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

Contents of /code/trunk/pcre_compile.c

Parent Directory Parent Directory | Revision Log Revision Log


Revision 758 - (show annotations)
Mon Nov 21 12:05:36 2011 UTC (7 years, 9 months ago) by ph10
File MIME type: text/plain
File size: 251043 byte(s)
Disallow \N in character classes, for Perl compatibility.
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 "\\N is not supported in a class\0"
416 ;
417
418 /* Table to identify digits and hex digits. This is used when compiling
419 patterns. Note that the tables in chartables are dependent on the locale, and
420 may mark arbitrary characters as digits - but the PCRE compiling code expects
421 to handle only 0-9, a-z, and A-Z as digits when compiling. That is why we have
422 a private table here. It costs 256 bytes, but it is a lot faster than doing
423 character value tests (at least in some simple cases I timed), and in some
424 applications one wants PCRE to compile efficiently as well as match
425 efficiently.
426
427 For convenience, we use the same bit definitions as in chartables:
428
429 0x04 decimal digit
430 0x08 hexadecimal digit
431
432 Then we can use ctype_digit and ctype_xdigit in the code. */
433
434 #ifndef EBCDIC
435
436 /* This is the "normal" case, for ASCII systems, and EBCDIC systems running in
437 UTF-8 mode. */
438
439 static const unsigned char digitab[] =
440 {
441 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 0- 7 */
442 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 8- 15 */
443 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 16- 23 */
444 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 24- 31 */
445 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - ' */
446 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* ( - / */
447 0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c, /* 0 - 7 */
448 0x0c,0x0c,0x00,0x00,0x00,0x00,0x00,0x00, /* 8 - ? */
449 0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* @ - G */
450 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* H - O */
451 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* P - W */
452 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* X - _ */
453 0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* ` - g */
454 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* h - o */
455 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* p - w */
456 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* x -127 */
457 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 128-135 */
458 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 136-143 */
459 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 144-151 */
460 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 152-159 */
461 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 160-167 */
462 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 168-175 */
463 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 176-183 */
464 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 184-191 */
465 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 192-199 */
466 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 200-207 */
467 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 208-215 */
468 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 216-223 */
469 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 224-231 */
470 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 232-239 */
471 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 240-247 */
472 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00};/* 248-255 */
473
474 #else
475
476 /* This is the "abnormal" case, for EBCDIC systems not running in UTF-8 mode. */
477
478 static const unsigned char digitab[] =
479 {
480 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 0- 7 0 */
481 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 8- 15 */
482 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 16- 23 10 */
483 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 24- 31 */
484 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 32- 39 20 */
485 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 40- 47 */
486 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 48- 55 30 */
487 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 56- 63 */
488 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - 71 40 */
489 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 72- | */
490 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* & - 87 50 */
491 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 88- 95 */
492 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - -103 60 */
493 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 104- ? */
494 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 112-119 70 */
495 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 120- " */
496 0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* 128- g 80 */
497 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* h -143 */
498 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 144- p 90 */
499 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* q -159 */
500 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 160- x A0 */
501 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* y -175 */
502 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* ^ -183 B0 */
503 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 184-191 */
504 0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* { - G C0 */
505 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* H -207 */
506 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* } - P D0 */
507 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* Q -223 */
508 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* \ - X E0 */
509 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* Y -239 */
510 0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c, /* 0 - 7 F0 */
511 0x0c,0x0c,0x00,0x00,0x00,0x00,0x00,0x00};/* 8 -255 */
512
513 static const unsigned char ebcdic_chartab[] = { /* chartable partial dup */
514 0x80,0x00,0x00,0x00,0x00,0x01,0x00,0x00, /* 0- 7 */
515 0x00,0x00,0x00,0x00,0x01,0x01,0x00,0x00, /* 8- 15 */
516 0x00,0x00,0x00,0x00,0x00,0x01,0x00,0x00, /* 16- 23 */
517 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 24- 31 */
518 0x00,0x00,0x00,0x00,0x00,0x01,0x00,0x00, /* 32- 39 */
519 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 40- 47 */
520 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 48- 55 */
521 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 56- 63 */
522 0x01,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - 71 */
523 0x00,0x00,0x00,0x80,0x00,0x80,0x80,0x80, /* 72- | */
524 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* & - 87 */
525 0x00,0x00,0x00,0x80,0x80,0x80,0x00,0x00, /* 88- 95 */
526 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - -103 */
527 0x00,0x00,0x00,0x00,0x00,0x10,0x00,0x80, /* 104- ? */
528 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 112-119 */
529 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 120- " */
530 0x00,0x1a,0x1a,0x1a,0x1a,0x1a,0x1a,0x12, /* 128- g */
531 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* h -143 */
532 0x00,0x12,0x12,0x12,0x12,0x12,0x12,0x12, /* 144- p */
533 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* q -159 */
534 0x00,0x00,0x12,0x12,0x12,0x12,0x12,0x12, /* 160- x */
535 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* y -175 */
536 0x80,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* ^ -183 */
537 0x00,0x00,0x80,0x00,0x00,0x00,0x00,0x00, /* 184-191 */
538 0x80,0x1a,0x1a,0x1a,0x1a,0x1a,0x1a,0x12, /* { - G */
539 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* H -207 */
540 0x00,0x12,0x12,0x12,0x12,0x12,0x12,0x12, /* } - P */
541 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* Q -223 */
542 0x00,0x00,0x12,0x12,0x12,0x12,0x12,0x12, /* \ - X */
543 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* Y -239 */
544 0x1c,0x1c,0x1c,0x1c,0x1c,0x1c,0x1c,0x1c, /* 0 - 7 */
545 0x1c,0x1c,0x00,0x00,0x00,0x00,0x00,0x00};/* 8 -255 */
546 #endif
547
548
549 /* Definition to allow mutual recursion */
550
551 static BOOL
552 compile_regex(int, uschar **, const uschar **, int *, BOOL, BOOL, int, int,
553 int *, int *, branch_chain *, compile_data *, int *);
554
555
556
557 /*************************************************
558 * Find an error text *
559 *************************************************/
560
561 /* The error texts are now all in one long string, to save on relocations. As
562 some of the text is of unknown length, we can't use a table of offsets.
563 Instead, just count through the strings. This is not a performance issue
564 because it happens only when there has been a compilation error.
565
566 Argument: the error number
567 Returns: pointer to the error string
568 */
569
570 static const char *
571 find_error_text(int n)
572 {
573 const char *s = error_texts;
574 for (; n > 0; n--)
575 {
576 while (*s++ != 0) {};
577 if (*s == 0) return "Error text not found (please report)";
578 }
579 return s;
580 }
581
582
583 /*************************************************
584 * Check for counted repeat *
585 *************************************************/
586
587 /* This function is called when a '{' is encountered in a place where it might
588 start a quantifier. It looks ahead to see if it really is a quantifier or not.
589 It is only a quantifier if it is one of the forms {ddd} {ddd,} or {ddd,ddd}
590 where the ddds are digits.
591
592 Arguments:
593 p pointer to the first char after '{'
594
595 Returns: TRUE or FALSE
596 */
597
598 static BOOL
599 is_counted_repeat(const uschar *p)
600 {
601 if ((digitab[*p++] & ctype_digit) == 0) return FALSE;
602 while ((digitab[*p] & ctype_digit) != 0) p++;
603 if (*p == CHAR_RIGHT_CURLY_BRACKET) return TRUE;
604
605 if (*p++ != CHAR_COMMA) return FALSE;
606 if (*p == CHAR_RIGHT_CURLY_BRACKET) return TRUE;
607
608 if ((digitab[*p++] & ctype_digit) == 0) return FALSE;
609 while ((digitab[*p] & ctype_digit) != 0) p++;
610
611 return (*p == CHAR_RIGHT_CURLY_BRACKET);
612 }
613
614
615
616 /*************************************************
617 * Handle escapes *
618 *************************************************/
619
620 /* This function is called when a \ has been encountered. It either returns a
621 positive value for a simple escape such as \n, or a negative value which
622 encodes one of the more complicated things such as \d. A backreference to group
623 n is returned as -(ESC_REF + n); ESC_REF is the highest ESC_xxx macro. When
624 UTF-8 is enabled, a positive value greater than 255 may be returned. On entry,
625 ptr is pointing at the \. On exit, it is on the final character of the escape
626 sequence.
627
628 Arguments:
629 ptrptr points to the pattern position pointer
630 errorcodeptr points to the errorcode variable
631 bracount number of previous extracting brackets
632 options the options bits
633 isclass TRUE if inside a character class
634
635 Returns: zero or positive => a data character
636 negative => a special escape sequence
637 on error, errorcodeptr is set
638 */
639
640 static int
641 check_escape(const uschar **ptrptr, int *errorcodeptr, int bracount,
642 int options, BOOL isclass)
643 {
644 BOOL utf8 = (options & PCRE_UTF8) != 0;
645 const uschar *ptr = *ptrptr + 1;
646 int c, i;
647
648 GETCHARINCTEST(c, ptr); /* Get character value, increment pointer */
649 ptr--; /* Set pointer back to the last byte */
650
651 /* If backslash is at the end of the pattern, it's an error. */
652
653 if (c == 0) *errorcodeptr = ERR1;
654
655 /* Non-alphanumerics are literals. For digits or letters, do an initial lookup
656 in a table. A non-zero result is something that can be returned immediately.
657 Otherwise further processing may be required. */
658
659 #ifndef EBCDIC /* ASCII/UTF-8 coding */
660 else if (c < CHAR_0 || c > CHAR_z) {} /* Not alphanumeric */
661 else if ((i = escapes[c - CHAR_0]) != 0) c = i;
662
663 #else /* EBCDIC coding */
664 else if (c < 'a' || (ebcdic_chartab[c] & 0x0E) == 0) {} /* Not alphanumeric */
665 else if ((i = escapes[c - 0x48]) != 0) c = i;
666 #endif
667
668 /* Escapes that need further processing, or are illegal. */
669
670 else
671 {
672 const uschar *oldptr;
673 BOOL braced, negated;
674
675 switch (c)
676 {
677 /* A number of Perl escapes are not handled by PCRE. We give an explicit
678 error. */
679
680 case CHAR_l:
681 case CHAR_L:
682 *errorcodeptr = ERR37;
683 break;
684
685 case CHAR_u:
686 if ((options & PCRE_JAVASCRIPT_COMPAT) != 0)
687 {
688 /* In JavaScript, \u must be followed by four hexadecimal numbers.
689 Otherwise it is a lowercase u letter. */
690 if ((digitab[ptr[1]] & ctype_xdigit) != 0 && (digitab[ptr[2]] & ctype_xdigit) != 0
691 && (digitab[ptr[3]] & ctype_xdigit) != 0 && (digitab[ptr[4]] & ctype_xdigit) != 0)
692 {
693 c = 0;
694 for (i = 0; i < 4; ++i)
695 {
696 register int cc = *(++ptr);
697 #ifndef EBCDIC /* ASCII/UTF-8 coding */
698 if (cc >= CHAR_a) cc -= 32; /* Convert to upper case */
699 c = (c << 4) + cc - ((cc < CHAR_A)? CHAR_0 : (CHAR_A - 10));
700 #else /* EBCDIC coding */
701 if (cc >= CHAR_a && cc <= CHAR_z) cc += 64; /* Convert to upper case */
702 c = (c << 4) + cc - ((cc >= CHAR_0)? CHAR_0 : (CHAR_A - 10));
703 #endif
704 }
705 }
706 }
707 else
708 *errorcodeptr = ERR37;
709 break;
710
711 case CHAR_U:
712 /* In JavaScript, \U is an uppercase U letter. */
713 if ((options & PCRE_JAVASCRIPT_COMPAT) == 0) *errorcodeptr = ERR37;
714 break;
715
716 /* In a character class, \g is just a literal "g". Outside a character
717 class, \g must be followed by one of a number of specific things:
718
719 (1) A number, either plain or braced. If positive, it is an absolute
720 backreference. If negative, it is a relative backreference. This is a Perl
721 5.10 feature.
722
723 (2) Perl 5.10 also supports \g{name} as a reference to a named group. This
724 is part of Perl's movement towards a unified syntax for back references. As
725 this is synonymous with \k{name}, we fudge it up by pretending it really
726 was \k.
727
728 (3) For Oniguruma compatibility we also support \g followed by a name or a
729 number either in angle brackets or in single quotes. However, these are
730 (possibly recursive) subroutine calls, _not_ backreferences. Just return
731 the -ESC_g code (cf \k). */
732
733 case CHAR_g:
734 if (isclass) break;
735 if (ptr[1] == CHAR_LESS_THAN_SIGN || ptr[1] == CHAR_APOSTROPHE)
736 {
737 c = -ESC_g;
738 break;
739 }
740
741 /* Handle the Perl-compatible cases */
742
743 if (ptr[1] == CHAR_LEFT_CURLY_BRACKET)
744 {
745 const uschar *p;
746 for (p = ptr+2; *p != 0 && *p != CHAR_RIGHT_CURLY_BRACKET; p++)
747 if (*p != CHAR_MINUS && (digitab[*p] & ctype_digit) == 0) break;
748 if (*p != 0 && *p != CHAR_RIGHT_CURLY_BRACKET)
749 {
750 c = -ESC_k;
751 break;
752 }
753 braced = TRUE;
754 ptr++;
755 }
756 else braced = FALSE;
757
758 if (ptr[1] == CHAR_MINUS)
759 {
760 negated = TRUE;
761 ptr++;
762 }
763 else negated = FALSE;
764
765 c = 0;
766 while ((digitab[ptr[1]] & ctype_digit) != 0)
767 c = c * 10 + *(++ptr) - CHAR_0;
768
769 if (c < 0) /* Integer overflow */
770 {
771 *errorcodeptr = ERR61;
772 break;
773 }
774
775 if (braced && *(++ptr) != CHAR_RIGHT_CURLY_BRACKET)
776 {
777 *errorcodeptr = ERR57;
778 break;
779 }
780
781 if (c == 0)
782 {
783 *errorcodeptr = ERR58;
784 break;
785 }
786
787 if (negated)
788 {
789 if (c > bracount)
790 {
791 *errorcodeptr = ERR15;
792 break;
793 }
794 c = bracount - (c - 1);
795 }
796
797 c = -(ESC_REF + c);
798 break;
799
800 /* The handling of escape sequences consisting of a string of digits
801 starting with one that is not zero is not straightforward. By experiment,
802 the way Perl works seems to be as follows:
803
804 Outside a character class, the digits are read as a decimal number. If the
805 number is less than 10, or if there are that many previous extracting
806 left brackets, then it is a back reference. Otherwise, up to three octal
807 digits are read to form an escaped byte. Thus \123 is likely to be octal
808 123 (cf \0123, which is octal 012 followed by the literal 3). If the octal
809 value is greater than 377, the least significant 8 bits are taken. Inside a
810 character class, \ followed by a digit is always an octal number. */
811
812 case CHAR_1: case CHAR_2: case CHAR_3: case CHAR_4: case CHAR_5:
813 case CHAR_6: case CHAR_7: case CHAR_8: case CHAR_9:
814
815 if (!isclass)
816 {
817 oldptr = ptr;
818 c -= CHAR_0;
819 while ((digitab[ptr[1]] & ctype_digit) != 0)
820 c = c * 10 + *(++ptr) - CHAR_0;
821 if (c < 0) /* Integer overflow */
822 {
823 *errorcodeptr = ERR61;
824 break;
825 }
826 if (c < 10 || c <= bracount)
827 {
828 c = -(ESC_REF + c);
829 break;
830 }
831 ptr = oldptr; /* Put the pointer back and fall through */
832 }
833
834 /* Handle an octal number following \. If the first digit is 8 or 9, Perl
835 generates a binary zero byte and treats the digit as a following literal.
836 Thus we have to pull back the pointer by one. */
837
838 if ((c = *ptr) >= CHAR_8)
839 {
840 ptr--;
841 c = 0;
842 break;
843 }
844
845 /* \0 always starts an octal number, but we may drop through to here with a
846 larger first octal digit. The original code used just to take the least
847 significant 8 bits of octal numbers (I think this is what early Perls used
848 to do). Nowadays we allow for larger numbers in UTF-8 mode, but no more
849 than 3 octal digits. */
850
851 case CHAR_0:
852 c -= CHAR_0;
853 while(i++ < 2 && ptr[1] >= CHAR_0 && ptr[1] <= CHAR_7)
854 c = c * 8 + *(++ptr) - CHAR_0;
855 if (!utf8 && c > 255) *errorcodeptr = ERR51;
856 break;
857
858 /* \x is complicated. \x{ddd} is a character number which can be greater
859 than 0xff in utf8 mode, but only if the ddd are hex digits. If not, { is
860 treated as a data character. */
861
862 case CHAR_x:
863 if ((options & PCRE_JAVASCRIPT_COMPAT) != 0)
864 {
865 /* In JavaScript, \x must be followed by two hexadecimal numbers.
866 Otherwise it is a lowercase x letter. */
867 if ((digitab[ptr[1]] & ctype_xdigit) != 0 && (digitab[ptr[2]] & ctype_xdigit) != 0)
868 {
869 c = 0;
870 for (i = 0; i < 2; ++i)
871 {
872 register int cc = *(++ptr);
873 #ifndef EBCDIC /* ASCII/UTF-8 coding */
874 if (cc >= CHAR_a) cc -= 32; /* Convert to upper case */
875 c = (c << 4) + cc - ((cc < CHAR_A)? CHAR_0 : (CHAR_A - 10));
876 #else /* EBCDIC coding */
877 if (cc >= CHAR_a && cc <= CHAR_z) cc += 64; /* Convert to upper case */
878 c = (c << 4) + cc - ((cc >= CHAR_0)? CHAR_0 : (CHAR_A - 10));
879 #endif
880 }
881 }
882 break;
883 }
884
885 if (ptr[1] == CHAR_LEFT_CURLY_BRACKET)
886 {
887 const uschar *pt = ptr + 2;
888 int count = 0;
889
890 c = 0;
891 while ((digitab[*pt] & ctype_xdigit) != 0)
892 {
893 register int cc = *pt++;
894 if (c == 0 && cc == CHAR_0) continue; /* Leading zeroes */
895 count++;
896
897 #ifndef EBCDIC /* ASCII/UTF-8 coding */
898 if (cc >= CHAR_a) cc -= 32; /* Convert to upper case */
899 c = (c << 4) + cc - ((cc < CHAR_A)? CHAR_0 : (CHAR_A - 10));
900 #else /* EBCDIC coding */
901 if (cc >= CHAR_a && cc <= CHAR_z) cc += 64; /* Convert to upper case */
902 c = (c << 4) + cc - ((cc >= CHAR_0)? CHAR_0 : (CHAR_A - 10));
903 #endif
904 }
905
906 if (*pt == CHAR_RIGHT_CURLY_BRACKET)
907 {
908 if (c < 0 || count > (utf8? 8 : 2)) *errorcodeptr = ERR34;
909 ptr = pt;
910 break;
911 }
912
913 /* If the sequence of hex digits does not end with '}', then we don't
914 recognize this construct; fall through to the normal \x handling. */
915 }
916
917 /* Read just a single-byte hex-defined char */
918
919 c = 0;
920 while (i++ < 2 && (digitab[ptr[1]] & ctype_xdigit) != 0)
921 {
922 int cc; /* Some compilers don't like */
923 cc = *(++ptr); /* ++ in initializers */
924 #ifndef EBCDIC /* ASCII/UTF-8 coding */
925 if (cc >= CHAR_a) cc -= 32; /* Convert to upper case */
926 c = c * 16 + cc - ((cc < CHAR_A)? CHAR_0 : (CHAR_A - 10));
927 #else /* EBCDIC coding */
928 if (cc <= CHAR_z) cc += 64; /* Convert to upper case */
929 c = c * 16 + cc - ((cc >= CHAR_0)? CHAR_0 : (CHAR_A - 10));
930 #endif
931 }
932 break;
933
934 /* For \c, a following letter is upper-cased; then the 0x40 bit is flipped.
935 An error is given if the byte following \c is not an ASCII character. This
936 coding is ASCII-specific, but then the whole concept of \cx is
937 ASCII-specific. (However, an EBCDIC equivalent has now been added.) */
938
939 case CHAR_c:
940 c = *(++ptr);
941 if (c == 0)
942 {
943 *errorcodeptr = ERR2;
944 break;
945 }
946 #ifndef EBCDIC /* ASCII/UTF-8 coding */
947 if (c > 127) /* Excludes all non-ASCII in either mode */
948 {
949 *errorcodeptr = ERR68;
950 break;
951 }
952 if (c >= CHAR_a && c <= CHAR_z) c -= 32;
953 c ^= 0x40;
954 #else /* EBCDIC coding */
955 if (c >= CHAR_a && c <= CHAR_z) c += 64;
956 c ^= 0xC0;
957 #endif
958 break;
959
960 /* PCRE_EXTRA enables extensions to Perl in the matter of escapes. Any
961 other alphanumeric following \ is an error if PCRE_EXTRA was set;
962 otherwise, for Perl compatibility, it is a literal. This code looks a bit
963 odd, but there used to be some cases other than the default, and there may
964 be again in future, so I haven't "optimized" it. */
965
966 default:
967 if ((options & PCRE_EXTRA) != 0) switch(c)
968 {
969 default:
970 *errorcodeptr = ERR3;
971 break;
972 }
973 break;
974 }
975 }
976
977 /* Perl supports \N{name} for character names, as well as plain \N for "not
978 newline". PCRE does not support \N{name}. However, it does support
979 quantification such as \N{2,3}. */
980
981 if (c == -ESC_N && ptr[1] == CHAR_LEFT_CURLY_BRACKET &&
982 !is_counted_repeat(ptr+2))
983 *errorcodeptr = ERR37;
984
985 /* If PCRE_UCP is set, we change the values for \d etc. */
986
987 if ((options & PCRE_UCP) != 0 && c <= -ESC_D && c >= -ESC_w)
988 c -= (ESC_DU - ESC_D);
989
990 /* Set the pointer to the final character before returning. */
991
992 *ptrptr = ptr;
993 return c;
994 }
995
996
997
998 #ifdef SUPPORT_UCP
999 /*************************************************
1000 * Handle \P and \p *
1001 *************************************************/
1002
1003 /* This function is called after \P or \p has been encountered, provided that
1004 PCRE is compiled with support for Unicode properties. On entry, ptrptr is
1005 pointing at the P or p. On exit, it is pointing at the final character of the
1006 escape sequence.
1007
1008 Argument:
1009 ptrptr points to the pattern position pointer
1010 negptr points to a boolean that is set TRUE for negation else FALSE
1011 dptr points to an int that is set to the detailed property value
1012 errorcodeptr points to the error code variable
1013
1014 Returns: type value from ucp_type_table, or -1 for an invalid type
1015 */
1016
1017 static int
1018 get_ucp(const uschar **ptrptr, BOOL *negptr, int *dptr, int *errorcodeptr)
1019 {
1020 int c, i, bot, top;
1021 const uschar *ptr = *ptrptr;
1022 char name[32];
1023
1024 c = *(++ptr);
1025 if (c == 0) goto ERROR_RETURN;
1026
1027 *negptr = FALSE;
1028
1029 /* \P or \p can be followed by a name in {}, optionally preceded by ^ for
1030 negation. */
1031
1032 if (c == CHAR_LEFT_CURLY_BRACKET)
1033 {
1034 if (ptr[1] == CHAR_CIRCUMFLEX_ACCENT)
1035 {
1036 *negptr = TRUE;
1037 ptr++;
1038 }
1039 for (i = 0; i < (int)sizeof(name) - 1; i++)
1040 {
1041 c = *(++ptr);
1042 if (c == 0) goto ERROR_RETURN;
1043 if (c == CHAR_RIGHT_CURLY_BRACKET) break;
1044 name[i] = c;
1045 }
1046 if (c != CHAR_RIGHT_CURLY_BRACKET) goto ERROR_RETURN;
1047 name[i] = 0;
1048 }
1049
1050 /* Otherwise there is just one following character */
1051
1052 else
1053 {
1054 name[0] = c;
1055 name[1] = 0;
1056 }
1057
1058 *ptrptr = ptr;
1059
1060 /* Search for a recognized property name using binary chop */
1061
1062 bot = 0;
1063 top = _pcre_utt_size;
1064
1065 while (bot < top)
1066 {
1067 i = (bot + top) >> 1;
1068 c = strcmp(name, _pcre_utt_names + _pcre_utt[i].name_offset);
1069 if (c == 0)
1070 {
1071 *dptr = _pcre_utt[i].value;
1072 return _pcre_utt[i].type;
1073 }
1074 if (c > 0) bot = i + 1; else top = i;
1075 }
1076
1077 *errorcodeptr = ERR47;
1078 *ptrptr = ptr;
1079 return -1;
1080
1081 ERROR_RETURN:
1082 *errorcodeptr = ERR46;
1083 *ptrptr = ptr;
1084 return -1;
1085 }
1086 #endif
1087
1088
1089
1090
1091 /*************************************************
1092 * Read repeat counts *
1093 *************************************************/
1094
1095 /* Read an item of the form {n,m} and return the values. This is called only
1096 after is_counted_repeat() has confirmed that a repeat-count quantifier exists,
1097 so the syntax is guaranteed to be correct, but we need to check the values.
1098
1099 Arguments:
1100 p pointer to first char after '{'
1101 minp pointer to int for min
1102 maxp pointer to int for max
1103 returned as -1 if no max
1104 errorcodeptr points to error code variable
1105
1106 Returns: pointer to '}' on success;
1107 current ptr on error, with errorcodeptr set non-zero
1108 */
1109
1110 static const uschar *
1111 read_repeat_counts(const uschar *p, int *minp, int *maxp, int *errorcodeptr)
1112 {
1113 int min = 0;
1114 int max = -1;
1115
1116 /* Read the minimum value and do a paranoid check: a negative value indicates
1117 an integer overflow. */
1118
1119 while ((digitab[*p] & ctype_digit) != 0) min = min * 10 + *p++ - CHAR_0;
1120 if (min < 0 || min > 65535)
1121 {
1122 *errorcodeptr = ERR5;
1123 return p;
1124 }
1125
1126 /* Read the maximum value if there is one, and again do a paranoid on its size.
1127 Also, max must not be less than min. */
1128
1129 if (*p == CHAR_RIGHT_CURLY_BRACKET) max = min; else
1130 {
1131 if (*(++p) != CHAR_RIGHT_CURLY_BRACKET)
1132 {
1133 max = 0;
1134 while((digitab[*p] & ctype_digit) != 0) max = max * 10 + *p++ - CHAR_0;
1135 if (max < 0 || max > 65535)
1136 {
1137 *errorcodeptr = ERR5;
1138 return p;
1139 }
1140 if (max < min)
1141 {
1142 *errorcodeptr = ERR4;
1143 return p;
1144 }
1145 }
1146 }
1147
1148 /* Fill in the required variables, and pass back the pointer to the terminating
1149 '}'. */
1150
1151 *minp = min;
1152 *maxp = max;
1153 return p;
1154 }
1155
1156
1157
1158 /*************************************************
1159 * Subroutine for finding forward reference *
1160 *************************************************/
1161
1162 /* This recursive function is called only from find_parens() below. The
1163 top-level call starts at the beginning of the pattern. All other calls must
1164 start at a parenthesis. It scans along a pattern's text looking for capturing
1165 subpatterns, and counting them. If it finds a named pattern that matches the
1166 name it is given, it returns its number. Alternatively, if the name is NULL, it
1167 returns when it reaches a given numbered subpattern. Recursion is used to keep
1168 track of subpatterns that reset the capturing group numbers - the (?| feature.
1169
1170 This function was originally called only from the second pass, in which we know
1171 that if (?< or (?' or (?P< is encountered, the name will be correctly
1172 terminated because that is checked in the first pass. There is now one call to
1173 this function in the first pass, to check for a recursive back reference by
1174 name (so that we can make the whole group atomic). In this case, we need check
1175 only up to the current position in the pattern, and that is still OK because
1176 and previous occurrences will have been checked. To make this work, the test
1177 for "end of pattern" is a check against cd->end_pattern in the main loop,
1178 instead of looking for a binary zero. This means that the special first-pass
1179 call can adjust cd->end_pattern temporarily. (Checks for binary zero while
1180 processing items within the loop are OK, because afterwards the main loop will
1181 terminate.)
1182
1183 Arguments:
1184 ptrptr address of the current character pointer (updated)
1185 cd compile background data
1186 name name to seek, or NULL if seeking a numbered subpattern
1187 lorn name length, or subpattern number if name is NULL
1188 xmode TRUE if we are in /x mode
1189 utf8 TRUE if we are in UTF-8 mode
1190 count pointer to the current capturing subpattern number (updated)
1191
1192 Returns: the number of the named subpattern, or -1 if not found
1193 */
1194
1195 static int
1196 find_parens_sub(uschar **ptrptr, compile_data *cd, const uschar *name, int lorn,
1197 BOOL xmode, BOOL utf8, int *count)
1198 {
1199 uschar *ptr = *ptrptr;
1200 int start_count = *count;
1201 int hwm_count = start_count;
1202 BOOL dup_parens = FALSE;
1203
1204 /* If the first character is a parenthesis, check on the type of group we are
1205 dealing with. The very first call may not start with a parenthesis. */
1206
1207 if (ptr[0] == CHAR_LEFT_PARENTHESIS)
1208 {
1209 /* Handle specials such as (*SKIP) or (*UTF8) etc. */
1210
1211 if (ptr[1] == CHAR_ASTERISK) ptr += 2;
1212
1213 /* Handle a normal, unnamed capturing parenthesis. */
1214
1215 else if (ptr[1] != CHAR_QUESTION_MARK)
1216 {
1217 *count += 1;
1218 if (name == NULL && *count == lorn) return *count;
1219 ptr++;
1220 }
1221
1222 /* All cases now have (? at the start. Remember when we are in a group
1223 where the parenthesis numbers are duplicated. */
1224
1225 else if (ptr[2] == CHAR_VERTICAL_LINE)
1226 {
1227 ptr += 3;
1228 dup_parens = TRUE;
1229 }
1230
1231 /* Handle comments; all characters are allowed until a ket is reached. */
1232
1233 else if (ptr[2] == CHAR_NUMBER_SIGN)
1234 {
1235 for (ptr += 3; *ptr != 0; ptr++) if (*ptr == CHAR_RIGHT_PARENTHESIS) break;
1236 goto FAIL_EXIT;
1237 }
1238
1239 /* Handle a condition. If it is an assertion, just carry on so that it
1240 is processed as normal. If not, skip to the closing parenthesis of the
1241 condition (there can't be any nested parens). */
1242
1243 else if (ptr[2] == CHAR_LEFT_PARENTHESIS)
1244 {
1245 ptr += 2;
1246 if (ptr[1] != CHAR_QUESTION_MARK)
1247 {
1248 while (*ptr != 0 && *ptr != CHAR_RIGHT_PARENTHESIS) ptr++;
1249 if (*ptr != 0) ptr++;
1250 }
1251 }
1252
1253 /* Start with (? but not a condition. */
1254
1255 else
1256 {
1257 ptr += 2;
1258 if (*ptr == CHAR_P) ptr++; /* Allow optional P */
1259
1260 /* We have to disambiguate (?<! and (?<= from (?<name> for named groups */
1261
1262 if ((*ptr == CHAR_LESS_THAN_SIGN && ptr[1] != CHAR_EXCLAMATION_MARK &&
1263 ptr[1] != CHAR_EQUALS_SIGN) || *ptr == CHAR_APOSTROPHE)
1264 {
1265 int term;
1266 const uschar *thisname;
1267 *count += 1;
1268 if (name == NULL && *count == lorn) return *count;
1269 term = *ptr++;
1270 if (term == CHAR_LESS_THAN_SIGN) term = CHAR_GREATER_THAN_SIGN;
1271 thisname = ptr;
1272 while (*ptr != term) ptr++;
1273 if (name != NULL && lorn == ptr - thisname &&
1274 strncmp((const char *)name, (const char *)thisname, lorn) == 0)
1275 return *count;
1276 term++;
1277 }
1278 }
1279 }
1280
1281 /* Past any initial parenthesis handling, scan for parentheses or vertical
1282 bars. Stop if we get to cd->end_pattern. Note that this is important for the
1283 first-pass call when this value is temporarily adjusted to stop at the current
1284 position. So DO NOT change this to a test for binary zero. */
1285
1286 for (; ptr < cd->end_pattern; ptr++)
1287 {
1288 /* Skip over backslashed characters and also entire \Q...\E */
1289
1290 if (*ptr == CHAR_BACKSLASH)
1291 {
1292 if (*(++ptr) == 0) goto FAIL_EXIT;
1293 if (*ptr == CHAR_Q) for (;;)
1294 {
1295 while (*(++ptr) != 0 && *ptr != CHAR_BACKSLASH) {};
1296 if (*ptr == 0) goto FAIL_EXIT;
1297 if (*(++ptr) == CHAR_E) break;
1298 }
1299 continue;
1300 }
1301
1302 /* Skip over character classes; this logic must be similar to the way they
1303 are handled for real. If the first character is '^', skip it. Also, if the
1304 first few characters (either before or after ^) are \Q\E or \E we skip them
1305 too. This makes for compatibility with Perl. Note the use of STR macros to
1306 encode "Q\\E" so that it works in UTF-8 on EBCDIC platforms. */
1307
1308 if (*ptr == CHAR_LEFT_SQUARE_BRACKET)
1309 {
1310 BOOL negate_class = FALSE;
1311 for (;;)
1312 {
1313 if (ptr[1] == CHAR_BACKSLASH)
1314 {
1315 if (ptr[2] == CHAR_E)
1316 ptr+= 2;
1317 else if (strncmp((const char *)ptr+2,
1318 STR_Q STR_BACKSLASH STR_E, 3) == 0)
1319 ptr += 4;
1320 else
1321 break;
1322 }
1323 else if (!negate_class && ptr[1] == CHAR_CIRCUMFLEX_ACCENT)
1324 {
1325 negate_class = TRUE;
1326 ptr++;
1327 }
1328 else break;
1329 }
1330
1331 /* If the next character is ']', it is a data character that must be
1332 skipped, except in JavaScript compatibility mode. */
1333
1334 if (ptr[1] == CHAR_RIGHT_SQUARE_BRACKET &&
1335 (cd->external_options & PCRE_JAVASCRIPT_COMPAT) == 0)
1336 ptr++;
1337
1338 while (*(++ptr) != CHAR_RIGHT_SQUARE_BRACKET)
1339 {
1340 if (*ptr == 0) return -1;
1341 if (*ptr == CHAR_BACKSLASH)
1342 {
1343 if (*(++ptr) == 0) goto FAIL_EXIT;
1344 if (*ptr == CHAR_Q) for (;;)
1345 {
1346 while (*(++ptr) != 0 && *ptr != CHAR_BACKSLASH) {};
1347 if (*ptr == 0) goto FAIL_EXIT;
1348 if (*(++ptr) == CHAR_E) break;
1349 }
1350 continue;
1351 }
1352 }
1353 continue;
1354 }
1355
1356 /* Skip comments in /x mode */
1357
1358 if (xmode && *ptr == CHAR_NUMBER_SIGN)
1359 {
1360 ptr++;
1361 while (*ptr != 0)
1362 {
1363 if (IS_NEWLINE(ptr)) { ptr += cd->nllen - 1; break; }
1364 ptr++;
1365 #ifdef SUPPORT_UTF8
1366 if (utf8) while ((*ptr & 0xc0) == 0x80) ptr++;
1367 #endif
1368 }
1369 if (*ptr == 0) goto FAIL_EXIT;
1370 continue;
1371 }
1372
1373 /* Check for the special metacharacters */
1374
1375 if (*ptr == CHAR_LEFT_PARENTHESIS)
1376 {
1377 int rc = find_parens_sub(&ptr, cd, name, lorn, xmode, utf8, count);
1378 if (rc > 0) return rc;
1379 if (*ptr == 0) goto FAIL_EXIT;
1380 }
1381
1382 else if (*ptr == CHAR_RIGHT_PARENTHESIS)
1383 {
1384 if (dup_parens && *count < hwm_count) *count = hwm_count;
1385 goto FAIL_EXIT;
1386 }
1387
1388 else if (*ptr == CHAR_VERTICAL_LINE && dup_parens)
1389 {
1390 if (*count > hwm_count) hwm_count = *count;
1391 *count = start_count;
1392 }
1393 }
1394
1395 FAIL_EXIT:
1396 *ptrptr = ptr;
1397 return -1;
1398 }
1399
1400
1401
1402
1403 /*************************************************
1404 * Find forward referenced subpattern *
1405 *************************************************/
1406
1407 /* This function scans along a pattern's text looking for capturing
1408 subpatterns, and counting them. If it finds a named pattern that matches the
1409 name it is given, it returns its number. Alternatively, if the name is NULL, it
1410 returns when it reaches a given numbered subpattern. This is used for forward
1411 references to subpatterns. We used to be able to start this scan from the
1412 current compiling point, using the current count value from cd->bracount, and
1413 do it all in a single loop, but the addition of the possibility of duplicate
1414 subpattern numbers means that we have to scan from the very start, in order to
1415 take account of such duplicates, and to use a recursive function to keep track
1416 of the different types of group.
1417
1418 Arguments:
1419 cd compile background data
1420 name name to seek, or NULL if seeking a numbered subpattern
1421 lorn name length, or subpattern number if name is NULL
1422 xmode TRUE if we are in /x mode
1423 utf8 TRUE if we are in UTF-8 mode
1424
1425 Returns: the number of the found subpattern, or -1 if not found
1426 */
1427
1428 static int
1429 find_parens(compile_data *cd, const uschar *name, int lorn, BOOL xmode,
1430 BOOL utf8)
1431 {
1432 uschar *ptr = (uschar *)cd->start_pattern;
1433 int count = 0;
1434 int rc;
1435
1436 /* If the pattern does not start with an opening parenthesis, the first call
1437 to find_parens_sub() will scan right to the end (if necessary). However, if it
1438 does start with a parenthesis, find_parens_sub() will return when it hits the
1439 matching closing parens. That is why we have to have a loop. */
1440
1441 for (;;)
1442 {
1443 rc = find_parens_sub(&ptr, cd, name, lorn, xmode, utf8, &count);
1444 if (rc > 0 || *ptr++ == 0) break;
1445 }
1446
1447 return rc;
1448 }
1449
1450
1451
1452
1453 /*************************************************
1454 * Find first significant op code *
1455 *************************************************/
1456
1457 /* This is called by several functions that scan a compiled expression looking
1458 for a fixed first character, or an anchoring op code etc. It skips over things
1459 that do not influence this. For some calls, it makes sense to skip negative
1460 forward and all backward assertions, and also the \b assertion; for others it
1461 does not.
1462
1463 Arguments:
1464 code pointer to the start of the group
1465 skipassert TRUE if certain assertions are to be skipped
1466
1467 Returns: pointer to the first significant opcode
1468 */
1469
1470 static const uschar*
1471 first_significant_code(const uschar *code, BOOL skipassert)
1472 {
1473 for (;;)
1474 {
1475 switch ((int)*code)
1476 {
1477 case OP_ASSERT_NOT:
1478 case OP_ASSERTBACK:
1479 case OP_ASSERTBACK_NOT:
1480 if (!skipassert) return code;
1481 do code += GET(code, 1); while (*code == OP_ALT);
1482 code += _pcre_OP_lengths[*code];
1483 break;
1484
1485 case OP_WORD_BOUNDARY:
1486 case OP_NOT_WORD_BOUNDARY:
1487 if (!skipassert) return code;
1488 /* Fall through */
1489
1490 case OP_CALLOUT:
1491 case OP_CREF:
1492 case OP_NCREF:
1493 case OP_RREF:
1494 case OP_NRREF:
1495 case OP_DEF:
1496 code += _pcre_OP_lengths[*code];
1497 break;
1498
1499 default:
1500 return code;
1501 }
1502 }
1503 /* Control never reaches here */
1504 }
1505
1506
1507
1508
1509 /*************************************************
1510 * Find the fixed length of a branch *
1511 *************************************************/
1512
1513 /* Scan a branch and compute the fixed length of subject that will match it,
1514 if the length is fixed. This is needed for dealing with backward assertions.
1515 In UTF8 mode, the result is in characters rather than bytes. The branch is
1516 temporarily terminated with OP_END when this function is called.
1517
1518 This function is called when a backward assertion is encountered, so that if it
1519 fails, the error message can point to the correct place in the pattern.
1520 However, we cannot do this when the assertion contains subroutine calls,
1521 because they can be forward references. We solve this by remembering this case
1522 and doing the check at the end; a flag specifies which mode we are running in.
1523
1524 Arguments:
1525 code points to the start of the pattern (the bracket)
1526 utf8 TRUE in UTF-8 mode
1527 atend TRUE if called when the pattern is complete
1528 cd the "compile data" structure
1529
1530 Returns: the fixed length,
1531 or -1 if there is no fixed length,
1532 or -2 if \C was encountered (in UTF-8 mode only)
1533 or -3 if an OP_RECURSE item was encountered and atend is FALSE
1534 or -4 if an unknown opcode was encountered (internal error)
1535 */
1536
1537 static int
1538 find_fixedlength(uschar *code, BOOL utf8, BOOL atend, compile_data *cd)
1539 {
1540 int length = -1;
1541
1542 register int branchlength = 0;
1543 register uschar *cc = code + 1 + LINK_SIZE;
1544
1545 /* Scan along the opcodes for this branch. If we get to the end of the
1546 branch, check the length against that of the other branches. */
1547
1548 for (;;)
1549 {
1550 int d;
1551 uschar *ce, *cs;
1552 register int op = *cc;
1553 switch (op)
1554 {
1555 /* We only need to continue for OP_CBRA (normal capturing bracket) and
1556 OP_BRA (normal non-capturing bracket) because the other variants of these
1557 opcodes are all concerned with unlimited repeated groups, which of course
1558 are not of fixed length. */
1559
1560 case OP_CBRA:
1561 case OP_BRA:
1562 case OP_ONCE:
1563 case OP_ONCE_NC:
1564 case OP_COND:
1565 d = find_fixedlength(cc + ((op == OP_CBRA)? 2:0), utf8, atend, cd);
1566 if (d < 0) return d;
1567 branchlength += d;
1568 do cc += GET(cc, 1); while (*cc == OP_ALT);
1569 cc += 1 + LINK_SIZE;
1570 break;
1571
1572 /* Reached end of a branch; if it's a ket it is the end of a nested call.
1573 If it's ALT it is an alternation in a nested call. An ACCEPT is effectively
1574 an ALT. If it is END it's the end of the outer call. All can be handled by
1575 the same code. Note that we must not include the OP_KETRxxx opcodes here,
1576 because they all imply an unlimited repeat. */
1577
1578 case OP_ALT:
1579 case OP_KET:
1580 case OP_END:
1581 case OP_ACCEPT:
1582 case OP_ASSERT_ACCEPT:
1583 if (length < 0) length = branchlength;
1584 else if (length != branchlength) return -1;
1585 if (*cc != OP_ALT) return length;
1586 cc += 1 + LINK_SIZE;
1587 branchlength = 0;
1588 break;
1589
1590 /* A true recursion implies not fixed length, but a subroutine call may
1591 be OK. If the subroutine is a forward reference, we can't deal with
1592 it until the end of the pattern, so return -3. */
1593
1594 case OP_RECURSE:
1595 if (!atend) return -3;
1596 cs = ce = (uschar *)cd->start_code + GET(cc, 1); /* Start subpattern */
1597 do ce += GET(ce, 1); while (*ce == OP_ALT); /* End subpattern */
1598 if (cc > cs && cc < ce) return -1; /* Recursion */
1599 d = find_fixedlength(cs + 2, utf8, atend, cd);
1600 if (d < 0) return d;
1601 branchlength += d;
1602 cc += 1 + LINK_SIZE;
1603 break;
1604
1605 /* Skip over assertive subpatterns */
1606
1607 case OP_ASSERT:
1608 case OP_ASSERT_NOT:
1609 case OP_ASSERTBACK:
1610 case OP_ASSERTBACK_NOT:
1611 do cc += GET(cc, 1); while (*cc == OP_ALT);
1612 /* Fall through */
1613
1614 /* Skip over things that don't match chars */
1615
1616 case OP_MARK:
1617 case OP_PRUNE_ARG:
1618 case OP_SKIP_ARG:
1619 case OP_THEN_ARG:
1620 cc += cc[1] + _pcre_OP_lengths[*cc];
1621 break;
1622
1623 case OP_CALLOUT:
1624 case OP_CIRC:
1625 case OP_CIRCM:
1626 case OP_CLOSE:
1627 case OP_COMMIT:
1628 case OP_CREF:
1629 case OP_DEF:
1630 case OP_DOLL:
1631 case OP_DOLLM:
1632 case OP_EOD:
1633 case OP_EODN:
1634 case OP_FAIL:
1635 case OP_NCREF:
1636 case OP_NRREF:
1637 case OP_NOT_WORD_BOUNDARY:
1638 case OP_PRUNE:
1639 case OP_REVERSE:
1640 case OP_RREF:
1641 case OP_SET_SOM:
1642 case OP_SKIP:
1643 case OP_SOD:
1644 case OP_SOM:
1645 case OP_THEN:
1646 case OP_WORD_BOUNDARY:
1647 cc += _pcre_OP_lengths[*cc];
1648 break;
1649
1650 /* Handle literal characters */
1651
1652 case OP_CHAR:
1653 case OP_CHARI:
1654 case OP_NOT:
1655 case OP_NOTI:
1656 branchlength++;
1657 cc += 2;
1658 #ifdef SUPPORT_UTF8
1659 if (utf8 && cc[-1] >= 0xc0) cc += _pcre_utf8_table4[cc[-1] & 0x3f];
1660 #endif
1661 break;
1662
1663 /* Handle exact repetitions. The count is already in characters, but we
1664 need to skip over a multibyte character in UTF8 mode. */
1665
1666 case OP_EXACT:
1667 case OP_EXACTI:
1668 case OP_NOTEXACT:
1669 case OP_NOTEXACTI:
1670 branchlength += GET2(cc,1);
1671 cc += 4;
1672 #ifdef SUPPORT_UTF8
1673 if (utf8 && cc[-1] >= 0xc0) cc += _pcre_utf8_table4[cc[-1] & 0x3f];
1674 #endif
1675 break;
1676
1677 case OP_TYPEEXACT:
1678 branchlength += GET2(cc,1);
1679 if (cc[3] == OP_PROP || cc[3] == OP_NOTPROP) cc += 2;
1680 cc += 4;
1681 break;
1682
1683 /* Handle single-char matchers */
1684
1685 case OP_PROP:
1686 case OP_NOTPROP:
1687 cc += 2;
1688 /* Fall through */
1689
1690 case OP_HSPACE:
1691 case OP_VSPACE:
1692 case OP_NOT_HSPACE:
1693 case OP_NOT_VSPACE:
1694 case OP_NOT_DIGIT:
1695 case OP_DIGIT:
1696 case OP_NOT_WHITESPACE:
1697 case OP_WHITESPACE:
1698 case OP_NOT_WORDCHAR:
1699 case OP_WORDCHAR:
1700 case OP_ANY:
1701 case OP_ALLANY:
1702 branchlength++;
1703 cc++;
1704 break;
1705
1706 /* The single-byte matcher isn't allowed. This only happens in UTF-8 mode;
1707 otherwise \C is coded as OP_ALLANY. */
1708
1709 case OP_ANYBYTE:
1710 return -2;
1711
1712 /* Check a class for variable quantification */
1713
1714 #ifdef SUPPORT_UTF8
1715 case OP_XCLASS:
1716 cc += GET(cc, 1) - 33;
1717 /* Fall through */
1718 #endif
1719
1720 case OP_CLASS:
1721 case OP_NCLASS:
1722 cc += 33;
1723
1724 switch (*cc)
1725 {
1726 case OP_CRPLUS:
1727 case OP_CRMINPLUS:
1728 case OP_CRSTAR:
1729 case OP_CRMINSTAR:
1730 case OP_CRQUERY:
1731 case OP_CRMINQUERY:
1732 return -1;
1733
1734 case OP_CRRANGE:
1735 case OP_CRMINRANGE:
1736 if (GET2(cc,1) != GET2(cc,3)) return -1;
1737 branchlength += GET2(cc,1);
1738 cc += 5;
1739 break;
1740
1741 default:
1742 branchlength++;
1743 }
1744 break;
1745
1746 /* Anything else is variable length */
1747
1748 case OP_ANYNL:
1749 case OP_BRAMINZERO:
1750 case OP_BRAPOS:
1751 case OP_BRAPOSZERO:
1752 case OP_BRAZERO:
1753 case OP_CBRAPOS:
1754 case OP_EXTUNI:
1755 case OP_KETRMAX:
1756 case OP_KETRMIN:
1757 case OP_KETRPOS:
1758 case OP_MINPLUS:
1759 case OP_MINPLUSI:
1760 case OP_MINQUERY:
1761 case OP_MINQUERYI:
1762 case OP_MINSTAR:
1763 case OP_MINSTARI:
1764 case OP_MINUPTO:
1765 case OP_MINUPTOI:
1766 case OP_NOTMINPLUS:
1767 case OP_NOTMINPLUSI:
1768 case OP_NOTMINQUERY:
1769 case OP_NOTMINQUERYI:
1770 case OP_NOTMINSTAR:
1771 case OP_NOTMINSTARI:
1772 case OP_NOTMINUPTO:
1773 case OP_NOTMINUPTOI:
1774 case OP_NOTPLUS:
1775 case OP_NOTPLUSI:
1776 case OP_NOTPOSPLUS:
1777 case OP_NOTPOSPLUSI:
1778 case OP_NOTPOSQUERY:
1779 case OP_NOTPOSQUERYI:
1780 case OP_NOTPOSSTAR:
1781 case OP_NOTPOSSTARI:
1782 case OP_NOTPOSUPTO:
1783 case OP_NOTPOSUPTOI:
1784 case OP_NOTQUERY:
1785 case OP_NOTQUERYI:
1786 case OP_NOTSTAR:
1787 case OP_NOTSTARI:
1788 case OP_NOTUPTO:
1789 case OP_NOTUPTOI:
1790 case OP_PLUS:
1791 case OP_PLUSI:
1792 case OP_POSPLUS:
1793 case OP_POSPLUSI:
1794 case OP_POSQUERY:
1795 case OP_POSQUERYI:
1796 case OP_POSSTAR:
1797 case OP_POSSTARI:
1798 case OP_POSUPTO:
1799 case OP_POSUPTOI:
1800 case OP_QUERY:
1801 case OP_QUERYI:
1802 case OP_REF:
1803 case OP_REFI:
1804 case OP_SBRA:
1805 case OP_SBRAPOS:
1806 case OP_SCBRA:
1807 case OP_SCBRAPOS:
1808 case OP_SCOND:
1809 case OP_SKIPZERO:
1810 case OP_STAR:
1811 case OP_STARI:
1812 case OP_TYPEMINPLUS:
1813 case OP_TYPEMINQUERY:
1814 case OP_TYPEMINSTAR:
1815 case OP_TYPEMINUPTO:
1816 case OP_TYPEPLUS:
1817 case OP_TYPEPOSPLUS:
1818 case OP_TYPEPOSQUERY:
1819 case OP_TYPEPOSSTAR:
1820 case OP_TYPEPOSUPTO:
1821 case OP_TYPEQUERY:
1822 case OP_TYPESTAR:
1823 case OP_TYPEUPTO:
1824 case OP_UPTO:
1825 case OP_UPTOI:
1826 return -1;
1827
1828 /* Catch unrecognized opcodes so that when new ones are added they
1829 are not forgotten, as has happened in the past. */
1830
1831 default:
1832 return -4;
1833 }
1834 }
1835 /* Control never gets here */
1836 }
1837
1838
1839
1840
1841 /*************************************************
1842 * Scan compiled regex for specific bracket *
1843 *************************************************/
1844
1845 /* This little function scans through a compiled pattern until it finds a
1846 capturing bracket with the given number, or, if the number is negative, an
1847 instance of OP_REVERSE for a lookbehind. The function is global in the C sense
1848 so that it can be called from pcre_study() when finding the minimum matching
1849 length.
1850
1851 Arguments:
1852 code points to start of expression
1853 utf8 TRUE in UTF-8 mode
1854 number the required bracket number or negative to find a lookbehind
1855
1856 Returns: pointer to the opcode for the bracket, or NULL if not found
1857 */
1858
1859 const uschar *
1860 _pcre_find_bracket(const uschar *code, BOOL utf8, int number)
1861 {
1862 for (;;)
1863 {
1864 register int c = *code;
1865
1866 if (c == OP_END) return NULL;
1867
1868 /* XCLASS is used for classes that cannot be represented just by a bit
1869 map. This includes negated single high-valued characters. The length in
1870 the table is zero; the actual length is stored in the compiled code. */
1871
1872 if (c == OP_XCLASS) code += GET(code, 1);
1873
1874 /* Handle recursion */
1875
1876 else if (c == OP_REVERSE)
1877 {
1878 if (number < 0) return (uschar *)code;
1879 code += _pcre_OP_lengths[c];
1880 }
1881
1882 /* Handle capturing bracket */
1883
1884 else if (c == OP_CBRA || c == OP_SCBRA ||
1885 c == OP_CBRAPOS || c == OP_SCBRAPOS)
1886 {
1887 int n = GET2(code, 1+LINK_SIZE);
1888 if (n == number) return (uschar *)code;
1889 code += _pcre_OP_lengths[c];
1890 }
1891
1892 /* Otherwise, we can get the item's length from the table, except that for
1893 repeated character types, we have to test for \p and \P, which have an extra
1894 two bytes of parameters, and for MARK/PRUNE/SKIP/THEN with an argument, we
1895 must add in its length. */
1896
1897 else
1898 {
1899 switch(c)
1900 {
1901 case OP_TYPESTAR:
1902 case OP_TYPEMINSTAR:
1903 case OP_TYPEPLUS:
1904 case OP_TYPEMINPLUS:
1905 case OP_TYPEQUERY:
1906 case OP_TYPEMINQUERY:
1907 case OP_TYPEPOSSTAR:
1908 case OP_TYPEPOSPLUS:
1909 case OP_TYPEPOSQUERY:
1910 if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
1911 break;
1912
1913 case OP_TYPEUPTO:
1914 case OP_TYPEMINUPTO:
1915 case OP_TYPEEXACT:
1916 case OP_TYPEPOSUPTO:
1917 if (code[3] == OP_PROP || code[3] == OP_NOTPROP) code += 2;
1918 break;
1919
1920 case OP_MARK:
1921 case OP_PRUNE_ARG:
1922 case OP_SKIP_ARG:
1923 code += code[1];
1924 break;
1925
1926 case OP_THEN_ARG:
1927 code += code[1];
1928 break;
1929 }
1930
1931 /* Add in the fixed length from the table */
1932
1933 code += _pcre_OP_lengths[c];
1934
1935 /* In UTF-8 mode, opcodes that are followed by a character may be followed by
1936 a multi-byte character. The length in the table is a minimum, so we have to
1937 arrange to skip the extra bytes. */
1938
1939 #ifdef SUPPORT_UTF8
1940 if (utf8) switch(c)
1941 {
1942 case OP_CHAR:
1943 case OP_CHARI:
1944 case OP_EXACT:
1945 case OP_EXACTI:
1946 case OP_UPTO:
1947 case OP_UPTOI:
1948 case OP_MINUPTO:
1949 case OP_MINUPTOI:
1950 case OP_POSUPTO:
1951 case OP_POSUPTOI:
1952 case OP_STAR:
1953 case OP_STARI:
1954 case OP_MINSTAR:
1955 case OP_MINSTARI:
1956 case OP_POSSTAR:
1957 case OP_POSSTARI:
1958 case OP_PLUS:
1959 case OP_PLUSI:
1960 case OP_MINPLUS:
1961 case OP_MINPLUSI:
1962 case OP_POSPLUS:
1963 case OP_POSPLUSI:
1964 case OP_QUERY:
1965 case OP_QUERYI:
1966 case OP_MINQUERY:
1967 case OP_MINQUERYI:
1968 case OP_POSQUERY:
1969 case OP_POSQUERYI:
1970 if (code[-1] >= 0xc0) code += _pcre_utf8_table4[code[-1] & 0x3f];
1971 break;
1972 }
1973 #else
1974 (void)(utf8); /* Keep compiler happy by referencing function argument */
1975 #endif
1976 }
1977 }
1978 }
1979
1980
1981
1982 /*************************************************
1983 * Scan compiled regex for recursion reference *
1984 *************************************************/
1985
1986 /* This little function scans through a compiled pattern until it finds an
1987 instance of OP_RECURSE.
1988
1989 Arguments:
1990 code points to start of expression
1991 utf8 TRUE in UTF-8 mode
1992
1993 Returns: pointer to the opcode for OP_RECURSE, or NULL if not found
1994 */
1995
1996 static const uschar *
1997 find_recurse(const uschar *code, BOOL utf8)
1998 {
1999 for (;;)
2000 {
2001 register int c = *code;
2002 if (c == OP_END) return NULL;
2003 if (c == OP_RECURSE) return code;
2004
2005 /* XCLASS is used for classes that cannot be represented just by a bit
2006 map. This includes negated single high-valued characters. The length in
2007 the table is zero; the actual length is stored in the compiled code. */
2008
2009 if (c == OP_XCLASS) code += GET(code, 1);
2010
2011 /* Otherwise, we can get the item's length from the table, except that for
2012 repeated character types, we have to test for \p and \P, which have an extra
2013 two bytes of parameters, and for MARK/PRUNE/SKIP/THEN with an argument, we
2014 must add in its length. */
2015
2016 else
2017 {
2018 switch(c)
2019 {
2020 case OP_TYPESTAR:
2021 case OP_TYPEMINSTAR:
2022 case OP_TYPEPLUS:
2023 case OP_TYPEMINPLUS:
2024 case OP_TYPEQUERY:
2025 case OP_TYPEMINQUERY:
2026 case OP_TYPEPOSSTAR:
2027 case OP_TYPEPOSPLUS:
2028 case OP_TYPEPOSQUERY:
2029 if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
2030 break;
2031
2032 case OP_TYPEPOSUPTO:
2033 case OP_TYPEUPTO:
2034 case OP_TYPEMINUPTO:
2035 case OP_TYPEEXACT:
2036 if (code[3] == OP_PROP || code[3] == OP_NOTPROP) code += 2;
2037 break;
2038
2039 case OP_MARK:
2040 case OP_PRUNE_ARG:
2041 case OP_SKIP_ARG:
2042 code += code[1];
2043 break;
2044
2045 case OP_THEN_ARG:
2046 code += code[1];
2047 break;
2048 }
2049
2050 /* Add in the fixed length from the table */
2051
2052 code += _pcre_OP_lengths[c];
2053
2054 /* In UTF-8 mode, opcodes that are followed by a character may be followed
2055 by a multi-byte character. The length in the table is a minimum, so we have
2056 to arrange to skip the extra bytes. */
2057
2058 #ifdef SUPPORT_UTF8
2059 if (utf8) switch(c)
2060 {
2061 case OP_CHAR:
2062 case OP_CHARI:
2063 case OP_EXACT:
2064 case OP_EXACTI:
2065 case OP_UPTO:
2066 case OP_UPTOI:
2067 case OP_MINUPTO:
2068 case OP_MINUPTOI:
2069 case OP_POSUPTO:
2070 case OP_POSUPTOI:
2071 case OP_STAR:
2072 case OP_STARI:
2073 case OP_MINSTAR:
2074 case OP_MINSTARI:
2075 case OP_POSSTAR:
2076 case OP_POSSTARI:
2077 case OP_PLUS:
2078 case OP_PLUSI:
2079 case OP_MINPLUS:
2080 case OP_MINPLUSI:
2081 case OP_POSPLUS:
2082 case OP_POSPLUSI:
2083 case OP_QUERY:
2084 case OP_QUERYI:
2085 case OP_MINQUERY:
2086 case OP_MINQUERYI:
2087 case OP_POSQUERY:
2088 case OP_POSQUERYI:
2089 if (code[-1] >= 0xc0) code += _pcre_utf8_table4[code[-1] & 0x3f];
2090 break;
2091 }
2092 #else
2093 (void)(utf8); /* Keep compiler happy by referencing function argument */
2094 #endif
2095 }
2096 }
2097 }
2098
2099
2100
2101 /*************************************************
2102 * Scan compiled branch for non-emptiness *
2103 *************************************************/
2104
2105 /* This function scans through a branch of a compiled pattern to see whether it
2106 can match the empty string or not. It is called from could_be_empty()
2107 below and from compile_branch() when checking for an unlimited repeat of a
2108 group that can match nothing. Note that first_significant_code() skips over
2109 backward and negative forward assertions when its final argument is TRUE. If we
2110 hit an unclosed bracket, we return "empty" - this means we've struck an inner
2111 bracket whose current branch will already have been scanned.
2112
2113 Arguments:
2114 code points to start of search
2115 endcode points to where to stop
2116 utf8 TRUE if in UTF8 mode
2117 cd contains pointers to tables etc.
2118
2119 Returns: TRUE if what is matched could be empty
2120 */
2121
2122 static BOOL
2123 could_be_empty_branch(const uschar *code, const uschar *endcode, BOOL utf8,
2124 compile_data *cd)
2125 {
2126 register int c;
2127 for (code = first_significant_code(code + _pcre_OP_lengths[*code], TRUE);
2128 code < endcode;
2129 code = first_significant_code(code + _pcre_OP_lengths[c], TRUE))
2130 {
2131 const uschar *ccode;
2132
2133 c = *code;
2134
2135 /* Skip over forward assertions; the other assertions are skipped by
2136 first_significant_code() with a TRUE final argument. */
2137
2138 if (c == OP_ASSERT)
2139 {
2140 do code += GET(code, 1); while (*code == OP_ALT);
2141 c = *code;
2142 continue;
2143 }
2144
2145 /* For a recursion/subroutine call, if its end has been reached, which
2146 implies a backward reference subroutine call, we can scan it. If it's a
2147 forward reference subroutine call, we can't. To detect forward reference
2148 we have to scan up the list that is kept in the workspace. This function is
2149 called only when doing the real compile, not during the pre-compile that
2150 measures the size of the compiled pattern. */
2151
2152 if (c == OP_RECURSE)
2153 {
2154 const uschar *scode;
2155 BOOL empty_branch;
2156
2157 /* Test for forward reference */
2158
2159 for (scode = cd->start_workspace; scode < cd->hwm; scode += LINK_SIZE)
2160 if (GET(scode, 0) == code + 1 - cd->start_code) return TRUE;
2161
2162 /* Not a forward reference, test for completed backward reference */
2163
2164 empty_branch = FALSE;
2165 scode = cd->start_code + GET(code, 1);
2166 if (GET(scode, 1) == 0) return TRUE; /* Unclosed */
2167
2168 /* Completed backwards reference */
2169
2170 do
2171 {
2172 if (could_be_empty_branch(scode, endcode, utf8, cd))
2173 {
2174 empty_branch = TRUE;
2175 break;
2176 }
2177 scode += GET(scode, 1);
2178 }
2179 while (*scode == OP_ALT);
2180
2181 if (!empty_branch) return FALSE; /* All branches are non-empty */
2182 continue;
2183 }
2184
2185 /* Groups with zero repeats can of course be empty; skip them. */
2186
2187 if (c == OP_BRAZERO || c == OP_BRAMINZERO || c == OP_SKIPZERO ||
2188 c == OP_BRAPOSZERO)
2189 {
2190 code += _pcre_OP_lengths[c];
2191 do code += GET(code, 1); while (*code == OP_ALT);
2192 c = *code;
2193 continue;
2194 }
2195
2196 /* A nested group that is already marked as "could be empty" can just be
2197 skipped. */
2198
2199 if (c == OP_SBRA || c == OP_SBRAPOS ||
2200 c == OP_SCBRA || c == OP_SCBRAPOS)
2201 {
2202 do code += GET(code, 1); while (*code == OP_ALT);
2203 c = *code;
2204 continue;
2205 }
2206
2207 /* For other groups, scan the branches. */
2208
2209 if (c == OP_BRA || c == OP_BRAPOS ||
2210 c == OP_CBRA || c == OP_CBRAPOS ||
2211 c == OP_ONCE || c == OP_ONCE_NC ||
2212 c == OP_COND)
2213 {
2214 BOOL empty_branch;
2215 if (GET(code, 1) == 0) return TRUE; /* Hit unclosed bracket */
2216
2217 /* If a conditional group has only one branch, there is a second, implied,
2218 empty branch, so just skip over the conditional, because it could be empty.
2219 Otherwise, scan the individual branches of the group. */
2220
2221 if (c == OP_COND && code[GET(code, 1)] != OP_ALT)
2222 code += GET(code, 1);
2223 else
2224 {
2225 empty_branch = FALSE;
2226 do
2227 {
2228 if (!empty_branch && could_be_empty_branch(code, endcode, utf8, cd))
2229 empty_branch = TRUE;
2230 code += GET(code, 1);
2231 }
2232 while (*code == OP_ALT);
2233 if (!empty_branch) return FALSE; /* All branches are non-empty */
2234 }
2235
2236 c = *code;
2237 continue;
2238 }
2239
2240 /* Handle the other opcodes */
2241
2242 switch (c)
2243 {
2244 /* Check for quantifiers after a class. XCLASS is used for classes that
2245 cannot be represented just by a bit map. This includes negated single
2246 high-valued characters. The length in _pcre_OP_lengths[] is zero; the
2247 actual length is stored in the compiled code, so we must update "code"
2248 here. */
2249
2250 #ifdef SUPPORT_UTF8
2251 case OP_XCLASS:
2252 ccode = code += GET(code, 1);
2253 goto CHECK_CLASS_REPEAT;
2254 #endif
2255
2256 case OP_CLASS:
2257 case OP_NCLASS:
2258 ccode = code + 33;
2259
2260 #ifdef SUPPORT_UTF8
2261 CHECK_CLASS_REPEAT:
2262 #endif
2263
2264 switch (*ccode)
2265 {
2266 case OP_CRSTAR: /* These could be empty; continue */
2267 case OP_CRMINSTAR:
2268 case OP_CRQUERY:
2269 case OP_CRMINQUERY:
2270 break;
2271
2272 default: /* Non-repeat => class must match */
2273 case OP_CRPLUS: /* These repeats aren't empty */
2274 case OP_CRMINPLUS:
2275 return FALSE;
2276
2277 case OP_CRRANGE:
2278 case OP_CRMINRANGE:
2279 if (GET2(ccode, 1) > 0) return FALSE; /* Minimum > 0 */
2280 break;
2281 }
2282 break;
2283
2284 /* Opcodes that must match a character */
2285
2286 case OP_PROP:
2287 case OP_NOTPROP:
2288 case OP_EXTUNI:
2289 case OP_NOT_DIGIT:
2290 case OP_DIGIT:
2291 case OP_NOT_WHITESPACE:
2292 case OP_WHITESPACE:
2293 case OP_NOT_WORDCHAR:
2294 case OP_WORDCHAR:
2295 case OP_ANY:
2296 case OP_ALLANY:
2297 case OP_ANYBYTE:
2298 case OP_CHAR:
2299 case OP_CHARI:
2300 case OP_NOT:
2301 case OP_NOTI:
2302 case OP_PLUS:
2303 case OP_MINPLUS:
2304 case OP_POSPLUS:
2305 case OP_EXACT:
2306 case OP_NOTPLUS:
2307 case OP_NOTMINPLUS:
2308 case OP_NOTPOSPLUS:
2309 case OP_NOTEXACT:
2310 case OP_TYPEPLUS:
2311 case OP_TYPEMINPLUS:
2312 case OP_TYPEPOSPLUS:
2313 case OP_TYPEEXACT:
2314 return FALSE;
2315
2316 /* These are going to continue, as they may be empty, but we have to
2317 fudge the length for the \p and \P cases. */
2318
2319 case OP_TYPESTAR:
2320 case OP_TYPEMINSTAR:
2321 case OP_TYPEPOSSTAR:
2322 case OP_TYPEQUERY:
2323 case OP_TYPEMINQUERY:
2324 case OP_TYPEPOSQUERY:
2325 if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
2326 break;
2327
2328 /* Same for these */
2329
2330 case OP_TYPEUPTO:
2331 case OP_TYPEMINUPTO:
2332 case OP_TYPEPOSUPTO:
2333 if (code[3] == OP_PROP || code[3] == OP_NOTPROP) code += 2;
2334 break;
2335
2336 /* End of branch */
2337
2338 case OP_KET:
2339 case OP_KETRMAX:
2340 case OP_KETRMIN:
2341 case OP_KETRPOS:
2342 case OP_ALT:
2343 return TRUE;
2344
2345 /* In UTF-8 mode, STAR, MINSTAR, POSSTAR, QUERY, MINQUERY, POSQUERY, UPTO,
2346 MINUPTO, and POSUPTO may be followed by a multibyte character */
2347
2348 #ifdef SUPPORT_UTF8
2349 case OP_STAR:
2350 case OP_STARI:
2351 case OP_MINSTAR:
2352 case OP_MINSTARI:
2353 case OP_POSSTAR:
2354 case OP_POSSTARI:
2355 case OP_QUERY:
2356 case OP_QUERYI:
2357 case OP_MINQUERY:
2358 case OP_MINQUERYI:
2359 case OP_POSQUERY:
2360 case OP_POSQUERYI:
2361 if (utf8 && code[1] >= 0xc0) code += _pcre_utf8_table4[code[1] & 0x3f];
2362 break;
2363
2364 case OP_UPTO:
2365 case OP_UPTOI:
2366 case OP_MINUPTO:
2367 case OP_MINUPTOI:
2368 case OP_POSUPTO:
2369 case OP_POSUPTOI:
2370 if (utf8 && code[3] >= 0xc0) code += _pcre_utf8_table4[code[3] & 0x3f];
2371 break;
2372 #endif
2373
2374 /* MARK, and PRUNE/SKIP/THEN with an argument must skip over the argument
2375 string. */
2376
2377 case OP_MARK:
2378 case OP_PRUNE_ARG:
2379 case OP_SKIP_ARG:
2380 code += code[1];
2381 break;
2382
2383 case OP_THEN_ARG:
2384 code += code[1];
2385 break;
2386
2387 /* None of the remaining opcodes are required to match a character. */
2388
2389 default:
2390 break;
2391 }
2392 }
2393
2394 return TRUE;
2395 }
2396
2397
2398
2399 /*************************************************
2400 * Scan compiled regex for non-emptiness *
2401 *************************************************/
2402
2403 /* This function is called to check for left recursive calls. We want to check
2404 the current branch of the current pattern to see if it could match the empty
2405 string. If it could, we must look outwards for branches at other levels,
2406 stopping when we pass beyond the bracket which is the subject of the recursion.
2407 This function is called only during the real compile, not during the
2408 pre-compile.
2409
2410 Arguments:
2411 code points to start of the recursion
2412 endcode points to where to stop (current RECURSE item)
2413 bcptr points to the chain of current (unclosed) branch starts
2414 utf8 TRUE if in UTF-8 mode
2415 cd pointers to tables etc
2416
2417 Returns: TRUE if what is matched could be empty
2418 */
2419
2420 static BOOL
2421 could_be_empty(const uschar *code, const uschar *endcode, branch_chain *bcptr,
2422 BOOL utf8, compile_data *cd)
2423 {
2424 while (bcptr != NULL && bcptr->current_branch >= code)
2425 {
2426 if (!could_be_empty_branch(bcptr->current_branch, endcode, utf8, cd))
2427 return FALSE;
2428 bcptr = bcptr->outer;
2429 }
2430 return TRUE;
2431 }
2432
2433
2434
2435 /*************************************************
2436 * Check for POSIX class syntax *
2437 *************************************************/
2438
2439 /* This function is called when the sequence "[:" or "[." or "[=" is
2440 encountered in a character class. It checks whether this is followed by a
2441 sequence of characters terminated by a matching ":]" or ".]" or "=]". If we
2442 reach an unescaped ']' without the special preceding character, return FALSE.
2443
2444 Originally, this function only recognized a sequence of letters between the
2445 terminators, but it seems that Perl recognizes any sequence of characters,
2446 though of course unknown POSIX names are subsequently rejected. Perl gives an
2447 "Unknown POSIX class" error for [:f\oo:] for example, where previously PCRE
2448 didn't consider this to be a POSIX class. Likewise for [:1234:].
2449
2450 The problem in trying to be exactly like Perl is in the handling of escapes. We
2451 have to be sure that [abc[:x\]pqr] is *not* treated as containing a POSIX
2452 class, but [abc[:x\]pqr:]] is (so that an error can be generated). The code
2453 below handles the special case of \], but does not try to do any other escape
2454 processing. This makes it different from Perl for cases such as [:l\ower:]
2455 where Perl recognizes it as the POSIX class "lower" but PCRE does not recognize
2456 "l\ower". This is a lesser evil that not diagnosing bad classes when Perl does,
2457 I think.
2458
2459 A user pointed out that PCRE was rejecting [:a[:digit:]] whereas Perl was not.
2460 It seems that the appearance of a nested POSIX class supersedes an apparent
2461 external class. For example, [:a[:digit:]b:] matches "a", "b", ":", or
2462 a digit.
2463
2464 In Perl, unescaped square brackets may also appear as part of class names. For
2465 example, [:a[:abc]b:] gives unknown POSIX class "[:abc]b:]". However, for
2466 [:a[:abc]b][b:] it gives unknown POSIX class "[:abc]b][b:]", which does not
2467 seem right at all. PCRE does not allow closing square brackets in POSIX class
2468 names.
2469
2470 Arguments:
2471 ptr pointer to the initial [
2472 endptr where to return the end pointer
2473
2474 Returns: TRUE or FALSE
2475 */
2476
2477 static BOOL
2478 check_posix_syntax(const uschar *ptr, const uschar **endptr)
2479 {
2480 int terminator; /* Don't combine these lines; the Solaris cc */
2481 terminator = *(++ptr); /* compiler warns about "non-constant" initializer. */
2482 for (++ptr; *ptr != 0; ptr++)
2483 {
2484 if (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET)
2485 ptr++;
2486 else if (*ptr == CHAR_RIGHT_SQUARE_BRACKET) return FALSE;
2487 else
2488 {
2489 if (*ptr == terminator && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET)
2490 {
2491 *endptr = ptr;
2492 return TRUE;
2493 }
2494 if (*ptr == CHAR_LEFT_SQUARE_BRACKET &&
2495 (ptr[1] == CHAR_COLON || ptr[1] == CHAR_DOT ||
2496 ptr[1] == CHAR_EQUALS_SIGN) &&
2497 check_posix_syntax(ptr, endptr))
2498 return FALSE;
2499 }
2500 }
2501 return FALSE;
2502 }
2503
2504
2505
2506
2507 /*************************************************
2508 * Check POSIX class name *
2509 *************************************************/
2510
2511 /* This function is called to check the name given in a POSIX-style class entry
2512 such as [:alnum:].
2513
2514 Arguments:
2515 ptr points to the first letter
2516 len the length of the name
2517
2518 Returns: a value representing the name, or -1 if unknown
2519 */
2520
2521 static int
2522 check_posix_name(const uschar *ptr, int len)
2523 {
2524 const char *pn = posix_names;
2525 register int yield = 0;
2526 while (posix_name_lengths[yield] != 0)
2527 {
2528 if (len == posix_name_lengths[yield] &&
2529 strncmp((const char *)ptr, pn, len) == 0) return yield;
2530 pn += posix_name_lengths[yield] + 1;
2531 yield++;
2532 }
2533 return -1;
2534 }
2535
2536
2537 /*************************************************
2538 * Adjust OP_RECURSE items in repeated group *
2539 *************************************************/
2540
2541 /* OP_RECURSE items contain an offset from the start of the regex to the group
2542 that is referenced. This means that groups can be replicated for fixed
2543 repetition simply by copying (because the recursion is allowed to refer to
2544 earlier groups that are outside the current group). However, when a group is
2545 optional (i.e. the minimum quantifier is zero), OP_BRAZERO or OP_SKIPZERO is
2546 inserted before it, after it has been compiled. This means that any OP_RECURSE
2547 items within it that refer to the group itself or any contained groups have to
2548 have their offsets adjusted. That one of the jobs of this function. Before it
2549 is called, the partially compiled regex must be temporarily terminated with
2550 OP_END.
2551
2552 This function has been extended with the possibility of forward references for
2553 recursions and subroutine calls. It must also check the list of such references
2554 for the group we are dealing with. If it finds that one of the recursions in
2555 the current group is on this list, it adjusts the offset in the list, not the
2556 value in the reference (which is a group number).
2557
2558 Arguments:
2559 group points to the start of the group
2560 adjust the amount by which the group is to be moved
2561 utf8 TRUE in UTF-8 mode
2562 cd contains pointers to tables etc.
2563 save_hwm the hwm forward reference pointer at the start of the group
2564
2565 Returns: nothing
2566 */
2567
2568 static void
2569 adjust_recurse(uschar *group, int adjust, BOOL utf8, compile_data *cd,
2570 uschar *save_hwm)
2571 {
2572 uschar *ptr = group;
2573
2574 while ((ptr = (uschar *)find_recurse(ptr, utf8)) != NULL)
2575 {
2576 int offset;
2577 uschar *hc;
2578
2579 /* See if this recursion is on the forward reference list. If so, adjust the
2580 reference. */
2581
2582 for (hc = save_hwm; hc < cd->hwm; hc += LINK_SIZE)
2583 {
2584 offset = GET(hc, 0);
2585 if (cd->start_code + offset == ptr + 1)
2586 {
2587 PUT(hc, 0, offset + adjust);
2588 break;
2589 }
2590 }
2591
2592 /* Otherwise, adjust the recursion offset if it's after the start of this
2593 group. */
2594
2595 if (hc >= cd->hwm)
2596 {
2597 offset = GET(ptr, 1);
2598 if (cd->start_code + offset >= group) PUT(ptr, 1, offset + adjust);
2599 }
2600
2601 ptr += 1 + LINK_SIZE;
2602 }
2603 }
2604
2605
2606
2607 /*************************************************
2608 * Insert an automatic callout point *
2609 *************************************************/
2610
2611 /* This function is called when the PCRE_AUTO_CALLOUT option is set, to insert
2612 callout points before each pattern item.
2613
2614 Arguments:
2615 code current code pointer
2616 ptr current pattern pointer
2617 cd pointers to tables etc
2618
2619 Returns: new code pointer
2620 */
2621
2622 static uschar *
2623 auto_callout(uschar *code, const uschar *ptr, compile_data *cd)
2624 {
2625 *code++ = OP_CALLOUT;
2626 *code++ = 255;
2627 PUT(code, 0, (int)(ptr - cd->start_pattern)); /* Pattern offset */
2628 PUT(code, LINK_SIZE, 0); /* Default length */
2629 return code + 2*LINK_SIZE;
2630 }
2631
2632
2633
2634 /*************************************************
2635 * Complete a callout item *
2636 *************************************************/
2637
2638 /* A callout item contains the length of the next item in the pattern, which
2639 we can't fill in till after we have reached the relevant point. This is used
2640 for both automatic and manual callouts.
2641
2642 Arguments:
2643 previous_callout points to previous callout item
2644 ptr current pattern pointer
2645 cd pointers to tables etc
2646
2647 Returns: nothing
2648 */
2649
2650 static void
2651 complete_callout(uschar *previous_callout, const uschar *ptr, compile_data *cd)
2652 {
2653 int length = (int)(ptr - cd->start_pattern - GET(previous_callout, 2));
2654 PUT(previous_callout, 2 + LINK_SIZE, length);
2655 }
2656
2657
2658
2659 #ifdef SUPPORT_UCP
2660 /*************************************************
2661 * Get othercase range *
2662 *************************************************/
2663
2664 /* This function is passed the start and end of a class range, in UTF-8 mode
2665 with UCP support. It searches up the characters, looking for internal ranges of
2666 characters in the "other" case. Each call returns the next one, updating the
2667 start address.
2668
2669 Arguments:
2670 cptr points to starting character value; updated
2671 d end value
2672 ocptr where to put start of othercase range
2673 odptr where to put end of othercase range
2674
2675 Yield: TRUE when range returned; FALSE when no more
2676 */
2677
2678 static BOOL
2679 get_othercase_range(unsigned int *cptr, unsigned int d, unsigned int *ocptr,
2680 unsigned int *odptr)
2681 {
2682 unsigned int c, othercase, next;
2683
2684 for (c = *cptr; c <= d; c++)
2685 { if ((othercase = UCD_OTHERCASE(c)) != c) break; }
2686
2687 if (c > d) return FALSE;
2688
2689 *ocptr = othercase;
2690 next = othercase + 1;
2691
2692 for (++c; c <= d; c++)
2693 {
2694 if (UCD_OTHERCASE(c) != next) break;
2695 next++;
2696 }
2697
2698 *odptr = next - 1;
2699 *cptr = c;
2700
2701 return TRUE;
2702 }
2703
2704
2705
2706 /*************************************************
2707 * Check a character and a property *
2708 *************************************************/
2709
2710 /* This function is called by check_auto_possessive() when a property item
2711 is adjacent to a fixed character.
2712
2713 Arguments:
2714 c the character
2715 ptype the property type
2716 pdata the data for the type
2717 negated TRUE if it's a negated property (\P or \p{^)
2718
2719 Returns: TRUE if auto-possessifying is OK
2720 */
2721
2722 static BOOL
2723 check_char_prop(int c, int ptype, int pdata, BOOL negated)
2724 {
2725 const ucd_record *prop = GET_UCD(c);
2726 switch(ptype)
2727 {
2728 case PT_LAMP:
2729 return (prop->chartype == ucp_Lu ||
2730 prop->chartype == ucp_Ll ||
2731 prop->chartype == ucp_Lt) == negated;
2732
2733 case PT_GC:
2734 return (pdata == _pcre_ucp_gentype[prop->chartype]) == negated;
2735
2736 case PT_PC:
2737 return (pdata == prop->chartype) == negated;
2738
2739 case PT_SC:
2740 return (pdata == prop->script) == negated;
2741
2742 /* These are specials */
2743
2744 case PT_ALNUM:
2745 return (_pcre_ucp_gentype[prop->chartype] == ucp_L ||
2746 _pcre_ucp_gentype[prop->chartype] == ucp_N) == negated;
2747
2748 case PT_SPACE: /* Perl space */
2749 return (_pcre_ucp_gentype[prop->chartype] == ucp_Z ||
2750 c == CHAR_HT || c == CHAR_NL || c == CHAR_FF || c == CHAR_CR)
2751 == negated;
2752
2753 case PT_PXSPACE: /* POSIX space */
2754 return (_pcre_ucp_gentype[prop->chartype] == ucp_Z ||
2755 c == CHAR_HT || c == CHAR_NL || c == CHAR_VT ||
2756 c == CHAR_FF || c == CHAR_CR)
2757 == negated;
2758
2759 case PT_WORD:
2760 return (_pcre_ucp_gentype[prop->chartype] == ucp_L ||
2761 _pcre_ucp_gentype[prop->chartype] == ucp_N ||
2762 c == CHAR_UNDERSCORE) == negated;
2763 }
2764 return FALSE;
2765 }
2766 #endif /* SUPPORT_UCP */
2767
2768
2769
2770 /*************************************************
2771 * Check if auto-possessifying is possible *
2772 *************************************************/
2773
2774 /* This function is called for unlimited repeats of certain items, to see
2775 whether the next thing could possibly match the repeated item. If not, it makes
2776 sense to automatically possessify the repeated item.
2777
2778 Arguments:
2779 previous pointer to the repeated opcode
2780 utf8 TRUE in UTF-8 mode
2781 ptr next character in pattern
2782 options options bits
2783 cd contains pointers to tables etc.
2784
2785 Returns: TRUE if possessifying is wanted
2786 */
2787
2788 static BOOL
2789 check_auto_possessive(const uschar *previous, BOOL utf8, const uschar *ptr,
2790 int options, compile_data *cd)
2791 {
2792 int c, next;
2793 int op_code = *previous++;
2794
2795 /* Skip whitespace and comments in extended mode */
2796
2797 if ((options & PCRE_EXTENDED) != 0)
2798 {
2799 for (;;)
2800 {
2801 while ((cd->ctypes[*ptr] & ctype_space) != 0) ptr++;
2802 if (*ptr == CHAR_NUMBER_SIGN)
2803 {
2804 ptr++;
2805 while (*ptr != 0)
2806 {
2807 if (IS_NEWLINE(ptr)) { ptr += cd->nllen; break; }
2808 ptr++;
2809 #ifdef SUPPORT_UTF8
2810 if (utf8) while ((*ptr & 0xc0) == 0x80) ptr++;
2811 #endif
2812 }
2813 }
2814 else break;
2815 }
2816 }
2817
2818 /* If the next item is one that we can handle, get its value. A non-negative
2819 value is a character, a negative value is an escape value. */
2820
2821 if (*ptr == CHAR_BACKSLASH)
2822 {
2823 int temperrorcode = 0;
2824 next = check_escape(&ptr, &temperrorcode, cd->bracount, options, FALSE);
2825 if (temperrorcode != 0) return FALSE;
2826 ptr++; /* Point after the escape sequence */
2827 }
2828
2829 else if ((cd->ctypes[*ptr] & ctype_meta) == 0)
2830 {
2831 #ifdef SUPPORT_UTF8
2832 if (utf8) { GETCHARINC(next, ptr); } else
2833 #endif
2834 next = *ptr++;
2835 }
2836
2837 else return FALSE;
2838
2839 /* Skip whitespace and comments in extended mode */
2840
2841 if ((options & PCRE_EXTENDED) != 0)
2842 {
2843 for (;;)
2844 {
2845 while ((cd->ctypes[*ptr] & ctype_space) != 0) ptr++;
2846 if (*ptr == CHAR_NUMBER_SIGN)
2847 {
2848 ptr++;
2849 while (*ptr != 0)
2850 {
2851 if (IS_NEWLINE(ptr)) { ptr += cd->nllen; break; }
2852 ptr++;
2853 #ifdef SUPPORT_UTF8
2854 if (utf8) while ((*ptr & 0xc0) == 0x80) ptr++;
2855 #endif
2856 }
2857 }
2858 else break;
2859 }
2860 }
2861
2862 /* If the next thing is itself optional, we have to give up. */
2863
2864 if (*ptr == CHAR_ASTERISK || *ptr == CHAR_QUESTION_MARK ||
2865 strncmp((char *)ptr, STR_LEFT_CURLY_BRACKET STR_0 STR_COMMA, 3) == 0)
2866 return FALSE;
2867
2868 /* Now compare the next item with the previous opcode. First, handle cases when
2869 the next item is a character. */
2870
2871 if (next >= 0) switch(op_code)
2872 {
2873 case OP_CHAR:
2874 #ifdef SUPPORT_UTF8
2875 GETCHARTEST(c, previous);
2876 #else
2877 c = *previous;
2878 #endif
2879 return c != next;
2880
2881 /* For CHARI (caseless character) we must check the other case. If we have
2882 Unicode property support, we can use it to test the other case of
2883 high-valued characters. */
2884
2885 case OP_CHARI:
2886 #ifdef SUPPORT_UTF8
2887 GETCHARTEST(c, previous);
2888 #else
2889 c = *previous;
2890 #endif
2891 if (c == next) return FALSE;
2892 #ifdef SUPPORT_UTF8
2893 if (utf8)
2894 {
2895 unsigned int othercase;
2896 if (next < 128) othercase = cd->fcc[next]; else
2897 #ifdef SUPPORT_UCP
2898 othercase = UCD_OTHERCASE((unsigned int)next);
2899 #else
2900 othercase = NOTACHAR;
2901 #endif
2902 return (unsigned int)c != othercase;
2903 }
2904 else
2905 #endif /* SUPPORT_UTF8 */
2906 return (c != cd->fcc[next]); /* Non-UTF-8 mode */
2907
2908 /* For OP_NOT and OP_NOTI, the data is always a single-byte character. These
2909 opcodes are not used for multi-byte characters, because they are coded using
2910 an XCLASS instead. */
2911
2912 case OP_NOT:
2913 return (c = *previous) == next;
2914
2915 case OP_NOTI:
2916 if ((c = *previous) == next) return TRUE;
2917 #ifdef SUPPORT_UTF8
2918 if (utf8)
2919 {
2920 unsigned int othercase;
2921 if (next < 128) othercase = cd->fcc[next]; else
2922 #ifdef SUPPORT_UCP
2923 othercase = UCD_OTHERCASE(next);
2924 #else
2925 othercase = NOTACHAR;
2926 #endif
2927 return (unsigned int)c == othercase;
2928 }
2929 else
2930 #endif /* SUPPORT_UTF8 */
2931 return (c == cd->fcc[next]); /* Non-UTF-8 mode */
2932
2933 /* Note that OP_DIGIT etc. are generated only when PCRE_UCP is *not* set.
2934 When it is set, \d etc. are converted into OP_(NOT_)PROP codes. */
2935
2936 case OP_DIGIT:
2937 return next > 127 || (cd->ctypes[next] & ctype_digit) == 0;
2938
2939 case OP_NOT_DIGIT:
2940 return next <= 127 && (cd->ctypes[next] & ctype_digit) != 0;
2941
2942 case OP_WHITESPACE:
2943 return next > 127 || (cd->ctypes[next] & ctype_space) == 0;
2944
2945 case OP_NOT_WHITESPACE:
2946 return next <= 127 && (cd->ctypes[next] & ctype_space) != 0;
2947
2948 case OP_WORDCHAR:
2949 return next > 127 || (cd->ctypes[next] & ctype_word) == 0;
2950
2951 case OP_NOT_WORDCHAR:
2952 return next <= 127 && (cd->ctypes[next] & ctype_word) != 0;
2953
2954 case OP_HSPACE:
2955 case OP_NOT_HSPACE:
2956 switch(next)
2957 {
2958 case 0x09:
2959 case 0x20:
2960 case 0xa0:
2961 case 0x1680:
2962 case 0x180e:
2963 case 0x2000:
2964 case 0x2001:
2965 case 0x2002:
2966 case 0x2003:
2967 case 0x2004:
2968 case 0x2005:
2969 case 0x2006:
2970 case 0x2007:
2971 case 0x2008:
2972 case 0x2009:
2973 case 0x200A:
2974 case 0x202f:
2975 case 0x205f:
2976 case 0x3000:
2977 return op_code == OP_NOT_HSPACE;
2978 default:
2979 return op_code != OP_NOT_HSPACE;
2980 }
2981
2982 case OP_ANYNL:
2983 case OP_VSPACE:
2984 case OP_NOT_VSPACE:
2985 switch(next)
2986 {
2987 case 0x0a:
2988 case 0x0b:
2989 case 0x0c:
2990 case 0x0d:
2991 case 0x85:
2992 case 0x2028:
2993 case 0x2029:
2994 return op_code == OP_NOT_VSPACE;
2995 default:
2996 return op_code != OP_NOT_VSPACE;
2997 }
2998
2999 #ifdef SUPPORT_UCP
3000 case OP_PROP:
3001 return check_char_prop(next, previous[0], previous[1], FALSE);
3002
3003 case OP_NOTPROP:
3004 return check_char_prop(next, previous[0], previous[1], TRUE);
3005 #endif
3006
3007 default:
3008 return FALSE;
3009 }
3010
3011
3012 /* Handle the case when the next item is \d, \s, etc. Note that when PCRE_UCP
3013 is set, \d turns into ESC_du rather than ESC_d, etc., so ESC_d etc. are
3014 generated only when PCRE_UCP is *not* set, that is, when only ASCII
3015 characteristics are recognized. Similarly, the opcodes OP_DIGIT etc. are
3016 replaced by OP_PROP codes when PCRE_UCP is set. */
3017
3018 switch(op_code)
3019 {
3020 case OP_CHAR:
3021 case OP_CHARI:
3022 #ifdef SUPPORT_UTF8
3023 GETCHARTEST(c, previous);
3024 #else
3025 c = *previous;
3026 #endif
3027 switch(-next)
3028 {
3029 case ESC_d:
3030 return c > 127 || (cd->ctypes[c] & ctype_digit) == 0;
3031
3032 case ESC_D:
3033 return c <= 127 && (cd->ctypes[c] & ctype_digit) != 0;
3034
3035 case ESC_s:
3036 return c > 127 || (cd->ctypes[c] & ctype_space) == 0;
3037
3038 case ESC_S:
3039 return c <= 127 && (cd->ctypes[c] & ctype_space) != 0;
3040
3041 case ESC_w:
3042 return c > 127 || (cd->ctypes[c] & ctype_word) == 0;
3043
3044 case ESC_W:
3045 return c <= 127 && (cd->ctypes[c] & ctype_word) != 0;
3046
3047 case ESC_h:
3048 case ESC_H:
3049 switch(c)
3050 {
3051 case 0x09:
3052 case 0x20:
3053 case 0xa0:
3054 case 0x1680:
3055 case 0x180e:
3056 case 0x2000:
3057 case 0x2001:
3058 case 0x2002:
3059 case 0x2003:
3060 case 0x2004:
3061 case 0x2005:
3062 case 0x2006:
3063 case 0x2007:
3064 case 0x2008:
3065 case 0x2009:
3066 case 0x200A:
3067 case 0x202f:
3068 case 0x205f:
3069 case 0x3000:
3070 return -next != ESC_h;
3071 default:
3072 return -next == ESC_h;
3073 }
3074
3075 case ESC_v:
3076 case ESC_V:
3077 switch(c)
3078 {
3079 case 0x0a:
3080 case 0x0b:
3081 case 0x0c:
3082 case 0x0d:
3083 case 0x85:
3084 case 0x2028:
3085 case 0x2029:
3086 return -next != ESC_v;
3087 default:
3088 return -next == ESC_v;
3089 }
3090
3091 /* When PCRE_UCP is set, these values get generated for \d etc. Find
3092 their substitutions and process them. The result will always be either
3093 -ESC_p or -ESC_P. Then fall through to process those values. */
3094
3095 #ifdef SUPPORT_UCP
3096 case ESC_du:
3097 case ESC_DU:
3098 case ESC_wu:
3099 case ESC_WU:
3100 case ESC_su:
3101 case ESC_SU:
3102 {
3103 int temperrorcode = 0;
3104 ptr = substitutes[-next - ESC_DU];
3105 next = check_escape(&ptr, &temperrorcode, 0, options, FALSE);
3106 if (temperrorcode != 0) return FALSE;
3107 ptr++; /* For compatibility */
3108 }
3109 /* Fall through */
3110
3111 case ESC_p:
3112 case ESC_P:
3113 {
3114 int ptype, pdata, errorcodeptr;
3115 BOOL negated;
3116
3117 ptr--; /* Make ptr point at the p or P */
3118 ptype = get_ucp(&ptr, &negated, &pdata, &errorcodeptr);
3119 if (ptype < 0) return FALSE;
3120 ptr++; /* Point past the final curly ket */
3121
3122 /* If the property item is optional, we have to give up. (When generated
3123 from \d etc by PCRE_UCP, this test will have been applied much earlier,
3124 to the original \d etc. At this point, ptr will point to a zero byte. */
3125
3126 if (*ptr == CHAR_ASTERISK || *ptr == CHAR_QUESTION_MARK ||
3127 strncmp((char *)ptr, STR_LEFT_CURLY_BRACKET STR_0 STR_COMMA, 3) == 0)
3128 return FALSE;
3129
3130 /* Do the property check. */
3131
3132 return check_char_prop(c, ptype, pdata, (next == -ESC_P) != negated);
3133 }
3134 #endif
3135
3136 default:
3137 return FALSE;
3138 }
3139
3140 /* In principle, support for Unicode properties should be integrated here as
3141 well. It means re-organizing the above code so as to get hold of the property
3142 values before switching on the op-code. However, I wonder how many patterns
3143 combine ASCII \d etc with Unicode properties? (Note that if PCRE_UCP is set,
3144 these op-codes are never generated.) */
3145
3146 case OP_DIGIT:
3147 return next == -ESC_D || next == -ESC_s || next == -ESC_W ||
3148 next == -ESC_h || next == -ESC_v || next == -ESC_R;
3149
3150 case OP_NOT_DIGIT:
3151 return next == -ESC_d;
3152
3153 case OP_WHITESPACE:
3154 return next == -ESC_S || next == -ESC_d || next == -ESC_w || next == -ESC_R;
3155
3156 case OP_NOT_WHITESPACE:
3157 return next == -ESC_s || next == -ESC_h || next == -ESC_v;
3158
3159 case OP_HSPACE:
3160 return next == -ESC_S || next == -ESC_H || next == -ESC_d ||
3161 next == -ESC_w || next == -ESC_v || next == -ESC_R;
3162
3163 case OP_NOT_HSPACE:
3164 return next == -ESC_h;
3165
3166 /* Can't have \S in here because VT matches \S (Perl anomaly) */
3167 case OP_ANYNL:
3168 case OP_VSPACE:
3169 return next == -ESC_V || next == -ESC_d || next == -ESC_w;
3170
3171 case OP_NOT_VSPACE:
3172 return next == -ESC_v || next == -ESC_R;
3173
3174 case OP_WORDCHAR:
3175 return next == -ESC_W || next == -ESC_s || next == -ESC_h ||
3176 next == -ESC_v || next == -ESC_R;
3177
3178 case OP_NOT_WORDCHAR:
3179 return next == -ESC_w || next == -ESC_d;
3180
3181 default:
3182 return FALSE;
3183 }
3184
3185 /* Control does not reach here */
3186 }
3187
3188
3189
3190 /*************************************************
3191 * Compile one branch *
3192 *************************************************/
3193
3194 /* Scan the pattern, compiling it into the a vector. If the options are
3195 changed during the branch, the pointer is used to change the external options
3196 bits. This function is used during the pre-compile phase when we are trying
3197 to find out the amount of memory needed, as well as during the real compile
3198 phase. The value of lengthptr distinguishes the two phases.
3199
3200 Arguments:
3201 optionsptr pointer to the option bits
3202 codeptr points to the pointer to the current code point
3203 ptrptr points to the current pattern pointer
3204 errorcodeptr points to error code variable
3205 firstbyteptr set to initial literal character, or < 0 (REQ_UNSET, REQ_NONE)
3206 reqbyteptr set to the last literal character required, else < 0
3207 bcptr points to current branch chain
3208 cond_depth conditional nesting depth
3209 cd contains pointers to tables etc.
3210 lengthptr NULL during the real compile phase
3211 points to length accumulator during pre-compile phase
3212
3213 Returns: TRUE on success
3214 FALSE, with *errorcodeptr set non-zero on error
3215 */
3216
3217 static BOOL
3218 compile_branch(int *optionsptr, uschar **codeptr, const uschar **ptrptr,
3219 int *errorcodeptr, int *firstbyteptr, int *reqbyteptr, branch_chain *bcptr,
3220 int cond_depth, compile_data *cd, int *lengthptr)
3221 {
3222 int repeat_type, op_type;
3223 int repeat_min = 0, repeat_max = 0; /* To please picky compilers */
3224 int bravalue = 0;
3225 int greedy_default, greedy_non_default;
3226 int firstbyte, reqbyte;
3227 int zeroreqbyte, zerofirstbyte;
3228 int req_caseopt, reqvary, tempreqvary;
3229 int options = *optionsptr; /* May change dynamically */
3230 int after_manual_callout = 0;
3231 int length_prevgroup = 0;
3232 register int c;
3233 register uschar *code = *codeptr;
3234 uschar *last_code = code;
3235 uschar *orig_code = code;
3236 uschar *tempcode;
3237 BOOL inescq = FALSE;
3238 BOOL groupsetfirstbyte = FALSE;
3239 const uschar *ptr = *ptrptr;
3240 const uschar *tempptr;
3241 const uschar *nestptr = NULL;
3242 uschar *previous = NULL;
3243 uschar *previous_callout = NULL;
3244 uschar *save_hwm = NULL;
3245 uschar classbits[32];
3246
3247 /* We can fish out the UTF-8 setting once and for all into a BOOL, but we
3248 must not do this for other options (e.g. PCRE_EXTENDED) because they may change
3249 dynamically as we process the pattern. */
3250
3251 #ifdef SUPPORT_UTF8
3252 BOOL class_utf8;
3253 BOOL utf8 = (options & PCRE_UTF8) != 0;
3254 uschar *class_utf8data;
3255 uschar *class_utf8data_base;
3256 uschar utf8_char[6];
3257 #else
3258 BOOL utf8 = FALSE;
3259 #endif
3260
3261 #ifdef PCRE_DEBUG
3262 if (lengthptr != NULL) DPRINTF((">> start branch\n"));
3263 #endif
3264
3265 /* Set up the default and non-default settings for greediness */
3266
3267 greedy_default = ((options & PCRE_UNGREEDY) != 0);
3268 greedy_non_default = greedy_default ^ 1;
3269
3270 /* Initialize no first byte, no required byte. REQ_UNSET means "no char
3271 matching encountered yet". It gets changed to REQ_NONE if we hit something that
3272 matches a non-fixed char first char; reqbyte just remains unset if we never
3273 find one.
3274
3275 When we hit a repeat whose minimum is zero, we may have to adjust these values
3276 to take the zero repeat into account. This is implemented by setting them to
3277 zerofirstbyte and zeroreqbyte when such a repeat is encountered. The individual
3278 item types that can be repeated set these backoff variables appropriately. */
3279
3280 firstbyte = reqbyte = zerofirstbyte = zeroreqbyte = REQ_UNSET;
3281
3282 /* The variable req_caseopt contains either the REQ_CASELESS value or zero,
3283 according to the current setting of the caseless flag. REQ_CASELESS is a bit
3284 value > 255. It is added into the firstbyte or reqbyte variables to record the
3285 case status of the value. This is used only for ASCII characters. */
3286
3287 req_caseopt = ((options & PCRE_CASELESS) != 0)? REQ_CASELESS : 0;
3288
3289 /* Switch on next character until the end of the branch */
3290
3291 for (;; ptr++)
3292 {
3293 BOOL negate_class;
3294 BOOL should_flip_negation;
3295 BOOL possessive_quantifier;
3296 BOOL is_quantifier;
3297 BOOL is_recurse;
3298 BOOL reset_bracount;
3299 int class_charcount;
3300 int class_lastchar;
3301 int newoptions;
3302 int recno;
3303 int refsign;
3304 int skipbytes;
3305 int subreqbyte;
3306 int subfirstbyte;
3307 int terminator;
3308 int mclength;
3309 int tempbracount;
3310 uschar mcbuffer[8];
3311
3312 /* Get next byte in the pattern */
3313
3314 c = *ptr;
3315
3316 /* If we are at the end of a nested substitution, revert to the outer level
3317 string. Nesting only happens one level deep. */
3318
3319 if (c == 0 && nestptr != NULL)
3320 {
3321 ptr = nestptr;
3322 nestptr = NULL;
3323 c = *ptr;
3324 }
3325
3326 /* If we are in the pre-compile phase, accumulate the length used for the
3327 previous cycle of this loop. */
3328
3329 if (lengthptr != NULL)
3330 {
3331 #ifdef PCRE_DEBUG
3332 if (code > cd->hwm) cd->hwm = code; /* High water info */
3333 #endif
3334 if (code > cd->start_workspace + WORK_SIZE_CHECK) /* Check for overrun */
3335 {
3336 *errorcodeptr = ERR52;
3337 goto FAILED;
3338 }
3339
3340 /* There is at least one situation where code goes backwards: this is the
3341 case of a zero quantifier after a class (e.g. [ab]{0}). At compile time,
3342 the class is simply eliminated. However, it is created first, so we have to
3343 allow memory for it. Therefore, don't ever reduce the length at this point.
3344 */
3345
3346 if (code < last_code) code = last_code;
3347
3348 /* Paranoid check for integer overflow */
3349
3350 if (OFLOW_MAX - *lengthptr < code - last_code)
3351 {
3352 *errorcodeptr = ERR20;
3353 goto FAILED;
3354 }
3355
3356 *lengthptr += (int)(code - last_code);
3357 DPRINTF(("length=%d added %d c=%c\n", *lengthptr, (int)(code - last_code),
3358 c));
3359
3360 /* If "previous" is set and it is not at the start of the work space, move
3361 it back to there, in order to avoid filling up the work space. Otherwise,
3362 if "previous" is NULL, reset the current code pointer to the start. */
3363
3364 if (previous != NULL)
3365 {
3366 if (previous > orig_code)
3367 {
3368 memmove(orig_code, previous, code - previous);
3369 code -= previous - orig_code;
3370 previous = orig_code;
3371 }
3372 }
3373 else code = orig_code;
3374
3375 /* Remember where this code item starts so we can pick up the length
3376 next time round. */
3377
3378 last_code = code;
3379 }
3380
3381 /* In the real compile phase, just check the workspace used by the forward
3382 reference list. */
3383
3384 else if (cd->hwm > cd->start_workspace + WORK_SIZE_CHECK)
3385 {
3386 *errorcodeptr = ERR52;
3387 goto FAILED;
3388 }
3389
3390 /* If in \Q...\E, check for the end; if not, we have a literal */
3391
3392 if (inescq && c != 0)
3393 {
3394 if (c == CHAR_BACKSLASH && ptr[1] == CHAR_E)
3395 {
3396 inescq = FALSE;
3397 ptr++;
3398 continue;
3399 }
3400 else
3401 {
3402 if (previous_callout != NULL)
3403 {
3404 if (lengthptr == NULL) /* Don't attempt in pre-compile phase */
3405 complete_callout(previous_callout, ptr, cd);
3406 previous_callout = NULL;
3407 }
3408 if ((options & PCRE_AUTO_CALLOUT) != 0)
3409 {
3410 previous_callout = code;
3411 code = auto_callout(code, ptr, cd);
3412 }
3413 goto NORMAL_CHAR;
3414 }
3415 }
3416
3417 /* Fill in length of a previous callout, except when the next thing is
3418 a quantifier. */
3419
3420 is_quantifier =
3421 c == CHAR_ASTERISK || c == CHAR_PLUS || c == CHAR_QUESTION_MARK ||
3422 (c == CHAR_LEFT_CURLY_BRACKET && is_counted_repeat(ptr+1));
3423
3424 if (!is_quantifier && previous_callout != NULL &&
3425 after_manual_callout-- <= 0)
3426 {
3427 if (lengthptr == NULL) /* Don't attempt in pre-compile phase */
3428 complete_callout(previous_callout, ptr, cd);
3429 previous_callout = NULL;
3430 }
3431
3432 /* In extended mode, skip white space and comments. */
3433
3434 if ((options & PCRE_EXTENDED) != 0)
3435 {
3436 if ((cd->ctypes[c] & ctype_space) != 0) continue;
3437 if (c == CHAR_NUMBER_SIGN)
3438 {
3439 ptr++;
3440 while (*ptr != 0)
3441 {
3442 if (IS_NEWLINE(ptr)) { ptr += cd->nllen - 1; break; }
3443 ptr++;
3444 #ifdef SUPPORT_UTF8
3445 if (utf8) while ((*ptr & 0xc0) == 0x80) ptr++;
3446 #endif
3447 }
3448 if (*ptr != 0) continue;
3449
3450 /* Else fall through to handle end of string */
3451 c = 0;
3452 }
3453 }
3454
3455 /* No auto callout for quantifiers. */
3456
3457 if ((options & PCRE_AUTO_CALLOUT) != 0 && !is_quantifier)
3458 {
3459 previous_callout = code;
3460 code = auto_callout(code, ptr, cd);
3461 }
3462
3463 switch(c)
3464 {
3465 /* ===================================================================*/
3466 case 0: /* The branch terminates at string end */
3467 case CHAR_VERTICAL_LINE: /* or | or ) */
3468 case CHAR_RIGHT_PARENTHESIS:
3469 *firstbyteptr = firstbyte;
3470 *reqbyteptr = reqbyte;
3471 *codeptr = code;
3472 *ptrptr = ptr;
3473 if (lengthptr != NULL)
3474 {
3475 if (OFLOW_MAX - *lengthptr < code - last_code)
3476 {
3477 *errorcodeptr = ERR20;
3478 goto FAILED;
3479 }
3480 *lengthptr += (int)(code - last_code); /* To include callout length */
3481 DPRINTF((">> end branch\n"));
3482 }
3483 return TRUE;
3484
3485
3486 /* ===================================================================*/
3487 /* Handle single-character metacharacters. In multiline mode, ^ disables
3488 the setting of any following char as a first character. */
3489
3490 case CHAR_CIRCUMFLEX_ACCENT:
3491 previous = NULL;
3492 if ((options & PCRE_MULTILINE) != 0)
3493 {
3494 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
3495 *code++ = OP_CIRCM;
3496 }
3497 else *code++ = OP_CIRC;
3498 break;
3499
3500 case CHAR_DOLLAR_SIGN:
3501 previous = NULL;
3502 *code++ = ((options & PCRE_MULTILINE) != 0)? OP_DOLLM : OP_DOLL;
3503 break;
3504
3505 /* There can never be a first char if '.' is first, whatever happens about
3506 repeats. The value of reqbyte doesn't change either. */
3507
3508 case CHAR_DOT:
3509 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
3510 zerofirstbyte = firstbyte;
3511 zeroreqbyte = reqbyte;
3512 previous = code;
3513 *code++ = ((options & PCRE_DOTALL) != 0)? OP_ALLANY: OP_ANY;
3514 break;
3515
3516
3517 /* ===================================================================*/
3518 /* Character classes. If the included characters are all < 256, we build a
3519 32-byte bitmap of the permitted characters, except in the special case
3520 where there is only one such character. For negated classes, we build the
3521 map as usual, then invert it at the end. However, we use a different opcode
3522 so that data characters > 255 can be handled correctly.
3523
3524 If the class contains characters outside the 0-255 range, a different
3525 opcode is compiled. It may optionally have a bit map for characters < 256,
3526 but those above are are explicitly listed afterwards. A flag byte tells
3527 whether the bitmap is present, and whether this is a negated class or not.
3528
3529 In JavaScript compatibility mode, an isolated ']' causes an error. In
3530 default (Perl) mode, it is treated as a data character. */
3531
3532 case CHAR_RIGHT_SQUARE_BRACKET:
3533 if ((cd->external_options & PCRE_JAVASCRIPT_COMPAT) != 0)
3534 {
3535 *errorcodeptr = ERR64;
3536 goto FAILED;
3537 }
3538 goto NORMAL_CHAR;
3539
3540 case CHAR_LEFT_SQUARE_BRACKET:
3541 previous = code;
3542
3543 /* PCRE supports POSIX class stuff inside a class. Perl gives an error if
3544 they are encountered at the top level, so we'll do that too. */
3545
3546 if ((ptr[1] == CHAR_COLON || ptr[1] == CHAR_DOT ||
3547 ptr[1] == CHAR_EQUALS_SIGN) &&
3548 check_posix_syntax(ptr, &tempptr))
3549 {
3550 *errorcodeptr = (ptr[1] == CHAR_COLON)? ERR13 : ERR31;
3551 goto FAILED;
3552 }
3553
3554 /* If the first character is '^', set the negation flag and skip it. Also,
3555 if the first few characters (either before or after ^) are \Q\E or \E we
3556 skip them too. This makes for compatibility with Perl. */
3557
3558 negate_class = FALSE;
3559 for (;;)
3560 {
3561 c = *(++ptr);
3562 if (c == CHAR_BACKSLASH)
3563 {
3564 if (ptr[1] == CHAR_E)
3565 ptr++;
3566 else if (strncmp((const char *)ptr+1,
3567 STR_Q STR_BACKSLASH STR_E, 3) == 0)
3568 ptr += 3;
3569 else
3570 break;
3571 }
3572 else if (!negate_class && c == CHAR_CIRCUMFLEX_ACCENT)
3573 negate_class = TRUE;
3574 else break;
3575 }
3576
3577 /* Empty classes are allowed in JavaScript compatibility mode. Otherwise,
3578 an initial ']' is taken as a data character -- the code below handles
3579 that. In JS mode, [] must always fail, so generate OP_FAIL, whereas
3580 [^] must match any character, so generate OP_ALLANY. */
3581
3582 if (c == CHAR_RIGHT_SQUARE_BRACKET &&
3583 (cd->external_options & PCRE_JAVASCRIPT_COMPAT) != 0)
3584 {
3585 *code++ = negate_class? OP_ALLANY : OP_FAIL;
3586 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
3587 zerofirstbyte = firstbyte;
3588 break;
3589 }
3590
3591 /* If a class contains a negative special such as \S, we need to flip the
3592 negation flag at the end, so that support for characters > 255 works
3593 correctly (they are all included in the class). */
3594
3595 should_flip_negation = FALSE;
3596
3597 /* Keep a count of chars with values < 256 so that we can optimize the case
3598 of just a single character (as long as it's < 256). However, For higher
3599 valued UTF-8 characters, we don't yet do any optimization. */
3600
3601 class_charcount = 0;
3602 class_lastchar = -1;
3603
3604 /* Initialize the 32-char bit map to all zeros. We build the map in a
3605 temporary bit of memory, in case the class contains only 1 character (less
3606 than 256), because in that case the compiled code doesn't use the bit map.
3607 */
3608
3609 memset(classbits, 0, 32 * sizeof(uschar));
3610
3611 #ifdef SUPPORT_UTF8
3612 class_utf8 = FALSE; /* No chars >= 256 */
3613 class_utf8data = code + LINK_SIZE + 2; /* For UTF-8 items */
3614 class_utf8data_base = class_utf8data; /* For resetting in pass 1 */
3615 #endif
3616
3617 /* Process characters until ] is reached. By writing this as a "do" it
3618 means that an initial ] is taken as a data character. At the start of the
3619 loop, c contains the first byte of the character. */
3620
3621 if (c != 0) do
3622 {
3623 const uschar *oldptr;
3624
3625 #ifdef SUPPORT_UTF8
3626 if (utf8 && c > 127)
3627 { /* Braces are required because the */
3628 GETCHARLEN(c, ptr, ptr); /* macro generates multiple statements */
3629 }
3630
3631 /* In the pre-compile phase, accumulate the length of any UTF-8 extra
3632 data and reset the pointer. This is so that very large classes that
3633 contain a zillion UTF-8 characters no longer overwrite the work space
3634 (which is on the stack). */
3635
3636 if (lengthptr != NULL)
3637 {
3638 *lengthptr += class_utf8data - class_utf8data_base;
3639 class_utf8data = class_utf8data_base;
3640 }
3641
3642 #endif
3643
3644 /* Inside \Q...\E everything is literal except \E */
3645
3646 if (inescq)
3647 {
3648 if (c == CHAR_BACKSLASH && ptr[1] == CHAR_E) /* If we are at \E */
3649 {
3650 inescq = FALSE; /* Reset literal state */
3651 ptr++; /* Skip the 'E' */
3652 continue; /* Carry on with next */
3653 }
3654 goto CHECK_RANGE; /* Could be range if \E follows */
3655 }
3656
3657 /* Handle POSIX class names. Perl allows a negation extension of the
3658 form [:^name:]. A square bracket that doesn't match the syntax is
3659 treated as a literal. We also recognize the POSIX constructions
3660 [.ch.] and [=ch=] ("collating elements") and fault them, as Perl
3661 5.6 and 5.8 do. */
3662
3663 if (c == CHAR_LEFT_SQUARE_BRACKET &&
3664 (ptr[1] == CHAR_COLON || ptr[1] == CHAR_DOT ||
3665 ptr[1] == CHAR_EQUALS_SIGN) && check_posix_syntax(ptr, &tempptr))
3666 {
3667 BOOL local_negate = FALSE;
3668 int posix_class, taboffset, tabopt;
3669 register const uschar *cbits = cd->cbits;
3670 uschar pbits[32];
3671
3672 if (ptr[1] != CHAR_COLON)
3673 {
3674 *errorcodeptr = ERR31;
3675 goto FAILED;
3676 }
3677
3678 ptr += 2;
3679 if (*ptr == CHAR_CIRCUMFLEX_ACCENT)
3680 {
3681 local_negate = TRUE;
3682 should_flip_negation = TRUE; /* Note negative special */
3683 ptr++;
3684 }
3685
3686 posix_class = check_posix_name(ptr, (int)(tempptr - ptr));
3687 if (posix_class < 0)
3688 {
3689 *errorcodeptr = ERR30;
3690 goto FAILED;
3691 }
3692
3693 /* If matching is caseless, upper and lower are converted to
3694 alpha. This relies on the fact that the class table starts with
3695 alpha, lower, upper as the first 3 entries. */
3696
3697 if ((options & PCRE_CASELESS) != 0 && posix_class <= 2)
3698 posix_class = 0;
3699
3700 /* When PCRE_UCP is set, some of the POSIX classes are converted to
3701 different escape sequences that use Unicode properties. */
3702
3703 #ifdef SUPPORT_UCP
3704 if ((options & PCRE_UCP) != 0)
3705 {
3706 int pc = posix_class + ((local_negate)? POSIX_SUBSIZE/2 : 0);
3707 if (posix_substitutes[pc] != NULL)
3708 {
3709 nestptr = tempptr + 1;
3710 ptr = posix_substitutes[pc] - 1;
3711 continue;
3712 }
3713 }
3714 #endif
3715 /* In the non-UCP case, we build the bit map for the POSIX class in a
3716 chunk of local store because we may be adding and subtracting from it,
3717 and we don't want to subtract bits that may be in the main map already.
3718 At the end we or the result into the bit map that is being built. */
3719
3720 posix_class *= 3;
3721
3722 /* Copy in the first table (always present) */
3723
3724 memcpy(pbits, cbits + posix_class_maps[posix_class],
3725 32 * sizeof(uschar));
3726
3727 /* If there is a second table, add or remove it as required. */
3728
3729 taboffset = posix_class_maps[posix_class + 1];
3730 tabopt = posix_class_maps[posix_class + 2];
3731
3732 if (taboffset >= 0)
3733 {
3734 if (tabopt >= 0)
3735 for (c = 0; c < 32; c++) pbits[c] |= cbits[c + taboffset];
3736 else
3737 for (c = 0; c < 32; c++) pbits[c] &= ~cbits[c + taboffset];
3738 }
3739
3740 /* Not see if we need to remove any special characters. An option
3741 value of 1 removes vertical space and 2 removes underscore. */
3742
3743 if (tabopt < 0) tabopt = -tabopt;
3744 if (tabopt == 1) pbits[1] &= ~0x3c;
3745 else if (tabopt == 2) pbits[11] &= 0x7f;
3746
3747 /* Add the POSIX table or its complement into the main table that is
3748 being built and we are done. */
3749
3750 if (local_negate)
3751 for (c = 0; c < 32; c++) classbits[c] |= ~pbits[c];
3752 else
3753 for (c = 0; c < 32; c++) classbits[c] |= pbits[c];
3754
3755 ptr = tempptr + 1;
3756 class_charcount = 10; /* Set > 1; assumes more than 1 per class */
3757 continue; /* End of POSIX syntax handling */
3758 }
3759
3760 /* Backslash may introduce a single character, or it may introduce one
3761 of the specials, which just set a flag. The sequence \b is a special
3762 case. Inside a class (and only there) it is treated as backspace. We
3763 assume that other escapes have more than one character in them, so set
3764 class_charcount bigger than one. Unrecognized escapes fall through and
3765 are either treated as literal characters (by default), or are faulted if
3766 PCRE_EXTRA is set. */
3767
3768 if (c == CHAR_BACKSLASH)
3769 {
3770 c = check_escape(&ptr, errorcodeptr, cd->bracount, options, TRUE);
3771 if (*errorcodeptr != 0) goto FAILED;
3772
3773 if (-c == ESC_b) c = CHAR_BS; /* \b is backspace in a class */
3774 else if (-c == ESC_N) /* \N is not supported in a class */
3775 {
3776 *errorcodeptr = ERR71;
3777 goto FAILED;
3778 }
3779 else if (-c == ESC_Q) /* Handle start of quoted string */
3780 {
3781 if (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
3782 {
3783 ptr += 2; /* avoid empty string */
3784 }
3785 else inescq = TRUE;
3786 continue;
3787 }
3788 else if (-c == ESC_E) continue; /* Ignore orphan \E */
3789
3790 if (c < 0)
3791 {
3792 register const uschar *cbits = cd->cbits;
3793 class_charcount += 2; /* Greater than 1 is what matters */
3794
3795 switch (-c)
3796 {
3797 #ifdef SUPPORT_UCP
3798 case ESC_du: /* These are the values given for \d etc */
3799 case ESC_DU: /* when PCRE_UCP is set. We replace the */
3800 case ESC_wu: /* escape sequence with an appropriate \p */
3801 case ESC_WU: /* or \P to test Unicode properties instead */
3802 case ESC_su: /* of the default ASCII testing. */
3803 case ESC_SU:
3804 nestptr = ptr;
3805 ptr = substitutes[-c - ESC_DU] - 1; /* Just before substitute */
3806 class_charcount -= 2; /* Undo! */
3807 continue;
3808 #endif
3809 case ESC_d:
3810 for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_digit];
3811 continue;
3812
3813 case ESC_D:
3814 should_flip_negation = TRUE;
3815 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_digit];
3816 continue;
3817
3818 case ESC_w:
3819 for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_word];
3820 continue;
3821
3822 case ESC_W:
3823 should_flip_negation = TRUE;
3824 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_word];
3825 continue;
3826
3827 /* Perl 5.004 onwards omits VT from \s, but we must preserve it
3828 if it was previously set by something earlier in the character
3829 class. */
3830
3831 case ESC_s:
3832 classbits[0] |= cbits[cbit_space];
3833 classbits[1] |= cbits[cbit_space+1] & ~0x08;
3834 for (c = 2; c < 32; c++) classbits[c] |= cbits[c+cbit_space];
3835 continue;
3836
3837 case ESC_S:
3838 should_flip_negation = TRUE;
3839 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_space];
3840 classbits[1] |= 0x08; /* Perl 5.004 onwards omits VT from \s */
3841 continue;
3842
3843 case ESC_h:
3844 SETBIT(classbits, 0x09); /* VT */
3845 SETBIT(classbits, 0x20); /* SPACE */
3846 SETBIT(classbits, 0xa0); /* NSBP */
3847 #ifdef SUPPORT_UTF8
3848 if (utf8)
3849 {
3850 class_utf8 = TRUE;
3851 *class_utf8data++ = XCL_SINGLE;
3852 class_utf8data += _pcre_ord2utf8(0x1680, class_utf8data);
3853 *class_utf8data++ = XCL_SINGLE;
3854 class_utf8data += _pcre_ord2utf8(0x180e, class_utf8data);
3855 *class_utf8data++ = XCL_RANGE;
3856 class_utf8data += _pcre_ord2utf8(0x2000, class_utf8data);
3857 class_utf8data += _pcre_ord2utf8(0x200A, class_utf8data);
3858 *class_utf8data++ = XCL_SINGLE;
3859 class_utf8data += _pcre_ord2utf8(0x202f, class_utf8data);
3860 *class_utf8data++ = XCL_SINGLE;
3861 class_utf8data += _pcre_ord2utf8(0x205f, class_utf8data);
3862 *class_utf8data++ = XCL_SINGLE;
3863 class_utf8data += _pcre_ord2utf8(0x3000, class_utf8data);
3864 }
3865 #endif
3866 continue;
3867
3868 case ESC_H:
3869 for (c = 0; c < 32; c++)
3870 {
3871 int x = 0xff;
3872 switch (c)
3873 {
3874 case 0x09/8: x ^= 1 << (0x09%8); break;
3875 case 0x20/8: x ^= 1 << (0x20%8); break;
3876 case 0xa0/8: x ^= 1 << (0xa0%8); break;
3877 default: break;
3878 }
3879 classbits[c] |= x;
3880 }
3881
3882 #ifdef SUPPORT_UTF8
3883 if (utf8)
3884 {
3885 class_utf8 = TRUE;
3886 *class_utf8data++ = XCL_RANGE;
3887 class_utf8data += _pcre_ord2utf8(0x0100, class_utf8data);
3888 class_utf8data += _pcre_ord2utf8(0x167f, class_utf8data);
3889 *class_utf8data++ = XCL_RANGE;
3890 class_utf8data += _pcre_ord2utf8(0x1681, class_utf8data);
3891 class_utf8data += _pcre_ord2utf8(0x180d, class_utf8data);
3892 *class_utf8data++ = XCL_RANGE;
3893 class_utf8data += _pcre_ord2utf8(0x180f, class_utf8data);
3894 class_utf8data += _pcre_ord2utf8(0x1fff, class_utf8data);
3895 *class_utf8data++ = XCL_RANGE;
3896 class_utf8data += _pcre_ord2utf8(0x200B, class_utf8data);
3897 class_utf8data += _pcre_ord2utf8(0x202e, class_utf8data);
3898 *class_utf8data++ = XCL_RANGE;
3899 class_utf8data += _pcre_ord2utf8(0x2030, class_utf8data);
3900 class_utf8data += _pcre_ord2utf8(0x205e, class_utf8data);
3901 *class_utf8data++ = XCL_RANGE;
3902 class_utf8data += _pcre_ord2utf8(0x2060, class_utf8data);
3903 class_utf8data += _pcre_ord2utf8(0x2fff, class_utf8data);
3904 *class_utf8data++ = XCL_RANGE;
3905 class_utf8data += _pcre_ord2utf8(0x3001, class_utf8data);
3906 class_utf8data += _pcre_ord2utf8(0x7fffffff, class_utf8data);
3907 }
3908 #endif
3909 continue;
3910
3911 case ESC_v:
3912 SETBIT(classbits, 0x0a); /* LF */
3913 SETBIT(classbits, 0x0b); /* VT */
3914 SETBIT(classbits, 0x0c); /* FF */
3915 SETBIT(classbits, 0x0d); /* CR */
3916 SETBIT(classbits, 0x85); /* NEL */
3917 #ifdef SUPPORT_UTF8
3918 if (utf8)
3919 {
3920 class_utf8 = TRUE;
3921 *class_utf8data++ = XCL_RANGE;
3922 class_utf8data += _pcre_ord2utf8(0x2028, class_utf8data);
3923 class_utf8data += _pcre_ord2utf8(0x2029, class_utf8data);
3924 }
3925 #endif
3926 continue;
3927
3928 case ESC_V:
3929 for (c = 0; c < 32; c++)
3930 {
3931 int x = 0xff;
3932 switch (c)
3933 {
3934 case 0x0a/8: x ^= 1 << (0x0a%8);
3935 x ^= 1 << (0x0b%8);
3936 x ^= 1 << (0x0c%8);
3937 x ^= 1 << (0x0d%8);
3938 break;
3939 case 0x85/8: x ^= 1 << (0x85%8); break;
3940 default: break;
3941 }
3942 classbits[c] |= x;
3943 }
3944
3945 #ifdef SUPPORT_UTF8
3946 if (utf8)
3947 {
3948 class_utf8 = TRUE;
3949 *class_utf8data++ = XCL_RANGE;
3950 class_utf8data += _pcre_ord2utf8(0x0100, class_utf8data);
3951 class_utf8data += _pcre_ord2utf8(0x2027, class_utf8data);
3952 *class_utf8data++ = XCL_RANGE;
3953 class_utf8data += _pcre_ord2utf8(0x2029, class_utf8data);
3954 class_utf8data += _pcre_ord2utf8(0x7fffffff, class_utf8data);
3955 }
3956 #endif
3957 continue;
3958
3959 #ifdef SUPPORT_UCP
3960 case ESC_p:
3961 case ESC_P:
3962 {
3963 BOOL negated;
3964 int pdata;
3965 int ptype = get_ucp(&ptr, &negated, &pdata, errorcodeptr);
3966 if (ptype < 0) goto FAILED;
3967 class_utf8 = TRUE;
3968 *class_utf8data++ = ((-c == ESC_p) != negated)?
3969 XCL_PROP : XCL_NOTPROP;
3970 *class_utf8data++ = ptype;
3971 *class_utf8data++ = pdata;
3972 class_charcount -= 2; /* Not a < 256 character */
3973 continue;
3974 }
3975 #endif
3976 /* Unrecognized escapes are faulted if PCRE is running in its
3977 strict mode. By default, for compatibility with Perl, they are
3978 treated as literals. */
3979
3980 default:
3981 if ((options & PCRE_EXTRA) != 0)
3982 {
3983 *errorcodeptr = ERR7;
3984 goto FAILED;
3985 }
3986 class_charcount -= 2; /* Undo the default count from above */
3987 c = *ptr; /* Get the final character and fall through */
3988 break;
3989 }
3990 }
3991
3992 /* Fall through if we have a single character (c >= 0). This may be
3993 greater than 256 in UTF-8 mode. */
3994
3995 } /* End of backslash handling */
3996
3997 /* A single character may be followed by '-' to form a range. However,
3998 Perl does not permit ']' to be the end of the range. A '-' character
3999 at the end is treated as a literal. Perl ignores orphaned \E sequences
4000 entirely. The code for handling \Q and \E is messy. */
4001
4002 CHECK_RANGE:
4003 while (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
4004 {
4005 inescq = FALSE;
4006 ptr += 2;
4007 }
4008
4009 oldptr = ptr;
4010
4011 /* Remember \r or \n */
4012
4013 if (c == CHAR_CR || c == CHAR_NL) cd->external_flags |= PCRE_HASCRORLF;
4014
4015 /* Check for range */
4016
4017 if (!inescq && ptr[1] == CHAR_MINUS)
4018 {
4019 int d;
4020 ptr += 2;
4021 while (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_E) ptr += 2;
4022
4023 /* If we hit \Q (not followed by \E) at this point, go into escaped
4024 mode. */
4025
4026 while (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_Q)
4027 {
4028 ptr += 2;
4029 if (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_E)
4030 { ptr += 2; continue; }
4031 inescq = TRUE;
4032 break;
4033 }
4034
4035 if (*ptr == 0 || (!inescq && *ptr == CHAR_RIGHT_SQUARE_BRACKET))
4036 {
4037 ptr = oldptr;
4038 goto LONE_SINGLE_CHARACTER;
4039 }
4040
4041 #ifdef SUPPORT_UTF8
4042 if (utf8)
4043 { /* Braces are required because the */
4044 GETCHARLEN(d, ptr, ptr); /* macro generates multiple statements */
4045 }
4046 else
4047 #endif
4048 d = *ptr; /* Not UTF-8 mode */
4049
4050 /* The second part of a range can be a single-character escape, but
4051 not any of the other escapes. Perl 5.6 treats a hyphen as a literal
4052 in such circumstances. */
4053
4054 if (!inescq && d == CHAR_BACKSLASH)
4055 {
4056 d = check_escape(&ptr, errorcodeptr, cd->bracount, options, TRUE);
4057 if (*errorcodeptr != 0) goto FAILED;
4058
4059 /* \b is backspace; any other special means the '-' was literal */
4060
4061 if (d < 0)
4062 {
4063 if (d == -ESC_b) d = CHAR_BS; else
4064 {
4065 ptr = oldptr;
4066 goto LONE_SINGLE_CHARACTER; /* A few lines below */
4067 }
4068 }
4069 }
4070
4071 /* Check that the two values are in the correct order. Optimize
4072 one-character ranges */
4073
4074 if (d < c)
4075 {
4076 *errorcodeptr = ERR8;
4077 goto FAILED;
4078 }
4079
4080 if (d == c) goto LONE_SINGLE_CHARACTER; /* A few lines below */
4081
4082 /* Remember \r or \n */
4083
4084 if (d == CHAR_CR || d == CHAR_NL) cd->external_flags |= PCRE_HASCRORLF;
4085
4086 /* In UTF-8 mode, if the upper limit is > 255, or > 127 for caseless
4087 matching, we have to use an XCLASS with extra data items. Caseless
4088 matching for characters > 127 is available only if UCP support is
4089 available. */
4090
4091 #ifdef SUPPORT_UTF8
4092 if (utf8 && (d > 255 || ((options & PCRE_CASELESS) != 0 && d > 127)))
4093 {
4094 class_utf8 = TRUE;
4095
4096 /* With UCP support, we can find the other case equivalents of
4097 the relevant characters. There may be several ranges. Optimize how
4098 they fit with the basic range. */
4099
4100 #ifdef SUPPORT_UCP
4101 if ((options & PCRE_CASELESS) != 0)
4102 {
4103 unsigned int occ, ocd;
4104 unsigned int cc = c;
4105 unsigned int origd = d;
4106 while (get_othercase_range(&cc, origd, &occ, &ocd))
4107 {
4108 if (occ >= (unsigned int)c &&
4109 ocd <= (unsigned int)d)
4110 continue; /* Skip embedded ranges */
4111
4112 if (occ < (unsigned int)c &&
4113 ocd >= (unsigned int)c - 1) /* Extend the basic range */
4114 { /* if there is overlap, */
4115 c = occ; /* noting that if occ < c */
4116 continue; /* we can't have ocd > d */
4117 } /* because a subrange is */
4118 if (ocd > (unsigned int)d &&
4119 occ <= (unsigned int)d + 1) /* always shorter than */
4120 { /* the basic range. */
4121 d = ocd;
4122 continue;
4123 }
4124
4125 if (occ == ocd)
4126 {
4127 *class_utf8data++ = XCL_SINGLE;
4128 }
4129 else
4130 {
4131 *class_utf8data++ = XCL_RANGE;
4132 class_utf8data += _pcre_ord2utf8(occ, class_utf8data);
4133 }
4134 class_utf8data += _pcre_ord2utf8(ocd, class_utf8data);
4135 }
4136 }
4137 #endif /* SUPPORT_UCP */
4138
4139 /* Now record the original range, possibly modified for UCP caseless
4140 overlapping ranges. */
4141
4142 *class_utf8data++ = XCL_RANGE;
4143 class_utf8data += _pcre_ord2utf8(c, class_utf8data);
4144 class_utf8data += _pcre_ord2utf8(d, class_utf8data);
4145
4146 /* With UCP support, we are done. Without UCP support, there is no
4147 caseless matching for UTF-8 characters > 127; we can use the bit map
4148 for the smaller ones. */
4149
4150 #ifdef SUPPORT_UCP
4151 continue; /* With next character in the class */
4152 #else
4153 if ((options & PCRE_CASELESS) == 0 || c > 127) continue;
4154
4155 /* Adjust upper limit and fall through to set up the map */
4156
4157 d = 127;
4158
4159 #endif /* SUPPORT_UCP */
4160 }
4161 #endif /* SUPPORT_UTF8 */
4162
4163 /* We use the bit map for all cases when not in UTF-8 mode; else
4164 ranges that lie entirely within 0-127 when there is UCP support; else
4165 for partial ranges without UCP support. */
4166
4167 class_charcount += d - c + 1;
4168 class_lastchar = d;
4169
4170 /* We can save a bit of time by skipping this in the pre-compile. */
4171
4172 if (lengthptr == NULL) for (; c <= d; c++)
4173 {
4174 classbits[c/8] |= (1 << (c&7));
4175 if ((options & PCRE_CASELESS) != 0)
4176 {
4177 int uc = cd->fcc[c]; /* flip case */
4178 classbits[uc/8] |= (1 << (uc&7));
4179 }
4180 }
4181
4182 continue; /* Go get the next char in the class */
4183 }
4184
4185 /* Handle a lone single character - we can get here for a normal
4186 non-escape char, or after \ that introduces a single character or for an
4187 apparent range that isn't. */
4188
4189 LONE_SINGLE_CHARACTER:
4190
4191 /* Handle a character that cannot go in the bit map */
4192
4193 #ifdef SUPPORT_UTF8
4194 if (utf8 && (c > 255 || ((options & PCRE_CASELESS) != 0 && c > 127)))
4195 {
4196 class_utf8 = TRUE;
4197 *class_utf8data++ = XCL_SINGLE;
4198 class_utf8data += _pcre_ord2utf8(c, class_utf8data);
4199
4200 #ifdef SUPPORT_UCP
4201 if ((options & PCRE_CASELESS) != 0)
4202 {
4203 unsigned int othercase;
4204 if ((othercase = UCD_OTHERCASE(c)) != c)
4205 {
4206 *class_utf8data++ = XCL_SINGLE;
4207 class_utf8data += _pcre_ord2utf8(othercase, class_utf8data);
4208 }
4209 }
4210 #endif /* SUPPORT_UCP */
4211
4212 }
4213 else
4214 #endif /* SUPPORT_UTF8 */
4215
4216 /* Handle a single-byte character */
4217 {
4218 classbits[c/8] |= (1 << (c&7));
4219 if ((options & PCRE_CASELESS) != 0)
4220 {
4221 c = cd->fcc[c]; /* flip case */
4222 classbits[c/8] |= (1 << (c&7));
4223 }
4224 class_charcount++;
4225 class_lastchar = c;
4226 }
4227 }
4228
4229 /* Loop until ']' reached. This "while" is the end of the "do" far above.
4230 If we are at the end of an internal nested string, revert to the outer
4231 string. */
4232
4233 while (((c = *(++ptr)) != 0 ||
4234 (nestptr != NULL &&
4235 (ptr = nestptr, nestptr = NULL, c = *(++ptr)) != 0)) &&
4236 (c != CHAR_RIGHT_SQUARE_BRACKET || inescq));
4237
4238 /* Check for missing terminating ']' */
4239
4240 if (c == 0)
4241 {
4242 *errorcodeptr = ERR6;
4243 goto FAILED;
4244 }
4245
4246 /* If class_charcount is 1, we saw precisely one character whose value is
4247 less than 256. As long as there were no characters >= 128 and there was no
4248 use of \p or \P, in other words, no use of any XCLASS features, we can
4249 optimize.
4250
4251 In UTF-8 mode, we can optimize the negative case only if there were no
4252 characters >= 128 because OP_NOT and the related opcodes like OP_NOTSTAR
4253 operate on single-bytes characters only. This is an historical hangover.
4254 Maybe one day we can tidy these opcodes to handle multi-byte characters.
4255
4256 The optimization throws away the bit map. We turn the item into a
4257 1-character OP_CHAR[I] if it's positive, or OP_NOT[I] if it's negative.
4258 Note that OP_NOT[I] does not support multibyte characters. In the positive
4259 case, it can cause firstbyte to be set. Otherwise, there can be no first
4260 char if this item is first, whatever repeat count may follow. In the case
4261 of reqbyte, save the previous value for reinstating. */
4262
4263 #ifdef SUPPORT_UTF8
4264 if (class_charcount == 1 && !class_utf8 &&
4265 (!utf8 || !negate_class || class_lastchar < 128))
4266 #else
4267 if (class_charcount == 1)
4268 #endif
4269 {
4270 zeroreqbyte = reqbyte;
4271
4272 /* The OP_NOT[I] opcodes work on one-byte characters only. */
4273
4274 if (negate_class)
4275 {
4276 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
4277 zerofirstbyte = firstbyte;
4278 *code++ = ((options & PCRE_CASELESS) != 0)? OP_NOTI: OP_NOT;
4279 *code++ = class_lastchar;
4280 break;
4281 }
4282
4283 /* For a single, positive character, get the value into mcbuffer, and
4284 then we can handle this with the normal one-character code. */
4285
4286 #ifdef SUPPORT_UTF8
4287 if (utf8 && class_lastchar > 127)
4288 mclength = _pcre_ord2utf8(class_lastchar, mcbuffer);
4289 else
4290 #endif
4291 {
4292 mcbuffer[0] = class_lastchar;
4293 mclength = 1;
4294 }
4295 goto ONE_CHAR;
4296 } /* End of 1-char optimization */
4297
4298 /* The general case - not the one-char optimization. If this is the first
4299 thing in the branch, there can be no first char setting, whatever the
4300 repeat count. Any reqbyte setting must remain unchanged after any kind of
4301 repeat. */
4302
4303 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
4304 zerofirstbyte = firstbyte;
4305 zeroreqbyte = reqbyte;
4306
4307 /* If there are characters with values > 255, we have to compile an
4308 extended class, with its own opcode, unless there was a negated special
4309 such as \S in the class, and PCRE_UCP is not set, because in that case all
4310 characters > 255 are in the class, so any that were explicitly given as
4311 well can be ignored. If (when there are explicit characters > 255 that must
4312 be listed) there are no characters < 256, we can omit the bitmap in the
4313 actual compiled code. */
4314
4315 #ifdef SUPPORT_UTF8
4316 if (class_utf8 && (!should_flip_negation || (options & PCRE_UCP) != 0))
4317 {
4318 *class_utf8data++ = XCL_END; /* Marks the end of extra data */
4319 *code++ = OP_XCLASS;
4320 code += LINK_SIZE;
4321 *code = negate_class? XCL_NOT : 0;
4322
4323 /* If the map is required, move up the extra data to make room for it;
4324 otherwise just move the code pointer to the end of the extra data. */
4325
4326 if (class_charcount > 0)
4327 {
4328 *code++ |= XCL_MAP;
4329 memmove(code + 32, code, class_utf8data - code);
4330 memcpy(code, classbits, 32);
4331 code = class_utf8data + 32;
4332 }
4333 else code = class_utf8data;
4334
4335 /* Now fill in the complete length of the item */
4336
4337 PUT(previous, 1, code - previous);
4338 break; /* End of class handling */
4339 }
4340 #endif
4341
4342 /* If there are no characters > 255, or they are all to be included or
4343 excluded, set the opcode to OP_CLASS or OP_NCLASS, depending on whether the
4344 whole class was negated and whether there were negative specials such as \S
4345 (non-UCP) in the class. Then copy the 32-byte map into the code vector,
4346 negating it if necessary. */
4347
4348 *code++ = (negate_class == should_flip_negation) ? OP_CLASS : OP_NCLASS;
4349 if (negate_class)
4350 {
4351 if (lengthptr == NULL) /* Save time in the pre-compile phase */
4352 for (c = 0; c < 32; c++) code[c] = ~classbits[c];
4353 }
4354 else
4355 {
4356 memcpy(code, classbits, 32);
4357 }
4358 code += 32;
4359 break;
4360
4361
4362 /* ===================================================================*/
4363 /* Various kinds of repeat; '{' is not necessarily a quantifier, but this
4364 has been tested above. */
4365
4366 case CHAR_LEFT_CURLY_BRACKET:
4367 if (!is_quantifier) goto NORMAL_CHAR;
4368 ptr = read_repeat_counts(ptr+1, &repeat_min, &repeat_max, errorcodeptr);
4369 if (*errorcodeptr != 0) goto FAILED;
4370 goto REPEAT;
4371
4372 case CHAR_ASTERISK:
4373 repeat_min = 0;
4374 repeat_max = -1;
4375 goto REPEAT;
4376
4377 case CHAR_PLUS:
4378 repeat_min = 1;
4379 repeat_max = -1;
4380 goto REPEAT;
4381
4382 case CHAR_QUESTION_MARK:
4383 repeat_min = 0;
4384 repeat_max = 1;
4385
4386 REPEAT:
4387 if (previous == NULL)
4388 {
4389 *errorcodeptr = ERR9;
4390 goto FAILED;
4391 }
4392
4393 if (repeat_min == 0)
4394 {
4395 firstbyte = zerofirstbyte; /* Adjust for zero repeat */
4396 reqbyte = zeroreqbyte; /* Ditto */
4397 }
4398
4399 /* Remember whether this is a variable length repeat */
4400
4401 reqvary = (repeat_min == repeat_max)? 0 : REQ_VARY;
4402
4403 op_type = 0; /* Default single-char op codes */
4404 possessive_quantifier = FALSE; /* Default not possessive quantifier */
4405
4406 /* Save start of previous item, in case we have to move it up in order to
4407 insert something before it. */
4408
4409 tempcode = previous;
4410
4411 /* If the next character is '+', we have a possessive quantifier. This
4412 implies greediness, whatever the setting of the PCRE_UNGREEDY option.
4413 If the next character is '?' this is a minimizing repeat, by default,
4414 but if PCRE_UNGREEDY is set, it works the other way round. We change the
4415 repeat type to the non-default. */
4416
4417 if (ptr[1] == CHAR_PLUS)
4418 {
4419 repeat_type = 0; /* Force greedy */
4420 possessive_quantifier = TRUE;
4421 ptr++;
4422 }
4423 else if (ptr[1] == CHAR_QUESTION_MARK)
4424 {
4425 repeat_type = greedy_non_default;
4426 ptr++;
4427 }
4428 else repeat_type = greedy_default;
4429
4430 /* If previous was a recursion call, wrap it in atomic brackets so that
4431 previous becomes the atomic group. All recursions were so wrapped in the
4432 past, but it no longer happens for non-repeated recursions. In fact, the
4433 repeated ones could be re-implemented independently so as not to need this,
4434 but for the moment we rely on the code for repeating groups. */
4435
4436 if (*previous == OP_RECURSE)
4437 {
4438 memmove(previous + 1 + LINK_SIZE, previous, 1 + LINK_SIZE);
4439 *previous = OP_ONCE;
4440 PUT(previous, 1, 2 + 2*LINK_SIZE);
4441 previous[2 + 2*LINK_SIZE] = OP_KET;
4442 PUT(previous, 3 + 2*LINK_SIZE, 2 + 2*LINK_SIZE);
4443 code += 2 + 2 * LINK_SIZE;
4444 length_prevgroup = 3 + 3*LINK_SIZE;
4445
4446 /* When actually compiling, we need to check whether this was a forward
4447 reference, and if so, adjust the offset. */
4448
4449 if (lengthptr == NULL && cd->hwm >= cd->start_workspace + LINK_SIZE)
4450 {
4451 int offset = GET(cd->hwm, -LINK_SIZE);
4452 if (offset == previous + 1 - cd->start_code)
4453 PUT(cd->hwm, -LINK_SIZE, offset + 1 + LINK_SIZE);
4454 }
4455 }
4456
4457 /* Now handle repetition for the different types of item. */
4458
4459 /* If previous was a character match, abolish the item and generate a
4460 repeat item instead. If a char item has a minumum of more than one, ensure
4461 that it is set in reqbyte - it might not be if a sequence such as x{3} is
4462 the first thing in a branch because the x will have gone into firstbyte
4463 instead. */
4464
4465 if (*previous == OP_CHAR || *previous == OP_CHARI)
4466 {
4467 op_type = (*previous == OP_CHAR)? 0 : OP_STARI - OP_STAR;
4468
4469 /* Deal with UTF-8 characters that take up more than one byte. It's
4470 easier to write this out separately than try to macrify it. Use c to
4471 hold the length of the character in bytes, plus 0x80 to flag that it's a
4472 length rather than a small character. */
4473
4474 #ifdef SUPPORT_UTF8
4475 if (utf8 && (code[-1] & 0x80) != 0)
4476 {
4477 uschar *lastchar = code - 1;
4478 while((*lastchar & 0xc0) == 0x80) lastchar--;
4479 c = code - lastchar; /* Length of UTF-8 character */
4480 memcpy(utf8_char, lastchar, c); /* Save the char */
4481 c |= 0x80; /* Flag c as a length */
4482 }
4483 else
4484 #endif
4485
4486 /* Handle the case of a single byte - either with no UTF8 support, or
4487 with UTF-8 disabled, or for a UTF-8 character < 128. */
4488
4489 {
4490 c = code[-1];
4491 if (repeat_min > 1) reqbyte = c | req_caseopt | cd->req_varyopt;
4492 }
4493
4494 /* If the repetition is unlimited, it pays to see if the next thing on
4495 the line is something that cannot possibly match this character. If so,
4496 automatically possessifying this item gains some performance in the case
4497 where the match fails. */
4498
4499 if (!possessive_quantifier &&
4500 repeat_max < 0 &&
4501 check_auto_possessive(previous, utf8, ptr + 1, options, cd))
4502 {
4503 repeat_type = 0; /* Force greedy */
4504 possessive_quantifier = TRUE;
4505 }
4506
4507 goto OUTPUT_SINGLE_REPEAT; /* Code shared with single character types */
4508 }
4509
4510 /* If previous was a single negated character ([^a] or similar), we use
4511 one of the special opcodes, replacing it. The code is shared with single-
4512 character repeats by setting opt_type to add a suitable offset into
4513 repeat_type. We can also test for auto-possessification. OP_NOT and OP_NOTI
4514 are currently used only for single-byte chars. */
4515
4516 else if (*previous == OP_NOT || *previous == OP_NOTI)
4517 {
4518 op_type = ((*previous == OP_NOT)? OP_NOTSTAR : OP_NOTSTARI) - OP_STAR;
4519 c = previous[1];
4520 if (!possessive_quantifier &&
4521 repeat_max < 0 &&
4522 check_auto_possessive(previous, utf8, ptr + 1, options, cd))
4523 {
4524 repeat_type = 0; /* Force greedy */
4525 possessive_quantifier = TRUE;
4526 }
4527 goto OUTPUT_SINGLE_REPEAT;
4528 }
4529
4530 /* If previous was a character type match (\d or similar), abolish it and
4531 create a suitable repeat item. The code is shared with single-character
4532 repeats by setting op_type to add a suitable offset into repeat_type. Note
4533 the the Unicode property types will be present only when SUPPORT_UCP is
4534 defined, but we don't wrap the little bits of code here because it just
4535 makes it horribly messy. */
4536
4537 else if (*previous < OP_EODN)
4538 {
4539 uschar *oldcode;
4540 int prop_type, prop_value;
4541 op_type = OP_TYPESTAR - OP_STAR; /* Use type opcodes */
4542 c = *previous;
4543
4544 if (!possessive_quantifier &&
4545 repeat_max < 0 &&
4546 check_auto_possessive(previous, utf8, ptr + 1, options, cd))
4547 {
4548 repeat_type = 0; /* Force greedy */
4549 possessive_quantifier = TRUE;
4550 }
4551
4552 OUTPUT_SINGLE_REPEAT:
4553 if (*previous == OP_PROP || *previous == OP_NOTPROP)
4554 {
4555 prop_type = previous[1];
4556 prop_value = previous[2];
4557 }
4558 else prop_type = prop_value = -1;
4559
4560 oldcode = code;
4561 code = previous; /* Usually overwrite previous item */
4562
4563 /* If the maximum is zero then the minimum must also be zero; Perl allows
4564 this case, so we do too - by simply omitting the item altogether. */
4565
4566 if (repeat_max == 0) goto END_REPEAT;
4567
4568 /*--------------------------------------------------------------------*/
4569 /* This code is obsolete from release 8.00; the restriction was finally
4570 removed: */
4571
4572 /* All real repeats make it impossible to handle partial matching (maybe
4573 one day we will be able to remove this restriction). */
4574
4575 /* if (repeat_max != 1) cd->external_flags |= PCRE_NOPARTIAL; */
4576 /*--------------------------------------------------------------------*/
4577
4578 /* Combine the op_type with the repeat_type */
4579
4580 repeat_type += op_type;
4581
4582 /* A minimum of zero is handled either as the special case * or ?, or as
4583 an UPTO, with the maximum given. */
4584
4585 if (repeat_min == 0)
4586 {
4587 if (repeat_max == -1) *code++ = OP_STAR + repeat_type;
4588 else if (repeat_max == 1) *code++ = OP_QUERY + repeat_type;
4589 else
4590 {
4591 *code++ = OP_UPTO + repeat_type;
4592 PUT2INC(code, 0, repeat_max);
4593 }
4594 }
4595
4596 /* A repeat minimum of 1 is optimized into some special cases. If the
4597 maximum is unlimited, we use OP_PLUS. Otherwise, the original item is
4598 left in place and, if the maximum is greater than 1, we use OP_UPTO with
4599 one less than the maximum. */
4600
4601 else if (repeat_min == 1)
4602 {
4603 if (repeat_max == -1)
4604 *code++ = OP_PLUS + repeat_type;
4605 else
4606 {
4607 code = oldcode; /* leave previous item in place */
4608 if (repeat_max == 1) goto END_REPEAT;
4609 *code++ = OP_UPTO + repeat_type;
4610 PUT2INC(code, 0, repeat_max - 1);
4611 }
4612 }
4613
4614 /* The case {n,n} is just an EXACT, while the general case {n,m} is
4615 handled as an EXACT followed by an UPTO. */
4616
4617 else
4618 {
4619 *code++ = OP_EXACT + op_type; /* NB EXACT doesn't have repeat_type */
4620 PUT2INC(code, 0, repeat_min);
4621
4622 /* If the maximum is unlimited, insert an OP_STAR. Before doing so,
4623 we have to insert the character for the previous code. For a repeated
4624 Unicode property match, there are two extra bytes that define the
4625 required property. In UTF-8 mode, long characters have their length in
4626 c, with the 0x80 bit as a flag. */
4627
4628 if (repeat_max < 0)
4629 {
4630 #ifdef SUPPORT_UTF8
4631 if (utf8 && c >= 128)
4632 {
4633 memcpy(code, utf8_char, c & 7);
4634 code += c & 7;
4635 }
4636 else
4637 #endif
4638 {
4639 *code++ = c;
4640 if (prop_type >= 0)
4641 {
4642 *code++ = prop_type;
4643 *code++ = prop_value;
4644 }
4645 }
4646 *code++ = OP_STAR + repeat_type;
4647 }
4648
4649 /* Else insert an UPTO if the max is greater than the min, again
4650 preceded by the character, for the previously inserted code. If the
4651 UPTO is just for 1 instance, we can use QUERY instead. */
4652
4653 else if (repeat_max != repeat_min)
4654 {
4655 #ifdef SUPPORT_UTF8
4656 if (utf8 && c >= 128)
4657 {
4658 memcpy(code, utf8_char, c & 7);
4659 code += c & 7;
4660 }
4661 else
4662 #endif
4663 *code++ = c;
4664 if (prop_type >= 0)
4665 {
4666 *code++ = prop_type;
4667 *code++ = prop_value;
4668 }
4669 repeat_max -= repeat_min;
4670
4671 if (repeat_max == 1)
4672 {
4673 *code++ = OP_QUERY + repeat_type;
4674 }
4675 else
4676 {
4677 *code++ = OP_UPTO + repeat_type;
4678 PUT2INC(code, 0, repeat_max);
4679 }
4680 }
4681 }
4682
4683 /* The character or character type itself comes last in all cases. */
4684
4685 #ifdef SUPPORT_UTF8
4686 if (utf8 && c >= 128)
4687 {
4688 memcpy(code, utf8_char, c & 7);
4689 code += c & 7;
4690 }
4691 else
4692 #endif
4693 *code++ = c;
4694
4695 /* For a repeated Unicode property match, there are two extra bytes that
4696 define the required property. */
4697
4698 #ifdef SUPPORT_UCP
4699 if (prop_type >= 0)
4700 {
4701 *code++ = prop_type;
4702 *code++ = prop_value;
4703 }
4704 #endif
4705 }
4706
4707 /* If previous was a character class or a back reference, we put the repeat
4708 stuff after it, but just skip the item if the repeat was {0,0}. */
4709
4710 else if (*previous == OP_CLASS ||
4711 *previous == OP_NCLASS ||
4712 #ifdef SUPPORT_UTF8
4713 *previous == OP_XCLASS ||
4714 #endif
4715 *previous == OP_REF ||
4716 *previous == OP_REFI)
4717 {
4718 if (repeat_max == 0)
4719 {
4720 code = previous;
4721 goto END_REPEAT;
4722 }
4723
4724 /*--------------------------------------------------------------------*/
4725 /* This code is obsolete from release 8.00; the restriction was finally
4726 removed: */
4727
4728 /* All real repeats make it impossible to handle partial matching (maybe
4729 one day we will be able to remove this restriction). */
4730
4731 /* if (repeat_max != 1) cd->external_flags |= PCRE_NOPARTIAL; */
4732 /*--------------------------------------------------------------------*/
4733
4734 if (repeat_min == 0 && repeat_max == -1)
4735 *code++ = OP_CRSTAR + repeat_type;
4736 else if (repeat_min == 1 && repeat_max == -1)
4737 *code++ = OP_CRPLUS + repeat_type;
4738 else if (repeat_min == 0 && repeat_max == 1)
4739 *code++ = OP_CRQUERY + repeat_type;
4740 else
4741 {
4742 *code++ = OP_CRRANGE + repeat_type;
4743 PUT2INC(code, 0, repeat_min);
4744 if (repeat_max == -1) repeat_max = 0; /* 2-byte encoding for max */
4745 PUT2INC(code, 0, repeat_max);
4746 }
4747 }
4748
4749 /* If previous was a bracket group, we may have to replicate it in certain
4750 cases. Note that at this point we can encounter only the "basic" bracket
4751 opcodes such as BRA and CBRA, as this is the place where they get converted
4752 into the more special varieties such as BRAPOS and SBRA. A test for >=
4753 OP_ASSERT and <= OP_COND includes ASSERT, ASSERT_NOT, ASSERTBACK,
4754 ASSERTBACK_NOT, ONCE, BRA, CBRA, and COND. Originally, PCRE did not allow
4755 repetition of assertions, but now it does, for Perl compatibility. */
4756
4757 else if (*previous >= OP_ASSERT && *previous <= OP_COND)
4758 {
4759 register int i;
4760 int len = (int)(code - previous);
4761 uschar *bralink = NULL;
4762 uschar *brazeroptr = NULL;
4763
4764 /* Repeating a DEFINE group is pointless, but Perl allows the syntax, so
4765 we just ignore the repeat. */
4766
4767 if (*previous == OP_COND && previous[LINK_SIZE+1] == OP_DEF)
4768 goto END_REPEAT;
4769
4770 /* There is no sense in actually repeating assertions. The only potential
4771 use of repetition is in cases when the assertion is optional. Therefore,
4772 if the minimum is greater than zero, just ignore the repeat. If the
4773 maximum is not not zero or one, set it to 1. */
4774
4775 if (*previous < OP_ONCE) /* Assertion */
4776 {
4777 if (repeat_min > 0) goto END_REPEAT;
4778 if (repeat_max < 0 || repeat_max > 1) repeat_max = 1;
4779 }
4780
4781 /* The case of a zero minimum is special because of the need to stick
4782 OP_BRAZERO in front of it, and because the group appears once in the
4783 data, whereas in other cases it appears the minimum number of times. For
4784 this reason, it is simplest to treat this case separately, as otherwise
4785 the code gets far too messy. There are several special subcases when the
4786 minimum is zero. */
4787
4788 if (repeat_min == 0)
4789 {
4790 /* If the maximum is also zero, we used to just omit the group from the
4791 output altogether, like this:
4792
4793 ** if (repeat_max == 0)
4794 ** {
4795 ** code = previous;
4796 ** goto END_REPEAT;
4797 ** }
4798
4799 However, that fails when a group or a subgroup within it is referenced
4800 as a subroutine from elsewhere in the pattern, so now we stick in
4801 OP_SKIPZERO in front of it so that it is skipped on execution. As we
4802 don't have a list of which groups are referenced, we cannot do this
4803 selectively.
4804
4805 If the maximum is 1 or unlimited, we just have to stick in the BRAZERO
4806 and do no more at this point. However, we do need to adjust any
4807 OP_RECURSE calls inside the group that refer to the group itself or any
4808 internal or forward referenced group, because the offset is from the
4809 start of the whole regex. Temporarily terminate the pattern while doing
4810 this. */
4811
4812 if (repeat_max <= 1) /* Covers 0, 1, and unlimited */
4813 {
4814 *code = OP_END;
4815 adjust_recurse(previous, 1, utf8, cd, save_hwm);
4816 memmove(previous+1, previous, len);
4817 code++;
4818 if (repeat_max == 0)
4819 {
4820 *previous++ = OP_SKIPZERO;
4821 goto END_REPEAT;
4822 }
4823 brazeroptr = previous; /* Save for possessive optimizing */
4824 *previous++ = OP_BRAZERO + repeat_type;
4825 }
4826
4827 /* If the maximum is greater than 1 and limited, we have to replicate
4828 in a nested fashion, sticking OP_BRAZERO before each set of brackets.
4829 The first one has to be handled carefully because it's the original
4830 copy, which has to be moved up. The remainder can be handled by code
4831 that is common with the non-zero minimum case below. We have to
4832 adjust the value or repeat_max, since one less copy is required. Once
4833 again, we may have to adjust any OP_RECURSE calls inside the group. */
4834
4835 else
4836 {
4837 int offset;
4838 *code = OP_END;
4839 adjust_recurse(previous, 2 + LINK_SIZE, utf8, cd, save_hwm);
4840 memmove(previous + 2 + LINK_SIZE, previous, len);
4841 code += 2 + LINK_SIZE;
4842 *previous++ = OP_BRAZERO + repeat_type;
4843 *previous++ = OP_BRA;
4844
4845 /* We chain together the bracket offset fields that have to be
4846 filled in later when the ends of the brackets are reached. */
4847
4848 offset = (bralink == NULL)? 0 : (int)(previous - bralink);
4849 bralink = previous;
4850 PUTINC(previous, 0, offset);
4851 }
4852
4853 repeat_max--;
4854 }
4855
4856 /* If the minimum is greater than zero, replicate the group as many
4857 times as necessary, and adjust the maximum to the number of subsequent
4858 copies that we need. If we set a first char from the group, and didn't
4859 set a required char, copy the latter from the former. If there are any
4860 forward reference subroutine calls in the group, there will be entries on
4861 the workspace list; replicate these with an appropriate increment. */
4862
4863 else
4864 {
4865 if (repeat_min > 1)
4866 {
4867 /* In the pre-compile phase, we don't actually do the replication. We
4868 just adjust the length as if we had. Do some paranoid checks for
4869 potential integer overflow. The INT64_OR_DOUBLE type is a 64-bit
4870 integer type when available, otherwise double. */
4871
4872 if (lengthptr != NULL)
4873 {
4874 int delta = (repeat_min - 1)*length_prevgroup;
4875 if ((INT64_OR_DOUBLE)(repeat_min - 1)*
4876 (INT64_OR_DOUBLE)length_prevgroup >
4877 (INT64_OR_DOUBLE)INT_MAX ||
4878 OFLOW_MAX - *lengthptr < delta)
4879 {
4880 *errorcodeptr = ERR20;
4881 goto FAILED;
4882 }
4883 *lengthptr += delta;
4884 }
4885
4886 /* This is compiling for real */
4887
4888 else
4889 {
4890 if (groupsetfirstbyte && reqbyte < 0) reqbyte = firstbyte;
4891 for (i = 1; i < repeat_min; i++)
4892 {
4893 uschar *hc;
4894 uschar *this_hwm = cd->hwm;
4895 memcpy(code, previous, len);
4896 for (hc = save_hwm; hc < this_hwm; hc += LINK_SIZE)
4897 {
4898 PUT(cd->hwm, 0, GET(hc, 0) + len);
4899 cd->hwm += LINK_SIZE;
4900 }
4901 save_hwm = this_hwm;
4902 code += len;
4903 }
4904 }
4905 }
4906
4907 if (repeat_max > 0) repeat_max -= repeat_min;
4908 }
4909
4910 /* This code is common to both the zero and non-zero minimum cases. If
4911 the maximum is limited, it replicates the group in a nested fashion,
4912 remembering the bracket starts on a stack. In the case of a zero minimum,
4913 the first one was set up above. In all cases the repeat_max now specifies
4914 the number of additional copies needed. Again, we must remember to
4915 replicate entries on the forward reference list. */
4916
4917 if (repeat_max >= 0)
4918 {
4919 /* In the pre-compile phase, we don't actually do the replication. We
4920 just adjust the length as if we had. For each repetition we must add 1
4921 to the length for BRAZERO and for all but the last repetition we must
4922 add 2 + 2*LINKSIZE to allow for the nesting that occurs. Do some
4923 paranoid checks to avoid integer overflow. The INT64_OR_DOUBLE type is
4924 a 64-bit integer type when available, otherwise double. */
4925
4926 if (lengthptr != NULL && repeat_max > 0)
4927 {
4928 int delta = repeat_max * (length_prevgroup + 1 + 2 + 2*LINK_SIZE) -
4929 2 - 2*LINK_SIZE; /* Last one doesn't nest */
4930 if ((INT64_OR_DOUBLE)repeat_max *
4931 (INT64_OR_DOUBLE)(length_prevgroup + 1 + 2 + 2*LINK_SIZE)
4932 > (INT64_OR_DOUBLE)INT_MAX ||
4933 OFLOW_MAX - *lengthptr < delta)
4934 {
4935 *errorcodeptr = ERR20;
4936 goto FAILED;
4937 }
4938 *lengthptr += delta;
4939 }
4940
4941 /* This is compiling for real */
4942
4943 else for (i = repeat_max - 1; i >= 0; i--)
4944 {
4945 uschar *hc;
4946 uschar *this_hwm = cd->hwm;
4947
4948 *code++ = OP_BRAZERO + repeat_type;
4949
4950 /* All but the final copy start a new nesting, maintaining the
4951 chain of brackets outstanding. */
4952
4953 if (i != 0)
4954 {
4955 int offset;
4956 *code++ = OP_BRA;
4957 offset = (bralink == NULL)? 0 : (int)(code - bralink);
4958 bralink = code;
4959 PUTINC(code, 0, offset);
4960 }
4961
4962 memcpy(code, previous, len);
4963 for (hc = save_hwm; hc < this_hwm; hc += LINK_SIZE)
4964 {
4965 PUT(cd->hwm, 0, GET(hc, 0) + len + ((i != 0)? 2+LINK_SIZE : 1));
4966 cd->hwm += LINK_SIZE;
4967 }
4968 save_hwm = this_hwm;
4969 code += len;
4970 }
4971
4972 /* Now chain through the pending brackets, and fill in their length
4973 fields (which are holding the chain links pro tem). */
4974
4975 while (bralink != NULL)
4976 {
4977 int oldlinkoffset;
4978 int offset = (int)(code - bralink + 1);
4979 uschar *bra = code - offset;
4980 oldlinkoffset = GET(bra, 1);
4981 bralink = (oldlinkoffset == 0)? NULL : bralink - oldlinkoffset;
4982 *code++ = OP_KET;
4983 PUTINC(code, 0, offset);
4984 PUT(bra, 1, offset);
4985 }
4986 }
4987
4988 /* If the maximum is unlimited, set a repeater in the final copy. For
4989 ONCE brackets, that's all we need to do. However, possessively repeated
4990 ONCE brackets can be converted into non-capturing brackets, as the
4991 behaviour of (?:xx)++ is the same as (?>xx)++ and this saves having to
4992 deal with possessive ONCEs specially.
4993
4994 Otherwise, when we are doing the actual compile phase, check to see
4995 whether this group is one that could match an empty string. If so,
4996 convert the initial operator to the S form (e.g. OP_BRA -> OP_SBRA) so
4997 that runtime checking can be done. [This check is also applied to ONCE
4998 groups at runtime, but in a different way.]
4999
5000 Then, if the quantifier was possessive and the bracket is not a
5001 conditional, we convert the BRA code to the POS form, and the KET code to
5002 KETRPOS. (It turns out to be convenient at runtime to detect this kind of
5003 subpattern at both the start and at the end.) The use of special opcodes
5004 makes it possible to reduce greatly the stack usage in pcre_exec(). If
5005 the group is preceded by OP_BRAZERO, convert this to OP_BRAPOSZERO.
5006
5007 Then, if the minimum number of matches is 1 or 0, cancel the possessive
5008 flag so that the default action below, of wrapping everything inside
5009 atomic brackets, does not happen. When the minimum is greater than 1,
5010 there will be earlier copies of the group, and so we still have to wrap
5011 the whole thing. */
5012
5013 else
5014 {
5015 uschar *ketcode = code - 1 - LINK_SIZE;
5016 uschar *bracode = ketcode - GET(ketcode, 1);
5017
5018 /* Convert possessive ONCE brackets to non-capturing */
5019
5020 if ((*bracode == OP_ONCE || *bracode == OP_ONCE_NC) &&
5021 possessive_quantifier) *bracode = OP_BRA;
5022
5023 /* For non-possessive ONCE brackets, all we need to do is to
5024 set the KET. */
5025
5026 if (*bracode == OP_ONCE || *bracode == OP_ONCE_NC)
5027 *ketcode = OP_KETRMAX + repeat_type;
5028
5029 /* Handle non-ONCE brackets and possessive ONCEs (which have been
5030 converted to non-capturing above). */
5031
5032 else
5033 {
5034 /* In the compile phase, check for empty string matching. */
5035
5036 if (lengthptr == NULL)
5037 {
5038 uschar *scode = bracode;
5039 do
5040 {
5041 if (could_be_empty_branch(scode, ketcode, utf8, cd))
5042 {
5043 *bracode += OP_SBRA - OP_BRA;
5044 break;
5045 }
5046 scode += GET(scode, 1);
5047 }
5048 while (*scode == OP_ALT);
5049 }
5050
5051 /* Handle possessive quantifiers. */
5052
5053 if (possessive_quantifier)
5054 {
5055 /* For COND brackets, we wrap the whole thing in a possessively
5056 repeated non-capturing bracket, because we have not invented POS
5057 versions of the COND opcodes. Because we are moving code along, we
5058 must ensure that any pending recursive references are updated. */
5059
5060 if (*bracode == OP_COND || *bracode == OP_SCOND)
5061 {
5062 int nlen = (int)(code - bracode);
5063 *code = OP_END;
5064 adjust_recurse(bracode, 1 + LINK_SIZE, utf8, cd, save_hwm);
5065 memmove(bracode + 1+LINK_SIZE, bracode, nlen);
5066 code += 1 + LINK_SIZE;
5067 nlen += 1 + LINK_SIZE;
5068 *bracode = OP_BRAPOS;
5069 *code++ = OP_KETRPOS;
5070 PUTINC(code, 0, nlen);
5071 PUT(bracode, 1, nlen);
5072 }
5073
5074 /* For non-COND brackets, we modify the BRA code and use KETRPOS. */
5075
5076 else
5077 {
5078 *bracode += 1; /* Switch to xxxPOS opcodes */
5079 *ketcode = OP_KETRPOS;
5080 }
5081
5082 /* If the minimum is zero, mark it as possessive, then unset the
5083 possessive flag when the minimum is 0 or 1. */
5084
5085 if (brazeroptr != NULL) *brazeroptr = OP_BRAPOSZERO;
5086 if (repeat_min < 2) possessive_quantifier = FALSE;
5087 }
5088
5089 /* Non-possessive quantifier */
5090
5091 else *ketcode = OP_KETRMAX + repeat_type;
5092 }
5093 }
5094 }
5095
5096 /* If previous is OP_FAIL, it was generated by an empty class [] in
5097 JavaScript mode. The other ways in which OP_FAIL can be generated, that is
5098 by (*FAIL) or (?!) set previous to NULL, which gives a "nothing to repeat"
5099 error above. We can just ignore the repeat in JS case. */
5100
5101 else if (*previous == OP_FAIL) goto END_REPEAT;
5102
5103 /* Else there's some kind of shambles */
5104
5105 else
5106 {
5107 *errorcodeptr = ERR11;
5108 goto FAILED;
5109 }
5110
5111 /* If the character following a repeat is '+', or if certain optimization
5112 tests above succeeded, possessive_quantifier is TRUE. For some opcodes,
5113 there are special alternative opcodes for this case. For anything else, we
5114 wrap the entire repeated item inside OP_ONCE brackets. Logically, the '+'
5115 notation is just syntactic sugar, taken from Sun's Java package, but the
5116 special opcodes can optimize it.
5117
5118 Some (but not all) possessively repeated subpatterns have already been
5119 completely handled in the code just above. For them, possessive_quantifier
5120 is always FALSE at this stage.
5121
5122 Note that the repeated item starts at tempcode, not at previous, which
5123 might be the first part of a string whose (former) last char we repeated.
5124
5125 Possessifying an 'exact' quantifier has no effect, so we can ignore it. But
5126 an 'upto' may follow. We skip over an 'exact' item, and then test the
5127 length of what remains before proceeding. */
5128
5129 if (possessive_quantifier)
5130 {
5131 int len;
5132
5133 if (*tempcode == OP_TYPEEXACT)
5134 tempcode += _pcre_OP_lengths[*tempcode] +
5135 ((tempcode[3] == OP_PROP || tempcode[3] == OP_NOTPROP)? 2 : 0);
5136
5137 else if (*tempcode == OP_EXACT || *tempcode == OP_NOTEXACT)
5138 {
5139 tempcode += _pcre_OP_lengths[*tempcode];
5140 #ifdef SUPPORT_UTF8
5141 if (utf8 && tempcode[-1] >= 0xc0)
5142 tempcode += _pcre_utf8_table4[tempcode[-1] & 0x3f];
5143 #endif
5144 }
5145
5146 len = (int)(code - tempcode);
5147 if (len > 0) switch (*tempcode)
5148 {
5149 case OP_STAR: *tempcode = OP_POSSTAR; break;
5150 case OP_PLUS: *tempcode = OP_POSPLUS; break;
5151 case OP_QUERY: *tempcode = OP_POSQUERY; break;
5152 case OP_UPTO: *tempcode = OP_POSUPTO; break;
5153
5154 case OP_STARI: *tempcode = OP_POSSTARI; break;
5155 case OP_PLUSI: *tempcode = OP_POSPLUSI; break;
5156 case OP_QUERYI: *tempcode = OP_POSQUERYI; break;
5157 case OP_UPTOI: *tempcode = OP_POSUPTOI; break;
5158
5159 case OP_NOTSTAR: *tempcode = OP_NOTPOSSTAR; break;
5160 case OP_NOTPLUS: *tempcode = OP_NOTPOSPLUS; break;
5161 case OP_NOTQUERY: *tempcode = OP_NOTPOSQUERY; break;
5162 case OP_NOTUPTO: *tempcode = OP_NOTPOSUPTO; break;
5163
5164 case OP_NOTSTARI: *tempcode = OP_NOTPOSSTARI; break;
5165 case OP_NOTPLUSI: *tempcode = OP_NOTPOSPLUSI; break;
5166 case OP_NOTQUERYI: *tempcode = OP_NOTPOSQUERYI; break;
5167 case OP_NOTUPTOI: *tempcode = OP_NOTPOSUPTOI; break;
5168
5169 case OP_TYPESTAR: *tempcode = OP_TYPEPOSSTAR; break;
5170 case OP_TYPEPLUS: *tempcode = OP_TYPEPOSPLUS; break;
5171 case OP_TYPEQUERY: *tempcode = OP_TYPEPOSQUERY; break;
5172 case OP_TYPEUPTO: *tempcode = OP_TYPEPOSUPTO; break;
5173
5174 /* Because we are moving code along, we must ensure that any
5175 pending recursive references are updated. */
5176
5177 default:
5178 *code = OP_END;
5179 adjust_recurse(tempcode, 1 + LINK_SIZE, utf8, cd, save_hwm);
5180 memmove(tempcode + 1+LINK_SIZE, tempcode, len);
5181 code += 1 + LINK_SIZE;
5182 len += 1 + LINK_SIZE;
5183 tempcode[0] = OP_ONCE;
5184 *code++ = OP_KET;
5185 PUTINC(code, 0, len);
5186 PUT(tempcode, 1, len);
5187 break;
5188 }
5189 }
5190
5191 /* In all case we no longer have a previous item. We also set the
5192 "follows varying string" flag for subsequently encountered reqbytes if
5193 it isn't already set and we have just passed a varying length item. */
5194
5195 END_REPEAT:
5196 previous = NULL;
5197 cd->req_varyopt |= reqvary;
5198 break;
5199
5200
5201 /* ===================================================================*/
5202 /* Start of nested parenthesized sub-expression, or comment or lookahead or
5203 lookbehind or option setting or condition or all the other extended
5204 parenthesis forms. */
5205
5206 case CHAR_LEFT_PARENTHESIS:
5207 newoptions = options;
5208 skipbytes = 0;
5209 bravalue = OP_CBRA;
5210 save_hwm = cd->hwm;
5211 reset_bracount = FALSE;
5212
5213 /* First deal with various "verbs" that can be introduced by '*'. */
5214
5215 if (*(++ptr) == CHAR_ASTERISK &&
5216 ((cd->ctypes[ptr[1]] & ctype_letter) != 0 || ptr[1] == ':'))
5217 {
5218 int i, namelen;
5219 int arglen = 0;
5220 const char *vn = verbnames;
5221 const uschar *name = ptr + 1;
5222 const uschar *arg = NULL;
5223 previous = NULL;
5224 while ((cd->ctypes[*++ptr] & ctype_letter) != 0) {};
5225 namelen = (int)(ptr - name);
5226
5227 /* It appears that Perl allows any characters whatsoever, other than
5228 a closing parenthesis, to appear in arguments, so we no longer insist on
5229 letters, digits, and underscores. */
5230
5231 if (*ptr == CHAR_COLON)
5232 {
5233 arg = ++ptr;
5234 while (*ptr != 0 && *ptr != CHAR_RIGHT_PARENTHESIS) ptr++;
5235 arglen = (int)(ptr - arg);
5236 }
5237
5238 if (*ptr != CHAR_RIGHT_PARENTHESIS)
5239 {
5240 *errorcodeptr = ERR60;
5241 goto FAILED;
5242 }
5243
5244 /* Scan the table of verb names */
5245
5246 for (i = 0; i < verbcount; i++)
5247 {
5248 if (namelen == verbs[i].len &&
5249 strncmp((char *)name, vn, namelen) == 0)
5250 {
5251 /* Check for open captures before ACCEPT and convert it to
5252 ASSERT_ACCEPT if in an assertion. */
5253
5254 if (verbs[i].op == OP_ACCEPT)
5255 {
5256 open_capitem *oc;
5257 if (arglen != 0)
5258 {
5259 *errorcodeptr = ERR59;
5260 goto FAILED;
5261 }
5262 cd->had_accept = TRUE;
5263 for (oc = cd->open_caps; oc != NULL; oc = oc->next)
5264 {
5265 *code++ = OP_CLOSE;
5266 PUT2INC(code, 0, oc->number);
5267 }
5268 *code++ = (cd->assert_depth > 0)? OP_ASSERT_ACCEPT : OP_ACCEPT;
5269
5270 /* Do not set firstbyte after *ACCEPT */
5271 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
5272 }
5273
5274 /* Handle other cases with/without an argument */
5275
5276 else if (arglen == 0)
5277 {
5278 if (verbs[i].op < 0) /* Argument is mandatory */
5279 {
5280 *errorcodeptr = ERR66;
5281 goto FAILED;
5282 }
5283 *code = verbs[i].op;
5284 if (*code++ == OP_THEN) cd->external_flags |= PCRE_HASTHEN;
5285 }
5286
5287 else
5288 {
5289 if (verbs[i].op_arg < 0) /* Argument is forbidden */
5290 {
5291 *errorcodeptr = ERR59;
5292 goto FAILED;
5293 }
5294 *code = verbs[i].op_arg;
5295 if (*code++ == OP_THEN_ARG) cd->external_flags |= PCRE_HASTHEN;
5296 *code++ = arglen;
5297 memcpy(code, arg, arglen);
5298 code += arglen;
5299 *code++ = 0;
5300 }
5301
5302 break; /* Found verb, exit loop */
5303 }
5304
5305 vn += verbs[i].len + 1;
5306 }
5307
5308 if (i < verbcount) continue; /* Successfully handled a verb */
5309 *errorcodeptr = ERR60; /* Verb not recognized */
5310 goto FAILED;
5311 }
5312
5313 /* Deal with the extended parentheses; all are introduced by '?', and the
5314 appearance of any of them means that this is not a capturing group. */
5315
5316 else if (*ptr == CHAR_QUESTION_MARK)
5317 {
5318 int i, set, unset, namelen;
5319 int *optset;
5320 const uschar *name;
5321 uschar *slot;
5322
5323 switch (*(++ptr))
5324 {
5325 case CHAR_NUMBER_SIGN: /* Comment; skip to ket */
5326 ptr++;
5327 while (*ptr != 0 && *ptr != CHAR_RIGHT_PARENTHESIS) ptr++;
5328 if (*ptr == 0)
5329 {
5330 *errorcodeptr = ERR18;
5331 goto FAILED;
5332 }
5333 continue;
5334
5335
5336 /* ------------------------------------------------------------ */
5337 case CHAR_VERTICAL_LINE: /* Reset capture count for each branch */
5338 reset_bracount = TRUE;
5339 /* Fall through */
5340
5341 /* ------------------------------------------------------------ */
5342 case CHAR_COLON: /* Non-capturing bracket */
5343 bravalue = OP_BRA;
5344 ptr++;
5345 break;
5346
5347
5348 /* ------------------------------------------------------------ */
5349 case CHAR_LEFT_PARENTHESIS:
5350 bravalue = OP_COND; /* Conditional group */
5351
5352 /* A condition can be an assertion, a number (referring to a numbered
5353 group), a name (referring to a named group), or 'R', referring to
5354 recursion. R<digits> and R&name are also permitted for recursion tests.
5355
5356 There are several syntaxes for testing a named group: (?(name)) is used
5357 by Python; Perl 5.10 onwards uses (?(<name>) or (?('name')).
5358
5359 There are two unfortunate ambiguities, caused by history. (a) 'R' can
5360 be the recursive thing or the name 'R' (and similarly for 'R' followed
5361 by digits), and (b) a number could be a name that consists of digits.
5362 In both cases, we look for a name first; if not found, we try the other
5363 cases. */
5364
5365 /* For conditions that are assertions, check the syntax, and then exit
5366 the switch. This will take control down to where bracketed groups,
5367 including assertions, are processed. */
5368
5369 if (ptr[1] == CHAR_QUESTION_MARK && (ptr[2] == CHAR_EQUALS_SIGN ||
5370 ptr[2] == CHAR_EXCLAMATION_MARK || ptr[2] == CHAR_LESS_THAN_SIGN))
5371 break;
5372
5373 /* Most other conditions use OP_CREF (a couple change to OP_RREF
5374 below), and all need to skip 3 bytes at the start of the group. */
5375
5376 code[1+LINK_SIZE] = OP_CREF;
5377 skipbytes = 3;
5378 refsign = -1;
5379
5380 /* Check for a test for recursion in a named group. */
5381
5382 if (ptr[1] == CHAR_R && ptr[2] == CHAR_AMPERSAND)
5383 {
5384 terminator = -1;
5385 ptr += 2;
5386 code[1+LINK_SIZE] = OP_RREF; /* Change the type of test */
5387 }
5388
5389 /* Check for a test for a named group's having been set, using the Perl
5390 syntax (?(<name>) or (?('name') */
5391
5392 else if (ptr[1] == CHAR_LESS_THAN_SIGN)
5393 {
5394 terminator = CHAR_GREATER_THAN_SIGN;
5395 ptr++;
5396 }
5397 else if (ptr[1] == CHAR_APOSTROPHE)
5398 {
5399 terminator = CHAR_APOSTROPHE;
5400 ptr++;
5401 }
5402 else
5403 {
5404 terminator = 0;
5405 if (ptr[1] == CHAR_MINUS || ptr[1] == CHAR_PLUS) refsign = *(++ptr);
5406 }
5407
5408 /* We now expect to read a name; any thing else is an error */
5409
5410 if ((cd->ctypes[ptr[1]] & ctype_word) == 0)
5411 {
5412 ptr += 1; /* To get the right offset */
5413 *errorcodeptr = ERR28;
5414 goto FAILED;
5415 }
5416
5417 /* Read the name, but also get it as a number if it's all digits */
5418
5419 recno = 0;
5420 name = ++ptr;
5421 while ((cd->ctypes[*ptr] & ctype_word) != 0)
5422 {
5423 if (recno >= 0)
5424 recno = ((digitab[*ptr] & ctype_digit) != 0)?
5425 recno * 10 + *ptr - CHAR_0 : -1;
5426 ptr++;
5427 }
5428 namelen = (int)(ptr - name);
5429
5430 if ((terminator > 0 && *ptr++ != terminator) ||
5431 *ptr++ != CHAR_RIGHT_PARENTHESIS)
5432 {
5433 ptr--; /* Error offset */
5434 *errorcodeptr = ERR26;
5435 goto FAILED;
5436 }
5437
5438 /* Do no further checking in the pre-compile phase. */
5439
5440 if (lengthptr != NULL) break;
5441
5442 /* In the real compile we do the work of looking for the actual
5443 reference. If the string started with "+" or "-" we require the rest to
5444 be digits, in which case recno will be set. */
5445
5446 if (refsign > 0)
5447 {
5448 if (recno <= 0)
5449 {
5450 *errorcodeptr = ERR58;
5451 goto FAILED;
5452 }
5453 recno = (refsign == CHAR_MINUS)?
5454 cd->bracount - recno + 1 : recno +cd->bracount;
5455 if (recno <= 0 || recno > cd->final_bracount)
5456 {
5457 *errorcodeptr = ERR15;
5458 goto FAILED;
5459 }
5460 PUT2(code, 2+LINK_SIZE, recno);
5461 break;
5462 }
5463
5464 /* Otherwise (did not start with "+" or "-"), start by looking for the
5465 name. If we find a name, add one to the opcode to change OP_CREF or
5466 OP_RREF into OP_NCREF or OP_NRREF. These behave exactly the same,
5467 except they record that the reference was originally to a name. The
5468 information is used to check duplicate names. */
5469
5470 slot = cd->name_table;
5471 for (i = 0; i < cd->names_found; i++)
5472 {
5473 if (strncmp((char *)name, (char *)slot+2, namelen) == 0) break;
5474 slot += cd->name_entry_size;
5475 }
5476
5477 /* Found a previous named subpattern */
5478
5479 if (i < cd->names_found)
5480 {
5481 recno = GET2(slot, 0);
5482 PUT2(code, 2+LINK_SIZE, recno);
5483 code[1+LINK_SIZE]++;
5484 }
5485
5486 /* Search the pattern for a forward reference */
5487
5488 else if ((i = find_parens(cd, name, namelen,
5489 (options & PCRE_EXTENDED) != 0, utf8)) > 0)
5490 {
5491 PUT2(code, 2+LINK_SIZE, i);
5492 code[1+LINK_SIZE]++;
5493 }
5494
5495 /* If terminator == 0 it means that the name followed directly after
5496 the opening parenthesis [e.g. (?(abc)...] and in this case there are
5497 some further alternatives to try. For the cases where terminator != 0
5498 [things like (?(<name>... or (?('name')... or (?(R&name)... ] we have
5499 now checked all the possibilities, so give an error. */
5500
5501 else if (terminator != 0)
5502 {
5503 *errorcodeptr = ERR15;
5504 goto FAILED;
5505 }
5506
5507 /* Check for (?(R) for recursion. Allow digits after R to specify a
5508 specific group number. */
5509
5510 else if (*name == CHAR_R)
5511 {
5512 recno = 0;
5513 for (i = 1; i < namelen; i++)
5514 {
5515 if ((digitab[name[i]] & ctype_digit) == 0)
5516 {
5517 *errorcodeptr = ERR15;
5518 goto FAILED;
5519 }
5520 recno = recno * 10 + name[i] - CHAR_0;
5521 }
5522 if (recno == 0) recno = RREF_ANY;
5523 code[1+LINK_SIZE] = OP_RREF; /* Change test type */
5524 PUT2(code, 2+LINK_SIZE, recno);
5525 }
5526
5527 /* Similarly, check for the (?(DEFINE) "condition", which is always
5528 false. */
5529
5530 else if (namelen == 6 && strncmp((char *)name, STRING_DEFINE, 6) == 0)
5531 {
5532 code[1+LINK_SIZE] = OP_DEF;
5533 skipbytes = 1;
5534 }
5535
5536 /* Check for the "name" actually being a subpattern number. We are
5537 in the second pass here, so final_bracount is set. */
5538
5539 else if (recno > 0 && recno <= cd->final_bracount)
5540 {
5541 PUT2(code, 2+LINK_SIZE, recno);
5542 }
5543
5544 /* Either an unidentified subpattern, or a reference to (?(0) */
5545
5546 else
5547 {
5548 *errorcodeptr = (recno == 0)? ERR35: ERR15;
5549 goto FAILED;
5550 }
5551 break;
5552
5553
5554 /* ------------------------------------------------------------ */
5555 case CHAR_EQUALS_SIGN: /* Positive lookahead */
5556 bravalue = OP_ASSERT;
5557 cd->assert_depth += 1;
5558 ptr++;
5559 break;
5560
5561
5562 /* ------------------------------------------------------------ */
5563 case CHAR_EXCLAMATION_MARK: /* Negative lookahead */
5564 ptr++;
5565 if (*ptr == CHAR_RIGHT_PARENTHESIS) /* Optimize (?!) */
5566 {
5567 *code++ = OP_FAIL;
5568 previous = NULL;
5569 continue;
5570 }
5571 bravalue = OP_ASSERT_NOT;
5572 cd->assert_depth += 1;
5573 break;
5574
5575
5576 /* ------------------------------------------------------------ */
5577 case CHAR_LESS_THAN_SIGN: /* Lookbehind or named define */
5578 switch (ptr[1])
5579 {
5580 case CHAR_EQUALS_SIGN: /* Positive lookbehind */
5581 bravalue = OP_ASSERTBACK;
5582 cd->assert_depth += 1;
5583 ptr += 2;
5584 break;
5585
5586 case CHAR_EXCLAMATION_MARK: /* Negative lookbehind */
5587 bravalue = OP_ASSERTBACK_NOT;
5588 cd->assert_depth += 1;
5589 ptr += 2;
5590 break;
5591
5592 default: /* Could be name define, else bad */
5593 if ((cd->ctypes[ptr[1]] & ctype_word) != 0) goto DEFINE_NAME;
5594 ptr++; /* Correct offset for error */
5595 *errorcodeptr = ERR24;
5596 goto FAILED;
5597 }
5598 break;
5599
5600
5601 /* ------------------------------------------------------------ */
5602 case CHAR_GREATER_THAN_SIGN: /* One-time brackets */
5603 bravalue = OP_ONCE;
5604 ptr++;
5605 break;
5606
5607
5608 /* ------------------------------------------------------------ */
5609 case CHAR_C: /* Callout - may be followed by digits; */
5610 previous_callout = code; /* Save for later completion */
5611 after_manual_callout = 1; /* Skip one item before completing */
5612 *code++ = OP_CALLOUT;
5613 {
5614 int n = 0;
5615 while ((digitab[*(++ptr)] & ctype_digit) != 0)
5616 n = n * 10 + *ptr - CHAR_0;
5617 if (*ptr != CHAR_RIGHT_PARENTHESIS)
5618 {
5619 *errorcodeptr = ERR39;
5620 goto FAILED;
5621 }
5622 if (n > 255)
5623 {
5624 *errorcodeptr = ERR38;
5625 goto FAILED;
5626 }
5627 *code++ = n;
5628 PUT(code, 0, (int)(ptr - cd->start_pattern + 1)); /* Pattern offset */
5629 PUT(code, LINK_SIZE, 0); /* Default length */
5630 code += 2 * LINK_SIZE;
5631 }
5632 previous = NULL;
5633 continue;
5634
5635
5636 /* ------------------------------------------------------------ */
5637 case CHAR_P: /* Python-style named subpattern handling */
5638 if (*(++ptr) == CHAR_EQUALS_SIGN ||
5639 *ptr == CHAR_GREATER_THAN_SIGN) /* Reference or recursion */
5640 {
5641 is_recurse = *ptr == CHAR_GREATER_THAN_SIGN;
5642 terminator = CHAR_RIGHT_PARENTHESIS;
5643 goto NAMED_REF_OR_RECURSE;
5644 }
5645 else if (*ptr != CHAR_LESS_THAN_SIGN) /* Test for Python-style defn */
5646 {
5647 *errorcodeptr = ERR41;
5648 goto FAILED;
5649 }
5650 /* Fall through to handle (?P< as (?< is handled */
5651
5652
5653 /* ------------------------------------------------------------ */
5654 DEFINE_NAME: /* Come here from (?< handling */
5655 case CHAR_APOSTROPHE:
5656 {
5657 terminator = (*ptr == CHAR_LESS_THAN_SIGN)?
5658 CHAR_GREATER_THAN_SIGN : CHAR_APOSTROPHE;
5659 name = ++ptr;
5660
5661 while ((cd->ctypes[*ptr] & ctype_word) != 0) ptr++;
5662 namelen = (int)(ptr - name);
5663
5664 /* In the pre-compile phase, just do a syntax check. */
5665
5666 if (lengthptr != NULL)
5667 {
5668 if (*ptr != terminator)
5669 {
5670 *errorcodeptr = ERR42;
5671 goto FAILED;
5672 }
5673 if (cd->names_found >= MAX_NAME_COUNT)
5674 {
5675 *errorcodeptr = ERR49;
5676 goto FAILED;
5677 }
5678 if (namelen + 3 > cd->name_entry_size)
5679 {
5680 cd->name_entry_size = namelen + 3;
5681 if (namelen > MAX_NAME_SIZE)
5682 {
5683 *errorcodeptr = ERR48;
5684 goto FAILED;
5685 }
5686 }
5687 }
5688
5689 /* In the real compile, create the entry in the table, maintaining
5690 alphabetical order. Duplicate names for different numbers are
5691 permitted only if PCRE_DUPNAMES is set. Duplicate names for the same
5692 number are always OK. (An existing number can be re-used if (?|
5693 appears in the pattern.) In either event, a duplicate name results in
5694 a duplicate entry in the table, even if the number is the same. This
5695 is because the number of names, and hence the table size, is computed
5696 in the pre-compile, and it affects various numbers and pointers which
5697 would all have to be modified, and the compiled code moved down, if
5698 duplicates with the same number were omitted from the table. This
5699 doesn't seem worth the hassle. However, *different* names for the
5700 same number are not permitted. */
5701
5702 else
5703 {
5704 BOOL dupname = FALSE;
5705 slot = cd->name_table;
5706
5707 for (i = 0; i < cd->names_found; i++)
5708 {
5709 int crc = memcmp(name, slot+2, namelen);
5710 if (crc == 0)
5711 {
5712 if (slot[2+namelen] == 0)
5713 {
5714 if (GET2(slot, 0) != cd->bracount + 1 &&
5715 (options & PCRE_DUPNAMES) == 0)
5716 {
5717 *errorcodeptr = ERR43;
5718 goto FAILED;
5719 }
5720 else dupname = TRUE;
5721 }
5722 else crc = -1; /* Current name is a substring */
5723 }
5724
5725 /* Make space in the table and break the loop for an earlier
5726 name. For a duplicate or later name, carry on. We do this for
5727 duplicates so that in the simple case (when ?(| is not used) they
5728 are in order of their numbers. */
5729
5730 if (crc < 0)
5731 {
5732 memmove(slot + cd->name_entry_size, slot,
5733 (cd->names_found - i) * cd->name_entry_size);
5734 break;
5735 }
5736
5737 /* Continue the loop for a later or duplicate name */
5738
5739 slot += cd->name_entry_size;
5740 }
5741
5742 /* For non-duplicate names, check for a duplicate number before
5743 adding the new name. */
5744
5745 if (!dupname)
5746 {
5747 uschar *cslot = cd->name_table;
5748 for (i = 0; i < cd->names_found; i++)
5749 {
5750 if (cslot != slot)
5751 {
5752 if (GET2(cslot, 0) == cd->bracount + 1)
5753 {
5754 *errorcodeptr = ERR65;
5755 goto FAILED;
5756 }
5757 }
5758 else i--;
5759 cslot += cd->name_entry_size;
5760 }
5761 }
5762
5763 PUT2(slot, 0, cd->bracount + 1);
5764 memcpy(slot + 2, name, namelen);
5765 slot[2+namelen] = 0;
5766 }
5767 }
5768
5769 /* In both pre-compile and compile, count the number of names we've
5770 encountered. */
5771
5772 cd->names_found++;
5773 ptr++; /* Move past > or ' */
5774 goto NUMBERED_GROUP;
5775
5776
5777 /* ------------------------------------------------------------ */
5778 case CHAR_AMPERSAND: /* Perl recursion/subroutine syntax */
5779 terminator = CHAR_RIGHT_PARENTHESIS;
5780 is_recurse = TRUE;
5781 /* Fall through */
5782
5783 /* We come here from the Python syntax above that handles both
5784 references (?P=name) and recursion (?P>name), as well as falling
5785 through from the Perl recursion syntax (?&name). We also come here from
5786 the Perl \k<name> or \k'name' back reference syntax and the \k{name}
5787 .NET syntax, and the Oniguruma \g<...> and \g'...' subroutine syntax. */
5788
5789 NAMED_REF_OR_RECURSE:
5790 name = ++ptr;
5791 while ((cd->ctypes[*ptr] & ctype_word) != 0) ptr++;
5792 namelen = (int)(ptr - name);
5793
5794 /* In the pre-compile phase, do a syntax check. We used to just set
5795 a dummy reference number, because it was not used in the first pass.
5796 However, with the change of recursive back references to be atomic,
5797 we have to look for the number so that this state can be identified, as
5798 otherwise the incorrect length is computed. If it's not a backwards
5799 reference, the dummy number will do. */
5800
5801 if (lengthptr != NULL)
5802 {
5803 const uschar *temp;
5804
5805 if (namelen == 0)
5806 {
5807 *errorcodeptr = ERR62;
5808 goto FAILED;
5809 }
5810 if (*ptr != terminator)
5811 {
5812 *errorcodeptr = ERR42;
5813 goto FAILED;
5814 }
5815 if (namelen > MAX_NAME_SIZE)
5816 {
5817 *errorcodeptr = ERR48;
5818 goto FAILED;
5819 }
5820
5821 /* The name table does not exist in the first pass, so we cannot
5822 do a simple search as in the code below. Instead, we have to scan the
5823 pattern to find the number. It is important that we scan it only as
5824 far as we have got because the syntax of named subpatterns has not
5825 been checked for the rest of the pattern, and find_parens() assumes
5826 correct syntax. In any case, it's a waste of resources to scan
5827 further. We stop the scan at the current point by temporarily
5828 adjusting the value of cd->endpattern. */
5829
5830 temp = cd->end_pattern;
5831 cd->end_pattern = ptr;
5832 recno = find_parens(cd, name, namelen,
5833 (options & PCRE_EXTENDED) != 0, utf8);
5834 cd->end_pattern = temp;
5835 if (recno < 0) recno = 0; /* Forward ref; set dummy number */
5836 }
5837
5838 /* In the real compile, seek the name in the table. We check the name
5839 first, and then check that we have reached the end of the name in the
5840 table. That way, if the name that is longer than any in the table,
5841 the comparison will fail without reading beyond the table entry. */
5842
5843 else
5844 {
5845 slot = cd->name_table;
5846 for (i = 0; i < cd->names_found; i++)
5847 {
5848 if (strncmp((char *)name, (char *)slot+2, namelen) == 0 &&
5849 slot[2+namelen] == 0)
5850 break;
5851 slot += cd->name_entry_size;
5852 }
5853
5854 if (i < cd->names_found) /* Back reference */
5855 {
5856 recno = GET2(slot, 0);
5857 }
5858 else if ((recno = /* Forward back reference */
5859 find_parens(cd, name, namelen,
5860 (options & PCRE_EXTENDED) != 0, utf8)) <= 0)
5861 {
5862 *errorcodeptr = ERR15;
5863 goto FAILED;
5864 }
5865 }
5866
5867 /* In both phases, we can now go to the code than handles numerical
5868 recursion or backreferences. */
5869
5870 if (is_recurse) goto HANDLE_RECURSION;
5871 else goto HANDLE_REFERENCE;
5872
5873
5874 /* ------------------------------------------------------------ */
5875 case CHAR_R: /* Recursion */
5876 ptr++; /* Same as (?0) */
5877 /* Fall through */
5878
5879
5880 /* ------------------------------------------------------------ */
5881 case CHAR_MINUS: case CHAR_PLUS: /* Recursion or subroutine */
5882 case CHAR_0: case CHAR_1: case CHAR_2: case CHAR_3: case CHAR_4:
5883 case CHAR_5: case CHAR_6: case CHAR_7: case CHAR_8: case CHAR_9:
5884 {
5885 const uschar *called;
5886 terminator = CHAR_RIGHT_PARENTHESIS;
5887
5888 /* Come here from the \g<...> and \g'...' code (Oniguruma
5889 compatibility). However, the syntax has been checked to ensure that
5890 the ... are a (signed) number, so that neither ERR63 nor ERR29 will
5891 be called on this path, nor with the jump to OTHER_CHAR_AFTER_QUERY
5892 ever be taken. */
5893
5894 HANDLE_NUMERICAL_RECURSION:
5895
5896 if ((refsign = *ptr) == CHAR_PLUS)
5897 {
5898 ptr++;
5899 if ((digitab[*ptr] & ctype_digit) == 0)
5900 {
5901 *errorcodeptr = ERR63;
5902 goto FAILED;
5903 }
5904 }
5905 else if (refsign == CHAR_MINUS)
5906 {
5907 if ((digitab[ptr[1]] & ctype_digit) == 0)
5908 goto OTHER_CHAR_AFTER_QUERY;
5909 ptr++;
5910 }
5911
5912 recno = 0;
5913 while((digitab[*ptr] & ctype_digit) != 0)
5914 recno = recno * 10 + *ptr++ - CHAR_0;
5915
5916 if (*ptr != terminator)
5917 {
5918 *errorcodeptr = ERR29;
5919 goto FAILED;
5920 }
5921
5922 if (refsign == CHAR_MINUS)
5923 {
5924 if (recno == 0)
5925 {
5926 *errorcodeptr = ERR58;
5927 goto FAILED;
5928 }
5929 recno = cd->bracount - recno + 1;
5930 if (recno <= 0)
5931 {
5932 *errorcodeptr = ERR15;
5933 goto FAILED;
5934 }
5935 }
5936 else if (refsign == CHAR_PLUS)
5937 {
5938 if (recno == 0)
5939 {
5940 *errorcodeptr = ERR58;
5941 goto FAILED;
5942 }
5943 recno += cd->bracount;
5944 }
5945
5946 /* Come here from code above that handles a named recursion */
5947
5948 HANDLE_RECURSION:
5949
5950 previous = code;
5951 called = cd->start_code;
5952
5953 /* When we are actually compiling, find the bracket that is being
5954 referenced. Temporarily end the regex in case it doesn't exist before
5955 this point. If we end up with a forward reference, first check that
5956 the bracket does occur later so we can give the error (and position)
5957 now. Then remember this forward reference in the workspace so it can
5958 be filled in at the end. */
5959
5960 if (lengthptr == NULL)
5961 {
5962 *code = OP_END;
5963 if (recno != 0)
5964 called = _pcre_find_bracket(cd->start_code, utf8, recno);
5965
5966 /* Forward reference */
5967
5968 if (called == NULL)
5969 {
5970 if (find_parens(cd, NULL, recno,
5971 (options & PCRE_EXTENDED) != 0, utf8) < 0)
5972 {
5973 *errorcodeptr = ERR15;
5974 goto FAILED;
5975 }
5976
5977 /* Fudge the value of "called" so that when it is inserted as an
5978 offset below, what it actually inserted is the reference number
5979 of the group. Then remember the forward reference. */
5980
5981 called = cd->start_code + recno;
5982 PUTINC(cd->hwm, 0, (int)(code + 1 - cd->start_code));
5983 }
5984
5985 /* If not a forward reference, and the subpattern is still open,
5986 this is a recursive call. We check to see if this is a left
5987 recursion that could loop for ever, and diagnose that case. We
5988 must not, however, do this check if we are in a conditional
5989 subpattern because the condition might be testing for recursion in
5990 a pattern such as /(?(R)a+|(?R)b)/, which is perfectly valid.
5991 Forever loops are also detected at runtime, so those that occur in
5992 conditional subpatterns will be picked up then. */
5993
5994 else if (GET(called, 1) == 0 && cond_depth <= 0 &&
5995 could_be_empty(called, code, bcptr, utf8, cd))
5996 {
5997 *errorcodeptr = ERR40;
5998 goto FAILED;
5999 }
6000 }
6001
6002 /* Insert the recursion/subroutine item. */
6003
6004 *code = OP_RECURSE;
6005 PUT(code, 1, (int)(called - cd->start_code));
6006 code += 1 + LINK_SIZE;
6007 }
6008
6009 /* Can't determine a first byte now */
6010
6011 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
6012 continue;
6013
6014
6015 /* ------------------------------------------------------------ */
6016 default: /* Other characters: check option setting */
6017 OTHER_CHAR_AFTER_QUERY:
6018 set = unset = 0;
6019 optset = &set;
6020
6021 while (*ptr != CHAR_RIGHT_PARENTHESIS && *ptr != CHAR_COLON)
6022 {
6023 switch (*ptr++)
6024 {
6025 case CHAR_MINUS: optset = &unset; break;
6026
6027 case CHAR_J: /* Record that it changed in the external options */
6028 *optset |= PCRE_DUPNAMES;
6029 cd->external_flags |= PCRE_JCHANGED;
6030 break;
6031
6032 case CHAR_i: *optset |= PCRE_CASELESS; break;
6033 case CHAR_m: *optset |= PCRE_MULTILINE; break;
6034 case CHAR_s: *optset |= PCRE_DOTALL; break;
6035 case CHAR_x: *optset |= PCRE_EXTENDED; break;
6036 case CHAR_U: *optset |= PCRE_UNGREEDY; break;
6037 case CHAR_X: *optset |= PCRE_EXTRA; break;
6038
6039 default: *errorcodeptr = ERR12;
6040 ptr--; /* Correct the offset */
6041 goto FAILED;
6042 }
6043 }
6044
6045 /* Set up the changed option bits, but don't change anything yet. */
6046
6047 newoptions = (options | set) & (~unset);
6048
6049 /* If the options ended with ')' this is not the start of a nested
6050 group with option changes, so the options change at this level. If this
6051 item is right at the start of the pattern, the options can be
6052 abstracted and made external in the pre-compile phase, and ignored in
6053 the compile phase. This can be helpful when matching -- for instance in
6054 caseless checking of required bytes.
6055
6056 If the code pointer is not (cd->start_code + 1 + LINK_SIZE), we are
6057 definitely *not* at the start of the pattern because something has been
6058 compiled. In the pre-compile phase, however, the code pointer can have
6059 that value after the start, because it gets reset as code is discarded
6060 during the pre-compile. However, this can happen only at top level - if
6061 we are within parentheses, the starting BRA will still be present. At
6062 any parenthesis level, the length value can be used to test if anything
6063 has been compiled at that level. Thus, a test for both these conditions
6064 is necessary to ensure we correctly detect the start of the pattern in
6065 both phases.
6066
6067 If we are not at the pattern start, reset the greedy defaults and the
6068 case value for firstbyte and reqbyte. */
6069
6070 if (*ptr == CHAR_RIGHT_PARENTHESIS)
6071 {
6072 if (code == cd->start_code + 1 + LINK_SIZE &&
6073 (lengthptr == NULL || *lengthptr == 2 + 2*LINK_SIZE))
6074 {
6075 cd->external_options = newoptions;
6076 }
6077 else
6078 {
6079 greedy_default = ((newoptions & PCRE_UNGREEDY) != 0);
6080 greedy_non_default = greedy_default ^ 1;
6081 req_caseopt = ((newoptions & PCRE_CASELESS) != 0)? REQ_CASELESS : 0;
6082 }
6083
6084 /* Change options at this level, and pass them back for use
6085 in subsequent branches. */
6086
6087 *optionsptr = options = newoptions;
6088 previous = NULL; /* This item can't be repeated */
6089 continue; /* It is complete */
6090 }
6091
6092 /* If the options ended with ':' we are heading into a nested group
6093 with possible change of options. Such groups are non-capturing and are
6094 not assertions of any kind. All we need to do is skip over the ':';
6095 the newoptions value is handled below. */
6096
6097 bravalue = OP_BRA;
6098 ptr++;
6099 } /* End of switch for character following (? */
6100 } /* End of (? handling */
6101
6102 /* Opening parenthesis not followed by '*' or '?'. If PCRE_NO_AUTO_CAPTURE
6103 is set, all unadorned brackets become non-capturing and behave like (?:...)
6104 brackets. */
6105
6106 else if ((options & PCRE_NO_AUTO_CAPTURE) != 0)
6107 {
6108 bravalue = OP_BRA;
6109 }
6110
6111 /* Else we have a capturing group. */
6112
6113 else
6114 {
6115 NUMBERED_GROUP:
6116 cd->bracount += 1;
6117 PUT2(code, 1+LINK_SIZE, cd->bracount);
6118 skipbytes = 2;
6119 }
6120
6121 /* Process nested bracketed regex. Assertions used not to be repeatable,
6122 but this was changed for Perl compatibility, so all kinds can now be
6123 repeated. We copy code into a non-register variable (tempcode) in order to
6124 be able to pass its address because some compilers complain otherwise. */
6125
6126 previous = code; /* For handling repetition */
6127 *code = bravalue;
6128 tempcode = code;
6129 tempreqvary = cd->req_varyopt; /* Save value before bracket */
6130 tempbracount = cd->bracount; /* Save value before bracket */
6131 length_prevgroup = 0; /* Initialize for pre-compile phase */
6132
6133 if (!compile_regex(
6134 newoptions, /* The complete new option state */
6135 &tempcode, /* Where to put code (updated) */
6136 &ptr, /* Input pointer (updated) */
6137 errorcodeptr, /* Where to put an error message */
6138 (bravalue == OP_ASSERTBACK ||
6139 bravalue == OP_ASSERTBACK_NOT), /* TRUE if back assert */
6140 reset_bracount, /* True if (?| group */
6141 skipbytes, /* Skip over bracket number */
6142 cond_depth +
6143 ((bravalue == OP_COND)?1:0), /* Depth of condition subpatterns */
6144 &subfirstbyte, /* For possible first char */
6145 &subreqbyte, /* For possible last char */
6146 bcptr, /* Current branch chain */
6147 cd, /* Tables block */
6148 (lengthptr == NULL)? NULL : /* Actual compile phase */
6149 &length_prevgroup /* Pre-compile phase */
6150 ))
6151 goto FAILED;
6152
6153 /* If this was an atomic group and there are no capturing groups within it,
6154 generate OP_ONCE_NC instead of OP_ONCE. */
6155
6156 if (bravalue == OP_ONCE && cd->bracount <= tempbracount)
6157 *code = OP_ONCE_NC;
6158
6159 if (bravalue >= OP_ASSERT && bravalue <= OP_ASSERTBACK_NOT)
6160 cd->assert_depth -= 1;
6161
6162 /* At the end of compiling, code is still pointing to the start of the
6163 group, while tempcode has been updated to point past the end of the group.
6164 The pattern pointer (ptr) is on the bracket.
6165
6166 If this is a conditional bracket, check that there are no more than
6167 two branches in the group, or just one if it's a DEFINE group. We do this
6168 in the real compile phase, not in the pre-pass, where the whole group may
6169 not be available. */
6170
6171 if (bravalue == OP_COND && lengthptr == NULL)
6172 {
6173 uschar *tc = code;
6174 int condcount = 0;
6175
6176 do {
6177 condcount++;
6178 tc += GET(tc,1);
6179 }
6180 while (*tc != OP_KET);
6181
6182 /* A DEFINE group is never obeyed inline (the "condition" is always
6183 false). It must have only one branch. */
6184
6185 if (code[LINK_SIZE+1] == OP_DEF)
6186 {
6187 if (condcount > 1)
6188 {
6189 *errorcodeptr = ERR54;
6190 goto FAILED;
6191 }
6192 bravalue = OP_DEF; /* Just a flag to suppress char handling below */
6193 }
6194
6195 /* A "normal" conditional group. If there is just one branch, we must not
6196 make use of its firstbyte or reqbyte, because this is equivalent to an
6197 empty second branch. */
6198
6199 else
6200 {
6201 if (condcount > 2)
6202 {
6203 *errorcodeptr = ERR27;
6204 goto FAILED;
6205 }
6206 if (condcount == 1) subfirstbyte = subreqbyte = REQ_NONE;
6207 }
6208 }
6209
6210 /* Error if hit end of pattern */
6211
6212 if (*ptr != CHAR_RIGHT_PARENTHESIS)
6213 {
6214 *errorcodeptr = ERR14;
6215 goto FAILED;
6216 }
6217
6218 /* In the pre-compile phase, update the length by the length of the group,
6219 less the brackets at either end. Then reduce the compiled code to just a
6220 set of non-capturing brackets so that it doesn't use much memory if it is
6221 duplicated by a quantifier.*/
6222
6223 if (lengthptr != NULL)
6224 {
6225 if (OFLOW_MAX - *lengthptr < length_prevgroup - 2 - 2*LINK_SIZE)
6226 {
6227 *errorcodeptr = ERR20;
6228 goto FAILED;
6229 }
6230 *lengthptr += length_prevgroup - 2 - 2*LINK_SIZE;
6231 code++; /* This already contains bravalue */
6232 PUTINC(code, 0, 1 + LINK_SIZE);
6233 *code++ = OP_KET;
6234 PUTINC(code, 0, 1 + LINK_SIZE);
6235 break; /* No need to waste time with special character handling */
6236 }
6237
6238 /* Otherwise update the main code pointer to the end of the group. */
6239
6240 code = tempcode;
6241
6242 /* For a DEFINE group, required and first character settings are not
6243 relevant. */
6244
6245 if (bravalue == OP_DEF) break;
6246
6247 /* Handle updating of the required and first characters for other types of
6248 group. Update for normal brackets of all kinds, and conditions with two
6249 branches (see code above). If the bracket is followed by a quantifier with
6250 zero repeat, we have to back off. Hence the definition of zeroreqbyte and
6251 zerofirstbyte outside the main loop so that they can be accessed for the
6252 back off. */
6253
6254 zeroreqbyte = reqbyte;
6255 zerofirstbyte = firstbyte;
6256 groupsetfirstbyte = FALSE;
6257
6258 if (bravalue >= OP_ONCE)
6259 {
6260 /* If we have not yet set a firstbyte in this branch, take it from the
6261 subpattern, remembering that it was set here so that a repeat of more
6262 than one can replicate it as reqbyte if necessary. If the subpattern has
6263 no firstbyte, set "none" for the whole branch. In both cases, a zero
6264 repeat forces firstbyte to "none". */
6265
6266 if (firstbyte == REQ_UNSET)
6267 {
6268 if (subfirstbyte >= 0)
6269 {
6270 firstbyte = subfirstbyte;
6271 groupsetfirstbyte = TRUE;
6272 }
6273 else firstbyte = REQ_NONE;
6274 zerofirstbyte = REQ_NONE;
6275 }
6276
6277 /* If firstbyte was previously set, convert the subpattern's firstbyte
6278 into reqbyte if there wasn't one, using the vary flag that was in
6279 existence beforehand. */
6280
6281 else if (subfirstbyte >= 0 && subreqbyte < 0)
6282 subreqbyte = subfirstbyte | tempreqvary;
6283
6284 /* If the subpattern set a required byte (or set a first byte that isn't
6285 really the first byte - see above), set it. */
6286
6287 if (subreqbyte >= 0) reqbyte = subreqbyte;
6288 }
6289
6290 /* For a forward assertion, we take the reqbyte, if set. This can be
6291 helpful if the pattern that follows the assertion doesn't set a different
6292 char. For example, it's useful for /(?=abcde).+/. We can't set firstbyte
6293 for an assertion, however because it leads to incorrect effect for patterns
6294 such as /(?=a)a.+/ when the "real" "a" would then become a reqbyte instead
6295 of a firstbyte. This is overcome by a scan at the end if there's no
6296 firstbyte, looking for an asserted first char. */
6297
6298 else if (bravalue == OP_ASSERT && subreqbyte >= 0) reqbyte = subreqbyte;
6299 break; /* End of processing '(' */
6300
6301
6302 /* ===================================================================*/
6303 /* Handle metasequences introduced by \. For ones like \d, the ESC_ values
6304 are arranged to be the negation of the corresponding OP_values in the
6305 default case when PCRE_UCP is not set. For the back references, the values
6306 are ESC_REF plus the reference number. Only back references and those types
6307 that consume a character may be repeated. We can test for values between
6308 ESC_b and ESC_Z for the latter; this may have to change if any new ones are
6309 ever created. */
6310
6311 case CHAR_BACKSLASH:
6312 tempptr = ptr;
6313 c = check_escape(&ptr, errorcodeptr, cd->bracount, options, FALSE);
6314 if (*errorcodeptr != 0) goto FAILED;
6315
6316 if (c < 0)
6317 {
6318 if (-c == ESC_Q) /* Handle start of quoted string */
6319 {
6320 if (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
6321 ptr += 2; /* avoid empty string */
6322 else inescq = TRUE;
6323 continue;
6324 }
6325
6326 if (-c == ESC_E) continue; /* Perl ignores an orphan \E */
6327
6328 /* For metasequences that actually match a character, we disable the
6329 setting of a first character if it hasn't already been set. */
6330
6331 if (firstbyte == REQ_UNSET && -c > ESC_b && -c < ESC_Z)
6332 firstbyte = REQ_NONE;
6333
6334 /* Set values to reset to if this is followed by a zero repeat. */
6335
6336 zerofirstbyte = firstbyte;
6337 zeroreqbyte = reqbyte;
6338
6339 /* \g<name> or \g'name' is a subroutine call by name and \g<n> or \g'n'
6340 is a subroutine call by number (Oniguruma syntax). In fact, the value
6341 -ESC_g is returned only for these cases. So we don't need to check for <
6342 or ' if the value is -ESC_g. For the Perl syntax \g{n} the value is
6343 -ESC_REF+n, and for the Perl syntax \g{name} the result is -ESC_k (as
6344 that is a synonym for a named back reference). */
6345
6346 if (-c == ESC_g)
6347 {
6348 const uschar *p;
6349 save_hwm = cd->hwm; /* Normally this is set when '(' is read */
6350 terminator = (*(++ptr) == CHAR_LESS_THAN_SIGN)?
6351 CHAR_GREATER_THAN_SIGN : CHAR_APOSTROPHE;
6352
6353 /* These two statements stop the compiler for warning about possibly
6354 unset variables caused by the jump to HANDLE_NUMERICAL_RECURSION. In
6355 fact, because we actually check for a number below, the paths that
6356 would actually be in error are never taken. */
6357
6358 skipbytes = 0;
6359 reset_bracount = FALSE;
6360
6361 /* Test for a name */
6362
6363 if (ptr[1] != CHAR_PLUS && ptr[1] != CHAR_MINUS)
6364 {
6365 BOOL isnumber = TRUE;
6366 for (p = ptr + 1; *p != 0 && *p != terminator; p++)
6367 {
6368 if ((cd->ctypes[*p] & ctype_digit) == 0) isnumber = FALSE;
6369 if ((cd->ctypes[*p] & ctype_word) == 0) break;
6370 }
6371 if (*p != terminator)
6372 {
6373 *errorcodeptr = ERR57;
6374 break;
6375 }
6376 if (isnumber)
6377 {
6378 ptr++;
6379 goto HANDLE_NUMERICAL_RECURSION;
6380 }
6381 is_recurse = TRUE;
6382 goto NAMED_REF_OR_RECURSE;
6383 }
6384
6385 /* Test a signed number in angle brackets or quotes. */
6386
6387 p = ptr + 2;
6388 while ((digitab[*p] & ctype_digit) != 0) p++;
6389 if (*p != terminator)
6390 {
6391 *errorcodeptr = ERR57;
6392 break;
6393 }
6394 ptr++;
6395 goto HANDLE_NUMERICAL_RECURSION;
6396 }
6397
6398 /* \k<name> or \k'name' is a back reference by name (Perl syntax).
6399 We also support \k{name} (.NET syntax). */
6400
6401 if (-c == ESC_k)
6402 {
6403 if ((ptr[1] != CHAR_LESS_THAN_SIGN &&
6404 ptr[1] != CHAR_APOSTROPHE && ptr[1] != CHAR_LEFT_CURLY_BRACKET))
6405 {
6406 *errorcodeptr = ERR69;
6407 break;
6408 }
6409 is_recurse = FALSE;
6410 terminator = (*(++ptr) == CHAR_LESS_THAN_SIGN)?
6411 CHAR_GREATER_THAN_SIGN : (*ptr == CHAR_APOSTROPHE)?
6412 CHAR_APOSTROPHE : CHAR_RIGHT_CURLY_BRACKET;
6413 goto NAMED_REF_OR_RECURSE;
6414 }
6415
6416 /* Back references are handled specially; must disable firstbyte if
6417 not set to cope with cases like (?=(\w+))\1: which would otherwise set
6418 ':' later. */
6419
6420 if (-c >= ESC_REF)
6421 {
6422 open_capitem *oc;
6423 recno = -c - ESC_REF;
6424
6425 HANDLE_REFERENCE: /* Come here from named backref handling */
6426 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
6427 previous = code;
6428 *code++ = ((options & PCRE_CASELESS) != 0)? OP_REFI : OP_REF;
6429 PUT2INC(code, 0, recno);
6430 cd->backref_map |= (recno < 32)? (1 << recno) : 1;
6431 if (recno > cd->top_backref) cd->top_backref = recno;
6432
6433 /* Check to see if this back reference is recursive, that it, it
6434 is inside the group that it references. A flag is set so that the
6435 group can be made atomic. */
6436
6437 for (oc = cd->open_caps; oc != NULL; oc = oc->next)
6438 {
6439 if (oc->number == recno)
6440 {
6441 oc->flag = TRUE;
6442 break;
6443 }
6444 }
6445 }
6446
6447 /* So are Unicode property matches, if supported. */
6448
6449 #ifdef SUPPORT_UCP
6450 else if (-c == ESC_P || -c == ESC_p)
6451 {
6452 BOOL negated;
6453 int pdata;
6454 int ptype = get_ucp(&ptr, &negated, &pdata, errorcodeptr);
6455 if (ptype < 0) goto FAILED;
6456 previous = code;
6457 *code++ = ((-c == ESC_p) != negated)? OP_PROP : OP_NOTPROP;
6458 *code++ = ptype;
6459 *code++ = pdata;
6460 }
6461 #else
6462
6463 /* If Unicode properties are not supported, \X, \P, and \p are not
6464 allowed. */
6465
6466 else if (-c == ESC_X || -c == ESC_P || -c == ESC_p)
6467 {
6468 *errorcodeptr = ERR45;
6469 goto FAILED;
6470 }
6471 #endif
6472
6473 /* For the rest (including \X when Unicode properties are supported), we
6474 can obtain the OP value by negating the escape value in the default
6475 situation when PCRE_UCP is not set. When it *is* set, we substitute
6476 Unicode property tests. */
6477
6478 else
6479 {
6480 #ifdef SUPPORT_UCP
6481 if (-c >= ESC_DU && -c <= ESC_wu)
6482 {
6483 nestptr = ptr + 1; /* Where to resume */
6484 ptr = substitutes[-c - ESC_DU] - 1; /* Just before substitute */
6485 }
6486 else
6487 #endif
6488 /* In non-UTF-8 mode, we turn \C into OP_ALLANY instead of OP_ANYBYTE
6489 so that it works in DFA mode and in lookbehinds. */
6490
6491 {
6492 previous = (-c > ESC_b && -c < ESC_Z)? code : NULL;
6493 *code++ = (!utf8 && c == -ESC_C)? OP_ALLANY : -c;
6494 }
6495 }
6496 continue;
6497 }
6498
6499 /* We have a data character whose value is in c. In UTF-8 mode it may have
6500 a value > 127. We set its representation in the length/buffer, and then
6501 handle it as a data character. */
6502
6503 #ifdef SUPPORT_UTF8
6504 if (utf8 && c > 127)
6505 mclength = _pcre_ord2utf8(c, mcbuffer);
6506 else
6507 #endif
6508
6509 {
6510 mcbuffer[0] = c;
6511 mclength = 1;
6512 }
6513 goto ONE_CHAR;
6514
6515
6516 /* ===================================================================*/
6517 /* Handle a literal character. It is guaranteed not to be whitespace or #
6518 when the extended flag is set. If we are in UTF-8 mode, it may be a
6519 multi-byte literal character. */
6520
6521 default:
6522 NORMAL_CHAR:
6523 mclength = 1;
6524 mcbuffer[0] = c;
6525
6526 #ifdef SUPPORT_UTF8
6527 if (utf8 && c >= 0xc0)
6528 {
6529 while ((ptr[1] & 0xc0) == 0x80)
6530 mcbuffer[mclength++] = *(++ptr);
6531 }
6532 #endif
6533
6534 /* At this point we have the character's bytes in mcbuffer, and the length
6535 in mclength. When not in UTF-8 mode, the length is always 1. */
6536
6537 ONE_CHAR:
6538 previous = code;
6539 *code++ = ((options & PCRE_CASELESS) != 0)? OP_CHARI : OP_CHAR;
6540 for (c = 0; c < mclength; c++) *code++ = mcbuffer[c];
6541
6542 /* Remember if \r or \n were seen */
6543
6544 if (mcbuffer[0] == CHAR_CR || mcbuffer[0] == CHAR_NL)
6545 cd->external_flags |= PCRE_HASCRORLF;
6546
6547 /* Set the first and required bytes appropriately. If no previous first
6548 byte, set it from this character, but revert to none on a zero repeat.
6549 Otherwise, leave the firstbyte value alone, and don't change it on a zero
6550 repeat. */
6551
6552 if (firstbyte == REQ_UNSET)
6553 {
6554 zerofirstbyte = REQ_NONE;
6555 zeroreqbyte = reqbyte;
6556
6557 /* If the character is more than one byte long, we can set firstbyte
6558 only if it is not to be matched caselessly. */
6559
6560 if (mclength == 1 || req_caseopt == 0)
6561 {
6562 firstbyte = mcbuffer[0] | req_caseopt;
6563 if (mclength != 1) reqbyte = code[-1] | cd->req_varyopt;
6564 }
6565 else firstbyte = reqbyte = REQ_NONE;
6566 }
6567
6568 /* firstbyte was previously set; we can set reqbyte only if the length is
6569 1 or the matching is caseful. */
6570
6571 else
6572 {
6573 zerofirstbyte = firstbyte;
6574 zeroreqbyte = reqbyte;
6575 if (mclength == 1 || req_caseopt == 0)
6576 reqbyte = code[-1] | req_caseopt | cd->req_varyopt;
6577 }
6578
6579 break; /* End of literal character handling */
6580 }
6581 } /* end of big loop */
6582
6583
6584 /* Control never reaches here by falling through, only by a goto for all the
6585 error states. Pass back the position in the pattern so that it can be displayed
6586 to the user for diagnosing the error. */
6587
6588 FAILED:
6589 *ptrptr = ptr;
6590 return FALSE;
6591 }
6592
6593
6594
6595
6596 /*************************************************
6597 * Compile sequence of alternatives *
6598 *************************************************/
6599
6600 /* On entry, ptr is pointing past the bracket character, but on return it
6601 points to the closing bracket, or vertical bar, or end of string. The code
6602 variable is pointing at the byte into which the BRA operator has been stored.
6603 This function is used during the pre-compile phase when we are trying to find
6604 out the amount of memory needed, as well as during the real compile phase. The
6605 value of lengthptr distinguishes the two phases.
6606
6607 Arguments:
6608 options option bits, including any changes for this subpattern
6609 codeptr -> the address of the current code pointer
6610 ptrptr -> the address of the current pattern pointer
6611 errorcodeptr -> pointer to error code variable
6612 lookbehind TRUE if this is a lookbehind assertion
6613 reset_bracount TRUE to reset the count for each branch
6614 skipbytes skip this many bytes at start (for brackets and OP_COND)
6615 cond_depth depth of nesting for conditional subpatterns
6616 firstbyteptr place to put the first required character, or a negative number
6617 reqbyteptr place to put the last required character, or a negative number
6618 bcptr pointer to the chain of currently open branches
6619 cd points to the data block with tables pointers etc.
6620 lengthptr NULL during the real compile phase
6621 points to length accumulator during pre-compile phase
6622
6623 Returns: TRUE on success
6624 */
6625
6626 static BOOL
6627 compile_regex(int options, uschar **codeptr, const uschar **ptrptr,
6628 int *errorcodeptr, BOOL lookbehind, BOOL reset_bracount, int skipbytes,
6629 int cond_depth, int *firstbyteptr, int *reqbyteptr, branch_chain *bcptr,
6630 compile_data *cd, int *lengthptr)
6631 {
6632 const uschar *ptr = *ptrptr;
6633 uschar *code = *codeptr;
6634 uschar *last_branch = code;
6635 uschar *start_bracket = code;
6636 uschar *reverse_count = NULL;
6637 open_capitem capitem;
6638 int capnumber = 0;
6639 int firstbyte, reqbyte;
6640 int branchfirstbyte, branchreqbyte;
6641 int length;
6642 int orig_bracount;
6643 int max_bracount;
6644 branch_chain bc;
6645
6646 bc.outer = bcptr;
6647 bc.current_branch = code;
6648
6649 firstbyte = reqbyte = REQ_UNSET;
6650
6651 /* Accumulate the length for use in the pre-compile phase. Start with the
6652 length of the BRA and KET and any extra bytes t