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

Contents of /code/trunk/pcre_compile.c

Parent Directory Parent Directory | Revision Log Revision Log


Revision 751 - (show annotations)
Fri Nov 18 11:13:30 2011 UTC (7 years, 9 months ago) by ph10
File MIME type: text/plain
File size: 250553 byte(s)
Error occurred while calculating annotation data.
Code tidies.
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. */
92
93 #define COMPILE_WORK_SIZE (4096)
94
95 /* The overrun tests check for a slightly smaller size so that they detect the
96 overrun before it actually does run off the end of the data block. */
97
98 #define WORK_SIZE_CHECK (COMPILE_WORK_SIZE - 100)
99
100
101 /* Table for handling escaped characters in the range '0'-'z'. Positive returns
102 are simple data values; negative values are for special things like \d and so
103 on. Zero means further processing is needed (for things like \x), or the escape
104 is invalid. */
105
106 #ifndef EBCDIC
107
108 /* This is the "normal" table for ASCII systems or for EBCDIC systems running
109 in UTF-8 mode. */
110
111 static const short int escapes[] = {
112 0, 0,
113 0, 0,
114 0, 0,
115 0, 0,
116 0, 0,
117 CHAR_COLON, CHAR_SEMICOLON,
118 CHAR_LESS_THAN_SIGN, CHAR_EQUALS_SIGN,
119 CHAR_GREATER_THAN_SIGN, CHAR_QUESTION_MARK,
120 CHAR_COMMERCIAL_AT, -ESC_A,
121 -ESC_B, -ESC_C,
122 -ESC_D, -ESC_E,
123 0, -ESC_G,
124 -ESC_H, 0,
125 0, -ESC_K,
126 0, 0,
127 -ESC_N, 0,
128 -ESC_P, -ESC_Q,
129 -ESC_R, -ESC_S,
130 0, 0,
131 -ESC_V, -ESC_W,
132 -ESC_X, 0,
133 -ESC_Z, CHAR_LEFT_SQUARE_BRACKET,
134 CHAR_BACKSLASH, CHAR_RIGHT_SQUARE_BRACKET,
135 CHAR_CIRCUMFLEX_ACCENT, CHAR_UNDERSCORE,
136 CHAR_GRAVE_ACCENT, 7,
137 -ESC_b, 0,
138 -ESC_d, ESC_e,
139 ESC_f, 0,
140 -ESC_h, 0,
141 0, -ESC_k,
142 0, 0,
143 ESC_n, 0,
144 -ESC_p, 0,
145 ESC_r, -ESC_s,
146 ESC_tee, 0,
147 -ESC_v, -ESC_w,
148 0, 0,
149 -ESC_z
150 };
151
152 #else
153
154 /* This is the "abnormal" table for EBCDIC systems without UTF-8 support. */
155
156 static const short int escapes[] = {
157 /* 48 */ 0, 0, 0, '.', '<', '(', '+', '|',
158 /* 50 */ '&', 0, 0, 0, 0, 0, 0, 0,
159 /* 58 */ 0, 0, '!', '$', '*', ')', ';', '~',
160 /* 60 */ '-', '/', 0, 0, 0, 0, 0, 0,
161 /* 68 */ 0, 0, '|', ',', '%', '_', '>', '?',
162 /* 70 */ 0, 0, 0, 0, 0, 0, 0, 0,
163 /* 78 */ 0, '`', ':', '#', '@', '\'', '=', '"',
164 /* 80 */ 0, 7, -ESC_b, 0, -ESC_d, ESC_e, ESC_f, 0,
165 /* 88 */-ESC_h, 0, 0, '{', 0, 0, 0, 0,
166 /* 90 */ 0, 0, -ESC_k, 'l', 0, ESC_n, 0, -ESC_p,
167 /* 98 */ 0, ESC_r, 0, '}', 0, 0, 0, 0,
168 /* A0 */ 0, '~', -ESC_s, ESC_tee, 0,-ESC_v, -ESC_w, 0,
169 /* A8 */ 0,-ESC_z, 0, 0, 0, '[', 0, 0,
170 /* B0 */ 0, 0, 0, 0, 0, 0, 0, 0,
171 /* B8 */ 0, 0, 0, 0, 0, ']', '=', '-',
172 /* C0 */ '{',-ESC_A, -ESC_B, -ESC_C, -ESC_D,-ESC_E, 0, -ESC_G,
173 /* C8 */-ESC_H, 0, 0, 0, 0, 0, 0, 0,
174 /* D0 */ '}', 0, -ESC_K, 0, 0,-ESC_N, 0, -ESC_P,
175 /* D8 */-ESC_Q,-ESC_R, 0, 0, 0, 0, 0, 0,
176 /* E0 */ '\\', 0, -ESC_S, 0, 0,-ESC_V, -ESC_W, -ESC_X,
177 /* E8 */ 0,-ESC_Z, 0, 0, 0, 0, 0, 0,
178 /* F0 */ 0, 0, 0, 0, 0, 0, 0, 0,
179 /* F8 */ 0, 0, 0, 0, 0, 0, 0, 0
180 };
181 #endif
182
183
184 /* Table of special "verbs" like (*PRUNE). This is a short table, so it is
185 searched linearly. Put all the names into a single string, in order to reduce
186 the number of relocations when a shared library is dynamically linked. The
187 string is built from string macros so that it works in UTF-8 mode on EBCDIC
188 platforms. */
189
190 typedef struct verbitem {
191 int len; /* Length of verb name */
192 int op; /* Op when no arg, or -1 if arg mandatory */
193 int op_arg; /* Op when arg present, or -1 if not allowed */
194 } verbitem;
195
196 static const char verbnames[] =
197 "\0" /* Empty name is a shorthand for MARK */
198 STRING_MARK0
199 STRING_ACCEPT0
200 STRING_COMMIT0
201 STRING_F0
202 STRING_FAIL0
203 STRING_PRUNE0
204 STRING_SKIP0
205 STRING_THEN;
206
207 static const verbitem verbs[] = {
208 { 0, -1, OP_MARK },
209 { 4, -1, OP_MARK },
210 { 6, OP_ACCEPT, -1 },
211 { 6, OP_COMMIT, -1 },
212 { 1, OP_FAIL, -1 },
213 { 4, OP_FAIL, -1 },
214 { 5, OP_PRUNE, OP_PRUNE_ARG },
215 { 4, OP_SKIP, OP_SKIP_ARG },
216 { 4, OP_THEN, OP_THEN_ARG }
217 };
218
219 static const int verbcount = sizeof(verbs)/sizeof(verbitem);
220
221
222 /* Tables of names of POSIX character classes and their lengths. The names are
223 now all in a single string, to reduce the number of relocations when a shared
224 library is dynamically loaded. The list of lengths is terminated by a zero
225 length entry. The first three must be alpha, lower, upper, as this is assumed
226 for handling case independence. */
227
228 static const char posix_names[] =
229 STRING_alpha0 STRING_lower0 STRING_upper0 STRING_alnum0
230 STRING_ascii0 STRING_blank0 STRING_cntrl0 STRING_digit0
231 STRING_graph0 STRING_print0 STRING_punct0 STRING_space0
232 STRING_word0 STRING_xdigit;
233
234 static const uschar posix_name_lengths[] = {
235 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 4, 6, 0 };
236
237 /* Table of class bit maps for each POSIX class. Each class is formed from a
238 base map, with an optional addition or removal of another map. Then, for some
239 classes, there is some additional tweaking: for [:blank:] the vertical space
240 characters are removed, and for [:alpha:] and [:alnum:] the underscore
241 character is removed. The triples in the table consist of the base map offset,
242 second map offset or -1 if no second map, and a non-negative value for map
243 addition or a negative value for map subtraction (if there are two maps). The
244 absolute value of the third field has these meanings: 0 => no tweaking, 1 =>
245 remove vertical space characters, 2 => remove underscore. */
246
247 static const int posix_class_maps[] = {
248 cbit_word, cbit_digit, -2, /* alpha */
249 cbit_lower, -1, 0, /* lower */
250 cbit_upper, -1, 0, /* upper */
251 cbit_word, -1, 2, /* alnum - word without underscore */
252 cbit_print, cbit_cntrl, 0, /* ascii */
253 cbit_space, -1, 1, /* blank - a GNU extension */
254 cbit_cntrl, -1, 0, /* cntrl */
255 cbit_digit, -1, 0, /* digit */
256 cbit_graph, -1, 0, /* graph */
257 cbit_print, -1, 0, /* print */
258 cbit_punct, -1, 0, /* punct */
259 cbit_space, -1, 0, /* space */
260 cbit_word, -1, 0, /* word - a Perl extension */
261 cbit_xdigit,-1, 0 /* xdigit */
262 };
263
264 /* Table of substitutes for \d etc when PCRE_UCP is set. The POSIX class
265 substitutes must be in the order of the names, defined above, and there are
266 both positive and negative cases. NULL means no substitute. */
267
268 #ifdef SUPPORT_UCP
269 static const uschar *substitutes[] = {
270 (uschar *)"\\P{Nd}", /* \D */
271 (uschar *)"\\p{Nd}", /* \d */
272 (uschar *)"\\P{Xsp}", /* \S */ /* NOTE: Xsp is Perl space */
273 (uschar *)"\\p{Xsp}", /* \s */
274 (uschar *)"\\P{Xwd}", /* \W */
275 (uschar *)"\\p{Xwd}" /* \w */
276 };
277
278 static const uschar *posix_substitutes[] = {
279 (uschar *)"\\p{L}", /* alpha */
280 (uschar *)"\\p{Ll}", /* lower */
281 (uschar *)"\\p{Lu}", /* upper */
282 (uschar *)"\\p{Xan}", /* alnum */
283 NULL, /* ascii */
284 (uschar *)"\\h", /* blank */
285 NULL, /* cntrl */
286 (uschar *)"\\p{Nd}", /* digit */
287 NULL, /* graph */
288 NULL, /* print */
289 NULL, /* punct */
290 (uschar *)"\\p{Xps}", /* space */ /* NOTE: Xps is POSIX space */
291 (uschar *)"\\p{Xwd}", /* word */
292 NULL, /* xdigit */
293 /* Negated cases */
294 (uschar *)"\\P{L}", /* ^alpha */
295 (uschar *)"\\P{Ll}", /* ^lower */
296 (uschar *)"\\P{Lu}", /* ^upper */
297 (uschar *)"\\P{Xan}", /* ^alnum */
298 NULL, /* ^ascii */
299 (uschar *)"\\H", /* ^blank */
300 NULL, /* ^cntrl */
301 (uschar *)"\\P{Nd}", /* ^digit */
302 NULL, /* ^graph */
303 NULL, /* ^print */
304 NULL, /* ^punct */
305 (uschar *)"\\P{Xps}", /* ^space */ /* NOTE: Xps is POSIX space */
306 (uschar *)"\\P{Xwd}", /* ^word */
307 NULL /* ^xdigit */
308 };
309 #define POSIX_SUBSIZE (sizeof(posix_substitutes)/sizeof(uschar *))
310 #endif
311
312 #define STRING(a) # a
313 #define XSTRING(s) STRING(s)
314
315 /* The texts of compile-time error messages. These are "char *" because they
316 are passed to the outside world. Do not ever re-use any error number, because
317 they are documented. Always add a new error instead. Messages marked DEAD below
318 are no longer used. This used to be a table of strings, but in order to reduce
319 the number of relocations needed when a shared library is loaded dynamically,
320 it is now one long string. We cannot use a table of offsets, because the
321 lengths of inserts such as XSTRING(MAX_NAME_SIZE) are not known. Instead, we
322 simply count through to the one we want - this isn't a performance issue
323 because these strings are used only when there is a compilation error.
324
325 Each substring ends with \0 to insert a null character. This includes the final
326 substring, so that the whole string ends with \0\0, which can be detected when
327 counting through. */
328
329 static const char error_texts[] =
330 "no error\0"
331 "\\ at end of pattern\0"
332 "\\c at end of pattern\0"
333 "unrecognized character follows \\\0"
334 "numbers out of order in {} quantifier\0"
335 /* 5 */
336 "number too big in {} quantifier\0"
337 "missing terminating ] for character class\0"
338 "invalid escape sequence in character class\0"
339 "range out of order in character class\0"
340 "nothing to repeat\0"
341 /* 10 */
342 "operand of unlimited repeat could match the empty string\0" /** DEAD **/
343 "internal error: unexpected repeat\0"
344 "unrecognized character after (? or (?-\0"
345 "POSIX named classes are supported only within a class\0"
346 "missing )\0"
347 /* 15 */
348 "reference to non-existent subpattern\0"
349 "erroffset passed as NULL\0"
350 "unknown option bit(s) set\0"
351 "missing ) after comment\0"
352 "parentheses nested too deeply\0" /** DEAD **/
353 /* 20 */
354 "regular expression is too large\0"
355 "failed to get memory\0"
356 "unmatched parentheses\0"
357 "internal error: code overflow\0"
358 "unrecognized character after (?<\0"
359 /* 25 */
360 "lookbehind assertion is not fixed length\0"
361 "malformed number or name after (?(\0"
362 "conditional group contains more than two branches\0"
363 "assertion expected after (?(\0"
364 "(?R or (?[+-]digits must be followed by )\0"
365 /* 30 */
366 "unknown POSIX class name\0"
367 "POSIX collating elements are not supported\0"
368 "this version of PCRE is not compiled with PCRE_UTF8 support\0"
369 "spare error\0" /** DEAD **/
370 "character value in \\x{...} sequence is too large\0"
371 /* 35 */
372 "invalid condition (?(0)\0"
373 "\\C not allowed in lookbehind assertion\0"
374 "PCRE does not support \\L, \\l, \\N{name}, \\U, or \\u\0"
375 "number after (?C is > 255\0"
376 "closing ) for (?C expected\0"
377 /* 40 */
378 "recursive call could loop indefinitely\0"
379 "unrecognized character after (?P\0"
380 "syntax error in subpattern name (missing terminator)\0"
381 "two named subpatterns have the same name\0"
382 "invalid UTF-8 string\0"
383 /* 45 */
384 "support for \\P, \\p, and \\X has not been compiled\0"
385 "malformed \\P or \\p sequence\0"
386 "unknown property name after \\P or \\p\0"
387 "subpattern name is too long (maximum " XSTRING(MAX_NAME_SIZE) " characters)\0"
388 "too many named subpatterns (maximum " XSTRING(MAX_NAME_COUNT) ")\0"
389 /* 50 */
390 "repeated subpattern is too long\0" /** DEAD **/
391 "octal value is greater than \\377 (not in UTF-8 mode)\0"
392 "internal error: overran compiling workspace\0"
393 "internal error: previously-checked referenced subpattern not found\0"
394 "DEFINE group contains more than one branch\0"
395 /* 55 */
396 "repeating a DEFINE group is not allowed\0" /** DEAD **/
397 "inconsistent NEWLINE options\0"
398 "\\g is not followed by a braced, angle-bracketed, or quoted name/number or by a plain number\0"
399 "a numbered reference must not be zero\0"
400 "an argument is not allowed for (*ACCEPT), (*FAIL), or (*COMMIT)\0"
401 /* 60 */
402 "(*VERB) not recognized\0"
403 "number is too big\0"
404 "subpattern name expected\0"
405 "digit expected after (?+\0"
406 "] is an invalid data character in JavaScript compatibility mode\0"
407 /* 65 */
408 "different names for subpatterns of the same number are not allowed\0"
409 "(*MARK) must have an argument\0"
410 "this version of PCRE is not compiled with PCRE_UCP support\0"
411 "\\c must be followed by an ASCII character\0"
412 "\\k is not followed by a braced, angle-bracketed, or quoted name\0"
413 /* 70 */
414 "internal error: unknown opcode in find_fixedlength()\0"
415 ;
416
417 /* Table to identify digits and hex digits. This is used when compiling
418 patterns. Note that the tables in chartables are dependent on the locale, and
419 may mark arbitrary characters as digits - but the PCRE compiling code expects
420 to handle only 0-9, a-z, and A-Z as digits when compiling. That is why we have
421 a private table here. It costs 256 bytes, but it is a lot faster than doing
422 character value tests (at least in some simple cases I timed), and in some
423 applications one wants PCRE to compile efficiently as well as match
424 efficiently.
425
426 For convenience, we use the same bit definitions as in chartables:
427
428 0x04 decimal digit
429 0x08 hexadecimal digit
430
431 Then we can use ctype_digit and ctype_xdigit in the code. */
432
433 #ifndef EBCDIC
434
435 /* This is the "normal" case, for ASCII systems, and EBCDIC systems running in
436 UTF-8 mode. */
437
438 static const unsigned char digitab[] =
439 {
440 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 0- 7 */
441 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 8- 15 */
442 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 16- 23 */
443 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 24- 31 */
444 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - ' */
445 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* ( - / */
446 0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c, /* 0 - 7 */
447 0x0c,0x0c,0x00,0x00,0x00,0x00,0x00,0x00, /* 8 - ? */
448 0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* @ - G */
449 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* H - O */
450 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* P - W */
451 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* X - _ */
452 0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* ` - g */
453 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* h - o */
454 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* p - w */
455 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* x -127 */
456 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 128-135 */
457 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 136-143 */
458 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 144-151 */
459 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 152-159 */
460 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 160-167 */
461 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 168-175 */
462 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 176-183 */
463 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 184-191 */
464 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 192-199 */
465 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 200-207 */
466 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 208-215 */
467 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 216-223 */
468 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 224-231 */
469 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 232-239 */
470 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 240-247 */
471 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00};/* 248-255 */
472
473 #else
474
475 /* This is the "abnormal" case, for EBCDIC systems not running in UTF-8 mode. */
476
477 static const unsigned char digitab[] =
478 {
479 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 0- 7 0 */
480 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 8- 15 */
481 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 16- 23 10 */
482 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 24- 31 */
483 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 32- 39 20 */
484 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 40- 47 */
485 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 48- 55 30 */
486 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 56- 63 */
487 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - 71 40 */
488 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 72- | */
489 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* & - 87 50 */
490 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 88- 95 */
491 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - -103 60 */
492 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 104- ? */
493 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 112-119 70 */
494 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 120- " */
495 0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* 128- g 80 */
496 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* h -143 */
497 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 144- p 90 */
498 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* q -159 */
499 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 160- x A0 */
500 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* y -175 */
501 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* ^ -183 B0 */
502 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 184-191 */
503 0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* { - G C0 */
504 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* H -207 */
505 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* } - P D0 */
506 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* Q -223 */
507 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* \ - X E0 */
508 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* Y -239 */
509 0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c, /* 0 - 7 F0 */
510 0x0c,0x0c,0x00,0x00,0x00,0x00,0x00,0x00};/* 8 -255 */
511
512 static const unsigned char ebcdic_chartab[] = { /* chartable partial dup */
513 0x80,0x00,0x00,0x00,0x00,0x01,0x00,0x00, /* 0- 7 */
514 0x00,0x00,0x00,0x00,0x01,0x01,0x00,0x00, /* 8- 15 */
515 0x00,0x00,0x00,0x00,0x00,0x01,0x00,0x00, /* 16- 23 */
516 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 24- 31 */
517 0x00,0x00,0x00,0x00,0x00,0x01,0x00,0x00, /* 32- 39 */
518 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 40- 47 */
519 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 48- 55 */
520 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 56- 63 */
521 0x01,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - 71 */
522 0x00,0x00,0x00,0x80,0x00,0x80,0x80,0x80, /* 72- | */
523 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* & - 87 */
524 0x00,0x00,0x00,0x80,0x80,0x80,0x00,0x00, /* 88- 95 */
525 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - -103 */
526 0x00,0x00,0x00,0x00,0x00,0x10,0x00,0x80, /* 104- ? */
527 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 112-119 */
528 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 120- " */
529 0x00,0x1a,0x1a,0x1a,0x1a,0x1a,0x1a,0x12, /* 128- g */
530 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* h -143 */
531 0x00,0x12,0x12,0x12,0x12,0x12,0x12,0x12, /* 144- p */
532 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* q -159 */
533 0x00,0x00,0x12,0x12,0x12,0x12,0x12,0x12, /* 160- x */
534 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* y -175 */
535 0x80,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* ^ -183 */
536 0x00,0x00,0x80,0x00,0x00,0x00,0x00,0x00, /* 184-191 */
537 0x80,0x1a,0x1a,0x1a,0x1a,0x1a,0x1a,0x12, /* { - G */
538 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* H -207 */
539 0x00,0x12,0x12,0x12,0x12,0x12,0x12,0x12, /* } - P */
540 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* Q -223 */
541 0x00,0x00,0x12,0x12,0x12,0x12,0x12,0x12, /* \ - X */
542 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* Y -239 */
543 0x1c,0x1c,0x1c,0x1c,0x1c,0x1c,0x1c,0x1c, /* 0 - 7 */
544 0x1c,0x1c,0x00,0x00,0x00,0x00,0x00,0x00};/* 8 -255 */
545 #endif
546
547
548 /* Definition to allow mutual recursion */
549
550 static BOOL
551 compile_regex(int, uschar **, const uschar **, int *, BOOL, BOOL, int, int,
552 int *, int *, branch_chain *, compile_data *, int *);
553
554
555
556 /*************************************************
557 * Find an error text *
558 *************************************************/
559
560 /* The error texts are now all in one long string, to save on relocations. As
561 some of the text is of unknown length, we can't use a table of offsets.
562 Instead, just count through the strings. This is not a performance issue
563 because it happens only when there has been a compilation error.
564
565 Argument: the error number
566 Returns: pointer to the error string
567 */
568
569 static const char *
570 find_error_text(int n)
571 {
572 const char *s = error_texts;
573 for (; n > 0; n--)
574 {
575 while (*s++ != 0) {};
576 if (*s == 0) return "Error text not found (please report)";
577 }
578 return s;
579 }
580
581
582 /*************************************************
583 * Check for counted repeat *
584 *************************************************/
585
586 /* This function is called when a '{' is encountered in a place where it might
587 start a quantifier. It looks ahead to see if it really is a quantifier or not.
588 It is only a quantifier if it is one of the forms {ddd} {ddd,} or {ddd,ddd}
589 where the ddds are digits.
590
591 Arguments:
592 p pointer to the first char after '{'
593
594 Returns: TRUE or FALSE
595 */
596
597 static BOOL
598 is_counted_repeat(const uschar *p)
599 {
600 if ((digitab[*p++] & ctype_digit) == 0) return FALSE;
601 while ((digitab[*p] & ctype_digit) != 0) p++;
602 if (*p == CHAR_RIGHT_CURLY_BRACKET) return TRUE;
603
604 if (*p++ != CHAR_COMMA) return FALSE;
605 if (*p == CHAR_RIGHT_CURLY_BRACKET) return TRUE;
606
607 if ((digitab[*p++] & ctype_digit) == 0) return FALSE;
608 while ((digitab[*p] & ctype_digit) != 0) p++;
609
610 return (*p == CHAR_RIGHT_CURLY_BRACKET);
611 }
612
613
614
615 /*************************************************
616 * Handle escapes *
617 *************************************************/
618
619 /* This function is called when a \ has been encountered. It either returns a
620 positive value for a simple escape such as \n, or a negative value which
621 encodes one of the more complicated things such as \d. A backreference to group
622 n is returned as -(ESC_REF + n); ESC_REF is the highest ESC_xxx macro. When
623 UTF-8 is enabled, a positive value greater than 255 may be returned. On entry,
624 ptr is pointing at the \. On exit, it is on the final character of the escape
625 sequence.
626
627 Arguments:
628 ptrptr points to the pattern position pointer
629 errorcodeptr points to the errorcode variable
630 bracount number of previous extracting brackets
631 options the options bits
632 isclass TRUE if inside a character class
633
634 Returns: zero or positive => a data character
635 negative => a special escape sequence
636 on error, errorcodeptr is set
637 */
638
639 static int
640 check_escape(const uschar **ptrptr, int *errorcodeptr, int bracount,
641 int options, BOOL isclass)
642 {
643 BOOL utf8 = (options & PCRE_UTF8) != 0;
644 const uschar *ptr = *ptrptr + 1;
645 int c, i;
646
647 GETCHARINCTEST(c, ptr); /* Get character value, increment pointer */
648 ptr--; /* Set pointer back to the last byte */
649
650 /* If backslash is at the end of the pattern, it's an error. */
651
652 if (c == 0) *errorcodeptr = ERR1;
653
654 /* Non-alphanumerics are literals. For digits or letters, do an initial lookup
655 in a table. A non-zero result is something that can be returned immediately.
656 Otherwise further processing may be required. */
657
658 #ifndef EBCDIC /* ASCII/UTF-8 coding */
659 else if (c < CHAR_0 || c > CHAR_z) {} /* Not alphanumeric */
660 else if ((i = escapes[c - CHAR_0]) != 0) c = i;
661
662 #else /* EBCDIC coding */
663 else if (c < 'a' || (ebcdic_chartab[c] & 0x0E) == 0) {} /* Not alphanumeric */
664 else if ((i = escapes[c - 0x48]) != 0) c = i;
665 #endif
666
667 /* Escapes that need further processing, or are illegal. */
668
669 else
670 {
671 const uschar *oldptr;
672 BOOL braced, negated;
673
674 switch (c)
675 {
676 /* A number of Perl escapes are not handled by PCRE. We give an explicit
677 error. */
678
679 case CHAR_l:
680 case CHAR_L:
681 *errorcodeptr = ERR37;
682 break;
683
684 case CHAR_u:
685 if ((options & PCRE_JAVASCRIPT_COMPAT) != 0)
686 {
687 /* In JavaScript, \u must be followed by four hexadecimal numbers.
688 Otherwise it is a lowercase u letter. */
689 if ((digitab[ptr[1]] & ctype_xdigit) != 0 && (digitab[ptr[2]] & ctype_xdigit) != 0
690 && (digitab[ptr[3]] & ctype_xdigit) != 0 && (digitab[ptr[4]] & ctype_xdigit) != 0)
691 {
692 c = 0;
693 for (i = 0; i < 4; ++i)
694 {
695 register int cc = *(++ptr);
696 #ifndef EBCDIC /* ASCII/UTF-8 coding */
697 if (cc >= CHAR_a) cc -= 32; /* Convert to upper case */
698 c = (c << 4) + cc - ((cc < CHAR_A)? CHAR_0 : (CHAR_A - 10));
699 #else /* EBCDIC coding */
700 if (cc >= CHAR_a && cc <= CHAR_z) cc += 64; /* Convert to upper case */
701 c = (c << 4) + cc - ((cc >= CHAR_0)? CHAR_0 : (CHAR_A - 10));
702 #endif
703 }
704 }
705 }
706 else
707 *errorcodeptr = ERR37;
708 break;
709
710 case CHAR_U:
711 /* In JavaScript, \U is an uppercase U letter. */
712 if ((options & PCRE_JAVASCRIPT_COMPAT) == 0) *errorcodeptr = ERR37;
713 break;
714
715 /* In a character class, \g is just a literal "g". Outside a character
716 class, \g must be followed by one of a number of specific things:
717
718 (1) A number, either plain or braced. If positive, it is an absolute
719 backreference. If negative, it is a relative backreference. This is a Perl
720 5.10 feature.
721
722 (2) Perl 5.10 also supports \g{name} as a reference to a named group. This
723 is part of Perl's movement towards a unified syntax for back references. As
724 this is synonymous with \k{name}, we fudge it up by pretending it really
725 was \k.
726
727 (3) For Oniguruma compatibility we also support \g followed by a name or a
728 number either in angle brackets or in single quotes. However, these are
729 (possibly recursive) subroutine calls, _not_ backreferences. Just return
730 the -ESC_g code (cf \k). */
731
732 case CHAR_g:
733 if (isclass) break;
734 if (ptr[1] == CHAR_LESS_THAN_SIGN || ptr[1] == CHAR_APOSTROPHE)
735 {
736 c = -ESC_g;
737 break;
738 }
739
740 /* Handle the Perl-compatible cases */
741
742 if (ptr[1] == CHAR_LEFT_CURLY_BRACKET)
743 {
744 const uschar *p;
745 for (p = ptr+2; *p != 0 && *p != CHAR_RIGHT_CURLY_BRACKET; p++)
746 if (*p != CHAR_MINUS && (digitab[*p] & ctype_digit) == 0) break;
747 if (*p != 0 && *p != CHAR_RIGHT_CURLY_BRACKET)
748 {
749 c = -ESC_k;
750 break;
751 }
752 braced = TRUE;
753 ptr++;
754 }
755 else braced = FALSE;
756
757 if (ptr[1] == CHAR_MINUS)
758 {
759 negated = TRUE;
760 ptr++;
761 }
762 else negated = FALSE;
763
764 c = 0;
765 while ((digitab[ptr[1]] & ctype_digit) != 0)
766 c = c * 10 + *(++ptr) - CHAR_0;
767
768 if (c < 0) /* Integer overflow */
769 {
770 *errorcodeptr = ERR61;
771 break;
772 }
773
774 if (braced && *(++ptr) != CHAR_RIGHT_CURLY_BRACKET)
775 {
776 *errorcodeptr = ERR57;
777 break;
778 }
779
780 if (c == 0)
781 {
782 *errorcodeptr = ERR58;
783 break;
784 }
785
786 if (negated)
787 {
788 if (c > bracount)
789 {
790 *errorcodeptr = ERR15;
791 break;
792 }
793 c = bracount - (c - 1);
794 }
795
796 c = -(ESC_REF + c);
797 break;
798
799 /* The handling of escape sequences consisting of a string of digits
800 starting with one that is not zero is not straightforward. By experiment,
801 the way Perl works seems to be as follows:
802
803 Outside a character class, the digits are read as a decimal number. If the
804 number is less than 10, or if there are that many previous extracting
805 left brackets, then it is a back reference. Otherwise, up to three octal
806 digits are read to form an escaped byte. Thus \123 is likely to be octal
807 123 (cf \0123, which is octal 012 followed by the literal 3). If the octal
808 value is greater than 377, the least significant 8 bits are taken. Inside a
809 character class, \ followed by a digit is always an octal number. */
810
811 case CHAR_1: case CHAR_2: case CHAR_3: case CHAR_4: case CHAR_5:
812 case CHAR_6: case CHAR_7: case CHAR_8: case CHAR_9:
813
814 if (!isclass)
815 {
816 oldptr = ptr;
817 c -= CHAR_0;
818 while ((digitab[ptr[1]] & ctype_digit) != 0)
819 c = c * 10 + *(++ptr) - CHAR_0;
820 if (c < 0) /* Integer overflow */
821 {
822 *errorcodeptr = ERR61;
823 break;
824 }
825 if (c < 10 || c <= bracount)
826 {
827 c = -(ESC_REF + c);
828 break;
829 }
830 ptr = oldptr; /* Put the pointer back and fall through */
831 }
832
833 /* Handle an octal number following \. If the first digit is 8 or 9, Perl
834 generates a binary zero byte and treats the digit as a following literal.
835 Thus we have to pull back the pointer by one. */
836
837 if ((c = *ptr) >= CHAR_8)
838 {
839 ptr--;
840 c = 0;
841 break;
842 }
843
844 /* \0 always starts an octal number, but we may drop through to here with a
845 larger first octal digit. The original code used just to take the least
846 significant 8 bits of octal numbers (I think this is what early Perls used
847 to do). Nowadays we allow for larger numbers in UTF-8 mode, but no more
848 than 3 octal digits. */
849
850 case CHAR_0:
851 c -= CHAR_0;
852 while(i++ < 2 && ptr[1] >= CHAR_0 && ptr[1] <= CHAR_7)
853 c = c * 8 + *(++ptr) - CHAR_0;
854 if (!utf8 && c > 255) *errorcodeptr = ERR51;
855 break;
856
857 /* \x is complicated. \x{ddd} is a character number which can be greater
858 than 0xff in utf8 mode, but only if the ddd are hex digits. If not, { is
859 treated as a data character. */
860
861 case CHAR_x:
862 if ((options & PCRE_JAVASCRIPT_COMPAT) != 0)
863 {
864 /* In JavaScript, \x must be followed by two hexadecimal numbers.
865 Otherwise it is a lowercase x letter. */
866 if ((digitab[ptr[1]] & ctype_xdigit) != 0 && (digitab[ptr[2]] & ctype_xdigit) != 0)
867 {
868 c = 0;
869 for (i = 0; i < 2; ++i)
870 {
871 register int cc = *(++ptr);
872 #ifndef EBCDIC /* ASCII/UTF-8 coding */
873 if (cc >= CHAR_a) cc -= 32; /* Convert to upper case */
874 c = (c << 4) + cc - ((cc < CHAR_A)? CHAR_0 : (CHAR_A - 10));
875 #else /* EBCDIC coding */
876 if (cc >= CHAR_a && cc <= CHAR_z) cc += 64; /* Convert to upper case */
877 c = (c << 4) + cc - ((cc >= CHAR_0)? CHAR_0 : (CHAR_A - 10));
878 #endif
879 }
880 }
881 break;
882 }
883
884 if (ptr[1] == CHAR_LEFT_CURLY_BRACKET)
885 {
886 const uschar *pt = ptr + 2;
887 int count = 0;
888
889 c = 0;
890 while ((digitab[*pt] & ctype_xdigit) != 0)
891 {
892 register int cc = *pt++;
893 if (c == 0 && cc == CHAR_0) continue; /* Leading zeroes */
894 count++;
895
896 #ifndef EBCDIC /* ASCII/UTF-8 coding */
897 if (cc >= CHAR_a) cc -= 32; /* Convert to upper case */
898 c = (c << 4) + cc - ((cc < CHAR_A)? CHAR_0 : (CHAR_A - 10));
899 #else /* EBCDIC coding */
900 if (cc >= CHAR_a && cc <= CHAR_z) cc += 64; /* Convert to upper case */
901 c = (c << 4) + cc - ((cc >= CHAR_0)? CHAR_0 : (CHAR_A - 10));
902 #endif
903 }
904
905 if (*pt == CHAR_RIGHT_CURLY_BRACKET)
906 {
907 if (c < 0 || count > (utf8? 8 : 2)) *errorcodeptr = ERR34;
908 ptr = pt;
909 break;
910 }
911
912 /* If the sequence of hex digits does not end with '}', then we don't
913 recognize this construct; fall through to the normal \x handling. */
914 }
915
916 /* Read just a single-byte hex-defined char */
917
918 c = 0;
919 while (i++ < 2 && (digitab[ptr[1]] & ctype_xdigit) != 0)
920 {
921 int cc; /* Some compilers don't like */
922 cc = *(++ptr); /* ++ in initializers */
923 #ifndef EBCDIC /* ASCII/UTF-8 coding */
924 if (cc >= CHAR_a) cc -= 32; /* Convert to upper case */
925 c = c * 16 + cc - ((cc < CHAR_A)? CHAR_0 : (CHAR_A - 10));
926 #else /* EBCDIC coding */
927 if (cc <= CHAR_z) cc += 64; /* Convert to upper case */
928 c = c * 16 + cc - ((cc >= CHAR_0)? CHAR_0 : (CHAR_A - 10));
929 #endif
930 }
931 break;
932
933 /* For \c, a following letter is upper-cased; then the 0x40 bit is flipped.
934 An error is given if the byte following \c is not an ASCII character. This
935 coding is ASCII-specific, but then the whole concept of \cx is
936 ASCII-specific. (However, an EBCDIC equivalent has now been added.) */
937
938 case CHAR_c:
939 c = *(++ptr);
940 if (c == 0)
941 {
942 *errorcodeptr = ERR2;
943 break;
944 }
945 #ifndef EBCDIC /* ASCII/UTF-8 coding */
946 if (c > 127) /* Excludes all non-ASCII in either mode */
947 {
948 *errorcodeptr = ERR68;
949 break;
950 }
951 if (c >= CHAR_a && c <= CHAR_z) c -= 32;
952 c ^= 0x40;
953 #else /* EBCDIC coding */
954 if (c >= CHAR_a && c <= CHAR_z) c += 64;
955 c ^= 0xC0;
956 #endif
957 break;
958
959 /* PCRE_EXTRA enables extensions to Perl in the matter of escapes. Any
960 other alphanumeric following \ is an error if PCRE_EXTRA was set;
961 otherwise, for Perl compatibility, it is a literal. This code looks a bit
962 odd, but there used to be some cases other than the default, and there may
963 be again in future, so I haven't "optimized" it. */
964
965 default:
966 if ((options & PCRE_EXTRA) != 0) switch(c)
967 {
968 default:
969 *errorcodeptr = ERR3;
970 break;
971 }
972 break;
973 }
974 }
975
976 /* Perl supports \N{name} for character names, as well as plain \N for "not
977 newline". PCRE does not support \N{name}. However, it does support
978 quantification such as \N{2,3}. */
979
980 if (c == -ESC_N && ptr[1] == CHAR_LEFT_CURLY_BRACKET &&
981 !is_counted_repeat(ptr+2))
982 *errorcodeptr = ERR37;
983
984 /* If PCRE_UCP is set, we change the values for \d etc. */
985
986 if ((options & PCRE_UCP) != 0 && c <= -ESC_D && c >= -ESC_w)
987 c -= (ESC_DU - ESC_D);
988
989 /* Set the pointer to the final character before returning. */
990
991 *ptrptr = ptr;
992 return c;
993 }
994
995
996
997 #ifdef SUPPORT_UCP
998 /*************************************************
999 * Handle \P and \p *
1000 *************************************************/
1001
1002 /* This function is called after \P or \p has been encountered, provided that
1003 PCRE is compiled with support for Unicode properties. On entry, ptrptr is
1004 pointing at the P or p. On exit, it is pointing at the final character of the
1005 escape sequence.
1006
1007 Argument:
1008 ptrptr points to the pattern position pointer
1009 negptr points to a boolean that is set TRUE for negation else FALSE
1010 dptr points to an int that is set to the detailed property value
1011 errorcodeptr points to the error code variable
1012
1013 Returns: type value from ucp_type_table, or -1 for an invalid type
1014 */
1015
1016 static int
1017 get_ucp(const uschar **ptrptr, BOOL *negptr, int *dptr, int *errorcodeptr)
1018 {
1019 int c, i, bot, top;
1020 const uschar *ptr = *ptrptr;
1021 char name[32];
1022
1023 c = *(++ptr);
1024 if (c == 0) goto ERROR_RETURN;
1025
1026 *negptr = FALSE;
1027
1028 /* \P or \p can be followed by a name in {}, optionally preceded by ^ for
1029 negation. */
1030
1031 if (c == CHAR_LEFT_CURLY_BRACKET)
1032 {
1033 if (ptr[1] == CHAR_CIRCUMFLEX_ACCENT)
1034 {
1035 *negptr = TRUE;
1036 ptr++;
1037 }
1038 for (i = 0; i < (int)sizeof(name) - 1; i++)
1039 {
1040 c = *(++ptr);
1041 if (c == 0) goto ERROR_RETURN;
1042 if (c == CHAR_RIGHT_CURLY_BRACKET) break;
1043 name[i] = c;
1044 }
1045 if (c != CHAR_RIGHT_CURLY_BRACKET) goto ERROR_RETURN;
1046 name[i] = 0;
1047 }
1048
1049 /* Otherwise there is just one following character */
1050
1051 else
1052 {
1053 name[0] = c;
1054 name[1] = 0;
1055 }
1056
1057 *ptrptr = ptr;
1058
1059 /* Search for a recognized property name using binary chop */
1060
1061 bot = 0;
1062 top = _pcre_utt_size;
1063
1064 while (bot < top)
1065 {
1066 i = (bot + top) >> 1;
1067 c = strcmp(name, _pcre_utt_names + _pcre_utt[i].name_offset);
1068 if (c == 0)
1069 {
1070 *dptr = _pcre_utt[i].value;
1071 return _pcre_utt[i].type;
1072 }
1073 if (c > 0) bot = i + 1; else top = i;
1074 }
1075
1076 *errorcodeptr = ERR47;
1077 *ptrptr = ptr;
1078 return -1;
1079
1080 ERROR_RETURN:
1081 *errorcodeptr = ERR46;
1082 *ptrptr = ptr;
1083 return -1;
1084 }
1085 #endif
1086
1087
1088
1089
1090 /*************************************************
1091 * Read repeat counts *
1092 *************************************************/
1093
1094 /* Read an item of the form {n,m} and return the values. This is called only
1095 after is_counted_repeat() has confirmed that a repeat-count quantifier exists,
1096 so the syntax is guaranteed to be correct, but we need to check the values.
1097
1098 Arguments:
1099 p pointer to first char after '{'
1100 minp pointer to int for min
1101 maxp pointer to int for max
1102 returned as -1 if no max
1103 errorcodeptr points to error code variable
1104
1105 Returns: pointer to '}' on success;
1106 current ptr on error, with errorcodeptr set non-zero
1107 */
1108
1109 static const uschar *
1110 read_repeat_counts(const uschar *p, int *minp, int *maxp, int *errorcodeptr)
1111 {
1112 int min = 0;
1113 int max = -1;
1114
1115 /* Read the minimum value and do a paranoid check: a negative value indicates
1116 an integer overflow. */
1117
1118 while ((digitab[*p] & ctype_digit) != 0) min = min * 10 + *p++ - CHAR_0;
1119 if (min < 0 || min > 65535)
1120 {
1121 *errorcodeptr = ERR5;
1122 return p;
1123 }
1124
1125 /* Read the maximum value if there is one, and again do a paranoid on its size.
1126 Also, max must not be less than min. */
1127
1128 if (*p == CHAR_RIGHT_CURLY_BRACKET) max = min; else
1129 {
1130 if (*(++p) != CHAR_RIGHT_CURLY_BRACKET)
1131 {
1132 max = 0;
1133 while((digitab[*p] & ctype_digit) != 0) max = max * 10 + *p++ - CHAR_0;
1134 if (max < 0 || max > 65535)
1135 {
1136 *errorcodeptr = ERR5;
1137 return p;
1138 }
1139 if (max < min)
1140 {
1141 *errorcodeptr = ERR4;
1142 return p;
1143 }
1144 }
1145 }
1146
1147 /* Fill in the required variables, and pass back the pointer to the terminating
1148 '}'. */
1149
1150 *minp = min;
1151 *maxp = max;
1152 return p;
1153 }
1154
1155
1156
1157 /*************************************************
1158 * Subroutine for finding forward reference *
1159 *************************************************/
1160
1161 /* This recursive function is called only from find_parens() below. The
1162 top-level call starts at the beginning of the pattern. All other calls must
1163 start at a parenthesis. It scans along a pattern's text looking for capturing
1164 subpatterns, and counting them. If it finds a named pattern that matches the
1165 name it is given, it returns its number. Alternatively, if the name is NULL, it
1166 returns when it reaches a given numbered subpattern. Recursion is used to keep
1167 track of subpatterns that reset the capturing group numbers - the (?| feature.
1168
1169 This function was originally called only from the second pass, in which we know
1170 that if (?< or (?' or (?P< is encountered, the name will be correctly
1171 terminated because that is checked in the first pass. There is now one call to
1172 this function in the first pass, to check for a recursive back reference by
1173 name (so that we can make the whole group atomic). In this case, we need check
1174 only up to the current position in the pattern, and that is still OK because
1175 and previous occurrences will have been checked. To make this work, the test
1176 for "end of pattern" is a check against cd->end_pattern in the main loop,
1177 instead of looking for a binary zero. This means that the special first-pass
1178 call can adjust cd->end_pattern temporarily. (Checks for binary zero while
1179 processing items within the loop are OK, because afterwards the main loop will
1180 terminate.)
1181
1182 Arguments:
1183 ptrptr address of the current character pointer (updated)
1184 cd compile background data
1185 name name to seek, or NULL if seeking a numbered subpattern
1186 lorn name length, or subpattern number if name is NULL
1187 xmode TRUE if we are in /x mode
1188 utf8 TRUE if we are in UTF-8 mode
1189 count pointer to the current capturing subpattern number (updated)
1190
1191 Returns: the number of the named subpattern, or -1 if not found
1192 */
1193
1194 static int
1195 find_parens_sub(uschar **ptrptr, compile_data *cd, const uschar *name, int lorn,
1196 BOOL xmode, BOOL utf8, int *count)
1197 {
1198 uschar *ptr = *ptrptr;
1199 int start_count = *count;
1200 int hwm_count = start_count;
1201 BOOL dup_parens = FALSE;
1202
1203 /* If the first character is a parenthesis, check on the type of group we are
1204 dealing with. The very first call may not start with a parenthesis. */
1205
1206 if (ptr[0] == CHAR_LEFT_PARENTHESIS)
1207 {
1208 /* Handle specials such as (*SKIP) or (*UTF8) etc. */
1209
1210 if (ptr[1] == CHAR_ASTERISK) ptr += 2;
1211
1212 /* Handle a normal, unnamed capturing parenthesis. */
1213
1214 else if (ptr[1] != CHAR_QUESTION_MARK)
1215 {
1216 *count += 1;
1217 if (name == NULL && *count == lorn) return *count;
1218 ptr++;
1219 }
1220
1221 /* All cases now have (? at the start. Remember when we are in a group
1222 where the parenthesis numbers are duplicated. */
1223
1224 else if (ptr[2] == CHAR_VERTICAL_LINE)
1225 {
1226 ptr += 3;
1227 dup_parens = TRUE;
1228 }
1229
1230 /* Handle comments; all characters are allowed until a ket is reached. */
1231
1232 else if (ptr[2] == CHAR_NUMBER_SIGN)
1233 {
1234 for (ptr += 3; *ptr != 0; ptr++) if (*ptr == CHAR_RIGHT_PARENTHESIS) break;
1235 goto FAIL_EXIT;
1236 }
1237
1238 /* Handle a condition. If it is an assertion, just carry on so that it
1239 is processed as normal. If not, skip to the closing parenthesis of the
1240 condition (there can't be any nested parens). */
1241
1242 else if (ptr[2] == CHAR_LEFT_PARENTHESIS)
1243 {
1244 ptr += 2;
1245 if (ptr[1] != CHAR_QUESTION_MARK)
1246 {
1247 while (*ptr != 0 && *ptr != CHAR_RIGHT_PARENTHESIS) ptr++;
1248 if (*ptr != 0) ptr++;
1249 }
1250 }
1251
1252 /* Start with (? but not a condition. */
1253
1254 else
1255 {
1256 ptr += 2;
1257 if (*ptr == CHAR_P) ptr++; /* Allow optional P */
1258
1259 /* We have to disambiguate (?<! and (?<= from (?<name> for named groups */
1260
1261 if ((*ptr == CHAR_LESS_THAN_SIGN && ptr[1] != CHAR_EXCLAMATION_MARK &&
1262 ptr[1] != CHAR_EQUALS_SIGN) || *ptr == CHAR_APOSTROPHE)
1263 {
1264 int term;
1265 const uschar *thisname;
1266 *count += 1;
1267 if (name == NULL && *count == lorn) return *count;
1268 term = *ptr++;
1269 if (term == CHAR_LESS_THAN_SIGN) term = CHAR_GREATER_THAN_SIGN;
1270 thisname = ptr;
1271 while (*ptr != term) ptr++;
1272 if (name != NULL && lorn == ptr - thisname &&
1273 strncmp((const char *)name, (const char *)thisname, lorn) == 0)
1274 return *count;
1275 term++;
1276 }
1277 }
1278 }
1279
1280 /* Past any initial parenthesis handling, scan for parentheses or vertical
1281 bars. Stop if we get to cd->end_pattern. Note that this is important for the
1282 first-pass call when this value is temporarily adjusted to stop at the current
1283 position. So DO NOT change this to a test for binary zero. */
1284
1285 for (; ptr < cd->end_pattern; ptr++)
1286 {
1287 /* Skip over backslashed characters and also entire \Q...\E */
1288
1289 if (*ptr == CHAR_BACKSLASH)
1290 {
1291 if (*(++ptr) == 0) goto FAIL_EXIT;
1292 if (*ptr == CHAR_Q) for (;;)
1293 {
1294 while (*(++ptr) != 0 && *ptr != CHAR_BACKSLASH) {};
1295 if (*ptr == 0) goto FAIL_EXIT;
1296 if (*(++ptr) == CHAR_E) break;
1297 }
1298 continue;
1299 }
1300
1301 /* Skip over character classes; this logic must be similar to the way they
1302 are handled for real. If the first character is '^', skip it. Also, if the
1303 first few characters (either before or after ^) are \Q\E or \E we skip them
1304 too. This makes for compatibility with Perl. Note the use of STR macros to
1305 encode "Q\\E" so that it works in UTF-8 on EBCDIC platforms. */
1306
1307 if (*ptr == CHAR_LEFT_SQUARE_BRACKET)
1308 {
1309 BOOL negate_class = FALSE;
1310 for (;;)
1311 {
1312 if (ptr[1] == CHAR_BACKSLASH)
1313 {
1314 if (ptr[2] == CHAR_E)
1315 ptr+= 2;
1316 else if (strncmp((const char *)ptr+2,
1317 STR_Q STR_BACKSLASH STR_E, 3) == 0)
1318 ptr += 4;
1319 else
1320 break;
1321 }
1322 else if (!negate_class && ptr[1] == CHAR_CIRCUMFLEX_ACCENT)
1323 {
1324 negate_class = TRUE;
1325 ptr++;
1326 }
1327 else break;
1328 }
1329
1330 /* If the next character is ']', it is a data character that must be
1331 skipped, except in JavaScript compatibility mode. */
1332
1333 if (ptr[1] == CHAR_RIGHT_SQUARE_BRACKET &&
1334 (cd->external_options & PCRE_JAVASCRIPT_COMPAT) == 0)
1335 ptr++;
1336
1337 while (*(++ptr) != CHAR_RIGHT_SQUARE_BRACKET)
1338 {
1339 if (*ptr == 0) return -1;
1340 if (*ptr == CHAR_BACKSLASH)
1341 {
1342 if (*(++ptr) == 0) goto FAIL_EXIT;
1343 if (*ptr == CHAR_Q) for (;;)
1344 {
1345 while (*(++ptr) != 0 && *ptr != CHAR_BACKSLASH) {};
1346 if (*ptr == 0) goto FAIL_EXIT;
1347 if (*(++ptr) == CHAR_E) break;
1348 }
1349 continue;
1350 }
1351 }
1352 continue;
1353 }
1354
1355 /* Skip comments in /x mode */
1356
1357 if (xmode && *ptr == CHAR_NUMBER_SIGN)
1358 {
1359 ptr++;
1360 while (*ptr != 0)
1361 {
1362 if (IS_NEWLINE(ptr)) { ptr += cd->nllen - 1; break; }
1363 ptr++;
1364 #ifdef SUPPORT_UTF8
1365 if (utf8) while ((*ptr & 0xc0) == 0x80) ptr++;
1366 #endif
1367 }
1368 if (*ptr == 0) goto FAIL_EXIT;
1369 continue;
1370 }
1371
1372 /* Check for the special metacharacters */
1373
1374 if (*ptr == CHAR_LEFT_PARENTHESIS)
1375 {
1376 int rc = find_parens_sub(&ptr, cd, name, lorn, xmode, utf8, count);
1377 if (rc > 0) return rc;
1378 if (*ptr == 0) goto FAIL_EXIT;
1379 }
1380
1381 else if (*ptr == CHAR_RIGHT_PARENTHESIS)
1382 {
1383 if (dup_parens && *count < hwm_count) *count = hwm_count;
1384 goto FAIL_EXIT;
1385 }
1386
1387 else if (*ptr == CHAR_VERTICAL_LINE && dup_parens)
1388 {
1389 if (*count > hwm_count) hwm_count = *count;
1390 *count = start_count;
1391 }
1392 }
1393
1394 FAIL_EXIT:
1395 *ptrptr = ptr;
1396 return -1;
1397 }
1398
1399
1400
1401
1402 /*************************************************
1403 * Find forward referenced subpattern *
1404 *************************************************/
1405
1406 /* This function scans along a pattern's text looking for capturing
1407 subpatterns, and counting them. If it finds a named pattern that matches the
1408 name it is given, it returns its number. Alternatively, if the name is NULL, it
1409 returns when it reaches a given numbered subpattern. This is used for forward
1410 references to subpatterns. We used to be able to start this scan from the
1411 current compiling point, using the current count value from cd->bracount, and
1412 do it all in a single loop, but the addition of the possibility of duplicate
1413 subpattern numbers means that we have to scan from the very start, in order to
1414 take account of such duplicates, and to use a recursive function to keep track
1415 of the different types of group.
1416
1417 Arguments:
1418 cd compile background data
1419 name name to seek, or NULL if seeking a numbered subpattern
1420 lorn name length, or subpattern number if name is NULL
1421 xmode TRUE if we are in /x mode
1422 utf8 TRUE if we are in UTF-8 mode
1423
1424 Returns: the number of the found subpattern, or -1 if not found
1425 */
1426
1427 static int
1428 find_parens(compile_data *cd, const uschar *name, int lorn, BOOL xmode,
1429 BOOL utf8)
1430 {
1431 uschar *ptr = (uschar *)cd->start_pattern;
1432 int count = 0;
1433 int rc;
1434
1435 /* If the pattern does not start with an opening parenthesis, the first call
1436 to find_parens_sub() will scan right to the end (if necessary). However, if it
1437 does start with a parenthesis, find_parens_sub() will return when it hits the
1438 matching closing parens. That is why we have to have a loop. */
1439
1440 for (;;)
1441 {
1442 rc = find_parens_sub(&ptr, cd, name, lorn, xmode, utf8, &count);
1443 if (rc > 0 || *ptr++ == 0) break;
1444 }
1445
1446 return rc;
1447 }
1448
1449
1450
1451
1452 /*************************************************
1453 * Find first significant op code *
1454 *************************************************/
1455
1456 /* This is called by several functions that scan a compiled expression looking
1457 for a fixed first character, or an anchoring op code etc. It skips over things
1458 that do not influence this. For some calls, it makes sense to skip negative
1459 forward and all backward assertions, and also the \b assertion; for others it
1460 does not.
1461
1462 Arguments:
1463 code pointer to the start of the group
1464 skipassert TRUE if certain assertions are to be skipped
1465
1466 Returns: pointer to the first significant opcode
1467 */
1468
1469 static const uschar*
1470 first_significant_code(const uschar *code, BOOL skipassert)
1471 {
1472 for (;;)
1473 {
1474 switch ((int)*code)
1475 {
1476 case OP_ASSERT_NOT:
1477 case OP_ASSERTBACK:
1478 case OP_ASSERTBACK_NOT:
1479 if (!skipassert) return code;
1480 do code += GET(code, 1); while (*code == OP_ALT);
1481 code += _pcre_OP_lengths[*code];
1482 break;
1483
1484 case OP_WORD_BOUNDARY:
1485 case OP_NOT_WORD_BOUNDARY:
1486 if (!skipassert) return code;
1487 /* Fall through */
1488
1489 case OP_CALLOUT:
1490 case OP_CREF:
1491 case OP_NCREF:
1492 case OP_RREF:
1493 case OP_NRREF:
1494 case OP_DEF:
1495 code += _pcre_OP_lengths[*code];
1496 break;
1497
1498 default:
1499 return code;
1500 }
1501 }
1502 /* Control never reaches here */
1503 }
1504
1505
1506
1507
1508 /*************************************************
1509 * Find the fixed length of a branch *
1510 *************************************************/
1511
1512 /* Scan a branch and compute the fixed length of subject that will match it,
1513 if the length is fixed. This is needed for dealing with backward assertions.
1514 In UTF8 mode, the result is in characters rather than bytes. The branch is
1515 temporarily terminated with OP_END when this function is called.
1516
1517 This function is called when a backward assertion is encountered, so that if it
1518 fails, the error message can point to the correct place in the pattern.
1519 However, we cannot do this when the assertion contains subroutine calls,
1520 because they can be forward references. We solve this by remembering this case
1521 and doing the check at the end; a flag specifies which mode we are running in.
1522
1523 Arguments:
1524 code points to the start of the pattern (the bracket)
1525 utf8 TRUE in UTF-8 mode
1526 atend TRUE if called when the pattern is complete
1527 cd the "compile data" structure
1528
1529 Returns: the fixed length,
1530 or -1 if there is no fixed length,
1531 or -2 if \C was encountered
1532 or -3 if an OP_RECURSE item was encountered and atend is FALSE
1533 or -4 if an unknown opcode was encountered (internal error)
1534 */
1535
1536 static int
1537 find_fixedlength(uschar *code, BOOL utf8, BOOL atend, compile_data *cd)
1538 {
1539 int length = -1;
1540
1541 register int branchlength = 0;
1542 register uschar *cc = code + 1 + LINK_SIZE;
1543
1544 /* Scan along the opcodes for this branch. If we get to the end of the
1545 branch, check the length against that of the other branches. */
1546
1547 for (;;)
1548 {
1549 int d;
1550 uschar *ce, *cs;
1551 register int op = *cc;
1552 switch (op)
1553 {
1554 /* We only need to continue for OP_CBRA (normal capturing bracket) and
1555 OP_BRA (normal non-capturing bracket) because the other variants of these
1556 opcodes are all concerned with unlimited repeated groups, which of course
1557 are not of fixed length. */
1558
1559 case OP_CBRA:
1560 case OP_BRA:
1561 case OP_ONCE:
1562 case OP_ONCE_NC:
1563 case OP_COND:
1564 d = find_fixedlength(cc + ((op == OP_CBRA)? 2:0), utf8, atend, cd);
1565 if (d < 0) return d;
1566 branchlength += d;
1567 do cc += GET(cc, 1); while (*cc == OP_ALT);
1568 cc += 1 + LINK_SIZE;
1569 break;
1570
1571 /* Reached end of a branch; if it's a ket it is the end of a nested call.
1572 If it's ALT it is an alternation in a nested call. An ACCEPT is effectively
1573 an ALT. If it is END it's the end of the outer call. All can be handled by
1574 the same code. Note that we must not include the OP_KETRxxx opcodes here,
1575 because they all imply an unlimited repeat. */
1576
1577 case OP_ALT:
1578 case OP_KET:
1579 case OP_END:
1580 case OP_ACCEPT:
1581 case OP_ASSERT_ACCEPT:
1582 if (length < 0) length = branchlength;
1583 else if (length != branchlength) return -1;
1584 if (*cc != OP_ALT) return length;
1585 cc += 1 + LINK_SIZE;
1586 branchlength = 0;
1587 break;
1588
1589 /* A true recursion implies not fixed length, but a subroutine call may
1590 be OK. If the subroutine is a forward reference, we can't deal with
1591 it until the end of the pattern, so return -3. */
1592
1593 case OP_RECURSE:
1594 if (!atend) return -3;
1595 cs = ce = (uschar *)cd->start_code + GET(cc, 1); /* Start subpattern */
1596 do ce += GET(ce, 1); while (*ce == OP_ALT); /* End subpattern */
1597 if (cc > cs && cc < ce) return -1; /* Recursion */
1598 d = find_fixedlength(cs + 2, utf8, atend, cd);
1599 if (d < 0) return d;
1600 branchlength += d;
1601 cc += 1 + LINK_SIZE;
1602 break;
1603
1604 /* Skip over assertive subpatterns */
1605
1606 case OP_ASSERT:
1607 case OP_ASSERT_NOT:
1608 case OP_ASSERTBACK:
1609 case OP_ASSERTBACK_NOT:
1610 do cc += GET(cc, 1); while (*cc == OP_ALT);
1611 /* Fall through */
1612
1613 /* Skip over things that don't match chars */
1614
1615 case OP_MARK:
1616 case OP_PRUNE_ARG:
1617 case OP_SKIP_ARG:
1618 case OP_THEN_ARG:
1619 cc += cc[1] + _pcre_OP_lengths[*cc];
1620 break;
1621
1622 case OP_CALLOUT:
1623 case OP_CIRC:
1624 case OP_CIRCM:
1625 case OP_CLOSE:
1626 case OP_COMMIT:
1627 case OP_CREF:
1628 case OP_DEF:
1629 case OP_DOLL:
1630 case OP_DOLLM:
1631 case OP_EOD:
1632 case OP_EODN:
1633 case OP_FAIL:
1634 case OP_NCREF:
1635 case OP_NRREF:
1636 case OP_NOT_WORD_BOUNDARY:
1637 case OP_PRUNE:
1638 case OP_REVERSE:
1639 case OP_RREF:
1640 case OP_SET_SOM:
1641 case OP_SKIP:
1642 case OP_SOD:
1643 case OP_SOM:
1644 case OP_THEN:
1645 case OP_WORD_BOUNDARY:
1646 cc += _pcre_OP_lengths[*cc];
1647 break;
1648
1649 /* Handle literal characters */
1650
1651 case OP_CHAR:
1652 case OP_CHARI:
1653 case OP_NOT:
1654 case OP_NOTI:
1655 branchlength++;
1656 cc += 2;
1657 #ifdef SUPPORT_UTF8
1658 if (utf8 && cc[-1] >= 0xc0) cc += _pcre_utf8_table4[cc[-1] & 0x3f];
1659 #endif
1660 break;
1661
1662 /* Handle exact repetitions. The count is already in characters, but we
1663 need to skip over a multibyte character in UTF8 mode. */
1664
1665 case OP_EXACT:
1666 case OP_EXACTI:
1667 case OP_NOTEXACT:
1668 case OP_NOTEXACTI:
1669 branchlength += GET2(cc,1);
1670 cc += 4;
1671 #ifdef SUPPORT_UTF8
1672 if (utf8 && cc[-1] >= 0xc0) cc += _pcre_utf8_table4[cc[-1] & 0x3f];
1673 #endif
1674 break;
1675
1676 case OP_TYPEEXACT:
1677 branchlength += GET2(cc,1);
1678 if (cc[3] == OP_PROP || cc[3] == OP_NOTPROP) cc += 2;
1679 cc += 4;
1680 break;
1681
1682 /* Handle single-char matchers */
1683
1684 case OP_PROP:
1685 case OP_NOTPROP:
1686 cc += 2;
1687 /* Fall through */
1688
1689 case OP_HSPACE:
1690 case OP_VSPACE:
1691 case OP_NOT_HSPACE:
1692 case OP_NOT_VSPACE:
1693 case OP_NOT_DIGIT:
1694 case OP_DIGIT:
1695 case OP_NOT_WHITESPACE:
1696 case OP_WHITESPACE:
1697 case OP_NOT_WORDCHAR:
1698 case OP_WORDCHAR:
1699 case OP_ANY:
1700 case OP_ALLANY:
1701 branchlength++;
1702 cc++;
1703 break;
1704
1705 /* The single-byte matcher isn't allowed */
1706
1707 case OP_ANYBYTE:
1708 return -2;
1709
1710 /* Check a class for variable quantification */
1711
1712 #ifdef SUPPORT_UTF8
1713 case OP_XCLASS:
1714 cc += GET(cc, 1) - 33;
1715 /* Fall through */
1716 #endif
1717
1718 case OP_CLASS:
1719 case OP_NCLASS:
1720 cc += 33;
1721
1722 switch (*cc)
1723 {
1724 case OP_CRPLUS:
1725 case OP_CRMINPLUS:
1726 case OP_CRSTAR:
1727 case OP_CRMINSTAR:
1728 case OP_CRQUERY:
1729 case OP_CRMINQUERY:
1730 return -1;
1731
1732 case OP_CRRANGE:
1733 case OP_CRMINRANGE:
1734 if (GET2(cc,1) != GET2(cc,3)) return -1;
1735 branchlength += GET2(cc,1);
1736 cc += 5;
1737 break;
1738
1739 default:
1740 branchlength++;
1741 }
1742 break;
1743
1744 /* Anything else is variable length */
1745
1746 case OP_ANYNL:
1747 case OP_BRAMINZERO:
1748 case OP_BRAPOS:
1749 case OP_BRAPOSZERO:
1750 case OP_BRAZERO:
1751 case OP_CBRAPOS:
1752 case OP_EXTUNI:
1753 case OP_KETRMAX:
1754 case OP_KETRMIN:
1755 case OP_KETRPOS:
1756 case OP_MINPLUS:
1757 case OP_MINPLUSI:
1758 case OP_MINQUERY:
1759 case OP_MINQUERYI:
1760 case OP_MINSTAR:
1761 case OP_MINSTARI:
1762 case OP_MINUPTO:
1763 case OP_MINUPTOI:
1764 case OP_NOTMINPLUS:
1765 case OP_NOTMINPLUSI:
1766 case OP_NOTMINQUERY:
1767 case OP_NOTMINQUERYI:
1768 case OP_NOTMINSTAR:
1769 case OP_NOTMINSTARI:
1770 case OP_NOTMINUPTO:
1771 case OP_NOTMINUPTOI:
1772 case OP_NOTPLUS:
1773 case OP_NOTPLUSI:
1774 case OP_NOTPOSPLUS:
1775 case OP_NOTPOSPLUSI:
1776 case OP_NOTPOSQUERY:
1777 case OP_NOTPOSQUERYI:
1778 case OP_NOTPOSSTAR:
1779 case OP_NOTPOSSTARI:
1780 case OP_NOTPOSUPTO:
1781 case OP_NOTPOSUPTOI:
1782 case OP_NOTQUERY:
1783 case OP_NOTQUERYI:
1784 case OP_NOTSTAR:
1785 case OP_NOTSTARI:
1786 case OP_NOTUPTO:
1787 case OP_NOTUPTOI:
1788 case OP_PLUS:
1789 case OP_PLUSI:
1790 case OP_POSPLUS:
1791 case OP_POSPLUSI:
1792 case OP_POSQUERY:
1793 case OP_POSQUERYI:
1794 case OP_POSSTAR:
1795 case OP_POSSTARI:
1796 case OP_POSUPTO:
1797 case OP_POSUPTOI:
1798 case OP_QUERY:
1799 case OP_QUERYI:
1800 case OP_REF:
1801 case OP_REFI:
1802 case OP_SBRA:
1803 case OP_SBRAPOS:
1804 case OP_SCBRA:
1805 case OP_SCBRAPOS:
1806 case OP_SCOND:
1807 case OP_SKIPZERO:
1808 case OP_STAR:
1809 case OP_STARI:
1810 case OP_TYPEMINPLUS:
1811 case OP_TYPEMINQUERY:
1812 case OP_TYPEMINSTAR:
1813 case OP_TYPEMINUPTO:
1814 case OP_TYPEPLUS:
1815 case OP_TYPEPOSPLUS:
1816 case OP_TYPEPOSQUERY:
1817 case OP_TYPEPOSSTAR:
1818 case OP_TYPEPOSUPTO:
1819 case OP_TYPEQUERY:
1820 case OP_TYPESTAR:
1821 case OP_TYPEUPTO:
1822 case OP_UPTO:
1823 case OP_UPTOI:
1824 return -1;
1825
1826 /* Catch unrecognized opcodes so that when new ones are added they
1827 are not forgotten, as has happened in the past. */
1828
1829 default:
1830 return -4;
1831 }
1832 }
1833 /* Control never gets here */
1834 }
1835
1836
1837
1838
1839 /*************************************************
1840 * Scan compiled regex for specific bracket *
1841 *************************************************/
1842
1843 /* This little function scans through a compiled pattern until it finds a
1844 capturing bracket with the given number, or, if the number is negative, an
1845 instance of OP_REVERSE for a lookbehind. The function is global in the C sense
1846 so that it can be called from pcre_study() when finding the minimum matching
1847 length.
1848
1849 Arguments:
1850 code points to start of expression
1851 utf8 TRUE in UTF-8 mode
1852 number the required bracket number or negative to find a lookbehind
1853
1854 Returns: pointer to the opcode for the bracket, or NULL if not found
1855 */
1856
1857 const uschar *
1858 _pcre_find_bracket(const uschar *code, BOOL utf8, int number)
1859 {
1860 for (;;)
1861 {
1862 register int c = *code;
1863
1864 if (c == OP_END) return NULL;
1865
1866 /* XCLASS is used for classes that cannot be represented just by a bit
1867 map. This includes negated single high-valued characters. The length in
1868 the table is zero; the actual length is stored in the compiled code. */
1869
1870 if (c == OP_XCLASS) code += GET(code, 1);
1871
1872 /* Handle recursion */
1873
1874 else if (c == OP_REVERSE)
1875 {
1876 if (number < 0) return (uschar *)code;
1877 code += _pcre_OP_lengths[c];
1878 }
1879
1880 /* Handle capturing bracket */
1881
1882 else if (c == OP_CBRA || c == OP_SCBRA ||
1883 c == OP_CBRAPOS || c == OP_SCBRAPOS)
1884 {
1885 int n = GET2(code, 1+LINK_SIZE);
1886 if (n == number) return (uschar *)code;
1887 code += _pcre_OP_lengths[c];
1888 }
1889
1890 /* Otherwise, we can get the item's length from the table, except that for
1891 repeated character types, we have to test for \p and \P, which have an extra
1892 two bytes of parameters, and for MARK/PRUNE/SKIP/THEN with an argument, we
1893 must add in its length. */
1894
1895 else
1896 {
1897 switch(c)
1898 {
1899 case OP_TYPESTAR:
1900 case OP_TYPEMINSTAR:
1901 case OP_TYPEPLUS:
1902 case OP_TYPEMINPLUS:
1903 case OP_TYPEQUERY:
1904 case OP_TYPEMINQUERY:
1905 case OP_TYPEPOSSTAR:
1906 case OP_TYPEPOSPLUS:
1907 case OP_TYPEPOSQUERY:
1908 if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
1909 break;
1910
1911 case OP_TYPEUPTO:
1912 case OP_TYPEMINUPTO:
1913 case OP_TYPEEXACT:
1914 case OP_TYPEPOSUPTO:
1915 if (code[3] == OP_PROP || code[3] == OP_NOTPROP) code += 2;
1916 break;
1917
1918 case OP_MARK:
1919 case OP_PRUNE_ARG:
1920 case OP_SKIP_ARG:
1921 code += code[1];
1922 break;
1923
1924 case OP_THEN_ARG:
1925 code += code[1];
1926 break;
1927 }
1928
1929 /* Add in the fixed length from the table */
1930
1931 code += _pcre_OP_lengths[c];
1932
1933 /* In UTF-8 mode, opcodes that are followed by a character may be followed by
1934 a multi-byte character. The length in the table is a minimum, so we have to
1935 arrange to skip the extra bytes. */
1936
1937 #ifdef SUPPORT_UTF8
1938 if (utf8) switch(c)
1939 {
1940 case OP_CHAR:
1941 case OP_CHARI:
1942 case OP_EXACT:
1943 case OP_EXACTI:
1944 case OP_UPTO:
1945 case OP_UPTOI:
1946 case OP_MINUPTO:
1947 case OP_MINUPTOI:
1948 case OP_POSUPTO:
1949 case OP_POSUPTOI:
1950 case OP_STAR:
1951 case OP_STARI:
1952 case OP_MINSTAR:
1953 case OP_MINSTARI:
1954 case OP_POSSTAR:
1955 case OP_POSSTARI:
1956 case OP_PLUS:
1957 case OP_PLUSI:
1958 case OP_MINPLUS:
1959 case OP_MINPLUSI:
1960 case OP_POSPLUS:
1961 case OP_POSPLUSI:
1962 case OP_QUERY:
1963 case OP_QUERYI:
1964 case OP_MINQUERY:
1965 case OP_MINQUERYI:
1966 case OP_POSQUERY:
1967 case OP_POSQUERYI:
1968 if (code[-1] >= 0xc0) code += _pcre_utf8_table4[code[-1] & 0x3f];
1969 break;
1970 }
1971 #else
1972 (void)(utf8); /* Keep compiler happy by referencing function argument */
1973 #endif
1974 }
1975 }
1976 }
1977
1978
1979
1980 /*************************************************
1981 * Scan compiled regex for recursion reference *
1982 *************************************************/
1983
1984 /* This little function scans through a compiled pattern until it finds an
1985 instance of OP_RECURSE.
1986
1987 Arguments:
1988 code points to start of expression
1989 utf8 TRUE in UTF-8 mode
1990
1991 Returns: pointer to the opcode for OP_RECURSE, or NULL if not found
1992 */
1993
1994 static const uschar *
1995 find_recurse(const uschar *code, BOOL utf8)
1996 {
1997 for (;;)
1998 {
1999 register int c = *code;
2000 if (c == OP_END) return NULL;
2001 if (c == OP_RECURSE) return code;
2002
2003 /* XCLASS is used for classes that cannot be represented just by a bit
2004 map. This includes negated single high-valued characters. The length in
2005 the table is zero; the actual length is stored in the compiled code. */
2006
2007 if (c == OP_XCLASS) code += GET(code, 1);
2008
2009 /* Otherwise, we can get the item's length from the table, except that for
2010 repeated character types, we have to test for \p and \P, which have an extra
2011 two bytes of parameters, and for MARK/PRUNE/SKIP/THEN with an argument, we
2012 must add in its length. */
2013
2014 else
2015 {
2016 switch(c)
2017 {
2018 case OP_TYPESTAR:
2019 case OP_TYPEMINSTAR:
2020 case OP_TYPEPLUS:
2021 case OP_TYPEMINPLUS:
2022 case OP_TYPEQUERY:
2023 case OP_TYPEMINQUERY:
2024 case OP_TYPEPOSSTAR:
2025 case OP_TYPEPOSPLUS:
2026 case OP_TYPEPOSQUERY:
2027 if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
2028 break;
2029
2030 case OP_TYPEPOSUPTO:
2031 case OP_TYPEUPTO:
2032 case OP_TYPEMINUPTO:
2033 case OP_TYPEEXACT:
2034 if (code[3] == OP_PROP || code[3] == OP_NOTPROP) code += 2;
2035 break;
2036
2037 case OP_MARK:
2038 case OP_PRUNE_ARG:
2039 case OP_SKIP_ARG:
2040 code += code[1];
2041 break;
2042
2043 case OP_THEN_ARG:
2044 code += code[1];
2045 break;
2046 }
2047
2048 /* Add in the fixed length from the table */
2049
2050 code += _pcre_OP_lengths[c];
2051
2052 /* In UTF-8 mode, opcodes that are followed by a character may be followed
2053 by a multi-byte character. The length in the table is a minimum, so we have
2054 to arrange to skip the extra bytes. */
2055
2056 #ifdef SUPPORT_UTF8
2057 if (utf8) switch(c)
2058 {
2059 case OP_CHAR:
2060 case OP_CHARI:
2061 case OP_EXACT:
2062 case OP_EXACTI:
2063 case OP_UPTO:
2064 case OP_UPTOI:
2065 case OP_MINUPTO:
2066 case OP_MINUPTOI:
2067 case OP_POSUPTO:
2068 case OP_POSUPTOI:
2069 case OP_STAR:
2070 case OP_STARI:
2071 case OP_MINSTAR:
2072 case OP_MINSTARI:
2073 case OP_POSSTAR:
2074 case OP_POSSTARI:
2075 case OP_PLUS:
2076 case OP_PLUSI:
2077 case OP_MINPLUS:
2078 case OP_MINPLUSI:
2079 case OP_POSPLUS:
2080 case OP_POSPLUSI:
2081 case OP_QUERY:
2082 case OP_QUERYI:
2083 case OP_MINQUERY:
2084 case OP_MINQUERYI:
2085 case OP_POSQUERY:
2086 case OP_POSQUERYI:
2087 if (code[-1] >= 0xc0) code += _pcre_utf8_table4[code[-1] & 0x3f];
2088 break;
2089 }
2090 #else
2091 (void)(utf8); /* Keep compiler happy by referencing function argument */
2092 #endif
2093 }
2094 }
2095 }
2096
2097
2098
2099 /*************************************************
2100 * Scan compiled branch for non-emptiness *
2101 *************************************************/
2102
2103 /* This function scans through a branch of a compiled pattern to see whether it
2104 can match the empty string or not. It is called from could_be_empty()
2105 below and from compile_branch() when checking for an unlimited repeat of a
2106 group that can match nothing. Note that first_significant_code() skips over
2107 backward and negative forward assertions when its final argument is TRUE. If we
2108 hit an unclosed bracket, we return "empty" - this means we've struck an inner
2109 bracket whose current branch will already have been scanned.
2110
2111 Arguments:
2112 code points to start of search
2113 endcode points to where to stop
2114 utf8 TRUE if in UTF8 mode
2115 cd contains pointers to tables etc.
2116
2117 Returns: TRUE if what is matched could be empty
2118 */
2119
2120 static BOOL
2121 could_be_empty_branch(const uschar *code, const uschar *endcode, BOOL utf8,
2122 compile_data *cd)
2123 {
2124 register int c;
2125 for (code = first_significant_code(code + _pcre_OP_lengths[*code], TRUE);
2126 code < endcode;
2127 code = first_significant_code(code + _pcre_OP_lengths[c], TRUE))
2128 {
2129 const uschar *ccode;
2130
2131 c = *code;
2132
2133 /* Skip over forward assertions; the other assertions are skipped by
2134 first_significant_code() with a TRUE final argument. */
2135
2136 if (c == OP_ASSERT)
2137 {
2138 do code += GET(code, 1); while (*code == OP_ALT);
2139 c = *code;
2140 continue;
2141 }
2142
2143 /* For a recursion/subroutine call, if its end has been reached, which
2144 implies a backward reference subroutine call, we can scan it. If it's a
2145 forward reference subroutine call, we can't. To detect forward reference
2146 we have to scan up the list that is kept in the workspace. This function is
2147 called only when doing the real compile, not during the pre-compile that
2148 measures the size of the compiled pattern. */
2149
2150 if (c == OP_RECURSE)
2151 {
2152 const uschar *scode;
2153 BOOL empty_branch;
2154
2155 /* Test for forward reference */
2156
2157 for (scode = cd->start_workspace; scode < cd->hwm; scode += LINK_SIZE)
2158 if (GET(scode, 0) == code + 1 - cd->start_code) return TRUE;
2159
2160 /* Not a forward reference, test for completed backward reference */
2161
2162 empty_branch = FALSE;
2163 scode = cd->start_code + GET(code, 1);
2164 if (GET(scode, 1) == 0) return TRUE; /* Unclosed */
2165
2166 /* Completed backwards reference */
2167
2168 do
2169 {
2170 if (could_be_empty_branch(scode, endcode, utf8, cd))
2171 {
2172 empty_branch = TRUE;
2173 break;
2174 }
2175 scode += GET(scode, 1);
2176 }
2177 while (*scode == OP_ALT);
2178
2179 if (!empty_branch) return FALSE; /* All branches are non-empty */
2180 continue;
2181 }
2182
2183 /* Groups with zero repeats can of course be empty; skip them. */
2184
2185 if (c == OP_BRAZERO || c == OP_BRAMINZERO || c == OP_SKIPZERO ||
2186 c == OP_BRAPOSZERO)
2187 {
2188 code += _pcre_OP_lengths[c];
2189 do code += GET(code, 1); while (*code == OP_ALT);
2190 c = *code;
2191 continue;
2192 }
2193
2194 /* A nested group that is already marked as "could be empty" can just be
2195 skipped. */
2196
2197 if (c == OP_SBRA || c == OP_SBRAPOS ||
2198 c == OP_SCBRA || c == OP_SCBRAPOS)
2199 {
2200 do code += GET(code, 1); while (*code == OP_ALT);
2201 c = *code;
2202 continue;
2203 }
2204
2205 /* For other groups, scan the branches. */
2206
2207 if (c == OP_BRA || c == OP_BRAPOS ||
2208 c == OP_CBRA || c == OP_CBRAPOS ||
2209 c == OP_ONCE || c == OP_ONCE_NC ||
2210 c == OP_COND)
2211 {
2212 BOOL empty_branch;
2213 if (GET(code, 1) == 0) return TRUE; /* Hit unclosed bracket */
2214
2215 /* If a conditional group has only one branch, there is a second, implied,
2216 empty branch, so just skip over the conditional, because it could be empty.
2217 Otherwise, scan the individual branches of the group. */
2218
2219 if (c == OP_COND && code[GET(code, 1)] != OP_ALT)
2220 code += GET(code, 1);
2221 else
2222 {
2223 empty_branch = FALSE;
2224 do
2225 {
2226 if (!empty_branch && could_be_empty_branch(code, endcode, utf8, cd))
2227 empty_branch = TRUE;
2228 code += GET(code, 1);
2229 }
2230 while (*code == OP_ALT);
2231 if (!empty_branch) return FALSE; /* All branches are non-empty */
2232 }
2233
2234 c = *code;
2235 continue;
2236 }
2237
2238 /* Handle the other opcodes */
2239
2240 switch (c)
2241 {
2242 /* Check for quantifiers after a class. XCLASS is used for classes that
2243 cannot be represented just by a bit map. This includes negated single
2244 high-valued characters. The length in _pcre_OP_lengths[] is zero; the
2245 actual length is stored in the compiled code, so we must update "code"
2246 here. */
2247
2248 #ifdef SUPPORT_UTF8
2249 case OP_XCLASS:
2250 ccode = code += GET(code, 1);
2251 goto CHECK_CLASS_REPEAT;
2252 #endif
2253
2254 case OP_CLASS:
2255 case OP_NCLASS:
2256 ccode = code + 33;
2257
2258 #ifdef SUPPORT_UTF8
2259 CHECK_CLASS_REPEAT:
2260 #endif
2261
2262 switch (*ccode)
2263 {
2264 case OP_CRSTAR: /* These could be empty; continue */
2265 case OP_CRMINSTAR:
2266 case OP_CRQUERY:
2267 case OP_CRMINQUERY:
2268 break;
2269
2270 default: /* Non-repeat => class must match */
2271 case OP_CRPLUS: /* These repeats aren't empty */
2272 case OP_CRMINPLUS:
2273 return FALSE;
2274
2275 case OP_CRRANGE:
2276 case OP_CRMINRANGE:
2277 if (GET2(ccode, 1) > 0) return FALSE; /* Minimum > 0 */
2278 break;
2279 }
2280 break;
2281
2282 /* Opcodes that must match a character */
2283
2284 case OP_PROP:
2285 case OP_NOTPROP:
2286 case OP_EXTUNI:
2287 case OP_NOT_DIGIT:
2288 case OP_DIGIT:
2289 case OP_NOT_WHITESPACE:
2290 case OP_WHITESPACE:
2291 case OP_NOT_WORDCHAR:
2292 case OP_WORDCHAR:
2293 case OP_ANY:
2294 case OP_ALLANY:
2295 case OP_ANYBYTE:
2296 case OP_CHAR:
2297 case OP_CHARI:
2298 case OP_NOT:
2299 case OP_NOTI:
2300 case OP_PLUS:
2301 case OP_MINPLUS:
2302 case OP_POSPLUS:
2303 case OP_EXACT:
2304 case OP_NOTPLUS:
2305 case OP_NOTMINPLUS:
2306 case OP_NOTPOSPLUS:
2307 case OP_NOTEXACT:
2308 case OP_TYPEPLUS:
2309 case OP_TYPEMINPLUS:
2310 case OP_TYPEPOSPLUS:
2311 case OP_TYPEEXACT:
2312 return FALSE;
2313
2314 /* These are going to continue, as they may be empty, but we have to
2315 fudge the length for the \p and \P cases. */
2316
2317 case OP_TYPESTAR:
2318 case OP_TYPEMINSTAR:
2319 case OP_TYPEPOSSTAR:
2320 case OP_TYPEQUERY:
2321 case OP_TYPEMINQUERY:
2322 case OP_TYPEPOSQUERY:
2323 if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
2324 break;
2325
2326 /* Same for these */
2327
2328 case OP_TYPEUPTO:
2329 case OP_TYPEMINUPTO:
2330 case OP_TYPEPOSUPTO:
2331 if (code[3] == OP_PROP || code[3] == OP_NOTPROP) code += 2;
2332 break;
2333
2334 /* End of branch */
2335
2336 case OP_KET:
2337 case OP_KETRMAX:
2338 case OP_KETRMIN:
2339 case OP_KETRPOS:
2340 case OP_ALT:
2341 return TRUE;
2342
2343 /* In UTF-8 mode, STAR, MINSTAR, POSSTAR, QUERY, MINQUERY, POSQUERY, UPTO,
2344 MINUPTO, and POSUPTO may be followed by a multibyte character */
2345
2346 #ifdef SUPPORT_UTF8
2347 case OP_STAR:
2348 case OP_STARI:
2349 case OP_MINSTAR:
2350 case OP_MINSTARI:
2351 case OP_POSSTAR:
2352 case OP_POSSTARI:
2353 case OP_QUERY:
2354 case OP_QUERYI:
2355 case OP_MINQUERY:
2356 case OP_MINQUERYI:
2357 case OP_POSQUERY:
2358 case OP_POSQUERYI:
2359 if (utf8 && code[1] >= 0xc0) code += _pcre_utf8_table4[code[1] & 0x3f];
2360 break;
2361
2362 case OP_UPTO:
2363 case OP_UPTOI:
2364 case OP_MINUPTO:
2365 case OP_MINUPTOI:
2366 case OP_POSUPTO:
2367 case OP_POSUPTOI:
2368 if (utf8 && code[3] >= 0xc0) code += _pcre_utf8_table4[code[3] & 0x3f];
2369 break;
2370 #endif
2371
2372 /* MARK, and PRUNE/SKIP/THEN with an argument must skip over the argument
2373 string. */
2374
2375 case OP_MARK:
2376 case OP_PRUNE_ARG:
2377 case OP_SKIP_ARG:
2378 code += code[1];
2379 break;
2380
2381 case OP_THEN_ARG:
2382 code += code[1];
2383 break;
2384
2385 /* None of the remaining opcodes are required to match a character. */
2386
2387 default:
2388 break;
2389 }
2390 }
2391
2392 return TRUE;
2393 }
2394
2395
2396
2397 /*************************************************
2398 * Scan compiled regex for non-emptiness *
2399 *************************************************/
2400
2401 /* This function is called to check for left recursive calls. We want to check
2402 the current branch of the current pattern to see if it could match the empty
2403 string. If it could, we must look outwards for branches at other levels,
2404 stopping when we pass beyond the bracket which is the subject of the recursion.
2405 This function is called only during the real compile, not during the
2406 pre-compile.
2407
2408 Arguments:
2409 code points to start of the recursion
2410 endcode points to where to stop (current RECURSE item)
2411 bcptr points to the chain of current (unclosed) branch starts
2412 utf8 TRUE if in UTF-8 mode
2413 cd pointers to tables etc
2414
2415 Returns: TRUE if what is matched could be empty
2416 */
2417
2418 static BOOL
2419 could_be_empty(const uschar *code, const uschar *endcode, branch_chain *bcptr,
2420 BOOL utf8, compile_data *cd)
2421 {
2422 while (bcptr != NULL && bcptr->current_branch >= code)
2423 {
2424 if (!could_be_empty_branch(bcptr->current_branch, endcode, utf8, cd))
2425 return FALSE;
2426 bcptr = bcptr->outer;
2427 }
2428 return TRUE;
2429 }
2430
2431
2432
2433 /*************************************************
2434 * Check for POSIX class syntax *
2435 *************************************************/
2436
2437 /* This function is called when the sequence "[:" or "[." or "[=" is
2438 encountered in a character class. It checks whether this is followed by a
2439 sequence of characters terminated by a matching ":]" or ".]" or "=]". If we
2440 reach an unescaped ']' without the special preceding character, return FALSE.
2441
2442 Originally, this function only recognized a sequence of letters between the
2443 terminators, but it seems that Perl recognizes any sequence of characters,
2444 though of course unknown POSIX names are subsequently rejected. Perl gives an
2445 "Unknown POSIX class" error for [:f\oo:] for example, where previously PCRE
2446 didn't consider this to be a POSIX class. Likewise for [:1234:].
2447
2448 The problem in trying to be exactly like Perl is in the handling of escapes. We
2449 have to be sure that [abc[:x\]pqr] is *not* treated as containing a POSIX
2450 class, but [abc[:x\]pqr:]] is (so that an error can be generated). The code
2451 below handles the special case of \], but does not try to do any other escape
2452 processing. This makes it different from Perl for cases such as [:l\ower:]
2453 where Perl recognizes it as the POSIX class "lower" but PCRE does not recognize
2454 "l\ower". This is a lesser evil that not diagnosing bad classes when Perl does,
2455 I think.
2456
2457 A user pointed out that PCRE was rejecting [:a[:digit:]] whereas Perl was not.
2458 It seems that the appearance of a nested POSIX class supersedes an apparent
2459 external class. For example, [:a[:digit:]b:] matches "a", "b", ":", or
2460 a digit.
2461
2462 In Perl, unescaped square brackets may also appear as part of class names. For
2463 example, [:a[:abc]b:] gives unknown POSIX class "[:abc]b:]". However, for
2464 [:a[:abc]b][b:] it gives unknown POSIX class "[:abc]b][b:]", which does not
2465 seem right at all. PCRE does not allow closing square brackets in POSIX class
2466 names.
2467
2468 Arguments:
2469 ptr pointer to the initial [
2470 endptr where to return the end pointer
2471
2472 Returns: TRUE or FALSE
2473 */
2474
2475 static BOOL
2476 check_posix_syntax(const uschar *ptr, const uschar **endptr)
2477 {
2478 int terminator; /* Don't combine these lines; the Solaris cc */
2479 terminator = *(++ptr); /* compiler warns about "non-constant" initializer. */
2480 for (++ptr; *ptr != 0; ptr++)
2481 {
2482 if (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET)
2483 ptr++;
2484 else if (*ptr == CHAR_RIGHT_SQUARE_BRACKET) return FALSE;
2485 else
2486 {
2487 if (*ptr == terminator && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET)
2488 {
2489 *endptr = ptr;
2490 return TRUE;
2491 }
2492 if (*ptr == CHAR_LEFT_SQUARE_BRACKET &&
2493 (ptr[1] == CHAR_COLON || ptr[1] == CHAR_DOT ||
2494 ptr[1] == CHAR_EQUALS_SIGN) &&
2495 check_posix_syntax(ptr, endptr))
2496 return FALSE;
2497 }
2498 }
2499 return FALSE;
2500 }
2501
2502
2503
2504
2505 /*************************************************
2506 * Check POSIX class name *
2507 *************************************************/
2508
2509 /* This function is called to check the name given in a POSIX-style class entry
2510 such as [:alnum:].
2511
2512 Arguments:
2513 ptr points to the first letter
2514 len the length of the name
2515
2516 Returns: a value representing the name, or -1 if unknown
2517 */
2518
2519 static int
2520 check_posix_name(const uschar *ptr, int len)
2521 {
2522 const char *pn = posix_names;
2523 register int yield = 0;
2524 while (posix_name_lengths[yield] != 0)
2525 {
2526 if (len == posix_name_lengths[yield] &&
2527 strncmp((const char *)ptr, pn, len) == 0) return yield;
2528 pn += posix_name_lengths[yield] + 1;
2529 yield++;
2530 }
2531 return -1;
2532 }
2533
2534
2535 /*************************************************
2536 * Adjust OP_RECURSE items in repeated group *
2537 *************************************************/
2538
2539 /* OP_RECURSE items contain an offset from the start of the regex to the group
2540 that is referenced. This means that groups can be replicated for fixed
2541 repetition simply by copying (because the recursion is allowed to refer to
2542 earlier groups that are outside the current group). However, when a group is
2543 optional (i.e. the minimum quantifier is zero), OP_BRAZERO or OP_SKIPZERO is
2544 inserted before it, after it has been compiled. This means that any OP_RECURSE
2545 items within it that refer to the group itself or any contained groups have to
2546 have their offsets adjusted. That one of the jobs of this function. Before it
2547 is called, the partially compiled regex must be temporarily terminated with
2548 OP_END.
2549
2550 This function has been extended with the possibility of forward references for
2551 recursions and subroutine calls. It must also check the list of such references
2552 for the group we are dealing with. If it finds that one of the recursions in
2553 the current group is on this list, it adjusts the offset in the list, not the
2554 value in the reference (which is a group number).
2555
2556 Arguments:
2557 group points to the start of the group
2558 adjust the amount by which the group is to be moved
2559 utf8 TRUE in UTF-8 mode
2560 cd contains pointers to tables etc.
2561 save_hwm the hwm forward reference pointer at the start of the group
2562
2563 Returns: nothing
2564 */
2565
2566 static void
2567 adjust_recurse(uschar *group, int adjust, BOOL utf8, compile_data *cd,
2568 uschar *save_hwm)
2569 {
2570 uschar *ptr = group;
2571
2572 while ((ptr = (uschar *)find_recurse(ptr, utf8)) != NULL)
2573 {
2574 int offset;
2575 uschar *hc;
2576
2577 /* See if this recursion is on the forward reference list. If so, adjust the
2578 reference. */
2579
2580 for (hc = save_hwm; hc < cd->hwm; hc += LINK_SIZE)
2581 {
2582 offset = GET(hc, 0);
2583 if (cd->start_code + offset == ptr + 1)
2584 {
2585 PUT(hc, 0, offset + adjust);
2586 break;
2587 }
2588 }
2589
2590 /* Otherwise, adjust the recursion offset if it's after the start of this
2591 group. */
2592
2593 if (hc >= cd->hwm)
2594 {
2595 offset = GET(ptr, 1);
2596 if (cd->start_code + offset >= group) PUT(ptr, 1, offset + adjust);
2597 }
2598
2599 ptr += 1 + LINK_SIZE;
2600 }
2601 }
2602
2603
2604
2605 /*************************************************
2606 * Insert an automatic callout point *
2607 *************************************************/
2608
2609 /* This function is called when the PCRE_AUTO_CALLOUT option is set, to insert
2610 callout points before each pattern item.
2611
2612 Arguments:
2613 code current code pointer
2614 ptr current pattern pointer
2615 cd pointers to tables etc
2616
2617 Returns: new code pointer
2618 */
2619
2620 static uschar *
2621 auto_callout(uschar *code, const uschar *ptr, compile_data *cd)
2622 {
2623 *code++ = OP_CALLOUT;
2624 *code++ = 255;
2625 PUT(code, 0, (int)(ptr - cd->start_pattern)); /* Pattern offset */
2626 PUT(code, LINK_SIZE, 0); /* Default length */
2627 return code + 2*LINK_SIZE;
2628 }
2629
2630
2631
2632 /*************************************************
2633 * Complete a callout item *
2634 *************************************************/
2635
2636 /* A callout item contains the length of the next item in the pattern, which
2637 we can't fill in till after we have reached the relevant point. This is used
2638 for both automatic and manual callouts.
2639
2640 Arguments:
2641 previous_callout points to previous callout item
2642 ptr current pattern pointer
2643 cd pointers to tables etc
2644
2645 Returns: nothing
2646 */
2647
2648 static void
2649 complete_callout(uschar *previous_callout, const uschar *ptr, compile_data *cd)
2650 {
2651 int length = (int)(ptr - cd->start_pattern - GET(previous_callout, 2));
2652 PUT(previous_callout, 2 + LINK_SIZE, length);
2653 }
2654
2655
2656
2657 #ifdef SUPPORT_UCP
2658 /*************************************************
2659 * Get othercase range *
2660 *************************************************/
2661
2662 /* This function is passed the start and end of a class range, in UTF-8 mode
2663 with UCP support. It searches up the characters, looking for internal ranges of
2664 characters in the "other" case. Each call returns the next one, updating the
2665 start address.
2666
2667 Arguments:
2668 cptr points to starting character value; updated
2669 d end value
2670 ocptr where to put start of othercase range
2671 odptr where to put end of othercase range
2672
2673 Yield: TRUE when range returned; FALSE when no more
2674 */
2675
2676 static BOOL
2677 get_othercase_range(unsigned int *cptr, unsigned int d, unsigned int *ocptr,
2678 unsigned int *odptr)
2679 {
2680 unsigned int c, othercase, next;
2681
2682 for (c = *cptr; c <= d; c++)
2683 { if ((othercase = UCD_OTHERCASE(c)) != c) break; }
2684
2685 if (c > d) return FALSE;
2686
2687 *ocptr = othercase;
2688 next = othercase + 1;
2689
2690 for (++c; c <= d; c++)
2691 {
2692 if (UCD_OTHERCASE(c) != next) break;
2693 next++;
2694 }
2695
2696 *odptr = next - 1;
2697 *cptr = c;
2698
2699 return TRUE;
2700 }
2701
2702
2703
2704 /*************************************************
2705 * Check a character and a property *
2706 *************************************************/
2707
2708 /* This function is called by check_auto_possessive() when a property item
2709 is adjacent to a fixed character.
2710
2711 Arguments:
2712 c the character
2713 ptype the property type
2714 pdata the data for the type
2715 negated TRUE if it's a negated property (\P or \p{^)
2716
2717 Returns: TRUE if auto-possessifying is OK
2718 */
2719
2720 static BOOL
2721 check_char_prop(int c, int ptype, int pdata, BOOL negated)
2722 {
2723 const ucd_record *prop = GET_UCD(c);
2724 switch(ptype)
2725 {
2726 case PT_LAMP:
2727 return (prop->chartype == ucp_Lu ||
2728 prop->chartype == ucp_Ll ||
2729 prop->chartype == ucp_Lt) == negated;
2730
2731 case PT_GC:
2732 return (pdata == _pcre_ucp_gentype[prop->chartype]) == negated;
2733
2734 case PT_PC:
2735 return (pdata == prop->chartype) == negated;
2736
2737 case PT_SC:
2738 return (pdata == prop->script) == negated;
2739
2740 /* These are specials */
2741
2742 case PT_ALNUM:
2743 return (_pcre_ucp_gentype[prop->chartype] == ucp_L ||
2744 _pcre_ucp_gentype[prop->chartype] == ucp_N) == negated;
2745
2746 case PT_SPACE: /* Perl space */
2747 return (_pcre_ucp_gentype[prop->chartype] == ucp_Z ||
2748 c == CHAR_HT || c == CHAR_NL || c == CHAR_FF || c == CHAR_CR)
2749 == negated;
2750
2751 case PT_PXSPACE: /* POSIX space */
2752 return (_pcre_ucp_gentype[prop->chartype] == ucp_Z ||
2753 c == CHAR_HT || c == CHAR_NL || c == CHAR_VT ||
2754 c == CHAR_FF || c == CHAR_CR)
2755 == negated;
2756
2757 case PT_WORD:
2758 return (_pcre_ucp_gentype[prop->chartype] == ucp_L ||
2759 _pcre_ucp_gentype[prop->chartype] == ucp_N ||
2760 c == CHAR_UNDERSCORE) == negated;
2761 }
2762 return FALSE;
2763 }
2764 #endif /* SUPPORT_UCP */
2765
2766
2767
2768 /*************************************************
2769 * Check if auto-possessifying is possible *
2770 *************************************************/
2771
2772 /* This function is called for unlimited repeats of certain items, to see
2773 whether the next thing could possibly match the repeated item. If not, it makes
2774 sense to automatically possessify the repeated item.
2775
2776 Arguments:
2777 previous pointer to the repeated opcode
2778 utf8 TRUE in UTF-8 mode
2779 ptr next character in pattern
2780 options options bits
2781 cd contains pointers to tables etc.
2782
2783 Returns: TRUE if possessifying is wanted
2784 */
2785
2786 static BOOL
2787 check_auto_possessive(const uschar *previous, BOOL utf8, const uschar *ptr,
2788 int options, compile_data *cd)
2789 {
2790 int c, next;
2791 int op_code = *previous++;
2792
2793 /* Skip whitespace and comments in extended mode */
2794
2795 if ((options & PCRE_EXTENDED) != 0)
2796 {
2797 for (;;)
2798 {
2799 while ((cd->ctypes[*ptr] & ctype_space) != 0) ptr++;
2800 if (*ptr == CHAR_NUMBER_SIGN)
2801 {
2802 ptr++;
2803 while (*ptr != 0)
2804 {
2805 if (IS_NEWLINE(ptr)) { ptr += cd->nllen; break; }
2806 ptr++;
2807 #ifdef SUPPORT_UTF8
2808 if (utf8) while ((*ptr & 0xc0) == 0x80) ptr++;
2809 #endif
2810 }
2811 }
2812 else break;
2813 }
2814 }
2815
2816 /* If the next item is one that we can handle, get its value. A non-negative
2817 value is a character, a negative value is an escape value. */
2818
2819 if (*ptr == CHAR_BACKSLASH)
2820 {
2821 int temperrorcode = 0;
2822 next = check_escape(&ptr, &temperrorcode, cd->bracount, options, FALSE);
2823 if (temperrorcode != 0) return FALSE;
2824 ptr++; /* Point after the escape sequence */
2825 }
2826
2827 else if ((cd->ctypes[*ptr] & ctype_meta) == 0)
2828 {
2829 #ifdef SUPPORT_UTF8
2830 if (utf8) { GETCHARINC(next, ptr); } else
2831 #endif
2832 next = *ptr++;
2833 }
2834
2835 else return FALSE;
2836
2837 /* Skip whitespace and comments in extended mode */
2838
2839 if ((options & PCRE_EXTENDED) != 0)
2840 {
2841 for (;;)
2842 {
2843 while ((cd->ctypes[*ptr] & ctype_space) != 0) ptr++;
2844 if (*ptr == CHAR_NUMBER_SIGN)
2845 {
2846 ptr++;
2847 while (*ptr != 0)
2848 {
2849 if (IS_NEWLINE(ptr)) { ptr += cd->nllen; break; }
2850 ptr++;
2851 #ifdef SUPPORT_UTF8
2852 if (utf8) while ((*ptr & 0xc0) == 0x80) ptr++;
2853 #endif
2854 }
2855 }
2856 else break;
2857 }
2858 }
2859
2860 /* If the next thing is itself optional, we have to give up. */
2861
2862 if (*ptr == CHAR_ASTERISK || *ptr == CHAR_QUESTION_MARK ||
2863 strncmp((char *)ptr, STR_LEFT_CURLY_BRACKET STR_0 STR_COMMA, 3) == 0)
2864 return FALSE;
2865
2866 /* Now compare the next item with the previous opcode. First, handle cases when
2867 the next item is a character. */
2868
2869 if (next >= 0) switch(op_code)
2870 {
2871 case OP_CHAR:
2872 #ifdef SUPPORT_UTF8
2873 GETCHARTEST(c, previous);
2874 #else
2875 c = *previous;
2876 #endif
2877 return c != next;
2878
2879 /* For CHARI (caseless character) we must check the other case. If we have
2880 Unicode property support, we can use it to test the other case of
2881 high-valued characters. */
2882
2883 case OP_CHARI:
2884 #ifdef SUPPORT_UTF8
2885 GETCHARTEST(c, previous);
2886 #else
2887 c = *previous;
2888 #endif
2889 if (c == next) return FALSE;
2890 #ifdef SUPPORT_UTF8
2891 if (utf8)
2892 {
2893 unsigned int othercase;
2894 if (next < 128) othercase = cd->fcc[next]; else
2895 #ifdef SUPPORT_UCP
2896 othercase = UCD_OTHERCASE((unsigned int)next);
2897 #else
2898 othercase = NOTACHAR;
2899 #endif
2900 return (unsigned int)c != othercase;
2901 }
2902 else
2903 #endif /* SUPPORT_UTF8 */
2904 return (c != cd->fcc[next]); /* Non-UTF-8 mode */
2905
2906 /* For OP_NOT and OP_NOTI, the data is always a single-byte character. These
2907 opcodes are not used for multi-byte characters, because they are coded using
2908 an XCLASS instead. */
2909
2910 case OP_NOT:
2911 return (c = *previous) == next;
2912
2913 case OP_NOTI:
2914 if ((c = *previous) == next) return TRUE;
2915 #ifdef SUPPORT_UTF8
2916 if (utf8)
2917 {
2918 unsigned int othercase;
2919 if (next < 128) othercase = cd->fcc[next]; else
2920 #ifdef SUPPORT_UCP
2921 othercase = UCD_OTHERCASE(next);
2922 #else
2923 othercase = NOTACHAR;
2924 #endif
2925 return (unsigned int)c == othercase;
2926 }
2927 else
2928 #endif /* SUPPORT_UTF8 */
2929 return (c == cd->fcc[next]); /* Non-UTF-8 mode */
2930
2931 /* Note that OP_DIGIT etc. are generated only when PCRE_UCP is *not* set.
2932 When it is set, \d etc. are converted into OP_(NOT_)PROP codes. */
2933
2934 case OP_DIGIT:
2935 return next > 127 || (cd->ctypes[next] & ctype_digit) == 0;
2936
2937 case OP_NOT_DIGIT:
2938 return next <= 127 && (cd->ctypes[next] & ctype_digit) != 0;
2939
2940 case OP_WHITESPACE:
2941 return next > 127 || (cd->ctypes[next] & ctype_space) == 0;
2942
2943 case OP_NOT_WHITESPACE:
2944 return next <= 127 && (cd->ctypes[next] & ctype_space) != 0;
2945
2946 case OP_WORDCHAR:
2947 return next > 127 || (cd->ctypes[next] & ctype_word) == 0;
2948
2949 case OP_NOT_WORDCHAR:
2950 return next <= 127 && (cd->ctypes[next] & ctype_word) != 0;
2951
2952 case OP_HSPACE:
2953 case OP_NOT_HSPACE:
2954 switch(next)
2955 {
2956 case 0x09:
2957 case 0x20:
2958 case 0xa0:
2959 case 0x1680:
2960 case 0x180e:
2961 case 0x2000:
2962 case 0x2001:
2963 case 0x2002:
2964 case 0x2003:
2965 case 0x2004:
2966 case 0x2005:
2967 case 0x2006:
2968 case 0x2007:
2969 case 0x2008:
2970 case 0x2009:
2971 case 0x200A:
2972 case 0x202f:
2973 case 0x205f:
2974 case 0x3000:
2975 return op_code == OP_NOT_HSPACE;
2976 default:
2977 return op_code != OP_NOT_HSPACE;
2978 }
2979
2980 case OP_ANYNL:
2981 case OP_VSPACE:
2982 case OP_NOT_VSPACE:
2983 switch(next)
2984 {
2985 case 0x0a:
2986 case 0x0b:
2987 case 0x0c:
2988 case 0x0d:
2989 case 0x85:
2990 case 0x2028:
2991 case 0x2029:
2992 return op_code == OP_NOT_VSPACE;
2993 default:
2994 return op_code != OP_NOT_VSPACE;
2995 }
2996
2997 #ifdef SUPPORT_UCP
2998 case OP_PROP:
2999 return check_char_prop(next, previous[0], previous[1], FALSE);
3000
3001 case OP_NOTPROP:
3002 return check_char_prop(next, previous[0], previous[1], TRUE);
3003 #endif
3004
3005 default:
3006 return FALSE;
3007 }
3008
3009
3010 /* Handle the case when the next item is \d, \s, etc. Note that when PCRE_UCP
3011 is set, \d turns into ESC_du rather than ESC_d, etc., so ESC_d etc. are
3012 generated only when PCRE_UCP is *not* set, that is, when only ASCII
3013 characteristics are recognized. Similarly, the opcodes OP_DIGIT etc. are
3014 replaced by OP_PROP codes when PCRE_UCP is set. */
3015
3016 switch(op_code)
3017 {
3018 case OP_CHAR:
3019 case OP_CHARI:
3020 #ifdef SUPPORT_UTF8
3021 GETCHARTEST(c, previous);
3022 #else
3023 c = *previous;
3024 #endif
3025 switch(-next)
3026 {
3027 case ESC_d:
3028 return c > 127 || (cd->ctypes[c] & ctype_digit) == 0;
3029
3030 case ESC_D:
3031 return c <= 127 && (cd->ctypes[c] & ctype_digit) != 0;
3032
3033 case ESC_s:
3034 return c > 127 || (cd->ctypes[c] & ctype_space) == 0;
3035
3036 case ESC_S:
3037 return c <= 127 && (cd->ctypes[c] & ctype_space) != 0;
3038
3039 case ESC_w:
3040 return c > 127 || (cd->ctypes[c] & ctype_word) == 0;
3041
3042 case ESC_W:
3043 return c <= 127 && (cd->ctypes[c] & ctype_word) != 0;
3044
3045 case ESC_h:
3046 case ESC_H:
3047 switch(c)
3048 {
3049 case 0x09:
3050 case 0x20:
3051 case 0xa0:
3052 case 0x1680:
3053 case 0x180e:
3054 case 0x2000:
3055 case 0x2001:
3056 case 0x2002:
3057 case 0x2003:
3058 case 0x2004:
3059 case 0x2005:
3060 case 0x2006:
3061 case 0x2007:
3062 case 0x2008:
3063 case 0x2009:
3064 case 0x200A:
3065 case 0x202f:
3066 case 0x205f:
3067 case 0x3000:
3068 return -next != ESC_h;
3069 default:
3070 return -next == ESC_h;
3071 }
3072
3073 case ESC_v:
3074 case ESC_V:
3075 switch(c)
3076 {
3077 case 0x0a:
3078 case 0x0b:
3079 case 0x0c:
3080 case 0x0d:
3081 case 0x85:
3082 case 0x2028:
3083 case 0x2029:
3084 return -next != ESC_v;
3085 default:
3086 return -next == ESC_v;
3087 }
3088
3089 /* When PCRE_UCP is set, these values get generated for \d etc. Find
3090 their substitutions and process them. The result will always be either
3091 -ESC_p or -ESC_P. Then fall through to process those values. */
3092
3093 #ifdef SUPPORT_UCP
3094 case ESC_du:
3095 case ESC_DU:
3096 case ESC_wu:
3097 case ESC_WU:
3098 case ESC_su:
3099 case ESC_SU:
3100 {
3101 int temperrorcode = 0;
3102 ptr = substitutes[-next - ESC_DU];
3103 next = check_escape(&ptr, &temperrorcode, 0, options, FALSE);
3104 if (temperrorcode != 0) return FALSE;
3105 ptr++; /* For compatibility */
3106 }
3107 /* Fall through */
3108
3109 case ESC_p:
3110 case ESC_P:
3111 {
3112 int ptype, pdata, errorcodeptr;
3113 BOOL negated;
3114
3115 ptr--; /* Make ptr point at the p or P */
3116 ptype = get_ucp(&ptr, &negated, &pdata, &errorcodeptr);
3117 if (ptype < 0) return FALSE;
3118 ptr++; /* Point past the final curly ket */
3119
3120 /* If the property item is optional, we have to give up. (When generated
3121 from \d etc by PCRE_UCP, this test will have been applied much earlier,
3122 to the original \d etc. At this point, ptr will point to a zero byte. */
3123
3124 if (*ptr == CHAR_ASTERISK || *ptr == CHAR_QUESTION_MARK ||
3125 strncmp((char *)ptr, STR_LEFT_CURLY_BRACKET STR_0 STR_COMMA, 3) == 0)
3126 return FALSE;
3127
3128 /* Do the property check. */
3129
3130 return check_char_prop(c, ptype, pdata, (next == -ESC_P) != negated);
3131 }
3132 #endif
3133
3134 default:
3135 return FALSE;
3136 }
3137
3138 /* In principle, support for Unicode properties should be integrated here as
3139 well. It means re-organizing the above code so as to get hold of the property
3140 values before switching on the op-code. However, I wonder how many patterns
3141 combine ASCII \d etc with Unicode properties? (Note that if PCRE_UCP is set,
3142 these op-codes are never generated.) */
3143
3144 case OP_DIGIT:
3145 return next == -ESC_D || next == -ESC_s || next == -ESC_W ||
3146 next == -ESC_h || next == -ESC_v || next == -ESC_R;
3147
3148 case OP_NOT_DIGIT:
3149 return next == -ESC_d;
3150
3151 case OP_WHITESPACE:
3152 return next == -ESC_S || next == -ESC_d || next == -ESC_w || next == -ESC_R;
3153
3154 case OP_NOT_WHITESPACE:
3155 return next == -ESC_s || next == -ESC_h || next == -ESC_v;
3156
3157 case OP_HSPACE:
3158 return next == -ESC_S || next == -ESC_H || next == -ESC_d ||
3159 next == -ESC_w || next == -ESC_v || next == -ESC_R;
3160
3161 case OP_NOT_HSPACE:
3162 return next == -ESC_h;
3163
3164 /* Can't have \S in here because VT matches \S (Perl anomaly) */
3165 case OP_ANYNL:
3166 case OP_VSPACE:
3167 return next == -ESC_V || next == -ESC_d || next == -ESC_w;
3168
3169 case OP_NOT_VSPACE:
3170 return next == -ESC_v || next == -ESC_R;
3171
3172 case OP_WORDCHAR:
3173 return next == -ESC_W || next == -ESC_s || next == -ESC_h ||
3174 next == -ESC_v || next == -ESC_R;
3175
3176 case OP_NOT_WORDCHAR:
3177 return next == -ESC_w || next == -ESC_d;
3178
3179 default:
3180 return FALSE;
3181 }
3182
3183 /* Control does not reach here */
3184 }
3185
3186
3187
3188 /*************************************************
3189 * Compile one branch *
3190 *************************************************/
3191
3192 /* Scan the pattern, compiling it into the a vector. If the options are
3193 changed during the branch, the pointer is used to change the external options
3194 bits. This function is used during the pre-compile phase when we are trying
3195 to find out the amount of memory needed, as well as during the real compile
3196 phase. The value of lengthptr distinguishes the two phases.
3197
3198 Arguments:
3199 optionsptr pointer to the option bits
3200 codeptr points to the pointer to the current code point
3201 ptrptr points to the current pattern pointer
3202 errorcodeptr points to error code variable
3203 firstbyteptr set to initial literal character, or < 0 (REQ_UNSET, REQ_NONE)
3204 reqbyteptr set to the last literal character required, else < 0
3205 bcptr points to current branch chain
3206 cond_depth conditional nesting depth
3207 cd contains pointers to tables etc.
3208 lengthptr NULL during the real compile phase
3209 points to length accumulator during pre-compile phase
3210
3211 Returns: TRUE on success
3212 FALSE, with *errorcodeptr set non-zero on error
3213 */
3214
3215 static BOOL
3216 compile_branch(int *optionsptr, uschar **codeptr, const uschar **ptrptr,
3217 int *errorcodeptr, int *firstbyteptr, int *reqbyteptr, branch_chain *bcptr,
3218 int cond_depth, compile_data *cd, int *lengthptr)
3219 {
3220 int repeat_type, op_type;
3221 int repeat_min = 0, repeat_max = 0; /* To please picky compilers */
3222 int bravalue = 0;
3223 int greedy_default, greedy_non_default;
3224 int firstbyte, reqbyte;
3225 int zeroreqbyte, zerofirstbyte;
3226 int req_caseopt, reqvary, tempreqvary;
3227 int options = *optionsptr; /* May change dynamically */
3228 int after_manual_callout = 0;
3229 int length_prevgroup = 0;
3230 register int c;
3231 register uschar *code = *codeptr;
3232 uschar *last_code = code;
3233 uschar *orig_code = code;
3234 uschar *tempcode;
3235 BOOL inescq = FALSE;
3236 BOOL groupsetfirstbyte = FALSE;
3237 const uschar *ptr = *ptrptr;
3238 const uschar *tempptr;
3239 const uschar *nestptr = NULL;
3240 uschar *previous = NULL;
3241 uschar *previous_callout = NULL;
3242 uschar *save_hwm = NULL;
3243 uschar classbits[32];
3244
3245 /* We can fish out the UTF-8 setting once and for all into a BOOL, but we
3246 must not do this for other options (e.g. PCRE_EXTENDED) because they may change
3247 dynamically as we process the pattern. */
3248
3249 #ifdef SUPPORT_UTF8
3250 BOOL class_utf8;
3251 BOOL utf8 = (options & PCRE_UTF8) != 0;
3252 uschar *class_utf8data;
3253 uschar *class_utf8data_base;
3254 uschar utf8_char[6];
3255 #else
3256 BOOL utf8 = FALSE;
3257 #endif
3258
3259 #ifdef PCRE_DEBUG
3260 if (lengthptr != NULL) DPRINTF((">> start branch\n"));
3261 #endif
3262
3263 /* Set up the default and non-default settings for greediness */
3264
3265 greedy_default = ((options & PCRE_UNGREEDY) != 0);
3266 greedy_non_default = greedy_default ^ 1;
3267
3268 /* Initialize no first byte, no required byte. REQ_UNSET means "no char
3269 matching encountered yet". It gets changed to REQ_NONE if we hit something that
3270 matches a non-fixed char first char; reqbyte just remains unset if we never
3271 find one.
3272
3273 When we hit a repeat whose minimum is zero, we may have to adjust these values
3274 to take the zero repeat into account. This is implemented by setting them to
3275 zerofirstbyte and zeroreqbyte when such a repeat is encountered. The individual
3276 item types that can be repeated set these backoff variables appropriately. */
3277
3278 firstbyte = reqbyte = zerofirstbyte = zeroreqbyte = REQ_UNSET;
3279
3280 /* The variable req_caseopt contains either the REQ_CASELESS value or zero,
3281 according to the current setting of the caseless flag. REQ_CASELESS is a bit
3282 value > 255. It is added into the firstbyte or reqbyte variables to record the
3283 case status of the value. This is used only for ASCII characters. */
3284
3285 req_caseopt = ((options & PCRE_CASELESS) != 0)? REQ_CASELESS : 0;
3286
3287 /* Switch on next character until the end of the branch */
3288
3289 for (;; ptr++)
3290 {
3291 BOOL negate_class;
3292 BOOL should_flip_negation;
3293 BOOL possessive_quantifier;
3294 BOOL is_quantifier;
3295 BOOL is_recurse;
3296 BOOL reset_bracount;
3297 int class_charcount;
3298 int class_lastchar;
3299 int newoptions;
3300 int recno;
3301 int refsign;
3302 int skipbytes;
3303 int subreqbyte;
3304 int subfirstbyte;
3305 int terminator;
3306 int mclength;
3307 int tempbracount;
3308 uschar mcbuffer[8];
3309
3310 /* Get next byte in the pattern */
3311
3312 c = *ptr;
3313
3314 /* If we are at the end of a nested substitution, revert to the outer level
3315 string. Nesting only happens one level deep. */
3316
3317 if (c == 0 && nestptr != NULL)
3318 {
3319 ptr = nestptr;
3320 nestptr = NULL;
3321 c = *ptr;
3322 }
3323
3324 /* If we are in the pre-compile phase, accumulate the length used for the
3325 previous cycle of this loop. */
3326
3327 if (lengthptr != NULL)
3328 {
3329 #ifdef PCRE_DEBUG
3330 if (code > cd->hwm) cd->hwm = code; /* High water info */
3331 #endif
3332 if (code > cd->start_workspace + WORK_SIZE_CHECK) /* Check for overrun */
3333 {
3334 *errorcodeptr = ERR52;
3335 goto FAILED;
3336 }
3337
3338 /* There is at least one situation where code goes backwards: this is the
3339 case of a zero quantifier after a class (e.g. [ab]{0}). At compile time,
3340 the class is simply eliminated. However, it is created first, so we have to
3341 allow memory for it. Therefore, don't ever reduce the length at this point.
3342 */
3343
3344 if (code < last_code) code = last_code;
3345
3346 /* Paranoid check for integer overflow */
3347
3348 if (OFLOW_MAX - *lengthptr < code - last_code)
3349 {
3350 *errorcodeptr = ERR20;
3351 goto FAILED;
3352 }
3353
3354 *lengthptr += (int)(code - last_code);
3355 DPRINTF(("length=%d added %d c=%c\n", *lengthptr, (int)(code - last_code),
3356 c));
3357
3358 /* If "previous" is set and it is not at the start of the work space, move
3359 it back to there, in order to avoid filling up the work space. Otherwise,
3360 if "previous" is NULL, reset the current code pointer to the start. */
3361
3362 if (previous != NULL)
3363 {
3364 if (previous > orig_code)
3365 {
3366 memmove(orig_code, previous, code - previous);
3367 code -= previous - orig_code;
3368 previous = orig_code;
3369 }
3370 }
3371 else code = orig_code;
3372
3373 /* Remember where this code item starts so we can pick up the length
3374 next time round. */
3375
3376 last_code = code;
3377 }
3378
3379 /* In the real compile phase, just check the workspace used by the forward
3380 reference list. */
3381
3382 else if (cd->hwm > cd->start_workspace + WORK_SIZE_CHECK)
3383 {
3384 *errorcodeptr = ERR52;
3385 goto FAILED;
3386 }
3387
3388 /* If in \Q...\E, check for the end; if not, we have a literal */
3389
3390 if (inescq && c != 0)
3391 {
3392 if (c == CHAR_BACKSLASH && ptr[1] == CHAR_E)
3393 {
3394 inescq = FALSE;
3395 ptr++;
3396 continue;
3397 }
3398 else
3399 {
3400 if (previous_callout != NULL)
3401 {
3402 if (lengthptr == NULL) /* Don't attempt in pre-compile phase */
3403 complete_callout(previous_callout, ptr, cd);
3404 previous_callout = NULL;
3405 }
3406 if ((options & PCRE_AUTO_CALLOUT) != 0)
3407 {
3408 previous_callout = code;
3409 code = auto_callout(code, ptr, cd);
3410 }
3411 goto NORMAL_CHAR;
3412 }
3413 }
3414
3415 /* Fill in length of a previous callout, except when the next thing is
3416 a quantifier. */
3417
3418 is_quantifier =
3419 c == CHAR_ASTERISK || c == CHAR_PLUS || c == CHAR_QUESTION_MARK ||
3420 (c == CHAR_LEFT_CURLY_BRACKET && is_counted_repeat(ptr+1));
3421
3422 if (!is_quantifier && previous_callout != NULL &&
3423 after_manual_callout-- <= 0)
3424 {
3425 if (lengthptr == NULL) /* Don't attempt in pre-compile phase */
3426 complete_callout(previous_callout, ptr, cd);
3427 previous_callout = NULL;
3428 }
3429
3430 /* In extended mode, skip white space and comments. */
3431
3432 if ((options & PCRE_EXTENDED) != 0)
3433 {
3434 if ((cd->ctypes[c] & ctype_space) != 0) continue;
3435 if (c == CHAR_NUMBER_SIGN)
3436 {
3437 ptr++;
3438 while (*ptr != 0)
3439 {
3440 if (IS_NEWLINE(ptr)) { ptr += cd->nllen - 1; break; }
3441 ptr++;
3442 #ifdef SUPPORT_UTF8
3443 if (utf8) while ((*ptr & 0xc0) == 0x80) ptr++;
3444 #endif
3445 }
3446 if (*ptr != 0) continue;
3447
3448 /* Else fall through to handle end of string */
3449 c = 0;
3450 }
3451 }
3452
3453 /* No auto callout for quantifiers. */
3454
3455 if ((options & PCRE_AUTO_CALLOUT) != 0 && !is_quantifier)
3456 {
3457 previous_callout = code;
3458 code = auto_callout(code, ptr, cd);
3459 }
3460
3461 switch(c)
3462 {
3463 /* ===================================================================*/
3464 case 0: /* The branch terminates at string end */
3465 case CHAR_VERTICAL_LINE: /* or | or ) */
3466 case CHAR_RIGHT_PARENTHESIS:
3467 *firstbyteptr = firstbyte;
3468 *reqbyteptr = reqbyte;
3469 *codeptr = code;
3470 *ptrptr = ptr;
3471 if (lengthptr != NULL)
3472 {
3473 if (OFLOW_MAX - *lengthptr < code - last_code)
3474 {
3475 *errorcodeptr = ERR20;
3476 goto FAILED;
3477 }
3478 *lengthptr += (int)(code - last_code); /* To include callout length */
3479 DPRINTF((">> end branch\n"));
3480 }
3481 return TRUE;
3482
3483
3484 /* ===================================================================*/
3485 /* Handle single-character metacharacters. In multiline mode, ^ disables
3486 the setting of any following char as a first character. */
3487
3488 case CHAR_CIRCUMFLEX_ACCENT:
3489 previous = NULL;
3490 if ((options & PCRE_MULTILINE) != 0)
3491 {
3492 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
3493 *code++ = OP_CIRCM;
3494 }
3495 else *code++ = OP_CIRC;
3496 break;
3497
3498 case CHAR_DOLLAR_SIGN:
3499 previous = NULL;
3500 *code++ = ((options & PCRE_MULTILINE) != 0)? OP_DOLLM : OP_DOLL;
3501 break;
3502
3503 /* There can never be a first char if '.' is first, whatever happens about
3504 repeats. The value of reqbyte doesn't change either. */
3505
3506 case CHAR_DOT:
3507 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
3508 zerofirstbyte = firstbyte;
3509 zeroreqbyte = reqbyte;
3510 previous = code;
3511 *code++ = ((options & PCRE_DOTALL) != 0)? OP_ALLANY: OP_ANY;
3512 break;
3513
3514
3515 /* ===================================================================*/
3516 /* Character classes. If the included characters are all < 256, we build a
3517 32-byte bitmap of the permitted characters, except in the special case
3518 where there is only one such character. For negated classes, we build the
3519 map as usual, then invert it at the end. However, we use a different opcode
3520 so that data characters > 255 can be handled correctly.
3521
3522 If the class contains characters outside the 0-255 range, a different
3523 opcode is compiled. It may optionally have a bit map for characters < 256,
3524 but those above are are explicitly listed afterwards. A flag byte tells
3525 whether the bitmap is present, and whether this is a negated class or not.
3526
3527 In JavaScript compatibility mode, an isolated ']' causes an error. In
3528 default (Perl) mode, it is treated as a data character. */
3529
3530 case CHAR_RIGHT_SQUARE_BRACKET:
3531 if ((cd->external_options & PCRE_JAVASCRIPT_COMPAT) != 0)
3532 {
3533 *errorcodeptr = ERR64;
3534 goto FAILED;
3535 }
3536 goto NORMAL_CHAR;
3537
3538 case CHAR_LEFT_SQUARE_BRACKET:
3539 previous = code;
3540
3541 /* PCRE supports POSIX class stuff inside a class. Perl gives an error if
3542 they are encountered at the top level, so we'll do that too. */
3543
3544 if ((ptr[1] == CHAR_COLON || ptr[1] == CHAR_DOT ||
3545 ptr[1] == CHAR_EQUALS_SIGN) &&
3546 check_posix_syntax(ptr, &tempptr))
3547 {
3548 *errorcodeptr = (ptr[1] == CHAR_COLON)? ERR13 : ERR31;
3549 goto FAILED;
3550 }
3551
3552 /* If the first character is '^', set the negation flag and skip it. Also,
3553 if the first few characters (either before or after ^) are \Q\E or \E we
3554 skip them too. This makes for compatibility with Perl. */
3555
3556 negate_class = FALSE;
3557 for (;;)
3558 {
3559 c = *(++ptr);
3560 if (c == CHAR_BACKSLASH)
3561 {
3562 if (ptr[1] == CHAR_E)
3563 ptr++;
3564 else if (strncmp((const char *)ptr+1,
3565 STR_Q STR_BACKSLASH STR_E, 3) == 0)
3566 ptr += 3;
3567 else
3568 break;
3569 }
3570 else if (!negate_class && c == CHAR_CIRCUMFLEX_ACCENT)
3571 negate_class = TRUE;
3572 else break;
3573 }
3574
3575 /* Empty classes are allowed in JavaScript compatibility mode. Otherwise,
3576 an initial ']' is taken as a data character -- the code below handles
3577 that. In JS mode, [] must always fail, so generate OP_FAIL, whereas
3578 [^] must match any character, so generate OP_ALLANY. */
3579
3580 if (c == CHAR_RIGHT_SQUARE_BRACKET &&
3581 (cd->external_options & PCRE_JAVASCRIPT_COMPAT) != 0)
3582 {
3583 *code++ = negate_class? OP_ALLANY : OP_FAIL;
3584 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
3585 zerofirstbyte = firstbyte;
3586 break;
3587 }
3588
3589 /* If a class contains a negative special such as \S, we need to flip the
3590 negation flag at the end, so that support for characters > 255 works
3591 correctly (they are all included in the class). */
3592
3593 should_flip_negation = FALSE;
3594
3595 /* Keep a count of chars with values < 256 so that we can optimize the case
3596 of just a single character (as long as it's < 256). However, For higher
3597 valued UTF-8 characters, we don't yet do any optimization. */
3598
3599 class_charcount = 0;
3600 class_lastchar = -1;
3601
3602 /* Initialize the 32-char bit map to all zeros. We build the map in a
3603 temporary bit of memory, in case the class contains only 1 character (less
3604 than 256), because in that case the compiled code doesn't use the bit map.
3605 */
3606
3607 memset(classbits, 0, 32 * sizeof(uschar));
3608
3609 #ifdef SUPPORT_UTF8
3610 class_utf8 = FALSE; /* No chars >= 256 */
3611 class_utf8data = code + LINK_SIZE + 2; /* For UTF-8 items */
3612 class_utf8data_base = class_utf8data; /* For resetting in pass 1 */
3613 #endif
3614
3615 /* Process characters until ] is reached. By writing this as a "do" it
3616 means that an initial ] is taken as a data character. At the start of the
3617 loop, c contains the first byte of the character. */
3618
3619 if (c != 0) do
3620 {
3621 const uschar *oldptr;
3622
3623 #ifdef SUPPORT_UTF8
3624 if (utf8 && c > 127)
3625 { /* Braces are required because the */
3626 GETCHARLEN(c, ptr, ptr); /* macro generates multiple statements */
3627 }
3628
3629 /* In the pre-compile phase, accumulate the length of any UTF-8 extra
3630 data and reset the pointer. This is so that very large classes that
3631 contain a zillion UTF-8 characters no longer overwrite the work space
3632 (which is on the stack). */
3633
3634 if (lengthptr != NULL)
3635 {
3636 *lengthptr += class_utf8data - class_utf8data_base;
3637 class_utf8data = class_utf8data_base;
3638 }
3639
3640 #endif
3641
3642 /* Inside \Q...\E everything is literal except \E */
3643
3644 if (inescq)
3645 {
3646 if (c == CHAR_BACKSLASH && ptr[1] == CHAR_E) /* If we are at \E */
3647 {
3648 inescq = FALSE; /* Reset literal state */
3649 ptr++; /* Skip the 'E' */
3650 continue; /* Carry on with next */
3651 }
3652 goto CHECK_RANGE; /* Could be range if \E follows */
3653 }
3654
3655 /* Handle POSIX class names. Perl allows a negation extension of the
3656 form [:^name:]. A square bracket that doesn't match the syntax is
3657 treated as a literal. We also recognize the POSIX constructions
3658 [.ch.] and [=ch=] ("collating elements") and fault them, as Perl
3659 5.6 and 5.8 do. */
3660
3661 if (c == CHAR_LEFT_SQUARE_BRACKET &&
3662 (ptr[1] == CHAR_COLON || ptr[1] == CHAR_DOT ||
3663 ptr[1] == CHAR_EQUALS_SIGN) && check_posix_syntax(ptr, &tempptr))
3664 {
3665 BOOL local_negate = FALSE;
3666 int posix_class, taboffset, tabopt;
3667 register const uschar *cbits = cd->cbits;
3668 uschar pbits[32];
3669
3670 if (ptr[1] != CHAR_COLON)
3671 {
3672 *errorcodeptr = ERR31;
3673 goto FAILED;
3674 }
3675
3676 ptr += 2;
3677 if (*ptr == CHAR_CIRCUMFLEX_ACCENT)
3678 {
3679 local_negate = TRUE;
3680 should_flip_negation = TRUE; /* Note negative special */
3681 ptr++;
3682 }
3683
3684 posix_class = check_posix_name(ptr, (int)(tempptr - ptr));
3685 if (posix_class < 0)
3686 {
3687 *errorcodeptr = ERR30;
3688 goto FAILED;
3689 }
3690
3691 /* If matching is caseless, upper and lower are converted to
3692 alpha. This relies on the fact that the class table starts with
3693 alpha, lower, upper as the first 3 entries. */
3694
3695 if ((options & PCRE_CASELESS) != 0 && posix_class <= 2)
3696 posix_class = 0;
3697
3698 /* When PCRE_UCP is set, some of the POSIX classes are converted to
3699 different escape sequences that use Unicode properties. */
3700
3701 #ifdef SUPPORT_UCP
3702 if ((options & PCRE_UCP) != 0)
3703 {
3704 int pc = posix_class + ((local_negate)? POSIX_SUBSIZE/2 : 0);
3705 if (posix_substitutes[pc] != NULL)
3706 {
3707 nestptr = tempptr + 1;
3708 ptr = posix_substitutes[pc] - 1;
3709 continue;
3710 }
3711 }
3712 #endif
3713 /* In the non-UCP case, we build the bit map for the POSIX class in a
3714 chunk of local store because we may be adding and subtracting from it,
3715 and we don't want to subtract bits that may be in the main map already.
3716 At the end we or the result into the bit map that is being built. */
3717
3718 posix_class *= 3;
3719
3720 /* Copy in the first table (always present) */
3721
3722 memcpy(pbits, cbits + posix_class_maps[posix_class],
3723 32 * sizeof(uschar));
3724
3725 /* If there is a second table, add or remove it as required. */
3726
3727 taboffset = posix_class_maps[posix_class + 1];
3728 tabopt = posix_class_maps[posix_class + 2];
3729
3730 if (taboffset >= 0)
3731 {
3732 if (tabopt >= 0)
3733 for (c = 0; c < 32; c++) pbits[c] |= cbits[c + taboffset];
3734 else
3735 for (c = 0; c < 32; c++) pbits[c] &= ~cbits[c + taboffset];
3736 }
3737
3738 /* Not see if we need to remove any special characters. An option
3739 value of 1 removes vertical space and 2 removes underscore. */
3740
3741 if (tabopt < 0) tabopt = -tabopt;
3742 if (tabopt == 1) pbits[1] &= ~0x3c;
3743 else if (tabopt == 2) pbits[11] &= 0x7f;
3744
3745 /* Add the POSIX table or its complement into the main table that is
3746 being built and we are done. */
3747
3748 if (local_negate)
3749 for (c = 0; c < 32; c++) classbits[c] |= ~pbits[c];
3750 else
3751 for (c = 0; c < 32; c++) classbits[c] |= pbits[c];
3752
3753 ptr = tempptr + 1;
3754 class_charcount = 10; /* Set > 1; assumes more than 1 per class */
3755 continue; /* End of POSIX syntax handling */
3756 }
3757
3758 /* Backslash may introduce a single character, or it may introduce one
3759 of the specials, which just set a flag. The sequence \b is a special
3760 case. Inside a class (and only there) it is treated as backspace. We
3761 assume that other escapes have more than one character in them, so set
3762 class_charcount bigger than one. Unrecognized escapes fall through and
3763 are either treated as literal characters (by default), or are faulted if
3764 PCRE_EXTRA is set. */
3765
3766 if (c == CHAR_BACKSLASH)
3767 {
3768 c = check_escape(&ptr, errorcodeptr, cd->bracount, options, TRUE);
3769 if (*errorcodeptr != 0) goto FAILED;
3770
3771 if (-c == ESC_b) c = CHAR_BS; /* \b is backspace in a class */
3772 else if (-c == ESC_Q) /* Handle start of quoted string */
3773 {
3774 if (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
3775 {
3776 ptr += 2; /* avoid empty string */
3777 }
3778 else inescq = TRUE;
3779 continue;
3780 }
3781 else if (-c == ESC_E) continue; /* Ignore orphan \E */
3782
3783 if (c < 0)
3784 {
3785 register const uschar *cbits = cd->cbits;
3786 class_charcount += 2; /* Greater than 1 is what matters */
3787
3788 switch (-c)
3789 {
3790 #ifdef SUPPORT_UCP
3791 case ESC_du: /* These are the values given for \d etc */
3792 case ESC_DU: /* when PCRE_UCP is set. We replace the */
3793 case ESC_wu: /* escape sequence with an appropriate \p */
3794 case ESC_WU: /* or \P to test Unicode properties instead */
3795 case ESC_su: /* of the default ASCII testing. */
3796 case ESC_SU:
3797 nestptr = ptr;
3798 ptr = substitutes[-c - ESC_DU] - 1; /* Just before substitute */
3799 class_charcount -= 2; /* Undo! */
3800 continue;
3801 #endif
3802 case ESC_d:
3803 for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_digit];
3804 continue;
3805
3806 case ESC_D:
3807 should_flip_negation = TRUE;
3808 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_digit];
3809 continue;
3810
3811 case ESC_w:
3812 for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_word];
3813 continue;
3814
3815 case ESC_W:
3816 should_flip_negation = TRUE;
3817 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_word];
3818 continue;
3819
3820 /* Perl 5.004 onwards omits VT from \s, but we must preserve it
3821 if it was previously set by something earlier in the character
3822 class. */
3823
3824 case ESC_s:
3825 classbits[0] |= cbits[cbit_space];
3826 classbits[1] |= cbits[cbit_space+1] & ~0x08;
3827 for (c = 2; c < 32; c++) classbits[c] |= cbits[c+cbit_space];
3828 continue;
3829
3830 case ESC_S:
3831 should_flip_negation = TRUE;
3832 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_space];
3833 classbits[1] |= 0x08; /* Perl 5.004 onwards omits VT from \s */
3834 continue;
3835
3836 case ESC_h:
3837 SETBIT(classbits, 0x09); /* VT */
3838 SETBIT(classbits, 0x20); /* SPACE */
3839 SETBIT(classbits, 0xa0); /* NSBP */
3840 #ifdef SUPPORT_UTF8
3841 if (utf8)
3842 {
3843 class_utf8 = TRUE;
3844 *class_utf8data++ = XCL_SINGLE;
3845 class_utf8data += _pcre_ord2utf8(0x1680, class_utf8data);
3846 *class_utf8data++ = XCL_SINGLE;
3847 class_utf8data += _pcre_ord2utf8(0x180e, class_utf8data);
3848 *class_utf8data++ = XCL_RANGE;
3849 class_utf8data += _pcre_ord2utf8(0x2000, class_utf8data);
3850 class_utf8data += _pcre_ord2utf8(0x200A, class_utf8data);
3851 *class_utf8data++ = XCL_SINGLE;
3852 class_utf8data += _pcre_ord2utf8(0x202f, class_utf8data);
3853 *class_utf8data++ = XCL_SINGLE;
3854 class_utf8data += _pcre_ord2utf8(0x205f, class_utf8data);
3855 *class_utf8data++ = XCL_SINGLE;
3856 class_utf8data += _pcre_ord2utf8(0x3000, class_utf8data);
3857 }
3858 #endif
3859 continue;
3860
3861 case ESC_H:
3862 for (c = 0; c < 32; c++)
3863 {
3864 int x = 0xff;
3865 switch (c)
3866 {
3867 case 0x09/8: x ^= 1 << (0x09%8); break;
3868 case 0x20/8: x ^= 1 << (0x20%8); break;
3869 case 0xa0/8: x ^= 1 << (0xa0%8); break;
3870 default: break;
3871 }
3872 classbits[c] |= x;
3873 }
3874
3875 #ifdef SUPPORT_UTF8
3876 if (utf8)
3877 {
3878 class_utf8 = TRUE;
3879 *class_utf8data++ = XCL_RANGE;
3880 class_utf8data += _pcre_ord2utf8(0x0100, class_utf8data);
3881 class_utf8data += _pcre_ord2utf8(0x167f, class_utf8data);
3882 *class_utf8data++ = XCL_RANGE;
3883 class_utf8data += _pcre_ord2utf8(0x1681, class_utf8data);
3884 class_utf8data += _pcre_ord2utf8(0x180d, class_utf8data);
3885 *class_utf8data++ = XCL_RANGE;
3886 class_utf8data += _pcre_ord2utf8(0x180f, class_utf8data);
3887 class_utf8data += _pcre_ord2utf8(0x1fff, class_utf8data);
3888 *class_utf8data++ = XCL_RANGE;
3889 class_utf8data += _pcre_ord2utf8(0x200B, class_utf8data);
3890 class_utf8data += _pcre_ord2utf8(0x202e, class_utf8data);
3891 *class_utf8data++ = XCL_RANGE;
3892 class_utf8data += _pcre_ord2utf8(0x2030, class_utf8data);
3893 class_utf8data += _pcre_ord2utf8(0x205e, class_utf8data);
3894 *class_utf8data++ = XCL_RANGE;
3895 class_utf8data += _pcre_ord2utf8(0x2060, class_utf8data);
3896 class_utf8data += _pcre_ord2utf8(0x2fff, class_utf8data);
3897 *class_utf8data++ = XCL_RANGE;
3898 class_utf8data += _pcre_ord2utf8(0x3001, class_utf8data);
3899 class_utf8data += _pcre_ord2utf8(0x7fffffff, class_utf8data);
3900 }
3901 #endif
3902 continue;
3903
3904 case ESC_v:
3905 SETBIT(classbits, 0x0a); /* LF */
3906 SETBIT(classbits, 0x0b); /* VT */
3907 SETBIT(classbits, 0x0c); /* FF */
3908 SETBIT(classbits, 0x0d); /* CR */
3909 SETBIT(classbits, 0x85); /* NEL */
3910 #ifdef SUPPORT_UTF8
3911 if (utf8)
3912 {
3913 class_utf8 = TRUE;
3914 *class_utf8data++ = XCL_RANGE;
3915 class_utf8data += _pcre_ord2utf8(0x2028, class_utf8data);
3916 class_utf8data += _pcre_ord2utf8(0x2029, class_utf8data);
3917 }
3918 #endif
3919 continue;
3920
3921 case ESC_V:
3922 for (c = 0; c < 32; c++)
3923 {
3924 int x = 0xff;
3925 switch (c)
3926 {
3927 case 0x0a/8: x ^= 1 << (0x0a%8);
3928 x ^= 1 << (0x0b%8);
3929 x ^= 1 << (0x0c%8);
3930 x ^= 1 << (0x0d%8);
3931 break;
3932 case 0x85/8: x ^= 1 << (0x85%8); break;
3933 default: break;
3934 }
3935 classbits[c] |= x;
3936 }
3937
3938 #ifdef SUPPORT_UTF8
3939 if (utf8)
3940 {
3941 class_utf8 = TRUE;
3942 *class_utf8data++ = XCL_RANGE;
3943 class_utf8data += _pcre_ord2utf8(0x0100, class_utf8data);
3944 class_utf8data += _pcre_ord2utf8(0x2027, class_utf8data);
3945 *class_utf8data++ = XCL_RANGE;
3946 class_utf8data += _pcre_ord2utf8(0x2029, class_utf8data);
3947 class_utf8data += _pcre_ord2utf8(0x7fffffff, class_utf8data);
3948 }
3949 #endif
3950 continue;
3951
3952 #ifdef SUPPORT_UCP
3953 case ESC_p:
3954 case ESC_P:
3955 {
3956 BOOL negated;
3957 int pdata;
3958 int ptype = get_ucp(&ptr, &negated, &pdata, errorcodeptr);
3959 if (ptype < 0) goto FAILED;
3960 class_utf8 = TRUE;
3961 *class_utf8data++ = ((-c == ESC_p) != negated)?
3962 XCL_PROP : XCL_NOTPROP;
3963 *class_utf8data++ = ptype;
3964 *class_utf8data++ = pdata;
3965 class_charcount -= 2; /* Not a < 256 character */
3966 continue;
3967 }
3968 #endif
3969 /* Unrecognized escapes are faulted if PCRE is running in its
3970 strict mode. By default, for compatibility with Perl, they are
3971 treated as literals. */
3972
3973 default:
3974 if ((options & PCRE_EXTRA) != 0)
3975 {
3976 *errorcodeptr = ERR7;
3977 goto FAILED;
3978 }
3979 class_charcount -= 2; /* Undo the default count from above */
3980 c = *ptr; /* Get the final character and fall through */
3981 break;
3982 }
3983 }
3984
3985 /* Fall through if we have a single character (c >= 0). This may be
3986 greater than 256 in UTF-8 mode. */
3987
3988 } /* End of backslash handling */
3989
3990 /* A single character may be followed by '-' to form a range. However,
3991 Perl does not permit ']' to be the end of the range. A '-' character
3992 at the end is treated as a literal. Perl ignores orphaned \E sequences
3993 entirely. The code for handling \Q and \E is messy. */
3994
3995 CHECK_RANGE:
3996 while (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
3997 {
3998 inescq = FALSE;
3999 ptr += 2;
4000 }
4001
4002 oldptr = ptr;
4003
4004 /* Remember \r or \n */
4005
4006 if (c == CHAR_CR || c == CHAR_NL) cd->external_flags |= PCRE_HASCRORLF;
4007
4008 /* Check for range */
4009
4010 if (!inescq && ptr[1] == CHAR_MINUS)
4011 {
4012 int d;
4013 ptr += 2;
4014 while (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_E) ptr += 2;
4015
4016 /* If we hit \Q (not followed by \E) at this point, go into escaped
4017 mode. */
4018
4019 while (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_Q)
4020 {
4021 ptr += 2;
4022 if (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_E)
4023 { ptr += 2; continue; }
4024 inescq = TRUE;
4025 break;
4026 }
4027
4028 if (*ptr == 0 || (!inescq && *ptr == CHAR_RIGHT_SQUARE_BRACKET))
4029 {
4030 ptr = oldptr;
4031 goto LONE_SINGLE_CHARACTER;
4032 }
4033
4034 #ifdef SUPPORT_UTF8
4035 if (utf8)
4036 { /* Braces are required because the */
4037 GETCHARLEN(d, ptr, ptr); /* macro generates multiple statements */
4038 }
4039 else
4040 #endif
4041 d = *ptr; /* Not UTF-8 mode */
4042
4043 /* The second part of a range can be a single-character escape, but
4044 not any of the other escapes. Perl 5.6 treats a hyphen as a literal
4045 in such circumstances. */
4046
4047 if (!inescq && d == CHAR_BACKSLASH)
4048 {
4049 d = check_escape(&ptr, errorcodeptr, cd->bracount, options, TRUE);
4050 if (*errorcodeptr != 0) goto FAILED;
4051
4052 /* \b is backspace; any other special means the '-' was literal */
4053
4054 if (d < 0)
4055 {
4056 if (d == -ESC_b) d = CHAR_BS; else
4057 {
4058 ptr = oldptr;
4059 goto LONE_SINGLE_CHARACTER; /* A few lines below */
4060 }
4061 }
4062 }
4063
4064 /* Check that the two values are in the correct order. Optimize
4065 one-character ranges */
4066
4067 if (d < c)
4068 {
4069 *errorcodeptr = ERR8;
4070 goto FAILED;
4071 }
4072
4073 if (d == c) goto LONE_SINGLE_CHARACTER; /* A few lines below */
4074
4075 /* Remember \r or \n */
4076
4077 if (d == CHAR_CR || d == CHAR_NL) cd->external_flags |= PCRE_HASCRORLF;
4078
4079 /* In UTF-8 mode, if the upper limit is > 255, or > 127 for caseless
4080 matching, we have to use an XCLASS with extra data items. Caseless
4081 matching for characters > 127 is available only if UCP support is
4082 available. */
4083
4084 #ifdef SUPPORT_UTF8
4085 if (utf8 && (d > 255 || ((options & PCRE_CASELESS) != 0 && d > 127)))
4086 {
4087 class_utf8 = TRUE;
4088
4089 /* With UCP support, we can find the other case equivalents of
4090 the relevant characters. There may be several ranges. Optimize how
4091 they fit with the basic range. */
4092
4093 #ifdef SUPPORT_UCP
4094 if ((options & PCRE_CASELESS) != 0)
4095 {
4096 unsigned int occ, ocd;
4097 unsigned int cc = c;
4098 unsigned int origd = d;
4099 while (get_othercase_range(&cc, origd, &occ, &ocd))
4100 {
4101 if (occ >= (unsigned int)c &&
4102 ocd <= (unsigned int)d)
4103 continue; /* Skip embedded ranges */
4104
4105 if (occ < (unsigned int)c &&
4106 ocd >= (unsigned int)c - 1) /* Extend the basic range */
4107 { /* if there is overlap, */
4108 c = occ; /* noting that if occ < c */
4109 continue; /* we can't have ocd > d */
4110 } /* because a subrange is */
4111 if (ocd > (unsigned int)d &&
4112 occ <= (unsigned int)d + 1) /* always shorter than */
4113 { /* the basic range. */
4114 d = ocd;
4115 continue;
4116 }
4117
4118 if (occ == ocd)
4119 {
4120 *class_utf8data++ = XCL_SINGLE;
4121 }
4122 else
4123 {
4124 *class_utf8data++ = XCL_RANGE;
4125 class_utf8data += _pcre_ord2utf8(occ, class_utf8data);
4126 }
4127 class_utf8data += _pcre_ord2utf8(ocd, class_utf8data);
4128 }
4129 }
4130 #endif /* SUPPORT_UCP */
4131
4132 /* Now record the original range, possibly modified for UCP caseless
4133 overlapping ranges. */
4134
4135 *class_utf8data++ = XCL_RANGE;
4136 class_utf8data += _pcre_ord2utf8(c, class_utf8data);
4137 class_utf8data += _pcre_ord2utf8(d, class_utf8data);
4138
4139 /* With UCP support, we are done. Without UCP support, there is no
4140 caseless matching for UTF-8 characters > 127; we can use the bit map
4141 for the smaller ones. */
4142
4143 #ifdef SUPPORT_UCP
4144 continue; /* With next character in the class */
4145 #else
4146 if ((options & PCRE_CASELESS) == 0 || c > 127) continue;
4147
4148 /* Adjust upper limit and fall through to set up the map */
4149
4150 d = 127;
4151
4152 #endif /* SUPPORT_UCP */
4153 }
4154 #endif /* SUPPORT_UTF8 */
4155
4156 /* We use the bit map for all cases when not in UTF-8 mode; else
4157 ranges that lie entirely within 0-127 when there is UCP support; else
4158 for partial ranges without UCP support. */
4159
4160 class_charcount += d - c + 1;
4161 class_lastchar = d;
4162
4163 /* We can save a bit of time by skipping this in the pre-compile. */
4164
4165 if (lengthptr == NULL) for (; c <= d; c++)
4166 {
4167 classbits[c/8] |= (1 << (c&7));
4168 if ((options & PCRE_CASELESS) != 0)
4169 {
4170 int uc = cd->fcc[c]; /* flip case */
4171 classbits[uc/8] |= (1 << (uc&7));
4172 }
4173 }
4174
4175 continue; /* Go get the next char in the class */
4176 }
4177
4178 /* Handle a lone single character - we can get here for a normal
4179 non-escape char, or after \ that introduces a single character or for an
4180 apparent range that isn't. */
4181
4182 LONE_SINGLE_CHARACTER:
4183
4184 /* Handle a character that cannot go in the bit map */
4185
4186 #ifdef SUPPORT_UTF8
4187 if (utf8 && (c > 255 || ((options & PCRE_CASELESS) != 0 && c > 127)))
4188 {
4189 class_utf8 = TRUE;
4190 *class_utf8data++ = XCL_SINGLE;
4191 class_utf8data += _pcre_ord2utf8(c, class_utf8data);
4192
4193 #ifdef SUPPORT_UCP
4194 if ((options & PCRE_CASELESS) != 0)
4195 {
4196 unsigned int othercase;
4197 if ((othercase = UCD_OTHERCASE(c)) != c)
4198 {
4199 *class_utf8data++ = XCL_SINGLE;
4200 class_utf8data += _pcre_ord2utf8(othercase, class_utf8data);
4201 }
4202 }
4203 #endif /* SUPPORT_UCP */
4204
4205 }
4206 else
4207 #endif /* SUPPORT_UTF8 */
4208
4209 /* Handle a single-byte character */
4210 {
4211 classbits[c/8] |= (1 << (c&7));
4212 if ((options & PCRE_CASELESS) != 0)
4213 {
4214 c = cd->fcc[c]; /* flip case */
4215 classbits[c/8] |= (1 << (c&7));
4216 }
4217 class_charcount++;
4218 class_lastchar = c;
4219 }
4220 }
4221
4222 /* Loop until ']' reached. This "while" is the end of the "do" far above.
4223 If we are at the end of an internal nested string, revert to the outer
4224 string. */
4225
4226 while (((c = *(++ptr)) != 0 ||
4227 (nestptr != NULL &&
4228 (ptr = nestptr, nestptr = NULL, c = *(++ptr)) != 0)) &&
4229 (c != CHAR_RIGHT_SQUARE_BRACKET || inescq));
4230
4231 /* Check for missing terminating ']' */
4232
4233 if (c == 0)
4234 {
4235 *errorcodeptr = ERR6;
4236 goto FAILED;
4237 }
4238
4239 /* If class_charcount is 1, we saw precisely one character whose value is
4240 less than 256. As long as there were no characters >= 128 and there was no
4241 use of \p or \P, in other words, no use of any XCLASS features, we can
4242 optimize.
4243
4244 In UTF-8 mode, we can optimize the negative case only if there were no
4245 characters >= 128 because OP_NOT and the related opcodes like OP_NOTSTAR
4246 operate on single-bytes characters only. This is an historical hangover.
4247 Maybe one day we can tidy these opcodes to handle multi-byte characters.
4248
4249 The optimization throws away the bit map. We turn the item into a
4250 1-character OP_CHAR[I] if it's positive, or OP_NOT[I] if it's negative.
4251 Note that OP_NOT[I] does not support multibyte characters. In the positive
4252 case, it can cause firstbyte to be set. Otherwise, there can be no first
4253 char if this item is first, whatever repeat count may follow. In the case
4254 of reqbyte, save the previous value for reinstating. */
4255
4256 #ifdef SUPPORT_UTF8
4257 if (class_charcount == 1 && !class_utf8 &&
4258 (!utf8 || !negate_class || class_lastchar < 128))
4259 #else
4260 if (class_charcount == 1)
4261 #endif
4262 {
4263 zeroreqbyte = reqbyte;
4264
4265 /* The OP_NOT[I] opcodes work on one-byte characters only. */
4266
4267 if (negate_class)
4268 {
4269 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
4270 zerofirstbyte = firstbyte;
4271 *code++ = ((options & PCRE_CASELESS) != 0)? OP_NOTI: OP_NOT;
4272 *code++ = class_lastchar;
4273 break;
4274 }
4275
4276 /* For a single, positive character, get the value into mcbuffer, and
4277 then we can handle this with the normal one-character code. */
4278
4279 #ifdef SUPPORT_UTF8
4280 if (utf8 && class_lastchar > 127)
4281 mclength = _pcre_ord2utf8(class_lastchar, mcbuffer);
4282 else
4283 #endif
4284 {
4285 mcbuffer[0] = class_lastchar;
4286 mclength = 1;
4287 }
4288 goto ONE_CHAR;
4289 } /* End of 1-char optimization */
4290
4291 /* The general case - not the one-char optimization. If this is the first
4292 thing in the branch, there can be no first char setting, whatever the
4293 repeat count. Any reqbyte setting must remain unchanged after any kind of
4294 repeat. */
4295
4296 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
4297 zerofirstbyte = firstbyte;
4298 zeroreqbyte = reqbyte;
4299
4300 /* If there are characters with values > 255, we have to compile an
4301 extended class, with its own opcode, unless there was a negated special
4302 such as \S in the class, and PCRE_UCP is not set, because in that case all
4303 characters > 255 are in the class, so any that were explicitly given as
4304 well can be ignored. If (when there are explicit characters > 255 that must
4305 be listed) there are no characters < 256, we can omit the bitmap in the
4306 actual compiled code. */
4307
4308 #ifdef SUPPORT_UTF8
4309 if (class_utf8 && (!should_flip_negation || (options & PCRE_UCP) != 0))
4310 {
4311 *class_utf8data++ = XCL_END; /* Marks the end of extra data */
4312 *code++ = OP_XCLASS;
4313 code += LINK_SIZE;
4314 *code = negate_class? XCL_NOT : 0;
4315
4316 /* If the map is required, move up the extra data to make room for it;
4317 otherwise just move the code pointer to the end of the extra data. */
4318
4319 if (class_charcount > 0)
4320 {
4321 *code++ |= XCL_MAP;
4322 memmove(code + 32, code, class_utf8data - code);
4323 memcpy(code, classbits, 32);
4324 code = class_utf8data + 32;
4325 }
4326 else code = class_utf8data;
4327
4328 /* Now fill in the complete length of the item */
4329
4330 PUT(previous, 1, code - previous);
4331 break; /* End of class handling */
4332 }
4333 #endif
4334
4335 /* If there are no characters > 255, or they are all to be included or
4336 excluded, set the opcode to OP_CLASS or OP_NCLASS, depending on whether the
4337 whole class was negated and whether there were negative specials such as \S
4338 (non-UCP) in the class. Then copy the 32-byte map into the code vector,
4339 negating it if necessary. */
4340
4341 *code++ = (negate_class == should_flip_negation) ? OP_CLASS : OP_NCLASS;
4342 if (negate_class)
4343 {
4344 if (lengthptr == NULL) /* Save time in the pre-compile phase */
4345 for (c = 0; c < 32; c++) code[c] = ~classbits[c];
4346 }
4347 else
4348 {
4349 memcpy(code, classbits, 32);
4350 }
4351 code += 32;
4352 break;
4353
4354
4355 /* ===================================================================*/
4356 /* Various kinds of repeat; '{' is not necessarily a quantifier, but this
4357 has been tested above. */
4358
4359 case CHAR_LEFT_CURLY_BRACKET:
4360 if (!is_quantifier) goto NORMAL_CHAR;
4361 ptr = read_repeat_counts(ptr+1, &repeat_min, &repeat_max, errorcodeptr);
4362 if (*errorcodeptr != 0) goto FAILED;
4363 goto REPEAT;
4364
4365 case CHAR_ASTERISK:
4366 repeat_min = 0;
4367 repeat_max = -1;
4368 goto REPEAT;
4369
4370 case CHAR_PLUS:
4371 repeat_min = 1;
4372 repeat_max = -1;
4373 goto REPEAT;
4374
4375 case CHAR_QUESTION_MARK:
4376 repeat_min = 0;
4377 repeat_max = 1;
4378
4379 REPEAT:
4380 if (previous == NULL)
4381 {
4382 *errorcodeptr = ERR9;
4383 goto FAILED;
4384 }
4385
4386 if (repeat_min == 0)
4387 {
4388 firstbyte = zerofirstbyte; /* Adjust for zero repeat */
4389 reqbyte = zeroreqbyte; /* Ditto */
4390 }
4391
4392 /* Remember whether this is a variable length repeat */
4393
4394 reqvary = (repeat_min == repeat_max)? 0 : REQ_VARY;
4395
4396 op_type = 0; /* Default single-char op codes */
4397 possessive_quantifier = FALSE; /* Default not possessive quantifier */
4398
4399 /* Save start of previous item, in case we have to move it up in order to
4400 insert something before it. */
4401
4402 tempcode = previous;
4403
4404 /* If the next character is '+', we have a possessive quantifier. This
4405 implies greediness, whatever the setting of the PCRE_UNGREEDY option.
4406 If the next character is '?' this is a minimizing repeat, by default,
4407 but if PCRE_UNGREEDY is set, it works the other way round. We change the
4408 repeat type to the non-default. */
4409
4410 if (ptr[1] == CHAR_PLUS)
4411 {
4412 repeat_type = 0; /* Force greedy */
4413 possessive_quantifier = TRUE;
4414 ptr++;
4415 }
4416 else if (ptr[1] == CHAR_QUESTION_MARK)
4417 {
4418 repeat_type = greedy_non_default;
4419 ptr++;
4420 }
4421 else repeat_type = greedy_default;
4422
4423 /* If previous was a recursion call, wrap it in atomic brackets so that
4424 previous becomes the atomic group. All recursions were so wrapped in the
4425 past, but it no longer happens for non-repeated recursions. In fact, the
4426 repeated ones could be re-implemented independently so as not to need this,
4427 but for the moment we rely on the code for repeating groups. */
4428
4429 if (*previous == OP_RECURSE)
4430 {
4431 memmove(previous + 1 + LINK_SIZE, previous, 1 + LINK_SIZE);
4432 *previous = OP_ONCE;
4433 PUT(previous, 1, 2 + 2*LINK_SIZE);
4434 previous[2 + 2*LINK_SIZE] = OP_KET;
4435 PUT(previous, 3 + 2*LINK_SIZE, 2 + 2*LINK_SIZE);
4436 code += 2 + 2 * LINK_SIZE;
4437 length_prevgroup = 3 + 3*LINK_SIZE;
4438
4439 /* When actually compiling, we need to check whether this was a forward
4440 reference, and if so, adjust the offset. */
4441
4442 if (lengthptr == NULL && cd->hwm >= cd->start_workspace + LINK_SIZE)
4443 {
4444 int offset = GET(cd->hwm, -LINK_SIZE);
4445 if (offset == previous + 1 - cd->start_code)
4446 PUT(cd->hwm, -LINK_SIZE, offset + 1 + LINK_SIZE);
4447 }
4448 }
4449
4450 /* Now handle repetition for the different types of item. */
4451
4452 /* If previous was a character match, abolish the item and generate a
4453 repeat item instead. If a char item has a minumum of more than one, ensure
4454 that it is set in reqbyte - it might not be if a sequence such as x{3} is
4455 the first thing in a branch because the x will have gone into firstbyte
4456 instead. */
4457
4458 if (*previous == OP_CHAR || *previous == OP_CHARI)
4459 {
4460 op_type = (*previous == OP_CHAR)? 0 : OP_STARI - OP_STAR;
4461
4462 /* Deal with UTF-8 characters that take up more than one byte. It's
4463 easier to write this out separately than try to macrify it. Use c to
4464 hold the length of the character in bytes, plus 0x80 to flag that it's a
4465 length rather than a small character. */
4466
4467 #ifdef SUPPORT_UTF8
4468 if (utf8 && (code[-1] & 0x80) != 0)
4469 {
4470 uschar *lastchar = code - 1;
4471 while((*lastchar & 0xc0) == 0x80) lastchar--;
4472 c = code - lastchar; /* Length of UTF-8 character */
4473 memcpy(utf8_char, lastchar, c); /* Save the char */
4474 c |= 0x80; /* Flag c as a length */
4475 }
4476 else
4477 #endif
4478
4479 /* Handle the case of a single byte - either with no UTF8 support, or
4480 with UTF-8 disabled, or for a UTF-8 character < 128. */
4481
4482 {
4483 c = code[-1];
4484 if (repeat_min > 1) reqbyte = c | req_caseopt | cd->req_varyopt;
4485 }
4486
4487 /* If the repetition is unlimited, it pays to see if the next thing on
4488 the line is something that cannot possibly match this character. If so,
4489 automatically possessifying this item gains some performance in the case
4490 where the match fails. */
4491
4492 if (!possessive_quantifier &&
4493 repeat_max < 0 &&
4494 check_auto_possessive(previous, utf8, ptr + 1, options, cd))
4495 {
4496 repeat_type = 0; /* Force greedy */
4497 possessive_quantifier = TRUE;
4498 }
4499
4500 goto OUTPUT_SINGLE_REPEAT; /* Code shared with single character types */
4501 }
4502
4503 /* If previous was a single negated character ([^a] or similar), we use
4504 one of the special opcodes, replacing it. The code is shared with single-
4505 character repeats by setting opt_type to add a suitable offset into
4506 repeat_type. We can also test for auto-possessification. OP_NOT and OP_NOTI
4507 are currently used only for single-byte chars. */
4508
4509 else if (*previous == OP_NOT || *previous == OP_NOTI)
4510 {
4511 op_type = ((*previous == OP_NOT)? OP_NOTSTAR : OP_NOTSTARI) - OP_STAR;
4512 c = previous[1];
4513 if (!possessive_quantifier &&
4514 repeat_max < 0 &&
4515 check_auto_possessive(previous, utf8, ptr + 1, options, cd))
4516 {
4517 repeat_type = 0; /* Force greedy */
4518 possessive_quantifier = TRUE;
4519 }
4520 goto OUTPUT_SINGLE_REPEAT;
4521 }
4522
4523 /* If previous was a character type match (\d or similar), abolish it and
4524 create a suitable repeat item. The code is shared with single-character
4525 repeats by setting op_type to add a suitable offset into repeat_type. Note
4526 the the Unicode property types will be present only when SUPPORT_UCP is
4527 defined, but we don't wrap the little bits of code here because it just
4528 makes it horribly messy. */
4529
4530 else if (*previous < OP_EODN)
4531 {
4532 uschar *oldcode;
4533 int prop_type, prop_value;
4534 op_type = OP_TYPESTAR - OP_STAR; /* Use type opcodes */
4535 c = *previous;
4536
4537 if (!possessive_quantifier &&
4538 repeat_max < 0 &&
4539 check_auto_possessive(previous, utf8, ptr + 1, options, cd))
4540 {
4541 repeat_type = 0; /* Force greedy */
4542 possessive_quantifier = TRUE;
4543 }
4544
4545 OUTPUT_SINGLE_REPEAT:
4546 if (*previous == OP_PROP || *previous == OP_NOTPROP)
4547 {
4548 prop_type = previous[1];
4549 prop_value = previous[2];
4550 }
4551 else prop_type = prop_value = -1;
4552
4553 oldcode = code;
4554 code = previous; /* Usually overwrite previous item */
4555
4556 /* If the maximum is zero then the minimum must also be zero; Perl allows
4557 this case, so we do too - by simply omitting the item altogether. */
4558
4559 if (repeat_max == 0) goto END_REPEAT;
4560
4561 /*--------------------------------------------------------------------*/
4562 /* This code is obsolete from release 8.00; the restriction was finally
4563 removed: */
4564
4565 /* All real repeats make it impossible to handle partial matching (maybe
4566 one day we will be able to remove this restriction). */
4567
4568 /* if (repeat_max != 1) cd->external_flags |= PCRE_NOPARTIAL; */
4569 /*--------------------------------------------------------------------*/
4570
4571 /* Combine the op_type with the repeat_type */
4572
4573 repeat_type += op_type;
4574
4575 /* A minimum of zero is handled either as the special case * or ?, or as
4576 an UPTO, with the maximum given. */
4577
4578 if (repeat_min == 0)
4579 {
4580 if (repeat_max == -1) *code++ = OP_STAR + repeat_type;
4581 else if (repeat_max == 1) *code++ = OP_QUERY + repeat_type;
4582 else
4583 {
4584 *code++ = OP_UPTO + repeat_type;
4585 PUT2INC(code, 0, repeat_max);
4586 }
4587 }
4588
4589 /* A repeat minimum of 1 is optimized into some special cases. If the
4590 maximum is unlimited, we use OP_PLUS. Otherwise, the original item is
4591 left in place and, if the maximum is greater than 1, we use OP_UPTO with
4592 one less than the maximum. */
4593
4594 else if (repeat_min == 1)
4595 {
4596 if (repeat_max == -1)
4597 *code++ = OP_PLUS + repeat_type;
4598 else
4599 {
4600 code = oldcode; /* leave previous item in place */
4601 if (repeat_max == 1) goto END_REPEAT;
4602 *code++ = OP_UPTO + repeat_type;
4603 PUT2INC(code, 0, repeat_max - 1);
4604 }
4605 }
4606
4607 /* The case {n,n} is just an EXACT, while the general case {n,m} is
4608 handled as an EXACT followed by an UPTO. */
4609
4610 else
4611 {
4612 *code++ = OP_EXACT + op_type; /* NB EXACT doesn't have repeat_type */
4613 PUT2INC(code, 0, repeat_min);
4614
4615 /* If the maximum is unlimited, insert an OP_STAR. Before doing so,
4616 we have to insert the character for the previous code. For a repeated
4617 Unicode property match, there are two extra bytes that define the
4618 required property. In UTF-8 mode, long characters have their length in
4619 c, with the 0x80 bit as a flag. */
4620
4621 if (repeat_max < 0)
4622 {
4623 #ifdef SUPPORT_UTF8
4624 if (utf8 && c >= 128)
4625 {
4626 memcpy(code, utf8_char, c & 7);
4627 code += c & 7;
4628 }
4629 else
4630 #endif
4631 {
4632 *code++ = c;
4633 if (prop_type >= 0)
4634 {
4635 *code++ = prop_type;
4636 *code++ = prop_value;
4637 }
4638 }
4639 *code++ = OP_STAR + repeat_type;
4640 }
4641
4642 /* Else insert an UPTO if the max is greater than the min, again
4643 preceded by the character, for the previously inserted code. If the
4644 UPTO is just for 1 instance, we can use QUERY instead. */
4645
4646 else if (repeat_max != repeat_min)
4647 {
4648 #ifdef SUPPORT_UTF8
4649 if (utf8 && c >= 128)
4650 {
4651 memcpy(code, utf8_char, c & 7);
4652 code += c & 7;
4653 }
4654 else
4655 #endif
4656 *code++ = c;
4657 if (prop_type >= 0)
4658 {
4659 *code++ = prop_type;
4660 *code++ = prop_value;
4661 }
4662 repeat_max -= repeat_min;
4663
4664 if (repeat_max == 1)
4665 {
4666 *code++ = OP_QUERY + repeat_type;
4667 }
4668 else
4669 {
4670 *code++ = OP_UPTO + repeat_type;
4671 PUT2INC(code, 0, repeat_max);
4672 }
4673 }
4674 }
4675
4676 /* The character or character type itself comes last in all cases. */
4677
4678 #ifdef SUPPORT_UTF8
4679 if (utf8 && c >= 128)
4680 {
4681 memcpy(code, utf8_char, c & 7);
4682 code += c & 7;
4683 }
4684 else
4685 #endif
4686 *code++ = c;
4687
4688 /* For a repeated Unicode property match, there are two extra bytes that
4689 define the required property. */
4690
4691 #ifdef SUPPORT_UCP
4692 if (prop_type >= 0)
4693 {
4694 *code++ = prop_type;
4695 *code++ = prop_value;
4696 }
4697 #endif
4698 }
4699
4700 /* If previous was a character class or a back reference, we put the repeat
4701 stuff after it, but just skip the item if the repeat was {0,0}. */
4702
4703 else if (*previous == OP_CLASS ||
4704 *previous == OP_NCLASS ||
4705 #ifdef SUPPORT_UTF8
4706 *previous == OP_XCLASS ||
4707 #endif
4708 *previous == OP_REF ||
4709 *previous == OP_REFI)
4710 {
4711 if (repeat_max == 0)
4712 {
4713 code = previous;
4714 goto END_REPEAT;
4715 }
4716
4717 /*--------------------------------------------------------------------*/
4718 /* This code is obsolete from release 8.00; the restriction was finally
4719 removed: */
4720
4721 /* All real repeats make it impossible to handle partial matching (maybe
4722 one day we will be able to remove this restriction). */
4723
4724 /* if (repeat_max != 1) cd->external_flags |= PCRE_NOPARTIAL; */
4725 /*--------------------------------------------------------------------*/
4726
4727 if (repeat_min == 0 && repeat_max == -1)
4728 *code++ = OP_CRSTAR + repeat_type;
4729 else if (repeat_min == 1 && repeat_max == -1)
4730 *code++ = OP_CRPLUS + repeat_type;
4731 else if (repeat_min == 0 && repeat_max == 1)
4732 *code++ = OP_CRQUERY + repeat_type;
4733 else
4734 {
4735 *code++ = OP_CRRANGE + repeat_type;
4736 PUT2INC(code, 0, repeat_min);
4737 if (repeat_max == -1) repeat_max = 0; /* 2-byte encoding for max */
4738 PUT2INC(code, 0, repeat_max);
4739 }
4740 }
4741
4742 /* If previous was a bracket group, we may have to replicate it in certain
4743 cases. Note that at this point we can encounter only the "basic" bracket
4744 opcodes such as BRA and CBRA, as this is the place where they get converted
4745 into the more special varieties such as BRAPOS and SBRA. A test for >=
4746 OP_ASSERT and <= OP_COND includes ASSERT, ASSERT_NOT, ASSERTBACK,
4747 ASSERTBACK_NOT, ONCE, BRA, CBRA, and COND. Originally, PCRE did not allow
4748 repetition of assertions, but now it does, for Perl compatibility. */
4749
4750 else if (*previous >= OP_ASSERT && *previous <= OP_COND)
4751 {
4752 register int i;
4753 int len = (int)(code - previous);
4754 uschar *bralink = NULL;
4755 uschar *brazeroptr = NULL;
4756
4757 /* Repeating a DEFINE group is pointless, but Perl allows the syntax, so
4758 we just ignore the repeat. */
4759
4760 if (*previous == OP_COND && previous[LINK_SIZE+1] == OP_DEF)
4761 goto END_REPEAT;
4762
4763 /* There is no sense in actually repeating assertions. The only potential
4764 use of repetition is in cases when the assertion is optional. Therefore,
4765 if the minimum is greater than zero, just ignore the repeat. If the
4766 maximum is not not zero or one, set it to 1. */
4767
4768 if (*previous < OP_ONCE) /* Assertion */
4769 {
4770 if (repeat_min > 0) goto END_REPEAT;
4771 if (repeat_max < 0 || repeat_max > 1) repeat_max = 1;
4772 }
4773
4774 /* The case of a zero minimum is special because of the need to stick
4775 OP_BRAZERO in front of it, and because the group appears once in the
4776 data, whereas in other cases it appears the minimum number of times. For
4777 this reason, it is simplest to treat this case separately, as otherwise
4778 the code gets far too messy. There are several special subcases when the
4779 minimum is zero. */
4780
4781 if (repeat_min == 0)
4782 {
4783 /* If the maximum is also zero, we used to just omit the group from the
4784 output altogether, like this:
4785
4786 ** if (repeat_max == 0)
4787 ** {
4788 ** code = previous;
4789 ** goto END_REPEAT;
4790 ** }
4791
4792 However, that fails when a group or a subgroup within it is referenced
4793 as a subroutine from elsewhere in the pattern, so now we stick in
4794 OP_SKIPZERO in front of it so that it is skipped on execution. As we
4795 don't have a list of which groups are referenced, we cannot do this
4796 selectively.
4797
4798 If the maximum is 1 or unlimited, we just have to stick in the BRAZERO
4799 and do no more at this point. However, we do need to adjust any
4800 OP_RECURSE calls inside the group that refer to the group itself or any
4801 internal or forward referenced group, because the offset is from the
4802 start of the whole regex. Temporarily terminate the pattern while doing
4803 this. */
4804
4805 if (repeat_max <= 1) /* Covers 0, 1, and unlimited */
4806 {
4807 *code = OP_END;
4808 adjust_recurse(previous, 1, utf8, cd, save_hwm);
4809 memmove(previous+1, previous, len);
4810 code++;
4811 if (repeat_max == 0)
4812 {
4813 *previous++ = OP_SKIPZERO;
4814 goto END_REPEAT;
4815 }
4816 brazeroptr = previous; /* Save for possessive optimizing */
4817 *previous++ = OP_BRAZERO + repeat_type;
4818 }
4819
4820 /* If the maximum is greater than 1 and limited, we have to replicate
4821 in a nested fashion, sticking OP_BRAZERO before each set of brackets.
4822 The first one has to be handled carefully because it's the original
4823 copy, which has to be moved up. The remainder can be handled by code
4824 that is common with the non-zero minimum case below. We have to
4825 adjust the value or repeat_max, since one less copy is required. Once
4826 again, we may have to adjust any OP_RECURSE calls inside the group. */
4827
4828 else
4829 {
4830 int offset;
4831 *code = OP_END;
4832 adjust_recurse(previous, 2 + LINK_SIZE, utf8, cd, save_hwm);
4833 memmove(previous + 2 + LINK_SIZE, previous, len);
4834 code += 2 + LINK_SIZE;
4835 *previous++ = OP_BRAZERO + repeat_type;
4836 *previous++ = OP_BRA;
4837
4838 /* We chain together the bracket offset fields that have to be
4839 filled in later when the ends of the brackets are reached. */
4840
4841 offset = (bralink == NULL)? 0 : (int)(previous - bralink);
4842 bralink = previous;
4843 PUTINC(previous, 0, offset);
4844 }
4845
4846 repeat_max--;
4847 }
4848
4849 /* If the minimum is greater than zero, replicate the group as many
4850 times as necessary, and adjust the maximum to the number of subsequent
4851 copies that we need. If we set a first char from the group, and didn't
4852 set a required char, copy the latter from the former. If there are any
4853 forward reference subroutine calls in the group, there will be entries on
4854 the workspace list; replicate these with an appropriate increment. */
4855
4856 else
4857 {
4858 if (repeat_min > 1)
4859 {
4860 /* In the pre-compile phase, we don't actually do the replication. We
4861 just adjust the length as if we had. Do some paranoid checks for
4862 potential integer overflow. The INT64_OR_DOUBLE type is a 64-bit
4863 integer type when available, otherwise double. */
4864
4865 if (lengthptr != NULL)
4866 {
4867 int delta = (repeat_min - 1)*length_prevgroup;
4868 if ((INT64_OR_DOUBLE)(repeat_min - 1)*
4869 (INT64_OR_DOUBLE)length_prevgroup >
4870 (INT64_OR_DOUBLE)INT_MAX ||
4871 OFLOW_MAX - *lengthptr < delta)
4872 {
4873 *errorcodeptr = ERR20;
4874 goto FAILED;
4875 }
4876 *lengthptr += delta;
4877 }
4878
4879 /* This is compiling for real */
4880
4881 else
4882 {
4883 if (groupsetfirstbyte && reqbyte < 0) reqbyte = firstbyte;
4884 for (i = 1; i < repeat_min; i++)
4885 {
4886 uschar *hc;
4887 uschar *this_hwm = cd->hwm;
4888 memcpy(code, previous, len);
4889 for (hc = save_hwm; hc < this_hwm; hc += LINK_SIZE)
4890 {
4891 PUT(cd->hwm, 0, GET(hc, 0) + len);
4892 cd->hwm += LINK_SIZE;
4893 }
4894 save_hwm = this_hwm;
4895 code += len;
4896 }
4897 }
4898 }
4899
4900 if (repeat_max > 0) repeat_max -= repeat_min;
4901 }
4902
4903 /* This code is common to both the zero and non-zero minimum cases. If
4904 the maximum is limited, it replicates the group in a nested fashion,
4905 remembering the bracket starts on a stack. In the case of a zero minimum,
4906 the first one was set up above. In all cases the repeat_max now specifies
4907 the number of additional copies needed. Again, we must remember to
4908 replicate entries on the forward reference list. */
4909
4910 if (repeat_max >= 0)
4911 {
4912 /* In the pre-compile phase, we don't actually do the replication. We
4913 just adjust the length as if we had. For each repetition we must add 1
4914 to the length for BRAZERO and for all but the last repetition we must
4915 add 2 + 2*LINKSIZE to allow for the nesting that occurs. Do some
4916 paranoid checks to avoid integer overflow. The INT64_OR_DOUBLE type is
4917 a 64-bit integer type when available, otherwise double. */
4918
4919 if (lengthptr != NULL && repeat_max > 0)
4920 {
4921 int delta = repeat_max * (length_prevgroup + 1 + 2 + 2*LINK_SIZE) -
4922 2 - 2*LINK_SIZE; /* Last one doesn't nest */
4923 if ((INT64_OR_DOUBLE)repeat_max *
4924 (INT64_OR_DOUBLE)(length_prevgroup + 1 + 2 + 2*LINK_SIZE)
4925 > (INT64_OR_DOUBLE)INT_MAX ||
4926 OFLOW_MAX - *lengthptr < delta)
4927 {
4928 *errorcodeptr = ERR20;
4929 goto FAILED;
4930 }
4931 *lengthptr += delta;
4932 }
4933
4934 /* This is compiling for real */
4935
4936 else for (i = repeat_max - 1; i >= 0; i--)
4937 {
4938 uschar *hc;
4939 uschar *this_hwm = cd->hwm;
4940
4941 *code++ = OP_BRAZERO + repeat_type;
4942
4943 /* All but the final copy start a new nesting, maintaining the
4944 chain of brackets outstanding. */
4945
4946 if (i != 0)
4947 {
4948 int offset;
4949 *code++ = OP_BRA;
4950 offset = (bralink == NULL)? 0 : (int)(code - bralink);
4951 bralink = code;
4952 PUTINC(code, 0, offset);
4953 }
4954
4955 memcpy(code, previous, len);
4956 for (hc = save_hwm; hc < this_hwm; hc += LINK_SIZE)
4957 {
4958 PUT(cd->hwm, 0, GET(hc, 0) + len + ((i != 0)? 2+LINK_SIZE : 1));
4959 cd->hwm += LINK_SIZE;
4960 }
4961 save_hwm = this_hwm;
4962 code += len;
4963 }
4964
4965 /* Now chain through the pending brackets, and fill in their length
4966 fields (which are holding the chain links pro tem). */
4967
4968 while (bralink != NULL)
4969 {
4970 int oldlinkoffset;
4971 int offset = (int)(code - bralink + 1);
4972 uschar *bra = code - offset;
4973 oldlinkoffset = GET(bra, 1);
4974 bralink = (oldlinkoffset == 0)? NULL : bralink - oldlinkoffset;
4975 *code++ = OP_KET;
4976 PUTINC(code, 0, offset);
4977 PUT(bra, 1, offset);
4978 }
4979 }
4980
4981 /* If the maximum is unlimited, set a repeater in the final copy. For
4982 ONCE brackets, that's all we need to do. However, possessively repeated
4983 ONCE brackets can be converted into non-capturing brackets, as the
4984 behaviour of (?:xx)++ is the same as (?>xx)++ and this saves having to
4985 deal with possessive ONCEs specially.
4986
4987 Otherwise, when we are doing the actual compile phase, check to see
4988 whether this group is one that could match an empty string. If so,
4989 convert the initial operator to the S form (e.g. OP_BRA -> OP_SBRA) so
4990 that runtime checking can be done. [This check is also applied to ONCE
4991 groups at runtime, but in a different way.]
4992
4993 Then, if the quantifier was possessive and the bracket is not a
4994 conditional, we convert the BRA code to the POS form, and the KET code to
4995 KETRPOS. (It turns out to be convenient at runtime to detect this kind of
4996 subpattern at both the start and at the end.) The use of special opcodes
4997 makes it possible to reduce greatly the stack usage in pcre_exec(). If
4998 the group is preceded by OP_BRAZERO, convert this to OP_BRAPOSZERO.
4999
5000 Then, if the minimum number of matches is 1 or 0, cancel the possessive
5001 flag so that the default action below, of wrapping everything inside
5002 atomic brackets, does not happen. When the minimum is greater than 1,
5003 there will be earlier copies of the group, and so we still have to wrap
5004 the whole thing. */
5005
5006 else
5007 {
5008 uschar *ketcode = code - 1 - LINK_SIZE;
5009 uschar *bracode = ketcode - GET(ketcode, 1);
5010
5011 /* Convert possessive ONCE brackets to non-capturing */
5012
5013 if ((*bracode == OP_ONCE || *bracode == OP_ONCE_NC) &&
5014 possessive_quantifier) *bracode = OP_BRA;
5015
5016 /* For non-possessive ONCE brackets, all we need to do is to
5017 set the KET. */
5018
5019 if (*bracode == OP_ONCE || *bracode == OP_ONCE_NC)
5020 *ketcode = OP_KETRMAX + repeat_type;
5021
5022 /* Handle non-ONCE brackets and possessive ONCEs (which have been
5023 converted to non-capturing above). */
5024
5025 else
5026 {
5027 /* In the compile phase, check for empty string matching. */
5028
5029 if (lengthptr == NULL)
5030 {
5031 uschar *scode = bracode;
5032 do
5033 {
5034 if (could_be_empty_branch(scode, ketcode, utf8, cd))
5035 {
5036 *bracode += OP_SBRA - OP_BRA;
5037 break;
5038 }
5039 scode += GET(scode, 1);
5040 }
5041 while (*scode == OP_ALT);
5042 }
5043
5044 /* Handle possessive quantifiers. */
5045
5046 if (possessive_quantifier)
5047 {
5048 /* For COND brackets, we wrap the whole thing in a possessively
5049 repeated non-capturing bracket, because we have not invented POS
5050 versions of the COND opcodes. Because we are moving code along, we
5051 must ensure that any pending recursive references are updated. */
5052
5053 if (*bracode == OP_COND || *bracode == OP_SCOND)
5054 {
5055 int nlen = (int)(code - bracode);
5056 *code = OP_END;
5057 adjust_recurse(bracode, 1 + LINK_SIZE, utf8, cd, save_hwm);
5058 memmove(bracode + 1+LINK_SIZE, bracode, nlen);
5059 code += 1 + LINK_SIZE;
5060 nlen += 1 + LINK_SIZE;
5061 *bracode = OP_BRAPOS;
5062 *code++ = OP_KETRPOS;
5063 PUTINC(code, 0, nlen);
5064 PUT(bracode, 1, nlen);
5065 }
5066
5067 /* For non-COND brackets, we modify the BRA code and use KETRPOS. */
5068
5069 else
5070 {
5071 *bracode += 1; /* Switch to xxxPOS opcodes */
5072 *ketcode = OP_KETRPOS;
5073 }
5074
5075 /* If the minimum is zero, mark it as possessive, then unset the
5076 possessive flag when the minimum is 0 or 1. */
5077
5078 if (brazeroptr != NULL) *brazeroptr = OP_BRAPOSZERO;
5079 if (repeat_min < 2) possessive_quantifier = FALSE;
5080 }
5081
5082 /* Non-possessive quantifier */
5083
5084 else *ketcode = OP_KETRMAX + repeat_type;
5085 }
5086 }
5087 }
5088
5089 /* If previous is OP_FAIL, it was generated by an empty class [] in
5090 JavaScript mode. The other ways in which OP_FAIL can be generated, that is
5091 by (*FAIL) or (?!) set previous to NULL, which gives a "nothing to repeat"
5092 error above. We can just ignore the repeat in JS case. */
5093
5094 else if (*previous == OP_FAIL) goto END_REPEAT;
5095
5096 /* Else there's some kind of shambles */
5097
5098 else
5099 {
5100 *errorcodeptr = ERR11;
5101 goto FAILED;
5102 }
5103
5104 /* If the character following a repeat is '+', or if certain optimization
5105 tests above succeeded, possessive_quantifier is TRUE. For some opcodes,
5106 there are special alternative opcodes for this case. For anything else, we
5107 wrap the entire repeated item inside OP_ONCE brackets. Logically, the '+'
5108 notation is just syntactic sugar, taken from Sun's Java package, but the
5109 special opcodes can optimize it.
5110
5111 Some (but not all) possessively repeated subpatterns have already been
5112 completely handled in the code just above. For them, possessive_quantifier
5113 is always FALSE at this stage.
5114
5115 Note that the repeated item starts at tempcode, not at previous, which
5116 might be the first part of a string whose (former) last char we repeated.
5117
5118 Possessifying an 'exact' quantifier has no effect, so we can ignore it. But
5119 an 'upto' may follow. We skip over an 'exact' item, and then test the
5120 length of what remains before proceeding. */
5121
5122 if (possessive_quantifier)
5123 {
5124 int len;
5125
5126 if (*tempcode == OP_TYPEEXACT)
5127 tempcode += _pcre_OP_lengths[*tempcode] +
5128 ((tempcode[3] == OP_PROP || tempcode[3] == OP_NOTPROP)? 2 : 0);
5129
5130 else if (*tempcode == OP_EXACT || *tempcode == OP_NOTEXACT)
5131 {
5132 tempcode += _pcre_OP_lengths[*tempcode];
5133 #ifdef SUPPORT_UTF8
5134 if (utf8 && tempcode[-1] >= 0xc0)
5135 tempcode += _pcre_utf8_table4[tempcode[-1] & 0x3f];
5136 #endif
5137 }
5138
5139 len = (int)(code - tempcode);
5140 if (len > 0) switch (*tempcode)
5141 {
5142 case OP_STAR: *tempcode = OP_POSSTAR; break;
5143 case OP_PLUS: *tempcode = OP_POSPLUS; break;
5144 case OP_QUERY: *tempcode = OP_POSQUERY; break;
5145 case OP_UPTO: *tempcode = OP_POSUPTO; break;
5146
5147 case OP_STARI: *tempcode = OP_POSSTARI; break;
5148 case OP_PLUSI: *tempcode = OP_POSPLUSI; break;
5149 case OP_QUERYI: *tempcode = OP_POSQUERYI; break;
5150 case OP_UPTOI: *tempcode = OP_POSUPTOI; break;
5151
5152 case OP_NOTSTAR: *tempcode = OP_NOTPOSSTAR; break;
5153 case OP_NOTPLUS: *tempcode = OP_NOTPOSPLUS; break;
5154 case OP_NOTQUERY: *tempcode = OP_NOTPOSQUERY; break;
5155 case OP_NOTUPTO: *tempcode = OP_NOTPOSUPTO; break;
5156
5157 case OP_NOTSTARI: *tempcode = OP_NOTPOSSTARI; break;
5158 case OP_NOTPLUSI: *tempcode = OP_NOTPOSPLUSI; break;
5159 case OP_NOTQUERYI: *tempcode = OP_NOTPOSQUERYI; break;
5160 case OP_NOTUPTOI: *tempcode = OP_NOTPOSUPTOI; break;
5161
5162 case OP_TYPESTAR: *tempcode = OP_TYPEPOSSTAR; break;
5163 case OP_TYPEPLUS: *tempcode = OP_TYPEPOSPLUS; break;
5164 case OP_TYPEQUERY: *tempcode = OP_TYPEPOSQUERY; break;
5165 case OP_TYPEUPTO: *tempcode = OP_TYPEPOSUPTO; break;
5166
5167 /* Because we are moving code along, we must ensure that any
5168 pending recursive references are updated. */
5169
5170 default:
5171 *code = OP_END;
5172 adjust_recurse(tempcode, 1 + LINK_SIZE, utf8, cd, save_hwm);
5173 memmove(tempcode + 1+LINK_SIZE, tempcode, len);
5174 code += 1 + LINK_SIZE;
5175 len += 1 + LINK_SIZE;
5176 tempcode[0] = OP_ONCE;
5177 *code++ = OP_KET;
5178 PUTINC(code, 0, len);
5179 PUT(tempcode, 1, len);
5180 break;
5181 }
5182 }
5183
5184 /* In all case we no longer have a previous item. We also set the
5185 "follows varying string" flag for subsequently encountered reqbytes if
5186 it isn't already set and we have just passed a varying length item. */
5187
5188 END_REPEAT:
5189 previous = NULL;
5190 cd->req_varyopt |= reqvary;
5191 break;
5192
5193
5194 /* ===================================================================*/
5195 /* Start of nested parenthesized sub-expression, or comment or lookahead or
5196 lookbehind or option setting or condition or all the other extended
5197 parenthesis forms. */
5198
5199 case CHAR_LEFT_PARENTHESIS:
5200 newoptions = options;
5201 skipbytes = 0;
5202 bravalue = OP_CBRA;
5203 save_hwm = cd->hwm;
5204 reset_bracount = FALSE;
5205
5206 /* First deal with various "verbs" that can be introduced by '*'. */
5207
5208 if (*(++ptr) == CHAR_ASTERISK &&
5209 ((cd->ctypes[ptr[1]] & ctype_letter) != 0 || ptr[1] == ':'))
5210 {
5211 int i, namelen;
5212 int arglen = 0;
5213 const char *vn = verbnames;
5214 const uschar *name = ptr + 1;
5215 const uschar *arg = NULL;
5216 previous = NULL;
5217 while ((cd->ctypes[*++ptr] & ctype_letter) != 0) {};
5218 namelen = (int)(ptr - name);
5219
5220 /* It appears that Perl allows any characters whatsoever, other than
5221 a closing parenthesis, to appear in arguments, so we no longer insist on
5222 letters, digits, and underscores. */
5223
5224 if (*ptr == CHAR_COLON)
5225 {
5226 arg = ++ptr;
5227 while (*ptr != 0 && *ptr != CHAR_RIGHT_PARENTHESIS) ptr++;
5228 arglen = (int)(ptr - arg);
5229 }
5230
5231 if (*ptr != CHAR_RIGHT_PARENTHESIS)
5232 {
5233 *errorcodeptr = ERR60;
5234 goto FAILED;
5235 }
5236
5237 /* Scan the table of verb names */
5238
5239 for (i = 0; i < verbcount; i++)
5240 {
5241 if (namelen == verbs[i].len &&
5242 strncmp((char *)name, vn, namelen) == 0)
5243 {
5244 /* Check for open captures before ACCEPT and convert it to
5245 ASSERT_ACCEPT if in an assertion. */
5246
5247 if (verbs[i].op == OP_ACCEPT)
5248 {
5249 open_capitem *oc;
5250 if (arglen != 0)
5251 {
5252 *errorcodeptr = ERR59;
5253 goto FAILED;
5254 }
5255 cd->had_accept = TRUE;
5256 for (oc = cd->open_caps; oc != NULL; oc = oc->next)
5257 {
5258 *code++ = OP_CLOSE;
5259 PUT2INC(code, 0, oc->number);
5260 }
5261 *code++ = (cd->assert_depth > 0)? OP_ASSERT_ACCEPT : OP_ACCEPT;
5262
5263 /* Do not set firstbyte after *ACCEPT */
5264 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
5265 }
5266
5267 /* Handle other cases with/without an argument */
5268
5269 else if (arglen == 0)
5270 {
5271 if (verbs[i].op < 0) /* Argument is mandatory */
5272 {
5273 *errorcodeptr = ERR66;
5274 goto FAILED;
5275 }
5276 *code = verbs[i].op;
5277 if (*code++ == OP_THEN) cd->external_flags |= PCRE_HASTHEN;
5278 }
5279
5280 else
5281 {
5282 if (verbs[i].op_arg < 0) /* Argument is forbidden */
5283 {
5284 *errorcodeptr = ERR59;
5285 goto FAILED;
5286 }
5287 *code = verbs[i].op_arg;
5288 if (*code++ == OP_THEN_ARG) cd->external_flags |= PCRE_HASTHEN;
5289 *code++ = arglen;
5290 memcpy(code, arg, arglen);
5291 code += arglen;
5292 *code++ = 0;
5293 }
5294
5295 break; /* Found verb, exit loop */
5296 }
5297
5298 vn += verbs[i].len + 1;
5299 }
5300
5301 if (i < verbcount) continue; /* Successfully handled a verb */
5302 *errorcodeptr = ERR60; /* Verb not recognized */
5303 goto FAILED;
5304 }
5305
5306 /* Deal with the extended parentheses; all are introduced by '?', and the
5307 appearance of any of them means that this is not a capturing group. */
5308
5309 else if (*ptr == CHAR_QUESTION_MARK)
5310 {
5311 int i, set, unset, namelen;
5312 int *optset;
5313 const uschar *name;
5314 uschar *slot;
5315
5316 switch (*(++ptr))
5317 {
5318 case CHAR_NUMBER_SIGN: /* Comment; skip to ket */
5319 ptr++;
5320 while (*ptr != 0 && *ptr != CHAR_RIGHT_PARENTHESIS) ptr++;
5321 if (*ptr == 0)
5322 {
5323 *errorcodeptr = ERR18;
5324 goto FAILED;
5325 }
5326 continue;
5327
5328
5329 /* ------------------------------------------------------------ */
5330 case CHAR_VERTICAL_LINE: /* Reset capture count for each branch */
5331 reset_bracount = TRUE;
5332 /* Fall through */
5333
5334 /* ------------------------------------------------------------ */
5335 case CHAR_COLON: /* Non-capturing bracket */
5336 bravalue = OP_BRA;
5337 ptr++;
5338 break;
5339
5340
5341 /* ------------------------------------------------------------ */
5342 case CHAR_LEFT_PARENTHESIS:
5343 bravalue = OP_COND; /* Conditional group */
5344
5345 /* A condition can be an assertion, a number (referring to a numbered
5346 group), a name (referring to a named group), or 'R', referring to
5347 recursion. R<digits> and R&name are also permitted for recursion tests.
5348
5349 There are several syntaxes for testing a named group: (?(name)) is used
5350 by Python; Perl 5.10 onwards uses (?(<name>) or (?('name')).
5351
5352 There are two unfortunate ambiguities, caused by history. (a) 'R' can
5353 be the recursive thing or the name 'R' (and similarly for 'R' followed
5354 by digits), and (b) a number could be a name that consists of digits.
5355 In both cases, we look for a name first; if not found, we try the other
5356 cases. */
5357
5358 /* For conditions that are assertions, check the syntax, and then exit
5359 the switch. This will take control down to where bracketed groups,
5360 including assertions, are processed. */
5361
5362 if (ptr[1] == CHAR_QUESTION_MARK && (ptr[2] == CHAR_EQUALS_SIGN ||
5363 ptr[2] == CHAR_EXCLAMATION_MARK || ptr[2] == CHAR_LESS_THAN_SIGN))
5364 break;
5365
5366 /* Most other conditions use OP_CREF (a couple change to OP_RREF
5367 below), and all need to skip 3 bytes at the start of the group. */
5368
5369 code[1+LINK_SIZE] = OP_CREF;
5370 skipbytes = 3;
5371 refsign = -1;
5372
5373 /* Check for a test for recursion in a named group. */
5374
5375 if (ptr[1] == CHAR_R && ptr[2] == CHAR_AMPERSAND)
5376 {
5377 terminator = -1;
5378 ptr += 2;
5379 code[1+LINK_SIZE] = OP_RREF; /* Change the type of test */
5380 }
5381
5382 /* Check for a test for a named group's having been set, using the Perl
5383 syntax (?(<name>) or (?('name') */
5384
5385 else if (ptr[1] == CHAR_LESS_THAN_SIGN)
5386 {
5387 terminator = CHAR_GREATER_THAN_SIGN;
5388 ptr++;
5389 }
5390 else if (ptr[1] == CHAR_APOSTROPHE)
5391 {
5392 terminator = CHAR_APOSTROPHE;
5393 ptr++;
5394 }
5395 else
5396 {
5397 terminator = 0;
5398 if (ptr[1] == CHAR_MINUS || ptr[1] == CHAR_PLUS) refsign = *(++ptr);
5399 }
5400
5401 /* We now expect to read a name; any thing else is an error */
5402
5403 if ((cd->ctypes[ptr[1]] & ctype_word) == 0)
5404 {
5405 ptr += 1; /* To get the right offset */
5406 *errorcodeptr = ERR28;
5407 goto FAILED;
5408 }
5409
5410 /* Read the name, but also get it as a number if it's all digits */
5411
5412 recno = 0;
5413 name = ++ptr;
5414 while ((cd->ctypes[*ptr] & ctype_word) != 0)
5415 {
5416 if (recno >= 0)
5417 recno = ((digitab[*ptr] & ctype_digit) != 0)?
5418 recno * 10 + *ptr - CHAR_0 : -1;
5419 ptr++;
5420 }
5421 namelen = (int)(ptr - name);
5422
5423 if ((terminator > 0 && *ptr++ != terminator) ||
5424 *ptr++ != CHAR_RIGHT_PARENTHESIS)
5425 {
5426 ptr--; /* Error offset */
5427 *errorcodeptr = ERR26;
5428 goto FAILED;
5429 }
5430
5431 /* Do no further checking in the pre-compile phase. */
5432
5433 if (lengthptr != NULL) break;
5434
5435 /* In the real compile we do the work of looking for the actual
5436 reference. If the string started with "+" or "-" we require the rest to
5437 be digits, in which case recno will be set. */
5438
5439 if (refsign > 0)
5440 {
5441 if (recno <= 0)
5442 {
5443 *errorcodeptr = ERR58;
5444 goto FAILED;
5445 }
5446 recno = (refsign == CHAR_MINUS)?
5447 cd->bracount - recno + 1 : recno +cd->bracount;
5448 if (recno <= 0 || recno > cd->final_bracount)
5449 {
5450 *errorcodeptr = ERR15;
5451 goto FAILED;
5452 }
5453 PUT2(code, 2+LINK_SIZE, recno);
5454 break;
5455 }
5456
5457 /* Otherwise (did not start with "+" or "-"), start by looking for the
5458 name. If we find a name, add one to the opcode to change OP_CREF or
5459 OP_RREF into OP_NCREF or OP_NRREF. These behave exactly the same,
5460 except they record that the reference was originally to a name. The
5461 information is used to check duplicate names. */
5462
5463 slot = cd->name_table;
5464 for (i = 0; i < cd->names_found; i++)
5465 {
5466 if (strncmp((char *)name, (char *)slot+2, namelen) == 0) break;
5467 slot += cd->name_entry_size;
5468 }
5469
5470 /* Found a previous named subpattern */
5471
5472 if (i < cd->names_found)
5473 {
5474 recno = GET2(slot, 0);
5475 PUT2(code, 2+LINK_SIZE, recno);
5476 code[1+LINK_SIZE]++;
5477 }
5478
5479 /* Search the pattern for a forward reference */
5480
5481 else if ((i = find_parens(cd, name, namelen,
5482 (options & PCRE_EXTENDED) != 0, utf8)) > 0)
5483 {
5484 PUT2(code, 2+LINK_SIZE, i);
5485 code[1+LINK_SIZE]++;
5486 }
5487
5488 /* If terminator == 0 it means that the name followed directly after
5489 the opening parenthesis [e.g. (?(abc)...] and in this case there are
5490 some further alternatives to try. For the cases where terminator != 0
5491 [things like (?(<name>... or (?('name')... or (?(R&name)... ] we have
5492 now checked all the possibilities, so give an error. */
5493
5494 else if (terminator != 0)
5495 {
5496 *errorcodeptr = ERR15;
5497 goto FAILED;
5498 }
5499
5500 /* Check for (?(R) for recursion. Allow digits after R to specify a
5501 specific group number. */
5502
5503 else if (*name == CHAR_R)
5504 {
5505 recno = 0;
5506 for (i = 1; i < namelen; i++)
5507 {
5508 if ((digitab[name[i]] & ctype_digit) == 0)
5509 {
5510 *errorcodeptr = ERR15;
5511 goto FAILED;
5512 }
5513 recno = recno * 10 + name[i] - CHAR_0;
5514 }
5515 if (recno == 0) recno = RREF_ANY;
5516 code[1+LINK_SIZE] = OP_RREF; /* Change test type */
5517 PUT2(code, 2+LINK_SIZE, recno);
5518 }
5519
5520 /* Similarly, check for the (?(DEFINE) "condition", which is always
5521 false. */
5522
5523 else if (namelen == 6 && strncmp((char *)name, STRING_DEFINE, 6) == 0)
5524 {
5525 code[1+LINK_SIZE] = OP_DEF;
5526 skipbytes = 1;
5527 }
5528
5529 /* Check for the "name" actually being a subpattern number. We are
5530 in the second pass here, so final_bracount is set. */
5531
5532 else if (recno > 0 && recno <= cd->final_bracount)
5533 {
5534 PUT2(code, 2+LINK_SIZE, recno);
5535 }
5536
5537 /* Either an unidentified subpattern, or a reference to (?(0) */
5538
5539 else
5540 {
5541 *errorcodeptr = (recno == 0)? ERR35: ERR15;
5542 goto FAILED;
5543 }
5544 break;
5545
5546
5547 /* ------------------------------------------------------------ */
5548 case CHAR_EQUALS_SIGN: /* Positive lookahead */
5549 bravalue = OP_ASSERT;
5550 cd->assert_depth += 1;
5551 ptr++;
5552 break;
5553
5554
5555 /* ------------------------------------------------------------ */
5556 case CHAR_EXCLAMATION_MARK: /* Negative lookahead */
5557 ptr++;
5558 if (*ptr == CHAR_RIGHT_PARENTHESIS) /* Optimize (?!) */
5559 {
5560 *code++ = OP_FAIL;
5561 previous = NULL;
5562 continue;
5563 }
5564 bravalue = OP_ASSERT_NOT;
5565 cd->assert_depth += 1;
5566 break;
5567
5568
5569 /* ------------------------------------------------------------ */
5570 case CHAR_LESS_THAN_SIGN: /* Lookbehind or named define */
5571 switch (ptr[1])
5572 {
5573 case CHAR_EQUALS_SIGN: /* Positive lookbehind */
5574 bravalue = OP_ASSERTBACK;
5575 cd->assert_depth += 1;
5576 ptr += 2;
5577 break;
5578
5579 case CHAR_EXCLAMATION_MARK: /* Negative lookbehind */
5580 bravalue = OP_ASSERTBACK_NOT;
5581 cd->assert_depth += 1;
5582 ptr += 2;
5583 break;
5584
5585 default: /* Could be name define, else bad */
5586 if ((cd->ctypes[ptr[1]] & ctype_word) != 0) goto DEFINE_NAME;
5587 ptr++; /* Correct offset for error */
5588 *errorcodeptr = ERR24;
5589 goto FAILED;
5590 }
5591 break;
5592
5593
5594 /* ------------------------------------------------------------ */
5595 case CHAR_GREATER_THAN_SIGN: /* One-time brackets */
5596 bravalue = OP_ONCE;
5597 ptr++;
5598 break;
5599
5600
5601 /* ------------------------------------------------------------ */
5602 case CHAR_C: /* Callout - may be followed by digits; */
5603 previous_callout = code; /* Save for later completion */
5604 after_manual_callout = 1; /* Skip one item before completing */
5605 *code++ = OP_CALLOUT;
5606 {
5607 int n = 0;
5608 while ((digitab[*(++ptr)] & ctype_digit) != 0)
5609 n = n * 10 + *ptr - CHAR_0;
5610 if (*ptr != CHAR_RIGHT_PARENTHESIS)
5611 {
5612 *errorcodeptr = ERR39;
5613 goto FAILED;
5614 }
5615 if (n > 255)
5616 {
5617 *errorcodeptr = ERR38;
5618 goto FAILED;
5619 }
5620 *code++ = n;
5621 PUT(code, 0, (int)(ptr - cd->start_pattern + 1)); /* Pattern offset */
5622 PUT(code, LINK_SIZE, 0); /* Default length */
5623 code += 2 * LINK_SIZE;
5624 }
5625 previous = NULL;
5626 continue;
5627
5628
5629 /* ------------------------------------------------------------ */
5630 case CHAR_P: /* Python-style named subpattern handling */
5631 if (*(++ptr) == CHAR_EQUALS_SIGN ||
5632 *ptr == CHAR_GREATER_THAN_SIGN) /* Reference or recursion */
5633 {
5634 is_recurse = *ptr == CHAR_GREATER_THAN_SIGN;
5635 terminator = CHAR_RIGHT_PARENTHESIS;
5636 goto NAMED_REF_OR_RECURSE;
5637 }
5638 else if (*ptr != CHAR_LESS_THAN_SIGN) /* Test for Python-style defn */
5639 {
5640 *errorcodeptr = ERR41;
5641 goto FAILED;
5642 }
5643 /* Fall through to handle (?P< as (?< is handled */
5644
5645
5646 /* ------------------------------------------------------------ */
5647 DEFINE_NAME: /* Come here from (?< handling */
5648 case CHAR_APOSTROPHE:
5649 {
5650 terminator = (*ptr == CHAR_LESS_THAN_SIGN)?
5651 CHAR_GREATER_THAN_SIGN : CHAR_APOSTROPHE;
5652 name = ++ptr;
5653
5654 while ((cd->ctypes[*ptr] & ctype_word) != 0) ptr++;
5655 namelen = (int)(ptr - name);
5656
5657 /* In the pre-compile phase, just do a syntax check. */
5658
5659 if (lengthptr != NULL)
5660 {
5661 if (*ptr != terminator)
5662 {
5663 *errorcodeptr = ERR42;
5664 goto FAILED;
5665 }
5666 if (cd->names_found >= MAX_NAME_COUNT)
5667 {
5668 *errorcodeptr = ERR49;
5669 goto FAILED;
5670 }
5671 if (namelen + 3 > cd->name_entry_size)
5672 {
5673 cd->name_entry_size = namelen + 3;
5674 if (namelen > MAX_NAME_SIZE)
5675 {
5676 *errorcodeptr = ERR48;
5677 goto FAILED;
5678 }
5679 }
5680 }
5681
5682 /* In the real compile, create the entry in the table, maintaining
5683 alphabetical order. Duplicate names for different numbers are
5684 permitted only if PCRE_DUPNAMES is set. Duplicate names for the same
5685 number are always OK. (An existing number can be re-used if (?|
5686 appears in the pattern.) In either event, a duplicate name results in
5687 a duplicate entry in the table, even if the number is the same. This
5688 is because the number of names, and hence the table size, is computed
5689 in the pre-compile, and it affects various numbers and pointers which
5690 would all have to be modified, and the compiled code moved down, if
5691 duplicates with the same number were omitted from the table. This
5692 doesn't seem worth the hassle. However, *different* names for the
5693 same number are not permitted. */
5694
5695 else
5696 {
5697 BOOL dupname = FALSE;
5698 slot = cd->name_table;
5699
5700 for (i = 0; i < cd->names_found; i++)
5701 {
5702 int crc = memcmp(name, slot+2, namelen);
5703 if (crc == 0)
5704 {
5705 if (slot[2+namelen] == 0)
5706 {
5707 if (GET2(slot, 0) != cd->bracount + 1 &&
5708 (options & PCRE_DUPNAMES) == 0)
5709 {
5710 *errorcodeptr = ERR43;
5711 goto FAILED;
5712 }
5713 else dupname = TRUE;
5714 }
5715 else crc = -1; /* Current name is a substring */
5716 }
5717
5718 /* Make space in the table and break the loop for an earlier
5719 name. For a duplicate or later name, carry on. We do this for
5720 duplicates so that in the simple case (when ?(| is not used) they
5721 are in order of their numbers. */
5722
5723 if (crc < 0)
5724 {
5725 memmove(slot + cd->name_entry_size, slot,
5726 (cd->names_found - i) * cd->name_entry_size);
5727 break;
5728 }
5729
5730 /* Continue the loop for a later or duplicate name */
5731
5732 slot += cd->name_entry_size;
5733 }
5734
5735 /* For non-duplicate names, check for a duplicate number before
5736 adding the new name. */
5737
5738 if (!dupname)
5739 {
5740 uschar *cslot = cd->name_table;
5741 for (i = 0; i < cd->names_found; i++)
5742 {
5743 if (cslot != slot)
5744 {
5745 if (GET2(cslot, 0) == cd->bracount + 1)
5746 {
5747 *errorcodeptr = ERR65;
5748 goto FAILED;
5749 }
5750 }
5751 else i--;
5752 cslot += cd->name_entry_size;
5753 }
5754 }
5755
5756 PUT2(slot, 0, cd->bracount + 1);
5757 memcpy(slot + 2, name, namelen);
5758 slot[2+namelen] = 0;
5759 }
5760 }
5761
5762 /* In both pre-compile and compile, count the number of names we've
5763 encountered. */
5764
5765 cd->names_found++;
5766 ptr++; /* Move past > or ' */
5767 goto NUMBERED_GROUP;
5768
5769
5770 /* ------------------------------------------------------------ */
5771 case CHAR_AMPERSAND: /* Perl recursion/subroutine syntax */
5772 terminator = CHAR_RIGHT_PARENTHESIS;
5773 is_recurse = TRUE;
5774 /* Fall through */
5775
5776 /* We come here from the Python syntax above that handles both
5777 references (?P=name) and recursion (?P>name), as well as falling
5778 through from the Perl recursion syntax (?&name). We also come here from
5779 the Perl \k<name> or \k'name' back reference syntax and the \k{name}
5780 .NET syntax, and the Oniguruma \g<...> and \g'...' subroutine syntax. */
5781
5782 NAMED_REF_OR_RECURSE:
5783 name = ++ptr;
5784 while ((cd->ctypes[*ptr] & ctype_word) != 0) ptr++;
5785 namelen = (int)(ptr - name);
5786
5787 /* In the pre-compile phase, do a syntax check. We used to just set
5788 a dummy reference number, because it was not used in the first pass.
5789 However, with the change of recursive back references to be atomic,
5790 we have to look for the number so that this state can be identified, as
5791 otherwise the incorrect length is computed. If it's not a backwards
5792 reference, the dummy number will do. */
5793
5794 if (lengthptr != NULL)
5795 {
5796 const uschar *temp;
5797
5798 if (namelen == 0)
5799 {
5800 *errorcodeptr = ERR62;
5801 goto FAILED;
5802 }
5803 if (*ptr != terminator)
5804 {
5805 *errorcodeptr = ERR42;
5806 goto FAILED;
5807 }
5808 if (namelen > MAX_NAME_SIZE)
5809 {
5810 *errorcodeptr = ERR48;
5811 goto FAILED;
5812 }
5813
5814 /* The name table does not exist in the first pass, so we cannot
5815 do a simple search as in the code below. Instead, we have to scan the
5816 pattern to find the number. It is important that we scan it only as
5817 far as we have got because the syntax of named subpatterns has not
5818 been checked for the rest of the pattern, and find_parens() assumes
5819 correct syntax. In any case, it's a waste of resources to scan
5820 further. We stop the scan at the current point by temporarily
5821 adjusting the value of cd->endpattern. */
5822
5823 temp = cd->end_pattern;
5824 cd->end_pattern = ptr;
5825 recno = find_parens(cd, name, namelen,
5826 (options & PCRE_EXTENDED) != 0, utf8);
5827 cd->end_pattern = temp;
5828 if (recno < 0) recno = 0; /* Forward ref; set dummy number */
5829 }
5830
5831 /* In the real compile, seek the name in the table. We check the name
5832 first, and then check that we have reached the end of the name in the
5833 table. That way, if the name that is longer than any in the table,
5834 the comparison will fail without reading beyond the table entry. */
5835
5836 else
5837 {
5838 slot = cd->name_table;
5839 for (i = 0; i < cd->names_found; i++)
5840 {
5841 if (strncmp((char *)name, (char *)slot+2, namelen) == 0 &&
5842 slot[2+namelen] == 0)
5843 break;
5844 slot += cd->name_entry_size;
5845 }
5846
5847 if (i < cd->names_found) /* Back reference */
5848 {
5849 recno = GET2(slot, 0);
5850 }
5851 else if ((recno = /* Forward back reference */
5852 find_parens(cd, name, namelen,
5853 (options & PCRE_EXTENDED) != 0, utf8)) <= 0)
5854 {
5855 *errorcodeptr = ERR15;
5856 goto FAILED;
5857 }
5858 }
5859
5860 /* In both phases, we can now go to the code than handles numerical
5861 recursion or backreferences. */
5862
5863 if (is_recurse) goto HANDLE_RECURSION;
5864 else goto HANDLE_REFERENCE;
5865
5866
5867 /* ------------------------------------------------------------ */
5868 case CHAR_R: /* Recursion */
5869 ptr++; /* Same as (?0) */
5870 /* Fall through */
5871
5872
5873 /* ------------------------------------------------------------ */
5874 case CHAR_MINUS: case CHAR_PLUS: /* Recursion or subroutine */
5875 case CHAR_0: case CHAR_1: case CHAR_2: case CHAR_3: case CHAR_4:
5876 case CHAR_5: case CHAR_6: case CHAR_7: case CHAR_8: case CHAR_9:
5877 {
5878 const uschar *called;
5879 terminator = CHAR_RIGHT_PARENTHESIS;
5880
5881 /* Come here from the \g<...> and \g'...' code (Oniguruma
5882 compatibility). However, the syntax has been checked to ensure that
5883 the ... are a (signed) number, so that neither ERR63 nor ERR29 will
5884 be called on this path, nor with the jump to OTHER_CHAR_AFTER_QUERY
5885 ever be taken. */
5886
5887 HANDLE_NUMERICAL_RECURSION:
5888
5889 if ((refsign = *ptr) == CHAR_PLUS)
5890 {
5891 ptr++;
5892 if ((digitab[*ptr] & ctype_digit) == 0)
5893 {
5894 *errorcodeptr = ERR63;
5895 goto FAILED;
5896 }
5897 }
5898 else if (refsign == CHAR_MINUS)
5899 {
5900 if ((digitab[ptr[1]] & ctype_digit) == 0)
5901 goto OTHER_CHAR_AFTER_QUERY;
5902 ptr++;
5903 }
5904
5905 recno = 0;
5906 while((digitab[*ptr] & ctype_digit) != 0)
5907 recno = recno * 10 + *ptr++ - CHAR_0;
5908
5909 if (*ptr != terminator)
5910 {
5911 *errorcodeptr = ERR29;
5912 goto FAILED;
5913 }
5914
5915 if (refsign == CHAR_MINUS)
5916 {
5917 if (recno == 0)
5918 {
5919 *errorcodeptr = ERR58;
5920 goto FAILED;
5921 }
5922 recno = cd->bracount - recno + 1;
5923 if (recno <= 0)
5924 {
5925 *errorcodeptr = ERR15;
5926 goto FAILED;
5927 }
5928 }
5929 else if (refsign == CHAR_PLUS)
5930 {
5931 if (recno == 0)
5932 {
5933 *errorcodeptr = ERR58;
5934 goto FAILED;
5935 }
5936 recno += cd->bracount;
5937 }
5938
5939 /* Come here from code above that handles a named recursion */
5940
5941 HANDLE_RECURSION:
5942
5943 previous = code;
5944 called = cd->start_code;
5945
5946 /* When we are actually compiling, find the bracket that is being
5947 referenced. Temporarily end the regex in case it doesn't exist before
5948 this point. If we end up with a forward reference, first check that
5949 the bracket does occur later so we can give the error (and position)
5950 now. Then remember this forward reference in the workspace so it can
5951 be filled in at the end. */
5952
5953 if (lengthptr == NULL)
5954 {
5955 *code = OP_END;
5956 if (recno != 0)
5957 called = _pcre_find_bracket(cd->start_code, utf8, recno);
5958
5959 /* Forward reference */
5960
5961 if (called == NULL)
5962 {
5963 if (find_parens(cd, NULL, recno,
5964 (options & PCRE_EXTENDED) != 0, utf8) < 0)
5965 {
5966 *errorcodeptr = ERR15;
5967 goto FAILED;
5968 }
5969
5970 /* Fudge the value of "called" so that when it is inserted as an
5971 offset below, what it actually inserted is the reference number
5972 of the group. Then remember the forward reference. */
5973
5974 called = cd->start_code + recno;
5975 PUTINC(cd->hwm, 0, (int)(code + 1 - cd->start_code));
5976 }
5977
5978 /* If not a forward reference, and the subpattern is still open,
5979 this is a recursive call. We check to see if this is a left
5980 recursion that could loop for ever, and diagnose that case. We
5981 must not, however, do this check if we are in a conditional
5982 subpattern because the condition might be testing for recursion in
5983 a pattern such as /(?(R)a+|(?R)b)/, which is perfectly valid.
5984 Forever loops are also detected at runtime, so those that occur in
5985 conditional subpatterns will be picked up then. */
5986
5987 else if (GET(called, 1) == 0 && cond_depth <= 0 &&
5988 could_be_empty(called, code, bcptr, utf8, cd))
5989 {
5990 *errorcodeptr = ERR40;
5991 goto FAILED;
5992 }
5993 }
5994
5995 /* Insert the recursion/subroutine item. */
5996
5997 *code = OP_RECURSE;
5998 PUT(code, 1, (int)(called - cd->start_code));
5999 code += 1 + LINK_SIZE;
6000 }
6001
6002 /* Can't determine a first byte now */
6003
6004 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
6005 continue;
6006
6007
6008 /* ------------------------------------------------------------ */
6009 default: /* Other characters: check option setting */
6010 OTHER_CHAR_AFTER_QUERY:
6011 set = unset = 0;
6012 optset = &set;
6013
6014 while (*ptr != CHAR_RIGHT_PARENTHESIS && *ptr != CHAR_COLON)
6015 {
6016 switch (*ptr++)
6017 {
6018 case CHAR_MINUS: optset = &unset; break;
6019
6020 case CHAR_J: /* Record that it changed in the external options */
6021 *optset |= PCRE_DUPNAMES;
6022 cd->external_flags |= PCRE_JCHANGED;
6023 break;
6024
6025 case CHAR_i: *optset |= PCRE_CASELESS; break;
6026 case CHAR_m: *optset |= PCRE_MULTILINE; break;
6027 case CHAR_s: *optset |= PCRE_DOTALL; break;
6028 case CHAR_x: *optset |= PCRE_EXTENDED; break;
6029 case CHAR_U: *optset |= PCRE_UNGREEDY; break;
6030 case CHAR_X: *optset |= PCRE_EXTRA; break;
6031
6032 default: *errorcodeptr = ERR12;
6033 ptr--; /* Correct the offset */
6034 goto FAILED;
6035 }
6036 }
6037
6038 /* Set up the changed option bits, but don't change anything yet. */
6039
6040 newoptions = (options | set) & (~unset);
6041
6042 /* If the options ended with ')' this is not the start of a nested
6043 group with option changes, so the options change at this level. If this
6044 item is right at the start of the pattern, the options can be
6045 abstracted and made external in the pre-compile phase, and ignored in
6046 the compile phase. This can be helpful when matching -- for instance in
6047 caseless checking of required bytes.
6048
6049 If the code pointer is not (cd->start_code + 1 + LINK_SIZE), we are
6050 definitely *not* at the start of the pattern because something has been
6051 compiled. In the pre-compile phase, however, the code pointer can have
6052 that value after the start, because it gets reset as code is discarded
6053 during the pre-compile. However, this can happen only at top level - if
6054 we are within parentheses, the starting BRA will still be present. At
6055 any parenthesis level, the length value can be used to test if anything
6056 has been compiled at that level. Thus, a test for both these conditions
6057 is necessary to ensure we correctly detect the start of the pattern in
6058 both phases.
6059
6060 If we are not at the pattern start, reset the greedy defaults and the
6061 case value for firstbyte and reqbyte. */
6062
6063 if (*ptr == CHAR_RIGHT_PARENTHESIS)
6064 {
6065 if (code == cd->start_code + 1 + LINK_SIZE &&
6066 (lengthptr == NULL || *lengthptr == 2 + 2*LINK_SIZE))
6067 {
6068 cd->external_options = newoptions;
6069 }
6070 else
6071 {
6072 greedy_default = ((newoptions & PCRE_UNGREEDY) != 0);
6073 greedy_non_default = greedy_default ^ 1;
6074 req_caseopt = ((newoptions & PCRE_CASELESS) != 0)? REQ_CASELESS : 0;
6075 }
6076
6077 /* Change options at this level, and pass them back for use
6078 in subsequent branches. */
6079
6080 *optionsptr = options = newoptions;
6081 previous = NULL; /* This item can't be repeated */
6082 continue; /* It is complete */
6083 }
6084
6085 /* If the options ended with ':' we are heading into a nested group
6086 with possible change of options. Such groups are non-capturing and are
6087 not assertions of any kind. All we need to do is skip over the ':';
6088 the newoptions value is handled below. */
6089
6090 bravalue = OP_BRA;
6091 ptr++;
6092 } /* End of switch for character following (? */
6093 } /* End of (? handling */
6094
6095 /* Opening parenthesis not followed by '*' or '?'. If PCRE_NO_AUTO_CAPTURE
6096 is set, all unadorned brackets become non-capturing and behave like (?:...)
6097 brackets. */
6098
6099 else if ((options & PCRE_NO_AUTO_CAPTURE) != 0)
6100 {
6101 bravalue = OP_BRA;
6102 }
6103
6104 /* Else we have a capturing group. */
6105
6106 else
6107 {
6108 NUMBERED_GROUP:
6109 cd->bracount += 1;
6110 PUT2(code, 1+LINK_SIZE, cd->bracount);
6111 skipbytes = 2;
6112 }
6113
6114 /* Process nested bracketed regex. Assertions used not to be repeatable,
6115 but this was changed for Perl compatibility, so all kinds can now be
6116 repeated. We copy code into a non-register variable (tempcode) in order to
6117 be able to pass its address because some compilers complain otherwise. */
6118
6119 previous = code; /* For handling repetition */
6120 *code = bravalue;
6121 tempcode = code;
6122 tempreqvary = cd->req_varyopt; /* Save value before bracket */
6123 tempbracount = cd->bracount; /* Save value before bracket */
6124 length_prevgroup = 0; /* Initialize for pre-compile phase */
6125
6126 if (!compile_regex(
6127 newoptions, /* The complete new option state */
6128 &tempcode, /* Where to put code (updated) */
6129 &ptr, /* Input pointer (updated) */
6130 errorcodeptr, /* Where to put an error message */
6131 (bravalue == OP_ASSERTBACK ||
6132 bravalue == OP_ASSERTBACK_NOT), /* TRUE if back assert */
6133 reset_bracount, /* True if (?| group */
6134 skipbytes, /* Skip over bracket number */
6135 cond_depth +
6136 ((bravalue == OP_COND)?1:0), /* Depth of condition subpatterns */
6137 &subfirstbyte, /* For possible first char */
6138 &subreqbyte, /* For possible last char */
6139 bcptr, /* Current branch chain */
6140 cd, /* Tables block */
6141 (lengthptr == NULL)? NULL : /* Actual compile phase */
6142 &length_prevgroup /* Pre-compile phase */
6143 ))
6144 goto FAILED;
6145
6146 /* If this was an atomic group and there are no capturing groups within it,
6147 generate OP_ONCE_NC instead of OP_ONCE. */
6148
6149 if (bravalue == OP_ONCE && cd->bracount <= tempbracount)
6150 *code = OP_ONCE_NC;
6151
6152 if (bravalue >= OP_ASSERT && bravalue <= OP_ASSERTBACK_NOT)
6153 cd->assert_depth -= 1;
6154
6155 /* At the end of compiling, code is still pointing to the start of the
6156 group, while tempcode has been updated to point past the end of the group.
6157 The pattern pointer (ptr) is on the bracket.
6158
6159 If this is a conditional bracket, check that there are no more than
6160 two branches in the group, or just one if it's a DEFINE group. We do this
6161 in the real compile phase, not in the pre-pass, where the whole group may
6162 not be available. */
6163
6164 if (bravalue == OP_COND && lengthptr == NULL)
6165 {
6166 uschar *tc = code;
6167 int condcount = 0;
6168
6169 do {
6170 condcount++;
6171 tc += GET(tc,1);
6172 }
6173 while (*tc != OP_KET);
6174
6175 /* A DEFINE group is never obeyed inline (the "condition" is always
6176 false). It must have only one branch. */
6177
6178 if (code[LINK_SIZE+1] == OP_DEF)
6179 {
6180 if (condcount > 1)
6181 {
6182 *errorcodeptr = ERR54;
6183 goto FAILED;
6184 }
6185 bravalue = OP_DEF; /* Just a flag to suppress char handling below */
6186 }
6187
6188 /* A "normal" conditional group. If there is just one branch, we must not
6189 make use of its firstbyte or reqbyte, because this is equivalent to an
6190 empty second branch. */
6191
6192 else
6193 {
6194 if (condcount > 2)
6195 {
6196 *errorcodeptr = ERR27;
6197 goto FAILED;
6198 }
6199 if (condcount == 1) subfirstbyte = subreqbyte = REQ_NONE;
6200 }
6201 }
6202
6203 /* Error if hit end of pattern */
6204
6205 if (*ptr != CHAR_RIGHT_PARENTHESIS)
6206 {
6207 *errorcodeptr = ERR14;
6208 goto FAILED;
6209 }
6210
6211 /* In the pre-compile phase, update the length by the length of the group,
6212 less the brackets at either end. Then reduce the compiled code to just a
6213 set of non-capturing brackets so that it doesn't use much memory if it is
6214 duplicated by a quantifier.*/
6215
6216 if (lengthptr != NULL)
6217 {
6218 if (OFLOW_MAX - *lengthptr < length_prevgroup - 2 - 2*LINK_SIZE)
6219 {
6220 *errorcodeptr = ERR20;
6221 goto FAILED;
6222 }
6223 *lengthptr += length_prevgroup - 2 - 2*LINK_SIZE;
6224 code++; /* This already contains bravalue */
6225 PUTINC(code, 0, 1 + LINK_SIZE);
6226 *code++ = OP_KET;
6227 PUTINC(code, 0, 1 + LINK_SIZE);
6228 break; /* No need to waste time with special character handling */
6229 }
6230
6231 /* Otherwise update the main code pointer to the end of the group. */
6232
6233 code = tempcode;
6234
6235 /* For a DEFINE group, required and first character settings are not
6236 relevant. */
6237
6238 if (bravalue == OP_DEF) break;
6239
6240 /* Handle updating of the required and first characters for other types of
6241 group. Update for normal brackets of all kinds, and conditions with two
6242 branches (see code above). If the bracket is followed by a quantifier with
6243 zero repeat, we have to back off. Hence the definition of zeroreqbyte and
6244 zerofirstbyte outside the main loop so that they can be accessed for the
6245 back off. */
6246
6247 zeroreqbyte = reqbyte;
6248 zerofirstbyte = firstbyte;
6249 groupsetfirstbyte = FALSE;
6250
6251 if (bravalue >= OP_ONCE)
6252 {
6253 /* If we have not yet set a firstbyte in this branch, take it from the
6254 subpattern, remembering that it was set here so that a repeat of more
6255 than one can replicate it as reqbyte if necessary. If the subpattern has
6256 no firstbyte, set "none" for the whole branch. In both cases, a zero
6257 repeat forces firstbyte to "none". */
6258
6259 if (firstbyte == REQ_UNSET)
6260 {
6261 if (subfirstbyte >= 0)
6262 {
6263 firstbyte = subfirstbyte;
6264 groupsetfirstbyte = TRUE;
6265 }
6266 else firstbyte = REQ_NONE;
6267 zerofirstbyte = REQ_NONE;
6268 }
6269
6270 /* If firstbyte was previously set, convert the subpattern's firstbyte
6271 into reqbyte if there wasn't one, using the vary flag that was in
6272 existence beforehand. */
6273
6274 else if (subfirstbyte >= 0 && subreqbyte < 0)
6275 subreqbyte = subfirstbyte | tempreqvary;
6276
6277 /* If the subpattern set a required byte (or set a first byte that isn't
6278 really the first byte - see above), set it. */
6279
6280 if (subreqbyte >= 0) reqbyte = subreqbyte;
6281 }
6282
6283 /* For a forward assertion, we take the reqbyte, if set. This can be
6284 helpful if the pattern that follows the assertion doesn't set a different
6285 char. For example, it's useful for /(?=abcde).+/. We can't set firstbyte
6286 for an assertion, however because it leads to incorrect effect for patterns
6287 such as /(?=a)a.+/ when the "real" "a" would then become a reqbyte instead
6288 of a firstbyte. This is overcome by a scan at the end if there's no
6289 firstbyte, looking for an asserted first char. */
6290
6291 else if (bravalue == OP_ASSERT && subreqbyte >= 0) reqbyte = subreqbyte;
6292 break; /* End of processing '(' */
6293
6294
6295 /* ===================================================================*/
6296 /* Handle metasequences introduced by \. For ones like \d, the ESC_ values
6297 are arranged to be the negation of the corresponding OP_values in the
6298 default case when PCRE_UCP is not set. For the back references, the values
6299 are ESC_REF plus the reference number. Only back references and those types
6300 that consume a character may be repeated. We can test for values between
6301 ESC_b and ESC_Z for the latter; this may have to change if any new ones are
6302 ever created. */
6303
6304 case CHAR_BACKSLASH:
6305 tempptr = ptr;
6306 c = check_escape(&ptr, errorcodeptr, cd->bracount, options, FALSE);
6307 if (*errorcodeptr != 0) goto FAILED;
6308
6309 if (c < 0)
6310 {
6311 if (-c == ESC_Q) /* Handle start of quoted string */
6312 {
6313 if (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
6314 ptr += 2; /* avoid empty string */
6315 else inescq = TRUE;
6316 continue;
6317 }
6318
6319 if (-c == ESC_E) continue; /* Perl ignores an orphan \E */
6320
6321 /* For metasequences that actually match a character, we disable the
6322 setting of a first character if it hasn't already been set. */
6323
6324 if (firstbyte == REQ_UNSET && -c > ESC_b && -c < ESC_Z)
6325 firstbyte = REQ_NONE;
6326
6327 /* Set values to reset to if this is followed by a zero repeat. */
6328
6329 zerofirstbyte = firstbyte;
6330 zeroreqbyte = reqbyte;
6331
6332 /* \g<name> or \g'name' is a subroutine call by name and \g<n> or \g'n'
6333 is a subroutine call by number (Oniguruma syntax). In fact, the value
6334 -ESC_g is returned only for these cases. So we don't need to check for <
6335 or ' if the value is -ESC_g. For the Perl syntax \g{n} the value is
6336 -ESC_REF+n, and for the Perl syntax \g{name} the result is -ESC_k (as
6337 that is a synonym for a named back reference). */
6338
6339 if (-c == ESC_g)
6340 {
6341 const uschar *p;
6342 save_hwm = cd->hwm; /* Normally this is set when '(' is read */
6343 terminator = (*(++ptr) == CHAR_LESS_THAN_SIGN)?
6344 CHAR_GREATER_THAN_SIGN : CHAR_APOSTROPHE;
6345
6346 /* These two statements stop the compiler for warning about possibly
6347 unset variables caused by the jump to HANDLE_NUMERICAL_RECURSION. In
6348 fact, because we actually check for a number below, the paths that
6349 would actually be in error are never taken. */
6350
6351 skipbytes = 0;
6352 reset_bracount = FALSE;
6353
6354 /* Test for a name */
6355
6356 if (ptr[1] != CHAR_PLUS && ptr[1] != CHAR_MINUS)
6357 {
6358 BOOL isnumber = TRUE;
6359 for (p = ptr + 1; *p != 0 && *p != terminator; p++)
6360 {
6361 if ((cd->ctypes[*p] & ctype_digit) == 0) isnumber = FALSE;
6362 if ((cd->ctypes[*p] & ctype_word) == 0) break;
6363 }
6364 if (*p != terminator)
6365 {
6366 *errorcodeptr = ERR57;
6367 break;
6368 }
6369 if (isnumber)
6370 {
6371 ptr++;
6372 goto HANDLE_NUMERICAL_RECURSION;
6373 }
6374 is_recurse = TRUE;
6375 goto NAMED_REF_OR_RECURSE;
6376 }
6377
6378 /* Test a signed number in angle brackets or quotes. */
6379
6380 p = ptr + 2;
6381 while ((digitab[*p] & ctype_digit) != 0) p++;
6382 if (*p != terminator)
6383 {
6384 *errorcodeptr = ERR57;
6385 break;
6386 }
6387 ptr++;
6388 goto HANDLE_NUMERICAL_RECURSION;
6389 }
6390
6391 /* \k<name> or \k'name' is a back reference by name (Perl syntax).
6392 We also support \k{name} (.NET syntax). */
6393
6394 if (-c == ESC_k)
6395 {
6396 if ((ptr[1] != CHAR_LESS_THAN_SIGN &&
6397 ptr[1] != CHAR_APOSTROPHE && ptr[1] != CHAR_LEFT_CURLY_BRACKET))
6398 {
6399 *errorcodeptr = ERR69;
6400 break;
6401 }
6402 is_recurse = FALSE;
6403 terminator = (*(++ptr) == CHAR_LESS_THAN_SIGN)?
6404 CHAR_GREATER_THAN_SIGN : (*ptr == CHAR_APOSTROPHE)?
6405 CHAR_APOSTROPHE : CHAR_RIGHT_CURLY_BRACKET;
6406 goto NAMED_REF_OR_RECURSE;
6407 }
6408
6409 /* Back references are handled specially; must disable firstbyte if
6410 not set to cope with cases like (?=(\w+))\1: which would otherwise set
6411 ':' later. */
6412
6413 if (-c >= ESC_REF)
6414 {
6415 open_capitem *oc;
6416 recno = -c - ESC_REF;
6417
6418 HANDLE_REFERENCE: /* Come here from named backref handling */
6419 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
6420 previous = code;
6421 *code++ = ((options & PCRE_CASELESS) != 0)? OP_REFI : OP_REF;
6422 PUT2INC(code, 0, recno);
6423 cd->backref_map |= (recno < 32)? (1 << recno) : 1;
6424 if (recno > cd->top_backref) cd->top_backref = recno;
6425
6426 /* Check to see if this back reference is recursive, that it, it
6427 is inside the group that it references. A flag is set so that the
6428 group can be made atomic. */
6429
6430 for (oc = cd->open_caps; oc != NULL; oc = oc->next)
6431 {
6432 if (oc->number == recno)
6433 {
6434 oc->flag = TRUE;
6435 break;
6436 }
6437 }
6438 }
6439
6440 /* So are Unicode property matches, if supported. */
6441
6442 #ifdef SUPPORT_UCP
6443 else if (-c == ESC_P || -c == ESC_p)
6444 {
6445 BOOL negated;
6446 int pdata;
6447 int ptype = get_ucp(&ptr, &negated, &pdata, errorcodeptr);
6448 if (ptype < 0) goto FAILED;
6449 previous = code;
6450 *code++ = ((-c == ESC_p) != negated)? OP_PROP : OP_NOTPROP;
6451 *code++ = ptype;
6452 *code++ = pdata;
6453 }
6454 #else
6455
6456 /* If Unicode properties are not supported, \X, \P, and \p are not
6457 allowed. */
6458
6459 else if (-c == ESC_X || -c == ESC_P || -c == ESC_p)
6460 {
6461 *errorcodeptr = ERR45;
6462 goto FAILED;
6463 }
6464 #endif
6465
6466 /* For the rest (including \X when Unicode properties are supported), we
6467 can obtain the OP value by negating the escape value in the default
6468 situation when PCRE_UCP is not set. When it *is* set, we substitute
6469 Unicode property tests. */
6470
6471 else
6472 {
6473 #ifdef SUPPORT_UCP
6474 if (-c >= ESC_DU && -c <= ESC_wu)
6475 {
6476 nestptr = ptr + 1; /* Where to resume */
6477 ptr = substitutes[-c - ESC_DU] - 1; /* Just before substitute */
6478 }
6479 else
6480 #endif
6481 {
6482 previous = (-c > ESC_b && -c < ESC_Z)? code : NULL;
6483 *code++ = -c;
6484 }
6485 }
6486 continue;
6487 }
6488
6489 /* We have a data character whose value is in c. In UTF-8 mode it may have
6490 a value > 127. We set its representation in the length/buffer, and then
6491 handle it as a data character. */
6492
6493 #ifdef SUPPORT_UTF8
6494 if (utf8 && c > 127)
6495 mclength = _pcre_ord2utf8(c, mcbuffer);
6496 else
6497 #endif
6498
6499 {
6500 mcbuffer[0] = c;
6501 mclength = 1;
6502 }
6503 goto ONE_CHAR;
6504
6505
6506 /* ===================================================================*/
6507 /* Handle a literal character. It is guaranteed not to be whitespace or #
6508 when the extended flag is set. If we are in UTF-8 mode, it may be a
6509 multi-byte literal character. */
6510
6511 default:
6512 NORMAL_CHAR:
6513 mclength = 1;
6514 mcbuffer[0] = c;
6515
6516 #ifdef SUPPORT_UTF8
6517 if (utf8 && c >= 0xc0)
6518 {
6519 while ((ptr[1] & 0xc0) == 0x80)
6520 mcbuffer[mclength++] = *(++ptr);
6521 }
6522 #endif
6523
6524 /* At this point we have the character's bytes in mcbuffer, and the length
6525 in mclength. When not in UTF-8 mode, the length is always 1. */
6526
6527 ONE_CHAR:
6528 previous = code;
6529 *code++ = ((options & PCRE_CASELESS) != 0)? OP_CHARI : OP_CHAR;
6530 for (c = 0; c < mclength; c++) *code++ = mcbuffer[c];
6531
6532 /* Remember if \r or \n were seen */
6533
6534 if (mcbuffer[0] == CHAR_CR || mcbuffer[0] == CHAR_NL)
6535 cd->external_flags |= PCRE_HASCRORLF;
6536
6537 /* Set the first and required bytes appropriately. If no previous first
6538 byte, set it from this character, but revert to none on a zero repeat.
6539 Otherwise, leave the firstbyte value alone, and don't change it on a zero
6540 repeat. */
6541
6542 if (firstbyte == REQ_UNSET)
6543 {
6544 zerofirstbyte = REQ_NONE;
6545 zeroreqbyte = reqbyte;
6546
6547 /* If the character is more than one byte long, we can set firstbyte
6548 only if it is not to be matched caselessly. */
6549
6550 if (mclength == 1 || req_caseopt == 0)
6551 {
6552 firstbyte = mcbuffer[0] | req_caseopt;
6553 if (mclength != 1) reqbyte = code[-1] | cd->req_varyopt;
6554 }
6555 else firstbyte = reqbyte = REQ_NONE;
6556 }
6557
6558 /* firstbyte was previously set; we can set reqbyte only if the length is
6559 1 or the matching is caseful. */
6560
6561 else
6562 {
6563 zerofirstbyte = firstbyte;
6564 zeroreqbyte = reqbyte;
6565 if (mclength == 1 || req_caseopt == 0)
6566 reqbyte = code[-1] | req_caseopt | cd->req_varyopt;
6567 }
6568
6569 break; /* End of literal character handling */
6570 }
6571 } /* end of big loop */
6572
6573
6574 /* Control never reaches here by falling through, only by a goto for all the
6575 error states. Pass back the position in the pattern so that it can be displayed
6576 to the user for diagnosing the error. */
6577
6578 FAILED:
6579 *ptrptr = ptr;
6580 return FALSE;
6581 }
6582
6583
6584
6585
6586 /*************************************************
6587 * Compile sequence of alternatives *
6588 *************************************************/
6589
6590 /* On entry, ptr is pointing past the bracket character, but on return it
6591 points to the closing bracket, or vertical bar, or end of string. The code
6592 variable is pointing at the byte into which the BRA operator has been stored.
6593 This function is used during the pre-compile phase when we are trying to find
6594 out the amount of memory needed, as well as during the real compile phase. The
6595 value of lengthptr distinguishes the two phases.
6596
6597 Arguments:
6598 options option bits, including any changes for this subpattern
6599 codeptr -> the address of the current code pointer
6600 ptrptr -> the address of the current pattern pointer
6601 errorcodeptr -> pointer to error code variable
6602 lookbehind TRUE if this is a lookbehind assertion
6603 reset_bracount TRUE to reset the count for each branch
6604 skipbytes skip this many bytes at start (for brackets and OP_COND)
6605 cond_depth depth of nesting for conditional subpatterns
6606 firstbyteptr place to put the first required character, or a negative number
6607 reqbyteptr place to put the last required character, or a negative number
6608 bcptr pointer to the chain of currently open branches
6609 cd points to the data block with tables pointers etc.
6610 lengthptr NULL during the real compile phase
6611 points to length accumulator during pre-compile phase
6612
6613 Returns: TRUE on success
6614 */
6615
6616 static BOOL
6617 compile_regex(int options, uschar **codeptr, const uschar **ptrptr,
6618 int *errorcodeptr, BOOL lookbehind, BOOL reset_bracount, int skipbytes,
6619 int cond_depth, int *firstbyteptr, int *reqbyteptr, branch_chain *bcptr,
6620 compile_data *cd, int *lengthptr)
6621 {
6622 const uschar *ptr = *ptrptr;
6623 uschar *code = *codeptr;
6624 uschar *last_branch = code;
6625 uschar *start_bracket = code;
6626 uschar *reverse_count = NULL;
6627 open_capitem capitem;
6628 int capnumber = 0;
6629 int firstbyte, reqbyte;
6630 int branchfirstbyte, branchreqbyte;
6631 int length;
6632 int orig_bracount;
6633 int max_bracount;
6634 branch_chain bc;
6635
6636 bc.outer = bcptr;
6637 bc.current_branch = code;
6638
6639 firstbyte = reqbyte = REQ_UNSET;
6640
6641 /* Accumulate the length for use in the pre-compile phase. Start with the
6642 length of the BRA and KET and any extra bytes that are required at the
6643 beginning. We accumulate in a local variable to save frequent testing of
6644 lenthptr for NULL. We cannot do this by looking at the value of code at the
6645 start and end of each alternative, because compiled items are discarded during
6646 the pre-compile phase so that the work space is not exceeded. */
6647
6648 length = 2 + 2*LINK_SIZE + skipbytes;
6649
6650 /* WARNING: If the above line is changed for any reason, you must also change
6651 the code that abstracts option settings at the start of the pattern and makes
6652