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

Contents of /code/trunk/pcre_compile.c

Parent Directory Parent Directory | Revision Log Revision Log


Revision 754 - (show annotations)
Sat Nov 19 18:32:18 2011 UTC (3 years, 9 months ago) by ph10
File MIME type: text/plain
File size: 250841 byte(s)
Error occurred while calculating annotation data.
Support \C in lookbehinds and DFA matching when not in UTF-8 mode.
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 (in UTF-8 mode only)
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. This only happens in UTF-8 mode;
1706 otherwise \C is coded as OP_ALLANY. */
1707
1708 case OP_ANYBYTE:
1709 return -2;
1710
1711 /* Check a class for variable quantification */
1712
1713 #ifdef SUPPORT_UTF8
1714 case OP_XCLASS:
1715 cc += GET(cc, 1) - 33;
1716 /* Fall through */
1717 #endif
1718
1719 case OP_CLASS:
1720 case OP_NCLASS:
1721 cc += 33;
1722
1723 switch (*cc)
1724 {
1725 case OP_CRPLUS:
1726 case OP_CRMINPLUS:
1727 case OP_CRSTAR:
1728 case OP_CRMINSTAR:
1729 case OP_CRQUERY:
1730 case OP_CRMINQUERY:
1731 return -1;
1732
1733 case OP_CRRANGE:
1734 case OP_CRMINRANGE:
1735 if (GET2(cc,1) != GET2(cc,3)) return -1;
1736 branchlength += GET2(cc,1);
1737 cc += 5;
1738 break;
1739
1740 default:
1741 branchlength++;
1742 }
1743 break;
1744
1745 /* Anything else is variable length */
1746
1747 case OP_ANYNL:
1748 case OP_BRAMINZERO:
1749 case OP_BRAPOS:
1750 case OP_BRAPOSZERO:
1751 case OP_BRAZERO:
1752 case OP_CBRAPOS:
1753 case OP_EXTUNI:
1754 case OP_KETRMAX:
1755 case OP_KETRMIN:
1756 case OP_KETRPOS:
1757 case OP_MINPLUS:
1758 case OP_MINPLUSI:
1759 case OP_MINQUERY:
1760 case OP_MINQUERYI:
1761 case OP_MINSTAR:
1762 case OP_MINSTARI:
1763 case OP_MINUPTO:
1764 case OP_MINUPTOI:
1765 case OP_NOTMINPLUS:
1766 case OP_NOTMINPLUSI:
1767 case OP_NOTMINQUERY:
1768 case OP_NOTMINQUERYI:
1769 case OP_NOTMINSTAR:
1770 case OP_NOTMINSTARI:
1771 case OP_NOTMINUPTO:
1772 case OP_NOTMINUPTOI:
1773 case OP_NOTPLUS:
1774 case OP_NOTPLUSI:
1775 case OP_NOTPOSPLUS:
1776 case OP_NOTPOSPLUSI:
1777 case OP_NOTPOSQUERY:
1778 case OP_NOTPOSQUERYI:
1779 case OP_NOTPOSSTAR:
1780 case OP_NOTPOSSTARI:
1781 case OP_NOTPOSUPTO:
1782 case OP_NOTPOSUPTOI:
1783 case OP_NOTQUERY:
1784 case OP_NOTQUERYI:
1785 case OP_NOTSTAR:
1786 case OP_NOTSTARI:
1787 case OP_NOTUPTO:
1788 case OP_NOTUPTOI:
1789 case OP_PLUS:
1790 case OP_PLUSI:
1791 case OP_POSPLUS:
1792 case OP_POSPLUSI:
1793 case OP_POSQUERY:
1794 case OP_POSQUERYI:
1795 case OP_POSSTAR:
1796 case OP_POSSTARI:
1797 case OP_POSUPTO:
1798 case OP_POSUPTOI:
1799 case OP_QUERY:
1800 case OP_QUERYI:
1801 case OP_REF:
1802 case OP_REFI:
1803 case OP_SBRA:
1804 case OP_SBRAPOS:
1805 case OP_SCBRA:
1806 case OP_SCBRAPOS:
1807 case OP_SCOND:
1808 case OP_SKIPZERO:
1809 case OP_STAR:
1810 case OP_STARI:
1811 case OP_TYPEMINPLUS:
1812 case OP_TYPEMINQUERY:
1813 case OP_TYPEMINSTAR:
1814 case OP_TYPEMINUPTO:
1815 case OP_TYPEPLUS:
1816 case OP_TYPEPOSPLUS:
1817 case OP_TYPEPOSQUERY:
1818 case OP_TYPEPOSSTAR:
1819 case OP_TYPEPOSUPTO:
1820 case OP_TYPEQUERY:
1821 case OP_TYPESTAR:
1822 case OP_TYPEUPTO:
1823 case OP_UPTO:
1824 case OP_UPTOI:
1825 return -1;
1826
1827 /* Catch unrecognized opcodes so that when new ones are added they
1828 are not forgotten, as has happened in the past. */
1829
1830 default:
1831 return -4;
1832 }
1833 }
1834 /* Control never gets here */
1835 }
1836
1837
1838
1839
1840 /*************************************************
1841 * Scan compiled regex for specific bracket *
1842 *************************************************/
1843
1844 /* This little function scans through a compiled pattern until it finds a
1845 capturing bracket with the given number, or, if the number is negative, an
1846 instance of OP_REVERSE for a lookbehind. The function is global in the C sense
1847 so that it can be called from pcre_study() when finding the minimum matching
1848 length.
1849
1850 Arguments:
1851 code points to start of expression
1852 utf8 TRUE in UTF-8 mode
1853 number the required bracket number or negative to find a lookbehind
1854
1855 Returns: pointer to the opcode for the bracket, or NULL if not found
1856 */
1857
1858 const uschar *
1859 _pcre_find_bracket(const uschar *code, BOOL utf8, int number)
1860 {
1861 for (;;)
1862 {
1863 register int c = *code;
1864
1865 if (c == OP_END) return NULL;
1866
1867 /* XCLASS is used for classes that cannot be represented just by a bit
1868 map. This includes negated single high-valued characters. The length in
1869 the table is zero; the actual length is stored in the compiled code. */
1870
1871 if (c == OP_XCLASS) code += GET(code, 1);
1872
1873 /* Handle recursion */
1874
1875 else if (c == OP_REVERSE)
1876 {
1877 if (number < 0) return (uschar *)code;
1878 code += _pcre_OP_lengths[c];
1879 }
1880
1881 /* Handle capturing bracket */
1882
1883 else if (c == OP_CBRA || c == OP_SCBRA ||
1884 c == OP_CBRAPOS || c == OP_SCBRAPOS)
1885 {
1886 int n = GET2(code, 1+LINK_SIZE);
1887 if (n == number) return (uschar *)code;
1888 code += _pcre_OP_lengths[c];
1889 }
1890
1891 /* Otherwise, we can get the item's length from the table, except that for
1892 repeated character types, we have to test for \p and \P, which have an extra
1893 two bytes of parameters, and for MARK/PRUNE/SKIP/THEN with an argument, we
1894 must add in its length. */
1895
1896 else
1897 {
1898 switch(c)
1899 {
1900 case OP_TYPESTAR:
1901 case OP_TYPEMINSTAR:
1902 case OP_TYPEPLUS:
1903 case OP_TYPEMINPLUS:
1904 case OP_TYPEQUERY:
1905 case OP_TYPEMINQUERY:
1906 case OP_TYPEPOSSTAR:
1907 case OP_TYPEPOSPLUS:
1908 case OP_TYPEPOSQUERY:
1909 if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
1910 break;
1911
1912 case OP_TYPEUPTO:
1913 case OP_TYPEMINUPTO:
1914 case OP_TYPEEXACT:
1915 case OP_TYPEPOSUPTO:
1916 if (code[3] == OP_PROP || code[3] == OP_NOTPROP) code += 2;
1917 break;
1918
1919 case OP_MARK:
1920 case OP_PRUNE_ARG:
1921 case OP_SKIP_ARG:
1922 code += code[1];
1923 break;
1924
1925 case OP_THEN_ARG:
1926 code += code[1];
1927 break;
1928 }
1929
1930 /* Add in the fixed length from the table */
1931
1932 code += _pcre_OP_lengths[c];
1933
1934 /* In UTF-8 mode, opcodes that are followed by a character may be followed by
1935 a multi-byte character. The length in the table is a minimum, so we have to
1936 arrange to skip the extra bytes. */
1937
1938 #ifdef SUPPORT_UTF8
1939 if (utf8) switch(c)
1940 {
1941 case OP_CHAR:
1942 case OP_CHARI:
1943 case OP_EXACT:
1944 case OP_EXACTI:
1945 case OP_UPTO:
1946 case OP_UPTOI:
1947 case OP_MINUPTO:
1948 case OP_MINUPTOI:
1949 case OP_POSUPTO:
1950 case OP_POSUPTOI:
1951 case OP_STAR:
1952 case OP_STARI:
1953 case OP_MINSTAR:
1954 case OP_MINSTARI:
1955 case OP_POSSTAR:
1956 case OP_POSSTARI:
1957 case OP_PLUS:
1958 case OP_PLUSI:
1959 case OP_MINPLUS:
1960 case OP_MINPLUSI:
1961 case OP_POSPLUS:
1962 case OP_POSPLUSI:
1963 case OP_QUERY:
1964 case OP_QUERYI:
1965 case OP_MINQUERY:
1966 case OP_MINQUERYI:
1967 case OP_POSQUERY:
1968 case OP_POSQUERYI:
1969 if (code[-1] >= 0xc0) code += _pcre_utf8_table4[code[-1] & 0x3f];
1970 break;
1971 }
1972 #else
1973 (void)(utf8); /* Keep compiler happy by referencing function argument */
1974 #endif
1975 }
1976 }
1977 }
1978
1979
1980
1981 /*************************************************
1982 * Scan compiled regex for recursion reference *
1983 *************************************************/
1984
1985 /* This little function scans through a compiled pattern until it finds an
1986 instance of OP_RECURSE.
1987
1988 Arguments:
1989 code points to start of expression
1990 utf8 TRUE in UTF-8 mode
1991
1992 Returns: pointer to the opcode for OP_RECURSE, or NULL if not found
1993 */
1994
1995 static const uschar *
1996 find_recurse(const uschar *code, BOOL utf8)
1997 {
1998 for (;;)
1999 {
2000 register int c = *code;
2001 if (c == OP_END) return NULL;
2002 if (c == OP_RECURSE) return code;
2003
2004 /* XCLASS is used for classes that cannot be represented just by a bit
2005 map. This includes negated single high-valued characters. The length in
2006 the table is zero; the actual length is stored in the compiled code. */
2007
2008 if (c == OP_XCLASS) code += GET(code, 1);
2009
2010 /* Otherwise, we can get the item's length from the table, except that for
2011 repeated character types, we have to test for \p and \P, which have an extra
2012 two bytes of parameters, and for MARK/PRUNE/SKIP/THEN with an argument, we
2013 must add in its length. */
2014
2015 else
2016 {
2017 switch(c)
2018 {
2019 case OP_TYPESTAR:
2020 case OP_TYPEMINSTAR:
2021 case OP_TYPEPLUS:
2022 case OP_TYPEMINPLUS:
2023 case OP_TYPEQUERY:
2024 case OP_TYPEMINQUERY:
2025 case OP_TYPEPOSSTAR:
2026 case OP_TYPEPOSPLUS:
2027 case OP_TYPEPOSQUERY:
2028 if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
2029 break;
2030
2031 case OP_TYPEPOSUPTO:
2032 case OP_TYPEUPTO:
2033 case OP_TYPEMINUPTO:
2034 case OP_TYPEEXACT:
2035 if (code[3] == OP_PROP || code[3] == OP_NOTPROP) code += 2;
2036 break;
2037
2038 case OP_MARK:
2039 case OP_PRUNE_ARG:
2040 case OP_SKIP_ARG:
2041 code += code[1];
2042 break;
2043
2044 case OP_THEN_ARG:
2045 code += code[1];
2046 break;
2047 }
2048
2049 /* Add in the fixed length from the table */
2050
2051 code += _pcre_OP_lengths[c];
2052
2053 /* In UTF-8 mode, opcodes that are followed by a character may be followed
2054 by a multi-byte character. The length in the table is a minimum, so we have
2055 to arrange to skip the extra bytes. */
2056
2057 #ifdef SUPPORT_UTF8
2058 if (utf8) switch(c)
2059 {
2060 case OP_CHAR:
2061 case OP_CHARI:
2062 case OP_EXACT:
2063 case OP_EXACTI:
2064 case OP_UPTO:
2065 case OP_UPTOI:
2066 case OP_MINUPTO:
2067 case OP_MINUPTOI:
2068 case OP_POSUPTO:
2069 case OP_POSUPTOI:
2070 case OP_STAR:
2071 case OP_STARI:
2072 case OP_MINSTAR:
2073 case OP_MINSTARI:
2074 case OP_POSSTAR:
2075 case OP_POSSTARI:
2076 case OP_PLUS:
2077 case OP_PLUSI:
2078 case OP_MINPLUS:
2079 case OP_MINPLUSI:
2080 case OP_POSPLUS:
2081 case OP_POSPLUSI:
2082 case OP_QUERY:
2083 case OP_QUERYI:
2084 case OP_MINQUERY:
2085 case OP_MINQUERYI:
2086 case OP_POSQUERY:
2087 case OP_POSQUERYI:
2088 if (code[-1] >= 0xc0) code += _pcre_utf8_table4[code[-1] & 0x3f];
2089 break;
2090 }
2091 #else
2092 (void)(utf8); /* Keep compiler happy by referencing function argument */
2093 #endif
2094 }
2095 }
2096 }
2097
2098
2099
2100 /*************************************************
2101 * Scan compiled branch for non-emptiness *
2102 *************************************************/
2103
2104 /* This function scans through a branch of a compiled pattern to see whether it
2105 can match the empty string or not. It is called from could_be_empty()
2106 below and from compile_branch() when checking for an unlimited repeat of a
2107 group that can match nothing. Note that first_significant_code() skips over
2108 backward and negative forward assertions when its final argument is TRUE. If we
2109 hit an unclosed bracket, we return "empty" - this means we've struck an inner
2110 bracket whose current branch will already have been scanned.
2111
2112 Arguments:
2113 code points to start of search
2114 endcode points to where to stop
2115 utf8 TRUE if in UTF8 mode
2116 cd contains pointers to tables etc.
2117
2118 Returns: TRUE if what is matched could be empty
2119 */
2120
2121 static BOOL
2122 could_be_empty_branch(const uschar *code, const uschar *endcode, BOOL utf8,
2123 compile_data *cd)
2124 {
2125 register int c;
2126 for (code = first_significant_code(code + _pcre_OP_lengths[*code], TRUE);
2127 code < endcode;
2128 code = first_significant_code(code + _pcre_OP_lengths[c], TRUE))
2129 {
2130 const uschar *ccode;
2131
2132 c = *code;
2133
2134 /* Skip over forward assertions; the other assertions are skipped by
2135 first_significant_code() with a TRUE final argument. */
2136
2137 if (c == OP_ASSERT)
2138 {
2139 do code += GET(code, 1); while (*code == OP_ALT);
2140 c = *code;
2141 continue;
2142 }
2143
2144 /* For a recursion/subroutine call, if its end has been reached, which
2145 implies a backward reference subroutine call, we can scan it. If it's a
2146 forward reference subroutine call, we can't. To detect forward reference
2147 we have to scan up the list that is kept in the workspace. This function is
2148 called only when doing the real compile, not during the pre-compile that
2149 measures the size of the compiled pattern. */
2150
2151 if (c == OP_RECURSE)
2152 {
2153 const uschar *scode;
2154 BOOL empty_branch;
2155
2156 /* Test for forward reference */
2157
2158 for (scode = cd->start_workspace; scode < cd->hwm; scode += LINK_SIZE)
2159 if (GET(scode, 0) == code + 1 - cd->start_code) return TRUE;
2160
2161 /* Not a forward reference, test for completed backward reference */
2162
2163 empty_branch = FALSE;
2164 scode = cd->start_code + GET(code, 1);
2165 if (GET(scode, 1) == 0) return TRUE; /* Unclosed */
2166
2167 /* Completed backwards reference */
2168
2169 do
2170 {
2171 if (could_be_empty_branch(scode, endcode, utf8, cd))
2172 {
2173 empty_branch = TRUE;
2174 break;
2175 }
2176 scode += GET(scode, 1);
2177 }
2178 while (*scode == OP_ALT);
2179
2180 if (!empty_branch) return FALSE; /* All branches are non-empty */
2181 continue;
2182 }
2183
2184 /* Groups with zero repeats can of course be empty; skip them. */
2185
2186 if (c == OP_BRAZERO || c == OP_BRAMINZERO || c == OP_SKIPZERO ||
2187 c == OP_BRAPOSZERO)
2188 {
2189 code += _pcre_OP_lengths[c];
2190 do code += GET(code, 1); while (*code == OP_ALT);
2191 c = *code;
2192 continue;
2193 }
2194
2195 /* A nested group that is already marked as "could be empty" can just be
2196 skipped. */
2197
2198 if (c == OP_SBRA || c == OP_SBRAPOS ||
2199 c == OP_SCBRA || c == OP_SCBRAPOS)
2200 {
2201 do code += GET(code, 1); while (*code == OP_ALT);
2202 c = *code;
2203 continue;
2204 }
2205
2206 /* For other groups, scan the branches. */
2207
2208 if (c == OP_BRA || c == OP_BRAPOS ||
2209 c == OP_CBRA || c == OP_CBRAPOS ||
2210 c == OP_ONCE || c == OP_ONCE_NC ||
2211 c == OP_COND)
2212 {
2213 BOOL empty_branch;
2214 if (GET(code, 1) == 0) return TRUE; /* Hit unclosed bracket */
2215
2216 /* If a conditional group has only one branch, there is a second, implied,
2217 empty branch, so just skip over the conditional, because it could be empty.
2218 Otherwise, scan the individual branches of the group. */
2219
2220 if (c == OP_COND && code[GET(code, 1)] != OP_ALT)
2221 code += GET(code, 1);
2222 else
2223 {
2224 empty_branch = FALSE;
2225 do
2226 {
2227 if (!empty_branch && could_be_empty_branch(code, endcode, utf8, cd))
2228 empty_branch = TRUE;
2229 code += GET(code, 1);
2230 }
2231 while (*code == OP_ALT);
2232 if (!empty_branch) return FALSE; /* All branches are non-empty */
2233 }
2234
2235 c = *code;
2236 continue;
2237 }
2238
2239 /* Handle the other opcodes */
2240
2241 switch (c)
2242 {
2243 /* Check for quantifiers after a class. XCLASS is used for classes that
2244 cannot be represented just by a bit map. This includes negated single
2245 high-valued characters. The length in _pcre_OP_lengths[] is zero; the
2246 actual length is stored in the compiled code, so we must update "code"
2247 here. */
2248
2249 #ifdef SUPPORT_UTF8
2250 case OP_XCLASS:
2251 ccode = code += GET(code, 1);
2252 goto CHECK_CLASS_REPEAT;
2253 #endif
2254
2255 case OP_CLASS:
2256 case OP_NCLASS:
2257 ccode = code + 33;
2258
2259 #ifdef SUPPORT_UTF8
2260 CHECK_CLASS_REPEAT:
2261 #endif
2262
2263 switch (*ccode)
2264 {
2265 case OP_CRSTAR: /* These could be empty; continue */
2266 case OP_CRMINSTAR:
2267 case OP_CRQUERY:
2268 case OP_CRMINQUERY:
2269 break;
2270
2271 default: /* Non-repeat => class must match */
2272 case OP_CRPLUS: /* These repeats aren't empty */
2273 case OP_CRMINPLUS:
2274 return FALSE;
2275
2276 case OP_CRRANGE:
2277 case OP_CRMINRANGE:
2278 if (GET2(ccode, 1) > 0) return FALSE; /* Minimum > 0 */
2279 break;
2280 }
2281 break;
2282
2283 /* Opcodes that must match a character */
2284
2285 case OP_PROP:
2286 case OP_NOTPROP:
2287 case OP_EXTUNI:
2288 case OP_NOT_DIGIT:
2289 case OP_DIGIT:
2290 case OP_NOT_WHITESPACE:
2291 case OP_WHITESPACE:
2292 case OP_NOT_WORDCHAR:
2293 case OP_WORDCHAR:
2294 case OP_ANY:
2295 case OP_ALLANY:
2296 case OP_ANYBYTE:
2297 case OP_CHAR:
2298 case OP_CHARI:
2299 case OP_NOT:
2300 case OP_NOTI:
2301 case OP_PLUS:
2302 case OP_MINPLUS:
2303 case OP_POSPLUS:
2304 case OP_EXACT:
2305 case OP_NOTPLUS:
2306 case OP_NOTMINPLUS:
2307 case OP_NOTPOSPLUS:
2308 case OP_NOTEXACT:
2309 case OP_TYPEPLUS:
2310 case OP_TYPEMINPLUS:
2311 case OP_TYPEPOSPLUS:
2312 case OP_TYPEEXACT:
2313 return FALSE;
2314
2315 /* These are going to continue, as they may be empty, but we have to
2316 fudge the length for the \p and \P cases. */
2317
2318 case OP_TYPESTAR:
2319 case OP_TYPEMINSTAR:
2320 case OP_TYPEPOSSTAR:
2321 case OP_TYPEQUERY:
2322 case OP_TYPEMINQUERY:
2323 case OP_TYPEPOSQUERY:
2324 if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
2325 break;
2326
2327 /* Same for these */
2328
2329 case OP_TYPEUPTO:
2330 case OP_TYPEMINUPTO:
2331 case OP_TYPEPOSUPTO:
2332 if (code[3] == OP_PROP || code[3] == OP_NOTPROP) code += 2;
2333 break;
2334
2335 /* End of branch */
2336
2337 case OP_KET:
2338 case OP_KETRMAX:
2339 case OP_KETRMIN:
2340 case OP_KETRPOS:
2341 case OP_ALT:
2342 return TRUE;
2343
2344 /* In UTF-8 mode, STAR, MINSTAR, POSSTAR, QUERY, MINQUERY, POSQUERY, UPTO,
2345 MINUPTO, and POSUPTO may be followed by a multibyte character */
2346
2347 #ifdef SUPPORT_UTF8
2348 case OP_STAR:
2349 case OP_STARI:
2350 case OP_MINSTAR:
2351 case OP_MINSTARI:
2352 case OP_POSSTAR:
2353 case OP_POSSTARI:
2354 case OP_QUERY:
2355 case OP_QUERYI:
2356 case OP_MINQUERY:
2357 case OP_MINQUERYI:
2358 case OP_POSQUERY:
2359 case OP_POSQUERYI:
2360 if (utf8 && code[1] >= 0xc0) code += _pcre_utf8_table4[code[1] & 0x3f];
2361 break;
2362
2363 case OP_UPTO:
2364 case OP_UPTOI:
2365 case OP_MINUPTO:
2366 case OP_MINUPTOI:
2367 case OP_POSUPTO:
2368 case OP_POSUPTOI:
2369 if (utf8 && code[3] >= 0xc0) code += _pcre_utf8_table4[code[3] & 0x3f];
2370 break;
2371 #endif
2372
2373 /* MARK, and PRUNE/SKIP/THEN with an argument must skip over the argument
2374 string. */
2375
2376 case OP_MARK:
2377 case OP_PRUNE_ARG:
2378 case OP_SKIP_ARG:
2379 code += code[1];
2380 break;
2381
2382 case OP_THEN_ARG:
2383 code += code[1];
2384 break;
2385
2386 /* None of the remaining opcodes are required to match a character. */
2387
2388 default:
2389 break;
2390 }
2391 }
2392
2393 return TRUE;
2394 }
2395
2396
2397
2398 /*************************************************
2399 * Scan compiled regex for non-emptiness *
2400 *************************************************/
2401
2402 /* This function is called to check for left recursive calls. We want to check
2403 the current branch of the current pattern to see if it could match the empty
2404 string. If it could, we must look outwards for branches at other levels,
2405 stopping when we pass beyond the bracket which is the subject of the recursion.
2406 This function is called only during the real compile, not during the
2407 pre-compile.
2408
2409 Arguments:
2410 code points to start of the recursion
2411 endcode points to where to stop (current RECURSE item)
2412 bcptr points to the chain of current (unclosed) branch starts
2413 utf8 TRUE if in UTF-8 mode
2414 cd pointers to tables etc
2415
2416 Returns: TRUE if what is matched could be empty
2417 */
2418
2419 static BOOL
2420 could_be_empty(const uschar *code, const uschar *endcode, branch_chain *bcptr,
2421 BOOL utf8, compile_data *cd)
2422 {
2423 while (bcptr != NULL && bcptr->current_branch >= code)
2424 {
2425 if (!could_be_empty_branch(bcptr->current_branch, endcode, utf8, cd))
2426 return FALSE;
2427 bcptr = bcptr->outer;
2428 }
2429 return TRUE;
2430 }
2431
2432
2433
2434 /*************************************************
2435 * Check for POSIX class syntax *
2436 *************************************************/
2437
2438 /* This function is called when the sequence "[:" or "[." or "[=" is
2439 encountered in a character class. It checks whether this is followed by a
2440 sequence of characters terminated by a matching ":]" or ".]" or "=]". If we
2441 reach an unescaped ']' without the special preceding character, return FALSE.
2442
2443 Originally, this function only recognized a sequence of letters between the
2444 terminators, but it seems that Perl recognizes any sequence of characters,
2445 though of course unknown POSIX names are subsequently rejected. Perl gives an
2446 "Unknown POSIX class" error for [:f\oo:] for example, where previously PCRE
2447 didn't consider this to be a POSIX class. Likewise for [:1234:].
2448
2449 The problem in trying to be exactly like Perl is in the handling of escapes. We
2450 have to be sure that [abc[:x\]pqr] is *not* treated as containing a POSIX
2451 class, but [abc[:x\]pqr:]] is (so that an error can be generated). The code
2452 below handles the special case of \], but does not try to do any other escape
2453 processing. This makes it different from Perl for cases such as [:l\ower:]
2454 where Perl recognizes it as the POSIX class "lower" but PCRE does not recognize
2455 "l\ower". This is a lesser evil that not diagnosing bad classes when Perl does,
2456 I think.
2457
2458 A user pointed out that PCRE was rejecting [:a[:digit:]] whereas Perl was not.
2459 It seems that the appearance of a nested POSIX class supersedes an apparent
2460 external class. For example, [:a[:digit:]b:] matches "a", "b", ":", or
2461 a digit.
2462
2463 In Perl, unescaped square brackets may also appear as part of class names. For
2464 example, [:a[:abc]b:] gives unknown POSIX class "[:abc]b:]". However, for
2465 [:a[:abc]b][b:] it gives unknown POSIX class "[:abc]b][b:]", which does not
2466 seem right at all. PCRE does not allow closing square brackets in POSIX class
2467 names.
2468
2469 Arguments:
2470 ptr pointer to the initial [
2471 endptr where to return the end pointer
2472
2473 Returns: TRUE or FALSE
2474 */
2475
2476 static BOOL
2477 check_posix_syntax(const uschar *ptr, const uschar **endptr)
2478 {
2479 int terminator; /* Don't combine these lines; the Solaris cc */
2480 terminator = *(++ptr); /* compiler warns about "non-constant" initializer. */
2481 for (++ptr; *ptr != 0; ptr++)
2482 {
2483 if (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET)
2484 ptr++;
2485 else if (*ptr == CHAR_RIGHT_SQUARE_BRACKET) return FALSE;
2486 else
2487 {
2488 if (*ptr == terminator && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET)
2489 {
2490 *endptr = ptr;
2491 return TRUE;
2492 }
2493 if (*ptr == CHAR_LEFT_SQUARE_BRACKET &&
2494 (ptr[1] == CHAR_COLON || ptr[1] == CHAR_DOT ||
2495 ptr[1] == CHAR_EQUALS_SIGN) &&
2496 check_posix_syntax(ptr, endptr))
2497 return FALSE;
2498 }
2499 }
2500 return FALSE;
2501 }
2502
2503
2504
2505
2506 /*************************************************
2507 * Check POSIX class name *
2508 *************************************************/
2509
2510 /* This function is called to check the name given in a POSIX-style class entry
2511 such as [:alnum:].
2512
2513 Arguments:
2514 ptr points to the first letter
2515 len the length of the name
2516
2517 Returns: a value representing the name, or -1 if unknown
2518 */
2519
2520 static int
2521 check_posix_name(const uschar *ptr, int len)
2522 {
2523 const char *pn = posix_names;
2524 register int yield = 0;
2525 while (posix_name_lengths[yield] != 0)
2526 {
2527 if (len == posix_name_lengths[yield] &&
2528 strncmp((const char *)ptr, pn, len) == 0) return yield;
2529 pn += posix_name_lengths[yield] + 1;
2530 yield++;
2531 }
2532 return -1;
2533 }
2534
2535
2536 /*************************************************
2537 * Adjust OP_RECURSE items in repeated group *
2538 *************************************************/
2539
2540 /* OP_RECURSE items contain an offset from the start of the regex to the group
2541 that is referenced. This means that groups can be replicated for fixed
2542 repetition simply by copying (because the recursion is allowed to refer to
2543 earlier groups that are outside the current group). However, when a group is
2544 optional (i.e. the minimum quantifier is zero), OP_BRAZERO or OP_SKIPZERO is
2545 inserted before it, after it has been compiled. This means that any OP_RECURSE
2546 items within it that refer to the group itself or any contained groups have to
2547 have their offsets adjusted. That one of the jobs of this function. Before it
2548 is called, the partially compiled regex must be temporarily terminated with
2549 OP_END.
2550
2551 This function has been extended with the possibility of forward references for
2552 recursions and subroutine calls. It must also check the list of such references
2553 for the group we are dealing with. If it finds that one of the recursions in
2554 the current group is on this list, it adjusts the offset in the list, not the
2555 value in the reference (which is a group number).
2556
2557 Arguments:
2558 group points to the start of the group
2559 adjust the amount by which the group is to be moved
2560 utf8 TRUE in UTF-8 mode
2561 cd contains pointers to tables etc.
2562 save_hwm the hwm forward reference pointer at the start of the group
2563
2564 Returns: nothing
2565 */
2566
2567 static void
2568 adjust_recurse(uschar *group, int adjust, BOOL utf8, compile_data *cd,
2569 uschar *save_hwm)
2570 {
2571 uschar *ptr = group;
2572
2573 while ((ptr = (uschar *)find_recurse(ptr, utf8)) != NULL)
2574 {
2575 int offset;
2576 uschar *hc;
2577
2578 /* See if this recursion is on the forward reference list. If so, adjust the
2579 reference. */
2580
2581 for (hc = save_hwm; hc < cd->hwm; hc += LINK_SIZE)
2582 {
2583 offset = GET(hc, 0);
2584 if (cd->start_code + offset == ptr + 1)
2585 {
2586 PUT(hc, 0, offset + adjust);
2587 break;
2588 }
2589 }
2590
2591 /* Otherwise, adjust the recursion offset if it's after the start of this
2592 group. */
2593
2594 if (hc >= cd->hwm)
2595 {
2596 offset = GET(ptr, 1);
2597 if (cd->start_code + offset >= group) PUT(ptr, 1, offset + adjust);
2598 }
2599
2600 ptr += 1 + LINK_SIZE;
2601 }
2602 }
2603
2604
2605
2606 /*************************************************
2607 * Insert an automatic callout point *
2608 *************************************************/
2609
2610 /* This function is called when the PCRE_AUTO_CALLOUT option is set, to insert
2611 callout points before each pattern item.
2612
2613 Arguments:
2614 code current code pointer
2615 ptr current pattern pointer
2616 cd pointers to tables etc
2617
2618 Returns: new code pointer
2619 */
2620
2621 static uschar *
2622 auto_callout(uschar *code, const uschar *ptr, compile_data *cd)
2623 {
2624 *code++ = OP_CALLOUT;
2625 *code++ = 255;
2626 PUT(code, 0, (int)(ptr - cd->start_pattern)); /* Pattern offset */
2627 PUT(code, LINK_SIZE, 0); /* Default length */
2628 return code + 2*LINK_SIZE;
2629 }
2630
2631
2632
2633 /*************************************************
2634 * Complete a callout item *
2635 *************************************************/
2636
2637 /* A callout item contains the length of the next item in the pattern, which
2638 we can't fill in till after we have reached the relevant point. This is used
2639 for both automatic and manual callouts.
2640
2641 Arguments:
2642 previous_callout points to previous callout item
2643 ptr current pattern pointer
2644 cd pointers to tables etc
2645
2646 Returns: nothing
2647 */
2648
2649 static void
2650 complete_callout(uschar *previous_callout, const uschar *ptr, compile_data *cd)
2651 {
2652 int length = (int)(ptr - cd->start_pattern - GET(previous_callout, 2));
2653 PUT(previous_callout, 2 + LINK_SIZE, length);
2654 }
2655
2656
2657
2658 #ifdef SUPPORT_UCP
2659 /*************************************************
2660 * Get othercase range *
2661 *************************************************/
2662
2663 /* This function is passed the start and end of a class range, in UTF-8 mode
2664 with UCP support. It searches up the characters, looking for internal ranges of
2665 characters in the "other" case. Each call returns the next one, updating the
2666 start address.
2667
2668 Arguments:
2669 cptr points to starting character value; updated
2670 d end value
2671 ocptr where to put start of othercase range
2672 odptr where to put end of othercase range
2673
2674 Yield: TRUE when range returned; FALSE when no more
2675 */
2676
2677 static BOOL
2678 get_othercase_range(unsigned int *cptr, unsigned int d, unsigned int *ocptr,
2679 unsigned int *odptr)
2680 {
2681 unsigned int c, othercase, next;
2682
2683 for (c = *cptr; c <= d; c++)
2684 { if ((othercase = UCD_OTHERCASE(c)) != c) break; }
2685
2686 if (c > d) return FALSE;
2687
2688 *ocptr = othercase;
2689 next = othercase + 1;
2690
2691 for (++c; c <= d; c++)
2692 {
2693 if (UCD_OTHERCASE(c) != next) break;
2694 next++;
2695 }
2696
2697 *odptr = next - 1;
2698 *cptr = c;
2699
2700 return TRUE;
2701 }
2702
2703
2704
2705 /*************************************************
2706 * Check a character and a property *
2707 *************************************************/
2708
2709 /* This function is called by check_auto_possessive() when a property item
2710 is adjacent to a fixed character.
2711
2712 Arguments:
2713 c the character
2714 ptype the property type
2715 pdata the data for the type
2716 negated TRUE if it's a negated property (\P or \p{^)
2717
2718 Returns: TRUE if auto-possessifying is OK
2719 */
2720
2721 static BOOL
2722 check_char_prop(int c, int ptype, int pdata, BOOL negated)
2723 {
2724 const ucd_record *prop = GET_UCD(c);
2725 switch(ptype)
2726 {
2727 case PT_LAMP:
2728 return (prop->chartype == ucp_Lu ||
2729 prop->chartype == ucp_Ll ||
2730 prop->chartype == ucp_Lt) == negated;
2731
2732 case PT_GC:
2733 return (pdata == _pcre_ucp_gentype[prop->chartype]) == negated;
2734
2735 case PT_PC:
2736 return (pdata == prop->chartype) == negated;
2737
2738 case PT_SC:
2739 return (pdata == prop->script) == negated;
2740
2741 /* These are specials */
2742
2743 case PT_ALNUM:
2744 return (_pcre_ucp_gentype[prop->chartype] == ucp_L ||
2745 _pcre_ucp_gentype[prop->chartype] == ucp_N) == negated;
2746
2747 case PT_SPACE: /* Perl space */
2748 return (_pcre_ucp_gentype[prop->chartype] == ucp_Z ||
2749 c == CHAR_HT || c == CHAR_NL || c == CHAR_FF || c == CHAR_CR)
2750 == negated;
2751
2752 case PT_PXSPACE: /* POSIX space */
2753 return (_pcre_ucp_gentype[prop->chartype] == ucp_Z ||
2754 c == CHAR_HT || c == CHAR_NL || c == CHAR_VT ||
2755 c == CHAR_FF || c == CHAR_CR)
2756 == negated;
2757
2758 case PT_WORD:
2759 return (_pcre_ucp_gentype[prop->chartype] == ucp_L ||
2760 _pcre_ucp_gentype[prop->chartype] == ucp_N ||
2761 c == CHAR_UNDERSCORE) == negated;
2762 }
2763 return FALSE;
2764 }
2765 #endif /* SUPPORT_UCP */
2766
2767
2768
2769 /*************************************************
2770 * Check if auto-possessifying is possible *
2771 *************************************************/
2772
2773 /* This function is called for unlimited repeats of certain items, to see
2774 whether the next thing could possibly match the repeated item. If not, it makes
2775 sense to automatically possessify the repeated item.
2776
2777 Arguments:
2778 previous pointer to the repeated opcode
2779 utf8 TRUE in UTF-8 mode
2780 ptr next character in pattern
2781 options options bits
2782 cd contains pointers to tables etc.
2783
2784 Returns: TRUE if possessifying is wanted
2785 */
2786
2787 static BOOL
2788 check_auto_possessive(const uschar *previous, BOOL utf8, const uschar *ptr,
2789 int options, compile_data *cd)
2790 {
2791 int c, next;
2792 int op_code = *previous++;
2793
2794 /* Skip whitespace and comments in extended mode */
2795
2796 if ((options & PCRE_EXTENDED) != 0)
2797 {
2798 for (;;)
2799 {
2800 while ((cd->ctypes[*ptr] & ctype_space) != 0) ptr++;
2801 if (*ptr == CHAR_NUMBER_SIGN)
2802 {
2803 ptr++;
2804 while (*ptr != 0)
2805 {
2806 if (IS_NEWLINE(ptr)) { ptr += cd->nllen; break; }
2807 ptr++;
2808 #ifdef SUPPORT_UTF8
2809 if (utf8) while ((*ptr & 0xc0) == 0x80) ptr++;
2810 #endif
2811 }
2812 }
2813 else break;
2814 }
2815 }
2816
2817 /* If the next item is one that we can handle, get its value. A non-negative
2818 value is a character, a negative value is an escape value. */
2819
2820 if (*ptr == CHAR_BACKSLASH)
2821 {
2822 int temperrorcode = 0;
2823 next = check_escape(&ptr, &temperrorcode, cd->bracount, options, FALSE);
2824 if (temperrorcode != 0) return FALSE;
2825 ptr++; /* Point after the escape sequence */
2826 }
2827
2828 else if ((cd->ctypes[*ptr] & ctype_meta) == 0)
2829 {
2830 #ifdef SUPPORT_UTF8
2831 if (utf8) { GETCHARINC(next, ptr); } else
2832 #endif
2833 next = *ptr++;
2834 }
2835
2836 else return FALSE;
2837
2838 /* Skip whitespace and comments in extended mode */
2839
2840 if ((options & PCRE_EXTENDED) != 0)
2841 {
2842 for (;;)
2843 {
2844 while ((cd->ctypes[*ptr] & ctype_space) != 0) ptr++;
2845 if (*ptr == CHAR_NUMBER_SIGN)
2846 {
2847 ptr++;
2848 while (*ptr != 0)
2849 {
2850 if (IS_NEWLINE(ptr)) { ptr += cd->nllen; break; }
2851 ptr++;
2852 #ifdef SUPPORT_UTF8
2853 if (utf8) while ((*ptr & 0xc0) == 0x80) ptr++;
2854 #endif
2855 }
2856 }
2857 else break;
2858 }
2859 }
2860
2861 /* If the next thing is itself optional, we have to give up. */
2862
2863 if (*ptr == CHAR_ASTERISK || *ptr == CHAR_QUESTION_MARK ||
2864 strncmp((char *)ptr, STR_LEFT_CURLY_BRACKET STR_0 STR_COMMA, 3) == 0)
2865 return FALSE;
2866
2867 /* Now compare the next item with the previous opcode. First, handle cases when
2868 the next item is a character. */
2869
2870 if (next >= 0) switch(op_code)
2871 {
2872 case OP_CHAR:
2873 #ifdef SUPPORT_UTF8
2874 GETCHARTEST(c, previous);
2875 #else
2876 c = *previous;
2877 #endif
2878 return c != next;
2879
2880 /* For CHARI (caseless character) we must check the other case. If we have
2881 Unicode property support, we can use it to test the other case of
2882 high-valued characters. */
2883
2884 case OP_CHARI:
2885 #ifdef SUPPORT_UTF8
2886 GETCHARTEST(c, previous);
2887 #else
2888 c = *previous;
2889 #endif
2890 if (c == next) return FALSE;
2891 #ifdef SUPPORT_UTF8
2892 if (utf8)
2893 {
2894 unsigned int othercase;
2895 if (next < 128) othercase = cd->fcc[next]; else
2896 #ifdef SUPPORT_UCP
2897 othercase = UCD_OTHERCASE((unsigned int)next);
2898 #else
2899 othercase = NOTACHAR;
2900 #endif
2901 return (unsigned int)c != othercase;
2902 }
2903 else
2904 #endif /* SUPPORT_UTF8 */
2905 return (c != cd->fcc[next]); /* Non-UTF-8 mode */
2906
2907 /* For OP_NOT and OP_NOTI, the data is always a single-byte character. These
2908 opcodes are not used for multi-byte characters, because they are coded using
2909 an XCLASS instead. */
2910
2911 case OP_NOT:
2912 return (c = *previous) == next;
2913
2914 case OP_NOTI:
2915 if ((c = *previous) == next) return TRUE;
2916 #ifdef SUPPORT_UTF8
2917 if (utf8)
2918 {
2919 unsigned int othercase;
2920 if (next < 128) othercase = cd->fcc[next]; else
2921 #ifdef SUPPORT_UCP
2922 othercase = UCD_OTHERCASE(next);
2923 #else
2924 othercase = NOTACHAR;
2925 #endif
2926 return (unsigned int)c == othercase;
2927 }
2928 else
2929 #endif /* SUPPORT_UTF8 */
2930 return (c == cd->fcc[next]); /* Non-UTF-8 mode */
2931
2932 /* Note that OP_DIGIT etc. are generated only when PCRE_UCP is *not* set.
2933 When it is set, \d etc. are converted into OP_(NOT_)PROP codes. */
2934
2935 case OP_DIGIT:
2936 return next > 127 || (cd->ctypes[next] & ctype_digit) == 0;
2937
2938 case OP_NOT_DIGIT:
2939 return next <= 127 && (cd->ctypes[next] & ctype_digit) != 0;
2940
2941 case OP_WHITESPACE:
2942 return next > 127 || (cd->ctypes[next] & ctype_space) == 0;
2943
2944 case OP_NOT_WHITESPACE:
2945 return next <= 127 && (cd->ctypes[next] & ctype_space) != 0;
2946
2947 case OP_WORDCHAR:
2948 return next > 127 || (cd->ctypes[next] & ctype_word) == 0;
2949
2950 case OP_NOT_WORDCHAR:
2951 return next <= 127 && (cd->ctypes[next] & ctype_word) != 0;
2952
2953 case OP_HSPACE:
2954 case OP_NOT_HSPACE:
2955 switch(next)
2956 {
2957 case 0x09:
2958 case 0x20:
2959 case 0xa0:
2960 case 0x1680:
2961 case 0x180e:
2962 case 0x2000:
2963 case 0x2001:
2964 case 0x2002:
2965 case 0x2003:
2966 case 0x2004:
2967 case 0x2005:
2968 case 0x2006:
2969 case 0x2007:
2970 case 0x2008:
2971 case 0x2009:
2972 case 0x200A:
2973 case 0x202f:
2974 case 0x205f:
2975 case 0x3000:
2976 return op_code == OP_NOT_HSPACE;
2977 default:
2978 return op_code != OP_NOT_HSPACE;
2979 }
2980
2981 case OP_ANYNL:
2982 case OP_VSPACE:
2983 case OP_NOT_VSPACE:
2984 switch(next)
2985 {
2986 case 0x0a:
2987 case 0x0b:
2988 case 0x0c:
2989 case 0x0d:
2990 case 0x85:
2991 case 0x2028:
2992 case 0x2029:
2993 return op_code == OP_NOT_VSPACE;
2994 default:
2995 return op_code != OP_NOT_VSPACE;
2996 }
2997
2998 #ifdef SUPPORT_UCP
2999 case OP_PROP:
3000 return check_char_prop(next, previous[0], previous[1], FALSE);
3001
3002 case OP_NOTPROP:
3003 return check_char_prop(next, previous[0], previous[1], TRUE);
3004 #endif
3005
3006 default:
3007 return FALSE;
3008 }
3009
3010
3011 /* Handle the case when the next item is \d, \s, etc. Note that when PCRE_UCP
3012 is set, \d turns into ESC_du rather than ESC_d, etc., so ESC_d etc. are
3013 generated only when PCRE_UCP is *not* set, that is, when only ASCII
3014 characteristics are recognized. Similarly, the opcodes OP_DIGIT etc. are
3015 replaced by OP_PROP codes when PCRE_UCP is set. */
3016
3017 switch(op_code)
3018 {
3019 case OP_CHAR:
3020 case OP_CHARI:
3021 #ifdef SUPPORT_UTF8
3022 GETCHARTEST(c, previous);
3023 #else
3024 c = *previous;
3025 #endif
3026 switch(-next)
3027 {
3028 case ESC_d:
3029 return c > 127 || (cd->ctypes[c] & ctype_digit) == 0;
3030
3031 case ESC_D:
3032 return c <= 127 && (cd->ctypes[c] & ctype_digit) != 0;
3033
3034 case ESC_s:
3035 return c > 127 || (cd->ctypes[c] & ctype_space) == 0;
3036
3037 case ESC_S:
3038 return c <= 127 && (cd->ctypes[c] & ctype_space) != 0;
3039
3040 case ESC_w:
3041 return c > 127 || (cd->ctypes[c] & ctype_word) == 0;
3042
3043 case ESC_W:
3044 return c <= 127 && (cd->ctypes[c] & ctype_word) != 0;
3045
3046 case ESC_h:
3047 case ESC_H:
3048 switch(c)
3049 {
3050 case 0x09:
3051 case 0x20:
3052 case 0xa0:
3053 case 0x1680:
3054 case 0x180e:
3055 case 0x2000:
3056 case 0x2001:
3057 case 0x2002:
3058 case 0x2003:
3059 case 0x2004:
3060 case 0x2005:
3061 case 0x2006:
3062 case 0x2007:
3063 case 0x2008:
3064 case 0x2009:
3065 case 0x200A:
3066 case 0x202f:
3067 case 0x205f:
3068 case 0x3000:
3069 return -next != ESC_h;
3070 default:
3071 return -next == ESC_h;
3072 }
3073
3074 case ESC_v:
3075 case ESC_V:
3076 switch(c)
3077 {
3078 case 0x0a:
3079 case 0x0b:
3080 case 0x0c:
3081 case 0x0d:
3082 case 0x85:
3083 case 0x2028:
3084 case 0x2029:
3085 return -next != ESC_v;
3086 default:
3087 return -next == ESC_v;
3088 }
3089
3090 /* When PCRE_UCP is set, these values get generated for \d etc. Find
3091 their substitutions and process them. The result will always be either
3092 -ESC_p or -ESC_P. Then fall through to process those values. */
3093
3094 #ifdef SUPPORT_UCP
3095 case ESC_du:
3096 case ESC_DU:
3097 case ESC_wu:
3098 case ESC_WU:
3099 case ESC_su:
3100 case ESC_SU:
3101 {
3102 int temperrorcode = 0;
3103 ptr = substitutes[-next - ESC_DU];
3104 next = check_escape(&ptr, &temperrorcode, 0, options, FALSE);
3105 if (temperrorcode != 0) return FALSE;
3106 ptr++; /* For compatibility */
3107 }
3108 /* Fall through */
3109
3110 case ESC_p:
3111 case ESC_P:
3112 {
3113 int ptype, pdata, errorcodeptr;
3114 BOOL negated;
3115
3116 ptr--; /* Make ptr point at the p or P */
3117 ptype = get_ucp(&ptr, &negated, &pdata, &errorcodeptr);
3118 if (ptype < 0) return FALSE;
3119 ptr++; /* Point past the final curly ket */
3120
3121 /* If the property item is optional, we have to give up. (When generated
3122 from \d etc by PCRE_UCP, this test will have been applied much earlier,
3123 to the original \d etc. At this point, ptr will point to a zero byte. */
3124
3125 if (*ptr == CHAR_ASTERISK || *ptr == CHAR_QUESTION_MARK ||
3126 strncmp((char *)ptr, STR_LEFT_CURLY_BRACKET STR_0 STR_COMMA, 3) == 0)
3127 return FALSE;
3128
3129 /* Do the property check. */
3130
3131 return check_char_prop(c, ptype, pdata, (next == -ESC_P) != negated);
3132 }
3133 #endif
3134
3135 default:
3136 return FALSE;
3137 }
3138
3139 /* In principle, support for Unicode properties should be integrated here as
3140 well. It means re-organizing the above code so as to get hold of the property
3141 values before switching on the op-code. However, I wonder how many patterns
3142 combine ASCII \d etc with Unicode properties? (Note that if PCRE_UCP is set,
3143 these op-codes are never generated.) */
3144
3145 case OP_DIGIT:
3146 return next == -ESC_D || next == -ESC_s || next == -ESC_W ||
3147 next == -ESC_h || next == -ESC_v || next == -ESC_R;
3148
3149 case OP_NOT_DIGIT:
3150 return next == -ESC_d;
3151
3152 case OP_WHITESPACE:
3153 return next == -ESC_S || next == -ESC_d || next == -ESC_w || next == -ESC_R;
3154
3155 case OP_NOT_WHITESPACE:
3156 return next == -ESC_s || next == -ESC_h || next == -ESC_v;
3157
3158 case OP_HSPACE:
3159 return next == -ESC_S || next == -ESC_H || next == -ESC_d ||
3160 next == -ESC_w || next == -ESC_v || next == -ESC_R;
3161
3162 case OP_NOT_HSPACE:
3163 return next == -ESC_h;
3164
3165 /* Can't have \S in here because VT matches \S (Perl anomaly) */
3166 case OP_ANYNL:
3167 case OP_VSPACE:
3168 return next == -ESC_V || next == -ESC_d || next == -ESC_w;
3169
3170 case OP_NOT_VSPACE:
3171 return next == -ESC_v || next == -ESC_R;
3172
3173 case OP_WORDCHAR:
3174 return next == -ESC_W || next == -ESC_s || next == -ESC_h ||
3175 next == -ESC_v || next == -ESC_R;
3176
3177 case OP_NOT_WORDCHAR:
3178 return next == -ESC_w || next == -ESC_d;
3179
3180 default:
3181 return FALSE;
3182 }
3183
3184 /* Control does not reach here */
3185 }
3186
3187
3188
3189 /*************************************************
3190 * Compile one branch *
3191 *************************************************/
3192
3193 /* Scan the pattern, compiling it into the a vector. If the options are
3194 changed during the branch, the pointer is used to change the external options
3195 bits. This function is used during the pre-compile phase when we are trying
3196 to find out the amount of memory needed, as well as during the real compile
3197 phase. The value of lengthptr distinguishes the two phases.
3198
3199 Arguments:
3200 optionsptr pointer to the option bits
3201 codeptr points to the pointer to the current code point
3202 ptrptr points to the current pattern pointer
3203 errorcodeptr points to error code variable
3204 firstbyteptr set to initial literal character, or < 0 (REQ_UNSET, REQ_NONE)
3205 reqbyteptr set to the last literal character required, else < 0
3206 bcptr points to current branch chain
3207 cond_depth conditional nesting depth
3208 cd contains pointers to tables etc.
3209 lengthptr NULL during the real compile phase
3210 points to length accumulator during pre-compile phase
3211
3212 Returns: TRUE on success
3213 FALSE, with *errorcodeptr set non-zero on error
3214 */
3215
3216 static BOOL
3217 compile_branch(int *optionsptr, uschar **codeptr, const uschar **ptrptr,
3218 int *errorcodeptr, int *firstbyteptr, int *reqbyteptr, branch_chain *bcptr,
3219 int cond_depth, compile_data *cd, int *lengthptr)
3220 {
3221 int repeat_type, op_type;
3222 int repeat_min = 0, repeat_max = 0; /* To please picky compilers */
3223 int bravalue = 0;
3224 int greedy_default, greedy_non_default;
3225 int firstbyte, reqbyte;
3226 int zeroreqbyte, zerofirstbyte;
3227 int req_caseopt, reqvary, tempreqvary;
3228 int options = *optionsptr; /* May change dynamically */
3229 int after_manual_callout = 0;
3230 int length_prevgroup = 0;
3231 register int c;
3232 register uschar *code = *codeptr;
3233 uschar *last_code = code;
3234 uschar *orig_code = code;
3235 uschar *tempcode;
3236 BOOL inescq = FALSE;
3237 BOOL groupsetfirstbyte = FALSE;
3238 const uschar *ptr = *ptrptr;
3239 const uschar *tempptr;
3240 const uschar *nestptr = NULL;
3241 uschar *previous = NULL;
3242 uschar *previous_callout = NULL;
3243 uschar *save_hwm = NULL;
3244 uschar classbits[32];
3245
3246 /* We can fish out the UTF-8 setting once and for all into a BOOL, but we
3247 must not do this for other options (e.g. PCRE_EXTENDED) because they may change
3248 dynamically as we process the pattern. */
3249
3250 #ifdef SUPPORT_UTF8
3251 BOOL class_utf8;
3252 BOOL utf8 = (options & PCRE_UTF8) != 0;
3253 uschar *class_utf8data;
3254 uschar *class_utf8data_base;
3255 uschar utf8_char[6];
3256 #else
3257 BOOL utf8 = FALSE;
3258 #endif
3259
3260 #ifdef PCRE_DEBUG
3261 if (lengthptr != NULL) DPRINTF((">> start branch\n"));
3262 #endif
3263
3264 /* Set up the default and non-default settings for greediness */
3265
3266 greedy_default = ((options & PCRE_UNGREEDY) != 0);
3267 greedy_non_default = greedy_default ^ 1;
3268
3269 /* Initialize no first byte, no required byte. REQ_UNSET means "no char
3270 matching encountered yet". It gets changed to REQ_NONE if we hit something that
3271 matches a non-fixed char first char; reqbyte just remains unset if we never
3272 find one.
3273
3274 When we hit a repeat whose minimum is zero, we may have to adjust these values
3275 to take the zero repeat into account. This is implemented by setting them to
3276 zerofirstbyte and zeroreqbyte when such a repeat is encountered. The individual
3277 item types that can be repeated set these backoff variables appropriately. */
3278
3279 firstbyte = reqbyte = zerofirstbyte = zeroreqbyte = REQ_UNSET;
3280
3281 /* The variable req_caseopt contains either the REQ_CASELESS value or zero,
3282 according to the current setting of the caseless flag. REQ_CASELESS is a bit
3283 value > 255. It is added into the firstbyte or reqbyte variables to record the
3284 case status of the value. This is used only for ASCII characters. */
3285
3286 req_caseopt = ((options & PCRE_CASELESS) != 0)? REQ_CASELESS : 0;
3287
3288 /* Switch on next character until the end of the branch */
3289
3290 for (;; ptr++)
3291 {
3292 BOOL negate_class;
3293 BOOL should_flip_negation;
3294 BOOL possessive_quantifier;
3295 BOOL is_quantifier;
3296 BOOL is_recurse;
3297 BOOL reset_bracount;
3298 int class_charcount;
3299 int class_lastchar;
3300 int newoptions;
3301 int recno;
3302 int refsign;
3303 int skipbytes;
3304 int subreqbyte;
3305 int subfirstbyte;
3306 int terminator;
3307 int mclength;
3308 int tempbracount;
3309 uschar mcbuffer[8];
3310
3311 /* Get next byte in the pattern */
3312
3313 c = *ptr;
3314
3315 /* If we are at the end of a nested substitution, revert to the outer level
3316 string. Nesting only happens one level deep. */
3317
3318 if (c == 0 && nestptr != NULL)
3319 {
3320 ptr = nestptr;
3321 nestptr = NULL;
3322 c = *ptr;
3323 }
3324
3325 /* If we are in the pre-compile phase, accumulate the length used for the
3326 previous cycle of this loop. */
3327
3328 if (lengthptr != NULL)
3329 {
3330 #ifdef PCRE_DEBUG
3331 if (code > cd->hwm) cd->hwm = code; /* High water info */
3332 #endif
3333 if (code > cd->start_workspace + WORK_SIZE_CHECK) /* Check for overrun */
3334 {
3335 *errorcodeptr = ERR52;
3336 goto FAILED;
3337 }
3338
3339 /* There is at least one situation where code goes backwards: this is the
3340 case of a zero quantifier after a class (e.g. [ab]{0}). At compile time,
3341 the class is simply eliminated. However, it is created first, so we have to
3342 allow memory for it. Therefore, don't ever reduce the length at this point.
3343 */
3344
3345 if (code < last_code) code = last_code;
3346
3347 /* Paranoid check for integer overflow */
3348
3349 if (OFLOW_MAX - *lengthptr < code - last_code)
3350 {
3351 *errorcodeptr = ERR20;
3352 goto FAILED;
3353 }
3354
3355 *lengthptr += (int)(code - last_code);
3356 DPRINTF(("length=%d added %d c=%c\n", *lengthptr, (int)(code - last_code),
3357 c));
3358
3359 /* If "previous" is set and it is not at the start of the work space, move
3360 it back to there, in order to avoid filling up the work space. Otherwise,
3361 if "previous" is NULL, reset the current code pointer to the start. */
3362
3363 if (previous != NULL)
3364 {
3365 if (previous > orig_code)
3366 {
3367 memmove(orig_code, previous, code - previous);
3368 code -= previous - orig_code;
3369 previous = orig_code;
3370 }
3371 }
3372 else code = orig_code;
3373
3374 /* Remember where this code item starts so we can pick up the length
3375 next time round. */
3376
3377 last_code = code;
3378 }
3379
3380 /* In the real compile phase, just check the workspace used by the forward
3381 reference list. */
3382
3383 else if (cd->hwm > cd->start_workspace + WORK_SIZE_CHECK)
3384 {
3385 *errorcodeptr = ERR52;
3386 goto FAILED;
3387 }
3388
3389 /* If in \Q...\E, check for the end; if not, we have a literal */
3390
3391 if (inescq && c != 0)
3392 {
3393 if (c == CHAR_BACKSLASH && ptr[1] == CHAR_E)
3394 {
3395 inescq = FALSE;
3396 ptr++;
3397 continue;
3398 }
3399 else
3400 {
3401 if (previous_callout != NULL)
3402 {
3403 if (lengthptr == NULL) /* Don't attempt in pre-compile phase */
3404 complete_callout(previous_callout, ptr, cd);
3405 previous_callout = NULL;
3406 }
3407 if ((options & PCRE_AUTO_CALLOUT) != 0)
3408 {
3409 previous_callout = code;
3410 code = auto_callout(code, ptr, cd);
3411 }
3412 goto NORMAL_CHAR;
3413 }
3414 }
3415
3416 /* Fill in length of a previous callout, except when the next thing is
3417 a quantifier. */
3418
3419 is_quantifier =
3420 c == CHAR_ASTERISK || c == CHAR_PLUS || c == CHAR_QUESTION_MARK ||
3421 (c == CHAR_LEFT_CURLY_BRACKET && is_counted_repeat(ptr+1));
3422
3423 if (!is_quantifier && previous_callout != NULL &&
3424 after_manual_callout-- <= 0)
3425 {
3426 if (lengthptr == NULL) /* Don't attempt in pre-compile phase */
3427 complete_callout(previous_callout, ptr, cd);
3428 previous_callout = NULL;
3429 }
3430
3431 /* In extended mode, skip white space and comments. */
3432
3433 if ((options & PCRE_EXTENDED) != 0)
3434 {
3435 if ((cd->ctypes[c] & ctype_space) != 0) continue;
3436 if (c == CHAR_NUMBER_SIGN)
3437 {
3438 ptr++;
3439 while (*ptr != 0)
3440 {
3441 if (IS_NEWLINE(ptr)) { ptr += cd->nllen - 1; break; }
3442 ptr++;
3443 #ifdef SUPPORT_UTF8
3444 if (utf8) while ((*ptr & 0xc0) == 0x80) ptr++;
3445 #endif
3446 }
3447 if (*ptr != 0) continue;
3448
3449 /* Else fall through to handle end of string */
3450 c = 0;
3451 }
3452 }
3453
3454 /* No auto callout for quantifiers. */
3455
3456 if ((options & PCRE_AUTO_CALLOUT) != 0 && !is_quantifier)
3457 {
3458 previous_callout = code;
3459 code = auto_callout(code, ptr, cd);
3460 }
3461
3462 switch(c)
3463 {
3464 /* ===================================================================*/
3465 case 0: /* The branch terminates at string end */
3466 case CHAR_VERTICAL_LINE: /* or | or ) */
3467 case CHAR_RIGHT_PARENTHESIS:
3468 *firstbyteptr = firstbyte;
3469 *reqbyteptr = reqbyte;
3470 *codeptr = code;
3471 *ptrptr = ptr;
3472 if (lengthptr != NULL)
3473 {
3474 if (OFLOW_MAX - *lengthptr < code - last_code)
3475 {
3476 *errorcodeptr = ERR20;
3477 goto FAILED;
3478 }
3479 *lengthptr += (int)(code - last_code); /* To include callout length */
3480 DPRINTF((">> end branch\n"));
3481 }
3482 return TRUE;
3483
3484
3485 /* ===================================================================*/
3486 /* Handle single-character metacharacters. In multiline mode, ^ disables
3487 the setting of any following char as a first character. */
3488
3489 case CHAR_CIRCUMFLEX_ACCENT:
3490 previous = NULL;
3491 if ((options & PCRE_MULTILINE) != 0)
3492 {
3493 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
3494 *code++ = OP_CIRCM;
3495 }
3496 else *code++ = OP_CIRC;
3497 break;
3498
3499 case CHAR_DOLLAR_SIGN:
3500 previous = NULL;
3501 *code++ = ((options & PCRE_MULTILINE) != 0)? OP_DOLLM : OP_DOLL;
3502 break;
3503
3504 /* There can never be a first char if '.' is first, whatever happens about
3505 repeats. The value of reqbyte doesn't change either. */
3506
3507 case CHAR_DOT:
3508 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
3509 zerofirstbyte = firstbyte;
3510 zeroreqbyte = reqbyte;
3511 previous = code;
3512 *code++ = ((options & PCRE_DOTALL) != 0)? OP_ALLANY: OP_ANY;
3513 break;
3514
3515
3516 /* ===================================================================*/
3517 /* Character classes. If the included characters are all < 256, we build a
3518 32-byte bitmap of the permitted characters, except in the special case
3519 where there is only one such character. For negated classes, we build the
3520 map as usual, then invert it at the end. However, we use a different opcode
3521 so that data characters > 255 can be handled correctly.
3522
3523 If the class contains characters outside the 0-255 range, a different
3524 opcode is compiled. It may optionally have a bit map for characters < 256,
3525 but those above are are explicitly listed afterwards. A flag byte tells
3526 whether the bitmap is present, and whether this is a negated class or not.
3527
3528 In JavaScript compatibility mode, an isolated ']' causes an error. In
3529 default (Perl) mode, it is treated as a data character. */
3530
3531 case CHAR_RIGHT_SQUARE_BRACKET:
3532 if ((cd->external_options & PCRE_JAVASCRIPT_COMPAT) != 0)
3533 {
3534 *errorcodeptr = ERR64;
3535 goto FAILED;
3536 }
3537 goto NORMAL_CHAR;
3538
3539 case CHAR_LEFT_SQUARE_BRACKET:
3540 previous = code;
3541
3542 /* PCRE supports POSIX class stuff inside a class. Perl gives an error if
3543 they are encountered at the top level, so we'll do that too. */
3544
3545 if ((ptr[1] == CHAR_COLON || ptr[1] == CHAR_DOT ||
3546 ptr[1] == CHAR_EQUALS_SIGN) &&
3547 check_posix_syntax(ptr, &tempptr))
3548 {
3549 *errorcodeptr = (ptr[1] == CHAR_COLON)? ERR13 : ERR31;
3550 goto FAILED;
3551 }
3552
3553 /* If the first character is '^', set the negation flag and skip it. Also,
3554 if the first few characters (either before or after ^) are \Q\E or \E we
3555 skip them too. This makes for compatibility with Perl. */
3556
3557 negate_class = FALSE;
3558 for (;;)
3559 {
3560 c = *(++ptr);
3561 if (c == CHAR_BACKSLASH)
3562 {
3563 if (ptr[1] == CHAR_E)
3564 ptr++;
3565 else if (strncmp((const char *)ptr+1,
3566 STR_Q STR_BACKSLASH STR_E, 3) == 0)
3567 ptr += 3;
3568 else
3569 break;
3570 }
3571 else if (!negate_class && c == CHAR_CIRCUMFLEX_ACCENT)
3572 negate_class = TRUE;
3573 else break;
3574 }
3575
3576 /* Empty classes are allowed in JavaScript compatibility mode. Otherwise,
3577 an initial ']' is taken as a data character -- the code below handles
3578 that. In JS mode, [] must always fail, so generate OP_FAIL, whereas
3579 [^] must match any character, so generate OP_ALLANY. */
3580
3581 if (c == CHAR_RIGHT_SQUARE_BRACKET &&
3582 (cd->external_options & PCRE_JAVASCRIPT_COMPAT) != 0)
3583 {
3584 *code++ = negate_class? OP_ALLANY : OP_FAIL;
3585 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
3586 zerofirstbyte = firstbyte;
3587 break;
3588 }
3589
3590 /* If a class contains a negative special such as \S, we need to flip the
3591 negation flag at the end, so that support for characters > 255 works
3592 correctly (they are all included in the class). */
3593
3594 should_flip_negation = FALSE;
3595
3596 /* Keep a count of chars with values < 256 so that we can optimize the case
3597 of just a single character (as long as it's < 256). However, For higher
3598 valued UTF-8 characters, we don't yet do any optimization. */
3599
3600 class_charcount = 0;
3601 class_lastchar = -1;
3602
3603 /* Initialize the 32-char bit map to all zeros. We build the map in a
3604 temporary bit of memory, in case the class contains only 1 character (less
3605 than 256), because in that case the compiled code doesn't use the bit map.
3606 */
3607
3608 memset(classbits, 0, 32 * sizeof(uschar));
3609
3610 #ifdef SUPPORT_UTF8
3611 class_utf8 = FALSE; /* No chars >= 256 */
3612 class_utf8data = code + LINK_SIZE + 2; /* For UTF-8 items */
3613 class_utf8data_base = class_utf8data; /* For resetting in pass 1 */
3614 #endif
3615
3616 /* Process characters until ] is reached. By writing this as a "do" it
3617 means that an initial ] is taken as a data character. At the start of the
3618 loop, c contains the first byte of the character. */
3619
3620 if (c != 0) do
3621 {
3622 const uschar *oldptr;
3623
3624 #ifdef SUPPORT_UTF8
3625 if (utf8 && c > 127)
3626 { /* Braces are required because the */
3627 GETCHARLEN(c, ptr, ptr); /* macro generates multiple statements */
3628 }
3629
3630 /* In the pre-compile phase, accumulate the length of any UTF-8 extra
3631 data and reset the pointer. This is so that very large classes that
3632 contain a zillion UTF-8 characters no longer overwrite the work space
3633 (which is on the stack). */
3634
3635 if (lengthptr != NULL)
3636 {
3637 *lengthptr += class_utf8data - class_utf8data_base;
3638 class_utf8data = class_utf8data_base;
3639 }
3640
3641 #endif
3642
3643 /* Inside \Q...\E everything is literal except \E */
3644
3645 if (inescq)
3646 {
3647 if (c == CHAR_BACKSLASH && ptr[1] == CHAR_E) /* If we are at \E */
3648 {
3649 inescq = FALSE; /* Reset literal state */
3650 ptr++; /* Skip the 'E' */
3651 continue; /* Carry on with next */
3652 }
3653 goto CHECK_RANGE; /* Could be range if \E follows */
3654 }
3655
3656 /* Handle POSIX class names. Perl allows a negation extension of the
3657 form [:^name:]. A square bracket that doesn't match the syntax is
3658 treated as a literal. We also recognize the POSIX constructions
3659 [.ch.] and [=ch=] ("collating elements") and fault them, as Perl
3660 5.6 and 5.8 do. */
3661
3662 if (c == CHAR_LEFT_SQUARE_BRACKET &&
3663 (ptr[1] == CHAR_COLON || ptr[1] == CHAR_DOT ||
3664 ptr[1] == CHAR_EQUALS_SIGN) && check_posix_syntax(ptr, &tempptr))
3665 {
3666 BOOL local_negate = FALSE;
3667 int posix_class, taboffset, tabopt;
3668 register const uschar *cbits = cd->cbits;
3669 uschar pbits[32];
3670
3671 if (ptr[1] != CHAR_COLON)
3672 {
3673 *errorcodeptr = ERR31;
3674 goto FAILED;
3675 }
3676
3677 ptr += 2;
3678 if (*ptr == CHAR_CIRCUMFLEX_ACCENT)
3679 {
3680 local_negate = TRUE;
3681 should_flip_negation = TRUE; /* Note negative special */
3682 ptr++;
3683 }
3684
3685 posix_class = check_posix_name(ptr, (int)(tempptr - ptr));
3686 if (posix_class < 0)
3687 {
3688 *errorcodeptr = ERR30;
3689 goto FAILED;
3690 }
3691
3692 /* If matching is caseless, upper and lower are converted to
3693 alpha. This relies on the fact that the class table starts with
3694 alpha, lower, upper as the first 3 entries. */
3695
3696 if ((options & PCRE_CASELESS) != 0 && posix_class <= 2)
3697 posix_class = 0;
3698
3699 /* When PCRE_UCP is set, some of the POSIX classes are converted to
3700 different escape sequences that use Unicode properties. */
3701
3702 #ifdef SUPPORT_UCP
3703 if ((options & PCRE_UCP) != 0)
3704 {
3705 int pc = posix_class + ((local_negate)? POSIX_SUBSIZE/2 : 0);
3706 if (posix_substitutes[pc] != NULL)
3707 {
3708 nestptr = tempptr + 1;
3709 ptr = posix_substitutes[pc] - 1;
3710 continue;
3711 }
3712 }
3713 #endif
3714 /* In the non-UCP case, we build the bit map for the POSIX class in a
3715 chunk of local store because we may be adding and subtracting from it,
3716 and we don't want to subtract bits that may be in the main map already.
3717 At the end we or the result into the bit map that is being built. */
3718
3719 posix_class *= 3;
3720
3721 /* Copy in the first table (always present) */
3722
3723 memcpy(pbits, cbits + posix_class_maps[posix_class],
3724 32 * sizeof(uschar));
3725
3726 /* If there is a second table, add or remove it as required. */
3727
3728 taboffset = posix_class_maps[posix_class + 1];
3729 tabopt = posix_class_maps[posix_class + 2];
3730
3731 if (taboffset >= 0)
3732 {
3733 if (tabopt >= 0)
3734 for (c = 0; c < 32; c++) pbits[c] |= cbits[c + taboffset];
3735 else
3736 for (c = 0; c < 32; c++) pbits[c] &= ~cbits[c + taboffset];
3737 }
3738
3739 /* Not see if we need to remove any special characters. An option
3740 value of 1 removes vertical space and 2 removes underscore. */
3741
3742 if (tabopt < 0) tabopt = -tabopt;
3743 if (tabopt == 1) pbits[1] &= ~0x3c;
3744 else if (tabopt == 2) pbits[11] &= 0x7f;
3745
3746 /* Add the POSIX table or its complement into the main table that is
3747 being built and we are done. */
3748
3749 if (local_negate)
3750 for (c = 0; c < 32; c++) classbits[c] |= ~pbits[c];
3751 else
3752 for (c = 0; c < 32; c++) classbits[c] |= pbits[c];
3753
3754 ptr = tempptr + 1;
3755 class_charcount = 10; /* Set > 1; assumes more than 1 per class */
3756 continue; /* End of POSIX syntax handling */
3757 }
3758
3759 /* Backslash may introduce a single character, or it may introduce one
3760 of the specials, which just set a flag. The sequence \b is a special
3761 case. Inside a class (and only there) it is treated as backspace. We
3762 assume that other escapes have more than one character in them, so set
3763 class_charcount bigger than one. Unrecognized escapes fall through and
3764 are either treated as literal characters (by default), or are faulted if
3765 PCRE_EXTRA is set. */
3766
3767 if (c == CHAR_BACKSLASH)
3768 {
3769 c = check_escape(&ptr, errorcodeptr, cd->bracount, options, TRUE);
3770 if (*errorcodeptr != 0) goto FAILED;
3771
3772 if (-c == ESC_b) c = CHAR_BS; /* \b is backspace in a class */
3773 else if (-c == ESC_Q) /* Handle start of quoted string */
3774 {
3775 if (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
3776 {
3777 ptr += 2; /* avoid empty string */
3778 }
3779 else inescq = TRUE;
3780 continue;
3781 }
3782 else if (-c == ESC_E) continue; /* Ignore orphan \E */
3783
3784 if (c < 0)
3785 {
3786 register const uschar *cbits = cd->cbits;
3787 class_charcount += 2; /* Greater than 1 is what matters */
3788
3789 switch (-c)
3790 {
3791 #ifdef SUPPORT_UCP
3792 case ESC_du: /* These are the values given for \d etc */
3793 case ESC_DU: /* when PCRE_UCP is set. We replace the */
3794 case ESC_wu: /* escape sequence with an appropriate \p */
3795 case ESC_WU: /* or \P to test Unicode properties instead */
3796 case ESC_su: /* of the default ASCII testing. */
3797 case ESC_SU:
3798 nestptr = ptr;
3799 ptr = substitutes[-c - ESC_DU] - 1; /* Just before substitute */
3800 class_charcount -= 2; /* Undo! */
3801 continue;
3802 #endif
3803 case ESC_d:
3804 for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_digit];
3805 continue;
3806
3807 case ESC_D:
3808 should_flip_negation = TRUE;
3809 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_digit];
3810 continue;
3811
3812 case ESC_w:
3813 for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_word];
3814 continue;
3815
3816 case ESC_W:
3817 should_flip_negation = TRUE;
3818 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_word];
3819 continue;
3820
3821 /* Perl 5.004 onwards omits VT from \s, but we must preserve it
3822 if it was previously set by something earlier in the character
3823 class. */
3824
3825 case ESC_s:
3826 classbits[0] |= cbits[cbit_space];
3827 classbits[1] |= cbits[cbit_space+1] & ~0x08;
3828 for (c = 2; c < 32; c++) classbits[c] |= cbits[c+cbit_space];
3829 continue;
3830
3831 case ESC_S:
3832 should_flip_negation = TRUE;
3833 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_space];
3834 classbits[1] |= 0x08; /* Perl 5.004 onwards omits VT from \s */
3835 continue;
3836
3837 case ESC_h:
3838 SETBIT(classbits, 0x09); /* VT */
3839 SETBIT(classbits, 0x20); /* SPACE */
3840 SETBIT(classbits, 0xa0); /* NSBP */
3841 #ifdef SUPPORT_UTF8
3842 if (utf8)
3843 {
3844 class_utf8 = TRUE;
3845 *class_utf8data++ = XCL_SINGLE;
3846 class_utf8data += _pcre_ord2utf8(0x1680, class_utf8data);
3847 *class_utf8data++ = XCL_SINGLE;
3848 class_utf8data += _pcre_ord2utf8(0x180e, class_utf8data);
3849 *class_utf8data++ = XCL_RANGE;
3850 class_utf8data += _pcre_ord2utf8(0x2000, class_utf8data);
3851 class_utf8data += _pcre_ord2utf8(0x200A, class_utf8data);
3852 *class_utf8data++ = XCL_SINGLE;
3853 class_utf8data += _pcre_ord2utf8(0x202f, class_utf8data);
3854 *class_utf8data++ = XCL_SINGLE;
3855 class_utf8data += _pcre_ord2utf8(0x205f, class_utf8data);
3856 *class_utf8data++ = XCL_SINGLE;
3857 class_utf8data += _pcre_ord2utf8(0x3000, class_utf8data);
3858 }
3859 #endif
3860 continue;
3861
3862 case ESC_H:
3863 for (c = 0; c < 32; c++)
3864 {
3865 int x = 0xff;
3866 switch (c)
3867 {
3868 case 0x09/8: x ^= 1 << (0x09%8); break;
3869 case 0x20/8: x ^= 1 << (0x20%8); break;
3870 case 0xa0/8: x ^= 1 << (0xa0%8); break;
3871 default: break;
3872 }
3873 classbits[c] |= x;
3874 }
3875
3876 #ifdef SUPPORT_UTF8
3877 if (utf8)
3878 {
3879 class_utf8 = TRUE;
3880 *class_utf8data++ = XCL_RANGE;
3881 class_utf8data += _pcre_ord2utf8(0x0100, class_utf8data);
3882 class_utf8data += _pcre_ord2utf8(0x167f, class_utf8data);
3883 *class_utf8data++ = XCL_RANGE;
3884 class_utf8data += _pcre_ord2utf8(0x1681, class_utf8data);
3885 class_utf8data += _pcre_ord2utf8(0x180d, class_utf8data);
3886 *class_utf8data++ = XCL_RANGE;
3887 class_utf8data += _pcre_ord2utf8(0x180f, class_utf8data);
3888 class_utf8data += _pcre_ord2utf8(0x1fff, class_utf8data);
3889 *class_utf8data++ = XCL_RANGE;
3890 class_utf8data += _pcre_ord2utf8(0x200B, class_utf8data);
3891 class_utf8data += _pcre_ord2utf8(0x202e, class_utf8data);
3892 *class_utf8data++ = XCL_RANGE;
3893 class_utf8data += _pcre_ord2utf8(0x2030, class_utf8data);
3894 class_utf8data += _pcre_ord2utf8(0x205e, class_utf8data);
3895 *class_utf8data++ = XCL_RANGE;
3896 class_utf8data += _pcre_ord2utf8(0x2060, class_utf8data);
3897 class_utf8data += _pcre_ord2utf8(0x2fff, class_utf8data);
3898 *class_utf8data++ = XCL_RANGE;
3899 class_utf8data += _pcre_ord2utf8(0x3001, class_utf8data);
3900 class_utf8data += _pcre_ord2utf8(0x7fffffff, class_utf8data);
3901 }
3902 #endif
3903 continue;
3904
3905 case ESC_v:
3906 SETBIT(classbits, 0x0a); /* LF */
3907 SETBIT(classbits, 0x0b); /* VT */
3908 SETBIT(classbits, 0x0c); /* FF */
3909 SETBIT(classbits, 0x0d); /* CR */
3910 SETBIT(classbits, 0x85); /* NEL */
3911 #ifdef SUPPORT_UTF8
3912 if (utf8)
3913 {
3914 class_utf8 = TRUE;
3915 *class_utf8data++ = XCL_RANGE;
3916 class_utf8data += _pcre_ord2utf8(0x2028, class_utf8data);
3917 class_utf8data += _pcre_ord2utf8(0x2029, class_utf8data);
3918 }
3919 #endif
3920 continue;
3921
3922 case ESC_V:
3923 for (c = 0; c < 32; c++)
3924 {
3925 int x = 0xff;
3926 switch (c)
3927 {
3928 case 0x0a/8: x ^= 1 << (0x0a%8);
3929 x ^= 1 << (0x0b%8);
3930 x ^= 1 << (0x0c%8);
3931 x ^= 1 << (0x0d%8);
3932 break;
3933 case 0x85/8: x ^= 1 << (0x85%8); break;
3934 default: break;
3935 }
3936 classbits[c] |= x;
3937 }
3938
3939 #ifdef SUPPORT_UTF8
3940 if (utf8)
3941 {
3942 class_utf8 = TRUE;
3943 *class_utf8data++ = XCL_RANGE;
3944 class_utf8data += _pcre_ord2utf8(0x0100, class_utf8data);
3945 class_utf8data += _pcre_ord2utf8(0x2027, class_utf8data);
3946 *class_utf8data++ = XCL_RANGE;
3947 class_utf8data += _pcre_ord2utf8(0x2029, class_utf8data);
3948 class_utf8data += _pcre_ord2utf8(0x7fffffff, class_utf8data);
3949 }
3950 #endif
3951 continue;
3952
3953 #ifdef SUPPORT_UCP
3954 case ESC_p:
3955 case ESC_P:
3956 {
3957 BOOL negated;
3958 int pdata;
3959 int ptype = get_ucp(&ptr, &negated, &pdata, errorcodeptr);
3960 if (ptype < 0) goto FAILED;
3961 class_utf8 = TRUE;
3962 *class_utf8data++ = ((-c == ESC_p) != negated)?
3963 XCL_PROP : XCL_NOTPROP;
3964 *class_utf8data++ = ptype;
3965 *class_utf8data++ = pdata;
3966 class_charcount -= 2; /* Not a < 256 character */
3967 continue;
3968 }
3969 #endif
3970 /* Unrecognized escapes are faulted if PCRE is running in its
3971 strict mode. By default, for compatibility with Perl, they are
3972 treated as literals. */
3973
3974 default:
3975 if ((options & PCRE_EXTRA) != 0)
3976 {
3977 *errorcodeptr = ERR7;
3978 goto FAILED;
3979 }
3980 class_charcount -= 2; /* Undo the default count from above */
3981 c = *ptr; /* Get the final character and fall through */
3982 break;
3983 }
3984 }
3985
3986 /* Fall through if we have a single character (c >= 0). This may be
3987 greater than 256 in UTF-8 mode. */
3988
3989 } /* End of backslash handling */
3990
3991 /* A single character may be followed by '-' to form a range. However,
3992 Perl does not permit ']' to be the end of the range. A '-' character
3993 at the end is treated as a literal. Perl ignores orphaned \E sequences
3994 entirely. The code for handling \Q and \E is messy. */
3995
3996 CHECK_RANGE:
3997 while (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
3998 {
3999 inescq = FALSE;
4000 ptr += 2;
4001 }
4002
4003 oldptr = ptr;
4004
4005 /* Remember \r or \n */
4006
4007 if (c == CHAR_CR || c == CHAR_NL) cd->external_flags |= PCRE_HASCRORLF;
4008
4009 /* Check for range */
4010
4011 if (!inescq && ptr[1] == CHAR_MINUS)
4012 {
4013 int d;
4014 ptr += 2;
4015 while (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_E) ptr += 2;
4016
4017 /* If we hit \Q (not followed by \E) at this point, go into escaped
4018 mode. */
4019
4020 while (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_Q)
4021 {
4022 ptr += 2;
4023 if (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_E)
4024 { ptr += 2; continue; }
4025 inescq = TRUE;
4026 break;
4027 }
4028
4029 if (*ptr == 0 || (!inescq && *ptr == CHAR_RIGHT_SQUARE_BRACKET))
4030 {
4031 ptr = oldptr;
4032 goto LONE_SINGLE_CHARACTER;
4033 }
4034
4035 #ifdef SUPPORT_UTF8
4036 if (utf8)
4037 { /* Braces are required because the */
4038 GETCHARLEN(d, ptr, ptr); /* macro generates multiple statements */
4039 }
4040 else
4041 #endif
4042 d = *ptr; /* Not UTF-8 mode */
4043
4044 /* The second part of a range can be a single-character escape, but
4045 not any of the other escapes. Perl 5.6 treats a hyphen as a literal
4046 in such circumstances. */
4047
4048 if (!inescq && d == CHAR_BACKSLASH)
4049 {
4050 d = check_escape(&ptr, errorcodeptr, cd->bracount, options, TRUE);
4051 if (*errorcodeptr != 0) goto FAILED;
4052
4053 /* \b is backspace; any other special means the '-' was literal */
4054
4055 if (d < 0)
4056 {
4057 if (d == -ESC_b) d = CHAR_BS; else
4058 {
4059 ptr = oldptr;
4060 goto LONE_SINGLE_CHARACTER; /* A few lines below */
4061 }
4062 }
4063 }
4064
4065 /* Check that the two values are in the correct order. Optimize
4066 one-character ranges */
4067
4068 if (d < c)
4069 {
4070 *errorcodeptr = ERR8;
4071 goto FAILED;
4072 }
4073
4074 if (d == c) goto LONE_SINGLE_CHARACTER; /* A few lines below */
4075
4076 /* Remember \r or \n */
4077
4078 if (d == CHAR_CR || d == CHAR_NL) cd->external_flags |= PCRE_HASCRORLF;
4079
4080 /* In UTF-8 mode, if the upper limit is > 255, or > 127 for caseless
4081 matching, we have to use an XCLASS with extra data items. Caseless
4082 matching for characters > 127 is available only if UCP support is
4083 available. */
4084
4085 #ifdef SUPPORT_UTF8
4086 if (utf8 && (d > 255 || ((options & PCRE_CASELESS) != 0 && d > 127)))
4087 {
4088 class_utf8 = TRUE;
4089
4090 /* With UCP support, we can find the other case equivalents of
4091 the relevant characters. There may be several ranges. Optimize how
4092 they fit with the basic range. */
4093
4094 #ifdef SUPPORT_UCP
4095 if ((options & PCRE_CASELESS) != 0)
4096 {
4097 unsigned int occ, ocd;
4098 unsigned int cc = c;
4099 unsigned int origd = d;
4100 while (get_othercase_range(&cc, origd, &occ, &ocd))
4101 {
4102 if (occ >= (unsigned int)c &&
4103 ocd <= (unsigned int)d)
4104 continue; /* Skip embedded ranges */
4105
4106 if (occ < (unsigned int)c &&
4107 ocd >= (unsigned int)c - 1) /* Extend the basic range */
4108 { /* if there is overlap, */
4109 c = occ; /* noting that if occ < c */
4110 continue; /* we can't have ocd > d */
4111 } /* because a subrange is */
4112 if (ocd > (unsigned int)d &&
4113 occ <= (unsigned int)d + 1) /* always shorter than */
4114 { /* the basic range. */
4115 d = ocd;
4116 continue;
4117 }
4118
4119 if (occ == ocd)
4120 {
4121 *class_utf8data++ = XCL_SINGLE;
4122 }
4123 else
4124 {
4125 *class_utf8data++ = XCL_RANGE;
4126 class_utf8data += _pcre_ord2utf8(occ, class_utf8data);
4127 }
4128 class_utf8data += _pcre_ord2utf8(ocd, class_utf8data);
4129 }
4130 }
4131 #endif /* SUPPORT_UCP */
4132
4133 /* Now record the original range, possibly modified for UCP caseless
4134 overlapping ranges. */
4135
4136 *class_utf8data++ = XCL_RANGE;
4137 class_utf8data += _pcre_ord2utf8(c, class_utf8data);
4138 class_utf8data += _pcre_ord2utf8(d, class_utf8data);
4139
4140 /* With UCP support, we are done. Without UCP support, there is no
4141 caseless matching for UTF-8 characters > 127; we can use the bit map
4142 for the smaller ones. */
4143
4144 #ifdef SUPPORT_UCP
4145 continue; /* With next character in the class */
4146 #else
4147 if ((options & PCRE_CASELESS) == 0 || c > 127) continue;
4148
4149 /* Adjust upper limit and fall through to set up the map */
4150
4151 d = 127;
4152
4153 #endif /* SUPPORT_UCP */
4154 }
4155 #endif /* SUPPORT_UTF8 */
4156
4157 /* We use the bit map for all cases when not in UTF-8 mode; else
4158 ranges that lie entirely within 0-127 when there is UCP support; else
4159 for partial ranges without UCP support. */
4160
4161 class_charcount += d - c + 1;
4162 class_lastchar = d;
4163
4164 /* We can save a bit of time by skipping this in the pre-compile. */
4165
4166 if (lengthptr == NULL) for (; c <= d; c++)
4167 {
4168 classbits[c/8] |= (1 << (c&7));
4169 if ((options & PCRE_CASELESS) != 0)
4170 {
4171 int uc = cd->fcc[c]; /* flip case */
4172 classbits[uc/8] |= (1 << (uc&7));
4173 }
4174 }
4175
4176 continue; /* Go get the next char in the class */
4177 }
4178
4179 /* Handle a lone single character - we can get here for a normal
4180 non-escape char, or after \ that introduces a single character or for an
4181 apparent range that isn't. */
4182
4183 LONE_SINGLE_CHARACTER:
4184
4185 /* Handle a character that cannot go in the bit map */
4186
4187 #ifdef SUPPORT_UTF8
4188 if (utf8 && (c > 255 || ((options & PCRE_CASELESS) != 0 && c > 127)))
4189 {
4190 class_utf8 = TRUE;
4191 *class_utf8data++ = XCL_SINGLE;
4192 class_utf8data += _pcre_ord2utf8(c, class_utf8data);
4193
4194 #ifdef SUPPORT_UCP
4195 if ((options & PCRE_CASELESS) != 0)
4196 {
4197 unsigned int othercase;
4198 if ((othercase = UCD_OTHERCASE(c)) != c)
4199 {
4200 *class_utf8data++ = XCL_SINGLE;
4201 class_utf8data += _pcre_ord2utf8(othercase, class_utf8data);
4202 }
4203 }
4204 #endif /* SUPPORT_UCP */
4205
4206 }
4207 else
4208 #endif /* SUPPORT_UTF8 */
4209
4210 /* Handle a single-byte character */
4211 {
4212 classbits[c/8] |= (1 << (c&7));
4213 if ((options & PCRE_CASELESS) != 0)
4214 {
4215 c = cd->fcc[c]; /* flip case */
4216 classbits[c/8] |= (1 << (c&7));
4217 }
4218 class_charcount++;
4219 class_lastchar = c;
4220 }
4221 }
4222
4223 /* Loop until ']' reached. This "while" is the end of the "do" far above.
4224 If we are at the end of an internal nested string, revert to the outer
4225 string. */
4226
4227 while (((c = *(++ptr)) != 0 ||
4228 (nestptr != NULL &&
4229 (ptr = nestptr, nestptr = NULL, c = *(++ptr)) != 0)) &&
4230 (c != CHAR_RIGHT_SQUARE_BRACKET || inescq));
4231
4232 /* Check for missing terminating ']' */
4233
4234 if (c == 0)
4235 {
4236 *errorcodeptr = ERR6;
4237 goto FAILED;
4238 }
4239
4240 /* If class_charcount is 1, we saw precisely one character whose value is
4241 less than 256. As long as there were no characters >= 128 and there was no
4242 use of \p or \P, in other words, no use of any XCLASS features, we can
4243 optimize.
4244
4245 In UTF-8 mode, we can optimize the negative case only if there were no
4246 characters >= 128 because OP_NOT and the related opcodes like OP_NOTSTAR
4247 operate on single-bytes characters only. This is an historical hangover.
4248 Maybe one day we can tidy these opcodes to handle multi-byte characters.
4249
4250 The optimization throws away the bit map. We turn the item into a
4251 1-character OP_CHAR[I] if it's positive, or OP_NOT[I] if it's negative.
4252 Note that OP_NOT[I] does not support multibyte characters. In the positive
4253 case, it can cause firstbyte to be set. Otherwise, there can be no first
4254 char if this item is first, whatever repeat count may follow. In the case
4255 of reqbyte, save the previous value for reinstating. */
4256
4257 #ifdef SUPPORT_UTF8
4258 if (class_charcount == 1 && !class_utf8 &&
4259 (!utf8 || !negate_class || class_lastchar < 128))
4260 #else
4261 if (class_charcount == 1)
4262 #endif
4263 {
4264 zeroreqbyte = reqbyte;
4265
4266 /* The OP_NOT[I] opcodes work on one-byte characters only. */
4267
4268 if (negate_class)
4269 {
4270 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
4271 zerofirstbyte = firstbyte;
4272 *code++ = ((options & PCRE_CASELESS) != 0)? OP_NOTI: OP_NOT;
4273 *code++ = class_lastchar;
4274 break;
4275 }
4276
4277 /* For a single, positive character, get the value into mcbuffer, and
4278 then we can handle this with the normal one-character code. */
4279
4280 #ifdef SUPPORT_UTF8
4281 if (utf8 && class_lastchar > 127)
4282 mclength = _pcre_ord2utf8(class_lastchar, mcbuffer);
4283 else
4284 #endif
4285 {
4286 mcbuffer[0] = class_lastchar;
4287 mclength = 1;
4288 }
4289 goto ONE_CHAR;
4290 } /* End of 1-char optimization */
4291
4292 /* The general case - not the one-char optimization. If this is the first
4293 thing in the branch, there can be no first char setting, whatever the
4294 repeat count. Any reqbyte setting must remain unchanged after any kind of
4295 repeat. */
4296
4297 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
4298 zerofirstbyte = firstbyte;
4299 zeroreqbyte = reqbyte;
4300
4301 /* If there are characters with values > 255, we have to compile an
4302 extended class, with its own opcode, unless there was a negated special
4303 such as \S in the class, and PCRE_UCP is not set, because in that case all
4304 characters > 255 are in the class, so any that were explicitly given as
4305 well can be ignored. If (when there are explicit characters > 255 that must
4306 be listed) there are no characters < 256, we can omit the bitmap in the
4307 actual compiled code. */
4308
4309 #ifdef SUPPORT_UTF8
4310 if (class_utf8 && (!should_flip_negation || (options & PCRE_UCP) != 0))
4311 {
4312 *class_utf8data++ = XCL_END; /* Marks the end of extra data */
4313 *code++ = OP_XCLASS;
4314 code += LINK_SIZE;
4315 *code = negate_class? XCL_NOT : 0;
4316
4317 /* If the map is required, move up the extra data to make room for it;
4318 otherwise just move the code pointer to the end of the extra data. */
4319
4320 if (class_charcount > 0)
4321 {
4322 *code++ |= XCL_MAP;
4323 memmove(code + 32, code, class_utf8data - code);
4324 memcpy(code, classbits, 32);
4325 code = class_utf8data + 32;
4326 }
4327 else code = class_utf8data;
4328
4329 /* Now fill in the complete length of the item */
4330
4331 PUT(previous, 1, code - previous);
4332 break; /* End of class handling */
4333 }
4334 #endif
4335
4336 /* If there are no characters > 255, or they are all to be included or
4337 excluded, set the opcode to OP_CLASS or OP_NCLASS, depending on whether the
4338 whole class was negated and whether there were negative specials such as \S
4339 (non-UCP) in the class. Then copy the 32-byte map into the code vector,
4340 negating it if necessary. */
4341
4342 *code++ = (negate_class == should_flip_negation) ? OP_CLASS : OP_NCLASS;
4343 if (negate_class)
4344 {
4345 if (lengthptr == NULL) /* Save time in the pre-compile phase */
4346 for (c = 0; c < 32; c++) code[c] = ~classbits[c];
4347 }
4348 else
4349 {
4350 memcpy(code, classbits, 32);
4351 }
4352 code += 32;
4353 break;
4354
4355
4356 /* ===================================================================*/
4357 /* Various kinds of repeat; '{' is not necessarily a quantifier, but this
4358 has been tested above. */
4359
4360 case CHAR_LEFT_CURLY_BRACKET:
4361 if (!is_quantifier) goto NORMAL_CHAR;
4362 ptr = read_repeat_counts(ptr+1, &repeat_min, &repeat_max, errorcodeptr);
4363 if (*errorcodeptr != 0) goto FAILED;
4364 goto REPEAT;
4365
4366 case CHAR_ASTERISK:
4367 repeat_min = 0;
4368 repeat_max = -1;
4369 goto REPEAT;
4370
4371 case CHAR_PLUS:
4372 repeat_min = 1;
4373 repeat_max = -1;
4374 goto REPEAT;
4375
4376 case CHAR_QUESTION_MARK:
4377 repeat_min = 0;
4378 repeat_max = 1;
4379
4380 REPEAT:
4381 if (previous == NULL)
4382 {
4383 *errorcodeptr = ERR9;
4384 goto FAILED;
4385 }
4386
4387 if (repeat_min == 0)
4388 {
4389 firstbyte = zerofirstbyte; /* Adjust for zero repeat */
4390 reqbyte = zeroreqbyte; /* Ditto */
4391 }
4392
4393 /* Remember whether this is a variable length repeat */
4394
4395 reqvary = (repeat_min == repeat_max)? 0 : REQ_VARY;
4396
4397 op_type = 0; /* Default single-char op codes */
4398 possessive_quantifier = FALSE; /* Default not possessive quantifier */
4399
4400 /* Save start of previous item, in case we have to move it up in order to
4401 insert something before it. */
4402
4403 tempcode = previous;
4404
4405 /* If the next character is '+', we have a possessive quantifier. This
4406 implies greediness, whatever the setting of the PCRE_UNGREEDY option.
4407 If the next character is '?' this is a minimizing repeat, by default,
4408 but if PCRE_UNGREEDY is set, it works the other way round. We change the
4409 repeat type to the non-default. */
4410
4411 if (ptr[1] == CHAR_PLUS)
4412 {
4413 repeat_type = 0; /* Force greedy */
4414 possessive_quantifier = TRUE;
4415 ptr++;
4416 }
4417 else if (ptr[1] == CHAR_QUESTION_MARK)
4418 {
4419 repeat_type = greedy_non_default;
4420 ptr++;
4421 }
4422 else repeat_type = greedy_default;
4423
4424 /* If previous was a recursion call, wrap it in atomic brackets so that
4425 previous becomes the atomic group. All recursions were so wrapped in the
4426 past, but it no longer happens for non-repeated recursions. In fact, the
4427 repeated ones could be re-implemented independently so as not to need this,
4428 but for the moment we rely on the code for repeating groups. */
4429
4430 if (*previous == OP_RECURSE)
4431 {
4432 memmove(previous + 1 + LINK_SIZE, previous, 1 + LINK_SIZE);
4433 *previous = OP_ONCE;
4434 PUT(previous, 1, 2 + 2*LINK_SIZE);
4435 previous[2 + 2*LINK_SIZE] = OP_KET;
4436 PUT(previous, 3 + 2*LINK_SIZE, 2 + 2*LINK_SIZE);
4437 code += 2 + 2 * LINK_SIZE;
4438 length_prevgroup = 3 + 3*LINK_SIZE;
4439
4440 /* When actually compiling, we need to check whether this was a forward
4441 reference, and if so, adjust the offset. */
4442
4443 if (lengthptr == NULL && cd->hwm >= cd->start_workspace + LINK_SIZE)
4444 {
4445 int offset = GET(cd->hwm, -LINK_SIZE);
4446 if (offset == previous + 1 - cd->start_code)
4447 PUT(cd->hwm, -LINK_SIZE, offset + 1 + LINK_SIZE);
4448 }
4449 }
4450
4451 /* Now handle repetition for the different types of item. */
4452
4453 /* If previous was a character match, abolish the item and generate a
4454 repeat item instead. If a char item has a minumum of more than one, ensure
4455 that it is set in reqbyte - it might not be if a sequence such as x{3} is
4456 the first thing in a branch because the x will have gone into firstbyte
4457 instead. */
4458
4459 if (*previous == OP_CHAR || *previous == OP_CHARI)
4460 {
4461 op_type = (*previous == OP_CHAR)? 0 : OP_STARI - OP_STAR;
4462
4463 /* Deal with UTF-8 characters that take up more than one byte. It's
4464 easier to write this out separately than try to macrify it. Use c to
4465 hold the length of the character in bytes, plus 0x80 to flag that it's a
4466 length rather than a small character. */
4467
4468 #ifdef SUPPORT_UTF8
4469 if (utf8 && (code[-1] & 0x80) != 0)
4470 {
4471 uschar *lastchar = code - 1;
4472 while((*lastchar & 0xc0) == 0x80) lastchar--;
4473 c = code - lastchar; /* Length of UTF-8 character */
4474 memcpy(utf8_char, lastchar, c); /* Save the char */
4475 c |= 0x80; /* Flag c as a length */
4476 }
4477 else
4478 #endif
4479
4480 /* Handle the case of a single byte - either with no UTF8 support, or
4481 with UTF-8 disabled, or for a UTF-8 character < 128. */
4482
4483 {
4484 c = code[-1];
4485 if (repeat_min > 1) reqbyte = c | req_caseopt | cd->req_varyopt;
4486 }
4487
4488 /* If the repetition is unlimited, it pays to see if the next thing on
4489 the line is something that cannot possibly match this character. If so,
4490 automatically possessifying this item gains some performance in the case
4491 where the match fails. */
4492
4493 if (!possessive_quantifier &&
4494 repeat_max < 0 &&
4495 check_auto_possessive(previous, utf8, ptr + 1, options, cd))
4496 {
4497 repeat_type = 0; /* Force greedy */
4498 possessive_quantifier = TRUE;
4499 }
4500
4501 goto OUTPUT_SINGLE_REPEAT; /* Code shared with single character types */
4502 }
4503
4504 /* If previous was a single negated character ([^a] or similar), we use
4505 one of the special opcodes, replacing it. The code is shared with single-
4506 character repeats by setting opt_type to add a suitable offset into
4507 repeat_type. We can also test for auto-possessification. OP_NOT and OP_NOTI
4508 are currently used only for single-byte chars. */
4509
4510 else if (*previous == OP_NOT || *previous == OP_NOTI)
4511 {
4512 op_type = ((*previous == OP_NOT)? OP_NOTSTAR : OP_NOTSTARI) - OP_STAR;
4513 c = previous[1];
4514 if (!possessive_quantifier &&
4515 repeat_max < 0 &&
4516 check_auto_possessive(previous, utf8, ptr + 1, options, cd))
4517 {
4518 repeat_type = 0; /* Force greedy */
4519 possessive_quantifier = TRUE;
4520 }
4521 goto OUTPUT_SINGLE_REPEAT;
4522 }
4523
4524 /* If previous was a character type match (\d or similar), abolish it and
4525 create a suitable repeat item. The code is shared with single-character
4526 repeats by setting op_type to add a suitable offset into repeat_type. Note
4527 the the Unicode property types will be present only when SUPPORT_UCP is
4528 defined, but we don't wrap the little bits of code here because it just
4529 makes it horribly messy. */
4530
4531 else if (*previous < OP_EODN)
4532 {
4533 uschar *oldcode;
4534 int prop_type, prop_value;
4535 op_type = OP_TYPESTAR - OP_STAR; /* Use type opcodes */
4536 c = *previous;
4537
4538 if (!possessive_quantifier &&
4539 repeat_max < 0 &&
4540 check_auto_possessive(previous, utf8, ptr + 1, options, cd))
4541 {
4542 repeat_type = 0; /* Force greedy */
4543 possessive_quantifier = TRUE;
4544 }
4545
4546 OUTPUT_SINGLE_REPEAT:
4547 if (*previous == OP_PROP || *previous == OP_NOTPROP)
4548 {
4549 prop_type = previous[1];
4550 prop_value = previous[2];
4551 }
4552 else prop_type = prop_value = -1;
4553
4554 oldcode = code;
4555 code = previous; /* Usually overwrite previous item */
4556
4557 /* If the maximum is zero then the minimum must also be zero; Perl allows
4558 this case, so we do too - by simply omitting the item altogether. */
4559
4560 if (repeat_max == 0) goto END_REPEAT;
4561
4562 /*--------------------------------------------------------------------*/
4563 /* This code is obsolete from release 8.00; the restriction was finally
4564 removed: */
4565
4566 /* All real repeats make it impossible to handle partial matching (maybe
4567 one day we will be able to remove this restriction). */
4568
4569 /* if (repeat_max != 1) cd->external_flags |= PCRE_NOPARTIAL; */
4570 /*--------------------------------------------------------------------*/
4571
4572 /* Combine the op_type with the repeat_type */
4573
4574 repeat_type += op_type;
4575
4576 /* A minimum of zero is handled either as the special case * or ?, or as
4577 an UPTO, with the maximum given. */
4578
4579 if (repeat_min == 0)
4580 {
4581 if (repeat_max == -1) *code++ = OP_STAR + repeat_type;
4582 else if (repeat_max == 1) *code++ = OP_QUERY + repeat_type;
4583 else
4584 {
4585 *code++ = OP_UPTO + repeat_type;
4586 PUT2INC(code, 0, repeat_max);
4587 }
4588 }
4589
4590 /* A repeat minimum of 1 is optimized into some special cases. If the
4591 maximum is unlimited, we use OP_PLUS. Otherwise, the original item is
4592 left in place and, if the maximum is greater than 1, we use OP_UPTO with
4593 one less than the maximum. */
4594
4595 else if (repeat_min == 1)
4596 {
4597 if (repeat_max == -1)
4598 *code++ = OP_PLUS + repeat_type;
4599 else
4600 {
4601 code = oldcode; /* leave previous item in place */
4602 if (repeat_max == 1) goto END_REPEAT;
4603 *code++ = OP_UPTO + repeat_type;
4604 PUT2INC(code, 0, repeat_max - 1);
4605 }
4606 }
4607
4608 /* The case {n,n} is just an EXACT, while the general case {n,m} is
4609 handled as an EXACT followed by an UPTO. */
4610
4611 else
4612 {
4613 *code++ = OP_EXACT + op_type; /* NB EXACT doesn't have repeat_type */
4614 PUT2INC(code, 0, repeat_min);
4615
4616 /* If the maximum is unlimited, insert an OP_STAR. Before doing so,
4617 we have to insert the character for the previous code. For a repeated
4618 Unicode property match, there are two extra bytes that define the
4619 required property. In UTF-8 mode, long characters have their length in
4620 c, with the 0x80 bit as a flag. */
4621
4622 if (repeat_max < 0)
4623 {
4624 #ifdef SUPPORT_UTF8
4625 if (utf8 && c >= 128)
4626 {
4627 memcpy(code, utf8_char, c & 7);
4628 code += c & 7;
4629 }
4630 else
4631 #endif
4632 {
4633 *code++ = c;
4634 if (prop_type >= 0)
4635 {
4636 *code++ = prop_type;
4637 *code++ = prop_value;
4638 }
4639 }
4640 *code++ = OP_STAR + repeat_type;
4641 }
4642
4643 /* Else insert an UPTO if the max is greater than the min, again
4644 preceded by the character, for the previously inserted code. If the
4645 UPTO is just for 1 instance, we can use QUERY instead. */
4646
4647 else if (repeat_max != repeat_min)
4648 {
4649 #ifdef SUPPORT_UTF8
4650 if (utf8 && c >= 128)
4651 {
4652 memcpy(code, utf8_char, c & 7);
4653 code += c & 7;
4654 }
4655 else
4656 #endif
4657 *code++ = c;
4658 if (prop_type >= 0)
4659 {
4660 *code++ = prop_type;
4661 *code++ = prop_value;
4662 }
4663 repeat_max -= repeat_min;
4664
4665 if (repeat_max == 1)
4666 {
4667 *code++ = OP_QUERY + repeat_type;
4668 }
4669 else
4670 {
4671 *code++ = OP_UPTO + repeat_type;
4672 PUT2INC(code, 0, repeat_max);
4673 }
4674 }
4675 }
4676
4677 /* The character or character type itself comes last in all cases. */
4678
4679 #ifdef SUPPORT_UTF8
4680 if (utf8 && c >= 128)
4681 {
4682 memcpy(code, utf8_char, c & 7);
4683 code += c & 7;
4684 }
4685 else
4686 #endif
4687 *code++ = c;
4688
4689 /* For a repeated Unicode property match, there are two extra bytes that
4690 define the required property. */
4691
4692 #ifdef SUPPORT_UCP
4693 if (prop_type >= 0)
4694 {
4695 *code++ = prop_type;
4696 *code++ = prop_value;
4697 }
4698 #endif
4699 }
4700
4701 /* If previous was a character class or a back reference, we put the repeat
4702 stuff after it, but just skip the item if the repeat was {0,0}. */
4703
4704 else if (*previous == OP_CLASS ||
4705 *previous == OP_NCLASS ||
4706 #ifdef SUPPORT_UTF8
4707 *previous == OP_XCLASS ||
4708 #endif
4709 *previous == OP_REF ||
4710 *previous == OP_REFI)
4711 {
4712 if (repeat_max == 0)
4713 {
4714 code = previous;
4715 goto END_REPEAT;
4716 }
4717
4718 /*--------------------------------------------------------------------*/
4719 /* This code is obsolete from release 8.00; the restriction was finally
4720 removed: */
4721
4722 /* All real repeats make it impossible to handle partial matching (maybe
4723 one day we will be able to remove this restriction). */
4724
4725 /* if (repeat_max != 1) cd->external_flags |= PCRE_NOPARTIAL; */
4726 /*--------------------------------------------------------------------*/
4727
4728 if (repeat_min == 0 && repeat_max == -1)
4729 *code++ = OP_CRSTAR + repeat_type;
4730 else if (repeat_min == 1 && repeat_max == -1)
4731 *code++ = OP_CRPLUS + repeat_type;
4732 else if (repeat_min == 0 && repeat_max == 1)
4733 *code++ = OP_CRQUERY + repeat_type;
4734 else
4735 {
4736 *code++ = OP_CRRANGE + repeat_type;
4737 PUT2INC(code, 0, repeat_min);
4738 if (repeat_max == -1) repeat_max = 0; /* 2-byte encoding for max */
4739 PUT2INC(code, 0, repeat_max);
4740 }
4741 }
4742
4743 /* If previous was a bracket group, we may have to replicate it in certain
4744 cases. Note that at this point we can encounter only the "basic" bracket
4745 opcodes such as BRA and CBRA, as this is the place where they get converted
4746 into the more special varieties such as BRAPOS and SBRA. A test for >=
4747 OP_ASSERT and <= OP_COND includes ASSERT, ASSERT_NOT, ASSERTBACK,
4748 ASSERTBACK_NOT, ONCE, BRA, CBRA, and COND. Originally, PCRE did not allow
4749 repetition of assertions, but now it does, for Perl compatibility. */
4750
4751 else if (*previous >= OP_ASSERT && *previous <= OP_COND)
4752 {
4753 register int i;
4754 int len = (int)(code - previous);
4755 uschar *bralink = NULL;
4756 uschar *brazeroptr = NULL;
4757
4758 /* Repeating a DEFINE group is pointless, but Perl allows the syntax, so
4759 we just ignore the repeat. */
4760
4761 if (*previous == OP_COND && previous[LINK_SIZE+1] == OP_DEF)
4762 goto END_REPEAT;
4763
4764 /* There is no sense in actually repeating assertions. The only potential
4765 use of repetition is in cases when the assertion is optional. Therefore,
4766 if the minimum is greater than zero, just ignore the repeat. If the
4767 maximum is not not zero or one, set it to 1. */
4768
4769 if (*previous < OP_ONCE) /* Assertion */
4770 {
4771 if (repeat_min > 0) goto END_REPEAT;
4772 if (repeat_max < 0 || repeat_max > 1) repeat_max = 1;
4773 }
4774
4775 /* The case of a zero minimum is special because of the need to stick
4776 OP_BRAZERO in front of it, and because the group appears once in the
4777 data, whereas in other cases it appears the minimum number of times. For
4778 this reason, it is simplest to treat this case separately, as otherwise
4779 the code gets far too messy. There are several special subcases when the
4780 minimum is zero. */
4781
4782 if (repeat_min == 0)
4783 {
4784 /* If the maximum is also zero, we used to just omit the group from the
4785 output altogether, like this:
4786
4787 ** if (repeat_max == 0)
4788 ** {
4789 ** code = previous;
4790 ** goto END_REPEAT;
4791 ** }
4792
4793 However, that fails when a group or a subgroup within it is referenced
4794 as a subroutine from elsewhere in the pattern, so now we stick in
4795 OP_SKIPZERO in front of it so that it is skipped on execution. As we
4796 don't have a list of which groups are referenced, we cannot do this
4797 selectively.
4798
4799 If the maximum is 1 or unlimited, we just have to stick in the BRAZERO
4800 and do no more at this point. However, we do need to adjust any
4801 OP_RECURSE calls inside the group that refer to the group itself or any
4802 internal or forward referenced group, because the offset is from the
4803 start of the whole regex. Temporarily terminate the pattern while doing
4804 this. */
4805
4806 if (repeat_max <= 1) /* Covers 0, 1, and unlimited */
4807 {
4808 *code = OP_END;
4809 adjust_recurse(previous, 1, utf8, cd, save_hwm);
4810 memmove(previous+1, previous, len);
4811 code++;
4812 if (repeat_max == 0)
4813 {
4814 *previous++ = OP_SKIPZERO;
4815 goto END_REPEAT;
4816 }
4817 brazeroptr = previous; /* Save for possessive optimizing */
4818 *previous++ = OP_BRAZERO + repeat_type;
4819 }
4820
4821 /* If the maximum is greater than 1 and limited, we have to replicate
4822 in a nested fashion, sticking OP_BRAZERO before each set of brackets.
4823 The first one has to be handled carefully because it's the original
4824 copy, which has to be moved up. The remainder can be handled by code
4825 that is common with the non-zero minimum case below. We have to
4826 adjust the value or repeat_max, since one less copy is required. Once
4827 again, we may have to adjust any OP_RECURSE calls inside the group. */
4828
4829 else
4830 {
4831 int offset;
4832 *code = OP_END;
4833 adjust_recurse(previous, 2 + LINK_SIZE, utf8, cd, save_hwm);
4834 memmove(previous + 2 + LINK_SIZE, previous, len);
4835 code += 2 + LINK_SIZE;
4836 *previous++ = OP_BRAZERO + repeat_type;
4837 *previous++ = OP_BRA;
4838
4839 /* We chain together the bracket offset fields that have to be
4840 filled in later when the ends of the brackets are reached. */
4841
4842 offset = (bralink == NULL)? 0 : (int)(previous - bralink);
4843 bralink = previous;
4844 PUTINC(previous, 0, offset);
4845 }
4846
4847 repeat_max--;
4848 }
4849
4850 /* If the minimum is greater than zero, replicate the group as many
4851 times as necessary, and adjust the maximum to the number of subsequent
4852 copies that we need. If we set a first char from the group, and didn't
4853 set a required char, copy the latter from the former. If there are any
4854 forward reference subroutine calls in the group, there will be entries on
4855 the workspace list; replicate these with an appropriate increment. */
4856
4857 else
4858 {
4859 if (repeat_min > 1)
4860 {
4861 /* In the pre-compile phase, we don't actually do the replication. We
4862 just adjust the length as if we had. Do some paranoid checks for
4863 potential integer overflow. The INT64_OR_DOUBLE type is a 64-bit
4864 integer type when available, otherwise double. */
4865
4866 if (lengthptr != NULL)
4867 {
4868 int delta = (repeat_min - 1)*length_prevgroup;
4869 if ((INT64_OR_DOUBLE)(repeat_min - 1)*
4870 (INT64_OR_DOUBLE)length_prevgroup >
4871 (INT64_OR_DOUBLE)INT_MAX ||
4872 OFLOW_MAX - *lengthptr < delta)
4873 {
4874 *errorcodeptr = ERR20;
4875 goto FAILED;
4876 }
4877 *lengthptr += delta;
4878 }
4879
4880 /* This is compiling for real */
4881
4882 else
4883 {
4884 if (groupsetfirstbyte && reqbyte < 0) reqbyte = firstbyte;
4885 for (i = 1; i < repeat_min; i++)
4886 {
4887 uschar *hc;
4888 uschar *this_hwm = cd->hwm;
4889 memcpy(code, previous, len);
4890 for (hc = save_hwm; hc < this_hwm; hc += LINK_SIZE)
4891 {
4892 PUT(cd->hwm, 0, GET(hc, 0) + len);
4893 cd->hwm += LINK_SIZE;
4894 }
4895 save_hwm = this_hwm;
4896 code += len;
4897 }
4898 }
4899 }
4900
4901 if (repeat_max > 0) repeat_max -= repeat_min;
4902 }
4903
4904 /* This code is common to both the zero and non-zero minimum cases. If
4905 the maximum is limited, it replicates the group in a nested fashion,
4906 remembering the bracket starts on a stack. In the case of a zero minimum,
4907 the first one was set up above. In all cases the repeat_max now specifies
4908 the number of additional copies needed. Again, we must remember to
4909 replicate entries on the forward reference list. */
4910
4911 if (repeat_max >= 0)
4912 {
4913 /* In the pre-compile phase, we don't actually do the replication. We
4914 just adjust the length as if we had. For each repetition we must add 1
4915 to the length for BRAZERO and for all but the last repetition we must
4916 add 2 + 2*LINKSIZE to allow for the nesting that occurs. Do some
4917 paranoid checks to avoid integer overflow. The INT64_OR_DOUBLE type is
4918 a 64-bit integer type when available, otherwise double. */
4919
4920 if (lengthptr != NULL && repeat_max > 0)
4921 {
4922 int delta = repeat_max * (length_prevgroup + 1 + 2 + 2*LINK_SIZE) -
4923 2 - 2*LINK_SIZE; /* Last one doesn't nest */
4924 if ((INT64_OR_DOUBLE)repeat_max *
4925 (INT64_OR_DOUBLE)(length_prevgroup + 1 + 2 + 2*LINK_SIZE)
4926 > (INT64_OR_DOUBLE)INT_MAX ||
4927 OFLOW_MAX - *lengthptr < delta)
4928 {
4929 *errorcodeptr = ERR20;
4930 goto FAILED;
4931 }
4932 *lengthptr += delta;
4933 }
4934
4935 /* This is compiling for real */
4936
4937 else for (i = repeat_max - 1; i >= 0; i--)
4938 {
4939 uschar *hc;
4940 uschar *this_hwm = cd->hwm;
4941
4942 *code++ = OP_BRAZERO + repeat_type;
4943
4944 /* All but the final copy start a new nesting, maintaining the
4945 chain of brackets outstanding. */
4946
4947 if (i != 0)
4948 {
4949 int offset;
4950 *code++ = OP_BRA;
4951 offset = (bralink == NULL)? 0 : (int)(code - bralink);
4952 bralink = code;
4953 PUTINC(code, 0, offset);
4954 }
4955
4956 memcpy(code, previous, len);
4957 for (hc = save_hwm; hc < this_hwm; hc += LINK_SIZE)
4958 {
4959 PUT(cd->hwm, 0, GET(hc, 0) + len + ((i != 0)? 2+LINK_SIZE : 1));
4960 cd->hwm += LINK_SIZE;
4961 }
4962 save_hwm = this_hwm;
4963 code += len;
4964 }
4965
4966 /* Now chain through the pending brackets, and fill in their length
4967 fields (which are holding the chain links pro tem). */
4968
4969 while (bralink != NULL)
4970 {
4971 int oldlinkoffset;
4972 int offset = (int)(code - bralink + 1);
4973 uschar *bra = code - offset;
4974 oldlinkoffset = GET(bra, 1);
4975 bralink = (oldlinkoffset == 0)? NULL : bralink - oldlinkoffset;
4976 *code++ = OP_KET;
4977 PUTINC(code, 0, offset);
4978 PUT(bra, 1, offset);
4979 }
4980 }
4981
4982 /* If the maximum is unlimited, set a repeater in the final copy. For
4983 ONCE brackets, that's all we need to do. However, possessively repeated
4984 ONCE brackets can be converted into non-capturing brackets, as the
4985 behaviour of (?:xx)++ is the same as (?>xx)++ and this saves having to
4986 deal with possessive ONCEs specially.
4987
4988 Otherwise, when we are doing the actual compile phase, check to see
4989 whether this group is one that could match an empty string. If so,
4990 convert the initial operator to the S form (e.g. OP_BRA -> OP_SBRA) so
4991 that runtime checking can be done. [This check is also applied to ONCE
4992 groups at runtime, but in a different way.]
4993
4994 Then, if the quantifier was possessive and the bracket is not a
4995 conditional, we convert the BRA code to the POS form, and the KET code to
4996 KETRPOS. (It turns out to be convenient at runtime to detect this kind of
4997 subpattern at both the start and at the end.) The use of special opcodes
4998 makes it possible to reduce greatly the stack usage in pcre_exec(). If
4999 the group is preceded by OP_BRAZERO, convert this to OP_BRAPOSZERO.
5000
5001 Then, if the minimum number of matches is 1 or 0, cancel the possessive
5002 flag so that the default action below, of wrapping everything inside
5003 atomic brackets, does not happen. When the minimum is greater than 1,
5004 there will be earlier copies of the group, and so we still have to wrap
5005 the whole thing. */
5006
5007 else
5008 {
5009 uschar *ketcode = code - 1 - LINK_SIZE;
5010 uschar *bracode = ketcode - GET(ketcode, 1);
5011
5012 /* Convert possessive ONCE brackets to non-capturing */
5013
5014 if ((*bracode == OP_ONCE || *bracode == OP_ONCE_NC) &&
5015 possessive_quantifier) *bracode = OP_BRA;
5016
5017 /* For non-possessive ONCE brackets, all we need to do is to
5018 set the KET. */
5019
5020 if (*bracode == OP_ONCE || *bracode == OP_ONCE_NC)
5021 *ketcode = OP_KETRMAX + repeat_type;
5022
5023 /* Handle non-ONCE brackets and possessive ONCEs (which have been
5024 converted to non-capturing above). */
5025
5026 else
5027 {
5028 /* In the compile phase, check for empty string matching. */
5029
5030 if (lengthptr == NULL)
5031 {
5032 uschar *scode = bracode;
5033 do
5034 {
5035 if (could_be_empty_branch(scode, ketcode, utf8, cd))
5036 {
5037 *bracode += OP_SBRA - OP_BRA;
5038 break;
5039 }
5040 scode += GET(scode, 1);
5041 }
5042 while (*scode == OP_ALT);
5043 }
5044
5045 /* Handle possessive quantifiers. */
5046
5047 if (possessive_quantifier)
5048 {
5049 /* For COND brackets, we wrap the whole thing in a possessively
5050 repeated non-capturing bracket, because we have not invented POS
5051 versions of the COND opcodes. Because we are moving code along, we
5052 must ensure that any pending recursive references are updated. */
5053
5054 if (*bracode == OP_COND || *bracode == OP_SCOND)
5055 {
5056 int nlen = (int)(code - bracode);
5057 *code = OP_END;
5058 adjust_recurse(bracode, 1 + LINK_SIZE, utf8, cd, save_hwm);
5059 memmove(bracode + 1+LINK_SIZE, bracode, nlen);
5060 code += 1 + LINK_SIZE;
5061 nlen += 1 + LINK_SIZE;
5062 *bracode = OP_BRAPOS;
5063 *code++ = OP_KETRPOS;
5064 PUTINC(code, 0, nlen);
5065 PUT(bracode, 1, nlen);
5066 }
5067
5068 /* For non-COND brackets, we modify the BRA code and use KETRPOS. */
5069
5070 else
5071 {
5072 *bracode += 1; /* Switch to xxxPOS opcodes */
5073 *ketcode = OP_KETRPOS;
5074 }
5075
5076 /* If the minimum is zero, mark it as possessive, then unset the
5077 possessive flag when the minimum is 0 or 1. */
5078
5079 if (brazeroptr != NULL) *brazeroptr = OP_BRAPOSZERO;
5080 if (repeat_min < 2) possessive_quantifier = FALSE;
5081 }
5082
5083 /* Non-possessive quantifier */
5084
5085 else *ketcode = OP_KETRMAX + repeat_type;
5086 }
5087 }
5088 }
5089
5090 /* If previous is OP_FAIL, it was generated by an empty class [] in
5091 JavaScript mode. The other ways in which OP_FAIL can be generated, that is
5092 by (*FAIL) or (?!) set previous to NULL, which gives a "nothing to repeat"
5093 error above. We can just ignore the repeat in JS case. */
5094
5095 else if (*previous == OP_FAIL) goto END_REPEAT;
5096
5097 /* Else there's some kind of shambles */
5098
5099 else
5100 {
5101 *errorcodeptr = ERR11;
5102 goto FAILED;
5103 }
5104
5105 /* If the character following a repeat is '+', or if certain optimization
5106 tests above succeeded, possessive_quantifier is TRUE. For some opcodes,
5107 there are special alternative opcodes for this case. For anything else, we
5108 wrap the entire repeated item inside OP_ONCE brackets. Logically, the '+'
5109 notation is just syntactic sugar, taken from Sun's Java package, but the
5110 special opcodes can optimize it.
5111
5112 Some (but not all) possessively repeated subpatterns have already been
5113 completely handled in the code just above. For them, possessive_quantifier
5114 is always FALSE at this stage.
5115
5116 Note that the repeated item starts at tempcode, not at previous, which
5117 might be the first part of a string whose (former) last char we repeated.
5118
5119 Possessifying an 'exact' quantifier has no effect, so we can ignore it. But
5120 an 'upto' may follow. We skip over an 'exact' item, and then test the
5121 length of what remains before proceeding. */
5122
5123 if (possessive_quantifier)
5124 {
5125 int len;
5126
5127 if (*tempcode == OP_TYPEEXACT)
5128 tempcode += _pcre_OP_lengths[*tempcode] +
5129 ((tempcode[3] == OP_PROP || tempcode[3] == OP_NOTPROP)? 2 : 0);
5130
5131 else if (*tempcode == OP_EXACT || *tempcode == OP_NOTEXACT)
5132 {
5133 tempcode += _pcre_OP_lengths[*tempcode];
5134 #ifdef SUPPORT_UTF8
5135 if (utf8 && tempcode[-1] >= 0xc0)
5136 tempcode += _pcre_utf8_table4[tempcode[-1] & 0x3f];
5137 #endif
5138 }
5139
5140 len = (int)(code - tempcode);
5141 if (len > 0) switch (*tempcode)
5142 {
5143 case OP_STAR: *tempcode = OP_POSSTAR; break;
5144 case OP_PLUS: *tempcode = OP_POSPLUS; break;
5145 case OP_QUERY: *tempcode = OP_POSQUERY; break;
5146 case OP_UPTO: *tempcode = OP_POSUPTO; break;
5147
5148 case OP_STARI: *tempcode = OP_POSSTARI; break;
5149 case OP_PLUSI: *tempcode = OP_POSPLUSI; break;
5150 case OP_QUERYI: *tempcode = OP_POSQUERYI; break;
5151 case OP_UPTOI: *tempcode = OP_POSUPTOI; break;
5152
5153 case OP_NOTSTAR: *tempcode = OP_NOTPOSSTAR; break;
5154 case OP_NOTPLUS: *tempcode = OP_NOTPOSPLUS; break;
5155 case OP_NOTQUERY: *tempcode = OP_NOTPOSQUERY; break;
5156 case OP_NOTUPTO: *tempcode = OP_NOTPOSUPTO; break;
5157
5158 case OP_NOTSTARI: *tempcode = OP_NOTPOSSTARI; break;
5159 case OP_NOTPLUSI: *tempcode = OP_NOTPOSPLUSI; break;
5160 case OP_NOTQUERYI: *tempcode = OP_NOTPOSQUERYI; break;
5161 case OP_NOTUPTOI: *tempcode = OP_NOTPOSUPTOI; break;
5162
5163 case OP_TYPESTAR: *tempcode = OP_TYPEPOSSTAR; break;
5164 case OP_TYPEPLUS: *tempcode = OP_TYPEPOSPLUS; break;
5165 case OP_TYPEQUERY: *tempcode = OP_TYPEPOSQUERY; break;
5166 case OP_TYPEUPTO: *tempcode = OP_TYPEPOSUPTO; break;
5167
5168 /* Because we are moving code along, we must ensure that any
5169 pending recursive references are updated. */
5170
5171 default:
5172 *code = OP_END;
5173 adjust_recurse(tempcode, 1 + LINK_SIZE, utf8, cd, save_hwm);
5174 memmove(tempcode + 1+LINK_SIZE, tempcode, len);
5175 code += 1 + LINK_SIZE;
5176 len += 1 + LINK_SIZE;
5177 tempcode[0] = OP_ONCE;
5178 *code++ = OP_KET;
5179 PUTINC(code, 0, len);
5180 PUT(tempcode, 1, len);
5181 break;
5182 }
5183 }
5184
5185 /* In all case we no longer have a previous item. We also set the
5186 "follows varying string" flag for subsequently encountered reqbytes if
5187 it isn't already set and we have just passed a varying length item. */
5188
5189 END_REPEAT:
5190 previous = NULL;
5191 cd->req_varyopt |= reqvary;
5192 break;
5193
5194
5195 /* ===================================================================*/
5196 /* Start of nested parenthesized sub-expression, or comment or lookahead or
5197 lookbehind or option setting or condition or all the other extended
5198 parenthesis forms. */
5199
5200 case CHAR_LEFT_PARENTHESIS:
5201 newoptions = options;
5202 skipbytes = 0;
5203 bravalue = OP_CBRA;
5204 save_hwm = cd->hwm;
5205 reset_bracount = FALSE;
5206
5207 /* First deal with various "verbs" that can be introduced by '*'. */
5208
5209 if (*(++ptr) == CHAR_ASTERISK &&
5210 ((cd->ctypes[ptr[1]] & ctype_letter) != 0 || ptr[1] == ':'))
5211 {
5212 int i, namelen;
5213 int arglen = 0;
5214 const char *vn = verbnames;
5215 const uschar *name = ptr + 1;
5216 const uschar *arg = NULL;
5217 previous = NULL;
5218 while ((cd->ctypes[*++ptr] & ctype_letter) != 0) {};
5219 namelen = (int)(ptr - name);
5220
5221 /* It appears that Perl allows any characters whatsoever, other than
5222 a closing parenthesis, to appear in arguments, so we no longer insist on
5223 letters, digits, and underscores. */
5224
5225 if (*ptr == CHAR_COLON)
5226 {
5227 arg = ++ptr;
5228 while (*ptr != 0 && *ptr != CHAR_RIGHT_PARENTHESIS) ptr++;
5229 arglen = (int)(ptr - arg);
5230 }
5231
5232 if (*ptr != CHAR_RIGHT_PARENTHESIS)
5233 {
5234 *errorcodeptr = ERR60;
5235 goto FAILED;
5236 }
5237
5238 /* Scan the table of verb names */
5239
5240 for (i = 0; i < verbcount; i++)
5241 {
5242 if (namelen == verbs[i].len &&
5243 strncmp((char *)name, vn, namelen) == 0)
5244 {
5245 /* Check for open captures before ACCEPT and convert it to
5246 ASSERT_ACCEPT if in an assertion. */
5247
5248 if (verbs[i].op == OP_ACCEPT)
5249 {
5250 open_capitem *oc;
5251 if (arglen != 0)
5252 {
5253 *errorcodeptr = ERR59;
5254 goto FAILED;
5255 }
5256 cd->had_accept = TRUE;
5257 for (oc = cd->open_caps; oc != NULL; oc = oc->next)
5258 {
5259 *code++ = OP_CLOSE;
5260 PUT2INC(code, 0, oc->number);
5261 }
5262 *code++ = (cd->assert_depth > 0)? OP_ASSERT_ACCEPT : OP_ACCEPT;
5263
5264 /* Do not set firstbyte after *ACCEPT */
5265 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
5266 }
5267
5268 /* Handle other cases with/without an argument */
5269
5270 else if (arglen == 0)
5271 {
5272 if (verbs[i].op < 0) /* Argument is mandatory */
5273 {
5274 *errorcodeptr = ERR66;
5275 goto FAILED;
5276 }
5277 *code = verbs[i].op;
5278 if (*code++ == OP_THEN) cd->external_flags |= PCRE_HASTHEN;
5279 }
5280
5281 else
5282 {
5283 if (verbs[i].op_arg < 0) /* Argument is forbidden */
5284 {
5285 *errorcodeptr = ERR59;
5286 goto FAILED;
5287 }
5288 *code = verbs[i].op_arg;
5289 if (*code++ == OP_THEN_ARG) cd->external_flags |= PCRE_HASTHEN;
5290 *code++ = arglen;
5291 memcpy(code, arg, arglen);
5292 code += arglen;
5293 *code++ = 0;
5294 }
5295
5296 break; /* Found verb, exit loop */
5297 }
5298
5299 vn += verbs[i].len + 1;
5300 }
5301
5302 if (i < verbcount) continue; /* Successfully handled a verb */
5303 *errorcodeptr = ERR60; /* Verb not recognized */
5304 goto FAILED;
5305 }
5306
5307 /* Deal with the extended parentheses; all are introduced by '?', and the
5308 appearance of any of them means that this is not a capturing group. */
5309
5310 else if (*ptr == CHAR_QUESTION_MARK)
5311 {
5312 int i, set, unset, namelen;
5313 int *optset;
5314 const uschar *name;
5315 uschar *slot;
5316
5317 switch (*(++ptr))
5318 {
5319 case CHAR_NUMBER_SIGN: /* Comment; skip to ket */
5320 ptr++;
5321 while (*ptr != 0 && *ptr != CHAR_RIGHT_PARENTHESIS) ptr++;
5322 if (*ptr == 0)
5323 {
5324 *errorcodeptr = ERR18;
5325 goto FAILED;
5326 }
5327 continue;
5328
5329
5330 /* ------------------------------------------------------------ */
5331 case CHAR_VERTICAL_LINE: /* Reset capture count for each branch */
5332 reset_bracount = TRUE;
5333 /* Fall through */
5334
5335 /* ------------------------------------------------------------ */
5336 case CHAR_COLON: /* Non-capturing bracket */
5337 bravalue = OP_BRA;
5338 ptr++;
5339 break;
5340
5341
5342 /* ------------------------------------------------------------ */
5343 case CHAR_LEFT_PARENTHESIS:
5344 bravalue = OP_COND; /* Conditional group */
5345
5346 /* A condition can be an assertion, a number (referring to a numbered
5347 group), a name (referring to a named group), or 'R', referring to
5348 recursion. R<digits> and R&name are also permitted for recursion tests.
5349
5350 There are several syntaxes for testing a named group: (?(name)) is used
5351 by Python; Perl 5.10 onwards uses (?(<name>) or (?('name')).
5352
5353 There are two unfortunate ambiguities, caused by history. (a) 'R' can
5354 be the recursive thing or the name 'R' (and similarly for 'R' followed
5355 by digits), and (b) a number could be a name that consists of digits.
5356 In both cases, we look for a name first; if not found, we try the other
5357 cases. */
5358
5359 /* For conditions that are assertions, check the syntax, and then exit
5360 the switch. This will take control down to where bracketed groups,
5361 including assertions, are processed. */
5362
5363 if (ptr[1] == CHAR_QUESTION_MARK && (ptr[2] == CHAR_EQUALS_SIGN ||
5364 ptr[2] == CHAR_EXCLAMATION_MARK || ptr[2] == CHAR_LESS_THAN_SIGN))
5365 break;
5366
5367 /* Most other conditions use OP_CREF (a couple change to OP_RREF
5368 below), and all need to skip 3 bytes at the start of the group. */
5369
5370 code[1+LINK_SIZE] = OP_CREF;
5371 skipbytes = 3;
5372 refsign = -1;
5373
5374 /* Check for a test for recursion in a named group. */
5375
5376 if (ptr[1] == CHAR_R && ptr[2] == CHAR_AMPERSAND)
5377 {
5378 terminator = -1;
5379 ptr += 2;
5380 code[1+LINK_SIZE] = OP_RREF; /* Change the type of test */
5381 }
5382
5383 /* Check for a test for a named group's having been set, using the Perl
5384 syntax (?(<name>) or (?('name') */
5385
5386 else if (ptr[1] == CHAR_LESS_THAN_SIGN)
5387 {
5388 terminator = CHAR_GREATER_THAN_SIGN;
5389 ptr++;
5390 }
5391 else if (ptr[1] == CHAR_APOSTROPHE)
5392 {
5393 terminator = CHAR_APOSTROPHE;
5394 ptr++;
5395 }
5396 else
5397 {
5398 terminator = 0;
5399 if (ptr[1] == CHAR_MINUS || ptr[1] == CHAR_PLUS) refsign = *(++ptr);
5400 }
5401
5402 /* We now expect to read a name; any thing else is an error */
5403
5404 if ((cd->ctypes[ptr[1]] & ctype_word) == 0)
5405 {
5406 ptr += 1; /* To get the right offset */
5407 *errorcodeptr = ERR28;
5408 goto FAILED;
5409 }
5410
5411 /* Read the name, but also get it as a number if it's all digits */
5412
5413 recno = 0;
5414 name = ++ptr;
5415 while ((cd->ctypes[*ptr] & ctype_word) != 0)
5416 {
5417 if (recno >= 0)
5418 recno = ((digitab[*ptr] & ctype_digit) != 0)?
5419 recno * 10 + *ptr - CHAR_0 : -1;
5420 ptr++;
5421 }
5422 namelen = (int)(ptr - name);
5423
5424 if ((terminator > 0 && *ptr++ != terminator) ||
5425 *ptr++ != CHAR_RIGHT_PARENTHESIS)
5426 {
5427 ptr--; /* Error offset */
5428 *errorcodeptr = ERR26;
5429 goto FAILED;
5430 }
5431
5432 /* Do no further checking in the pre-compile phase. */
5433
5434 if (lengthptr != NULL) break;
5435
5436 /* In the real compile we do the work of looking for the actual
5437 reference. If the string started with "+" or "-" we require the rest to
5438 be digits, in which case recno will be set. */
5439
5440 if (refsign > 0)
5441 {
5442 if (recno <= 0)
5443 {
5444 *errorcodeptr = ERR58;
5445 goto FAILED;
5446 }
5447 recno = (refsign == CHAR_MINUS)?
5448 cd->bracount - recno + 1 : recno +cd->bracount;
5449 if (recno <= 0 || recno > cd->final_bracount)
5450 {
5451 *errorcodeptr = ERR15;
5452 goto FAILED;
5453 }
5454 PUT2(code, 2+LINK_SIZE, recno);
5455 break;
5456 }
5457
5458 /* Otherwise (did not start with "+" or "-"), start by looking for the
5459 name. If we find a name, add one to the opcode to change OP_CREF or
5460 OP_RREF into OP_NCREF or OP_NRREF. These behave exactly the same,
5461 except they record that the reference was originally to a name. The
5462 information is used to check duplicate names. */
5463
5464 slot = cd->name_table;
5465 for (i = 0; i < cd->names_found; i++)
5466 {
5467 if (strncmp((char *)name, (char *)slot+2, namelen) == 0) break;
5468 slot += cd->name_entry_size;
5469 }
5470
5471 /* Found a previous named subpattern */
5472
5473 if (i < cd->names_found)
5474 {
5475 recno = GET2(slot, 0);
5476 PUT2(code, 2+LINK_SIZE, recno);
5477 code[1+LINK_SIZE]++;
5478 }
5479
5480 /* Search the pattern for a forward reference */
5481
5482 else if ((i = find_parens(cd, name, namelen,
5483 (options & PCRE_EXTENDED) != 0, utf8)) > 0)
5484 {
5485 PUT2(code, 2+LINK_SIZE, i);
5486 code[1+LINK_SIZE]++;
5487 }
5488
5489 /* If terminator == 0 it means that the name followed directly after
5490 the opening parenthesis [e.g. (?(abc)...] and in this case there are
5491 some further alternatives to try. For the cases where terminator != 0
5492 [things like (?(<name>... or (?('name')... or (?(R&name)... ] we have
5493 now checked all the possibilities, so give an error. */
5494
5495 else if (terminator != 0)
5496 {
5497 *errorcodeptr = ERR15;
5498 goto FAILED;
5499 }
5500
5501 /* Check for (?(R) for recursion. Allow digits after R to specify a
5502 specific group number. */
5503
5504 else if (*name == CHAR_R)
5505 {
5506 recno = 0;
5507 for (i = 1; i < namelen; i++)
5508 {
5509 if ((digitab[name[i]] & ctype_digit) == 0)
5510 {
5511 *errorcodeptr = ERR15;
5512 goto FAILED;
5513 }
5514 recno = recno * 10 + name[i] - CHAR_0;
5515 }
5516 if (recno == 0) recno = RREF_ANY;
5517 code[1+LINK_SIZE] = OP_RREF; /* Change test type */
5518 PUT2(code, 2+LINK_SIZE, recno);
5519 }
5520
5521 /* Similarly, check for the (?(DEFINE) "condition", which is always
5522 false. */
5523
5524 else if (namelen == 6 && strncmp((char *)name, STRING_DEFINE, 6) == 0)
5525 {
5526 code[1+LINK_SIZE] = OP_DEF;
5527 skipbytes = 1;
5528 }
5529
5530 /* Check for the "name" actually being a subpattern number. We are
5531 in the second pass here, so final_bracount is set. */
5532
5533 else if (recno > 0 && recno <= cd->final_bracount)
5534 {
5535 PUT2(code, 2+LINK_SIZE, recno);
5536 }
5537
5538 /* Either an unidentified subpattern, or a reference to (?(0) */
5539
5540 else
5541 {
5542 *errorcodeptr = (recno == 0)? ERR35: ERR15;
5543 goto FAILED;
5544 }
5545 break;
5546
5547
5548 /* ------------------------------------------------------------ */
5549 case CHAR_EQUALS_SIGN: /* Positive lookahead */
5550 bravalue = OP_ASSERT;
5551 cd->assert_depth += 1;
5552 ptr++;
5553 break;
5554
5555
5556 /* ------------------------------------------------------------ */
5557 case CHAR_EXCLAMATION_MARK: /* Negative lookahead */
5558 ptr++;
5559 if (*ptr == CHAR_RIGHT_PARENTHESIS) /* Optimize (?!) */
5560 {
5561 *code++ = OP_FAIL;
5562 previous = NULL;
5563 continue;
5564 }
5565 bravalue = OP_ASSERT_NOT;
5566 cd->assert_depth += 1;
5567 break;
5568
5569
5570 /* ------------------------------------------------------------ */
5571 case CHAR_LESS_THAN_SIGN: /* Lookbehind or named define */
5572 switch (ptr[1])
5573 {
5574 case CHAR_EQUALS_SIGN: /* Positive lookbehind */
5575 bravalue = OP_ASSERTBACK;
5576 cd->assert_depth += 1;
5577 ptr += 2;
5578 break;
5579
5580 case CHAR_EXCLAMATION_MARK: /* Negative lookbehind */
5581 bravalue = OP_ASSERTBACK_NOT;
5582 cd->assert_depth += 1;
5583 ptr += 2;
5584 break;
5585
5586 default: /* Could be name define, else bad */
5587 if ((cd->ctypes[ptr[1]] & ctype_word) != 0) goto DEFINE_NAME;
5588 ptr++; /* Correct offset for error */
5589 *errorcodeptr = ERR24;
5590 goto FAILED;
5591 }
5592 break;
5593
5594
5595 /* ------------------------------------------------------------ */
5596 case CHAR_GREATER_THAN_SIGN: /* One-time brackets */
5597 bravalue = OP_ONCE;
5598 ptr++;
5599 break;
5600
5601
5602 /* ------------------------------------------------------------ */
5603 case CHAR_C: /* Callout - may be followed by digits; */
5604 previous_callout = code; /* Save for later completion */
5605 after_manual_callout = 1; /* Skip one item before completing */
5606 *code++ = OP_CALLOUT;
5607 {
5608 int n = 0;
5609 while ((digitab[*(++ptr)] & ctype_digit) != 0)
5610 n = n * 10 + *ptr - CHAR_0;
5611 if (*ptr != CHAR_RIGHT_PARENTHESIS)
5612 {
5613 *errorcodeptr = ERR39;
5614 goto FAILED;
5615 }
5616 if (n > 255)
5617 {
5618 *errorcodeptr = ERR38;
5619 goto FAILED;
5620 }
5621 *code++ = n;
5622 PUT(code, 0, (int)(ptr - cd->start_pattern + 1)); /* Pattern offset */
5623 PUT(code, LINK_SIZE, 0); /* Default length */
5624 code += 2 * LINK_SIZE;
5625 }
5626 previous = NULL;
5627 continue;
5628
5629
5630 /* ------------------------------------------------------------ */
5631 case CHAR_P: /* Python-style named subpattern handling */
5632 if (*(++ptr) == CHAR_EQUALS_SIGN ||
5633 *ptr == CHAR_GREATER_THAN_SIGN) /* Reference or recursion */
5634 {
5635 is_recurse = *ptr == CHAR_GREATER_THAN_SIGN;
5636 terminator = CHAR_RIGHT_PARENTHESIS;
5637 goto NAMED_REF_OR_RECURSE;
5638 }
5639 else if (*ptr != CHAR_LESS_THAN_SIGN) /* Test for Python-style defn */
5640 {
5641 *errorcodeptr = ERR41;
5642 goto FAILED;
5643 }
5644 /* Fall through to handle (?P< as (?< is handled */
5645
5646
5647 /* ------------------------------------------------------------ */
5648 DEFINE_NAME: /* Come here from (?< handling */
5649 case CHAR_APOSTROPHE:
5650 {
5651 terminator = (*ptr == CHAR_LESS_THAN_SIGN)?
5652 CHAR_GREATER_THAN_SIGN : CHAR_APOSTROPHE;
5653 name = ++ptr;
5654
5655 while ((cd->ctypes[*ptr] & ctype_word) != 0) ptr++;
5656 namelen = (int)(ptr - name);
5657
5658 /* In the pre-compile phase, just do a syntax check. */
5659
5660 if (lengthptr != NULL)
5661 {
5662 if (*ptr != terminator)
5663 {
5664 *errorcodeptr = ERR42;
5665 goto FAILED;
5666 }
5667 if (cd->names_found >= MAX_NAME_COUNT)
5668 {
5669 *errorcodeptr = ERR49;
5670 goto FAILED;
5671 }
5672 if (namelen + 3 > cd->name_entry_size)
5673 {
5674 cd->name_entry_size = namelen + 3;
5675 if (namelen > MAX_NAME_SIZE)
5676 {
5677 *errorcodeptr = ERR48;
5678 goto FAILED;
5679 }
5680 }
5681 }
5682
5683 /* In the real compile, create the entry in the table, maintaining
5684 alphabetical order. Duplicate names for different numbers are
5685 permitted only if PCRE_DUPNAMES is set. Duplicate names for the same
5686 number are always OK. (An existing number can be re-used if (?|
5687 appears in the pattern.) In either event, a duplicate name results in
5688 a duplicate entry in the table, even if the number is the same. This
5689 is because the number of names, and hence the table size, is computed
5690 in the pre-compile, and it affects various numbers and pointers which
5691 would all have to be modified, and the compiled code moved down, if
5692 duplicates with the same number were omitted from the table. This
5693 doesn't seem worth the hassle. However, *different* names for the
5694 same number are not permitted. */
5695
5696 else
5697 {
5698 BOOL dupname = FALSE;
5699 slot = cd->name_table;
5700
5701 for (i = 0; i < cd->names_found; i++)
5702 {
5703 int crc = memcmp(name, slot+2, namelen);
5704 if (crc == 0)
5705 {
5706 if (slot[2+namelen] == 0)
5707 {
5708 if (GET2(slot, 0) != cd->bracount + 1 &&
5709 (options & PCRE_DUPNAMES) == 0)
5710 {
5711 *errorcodeptr = ERR43;
5712 goto FAILED;
5713 }
5714 else dupname = TRUE;
5715 }
5716 else crc = -1; /* Current name is a substring */
5717 }
5718
5719 /* Make space in the table and break the loop for an earlier
5720 name. For a duplicate or later name, carry on. We do this for
5721 duplicates so that in the simple case (when ?(| is not used) they
5722 are in order of their numbers. */
5723
5724 if (crc < 0)
5725 {
5726 memmove(slot + cd->name_entry_size, slot,
5727 (cd->names_found - i) * cd->name_entry_size);
5728 break;
5729 }
5730
5731 /* Continue the loop for a later or duplicate name */
5732
5733 slot += cd->name_entry_size;
5734 }
5735
5736 /* For non-duplicate names, check for a duplicate number before
5737 adding the new name. */
5738
5739 if (!dupname)
5740 {
5741 uschar *cslot = cd->name_table;
5742 for (i = 0; i < cd->names_found; i++)
5743 {
5744 if (cslot != slot)
5745 {
5746 if (GET2(cslot, 0) == cd->bracount + 1)
5747 {
5748 *errorcodeptr = ERR65;
5749 goto FAILED;
5750 }
5751 }
5752 else i--;
5753 cslot += cd->name_entry_size;
5754 }
5755 }
5756
5757 PUT2(slot, 0, cd->bracount + 1);
5758 memcpy(slot + 2, name, namelen);
5759 slot[2+namelen] = 0;
5760 }
5761 }
5762
5763 /* In both pre-compile and compile, count the number of names we've
5764 encountered. */
5765
5766 cd->names_found++;
5767 ptr++; /* Move past > or ' */
5768 goto NUMBERED_GROUP;
5769
5770
5771 /* ------------------------------------------------------------ */
5772 case CHAR_AMPERSAND: /* Perl recursion/subroutine syntax */
5773 terminator = CHAR_RIGHT_PARENTHESIS;
5774 is_recurse = TRUE;
5775 /* Fall through */
5776
5777 /* We come here from the Python syntax above that handles both
5778 references (?P=name) and recursion (?P>name), as well as falling
5779 through from the Perl recursion syntax (?&name). We also come here from
5780 the Perl \k<name> or \k'name' back reference syntax and the \k{name}
5781 .NET syntax, and the Oniguruma \g<...> and \g'...' subroutine syntax. */
5782
5783 NAMED_REF_OR_RECURSE:
5784 name = ++ptr;
5785 while ((cd->ctypes[*ptr] & ctype_word) != 0) ptr++;
5786 namelen = (int)(ptr - name);
5787
5788 /* In the pre-compile phase, do a syntax check. We used to just set
5789 a dummy reference number, because it was not used in the first pass.
5790 However, with the change of recursive back references to be atomic,
5791 we have to look for the number so that this state can be identified, as
5792 otherwise the incorrect length is computed. If it's not a backwards
5793 reference, the dummy number will do. */
5794
5795 if (lengthptr != NULL)
5796 {
5797 const uschar *temp;
5798
5799 if (namelen == 0)
5800 {
5801 *errorcodeptr = ERR62;
5802 goto FAILED;
5803 }
5804 if (*ptr != terminator)
5805 {
5806 *errorcodeptr = ERR42;
5807 goto FAILED;
5808 }
5809 if (namelen > MAX_NAME_SIZE)
5810 {
5811 *errorcodeptr = ERR48;
5812 goto FAILED;
5813 }
5814
5815 /* The name table does not exist in the first pass, so we cannot
5816 do a simple search as in the code below. Instead, we have to scan the
5817 pattern to find the number. It is important that we scan it only as
5818 far as we have got because the syntax of named subpatterns has not
5819 been checked for the rest of the pattern, and find_parens() assumes
5820 correct syntax. In any case, it's a waste of resources to scan
5821 further. We stop the scan at the current point by temporarily
5822 adjusting the value of cd->endpattern. */
5823
5824 temp = cd->end_pattern;
5825 cd->end_pattern = ptr;
5826 recno = find_parens(cd, name, namelen,
5827 (options & PCRE_EXTENDED) != 0, utf8);
5828 cd->end_pattern = temp;
5829 if (recno < 0) recno = 0; /* Forward ref; set dummy number */
5830 }
5831
5832 /* In the real compile, seek the name in the table. We check the name
5833 first, and then check that we have reached the end of the name in the
5834 table. That way, if the name that is longer than any in the table,
5835 the comparison will fail without reading beyond the table entry. */
5836
5837 else
5838 {
5839 slot = cd->name_table;
5840 for (i = 0; i < cd->names_found; i++)
5841 {
5842 if (strncmp((char *)name, (char *)slot+2, namelen) == 0 &&
5843 slot[2+namelen] == 0)
5844 break;
5845 slot += cd->name_entry_size;
5846 }
5847
5848 if (i < cd->names_found) /* Back reference */
5849 {
5850 recno = GET2(slot, 0);
5851 }
5852 else if ((recno = /* Forward back reference */
5853 find_parens(cd, name, namelen,
5854 (options & PCRE_EXTENDED) != 0, utf8)) <= 0)
5855 {
5856 *errorcodeptr = ERR15;
5857 goto FAILED;
5858 }
5859 }
5860
5861 /* In both phases, we can now go to the code than handles numerical
5862 recursion or backreferences. */
5863
5864 if (is_recurse) goto HANDLE_RECURSION;
5865 else goto HANDLE_REFERENCE;
5866
5867
5868 /* ------------------------------------------------------------ */
5869 case CHAR_R: /* Recursion */
5870 ptr++; /* Same as (?0) */
5871 /* Fall through */
5872
5873
5874 /* ------------------------------------------------------------ */
5875 case CHAR_MINUS: case CHAR_PLUS: /* Recursion or subroutine */
5876 case CHAR_0: case CHAR_1: case CHAR_2: case CHAR_3: case CHAR_4:
5877 case CHAR_5: case CHAR_6: case CHAR_7: case CHAR_8: case CHAR_9:
5878 {
5879 const uschar *called;
5880 terminator = CHAR_RIGHT_PARENTHESIS;
5881
5882 /* Come here from the \g<...> and \g'...' code (Oniguruma
5883 compatibility). However, the syntax has been checked to ensure that
5884 the ... are a (signed) number, so that neither ERR63 nor ERR29 will
5885 be called on this path, nor with the jump to OTHER_CHAR_AFTER_QUERY
5886 ever be taken. */
5887
5888 HANDLE_NUMERICAL_RECURSION:
5889
5890 if ((refsign = *ptr) == CHAR_PLUS)
5891 {
5892 ptr++;
5893 if ((digitab[*ptr] & ctype_digit) == 0)
5894 {
5895 *errorcodeptr = ERR63;
5896 goto FAILED;
5897 }
5898 }
5899 else if (refsign == CHAR_MINUS)
5900 {
5901 if ((digitab[ptr[1]] & ctype_digit) == 0)
5902 goto OTHER_CHAR_AFTER_QUERY;
5903 ptr++;
5904 }
5905
5906 recno = 0;
5907 while((digitab[*ptr] & ctype_digit) != 0)
5908 recno = recno * 10 + *ptr++ - CHAR_0;
5909
5910 if (*ptr != terminator)
5911 {
5912 *errorcodeptr = ERR29;
5913 goto FAILED;
5914 }
5915
5916 if (refsign == CHAR_MINUS)
5917 {
5918 if (recno == 0)
5919 {
5920 *errorcodeptr = ERR58;
5921 goto FAILED;
5922 }
5923 recno = cd->bracount - recno + 1;
5924 if (recno <= 0)
5925 {
5926 *errorcodeptr = ERR15;
5927 goto FAILED;
5928 }
5929 }
5930 else if (refsign == CHAR_PLUS)
5931 {
5932 if (recno == 0)
5933 {
5934 *errorcodeptr = ERR58;
5935 goto FAILED;
5936 }
5937 recno += cd->bracount;
5938 }
5939
5940 /* Come here from code above that handles a named recursion */
5941
5942 HANDLE_RECURSION:
5943
5944 previous = code;
5945 called = cd->start_code;
5946
5947 /* When we are actually compiling, find the bracket that is being
5948 referenced. Temporarily end the regex in case it doesn't exist before
5949 this point. If we end up with a forward reference, first check that
5950 the bracket does occur later so we can give the error (and position)
5951 now. Then remember this forward reference in the workspace so it can
5952 be filled in at the end. */
5953
5954 if (lengthptr == NULL)
5955 {
5956 *code = OP_END;
5957 if (recno != 0)
5958 called = _pcre_find_bracket(cd->start_code, utf8, recno);
5959
5960 /* Forward reference */
5961
5962 if (called == NULL)
5963 {
5964 if (find_parens(cd, NULL, recno,
5965 (options & PCRE_EXTENDED) != 0, utf8) < 0)
5966 {
5967 *errorcodeptr = ERR15;
5968 goto FAILED;
5969 }
5970
5971 /* Fudge the value of "called" so that when it is inserted as an
5972 offset below, what it actually inserted is the reference number
5973 of the group. Then remember the forward reference. */
5974
5975 called = cd->start_code + recno;
5976 PUTINC(cd->hwm, 0, (int)(code + 1 - cd->start_code));
5977 }
5978
5979 /* If not a forward reference, and the subpattern is still open,
5980 this is a recursive call. We check to see if this is a left
5981 recursion that could loop for ever, and diagnose that case. We
5982 must not, however, do this check if we are in a conditional
5983 subpattern because the condition might be testing for recursion in
5984 a pattern such as /(?(R)a+|(?R)b)/, which is perfectly valid.
5985 Forever loops are also detected at runtime, so those that occur in
5986 conditional subpatterns will be picked up then. */
5987
5988 else if (GET(called, 1) == 0 && cond_depth <= 0 &&
5989 could_be_empty(called, code, bcptr, utf8, cd))
5990 {
5991 *errorcodeptr = ERR40;
5992 goto FAILED;
5993 }
5994 }
5995
5996 /* Insert the recursion/subroutine item. */
5997
5998 *code = OP_RECURSE;
5999 PUT(code, 1, (int)(called - cd->start_code));
6000 code += 1 + LINK_SIZE;
6001 }
6002
6003 /* Can't determine a first byte now */
6004
6005 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
6006 continue;
6007
6008
6009 /* ------------------------------------------------------------ */
6010 default: /* Other characters: check option setting */
6011 OTHER_CHAR_AFTER_QUERY:
6012 set = unset = 0;
6013 optset = &set;
6014
6015 while (*ptr != CHAR_RIGHT_PARENTHESIS && *ptr != CHAR_COLON)
6016 {
6017 switch (*ptr++)
6018 {
6019 case CHAR_MINUS: optset = &unset; break;
6020
6021 case CHAR_J: /* Record that it changed in the external options */
6022 *optset |= PCRE_DUPNAMES;
6023 cd->external_flags |= PCRE_JCHANGED;
6024 break;
6025
6026 case CHAR_i: *optset |= PCRE_CASELESS; break;
6027 case CHAR_m: *optset |= PCRE_MULTILINE; break;
6028 case CHAR_s: *optset |= PCRE_DOTALL; break;
6029 case CHAR_x: *optset |= PCRE_EXTENDED; break;
6030 case CHAR_U: *optset |= PCRE_UNGREEDY; break;
6031 case CHAR_X: *optset |= PCRE_EXTRA; break;
6032
6033 default: *errorcodeptr = ERR12;
6034 ptr--; /* Correct the offset */
6035 goto FAILED;
6036 }
6037 }
6038
6039 /* Set up the changed option bits, but don't change anything yet. */
6040
6041 newoptions = (options | set) & (~unset);
6042
6043 /* If the options ended with ')' this is not the start of a nested
6044 group with option changes, so the options change at this level. If this
6045 item is right at the start of the pattern, the options can be
6046 abstracted and made external in the pre-compile phase, and ignored in
6047 the compile phase. This can be helpful when matching -- for instance in
6048 caseless checking of required bytes.
6049
6050 If the code pointer is not (cd->start_code + 1 + LINK_SIZE), we are
6051 definitely *not* at the start of the pattern because something has been
6052 compiled. In the pre-compile phase, however, the code pointer can have
6053 that value after the start, because it gets reset as code is discarded
6054 during the pre-compile. However, this can happen only at top level - if
6055 we are within parentheses, the starting BRA will still be present. At
6056 any parenthesis level, the length value can be used to test if anything
6057 has been compiled at that level. Thus, a test for both these conditions
6058 is necessary to ensure we correctly detect the start of the pattern in
6059 both phases.
6060
6061 If we are not at the pattern start, reset the greedy defaults and the
6062 case value for firstbyte and reqbyte. */
6063
6064 if (*ptr == CHAR_RIGHT_PARENTHESIS)
6065 {
6066 if (code == cd->start_code + 1 + LINK_SIZE &&
6067 (lengthptr == NULL || *lengthptr == 2 + 2*LINK_SIZE))
6068 {
6069 cd->external_options = newoptions;
6070 }
6071 else
6072 {
6073 greedy_default = ((newoptions & PCRE_UNGREEDY) != 0);
6074 greedy_non_default = greedy_default ^ 1;
6075 req_caseopt = ((newoptions & PCRE_CASELESS) != 0)? REQ_CASELESS : 0;
6076 }
6077
6078 /* Change options at this level, and pass them back for use
6079 in subsequent branches. */
6080
6081 *optionsptr = options = newoptions;
6082 previous = NULL; /* This item can't be repeated */
6083 continue; /* It is complete */
6084 }
6085
6086 /* If the options ended with ':' we are heading into a nested group
6087 with possible change of options. Such groups are non-capturing and are
6088 not assertions of any kind. All we need to do is skip over the ':';
6089 the newoptions value is handled below. */
6090
6091 bravalue = OP_BRA;
6092 ptr++;
6093 } /* End of switch for character following (? */
6094 } /* End of (? handling */
6095
6096 /* Opening parenthesis not followed by '*' or '?'. If PCRE_NO_AUTO_CAPTURE
6097 is set, all unadorned brackets become non-capturing and behave like (?:...)
6098 brackets. */
6099
6100 else if ((options & PCRE_NO_AUTO_CAPTURE) != 0)
6101 {
6102 bravalue = OP_BRA;
6103 }
6104
6105 /* Else we have a capturing group. */
6106
6107 else
6108 {
6109 NUMBERED_GROUP:
6110 cd->bracount += 1;
6111 PUT2(code, 1+LINK_SIZE, cd->bracount);
6112 skipbytes = 2;
6113 }
6114
6115 /* Process nested bracketed regex. Assertions used not to be repeatable,
6116 but this was changed for Perl compatibility, so all kinds can now be
6117 repeated. We copy code into a non-register variable (tempcode) in order to
6118 be able to pass its address because some compilers complain otherwise. */
6119
6120 previous = code; /* For handling repetition */
6121 *code = bravalue;
6122 tempcode = code;
6123 tempreqvary = cd->req_varyopt; /* Save value before bracket */
6124 tempbracount = cd->bracount; /* Save value before bracket */
6125 length_prevgroup = 0; /* Initialize for pre-compile phase */
6126
6127 if (!compile_regex(
6128 newoptions, /* The complete new option state */
6129 &tempcode, /* Where to put code (updated) */
6130 &ptr, /* Input pointer (updated) */
6131 errorcodeptr, /* Where to put an error message */
6132 (bravalue == OP_ASSERTBACK ||
6133 bravalue == OP_ASSERTBACK_NOT), /* TRUE if back assert */
6134 reset_bracount, /* True if (?| group */
6135 skipbytes, /* Skip over bracket number */
6136 cond_depth +
6137 ((bravalue == OP_COND)?1:0), /* Depth of condition subpatterns */
6138 &subfirstbyte, /* For possible first char */
6139 &subreqbyte, /* For possible last char */
6140 bcptr, /* Current branch chain */
6141 cd, /* Tables block */
6142 (lengthptr == NULL)? NULL : /* Actual compile phase */
6143 &length_prevgroup /* Pre-compile phase */
6144 ))
6145 goto FAILED;
6146
6147 /* If this was an atomic group and there are no capturing groups within it,
6148 generate OP_ONCE_NC instead of OP_ONCE. */
6149
6150 if (bravalue == OP_ONCE && cd->bracount <= tempbracount)
6151 *code = OP_ONCE_NC;
6152
6153 if (bravalue >= OP_ASSERT && bravalue <= OP_ASSERTBACK_NOT)
6154 cd->assert_depth -= 1;
6155
6156 /* At the end of compiling, code is still pointing to the start of the
6157 group, while tempcode has been updated to point past the end of the group.
6158 The pattern pointer (ptr) is on the bracket.
6159
6160 If this is a conditional bracket, check that there are no more than
6161 two branches in the group, or just one if it's a DEFINE group. We do this
6162 in the real compile phase, not in the pre-pass, where the whole group may
6163 not be available. */
6164
6165 if (bravalue == OP_COND && lengthptr == NULL)
6166 {
6167 uschar *tc = code;
6168 int condcount = 0;
6169
6170 do {
6171 condcount++;
6172 tc += GET(tc,1);
6173 }
6174 while (*tc != OP_KET);
6175
6176 /* A DEFINE group is never obeyed inline (the "condition" is always
6177 false). It must have only one branch. */
6178
6179 if (code[LINK_SIZE+1] == OP_DEF)
6180 {
6181 if (condcount > 1)
6182 {
6183 *errorcodeptr = ERR54;
6184 goto FAILED;
6185 }
6186 bravalue = OP_DEF; /* Just a flag to suppress char handling below */
6187 }
6188
6189 /* A "normal" conditional group. If there is just one branch, we must not
6190 make use of its firstbyte or reqbyte, because this is equivalent to an
6191 empty second branch. */
6192
6193 else
6194 {
6195 if (condcount > 2)
6196 {
6197 *errorcodeptr = ERR27;
6198 goto FAILED;
6199 }
6200 if (condcount == 1) subfirstbyte = subreqbyte = REQ_NONE;
6201 }
6202 }
6203
6204 /* Error if hit end of pattern */
6205
6206 if (*ptr != CHAR_RIGHT_PARENTHESIS)
6207 {
6208 *errorcodeptr = ERR14;
6209 goto FAILED;
6210 }
6211
6212 /* In the pre-compile phase, update the length by the length of the group,
6213 less the brackets at either end. Then reduce the compiled code to just a
6214 set of non-capturing brackets so that it doesn't use much memory if it is
6215 duplicated by a quantifier.*/
6216
6217 if (lengthptr != NULL)
6218 {
6219 if (OFLOW_MAX - *lengthptr < length_prevgroup - 2 - 2*LINK_SIZE)
6220 {
6221 *errorcodeptr = ERR20;
6222 goto FAILED;
6223 }
6224 *lengthptr += length_prevgroup - 2 - 2*LINK_SIZE;
6225 code++; /* This already contains bravalue */
6226 PUTINC(code, 0, 1 + LINK_SIZE);
6227 *code++ = OP_KET;
6228 PUTINC(code, 0, 1 + LINK_SIZE);
6229 break; /* No need to waste time with special character handling */
6230 }
6231
6232 /* Otherwise update the main code pointer to the end of the group. */
6233
6234 code = tempcode;
6235
6236 /* For a DEFINE group, required and first character settings are not
6237 relevant. */
6238
6239 if (bravalue == OP_DEF) break;
6240
6241 /* Handle updating of the required and first characters for other types of
6242 group. Update for normal brackets of all kinds, and conditions with two
6243 branches (see code above). If the bracket is followed by a quantifier with
6244 zero repeat, we have to back off. Hence the definition of zeroreqbyte and
6245 zerofirstbyte outside the main loop so that they can be accessed for the
6246 back off. */
6247
6248 zeroreqbyte = reqbyte;
6249 zerofirstbyte = firstbyte;
6250 groupsetfirstbyte = FALSE;
6251
6252 if (bravalue >= OP_ONCE)
6253 {
6254 /* If we have not yet set a firstbyte in this branch, take it from the
6255 subpattern, remembering that it was set here so that a repeat of more
6256 than one can replicate it as reqbyte if necessary. If the subpattern has
6257 no firstbyte, set "none" for the whole branch. In both cases, a zero
6258 repeat forces firstbyte to "none". */
6259
6260 if (firstbyte == REQ_UNSET)
6261 {
6262 if (subfirstbyte >= 0)
6263 {
6264 firstbyte = subfirstbyte;
6265 groupsetfirstbyte = TRUE;
6266 }
6267 else firstbyte = REQ_NONE;
6268 zerofirstbyte = REQ_NONE;
6269 }
6270
6271 /* If firstbyte was previously set, convert the subpattern's firstbyte
6272 into reqbyte if there wasn't one, using the vary flag that was in
6273 existence beforehand. */
6274
6275 else if (subfirstbyte >= 0 && subreqbyte < 0)
6276 subreqbyte = subfirstbyte | tempreqvary;
6277
6278 /* If the subpattern set a required byte (or set a first byte that isn't
6279 really the first byte - see above), set it. */
6280
6281 if (subreqbyte >= 0) reqbyte = subreqbyte;
6282 }
6283
6284 /* For a forward assertion, we take the reqbyte, if set. This can be
6285 helpful if the pattern that follows the assertion doesn't set a different
6286 char. For example, it's useful for /(?=abcde).+/. We can't set firstbyte
6287 for an assertion, however because it leads to incorrect effect for patterns
6288 such as /(?=a)a.+/ when the "real" "a" would then become a reqbyte instead
6289 of a firstbyte. This is overcome by a scan at the end if there's no
6290 firstbyte, looking for an asserted first char. */
6291
6292 else if (bravalue == OP_ASSERT && subreqbyte >= 0) reqbyte = subreqbyte;
6293 break; /* End of processing '(' */
6294
6295
6296 /* ===================================================================*/
6297 /* Handle metasequences introduced by \. For ones like \d, the ESC_ values
6298 are arranged to be the negation of the corresponding OP_values in the
6299 default case when PCRE_UCP is not set. For the back references, the values
6300 are ESC_REF plus the reference number. Only back references and those types
6301 that consume a character may be repeated. We can test for values between
6302 ESC_b and ESC_Z for the latter; this may have to change if any new ones are
6303 ever created. */
6304
6305 case CHAR_BACKSLASH:
6306 tempptr = ptr;
6307 c = check_escape(&ptr, errorcodeptr, cd->bracount, options, FALSE);
6308 if (*errorcodeptr != 0) goto FAILED;
6309
6310 if (c < 0)
6311 {
6312 if (-c == ESC_Q) /* Handle start of quoted string */
6313 {
6314 if (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
6315 ptr += 2; /* avoid empty string */
6316 else inescq = TRUE;
6317 continue;
6318 }
6319
6320 if (-c == ESC_E) continue; /* Perl ignores an orphan \E */
6321
6322 /* For metasequences that actually match a character, we disable the
6323 setting of a first character if it hasn't already been set. */
6324
6325 if (firstbyte == REQ_UNSET && -c > ESC_b && -c < ESC_Z)
6326 firstbyte = REQ_NONE;
6327
6328 /* Set values to reset to if this is followed by a zero repeat. */
6329
6330 zerofirstbyte = firstbyte;
6331 zeroreqbyte = reqbyte;
6332
6333 /* \g<name> or \g'name' is a subroutine call by name and \g<n> or \g'n'
6334 is a subroutine call by number (Oniguruma syntax). In fact, the value
6335 -ESC_g is returned only for these cases. So we don't need to check for <
6336 or ' if the value is -ESC_g. For the Perl syntax \g{n} the value is
6337 -ESC_REF+n, and for the Perl syntax \g{name} the result is -ESC_k (as
6338 that is a synonym for a named back reference). */
6339
6340 if (-c == ESC_g)
6341 {
6342 const uschar *p;
6343 save_hwm = cd->hwm; /* Normally this is set when '(' is read */
6344 terminator = (*(++ptr) == CHAR_LESS_THAN_SIGN)?
6345 CHAR_GREATER_THAN_SIGN : CHAR_APOSTROPHE;
6346
6347 /* These two statements stop the compiler for warning about possibly
6348 unset variables caused by the jump to HANDLE_NUMERICAL_RECURSION. In
6349 fact, because we actually check for a number below, the paths that
6350 would actually be in error are never taken. */
6351
6352 skipbytes = 0;
6353 reset_bracount = FALSE;
6354
6355 /* Test for a name */
6356
6357 if (ptr[1] != CHAR_PLUS && ptr[1] != CHAR_MINUS)
6358 {
6359 BOOL isnumber = TRUE;
6360 for (p = ptr + 1; *p != 0 && *p != terminator; p++)
6361 {
6362 if ((cd->ctypes[*p] & ctype_digit) == 0) isnumber = FALSE;
6363 if ((cd->ctypes[*p] & ctype_word) == 0) break;
6364 }
6365 if (*p != terminator)
6366 {
6367 *errorcodeptr = ERR57;
6368 break;
6369 }
6370 if (isnumber)
6371 {
6372 ptr++;
6373 goto HANDLE_NUMERICAL_RECURSION;
6374 }
6375 is_recurse = TRUE;
6376 goto NAMED_REF_OR_RECURSE;
6377 }
6378
6379 /* Test a signed number in angle brackets or quotes. */
6380
6381 p = ptr + 2;
6382 while ((digitab[*p] & ctype_digit) != 0) p++;
6383 if (*p != terminator)
6384 {
6385 *errorcodeptr = ERR57;
6386 break;
6387 }
6388 ptr++;
6389 goto HANDLE_NUMERICAL_RECURSION;
6390 }
6391
6392 /* \k<name> or \k'name' is a back reference by name (Perl syntax).
6393 We also support \k{name} (.NET syntax). */
6394
6395 if (-c == ESC_k)
6396 {
6397 if ((ptr[1] != CHAR_LESS_THAN_SIGN &&
6398 ptr[1] != CHAR_APOSTROPHE && ptr[1] != CHAR_LEFT_CURLY_BRACKET))
6399 {
6400 *errorcodeptr = ERR69;
6401 break;
6402 }
6403 is_recurse = FALSE;
6404 terminator = (*(++ptr) == CHAR_LESS_THAN_SIGN)?
6405 CHAR_GREATER_THAN_SIGN : (*ptr == CHAR_APOSTROPHE)?
6406 CHAR_APOSTROPHE : CHAR_RIGHT_CURLY_BRACKET;
6407 goto NAMED_REF_OR_RECURSE;
6408 }
6409
6410 /* Back references are handled specially; must disable firstbyte if
6411 not set to cope with cases like (?=(\w+))\1: which would otherwise set
6412 ':' later. */
6413
6414 if (-c >= ESC_REF)
6415 {
6416 open_capitem *oc;
6417 recno = -c - ESC_REF;
6418
6419 HANDLE_REFERENCE: /* Come here from named backref handling */
6420 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
6421 previous = code;
6422 *code++ = ((options & PCRE_CASELESS) != 0)? OP_REFI : OP_REF;
6423 PUT2INC(code, 0, recno);
6424 cd->backref_map |= (recno < 32)? (1 << recno) : 1;
6425 if (recno > cd->top_backref) cd->top_backref = recno;
6426
6427 /* Check to see if this back reference is recursive, that it, it
6428 is inside the group that it references. A flag is set so that the
6429 group can be made atomic. */
6430
6431 for (oc = cd->open_caps; oc != NULL; oc = oc->next)
6432 {
6433 if (oc->number == recno)
6434 {
6435 oc->flag = TRUE;
6436 break;
6437 }
6438 }
6439 }
6440
6441 /* So are Unicode property matches, if supported. */
6442
6443 #ifdef SUPPORT_UCP
6444 else if (-c == ESC_P || -c == ESC_p)
6445 {
6446 BOOL negated;
6447 int pdata;
6448 int ptype = get_ucp(&ptr, &negated, &pdata, errorcodeptr);
6449 if (ptype < 0) goto FAILED;
6450 previous = code;
6451 *code++ = ((-c == ESC_p) != negated)? OP_PROP : OP_NOTPROP;
6452 *code++ = ptype;
6453 *code++ = pdata;
6454 }
6455 #else
6456
6457 /* If Unicode properties are not supported, \X, \P, and \p are not
6458 allowed. */
6459
6460 else if (-c == ESC_X || -c == ESC_P || -c == ESC_p)
6461 {
6462 *errorcodeptr = ERR45;
6463 goto FAILED;
6464 }
6465 #endif
6466
6467 /* For the rest (including \X when Unicode properties are supported), we
6468 can obtain the OP value by negating the escape value in the default
6469 situation when PCRE_UCP is not set. When it *is* set, we substitute
6470 Unicode property tests. */
6471
6472 else
6473 {
6474 #ifdef SUPPORT_UCP
6475 if (-c >= ESC_DU && -c <= ESC_wu)
6476 {
6477 nestptr = ptr + 1; /* Where to resume */
6478 ptr = substitutes[-c - ESC_DU] - 1; /* Just before substitute */
6479 }
6480 else
6481 #endif
6482 /* In non-UTF-8 mode, we turn \C into OP_ALLANY instead of OP_ANYBYTE
6483 so that it works in DFA mode and in lookbehinds. */
6484
6485 {
6486 previous = (-c > ESC_b && -c < ESC_Z)? code : NULL;
6487 *code++ = (!utf8 && c == -ESC_C)? OP_ALLANY : -c;
6488 }
6489 }
6490 continue;
6491 }
6492
6493 /* We have a data character whose value is in c. In UTF-8 mode it may have
6494 a value > 127. We set its representation in the length/buffer, and then
6495 handle it as a data character. */
6496
6497 #ifdef SUPPORT_UTF8
6498 if (utf8 && c > 127)
6499 mclength = _pcre_ord2utf8(c, mcbuffer);
6500 else
6501 #endif
6502
6503 {
6504 mcbuffer[0] = c;
6505 mclength = 1;
6506 }
6507 goto ONE_CHAR;
6508
6509
6510 /* ===================================================================*/
6511 /* Handle a literal character. It is guaranteed not to be whitespace or #
6512 when the extended flag is set. If we are in UTF-8 mode, it may be a
6513 multi-byte literal character. */
6514
6515 default:
6516 NORMAL_CHAR:
6517 mclength = 1;
6518 mcbuffer[0] = c;
6519
6520 #ifdef SUPPORT_UTF8
6521 if (utf8 && c >= 0xc0)
6522 {
6523 while ((ptr[1] & 0xc0) == 0x80)
6524 mcbuffer[mclength++] = *(++ptr);
6525 }
6526 #endif
6527
6528 /* At this point we have the character's bytes in mcbuffer, and the length
6529 in mclength. When not in UTF-8 mode, the length is always 1. */
6530
6531 ONE_CHAR:
6532 previous = code;
6533 *code++ = ((options & PCRE_CASELESS) != 0)? OP_CHARI : OP_CHAR;
6534 for (c = 0; c < mclength; c++) *code++ = mcbuffer[c];
6535
6536 /* Remember if \r or \n were seen */
6537
6538 if (mcbuffer[0] == CHAR_CR || mcbuffer[0] == CHAR_NL)
6539 cd->external_flags |= PCRE_HASCRORLF;
6540
6541 /* Set the first and required bytes appropriately. If no previous first
6542 byte, set it from this character, but revert to none on a zero repeat.
6543 Otherwise, leave the firstbyte value alone, and don't change it on a zero
6544 repeat. */
6545
6546 if (firstbyte == REQ_UNSET)
6547 {
6548 zerofirstbyte = REQ_NONE;
6549 zeroreqbyte = reqbyte;
6550
6551 /* If the character is more than one byte long, we can set firstbyte
6552 only if it is not to be matched caselessly. */
6553
6554 if (mclength == 1 || req_caseopt == 0)
6555 {
6556 firstbyte = mcbuffer[0] | req_caseopt;
6557 if (mclength != 1) reqbyte = code[-1] | cd->req_varyopt;
6558 }
6559 else firstbyte = reqbyte = REQ_NONE;
6560 }
6561
6562 /* firstbyte was previously set; we can set reqbyte only if the length is
6563 1 or the matching is caseful. */
6564
6565 else
6566 {
6567 zerofirstbyte = firstbyte;
6568 zeroreqbyte = reqbyte;
6569 if (mclength == 1 || req_caseopt == 0)
6570 reqbyte = code[-1] | req_caseopt | cd->req_varyopt;
6571 }
6572
6573 break; /* End of literal character handling */
6574 }
6575 } /* end of big loop */
6576
6577
6578 /* Control never reaches here by falling through, only by a goto for all the
6579 error states. Pass back the position in the pattern so that it can be displayed
6580 to the user for diagnosing the error. */
6581
6582 FAILED:
6583 *ptrptr = ptr;
6584 return FALSE;
6585 }
6586
6587
6588
6589
6590 /*************************************************
6591 * Compile sequence of alternatives *
6592 *************************************************/
6593
6594 /* On entry, ptr is pointing past the bracket character, but on return it
6595 points to the closing bracket, or vertical bar, or end of string. The code
6596 variable is pointing at the byte into which the BRA operator has been stored.
6597 This function is used during the pre-compile phase when we are trying to find
6598 out the amount of memory needed, as well as during the real compile phase. The
6599 value of lengthptr distinguishes the two phases.
6600
6601 Arguments:
6602 options option bits, including any changes for this subpattern
6603 codeptr -> the address of the current code pointer
6604 ptrptr -> the address of the current pattern pointer
6605 errorcodeptr -> pointer to error code variable
6606 lookbehind TRUE if this is a lookbehind assertion