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

Contents of /code/trunk/pcre_compile.c

Parent Directory Parent Directory | Revision Log Revision Log


Revision 807 - (show annotations)
Sun Dec 18 10:03:38 2011 UTC (7 years, 9 months ago) by ph10
File MIME type: text/plain
File size: 255196 byte(s)
Renamed isnumber in pcre_compile to avoid a clash with ctype.h in Macs, and 
fixed a bug in fixed-length calculation for lookbehinds that would show up only 
in quite long subpatterns.
1 /*************************************************
2 * Perl-Compatible Regular Expressions *
3 *************************************************/
4
5 /* PCRE is a library of functions to support regular expressions whose syntax
6 and semantics are as close as possible to those of the Perl 5 language.
7
8 Written by Philip Hazel
9 Copyright (c) 1997-2011 University of Cambridge
10
11 -----------------------------------------------------------------------------
12 Redistribution and use in source and binary forms, with or without
13 modification, are permitted provided that the following conditions are met:
14
15 * Redistributions of source code must retain the above copyright notice,
16 this list of conditions and the following disclaimer.
17
18 * Redistributions in binary form must reproduce the above copyright
19 notice, this list of conditions and the following disclaimer in the
20 documentation and/or other materials provided with the distribution.
21
22 * Neither the name of the University of Cambridge nor the names of its
23 contributors may be used to endorse or promote products derived from
24 this software without specific prior written permission.
25
26 THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
27 AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28 IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29 ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE
30 LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
31 CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
32 SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
33 INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
34 CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
35 ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
36 POSSIBILITY OF SUCH DAMAGE.
37 -----------------------------------------------------------------------------
38 */
39
40
41 /* This module contains the external function pcre_compile(), along with
42 supporting internal functions that are not used by other modules. */
43
44
45 #ifdef HAVE_CONFIG_H
46 #include "config.h"
47 #endif
48
49 #define NLBLOCK cd /* Block containing newline information */
50 #define PSSTART start_pattern /* Field containing processed string start */
51 #define PSEND end_pattern /* Field containing processed string end */
52
53 #include "pcre_internal.h"
54
55
56 /* When PCRE_DEBUG is defined, we need the pcre_printint() function, which is
57 also used by pcretest. PCRE_DEBUG is not defined when building a production
58 library. */
59
60 #ifdef PCRE_DEBUG
61 #include "pcre_printint.src"
62 #endif
63
64
65 /* Macro for setting individual bits in class bitmaps. */
66
67 #define SETBIT(a,b) a[b/8] |= (1 << (b%8))
68
69 /* Maximum length value to check against when making sure that the integer that
70 holds the compiled pattern length does not overflow. We make it a bit less than
71 INT_MAX to allow for adding in group terminating bytes, so that we don't have
72 to check them every time. */
73
74 #define OFLOW_MAX (INT_MAX - 20)
75
76
77 /*************************************************
78 * Code parameters and static tables *
79 *************************************************/
80
81 /* This value specifies the size of stack workspace that is used during the
82 first pre-compile phase that determines how much memory is required. The regex
83 is partly compiled into this space, but the compiled parts are discarded as
84 soon as they can be, so that hopefully there will never be an overrun. The code
85 does, however, check for an overrun. The largest amount I've seen used is 218,
86 so this number is very generous.
87
88 The same workspace is used during the second, actual compile phase for
89 remembering forward references to groups so that they can be filled in at the
90 end. Each entry in this list occupies LINK_SIZE bytes, so even when LINK_SIZE
91 is 4 there is plenty of room for most patterns. However, the memory can get
92 filled up by repetitions of forward references, for example patterns like
93 /(?1){0,1999}(b)/, and one user did hit the limit. The code has been changed so
94 that the workspace is expanded using malloc() in this situation. The value
95 below is therefore a minimum, and we put a maximum on it for safety. The
96 minimum is now also defined in terms of LINK_SIZE so that the use of malloc()
97 kicks in at the same number of forward references in all cases. */
98
99 #define COMPILE_WORK_SIZE (2048*LINK_SIZE)
100 #define COMPILE_WORK_SIZE_MAX (100*COMPILE_WORK_SIZE)
101
102 /* The overrun tests check for a slightly smaller size so that they detect the
103 overrun before it actually does run off the end of the data block. */
104
105 #define WORK_SIZE_SAFETY_MARGIN (100)
106
107
108 /* Table for handling escaped characters in the range '0'-'z'. Positive returns
109 are simple data values; negative values are for special things like \d and so
110 on. Zero means further processing is needed (for things like \x), or the escape
111 is invalid. */
112
113 #ifndef EBCDIC
114
115 /* This is the "normal" table for ASCII systems or for EBCDIC systems running
116 in UTF-8 mode. */
117
118 static const short int escapes[] = {
119 0, 0,
120 0, 0,
121 0, 0,
122 0, 0,
123 0, 0,
124 CHAR_COLON, CHAR_SEMICOLON,
125 CHAR_LESS_THAN_SIGN, CHAR_EQUALS_SIGN,
126 CHAR_GREATER_THAN_SIGN, CHAR_QUESTION_MARK,
127 CHAR_COMMERCIAL_AT, -ESC_A,
128 -ESC_B, -ESC_C,
129 -ESC_D, -ESC_E,
130 0, -ESC_G,
131 -ESC_H, 0,
132 0, -ESC_K,
133 0, 0,
134 -ESC_N, 0,
135 -ESC_P, -ESC_Q,
136 -ESC_R, -ESC_S,
137 0, 0,
138 -ESC_V, -ESC_W,
139 -ESC_X, 0,
140 -ESC_Z, CHAR_LEFT_SQUARE_BRACKET,
141 CHAR_BACKSLASH, CHAR_RIGHT_SQUARE_BRACKET,
142 CHAR_CIRCUMFLEX_ACCENT, CHAR_UNDERSCORE,
143 CHAR_GRAVE_ACCENT, 7,
144 -ESC_b, 0,
145 -ESC_d, ESC_e,
146 ESC_f, 0,
147 -ESC_h, 0,
148 0, -ESC_k,
149 0, 0,
150 ESC_n, 0,
151 -ESC_p, 0,
152 ESC_r, -ESC_s,
153 ESC_tee, 0,
154 -ESC_v, -ESC_w,
155 0, 0,
156 -ESC_z
157 };
158
159 #else
160
161 /* This is the "abnormal" table for EBCDIC systems without UTF-8 support. */
162
163 static const short int escapes[] = {
164 /* 48 */ 0, 0, 0, '.', '<', '(', '+', '|',
165 /* 50 */ '&', 0, 0, 0, 0, 0, 0, 0,
166 /* 58 */ 0, 0, '!', '$', '*', ')', ';', '~',
167 /* 60 */ '-', '/', 0, 0, 0, 0, 0, 0,
168 /* 68 */ 0, 0, '|', ',', '%', '_', '>', '?',
169 /* 70 */ 0, 0, 0, 0, 0, 0, 0, 0,
170 /* 78 */ 0, '`', ':', '#', '@', '\'', '=', '"',
171 /* 80 */ 0, 7, -ESC_b, 0, -ESC_d, ESC_e, ESC_f, 0,
172 /* 88 */-ESC_h, 0, 0, '{', 0, 0, 0, 0,
173 /* 90 */ 0, 0, -ESC_k, 'l', 0, ESC_n, 0, -ESC_p,
174 /* 98 */ 0, ESC_r, 0, '}', 0, 0, 0, 0,
175 /* A0 */ 0, '~', -ESC_s, ESC_tee, 0,-ESC_v, -ESC_w, 0,
176 /* A8 */ 0,-ESC_z, 0, 0, 0, '[', 0, 0,
177 /* B0 */ 0, 0, 0, 0, 0, 0, 0, 0,
178 /* B8 */ 0, 0, 0, 0, 0, ']', '=', '-',
179 /* C0 */ '{',-ESC_A, -ESC_B, -ESC_C, -ESC_D,-ESC_E, 0, -ESC_G,
180 /* C8 */-ESC_H, 0, 0, 0, 0, 0, 0, 0,
181 /* D0 */ '}', 0, -ESC_K, 0, 0,-ESC_N, 0, -ESC_P,
182 /* D8 */-ESC_Q,-ESC_R, 0, 0, 0, 0, 0, 0,
183 /* E0 */ '\\', 0, -ESC_S, 0, 0,-ESC_V, -ESC_W, -ESC_X,
184 /* E8 */ 0,-ESC_Z, 0, 0, 0, 0, 0, 0,
185 /* F0 */ 0, 0, 0, 0, 0, 0, 0, 0,
186 /* F8 */ 0, 0, 0, 0, 0, 0, 0, 0
187 };
188 #endif
189
190
191 /* Table of special "verbs" like (*PRUNE). This is a short table, so it is
192 searched linearly. Put all the names into a single string, in order to reduce
193 the number of relocations when a shared library is dynamically linked. The
194 string is built from string macros so that it works in UTF-8 mode on EBCDIC
195 platforms. */
196
197 typedef struct verbitem {
198 int len; /* Length of verb name */
199 int op; /* Op when no arg, or -1 if arg mandatory */
200 int op_arg; /* Op when arg present, or -1 if not allowed */
201 } verbitem;
202
203 static const char verbnames[] =
204 "\0" /* Empty name is a shorthand for MARK */
205 STRING_MARK0
206 STRING_ACCEPT0
207 STRING_COMMIT0
208 STRING_F0
209 STRING_FAIL0
210 STRING_PRUNE0
211 STRING_SKIP0
212 STRING_THEN;
213
214 static const verbitem verbs[] = {
215 { 0, -1, OP_MARK },
216 { 4, -1, OP_MARK },
217 { 6, OP_ACCEPT, -1 },
218 { 6, OP_COMMIT, -1 },
219 { 1, OP_FAIL, -1 },
220 { 4, OP_FAIL, -1 },
221 { 5, OP_PRUNE, OP_PRUNE_ARG },
222 { 4, OP_SKIP, OP_SKIP_ARG },
223 { 4, OP_THEN, OP_THEN_ARG }
224 };
225
226 static const int verbcount = sizeof(verbs)/sizeof(verbitem);
227
228
229 /* Tables of names of POSIX character classes and their lengths. The names are
230 now all in a single string, to reduce the number of relocations when a shared
231 library is dynamically loaded. The list of lengths is terminated by a zero
232 length entry. The first three must be alpha, lower, upper, as this is assumed
233 for handling case independence. */
234
235 static const char posix_names[] =
236 STRING_alpha0 STRING_lower0 STRING_upper0 STRING_alnum0
237 STRING_ascii0 STRING_blank0 STRING_cntrl0 STRING_digit0
238 STRING_graph0 STRING_print0 STRING_punct0 STRING_space0
239 STRING_word0 STRING_xdigit;
240
241 static const uschar posix_name_lengths[] = {
242 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 4, 6, 0 };
243
244 /* Table of class bit maps for each POSIX class. Each class is formed from a
245 base map, with an optional addition or removal of another map. Then, for some
246 classes, there is some additional tweaking: for [:blank:] the vertical space
247 characters are removed, and for [:alpha:] and [:alnum:] the underscore
248 character is removed. The triples in the table consist of the base map offset,
249 second map offset or -1 if no second map, and a non-negative value for map
250 addition or a negative value for map subtraction (if there are two maps). The
251 absolute value of the third field has these meanings: 0 => no tweaking, 1 =>
252 remove vertical space characters, 2 => remove underscore. */
253
254 static const int posix_class_maps[] = {
255 cbit_word, cbit_digit, -2, /* alpha */
256 cbit_lower, -1, 0, /* lower */
257 cbit_upper, -1, 0, /* upper */
258 cbit_word, -1, 2, /* alnum - word without underscore */
259 cbit_print, cbit_cntrl, 0, /* ascii */
260 cbit_space, -1, 1, /* blank - a GNU extension */
261 cbit_cntrl, -1, 0, /* cntrl */
262 cbit_digit, -1, 0, /* digit */
263 cbit_graph, -1, 0, /* graph */
264 cbit_print, -1, 0, /* print */
265 cbit_punct, -1, 0, /* punct */
266 cbit_space, -1, 0, /* space */
267 cbit_word, -1, 0, /* word - a Perl extension */
268 cbit_xdigit,-1, 0 /* xdigit */
269 };
270
271 /* Table of substitutes for \d etc when PCRE_UCP is set. The POSIX class
272 substitutes must be in the order of the names, defined above, and there are
273 both positive and negative cases. NULL means no substitute. */
274
275 #ifdef SUPPORT_UCP
276 static const uschar *substitutes[] = {
277 (uschar *)"\\P{Nd}", /* \D */
278 (uschar *)"\\p{Nd}", /* \d */
279 (uschar *)"\\P{Xsp}", /* \S */ /* NOTE: Xsp is Perl space */
280 (uschar *)"\\p{Xsp}", /* \s */
281 (uschar *)"\\P{Xwd}", /* \W */
282 (uschar *)"\\p{Xwd}" /* \w */
283 };
284
285 static const uschar *posix_substitutes[] = {
286 (uschar *)"\\p{L}", /* alpha */
287 (uschar *)"\\p{Ll}", /* lower */
288 (uschar *)"\\p{Lu}", /* upper */
289 (uschar *)"\\p{Xan}", /* alnum */
290 NULL, /* ascii */
291 (uschar *)"\\h", /* blank */
292 NULL, /* cntrl */
293 (uschar *)"\\p{Nd}", /* digit */
294 NULL, /* graph */
295 NULL, /* print */
296 NULL, /* punct */
297 (uschar *)"\\p{Xps}", /* space */ /* NOTE: Xps is POSIX space */
298 (uschar *)"\\p{Xwd}", /* word */
299 NULL, /* xdigit */
300 /* Negated cases */
301 (uschar *)"\\P{L}", /* ^alpha */
302 (uschar *)"\\P{Ll}", /* ^lower */
303 (uschar *)"\\P{Lu}", /* ^upper */
304 (uschar *)"\\P{Xan}", /* ^alnum */
305 NULL, /* ^ascii */
306 (uschar *)"\\H", /* ^blank */
307 NULL, /* ^cntrl */
308 (uschar *)"\\P{Nd}", /* ^digit */
309 NULL, /* ^graph */
310 NULL, /* ^print */
311 NULL, /* ^punct */
312 (uschar *)"\\P{Xps}", /* ^space */ /* NOTE: Xps is POSIX space */
313 (uschar *)"\\P{Xwd}", /* ^word */
314 NULL /* ^xdigit */
315 };
316 #define POSIX_SUBSIZE (sizeof(posix_substitutes)/sizeof(uschar *))
317 #endif
318
319 #define STRING(a) # a
320 #define XSTRING(s) STRING(s)
321
322 /* The texts of compile-time error messages. These are "char *" because they
323 are passed to the outside world. Do not ever re-use any error number, because
324 they are documented. Always add a new error instead. Messages marked DEAD below
325 are no longer used. This used to be a table of strings, but in order to reduce
326 the number of relocations needed when a shared library is loaded dynamically,
327 it is now one long string. We cannot use a table of offsets, because the
328 lengths of inserts such as XSTRING(MAX_NAME_SIZE) are not known. Instead, we
329 simply count through to the one we want - this isn't a performance issue
330 because these strings are used only when there is a compilation error.
331
332 Each substring ends with \0 to insert a null character. This includes the final
333 substring, so that the whole string ends with \0\0, which can be detected when
334 counting through. */
335
336 static const char error_texts[] =
337 "no error\0"
338 "\\ at end of pattern\0"
339 "\\c at end of pattern\0"
340 "unrecognized character follows \\\0"
341 "numbers out of order in {} quantifier\0"
342 /* 5 */
343 "number too big in {} quantifier\0"
344 "missing terminating ] for character class\0"
345 "invalid escape sequence in character class\0"
346 "range out of order in character class\0"
347 "nothing to repeat\0"
348 /* 10 */
349 "operand of unlimited repeat could match the empty string\0" /** DEAD **/
350 "internal error: unexpected repeat\0"
351 "unrecognized character after (? or (?-\0"
352 "POSIX named classes are supported only within a class\0"
353 "missing )\0"
354 /* 15 */
355 "reference to non-existent subpattern\0"
356 "erroffset passed as NULL\0"
357 "unknown option bit(s) set\0"
358 "missing ) after comment\0"
359 "parentheses nested too deeply\0" /** DEAD **/
360 /* 20 */
361 "regular expression is too large\0"
362 "failed to get memory\0"
363 "unmatched parentheses\0"
364 "internal error: code overflow\0"
365 "unrecognized character after (?<\0"
366 /* 25 */
367 "lookbehind assertion is not fixed length\0"
368 "malformed number or name after (?(\0"
369 "conditional group contains more than two branches\0"
370 "assertion expected after (?(\0"
371 "(?R or (?[+-]digits must be followed by )\0"
372 /* 30 */
373 "unknown POSIX class name\0"
374 "POSIX collating elements are not supported\0"
375 "this version of PCRE is not compiled with PCRE_UTF8 support\0"
376 "spare error\0" /** DEAD **/
377 "character value in \\x{...} sequence is too large\0"
378 /* 35 */
379 "invalid condition (?(0)\0"
380 "\\C not allowed in lookbehind assertion\0"
381 "PCRE does not support \\L, \\l, \\N{name}, \\U, or \\u\0"
382 "number after (?C is > 255\0"
383 "closing ) for (?C expected\0"
384 /* 40 */
385 "recursive call could loop indefinitely\0"
386 "unrecognized character after (?P\0"
387 "syntax error in subpattern name (missing terminator)\0"
388 "two named subpatterns have the same name\0"
389 "invalid UTF-8 string\0"
390 /* 45 */
391 "support for \\P, \\p, and \\X has not been compiled\0"
392 "malformed \\P or \\p sequence\0"
393 "unknown property name after \\P or \\p\0"
394 "subpattern name is too long (maximum " XSTRING(MAX_NAME_SIZE) " characters)\0"
395 "too many named subpatterns (maximum " XSTRING(MAX_NAME_COUNT) ")\0"
396 /* 50 */
397 "repeated subpattern is too long\0" /** DEAD **/
398 "octal value is greater than \\377 (not in UTF-8 mode)\0"
399 "internal error: overran compiling workspace\0"
400 "internal error: previously-checked referenced subpattern not found\0"
401 "DEFINE group contains more than one branch\0"
402 /* 55 */
403 "repeating a DEFINE group is not allowed\0" /** DEAD **/
404 "inconsistent NEWLINE options\0"
405 "\\g is not followed by a braced, angle-bracketed, or quoted name/number or by a plain number\0"
406 "a numbered reference must not be zero\0"
407 "an argument is not allowed for (*ACCEPT), (*FAIL), or (*COMMIT)\0"
408 /* 60 */
409 "(*VERB) not recognized\0"
410 "number is too big\0"
411 "subpattern name expected\0"
412 "digit expected after (?+\0"
413 "] is an invalid data character in JavaScript compatibility mode\0"
414 /* 65 */
415 "different names for subpatterns of the same number are not allowed\0"
416 "(*MARK) must have an argument\0"
417 "this version of PCRE is not compiled with PCRE_UCP support\0"
418 "\\c must be followed by an ASCII character\0"
419 "\\k is not followed by a braced, angle-bracketed, or quoted name\0"
420 /* 70 */
421 "internal error: unknown opcode in find_fixedlength()\0"
422 "\\N is not supported in a class\0"
423 "too many forward references\0"
424 ;
425
426 /* Table to identify digits and hex digits. This is used when compiling
427 patterns. Note that the tables in chartables are dependent on the locale, and
428 may mark arbitrary characters as digits - but the PCRE compiling code expects
429 to handle only 0-9, a-z, and A-Z as digits when compiling. That is why we have
430 a private table here. It costs 256 bytes, but it is a lot faster than doing
431 character value tests (at least in some simple cases I timed), and in some
432 applications one wants PCRE to compile efficiently as well as match
433 efficiently.
434
435 For convenience, we use the same bit definitions as in chartables:
436
437 0x04 decimal digit
438 0x08 hexadecimal digit
439
440 Then we can use ctype_digit and ctype_xdigit in the code. */
441
442 #ifndef EBCDIC
443
444 /* This is the "normal" case, for ASCII systems, and EBCDIC systems running in
445 UTF-8 mode. */
446
447 static const unsigned char digitab[] =
448 {
449 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 0- 7 */
450 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 8- 15 */
451 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 16- 23 */
452 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 24- 31 */
453 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - ' */
454 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* ( - / */
455 0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c, /* 0 - 7 */
456 0x0c,0x0c,0x00,0x00,0x00,0x00,0x00,0x00, /* 8 - ? */
457 0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* @ - G */
458 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* H - O */
459 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* P - W */
460 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* X - _ */
461 0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* ` - g */
462 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* h - o */
463 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* p - w */
464 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* x -127 */
465 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 128-135 */
466 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 136-143 */
467 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 144-151 */
468 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 152-159 */
469 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 160-167 */
470 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 168-175 */
471 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 176-183 */
472 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 184-191 */
473 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 192-199 */
474 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 200-207 */
475 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 208-215 */
476 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 216-223 */
477 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 224-231 */
478 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 232-239 */
479 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 240-247 */
480 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00};/* 248-255 */
481
482 #else
483
484 /* This is the "abnormal" case, for EBCDIC systems not running in UTF-8 mode. */
485
486 static const unsigned char digitab[] =
487 {
488 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 0- 7 0 */
489 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 8- 15 */
490 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 16- 23 10 */
491 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 24- 31 */
492 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 32- 39 20 */
493 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 40- 47 */
494 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 48- 55 30 */
495 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 56- 63 */
496 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - 71 40 */
497 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 72- | */
498 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* & - 87 50 */
499 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 88- 95 */
500 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - -103 60 */
501 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 104- ? */
502 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 112-119 70 */
503 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 120- " */
504 0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* 128- g 80 */
505 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* h -143 */
506 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 144- p 90 */
507 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* q -159 */
508 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 160- x A0 */
509 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* y -175 */
510 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* ^ -183 B0 */
511 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 184-191 */
512 0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* { - G C0 */
513 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* H -207 */
514 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* } - P D0 */
515 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* Q -223 */
516 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* \ - X E0 */
517 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* Y -239 */
518 0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c, /* 0 - 7 F0 */
519 0x0c,0x0c,0x00,0x00,0x00,0x00,0x00,0x00};/* 8 -255 */
520
521 static const unsigned char ebcdic_chartab[] = { /* chartable partial dup */
522 0x80,0x00,0x00,0x00,0x00,0x01,0x00,0x00, /* 0- 7 */
523 0x00,0x00,0x00,0x00,0x01,0x01,0x00,0x00, /* 8- 15 */
524 0x00,0x00,0x00,0x00,0x00,0x01,0x00,0x00, /* 16- 23 */
525 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 24- 31 */
526 0x00,0x00,0x00,0x00,0x00,0x01,0x00,0x00, /* 32- 39 */
527 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 40- 47 */
528 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 48- 55 */
529 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 56- 63 */
530 0x01,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - 71 */
531 0x00,0x00,0x00,0x80,0x00,0x80,0x80,0x80, /* 72- | */
532 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* & - 87 */
533 0x00,0x00,0x00,0x80,0x80,0x80,0x00,0x00, /* 88- 95 */
534 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - -103 */
535 0x00,0x00,0x00,0x00,0x00,0x10,0x00,0x80, /* 104- ? */
536 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 112-119 */
537 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 120- " */
538 0x00,0x1a,0x1a,0x1a,0x1a,0x1a,0x1a,0x12, /* 128- g */
539 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* h -143 */
540 0x00,0x12,0x12,0x12,0x12,0x12,0x12,0x12, /* 144- p */
541 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* q -159 */
542 0x00,0x00,0x12,0x12,0x12,0x12,0x12,0x12, /* 160- x */
543 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* y -175 */
544 0x80,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* ^ -183 */
545 0x00,0x00,0x80,0x00,0x00,0x00,0x00,0x00, /* 184-191 */
546 0x80,0x1a,0x1a,0x1a,0x1a,0x1a,0x1a,0x12, /* { - G */
547 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* H -207 */
548 0x00,0x12,0x12,0x12,0x12,0x12,0x12,0x12, /* } - P */
549 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* Q -223 */
550 0x00,0x00,0x12,0x12,0x12,0x12,0x12,0x12, /* \ - X */
551 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* Y -239 */
552 0x1c,0x1c,0x1c,0x1c,0x1c,0x1c,0x1c,0x1c, /* 0 - 7 */
553 0x1c,0x1c,0x00,0x00,0x00,0x00,0x00,0x00};/* 8 -255 */
554 #endif
555
556
557 /* Definition to allow mutual recursion */
558
559 static BOOL
560 compile_regex(int, uschar **, const uschar **, int *, BOOL, BOOL, int, int,
561 int *, int *, branch_chain *, compile_data *, int *);
562
563
564
565 /*************************************************
566 * Find an error text *
567 *************************************************/
568
569 /* The error texts are now all in one long string, to save on relocations. As
570 some of the text is of unknown length, we can't use a table of offsets.
571 Instead, just count through the strings. This is not a performance issue
572 because it happens only when there has been a compilation error.
573
574 Argument: the error number
575 Returns: pointer to the error string
576 */
577
578 static const char *
579 find_error_text(int n)
580 {
581 const char *s = error_texts;
582 for (; n > 0; n--)
583 {
584 while (*s++ != 0) {};
585 if (*s == 0) return "Error text not found (please report)";
586 }
587 return s;
588 }
589
590
591 /*************************************************
592 * Expand the workspace *
593 *************************************************/
594
595 /* This function is called during the second compiling phase, if the number of
596 forward references fills the existing workspace, which is originally a block on
597 the stack. A larger block is obtained from malloc() unless the ultimate limit
598 has been reached or the increase will be rather small.
599
600 Argument: pointer to the compile data block
601 Returns: 0 if all went well, else an error number
602 */
603
604 static int
605 expand_workspace(compile_data *cd)
606 {
607 uschar *newspace;
608 int newsize = cd->workspace_size * 2;
609
610 if (newsize > COMPILE_WORK_SIZE_MAX) newsize = COMPILE_WORK_SIZE_MAX;
611 if (cd->workspace_size >= COMPILE_WORK_SIZE_MAX ||
612 newsize - cd->workspace_size < WORK_SIZE_SAFETY_MARGIN)
613 return ERR72;
614
615 newspace = (pcre_malloc)(newsize);
616 if (newspace == NULL) return ERR21;
617
618 memcpy(newspace, cd->start_workspace, cd->workspace_size);
619 cd->hwm = (uschar *)newspace + (cd->hwm - cd->start_workspace);
620 if (cd->workspace_size > COMPILE_WORK_SIZE)
621 (pcre_free)((void *)cd->start_workspace);
622 cd->start_workspace = newspace;
623 cd->workspace_size = newsize;
624 return 0;
625 }
626
627
628
629 /*************************************************
630 * Check for counted repeat *
631 *************************************************/
632
633 /* This function is called when a '{' is encountered in a place where it might
634 start a quantifier. It looks ahead to see if it really is a quantifier or not.
635 It is only a quantifier if it is one of the forms {ddd} {ddd,} or {ddd,ddd}
636 where the ddds are digits.
637
638 Arguments:
639 p pointer to the first char after '{'
640
641 Returns: TRUE or FALSE
642 */
643
644 static BOOL
645 is_counted_repeat(const uschar *p)
646 {
647 if ((digitab[*p++] & ctype_digit) == 0) return FALSE;
648 while ((digitab[*p] & ctype_digit) != 0) p++;
649 if (*p == CHAR_RIGHT_CURLY_BRACKET) return TRUE;
650
651 if (*p++ != CHAR_COMMA) return FALSE;
652 if (*p == CHAR_RIGHT_CURLY_BRACKET) return TRUE;
653
654 if ((digitab[*p++] & ctype_digit) == 0) return FALSE;
655 while ((digitab[*p] & ctype_digit) != 0) p++;
656
657 return (*p == CHAR_RIGHT_CURLY_BRACKET);
658 }
659
660
661
662 /*************************************************
663 * Handle escapes *
664 *************************************************/
665
666 /* This function is called when a \ has been encountered. It either returns a
667 positive value for a simple escape such as \n, or a negative value which
668 encodes one of the more complicated things such as \d. A backreference to group
669 n is returned as -(ESC_REF + n); ESC_REF is the highest ESC_xxx macro. When
670 UTF-8 is enabled, a positive value greater than 255 may be returned. On entry,
671 ptr is pointing at the \. On exit, it is on the final character of the escape
672 sequence.
673
674 Arguments:
675 ptrptr points to the pattern position pointer
676 errorcodeptr points to the errorcode variable
677 bracount number of previous extracting brackets
678 options the options bits
679 isclass TRUE if inside a character class
680
681 Returns: zero or positive => a data character
682 negative => a special escape sequence
683 on error, errorcodeptr is set
684 */
685
686 static int
687 check_escape(const uschar **ptrptr, int *errorcodeptr, int bracount,
688 int options, BOOL isclass)
689 {
690 BOOL utf8 = (options & PCRE_UTF8) != 0;
691 const uschar *ptr = *ptrptr + 1;
692 int c, i;
693
694 GETCHARINCTEST(c, ptr); /* Get character value, increment pointer */
695 ptr--; /* Set pointer back to the last byte */
696
697 /* If backslash is at the end of the pattern, it's an error. */
698
699 if (c == 0) *errorcodeptr = ERR1;
700
701 /* Non-alphanumerics are literals. For digits or letters, do an initial lookup
702 in a table. A non-zero result is something that can be returned immediately.
703 Otherwise further processing may be required. */
704
705 #ifndef EBCDIC /* ASCII/UTF-8 coding */
706 else if (c < CHAR_0 || c > CHAR_z) {} /* Not alphanumeric */
707 else if ((i = escapes[c - CHAR_0]) != 0) c = i;
708
709 #else /* EBCDIC coding */
710 else if (c < 'a' || (ebcdic_chartab[c] & 0x0E) == 0) {} /* Not alphanumeric */
711 else if ((i = escapes[c - 0x48]) != 0) c = i;
712 #endif
713
714 /* Escapes that need further processing, or are illegal. */
715
716 else
717 {
718 const uschar *oldptr;
719 BOOL braced, negated;
720
721 switch (c)
722 {
723 /* A number of Perl escapes are not handled by PCRE. We give an explicit
724 error. */
725
726 case CHAR_l:
727 case CHAR_L:
728 *errorcodeptr = ERR37;
729 break;
730
731 case CHAR_u:
732 if ((options & PCRE_JAVASCRIPT_COMPAT) != 0)
733 {
734 /* In JavaScript, \u must be followed by four hexadecimal numbers.
735 Otherwise it is a lowercase u letter. */
736 if ((digitab[ptr[1]] & ctype_xdigit) != 0 && (digitab[ptr[2]] & ctype_xdigit) != 0
737 && (digitab[ptr[3]] & ctype_xdigit) != 0 && (digitab[ptr[4]] & ctype_xdigit) != 0)
738 {
739 c = 0;
740 for (i = 0; i < 4; ++i)
741 {
742 register int cc = *(++ptr);
743 #ifndef EBCDIC /* ASCII/UTF-8 coding */
744 if (cc >= CHAR_a) cc -= 32; /* Convert to upper case */
745 c = (c << 4) + cc - ((cc < CHAR_A)? CHAR_0 : (CHAR_A - 10));
746 #else /* EBCDIC coding */
747 if (cc >= CHAR_a && cc <= CHAR_z) cc += 64; /* Convert to upper case */
748 c = (c << 4) + cc - ((cc >= CHAR_0)? CHAR_0 : (CHAR_A - 10));
749 #endif
750 }
751 }
752 }
753 else
754 *errorcodeptr = ERR37;
755 break;
756
757 case CHAR_U:
758 /* In JavaScript, \U is an uppercase U letter. */
759 if ((options & PCRE_JAVASCRIPT_COMPAT) == 0) *errorcodeptr = ERR37;
760 break;
761
762 /* In a character class, \g is just a literal "g". Outside a character
763 class, \g must be followed by one of a number of specific things:
764
765 (1) A number, either plain or braced. If positive, it is an absolute
766 backreference. If negative, it is a relative backreference. This is a Perl
767 5.10 feature.
768
769 (2) Perl 5.10 also supports \g{name} as a reference to a named group. This
770 is part of Perl's movement towards a unified syntax for back references. As
771 this is synonymous with \k{name}, we fudge it up by pretending it really
772 was \k.
773
774 (3) For Oniguruma compatibility we also support \g followed by a name or a
775 number either in angle brackets or in single quotes. However, these are
776 (possibly recursive) subroutine calls, _not_ backreferences. Just return
777 the -ESC_g code (cf \k). */
778
779 case CHAR_g:
780 if (isclass) break;
781 if (ptr[1] == CHAR_LESS_THAN_SIGN || ptr[1] == CHAR_APOSTROPHE)
782 {
783 c = -ESC_g;
784 break;
785 }
786
787 /* Handle the Perl-compatible cases */
788
789 if (ptr[1] == CHAR_LEFT_CURLY_BRACKET)
790 {
791 const uschar *p;
792 for (p = ptr+2; *p != 0 && *p != CHAR_RIGHT_CURLY_BRACKET; p++)
793 if (*p != CHAR_MINUS && (digitab[*p] & ctype_digit) == 0) break;
794 if (*p != 0 && *p != CHAR_RIGHT_CURLY_BRACKET)
795 {
796 c = -ESC_k;
797 break;
798 }
799 braced = TRUE;
800 ptr++;
801 }
802 else braced = FALSE;
803
804 if (ptr[1] == CHAR_MINUS)
805 {
806 negated = TRUE;
807 ptr++;
808 }
809 else negated = FALSE;
810
811 c = 0;
812 while ((digitab[ptr[1]] & ctype_digit) != 0)
813 c = c * 10 + *(++ptr) - CHAR_0;
814
815 if (c < 0) /* Integer overflow */
816 {
817 *errorcodeptr = ERR61;
818 break;
819 }
820
821 if (braced && *(++ptr) != CHAR_RIGHT_CURLY_BRACKET)
822 {
823 *errorcodeptr = ERR57;
824 break;
825 }
826
827 if (c == 0)
828 {
829 *errorcodeptr = ERR58;
830 break;
831 }
832
833 if (negated)
834 {
835 if (c > bracount)
836 {
837 *errorcodeptr = ERR15;
838 break;
839 }
840 c = bracount - (c - 1);
841 }
842
843 c = -(ESC_REF + c);
844 break;
845
846 /* The handling of escape sequences consisting of a string of digits
847 starting with one that is not zero is not straightforward. By experiment,
848 the way Perl works seems to be as follows:
849
850 Outside a character class, the digits are read as a decimal number. If the
851 number is less than 10, or if there are that many previous extracting
852 left brackets, then it is a back reference. Otherwise, up to three octal
853 digits are read to form an escaped byte. Thus \123 is likely to be octal
854 123 (cf \0123, which is octal 012 followed by the literal 3). If the octal
855 value is greater than 377, the least significant 8 bits are taken. Inside a
856 character class, \ followed by a digit is always an octal number. */
857
858 case CHAR_1: case CHAR_2: case CHAR_3: case CHAR_4: case CHAR_5:
859 case CHAR_6: case CHAR_7: case CHAR_8: case CHAR_9:
860
861 if (!isclass)
862 {
863 oldptr = ptr;
864 c -= CHAR_0;
865 while ((digitab[ptr[1]] & ctype_digit) != 0)
866 c = c * 10 + *(++ptr) - CHAR_0;
867 if (c < 0) /* Integer overflow */
868 {
869 *errorcodeptr = ERR61;
870 break;
871 }
872 if (c < 10 || c <= bracount)
873 {
874 c = -(ESC_REF + c);
875 break;
876 }
877 ptr = oldptr; /* Put the pointer back and fall through */
878 }
879
880 /* Handle an octal number following \. If the first digit is 8 or 9, Perl
881 generates a binary zero byte and treats the digit as a following literal.
882 Thus we have to pull back the pointer by one. */
883
884 if ((c = *ptr) >= CHAR_8)
885 {
886 ptr--;
887 c = 0;
888 break;
889 }
890
891 /* \0 always starts an octal number, but we may drop through to here with a
892 larger first octal digit. The original code used just to take the least
893 significant 8 bits of octal numbers (I think this is what early Perls used
894 to do). Nowadays we allow for larger numbers in UTF-8 mode, but no more
895 than 3 octal digits. */
896
897 case CHAR_0:
898 c -= CHAR_0;
899 while(i++ < 2 && ptr[1] >= CHAR_0 && ptr[1] <= CHAR_7)
900 c = c * 8 + *(++ptr) - CHAR_0;
901 if (!utf8 && c > 255) *errorcodeptr = ERR51;
902 break;
903
904 /* \x is complicated. \x{ddd} is a character number which can be greater
905 than 0xff in utf8 mode, but only if the ddd are hex digits. If not, { is
906 treated as a data character. */
907
908 case CHAR_x:
909 if ((options & PCRE_JAVASCRIPT_COMPAT) != 0)
910 {
911 /* In JavaScript, \x must be followed by two hexadecimal numbers.
912 Otherwise it is a lowercase x letter. */
913 if ((digitab[ptr[1]] & ctype_xdigit) != 0 && (digitab[ptr[2]] & ctype_xdigit) != 0)
914 {
915 c = 0;
916 for (i = 0; i < 2; ++i)
917 {
918 register int cc = *(++ptr);
919 #ifndef EBCDIC /* ASCII/UTF-8 coding */
920 if (cc >= CHAR_a) cc -= 32; /* Convert to upper case */
921 c = (c << 4) + cc - ((cc < CHAR_A)? CHAR_0 : (CHAR_A - 10));
922 #else /* EBCDIC coding */
923 if (cc >= CHAR_a && cc <= CHAR_z) cc += 64; /* Convert to upper case */
924 c = (c << 4) + cc - ((cc >= CHAR_0)? CHAR_0 : (CHAR_A - 10));
925 #endif
926 }
927 }
928 break;
929 }
930
931 if (ptr[1] == CHAR_LEFT_CURLY_BRACKET)
932 {
933 const uschar *pt = ptr + 2;
934 int count = 0;
935
936 c = 0;
937 while ((digitab[*pt] & ctype_xdigit) != 0)
938 {
939 register int cc = *pt++;
940 if (c == 0 && cc == CHAR_0) continue; /* Leading zeroes */
941 count++;
942
943 #ifndef EBCDIC /* ASCII/UTF-8 coding */
944 if (cc >= CHAR_a) cc -= 32; /* Convert to upper case */
945 c = (c << 4) + cc - ((cc < CHAR_A)? CHAR_0 : (CHAR_A - 10));
946 #else /* EBCDIC coding */
947 if (cc >= CHAR_a && cc <= CHAR_z) cc += 64; /* Convert to upper case */
948 c = (c << 4) + cc - ((cc >= CHAR_0)? CHAR_0 : (CHAR_A - 10));
949 #endif
950 }
951
952 if (*pt == CHAR_RIGHT_CURLY_BRACKET)
953 {
954 if (c < 0 || count > (utf8? 8 : 2)) *errorcodeptr = ERR34;
955 ptr = pt;
956 break;
957 }
958
959 /* If the sequence of hex digits does not end with '}', then we don't
960 recognize this construct; fall through to the normal \x handling. */
961 }
962
963 /* Read just a single-byte hex-defined char */
964
965 c = 0;
966 while (i++ < 2 && (digitab[ptr[1]] & ctype_xdigit) != 0)
967 {
968 int cc; /* Some compilers don't like */
969 cc = *(++ptr); /* ++ in initializers */
970 #ifndef EBCDIC /* ASCII/UTF-8 coding */
971 if (cc >= CHAR_a) cc -= 32; /* Convert to upper case */
972 c = c * 16 + cc - ((cc < CHAR_A)? CHAR_0 : (CHAR_A - 10));
973 #else /* EBCDIC coding */
974 if (cc <= CHAR_z) cc += 64; /* Convert to upper case */
975 c = c * 16 + cc - ((cc >= CHAR_0)? CHAR_0 : (CHAR_A - 10));
976 #endif
977 }
978 break;
979
980 /* For \c, a following letter is upper-cased; then the 0x40 bit is flipped.
981 An error is given if the byte following \c is not an ASCII character. This
982 coding is ASCII-specific, but then the whole concept of \cx is
983 ASCII-specific. (However, an EBCDIC equivalent has now been added.) */
984
985 case CHAR_c:
986 c = *(++ptr);
987 if (c == 0)
988 {
989 *errorcodeptr = ERR2;
990 break;
991 }
992 #ifndef EBCDIC /* ASCII/UTF-8 coding */
993 if (c > 127) /* Excludes all non-ASCII in either mode */
994 {
995 *errorcodeptr = ERR68;
996 break;
997 }
998 if (c >= CHAR_a && c <= CHAR_z) c -= 32;
999 c ^= 0x40;
1000 #else /* EBCDIC coding */
1001 if (c >= CHAR_a && c <= CHAR_z) c += 64;
1002 c ^= 0xC0;
1003 #endif
1004 break;
1005
1006 /* PCRE_EXTRA enables extensions to Perl in the matter of escapes. Any
1007 other alphanumeric following \ is an error if PCRE_EXTRA was set;
1008 otherwise, for Perl compatibility, it is a literal. This code looks a bit
1009 odd, but there used to be some cases other than the default, and there may
1010 be again in future, so I haven't "optimized" it. */
1011
1012 default:
1013 if ((options & PCRE_EXTRA) != 0) switch(c)
1014 {
1015 default:
1016 *errorcodeptr = ERR3;
1017 break;
1018 }
1019 break;
1020 }
1021 }
1022
1023 /* Perl supports \N{name} for character names, as well as plain \N for "not
1024 newline". PCRE does not support \N{name}. However, it does support
1025 quantification such as \N{2,3}. */
1026
1027 if (c == -ESC_N && ptr[1] == CHAR_LEFT_CURLY_BRACKET &&
1028 !is_counted_repeat(ptr+2))
1029 *errorcodeptr = ERR37;
1030
1031 /* If PCRE_UCP is set, we change the values for \d etc. */
1032
1033 if ((options & PCRE_UCP) != 0 && c <= -ESC_D && c >= -ESC_w)
1034 c -= (ESC_DU - ESC_D);
1035
1036 /* Set the pointer to the final character before returning. */
1037
1038 *ptrptr = ptr;
1039 return c;
1040 }
1041
1042
1043
1044 #ifdef SUPPORT_UCP
1045 /*************************************************
1046 * Handle \P and \p *
1047 *************************************************/
1048
1049 /* This function is called after \P or \p has been encountered, provided that
1050 PCRE is compiled with support for Unicode properties. On entry, ptrptr is
1051 pointing at the P or p. On exit, it is pointing at the final character of the
1052 escape sequence.
1053
1054 Argument:
1055 ptrptr points to the pattern position pointer
1056 negptr points to a boolean that is set TRUE for negation else FALSE
1057 dptr points to an int that is set to the detailed property value
1058 errorcodeptr points to the error code variable
1059
1060 Returns: type value from ucp_type_table, or -1 for an invalid type
1061 */
1062
1063 static int
1064 get_ucp(const uschar **ptrptr, BOOL *negptr, int *dptr, int *errorcodeptr)
1065 {
1066 int c, i, bot, top;
1067 const uschar *ptr = *ptrptr;
1068 char name[32];
1069
1070 c = *(++ptr);
1071 if (c == 0) goto ERROR_RETURN;
1072
1073 *negptr = FALSE;
1074
1075 /* \P or \p can be followed by a name in {}, optionally preceded by ^ for
1076 negation. */
1077
1078 if (c == CHAR_LEFT_CURLY_BRACKET)
1079 {
1080 if (ptr[1] == CHAR_CIRCUMFLEX_ACCENT)
1081 {
1082 *negptr = TRUE;
1083 ptr++;
1084 }
1085 for (i = 0; i < (int)sizeof(name) - 1; i++)
1086 {
1087 c = *(++ptr);
1088 if (c == 0) goto ERROR_RETURN;
1089 if (c == CHAR_RIGHT_CURLY_BRACKET) break;
1090 name[i] = c;
1091 }
1092 if (c != CHAR_RIGHT_CURLY_BRACKET) goto ERROR_RETURN;
1093 name[i] = 0;
1094 }
1095
1096 /* Otherwise there is just one following character */
1097
1098 else
1099 {
1100 name[0] = c;
1101 name[1] = 0;
1102 }
1103
1104 *ptrptr = ptr;
1105
1106 /* Search for a recognized property name using binary chop */
1107
1108 bot = 0;
1109 top = _pcre_utt_size;
1110
1111 while (bot < top)
1112 {
1113 i = (bot + top) >> 1;
1114 c = strcmp(name, _pcre_utt_names + _pcre_utt[i].name_offset);
1115 if (c == 0)
1116 {
1117 *dptr = _pcre_utt[i].value;
1118 return _pcre_utt[i].type;
1119 }
1120 if (c > 0) bot = i + 1; else top = i;
1121 }
1122
1123 *errorcodeptr = ERR47;
1124 *ptrptr = ptr;
1125 return -1;
1126
1127 ERROR_RETURN:
1128 *errorcodeptr = ERR46;
1129 *ptrptr = ptr;
1130 return -1;
1131 }
1132 #endif
1133
1134
1135
1136
1137 /*************************************************
1138 * Read repeat counts *
1139 *************************************************/
1140
1141 /* Read an item of the form {n,m} and return the values. This is called only
1142 after is_counted_repeat() has confirmed that a repeat-count quantifier exists,
1143 so the syntax is guaranteed to be correct, but we need to check the values.
1144
1145 Arguments:
1146 p pointer to first char after '{'
1147 minp pointer to int for min
1148 maxp pointer to int for max
1149 returned as -1 if no max
1150 errorcodeptr points to error code variable
1151
1152 Returns: pointer to '}' on success;
1153 current ptr on error, with errorcodeptr set non-zero
1154 */
1155
1156 static const uschar *
1157 read_repeat_counts(const uschar *p, int *minp, int *maxp, int *errorcodeptr)
1158 {
1159 int min = 0;
1160 int max = -1;
1161
1162 /* Read the minimum value and do a paranoid check: a negative value indicates
1163 an integer overflow. */
1164
1165 while ((digitab[*p] & ctype_digit) != 0) min = min * 10 + *p++ - CHAR_0;
1166 if (min < 0 || min > 65535)
1167 {
1168 *errorcodeptr = ERR5;
1169 return p;
1170 }
1171
1172 /* Read the maximum value if there is one, and again do a paranoid on its size.
1173 Also, max must not be less than min. */
1174
1175 if (*p == CHAR_RIGHT_CURLY_BRACKET) max = min; else
1176 {
1177 if (*(++p) != CHAR_RIGHT_CURLY_BRACKET)
1178 {
1179 max = 0;
1180 while((digitab[*p] & ctype_digit) != 0) max = max * 10 + *p++ - CHAR_0;
1181 if (max < 0 || max > 65535)
1182 {
1183 *errorcodeptr = ERR5;
1184 return p;
1185 }
1186 if (max < min)
1187 {
1188 *errorcodeptr = ERR4;
1189 return p;
1190 }
1191 }
1192 }
1193
1194 /* Fill in the required variables, and pass back the pointer to the terminating
1195 '}'. */
1196
1197 *minp = min;
1198 *maxp = max;
1199 return p;
1200 }
1201
1202
1203
1204 /*************************************************
1205 * Subroutine for finding forward reference *
1206 *************************************************/
1207
1208 /* This recursive function is called only from find_parens() below. The
1209 top-level call starts at the beginning of the pattern. All other calls must
1210 start at a parenthesis. It scans along a pattern's text looking for capturing
1211 subpatterns, and counting them. If it finds a named pattern that matches the
1212 name it is given, it returns its number. Alternatively, if the name is NULL, it
1213 returns when it reaches a given numbered subpattern. Recursion is used to keep
1214 track of subpatterns that reset the capturing group numbers - the (?| feature.
1215
1216 This function was originally called only from the second pass, in which we know
1217 that if (?< or (?' or (?P< is encountered, the name will be correctly
1218 terminated because that is checked in the first pass. There is now one call to
1219 this function in the first pass, to check for a recursive back reference by
1220 name (so that we can make the whole group atomic). In this case, we need check
1221 only up to the current position in the pattern, and that is still OK because
1222 and previous occurrences will have been checked. To make this work, the test
1223 for "end of pattern" is a check against cd->end_pattern in the main loop,
1224 instead of looking for a binary zero. This means that the special first-pass
1225 call can adjust cd->end_pattern temporarily. (Checks for binary zero while
1226 processing items within the loop are OK, because afterwards the main loop will
1227 terminate.)
1228
1229 Arguments:
1230 ptrptr address of the current character pointer (updated)
1231 cd compile background data
1232 name name to seek, or NULL if seeking a numbered subpattern
1233 lorn name length, or subpattern number if name is NULL
1234 xmode TRUE if we are in /x mode
1235 utf8 TRUE if we are in UTF-8 mode
1236 count pointer to the current capturing subpattern number (updated)
1237
1238 Returns: the number of the named subpattern, or -1 if not found
1239 */
1240
1241 static int
1242 find_parens_sub(uschar **ptrptr, compile_data *cd, const uschar *name, int lorn,
1243 BOOL xmode, BOOL utf8, int *count)
1244 {
1245 uschar *ptr = *ptrptr;
1246 int start_count = *count;
1247 int hwm_count = start_count;
1248 BOOL dup_parens = FALSE;
1249
1250 /* If the first character is a parenthesis, check on the type of group we are
1251 dealing with. The very first call may not start with a parenthesis. */
1252
1253 if (ptr[0] == CHAR_LEFT_PARENTHESIS)
1254 {
1255 /* Handle specials such as (*SKIP) or (*UTF8) etc. */
1256
1257 if (ptr[1] == CHAR_ASTERISK) ptr += 2;
1258
1259 /* Handle a normal, unnamed capturing parenthesis. */
1260
1261 else if (ptr[1] != CHAR_QUESTION_MARK)
1262 {
1263 *count += 1;
1264 if (name == NULL && *count == lorn) return *count;
1265 ptr++;
1266 }
1267
1268 /* All cases now have (? at the start. Remember when we are in a group
1269 where the parenthesis numbers are duplicated. */
1270
1271 else if (ptr[2] == CHAR_VERTICAL_LINE)
1272 {
1273 ptr += 3;
1274 dup_parens = TRUE;
1275 }
1276
1277 /* Handle comments; all characters are allowed until a ket is reached. */
1278
1279 else if (ptr[2] == CHAR_NUMBER_SIGN)
1280 {
1281 for (ptr += 3; *ptr != 0; ptr++) if (*ptr == CHAR_RIGHT_PARENTHESIS) break;
1282 goto FAIL_EXIT;
1283 }
1284
1285 /* Handle a condition. If it is an assertion, just carry on so that it
1286 is processed as normal. If not, skip to the closing parenthesis of the
1287 condition (there can't be any nested parens). */
1288
1289 else if (ptr[2] == CHAR_LEFT_PARENTHESIS)
1290 {
1291 ptr += 2;
1292 if (ptr[1] != CHAR_QUESTION_MARK)
1293 {
1294 while (*ptr != 0 && *ptr != CHAR_RIGHT_PARENTHESIS) ptr++;
1295 if (*ptr != 0) ptr++;
1296 }
1297 }
1298
1299 /* Start with (? but not a condition. */
1300
1301 else
1302 {
1303 ptr += 2;
1304 if (*ptr == CHAR_P) ptr++; /* Allow optional P */
1305
1306 /* We have to disambiguate (?<! and (?<= from (?<name> for named groups */
1307
1308 if ((*ptr == CHAR_LESS_THAN_SIGN && ptr[1] != CHAR_EXCLAMATION_MARK &&
1309 ptr[1] != CHAR_EQUALS_SIGN) || *ptr == CHAR_APOSTROPHE)
1310 {
1311 int term;
1312 const uschar *thisname;
1313 *count += 1;
1314 if (name == NULL && *count == lorn) return *count;
1315 term = *ptr++;
1316 if (term == CHAR_LESS_THAN_SIGN) term = CHAR_GREATER_THAN_SIGN;
1317 thisname = ptr;
1318 while (*ptr != term) ptr++;
1319 if (name != NULL && lorn == ptr - thisname &&
1320 strncmp((const char *)name, (const char *)thisname, lorn) == 0)
1321 return *count;
1322 term++;
1323 }
1324 }
1325 }
1326
1327 /* Past any initial parenthesis handling, scan for parentheses or vertical
1328 bars. Stop if we get to cd->end_pattern. Note that this is important for the
1329 first-pass call when this value is temporarily adjusted to stop at the current
1330 position. So DO NOT change this to a test for binary zero. */
1331
1332 for (; ptr < cd->end_pattern; ptr++)
1333 {
1334 /* Skip over backslashed characters and also entire \Q...\E */
1335
1336 if (*ptr == CHAR_BACKSLASH)
1337 {
1338 if (*(++ptr) == 0) goto FAIL_EXIT;
1339 if (*ptr == CHAR_Q) for (;;)
1340 {
1341 while (*(++ptr) != 0 && *ptr != CHAR_BACKSLASH) {};
1342 if (*ptr == 0) goto FAIL_EXIT;
1343 if (*(++ptr) == CHAR_E) break;
1344 }
1345 continue;
1346 }
1347
1348 /* Skip over character classes; this logic must be similar to the way they
1349 are handled for real. If the first character is '^', skip it. Also, if the
1350 first few characters (either before or after ^) are \Q\E or \E we skip them
1351 too. This makes for compatibility with Perl. Note the use of STR macros to
1352 encode "Q\\E" so that it works in UTF-8 on EBCDIC platforms. */
1353
1354 if (*ptr == CHAR_LEFT_SQUARE_BRACKET)
1355 {
1356 BOOL negate_class = FALSE;
1357 for (;;)
1358 {
1359 if (ptr[1] == CHAR_BACKSLASH)
1360 {
1361 if (ptr[2] == CHAR_E)
1362 ptr+= 2;
1363 else if (strncmp((const char *)ptr+2,
1364 STR_Q STR_BACKSLASH STR_E, 3) == 0)
1365 ptr += 4;
1366 else
1367 break;
1368 }
1369 else if (!negate_class && ptr[1] == CHAR_CIRCUMFLEX_ACCENT)
1370 {
1371 negate_class = TRUE;
1372 ptr++;
1373 }
1374 else break;
1375 }
1376
1377 /* If the next character is ']', it is a data character that must be
1378 skipped, except in JavaScript compatibility mode. */
1379
1380 if (ptr[1] == CHAR_RIGHT_SQUARE_BRACKET &&
1381 (cd->external_options & PCRE_JAVASCRIPT_COMPAT) == 0)
1382 ptr++;
1383
1384 while (*(++ptr) != CHAR_RIGHT_SQUARE_BRACKET)
1385 {
1386 if (*ptr == 0) return -1;
1387 if (*ptr == CHAR_BACKSLASH)
1388 {
1389 if (*(++ptr) == 0) goto FAIL_EXIT;
1390 if (*ptr == CHAR_Q) for (;;)
1391 {
1392 while (*(++ptr) != 0 && *ptr != CHAR_BACKSLASH) {};
1393 if (*ptr == 0) goto FAIL_EXIT;
1394 if (*(++ptr) == CHAR_E) break;
1395 }
1396 continue;
1397 }
1398 }
1399 continue;
1400 }
1401
1402 /* Skip comments in /x mode */
1403
1404 if (xmode && *ptr == CHAR_NUMBER_SIGN)
1405 {
1406 ptr++;
1407 while (*ptr != 0)
1408 {
1409 if (IS_NEWLINE(ptr)) { ptr += cd->nllen - 1; break; }
1410 ptr++;
1411 #ifdef SUPPORT_UTF8
1412 if (utf8) while ((*ptr & 0xc0) == 0x80) ptr++;
1413 #endif
1414 }
1415 if (*ptr == 0) goto FAIL_EXIT;
1416 continue;
1417 }
1418
1419 /* Check for the special metacharacters */
1420
1421 if (*ptr == CHAR_LEFT_PARENTHESIS)
1422 {
1423 int rc = find_parens_sub(&ptr, cd, name, lorn, xmode, utf8, count);
1424 if (rc > 0) return rc;
1425 if (*ptr == 0) goto FAIL_EXIT;
1426 }
1427
1428 else if (*ptr == CHAR_RIGHT_PARENTHESIS)
1429 {
1430 if (dup_parens && *count < hwm_count) *count = hwm_count;
1431 goto FAIL_EXIT;
1432 }
1433
1434 else if (*ptr == CHAR_VERTICAL_LINE && dup_parens)
1435 {
1436 if (*count > hwm_count) hwm_count = *count;
1437 *count = start_count;
1438 }
1439 }
1440
1441 FAIL_EXIT:
1442 *ptrptr = ptr;
1443 return -1;
1444 }
1445
1446
1447
1448
1449 /*************************************************
1450 * Find forward referenced subpattern *
1451 *************************************************/
1452
1453 /* This function scans along a pattern's text looking for capturing
1454 subpatterns, and counting them. If it finds a named pattern that matches the
1455 name it is given, it returns its number. Alternatively, if the name is NULL, it
1456 returns when it reaches a given numbered subpattern. This is used for forward
1457 references to subpatterns. We used to be able to start this scan from the
1458 current compiling point, using the current count value from cd->bracount, and
1459 do it all in a single loop, but the addition of the possibility of duplicate
1460 subpattern numbers means that we have to scan from the very start, in order to
1461 take account of such duplicates, and to use a recursive function to keep track
1462 of the different types of group.
1463
1464 Arguments:
1465 cd compile background data
1466 name name to seek, or NULL if seeking a numbered subpattern
1467 lorn name length, or subpattern number if name is NULL
1468 xmode TRUE if we are in /x mode
1469 utf8 TRUE if we are in UTF-8 mode
1470
1471 Returns: the number of the found subpattern, or -1 if not found
1472 */
1473
1474 static int
1475 find_parens(compile_data *cd, const uschar *name, int lorn, BOOL xmode,
1476 BOOL utf8)
1477 {
1478 uschar *ptr = (uschar *)cd->start_pattern;
1479 int count = 0;
1480 int rc;
1481
1482 /* If the pattern does not start with an opening parenthesis, the first call
1483 to find_parens_sub() will scan right to the end (if necessary). However, if it
1484 does start with a parenthesis, find_parens_sub() will return when it hits the
1485 matching closing parens. That is why we have to have a loop. */
1486
1487 for (;;)
1488 {
1489 rc = find_parens_sub(&ptr, cd, name, lorn, xmode, utf8, &count);
1490 if (rc > 0 || *ptr++ == 0) break;
1491 }
1492
1493 return rc;
1494 }
1495
1496
1497
1498
1499 /*************************************************
1500 * Find first significant op code *
1501 *************************************************/
1502
1503 /* This is called by several functions that scan a compiled expression looking
1504 for a fixed first character, or an anchoring op code etc. It skips over things
1505 that do not influence this. For some calls, it makes sense to skip negative
1506 forward and all backward assertions, and also the \b assertion; for others it
1507 does not.
1508
1509 Arguments:
1510 code pointer to the start of the group
1511 skipassert TRUE if certain assertions are to be skipped
1512
1513 Returns: pointer to the first significant opcode
1514 */
1515
1516 static const uschar*
1517 first_significant_code(const uschar *code, BOOL skipassert)
1518 {
1519 for (;;)
1520 {
1521 switch ((int)*code)
1522 {
1523 case OP_ASSERT_NOT:
1524 case OP_ASSERTBACK:
1525 case OP_ASSERTBACK_NOT:
1526 if (!skipassert) return code;
1527 do code += GET(code, 1); while (*code == OP_ALT);
1528 code += _pcre_OP_lengths[*code];
1529 break;
1530
1531 case OP_WORD_BOUNDARY:
1532 case OP_NOT_WORD_BOUNDARY:
1533 if (!skipassert) return code;
1534 /* Fall through */
1535
1536 case OP_CALLOUT:
1537 case OP_CREF:
1538 case OP_NCREF:
1539 case OP_RREF:
1540 case OP_NRREF:
1541 case OP_DEF:
1542 code += _pcre_OP_lengths[*code];
1543 break;
1544
1545 default:
1546 return code;
1547 }
1548 }
1549 /* Control never reaches here */
1550 }
1551
1552
1553
1554
1555 /*************************************************
1556 * Find the fixed length of a branch *
1557 *************************************************/
1558
1559 /* Scan a branch and compute the fixed length of subject that will match it,
1560 if the length is fixed. This is needed for dealing with backward assertions.
1561 In UTF8 mode, the result is in characters rather than bytes. The branch is
1562 temporarily terminated with OP_END when this function is called.
1563
1564 This function is called when a backward assertion is encountered, so that if it
1565 fails, the error message can point to the correct place in the pattern.
1566 However, we cannot do this when the assertion contains subroutine calls,
1567 because they can be forward references. We solve this by remembering this case
1568 and doing the check at the end; a flag specifies which mode we are running in.
1569
1570 Arguments:
1571 code points to the start of the pattern (the bracket)
1572 utf8 TRUE in UTF-8 mode
1573 atend TRUE if called when the pattern is complete
1574 cd the "compile data" structure
1575
1576 Returns: the fixed length,
1577 or -1 if there is no fixed length,
1578 or -2 if \C was encountered (in UTF-8 mode only)
1579 or -3 if an OP_RECURSE item was encountered and atend is FALSE
1580 or -4 if an unknown opcode was encountered (internal error)
1581 */
1582
1583 static int
1584 find_fixedlength(uschar *code, BOOL utf8, BOOL atend, compile_data *cd)
1585 {
1586 int length = -1;
1587
1588 register int branchlength = 0;
1589 register uschar *cc = code + 1 + LINK_SIZE;
1590
1591 /* Scan along the opcodes for this branch. If we get to the end of the
1592 branch, check the length against that of the other branches. */
1593
1594 for (;;)
1595 {
1596 int d;
1597 uschar *ce, *cs;
1598 register int op = *cc;
1599 switch (op)
1600 {
1601 /* We only need to continue for OP_CBRA (normal capturing bracket) and
1602 OP_BRA (normal non-capturing bracket) because the other variants of these
1603 opcodes are all concerned with unlimited repeated groups, which of course
1604 are not of fixed length. */
1605
1606 case OP_CBRA:
1607 case OP_BRA:
1608 case OP_ONCE:
1609 case OP_ONCE_NC:
1610 case OP_COND:
1611 d = find_fixedlength(cc + ((op == OP_CBRA)? 2:0), utf8, atend, cd);
1612 if (d < 0) return d;
1613 branchlength += d;
1614 do cc += GET(cc, 1); while (*cc == OP_ALT);
1615 cc += 1 + LINK_SIZE;
1616 break;
1617
1618 /* Reached end of a branch; if it's a ket it is the end of a nested call.
1619 If it's ALT it is an alternation in a nested call. An ACCEPT is effectively
1620 an ALT. If it is END it's the end of the outer call. All can be handled by
1621 the same code. Note that we must not include the OP_KETRxxx opcodes here,
1622 because they all imply an unlimited repeat. */
1623
1624 case OP_ALT:
1625 case OP_KET:
1626 case OP_END:
1627 case OP_ACCEPT:
1628 case OP_ASSERT_ACCEPT:
1629 if (length < 0) length = branchlength;
1630 else if (length != branchlength) return -1;
1631 if (*cc != OP_ALT) return length;
1632 cc += 1 + LINK_SIZE;
1633 branchlength = 0;
1634 break;
1635
1636 /* A true recursion implies not fixed length, but a subroutine call may
1637 be OK. If the subroutine is a forward reference, we can't deal with
1638 it until the end of the pattern, so return -3. */
1639
1640 case OP_RECURSE:
1641 if (!atend) return -3;
1642 cs = ce = (uschar *)cd->start_code + GET(cc, 1); /* Start subpattern */
1643 do ce += GET(ce, 1); while (*ce == OP_ALT); /* End subpattern */
1644 if (cc > cs && cc < ce) return -1; /* Recursion */
1645 d = find_fixedlength(cs + 2, utf8, atend, cd);
1646 if (d < 0) return d;
1647 branchlength += d;
1648 cc += 1 + LINK_SIZE;
1649 break;
1650
1651 /* Skip over assertive subpatterns */
1652
1653 case OP_ASSERT:
1654 case OP_ASSERT_NOT:
1655 case OP_ASSERTBACK:
1656 case OP_ASSERTBACK_NOT:
1657 do cc += GET(cc, 1); while (*cc == OP_ALT);
1658 cc += _pcre_OP_lengths[*cc];
1659 break;
1660
1661 /* Skip over things that don't match chars */
1662
1663 case OP_MARK:
1664 case OP_PRUNE_ARG:
1665 case OP_SKIP_ARG:
1666 case OP_THEN_ARG:
1667 cc += cc[1] + _pcre_OP_lengths[*cc];
1668 break;
1669
1670 case OP_CALLOUT:
1671 case OP_CIRC:
1672 case OP_CIRCM:
1673 case OP_CLOSE:
1674 case OP_COMMIT:
1675 case OP_CREF:
1676 case OP_DEF:
1677 case OP_DOLL:
1678 case OP_DOLLM:
1679 case OP_EOD:
1680 case OP_EODN:
1681 case OP_FAIL:
1682 case OP_NCREF:
1683 case OP_NRREF:
1684 case OP_NOT_WORD_BOUNDARY:
1685 case OP_PRUNE:
1686 case OP_REVERSE:
1687 case OP_RREF:
1688 case OP_SET_SOM:
1689 case OP_SKIP:
1690 case OP_SOD:
1691 case OP_SOM:
1692 case OP_THEN:
1693 case OP_WORD_BOUNDARY:
1694 cc += _pcre_OP_lengths[*cc];
1695 break;
1696
1697 /* Handle literal characters */
1698
1699 case OP_CHAR:
1700 case OP_CHARI:
1701 case OP_NOT:
1702 case OP_NOTI:
1703 branchlength++;
1704 cc += 2;
1705 #ifdef SUPPORT_UTF8
1706 if (utf8 && cc[-1] >= 0xc0) cc += _pcre_utf8_table4[cc[-1] & 0x3f];
1707 #endif
1708 break;
1709
1710 /* Handle exact repetitions. The count is already in characters, but we
1711 need to skip over a multibyte character in UTF8 mode. */
1712
1713 case OP_EXACT:
1714 case OP_EXACTI:
1715 case OP_NOTEXACT:
1716 case OP_NOTEXACTI:
1717 branchlength += GET2(cc,1);
1718 cc += 4;
1719 #ifdef SUPPORT_UTF8
1720 if (utf8 && cc[-1] >= 0xc0) cc += _pcre_utf8_table4[cc[-1] & 0x3f];
1721 #endif
1722 break;
1723
1724 case OP_TYPEEXACT:
1725 branchlength += GET2(cc,1);
1726 if (cc[3] == OP_PROP || cc[3] == OP_NOTPROP) cc += 2;
1727 cc += 4;
1728 break;
1729
1730 /* Handle single-char matchers */
1731
1732 case OP_PROP:
1733 case OP_NOTPROP:
1734 cc += 2;
1735 /* Fall through */
1736
1737 case OP_HSPACE:
1738 case OP_VSPACE:
1739 case OP_NOT_HSPACE:
1740 case OP_NOT_VSPACE:
1741 case OP_NOT_DIGIT:
1742 case OP_DIGIT:
1743 case OP_NOT_WHITESPACE:
1744 case OP_WHITESPACE:
1745 case OP_NOT_WORDCHAR:
1746 case OP_WORDCHAR:
1747 case OP_ANY:
1748 case OP_ALLANY:
1749 branchlength++;
1750 cc++;
1751 break;
1752
1753 /* The single-byte matcher isn't allowed. This only happens in UTF-8 mode;
1754 otherwise \C is coded as OP_ALLANY. */
1755
1756 case OP_ANYBYTE:
1757 return -2;
1758
1759 /* Check a class for variable quantification */
1760
1761 #ifdef SUPPORT_UTF8
1762 case OP_XCLASS:
1763 cc += GET(cc, 1) - 33;
1764 /* Fall through */
1765 #endif
1766
1767 case OP_CLASS:
1768 case OP_NCLASS:
1769 cc += 33;
1770
1771 switch (*cc)
1772 {
1773 case OP_CRPLUS:
1774 case OP_CRMINPLUS:
1775 case OP_CRSTAR:
1776 case OP_CRMINSTAR:
1777 case OP_CRQUERY:
1778 case OP_CRMINQUERY:
1779 return -1;
1780
1781 case OP_CRRANGE:
1782 case OP_CRMINRANGE:
1783 if (GET2(cc,1) != GET2(cc,3)) return -1;
1784 branchlength += GET2(cc,1);
1785 cc += 5;
1786 break;
1787
1788 default:
1789 branchlength++;
1790 }
1791 break;
1792
1793 /* Anything else is variable length */
1794
1795 case OP_ANYNL:
1796 case OP_BRAMINZERO:
1797 case OP_BRAPOS:
1798 case OP_BRAPOSZERO:
1799 case OP_BRAZERO:
1800 case OP_CBRAPOS:
1801 case OP_EXTUNI:
1802 case OP_KETRMAX:
1803 case OP_KETRMIN:
1804 case OP_KETRPOS:
1805 case OP_MINPLUS:
1806 case OP_MINPLUSI:
1807 case OP_MINQUERY:
1808 case OP_MINQUERYI:
1809 case OP_MINSTAR:
1810 case OP_MINSTARI:
1811 case OP_MINUPTO:
1812 case OP_MINUPTOI:
1813 case OP_NOTMINPLUS:
1814 case OP_NOTMINPLUSI:
1815 case OP_NOTMINQUERY:
1816 case OP_NOTMINQUERYI:
1817 case OP_NOTMINSTAR:
1818 case OP_NOTMINSTARI:
1819 case OP_NOTMINUPTO:
1820 case OP_NOTMINUPTOI:
1821 case OP_NOTPLUS:
1822 case OP_NOTPLUSI:
1823 case OP_NOTPOSPLUS:
1824 case OP_NOTPOSPLUSI:
1825 case OP_NOTPOSQUERY:
1826 case OP_NOTPOSQUERYI:
1827 case OP_NOTPOSSTAR:
1828 case OP_NOTPOSSTARI:
1829 case OP_NOTPOSUPTO:
1830 case OP_NOTPOSUPTOI:
1831 case OP_NOTQUERY:
1832 case OP_NOTQUERYI:
1833 case OP_NOTSTAR:
1834 case OP_NOTSTARI:
1835 case OP_NOTUPTO:
1836 case OP_NOTUPTOI:
1837 case OP_PLUS:
1838 case OP_PLUSI:
1839 case OP_POSPLUS:
1840 case OP_POSPLUSI:
1841 case OP_POSQUERY:
1842 case OP_POSQUERYI:
1843 case OP_POSSTAR:
1844 case OP_POSSTARI:
1845 case OP_POSUPTO:
1846 case OP_POSUPTOI:
1847 case OP_QUERY:
1848 case OP_QUERYI:
1849 case OP_REF:
1850 case OP_REFI:
1851 case OP_SBRA:
1852 case OP_SBRAPOS:
1853 case OP_SCBRA:
1854 case OP_SCBRAPOS:
1855 case OP_SCOND:
1856 case OP_SKIPZERO:
1857 case OP_STAR:
1858 case OP_STARI:
1859 case OP_TYPEMINPLUS:
1860 case OP_TYPEMINQUERY:
1861 case OP_TYPEMINSTAR:
1862 case OP_TYPEMINUPTO:
1863 case OP_TYPEPLUS:
1864 case OP_TYPEPOSPLUS:
1865 case OP_TYPEPOSQUERY:
1866 case OP_TYPEPOSSTAR:
1867 case OP_TYPEPOSUPTO:
1868 case OP_TYPEQUERY:
1869 case OP_TYPESTAR:
1870 case OP_TYPEUPTO:
1871 case OP_UPTO:
1872 case OP_UPTOI:
1873 return -1;
1874
1875 /* Catch unrecognized opcodes so that when new ones are added they
1876 are not forgotten, as has happened in the past. */
1877
1878 default:
1879 return -4;
1880 }
1881 }
1882 /* Control never gets here */
1883 }
1884
1885
1886
1887
1888 /*************************************************
1889 * Scan compiled regex for specific bracket *
1890 *************************************************/
1891
1892 /* This little function scans through a compiled pattern until it finds a
1893 capturing bracket with the given number, or, if the number is negative, an
1894 instance of OP_REVERSE for a lookbehind. The function is global in the C sense
1895 so that it can be called from pcre_study() when finding the minimum matching
1896 length.
1897
1898 Arguments:
1899 code points to start of expression
1900 utf8 TRUE in UTF-8 mode
1901 number the required bracket number or negative to find a lookbehind
1902
1903 Returns: pointer to the opcode for the bracket, or NULL if not found
1904 */
1905
1906 const uschar *
1907 _pcre_find_bracket(const uschar *code, BOOL utf8, int number)
1908 {
1909 for (;;)
1910 {
1911 register int c = *code;
1912
1913 if (c == OP_END) return NULL;
1914
1915 /* XCLASS is used for classes that cannot be represented just by a bit
1916 map. This includes negated single high-valued characters. The length in
1917 the table is zero; the actual length is stored in the compiled code. */
1918
1919 if (c == OP_XCLASS) code += GET(code, 1);
1920
1921 /* Handle recursion */
1922
1923 else if (c == OP_REVERSE)
1924 {
1925 if (number < 0) return (uschar *)code;
1926 code += _pcre_OP_lengths[c];
1927 }
1928
1929 /* Handle capturing bracket */
1930
1931 else if (c == OP_CBRA || c == OP_SCBRA ||
1932 c == OP_CBRAPOS || c == OP_SCBRAPOS)
1933 {
1934 int n = GET2(code, 1+LINK_SIZE);
1935 if (n == number) return (uschar *)code;
1936 code += _pcre_OP_lengths[c];
1937 }
1938
1939 /* Otherwise, we can get the item's length from the table, except that for
1940 repeated character types, we have to test for \p and \P, which have an extra
1941 two bytes of parameters, and for MARK/PRUNE/SKIP/THEN with an argument, we
1942 must add in its length. */
1943
1944 else
1945 {
1946 switch(c)
1947 {
1948 case OP_TYPESTAR:
1949 case OP_TYPEMINSTAR:
1950 case OP_TYPEPLUS:
1951 case OP_TYPEMINPLUS:
1952 case OP_TYPEQUERY:
1953 case OP_TYPEMINQUERY:
1954 case OP_TYPEPOSSTAR:
1955 case OP_TYPEPOSPLUS:
1956 case OP_TYPEPOSQUERY:
1957 if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
1958 break;
1959
1960 case OP_TYPEUPTO:
1961 case OP_TYPEMINUPTO:
1962 case OP_TYPEEXACT:
1963 case OP_TYPEPOSUPTO:
1964 if (code[3] == OP_PROP || code[3] == OP_NOTPROP) code += 2;
1965 break;
1966
1967 case OP_MARK:
1968 case OP_PRUNE_ARG:
1969 case OP_SKIP_ARG:
1970 code += code[1];
1971 break;
1972
1973 case OP_THEN_ARG:
1974 code += code[1];
1975 break;
1976 }
1977
1978 /* Add in the fixed length from the table */
1979
1980 code += _pcre_OP_lengths[c];
1981
1982 /* In UTF-8 mode, opcodes that are followed by a character may be followed by
1983 a multi-byte character. The length in the table is a minimum, so we have to
1984 arrange to skip the extra bytes. */
1985
1986 #ifdef SUPPORT_UTF8
1987 if (utf8) switch(c)
1988 {
1989 case OP_CHAR:
1990 case OP_CHARI:
1991 case OP_EXACT:
1992 case OP_EXACTI:
1993 case OP_UPTO:
1994 case OP_UPTOI:
1995 case OP_MINUPTO:
1996 case OP_MINUPTOI:
1997 case OP_POSUPTO:
1998 case OP_POSUPTOI:
1999 case OP_STAR:
2000 case OP_STARI:
2001 case OP_MINSTAR:
2002 case OP_MINSTARI:
2003 case OP_POSSTAR:
2004 case OP_POSSTARI:
2005 case OP_PLUS:
2006 case OP_PLUSI:
2007 case OP_MINPLUS:
2008 case OP_MINPLUSI:
2009 case OP_POSPLUS:
2010 case OP_POSPLUSI:
2011 case OP_QUERY:
2012 case OP_QUERYI:
2013 case OP_MINQUERY:
2014 case OP_MINQUERYI:
2015 case OP_POSQUERY:
2016 case OP_POSQUERYI:
2017 if (code[-1] >= 0xc0) code += _pcre_utf8_table4[code[-1] & 0x3f];
2018 break;
2019 }
2020 #else
2021 (void)(utf8); /* Keep compiler happy by referencing function argument */
2022 #endif
2023 }
2024 }
2025 }
2026
2027
2028
2029 /*************************************************
2030 * Scan compiled regex for recursion reference *
2031 *************************************************/
2032
2033 /* This little function scans through a compiled pattern until it finds an
2034 instance of OP_RECURSE.
2035
2036 Arguments:
2037 code points to start of expression
2038 utf8 TRUE in UTF-8 mode
2039
2040 Returns: pointer to the opcode for OP_RECURSE, or NULL if not found
2041 */
2042
2043 static const uschar *
2044 find_recurse(const uschar *code, BOOL utf8)
2045 {
2046 for (;;)
2047 {
2048 register int c = *code;
2049 if (c == OP_END) return NULL;
2050 if (c == OP_RECURSE) return code;
2051
2052 /* XCLASS is used for classes that cannot be represented just by a bit
2053 map. This includes negated single high-valued characters. The length in
2054 the table is zero; the actual length is stored in the compiled code. */
2055
2056 if (c == OP_XCLASS) code += GET(code, 1);
2057
2058 /* Otherwise, we can get the item's length from the table, except that for
2059 repeated character types, we have to test for \p and \P, which have an extra
2060 two bytes of parameters, and for MARK/PRUNE/SKIP/THEN with an argument, we
2061 must add in its length. */
2062
2063 else
2064 {
2065 switch(c)
2066 {
2067 case OP_TYPESTAR:
2068 case OP_TYPEMINSTAR:
2069 case OP_TYPEPLUS:
2070 case OP_TYPEMINPLUS:
2071 case OP_TYPEQUERY:
2072 case OP_TYPEMINQUERY:
2073 case OP_TYPEPOSSTAR:
2074 case OP_TYPEPOSPLUS:
2075 case OP_TYPEPOSQUERY:
2076 if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
2077 break;
2078
2079 case OP_TYPEPOSUPTO:
2080 case OP_TYPEUPTO:
2081 case OP_TYPEMINUPTO:
2082 case OP_TYPEEXACT:
2083 if (code[3] == OP_PROP || code[3] == OP_NOTPROP) code += 2;
2084 break;
2085
2086 case OP_MARK:
2087 case OP_PRUNE_ARG:
2088 case OP_SKIP_ARG:
2089 code += code[1];
2090 break;
2091
2092 case OP_THEN_ARG:
2093 code += code[1];
2094 break;
2095 }
2096
2097 /* Add in the fixed length from the table */
2098
2099 code += _pcre_OP_lengths[c];
2100
2101 /* In UTF-8 mode, opcodes that are followed by a character may be followed
2102 by a multi-byte character. The length in the table is a minimum, so we have
2103 to arrange to skip the extra bytes. */
2104
2105 #ifdef SUPPORT_UTF8
2106 if (utf8) switch(c)
2107 {
2108 case OP_CHAR:
2109 case OP_CHARI:
2110 case OP_EXACT:
2111 case OP_EXACTI:
2112 case OP_UPTO:
2113 case OP_UPTOI:
2114 case OP_MINUPTO:
2115 case OP_MINUPTOI:
2116 case OP_POSUPTO:
2117 case OP_POSUPTOI:
2118 case OP_STAR:
2119 case OP_STARI:
2120 case OP_MINSTAR:
2121 case OP_MINSTARI:
2122 case OP_POSSTAR:
2123 case OP_POSSTARI:
2124 case OP_PLUS:
2125 case OP_PLUSI:
2126 case OP_MINPLUS:
2127 case OP_MINPLUSI:
2128 case OP_POSPLUS:
2129 case OP_POSPLUSI:
2130 case OP_QUERY:
2131 case OP_QUERYI:
2132 case OP_MINQUERY:
2133 case OP_MINQUERYI:
2134 case OP_POSQUERY:
2135 case OP_POSQUERYI:
2136 if (code[-1] >= 0xc0) code += _pcre_utf8_table4[code[-1] & 0x3f];
2137 break;
2138 }
2139 #else
2140 (void)(utf8); /* Keep compiler happy by referencing function argument */
2141 #endif
2142 }
2143 }
2144 }
2145
2146
2147
2148 /*************************************************
2149 * Scan compiled branch for non-emptiness *
2150 *************************************************/
2151
2152 /* This function scans through a branch of a compiled pattern to see whether it
2153 can match the empty string or not. It is called from could_be_empty()
2154 below and from compile_branch() when checking for an unlimited repeat of a
2155 group that can match nothing. Note that first_significant_code() skips over
2156 backward and negative forward assertions when its final argument is TRUE. If we
2157 hit an unclosed bracket, we return "empty" - this means we've struck an inner
2158 bracket whose current branch will already have been scanned.
2159
2160 Arguments:
2161 code points to start of search
2162 endcode points to where to stop
2163 utf8 TRUE if in UTF8 mode
2164 cd contains pointers to tables etc.
2165
2166 Returns: TRUE if what is matched could be empty
2167 */
2168
2169 static BOOL
2170 could_be_empty_branch(const uschar *code, const uschar *endcode, BOOL utf8,
2171 compile_data *cd)
2172 {
2173 register int c;
2174 for (code = first_significant_code(code + _pcre_OP_lengths[*code], TRUE);
2175 code < endcode;
2176 code = first_significant_code(code + _pcre_OP_lengths[c], TRUE))
2177 {
2178 const uschar *ccode;
2179
2180 c = *code;
2181
2182 /* Skip over forward assertions; the other assertions are skipped by
2183 first_significant_code() with a TRUE final argument. */
2184
2185 if (c == OP_ASSERT)
2186 {
2187 do code += GET(code, 1); while (*code == OP_ALT);
2188 c = *code;
2189 continue;
2190 }
2191
2192 /* For a recursion/subroutine call, if its end has been reached, which
2193 implies a backward reference subroutine call, we can scan it. If it's a
2194 forward reference subroutine call, we can't. To detect forward reference
2195 we have to scan up the list that is kept in the workspace. This function is
2196 called only when doing the real compile, not during the pre-compile that
2197 measures the size of the compiled pattern. */
2198
2199 if (c == OP_RECURSE)
2200 {
2201 const uschar *scode;
2202 BOOL empty_branch;
2203
2204 /* Test for forward reference */
2205
2206 for (scode = cd->start_workspace; scode < cd->hwm; scode += LINK_SIZE)
2207 if (GET(scode, 0) == code + 1 - cd->start_code) return TRUE;
2208
2209 /* Not a forward reference, test for completed backward reference */
2210
2211 empty_branch = FALSE;
2212 scode = cd->start_code + GET(code, 1);
2213 if (GET(scode, 1) == 0) return TRUE; /* Unclosed */
2214
2215 /* Completed backwards reference */
2216
2217 do
2218 {
2219 if (could_be_empty_branch(scode, endcode, utf8, cd))
2220 {
2221 empty_branch = TRUE;
2222 break;
2223 }
2224 scode += GET(scode, 1);
2225 }
2226 while (*scode == OP_ALT);
2227
2228 if (!empty_branch) return FALSE; /* All branches are non-empty */
2229 continue;
2230 }
2231
2232 /* Groups with zero repeats can of course be empty; skip them. */
2233
2234 if (c == OP_BRAZERO || c == OP_BRAMINZERO || c == OP_SKIPZERO ||
2235 c == OP_BRAPOSZERO)
2236 {
2237 code += _pcre_OP_lengths[c];
2238 do code += GET(code, 1); while (*code == OP_ALT);
2239 c = *code;
2240 continue;
2241 }
2242
2243 /* A nested group that is already marked as "could be empty" can just be
2244 skipped. */
2245
2246 if (c == OP_SBRA || c == OP_SBRAPOS ||
2247 c == OP_SCBRA || c == OP_SCBRAPOS)
2248 {
2249 do code += GET(code, 1); while (*code == OP_ALT);
2250 c = *code;
2251 continue;
2252 }
2253
2254 /* For other groups, scan the branches. */
2255
2256 if (c == OP_BRA || c == OP_BRAPOS ||
2257 c == OP_CBRA || c == OP_CBRAPOS ||
2258 c == OP_ONCE || c == OP_ONCE_NC ||
2259 c == OP_COND)
2260 {
2261 BOOL empty_branch;
2262 if (GET(code, 1) == 0) return TRUE; /* Hit unclosed bracket */
2263
2264 /* If a conditional group has only one branch, there is a second, implied,
2265 empty branch, so just skip over the conditional, because it could be empty.
2266 Otherwise, scan the individual branches of the group. */
2267
2268 if (c == OP_COND && code[GET(code, 1)] != OP_ALT)
2269 code += GET(code, 1);
2270 else
2271 {
2272 empty_branch = FALSE;
2273 do
2274 {
2275 if (!empty_branch && could_be_empty_branch(code, endcode, utf8, cd))
2276 empty_branch = TRUE;
2277 code += GET(code, 1);
2278 }
2279 while (*code == OP_ALT);
2280 if (!empty_branch) return FALSE; /* All branches are non-empty */
2281 }
2282
2283 c = *code;
2284 continue;
2285 }
2286
2287 /* Handle the other opcodes */
2288
2289 switch (c)
2290 {
2291 /* Check for quantifiers after a class. XCLASS is used for classes that
2292 cannot be represented just by a bit map. This includes negated single
2293 high-valued characters. The length in _pcre_OP_lengths[] is zero; the
2294 actual length is stored in the compiled code, so we must update "code"
2295 here. */
2296
2297 #ifdef SUPPORT_UTF8
2298 case OP_XCLASS:
2299 ccode = code += GET(code, 1);
2300 goto CHECK_CLASS_REPEAT;
2301 #endif
2302
2303 case OP_CLASS:
2304 case OP_NCLASS:
2305 ccode = code + 33;
2306
2307 #ifdef SUPPORT_UTF8
2308 CHECK_CLASS_REPEAT:
2309 #endif
2310
2311 switch (*ccode)
2312 {
2313 case OP_CRSTAR: /* These could be empty; continue */
2314 case OP_CRMINSTAR:
2315 case OP_CRQUERY:
2316 case OP_CRMINQUERY:
2317 break;
2318
2319 default: /* Non-repeat => class must match */
2320 case OP_CRPLUS: /* These repeats aren't empty */
2321 case OP_CRMINPLUS:
2322 return FALSE;
2323
2324 case OP_CRRANGE:
2325 case OP_CRMINRANGE:
2326 if (GET2(ccode, 1) > 0) return FALSE; /* Minimum > 0 */
2327 break;
2328 }
2329 break;
2330
2331 /* Opcodes that must match a character */
2332
2333 case OP_PROP:
2334 case OP_NOTPROP:
2335 case OP_EXTUNI:
2336 case OP_NOT_DIGIT:
2337 case OP_DIGIT:
2338 case OP_NOT_WHITESPACE:
2339 case OP_WHITESPACE:
2340 case OP_NOT_WORDCHAR:
2341 case OP_WORDCHAR:
2342 case OP_ANY:
2343 case OP_ALLANY:
2344 case OP_ANYBYTE:
2345 case OP_CHAR:
2346 case OP_CHARI:
2347 case OP_NOT:
2348 case OP_NOTI:
2349 case OP_PLUS:
2350 case OP_MINPLUS:
2351 case OP_POSPLUS:
2352 case OP_EXACT:
2353 case OP_NOTPLUS:
2354 case OP_NOTMINPLUS:
2355 case OP_NOTPOSPLUS:
2356 case OP_NOTEXACT:
2357 case OP_TYPEPLUS:
2358 case OP_TYPEMINPLUS:
2359 case OP_TYPEPOSPLUS:
2360 case OP_TYPEEXACT:
2361 return FALSE;
2362
2363 /* These are going to continue, as they may be empty, but we have to
2364 fudge the length for the \p and \P cases. */
2365
2366 case OP_TYPESTAR:
2367 case OP_TYPEMINSTAR:
2368 case OP_TYPEPOSSTAR:
2369 case OP_TYPEQUERY:
2370 case OP_TYPEMINQUERY:
2371 case OP_TYPEPOSQUERY:
2372 if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
2373 break;
2374
2375 /* Same for these */
2376
2377 case OP_TYPEUPTO:
2378 case OP_TYPEMINUPTO:
2379 case OP_TYPEPOSUPTO:
2380 if (code[3] == OP_PROP || code[3] == OP_NOTPROP) code += 2;
2381 break;
2382
2383 /* End of branch */
2384
2385 case OP_KET:
2386 case OP_KETRMAX:
2387 case OP_KETRMIN:
2388 case OP_KETRPOS:
2389 case OP_ALT:
2390 return TRUE;
2391
2392 /* In UTF-8 mode, STAR, MINSTAR, POSSTAR, QUERY, MINQUERY, POSQUERY, UPTO,
2393 MINUPTO, and POSUPTO may be followed by a multibyte character */
2394
2395 #ifdef SUPPORT_UTF8
2396 case OP_STAR:
2397 case OP_STARI:
2398 case OP_MINSTAR:
2399 case OP_MINSTARI:
2400 case OP_POSSTAR:
2401 case OP_POSSTARI:
2402 case OP_QUERY:
2403 case OP_QUERYI:
2404 case OP_MINQUERY:
2405 case OP_MINQUERYI:
2406 case OP_POSQUERY:
2407 case OP_POSQUERYI:
2408 if (utf8 && code[1] >= 0xc0) code += _pcre_utf8_table4[code[1] & 0x3f];
2409 break;
2410
2411 case OP_UPTO:
2412 case OP_UPTOI:
2413 case OP_MINUPTO:
2414 case OP_MINUPTOI:
2415 case OP_POSUPTO:
2416 case OP_POSUPTOI:
2417 if (utf8 && code[3] >= 0xc0) code += _pcre_utf8_table4[code[3] & 0x3f];
2418 break;
2419 #endif
2420
2421 /* MARK, and PRUNE/SKIP/THEN with an argument must skip over the argument
2422 string. */
2423
2424 case OP_MARK:
2425 case OP_PRUNE_ARG:
2426 case OP_SKIP_ARG:
2427 code += code[1];
2428 break;
2429
2430 case OP_THEN_ARG:
2431 code += code[1];
2432 break;
2433
2434 /* None of the remaining opcodes are required to match a character. */
2435
2436 default:
2437 break;
2438 }
2439 }
2440
2441 return TRUE;
2442 }
2443
2444
2445
2446 /*************************************************
2447 * Scan compiled regex for non-emptiness *
2448 *************************************************/
2449
2450 /* This function is called to check for left recursive calls. We want to check
2451 the current branch of the current pattern to see if it could match the empty
2452 string. If it could, we must look outwards for branches at other levels,
2453 stopping when we pass beyond the bracket which is the subject of the recursion.
2454 This function is called only during the real compile, not during the
2455 pre-compile.
2456
2457 Arguments:
2458 code points to start of the recursion
2459 endcode points to where to stop (current RECURSE item)
2460 bcptr points to the chain of current (unclosed) branch starts
2461 utf8 TRUE if in UTF-8 mode
2462 cd pointers to tables etc
2463
2464 Returns: TRUE if what is matched could be empty
2465 */
2466
2467 static BOOL
2468 could_be_empty(const uschar *code, const uschar *endcode, branch_chain *bcptr,
2469 BOOL utf8, compile_data *cd)
2470 {
2471 while (bcptr != NULL && bcptr->current_branch >= code)
2472 {
2473 if (!could_be_empty_branch(bcptr->current_branch, endcode, utf8, cd))
2474 return FALSE;
2475 bcptr = bcptr->outer;
2476 }
2477 return TRUE;
2478 }
2479
2480
2481
2482 /*************************************************
2483 * Check for POSIX class syntax *
2484 *************************************************/
2485
2486 /* This function is called when the sequence "[:" or "[." or "[=" is
2487 encountered in a character class. It checks whether this is followed by a
2488 sequence of characters terminated by a matching ":]" or ".]" or "=]". If we
2489 reach an unescaped ']' without the special preceding character, return FALSE.
2490
2491 Originally, this function only recognized a sequence of letters between the
2492 terminators, but it seems that Perl recognizes any sequence of characters,
2493 though of course unknown POSIX names are subsequently rejected. Perl gives an
2494 "Unknown POSIX class" error for [:f\oo:] for example, where previously PCRE
2495 didn't consider this to be a POSIX class. Likewise for [:1234:].
2496
2497 The problem in trying to be exactly like Perl is in the handling of escapes. We
2498 have to be sure that [abc[:x\]pqr] is *not* treated as containing a POSIX
2499 class, but [abc[:x\]pqr:]] is (so that an error can be generated). The code
2500 below handles the special case of \], but does not try to do any other escape
2501 processing. This makes it different from Perl for cases such as [:l\ower:]
2502 where Perl recognizes it as the POSIX class "lower" but PCRE does not recognize
2503 "l\ower". This is a lesser evil that not diagnosing bad classes when Perl does,
2504 I think.
2505
2506 A user pointed out that PCRE was rejecting [:a[:digit:]] whereas Perl was not.
2507 It seems that the appearance of a nested POSIX class supersedes an apparent
2508 external class. For example, [:a[:digit:]b:] matches "a", "b", ":", or
2509 a digit.
2510
2511 In Perl, unescaped square brackets may also appear as part of class names. For
2512 example, [:a[:abc]b:] gives unknown POSIX class "[:abc]b:]". However, for
2513 [:a[:abc]b][b:] it gives unknown POSIX class "[:abc]b][b:]", which does not
2514 seem right at all. PCRE does not allow closing square brackets in POSIX class
2515 names.
2516
2517 Arguments:
2518 ptr pointer to the initial [
2519 endptr where to return the end pointer
2520
2521 Returns: TRUE or FALSE
2522 */
2523
2524 static BOOL
2525 check_posix_syntax(const uschar *ptr, const uschar **endptr)
2526 {
2527 int terminator; /* Don't combine these lines; the Solaris cc */
2528 terminator = *(++ptr); /* compiler warns about "non-constant" initializer. */
2529 for (++ptr; *ptr != 0; ptr++)
2530 {
2531 if (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET)
2532 ptr++;
2533 else if (*ptr == CHAR_RIGHT_SQUARE_BRACKET) return FALSE;
2534 else
2535 {
2536 if (*ptr == terminator && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET)
2537 {
2538 *endptr = ptr;
2539 return TRUE;
2540 }
2541 if (*ptr == CHAR_LEFT_SQUARE_BRACKET &&
2542 (ptr[1] == CHAR_COLON || ptr[1] == CHAR_DOT ||
2543 ptr[1] == CHAR_EQUALS_SIGN) &&
2544 check_posix_syntax(ptr, endptr))
2545 return FALSE;
2546 }
2547 }
2548 return FALSE;
2549 }
2550
2551
2552
2553
2554 /*************************************************
2555 * Check POSIX class name *
2556 *************************************************/
2557
2558 /* This function is called to check the name given in a POSIX-style class entry
2559 such as [:alnum:].
2560
2561 Arguments:
2562 ptr points to the first letter
2563 len the length of the name
2564
2565 Returns: a value representing the name, or -1 if unknown
2566 */
2567
2568 static int
2569 check_posix_name(const uschar *ptr, int len)
2570 {
2571 const char *pn = posix_names;
2572 register int yield = 0;
2573 while (posix_name_lengths[yield] != 0)
2574 {
2575 if (len == posix_name_lengths[yield] &&
2576 strncmp((const char *)ptr, pn, len) == 0) return yield;
2577 pn += posix_name_lengths[yield] + 1;
2578 yield++;
2579 }
2580 return -1;
2581 }
2582
2583
2584 /*************************************************
2585 * Adjust OP_RECURSE items in repeated group *
2586 *************************************************/
2587
2588 /* OP_RECURSE items contain an offset from the start of the regex to the group
2589 that is referenced. This means that groups can be replicated for fixed
2590 repetition simply by copying (because the recursion is allowed to refer to
2591 earlier groups that are outside the current group). However, when a group is
2592 optional (i.e. the minimum quantifier is zero), OP_BRAZERO or OP_SKIPZERO is
2593 inserted before it, after it has been compiled. This means that any OP_RECURSE
2594 items within it that refer to the group itself or any contained groups have to
2595 have their offsets adjusted. That one of the jobs of this function. Before it
2596 is called, the partially compiled regex must be temporarily terminated with
2597 OP_END.
2598
2599 This function has been extended with the possibility of forward references for
2600 recursions and subroutine calls. It must also check the list of such references
2601 for the group we are dealing with. If it finds that one of the recursions in
2602 the current group is on this list, it adjusts the offset in the list, not the
2603 value in the reference (which is a group number).
2604
2605 Arguments:
2606 group points to the start of the group
2607 adjust the amount by which the group is to be moved
2608 utf8 TRUE in UTF-8 mode
2609 cd contains pointers to tables etc.
2610 save_hwm the hwm forward reference pointer at the start of the group
2611
2612 Returns: nothing
2613 */
2614
2615 static void
2616 adjust_recurse(uschar *group, int adjust, BOOL utf8, compile_data *cd,
2617 uschar *save_hwm)
2618 {
2619 uschar *ptr = group;
2620
2621 while ((ptr = (uschar *)find_recurse(ptr, utf8)) != NULL)
2622 {
2623 int offset;
2624 uschar *hc;
2625
2626 /* See if this recursion is on the forward reference list. If so, adjust the
2627 reference. */
2628
2629 for (hc = save_hwm; hc < cd->hwm; hc += LINK_SIZE)
2630 {
2631 offset = GET(hc, 0);
2632 if (cd->start_code + offset == ptr + 1)
2633 {
2634 PUT(hc, 0, offset + adjust);
2635 break;
2636 }
2637 }
2638
2639 /* Otherwise, adjust the recursion offset if it's after the start of this
2640 group. */
2641
2642 if (hc >= cd->hwm)
2643 {
2644 offset = GET(ptr, 1);
2645 if (cd->start_code + offset >= group) PUT(ptr, 1, offset + adjust);
2646 }
2647
2648 ptr += 1 + LINK_SIZE;
2649 }
2650 }
2651
2652
2653
2654 /*************************************************
2655 * Insert an automatic callout point *
2656 *************************************************/
2657
2658 /* This function is called when the PCRE_AUTO_CALLOUT option is set, to insert
2659 callout points before each pattern item.
2660
2661 Arguments:
2662 code current code pointer
2663 ptr current pattern pointer
2664 cd pointers to tables etc
2665
2666 Returns: new code pointer
2667 */
2668
2669 static uschar *
2670 auto_callout(uschar *code, const uschar *ptr, compile_data *cd)
2671 {
2672 *code++ = OP_CALLOUT;
2673 *code++ = 255;
2674 PUT(code, 0, (int)(ptr - cd->start_pattern)); /* Pattern offset */
2675 PUT(code, LINK_SIZE, 0); /* Default length */
2676 return code + 2*LINK_SIZE;
2677 }
2678
2679
2680
2681 /*************************************************
2682 * Complete a callout item *
2683 *************************************************/
2684
2685 /* A callout item contains the length of the next item in the pattern, which
2686 we can't fill in till after we have reached the relevant point. This is used
2687 for both automatic and manual callouts.
2688
2689 Arguments:
2690 previous_callout points to previous callout item
2691 ptr current pattern pointer
2692 cd pointers to tables etc
2693
2694 Returns: nothing
2695 */
2696
2697 static void
2698 complete_callout(uschar *previous_callout, const uschar *ptr, compile_data *cd)
2699 {
2700 int length = (int)(ptr - cd->start_pattern - GET(previous_callout, 2));
2701 PUT(previous_callout, 2 + LINK_SIZE, length);
2702 }
2703
2704
2705
2706 #ifdef SUPPORT_UCP
2707 /*************************************************
2708 * Get othercase range *
2709 *************************************************/
2710
2711 /* This function is passed the start and end of a class range, in UTF-8 mode
2712 with UCP support. It searches up the characters, looking for internal ranges of
2713 characters in the "other" case. Each call returns the next one, updating the
2714 start address.
2715
2716 Arguments:
2717 cptr points to starting character value; updated
2718 d end value
2719 ocptr where to put start of othercase range
2720 odptr where to put end of othercase range
2721
2722 Yield: TRUE when range returned; FALSE when no more
2723 */
2724
2725 static BOOL
2726 get_othercase_range(unsigned int *cptr, unsigned int d, unsigned int *ocptr,
2727 unsigned int *odptr)
2728 {
2729 unsigned int c, othercase, next;
2730
2731 for (c = *cptr; c <= d; c++)
2732 { if ((othercase = UCD_OTHERCASE(c)) != c) break; }
2733
2734 if (c > d) return FALSE;
2735
2736 *ocptr = othercase;
2737 next = othercase + 1;
2738
2739 for (++c; c <= d; c++)
2740 {
2741 if (UCD_OTHERCASE(c) != next) break;
2742 next++;
2743 }
2744
2745 *odptr = next - 1;
2746 *cptr = c;
2747
2748 return TRUE;
2749 }
2750
2751
2752
2753 /*************************************************
2754 * Check a character and a property *
2755 *************************************************/
2756
2757 /* This function is called by check_auto_possessive() when a property item
2758 is adjacent to a fixed character.
2759
2760 Arguments:
2761 c the character
2762 ptype the property type
2763 pdata the data for the type
2764 negated TRUE if it's a negated property (\P or \p{^)
2765
2766 Returns: TRUE if auto-possessifying is OK
2767 */
2768
2769 static BOOL
2770 check_char_prop(int c, int ptype, int pdata, BOOL negated)
2771 {
2772 const ucd_record *prop = GET_UCD(c);
2773 switch(ptype)
2774 {
2775 case PT_LAMP:
2776 return (prop->chartype == ucp_Lu ||
2777 prop->chartype == ucp_Ll ||
2778 prop->chartype == ucp_Lt) == negated;
2779
2780 case PT_GC:
2781 return (pdata == _pcre_ucp_gentype[prop->chartype]) == negated;
2782
2783 case PT_PC:
2784 return (pdata == prop->chartype) == negated;
2785
2786 case PT_SC:
2787 return (pdata == prop->script) == negated;
2788
2789 /* These are specials */
2790
2791 case PT_ALNUM:
2792 return (_pcre_ucp_gentype[prop->chartype] == ucp_L ||
2793 _pcre_ucp_gentype[prop->chartype] == ucp_N) == negated;
2794
2795 case PT_SPACE: /* Perl space */
2796 return (_pcre_ucp_gentype[prop->chartype] == ucp_Z ||
2797 c == CHAR_HT || c == CHAR_NL || c == CHAR_FF || c == CHAR_CR)
2798 == negated;
2799
2800 case PT_PXSPACE: /* POSIX space */
2801 return (_pcre_ucp_gentype[prop->chartype] == ucp_Z ||
2802 c == CHAR_HT || c == CHAR_NL || c == CHAR_VT ||
2803 c == CHAR_FF || c == CHAR_CR)
2804 == negated;
2805
2806 case PT_WORD:
2807 return (_pcre_ucp_gentype[prop->chartype] == ucp_L ||
2808 _pcre_ucp_gentype[prop->chartype] == ucp_N ||
2809 c == CHAR_UNDERSCORE) == negated;
2810 }
2811 return FALSE;
2812 }
2813 #endif /* SUPPORT_UCP */
2814
2815
2816
2817 /*************************************************
2818 * Check if auto-possessifying is possible *
2819 *************************************************/
2820
2821 /* This function is called for unlimited repeats of certain items, to see
2822 whether the next thing could possibly match the repeated item. If not, it makes
2823 sense to automatically possessify the repeated item.
2824
2825 Arguments:
2826 previous pointer to the repeated opcode
2827 utf8 TRUE in UTF-8 mode
2828 ptr next character in pattern
2829 options options bits
2830 cd contains pointers to tables etc.
2831
2832 Returns: TRUE if possessifying is wanted
2833 */
2834
2835 static BOOL
2836 check_auto_possessive(const uschar *previous, BOOL utf8, const uschar *ptr,
2837 int options, compile_data *cd)
2838 {
2839 int c, next;
2840 int op_code = *previous++;
2841
2842 /* Skip whitespace and comments in extended mode */
2843
2844 if ((options & PCRE_EXTENDED) != 0)
2845 {
2846 for (;;)
2847 {
2848 while ((cd->ctypes[*ptr] & ctype_space) != 0) ptr++;
2849 if (*ptr == CHAR_NUMBER_SIGN)
2850 {
2851 ptr++;
2852 while (*ptr != 0)
2853 {
2854 if (IS_NEWLINE(ptr)) { ptr += cd->nllen; break; }
2855 ptr++;
2856 #ifdef SUPPORT_UTF8
2857 if (utf8) while ((*ptr & 0xc0) == 0x80) ptr++;
2858 #endif
2859 }
2860 }
2861 else break;
2862 }
2863 }
2864
2865 /* If the next item is one that we can handle, get its value. A non-negative
2866 value is a character, a negative value is an escape value. */
2867
2868 if (*ptr == CHAR_BACKSLASH)
2869 {
2870 int temperrorcode = 0;
2871 next = check_escape(&ptr, &temperrorcode, cd->bracount, options, FALSE);
2872 if (temperrorcode != 0) return FALSE;
2873 ptr++; /* Point after the escape sequence */
2874 }
2875
2876 else if ((cd->ctypes[*ptr] & ctype_meta) == 0)
2877 {
2878 #ifdef SUPPORT_UTF8
2879 if (utf8) { GETCHARINC(next, ptr); } else
2880 #endif
2881 next = *ptr++;
2882 }
2883
2884 else return FALSE;
2885
2886 /* Skip whitespace and comments in extended mode */
2887
2888 if ((options & PCRE_EXTENDED) != 0)
2889 {
2890 for (;;)
2891 {
2892 while ((cd->ctypes[*ptr] & ctype_space) != 0) ptr++;
2893 if (*ptr == CHAR_NUMBER_SIGN)
2894 {
2895 ptr++;
2896 while (*ptr != 0)
2897 {
2898 if (IS_NEWLINE(ptr)) { ptr += cd->nllen; break; }
2899 ptr++;
2900 #ifdef SUPPORT_UTF8
2901 if (utf8) while ((*ptr & 0xc0) == 0x80) ptr++;
2902 #endif
2903 }
2904 }
2905 else break;
2906 }
2907 }
2908
2909 /* If the next thing is itself optional, we have to give up. */
2910
2911 if (*ptr == CHAR_ASTERISK || *ptr == CHAR_QUESTION_MARK ||
2912 strncmp((char *)ptr, STR_LEFT_CURLY_BRACKET STR_0 STR_COMMA, 3) == 0)
2913 return FALSE;
2914
2915 /* Now compare the next item with the previous opcode. First, handle cases when
2916 the next item is a character. */
2917
2918 if (next >= 0) switch(op_code)
2919 {
2920 case OP_CHAR:
2921 #ifdef SUPPORT_UTF8
2922 GETCHARTEST(c, previous);
2923 #else
2924 c = *previous;
2925 #endif
2926 return c != next;
2927
2928 /* For CHARI (caseless character) we must check the other case. If we have
2929 Unicode property support, we can use it to test the other case of
2930 high-valued characters. */
2931
2932 case OP_CHARI:
2933 #ifdef SUPPORT_UTF8
2934 GETCHARTEST(c, previous);
2935 #else
2936 c = *previous;
2937 #endif
2938 if (c == next) return FALSE;
2939 #ifdef SUPPORT_UTF8
2940 if (utf8)
2941 {
2942 unsigned int othercase;
2943 if (next < 128) othercase = cd->fcc[next]; else
2944 #ifdef SUPPORT_UCP
2945 othercase = UCD_OTHERCASE((unsigned int)next);
2946 #else
2947 othercase = NOTACHAR;
2948 #endif
2949 return (unsigned int)c != othercase;
2950 }
2951 else
2952 #endif /* SUPPORT_UTF8 */
2953 return (c != cd->fcc[next]); /* Non-UTF-8 mode */
2954
2955 /* For OP_NOT and OP_NOTI, the data is always a single-byte character. These
2956 opcodes are not used for multi-byte characters, because they are coded using
2957 an XCLASS instead. */
2958
2959 case OP_NOT:
2960 return (c = *previous) == next;
2961
2962 case OP_NOTI:
2963 if ((c = *previous) == next) return TRUE;
2964 #ifdef SUPPORT_UTF8
2965 if (utf8)
2966 {
2967 unsigned int othercase;
2968 if (next < 128) othercase = cd->fcc[next]; else
2969 #ifdef SUPPORT_UCP
2970 othercase = UCD_OTHERCASE(next);
2971 #else
2972 othercase = NOTACHAR;
2973 #endif
2974 return (unsigned int)c == othercase;
2975 }
2976 else
2977 #endif /* SUPPORT_UTF8 */
2978 return (c == cd->fcc[next]); /* Non-UTF-8 mode */
2979
2980 /* Note that OP_DIGIT etc. are generated only when PCRE_UCP is *not* set.
2981 When it is set, \d etc. are converted into OP_(NOT_)PROP codes. */
2982
2983 case OP_DIGIT:
2984 return next > 127 || (cd->ctypes[next] & ctype_digit) == 0;
2985
2986 case OP_NOT_DIGIT:
2987 return next <= 127 && (cd->ctypes[next] & ctype_digit) != 0;
2988
2989 case OP_WHITESPACE:
2990 return next > 127 || (cd->ctypes[next] & ctype_space) == 0;
2991
2992 case OP_NOT_WHITESPACE:
2993 return next <= 127 && (cd->ctypes[next] & ctype_space) != 0;
2994
2995 case OP_WORDCHAR:
2996 return next > 127 || (cd->ctypes[next] & ctype_word) == 0;
2997
2998 case OP_NOT_WORDCHAR:
2999 return next <= 127 && (cd->ctypes[next] & ctype_word) != 0;
3000
3001 case OP_HSPACE:
3002 case OP_NOT_HSPACE:
3003 switch(next)
3004 {
3005 case 0x09:
3006 case 0x20:
3007 case 0xa0:
3008 case 0x1680:
3009 case 0x180e:
3010 case 0x2000:
3011 case 0x2001:
3012 case 0x2002:
3013 case 0x2003:
3014 case 0x2004:
3015 case 0x2005:
3016 case 0x2006:
3017 case 0x2007:
3018 case 0x2008:
3019 case 0x2009:
3020 case 0x200A:
3021 case 0x202f:
3022 case 0x205f:
3023 case 0x3000:
3024 return op_code == OP_NOT_HSPACE;
3025 default:
3026 return op_code != OP_NOT_HSPACE;
3027 }
3028
3029 case OP_ANYNL:
3030 case OP_VSPACE:
3031 case OP_NOT_VSPACE:
3032 switch(next)
3033 {
3034 case 0x0a:
3035 case 0x0b:
3036 case 0x0c:
3037 case 0x0d:
3038 case 0x85:
3039 case 0x2028:
3040 case 0x2029:
3041 return op_code == OP_NOT_VSPACE;
3042 default:
3043 return op_code != OP_NOT_VSPACE;
3044 }
3045
3046 #ifdef SUPPORT_UCP
3047 case OP_PROP:
3048 return check_char_prop(next, previous[0], previous[1], FALSE);
3049
3050 case OP_NOTPROP:
3051 return check_char_prop(next, previous[0], previous[1], TRUE);
3052 #endif
3053
3054 default:
3055 return FALSE;
3056 }
3057
3058
3059 /* Handle the case when the next item is \d, \s, etc. Note that when PCRE_UCP
3060 is set, \d turns into ESC_du rather than ESC_d, etc., so ESC_d etc. are
3061 generated only when PCRE_UCP is *not* set, that is, when only ASCII
3062 characteristics are recognized. Similarly, the opcodes OP_DIGIT etc. are
3063 replaced by OP_PROP codes when PCRE_UCP is set. */
3064
3065 switch(op_code)
3066 {
3067 case OP_CHAR:
3068 case OP_CHARI:
3069 #ifdef SUPPORT_UTF8
3070 GETCHARTEST(c, previous);
3071 #else
3072 c = *previous;
3073 #endif
3074 switch(-next)
3075 {
3076 case ESC_d:
3077 return c > 127 || (cd->ctypes[c] & ctype_digit) == 0;
3078
3079 case ESC_D:
3080 return c <= 127 && (cd->ctypes[c] & ctype_digit) != 0;
3081
3082 case ESC_s:
3083 return c > 127 || (cd->ctypes[c] & ctype_space) == 0;
3084
3085 case ESC_S:
3086 return c <= 127 && (cd->ctypes[c] & ctype_space) != 0;
3087
3088 case ESC_w:
3089 return c > 127 || (cd->ctypes[c] & ctype_word) == 0;
3090
3091 case ESC_W:
3092 return c <= 127 && (cd->ctypes[c] & ctype_word) != 0;
3093
3094 case ESC_h:
3095 case ESC_H:
3096 switch(c)
3097 {
3098 case 0x09:
3099 case 0x20:
3100 case 0xa0:
3101 case 0x1680:
3102 case 0x180e:
3103 case 0x2000:
3104 case 0x2001:
3105 case 0x2002:
3106 case 0x2003:
3107 case 0x2004:
3108 case 0x2005:
3109 case 0x2006:
3110 case 0x2007:
3111 case 0x2008:
3112 case 0x2009:
3113 case 0x200A:
3114 case 0x202f:
3115 case 0x205f:
3116 case 0x3000:
3117 return -next != ESC_h;
3118 default:
3119 return -next == ESC_h;
3120 }
3121
3122 case ESC_v:
3123 case ESC_V:
3124 switch(c)
3125 {
3126 case 0x0a:
3127 case 0x0b:
3128 case 0x0c:
3129 case 0x0d:
3130 case 0x85:
3131 case 0x2028:
3132 case 0x2029:
3133 return -next != ESC_v;
3134 default:
3135 return -next == ESC_v;
3136 }
3137
3138 /* When PCRE_UCP is set, these values get generated for \d etc. Find
3139 their substitutions and process them. The result will always be either
3140 -ESC_p or -ESC_P. Then fall through to process those values. */
3141
3142 #ifdef SUPPORT_UCP
3143 case ESC_du:
3144 case ESC_DU:
3145 case ESC_wu:
3146 case ESC_WU:
3147 case ESC_su:
3148 case ESC_SU:
3149 {
3150 int temperrorcode = 0;
3151 ptr = substitutes[-next - ESC_DU];
3152 next = check_escape(&ptr, &temperrorcode, 0, options, FALSE);
3153 if (temperrorcode != 0) return FALSE;
3154 ptr++; /* For compatibility */
3155 }
3156 /* Fall through */
3157
3158 case ESC_p:
3159 case ESC_P:
3160 {
3161 int ptype, pdata, errorcodeptr;
3162 BOOL negated;
3163
3164 ptr--; /* Make ptr point at the p or P */
3165 ptype = get_ucp(&ptr, &negated, &pdata, &errorcodeptr);
3166 if (ptype < 0) return FALSE;
3167 ptr++; /* Point past the final curly ket */
3168
3169 /* If the property item is optional, we have to give up. (When generated
3170 from \d etc by PCRE_UCP, this test will have been applied much earlier,
3171 to the original \d etc. At this point, ptr will point to a zero byte. */
3172
3173 if (*ptr == CHAR_ASTERISK || *ptr == CHAR_QUESTION_MARK ||
3174 strncmp((char *)ptr, STR_LEFT_CURLY_BRACKET STR_0 STR_COMMA, 3) == 0)
3175 return FALSE;
3176
3177 /* Do the property check. */
3178
3179 return check_char_prop(c, ptype, pdata, (next == -ESC_P) != negated);
3180 }
3181 #endif
3182
3183 default:
3184 return FALSE;
3185 }
3186
3187 /* In principle, support for Unicode properties should be integrated here as
3188 well. It means re-organizing the above code so as to get hold of the property
3189 values before switching on the op-code. However, I wonder how many patterns
3190 combine ASCII \d etc with Unicode properties? (Note that if PCRE_UCP is set,
3191 these op-codes are never generated.) */
3192
3193 case OP_DIGIT:
3194 return next == -ESC_D || next == -ESC_s || next == -ESC_W ||
3195 next == -ESC_h || next == -ESC_v || next == -ESC_R;
3196
3197 case OP_NOT_DIGIT:
3198 return next == -ESC_d;
3199
3200 case OP_WHITESPACE:
3201 return next == -ESC_S || next == -ESC_d || next == -ESC_w || next == -ESC_R;
3202
3203 case OP_NOT_WHITESPACE:
3204 return next == -ESC_s || next == -ESC_h || next == -ESC_v;
3205
3206 case OP_HSPACE:
3207 return next == -ESC_S || next == -ESC_H || next == -ESC_d ||
3208 next == -ESC_w || next == -ESC_v || next == -ESC_R;
3209
3210 case OP_NOT_HSPACE:
3211 return next == -ESC_h;
3212
3213 /* Can't have \S in here because VT matches \S (Perl anomaly) */
3214 case OP_ANYNL:
3215 case OP_VSPACE:
3216 return next == -ESC_V || next == -ESC_d || next == -ESC_w;
3217
3218 case OP_NOT_VSPACE:
3219 return next == -ESC_v || next == -ESC_R;
3220
3221 case OP_WORDCHAR:
3222 return next == -ESC_W || next == -ESC_s || next == -ESC_h ||
3223 next == -ESC_v || next == -ESC_R;
3224
3225 case OP_NOT_WORDCHAR:
3226 return next == -ESC_w || next == -ESC_d;
3227
3228 default:
3229 return FALSE;
3230 }
3231
3232 /* Control does not reach here */
3233 }
3234
3235
3236
3237 /*************************************************
3238 * Compile one branch *
3239 *************************************************/
3240
3241 /* Scan the pattern, compiling it into the a vector. If the options are
3242 changed during the branch, the pointer is used to change the external options
3243 bits. This function is used during the pre-compile phase when we are trying
3244 to find out the amount of memory needed, as well as during the real compile
3245 phase. The value of lengthptr distinguishes the two phases.
3246
3247 Arguments:
3248 optionsptr pointer to the option bits
3249 codeptr points to the pointer to the current code point
3250 ptrptr points to the current pattern pointer
3251 errorcodeptr points to error code variable
3252 firstbyteptr set to initial literal character, or < 0 (REQ_UNSET, REQ_NONE)
3253 reqbyteptr set to the last literal character required, else < 0
3254 bcptr points to current branch chain
3255 cond_depth conditional nesting depth
3256 cd contains pointers to tables etc.
3257 lengthptr NULL during the real compile phase
3258 points to length accumulator during pre-compile phase
3259
3260 Returns: TRUE on success
3261 FALSE, with *errorcodeptr set non-zero on error
3262 */
3263
3264 static BOOL
3265 compile_branch(int *optionsptr, uschar **codeptr, const uschar **ptrptr,
3266 int *errorcodeptr, int *firstbyteptr, int *reqbyteptr, branch_chain *bcptr,
3267 int cond_depth, compile_data *cd, int *lengthptr)
3268 {
3269 int repeat_type, op_type;
3270 int repeat_min = 0, repeat_max = 0; /* To please picky compilers */
3271 int bravalue = 0;
3272 int greedy_default, greedy_non_default;
3273 int firstbyte, reqbyte;
3274 int zeroreqbyte, zerofirstbyte;
3275 int req_caseopt, reqvary, tempreqvary;
3276 int options = *optionsptr; /* May change dynamically */
3277 int after_manual_callout = 0;
3278 int length_prevgroup = 0;
3279 register int c;
3280 register uschar *code = *codeptr;
3281 uschar *last_code = code;
3282 uschar *orig_code = code;
3283 uschar *tempcode;
3284 BOOL inescq = FALSE;
3285 BOOL groupsetfirstbyte = FALSE;
3286 const uschar *ptr = *ptrptr;
3287 const uschar *tempptr;
3288 const uschar *nestptr = NULL;
3289 uschar *previous = NULL;
3290 uschar *previous_callout = NULL;
3291 uschar *save_hwm = NULL;
3292 uschar classbits[32];
3293
3294 /* We can fish out the UTF-8 setting once and for all into a BOOL, but we
3295 must not do this for other options (e.g. PCRE_EXTENDED) because they may change
3296 dynamically as we process the pattern. */
3297
3298 #ifdef SUPPORT_UTF8
3299 BOOL class_utf8;
3300 BOOL utf8 = (options & PCRE_UTF8) != 0;
3301 uschar *class_utf8data;
3302 uschar *class_utf8data_base;
3303 uschar utf8_char[6];
3304 #else
3305 BOOL utf8 = FALSE;
3306 #endif
3307
3308 #ifdef PCRE_DEBUG
3309 if (lengthptr != NULL) DPRINTF((">> start branch\n"));
3310 #endif
3311
3312 /* Set up the default and non-default settings for greediness */
3313
3314 greedy_default = ((options & PCRE_UNGREEDY) != 0);
3315 greedy_non_default = greedy_default ^ 1;
3316
3317 /* Initialize no first byte, no required byte. REQ_UNSET means "no char
3318 matching encountered yet". It gets changed to REQ_NONE if we hit something that
3319 matches a non-fixed char first char; reqbyte just remains unset if we never
3320 find one.
3321
3322 When we hit a repeat whose minimum is zero, we may have to adjust these values
3323 to take the zero repeat into account. This is implemented by setting them to
3324 zerofirstbyte and zeroreqbyte when such a repeat is encountered. The individual
3325 item types that can be repeated set these backoff variables appropriately. */
3326
3327 firstbyte = reqbyte = zerofirstbyte = zeroreqbyte = REQ_UNSET;
3328
3329 /* The variable req_caseopt contains either the REQ_CASELESS value or zero,
3330 according to the current setting of the caseless flag. REQ_CASELESS is a bit
3331 value > 255. It is added into the firstbyte or reqbyte variables to record the
3332 case status of the value. This is used only for ASCII characters. */
3333
3334 req_caseopt = ((options & PCRE_CASELESS) != 0)? REQ_CASELESS : 0;
3335
3336 /* Switch on next character until the end of the branch */
3337
3338 for (;; ptr++)
3339 {
3340 BOOL negate_class;
3341 BOOL should_flip_negation;
3342 BOOL possessive_quantifier;
3343 BOOL is_quantifier;
3344 BOOL is_recurse;
3345 BOOL reset_bracount;
3346 int class_charcount;
3347 int class_lastchar;
3348 int newoptions;
3349 int recno;
3350 int refsign;
3351 int skipbytes;
3352 int subreqbyte;
3353 int subfirstbyte;
3354 int terminator;
3355 int mclength;
3356 int tempbracount;
3357 uschar mcbuffer[8];
3358
3359 /* Get next byte in the pattern */
3360
3361 c = *ptr;
3362
3363 /* If we are at the end of a nested substitution, revert to the outer level
3364 string. Nesting only happens one level deep. */
3365
3366 if (c == 0 && nestptr != NULL)
3367 {
3368 ptr = nestptr;
3369 nestptr = NULL;
3370 c = *ptr;
3371 }
3372
3373 /* If we are in the pre-compile phase, accumulate the length used for the
3374 previous cycle of this loop. */
3375
3376 if (lengthptr != NULL)
3377 {
3378 #ifdef PCRE_DEBUG
3379 if (code > cd->hwm) cd->hwm = code; /* High water info */
3380 #endif
3381 if (code > cd->start_workspace + cd->workspace_size -
3382 WORK_SIZE_SAFETY_MARGIN) /* Check for overrun */
3383 {
3384 *errorcodeptr = ERR52;
3385 goto FAILED;
3386 }
3387
3388 /* There is at least one situation where code goes backwards: this is the
3389 case of a zero quantifier after a class (e.g. [ab]{0}). At compile time,
3390 the class is simply eliminated. However, it is created first, so we have to
3391 allow memory for it. Therefore, don't ever reduce the length at this point.
3392 */
3393
3394 if (code < last_code) code = last_code;
3395
3396 /* Paranoid check for integer overflow */
3397
3398 if (OFLOW_MAX - *lengthptr < code - last_code)
3399 {
3400 *errorcodeptr = ERR20;
3401 goto FAILED;
3402 }
3403
3404 *lengthptr += (int)(code - last_code);
3405 DPRINTF(("length=%d added %d c=%c\n", *lengthptr, (int)(code - last_code),
3406 c));
3407
3408 /* If "previous" is set and it is not at the start of the work space, move
3409 it back to there, in order to avoid filling up the work space. Otherwise,
3410 if "previous" is NULL, reset the current code pointer to the start. */
3411
3412 if (previous != NULL)
3413 {
3414 if (previous > orig_code)
3415 {
3416 memmove(orig_code, previous, code - previous);
3417 code -= previous - orig_code;
3418 previous = orig_code;
3419 }
3420 }
3421 else code = orig_code;
3422
3423 /* Remember where this code item starts so we can pick up the length
3424 next time round. */
3425
3426 last_code = code;
3427 }
3428
3429 /* In the real compile phase, just check the workspace used by the forward
3430 reference list. */
3431
3432 else if (cd->hwm > cd->start_workspace + cd->workspace_size -
3433 WORK_SIZE_SAFETY_MARGIN)
3434 {
3435 *errorcodeptr = ERR52;
3436 goto FAILED;
3437 }
3438
3439 /* If in \Q...\E, check for the end; if not, we have a literal */
3440
3441 if (inescq && c != 0)
3442 {
3443 if (c == CHAR_BACKSLASH && ptr[1] == CHAR_E)
3444 {
3445 inescq = FALSE;
3446 ptr++;
3447 continue;
3448 }
3449 else
3450 {
3451 if (previous_callout != NULL)
3452 {
3453 if (lengthptr == NULL) /* Don't attempt in pre-compile phase */
3454 complete_callout(previous_callout, ptr, cd);
3455 previous_callout = NULL;
3456 }
3457 if ((options & PCRE_AUTO_CALLOUT) != 0)
3458 {
3459 previous_callout = code;
3460 code = auto_callout(code, ptr, cd);
3461 }
3462 goto NORMAL_CHAR;
3463 }
3464 }
3465
3466 /* Fill in length of a previous callout, except when the next thing is
3467 a quantifier. */
3468
3469 is_quantifier =
3470 c == CHAR_ASTERISK || c == CHAR_PLUS || c == CHAR_QUESTION_MARK ||
3471 (c == CHAR_LEFT_CURLY_BRACKET && is_counted_repeat(ptr+1));
3472
3473 if (!is_quantifier && previous_callout != NULL &&
3474 after_manual_callout-- <= 0)
3475 {
3476 if (lengthptr == NULL) /* Don't attempt in pre-compile phase */
3477 complete_callout(previous_callout, ptr, cd);
3478 previous_callout = NULL;
3479 }
3480
3481 /* In extended mode, skip white space and comments. */
3482
3483 if ((options & PCRE_EXTENDED) != 0)
3484 {
3485 if ((cd->ctypes[c] & ctype_space) != 0) continue;
3486 if (c == CHAR_NUMBER_SIGN)
3487 {
3488 ptr++;
3489 while (*ptr != 0)
3490 {
3491 if (IS_NEWLINE(ptr)) { ptr += cd->nllen - 1; break; }
3492 ptr++;
3493 #ifdef SUPPORT_UTF8
3494 if (utf8) while ((*ptr & 0xc0) == 0x80) ptr++;
3495 #endif
3496 }
3497 if (*ptr != 0) continue;
3498
3499 /* Else fall through to handle end of string */
3500 c = 0;
3501 }
3502 }
3503
3504 /* No auto callout for quantifiers. */
3505
3506 if ((options & PCRE_AUTO_CALLOUT) != 0 && !is_quantifier)
3507 {
3508 previous_callout = code;
3509 code = auto_callout(code, ptr, cd);
3510 }
3511
3512 switch(c)
3513 {
3514 /* ===================================================================*/
3515 case 0: /* The branch terminates at string end */
3516 case CHAR_VERTICAL_LINE: /* or | or ) */
3517 case CHAR_RIGHT_PARENTHESIS:
3518 *firstbyteptr = firstbyte;
3519 *reqbyteptr = reqbyte;
3520 *codeptr = code;
3521 *ptrptr = ptr;
3522 if (lengthptr != NULL)
3523 {
3524 if (OFLOW_MAX - *lengthptr < code - last_code)
3525 {
3526 *errorcodeptr = ERR20;
3527 goto FAILED;
3528 }
3529 *lengthptr += (int)(code - last_code); /* To include callout length */
3530 DPRINTF((">> end branch\n"));
3531 }
3532 return TRUE;
3533
3534
3535 /* ===================================================================*/
3536 /* Handle single-character metacharacters. In multiline mode, ^ disables
3537 the setting of any following char as a first character. */
3538
3539 case CHAR_CIRCUMFLEX_ACCENT:
3540 previous = NULL;
3541 if ((options & PCRE_MULTILINE) != 0)
3542 {
3543 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
3544 *code++ = OP_CIRCM;
3545 }
3546 else *code++ = OP_CIRC;
3547 break;
3548
3549 case CHAR_DOLLAR_SIGN:
3550 previous = NULL;
3551 *code++ = ((options & PCRE_MULTILINE) != 0)? OP_DOLLM : OP_DOLL;
3552 break;
3553
3554 /* There can never be a first char if '.' is first, whatever happens about
3555 repeats. The value of reqbyte doesn't change either. */
3556
3557 case CHAR_DOT:
3558 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
3559 zerofirstbyte = firstbyte;
3560 zeroreqbyte = reqbyte;
3561 previous = code;
3562 *code++ = ((options & PCRE_DOTALL) != 0)? OP_ALLANY: OP_ANY;
3563 break;
3564
3565
3566 /* ===================================================================*/
3567 /* Character classes. If the included characters are all < 256, we build a
3568 32-byte bitmap of the permitted characters, except in the special case
3569 where there is only one such character. For negated classes, we build the
3570 map as usual, then invert it at the end. However, we use a different opcode
3571 so that data characters > 255 can be handled correctly.
3572
3573 If the class contains characters outside the 0-255 range, a different
3574 opcode is compiled. It may optionally have a bit map for characters < 256,
3575 but those above are are explicitly listed afterwards. A flag byte tells
3576 whether the bitmap is present, and whether this is a negated class or not.
3577
3578 In JavaScript compatibility mode, an isolated ']' causes an error. In
3579 default (Perl) mode, it is treated as a data character. */
3580
3581 case CHAR_RIGHT_SQUARE_BRACKET:
3582 if ((cd->external_options & PCRE_JAVASCRIPT_COMPAT) != 0)
3583 {
3584 *errorcodeptr = ERR64;
3585 goto FAILED;
3586 }
3587 goto NORMAL_CHAR;
3588
3589 case CHAR_LEFT_SQUARE_BRACKET:
3590 previous = code;
3591
3592 /* PCRE supports POSIX class stuff inside a class. Perl gives an error if
3593 they are encountered at the top level, so we'll do that too. */
3594
3595 if ((ptr[1] == CHAR_COLON || ptr[1] == CHAR_DOT ||
3596 ptr[1] == CHAR_EQUALS_SIGN) &&
3597 check_posix_syntax(ptr, &tempptr))
3598 {
3599 *errorcodeptr = (ptr[1] == CHAR_COLON)? ERR13 : ERR31;
3600 goto FAILED;
3601 }
3602
3603 /* If the first character is '^', set the negation flag and skip it. Also,
3604 if the first few characters (either before or after ^) are \Q\E or \E we
3605 skip them too. This makes for compatibility with Perl. */
3606
3607 negate_class = FALSE;
3608 for (;;)
3609 {
3610 c = *(++ptr);
3611 if (c == CHAR_BACKSLASH)
3612 {
3613 if (ptr[1] == CHAR_E)
3614 ptr++;
3615 else if (strncmp((const char *)ptr+1,
3616 STR_Q STR_BACKSLASH STR_E, 3) == 0)
3617 ptr += 3;
3618 else
3619 break;
3620 }
3621 else if (!negate_class && c == CHAR_CIRCUMFLEX_ACCENT)
3622 negate_class = TRUE;
3623 else break;
3624 }
3625
3626 /* Empty classes are allowed in JavaScript compatibility mode. Otherwise,
3627 an initial ']' is taken as a data character -- the code below handles
3628 that. In JS mode, [] must always fail, so generate OP_FAIL, whereas
3629 [^] must match any character, so generate OP_ALLANY. */
3630
3631 if (c == CHAR_RIGHT_SQUARE_BRACKET &&
3632 (cd->external_options & PCRE_JAVASCRIPT_COMPAT) != 0)
3633 {
3634 *code++ = negate_class? OP_ALLANY : OP_FAIL;
3635 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
3636 zerofirstbyte = firstbyte;
3637 break;
3638 }
3639
3640 /* If a class contains a negative special such as \S, we need to flip the
3641 negation flag at the end, so that support for characters > 255 works
3642 correctly (they are all included in the class). */
3643
3644 should_flip_negation = FALSE;
3645
3646 /* Keep a count of chars with values < 256 so that we can optimize the case
3647 of just a single character (as long as it's < 256). However, For higher
3648 valued UTF-8 characters, we don't yet do any optimization. */
3649
3650 class_charcount = 0;
3651 class_lastchar = -1;
3652
3653 /* Initialize the 32-char bit map to all zeros. We build the map in a
3654 temporary bit of memory, in case the class contains only 1 character (less
3655 than 256), because in that case the compiled code doesn't use the bit map.
3656 */
3657
3658 memset(classbits, 0, 32 * sizeof(uschar));
3659
3660 #ifdef SUPPORT_UTF8
3661 class_utf8 = FALSE; /* No chars >= 256 */
3662 class_utf8data = code + LINK_SIZE + 2; /* For UTF-8 items */
3663 class_utf8data_base = class_utf8data; /* For resetting in pass 1 */
3664 #endif
3665
3666 /* Process characters until ] is reached. By writing this as a "do" it
3667 means that an initial ] is taken as a data character. At the start of the
3668 loop, c contains the first byte of the character. */
3669
3670 if (c != 0) do
3671 {
3672 const uschar *oldptr;
3673
3674 #ifdef SUPPORT_UTF8
3675 if (utf8 && c > 127)
3676 { /* Braces are required because the */
3677 GETCHARLEN(c, ptr, ptr); /* macro generates multiple statements */
3678 }
3679
3680 /* In the pre-compile phase, accumulate the length of any UTF-8 extra
3681 data and reset the pointer. This is so that very large classes that
3682 contain a zillion UTF-8 characters no longer overwrite the work space
3683 (which is on the stack). */
3684
3685 if (lengthptr != NULL)
3686 {
3687 *lengthptr += (int)(class_utf8data - class_utf8data_base);
3688 class_utf8data = class_utf8data_base;
3689 }
3690
3691 #endif
3692
3693 /* Inside \Q...\E everything is literal except \E */
3694
3695 if (inescq)
3696 {
3697 if (c == CHAR_BACKSLASH && ptr[1] == CHAR_E) /* If we are at \E */
3698 {
3699 inescq = FALSE; /* Reset literal state */
3700 ptr++; /* Skip the 'E' */
3701 continue; /* Carry on with next */
3702 }
3703 goto CHECK_RANGE; /* Could be range if \E follows */
3704 }
3705
3706 /* Handle POSIX class names. Perl allows a negation extension of the
3707 form [:^name:]. A square bracket that doesn't match the syntax is
3708 treated as a literal. We also recognize the POSIX constructions
3709 [.ch.] and [=ch=] ("collating elements") and fault them, as Perl
3710 5.6 and 5.8 do. */
3711
3712 if (c == CHAR_LEFT_SQUARE_BRACKET &&
3713 (ptr[1] == CHAR_COLON || ptr[1] == CHAR_DOT ||
3714 ptr[1] == CHAR_EQUALS_SIGN) && check_posix_syntax(ptr, &tempptr))
3715 {
3716 BOOL local_negate = FALSE;
3717 int posix_class, taboffset, tabopt;
3718 register const uschar *cbits = cd->cbits;
3719 uschar pbits[32];
3720
3721 if (ptr[1] != CHAR_COLON)
3722 {
3723 *errorcodeptr = ERR31;
3724 goto FAILED;
3725 }
3726
3727 ptr += 2;
3728 if (*ptr == CHAR_CIRCUMFLEX_ACCENT)
3729 {
3730 local_negate = TRUE;
3731 should_flip_negation = TRUE; /* Note negative special */
3732 ptr++;
3733 }
3734
3735 posix_class = check_posix_name(ptr, (int)(tempptr - ptr));
3736 if (posix_class < 0)
3737 {
3738 *errorcodeptr = ERR30;
3739 goto FAILED;
3740 }
3741
3742 /* If matching is caseless, upper and lower are converted to
3743 alpha. This relies on the fact that the class table starts with
3744 alpha, lower, upper as the first 3 entries. */
3745
3746 if ((options & PCRE_CASELESS) != 0 && posix_class <= 2)
3747 posix_class = 0;
3748
3749 /* When PCRE_UCP is set, some of the POSIX classes are converted to
3750 different escape sequences that use Unicode properties. */
3751
3752 #ifdef SUPPORT_UCP
3753 if ((options & PCRE_UCP) != 0)
3754 {
3755 int pc = posix_class + ((local_negate)? POSIX_SUBSIZE/2 : 0);
3756 if (posix_substitutes[pc] != NULL)
3757 {
3758 nestptr = tempptr + 1;
3759 ptr = posix_substitutes[pc] - 1;
3760 continue;
3761 }
3762 }
3763 #endif
3764 /* In the non-UCP case, we build the bit map for the POSIX class in a
3765 chunk of local store because we may be adding and subtracting from it,
3766 and we don't want to subtract bits that may be in the main map already.
3767 At the end we or the result into the bit map that is being built. */
3768
3769 posix_class *= 3;
3770
3771 /* Copy in the first table (always present) */
3772
3773 memcpy(pbits, cbits + posix_class_maps[posix_class],
3774 32 * sizeof(uschar));
3775
3776 /* If there is a second table, add or remove it as required. */
3777
3778 taboffset = posix_class_maps[posix_class + 1];
3779 tabopt = posix_class_maps[posix_class + 2];
3780
3781 if (taboffset >= 0)
3782 {
3783 if (tabopt >= 0)
3784 for (c = 0; c < 32; c++) pbits[c] |= cbits[c + taboffset];
3785 else
3786 for (c = 0; c < 32; c++) pbits[c] &= ~cbits[c + taboffset];
3787 }
3788
3789 /* Not see if we need to remove any special characters. An option
3790 value of 1 removes vertical space and 2 removes underscore. */
3791
3792 if (tabopt < 0) tabopt = -tabopt;
3793 if (tabopt == 1) pbits[1] &= ~0x3c;
3794 else if (tabopt == 2) pbits[11] &= 0x7f;
3795
3796 /* Add the POSIX table or its complement into the main table that is
3797 being built and we are done. */
3798
3799 if (local_negate)
3800 for (c = 0; c < 32; c++) classbits[c] |= ~pbits[c];
3801 else
3802 for (c = 0; c < 32; c++) classbits[c] |= pbits[c];
3803
3804 ptr = tempptr + 1;
3805 class_charcount = 10; /* Set > 1; assumes more than 1 per class */
3806 continue; /* End of POSIX syntax handling */
3807 }
3808
3809 /* Backslash may introduce a single character, or it may introduce one
3810 of the specials, which just set a flag. The sequence \b is a special
3811 case. Inside a class (and only there) it is treated as backspace. We
3812 assume that other escapes have more than one character in them, so set
3813 class_charcount bigger than one. Unrecognized escapes fall through and
3814 are either treated as literal characters (by default), or are faulted if
3815 PCRE_EXTRA is set. */
3816
3817 if (c == CHAR_BACKSLASH)
3818 {
3819 c = check_escape(&ptr, errorcodeptr, cd->bracount, options, TRUE);
3820 if (*errorcodeptr != 0) goto FAILED;
3821
3822 if (-c == ESC_b) c = CHAR_BS; /* \b is backspace in a class */
3823 else if (-c == ESC_N) /* \N is not supported in a class */
3824 {
3825 *errorcodeptr = ERR71;
3826 goto FAILED;
3827 }
3828 else if (-c == ESC_Q) /* Handle start of quoted string */
3829 {
3830 if (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
3831 {
3832 ptr += 2; /* avoid empty string */
3833 }
3834 else inescq = TRUE;
3835 continue;
3836 }
3837 else if (-c == ESC_E) continue; /* Ignore orphan \E */
3838
3839 if (c < 0)
3840 {
3841 register const uschar *cbits = cd->cbits;
3842 class_charcount += 2; /* Greater than 1 is what matters */
3843
3844 switch (-c)
3845 {
3846 #ifdef SUPPORT_UCP
3847 case ESC_du: /* These are the values given for \d etc */
3848 case ESC_DU: /* when PCRE_UCP is set. We replace the */
3849 case ESC_wu: /* escape sequence with an appropriate \p */
3850 case ESC_WU: /* or \P to test Unicode properties instead */
3851 case ESC_su: /* of the default ASCII testing. */
3852 case ESC_SU:
3853 nestptr = ptr;
3854 ptr = substitutes[-c - ESC_DU] - 1; /* Just before substitute */
3855 class_charcount -= 2; /* Undo! */
3856 continue;
3857 #endif
3858 case ESC_d:
3859 for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_digit];
3860 continue;
3861
3862 case ESC_D:
3863 should_flip_negation = TRUE;
3864 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_digit];
3865 continue;
3866
3867 case ESC_w:
3868 for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_word];
3869 continue;
3870
3871 case ESC_W:
3872 should_flip_negation = TRUE;
3873 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_word];
3874 continue;
3875
3876 /* Perl 5.004 onwards omits VT from \s, but we must preserve it
3877 if it was previously set by something earlier in the character
3878 class. */
3879
3880 case ESC_s:
3881 classbits[0] |= cbits[cbit_space];
3882 classbits[1] |= cbits[cbit_space+1] & ~0x08;
3883 for (c = 2; c < 32; c++) classbits[c] |= cbits[c+cbit_space];
3884 continue;
3885
3886 case ESC_S:
3887 should_flip_negation = TRUE;
3888 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_space];
3889 classbits[1] |= 0x08; /* Perl 5.004 onwards omits VT from \s */
3890 continue;
3891
3892 case ESC_h:
3893 SETBIT(classbits, 0x09); /* VT */
3894 SETBIT(classbits, 0x20); /* SPACE */
3895 SETBIT(classbits, 0xa0); /* NSBP */
3896 #ifdef SUPPORT_UTF8
3897 if (utf8)
3898 {
3899 class_utf8 = TRUE;
3900 *class_utf8data++ = XCL_SINGLE;
3901 class_utf8data += _pcre_ord2utf8(0x1680, class_utf8data);
3902 *class_utf8data++ = XCL_SINGLE;
3903 class_utf8data += _pcre_ord2utf8(0x180e, class_utf8data);
3904 *class_utf8data++ = XCL_RANGE;
3905 class_utf8data += _pcre_ord2utf8(0x2000, class_utf8data);
3906 class_utf8data += _pcre_ord2utf8(0x200A, class_utf8data);
3907 *class_utf8data++ = XCL_SINGLE;
3908 class_utf8data += _pcre_ord2utf8(0x202f, class_utf8data);
3909 *class_utf8data++ = XCL_SINGLE;
3910 class_utf8data += _pcre_ord2utf8(0x205f, class_utf8data);
3911 *class_utf8data++ = XCL_SINGLE;
3912 class_utf8data += _pcre_ord2utf8(0x3000, class_utf8data);
3913 }
3914 #endif
3915 continue;
3916
3917 case ESC_H:
3918 for (c = 0; c < 32; c++)
3919 {
3920 int x = 0xff;
3921 switch (c)
3922 {
3923 case 0x09/8: x ^= 1 << (0x09%8); break;
3924 case 0x20/8: x ^= 1 << (0x20%8); break;
3925 case 0xa0/8: x ^= 1 << (0xa0%8); break;
3926 default: break;
3927 }
3928 classbits[c] |= x;
3929 }
3930
3931 #ifdef SUPPORT_UTF8
3932 if (utf8)
3933 {
3934 class_utf8 = TRUE;
3935 *class_utf8data++ = XCL_RANGE;
3936 class_utf8data += _pcre_ord2utf8(0x0100, class_utf8data);
3937 class_utf8data += _pcre_ord2utf8(0x167f, class_utf8data);
3938 *class_utf8data++ = XCL_RANGE;
3939 class_utf8data += _pcre_ord2utf8(0x1681, class_utf8data);
3940 class_utf8data += _pcre_ord2utf8(0x180d, class_utf8data);
3941 *class_utf8data++ = XCL_RANGE;
3942 class_utf8data += _pcre_ord2utf8(0x180f, class_utf8data);
3943 class_utf8data += _pcre_ord2utf8(0x1fff, class_utf8data);
3944 *class_utf8data++ = XCL_RANGE;
3945 class_utf8data += _pcre_ord2utf8(0x200B, class_utf8data);
3946 class_utf8data += _pcre_ord2utf8(0x202e, class_utf8data);
3947 *class_utf8data++ = XCL_RANGE;
3948 class_utf8data += _pcre_ord2utf8(0x2030, class_utf8data);
3949 class_utf8data += _pcre_ord2utf8(0x205e, class_utf8data);
3950 *class_utf8data++ = XCL_RANGE;
3951 class_utf8data += _pcre_ord2utf8(0x2060, class_utf8data);
3952 class_utf8data += _pcre_ord2utf8(0x2fff, class_utf8data);
3953 *class_utf8data++ = XCL_RANGE;
3954 class_utf8data += _pcre_ord2utf8(0x3001, class_utf8data);
3955 class_utf8data += _pcre_ord2utf8(0x7fffffff, class_utf8data);
3956 }
3957 #endif
3958 continue;
3959
3960 case ESC_v:
3961 SETBIT(classbits, 0x0a); /* LF */
3962 SETBIT(classbits, 0x0b); /* VT */
3963 SETBIT(classbits, 0x0c); /* FF */
3964 SETBIT(classbits, 0x0d); /* CR */
3965 SETBIT(classbits, 0x85); /* NEL */
3966 #ifdef SUPPORT_UTF8
3967 if (utf8)
3968 {
3969 class_utf8 = TRUE;
3970 *class_utf8data++ = XCL_RANGE;
3971 class_utf8data += _pcre_ord2utf8(0x2028, class_utf8data);
3972 class_utf8data += _pcre_ord2utf8(0x2029, class_utf8data);
3973 }
3974 #endif
3975 continue;
3976
3977 case ESC_V:
3978 for (c = 0; c < 32; c++)
3979 {
3980 int x = 0xff;
3981 switch (c)
3982 {
3983 case 0x0a/8: x ^= 1 << (0x0a%8);
3984 x ^= 1 << (0x0b%8);
3985 x ^= 1 << (0x0c%8);
3986 x ^= 1 << (0x0d%8);
3987 break;
3988 case 0x85/8: x ^= 1 << (0x85%8); break;
3989 default: break;
3990 }
3991 classbits[c] |= x;
3992 }
3993
3994 #ifdef SUPPORT_UTF8
3995 if (utf8)
3996 {
3997 class_utf8 = TRUE;
3998 *class_utf8data++ = XCL_RANGE;
3999 class_utf8data += _pcre_ord2utf8(0x0100, class_utf8data);
4000 class_utf8data += _pcre_ord2utf8(0x2027, class_utf8data);
4001 *class_utf8data++ = XCL_RANGE;
4002 class_utf8data += _pcre_ord2utf8(0x2029, class_utf8data);
4003 class_utf8data += _pcre_ord2utf8(0x7fffffff, class_utf8data);
4004 }
4005 #endif
4006 continue;
4007
4008 #ifdef SUPPORT_UCP
4009 case ESC_p:
4010 case ESC_P:
4011 {
4012 BOOL negated;
4013 int pdata;
4014 int ptype = get_ucp(&ptr, &negated, &pdata, errorcodeptr);
4015 if (ptype < 0) goto FAILED;
4016 class_utf8 = TRUE;
4017 *class_utf8data++ = ((-c == ESC_p) != negated)?
4018 XCL_PROP : XCL_NOTPROP;
4019 *class_utf8data++ = ptype;
4020 *class_utf8data++ = pdata;
4021 class_charcount -= 2; /* Not a < 256 character */
4022 continue;
4023 }
4024 #endif
4025 /* Unrecognized escapes are faulted if PCRE is running in its
4026 strict mode. By default, for compatibility with Perl, they are
4027 treated as literals. */
4028
4029 default:
4030 if ((options & PCRE_EXTRA) != 0)
4031 {
4032 *errorcodeptr = ERR7;
4033 goto FAILED;
4034 }
4035 class_charcount -= 2; /* Undo the default count from above */
4036 c = *ptr; /* Get the final character and fall through */
4037 break;
4038 }
4039 }
4040
4041 /* Fall through if we have a single character (c >= 0). This may be
4042 greater than 256 in UTF-8 mode. */
4043
4044 } /* End of backslash handling */
4045
4046 /* A single character may be followed by '-' to form a range. However,
4047 Perl does not permit ']' to be the end of the range. A '-' character
4048 at the end is treated as a literal. Perl ignores orphaned \E sequences
4049 entirely. The code for handling \Q and \E is messy. */
4050
4051 CHECK_RANGE:
4052 while (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
4053 {
4054 inescq = FALSE;
4055 ptr += 2;
4056 }
4057
4058 oldptr = ptr;
4059
4060 /* Remember \r or \n */
4061
4062 if (c == CHAR_CR || c == CHAR_NL) cd->external_flags |= PCRE_HASCRORLF;
4063
4064 /* Check for range */
4065
4066 if (!inescq && ptr[1] == CHAR_MINUS)
4067 {
4068 int d;
4069 ptr += 2;
4070 while (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_E) ptr += 2;
4071
4072 /* If we hit \Q (not followed by \E) at this point, go into escaped
4073 mode. */
4074
4075 while (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_Q)
4076 {
4077 ptr += 2;
4078 if (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_E)
4079 { ptr += 2; continue; }
4080 inescq = TRUE;
4081 break;
4082 }
4083
4084 if (*ptr == 0 || (!inescq && *ptr == CHAR_RIGHT_SQUARE_BRACKET))
4085 {
4086 ptr = oldptr;
4087 goto LONE_SINGLE_CHARACTER;
4088 }
4089
4090 #ifdef SUPPORT_UTF8
4091 if (utf8)
4092 { /* Braces are required because the */
4093 GETCHARLEN(d, ptr, ptr); /* macro generates multiple statements */
4094 }
4095 else
4096 #endif
4097 d = *ptr; /* Not UTF-8 mode */
4098
4099 /* The second part of a range can be a single-character escape, but
4100 not any of the other escapes. Perl 5.6 treats a hyphen as a literal
4101 in such circumstances. */
4102
4103 if (!inescq && d == CHAR_BACKSLASH)
4104 {
4105 d = check_escape(&ptr, errorcodeptr, cd->bracount, options, TRUE);
4106 if (*errorcodeptr != 0) goto FAILED;
4107
4108 /* \b is backspace; any other special means the '-' was literal */
4109
4110 if (d < 0)
4111 {
4112 if (d == -ESC_b) d = CHAR_BS; else
4113 {
4114 ptr = oldptr;
4115 goto LONE_SINGLE_CHARACTER; /* A few lines below */
4116 }
4117 }
4118 }
4119
4120 /* Check that the two values are in the correct order. Optimize
4121 one-character ranges */
4122
4123 if (d < c)
4124 {
4125 *errorcodeptr = ERR8;
4126 goto FAILED;
4127 }
4128
4129 if (d == c) goto LONE_SINGLE_CHARACTER; /* A few lines below */
4130
4131 /* Remember \r or \n */
4132
4133 if (d == CHAR_CR || d == CHAR_NL) cd->external_flags |= PCRE_HASCRORLF;
4134
4135 /* In UTF-8 mode, if the upper limit is > 255, or > 127 for caseless
4136 matching, we have to use an XCLASS with extra data items. Caseless
4137 matching for characters > 127 is available only if UCP support is
4138 available. */
4139
4140 #ifdef SUPPORT_UTF8
4141 if (utf8 && (d > 255 || ((options & PCRE_CASELESS) != 0 && d > 127)))
4142 {
4143 class_utf8 = TRUE;
4144
4145 /* With UCP support, we can find the other case equivalents of
4146 the relevant characters. There may be several ranges. Optimize how
4147 they fit with the basic range. */
4148
4149 #ifdef SUPPORT_UCP
4150 if ((options & PCRE_CASELESS) != 0)
4151 {
4152 unsigned int occ, ocd;
4153 unsigned int cc = c;
4154 unsigned int origd = d;
4155 while (get_othercase_range(&cc, origd, &occ, &ocd))
4156 {
4157 if (occ >= (unsigned int)c &&
4158 ocd <= (unsigned int)d)
4159 continue; /* Skip embedded ranges */
4160
4161 if (occ < (unsigned int)c &&
4162 ocd >= (unsigned int)c - 1) /* Extend the basic range */
4163 { /* if there is overlap, */
4164 c = occ; /* noting that if occ < c */
4165 continue; /* we can't have ocd > d */
4166 } /* because a subrange is */
4167 if (ocd > (unsigned int)d &&
4168 occ <= (unsigned int)d + 1) /* always shorter than */
4169 { /* the basic range. */
4170 d = ocd;
4171 continue;
4172 }
4173
4174 if (occ == ocd)
4175 {
4176 *class_utf8data++ = XCL_SINGLE;
4177 }
4178 else
4179 {
4180 *class_utf8data++ = XCL_RANGE;
4181 class_utf8data += _pcre_ord2utf8(occ, class_utf8data);
4182 }
4183 class_utf8data += _pcre_ord2utf8(ocd, class_utf8data);
4184 }
4185 }
4186 #endif /* SUPPORT_UCP */
4187
4188 /* Now record the original range, possibly modified for UCP caseless
4189 overlapping ranges. */
4190
4191 *class_utf8data++ = XCL_RANGE;
4192 class_utf8data += _pcre_ord2utf8(c, class_utf8data);
4193 class_utf8data += _pcre_ord2utf8(d, class_utf8data);
4194
4195 /* With UCP support, we are done. Without UCP support, there is no
4196 caseless matching for UTF-8 characters > 127; we can use the bit map
4197 for the smaller ones. */
4198
4199 #ifdef SUPPORT_UCP
4200 continue; /* With next character in the class */
4201 #else
4202 if ((options & PCRE_CASELESS) == 0 || c > 127) continue;
4203
4204 /* Adjust upper limit and fall through to set up the map */
4205
4206 d = 127;
4207
4208 #endif /* SUPPORT_UCP */
4209 }
4210 #endif /* SUPPORT_UTF8 */
4211
4212 /* We use the bit map for all cases when not in UTF-8 mode; else
4213 ranges that lie entirely within 0-127 when there is UCP support; else
4214 for partial ranges without UCP support. */
4215
4216 class_charcount += d - c + 1;
4217 class_lastchar = d;
4218
4219 /* We can save a bit of time by skipping this in the pre-compile. */
4220
4221 if (lengthptr == NULL) for (; c <= d; c++)
4222 {
4223 classbits[c/8] |= (1 << (c&7));
4224 if ((options & PCRE_CASELESS) != 0)
4225 {
4226 int uc = cd->fcc[c]; /* flip case */
4227 classbits[uc/8] |= (1 << (uc&7));
4228 }
4229 }
4230
4231 continue; /* Go get the next char in the class */
4232 }
4233
4234 /* Handle a lone single character - we can get here for a normal
4235 non-escape char, or after \ that introduces a single character or for an
4236 apparent range that isn't. */
4237
4238 LONE_SINGLE_CHARACTER:
4239
4240 /* Handle a character that cannot go in the bit map */
4241
4242 #ifdef SUPPORT_UTF8
4243 if (utf8 && (c > 255 || ((options & PCRE_CASELESS) != 0 && c > 127)))
4244 {
4245 class_utf8 = TRUE;
4246 *class_utf8data++ = XCL_SINGLE;
4247 class_utf8data += _pcre_ord2utf8(c, class_utf8data);
4248
4249 #ifdef SUPPORT_UCP
4250 if ((options & PCRE_CASELESS) != 0)
4251 {
4252 unsigned int othercase;
4253 if ((othercase = UCD_OTHERCASE(c)) != c)
4254 {
4255 *class_utf8data++ = XCL_SINGLE;
4256 class_utf8data += _pcre_ord2utf8(othercase, class_utf8data);
4257 }
4258 }
4259 #endif /* SUPPORT_UCP */
4260
4261 }
4262 else
4263 #endif /* SUPPORT_UTF8 */
4264
4265 /* Handle a single-byte character */
4266 {
4267 classbits[c/8] |= (1 << (c&7));
4268 if ((options & PCRE_CASELESS) != 0)
4269 {
4270 c = cd->fcc[c]; /* flip case */
4271 classbits[c/8] |= (1 << (c&7));
4272 }
4273 class_charcount++;
4274 class_lastchar = c;
4275 }
4276 }
4277
4278 /* Loop until ']' reached. This "while" is the end of the "do" far above.
4279 If we are at the end of an internal nested string, revert to the outer
4280 string. */
4281
4282 while (((c = *(++ptr)) != 0 ||
4283 (nestptr != NULL &&
4284 (ptr = nestptr, nestptr = NULL, c = *(++ptr)) != 0)) &&
4285 (c != CHAR_RIGHT_SQUARE_BRACKET || inescq));
4286
4287 /* Check for missing terminating ']' */
4288
4289 if (c == 0)
4290 {
4291 *errorcodeptr = ERR6;
4292 goto FAILED;
4293 }
4294
4295 /* If class_charcount is 1, we saw precisely one character whose value is
4296 less than 256. As long as there were no characters >= 128 and there was no
4297 use of \p or \P, in other words, no use of any XCLASS features, we can
4298 optimize.
4299
4300 In UTF-8 mode, we can optimize the negative case only if there were no
4301 characters >= 128 because OP_NOT and the related opcodes like OP_NOTSTAR
4302 operate on single-bytes characters only. This is an historical hangover.
4303 Maybe one day we can tidy these opcodes to handle multi-byte characters.
4304
4305 The optimization throws away the bit map. We turn the item into a
4306 1-character OP_CHAR[I] if it's positive, or OP_NOT[I] if it's negative.
4307 Note that OP_NOT[I] does not support multibyte characters. In the positive
4308 case, it can cause firstbyte to be set. Otherwise, there can be no first
4309 char if this item is first, whatever repeat count may follow. In the case
4310 of reqbyte, save the previous value for reinstating. */
4311
4312 #ifdef SUPPORT_UTF8
4313 if (class_charcount == 1 && !class_utf8 &&
4314 (!utf8 || !negate_class || class_lastchar < 128))
4315 #else
4316 if (class_charcount == 1)
4317 #endif
4318 {
4319 zeroreqbyte = reqbyte;
4320
4321 /* The OP_NOT[I] opcodes work on one-byte characters only. */
4322
4323 if (negate_class)
4324 {
4325 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
4326 zerofirstbyte = firstbyte;
4327 *code++ = ((options & PCRE_CASELESS) != 0)? OP_NOTI: OP_NOT;
4328 *code++ = class_lastchar;
4329 break;
4330 }
4331
4332 /* For a single, positive character, get the value into mcbuffer, and
4333 then we can handle this with the normal one-character code. */
4334
4335 #ifdef SUPPORT_UTF8
4336 if (utf8 && class_lastchar > 127)
4337 mclength = _pcre_ord2utf8(class_lastchar, mcbuffer);
4338 else
4339 #endif
4340 {
4341 mcbuffer[0] = class_lastchar;
4342 mclength = 1;
4343 }
4344 goto ONE_CHAR;
4345 } /* End of 1-char optimization */
4346
4347 /* The general case - not the one-char optimization. If this is the first
4348 thing in the branch, there can be no first char setting, whatever the
4349 repeat count. Any reqbyte setting must remain unchanged after any kind of
4350 repeat. */
4351
4352 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
4353 zerofirstbyte = firstbyte;
4354 zeroreqbyte = reqbyte;
4355
4356 /* If there are characters with values > 255, we have to compile an
4357 extended class, with its own opcode, unless there was a negated special
4358 such as \S in the class, and PCRE_UCP is not set, because in that case all
4359 characters > 255 are in the class, so any that were explicitly given as
4360 well can be ignored. If (when there are explicit characters > 255 that must
4361 be listed) there are no characters < 256, we can omit the bitmap in the
4362 actual compiled code. */
4363
4364 #ifdef SUPPORT_UTF8
4365 if (class_utf8 && (!should_flip_negation || (options & PCRE_UCP) != 0))
4366 {
4367 *class_utf8data++ = XCL_END; /* Marks the end of extra data */
4368 *code++ = OP_XCLASS;
4369 code += LINK_SIZE;
4370 *code = negate_class? XCL_NOT : 0;
4371
4372 /* If the map is required, move up the extra data to make room for it;
4373 otherwise just move the code pointer to the end of the extra data. */
4374
4375 if (class_charcount > 0)
4376 {
4377 *code++ |= XCL_MAP;
4378 memmove(code + 32, code, class_utf8data - code);
4379 memcpy(code, classbits, 32);
4380 code = class_utf8data + 32;
4381 }
4382 else code = class_utf8data;
4383
4384 /* Now fill in the complete length of the item */
4385
4386 PUT(previous, 1, (int)(code - previous));
4387 break; /* End of class handling */
4388 }
4389 #endif
4390
4391 /* If there are no characters > 255, or they are all to be included or
4392 excluded, set the opcode to OP_CLASS or OP_NCLASS, depending on whether the
4393 whole class was negated and whether there were negative specials such as \S
4394 (non-UCP) in the class. Then copy the 32-byte map into the code vector,
4395 negating it if necessary. */
4396
4397 *code++ = (negate_class == should_flip_negation) ? OP_CLASS : OP_NCLASS;
4398 if (negate_class)
4399 {
4400 if (lengthptr == NULL) /* Save time in the pre-compile phase */
4401 for (c = 0; c < 32; c++) code[c] = ~classbits[c];
4402 }
4403 else
4404 {
4405 memcpy(code, classbits, 32);
4406 }
4407 code += 32;
4408 break;
4409
4410
4411 /* ===================================================================*/
4412 /* Various kinds of repeat; '{' is not necessarily a quantifier, but this
4413 has been tested above. */
4414
4415 case CHAR_LEFT_CURLY_BRACKET:
4416 if (!is_quantifier) goto NORMAL_CHAR;
4417 ptr = read_repeat_counts(ptr+1, &repeat_min, &repeat_max, errorcodeptr);
4418 if (*errorcodeptr != 0) goto FAILED;
4419 goto REPEAT;
4420
4421 case CHAR_ASTERISK:
4422 repeat_min = 0;
4423 repeat_max = -1;
4424 goto REPEAT;
4425
4426 case CHAR_PLUS:
4427 repeat_min = 1;
4428 repeat_max = -1;
4429 goto REPEAT;
4430
4431 case CHAR_QUESTION_MARK:
4432 repeat_min = 0;
4433 repeat_max = 1;
4434
4435 REPEAT:
4436 if (previous == NULL)
4437 {
4438 *errorcodeptr = ERR9;
4439 goto FAILED;
4440 }
4441
4442 if (repeat_min == 0)
4443 {
4444 firstbyte = zerofirstbyte; /* Adjust for zero repeat */
4445 reqbyte = zeroreqbyte; /* Ditto */
4446 }
4447
4448 /* Remember whether this is a variable length repeat */
4449
4450 reqvary = (repeat_min == repeat_max)? 0 : REQ_VARY;
4451
4452 op_type = 0; /* Default single-char op codes */
4453 possessive_quantifier = FALSE; /* Default not possessive quantifier */
4454
4455 /* Save start of previous item, in case we have to move it up in order to
4456 insert something before it. */
4457
4458 tempcode = previous;
4459
4460 /* If the next character is '+', we have a possessive quantifier. This
4461 implies greediness, whatever the setting of the PCRE_UNGREEDY option.
4462 If the next character is '?' this is a minimizing repeat, by default,
4463 but if PCRE_UNGREEDY is set, it works the other way round. We change the
4464 repeat type to the non-default. */
4465
4466 if (ptr[1] == CHAR_PLUS)
4467 {
4468 repeat_type = 0; /* Force greedy */
4469 possessive_quantifier = TRUE;
4470 ptr++;
4471 }
4472 else if (ptr[1] == CHAR_QUESTION_MARK)
4473 {
4474 repeat_type = greedy_non_default;
4475 ptr++;
4476 }
4477 else repeat_type = greedy_default;
4478
4479 /* If previous was a recursion call, wrap it in atomic brackets so that
4480 previous becomes the atomic group. All recursions were so wrapped in the
4481 past, but it no longer happens for non-repeated recursions. In fact, the
4482 repeated ones could be re-implemented independently so as not to need this,
4483 but for the moment we rely on the code for repeating groups. */
4484
4485 if (*previous == OP_RECURSE)
4486 {
4487 memmove(previous + 1 + LINK_SIZE, previous, 1 + LINK_SIZE);
4488 *previous = OP_ONCE;
4489 PUT(previous, 1, 2 + 2*LINK_SIZE);
4490 previous[2 + 2*LINK_SIZE] = OP_KET;
4491 PUT(previous, 3 + 2*LINK_SIZE, 2 + 2*LINK_SIZE);
4492 code += 2 + 2 * LINK_SIZE;
4493 length_prevgroup = 3 + 3*LINK_SIZE;
4494
4495 /* When actually compiling, we need to check whether this was a forward
4496 reference, and if so, adjust the offset. */
4497
4498 if (lengthptr == NULL && cd->hwm >= cd->start_workspace + LINK_SIZE)
4499 {
4500 int offset = GET(cd->hwm, -LINK_SIZE);
4501 if (offset == previous + 1 - cd->start_code)
4502 PUT(cd->hwm, -LINK_SIZE, offset + 1 + LINK_SIZE);
4503 }
4504 }
4505
4506 /* Now handle repetition for the different types of item. */
4507
4508 /* If previous was a character match, abolish the item and generate a
4509 repeat item instead. If a char item has a minumum of more than one, ensure
4510 that it is set in reqbyte - it might not be if a sequence such as x{3} is
4511 the first thing in a branch because the x will have gone into firstbyte
4512 instead. */
4513
4514 if (*previous == OP_CHAR || *previous == OP_CHARI)
4515 {
4516 op_type = (*previous == OP_CHAR)? 0 : OP_STARI - OP_STAR;
4517
4518 /* Deal with UTF-8 characters that take up more than one byte. It's
4519 easier to write this out separately than try to macrify it. Use c to
4520 hold the length of the character in bytes, plus 0x80 to flag that it's a
4521 length rather than a small character. */
4522
4523 #ifdef SUPPORT_UTF8
4524 if (utf8 && (code[-1] & 0x80) != 0)
4525 {
4526 uschar *lastchar = code - 1;
4527 while((*lastchar & 0xc0) == 0x80) lastchar--;
4528 c = (int)(code - lastchar); /* Length of UTF-8 character */
4529 memcpy(utf8_char, lastchar, c); /* Save the char */
4530 c |= 0x80; /* Flag c as a length */
4531 }
4532 else
4533 #endif
4534
4535 /* Handle the case of a single byte - either with no UTF8 support, or
4536 with UTF-8 disabled, or for a UTF-8 character < 128. */
4537
4538 {
4539 c = code[-1];
4540 if (repeat_min > 1) reqbyte = c | req_caseopt | cd->req_varyopt;
4541 }
4542
4543 /* If the repetition is unlimited, it pays to see if the next thing on
4544 the line is something that cannot possibly match this character. If so,
4545 automatically possessifying this item gains some performance in the case
4546 where the match fails. */
4547
4548 if (!possessive_quantifier &&
4549 repeat_max < 0 &&
4550 check_auto_possessive(previous, utf8, ptr + 1, options, cd))
4551 {
4552 repeat_type = 0; /* Force greedy */
4553 possessive_quantifier = TRUE;
4554 }
4555
4556 goto OUTPUT_SINGLE_REPEAT; /* Code shared with single character types */
4557 }
4558
4559 /* If previous was a single negated character ([^a] or similar), we use
4560 one of the special opcodes, replacing it. The code is shared with single-
4561 character repeats by setting opt_type to add a suitable offset into
4562 repeat_type. We can also test for auto-possessification. OP_NOT and OP_NOTI
4563 are currently used only for single-byte chars. */
4564
4565 else if (*previous == OP_NOT || *previous == OP_NOTI)
4566 {
4567 op_type = ((*previous == OP_NOT)? OP_NOTSTAR : OP_NOTSTARI) - OP_STAR;
4568 c = previous[1];
4569 if (!possessive_quantifier &&
4570 repeat_max < 0 &&
4571 check_auto_possessive(previous, utf8, ptr + 1, options, cd))
4572 {
4573 repeat_type = 0; /* Force greedy */
4574 possessive_quantifier = TRUE;
4575 }
4576 goto OUTPUT_SINGLE_REPEAT;
4577 }
4578
4579 /* If previous was a character type match (\d or similar), abolish it and
4580 create a suitable repeat item. The code is shared with single-character
4581 repeats by setting op_type to add a suitable offset into repeat_type. Note
4582 the the Unicode property types will be present only when SUPPORT_UCP is
4583 defined, but we don't wrap the little bits of code here because it just
4584 makes it horribly messy. */
4585
4586 else if (*previous < OP_EODN)
4587 {
4588 uschar *oldcode;
4589 int prop_type, prop_value;
4590 op_type = OP_TYPESTAR - OP_STAR; /* Use type opcodes */
4591 c = *previous;
4592
4593 if (!possessive_quantifier &&
4594 repeat_max < 0 &&
4595 check_auto_possessive(previous, utf8, ptr + 1, options, cd))
4596 {
4597 repeat_type = 0; /* Force greedy */
4598 possessive_quantifier = TRUE;
4599 }
4600
4601 OUTPUT_SINGLE_REPEAT:
4602 if (*previous == OP_PROP || *previous == OP_NOTPROP)
4603 {
4604 prop_type = previous[1];
4605 prop_value = previous[2];
4606 }
4607 else prop_type = prop_value = -1;
4608
4609 oldcode = code;
4610 code = previous; /* Usually overwrite previous item */
4611
4612 /* If the maximum is zero then the minimum must also be zero; Perl allows
4613 this case, so we do too - by simply omitting the item altogether. */
4614
4615 if (repeat_max == 0) goto END_REPEAT;
4616
4617 /*--------------------------------------------------------------------*/
4618 /* This code is obsolete from release 8.00; the restriction was finally
4619 removed: */
4620
4621 /* All real repeats make it impossible to handle partial matching (maybe
4622 one day we will be able to remove this restriction). */
4623
4624 /* if (repeat_max != 1) cd->external_flags |= PCRE_NOPARTIAL; */
4625 /*--------------------------------------------------------------------*/
4626
4627 /* Combine the op_type with the repeat_type */
4628
4629 repeat_type += op_type;
4630
4631 /* A minimum of zero is handled either as the special case * or ?, or as
4632 an UPTO, with the maximum given. */
4633
4634 if (repeat_min == 0)
4635 {
4636 if (repeat_max == -1) *code++ = OP_STAR + repeat_type;
4637 else if (repeat_max == 1) *code++ = OP_QUERY + repeat_type;
4638 else
4639 {
4640 *code++ = OP_UPTO + repeat_type;
4641 PUT2INC(code, 0, repeat_max);
4642 }
4643 }
4644
4645 /* A repeat minimum of 1 is optimized into some special cases. If the
4646 maximum is unlimited, we use OP_PLUS. Otherwise, the original item is
4647 left in place and, if the maximum is greater than 1, we use OP_UPTO with
4648 one less than the maximum. */
4649
4650 else if (repeat_min == 1)
4651 {
4652 if (repeat_max == -1)
4653 *code++ = OP_PLUS + repeat_type;
4654 else
4655 {
4656 code = oldcode; /* leave previous item in place */
4657 if (repeat_max == 1) goto END_REPEAT;
4658 *code++ = OP_UPTO + repeat_type;
4659 PUT2INC(code, 0, repeat_max - 1);
4660 }
4661 }
4662
4663 /* The case {n,n} is just an EXACT, while the general case {n,m} is
4664 handled as an EXACT followed by an UPTO. */
4665
4666 else
4667 {
4668 *code++ = OP_EXACT + op_type; /* NB EXACT doesn't have repeat_type */
4669 PUT2INC(code, 0, repeat_min);
4670
4671 /* If the maximum is unlimited, insert an OP_STAR. Before doing so,
4672 we have to insert the character for the previous code. For a repeated
4673 Unicode property match, there are two extra bytes that define the
4674 required property. In UTF-8 mode, long characters have their length in
4675 c, with the 0x80 bit as a flag. */
4676
4677 if (repeat_max < 0)
4678 {
4679 #ifdef SUPPORT_UTF8
4680 if (utf8 && c >= 128)
4681 {
4682 memcpy(code, utf8_char, c & 7);
4683 code += c & 7;
4684 }
4685 else
4686 #endif
4687 {
4688 *code++ = c;
4689 if (prop_type >= 0)
4690 {
4691 *code++ = prop_type;
4692 *code++ = prop_value;
4693 }
4694 }
4695 *code++ = OP_STAR + repeat_type;
4696 }
4697
4698 /* Else insert an UPTO if the max is greater than the min, again
4699 preceded by the character, for the previously inserted code. If the
4700 UPTO is just for 1 instance, we can use QUERY instead. */
4701
4702 else if (repeat_max != repeat_min)
4703 {
4704 #ifdef SUPPORT_UTF8
4705 if (utf8 && c >= 128)
4706 {
4707 memcpy(code, utf8_char, c & 7);
4708 code += c & 7;
4709 }
4710 else
4711 #endif
4712 *code++ = c;
4713 if (prop_type >= 0)
4714 {
4715 *code++ = prop_type;
4716 *code++ = prop_value;
4717 }
4718 repeat_max -= repeat_min;
4719
4720 if (repeat_max == 1)
4721 {
4722 *code++ = OP_QUERY + repeat_type;
4723 }
4724 else
4725 {
4726 *code++ = OP_UPTO + repeat_type;
4727 PUT2INC(code, 0, repeat_max);
4728 }
4729 }
4730 }
4731
4732 /* The character or character type itself comes last in all cases. */
4733
4734 #ifdef SUPPORT_UTF8
4735 if (utf8 && c >= 128)
4736 {
4737 memcpy(code, utf8_char, c & 7);
4738 code += c & 7;
4739 }
4740 else
4741 #endif
4742 *code++ = c;
4743
4744 /* For a repeated Unicode property match, there are two extra bytes that
4745 define the required property. */
4746
4747 #ifdef SUPPORT_UCP
4748 if (prop_type >= 0)
4749 {
4750 *code++ = prop_type;
4751 *code++ = prop_value;
4752 }
4753 #endif
4754 }
4755
4756 /* If previous was a character class or a back reference, we put the repeat
4757 stuff after it, but just skip the item if the repeat was {0,0}. */
4758
4759 else if (*previous == OP_CLASS ||
4760 *previous == OP_NCLASS ||
4761 #ifdef SUPPORT_UTF8
4762 *previous == OP_XCLASS ||
4763 #endif
4764 *previous == OP_REF ||
4765 *previous == OP_REFI)
4766 {
4767 if (repeat_max == 0)
4768 {
4769 code = previous;
4770 goto END_REPEAT;
4771 }
4772
4773 /*--------------------------------------------------------------------*/
4774 /* This code is obsolete from release 8.00; the restriction was finally
4775 removed: */
4776
4777 /* All real repeats make it impossible to handle partial matching (maybe
4778 one day we will be able to remove this restriction). */
4779
4780 /* if (repeat_max != 1) cd->external_flags |= PCRE_NOPARTIAL; */
4781 /*--------------------------------------------------------------------*/
4782
4783 if (repeat_min == 0 && repeat_max == -1)
4784 *code++ = OP_CRSTAR + repeat_type;
4785 else if (repeat_min == 1 && repeat_max == -1)
4786 *code++ = OP_CRPLUS + repeat_type;
4787 else if (repeat_min == 0 && repeat_max == 1)
4788 *code++ = OP_CRQUERY + repeat_type;
4789 else
4790 {
4791 *code++ = OP_CRRANGE + repeat_type;
4792 PUT2INC(code, 0, repeat_min);
4793 if (repeat_max == -1) repeat_max = 0; /* 2-byte encoding for max */
4794 PUT2INC(code, 0, repeat_max);
4795 }
4796 }
4797
4798 /* If previous was a bracket group, we may have to replicate it in certain
4799 cases. Note that at this point we can encounter only the "basic" bracket
4800 opcodes such as BRA and CBRA, as this is the place where they get converted
4801 into the more special varieties such as BRAPOS and SBRA. A test for >=
4802 OP_ASSERT and <= OP_COND includes ASSERT, ASSERT_NOT, ASSERTBACK,
4803 ASSERTBACK_NOT, ONCE, BRA, CBRA, and COND. Originally, PCRE did not allow
4804 repetition of assertions, but now it does, for Perl compatibility. */
4805
4806 else if (*previous >= OP_ASSERT && *previous <= OP_COND)
4807 {
4808 register int i;
4809 int len = (int)(code - previous);
4810 uschar *bralink = NULL;
4811 uschar *brazeroptr = NULL;
4812
4813 /* Repeating a DEFINE group is pointless, but Perl allows the syntax, so
4814 we just ignore the repeat. */
4815
4816 if (*previous == OP_COND && previous[LINK_SIZE+1] == OP_DEF)
4817 goto END_REPEAT;
4818
4819 /* There is no sense in actually repeating assertions. The only potential
4820 use of repetition is in cases when the assertion is optional. Therefore,
4821 if the minimum is greater than zero, just ignore the repeat. If the
4822 maximum is not not zero or one, set it to 1. */
4823
4824 if (*previous < OP_ONCE) /* Assertion */
4825 {
4826 if (repeat_min > 0) goto END_REPEAT;
4827 if (repeat_max < 0 || repeat_max > 1) repeat_max = 1;
4828 }
4829
4830 /* The case of a zero minimum is special because of the need to stick
4831 OP_BRAZERO in front of it, and because the group appears once in the
4832 data, whereas in other cases it appears the minimum number of times. For
4833 this reason, it is simplest to treat this case separately, as otherwise
4834 the code gets far too messy. There are several special subcases when the
4835 minimum is zero. */
4836
4837 if (repeat_min == 0)
4838 {
4839 /* If the maximum is also zero, we used to just omit the group from the
4840 output altogether, like this:
4841
4842 ** if (repeat_max == 0)
4843 ** {
4844 ** code = previous;
4845 ** goto END_REPEAT;
4846 ** }
4847
4848 However, that fails when a group or a subgroup within it is referenced
4849 as a subroutine from elsewhere in the pattern, so now we stick in
4850 OP_SKIPZERO in front of it so that it is skipped on execution. As we
4851 don't have a list of which groups are referenced, we cannot do this
4852 selectively.
4853
4854 If the maximum is 1 or unlimited, we just have to stick in the BRAZERO
4855 and do no more at this point. However, we do need to adjust any
4856 OP_RECURSE calls inside the group that refer to the group itself or any
4857 internal or forward referenced group, because the offset is from the
4858 start of the whole regex. Temporarily terminate the pattern while doing
4859 this. */
4860
4861 if (repeat_max <= 1) /* Covers 0, 1, and unlimited */
4862 {
4863 *code = OP_END;
4864 adjust_recurse(previous, 1, utf8, cd, save_hwm);
4865 memmove(previous+1, previous, len);
4866 code++;
4867 if (repeat_max == 0)
4868 {
4869 *previous++ = OP_SKIPZERO;
4870 goto END_REPEAT;
4871 }
4872 brazeroptr = previous; /* Save for possessive optimizing */
4873 *previous++ = OP_BRAZERO + repeat_type;
4874 }
4875
4876 /* If the maximum is greater than 1 and limited, we have to replicate
4877 in a nested fashion, sticking OP_BRAZERO before each set of brackets.
4878 The first one has to be handled carefully because it's the original
4879 copy, which has to be moved up. The remainder can be handled by code
4880 that is common with the non-zero minimum case below. We have to
4881 adjust the value or repeat_max, since one less copy is required. Once
4882 again, we may have to adjust any OP_RECURSE calls inside the group. */
4883
4884 else
4885 {
4886 int offset;
4887 *code = OP_END;
4888 adjust_recurse(previous, 2 + LINK_SIZE, utf8, cd, save_hwm);
4889 memmove(previous + 2 + LINK_SIZE, previous, len);
4890 code += 2 + LINK_SIZE;
4891 *previous++ = OP_BRAZERO + repeat_type;
4892 *previous++ = OP_BRA;
4893
4894 /* We chain together the bracket offset fields that have to be
4895 filled in later when the ends of the brackets are reached. */
4896
4897 offset = (bralink == NULL)? 0 : (int)(previous - bralink);
4898 bralink = previous;
4899 PUTINC(previous, 0, offset);
4900 }
4901
4902 repeat_max--;
4903 }
4904
4905 /* If the minimum is greater than zero, replicate the group as many
4906 times as necessary, and adjust the maximum to the number of subsequent
4907 copies that we need. If we set a first char from the group, and didn't
4908 set a required char, copy the latter from the former. If there are any
4909 forward reference subroutine calls in the group, there will be entries on
4910 the workspace list; replicate these with an appropriate increment. */
4911
4912 else
4913 {
4914 if (repeat_min > 1)
4915 {
4916 /* In the pre-compile phase, we don't actually do the replication. We
4917 just adjust the length as if we had. Do some paranoid checks for
4918 potential integer overflow. The INT64_OR_DOUBLE type is a 64-bit
4919 integer type when available, otherwise double. */
4920
4921 if (lengthptr != NULL)
4922 {
4923 int delta = (repeat_min - 1)*length_prevgroup;
4924 if ((INT64_OR_DOUBLE)(repeat_min - 1)*
4925 (INT64_OR_DOUBLE)length_prevgroup >
4926 (INT64_OR_DOUBLE)INT_MAX ||
4927 OFLOW_MAX - *lengthptr < delta)
4928 {
4929 *errorcodeptr = ERR20;
4930 goto FAILED;
4931 }
4932 *lengthptr += delta;
4933 }
4934
4935 /* This is compiling for real. If there is a set first byte for
4936 the group, and we have not yet set a "required byte", set it. Make
4937 sure there is enough workspace for copying forward references before
4938 doing the copy. */
4939
4940 else
4941 {
4942 if (groupsetfirstbyte && reqbyte < 0) reqbyte = firstbyte;
4943
4944 for (i = 1; i < repeat_min; i++)
4945 {
4946 uschar *hc;
4947 uschar *this_hwm = cd->hwm;
4948 memcpy(code, previous, len);
4949
4950 while (cd->hwm > cd->start_workspace + cd->workspace_size -
4951 WORK_SIZE_SAFETY_MARGIN - (this_hwm - save_hwm))
4952 {
4953 int save_offset = save_hwm - cd->start_workspace;
4954 int this_offset = this_hwm - cd->start_workspace;
4955 *errorcodeptr = expand_workspace(cd);
4956 if (*errorcodeptr != 0) goto FAILED;
4957 save_hwm = (uschar *)cd->start_workspace + save_offset;
4958 this_hwm = (uschar *)cd->start_workspace + this_offset;
4959 }
4960
4961 for (hc = save_hwm; hc < this_hwm; hc += LINK_SIZE)
4962 {
4963 PUT(cd->hwm, 0, GET(hc, 0) + len);
4964 cd->hwm += LINK_SIZE;
4965 }
4966 save_hwm = this_hwm;
4967 code += len;
4968 }
4969 }
4970 }
4971
4972 if (repeat_max > 0) repeat_max -= repeat_min;
4973 }
4974
4975 /* This code is common to both the zero and non-zero minimum cases. If
4976 the maximum is limited, it replicates the group in a nested fashion,
4977 remembering the bracket starts on a stack. In the case of a zero minimum,
4978 the first one was set up above. In all cases the repeat_max now specifies
4979 the number of additional copies needed. Again, we must remember to
4980 replicate entries on the forward reference list. */
4981
4982 if (repeat_max >= 0)
4983 {
4984 /* In the pre-compile phase, we don't actually do the replication. We
4985 just adjust the length as if we had. For each repetition we must add 1
4986 to the length for BRAZERO and for all but the last repetition we must
4987 add 2 + 2*LINKSIZE to allow for the nesting that occurs. Do some
4988 paranoid checks to avoid integer overflow. The INT64_OR_DOUBLE type is
4989 a 64-bit integer type when available, otherwise double. */
4990
4991 if (lengthptr != NULL && repeat_max > 0)
4992 {
4993 int delta = repeat_max * (length_prevgroup + 1 + 2 + 2*LINK_SIZE) -
4994 2 - 2*LINK_SIZE; /* Last one doesn't nest */
4995 if ((INT64_OR_DOUBLE)repeat_max *
4996 (INT64_OR_DOUBLE)(length_prevgroup + 1 + 2 + 2*LINK_SIZE)
4997 > (INT64_OR_DOUBLE)INT_MAX ||
4998 OFLOW_MAX - *lengthptr < delta)
4999 {
5000 *errorcodeptr = ERR20;
5001 goto FAILED;
5002 }
5003 *lengthptr += delta;
5004 }
5005
5006 /* This is compiling for real */
5007
5008 else for (i = repeat_max - 1; i >= 0; i--)
5009 {
5010 uschar *hc;
5011 uschar *this_hwm = cd->hwm;
5012
5013 *code++ = OP_BRAZERO + repeat_type;
5014
5015 /* All but the final copy start a new nesting, maintaining the
5016 chain of brackets outstanding. */
5017
5018 if (i != 0)
5019 {
5020 int offset;
5021 *code++ = OP_BRA;
5022 offset = (bralink == NULL)? 0 : (int)(code - bralink);
5023 bralink = code;
5024 PUTINC(code, 0, offset);
5025 }
5026
5027 memcpy(code, previous, len);
5028
5029 /* Ensure there is enough workspace for forward references before
5030 copying them. */
5031
5032 while (cd->hwm > cd->start_workspace + cd->workspace_size -
5033 WORK_SIZE_SAFETY_MARGIN - (this_hwm - save_hwm))
5034 {
5035 int save_offset = save_hwm - cd->start_workspace;
5036 int this_offset = this_hwm - cd->start_workspace;
5037 *errorcodeptr = expand_workspace(cd);
5038 if (*errorcodeptr != 0) goto FAILED;
5039 save_hwm = (uschar *)cd->start_workspace + save_offset;
5040 this_hwm = (uschar *)cd->start_workspace + this_offset;
5041 }
5042
5043 for (hc = save_hwm; hc < this_hwm; hc += LINK_SIZE)
5044 {
5045 PUT(cd->hwm, 0, GET(hc, 0) + len + ((i != 0)? 2+LINK_SIZE : 1));
5046 cd->hwm += LINK_SIZE;
5047 }
5048 save_hwm = this_hwm;
5049 code += len;
5050 }
5051
5052 /* Now chain through the pending brackets, and fill in their length
5053 fields (which are holding the chain links pro tem). */
5054
5055 while (bralink != NULL)
5056 {
5057 int oldlinkoffset;
5058 int offset = (int)(code - bralink + 1);
5059 uschar *bra = code - offset;
5060 oldlinkoffset = GET(bra, 1);
5061 bralink = (oldlinkoffset == 0)? NULL : bralink - oldlinkoffset;
5062 *code++ = OP_KET;
5063 PUTINC(code, 0, offset);
5064 PUT(bra, 1, offset);
5065 }
5066 }
5067
5068 /* If the maximum is unlimited, set a repeater in the final copy. For
5069 ONCE brackets, that's all we need to do. However, possessively repeated
5070 ONCE brackets can be converted into non-capturing brackets, as the
5071 behaviour of (?:xx)++ is the same as (?>xx)++ and this saves having to
5072 deal with possessive ONCEs specially.
5073
5074 Otherwise, when we are doing the actual compile phase, check to see
5075 whether this group is one that could match an empty string. If so,
5076 convert the initial operator to the S form (e.g. OP_BRA -> OP_SBRA) so
5077 that runtime checking can be done. [This check is also applied to ONCE
5078 groups at runtime, but in a different way.]
5079
5080 Then, if the quantifier was possessive and the bracket is not a
5081 conditional, we convert the BRA code to the POS form, and the KET code to
5082 KETRPOS. (It turns out to be convenient at runtime to detect this kind of
5083 subpattern at both the start and at the end.) The use of special opcodes
5084 makes it possible to reduce greatly the stack usage in pcre_exec(). If
5085 the group is preceded by OP_BRAZERO, convert this to OP_BRAPOSZERO.
5086
5087 Then, if the minimum number of matches is 1 or 0, cancel the possessive
5088 flag so that the default action below, of wrapping everything inside
5089 atomic brackets, does not happen. When the minimum is greater than 1,
5090 there will be earlier copies of the group, and so we still have to wrap
5091 the whole thing. */
5092
5093 else
5094 {
5095 uschar *ketcode = code - 1 - LINK_SIZE;
5096 uschar *bracode = ketcode - GET(ketcode, 1);
5097
5098 /* Convert possessive ONCE brackets to non-capturing */
5099
5100 if ((*bracode == OP_ONCE || *bracode == OP_ONCE_NC) &&
5101 possessive_quantifier) *bracode = OP_BRA;
5102
5103 /* For non-possessive ONCE brackets, all we need to do is to
5104 set the KET. */
5105
5106 if (*bracode == OP_ONCE || *bracode == OP_ONCE_NC)
5107 *ketcode = OP_KETRMAX + repeat_type;
5108
5109 /* Handle non-ONCE brackets and possessive ONCEs (which have been
5110 converted to non-capturing above). */
5111
5112 else
5113 {
5114 /* In the compile phase, check for empty string matching. */
5115
5116 if (lengthptr == NULL)
5117 {
5118 uschar *scode = bracode;
5119 do
5120 {
5121 if (could_be_empty_branch(scode, ketcode, utf8, cd))
5122 {
5123 *bracode += OP_SBRA - OP_BRA;
5124 break;
5125 }
5126 scode += GET(scode, 1);
5127 }
5128 while (*scode == OP_ALT);
5129 }
5130
5131 /* Handle possessive quantifiers. */
5132
5133 if (possessive_quantifier)
5134 {
5135 /* For COND brackets, we wrap the whole thing in a possessively
5136 repeated non-capturing bracket, because we have not invented POS
5137 versions of the COND opcodes. Because we are moving code along, we
5138 must ensure that any pending recursive references are updated. */
5139
5140 if (*bracode == OP_COND || *bracode == OP_SCOND)
5141 {
5142 int nlen = (int)(code - bracode);
5143 *code = OP_END;
5144 adjust_recurse(bracode, 1 + LINK_SIZE, utf8, cd, save_hwm);
5145 memmove(bracode + 1+LINK_SIZE, bracode, nlen);
5146 code += 1 + LINK_SIZE;
5147 nlen += 1 + LINK_SIZE;
5148 *bracode = OP_BRAPOS;
5149 *code++ = OP_KETRPOS;
5150 PUTINC(code, 0, nlen);
5151 PUT(bracode, 1, nlen);
5152 }
5153
5154 /* For non-COND brackets, we modify the BRA code and use KETRPOS. */
5155
5156 else
5157 {
5158 *bracode += 1; /* Switch to xxxPOS opcodes */
5159 *ketcode = OP_KETRPOS;
5160 }
5161
5162 /* If the minimum is zero, mark it as possessive, then unset the
5163 possessive flag when the minimum is 0 or 1. */
5164
5165 if (brazeroptr != NULL) *brazeroptr = OP_BRAPOSZERO;
5166 if (repeat_min < 2) possessive_quantifier = FALSE;
5167 }
5168
5169 /* Non-possessive quantifier */
5170
5171 else *ketcode = OP_KETRMAX + repeat_type;
5172 }
5173 }
5174 }
5175
5176 /* If previous is OP_FAIL, it was generated by an empty class [] in
5177 JavaScript mode. The other ways in which OP_FAIL can be generated, that is
5178 by (*FAIL) or (?!) set previous to NULL, which gives a "nothing to repeat"
5179 error above. We can just ignore the repeat in JS case. */
5180
5181 else if (*previous == OP_FAIL) goto END_REPEAT;
5182
5183 /* Else there's some kind of shambles */
5184
5185 else
5186 {
5187 *errorcodeptr = ERR11;
5188 goto FAILED;
5189 }
5190
5191 /* If the character following a repeat is '+', or if certain optimization
5192 tests above succeeded, possessive_quantifier is TRUE. For some opcodes,
5193 there are special alternative opcodes for this case. For anything else, we
5194 wrap the entire repeated item inside OP_ONCE brackets. Logically, the '+'
5195 notation is just syntactic sugar, taken from Sun's Java package, but the
5196 special opcodes can optimize it.
5197
5198 Some (but not all) possessively repeated subpatterns have already been
5199 completely handled in the code just above. For them, possessive_quantifier
5200 is always FALSE at this stage.
5201
5202 Note that the repeated item starts at tempcode, not at previous, which
5203 might be the first part of a string whose (former) last char we repeated.
5204
5205 Possessifying an 'exact' quantifier has no effect, so we can ignore it. But
5206 an 'upto' may follow. We skip over an 'exact' item, and then test the
5207 length of what remains before proceeding. */
5208
5209 if (possessive_quantifier)
5210 {
5211 int len;
5212
5213 if (*tempcode == OP_TYPEEXACT)
5214 tempcode += _pcre_OP_lengths[*tempcode] +
5215 ((tempcode[3] == OP_PROP || tempcode[3] == OP_NOTPROP)? 2 : 0);
5216
5217 else if (*tempcode == OP_EXACT || *tempcode == OP_NOTEXACT)
5218 {
5219 tempcode += _pcre_OP_lengths[*tempcode];
5220 #ifdef SUPPORT_UTF8
5221 if (utf8 && tempcode[-1] >= 0xc0)
5222 tempcode += _pcre_utf8_table4[tempcode[-1] & 0x3f];
5223 #endif
5224 }
5225
5226 len = (int)(code - tempcode);
5227 if (len > 0) switch (*tempcode)
5228 {
5229 case OP_STAR: *tempcode = OP_POSSTAR; break;
5230 case OP_PLUS: *tempcode = OP_POSPLUS; break;
5231 case OP_QUERY: *tempcode = OP_POSQUERY; break;
5232 case OP_UPTO: *tempcode = OP_POSUPTO; break;
5233
5234 case OP_STARI: *tempcode = OP_POSSTARI; break;
5235 case OP_PLUSI: *tempcode = OP_POSPLUSI; break;
5236 case OP_QUERYI: *tempcode = OP_POSQUERYI; break;
5237 case OP_UPTOI: *tempcode = OP_POSUPTOI; break;
5238
5239 case OP_NOTSTAR: *tempcode = OP_NOTPOSSTAR; break;
5240 case OP_NOTPLUS: *tempcode = OP_NOTPOSPLUS; break;
5241 case OP_NOTQUERY: *tempcode = OP_NOTPOSQUERY; break;
5242 case OP_NOTUPTO: *tempcode = OP_NOTPOSUPTO; break;
5243
5244 case OP_NOTSTARI: *tempcode = OP_NOTPOSSTARI; break;
5245 case OP_NOTPLUSI: *tempcode = OP_NOTPOSPLUSI; break;
5246 case OP_NOTQUERYI: *tempcode = OP_NOTPOSQUERYI; break;
5247 case OP_NOTUPTOI: *tempcode = OP_NOTPOSUPTOI; break;
5248
5249 case OP_TYPESTAR: *tempcode = OP_TYPEPOSSTAR; break;
5250 case OP_TYPEPLUS: *tempcode = OP_TYPEPOSPLUS; break;
5251 case OP_TYPEQUERY: *tempcode = OP_TYPEPOSQUERY; break;
5252 case OP_TYPEUPTO: *tempcode = OP_TYPEPOSUPTO; break;
5253
5254 /* Because we are moving code along, we must ensure that any
5255 pending recursive references are updated. */
5256
5257 default:
5258 *code = OP_END;
5259 adjust_recurse(tempcode, 1 + LINK_SIZE, utf8, cd, save_hwm);
5260 memmove(tempcode + 1+LINK_SIZE, tempcode, len);
5261 code += 1 + LINK_SIZE;
5262 len += 1 + LINK_SIZE;
5263 tempcode[0] = OP_ONCE;
5264 *code++ = OP_KET;
5265 PUTINC(code, 0, len);
5266 PUT(tempcode, 1, len);
5267 break;
5268 }
5269 }
5270
5271 /* In all case we no longer have a previous item. We also set the
5272 "follows varying string" flag for subsequently encountered reqbytes if
5273 it isn't already set and we have just passed a varying length item. */
5274
5275 END_REPEAT:
5276 previous = NULL;
5277 cd->req_varyopt |= reqvary;
5278 break;
5279
5280
5281 /* ===================================================================*/
5282 /* Start of nested parenthesized sub-expression, or comment or lookahead or
5283 lookbehind or option setting or condition or all the other extended
5284 parenthesis forms. */
5285
5286 case CHAR_LEFT_PARENTHESIS:
5287 newoptions = options;
5288 skipbytes = 0;
5289 bravalue = OP_CBRA;
5290 save_hwm = cd->hwm;
5291 reset_bracount = FALSE;
5292
5293 /* First deal with various "verbs" that can be introduced by '*'. */
5294
5295 if (*(++ptr) == CHAR_ASTERISK &&
5296 ((cd->ctypes[ptr[1]] & ctype_letter) != 0 || ptr[1] == ':'))
5297 {
5298 int i, namelen;
5299 int arglen = 0;
5300 const char *vn = verbnames;
5301 const uschar *name = ptr + 1;
5302 const uschar *arg = NULL;
5303 previous = NULL;
5304 while ((cd->ctypes[*++ptr] & ctype_letter) != 0) {};
5305 namelen = (int)(ptr - name);
5306
5307 /* It appears that Perl allows any characters whatsoever, other than
5308 a closing parenthesis, to appear in arguments, so we no longer insist on
5309 letters, digits, and underscores. */
5310
5311 if (*ptr == CHAR_COLON)
5312 {
5313 arg = ++ptr;
5314 while (*ptr != 0 && *ptr != CHAR_RIGHT_PARENTHESIS) ptr++;
5315 arglen = (int)(ptr - arg);
5316 }
5317
5318 if (*ptr != CHAR_RIGHT_PARENTHESIS)
5319 {
5320 *errorcodeptr = ERR60;
5321 goto FAILED;
5322 }
5323
5324 /* Scan the table of verb names */
5325
5326 for (i = 0; i < verbcount; i++)
5327 {
5328 if (namelen == verbs[i].len &&
5329 strncmp((char *)name, vn, namelen) == 0)
5330 {
5331 /* Check for open captures before ACCEPT and convert it to
5332 ASSERT_ACCEPT if in an assertion. */
5333
5334 if (verbs[i].op == OP_ACCEPT)
5335 {
5336 open_capitem *oc;
5337 if (arglen != 0)
5338 {
5339 *errorcodeptr = ERR59;
5340 goto FAILED;
5341 }
5342 cd->had_accept = TRUE;
5343 for (oc = cd->open_caps; oc != NULL; oc = oc->next)
5344 {
5345 *code++ = OP_CLOSE;
5346 PUT2INC(code, 0, oc->number);
5347 }
5348 *code++ = (cd->assert_depth > 0)? OP_ASSERT_ACCEPT : OP_ACCEPT;
5349
5350 /* Do not set firstbyte after *ACCEPT */
5351 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
5352 }
5353
5354 /* Handle other cases with/without an argument */
5355
5356 else if (arglen == 0)
5357 {
5358 if (verbs[i].op < 0) /* Argument is mandatory */
5359 {
5360 *errorcodeptr = ERR66;
5361 goto FAILED;
5362 }
5363 *code = verbs[i].op;
5364 if (*code++ == OP_THEN) cd->external_flags |= PCRE_HASTHEN;
5365 }
5366
5367 else
5368 {
5369 if (verbs[i].op_arg < 0) /* Argument is forbidden */
5370 {
5371 *errorcodeptr = ERR59;
5372 goto FAILED;
5373 }
5374 *code = verbs[i].op_arg;
5375 if (*code++ == OP_THEN_ARG) cd->external_flags |= PCRE_HASTHEN;
5376 *code++ = arglen;
5377 memcpy(code, arg, arglen);
5378 code += arglen;
5379 *code++ = 0;
5380 }
5381
5382 break; /* Found verb, exit loop */
5383 }
5384
5385 vn += verbs[i].len + 1;
5386 }
5387
5388 if (i < verbcount) continue; /* Successfully handled a verb */
5389 *errorcodeptr = ERR60; /* Verb not recognized */
5390 goto FAILED;
5391 }
5392
5393 /* Deal with the extended parentheses; all are introduced by '?', and the
5394 appearance of any of them means that this is not a capturing group. */
5395
5396 else if (*ptr == CHAR_QUESTION_MARK)
5397 {
5398 int i, set, unset, namelen;
5399 int *optset;
5400 const uschar *name;
5401 uschar *slot;
5402
5403 switch (*(++ptr))
5404 {
5405 case CHAR_NUMBER_SIGN: /* Comment; skip to ket */
5406 ptr++;
5407 while (*ptr != 0 && *ptr != CHAR_RIGHT_PARENTHESIS) ptr++;
5408 if (*ptr == 0)
5409 {
5410 *errorcodeptr = ERR18;
5411 goto FAILED;
5412 }
5413 continue;
5414
5415
5416 /* ------------------------------------------------------------ */
5417 case CHAR_VERTICAL_LINE: /* Reset capture count for each branch */
5418 reset_bracount = TRUE;
5419 /* Fall through */
5420
5421 /* ------------------------------------------------------------ */
5422 case CHAR_COLON: /* Non-capturing bracket */
5423 bravalue = OP_BRA;
5424 ptr++;
5425 break;
5426
5427
5428 /* ------------------------------------------------------------ */
5429 case CHAR_LEFT_PARENTHESIS:
5430 bravalue = OP_COND; /* Conditional group */
5431
5432 /* A condition can be an assertion, a number (referring to a numbered
5433 group), a name (referring to a named group), or 'R', referring to
5434 recursion. R<digits> and R&name are also permitted for recursion tests.
5435
5436 There are several syntaxes for testing a named group: (?(name)) is used
5437 by Python; Perl 5.10 onwards uses (?(<name>) or (?('name')).
5438
5439 There are two unfortunate ambiguities, caused by history. (a) 'R' can
5440 be the recursive thing or the name 'R' (and similarly for 'R' followed
5441 by digits), and (b) a number could be a name that consists of digits.
5442 In both cases, we look for a name first; if not found, we try the other
5443 cases. */
5444
5445 /* For conditions that are assertions, check the syntax, and then exit
5446 the switch. This will take control down to where bracketed groups,
5447 including assertions, are processed. */
5448
5449 if (ptr[1] == CHAR_QUESTION_MARK && (ptr[2] == CHAR_EQUALS_SIGN ||
5450 ptr[2] == CHAR_EXCLAMATION_MARK || ptr[2] == CHAR_LESS_THAN_SIGN))
5451 break;
5452
5453 /* Most other conditions use OP_CREF (a couple change to OP_RREF
5454 below), and all need to skip 3 bytes at the start of the group. */
5455
5456 code[1+LINK_SIZE] = OP_CREF;
5457 skipbytes = 3;
5458 refsign = -1;
5459
5460 /* Check for a test for recursion in a named group. */
5461
5462 if (ptr[1] == CHAR_R && ptr[2] == CHAR_AMPERSAND)
5463 {
5464 terminator = -1;
5465 ptr += 2;
5466 code[1+LINK_SIZE] = OP_RREF; /* Change the type of test */
5467 }
5468
5469 /* Check for a test for a named group's having been set, using the Perl
5470 syntax (?(<name>) or (?('name') */
5471
5472 else if (ptr[1] == CHAR_LESS_THAN_SIGN)
5473 {
5474 terminator = CHAR_GREATER_THAN_SIGN;
5475 ptr++;
5476 }
5477 else if (ptr[1] == CHAR_APOSTROPHE)
5478 {
5479 terminator = CHAR_APOSTROPHE;
5480 ptr++;
5481 }
5482 else
5483 {
5484 terminator = 0;
5485 if (ptr[1] == CHAR_MINUS || ptr[1] == CHAR_PLUS) refsign = *(++ptr);
5486 }
5487
5488 /* We now expect to read a name; any thing else is an error */
5489
5490 if ((cd->ctypes[ptr[1]] & ctype_word) == 0)
5491 {
5492 ptr += 1; /* To get the right offset */
5493 *errorcodeptr = ERR28;
5494 goto FAILED;
5495 }
5496
5497 /* Read the name, but also get it as a number if it's all digits */
5498
5499 recno = 0;
5500 name = ++ptr;
5501 while ((cd->ctypes[*ptr] & ctype_word) != 0)
5502 {
5503 if (recno >= 0)
5504 recno = ((digitab[*ptr] & ctype_digit) != 0)?
5505 recno * 10 + *ptr - CHAR_0 : -1;
5506 ptr++;
5507 }
5508 namelen = (int)(ptr - name);
5509
5510 if ((terminator > 0 && *ptr++ != terminator) ||
5511 *ptr++ != CHAR_RIGHT_PARENTHESIS)
5512 {
5513 ptr--; /* Error offset */
5514 *errorcodeptr = ERR26;
5515 goto FAILED;
5516 }
5517
5518 /* Do no further checking in the pre-compile phase. */
5519
5520 if (lengthptr != NULL) break;
5521
5522 /* In the real compile we do the work of looking for the actual
5523 reference. If the string started with "+" or "-" we require the rest to
5524 be digits, in which case recno will be set. */
5525
5526 if (refsign > 0)
5527 {
5528 if (recno <= 0)
5529 {
5530 *errorcodeptr = ERR58;
5531 goto FAILED;
5532 }
5533 recno = (refsign == CHAR_MINUS)?
5534 cd->bracount - recno + 1 : recno +cd->bracount;
5535 if (recno <= 0 || recno > cd->final_bracount)
5536 {
5537 *errorcodeptr = ERR15;
5538 goto FAILED;
5539 }
5540 PUT2(code, 2+LINK_SIZE, recno);
5541 break;
5542 }
5543
5544 /* Otherwise (did not start with "+" or "-"), start by looking for the
5545 name. If we find a name, add one to the opcode to change OP_CREF or
5546 OP_RREF into OP_NCREF or OP_NRREF. These behave exactly the same,
5547 except they record that the reference was originally to a name. The
5548 information is used to check duplicate names. */
5549
5550 slot = cd->name_table;
5551 for (i = 0; i < cd->names_found; i++)
5552 {
5553 if (strncmp((char *)name, (char *)slot+2, namelen) == 0) break;
5554 slot += cd->name_entry_size;
5555 }
5556
5557 /* Found a previous named subpattern */
5558
5559 if (i < cd->names_found)
5560 {
5561 recno = GET2(slot, 0);
5562 PUT2(code, 2+LINK_SIZE, recno);
5563 code[1+LINK_SIZE]++;
5564 }
5565
5566 /* Search the pattern for a forward reference */
5567
5568 else if ((i = find_parens(cd, name, namelen,
5569 (options & PCRE_EXTENDED) != 0, utf8)) > 0)
5570 {
5571 PUT2(code, 2+LINK_SIZE, i);
5572 code[1+LINK_SIZE]++;
5573 }
5574
5575 /* If terminator == 0 it means that the name followed directly after
5576 the opening parenthesis [e.g. (?(abc)...] and in this case there are
5577 some further alternatives to try. For the cases where terminator != 0
5578 [things like (?(<name>... or (?('name')... or (?(R&name)... ] we have
5579 now checked all the possibilities, so give an error. */
5580
5581 else if (terminator != 0)
5582 {
5583 *errorcodeptr = ERR15;
5584 goto FAILED;
5585 }
5586
5587 /* Check for (?(R) for recursion. Allow digits after R to specify a
5588 specific group number. */
5589
5590 else if (*name == CHAR_R)
5591 {
5592 recno = 0;
5593 for (i = 1; i < namelen; i++)
5594 {
5595 if ((digitab[name[i]] & ctype_digit) == 0)
5596 {
5597 *errorcodeptr = ERR15;
5598 goto FAILED;
5599 }
5600 recno = recno * 10 + name[i] - CHAR_0;
5601 }
5602 if (recno == 0) recno = RREF_ANY;
5603 code[1+LINK_SIZE] = OP_RREF; /* Change test type */
5604 PUT2(code, 2+LINK_SIZE, recno);
5605 }
5606
5607 /* Similarly, check for the (?(DEFINE) "condition", which is always
5608 false. */
5609
5610 else if (namelen == 6 && strncmp((char *)name, STRING_DEFINE, 6) == 0)
5611 {
5612 code[1+LINK_SIZE] = OP_DEF;
5613 skipbytes = 1;
5614 }
5615
5616 /* Check for the "name" actually being a subpattern number. We are
5617 in the second pass here, so final_bracount is set. */
5618
5619 else if (recno > 0 && recno <= cd->final_bracount)
5620 {
5621 PUT2(code, 2+LINK_SIZE, recno);
5622 }
5623
5624 /* Either an unidentified subpattern, or a reference to (?(0) */
5625
5626 else
5627 {
5628 *errorcodeptr = (recno == 0)? ERR35: ERR15;
5629 goto FAILED;
5630 }
5631 break;
5632
5633
5634 /* ------------------------------------------------------------ */
5635 case CHAR_EQUALS_SIGN: /* Positive lookahead */
5636 bravalue = OP_ASSERT;
5637 cd->assert_depth += 1;
5638 ptr++;
5639 break;
5640
5641
5642 /* ------------------------------------------------------------ */
5643 case CHAR_EXCLAMATION_MARK: /* Negative lookahead */
5644 ptr++;
5645 if (*ptr == CHAR_RIGHT_PARENTHESIS) /* Optimize (?!) */
5646 {
5647 *code++ = OP_FAIL;
5648 previous = NULL;
5649 continue;
5650 }
5651 bravalue = OP_ASSERT_NOT;
5652 cd->assert_depth += 1;
5653 break;
5654
5655
5656 /* ------------------------------------------------------------ */
5657 case CHAR_LESS_THAN_SIGN: /* Lookbehind or named define */
5658 switch (ptr[1])
5659 {
5660 case CHAR_EQUALS_SIGN: /* Positive lookbehind */
5661 bravalue = OP_ASSERTBACK;
5662 cd->assert_depth += 1;
5663 ptr += 2;
5664 break;
5665
5666 case CHAR_EXCLAMATION_MARK: /* Negative lookbehind */
5667 bravalue = OP_ASSERTBACK_NOT;
5668 cd->assert_depth += 1;
5669 ptr += 2;
5670 break;
5671
5672 default: /* Could be name define, else bad */
5673 if ((cd->ctypes[ptr[1]] & ctype_word) != 0) goto DEFINE_NAME;
5674 ptr++; /* Correct offset for error */
5675 *errorcodeptr = ERR24;
5676 goto FAILED;
5677 }
5678 break;
5679
5680
5681 /* ------------------------------------------------------------ */
5682 case CHAR_GREATER_THAN_SIGN: /* One-time brackets */
5683 bravalue = OP_ONCE;
5684 ptr++;
5685 break;
5686
5687
5688 /* ------------------------------------------------------------ */
5689 case CHAR_C: /* Callout - may be followed by digits; */
5690 previous_callout = code; /* Save for later completion */
5691 after_manual_callout = 1; /* Skip one item before completing */
5692 *code++ = OP_CALLOUT;
5693 {
5694 int n = 0;
5695 while ((digitab[*(++ptr)] & ctype_digit) != 0)
5696 n = n * 10 + *ptr - CHAR_0;
5697 if (*ptr != CHAR_RIGHT_PARENTHESIS)
5698 {
5699 *errorcodeptr = ERR39;
5700 goto FAILED;
5701 }
5702 if (n > 255)
5703 {
5704 *errorcodeptr = ERR38;
5705 goto FAILED;
5706 }
5707 *code++ = n;
5708 PUT(code, 0, (int)(ptr - cd->start_pattern + 1)); /* Pattern offset */
5709 PUT(code, LINK_SIZE, 0); /* Default length */
5710 code += 2 * LINK_SIZE;
5711 }
5712 previous = NULL;
5713 continue;
5714
5715
5716 /* ------------------------------------------------------------ */
5717 case CHAR_P: /* Python-style named subpattern handling */
5718 if (*(++ptr) == CHAR_EQUALS_SIGN ||
5719 *ptr == CHAR_GREATER_THAN_SIGN) /* Reference or recursion */
5720 {
5721 is_recurse = *ptr == CHAR_GREATER_THAN_SIGN;
5722 terminator = CHAR_RIGHT_PARENTHESIS;
5723 goto NAMED_REF_OR_RECURSE;
5724 }
5725 else if (*ptr != CHAR_LESS_THAN_SIGN) /* Test for Python-style defn */
5726 {
5727 *errorcodeptr = ERR41;
5728 goto FAILED;
5729 }
5730 /* Fall through to handle (?P< as (?< is handled */
5731
5732
5733 /* ------------------------------------------------------------ */
5734 DEFINE_NAME: /* Come here from (?< handling */
5735 case CHAR_APOSTROPHE:
5736 {
5737 terminator = (*ptr == CHAR_LESS_THAN_SIGN)?
5738 CHAR_GREATER_THAN_SIGN : CHAR_APOSTROPHE;
5739 name = ++ptr;
5740
5741 while ((cd->ctypes[*ptr] & ctype_word) != 0) ptr++;
5742 namelen = (int)(ptr - name);
5743
5744 /* In the pre-compile phase, just do a syntax check. */
5745
5746 if (lengthptr != NULL)
5747 {
5748 if (*ptr != terminator)
5749 {
5750 *errorcodeptr = ERR42;
5751 goto FAILED;
5752 }
5753 if (cd->names_found >= MAX_NAME_COUNT)
5754 {
5755 *errorcodeptr = ERR49;
5756 goto FAILED;
5757 }
5758 if (namelen + 3 > cd->name_entry_size)
5759 {
5760 cd->name_entry_size = namelen + 3;
5761 if (namelen > MAX_NAME_SIZE)
5762 {
5763 *errorcodeptr = ERR48;
5764 goto FAILED;
5765 }
5766 }
5767 }
5768
5769 /* In the real compile, create the entry in the table, maintaining
5770 alphabetical order. Duplicate names for different numbers are
5771 permitted only if PCRE_DUPNAMES is set. Duplicate names for the same
5772 number are always OK. (An existing number can be re-used if (?|
5773 appears in the pattern.) In either event, a duplicate name results in
5774 a duplicate entry in the table, even if the number is the same. This
5775 is because the number of names, and hence the table size, is computed
5776 in the pre-compile, and it affects various numbers and pointers which
5777 would all have to be modified, and the compiled code moved down, if
5778 duplicates with the same number were omitted from the table. This
5779 doesn't seem worth the hassle. However, *different* names for the
5780 same number are not permitted. */
5781
5782 else
5783 {
5784 BOOL dupname = FALSE;
5785 slot = cd->name_table;
5786
5787 for (i = 0; i < cd->names_found; i++)
5788 {
5789 int crc = memcmp(name, slot+2, namelen);
5790 if (crc == 0)
5791 {
5792 if (slot[2+namelen] == 0)
5793 {
5794 if (GET2(slot, 0) != cd->bracount + 1 &&
5795 (options & PCRE_DUPNAMES) == 0)
5796 {
5797 *errorcodeptr = ERR43;
5798 goto FAILED;
5799 }
5800 else dupname = TRUE;
5801 }
5802 else crc = -1; /* Current name is a substring */
5803 }
5804
5805 /* Make space in the table and break the loop for an earlier
5806 name. For a duplicate or later name, carry on. We do this for
5807 duplicates so that in the simple case (when ?(| is not used) they
5808 are in order of their numbers. */
5809
5810 if (crc < 0)
5811 {
5812 memmove(slot + cd->name_entry_size, slot,
5813 (cd->names_found - i) * cd->name_entry_size);
5814 break;
5815 }
5816
5817 /* Continue the loop for a later or duplicate name */
5818
5819 slot += cd->name_entry_size;
5820 }
5821
5822 /* For non-duplicate names, check for a duplicate number before
5823 adding the new name. */
5824
5825 if (!dupname)
5826 {
5827 uschar *cslot = cd->name_table;
5828 for (i = 0; i < cd->names_found; i++)
5829 {
5830 if (cslot != slot)
5831 {
5832 if (GET2(cslot, 0) == cd->bracount + 1)
5833 {
5834 *errorcodeptr = ERR65;
5835 goto FAILED;
5836 }
5837 }
5838 else i--;
5839 cslot += cd->name_entry_size;
5840 }
5841 }
5842
5843 PUT2(slot, 0, cd->bracount + 1);
5844 memcpy(slot + 2, name, namelen);
5845 slot[2+namelen] = 0;
5846 }
5847 }
5848
5849 /* In both pre-compile and compile, count the number of names we've
5850 encountered. */
5851
5852 cd->names_found++;
5853 ptr++; /* Move past > or ' */
5854 goto NUMBERED_GROUP;
5855
5856
5857 /* ------------------------------------------------------------ */
5858 case CHAR_AMPERSAND: /* Perl recursion/subroutine syntax */
5859 terminator = CHAR_RIGHT_PARENTHESIS;
5860 is_recurse = TRUE;
5861 /* Fall through */
5862
5863 /* We come here from the Python syntax above that handles both
5864 references (?P=name) and recursion (?P>name), as well as falling
5865 through from the Perl recursion syntax (?&name). We also come here from
5866 the Perl \k<name> or \k'name' back reference syntax and the \k{name}
5867 .NET syntax, and the Oniguruma \g<...> and \g'...' subroutine syntax. */
5868
5869 NAMED_REF_OR_RECURSE:
5870 name = ++ptr;
5871 while ((cd->ctypes[*ptr] & ctype_word) != 0) ptr++;
5872 namelen = (int)(ptr - name);
5873
5874 /* In the pre-compile phase, do a syntax check. We used to just set
5875 a dummy reference number, because it was not used in the first pass.
5876 However, with the change of recursive back references to be atomic,
5877 we have to look for the number so that this state can be identified, as
5878 otherwise the incorrect length is computed. If it's not a backwards
5879 reference, the dummy number will do. */
5880
5881 if (lengthptr != NULL)
5882 {
5883 const uschar *temp;
5884
5885 if (namelen == 0)
5886 {
5887 *errorcodeptr = ERR62;
5888 goto FAILED;
5889 }
5890 if (*ptr != terminator)
5891 {
5892 *errorcodeptr = ERR42;
5893 goto FAILED;
5894 }
5895 if (namelen > MAX_NAME_SIZE)
5896 {
5897 *errorcodeptr = ERR48;
5898 goto FAILED;
5899 }
5900
5901 /* The name table does not exist in the first pass, so we cannot
5902 do a simple search as in the code below. Instead, we have to scan the
5903 pattern to find the number. It is important that we scan it only as
5904 far as we have got because the syntax of named subpatterns has not
5905 been checked for the rest of the pattern, and find_parens() assumes
5906 correct syntax. In any case, it's a waste of resources to scan
5907 further. We stop the scan at the current point by temporarily
5908 adjusting the value of cd->endpattern. */
5909
5910 temp = cd->end_pattern;
5911 cd->end_pattern = ptr;
5912 recno = find_parens(cd, name, namelen,
5913 (options & PCRE_EXTENDED) != 0, utf8);
5914 cd->end_pattern = temp;
5915 if (recno < 0) recno = 0; /* Forward ref; set dummy number */
5916 }
5917
5918 /* In the real compile, seek the name in the table. We check the name
5919 first, and then check that we have reached the end of the name in the
5920 table. That way, if the name that is longer than any in the table,
5921 the comparison will fail without reading beyond the table entry. */
5922
5923 else
5924 {
5925 slot = cd->name_table;
5926 for (i = 0; i < cd->names_found; i++)
5927 {
5928 if (strncmp((char *)name, (char *)slot+2, namelen) == 0 &&
5929 slot[2+namelen] == 0)
5930 break;
5931 slot += cd->name_entry_size;
5932 }
5933
5934 if (i < cd->names_found) /* Back reference */
5935 {
5936 recno = GET2(slot, 0);
5937 }
5938 else if ((recno = /* Forward back reference */
5939 find_parens(cd, name, namelen,
5940 (options & PCRE_EXTENDED) != 0, utf8)) <= 0)
5941 {
5942 *errorcodeptr = ERR15;
5943 goto FAILED;
5944 }
5945 }
5946
5947 /* In both phases, we can now go to the code than handles numerical
5948 recursion or backreferences. */
5949
5950 if (is_recurse) goto HANDLE_RECURSION;
5951 else goto HANDLE_REFERENCE;
5952
5953
5954 /* ------------------------------------------------------------ */
5955 case CHAR_R: /* Recursion */
5956 ptr++; /* Same as (?0) */
5957 /* Fall through */
5958
5959
5960 /* ------------------------------------------------------------ */
5961 case CHAR_MINUS: case CHAR_PLUS: /* Recursion or subroutine */
5962 case CHAR_0: case CHAR_1: case CHAR_2: case CHAR_3: case CHAR_4:
5963 case CHAR_5: case CHAR_6: case CHAR_7: case CHAR_8: case CHAR_9:
5964 {
5965 const uschar *called;
5966 terminator = CHAR_RIGHT_PARENTHESIS;
5967
5968 /* Come here from the \g<...> and \g'...' code (Oniguruma
5969 compatibility). However, the syntax has been checked to ensure that
5970 the ... are a (signed) number, so that neither ERR63 nor ERR29 will
5971 be called on this path, nor with the jump to OTHER_CHAR_AFTER_QUERY
5972 ever be taken. */
5973
5974 HANDLE_NUMERICAL_RECURSION:
5975
5976 if ((refsign = *ptr) == CHAR_PLUS)
5977 {
5978 ptr++;
5979 if ((digitab[*ptr] & ctype_digit) == 0)
5980 {
5981 *errorcodeptr = ERR63;
5982 goto FAILED;
5983 }
5984 }
5985 else if (refsign == CHAR_MINUS)
5986 {
5987 if ((digitab[ptr[1]] & ctype_digit) == 0)
5988 goto OTHER_CHAR_AFTER_QUERY;
5989 ptr++;
5990 }
5991
5992 recno = 0;
5993 while((digitab[*ptr] & ctype_digit) != 0)
5994 recno = recno * 10 + *ptr++ - CHAR_0;
5995
5996 if (*ptr != terminator)
5997 {
5998 *errorcodeptr = ERR29;
5999 goto FAILED;
6000 }
6001
6002 if (refsign == CHAR_MINUS)
6003 {
6004 if (recno == 0)
6005 {
6006 *errorcodeptr = ERR58;
6007 goto FAILED;
6008 }
6009 recno = cd->bracount - recno + 1;
6010 if (recno <= 0)
6011 {
6012 *errorcodeptr = ERR15;
6013 goto FAILED;
6014 }
6015 }
6016 else if (refsign == CHAR_PLUS)
6017 {
6018 if (recno == 0)
6019 {
6020 *errorcodeptr = ERR58;
6021 goto FAILED;
6022 }
6023 recno += cd->bracount;
6024 }
6025
6026 /* Come here from code above that handles a named recursion */
6027
6028 HANDLE_RECURSION:
6029
6030 previous = code;
6031 called = cd->start_code;
6032
6033 /* When we are actually compiling, find the bracket that is being
6034 referenced. Temporarily end the regex in case it doesn't exist before
6035 this point. If we end up with a forward reference, first check that
6036 the bracket does occur later so we can give the error (and position)
6037 now. Then remember this forward reference in the workspace so it can
6038 be filled in at the end. */
6039
6040 if (lengthptr == NULL)
6041 {
6042 *code = OP_END;
6043 if (recno != 0)
6044 called = _pcre_find_bracket(cd->start_code, utf8, recno);
6045
6046 /* Forward reference */
6047
6048 if (called == NULL)
6049 {
6050 if (find_parens(cd, NULL, recno,
6051 (options & PCRE_EXTENDED) != 0, utf8) < 0)
6052 {
6053 *errorcodeptr = ERR15;
6054 goto FAILED;
6055 }
6056
6057 /* Fudge the value of "called" so that when it is inserted as an
6058 offset below, what it actually inserted is the reference number
6059 of the group. Then remember the forward reference. */
6060
6061 called = cd->start_code + recno;
6062 if (cd->hwm >= cd->start_workspace + cd->workspace_size -
6063 WORK_SIZE_SAFETY_MARGIN)
6064 {
6065 *errorcodeptr = expand_workspace(cd);
6066 if (*errorcodeptr != 0) goto FAILED;
6067 }
6068 PUTINC(cd->hwm, 0, (int)(code + 1 - cd->start_code));
6069 }
6070
6071 /* If not a forward reference, and the subpattern is still open,
6072 this is a recursive call. We check to see if this is a left
6073 recursion that could loop for ever, and diagnose that case. We
6074 must not, however, do this check if we are in a conditional
6075 subpattern because the condition might be testing for recursion in
6076 a pattern such as /(?(R)a+|(?R)b)/, which is perfectly valid.
6077 Forever loops are also detected at runtime, so those that occur in
6078 conditional subpatterns will be picked up then. */
6079
6080 else if (GET(called, 1) == 0 && cond_depth <= 0 &&
6081 could_be_empty(called, code, bcptr, utf8, cd))
6082 {
6083 *errorcodeptr = ERR40;
6084 goto FAILED;
6085 }
6086 }
6087
6088 /* Insert the recursion/subroutine item. It does not have a set first
6089 byte (relevant if it is repeated, because it will then be wrapped
6090 with ONCE brackets). */
6091
6092 *code = OP_RECURSE;
6093 PUT(code, 1, (int)(called - cd->start_code));
6094 code += 1 + LINK_SIZE;
6095 groupsetfirstbyte = FALSE;
6096 }
6097
6098 /* Can't determine a first byte now */
6099
6100 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
6101 continue;
6102
6103
6104 /* ------------------------------------------------------------ */
6105 default: /* Other characters: check option setting */
6106 OTHER_CHAR_AFTER_QUERY:
6107 set = unset = 0;
6108 optset = &set;
6109
6110 while (*ptr != CHAR_RIGHT_PARENTHESIS && *ptr != CHAR_COLON)
6111 {
6112 switch (*ptr++)
6113 {
6114 case CHAR_MINUS: optset = &unset; break;
6115
6116 case CHAR_J: /* Record that it changed in the external options */
6117 *optset |= PCRE_DUPNAMES;
6118 cd->external_flags |= PCRE_JCHANGED;
6119 break;
6120
6121 case CHAR_i: *optset |= PCRE_CASELESS; break;
6122 case CHAR_m: *optset |= PCRE_MULTILINE; break;
6123 case CHAR_s: *optset |= PCRE_DOTALL; break;
6124 case CHAR_x: *optset |= PCRE_EXTENDED; break;
6125 case CHAR_U: *optset |= PCRE_UNGREEDY; break;
6126 case CHAR_X: *optset |= PCRE_EXTRA; break;
6127
6128 default: *errorcodeptr = ERR12;
6129 ptr--; /* Correct the offset */
6130 goto FAILED;
6131 }
6132 }
6133
6134 /* Set up the changed option bits, but don't change anything yet. */
6135
6136 newoptions = (options | set) & (~unset);
6137
6138 /* If the options ended with ')' this is not the start of a nested
6139 group with option changes, so the options change at this level. If this
6140 item is right at the start of the pattern, the options can be
6141 abstracted and made external in the pre-compile phase, and ignored in
6142 the compile phase. This can be helpful when matching -- for instance in
6143 caseless checking of required bytes.
6144
6145 If the code pointer is not (cd->start_code + 1 + LINK_SIZE), we are
6146 definitely *not* at the start of the pattern because something has been
6147 compiled. In the pre-compile phase, however, the code pointer can have
6148 that value after the start, because it gets reset as code is discarded
6149 during the pre-compile. However, this can happen only at top level - if
6150 we are within parentheses, the starting BRA will still be present. At
6151 any parenthesis level, the length value can be used to test if anything
6152 has been compiled at that level. Thus, a test for both these conditions
6153 is necessary to ensure we correctly detect the start of the pattern in
6154 both phases.
6155
6156 If we are not at the pattern start, reset the greedy defaults and the
6157 case value for firstbyte and reqbyte. */
6158
6159 if (*ptr == CHAR_RIGHT_PARENTHESIS)
6160 {
6161 if (code == cd->start_code + 1 + LINK_SIZE &&
6162 (lengthptr == NULL || *lengthptr == 2 + 2*LINK_SIZE))
6163 {
6164 cd->external_options = newoptions;
6165 }
6166 else
6167 {
6168 greedy_default = ((newoptions & PCRE_UNGREEDY) != 0);
6169 greedy_non_default = greedy_default ^ 1;
6170 req_caseopt = ((newoptions & PCRE_CASELESS) != 0)? REQ_CASELESS : 0;
6171 }
6172
6173 /* Change options at this level, and pass them back for use
6174 in subsequent branches. */
6175
6176 *optionsptr = options = newoptions;
6177 previous = NULL; /* This item can't be repeated */
6178 continue; /* It is complete */
6179 }
6180
6181 /* If the options ended with ':' we are heading into a nested group
6182 with possible change of options. Such groups are non-capturing and are
6183 not assertions of any kind. All we need to do is skip over the ':';
6184 the newoptions value is handled below. */
6185
6186 bravalue = OP_BRA;
6187 ptr++;
6188 } /* End of switch for character following (? */
6189 } /* End of (? handling */
6190
6191 /* Opening parenthesis not followed by '*' or '?'. If PCRE_NO_AUTO_CAPTURE
6192 is set, all unadorned brackets become non-capturing and behave like (?:...)
6193 brackets. */
6194
6195 else if ((options & PCRE_NO_AUTO_CAPTURE) != 0)
6196 {
6197 bravalue = OP_BRA;
6198 }
6199
6200 /* Else we have a capturing group. */
6201
6202 else
6203 {
6204 NUMBERED_GROUP:
6205 cd->bracount += 1;
6206 PUT2(code, 1+LINK_SIZE, cd->bracount);
6207 skipbytes = 2;
6208 }
6209
6210 /* Process nested bracketed regex. Assertions used not to be repeatable,
6211 but this was changed for Perl compatibility, so all kinds can now be
6212 repeated. We copy code into a non-register variable (tempcode) in order to
6213 be able to pass its address because some compilers complain otherwise. */
6214
6215 previous = code; /* For handling repetition */
6216 *code = bravalue;
6217 tempcode = code;
6218 tempreqvary = cd->req_varyopt; /* Save value before bracket */
6219 tempbracount = cd->bracount; /* Save value before bracket */
6220 length_prevgroup = 0; /* Initialize for pre-compile phase */
6221
6222 if (!compile_regex(
6223 newoptions, /* The complete new option state */
6224 &tempcode, /* Where to put code (updated) */
6225 &ptr, /* Input pointer (updated) */
6226 errorcodeptr, /* Where to put an error message */
6227 (bravalue == OP_ASSERTBACK ||
6228 bravalue == OP_ASSERTBACK_NOT), /* TRUE if back assert */
6229 reset_bracount, /* True if (?| group */
6230 skipbytes, /* Skip over bracket number */
6231 cond_depth +
6232 ((bravalue == OP_COND)?1:0), /* Depth of condition subpatterns */
6233 &subfirstbyte, /* For possible first char */
6234 &subreqbyte, /* For possible last char */
6235 bcptr, /* Current branch chain */
6236 cd, /* Tables block */
6237 (lengthptr == NULL)? NULL : /* Actual compile phase */
6238 &length_prevgroup /* Pre-compile phase */
6239 ))
6240 goto FAILED;
6241
6242 /* If this was an atomic group and there are no capturing groups within it,
6243 generate OP_ONCE_NC instead of OP_ONCE. */
6244
6245 if (bravalue == OP_ONCE && cd->bracount <= tempbracount)
6246 *code = OP_ONCE_NC;
6247
6248 if (bravalue >= OP_ASSERT && bravalue <= OP_ASSERTBACK_NOT)
6249 cd->assert_depth -= 1;
6250
6251 /* At the end of compiling, code is still pointing to the start of the
6252 group, while tempcode has been updated to point past the end of the group.
6253 The pattern pointer (ptr) is on the bracket.
6254
6255 If this is a conditional bracket, check that there are no more than
6256 two branches in the group, or just one if it's a DEFINE group. We do this
6257 in the real compile phase, not in the pre-pass, where the whole group may
6258 not be available. */
6259
6260 if (bravalue == OP_COND && lengthptr == NULL)
6261 {
6262 uschar *tc = code;
6263 int condcount = 0;
6264
6265 do {
6266 condcount++;
6267 tc += GET(tc,1);
6268 }
6269 while (*tc != OP_KET);
6270
6271 /* A DEFINE group is never obeyed inline (the "condition" is always
6272 false). It must have only one branch. */
6273
6274 if (code[LINK_SIZE+1] == OP_DEF)
6275 {
6276 if (condcount > 1)
6277 {
6278 *errorcodeptr = ERR54;
6279 goto FAILED;
6280 }
6281 bravalue = OP_DEF; /* Just a flag to suppress char handling below */
6282 }
6283
6284 /* A "normal" conditional group. If there is just one branch, we must not
6285 make use of its firstbyte or reqbyte, because this is equivalent to an
6286 empty second branch. */
6287
6288 else
6289 {
6290 if (condcount > 2)
6291 {
6292 *errorcodeptr = ERR27;
6293 goto FAILED;
6294 }
6295 if (condcount == 1) subfirstbyte = subreqbyte = REQ_NONE;
6296 }
6297 }
6298
6299 /* Error if hit end of pattern */
6300
6301 if (*ptr != CHAR_RIGHT_PARENTHESIS)
6302 {
6303 *errorcodeptr = ERR14;
6304 goto FAILED;
6305 }
6306
6307 /* In the pre-compile phase, update the length by the length of the group,
6308 less the brackets at either end. Then reduce the compiled code to just a
6309 set of non-capturing brackets so that it doesn't use much memory if it is
6310 duplicated by a quantifier.*/
6311
6312 if (lengthptr != NULL)
6313 {
6314 if (OFLOW_MAX - *lengthptr < length_prevgroup - 2 - 2*LINK_SIZE)
6315 {
6316 *errorcodeptr = ERR20;
6317 goto FAILED;
6318 }
6319 *lengthptr += length_prevgroup - 2 - 2*LINK_SIZE;
6320 code++; /* This already contains bravalue */
6321 PUTINC(code, 0, 1 + LINK_SIZE);
6322 *code++ = OP_KET;
6323 PUTINC(code, 0, 1 + LINK_SIZE);
6324 break; /* No need to waste time with special character handling */
6325 }
6326
6327 /* Otherwise update the main code pointer to the end of the group. */
6328
6329 code = tempcode;
6330
6331 /* For a DEFINE group, required and first character settings are not
6332 relevant. */
6333
6334 if (bravalue == OP_DEF) break;
6335
6336 /* Handle updating of the required and first characters for other types of
6337 group. Update for normal brackets of all kinds, and conditions with two
6338 branches (see code above). If the bracket is followed by a quantifier with
6339 zero repeat, we have to back off. Hence the definition of zeroreqbyte and
6340 zerofirstbyte outside the main loop so that they can be accessed for the
6341 back off. */
6342
6343 zeroreqbyte = reqbyte;
6344 zerofirstbyte = firstbyte;
6345 groupsetfirstbyte = FALSE;
6346
6347 if (bravalue >= OP_ONCE)
6348 {
6349 /* If we have not yet set a firstbyte in this branch, take it from the
6350 subpattern, remembering that it was set here so that a repeat of more
6351 than one can replicate it as reqbyte if necessary. If the subpattern has
6352 no firstbyte, set "none" for the whole branch. In both cases, a zero
6353 repeat forces firstbyte to "none". */
6354
6355 if (firstbyte == REQ_UNSET)
6356 {
6357 if (subfirstbyte >= 0)
6358 {
6359 firstbyte = subfirstbyte;
6360 groupsetfirstbyte = TRUE;
6361 }
6362 else firstbyte = REQ_NONE;
6363 zerofirstbyte = REQ_NONE;
6364 }
6365
6366 /* If firstbyte was previously set, convert the subpattern's firstbyte
6367 into reqbyte if there wasn't one, using the vary flag that was in
6368 existence beforehand. */
6369
6370 else if (subfirstbyte >= 0 && subreqbyte < 0)
6371 subreqbyte = subfirstbyte | tempreqvary;
6372
6373 /* If the subpattern set a required byte (or set a first byte that isn't
6374 really the first byte - see above), set it. */
6375
6376 if (subreqbyte >= 0) reqbyte = subreqbyte;
6377 }
6378
6379 /* For a forward assertion, we take the reqbyte, if set. This can be
6380 helpful if the pattern that follows the assertion doesn't set a different
6381 char. For example, it's useful for /(?=abcde).+/. We can't set firstbyte
6382 for an assertion, however because it leads to incorrect effect for patterns
6383 such as /(?=a)a.+/ when the "real" "a" would then become a reqbyte instead
6384 of a firstbyte. This is overcome by a scan at the end if there's no
6385 firstbyte, looking for an asserted first char. */
6386
6387 else if (bravalue == OP_ASSERT && subreqbyte >= 0) reqbyte = subreqbyte;
6388 break; /* End of processing '(' */
6389
6390
6391 /* ===================================================================*/
6392 /* Handle metasequences introduced by \. For ones like \d, the ESC_ values
6393 are arranged to be the negation of the corresponding OP_values in the
6394 default case when PCRE_UCP is not set. For the back references, the values
6395 are ESC_REF plus the reference number. Only back references and those types
6396 that consume a character may be repeated. We can test for values between
6397 ESC_b and ESC_Z for the latter; this may have to change if any new ones are
6398 ever created. */
6399
6400 case CHAR_BACKSLASH:
6401 tempptr = ptr;
6402 c = check_escape(&ptr, errorcodeptr, cd->bracount, options, FALSE);
6403 if (*errorcodeptr != 0) goto FAILED;
6404
6405 if (c < 0)
6406 {
6407 if (-c == ESC_Q) /* Handle start of quoted string */
6408 {
6409 if (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
6410 ptr += 2; /* avoid empty string */
6411 else inescq = TRUE;
6412 continue;
6413 }
6414
6415 if (-c == ESC_E) continue; /* Perl ignores an orphan \E */
6416
6417 /* For metasequences that actually match a character, we disable the
6418 setting of a first character if it hasn't already been set. */
6419
6420 if (firstbyte == REQ_UNSET && -c > ESC_b && -c < ESC_Z)
6421 firstbyte = REQ_NONE;
6422
6423 /* Set values to reset to if this is followed by a zero repeat. */
6424
6425 zerofirstbyte = firstbyte;
6426 zeroreqbyte = reqbyte;
6427
6428 /* \g<name> or \g'name' is a subroutine call by name and \g<n> or \g'n'
6429 is a subroutine call by number (Oniguruma syntax). In fact, the value
6430 -ESC_g is returned only for these cases. So we don't need to check for <
6431 or ' if the value is -ESC_g. For the Perl syntax \g{n} the value is
6432 -ESC_REF+n, and for the Perl syntax \g{name} the result is -ESC_k (as
6433 that is a synonym for a named back reference). */
6434
6435 if (-c == ESC_g)
6436 {
6437 const uschar *p;
6438 save_hwm = cd->hwm; /* Normally this is set when '(' is read */
6439 terminator = (*(++ptr) == CHAR_LESS_THAN_SIGN)?
6440 CHAR_GREATER_THAN_SIGN : CHAR_APOSTROPHE;
6441
6442 /* These two statements stop the compiler for warning about possibly
6443 unset variables caused by the jump to HANDLE_NUMERICAL_RECURSION. In
6444 fact, because we actually check for a number below, the paths that
6445 would actually be in error are never taken. */
6446
6447 skipbytes = 0;
6448 reset_bracount = FALSE;
6449
6450 /* Test for a name */
6451
6452 if (ptr[1] != CHAR_PLUS && ptr[1] != CHAR_MINUS)
6453 {
6454 BOOL is_a_number = TRUE;
6455 for (p = ptr + 1; *p != 0 && *p != terminator; p++)
6456 {
6457 if ((cd->ctypes[*p] & ctype_digit) == 0) is_a_number = FALSE;
6458 if ((cd->ctypes[*p] & ctype_word) == 0) break;
6459 }
6460 if (*p != terminator)
6461 {
6462 *errorcodeptr = ERR57;
6463 break;
6464 }
6465 if (is_a_number)
6466 {
6467 ptr++;
6468 goto HANDLE_NUMERICAL_RECURSION;
6469 }
6470 is_recurse = TRUE;
6471 goto NAMED_REF_OR_RECURSE;
6472 }
6473
6474 /* Test a signed number in angle brackets or quotes. */
6475
6476 p = ptr + 2;
6477 while ((digitab[*p] & ctype_digit) != 0) p++;
6478 if (*p != terminator)
6479 {
6480 *errorcodeptr = ERR57;
6481 break;
6482 }
6483 ptr++;
6484 goto HANDLE_NUMERICAL_RECURSION;
6485 }
6486
6487 /* \k<name> or \k'name' is a back reference by name (Perl syntax).
6488 We also support \k{name} (.NET syntax). */
6489
6490 if (-c == ESC_k)
6491 {
6492 if ((ptr[1] != CHAR_LESS_THAN_SIGN &&
6493 ptr[1] != CHAR_APOSTROPHE && ptr[1] != CHAR_LEFT_CURLY_BRACKET))
6494 {
6495 *errorcodeptr = ERR69;
6496 break;
6497 }
6498 is_recurse = FALSE;
6499 terminator = (*(++ptr) == CHAR_LESS_THAN_SIGN)?
6500 CHAR_GREATER_THAN_SIGN : (*ptr == CHAR_APOSTROPHE)?
6501 CHAR_APOSTROPHE : CHAR_RIGHT_CURLY_BRACKET;
6502 goto NAMED_REF_OR_RECURSE;
6503 }
6504
6505 /* Back references are handled specially; must disable firstbyte if
6506 not set to cope with cases like (?=(\w+))\1: which would otherwise set
6507 ':' later. */
6508
6509 if (-c >= ESC_REF)
6510 {
6511 open_capitem *oc;
6512 recno = -c - ESC_REF;
6513
6514 HANDLE_REFERENCE: /* Come here from named backref handling */
6515 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
6516 previous = code;
6517 *code++ = ((options & PCRE_CASELESS) != 0)? OP_REFI : OP_REF;
6518 PUT2INC(code, 0, recno);
6519 cd->backref_map |= (recno < 32)? (1 << recno) : 1;
6520 if (recno > cd->top_backref) cd->top_backref = recno;
6521
6522 /* Check to see if this back reference is recursive, that it, it
6523 is inside the group that it references. A flag is set so that the
6524 group can be made atomic. */
6525
6526 for (oc = cd->open_caps; oc != NULL; oc = oc->next)
6527 {
6528 if (oc->number == recno)
6529 {
6530 oc->flag = TRUE;
6531 break;
6532 }
6533 }
6534 }
6535
6536 /* So are Unicode property matches, if supported. */
6537
6538 #ifdef SUPPORT_UCP
6539 else if (-c == ESC_P || -c == ESC_p)
6540 {
6541 BOOL negated;
6542 int pdata;
6543 int ptype = get_ucp(&ptr, &negated, &pdata, errorcodeptr);
6544 if (ptype < 0) goto FAILED;
6545 previous = code;
6546 *code++ = ((-c == ESC_p) != negated)? OP_PROP : OP_NOTPROP;
6547 *code++ = ptype;
6548 *code++ = pdata;
6549 }
6550 #else
6551
6552 /* If Unicode properties are not supported, \X, \P, and \p are not
6553 allowed. */
6554
6555 else if (-c == ESC_X || -c == ESC_P || -c == ESC_p)
6556 {
6557 *errorcodeptr = ERR45;
6558 goto FAILED;
6559 }
6560 #endif
6561
6562 /* For the rest (including \X when Unicode properties are supported), we
6563 can obtain the OP value by negating the escape value in the default
6564 situation when PCRE_UCP is not set. When it *is* set, we substitute
6565 Unicode property tests. */
6566
6567 else
6568 {
6569 #ifdef SUPPORT_UCP
6570 if (-c >= ESC_DU && -c <= ESC_wu)
6571 {
6572 nestptr = ptr + 1; /* Where to resume */
6573 ptr = substitutes[-c - ESC_DU] - 1; /* Just before substitute */
6574 }
6575 else
6576 #endif
6577 /* In non-UTF-8 mode, we turn \C into OP_ALLANY instead of OP_ANYBYTE
6578 so that it works in DFA mode and in lookbehinds. */
6579
6580 {
6581 previous = (-c > ESC_b && -c < ESC_Z)? code : NULL;
6582 *code++ = (!utf8 && c == -ESC_C)? OP_ALLANY : -c;
6583 }
6584 }
6585 continue;
6586 }
6587
6588 /* We have a data character whose value is in c. In UTF-8 mode it may have
6589 a value > 127. We set its representation in the length/buffer, and then
6590 handle it as a data character. */
6591
6592 #ifdef SUPPORT_UTF8
6593 if (utf8 && c > 127)
6594 mclength = _pcre_ord2utf8(c, mcbuffer);
6595 else
6596 #endif
6597
6598 {
6599 mcbuffer[0] = c;
6600 mclength = 1;
6601 }
6602 goto ONE_CHAR;
6603
6604
6605 /* ===================================================================*/
6606 /* Handle a literal character. It is guaranteed not to be whitespace or #
6607 when the extended flag is set. If we are in UTF-8 mode, it may be a
6608 multi-byte literal character. */
6609
6610 default:
6611 NORMAL_CHAR:
6612 mclength = 1;
6613 mcbuffer[0] = c;
6614
6615 #ifdef SUPPORT_UTF8
6616 if (utf8 && c >= 0xc0)
6617 {
6618 while ((ptr[1] & 0xc0) == 0x80)
6619 mcbuffer[mclength++] = *(++ptr);
6620 }
6621 #endif
6622
6623 /* At this point we have the character's bytes in mcbuffer, and the length
6624 in mclength. When not in UTF-8 mode, the length is always 1. */
6625
6626 ONE_CHAR:
6627 previous = code;
6628 *code++ = ((options & PCRE_CASELESS) != 0)? OP_CHARI : OP_CHAR;
6629 for (c = 0; c < mclength; c++) *code++ = mcbuffer[c];
6630
6631 /* Remember if \r or \n were seen */
6632
6633 if (mcbuffer[0] == CHAR_CR || mcbuffer[0] == CHAR_NL)
6634 cd->external_flags |= PCRE_HASCRORLF;
6635
6636 /* Set the first and required bytes appropriately. If no previous first
6637 byte, set it from this character, but revert to none on a zero repeat.
6638 Otherwise, leave the firstbyte value alone, and don't change it on a zero
6639 repeat. */
6640
6641 if (firstbyte == REQ_UNSET)
6642 {
6643 zerofirstbyte = REQ_NONE;
6644 zeroreqbyte = reqbyte;
6645
6646 /* If the character is more than one byte long, we can set firstbyte <