqbyte variables to record the
3021 case status of the value. This is used only for ASCII characters. */
3023 req_caseopt = ((options & PCRE_CASELESS) != 0)? REQ_CASELESS : 0;
3025 /* Switch on next character until the end of the branch */
3030 BOOL should_flip_negation;
3031 BOOL possessive_quantifier;
3034 BOOL reset_bracount;
3035 int class_charcount;
3047 /* Get next byte in the pattern */
3051 /* If we are at the end of a nested substitution, revert to the outer level
3052 string. Nesting only happens one level deep. */
3054 if (c == 0 && nestptr != NULL)
3061 /* If we are in the pre-compile phase, accumulate the length used for the
3062 previous cycle of this loop. */
3064 if (lengthptr != NULL)
3067 if (code > cd->hwm) cd->hwm = code; /* High water info */
3069 if (code > cd->start_workspace + WORK_SIZE_CHECK) /* Check for overrun */
3071 *errorcodeptr = ERR52;
3075 /* There is at least one situation where code goes backwards: this is the
3076 case of a zero quantifier after a class (e.g. [ab]{0}). At compile time,
3077 the class is simply eliminated. However, it is created first, so we have to
3078 allow memory for it. Therefore, don't ever reduce the length at this point.
3081 if (code < last_code) code = last_code;
3083 /* Paranoid check for integer overflow */
3085 if (OFLOW_MAX - *lengthptr < code - last_code)
3087 *errorcodeptr = ERR20;
3091 *lengthptr += (int)(code - last_code);
3092 DPRINTF(("length=%d added %d c=%c\n", *lengthptr, code - last_code, c));
3094 /* If "previous" is set and it is not at the start of the work space, move
3095 it back to there, in order to avoid filling up the work space. Otherwise,
3096 if "previous" is NULL, reset the current code pointer to the start. */
3098 if (previous != NULL)
3100 if (previous > orig_code)
3102 memmove(orig_code, previous, code - previous);
3103 code -= previous - orig_code;
3104 previous = orig_code;
3107 else code = orig_code;
3109 /* Remember where this code item starts so we can pick up the length
3115 /* In the real compile phase, just check the workspace used by the forward
3118 else if (cd->hwm > cd->start_workspace + WORK_SIZE_CHECK)
3120 *errorcodeptr = ERR52;
3124 /* If in \Q...\E, check for the end; if not, we have a literal */
3126 if (inescq && c != 0)
3128 if (c == CHAR_BACKSLASH && ptr[1] == CHAR_E)
3136 if (previous_callout != NULL)
3138 if (lengthptr == NULL) /* Don't attempt in pre-compile phase */
3139 complete_callout(previous_callout, ptr, cd);
3140 previous_callout = NULL;
3142 if ((options & PCRE_AUTO_CALLOUT) != 0)
3144 previous_callout = code;
3145 code = auto_callout(code, ptr, cd);
3151 /* Fill in length of a previous callout, except when the next thing is
3155 c == CHAR_ASTERISK || c == CHAR_PLUS || c == CHAR_QUESTION_MARK ||
3156 (c == CHAR_LEFT_CURLY_BRACKET && is_counted_repeat(ptr+1));
3158 if (!is_quantifier && previous_callout != NULL &&
3159 after_manual_callout-- <= 0)
3161 if (lengthptr == NULL) /* Don't attempt in pre-compile phase */
3162 complete_callout(previous_callout, ptr, cd);
3163 previous_callout = NULL;
3166 /* In extended mode, skip white space and comments */
3168 if ((options & PCRE_EXTENDED) != 0)
3170 if ((cd->ctypes[c] & ctype_space) != 0) continue;
3171 if (c == CHAR_NUMBER_SIGN)
3176 if (IS_NEWLINE(ptr)) { ptr += cd->nllen - 1; break; }
3179 if (utf8) while ((*ptr & 0xc0) == 0x80) ptr++;
3182 if (*ptr != 0) continue;
3184 /* Else fall through to handle end of string */
3189 /* No auto callout for quantifiers. */
3191 if ((options & PCRE_AUTO_CALLOUT) != 0 && !is_quantifier)
3193 previous_callout = code;
3194 code = auto_callout(code, ptr, cd);
3199 /* ===================================================================*/
3200 case 0: /* The branch terminates at string end */
3201 case CHAR_VERTICAL_LINE: /* or | or ) */
3202 case CHAR_RIGHT_PARENTHESIS:
3203 *firstbyteptr = firstbyte;
3204 *reqbyteptr = reqbyte;
3207 if (lengthptr != NULL)
3209 if (OFLOW_MAX - *lengthptr < code - last_code)
3211 *errorcodeptr = ERR20;
3214 *lengthptr += (int)(code - last_code); /* To include callout length */
3215 DPRINTF((">> end branch\n"));
3220 /* ===================================================================*/
3221 /* Handle single-character metacharacters. In multiline mode, ^ disables
3222 the setting of any following char as a first character. */
3224 case CHAR_CIRCUMFLEX_ACCENT:
3225 if ((options & PCRE_MULTILINE) != 0)
3227 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
3233 case CHAR_DOLLAR_SIGN:
3238 /* There can never be a first char if '.' is first, whatever happens about
3239 repeats. The value of reqbyte doesn't change either. */
3242 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
3243 zerofirstbyte = firstbyte;
3244 zeroreqbyte = reqbyte;
3246 *code++ = ((options & PCRE_DOTALL) != 0)? OP_ALLANY: OP_ANY;
3250 /* ===================================================================*/
3251 /* Character classes. If the included characters are all < 256, we build a
3252 32-byte bitmap of the permitted characters, except in the special case
3253 where there is only one such character. For negated classes, we build the
3254 map as usual, then invert it at the end. However, we use a different opcode
3255 so that data characters > 255 can be handled correctly.
3257 If the class contains characters outside the 0-255 range, a different
3258 opcode is compiled. It may optionally have a bit map for characters < 256,
3259 but those above are are explicitly listed afterwards. A flag byte tells
3260 whether the bitmap is present, and whether this is a negated class or not.
3262 In JavaScript compatibility mode, an isolated ']' causes an error. In
3263 default (Perl) mode, it is treated as a data character. */
3265 case CHAR_RIGHT_SQUARE_BRACKET:
3266 if ((cd->external_options & PCRE_JAVASCRIPT_COMPAT) != 0)
3268 *errorcodeptr = ERR64;
3273 case CHAR_LEFT_SQUARE_BRACKET:
3276 /* PCRE supports POSIX class stuff inside a class. Perl gives an error if
3277 they are encountered at the top level, so we'll do that too. */
3279 if ((ptr[1] == CHAR_COLON || ptr[1] == CHAR_DOT ||
3280 ptr[1] == CHAR_EQUALS_SIGN) &&
3281 check_posix_syntax(ptr, &tempptr))
3283 *errorcodeptr = (ptr[1] == CHAR_COLON)? ERR13 : ERR31;
3287 /* If the first character is '^', set the negation flag and skip it. Also,
3288 if the first few characters (either before or after ^) are \Q\E or \E we
3289 skip them too. This makes for compatibility with Perl. */
3291 negate_class = FALSE;
3295 if (c == CHAR_BACKSLASH)
3297 if (ptr[1] == CHAR_E)
3299 else if (strncmp((const char *)ptr+1,
3300 STR_Q STR_BACKSLASH STR_E, 3) == 0)
3305 else if (!negate_class && c == CHAR_CIRCUMFLEX_ACCENT)
3306 negate_class = TRUE;
3310 /* Empty classes are allowed in JavaScript compatibility mode. Otherwise,
3311 an initial ']' is taken as a data character -- the code below handles
3312 that. In JS mode, [] must always fail, so generate OP_FAIL, whereas
3313 [^] must match any character, so generate OP_ALLANY. */
3315 if (c == CHAR_RIGHT_SQUARE_BRACKET &&
3316 (cd->external_options & PCRE_JAVASCRIPT_COMPAT) != 0)
3318 *code++ = negate_class? OP_ALLANY : OP_FAIL;
3319 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
3320 zerofirstbyte = firstbyte;
3324 /* If a class contains a negative special such as \S, we need to flip the
3325 negation flag at the end, so that support for characters > 255 works
3326 correctly (they are all included in the class). */
3328 should_flip_negation = FALSE;
3330 /* Keep a count of chars with values < 256 so that we can optimize the case
3331 of just a single character (as long as it's < 256). However, For higher
3332 valued UTF-8 characters, we don't yet do any optimization. */
3334 class_charcount = 0;
3335 class_lastchar = -1;
3337 /* Initialize the 32-char bit map to all zeros. We build the map in a
3338 temporary bit of memory, in case the class contains only 1 character (less
3339 than 256), because in that case the compiled code doesn't use the bit map.
3342 memset(classbits, 0, 32 * sizeof(uschar));
3345 class_utf8 = FALSE; /* No chars >= 256 */
3346 class_utf8data = code + LINK_SIZE + 2; /* For UTF-8 items */
3347 class_utf8data_base = class_utf8data; /* For resetting in pass 1 */
3350 /* Process characters until ] is reached. By writing this as a "do" it
3351 means that an initial ] is taken as a data character. At the start of the
3352 loop, c contains the first byte of the character. */
3356 const uschar *oldptr;
3359 if (utf8 && c > 127)
3360 { /* Braces are required because the */
3361 GETCHARLEN(c, ptr, ptr); /* macro generates multiple statements */
3364 /* In the pre-compile phase, accumulate the length of any UTF-8 extra
3365 data and reset the pointer. This is so that very large classes that
3366 contain a zillion UTF-8 characters no longer overwrite the work space
3367 (which is on the stack). */
3369 if (lengthptr != NULL)
3371 *lengthptr += class_utf8data - class_utf8data_base;
3372 class_utf8data = class_utf8data_base;
3377 /* Inside \Q...\E everything is literal except \E */
3381 if (c == CHAR_BACKSLASH && ptr[1] == CHAR_E) /* If we are at \E */
3383 inescq = FALSE; /* Reset literal state */
3384 ptr++; /* Skip the 'E' */
3385 continue; /* Carry on with next */
3387 goto CHECK_RANGE; /* Could be range if \E follows */
3390 /* Handle POSIX class names. Perl allows a negation extension of the
3391 form [:^name:]. A square bracket that doesn't match the syntax is
3392 treated as a literal. We also recognize the POSIX constructions
3393 [.ch.] and [=ch=] ("collating elements") and fault them, as Perl
3396 if (c == CHAR_LEFT_SQUARE_BRACKET &&
3397 (ptr[1] == CHAR_COLON || ptr[1] == CHAR_DOT ||
3398 ptr[1] == CHAR_EQUALS_SIGN) && check_posix_syntax(ptr, &tempptr))
3400 BOOL local_negate = FALSE;
3401 int posix_class, taboffset, tabopt;
3402 register const uschar *cbits = cd->cbits;
3405 if (ptr[1] != CHAR_COLON)
3407 *errorcodeptr = ERR31;
3412 if (*ptr == CHAR_CIRCUMFLEX_ACCENT)
3414 local_negate = TRUE;
3415 should_flip_negation = TRUE; /* Note negative special */
3419 posix_class = check_posix_name(ptr, (int)(tempptr - ptr));
3420 if (posix_class < 0)
3422 *errorcodeptr = ERR30;
3426 /* If matching is caseless, upper and lower are converted to
3427 alpha. This relies on the fact that the class table starts with
3428 alpha, lower, upper as the first 3 entries. */
3430 if ((options & PCRE_CASELESS) != 0 && posix_class <= 2)
3433 /* When PCRE_UCP is set, some of the POSIX classes are converted to
3434 different escape sequences that use Unicode properties. */
3437 if ((options & PCRE_UCP) != 0)
3439 int pc = posix_class + ((local_negate)? POSIX_SUBSIZE/2 : 0);
3440 if (posix_substitutes[pc] != NULL)
3442 nestptr = tempptr + 1;
3443 ptr = posix_substitutes[pc] - 1;
3448 /* In the non-UCP case, we build the bit map for the POSIX class in a
3449 chunk of local store because we may be adding and subtracting from it,
3450 and we don't want to subtract bits that may be in the main map already.
3451 At the end we or the result into the bit map that is being built. */
3455 /* Copy in the first table (always present) */
3457 memcpy(pbits, cbits + posix_class_maps[posix_class],
3458 32 * sizeof(uschar));
3460 /* If there is a second table, add or remove it as required. */
3462 taboffset = posix_class_maps[posix_class + 1];
3463 tabopt = posix_class_maps[posix_class + 2];
3468 for (c = 0; c < 32; c++) pbits[c] |= cbits[c + taboffset];
3470 for (c = 0; c < 32; c++) pbits[c] &= ~cbits[c + taboffset];
3473 /* Not see if we need to remove any special characters. An option
3474 value of 1 removes vertical space and 2 removes underscore. */
3476 if (tabopt < 0) tabopt = -tabopt;
3477 if (tabopt == 1) pbits[1] &= ~0x3c;
3478 else if (tabopt == 2) pbits[11] &= 0x7f;
3480 /* Add the POSIX table or its complement into the main table that is
3481 being built and we are done. */
3484 for (c = 0; c < 32; c++) classbits[c] |= ~pbits[c];
3486 for (c = 0; c < 32; c++) classbits[c] |= pbits[c];
3489 class_charcount = 10; /* Set > 1; assumes more than 1 per class */
3490 continue; /* End of POSIX syntax handling */
3493 /* Backslash may introduce a single character, or it may introduce one
3494 of the specials, which just set a flag. The sequence \b is a special
3495 case. Inside a class (and only there) it is treated as backspace. We
3496 assume that other escapes have more than one character in them, so set
3497 class_charcount bigger than one. Unrecognized escapes fall through and
3498 are either treated as literal characters (by default), or are faulted if
3499 PCRE_EXTRA is set. */
3501 if (c == CHAR_BACKSLASH)
3503 c = check_escape(&ptr, errorcodeptr, cd->bracount, options, TRUE);
3504 if (*errorcodeptr != 0) goto FAILED;
3506 if (-c == ESC_b) c = CHAR_BS; /* \b is backspace in a class */
3507 else if (-c == ESC_Q) /* Handle start of quoted string */
3509 if (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
3511 ptr += 2; /* avoid empty string */
3516 else if (-c == ESC_E) continue; /* Ignore orphan \E */
3520 register const uschar *cbits = cd->cbits;
3521 class_charcount += 2; /* Greater than 1 is what matters */
3526 case ESC_du: /* These are the values given for \d etc */
3527 case ESC_DU: /* when PCRE_UCP is set. We replace the */
3528 case ESC_wu: /* escape sequence with an appropriate \p */
3529 case ESC_WU: /* or \P to test Unicode properties instead */
3530 case ESC_su: /* of the default ASCII testing. */
3533 ptr = substitutes[-c - ESC_DU] - 1; /* Just before substitute */
3534 class_charcount -= 2; /* Undo! */
3538 for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_digit];
3542 should_flip_negation = TRUE;
3543 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_digit];
3547 for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_word];
3551 should_flip_negation = TRUE;
3552 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_word];
3555 /* Perl 5.004 onwards omits VT from \s, but we must preserve it
3556 if it was previously set by something earlier in the character
3560 classbits[0] |= cbits[cbit_space];
3561 classbits[1] |= cbits[cbit_space+1] & ~0x08;
3562 for (c = 2; c < 32; c++) classbits[c] |= cbits[c+cbit_space];
3566 should_flip_negation = TRUE;
3567 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_space];
3568 classbits[1] |= 0x08; /* Perl 5.004 onwards omits VT from \s */
3572 SETBIT(classbits, 0x09); /* VT */
3573 SETBIT(classbits, 0x20); /* SPACE */
3574 SETBIT(classbits, 0xa0); /* NSBP */
3579 *class_utf8data++ = XCL_SINGLE;
3580 class_utf8data += _pcre_ord2utf8(0x1680, class_utf8data);
3581 *class_utf8data++ = XCL_SINGLE;
3582 class_utf8data += _pcre_ord2utf8(0x180e, class_utf8data);
3583 *class_utf8data++ = XCL_RANGE;
3584 class_utf8data += _pcre_ord2utf8(0x2000, class_utf8data);
3585 class_utf8data += _pcre_ord2utf8(0x200A, class_utf8data);
3586 *class_utf8data++ = XCL_SINGLE;
3587 class_utf8data += _pcre_ord2utf8(0x202f, class_utf8data);
3588 *class_utf8data++ = XCL_SINGLE;
3589 class_utf8data += _pcre_ord2utf8(0x205f, class_utf8data);
3590 *class_utf8data++ = XCL_SINGLE;
3591 class_utf8data += _pcre_ord2utf8(0x3000, class_utf8data);
3597 for (c = 0; c < 32; c++)
3602 case 0x09/8: x ^= 1 << (0x09%8); break;
3603 case 0x20/8: x ^= 1 << (0x20%8); break;
3604 case 0xa0/8: x ^= 1 << (0xa0%8); break;
3614 *class_utf8data++ = XCL_RANGE;
3615 class_utf8data += _pcre_ord2utf8(0x0100, class_utf8data);
3616 class_utf8data += _pcre_ord2utf8(0x167f, class_utf8data);
3617 *class_utf8data++ = XCL_RANGE;
3618 class_utf8data += _pcre_ord2utf8(0x1681, class_utf8data);
3619 class_utf8data += _pcre_ord2utf8(0x180d, class_utf8data);
3620 *class_utf8data++ = XCL_RANGE;
3621 class_utf8data += _pcre_ord2utf8(0x180f, class_utf8data);
3622 class_utf8data += _pcre_ord2utf8(0x1fff, class_utf8data);
3623 *class_utf8data++ = XCL_RANGE;
3624 class_utf8data += _pcre_ord2utf8(0x200B, class_utf8data);
3625 class_utf8data += _pcre_ord2utf8(0x202e, class_utf8data);
3626 *class_utf8data++ = XCL_RANGE;
3627 class_utf8data += _pcre_ord2utf8(0x2030, class_utf8data);
3628 class_utf8data += _pcre_ord2utf8(0x205e, class_utf8data);
3629 *class_utf8data++ = XCL_RANGE;
3630 class_utf8data += _pcre_ord2utf8(0x2060, class_utf8data);
3631 class_utf8data += _pcre_ord2utf8(0x2fff, class_utf8data);
3632 *class_utf8data++ = XCL_RANGE;
3633 class_utf8data += _pcre_ord2utf8(0x3001, class_utf8data);
3634 class_utf8data += _pcre_ord2utf8(0x7fffffff, class_utf8data);
3640 SETBIT(classbits, 0x0a); /* LF */
3641 SETBIT(classbits, 0x0b); /* VT */
3642 SETBIT(classbits, 0x0c); /* FF */
3643 SETBIT(classbits, 0x0d); /* CR */
3644 SETBIT(classbits, 0x85); /* NEL */
3649 *class_utf8data++ = XCL_RANGE;
3650 class_utf8data += _pcre_ord2utf8(0x2028, class_utf8data);
3651 class_utf8data += _pcre_ord2utf8(0x2029, class_utf8data);
3657 for (c = 0; c < 32; c++)
3662 case 0x0a/8: x ^= 1 << (0x0a%8);
3667 case 0x85/8: x ^= 1 << (0x85%8); break;
3677 *class_utf8data++ = XCL_RANGE;
3678 class_utf8data += _pcre_ord2utf8(0x0100, class_utf8data);
3679 class_utf8data += _pcre_ord2utf8(0x2027, class_utf8data);
3680 *class_utf8data++ = XCL_RANGE;
3681 class_utf8data += _pcre_ord2utf8(0x2029, class_utf8data);
3682 class_utf8data += _pcre_ord2utf8(0x7fffffff, class_utf8data);
3693 int ptype = get_ucp(&ptr, &negated, &pdata, errorcodeptr);
3694 if (ptype < 0) goto FAILED;
3696 *class_utf8data++ = ((-c == ESC_p) != negated)?
3697 XCL_PROP : XCL_NOTPROP;
3698 *class_utf8data++ = ptype;
3699 *class_utf8data++ = pdata;
3700 class_charcount -= 2; /* Not a < 256 character */
3704 /* Unrecognized escapes are faulted if PCRE is running in its
3705 strict mode. By default, for compatibility with Perl, they are
3706 treated as literals. */
3709 if ((options & PCRE_EXTRA) != 0)
3711 *errorcodeptr = ERR7;
3714 class_charcount -= 2; /* Undo the default count from above */
3715 c = *ptr; /* Get the final character and fall through */
3720 /* Fall through if we have a single character (c >= 0). This may be
3721 greater than 256 in UTF-8 mode. */
3723 } /* End of backslash handling */
3725 /* A single character may be followed by '-' to form a range. However,
3726 Perl does not permit ']' to be the end of the range. A '-' character
3727 at the end is treated as a literal. Perl ignores orphaned \E sequences
3728 entirely. The code for handling \Q and \E is messy. */
3731 while (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
3739 /* Remember \r or \n */
3741 if (c == CHAR_CR || c == CHAR_NL) cd->external_flags |= PCRE_HASCRORLF;
3743 /* Check for range */
3745 if (!inescq && ptr[1] == CHAR_MINUS)
3749 while (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_E) ptr += 2;
3751 /* If we hit \Q (not followed by \E) at this point, go into escaped
3754 while (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_Q)
3757 if (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_E)
3758 { ptr += 2; continue; }
3763 if (*ptr == 0 || (!inescq && *ptr == CHAR_RIGHT_SQUARE_BRACKET))
3766 goto LONE_SINGLE_CHARACTER;
3771 { /* Braces are required because the */
3772 GETCHARLEN(d, ptr, ptr); /* macro generates multiple statements */
3776 d = *ptr; /* Not UTF-8 mode */
3778 /* The second part of a range can be a single-character escape, but
3779 not any of the other escapes. Perl 5.6 treats a hyphen as a literal
3780 in such circumstances. */
3782 if (!inescq && d == CHAR_BACKSLASH)
3784 d = check_escape(&ptr, errorcodeptr, cd->bracount, options, TRUE);
3785 if (*errorcodeptr != 0) goto FAILED;
3787 /* \b is backspace; any other special means the '-' was literal */
3791 if (d == -ESC_b) d = CHAR_BS; else
3794 goto LONE_SINGLE_CHARACTER; /* A few lines below */
3799 /* Check that the two values are in the correct order. Optimize
3800 one-character ranges */
3804 *errorcodeptr = ERR8;
3808 if (d == c) goto LONE_SINGLE_CHARACTER; /* A few lines below */
3810 /* Remember \r or \n */
3812 if (d == CHAR_CR || d == CHAR_NL) cd->external_flags |= PCRE_HASCRORLF;
3814 /* In UTF-8 mode, if the upper limit is > 255, or > 127 for caseless
3815 matching, we have to use an XCLASS with extra data items. Caseless
3816 matching for characters > 127 is available only if UCP support is
3820 if (utf8 && (d > 255 || ((options & PCRE_CASELESS) != 0 && d > 127)))
3824 /* With UCP support, we can find the other case equivalents of
3825 the relevant characters. There may be several ranges. Optimize how
3826 they fit with the basic range. */
3829 if ((options & PCRE_CASELESS) != 0)
3831 unsigned int occ, ocd;
3832 unsigned int cc = c;
3833 unsigned int origd = d;
3834 while (get_othercase_range(&cc, origd, &occ, &ocd))
3836 if (occ >= (unsigned int)c &&
3837 ocd <= (unsigned int)d)
3838 continue; /* Skip embedded ranges */
3840 if (occ < (unsigned int)c &&
3841 ocd >= (unsigned int)c - 1) /* Extend the basic range */
3842 { /* if there is overlap, */
3843 c = occ; /* noting that if occ < c */
3844 continue; /* we can't have ocd > d */
3845 } /* because a subrange is */
3846 if (ocd > (unsigned int)d &&
3847 occ <= (unsigned int)d + 1) /* always shorter than */
3848 { /* the basic range. */
3855 *class_utf8data++ = XCL_SINGLE;
3859 *class_utf8data++ = XCL_RANGE;
3860 class_utf8data += _pcre_ord2utf8(occ, class_utf8data);
3862 class_utf8data += _pcre_ord2utf8(ocd, class_utf8data);
3865 #endif /* SUPPORT_UCP */
3867 /* Now record the original range, possibly modified for UCP caseless
3868 overlapping ranges. */
3870 *class_utf8data++ = XCL_RANGE;
3871 class_utf8data += _pcre_ord2utf8(c, class_utf8data);
3872 class_utf8data += _pcre_ord2utf8(d, class_utf8data);
3874 /* With UCP support, we are done. Without UCP support, there is no
3875 caseless matching for UTF-8 characters > 127; we can use the bit map
3876 for the smaller ones. */
3879 continue; /* With next character in the class */
3881 if ((options & PCRE_CASELESS) == 0 || c > 127) continue;
3883 /* Adjust upper limit and fall through to set up the map */
3887 #endif /* SUPPORT_UCP */
3889 #endif /* SUPPORT_UTF8 */
3891 /* We use the bit map for all cases when not in UTF-8 mode; else
3892 ranges that lie entirely within 0-127 when there is UCP support; else
3893 for partial ranges without UCP support. */
3895 class_charcount += d - c + 1;
3898 /* We can save a bit of time by skipping this in the pre-compile. */
3900 if (lengthptr == NULL) for (; c <= d; c++)
3902 classbits[c/8] |= (1 << (c&7));
3903 if ((options & PCRE_CASELESS) != 0)
3905 int uc = cd->fcc[c]; /* flip case */
3906 classbits[uc/8] |= (1 << (uc&7));
3910 continue; /* Go get the next char in the class */
3913 /* Handle a lone single character - we can get here for a normal
3914 non-escape char, or after \ that introduces a single character or for an
3915 apparent range that isn't. */
3917 LONE_SINGLE_CHARACTER:
3919 /* Handle a character that cannot go in the bit map */
3922 if (utf8 && (c > 255 || ((options & PCRE_CASELESS) != 0 && c > 127)))
3925 *class_utf8data++ = XCL_SINGLE;
3926 class_utf8data += _pcre_ord2utf8(c, class_utf8data);
3929 if ((options & PCRE_CASELESS) != 0)
3931 unsigned int othercase;
3932 if ((othercase = UCD_OTHERCASE(c)) != c)
3934 *class_utf8data++ = XCL_SINGLE;
3935 class_utf8data += _pcre_ord2utf8(othercase, class_utf8data);
3938 #endif /* SUPPORT_UCP */
3942 #endif /* SUPPORT_UTF8 */
3944 /* Handle a single-byte character */
3946 classbits[c/8] |= (1 << (c&7));
3947 if ((options & PCRE_CASELESS) != 0)
3949 c = cd->fcc[c]; /* flip case */
3950 classbits[c/8] |= (1 << (c&7));
3957 /* Loop until ']' reached. This "while" is the end of the "do" far above.
3958 If we are at the end of an internal nested string, revert to the outer
3961 while (((c = *(++ptr)) != 0 ||
3963 (ptr = nestptr, nestptr = NULL, c = *(++ptr)) != 0)) &&
3964 (c != CHAR_RIGHT_SQUARE_BRACKET || inescq));
3966 /* Check for missing terminating ']' */
3970 *errorcodeptr = ERR6;
3974 /* If class_charcount is 1, we saw precisely one character whose value is
3975 less than 256. As long as there were no characters >= 128 and there was no
3976 use of \p or \P, in other words, no use of any XCLASS features, we can
3979 In UTF-8 mode, we can optimize the negative case only if there were no
3980 characters >= 128 because OP_NOT and the related opcodes like OP_NOTSTAR
3981 operate on single-bytes only. This is an historical hangover. Maybe one day
3982 we can tidy these opcodes to handle multi-byte characters.
3984 The optimization throws away the bit map. We turn the item into a
3985 1-character OP_CHAR[NC] if it's positive, or OP_NOT if it's negative. Note
3986 that OP_NOT does not support multibyte characters. In the positive case, it
3987 can cause firstbyte to be set. Otherwise, there can be no first char if
3988 this item is first, whatever repeat count may follow. In the case of
3989 reqbyte, save the previous value for reinstating. */
3992 if (class_charcount == 1 && !class_utf8 &&
3993 (!utf8 || !negate_class || class_lastchar < 128))
3995 if (class_charcount == 1)
3998 zeroreqbyte = reqbyte;
4000 /* The OP_NOT opcode works on one-byte characters only. */
4004 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
4005 zerofirstbyte = firstbyte;
4007 *code++ = class_lastchar;
4011 /* For a single, positive character, get the value into mcbuffer, and
4012 then we can handle this with the normal one-character code. */
4015 if (utf8 && class_lastchar > 127)
4016 mclength = _pcre_ord2utf8(class_lastchar, mcbuffer);
4020 mcbuffer[0] = class_lastchar;
4024 } /* End of 1-char optimization */
4026 /* The general case - not the one-char optimization. If this is the first
4027 thing in the branch, there can be no first char setting, whatever the
4028 repeat count. Any reqbyte setting must remain unchanged after any kind of
4031 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
4032 zerofirstbyte = firstbyte;
4033 zeroreqbyte = reqbyte;
4035 /* If there are characters with values > 255, we have to compile an
4036 extended class, with its own opcode, unless there was a negated special
4037 such as \S in the class, and PCRE_UCP is not set, because in that case all
4038 characters > 255 are in the class, so any that were explicitly given as
4039 well can be ignored. If (when there are explicit characters > 255 that must
4040 be listed) there are no characters < 256, we can omit the bitmap in the
4041 actual compiled code. */
4044 if (class_utf8 && (!should_flip_negation || (options & PCRE_UCP) != 0))
4046 *class_utf8data++ = XCL_END; /* Marks the end of extra data */
4047 *code++ = OP_XCLASS;
4049 *code = negate_class? XCL_NOT : 0;
4051 /* If the map is required, move up the extra data to make room for it;
4052 otherwise just move the code pointer to the end of the extra data. */
4054 if (class_charcount > 0)
4057 memmove(code + 32, code, class_utf8data - code);
4058 memcpy(code, classbits, 32);
4059 code = class_utf8data + 32;
4061 else code = class_utf8data;
4063 /* Now fill in the complete length of the item */
4065 PUT(previous, 1, code - previous);
4066 break; /* End of class handling */
4070 /* If there are no characters > 255, or they are all to be included or
4071 excluded, set the opcode to OP_CLASS or OP_NCLASS, depending on whether the
4072 whole class was negated and whether there were negative specials such as \S
4073 (non-UCP) in the class. Then copy the 32-byte map into the code vector,
4074 negating it if necessary. */
4076 *code++ = (negate_class == should_flip_negation) ? OP_CLASS : OP_NCLASS;
4079 if (lengthptr == NULL) /* Save time in the pre-compile phase */
4080 for (c = 0; c < 32; c++) code[c] = ~classbits[c];
4084 memcpy(code, classbits, 32);
4090 /* ===================================================================*/
4091 /* Various kinds of repeat; '{' is not necessarily a quantifier, but this
4092 has been tested above. */
4094 case CHAR_LEFT_CURLY_BRACKET:
4095 if (!is_quantifier) goto NORMAL_CHAR;
4096 ptr = read_repeat_counts(ptr+1, &repeat_min, &repeat_max, errorcodeptr);
4097 if (*errorcodeptr != 0) goto FAILED;
4110 case CHAR_QUESTION_MARK:
4115 if (previous == NULL)
4117 *errorcodeptr = ERR9;
4121 if (repeat_min == 0)
4123 firstbyte = zerofirstbyte; /* Adjust for zero repeat */
4124 reqbyte = zeroreqbyte; /* Ditto */
4127 /* Remember whether this is a variable length repeat */
4129 reqvary = (repeat_min == repeat_max)? 0 : REQ_VARY;
4131 op_type = 0; /* Default single-char op codes */
4132 possessive_quantifier = FALSE; /* Default not possessive quantifier */
4134 /* Save start of previous item, in case we have to move it up to make space
4135 for an inserted OP_ONCE for the additional '+' extension. */
4137 tempcode = previous;
4139 /* If the next character is '+', we have a possessive quantifier. This
4140 implies greediness, whatever the setting of the PCRE_UNGREEDY option.
4141 If the next character is '?' this is a minimizing repeat, by default,
4142 but if PCRE_UNGREEDY is set, it works the other way round. We change the
4143 repeat type to the non-default. */
4145 if (ptr[1] == CHAR_PLUS)
4147 repeat_type = 0; /* Force greedy */
4148 possessive_quantifier = TRUE;
4151 else if (ptr[1] == CHAR_QUESTION_MARK)
4153 repeat_type = greedy_non_default;
4156 else repeat_type = greedy_default;
4158 /* If previous was a character match, abolish the item and generate a
4159 repeat item instead. If a char item has a minumum of more than one, ensure
4160 that it is set in reqbyte - it might not be if a sequence such as x{3} is
4161 the first thing in a branch because the x will have gone into firstbyte
4164 if (*previous == OP_CHAR || *previous == OP_CHARNC)
4166 /* Deal with UTF-8 characters that take up more than one byte. It's
4167 easier to write this out separately than try to macrify it. Use c to
4168 hold the length of the character in bytes, plus 0x80 to flag that it's a
4169 length rather than a small character. */
4172 if (utf8 && (code[-1] & 0x80) != 0)
4174 uschar *lastchar = code - 1;
4175 while((*lastchar & 0xc0) == 0x80) lastchar--;
4176 c = code - lastchar; /* Length of UTF-8 character */
4177 memcpy(utf8_char, lastchar, c); /* Save the char */
4178 c |= 0x80; /* Flag c as a length */
4183 /* Handle the case of a single byte - either with no UTF8 support, or
4184 with UTF-8 disabled, or for a UTF-8 character < 128. */
4188 if (repeat_min > 1) reqbyte = c | req_caseopt | cd->req_varyopt;
4191 /* If the repetition is unlimited, it pays to see if the next thing on
4192 the line is something that cannot possibly match this character. If so,
4193 automatically possessifying this item gains some performance in the case
4194 where the match fails. */
4196 if (!possessive_quantifier &&
4198 check_auto_possessive(previous, utf8, ptr + 1, options, cd))
4200 repeat_type = 0; /* Force greedy */
4201 possessive_quantifier = TRUE;
4204 goto OUTPUT_SINGLE_REPEAT; /* Code shared with single character types */
4207 /* If previous was a single negated character ([^a] or similar), we use
4208 one of the special opcodes, replacing it. The code is shared with single-
4209 character repeats by setting opt_type to add a suitable offset into
4210 repeat_type. We can also test for auto-possessification. OP_NOT is
4211 currently used only for single-byte chars. */
4213 else if (*previous == OP_NOT)
4215 op_type = OP_NOTSTAR - OP_STAR; /* Use "not" opcodes */
4217 if (!possessive_quantifier &&
4219 check_auto_possessive(previous, utf8, ptr + 1, options, cd))
4221 repeat_type = 0; /* Force greedy */
4222 possessive_quantifier = TRUE;
4224 goto OUTPUT_SINGLE_REPEAT;
4227 /* If previous was a character type match (\d or similar), abolish it and
4228 create a suitable repeat item. The code is shared with single-character
4229 repeats by setting op_type to add a suitable offset into repeat_type. Note
4230 the the Unicode property types will be present only when SUPPORT_UCP is
4231 defined, but we don't wrap the little bits of code here because it just
4232 makes it horribly messy. */
4234 else if (*previous < OP_EODN)
4237 int prop_type, prop_value;
4238 op_type = OP_TYPESTAR - OP_STAR; /* Use type opcodes */
4241 if (!possessive_quantifier &&
4243 check_auto_possessive(previous, utf8, ptr + 1, options, cd))
4245 repeat_type = 0; /* Force greedy */
4246 possessive_quantifier = TRUE;
4249 OUTPUT_SINGLE_REPEAT:
4250 if (*previous == OP_PROP || *previous == OP_NOTPROP)
4252 prop_type = previous[1];
4253 prop_value = previous[2];
4255 else prop_type = prop_value = -1;
4258 code = previous; /* Usually overwrite previous item */
4260 /* If the maximum is zero then the minimum must also be zero; Perl allows
4261 this case, so we do too - by simply omitting the item altogether. */
4263 if (repeat_max == 0) goto END_REPEAT;
4265 /*--------------------------------------------------------------------*/
4266 /* This code is obsolete from release 8.00; the restriction was finally
4269 /* All real repeats make it impossible to handle partial matching (maybe
4270 one day we will be able to remove this restriction). */
4272 /* if (repeat_max != 1) cd->external_flags |= PCRE_NOPARTIAL; */
4273 /*--------------------------------------------------------------------*/
4275 /* Combine the op_type with the repeat_type */
4277 repeat_type += op_type;
4279 /* A minimum of zero is handled either as the special case * or ?, or as
4280 an UPTO, with the maximum given. */
4282 if (repeat_min == 0)
4284 if (repeat_max == -1) *code++ = OP_STAR + repeat_type;
4285 else if (repeat_max == 1) *code++ = OP_QUERY + repeat_type;
4288 *code++ = OP_UPTO + repeat_type;
4289 PUT2INC(code, 0, repeat_max);
4293 /* A repeat minimum of 1 is optimized into some special cases. If the
4294 maximum is unlimited, we use OP_PLUS. Otherwise, the original item is
4295 left in place and, if the maximum is greater than 1, we use OP_UPTO with
4296 one less than the maximum. */
4298 else if (repeat_min == 1)
4300 if (repeat_max == -1)
4301 *code++ = OP_PLUS + repeat_type;
4304 code = oldcode; /* leave previous item in place */
4305 if (repeat_max == 1) goto END_REPEAT;
4306 *code++ = OP_UPTO + repeat_type;
4307 PUT2INC(code, 0, repeat_max - 1);
4311 /* The case {n,n} is just an EXACT, while the general case {n,m} is
4312 handled as an EXACT followed by an UPTO. */
4316 *code++ = OP_EXACT + op_type; /* NB EXACT doesn't have repeat_type */
4317 PUT2INC(code, 0, repeat_min);
4319 /* If the maximum is unlimited, insert an OP_STAR. Before doing so,
4320 we have to insert the character for the previous code. For a repeated
4321 Unicode property match, there are two extra bytes that define the
4322 required property. In UTF-8 mode, long characters have their length in
4323 c, with the 0x80 bit as a flag. */
4328 if (utf8 && c >= 128)
4330 memcpy(code, utf8_char, c & 7);
4339 *code++ = prop_type;
4340 *code++ = prop_value;
4343 *code++ = OP_STAR + repeat_type;
4346 /* Else insert an UPTO if the max is greater than the min, again
4347 preceded by the character, for the previously inserted code. If the
4348 UPTO is just for 1 instance, we can use QUERY instead. */
4350 else if (repeat_max != repeat_min)
4353 if (utf8 && c >= 128)
4355 memcpy(code, utf8_char, c & 7);
4363 *code++ = prop_type;
4364 *code++ = prop_value;
4366 repeat_max -= repeat_min;
4368 if (repeat_max == 1)
4370 *code++ = OP_QUERY + repeat_type;
4374 *code++ = OP_UPTO + repeat_type;
4375 PUT2INC(code, 0, repeat_max);
4380 /* The character or character type itself comes last in all cases. */
4383 if (utf8 && c >= 128)
4385 memcpy(code, utf8_char, c & 7);
4392 /* For a repeated Unicode property match, there are two extra bytes that
4393 define the required property. */
4398 *code++ = prop_type;
4399 *code++ = prop_value;
4404 /* If previous was a character class or a back reference, we put the repeat
4405 stuff after it, but just skip the item if the repeat was {0,0}. */
4407 else if (*previous == OP_CLASS ||
4408 *previous == OP_NCLASS ||
4410 *previous == OP_XCLASS ||
4412 *previous == OP_REF)
4414 if (repeat_max == 0)
4420 /*--------------------------------------------------------------------*/
4421 /* This code is obsolete from release 8.00; the restriction was finally
4424 /* All real repeats make it impossible to handle partial matching (maybe
4425 one day we will be able to remove this restriction). */
4427 /* if (repeat_max != 1) cd->external_flags |= PCRE_NOPARTIAL; */
4428 /*--------------------------------------------------------------------*/
4430 if (repeat_min == 0 && repeat_max == -1)
4431 *code++ = OP_CRSTAR + repeat_type;
4432 else if (repeat_min == 1 && repeat_max == -1)
4433 *code++ = OP_CRPLUS + repeat_type;
4434 else if (repeat_min == 0 && repeat_max == 1)
4435 *code++ = OP_CRQUERY + repeat_type;
4438 *code++ = OP_CRRANGE + repeat_type;
4439 PUT2INC(code, 0, repeat_min);
4440 if (repeat_max == -1) repeat_max = 0; /* 2-byte encoding for max */
4441 PUT2INC(code, 0, repeat_max);
4445 /* If previous was a bracket group, we may have to replicate it in certain
4448 else if (*previous == OP_BRA || *previous == OP_CBRA ||
4449 *previous == OP_ONCE || *previous == OP_COND)
4453 int len = (int)(code - previous);
4454 uschar *bralink = NULL;
4456 /* Repeating a DEFINE group is pointless */
4458 if (*previous == OP_COND && previous[LINK_SIZE+1] == OP_DEF)
4460 *errorcodeptr = ERR55;
4464 /* If the maximum repeat count is unlimited, find the end of the bracket
4465 by scanning through from the start, and compute the offset back to it
4466 from the current code pointer. There may be an OP_OPT setting following
4467 the final KET, so we can't find the end just by going back from the code
4470 if (repeat_max == -1)
4472 register uschar *ket = previous;
4473 do ket += GET(ket, 1); while (*ket != OP_KET);
4474 ketoffset = (int)(code - ket);
4477 /* The case of a zero minimum is special because of the need to stick
4478 OP_BRAZERO in front of it, and because the group appears once in the
4479 data, whereas in other cases it appears the minimum number of times. For
4480 this reason, it is simplest to treat this case separately, as otherwise
4481 the code gets far too messy. There are several special subcases when the
4484 if (repeat_min == 0)
4486 /* If the maximum is also zero, we used to just omit the group from the
4487 output altogether, like this:
4489 ** if (repeat_max == 0)
4495 However, that fails when a group is referenced as a subroutine from
4496 elsewhere in the pattern, so now we stick in OP_SKIPZERO in front of it
4497 so that it is skipped on execution. As we don't have a list of which
4498 groups are referenced, we cannot do this selectively.
4500 If the maximum is 1 or unlimited, we just have to stick in the BRAZERO
4501 and do no more at this point. However, we do need to adjust any
4502 OP_RECURSE calls inside the group that refer to the group itself or any
4503 internal or forward referenced group, because the offset is from the
4504 start of the whole regex. Temporarily terminate the pattern while doing
4507 if (repeat_max <= 1) /* Covers 0, 1, and unlimited */
4510 adjust_recurse(previous, 1, utf8, cd, save_hwm);
4511 memmove(previous+1, previous, len);
4513 if (repeat_max == 0)
4515 *previous++ = OP_SKIPZERO;
4518 *previous++ = OP_BRAZERO + repeat_type;
4521 /* If the maximum is greater than 1 and limited, we have to replicate
4522 in a nested fashion, sticking OP_BRAZERO before each set of brackets.
4523 The first one has to be handled carefully because it's the original
4524 copy, which has to be moved up. The remainder can be handled by code
4525 that is common with the non-zero minimum case below. We have to
4526 adjust the value or repeat_max, since one less copy is required. Once
4527 again, we may have to adjust any OP_RECURSE calls inside the group. */
4533 adjust_recurse(previous, 2 + LINK_SIZE, utf8, cd, save_hwm);
4534 memmove(previous + 2 + LINK_SIZE, previous, len);
4535 code += 2 + LINK_SIZE;
4536 *previous++ = OP_BRAZERO + repeat_type;
4537 *previous++ = OP_BRA;
4539 /* We chain together the bracket offset fields that have to be
4540 filled in later when the ends of the brackets are reached. */
4542 offset = (bralink == NULL)? 0 : (int)(previous - bralink);
4544 PUTINC(previous, 0, offset);
4550 /* If the minimum is greater than zero, replicate the group as many
4551 times as necessary, and adjust the maximum to the number of subsequent
4552 copies that we need. If we set a first char from the group, and didn't
4553 set a required char, copy the latter from the former. If there are any
4554 forward reference subroutine calls in the group, there will be entries on
4555 the workspace list; replicate these with an appropriate increment. */
4561 /* In the pre-compile phase, we don't actually do the replication. We
4562 just adjust the length as if we had. Do some paranoid checks for
4563 potential integer overflow. The INT64_OR_DOUBLE type is a 64-bit
4564 integer type when available, otherwise double. */
4566 if (lengthptr != NULL)
4568 int delta = (repeat_min - 1)*length_prevgroup;
4569 if ((INT64_OR_DOUBLE)(repeat_min - 1)*
4570 (INT64_OR_DOUBLE)length_prevgroup >
4571 (INT64_OR_DOUBLE)INT_MAX ||
4572 OFLOW_MAX - *lengthptr < delta)
4574 *errorcodeptr = ERR20;
4577 *lengthptr += delta;
4580 /* This is compiling for real */
4584 if (groupsetfirstbyte && reqbyte < 0) reqbyte = firstbyte;
4585 for (i = 1; i < repeat_min; i++)
4588 uschar *this_hwm = cd->hwm;
4589 memcpy(code, previous, len);
4590 for (hc = save_hwm; hc < this_hwm; hc += LINK_SIZE)
4592 PUT(cd->hwm, 0, GET(hc, 0) + len);
4593 cd->hwm += LINK_SIZE;
4595 save_hwm = this_hwm;
4601 if (repeat_max > 0) repeat_max -= repeat_min;
4604 /* This code is common to both the zero and non-zero minimum cases. If
4605 the maximum is limited, it replicates the group in a nested fashion,
4606 remembering the bracket starts on a stack. In the case of a zero minimum,
4607 the first one was set up above. In all cases the repeat_max now specifies
4608 the number of additional copies needed. Again, we must remember to
4609 replicate entries on the forward reference list. */
4611 if (repeat_max >= 0)
4613 /* In the pre-compile phase, we don't actually do the replication. We
4614 just adjust the length as if we had. For each repetition we must add 1
4615 to the length for BRAZERO and for all but the last repetition we must
4616 add 2 + 2*LINKSIZE to allow for the nesting that occurs. Do some
4617 paranoid checks to avoid integer overflow. The INT64_OR_DOUBLE type is
4618 a 64-bit integer type when available, otherwise double. */
4620 if (lengthptr != NULL && repeat_max > 0)
4622 int delta = repeat_max * (length_prevgroup + 1 + 2 + 2*LINK_SIZE) -
4623 2 - 2*LINK_SIZE; /* Last one doesn't nest */
4624 if ((INT64_OR_DOUBLE)repeat_max *
4625 (INT64_OR_DOUBLE)(length_prevgroup + 1 + 2 + 2*LINK_SIZE)
4626 > (INT64_OR_DOUBLE)INT_MAX ||
4627 OFLOW_MAX - *lengthptr < delta)
4629 *errorcodeptr = ERR20;
4632 *lengthptr += delta;
4635 /* This is compiling for real */
4637 else for (i = repeat_max - 1; i >= 0; i--)
4640 uschar *this_hwm = cd->hwm;
4642 *code++ = OP_BRAZERO + repeat_type;
4644 /* All but the final copy start a new nesting, maintaining the
4645 chain of brackets outstanding. */
4651 offset = (bralink == NULL)? 0 : (int)(code - bralink);
4653 PUTINC(code, 0, offset);
4656 memcpy(code, previous, len);
4657 for (hc = save_hwm; hc < this_hwm; hc += LINK_SIZE)
4659 PUT(cd->hwm, 0, GET(hc, 0) + len + ((i != 0)? 2+LINK_SIZE : 1));
4660 cd->hwm += LINK_SIZE;
4662 save_hwm = this_hwm;
4666 /* Now chain through the pending brackets, and fill in their length
4667 fields (which are holding the chain links pro tem). */
4669 while (bralink != NULL)
4672 int offset = (int)(code - bralink + 1);
4673 uschar *bra = code - offset;
4674 oldlinkoffset = GET(bra, 1);
4675 bralink = (oldlinkoffset == 0)? NULL : bralink - oldlinkoffset;
4677 PUTINC(code, 0, offset);
4678 PUT(bra, 1, offset);
4682 /* If the maximum is unlimited, set a repeater in the final copy. We
4683 can't just offset backwards from the current code point, because we
4684 don't know if there's been an options resetting after the ket. The
4685 correct offset was computed above.
4687 Then, when we are doing the actual compile phase, check to see whether
4688 this group is a non-atomic one that could match an empty string. If so,
4689 convert the initial operator to the S form (e.g. OP_BRA -> OP_SBRA) so
4690 that runtime checking can be done. [This check is also applied to
4691 atomic groups at runtime, but in a different way.] */
4695 uschar *ketcode = code - ketoffset;
4696 uschar *bracode = ketcode - GET(ketcode, 1);
4697 *ketcode = OP_KETRMAX + repeat_type;
4698 if (lengthptr == NULL && *bracode != OP_ONCE)
4700 uschar *scode = bracode;
4703 if (could_be_empty_branch(scode, ketcode, utf8, cd))
4705 *bracode += OP_SBRA - OP_BRA;
4708 scode += GET(scode, 1);
4710 while (*scode == OP_ALT);
4715 /* If previous is OP_FAIL, it was generated by an empty class [] in
4716 JavaScript mode. The other ways in which OP_FAIL can be generated, that is
4717 by (*FAIL) or (?!) set previous to NULL, which gives a "nothing to repeat"
4718 error above. We can just ignore the repeat in JS case. */
4720 else if (*previous == OP_FAIL) goto END_REPEAT;
4722 /* Else there's some kind of shambles */
4726 *errorcodeptr = ERR11;
4730 /* If the character following a repeat is '+', or if certain optimization
4731 tests above succeeded, possessive_quantifier is TRUE. For some of the
4732 simpler opcodes, there is an special alternative opcode for this. For
4733 anything else, we wrap the entire repeated item inside OP_ONCE brackets.
4734 The '+' notation is just syntactic sugar, taken from Sun's Java package,
4735 but the special opcodes can optimize it a bit. The repeated item starts at
4736 tempcode, not at previous, which might be the first part of a string whose
4737 (former) last char we repeated.
4739 Possessifying an 'exact' quantifier has no effect, so we can ignore it. But
4740 an 'upto' may follow. We skip over an 'exact' item, and then test the
4741 length of what remains before proceeding. */
4743 if (possessive_quantifier)
4747 if (*tempcode == OP_TYPEEXACT)
4748 tempcode += _pcre_OP_lengths[*tempcode] +
4749 ((tempcode[3] == OP_PROP || tempcode[3] == OP_NOTPROP)? 2 : 0);
4751 else if (*tempcode == OP_EXACT || *tempcode == OP_NOTEXACT)
4753 tempcode += _pcre_OP_lengths[*tempcode];
4755 if (utf8 && tempcode[-1] >= 0xc0)
4756 tempcode += _pcre_utf8_table4[tempcode[-1] & 0x3f];
4760 len = (int)(code - tempcode);
4761 if (len > 0) switch (*tempcode)
4763 case OP_STAR: *tempcode = OP_POSSTAR; break;
4764 case OP_PLUS: *tempcode = OP_POSPLUS; break;
4765 case OP_QUERY: *tempcode = OP_POSQUERY; break;
4766 case OP_UPTO: *tempcode = OP_POSUPTO; break;
4768 case OP_TYPESTAR: *tempcode = OP_TYPEPOSSTAR; break;
4769 case OP_TYPEPLUS: *tempcode = OP_TYPEPOSPLUS; break;
4770 case OP_TYPEQUERY: *tempcode = OP_TYPEPOSQUERY; break;
4771 case OP_TYPEUPTO: *tempcode = OP_TYPEPOSUPTO; break;
4773 case OP_NOTSTAR: *tempcode = OP_NOTPOSSTAR; break;
4774 case OP_NOTPLUS: *tempcode = OP_NOTPOSPLUS; break;
4775 case OP_NOTQUERY: *tempcode = OP_NOTPOSQUERY; break;
4776 case OP_NOTUPTO: *tempcode = OP_NOTPOSUPTO; break;
4778 /* Because we are moving code along, we must ensure that any
4779 pending recursive references are updated. */
4783 adjust_recurse(tempcode, 1 + LINK_SIZE, utf8, cd, save_hwm);
4784 memmove(tempcode + 1+LINK_SIZE, tempcode, len);
4785 code += 1 + LINK_SIZE;
4786 len += 1 + LINK_SIZE;
4787 tempcode[0] = OP_ONCE;
4789 PUTINC(code, 0, len);
4790 PUT(tempcode, 1, len);
4795 /* In all case we no longer have a previous item. We also set the
4796 "follows varying string" flag for subsequently encountered reqbytes if
4797 it isn't already set and we have just passed a varying length item. */
4801 cd->req_varyopt |= reqvary;
4805 /* ===================================================================*/
4806 /* Start of nested parenthesized sub-expression, or comment or lookahead or
4807 lookbehind or option setting or condition or all the other extended
4808 parenthesis forms. */
4810 case CHAR_LEFT_PARENTHESIS:
4811 newoptions = options;
4815 reset_bracount = FALSE;
4817 /* First deal with various "verbs" that can be introduced by '*'. */
4819 if (*(++ptr) == CHAR_ASTERISK &&
4820 ((cd->ctypes[ptr[1]] & ctype_letter) != 0 || ptr[1] == ':'))
4824 const char *vn = verbnames;
4825 const uschar *name = ptr + 1;
4826 const uschar *arg = NULL;
4828 while ((cd->ctypes[*++ptr] & ctype_letter) != 0) {};
4829 namelen = (int)(ptr - name);
4831 if (*ptr == CHAR_COLON)
4834 while ((cd->ctypes[*ptr] & (ctype_letter|ctype_digit)) != 0
4835 || *ptr == '_') ptr++;
4836 arglen = (int)(ptr - arg);
4839 if (*ptr != CHAR_RIGHT_PARENTHESIS)
4841 *errorcodeptr = ERR60;
4845 /* Scan the table of verb names */
4847 for (i = 0; i < verbcount; i++)
4849 if (namelen == verbs[i].len &&
4850 strncmp((char *)name, vn, namelen) == 0)
4852 /* Check for open captures before ACCEPT */
4854 if (verbs[i].op == OP_ACCEPT)
4857 cd->had_accept = TRUE;
4858 for (oc = cd->open_caps; oc != NULL; oc = oc->next)
4861 PUT2INC(code, 0, oc->number);
4865 /* Handle the cases with/without an argument */
4869 if (verbs[i].op < 0) /* Argument is mandatory */
4871 *errorcodeptr = ERR66;
4874 *code = verbs[i].op;
4875 if (*code++ == OP_THEN)
4877 PUT(code, 0, code - bcptr->current_branch - 1);
4884 if (verbs[i].op_arg < 0) /* Argument is forbidden */
4886 *errorcodeptr = ERR59;
4889 *code = verbs[i].op_arg;
4890 if (*code++ == OP_THEN_ARG)
4892 PUT(code, 0, code - bcptr->current_branch - 1);
4896 memcpy(code, arg, arglen);
4901 break; /* Found verb, exit loop */
4904 vn += verbs[i].len + 1;
4907 if (i < verbcount) continue; /* Successfully handled a verb */
4908 *errorcodeptr = ERR60; /* Verb not recognized */
4912 /* Deal with the extended parentheses; all are introduced by '?', and the
4913 appearance of any of them means that this is not a capturing group. */
4915 else if (*ptr == CHAR_QUESTION_MARK)
4917 int i, set, unset, namelen;
4924 case CHAR_NUMBER_SIGN: /* Comment; skip to ket */
4926 while (*ptr != 0 && *ptr != CHAR_RIGHT_PARENTHESIS) ptr++;
4929 *errorcodeptr = ERR18;
4935 /* ------------------------------------------------------------ */
4936 case CHAR_VERTICAL_LINE: /* Reset capture count for each branch */
4937 reset_bracount = TRUE;
4940 /* ------------------------------------------------------------ */
4941 case CHAR_COLON: /* Non-capturing bracket */
4947 /* ------------------------------------------------------------ */
4948 case CHAR_LEFT_PARENTHESIS:
4949 bravalue = OP_COND; /* Conditional group */
4951 /* A condition can be an assertion, a number (referring to a numbered
4952 group), a name (referring to a named group), or 'R', referring to
4953 recursion. R<digits> and R&name are also permitted for recursion tests.
4955 There are several syntaxes for testing a named group: (?(name)) is used
4956 by Python; Perl 5.10 onwards uses (?(<name>) or (?('name')).
4958 There are two unfortunate ambiguities, caused by history. (a) 'R' can
4959 be the recursive thing or the name 'R' (and similarly for 'R' followed
4960 by digits), and (b) a number could be a name that consists of digits.
4961 In both cases, we look for a name first; if not found, we try the other
4964 /* For conditions that are assertions, check the syntax, and then exit
4965 the switch. This will take control down to where bracketed groups,
4966 including assertions, are processed. */
4968 if (ptr[1] == CHAR_QUESTION_MARK && (ptr[2] == CHAR_EQUALS_SIGN ||
4969 ptr[2] == CHAR_EXCLAMATION_MARK || ptr[2] == CHAR_LESS_THAN_SIGN))
4972 /* Most other conditions use OP_CREF (a couple change to OP_RREF
4973 below), and all need to skip 3 bytes at the start of the group. */
4975 code[1+LINK_SIZE] = OP_CREF;
4979 /* Check for a test for recursion in a named group. */
4981 if (ptr[1] == CHAR_R && ptr[2] == CHAR_AMPERSAND)
4985 code[1+LINK_SIZE] = OP_RREF; /* Change the type of test */
4988 /* Check for a test for a named group's having been set, using the Perl
4989 syntax (?(<name>) or (?('name') */
4991 else if (ptr[1] == CHAR_LESS_THAN_SIGN)
4993 terminator = CHAR_GREATER_THAN_SIGN;
4996 else if (ptr[1] == CHAR_APOSTROPHE)
4998 terminator = CHAR_APOSTROPHE;
5004 if (ptr[1] == CHAR_MINUS || ptr[1] == CHAR_PLUS) refsign = *(++ptr);
5007 /* We now expect to read a name; any thing else is an error */
5009 if ((cd->ctypes[ptr[1]] & ctype_word) == 0)
5011 ptr += 1; /* To get the right offset */
5012 *errorcodeptr = ERR28;
5016 /* Read the name, but also get it as a number if it's all digits */
5020 while ((cd->ctypes[*ptr] & ctype_word) != 0)
5023 recno = ((digitab[*ptr] & ctype_digit) != 0)?
5024 recno * 10 + *ptr - CHAR_0 : -1;
5027 namelen = (int)(ptr - name);
5029 if ((terminator > 0 && *ptr++ != terminator) ||
5030 *ptr++ != CHAR_RIGHT_PARENTHESIS)
5032 ptr--; /* Error offset */
5033 *errorcodeptr = ERR26;
5037 /* Do no further checking in the pre-compile phase. */
5039 if (lengthptr != NULL) break;
5041 /* In the real compile we do the work of looking for the actual
5042 reference. If the string started with "+" or "-" we require the rest to
5043 be digits, in which case recno will be set. */
5049 *errorcodeptr = ERR58;
5052 recno = (refsign == CHAR_MINUS)?
5053 cd->bracount - recno + 1 : recno +cd->bracount;
5054 if (recno <= 0 || recno > cd->final_bracount)
5056 *errorcodeptr = ERR15;
5059 PUT2(code, 2+LINK_SIZE, recno);
5063 /* Otherwise (did not start with "+" or "-"), start by looking for the
5064 name. If we find a name, add one to the opcode to change OP_CREF or
5065 OP_RREF into OP_NCREF or OP_NRREF. These behave exactly the same,
5066 except they record that the reference was originally to a name. The
5067 information is used to check duplicate names. */
5069 slot = cd->name_table;
5070 for (i = 0; i < cd->names_found; i++)
5072 if (strncmp((char *)name, (char *)slot+2, namelen) == 0) break;
5073 slot += cd->name_entry_size;
5076 /* Found a previous named subpattern */
5078 if (i < cd->names_found)
5080 recno = GET2(slot, 0);
5081 PUT2(code, 2+LINK_SIZE, recno);
5082 code[1+LINK_SIZE]++;
5085 /* Search the pattern for a forward reference */
5087 else if ((i = find_parens(cd, name, namelen,
5088 (options & PCRE_EXTENDED) != 0, utf8)) > 0)
5090 PUT2(code, 2+LINK_SIZE, i);
5091 code[1+LINK_SIZE]++;
5094 /* If terminator == 0 it means that the name followed directly after
5095 the opening parenthesis [e.g. (?(abc)...] and in this case there are
5096 some further alternatives to try. For the cases where terminator != 0
5097 [things like (?(<name>... or (?('name')... or (?(R&name)... ] we have
5098 now checked all the possibilities, so give an error. */
5100 else if (terminator != 0)
5102 *errorcodeptr = ERR15;
5106 /* Check for (?(R) for recursion. Allow digits after R to specify a
5107 specific group number. */
5109 else if (*name == CHAR_R)
5112 for (i = 1; i < namelen; i++)
5114 if ((digitab[name[i]] & ctype_digit) == 0)
5116 *errorcodeptr = ERR15;
5119 recno = recno * 10 + name[i] - CHAR_0;
5121 if (recno == 0) recno = RREF_ANY;
5122 code[1+LINK_SIZE] = OP_RREF; /* Change test type */
5123 PUT2(code, 2+LINK_SIZE, recno);
5126 /* Similarly, check for the (?(DEFINE) "condition", which is always
5129 else if (namelen == 6 && strncmp((char *)name, STRING_DEFINE, 6) == 0)
5131 code[1+LINK_SIZE] = OP_DEF;
5135 /* Check for the "name" actually being a subpattern number. We are
5136 in the second pass here, so final_bracount is set. */
5138 else if (recno > 0 && recno <= cd->final_bracount)
5140 PUT2(code, 2+LINK_SIZE, recno);
5143 /* Either an unidentified subpattern, or a reference to (?(0) */
5147 *errorcodeptr = (recno == 0)? ERR35: ERR15;
5153 /* ------------------------------------------------------------ */
5154 case CHAR_EQUALS_SIGN: /* Positive lookahead */
5155 bravalue = OP_ASSERT;
5160 /* ------------------------------------------------------------ */
5161 case CHAR_EXCLAMATION_MARK: /* Negative lookahead */
5163 if (*ptr == CHAR_RIGHT_PARENTHESIS) /* Optimize (?!) */
5169 bravalue = OP_ASSERT_NOT;
5173 /* ------------------------------------------------------------ */
5174 case CHAR_LESS_THAN_SIGN: /* Lookbehind or named define */
5177 case CHAR_EQUALS_SIGN: /* Positive lookbehind */
5178 bravalue = OP_ASSERTBACK;
5182 case CHAR_EXCLAMATION_MARK: /* Negative lookbehind */
5183 bravalue = OP_ASSERTBACK_NOT;
5187 default: /* Could be name define, else bad */
5188 if ((cd->ctypes[ptr[1]] & ctype_word) != 0) goto DEFINE_NAME;
5189 ptr++; /* Correct offset for error */
5190 *errorcodeptr = ERR24;
5196 /* ------------------------------------------------------------ */
5197 case CHAR_GREATER_THAN_SIGN: /* One-time brackets */
5203 /* ------------------------------------------------------------ */
5204 case CHAR_C: /* Callout - may be followed by digits; */
5205 previous_callout = code; /* Save for later completion */
5206 after_manual_callout = 1; /* Skip one item before completing */
5207 *code++ = OP_CALLOUT;
5210 while ((digitab[*(++ptr)] & ctype_digit) != 0)
5211 n = n * 10 + *ptr - CHAR_0;
5212 if (*ptr != CHAR_RIGHT_PARENTHESIS)
5214 *errorcodeptr = ERR39;
5219 *errorcodeptr = ERR38;
5223 PUT(code, 0, (int)(ptr - cd->start_pattern + 1)); /* Pattern offset */
5224 PUT(code, LINK_SIZE, 0); /* Default length */
5225 code += 2 * LINK_SIZE;
5231 /* ------------------------------------------------------------ */
5232 case CHAR_P: /* Python-style named subpattern handling */
5233 if (*(++ptr) == CHAR_EQUALS_SIGN ||
5234 *ptr == CHAR_GREATER_THAN_SIGN) /* Reference or recursion */
5236 is_recurse = *ptr == CHAR_GREATER_THAN_SIGN;
5237 terminator = CHAR_RIGHT_PARENTHESIS;
5238 goto NAMED_REF_OR_RECURSE;
5240 else if (*ptr != CHAR_LESS_THAN_SIGN) /* Test for Python-style defn */
5242 *errorcodeptr = ERR41;
5245 /* Fall through to handle (?P< as (?< is handled */
5248 /* ------------------------------------------------------------ */
5249 DEFINE_NAME: /* Come here from (?< handling */
5250 case CHAR_APOSTROPHE:
5252 terminator = (*ptr == CHAR_LESS_THAN_SIGN)?
5253 CHAR_GREATER_THAN_SIGN : CHAR_APOSTROPHE;
5256 while ((cd->ctypes[*ptr] & ctype_word) != 0) ptr++;
5257 namelen = (int)(ptr - name);
5259 /* In the pre-compile phase, just do a syntax check. */
5261 if (lengthptr != NULL)
5263 if (*ptr != terminator)
5265 *errorcodeptr = ERR42;
5268 if (cd->names_found >= MAX_NAME_COUNT)
5270 *errorcodeptr = ERR49;
5273 if (namelen + 3 > cd->name_entry_size)
5275 cd->name_entry_size = namelen + 3;
5276 if (namelen > MAX_NAME_SIZE)
5278 *errorcodeptr = ERR48;
5284 /* In the real compile, create the entry in the table, maintaining
5285 alphabetical order. Duplicate names for different numbers are
5286 permitted only if PCRE_DUPNAMES is set. Duplicate names for the same
5287 number are always OK. (An existing number can be re-used if (?|
5288 appears in the pattern.) In either event, a duplicate name results in
5289 a duplicate entry in the table, even if the number is the same. This
5290 is because the number of names, and hence the table size, is computed
5291 in the pre-compile, and it affects various numbers and pointers which
5292 would all have to be modified, and the compiled code moved down, if
5293 duplicates with the same number were omitted from the table. This
5294 doesn't seem worth the hassle. However, *different* names for the
5295 same number are not permitted. */
5299 BOOL dupname = FALSE;
5300 slot = cd->name_table;
5302 for (i = 0; i < cd->names_found; i++)
5304 int crc = memcmp(name, slot+2, namelen);
5307 if (slot[2+namelen] == 0)
5309 if (GET2(slot, 0) != cd->bracount + 1 &&
5310 (options & PCRE_DUPNAMES) == 0)
5312 *errorcodeptr = ERR43;
5315 else dupname = TRUE;
5317 else crc = -1; /* Current name is a substring */
5320 /* Make space in the table and break the loop for an earlier
5321 name. For a duplicate or later name, carry on. We do this for
5322 duplicates so that in the simple case (when ?(| is not used) they
5323 are in order of their numbers. */
5327 memmove(slot + cd->name_entry_size, slot,
5328 (cd->names_found - i) * cd->name_entry_size);
5332 /* Continue the loop for a later or duplicate name */
5334 slot += cd->name_entry_size;
5337 /* For non-duplicate names, check for a duplicate number before
5338 adding the new name. */
5342 uschar *cslot = cd->name_table;
5343 for (i = 0; i < cd->names_found; i++)
5347 if (GET2(cslot, 0) == cd->bracount + 1)
5349 *errorcodeptr = ERR65;
5354 cslot += cd->name_entry_size;
5358 PUT2(slot, 0, cd->bracount + 1);
5359 memcpy(slot + 2, name, namelen);
5360 slot[2+namelen] = 0;
5364 /* In both pre-compile and compile, count the number of names we've
5368 ptr++; /* Move past > or ' */
5369 goto NUMBERED_GROUP;
5372 /* ------------------------------------------------------------ */
5373 case CHAR_AMPERSAND: /* Perl recursion/subroutine syntax */
5374 terminator = CHAR_RIGHT_PARENTHESIS;
5378 /* We come here from the Python syntax above that handles both
5379 references (?P=name) and recursion (?P>name), as well as falling
5380 through from the Perl recursion syntax (?&name). We also come here from
5381 the Perl \k<name> or \k'name' back reference syntax and the \k{name}
5382 .NET syntax, and the Oniguruma \g<...> and \g'...' subroutine syntax. */
5384 NAMED_REF_OR_RECURSE:
5386 while ((cd->ctypes[*ptr] & ctype_word) != 0) ptr++;
5387 namelen = (int)(ptr - name);
5389 /* In the pre-compile phase, do a syntax check. We used to just set
5390 a dummy reference number, because it was not used in the first pass.
5391 However, with the change of recursive back references to be atomic,
5392 we have to look for the number so that this state can be identified, as
5393 otherwise the incorrect length is computed. If it's not a backwards
5394 reference, the dummy number will do. */
5396 if (lengthptr != NULL)
5402 *errorcodeptr = ERR62;
5405 if (*ptr != terminator)
5407 *errorcodeptr = ERR42;
5410 if (namelen > MAX_NAME_SIZE)
5412 *errorcodeptr = ERR48;
5416 /* The name table does not exist in the first pass, so we cannot
5417 do a simple search as in the code below. Instead, we have to scan the
5418 pattern to find the number. It is important that we scan it only as
5419 far as we have got because the syntax of named subpatterns has not
5420 been checked for the rest of the pattern, and find_parens() assumes
5421 correct syntax. In any case, it's a waste of resources to scan
5422 further. We stop the scan at the current point by temporarily
5423 adjusting the value of cd->endpattern. */
5425 temp = cd->end_pattern;
5426 cd->end_pattern = ptr;
5427 recno = find_parens(cd, name, namelen,
5428 (options & PCRE_EXTENDED) != 0, utf8);
5429 cd->end_pattern = temp;
5430 if (recno < 0) recno = 0; /* Forward ref; set dummy number */
5433 /* In the real compile, seek the name in the table. We check the name
5434 first, and then check that we have reached the end of the name in the
5435 table. That way, if the name that is longer than any in the table,
5436 the comparison will fail without reading beyond the table entry. */
5440 slot = cd->name_table;
5441 for (i = 0; i < cd->names_found; i++)
5443 if (strncmp((char *)name, (char *)slot+2, namelen) == 0 &&
5444 slot[2+namelen] == 0)
5446 slot += cd->name_entry_size;
5449 if (i < cd->names_found) /* Back reference */
5451 recno = GET2(slot, 0);
5453 else if ((recno = /* Forward back reference */
5454 find_parens(cd, name, namelen,
5455 (options & PCRE_EXTENDED) != 0, utf8)) <= 0)
5457 *errorcodeptr = ERR15;
5462 /* In both phases, we can now go to the code than handles numerical
5463 recursion or backreferences. */
5465 if (is_recurse) goto HANDLE_RECURSION;
5466 else goto HANDLE_REFERENCE;
5469 /* ------------------------------------------------------------ */
5470 case CHAR_R: /* Recursion */
5471 ptr++; /* Same as (?0) */
5475 /* ------------------------------------------------------------ */
5476 case CHAR_MINUS: case CHAR_PLUS: /* Recursion or subroutine */
5477 case CHAR_0: case CHAR_1: case CHAR_2: case CHAR_3: case CHAR_4:
5478 case CHAR_5: case CHAR_6: case CHAR_7: case CHAR_8: case CHAR_9:
5480 const uschar *called;
5481 terminator = CHAR_RIGHT_PARENTHESIS;
5483 /* Come here from the \g<...> and \g'...' code (Oniguruma
5484 compatibility). However, the syntax has been checked to ensure that
5485 the ... are a (signed) number, so that neither ERR63 nor ERR29 will
5486 be called on this path, nor with the jump to OTHER_CHAR_AFTER_QUERY
5489 HANDLE_NUMERICAL_RECURSION:
5491 if ((refsign = *ptr) == CHAR_PLUS)
5494 if ((digitab[*ptr] & ctype_digit) == 0)
5496 *errorcodeptr = ERR63;
5500 else if (refsign == CHAR_MINUS)
5502 if ((digitab[ptr[1]] & ctype_digit) == 0)
5503 goto OTHER_CHAR_AFTER_QUERY;
5508 while((digitab[*ptr] & ctype_digit) != 0)
5509 recno = recno * 10 + *ptr++ - CHAR_0;
5511 if (*ptr != terminator)
5513 *errorcodeptr = ERR29;
5517 if (refsign == CHAR_MINUS)
5521 *errorcodeptr = ERR58;
5524 recno = cd->bracount - recno + 1;
5527 *errorcodeptr = ERR15;
5531 else if (refsign == CHAR_PLUS)
5535 *errorcodeptr = ERR58;
5538 recno += cd->bracount;
5541 /* Come here from code above that handles a named recursion */
5546 called = cd->start_code;
5548 /* When we are actually compiling, find the bracket that is being
5549 referenced. Temporarily end the regex in case it doesn't exist before
5550 this point. If we end up with a forward reference, first check that
5551 the bracket does occur later so we can give the error (and position)
5552 now. Then remember this forward reference in the workspace so it can
5553 be filled in at the end. */
5555 if (lengthptr == NULL)
5559 called = _pcre_find_bracket(cd->start_code, utf8, recno);
5561 /* Forward reference */
5565 if (find_parens(cd, NULL, recno,
5566 (options & PCRE_EXTENDED) != 0, utf8) < 0)
5568 *errorcodeptr = ERR15;
5572 /* Fudge the value of "called" so that when it is inserted as an
5573 offset below, what it actually inserted is the reference number
5576 called = cd->start_code + recno;
5577 PUTINC(cd->hwm, 0, (int)(code + 2 + LINK_SIZE - cd->start_code));
5580 /* If not a forward reference, and the subpattern is still open,
5581 this is a recursive call. We check to see if this is a left
5582 recursion that could loop for ever, and diagnose that case. */
5584 else if (GET(called, 1) == 0 &&
5585 could_be_empty(called, code, bcptr, utf8, cd))
5587 *errorcodeptr = ERR40;
5592 /* Insert the recursion/subroutine item, automatically wrapped inside
5593 "once" brackets. Set up a "previous group" length so that a
5594 subsequent quantifier will work. */
5597 PUT(code, 1, 2 + 2*LINK_SIZE);
5598 code += 1 + LINK_SIZE;
5601 PUT(code, 1, (int)(called - cd->start_code));
5602 code += 1 + LINK_SIZE;
5605 PUT(code, 1, 2 + 2*LINK_SIZE);
5606 code += 1 + LINK_SIZE;
5608 length_prevgroup = 3 + 3*LINK_SIZE;
5611 /* Can't determine a first byte now */
5613 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
5617 /* ------------------------------------------------------------ */
5618 default: /* Other characters: check option setting */
5619 OTHER_CHAR_AFTER_QUERY:
5623 while (*ptr != CHAR_RIGHT_PARENTHESIS && *ptr != CHAR_COLON)
5627 case CHAR_MINUS: optset = &unset; break;
5629 case CHAR_J: /* Record that it changed in the external options */
5630 *optset |= PCRE_DUPNAMES;
5631 cd->external_flags |= PCRE_JCHANGED;
5634 case CHAR_i: *optset |= PCRE_CASELESS; break;
5635 case CHAR_m: *optset |= PCRE_MULTILINE; break;
5636 case CHAR_s: *optset |= PCRE_DOTALL; break;
5637 case CHAR_x: *optset |= PCRE_EXTENDED; break;
5638 case CHAR_U: *optset |= PCRE_UNGREEDY; break;
5639 case CHAR_X: *optset |= PCRE_EXTRA; break;
5641 default: *errorcodeptr = ERR12;
5642 ptr--; /* Correct the offset */
5647 /* Set up the changed option bits, but don't change anything yet. */
5649 newoptions = (options | set) & (~unset);
5651 /* If the options ended with ')' this is not the start of a nested
5652 group with option changes, so the options change at this level. If this
5653 item is right at the start of the pattern, the options can be
5654 abstracted and made external in the pre-compile phase, and ignored in
5655 the compile phase. This can be helpful when matching -- for instance in
5656 caseless checking of required bytes.
5658 If the code pointer is not (cd->start_code + 1 + LINK_SIZE), we are
5659 definitely *not* at the start of the pattern because something has been
5660 compiled. In the pre-compile phase, however, the code pointer can have
5661 that value after the start, because it gets reset as code is discarded
5662 during the pre-compile. However, this can happen only at top level - if
5663 we are within parentheses, the starting BRA will still be present. At
5664 any parenthesis level, the length value can be used to test if anything
5665 has been compiled at that level. Thus, a test for both these conditions
5666 is necessary to ensure we correctly detect the start of the pattern in
5669 If we are not at the pattern start, compile code to change the ims
5670 options if this setting actually changes any of them, and reset the
5671 greedy defaults and the case value for firstbyte and reqbyte. */
5673 if (*ptr == CHAR_RIGHT_PARENTHESIS)
5675 if (code == cd->start_code + 1 + LINK_SIZE &&
5676 (lengthptr == NULL || *lengthptr == 2 + 2*LINK_SIZE))
5678 cd->external_options = newoptions;
5682 if ((options & PCRE_IMS) != (newoptions & PCRE_IMS))
5685 *code++ = newoptions & PCRE_IMS;
5687 greedy_default = ((newoptions & PCRE_UNGREEDY) != 0);
5688 greedy_non_default = greedy_default ^ 1;
5689 req_caseopt = ((newoptions & PCRE_CASELESS) != 0)? REQ_CASELESS : 0;
5692 /* Change options at this level, and pass them back for use
5693 in subsequent branches. When not at the start of the pattern, this
5694 information is also necessary so that a resetting item can be
5695 compiled at the end of a group (if we are in a group). */
5697 *optionsptr = options = newoptions;
5698 previous = NULL; /* This item can't be repeated */
5699 continue; /* It is complete */
5702 /* If the options ended with ':' we are heading into a nested group
5703 with possible change of options. Such groups are non-capturing and are
5704 not assertions of any kind. All we need to do is skip over the ':';
5705 the newoptions value is handled below. */
5709 } /* End of switch for character following (? */
5710 } /* End of (? handling */
5712 /* Opening parenthesis not followed by '*' or '?'. If PCRE_NO_AUTO_CAPTURE
5713 is set, all unadorned brackets become non-capturing and behave like (?:...)
5716 else if ((options & PCRE_NO_AUTO_CAPTURE) != 0)
5721 /* Else we have a capturing group. */
5727 PUT2(code, 1+LINK_SIZE, cd->bracount);
5731 /* Process nested bracketed regex. Assertions may not be repeated, but
5732 other kinds can be. All their opcodes are >= OP_ONCE. We copy code into a
5733 non-register variable in order to be able to pass its address because some
5734 compilers complain otherwise. Pass in a new setting for the ims options if
5735 they have changed. */
5737 previous = (bravalue >= OP_ONCE)? code : NULL;
5740 tempreqvary = cd->req_varyopt; /* Save value before bracket */
5741 length_prevgroup = 0; /* Initialize for pre-compile phase */
5744 newoptions, /* The complete new option state */
5745 options & PCRE_IMS, /* The previous ims option state */
5746 &tempcode, /* Where to put code (updated) */
5747 &ptr, /* Input pointer (updated) */
5748 errorcodeptr, /* Where to put an error message */
5749 (bravalue == OP_ASSERTBACK ||
5750 bravalue == OP_ASSERTBACK_NOT), /* TRUE if back assert */
5751 reset_bracount, /* True if (?| group */
5752 skipbytes, /* Skip over bracket number */
5753 &subfirstbyte, /* For possible first char */
5754 &subreqbyte, /* For possible last char */
5755 bcptr, /* Current branch chain */
5756 cd, /* Tables block */
5757 (lengthptr == NULL)? NULL : /* Actual compile phase */
5758 &length_prevgroup /* Pre-compile phase */
5762 /* At the end of compiling, code is still pointing to the start of the
5763 group, while tempcode has been updated to point past the end of the group
5764 and any option resetting that may follow it. The pattern pointer (ptr)
5765 is on the bracket. */
5767 /* If this is a conditional bracket, check that there are no more than
5768 two branches in the group, or just one if it's a DEFINE group. We do this
5769 in the real compile phase, not in the pre-pass, where the whole group may
5770 not be available. */
5772 if (bravalue == OP_COND && lengthptr == NULL)
5781 while (*tc != OP_KET);
5783 /* A DEFINE group is never obeyed inline (the "condition" is always
5784 false). It must have only one branch. */
5786 if (code[LINK_SIZE+1] == OP_DEF)
5790 *errorcodeptr = ERR54;
5793 bravalue = OP_DEF; /* Just a flag to suppress char handling below */
5796 /* A "normal" conditional group. If there is just one branch, we must not
5797 make use of its firstbyte or reqbyte, because this is equivalent to an
5798 empty second branch. */
5804 *errorcodeptr = ERR27;
5807 if (condcount == 1) subfirstbyte = subreqbyte = REQ_NONE;
5811 /* Error if hit end of pattern */
5813 if (*ptr != CHAR_RIGHT_PARENTHESIS)
5815 *errorcodeptr = ERR14;
5819 /* In the pre-compile phase, update the length by the length of the group,
5820 less the brackets at either end. Then reduce the compiled code to just a
5821 set of non-capturing brackets so that it doesn't use much memory if it is
5822 duplicated by a quantifier.*/
5824 if (lengthptr != NULL)
5826 if (OFLOW_MAX - *lengthptr < length_prevgroup - 2 - 2*LINK_SIZE)
5828 *errorcodeptr = ERR20;
5831 *lengthptr += length_prevgroup - 2 - 2*LINK_SIZE;
5833 PUTINC(code, 0, 1 + LINK_SIZE);
5835 PUTINC(code, 0, 1 + LINK_SIZE);
5836 break; /* No need to waste time with special character handling */
5839 /* Otherwise update the main code pointer to the end of the group. */
5843 /* For a DEFINE group, required and first character settings are not
5846 if (bravalue == OP_DEF) break;
5848 /* Handle updating of the required and first characters for other types of
5849 group. Update for normal brackets of all kinds, and conditions with two
5850 branches (see code above). If the bracket is followed by a quantifier with
5851 zero repeat, we have to back off. Hence the definition of zeroreqbyte and
5852 zerofirstbyte outside the main loop so that they can be accessed for the
5855 zeroreqbyte = reqbyte;
5856 zerofirstbyte = firstbyte;
5857 groupsetfirstbyte = FALSE;
5859 if (bravalue >= OP_ONCE)
5861 /* If we have not yet set a firstbyte in this branch, take it from the
5862 subpattern, remembering that it was set here so that a repeat of more
5863 than one can replicate it as reqbyte if necessary. If the subpattern has
5864 no firstbyte, set "none" for the whole branch. In both cases, a zero
5865 repeat forces firstbyte to "none". */
5867 if (firstbyte == REQ_UNSET)
5869 if (subfirstbyte >= 0)
5871 firstbyte = subfirstbyte;
5872 groupsetfirstbyte = TRUE;
5874 else firstbyte = REQ_NONE;
5875 zerofirstbyte = REQ_NONE;
5878 /* If firstbyte was previously set, convert the subpattern's firstbyte
5879 into reqbyte if there wasn't one, using the vary flag that was in
5880 existence beforehand. */
5882 else if (subfirstbyte >= 0 && subreqbyte < 0)
5883 subreqbyte = subfirstbyte | tempreqvary;
5885 /* If the subpattern set a required byte (or set a first byte that isn't
5886 really the first byte - see above), set it. */
5888 if (subreqbyte >= 0) reqbyte = subreqbyte;
5891 /* For a forward assertion, we take the reqbyte, if set. This can be
5892 helpful if the pattern that follows the assertion doesn't set a different
5893 char. For example, it's useful for /(?=abcde).+/. We can't set firstbyte
5894 for an assertion, however because it leads to incorrect effect for patterns
5895 such as /(?=a)a.+/ when the "real" "a" would then become a reqbyte instead
5896 of a firstbyte. This is overcome by a scan at the end if there's no
5897 firstbyte, looking for an asserted first char. */
5899 else if (bravalue == OP_ASSERT && subreqbyte >= 0) reqbyte = subreqbyte;
5900 break; /* End of processing '(' */
5903 /* ===================================================================*/
5904 /* Handle metasequences introduced by \. For ones like \d, the ESC_ values
5905 are arranged to be the negation of the corresponding OP_values in the
5906 default case when PCRE_UCP is not set. For the back references, the values
5907 are ESC_REF plus the reference number. Only back references and those types
5908 that consume a character may be repeated. We can test for values between
5909 ESC_b and ESC_Z for the latter; this may have to change if any new ones are
5912 case CHAR_BACKSLASH:
5914 c = check_escape(&ptr, errorcodeptr, cd->bracount, options, FALSE);
5915 if (*errorcodeptr != 0) goto FAILED;
5919 if (-c == ESC_Q) /* Handle start of quoted string */
5921 if (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
5922 ptr += 2; /* avoid empty string */
5927 if (-c == ESC_E) continue; /* Perl ignores an orphan \E */
5929 /* For metasequences that actually match a character, we disable the
5930 setting of a first character if it hasn't already been set. */
5932 if (firstbyte == REQ_UNSET && -c > ESC_b && -c < ESC_Z)
5933 firstbyte = REQ_NONE;
5935 /* Set values to reset to if this is followed by a zero repeat. */
5937 zerofirstbyte = firstbyte;
5938 zeroreqbyte = reqbyte;
5940 /* \g<name> or \g'name' is a subroutine call by name and \g<n> or \g'n'
5941 is a subroutine call by number (Oniguruma syntax). In fact, the value
5942 -ESC_g is returned only for these cases. So we don't need to check for <
5943 or ' if the value is -ESC_g. For the Perl syntax \g{n} the value is
5944 -ESC_REF+n, and for the Perl syntax \g{name} the result is -ESC_k (as
5945 that is a synonym for a named back reference). */
5950 save_hwm = cd->hwm; /* Normally this is set when '(' is read */
5951 terminator = (*(++ptr) == CHAR_LESS_THAN_SIGN)?
5952 CHAR_GREATER_THAN_SIGN : CHAR_APOSTROPHE;
5954 /* These two statements stop the compiler for warning about possibly
5955 unset variables caused by the jump to HANDLE_NUMERICAL_RECURSION. In
5956 fact, because we actually check for a number below, the paths that
5957 would actually be in error are never taken. */
5960 reset_bracount = FALSE;
5962 /* Test for a name */
5964 if (ptr[1] != CHAR_PLUS && ptr[1] != CHAR_MINUS)
5966 BOOL isnumber = TRUE;
5967 for (p = ptr + 1; *p != 0 && *p != terminator; p++)
5969 if ((cd->ctypes[*p] & ctype_digit) == 0) isnumber = FALSE;
5970 if ((cd->ctypes[*p] & ctype_word) == 0) break;
5972 if (*p != terminator)
5974 *errorcodeptr = ERR57;
5980 goto HANDLE_NUMERICAL_RECURSION;
5983 goto NAMED_REF_OR_RECURSE;
5986 /* Test a signed number in angle brackets or quotes. */
5989 while ((digitab[*p] & ctype_digit) != 0) p++;
5990 if (*p != terminator)
5992 *errorcodeptr = ERR57;
5996 goto HANDLE_NUMERICAL_RECURSION;
5999 /* \k<name> or \k'name' is a back reference by name (Perl syntax).
6000 We also support \k{name} (.NET syntax) */
6002 if (-c == ESC_k && (ptr[1] == CHAR_LESS_THAN_SIGN ||
6003 ptr[1] == CHAR_APOSTROPHE || ptr[1] == CHAR_LEFT_CURLY_BRACKET))
6006 terminator = (*(++ptr) == CHAR_LESS_THAN_SIGN)?
6007 CHAR_GREATER_THAN_SIGN : (*ptr == CHAR_APOSTROPHE)?
6008 CHAR_APOSTROPHE : CHAR_RIGHT_CURLY_BRACKET;
6009 goto NAMED_REF_OR_RECURSE;
6012 /* Back references are handled specially; must disable firstbyte if
6013 not set to cope with cases like (?=(\w+))\1: which would otherwise set
6019 recno = -c - ESC_REF;
6021 HANDLE_REFERENCE: /* Come here from named backref handling */
6022 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
6025 PUT2INC(code, 0, recno);
6026 cd->backref_map |= (recno < 32)? (1 << recno) : 1;
6027 if (recno > cd->top_backref) cd->top_backref = recno;
6029 /* Check to see if this back reference is recursive, that it, it
6030 is inside the group that it references. A flag is set so that the
6031 group can be made atomic. */
6033 for (oc = cd->open_caps; oc != NULL; oc = oc->next)
6035 if (oc->number == recno)
6043 /* So are Unicode property matches, if supported. */
6046 else if (-c == ESC_P || -c == ESC_p)
6050 int ptype = get_ucp(&ptr, &negated, &pdata, errorcodeptr);
6051 if (ptype < 0) goto FAILED;
6053 *code++ = ((-c == ESC_p) != negated)? OP_PROP : OP_NOTPROP;
6059 /* If Unicode properties are not supported, \X, \P, and \p are not
6062 else if (-c == ESC_X || -c == ESC_P || -c == ESC_p)
6064 *errorcodeptr = ERR45;
6069 /* For the rest (including \X when Unicode properties are supported), we
6070 can obtain the OP value by negating the escape value in the default
6071 situation when PCRE_UCP is not set. When it *is* set, we substitute
6072 Unicode property tests. */
6077 if (-c >= ESC_DU && -c <= ESC_wu)
6079 nestptr = ptr + 1; /* Where to resume */
6080 ptr = substitutes[-c - ESC_DU] - 1; /* Just before substitute */
6085 previous = (-c > ESC_b && -c < ESC_Z)? code : NULL;
6092 /* We have a data character whose value is in c. In UTF-8 mode it may have
6093 a value > 127. We set its representation in the length/buffer, and then
6094 handle it as a data character. */
6097 if (utf8 && c > 127)
6098 mclength = _pcre_ord2utf8(c, mcbuffer);
6109 /* ===================================================================*/
6110 /* Handle a literal character. It is guaranteed not to be whitespace or #
6111 when the extended flag is set. If we are in UTF-8 mode, it may be a
6112 multi-byte literal character. */
6120 if (utf8 && c >= 0xc0)
6122 while ((ptr[1] & 0xc0) == 0x80)
6123 mcbuffer[mclength++] = *(++ptr);
6127 /* At this point we have the character's bytes in mcbuffer, and the length
6128 in mclength. When not in UTF-8 mode, the length is always 1. */
6132 *code++ = ((options & PCRE_CASELESS) != 0)? OP_CHARNC : OP_CHAR;
6133 for (c = 0; c < mclength; c++) *code++ = mcbuffer[c];
6135 /* Remember if \r or \n were seen */
6137 if (mcbuffer[0] == CHAR_CR || mcbuffer[0] == CHAR_NL)
6138 cd->external_flags |= PCRE_HASCRORLF;
6140 /* Set the first and required bytes appropriately. If no previous first
6141 byte, set it from this character, but revert to none on a zero repeat.
6142 Otherwise, leave the firstbyte value alone, and don't change it on a zero
6145 if (firstbyte == REQ_UNSET)
6147 zerofirstbyte = REQ_NONE;
6148 zeroreqbyte = reqbyte;
6150 /* If the character is more than one byte long, we can set firstbyte
6151 only if it is not to be matched caselessly. */
6153 if (mclength == 1 || req_caseopt == 0)
6155 firstbyte = mcbuffer[0] | req_caseopt;
6156 if (mclength != 1) reqbyte = code[-1] | cd->req_varyopt;
6158 else firstbyte = reqbyte = REQ_NONE;
6161 /* firstbyte was previously set; we can set reqbyte only the length is
6162 1 or the matching is caseful. */
6166 zerofirstbyte = firstbyte;
6167 zeroreqbyte = reqbyte;
6168 if (mclength == 1 || req_caseopt == 0)
6169 reqbyte = code[-1] | req_caseopt | cd->req_varyopt;
6172 break; /* End of literal character handling */
6174 } /* end of big loop */
6177 /* Control never reaches here by falling through, only by a goto for all the
6178 error states. Pass back the position in the pattern so that it can be displayed
6179 to the user for diagnosing the error. */
6189 /*************************************************
6190 * Compile sequence of alternatives *
6191 *************************************************/
6193 /* On entry, ptr is pointing past the bracket character, but on return it
6194 points to the closing bracket, or vertical bar, or end of string. The code
6195 variable is pointing at the byte into which the BRA operator has been stored.
6196 If the ims options are changed at the start (for a (?ims: group) or during any
6197 branch, we need to insert an OP_OPT item at the start of every following branch
6198 to ensure they get set correctly at run time, and also pass the new options
6199 into every subsequent branch compile.
6201 This function is used during the pre-compile phase when we are trying to find
6202 out the amount of memory needed, as well as during the real compile phase. The
6203 value of lengthptr distinguishes the two phases.
6206 options option bits, including any changes for this subpattern
6207 oldims previous settings of ims option bits
6208 codeptr -> the address of the current code pointer
6209 ptrptr -> the address of the current pattern pointer
6210 errorcodeptr -> pointer to error code variable
6211 lookbehind TRUE if this is a lookbehind assertion
6212 reset_bracount TRUE to reset the count for each branch
6213 skipbytes skip this many bytes at start (for brackets and OP_COND)
6214 firstbyteptr place to put the first required character, or a negative number
6215 reqbyteptr place to put the last required character, or a negative number
6216 bcptr pointer to the chain of currently open branches
6217 cd points to the data block with tables pointers etc.
6218 lengthptr NULL during the real compile phase
6219 points to length accumulator during pre-compile phase
6221 Returns: TRUE on success
6225 compile_regex(int options, int oldims, uschar **codeptr, const uschar **ptrptr,
6226 int *errorcodeptr, BOOL lookbehind, BOOL reset_bracount, int skipbytes,
6227 int *firstbyteptr, int *reqbyteptr, branch_chain *bcptr, compile_data *cd,
6230 const uschar *ptr = *ptrptr;
6231 uschar *code = *codeptr;
6232 uschar *last_branch = code;
6233 uschar *start_bracket = code;
6234 uschar *reverse_count = NULL;
6235 open_capitem capitem;
6237 int firstbyte, reqbyte;
6238 int branchfirstbyte, branchreqbyte;
6242 int old_external_options = cd->external_options;
6246 bc.current_branch = code;
6248 firstbyte = reqbyte = REQ_UNSET;
6250 /* Accumulate the length for use in the pre-compile phase. Start with the
6251 length of the BRA and KET and any extra bytes that are required at the
6252 beginning. We accumulate in a local variable to save frequent testing of
6253 lenthptr for NULL. We cannot do this by looking at the value of code at the
6254 start and end of each alternative, because compiled items are discarded during
6255 the pre-compile phase so that the work space is not exceeded. */
6257 length = 2 + 2*LINK_SIZE + skipbytes;
6259 /* WARNING: If the above line is changed for any reason, you must also change
6260 the code that abstracts option settings at the start of the pattern and makes
6261 them global. It tests the value of length for (2 + 2*LINK_SIZE) in the
6262 pre-compile phase to find out whether anything has yet been compiled or not. */
6264 /* If this is a capturing subpattern, add to the chain of open capturing items
6265 so that we can detect them if (*ACCEPT) is encountered. This is also used to
6266 detect groups that contain recursive back references to themselves. */
6268 if (*code == OP_CBRA)
6270 capnumber = GET2(code, 1 + LINK_SIZE);
6271 capitem.number = capnumber;
6272 capitem.next = cd->open_caps;
6273 capitem.flag = FALSE;
6274 cd->open_caps = &capitem;
6277 /* Offset is set zero to mark that this bracket is still open */
6280 code += 1 + LINK_SIZE + skipbytes;
6282 /* Loop for each alternative branch */
6284 orig_bracount = max_bracount = cd->bracount;
6287 /* For a (?| group, reset the capturing bracket count so that each branch
6288 uses the same numbers. */
6290 if (reset_bracount) cd->bracount = orig_bracount;
6292 /* Handle a change of ims options at the start of the branch */
6294 if ((options & PCRE_IMS) != oldims)
6297 *code++ = options & PCRE_IMS;
6301 /* Set up dummy OP_REVERSE if lookbehind assertion */
6305 *code++ = OP_REVERSE;
6306 reverse_count = code;
6308 length += 1 + LINK_SIZE;
6311 /* Now compile the branch; in the pre-compile phase its length gets added
6314 if (!compile_branch(&options, &code, &ptr, errorcodeptr, &branchfirstbyte,
6315 &branchreqbyte, &bc, cd, (lengthptr == NULL)? NULL : &length))
6321 /* If the external options have changed during this branch, it means that we
6322 are at the top level, and a leading option setting has been encountered. We
6323 need to re-set the original option values to take account of this so that,
6324 during the pre-compile phase, we know to allow for a re-set at the start of
6325 subsequent branches. */
6327 if (old_external_options != cd->external_options)
6328 oldims = cd->external_options & PCRE_IMS;
6330 /* Keep the highest bracket count in case (?| was used and some branch
6331 has fewer than the rest. */
6333 if (cd->bracount > max_bracount) max_bracount = cd->bracount;
6335 /* In the real compile phase, there is some post-processing to be done. */
6337 if (lengthptr == NULL)
6339 /* If this is the first branch, the firstbyte and reqbyte values for the
6340 branch become the values for the regex. */
6342 if (*last_branch != OP_ALT)
6344 firstbyte = branchfirstbyte;
6345 reqbyte = branchreqbyte;
6348 /* If this is not the first branch, the first char and reqbyte have to
6349 match the values from all the previous branches, except that if the
6350 previous value for reqbyte didn't have REQ_VARY set, it can still match,
6351 and we set REQ_VARY for the regex. */
6355 /* If we previously had a firstbyte, but it doesn't match the new branch,
6356 we have to abandon the firstbyte for the regex, but if there was
6357 previously no reqbyte, it takes on the value of the old firstbyte. */
6359 if (firstbyte >= 0 && firstbyte != branchfirstbyte)
6361 if (reqbyte < 0) reqbyte = firstbyte;
6362 firstbyte = REQ_NONE;
6365 /* If we (now or from before) have no firstbyte, a firstbyte from the
6366 branch becomes a reqbyte if there isn't a branch reqbyte. */
6368 if (firstbyte < 0 && branchfirstbyte >= 0 && branchreqbyte < 0)
6369 branchreqbyte = branchfirstbyte;
6371 /* Now ensure that the reqbytes match */
6373 if ((reqbyte & ~REQ_VARY) != (branchreqbyte & ~REQ_VARY))
6375 else reqbyte |= branchreqbyte; /* To "or" REQ_VARY */
6378 /* If lookbehind, check that this branch matches a fixed-length string, and
6379 put the length into the OP_REVERSE item. Temporarily mark the end of the
6380 branch with OP_END. If the branch contains OP_RECURSE, the result is -3
6381 because there may be forward references that we can't check here. Set a
6382 flag to cause another lookbehind check at the end. Why not do it all at the
6383 end? Because common, erroneous checks are picked up here and the offset of
6384 the problem can be shown. */
6390 fixed_length = find_fixedlength(last_branch, options, FALSE, cd);
6391 DPRINTF(("fixed length = %d\n", fixed_length));
6392 if (fixed_length == -3)
6394 cd->check_lookbehind = TRUE;
6396 else if (fixed_length < 0)
6398 *errorcodeptr = (fixed_length == -2)? ERR36 : ERR25;
6402 else { PUT(reverse_count, 0, fixed_length); }
6406 /* Reached end of expression, either ')' or end of pattern. In the real
6407 compile phase, go back through the alternative branches and reverse the chain
6408 of offsets, with the field in the BRA item now becoming an offset to the
6409 first alternative. If there are no alternatives, it points to the end of the
6410 group. The length in the terminating ket is always the length of the whole
6411 bracketed item. If any of the ims options were changed inside the group,
6412 compile a resetting op-code following, except at the very end of the pattern.
6413 Return leaving the pointer at the terminating char. */
6415 if (*ptr != CHAR_VERTICAL_LINE)
6417 if (lengthptr == NULL)
6419 int branch_length = (int)(code - last_branch);
6422 int prev_length = GET(last_branch, 1);
6423 PUT(last_branch, 1, branch_length);
6424 branch_length = prev_length;
6425 last_branch -= branch_length;
6427 while (branch_length > 0);
6430 /* Fill in the ket */
6433 PUT(code, 1, (int)(code - start_bracket));
6434 code += 1 + LINK_SIZE;
6436 /* If it was a capturing subpattern, check to see if it contained any
6437 recursive back references. If so, we must wrap it in atomic brackets.
6438 In any event, remove the block from the chain. */
6442 if (cd->open_caps->flag)
6444 memmove(start_bracket + 1 + LINK_SIZE, start_bracket,
6445 code - start_bracket);
6446 *start_bracket = OP_ONCE;
6447 code += 1 + LINK_SIZE;
6448 PUT(start_bracket, 1, (int)(code - start_bracket));
6450 PUT(code, 1, (int)(code - start_bracket));
6451 code += 1 + LINK_SIZE;
6452 length += 2 + 2*LINK_SIZE;
6454 cd->open_caps = cd->open_caps->next;
6457 /* Reset options if needed. */
6459 if ((options & PCRE_IMS) != oldims && *ptr == CHAR_RIGHT_PARENTHESIS)
6466 /* Retain the highest bracket number, in case resetting was used. */
6468 cd->bracount = max_bracount;
6470 /* Set values to pass back */
6474 *firstbyteptr = firstbyte;
6475 *reqbyteptr = reqbyte;
6476 if (lengthptr != NULL)
6478 if (OFLOW_MAX - *lengthptr < length)
6480 *errorcodeptr = ERR20;
6483 *lengthptr += length;
6488 /* Another branch follows. In the pre-compile phase, we can move the code
6489 pointer back to where it was for the start of the first branch. (That is,
6490 pretend that each branch is the only one.)
6492 In the real compile phase, insert an ALT node. Its length field points back
6493 to the previous branch while the bracket remains open. At the end the chain
6494 is reversed. It's done like this so that the start of the bracket has a
6495 zero offset until it is closed, making it possible to detect recursion. */
6497 if (lengthptr != NULL)
6499 code = *codeptr + 1 + LINK_SIZE + skipbytes;
6500 length += 1 + LINK_SIZE;
6505 PUT(code, 1, (int)(code - last_branch));
6506 bc.current_branch = last_branch = code;
6507 code += 1 + LINK_SIZE;
6512 /* Control never reaches here */
6518 /*************************************************
6519 * Check for anchored expression *
6520 *************************************************/
6522 /* Try to find out if this is an anchored regular expression. Consider each
6523 alternative branch. If they all start with OP_SOD or OP_CIRC, or with a bracket
6524 all of whose alternatives start with OP_SOD or OP_CIRC (recurse ad lib), then
6525 it's anchored. However, if this is a multiline pattern, then only OP_SOD
6526 counts, since OP_CIRC can match in the middle.
6528 We can also consider a regex to be anchored if OP_SOM starts all its branches.
6529 This is the code for \G, which means "match at start of match position, taking
6530 into account the match offset".
6532 A branch is also implicitly anchored if it starts with .* and DOTALL is set,
6533 because that will try the rest of the pattern at all possible matching points,
6534 so there is no point trying again.... er ....
6536 .... except when the .* appears inside capturing parentheses, and there is a
6537 subsequent back reference to those parentheses. We haven't enough information
6538 to catch that case precisely.
6540 At first, the best we could do was to detect when .* was in capturing brackets
6541 and the highest back reference was greater than or equal to that level.
6542 However, by keeping a bitmap of the first 31 back references, we can catch some
6543 of the more common cases more precisely.
6546 code points to start of expression (the bracket)
6547 options points to the options setting
6548 bracket_map a bitmap of which brackets we are inside while testing; this
6549 handles up to substring 31; after that we just have to take
6550 the less precise approach
6551 backref_map the back reference bitmap
6553 Returns: TRUE or FALSE
6557 is_anchored(register const uschar *code, int *options, unsigned int bracket_map,
6558 unsigned int backref_map)
6561 const uschar *scode = first_significant_code(code + _pcre_OP_lengths[*code],
6562 options, PCRE_MULTILINE, FALSE);
6563 register int op = *scode;
6565 /* Non-capturing brackets */
6569 if (!is_anchored(scode, options, bracket_map, backref_map)) return FALSE;
6572 /* Capturing brackets */
6574 else if (op == OP_CBRA)
6576 int n = GET2(scode, 1+LINK_SIZE);
6577 int new_map = bracket_map | ((n < 32)? (1 << n) : 1);
6578 if (!is_anchored(scode, options, new_map, backref_map)) return FALSE;
6581 /* Other brackets */
6583 else if (op == OP_ASSERT || op == OP_ONCE || op == OP_COND)
6585 if (!is_anchored(scode, options, bracket_map, backref_map)) return FALSE;
6588 /* .* is not anchored unless DOTALL is set (which generates OP_ALLANY) and
6589 it isn't in brackets that are or may be referenced. */
6591 else if ((op == OP_TYPESTAR || op == OP_TYPEMINSTAR ||
6592 op == OP_TYPEPOSSTAR))
6594 if (scode[1] != OP_ALLANY || (bracket_map & backref_map) != 0)
6598 /* Check for explicit anchoring */
6600 else if (op != OP_SOD && op != OP_SOM &&
6601 ((*options & PCRE_MULTILINE) != 0 || op != OP_CIRC))
6603 code += GET(code, 1);
6605 while (*code == OP_ALT); /* Loop for each alternative */
6611 /*************************************************
6612 * Check for starting with ^ or .* *
6613 *************************************************/
6615 /* This is called to find out if every branch starts with ^ or .* so that
6616 "first char" processing can be done to speed things up in multiline
6617 matching and for non-DOTALL patterns that start with .* (which must start at
6618 the beginning or after \n). As in the case of is_anchored() (see above), we
6619 have to take account of back references to capturing brackets that contain .*
6620 because in that case we can't make the assumption.
6623 code points to start of expression (the bracket)
6624 bracket_map a bitmap of which brackets we are inside while testing; this
6625 handles up to substring 31; after that we just have to take
6626 the less precise approach
6627 backref_map the back reference bitmap
6629 Returns: TRUE or FALSE
6633 is_startline(const uschar *code, unsigned int bracket_map,
6634 unsigned int backref_map)
6637 const uschar *scode = first_significant_code(code + _pcre_OP_lengths[*code],
6639 register int op = *scode;
6641 /* If we are at the start of a conditional assertion group, *both* the
6642 conditional assertion *and* what follows the condition must satisfy the test
6643 for start of line. Other kinds of condition fail. Note that there may be an
6644 auto-callout at the start of a condition. */
6648 scode += 1 + LINK_SIZE;
6649 if (*scode == OP_CALLOUT) scode += _pcre_OP_lengths[OP_CALLOUT];
6659 default: /* Assertion */
6660 if (!is_startline(scode, bracket_map, backref_map)) return FALSE;
6661 do scode += GET(scode, 1); while (*scode == OP_ALT);
6662 scode += 1 + LINK_SIZE;
6665 scode = first_significant_code(scode, NULL, 0, FALSE);
6669 /* Non-capturing brackets */
6673 if (!is_startline(scode, bracket_map, backref_map)) return FALSE;
6676 /* Capturing brackets */
6678 else if (op == OP_CBRA)
6680 int n = GET2(scode, 1+LINK_SIZE);
6681 int new_map = bracket_map | ((n < 32)? (1 << n) : 1);
6682 if (!is_startline(scode, new_map, backref_map)) return FALSE;
6685 /* Other brackets */
6687 else if (op == OP_ASSERT || op == OP_ONCE)
6689 if (!is_startline(scode, bracket_map, backref_map)) return FALSE;
6692 /* .* means "start at start or after \n" if it isn't in brackets that
6693 may be referenced. */
6695 else if (op == OP_TYPESTAR || op == OP_TYPEMINSTAR || op == OP_TYPEPOSSTAR)
6697 if (scode[1] != OP_ANY || (bracket_map & backref_map) != 0) return FALSE;
6700 /* Check for explicit circumflex */
6702 else if (op != OP_CIRC) return FALSE;
6704 /* Move on to the next alternative */
6706 code += GET(code, 1);
6708 while (*code == OP_ALT); /* Loop for each alternative */
6714 /*************************************************
6715 * Check for asserted fixed first char *
6716 *************************************************/
6718 /* During compilation, the "first char" settings from forward assertions are
6719 discarded, because they can cause conflicts with actual literals that follow.
6720 However, if we end up without a first char setting for an unanchored pattern,
6721 it is worth scanning the regex to see if there is an initial asserted first
6722 char. If all branches start with the same asserted char, or with a bracket all
6723 of whose alternatives start with the same asserted char (recurse ad lib), then
6724 we return that char, otherwise -1.
6727 code points to start of expression (the bracket)
6728 options pointer to the options (used to check casing changes)
6729 inassert TRUE if in an assertion
6731 Returns: -1 or the fixed first char
6735 find_firstassertedchar(const uschar *code, int *options, BOOL inassert)
6737 register int c = -1;
6740 const uschar *scode =
6741 first_significant_code(code + 1+LINK_SIZE, options, PCRE_CASELESS, TRUE);
6742 register int op = *scode;
6754 if ((d = find_firstassertedchar(scode, options, op == OP_ASSERT)) < 0)
6756 if (c < 0) c = d; else if (c != d) return -1;
6759 case OP_EXACT: /* Fall through */
6767 if (!inassert) return -1;
6771 if ((*options & PCRE_CASELESS) != 0) c |= REQ_CASELESS;
6773 else if (c != scode[1]) return -1;
6777 code += GET(code, 1);
6779 while (*code == OP_ALT);
6785 /*************************************************
6786 * Compile a Regular Expression *
6787 *************************************************/
6789 /* This function takes a string and returns a pointer to a block of store
6790 holding a compiled version of the expression. The original API for this
6791 function had no error code return variable; it is retained for backwards
6792 compatibility. The new function is given a new name.
6795 pattern the regular expression
6796 options various option bits
6797 errorcodeptr pointer to error code variable (pcre_compile2() only)
6798 can be NULL if you don't want a code value
6799 errorptr pointer to pointer to error text
6800 erroroffset ptr offset in pattern where error was detected
6801 tables pointer to character tables or NULL
6803 Returns: pointer to compiled data block, or NULL on error,
6804 with errorptr and erroroffset set
6807 PCRE_EXP_DEFN pcre * PCRE_CALL_CONVENTION
6808 pcre_compile(const char *pattern, int options, const char **errorptr,
6809 int *erroroffset, const unsigned char *tables)
6811 return pcre_compile2(pattern, options, NULL, errorptr, erroroffset, tables);
6815 PCRE_EXP_DEFN pcre * PCRE_CALL_CONVENTION
6816 pcre_compile2(const char *pattern, int options, int *errorcodeptr,
6817 const char **errorptr, int *erroroffset, const unsigned char *tables)
6820 int length = 1; /* For final END opcode */
6821 int firstbyte, reqbyte, newline;
6823 int skipatstart = 0;
6827 const uschar *codestart;
6829 compile_data compile_block;
6830 compile_data *cd = &compile_block;
6832 /* This space is used for "compiling" into during the first phase, when we are
6833 computing the amount of memory that is needed. Compiled items are thrown away
6834 as soon as possible, so that a fairly large buffer should be sufficient for
6835 this purpose. The same space is used in the second phase for remembering where
6836 to fill in forward references to subpatterns. */
6838 uschar cworkspace[COMPILE_WORK_SIZE];
6840 /* Set this early so that early errors get offset 0. */
6842 ptr = (const uschar *)pattern;
6844 /* We can't pass back an error message if errorptr is NULL; I guess the best we
6845 can do is just return NULL, but we can set a code value if there is a code
6848 if (errorptr == NULL)
6850 if (errorcodeptr != NULL) *errorcodeptr = 99;
6855 if (errorcodeptr != NULL) *errorcodeptr = ERR0;
6857 /* However, we can give a message for this error */
6859 if (erroroffset == NULL)
6862 goto PCRE_EARLY_ERROR_RETURN2;
6867 /* Set up pointers to the individual character tables */
6869 if (tables == NULL) tables = _pcre_default_tables;
6870 cd->lcc = tables + lcc_offset;
6871 cd->fcc = tables + fcc_offset;
6872 cd->cbits = tables + cbits_offset;
6873 cd->ctypes = tables + ctypes_offset;
6875 /* Check that all undefined public option bits are zero */
6877 if ((options & ~PUBLIC_COMPILE_OPTIONS) != 0)
6880 goto PCRE_EARLY_ERROR_RETURN;
6883 /* Check for global one-time settings at the start of the pattern, and remember
6884 the offset for later. */
6886 while (ptr[skipatstart] == CHAR_LEFT_PARENTHESIS &&
6887 ptr[skipatstart+1] == CHAR_ASTERISK)
6892 if (strncmp((char *)(ptr+skipatstart+2), STRING_UTF8_RIGHTPAR, 5) == 0)
6893 { skipatstart += 7; options |= PCRE_UTF8; continue; }
6894 else if (strncmp((char *)(ptr+skipatstart+2), STRING_UCP_RIGHTPAR, 4) == 0)
6895 { skipatstart += 6; options |= PCRE_UCP; continue; }
6896 else if (strncmp((char *)(ptr+skipatstart+2), STRING_NO_START_OPT_RIGHTPAR, 13) == 0)
6897 { skipatstart += 15; options |= PCRE_NO_START_OPTIMIZE; continue; }
6899 if (strncmp((char *)(ptr+skipatstart+2), STRING_CR_RIGHTPAR, 3) == 0)
6900 { skipatstart += 5; newnl = PCRE_NEWLINE_CR; }
6901 else if (strncmp((char *)(ptr+skipatstart+2), STRING_LF_RIGHTPAR, 3) == 0)
6902 { skipatstart += 5; newnl = PCRE_NEWLINE_LF; }
6903 else if (strncmp((char *)(ptr+skipatstart+2), STRING_CRLF_RIGHTPAR, 5) == 0)
6904 { skipatstart += 7; newnl = PCRE_NEWLINE_CR + PCRE_NEWLINE_LF; }
6905 else if (strncmp((char *)(ptr+skipatstart+2), STRING_ANY_RIGHTPAR, 4) == 0)
6906 { skipatstart += 6; newnl = PCRE_NEWLINE_ANY; }
6907 else if (strncmp((char *)(ptr+skipatstart+2), STRING_ANYCRLF_RIGHTPAR, 8) == 0)
6908 { skipatstart += 10; newnl = PCRE_NEWLINE_ANYCRLF; }
6910 else if (strncmp((char *)(ptr+skipatstart+2), STRING_BSR_ANYCRLF_RIGHTPAR, 12) == 0)
6911 { skipatstart += 14; newbsr = PCRE_BSR_ANYCRLF; }
6912 else if (strncmp((char *)(ptr+skipatstart+2), STRING_BSR_UNICODE_RIGHTPAR, 12) == 0)
6913 { skipatstart += 14; newbsr = PCRE_BSR_UNICODE; }
6916 options = (options & ~PCRE_NEWLINE_BITS) | newnl;
6917 else if (newbsr != 0)
6918 options = (options & ~(PCRE_BSR_ANYCRLF|PCRE_BSR_UNICODE)) | newbsr;
6922 utf8 = (options & PCRE_UTF8) != 0;
6924 /* Can't support UTF8 unless PCRE has been compiled to include the code. */
6927 if (utf8 && (options & PCRE_NO_UTF8_CHECK) == 0 &&
6928 (*erroroffset = _pcre_valid_utf8((USPTR)pattern, -1)) >= 0)
6931 goto PCRE_EARLY_ERROR_RETURN2;
6937 goto PCRE_EARLY_ERROR_RETURN;
6941 /* Can't support UCP unless PCRE has been compiled to include the code. */
6944 if ((options & PCRE_UCP) != 0)
6947 goto PCRE_EARLY_ERROR_RETURN;
6951 /* Check validity of \R options. */
6953 switch (options & (PCRE_BSR_ANYCRLF|PCRE_BSR_UNICODE))
6956 case PCRE_BSR_ANYCRLF:
6957 case PCRE_BSR_UNICODE:
6959 default: errorcode = ERR56; goto PCRE_EARLY_ERROR_RETURN;
6962 /* Handle different types of newline. The three bits give seven cases. The
6963 current code allows for fixed one- or two-byte sequences, plus "any" and
6966 switch (options & PCRE_NEWLINE_BITS)
6968 case 0: newline = NEWLINE; break; /* Build-time default */
6969 case PCRE_NEWLINE_CR: newline = CHAR_CR; break;
6970 case PCRE_NEWLINE_LF: newline = CHAR_NL; break;
6971 case PCRE_NEWLINE_CR+
6972 PCRE_NEWLINE_LF: newline = (CHAR_CR << 8) | CHAR_NL; break;
6973 case PCRE_NEWLINE_ANY: newline = -1; break;
6974 case PCRE_NEWLINE_ANYCRLF: newline = -2; break;
6975 default: errorcode = ERR56; goto PCRE_EARLY_ERROR_RETURN;
6980 cd->nltype = NLTYPE_ANYCRLF;
6982 else if (newline < 0)
6984 cd->nltype = NLTYPE_ANY;
6988 cd->nltype = NLTYPE_FIXED;
6992 cd->nl[0] = (newline >> 8) & 255;
6993 cd->nl[1] = newline & 255;
6998 cd->nl[0] = newline;
7002 /* Maximum back reference and backref bitmap. The bitmap records up to 31 back
7003 references to help in deciding whether (.*) can be treated as anchored or not.
7006 cd->top_backref = 0;
7007 cd->backref_map = 0;
7009 /* Reflect pattern for debugging output */
7011 DPRINTF(("------------------------------------------------------------------\n"));
7012 DPRINTF(("%s\n", pattern));
7014 /* Pretend to compile the pattern while actually just accumulating the length
7015 of memory required. This behaviour is triggered by passing a non-NULL final
7016 argument to compile_regex(). We pass a block of workspace (cworkspace) for it
7017 to compile parts of the pattern into; the compiled code is discarded when it is
7018 no longer needed, so hopefully this workspace will never overflow, though there
7019 is a test for its doing so. */
7021 cd->bracount = cd->final_bracount = 0;
7022 cd->names_found = 0;
7023 cd->name_entry_size = 0;
7024 cd->name_table = NULL;
7025 cd->start_workspace = cworkspace;
7026 cd->start_code = cworkspace;
7027 cd->hwm = cworkspace;
7028 cd->start_pattern = (const uschar *)pattern;
7029 cd->end_pattern = (const uschar *)(pattern + strlen(pattern));
7030 cd->req_varyopt = 0;
7031 cd->external_options = options;
7032 cd->external_flags = 0;
7033 cd->open_caps = NULL;
7035 /* Now do the pre-compile. On error, errorcode will be set non-zero, so we
7036 don't need to look at the result of the function here. The initial options have
7037 been put into the cd block so that they can be changed if an option setting is
7038 found within the regex right at the beginning. Bringing initial option settings
7039 outside can help speed up starting point checks. */
7044 (void)compile_regex(cd->external_options, cd->external_options & PCRE_IMS,
7045 &code, &ptr, &errorcode, FALSE, FALSE, 0, &firstbyte, &reqbyte, NULL, cd,
7047 if (errorcode != 0) goto PCRE_EARLY_ERROR_RETURN;
7049 DPRINTF(("end pre-compile: length=%d workspace=%d\n", length,
7050 cd->hwm - cworkspace));
7052 if (length > MAX_PATTERN_SIZE)
7055 goto PCRE_EARLY_ERROR_RETURN;
7058 /* Compute the size of data block needed and get it, either from malloc or
7059 externally provided function. Integer overflow should no longer be possible
7060 because nowadays we limit the maximum value of cd->names_found and
7061 cd->name_entry_size. */
7063 size = length + sizeof(real_pcre) + cd->names_found * (cd->name_entry_size + 3);
7064 re = (real_pcre *)(pcre_malloc)(size);
7069 goto PCRE_EARLY_ERROR_RETURN;
7072 /* Put in the magic number, and save the sizes, initial options, internal
7073 flags, and character table pointer. NULL is used for the default character
7074 tables. The nullpad field is at the end; it's there to help in the case when a
7075 regex compiled on a system with 4-byte pointers is run on another with 8-byte
7078 re->magic_number = MAGIC_NUMBER;
7079 re->size = (int)size;
7080 re->options = cd->external_options;
7081 re->flags = cd->external_flags;
7085 re->name_table_offset = sizeof(real_pcre);
7086 re->name_entry_size = cd->name_entry_size;
7087 re->name_count = cd->names_found;
7089 re->tables = (tables == _pcre_default_tables)? NULL : tables;
7092 /* The starting points of the name/number translation table and of the code are
7093 passed around in the compile data block. The start/end pattern and initial
7094 options are already set from the pre-compile phase, as is the name_entry_size
7095 field. Reset the bracket count and the names_found field. Also reset the hwm
7096 field; this time it's used for remembering forward references to subpatterns.
7099 cd->final_bracount = cd->bracount; /* Save for checking forward references */
7101 cd->names_found = 0;
7102 cd->name_table = (uschar *)re + re->name_table_offset;
7103 codestart = cd->name_table + re->name_entry_size * re->name_count;
7104 cd->start_code = codestart;
7105 cd->hwm = cworkspace;
7106 cd->req_varyopt = 0;
7107 cd->had_accept = FALSE;
7108 cd->check_lookbehind = FALSE;
7109 cd->open_caps = NULL;
7111 /* Set up a starting, non-extracting bracket, then compile the expression. On
7112 error, errorcode will be set non-zero, so we don't need to look at the result
7113 of the function here. */
7115 ptr = (const uschar *)pattern + skipatstart;
7116 code = (uschar *)codestart;
7118 (void)compile_regex(re->options, re->options & PCRE_IMS, &code, &ptr,
7119 &errorcode, FALSE, FALSE, 0, &firstbyte, &reqbyte, NULL, cd, NULL);
7120 re->top_bracket = cd->bracount;
7121 re->top_backref = cd->top_backref;
7122 re->flags = cd->external_flags;
7124 if (cd->had_accept) reqbyte = -1; /* Must disable after (*ACCEPT) */
7126 /* If not reached end of pattern on success, there's an excess bracket. */
7128 if (errorcode == 0 && *ptr != 0) errorcode = ERR22;
7130 /* Fill in the terminating state and check for disastrous overflow, but
7131 if debugging, leave the test till after things are printed out. */
7136 if (code - codestart > length) errorcode = ERR23;
7139 /* Fill in any forward references that are required. */
7141 while (errorcode == 0 && cd->hwm > cworkspace)
7144 const uschar *groupptr;
7145 cd->hwm -= LINK_SIZE;
7146 offset = GET(cd->hwm, 0);
7147 recno = GET(codestart, offset);
7148 groupptr = _pcre_find_bracket(codestart, utf8, recno);
7149 if (groupptr == NULL) errorcode = ERR53;
7150 else PUT(((uschar *)codestart), offset, (int)(groupptr - codestart));
7153 /* Give an error if there's back reference to a non-existent capturing
7156 if (errorcode == 0 && re->top_backref > re->top_bracket) errorcode = ERR15;
7158 /* If there were any lookbehind assertions that contained OP_RECURSE
7159 (recursions or subroutine calls), a flag is set for them to be checked here,
7160 because they may contain forward references. Actual recursions can't be fixed
7161 length, but subroutine calls can. It is done like this so that those without
7162 OP_RECURSE that are not fixed length get a diagnosic with a useful offset. The
7163 exceptional ones forgo this. We scan the pattern to check that they are fixed
7164 length, and set their lengths. */
7166 if (cd->check_lookbehind)
7168 uschar *cc = (uschar *)codestart;
7170 /* Loop, searching for OP_REVERSE items, and process those that do not have
7171 their length set. (Actually, it will also re-process any that have a length
7172 of zero, but that is a pathological case, and it does no harm.) When we find
7173 one, we temporarily terminate the branch it is in while we scan it. */
7175 for (cc = (uschar *)_pcre_find_bracket(codestart, utf8, -1);
7177 cc = (uschar *)_pcre_find_bracket(cc, utf8, -1))
7179 if (GET(cc, 1) == 0)
7182 uschar *be = cc - 1 - LINK_SIZE + GET(cc, -LINK_SIZE);
7185 fixed_length = find_fixedlength(cc, re->options, TRUE, cd);
7187 DPRINTF(("fixed length = %d\n", fixed_length));
7188 if (fixed_length < 0)
7190 errorcode = (fixed_length == -2)? ERR36 : ERR25;
7193 PUT(cc, 1, fixed_length);
7195 cc += 1 + LINK_SIZE;
7199 /* Failed to compile, or error while post-processing */
7204 PCRE_EARLY_ERROR_RETURN:
7205 *erroroffset = (int)(ptr - (const uschar *)pattern);
7206 PCRE_EARLY_ERROR_RETURN2:
7207 *errorptr = find_error_text(errorcode);
7208 if (errorcodeptr != NULL) *errorcodeptr = errorcode;
7212 /* If the anchored option was not passed, set the flag if we can determine that
7213 the pattern is anchored by virtue of ^ characters or \A or anything else (such
7214 as starting with .* when DOTALL is set).
7216 Otherwise, if we know what the first byte has to be, save it, because that
7217 speeds up unanchored matches no end. If not, see if we can set the
7218 PCRE_STARTLINE flag. This is helpful for multiline matches when all branches
7219 start with ^. and also when all branches start with .* for non-DOTALL matches.
7222 if ((re->options & PCRE_ANCHORED) == 0)
7224 int temp_options = re->options; /* May get changed during these scans */
7225 if (is_anchored(codestart, &temp_options, 0, cd->backref_map))
7226 re->options |= PCRE_ANCHORED;
7230 firstbyte = find_firstassertedchar(codestart, &temp_options, FALSE);
7231 if (firstbyte >= 0) /* Remove caseless flag for non-caseable chars */
7233 int ch = firstbyte & 255;
7234 re->first_byte = ((firstbyte & REQ_CASELESS) != 0 &&
7235 cd->fcc[ch] == ch)? ch : firstbyte;
7236 re->flags |= PCRE_FIRSTSET;
7238 else if (is_startline(codestart, 0, cd->backref_map))
7239 re->flags |= PCRE_STARTLINE;
7243 /* For an anchored pattern, we use the "required byte" only if it follows a
7244 variable length item in the regex. Remove the caseless flag for non-caseable
7248 ((re->options & PCRE_ANCHORED) == 0 || (reqbyte & REQ_VARY) != 0))
7250 int ch = reqbyte & 255;
7251 re->req_byte = ((reqbyte & REQ_CASELESS) != 0 &&
7252 cd->fcc[ch] == ch)? (reqbyte & ~REQ_CASELESS) : reqbyte;
7253 re->flags |= PCRE_REQCHSET;
7256 /* Print out the compiled data if debugging is enabled. This is never the
7257 case when building a production library. */
7260 printf("Length = %d top_bracket = %d top_backref = %d\n",
7261 length, re->top_bracket, re->top_backref);
7263 printf("Options=%08x\n", re->options);
7265 if ((re->flags & PCRE_FIRSTSET) != 0)
7267 int ch = re->first_byte & 255;
7268 const char *caseless = ((re->first_byte & REQ_CASELESS) == 0)?
7270 if (isprint(ch)) printf("First char = %c%s\n", ch, caseless);
7271 else printf("First char = \\x%02x%s\n", ch, caseless);
7274 if ((re->flags & PCRE_REQCHSET) != 0)
7276 int ch = re->req_byte & 255;
7277 const char *caseless = ((re->req_byte & REQ_CASELESS) == 0)?
7279 if (isprint(ch)) printf("Req char = %c%s\n", ch, caseless);
7280 else printf("Req char = \\x%02x%s\n", ch, caseless);
7283 pcre_printint(re, stdout, TRUE);
7285 /* This check is done here in the debugging case so that the code that
7286 was compiled can be seen. */
7288 if (code - codestart > length)
7291 *errorptr = find_error_text(ERR23);
7292 *erroroffset = ptr - (uschar *)pattern;
7293 if (errorcodeptr != NULL) *errorcodeptr = ERR23;
7296 #endif /* PCRE_DEBUG */
7301 /* End of pcre_compile.c */